~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/ocfs2/cluster/tcp.c

Version: ~ [ linux-5.10-rc1 ] ~ [ linux-5.9.1 ] ~ [ linux-5.8.16 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.72 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.152 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.202 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.240 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.240 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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/sched/mm.h>
 58 #include <linux/jiffies.h>
 59 #include <linux/slab.h>
 60 #include <linux/idr.h>
 61 #include <linux/kref.h>
 62 #include <linux/net.h>
 63 #include <linux/export.h>
 64 #include <net/tcp.h>
 65 
 66 #include <linux/uaccess.h>
 67 
 68 #include "heartbeat.h"
 69 #include "tcp.h"
 70 #include "nodemanager.h"
 71 #define MLOG_MASK_PREFIX ML_TCP
 72 #include "masklog.h"
 73 #include "quorum.h"
 74 
 75 #include "tcp_internal.h"
 76 
 77 #define SC_NODEF_FMT "node %s (num %u) at %pI4:%u"
 78 #define SC_NODEF_ARGS(sc) sc->sc_node->nd_name, sc->sc_node->nd_num,    \
 79                           &sc->sc_node->nd_ipv4_address,                \
 80                           ntohs(sc->sc_node->nd_ipv4_port)
 81 
 82 /*
 83  * In the following two log macros, the whitespace after the ',' just
 84  * before ##args is intentional. Otherwise, gcc 2.95 will eat the
 85  * previous token if args expands to nothing.
 86  */
 87 #define msglog(hdr, fmt, args...) do {                                  \
 88         typeof(hdr) __hdr = (hdr);                                      \
 89         mlog(ML_MSG, "[mag %u len %u typ %u stat %d sys_stat %d "       \
 90              "key %08x num %u] " fmt,                                   \
 91              be16_to_cpu(__hdr->magic), be16_to_cpu(__hdr->data_len),   \
 92              be16_to_cpu(__hdr->msg_type), be32_to_cpu(__hdr->status),  \
 93              be32_to_cpu(__hdr->sys_status), be32_to_cpu(__hdr->key),   \
 94              be32_to_cpu(__hdr->msg_num) ,  ##args);                    \
 95 } while (0)
 96 
 97 #define sclog(sc, fmt, args...) do {                                    \
 98         typeof(sc) __sc = (sc);                                         \
 99         mlog(ML_SOCKET, "[sc %p refs %d sock %p node %u page %p "       \
100              "pg_off %zu] " fmt, __sc,                                  \
101              kref_read(&__sc->sc_kref), __sc->sc_sock,  \
102             __sc->sc_node->nd_num, __sc->sc_page, __sc->sc_page_off ,   \
103             ##args);                                                    \
104 } while (0)
105 
106 static DEFINE_RWLOCK(o2net_handler_lock);
107 static struct rb_root o2net_handler_tree = RB_ROOT;
108 
109 static struct o2net_node o2net_nodes[O2NM_MAX_NODES];
110 
111 /* XXX someday we'll need better accounting */
112 static struct socket *o2net_listen_sock;
113 
114 /*
115  * listen work is only queued by the listening socket callbacks on the
116  * o2net_wq.  teardown detaches the callbacks before destroying the workqueue.
117  * quorum work is queued as sock containers are shutdown.. stop_listening
118  * tears down all the node's sock containers, preventing future shutdowns
119  * and queued quroum work, before canceling delayed quorum work and
120  * destroying the work queue.
121  */
122 static struct workqueue_struct *o2net_wq;
123 static struct work_struct o2net_listen_work;
124 
125 static struct o2hb_callback_func o2net_hb_up, o2net_hb_down;
126 #define O2NET_HB_PRI 0x1
127 
128 static struct o2net_handshake *o2net_hand;
129 static struct o2net_msg *o2net_keep_req, *o2net_keep_resp;
130 
131 static int o2net_sys_err_translations[O2NET_ERR_MAX] =
132                 {[O2NET_ERR_NONE]       = 0,
133                  [O2NET_ERR_NO_HNDLR]   = -ENOPROTOOPT,
134                  [O2NET_ERR_OVERFLOW]   = -EOVERFLOW,
135                  [O2NET_ERR_DIED]       = -EHOSTDOWN,};
136 
137 /* can't quite avoid *all* internal declarations :/ */
138 static void o2net_sc_connect_completed(struct work_struct *work);
139 static void o2net_rx_until_empty(struct work_struct *work);
140 static void o2net_shutdown_sc(struct work_struct *work);
141 static void o2net_listen_data_ready(struct sock *sk);
142 static void o2net_sc_send_keep_req(struct work_struct *work);
143 static void o2net_idle_timer(struct timer_list *t);
144 static void o2net_sc_postpone_idle(struct o2net_sock_container *sc);
145 static void o2net_sc_reset_idle_timer(struct o2net_sock_container *sc);
146 
147 #ifdef CONFIG_DEBUG_FS
148 static void o2net_init_nst(struct o2net_send_tracking *nst, u32 msgtype,
149                            u32 msgkey, struct task_struct *task, u8 node)
150 {
151         INIT_LIST_HEAD(&nst->st_net_debug_item);
152         nst->st_task = task;
153         nst->st_msg_type = msgtype;
154         nst->st_msg_key = msgkey;
155         nst->st_node = node;
156 }
157 
158 static inline void o2net_set_nst_sock_time(struct o2net_send_tracking *nst)
159 {
160         nst->st_sock_time = ktime_get();
161 }
162 
163 static inline void o2net_set_nst_send_time(struct o2net_send_tracking *nst)
164 {
165         nst->st_send_time = ktime_get();
166 }
167 
168 static inline void o2net_set_nst_status_time(struct o2net_send_tracking *nst)
169 {
170         nst->st_status_time = ktime_get();
171 }
172 
173 static inline void o2net_set_nst_sock_container(struct o2net_send_tracking *nst,
174                                                 struct o2net_sock_container *sc)
175 {
176         nst->st_sc = sc;
177 }
178 
179 static inline void o2net_set_nst_msg_id(struct o2net_send_tracking *nst,
180                                         u32 msg_id)
181 {
182         nst->st_id = msg_id;
183 }
184 
185 static inline void o2net_set_sock_timer(struct o2net_sock_container *sc)
186 {
187         sc->sc_tv_timer = ktime_get();
188 }
189 
190 static inline void o2net_set_data_ready_time(struct o2net_sock_container *sc)
191 {
192         sc->sc_tv_data_ready = ktime_get();
193 }
194 
195 static inline void o2net_set_advance_start_time(struct o2net_sock_container *sc)
196 {
197         sc->sc_tv_advance_start = ktime_get();
198 }
199 
200 static inline void o2net_set_advance_stop_time(struct o2net_sock_container *sc)
201 {
202         sc->sc_tv_advance_stop = ktime_get();
203 }
204 
205 static inline void o2net_set_func_start_time(struct o2net_sock_container *sc)
206 {
207         sc->sc_tv_func_start = ktime_get();
208 }
209 
210 static inline void o2net_set_func_stop_time(struct o2net_sock_container *sc)
211 {
212         sc->sc_tv_func_stop = ktime_get();
213 }
214 
215 #else  /* CONFIG_DEBUG_FS */
216 # define o2net_init_nst(a, b, c, d, e)
217 # define o2net_set_nst_sock_time(a)
218 # define o2net_set_nst_send_time(a)
219 # define o2net_set_nst_status_time(a)
220 # define o2net_set_nst_sock_container(a, b)
221 # define o2net_set_nst_msg_id(a, b)
222 # define o2net_set_sock_timer(a)
223 # define o2net_set_data_ready_time(a)
224 # define o2net_set_advance_start_time(a)
225 # define o2net_set_advance_stop_time(a)
226 # define o2net_set_func_start_time(a)
227 # define o2net_set_func_stop_time(a)
228 #endif /* CONFIG_DEBUG_FS */
229 
230 #ifdef CONFIG_OCFS2_FS_STATS
231 static ktime_t o2net_get_func_run_time(struct o2net_sock_container *sc)
232 {
233         return ktime_sub(sc->sc_tv_func_stop, sc->sc_tv_func_start);
234 }
235 
236 static void o2net_update_send_stats(struct o2net_send_tracking *nst,
237                                     struct o2net_sock_container *sc)
238 {
239         sc->sc_tv_status_total = ktime_add(sc->sc_tv_status_total,
240                                            ktime_sub(ktime_get(),
241                                                      nst->st_status_time));
242         sc->sc_tv_send_total = ktime_add(sc->sc_tv_send_total,
243                                          ktime_sub(nst->st_status_time,
244                                                    nst->st_send_time));
245         sc->sc_tv_acquiry_total = ktime_add(sc->sc_tv_acquiry_total,
246                                             ktime_sub(nst->st_send_time,
247                                                       nst->st_sock_time));
248         sc->sc_send_count++;
249 }
250 
251 static void o2net_update_recv_stats(struct o2net_sock_container *sc)
252 {
253         sc->sc_tv_process_total = ktime_add(sc->sc_tv_process_total,
254                                             o2net_get_func_run_time(sc));
255         sc->sc_recv_count++;
256 }
257 
258 #else
259 
260 # define o2net_update_send_stats(a, b)
261 
262 # define o2net_update_recv_stats(sc)
263 
264 #endif /* CONFIG_OCFS2_FS_STATS */
265 
266 static inline unsigned int o2net_reconnect_delay(void)
267 {
268         return o2nm_single_cluster->cl_reconnect_delay_ms;
269 }
270 
271 static inline unsigned int o2net_keepalive_delay(void)
272 {
273         return o2nm_single_cluster->cl_keepalive_delay_ms;
274 }
275 
276 static inline unsigned int o2net_idle_timeout(void)
277 {
278         return o2nm_single_cluster->cl_idle_timeout_ms;
279 }
280 
281 static inline int o2net_sys_err_to_errno(enum o2net_system_error err)
282 {
283         int trans;
284         BUG_ON(err >= O2NET_ERR_MAX);
285         trans = o2net_sys_err_translations[err];
286 
287         /* Just in case we mess up the translation table above */
288         BUG_ON(err != O2NET_ERR_NONE && trans == 0);
289         return trans;
290 }
291 
292 static struct o2net_node * o2net_nn_from_num(u8 node_num)
293 {
294         BUG_ON(node_num >= ARRAY_SIZE(o2net_nodes));
295         return &o2net_nodes[node_num];
296 }
297 
298 static u8 o2net_num_from_nn(struct o2net_node *nn)
299 {
300         BUG_ON(nn == NULL);
301         return nn - o2net_nodes;
302 }
303 
304 /* ------------------------------------------------------------ */
305 
306 static int o2net_prep_nsw(struct o2net_node *nn, struct o2net_status_wait *nsw)
307 {
308         int ret;
309 
310         spin_lock(&nn->nn_lock);
311         ret = idr_alloc(&nn->nn_status_idr, nsw, 0, 0, GFP_ATOMIC);
312         if (ret >= 0) {
313                 nsw->ns_id = ret;
