git.oblomov.eu Git - linux-2.6/blob - fs/dlm/lowcomms-tcp.c

   1 /******************************************************************************
   2 *******************************************************************************
   3 **
   4 **  Copyright (C) Sistina Software, Inc.  1997-2003  All rights reserved.
   5 **  Copyright (C) 2004-2007 Red Hat, Inc.  All rights reserved.
   6 **
   7 **  This copyrighted material is made available to anyone wishing to use,
   8 **  modify, copy, or redistribute it subject to the terms and conditions
   9 **  of the GNU General Public License v.2.
  10 **
  11 *******************************************************************************
  12 ******************************************************************************/
  13
  14 /*
  15  * lowcomms.c
  16  *
  17  * This is the "low-level" comms layer.
  18  *
  19  * It is responsible for sending/receiving messages
  20  * from other nodes in the cluster.
  21  *
  22  * Cluster nodes are referred to by their nodeids. nodeids are
  23  * simply 32 bit numbers to the locking module - if they need to
  24  * be expanded for the cluster infrastructure then that is it's
  25  * responsibility. It is this layer's
  26  * responsibility to resolve these into IP address or
  27  * whatever it needs for inter-node communication.
  28  *
  29  * The comms level is two kernel threads that deal mainly with
  30  * the receiving of messages from other nodes and passing them
  31  * up to the mid-level comms layer (which understands the
  32  * message format) for execution by the locking core, and
  33  * a send thread which does all the setting up of connections
  34  * to remote nodes and the sending of data. Threads are not allowed
  35  * to send their own data because it may cause them to wait in times
  36  * of high load. Also, this way, the sending thread can collect together
  37  * messages bound for one node and send them in one block.
  38  *
  39  * I don't see any problem with the recv thread executing the locking
  40  * code on behalf of remote processes as the locking code is
  41  * short, efficient and never waits.
  42  *
  43  */
  44
  45
  46 #include <asm/ioctls.h>
  47 #include <net/sock.h>
  48 #include <net/tcp.h>
  49 #include <linux/pagemap.h>
  50
  51 #include "dlm_internal.h"
  52 #include "lowcomms.h"
  53 #include "midcomms.h"
  54 #include "config.h"
  55
  56 struct cbuf {
  57         unsigned int base;
  58         unsigned int len;
  59         unsigned int mask;
  60 };
  61
  62 #define NODE_INCREMENT 32
  63 static void cbuf_add(struct cbuf *cb, int n)
  64 {
  65         cb->len += n;
  66 }
  67
  68 static int cbuf_data(struct cbuf *cb)
  69 {
  70         return ((cb->base + cb->len) & cb->mask);
  71 }
  72
  73 static void cbuf_init(struct cbuf *cb, int size)
  74 {
  75         cb->base = cb->len = 0;
  76         cb->mask = size-1;
  77 }
  78
  79 static void cbuf_eat(struct cbuf *cb, int n)
  80 {
  81         cb->len  -= n;
  82         cb->base += n;
  83         cb->base &= cb->mask;
  84 }
  85
  86 static bool cbuf_empty(struct cbuf *cb)
  87 {
  88         return cb->len == 0;
  89 }
  90
  91 /* Maximum number of incoming messages to process before
  92    doing a cond_resched()
  93 */
  94 #define MAX_RX_MSG_COUNT 25
  95
  96 struct connection {
  97         struct socket *sock;    /* NULL if not connected */
  98         uint32_t nodeid;        /* So we know who we are in the list */
  99         struct mutex sock_mutex;
 100         unsigned long flags;    /* bit 1,2 = We are on the read/write lists */
 101 #define CF_READ_PENDING 1
 102 #define CF_WRITE_PENDING 2
 103 #define CF_CONNECT_PENDING 3
 104 #define CF_IS_OTHERCON 4
 105         struct list_head writequeue;  /* List of outgoing writequeue_entries */
 106         struct list_head listenlist;  /* List of allocated listening sockets */
 107         spinlock_t writequeue_lock;
 108         int (*rx_action) (struct connection *); /* What to do when active */
 109         struct page *rx_page;
 110         struct cbuf cb;
 111         int retries;
 112 #define MAX_CONNECT_RETRIES 3
 113         struct connection *othercon;
 114         struct work_struct rwork; /* Receive workqueue */
 115         struct work_struct swork; /* Send workqueue */
 116 };
 117 #define sock2con(x) ((struct connection *)(x)->sk_user_data)
 118
 119 /* An entry waiting to be sent */
