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;
 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, 0, sizeof(_xid), &_xid);
1006         if (xp == NULL)
1007                 return;
1008
1009         /* Look up and lock the request corresponding to the given XID */
1010         spin_lock_bh(&xprt->transport_lock);
1011         rovr = xprt_lookup_rqst(xprt, *xp);
1012         if (!rovr)
1013                 goto out_unlock;
1014         task = rovr->rq_task;
1015
1016         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1017                 copied = repsize;
1018
1019         /* Suck it into the iovec, verify checksum if not done by hw. */
1020         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1021                 __UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1022                 goto out_unlock;
1023         }
1024
1025         __UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1026
1027         xprt_adjust_cwnd(xprt, task, copied);
1028         xprt_complete_rqst(task, copied);
1029
1030  out_unlock:
1031         spin_unlock_bh(&xprt->transport_lock);
1032 }
1033
1034 static void xs_udp_data_receive(struct sock_xprt *transport)
1035 {
1036         struct sk_buff *skb;
1037         struct sock *sk;
1038         int err;
1039
1040         mutex_lock(&transport->recv_mutex);
1041         sk = transport->inet;
1042         if (sk == NULL)
1043                 goto out;
1044         for (;;) {
1045                 skb = skb_recv_datagram(sk, 0, 1, &err);
1046                 if (skb == NULL)
1047                         break;
1048                 xs_udp_data_read_skb(&transport->xprt, sk, skb);
1049                 skb_free_datagram(sk, skb);
1050         }
1051 out:
1052         mutex_unlock(&transport->recv_mutex);
1053 }
1054
1055 static void xs_udp_data_receive_workfn(struct work_struct *work)
1056 {
1057         struct sock_xprt *transport =
1058                 container_of(work, struct sock_xprt, recv_worker);
1059         xs_udp_data_receive(transport);
1060 }
1061
1062 /**
1063  * xs_data_ready - "data ready" callback for UDP sockets
1064  * @sk: socket with data to read
1065  *
1066  */
1067 static void xs_data_ready(struct sock *sk)
1068 {
1069         struct rpc_xprt *xprt;
1070
1071         read_lock_bh(&sk->sk_callback_lock);
1072         dprintk("RPC:       xs_data_ready...\n");
1073         xprt = xprt_from_sock(sk);
1074         if (xprt != NULL) {
1075                 struct sock_xprt *transport = container_of(xprt,
1076                                 struct sock_xprt, xprt);
1077                 queue_work(rpciod_workqueue, &transport->recv_worker);
1078         }
1079         read_unlock_bh(&sk->sk_callback_lock);
1080 }
1081
1082 /*
1083  * Helper function to force a TCP close if the server is sending
1084  * junk and/or it has put us in CLOSE_WAIT
1085  */
1086 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1087 {
1088         xprt_force_disconnect(xprt);
1089 }
1090
1091 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1092 {
1093         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1094         size_t len, used;
1095         char *p;
1096
1097         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1098         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1099         used = xdr_skb_read_bits(desc, p, len);
1100         transport->tcp_offset += used;
1101         if (used != len)
1102                 return;
1103
1104         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1105         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1106                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1107         else
1108                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1109         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1110
1111         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1112         transport->tcp_offset = 0;
1113
1114         /* Sanity check of the record length */
1115         if (unlikely(transport->tcp_reclen < 8)) {
1116                 dprintk("RPC:       invalid TCP record fragment length\n");
1117                 xs_tcp_force_close(xprt);
1118                 return;
1119         }
1120         dprintk("RPC:       reading TCP record fragment of length %d\n",
1121                         transport->tcp_reclen);
1122 }
1123
1124 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1125 {
1126         if (transport->tcp_offset == transport->tcp_reclen) {
1127                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1128                 transport->tcp_offset = 0;
1129                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1130                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1131                         transport->tcp_flags |= TCP_RCV_COPY_XID;
1132                         transport->tcp_copied = 0;
1133                 }
1134         }
1135 }
1136
1137 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1138 {
1139         size_t len, used;
1140         char *p;
1141
1142         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1143         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
1144         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1145         used = xdr_skb_read_bits(desc, p, len);
1146         transport->tcp_offset += used;
1147         if (used != len)
1148                 return;
1149         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1150         transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1151         transport->tcp_copied = 4;
1152         dprintk("RPC:       reading %s XID %08x\n",
1153                         (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1154                                                               : "request with",
1155                         ntohl(transport->tcp_xid));
1156         xs_tcp_check_fraghdr(transport);
1157 }
1158
1159 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1160                                        struct xdr_skb_reader *desc)
1161 {
1162         size_t len, used;
1163         u32 offset;
1164         char *p;
1165
1166         /*
1167          * We want transport->tcp_offset to be 8 at the end of this routine
1168          * (4 bytes for the xid and 4 bytes for the call/reply flag).
1169          * When this function is called for the first time,
1170          * transport->tcp_offset is 4 (after having already read the xid).
1171          */
1172         offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1173         len = sizeof(transport->tcp_calldir) - offset;
1174         dprintk("RPC:       reading CALL/REPLY flag (%Zu bytes)\n", len);
1175         p = ((char *) &transport->tcp_calldir) + offset;
1176         used = xdr_skb_read_bits(desc, p, len);
1177         transport->tcp_offset += used;
1178         if (used != len)
1179                 return;
1180         transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1181         /*
1182          * We don't yet have the XDR buffer, so we will write the calldir
1183          * out after we get the buffer from the 'struct rpc_rqst'
1184          */
1185         switch (ntohl(transport->tcp_calldir)) {
1186         case RPC_REPLY:
1187                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1188                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1189                 transport->tcp_flags |= TCP_RPC_REPLY;
1190                 break;
1191         case RPC_CALL:
1192                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1193                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1194                 transport->tcp_flags &= ~TCP_RPC_REPLY;
1195                 break;
1196         default:
1197                 dprintk("RPC:       invalid request message type\n");
1198                 xs_tcp_force_close(&transport->xprt);
1199         }
1200         xs_tcp_check_fraghdr(transport);
1201 }
1202
1203 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1204                                      struct xdr_skb_reader *desc,
1205                                      struct rpc_rqst *req)
1206 {
1207         struct sock_xprt *transport =
1208                                 container_of(xprt, struct sock_xprt, xprt);
1209         struct xdr_buf *rcvbuf;
1210         size_t len;
1211         ssize_t r;
1212
1213         rcvbuf = &req->rq_private_buf;
1214
1215         if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1216                 /*
1217                  * Save the RPC direction in the XDR buffer
1218                  */
1219                 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1220                         &transport->tcp_calldir,
1221                         sizeof(transport->tcp_calldir));
1222                 transport->tcp_copied += sizeof(transport->tcp_calldir);
1223                 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1224         }
1225
1226         len = desc->count;
1227         if (len > transport->tcp_reclen - transport->tcp_offset) {
1228                 struct xdr_skb_reader my_desc;
1229
1230                 len = transport->tcp_reclen - transport->tcp_offset;
1231                 memcpy(&my_desc, desc, sizeof(my_desc));
1232                 my_desc.count = len;
1233                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1234                                           &my_desc, xdr_skb_read_bits);
1235                 desc->count -= r;
1236                 desc->offset += r;
1237         } else
1238                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1239                                           desc, xdr_skb_read_bits);
1240
1241         if (r > 0) {
1242                 transport->tcp_copied += r;
1243                 transport->tcp_offset += r;
1244         }
1245         if (r != len) {
1246                 /* Error when copying to the receive buffer,
1247                  * usually because we weren't able to allocate
1248                  * additional buffer pages. All we can do now
1249                  * is turn off TCP_RCV_COPY_DATA, so the request
1250                  * will not receive any additional updates,
1251                  * and time out.
1252                  * Any remaining data from this record will
1253                  * be discarded.
1254                  */
1255                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1256                 dprintk("RPC:       XID %08x truncated request\n",
1257                                 ntohl(transport->tcp_xid));
1258                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1259                                 "tcp_offset = %u, tcp_reclen = %u\n",
1260                                 xprt, transport->tcp_copied,
1261                                 transport->tcp_offset, transport->tcp_reclen);
1262                 return;
1263         }
1264
1265         dprintk("RPC:       XID %08x read %Zd bytes\n",
1266                         ntohl(transport->tcp_xid), r);
1267         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1268                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1269                         transport->tcp_offset, transport->tcp_reclen);
1270
1271         if (transport->tcp_copied == req->rq_private_buf.buflen)
1272                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1273         else if (transport->tcp_offset == transport->tcp_reclen) {
1274                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1275                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1276         }
1277 }
1278
1279 /*
1280  * Finds the request corresponding to the RPC xid and invokes the common
1281  * tcp read code to read the data.