314                 list_add_tail(&nsw->ns_node_item, &nn->nn_status_list);
315         }
316         spin_unlock(&nn->nn_lock);
317         if (ret < 0)
318                 return ret;
319 
320         init_waitqueue_head(&nsw->ns_wq);
321         nsw->ns_sys_status = O2NET_ERR_NONE;
322         nsw->ns_status = 0;
323         return 0;
324 }
325 
326 static void o2net_complete_nsw_locked(struct o2net_node *nn,
327                                       struct o2net_status_wait *nsw,
328                                       enum o2net_system_error sys_status,
329                                       s32 status)
330 {
331         assert_spin_locked(&nn->nn_lock);
332 
333         if (!list_empty(&nsw->ns_node_item)) {
334                 list_del_init(&nsw->ns_node_item);
335                 nsw->ns_sys_status = sys_status;
336                 nsw->ns_status = status;
337                 idr_remove(&nn->nn_status_idr, nsw->ns_id);
338                 wake_up(&nsw->ns_wq);
339         }
340 }
341 
342 static void o2net_complete_nsw(struct o2net_node *nn,
343                                struct o2net_status_wait *nsw,
344                                u64 id, enum o2net_system_error sys_status,
345                                s32 status)
346 {
347         spin_lock(&nn->nn_lock);
348         if (nsw == NULL) {
349                 if (id > INT_MAX)
350                         goto out;
351 
352                 nsw = idr_find(&nn->nn_status_idr, id);
353                 if (nsw == NULL)
354                         goto out;
355         }
356 
357         o2net_complete_nsw_locked(nn, nsw, sys_status, status);
358 
359 out:
360         spin_unlock(&nn->nn_lock);
361         return;
362 }
363 
364 static void o2net_complete_nodes_nsw(struct o2net_node *nn)
365 {
366         struct o2net_status_wait *nsw, *tmp;
367         unsigned int num_kills = 0;
368 
369         assert_spin_locked(&nn->nn_lock);
370 
371         list_for_each_entry_safe(nsw, tmp, &nn->nn_status_list, ns_node_item) {
372                 o2net_complete_nsw_locked(nn, nsw, O2NET_ERR_DIED, 0);
373                 num_kills++;
374         }
375 
376         mlog(0, "completed %d messages for node %u\n", num_kills,
377              o2net_num_from_nn(nn));
378 }
379 
380 static int o2net_nsw_completed(struct o2net_node *nn,
381                                struct o2net_status_wait *nsw)
382 {
383         int completed;
384         spin_lock(&nn->nn_lock);
385         completed = list_empty(&nsw->ns_node_item);
386         spin_unlock(&nn->nn_lock);
387         return completed;
388 }
389 
390 /* ------------------------------------------------------------ */
391 
392 static void sc_kref_release(struct kref *kref)
393 {
394         struct o2net_sock_container *sc = container_of(kref,
395                                         struct o2net_sock_container, sc_kref);
396         BUG_ON(timer_pending(&sc->sc_idle_timeout));
397 
398         sclog(sc, "releasing\n");
399 
400         if (sc->sc_sock) {
401                 sock_release(sc->sc_sock);
402                 sc->sc_sock = NULL;
403         }
404 
405         o2nm_undepend_item(&sc->sc_node->nd_item);
406         o2nm_node_put(sc->sc_node);
407         sc->sc_node = NULL;
408 
409         o2net_debug_del_sc(sc);
410 
411         if (sc->sc_page)
412                 __free_page(sc->sc_page);
413         kfree(sc);
414 }
415 
416 static void sc_put(struct o2net_sock_container *sc)
417 {
418         sclog(sc, "put\n");
419         kref_put(&sc->sc_kref, sc_kref_release);
420 }
421 static void sc_get(struct o2net_sock_container *sc)
422 {
423         sclog(sc, "get\n");
424         kref_get(&sc->sc_kref);
425 }
426 static struct o2net_sock_container *sc_alloc(struct o2nm_node *node)
427 {
428         struct o2net_sock_container *sc, *ret = NULL;
429         struct page *page = NULL;
430         int status = 0;
431 
432         page = alloc_page(GFP_NOFS);
433         sc = kzalloc(sizeof(*sc), GFP_NOFS);
434         if (sc == NULL || page == NULL)
435                 goto out;
436 
437         kref_init(&sc->sc_kref);
438         o2nm_node_get(node);
439         sc->sc_node = node;
440 
441         /* pin the node item of the remote node */
442         status = o2nm_depend_item(&node->nd_item);
443         if (status) {
444                 mlog_errno(status);
445                 o2nm_node_put(node);
446                 goto out;
447         }
448         INIT_WORK(&sc->sc_connect_work, o2net_sc_connect_completed);
449         INIT_WORK(&sc->sc_rx_work, o2net_rx_until_empty);
450         INIT_WORK(&sc->sc_shutdown_work, o2net_shutdown_sc);
451         INIT_DELAYED_WORK(&sc->sc_keepalive_work, o2net_sc_send_keep_req);
452 
453         timer_setup(&sc->sc_idle_timeout, o2net_idle_timer, 0);
454 
455         sclog(sc, "alloced\n");
456 
457         ret = sc;
458         sc->sc_page = page;
459         o2net_debug_add_sc(sc);
460         sc = NULL;
461         page = NULL;
462 
463 out:
464         if (page)
465                 __free_page(page);
466         kfree(sc);
467 
468         return ret;
469 }
470 
471 /* ------------------------------------------------------------ */
472 
473 static void o2net_sc_queue_work(struct o2net_sock_container *sc,
474                                 struct work_struct *work)
475 {
476         sc_get(sc);
477         if (!queue_work(o2net_wq, work))
478                 sc_put(sc);
479 }
480 static void o2net_sc_queue_delayed_work(struct o2net_sock_container *sc,
481                                         struct delayed_work *work,
482                                         int delay)
483 {
484         sc_get(sc);
485         if (!queue_delayed_work(o2net_wq, work, delay))
486                 sc_put(sc);
487 }
488 static void o2net_sc_cancel_delayed_work(struct o2net_sock_container *sc,
489                                          struct delayed_work *work)
490 {
491         if (cancel_delayed_work(work))
492                 sc_put(sc);
493 }
494 
495 static atomic_t o2net_connected_peers = ATOMIC_INIT(0);
496 
497 int o2net_num_connected_peers(void)
498 {
499         return atomic_read(&o2net_connected_peers);
500 }
501 
502 static void o2net_set_nn_state(struct o2net_node *nn,
503                                struct o2net_sock_container *sc,
504                                unsigned valid, int err)
505 {
506         int was_valid = nn->nn_sc_valid;
507         int was_err = nn->nn_persistent_error;
508         struct o2net_sock_container *old_sc = nn->nn_sc;
509 
510         assert_spin_locked(&nn->nn_lock);
511 
512         if (old_sc && !sc)
513                 atomic_dec(&o2net_connected_peers);
514         else if (!old_sc && sc)
515                 atomic_inc(&o2net_connected_peers);
516 
517         /* the node num comparison and single connect/accept path should stop
518          * an non-null sc from being overwritten with another */
519         BUG_ON(sc && nn->nn_sc && nn->nn_sc != sc);
520         mlog_bug_on_msg(err && valid, "err %d valid %u\n", err, valid);
521         mlog_bug_on_msg(valid && !sc, "valid %u sc %p\n", valid, sc);
522 
523         if (was_valid && !valid && err == 0)
524                 err = -ENOTCONN;
525 
526         mlog(ML_CONN, "node %u sc: %p -> %p, valid %u -> %u, err %d -> %d\n",
527              o2net_num_from_nn(nn), nn->nn_sc, sc, nn->nn_sc_valid, valid,
528              nn->nn_persistent_error, err);
529 
530         nn->nn_sc = sc;
531         nn->nn_sc_valid = valid ? 1 : 0;
532         nn->nn_persistent_error = err;
533 
534         /* mirrors o2net_tx_can_proceed() */
535         if (nn->nn_persistent_error || nn->nn_sc_valid)
536                 wake_up(&nn->nn_sc_wq);
537 
538         if (was_valid && !was_err && nn->nn_persistent_error) {
539                 o2quo_conn_err(o2net_num_from_nn(nn));
540                 queue_delayed_work(o2net_wq, &nn->nn_still_up,
541                                    msecs_to_jiffies(O2NET_QUORUM_DELAY_MS));
542         }
543 
544         if (was_valid && !valid) {
545                 if (old_sc)
546                         printk(KERN_NOTICE "o2net: No longer connected to "
547                                 SC_NODEF_FMT "\n", SC_NODEF_ARGS(old_sc));
548                 o2net_complete_nodes_nsw(nn);
549         }
550 
551         if (!was_valid && valid) {
552                 o2quo_conn_up(o2net_num_from_nn(nn));
553                 cancel_delayed_work(&nn->nn_connect_expired);
554                 printk(KERN_NOTICE "o2net: %s " SC_NODEF_FMT "\n",
555                        o2nm_this_node() > sc->sc_node->nd_num ?
556                        "Connected to" : "Accepted connection from",
557                        SC_NODEF_ARGS(sc));
558         }
559 
560         /* trigger the connecting worker func as long as we're not valid,
561          * it will back off if it shouldn't connect.  This can be called
562          * from node config teardown and so needs to be careful about
563          * the work queue actually being up. */
564         if (!valid && o2net_wq) {
565                 unsigned long delay;
566                 /* delay if we're within a RECONNECT_DELAY of the
567                  * last attempt */
568                 delay = (nn->nn_last_connect_attempt +
569                          msecs_to_jiffies(o2net_reconnect_delay()))
570                         - jiffies;
571                 if (delay > msecs_to_jiffies(o2net_reconnect_delay()))
572                         delay = 0;
573                 mlog(ML_CONN, "queueing conn attempt in %lu jiffies\n", delay);
574                 queue_delayed_work(o2net_wq, &nn->nn_connect_work, delay);
575 
576                 /*
577                  * Delay the expired work after idle timeout.
578                  *
579                  * We might have lots of failed connection attempts that run
580                  * through here but we only cancel the connect_expired work when
581                  * a connection attempt succeeds.  So only the first enqueue of
582                  * the connect_expired work will do anything.  The rest will see
583                  * that it's already queued and do nothing.
