blob: 9fbe849f634419d9f5418d6097e789906ba460b1 [file] [log] [blame]
Zach Brown98211482005-12-15 14:31:23 -08001/* -*- 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>
Trond Myklebust91cf45f2007-11-12 18:10:39 -080061#include <linux/net.h>
Zach Brown98211482005-12-15 14:31:23 -080062#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
Zach Brown98211482005-12-15 14:31:23 -080075#define SC_NODEF_FMT "node %s (num %u) at %u.%u.%u.%u:%u"
76#define SC_NODEF_ARGS(sc) sc->sc_node->nd_name, sc->sc_node->nd_num, \
77 NIPQUAD(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
Ingo Molnar34af9462006-06-27 02:53:55 -0700104static DEFINE_RWLOCK(o2net_handler_lock);
Zach Brown98211482005-12-15 14:31:23 -0800105static 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 :/ */
David Howellsc4028952006-11-22 14:57:56 +0000136static 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);
Zach Brown98211482005-12-15 14:31:23 -0800139static void o2net_listen_data_ready(struct sock *sk, int bytes);
David Howellsc4028952006-11-22 14:57:56 +0000140static void o2net_sc_send_keep_req(struct work_struct *work);
Zach Brown98211482005-12-15 14:31:23 -0800141static void o2net_idle_timer(unsigned long data);
142static void o2net_sc_postpone_idle(struct o2net_sock_container *sc);
Jeff Mahoneyb5dd8032006-12-04 14:04:54 +0100143static void o2net_sc_reset_idle_timer(struct o2net_sock_container *sc);
144
Sunil Mushran2309e9e2008-04-14 10:46:19 -0700145#ifdef CONFIG_DEBUG_FS
Adrian Bunk18496e82008-08-07 00:11:12 +0300146static void o2net_init_nst(struct o2net_send_tracking *nst, u32 msgtype,
147 u32 msgkey, struct task_struct *task, u8 node)
Sunil Mushran0f475b22008-05-12 18:31:37 -0700148{
Sunil Mushran2309e9e2008-04-14 10:46:19 -0700149 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;
Sunil Mushran2309e9e2008-04-14 10:46:19 -0700154}
155
Adrian Bunk18496e82008-08-07 00:11:12 +0300156static void o2net_set_nst_sock_time(struct o2net_send_tracking *nst)
Sunil Mushran2309e9e2008-04-14 10:46:19 -0700157{
Sunil Mushran2309e9e2008-04-14 10:46:19 -0700158 do_gettimeofday(&nst->st_sock_time);
Sunil Mushran2309e9e2008-04-14 10:46:19 -0700159}
160
Adrian Bunk18496e82008-08-07 00:11:12 +0300161static void o2net_set_nst_send_time(struct o2net_send_tracking *nst)
Sunil Mushran2309e9e2008-04-14 10:46:19 -0700162{
Sunil Mushran2309e9e2008-04-14 10:46:19 -0700163 do_gettimeofday(&nst->st_send_time);
Sunil Mushran2309e9e2008-04-14 10:46:19 -0700164}
165
Adrian Bunk18496e82008-08-07 00:11:12 +0300166static void o2net_set_nst_status_time(struct o2net_send_tracking *nst)
Sunil Mushran2309e9e2008-04-14 10:46:19 -0700167{
Sunil Mushran2309e9e2008-04-14 10:46:19 -0700168 do_gettimeofday(&nst->st_status_time);
Sunil Mushran2309e9e2008-04-14 10:46:19 -0700169}
170
Adrian Bunk18496e82008-08-07 00:11:12 +0300171static void o2net_set_nst_sock_container(struct o2net_send_tracking *nst,
Sunil Mushran2309e9e2008-04-14 10:46:19 -0700172 struct o2net_sock_container *sc)
173{
Sunil Mushran2309e9e2008-04-14 10:46:19 -0700174 nst->st_sc = sc;
Sunil Mushran2309e9e2008-04-14 10:46:19 -0700175}
176
Adrian Bunk18496e82008-08-07 00:11:12 +0300177static void o2net_set_nst_msg_id(struct o2net_send_tracking *nst, u32 msg_id)
Sunil Mushran2309e9e2008-04-14 10:46:19 -0700178{
Sunil Mushran2309e9e2008-04-14 10:46:19 -0700179 nst->st_id = msg_id;
Sunil Mushran2309e9e2008-04-14 10:46:19 -0700180}
Adrian Bunk18496e82008-08-07 00:11:12 +0300181
182#else /* CONFIG_DEBUG_FS */
183
184static inline void o2net_init_nst(struct o2net_send_tracking *nst, u32 msgtype,
185 u32 msgkey, struct task_struct *task, u8 node)
186{
187}
188
189static inline void o2net_set_nst_sock_time(struct o2net_send_tracking *nst)
190{
191}
192
193static inline void o2net_set_nst_send_time(struct o2net_send_tracking *nst)
194{
195}
196
197static inline void o2net_set_nst_status_time(struct o2net_send_tracking *nst)
198{
199}
200
201static inline void o2net_set_nst_sock_container(struct o2net_send_tracking *nst,
202 struct o2net_sock_container *sc)
203{
204}
205
206static inline void o2net_set_nst_msg_id(struct o2net_send_tracking *nst,
207 u32 msg_id)
208{
209}
210
Sunil Mushran0f475b22008-05-12 18:31:37 -0700211#endif /* CONFIG_DEBUG_FS */
Sunil Mushran2309e9e2008-04-14 10:46:19 -0700212
Jeff Mahoney409753b2008-03-28 16:44:13 -0700213static inline int o2net_reconnect_delay(void)
Jeff Mahoneyb5dd8032006-12-04 14:04:54 +0100214{
215 return o2nm_single_cluster->cl_reconnect_delay_ms;
216}
217
Jeff Mahoney409753b2008-03-28 16:44:13 -0700218static inline int o2net_keepalive_delay(void)
Jeff Mahoneyb5dd8032006-12-04 14:04:54 +0100219{
220 return o2nm_single_cluster->cl_keepalive_delay_ms;
221}
222
Jeff Mahoney409753b2008-03-28 16:44:13 -0700223static inline int o2net_idle_timeout(void)
Jeff Mahoneyb5dd8032006-12-04 14:04:54 +0100224{
225 return o2nm_single_cluster->cl_idle_timeout_ms;
226}
Zach Brown98211482005-12-15 14:31:23 -0800227
228static inline int o2net_sys_err_to_errno(enum o2net_system_error err)
229{
230 int trans;
231 BUG_ON(err >= O2NET_ERR_MAX);
232 trans = o2net_sys_err_translations[err];
233
234 /* Just in case we mess up the translation table above */
235 BUG_ON(err != O2NET_ERR_NONE && trans == 0);
236 return trans;
237}
238
239static struct o2net_node * o2net_nn_from_num(u8 node_num)
240{
241 BUG_ON(node_num >= ARRAY_SIZE(o2net_nodes));
242 return &o2net_nodes[node_num];
243}
244
245static u8 o2net_num_from_nn(struct o2net_node *nn)
246{
247 BUG_ON(nn == NULL);
248 return nn - o2net_nodes;
249}
250
251/* ------------------------------------------------------------ */
252
253static int o2net_prep_nsw(struct o2net_node *nn, struct o2net_status_wait *nsw)
254{
255 int ret = 0;
256
257 do {
258 if (!idr_pre_get(&nn->nn_status_idr, GFP_ATOMIC)) {
259 ret = -EAGAIN;
260 break;
261 }
262 spin_lock(&nn->nn_lock);
263 ret = idr_get_new(&nn->nn_status_idr, nsw, &nsw->ns_id);
264 if (ret == 0)
265 list_add_tail(&nsw->ns_node_item,
266 &nn->nn_status_list);
267 spin_unlock(&nn->nn_lock);
268 } while (ret == -EAGAIN);
269
270 if (ret == 0) {
271 init_waitqueue_head(&nsw->ns_wq);
272 nsw->ns_sys_status = O2NET_ERR_NONE;
273 nsw->ns_status = 0;
274 }
275
276 return ret;
277}
278
279static void o2net_complete_nsw_locked(struct o2net_node *nn,
280 struct o2net_status_wait *nsw,
281 enum o2net_system_error sys_status,
282 s32 status)
283{
284 assert_spin_locked(&nn->nn_lock);
285
286 if (!list_empty(&nsw->ns_node_item)) {
287 list_del_init(&nsw->ns_node_item);
288 nsw->ns_sys_status = sys_status;
289 nsw->ns_status = status;
290 idr_remove(&nn->nn_status_idr, nsw->ns_id);
291 wake_up(&nsw->ns_wq);
292 }
293}
294
295static void o2net_complete_nsw(struct o2net_node *nn,
296 struct o2net_status_wait *nsw,
297 u64 id, enum o2net_system_error sys_status,
298 s32 status)
299{
300 spin_lock(&nn->nn_lock);
301 if (nsw == NULL) {
302 if (id > INT_MAX)
303 goto out;
304
305 nsw = idr_find(&nn->nn_status_idr, id);
306 if (nsw == NULL)
307 goto out;
308 }
309
310 o2net_complete_nsw_locked(nn, nsw, sys_status, status);
311
312out:
313 spin_unlock(&nn->nn_lock);
314 return;
315}
316
317static void o2net_complete_nodes_nsw(struct o2net_node *nn)
318{
Christoph Hellwig800deef2007-05-17 16:03:13 +0200319 struct o2net_status_wait *nsw, *tmp;
Zach Brown98211482005-12-15 14:31:23 -0800320 unsigned int num_kills = 0;
Zach Brown98211482005-12-15 14:31:23 -0800321
322 assert_spin_locked(&nn->nn_lock);
323
Christoph Hellwig800deef2007-05-17 16:03:13 +0200324 list_for_each_entry_safe(nsw, tmp, &nn->nn_status_list, ns_node_item) {
Zach Brown98211482005-12-15 14:31:23 -0800325 o2net_complete_nsw_locked(nn, nsw, O2NET_ERR_DIED, 0);
326 num_kills++;
327 }
328
329 mlog(0, "completed %d messages for node %u\n", num_kills,
330 o2net_num_from_nn(nn));
331}
332
333static int o2net_nsw_completed(struct o2net_node *nn,
334 struct o2net_status_wait *nsw)
335{
336 int completed;
337 spin_lock(&nn->nn_lock);
338 completed = list_empty(&nsw->ns_node_item);
339 spin_unlock(&nn->nn_lock);
340 return completed;
341}
342
343/* ------------------------------------------------------------ */
344
345static void sc_kref_release(struct kref *kref)
346{
347 struct o2net_sock_container *sc = container_of(kref,
348 struct o2net_sock_container, sc_kref);
Jeff Mahoneyb5dd8032006-12-04 14:04:54 +0100349 BUG_ON(timer_pending(&sc->sc_idle_timeout));
350
Zach Brown98211482005-12-15 14:31:23 -0800351 sclog(sc, "releasing\n");
352
353 if (sc->sc_sock) {
354 sock_release(sc->sc_sock);
355 sc->sc_sock = NULL;
356 }
357
358 o2nm_node_put(sc->sc_node);
359 sc->sc_node = NULL;
360
Sunil Mushran2309e9e2008-04-14 10:46:19 -0700361 o2net_debug_del_sc(sc);
Zach Brown98211482005-12-15 14:31:23 -0800362 kfree(sc);
363}
364
365static void sc_put(struct o2net_sock_container *sc)
366{
367 sclog(sc, "put\n");
368 kref_put(&sc->sc_kref, sc_kref_release);
369}
370static void sc_get(struct o2net_sock_container *sc)
371{
372 sclog(sc, "get\n");
373 kref_get(&sc->sc_kref);
374}
375static struct o2net_sock_container *sc_alloc(struct o2nm_node *node)
376{
377 struct o2net_sock_container *sc, *ret = NULL;
378 struct page *page = NULL;
379
380 page = alloc_page(GFP_NOFS);
