Commit | Line | Data |
---|---|---|
98211489 ZB |
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> | |
5b3cc15a | 57 | #include <linux/sched/mm.h> |
98211489 ZB |
58 | #include <linux/jiffies.h> |
59 | #include <linux/slab.h> | |
60 | #include <linux/idr.h> | |
61 | #include <linux/kref.h> | |
91cf45f0 | 62 | #include <linux/net.h> |
afeacc8c | 63 | #include <linux/export.h> |
98211489 ZB |
64 | #include <net/tcp.h> |
65 | ||
7c0f6ba6 | 66 | #include <linux/uaccess.h> |
98211489 ZB |
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 | ||
03affdef | 77 | #define SC_NODEF_FMT "node %s (num %u) at %pI4:%u" |
98211489 | 78 | #define SC_NODEF_ARGS(sc) sc->sc_node->nd_name, sc->sc_node->nd_num, \ |
03affdef | 79 | &sc->sc_node->nd_ipv4_address, \ |
98211489 ZB |
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, \ | |
2c935bc5 | 101 | kref_read(&__sc->sc_kref), __sc->sc_sock, \ |
98211489 ZB |
102 | __sc->sc_node->nd_num, __sc->sc_page, __sc->sc_page_off , \ |
103 | ##args); \ | |
104 | } while (0) | |
105 | ||
34af946a | 106 | static DEFINE_RWLOCK(o2net_handler_lock); |
98211489 ZB |
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 */ | |
1a5c4e2a | 112 | static struct socket *o2net_listen_sock; |
98211489 ZB |
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 :/ */ | |
c4028958 DH |
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); | |
676d2369 | 141 | static void o2net_listen_data_ready(struct sock *sk); |
c4028958 | 142 | static void o2net_sc_send_keep_req(struct work_struct *work); |
98211489 ZB |
143 | static void o2net_idle_timer(unsigned long data); |
144 | static void o2net_sc_postpone_idle(struct o2net_sock_container *sc); | |
b5dd8030 JM |
145 | static void o2net_sc_reset_idle_timer(struct o2net_sock_container *sc); |
146 | ||
2309e9e0 | 147 | #ifdef CONFIG_DEBUG_FS |
18496e80 AB |
148 | static void o2net_init_nst(struct o2net_send_tracking *nst, u32 msgtype, |
149 | u32 msgkey, struct task_struct *task, u8 node) | |
0f475b2a | 150 | { |
2309e9e0 SM |
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; | |
2309e9e0 SM |
156 | } |
157 | ||
ff1becbf | 158 | static inline void o2net_set_nst_sock_time(struct o2net_send_tracking *nst) |
2309e9e0 | 159 | { |
3f9c14fa | 160 | nst->st_sock_time = ktime_get(); |
2309e9e0 SM |
161 | } |
162 | ||
ff1becbf | 163 | static inline void o2net_set_nst_send_time(struct o2net_send_tracking *nst) |
2309e9e0 | 164 | { |
3f9c14fa | 165 | nst->st_send_time = ktime_get(); |
2309e9e0 SM |
166 | } |
167 | ||
ff1becbf | 168 | static inline void o2net_set_nst_status_time(struct o2net_send_tracking *nst) |
2309e9e0 | 169 | { |
3f9c14fa | 170 | nst->st_status_time = ktime_get(); |
2309e9e0 SM |
171 | } |
172 | ||
ff1becbf SM |
173 | static inline void o2net_set_nst_sock_container(struct o2net_send_tracking *nst, |
174 | struct o2net_sock_container *sc) | |
2309e9e0 | 175 | { |
2309e9e0 | 176 | nst->st_sc = sc; |
2309e9e0 SM |
177 | } |
178 | ||
ff1becbf SM |
179 | static inline void o2net_set_nst_msg_id(struct o2net_send_tracking *nst, |
180 | u32 msg_id) | |
2309e9e0 | 181 | { |
2309e9e0 | 182 | nst->st_id = msg_id; |
2309e9e0 | 183 | } |
18496e80 | 184 | |
ff1becbf | 185 | static inline void o2net_set_sock_timer(struct o2net_sock_container *sc) |
18496e80 | 186 | { |
ff1becbf | 187 | sc->sc_tv_timer = ktime_get(); |
18496e80 AB |
188 | } |
189 | ||
ff1becbf | 190 | static inline void o2net_set_data_ready_time(struct o2net_sock_container *sc) |
18496e80 | 191 | { |
ff1becbf | 192 | sc->sc_tv_data_ready = ktime_get(); |
18496e80 AB |
193 | } |
194 | ||
ff1becbf | 195 | static inline void o2net_set_advance_start_time(struct o2net_sock_container *sc) |
18496e80 | 196 | { |
ff1becbf | 197 | sc->sc_tv_advance_start = ktime_get(); |
18496e80 AB |
198 | } |
199 | ||
ff1becbf | 200 | static inline void o2net_set_advance_stop_time(struct o2net_sock_container *sc) |
18496e80 | 201 | { |
ff1becbf | 202 | sc->sc_tv_advance_stop = ktime_get(); |
18496e80 AB |
203 | } |
204 | ||
ff1becbf | 205 | static inline void o2net_set_func_start_time(struct o2net_sock_container *sc) |
18496e80 | 206 | { |
ff1becbf | 207 | sc->sc_tv_func_start = ktime_get(); |
18496e80 AB |
208 | } |
209 | ||
ff1becbf | 210 | static inline void o2net_set_func_stop_time(struct o2net_sock_container *sc) |
18496e80 | 211 | { |
ff1becbf | 212 | sc->sc_tv_func_stop = ktime_get(); |
18496e80 | 213 | } |
e453039f | 214 | |
ff1becbf SM |
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) | |
0f475b2a | 228 | #endif /* CONFIG_DEBUG_FS */ |
2309e9e0 | 229 | |
3c193b38 | 230 | #ifdef CONFIG_OCFS2_FS_STATS |
ed59992e RM |
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 | ||
3c193b38 SM |
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 | ||
e453039f SM |
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 | ||
3c193b38 SM |
258 | #else |
259 | ||
260 | # define o2net_update_send_stats(a, b) | |
261 | ||
e70d8450 RD |
262 | # define o2net_update_recv_stats(sc) |
263 | ||
3c193b38 SM |
264 | #endif /* CONFIG_OCFS2_FS_STATS */ |
265 | ||
41b63efb | 266 | static inline unsigned int o2net_reconnect_delay(void) |
b5dd8030 JM |
267 | { |
268 | return o2nm_single_cluster->cl_reconnect_delay_ms; | |
269 | } | |
270 | ||
41b63efb | 271 | static inline unsigned int o2net_keepalive_delay(void) |
b5dd8030 JM |
272 | { |
273 | return o2nm_single_cluster->cl_keepalive_delay_ms; | |
274 | } | |
275 | ||
41b63efb | 276 | static inline unsigned int o2net_idle_timeout(void) |
b5dd8030 JM |
277 | { |
278 | return o2nm_single_cluster->cl_idle_timeout_ms; | |
279 | } | |
98211489 ZB |
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 | { | |
6b207ba3 TH |
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); | |
98211489 | 315 | } |
6b207ba3 TH |
316 | spin_unlock(&nn->nn_lock); |
317 | if (ret < 0) | |
318 | return ret; | |
98211489 | 319 | |
6b207ba3 TH |
320 | init_waitqueue_head(&nsw->ns_wq); |
321 | nsw->ns_sys_status = O2NET_ERR_NONE; | |
322 | nsw->ns_status = 0; | |
323 | return 0; | |
98211489 ZB |
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 | { | |
800deef3 | 366 | struct o2net_status_wait *nsw, *tmp; |
98211489 | 367 | unsigned int num_kills = 0; |
98211489 ZB |
368 | |
369 | assert_spin_locked(&nn->nn_lock); | |
370 | ||
800deef3 | 371 | list_for_each_entry_safe(nsw, tmp, &nn->nn_status_list, ns_node_item) { |
98211489 ZB |
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); | |
b5dd8030 JM |
396 | BUG_ON(timer_pending(&sc->sc_idle_timeout)); |
397 | ||
98211489 ZB |
398 | sclog(sc, "releasing\n"); |
399 | ||
400 | if (sc->sc_sock) { | |
401 | sock_release(sc->sc_sock); | |
402 | sc->sc_sock = NULL; | |
403 | } | |
404 | ||
2b190ce9 | 405 | o2nm_undepend_item(&sc->sc_node->nd_item); |
98211489 ZB |
406 | o2nm_node_put(sc->sc_node); |
407 | sc->sc_node = NULL; | |
408 | ||
2309e9e0 | 409 | o2net_debug_del_sc(sc); |
d3e3b41b YL |
410 | |
411 | if (sc->sc_page) | |
412 | __free_page(sc->sc_page); | |
98211489 ZB |
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; | |
2b190ce9 | 430 | int status = 0; |
98211489 ZB |
431 | |
432 | page = alloc_page(GFP_NOFS); | |
cd861280 | 433 | sc = kzalloc(sizeof(*sc), GFP_NOFS); |
98211489 ZB |
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 | ||
2b190ce9 SM |
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 | } | |
c4028958 DH |
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); | |
98211489 ZB |
452 | |
453 | init_timer(&sc->sc_idle_timeout); | |
454 | sc->sc_idle_timeout.function = o2net_idle_timer; | |
455 | sc->sc_idle_timeout.data = (unsigned long)sc; | |
456 | ||
457 | sclog(sc, "alloced\n"); | |
458 | ||
459 | ret = sc; | |
460 | sc->sc_page = page; | |
2309e9e0 | 461 | o2net_debug_add_sc(sc); |
98211489 ZB |
462 | sc = NULL; |
463 | page = NULL; | |
464 | ||
465 | out: | |
466 | if (page) | |
467 | __free_page(page); | |
468 | kfree(sc); | |
469 | ||
470 | return ret; | |
471 | } | |
472 | ||
473 | /* ------------------------------------------------------------ */ | |
474 | ||
475 | static void o2net_sc_queue_work(struct o2net_sock_container *sc, | |
476 | struct work_struct *work) | |
477 | { | |
478 | sc_get(sc); | |
479 | if (!queue_work(o2net_wq, work)) | |
480 | sc_put(sc); | |
481 | } | |
482 | static void o2net_sc_queue_delayed_work(struct o2net_sock_container *sc, | |
c4028958 | 483 | struct delayed_work *work, |
98211489 ZB |
484 | int delay) |
485 | { | |
486 | sc_get(sc); | |
487 | if (!queue_delayed_work(o2net_wq, work, delay)) | |
488 | sc_put(sc); | |
489 | } | |
490 | static void o2net_sc_cancel_delayed_work(struct o2net_sock_container *sc, | |
c4028958 | 491 | struct delayed_work *work) |
98211489 ZB |
492 | { |
493 | if (cancel_delayed_work(work)) | |
494 | sc_put(sc); | |
495 | } | |
496 | ||
828ae6af AB |
497 | static atomic_t o2net_connected_peers = ATOMIC_INIT(0); |
498 | ||
499 | int o2net_num_connected_peers(void) | |
500 | { | |
501 | return atomic_read(&o2net_connected_peers); | |
502 | } | |
503 | ||
98211489 ZB |
504 | static void o2net_set_nn_state(struct o2net_node *nn, |
505 | struct o2net_sock_container *sc, | |
506 | unsigned valid, int err) | |
507 | { | |
508 | int was_valid = nn->nn_sc_valid; | |
509 | int was_err = nn->nn_persistent_error; | |
510 | struct o2net_sock_container *old_sc = nn->nn_sc; | |
511 | ||
512 | assert_spin_locked(&nn->nn_lock); | |
513 | ||
828ae6af AB |
514 | if (old_sc && !sc) |
515 | atomic_dec(&o2net_connected_peers); | |
