2 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
4 * Copyright (c) 2011, Whamcloud, Inc.
6 * Author: Zach Brown <zab@zabbo.net>
7 * Author: Peter J. Braam <braam@clusterfs.com>
8 * Author: Phil Schwan <phil@clusterfs.com>
9 * Author: Eric Barton <eric@bartonsoftware.com>
11 * This file is part of Lustre, http://www.lustre.org
13 * Portals is free software; you can redistribute it and/or
14 * modify it under the terms of version 2 of the GNU General Public
15 * License as published by the Free Software Foundation.
17 * Portals is distributed in the hope that it will be useful,
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
20 * GNU General Public License for more details.
22 * You should have received a copy of the GNU General Public License
23 * along with Portals; if not, write to the Free Software
24 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
28 #define DEBUG_PORTAL_ALLOC
29 #define DEBUG_SUBSYSTEM S_LND
31 #if defined(__linux__)
32 #include "socklnd_lib-linux.h"
33 #elif defined(__APPLE__)
34 #include "socklnd_lib-darwin.h"
35 #elif defined(__WINNT__)
36 #include "socklnd_lib-winnt.h"
38 #error Unsupported Operating System
41 #include <libcfs/libcfs.h>
42 #include <lnet/lnet.h>
43 #include <lnet/lib-lnet.h>
44 #include <lnet/socklnd.h>
45 #include <lnet/lnet-sysctl.h>
47 #define SOCKNAL_PEER_HASH_SIZE 101 /* # peer lists */
48 #define SOCKNAL_RESCHED 100 /* # scheduler loops before reschedule */
49 #define SOCKNAL_INSANITY_RECONN 5000 /* connd is trying on reconn infinitely */
50 #define SOCKNAL_ENOMEM_RETRY CFS_TICK /* jiffies between retries */
52 #define SOCKNAL_SINGLE_FRAG_TX 0 /* disable multi-fragment sends */
53 #define SOCKNAL_SINGLE_FRAG_RX 0 /* disable multi-fragment receives */
55 #define SOCKNAL_VERSION_DEBUG 0 /* enable protocol version debugging */
57 /* risk kmap deadlock on multi-frag I/O (backs off to single-frag if disabled).
58 * no risk if we're not running on a CONFIG_HIGHMEM platform. */
60 # define SOCKNAL_RISK_KMAP_DEADLOCK 0
62 # define SOCKNAL_RISK_KMAP_DEADLOCK 1
65 struct ksock_sched_info;
67 typedef struct /* per scheduler state */
69 spinlock_t kss_lock; /* serialise */
70 cfs_list_t kss_rx_conns; /* conn waiting to be read */
71 /* conn waiting to be written */
72 cfs_list_t kss_tx_conns;
73 /* zombie noop tx list */
74 cfs_list_t kss_zombie_noop_txs;
75 cfs_waitq_t kss_waitq; /* where scheduler sleeps */
76 /* # connections assigned to this scheduler */
78 struct ksock_sched_info *kss_info; /* owner of it */
79 #if !SOCKNAL_SINGLE_FRAG_RX
80 struct page *kss_rx_scratch_pgs[LNET_MAX_IOV];
82 #if !SOCKNAL_SINGLE_FRAG_TX || !SOCKNAL_SINGLE_FRAG_RX
83 struct iovec kss_scratch_iov[LNET_MAX_IOV];
87 struct ksock_sched_info {
88 int ksi_nthreads_max; /* max allowed threads */
89 int ksi_nthreads; /* number of threads */
90 int ksi_cpt; /* CPT id */
91 ksock_sched_t *ksi_scheds; /* array of schedulers */
94 #define KSOCK_CPT_SHIFT 16
95 #define KSOCK_THREAD_ID(cpt, sid) (((cpt) << KSOCK_CPT_SHIFT) | (sid))