 120 struct writequeue_entry {
 121         struct list_head list;
 122         struct page *page;
 123         int offset;
 124         int len;
 125         int end;
 126         int users;
 127         struct connection *con;
 128 };
 129
 130 static struct sockaddr_storage dlm_local_addr;
 131
 132 /* Work queues */
 133 static struct workqueue_struct *recv_workqueue;
 134 static struct workqueue_struct *send_workqueue;
 135
 136 /* An array of pointers to connections, indexed by NODEID */
 137 static struct connection **connections;
 138 static DECLARE_MUTEX(connections_lock);
 139 static struct kmem_cache *con_cache;
 140 static int conn_array_size;
 141
 142 static void process_recv_sockets(struct work_struct *work);
 143 static void process_send_sockets(struct work_struct *work);
 144
 145 static struct connection *nodeid2con(int nodeid, gfp_t allocation)
 146 {
 147         struct connection *con = NULL;
 148
 149         down(&connections_lock);
 150         if (nodeid >= conn_array_size) {
 151                 int new_size = nodeid + NODE_INCREMENT;
 152                 struct connection **new_conns;
 153
 154                 new_conns = kzalloc(sizeof(struct connection *) *
 155                                     new_size, allocation);
 156                 if (!new_conns)
 157                         goto finish;
 158
 159                 memcpy(new_conns, connections,  sizeof(struct connection *) * conn_array_size);
 160                 conn_array_size = new_size;
 161                 kfree(connections);
 162                 connections = new_conns;
 163
 164         }
 165
 166         con = connections[nodeid];
 167         if (con == NULL && allocation) {
 168                 con = kmem_cache_zalloc(con_cache, allocation);
 169                 if (!con)
 170                         goto finish;
 171
 172                 con->nodeid = nodeid;
 173                 mutex_init(&con->sock_mutex);
 174                 INIT_LIST_HEAD(&con->writequeue);
 175                 spin_lock_init(&con->writequeue_lock);
 176                 INIT_WORK(&con->swork, process_send_sockets);
 177                 INIT_WORK(&con->rwork, process_recv_sockets);
 178
 179                 connections[nodeid] = con;
 180         }
 181
 182 finish:
 183         up(&connections_lock);
 184         return con;
 185 }
 186
 187 /* Data available on socket or listen socket received a connect */
 188 static void lowcomms_data_ready(struct sock *sk, int count_unused)
 189 {
 190         struct connection *con = sock2con(sk);
 191
 192         if (!test_and_set_bit(CF_READ_PENDING, &con->flags))
 193                 queue_work(recv_workqueue, &con->rwork);
 194 }
 195
 196 static void lowcomms_write_space(struct sock *sk)
 197 {
 198         struct connection *con = sock2con(sk);
 199
 200         if (!test_and_set_bit(CF_WRITE_PENDING, &con->flags))
 201                 queue_work(send_workqueue, &con->swork);
 202 }
 203
 204 static inline void lowcomms_connect_sock(struct connection *con)
 205 {
 206         if (!test_and_set_bit(CF_CONNECT_PENDING, &con->flags))
 207                 queue_work(send_workqueue, &con->swork);
 208 }
 209
 210 static void lowcomms_state_change(struct sock *sk)
 211 {
 212         if (sk->sk_state == TCP_ESTABLISHED)
 213                 lowcomms_write_space(sk);
 214 }
 215
 216 /* Make a socket active */
 217 static int add_sock(struct socket *sock, struct connection *con)
 218 {
 219         con->sock = sock;
 220
 221         /* Install a data_ready callback */
 222         con->sock->sk->sk_data_ready = lowcomms_data_ready;
 223         con->sock->sk->sk_write_space = lowcomms_write_space;
 224         con->sock->sk->sk_state_change = lowcomms_state_change;
 225
 226         return 0;
 227 }
 228
 229 /* Add the port number to an IP6 or 4 sockaddr and return the address
 230    length */
 231 static void make_sockaddr(struct sockaddr_storage *saddr, uint16_t port,
 232                           int *addr_len)
 233 {
 234         saddr->ss_family =  dlm_local_addr.ss_family;
 235         if (saddr->ss_family == AF_INET) {
 236                 struct sockaddr_in *in4_addr = (struct sockaddr_in *)saddr;
 237                 in4_addr->sin_port = cpu_to_be16(port);
 238                 *addr_len = sizeof(struct sockaddr_in);
 239         } else {
 240                 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)saddr;
 241                 in6_addr->sin6_port = cpu_to_be16(port);