516 | else if (!old_sc && sc) | |
517 | atomic_inc(&o2net_connected_peers); | |
518 | ||
98211489 ZB |
519 | /* the node num comparison and single connect/accept path should stop |
520 | * an non-null sc from being overwritten with another */ | |
521 | BUG_ON(sc && nn->nn_sc && nn->nn_sc != sc); | |
522 | mlog_bug_on_msg(err && valid, "err %d valid %u\n", err, valid); | |
523 | mlog_bug_on_msg(valid && !sc, "valid %u sc %p\n", valid, sc); | |
524 | ||
98211489 ZB |
525 | if (was_valid && !valid && err == 0) |
526 | err = -ENOTCONN; | |
527 | ||
528 | mlog(ML_CONN, "node %u sc: %p -> %p, valid %u -> %u, err %d -> %d\n", | |
529 | o2net_num_from_nn(nn), nn->nn_sc, sc, nn->nn_sc_valid, valid, | |
530 | nn->nn_persistent_error, err); | |
531 | ||
532 | nn->nn_sc = sc; | |
533 | nn->nn_sc_valid = valid ? 1 : 0; | |
534 | nn->nn_persistent_error = err; | |
535 | ||
536 | /* mirrors o2net_tx_can_proceed() */ | |
537 | if (nn->nn_persistent_error || nn->nn_sc_valid) | |
538 | wake_up(&nn->nn_sc_wq); | |
539 | ||
5046f18d | 540 | if (was_valid && !was_err && nn->nn_persistent_error) { |
98211489 ZB |
541 | o2quo_conn_err(o2net_num_from_nn(nn)); |
542 | queue_delayed_work(o2net_wq, &nn->nn_still_up, | |
543 | msecs_to_jiffies(O2NET_QUORUM_DELAY_MS)); | |
544 | } | |
545 | ||
546 | if (was_valid && !valid) { | |
8dd7903e SM |
547 | if (old_sc) |
548 | printk(KERN_NOTICE "o2net: No longer connected to " | |
549 | SC_NODEF_FMT "\n", SC_NODEF_ARGS(old_sc)); | |
98211489 ZB |
550 | o2net_complete_nodes_nsw(nn); |
551 | } | |
552 | ||
553 | if (!was_valid && valid) { | |
554 | o2quo_conn_up(o2net_num_from_nn(nn)); | |
98211489 | 555 | cancel_delayed_work(&nn->nn_connect_expired); |
6efd8066 | 556 | printk(KERN_NOTICE "o2net: %s " SC_NODEF_FMT "\n", |
781ee3e2 | 557 | o2nm_this_node() > sc->sc_node->nd_num ? |
1dfecf81 | 558 | "Connected to" : "Accepted connection from", |
781ee3e2 | 559 | SC_NODEF_ARGS(sc)); |
98211489 ZB |
560 | } |
561 | ||
562 | /* trigger the connecting worker func as long as we're not valid, | |
563 | * it will back off if it shouldn't connect. This can be called | |
564 | * from node config teardown and so needs to be careful about | |
565 | * the work queue actually being up. */ | |
566 | if (!valid && o2net_wq) { | |
567 | unsigned long delay; | |
25985edc | 568 | /* delay if we're within a RECONNECT_DELAY of the |
98211489 ZB |
569 | * last attempt */ |
570 | delay = (nn->nn_last_connect_attempt + | |
409753bf | 571 | msecs_to_jiffies(o2net_reconnect_delay())) |
98211489 | 572 | - jiffies; |
409753bf | 573 | if (delay > msecs_to_jiffies(o2net_reconnect_delay())) |
98211489 ZB |
574 | delay = 0; |
575 | mlog(ML_CONN, "queueing conn attempt in %lu jiffies\n", delay); | |
576 | queue_delayed_work(o2net_wq, &nn->nn_connect_work, delay); | |
5cc3bf27 TM |
577 | |
578 | /* | |
579 | * Delay the expired work after idle timeout. | |
580 | * | |
581 | * We might have lots of failed connection attempts that run | |
582 | * through here but we only cancel the connect_expired work when | |
583 | * a connection attempt succeeds. So only the first enqueue of | |
584 | * the connect_expired work will do anything. The rest will see | |
585 | * that it's already queued and do nothing. | |
586 | */ | |
409753bf | 587 | delay += msecs_to_jiffies(o2net_idle_timeout()); |
5cc3bf27 | 588 | queue_delayed_work(o2net_wq, &nn->nn_connect_expired, delay); |
98211489 ZB |
589 | } |
590 | ||
591 | /* keep track of the nn's sc ref for the caller */ | |
592 | if ((old_sc == NULL) && sc) | |
593 | sc_get(sc); | |
594 | if (old_sc && (old_sc != sc)) { | |
595 | o2net_sc_queue_work(old_sc, &old_sc->sc_shutdown_work); | |
596 | sc_put(old_sc); | |
597 | } | |
598 | } | |
599 | ||
600 | /* see o2net_register_callbacks() */ | |
676d2369 | 601 | static void o2net_data_ready(struct sock *sk) |
98211489 | 602 | { |
676d2369 | 603 | void (*ready)(struct sock *sk); |
e1daca28 | 604 | struct o2net_sock_container *sc; |
98211489 | 605 | |
e1daca28 ED |
606 | read_lock_bh(&sk->sk_callback_lock); |
607 | sc = sk->sk_user_data; | |
608 | if (sc) { | |
98211489 | 609 | sclog(sc, "data_ready hit\n"); |
ff1becbf | 610 | o2net_set_data_ready_time(sc); |
98211489 ZB |
611 | o2net_sc_queue_work(sc, &sc->sc_rx_work); |
612 | ready = sc->sc_data_ready; | |
613 | } else { | |
614 | ready = sk->sk_data_ready; | |
615 | } | |
e1daca28 | 616 | read_unlock_bh(&sk->sk_callback_lock); |
98211489 | 617 | |
676d2369 | 618 | ready(sk); |
98211489 ZB |
619 | } |
620 | ||
621 | /* see o2net_register_callbacks() */ | |
622 | static void o2net_state_change(struct sock *sk) | |
623 | { | |
624 | void (*state_change)(struct sock *sk); | |
625 | struct o2net_sock_container *sc; | |
626 | ||
e1daca28 | 627 | read_lock_bh(&sk->sk_callback_lock); |
98211489 ZB |
628 | sc = sk->sk_user_data; |
629 | if (sc == NULL) { | |
630 | state_change = sk->sk_state_change; | |
631 | goto out; | |
632 | } | |
633 | ||
634 | sclog(sc, "state_change to %d\n", sk->sk_state); | |
635 | ||
636 | state_change = sc->sc_state_change; | |
637 | ||
638 | switch(sk->sk_state) { | |
44e89cb8 JL |
639 | /* ignore connecting sockets as they make progress */ |
640 | case TCP_SYN_SENT: | |
641 | case TCP_SYN_RECV: | |
642 | break; | |
643 | case TCP_ESTABLISHED: | |
644 | o2net_sc_queue_work(sc, &sc->sc_connect_work); | |
645 | break; | |
646 | default: | |
647 | printk(KERN_INFO "o2net: Connection to " SC_NODEF_FMT | |
648 | " shutdown, state %d\n", | |
649 | SC_NODEF_ARGS(sc), sk->sk_state); | |
650 | o2net_sc_queue_work(sc, &sc->sc_shutdown_work); | |
651 | break; | |
98211489 ZB |
652 | } |
653 | out: | |
e1daca28 | 654 | read_unlock_bh(&sk->sk_callback_lock); |
98211489 ZB |
655 | state_change(sk); |
656 | } | |
657 | ||
658 | /* | |
659 | * we register callbacks so we can queue work on events before calling | |
660 | * the original callbacks. our callbacks our careful to test user_data | |
661 | * to discover when they've reaced with o2net_unregister_callbacks(). | |
662 | */ | |
663 | static void o2net_register_callbacks(struct sock *sk, | |
664 | struct o2net_sock_container *sc) | |
665 | { | |
666 | write_lock_bh(&sk->sk_callback_lock); | |
667 | ||
668 | /* accepted sockets inherit the old listen socket data ready */ | |
669 | if (sk->sk_data_ready == o2net_listen_data_ready) { | |
670 | sk->sk_data_ready = sk->sk_user_data; | |
671 | sk->sk_user_data = NULL; | |
672 | } | |
673 | ||
674 | BUG_ON(sk->sk_user_data != NULL); | |
675 | sk->sk_user_data = sc; | |
676 | sc_get(sc); | |
677 | ||
678 | sc->sc_data_ready = sk->sk_data_ready; | |
679 | sc->sc_state_change = sk->sk_state_change; | |
680 | sk->sk_data_ready = o2net_data_ready; | |
681 | sk->sk_state_change = o2net_state_change; | |
682 | ||
925037bc ZW |
683 | mutex_init(&sc->sc_send_lock); |
684 | ||
98211489 ZB |
685 | write_unlock_bh(&sk->sk_callback_lock); |
686 | } | |
687 | ||
688 | static int o2net_unregister_callbacks(struct sock *sk, | |
689 | struct o2net_sock_container *sc) | |
690 | { | |
691 | int ret = 0; | |
692 | ||
693 | write_lock_bh(&sk->sk_callback_lock); | |
694 | if (sk->sk_user_data == sc) { | |
695 | ret = 1; | |
696 | sk->sk_user_data = NULL; | |
697 | sk->sk_data_ready = sc->sc_data_ready; | |
698 | sk->sk_state_change = sc->sc_state_change; | |
699 | } | |
700 | write_unlock_bh(&sk->sk_callback_lock); | |
701 | ||
702 | return ret; | |
703 | } | |
704 | ||
705 | /* | |
706 | * this is a little helper that is called by callers who have seen a problem | |
707 | * with an sc and want to detach it from the nn if someone already hasn't beat | |
708 | * them to it. if an error is given then the shutdown will be persistent | |
709 | * and pending transmits will be canceled. | |
710 | */ | |
711 | static void o2net_ensure_shutdown(struct o2net_node *nn, | |
712 | struct o2net_sock_container *sc, | |
713 | int err) | |
714 | { | |
715 | spin_lock(&nn->nn_lock); | |
716 | if (nn->nn_sc == sc) | |
717 | o2net_set_nn_state(nn, NULL, 0, err); | |
718 | spin_unlock(&nn->nn_lock); | |
719 | } | |
720 | ||
721 | /* | |
722 | * This work queue function performs the blocking parts of socket shutdown. A | |
723 | * few paths lead here. set_nn_state will trigger this callback if it sees an | |
724 | * sc detached from the nn. state_change will also trigger this callback | |
725 | * directly when it sees errors. In that case we need to call set_nn_state | |
726 | * ourselves as state_change couldn't get the nn_lock and call set_nn_state | |
727 | * itself. | |
728 | */ | |
c4028958 | 729 | static void o2net_shutdown_sc(struct work_struct *work) |
98211489 | 730 | { |
c4028958 DH |
731 | struct o2net_sock_container *sc = |
732 | container_of(work, struct o2net_sock_container, | |
733 | sc_shutdown_work); | |
98211489 ZB |
734 | struct o2net_node *nn = o2net_nn_from_num(sc->sc_node->nd_num); |
735 | ||
736 | sclog(sc, "shutting down\n"); | |
737 | ||
738 | /* drop the callbacks ref and call shutdown only once */ | |
739 | if (o2net_unregister_callbacks(sc->sc_sock->sk, sc)) { | |
740 | /* we shouldn't flush as we're in the thread, the | |
741 | * races with pending sc work structs are harmless */ | |
742 | del_timer_sync(&sc->sc_idle_timeout); | |
743 | o2net_sc_cancel_delayed_work(sc, &sc->sc_keepalive_work); | |
744 | sc_put(sc); | |
91cf45f0 | 745 | kernel_sock_shutdown(sc->sc_sock, SHUT_RDWR); |
98211489 ZB |
746 | } |
747 | ||
748 | /* not fatal so failed connects before the other guy has our | |
