1/* -*- mode: c; c-basic-offset: 8; -*-
   2 *
   3 * vim: noexpandtab sw=8 ts=8 sts=0:
   4 *
   5 * Copyright (C) 2004 Oracle.  All rights reserved.
   6 *
   7 * This program is free software; you can redistribute it and/or
   8 * modify it under the terms of the GNU General Public
   9 * License as published by the Free Software Foundation; either
  10 * version 2 of the License, or (at your option) any later version.
  11 *
  12 * This program is distributed in the hope that it will be useful,
  13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15 * General Public License for more details.
  16 *
  17 * You should have received a copy of the GNU General Public
  18 * License along with this program; if not, write to the
  19 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  20 * Boston, MA 021110-1307, USA.
  21 *
  22 * ----
  23 *
  24 * Callers for this were originally written against a very simple synchronus
  25 * API.  This implementation reflects those simple callers.  Some day I'm sure
  26 * we'll need to move to a more robust posting/callback mechanism.
  27 *
  28 * Transmit calls pass in kernel virtual addresses and block copying this into
  29 * the socket's tx buffers via a usual blocking sendmsg.  They'll block waiting
  30 * for a failed socket to timeout.  TX callers can also pass in a poniter to an
  31 * 'int' which gets filled with an errno off the wire in response to the
  32 * message they send.
  33 *
  34 * Handlers for unsolicited messages are registered.  Each socket has a page
  35 * that incoming data is copied into.  First the header, then the data.
  36 * Handlers are called from only one thread with a reference to this per-socket
  37 * page.  This page is destroyed after the handler call, so it can't be
  38 * referenced beyond the call.  Handlers may block but are discouraged from
  39 * doing so.
  40 *
  41 * Any framing errors (bad magic, large payload lengths) close a connection.
  42 *
  43 * Our sock_container holds the state we associate with a socket.  It's current
  44 * framing state is held there as well as the refcounting we do around when it
  45 * is safe to tear down the socket.  The socket is only finally torn down from
  46 * the container when the container loses all of its references -- so as long
  47 * as you hold a ref on the container you can trust that the socket is valid
  48 * for use with kernel socket APIs.
  49 *
  50 * Connections are initiated between a pair of nodes when the node with the
  51 * higher node number gets a heartbeat callback which indicates that the lower
  52 * numbered node has started heartbeating.  The lower numbered node is passive
  53 * and only accepts the connection if the higher numbered node is heartbeating.
  54 */
  55
  56#include <linux/kernel.h>
  57#include <linux/jiffies.h>
  58#include <linux/slab.h>
  59#include <linux/idr.h>
  60#include <linux/kref.h>
  61#include <linux/net.h>
 
  62#include <net/tcp.h>
  63
  64#include <asm/uaccess.h>
  65
  66#include "heartbeat.h"
  67#include "tcp.h"
  68#include "nodemanager.h"
  69#define MLOG_MASK_PREFIX ML_TCP
  70#include "masklog.h"
  71#include "quorum.h"
  72
  73#include "tcp_internal.h"
  74
  75#define SC_NODEF_FMT "node %s (num %u) at %pI4:%u"
  76#define SC_NODEF_ARGS(sc) sc->sc_node->nd_name, sc->sc_node->nd_num,	\
  77			  &sc->sc_node->nd_ipv4_address,		\
  78			  ntohs(sc->sc_node->nd_ipv4_port)
  79
  80/*
  81 * In the following two log macros, the whitespace after the ',' just
  82 * before ##args is intentional. Otherwise, gcc 2.95 will eat the
  83 * previous token if args expands to nothing.
  84 */
  85#define msglog(hdr, fmt, args...) do {					\
  86	typeof(hdr) __hdr = (hdr);					\
  87	mlog(ML_MSG, "[mag %u len %u typ %u stat %d sys_stat %d "	\
  88	     "key %08x num %u] " fmt,					\
  89	     be16_to_cpu(__hdr->magic), be16_to_cpu(__hdr->data_len), 	\
  90	     be16_to_cpu(__hdr->msg_type), be32_to_cpu(__hdr->status),	\
  91	     be32_to_cpu(__hdr->sys_status), be32_to_cpu(__hdr->key),	\
  92	     be32_to_cpu(__hdr->msg_num) ,  ##args);			\
  93} while (0)
  94
  95#define sclog(sc, fmt, args...) do {					\
  96	typeof(sc) __sc = (sc);						\
  97	mlog(ML_SOCKET, "[sc %p refs %d sock %p node %u page %p "	\
  98	     "pg_off %zu] " fmt, __sc,					\
  99	     atomic_read(&__sc->sc_kref.refcount), __sc->sc_sock,	\
 100	    __sc->sc_node->nd_num, __sc->sc_page, __sc->sc_page_off ,	\
 101	    ##args);							\
 102} while (0)
 103
 104static DEFINE_RWLOCK(o2net_handler_lock);
 105static struct rb_root o2net_handler_tree = RB_ROOT;
 106
 107static struct o2net_node o2net_nodes[O2NM_MAX_NODES];
 108
 109/* XXX someday we'll need better accounting */
 110static struct socket *o2net_listen_sock = NULL;
 111
 112/*
 113 * listen work is only queued by the listening socket callbacks on the
 114 * o2net_wq.  teardown detaches the callbacks before destroying the workqueue.
 115 * quorum work is queued as sock containers are shutdown.. stop_listening
 116 * tears down all the node's sock containers, preventing future shutdowns
 117 * and queued quroum work, before canceling delayed quorum work and
 118 * destroying the work queue.
 119 */
 120static struct workqueue_struct *o2net_wq;
 121static struct work_struct o2net_listen_work;
 122
 123static struct o2hb_callback_func o2net_hb_up, o2net_hb_down;
 124#define O2NET_HB_PRI 0x1
 125
 126static struct o2net_handshake *o2net_hand;
 127static struct o2net_msg *o2net_keep_req, *o2net_keep_resp;
 128
 129static int o2net_sys_err_translations[O2NET_ERR_MAX] =
 130		{[O2NET_ERR_NONE]	= 0,
 131		 [O2NET_ERR_NO_HNDLR]	= -ENOPROTOOPT,
 132		 [O2NET_ERR_OVERFLOW]	= -EOVERFLOW,
 133		 [O2NET_ERR_DIED]	= -EHOSTDOWN,};
 134
 135/* can't quite avoid *all* internal declarations :/ */
 136static void o2net_sc_connect_completed(struct work_struct *work);
 137static void o2net_rx_until_empty(struct work_struct *work);
 138static void o2net_shutdown_sc(struct work_struct *work);
 139static void o2net_listen_data_ready(struct sock *sk, int bytes);
 140static void o2net_sc_send_keep_req(struct work_struct *work);
 141static void o2net_idle_timer(unsigned long data);
 142static void o2net_sc_postpone_idle(struct o2net_sock_container *sc);
 143static void o2net_sc_reset_idle_timer(struct o2net_sock_container *sc);
 144
 145#ifdef CONFIG_DEBUG_FS
 146static void o2net_init_nst(struct o2net_send_tracking *nst, u32 msgtype,
 147			   u32 msgkey, struct task_struct *task, u8 node)
 148{
 149	INIT_LIST_HEAD(&nst->st_net_debug_item);
 150	nst->st_task = task;
 151	nst->st_msg_type = msgtype;
 152	nst->st_msg_key = msgkey;
 153	nst->st_node = node;
 154}
 155
 156static inline void o2net_set_nst_sock_time(struct o2net_send_tracking *nst)
 157{
 158	nst->st_sock_time = ktime_get();
 159}
 160
 161static inline void o2net_set_nst_send_time(struct o2net_send_tracking *nst)
 162{
 163	nst->st_send_time = ktime_get();
 164}
 165
 166static inline void o2net_set_nst_status_time(struct o2net_send_tracking *nst)
 167{
 168	nst->st_status_time = ktime_get();
 169}
 170
 171static inline void o2net_set_nst_sock_container(struct o2net_send_tracking *nst,
 172						struct o2net_sock_container *sc)
 173{
 174	nst->st_sc = sc;
 175}
 176
 177static inline void o2net_set_nst_msg_id(struct o2net_send_tracking *nst,
 178					u32 msg_id)
 179{
 180	nst->st_id = msg_id;
 181}
 182
 183static inline void o2net_set_sock_timer(struct o2net_sock_container *sc)
 184{
 185	sc->sc_tv_timer = ktime_get();
 186}
 187
 188static inline void o2net_set_data_ready_time(struct o2net_sock_container *sc)
 189{
 190	sc->sc_tv_data_ready = ktime_get();
 191}
 192
 193static inline void o2net_set_advance_start_time(struct o2net_sock_container *sc)
 194{
 195	sc->sc_tv_advance_start = ktime_get();
 196}
 197
 198static inline void o2net_set_advance_stop_time(struct o2net_sock_container *sc)
 199{
 200	sc->sc_tv_advance_stop = ktime_get();
 201}
 202
 203static inline void o2net_set_func_start_time(struct o2net_sock_container *sc)
 204{
 205	sc->sc_tv_func_start = ktime_get();
 206}
 207
 208static inline void o2net_set_func_stop_time(struct o2net_sock_container *sc)
 209{
 210	sc->sc_tv_func_stop = ktime_get();
 211}
 212
 213#else  /* CONFIG_DEBUG_FS */
 214# define o2net_init_nst(a, b, c, d, e)
 215# define o2net_set_nst_sock_time(a)
 216# define o2net_set_nst_send_time(a)
 217# define o2net_set_nst_status_time(a)
 218# define o2net_set_nst_sock_container(a, b)
 219# define o2net_set_nst_msg_id(a, b)
 220# define o2net_set_sock_timer(a)
 221# define o2net_set_data_ready_time(a)
 222# define o2net_set_advance_start_time(a)
 223# define o2net_set_advance_stop_time(a)
 224# define o2net_set_func_start_time(a)
 225# define o2net_set_func_stop_time(a)
 226#endif /* CONFIG_DEBUG_FS */
 227
 228#ifdef CONFIG_OCFS2_FS_STATS
 229static ktime_t o2net_get_func_run_time(struct o2net_sock_container *sc)
 230{
 231	return ktime_sub(sc->sc_tv_func_stop, sc->sc_tv_func_start);
 232}
 233
 234static void o2net_update_send_stats(struct o2net_send_tracking *nst,
 235				    struct o2net_sock_container *sc)
 236{
 237	sc->sc_tv_status_total = ktime_add(sc->sc_tv_status_total,
 238					   ktime_sub(ktime_get(),
 239						     nst->st_status_time));
 240	sc->sc_tv_send_total = ktime_add(sc->sc_tv_send_total,
 241					 ktime_sub(nst->st_status_time,
 242						   nst->st_send_time));
 243	sc->sc_tv_acquiry_total = ktime_add(sc->sc_tv_acquiry_total,
 244					    ktime_sub(nst->st_send_time,
 245						      nst->st_sock_time));
 246	sc->sc_send_count++;
 247}
 248
 249static void o2net_update_recv_stats(struct o2net_sock_container *sc)
 250{
 251	sc->sc_tv_process_total = ktime_add(sc->sc_tv_process_total,
 252					    o2net_get_func_run_time(sc));
 253	sc->sc_recv_count++;
 254}
 255
 256#else
 257
 258# define o2net_update_send_stats(a, b)
 259
 260# define o2net_update_recv_stats(sc)
 261
 262#endif /* CONFIG_OCFS2_FS_STATS */
 263
 264static inline int o2net_reconnect_delay(void)
 265{
 266	return o2nm_single_cluster->cl_reconnect_delay_ms;
 267}
 268
 269static inline int o2net_keepalive_delay(void)
 270{
 271	return o2nm_single_cluster->cl_keepalive_delay_ms;
 272}
 273
 274static inline int o2net_idle_timeout(void)
 275{
 276	return o2nm_single_cluster->cl_idle_timeout_ms;
 277}
 278
 279static inline int o2net_sys_err_to_errno(enum o2net_system_error err)
 280{
 281	int trans;
 282	BUG_ON(err >= O2NET_ERR_MAX);
 283	trans = o2net_sys_err_translations[err];
 284
 285	/* Just in case we mess up the translation table above */
 286	BUG_ON(err != O2NET_ERR_NONE && trans == 0);
 287	return trans;
 288}
 289
 290static struct o2net_node * o2net_nn_from_num(u8 node_num)
 291{
 292	BUG_ON(node_num >= ARRAY_SIZE(o2net_nodes));
 293	return &o2net_nodes[node_num];
 294}
 295
 296static u8 o2net_num_from_nn(struct o2net_node *nn)
 297{
 298	BUG_ON(nn == NULL);
 299	return nn - o2net_nodes;
 300}
 301
 302/* ------------------------------------------------------------ */
 303
 304static int o2net_prep_nsw(struct o2net_node *nn, struct o2net_status_wait *nsw)
 305{
 306	int ret = 0;
 307
 308	do {
 309		if (!idr_pre_get(&nn->nn_status_idr, GFP_ATOMIC)) {
 310			ret = -EAGAIN;
 311			break;
 312		}
 313		spin_lock(&nn->nn_lock);
 314		ret = idr_get_new(&nn->nn_status_idr, nsw, &nsw->ns_id);
 315		if (ret == 0)
 316			list_add_tail(&nsw->ns_node_item,
 317				      &nn->nn_status_list);
 318		spin_unlock(&nn->nn_lock);
 319	} while (ret == -EAGAIN);
 320
 321	if (ret == 0)  {
 322		init_waitqueue_head(&nsw->ns_wq);
 323		nsw->ns_sys_status = O2NET_ERR_NONE;
 324		nsw->ns_status = 0;
 
