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"
36 #error Unsupported Operating System
39 #include <libcfs/libcfs.h>
40 #include <lnet/lnet.h>
41 #include <lnet/lib-lnet.h>
42 #include <lnet/socklnd.h>
44 #define SOCKNAL_PEER_HASH_SIZE 101 /* # peer lists */
45 #define SOCKNAL_RESCHED 100 /* # scheduler loops before reschedule */
46 #define SOCKNAL_INSANITY_RECONN 5000 /* connd is trying on reconn infinitely */
47 #define SOCKNAL_ENOMEM_RETRY CFS_TICK /* jiffies between retries */
49 #define SOCKNAL_SINGLE_FRAG_TX 0 /* disable multi-fragment sends */
50 #define SOCKNAL_SINGLE_FRAG_RX 0 /* disable multi-fragment receives */
52 #define SOCKNAL_VERSION_DEBUG 0 /* enable protocol version debugging */
54 /* risk kmap deadlock on multi-frag I/O (backs off to single-frag if disabled).
55 * no risk if we're not running on a CONFIG_HIGHMEM platform. */
57 # define SOCKNAL_RISK_KMAP_DEADLOCK 0
59 # define SOCKNAL_RISK_KMAP_DEADLOCK 1
62 struct ksock_sched_info;
64 typedef struct /* per scheduler state */
66 spinlock_t kss_lock; /* serialise */
67 struct list_head kss_rx_conns; /* conn waiting to be read */
68 /* conn waiting to be written */
69 struct list_head kss_tx_conns;
70 /* zombie noop tx list */
71 struct list_head kss_zombie_noop_txs;
72 wait_queue_head_t kss_waitq; /* where scheduler sleeps */
73 /* # connections assigned to this scheduler */
75 struct ksock_sched_info *kss_info; /* owner of it */
76 #if !SOCKNAL_SINGLE_FRAG_RX
77 struct page *kss_rx_scratch_pgs[LNET_MAX_IOV];
79 #if !SOCKNAL_SINGLE_FRAG_TX || !SOCKNAL_SINGLE_FRAG_RX
80 struct iovec kss_scratch_iov[LNET_MAX_IOV];
84 struct ksock_sched_info {
85 int ksi_nthreads_max; /* max allowed threads */
86 int ksi_nthreads; /* number of threads */
87 int ksi_cpt; /* CPT id */
88 ksock_sched_t *ksi_scheds; /* array of schedulers */
91 #define KSOCK_CPT_SHIFT 16
92 #define KSOCK_THREAD_ID(cpt, sid) (((cpt) << KSOCK_CPT_SHIFT) | (sid))
93 #define KSOCK_THREAD_CPT(id) ((id) >> KSOCK_CPT_SHIFT)
94 #define KSOCK_THREAD_SID(id) ((id) & ((1UL << KSOCK_CPT_SHIFT) - 1))
96 typedef struct /* in-use interface */
98 __u32 ksni_ipaddr; /* interface's IP address */
99 __u32 ksni_netmask; /* interface's network mask */
100 int ksni_nroutes; /* # routes using (active) */
101 int ksni_npeers; /* # peers using (passive) */
102 char ksni_name[IFNAMSIZ]; /* interface name */
107 /* "stuck" socket timeout (seconds) */
109 /* # scheduler threads in each pool while starting */
111 int *ksnd_nconnds; /* # connection daemons */
112 int *ksnd_nconnds_max; /* max # connection daemons */
113 int *ksnd_min_reconnectms; /* first connection retry after (ms)... */
114 int *ksnd_max_reconnectms; /* ...exponentially increasing to this */
115 int *ksnd_eager_ack; /* make TCP ack eagerly? */
116 int *ksnd_typed_conns; /* drive sockets by type? */
117 int *ksnd_min_bulk; /* smallest "large" message */
118 int *ksnd_tx_buffer_size; /* socket tx buffer size */
119 int *ksnd_rx_buffer_size; /* socket rx buffer size */
120 int *ksnd_nagle; /* enable NAGLE? */
121 int *ksnd_round_robin; /* round robin for multiple interfaces */
122 int *ksnd_keepalive; /* # secs for sending keepalive NOOP */
123 int *ksnd_keepalive_idle; /* # idle secs before 1st probe */
124 int *ksnd_keepalive_count; /* # probes */
125 int *ksnd_keepalive_intvl; /* time between probes */
126 int *ksnd_credits; /* # concurrent sends */
127 int *ksnd_peertxcredits; /* # concurrent sends to 1 peer */
128 int *ksnd_peerrtrcredits; /* # per-peer router buffer credits */