Robert P. J. Daycd861282006-12-13 00:34:52 -0800381 sc = kzalloc(sizeof(*sc), GFP_NOFS);
Zach Brown98211482005-12-15 14:31:23 -0800382 if (sc == NULL || page == NULL)
383 goto out;
384
385 kref_init(&sc->sc_kref);
386 o2nm_node_get(node);
387 sc->sc_node = node;
388
David Howellsc4028952006-11-22 14:57:56 +0000389 INIT_WORK(&sc->sc_connect_work, o2net_sc_connect_completed);
390 INIT_WORK(&sc->sc_rx_work, o2net_rx_until_empty);
391 INIT_WORK(&sc->sc_shutdown_work, o2net_shutdown_sc);
392 INIT_DELAYED_WORK(&sc->sc_keepalive_work, o2net_sc_send_keep_req);
Zach Brown98211482005-12-15 14:31:23 -0800393
394 init_timer(&sc->sc_idle_timeout);
395 sc->sc_idle_timeout.function = o2net_idle_timer;
396 sc->sc_idle_timeout.data = (unsigned long)sc;
397
398 sclog(sc, "alloced\n");
399
400 ret = sc;
401 sc->sc_page = page;
Sunil Mushran2309e9e2008-04-14 10:46:19 -0700402 o2net_debug_add_sc(sc);
Zach Brown98211482005-12-15 14:31:23 -0800403 sc = NULL;
404 page = NULL;
405
406out:
407 if (page)
408 __free_page(page);
409 kfree(sc);
410
411 return ret;
412}
413
414/* ------------------------------------------------------------ */
415
416static void o2net_sc_queue_work(struct o2net_sock_container *sc,
417 struct work_struct *work)
418{
419 sc_get(sc);
420 if (!queue_work(o2net_wq, work))
421 sc_put(sc);
422}
423static void o2net_sc_queue_delayed_work(struct o2net_sock_container *sc,
David Howellsc4028952006-11-22 14:57:56 +0000424 struct delayed_work *work,
Zach Brown98211482005-12-15 14:31:23 -0800425 int delay)
426{
427 sc_get(sc);
428 if (!queue_delayed_work(o2net_wq, work, delay))
429 sc_put(sc);
430}
431static void o2net_sc_cancel_delayed_work(struct o2net_sock_container *sc,
David Howellsc4028952006-11-22 14:57:56 +0000432 struct delayed_work *work)
Zach Brown98211482005-12-15 14:31:23 -0800433{
434 if (cancel_delayed_work(work))
435 sc_put(sc);
436}
437
Andrew Beekhof828ae6a2006-12-04 14:04:55 +0100438static atomic_t o2net_connected_peers = ATOMIC_INIT(0);
439
440int o2net_num_connected_peers(void)
441{
442 return atomic_read(&o2net_connected_peers);
443}
444
Zach Brown98211482005-12-15 14:31:23 -0800445static void o2net_set_nn_state(struct o2net_node *nn,
446 struct o2net_sock_container *sc,
447 unsigned valid, int err)
448{
449 int was_valid = nn->nn_sc_valid;
450 int was_err = nn->nn_persistent_error;
451 struct o2net_sock_container *old_sc = nn->nn_sc;
452
453 assert_spin_locked(&nn->nn_lock);
454
Andrew Beekhof828ae6a2006-12-04 14:04:55 +0100455 if (old_sc && !sc)
456 atomic_dec(&o2net_connected_peers);
457 else if (!old_sc && sc)
458 atomic_inc(&o2net_connected_peers);
459
Zach Brown98211482005-12-15 14:31:23 -0800460 /* the node num comparison and single connect/accept path should stop
461 * an non-null sc from being overwritten with another */
462 BUG_ON(sc && nn->nn_sc && nn->nn_sc != sc);
463 mlog_bug_on_msg(err && valid, "err %d valid %u\n", err, valid);
464 mlog_bug_on_msg(valid && !sc, "valid %u sc %p\n", valid, sc);
465
Zach Brown98211482005-12-15 14:31:23 -0800466 if (was_valid && !valid && err == 0)
467 err = -ENOTCONN;
468
469 mlog(ML_CONN, "node %u sc: %p -> %p, valid %u -> %u, err %d -> %d\n",
470 o2net_num_from_nn(nn), nn->nn_sc, sc, nn->nn_sc_valid, valid,
471 nn->nn_persistent_error, err);
472
473 nn->nn_sc = sc;
474 nn->nn_sc_valid = valid ? 1 : 0;
475 nn->nn_persistent_error = err;
476
477 /* mirrors o2net_tx_can_proceed() */
478 if (nn->nn_persistent_error || nn->nn_sc_valid)
479 wake_up(&nn->nn_sc_wq);
480
481 if (!was_err && nn->nn_persistent_error) {
482 o2quo_conn_err(o2net_num_from_nn(nn));
483 queue_delayed_work(o2net_wq, &nn->nn_still_up,
484 msecs_to_jiffies(O2NET_QUORUM_DELAY_MS));
485 }
486
487 if (was_valid && !valid) {
Sunil Mushran781ee3e2006-04-27 16:41:31 -0700488 printk(KERN_INFO "o2net: no longer connected to "
489 SC_NODEF_FMT "\n", SC_NODEF_ARGS(old_sc));
Zach Brown98211482005-12-15 14:31:23 -0800490 o2net_complete_nodes_nsw(nn);
491 }
492
493 if (!was_valid && valid) {
494 o2quo_conn_up(o2net_num_from_nn(nn));
Zach Brown98211482005-12-15 14:31:23 -0800495 cancel_delayed_work(&nn->nn_connect_expired);
Sunil Mushran781ee3e2006-04-27 16:41:31 -0700496 printk(KERN_INFO "o2net: %s " SC_NODEF_FMT "\n",
497 o2nm_this_node() > sc->sc_node->nd_num ?
498 "connected to" : "accepted connection from",
499 SC_NODEF_ARGS(sc));
Zach Brown98211482005-12-15 14:31:23 -0800500 }
501
502 /* trigger the connecting worker func as long as we're not valid,
503 * it will back off if it shouldn't connect. This can be called
504 * from node config teardown and so needs to be careful about
505 * the work queue actually being up. */
506 if (!valid && o2net_wq) {
507 unsigned long delay;
508 /* delay if we're withing a RECONNECT_DELAY of the
509 * last attempt */
510 delay = (nn->nn_last_connect_attempt +
Jeff Mahoney409753b2008-03-28 16:44:13 -0700511 msecs_to_jiffies(o2net_reconnect_delay()))
Zach Brown98211482005-12-15 14:31:23 -0800512 - jiffies;
Jeff Mahoney409753b2008-03-28 16:44:13 -0700513 if (delay > msecs_to_jiffies(o2net_reconnect_delay()))
Zach Brown98211482005-12-15 14:31:23 -0800514 delay = 0;
515 mlog(ML_CONN, "queueing conn attempt in %lu jiffies\n", delay);
516 queue_delayed_work(o2net_wq, &nn->nn_connect_work, delay);
Tao Ma5cc3bf22008-03-05 15:50:12 +0800517
518 /*
519 * Delay the expired work after idle timeout.
520 *
521 * We might have lots of failed connection attempts that run
522 * through here but we only cancel the connect_expired work when
523 * a connection attempt succeeds. So only the first enqueue of
524 * the connect_expired work will do anything. The rest will see
525 * that it's already queued and do nothing.
526 */
Jeff Mahoney409753b2008-03-28 16:44:13 -0700527 delay += msecs_to_jiffies(o2net_idle_timeout());
Tao Ma5cc3bf22008-03-05 15:50:12 +0800528 queue_delayed_work(o2net_wq, &nn->nn_connect_expired, delay);
Zach Brown98211482005-12-15 14:31:23 -0800529 }
530
531 /* keep track of the nn's sc ref for the caller */
532 if ((old_sc == NULL) && sc)
533 sc_get(sc);
534 if (old_sc && (old_sc != sc)) {
535 o2net_sc_queue_work(old_sc, &old_sc->sc_shutdown_work);
536 sc_put(old_sc);
537 }
538}
539
540/* see o2net_register_callbacks() */
541static void o2net_data_ready(struct sock *sk, int bytes)
542{
543 void (*ready)(struct sock *sk, int bytes);
544
545 read_lock(&sk->sk_callback_lock);
546 if (sk->sk_user_data) {
547 struct o2net_sock_container *sc = sk->sk_user_data;
548 sclog(sc, "data_ready hit\n");
549 do_gettimeofday(&sc->sc_tv_data_ready);
550 o2net_sc_queue_work(sc, &sc->sc_rx_work);
551 ready = sc->sc_data_ready;
552 } else {
553 ready = sk->sk_data_ready;
554 }
555 read_unlock(&sk->sk_callback_lock);
556
557 ready(sk, bytes);
558}
559
560/* see o2net_register_callbacks() */
561static void o2net_state_change(struct sock *sk)
562{
563 void (*state_change)(struct sock *sk);
564 struct o2net_sock_container *sc;
565
566 read_lock(&sk->sk_callback_lock);
567 sc = sk->sk_user_data;
568 if (sc == NULL) {
569 state_change = sk->sk_state_change;
570 goto out;
571 }
572
573 sclog(sc, "state_change to %d\n", sk->sk_state);
574
575 state_change = sc->sc_state_change;
576
577 switch(sk->sk_state) {
578 /* ignore connecting sockets as they make progress */
579 case TCP_SYN_SENT:
580 case TCP_SYN_RECV:
581 break;
582 case TCP_ESTABLISHED:
583 o2net_sc_queue_work(sc, &sc->sc_connect_work);
584 break;
585 default:
586 o2net_sc_queue_work(sc, &sc->sc_shutdown_work);
587 break;
588 }
589out:
590 read_unlock(&sk->sk_callback_lock);
591 state_change(sk);
592}
593
594/*
595 * we register callbacks so we can queue work on events before calling
596 * the original callbacks. our callbacks our careful to test user_data
597 * to discover when they've reaced with o2net_unregister_callbacks().
598 */
599static void o2net_register_callbacks(struct sock *sk,
600 struct o2net_sock_container *sc)
601{
602 write_lock_bh(&sk->sk_callback_lock);
603
604 /* accepted sockets inherit the old listen socket data ready */
605 if (sk->sk_data_ready == o2net_listen_data_ready) {
606 sk->sk_data_ready = sk->sk_user_data;
607 sk->sk_user_data = NULL;
608 }
609
610 BUG_ON(sk->sk_user_data != NULL);
611 sk->sk_user_data = sc;
612 sc_get(sc);
613
614 sc->sc_data_ready = sk->sk_data_ready;
615 sc->sc_state_change = sk->sk_state_change;
616 sk->sk_data_ready = o2net_data_ready;
617 sk->sk_state_change = o2net_state_change;
618
Zhen Wei925037b2007-01-23 17:19:59 -0800619 mutex_init(&sc->sc_send_lock);
620
Zach Brown98211482005-12-15 14:31:23 -0800621 write_unlock_bh(&sk->sk_callback_lock);
622}
623
624static int o2net_unregister_callbacks(struct sock *sk,
625 struct o2net_sock_container *sc)
626{
627 int ret = 0;
628
629 write_lock_bh(&sk->sk_callback_lock);
630 if (sk->sk_user_data == sc) {
631 ret = 1;
632 sk->sk_user_data = NULL;
633 sk->sk_data_ready = sc->sc_data_ready;
634 sk->sk_state_change = sc->sc_state_change;
635 }
636 write_unlock_bh(&sk->sk_callback_lock);
637
638 return ret;
639}
640
641/*
642 * this is a little helper that is called by callers who have seen a problem
643 * with an sc and want to detach it from the nn if someone already hasn't beat
644 * them to it. if an error is given then the shutdown will be persistent
645 * and pending transmits will be canceled.