96 #define KSOCK_THREAD_CPT(id) ((id) >> KSOCK_CPT_SHIFT)
97 #define KSOCK_THREAD_SID(id) ((id) & ((1UL << KSOCK_CPT_SHIFT) - 1))
99 typedef struct /* in-use interface */
101 __u32 ksni_ipaddr; /* interface's IP address */
102 __u32 ksni_netmask; /* interface's network mask */
103 int ksni_nroutes; /* # routes using (active) */
104 int ksni_npeers; /* # peers using (passive) */
105 char ksni_name[IFNAMSIZ]; /* interface name */
110 /* "stuck" socket timeout (seconds) */
112 /* # scheduler threads in each pool while starting */
114 int *ksnd_nconnds; /* # connection daemons */
115 int *ksnd_nconnds_max; /* max # connection daemons */
116 int *ksnd_min_reconnectms; /* first connection retry after (ms)... */
117 int *ksnd_max_reconnectms; /* ...exponentially increasing to this */
118 int *ksnd_eager_ack; /* make TCP ack eagerly? */
119 int *ksnd_typed_conns; /* drive sockets by type? */
120 int *ksnd_min_bulk; /* smallest "large" message */
121 int *ksnd_tx_buffer_size; /* socket tx buffer size */
122 int *ksnd_rx_buffer_size; /* socket rx buffer size */
123 int *ksnd_nagle; /* enable NAGLE? */
124 int *ksnd_round_robin; /* round robin for multiple interfaces */
125 int *ksnd_keepalive; /* # secs for sending keepalive NOOP */
126 int *ksnd_keepalive_idle; /* # idle secs before 1st probe */
127 int *ksnd_keepalive_count; /* # probes */
128 int *ksnd_keepalive_intvl; /* time between probes */
129 int *ksnd_credits; /* # concurrent sends */
130 int *ksnd_peertxcredits; /* # concurrent sends to 1 peer */
131 int *ksnd_peerrtrcredits; /* # per-peer router buffer credits */
132 int *ksnd_peertimeout; /* seconds to consider peer dead */
133 int *ksnd_enable_csum; /* enable check sum */
134 int *ksnd_inject_csum_error; /* set non-zero to inject checksum error */
135 int *ksnd_nonblk_zcack; /* always send zc-ack on non-blocking connection */
136 unsigned int *ksnd_zc_min_payload; /* minimum zero copy payload size */
137 int *ksnd_zc_recv; /* enable ZC receive (for Chelsio TOE) */
138 int *ksnd_zc_recv_min_nfrags; /* minimum # of fragments to enable ZC receive */
140 int *ksnd_irq_affinity; /* enable IRQ affinity? */
142 #ifdef SOCKNAL_BACKOFF
143 int *ksnd_backoff_init; /* initial TCP backoff */
144 int *ksnd_backoff_max; /* maximum TCP backoff */
146 #if SOCKNAL_VERSION_DEBUG
147 int *ksnd_protocol; /* protocol version */
149 #if defined(CONFIG_SYSCTL) && !CFS_SYSFS_MODULE_PARM
150 cfs_sysctl_table_header_t *ksnd_sysctl; /* sysctl interface */
156 __u64 ksnn_incarnation; /* my epoch */
157 spinlock_t ksnn_lock; /* serialise */
158 cfs_list_t ksnn_list; /* chain on global list */
159 int ksnn_npeers; /* # peers */
160 int ksnn_shutdown; /* shutting down? */
161 int ksnn_ninterfaces; /* IP interfaces */
162 ksock_interface_t ksnn_interfaces[LNET_MAX_INTERFACES];
166 #define SOCKNAL_CONND_TIMEOUT 120
167 /** reserved thread for accepting & creating new connd */
168 #define SOCKNAL_CONND_RESV 1
172 int ksnd_init; /* initialisation state */
173 int ksnd_nnets; /* # networks set up */
174 cfs_list_t ksnd_nets; /* list of nets */