129 int *ksnd_peertimeout; /* seconds to consider peer dead */
130 int *ksnd_enable_csum; /* enable check sum */
131 int *ksnd_inject_csum_error; /* set non-zero to inject checksum error */
132 int *ksnd_nonblk_zcack; /* always send zc-ack on non-blocking connection */
133 unsigned int *ksnd_zc_min_payload; /* minimum zero copy payload size */
134 int *ksnd_zc_recv; /* enable ZC receive (for Chelsio TOE) */
135 int *ksnd_zc_recv_min_nfrags; /* minimum # of fragments to enable ZC receive */
137 int *ksnd_irq_affinity; /* enable IRQ affinity? */
139 #ifdef SOCKNAL_BACKOFF
140 int *ksnd_backoff_init; /* initial TCP backoff */
141 int *ksnd_backoff_max; /* maximum TCP backoff */
143 #if SOCKNAL_VERSION_DEBUG
144 int *ksnd_protocol; /* protocol version */
146 #if defined(CONFIG_SYSCTL) && !CFS_SYSFS_MODULE_PARM
147 struct ctl_table_header *ksnd_sysctl; /* sysctl interface */
153 __u64 ksnn_incarnation; /* my epoch */
154 spinlock_t ksnn_lock; /* serialise */
155 struct list_head ksnn_list; /* chain on global list */
156 int ksnn_npeers; /* # peers */
157 int ksnn_shutdown; /* shutting down? */
158 int ksnn_ninterfaces; /* IP interfaces */
159 ksock_interface_t ksnn_interfaces[LNET_MAX_INTERFACES];
163 #define SOCKNAL_CONND_TIMEOUT 120
164 /** reserved thread for accepting & creating new connd */
165 #define SOCKNAL_CONND_RESV 1
169 int ksnd_init; /* initialisation state */
170 int ksnd_nnets; /* # networks set up */
171 struct list_head ksnd_nets; /* list of nets */
172 /* stabilize peer/conn ops */
173 rwlock_t ksnd_global_lock;
174 /* hash table of all my known peers */
175 struct list_head *ksnd_peers;
176 int ksnd_peer_hash_size; /* size of ksnd_peers */
178 int ksnd_nthreads; /* # live threads */
179 int ksnd_shuttingdown; /* tell threads to exit */
180 /* schedulers information */
181 struct ksock_sched_info **ksnd_sched_info;
183 atomic_t ksnd_nactive_txs; /* #active txs */
185 /* conns to close: reaper_lock*/
186 struct list_head ksnd_deathrow_conns;
187 /* conns to free: reaper_lock */
188 struct list_head ksnd_zombie_conns;
189 /* conns to retry: reaper_lock*/
190 struct list_head ksnd_enomem_conns;
191 /* reaper sleeps here */
192 wait_queue_head_t ksnd_reaper_waitq;
193 /* when reaper will wake */
194 cfs_time_t ksnd_reaper_waketime;
196 spinlock_t ksnd_reaper_lock;
198 int ksnd_enomem_tx; /* test ENOMEM sender */
199 int ksnd_stall_tx; /* test sluggish sender */
200 int ksnd_stall_rx; /* test sluggish receiver */
202 /* incoming connection requests */
203 struct list_head ksnd_connd_connreqs;
204 /* routes waiting to be connected */
205 struct list_head ksnd_connd_routes;
206 /* connds sleep here */
207 wait_queue_head_t ksnd_connd_waitq;
208 /* # connds connecting */
209 int ksnd_connd_connecting;
210 /** time stamp of the last failed connecting attempt */
211 long ksnd_connd_failed_stamp;
212 /** # starting connd */
213 unsigned ksnd_connd_starting;
214 /** time stamp of the last starting connd */
215 long ksnd_connd_starting_stamp;
216 /** # running connd */
217 unsigned ksnd_connd_running;
219 spinlock_t ksnd_connd_lock;
221 /* list head for freed noop tx */
222 struct list_head ksnd_idle_noop_txs;
223 /* serialise, g_lock unsafe */
224 spinlock_t ksnd_tx_lock;
228 #define SOCKNAL_INIT_NOTHING 0
229 #define SOCKNAL_INIT_DATA 1
230 #define SOCKNAL_INIT_ALL 2
232 /* A packet just assembled for transmission is represented by 1 or more
233 * struct iovec fragments (the first frag contains the portals header),
234 * followed by 0 or more lnet_kiov_t fragments.