646 */
647static void o2net_ensure_shutdown(struct o2net_node *nn,
648 struct o2net_sock_container *sc,
649 int err)
650{
651 spin_lock(&nn->nn_lock);
652 if (nn->nn_sc == sc)
653 o2net_set_nn_state(nn, NULL, 0, err);
654 spin_unlock(&nn->nn_lock);
655}
656
657/*
658 * This work queue function performs the blocking parts of socket shutdown. A
659 * few paths lead here. set_nn_state will trigger this callback if it sees an
660 * sc detached from the nn. state_change will also trigger this callback
661 * directly when it sees errors. In that case we need to call set_nn_state
662 * ourselves as state_change couldn't get the nn_lock and call set_nn_state
663 * itself.
664 */
David Howellsc4028952006-11-22 14:57:56 +0000665static void o2net_shutdown_sc(struct work_struct *work)
Zach Brown98211482005-12-15 14:31:23 -0800666{
David Howellsc4028952006-11-22 14:57:56 +0000667 struct o2net_sock_container *sc =
668 container_of(work, struct o2net_sock_container,
669 sc_shutdown_work);
Zach Brown98211482005-12-15 14:31:23 -0800670 struct o2net_node *nn = o2net_nn_from_num(sc->sc_node->nd_num);
671
672 sclog(sc, "shutting down\n");
673
674 /* drop the callbacks ref and call shutdown only once */
675 if (o2net_unregister_callbacks(sc->sc_sock->sk, sc)) {
676 /* we shouldn't flush as we're in the thread, the
677 * races with pending sc work structs are harmless */
678 del_timer_sync(&sc->sc_idle_timeout);
679 o2net_sc_cancel_delayed_work(sc, &sc->sc_keepalive_work);
680 sc_put(sc);
Trond Myklebust91cf45f2007-11-12 18:10:39 -0800681 kernel_sock_shutdown(sc->sc_sock, SHUT_RDWR);
Zach Brown98211482005-12-15 14:31:23 -0800682 }
683
684 /* not fatal so failed connects before the other guy has our
685 * heartbeat can be retried */
686 o2net_ensure_shutdown(nn, sc, 0);
687 sc_put(sc);
688}
689
690/* ------------------------------------------------------------ */
691
692static int o2net_handler_cmp(struct o2net_msg_handler *nmh, u32 msg_type,
693 u32 key)
694{
695 int ret = memcmp(&nmh->nh_key, &key, sizeof(key));
696
697 if (ret == 0)
698 ret = memcmp(&nmh->nh_msg_type, &msg_type, sizeof(msg_type));
699
700 return ret;
701}
702
703static struct o2net_msg_handler *
704o2net_handler_tree_lookup(u32 msg_type, u32 key, struct rb_node ***ret_p,
705 struct rb_node **ret_parent)
706{
707 struct rb_node **p = &o2net_handler_tree.rb_node;
708 struct rb_node *parent = NULL;
709 struct o2net_msg_handler *nmh, *ret = NULL;
710 int cmp;
711
712 while (*p) {
713 parent = *p;
714 nmh = rb_entry(parent, struct o2net_msg_handler, nh_node);
715 cmp = o2net_handler_cmp(nmh, msg_type, key);
716
717 if (cmp < 0)
718 p = &(*p)->rb_left;
719 else if (cmp > 0)
720 p = &(*p)->rb_right;
721 else {
722 ret = nmh;
723 break;
724 }
725 }
726
727 if (ret_p != NULL)
728 *ret_p = p;
729 if (ret_parent != NULL)
730 *ret_parent = parent;
731
732 return ret;
733}
734
735static void o2net_handler_kref_release(struct kref *kref)
736{
737 struct o2net_msg_handler *nmh;
738 nmh = container_of(kref, struct o2net_msg_handler, nh_kref);
739
740 kfree(nmh);
741}
742
743static void o2net_handler_put(struct o2net_msg_handler *nmh)
744{
745 kref_put(&nmh->nh_kref, o2net_handler_kref_release);
746}
747
748/* max_len is protection for the handler func. incoming messages won't
749 * be given to the handler if their payload is longer than the max. */
750int o2net_register_handler(u32 msg_type, u32 key, u32 max_len,
751 o2net_msg_handler_func *func, void *data,
Kurt Hackeld74c9802007-01-17 17:04:25 -0800752 o2net_post_msg_handler_func *post_func,
Zach Brown98211482005-12-15 14:31:23 -0800753 struct list_head *unreg_list)
754{
755 struct o2net_msg_handler *nmh = NULL;
756 struct rb_node **p, *parent;
757 int ret = 0;
758
759 if (max_len > O2NET_MAX_PAYLOAD_BYTES) {
760 mlog(0, "max_len for message handler out of range: %u\n",
761 max_len);
762 ret = -EINVAL;
763 goto out;
764 }
765
766 if (!msg_type) {
767 mlog(0, "no message type provided: %u, %p\n", msg_type, func);
768 ret = -EINVAL;
769 goto out;
770
771 }
772 if (!func) {
773 mlog(0, "no message handler provided: %u, %p\n",
774 msg_type, func);
775 ret = -EINVAL;
776 goto out;
777 }
778
Robert P. J. Daycd861282006-12-13 00:34:52 -0800779 nmh = kzalloc(sizeof(struct o2net_msg_handler), GFP_NOFS);
Zach Brown98211482005-12-15 14:31:23 -0800780 if (nmh == NULL) {
781 ret = -ENOMEM;
782 goto out;
783 }
784
785 nmh->nh_func = func;
786 nmh->nh_func_data = data;
Kurt Hackeld74c9802007-01-17 17:04:25 -0800787 nmh->nh_post_func = post_func;
Zach Brown98211482005-12-15 14:31:23 -0800788 nmh->nh_msg_type = msg_type;
789 nmh->nh_max_len = max_len;
790 nmh->nh_key = key;
791 /* the tree and list get this ref.. they're both removed in
792 * unregister when this ref is dropped */
793 kref_init(&nmh->nh_kref);
794 INIT_LIST_HEAD(&nmh->nh_unregister_item);
795
796 write_lock(&o2net_handler_lock);
797 if (o2net_handler_tree_lookup(msg_type, key, &p, &parent))
798 ret = -EEXIST;
799 else {
800 rb_link_node(&nmh->nh_node, parent, p);
801 rb_insert_color(&nmh->nh_node, &o2net_handler_tree);
802 list_add_tail(&nmh->nh_unregister_item, unreg_list);
803
804 mlog(ML_TCP, "registered handler func %p type %u key %08x\n",
805 func, msg_type, key);
806 /* we've had some trouble with handlers seemingly vanishing. */
807 mlog_bug_on_msg(o2net_handler_tree_lookup(msg_type, key, &p,
808 &parent) == NULL,
809 "couldn't find handler we *just* registerd "
810 "for type %u key %08x\n", msg_type, key);
811 }
812 write_unlock(&o2net_handler_lock);
813 if (ret)
814 goto out;
815
816out:
817 if (ret)
818 kfree(nmh);
819
820 return ret;
821}
822EXPORT_SYMBOL_GPL(o2net_register_handler);
823
824void o2net_unregister_handler_list(struct list_head *list)
825{
Christoph Hellwig800deef2007-05-17 16:03:13 +0200826 struct o2net_msg_handler *nmh, *n;
Zach Brown98211482005-12-15 14:31:23 -0800827
828 write_lock(&o2net_handler_lock);
Christoph Hellwig800deef2007-05-17 16:03:13 +0200829 list_for_each_entry_safe(nmh, n, list, nh_unregister_item) {
Zach Brown98211482005-12-15 14:31:23 -0800830 mlog(ML_TCP, "unregistering handler func %p type %u key %08x\n",
831 nmh->nh_func, nmh->nh_msg_type, nmh->nh_key);
832 rb_erase(&nmh->nh_node, &o2net_handler_tree);
833 list_del_init(&nmh->nh_unregister_item);
834 kref_put(&nmh->nh_kref, o2net_handler_kref_release);
835 }
836 write_unlock(&o2net_handler_lock);
837}
838EXPORT_SYMBOL_GPL(o2net_unregister_handler_list);
839
840static struct o2net_msg_handler *o2net_handler_get(u32 msg_type, u32 key)
841{
842 struct o2net_msg_handler *nmh;
843
844 read_lock(&o2net_handler_lock);
845 nmh = o2net_handler_tree_lookup(msg_type, key, NULL, NULL);
846 if (nmh)
847 kref_get(&nmh->nh_kref);
848 read_unlock(&o2net_handler_lock);
849
850 return nmh;
851}
852
853/* ------------------------------------------------------------ */
854
855static int o2net_recv_tcp_msg(struct socket *sock, void *data, size_t len)
856{
857 int ret;
858 mm_segment_t oldfs;
859 struct kvec vec = {
860 .iov_len = len,
861 .iov_base = data,
862 };
863 struct msghdr msg = {
864 .msg_iovlen = 1,
865 .msg_iov = (struct iovec *)&vec,
866 .msg_flags = MSG_DONTWAIT,
867 };
868
869 oldfs = get_fs();
870 set_fs(get_ds());
871 ret = sock_recvmsg(sock, &msg, len, msg.msg_flags);
872 set_fs(oldfs);
873
874 return ret;
875}
876
877static int o2net_send_tcp_msg(struct socket *sock, struct kvec *vec,
878 size_t veclen, size_t total)
879{
880 int ret;
881 mm_segment_t oldfs;
882 struct msghdr msg = {
883 .msg_iov = (struct iovec *)vec,
884 .msg_iovlen = veclen,
885 };
886
887 if (sock == NULL) {
888 ret = -EINVAL;
889 goto out;
890 }
