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

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

Version: ~ [ linux-5.12-rc1 ] ~ [ linux-5.11.2 ] ~ [ linux-5.10.19 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.101 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.177 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.222 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.258 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.258 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ 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 
 56 /*
 57  * xprtsock tunables
 58  */
 59 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
 60 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
 61 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
 62 
 63 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
 64 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
 65 
 66 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
 67 
 68 #define XS_TCP_LINGER_TO        (15U * HZ)
 69 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
 70 
 71 /*
 72  * We can register our own files under /proc/sys/sunrpc by
 73  * calling register_sysctl_table() again.  The files in that
 74  * directory become the union of all files registered there.
 75  *
 76  * We simply need to make sure that we don't collide with
 77  * someone else's file names!
 78  */
 79 
 80 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
 81 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
 82 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
 83 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
 84 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
 85 
 86 static struct ctl_table_header *sunrpc_table_header;
 87 
 88 /*
 89  * FIXME: changing the UDP slot table size should also resize the UDP
 90  *        socket buffers for existing UDP transports
 91  */
 92 static struct ctl_table xs_tunables_table[] = {
 93         {
 94                 .procname       = "udp_slot_table_entries",
 95                 .data           = &xprt_udp_slot_table_entries,
 96                 .maxlen         = sizeof(unsigned int),
 97                 .mode           = 0644,
 98                 .proc_handler   = proc_dointvec_minmax,
 99                 .extra1         = &min_slot_table_size,
100                 .extra2         = &max_slot_table_size
101         },
102         {
103                 .procname       = "tcp_slot_table_entries",
104                 .data           = &xprt_tcp_slot_table_entries,
105                 .maxlen         = sizeof(unsigned int),
106                 .mode           = 0644,
107                 .proc_handler   = proc_dointvec_minmax,
108                 .extra1         = &min_slot_table_size,
109                 .extra2         = &max_slot_table_size
110         },
111         {
112                 .procname       = "tcp_max_slot_table_entries",
113                 .data           = &xprt_max_tcp_slot_table_entries,
114                 .maxlen         = sizeof(unsigned int),
115                 .mode           = 0644,
116                 .proc_handler   = proc_dointvec_minmax,
117                 .extra1         = &min_slot_table_size,
118                 .extra2         = &max_tcp_slot_table_limit
119         },
120         {
121                 .procname       = "min_resvport",
122                 .data           = &xprt_min_resvport,
123                 .maxlen         = sizeof(unsigned int),
124                 .mode           = 0644,
125                 .proc_handler   = proc_dointvec_minmax,
126                 .extra1         = &xprt_min_resvport_limit,
127                 .extra2         = &xprt_max_resvport
128         },
129         {
130                 .procname       = "max_resvport",
131                 .data           = &xprt_max_resvport,
132                 .maxlen         = sizeof(unsigned int),
133                 .mode           = 0644,
134                 .proc_handler   = proc_dointvec_minmax,
135                 .extra1         = &xprt_min_resvport,
136                 .extra2         = &xprt_max_resvport_limit
137         },
138         {
139                 .procname       = "tcp_fin_timeout",
140                 .data           = &xs_tcp_fin_timeout,
141                 .maxlen         = sizeof(xs_tcp_fin_timeout),
142                 .mode           = 0644,
143                 .proc_handler   = proc_dointvec_jiffies,
144         },
145         { },
146 };
147 
148 static struct ctl_table sunrpc_table[] = {
149         {
150                 .procname       = "sunrpc",
151                 .mode           = 0555,
152                 .child          = xs_tunables_table
153         },
154         { },
155 };
156 
157 #endif
158 
159 /*
160  * Wait duration for a reply from the RPC portmapper.
161  */
162 #define XS_BIND_TO              (60U * HZ)
163 
164 /*
165  * Delay if a UDP socket connect error occurs.  This is most likely some
166  * kind of resource problem on the local host.
167  */
168 #define XS_UDP_REEST_TO         (2U * HZ)
169 
170 /*
171  * The reestablish timeout allows clients to delay for a bit before attempting
172  * to reconnect to a server that just dropped our connection.
173  *
174  * We implement an exponential backoff when trying to reestablish a TCP
175  * transport connection with the server.  Some servers like to drop a TCP
176  * connection when they are overworked, so we start with a short timeout and
177  * increase over time if the server is down or not responding.
178  */
179 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
180 
181 /*
182  * TCP idle timeout; client drops the transport socket if it is idle
183  * for this long.  Note that we also timeout UDP sockets to prevent
184  * holding port numbers when there is no RPC traffic.
185  */
186 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
187 
188 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
189 # undef  RPC_DEBUG_DATA
190 # define RPCDBG_FACILITY        RPCDBG_TRANS
191 #endif
192 
193 #ifdef RPC_DEBUG_DATA
194 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
195 {
196         u8 *buf = (u8 *) packet;
197         int j;
198 
199         dprintk("RPC:       %s\n", msg);
200         for (j = 0; j < count && j < 128; j += 4) {
201                 if (!(j & 31)) {
202                         if (j)
203                                 dprintk("\n");
204                         dprintk("0x%04x ", j);
205                 }
206                 dprintk("%02x%02x%02x%02x ",
207                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
208         }
209         dprintk("\n");
210 }
211 #else
212 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
213 {
214         /* NOP */
215 }
216 #endif
217 
218 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
219 {
220         return (struct rpc_xprt *) sk->sk_user_data;
221 }
222 
223 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
224 {
225         return (struct sockaddr *) &xprt->addr;
226 }
227 
228 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
229 {
230         return (struct sockaddr_un *) &xprt->addr;
231 }
232 
233 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
234 {
235         return (struct sockaddr_in *) &xprt->addr;
236 }
237 
238 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
239 {
240         return (struct sockaddr_in6 *) &xprt->addr;
241 }
242 
243 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
244 {
245         struct sockaddr *sap = xs_addr(xprt);
246         struct sockaddr_in6 *sin6;
247         struct sockaddr_in *sin;
248         struct sockaddr_un *sun;
249         char buf[128];
250 
251         switch (sap->sa_family) {
252         case AF_LOCAL:
253                 sun = xs_addr_un(xprt);
254                 strlcpy(buf, sun->sun_path, sizeof(buf));
255                 xprt->address_strings[RPC_DISPLAY_ADDR] =
256                                                 kstrdup(buf, GFP_KERNEL);
257                 break;
258         case AF_INET:
259                 (void)rpc_ntop(sap, buf, sizeof(buf));
260                 xprt->address_strings[RPC_DISPLAY_ADDR] =
261                                                 kstrdup(buf, GFP_KERNEL);
262                 sin = xs_addr_in(xprt);
263                 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
264                 break;
265         case AF_INET6:
266                 (void)rpc_ntop(sap, buf, sizeof(buf));
267                 xprt->address_strings[RPC_DISPLAY_ADDR] =
268                                                 kstrdup(buf, GFP_KERNEL);
269                 sin6 = xs_addr_in6(xprt);
270                 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
271                 break;
272         default:
273                 BUG();
274         }
275 
276         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
277 }
278 
279 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
280 {
281         struct sockaddr *sap = xs_addr(xprt);
282         char buf[128];
283 
284         snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
285         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
286 
287         snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
288         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
289 }
290 
291 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
292                                      const char *protocol,
293                                      const char *netid)
294 {
295         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
296         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
297         xs_format_common_peer_addresses(xprt);
298         xs_format_common_peer_ports(xprt);
299 }
300 
301 static void xs_update_peer_port(struct rpc_xprt *xprt)
302 {
303         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
304         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
305 
306         xs_format_common_peer_ports(xprt);
307 }
308 
309 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
310 {
311         unsigned int i;
312 
313         for (i = 0; i < RPC_DISPLAY_MAX; i++)
314                 switch (i) {
315                 case RPC_DISPLAY_PROTO:
316                 case RPC_DISPLAY_NETID:
317                         continue;
318                 default:
319                         kfree(xprt->address_strings[i]);
320                 }
321 }
322 
323 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
324 
325 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
326 {
327         struct msghdr msg = {
328                 .msg_name       = addr,
329                 .msg_namelen    = addrlen,
330                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
331         };
332         struct kvec iov = {
333                 .iov_base       = vec->iov_base + base,
334                 .iov_len        = vec->iov_len - base,
335         };
336 
337         if (iov.iov_len != 0)
338                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
339         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
340 }
341 
342 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
343 {
344         ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
345                         int offset, size_t size, int flags);
346         struct page **ppage;
347         unsigned int remainder;
348         int err;
349 
350         remainder = xdr->page_len - base;
351         base += xdr->page_base;
352         ppage = xdr->pages + (base >> PAGE_SHIFT);
353         base &= ~PAGE_MASK;
354         do_sendpage = sock->ops->sendpage;
355         if (!zerocopy)
356                 do_sendpage = sock_no_sendpage;
357         for(;;) {
358                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
359                 int flags = XS_SENDMSG_FLAGS;
360 
361                 remainder -= len;
362                 if (more)
363                         flags |= MSG_MORE;
364                 if (remainder != 0)
365                         flags |= MSG_SENDPAGE_NOTLAST | MSG_MORE;
366                 err = do_sendpage(sock, *ppage, base, len, flags);
367                 if (remainder == 0 || err != len)
368                         break;
369                 *sent_p += err;
370                 ppage++;
371                 base = 0;
372         }
373         if (err > 0) {
374                 *sent_p += err;
375                 err = 0;
376         }
377         return err;
378 }
379 
380 /**
381  * xs_sendpages - write pages directly to a socket
382  * @sock: socket to send on
383  * @addr: UDP only -- address of destination
384  * @addrlen: UDP only -- length of destination address
385  * @xdr: buffer containing this request
386  * @base: starting position in the buffer
387  * @zerocopy: true if it is safe to use sendpage()
388  * @sent_p: return the total number of bytes successfully queued for sending
389  *
390  */
391 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
392 {
393         unsigned int remainder = xdr->len - base;
394         int err = 0;
395         int sent = 0;
396 
397         if (unlikely(!sock))
398                 return -ENOTSOCK;
399 
400         if (base != 0) {
401                 addr = NULL;
402                 addrlen = 0;
403         }
404 
405         if (base < xdr->head[0].iov_len || addr != NULL) {
406                 unsigned int len = xdr->head[0].iov_len - base;
407                 remainder -= len;
408                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
409                 if (remainder == 0 || err != len)
410                         goto out;
411                 *sent_p += err;
412                 base = 0;
413         } else
414                 base -= xdr->head[0].iov_len;
415 
416         if (base < xdr->page_len) {
417                 unsigned int len = xdr->page_len - base;
418                 remainder -= len;
419                 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent);
420                 *sent_p += sent;
421                 if (remainder == 0 || sent != len)
422                         goto out;
423                 base = 0;
424         } else
425                 base -= xdr->page_len;
426 
427         if (base >= xdr->tail[0].iov_len)
428                 return 0;
429         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
430 out:
431         if (err > 0) {
432                 *sent_p += err;
433                 err = 0;
434         }
435         return err;
436 }
437 
438 static void xs_nospace_callback(struct rpc_task *task)
439 {
440         struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
441 
442         transport->inet->sk_write_pending--;
443 }
444 
445 /**
446  * xs_nospace - place task on wait queue if transmit was incomplete
447  * @task: task to put to sleep
448  *
449  */
450 static int xs_nospace(struct rpc_task *task)
451 {
452         struct rpc_rqst *req = task->tk_rqstp;
453         struct rpc_xprt *xprt = req->rq_xprt;
454         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
455         struct sock *sk = transport->inet;
456         int ret = -EAGAIN;
457 
458         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
459                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
460                         req->rq_slen);
461 
462         /* Protect against races with write_space */
463         spin_lock_bh(&xprt->transport_lock);
464 
465         /* Don't race with disconnect */
466         if (xprt_connected(xprt)) {
467                 /* wait for more buffer space */
468                 sk->sk_write_pending++;
469                 xprt_wait_for_buffer_space(task, xs_nospace_callback);
470         } else
471                 ret = -ENOTCONN;
472 
473         spin_unlock_bh(&xprt->transport_lock);
474 
475         /* Race breaker in case memory is freed before above code is called */
476         if (ret == -EAGAIN) {
477                 struct socket_wq *wq;
478 
479                 rcu_read_lock();
480                 wq = rcu_dereference(sk->sk_wq);
481                 set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
482                 rcu_read_unlock();
483 
484                 sk->sk_write_space(sk);
485         }
486         return ret;
487 }
488 
489 /*
490  * Construct a stream transport record marker in @buf.