175 /* stabilize peer/conn ops */
176 rwlock_t ksnd_global_lock;
177 /* hash table of all my known peers */
178 cfs_list_t *ksnd_peers;
179 int ksnd_peer_hash_size; /* size of ksnd_peers */
181 int ksnd_nthreads; /* # live threads */
182 int ksnd_shuttingdown; /* tell threads to exit */
183 /* schedulers information */
184 struct ksock_sched_info **ksnd_sched_info;
186 cfs_atomic_t ksnd_nactive_txs; /* #active txs */
188 cfs_list_t ksnd_deathrow_conns; /* conns to close: reaper_lock*/
189 cfs_list_t ksnd_zombie_conns; /* conns to free: reaper_lock */
190 cfs_list_t ksnd_enomem_conns; /* conns to retry: reaper_lock*/
191 cfs_waitq_t ksnd_reaper_waitq; /* reaper sleeps here */
192 cfs_time_t ksnd_reaper_waketime;/* when reaper will wake */
193 spinlock_t ksnd_reaper_lock; /* serialise */
195 int ksnd_enomem_tx; /* test ENOMEM sender */
196 int ksnd_stall_tx; /* test sluggish sender */
197 int ksnd_stall_rx; /* test sluggish receiver */
199 cfs_list_t ksnd_connd_connreqs; /* incoming connection requests */
200 cfs_list_t ksnd_connd_routes; /* routes waiting to be connected */
201 cfs_waitq_t ksnd_connd_waitq; /* connds sleep here */
202 int ksnd_connd_connecting;/* # connds connecting */
203 /** time stamp of the last failed connecting attempt */
204 long ksnd_connd_failed_stamp;
205 /** # starting connd */
206 unsigned ksnd_connd_starting;
207 /** time stamp of the last starting connd */
208 long ksnd_connd_starting_stamp;
209 /** # running connd */
210 unsigned ksnd_connd_running;
211 spinlock_t ksnd_connd_lock; /* serialise */
213 cfs_list_t ksnd_idle_noop_txs; /* list head for freed noop tx */
214 spinlock_t ksnd_tx_lock; /* serialise, g_lock unsafe */
218 #define SOCKNAL_INIT_NOTHING 0
219 #define SOCKNAL_INIT_DATA 1
220 #define SOCKNAL_INIT_ALL 2
222 /* A packet just assembled for transmission is represented by 1 or more
223 * struct iovec fragments (the first frag contains the portals header),
224 * followed by 0 or more lnet_kiov_t fragments.
226 * On the receive side, initially 1 struct iovec fragment is posted for
227 * receive (the header). Once the header has been received, the payload is
228 * received into either struct iovec or lnet_kiov_t fragments, depending on
229 * what the header matched or whether the message needs forwarding. */
231 struct ksock_conn; /* forward ref */
232 struct ksock_peer; /* forward ref */
233 struct ksock_route; /* forward ref */
234 struct ksock_proto; /* forward ref */
236 typedef struct /* transmit packet */
238 cfs_list_t tx_list; /* queue on conn for transmission etc */
239 cfs_list_t tx_zc_list; /* queue on peer for ZC request */
240 cfs_atomic_t tx_refcount; /* tx reference count */
241 int tx_nob; /* # packet bytes */
242 int tx_resid; /* residual bytes */
243 int tx_niov; /* # packet iovec frags */
244 struct iovec *tx_iov; /* packet iovec frags */
245 int tx_nkiov; /* # packet page frags */
246 unsigned short tx_zc_aborted; /* aborted ZC request */
247 unsigned short tx_zc_capable:1; /* payload is large enough for ZC */