 242                 *addr_len = sizeof(struct sockaddr_in6);
 243         }
 244 }
 245
 246 /* Close a remote connection and tidy up */
 247 static void close_connection(struct connection *con, bool and_other)
 248 {
 249         mutex_lock(&con->sock_mutex);
 250
 251         if (con->sock) {
 252                 sock_release(con->sock);
 253                 con->sock = NULL;
 254         }
 255         if (con->othercon && and_other) {
 256                 /* Will only re-enter once. */
 257                 close_connection(con->othercon, false);
 258         }
 259         if (con->rx_page) {
 260                 __free_page(con->rx_page);
 261                 con->rx_page = NULL;
 262         }
 263         con->retries = 0;
 264         mutex_unlock(&con->sock_mutex);
 265 }
 266
 267 /* Data received from remote end */
 268 static int receive_from_sock(struct connection *con)
 269 {
 270         int ret = 0;
 271         struct msghdr msg = {};
 272         struct kvec iov[2];
 273         unsigned len;
 274         int r;
 275         int call_again_soon = 0;
 276         int nvec;
 277
 278         mutex_lock(&con->sock_mutex);
 279
 280         if (con->sock == NULL) {
 281                 ret = -EAGAIN;
 282                 goto out_close;
 283         }
 284
 285         if (con->rx_page == NULL) {
 286                 /*
 287                  * This doesn't need to be atomic, but I think it should
 288                  * improve performance if it is.
 289                  */
 290                 con->rx_page = alloc_page(GFP_ATOMIC);
 291                 if (con->rx_page == NULL)
 292                         goto out_resched;
 293                 cbuf_init(&con->cb, PAGE_CACHE_SIZE);
 294         }
 295
 296         /*
 297          * iov[0] is the bit of the circular buffer between the current end
 298          * point (cb.base + cb.len) and the end of the buffer.
 299          */
 300         iov[0].iov_len = con->cb.base - cbuf_data(&con->cb);
 301         iov[0].iov_base = page_address(con->rx_page) + cbuf_data(&con->cb);
 302         nvec = 1;
 303
 304         /*
 305          * iov[1] is the bit of the circular buffer between the start of the
 306          * buffer and the start of the currently used section (cb.base)
 307          */
 308         if (cbuf_data(&con->cb) >= con->cb.base) {
 309                 iov[0].iov_len = PAGE_CACHE_SIZE - cbuf_data(&con->cb);
 310                 iov[1].iov_len = con->cb.base;
 311                 iov[1].iov_base = page_address(con->rx_page);
 312                 nvec = 2;
 313         }
 314         len = iov[0].iov_len + iov[1].iov_len;
 315
 316         r = ret = kernel_recvmsg(con->sock, &msg, iov, nvec, len,
 317                                MSG_DONTWAIT | MSG_NOSIGNAL);
 318
 319         if (ret <= 0)
 320                 goto out_close;
 321         if (ret == -EAGAIN)
 322                 goto out_resched;
 323
 324         if (ret == len)
 325                 call_again_soon = 1;
 326         cbuf_add(&con->cb, ret);
 327         ret = dlm_process_incoming_buffer(con->nodeid,
 328                                           page_address(con->rx_page),
 329                                           con->cb.base, con->cb.len,
 330                                           PAGE_CACHE_SIZE);
 331         if (ret == -EBADMSG) {
 332                 printk(KERN_INFO "dlm: lowcomms: addr=%p, base=%u, len=%u, "
 333                        "iov_len=%u, iov_base[0]=%p, read=%d\n",
 334                        page_address(con->rx_page), con->cb.base, con->cb.len,
 335                        len, iov[0].iov_base, r);
 336         }
 337         if (ret < 0)
 338                 goto out_close;
 339         cbuf_eat(&con->cb, ret);
 340
 341         if (cbuf_empty(&con->cb) && !call_again_soon) {
 342                 __free_page(con->rx_page);
 343                 con->rx_page = NULL;
 344         }
 345
 346         if (call_again_soon)
 347                 goto out_resched;
 348         mutex_unlock(&con->sock_mutex);
 349         return 0;
 350
 351 out_resched:
 352         if (!test_and_set_bit(CF_READ_PENDING, &con->flags))
 353                 queue_work(recv_workqueue, &con->rwork);
 354         mutex_unlock(&con->sock_mutex);
 355         return -EAGAIN;
 356
 357 out_close:
 358         mutex_unlock(&con->sock_mutex);
 359         if (ret != -EAGAIN && !test_bit(CF_IS_OTHERCON, &con->flags)) {