749 | * heartbeat can be retried */ | |
750 | o2net_ensure_shutdown(nn, sc, 0); | |
751 | sc_put(sc); | |
752 | } | |
753 | ||
754 | /* ------------------------------------------------------------ */ | |
755 | ||
756 | static int o2net_handler_cmp(struct o2net_msg_handler *nmh, u32 msg_type, | |
757 | u32 key) | |
758 | { | |
759 | int ret = memcmp(&nmh->nh_key, &key, sizeof(key)); | |
760 | ||
761 | if (ret == 0) | |
762 | ret = memcmp(&nmh->nh_msg_type, &msg_type, sizeof(msg_type)); | |
763 | ||
764 | return ret; | |
765 | } | |
766 | ||
767 | static struct o2net_msg_handler * | |
768 | o2net_handler_tree_lookup(u32 msg_type, u32 key, struct rb_node ***ret_p, | |
769 | struct rb_node **ret_parent) | |
770 | { | |
9a239e4c JQ |
771 | struct rb_node **p = &o2net_handler_tree.rb_node; |
772 | struct rb_node *parent = NULL; | |
98211489 ZB |
773 | struct o2net_msg_handler *nmh, *ret = NULL; |
774 | int cmp; | |
775 | ||
9a239e4c JQ |
776 | while (*p) { |
777 | parent = *p; | |
778 | nmh = rb_entry(parent, struct o2net_msg_handler, nh_node); | |
98211489 ZB |
779 | cmp = o2net_handler_cmp(nmh, msg_type, key); |
780 | ||
9a239e4c JQ |
781 | if (cmp < 0) |
782 | p = &(*p)->rb_left; | |
783 | else if (cmp > 0) | |
784 | p = &(*p)->rb_right; | |
785 | else { | |
98211489 | 786 | ret = nmh; |
9a239e4c | 787 | break; |
98211489 | 788 | } |
9a239e4c | 789 | } |
98211489 | 790 | |
9a239e4c JQ |
791 | if (ret_p != NULL) |
792 | *ret_p = p; | |
793 | if (ret_parent != NULL) | |
794 | *ret_parent = parent; | |
98211489 | 795 | |
9a239e4c | 796 | return ret; |
98211489 ZB |
797 | } |
798 | ||
799 | static void o2net_handler_kref_release(struct kref *kref) | |
800 | { | |
801 | struct o2net_msg_handler *nmh; | |
802 | nmh = container_of(kref, struct o2net_msg_handler, nh_kref); | |
803 | ||
804 | kfree(nmh); | |
805 | } | |
806 | ||
807 | static void o2net_handler_put(struct o2net_msg_handler *nmh) | |
808 | { | |
809 | kref_put(&nmh->nh_kref, o2net_handler_kref_release); | |
810 | } | |
811 | ||
812 | /* max_len is protection for the handler func. incoming messages won't | |
813 | * be given to the handler if their payload is longer than the max. */ | |
814 | int o2net_register_handler(u32 msg_type, u32 key, u32 max_len, | |
815 | o2net_msg_handler_func *func, void *data, | |
d74c9803 | 816 | o2net_post_msg_handler_func *post_func, |
98211489 ZB |
817 | struct list_head *unreg_list) |
818 | { | |
819 | struct o2net_msg_handler *nmh = NULL; | |
820 | struct rb_node **p, *parent; | |
821 | int ret = 0; | |
822 | ||
823 | if (max_len > O2NET_MAX_PAYLOAD_BYTES) { | |
824 | mlog(0, "max_len for message handler out of range: %u\n", | |
825 | max_len); | |
826 | ret = -EINVAL; | |
827 | goto out; | |
828 | } | |
829 | ||
830 | if (!msg_type) { | |
831 | mlog(0, "no message type provided: %u, %p\n", msg_type, func); | |
832 | ret = -EINVAL; | |
833 | goto out; | |
834 | ||
835 | } | |
836 | if (!func) { | |
837 | mlog(0, "no message handler provided: %u, %p\n", | |
838 | msg_type, func); | |
839 | ret = -EINVAL; | |
840 | goto out; | |
841 | } | |
842 | ||
cd861280 | 843 | nmh = kzalloc(sizeof(struct o2net_msg_handler), GFP_NOFS); |
98211489 ZB |
844 | if (nmh == NULL) { |
845 | ret = -ENOMEM; | |
846 | goto out; | |
847 | } | |
848 | ||
849 | nmh->nh_func = func; | |
850 | nmh->nh_func_data = data; | |
d74c9803 | 851 | nmh->nh_post_func = post_func; |
98211489 ZB |
852 | nmh->nh_msg_type = msg_type; |
853 | nmh->nh_max_len = max_len; | |
854 | nmh->nh_key = key; | |
855 | /* the tree and list get this ref.. they're both removed in | |
856 | * unregister when this ref is dropped */ | |
857 | kref_init(&nmh->nh_kref); | |
858 | INIT_LIST_HEAD(&nmh->nh_unregister_item); | |
859 | ||
860 | write_lock(&o2net_handler_lock); | |
861 | if (o2net_handler_tree_lookup(msg_type, key, &p, &parent)) | |
862 | ret = -EEXIST; | |
863 | else { | |
864 | rb_link_node(&nmh->nh_node, parent, p); | |
865 | rb_insert_color(&nmh->nh_node, &o2net_handler_tree); | |
866 | list_add_tail(&nmh->nh_unregister_item, unreg_list); | |
867 | ||
868 | mlog(ML_TCP, "registered handler func %p type %u key %08x\n", | |
869 | func, msg_type, key); | |
870 | /* we've had some trouble with handlers seemingly vanishing. */ | |
871 | mlog_bug_on_msg(o2net_handler_tree_lookup(msg_type, key, &p, | |
872 | &parent) == NULL, | |
8a168ca7 | 873 | "couldn't find handler we *just* registered " |
98211489 ZB |
874 | "for type %u key %08x\n", msg_type, key); |
875 | } | |
876 | write_unlock(&o2net_handler_lock); | |
877 | if (ret) | |
878 | goto out; | |
879 | ||
880 | out: | |
881 | if (ret) | |
882 | kfree(nmh); | |
883 | ||
884 | return ret; | |
885 | } | |
886 | EXPORT_SYMBOL_GPL(o2net_register_handler); | |
887 | ||
888 | void o2net_unregister_handler_list(struct list_head *list) | |
889 | { | |
800deef3 | 890 | struct o2net_msg_handler *nmh, *n; |
98211489 ZB |
891 | |
892 | write_lock(&o2net_handler_lock); | |
800deef3 | 893 | list_for_each_entry_safe(nmh, n, list, nh_unregister_item) { |
98211489 ZB |
894 | mlog(ML_TCP, "unregistering handler func %p type %u key %08x\n", |
895 | nmh->nh_func, nmh->nh_msg_type, nmh->nh_key); | |
896 | rb_erase(&nmh->nh_node, &o2net_handler_tree); | |
897 | list_del_init(&nmh->nh_unregister_item); | |
898 | kref_put(&nmh->nh_kref, o2net_handler_kref_release); | |
899 | } | |
900 | write_unlock(&o2net_handler_lock); | |
901 | } | |
902 | EXPORT_SYMBOL_GPL(o2net_unregister_handler_list); | |
903 | ||
904 | static struct o2net_msg_handler *o2net_handler_get(u32 msg_type, u32 key) | |
905 | { | |
906 | struct o2net_msg_handler *nmh; | |
907 | ||
908 | read_lock(&o2net_handler_lock); | |
909 | nmh = o2net_handler_tree_lookup(msg_type, key, NULL, NULL); | |
910 | if (nmh) | |
911 | kref_get(&nmh->nh_kref); | |
912 | read_unlock(&o2net_handler_lock); | |
913 | ||
914 | return nmh; | |
915 | } | |
916 | ||
917 | /* ------------------------------------------------------------ */ | |
918 | ||
919 | static int o2net_recv_tcp_msg(struct socket *sock, void *data, size_t len) | |
920 | { | |
920220c1 AV |
921 | struct kvec vec = { .iov_len = len, .iov_base = data, }; |
922 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT, }; | |
923 | return kernel_recvmsg(sock, &msg, &vec, 1, len, msg.msg_flags); | |
98211489 ZB |
924 | } |
925 | ||
926 | static int o2net_send_tcp_msg(struct socket *sock, struct kvec *vec, | |
927 | size_t veclen, size_t total) | |
928 | { | |
929 | int ret; | |
7e8631e8 | 930 | struct msghdr msg = {.msg_flags = 0,}; |
98211489 ZB |
931 | |
932 | if (sock == NULL) { | |
933 | ret = -EINVAL; | |
934 | goto out; | |
935 | } | |
936 | ||
66f5dcef AV |
937 | ret = kernel_sendmsg(sock, &msg, vec, veclen, total); |
938 | if (likely(ret == total)) | |
939 | return 0; | |
940 | mlog(ML_ERROR, "sendmsg returned %d instead of %zu\n", ret, total); | |
941 | if (ret >= 0) | |
942 | ret = -EPIPE; /* should be smarter, I bet */ | |
98211489 | 943 | out: |
66f5dcef | 944 | mlog(0, "returning error: %d\n", ret); |
98211489 ZB |
945 | return ret; |
946 | } | |
947 | ||
948 | static void o2net_sendpage(struct o2net_sock_container *sc, | |
949 | void *kmalloced_virt, | |
950 | size_t size) | |
951 | { | |
952 | struct o2net_node *nn = o2net_nn_from_num(sc->sc_node->nd_num); | |
953 | ssize_t ret; | |
954 | ||
ce17204a SM |
955 | while (1) { |
956 | mutex_lock(&sc->sc_send_lock); | |
957 | ret = sc->sc_sock->ops->sendpage(sc->sc_sock, | |
958 | virt_to_page(kmalloced_virt), | |
959 | (long)kmalloced_virt & ~PAGE_MASK, | |
960 | size, MSG_DONTWAIT); | |
961 | mutex_unlock(&sc->sc_send_lock); | |
962 | if (ret == size) | |
963 | break; | |
964 | if (ret == (ssize_t)-EAGAIN) { | |
965 | mlog(0, "sendpage of size %zu to " SC_NODEF_FMT | |
966 | " returned EAGAIN\n", size, SC_NODEF_ARGS(sc)); | |
967 | cond_resched(); | |
968 | continue; | |
969 | } | |
2bd63216 | 970 | mlog(ML_ERROR, "sendpage of size %zu to " SC_NODEF_FMT |
98211489 ZB |
971 | " failed with %zd\n", size, SC_NODEF_ARGS(sc), ret); |
972 | o2net_ensure_shutdown(nn, sc, 0); | |
ce17204a | 973 | break; |
98211489 ZB |
974 | } |
975 | } | |
976 | ||
977 | static void o2net_init_msg(struct o2net_msg *msg, u16 data_len, u16 msg_type, u32 key) | |
978 | { | |
979 | memset(msg, 0, sizeof(struct o2net_msg)); | |
980 | msg->magic = cpu_to_be16(O2NET_MSG_MAGIC); | |
981 | msg->data_len = cpu_to_be16(data_len); | |
982 | msg->msg_type = cpu_to_be16(msg_type); | |
983 | msg->sys_status = cpu_to_be32(O2NET_ERR_NONE); | |
984 | msg->status = 0; | |
985 | msg->key = cpu_to_be32(key); | |
986 | } | |
987 | ||
988 | static int o2net_tx_can_proceed(struct o2net_node *nn, | |
989 | struct o2net_sock_container **sc_ret, | |
990 | int *error) | |
991 | { | |
992 | int ret = 0; | |
993 | ||
994 | spin_lock(&nn->nn_lock); | |
995 | if (nn->nn_persistent_error) { | |
996 | ret = 1; | |
997 | *sc_ret = NULL; | |
998 | *error = nn->nn_persistent_error; | |
999 | } else if (nn->nn_sc_valid) { | |
1000 | kref_get(&nn->nn_sc->sc_kref); | |
1001 | ||
1002 | ret = 1; | |
1003 | *sc_ret = nn->nn_sc; | |
1004 | *error = 0; | |
1005 | } | |
1006 | spin_unlock(&nn->nn_lock); | |
1007 | ||
1008 | return ret; | |
1009 | } | |
1010 | ||
3ba169cc SM |
1011 | /* Get a map of all nodes to which this node is currently connected to */ |
1012 | void o2net_fill_node_map(unsigned long *map, unsigned bytes) | |
1013 | { | |
1014 | struct o2net_sock_container *sc; | |