248 unsigned short tx_zc_checked:1; /* Have I checked if I should ZC? */
249 unsigned short tx_nonblk:1; /* it's a non-blocking ACK */
250 lnet_kiov_t *tx_kiov; /* packet page frags */
251 struct ksock_conn *tx_conn; /* owning conn */
252 lnet_msg_t *tx_lnetmsg; /* lnet message for lnet_finalize() */
253 cfs_time_t tx_deadline; /* when (in jiffies) tx times out */
254 ksock_msg_t tx_msg; /* socklnd message buffer */
255 int tx_desc_size; /* size of this descriptor */
258 struct iovec iov; /* virt hdr */
259 lnet_kiov_t kiov[0]; /* paged payload */
262 struct iovec iov[1]; /* virt hdr + payload */
267 #define KSOCK_NOOP_TX_SIZE ((int)offsetof(ksock_tx_t, tx_frags.paged.kiov[0]))
269 /* network zero copy callback descriptor embedded in ksock_tx_t */
271 /* space for the rx frag descriptors; we either read a single contiguous
272 * header, or up to LNET_MAX_IOV frags of payload of either type. */
274 struct iovec iov[LNET_MAX_IOV];
275 lnet_kiov_t kiov[LNET_MAX_IOV];
276 } ksock_rxiovspace_t;
278 #define SOCKNAL_RX_KSM_HEADER 1 /* reading ksock message header */
279 #define SOCKNAL_RX_LNET_HEADER 2 /* reading lnet message header */
280 #define SOCKNAL_RX_PARSE 3 /* Calling lnet_parse() */
281 #define SOCKNAL_RX_PARSE_WAIT 4 /* waiting to be told to read the body */
282 #define SOCKNAL_RX_LNET_PAYLOAD 5 /* reading lnet payload (to deliver here) */
283 #define SOCKNAL_RX_SLOP 6 /* skipping body */
285 typedef struct ksock_conn
287 struct ksock_peer *ksnc_peer; /* owning peer */
288 struct ksock_route *ksnc_route; /* owning route */
289 cfs_list_t ksnc_list; /* stash on peer's conn list */
290 cfs_socket_t *ksnc_sock; /* actual socket */
291 void *ksnc_saved_data_ready; /* socket's original data_ready() callback */
292 void *ksnc_saved_write_space; /* socket's original write_space() callback */
293 cfs_atomic_t ksnc_conn_refcount; /* conn refcount */
294 cfs_atomic_t ksnc_sock_refcount; /* sock refcount */
295 ksock_sched_t *ksnc_scheduler; /* who schedules this connection */
296 __u32 ksnc_myipaddr; /* my IP */
297 __u32 ksnc_ipaddr; /* peer's IP */
298 int ksnc_port; /* peer's port */
299 int ksnc_type:3; /* type of connection, should be signed value */
300 int ksnc_closing:1; /* being shut down */
301 int ksnc_flip:1; /* flip or not, only for V2.x */
302 int ksnc_zc_capable:1; /* enable to ZC */
303 struct ksock_proto *ksnc_proto; /* protocol for the connection */
306 cfs_list_t ksnc_rx_list; /* where I enq waiting input or a forwarding descriptor */
307 cfs_time_t ksnc_rx_deadline; /* when (in jiffies) receive times out */
308 __u8 ksnc_rx_started; /* started receiving a message */
309 __u8 ksnc_rx_ready; /* data ready to read */
310 __u8 ksnc_rx_scheduled;/* being progressed */
311 __u8 ksnc_rx_state; /* what is being read */
312 int ksnc_rx_nob_left; /* # bytes to next hdr/body */
313 int ksnc_rx_nob_wanted; /* bytes actually wanted */
314 int ksnc_rx_niov; /* # iovec frags */
315 struct iovec *ksnc_rx_iov; /* the iovec frags */
316 int ksnc_rx_nkiov; /* # page frags */