584                  */
585                 delay += msecs_to_jiffies(o2net_idle_timeout());
586                 queue_delayed_work(o2net_wq, &nn->nn_connect_expired, delay);
587         }
588 
589         /* keep track of the nn's sc ref for the caller */
590         if ((old_sc == NULL) && sc)
591                 sc_get(sc);
592         if (old_sc && (old_sc != sc)) {
593                 o2net_sc_queue_work(old_sc, &old_sc->sc_shutdown_work);
594                 sc_put(old_sc);
595         }
596 }
597 
598 /* see o2net_register_callbacks() */
599 static void o2net_data_ready(struct sock *sk)
600 {
601         void (*ready)(struct sock *sk);
602         struct o2net_sock_container *sc;
603 
604         read_lock_bh(&sk->sk_callback_lock);
605         sc = sk->sk_user_data;
606         if (sc) {
607                 sclog(sc, "data_ready hit\n");
608                 o2net_set_data_ready_time(sc);
609                 o2net_sc_queue_work(sc, &sc->sc_rx_work);
610                 ready = sc->sc_data_ready;
611         } else {
612                 ready = sk->sk_data_ready;
613         }
614         read_unlock_bh(&sk->sk_callback_lock);
615 
616         ready(sk);
617 }
618 
619 /* see o2net_register_callbacks() */
620 static void o2net_state_change(struct sock *sk)
621 {
622         void (*state_change)(struct sock *sk);
623         struct o2net_sock_container *sc;
624 
625         read_lock_bh(&sk->sk_callback_lock);
626         sc = sk->sk_user_data;
627         if (sc == NULL) {
628                 state_change = sk->sk_state_change;
629                 goto out;
630         }
631 
632         sclog(sc, "state_change to %d\n", sk->sk_state);
633 
634         state_change = sc->sc_state_change;
635 
636         switch(sk->sk_state) {
637         /* ignore connecting sockets as they make progress */
638         case TCP_SYN_SENT:
639         case TCP_SYN_RECV:
640                 break;
641         case TCP_ESTABLISHED:
642                 o2net_sc_queue_work(sc, &sc->sc_connect_work);
643                 break;
644         default:
645                 printk(KERN_INFO "o2net: Connection to " SC_NODEF_FMT
646                         " shutdown, state %d\n",
647                         SC_NODEF_ARGS(sc), sk->sk_state);
648                 o2net_sc_queue_work(sc, &sc->sc_shutdown_work);
649                 break;
650         }
651 out:
652         read_unlock_bh(&sk->sk_callback_lock);
653         state_change(sk);
654 }
655 
656 /*
657  * we register callbacks so we can queue work on events before calling
658  * the original callbacks.  our callbacks our careful to test user_data
659  * to discover when they've reaced with o2net_unregister_callbacks().
660  */
661 static void o2net_register_callbacks(struct sock *sk,
662                                      struct o2net_sock_container *sc)
663 {
664         write_lock_bh(&sk->sk_callback_lock);
665 
666         /* accepted sockets inherit the old listen socket data ready */
667         if (sk->sk_data_ready == o2net_listen_data_ready) {
668                 sk->sk_data_ready = sk->sk_user_data;
669                 sk->sk_user_data = NULL;
670         }
671 
672         BUG_ON(sk->sk_user_data != NULL);
673         sk->sk_user_data = sc;
674         sc_get(sc);
675 
676         sc->sc_data_ready = sk->sk_data_ready;
677         sc->sc_state_change = sk->sk_state_change;
678         sk->sk_data_ready = o2net_data_ready;
679         sk->sk_state_change = o2net_state_change;
680 
681         mutex_init(&sc->sc_send_lock);
682 
683         write_unlock_bh(&sk->sk_callback_lock);
684 }
685 
686 static int o2net_unregister_callbacks(struct sock *sk,
687                                    struct o2net_sock_container *sc)
688 {
689         int ret = 0;
690 
691         write_lock_bh(&sk->sk_callback_lock);
692         if (sk->sk_user_data == sc) {
693                 ret = 1;
694                 sk->sk_user_data = NULL;
695                 sk->sk_data_ready = sc->sc_data_ready;
696                 sk->sk_state_change = sc->sc_state_change;
697         }
698         write_unlock_bh(&sk->sk_callback_lock);
699 
700         return ret;
701 }
702 
703 /*
704  * this is a little helper that is called by callers who have seen a problem
705  * with an sc and want to detach it from the nn if someone already hasn't beat
706  * them to it.  if an error is given then the shutdown will be persistent
707  * and pending transmits will be canceled.
708  */
709 static void o2net_ensure_shutdown(struct o2net_node *nn,
710                                    struct o2net_sock_container *sc,
711                                    int err)
712 {
713         spin_lock(&nn->nn_lock);
714         if (nn->nn_sc == sc)
715                 o2net_set_nn_state(nn, NULL, 0, err);
716         spin_unlock(&nn->nn_lock);
717 }
718 
719 /*
720  * This work queue function performs the blocking parts of socket shutdown.  A
721  * few paths lead here.  set_nn_state will trigger this callback if it sees an
722  * sc detached from the nn.  state_change will also trigger this callback
723  * directly when it sees errors.  In that case we need to call set_nn_state
724  * ourselves as state_change couldn't get the nn_lock and call set_nn_state
725  * itself.
726  */
727 static void o2net_shutdown_sc(struct work_struct *work)
728 {
729         struct o2net_sock_container *sc =
730                 container_of(work, struct o2net_sock_container,
731                              sc_shutdown_work);
732         struct o2net_node *nn = o2net_nn_from_num(sc->sc_node->nd_num);
733 
734         sclog(sc, "shutting down\n");
735 
736         /* drop the callbacks ref and call shutdown only once */
737         if (o2net_unregister_callbacks(sc->sc_sock->sk, sc)) {
738                 /* we shouldn't flush as we're in the thread, the
739                  * races with pending sc work structs are harmless */
740                 del_timer_sync(&sc->sc_idle_timeout);
741                 o2net_sc_cancel_delayed_work(sc, &sc->sc_keepalive_work);
742                 sc_put(sc);
743                 kernel_sock_shutdown(sc->sc_sock, SHUT_RDWR);
744         }
745 
746         /* not fatal so failed connects before the other guy has our
747          * heartbeat can be retried */
748         o2net_ensure_shutdown(nn, sc, 0);
749         sc_put(sc);
750 }
751 
752 /* ------------------------------------------------------------ */
753 
754 static int o2net_handler_cmp(struct o2net_msg_handler *nmh, u32 msg_type,
755                              u32 key)
756 {
757         int ret = memcmp(&nmh->nh_key, &key, sizeof(key));
758 
759         if (ret == 0)
760                 ret = memcmp(&nmh->nh_msg_type, &msg_type, sizeof(msg_type));
761 
762         return ret;
763 }
764 
765 static struct o2net_msg_handler *
766 o2net_handler_tree_lookup(u32 msg_type, u32 key, struct rb_node ***ret_p,
767                           struct rb_node **ret_parent)
768 {
769         struct rb_node **p = &o2net_handler_tree.rb_node;
770         struct rb_node *parent = NULL;
771         struct o2net_msg_handler *nmh, *ret = NULL;
772         int cmp;
773 
774         while (*p) {
775                 parent = *p;
776                 nmh = rb_entry(parent, struct o2net_msg_handler, nh_node);
777                 cmp = o2net_handler_cmp(nmh, msg_type, key);
778 
779                 if (cmp < 0)
780                         p = &(*p)->rb_left;
781                 else if (cmp > 0)
782                         p = &(*p)->rb_right;
783                 else {
784                         ret = nmh;
785                         break;
786                 }
787         }
788 
789         if (ret_p != NULL)
790                 *ret_p = p;
791         if (ret_parent != NULL)
792                 *ret_parent = parent;
793 
794         return ret;
795 }
796 
797 static void o2net_handler_kref_release(struct kref *kref)
798 {
799         struct o2net_msg_handler *nmh;
800         nmh = container_of(kref, struct o2net_msg_handler, nh_kref);
801 
802         kfree(nmh);
803 }
804 
805 static void o2net_handler_put(struct o2net_msg_handler *nmh)
806 {
807         kref_put(&nmh->nh_kref, o2net_handler_kref_release);
808 }
809 
810 /* max_len is protection for the handler func.  incoming messages won't
811  * be given to the handler if their payload is longer than the max. */
812 int o2net_register_handler(u32 msg_type, u32 key, u32 max_len,
813                            o2net_msg_handler_func *func, void *data,
814                            o2net_post_msg_handler_func *post_func,
815                            struct list_head *unreg_list)
816 {
817         struct o2net_msg_handler *nmh = NULL;
818         struct rb_node **p, *parent;
819         int ret = 0;
820 
821         if (max_len > O2NET_MAX_PAYLOAD_BYTES) {
822                 mlog(0, "max_len for message handler out of range: %u\n",
823                         max_len);
824                 ret = -EINVAL;
825                 goto out;
826         }
827 
828         if (!msg_type) {
829                 mlog(0, "no message type provided: %u, %p\n", msg_type, func);
830                 ret = -EINVAL;
831                 goto out;
832 
833         }
834         if (!func) {
835                 mlog(0, "no message handler provided: %u, %p\n",
836                        msg_type, func);
837                 ret = -EINVAL;
838                 goto out;
839         }
840 
841         nmh = kzalloc(sizeof(struct o2net_msg_handler), GFP_NOFS);
842         if (nmh == NULL) {
843                 ret = -ENOMEM;
844                 goto out;
845         }
846 
847         nmh->nh_func = func;
848         nmh->nh_func_data = data;
849         nmh->nh_post_func = post_func;
850         nmh->nh_msg_type = msg_type;
851         nmh->nh_max_len = max_len;
852         nmh->nh_key = key;
853         /* the tree and list get this ref.. they're both removed in
854          * unregister when this ref is dropped */
855         kref_init(&nmh->nh_kref);
856         INIT_LIST_HEAD(&nmh->nh_unregister_item);
857 
858         write_lock(&o2net_handler_lock);
859         if (o2net_handler_tree_lookup(msg_type, key, &p, &parent))
860                 ret = -EEXIST;
861         else {
862                 rb_link_node(&nmh->nh_node, parent, p);
863                 rb_insert_color(&nmh->nh_node, &o2net_handler_tree);
864                 list_add_tail(&nmh->nh_unregister_item, unreg_list);
865 
866                 mlog(ML_TCP, "registered handler func %p type %u key %08x\n",
867                      func, msg_type, key);
868                 /* we've had some trouble with handlers seemingly vanishing. */
869                 mlog_bug_on_msg(o2net_handler_tree_lookup(msg_type, key, &p,
870                                                           &parent) == NULL,
871                                 "couldn't find handler we *just* registered "
872                                 "for type %u key %08x\n", msg_type, key);
873         }
874         write_unlock(&o2net_handler_lock);
875         if (ret)
876                 goto out;
877 
878 out:
879         if (ret)
880                 kfree(nmh);
881 
882         return ret;
883 }
884 EXPORT_SYMBOL_GPL(o2net_register_handler);
885 
886 void o2net_unregister_handler_list(struct list_head *list)
887 {