 325	}
 
 
 
 326
 327	return ret;
 
 
 
 328}
 329
 330static void o2net_complete_nsw_locked(struct o2net_node *nn,
 331				      struct o2net_status_wait *nsw,
 332				      enum o2net_system_error sys_status,
 333				      s32 status)
 334{
 335	assert_spin_locked(&nn->nn_lock);
 336
 337	if (!list_empty(&nsw->ns_node_item)) {
 338		list_del_init(&nsw->ns_node_item);
 339		nsw->ns_sys_status = sys_status;
 340		nsw->ns_status = status;
 341		idr_remove(&nn->nn_status_idr, nsw->ns_id);
 342		wake_up(&nsw->ns_wq);
 343	}
 344}
 345
 346static void o2net_complete_nsw(struct o2net_node *nn,
 347			       struct o2net_status_wait *nsw,
 348			       u64 id, enum o2net_system_error sys_status,
 349			       s32 status)
 350{
 351	spin_lock(&nn->nn_lock);
 352	if (nsw == NULL) {
 353		if (id > INT_MAX)
 354			goto out;
 355
 356		nsw = idr_find(&nn->nn_status_idr, id);
 357		if (nsw == NULL)
 358			goto out;
 359	}
 360
 361	o2net_complete_nsw_locked(nn, nsw, sys_status, status);
 362
 363out:
 364	spin_unlock(&nn->nn_lock);
 365	return;
 366}
 367
 368static void o2net_complete_nodes_nsw(struct o2net_node *nn)
 369{
 370	struct o2net_status_wait *nsw, *tmp;
 371	unsigned int num_kills = 0;
 372
 373	assert_spin_locked(&nn->nn_lock);
 374
 375	list_for_each_entry_safe(nsw, tmp, &nn->nn_status_list, ns_node_item) {
 376		o2net_complete_nsw_locked(nn, nsw, O2NET_ERR_DIED, 0);
 377		num_kills++;
 378	}
 379
 380	mlog(0, "completed %d messages for node %u\n", num_kills,
 381	     o2net_num_from_nn(nn));
 382}
 383
 384static int o2net_nsw_completed(struct o2net_node *nn,
 385			       struct o2net_status_wait *nsw)
 386{
 387	int completed;
 388	spin_lock(&nn->nn_lock);
 389	completed = list_empty(&nsw->ns_node_item);
 390	spin_unlock(&nn->nn_lock);
 391	return completed;
 392}
 393
 394/* ------------------------------------------------------------ */
 395
 396static void sc_kref_release(struct kref *kref)
 397{
 398	struct o2net_sock_container *sc = container_of(kref,
 399					struct o2net_sock_container, sc_kref);
 400	BUG_ON(timer_pending(&sc->sc_idle_timeout));
 401
 402	sclog(sc, "releasing\n");
 403
 404	if (sc->sc_sock) {
 405		sock_release(sc->sc_sock);
 406		sc->sc_sock = NULL;
 407	}
 408
 409	o2nm_undepend_item(&sc->sc_node->nd_item);
 410	o2nm_node_put(sc->sc_node);
 411	sc->sc_node = NULL;
 412
 413	o2net_debug_del_sc(sc);
 
 
 
 414	kfree(sc);
 415}
 416
 417static void sc_put(struct o2net_sock_container *sc)
 418{
 419	sclog(sc, "put\n");
 420	kref_put(&sc->sc_kref, sc_kref_release);
 421}
 422static void sc_get(struct o2net_sock_container *sc)
 423{
 424	sclog(sc, "get\n");
 425	kref_get(&sc->sc_kref);
 426}
 427static struct o2net_sock_container *sc_alloc(struct o2nm_node *node)
 428{
 429	struct o2net_sock_container *sc, *ret = NULL;
 430	struct page *page = NULL;
 431	int status = 0;
 432
 433	page = alloc_page(GFP_NOFS);
 434	sc = kzalloc(sizeof(*sc), GFP_NOFS);
 435	if (sc == NULL || page == NULL)
 436		goto out;
 437
 438	kref_init(&sc->sc_kref);
 439	o2nm_node_get(node);
 440	sc->sc_node = node;
 441
 442	/* pin the node item of the remote node */
 443	status = o2nm_depend_item(&node->nd_item);
 444	if (status) {
 445		mlog_errno(status);
 446		o2nm_node_put(node);
 447		goto out;
 448	}
 449	INIT_WORK(&sc->sc_connect_work, o2net_sc_connect_completed);
 450	INIT_WORK(&sc->sc_rx_work, o2net_rx_until_empty);
 451	INIT_WORK(&sc->sc_shutdown_work, o2net_shutdown_sc);
 452	INIT_DELAYED_WORK(&sc->sc_keepalive_work, o2net_sc_send_keep_req);
 453
 454	init_timer(&sc->sc_idle_timeout);
 455	sc->sc_idle_timeout.function = o2net_idle_timer;
 456	sc->sc_idle_timeout.data = (unsigned long)sc;
 457
 458	sclog(sc, "alloced\n");
 459
 460	ret = sc;
 461	sc->sc_page = page;
 462	o2net_debug_add_sc(sc);
 463	sc = NULL;
 464	page = NULL;
 465
 466out:
 467	if (page)
 468		__free_page(page);
 469	kfree(sc);
 470
 471	return ret;
 472}
 473
 474/* ------------------------------------------------------------ */
 475
 476static void o2net_sc_queue_work(struct o2net_sock_container *sc,
 477				struct work_struct *work)
 478{
 479	sc_get(sc);
 480	if (!queue_work(o2net_wq, work))
 481		sc_put(sc);
 482}
 483static void o2net_sc_queue_delayed_work(struct o2net_sock_container *sc,
 484					struct delayed_work *work,
 485					int delay)
 486{
 487	sc_get(sc);
 488	if (!queue_delayed_work(o2net_wq, work, delay))
 489		sc_put(sc);
 490}
 491static void o2net_sc_cancel_delayed_work(struct o2net_sock_container *sc,
 492					 struct delayed_work *work)
 493{
 494	if (cancel_delayed_work(work))
 495		sc_put(sc);
 496}
 497
 498static atomic_t o2net_connected_peers = ATOMIC_INIT(0);
 499
 500int o2net_num_connected_peers(void)
 501{
 502	return atomic_read(&o2net_connected_peers);
 503}
 504
 505static void o2net_set_nn_state(struct o2net_node *nn,
 506			       struct o2net_sock_container *sc,
 507			       unsigned valid, int err)
 508{
 509	int was_valid = nn->nn_sc_valid;
 510	int was_err = nn->nn_persistent_error;
 511	struct o2net_sock_container *old_sc = nn->nn_sc;
 512
 513	assert_spin_locked(&nn->nn_lock);
 514
 515	if (old_sc && !sc)
 516		atomic_dec(&o2net_connected_peers);
 517	else if (!old_sc && sc)
 518		atomic_inc(&o2net_connected_peers);
 519
 520	/* the node num comparison and single connect/accept path should stop
 521	 * an non-null sc from being overwritten with another */
 522	BUG_ON(sc && nn->nn_sc && nn->nn_sc != sc);
 523	mlog_bug_on_msg(err && valid, "err %d valid %u\n", err, valid);
 524	mlog_bug_on_msg(valid && !sc, "valid %u sc %p\n", valid, sc);
 525
 526	if (was_valid && !valid && err == 0)
 527		err = -ENOTCONN;
 528
 529	mlog(ML_CONN, "node %u sc: %p -> %p, valid %u -> %u, err %d -> %d\n",
 530	     o2net_num_from_nn(nn), nn->nn_sc, sc, nn->nn_sc_valid, valid,
 531	     nn->nn_persistent_error, err);
 532
 533	nn->nn_sc = sc;
 534	nn->nn_sc_valid = valid ? 1 : 0;
 535	nn->nn_persistent_error = err;
 536
 537	/* mirrors o2net_tx_can_proceed() */
 538	if (nn->nn_persistent_error || nn->nn_sc_valid)
 539		wake_up(&nn->nn_sc_wq);
 540
 541	if (!was_err && nn->nn_persistent_error) {
 542		o2quo_conn_err(o2net_num_from_nn(nn));
 543		queue_delayed_work(o2net_wq, &nn->nn_still_up,
 544				   msecs_to_jiffies(O2NET_QUORUM_DELAY_MS));
 545	}
 546
 547	if (was_valid && !valid) {
 548		printk(KERN_NOTICE "o2net: no longer connected to "
 549		       SC_NODEF_FMT "\n", SC_NODEF_ARGS(old_sc));
 