317 lnet_kiov_t *ksnc_rx_kiov; /* the page frags */
318 ksock_rxiovspace_t ksnc_rx_iov_space;/* space for frag descriptors */
319 __u32 ksnc_rx_csum; /* partial checksum for incoming data */
320 void *ksnc_cookie; /* rx lnet_finalize passthru arg */
321 ksock_msg_t ksnc_msg; /* incoming message buffer:
322 * V2.x message takes the
324 * V1.x message is a bare
325 * lnet_hdr_t, it's stored in
326 * ksnc_msg.ksm_u.lnetmsg */
329 cfs_list_t ksnc_tx_list; /* where I enq waiting for output space */
330 cfs_list_t ksnc_tx_queue; /* packets waiting to be sent */
331 ksock_tx_t *ksnc_tx_carrier; /* next TX that can carry a LNet message or ZC-ACK */
332 cfs_time_t ksnc_tx_deadline; /* when (in jiffies) tx times out */
333 int ksnc_tx_bufnob; /* send buffer marker */
334 cfs_atomic_t ksnc_tx_nob; /* # bytes queued */
335 int ksnc_tx_ready; /* write space */
336 int ksnc_tx_scheduled; /* being progressed */
337 cfs_time_t ksnc_tx_last_post; /* time stamp of the last posted TX */
340 typedef struct ksock_route
342 cfs_list_t ksnr_list; /* chain on peer route list */
343 cfs_list_t ksnr_connd_list; /* chain on ksnr_connd_routes */
344 struct ksock_peer *ksnr_peer; /* owning peer */
345 cfs_atomic_t ksnr_refcount; /* # users */
346 cfs_time_t ksnr_timeout; /* when (in jiffies) reconnection can happen next */
347 cfs_duration_t ksnr_retry_interval; /* how long between retries */
348 __u32 ksnr_myipaddr; /* my IP */
349 __u32 ksnr_ipaddr; /* IP address to connect to */
350 int ksnr_port; /* port to connect to */
351 unsigned int ksnr_scheduled:1; /* scheduled for attention */
352 unsigned int ksnr_connecting:1;/* connection establishment in progress */
353 unsigned int ksnr_connected:4; /* connections established by type */
354 unsigned int ksnr_deleted:1; /* been removed from peer? */
355 unsigned int ksnr_share_count; /* created explicitly? */
356 int ksnr_conn_count; /* # conns established by this route */
359 #define SOCKNAL_KEEPALIVE_PING 1 /* cookie for keepalive ping */
361 typedef struct ksock_peer
363 cfs_list_t ksnp_list; /* stash on global peer list */
364 cfs_time_t ksnp_last_alive; /* when (in jiffies) I was last alive */
365 lnet_process_id_t ksnp_id; /* who's on the other end(s) */
366 cfs_atomic_t ksnp_refcount; /* # users */
367 int ksnp_sharecount; /* lconf usage counter */
368 int ksnp_closing; /* being closed */
369 int ksnp_accepting;/* # passive connections pending */
370 int ksnp_error; /* errno on closing last conn */
371 __u64 ksnp_zc_next_cookie;/* ZC completion cookie */
372 __u64 ksnp_incarnation; /* latest known peer incarnation */
373 struct ksock_proto *ksnp_proto; /* latest known peer protocol */
374 cfs_list_t ksnp_conns; /* all active connections */
375 cfs_list_t ksnp_routes; /* routes */
376 cfs_list_t ksnp_tx_queue; /* waiting packets */
377 spinlock_t ksnp_lock; /* serialize, g_lock unsafe */
378 cfs_list_t ksnp_zc_req_list; /* zero copy requests wait for ACK */
379 cfs_time_t ksnp_send_keepalive; /* time to send keepalive */
380 lnet_ni_t *ksnp_ni; /* which network */
