2 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
4 * Copyright (c) 2011, 2012, Intel Corporation.
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 #include "socklnd_lib-linux.h"
33 #include "../../../include/linux/libcfs/libcfs.h"
34 #include "../../../include/linux/lnet/lnet.h"
35 #include "../../../include/linux/lnet/lib-lnet.h"
36 #include "../../../include/linux/lnet/socklnd.h"
37 #include "../../../include/linux/lnet/lnet-sysctl.h"
39 #define SOCKNAL_PEER_HASH_SIZE 101 /* # peer lists */
40 #define SOCKNAL_RESCHED 100 /* # scheduler loops before reschedule */
41 #define SOCKNAL_INSANITY_RECONN 5000 /* connd is trying on reconn infinitely */
42 #define SOCKNAL_ENOMEM_RETRY CFS_TICK /* jiffies between retries */
44 #define SOCKNAL_SINGLE_FRAG_TX 0 /* disable multi-fragment sends */
45 #define SOCKNAL_SINGLE_FRAG_RX 0 /* disable multi-fragment receives */
47 #define SOCKNAL_VERSION_DEBUG 0 /* enable protocol version debugging */
49 /* risk kmap deadlock on multi-frag I/O (backs off to single-frag if disabled).
50 * no risk if we're not running on a CONFIG_HIGHMEM platform. */
52 # define SOCKNAL_RISK_KMAP_DEADLOCK 0
54 # define SOCKNAL_RISK_KMAP_DEADLOCK 1
57 struct ksock_sched_info;
59 typedef struct /* per scheduler state */
61 spinlock_t kss_lock; /* serialise */
62 struct list_head kss_rx_conns; /* conn waiting to be read */
63 /* conn waiting to be written */
64 struct list_head kss_tx_conns;
65 /* zombie noop tx list */
66 struct list_head kss_zombie_noop_txs;
67 wait_queue_head_t kss_waitq; /* where scheduler sleeps */
68 /* # connections assigned to this scheduler */
70 struct ksock_sched_info *kss_info; /* owner of it */
71 struct page *kss_rx_scratch_pgs[LNET_MAX_IOV];
72 struct kvec kss_scratch_iov[LNET_MAX_IOV];
75 struct ksock_sched_info {
76 int ksi_nthreads_max; /* max allowed threads */
77 int ksi_nthreads; /* number of threads */
78 int ksi_cpt; /* CPT id */
79 ksock_sched_t *ksi_scheds; /* array of schedulers */
82 #define KSOCK_CPT_SHIFT 16
83 #define KSOCK_THREAD_ID(cpt, sid) (((cpt) << KSOCK_CPT_SHIFT) | (sid))
84 #define KSOCK_THREAD_CPT(id) ((id) >> KSOCK_CPT_SHIFT)
85 #define KSOCK_THREAD_SID(id) ((id) & ((1UL << KSOCK_CPT_SHIFT) - 1))
87 typedef struct /* in-use interface */
89 __u32 ksni_ipaddr; /* interface's IP address */
90 __u32 ksni_netmask; /* interface's network mask */
91 int ksni_nroutes; /* # routes using (active) */
92 int ksni_npeers; /* # peers using (passive) */
93 char ksni_name[IFNAMSIZ]; /* interface name */
97 /* "stuck" socket timeout (seconds) */
99 /* # scheduler threads in each pool while starting */
101 int *ksnd_nconnds; /* # connection daemons */
102 int *ksnd_nconnds_max; /* max # connection daemons */
103 int *ksnd_min_reconnectms; /* first connection retry after (ms)... */
104 int *ksnd_max_reconnectms; /* ...exponentially increasing to this */
105 int *ksnd_eager_ack; /* make TCP ack eagerly? */
106 int *ksnd_typed_conns; /* drive sockets by type? */
107 int *ksnd_min_bulk; /* smallest "large" message */
108 int *ksnd_tx_buffer_size; /* socket tx buffer size */
109 int *ksnd_rx_buffer_size; /* socket rx buffer size */
110 int *ksnd_nagle; /* enable NAGLE? */
111 int *ksnd_round_robin; /* round robin for multiple interfaces */
112 int *ksnd_keepalive; /* # secs for sending keepalive NOOP */
113 int *ksnd_keepalive_idle; /* # idle secs before 1st probe */