 550		o2net_complete_nodes_nsw(nn);
 551	}
 552
 553	if (!was_valid && valid) {
 554		o2quo_conn_up(o2net_num_from_nn(nn));
 555		cancel_delayed_work(&nn->nn_connect_expired);
 556		printk(KERN_NOTICE "o2net: %s " SC_NODEF_FMT "\n",
 557		       o2nm_this_node() > sc->sc_node->nd_num ?
 558		       		"connected to" : "accepted connection from",
 559		       SC_NODEF_ARGS(sc));
 560	}
 561
 562	/* trigger the connecting worker func as long as we're not valid,
 563	 * it will back off if it shouldn't connect.  This can be called
 564	 * from node config teardown and so needs to be careful about
 565	 * the work queue actually being up. */
 566	if (!valid && o2net_wq) {
 567		unsigned long delay;
 568		/* delay if we're within a RECONNECT_DELAY of the
 569		 * last attempt */
 570		delay = (nn->nn_last_connect_attempt +
 571			 msecs_to_jiffies(o2net_reconnect_delay()))
 572			- jiffies;
 573		if (delay > msecs_to_jiffies(o2net_reconnect_delay()))
 574			delay = 0;
 575		mlog(ML_CONN, "queueing conn attempt in %lu jiffies\n", delay);
 576		queue_delayed_work(o2net_wq, &nn->nn_connect_work, delay);
 577
 578		/*
 579		 * Delay the expired work after idle timeout.
 580		 *
 581		 * We might have lots of failed connection attempts that run
 582		 * through here but we only cancel the connect_expired work when
 583		 * a connection attempt succeeds.  So only the first enqueue of
 584		 * the connect_expired work will do anything.  The rest will see
 585		 * that it's already queued and do nothing.
 586		 */
 587		delay += msecs_to_jiffies(o2net_idle_timeout());
 588		queue_delayed_work(o2net_wq, &nn->nn_connect_expired, delay);
 589	}
 590
 591	/* keep track of the nn's sc ref for the caller */
 592	if ((old_sc == NULL) && sc)
 593		sc_get(sc);
 594	if (old_sc && (old_sc != sc)) {
 595		o2net_sc_queue_work(old_sc, &old_sc->sc_shutdown_work);
 596		sc_put(old_sc);
 597	}
 598}
 599
 600/* see o2net_register_callbacks() */
 601static void o2net_data_ready(struct sock *sk, int bytes)
 602{
 603	void (*ready)(struct sock *sk, int bytes);
 604
 605	read_lock(&sk->sk_callback_lock);
 606	if (sk->sk_user_data) {
 607		struct o2net_sock_container *sc = sk->sk_user_data;
 608		sclog(sc, "data_ready hit\n");
 609		o2net_set_data_ready_time(sc);
 610		o2net_sc_queue_work(sc, &sc->sc_rx_work);
 611		ready = sc->sc_data_ready;
 612	} else {
 613		ready = sk->sk_data_ready;
 614	}
 615	read_unlock(&sk->sk_callback_lock);
 616
 617	ready(sk, bytes);
 618}
 619
 620/* see o2net_register_callbacks() */
 621static void o2net_state_change(struct sock *sk)
 622{
 623	void (*state_change)(struct sock *sk);
 624	struct o2net_sock_container *sc;
 625
 626	read_lock(&sk->sk_callback_lock);
 627	sc = sk->sk_user_data;
 628	if (sc == NULL) {
 629		state_change = sk->sk_state_change;
 630		goto out;
 631	}
 632
 633	sclog(sc, "state_change to %d\n", sk->sk_state);
 634
 635	state_change = sc->sc_state_change;
 636
 637	switch(sk->sk_state) {
 638		/* ignore connecting sockets as they make progress */
 639		case TCP_SYN_SENT:
 640		case TCP_SYN_RECV:
 641			break;
 642		case TCP_ESTABLISHED:
 643			o2net_sc_queue_work(sc, &sc->sc_connect_work);
 644			break;
 645		default:
 646			printk(KERN_INFO "o2net: connection to " SC_NODEF_FMT
 647			      " shutdown, state %d\n",
 648			      SC_NODEF_ARGS(sc), sk->sk_state);
 649			o2net_sc_queue_work(sc, &sc->sc_shutdown_work);
 650			break;
 651	}
 652out:
 653	read_unlock(&sk->sk_callback_lock);
 654	state_change(sk);
 655}
 656
 657/*
 658 * we register callbacks so we can queue work on events before calling
 659 * the original callbacks.  our callbacks our careful to test user_data
 660 * to discover when they've reaced with o2net_unregister_callbacks().
 661 */
 662static void o2net_register_callbacks(struct sock *sk,
 663				     struct o2net_sock_container *sc)
 664{
 665	write_lock_bh(&sk->sk_callback_lock);
 666
 667	/* accepted sockets inherit the old listen socket data ready */
 668	if (sk->sk_data_ready == o2net_listen_data_ready) {
 669		sk->sk_data_ready = sk->sk_user_data;
 670		sk->sk_user_data = NULL;
 671	}
 672
 673	BUG_ON(sk->sk_user_data != NULL);
 674	sk->sk_user_data = sc;
 675	sc_get(sc);
 676
 677	sc->sc_data_ready = sk->sk_data_ready;
 678	sc->sc_state_change = sk->sk_state_change;
 679	sk->sk_data_ready = o2net_data_ready;
 680	sk->sk_state_change = o2net_state_change;
 681
 682	mutex_init(&sc->sc_send_lock);
 683
 684	write_unlock_bh(&sk->sk_callback_lock);
 685}
 686
 687static int o2net_unregister_callbacks(struct sock *sk,
 688			           struct o2net_sock_container *sc)
 689{
 690	int ret = 0;
 691
 692	write_lock_bh(&sk->sk_callback_lock);
 693	if (sk->sk_user_data == sc) {
 694		ret = 1;
 695		sk->sk_user_data = NULL;
 696		sk->sk_data_ready = sc->sc_data_ready;
 697		sk->sk_state_change = sc->sc_state_change;
 698	}
 699	write_unlock_bh(&sk->sk_callback_lock);
 700
 701	return ret;
 702}
 703
 704/*
 705 * this is a little helper that is called by callers who have seen a problem
 706 * with an sc and want to detach it from the nn if someone already hasn't beat
 707 * them to it.  if an error is given then the shutdown will be persistent
 708 * and pending transmits will be canceled.
 709 */
 710static void o2net_ensure_shutdown(struct o2net_node *nn,
 711			           struct o2net_sock_container *sc,
 712				   int err)
 713{
 714	spin_lock(&nn->nn_lock);
 715	if (nn->nn_sc == sc)
 716		o2net_set_nn_state(nn, NULL, 0, err);
 717	spin_unlock(&nn->nn_lock);
 718}
 719
 720/*
 721 * This work queue function performs the blocking parts of socket shutdown.  A
 722 * few paths lead here.  set_nn_state will trigger this callback if it sees an
 723 * sc detached from the nn.  state_change will also trigger this callback
 724 * directly when it sees errors.  In that case we need to call set_nn_state
 725 * ourselves as state_change couldn't get the nn_lock and call set_nn_state
 726 * itself.
 727 */
 728static void o2net_shutdown_sc(struct work_struct *work)
 729{
 730	struct o2net_sock_container *sc =
 731		container_of(work, struct o2net_sock_container,
 732			     sc_shutdown_work);
 733	struct o2net_node *nn = o2net_nn_from_num(sc->sc_node->nd_num);
 734
 735	sclog(sc, "shutting down\n");
 736
 737	/* drop the callbacks ref and call shutdown only once */
 738	if (o2net_unregister_callbacks(sc->sc_sock->sk, sc)) {
 739		/* we shouldn't flush as we're in the thread, the
 740		 * races with pending sc work structs are harmless */
 741		del_timer_sync(&sc->sc_idle_timeout);
 742		o2net_sc_cancel_delayed_work(sc, &sc->sc_keepalive_work);
 743		sc_put(sc);
 744		kernel_sock_shutdown(sc->sc_sock, SHUT_RDWR);
 745	}
 746
 747	/* not fatal so failed connects before the other guy has our
 748	 * heartbeat can be retried */
 749	o2net_ensure_shutdown(nn, sc, 0);
 750	sc_put(sc);
 751}
 752
 753/* ------------------------------------------------------------ */
 754
 755static int o2net_handler_cmp(struct o2net_msg_handler *nmh, u32 msg_type,
 756			     u32 key)
 757{
 758	int ret = memcmp(&nmh->nh_key, &key, sizeof(key));
 759
 760	if (ret == 0)
 761		ret = memcmp(&nmh->nh_msg_type, &msg_type, sizeof(msg_type));
 762
 763	return ret;
 764}
 765
 766static struct o2net_msg_handler *
 767o2net_handler_tree_lookup(u32 msg_type, u32 key, struct rb_node ***ret_p,
 768			  struct rb_node **ret_parent)
 769{
 770        struct rb_node **p = &o2net_handler_tree.rb_node;
 771        struct rb_node *parent = NULL;
 772	struct o2net_msg_handler *nmh, *ret = NULL;
 773	int cmp;
 774
 775        while (*p) {
 776                parent = *p;
 777                nmh = rb_entry(parent, struct o2net_msg_handler, nh_node);
 778		cmp = o2net_handler_cmp(nmh, msg_type, key);
 779
 780                if (cmp < 0)
 781                        p = &(*p)->rb_left;
 782                else if (cmp > 0)
 783                        p = &(*p)->rb_right;