891
892 oldfs = get_fs();
893 set_fs(get_ds());
894 ret = sock_sendmsg(sock, &msg, total);
895 set_fs(oldfs);
896 if (ret != total) {
897 mlog(ML_ERROR, "sendmsg returned %d instead of %zu\n", ret,
898 total);
899 if (ret >= 0)
900 ret = -EPIPE; /* should be smarter, I bet */
901 goto out;
902 }
903
904 ret = 0;
905out:
906 if (ret < 0)
907 mlog(0, "returning error: %d\n", ret);
908 return ret;
909}
910
911static void o2net_sendpage(struct o2net_sock_container *sc,
912 void *kmalloced_virt,
913 size_t size)
914{
915 struct o2net_node *nn = o2net_nn_from_num(sc->sc_node->nd_num);
916 ssize_t ret;
917
Sunil Mushrance17204a2007-07-30 11:02:50 -0700918 while (1) {
919 mutex_lock(&sc->sc_send_lock);
920 ret = sc->sc_sock->ops->sendpage(sc->sc_sock,
921 virt_to_page(kmalloced_virt),
922 (long)kmalloced_virt & ~PAGE_MASK,
923 size, MSG_DONTWAIT);
924 mutex_unlock(&sc->sc_send_lock);
925 if (ret == size)
926 break;
927 if (ret == (ssize_t)-EAGAIN) {
928 mlog(0, "sendpage of size %zu to " SC_NODEF_FMT
929 " returned EAGAIN\n", size, SC_NODEF_ARGS(sc));
930 cond_resched();
931 continue;
932 }
Zach Brown98211482005-12-15 14:31:23 -0800933 mlog(ML_ERROR, "sendpage of size %zu to " SC_NODEF_FMT
934 " failed with %zd\n", size, SC_NODEF_ARGS(sc), ret);
935 o2net_ensure_shutdown(nn, sc, 0);
Sunil Mushrance17204a2007-07-30 11:02:50 -0700936 break;
Zach Brown98211482005-12-15 14:31:23 -0800937 }
938}
939
940static void o2net_init_msg(struct o2net_msg *msg, u16 data_len, u16 msg_type, u32 key)
941{
942 memset(msg, 0, sizeof(struct o2net_msg));
943 msg->magic = cpu_to_be16(O2NET_MSG_MAGIC);
944 msg->data_len = cpu_to_be16(data_len);
945 msg->msg_type = cpu_to_be16(msg_type);
946 msg->sys_status = cpu_to_be32(O2NET_ERR_NONE);
947 msg->status = 0;
948 msg->key = cpu_to_be32(key);
949}
950
951static int o2net_tx_can_proceed(struct o2net_node *nn,
952 struct o2net_sock_container **sc_ret,
953 int *error)
954{
955 int ret = 0;
956
957 spin_lock(&nn->nn_lock);
958 if (nn->nn_persistent_error) {
959 ret = 1;
960 *sc_ret = NULL;
961 *error = nn->nn_persistent_error;
962 } else if (nn->nn_sc_valid) {
963 kref_get(&nn->nn_sc->sc_kref);
964
965 ret = 1;
966 *sc_ret = nn->nn_sc;
967 *error = 0;
968 }
969 spin_unlock(&nn->nn_lock);
970
971 return ret;
972}
973
974int o2net_send_message_vec(u32 msg_type, u32 key, struct kvec *caller_vec,
975 size_t caller_veclen, u8 target_node, int *status)
976{
977 int ret, error = 0;
978 struct o2net_msg *msg = NULL;
979 size_t veclen, caller_bytes = 0;
980 struct kvec *vec = NULL;
981 struct o2net_sock_container *sc = NULL;
982 struct o2net_node *nn = o2net_nn_from_num(target_node);
983 struct o2net_status_wait nsw = {
984 .ns_node_item = LIST_HEAD_INIT(nsw.ns_node_item),
985 };
Sunil Mushran2309e9e2008-04-14 10:46:19 -0700986 struct o2net_send_tracking nst;
987
988 o2net_init_nst(&nst, msg_type, key, current, target_node);
Zach Brown98211482005-12-15 14:31:23 -0800989
990 if (o2net_wq == NULL) {
991 mlog(0, "attempt to tx without o2netd running\n");
992 ret = -ESRCH;
993 goto out;
994 }
995
996 if (caller_veclen == 0) {
997 mlog(0, "bad kvec array length\n");
998 ret = -EINVAL;
999 goto out;
1000 }
1001
1002 caller_bytes = iov_length((struct iovec *)caller_vec, caller_veclen);
1003 if (caller_bytes > O2NET_MAX_PAYLOAD_BYTES) {
1004 mlog(0, "total payload len %zu too large\n", caller_bytes);
1005 ret = -EINVAL;
1006 goto out;
1007 }
1008
1009 if (target_node == o2nm_this_node()) {
1010 ret = -ELOOP;
1011 goto out;
1012 }
1013
Sunil Mushran2309e9e2008-04-14 10:46:19 -07001014 o2net_debug_add_nst(&nst);
1015
1016 o2net_set_nst_sock_time(&nst);
1017
Zach Brown98211482005-12-15 14:31:23 -08001018 ret = wait_event_interruptible(nn->nn_sc_wq,
1019 o2net_tx_can_proceed(nn, &sc, &error));
1020 if (!ret && error)
1021 ret = error;
1022 if (ret)
1023 goto out;
1024
Sunil Mushran2309e9e2008-04-14 10:46:19 -07001025 o2net_set_nst_sock_container(&nst, sc);
1026
Zach Brown98211482005-12-15 14:31:23 -08001027 veclen = caller_veclen + 1;
1028 vec = kmalloc(sizeof(struct kvec) * veclen, GFP_ATOMIC);
1029 if (vec == NULL) {
1030 mlog(0, "failed to %zu element kvec!\n", veclen);
1031 ret = -ENOMEM;
1032 goto out;
1033 }
1034
1035 msg = kmalloc(sizeof(struct o2net_msg), GFP_ATOMIC);
1036 if (!msg) {
1037 mlog(0, "failed to allocate a o2net_msg!\n");
1038 ret = -ENOMEM;
1039 goto out;
1040 }
1041
1042 o2net_init_msg(msg, caller_bytes, msg_type, key);
1043
1044 vec[0].iov_len = sizeof(struct o2net_msg);
1045 vec[0].iov_base = msg;
1046 memcpy(&vec[1], caller_vec, caller_veclen * sizeof(struct kvec));
1047
1048 ret = o2net_prep_nsw(nn, &nsw);
1049 if (ret)
1050 goto out;
1051
1052 msg->msg_num = cpu_to_be32(nsw.ns_id);
Sunil Mushran2309e9e2008-04-14 10:46:19 -07001053 o2net_set_nst_msg_id(&nst, nsw.ns_id);
1054
1055 o2net_set_nst_send_time(&nst);
Zach Brown98211482005-12-15 14:31:23 -08001056
1057 /* finally, convert the message header to network byte-order
1058 * and send */
Zhen Wei925037b2007-01-23 17:19:59 -08001059 mutex_lock(&sc->sc_send_lock);
Zach Brown98211482005-12-15 14:31:23 -08001060 ret = o2net_send_tcp_msg(sc->sc_sock, vec, veclen,
1061 sizeof(struct o2net_msg) + caller_bytes);
Zhen Wei925037b2007-01-23 17:19:59 -08001062 mutex_unlock(&sc->sc_send_lock);
Zach Brown98211482005-12-15 14:31:23 -08001063 msglog(msg, "sending returned %d\n", ret);
1064 if (ret < 0) {
1065 mlog(0, "error returned from o2net_send_tcp_msg=%d\n", ret);
1066 goto out;
1067 }
1068
1069 /* wait on other node's handler */
Sunil Mushran2309e9e2008-04-14 10:46:19 -07001070 o2net_set_nst_status_time(&nst);
Zach Brown98211482005-12-15 14:31:23 -08001071 wait_event(nsw.ns_wq, o2net_nsw_completed(nn, &nsw));
1072
1073 /* Note that we avoid overwriting the callers status return
1074 * variable if a system error was reported on the other
1075 * side. Callers beware. */
1076 ret = o2net_sys_err_to_errno(nsw.ns_sys_status);
1077 if (status && !ret)
1078 *status = nsw.ns_status;
1079
1080 mlog(0, "woken, returning system status %d, user status %d\n",
1081 ret, nsw.ns_status);
1082out:
Sunil Mushran2309e9e2008-04-14 10:46:19 -07001083 o2net_debug_del_nst(&nst); /* must be before dropping sc and node */
Zach Brown98211482005-12-15 14:31:23 -08001084 if (sc)
1085 sc_put(sc);
1086 if (vec)
1087 kfree(vec);
1088 if (msg)
1089 kfree(msg);
1090 o2net_complete_nsw(nn, &nsw, 0, 0, 0);
1091 return ret;
1092}
1093EXPORT_SYMBOL_GPL(o2net_send_message_vec);
1094
1095int o2net_send_message(u32 msg_type, u32 key, void *data, u32 len,
1096 u8 target_node, int *status)
1097{
1098 struct kvec vec = {
1099 .iov_base = data,
1100 .iov_len = len,
1101 };
1102 return o2net_send_message_vec(msg_type, key, &vec, 1,
1103 target_node, status);
1104}
1105EXPORT_SYMBOL_GPL(o2net_send_message);
1106
1107static int o2net_send_status_magic(struct socket *sock, struct o2net_msg *hdr,
1108 enum o2net_system_error syserr, int err)
1109{
1110 struct kvec vec = {
1111 .iov_base = hdr,
1112 .iov_len = sizeof(struct o2net_msg),
1113 };
1114
1115 BUG_ON(syserr >= O2NET_ERR_MAX);
1116
1117 /* leave other fields intact from the incoming message, msg_num
1118 * in particular */
1119 hdr->sys_status = cpu_to_be32(syserr);
1120 hdr->status = cpu_to_be32(err);
1121 hdr->magic = cpu_to_be16(O2NET_MSG_STATUS_MAGIC); // twiddle the magic
1122 hdr->data_len = 0;
1123
1124 msglog(hdr, "about to send status magic %d\n", err);
1125 /* hdr has been in host byteorder this whole time */
1126 return o2net_send_tcp_msg(sock, &vec, 1, sizeof(struct o2net_msg));
1127}
1128
1129/* this returns -errno if the header was unknown or too large, etc.