114 int *ksnd_keepalive_count; /* # probes */
115 int *ksnd_keepalive_intvl; /* time between probes */
116 int *ksnd_credits; /* # concurrent sends */
117 int *ksnd_peertxcredits; /* # concurrent sends to 1 peer */
118 int *ksnd_peerrtrcredits; /* # per-peer router buffer credits */
119 int *ksnd_peertimeout; /* seconds to consider peer dead */
120 int *ksnd_enable_csum; /* enable check sum */
121 int *ksnd_inject_csum_error; /* set non-zero to inject checksum error */
122 int *ksnd_nonblk_zcack; /* always send zc-ack on non-blocking connection */
123 unsigned int *ksnd_zc_min_payload; /* minimum zero copy payload size */
124 int *ksnd_zc_recv; /* enable ZC receive (for Chelsio TOE) */
125 int *ksnd_zc_recv_min_nfrags; /* minimum # of fragments to enable ZC receive */
129 __u64 ksnn_incarnation; /* my epoch */
130 spinlock_t ksnn_lock; /* serialise */
131 struct list_head ksnn_list; /* chain on global list */
132 int ksnn_npeers; /* # peers */
133 int ksnn_shutdown; /* shutting down? */
134 int ksnn_ninterfaces; /* IP interfaces */
135 ksock_interface_t ksnn_interfaces[LNET_MAX_INTERFACES];
139 #define SOCKNAL_CONND_TIMEOUT 120
140 /** reserved thread for accepting & creating new connd */
141 #define SOCKNAL_CONND_RESV 1
144 int ksnd_init; /* initialisation state */
145 int ksnd_nnets; /* # networks set up */
146 struct list_head ksnd_nets; /* list of nets */
147 /* stabilize peer/conn ops */
148 rwlock_t ksnd_global_lock;
149 /* hash table of all my known peers */
150 struct list_head *ksnd_peers;
151 int ksnd_peer_hash_size; /* size of ksnd_peers */
153 int ksnd_nthreads; /* # live threads */
154 int ksnd_shuttingdown; /* tell threads to exit */
155 /* schedulers information */
156 struct ksock_sched_info **ksnd_sched_info;
158 atomic_t ksnd_nactive_txs; /* #active txs */
160 struct list_head ksnd_deathrow_conns; /* conns to close: reaper_lock*/
161 struct list_head ksnd_zombie_conns; /* conns to free: reaper_lock */
162 struct list_head ksnd_enomem_conns; /* conns to retry: reaper_lock*/
163 wait_queue_head_t ksnd_reaper_waitq; /* reaper sleeps here */
164 unsigned long ksnd_reaper_waketime;/* when reaper will wake */
165 spinlock_t ksnd_reaper_lock; /* serialise */
167 int ksnd_enomem_tx; /* test ENOMEM sender */
168 int ksnd_stall_tx; /* test sluggish sender */
169 int ksnd_stall_rx; /* test sluggish receiver */
171 struct list_head ksnd_connd_connreqs; /* incoming connection requests */
172 struct list_head ksnd_connd_routes; /* routes waiting to be connected */
173 wait_queue_head_t ksnd_connd_waitq; /* connds sleep here */
174 int ksnd_connd_connecting;/* # connds connecting */
175 /** time stamp of the last failed connecting attempt */
176 long ksnd_connd_failed_stamp;
177 /** # starting connd */
178 unsigned ksnd_connd_starting;
179 /** time stamp of the last starting connd */
180 long ksnd_connd_starting_stamp;
181 /** # running connd */
182 unsigned ksnd_connd_running;
183 spinlock_t ksnd_connd_lock; /* serialise */
185 struct list_head ksnd_idle_noop_txs; /* list head for freed noop tx */
186 spinlock_t ksnd_tx_lock; /* serialise, g_lock unsafe */
190 #define SOCKNAL_INIT_NOTHING 0
191 #define SOCKNAL_INIT_DATA 1
192 #define SOCKNAL_INIT_ALL 2
194 /* A packet just assembled for transmission is represented by 1 or more
195 * struct iovec fragments (the first frag contains the portals header),
196 * followed by 0 or more lnet_kiov_t fragments.
