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 #include "socklnd_lib-linux.h"
33 #include <libcfs/libcfs.h>
34 #include <lnet/lnet.h>
35 #include <lnet/lib-lnet.h>
36 #include <lnet/socklnd.h>
38 #define SOCKNAL_PEER_HASH_SIZE 101 /* # peer lists */
39 #define SOCKNAL_RESCHED 100 /* # scheduler loops before reschedule */
40 #define SOCKNAL_INSANITY_RECONN 5000 /* connd is trying on reconn infinitely */
41 #define SOCKNAL_ENOMEM_RETRY CFS_TICK /* jiffies between retries */
43 #define SOCKNAL_SINGLE_FRAG_TX 0 /* disable multi-fragment sends */
44 #define SOCKNAL_SINGLE_FRAG_RX 0 /* disable multi-fragment receives */
46 #define SOCKNAL_VERSION_DEBUG 0 /* enable protocol version debugging */
48 /* risk kmap deadlock on multi-frag I/O (backs off to single-frag if disabled).
49 * no risk if we're not running on a CONFIG_HIGHMEM platform. */
51 # define SOCKNAL_RISK_KMAP_DEADLOCK 0
53 # define SOCKNAL_RISK_KMAP_DEADLOCK 1
56 struct ksock_sched_info;
58 typedef struct /* per scheduler state */
60 spinlock_t kss_lock; /* serialise */
61 struct list_head kss_rx_conns; /* conn waiting to be read */
62 /* conn waiting to be written */
63 struct list_head kss_tx_conns;
64 /* zombie noop tx list */
65 struct list_head kss_zombie_noop_txs;
66 wait_queue_head_t kss_waitq; /* where scheduler sleeps */
67 /* # connections assigned to this scheduler */
69 struct ksock_sched_info *kss_info; /* owner of it */
70 #if !SOCKNAL_SINGLE_FRAG_RX
71 struct page *kss_rx_scratch_pgs[LNET_MAX_IOV];
73 #if !SOCKNAL_SINGLE_FRAG_TX || !SOCKNAL_SINGLE_FRAG_RX
74 struct iovec kss_scratch_iov[LNET_MAX_IOV];
78 struct ksock_sched_info {
79 int ksi_nthreads_max; /* max allowed threads */
80 int ksi_nthreads; /* number of threads */
81 int ksi_cpt; /* CPT id */
82 ksock_sched_t *ksi_scheds; /* array of schedulers */
85 #define KSOCK_CPT_SHIFT 16
86 #define KSOCK_THREAD_ID(cpt, sid) (((cpt) << KSOCK_CPT_SHIFT) | (sid))
87 #define KSOCK_THREAD_CPT(id) ((id) >> KSOCK_CPT_SHIFT)
88 #define KSOCK_THREAD_SID(id) ((id) & ((1UL << KSOCK_CPT_SHIFT) - 1))
90 typedef struct /* in-use interface */
92 __u32 ksni_ipaddr; /* interface's IP address */
93 __u32 ksni_netmask; /* interface's network mask */
94 int ksni_nroutes; /* # routes using (active) */
95 int ksni_npeers; /* # peers using (passive) */
96 char ksni_name[IFNAMSIZ]; /* interface name */
101 /* "stuck" socket timeout (seconds) */
103 /* # scheduler threads in each pool while starting */
105 int *ksnd_nconnds; /* # connection daemons */
106 int *ksnd_nconnds_max; /* max # connection daemons */
107 int *ksnd_min_reconnectms; /* first connection retry after (ms)... */
108 int *ksnd_max_reconnectms; /* ...exponentially increasing to this */
109 int *ksnd_eager_ack; /* make TCP ack eagerly? */
110 int *ksnd_typed_conns; /* drive sockets by type? */
111 int *ksnd_min_bulk; /* smallest "large" message */
112 int *ksnd_tx_buffer_size; /* socket tx buffer size */
113 int *ksnd_rx_buffer_size; /* socket rx buffer size */
114 int *ksnd_nagle; /* enable NAGLE? */
115 int *ksnd_round_robin; /* round robin for multiple interfaces */
116 int *ksnd_keepalive; /* # secs for sending keepalive NOOP */
117 int *ksnd_keepalive_idle; /* # idle secs before 1st probe */
118 int *ksnd_keepalive_count; /* # probes */
119 int *ksnd_keepalive_intvl; /* time between probes */
120 int *ksnd_credits; /* # concurrent sends */
121 int *ksnd_peertxcredits; /* # concurrent sends to 1 peer */
122 int *ksnd_peerrtrcredits; /* # per-peer router buffer credits */