491  */
492 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
493 {
494         u32 reclen = buf->len - sizeof(rpc_fraghdr);
495         rpc_fraghdr *base = buf->head[0].iov_base;
496         *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
497 }
498 
499 /**
500  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
501  * @task: RPC task that manages the state of an RPC request
502  *
503  * Return values:
504  *        0:    The request has been sent
505  *   EAGAIN:    The socket was blocked, please call again later to
506  *              complete the request
507  * ENOTCONN:    Caller needs to invoke connect logic then call again
508  *    other:    Some other error occured, the request was not sent
509  */
510 static int xs_local_send_request(struct rpc_task *task)
511 {
512         struct rpc_rqst *req = task->tk_rqstp;
513         struct rpc_xprt *xprt = req->rq_xprt;
514         struct sock_xprt *transport =
515                                 container_of(xprt, struct sock_xprt, xprt);
516         struct xdr_buf *xdr = &req->rq_snd_buf;
517         int status;
518         int sent = 0;
519 
520         xs_encode_stream_record_marker(&req->rq_snd_buf);
521 
522         xs_pktdump("packet data:",
523                         req->rq_svec->iov_base, req->rq_svec->iov_len);
524 
525         status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent,
526                               true, &sent);
527         dprintk("RPC:       %s(%u) = %d\n",
528                         __func__, xdr->len - req->rq_bytes_sent, status);
529 
530         if (status == -EAGAIN && sock_writeable(transport->inet))
531                 status = -ENOBUFS;
532 
533         if (likely(sent > 0) || status == 0) {
534                 req->rq_bytes_sent += sent;
535                 req->rq_xmit_bytes_sent += sent;
536                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
537                         req->rq_bytes_sent = 0;
538                         return 0;
539                 }
540                 status = -EAGAIN;
541         }
542 
543         switch (status) {
544         case -ENOBUFS:
545                 break;
546         case -EAGAIN:
547                 status = xs_nospace(task);
548                 break;
549         default:
550                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
551                         -status);
552         case -EPIPE:
553                 xs_close(xprt);
554                 status = -ENOTCONN;
555         }
556 
557         return status;
558 }
559 
560 /**
561  * xs_udp_send_request - write an RPC request to a UDP socket
562  * @task: address of RPC task that manages the state of an RPC request
563  *
564  * Return values:
565  *        0:    The request has been sent
566  *   EAGAIN:    The socket was blocked, please call again later to
567  *              complete the request
568  * ENOTCONN:    Caller needs to invoke connect logic then call again
569  *    other:    Some other error occurred, the request was not sent
570  */
571 static int xs_udp_send_request(struct rpc_task *task)
572 {
573         struct rpc_rqst *req = task->tk_rqstp;
574         struct rpc_xprt *xprt = req->rq_xprt;
575         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
576         struct xdr_buf *xdr = &req->rq_snd_buf;
577         int sent = 0;
578         int status;
579 
580         xs_pktdump("packet data:",
581                                 req->rq_svec->iov_base,
582                                 req->rq_svec->iov_len);
583 
584         if (!xprt_bound(xprt))
585                 return -ENOTCONN;
586         status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
587                               xdr, req->rq_bytes_sent, true, &sent);
588 
589         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
590                         xdr->len - req->rq_bytes_sent, status);
591 
592         /* firewall is blocking us, don't return -EAGAIN or we end up looping */
593         if (status == -EPERM)
594                 goto process_status;
595 
596         if (status == -EAGAIN && sock_writeable(transport->inet))
597                 status = -ENOBUFS;
598 
599         if (sent > 0 || status == 0) {
600                 req->rq_xmit_bytes_sent += sent;
601                 if (sent >= req->rq_slen)
602                         return 0;
603                 /* Still some bytes left; set up for a retry later. */
604                 status = -EAGAIN;
605         }
606 
607 process_status:
608         switch (status) {
609         case -ENOTSOCK:
610                 status = -ENOTCONN;
611                 /* Should we call xs_close() here? */
612                 break;
613         case -EAGAIN:
614                 status = xs_nospace(task);
615                 break;
616         case -ENETUNREACH:
617         case -ENOBUFS:
618         case -EPIPE:
619         case -ECONNREFUSED:
620         case -EPERM:
621                 /* When the server has died, an ICMP port unreachable message
622                  * prompts ECONNREFUSED. */
623                 break;
624         default:
625                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
626                         -status);
627         }
628 
629         return status;
630 }
631 
632 /**
633  * xs_tcp_send_request - write an RPC request to a TCP socket
634  * @task: address of RPC task that manages the state of an RPC request
635  *
636  * Return values:
637  *        0:    The request has been sent
638  *   EAGAIN:    The socket was blocked, please call again later to
639  *              complete the request
640  * ENOTCONN:    Caller needs to invoke connect logic then call again
641  *    other:    Some other error occurred, the request was not sent
642  *
643  * XXX: In the case of soft timeouts, should we eventually give up
644  *      if sendmsg is not able to make progress?
645  */
646 static int xs_tcp_send_request(struct rpc_task *task)
647 {
648         struct rpc_rqst *req = task->tk_rqstp;
649         struct rpc_xprt *xprt = req->rq_xprt;
650         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
651         struct xdr_buf *xdr = &req->rq_snd_buf;
652         bool zerocopy = true;
653         bool vm_wait = false;
654         int status;
655         int sent;
656 
657         xs_encode_stream_record_marker(&req->rq_snd_buf);
658 
659         xs_pktdump("packet data:",
660                                 req->rq_svec->iov_base,
661                                 req->rq_svec->iov_len);
662         /* Don't use zero copy if this is a resend. If the RPC call
663          * completes while the socket holds a reference to the pages,
664          * then we may end up resending corrupted data.
665          */
666         if (task->tk_flags & RPC_TASK_SENT)
667                 zerocopy = false;
668 
669         /* Continue transmitting the packet/record. We must be careful
670          * to cope with writespace callbacks arriving _after_ we have
671          * called sendmsg(). */
672         while (1) {
673                 sent = 0;
674                 status = xs_sendpages(transport->sock, NULL, 0, xdr,
675                                       req->rq_bytes_sent, zerocopy, &sent);
676 
677                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
678                                 xdr->len - req->rq_bytes_sent, status);
679 
680                 /* If we've sent the entire packet, immediately
681                  * reset the count of bytes sent. */
682                 req->rq_bytes_sent += sent;
683                 req->rq_xmit_bytes_sent += sent;
684                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
685                         req->rq_bytes_sent = 0;
686                         return 0;
687                 }
688 
689                 WARN_ON_ONCE(sent == 0 && status == 0);
690 
691                 if (status == -EAGAIN ) {
692                         /*
693                          * Return EAGAIN if we're sure we're hitting the
694                          * socket send buffer limits.
695                          */
696                         if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
697                                 break;
698                         /*
699                          * Did we hit a memory allocation failure?
700                          */
701                         if (sent == 0) {
702                                 status = -ENOBUFS;
703                                 if (vm_wait)
704                                         break;
705                                 /* Retry, knowing now that we're below the
706                                  * socket send buffer limit
707                                  */
708                                 vm_wait = true;
709                         }
710                         continue;
711                 }
712                 if (status < 0)
713                         break;
714                 vm_wait = false;
715         }
716 
717         switch (status) {
718         case -ENOTSOCK:
719                 status = -ENOTCONN;
720                 /* Should we call xs_close() here? */
721                 break;
722         case -EAGAIN:
723                 status = xs_nospace(task);
724                 break;
725         case -ECONNRESET:
726         case -ECONNREFUSED:
727         case -ENOTCONN:
728         case -EADDRINUSE:
729         case -ENOBUFS:
730         case -EPIPE:
731                 break;
732         default:
733                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
734                         -status);
735         }
736 
737         return status;
738 }
739 
740 /**
741  * xs_tcp_release_xprt - clean up after a tcp transmission
742  * @xprt: transport
743  * @task: rpc task
744  *
745  * This cleans up if an error causes us to abort the transmission of a request.
746  * In this case, the socket may need to be reset in order to avoid confusing
747  * the server.
748  */
749 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
750 {
751         struct rpc_rqst *req;
752 
753         if (task != xprt->snd_task)
754                 return;
755         if (task == NULL)
756                 goto out_release;
757         req = task->tk_rqstp;
758         if (req == NULL)
759                 goto out_release;
760         if (req->rq_bytes_sent == 0)
761                 goto out_release;
762         if (req->rq_bytes_sent == req->rq_snd_buf.len)
763                 goto out_release;
764         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
765 out_release:
766         xprt_release_xprt(xprt, task);
767 }
768 
769 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
770 {
771         transport->old_data_ready = sk->sk_data_ready;
772         transport->old_state_change = sk->sk_state_change;
773         transport->old_write_space = sk->sk_write_space;
774         transport->old_error_report = sk->sk_error_report;
775 }
776 
777 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
778 {
779         sk->sk_data_ready = transport->old_data_ready;
780         sk->sk_state_change = transport->old_state_change;
781         sk->sk_write_space = transport->old_write_space;
782         sk->sk_error_report = transport->old_error_report;
783 }
784 
785 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
786 {
787         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
788 
789         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
790 }
791 
792 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
793 {
794         smp_mb__before_atomic();
795         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
796         clear_bit(XPRT_CLOSING, &xprt->state);
797         xs_sock_reset_state_flags(xprt);
798         smp_mb__after_atomic();
799 }
800 
801 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
802 {
803         xs_sock_reset_connection_flags(xprt);
804         /* Mark transport as closed and wake up all pending tasks */
805         xprt_disconnect_done(xprt);
806 }
807 
808 /**
809  * xs_error_report - callback to handle TCP socket state errors
810  * @sk: socket
811  *
812  * Note: we don't call sock_error() since there may be a rpc_task
813  * using the socket, and so we don't want to clear sk->sk_err.
814  */
815 static void xs_error_report(struct sock *sk)
816 {
817         struct rpc_xprt *xprt;
818         int err;
819 
820         read_lock_bh(&sk->sk_callback_lock);
821         if (!(xprt = xprt_from_sock(sk)))
822                 goto out;
823 
824         err = -sk->sk_err;
825         if (err == 0)
826                 goto out;
827         /* Is this a reset event? */
828         if (sk->sk_state == TCP_CLOSE)
829                 xs_sock_mark_closed(xprt);
830         dprintk("RPC:       xs_error_report client %p, error=%d...\n",
831                         xprt, -err);
832         trace_rpc_socket_error(xprt, sk->sk_socket, err);
833         xprt_wake_pending_tasks(xprt, err);
834  out:
835         read_unlock_bh(&sk->sk_callback_lock);
836 }
837 
838 static void xs_reset_transport(struct sock_xprt *transport)
839 {
840         struct socket *sock = transport->sock;
841         struct sock *sk = transport->inet;
842         struct rpc_xprt *xprt = &transport->xprt;
843 
844         if (sk == NULL)
845                 return;
846 
847         if (atomic_read(&transport->xprt.swapper))
848                 sk_clear_memalloc(sk);
849 
850         kernel_sock_shutdown(sock, SHUT_RDWR);
851 
852         mutex_lock(&transport->recv_mutex);
853         write_lock_bh(&sk->sk_callback_lock);
854         transport->inet = NULL;
855         transport->sock = NULL;
856 
857         sk->sk_user_data = NULL;
858 
859         xs_restore_old_callbacks(transport, sk);
860         xprt_clear_connected(xprt);
861         write_unlock_bh(&sk->sk_callback_lock);
862         xs_sock_reset_connection_flags(xprt);
863         mutex_unlock(&transport->recv_mutex);
864 
865         trace_rpc_socket_close(xprt, sock);
866         sock_release(sock);
867 }
868 
869 /**
870  * xs_close - close a socket
871  * @xprt: transport
872  *
873  * This is used when all requests are complete; ie, no DRC state remains
874  * on the server we want to save.
875  *
876  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
877  * xs_reset_transport() zeroing the socket from underneath a writer.