 360                 close_connection(con, false);
 361                 /* Reconnect when there is something to send */
 362         }
 363         /* Don't return success if we really got EOF */
 364         if (ret == 0)
 365                 ret = -EAGAIN;
 366
 367         return ret;
 368 }
 369
 370 /* Listening socket is busy, accept a connection */
 371 static int accept_from_sock(struct connection *con)
 372 {
 373         int result;
 374         struct sockaddr_storage peeraddr;
 375         struct socket *newsock;
 376         int len;
 377         int nodeid;
 378         struct connection *newcon;
 379         struct connection *addcon;
 380
 381         memset(&peeraddr, 0, sizeof(peeraddr));
 382         result = sock_create_kern(dlm_local_addr.ss_family, SOCK_STREAM,
 383                                   IPPROTO_TCP, &newsock);
 384         if (result < 0)
 385                 return -ENOMEM;
 386
 387         mutex_lock_nested(&con->sock_mutex, 0);
 388
 389         result = -ENOTCONN;
 390         if (con->sock == NULL)
 391                 goto accept_err;
 392
 393         newsock->type = con->sock->type;
 394         newsock->ops = con->sock->ops;
 395
 396         result = con->sock->ops->accept(con->sock, newsock, O_NONBLOCK);
 397         if (result < 0)
 398                 goto accept_err;
 399
 400         /* Get the connected socket's peer */
 401         memset(&peeraddr, 0, sizeof(peeraddr));
 402         if (newsock->ops->getname(newsock, (struct sockaddr *)&peeraddr,
 403                                   &len, 2)) {
 404                 result = -ECONNABORTED;
 405                 goto accept_err;
 406         }
 407
 408         /* Get the new node's NODEID */
 409         make_sockaddr(&peeraddr, 0, &len);
 410         if (dlm_addr_to_nodeid(&peeraddr, &nodeid)) {
 411                 printk("dlm: connect from non cluster node\n");
 412                 sock_release(newsock);
 413                 mutex_unlock(&con->sock_mutex);
 414                 return -1;
 415         }
 416
 417         log_print("got connection from %d", nodeid);
 418
 419         /*  Check to see if we already have a connection to this node. This
 420          *  could happen if the two nodes initiate a connection at roughly
 421          *  the same time and the connections cross on the wire.
 422          * TEMPORARY FIX:
 423          *  In this case we store the incoming one in "othercon"
 424          */
 425         newcon = nodeid2con(nodeid, GFP_KERNEL);
 426         if (!newcon) {
 427                 result = -ENOMEM;
 428                 goto accept_err;
 429         }
 430         mutex_lock_nested(&newcon->sock_mutex, 1);
 431         if (newcon->sock) {
 432                 struct connection *othercon = newcon->othercon;
 433
 434                 if (!othercon) {
 435                         othercon = kmem_cache_zalloc(con_cache, GFP_KERNEL);
 436                         if (!othercon) {
 437                                 printk("dlm: failed to allocate incoming socket\n");
 438                                 mutex_unlock(&newcon->sock_mutex);
 439                                 result = -ENOMEM;
 440                                 goto accept_err;
 441                         }
 442                         othercon->nodeid = nodeid;
 443                         othercon->rx_action = receive_from_sock;
 444                         mutex_init(&othercon->sock_mutex);
 445                         INIT_WORK(&othercon->swork, process_send_sockets);
 446                         INIT_WORK(&othercon->rwork, process_recv_sockets);
 447                         set_bit(CF_IS_OTHERCON, &othercon->flags);
 448                         newcon->othercon = othercon;
 449                 }
 450                 othercon->sock = newsock;
 451                 newsock->sk->sk_user_data = othercon;
 452                 add_sock(newsock, othercon);
 453                 addcon = othercon;
 454         }
 455         else {
 456                 newsock->sk->sk_user_data = newcon;
 457                 newcon->rx_action = receive_from_sock;
 458                 add_sock(newsock, newcon);
 459                 addcon = newcon;
 460         }
 461
 462         mutex_unlock(&newcon->sock_mutex);
 463
 464         /*
 465          * Add it to the active queue in case we got data
 466          * beween processing the accept adding the socket
 467          * to the read_sockets list
 468          */