198 * On the receive side, initially 1 struct iovec fragment is posted for
199 * receive (the header). Once the header has been received, the payload is
200 * received into either struct iovec or lnet_kiov_t fragments, depending on
201 * what the header matched or whether the message needs forwarding. */
203 struct ksock_conn; /* forward ref */
204 struct ksock_peer; /* forward ref */
205 struct ksock_route; /* forward ref */
206 struct ksock_proto; /* forward ref */
208 typedef struct /* transmit packet */
210 struct list_head tx_list; /* queue on conn for transmission etc */
211 struct list_head tx_zc_list; /* queue on peer for ZC request */
212 atomic_t tx_refcount; /* tx reference count */
213 int tx_nob; /* # packet bytes */
214 int tx_resid; /* residual bytes */
215 int tx_niov; /* # packet iovec frags */
216 struct kvec *tx_iov; /* packet iovec frags */
217 int tx_nkiov; /* # packet page frags */
218 unsigned short tx_zc_aborted; /* aborted ZC request */
219 unsigned short tx_zc_capable:1; /* payload is large enough for ZC */
220 unsigned short tx_zc_checked:1; /* Have I checked if I should ZC? */
221 unsigned short tx_nonblk:1; /* it's a non-blocking ACK */
222 lnet_kiov_t *tx_kiov; /* packet page frags */
223 struct ksock_conn *tx_conn; /* owning conn */
224 lnet_msg_t *tx_lnetmsg; /* lnet message for lnet_finalize() */
225 unsigned long tx_deadline; /* when (in jiffies) tx times out */
226 ksock_msg_t tx_msg; /* socklnd message buffer */
227 int tx_desc_size; /* size of this descriptor */
230 struct kvec iov; /* virt hdr */
231 lnet_kiov_t kiov[0]; /* paged payload */
234 struct kvec iov[1]; /* virt hdr + payload */
239 #define KSOCK_NOOP_TX_SIZE ((int)offsetof(ksock_tx_t, tx_frags.paged.kiov[0]))
241 /* network zero copy callback descriptor embedded in ksock_tx_t */
243 /* space for the rx frag descriptors; we either read a single contiguous
244 * header, or up to LNET_MAX_IOV frags of payload of either type. */
246 struct kvec iov[LNET_MAX_IOV];
247 lnet_kiov_t kiov[LNET_MAX_IOV];
248 } ksock_rxiovspace_t;
250 #define SOCKNAL_RX_KSM_HEADER 1 /* reading ksock message header */
251 #define SOCKNAL_RX_LNET_HEADER 2 /* reading lnet message header */
252 #define SOCKNAL_RX_PARSE 3 /* Calling lnet_parse() */
253 #define SOCKNAL_RX_PARSE_WAIT 4 /* waiting to be told to read the body */
254 #define SOCKNAL_RX_LNET_PAYLOAD 5 /* reading lnet payload (to deliver here) */
255 #define SOCKNAL_RX_SLOP 6 /* skipping body */
257 typedef struct ksock_conn {
258 struct ksock_peer *ksnc_peer; /* owning peer */
259 struct ksock_route *ksnc_route; /* owning route */
260 struct list_head ksnc_list; /* stash on peer's conn list */
261 struct socket *ksnc_sock; /* actual socket */
262 void *ksnc_saved_data_ready; /* socket's original data_ready() callback */
263 void *ksnc_saved_write_space; /* socket's original write_space() callback */
264 atomic_t ksnc_conn_refcount; /* conn refcount */
265 atomic_t ksnc_sock_refcount; /* sock refcount */
266 ksock_sched_t *ksnc_scheduler; /* who schedules this connection */
267 __u32 ksnc_myipaddr; /* my IP */
268 __u32 ksnc_ipaddr; /* peer's IP */
269 int ksnc_port; /* peer's port */
270 signed int ksnc_type:3; /* type of connection,