878  */
879 static void xs_close(struct rpc_xprt *xprt)
880 {
881         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
882 
883         dprintk("RPC:       xs_close xprt %p\n", xprt);
884 
885         xs_reset_transport(transport);
886         xprt->reestablish_timeout = 0;
887 
888         xprt_disconnect_done(xprt);
889 }
890 
891 static void xs_inject_disconnect(struct rpc_xprt *xprt)
892 {
893         dprintk("RPC:       injecting transport disconnect on xprt=%p\n",
894                 xprt);
895         xprt_disconnect_done(xprt);
896 }
897 
898 static void xs_xprt_free(struct rpc_xprt *xprt)
899 {
900         xs_free_peer_addresses(xprt);
901         xprt_free(xprt);
902 }
903 
904 /**
905  * xs_destroy - prepare to shutdown a transport
906  * @xprt: doomed transport
907  *
908  */
909 static void xs_destroy(struct rpc_xprt *xprt)
910 {
911         struct sock_xprt *transport = container_of(xprt,
912                         struct sock_xprt, xprt);
913         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
914 
915         cancel_delayed_work_sync(&transport->connect_worker);
916         xs_close(xprt);
917         cancel_work_sync(&transport->recv_worker);
918         xs_xprt_free(xprt);
919         module_put(THIS_MODULE);
920 }
921 
922 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
923 {
924         struct xdr_skb_reader desc = {
925                 .skb            = skb,
926                 .offset         = sizeof(rpc_fraghdr),
927                 .count          = skb->len - sizeof(rpc_fraghdr),
928         };
929 
930         if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
931                 return -1;
932         if (desc.count)
933                 return -1;
934         return 0;
935 }
936 
937 /**
938  * xs_local_data_read_skb
939  * @xprt: transport
940  * @sk: socket
941  * @skb: skbuff
942  *
943  * Currently this assumes we can read the whole reply in a single gulp.
944  */
945 static void xs_local_data_read_skb(struct rpc_xprt *xprt,
946                 struct sock *sk,
947                 struct sk_buff *skb)
948 {
949         struct rpc_task *task;
950         struct rpc_rqst *rovr;
951         int repsize, copied;
952         u32 _xid;
953         __be32 *xp;
954 
955         repsize = skb->len - sizeof(rpc_fraghdr);
956         if (repsize < 4) {
957                 dprintk("RPC:       impossible RPC reply size %d\n", repsize);
958                 return;
959         }
960 
961         /* Copy the XID from the skb... */
962         xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
963         if (xp == NULL)
964                 return;
965 
966         /* Look up and lock the request corresponding to the given XID */
967         spin_lock_bh(&xprt->transport_lock);
968         rovr = xprt_lookup_rqst(xprt, *xp);
969         if (!rovr)
970                 goto out_unlock;
971         task = rovr->rq_task;
972 
973         copied = rovr->rq_private_buf.buflen;
974         if (copied > repsize)
975                 copied = repsize;
976 
977         if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
978                 dprintk("RPC:       sk_buff copy failed\n");
979                 goto out_unlock;
980         }
981 
982         xprt_complete_rqst(task, copied);
983 
984  out_unlock:
985         spin_unlock_bh(&xprt->transport_lock);
986 }
987 
988 static void xs_local_data_receive(struct sock_xprt *transport)
989 {
990         struct sk_buff *skb;
991         struct sock *sk;
992         int err;
993 
994         mutex_lock(&transport->recv_mutex);
995         sk = transport->inet;
996         if (sk == NULL)
997                 goto out;
998         for (;;) {
999                 skb = skb_recv_datagram(sk, 0, 1, &err);
1000                 if (skb != NULL) {
1001                         xs_local_data_read_skb(&transport->xprt, sk, skb);
1002                         skb_free_datagram(sk, skb);
1003                         continue;
1004                 }
1005                 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1006                         break;
1007         }
1008 out:
1009         mutex_unlock(&transport->recv_mutex);
1010 }
1011 
1012 static void xs_local_data_receive_workfn(struct work_struct *work)
1013 {
1014         struct sock_xprt *transport =
1015                 container_of(work, struct sock_xprt, recv_worker);
1016         xs_local_data_receive(transport);
1017 }
1018 
1019 /**
1020  * xs_udp_data_read_skb - receive callback for UDP sockets
1021  * @xprt: transport
1022  * @sk: socket
1023  * @skb: skbuff
1024  *
1025  */
1026 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1027                 struct sock *sk,
1028                 struct sk_buff *skb)
1029 {
1030         struct rpc_task *task;
1031         struct rpc_rqst *rovr;
1032         int repsize, copied;
1033         u32 _xid;
1034         __be32 *xp;
1035 
1036         repsize = skb->len;
1037         if (repsize < 4) {
1038                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
1039                 return;
1040         }
1041 
1042         /* Copy the XID from the skb... */
1043         xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1044         if (xp == NULL)
1045                 return;
1046 
1047         /* Look up and lock the request corresponding to the given XID */
1048         spin_lock_bh(&xprt->transport_lock);
1049         rovr = xprt_lookup_rqst(xprt, *xp);
1050         if (!rovr)
1051                 goto out_unlock;
1052         task = rovr->rq_task;
1053 
1054         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1055                 copied = repsize;
1056 
1057         /* Suck it into the iovec, verify checksum if not done by hw. */
1058         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1059                 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1060                 goto out_unlock;
1061         }
1062 
1063         __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1064 
1065         xprt_adjust_cwnd(xprt, task, copied);
1066         xprt_complete_rqst(task, copied);
1067 
1068  out_unlock:
1069         spin_unlock_bh(&xprt->transport_lock);
1070 }
1071 
1072 static void xs_udp_data_receive(struct sock_xprt *transport)
1073 {
1074         struct sk_buff *skb;
1075         struct sock *sk;
1076         int err;
1077 
1078         mutex_lock(&transport->recv_mutex);
1079         sk = transport->inet;
1080         if (sk == NULL)
1081                 goto out;
1082         for (;;) {
1083                 skb = skb_recv_datagram(sk, 0, 1, &err);
1084                 if (skb != NULL) {
1085                         xs_udp_data_read_skb(&transport->xprt, sk, skb);
1086                         skb_free_datagram_locked(sk, skb);
1087                         continue;
1088                 }
1089                 if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1090                         break;
1091         }
1092 out:
1093         mutex_unlock(&transport->recv_mutex);
1094 }
1095 
1096 static void xs_udp_data_receive_workfn(struct work_struct *work)
1097 {
1098         struct sock_xprt *transport =
1099                 container_of(work, struct sock_xprt, recv_worker);
1100         xs_udp_data_receive(transport);
1101 }
1102 
1103 /**
1104  * xs_data_ready - "data ready" callback for UDP sockets
1105  * @sk: socket with data to read
1106  *
1107  */
1108 static void xs_data_ready(struct sock *sk)
1109 {
1110         struct rpc_xprt *xprt;
1111 
1112         read_lock_bh(&sk->sk_callback_lock);
1113         dprintk("RPC:       xs_data_ready...\n");
1114         xprt = xprt_from_sock(sk);
1115         if (xprt != NULL) {
1116                 struct sock_xprt *transport = container_of(xprt,
1117                                 struct sock_xprt, xprt);
1118                 transport->old_data_ready(sk);
1119                 /* Any data means we had a useful conversation, so
1120                  * then we don't need to delay the next reconnect
1121                  */
1122                 if (xprt->reestablish_timeout)
1123                         xprt->reestablish_timeout = 0;
1124                 if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1125                         queue_work(xprtiod_workqueue, &transport->recv_worker);
1126         }
1127         read_unlock_bh(&sk->sk_callback_lock);
1128 }
1129 
1130 /*
1131  * Helper function to force a TCP close if the server is sending
1132  * junk and/or it has put us in CLOSE_WAIT
1133  */
1134 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1135 {
1136         xprt_force_disconnect(xprt);
1137 }
1138 
1139 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1140 {
1141         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1142         size_t len, used;
1143         char *p;
1144 
1145         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1146         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1147         used = xdr_skb_read_bits(desc, p, len);
1148         transport->tcp_offset += used;
1149         if (used != len)
1150                 return;
1151 
1152         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1153         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1154                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1155         else
1156                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1157         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1158 
1159         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1160         transport->tcp_offset = 0;
1161 
1162         /* Sanity check of the record length */
1163         if (unlikely(transport->tcp_reclen < 8)) {
1164                 dprintk("RPC:       invalid TCP record fragment length\n");
1165                 xs_tcp_force_close(xprt);
1166                 return;
1167         }
1168         dprintk("RPC:       reading TCP record fragment of length %d\n",
1169                         transport->tcp_reclen);
1170 }
1171 
1172 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1173 {
1174         if (transport->tcp_offset == transport->tcp_reclen) {
1175                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1176                 transport->tcp_offset = 0;
1177                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1178                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1179                         transport->tcp_flags |= TCP_RCV_COPY_XID;
1180                         transport->tcp_copied = 0;
1181                 }
1182         }
1183 }
1184 
1185 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1186 {
1187         size_t len, used;
1188         char *p;
1189 
1190         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1191         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
1192         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1193         used = xdr_skb_read_bits(desc, p, len);
1194         transport->tcp_offset += used;
1195         if (used != len)
1196                 return;
1197         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1198         transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1199         transport->tcp_copied = 4;
1200         dprintk("RPC:       reading %s XID %08x\n",
1201                         (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1202                                                               : "request with",
1203                         ntohl(transport->tcp_xid));
1204         xs_tcp_check_fraghdr(transport);
1205 }
1206 
1207 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1208                                        struct xdr_skb_reader *desc)
1209 {
1210         size_t len, used;
1211         u32 offset;
1212         char *p;
1213 
1214         /*
1215          * We want transport->tcp_offset to be 8 at the end of this routine
1216          * (4 bytes for the xid and 4 bytes for the call/reply flag).
1217          * When this function is called for the first time,
1218          * transport->tcp_offset is 4 (after having already read the xid).
1219          */
1220         offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1221         len = sizeof(transport->tcp_calldir) - offset;
1222         dprintk("RPC:       reading CALL/REPLY flag (%Zu bytes)\n", len);
1223         p = ((char *) &transport->tcp_calldir) + offset;
1224         used = xdr_skb_read_bits(desc, p, len);
1225         transport->tcp_offset += used;
1226         if (used != len)
1227                 return;
1228         transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1229         /*
1230          * We don't yet have the XDR buffer, so we will write the calldir
1231          * out after we get the buffer from the 'struct rpc_rqst'
1232          */
1233         switch (ntohl(transport->tcp_calldir)) {
1234         case RPC_REPLY:
1235                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1236                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1237                 transport->tcp_flags |= TCP_RPC_REPLY;
1238                 break;
1239         case RPC_CALL:
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         default:
1245                 dprintk("RPC:       invalid request message type\n");
1246                 xs_tcp_force_close(&transport->xprt);
1247         }
1248         xs_tcp_check_fraghdr(transport);
1249 }
1250 
1251 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1252                                      struct xdr_skb_reader *desc,
1253                                      struct rpc_rqst *req)
1254 {
1255         struct sock_xprt *transport =
1256                                 container_of(xprt, struct sock_xprt, xprt);
1257         struct xdr_buf *rcvbuf;
1258         size_t len;
1259         ssize_t r;
1260 
1261         rcvbuf = &req->rq_private_buf;
1262 
1263         if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1264                 /*
1265                  * Save the RPC direction in the XDR buffer
1266                  */
1267                 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1268                         &transport->tcp_calldir,
1269                         sizeof(transport->tcp_calldir));
1270                 transport->tcp_copied += sizeof(transport->tcp_calldir);
1271                 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1272         }
1273 
1274         len = desc->count;
1275         if (len > transport->tcp_reclen - transport->tcp_offset) {
1276                 struct xdr_skb_reader my_desc;
1277 
1278                 len = transport->tcp_reclen - transport->tcp_offset;
1279                 memcpy(&my_desc, desc, sizeof(my_desc));
1280                 my_desc.count = len;
1281                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1282                                           &my_desc, xdr_skb_read_bits);
1283                 desc->count -= r;
1284                 desc->offset += r;
1285         } else
1286                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1287                                           desc, xdr_skb_read_bits);
1288 
1289         if (r > 0) {
1290                 transport->tcp_copied += r;
1291                 transport->tcp_offset += r;
1292         }
1293         if (r != len) {
1294                 /* Error when copying to the receive buffer,
1295                  * usually because we weren't able to allocate
1296                  * additional buffer pages. All we can do now
1297                  * is turn off TCP_RCV_COPY_DATA, so the request
1298                  * will not receive any additional updates,
1299                  * and time out.
1300                  * Any remaining data from this record will
1301                  * be discarded.