 469         if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags))
 470                 queue_work(recv_workqueue, &addcon->rwork);
 471         mutex_unlock(&con->sock_mutex);
 472
 473         return 0;
 474
 475 accept_err:
 476         mutex_unlock(&con->sock_mutex);
 477         sock_release(newsock);
 478
 479         if (result != -EAGAIN)
 480                 printk("dlm: error accepting connection from node: %d\n", result);
 481         return result;
 482 }
 483
 484 /* Connect a new socket to its peer */
 485 static void connect_to_sock(struct connection *con)
 486 {
 487         int result = -EHOSTUNREACH;
 488         struct sockaddr_storage saddr;
 489         int addr_len;
 490         struct socket *sock;
 491
 492         if (con->nodeid == 0) {
 493                 log_print("attempt to connect sock 0 foiled");
 494                 return;
 495         }
 496
 497         mutex_lock(&con->sock_mutex);
 498         if (con->retries++ > MAX_CONNECT_RETRIES)
 499                 goto out;
 500
 501         /* Some odd races can cause double-connects, ignore them */
 502         if (con->sock) {
 503                 result = 0;
 504                 goto out;
 505         }
 506
 507         /* Create a socket to communicate with */
 508         result = sock_create_kern(dlm_local_addr.ss_family, SOCK_STREAM,
 509                                   IPPROTO_TCP, &sock);
 510         if (result < 0)
 511                 goto out_err;
 512
 513         memset(&saddr, 0, sizeof(saddr));
 514         if (dlm_nodeid_to_addr(con->nodeid, &saddr))
 515                 goto out_err;
 516
 517         sock->sk->sk_user_data = con;
 518         con->rx_action = receive_from_sock;
 519
 520         make_sockaddr(&saddr, dlm_config.ci_tcp_port, &addr_len);
 521
 522         add_sock(sock, con);
 523
 524         log_print("connecting to %d", con->nodeid);
 525         result =
 526                 sock->ops->connect(sock, (struct sockaddr *)&saddr, addr_len,
 527                                    O_NONBLOCK);
 528         if (result == -EINPROGRESS)
 529                 result = 0;
 530         if (result == 0)
 531                 goto out;
 532
 533 out_err:
 534         if (con->sock) {
 535                 sock_release(con->sock);
 536                 con->sock = NULL;
 537         }
 538         /*
 539          * Some errors are fatal and this list might need adjusting. For other
 540          * errors we try again until the max number of retries is reached.
 541          */
 542         if (result != -EHOSTUNREACH && result != -ENETUNREACH &&
 543             result != -ENETDOWN && result != EINVAL
 544             && result != -EPROTONOSUPPORT) {
 545                 lowcomms_connect_sock(con);
 546                 result = 0;
 547         }
 548 out:
 549         mutex_unlock(&con->sock_mutex);
 550         return;
 551 }
 552
 553 static struct socket *create_listen_sock(struct connection *con,
 554                                          struct sockaddr_storage *saddr)
 555 {
 556         struct socket *sock = NULL;
 557         mm_segment_t fs;
 558         int result = 0;
 559         int one = 1;
 560         int addr_len;
 561
 562         if (dlm_local_addr.ss_family == AF_INET)
 563                 addr_len = sizeof(struct sockaddr_in);
 564         else
 565                 addr_len = sizeof(struct sockaddr_in6);
 566
 567         /* Create a socket to communicate with */
 568         result = sock_create_kern(dlm_local_addr.ss_family, SOCK_STREAM, IPPROTO_TCP, &sock);
 569         if (result < 0) {
 570                 printk("dlm: Can't create listening comms socket\n");
 571                 goto create_out;
 572         }
 573
 574         fs = get_fs();
 575         set_fs(get_ds());
 576         result = sock_setsockopt(sock, SOL_SOCKET, SO_REUSEADDR,
 577                                  (char *)&one, sizeof(one));
 578         set_fs(fs);
 579         if (result < 0) {
 580                 printk("dlm: Failed to set SO_REUSEADDR on socket: result=%d\n",
 581                        result);
 582         }
 583         sock->sk->sk_user_data = con;
 584         con->rx_action = accept_from_sock;
 585         con->sock = sock;
 586
 587         /* Bind to our port */
 588         make_sockaddr(saddr, dlm_config.ci_tcp_port, &addr_len);
 589         result = sock->ops->bind(sock, (struct sockaddr *) saddr, addr_len);
 590         if (result < 0) {
