net/sunrpc/xprtsock.c

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