1302                  */
1303                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1304                 dprintk("RPC:       XID %08x truncated request\n",
1305                                 ntohl(transport->tcp_xid));
1306                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1307                                 "tcp_offset = %u, tcp_reclen = %u\n",
1308                                 xprt, transport->tcp_copied,
1309                                 transport->tcp_offset, transport->tcp_reclen);
1310                 return;
1311         }
1312 
1313         dprintk("RPC:       XID %08x read %Zd bytes\n",
1314                         ntohl(transport->tcp_xid), r);
1315         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1316                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1317                         transport->tcp_offset, transport->tcp_reclen);
1318 
1319         if (transport->tcp_copied == req->rq_private_buf.buflen)
1320                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1321         else if (transport->tcp_offset == transport->tcp_reclen) {
1322                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1323                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1324         }
1325 }
1326 
1327 /*
1328  * Finds the request corresponding to the RPC xid and invokes the common
1329  * tcp read code to read the data.
1330  */
1331 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1332                                     struct xdr_skb_reader *desc)
1333 {
1334         struct sock_xprt *transport =
1335                                 container_of(xprt, struct sock_xprt, xprt);
1336         struct rpc_rqst *req;
1337 
1338         dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1339 
1340         /* Find and lock the request corresponding to this xid */
1341         spin_lock_bh(&xprt->transport_lock);
1342         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1343         if (!req) {
1344                 dprintk("RPC:       XID %08x request not found!\n",
1345                                 ntohl(transport->tcp_xid));
1346                 spin_unlock_bh(&xprt->transport_lock);
1347                 return -1;
1348         }
1349 
1350         xs_tcp_read_common(xprt, desc, req);
1351 
1352         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1353                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1354 
1355         spin_unlock_bh(&xprt->transport_lock);
1356         return 0;
1357 }
1358 
1359 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1360 /*
1361  * Obtains an rpc_rqst previously allocated and invokes the common
1362  * tcp read code to read the data.  The result is placed in the callback
1363  * queue.
1364  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1365  * connection and return -1.
1366  */
1367 static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1368                                        struct xdr_skb_reader *desc)
1369 {
1370         struct sock_xprt *transport =
1371                                 container_of(xprt, struct sock_xprt, xprt);
1372         struct rpc_rqst *req;
1373 
1374         /* Look up and lock the request corresponding to the given XID */
1375         spin_lock_bh(&xprt->transport_lock);
1376         req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1377         if (req == NULL) {
1378                 spin_unlock_bh(&xprt->transport_lock);
1379                 printk(KERN_WARNING "Callback slot table overflowed\n");
1380                 xprt_force_disconnect(xprt);
1381                 return -1;
1382         }
1383 
1384         dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1385         xs_tcp_read_common(xprt, desc, req);
1386 
1387         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1388                 xprt_complete_bc_request(req, transport->tcp_copied);
1389         spin_unlock_bh(&xprt->transport_lock);
1390 
1391         return 0;
1392 }
1393 
1394 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1395                                         struct xdr_skb_reader *desc)
1396 {
1397         struct sock_xprt *transport =
1398                                 container_of(xprt, struct sock_xprt, xprt);
1399 
1400         return (transport->tcp_flags & TCP_RPC_REPLY) ?
1401                 xs_tcp_read_reply(xprt, desc) :
1402                 xs_tcp_read_callback(xprt, desc);
1403 }
1404 
1405 static int xs_tcp_bc_up(struct svc_serv *serv, struct net *net)
1406 {
1407         int ret;
1408 
1409         ret = svc_create_xprt(serv, "tcp-bc", net, PF_INET, 0,
1410                               SVC_SOCK_ANONYMOUS);
1411         if (ret < 0)
1412                 return ret;
1413         return 0;
1414 }
1415 
1416 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1417 {
1418         return PAGE_SIZE;
1419 }
1420 #else
1421 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1422                                         struct xdr_skb_reader *desc)
1423 {
1424         return xs_tcp_read_reply(xprt, desc);
1425 }
1426 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1427 
1428 /*
1429  * Read data off the transport.  This can be either an RPC_CALL or an
1430  * RPC_REPLY.  Relay the processing to helper functions.
1431  */
1432 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1433                                     struct xdr_skb_reader *desc)
1434 {
1435         struct sock_xprt *transport =
1436                                 container_of(xprt, struct sock_xprt, xprt);
1437 
1438         if (_xs_tcp_read_data(xprt, desc) == 0)
1439                 xs_tcp_check_fraghdr(transport);
1440         else {
1441                 /*
1442                  * The transport_lock protects the request handling.
1443                  * There's no need to hold it to update the tcp_flags.
1444                  */
1445                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1446         }
1447 }
1448 
1449 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1450 {
1451         size_t len;
1452 
1453         len = transport->tcp_reclen - transport->tcp_offset;
1454         if (len > desc->count)
1455                 len = desc->count;
1456         desc->count -= len;
1457         desc->offset += len;
1458         transport->tcp_offset += len;
1459         dprintk("RPC:       discarded %Zu bytes\n", len);
1460         xs_tcp_check_fraghdr(transport);
1461 }
1462 
1463 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1464 {
1465         struct rpc_xprt *xprt = rd_desc->arg.data;
1466         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1467         struct xdr_skb_reader desc = {
1468                 .skb    = skb,
1469                 .offset = offset,
1470                 .count  = len,
1471         };
1472 
1473         dprintk("RPC:       xs_tcp_data_recv started\n");
1474         do {
1475                 trace_xs_tcp_data_recv(transport);
1476                 /* Read in a new fragment marker if necessary */
1477                 /* Can we ever really expect to get completely empty fragments? */
1478                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1479                         xs_tcp_read_fraghdr(xprt, &desc);
1480                         continue;
1481                 }
1482                 /* Read in the xid if necessary */
1483                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1484                         xs_tcp_read_xid(transport, &desc);
1485                         continue;
1486                 }
1487                 /* Read in the call/reply flag */
1488                 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1489                         xs_tcp_read_calldir(transport, &desc);
1490                         continue;
1491                 }
1492                 /* Read in the request data */
1493                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1494                         xs_tcp_read_data(xprt, &desc);
1495                         continue;
1496                 }
1497                 /* Skip over any trailing bytes on short reads */
1498                 xs_tcp_read_discard(transport, &desc);
1499         } while (desc.count);
1500         trace_xs_tcp_data_recv(transport);
1501         dprintk("RPC:       xs_tcp_data_recv done\n");
1502         return len - desc.count;
1503 }
1504 
1505 static void xs_tcp_data_receive(struct sock_xprt *transport)
1506 {
1507         struct rpc_xprt *xprt = &transport->xprt;
1508         struct sock *sk;
1509         read_descriptor_t rd_desc = {
1510                 .count = 2*1024*1024,
1511                 .arg.data = xprt,
1512         };
1513         unsigned long total = 0;
1514         int read = 0;
1515 
1516         mutex_lock(&transport->recv_mutex);
1517         sk = transport->inet;
1518         if (sk == NULL)
1519                 goto out;
1520 
1521         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1522         for (;;) {
1523                 lock_sock(sk);
1524                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1525                 if (read <= 0) {
1526                         clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1527                         release_sock(sk);
1528                         if (!test_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1529                                 break;
1530                 } else {
1531                         release_sock(sk);
1532                         total += read;
1533                 }
1534                 rd_desc.count = 65536;
1535         }
1536 out:
1537         mutex_unlock(&transport->recv_mutex);
1538         trace_xs_tcp_data_ready(xprt, read, total);
1539 }
1540 
1541 static void xs_tcp_data_receive_workfn(struct work_struct *work)
1542 {
1543         struct sock_xprt *transport =
1544                 container_of(work, struct sock_xprt, recv_worker);
1545         xs_tcp_data_receive(transport);
1546 }
1547 
1548 /**
1549  * xs_tcp_state_change - callback to handle TCP socket state changes
1550  * @sk: socket whose state has changed
1551  *
1552  */
1553 static void xs_tcp_state_change(struct sock *sk)
1554 {
1555         struct rpc_xprt *xprt;
1556         struct sock_xprt *transport;
1557 
1558         read_lock_bh(&sk->sk_callback_lock);
1559         if (!(xprt = xprt_from_sock(sk)))
1560                 goto out;
1561         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1562         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1563                         sk->sk_state, xprt_connected(xprt),
1564                         sock_flag(sk, SOCK_DEAD),
1565                         sock_flag(sk, SOCK_ZAPPED),
1566                         sk->sk_shutdown);
1567 
1568         transport = container_of(xprt, struct sock_xprt, xprt);
1569         trace_rpc_socket_state_change(xprt, sk->sk_socket);
1570         switch (sk->sk_state) {
1571         case TCP_ESTABLISHED:
1572                 spin_lock(&xprt->transport_lock);
1573                 if (!xprt_test_and_set_connected(xprt)) {
1574 
1575                         /* Reset TCP record info */
1576                         transport->tcp_offset = 0;
1577                         transport->tcp_reclen = 0;
1578                         transport->tcp_copied = 0;
1579                         transport->tcp_flags =
1580                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1581                         xprt->connect_cookie++;
1582                         clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1583                         xprt_clear_connecting(xprt);
1584 
1585                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1586                 }
1587                 spin_unlock(&xprt->transport_lock);
1588                 break;
1589         case TCP_FIN_WAIT1:
1590                 /* The client initiated a shutdown of the socket */
1591                 xprt->connect_cookie++;
1592                 xprt->reestablish_timeout = 0;
1593                 set_bit(XPRT_CLOSING, &xprt->state);
1594                 smp_mb__before_atomic();
1595                 clear_bit(XPRT_CONNECTED, &xprt->state);
1596                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1597                 smp_mb__after_atomic();
1598                 break;
1599         case TCP_CLOSE_WAIT:
1600                 /* The server initiated a shutdown of the socket */
1601                 xprt->connect_cookie++;
1602                 clear_bit(XPRT_CONNECTED, &xprt->state);
1603                 xs_tcp_force_close(xprt);
1604         case TCP_CLOSING:
1605                 /*
1606                  * If the server closed down the connection, make sure that
1607                  * we back off before reconnecting
1608                  */
1609                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1610                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1611                 break;
1612         case TCP_LAST_ACK:
1613                 set_bit(XPRT_CLOSING, &xprt->state);
1614                 smp_mb__before_atomic();
1615                 clear_bit(XPRT_CONNECTED, &xprt->state);
1616                 smp_mb__after_atomic();
1617                 break;
1618         case TCP_CLOSE:
1619                 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1620                                         &transport->sock_state))
1621                         xprt_clear_connecting(xprt);
1622                 xs_sock_mark_closed(xprt);
1623         }
1624  out:
1625         read_unlock_bh(&sk->sk_callback_lock);
1626 }
1627 
1628 static void xs_write_space(struct sock *sk)
1629 {
1630         struct socket_wq *wq;
1631         struct rpc_xprt *xprt;
1632 
1633         if (!sk->sk_socket)
1634                 return;
1635         clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1636 
1637         if (unlikely(!(xprt = xprt_from_sock(sk))))
1638                 return;
1639         rcu_read_lock();
1640         wq = rcu_dereference(sk->sk_wq);
1641         if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1642                 goto out;
1643 
1644         xprt_write_space(xprt);
1645 out:
1646         rcu_read_unlock();
1647 }
1648 
1649 /**
1650  * xs_udp_write_space - callback invoked when socket buffer space
1651  *                             becomes available
1652  * @sk: socket whose state has changed
1653  *
1654  * Called when more output buffer space is available for this socket.
1655  * We try not to wake our writers until they can make "significant"
1656  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1657  * with a bunch of small requests.
1658  */
1659 static void xs_udp_write_space(struct sock *sk)
1660 {
1661         read_lock_bh(&sk->sk_callback_lock);
1662 
1663         /* from net/core/sock.c:sock_def_write_space */
1664         if (sock_writeable(sk))
1665                 xs_write_space(sk);
1666 
1667         read_unlock_bh(&sk->sk_callback_lock);
1668 }
1669 
1670 /**
1671  * xs_tcp_write_space - callback invoked when socket buffer space
1672  *                             becomes available
1673  * @sk: socket whose state has changed
1674  *
1675  * Called when more output buffer space is available for this socket.
1676  * We try not to wake our writers until they can make "significant"
1677  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1678  * with a bunch of small requests.