123 int *ksnd_peertimeout; /* seconds to consider peer dead */
124 int *ksnd_enable_csum; /* enable check sum */
125 int *ksnd_inject_csum_error; /* set non-zero to inject checksum error */
126 int *ksnd_nonblk_zcack; /* always send zc-ack on non-blocking connection */
127 unsigned int *ksnd_zc_min_payload; /* minimum zero copy payload size */
128 int *ksnd_zc_recv; /* enable ZC receive (for Chelsio TOE) */
129 int *ksnd_zc_recv_min_nfrags; /* minimum # of fragments to enable ZC receive */
131 int *ksnd_irq_affinity; /* enable IRQ affinity? */
133 #ifdef SOCKNAL_BACKOFF
134 int *ksnd_backoff_init; /* initial TCP backoff */
135 int *ksnd_backoff_max; /* maximum TCP backoff */
137 #if SOCKNAL_VERSION_DEBUG
138 int *ksnd_protocol; /* protocol version */
140 #if defined(CONFIG_SYSCTL) && !CFS_SYSFS_MODULE_PARM
141 struct ctl_table_header *ksnd_sysctl; /* sysctl interface */
147 __u64 ksnn_incarnation; /* my epoch */
148 spinlock_t ksnn_lock; /* serialise */
149 struct list_head ksnn_list; /* chain on global list */
150 int ksnn_npeers; /* # peers */
151 int ksnn_shutdown; /* shutting down? */
152 int ksnn_ninterfaces; /* IP interfaces */
153 ksock_interface_t ksnn_interfaces[LNET_MAX_INTERFACES];
157 #define SOCKNAL_CONND_TIMEOUT 120
158 /** reserved thread for accepting & creating new connd */
159 #define SOCKNAL_CONND_RESV 1
163 int ksnd_init; /* initialisation state */
164 int ksnd_nnets; /* # networks set up */
165 struct list_head ksnd_nets; /* list of nets */
166 /* stabilize peer/conn ops */
167 rwlock_t ksnd_global_lock;
168 /* hash table of all my known peers */
169 struct list_head *ksnd_peers;
170 int ksnd_peer_hash_size; /* size of ksnd_peers */
172 int ksnd_nthreads; /* # live threads */
173 int ksnd_shuttingdown; /* tell threads to exit */
174 /* schedulers information */
175 struct ksock_sched_info **ksnd_sched_info;
177 atomic_t ksnd_nactive_txs; /* #active txs */
179 /* conns to close: reaper_lock*/
180 struct list_head ksnd_deathrow_conns;
181 /* conns to free: reaper_lock */
182 struct list_head ksnd_zombie_conns;
183 /* conns to retry: reaper_lock*/
184 struct list_head ksnd_enomem_conns;
185 /* reaper sleeps here */
186 wait_queue_head_t ksnd_reaper_waitq;
187 /* when reaper will wake */
188 cfs_time_t ksnd_reaper_waketime;
190 spinlock_t ksnd_reaper_lock;
192 int ksnd_enomem_tx; /* test ENOMEM sender */
193 int ksnd_stall_tx; /* test sluggish sender */
194 int ksnd_stall_rx; /* test sluggish receiver */
196 /* incoming connection requests */
197 struct list_head ksnd_connd_connreqs;
198 /* routes waiting to be connected */
199 struct list_head ksnd_connd_routes;
200 /* connds sleep here */
201 wait_queue_head_t ksnd_connd_waitq;
202 /* # connds connecting */
203 int ksnd_connd_connecting;
204 /** time stamp of the last failed connecting attempt */
205 long ksnd_connd_failed_stamp;
206 /** # starting connd */
207 unsigned ksnd_connd_starting;
208 /** time stamp of the last starting connd */
209 long ksnd_connd_starting_stamp;
210 /** # running connd */
211 unsigned ksnd_connd_running;
213 spinlock_t ksnd_connd_lock;
215 /* list head for freed noop tx */
216 struct list_head ksnd_idle_noop_txs;
217 /* serialise, g_lock unsafe */
218 spinlock_t ksnd_tx_lock;
222 #define SOCKNAL_INIT_NOTHING 0
223 #define SOCKNAL_INIT_DATA 1
224 #define SOCKNAL_INIT_ALL 2
226 /* A packet just assembled for transmission is represented by 1 or more
227 * struct iovec fragments (the first frag contains the portals header),
228 * followed by 0 or more lnet_kiov_t fragments.