888         struct o2net_msg_handler *nmh, *n;
889 
890         write_lock(&o2net_handler_lock);
891         list_for_each_entry_safe(nmh, n, list, nh_unregister_item) {
892                 mlog(ML_TCP, "unregistering handler func %p type %u key %08x\n",
893                      nmh->nh_func, nmh->nh_msg_type, nmh->nh_key);
894                 rb_erase(&nmh->nh_node, &o2net_handler_tree);
895                 list_del_init(&nmh->nh_unregister_item);
896                 kref_put(&nmh->nh_kref, o2net_handler_kref_release);
897         }
898         write_unlock(&o2net_handler_lock);
899 }
900 EXPORT_SYMBOL_GPL(o2net_unregister_handler_list);
901 
902 static struct o2net_msg_handler *o2net_handler_get(u32 msg_type, u32 key)
903 {
904         struct o2net_msg_handler *nmh;
905 
906         read_lock(&o2net_handler_lock);
907         nmh = o2net_handler_tree_lookup(msg_type, key, NULL, NULL);
908         if (nmh)
909                 kref_get(&nmh->nh_kref);
910         read_unlock(&o2net_handler_lock);
911 
912         return nmh;
913 }
914 
915 /* ------------------------------------------------------------ */
916 
917 static int o2net_recv_tcp_msg(struct socket *sock, void *data, size_t len)
918 {
919         struct kvec vec = { .iov_len = len, .iov_base = data, };
920         struct msghdr msg = { .msg_flags = MSG_DONTWAIT, };
921         return kernel_recvmsg(sock, &msg, &vec, 1, len, msg.msg_flags);
922 }
923 
924 static int o2net_send_tcp_msg(struct socket *sock, struct kvec *vec,
925                               size_t veclen, size_t total)
926 {
927         int ret;
928         struct msghdr msg = {.msg_flags = 0,};
929 
930         if (sock == NULL) {
931                 ret = -EINVAL;
932                 goto out;
933         }
934 
935         ret = kernel_sendmsg(sock, &msg, vec, veclen, total);
936         if (likely(ret == total))
937                 return 0;
938         mlog(ML_ERROR, "sendmsg returned %d instead of %zu\n", ret, total);
939         if (ret >= 0)
940                 ret = -EPIPE; /* should be smarter, I bet */
941 out:
942         mlog(0, "returning error: %d\n", ret);
943         return ret;
944 }
945 
946 static void o2net_sendpage(struct o2net_sock_container *sc,
947                            void *kmalloced_virt,
948                            size_t size)
949 {
950         struct o2net_node *nn = o2net_nn_from_num(sc->sc_node->nd_num);
951         ssize_t ret;
952 
953         while (1) {
954                 mutex_lock(&sc->sc_send_lock);
955                 ret = sc->sc_sock->ops->sendpage(sc->sc_sock,
956                                                  virt_to_page(kmalloced_virt),
957                                                  offset_in_page(kmalloced_virt),
958                                                  size, MSG_DONTWAIT);
959                 mutex_unlock(&sc->sc_send_lock);
960                 if (ret == size)
961                         break;
962                 if (ret == (ssize_t)-EAGAIN) {
963                         mlog(0, "sendpage of size %zu to " SC_NODEF_FMT
964                              " returned EAGAIN\n", size, SC_NODEF_ARGS(sc));
965                         cond_resched();
966                         continue;
967                 }
968                 mlog(ML_ERROR, "sendpage of size %zu to " SC_NODEF_FMT
969                      " failed with %zd\n", size, SC_NODEF_ARGS(sc), ret);
970                 o2net_ensure_shutdown(nn, sc, 0);
971                 break;
972         }
973 }
974 
975 static void o2net_init_msg(struct o2net_msg *msg, u16 data_len, u16 msg_type, u32 key)
976 {
977         memset(msg, 0, sizeof(struct o2net_msg));
978         msg->magic = cpu_to_be16(O2NET_MSG_MAGIC);
979         msg->data_len = cpu_to_be16(data_len);
980         msg->msg_type = cpu_to_be16(msg_type);
981         msg->sys_status = cpu_to_be32(O2NET_ERR_NONE);
982         msg->status = 0;
983         msg->key = cpu_to_be32(key);
984 }
985 
986 static int o2net_tx_can_proceed(struct o2net_node *nn,
987                                 struct o2net_sock_container **sc_ret,
988                                 int *error)
989 {
990         int ret = 0;
991 
992         spin_lock(&nn->nn_lock);
993         if (nn->nn_persistent_error) {
994                 ret = 1;
995                 *sc_ret = NULL;
996                 *error = nn->nn_persistent_error;
997         } else if (nn->nn_sc_valid) {
998                 kref_get(&nn->nn_sc->sc_kref);
999 
1000                 ret = 1;
1001                 *sc_ret = nn->nn_sc;
1002                 *error = 0;
1003         }
1004         spin_unlock(&nn->nn_lock);
1005 
1006         return ret;
1007 }
1008 
1009 /* Get a map of all nodes to which this node is currently connected to */
1010 void o2net_fill_node_map(unsigned long *map, unsigned bytes)
1011 {
1012         struct o2net_sock_container *sc;
1013         int node, ret;
1014 
1015         BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long)));
1016 
1017         memset(map, 0, bytes);
1018         for (node = 0; node < O2NM_MAX_NODES; ++node) {
1019                 if (!o2net_tx_can_proceed(o2net_nn_from_num(node), &sc, &ret))
1020                         continue;
1021                 if (!ret) {
1022                         set_bit(node, map);
1023                         sc_put(sc);
1024                 }
1025         }
1026 }
1027 EXPORT_SYMBOL_GPL(o2net_fill_node_map);
1028 
1029 int o2net_send_message_vec(u32 msg_type, u32 key, struct kvec *caller_vec,
1030                            size_t caller_veclen, u8 target_node, int *status)
1031 {
1032         int ret = 0;
1033         struct o2net_msg *msg = NULL;
1034         size_t veclen, caller_bytes = 0;
1035         struct kvec *vec = NULL;
1036         struct o2net_sock_container *sc = NULL;
1037         struct o2net_node *nn = o2net_nn_from_num(target_node);
1038         struct o2net_status_wait nsw = {
1039                 .ns_node_item = LIST_HEAD_INIT(nsw.ns_node_item),
1040         };
1041         struct o2net_send_tracking nst;
1042 
1043         o2net_init_nst(&nst, msg_type, key, current, target_node);
1044 
1045         if (o2net_wq == NULL) {
1046                 mlog(0, "attempt to tx without o2netd running\n");
1047                 ret = -ESRCH;
1048                 goto out;
1049         }
1050 
1051         if (caller_veclen == 0) {
1052                 mlog(0, "bad kvec array length\n");
1053                 ret = -EINVAL;
1054                 goto out;
1055         }
1056 
1057         caller_bytes = iov_length((struct iovec *)caller_vec, caller_veclen);
1058         if (caller_bytes > O2NET_MAX_PAYLOAD_BYTES) {
1059                 mlog(0, "total payload len %zu too large\n", caller_bytes);
1060                 ret = -EINVAL;
1061                 goto out;
1062         }
1063 
1064         if (target_node == o2nm_this_node()) {
1065                 ret = -ELOOP;
1066                 goto out;
1067         }
1068 
1069         o2net_debug_add_nst(&nst);
1070 
1071         o2net_set_nst_sock_time(&nst);
1072 
1073         wait_event(nn->nn_sc_wq, o2net_tx_can_proceed(nn, &sc, &ret));
1074         if (ret)
1075                 goto out;
1076 
1077         o2net_set_nst_sock_container(&nst, sc);
1078 
1079         veclen = caller_veclen + 1;
1080         vec = kmalloc(sizeof(struct kvec) * veclen, GFP_ATOMIC);
1081         if (vec == NULL) {
1082                 mlog(0, "failed to %zu element kvec!\n", veclen);
1083                 ret = -ENOMEM;
1084                 goto out;
1085         }
1086 
1087         msg = kmalloc(sizeof(struct o2net_msg), GFP_ATOMIC);
1088         if (!msg) {
1089                 mlog(0, "failed to allocate a o2net_msg!\n");
1090                 ret = -ENOMEM;
1091                 goto out;
1092         }
1093 
1094         o2net_init_msg(msg, caller_bytes, msg_type, key);
1095 
1096         vec[0].iov_len = sizeof(struct o2net_msg);
1097         vec[0].iov_base = msg;
1098         memcpy(&vec[1], caller_vec, caller_veclen * sizeof(struct kvec));
1099 
1100         ret = o2net_prep_nsw(nn, &nsw);
1101         if (ret)
1102                 goto out;
1103 
1104         msg->msg_num = cpu_to_be32(nsw.ns_id);
1105         o2net_set_nst_msg_id(&nst, nsw.ns_id);
1106 
1107         o2net_set_nst_send_time(&nst);
1108 
1109         /* finally, convert the message header to network byte-order
1110          * and send */
1111         mutex_lock(&sc->sc_send_lock);
1112         ret = o2net_send_tcp_msg(sc->sc_sock, vec, veclen,
1113                                  sizeof(struct o2net_msg) + caller_bytes);
1114         mutex_unlock(&sc->sc_send_lock);
1115         msglog(msg, "sending returned %d\n", ret);
1116         if (ret < 0) {
1117                 mlog(0, "error returned from o2net_send_tcp_msg=%d\n", ret);
1118                 goto out;
1119         }
1120 
1121         /* wait on other node's handler */
1122         o2net_set_nst_status_time(&nst);
1123         wait_event(nsw.ns_wq, o2net_nsw_completed(nn, &nsw));
1124 
1125         o2net_update_send_stats(&nst, sc);
1126 
1127         /* Note that we avoid overwriting the callers status return
1128          * variable if a system error was reported on the other
1129          * side. Callers beware. */
1130         ret = o2net_sys_err_to_errno(nsw.ns_sys_status);
1131         if (status && !ret)
1132                 *status = nsw.ns_status;
1133 
1134         mlog(0, "woken, returning system status %d, user status %d\n",
1135              ret, nsw.ns_status);
1136 out:
1137         o2net_debug_del_nst(&nst); /* must be before dropping sc and node */
1138         if (sc)
1139                 sc_put(sc);
1140         kfree(vec);
1141         kfree(msg);
1142         o2net_complete_nsw(nn, &nsw, 0, 0, 0);
1143         return ret;
1144 }
1145 EXPORT_SYMBOL_GPL(o2net_send_message_vec);
1146 
1147 int o2net_send_message(u32 msg_type, u32 key, void *data, u32 len,
1148                        u8 target_node, int *status)
1149 {
1150         struct kvec vec = {
1151                 .iov_base = data,
1152                 .iov_len = len,
1153         };
1154         return o2net_send_message_vec(msg_type, key, &vec, 1,
1155                                       target_node, status);
1156 }
1157 EXPORT_SYMBOL_GPL(o2net_send_message);
1158 
1159 static int o2net_send_status_magic(struct socket *sock, struct o2net_msg *hdr,
1160                                    enum o2net_system_error syserr, int err)
1161 {
1162         struct kvec vec = {
1163                 .iov_base = hdr,
1164                 .iov_len = sizeof(struct o2net_msg),
1165         };
1166 
1167         BUG_ON(syserr >= O2NET_ERR_MAX);
1168 
1169         /* leave other fields intact from the incoming message, msg_num
1170          * in particular */
1171         hdr->sys_status = cpu_to_be32(syserr);
1172         hdr->status = cpu_to_be32(err);
1173         hdr->magic = cpu_to_be16(O2NET_MSG_STATUS_MAGIC);  // twiddle the magic
1174         hdr->data_len = 0;
1175 
1176         msglog(hdr, "about to send status magic %d\n", err);
1177         /* hdr has been in host byteorder this whole time */
1178         return o2net_send_tcp_msg(sock, &vec, 1, sizeof(struct o2net_msg));
1179 }
1180 
1181 /* this returns -errno if the header was unknown or too large, etc.