381 int ksnp_n_passive_ips; /* # of... */
382 __u32 ksnp_passive_ips[LNET_MAX_INTERFACES]; /* preferred local interfaces */
385 typedef struct ksock_connreq
387 cfs_list_t ksncr_list; /* stash on ksnd_connd_connreqs */
388 lnet_ni_t *ksncr_ni; /* chosen NI */
389 cfs_socket_t *ksncr_sock; /* accepted socket */
392 extern ksock_nal_data_t ksocknal_data;
393 extern ksock_tunables_t ksocknal_tunables;
395 #define SOCKNAL_MATCH_NO 0 /* TX can't match type of connection */
396 #define SOCKNAL_MATCH_YES 1 /* TX matches type of connection */
397 #define SOCKNAL_MATCH_MAY 2 /* TX can be sent on the connection, but not preferred */
399 typedef struct ksock_proto
401 int pro_version; /* version number of protocol */
402 int (*pro_send_hello)(ksock_conn_t *, ksock_hello_msg_t *); /* handshake function */
403 int (*pro_recv_hello)(ksock_conn_t *, ksock_hello_msg_t *, int);/* handshake function */
404 void (*pro_pack)(ksock_tx_t *); /* message pack */
405 void (*pro_unpack)(ksock_msg_t *); /* message unpack */
406 ksock_tx_t *(*pro_queue_tx_msg)(ksock_conn_t *, ksock_tx_t *); /* queue tx on the connection */
407 int (*pro_queue_tx_zcack)(ksock_conn_t *, ksock_tx_t *, __u64); /* queue ZC ack on the connection */
408 int (*pro_handle_zcreq)(ksock_conn_t *, __u64, int); /* handle ZC request */
409 int (*pro_handle_zcack)(ksock_conn_t *, __u64, __u64); /* handle ZC ACK */
410 int (*pro_match_tx)(ksock_conn_t *, ksock_tx_t *, int); /* msg type matches the connection type:
412 * return MATCH_NO : no
413 * return MATCH_YES : matching type
414 * return MATCH_MAY : can be backup */
417 extern ksock_proto_t ksocknal_protocol_v1x;
418 extern ksock_proto_t ksocknal_protocol_v2x;
419 extern ksock_proto_t ksocknal_protocol_v3x;
421 #define KSOCK_PROTO_V1_MAJOR LNET_PROTO_TCP_VERSION_MAJOR
422 #define KSOCK_PROTO_V1_MINOR LNET_PROTO_TCP_VERSION_MINOR
423 #define KSOCK_PROTO_V1 KSOCK_PROTO_V1_MAJOR
425 #ifndef CPU_MASK_NONE
426 #define CPU_MASK_NONE 0UL
430 ksocknal_route_mask(void)
432 if (!*ksocknal_tunables.ksnd_typed_conns)
433 return (1 << SOCKLND_CONN_ANY);
435 return ((1 << SOCKLND_CONN_CONTROL) |
436 (1 << SOCKLND_CONN_BULK_IN) |
437 (1 << SOCKLND_CONN_BULK_OUT));
440 static inline cfs_list_t *
441 ksocknal_nid2peerlist (lnet_nid_t nid)
443 unsigned int hash = ((unsigned int)nid) % ksocknal_data.ksnd_peer_hash_size;
445 return (&ksocknal_data.ksnd_peers [hash]);
449 ksocknal_conn_addref (ksock_conn_t *conn)
451 LASSERT (cfs_atomic_read(&conn->ksnc_conn_refcount) > 0);
452 cfs_atomic_inc(&conn->ksnc_conn_refcount);
455 extern void ksocknal_queue_zombie_conn (ksock_conn_t *conn);
456 extern void ksocknal_finalize_zcreq(ksock_conn_t *conn);
459 ksocknal_conn_decref (ksock_conn_t *conn)
461 LASSERT (cfs_atomic_read(&conn->ksnc_conn_refcount) > 0);
462 if (cfs_atomic_dec_and_test(&conn->ksnc_conn_refcount))
463 ksocknal_queue_zombie_conn(conn);
467 ksocknal_connsock_addref (ksock_conn_t *conn)
471 read_lock(&ksocknal_data.ksnd_global_lock);
472 if (!conn->ksnc_closing) {
473 LASSERT(cfs_atomic_read(&conn->ksnc_sock_refcount) > 0);