230 * On the receive side, initially 1 struct iovec fragment is posted for
231 * receive (the header). Once the header has been received, the payload is
232 * received into either struct iovec or lnet_kiov_t fragments, depending on
233 * what the header matched or whether the message needs forwarding. */
235 struct ksock_conn; /* forward ref */
236 struct ksock_peer; /* forward ref */
237 struct ksock_route; /* forward ref */
238 struct ksock_proto; /* forward ref */
240 typedef struct /* transmit packet */
242 struct list_head tx_list; /* queue on conn for transmission etc */
243 struct list_head tx_zc_list; /* queue on peer for ZC request */
244 atomic_t tx_refcount; /* tx reference count */
245 int tx_nob; /* # packet bytes */
246 int tx_resid; /* residual bytes */
247 int tx_niov; /* # packet iovec frags */
248 struct iovec *tx_iov; /* packet iovec frags */
249 int tx_nkiov; /* # packet page frags */
250 unsigned short tx_zc_aborted; /* aborted ZC request */
251 unsigned short tx_zc_capable:1; /* payload is large enough for ZC */
252 unsigned short tx_zc_checked:1; /* Have I checked if I should ZC? */
253 unsigned short tx_nonblk:1; /* it's a non-blocking ACK */
254 lnet_kiov_t *tx_kiov; /* packet page frags */
255 struct ksock_conn *tx_conn; /* owning conn */
256 lnet_msg_t *tx_lnetmsg; /* lnet message for lnet_finalize() */
257 cfs_time_t tx_deadline; /* when (in jiffies) tx times out */
258 ksock_msg_t tx_msg; /* socklnd message buffer */
259 int tx_desc_size; /* size of this descriptor */
262 struct iovec iov; /* virt hdr */
263 lnet_kiov_t kiov[0]; /* paged payload */
266 struct iovec iov[1]; /* virt hdr + payload */
271 #define KSOCK_NOOP_TX_SIZE ((int)offsetof(ksock_tx_t, tx_frags.paged.kiov[0]))
273 /* network zero copy callback descriptor embedded in ksock_tx_t */
275 /* space for the rx frag descriptors; we either read a single contiguous
276 * header, or up to LNET_MAX_IOV frags of payload of either type. */
278 struct iovec iov[LNET_MAX_IOV];
279 lnet_kiov_t kiov[LNET_MAX_IOV];
280 } ksock_rxiovspace_t;
282 #define SOCKNAL_RX_KSM_HEADER 1 /* reading ksock message header */
283 #define SOCKNAL_RX_LNET_HEADER 2 /* reading lnet message header */
284 #define SOCKNAL_RX_PARSE 3 /* Calling lnet_parse() */
285 #define SOCKNAL_RX_PARSE_WAIT 4 /* waiting to be told to read the body */
286 #define SOCKNAL_RX_LNET_PAYLOAD 5 /* reading lnet payload (to deliver here) */
287 #define SOCKNAL_RX_SLOP 6 /* skipping body */
289 typedef struct ksock_conn
291 struct ksock_peer *ksnc_peer; /* owning peer */
292 struct ksock_route *ksnc_route; /* owning route */
293 struct list_head ksnc_list; /* stash on peer's conn list */
294 cfs_socket_t *ksnc_sock; /* actual socket */
295 void *ksnc_saved_data_ready; /* socket's original data_ready() callback */
296 void *ksnc_saved_write_space; /* socket's original write_space() callback */
297 atomic_t ksnc_conn_refcount; /* conn refcount */
298 atomic_t ksnc_sock_refcount; /* sock refcount */
299 ksock_sched_t *ksnc_scheduler; /* who schedules this connection */
300 __u32 ksnc_myipaddr; /* my IP */
301 __u32 ksnc_ipaddr; /* peer's IP */
302 int ksnc_port; /* peer's port */
303 signed int ksnc_type:3; /* type of connection,
304 * should be signed value */
305 unsigned int ksnc_closing:1; /* being shut down */
306 unsigned int ksnc_flip:1; /* flip or not, only for V2.x */
307 unsigned int ksnc_zc_capable:1; /* enable to ZC */
308 struct ksock_proto *ksnc_proto; /* protocol for the connection */
312 /* where I enq waiting input or a forwarding descriptor */
313 struct list_head ksnc_rx_list;