236 * On the receive side, initially 1 struct iovec fragment is posted for
237 * receive (the header). Once the header has been received, the payload is
238 * received into either struct iovec or lnet_kiov_t fragments, depending on
239 * what the header matched or whether the message needs forwarding. */
241 struct ksock_conn; /* forward ref */
242 struct ksock_peer; /* forward ref */
243 struct ksock_route; /* forward ref */
244 struct ksock_proto; /* forward ref */
246 typedef struct /* transmit packet */
248 struct list_head tx_list; /* queue on conn for transmission etc */
249 struct list_head tx_zc_list; /* queue on peer for ZC request */
250 atomic_t tx_refcount; /* tx reference count */
251 int tx_nob; /* # packet bytes */
252 int tx_resid; /* residual bytes */
253 int tx_niov; /* # packet iovec frags */
254 struct iovec *tx_iov; /* packet iovec frags */
255 int tx_nkiov; /* # packet page frags */
256 unsigned short tx_zc_aborted; /* aborted ZC request */
257 unsigned short tx_zc_capable:1; /* payload is large enough for ZC */
258 unsigned short tx_zc_checked:1; /* Have I checked if I should ZC? */
259 unsigned short tx_nonblk:1; /* it's a non-blocking ACK */
260 lnet_kiov_t *tx_kiov; /* packet page frags */
261 struct ksock_conn *tx_conn; /* owning conn */
262 lnet_msg_t *tx_lnetmsg; /* lnet message for lnet_finalize() */
263 cfs_time_t tx_deadline; /* when (in jiffies) tx times out */
264 ksock_msg_t tx_msg; /* socklnd message buffer */
265 int tx_desc_size; /* size of this descriptor */
268 struct iovec iov; /* virt hdr */
269 lnet_kiov_t kiov[0]; /* paged payload */
272 struct iovec iov[1]; /* virt hdr + payload */
277 #define KSOCK_NOOP_TX_SIZE ((int)offsetof(ksock_tx_t, tx_frags.paged.kiov[0]))
279 /* network zero copy callback descriptor embedded in ksock_tx_t */
281 /* space for the rx frag descriptors; we either read a single contiguous
282 * header, or up to LNET_MAX_IOV frags of payload of either type. */
284 struct iovec iov[LNET_MAX_IOV];
285 lnet_kiov_t kiov[LNET_MAX_IOV];
286 } ksock_rxiovspace_t;
288 #define SOCKNAL_RX_KSM_HEADER 1 /* reading ksock message header */
289 #define SOCKNAL_RX_LNET_HEADER 2 /* reading lnet message header */
290 #define SOCKNAL_RX_PARSE 3 /* Calling lnet_parse() */
291 #define SOCKNAL_RX_PARSE_WAIT 4 /* waiting to be told to read the body */
292 #define SOCKNAL_RX_LNET_PAYLOAD 5 /* reading lnet payload (to deliver here) */
293 #define SOCKNAL_RX_SLOP 6 /* skipping body */
295 typedef struct ksock_conn
297 struct ksock_peer *ksnc_peer; /* owning peer */
298 struct ksock_route *ksnc_route; /* owning route */
299 struct list_head ksnc_list; /* stash on peer's conn list */
300 cfs_socket_t *ksnc_sock; /* actual socket */
301 void *ksnc_saved_data_ready; /* socket's original data_ready() callback */
302 void *ksnc_saved_write_space; /* socket's original write_space() callback */
303 atomic_t ksnc_conn_refcount; /* conn refcount */
304 atomic_t ksnc_sock_refcount; /* sock refcount */
305 ksock_sched_t *ksnc_scheduler; /* who schedules this connection */
306 __u32 ksnc_myipaddr; /* my IP */
307 __u32 ksnc_ipaddr; /* peer's IP */
308 int ksnc_port; /* peer's port */
309 signed int ksnc_type:3; /* type of connection,
310 * should be signed value */