271 * should be signed value */
272 unsigned int ksnc_closing:1; /* being shut down */
273 unsigned int ksnc_flip:1; /* flip or not, only for V2.x */
274 unsigned int ksnc_zc_capable:1; /* enable to ZC */
275 struct ksock_proto *ksnc_proto; /* protocol for the connection */
278 struct list_head ksnc_rx_list; /* where I enq waiting input or a forwarding descriptor */
279 unsigned long ksnc_rx_deadline; /* when (in jiffies) receive times out */
280 __u8 ksnc_rx_started; /* started receiving a message */
281 __u8 ksnc_rx_ready; /* data ready to read */
282 __u8 ksnc_rx_scheduled;/* being progressed */
283 __u8 ksnc_rx_state; /* what is being read */
284 int ksnc_rx_nob_left; /* # bytes to next hdr/body */
285 int ksnc_rx_nob_wanted; /* bytes actually wanted */
286 int ksnc_rx_niov; /* # iovec frags */
287 struct kvec *ksnc_rx_iov; /* the iovec frags */
288 int ksnc_rx_nkiov; /* # page frags */
289 lnet_kiov_t *ksnc_rx_kiov; /* the page frags */
290 ksock_rxiovspace_t ksnc_rx_iov_space;/* space for frag descriptors */
291 __u32 ksnc_rx_csum; /* partial checksum for incoming data */
292 void *ksnc_cookie; /* rx lnet_finalize passthru arg */
293 ksock_msg_t ksnc_msg; /* incoming message buffer:
294 * V2.x message takes the
296 * V1.x message is a bare
297 * lnet_hdr_t, it's stored in
298 * ksnc_msg.ksm_u.lnetmsg */
301 struct list_head ksnc_tx_list; /* where I enq waiting for output space */
302 struct list_head ksnc_tx_queue; /* packets waiting to be sent */
303 ksock_tx_t *ksnc_tx_carrier; /* next TX that can carry a LNet message or ZC-ACK */
304 unsigned long ksnc_tx_deadline; /* when (in jiffies) tx times out */
305 int ksnc_tx_bufnob; /* send buffer marker */
306 atomic_t ksnc_tx_nob; /* # bytes queued */
307 int ksnc_tx_ready; /* write space */
308 int ksnc_tx_scheduled; /* being progressed */
309 unsigned long ksnc_tx_last_post; /* time stamp of the last posted TX */
312 typedef struct ksock_route {
313 struct list_head ksnr_list; /* chain on peer route list */
314 struct list_head ksnr_connd_list; /* chain on ksnr_connd_routes */
315 struct ksock_peer *ksnr_peer; /* owning peer */
316 atomic_t ksnr_refcount; /* # users */
317 unsigned long ksnr_timeout; /* when (in jiffies) reconnection can happen next */
318 long ksnr_retry_interval; /* how long between retries */
319 __u32 ksnr_myipaddr; /* my IP */
320 __u32 ksnr_ipaddr; /* IP address to connect to */
321 int ksnr_port; /* port to connect to */
322 unsigned int ksnr_scheduled:1; /* scheduled for attention */
323 unsigned int ksnr_connecting:1;/* connection establishment in progress */
324 unsigned int ksnr_connected:4; /* connections established by type */
325 unsigned int ksnr_deleted:1; /* been removed from peer? */
326 unsigned int ksnr_share_count; /* created explicitly? */
327 int ksnr_conn_count; /* # conns established by this route */
330 #define SOCKNAL_KEEPALIVE_PING 1 /* cookie for keepalive ping */
332 typedef struct ksock_peer {
333 struct list_head ksnp_list; /* stash on global peer list */
334 unsigned long ksnp_last_alive; /* when (in jiffies) I was last alive */
335 lnet_process_id_t ksnp_id; /* who's on the other end(s) */
336 atomic_t ksnp_refcount; /* # users */