1282  */
1283 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1284                                     struct xdr_skb_reader *desc)
1285 {
1286         struct sock_xprt *transport =
1287                                 container_of(xprt, struct sock_xprt, xprt);
1288         struct rpc_rqst *req;
1289
1290         dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1291
1292         /* Find and lock the request corresponding to this xid */
1293         spin_lock_bh(&xprt->transport_lock);
1294         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1295         if (!req) {
1296                 dprintk("RPC:       XID %08x request not found!\n",
1297                                 ntohl(transport->tcp_xid));
1298                 spin_unlock_bh(&xprt->transport_lock);
1299                 return -1;
1300         }
1301
1302         xs_tcp_read_common(xprt, desc, req);
1303
1304         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1305                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1306
1307         spin_unlock_bh(&xprt->transport_lock);
1308         return 0;
1309 }
1310
1311 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1312 /*
1313  * Obtains an rpc_rqst previously allocated and invokes the common
1314  * tcp read code to read the data.  The result is placed in the callback
1315  * queue.
1316  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1317  * connection and return -1.
1318  */
1319 static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1320                                        struct xdr_skb_reader *desc)
1321 {
1322         struct sock_xprt *transport =
1323                                 container_of(xprt, struct sock_xprt, xprt);
1324         struct rpc_rqst *req;
1325
1326         /* Look up and lock the request corresponding to the given XID */
1327         spin_lock_bh(&xprt->transport_lock);
1328         req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1329         if (req == NULL) {
1330                 spin_unlock_bh(&xprt->transport_lock);
1331                 printk(KERN_WARNING "Callback slot table overflowed\n");
1332                 xprt_force_disconnect(xprt);
1333                 return -1;
1334         }
1335
1336         dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1337         xs_tcp_read_common(xprt, desc, req);
1338
1339         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1340                 xprt_complete_bc_request(req, transport->tcp_copied);
1341         spin_unlock_bh(&xprt->transport_lock);
1342
1343         return 0;
1344 }
1345
1346 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1347                                         struct xdr_skb_reader *desc)
1348 {
1349         struct sock_xprt *transport =
1350                                 container_of(xprt, struct sock_xprt, xprt);
1351
1352         return (transport->tcp_flags & TCP_RPC_REPLY) ?
1353                 xs_tcp_read_reply(xprt, desc) :
1354                 xs_tcp_read_callback(xprt, desc);
1355 }
1356
1357 static int xs_tcp_bc_up(struct svc_serv *serv, struct net *net)
1358 {
1359         int ret;
1360
1361         ret = svc_create_xprt(serv, "tcp-bc", net, PF_INET, 0,
1362                               SVC_SOCK_ANONYMOUS);
1363         if (ret < 0)
1364                 return ret;
1365         return 0;
1366 }
1367 #else
1368 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1369                                         struct xdr_skb_reader *desc)
1370 {
1371         return xs_tcp_read_reply(xprt, desc);
1372 }
1373 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1374
1375 /*
1376  * Read data off the transport.  This can be either an RPC_CALL or an
1377  * RPC_REPLY.  Relay the processing to helper functions.
1378  */
1379 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1380                                     struct xdr_skb_reader *desc)
1381 {
1382         struct sock_xprt *transport =
1383                                 container_of(xprt, struct sock_xprt, xprt);
1384
1385         if (_xs_tcp_read_data(xprt, desc) == 0)
1386                 xs_tcp_check_fraghdr(transport);
1387         else {
1388                 /*
1389                  * The transport_lock protects the request handling.
1390                  * There's no need to hold it to update the tcp_flags.
1391                  */
1392                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1393         }
1394 }
1395
1396 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1397 {
1398         size_t len;
1399
1400         len = transport->tcp_reclen - transport->tcp_offset;
1401         if (len > desc->count)
1402                 len = desc->count;
1403         desc->count -= len;
1404         desc->offset += len;
1405         transport->tcp_offset += len;
1406         dprintk("RPC:       discarded %Zu bytes\n", len);
1407         xs_tcp_check_fraghdr(transport);
1408 }
1409
1410 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1411 {
1412         struct rpc_xprt *xprt = rd_desc->arg.data;
1413         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1414         struct xdr_skb_reader desc = {
1415                 .skb    = skb,
1416                 .offset = offset,
1417                 .count  = len,
1418         };
1419
1420         dprintk("RPC:       xs_tcp_data_recv started\n");
1421         do {
1422                 trace_xs_tcp_data_recv(transport);
1423                 /* Read in a new fragment marker if necessary */
1424                 /* Can we ever really expect to get completely empty fragments? */
1425                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1426                         xs_tcp_read_fraghdr(xprt, &desc);
1427                         continue;
1428                 }
1429                 /* Read in the xid if necessary */
1430                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1431                         xs_tcp_read_xid(transport, &desc);
1432                         continue;
1433                 }
1434                 /* Read in the call/reply flag */
1435                 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1436                         xs_tcp_read_calldir(transport, &desc);
1437                         continue;
1438                 }
1439                 /* Read in the request data */
1440                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1441                         xs_tcp_read_data(xprt, &desc);
1442                         continue;
1443                 }
1444                 /* Skip over any trailing bytes on short reads */
1445                 xs_tcp_read_discard(transport, &desc);
1446         } while (desc.count);
1447         trace_xs_tcp_data_recv(transport);
1448         dprintk("RPC:       xs_tcp_data_recv done\n");
1449         return len - desc.count;
1450 }
1451
1452 static void xs_tcp_data_receive(struct sock_xprt *transport)
1453 {
1454         struct rpc_xprt *xprt = &transport->xprt;
1455         struct sock *sk;
1456         read_descriptor_t rd_desc = {
1457                 .count = 2*1024*1024,
1458                 .arg.data = xprt,
1459         };
1460         unsigned long total = 0;
1461         int read = 0;
1462
1463         mutex_lock(&transport->recv_mutex);
1464         sk = transport->inet;
1465         if (sk == NULL)
1466                 goto out;
1467
1468         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1469         for (;;) {
1470                 lock_sock(sk);
1471                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1472                 release_sock(sk);
1473                 if (read <= 0)
1474                         break;
1475                 total += read;
1476                 rd_desc.count = 65536;
1477         }
1478 out:
1479         mutex_unlock(&transport->recv_mutex);
1480         trace_xs_tcp_data_ready(xprt, read, total);
1481 }
1482
1483 static void xs_tcp_data_receive_workfn(struct work_struct *work)
1484 {
1485         struct sock_xprt *transport =
1486                 container_of(work, struct sock_xprt, recv_worker);
1487         xs_tcp_data_receive(transport);
1488 }
1489
1490 /**
1491  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1492  * @sk: socket with data to read
1493  *
1494  */
1495 static void xs_tcp_data_ready(struct sock *sk)
1496 {
1497         struct sock_xprt *transport;
1498         struct rpc_xprt *xprt;
1499
1500         dprintk("RPC:       xs_tcp_data_ready...\n");
1501
1502         read_lock_bh(&sk->sk_callback_lock);
1503         if (!(xprt = xprt_from_sock(sk)))
1504                 goto out;
1505         transport = container_of(xprt, struct sock_xprt, xprt);
1506
1507         /* Any data means we had a useful conversation, so
1508          * the we don't need to delay the next reconnect
1509          */
1510         if (xprt->reestablish_timeout)
1511                 xprt->reestablish_timeout = 0;
1512         queue_work(rpciod_workqueue, &transport->recv_worker);
1513
1514 out:
1515         read_unlock_bh(&sk->sk_callback_lock);
1516 }
1517
1518 /**
1519  * xs_tcp_state_change - callback to handle TCP socket state changes
1520  * @sk: socket whose state has changed
1521  *
1522  */
1523 static void xs_tcp_state_change(struct sock *sk)
1524 {
1525         struct rpc_xprt *xprt;
1526         struct sock_xprt *transport;
1527
1528         read_lock_bh(&sk->sk_callback_lock);
1529         if (!(xprt = xprt_from_sock(sk)))
1530                 goto out;
1531         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1532         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1533                         sk->sk_state, xprt_connected(xprt),
1534                         sock_flag(sk, SOCK_DEAD),
1535                         sock_flag(sk, SOCK_ZAPPED),
1536                         sk->sk_shutdown);
1537
1538         transport = container_of(xprt, struct sock_xprt, xprt);
1539         trace_rpc_socket_state_change(xprt, sk->sk_socket);
1540         switch (sk->sk_state) {
1541         case TCP_ESTABLISHED:
1542                 spin_lock(&xprt->transport_lock);