311 unsigned int ksnc_closing:1; /* being shut down */
312 unsigned int ksnc_flip:1; /* flip or not, only for V2.x */
313 unsigned int ksnc_zc_capable:1; /* enable to ZC */
314 struct ksock_proto *ksnc_proto; /* protocol for the connection */
318 /* where I enq waiting input or a forwarding descriptor */
319 struct list_head ksnc_rx_list;
320 cfs_time_t ksnc_rx_deadline; /* when (in jiffies) receive times out */
321 __u8 ksnc_rx_started; /* started receiving a message */
322 __u8 ksnc_rx_ready; /* data ready to read */
323 __u8 ksnc_rx_scheduled;/* being progressed */
324 __u8 ksnc_rx_state; /* what is being read */
325 int ksnc_rx_nob_left; /* # bytes to next hdr/body */
326 int ksnc_rx_nob_wanted; /* bytes actually wanted */
327 int ksnc_rx_niov; /* # iovec frags */
328 struct iovec *ksnc_rx_iov; /* the iovec frags */
329 int ksnc_rx_nkiov; /* # page frags */
330 lnet_kiov_t *ksnc_rx_kiov; /* the page frags */
331 ksock_rxiovspace_t ksnc_rx_iov_space;/* space for frag descriptors */
332 __u32 ksnc_rx_csum; /* partial checksum for incoming data */
333 void *ksnc_cookie; /* rx lnet_finalize passthru arg */
334 ksock_msg_t ksnc_msg; /* incoming message buffer:
335 * V2.x message takes the
337 * V1.x message is a bare
338 * lnet_hdr_t, it's stored in
339 * ksnc_msg.ksm_u.lnetmsg */
342 /* where I enq waiting for output space */
343 struct list_head ksnc_tx_list;
344 /* packets waiting to be sent */
345 struct list_head ksnc_tx_queue;
346 /* next TX that can carry a LNet message or ZC-ACK */
347 ksock_tx_t *ksnc_tx_carrier;
348 /* when (in jiffies) tx times out */
349 cfs_time_t ksnc_tx_deadline;
350 /* send buffer marker */
353 atomic_t ksnc_tx_nob;
356 /* being progressed */
357 int ksnc_tx_scheduled;
358 /* time stamp of the last posted TX */
359 cfs_time_t ksnc_tx_last_post;
362 typedef struct ksock_route
364 struct list_head ksnr_list; /* chain on peer route list */
365 struct list_head ksnr_connd_list; /* chain on ksnr_connd_routes */
366 struct ksock_peer *ksnr_peer; /* owning peer */
367 atomic_t ksnr_refcount; /* # users */
368 cfs_time_t ksnr_timeout; /* when (in jiffies) reconnection can happen next */
369 cfs_duration_t ksnr_retry_interval; /* how long between retries */
370 __u32 ksnr_myipaddr; /* my IP */
371 __u32 ksnr_ipaddr; /* IP address to connect to */
372 int ksnr_port; /* port to connect to */
373 unsigned int ksnr_scheduled:1; /* scheduled for attention */
374 unsigned int ksnr_connecting:1;/* connection establishment in progress */
375 unsigned int ksnr_connected:4; /* connections established by type */
376 unsigned int ksnr_deleted:1; /* been removed from peer? */
377 unsigned int ksnr_share_count; /* created explicitly? */
378 int ksnr_conn_count; /* # conns established by this route */
381 #define SOCKNAL_KEEPALIVE_PING 1 /* cookie for keepalive ping */
383 typedef struct ksock_peer
385 struct list_head ksnp_list; /* stash on global peer list */
386 cfs_time_t ksnp_last_alive; /* when (in jiffies) I was last alive */
387 lnet_process_id_t ksnp_id; /* who's on the other end(s) */
388 atomic_t ksnp_refcount; /* # users */
389 int ksnp_sharecount; /* lconf usage counter */
390 int ksnp_closing; /* being closed */
391 int ksnp_accepting;/* # passive connections pending */