 591                 printk("dlm: Can't bind to port %d\n", dlm_config.ci_tcp_port);
 592                 sock_release(sock);
 593                 sock = NULL;
 594                 con->sock = NULL;
 595                 goto create_out;
 596         }
 597
 598         fs = get_fs();
 599         set_fs(get_ds());
 600
 601         result = sock_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
 602                                  (char *)&one, sizeof(one));
 603         set_fs(fs);
 604         if (result < 0) {
 605                 printk("dlm: Set keepalive failed: %d\n", result);
 606         }
 607
 608         result = sock->ops->listen(sock, 5);
 609         if (result < 0) {
 610                 printk("dlm: Can't listen on port %d\n", dlm_config.ci_tcp_port);
 611                 sock_release(sock);
 612                 sock = NULL;
 613                 goto create_out;
 614         }
 615
 616 create_out:
 617         return sock;
 618 }
 619
 620
 621 /* Listen on all interfaces */
 622 static int listen_for_all(void)
 623 {
 624         struct socket *sock = NULL;
 625         struct connection *con = nodeid2con(0, GFP_KERNEL);
 626         int result = -EINVAL;
 627
 628         /* We don't support multi-homed hosts */
 629         set_bit(CF_IS_OTHERCON, &con->flags);
 630
 631         sock = create_listen_sock(con, &dlm_local_addr);
 632         if (sock) {
 633                 add_sock(sock, con);
 634                 result = 0;
 635         }
 636         else {
 637                 result = -EADDRINUSE;
 638         }
 639
 640         return result;
 641 }
 642
 643
 644
 645 static struct writequeue_entry *new_writequeue_entry(struct connection *con,
 646                                                      gfp_t allocation)
 647 {
 648         struct writequeue_entry *entry;
 649
 650         entry = kmalloc(sizeof(struct writequeue_entry), allocation);
 651         if (!entry)
 652                 return NULL;
 653
 654         entry->page = alloc_page(allocation);
 655         if (!entry->page) {
 656                 kfree(entry);
 657                 return NULL;
 658         }
 659
 660         entry->offset = 0;
 661         entry->len = 0;
 662         entry->end = 0;
 663         entry->users = 0;
 664         entry->con = con;
 665
 666         return entry;
 667 }
 668
 669 void *dlm_lowcomms_get_buffer(int nodeid, int len,
 670                               gfp_t allocation, char **ppc)
 671 {
 672         struct connection *con;
 673         struct writequeue_entry *e;
 674         int offset = 0;
 675         int users = 0;
 676
 677         con = nodeid2con(nodeid, allocation);
 678         if (!con)
 679                 return NULL;
 680
 681         spin_lock(&con->writequeue_lock);
 682         e = list_entry(con->writequeue.prev, struct writequeue_entry, list);
 683         if ((&e->list == &con->writequeue) ||
 684             (PAGE_CACHE_SIZE - e->end < len)) {
 685                 e = NULL;
 686         } else {
 687                 offset = e->end;
 688                 e->end += len;
 689                 users = e->users++;
 690         }
 691         spin_unlock(&con->writequeue_lock);
 692
 693         if (e) {
 694         got_one:
 695                 if (users == 0)
 696                         kmap(e->page);
 697                 *ppc = page_address(e->page) + offset;
 698                 return e;
 699         }
 700
 701         e = new_writequeue_entry(con, allocation);
 702         if (e) {
 703                 spin_lock(&con->writequeue_lock);
 704                 offset = e->end;
 705                 e->end += len;
 706                 users = e->users++;
 707                 list_add_tail(&e->list, &con->writequeue);
 708                 spin_unlock(&con->writequeue_lock);
 709                 goto got_one;
 710         }
 711         return NULL;
 712 }
 713
 714 void dlm_lowcomms_commit_buffer(void *mh)
 715 {
 716         struct writequeue_entry *e = (struct writequeue_entry *)mh;
 717         struct connection *con = e->con;
 718         int users;
 719
 720         spin_lock(&con->writequeue_lock);
 721         users = --e->users;
 722         if (users)
 723                 goto out;
 724         e->len = e->end - e->offset;
 725         kunmap(e->page);
 726         spin_unlock(&con->writequeue_lock);
 727
 728         if (!test_and_set_bit(CF_WRITE_PENDING, &con->flags)) {
 729                 queue_work(send_workqueue, &con->swork);