1543                 if (!xprt_test_and_set_connected(xprt)) {
1544
1545                         /* Reset TCP record info */
1546                         transport->tcp_offset = 0;
1547                         transport->tcp_reclen = 0;
1548                         transport->tcp_copied = 0;
1549                         transport->tcp_flags =
1550                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1551                         xprt->connect_cookie++;
1552                         clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1553                         xprt_clear_connecting(xprt);
1554
1555                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1556                 }
1557                 spin_unlock(&xprt->transport_lock);
1558                 break;
1559         case TCP_FIN_WAIT1:
1560                 /* The client initiated a shutdown of the socket */
1561                 xprt->connect_cookie++;
1562                 xprt->reestablish_timeout = 0;
1563                 set_bit(XPRT_CLOSING, &xprt->state);
1564                 smp_mb__before_atomic();
1565                 clear_bit(XPRT_CONNECTED, &xprt->state);
1566                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1567                 smp_mb__after_atomic();
1568                 break;
1569         case TCP_CLOSE_WAIT:
1570                 /* The server initiated a shutdown of the socket */
1571                 xprt->connect_cookie++;
1572                 clear_bit(XPRT_CONNECTED, &xprt->state);
1573                 xs_tcp_force_close(xprt);
1574         case TCP_CLOSING:
1575                 /*
1576                  * If the server closed down the connection, make sure that
1577                  * we back off before reconnecting
1578                  */
1579                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1580                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1581                 break;
1582         case TCP_LAST_ACK:
1583                 set_bit(XPRT_CLOSING, &xprt->state);
1584                 smp_mb__before_atomic();
1585                 clear_bit(XPRT_CONNECTED, &xprt->state);
1586                 smp_mb__after_atomic();
1587                 break;
1588         case TCP_CLOSE:
1589                 if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1590                                         &transport->sock_state))
1591                         xprt_clear_connecting(xprt);
1592                 xs_sock_mark_closed(xprt);
1593         }
1594  out:
1595         read_unlock_bh(&sk->sk_callback_lock);
1596 }
1597
1598 static void xs_write_space(struct sock *sk)
1599 {
1600         struct socket_wq *wq;
1601         struct rpc_xprt *xprt;
1602
1603         if (!sk->sk_socket)
1604                 return;
1605         clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1606
1607         if (unlikely(!(xprt = xprt_from_sock(sk))))
1608                 return;
1609         rcu_read_lock();
1610         wq = rcu_dereference(sk->sk_wq);
1611         if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1612                 goto out;
1613
1614         xprt_write_space(xprt);
1615 out:
1616         rcu_read_unlock();
1617 }
1618
1619 /**
1620  * xs_udp_write_space - callback invoked when socket buffer space
1621  *                             becomes available
1622  * @sk: socket whose state has changed
1623  *
1624  * Called when more output buffer space is available for this socket.
1625  * We try not to wake our writers until they can make "significant"
1626  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1627  * with a bunch of small requests.
1628  */
1629 static void xs_udp_write_space(struct sock *sk)
1630 {
1631         read_lock_bh(&sk->sk_callback_lock);
1632
1633         /* from net/core/sock.c:sock_def_write_space */
1634         if (sock_writeable(sk))
1635                 xs_write_space(sk);
1636
1637         read_unlock_bh(&sk->sk_callback_lock);
1638 }
1639
1640 /**
1641  * xs_tcp_write_space - callback invoked when socket buffer space
1642  *                             becomes available
1643  * @sk: socket whose state has changed
1644  *
1645  * Called when more output buffer space is available for this socket.
1646  * We try not to wake our writers until they can make "significant"
1647  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1648  * with a bunch of small requests.
1649  */
1650 static void xs_tcp_write_space(struct sock *sk)
1651 {
1652         read_lock_bh(&sk->sk_callback_lock);
1653
1654         /* from net/core/stream.c:sk_stream_write_space */
1655         if (sk_stream_is_writeable(sk))
1656                 xs_write_space(sk);
1657
1658         read_unlock_bh(&sk->sk_callback_lock);
1659 }
1660
1661 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1662 {
1663         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1664         struct sock *sk = transport->inet;
1665
1666         if (transport->rcvsize) {
1667                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1668                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1669         }
1670         if (transport->sndsize) {
1671                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1672                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1673                 sk->sk_write_space(sk);
1674         }
1675 }
1676
1677 /**
1678  * xs_udp_set_buffer_size - set send and receive limits
1679  * @xprt: generic transport
1680  * @sndsize: requested size of send buffer, in bytes
1681  * @rcvsize: requested size of receive buffer, in bytes
1682  *
1683  * Set socket send and receive buffer size limits.
1684  */
1685 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1686 {
1687         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1688
1689         transport->sndsize = 0;
1690         if (sndsize)
1691                 transport->sndsize = sndsize + 1024;
1692         transport->rcvsize = 0;
1693         if (rcvsize)
1694                 transport->rcvsize = rcvsize + 1024;
1695
1696         xs_udp_do_set_buffer_size(xprt);
1697 }
1698
1699 /**
1700  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1701  * @task: task that timed out
1702  *
1703  * Adjust the congestion window after a retransmit timeout has occurred.
1704  */
1705 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1706 {
1707         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1708 }
1709
1710 static unsigned short xs_get_random_port(void)
1711 {
1712         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1713         unsigned short rand = (unsigned short) prandom_u32() % range;
1714         return rand + xprt_min_resvport;
1715 }
1716
1717 /**
1718  * xs_set_reuseaddr_port - set the socket's port and address reuse options
1719  * @sock: socket
1720  *
1721  * Note that this function has to be called on all sockets that share the
1722  * same port, and it must be called before binding.
1723  */
1724 static void xs_sock_set_reuseport(struct socket *sock)
1725 {
1726         int opt = 1;
1727
1728         kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1729                         (char *)&opt, sizeof(opt));
1730 }
1731
1732 static unsigned short xs_sock_getport(struct socket *sock)
1733 {
1734         struct sockaddr_storage buf;
1735         int buflen;
1736         unsigned short port = 0;
1737
1738         if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1739                 goto out;
1740         switch (buf.ss_family) {
1741         case AF_INET6:
1742                 port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1743                 break;
1744         case AF_INET:
1745                 port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1746         }
1747 out:
1748         return port;
1749 }
1750
1751 /**
1752  * xs_set_port - reset the port number in the remote endpoint address
1753  * @xprt: generic transport
1754  * @port: new port number
1755  *
1756  */
1757 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1758 {
1759         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1760
1761         rpc_set_port(xs_addr(xprt), port);
1762         xs_update_peer_port(xprt);
1763 }
1764
1765 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1766 {
1767         if (transport->srcport == 0)
1768                 transport->srcport = xs_sock_getport(sock);
1769 }
1770
1771 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1772 {
1773         unsigned short port = transport->srcport;
1774
1775         if (port == 0 && transport->xprt.resvport)
1776                 port = xs_get_random_port();
1777         return port;
1778 }
1779
1780 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1781 {
1782         if (transport->srcport != 0)
1783                 transport->srcport = 0;
1784         if (!transport->xprt.resvport)
1785                 return 0;
1786         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1787                 return xprt_max_resvport;
1788         return --port;
1789 }
1790 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1791 {
1792         struct sockaddr_storage myaddr;
1793         int err, nloop = 0;
1794         unsigned short port = xs_get_srcport(transport);
1795         unsigned short last;
1796
1797         /*
1798          * If we are asking for any ephemeral port (i.e. port == 0 &&
1799          * transport->xprt.resvport == 0), don't bind.  Let the local
1800          * port selection happen implicitly when the socket is used
1801          * (for example at connect time).
1802          *
1803          * This ensures that we can continue to establish TCP
1804          * connections even when all local ephemeral ports are already
1805          * a part of some TCP connection.  This makes no difference
1806          * for UDP sockets, but also doens't harm them.
1807          *
1808          * If we're asking for any reserved port (i.e. port == 0 &&
1809          * transport->xprt.resvport == 1) xs_get_srcport above will
1810          * ensure that port is non-zero and we will bind as needed.