1679  */
1680 static void xs_tcp_write_space(struct sock *sk)
1681 {
1682         read_lock_bh(&sk->sk_callback_lock);
1683 
1684         /* from net/core/stream.c:sk_stream_write_space */
1685         if (sk_stream_is_writeable(sk))
1686                 xs_write_space(sk);
1687 
1688         read_unlock_bh(&sk->sk_callback_lock);
1689 }
1690 
1691 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1692 {
1693         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1694         struct sock *sk = transport->inet;
1695 
1696         if (transport->rcvsize) {
1697                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1698                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1699         }
1700         if (transport->sndsize) {
1701                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1702                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1703                 sk->sk_write_space(sk);
1704         }
1705 }
1706 
1707 /**
1708  * xs_udp_set_buffer_size - set send and receive limits
1709  * @xprt: generic transport
1710  * @sndsize: requested size of send buffer, in bytes
1711  * @rcvsize: requested size of receive buffer, in bytes
1712  *
1713  * Set socket send and receive buffer size limits.
1714  */
1715 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1716 {
1717         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1718 
1719         transport->sndsize = 0;
1720         if (sndsize)
1721                 transport->sndsize = sndsize + 1024;
1722         transport->rcvsize = 0;
1723         if (rcvsize)
1724                 transport->rcvsize = rcvsize + 1024;
1725 
1726         xs_udp_do_set_buffer_size(xprt);
1727 }
1728 
1729 /**
1730  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1731  * @task: task that timed out
1732  *
1733  * Adjust the congestion window after a retransmit timeout has occurred.
1734  */
1735 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1736 {
1737         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1738 }
1739 
1740 static unsigned short xs_get_random_port(void)
1741 {
1742         unsigned short range = xprt_max_resvport - xprt_min_resvport + 1;
1743         unsigned short rand = (unsigned short) prandom_u32() % range;
1744         return rand + xprt_min_resvport;
1745 }
1746 
1747 /**
1748  * xs_set_reuseaddr_port - set the socket's port and address reuse options
1749  * @sock: socket
1750  *
1751  * Note that this function has to be called on all sockets that share the
1752  * same port, and it must be called before binding.
1753  */
1754 static void xs_sock_set_reuseport(struct socket *sock)
1755 {
1756         int opt = 1;
1757 
1758         kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1759                         (char *)&opt, sizeof(opt));
1760 }
1761 
1762 static unsigned short xs_sock_getport(struct socket *sock)
1763 {
1764         struct sockaddr_storage buf;
1765         int buflen;
1766         unsigned short port = 0;
1767 
1768         if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1769                 goto out;
1770         switch (buf.ss_family) {
1771         case AF_INET6:
1772                 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1773                 break;
1774         case AF_INET:
1775                 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1776         }
1777 out:
1778         return port;
1779 }
1780 
1781 /**
1782  * xs_set_port - reset the port number in the remote endpoint address
1783  * @xprt: generic transport
1784  * @port: new port number
1785  *
1786  */
1787 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1788 {
1789         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1790 
1791         rpc_set_port(xs_addr(xprt), port);
1792         xs_update_peer_port(xprt);
1793 }
1794 
1795 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1796 {
1797         if (transport->srcport == 0)
1798                 transport->srcport = xs_sock_getport(sock);
1799 }
1800 
1801 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1802 {
1803         unsigned short port = transport->srcport;
1804 
1805         if (port == 0 && transport->xprt.resvport)
1806                 port = xs_get_random_port();
1807         return port;
1808 }
1809 
1810 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1811 {
1812         if (transport->srcport != 0)
1813                 transport->srcport = 0;
1814         if (!transport->xprt.resvport)
1815                 return 0;
1816         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1817                 return xprt_max_resvport;
1818         return --port;
1819 }
1820 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1821 {
1822         struct sockaddr_storage myaddr;
1823         int err, nloop = 0;
1824         unsigned short port = xs_get_srcport(transport);
1825         unsigned short last;
1826 
1827         /*
1828          * If we are asking for any ephemeral port (i.e. port == 0 &&
1829          * transport->xprt.resvport == 0), don't bind.  Let the local
1830          * port selection happen implicitly when the socket is used
1831          * (for example at connect time).
1832          *
1833          * This ensures that we can continue to establish TCP
1834          * connections even when all local ephemeral ports are already
1835          * a part of some TCP connection.  This makes no difference
1836          * for UDP sockets, but also doens't harm them.
1837          *
1838          * If we're asking for any reserved port (i.e. port == 0 &&
1839          * transport->xprt.resvport == 1) xs_get_srcport above will
1840          * ensure that port is non-zero and we will bind as needed.
1841          */
1842         if (port == 0)
1843                 return 0;
1844 
1845         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1846         do {
1847                 rpc_set_port((struct sockaddr *)&myaddr, port);
1848                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1849                                 transport->xprt.addrlen);
1850                 if (err == 0) {
1851                         transport->srcport = port;
1852                         break;
1853                 }
1854                 last = port;
1855                 port = xs_next_srcport(transport, port);
1856                 if (port > last)
1857                         nloop++;
1858         } while (err == -EADDRINUSE && nloop != 2);
1859 
1860         if (myaddr.ss_family == AF_INET)
1861                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1862                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1863                                 port, err ? "failed" : "ok", err);
1864         else
1865                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1866                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1867                                 port, err ? "failed" : "ok", err);
1868         return err;
1869 }
1870 
1871 /*
1872  * We don't support autobind on AF_LOCAL sockets
1873  */
1874 static void xs_local_rpcbind(struct rpc_task *task)
1875 {
1876         xprt_set_bound(task->tk_xprt);
1877 }
1878 
1879 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1880 {
1881 }
1882 
1883 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1884 static struct lock_class_key xs_key[2];
1885 static struct lock_class_key xs_slock_key[2];
1886 
1887 static inline void xs_reclassify_socketu(struct socket *sock)
1888 {
1889         struct sock *sk = sock->sk;
1890 
1891         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1892                 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1893 }
1894 
1895 static inline void xs_reclassify_socket4(struct socket *sock)
1896 {
1897         struct sock *sk = sock->sk;
1898 
1899         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1900                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1901 }
1902 
1903 static inline void xs_reclassify_socket6(struct socket *sock)
1904 {
1905         struct sock *sk = sock->sk;
1906 
1907         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1908                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1909 }
1910 
1911 static inline void xs_reclassify_socket(int family, struct socket *sock)
1912 {
1913         if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1914                 return;
1915 
1916         switch (family) {
1917         case AF_LOCAL:
1918                 xs_reclassify_socketu(sock);
1919                 break;
1920         case AF_INET:
1921                 xs_reclassify_socket4(sock);
1922                 break;
1923         case AF_INET6:
1924                 xs_reclassify_socket6(sock);
1925                 break;
1926         }
1927 }
1928 #else
1929 static inline void xs_reclassify_socket(int family, struct socket *sock)
1930 {
1931 }
1932 #endif
1933 
1934 static void xs_dummy_setup_socket(struct work_struct *work)
1935 {
1936 }
1937 
1938 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1939                 struct sock_xprt *transport, int family, int type,
1940                 int protocol, bool reuseport)
1941 {
1942         struct socket *sock;
1943         int err;
1944 
1945         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1946         if (err < 0) {
1947                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1948                                 protocol, -err);
1949                 goto out;
1950         }
1951         xs_reclassify_socket(family, sock);
1952 
1953         if (reuseport)
1954                 xs_sock_set_reuseport(sock);
1955 
1956         err = xs_bind(transport, sock);
1957         if (err) {
1958                 sock_release(sock);
1959                 goto out;
1960         }
1961 
1962         return sock;
1963 out:
1964         return ERR_PTR(err);
1965 }
1966 
1967 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1968                                       struct socket *sock)
1969 {
1970         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1971                                                                         xprt);
1972 
1973         if (!transport->inet) {
1974                 struct sock *sk = sock->sk;
1975 
1976                 write_lock_bh(&sk->sk_callback_lock);
1977 
1978                 xs_save_old_callbacks(transport, sk);
1979 
1980                 sk->sk_user_data = xprt;
1981                 sk->sk_data_ready = xs_data_ready;
1982                 sk->sk_write_space = xs_udp_write_space;
1983                 sock_set_flag(sk, SOCK_FASYNC);
1984                 sk->sk_error_report = xs_error_report;
1985                 sk->sk_allocation = GFP_NOIO;
1986 
1987                 xprt_clear_connected(xprt);
1988 
1989                 /* Reset to new socket */
1990                 transport->sock = sock;
1991                 transport->inet = sk;
1992 
1993                 write_unlock_bh(&sk->sk_callback_lock);
1994         }
1995 
1996         /* Tell the socket layer to start connecting... */
1997         xprt->stat.connect_count++;
1998         xprt->stat.connect_start = jiffies;
1999         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
2000 }
2001 
2002 /**
2003  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
2004  * @transport: socket transport to connect
2005  */
2006 static int xs_local_setup_socket(struct sock_xprt *transport)
2007 {
2008         struct rpc_xprt *xprt = &transport->xprt;
2009         struct socket *sock;
2010         int status = -EIO;
2011 
2012         status = __sock_create(xprt->xprt_net, AF_LOCAL,
2013                                         SOCK_STREAM, 0, &sock, 1);
2014         if (status < 0) {
2015                 dprintk("RPC:       can't create AF_LOCAL "
2016                         "transport socket (%d).\n", -status);
2017                 goto out;
2018         }
2019         xs_reclassify_socket(AF_LOCAL, sock);
2020 
2021         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
2022                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2023 
2024         status = xs_local_finish_connecting(xprt, sock);
2025         trace_rpc_socket_connect(xprt, sock, status);
2026         switch (status) {
2027         case 0:
2028                 dprintk("RPC:       xprt %p connected to %s\n",
2029                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2030                 xprt_set_connected(xprt);
2031         case -ENOBUFS:
2032                 break;
2033         case -ENOENT:
2034                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
2035                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2036                 break;
2037         case -ECONNREFUSED:
2038                 dprintk("RPC:       xprt %p: connection refused for %s\n",
2039                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2040                 break;
2041         default:
2042                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2043                                 __func__, -status,
2044                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
2045         }
2046 
2047 out:
2048         xprt_clear_connecting(xprt);
2049         xprt_wake_pending_tasks(xprt, status);
2050         return status;
2051 }
2052 
2053 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2054 {
2055         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2056         int ret;
2057 
2058          if (RPC_IS_ASYNC(task)) {
2059                 /*
2060                  * We want the AF_LOCAL connect to be resolved in the
2061                  * filesystem namespace of the process making the rpc
2062                  * call.  Thus we connect synchronously.
2063                  *
2064                  * If we want to support asynchronous AF_LOCAL calls,
2065                  * we'll need to figure out how to pass a namespace to
2066                  * connect.
2067                  */
2068                 rpc_exit(task, -ENOTCONN);
2069                 return;
2070         }
2071         ret = xs_local_setup_socket(transport);
2072         if (ret && !RPC_IS_SOFTCONN(task))
2073                 msleep_interruptible(15000);
2074 }
2075 
2076 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2077 /*
2078  * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2079  * know that we have exclusive access to the socket), to guard against
2080  * races with xs_reset_transport.
2081  */
2082 static void xs_set_memalloc(struct rpc_xprt *xprt)
2083 {
2084         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2085                         xprt);
2086 
2087         /*
2088          * If there's no sock, then we have nothing to set. The
2089          * reconnecting process will get it for us.
2090          */
2091         if (!transport->inet)
2092                 return;
2093         if (atomic_read(&xprt->swapper))
2094                 sk_set_memalloc(transport->inet);
2095 }
2096 
2097 /**
2098  * xs_enable_swap - Tag this transport as being used for swap.
2099  * @xprt: transport to tag
2100  *
2101  * Take a reference to this transport on behalf of the rpc_clnt, and
2102  * optionally mark it for swapping if it wasn't already.
2103  */
2104 static int
2105 xs_enable_swap(struct rpc_xprt *xprt)
2106 {
2107         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2108 
2109         if (atomic_inc_return(&xprt->swapper) != 1)
2110                 return 0;
2111         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2112                 return -ERESTARTSYS;
2113         if (xs->inet)
2114                 sk_set_memalloc(xs->inet);
2115         xprt_release_xprt(xprt, NULL);
2116         return 0;
2117 }
2118 
2119 /**
2120  * xs_disable_swap - Untag this transport as being used for swap.