1015 | int node, ret; | |
1016 | ||
1017 | BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long))); | |
1018 | ||
1019 | memset(map, 0, bytes); | |
1020 | for (node = 0; node < O2NM_MAX_NODES; ++node) { | |
e6d9f86d DC |
1021 | if (!o2net_tx_can_proceed(o2net_nn_from_num(node), &sc, &ret)) |
1022 | continue; | |
3ba169cc SM |
1023 | if (!ret) { |
1024 | set_bit(node, map); | |
1025 | sc_put(sc); | |
1026 | } | |
1027 | } | |
1028 | } | |
1029 | EXPORT_SYMBOL_GPL(o2net_fill_node_map); | |
1030 | ||
98211489 ZB |
1031 | int o2net_send_message_vec(u32 msg_type, u32 key, struct kvec *caller_vec, |
1032 | size_t caller_veclen, u8 target_node, int *status) | |
1033 | { | |
50aff040 | 1034 | int ret = 0; |
98211489 ZB |
1035 | struct o2net_msg *msg = NULL; |
1036 | size_t veclen, caller_bytes = 0; | |
1037 | struct kvec *vec = NULL; | |
1038 | struct o2net_sock_container *sc = NULL; | |
1039 | struct o2net_node *nn = o2net_nn_from_num(target_node); | |
1040 | struct o2net_status_wait nsw = { | |
1041 | .ns_node_item = LIST_HEAD_INIT(nsw.ns_node_item), | |
1042 | }; | |
2309e9e0 SM |
1043 | struct o2net_send_tracking nst; |
1044 | ||
1045 | o2net_init_nst(&nst, msg_type, key, current, target_node); | |
98211489 ZB |
1046 | |
1047 | if (o2net_wq == NULL) { | |
1048 | mlog(0, "attempt to tx without o2netd running\n"); | |
1049 | ret = -ESRCH; | |
1050 | goto out; | |
1051 | } | |
1052 | ||
1053 | if (caller_veclen == 0) { | |
1054 | mlog(0, "bad kvec array length\n"); | |
1055 | ret = -EINVAL; | |
1056 | goto out; | |
1057 | } | |
1058 | ||
1059 | caller_bytes = iov_length((struct iovec *)caller_vec, caller_veclen); | |
1060 | if (caller_bytes > O2NET_MAX_PAYLOAD_BYTES) { | |
1061 | mlog(0, "total payload len %zu too large\n", caller_bytes); | |
1062 | ret = -EINVAL; | |
1063 | goto out; | |
1064 | } | |
1065 | ||
1066 | if (target_node == o2nm_this_node()) { | |
1067 | ret = -ELOOP; | |
1068 | goto out; | |
1069 | } | |
1070 | ||
2309e9e0 SM |
1071 | o2net_debug_add_nst(&nst); |
1072 | ||
1073 | o2net_set_nst_sock_time(&nst); | |
1074 | ||
9af0b38f | 1075 | wait_event(nn->nn_sc_wq, o2net_tx_can_proceed(nn, &sc, &ret)); |
98211489 ZB |
1076 | if (ret) |
1077 | goto out; | |
1078 | ||
2309e9e0 SM |
1079 | o2net_set_nst_sock_container(&nst, sc); |
1080 | ||
98211489 ZB |
1081 | veclen = caller_veclen + 1; |
1082 | vec = kmalloc(sizeof(struct kvec) * veclen, GFP_ATOMIC); | |
1083 | if (vec == NULL) { | |
1084 | mlog(0, "failed to %zu element kvec!\n", veclen); | |
1085 | ret = -ENOMEM; | |
1086 | goto out; | |
1087 | } | |
1088 | ||
1089 | msg = kmalloc(sizeof(struct o2net_msg), GFP_ATOMIC); | |
1090 | if (!msg) { | |
1091 | mlog(0, "failed to allocate a o2net_msg!\n"); | |
1092 | ret = -ENOMEM; | |
1093 | goto out; | |
1094 | } | |
1095 | ||
1096 | o2net_init_msg(msg, caller_bytes, msg_type, key); | |
1097 | ||
1098 | vec[0].iov_len = sizeof(struct o2net_msg); | |
1099 | vec[0].iov_base = msg; | |
1100 | memcpy(&vec[1], caller_vec, caller_veclen * sizeof(struct kvec)); | |
1101 | ||
1102 | ret = o2net_prep_nsw(nn, &nsw); | |
1103 | if (ret) | |
1104 | goto out; | |
1105 | ||
1106 | msg->msg_num = cpu_to_be32(nsw.ns_id); | |
2309e9e0 SM |
1107 | o2net_set_nst_msg_id(&nst, nsw.ns_id); |
1108 | ||
1109 | o2net_set_nst_send_time(&nst); | |
98211489 ZB |
1110 | |
1111 | /* finally, convert the message header to network byte-order | |
1112 | * and send */ | |
925037bc | 1113 | mutex_lock(&sc->sc_send_lock); |
98211489 ZB |
1114 | ret = o2net_send_tcp_msg(sc->sc_sock, vec, veclen, |
1115 | sizeof(struct o2net_msg) + caller_bytes); | |
925037bc | 1116 | mutex_unlock(&sc->sc_send_lock); |
98211489 ZB |
1117 | msglog(msg, "sending returned %d\n", ret); |
1118 | if (ret < 0) { | |
1119 | mlog(0, "error returned from o2net_send_tcp_msg=%d\n", ret); | |
1120 | goto out; | |
1121 | } | |
1122 | ||
1123 | /* wait on other node's handler */ | |
2309e9e0 | 1124 | o2net_set_nst_status_time(&nst); |
98211489 ZB |
1125 | wait_event(nsw.ns_wq, o2net_nsw_completed(nn, &nsw)); |
1126 | ||
3c193b38 SM |
1127 | o2net_update_send_stats(&nst, sc); |
1128 | ||
98211489 ZB |
1129 | /* Note that we avoid overwriting the callers status return |
1130 | * variable if a system error was reported on the other | |
1131 | * side. Callers beware. */ | |
1132 | ret = o2net_sys_err_to_errno(nsw.ns_sys_status); | |
1133 | if (status && !ret) | |
1134 | *status = nsw.ns_status; | |
1135 | ||
1136 | mlog(0, "woken, returning system status %d, user status %d\n", | |
1137 | ret, nsw.ns_status); | |
1138 | out: | |
2309e9e0 | 1139 | o2net_debug_del_nst(&nst); /* must be before dropping sc and node */ |
98211489 ZB |
1140 | if (sc) |
1141 | sc_put(sc); | |
d787ab09 TG |
1142 | kfree(vec); |
1143 | kfree(msg); | |
98211489 ZB |
1144 | o2net_complete_nsw(nn, &nsw, 0, 0, 0); |
1145 | return ret; | |
1146 | } | |
1147 | EXPORT_SYMBOL_GPL(o2net_send_message_vec); | |
1148 | ||
1149 | int o2net_send_message(u32 msg_type, u32 key, void *data, u32 len, | |
1150 | u8 target_node, int *status) | |
1151 | { | |
1152 | struct kvec vec = { | |
1153 | .iov_base = data, | |
1154 | .iov_len = len, | |
1155 | }; | |
1156 | return o2net_send_message_vec(msg_type, key, &vec, 1, | |
1157 | target_node, status); | |
1158 | } | |
1159 | EXPORT_SYMBOL_GPL(o2net_send_message); | |
1160 | ||
1161 | static int o2net_send_status_magic(struct socket *sock, struct o2net_msg *hdr, | |
1162 | enum o2net_system_error syserr, int err) | |
1163 | { | |
1164 | struct kvec vec = { | |
1165 | .iov_base = hdr, | |
1166 | .iov_len = sizeof(struct o2net_msg), | |
1167 | }; | |
1168 | ||
1169 | BUG_ON(syserr >= O2NET_ERR_MAX); | |
1170 | ||
1171 | /* leave other fields intact from the incoming message, msg_num | |
1172 | * in particular */ | |
1173 | hdr->sys_status = cpu_to_be32(syserr); | |
1174 | hdr->status = cpu_to_be32(err); | |
1175 | hdr->magic = cpu_to_be16(O2NET_MSG_STATUS_MAGIC); // twiddle the magic | |
1176 | hdr->data_len = 0; | |
1177 | ||
1178 | msglog(hdr, "about to send status magic %d\n", err); | |
1179 | /* hdr has been in host byteorder this whole time */ | |
1180 | return o2net_send_tcp_msg(sock, &vec, 1, sizeof(struct o2net_msg)); | |
1181 | } | |
1182 | ||
1183 | /* this returns -errno if the header was unknown or too large, etc. | |
1184 | * after this is called the buffer us reused for the next message */ | |
1185 | static int o2net_process_message(struct o2net_sock_container *sc, | |
1186 | struct o2net_msg *hdr) | |
1187 | { | |
1188 | struct o2net_node *nn = o2net_nn_from_num(sc->sc_node->nd_num); | |
1189 | int ret = 0, handler_status; | |
1190 | enum o2net_system_error syserr; | |
1191 | struct o2net_msg_handler *nmh = NULL; | |
d74c9803 | 1192 | void *ret_data = NULL; |
98211489 ZB |
1193 | |
1194 | msglog(hdr, "processing message\n"); | |
1195 | ||
1196 | o2net_sc_postpone_idle(sc); | |
1197 | ||
1198 | switch(be16_to_cpu(hdr->magic)) { | |
1199 | case O2NET_MSG_STATUS_MAGIC: | |
1200 | /* special type for returning message status */ | |
1201 | o2net_complete_nsw(nn, NULL, | |
1202 | be32_to_cpu(hdr->msg_num), | |
1203 | be32_to_cpu(hdr->sys_status), | |
1204 | be32_to_cpu(hdr->status)); | |
1205 | goto out; | |
1206 | case O2NET_MSG_KEEP_REQ_MAGIC: | |
1207 | o2net_sendpage(sc, o2net_keep_resp, | |
1208 | sizeof(*o2net_keep_resp)); | |
1209 | goto out; | |
1210 | case O2NET_MSG_KEEP_RESP_MAGIC: | |
1211 | goto out; | |
1212 | case O2NET_MSG_MAGIC: | |
1213 | break; | |
1214 | default: | |
1215 | msglog(hdr, "bad magic\n"); | |
1216 | ret = -EINVAL; | |
1217 | goto out; | |
1218 | break; | |
1219 | } | |
1220 | ||
1221 | /* find a handler for it */ | |
1222 | handler_status = 0; | |
1223 | nmh = o2net_handler_get(be16_to_cpu(hdr->msg_type), | |
1224 | be32_to_cpu(hdr->key)); | |
1225 | if (!nmh) { | |
1226 | mlog(ML_TCP, "couldn't find handler for type %u key %08x\n", | |
1227 | be16_to_cpu(hdr->msg_type), be32_to_cpu(hdr->key)); | |
1228 | syserr = O2NET_ERR_NO_HNDLR; | |
1229 | goto out_respond; | |
1230 | } | |
1231 | ||
1232 | syserr = O2NET_ERR_NONE; | |
1233 | ||
1234 | if (be16_to_cpu(hdr->data_len) > nmh->nh_max_len) | |
1235 | syserr = O2NET_ERR_OVERFLOW; | |
1236 | ||
1237 | if (syserr != O2NET_ERR_NONE) | |
1238 | goto out_respond; | |
1239 | ||
ff1becbf | 1240 | o2net_set_func_start_time(sc); |
98211489 ZB |
1241 | sc->sc_msg_key = be32_to_cpu(hdr->key); |
1242 | sc->sc_msg_type = be16_to_cpu(hdr->msg_type); | |
1243 | handler_status = (nmh->nh_func)(hdr, sizeof(struct o2net_msg) + | |
1244 | be16_to_cpu(hdr->data_len), | |
d74c9803 | 1245 | nmh->nh_func_data, &ret_data); |
ff1becbf | 1246 | o2net_set_func_stop_time(sc); |
98211489 | 1247 | |
e453039f SM |
1248 | o2net_update_recv_stats(sc); |
1249 | ||
98211489 ZB |
1250 | out_respond: |
1251 | /* this destroys the hdr, so don't use it after this */ | |
925037bc | 1252 | mutex_lock(&sc->sc_send_lock); |
98211489 ZB |
1253 | ret = o2net_send_status_magic(sc->sc_sock, hdr, syserr, |
1254 | handler_status); | |
925037bc | 1255 | mutex_unlock(&sc->sc_send_lock); |
98211489 ZB |
1256 | hdr = NULL; |
1257 | mlog(0, "sending handler status %d, syserr %d returned %d\n", | |
1258 | handler_status, syserr, ret); | |
1259 | ||
d74c9803 KH |
1260 | if (nmh) { |
1261 | BUG_ON(ret_data != NULL && nmh->nh_post_func == NULL); | |
1262 | if (nmh->nh_post_func) | |
1263 | (nmh->nh_post_func)(handler_status, nmh->nh_func_data, | |
1264 | ret_data); | |
1265 | } | |
1266 | ||
98211489 ZB |