1182  * after this is called the buffer us reused for the next message */
1183 static int o2net_process_message(struct o2net_sock_container *sc,
1184                                  struct o2net_msg *hdr)
1185 {
1186         struct o2net_node *nn = o2net_nn_from_num(sc->sc_node->nd_num);
1187         int ret = 0, handler_status;
1188         enum  o2net_system_error syserr;
1189         struct o2net_msg_handler *nmh = NULL;
1190         void *ret_data = NULL;
1191 
1192         msglog(hdr, "processing message\n");
1193 
1194         o2net_sc_postpone_idle(sc);
1195 
1196         switch(be16_to_cpu(hdr->magic)) {
1197                 case O2NET_MSG_STATUS_MAGIC:
1198                         /* special type for returning message status */
1199                         o2net_complete_nsw(nn, NULL,
1200                                            be32_to_cpu(hdr->msg_num),
1201                                            be32_to_cpu(hdr->sys_status),
1202                                            be32_to_cpu(hdr->status));
1203                         goto out;
1204                 case O2NET_MSG_KEEP_REQ_MAGIC:
1205                         o2net_sendpage(sc, o2net_keep_resp,
1206                                        sizeof(*o2net_keep_resp));
1207                         goto out;
1208                 case O2NET_MSG_KEEP_RESP_MAGIC:
1209                         goto out;
1210                 case O2NET_MSG_MAGIC:
1211                         break;
1212                 default:
1213                         msglog(hdr, "bad magic\n");
1214                         ret = -EINVAL;
1215                         goto out;
1216                         break;
1217         }
1218 
1219         /* find a handler for it */
1220         handler_status = 0;
1221         nmh = o2net_handler_get(be16_to_cpu(hdr->msg_type),
1222                                 be32_to_cpu(hdr->key));
1223         if (!nmh) {
1224                 mlog(ML_TCP, "couldn't find handler for type %u key %08x\n",
1225                      be16_to_cpu(hdr->msg_type), be32_to_cpu(hdr->key));
1226                 syserr = O2NET_ERR_NO_HNDLR;
1227                 goto out_respond;
1228         }
1229 
1230         syserr = O2NET_ERR_NONE;
1231 
1232         if (be16_to_cpu(hdr->data_len) > nmh->nh_max_len)
1233                 syserr = O2NET_ERR_OVERFLOW;
1234 
1235         if (syserr != O2NET_ERR_NONE)
1236                 goto out_respond;
1237 
1238         o2net_set_func_start_time(sc);
1239         sc->sc_msg_key = be32_to_cpu(hdr->key);
1240         sc->sc_msg_type = be16_to_cpu(hdr->msg_type);
1241         handler_status = (nmh->nh_func)(hdr, sizeof(struct o2net_msg) +
1242                                              be16_to_cpu(hdr->data_len),
1243                                         nmh->nh_func_data, &ret_data);
1244         o2net_set_func_stop_time(sc);
1245 
1246         o2net_update_recv_stats(sc);
1247 
1248 out_respond:
1249         /* this destroys the hdr, so don't use it after this */
1250         mutex_lock(&sc->sc_send_lock);
1251         ret = o2net_send_status_magic(sc->sc_sock, hdr, syserr,
1252                                       handler_status);
1253         mutex_unlock(&sc->sc_send_lock);
1254         hdr = NULL;
1255         mlog(0, "sending handler status %d, syserr %d returned %d\n",
1256              handler_status, syserr, ret);
1257 
1258         if (nmh) {
1259                 BUG_ON(ret_data != NULL && nmh->nh_post_func == NULL);
1260                 if (nmh->nh_post_func)
1261                         (nmh->nh_post_func)(handler_status, nmh->nh_func_data,
1262                                             ret_data);
1263         }
1264 
1265 out:
1266         if (nmh)
1267                 o2net_handler_put(nmh);
1268         return ret;
1269 }
1270 
1271 static int o2net_check_handshake(struct o2net_sock_container *sc)
1272 {
1273         struct o2net_handshake *hand = page_address(sc->sc_page);
1274         struct o2net_node *nn = o2net_nn_from_num(sc->sc_node->nd_num);
1275 
1276         if (hand->protocol_version != cpu_to_be64(O2NET_PROTOCOL_VERSION)) {
1277                 printk(KERN_NOTICE "o2net: " SC_NODEF_FMT " Advertised net "
1278                        "protocol version %llu but %llu is required. "
1279                        "Disconnecting.\n", SC_NODEF_ARGS(sc),
1280                        (unsigned long long)be64_to_cpu(hand->protocol_version),
1281                        O2NET_PROTOCOL_VERSION);
1282 
1283                 /* don't bother reconnecting if its the wrong version. */
1284                 o2net_ensure_shutdown(nn, sc, -ENOTCONN);
1285                 return -1;
1286         }
1287 
1288         /*
1289          * Ensure timeouts are consistent with other nodes, otherwise
1290          * we can end up with one node thinking that the other must be down,
1291          * but isn't. This can ultimately cause corruption.
1292          */
1293         if (be32_to_cpu(hand->o2net_idle_timeout_ms) !=
1294                                 o2net_idle_timeout()) {
1295                 printk(KERN_NOTICE "o2net: " SC_NODEF_FMT " uses a network "
1296                        "idle timeout of %u ms, but we use %u ms locally. "
1297                        "Disconnecting.\n", SC_NODEF_ARGS(sc),
1298                        be32_to_cpu(hand->o2net_idle_timeout_ms),
1299                        o2net_idle_timeout());
1300                 o2net_ensure_shutdown(nn, sc, -ENOTCONN);
1301                 return -1;
1302         }
1303 
1304         if (be32_to_cpu(hand->o2net_keepalive_delay_ms) !=
1305                         o2net_keepalive_delay()) {
1306                 printk(KERN_NOTICE "o2net: " SC_NODEF_FMT " uses a keepalive "
1307                        "delay of %u ms, but we use %u ms locally. "
1308                        "Disconnecting.\n", SC_NODEF_ARGS(sc),
1309                        be32_to_cpu(hand->o2net_keepalive_delay_ms),
1310                        o2net_keepalive_delay());
1311                 o2net_ensure_shutdown(nn, sc, -ENOTCONN);
1312                 return -1;
1313         }
1314 
1315         if (be32_to_cpu(hand->o2hb_heartbeat_timeout_ms) !=
1316                         O2HB_MAX_WRITE_TIMEOUT_MS) {
1317                 printk(KERN_NOTICE "o2net: " SC_NODEF_FMT " uses a heartbeat "
1318                        "timeout of %u ms, but we use %u ms locally. "
1319                        "Disconnecting.\n", SC_NODEF_ARGS(sc),
1320                        be32_to_cpu(hand->o2hb_heartbeat_timeout_ms),
1321                        O2HB_MAX_WRITE_TIMEOUT_MS);
1322                 o2net_ensure_shutdown(nn, sc, -ENOTCONN);
1323                 return -1;
1324         }
1325 
1326         sc->sc_handshake_ok = 1;
1327 
1328         spin_lock(&nn->nn_lock);
1329         /* set valid and queue the idle timers only if it hasn't been
1330          * shut down already */
1331         if (nn->nn_sc == sc) {
1332                 o2net_sc_reset_idle_timer(sc);
1333                 atomic_set(&nn->nn_timeout, 0);
1334                 o2net_set_nn_state(nn, sc, 1, 0);
1335         }
1336         spin_unlock(&nn->nn_lock);
1337 
1338         /* shift everything up as though it wasn't there */
1339         sc->sc_page_off -= sizeof(struct o2net_handshake);
1340         if (sc->sc_page_off)
1341                 memmove(hand, hand + 1, sc->sc_page_off);
1342 
1343         return 0;
1344 }
1345 
1346 /* this demuxes the queued rx bytes into header or payload bits and calls
1347  * handlers as each full message is read off the socket.  it returns -error,
1348  * == 0 eof, or > 0 for progress made.*/
1349 static int o2net_advance_rx(struct o2net_sock_container *sc)
1350 {
1351         struct o2net_msg *hdr;
1352         int ret = 0;
1353         void *data;
1354         size_t datalen;
1355 
1356         sclog(sc, "receiving\n");
1357         o2net_set_advance_start_time(sc);
1358 
1359         if (unlikely(sc->sc_handshake_ok == 0)) {
1360                 if(sc->sc_page_off < sizeof(struct o2net_handshake)) {
1361                         data = page_address(sc->sc_page) + sc->sc_page_off;
1362                         datalen = sizeof(struct o2net_handshake) - sc->sc_page_off;
1363                         ret = o2net_recv_tcp_msg(sc->sc_sock, data, datalen);
1364                         if (ret > 0)
1365                                 sc->sc_page_off += ret;
1366                 }
1367 
1368                 if (sc->sc_page_off == sizeof(struct o2net_handshake)) {
1369                         o2net_check_handshake(sc);
1370                         if (unlikely(sc->sc_handshake_ok == 0))
1371                                 ret = -EPROTO;
1372                 }
1373                 goto out;
1374         }
1375 
1376         /* do we need more header? */
1377         if (sc->sc_page_off < sizeof(struct o2net_msg)) {
1378                 data = page_address(sc->sc_page) + sc->sc_page_off;
1379                 datalen = sizeof(struct o2net_msg) - sc->sc_page_off;
1380                 ret = o2net_recv_tcp_msg(sc->sc_sock, data, datalen);
1381                 if (ret > 0) {
1382                         sc->sc_page_off += ret;
1383                         /* only swab incoming here.. we can
1384                          * only get here once as we cross from
1385                          * being under to over */
1386                         if (sc->sc_page_off == sizeof(struct o2net_msg)) {
1387                                 hdr = page_address(sc->sc_page);
1388                                 if (be16_to_cpu(hdr->data_len) >
1389                                     O2NET_MAX_PAYLOAD_BYTES)
1390                                         ret = -EOVERFLOW;