2121  * @xprt: transport to tag
2122  *
2123  * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2124  * swapper refcount goes to 0, untag the socket as a memalloc socket.
2125  */
2126 static void
2127 xs_disable_swap(struct rpc_xprt *xprt)
2128 {
2129         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2130 
2131         if (!atomic_dec_and_test(&xprt->swapper))
2132                 return;
2133         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2134                 return;
2135         if (xs->inet)
2136                 sk_clear_memalloc(xs->inet);
2137         xprt_release_xprt(xprt, NULL);
2138 }
2139 #else
2140 static void xs_set_memalloc(struct rpc_xprt *xprt)
2141 {
2142 }
2143 
2144 static int
2145 xs_enable_swap(struct rpc_xprt *xprt)
2146 {
2147         return -EINVAL;
2148 }
2149 
2150 static void
2151 xs_disable_swap(struct rpc_xprt *xprt)
2152 {
2153 }
2154 #endif
2155 
2156 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2157 {
2158         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2159 
2160         if (!transport->inet) {
2161                 struct sock *sk = sock->sk;
2162 
2163                 write_lock_bh(&sk->sk_callback_lock);
2164 
2165                 xs_save_old_callbacks(transport, sk);
2166 
2167                 sk->sk_user_data = xprt;
2168                 sk->sk_data_ready = xs_data_ready;
2169                 sk->sk_write_space = xs_udp_write_space;
2170                 sock_set_flag(sk, SOCK_FASYNC);
2171                 sk->sk_allocation = GFP_NOIO;
2172 
2173                 xprt_set_connected(xprt);
2174 
2175                 /* Reset to new socket */
2176                 transport->sock = sock;
2177                 transport->inet = sk;
2178 
2179                 xs_set_memalloc(xprt);
2180 
2181                 write_unlock_bh(&sk->sk_callback_lock);
2182         }
2183         xs_udp_do_set_buffer_size(xprt);
2184 
2185         xprt->stat.connect_start = jiffies;
2186 }
2187 
2188 static void xs_udp_setup_socket(struct work_struct *work)
2189 {
2190         struct sock_xprt *transport =
2191                 container_of(work, struct sock_xprt, connect_worker.work);
2192         struct rpc_xprt *xprt = &transport->xprt;
2193         struct socket *sock = transport->sock;
2194         int status = -EIO;
2195 
2196         sock = xs_create_sock(xprt, transport,
2197                         xs_addr(xprt)->sa_family, SOCK_DGRAM,
2198                         IPPROTO_UDP, false);
2199         if (IS_ERR(sock))
2200                 goto out;
2201 
2202         dprintk("RPC:       worker connecting xprt %p via %s to "
2203                                 "%s (port %s)\n", xprt,
2204                         xprt->address_strings[RPC_DISPLAY_PROTO],
2205                         xprt->address_strings[RPC_DISPLAY_ADDR],
2206                         xprt->address_strings[RPC_DISPLAY_PORT]);
2207 
2208         xs_udp_finish_connecting(xprt, sock);
2209         trace_rpc_socket_connect(xprt, sock, 0);
2210         status = 0;
2211 out:
2212         xprt_unlock_connect(xprt, transport);
2213         xprt_clear_connecting(xprt);
2214         xprt_wake_pending_tasks(xprt, status);
2215 }
2216 
2217 /**
2218  * xs_tcp_shutdown - gracefully shut down a TCP socket
2219  * @xprt: transport
2220  *
2221  * Initiates a graceful shutdown of the TCP socket by calling the
2222  * equivalent of shutdown(SHUT_RDWR);
2223  */
2224 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2225 {
2226         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2227         struct socket *sock = transport->sock;
2228 
2229         if (sock == NULL)
2230                 return;
2231         if (xprt_connected(xprt)) {
2232                 kernel_sock_shutdown(sock, SHUT_RDWR);
2233                 trace_rpc_socket_shutdown(xprt, sock);
2234         } else
2235                 xs_reset_transport(transport);
2236 }
2237 
2238 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2239 {
2240         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2241         int ret = -ENOTCONN;
2242 
2243         if (!transport->inet) {
2244                 struct sock *sk = sock->sk;
2245                 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2246                 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2247                 unsigned int opt_on = 1;
2248                 unsigned int timeo;
2249                 unsigned int addr_pref = IPV6_PREFER_SRC_PUBLIC;
2250 
2251                 /* TCP Keepalive options */
2252                 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2253                                 (char *)&opt_on, sizeof(opt_on));
2254                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2255                                 (char *)&keepidle, sizeof(keepidle));
2256                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2257                                 (char *)&keepidle, sizeof(keepidle));
2258                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2259                                 (char *)&keepcnt, sizeof(keepcnt));
2260 
2261                 /* Avoid temporary address, they are bad for long-lived
2262                  * connections such as NFS mounts.
2263                  * RFC4941, section 3.6 suggests that:
2264                  *    Individual applications, which have specific
2265                  *    knowledge about the normal duration of connections,
2266                  *    MAY override this as appropriate.
2267                  */
2268                 kernel_setsockopt(sock, SOL_IPV6, IPV6_ADDR_PREFERENCES,
2269                                 (char *)&addr_pref, sizeof(addr_pref));
2270 
2271                 /* TCP user timeout (see RFC5482) */
2272                 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2273                         (xprt->timeout->to_retries + 1);
2274                 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2275                                 (char *)&timeo, sizeof(timeo));
2276 
2277                 write_lock_bh(&sk->sk_callback_lock);
2278 
2279                 xs_save_old_callbacks(transport, sk);
2280 
2281                 sk->sk_user_data = xprt;
2282                 sk->sk_data_ready = xs_data_ready;
2283                 sk->sk_state_change = xs_tcp_state_change;
2284                 sk->sk_write_space = xs_tcp_write_space;
2285                 sock_set_flag(sk, SOCK_FASYNC);
2286                 sk->sk_error_report = xs_error_report;
2287                 sk->sk_allocation = GFP_NOIO;
2288 
2289                 /* socket options */
2290                 sock_reset_flag(sk, SOCK_LINGER);
2291                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2292 
2293                 xprt_clear_connected(xprt);
2294 
2295                 /* Reset to new socket */
2296                 transport->sock = sock;
2297                 transport->inet = sk;
2298 
2299                 write_unlock_bh(&sk->sk_callback_lock);
2300         }
2301 
2302         if (!xprt_bound(xprt))
2303                 goto out;
2304 
2305         xs_set_memalloc(xprt);
2306 
2307         /* Tell the socket layer to start connecting... */
2308         xprt->stat.connect_count++;
2309         xprt->stat.connect_start = jiffies;
2310         set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2311         ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2312         switch (ret) {
2313         case 0:
2314                 xs_set_srcport(transport, sock);
2315         case -EINPROGRESS:
2316                 /* SYN_SENT! */
2317                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2318                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2319                 break;
2320         case -EADDRNOTAVAIL:
2321                 /* Source port number is unavailable. Try a new one! */
2322                 transport->srcport = 0;
2323         }
2324 out:
2325         return ret;
2326 }
2327 
2328 /**
2329  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2330  *
2331  * Invoked by a work queue tasklet.
2332  */
2333 static void xs_tcp_setup_socket(struct work_struct *work)
2334 {
2335         struct sock_xprt *transport =
2336                 container_of(work, struct sock_xprt, connect_worker.work);
2337         struct socket *sock = transport->sock;
2338         struct rpc_xprt *xprt = &transport->xprt;
2339         int status = -EIO;
2340 
2341         if (!sock) {
2342                 sock = xs_create_sock(xprt, transport,
2343                                 xs_addr(xprt)->sa_family, SOCK_STREAM,
2344                                 IPPROTO_TCP, true);
2345                 if (IS_ERR(sock)) {
2346                         status = PTR_ERR(sock);
2347                         goto out;
2348                 }
2349         }
2350 
2351         dprintk("RPC:       worker connecting xprt %p via %s to "
2352                                 "%s (port %s)\n", xprt,
2353                         xprt->address_strings[RPC_DISPLAY_PROTO],
2354                         xprt->address_strings[RPC_DISPLAY_ADDR],
2355                         xprt->address_strings[RPC_DISPLAY_PORT]);
2356 
2357         status = xs_tcp_finish_connecting(xprt, sock);
2358         trace_rpc_socket_connect(xprt, sock, status);
2359         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2360                         xprt, -status, xprt_connected(xprt),
2361                         sock->sk->sk_state);
2362         switch (status) {
2363         default:
2364                 printk("%s: connect returned unhandled error %d\n",
2365                         __func__, status);
2366         case -EADDRNOTAVAIL:
2367                 /* We're probably in TIME_WAIT. Get rid of existing socket,
2368                  * and retry
2369                  */
2370                 xs_tcp_force_close(xprt);
2371                 break;
2372         case 0:
2373         case -EINPROGRESS:
2374         case -EALREADY:
2375                 xprt_unlock_connect(xprt, transport);
2376                 return;
2377         case -EINVAL:
2378                 /* Happens, for instance, if the user specified a link
2379                  * local IPv6 address without a scope-id.
2380                  */
2381         case -ECONNREFUSED:
2382         case -ECONNRESET:
2383         case -ENETUNREACH:
2384         case -EADDRINUSE:
2385         case -ENOBUFS:
2386                 /* retry with existing socket, after a delay */
2387                 xs_tcp_force_close(xprt);
2388                 goto out;
2389         }
2390         status = -EAGAIN;
2391 out:
2392         xprt_unlock_connect(xprt, transport);
2393         xprt_clear_connecting(xprt);
2394         xprt_wake_pending_tasks(xprt, status);
2395 }
2396 
2397 static unsigned long xs_reconnect_delay(const struct rpc_xprt *xprt)
2398 {
2399         unsigned long start, now = jiffies;
2400 
2401         start = xprt->stat.connect_start + xprt->reestablish_timeout;
2402         if (time_after(start, now))
2403                 return start - now;
2404         return 0;
2405 }
2406 
2407 static void xs_reconnect_backoff(struct rpc_xprt *xprt)
2408 {
2409         xprt->reestablish_timeout <<= 1;
2410         if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
2411                 xprt->reestablish_timeout = xprt->max_reconnect_timeout;
2412         if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2413                 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2414 }
2415 
2416 /**
2417  * xs_connect - connect a socket to a remote endpoint
2418  * @xprt: pointer to transport structure
2419  * @task: address of RPC task that manages state of connect request
2420  *
2421  * TCP: If the remote end dropped the connection, delay reconnecting.
2422  *
2423  * UDP socket connects are synchronous, but we use a work queue anyway
2424  * to guarantee that even unprivileged user processes can set up a
2425  * socket on a privileged port.
2426  *
2427  * If a UDP socket connect fails, the delay behavior here prevents
2428  * retry floods (hard mounts).
2429  */
2430 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2431 {
2432         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2433         unsigned long delay = 0;
2434 
2435         WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2436 
2437         if (transport->sock != NULL) {
2438                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2439                                 "seconds\n",
2440                                 xprt, xprt->reestablish_timeout / HZ);
2441 
2442                 /* Start by resetting any existing state */
2443                 xs_reset_transport(transport);
2444 
2445                 delay = xs_reconnect_delay(xprt);
2446                 xs_reconnect_backoff(xprt);
2447 
2448         } else
2449                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2450 
2451         queue_delayed_work(xprtiod_workqueue,
2452                         &transport->connect_worker,
2453                         delay);
2454 }
2455 
2456 /**
2457  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2458  * @xprt: rpc_xprt struct containing statistics
2459  * @seq: output file
2460  *
2461  */
2462 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2463 {
2464         long idle_time = 0;
2465 
2466         if (xprt_connected(xprt))
2467                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2468 
2469         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2470                         "%llu %llu %lu %llu %llu\n",
2471                         xprt->stat.bind_count,
2472                         xprt->stat.connect_count,
2473                         xprt->stat.connect_time,
2474                         idle_time,
2475                         xprt->stat.sends,
2476                         xprt->stat.recvs,
2477                         xprt->stat.bad_xids,
2478                         xprt->stat.req_u,
2479                         xprt->stat.bklog_u,
2480                         xprt->stat.max_slots,
2481                         xprt->stat.sending_u,
2482                         xprt->stat.pending_u);
2483 }
2484 
2485 /**
2486  * xs_udp_print_stats - display UDP socket-specifc stats
2487  * @xprt: rpc_xprt struct containing statistics
2488  * @seq: output file
2489  *
2490  */
2491 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2492 {
2493         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2494 
2495         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2496                         "%lu %llu %llu\n",
2497                         transport->srcport,
2498                         xprt->stat.bind_count,
2499                         xprt->stat.sends,
2500                         xprt->stat.recvs,
2501                         xprt->stat.bad_xids,
2502                         xprt->stat.req_u,
2503                         xprt->stat.bklog_u,
2504                         xprt->stat.max_slots,
2505                         xprt->stat.sending_u,
2506                         xprt->stat.pending_u);
2507 }
2508 
2509 /**
2510  * xs_tcp_print_stats - display TCP socket-specifc stats
2511  * @xprt: rpc_xprt struct containing statistics
2512  * @seq: output file
2513  *
2514  */
2515 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2516 {
2517         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2518         long idle_time = 0;
2519 
2520         if (xprt_connected(xprt))
2521                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2522 
2523         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2524                         "%llu %llu %lu %llu %llu\n",
2525                         transport->srcport,
2526                         xprt->stat.bind_count,
2527                         xprt->stat.connect_count,
2528                         xprt->stat.connect_time,
2529                         idle_time,
2530                         xprt->stat.sends,
2531                         xprt->stat.recvs,
2532                         xprt->stat.bad_xids,
2533                         xprt->stat.req_u,
2534                         xprt->stat.bklog_u,
2535                         xprt->stat.max_slots,
2536                         xprt->stat.sending_u,
2537                         xprt->stat.pending_u);
2538 }
2539 
2540 /*
2541  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2542  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2543  * to use the server side send routines.