1267 | out: |
1268 | if (nmh) | |
1269 | o2net_handler_put(nmh); | |
1270 | return ret; | |
1271 | } | |
1272 | ||
1273 | static int o2net_check_handshake(struct o2net_sock_container *sc) | |
1274 | { | |
1275 | struct o2net_handshake *hand = page_address(sc->sc_page); | |
1276 | struct o2net_node *nn = o2net_nn_from_num(sc->sc_node->nd_num); | |
1277 | ||
1278 | if (hand->protocol_version != cpu_to_be64(O2NET_PROTOCOL_VERSION)) { | |
1dfecf81 SM |
1279 | printk(KERN_NOTICE "o2net: " SC_NODEF_FMT " Advertised net " |
1280 | "protocol version %llu but %llu is required. " | |
1281 | "Disconnecting.\n", SC_NODEF_ARGS(sc), | |
1282 | (unsigned long long)be64_to_cpu(hand->protocol_version), | |
1283 | O2NET_PROTOCOL_VERSION); | |
98211489 ZB |
1284 | |
1285 | /* don't bother reconnecting if its the wrong version. */ | |
1286 | o2net_ensure_shutdown(nn, sc, -ENOTCONN); | |
1287 | return -1; | |
1288 | } | |
1289 | ||
828ae6af AB |
1290 | /* |
1291 | * Ensure timeouts are consistent with other nodes, otherwise | |
1292 | * we can end up with one node thinking that the other must be down, | |
1293 | * but isn't. This can ultimately cause corruption. | |
1294 | */ | |
1295 | if (be32_to_cpu(hand->o2net_idle_timeout_ms) != | |
409753bf | 1296 | o2net_idle_timeout()) { |
1dfecf81 SM |
1297 | printk(KERN_NOTICE "o2net: " SC_NODEF_FMT " uses a network " |
1298 | "idle timeout of %u ms, but we use %u ms locally. " | |
1299 | "Disconnecting.\n", SC_NODEF_ARGS(sc), | |
1300 | be32_to_cpu(hand->o2net_idle_timeout_ms), | |
1301 | o2net_idle_timeout()); | |
828ae6af AB |
1302 | o2net_ensure_shutdown(nn, sc, -ENOTCONN); |
1303 | return -1; | |
1304 | } | |
1305 | ||
1306 | if (be32_to_cpu(hand->o2net_keepalive_delay_ms) != | |
409753bf | 1307 | o2net_keepalive_delay()) { |
1dfecf81 SM |
1308 | printk(KERN_NOTICE "o2net: " SC_NODEF_FMT " uses a keepalive " |
1309 | "delay of %u ms, but we use %u ms locally. " | |
1310 | "Disconnecting.\n", SC_NODEF_ARGS(sc), | |
1311 | be32_to_cpu(hand->o2net_keepalive_delay_ms), | |
1312 | o2net_keepalive_delay()); | |
828ae6af AB |
1313 | o2net_ensure_shutdown(nn, sc, -ENOTCONN); |
1314 | return -1; | |
1315 | } | |
1316 | ||
1317 | if (be32_to_cpu(hand->o2hb_heartbeat_timeout_ms) != | |
1318 | O2HB_MAX_WRITE_TIMEOUT_MS) { | |
1dfecf81 SM |
1319 | printk(KERN_NOTICE "o2net: " SC_NODEF_FMT " uses a heartbeat " |
1320 | "timeout of %u ms, but we use %u ms locally. " | |
1321 | "Disconnecting.\n", SC_NODEF_ARGS(sc), | |
1322 | be32_to_cpu(hand->o2hb_heartbeat_timeout_ms), | |
1323 | O2HB_MAX_WRITE_TIMEOUT_MS); | |
828ae6af AB |
1324 | o2net_ensure_shutdown(nn, sc, -ENOTCONN); |
1325 | return -1; | |
1326 | } | |
1327 | ||
98211489 ZB |
1328 | sc->sc_handshake_ok = 1; |
1329 | ||
1330 | spin_lock(&nn->nn_lock); | |
1331 | /* set valid and queue the idle timers only if it hasn't been | |
1332 | * shut down already */ | |
1333 | if (nn->nn_sc == sc) { | |
b5dd8030 | 1334 | o2net_sc_reset_idle_timer(sc); |
5cc3bf27 | 1335 | atomic_set(&nn->nn_timeout, 0); |
98211489 ZB |
1336 | o2net_set_nn_state(nn, sc, 1, 0); |
1337 | } | |
1338 | spin_unlock(&nn->nn_lock); | |
1339 | ||
1340 | /* shift everything up as though it wasn't there */ | |
1341 | sc->sc_page_off -= sizeof(struct o2net_handshake); | |
1342 | if (sc->sc_page_off) | |
1343 | memmove(hand, hand + 1, sc->sc_page_off); | |
1344 | ||
1345 | return 0; | |
1346 | } | |
1347 | ||
1348 | /* this demuxes the queued rx bytes into header or payload bits and calls | |
1349 | * handlers as each full message is read off the socket. it returns -error, | |
1350 | * == 0 eof, or > 0 for progress made.*/ | |
1351 | static int o2net_advance_rx(struct o2net_sock_container *sc) | |
1352 | { | |
1353 | struct o2net_msg *hdr; | |
1354 | int ret = 0; | |
1355 | void *data; | |
1356 | size_t datalen; | |
1357 | ||
1358 | sclog(sc, "receiving\n"); | |
ff1becbf | 1359 | o2net_set_advance_start_time(sc); |
98211489 | 1360 | |
828ae6af AB |
1361 | if (unlikely(sc->sc_handshake_ok == 0)) { |
1362 | if(sc->sc_page_off < sizeof(struct o2net_handshake)) { | |
1363 | data = page_address(sc->sc_page) + sc->sc_page_off; | |
1364 | datalen = sizeof(struct o2net_handshake) - sc->sc_page_off; | |
1365 | ret = o2net_recv_tcp_msg(sc->sc_sock, data, datalen); | |
1366 | if (ret > 0) | |
1367 | sc->sc_page_off += ret; | |
1368 | } | |
1369 | ||
1370 | if (sc->sc_page_off == sizeof(struct o2net_handshake)) { | |
1371 | o2net_check_handshake(sc); | |
1372 | if (unlikely(sc->sc_handshake_ok == 0)) | |
1373 | ret = -EPROTO; | |
1374 | } | |
1375 | goto out; | |
1376 | } | |
1377 | ||
98211489 ZB |
1378 | /* do we need more header? */ |
1379 | if (sc->sc_page_off < sizeof(struct o2net_msg)) { | |
1380 | data = page_address(sc->sc_page) + sc->sc_page_off; | |
1381 | datalen = sizeof(struct o2net_msg) - sc->sc_page_off; | |
1382 | ret = o2net_recv_tcp_msg(sc->sc_sock, data, datalen); | |
1383 | if (ret > 0) { | |
1384 | sc->sc_page_off += ret; | |
98211489 ZB |
1385 | /* only swab incoming here.. we can |
1386 | * only get here once as we cross from | |
1387 | * being under to over */ | |
1388 | if (sc->sc_page_off == sizeof(struct o2net_msg)) { | |
1389 | hdr = page_address(sc->sc_page); | |
1390 | if (be16_to_cpu(hdr->data_len) > | |
1391 | O2NET_MAX_PAYLOAD_BYTES) | |
1392 | ret = -EOVERFLOW; | |
1393 | } | |
1394 | } | |
1395 | if (ret <= 0) | |
1396 | goto out; | |
1397 | } | |
1398 | ||
1399 | if (sc->sc_page_off < sizeof(struct o2net_msg)) { | |
1400 | /* oof, still don't have a header */ | |
1401 | goto out; | |
1402 | } | |
1403 | ||
1404 | /* this was swabbed above when we first read it */ | |
1405 | hdr = page_address(sc->sc_page); | |
1406 | ||
1407 | msglog(hdr, "at page_off %zu\n", sc->sc_page_off); | |
1408 | ||
1409 | /* do we need more payload? */ | |
1410 | if (sc->sc_page_off - sizeof(struct o2net_msg) < be16_to_cpu(hdr->data_len)) { | |
1411 | /* need more payload */ | |
1412 | data = page_address(sc->sc_page) + sc->sc_page_off; | |
1413 | datalen = (sizeof(struct o2net_msg) + be16_to_cpu(hdr->data_len)) - | |
1414 | sc->sc_page_off; | |
1415 | ret = o2net_recv_tcp_msg(sc->sc_sock, data, datalen); | |
1416 | if (ret > 0) | |
1417 | sc->sc_page_off += ret; | |
1418 | if (ret <= 0) | |
1419 | goto out; | |
1420 | } | |
1421 | ||
1422 | if (sc->sc_page_off - sizeof(struct o2net_msg) == be16_to_cpu(hdr->data_len)) { | |
1423 | /* we can only get here once, the first time we read | |
1424 | * the payload.. so set ret to progress if the handler | |
1425 | * works out. after calling this the message is toast */ | |
1426 | ret = o2net_process_message(sc, hdr); | |
1427 | if (ret == 0) | |
1428 | ret = 1; | |
1429 | sc->sc_page_off = 0; | |
1430 | } | |
1431 | ||
1432 | out: | |
1433 | sclog(sc, "ret = %d\n", ret); | |
ff1becbf | 1434 | o2net_set_advance_stop_time(sc); |
98211489 ZB |
1435 | return ret; |
1436 | } | |
1437 | ||
1438 | /* this work func is triggerd by data ready. it reads until it can read no | |
1439 | * more. it interprets 0, eof, as fatal. if data_ready hits while we're doing | |
1440 | * our work the work struct will be marked and we'll be called again. */ | |
c4028958 | 1441 | static void o2net_rx_until_empty(struct work_struct *work) |
98211489 | 1442 | { |
c4028958 DH |
1443 | struct o2net_sock_container *sc = |
1444 | container_of(work, struct o2net_sock_container, sc_rx_work); | |
98211489 ZB |
1445 | int ret; |
1446 | ||
1447 | do { | |
1448 | ret = o2net_advance_rx(sc); | |
1449 | } while (ret > 0); | |
1450 | ||
1451 | if (ret <= 0 && ret != -EAGAIN) { | |
1452 | struct o2net_node *nn = o2net_nn_from_num(sc->sc_node->nd_num); | |
1453 | sclog(sc, "saw error %d, closing\n", ret); | |
1454 | /* not permanent so read failed handshake can retry */ | |
1455 | o2net_ensure_shutdown(nn, sc, 0); | |
1456 | } | |
1457 | ||
1458 | sc_put(sc); | |
1459 | } | |
1460 | ||
1461 | static int o2net_set_nodelay(struct socket *sock) | |
1462 | { | |
1463 | int ret, val = 1; | |
1464 | mm_segment_t oldfs; | |
1465 | ||
1466 | oldfs = get_fs(); | |
1467 | set_fs(KERNEL_DS); | |
1468 | ||
1469 | /* | |
1470 | * Dear unsuspecting programmer, | |
1471 | * | |
1472 | * Don't use sock_setsockopt() for SOL_TCP. It doesn't check its level | |
1473 | * argument and assumes SOL_SOCKET so, say, your TCP_NODELAY will | |
1474 | * silently turn into SO_DEBUG. | |
1475 | * | |
1476 | * Yours, | |
1477 | * Keeper of hilariously fragile interfaces. | |
1478 | */ | |
1479 | ret = sock->ops->setsockopt(sock, SOL_TCP, TCP_NODELAY, | |
1480 | (char __user *)&val, sizeof(val)); | |
1481 | ||
1482 | set_fs(oldfs); | |
1483 | return ret; | |
1484 | } | |
1485 | ||
8e9801df JB |
1486 | static int o2net_set_usertimeout(struct socket *sock) |
1487 | { | |
1488 | int user_timeout = O2NET_TCP_USER_TIMEOUT; | |
1489 | ||
1490 | return kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT, | |
1491 | (char *)&user_timeout, sizeof(user_timeout)); | |
1492 | } | |
1493 | ||
828ae6af AB |
1494 | static void o2net_initialize_handshake(void) |
1495 | { | |
1496 | o2net_hand->o2hb_heartbeat_timeout_ms = cpu_to_be32( | |
1497 | O2HB_MAX_WRITE_TIMEOUT_MS); | |
409753bf | 1498 | o2net_hand->o2net_idle_timeout_ms = cpu_to_be32(o2net_idle_timeout()); |
828ae6af | 1499 | o2net_hand->o2net_keepalive_delay_ms = cpu_to_be32( |
409753bf | 1500 | o2net_keepalive_delay()); |
828ae6af | 1501 | o2net_hand->o2net_reconnect_delay_ms = cpu_to_be32( |
409753bf | 1502 | o2net_reconnect_delay()); |