1391                         }
1392                 }
1393                 if (ret <= 0)
1394                         goto out;
1395         }
1396 
1397         if (sc->sc_page_off < sizeof(struct o2net_msg)) {
1398                 /* oof, still don't have a header */
1399                 goto out;
1400         }
1401 
1402         /* this was swabbed above when we first read it */
1403         hdr = page_address(sc->sc_page);
1404 
1405         msglog(hdr, "at page_off %zu\n", sc->sc_page_off);
1406 
1407         /* do we need more payload? */
1408         if (sc->sc_page_off - sizeof(struct o2net_msg) < be16_to_cpu(hdr->data_len)) {
1409                 /* need more payload */
1410                 data = page_address(sc->sc_page) + sc->sc_page_off;
1411                 datalen = (sizeof(struct o2net_msg) + be16_to_cpu(hdr->data_len)) -
1412                           sc->sc_page_off;
1413                 ret = o2net_recv_tcp_msg(sc->sc_sock, data, datalen);
1414                 if (ret > 0)
1415                         sc->sc_page_off += ret;
1416                 if (ret <= 0)
1417                         goto out;
1418         }
1419 
1420         if (sc->sc_page_off - sizeof(struct o2net_msg) == be16_to_cpu(hdr->data_len)) {
1421                 /* we can only get here once, the first time we read
1422                  * the payload.. so set ret to progress if the handler
1423                  * works out. after calling this the message is toast */
1424                 ret = o2net_process_message(sc, hdr);
1425                 if (ret == 0)
1426                         ret = 1;
1427                 sc->sc_page_off = 0;
1428         }
1429 
1430 out:
1431         sclog(sc, "ret = %d\n", ret);
1432         o2net_set_advance_stop_time(sc);
1433         return ret;
1434 }
1435 
1436 /* this work func is triggerd by data ready.  it reads until it can read no
1437  * more.  it interprets 0, eof, as fatal.  if data_ready hits while we're doing
1438  * our work the work struct will be marked and we'll be called again. */
1439 static void o2net_rx_until_empty(struct work_struct *work)
1440 {
1441         struct o2net_sock_container *sc =
1442                 container_of(work, struct o2net_sock_container, sc_rx_work);
1443         int ret;
1444 
1445         do {
1446                 ret = o2net_advance_rx(sc);
1447         } while (ret > 0);
1448 
1449         if (ret <= 0 && ret != -EAGAIN) {
1450                 struct o2net_node *nn = o2net_nn_from_num(sc->sc_node->nd_num);
1451                 sclog(sc, "saw error %d, closing\n", ret);
1452                 /* not permanent so read failed handshake can retry */
1453                 o2net_ensure_shutdown(nn, sc, 0);
1454         }
1455 
1456         sc_put(sc);
1457 }
1458 
1459 static int o2net_set_nodelay(struct socket *sock)
1460 {
1461         int val = 1;
1462 
1463         return kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
1464                                     (void *)&val, sizeof(val));
1465 }
1466 
1467 static int o2net_set_usertimeout(struct socket *sock)
1468 {
1469         int user_timeout = O2NET_TCP_USER_TIMEOUT;
1470 
1471         return kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
1472                                 (void *)&user_timeout, sizeof(user_timeout));
1473 }
1474 
1475 static void o2net_initialize_handshake(void)
1476 {
1477         o2net_hand->o2hb_heartbeat_timeout_ms = cpu_to_be32(
1478                 O2HB_MAX_WRITE_TIMEOUT_MS);
1479         o2net_hand->o2net_idle_timeout_ms = cpu_to_be32(o2net_idle_timeout());
1480         o2net_hand->o2net_keepalive_delay_ms = cpu_to_be32(
1481                 o2net_keepalive_delay());
1482         o2net_hand->o2net_reconnect_delay_ms = cpu_to_be32(
1483                 o2net_reconnect_delay());
1484 }
1485 
1486 /* ------------------------------------------------------------ */
1487 
1488 /* called when a connect completes and after a sock is accepted.  the
1489  * rx path will see the response and mark the sc valid */
1490 static void o2net_sc_connect_completed(struct work_struct *work)
1491 {
1492         struct o2net_sock_container *sc =
1493                 container_of(work, struct o2net_sock_container,
1494                              sc_connect_work);
1495 
1496         mlog(ML_MSG, "sc sending handshake with ver %llu id %llx\n",
1497               (unsigned long long)O2NET_PROTOCOL_VERSION,
1498               (unsigned long long)be64_to_cpu(o2net_hand->connector_id));
1499 
1500         o2net_initialize_handshake();
1501         o2net_sendpage(sc, o2net_hand, sizeof(*o2net_hand));
1502         sc_put(sc);
1503 }
1504 
1505 /* this is called as a work_struct func. */
1506 static void o2net_sc_send_keep_req(struct work_struct *work)
1507 {
1508         struct o2net_sock_container *sc =
1509                 container_of(work, struct o2net_sock_container,
1510                              sc_keepalive_work.work);
1511 
1512         o2net_sendpage(sc, o2net_keep_req, sizeof(*o2net_keep_req));
1513         sc_put(sc);
1514 }
1515 
1516 /* socket shutdown does a del_timer_sync against this as it tears down.
1517  * we can't start this timer until we've got to the point in sc buildup
1518  * where shutdown is going to be involved */
1519 static void o2net_idle_timer(struct timer_list *t)
1520 {
1521         struct o2net_sock_container *sc = from_timer(sc, t, sc_idle_timeout);
1522         struct o2net_node *nn = o2net_nn_from_num(sc->sc_node->nd_num);
1523 #ifdef CONFIG_DEBUG_FS
1524         unsigned long msecs = ktime_to_ms(ktime_get()) -
1525                 ktime_to_ms(sc->sc_tv_timer);
1526 #else
1527         unsigned long msecs = o2net_idle_timeout();
1528 #endif
1529 
1530         printk(KERN_NOTICE "o2net: Connection to " SC_NODEF_FMT " has been "
1531                "idle for %lu.%lu secs.\n",
1532                SC_NODEF_ARGS(sc), msecs / 1000, msecs % 1000);
1533 
1534         /* idle timerout happen, don't shutdown the connection, but
1535          * make fence decision. Maybe the connection can recover before
1536          * the decision is made.
1537          */
1538         atomic_set(&nn->nn_timeout, 1);
1539         o2quo_conn_err(o2net_num_from_nn(nn));
1540         queue_delayed_work(o2net_wq, &nn->nn_still_up,
1541                         msecs_to_jiffies(O2NET_QUORUM_DELAY_MS));
1542 
1543         o2net_sc_reset_idle_timer(sc);
1544 
1545 }
1546 
1547 static void o2net_sc_reset_idle_timer(struct o2net_sock_container *sc)
1548 {
1549         o2net_sc_cancel_delayed_work(sc, &sc->sc_keepalive_work);
1550         o2net_sc_queue_delayed_work(sc, &sc->sc_keepalive_work,
1551                       msecs_to_jiffies(o2net_keepalive_delay()));
1552         o2net_set_sock_timer(sc);
1553         mod_timer(&sc->sc_idle_timeout,
1554                jiffies + msecs_to_jiffies(o2net_idle_timeout()));
1555 }
1556 
1557 static void o2net_sc_postpone_idle(struct o2net_sock_container *sc)
1558 {
1559         struct o2net_node *nn = o2net_nn_from_num(sc->sc_node->nd_num);
1560 
1561         /* clear fence decision since the connection recover from timeout*/
1562         if (atomic_read(&nn->nn_timeout)) {
1563                 o2quo_conn_up(o2net_num_from_nn(nn));
1564                 cancel_delayed_work(&nn->nn_still_up);
1565                 atomic_set(&nn->nn_timeout, 0);
1566         }
1567 
1568         /* Only push out an existing timer */
1569         if (timer_pending(&sc->sc_idle_timeout))
1570                 o2net_sc_reset_idle_timer(sc);
1571 }
1572 
1573 /* this work func is kicked whenever a path sets the nn state which doesn't
1574  * have valid set.  This includes seeing hb come up, losing a connection,
1575  * having a connect attempt fail, etc. This centralizes the logic which decides
1576  * if a connect attempt should be made or if we should give up and all future
1577  * transmit attempts should fail */
1578 static void o2net_start_connect(struct work_struct *work)
1579 {
1580         struct o2net_node *nn =
1581                 container_of(work, struct o2net_node, nn_connect_work.work);
1582         struct o2net_sock_container *sc = NULL;
1583         struct o2nm_node *node = NULL, *mynode = NULL;
1584         struct socket *sock = NULL;
1585         struct sockaddr_in myaddr = {0, }, remoteaddr = {0, };
1586         int ret = 0, stop;
1587         unsigned int timeout;
1588         unsigned int noio_flag;
1589 
1590         /*
1591          * sock_create allocates the sock with GFP_KERNEL. We must set
1592          * per-process flag PF_MEMALLOC_NOIO so that all allocations done
1593          * by this process are done as if GFP_NOIO was specified. So we
1594          * are not reentering filesystem while doing memory reclaim.
1595          */
1596         noio_flag = memalloc_noio_save();
1597         /* if we're greater we initiate tx, otherwise we accept */
1598         if (o2nm_this_node() <= o2net_num_from_nn(nn))
1599                 goto out;
1600 
1601         /* watch for racing with tearing a node down */
1602         node = o2nm_get_node_by_num(o2net_num_from_nn(nn));
1603         if (node == NULL)
1604                 goto out;
1605 
1606         mynode = o2nm_get_node_by_num(o2nm_this_node());
1607         if (mynode == NULL)
1608                 goto out;
1609 
1610         spin_lock(&nn->nn_lock);
1611         /*
1612          * see if we already have one pending or have given up.
1613          * For nn_timeout, it is set when we close the connection
1614          * because of the idle time out. So it means that we have
1615          * at least connected to that node successfully once,
1616          * now try to connect to it again.