474 cfs_atomic_inc(&conn->ksnc_sock_refcount);
477 read_unlock(&ksocknal_data.ksnd_global_lock);
483 ksocknal_connsock_decref (ksock_conn_t *conn)
485 LASSERT (cfs_atomic_read(&conn->ksnc_sock_refcount) > 0);
486 if (cfs_atomic_dec_and_test(&conn->ksnc_sock_refcount)) {
487 LASSERT (conn->ksnc_closing);
488 libcfs_sock_release(conn->ksnc_sock);
489 conn->ksnc_sock = NULL;
490 ksocknal_finalize_zcreq(conn);
495 ksocknal_tx_addref (ksock_tx_t *tx)
497 LASSERT (cfs_atomic_read(&tx->tx_refcount) > 0);
498 cfs_atomic_inc(&tx->tx_refcount);
501 extern void ksocknal_tx_prep (ksock_conn_t *, ksock_tx_t *tx);
502 extern void ksocknal_tx_done (lnet_ni_t *ni, ksock_tx_t *tx);
505 ksocknal_tx_decref (ksock_tx_t *tx)
507 LASSERT (cfs_atomic_read(&tx->tx_refcount) > 0);
508 if (cfs_atomic_dec_and_test(&tx->tx_refcount))
509 ksocknal_tx_done(NULL, tx);
513 ksocknal_route_addref (ksock_route_t *route)
515 LASSERT (cfs_atomic_read(&route->ksnr_refcount) > 0);
516 cfs_atomic_inc(&route->ksnr_refcount);
519 extern void ksocknal_destroy_route (ksock_route_t *route);
522 ksocknal_route_decref (ksock_route_t *route)
524 LASSERT (cfs_atomic_read (&route->ksnr_refcount) > 0);
525 if (cfs_atomic_dec_and_test(&route->ksnr_refcount))
526 ksocknal_destroy_route (route);
530 ksocknal_peer_addref (ksock_peer_t *peer)
532 LASSERT (cfs_atomic_read (&peer->ksnp_refcount) > 0);
533 cfs_atomic_inc(&peer->ksnp_refcount);
536 extern void ksocknal_destroy_peer (ksock_peer_t *peer);
539 ksocknal_peer_decref (ksock_peer_t *peer)
541 LASSERT (cfs_atomic_read (&peer->ksnp_refcount) > 0);
542 if (cfs_atomic_dec_and_test(&peer->ksnp_refcount))
543 ksocknal_destroy_peer (peer);
546 int ksocknal_startup (lnet_ni_t *ni);
547 void ksocknal_shutdown (lnet_ni_t *ni);
548 int ksocknal_ctl(lnet_ni_t *ni, unsigned int cmd, void *arg);
549 int ksocknal_send (lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg);
550 int ksocknal_recv(lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg,
551 int delayed, unsigned int niov,
552 struct iovec *iov, lnet_kiov_t *kiov,
553 unsigned int offset, unsigned int mlen, unsigned int rlen);
554 int ksocknal_accept(lnet_ni_t *ni, cfs_socket_t *sock);
556 extern int ksocknal_add_peer(lnet_ni_t *ni, lnet_process_id_t id, __u32 ip, int port);
557 extern ksock_peer_t *ksocknal_find_peer_locked (lnet_ni_t *ni, lnet_process_id_t id);
558 extern ksock_peer_t *ksocknal_find_peer (lnet_ni_t *ni, lnet_process_id_t id);
559 extern void ksocknal_peer_failed (ksock_peer_t *peer);
560 extern int ksocknal_create_conn (lnet_ni_t *ni, ksock_route_t *route,
561 cfs_socket_t *sock, int type);
562 extern void ksocknal_close_conn_locked (ksock_conn_t *conn, int why);
563 extern void ksocknal_terminate_conn (ksock_conn_t *conn);
564 extern void ksocknal_destroy_conn (ksock_conn_t *conn);
565 extern int ksocknal_close_peer_conns_locked (ksock_peer_t *peer,
566 __u32 ipaddr, int why);
567 extern int ksocknal_close_conn_and_siblings (ksock_conn_t *conn, int why);
568 extern int ksocknal_close_matching_conns (lnet_process_id_t id, __u32 ipaddr);