337 int ksnp_sharecount; /* lconf usage counter */
338 int ksnp_closing; /* being closed */
339 int ksnp_accepting;/* # passive connections pending */
340 int ksnp_error; /* errno on closing last conn */
341 __u64 ksnp_zc_next_cookie;/* ZC completion cookie */
342 __u64 ksnp_incarnation; /* latest known peer incarnation */
343 struct ksock_proto *ksnp_proto; /* latest known peer protocol */
344 struct list_head ksnp_conns; /* all active connections */
345 struct list_head ksnp_routes; /* routes */
346 struct list_head ksnp_tx_queue; /* waiting packets */
347 spinlock_t ksnp_lock; /* serialize, g_lock unsafe */
348 struct list_head ksnp_zc_req_list; /* zero copy requests wait for ACK */
349 unsigned long ksnp_send_keepalive; /* time to send keepalive */
350 lnet_ni_t *ksnp_ni; /* which network */
351 int ksnp_n_passive_ips; /* # of... */
352 __u32 ksnp_passive_ips[LNET_MAX_INTERFACES]; /* preferred local interfaces */
355 typedef struct ksock_connreq {
356 struct list_head ksncr_list; /* stash on ksnd_connd_connreqs */
357 lnet_ni_t *ksncr_ni; /* chosen NI */
358 struct socket *ksncr_sock; /* accepted socket */
361 extern ksock_nal_data_t ksocknal_data;
362 extern ksock_tunables_t ksocknal_tunables;
364 #define SOCKNAL_MATCH_NO 0 /* TX can't match type of connection */
365 #define SOCKNAL_MATCH_YES 1 /* TX matches type of connection */
366 #define SOCKNAL_MATCH_MAY 2 /* TX can be sent on the connection, but not preferred */
368 typedef struct ksock_proto {
369 int pro_version; /* version number of protocol */
370 int (*pro_send_hello)(ksock_conn_t *, ksock_hello_msg_t *); /* handshake function */
371 int (*pro_recv_hello)(ksock_conn_t *, ksock_hello_msg_t *, int);/* handshake function */
372 void (*pro_pack)(ksock_tx_t *); /* message pack */
373 void (*pro_unpack)(ksock_msg_t *); /* message unpack */
374 ksock_tx_t *(*pro_queue_tx_msg)(ksock_conn_t *, ksock_tx_t *); /* queue tx on the connection */
375 int (*pro_queue_tx_zcack)(ksock_conn_t *, ksock_tx_t *, __u64); /* queue ZC ack on the connection */
376 int (*pro_handle_zcreq)(ksock_conn_t *, __u64, int); /* handle ZC request */
377 int (*pro_handle_zcack)(ksock_conn_t *, __u64, __u64); /* handle ZC ACK */
378 int (*pro_match_tx)(ksock_conn_t *, ksock_tx_t *, int); /* msg type matches the connection type:
380 * return MATCH_NO : no
381 * return MATCH_YES : matching type
382 * return MATCH_MAY : can be backup */
385 extern ksock_proto_t ksocknal_protocol_v1x;
386 extern ksock_proto_t ksocknal_protocol_v2x;
387 extern ksock_proto_t ksocknal_protocol_v3x;
389 #define KSOCK_PROTO_V1_MAJOR LNET_PROTO_TCP_VERSION_MAJOR
390 #define KSOCK_PROTO_V1_MINOR LNET_PROTO_TCP_VERSION_MINOR
391 #define KSOCK_PROTO_V1 KSOCK_PROTO_V1_MAJOR
393 #ifndef CPU_MASK_NONE
394 #define CPU_MASK_NONE 0UL
398 ksocknal_route_mask(void)
400 if (!*ksocknal_tunables.ksnd_typed_conns)
401 return (1 << SOCKLND_CONN_ANY);
403 return ((1 << SOCKLND_CONN_CONTROL) |
404 (1 << SOCKLND_CONN_BULK_IN) |
405 (1 << SOCKLND_CONN_BULK_OUT));
408 static inline struct list_head *
409 ksocknal_nid2peerlist(lnet_nid_t nid)
411 unsigned int hash = ((unsigned int)nid) % ksocknal_data.ksnd_peer_hash_size;
413 return &ksocknal_data.ksnd_peers[hash];
417 ksocknal_conn_addref(ksock_conn_t *conn)