1811          */
1812         if (port == 0)
1813                 return 0;
1814
1815         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1816         do {
1817                 rpc_set_port((struct sockaddr *)&myaddr, port);
1818                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1819                                 transport->xprt.addrlen);
1820                 if (err == 0) {
1821                         transport->srcport = port;
1822                         break;
1823                 }
1824                 last = port;
1825                 port = xs_next_srcport(transport, port);
1826                 if (port > last)
1827                         nloop++;
1828         } while (err == -EADDRINUSE && nloop != 2);
1829
1830         if (myaddr.ss_family == AF_INET)
1831                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1832                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1833                                 port, err ? "failed" : "ok", err);
1834         else
1835                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1836                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1837                                 port, err ? "failed" : "ok", err);
1838         return err;
1839 }
1840
1841 /*
1842  * We don't support autobind on AF_LOCAL sockets
1843  */
1844 static void xs_local_rpcbind(struct rpc_task *task)
1845 {
1846         xprt_set_bound(task->tk_xprt);
1847 }
1848
1849 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1850 {
1851 }
1852
1853 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1854 static struct lock_class_key xs_key[2];
1855 static struct lock_class_key xs_slock_key[2];
1856
1857 static inline void xs_reclassify_socketu(struct socket *sock)
1858 {
1859         struct sock *sk = sock->sk;
1860
1861         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1862                 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1863 }
1864
1865 static inline void xs_reclassify_socket4(struct socket *sock)
1866 {
1867         struct sock *sk = sock->sk;
1868
1869         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1870                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1871 }
1872
1873 static inline void xs_reclassify_socket6(struct socket *sock)
1874 {
1875         struct sock *sk = sock->sk;
1876
1877         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1878                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1879 }
1880
1881 static inline void xs_reclassify_socket(int family, struct socket *sock)
1882 {
1883         if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1884                 return;
1885
1886         switch (family) {
1887         case AF_LOCAL:
1888                 xs_reclassify_socketu(sock);
1889                 break;
1890         case AF_INET:
1891                 xs_reclassify_socket4(sock);
1892                 break;
1893         case AF_INET6:
1894                 xs_reclassify_socket6(sock);
1895                 break;
1896         }
1897 }
1898 #else
1899 static inline void xs_reclassify_socket(int family, struct socket *sock)
1900 {
1901 }
1902 #endif
1903
1904 static void xs_dummy_setup_socket(struct work_struct *work)
1905 {
1906 }
1907
1908 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1909                 struct sock_xprt *transport, int family, int type,
1910                 int protocol, bool reuseport)
1911 {
1912         struct socket *sock;
1913         int err;
1914
1915         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1916         if (err < 0) {
1917                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1918                                 protocol, -err);
1919                 goto out;
1920         }
1921         xs_reclassify_socket(family, sock);
1922
1923         if (reuseport)
1924                 xs_sock_set_reuseport(sock);
1925
1926         err = xs_bind(transport, sock);
1927         if (err) {
1928                 sock_release(sock);
1929                 goto out;
1930         }
1931
1932         return sock;
1933 out:
1934         return ERR_PTR(err);
1935 }
1936
1937 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1938                                       struct socket *sock)
1939 {
1940         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1941                                                                         xprt);
1942
1943         if (!transport->inet) {
1944                 struct sock *sk = sock->sk;
1945
1946                 write_lock_bh(&sk->sk_callback_lock);
1947
1948                 xs_save_old_callbacks(transport, sk);
1949
1950                 sk->sk_user_data = xprt;
1951                 sk->sk_data_ready = xs_data_ready;
1952                 sk->sk_write_space = xs_udp_write_space;
1953                 sock_set_flag(sk, SOCK_FASYNC);
1954                 sk->sk_error_report = xs_error_report;
1955                 sk->sk_allocation = GFP_NOIO;
1956
1957                 xprt_clear_connected(xprt);
1958
1959                 /* Reset to new socket */
1960                 transport->sock = sock;
1961                 transport->inet = sk;
1962
1963                 write_unlock_bh(&sk->sk_callback_lock);
1964         }
1965
1966         /* Tell the socket layer to start connecting... */
1967         xprt->stat.connect_count++;
1968         xprt->stat.connect_start = jiffies;
1969         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1970 }
1971
1972 /**
1973  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1974  * @transport: socket transport to connect
1975  */
1976 static int xs_local_setup_socket(struct sock_xprt *transport)
1977 {
1978         struct rpc_xprt *xprt = &transport->xprt;
1979         struct socket *sock;
1980         int status = -EIO;
1981
1982         status = __sock_create(xprt->xprt_net, AF_LOCAL,
1983                                         SOCK_STREAM, 0, &sock, 1);
1984         if (status < 0) {
1985                 dprintk("RPC:       can't create AF_LOCAL "
1986                         "transport socket (%d).\n", -status);
1987                 goto out;
1988         }
1989         xs_reclassify_socket(AF_LOCAL, sock);
1990
1991         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
1992                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1993
1994         status = xs_local_finish_connecting(xprt, sock);
1995         trace_rpc_socket_connect(xprt, sock, status);
1996         switch (status) {
1997         case 0:
1998                 dprintk("RPC:       xprt %p connected to %s\n",
1999                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2000                 xprt_set_connected(xprt);
2001         case -ENOBUFS:
2002                 break;
2003         case -ENOENT:
2004                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
2005                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2006                 break;
2007         case -ECONNREFUSED:
2008                 dprintk("RPC:       xprt %p: connection refused for %s\n",
2009                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2010                 break;
2011         default:
2012                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2013                                 __func__, -status,
2014                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
2015         }
2016
2017 out:
2018         xprt_clear_connecting(xprt);
2019         xprt_wake_pending_tasks(xprt, status);
2020         return status;
2021 }
2022
2023 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2024 {
2025         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2026         int ret;
2027
2028          if (RPC_IS_ASYNC(task)) {
2029                 /*
2030                  * We want the AF_LOCAL connect to be resolved in the
2031                  * filesystem namespace of the process making the rpc
2032                  * call.  Thus we connect synchronously.
2033                  *
2034                  * If we want to support asynchronous AF_LOCAL calls,
2035                  * we'll need to figure out how to pass a namespace to
2036                  * connect.
2037                  */
2038                 rpc_exit(task, -ENOTCONN);
2039                 return;
2040         }
2041         ret = xs_local_setup_socket(transport);
2042         if (ret && !RPC_IS_SOFTCONN(task))
2043                 msleep_interruptible(15000);
2044 }
2045
2046 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2047 /*
2048  * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2049  * know that we have exclusive access to the socket), to guard against
2050  * races with xs_reset_transport.
2051  */
2052 static void xs_set_memalloc(struct rpc_xprt *xprt)
2053 {
2054         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2055                         xprt);
2056
2057         /*
2058          * If there's no sock, then we have nothing to set. The
2059          * reconnecting process will get it for us.
2060          */
2061         if (!transport->inet)
2062                 return;
2063         if (atomic_read(&xprt->swapper))
2064                 sk_set_memalloc(transport->inet);
2065 }
2066
2067 /**
2068  * xs_enable_swap - Tag this transport as being used for swap.
2069  * @xprt: transport to tag
2070  *
2071  * Take a reference to this transport on behalf of the rpc_clnt, and
2072  * optionally mark it for swapping if it wasn't already.
2073  */
2074 static int
2075 xs_enable_swap(struct rpc_xprt *xprt)
2076 {
2077         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2078
2079         if (atomic_inc_return(&xprt->swapper) != 1)
2080                 return 0;
2081         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2082                 return -ERESTARTSYS;
2083         if (xs->inet)
2084                 sk_set_memalloc(xs->inet);
2085         xprt_release_xprt(xprt, NULL);
2086         return 0;
2087 }
2088
2089 /**
2090  * xs_disable_swap - Untag this transport as being used for swap.
2091  * @xprt: transport to tag
2092  *
2093  * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2094  * swapper refcount goes to 0, untag the socket as a memalloc socket.