314 cfs_time_t ksnc_rx_deadline; /* when (in jiffies) receive times out */
315 __u8 ksnc_rx_started; /* started receiving a message */
316 __u8 ksnc_rx_ready; /* data ready to read */
317 __u8 ksnc_rx_scheduled;/* being progressed */
318 __u8 ksnc_rx_state; /* what is being read */
319 int ksnc_rx_nob_left; /* # bytes to next hdr/body */
320 int ksnc_rx_nob_wanted; /* bytes actually wanted */
321 int ksnc_rx_niov; /* # iovec frags */
322 struct iovec *ksnc_rx_iov; /* the iovec frags */
323 int ksnc_rx_nkiov; /* # page frags */
324 lnet_kiov_t *ksnc_rx_kiov; /* the page frags */
325 ksock_rxiovspace_t ksnc_rx_iov_space;/* space for frag descriptors */
326 __u32 ksnc_rx_csum; /* partial checksum for incoming data */
327 void *ksnc_cookie; /* rx lnet_finalize passthru arg */
328 ksock_msg_t ksnc_msg; /* incoming message buffer:
329 * V2.x message takes the
331 * V1.x message is a bare
332 * lnet_hdr_t, it's stored in
333 * ksnc_msg.ksm_u.lnetmsg */
336 /* where I enq waiting for output space */
337 struct list_head ksnc_tx_list;
338 /* packets waiting to be sent */
339 struct list_head ksnc_tx_queue;
340 /* next TX that can carry a LNet message or ZC-ACK */
341 ksock_tx_t *ksnc_tx_carrier;
342 /* when (in jiffies) tx times out */
343 cfs_time_t ksnc_tx_deadline;
344 /* send buffer marker */
347 atomic_t ksnc_tx_nob;
350 /* being progressed */
351 int ksnc_tx_scheduled;
352 /* time stamp of the last posted TX */
353 cfs_time_t ksnc_tx_last_post;
356 typedef struct ksock_route
358 struct list_head ksnr_list; /* chain on peer route list */
359 struct list_head ksnr_connd_list; /* chain on ksnr_connd_routes */
360 struct ksock_peer *ksnr_peer; /* owning peer */
361 atomic_t ksnr_refcount; /* # users */
362 cfs_time_t ksnr_timeout; /* when (in jiffies) reconnection can happen next */
363 cfs_duration_t ksnr_retry_interval; /* how long between retries */
364 __u32 ksnr_myipaddr; /* my IP */
365 __u32 ksnr_ipaddr; /* IP address to connect to */
366 int ksnr_port; /* port to connect to */
367 unsigned int ksnr_scheduled:1; /* scheduled for attention */
368 unsigned int ksnr_connecting:1;/* connection establishment in progress */
369 unsigned int ksnr_connected:4; /* connections established by type */
370 unsigned int ksnr_deleted:1; /* been removed from peer? */
371 unsigned int ksnr_share_count; /* created explicitly? */
372 int ksnr_conn_count; /* # conns established by this route */
375 #define SOCKNAL_KEEPALIVE_PING 1 /* cookie for keepalive ping */
377 typedef struct ksock_peer
379 struct list_head ksnp_list; /* stash on global peer list */
380 cfs_time_t ksnp_last_alive; /* when (in jiffies) I was last alive */
381 lnet_process_id_t ksnp_id; /* who's on the other end(s) */
382 atomic_t ksnp_refcount; /* # users */
383 int ksnp_sharecount; /* lconf usage counter */
384 int ksnp_closing; /* being closed */
385 int ksnp_accepting;/* # passive connections pending */
386 int ksnp_error; /* errno on closing last conn */
387 __u64 ksnp_zc_next_cookie;/* ZC completion cookie */
388 __u64 ksnp_incarnation; /* latest known peer incarnation */
389 struct ksock_proto *ksnp_proto; /* latest known peer protocol */
390 struct list_head ksnp_conns; /* all active connections */
391 struct list_head ksnp_routes; /* routes */
392 struct list_head ksnp_tx_queue; /* waiting packets */
393 spinlock_t ksnp_lock; /* serialize, g_lock unsafe */
394 /* zero copy requests wait for ACK */
395 struct list_head ksnp_zc_req_list;
396 cfs_time_t ksnp_send_keepalive; /* time to send keepalive */
397 lnet_ni_t *ksnp_ni; /* which network */
398 int ksnp_n_passive_ips; /* # of... */