 784                else {
 785			ret = nmh;
 786                        break;
 787		}
 788        }
 789
 790        if (ret_p != NULL)
 791                *ret_p = p;
 792        if (ret_parent != NULL)
 793                *ret_parent = parent;
 794
 795        return ret;
 796}
 797
 798static void o2net_handler_kref_release(struct kref *kref)
 799{
 800	struct o2net_msg_handler *nmh;
 801	nmh = container_of(kref, struct o2net_msg_handler, nh_kref);
 802
 803	kfree(nmh);
 804}
 805
 806static void o2net_handler_put(struct o2net_msg_handler *nmh)
 807{
 808	kref_put(&nmh->nh_kref, o2net_handler_kref_release);
 809}
 810
 811/* max_len is protection for the handler func.  incoming messages won't
 812 * be given to the handler if their payload is longer than the max. */
 813int o2net_register_handler(u32 msg_type, u32 key, u32 max_len,
 814			   o2net_msg_handler_func *func, void *data,
 815			   o2net_post_msg_handler_func *post_func,
 816			   struct list_head *unreg_list)
 817{
 818	struct o2net_msg_handler *nmh = NULL;
 819	struct rb_node **p, *parent;
 820	int ret = 0;
 821
 822	if (max_len > O2NET_MAX_PAYLOAD_BYTES) {
 823		mlog(0, "max_len for message handler out of range: %u\n",
 824			max_len);
 825		ret = -EINVAL;
 826		goto out;
 827	}
 828
 829	if (!msg_type) {
 830		mlog(0, "no message type provided: %u, %p\n", msg_type, func);
 831		ret = -EINVAL;
 832		goto out;
 833
 834	}
 835	if (!func) {
 836		mlog(0, "no message handler provided: %u, %p\n",
 837		       msg_type, func);
 838		ret = -EINVAL;
 839		goto out;
 840	}
 841
 842       	nmh = kzalloc(sizeof(struct o2net_msg_handler), GFP_NOFS);
 843	if (nmh == NULL) {
 844		ret = -ENOMEM;
 845		goto out;
 846	}
 847
 848	nmh->nh_func = func;
 849	nmh->nh_func_data = data;
 850	nmh->nh_post_func = post_func;
 851	nmh->nh_msg_type = msg_type;
 852	nmh->nh_max_len = max_len;
 853	nmh->nh_key = key;
 854	/* the tree and list get this ref.. they're both removed in
 855	 * unregister when this ref is dropped */
 856	kref_init(&nmh->nh_kref);
 857	INIT_LIST_HEAD(&nmh->nh_unregister_item);
 858
 859	write_lock(&o2net_handler_lock);
 860	if (o2net_handler_tree_lookup(msg_type, key, &p, &parent))
 861		ret = -EEXIST;
 862	else {
 863	        rb_link_node(&nmh->nh_node, parent, p);
 864		rb_insert_color(&nmh->nh_node, &o2net_handler_tree);
 865		list_add_tail(&nmh->nh_unregister_item, unreg_list);
 866
 867		mlog(ML_TCP, "registered handler func %p type %u key %08x\n",
 868		     func, msg_type, key);
 869		/* we've had some trouble with handlers seemingly vanishing. */
 870		mlog_bug_on_msg(o2net_handler_tree_lookup(msg_type, key, &p,
 871							  &parent) == NULL,
 872			        "couldn't find handler we *just* registerd "
 873				"for type %u key %08x\n", msg_type, key);
 874	}
 875	write_unlock(&o2net_handler_lock);
 876	if (ret)
 877		goto out;
 878
 879out:
 880	if (ret)
 881		kfree(nmh);
 882
 883	return ret;
 884}
 885EXPORT_SYMBOL_GPL(o2net_register_handler);
 886
 887void o2net_unregister_handler_list(struct list_head *list)
 888{
 889	struct o2net_msg_handler *nmh, *n;
 890
 891	write_lock(&o2net_handler_lock);
 892	list_for_each_entry_safe(nmh, n, list, nh_unregister_item) {
 893		mlog(ML_TCP, "unregistering handler func %p type %u key %08x\n",
 894		     nmh->nh_func, nmh->nh_msg_type, nmh->nh_key);
 895		rb_erase(&nmh->nh_node, &o2net_handler_tree);
 896		list_del_init(&nmh->nh_unregister_item);
 897		kref_put(&nmh->nh_kref, o2net_handler_kref_release);
 898	}
 899	write_unlock(&o2net_handler_lock);
 900}
 901EXPORT_SYMBOL_GPL(o2net_unregister_handler_list);
 902
 903static struct o2net_msg_handler *o2net_handler_get(u32 msg_type, u32 key)
 904{
 905	struct o2net_msg_handler *nmh;
 906
 907	read_lock(&o2net_handler_lock);
 908	nmh = o2net_handler_tree_lookup(msg_type, key, NULL, NULL);
 909	if (nmh)
 910		kref_get(&nmh->nh_kref);
 911	read_unlock(&o2net_handler_lock);
 912
 913	return nmh;
 914}
 915
 916/* ------------------------------------------------------------ */
 917
 918static int o2net_recv_tcp_msg(struct socket *sock, void *data, size_t len)
 919{
 920	int ret;
 921	mm_segment_t oldfs;
 922	struct kvec vec = {
 923		.iov_len = len,
 924		.iov_base = data,
 925	};
 926	struct msghdr msg = {
 927		.msg_iovlen = 1,
 928		.msg_iov = (struct iovec *)&vec,
 929       		.msg_flags = MSG_DONTWAIT,
 930	};
 931
 932	oldfs = get_fs();
 933	set_fs(get_ds());
 934	ret = sock_recvmsg(sock, &msg, len, msg.msg_flags);
 935	set_fs(oldfs);
 936
 937	return ret;
 938}
 939
 940static int o2net_send_tcp_msg(struct socket *sock, struct kvec *vec,
 941			      size_t veclen, size_t total)
 942{
 943	int ret;
 944	mm_segment_t oldfs;
 945	struct msghdr msg = {
 946		.msg_iov = (struct iovec *)vec,
 947		.msg_iovlen = veclen,
 948	};
 949
 950	if (sock == NULL) {
 951		ret = -EINVAL;
 952		goto out;
 953	}
 954
 955	oldfs = get_fs();
 956	set_fs(get_ds());
 957	ret = sock_sendmsg(sock, &msg, total);
 958	set_fs(oldfs);
 959	if (ret != total) {
 960		mlog(ML_ERROR, "sendmsg returned %d instead of %zu\n", ret,
 961		     total);
 962		if (ret >= 0)
 963			ret = -EPIPE; /* should be smarter, I bet */
 964		goto out;
 965	}
 966
 967	ret = 0;
 968out:
 969	if (ret < 0)
 970		mlog(0, "returning error: %d\n", ret);
 971	return ret;
 972}
 973
 974static void o2net_sendpage(struct o2net_sock_container *sc,
 975			   void *kmalloced_virt,
 976			   size_t size)
 977{
 978	struct o2net_node *nn = o2net_nn_from_num(sc->sc_node->nd_num);
 979	ssize_t ret;
 980
 981	while (1) {
 982		mutex_lock(&sc->sc_send_lock);
 983		ret = sc->sc_sock->ops->sendpage(sc->sc_sock,
 984						 virt_to_page(kmalloced_virt),
 985						 (long)kmalloced_virt & ~PAGE_MASK,
 986						 size, MSG_DONTWAIT);
 987		mutex_unlock(&sc->sc_send_lock);
 988		if (ret == size)
 989			break;
 990		if (ret == (ssize_t)-EAGAIN) {
 991			mlog(0, "sendpage of size %zu to " SC_NODEF_FMT
 992			     " returned EAGAIN\n", size, SC_NODEF_ARGS(sc));
 993			cond_resched();
 994			continue;
 995		}
 996		mlog(ML_ERROR, "sendpage of size %zu to " SC_NODEF_FMT
 997		     " failed with %zd\n", size, SC_NODEF_ARGS(sc), ret);
 998		o2net_ensure_shutdown(nn, sc, 0);
 999		break;
1000	}
1001}
1002
1003static void o2net_init_msg(struct o2net_msg *msg, u16 data_len, u16 msg_type, u32 key)
1004{
1005	memset(msg, 0, sizeof(struct o2net_msg));
1006	msg->magic = cpu_to_be16(O2NET_MSG_MAGIC);
1007	msg->data_len = cpu_to_be16(data_len);
1008	msg->msg_type = cpu_to_be16(msg_type);
1009	msg->sys_status = cpu_to_be32(O2NET_ERR_NONE);
1010	msg->status = 0;
1011	msg->key = cpu_to_be32(key);
1012}
1013
1014static int o2net_tx_can_proceed(struct o2net_node *nn,
1015			        struct o2net_sock_container **sc_ret,
1016				int *error)
1017{
1018	int ret = 0;
1019
1020	spin_lock(&nn->nn_lock);
1021	if (nn->nn_persistent_error) {
1022		ret = 1;
1023		*sc_ret = NULL;
1024		*error = nn->nn_persistent_error;
1025	} else if (nn->nn_sc_valid) {
1026		kref_get(&nn->nn_sc->sc_kref);
1027
1028		ret = 1;
1029		*sc_ret = nn->nn_sc;
1030		*error = 0;
1031	}
1032	spin_unlock(&nn->nn_lock);
1033
1034	return ret;
1035}
1036
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1037int o2net_send_message_vec(u32 msg_type, u32 key, struct kvec *caller_vec,
1038			   size_t caller_veclen, u8 target_node, int *status)
1039{
1040	int ret = 0;
1041	struct o2net_msg *msg = NULL;
1042	size_t veclen, caller_bytes = 0;
1043	struct kvec *vec = NULL;
1044	struct o2net_sock_container *sc = NULL;
1045	struct o2net_node *nn = o2net_nn_from_num(target_node);
1046	struct o2net_status_wait nsw = {
1047		.ns_node_item = LIST_HEAD_INIT(nsw.ns_node_item),
1048	};
1049	struct o2net_send_tracking nst;
1050
1051	o2net_init_nst(&nst, msg_type, key, current, target_node);
1052
1053	if (o2net_wq == NULL) {
1054		mlog(0, "attempt to tx without o2netd running\n");
1055		ret = -ESRCH;
1056		goto out;
1057	}
1058
1059	if (caller_veclen == 0) {
1060		mlog(0, "bad kvec array length\n");
1061		ret = -EINVAL;
1062		goto out;
1063	}
1064
1065	caller_bytes = iov_length((struct iovec *)caller_vec, caller_veclen);
1066	if (caller_bytes > O2NET_MAX_PAYLOAD_BYTES) {
1067		mlog(0, "total payload len %zu too large\n", caller_bytes);
1068		ret = -EINVAL;
1069		goto out;
1070	}
1071
1072	if (target_node == o2nm_this_node()) {
1073		ret = -ELOOP;
1074		goto out;
1075	}
1076
1077	o2net_debug_add_nst(&nst);
1078
1079	o2net_set_nst_sock_time(&nst);
1080
1081	wait_event(nn->nn_sc_wq, o2net_tx_can_proceed(nn, &sc, &ret));
1082	if (ret)
1083		goto out;
1084
1085	o2net_set_nst_sock_container(&nst, sc);
1086
1087	veclen = caller_veclen + 1;
1088	vec = kmalloc(sizeof(struct kvec) * veclen, GFP_ATOMIC);
1089	if (vec == NULL) {
1090		mlog(0, "failed to %zu element kvec!\n", veclen);
1091		ret = -ENOMEM;
1092		goto out;
1093	}
1094
1095	msg = kmalloc(sizeof(struct o2net_msg), GFP_ATOMIC);
1096	if (!msg) {
1097		mlog(0, "failed to allocate a o2net_msg!\n");
1098		ret = -ENOMEM;
1099		goto out;
1100	}
1101
1102	o2net_init_msg(msg, caller_bytes, msg_type, key);
1103
1104	vec[0].iov_len = sizeof(struct o2net_msg);
1105	vec[0].iov_base = msg;
1106	memcpy(&vec[1], caller_vec, caller_veclen * sizeof(struct kvec));
1107
1108	ret = o2net_prep_nsw(nn, &nsw);
1109	if (ret)
1110		goto out;
1111
1112	msg->msg_num = cpu_to_be32(nsw.ns_id);
1113	o2net_set_nst_msg_id(&nst, nsw.ns_id);
1114
1115	o2net_set_nst_send_time(&nst);
1116
1117	/* finally, convert the message header to network byte-order
1118	 * and send */
1119	mutex_lock(&sc->sc_send_lock);
1120	ret = o2net_send_tcp_msg(sc->sc_sock, vec, veclen,
1121				 sizeof(struct o2net_msg) + caller_bytes);
1122	mutex_unlock(&sc->sc_send_lock);
1123	msglog(msg, "sending returned %d\n", ret);
1124	if (ret < 0) {
1125		mlog(0, "error returned from o2net_send_tcp_msg=%d\n", ret);
1126		goto out;
1127	}
1128
1129	/* wait on other node's handler */
1130	o2net_set_nst_status_time(&nst);
1131	wait_event(nsw.ns_wq, o2net_nsw_completed(nn, &nsw));
1132
1133	o2net_update_send_stats(&nst, sc);
1134
1135	/* Note that we avoid overwriting the callers status return
1136	 * variable if a system error was reported on the other
1137	 * side. Callers beware. */
1138	ret = o2net_sys_err_to_errno(nsw.ns_sys_status);
1139	if (status && !ret)
1140		*status = nsw.ns_status;
1141
1142	mlog(0, "woken, returning system status %d, user status %d\n",
1143	     ret, nsw.ns_status);
1144out:
1145	o2net_debug_del_nst(&nst); /* must be before dropping sc and node */
1146	if (sc)
1147		sc_put(sc);
1148	if (vec)
1149		kfree(vec);
1150	if (msg)
1151		kfree(msg);
1152	o2net_complete_nsw(nn, &nsw, 0, 0, 0);
1153	return ret;
1154}
1155EXPORT_SYMBOL_GPL(o2net_send_message_vec);
1156
1157int o2net_send_message(u32 msg_type, u32 key, void *data, u32 len,
1158		       u8 target_node, int *status)
1159{
1160	struct kvec vec = {
1161		.iov_base = data,
1162		.iov_len = len,
1163	};
1164	return o2net_send_message_vec(msg_type, key, &vec, 1,
1165				      target_node, status);
1166}
1167EXPORT_SYMBOL_GPL(o2net_send_message);
1168
1169static int o2net_send_status_magic(struct socket *sock, struct o2net_msg *hdr,
1170				   enum o2net_system_error syserr, int err)
1171{
1172	struct kvec vec = {
1173		.iov_base = hdr,
1174		.iov_len = sizeof(struct o2net_msg),
1175	};
1176
1177	BUG_ON(syserr >= O2NET_ERR_MAX);
1178
1179	/* leave other fields intact from the incoming message, msg_num
1180	 * in particular */
1181	hdr->sys_status = cpu_to_be32(syserr);
1182	hdr->status = cpu_to_be32(err);
1183	hdr->magic = cpu_to_be16(O2NET_MSG_STATUS_MAGIC);  // twiddle the magic
1184	hdr->data_len = 0;
1185
1186	msglog(hdr, "about to send status magic %d\n", err);
1187	/* hdr has been in host byteorder this whole time */
1188	return o2net_send_tcp_msg(sock, &vec, 1, sizeof(struct o2net_msg));
1189}
1190
1191/* this returns -errno if the header was unknown or too large, etc.
1192 * after this is called the buffer us reused for the next message */
1193static int o2net_process_message(struct o2net_sock_container *sc,
1194				 struct o2net_msg *hdr)
1195{
1196	struct o2net_node *nn = o2net_nn_from_num(sc->sc_node->nd_num);
1197	int ret = 0, handler_status;
1198	enum  o2net_system_error syserr;
1199	struct o2net_msg_handler *nmh = NULL;
1200	void *ret_data = NULL;
1201
1202	msglog(hdr, "processing message\n");
1203
1204	o2net_sc_postpone_idle(sc);
1205
1206	switch(be16_to_cpu(hdr->magic)) {
1207		case O2NET_MSG_STATUS_MAGIC:
1208			/* special type for returning message status */
1209			o2net_complete_nsw(nn, NULL,
1210					   be32_to_cpu(hdr->msg_num),
1211					   be32_to_cpu(hdr->sys_status),
1212					   be32_to_cpu(hdr->status));
1213			goto out;
1214		case O2NET_MSG_KEEP_REQ_MAGIC:
1215			o2net_sendpage(sc, o2net_keep_resp,
1216				       sizeof(*o2net_keep_resp));
1217			goto out;
1218		case O2NET_MSG_KEEP_RESP_MAGIC:
1219			goto out;
1220		case O2NET_MSG_MAGIC:
1221			break;
1222		default:
1223			msglog(hdr, "bad magic\n");
1224			ret = -EINVAL;
1225			goto out;
1226			break;
1227	}
1228
1229	/* find a handler for it */
1230	handler_status = 0;
1231	nmh = o2net_handler_get(be16_to_cpu(hdr->msg_type),
1232				be32_to_cpu(hdr->key));
1233	if (!nmh) {
1234		mlog(ML_TCP, "couldn't find handler for type %u key %08x\n",
1235		     be16_to_cpu(hdr->msg_type), be32_to_cpu(hdr->key));
1236		syserr = O2NET_ERR_NO_HNDLR;
1237		goto out_respond;
1238	}
1239
1240	syserr = O2NET_ERR_NONE;
1241
1242	if (be16_to_cpu(hdr->data_len) > nmh->nh_max_len)
1243		syserr = O2NET_ERR_OVERFLOW;
1244
1245	if (syserr != O2NET_ERR_NONE)
1246		goto out_respond;
1247
1248	o2net_set_func_start_time(sc);
1249	sc->sc_msg_key = be32_to_cpu(hdr->key);
1250	sc->sc_msg_type = be16_to_cpu(hdr->msg_type);
1251	handler_status = (nmh->nh_func)(hdr, sizeof(struct o2net_msg) +
1252					     be16_to_cpu(hdr->data_len),
1253					nmh->nh_func_data, &ret_data);
1254	o2net_set_func_stop_time(sc);
1255
1256	o2net_update_recv_stats(sc);
1257
1258out_respond:
1259	/* this destroys the hdr, so don't use it after this */
1260	mutex_lock(&sc->sc_send_lock);
1261	ret = o2net_send_status_magic(sc->sc_sock, hdr, syserr,
1262				      handler_status);
1263	mutex_unlock(&sc->sc_send_lock);
1264	hdr = NULL;
1265	mlog(0, "sending handler status %d, syserr %d returned %d\n",
1266	     handler_status, syserr, ret);
1267
1268	if (nmh) {
1269		BUG_ON(ret_data != NULL && nmh->nh_post_func == NULL);
1270		if (nmh->nh_post_func)
1271			(nmh->nh_post_func)(handler_status, nmh->nh_func_data,
1272					    ret_data);
1273	}
1274
1275out:
1276	if (nmh)
1277		o2net_handler_put(nmh);
1278	return ret;
1279}
1280
1281static int o2net_check_handshake(struct o2net_sock_container *sc)
1282{
1283	struct o2net_handshake *hand = page_address(sc->sc_page);
1284	struct o2net_node *nn = o2net_nn_from_num(sc->sc_node->nd_num);
1285
1286	if (hand->protocol_version != cpu_to_be64(O2NET_PROTOCOL_VERSION)) {
1287		mlog(ML_NOTICE, SC_NODEF_FMT " advertised net protocol "
1288		     "version %llu but %llu is required, disconnecting\n",
1289		     SC_NODEF_ARGS(sc),
1290		     (unsigned long long)be64_to_cpu(hand->protocol_version),
1291		     O2NET_PROTOCOL_VERSION);
1292
1293		/* don't bother reconnecting if its the wrong version. */
1294		o2net_ensure_shutdown(nn, sc, -ENOTCONN);
1295		return -1;
1296	}
1297
1298	/*
1299	 * Ensure timeouts are consistent with other nodes, otherwise
1300	 * we can end up with one node thinking that the other must be down,
1301	 * but isn't. This can ultimately cause corruption.
1302	 */
1303	if (be32_to_cpu(hand->o2net_idle_timeout_ms) !=
1304				o2net_idle_timeout()) {
1305		mlog(ML_NOTICE, SC_NODEF_FMT " uses a network idle timeout of "
1306		     "%u ms, but we use %u ms locally.  disconnecting\n",
1307		     SC_NODEF_ARGS(sc),
1308		     be32_to_cpu(hand->o2net_idle_timeout_ms),
1309		     o2net_idle_timeout());
1310		o2net_ensure_shutdown(nn, sc, -ENOTCONN);
1311		return -1;
1312	}
1313
1314	if (be32_to_cpu(hand->o2net_keepalive_delay_ms) !=
1315			o2net_keepalive_delay()) {
1316		mlog(ML_NOTICE, SC_NODEF_FMT " uses a keepalive delay of "
1317		     "%u ms, but we use %u ms locally.  disconnecting\n",
1318		     SC_NODEF_ARGS(sc),
1319		     be32_to_cpu(hand->o2net_keepalive_delay_ms),
1320		     o2net_keepalive_delay());
1321		o2net_ensure_shutdown(nn, sc, -ENOTCONN);
1322		return -1;
1323	}
1324
1325	if (be32_to_cpu(hand->o2hb_heartbeat_timeout_ms) !=
1326			O2HB_MAX_WRITE_TIMEOUT_MS) {
1327		mlog(ML_NOTICE, SC_NODEF_FMT " uses a heartbeat timeout of "
1328		     "%u ms, but we use %u ms locally.  disconnecting\n",
1329		     SC_NODEF_ARGS(sc),
1330		     be32_to_cpu(hand->o2hb_heartbeat_timeout_ms),
1331		     O2HB_MAX_WRITE_TIMEOUT_MS);
1332		o2net_ensure_shutdown(nn, sc, -ENOTCONN);
1333		return -1;
1334	}
1335
1336	sc->sc_handshake_ok = 1;
1337
1338	spin_lock(&nn->nn_lock);
1339	/* set valid and queue the idle timers only if it hasn't been
1340	 * shut down already */
1341	if (nn->nn_sc == sc) {
1342		o2net_sc_reset_idle_timer(sc);
1343		atomic_set(&nn->nn_timeout, 0);
1344		o2net_set_nn_state(nn, sc, 1, 0);
1345	}
1346	spin_unlock(&nn->nn_lock);
1347
1348	/* shift everything up as though it wasn't there */
1349	sc->sc_page_off -= sizeof(struct o2net_handshake);
1350	if (sc->sc_page_off)
1351		memmove(hand, hand + 1, sc->sc_page_off);
1352
1353	return 0;
1354}
1355
1356/* this demuxes the queued rx bytes into header or payload bits and calls
1357 * handlers as each full message is read off the socket.  it returns -error,
1358 * == 0 eof, or > 0 for progress made.*/
1359static int o2net_advance_rx(struct o2net_sock_container *sc)
1360{
1361	struct o2net_msg *hdr;
1362	int ret = 0;
1363	void *data;
1364	size_t datalen;
1365
1366	sclog(sc, "receiving\n");
1367	o2net_set_advance_start_time(sc);
1368
1369	if (unlikely(sc->sc_handshake_ok == 0)) {
1370		if(sc->sc_page_off < sizeof(struct o2net_handshake)) {
1371			data = page_address(sc->sc_page) + sc->sc_page_off;
1372			datalen = sizeof(struct o2net_handshake) - sc->sc_page_off;
1373			ret = o2net_recv_tcp_msg(sc->sc_sock, data, datalen);
1374			if (ret > 0)
1375				sc->sc_page_off += ret;
1376		}
1377
1378		if (sc->sc_page_off == sizeof(struct o2net_handshake)) {
1379			o2net_check_handshake(sc);
1380			if (unlikely(sc->sc_handshake_ok == 0))
1381				ret = -EPROTO;
1382		}
1383		goto out;
1384	}
1385
1386	/* do we need more header? */
1387	if (sc->sc_page_off < sizeof(struct o2net_msg)) {
1388		data = page_address(sc->sc_page) + sc->sc_page_off;
1389		datalen = sizeof(struct o2net_msg) - sc->sc_page_off;
1390		ret = o2net_recv_tcp_msg(sc->sc_sock, data, datalen);
1391		if (ret > 0) {
1392			sc->sc_page_off += ret;
1393			/* only swab incoming here.. we can
1394			 * only get here once as we cross from
1395			 * being under to over */
1396			if (sc->sc_page_off == sizeof(struct o2net_msg)) {
1397				hdr = page_address(sc->sc_page);
1398				if (be16_to_cpu(hdr->data_len) >
1399				    O2NET_MAX_PAYLOAD_BYTES)
1400					ret = -EOVERFLOW;
1401			}
1402		}
1403		if (ret <= 0)
1404			goto out;
1405	}
1406
1407	if (sc->sc_page_off < sizeof(struct o2net_msg)) {
1408		/* oof, still don't have a header */
1409		goto out;
1410	}
1411
1412	/* this was swabbed above when we first read it */
1413	hdr = page_address(sc->sc_page);
1414
1415	msglog(hdr, "at page_off %zu\n", sc->sc_page_off);
1416
1417	/* do we need more payload? */
1418	if (sc->sc_page_off - sizeof(struct o2net_msg) < be16_to_cpu(hdr->data_len)) {
1419		/* need more payload */
1420		data = page_address(sc->sc_page) + sc->sc_page_off;
1421		datalen = (sizeof(struct o2net_msg) + be16_to_cpu(hdr->data_len)) -
1422			  sc->sc_page_off;
1423		ret = o2net_recv_tcp_msg(sc->sc_sock, data, datalen);
1424		if (ret > 0)
1425			sc->sc_page_off += ret;
1426		if (ret <= 0)
1427			goto out;
1428	}
1429
1430	if (sc->sc_page_off - sizeof(struct o2net_msg) == be16_to_cpu(hdr->data_len)) {
1431		/* we can only get here once, the first time we read
1432		 * the payload.. so set ret to progress if the handler
1433		 * works out. after calling this the message is toast */
1434		ret = o2net_process_message(sc, hdr);
1435		if (ret == 0)
1436			ret = 1;
1437		sc->sc_page_off = 0;
1438	}
1439
1440out:
1441	sclog(sc, "ret = %d\n", ret);
1442	o2net_set_advance_stop_time(sc);
1443	return ret;
1444}
1445
1446/* this work func is triggerd by data ready.  it reads until it can read no
1447 * more.  it interprets 0, eof, as fatal.  if data_ready hits while we're doing
1448 * our work the work struct will be marked and we'll be called again. */
1449static void o2net_rx_until_empty(struct work_struct *work)
1450{
1451	struct o2net_sock_container *sc =
1452		container_of(work, struct o2net_sock_container, sc_rx_work);
1453	int ret;
1454
1455	do {
1456		ret = o2net_advance_rx(sc);
1457	} while (ret > 0);
1458
1459	if (ret <= 0 && ret != -EAGAIN) {
1460		struct o2net_node *nn = o2net_nn_from_num(sc->sc_node->nd_num);
1461		sclog(sc, "saw error %d, closing\n", ret);
1462		/* not permanent so read failed handshake can retry */
1463		o2net_ensure_shutdown(nn, sc, 0);
1464	}
1465
1466	sc_put(sc);
1467}
1468
1469static int o2net_set_nodelay(struct socket *sock)
1470{
1471	int ret, val = 1;
1472	mm_segment_t oldfs;
1473
1474	oldfs = get_fs();
1475	set_fs(KERNEL_DS);
1476
1477	/*
1478	 * Dear unsuspecting programmer,
1479	 *
1480	 * Don't use sock_setsockopt() for SOL_TCP.  It doesn't check its level
1481	 * argument and assumes SOL_SOCKET so, say, your TCP_NODELAY will
1482	 * silently turn into SO_DEBUG.
1483	 *
1484	 * Yours,
1485	 * Keeper of hilariously fragile interfaces.
1486	 */
1487	ret = sock->ops->setsockopt(sock, SOL_TCP, TCP_NODELAY,
1488				    (char __user *)&val, sizeof(val));
1489
1490	set_fs(oldfs);
1491	return ret;
1492}
1493
1494static void o2net_initialize_handshake(void)
1495{
1496	o2net_hand->o2hb_heartbeat_timeout_ms = cpu_to_be32(
1497		O2HB_MAX_WRITE_TIMEOUT_MS);
1498	o2net_hand->o2net_idle_timeout_ms = cpu_to_be32(o2net_idle_timeout());
1499	o2net_hand->o2net_keepalive_delay_ms = cpu_to_be32(
1500		o2net_keepalive_delay());
1501	o2net_hand->o2net_reconnect_delay_ms = cpu_to_be32(
1502		o2net_reconnect_delay());
1503}
1504
1505/* ------------------------------------------------------------ */
1506
1507/* called when a connect completes and after a sock is accepted.  the
1508 * rx path will see the response and mark the sc valid */
1509static void o2net_sc_connect_completed(struct work_struct *work)
1510{
1511	struct o2net_sock_container *sc =
1512		container_of(work, struct o2net_sock_container,
1513			     sc_connect_work);
1514
1515	mlog(ML_MSG, "sc sending handshake with ver %llu id %llx\n",
1516              (unsigned long long)O2NET_PROTOCOL_VERSION,
1517	      (unsigned long long)be64_to_cpu(o2net_hand->connector_id));
1518
1519	o2net_initialize_handshake();
1520	o2net_sendpage(sc, o2net_hand, sizeof(*o2net_hand));
1521	sc_put(sc);
1522}
1523
1524/* this is called as a work_struct func. */
1525static void o2net_sc_send_keep_req(struct work_struct *work)
1526{
1527	struct o2net_sock_container *sc =
1528		container_of(work, struct o2net_sock_container,
1529			     sc_keepalive_work.work);
1530
1531	o2net_sendpage(sc, o2net_keep_req, sizeof(*o2net_keep_req));
1532	sc_put(sc);
1533}
1534
1535/* socket shutdown does a del_timer_sync against this as it tears down.
1536 * we can't start this timer until we've got to the point in sc buildup
1537 * where shutdown is going to be involved */
1538static void o2net_idle_timer(unsigned long data)
1539{
1540	struct o2net_sock_container *sc = (struct o2net_sock_container *)data;
1541	struct o2net_node *nn = o2net_nn_from_num(sc->sc_node->nd_num);
1542
1543#ifdef CONFIG_DEBUG_FS
1544	ktime_t now = ktime_get();
 
 
 
1545#endif
1546
1547	printk(KERN_NOTICE "o2net: connection to " SC_NODEF_FMT " has been idle for %u.%u "
1548	     "seconds, shutting it down.\n", SC_NODEF_ARGS(sc),
1549		     o2net_idle_timeout() / 1000,
1550		     o2net_idle_timeout() % 1000);
1551
1552#ifdef CONFIG_DEBUG_FS
1553	mlog(ML_NOTICE, "Here are some times that might help debug the "
1554	     "situation: (Timer: %lld, Now %lld, DataReady %lld, Advance %lld-%lld, "
1555	     "Key 0x%08x, Func %u, FuncTime %lld-%lld)\n",
1556	     (long long)ktime_to_us(sc->sc_tv_timer), (long long)ktime_to_us(now),
1557	     (long long)ktime_to_us(sc->sc_tv_data_ready),
1558	     (long long)ktime_to_us(sc->sc_tv_advance_start),
1559	     (long long)ktime_to_us(sc->sc_tv_advance_stop),
1560	     sc->sc_msg_key, sc->sc_msg_type,
1561	     (long long)ktime_to_us(sc->sc_tv_func_start),
1562	     (long long)ktime_to_us(sc->sc_tv_func_stop));
1563#endif
1564
1565	/*
1566	 * Initialize the nn_timeout so that the next connection attempt
1567	 * will continue in o2net_start_connect.
1568	 */
1569	atomic_set(&nn->nn_timeout, 1);
1570
1571	o2net_sc_queue_work(sc, &sc->sc_shutdown_work);
1572}
1573
1574static void o2net_sc_reset_idle_timer(struct o2net_sock_container *sc)
1575{
1576	o2net_sc_cancel_delayed_work(sc, &sc->sc_keepalive_work);
1577	o2net_sc_queue_delayed_work(sc, &sc->sc_keepalive_work,
1578		      msecs_to_jiffies(o2net_keepalive_delay()));
1579	o2net_set_sock_timer(sc);
1580	mod_timer(&sc->sc_idle_timeout,
1581	       jiffies + msecs_to_jiffies(o2net_idle_timeout()));
1582}
1583
1584static void o2net_sc_postpone_idle(struct o2net_sock_container *sc)
1585{
1586	/* Only push out an existing timer */
1587	if (timer_pending(&sc->sc_idle_timeout))
1588		o2net_sc_reset_idle_timer(sc);
1589}
1590
1591/* this work func is kicked whenever a path sets the nn state which doesn't
1592 * have valid set.  This includes seeing hb come up, losing a connection,
1593 * having a connect attempt fail, etc. This centralizes the logic which decides
1594 * if a connect attempt should be made or if we should give up and all future
1595 * transmit attempts should fail */
1596static void o2net_start_connect(struct work_struct *work)
1597{
1598	struct o2net_node *nn =
1599		container_of(work, struct o2net_node, nn_connect_work.work);
1600	struct o2net_sock_container *sc = NULL;
1601	struct o2nm_node *node = NULL, *mynode = NULL;
1602	struct socket *sock = NULL;
1603	struct sockaddr_in myaddr = {0, }, remoteaddr = {0, };
1604	int ret = 0, stop;
1605	unsigned int timeout;
1606
1607	/* if we're greater we initiate tx, otherwise we accept */
1608	if (o2nm_this_node() <= o2net_num_from_nn(nn))
1609		goto out;
1610
1611	/* watch for racing with tearing a node down */
1612	node = o2nm_get_node_by_num(o2net_num_from_nn(nn));
1613	if (node == NULL) {
1614		ret = 0;
1615		goto out;
1616	}
1617
1618	mynode = o2nm_get_node_by_num(o2nm_this_node());
1619	if (mynode == NULL) {
1620		ret = 0;
1621		goto out;
1622	}
1623
1624	spin_lock(&nn->nn_lock);
1625	/*
1626	 * see if we already have one pending or have given up.
1627	 * For nn_timeout, it is set when we close the connection
1628	 * because of the idle time out. So it means that we have
1629	 * at least connected to that node successfully once,
1630	 * now try to connect to it again.
1631	 */
1632	timeout = atomic_read(&nn->nn_timeout);
1633	stop = (nn->nn_sc ||
1634		(nn->nn_persistent_error &&
1635		(nn->nn_persistent_error != -ENOTCONN || timeout == 0)));
1636	spin_unlock(&nn->nn_lock);
1637	if (stop)
1638		goto out;
1639
1640	nn->nn_last_connect_attempt = jiffies;
1641
1642	sc = sc_alloc(node);
1643	if (sc == NULL) {
1644		mlog(0, "couldn't allocate sc\n");
1645		ret = -ENOMEM;
1646		goto out;
1647	}
1648
1649	ret = sock_create(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock);
1650	if (ret < 0) {
1651		mlog(0, "can't create socket: %d\n", ret);
1652		goto out;
1653	}
1654	sc->sc_sock = sock; /* freed by sc_kref_release */
1655
1656	sock->sk->sk_allocation = GFP_ATOMIC;
1657
1658	myaddr.sin_family = AF_INET;
1659	myaddr.sin_addr.s_addr = mynode->nd_ipv4_address;
1660	myaddr.sin_port = htons(0); /* any port */
1661
1662	ret = sock->ops->bind(sock, (struct sockaddr *)&myaddr,
1663			      sizeof(myaddr));
1664	if (ret) {
1665		mlog(ML_ERROR, "bind failed with %d at address %pI4\n",
1666		     ret, &mynode->nd_ipv4_address);
1667		goto out;
1668	}
1669
1670	ret = o2net_set_nodelay(sc->sc_sock);
1671	if (ret) {
1672		mlog(ML_ERROR, "setting TCP_NODELAY failed with %d\n", ret);
1673		goto out;
1674	}
1675
1676	o2net_register_callbacks(sc->sc_sock->sk, sc);
1677
1678	spin_lock(&nn->nn_lock);
1679	/* handshake completion will set nn->nn_sc_valid */
1680	o2net_set_nn_state(nn, sc, 0, 0);
1681	spin_unlock(&nn->nn_lock);
1682
1683	remoteaddr.sin_family = AF_INET;
1684	remoteaddr.sin_addr.s_addr = node->nd_ipv4_address;
1685	remoteaddr.sin_port = node->nd_ipv4_port;
1686
1687	ret = sc->sc_sock->ops->connect(sc->sc_sock,
1688					(struct sockaddr *)&remoteaddr,
1689					sizeof(remoteaddr),
1690					O_NONBLOCK);
1691	if (ret == -EINPROGRESS)
1692		ret = 0;
1693
1694out:
1695	if (ret) {
1696		mlog(ML_NOTICE, "connect attempt to " SC_NODEF_FMT " failed "
1697		     "with errno %d\n", SC_NODEF_ARGS(sc), ret);
1698		/* 0 err so that another will be queued and attempted
1699		 * from set_nn_state */
1700		if (sc)
1701			o2net_ensure_shutdown(nn, sc, 0);
1702	}
1703	if (sc)
1704		sc_put(sc);
1705	if (node)
1706		o2nm_node_put(node);
1707	if (mynode)
1708		o2nm_node_put(mynode);
1709
1710	return;
1711}
1712
1713static void o2net_connect_expired(struct work_struct *work)
1714{
1715	struct o2net_node *nn =
1716		container_of(work, struct o2net_node, nn_connect_expired.work);
1717
1718	spin_lock(&nn->nn_lock);
1719	if (!nn->nn_sc_valid) {
1720		mlog(ML_ERROR, "no connection established with node %u after "
1721		     "%u.%u seconds, giving up and returning errors.\n",
1722		     o2net_num_from_nn(nn),
1723		     o2net_idle_timeout() / 1000,
1724		     o2net_idle_timeout() % 1000);
1725
1726		o2net_set_nn_state(nn, NULL, 0, -ENOTCONN);
1727	}
1728	spin_unlock(&nn->nn_lock);
1729}
1730
1731static void o2net_still_up(struct work_struct *work)
1732{
1733	struct o2net_node *nn =
1734		container_of(work, struct o2net_node, nn_still_up.work);
1735
1736	o2quo_hb_still_up(o2net_num_from_nn(nn));
1737}
1738
1739/* ------------------------------------------------------------ */
1740
1741void o2net_disconnect_node(struct o2nm_node *node)
1742{
1743	struct o2net_node *nn = o2net_nn_from_num(node->nd_num);
1744
1745	/* don't reconnect until it's heartbeating again */
1746	spin_lock(&nn->nn_lock);
1747	atomic_set(&nn->nn_timeout, 0);
1748	o2net_set_nn_state(nn, NULL, 0, -ENOTCONN);
1749	spin_unlock(&nn->nn_lock);
1750
1751	if (o2net_wq) {
1752		cancel_delayed_work(&nn->nn_connect_expired);
1753		cancel_delayed_work(&nn->nn_connect_work);
1754		cancel_delayed_work(&nn->nn_still_up);
1755		flush_workqueue(o2net_wq);
1756	}
1757}
1758
1759static void o2net_hb_node_down_cb(struct o2nm_node *node, int node_num,
1760				  void *data)
1761{
1762	o2quo_hb_down(node_num);
1763
1764	if (!node)
1765		return;
1766
1767	if (node_num != o2nm_this_node())
1768		o2net_disconnect_node(node);
1769
1770	BUG_ON(atomic_read(&o2net_connected_peers) < 0);
1771}
1772
1773static void o2net_hb_node_up_cb(struct o2nm_node *node, int node_num,
1774				void *data)
1775{
1776	struct o2net_node *nn = o2net_nn_from_num(node_num);
1777
1778	o2quo_hb_up(node_num);
1779
1780	BUG_ON(!node);
1781
1782	/* ensure an immediate connect attempt */
1783	nn->nn_last_connect_attempt = jiffies -
1784		(msecs_to_jiffies(o2net_reconnect_delay()) + 1);
1785
1786	if (node_num != o2nm_this_node()) {
1787		/* believe it or not, accept and node hearbeating testing
1788		 * can succeed for this node before we got here.. so
1789		 * only use set_nn_state to clear the persistent error
1790		 * if that hasn't already happened */
1791		spin_lock(&nn->nn_lock);
1792		atomic_set(&nn->nn_timeout, 0);
1793		if (nn->nn_persistent_error)
1794			o2net_set_nn_state(nn, NULL, 0, 0);
1795		spin_unlock(&nn->nn_lock);
1796	}
1797}
1798
1799void o2net_unregister_hb_callbacks(void)
1800{
1801	o2hb_unregister_callback(NULL, &o2net_hb_up);
1802	o2hb_unregister_callback(NULL, &o2net_hb_down);
1803}
1804
1805int o2net_register_hb_callbacks(void)
1806{
1807	int ret;
1808
1809	o2hb_setup_callback(&o2net_hb_down, O2HB_NODE_DOWN_CB,
1810			    o2net_hb_node_down_cb, NULL, O2NET_HB_PRI);
1811	o2hb_setup_callback(&o2net_hb_up, O2HB_NODE_UP_CB,
1812			    o2net_hb_node_up_cb, NULL, O2NET_HB_PRI);
1813
1814	ret = o2hb_register_callback(NULL, &o2net_hb_up);
1815	if (ret == 0)
1816		ret = o2hb_register_callback(NULL, &o2net_hb_down);
1817
1818	if (ret)
1819		o2net_unregister_hb_callbacks();
1820
1821	return ret;
1822}
1823
1824/* ------------------------------------------------------------ */
1825
1826static int o2net_accept_one(struct socket *sock)
1827{
1828	int ret, slen;
1829	struct sockaddr_in sin;
1830	struct socket *new_sock = NULL;
1831	struct o2nm_node *node = NULL;
1832	struct o2nm_node *local_node = NULL;
1833	struct o2net_sock_container *sc = NULL;
1834	struct o2net_node *nn;
1835
1836	BUG_ON(sock == NULL);
1837	ret = sock_create_lite(sock->sk->sk_family, sock->sk->sk_type,
1838			       sock->sk->sk_protocol, &new_sock);
1839	if (ret)
1840		goto out;
1841
1842	new_sock->type = sock->type;
1843	new_sock->ops = sock->ops;
1844	ret = sock->ops->accept(sock, new_sock, O_NONBLOCK);
1845	if (ret < 0)
1846		goto out;
1847
1848	new_sock->sk->sk_allocation = GFP_ATOMIC;
1849
1850	ret = o2net_set_nodelay(new_sock);
1851	if (ret) {
1852		mlog(ML_ERROR, "setting TCP_NODELAY failed with %d\n", ret);
1853		goto out;
1854	}
1855
1856	slen = sizeof(sin);
1857	ret = new_sock->ops->getname(new_sock, (struct sockaddr *) &sin,
1858				       &slen, 1);
1859	if (ret < 0)
1860		goto out;
1861
1862	node = o2nm_get_node_by_ip(sin.sin_addr.s_addr);
1863	if (node == NULL) {
1864		mlog(ML_NOTICE, "attempt to connect from unknown node at %pI4:%d\n",
1865		     &sin.sin_addr.s_addr, ntohs(sin.sin_port));
 
1866		ret = -EINVAL;
1867		goto out;
1868	}
1869
1870	if (o2nm_this_node() >= node->nd_num) {
1871		local_node = o2nm_get_node_by_num(o2nm_this_node());
1872		mlog(ML_NOTICE, "unexpected connect attempt seen at node '%s' ("
1873		     "%u, %pI4:%d) from node '%s' (%u, %pI4:%d)\n",
1874		     local_node->nd_name, local_node->nd_num,
1875		     &(local_node->nd_ipv4_address),
1876		     ntohs(local_node->nd_ipv4_port),
1877		     node->nd_name, node->nd_num, &sin.sin_addr.s_addr,
1878		     ntohs(sin.sin_port));
 
 
 
1879		ret = -EINVAL;
1880		goto out;
1881	}
1882
1883	/* this happens all the time when the other node sees our heartbeat
1884	 * and tries to connect before we see their heartbeat */
1885	if (!o2hb_check_node_heartbeating_from_callback(node->nd_num)) {
1886		mlog(ML_CONN, "attempt to connect from node '%s' at "
1887		     "%pI4:%d but it isn't heartbeating\n",
1888		     node->nd_name, &sin.sin_addr.s_addr,
1889		     ntohs(sin.sin_port));
1890		ret = -EINVAL;
1891		goto out;
1892	}
1893
1894	nn = o2net_nn_from_num(node->nd_num);
1895
1896	spin_lock(&nn->nn_lock);
1897	if (nn->nn_sc)
1898		ret = -EBUSY;
1899	else
1900		ret = 0;
1901	spin_unlock(&nn->nn_lock);
1902	if (ret) {
1903		mlog(ML_NOTICE, "attempt to connect from node '%s' at "
1904		     "%pI4:%d but it already has an open connection\n",
1905		     node->nd_name, &sin.sin_addr.s_addr,
1906		     ntohs(sin.sin_port));
1907		goto out;
1908	}
1909
1910	sc = sc_alloc(node);
1911	if (sc == NULL) {
1912		ret = -ENOMEM;
1913		goto out;
1914	}
1915
1916	sc->sc_sock = new_sock;
1917	new_sock = NULL;
1918
1919	spin_lock(&nn->nn_lock);
1920	atomic_set(&nn->nn_timeout, 0);
1921	o2net_set_nn_state(nn, sc, 0, 0);
1922	spin_unlock(&nn->nn_lock);
1923
1924	o2net_register_callbacks(sc->sc_sock->sk, sc);
1925	o2net_sc_queue_work(sc, &sc->sc_rx_work);
1926
1927	o2net_initialize_handshake();
1928	o2net_sendpage(sc, o2net_hand, sizeof(*o2net_hand));
1929
1930out:
1931	if (new_sock)
1932		sock_release(new_sock);
1933	if (node)
1934		o2nm_node_put(node);
1935	if (local_node)
1936		o2nm_node_put(local_node);
1937	if (sc)
1938		sc_put(sc);
1939	return ret;
1940}
1941
1942static void o2net_accept_many(struct work_struct *work)
1943{
1944	struct socket *sock = o2net_listen_sock;
1945	while (o2net_accept_one(sock) == 0)
1946		cond_resched();
1947}
1948
1949static void o2net_listen_data_ready(struct sock *sk, int bytes)
1950{
1951	void (*ready)(struct sock *sk, int bytes);
1952
1953	read_lock(&sk->sk_callback_lock);
1954	ready = sk->sk_user_data;
1955	if (ready == NULL) { /* check for teardown race */
1956		ready = sk->sk_data_ready;
1957		goto out;
1958	}
1959
1960	/* ->sk_data_ready is also called for a newly established child socket
1961	 * before it has been accepted and the acceptor has set up their
1962	 * data_ready.. we only want to queue listen work for our listening
1963	 * socket */
 
 
 
 
 
 
 
 
 
1964	if (sk->sk_state == TCP_LISTEN) {
1965		mlog(ML_TCP, "bytes: %d\n", bytes);
1966		queue_work(o2net_wq, &o2net_listen_work);
 
 
1967	}
1968
1969out:
1970	read_unlock(&sk->sk_callback_lock);
1971	ready(sk, bytes);
 
1972}
1973
1974static int o2net_open_listening_sock(__be32 addr, __be16 port)
1975{
1976	struct socket *sock = NULL;
1977	int ret;
1978	struct sockaddr_in sin = {
1979		.sin_family = PF_INET,
1980		.sin_addr = { .s_addr = addr },
1981		.sin_port = port,
1982	};
1983
1984	ret = sock_create(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock);
1985	if (ret < 0) {
1986		mlog(ML_ERROR, "unable to create socket, ret=%d\n", ret);
1987		goto out;
1988	}
1989
1990	sock->sk->sk_allocation = GFP_ATOMIC;
1991
1992	write_lock_bh(&sock->sk->sk_callback_lock);
1993	sock->sk->sk_user_data = sock->sk->sk_data_ready;
1994	sock->sk->sk_data_ready = o2net_listen_data_ready;
1995	write_unlock_bh(&sock->sk->sk_callback_lock);
1996
1997	o2net_listen_sock = sock;
1998	INIT_WORK(&o2net_listen_work, o2net_accept_many);
1999
2000	sock->sk->sk_reuse = 1;
2001	ret = sock->ops->bind(sock, (struct sockaddr *)&sin, sizeof(sin));
2002	if (ret < 0) {
2003		mlog(ML_ERROR, "unable to bind socket at %pI4:%u, "
2004		     "ret=%d\n", &addr, ntohs(port), ret);
2005		goto out;
2006	}
2007
2008	ret = sock->ops->listen(sock, 64);
2009	if (ret < 0) {
2010		mlog(ML_ERROR, "unable to listen on %pI4:%u, ret=%d\n",
2011		     &addr, ntohs(port), ret);
2012	}
2013
2014out:
2015	if (ret) {
2016		o2net_listen_sock = NULL;
2017		if (sock)
2018			sock_release(sock);
2019	}
2020	return ret;
2021}
2022
2023/*
2024 * called from node manager when we should bring up our network listening
2025 * socket.  node manager handles all the serialization to only call this
2026 * once and to match it with o2net_stop_listening().  note,
2027 * o2nm_this_node() doesn't work yet as we're being called while it
2028 * is being set up.
2029 */
2030int o2net_start_listening(struct o2nm_node *node)
2031{
2032	int ret = 0;
2033
2034	BUG_ON(o2net_wq != NULL);
2035	BUG_ON(o2net_listen_sock != NULL);
2036
2037	mlog(ML_KTHREAD, "starting o2net thread...\n");
2038	o2net_wq = create_singlethread_workqueue("o2net");
2039	if (o2net_wq == NULL) {
2040		mlog(ML_ERROR, "unable to launch o2net thread\n");
2041		return -ENOMEM; /* ? */
2042	}
2043
2044	ret = o2net_open_listening_sock(node->nd_ipv4_address,
2045					node->nd_ipv4_port);
2046	if (ret) {
2047		destroy_workqueue(o2net_wq);
2048		o2net_wq = NULL;
2049	} else
2050		o2quo_conn_up(node->nd_num);
2051
2052	return ret;
2053}
2054
2055/* again, o2nm_this_node() doesn't work here as we're involved in
2056 * tearing it down */
2057void o2net_stop_listening(struct o2nm_node *node)
2058{
2059	struct socket *sock = o2net_listen_sock;
2060	size_t i;
2061
2062	BUG_ON(o2net_wq == NULL);
2063	BUG_ON(o2net_listen_sock == NULL);
2064
2065	/* stop the listening socket from generating work */
2066	write_lock_bh(&sock->sk->sk_callback_lock);
2067	sock->sk->sk_data_ready = sock->sk->sk_user_data;
2068	sock->sk->sk_user_data = NULL;
2069	write_unlock_bh(&sock->sk->sk_callback_lock);
2070
2071	for (i = 0; i < ARRAY_SIZE(o2net_nodes); i++) {
2072		struct o2nm_node *node = o2nm_get_node_by_num(i);
2073		if (node) {
2074			o2net_disconnect_node(node);
2075			o2nm_node_put(node);
2076		}
2077	}
2078
2079	/* finish all work and tear down the work queue */
2080	mlog(ML_KTHREAD, "waiting for o2net thread to exit....\n");
2081	destroy_workqueue(o2net_wq);
2082	o2net_wq = NULL;
2083
2084	sock_release(o2net_listen_sock);
2085	o2net_listen_sock = NULL;
2086
2087	o2quo_conn_err(node->nd_num);
2088}
2089
2090/* ------------------------------------------------------------ */
2091
2092int o2net_init(void)
2093{
2094	unsigned long i;
2095
2096	o2quo_init();
2097
2098	if (o2net_debugfs_init())
2099		return -ENOMEM;
2100
2101	o2net_hand = kzalloc(sizeof(struct o2net_handshake), GFP_KERNEL);
2102	o2net_keep_req = kzalloc(sizeof(struct o2net_msg), GFP_KERNEL);
2103	o2net_keep_resp = kzalloc(sizeof(struct o2net_msg), GFP_KERNEL);
2104	if (!o2net_hand || !o2net_keep_req || !o2net_keep_resp) {
2105		kfree(o2net_hand);
2106		kfree(o2net_keep_req);
2107		kfree(o2net_keep_resp);
2108		return -ENOMEM;
2109	}
2110
2111	o2net_hand->protocol_version = cpu_to_be64(O2NET_PROTOCOL_VERSION);
2112	o2net_hand->connector_id = cpu_to_be64(1);
2113
2114	o2net_keep_req->magic = cpu_to_be16(O2NET_MSG_KEEP_REQ_MAGIC);
2115	o2net_keep_resp->magic = cpu_to_be16(O2NET_MSG_KEEP_RESP_MAGIC);
2116
2117	for (i = 0; i < ARRAY_SIZE(o2net_nodes); i++) {
2118		struct o2net_node *nn = o2net_nn_from_num(i);
2119
2120		atomic_set(&nn->nn_timeout, 0);
2121		spin_lock_init(&nn->nn_lock);
2122		INIT_DELAYED_WORK(&nn->nn_connect_work, o2net_start_connect);
2123		INIT_DELAYED_WORK(&nn->nn_connect_expired,
2124				  o2net_connect_expired);
2125		INIT_DELAYED_WORK(&nn->nn_still_up, o2net_still_up);
2126		/* until we see hb from a node we'll return einval */
2127		nn->nn_persistent_error = -ENOTCONN;
2128		init_waitqueue_head(&nn->nn_sc_wq);
2129		idr_init(&nn->nn_status_idr);
2130		INIT_LIST_HEAD(&nn->nn_status_list);
2131	}
2132
2133	return 0;
2134}
2135
2136void o2net_exit(void)
2137{
2138	o2quo_exit();
2139	kfree(o2net_hand);
2140	kfree(o2net_keep_req);
2141	kfree(o2net_keep_resp);
2142	o2net_debugfs_exit();
2143}