2095  */
2096 static void
2097 xs_disable_swap(struct rpc_xprt *xprt)
2098 {
2099         struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2100
2101         if (!atomic_dec_and_test(&xprt->swapper))
2102                 return;
2103         if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2104                 return;
2105         if (xs->inet)
2106                 sk_clear_memalloc(xs->inet);
2107         xprt_release_xprt(xprt, NULL);
2108 }
2109 #else
2110 static void xs_set_memalloc(struct rpc_xprt *xprt)
2111 {
2112 }
2113
2114 static int
2115 xs_enable_swap(struct rpc_xprt *xprt)
2116 {
2117         return -EINVAL;
2118 }
2119
2120 static void
2121 xs_disable_swap(struct rpc_xprt *xprt)
2122 {
2123 }
2124 #endif
2125
2126 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2127 {
2128         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2129
2130         if (!transport->inet) {
2131                 struct sock *sk = sock->sk;
2132
2133                 write_lock_bh(&sk->sk_callback_lock);
2134
2135                 xs_save_old_callbacks(transport, sk);
2136
2137                 sk->sk_user_data = xprt;
2138                 sk->sk_data_ready = xs_data_ready;
2139                 sk->sk_write_space = xs_udp_write_space;
2140                 sock_set_flag(sk, SOCK_FASYNC);
2141                 sk->sk_allocation = GFP_NOIO;
2142
2143                 xprt_set_connected(xprt);
2144
2145                 /* Reset to new socket */
2146                 transport->sock = sock;
2147                 transport->inet = sk;
2148
2149                 xs_set_memalloc(xprt);
2150
2151                 write_unlock_bh(&sk->sk_callback_lock);
2152         }
2153         xs_udp_do_set_buffer_size(xprt);
2154 }
2155
2156 static void xs_udp_setup_socket(struct work_struct *work)
2157 {
2158         struct sock_xprt *transport =
2159                 container_of(work, struct sock_xprt, connect_worker.work);
2160         struct rpc_xprt *xprt = &transport->xprt;
2161         struct socket *sock = transport->sock;
2162         int status = -EIO;
2163
2164         sock = xs_create_sock(xprt, transport,
2165                         xs_addr(xprt)->sa_family, SOCK_DGRAM,
2166                         IPPROTO_UDP, false);
2167         if (IS_ERR(sock))
2168                 goto out;
2169
2170         dprintk("RPC:       worker connecting xprt %p via %s to "
2171                                 "%s (port %s)\n", xprt,
2172                         xprt->address_strings[RPC_DISPLAY_PROTO],
2173                         xprt->address_strings[RPC_DISPLAY_ADDR],
2174                         xprt->address_strings[RPC_DISPLAY_PORT]);
2175
2176         xs_udp_finish_connecting(xprt, sock);
2177         trace_rpc_socket_connect(xprt, sock, 0);
2178         status = 0;
2179 out:
2180         xprt_unlock_connect(xprt, transport);
2181         xprt_clear_connecting(xprt);
2182         xprt_wake_pending_tasks(xprt, status);
2183 }
2184
2185 /**
2186  * xs_tcp_shutdown - gracefully shut down a TCP socket
2187  * @xprt: transport
2188  *
2189  * Initiates a graceful shutdown of the TCP socket by calling the
2190  * equivalent of shutdown(SHUT_RDWR);
2191  */
2192 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2193 {
2194         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2195         struct socket *sock = transport->sock;
2196
2197         if (sock == NULL)
2198                 return;
2199         if (xprt_connected(xprt)) {
2200                 kernel_sock_shutdown(sock, SHUT_RDWR);
2201                 trace_rpc_socket_shutdown(xprt, sock);
2202         } else
2203                 xs_reset_transport(transport);
2204 }
2205
2206 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2207 {
2208         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2209         int ret = -ENOTCONN;
2210
2211         if (!transport->inet) {
2212                 struct sock *sk = sock->sk;
2213                 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2214                 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2215                 unsigned int opt_on = 1;
2216                 unsigned int timeo;
2217
2218                 /* TCP Keepalive options */
2219                 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2220                                 (char *)&opt_on, sizeof(opt_on));
2221                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2222                                 (char *)&keepidle, sizeof(keepidle));
2223                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2224                                 (char *)&keepidle, sizeof(keepidle));
2225                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2226                                 (char *)&keepcnt, sizeof(keepcnt));
2227
2228                 /* TCP user timeout (see RFC5482) */
2229                 timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2230                         (xprt->timeout->to_retries + 1);
2231                 kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2232                                 (char *)&timeo, sizeof(timeo));
2233
2234                 write_lock_bh(&sk->sk_callback_lock);
2235
2236                 xs_save_old_callbacks(transport, sk);
2237
2238                 sk->sk_user_data = xprt;
2239                 sk->sk_data_ready = xs_tcp_data_ready;
2240                 sk->sk_state_change = xs_tcp_state_change;
2241                 sk->sk_write_space = xs_tcp_write_space;
2242                 sock_set_flag(sk, SOCK_FASYNC);
2243                 sk->sk_error_report = xs_error_report;
2244                 sk->sk_allocation = GFP_NOIO;
2245
2246                 /* socket options */
2247                 sock_reset_flag(sk, SOCK_LINGER);
2248                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2249
2250                 xprt_clear_connected(xprt);
2251
2252                 /* Reset to new socket */
2253                 transport->sock = sock;
2254                 transport->inet = sk;
2255
2256                 write_unlock_bh(&sk->sk_callback_lock);
2257         }
2258
2259         if (!xprt_bound(xprt))
2260                 goto out;
2261
2262         xs_set_memalloc(xprt);
2263
2264         /* Tell the socket layer to start connecting... */
2265         xprt->stat.connect_count++;
2266         xprt->stat.connect_start = jiffies;
2267         set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2268         ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2269         switch (ret) {
2270         case 0:
2271                 xs_set_srcport(transport, sock);
2272         case -EINPROGRESS:
2273                 /* SYN_SENT! */
2274                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2275                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2276         }
2277 out:
2278         return ret;
2279 }
2280
2281 /**
2282  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2283  *
2284  * Invoked by a work queue tasklet.
2285  */
2286 static void xs_tcp_setup_socket(struct work_struct *work)
2287 {
2288         struct sock_xprt *transport =
2289                 container_of(work, struct sock_xprt, connect_worker.work);
2290         struct socket *sock = transport->sock;
2291         struct rpc_xprt *xprt = &transport->xprt;
2292         int status = -EIO;
2293
2294         if (!sock) {
2295                 sock = xs_create_sock(xprt, transport,
2296                                 xs_addr(xprt)->sa_family, SOCK_STREAM,
2297                                 IPPROTO_TCP, true);
2298                 if (IS_ERR(sock)) {
2299                         status = PTR_ERR(sock);
2300                         goto out;
2301                 }
2302         }
2303
2304         dprintk("RPC:       worker connecting xprt %p via %s to "
2305                                 "%s (port %s)\n", xprt,
2306                         xprt->address_strings[RPC_DISPLAY_PROTO],
2307                         xprt->address_strings[RPC_DISPLAY_ADDR],
2308                         xprt->address_strings[RPC_DISPLAY_PORT]);
2309
2310         status = xs_tcp_finish_connecting(xprt, sock);
2311         trace_rpc_socket_connect(xprt, sock, status);
2312         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2313                         xprt, -status, xprt_connected(xprt),
2314                         sock->sk->sk_state);
2315         switch (status) {
2316         default:
2317                 printk("%s: connect returned unhandled error %d\n",
2318                         __func__, status);
2319         case -EADDRNOTAVAIL:
2320                 /* We're probably in TIME_WAIT. Get rid of existing socket,
2321                  * and retry
2322                  */
2323                 xs_tcp_force_close(xprt);
2324                 break;
2325         case 0:
2326         case -EINPROGRESS:
2327         case -EALREADY:
2328                 xprt_unlock_connect(xprt, transport);
2329                 return;
2330         case -EINVAL:
2331                 /* Happens, for instance, if the user specified a link
2332                  * local IPv6 address without a scope-id.
2333                  */
2334         case -ECONNREFUSED:
2335         case -ECONNRESET:
2336         case -ENETUNREACH:
2337         case -EADDRINUSE:
2338         case -ENOBUFS:
2339                 /* retry with existing socket, after a delay */
2340                 xs_tcp_force_close(xprt);
2341                 goto out;
2342         }
2343         status = -EAGAIN;
2344 out:
2345         xprt_unlock_connect(xprt, transport);
2346         xprt_clear_connecting(xprt);
2347         xprt_wake_pending_tasks(xprt, status);
2348 }
2349
2350 /**
2351  * xs_connect - connect a socket to a remote endpoint
2352  * @xprt: pointer to transport structure
2353  * @task: address of RPC task that manages state of connect request
2354  *
2355  * TCP: If the remote end dropped the connection, delay reconnecting.
2356  *
2357  * UDP socket connects are synchronous, but we use a work queue anyway
2358  * to guarantee that even unprivileged user processes can set up a
2359  * socket on a privileged port.
2360  *
2361  * If a UDP socket connect fails, the delay behavior here prevents
2362  * retry floods (hard mounts).