 730         }
 731         return;
 732
 733 out:
 734         spin_unlock(&con->writequeue_lock);
 735         return;
 736 }
 737
 738 static void free_entry(struct writequeue_entry *e)
 739 {
 740         __free_page(e->page);
 741         kfree(e);
 742 }
 743
 744 /* Send a message */
 745 static void send_to_sock(struct connection *con)
 746 {
 747         int ret = 0;
 748         ssize_t(*sendpage) (struct socket *, struct page *, int, size_t, int);
 749         const int msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL;
 750         struct writequeue_entry *e;
 751         int len, offset;
 752
 753         mutex_lock(&con->sock_mutex);
 754         if (con->sock == NULL)
 755                 goto out_connect;
 756
 757         sendpage = con->sock->ops->sendpage;
 758
 759         spin_lock(&con->writequeue_lock);
 760         for (;;) {
 761                 e = list_entry(con->writequeue.next, struct writequeue_entry,
 762                                list);
 763                 if ((struct list_head *) e == &con->writequeue)
 764                         break;
 765
 766                 len = e->len;
 767                 offset = e->offset;
 768                 BUG_ON(len == 0 && e->users == 0);
 769                 spin_unlock(&con->writequeue_lock);
 770                 kmap(e->page);
 771
 772                 ret = 0;
 773                 if (len) {
 774                         ret = sendpage(con->sock, e->page, offset, len,
 775                                        msg_flags);
 776                         if (ret == -EAGAIN || ret == 0)
 777                                 goto out;
 778                         if (ret <= 0)
 779                                 goto send_error;
 780                 }
 781                 else {
 782                         /* Don't starve people filling buffers */
 783                         cond_resched();
 784                 }
 785
 786                 spin_lock(&con->writequeue_lock);
 787                 e->offset += ret;
 788                 e->len -= ret;
 789
 790                 if (e->len == 0 && e->users == 0) {
 791                         list_del(&e->list);
 792                         kunmap(e->page);
 793                         free_entry(e);
 794                         continue;
 795                 }
 796         }
 797         spin_unlock(&con->writequeue_lock);
 798 out:
 799         mutex_unlock(&con->sock_mutex);
 800         return;
 801
 802 send_error:
 803         mutex_unlock(&con->sock_mutex);
 804         close_connection(con, false);
 805         lowcomms_connect_sock(con);
 806         return;
 807
 808 out_connect:
 809         mutex_unlock(&con->sock_mutex);
 810         connect_to_sock(con);
 811         return;
 812 }
 813
 814 static void clean_one_writequeue(struct connection *con)
 815 {
 816         struct list_head *list;
 817         struct list_head *temp;
 818
 819         spin_lock(&con->writequeue_lock);
 820         list_for_each_safe(list, temp, &con->writequeue) {
 821                 struct writequeue_entry *e =
 822                         list_entry(list, struct writequeue_entry, list);
 823                 list_del(&e->list);
 824                 free_entry(e);
 825         }
 826         spin_unlock(&con->writequeue_lock);
 827 }
 828
 829 /* Called from recovery when it knows that a node has
 830    left the cluster */
 831 int dlm_lowcomms_close(int nodeid)
 832 {
 833         struct connection *con;
 834
 835         if (!connections)
 836                 goto out;
 837
 838         log_print("closing connection to node %d", nodeid);
 839         con = nodeid2con(nodeid, 0);
 840         if (con) {
 841                 clean_one_writequeue(con);
 842                 close_connection(con, true);
 843         }
 844         return 0;
 845
 846 out:
 847         return -1;
 848 }
 849
 850 /* Look for activity on active sockets */
 851 static void process_recv_sockets(struct work_struct *work)
 852 {
 853         struct connection *con = container_of(work, struct connection, rwork);
 854         int err;
 855
 856         clear_bit(CF_READ_PENDING, &con->flags);
 857         do {
 858                 err = con->rx_action(con);
 859         } while (!err);
 860 }
 861
 862
 863 static void process_send_sockets(struct work_struct *work)
 864 {
 865         struct connection *con = container_of(work, struct connection, swork);
 866