419 LASSERT(atomic_read(&conn->ksnc_conn_refcount) > 0);
420 atomic_inc(&conn->ksnc_conn_refcount);
423 extern void ksocknal_queue_zombie_conn(ksock_conn_t *conn);
424 extern void ksocknal_finalize_zcreq(ksock_conn_t *conn);
427 ksocknal_conn_decref(ksock_conn_t *conn)
429 LASSERT(atomic_read(&conn->ksnc_conn_refcount) > 0);
430 if (atomic_dec_and_test(&conn->ksnc_conn_refcount))
431 ksocknal_queue_zombie_conn(conn);
435 ksocknal_connsock_addref(ksock_conn_t *conn)
439 read_lock(&ksocknal_data.ksnd_global_lock);
440 if (!conn->ksnc_closing) {
441 LASSERT(atomic_read(&conn->ksnc_sock_refcount) > 0);
442 atomic_inc(&conn->ksnc_sock_refcount);
445 read_unlock(&ksocknal_data.ksnd_global_lock);
451 ksocknal_connsock_decref(ksock_conn_t *conn)
453 LASSERT(atomic_read(&conn->ksnc_sock_refcount) > 0);
454 if (atomic_dec_and_test(&conn->ksnc_sock_refcount)) {
455 LASSERT(conn->ksnc_closing);
456 libcfs_sock_release(conn->ksnc_sock);
457 conn->ksnc_sock = NULL;
458 ksocknal_finalize_zcreq(conn);
463 ksocknal_tx_addref(ksock_tx_t *tx)
465 LASSERT(atomic_read(&tx->tx_refcount) > 0);
466 atomic_inc(&tx->tx_refcount);
469 extern void ksocknal_tx_prep(ksock_conn_t *, ksock_tx_t *tx);
470 extern void ksocknal_tx_done(lnet_ni_t *ni, ksock_tx_t *tx);
473 ksocknal_tx_decref(ksock_tx_t *tx)
475 LASSERT(atomic_read(&tx->tx_refcount) > 0);
476 if (atomic_dec_and_test(&tx->tx_refcount))
477 ksocknal_tx_done(NULL, tx);
481 ksocknal_route_addref(ksock_route_t *route)
483 LASSERT(atomic_read(&route->ksnr_refcount) > 0);
484 atomic_inc(&route->ksnr_refcount);
487 extern void ksocknal_destroy_route(ksock_route_t *route);
490 ksocknal_route_decref(ksock_route_t *route)
492 LASSERT(atomic_read(&route->ksnr_refcount) > 0);
493 if (atomic_dec_and_test(&route->ksnr_refcount))
494 ksocknal_destroy_route(route);
498 ksocknal_peer_addref(ksock_peer_t *peer)
500 LASSERT(atomic_read(&peer->ksnp_refcount) > 0);
501 atomic_inc(&peer->ksnp_refcount);
504 extern void ksocknal_destroy_peer(ksock_peer_t *peer);
507 ksocknal_peer_decref(ksock_peer_t *peer)
509 LASSERT(atomic_read(&peer->ksnp_refcount) > 0);
510 if (atomic_dec_and_test(&peer->ksnp_refcount))
511 ksocknal_destroy_peer(peer);
514 int ksocknal_startup(lnet_ni_t *ni);
515 void ksocknal_shutdown(lnet_ni_t *ni);
516 int ksocknal_ctl(lnet_ni_t *ni, unsigned int cmd, void *arg);
517 int ksocknal_send(lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg);
518 int ksocknal_recv(lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg,
519 int delayed, unsigned int niov,
520 struct kvec *iov, lnet_kiov_t *kiov,
521 unsigned int offset, unsigned int mlen, unsigned int rlen);
522 int ksocknal_accept(lnet_ni_t *ni, struct socket *sock);
524 extern int ksocknal_add_peer(lnet_ni_t *ni, lnet_process_id_t id, __u32 ip, int port);
525 extern ksock_peer_t *ksocknal_find_peer_locked(lnet_ni_t *ni, lnet_process_id_t id);
526 extern ksock_peer_t *ksocknal_find_peer(lnet_ni_t *ni, lnet_process_id_t id);
527 extern void ksocknal_peer_failed(ksock_peer_t *peer);
528 extern int ksocknal_create_conn(lnet_ni_t *ni, ksock_route_t *route,
529 struct socket *sock, int type);
530 extern void ksocknal_close_conn_locked(ksock_conn_t *conn, int why);