2363  */
2364 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2365 {
2366         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2367
2368         WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2369
2370         if (transport->sock != NULL) {
2371                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2372                                 "seconds\n",
2373                                 xprt, xprt->reestablish_timeout / HZ);
2374
2375                 /* Start by resetting any existing state */
2376                 xs_reset_transport(transport);
2377
2378                 queue_delayed_work(rpciod_workqueue,
2379                                    &transport->connect_worker,
2380                                    xprt->reestablish_timeout);
2381                 xprt->reestablish_timeout <<= 1;
2382                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2383                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2384                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2385                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2386         } else {
2387                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2388                 queue_delayed_work(rpciod_workqueue,
2389                                    &transport->connect_worker, 0);
2390         }
2391 }
2392
2393 /**
2394  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2395  * @xprt: rpc_xprt struct containing statistics
2396  * @seq: output file
2397  *
2398  */
2399 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2400 {
2401         long idle_time = 0;
2402
2403         if (xprt_connected(xprt))
2404                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2405
2406         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2407                         "%llu %llu %lu %llu %llu\n",
2408                         xprt->stat.bind_count,
2409                         xprt->stat.connect_count,
2410                         xprt->stat.connect_time,
2411                         idle_time,
2412                         xprt->stat.sends,
2413                         xprt->stat.recvs,
2414                         xprt->stat.bad_xids,
2415                         xprt->stat.req_u,
2416                         xprt->stat.bklog_u,
2417                         xprt->stat.max_slots,
2418                         xprt->stat.sending_u,
2419                         xprt->stat.pending_u);
2420 }
2421
2422 /**
2423  * xs_udp_print_stats - display UDP socket-specifc stats
2424  * @xprt: rpc_xprt struct containing statistics
2425  * @seq: output file
2426  *
2427  */
2428 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2429 {
2430         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2431
2432         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2433                         "%lu %llu %llu\n",
2434                         transport->srcport,
2435                         xprt->stat.bind_count,
2436                         xprt->stat.sends,
2437                         xprt->stat.recvs,
2438                         xprt->stat.bad_xids,
2439                         xprt->stat.req_u,
2440                         xprt->stat.bklog_u,
2441                         xprt->stat.max_slots,
2442                         xprt->stat.sending_u,
2443                         xprt->stat.pending_u);
2444 }
2445
2446 /**
2447  * xs_tcp_print_stats - display TCP socket-specifc stats
2448  * @xprt: rpc_xprt struct containing statistics
2449  * @seq: output file
2450  *
2451  */
2452 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2453 {
2454         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2455         long idle_time = 0;
2456
2457         if (xprt_connected(xprt))
2458                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2459
2460         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2461                         "%llu %llu %lu %llu %llu\n",
2462                         transport->srcport,
2463                         xprt->stat.bind_count,
2464                         xprt->stat.connect_count,
2465                         xprt->stat.connect_time,
2466                         idle_time,
2467                         xprt->stat.sends,
2468                         xprt->stat.recvs,
2469                         xprt->stat.bad_xids,
2470                         xprt->stat.req_u,
2471                         xprt->stat.bklog_u,
2472                         xprt->stat.max_slots,
2473                         xprt->stat.sending_u,
2474                         xprt->stat.pending_u);
2475 }
2476
2477 /*
2478  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2479  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2480  * to use the server side send routines.
2481  */
2482 static void *bc_malloc(struct rpc_task *task, size_t size)
2483 {
2484         struct page *page;
2485         struct rpc_buffer *buf;
2486
2487         WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2488         if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2489                 return NULL;
2490
2491         page = alloc_page(GFP_KERNEL);
2492         if (!page)
2493                 return NULL;
2494
2495         buf = page_address(page);
2496         buf->len = PAGE_SIZE;
2497
2498         return buf->data;
2499 }
2500
2501 /*
2502  * Free the space allocated in the bc_alloc routine
2503  */
2504 static void bc_free(void *buffer)
2505 {
2506         struct rpc_buffer *buf;
2507
2508         if (!buffer)
2509                 return;
2510
2511         buf = container_of(buffer, struct rpc_buffer, data);
2512         free_page((unsigned long)buf);
2513 }
2514
2515 /*
2516  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2517  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2518  */
2519 static int bc_sendto(struct rpc_rqst *req)
2520 {
2521         int len;
2522         struct xdr_buf *xbufp = &req->rq_snd_buf;
2523         struct rpc_xprt *xprt = req->rq_xprt;
2524         struct sock_xprt *transport =
2525                                 container_of(xprt, struct sock_xprt, xprt);
2526         struct socket *sock = transport->sock;
2527         unsigned long headoff;
2528         unsigned long tailoff;
2529
2530         xs_encode_stream_record_marker(xbufp);
2531
2532         tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2533         headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2534         len = svc_send_common(sock, xbufp,
2535                               virt_to_page(xbufp->head[0].iov_base), headoff,
2536                               xbufp->tail[0].iov_base, tailoff);
2537
2538         if (len != xbufp->len) {
2539                 printk(KERN_NOTICE "Error sending entire callback!\n");
2540                 len = -EAGAIN;
2541         }
2542
2543         return len;
2544 }
2545
2546 /*
2547  * The send routine. Borrows from svc_send
2548  */
2549 static int bc_send_request(struct rpc_task *task)
2550 {
2551         struct rpc_rqst *req = task->tk_rqstp;
2552         struct svc_xprt *xprt;
2553         int len;
2554
2555         dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2556         /*
2557          * Get the server socket associated with this callback xprt
2558          */
2559         xprt = req->rq_xprt->bc_xprt;
2560
2561         /*
2562          * Grab the mutex to serialize data as the connection is shared
2563          * with the fore channel
2564          */
2565         if (!mutex_trylock(&xprt->xpt_mutex)) {
2566                 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2567                 if (!mutex_trylock(&xprt->xpt_mutex))
2568                         return -EAGAIN;
2569                 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2570         }
2571         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2572                 len = -ENOTCONN;
2573         else
2574                 len = bc_sendto(req);
2575         mutex_unlock(&xprt->xpt_mutex);
2576
2577         if (len > 0)
2578                 len = 0;
2579
2580         return len;
2581 }
2582
2583 /*
2584  * The close routine. Since this is client initiated, we do nothing
2585  */
2586
2587 static void bc_close(struct rpc_xprt *xprt)
2588 {
2589 }
2590
2591 /*
2592  * The xprt destroy routine. Again, because this connection is client
2593  * initiated, we do nothing
2594  */
2595
2596 static void bc_destroy(struct rpc_xprt *xprt)
2597 {
2598         dprintk("RPC:       bc_destroy xprt %p\n", xprt);
2599
2600         xs_xprt_free(xprt);
2601         module_put(THIS_MODULE);
2602 }
2603
2604 static struct rpc_xprt_ops xs_local_ops = {
2605         .reserve_xprt           = xprt_reserve_xprt,
2606         .release_xprt           = xs_tcp_release_xprt,
2607         .alloc_slot             = xprt_alloc_slot,
2608         .rpcbind                = xs_local_rpcbind,
2609         .set_port               = xs_local_set_port,
2610         .connect                = xs_local_connect,
2611         .buf_alloc              = rpc_malloc,
2612         .buf_free               = rpc_free,
2613         .send_request           = xs_local_send_request,
2614         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2615         .close                  = xs_close,
2616         .destroy                = xs_destroy,
2617         .print_stats            = xs_local_print_stats,
2618         .enable_swap            = xs_enable_swap,
2619         .disable_swap           = xs_disable_swap,
2620 };
2621
2622 static struct rpc_xprt_ops xs_udp_ops = {
2623         .set_buffer_size        = xs_udp_set_buffer_size,
2624         .reserve_xprt           = xprt_reserve_xprt_cong,
2625         .release_xprt           = xprt_release_xprt_cong,
2626         .alloc_slot             = xprt_alloc_slot,
2627         .rpcbind                = rpcb_getport_async,
2628         .set_port               = xs_set_port,
2629         .connect                = xs_connect,
2630         .buf_alloc              = rpc_malloc,
2631         .buf_free               = rpc_free,
2632         .send_request           = xs_udp_send_request,
2633         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
2634         .timer                  = xs_udp_timer,
2635         .release_request        = xprt_release_rqst_cong,
2636         .close                  = xs_close,
2637         .destroy                = xs_destroy,
2638         .print_stats            = xs_udp_print_stats,
2639         .enable_swap            = xs_enable_swap,
2640         .disable_swap           = xs_disable_swap,
2641         .inject_disconnect      = xs_inject_disconnect,
2642 };
2643