392 int ksnp_error; /* errno on closing last conn */
393 __u64 ksnp_zc_next_cookie;/* ZC completion cookie */
394 __u64 ksnp_incarnation; /* latest known peer incarnation */
395 struct ksock_proto *ksnp_proto; /* latest known peer protocol */
396 struct list_head ksnp_conns; /* all active connections */
397 struct list_head ksnp_routes; /* routes */
398 struct list_head ksnp_tx_queue; /* waiting packets */
399 spinlock_t ksnp_lock; /* serialize, g_lock unsafe */
400 /* zero copy requests wait for ACK */
401 struct list_head ksnp_zc_req_list;
402 cfs_time_t ksnp_send_keepalive; /* time to send keepalive */
403 lnet_ni_t *ksnp_ni; /* which network */
404 int ksnp_n_passive_ips; /* # of... */
405 __u32 ksnp_passive_ips[LNET_MAX_INTERFACES]; /* preferred local interfaces */
408 typedef struct ksock_connreq
410 /* stash on ksnd_connd_connreqs */
411 struct list_head ksncr_list;
414 /* accepted socket */
415 cfs_socket_t *ksncr_sock;
418 extern ksock_nal_data_t ksocknal_data;
419 extern ksock_tunables_t ksocknal_tunables;
421 #define SOCKNAL_MATCH_NO 0 /* TX can't match type of connection */
422 #define SOCKNAL_MATCH_YES 1 /* TX matches type of connection */
423 #define SOCKNAL_MATCH_MAY 2 /* TX can be sent on the connection, but not preferred */
425 typedef struct ksock_proto
427 int pro_version; /* version number of protocol */
428 int (*pro_send_hello)(ksock_conn_t *, ksock_hello_msg_t *); /* handshake function */
429 int (*pro_recv_hello)(ksock_conn_t *, ksock_hello_msg_t *, int);/* handshake function */
430 void (*pro_pack)(ksock_tx_t *); /* message pack */
431 void (*pro_unpack)(ksock_msg_t *); /* message unpack */
432 ksock_tx_t *(*pro_queue_tx_msg)(ksock_conn_t *, ksock_tx_t *); /* queue tx on the connection */
433 int (*pro_queue_tx_zcack)(ksock_conn_t *, ksock_tx_t *, __u64); /* queue ZC ack on the connection */
434 int (*pro_handle_zcreq)(ksock_conn_t *, __u64, int); /* handle ZC request */
435 int (*pro_handle_zcack)(ksock_conn_t *, __u64, __u64); /* handle ZC ACK */
436 int (*pro_match_tx)(ksock_conn_t *, ksock_tx_t *, int); /* msg type matches the connection type:
438 * return MATCH_NO : no
439 * return MATCH_YES : matching type
440 * return MATCH_MAY : can be backup */
443 extern ksock_proto_t ksocknal_protocol_v1x;
444 extern ksock_proto_t ksocknal_protocol_v2x;
445 extern ksock_proto_t ksocknal_protocol_v3x;
447 #define KSOCK_PROTO_V1_MAJOR LNET_PROTO_TCP_VERSION_MAJOR
448 #define KSOCK_PROTO_V1_MINOR LNET_PROTO_TCP_VERSION_MINOR
449 #define KSOCK_PROTO_V1 KSOCK_PROTO_V1_MAJOR
451 #ifndef CPU_MASK_NONE
452 #define CPU_MASK_NONE 0UL
456 ksocknal_route_mask(void)
458 if (!*ksocknal_tunables.ksnd_typed_conns)
459 return (1 << SOCKLND_CONN_ANY);
461 return ((1 << SOCKLND_CONN_CONTROL) |
462 (1 << SOCKLND_CONN_BULK_IN) |
463 (1 << SOCKLND_CONN_BULK_OUT));
466 static inline struct list_head *
467 ksocknal_nid2peerlist (lnet_nid_t nid)
469 unsigned int hash = ((unsigned int)nid) % ksocknal_data.ksnd_peer_hash_size;
471 return (&ksocknal_data.ksnd_peers [hash]);
475 ksocknal_conn_addref (ksock_conn_t *conn)
477 LASSERT (atomic_read(&conn->ksnc_conn_refcount) > 0);
478 atomic_inc(&conn->ksnc_conn_refcount);
481 extern void ksocknal_queue_zombie_conn (ksock_conn_t *conn);