531 extern void ksocknal_terminate_conn(ksock_conn_t *conn);
532 extern void ksocknal_destroy_conn(ksock_conn_t *conn);
533 extern int ksocknal_close_peer_conns_locked(ksock_peer_t *peer,
534 __u32 ipaddr, int why);
535 extern int ksocknal_close_conn_and_siblings(ksock_conn_t *conn, int why);
536 extern int ksocknal_close_matching_conns(lnet_process_id_t id, __u32 ipaddr);
537 extern ksock_conn_t *ksocknal_find_conn_locked(ksock_peer_t *peer,
538 ksock_tx_t *tx, int nonblk);
540 extern int ksocknal_launch_packet(lnet_ni_t *ni, ksock_tx_t *tx,
541 lnet_process_id_t id);
542 extern ksock_tx_t *ksocknal_alloc_tx(int type, int size);
543 extern void ksocknal_free_tx(ksock_tx_t *tx);
544 extern ksock_tx_t *ksocknal_alloc_tx_noop(__u64 cookie, int nonblk);
545 extern void ksocknal_next_tx_carrier(ksock_conn_t *conn);
546 extern void ksocknal_queue_tx_locked(ksock_tx_t *tx, ksock_conn_t *conn);
547 extern void ksocknal_txlist_done(lnet_ni_t *ni, struct list_head *txlist,
549 extern void ksocknal_notify(lnet_ni_t *ni, lnet_nid_t gw_nid, int alive);
550 extern void ksocknal_query(struct lnet_ni *ni, lnet_nid_t nid, unsigned long *when);
551 extern int ksocknal_thread_start(int (*fn)(void *arg), void *arg, char *name);
552 extern void ksocknal_thread_fini(void);
553 extern void ksocknal_launch_all_connections_locked(ksock_peer_t *peer);
554 extern ksock_route_t *ksocknal_find_connectable_route_locked(ksock_peer_t *peer);
555 extern ksock_route_t *ksocknal_find_connecting_route_locked(ksock_peer_t *peer);
556 extern int ksocknal_new_packet(ksock_conn_t *conn, int skip);
557 extern int ksocknal_scheduler(void *arg);
558 extern int ksocknal_connd(void *arg);
559 extern int ksocknal_reaper(void *arg);
560 extern int ksocknal_send_hello(lnet_ni_t *ni, ksock_conn_t *conn,
561 lnet_nid_t peer_nid, ksock_hello_msg_t *hello);
562 extern int ksocknal_recv_hello(lnet_ni_t *ni, ksock_conn_t *conn,
563 ksock_hello_msg_t *hello, lnet_process_id_t *id,
565 extern void ksocknal_read_callback(ksock_conn_t *conn);
566 extern void ksocknal_write_callback(ksock_conn_t *conn);
568 extern int ksocknal_lib_zc_capable(ksock_conn_t *conn);
569 extern void ksocknal_lib_save_callback(struct socket *sock, ksock_conn_t *conn);
570 extern void ksocknal_lib_set_callback(struct socket *sock, ksock_conn_t *conn);
571 extern void ksocknal_lib_reset_callback(struct socket *sock, ksock_conn_t *conn);
572 extern void ksocknal_lib_push_conn(ksock_conn_t *conn);
573 extern int ksocknal_lib_get_conn_addrs(ksock_conn_t *conn);
574 extern int ksocknal_lib_setup_sock(struct socket *so);
575 extern int ksocknal_lib_send_iov(ksock_conn_t *conn, ksock_tx_t *tx);
576 extern int ksocknal_lib_send_kiov(ksock_conn_t *conn, ksock_tx_t *tx);
577 extern void ksocknal_lib_eager_ack(ksock_conn_t *conn);
578 extern int ksocknal_lib_recv_iov(ksock_conn_t *conn);
579 extern int ksocknal_lib_recv_kiov(ksock_conn_t *conn);
580 extern int ksocknal_lib_get_conn_tunables(ksock_conn_t *conn, int *txmem,
581 int *rxmem, int *nagle);
583 extern int ksocknal_tunables_init(void);
585 extern void ksocknal_lib_csum_tx(ksock_tx_t *tx);
587 extern int ksocknal_lib_memory_pressure(ksock_conn_t *conn);
588 extern int ksocknal_lib_bind_thread_to_cpu(int id);