1617          */
1618         timeout = atomic_read(&nn->nn_timeout);
1619         stop = (nn->nn_sc ||
1620                 (nn->nn_persistent_error &&
1621                 (nn->nn_persistent_error != -ENOTCONN || timeout == 0)));
1622         spin_unlock(&nn->nn_lock);
1623         if (stop)
1624                 goto out;
1625 
1626         nn->nn_last_connect_attempt = jiffies;
1627 
1628         sc = sc_alloc(node);
1629         if (sc == NULL) {
1630                 mlog(0, "couldn't allocate sc\n");
1631                 ret = -ENOMEM;
1632                 goto out;
1633         }
1634 
1635         ret = sock_create(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock);
1636         if (ret < 0) {
1637                 mlog(0, "can't create socket: %d\n", ret);
1638                 goto out;
1639         }
1640         sc->sc_sock = sock; /* freed by sc_kref_release */
1641 
1642         sock->sk->sk_allocation = GFP_ATOMIC;
1643 
1644         myaddr.sin_family = AF_INET;
1645         myaddr.sin_addr.s_addr = mynode->nd_ipv4_address;
1646         myaddr.sin_port = htons(0); /* any port */
1647 
1648         ret = sock->ops->bind(sock, (struct sockaddr *)&myaddr,
1649                               sizeof(myaddr));
1650         if (ret) {
1651                 mlog(ML_ERROR, "bind failed with %d at address %pI4\n",
1652                      ret, &mynode->nd_ipv4_address);
1653                 goto out;
1654         }
1655 
1656         ret = o2net_set_nodelay(sc->sc_sock);
1657         if (ret) {
1658                 mlog(ML_ERROR, "setting TCP_NODELAY failed with %d\n", ret);
1659                 goto out;
1660         }
1661 
1662         ret = o2net_set_usertimeout(sock);
1663         if (ret) {
1664                 mlog(ML_ERROR, "set TCP_USER_TIMEOUT failed with %d\n", ret);
1665                 goto out;
1666         }
1667 
1668         o2net_register_callbacks(sc->sc_sock->sk, sc);
1669 
1670         spin_lock(&nn->nn_lock);
1671         /* handshake completion will set nn->nn_sc_valid */
1672         o2net_set_nn_state(nn, sc, 0, 0);
1673         spin_unlock(&nn->nn_lock);
1674 
1675         remoteaddr.sin_family = AF_INET;
1676         remoteaddr.sin_addr.s_addr = node->nd_ipv4_address;
1677         remoteaddr.sin_port = node->nd_ipv4_port;
1678 
1679         ret = sc->sc_sock->ops->connect(sc->sc_sock,
1680                                         (struct sockaddr *)&remoteaddr,
1681                                         sizeof(remoteaddr),
1682                                         O_NONBLOCK);
1683         if (ret == -EINPROGRESS)
1684                 ret = 0;
1685 
1686 out:
1687         if (ret && sc) {
1688                 printk(KERN_NOTICE "o2net: Connect attempt to " SC_NODEF_FMT
1689                        " failed with errno %d\n", SC_NODEF_ARGS(sc), ret);
1690                 /* 0 err so that another will be queued and attempted
1691                  * from set_nn_state */
1692                 o2net_ensure_shutdown(nn, sc, 0);
1693         }
1694         if (sc)
1695                 sc_put(sc);
1696         if (node)
1697                 o2nm_node_put(node);
1698         if (mynode)
1699                 o2nm_node_put(mynode);
1700 
1701         memalloc_noio_restore(noio_flag);
1702         return;
1703 }
1704 
1705 static void o2net_connect_expired(struct work_struct *work)
1706 {
1707         struct o2net_node *nn =
1708                 container_of(work, struct o2net_node, nn_connect_expired.work);
1709 
1710         spin_lock(&nn->nn_lock);
1711         if (!nn->nn_sc_valid) {
1712                 printk(KERN_NOTICE "o2net: No connection established with "
1713                        "node %u after %u.%u seconds, check network and"
1714                        " cluster configuration.\n",
1715                      o2net_num_from_nn(nn),
1716                      o2net_idle_timeout() / 1000,
1717                      o2net_idle_timeout() % 1000);
1718 
1719                 o2net_set_nn_state(nn, NULL, 0, 0);
1720         }
1721         spin_unlock(&nn->nn_lock);
1722 }
1723 
1724 static void o2net_still_up(struct work_struct *work)
1725 {
1726         struct o2net_node *nn =
1727                 container_of(work, struct o2net_node, nn_still_up.work);
1728 
1729         o2quo_hb_still_up(o2net_num_from_nn(nn));
1730 }
1731 
1732 /* ------------------------------------------------------------ */
1733 
1734 void o2net_disconnect_node(struct o2nm_node *node)
1735 {
1736         struct o2net_node *nn = o2net_nn_from_num(node->nd_num);
1737 
1738         /* don't reconnect until it's heartbeating again */
1739         spin_lock(&nn->nn_lock);
1740         atomic_set(&nn->nn_timeout, 0);
1741         o2net_set_nn_state(nn, NULL, 0, -ENOTCONN);
1742         spin_unlock(&nn->nn_lock);
1743 
1744         if (o2net_wq) {
1745                 cancel_delayed_work(&nn->nn_connect_expired);
1746                 cancel_delayed_work(&nn->nn_connect_work);
1747                 cancel_delayed_work(&nn->nn_still_up);
1748                 flush_workqueue(o2net_wq);
1749         }
1750 }
1751 
1752 static void o2net_hb_node_down_cb(struct o2nm_node *node, int node_num,
1753                                   void *data)
1754 {
1755         o2quo_hb_down(node_num);
1756 
1757         if (!node)
1758                 return;
1759 
1760         if (node_num != o2nm_this_node())
1761                 o2net_disconnect_node(node);
1762 
1763         BUG_ON(atomic_read(&o2net_connected_peers) < 0);
1764 }
1765 
1766 static void o2net_hb_node_up_cb(struct o2nm_node *node, int node_num,
1767                                 void *data)
1768 {
1769         struct o2net_node *nn = o2net_nn_from_num(node_num);
1770 
1771         o2quo_hb_up(node_num);
1772 
1773         BUG_ON(!node);
1774 
1775         /* ensure an immediate connect attempt */
1776         nn->nn_last_connect_attempt = jiffies -
1777                 (msecs_to_jiffies(o2net_reconnect_delay()) + 1);
1778 
1779         if (node_num != o2nm_this_node()) {
1780                 /* believe it or not, accept and node hearbeating testing
1781                  * can succeed for this node before we got here.. so
1782                  * only use set_nn_state to clear the persistent error
1783                  * if that hasn't already happened */
1784                 spin_lock(&nn->nn_lock);
1785                 atomic_set(&nn->nn_timeout, 0);
1786                 if (nn->nn_persistent_error)
1787                         o2net_set_nn_state(nn, NULL, 0, 0);
1788                 spin_unlock(&nn->nn_lock);
1789         }
1790 }
1791 
1792 void o2net_unregister_hb_callbacks(void)
1793 {
1794         o2hb_unregister_callback(NULL, &o2net_hb_up);
1795         o2hb_unregister_callback(NULL, &o2net_hb_down);
1796 }
1797 
1798 int o2net_register_hb_callbacks(void)
1799 {
1800         int ret;
1801 
1802         o2hb_setup_callback(&o2net_hb_down, O2HB_NODE_DOWN_CB,
1803                             o2net_hb_node_down_cb, NULL, O2NET_HB_PRI);
1804         o2hb_setup_callback(&o2net_hb_up, O2HB_NODE_UP_CB,
1805                             o2net_hb_node_up_cb, NULL, O2NET_HB_PRI);
1806 
1807         ret = o2hb_register_callback(NULL, &o2net_hb_up);
1808         if (ret == 0)
1809                 ret = o2hb_register_callback(NULL, &o2net_hb_down);
1810 
1811         if (ret)
1812                 o2net_unregister_hb_callbacks();
1813 
1814         return ret;
1815 }
1816 
1817 /* ------------------------------------------------------------ */
1818 
1819 static int o2net_accept_one(struct socket *sock, int *more)
1820 {
1821         int ret, slen;
1822         struct sockaddr_in sin;
1823         struct socket *new_sock = NULL;
1824         struct o2nm_node *node = NULL;
1825         struct o2nm_node *local_node = NULL;
1826         struct o2net_sock_container *sc = NULL;
1827         struct o2net_node *nn;
1828         unsigned int noio_flag;
1829 
1830         /*
1831          * sock_create_lite allocates the sock with GFP_KERNEL. We must set
1832          * per-process flag PF_MEMALLOC_NOIO so that all allocations done
1833          * by this process are done as if GFP_NOIO was specified. So we
1834          * are not reentering filesystem while doing memory reclaim.
1835          */
1836         noio_flag = memalloc_noio_save();
1837 
1838         BUG_ON(sock == NULL);
1839         *more = 0;
1840         ret = sock_create_lite(sock->sk->sk_family, sock->sk->sk_type,
1841                                sock->sk->sk_protocol, &new_sock);
1842         if (ret)
1843                 goto out;
1844 
1845         new_sock->type = sock->type;
1846         new_sock->ops = sock->ops;
1847         ret = sock->ops->accept(sock, new_sock, O_NONBLOCK, false);
1848         if (ret < 0)
1849                 goto out;
1850 
1851         *more = 1;
1852         new_sock->sk->sk_allocation = GFP_ATOMIC;
1853 
1854         ret = o2net_set_nodelay(new_sock);
1855         if (ret) {
1856                 mlog(ML_ERROR, "setting TCP_NODELAY failed with %d\n", ret);
1857                 goto out;
1858         }
1859 
1860         ret = o2net_set_usertimeout(new_sock);
1861         if (ret) {
1862                 mlog(ML_ERROR, "set TCP_USER_TIMEOUT failed with %d\n", ret);
1863                 goto out;
1864         }
1865 
1866         slen = sizeof(sin);
1867         ret = new_sock->ops->getname(new_sock, (struct sockaddr *) &sin,
1868                                        &slen, 1);
1869         if (ret < 0)
1870                 goto out;
1871 
1872         node = o2nm_get_node_by_ip(sin.sin_addr.s_addr);
1873         if (node == NULL) {
1874                 printk(KERN_NOTICE "o2net: Attempt to connect from unknown "
1875                        "node at %pI4:%d\n", &sin.sin_addr.s_addr,
1876                        ntohs(sin.sin_port));
1877                 ret = -EINVAL;
1878                 goto out;
1879         }
1880 
1881         if (o2nm_this_node() >= node->nd_num) {
1882                 local_node = o2nm_get_node_by_num(o2nm_this_node());
1883                 if (local_node)
1884                         printk(KERN_NOTICE "o2net: Unexpected connect attempt "
1885                                         "seen at node '%s' (%u, %pI4:%d) from "
1886                                         "node '%s' (%u, %pI4:%d)\n",
1887                                         local_node->nd_name, local_node->nd_num,
1888                                         &(local_node->nd_ipv4_address),
1889                                         ntohs(local_node->nd_ipv4_port),
1890                                         node->nd_name,
1891                                         node->nd_num, &sin.sin_addr.s_addr,
1892                                         ntohs(sin.sin_port));
1893                 ret = -EINVAL;
1894                 goto out;
1895         }
1896 
1897         /* this happens all the time when the other node sees our heartbeat
1898          * and tries to connect before we see their heartbeat */
1899         if (!o2hb_check_node_heartbeating_from_callback(node->nd_num)) {
1900                 mlog(ML_CONN, "attempt to connect from node '%s' at "
1901                      "%pI4:%d but it isn't heartbeating\n",
1902                      node->nd_name, &sin.sin_addr.s_addr,
1903                      ntohs(sin.sin_port));
1904                 ret = -EINVAL;
1905                 goto out;
1906         }
1907 
1908         nn = o2net_nn_from_num(node->nd_num);
1909 
1910         spin_lock(&nn->nn_lock);
1911         if (nn->nn_sc)
1912                 ret = -EBUSY;
1913         else
1914                 ret = 0;
1915         spin_unlock(&nn->nn_lock);
1916         if (ret) {
1917                 printk(KERN_NOTICE "o2net: Attempt to connect from node '%s' "
1918                        "at %pI4:%d but it already has an open connection\n",
1919                        node->nd_name, &sin.sin_addr.s_addr,
1920                        ntohs(sin.sin_port));
1921                 goto out;
1922         }
1923 
1924         sc = sc_alloc(node);
1925         if (sc == NULL) {
1926                 ret = -ENOMEM;
1927                 goto out;
1928         }
1929 
1930         sc->sc_sock = new_sock;
1931         new_sock = NULL;
1932 
1933         spin_lock(&nn->nn_lock);
1934         atomic_set(&nn->nn_timeout, 0);
1935         o2net_set_nn_state(nn, sc, 0, 0);
1936         spin_unlock(&nn->nn_lock);
1937 
1938         o2net_register_callbacks(sc->sc_sock->sk, sc);
1939         o2net_sc_queue_work(sc, &sc->sc_rx_work);
1940 
1941         o2net_initialize_handshake();
1942         o2net_sendpage(sc, o2net_hand, sizeof(*o2net_hand));
1943 
1944 out:
1945         if (new_sock)
1946                 sock_release(new_sock);
1947         if (node)
1948                 o2nm_node_put(node);
1949         if (local_node)
1950                 o2nm_node_put(local_node);
1951         if (sc)
1952                 sc_put(sc);
1953 
1954         memalloc_noio_restore(noio_flag);
1955         return ret;
1956 }
1957 
1958 /*
1959  * This function is invoked in response to one or more
1960  * pending accepts at softIRQ level. We must drain the
1961  * entire que before returning.