569 extern ksock_conn_t *ksocknal_find_conn_locked(ksock_peer_t *peer,
570 ksock_tx_t *tx, int nonblk);
572 extern int ksocknal_launch_packet(lnet_ni_t *ni, ksock_tx_t *tx,
573 lnet_process_id_t id);
574 extern ksock_tx_t *ksocknal_alloc_tx(int type, int size);
575 extern void ksocknal_free_tx (ksock_tx_t *tx);
576 extern ksock_tx_t *ksocknal_alloc_tx_noop(__u64 cookie, int nonblk);
577 extern void ksocknal_next_tx_carrier(ksock_conn_t *conn);
578 extern void ksocknal_queue_tx_locked (ksock_tx_t *tx, ksock_conn_t *conn);
579 extern void ksocknal_txlist_done (lnet_ni_t *ni, cfs_list_t *txlist,
581 extern void ksocknal_notify (lnet_ni_t *ni, lnet_nid_t gw_nid, int alive);
582 extern void ksocknal_query (struct lnet_ni *ni, lnet_nid_t nid, cfs_time_t *when);
583 extern int ksocknal_thread_start (int (*fn)(void *arg), void *arg);
584 extern void ksocknal_thread_fini (void);
585 extern void ksocknal_launch_all_connections_locked (ksock_peer_t *peer);
586 extern ksock_route_t *ksocknal_find_connectable_route_locked (ksock_peer_t *peer);
587 extern ksock_route_t *ksocknal_find_connecting_route_locked (ksock_peer_t *peer);
588 extern int ksocknal_new_packet (ksock_conn_t *conn, int skip);
589 extern int ksocknal_scheduler (void *arg);
590 extern int ksocknal_connd (void *arg);
591 extern int ksocknal_reaper (void *arg);
592 extern int ksocknal_send_hello (lnet_ni_t *ni, ksock_conn_t *conn,
593 lnet_nid_t peer_nid, ksock_hello_msg_t *hello);
594 extern int ksocknal_recv_hello (lnet_ni_t *ni, ksock_conn_t *conn,
595 ksock_hello_msg_t *hello, lnet_process_id_t *id,
597 extern void ksocknal_read_callback(ksock_conn_t *conn);
598 extern void ksocknal_write_callback(ksock_conn_t *conn);
600 extern int ksocknal_lib_zc_capable(ksock_conn_t *conn);
601 extern void ksocknal_lib_save_callback(cfs_socket_t *sock, ksock_conn_t *conn);
602 extern void ksocknal_lib_set_callback(cfs_socket_t *sock, ksock_conn_t *conn);
603 extern void ksocknal_lib_reset_callback(cfs_socket_t *sock, ksock_conn_t *conn);
604 extern void ksocknal_lib_push_conn (ksock_conn_t *conn);
605 extern int ksocknal_lib_get_conn_addrs (ksock_conn_t *conn);
606 extern int ksocknal_lib_setup_sock (cfs_socket_t *so);
607 extern int ksocknal_lib_send_iov (ksock_conn_t *conn, ksock_tx_t *tx);
608 extern int ksocknal_lib_send_kiov (ksock_conn_t *conn, ksock_tx_t *tx);
609 extern void ksocknal_lib_eager_ack (ksock_conn_t *conn);
610 extern int ksocknal_lib_recv_iov (ksock_conn_t *conn);
611 extern int ksocknal_lib_recv_kiov (ksock_conn_t *conn);
612 extern int ksocknal_lib_get_conn_tunables (ksock_conn_t *conn, int *txmem,
613 int *rxmem, int *nagle);
615 extern int ksocknal_tunables_init(void);
616 extern void ksocknal_tunables_fini(void);
617 extern int ksocknal_lib_tunables_init(void);
618 extern void ksocknal_lib_tunables_fini(void);
620 extern void ksocknal_lib_csum_tx(ksock_tx_t *tx);
622 extern int ksocknal_lib_memory_pressure(ksock_conn_t *conn);
623 extern int ksocknal_lib_bind_thread_to_cpu(int id);