482 extern void ksocknal_finalize_zcreq(ksock_conn_t *conn);
485 ksocknal_conn_decref (ksock_conn_t *conn)
487 LASSERT (atomic_read(&conn->ksnc_conn_refcount) > 0);
488 if (atomic_dec_and_test(&conn->ksnc_conn_refcount))
489 ksocknal_queue_zombie_conn(conn);
493 ksocknal_connsock_addref (ksock_conn_t *conn)
497 read_lock(&ksocknal_data.ksnd_global_lock);
498 if (!conn->ksnc_closing) {
499 LASSERT(atomic_read(&conn->ksnc_sock_refcount) > 0);
500 atomic_inc(&conn->ksnc_sock_refcount);
503 read_unlock(&ksocknal_data.ksnd_global_lock);
509 ksocknal_connsock_decref (ksock_conn_t *conn)
511 LASSERT (atomic_read(&conn->ksnc_sock_refcount) > 0);
512 if (atomic_dec_and_test(&conn->ksnc_sock_refcount)) {
513 LASSERT (conn->ksnc_closing);
514 libcfs_sock_release(conn->ksnc_sock);
515 conn->ksnc_sock = NULL;
516 ksocknal_finalize_zcreq(conn);
521 ksocknal_tx_addref (ksock_tx_t *tx)
523 LASSERT (atomic_read(&tx->tx_refcount) > 0);
524 atomic_inc(&tx->tx_refcount);
527 extern void ksocknal_tx_prep (ksock_conn_t *, ksock_tx_t *tx);
528 extern void ksocknal_tx_done (lnet_ni_t *ni, ksock_tx_t *tx);
531 ksocknal_tx_decref (ksock_tx_t *tx)
533 LASSERT (atomic_read(&tx->tx_refcount) > 0);
534 if (atomic_dec_and_test(&tx->tx_refcount))
535 ksocknal_tx_done(NULL, tx);
539 ksocknal_route_addref (ksock_route_t *route)
541 LASSERT (atomic_read(&route->ksnr_refcount) > 0);
542 atomic_inc(&route->ksnr_refcount);
545 extern void ksocknal_destroy_route (ksock_route_t *route);
548 ksocknal_route_decref (ksock_route_t *route)
550 LASSERT (atomic_read (&route->ksnr_refcount) > 0);
551 if (atomic_dec_and_test(&route->ksnr_refcount))
552 ksocknal_destroy_route (route);
556 ksocknal_peer_addref (ksock_peer_t *peer)
558 LASSERT (atomic_read (&peer->ksnp_refcount) > 0);
559 atomic_inc(&peer->ksnp_refcount);
562 extern void ksocknal_destroy_peer (ksock_peer_t *peer);
565 ksocknal_peer_decref (ksock_peer_t *peer)
567 LASSERT (atomic_read (&peer->ksnp_refcount) > 0);
568 if (atomic_dec_and_test(&peer->ksnp_refcount))
569 ksocknal_destroy_peer (peer);
572 int ksocknal_startup (lnet_ni_t *ni);
573 void ksocknal_shutdown (lnet_ni_t *ni);
574 int ksocknal_ctl(lnet_ni_t *ni, unsigned int cmd, void *arg);
575 int ksocknal_send (lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg);
576 int ksocknal_recv(lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg,
577 int delayed, unsigned int niov,
578 struct iovec *iov, lnet_kiov_t *kiov,
579 unsigned int offset, unsigned int mlen, unsigned int rlen);
580 int ksocknal_accept(lnet_ni_t *ni, cfs_socket_t *sock);
582 extern int ksocknal_add_peer(lnet_ni_t *ni, lnet_process_id_t id, __u32 ip, int port);
583 extern ksock_peer_t *ksocknal_find_peer_locked (lnet_ni_t *ni, lnet_process_id_t id);
584 extern ksock_peer_t *ksocknal_find_peer (lnet_ni_t *ni, lnet_process_id_t id);
585 extern void ksocknal_peer_failed (ksock_peer_t *peer);
586 extern int ksocknal_create_conn (lnet_ni_t *ni, ksock_route_t *route,
587 cfs_socket_t *sock, int type);
588 extern void ksocknal_close_conn_locked (ksock_conn_t *conn, int why);
589 extern void ksocknal_terminate_conn (ksock_conn_t *conn);
590 extern void ksocknal_destroy_conn (ksock_conn_t *conn);
591 extern int ksocknal_close_peer_conns_locked (ksock_peer_t *peer,