 867         if (test_and_clear_bit(CF_CONNECT_PENDING, &con->flags)) {
 868                 connect_to_sock(con);
 869         }
 870
 871         clear_bit(CF_WRITE_PENDING, &con->flags);
 872         send_to_sock(con);
 873 }
 874
 875
 876 /* Discard all entries on the write queues */
 877 static void clean_writequeues(void)
 878 {
 879         int nodeid;
 880
 881         for (nodeid = 1; nodeid < conn_array_size; nodeid++) {
 882                 struct connection *con = nodeid2con(nodeid, 0);
 883
 884                 if (con)
 885                         clean_one_writequeue(con);
 886         }
 887 }
 888
 889 static void work_stop(void)
 890 {
 891         destroy_workqueue(recv_workqueue);
 892         destroy_workqueue(send_workqueue);
 893 }
 894
 895 static int work_start(void)
 896 {
 897         int error;
 898         recv_workqueue = create_workqueue("dlm_recv");
 899         error = IS_ERR(recv_workqueue);
 900         if (error) {
 901                 log_print("can't start dlm_recv %d", error);
 902                 return error;
 903         }
 904
 905         send_workqueue = create_singlethread_workqueue("dlm_send");
 906         error = IS_ERR(send_workqueue);
 907         if (error) {
 908                 log_print("can't start dlm_send %d", error);
 909                 destroy_workqueue(recv_workqueue);
 910                 return error;
 911         }
 912
 913         return 0;
 914 }
 915
 916 void dlm_lowcomms_stop(void)
 917 {
 918         int i;
 919
 920         /* Set all the flags to prevent any
 921            socket activity.
 922         */
 923         for (i = 0; i < conn_array_size; i++) {
 924                 if (connections[i])
 925                         connections[i]->flags |= 0xFF;
 926         }
 927
 928         work_stop();
 929         clean_writequeues();
 930
 931         for (i = 0; i < conn_array_size; i++) {
 932                 if (connections[i]) {
 933                         close_connection(connections[i], true);
 934                         if (connections[i]->othercon)
 935                                 kmem_cache_free(con_cache, connections[i]->othercon);
 936                         kmem_cache_free(con_cache, connections[i]);
 937                 }
 938         }
 939
 940         kfree(connections);
 941         connections = NULL;
 942
 943         kmem_cache_destroy(con_cache);
 944 }
 945
 946 /* This is quite likely to sleep... */
 947 int dlm_lowcomms_start(void)
 948 {
 949         int error = 0;
 950
 951         error = -ENOMEM;
 952         connections = kzalloc(sizeof(struct connection *) *
 953                               NODE_INCREMENT, GFP_KERNEL);
 954         if (!connections)
 955                 goto out;
 956
 957         conn_array_size = NODE_INCREMENT;
 958
 959         if (dlm_our_addr(&dlm_local_addr, 0)) {
 960                 log_print("no local IP address has been set");
 961                 goto fail_free_conn;
 962         }
 963         if (!dlm_our_addr(&dlm_local_addr, 1)) {
 964                 log_print("This dlm comms module does not support multi-homed clustering");
 965                 goto fail_free_conn;
 966         }
 967
 968         con_cache = kmem_cache_create("dlm_conn", sizeof(struct connection),
 969                                       __alignof__(struct connection), 0,
 970                                       NULL, NULL);
 971         if (!con_cache)
 972                 goto fail_free_conn;
 973
 974
 975         /* Start listening */
 976         error = listen_for_all();
 977         if (error)
 978                 goto fail_unlisten;
 979
 980         error = work_start();
 981         if (error)
 982                 goto fail_unlisten;
 983
 984         return 0;
 985
 986 fail_unlisten:
 987         close_connection(connections[0], false);
 988         kmem_cache_free(con_cache, connections[0]);
 989         kmem_cache_destroy(con_cache);
 990
 991 fail_free_conn:
 992         kfree(connections);
 993
 994 out:
 995         return error;
 996 }
 997
 998 /*
 999  * Overrides for Emacs so that we follow Linus's tabbing style.
1000  * Emacs will notice this stuff at the end of the file and automatically
1001  * adjust the settings for this buffer only.  This must remain at the end
1002  * of the file.
1003  * ---------------------------------------------------------------------------
1004  * Local variables:
1005  * c-file-style: "linux"
1006  * End:
1007  */