828ae6af AB |
1503 | } |
1504 | ||
98211489 ZB |
1505 | /* ------------------------------------------------------------ */ |
1506 | ||
1507 | /* called when a connect completes and after a sock is accepted. the | |
1508 | * rx path will see the response and mark the sc valid */ | |
c4028958 | 1509 | static void o2net_sc_connect_completed(struct work_struct *work) |
98211489 | 1510 | { |
c4028958 DH |
1511 | struct o2net_sock_container *sc = |
1512 | container_of(work, struct o2net_sock_container, | |
1513 | sc_connect_work); | |
98211489 ZB |
1514 | |
1515 | mlog(ML_MSG, "sc sending handshake with ver %llu id %llx\n", | |
1516 | (unsigned long long)O2NET_PROTOCOL_VERSION, | |
1517 | (unsigned long long)be64_to_cpu(o2net_hand->connector_id)); | |
1518 | ||
828ae6af | 1519 | o2net_initialize_handshake(); |
98211489 ZB |
1520 | o2net_sendpage(sc, o2net_hand, sizeof(*o2net_hand)); |
1521 | sc_put(sc); | |
1522 | } | |
1523 | ||
1524 | /* this is called as a work_struct func. */ | |
c4028958 | 1525 | static void o2net_sc_send_keep_req(struct work_struct *work) |
98211489 | 1526 | { |
c4028958 DH |
1527 | struct o2net_sock_container *sc = |
1528 | container_of(work, struct o2net_sock_container, | |
1529 | sc_keepalive_work.work); | |
98211489 ZB |
1530 | |
1531 | o2net_sendpage(sc, o2net_keep_req, sizeof(*o2net_keep_req)); | |
1532 | sc_put(sc); | |
1533 | } | |
1534 | ||
1535 | /* socket shutdown does a del_timer_sync against this as it tears down. | |
1536 | * we can't start this timer until we've got to the point in sc buildup | |
1537 | * where shutdown is going to be involved */ | |
1538 | static void o2net_idle_timer(unsigned long data) | |
1539 | { | |
1540 | struct o2net_sock_container *sc = (struct o2net_sock_container *)data; | |
5cc3bf27 | 1541 | struct o2net_node *nn = o2net_nn_from_num(sc->sc_node->nd_num); |
ff1becbf | 1542 | #ifdef CONFIG_DEBUG_FS |
1dfecf81 SM |
1543 | unsigned long msecs = ktime_to_ms(ktime_get()) - |
1544 | ktime_to_ms(sc->sc_tv_timer); | |
1545 | #else | |
1546 | unsigned long msecs = o2net_idle_timeout(); | |
ff1becbf | 1547 | #endif |
98211489 | 1548 | |
1dfecf81 | 1549 | printk(KERN_NOTICE "o2net: Connection to " SC_NODEF_FMT " has been " |
c43c363d JB |
1550 | "idle for %lu.%lu secs.\n", |
1551 | SC_NODEF_ARGS(sc), msecs / 1000, msecs % 1000); | |
98211489 | 1552 | |
c43c363d JB |
1553 | /* idle timerout happen, don't shutdown the connection, but |
1554 | * make fence decision. Maybe the connection can recover before | |
1555 | * the decision is made. | |
5cc3bf27 TM |
1556 | */ |
1557 | atomic_set(&nn->nn_timeout, 1); | |
c43c363d JB |
1558 | o2quo_conn_err(o2net_num_from_nn(nn)); |
1559 | queue_delayed_work(o2net_wq, &nn->nn_still_up, | |
1560 | msecs_to_jiffies(O2NET_QUORUM_DELAY_MS)); | |
1561 | ||
1562 | o2net_sc_reset_idle_timer(sc); | |
5cc3bf27 | 1563 | |
98211489 ZB |
1564 | } |
1565 | ||
b5dd8030 | 1566 | static void o2net_sc_reset_idle_timer(struct o2net_sock_container *sc) |
98211489 ZB |
1567 | { |
1568 | o2net_sc_cancel_delayed_work(sc, &sc->sc_keepalive_work); | |
1569 | o2net_sc_queue_delayed_work(sc, &sc->sc_keepalive_work, | |
409753bf | 1570 | msecs_to_jiffies(o2net_keepalive_delay())); |
ff1becbf | 1571 | o2net_set_sock_timer(sc); |
98211489 | 1572 | mod_timer(&sc->sc_idle_timeout, |
409753bf | 1573 | jiffies + msecs_to_jiffies(o2net_idle_timeout())); |
b5dd8030 JM |
1574 | } |
1575 | ||
1576 | static void o2net_sc_postpone_idle(struct o2net_sock_container *sc) | |
1577 | { | |
c43c363d JB |
1578 | struct o2net_node *nn = o2net_nn_from_num(sc->sc_node->nd_num); |
1579 | ||
1580 | /* clear fence decision since the connection recover from timeout*/ | |
1581 | if (atomic_read(&nn->nn_timeout)) { | |
1582 | o2quo_conn_up(o2net_num_from_nn(nn)); | |
1583 | cancel_delayed_work(&nn->nn_still_up); | |
1584 | atomic_set(&nn->nn_timeout, 0); | |
1585 | } | |
1586 | ||
b5dd8030 JM |
1587 | /* Only push out an existing timer */ |
1588 | if (timer_pending(&sc->sc_idle_timeout)) | |
1589 | o2net_sc_reset_idle_timer(sc); | |
98211489 ZB |
1590 | } |
1591 | ||
1592 | /* this work func is kicked whenever a path sets the nn state which doesn't | |
1593 | * have valid set. This includes seeing hb come up, losing a connection, | |
1594 | * having a connect attempt fail, etc. This centralizes the logic which decides | |
1595 | * if a connect attempt should be made or if we should give up and all future | |
1596 | * transmit attempts should fail */ | |
c4028958 | 1597 | static void o2net_start_connect(struct work_struct *work) |
98211489 | 1598 | { |
c4028958 DH |
1599 | struct o2net_node *nn = |
1600 | container_of(work, struct o2net_node, nn_connect_work.work); | |
98211489 | 1601 | struct o2net_sock_container *sc = NULL; |
b7668c72 | 1602 | struct o2nm_node *node = NULL, *mynode = NULL; |
98211489 ZB |
1603 | struct socket *sock = NULL; |
1604 | struct sockaddr_in myaddr = {0, }, remoteaddr = {0, }; | |
c4028958 | 1605 | int ret = 0, stop; |
5cc3bf27 | 1606 | unsigned int timeout; |
b246d3d1 | 1607 | unsigned int noio_flag; |
98211489 | 1608 | |
b246d3d1 X |
1609 | /* |
1610 | * sock_create allocates the sock with GFP_KERNEL. We must set | |
1611 | * per-process flag PF_MEMALLOC_NOIO so that all allocations done | |
1612 | * by this process are done as if GFP_NOIO was specified. So we | |
1613 | * are not reentering filesystem while doing memory reclaim. | |
1614 | */ | |
1615 | noio_flag = memalloc_noio_save(); | |
98211489 ZB |
1616 | /* if we're greater we initiate tx, otherwise we accept */ |
1617 | if (o2nm_this_node() <= o2net_num_from_nn(nn)) | |
1618 | goto out; | |
1619 | ||
1620 | /* watch for racing with tearing a node down */ | |
1621 | node = o2nm_get_node_by_num(o2net_num_from_nn(nn)); | |
7d65b274 | 1622 | if (node == NULL) |
98211489 | 1623 | goto out; |
98211489 | 1624 | |
b7668c72 | 1625 | mynode = o2nm_get_node_by_num(o2nm_this_node()); |
7d65b274 | 1626 | if (mynode == NULL) |
b7668c72 | 1627 | goto out; |
b7668c72 | 1628 | |
98211489 | 1629 | spin_lock(&nn->nn_lock); |
5cc3bf27 TM |
1630 | /* |
1631 | * see if we already have one pending or have given up. | |
1632 | * For nn_timeout, it is set when we close the connection | |
1633 | * because of the idle time out. So it means that we have | |
1634 | * at least connected to that node successfully once, | |
1635 | * now try to connect to it again. | |
1636 | */ | |
1637 | timeout = atomic_read(&nn->nn_timeout); | |
1638 | stop = (nn->nn_sc || | |
1639 | (nn->nn_persistent_error && | |
1640 | (nn->nn_persistent_error != -ENOTCONN || timeout == 0))); | |
98211489 | 1641 | spin_unlock(&nn->nn_lock); |
c4028958 | 1642 | if (stop) |
98211489 ZB |
1643 | goto out; |
1644 | ||
1645 | nn->nn_last_connect_attempt = jiffies; | |
1646 | ||
1647 | sc = sc_alloc(node); | |
1648 | if (sc == NULL) { | |
1649 | mlog(0, "couldn't allocate sc\n"); | |
1650 | ret = -ENOMEM; | |
1651 | goto out; | |
1652 | } | |
1653 | ||
1654 | ret = sock_create(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock); | |
1655 | if (ret < 0) { | |
1656 | mlog(0, "can't create socket: %d\n", ret); | |
1657 | goto out; | |
1658 | } | |
1659 | sc->sc_sock = sock; /* freed by sc_kref_release */ | |
1660 | ||
1661 | sock->sk->sk_allocation = GFP_ATOMIC; | |
1662 | ||
1663 | myaddr.sin_family = AF_INET; | |
5fdf1e67 | 1664 | myaddr.sin_addr.s_addr = mynode->nd_ipv4_address; |
97bd7919 | 1665 | myaddr.sin_port = htons(0); /* any port */ |
98211489 ZB |
1666 | |
1667 | ret = sock->ops->bind(sock, (struct sockaddr *)&myaddr, | |
1668 | sizeof(myaddr)); | |
1669 | if (ret) { | |
be859405 HH |
1670 | mlog(ML_ERROR, "bind failed with %d at address %pI4\n", |
1671 | ret, &mynode->nd_ipv4_address); | |
98211489 ZB |
1672 | goto out; |
1673 | } | |
1674 | ||
1675 | ret = o2net_set_nodelay(sc->sc_sock); | |
1676 | if (ret) { | |
1677 | mlog(ML_ERROR, "setting TCP_NODELAY failed with %d\n", ret); | |
1678 | goto out; | |
1679 | } | |
1680 | ||
8e9801df JB |
1681 | ret = o2net_set_usertimeout(sock); |
1682 | if (ret) { | |
1683 | mlog(ML_ERROR, "set TCP_USER_TIMEOUT failed with %d\n", ret); | |
1684 | goto out; | |
1685 | } | |
1686 | ||
98211489 ZB |
1687 | o2net_register_callbacks(sc->sc_sock->sk, sc); |
1688 | ||
1689 | spin_lock(&nn->nn_lock); | |
1690 | /* handshake completion will set nn->nn_sc_valid */ | |
1691 | o2net_set_nn_state(nn, sc, 0, 0); | |
1692 | spin_unlock(&nn->nn_lock); | |
1693 | ||
1694 | remoteaddr.sin_family = AF_INET; | |
5fdf1e67 MF |
1695 | remoteaddr.sin_addr.s_addr = node->nd_ipv4_address; |
1696 | remoteaddr.sin_port = node->nd_ipv4_port; | |
98211489 ZB |
1697 | |
1698 | ret = sc->sc_sock->ops->connect(sc->sc_sock, | |
1699 | (struct sockaddr *)&remoteaddr, | |
1700 | sizeof(remoteaddr), | |
1701 | O_NONBLOCK); | |
1702 | if (ret == -EINPROGRESS) | |
1703 | ret = 0; | |
1704 | ||
1705 | out: | |
8dd7903e | 1706 | if (ret && sc) { |
1dfecf81 SM |
1707 | printk(KERN_NOTICE "o2net: Connect attempt to " SC_NODEF_FMT |
1708 | " failed with errno %d\n", SC_NODEF_ARGS(sc), ret); | |
98211489 ZB |
1709 | /* 0 err so that another will be queued and attempted |
1710 | * from set_nn_state */ | |
8dd7903e | 1711 | o2net_ensure_shutdown(nn, sc, 0); |
98211489 ZB |
1712 | } |
1713 | if (sc) | |
1714 | sc_put(sc); | |
1715 | if (node) | |
1716 | o2nm_node_put(node); | |
b7668c72 SM |
1717 | if (mynode) |
1718 | o2nm_node_put(mynode); | |
98211489 | 1719 | |
b246d3d1 | 1720 | memalloc_noio_restore(noio_flag); |