2544  */
2545 static void *bc_malloc(struct rpc_task *task, size_t size)
2546 {
2547         struct page *page;
2548         struct rpc_buffer *buf;
2549 
2550         WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2551         if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2552                 return NULL;
2553 
2554         page = alloc_page(GFP_KERNEL);
2555         if (!page)
2556                 return NULL;
2557 
2558         buf = page_address(page);
2559         buf->len = PAGE_SIZE;
2560 
2561         return buf->data;
2562 }
2563 
2564 /*
2565  * Free the space allocated in the bc_alloc routine
2566  */
2567 static void bc_free(void *buffer)
2568 {
2569         struct rpc_buffer *buf;
2570 
2571         if (!buffer)
2572                 return;
2573 
2574         buf = container_of(buffer, struct rpc_buffer, data);
2575         free_page((unsigned long)buf);
2576 }
2577 
2578 /*
2579  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2580  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2581  */
2582 static int bc_sendto(struct rpc_rqst *req)
2583 {
2584         int len;
2585         struct xdr_buf *xbufp = &req->rq_snd_buf;
2586         struct rpc_xprt *xprt = req->rq_xprt;
2587         struct sock_xprt *transport =
2588                                 container_of(xprt, struct sock_xprt, xprt);
2589         struct socket *sock = transport->sock;
2590         unsigned long headoff;
2591         unsigned long tailoff;
2592 
2593         xs_encode_stream_record_marker(xbufp);
2594 
2595         tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2596         headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2597         len = svc_send_common(sock, xbufp,
2598                               virt_to_page(xbufp->head[0].iov_base), headoff,
2599                               xbufp->tail[0].iov_base, tailoff);
2600 
2601         if (len != xbufp->len) {
2602                 printk(KERN_NOTICE "Error sending entire callback!\n");
2603                 len = -EAGAIN;
2604         }
2605 
2606         return len;
2607 }
2608 
2609 /*
2610  * The send routine. Borrows from svc_send
2611  */
2612 static int bc_send_request(struct rpc_task *task)
2613 {
2614         struct rpc_rqst *req = task->tk_rqstp;
2615         struct svc_xprt *xprt;
2616         int len;
2617 
2618         dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2619         /*
2620          * Get the server socket associated with this callback xprt
2621          */
2622         xprt = req->rq_xprt->bc_xprt;
2623 
2624         /*
2625          * Grab the mutex to serialize data as the connection is shared
2626          * with the fore channel
2627          */
2628         if (!mutex_trylock(&xprt->xpt_mutex)) {
2629                 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2630                 if (!mutex_trylock(&xprt->xpt_mutex))
2631                         return -EAGAIN;
2632                 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2633         }
2634         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2635                 len = -ENOTCONN;
2636         else
2637                 len = bc_sendto(req);
2638         mutex_unlock(&xprt->xpt_mutex);
2639 
2640         if (len > 0)
2641                 len = 0;
2642 
2643         return len;
2644 }
2645 
2646 /*
2647  * The close routine. Since this is client initiated, we do nothing
2648  */
2649 
2650 static void bc_close(struct rpc_xprt *xprt)
2651 {
2652 }
2653 
2654 /*
2655  * The xprt destroy routine. Again, because this connection is client
2656  * initiated, we do nothing
2657  */
2658 
2659 static void bc_destroy(struct rpc_xprt *xprt)
2660 {
2661         dprintk("RPC:       bc_destroy xprt %p\n", xprt);
2662 
2663         xs_xprt_free(xprt);
2664         module_put(THIS_MODULE);
2665 }
2666 
2667 static struct rpc_xprt_ops xs_local_ops = {
2668         .reserve_xprt           = xprt_reserve_xprt,
2669         .release_xprt           = xs_tcp_release_xprt,
2670         .alloc_slot             = xprt_alloc_slot,
2671         .rpcbind                = xs_local_rpcbind,
2672         .set_port               = xs_local_set_port,
2673         .connect                = xs_local_connect,
2674         .buf_alloc              = rpc_malloc,
2675         .buf_free               = rpc_free,
2676         .send_request           = xs_local_send_request,
2677         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2678         .close                  = xs_close,
2679         .destroy                = xs_destroy,
2680         .print_stats            = xs_local_print_stats,
2681         .enable_swap            = xs_enable_swap,
2682         .disable_swap           = xs_disable_swap,
2683 };
2684 
2685 static struct rpc_xprt_ops xs_udp_ops = {
2686         .set_buffer_size        = xs_udp_set_buffer_size,
2687         .reserve_xprt           = xprt_reserve_xprt_cong,
2688         .release_xprt           = xprt_release_xprt_cong,
2689         .alloc_slot             = xprt_alloc_slot,
2690         .rpcbind                = rpcb_getport_async,
2691         .set_port               = xs_set_port,
2692         .connect                = xs_connect,
2693         .buf_alloc              = rpc_malloc,
2694         .buf_free               = rpc_free,
2695         .send_request           = xs_udp_send_request,
2696         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
2697         .timer                  = xs_udp_timer,
2698         .release_request        = xprt_release_rqst_cong,
2699         .close                  = xs_close,
2700         .destroy                = xs_destroy,
2701         .print_stats            = xs_udp_print_stats,
2702         .enable_swap            = xs_enable_swap,
2703         .disable_swap           = xs_disable_swap,
2704         .inject_disconnect      = xs_inject_disconnect,
2705 };
2706 
2707 static struct rpc_xprt_ops xs_tcp_ops = {
2708         .reserve_xprt           = xprt_reserve_xprt,
2709         .release_xprt           = xs_tcp_release_xprt,
2710         .alloc_slot             = xprt_lock_and_alloc_slot,
2711         .rpcbind                = rpcb_getport_async,
2712         .set_port               = xs_set_port,
2713         .connect                = xs_connect,
2714         .buf_alloc              = rpc_malloc,
2715         .buf_free               = rpc_free,
2716         .send_request           = xs_tcp_send_request,
2717         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2718         .close                  = xs_tcp_shutdown,
2719         .destroy                = xs_destroy,
2720         .print_stats            = xs_tcp_print_stats,
2721         .enable_swap            = xs_enable_swap,
2722         .disable_swap           = xs_disable_swap,
2723         .inject_disconnect      = xs_inject_disconnect,
2724 #ifdef CONFIG_SUNRPC_BACKCHANNEL
2725         .bc_setup               = xprt_setup_bc,
2726         .bc_up                  = xs_tcp_bc_up,
2727         .bc_maxpayload          = xs_tcp_bc_maxpayload,
2728         .bc_free_rqst           = xprt_free_bc_rqst,
2729         .bc_destroy             = xprt_destroy_bc,
2730 #endif
2731 };
2732 
2733 /*
2734  * The rpc_xprt_ops for the server backchannel
2735  */
2736 
2737 static struct rpc_xprt_ops bc_tcp_ops = {
2738         .reserve_xprt           = xprt_reserve_xprt,
2739         .release_xprt           = xprt_release_xprt,
2740         .alloc_slot             = xprt_alloc_slot,
2741         .buf_alloc              = bc_malloc,
2742         .buf_free               = bc_free,
2743         .send_request           = bc_send_request,
2744         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2745         .close                  = bc_close,
2746         .destroy                = bc_destroy,
2747         .print_stats            = xs_tcp_print_stats,
2748         .enable_swap            = xs_enable_swap,
2749         .disable_swap           = xs_disable_swap,
2750         .inject_disconnect      = xs_inject_disconnect,
2751 };
2752 
2753 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2754 {
2755         static const struct sockaddr_in sin = {
2756                 .sin_family             = AF_INET,
2757                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2758         };
2759         static const struct sockaddr_in6 sin6 = {
2760                 .sin6_family            = AF_INET6,
2761                 .sin6_addr              = IN6ADDR_ANY_INIT,
2762         };
2763 
2764         switch (family) {
2765         case AF_LOCAL:
2766                 break;
2767         case AF_INET:
2768                 memcpy(sap, &sin, sizeof(sin));
2769                 break;
2770         case AF_INET6:
2771                 memcpy(sap, &sin6, sizeof(sin6));
2772                 break;
2773         default:
2774                 dprintk("RPC:       %s: Bad address family\n", __func__);
2775                 return -EAFNOSUPPORT;
2776         }
2777         return 0;
2778 }
2779 
2780 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2781                                       unsigned int slot_table_size,
2782                                       unsigned int max_slot_table_size)
2783 {
2784         struct rpc_xprt *xprt;
2785         struct sock_xprt *new;
2786 
2787         if (args->addrlen > sizeof(xprt->addr)) {
2788                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2789                 return ERR_PTR(-EBADF);
2790         }
2791 
2792         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2793                         max_slot_table_size);
2794         if (xprt == NULL) {
2795                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2796                                 "rpc_xprt\n");
2797                 return ERR_PTR(-ENOMEM);
2798         }
2799 
2800         new = container_of(xprt, struct sock_xprt, xprt);
2801         mutex_init(&new->recv_mutex);
2802         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2803         xprt->addrlen = args->addrlen;
2804         if (args->srcaddr)
2805                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2806         else {
2807                 int err;
2808                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2809                                         (struct sockaddr *)&new->srcaddr);
2810                 if (err != 0) {
2811                         xprt_free(xprt);
2812                         return ERR_PTR(err);
2813                 }
2814         }
2815 
2816         return xprt;
2817 }
2818 
2819 static const struct rpc_timeout xs_local_default_timeout = {
2820         .to_initval = 10 * HZ,
2821         .to_maxval = 10 * HZ,
2822         .to_retries = 2,
2823 };
2824 
2825 /**
2826  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2827  * @args: rpc transport creation arguments
2828  *
2829  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2830  */
2831 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2832 {
2833         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2834         struct sock_xprt *transport;
2835         struct rpc_xprt *xprt;
2836         struct rpc_xprt *ret;
2837 
2838         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2839                         xprt_max_tcp_slot_table_entries);
2840         if (IS_ERR(xprt))
2841                 return xprt;
2842         transport = container_of(xprt, struct sock_xprt, xprt);
2843 
2844         xprt->prot = 0;
2845         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2846         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2847 
2848         xprt->bind_timeout = XS_BIND_TO;
2849         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2850         xprt->idle_timeout = XS_IDLE_DISC_TO;
2851 
2852         xprt->ops = &xs_local_ops;
2853         xprt->timeout = &xs_local_default_timeout;
2854 
2855         INIT_WORK(&transport->recv_worker, xs_local_data_receive_workfn);
2856         INIT_DELAYED_WORK(&transport->connect_worker,
2857                         xs_dummy_setup_socket);
2858 
2859         switch (sun->sun_family) {
2860         case AF_LOCAL:
2861                 if (sun->sun_path[0] != '/') {
2862                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
2863                                         sun->sun_path);
2864                         ret = ERR_PTR(-EINVAL);
2865                         goto out_err;
2866                 }
2867                 xprt_set_bound(xprt);
2868                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2869                 ret = ERR_PTR(xs_local_setup_socket(transport));
2870                 if (ret)
2871                         goto out_err;
2872                 break;
2873         default:
2874                 ret = ERR_PTR(-EAFNOSUPPORT);
2875                 goto out_err;
2876         }
2877 
2878         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2879                         xprt->address_strings[RPC_DISPLAY_ADDR]);
2880 
2881         if (try_module_get(THIS_MODULE))
2882                 return xprt;
2883         ret = ERR_PTR(-EINVAL);
2884 out_err:
2885         xs_xprt_free(xprt);
2886         return ret;
2887 }
2888 
2889 static const struct rpc_timeout xs_udp_default_timeout = {
2890         .to_initval = 5 * HZ,
2891         .to_maxval = 30 * HZ,
2892         .to_increment = 5 * HZ,
2893         .to_retries = 5,
2894 };
2895 
2896 /**
2897  * xs_setup_udp - Set up transport to use a UDP socket
2898  * @args: rpc transport creation arguments
2899  *
2900  */
2901 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2902 {
2903         struct sockaddr *addr = args->dstaddr;
2904         struct rpc_xprt *xprt;
2905         struct sock_xprt *transport;
2906         struct rpc_xprt *ret;