1962  */
1963 
1964 static void o2net_accept_many(struct work_struct *work)
1965 {
1966         struct socket *sock = o2net_listen_sock;
1967         int     more;
1968         int     err;
1969 
1970         /*
1971          * It is critical to note that due to interrupt moderation
1972          * at the network driver level, we can't assume to get a
1973          * softIRQ for every single conn since tcp SYN packets
1974          * can arrive back-to-back, and therefore many pending
1975          * accepts may result in just 1 softIRQ. If we terminate
1976          * the o2net_accept_one() loop upon seeing an err, what happens
1977          * to the rest of the conns in the queue? If no new SYN
1978          * arrives for hours, no softIRQ  will be delivered,
1979          * and the connections will just sit in the queue.
1980          */
1981 
1982         for (;;) {
1983                 err = o2net_accept_one(sock, &more);
1984                 if (!more)
1985                         break;
1986                 cond_resched();
1987         }
1988 }
1989 
1990 static void o2net_listen_data_ready(struct sock *sk)
1991 {
1992         void (*ready)(struct sock *sk);
1993 
1994         read_lock_bh(&sk->sk_callback_lock);
1995         ready = sk->sk_user_data;
1996         if (ready == NULL) { /* check for teardown race */
1997                 ready = sk->sk_data_ready;
1998                 goto out;
1999         }
2000 
2001         /* This callback may called twice when a new connection
2002          * is  being established as a child socket inherits everything
2003          * from a parent LISTEN socket, including the data_ready cb of
2004          * the parent. This leads to a hazard. In o2net_accept_one()
2005          * we are still initializing the child socket but have not
2006          * changed the inherited data_ready callback yet when
2007          * data starts arriving.
2008          * We avoid this hazard by checking the state.
2009          * For the listening socket,  the state will be TCP_LISTEN; for the new
2010          * socket, will be  TCP_ESTABLISHED. Also, in this case,
2011          * sk->sk_user_data is not a valid function pointer.
2012          */
2013 
2014         if (sk->sk_state == TCP_LISTEN) {
2015                 queue_work(o2net_wq, &o2net_listen_work);
2016         } else {
2017                 ready = NULL;
2018         }
2019 
2020 out:
2021         read_unlock_bh(&sk->sk_callback_lock);
2022         if (ready != NULL)
2023                 ready(sk);
2024 }
2025 
2026 static int o2net_open_listening_sock(__be32 addr, __be16 port)
2027 {
2028         struct socket *sock = NULL;
2029         int ret;
2030         struct sockaddr_in sin = {
2031                 .sin_family = PF_INET,
2032                 .sin_addr = { .s_addr = addr },
2033                 .sin_port = port,
2034         };
2035 
2036         ret = sock_create(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock);
2037         if (ret < 0) {
2038                 printk(KERN_ERR "o2net: Error %d while creating socket\n", ret);
2039                 goto out;
2040         }
2041 
2042         sock->sk->sk_allocation = GFP_ATOMIC;
2043 
2044         write_lock_bh(&sock->sk->sk_callback_lock);
2045         sock->sk->sk_user_data = sock->sk->sk_data_ready;
2046         sock->sk->sk_data_ready = o2net_listen_data_ready;
2047         write_unlock_bh(&sock->sk->sk_callback_lock);
2048 
2049         o2net_listen_sock = sock;
2050         INIT_WORK(&o2net_listen_work, o2net_accept_many);
2051 
2052         sock->sk->sk_reuse = SK_CAN_REUSE;
2053         ret = sock->ops->bind(sock, (struct sockaddr *)&sin, sizeof(sin));
2054         if (ret < 0) {
2055                 printk(KERN_ERR "o2net: Error %d while binding socket at "
2056                        "%pI4:%u\n", ret, &addr, ntohs(port)); 
2057                 goto out;
2058         }
2059 
2060         ret = sock->ops->listen(sock, 64);
2061         if (ret < 0)
2062                 printk(KERN_ERR "o2net: Error %d while listening on %pI4:%u\n",
2063                        ret, &addr, ntohs(port));
2064 
2065 out:
2066         if (ret) {
2067                 o2net_listen_sock = NULL;
2068                 if (sock)
2069                         sock_release(sock);
2070         }
2071         return ret;
2072 }
2073 
2074 /*
2075  * called from node manager when we should bring up our network listening
2076  * socket.  node manager handles all the serialization to only call this
2077  * once and to match it with o2net_stop_listening().  note,
2078  * o2nm_this_node() doesn't work yet as we're being called while it
2079  * is being set up.
2080  */
2081 int o2net_start_listening(struct o2nm_node *node)
2082 {
2083         int ret = 0;
2084 
2085         BUG_ON(o2net_wq != NULL);
2086         BUG_ON(o2net_listen_sock != NULL);
2087 
2088         mlog(ML_KTHREAD, "starting o2net thread...\n");
2089         o2net_wq = alloc_ordered_workqueue("o2net", WQ_MEM_RECLAIM);
2090         if (o2net_wq == NULL) {
2091                 mlog(ML_ERROR, "unable to launch o2net thread\n");
2092                 return -ENOMEM; /* ? */
2093         }
2094 
2095         ret = o2net_open_listening_sock(node->nd_ipv4_address,
2096                                         node->nd_ipv4_port);
2097         if (ret) {
2098                 destroy_workqueue(o2net_wq);
2099                 o2net_wq = NULL;
2100         } else
2101                 o2quo_conn_up(node->nd_num);
2102 
2103         return ret;
2104 }
2105 
2106 /* again, o2nm_this_node() doesn't work here as we're involved in
2107  * tearing it down */
2108 void o2net_stop_listening(struct o2nm_node *node)
2109 {
2110         struct socket *sock = o2net_listen_sock;
2111         size_t i;
2112 
2113         BUG_ON(o2net_wq == NULL);
2114         BUG_ON(o2net_listen_sock == NULL);
2115 
2116         /* stop the listening socket from generating work */
2117         write_lock_bh(&sock->sk->sk_callback_lock);
2118         sock->sk->sk_data_ready = sock->sk->sk_user_data;
2119         sock->sk->sk_user_data = NULL;
2120         write_unlock_bh(&sock->sk->sk_callback_lock);
2121 
2122         for (i = 0; i < ARRAY_SIZE(o2net_nodes); i++) {
2123                 struct o2nm_node *node = o2nm_get_node_by_num(i);
2124                 if (node) {
2125                         o2net_disconnect_node(node);
2126                         o2nm_node_put(node);
2127                 }
2128         }
2129 
2130         /* finish all work and tear down the work queue */
2131         mlog(ML_KTHREAD, "waiting for o2net thread to exit....\n");
2132         destroy_workqueue(o2net_wq);
2133         o2net_wq = NULL;
2134 
2135         sock_release(o2net_listen_sock);
2136         o2net_listen_sock = NULL;
2137 
2138         o2quo_conn_err(node->nd_num);
2139 }
2140 
2141 /* ------------------------------------------------------------ */
2142 
2143 int o2net_init(void)
2144 {
2145         unsigned long i;
2146 
2147         o2quo_init();
2148 
2149         if (o2net_debugfs_init())
2150                 goto out;
2151 
2152         o2net_hand = kzalloc(sizeof(struct o2net_handshake), GFP_KERNEL);
2153         o2net_keep_req = kzalloc(sizeof(struct o2net_msg), GFP_KERNEL);
2154         o2net_keep_resp = kzalloc(sizeof(struct o2net_msg), GFP_KERNEL);
2155         if (!o2net_hand || !o2net_keep_req || !o2net_keep_resp)
2156                 goto out;
2157 
2158         o2net_hand->protocol_version = cpu_to_be64(O2NET_PROTOCOL_VERSION);
2159         o2net_hand->connector_id = cpu_to_be64(1);
2160 
2161         o2net_keep_req->magic = cpu_to_be16(O2NET_MSG_KEEP_REQ_MAGIC);
2162         o2net_keep_resp->magic = cpu_to_be16(O2NET_MSG_KEEP_RESP_MAGIC);
2163 
2164         for (i = 0; i < ARRAY_SIZE(o2net_nodes); i++) {
2165                 struct o2net_node *nn = o2net_nn_from_num(i);
2166 
2167                 atomic_set(&nn->nn_timeout, 0);
2168                 spin_lock_init(&nn->nn_lock);
2169                 INIT_DELAYED_WORK(&nn->nn_connect_work, o2net_start_connect);
2170                 INIT_DELAYED_WORK(&nn->nn_connect_expired,
2171                                   o2net_connect_expired);
2172                 INIT_DELAYED_WORK(&nn->nn_still_up, o2net_still_up);
2173                 /* until we see hb from a node we'll return einval */
2174                 nn->nn_persistent_error = -ENOTCONN;
2175                 init_waitqueue_head(&nn->nn_sc_wq);
2176                 idr_init(&nn->nn_status_idr);
2177                 INIT_LIST_HEAD(&nn->nn_status_list);
2178         }
2179 
2180         return 0;
2181 
2182 out:
2183         kfree(o2net_hand);
2184         kfree(o2net_keep_req);
2185         kfree(o2net_keep_resp);
2186         o2net_debugfs_exit();
2187         o2quo_exit();
2188         return -ENOMEM;
2189 }
2190 
2191 void o2net_exit(void)
2192 {
2193         o2quo_exit();
2194         kfree(o2net_hand);
2195         kfree(o2net_keep_req);
2196         kfree(o2net_keep_resp);
2197         o2net_debugfs_exit();
2198 }
2199 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp