2 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
4 * Copyright (c) 2011, 2013, 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 #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>
46 #define SOCKNAL_PEER_HASH_SIZE 101 /* # peer lists */
47 #define SOCKNAL_RESCHED 100 /* # scheduler loops before reschedule */
48 #define SOCKNAL_INSANITY_RECONN 5000 /* connd is trying on reconn infinitely */
49 #define SOCKNAL_ENOMEM_RETRY CFS_TICK /* jiffies between retries */
51 #define SOCKNAL_SINGLE_FRAG_TX 0 /* disable multi-fragment sends */
52 #define SOCKNAL_SINGLE_FRAG_RX 0 /* disable multi-fragment receives */
54 #define SOCKNAL_VERSION_DEBUG 0 /* enable protocol version debugging */
56 /* risk kmap deadlock on multi-frag I/O (backs off to single-frag if disabled).
57 * no risk if we're not running on a CONFIG_HIGHMEM platform. */
59 # define SOCKNAL_RISK_KMAP_DEADLOCK 0
61 # define SOCKNAL_RISK_KMAP_DEADLOCK 1
64 struct ksock_sched_info;
66 typedef struct /* per scheduler state */
68 spinlock_t kss_lock; /* serialise */
69 struct list_head kss_rx_conns; /* conn waiting to be read */
70 /* conn waiting to be written */
71 struct list_head kss_tx_conns;
72 /* zombie noop tx list */
73 struct list_head kss_zombie_noop_txs;
74 wait_queue_head_t kss_waitq; /* where scheduler sleeps */
75 /* # connections assigned to this scheduler */
77 struct ksock_sched_info *kss_info; /* owner of it */
78 #if !SOCKNAL_SINGLE_FRAG_RX
79 struct page *kss_rx_scratch_pgs[LNET_MAX_IOV];
81 #if !SOCKNAL_SINGLE_FRAG_TX || !SOCKNAL_SINGLE_FRAG_RX
82 struct iovec kss_scratch_iov[LNET_MAX_IOV];
86 struct ksock_sched_info {
87 int ksi_nthreads_max; /* max allowed threads */
88 int ksi_nthreads; /* number of threads */
89 int ksi_cpt; /* CPT id */
90 ksock_sched_t *ksi_scheds; /* array of schedulers */
93 #define KSOCK_CPT_SHIFT 16
94 #define KSOCK_THREAD_ID(cpt, sid) (((cpt) << KSOCK_CPT_SHIFT) | (sid))
95 #define KSOCK_THREAD_CPT(id) ((id) >> KSOCK_CPT_SHIFT)
96 #define KSOCK_THREAD_SID(id) ((id) & ((1UL << KSOCK_CPT_SHIFT) - 1))
98 typedef struct /* in-use interface */
100 __u32 ksni_ipaddr; /* interface's IP address */
101 __u32 ksni_netmask; /* interface's network mask */
102 int ksni_nroutes; /* # routes using (active) */
103 int ksni_npeers; /* # peers using (passive) */
104 char ksni_name[IFNAMSIZ]; /* interface name */
109 /* "stuck" socket timeout (seconds) */
111 /* # scheduler threads in each pool while starting */
113 int *ksnd_nconnds; /* # connection daemons */
114 int *ksnd_nconnds_max; /* max # connection daemons */
115 int *ksnd_min_reconnectms; /* first connection retry after (ms)... */
116 int *ksnd_max_reconnectms; /* ...exponentially increasing to this */
117 int *ksnd_eager_ack; /* make TCP ack eagerly? */
118 int *ksnd_typed_conns; /* drive sockets by type? */
119 int *ksnd_min_bulk; /* smallest "large" message */
120 int *ksnd_tx_buffer_size; /* socket tx buffer size */
121 int *ksnd_rx_buffer_size; /* socket rx buffer size */
122 int *ksnd_nagle; /* enable NAGLE? */
123 int *ksnd_round_robin; /* round robin for multiple interfaces */
124 int *ksnd_keepalive; /* # secs for sending keepalive NOOP */
125 int *ksnd_keepalive_idle; /* # idle secs before 1st probe */
126 int *ksnd_keepalive_count; /* # probes */
127 int *ksnd_keepalive_intvl; /* time between probes */
128 int *ksnd_credits; /* # concurrent sends */
129 int *ksnd_peertxcredits; /* # concurrent sends to 1 peer */
130 int *ksnd_peerrtrcredits; /* # per-peer router buffer credits */
131 int *ksnd_peertimeout; /* seconds to consider peer dead */
132 int *ksnd_enable_csum; /* enable check sum */
133 int *ksnd_inject_csum_error; /* set non-zero to inject checksum error */
134 int *ksnd_nonblk_zcack; /* always send zc-ack on non-blocking connection */
135 unsigned int *ksnd_zc_min_payload; /* minimum zero copy payload size */
136 int *ksnd_zc_recv; /* enable ZC receive (for Chelsio TOE) */
137 int *ksnd_zc_recv_min_nfrags; /* minimum # of fragments to enable ZC receive */
139 int *ksnd_irq_affinity; /* enable IRQ affinity? */
141 #ifdef SOCKNAL_BACKOFF
142 int *ksnd_backoff_init; /* initial TCP backoff */
143 int *ksnd_backoff_max; /* maximum TCP backoff */
145 #if SOCKNAL_VERSION_DEBUG
146 int *ksnd_protocol; /* protocol version */
148 #if defined(CONFIG_SYSCTL) && !CFS_SYSFS_MODULE_PARM
149 struct ctl_table_header *ksnd_sysctl; /* sysctl interface */
155 __u64 ksnn_incarnation; /* my epoch */
156 spinlock_t ksnn_lock; /* serialise */
157 struct list_head ksnn_list; /* chain on global list */
158 int ksnn_npeers; /* # peers */
159 int ksnn_shutdown; /* shutting down? */
160 int ksnn_ninterfaces; /* IP interfaces */
161 ksock_interface_t ksnn_interfaces[LNET_MAX_INTERFACES];
165 #define SOCKNAL_CONND_TIMEOUT 120
166 /** reserved thread for accepting & creating new connd */
167 #define SOCKNAL_CONND_RESV 1
171 int ksnd_init; /* initialisation state */
172 int ksnd_nnets; /* # networks set up */
173 struct list_head ksnd_nets; /* list of nets */
174 /* stabilize peer/conn ops */
175 rwlock_t ksnd_global_lock;
176 /* hash table of all my known peers */
177 struct list_head *ksnd_peers;
178 int ksnd_peer_hash_size; /* size of ksnd_peers */
180 int ksnd_nthreads; /* # live threads */
181 int ksnd_shuttingdown; /* tell threads to exit */
182 /* schedulers information */
183 struct ksock_sched_info **ksnd_sched_info;
185 atomic_t ksnd_nactive_txs; /* #active txs */
187 /* conns to close: reaper_lock*/
188 struct list_head ksnd_deathrow_conns;
189 /* conns to free: reaper_lock */
190 struct list_head ksnd_zombie_conns;
191 /* conns to retry: reaper_lock*/
192 struct list_head ksnd_enomem_conns;
193 /* reaper sleeps here */
194 wait_queue_head_t ksnd_reaper_waitq;
195 /* when reaper will wake */
196 cfs_time_t ksnd_reaper_waketime;
198 spinlock_t ksnd_reaper_lock;
200 int ksnd_enomem_tx; /* test ENOMEM sender */
201 int ksnd_stall_tx; /* test sluggish sender */
202 int ksnd_stall_rx; /* test sluggish receiver */
204 /* incoming connection requests */
205 struct list_head ksnd_connd_connreqs;
206 /* routes waiting to be connected */
207 struct list_head ksnd_connd_routes;
208 /* connds sleep here */
209 wait_queue_head_t ksnd_connd_waitq;
210 /* # connds connecting */
211 int ksnd_connd_connecting;
212 /** time stamp of the last failed connecting attempt */
213 long ksnd_connd_failed_stamp;
214 /** # starting connd */
215 unsigned ksnd_connd_starting;
216 /** time stamp of the last starting connd */
217 long ksnd_connd_starting_stamp;
218 /** # running connd */
219 unsigned ksnd_connd_running;
221 spinlock_t ksnd_connd_lock;
223 /* list head for freed noop tx */
224 struct list_head ksnd_idle_noop_txs;
225 /* serialise, g_lock unsafe */
226 spinlock_t ksnd_tx_lock;
230 #define SOCKNAL_INIT_NOTHING 0
231 #define SOCKNAL_INIT_DATA 1
232 #define SOCKNAL_INIT_ALL 2
234 /* A packet just assembled for transmission is represented by 1 or more
235 * struct iovec fragments (the first frag contains the portals header),
236 * followed by 0 or more lnet_kiov_t fragments.
238 * On the receive side, initially 1 struct iovec fragment is posted for
239 * receive (the header). Once the header has been received, the payload is
240 * received into either struct iovec or lnet_kiov_t fragments, depending on
241 * what the header matched or whether the message needs forwarding. */
243 struct ksock_conn; /* forward ref */
244 struct ksock_peer; /* forward ref */
245 struct ksock_route; /* forward ref */
246 struct ksock_proto; /* forward ref */
248 typedef struct /* transmit packet */
250 struct list_head tx_list; /* queue on conn for transmission etc */
251 struct list_head tx_zc_list; /* queue on peer for ZC request */
252 atomic_t tx_refcount; /* tx reference count */
253 int tx_nob; /* # packet bytes */
254 int tx_resid; /* residual bytes */
255 int tx_niov; /* # packet iovec frags */
256 struct iovec *tx_iov; /* packet iovec frags */
257 int tx_nkiov; /* # packet page frags */
258 unsigned short tx_zc_aborted; /* aborted ZC request */
259 unsigned short tx_zc_capable:1; /* payload is large enough for ZC */
260 unsigned short tx_zc_checked:1; /* Have I checked if I should ZC? */
261 unsigned short tx_nonblk:1; /* it's a non-blocking ACK */
262 lnet_kiov_t *tx_kiov; /* packet page frags */
263 struct ksock_conn *tx_conn; /* owning conn */
264 lnet_msg_t *tx_lnetmsg; /* lnet message for lnet_finalize() */
265 cfs_time_t tx_deadline; /* when (in jiffies) tx times out */
266 ksock_msg_t tx_msg; /* socklnd message buffer */
267 int tx_desc_size; /* size of this descriptor */
270 struct iovec iov; /* virt hdr */
271 lnet_kiov_t kiov[0]; /* paged payload */
274 struct iovec iov[1]; /* virt hdr + payload */
279 #define KSOCK_NOOP_TX_SIZE ((int)offsetof(ksock_tx_t, tx_frags.paged.kiov[0]))
281 /* network zero copy callback descriptor embedded in ksock_tx_t */
283 /* space for the rx frag descriptors; we either read a single contiguous
284 * header, or up to LNET_MAX_IOV frags of payload of either type. */
286 struct iovec iov[LNET_MAX_IOV];
287 lnet_kiov_t kiov[LNET_MAX_IOV];
288 } ksock_rxiovspace_t;
290 #define SOCKNAL_RX_KSM_HEADER 1 /* reading ksock message header */
291 #define SOCKNAL_RX_LNET_HEADER 2 /* reading lnet message header */
292 #define SOCKNAL_RX_PARSE 3 /* Calling lnet_parse() */
293 #define SOCKNAL_RX_PARSE_WAIT 4 /* waiting to be told to read the body */
294 #define SOCKNAL_RX_LNET_PAYLOAD 5 /* reading lnet payload (to deliver here) */
295 #define SOCKNAL_RX_SLOP 6 /* skipping body */
297 typedef struct ksock_conn
299 struct ksock_peer *ksnc_peer; /* owning peer */
300 struct ksock_route *ksnc_route; /* owning route */
301 struct list_head ksnc_list; /* stash on peer's conn list */
302 cfs_socket_t *ksnc_sock; /* actual socket */
303 void *ksnc_saved_data_ready; /* socket's original data_ready() callback */
304 void *ksnc_saved_write_space; /* socket's original write_space() callback */
305 atomic_t ksnc_conn_refcount; /* conn refcount */
306 atomic_t ksnc_sock_refcount; /* sock refcount */
307 ksock_sched_t *ksnc_scheduler; /* who schedules this connection */
308 __u32 ksnc_myipaddr; /* my IP */
309 __u32 ksnc_ipaddr; /* peer's IP */
310 int ksnc_port; /* peer's port */
311 signed int ksnc_type:3; /* type of connection,
312 * should be signed value */
313 unsigned int ksnc_closing:1; /* being shut down */
314 unsigned int ksnc_flip:1; /* flip or not, only for V2.x */
315 unsigned int ksnc_zc_capable:1; /* enable to ZC */
316 struct ksock_proto *ksnc_proto; /* protocol for the connection */
320 /* where I enq waiting input or a forwarding descriptor */
321 struct list_head ksnc_rx_list;
322 cfs_time_t ksnc_rx_deadline; /* when (in jiffies) receive times out */
323 __u8 ksnc_rx_started; /* started receiving a message */
324 __u8 ksnc_rx_ready; /* data ready to read */
325 __u8 ksnc_rx_scheduled;/* being progressed */
326 __u8 ksnc_rx_state; /* what is being read */
327 int ksnc_rx_nob_left; /* # bytes to next hdr/body */
328 int ksnc_rx_nob_wanted; /* bytes actually wanted */
329 int ksnc_rx_niov; /* # iovec frags */
330 struct iovec *ksnc_rx_iov; /* the iovec frags */
331 int ksnc_rx_nkiov; /* # page frags */
332 lnet_kiov_t *ksnc_rx_kiov; /* the page frags */
333 ksock_rxiovspace_t ksnc_rx_iov_space;/* space for frag descriptors */
334 __u32 ksnc_rx_csum; /* partial checksum for incoming data */
335 void *ksnc_cookie; /* rx lnet_finalize passthru arg */
336 ksock_msg_t ksnc_msg; /* incoming message buffer:
337 * V2.x message takes the
339 * V1.x message is a bare
340 * lnet_hdr_t, it's stored in
341 * ksnc_msg.ksm_u.lnetmsg */
344 /* where I enq waiting for output space */
345 struct list_head ksnc_tx_list;
346 /* packets waiting to be sent */
347 struct list_head ksnc_tx_queue;
348 /* next TX that can carry a LNet message or ZC-ACK */
349 ksock_tx_t *ksnc_tx_carrier;
350 /* when (in jiffies) tx times out */
351 cfs_time_t ksnc_tx_deadline;
352 /* send buffer marker */
355 atomic_t ksnc_tx_nob;
358 /* being progressed */
359 int ksnc_tx_scheduled;
360 /* time stamp of the last posted TX */
361 cfs_time_t ksnc_tx_last_post;
364 typedef struct ksock_route
366 struct list_head ksnr_list; /* chain on peer route list */
367 struct list_head ksnr_connd_list; /* chain on ksnr_connd_routes */
368 struct ksock_peer *ksnr_peer; /* owning peer */
369 atomic_t ksnr_refcount; /* # users */
370 cfs_time_t ksnr_timeout; /* when (in jiffies) reconnection can happen next */
371 cfs_duration_t ksnr_retry_interval; /* how long between retries */
372 __u32 ksnr_myipaddr; /* my IP */
373 __u32 ksnr_ipaddr; /* IP address to connect to */
374 int ksnr_port; /* port to connect to */
375 unsigned int ksnr_scheduled:1; /* scheduled for attention */
376 unsigned int ksnr_connecting:1;/* connection establishment in progress */
377 unsigned int ksnr_connected:4; /* connections established by type */
378 unsigned int ksnr_deleted:1; /* been removed from peer? */
379 unsigned int ksnr_share_count; /* created explicitly? */
380 int ksnr_conn_count; /* # conns established by this route */
383 #define SOCKNAL_KEEPALIVE_PING 1 /* cookie for keepalive ping */
385 typedef struct ksock_peer
387 struct list_head ksnp_list; /* stash on global peer list */
388 cfs_time_t ksnp_last_alive; /* when (in jiffies) I was last alive */
389 lnet_process_id_t ksnp_id; /* who's on the other end(s) */
390 atomic_t ksnp_refcount; /* # users */
391 int ksnp_sharecount; /* lconf usage counter */
392 int ksnp_closing; /* being closed */
393 int ksnp_accepting;/* # passive connections pending */
394 int ksnp_error; /* errno on closing last conn */
395 __u64 ksnp_zc_next_cookie;/* ZC completion cookie */
396 __u64 ksnp_incarnation; /* latest known peer incarnation */
397 struct ksock_proto *ksnp_proto; /* latest known peer protocol */
398 struct list_head ksnp_conns; /* all active connections */
399 struct list_head ksnp_routes; /* routes */
400 struct list_head ksnp_tx_queue; /* waiting packets */
401 spinlock_t ksnp_lock; /* serialize, g_lock unsafe */
402 /* zero copy requests wait for ACK */
403 struct list_head ksnp_zc_req_list;
404 cfs_time_t ksnp_send_keepalive; /* time to send keepalive */
405 lnet_ni_t *ksnp_ni; /* which network */
406 int ksnp_n_passive_ips; /* # of... */
407 __u32 ksnp_passive_ips[LNET_MAX_INTERFACES]; /* preferred local interfaces */
410 typedef struct ksock_connreq
412 /* stash on ksnd_connd_connreqs */
413 struct list_head ksncr_list;
416 /* accepted socket */
417 cfs_socket_t *ksncr_sock;
420 extern ksock_nal_data_t ksocknal_data;
421 extern ksock_tunables_t ksocknal_tunables;
423 #define SOCKNAL_MATCH_NO 0 /* TX can't match type of connection */
424 #define SOCKNAL_MATCH_YES 1 /* TX matches type of connection */
425 #define SOCKNAL_MATCH_MAY 2 /* TX can be sent on the connection, but not preferred */
427 typedef struct ksock_proto
429 int pro_version; /* version number of protocol */
430 int (*pro_send_hello)(ksock_conn_t *, ksock_hello_msg_t *); /* handshake function */
431 int (*pro_recv_hello)(ksock_conn_t *, ksock_hello_msg_t *, int);/* handshake function */
432 void (*pro_pack)(ksock_tx_t *); /* message pack */
433 void (*pro_unpack)(ksock_msg_t *); /* message unpack */
434 ksock_tx_t *(*pro_queue_tx_msg)(ksock_conn_t *, ksock_tx_t *); /* queue tx on the connection */
435 int (*pro_queue_tx_zcack)(ksock_conn_t *, ksock_tx_t *, __u64); /* queue ZC ack on the connection */
436 int (*pro_handle_zcreq)(ksock_conn_t *, __u64, int); /* handle ZC request */
437 int (*pro_handle_zcack)(ksock_conn_t *, __u64, __u64); /* handle ZC ACK */
438 int (*pro_match_tx)(ksock_conn_t *, ksock_tx_t *, int); /* msg type matches the connection type:
440 * return MATCH_NO : no
441 * return MATCH_YES : matching type
442 * return MATCH_MAY : can be backup */
445 extern ksock_proto_t ksocknal_protocol_v1x;
446 extern ksock_proto_t ksocknal_protocol_v2x;
447 extern ksock_proto_t ksocknal_protocol_v3x;
449 #define KSOCK_PROTO_V1_MAJOR LNET_PROTO_TCP_VERSION_MAJOR
450 #define KSOCK_PROTO_V1_MINOR LNET_PROTO_TCP_VERSION_MINOR
451 #define KSOCK_PROTO_V1 KSOCK_PROTO_V1_MAJOR
453 #ifndef CPU_MASK_NONE
454 #define CPU_MASK_NONE 0UL
458 ksocknal_route_mask(void)
460 if (!*ksocknal_tunables.ksnd_typed_conns)
461 return (1 << SOCKLND_CONN_ANY);
463 return ((1 << SOCKLND_CONN_CONTROL) |
464 (1 << SOCKLND_CONN_BULK_IN) |
465 (1 << SOCKLND_CONN_BULK_OUT));
468 static inline struct list_head *
469 ksocknal_nid2peerlist (lnet_nid_t nid)
471 unsigned int hash = ((unsigned int)nid) % ksocknal_data.ksnd_peer_hash_size;
473 return (&ksocknal_data.ksnd_peers [hash]);
477 ksocknal_conn_addref (ksock_conn_t *conn)
479 LASSERT (atomic_read(&conn->ksnc_conn_refcount) > 0);
480 atomic_inc(&conn->ksnc_conn_refcount);
483 extern void ksocknal_queue_zombie_conn (ksock_conn_t *conn);
484 extern void ksocknal_finalize_zcreq(ksock_conn_t *conn);
487 ksocknal_conn_decref (ksock_conn_t *conn)
489 LASSERT (atomic_read(&conn->ksnc_conn_refcount) > 0);
490 if (atomic_dec_and_test(&conn->ksnc_conn_refcount))
491 ksocknal_queue_zombie_conn(conn);
495 ksocknal_connsock_addref (ksock_conn_t *conn)
499 read_lock(&ksocknal_data.ksnd_global_lock);
500 if (!conn->ksnc_closing) {
501 LASSERT(atomic_read(&conn->ksnc_sock_refcount) > 0);
502 atomic_inc(&conn->ksnc_sock_refcount);
505 read_unlock(&ksocknal_data.ksnd_global_lock);
511 ksocknal_connsock_decref (ksock_conn_t *conn)
513 LASSERT (atomic_read(&conn->ksnc_sock_refcount) > 0);
514 if (atomic_dec_and_test(&conn->ksnc_sock_refcount)) {
515 LASSERT (conn->ksnc_closing);
516 libcfs_sock_release(conn->ksnc_sock);
517 conn->ksnc_sock = NULL;
518 ksocknal_finalize_zcreq(conn);
523 ksocknal_tx_addref (ksock_tx_t *tx)
525 LASSERT (atomic_read(&tx->tx_refcount) > 0);
526 atomic_inc(&tx->tx_refcount);
529 extern void ksocknal_tx_prep (ksock_conn_t *, ksock_tx_t *tx);
530 extern void ksocknal_tx_done (lnet_ni_t *ni, ksock_tx_t *tx);
533 ksocknal_tx_decref (ksock_tx_t *tx)
535 LASSERT (atomic_read(&tx->tx_refcount) > 0);
536 if (atomic_dec_and_test(&tx->tx_refcount))
537 ksocknal_tx_done(NULL, tx);
541 ksocknal_route_addref (ksock_route_t *route)
543 LASSERT (atomic_read(&route->ksnr_refcount) > 0);
544 atomic_inc(&route->ksnr_refcount);
547 extern void ksocknal_destroy_route (ksock_route_t *route);
550 ksocknal_route_decref (ksock_route_t *route)
552 LASSERT (atomic_read (&route->ksnr_refcount) > 0);
553 if (atomic_dec_and_test(&route->ksnr_refcount))
554 ksocknal_destroy_route (route);
558 ksocknal_peer_addref (ksock_peer_t *peer)
560 LASSERT (atomic_read (&peer->ksnp_refcount) > 0);
561 atomic_inc(&peer->ksnp_refcount);
564 extern void ksocknal_destroy_peer (ksock_peer_t *peer);
567 ksocknal_peer_decref (ksock_peer_t *peer)
569 LASSERT (atomic_read (&peer->ksnp_refcount) > 0);
570 if (atomic_dec_and_test(&peer->ksnp_refcount))
571 ksocknal_destroy_peer (peer);
574 int ksocknal_startup (lnet_ni_t *ni);
575 void ksocknal_shutdown (lnet_ni_t *ni);
576 int ksocknal_ctl(lnet_ni_t *ni, unsigned int cmd, void *arg);
577 int ksocknal_send (lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg);
578 int ksocknal_recv(lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg,
579 int delayed, unsigned int niov,
580 struct iovec *iov, lnet_kiov_t *kiov,
581 unsigned int offset, unsigned int mlen, unsigned int rlen);
582 int ksocknal_accept(lnet_ni_t *ni, cfs_socket_t *sock);
584 extern int ksocknal_add_peer(lnet_ni_t *ni, lnet_process_id_t id, __u32 ip, int port);
585 extern ksock_peer_t *ksocknal_find_peer_locked (lnet_ni_t *ni, lnet_process_id_t id);
586 extern ksock_peer_t *ksocknal_find_peer (lnet_ni_t *ni, lnet_process_id_t id);
587 extern void ksocknal_peer_failed (ksock_peer_t *peer);
588 extern int ksocknal_create_conn (lnet_ni_t *ni, ksock_route_t *route,
589 cfs_socket_t *sock, int type);
590 extern void ksocknal_close_conn_locked (ksock_conn_t *conn, int why);
591 extern void ksocknal_terminate_conn (ksock_conn_t *conn);
592 extern void ksocknal_destroy_conn (ksock_conn_t *conn);
593 extern int ksocknal_close_peer_conns_locked (ksock_peer_t *peer,