1130 * after this is called the buffer us reused for the next message */
1131static int o2net_process_message(struct o2net_sock_container *sc,
1132 struct o2net_msg *hdr)
1133{
1134 struct o2net_node *nn = o2net_nn_from_num(sc->sc_node->nd_num);
1135 int ret = 0, handler_status;
1136 enum o2net_system_error syserr;
1137 struct o2net_msg_handler *nmh = NULL;
Kurt Hackeld74c9802007-01-17 17:04:25 -08001138 void *ret_data = NULL;
Zach Brown98211482005-12-15 14:31:23 -08001139
1140 msglog(hdr, "processing message\n");
1141
1142 o2net_sc_postpone_idle(sc);
1143
1144 switch(be16_to_cpu(hdr->magic)) {
1145 case O2NET_MSG_STATUS_MAGIC:
1146 /* special type for returning message status */
1147 o2net_complete_nsw(nn, NULL,
1148 be32_to_cpu(hdr->msg_num),
1149 be32_to_cpu(hdr->sys_status),
1150 be32_to_cpu(hdr->status));
1151 goto out;
1152 case O2NET_MSG_KEEP_REQ_MAGIC:
1153 o2net_sendpage(sc, o2net_keep_resp,
1154 sizeof(*o2net_keep_resp));
1155 goto out;
1156 case O2NET_MSG_KEEP_RESP_MAGIC:
1157 goto out;
1158 case O2NET_MSG_MAGIC:
1159 break;
1160 default:
1161 msglog(hdr, "bad magic\n");
1162 ret = -EINVAL;
1163 goto out;
1164 break;
1165 }
1166
1167 /* find a handler for it */
1168 handler_status = 0;
1169 nmh = o2net_handler_get(be16_to_cpu(hdr->msg_type),
1170 be32_to_cpu(hdr->key));
1171 if (!nmh) {
1172 mlog(ML_TCP, "couldn't find handler for type %u key %08x\n",
1173 be16_to_cpu(hdr->msg_type), be32_to_cpu(hdr->key));
1174 syserr = O2NET_ERR_NO_HNDLR;
1175 goto out_respond;
1176 }
1177
1178 syserr = O2NET_ERR_NONE;
1179
1180 if (be16_to_cpu(hdr->data_len) > nmh->nh_max_len)
1181 syserr = O2NET_ERR_OVERFLOW;
1182
1183 if (syserr != O2NET_ERR_NONE)
1184 goto out_respond;
1185
1186 do_gettimeofday(&sc->sc_tv_func_start);
1187 sc->sc_msg_key = be32_to_cpu(hdr->key);
1188 sc->sc_msg_type = be16_to_cpu(hdr->msg_type);
1189 handler_status = (nmh->nh_func)(hdr, sizeof(struct o2net_msg) +
1190 be16_to_cpu(hdr->data_len),
Kurt Hackeld74c9802007-01-17 17:04:25 -08001191 nmh->nh_func_data, &ret_data);
Zach Brown98211482005-12-15 14:31:23 -08001192 do_gettimeofday(&sc->sc_tv_func_stop);
1193
1194out_respond:
1195 /* this destroys the hdr, so don't use it after this */
Zhen Wei925037b2007-01-23 17:19:59 -08001196 mutex_lock(&sc->sc_send_lock);
Zach Brown98211482005-12-15 14:31:23 -08001197 ret = o2net_send_status_magic(sc->sc_sock, hdr, syserr,
1198 handler_status);
Zhen Wei925037b2007-01-23 17:19:59 -08001199 mutex_unlock(&sc->sc_send_lock);
Zach Brown98211482005-12-15 14:31:23 -08001200 hdr = NULL;
1201 mlog(0, "sending handler status %d, syserr %d returned %d\n",
1202 handler_status, syserr, ret);
1203
Kurt Hackeld74c9802007-01-17 17:04:25 -08001204 if (nmh) {
1205 BUG_ON(ret_data != NULL && nmh->nh_post_func == NULL);
1206 if (nmh->nh_post_func)
1207 (nmh->nh_post_func)(handler_status, nmh->nh_func_data,
1208 ret_data);
1209 }
1210
Zach Brown98211482005-12-15 14:31:23 -08001211out:
1212 if (nmh)
1213 o2net_handler_put(nmh);
1214 return ret;
1215}
1216
1217static int o2net_check_handshake(struct o2net_sock_container *sc)
1218{
1219 struct o2net_handshake *hand = page_address(sc->sc_page);
1220 struct o2net_node *nn = o2net_nn_from_num(sc->sc_node->nd_num);
1221
1222 if (hand->protocol_version != cpu_to_be64(O2NET_PROTOCOL_VERSION)) {
1223 mlog(ML_NOTICE, SC_NODEF_FMT " advertised net protocol "
1224 "version %llu but %llu is required, disconnecting\n",
1225 SC_NODEF_ARGS(sc),
1226 (unsigned long long)be64_to_cpu(hand->protocol_version),
1227 O2NET_PROTOCOL_VERSION);
1228
1229 /* don't bother reconnecting if its the wrong version. */
1230 o2net_ensure_shutdown(nn, sc, -ENOTCONN);
1231 return -1;
1232 }
1233
Andrew Beekhof828ae6a2006-12-04 14:04:55 +01001234 /*
1235 * Ensure timeouts are consistent with other nodes, otherwise
1236 * we can end up with one node thinking that the other must be down,
1237 * but isn't. This can ultimately cause corruption.
1238 */
1239 if (be32_to_cpu(hand->o2net_idle_timeout_ms) !=
Jeff Mahoney409753b2008-03-28 16:44:13 -07001240 o2net_idle_timeout()) {
Andrew Beekhof828ae6a2006-12-04 14:04:55 +01001241 mlog(ML_NOTICE, SC_NODEF_FMT " uses a network idle timeout of "
1242 "%u ms, but we use %u ms locally. disconnecting\n",
1243 SC_NODEF_ARGS(sc),
1244 be32_to_cpu(hand->o2net_idle_timeout_ms),
Jeff Mahoney409753b2008-03-28 16:44:13 -07001245 o2net_idle_timeout());
Andrew Beekhof828ae6a2006-12-04 14:04:55 +01001246 o2net_ensure_shutdown(nn, sc, -ENOTCONN);
1247 return -1;
1248 }
1249
1250 if (be32_to_cpu(hand->o2net_keepalive_delay_ms) !=
Jeff Mahoney409753b2008-03-28 16:44:13 -07001251 o2net_keepalive_delay()) {
Andrew Beekhof828ae6a2006-12-04 14:04:55 +01001252 mlog(ML_NOTICE, SC_NODEF_FMT " uses a keepalive delay of "
1253 "%u ms, but we use %u ms locally. disconnecting\n",
1254 SC_NODEF_ARGS(sc),
1255 be32_to_cpu(hand->o2net_keepalive_delay_ms),
Jeff Mahoney409753b2008-03-28 16:44:13 -07001256 o2net_keepalive_delay());
Andrew Beekhof828ae6a2006-12-04 14:04:55 +01001257 o2net_ensure_shutdown(nn, sc, -ENOTCONN);
1258 return -1;
1259 }
1260
1261 if (be32_to_cpu(hand->o2hb_heartbeat_timeout_ms) !=
1262 O2HB_MAX_WRITE_TIMEOUT_MS) {
1263 mlog(ML_NOTICE, SC_NODEF_FMT " uses a heartbeat timeout of "
1264 "%u ms, but we use %u ms locally. disconnecting\n",
1265 SC_NODEF_ARGS(sc),
1266 be32_to_cpu(hand->o2hb_heartbeat_timeout_ms),
1267 O2HB_MAX_WRITE_TIMEOUT_MS);
1268 o2net_ensure_shutdown(nn, sc, -ENOTCONN);
1269 return -1;
1270 }
1271
Zach Brown98211482005-12-15 14:31:23 -08001272 sc->sc_handshake_ok = 1;
1273
1274 spin_lock(&nn->nn_lock);
1275 /* set valid and queue the idle timers only if it hasn't been
1276 * shut down already */
1277 if (nn->nn_sc == sc) {
Jeff Mahoneyb5dd8032006-12-04 14:04:54 +01001278 o2net_sc_reset_idle_timer(sc);
Tao Ma5cc3bf22008-03-05 15:50:12 +08001279 atomic_set(&nn->nn_timeout, 0);
Zach Brown98211482005-12-15 14:31:23 -08001280 o2net_set_nn_state(nn, sc, 1, 0);
1281 }
1282 spin_unlock(&nn->nn_lock);
1283
1284 /* shift everything up as though it wasn't there */
1285 sc->sc_page_off -= sizeof(struct o2net_handshake);
1286 if (sc->sc_page_off)
1287 memmove(hand, hand + 1, sc->sc_page_off);
1288
1289 return 0;
1290}
1291
1292/* this demuxes the queued rx bytes into header or payload bits and calls
1293 * handlers as each full message is read off the socket. it returns -error,
1294 * == 0 eof, or > 0 for progress made.*/
1295static int o2net_advance_rx(struct o2net_sock_container *sc)
1296{
1297 struct o2net_msg *hdr;
1298 int ret = 0;
1299 void *data;
1300 size_t datalen;
1301
1302 sclog(sc, "receiving\n");
1303 do_gettimeofday(&sc->sc_tv_advance_start);
1304
Andrew Beekhof828ae6a2006-12-04 14:04:55 +01001305 if (unlikely(sc->sc_handshake_ok == 0)) {
1306 if(sc->sc_page_off < sizeof(struct o2net_handshake)) {
1307 data = page_address(sc->sc_page) + sc->sc_page_off;
1308 datalen = sizeof(struct o2net_handshake) - sc->sc_page_off;
1309 ret = o2net_recv_tcp_msg(sc->sc_sock, data, datalen);
1310 if (ret > 0)
1311 sc->sc_page_off += ret;
1312 }
1313
1314 if (sc->sc_page_off == sizeof(struct o2net_handshake)) {
1315 o2net_check_handshake(sc);
1316 if (unlikely(sc->sc_handshake_ok == 0))
1317 ret = -EPROTO;
1318 }
1319 goto out;
1320 }
1321
Zach Brown98211482005-12-15 14:31:23 -08001322 /* do we need more header? */
1323 if (sc->sc_page_off < sizeof(struct o2net_msg)) {
1324 data = page_address(sc->sc_page) + sc->sc_page_off;
1325 datalen = sizeof(struct o2net_msg) - sc->sc_page_off;
1326 ret = o2net_recv_tcp_msg(sc->sc_sock, data, datalen);
1327 if (ret > 0) {
1328 sc->sc_page_off += ret;
Zach Brown98211482005-12-15 14:31:23 -08001329 /* only swab incoming here.. we can
1330 * only get here once as we cross from
1331 * being under to over */
1332 if (sc->sc_page_off == sizeof(struct o2net_msg)) {
1333 hdr = page_address(sc->sc_page);
1334 if (be16_to_cpu(hdr->data_len) >
1335 O2NET_MAX_PAYLOAD_BYTES)
1336 ret = -EOVERFLOW;
1337 }
1338 }
1339 if (ret <= 0)
1340 goto out;
1341 }
1342
1343 if (sc->sc_page_off < sizeof(struct o2net_msg)) {
1344 /* oof, still don't have a header */
1345 goto out;
1346 }
1347
1348 /* this was swabbed above when we first read it */
1349 hdr = page_address(sc->sc_page);
1350
1351 msglog(hdr, "at page_off %zu\n", sc->sc_page_off);
1352
1353 /* do we need more payload? */
1354 if (sc->sc_page_off - sizeof(struct o2net_msg) < be16_to_cpu(hdr->data_len)) {
1355 /* need more payload */
1356 data = page_address(sc->sc_page) + sc->sc_page_off;
1357 datalen = (sizeof(struct o2net_msg) + be16_to_cpu(hdr->data_len)) -
1358 sc->sc_page_off;
1359 ret = o2net_recv_tcp_msg(sc->sc_sock, data, datalen);
1360 if (ret > 0)
1361 sc->sc_page_off += ret;
1362 if (ret <= 0)
1363 goto out;
1364 }
1365
1366 if (sc->sc_page_off - sizeof(struct o2net_msg) == be16_to_cpu(hdr->data_len)) {
1367 /* we can only get here once, the first time we read
1368 * the payload.. so set ret to progress if the handler
1369 * works out. after calling this the message is toast */
1370 ret = o2net_process_message(sc, hdr);
1371 if (ret == 0)
1372 ret = 1;
1373 sc->sc_page_off = 0;
1374 }
1375
1376out:
1377 sclog(sc, "ret = %d\n", ret);
1378 do_gettimeofday(&sc->sc_tv_advance_stop);
1379 return ret;
1380}
1381
1382/* this work func is triggerd by data ready. it reads until it can read no
1383 * more. it interprets 0, eof, as fatal. if data_ready hits while we're doing
1384 * our work the work struct will be marked and we'll be called again. */
David Howellsc4028952006-11-22 14:57:56 +00001385static void o2net_rx_until_empty(struct work_struct *work)
Zach Brown98211482005-12-15 14:31:23 -08001386{
David Howellsc4028952006-11-22 14:57:56 +00001387 struct o2net_sock_container *sc =
1388 container_of(work, struct o2net_sock_container, sc_rx_work);
Zach Brown98211482005-12-15 14:31:23 -08001389 int ret;
1390
1391 do {
1392 ret = o2net_advance_rx(sc);
1393 } while (ret > 0);
1394
1395 if (ret <= 0 && ret != -EAGAIN) {
1396 struct o2net_node *nn = o2net_nn_from_num(sc->sc_node->nd_num);
1397 sclog(sc, "saw error %d, closing\n", ret);
1398 /* not permanent so read failed handshake can retry */
1399 o2net_ensure_shutdown(nn, sc, 0);
1400 }
1401
1402 sc_put(sc);
1403}
1404
1405static int o2net_set_nodelay(struct socket *sock)
1406{
1407 int ret, val = 1;
1408 mm_segment_t oldfs;
1409
1410 oldfs = get_fs();
1411 set_fs(KERNEL_DS);
1412
1413 /*
1414 * Dear unsuspecting programmer,
1415 *
1416 * Don't use sock_setsockopt() for SOL_TCP. It doesn't check its level
1417 * argument and assumes SOL_SOCKET so, say, your TCP_NODELAY will
1418 * silently turn into SO_DEBUG.