592 __u32 ipaddr, int why);
593 extern int ksocknal_close_conn_and_siblings (ksock_conn_t *conn, int why);
594 extern int ksocknal_close_matching_conns(lnet_process_id_t id, __u32 ipaddr);
595 extern ksock_conn_t *ksocknal_find_conn_locked(ksock_peer_t *peer,
596 ksock_tx_t *tx, int nonblk);
598 extern int ksocknal_launch_packet(lnet_ni_t *ni, ksock_tx_t *tx,
599 lnet_process_id_t id);
600 extern ksock_tx_t *ksocknal_alloc_tx(int type, int size);
601 extern void ksocknal_free_tx (ksock_tx_t *tx);
602 extern ksock_tx_t *ksocknal_alloc_tx_noop(__u64 cookie, int nonblk);
603 extern void ksocknal_next_tx_carrier(ksock_conn_t *conn);
604 extern void ksocknal_queue_tx_locked (ksock_tx_t *tx, ksock_conn_t *conn);
605 extern void ksocknal_txlist_done(lnet_ni_t *ni, struct list_head *txlist,
607 extern void ksocknal_notify (lnet_ni_t *ni, lnet_nid_t gw_nid, int alive);
608 extern void ksocknal_query (struct lnet_ni *ni, lnet_nid_t nid, cfs_time_t *when);
609 extern int ksocknal_thread_start(int (*fn)(void *arg), void *arg, char *name);
610 extern void ksocknal_thread_fini (void);
611 extern void ksocknal_launch_all_connections_locked (ksock_peer_t *peer);
612 extern ksock_route_t *ksocknal_find_connectable_route_locked (ksock_peer_t *peer);
613 extern ksock_route_t *ksocknal_find_connecting_route_locked (ksock_peer_t *peer);
614 extern int ksocknal_new_packet (ksock_conn_t *conn, int skip);
615 extern int ksocknal_scheduler (void *arg);
616 extern int ksocknal_connd (void *arg);
617 extern int ksocknal_reaper (void *arg);
618 extern int ksocknal_send_hello (lnet_ni_t *ni, ksock_conn_t *conn,
619 lnet_nid_t peer_nid, ksock_hello_msg_t *hello);
620 extern int ksocknal_recv_hello (lnet_ni_t *ni, ksock_conn_t *conn,
621 ksock_hello_msg_t *hello, lnet_process_id_t *id,
623 extern void ksocknal_read_callback(ksock_conn_t *conn);
624 extern void ksocknal_write_callback(ksock_conn_t *conn);
626 extern int ksocknal_lib_zc_capable(ksock_conn_t *conn);
627 extern void ksocknal_lib_save_callback(cfs_socket_t *sock, ksock_conn_t *conn);
628 extern void ksocknal_lib_set_callback(cfs_socket_t *sock, ksock_conn_t *conn);
629 extern void ksocknal_lib_reset_callback(cfs_socket_t *sock, ksock_conn_t *conn);
630 extern void ksocknal_lib_push_conn (ksock_conn_t *conn);
631 extern int ksocknal_lib_get_conn_addrs (ksock_conn_t *conn);
632 extern int ksocknal_lib_setup_sock (cfs_socket_t *so);
633 extern int ksocknal_lib_send_iov (ksock_conn_t *conn, ksock_tx_t *tx);
634 extern int ksocknal_lib_send_kiov (ksock_conn_t *conn, ksock_tx_t *tx);
635 extern void ksocknal_lib_eager_ack (ksock_conn_t *conn);
636 extern int ksocknal_lib_recv_iov (ksock_conn_t *conn);
637 extern int ksocknal_lib_recv_kiov (ksock_conn_t *conn);
638 extern int ksocknal_lib_get_conn_tunables (ksock_conn_t *conn, int *txmem,
639 int *rxmem, int *nagle);
641 extern int ksocknal_tunables_init(void);
642 extern void ksocknal_tunables_fini(void);
643 extern int ksocknal_lib_tunables_init(void);
644 extern void ksocknal_lib_tunables_fini(void);
646 extern void ksocknal_lib_csum_tx(ksock_tx_t *tx);
648 extern int ksocknal_lib_memory_pressure(ksock_conn_t *conn);
649 extern int ksocknal_lib_bind_thread_to_cpu(int id);