594 __u32 ipaddr, int why);
595 extern int ksocknal_close_conn_and_siblings (ksock_conn_t *conn, int why);
596 extern int ksocknal_close_matching_conns(lnet_process_id_t id, __u32 ipaddr);
597 extern ksock_conn_t *ksocknal_find_conn_locked(ksock_peer_t *peer,
598 ksock_tx_t *tx, int nonblk);
600 extern int ksocknal_launch_packet(lnet_ni_t *ni, ksock_tx_t *tx,
601 lnet_process_id_t id);
602 extern ksock_tx_t *ksocknal_alloc_tx(int type, int size);
603 extern void ksocknal_free_tx (ksock_tx_t *tx);
604 extern ksock_tx_t *ksocknal_alloc_tx_noop(__u64 cookie, int nonblk);
605 extern void ksocknal_next_tx_carrier(ksock_conn_t *conn);
606 extern void ksocknal_queue_tx_locked (ksock_tx_t *tx, ksock_conn_t *conn);
607 extern void ksocknal_txlist_done(lnet_ni_t *ni, struct list_head *txlist,
609 extern void ksocknal_notify (lnet_ni_t *ni, lnet_nid_t gw_nid, int alive);
610 extern void ksocknal_query (struct lnet_ni *ni, lnet_nid_t nid, cfs_time_t *when);
611 extern int ksocknal_thread_start(int (*fn)(void *arg), void *arg, char *name);
612 extern void ksocknal_thread_fini (void);
613 extern void ksocknal_launch_all_connections_locked (ksock_peer_t *peer);
614 extern ksock_route_t *ksocknal_find_connectable_route_locked (ksock_peer_t *peer);
615 extern ksock_route_t *ksocknal_find_connecting_route_locked (ksock_peer_t *peer);
616 extern int ksocknal_new_packet (ksock_conn_t *conn, int skip);
617 extern int ksocknal_scheduler (void *arg);
618 extern int ksocknal_connd (void *arg);
619 extern int ksocknal_reaper (void *arg);
620 extern int ksocknal_send_hello (lnet_ni_t *ni, ksock_conn_t *conn,
621 lnet_nid_t peer_nid, ksock_hello_msg_t *hello);
622 extern int ksocknal_recv_hello (lnet_ni_t *ni, ksock_conn_t *conn,
623 ksock_hello_msg_t *hello, lnet_process_id_t *id,
625 extern void ksocknal_read_callback(ksock_conn_t *conn);
626 extern void ksocknal_write_callback(ksock_conn_t *conn);
628 extern int ksocknal_lib_zc_capable(ksock_conn_t *conn);
629 extern void ksocknal_lib_save_callback(cfs_socket_t *sock, ksock_conn_t *conn);
630 extern void ksocknal_lib_set_callback(cfs_socket_t *sock, ksock_conn_t *conn);
631 extern void ksocknal_lib_reset_callback(cfs_socket_t *sock, ksock_conn_t *conn);
632 extern void ksocknal_lib_push_conn (ksock_conn_t *conn);
633 extern int ksocknal_lib_get_conn_addrs (ksock_conn_t *conn);
634 extern int ksocknal_lib_setup_sock (cfs_socket_t *so);
635 extern int ksocknal_lib_send_iov (ksock_conn_t *conn, ksock_tx_t *tx);
636 extern int ksocknal_lib_send_kiov (ksock_conn_t *conn, ksock_tx_t *tx);
637 extern void ksocknal_lib_eager_ack (ksock_conn_t *conn);
638 extern int ksocknal_lib_recv_iov (ksock_conn_t *conn);
639 extern int ksocknal_lib_recv_kiov (ksock_conn_t *conn);
640 extern int ksocknal_lib_get_conn_tunables (ksock_conn_t *conn, int *txmem,
641 int *rxmem, int *nagle);
643 extern int ksocknal_tunables_init(void);
644 extern void ksocknal_tunables_fini(void);
645 extern int ksocknal_lib_tunables_init(void);
646 extern void ksocknal_lib_tunables_fini(void);
648 extern void ksocknal_lib_csum_tx(ksock_tx_t *tx);
650 extern int ksocknal_lib_memory_pressure(ksock_conn_t *conn);
651 extern int ksocknal_lib_bind_thread_to_cpu(int id);