1419 *
1420 * Yours,
1421 * Keeper of hilariously fragile interfaces.
1422 */
1423 ret = sock->ops->setsockopt(sock, SOL_TCP, TCP_NODELAY,
1424 (char __user *)&val, sizeof(val));
1425
1426 set_fs(oldfs);
1427 return ret;
1428}
1429
Andrew Beekhof828ae6a2006-12-04 14:04:55 +01001430static void o2net_initialize_handshake(void)
1431{
1432 o2net_hand->o2hb_heartbeat_timeout_ms = cpu_to_be32(
1433 O2HB_MAX_WRITE_TIMEOUT_MS);
Jeff Mahoney409753b2008-03-28 16:44:13 -07001434 o2net_hand->o2net_idle_timeout_ms = cpu_to_be32(o2net_idle_timeout());
Andrew Beekhof828ae6a2006-12-04 14:04:55 +01001435 o2net_hand->o2net_keepalive_delay_ms = cpu_to_be32(
Jeff Mahoney409753b2008-03-28 16:44:13 -07001436 o2net_keepalive_delay());
Andrew Beekhof828ae6a2006-12-04 14:04:55 +01001437 o2net_hand->o2net_reconnect_delay_ms = cpu_to_be32(
Jeff Mahoney409753b2008-03-28 16:44:13 -07001438 o2net_reconnect_delay());
Andrew Beekhof828ae6a2006-12-04 14:04:55 +01001439}
1440
Zach Brown98211482005-12-15 14:31:23 -08001441/* ------------------------------------------------------------ */
1442
1443/* called when a connect completes and after a sock is accepted. the
1444 * rx path will see the response and mark the sc valid */
David Howellsc4028952006-11-22 14:57:56 +00001445static void o2net_sc_connect_completed(struct work_struct *work)
Zach Brown98211482005-12-15 14:31:23 -08001446{
David Howellsc4028952006-11-22 14:57:56 +00001447 struct o2net_sock_container *sc =
1448 container_of(work, struct o2net_sock_container,
1449 sc_connect_work);
Zach Brown98211482005-12-15 14:31:23 -08001450
1451 mlog(ML_MSG, "sc sending handshake with ver %llu id %llx\n",
1452 (unsigned long long)O2NET_PROTOCOL_VERSION,
1453 (unsigned long long)be64_to_cpu(o2net_hand->connector_id));
1454
Andrew Beekhof828ae6a2006-12-04 14:04:55 +01001455 o2net_initialize_handshake();
Zach Brown98211482005-12-15 14:31:23 -08001456 o2net_sendpage(sc, o2net_hand, sizeof(*o2net_hand));
1457 sc_put(sc);
1458}
1459
1460/* this is called as a work_struct func. */
David Howellsc4028952006-11-22 14:57:56 +00001461static void o2net_sc_send_keep_req(struct work_struct *work)
Zach Brown98211482005-12-15 14:31:23 -08001462{
David Howellsc4028952006-11-22 14:57:56 +00001463 struct o2net_sock_container *sc =
1464 container_of(work, struct o2net_sock_container,
1465 sc_keepalive_work.work);
Zach Brown98211482005-12-15 14:31:23 -08001466
1467 o2net_sendpage(sc, o2net_keep_req, sizeof(*o2net_keep_req));
1468 sc_put(sc);
1469}
1470
1471/* socket shutdown does a del_timer_sync against this as it tears down.
1472 * we can't start this timer until we've got to the point in sc buildup
1473 * where shutdown is going to be involved */
1474static void o2net_idle_timer(unsigned long data)
1475{
1476 struct o2net_sock_container *sc = (struct o2net_sock_container *)data;
Tao Ma5cc3bf22008-03-05 15:50:12 +08001477 struct o2net_node *nn = o2net_nn_from_num(sc->sc_node->nd_num);
Zach Brown98211482005-12-15 14:31:23 -08001478 struct timeval now;
1479
1480 do_gettimeofday(&now);
1481
Jeff Mahoneyb5dd8032006-12-04 14:04:54 +01001482 printk(KERN_INFO "o2net: connection to " SC_NODEF_FMT " has been idle for %u.%u "
1483 "seconds, shutting it down.\n", SC_NODEF_ARGS(sc),
Jeff Mahoney409753b2008-03-28 16:44:13 -07001484 o2net_idle_timeout() / 1000,
1485 o2net_idle_timeout() % 1000);
Zach Brown98211482005-12-15 14:31:23 -08001486 mlog(ML_NOTICE, "here are some times that might help debug the "
1487 "situation: (tmr %ld.%ld now %ld.%ld dr %ld.%ld adv "
1488 "%ld.%ld:%ld.%ld func (%08x:%u) %ld.%ld:%ld.%ld)\n",
Mark Fasheh215c7f92006-02-01 16:42:10 -08001489 sc->sc_tv_timer.tv_sec, (long) sc->sc_tv_timer.tv_usec,
1490 now.tv_sec, (long) now.tv_usec,
1491 sc->sc_tv_data_ready.tv_sec, (long) sc->sc_tv_data_ready.tv_usec,
1492 sc->sc_tv_advance_start.tv_sec,
1493 (long) sc->sc_tv_advance_start.tv_usec,
1494 sc->sc_tv_advance_stop.tv_sec,
1495 (long) sc->sc_tv_advance_stop.tv_usec,
Zach Brown98211482005-12-15 14:31:23 -08001496 sc->sc_msg_key, sc->sc_msg_type,
Mark Fasheh215c7f92006-02-01 16:42:10 -08001497 sc->sc_tv_func_start.tv_sec, (long) sc->sc_tv_func_start.tv_usec,
1498 sc->sc_tv_func_stop.tv_sec, (long) sc->sc_tv_func_stop.tv_usec);
Zach Brown98211482005-12-15 14:31:23 -08001499
Tao Ma5cc3bf22008-03-05 15:50:12 +08001500 /*
1501 * Initialize the nn_timeout so that the next connection attempt
1502 * will continue in o2net_start_connect.
1503 */
1504 atomic_set(&nn->nn_timeout, 1);
1505
Zach Brown98211482005-12-15 14:31:23 -08001506 o2net_sc_queue_work(sc, &sc->sc_shutdown_work);
1507}
1508
Jeff Mahoneyb5dd8032006-12-04 14:04:54 +01001509static void o2net_sc_reset_idle_timer(struct o2net_sock_container *sc)
Zach Brown98211482005-12-15 14:31:23 -08001510{
1511 o2net_sc_cancel_delayed_work(sc, &sc->sc_keepalive_work);
1512 o2net_sc_queue_delayed_work(sc, &sc->sc_keepalive_work,
Jeff Mahoney409753b2008-03-28 16:44:13 -07001513 msecs_to_jiffies(o2net_keepalive_delay()));
Zach Brown98211482005-12-15 14:31:23 -08001514 do_gettimeofday(&sc->sc_tv_timer);
1515 mod_timer(&sc->sc_idle_timeout,
Jeff Mahoney409753b2008-03-28 16:44:13 -07001516 jiffies + msecs_to_jiffies(o2net_idle_timeout()));
Jeff Mahoneyb5dd8032006-12-04 14:04:54 +01001517}
1518
1519static void o2net_sc_postpone_idle(struct o2net_sock_container *sc)
1520{
1521 /* Only push out an existing timer */
1522 if (timer_pending(&sc->sc_idle_timeout))
1523 o2net_sc_reset_idle_timer(sc);
Zach Brown98211482005-12-15 14:31:23 -08001524}
1525
1526/* this work func is kicked whenever a path sets the nn state which doesn't
1527 * have valid set. This includes seeing hb come up, losing a connection,
1528 * having a connect attempt fail, etc. This centralizes the logic which decides
1529 * if a connect attempt should be made or if we should give up and all future
1530 * transmit attempts should fail */
David Howellsc4028952006-11-22 14:57:56 +00001531static void o2net_start_connect(struct work_struct *work)
Zach Brown98211482005-12-15 14:31:23 -08001532{
David Howellsc4028952006-11-22 14:57:56 +00001533 struct o2net_node *nn =
1534 container_of(work, struct o2net_node, nn_connect_work.work);
Zach Brown98211482005-12-15 14:31:23 -08001535 struct o2net_sock_container *sc = NULL;
Sunil Mushranb7668c72006-02-28 23:28:01 -08001536 struct o2nm_node *node = NULL, *mynode = NULL;
Zach Brown98211482005-12-15 14:31:23 -08001537 struct socket *sock = NULL;
1538 struct sockaddr_in myaddr = {0, }, remoteaddr = {0, };
David Howellsc4028952006-11-22 14:57:56 +00001539 int ret = 0, stop;
Tao Ma5cc3bf22008-03-05 15:50:12 +08001540 unsigned int timeout;
Zach Brown98211482005-12-15 14:31:23 -08001541
1542 /* if we're greater we initiate tx, otherwise we accept */
1543 if (o2nm_this_node() <= o2net_num_from_nn(nn))
1544 goto out;
1545
1546 /* watch for racing with tearing a node down */
1547 node = o2nm_get_node_by_num(o2net_num_from_nn(nn));
1548 if (node == NULL) {
1549 ret = 0;
1550 goto out;
1551 }
1552
Sunil Mushranb7668c72006-02-28 23:28:01 -08001553 mynode = o2nm_get_node_by_num(o2nm_this_node());
1554 if (mynode == NULL) {
1555 ret = 0;
1556 goto out;
1557 }
1558
Zach Brown98211482005-12-15 14:31:23 -08001559 spin_lock(&nn->nn_lock);
Tao Ma5cc3bf22008-03-05 15:50:12 +08001560 /*
1561 * see if we already have one pending or have given up.
1562 * For nn_timeout, it is set when we close the connection
1563 * because of the idle time out. So it means that we have
1564 * at least connected to that node successfully once,
1565 * now try to connect to it again.