2907 
2908         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2909                         xprt_udp_slot_table_entries);
2910         if (IS_ERR(xprt))
2911                 return xprt;
2912         transport = container_of(xprt, struct sock_xprt, xprt);
2913 
2914         xprt->prot = IPPROTO_UDP;
2915         xprt->tsh_size = 0;
2916         /* XXX: header size can vary due to auth type, IPv6, etc. */
2917         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2918 
2919         xprt->bind_timeout = XS_BIND_TO;
2920         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2921         xprt->idle_timeout = XS_IDLE_DISC_TO;
2922 
2923         xprt->ops = &xs_udp_ops;
2924 
2925         xprt->timeout = &xs_udp_default_timeout;
2926 
2927         INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2928         INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2929 
2930         switch (addr->sa_family) {
2931         case AF_INET:
2932                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2933                         xprt_set_bound(xprt);
2934 
2935                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2936                 break;
2937         case AF_INET6:
2938                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2939                         xprt_set_bound(xprt);
2940 
2941                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2942                 break;
2943         default:
2944                 ret = ERR_PTR(-EAFNOSUPPORT);
2945                 goto out_err;
2946         }
2947 
2948         if (xprt_bound(xprt))
2949                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2950                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2951                                 xprt->address_strings[RPC_DISPLAY_PORT],
2952                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2953         else
2954                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2955                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2956                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2957 
2958         if (try_module_get(THIS_MODULE))
2959                 return xprt;
2960         ret = ERR_PTR(-EINVAL);
2961 out_err:
2962         xs_xprt_free(xprt);
2963         return ret;
2964 }
2965 
2966 static const struct rpc_timeout xs_tcp_default_timeout = {
2967         .to_initval = 60 * HZ,
2968         .to_maxval = 60 * HZ,
2969         .to_retries = 2,
2970 };
2971 
2972 /**
2973  * xs_setup_tcp - Set up transport to use a TCP socket
2974  * @args: rpc transport creation arguments
2975  *
2976  */
2977 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2978 {
2979         struct sockaddr *addr = args->dstaddr;
2980         struct rpc_xprt *xprt;
2981         struct sock_xprt *transport;
2982         struct rpc_xprt *ret;
2983         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2984 
2985         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2986                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2987 
2988         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2989                         max_slot_table_size);
2990         if (IS_ERR(xprt))
2991                 return xprt;
2992         transport = container_of(xprt, struct sock_xprt, xprt);
2993 
2994         xprt->prot = IPPROTO_TCP;
2995         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2996         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2997 
2998         xprt->bind_timeout = XS_BIND_TO;
2999         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3000         xprt->idle_timeout = XS_IDLE_DISC_TO;
3001 
3002         xprt->ops = &xs_tcp_ops;
3003         xprt->timeout = &xs_tcp_default_timeout;
3004 
3005         xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3006 
3007         INIT_WORK(&transport->recv_worker, xs_tcp_data_receive_workfn);
3008         INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3009 
3010         switch (addr->sa_family) {
3011         case AF_INET:
3012                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3013                         xprt_set_bound(xprt);
3014 
3015                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3016                 break;
3017         case AF_INET6:
3018                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3019                         xprt_set_bound(xprt);
3020 
3021                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3022                 break;
3023         default:
3024                 ret = ERR_PTR(-EAFNOSUPPORT);
3025                 goto out_err;
3026         }
3027 
3028         if (xprt_bound(xprt))
3029                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3030                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3031                                 xprt->address_strings[RPC_DISPLAY_PORT],
3032                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3033         else
3034                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3035                                 xprt->address_strings[RPC_DISPLAY_ADDR],
3036                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
3037 
3038         if (try_module_get(THIS_MODULE))
3039                 return xprt;
3040         ret = ERR_PTR(-EINVAL);
3041 out_err:
3042         xs_xprt_free(xprt);
3043         return ret;
3044 }
3045 
3046 /**
3047  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3048  * @args: rpc transport creation arguments
3049  *
3050  */
3051 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3052 {
3053         struct sockaddr *addr = args->dstaddr;
3054         struct rpc_xprt *xprt;
3055         struct sock_xprt *transport;
3056         struct svc_sock *bc_sock;
3057         struct rpc_xprt *ret;
3058 
3059         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3060                         xprt_tcp_slot_table_entries);
3061         if (IS_ERR(xprt))
3062                 return xprt;
3063         transport = container_of(xprt, struct sock_xprt, xprt);
3064 
3065         xprt->prot = IPPROTO_TCP;
3066         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3067         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3068         xprt->timeout = &xs_tcp_default_timeout;
3069 
3070         /* backchannel */
3071         xprt_set_bound(xprt);
3072         xprt->bind_timeout = 0;
3073         xprt->reestablish_timeout = 0;
3074         xprt->idle_timeout = 0;
3075 
3076         xprt->ops = &bc_tcp_ops;
3077 
3078         switch (addr->sa_family) {
3079         case AF_INET:
3080                 xs_format_peer_addresses(xprt, "tcp",
3081                                          RPCBIND_NETID_TCP);
3082                 break;
3083         case AF_INET6:
3084                 xs_format_peer_addresses(xprt, "tcp",
3085                                    RPCBIND_NETID_TCP6);
3086                 break;
3087         default:
3088                 ret = ERR_PTR(-EAFNOSUPPORT);
3089                 goto out_err;
3090         }
3091 
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 
3097         /*
3098          * Once we've associated a backchannel xprt with a connection,
3099          * we want to keep it around as long as the connection lasts,
3100          * in case we need to start using it for a backchannel again;
3101          * this reference won't be dropped until bc_xprt is destroyed.
3102          */
3103         xprt_get(xprt);
3104         args->bc_xprt->xpt_bc_xprt = xprt;
3105         xprt->bc_xprt = args->bc_xprt;
3106         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3107         transport->sock = bc_sock->sk_sock;
3108         transport->inet = bc_sock->sk_sk;
3109 
3110         /*
3111          * Since we don't want connections for the backchannel, we set
3112          * the xprt status to connected
3113          */
3114         xprt_set_connected(xprt);
3115 
3116         if (try_module_get(THIS_MODULE))
3117                 return xprt;
3118 
3119         args->bc_xprt->xpt_bc_xprt = NULL;
3120         args->bc_xprt->xpt_bc_xps = NULL;
3121         xprt_put(xprt);
3122         ret = ERR_PTR(-EINVAL);
3123 out_err:
3124         xs_xprt_free(xprt);
3125         return ret;
3126 }
3127 
3128 static struct xprt_class        xs_local_transport = {
3129         .list           = LIST_HEAD_INIT(xs_local_transport.list),
3130         .name           = "named UNIX socket",
3131         .owner          = THIS_MODULE,
3132         .ident          = XPRT_TRANSPORT_LOCAL,
3133         .setup          = xs_setup_local,
3134 };
3135 
3136 static struct xprt_class        xs_udp_transport = {
3137         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
3138         .name           = "udp",
3139         .owner          = THIS_MODULE,
3140         .ident          = XPRT_TRANSPORT_UDP,
3141         .setup          = xs_setup_udp,
3142 };
3143 
3144 static struct xprt_class        xs_tcp_transport = {
3145         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
3146         .name           = "tcp",
3147         .owner          = THIS_MODULE,
3148         .ident          = XPRT_TRANSPORT_TCP,
3149         .setup          = xs_setup_tcp,
3150 };
3151 
3152 static struct xprt_class        xs_bc_tcp_transport = {
3153         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3154         .name           = "tcp NFSv4.1 backchannel",
3155         .owner          = THIS_MODULE,
3156         .ident          = XPRT_TRANSPORT_BC_TCP,
3157         .setup          = xs_setup_bc_tcp,
3158 };
3159 
3160 /**
3161  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3162  *
3163  */
3164 int init_socket_xprt(void)
3165 {
3166 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3167         if (!sunrpc_table_header)
3168                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3169 #endif
3170 
3171         xprt_register_transport(&xs_local_transport);
3172         xprt_register_transport(&xs_udp_transport);
3173         xprt_register_transport(&xs_tcp_transport);
3174         xprt_register_transport(&xs_bc_tcp_transport);
3175 
3176         return 0;
3177 }
3178 
3179 /**
3180  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3181  *
3182  */
3183 void cleanup_socket_xprt(void)
3184 {
3185 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3186         if (sunrpc_table_header) {
3187                 unregister_sysctl_table(sunrpc_table_header);
3188                 sunrpc_table_header = NULL;
3189         }
3190 #endif
3191 
3192         xprt_unregister_transport(&xs_local_transport);
3193         xprt_unregister_transport(&xs_udp_transport);
3194         xprt_unregister_transport(&xs_tcp_transport);
3195         xprt_unregister_transport(&xs_bc_tcp_transport);
3196 }
3197 
3198 static int param_set_uint_minmax(const char *val,
3199                 const struct kernel_param *kp,
3200                 unsigned int min, unsigned int max)
3201 {
3202         unsigned int num;
3203         int ret;
3204 
3205         if (!val)
3206                 return -EINVAL;
3207         ret = kstrtouint(val, 0, &num);
3208         if (ret == -EINVAL || num < min || num > max)
3209                 return -EINVAL;
3210         *((unsigned int *)kp->arg) = num;
3211         return 0;
3212 }
3213 
3214 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3215 {
3216         if (kp->arg == &xprt_min_resvport)
3217                 return param_set_uint_minmax(val, kp,
3218                         RPC_MIN_RESVPORT,
3219                         xprt_max_resvport);
3220         return param_set_uint_minmax(val, kp,
3221                         xprt_min_resvport,
3222                         RPC_MAX_RESVPORT);
3223 }
3224 
3225 static const struct kernel_param_ops param_ops_portnr = {
3226         .set = param_set_portnr,
3227         .get = param_get_uint,
3228 };
3229 
3230 #define param_check_portnr(name, p) \
3231         __param_check(name, p, unsigned int);
3232 
3233 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3234 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3235 
3236 static int param_set_slot_table_size(const char *val,
3237                                      const struct kernel_param *kp)
3238 {
3239         return param_set_uint_minmax(val, kp,
3240                         RPC_MIN_SLOT_TABLE,
3241                         RPC_MAX_SLOT_TABLE);
3242 }
3243 
3244 static const struct kernel_param_ops param_ops_slot_table_size = {
3245         .set = param_set_slot_table_size,
3246         .get = param_get_uint,
3247 };
3248 
3249 #define param_check_slot_table_size(name, p) \
3250         __param_check(name, p, unsigned int);
3251 
3252 static int param_set_max_slot_table_size(const char *val,
3253                                      const struct kernel_param *kp)
3254 {
3255         return param_set_uint_minmax(val, kp,
3256                         RPC_MIN_SLOT_TABLE,
3257                         RPC_MAX_SLOT_TABLE_LIMIT);
3258 }
3259 
3260 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3261         .set = param_set_max_slot_table_size,
3262         .get = param_get_uint,
3263 };
3264 
3265 #define param_check_max_slot_table_size(name, p) \
3266         __param_check(name, p, unsigned int);
3267 
3268 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3269                    slot_table_size, 0644);
3270 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3271                    max_slot_table_size, 0644);
3272 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3273                    slot_table_size, 0644);
3274 
3275 

~ [ 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