2644 static struct rpc_xprt_ops xs_tcp_ops = {
2645         .reserve_xprt           = xprt_reserve_xprt,
2646         .release_xprt           = xs_tcp_release_xprt,
2647         .alloc_slot             = xprt_lock_and_alloc_slot,
2648         .rpcbind                = rpcb_getport_async,
2649         .set_port               = xs_set_port,
2650         .connect                = xs_connect,
2651         .buf_alloc              = rpc_malloc,
2652         .buf_free               = rpc_free,
2653         .send_request           = xs_tcp_send_request,
2654         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2655         .close                  = xs_tcp_shutdown,
2656         .destroy                = xs_destroy,
2657         .print_stats            = xs_tcp_print_stats,
2658         .enable_swap            = xs_enable_swap,
2659         .disable_swap           = xs_disable_swap,
2660         .inject_disconnect      = xs_inject_disconnect,
2661 #ifdef CONFIG_SUNRPC_BACKCHANNEL
2662         .bc_setup               = xprt_setup_bc,
2663         .bc_up                  = xs_tcp_bc_up,
2664         .bc_free_rqst           = xprt_free_bc_rqst,
2665         .bc_destroy             = xprt_destroy_bc,
2666 #endif
2667 };
2668
2669 /*
2670  * The rpc_xprt_ops for the server backchannel
2671  */
2672
2673 static struct rpc_xprt_ops bc_tcp_ops = {
2674         .reserve_xprt           = xprt_reserve_xprt,
2675         .release_xprt           = xprt_release_xprt,
2676         .alloc_slot             = xprt_alloc_slot,
2677         .buf_alloc              = bc_malloc,
2678         .buf_free               = bc_free,
2679         .send_request           = bc_send_request,
2680         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2681         .close                  = bc_close,
2682         .destroy                = bc_destroy,
2683         .print_stats            = xs_tcp_print_stats,
2684         .enable_swap            = xs_enable_swap,
2685         .disable_swap           = xs_disable_swap,
2686         .inject_disconnect      = xs_inject_disconnect,
2687 };
2688
2689 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2690 {
2691         static const struct sockaddr_in sin = {
2692                 .sin_family             = AF_INET,
2693                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2694         };
2695         static const struct sockaddr_in6 sin6 = {
2696                 .sin6_family            = AF_INET6,
2697                 .sin6_addr              = IN6ADDR_ANY_INIT,
2698         };
2699
2700         switch (family) {
2701         case AF_LOCAL:
2702                 break;
2703         case AF_INET:
2704                 memcpy(sap, &sin, sizeof(sin));
2705                 break;
2706         case AF_INET6:
2707                 memcpy(sap, &sin6, sizeof(sin6));
2708                 break;
2709         default:
2710                 dprintk("RPC:       %s: Bad address family\n", __func__);
2711                 return -EAFNOSUPPORT;
2712         }
2713         return 0;
2714 }
2715
2716 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2717                                       unsigned int slot_table_size,
2718                                       unsigned int max_slot_table_size)
2719 {
2720         struct rpc_xprt *xprt;
2721         struct sock_xprt *new;
2722
2723         if (args->addrlen > sizeof(xprt->addr)) {
2724                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2725                 return ERR_PTR(-EBADF);
2726         }
2727
2728         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2729                         max_slot_table_size);
2730         if (xprt == NULL) {
2731                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2732                                 "rpc_xprt\n");
2733                 return ERR_PTR(-ENOMEM);
2734         }
2735
2736         new = container_of(xprt, struct sock_xprt, xprt);
2737         mutex_init(&new->recv_mutex);
2738         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2739         xprt->addrlen = args->addrlen;
2740         if (args->srcaddr)
2741                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2742         else {
2743                 int err;
2744                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2745                                         (struct sockaddr *)&new->srcaddr);
2746                 if (err != 0) {
2747                         xprt_free(xprt);
2748                         return ERR_PTR(err);
2749                 }
2750         }
2751
2752         return xprt;
2753 }
2754
2755 static const struct rpc_timeout xs_local_default_timeout = {
2756         .to_initval = 10 * HZ,
2757         .to_maxval = 10 * HZ,
2758         .to_retries = 2,
2759 };
2760
2761 /**
2762  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2763  * @args: rpc transport creation arguments
2764  *
2765  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2766  */
2767 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2768 {
2769         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2770         struct sock_xprt *transport;
2771         struct rpc_xprt *xprt;
2772         struct rpc_xprt *ret;
2773
2774         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2775                         xprt_max_tcp_slot_table_entries);
2776         if (IS_ERR(xprt))
2777                 return xprt;
2778         transport = container_of(xprt, struct sock_xprt, xprt);
2779
2780         xprt->prot = 0;
2781         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2782         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2783
2784         xprt->bind_timeout = XS_BIND_TO;
2785         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2786         xprt->idle_timeout = XS_IDLE_DISC_TO;
2787
2788         xprt->ops = &xs_local_ops;
2789         xprt->timeout = &xs_local_default_timeout;
2790
2791         INIT_WORK(&transport->recv_worker, xs_local_data_receive_workfn);
2792         INIT_DELAYED_WORK(&transport->connect_worker,
2793                         xs_dummy_setup_socket);
2794
2795         switch (sun->sun_family) {
2796         case AF_LOCAL:
2797                 if (sun->sun_path[0] != '/') {
2798                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
2799                                         sun->sun_path);
2800                         ret = ERR_PTR(-EINVAL);
2801                         goto out_err;
2802                 }
2803                 xprt_set_bound(xprt);
2804                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2805                 ret = ERR_PTR(xs_local_setup_socket(transport));
2806                 if (ret)
2807                         goto out_err;
2808                 break;
2809         default:
2810                 ret = ERR_PTR(-EAFNOSUPPORT);
2811                 goto out_err;
2812         }
2813
2814         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2815                         xprt->address_strings[RPC_DISPLAY_ADDR]);
2816
2817         if (try_module_get(THIS_MODULE))
2818                 return xprt;
2819         ret = ERR_PTR(-EINVAL);
2820 out_err:
2821         xs_xprt_free(xprt);
2822         return ret;
2823 }
2824
2825 static const struct rpc_timeout xs_udp_default_timeout = {
2826         .to_initval = 5 * HZ,
2827         .to_maxval = 30 * HZ,
2828         .to_increment = 5 * HZ,
2829         .to_retries = 5,
2830 };
2831
2832 /**
2833  * xs_setup_udp - Set up transport to use a UDP socket
2834  * @args: rpc transport creation arguments
2835  *
2836  */
2837 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2838 {
2839         struct sockaddr *addr = args->dstaddr;
2840         struct rpc_xprt *xprt;
2841         struct sock_xprt *transport;
2842         struct rpc_xprt *ret;
2843
2844         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2845                         xprt_udp_slot_table_entries);
2846         if (IS_ERR(xprt))
2847                 return xprt;
2848         transport = container_of(xprt, struct sock_xprt, xprt);
2849
2850         xprt->prot = IPPROTO_UDP;
2851         xprt->tsh_size = 0;
2852         /* XXX: header size can vary due to auth type, IPv6, etc. */
2853         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2854
2855         xprt->bind_timeout = XS_BIND_TO;
2856         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2857         xprt->idle_timeout = XS_IDLE_DISC_TO;
2858
2859         xprt->ops = &xs_udp_ops;
2860
2861         xprt->timeout = &xs_udp_default_timeout;
2862
2863         INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2864         INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2865
2866         switch (addr->sa_family) {
2867         case AF_INET:
2868                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2869                         xprt_set_bound(xprt);
2870
2871                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2872                 break;
2873         case AF_INET6:
2874                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2875                         xprt_set_bound(xprt);
2876
2877                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2878                 break;
2879         default:
2880                 ret = ERR_PTR(-EAFNOSUPPORT);
2881                 goto out_err;
2882         }
2883
2884         if (xprt_bound(xprt))
2885                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2886                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2887                                 xprt->address_strings[RPC_DISPLAY_PORT],
2888                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2889         else
2890                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2891                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2892                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2893
2894         if (try_module_get(THIS_MODULE))
2895                 return xprt;
2896         ret = ERR_PTR(-EINVAL);
2897 out_err:
2898         xs_xprt_free(xprt);
2899         return ret;
2900 }
2901
2902 static const struct rpc_timeout xs_tcp_default_timeout = {
2903         .to_initval = 60 * HZ,
2904         .to_maxval = 60 * HZ,
2905         .to_retries = 2,
2906 };
2907
2908 /**
2909  * xs_setup_tcp - Set up transport to use a TCP socket
2910  * @args: rpc transport creation arguments
2911  *
2912  */
2913 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2914 {
2915         struct sockaddr *addr = args->dstaddr;
2916         struct rpc_xprt *xprt;
2917         struct sock_xprt *transport;