98211489 ZB |
1721 | return; |
1722 | } | |
1723 | ||
c4028958 | 1724 | static void o2net_connect_expired(struct work_struct *work) |
98211489 | 1725 | { |
c4028958 DH |
1726 | struct o2net_node *nn = |
1727 | container_of(work, struct o2net_node, nn_connect_expired.work); | |
98211489 ZB |
1728 | |
1729 | spin_lock(&nn->nn_lock); | |
1730 | if (!nn->nn_sc_valid) { | |
1dfecf81 | 1731 | printk(KERN_NOTICE "o2net: No connection established with " |
5046f18d JB |
1732 | "node %u after %u.%u seconds, check network and" |
1733 | " cluster configuration.\n", | |
b5dd8030 | 1734 | o2net_num_from_nn(nn), |
409753bf JM |
1735 | o2net_idle_timeout() / 1000, |
1736 | o2net_idle_timeout() % 1000); | |
98211489 | 1737 | |
196fe71d | 1738 | o2net_set_nn_state(nn, NULL, 0, 0); |
98211489 ZB |
1739 | } |
1740 | spin_unlock(&nn->nn_lock); | |
1741 | } | |
1742 | ||
c4028958 | 1743 | static void o2net_still_up(struct work_struct *work) |
98211489 | 1744 | { |
c4028958 DH |
1745 | struct o2net_node *nn = |
1746 | container_of(work, struct o2net_node, nn_still_up.work); | |
98211489 ZB |
1747 | |
1748 | o2quo_hb_still_up(o2net_num_from_nn(nn)); | |
1749 | } | |
1750 | ||
1751 | /* ------------------------------------------------------------ */ | |
1752 | ||
1753 | void o2net_disconnect_node(struct o2nm_node *node) | |
1754 | { | |
1755 | struct o2net_node *nn = o2net_nn_from_num(node->nd_num); | |
1756 | ||
1757 | /* don't reconnect until it's heartbeating again */ | |
1758 | spin_lock(&nn->nn_lock); | |
5cc3bf27 | 1759 | atomic_set(&nn->nn_timeout, 0); |
98211489 ZB |
1760 | o2net_set_nn_state(nn, NULL, 0, -ENOTCONN); |
1761 | spin_unlock(&nn->nn_lock); | |
1762 | ||
1763 | if (o2net_wq) { | |
1764 | cancel_delayed_work(&nn->nn_connect_expired); | |
1765 | cancel_delayed_work(&nn->nn_connect_work); | |
1766 | cancel_delayed_work(&nn->nn_still_up); | |
1767 | flush_workqueue(o2net_wq); | |
1768 | } | |
1769 | } | |
1770 | ||
1771 | static void o2net_hb_node_down_cb(struct o2nm_node *node, int node_num, | |
1772 | void *data) | |
1773 | { | |
1774 | o2quo_hb_down(node_num); | |
1775 | ||
0e105d37 SM |
1776 | if (!node) |
1777 | return; | |
1778 | ||
98211489 ZB |
1779 | if (node_num != o2nm_this_node()) |
1780 | o2net_disconnect_node(node); | |
828ae6af AB |
1781 | |
1782 | BUG_ON(atomic_read(&o2net_connected_peers) < 0); | |
98211489 ZB |
1783 | } |
1784 | ||
1785 | static void o2net_hb_node_up_cb(struct o2nm_node *node, int node_num, | |
1786 | void *data) | |
1787 | { | |
1788 | struct o2net_node *nn = o2net_nn_from_num(node_num); | |
1789 | ||
1790 | o2quo_hb_up(node_num); | |
1791 | ||
0e105d37 SM |
1792 | BUG_ON(!node); |
1793 | ||
98211489 ZB |
1794 | /* ensure an immediate connect attempt */ |
1795 | nn->nn_last_connect_attempt = jiffies - | |
409753bf | 1796 | (msecs_to_jiffies(o2net_reconnect_delay()) + 1); |
98211489 ZB |
1797 | |
1798 | if (node_num != o2nm_this_node()) { | |
98211489 ZB |
1799 | /* believe it or not, accept and node hearbeating testing |
1800 | * can succeed for this node before we got here.. so | |
1801 | * only use set_nn_state to clear the persistent error | |
1802 | * if that hasn't already happened */ | |
1803 | spin_lock(&nn->nn_lock); | |
5cc3bf27 | 1804 | atomic_set(&nn->nn_timeout, 0); |
98211489 ZB |
1805 | if (nn->nn_persistent_error) |
1806 | o2net_set_nn_state(nn, NULL, 0, 0); | |
1807 | spin_unlock(&nn->nn_lock); | |
1808 | } | |
1809 | } | |
1810 | ||
1811 | void o2net_unregister_hb_callbacks(void) | |
1812 | { | |
14829422 JB |
1813 | o2hb_unregister_callback(NULL, &o2net_hb_up); |
1814 | o2hb_unregister_callback(NULL, &o2net_hb_down); | |
98211489 ZB |
1815 | } |
1816 | ||
1817 | int o2net_register_hb_callbacks(void) | |
1818 | { | |
1819 | int ret; | |
1820 | ||
1821 | o2hb_setup_callback(&o2net_hb_down, O2HB_NODE_DOWN_CB, | |
1822 | o2net_hb_node_down_cb, NULL, O2NET_HB_PRI); | |
1823 | o2hb_setup_callback(&o2net_hb_up, O2HB_NODE_UP_CB, | |
1824 | o2net_hb_node_up_cb, NULL, O2NET_HB_PRI); | |
1825 | ||
14829422 | 1826 | ret = o2hb_register_callback(NULL, &o2net_hb_up); |
98211489 | 1827 | if (ret == 0) |
14829422 | 1828 | ret = o2hb_register_callback(NULL, &o2net_hb_down); |
98211489 ZB |
1829 | |
1830 | if (ret) | |
1831 | o2net_unregister_hb_callbacks(); | |
1832 | ||
1833 | return ret; | |
1834 | } | |
1835 | ||
1836 | /* ------------------------------------------------------------ */ | |
1837 | ||
79deb3c1 | 1838 | static int o2net_accept_one(struct socket *sock, int *more) |
98211489 ZB |
1839 | { |
1840 | int ret, slen; | |
1841 | struct sockaddr_in sin; | |
1842 | struct socket *new_sock = NULL; | |
1843 | struct o2nm_node *node = NULL; | |
415cf32c | 1844 | struct o2nm_node *local_node = NULL; |
98211489 ZB |
1845 | struct o2net_sock_container *sc = NULL; |
1846 | struct o2net_node *nn; | |
b246d3d1 X |
1847 | unsigned int noio_flag; |
1848 | ||
1849 | /* | |
1850 | * sock_create_lite allocates the sock with GFP_KERNEL. We must set | |
1851 | * per-process flag PF_MEMALLOC_NOIO so that all allocations done | |
1852 | * by this process are done as if GFP_NOIO was specified. So we | |
1853 | * are not reentering filesystem while doing memory reclaim. | |
1854 | */ | |
1855 | noio_flag = memalloc_noio_save(); | |
98211489 ZB |
1856 | |
1857 | BUG_ON(sock == NULL); | |
79deb3c1 | 1858 | *more = 0; |
98211489 ZB |
1859 | ret = sock_create_lite(sock->sk->sk_family, sock->sk->sk_type, |
1860 | sock->sk->sk_protocol, &new_sock); | |
1861 | if (ret) | |
1862 | goto out; | |
1863 | ||
1864 | new_sock->type = sock->type; | |
1865 | new_sock->ops = sock->ops; | |
1866 | ret = sock->ops->accept(sock, new_sock, O_NONBLOCK); | |
1867 | if (ret < 0) | |
1868 | goto out; | |
1869 | ||
79deb3c1 | 1870 | *more = 1; |
98211489 ZB |
1871 | new_sock->sk->sk_allocation = GFP_ATOMIC; |
1872 | ||
1873 | ret = o2net_set_nodelay(new_sock); | |
1874 | if (ret) { | |
1875 | mlog(ML_ERROR, "setting TCP_NODELAY failed with %d\n", ret); | |
1876 | goto out; | |
1877 | } | |
1878 | ||
8e9801df JB |
1879 | ret = o2net_set_usertimeout(new_sock); |
1880 | if (ret) { | |
1881 | mlog(ML_ERROR, "set TCP_USER_TIMEOUT failed with %d\n", ret); | |
1882 | goto out; | |
1883 | } | |
1884 | ||
98211489 ZB |
1885 | slen = sizeof(sin); |
1886 | ret = new_sock->ops->getname(new_sock, (struct sockaddr *) &sin, | |
1887 | &slen, 1); | |
1888 | if (ret < 0) | |
1889 | goto out; | |
1890 | ||
97bd7919 | 1891 | node = o2nm_get_node_by_ip(sin.sin_addr.s_addr); |
98211489 | 1892 | if (node == NULL) { |
1dfecf81 SM |
1893 | printk(KERN_NOTICE "o2net: Attempt to connect from unknown " |
1894 | "node at %pI4:%d\n", &sin.sin_addr.s_addr, | |
1895 | ntohs(sin.sin_port)); | |
98211489 ZB |
1896 | ret = -EINVAL; |
1897 | goto out; | |
1898 | } | |
1899 | ||
415cf32c TY |
1900 | if (o2nm_this_node() >= node->nd_num) { |
1901 | local_node = o2nm_get_node_by_num(o2nm_this_node()); | |
03dbe88a JQ |
1902 | if (local_node) |
1903 | printk(KERN_NOTICE "o2net: Unexpected connect attempt " | |
1904 | "seen at node '%s' (%u, %pI4:%d) from " | |
1905 | "node '%s' (%u, %pI4:%d)\n", | |
1906 | local_node->nd_name, local_node->nd_num, | |
1907 | &(local_node->nd_ipv4_address), | |
1908 | ntohs(local_node->nd_ipv4_port), | |
1909 | node->nd_name, | |
1910 | node->nd_num, &sin.sin_addr.s_addr, | |
1911 | ntohs(sin.sin_port)); | |
98211489 ZB |
1912 | ret = -EINVAL; |
1913 | goto out; | |
1914 | } | |
1915 | ||
1916 | /* this happens all the time when the other node sees our heartbeat | |
1917 | * and tries to connect before we see their heartbeat */ | |
1918 | if (!o2hb_check_node_heartbeating_from_callback(node->nd_num)) { | |
1919 | mlog(ML_CONN, "attempt to connect from node '%s' at " | |
be859405 HH |
1920 | "%pI4:%d but it isn't heartbeating\n", |
1921 | node->nd_name, &sin.sin_addr.s_addr, | |
97bd7919 | 1922 | ntohs(sin.sin_port)); |
98211489 ZB |
1923 | ret = -EINVAL; |
1924 | goto out; | |
1925 | } | |
1926 | ||
1927 | nn = o2net_nn_from_num(node->nd_num); | |
1928 | ||
1929 | spin_lock(&nn->nn_lock); | |
1930 | if (nn->nn_sc) | |
1931 | ret = -EBUSY; | |
1932 | else | |
1933 | ret = 0; | |
1934 | spin_unlock(&nn->nn_lock); | |
1935 | if (ret) { | |
1dfecf81 SM |
1936 | printk(KERN_NOTICE "o2net: Attempt to connect from node '%s' " |
1937 | "at %pI4:%d but it already has an open connection\n", | |
1938 | node->nd_name, &sin.sin_addr.s_addr, | |
1939 | ntohs(sin.sin_port)); | |
98211489 ZB |
1940 | goto out; |
1941 | } | |
1942 | ||
1943 | sc = sc_alloc(node); | |
1944 | if (sc == NULL) { | |
1945 | ret = -ENOMEM; | |
1946 | goto out; | |
1947 | } | |
1948 | ||
1949 | sc->sc_sock = new_sock; | |
1950 | new_sock = NULL; | |
1951 | ||
1952 | spin_lock(&nn->nn_lock); | |
5cc3bf27 | 1953 | atomic_set(&nn->nn_timeout, 0); |
98211489 ZB |
1954 | o2net_set_nn_state(nn, sc, 0, 0); |
1955 | spin_unlock(&nn->nn_lock); | |
1956 | ||
1957 | o2net_register_callbacks(sc->sc_sock->sk, sc); | |
1958 | o2net_sc_queue_work(sc, &sc->sc_rx_work); | |
1959 | ||
828ae6af | 1960 | o2net_initialize_handshake(); |
98211489 ZB |
1961 | o2net_sendpage(sc, o2net_hand, sizeof(*o2net_hand)); |
1962 | ||
1963 | out: | |
1964 | if (new_sock) | |
1965 | sock_release(new_sock); | |
1966 | if (node) | |
1967 | o2nm_node_put(node); | |
415cf32c TY |
1968 | if (local_node) |
1969 | o2nm_node_put(local_node); | |
98211489 ZB |
1970 | if (sc) |
1971 | sc_put(sc); | |
b246d3d1 X |
1972 | |
1973 | memalloc_noio_restore(noio_flag); | |