1566 */
1567 timeout = atomic_read(&nn->nn_timeout);
1568 stop = (nn->nn_sc ||
1569 (nn->nn_persistent_error &&
1570 (nn->nn_persistent_error != -ENOTCONN || timeout == 0)));
Zach Brown98211482005-12-15 14:31:23 -08001571 spin_unlock(&nn->nn_lock);
David Howellsc4028952006-11-22 14:57:56 +00001572 if (stop)
Zach Brown98211482005-12-15 14:31:23 -08001573 goto out;
1574
1575 nn->nn_last_connect_attempt = jiffies;
1576
1577 sc = sc_alloc(node);
1578 if (sc == NULL) {
1579 mlog(0, "couldn't allocate sc\n");
1580 ret = -ENOMEM;
1581 goto out;
1582 }
1583
1584 ret = sock_create(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock);
1585 if (ret < 0) {
1586 mlog(0, "can't create socket: %d\n", ret);
1587 goto out;
1588 }
1589 sc->sc_sock = sock; /* freed by sc_kref_release */
1590
1591 sock->sk->sk_allocation = GFP_ATOMIC;
1592
1593 myaddr.sin_family = AF_INET;
Mark Fasheh5fdf1e62007-04-27 16:50:03 -07001594 myaddr.sin_addr.s_addr = mynode->nd_ipv4_address;
Al Viro97bd7912007-12-05 08:46:47 +00001595 myaddr.sin_port = htons(0); /* any port */
Zach Brown98211482005-12-15 14:31:23 -08001596
1597 ret = sock->ops->bind(sock, (struct sockaddr *)&myaddr,
1598 sizeof(myaddr));
1599 if (ret) {
Harvey Harrisonbe859402008-10-31 00:56:28 -07001600 mlog(ML_ERROR, "bind failed with %d at address %pI4\n",
1601 ret, &mynode->nd_ipv4_address);
Zach Brown98211482005-12-15 14:31:23 -08001602 goto out;
1603 }
1604
1605 ret = o2net_set_nodelay(sc->sc_sock);
1606 if (ret) {
1607 mlog(ML_ERROR, "setting TCP_NODELAY failed with %d\n", ret);
1608 goto out;
1609 }
1610
1611 o2net_register_callbacks(sc->sc_sock->sk, sc);
1612
1613 spin_lock(&nn->nn_lock);
1614 /* handshake completion will set nn->nn_sc_valid */
1615 o2net_set_nn_state(nn, sc, 0, 0);
1616 spin_unlock(&nn->nn_lock);
1617
1618 remoteaddr.sin_family = AF_INET;
Mark Fasheh5fdf1e62007-04-27 16:50:03 -07001619 remoteaddr.sin_addr.s_addr = node->nd_ipv4_address;
1620 remoteaddr.sin_port = node->nd_ipv4_port;
Zach Brown98211482005-12-15 14:31:23 -08001621
1622 ret = sc->sc_sock->ops->connect(sc->sc_sock,
1623 (struct sockaddr *)&remoteaddr,
1624 sizeof(remoteaddr),
1625 O_NONBLOCK);
1626 if (ret == -EINPROGRESS)
1627 ret = 0;
1628
1629out:
1630 if (ret) {
1631 mlog(ML_NOTICE, "connect attempt to " SC_NODEF_FMT " failed "
1632 "with errno %d\n", SC_NODEF_ARGS(sc), ret);
1633 /* 0 err so that another will be queued and attempted
1634 * from set_nn_state */
1635 if (sc)
1636 o2net_ensure_shutdown(nn, sc, 0);
1637 }
1638 if (sc)
1639 sc_put(sc);
1640 if (node)
1641 o2nm_node_put(node);
Sunil Mushranb7668c72006-02-28 23:28:01 -08001642 if (mynode)
1643 o2nm_node_put(mynode);
Zach Brown98211482005-12-15 14:31:23 -08001644
1645 return;
1646}
1647
David Howellsc4028952006-11-22 14:57:56 +00001648static void o2net_connect_expired(struct work_struct *work)
Zach Brown98211482005-12-15 14:31:23 -08001649{
David Howellsc4028952006-11-22 14:57:56 +00001650 struct o2net_node *nn =
1651 container_of(work, struct o2net_node, nn_connect_expired.work);
Zach Brown98211482005-12-15 14:31:23 -08001652
1653 spin_lock(&nn->nn_lock);
1654 if (!nn->nn_sc_valid) {
1655 mlog(ML_ERROR, "no connection established with node %u after "
Jeff Mahoneyb5dd8032006-12-04 14:04:54 +01001656 "%u.%u seconds, giving up and returning errors.\n",
1657 o2net_num_from_nn(nn),
Jeff Mahoney409753b2008-03-28 16:44:13 -07001658 o2net_idle_timeout() / 1000,
1659 o2net_idle_timeout() % 1000);
Zach Brown98211482005-12-15 14:31:23 -08001660
1661 o2net_set_nn_state(nn, NULL, 0, -ENOTCONN);
1662 }
1663 spin_unlock(&nn->nn_lock);
1664}
1665
David Howellsc4028952006-11-22 14:57:56 +00001666static void o2net_still_up(struct work_struct *work)
Zach Brown98211482005-12-15 14:31:23 -08001667{
David Howellsc4028952006-11-22 14:57:56 +00001668 struct o2net_node *nn =
1669 container_of(work, struct o2net_node, nn_still_up.work);
Zach Brown98211482005-12-15 14:31:23 -08001670
1671 o2quo_hb_still_up(o2net_num_from_nn(nn));
1672}
1673
1674/* ------------------------------------------------------------ */
1675
1676void o2net_disconnect_node(struct o2nm_node *node)
1677{
1678 struct o2net_node *nn = o2net_nn_from_num(node->nd_num);
1679
1680 /* don't reconnect until it's heartbeating again */
1681 spin_lock(&nn->nn_lock);
Tao Ma5cc3bf22008-03-05 15:50:12 +08001682 atomic_set(&nn->nn_timeout, 0);
Zach Brown98211482005-12-15 14:31:23 -08001683 o2net_set_nn_state(nn, NULL, 0, -ENOTCONN);
1684 spin_unlock(&nn->nn_lock);
1685
1686 if (o2net_wq) {
1687 cancel_delayed_work(&nn->nn_connect_expired);
1688 cancel_delayed_work(&nn->nn_connect_work);
1689 cancel_delayed_work(&nn->nn_still_up);
1690 flush_workqueue(o2net_wq);
1691 }
1692}
1693
1694static void o2net_hb_node_down_cb(struct o2nm_node *node, int node_num,
1695 void *data)
1696{
1697 o2quo_hb_down(node_num);
1698
1699 if (node_num != o2nm_this_node())
1700 o2net_disconnect_node(node);
Andrew Beekhof828ae6a2006-12-04 14:04:55 +01001701
1702 BUG_ON(atomic_read(&o2net_connected_peers) < 0);
Zach Brown98211482005-12-15 14:31:23 -08001703}
1704
1705static void o2net_hb_node_up_cb(struct o2nm_node *node, int node_num,
1706 void *data)
1707{
1708 struct o2net_node *nn = o2net_nn_from_num(node_num);
1709
1710 o2quo_hb_up(node_num);
1711
1712 /* ensure an immediate connect attempt */
1713 nn->nn_last_connect_attempt = jiffies -
Jeff Mahoney409753b2008-03-28 16:44:13 -07001714 (msecs_to_jiffies(o2net_reconnect_delay()) + 1);
Zach Brown98211482005-12-15 14:31:23 -08001715
1716 if (node_num != o2nm_this_node()) {
Zach Brown98211482005-12-15 14:31:23 -08001717 /* believe it or not, accept and node hearbeating testing
1718 * can succeed for this node before we got here.. so
1719 * only use set_nn_state to clear the persistent error
1720 * if that hasn't already happened */
1721 spin_lock(&nn->nn_lock);
Tao Ma5cc3bf22008-03-05 15:50:12 +08001722 atomic_set(&nn->nn_timeout, 0);
Zach Brown98211482005-12-15 14:31:23 -08001723 if (nn->nn_persistent_error)
1724 o2net_set_nn_state(nn, NULL, 0, 0);
1725 spin_unlock(&nn->nn_lock);
1726 }
1727}
1728
1729void o2net_unregister_hb_callbacks(void)
1730{
Joel Becker14829422007-06-14 21:40:49 -07001731 o2hb_unregister_callback(NULL, &o2net_hb_up);
1732 o2hb_unregister_callback(NULL, &o2net_hb_down);
Zach Brown98211482005-12-15 14:31:23 -08001733}
1734
1735int o2net_register_hb_callbacks(void)
1736{
1737 int ret;
1738
1739 o2hb_setup_callback(&o2net_hb_down, O2HB_NODE_DOWN_CB,
1740 o2net_hb_node_down_cb, NULL, O2NET_HB_PRI);
1741 o2hb_setup_callback(&o2net_hb_up, O2HB_NODE_UP_CB,
1742 o2net_hb_node_up_cb, NULL, O2NET_HB_PRI);
1743
Joel Becker14829422007-06-14 21:40:49 -07001744 ret = o2hb_register_callback(NULL, &o2net_hb_up);
Zach Brown98211482005-12-15 14:31:23 -08001745 if (ret == 0)
Joel Becker14829422007-06-14 21:40:49 -07001746 ret = o2hb_register_callback(NULL, &o2net_hb_down);
Zach Brown98211482005-12-15 14:31:23 -08001747
1748 if (ret)
1749 o2net_unregister_hb_callbacks();
1750
1751 return ret;
1752}
1753
1754/* ------------------------------------------------------------ */
1755
1756static int o2net_accept_one(struct socket *sock)
1757{
1758 int ret, slen;
1759 struct sockaddr_in sin;
1760 struct socket *new_sock = NULL;
1761 struct o2nm_node *node = NULL;
1762 struct o2net_sock_container *sc = NULL;
1763 struct o2net_node *nn;
1764
1765 BUG_ON(sock == NULL);
1766 ret = sock_create_lite(sock->sk->sk_family, sock->sk->sk_type,
1767 sock->sk->sk_protocol, &new_sock);
1768 if (ret)
1769 goto out;
1770
1771 new_sock->type = sock->type;
1772 new_sock->ops = sock->ops;
1773 ret = sock->ops->accept(sock, new_sock, O_NONBLOCK);
1774 if (ret < 0)
1775 goto out;
1776
1777 new_sock->sk->sk_allocation = GFP_ATOMIC;
1778
1779 ret = o2net_set_nodelay(new_sock);
1780 if (ret) {
1781 mlog(ML_ERROR, "setting TCP_NODELAY failed with %d\n", ret);
1782 goto out;
1783 }
1784
1785 slen = sizeof(sin);
1786 ret = new_sock->ops->getname(new_sock, (struct sockaddr *) &sin,
1787 &slen, 1);
1788 if (ret < 0)
1789 goto out;
1790
Al Viro97bd7912007-12-05 08:46:47 +00001791 node = o2nm_get_node_by_ip(sin.sin_addr.s_addr);
Zach Brown98211482005-12-15 14:31:23 -08001792 if (node == NULL) {
Harvey Harrisonbe859402008-10-31 00:56:28 -07001793 mlog(ML_NOTICE, "attempt to connect from unknown node at %pI4:%d\n",
1794 &sin.sin_addr.s_addr, ntohs(sin.sin_port));
Zach Brown98211482005-12-15 14:31:23 -08001795 ret = -EINVAL;
1796 goto out;
1797 }
1798
1799 if (o2nm_this_node() > node->nd_num) {
1800 mlog(ML_NOTICE, "unexpected connect attempted from a lower "
Harvey Harrisonbe859402008-10-31 00:56:28 -07001801 "numbered node '%s' at " "%pI4:%d with num %u\n",
1802 node->nd_name, &sin.sin_addr.s_addr,
Al Viro97bd7912007-12-05 08:46:47 +00001803 ntohs(sin.sin_port), node->nd_num);
Zach Brown98211482005-12-15 14:31:23 -08001804 ret = -EINVAL;
1805 goto out;
1806 }
1807
1808 /* this happens all the time when the other node sees our heartbeat