2918         struct rpc_xprt *ret;
2919         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2920
2921         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2922                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2923
2924         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2925                         max_slot_table_size);
2926         if (IS_ERR(xprt))
2927                 return xprt;
2928         transport = container_of(xprt, struct sock_xprt, xprt);
2929
2930         xprt->prot = IPPROTO_TCP;
2931         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2932         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2933
2934         xprt->bind_timeout = XS_BIND_TO;
2935         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2936         xprt->idle_timeout = XS_IDLE_DISC_TO;
2937
2938         xprt->ops = &xs_tcp_ops;
2939         xprt->timeout = &xs_tcp_default_timeout;
2940
2941         INIT_WORK(&transport->recv_worker, xs_tcp_data_receive_workfn);
2942         INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
2943
2944         switch (addr->sa_family) {
2945         case AF_INET:
2946                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2947                         xprt_set_bound(xprt);
2948
2949                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2950                 break;
2951         case AF_INET6:
2952                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2953                         xprt_set_bound(xprt);
2954
2955                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2956                 break;
2957         default:
2958                 ret = ERR_PTR(-EAFNOSUPPORT);
2959                 goto out_err;
2960         }
2961
2962         if (xprt_bound(xprt))
2963                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2964                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2965                                 xprt->address_strings[RPC_DISPLAY_PORT],
2966                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2967         else
2968                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2969                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2970                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2971
2972         if (try_module_get(THIS_MODULE))
2973                 return xprt;
2974         ret = ERR_PTR(-EINVAL);
2975 out_err:
2976         xs_xprt_free(xprt);
2977         return ret;
2978 }
2979
2980 /**
2981  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2982  * @args: rpc transport creation arguments
2983  *
2984  */
2985 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2986 {
2987         struct sockaddr *addr = args->dstaddr;
2988         struct rpc_xprt *xprt;
2989         struct sock_xprt *transport;
2990         struct svc_sock *bc_sock;
2991         struct rpc_xprt *ret;
2992
2993         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2994                         xprt_tcp_slot_table_entries);
2995         if (IS_ERR(xprt))
2996                 return xprt;
2997         transport = container_of(xprt, struct sock_xprt, xprt);
2998
2999         xprt->prot = IPPROTO_TCP;
3000         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3001         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3002         xprt->timeout = &xs_tcp_default_timeout;
3003
3004         /* backchannel */
3005         xprt_set_bound(xprt);
3006         xprt->bind_timeout = 0;
3007         xprt->reestablish_timeout = 0;
3008         xprt->idle_timeout = 0;
3009
3010         xprt->ops = &bc_tcp_ops;
3011
3012         switch (addr->sa_family) {
3013         case AF_INET:
3014                 xs_format_peer_addresses(xprt, "tcp",
3015                                          RPCBIND_NETID_TCP);
3016                 break;
3017         case AF_INET6:
3018                 xs_format_peer_addresses(xprt, "tcp",
3019                                    RPCBIND_NETID_TCP6);
3020                 break;
3021         default:
3022                 ret = ERR_PTR(-EAFNOSUPPORT);
3023                 goto out_err;
3024         }
3025
3026         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3027                         xprt->address_strings[RPC_DISPLAY_ADDR],
3028                         xprt->address_strings[RPC_DISPLAY_PORT],
3029                         xprt->address_strings[RPC_DISPLAY_PROTO]);
3030
3031         /*
3032          * Once we've associated a backchannel xprt with a connection,
3033          * we want to keep it around as long as the connection lasts,
3034          * in case we need to start using it for a backchannel again;
3035          * this reference won't be dropped until bc_xprt is destroyed.
3036          */
3037         xprt_get(xprt);
3038         args->bc_xprt->xpt_bc_xprt = xprt;
3039         xprt->bc_xprt = args->bc_xprt;
3040         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3041         transport->sock = bc_sock->sk_sock;
3042         transport->inet = bc_sock->sk_sk;
3043
3044         /*
3045          * Since we don't want connections for the backchannel, we set
3046          * the xprt status to connected
3047          */
3048         xprt_set_connected(xprt);
3049
3050         if (try_module_get(THIS_MODULE))
3051                 return xprt;
3052
3053         args->bc_xprt->xpt_bc_xprt = NULL;
3054         xprt_put(xprt);
3055         ret = ERR_PTR(-EINVAL);
3056 out_err:
3057         xs_xprt_free(xprt);
3058         return ret;
3059 }
3060
3061 static struct xprt_class        xs_local_transport = {
3062         .list           = LIST_HEAD_INIT(xs_local_transport.list),
3063         .name           = "named UNIX socket",
3064         .owner          = THIS_MODULE,
3065         .ident          = XPRT_TRANSPORT_LOCAL,
3066         .setup          = xs_setup_local,
3067 };
3068
3069 static struct xprt_class        xs_udp_transport = {
3070         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
3071         .name           = "udp",
3072         .owner          = THIS_MODULE,
3073         .ident          = XPRT_TRANSPORT_UDP,
3074         .setup          = xs_setup_udp,
3075 };
3076
3077 static struct xprt_class        xs_tcp_transport = {
3078         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
3079         .name           = "tcp",
3080         .owner          = THIS_MODULE,
3081         .ident          = XPRT_TRANSPORT_TCP,
3082         .setup          = xs_setup_tcp,
3083 };
3084
3085 static struct xprt_class        xs_bc_tcp_transport = {
3086         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3087         .name           = "tcp NFSv4.1 backchannel",
3088         .owner          = THIS_MODULE,
3089         .ident          = XPRT_TRANSPORT_BC_TCP,
3090         .setup          = xs_setup_bc_tcp,
3091 };
3092
3093 /**
3094  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3095  *
3096  */
3097 int init_socket_xprt(void)
3098 {
3099 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3100         if (!sunrpc_table_header)
3101                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3102 #endif
3103
3104         xprt_register_transport(&xs_local_transport);
3105         xprt_register_transport(&xs_udp_transport);
3106         xprt_register_transport(&xs_tcp_transport);
3107         xprt_register_transport(&xs_bc_tcp_transport);
3108
3109         return 0;
3110 }
3111
3112 /**
3113  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3114  *
3115  */
3116 void cleanup_socket_xprt(void)
3117 {
3118 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3119         if (sunrpc_table_header) {
3120                 unregister_sysctl_table(sunrpc_table_header);
3121                 sunrpc_table_header = NULL;
3122         }
3123 #endif
3124
3125         xprt_unregister_transport(&xs_local_transport);
3126         xprt_unregister_transport(&xs_udp_transport);
3127         xprt_unregister_transport(&xs_tcp_transport);
3128         xprt_unregister_transport(&xs_bc_tcp_transport);
3129 }
3130
3131 static int param_set_uint_minmax(const char *val,
3132                 const struct kernel_param *kp,
3133                 unsigned int min, unsigned int max)
3134 {
3135         unsigned int num;
3136         int ret;
3137
3138         if (!val)
3139                 return -EINVAL;
3140         ret = kstrtouint(val, 0, &num);
3141         if (ret == -EINVAL || num < min || num > max)
3142                 return -EINVAL;
3143         *((unsigned int *)kp->arg) = num;
3144         return 0;
3145 }
3146
3147 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3148 {
3149         return param_set_uint_minmax(val, kp,
3150                         RPC_MIN_RESVPORT,
3151                         RPC_MAX_RESVPORT);
3152 }
3153
3154 static const struct kernel_param_ops param_ops_portnr = {
3155         .set = param_set_portnr,
3156         .get = param_get_uint,
3157 };
3158
3159 #define param_check_portnr(name, p) \
3160         __param_check(name, p, unsigned int);
3161
3162 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3163 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3164
3165 static int param_set_slot_table_size(const char *val,
3166                                      const struct kernel_param *kp)
3167 {
3168         return param_set_uint_minmax(val, kp,
3169                         RPC_MIN_SLOT_TABLE,
3170                         RPC_MAX_SLOT_TABLE);
3171 }
3172
3173 static const struct kernel_param_ops param_ops_slot_table_size = {
3174         .set = param_set_slot_table_size,
3175         .get = param_get_uint,
3176 };
3177
3178 #define param_check_slot_table_size(name, p) \
3179         __param_check(name, p, unsigned int);
3180
3181 static int param_set_max_slot_table_size(const char *val,
3182                                      const struct kernel_param *kp)
3183 {
3184         return param_set_uint_minmax(val, kp,
3185                         RPC_MIN_SLOT_TABLE,
3186                         RPC_MAX_SLOT_TABLE_LIMIT);
3187 }
3188
3189 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3190         .set = param_set_max_slot_table_size,
3191         .get = param_get_uint,
3192 };
3193
3194 #define param_check_max_slot_table_size(name, p) \
3195         __param_check(name, p, unsigned int);
3196
3197 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3198                    slot_table_size, 0644);
3199 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3200                    max_slot_table_size, 0644);
3201 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3202                    slot_table_size, 0644);
3203