98211489 ZB |
1974 | return ret; |
1975 | } | |
1976 | ||
79deb3c1 TS |
1977 | /* |
1978 | * This function is invoked in response to one or more | |
1979 | * pending accepts at softIRQ level. We must drain the | |
1980 | * entire que before returning. | |
1981 | */ | |
1982 | ||
c4028958 | 1983 | static void o2net_accept_many(struct work_struct *work) |
98211489 | 1984 | { |
c4028958 | 1985 | struct socket *sock = o2net_listen_sock; |
79deb3c1 TS |
1986 | int more; |
1987 | int err; | |
1988 | ||
1989 | /* | |
1990 | * It is critical to note that due to interrupt moderation | |
1991 | * at the network driver level, we can't assume to get a | |
1992 | * softIRQ for every single conn since tcp SYN packets | |
1993 | * can arrive back-to-back, and therefore many pending | |
1994 | * accepts may result in just 1 softIRQ. If we terminate | |
1995 | * the o2net_accept_one() loop upon seeing an err, what happens | |
1996 | * to the rest of the conns in the queue? If no new SYN | |
1997 | * arrives for hours, no softIRQ will be delivered, | |
1998 | * and the connections will just sit in the queue. | |
1999 | */ | |
2000 | ||
2001 | for (;;) { | |
2002 | err = o2net_accept_one(sock, &more); | |
2003 | if (!more) | |
2004 | break; | |
98211489 | 2005 | cond_resched(); |
79deb3c1 | 2006 | } |
98211489 ZB |
2007 | } |
2008 | ||
676d2369 | 2009 | static void o2net_listen_data_ready(struct sock *sk) |
98211489 | 2010 | { |
676d2369 | 2011 | void (*ready)(struct sock *sk); |
98211489 | 2012 | |
e1daca28 | 2013 | read_lock_bh(&sk->sk_callback_lock); |
98211489 ZB |
2014 | ready = sk->sk_user_data; |
2015 | if (ready == NULL) { /* check for teardown race */ | |
2016 | ready = sk->sk_data_ready; | |
2017 | goto out; | |
2018 | } | |
2019 | ||
da8ded40 TS |
2020 | /* This callback may called twice when a new connection |
2021 | * is being established as a child socket inherits everything | |
2022 | * from a parent LISTEN socket, including the data_ready cb of | |
2023 | * the parent. This leads to a hazard. In o2net_accept_one() | |
2024 | * we are still initializing the child socket but have not | |
2025 | * changed the inherited data_ready callback yet when | |
2026 | * data starts arriving. | |
2027 | * We avoid this hazard by checking the state. | |
2028 | * For the listening socket, the state will be TCP_LISTEN; for the new | |
2029 | * socket, will be TCP_ESTABLISHED. Also, in this case, | |
2030 | * sk->sk_user_data is not a valid function pointer. | |
2031 | */ | |
2032 | ||
98211489 | 2033 | if (sk->sk_state == TCP_LISTEN) { |
98211489 | 2034 | queue_work(o2net_wq, &o2net_listen_work); |
da8ded40 TS |
2035 | } else { |
2036 | ready = NULL; | |
98211489 ZB |
2037 | } |
2038 | ||
2039 | out: | |
e1daca28 | 2040 | read_unlock_bh(&sk->sk_callback_lock); |
da8ded40 | 2041 | if (ready != NULL) |
676d2369 | 2042 | ready(sk); |
98211489 ZB |
2043 | } |
2044 | ||
ab81afd3 | 2045 | static int o2net_open_listening_sock(__be32 addr, __be16 port) |
98211489 ZB |
2046 | { |
2047 | struct socket *sock = NULL; | |
2048 | int ret; | |
2049 | struct sockaddr_in sin = { | |
2050 | .sin_family = PF_INET, | |
5fdf1e67 MF |
2051 | .sin_addr = { .s_addr = addr }, |
2052 | .sin_port = port, | |
98211489 ZB |
2053 | }; |
2054 | ||
2055 | ret = sock_create(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock); | |
2056 | if (ret < 0) { | |
1dfecf81 | 2057 | printk(KERN_ERR "o2net: Error %d while creating socket\n", ret); |
98211489 ZB |
2058 | goto out; |
2059 | } | |
2060 | ||
2061 | sock->sk->sk_allocation = GFP_ATOMIC; | |
2062 | ||
2063 | write_lock_bh(&sock->sk->sk_callback_lock); | |
2064 | sock->sk->sk_user_data = sock->sk->sk_data_ready; | |
2065 | sock->sk->sk_data_ready = o2net_listen_data_ready; | |
2066 | write_unlock_bh(&sock->sk->sk_callback_lock); | |
2067 | ||
2068 | o2net_listen_sock = sock; | |
c4028958 | 2069 | INIT_WORK(&o2net_listen_work, o2net_accept_many); |
98211489 | 2070 | |
4a17fd52 | 2071 | sock->sk->sk_reuse = SK_CAN_REUSE; |
98211489 ZB |
2072 | ret = sock->ops->bind(sock, (struct sockaddr *)&sin, sizeof(sin)); |
2073 | if (ret < 0) { | |
1dfecf81 SM |
2074 | printk(KERN_ERR "o2net: Error %d while binding socket at " |
2075 | "%pI4:%u\n", ret, &addr, ntohs(port)); | |
98211489 ZB |
2076 | goto out; |
2077 | } | |
2078 | ||
2079 | ret = sock->ops->listen(sock, 64); | |
1dfecf81 SM |
2080 | if (ret < 0) |
2081 | printk(KERN_ERR "o2net: Error %d while listening on %pI4:%u\n", | |
2082 | ret, &addr, ntohs(port)); | |
98211489 ZB |
2083 | |
2084 | out: | |
2085 | if (ret) { | |
2086 | o2net_listen_sock = NULL; | |
2087 | if (sock) | |
2088 | sock_release(sock); | |
2089 | } | |
2090 | return ret; | |
2091 | } | |
2092 | ||
2093 | /* | |
2094 | * called from node manager when we should bring up our network listening | |
2095 | * socket. node manager handles all the serialization to only call this | |
2096 | * once and to match it with o2net_stop_listening(). note, | |
2097 | * o2nm_this_node() doesn't work yet as we're being called while it | |
2098 | * is being set up. | |
2099 | */ | |
2100 | int o2net_start_listening(struct o2nm_node *node) | |
2101 | { | |
2102 | int ret = 0; | |
2103 | ||
2104 | BUG_ON(o2net_wq != NULL); | |
2105 | BUG_ON(o2net_listen_sock != NULL); | |
2106 | ||
2107 | mlog(ML_KTHREAD, "starting o2net thread...\n"); | |
bf940776 | 2108 | o2net_wq = alloc_ordered_workqueue("o2net", WQ_MEM_RECLAIM); |
98211489 ZB |
2109 | if (o2net_wq == NULL) { |
2110 | mlog(ML_ERROR, "unable to launch o2net thread\n"); | |
2111 | return -ENOMEM; /* ? */ | |
2112 | } | |
2113 | ||
ab81afd3 SM |
2114 | ret = o2net_open_listening_sock(node->nd_ipv4_address, |
2115 | node->nd_ipv4_port); | |
98211489 ZB |
2116 | if (ret) { |
2117 | destroy_workqueue(o2net_wq); | |
2118 | o2net_wq = NULL; | |
2119 | } else | |
2120 | o2quo_conn_up(node->nd_num); | |
2121 | ||
2122 | return ret; | |
2123 | } | |
2124 | ||
2125 | /* again, o2nm_this_node() doesn't work here as we're involved in | |
2126 | * tearing it down */ | |
2127 | void o2net_stop_listening(struct o2nm_node *node) | |
2128 | { | |
2129 | struct socket *sock = o2net_listen_sock; | |
2130 | size_t i; | |
2131 | ||
2132 | BUG_ON(o2net_wq == NULL); | |
2133 | BUG_ON(o2net_listen_sock == NULL); | |
2134 | ||
2135 | /* stop the listening socket from generating work */ | |
2136 | write_lock_bh(&sock->sk->sk_callback_lock); | |
2137 | sock->sk->sk_data_ready = sock->sk->sk_user_data; | |
2138 | sock->sk->sk_user_data = NULL; | |
2139 | write_unlock_bh(&sock->sk->sk_callback_lock); | |
2140 | ||
2141 | for (i = 0; i < ARRAY_SIZE(o2net_nodes); i++) { | |
2142 | struct o2nm_node *node = o2nm_get_node_by_num(i); | |
2143 | if (node) { | |
2144 | o2net_disconnect_node(node); | |
2145 | o2nm_node_put(node); | |
2146 | } | |
2147 | } | |
2148 | ||
2149 | /* finish all work and tear down the work queue */ | |
2150 | mlog(ML_KTHREAD, "waiting for o2net thread to exit....\n"); | |
2151 | destroy_workqueue(o2net_wq); | |
2152 | o2net_wq = NULL; | |
2153 | ||
2154 | sock_release(o2net_listen_sock); | |
2155 | o2net_listen_sock = NULL; | |
2156 | ||
2157 | o2quo_conn_err(node->nd_num); | |
2158 | } | |
2159 | ||
2160 | /* ------------------------------------------------------------ */ | |
2161 | ||
2162 | int o2net_init(void) | |
2163 | { | |
2164 | unsigned long i; | |
2165 | ||
2166 | o2quo_init(); | |
2167 | ||
2309e9e0 | 2168 | if (o2net_debugfs_init()) |
4a4e07c1 | 2169 | goto out; |
2309e9e0 | 2170 | |
cd861280 RD |
2171 | o2net_hand = kzalloc(sizeof(struct o2net_handshake), GFP_KERNEL); |
2172 | o2net_keep_req = kzalloc(sizeof(struct o2net_msg), GFP_KERNEL); | |
2173 | o2net_keep_resp = kzalloc(sizeof(struct o2net_msg), GFP_KERNEL); | |
4a4e07c1 JQ |
2174 | if (!o2net_hand || !o2net_keep_req || !o2net_keep_resp) |
2175 | goto out; | |
98211489 ZB |
2176 | |
2177 | o2net_hand->protocol_version = cpu_to_be64(O2NET_PROTOCOL_VERSION); | |
2178 | o2net_hand->connector_id = cpu_to_be64(1); | |
2179 | ||
2180 | o2net_keep_req->magic = cpu_to_be16(O2NET_MSG_KEEP_REQ_MAGIC); | |
2181 | o2net_keep_resp->magic = cpu_to_be16(O2NET_MSG_KEEP_RESP_MAGIC); | |
2182 | ||
2183 | for (i = 0; i < ARRAY_SIZE(o2net_nodes); i++) { | |
2184 | struct o2net_node *nn = o2net_nn_from_num(i); | |
2185 | ||
5cc3bf27 | 2186 | atomic_set(&nn->nn_timeout, 0); |
98211489 | 2187 | spin_lock_init(&nn->nn_lock); |
c4028958 DH |
2188 | INIT_DELAYED_WORK(&nn->nn_connect_work, o2net_start_connect); |
2189 | INIT_DELAYED_WORK(&nn->nn_connect_expired, | |
2190 | o2net_connect_expired); | |
2191 | INIT_DELAYED_WORK(&nn->nn_still_up, o2net_still_up); | |
98211489 ZB |
2192 | /* until we see hb from a node we'll return einval */ |
2193 | nn->nn_persistent_error = -ENOTCONN; | |
2194 | init_waitqueue_head(&nn->nn_sc_wq); | |
2195 | idr_init(&nn->nn_status_idr); | |
2196 | INIT_LIST_HEAD(&nn->nn_status_list); | |
2197 | } | |
2198 | ||
2199 | return 0; | |
4a4e07c1 JQ |
2200 | |
2201 | out: | |
2202 | kfree(o2net_hand); | |
2203 | kfree(o2net_keep_req); | |
2204 | kfree(o2net_keep_resp); | |
fce56d84 | 2205 | o2net_debugfs_exit(); |
4a4e07c1 JQ |
2206 | o2quo_exit(); |
2207 | return -ENOMEM; | |
98211489 ZB |
2208 | } |
2209 | ||
2210 | void o2net_exit(void) | |
2211 | { | |
2212 | o2quo_exit(); | |
2213 | kfree(o2net_hand); | |
2214 | kfree(o2net_keep_req); | |
2215 | kfree(o2net_keep_resp); | |
2309e9e0 | 2216 | o2net_debugfs_exit(); |
98211489 | 2217 | } |