1809 * and tries to connect before we see their heartbeat */
1810 if (!o2hb_check_node_heartbeating_from_callback(node->nd_num)) {
1811 mlog(ML_CONN, "attempt to connect from node '%s' at "
Harvey Harrisonbe859402008-10-31 00:56:28 -07001812 "%pI4:%d but it isn't heartbeating\n",
1813 node->nd_name, &sin.sin_addr.s_addr,
Al Viro97bd7912007-12-05 08:46:47 +00001814 ntohs(sin.sin_port));
Zach Brown98211482005-12-15 14:31:23 -08001815 ret = -EINVAL;
1816 goto out;
1817 }
1818
1819 nn = o2net_nn_from_num(node->nd_num);
1820
1821 spin_lock(&nn->nn_lock);
1822 if (nn->nn_sc)
1823 ret = -EBUSY;
1824 else
1825 ret = 0;
1826 spin_unlock(&nn->nn_lock);
1827 if (ret) {
1828 mlog(ML_NOTICE, "attempt to connect from node '%s' at "
Harvey Harrisonbe859402008-10-31 00:56:28 -07001829 "%pI4:%d but it already has an open connection\n",
1830 node->nd_name, &sin.sin_addr.s_addr,
Al Viro97bd7912007-12-05 08:46:47 +00001831 ntohs(sin.sin_port));
Zach Brown98211482005-12-15 14:31:23 -08001832 goto out;
1833 }
1834
1835 sc = sc_alloc(node);
1836 if (sc == NULL) {
1837 ret = -ENOMEM;
1838 goto out;
1839 }
1840
1841 sc->sc_sock = new_sock;
1842 new_sock = NULL;
1843
1844 spin_lock(&nn->nn_lock);
Tao Ma5cc3bf22008-03-05 15:50:12 +08001845 atomic_set(&nn->nn_timeout, 0);
Zach Brown98211482005-12-15 14:31:23 -08001846 o2net_set_nn_state(nn, sc, 0, 0);
1847 spin_unlock(&nn->nn_lock);
1848
1849 o2net_register_callbacks(sc->sc_sock->sk, sc);
1850 o2net_sc_queue_work(sc, &sc->sc_rx_work);
1851
Andrew Beekhof828ae6a2006-12-04 14:04:55 +01001852 o2net_initialize_handshake();
Zach Brown98211482005-12-15 14:31:23 -08001853 o2net_sendpage(sc, o2net_hand, sizeof(*o2net_hand));
1854
1855out:
1856 if (new_sock)
1857 sock_release(new_sock);
1858 if (node)
1859 o2nm_node_put(node);
1860 if (sc)
1861 sc_put(sc);
1862 return ret;
1863}
1864
David Howellsc4028952006-11-22 14:57:56 +00001865static void o2net_accept_many(struct work_struct *work)
Zach Brown98211482005-12-15 14:31:23 -08001866{
David Howellsc4028952006-11-22 14:57:56 +00001867 struct socket *sock = o2net_listen_sock;
Zach Brown98211482005-12-15 14:31:23 -08001868 while (o2net_accept_one(sock) == 0)
1869 cond_resched();
1870}
1871
1872static void o2net_listen_data_ready(struct sock *sk, int bytes)
1873{
1874 void (*ready)(struct sock *sk, int bytes);
1875
1876 read_lock(&sk->sk_callback_lock);
1877 ready = sk->sk_user_data;
1878 if (ready == NULL) { /* check for teardown race */
1879 ready = sk->sk_data_ready;
1880 goto out;
1881 }
1882
1883 /* ->sk_data_ready is also called for a newly established child socket
1884 * before it has been accepted and the acceptor has set up their
1885 * data_ready.. we only want to queue listen work for our listening
1886 * socket */
1887 if (sk->sk_state == TCP_LISTEN) {
1888 mlog(ML_TCP, "bytes: %d\n", bytes);
1889 queue_work(o2net_wq, &o2net_listen_work);
1890 }
1891
1892out:
1893 read_unlock(&sk->sk_callback_lock);
1894 ready(sk, bytes);
1895}
1896
Sunil Mushranab81afd2007-01-29 14:57:14 -08001897static int o2net_open_listening_sock(__be32 addr, __be16 port)
Zach Brown98211482005-12-15 14:31:23 -08001898{
1899 struct socket *sock = NULL;
1900 int ret;
1901 struct sockaddr_in sin = {
1902 .sin_family = PF_INET,
Mark Fasheh5fdf1e62007-04-27 16:50:03 -07001903 .sin_addr = { .s_addr = addr },
1904 .sin_port = port,
Zach Brown98211482005-12-15 14:31:23 -08001905 };
1906
1907 ret = sock_create(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock);
1908 if (ret < 0) {
1909 mlog(ML_ERROR, "unable to create socket, ret=%d\n", ret);
1910 goto out;
1911 }
1912
1913 sock->sk->sk_allocation = GFP_ATOMIC;
1914
1915 write_lock_bh(&sock->sk->sk_callback_lock);
1916 sock->sk->sk_user_data = sock->sk->sk_data_ready;
1917 sock->sk->sk_data_ready = o2net_listen_data_ready;
1918 write_unlock_bh(&sock->sk->sk_callback_lock);
1919
1920 o2net_listen_sock = sock;
David Howellsc4028952006-11-22 14:57:56 +00001921 INIT_WORK(&o2net_listen_work, o2net_accept_many);
Zach Brown98211482005-12-15 14:31:23 -08001922
1923 sock->sk->sk_reuse = 1;
1924 ret = sock->ops->bind(sock, (struct sockaddr *)&sin, sizeof(sin));
1925 if (ret < 0) {
Harvey Harrisonbe859402008-10-31 00:56:28 -07001926 mlog(ML_ERROR, "unable to bind socket at %pI4:%u, "
1927 "ret=%d\n", &addr, ntohs(port), ret);
Zach Brown98211482005-12-15 14:31:23 -08001928 goto out;
1929 }
1930
1931 ret = sock->ops->listen(sock, 64);
1932 if (ret < 0) {
Harvey Harrisonbe859402008-10-31 00:56:28 -07001933 mlog(ML_ERROR, "unable to listen on %pI4:%u, ret=%d\n",
1934 &addr, ntohs(port), ret);
Zach Brown98211482005-12-15 14:31:23 -08001935 }
1936
1937out:
1938 if (ret) {
1939 o2net_listen_sock = NULL;
1940 if (sock)
1941 sock_release(sock);
1942 }
1943 return ret;
1944}
1945
1946/*
1947 * called from node manager when we should bring up our network listening
1948 * socket. node manager handles all the serialization to only call this
1949 * once and to match it with o2net_stop_listening(). note,
1950 * o2nm_this_node() doesn't work yet as we're being called while it
1951 * is being set up.
1952 */
1953int o2net_start_listening(struct o2nm_node *node)
1954{
1955 int ret = 0;
1956
1957 BUG_ON(o2net_wq != NULL);
1958 BUG_ON(o2net_listen_sock != NULL);
1959
1960 mlog(ML_KTHREAD, "starting o2net thread...\n");
1961 o2net_wq = create_singlethread_workqueue("o2net");
1962 if (o2net_wq == NULL) {
1963 mlog(ML_ERROR, "unable to launch o2net thread\n");
1964 return -ENOMEM; /* ? */
1965 }
1966
Sunil Mushranab81afd2007-01-29 14:57:14 -08001967 ret = o2net_open_listening_sock(node->nd_ipv4_address,
1968 node->nd_ipv4_port);
Zach Brown98211482005-12-15 14:31:23 -08001969 if (ret) {
1970 destroy_workqueue(o2net_wq);
1971 o2net_wq = NULL;
1972 } else
1973 o2quo_conn_up(node->nd_num);
1974
1975 return ret;
1976}
1977
1978/* again, o2nm_this_node() doesn't work here as we're involved in
1979 * tearing it down */
1980void o2net_stop_listening(struct o2nm_node *node)
1981{
1982 struct socket *sock = o2net_listen_sock;
1983 size_t i;
1984
1985 BUG_ON(o2net_wq == NULL);
1986 BUG_ON(o2net_listen_sock == NULL);
1987
1988 /* stop the listening socket from generating work */
1989 write_lock_bh(&sock->sk->sk_callback_lock);
1990 sock->sk->sk_data_ready = sock->sk->sk_user_data;
1991 sock->sk->sk_user_data = NULL;
1992 write_unlock_bh(&sock->sk->sk_callback_lock);
1993
1994 for (i = 0; i < ARRAY_SIZE(o2net_nodes); i++) {
1995 struct o2nm_node *node = o2nm_get_node_by_num(i);
1996 if (node) {
1997 o2net_disconnect_node(node);
1998 o2nm_node_put(node);
1999 }
2000 }
2001
2002 /* finish all work and tear down the work queue */
2003 mlog(ML_KTHREAD, "waiting for o2net thread to exit....\n");
2004 destroy_workqueue(o2net_wq);
2005 o2net_wq = NULL;
2006
2007 sock_release(o2net_listen_sock);
2008 o2net_listen_sock = NULL;
2009
2010 o2quo_conn_err(node->nd_num);
2011}
2012
2013/* ------------------------------------------------------------ */
2014
2015int o2net_init(void)
2016{
2017 unsigned long i;
2018
2019 o2quo_init();
2020
Sunil Mushran2309e9e2008-04-14 10:46:19 -07002021 if (o2net_debugfs_init())
2022 return -ENOMEM;
2023
Robert P. J. Daycd861282006-12-13 00:34:52 -08002024 o2net_hand = kzalloc(sizeof(struct o2net_handshake), GFP_KERNEL);
2025 o2net_keep_req = kzalloc(sizeof(struct o2net_msg), GFP_KERNEL);
2026 o2net_keep_resp = kzalloc(sizeof(struct o2net_msg), GFP_KERNEL);
Zach Brown98211482005-12-15 14:31:23 -08002027 if (!o2net_hand || !o2net_keep_req || !o2net_keep_resp) {
2028 kfree(o2net_hand);
2029 kfree(o2net_keep_req);
2030 kfree(o2net_keep_resp);
2031 return -ENOMEM;
2032 }
2033
2034 o2net_hand->protocol_version = cpu_to_be64(O2NET_PROTOCOL_VERSION);
2035 o2net_hand->connector_id = cpu_to_be64(1);
2036
2037 o2net_keep_req->magic = cpu_to_be16(O2NET_MSG_KEEP_REQ_MAGIC);
2038 o2net_keep_resp->magic = cpu_to_be16(O2NET_MSG_KEEP_RESP_MAGIC);
2039
2040 for (i = 0; i < ARRAY_SIZE(o2net_nodes); i++) {
2041 struct o2net_node *nn = o2net_nn_from_num(i);
2042
Tao Ma5cc3bf22008-03-05 15:50:12 +08002043 atomic_set(&nn->nn_timeout, 0);
Zach Brown98211482005-12-15 14:31:23 -08002044 spin_lock_init(&nn->nn_lock);
David Howellsc4028952006-11-22 14:57:56 +00002045 INIT_DELAYED_WORK(&nn->nn_connect_work, o2net_start_connect);
2046 INIT_DELAYED_WORK(&nn->nn_connect_expired,
2047 o2net_connect_expired);
2048 INIT_DELAYED_WORK(&nn->nn_still_up, o2net_still_up);
Zach Brown98211482005-12-15 14:31:23 -08002049 /* until we see hb from a node we'll return einval */
2050 nn->nn_persistent_error = -ENOTCONN;
2051 init_waitqueue_head(&nn->nn_sc_wq);
2052 idr_init(&nn->nn_status_idr);
2053 INIT_LIST_HEAD(&nn->nn_status_list);
2054 }
2055
2056 return 0;
2057}
2058
2059void o2net_exit(void)
2060{
2061 o2quo_exit();
2062 kfree(o2net_hand);
2063 kfree(o2net_keep_req);
2064 kfree(o2net_keep_resp);
Sunil Mushran2309e9e2008-04-14 10:46:19 -07002065 o2net_debugfs_exit();
Zach Brown98211482005-12-15 14:31:23 -08002066}