399 __u32 ksnp_passive_ips[LNET_MAX_INTERFACES]; /* preferred local interfaces */
402 typedef struct ksock_connreq
404 /* stash on ksnd_connd_connreqs */
405 struct list_head ksncr_list;
408 /* accepted socket */
409 cfs_socket_t *ksncr_sock;
412 extern ksock_nal_data_t ksocknal_data;
413 extern ksock_tunables_t ksocknal_tunables;
415 #define SOCKNAL_MATCH_NO 0 /* TX can't match type of connection */
416 #define SOCKNAL_MATCH_YES 1 /* TX matches type of connection */
417 #define SOCKNAL_MATCH_MAY 2 /* TX can be sent on the connection, but not preferred */
419 typedef struct ksock_proto
421 int pro_version; /* version number of protocol */
422 int (*pro_send_hello)(ksock_conn_t *, ksock_hello_msg_t *); /* handshake function */
423 int (*pro_recv_hello)(ksock_conn_t *, ksock_hello_msg_t *, int);/* handshake function */
424 void (*pro_pack)(ksock_tx_t *); /* message pack */
425 void (*pro_unpack)(ksock_msg_t *); /* message unpack */
426 ksock_tx_t *(*pro_queue_tx_msg)(ksock_conn_t *, ksock_tx_t *); /* queue tx on the connection */
427 int (*pro_queue_tx_zcack)(ksock_conn_t *, ksock_tx_t *, __u64); /* queue ZC ack on the connection */
428 int (*pro_handle_zcreq)(ksock_conn_t *, __u64, int); /* handle ZC request */
429 int (*pro_handle_zcack)(ksock_conn_t *, __u64, __u64); /* handle ZC ACK */
430 int (*pro_match_tx)(ksock_conn_t *, ksock_tx_t *, int); /* msg type matches the connection type:
432 * return MATCH_NO : no
433 * return MATCH_YES : matching type
434 * return MATCH_MAY : can be backup */
437 extern ksock_proto_t ksocknal_protocol_v1x;
438 extern ksock_proto_t ksocknal_protocol_v2x;
439 extern ksock_proto_t ksocknal_protocol_v3x;
441 #define KSOCK_PROTO_V1_MAJOR LNET_PROTO_TCP_VERSION_MAJOR
442 #define KSOCK_PROTO_V1_MINOR LNET_PROTO_TCP_VERSION_MINOR
443 #define KSOCK_PROTO_V1 KSOCK_PROTO_V1_MAJOR
445 #ifndef CPU_MASK_NONE
446 #define CPU_MASK_NONE 0UL
450 ksocknal_route_mask(void)
452 if (!*ksocknal_tunables.ksnd_typed_conns)
453 return (1 << SOCKLND_CONN_ANY);
455 return ((1 << SOCKLND_CONN_CONTROL) |
456 (1 << SOCKLND_CONN_BULK_IN) |
457 (1 << SOCKLND_CONN_BULK_OUT));
460 static inline struct list_head *
461 ksocknal_nid2peerlist (lnet_nid_t nid)
463 unsigned int hash = ((unsigned int)nid) % ksocknal_data.ksnd_peer_hash_size;
465 return (&ksocknal_data.ksnd_peers [hash]);
469 ksocknal_conn_addref (ksock_conn_t *conn)
471 LASSERT (atomic_read(&conn->ksnc_conn_refcount) > 0);
472 atomic_inc(&conn->ksnc_conn_refcount);
475 extern void ksocknal_queue_zombie_conn (ksock_conn_t *conn);
476 extern void ksocknal_finalize_zcreq(ksock_conn_t *conn);
479 ksocknal_conn_decref (ksock_conn_t *conn)
481 LASSERT (atomic_read(&conn->ksnc_conn_refcount) > 0);
482 if (atomic_dec_and_test(&conn->ksnc_conn_refcount))
483 ksocknal_queue_zombie_conn(conn);
487 ksocknal_connsock_addref (ksock_conn_t *conn)
491 read_lock(&ksocknal_data.ksnd_global_lock);
492 if (!conn->ksnc_closing) {
493 LASSERT(atomic_read(&conn->ksnc_sock_refcount) > 0);
494 atomic_inc(&conn->ksnc_sock_refcount);
497 read_unlock(&ksocknal_data.ksnd_global_lock);
503 ksocknal_connsock_decref (ksock_conn_t *conn)
505 LASSERT (atomic_read(&conn->ksnc_sock_refcount) > 0);
506 if (atomic_dec_and_test(&conn->ksnc_sock_refcount)) {
507 LASSERT (conn->ksnc_closing);
508 libcfs_sock_release(conn->ksnc_sock);
509 conn->ksnc_sock = NULL;
510 ksocknal_finalize_zcreq(conn);
515 ksocknal_tx_addref (ksock_tx_t *tx)
517 LASSERT (atomic_read(&tx->tx_refcount) > 0);
518 atomic_inc(&tx->tx_refcount);
521 extern void ksocknal_tx_prep (ksock_conn_t *, ksock_tx_t *tx);
522 extern void ksocknal_tx_done (lnet_ni_t *ni, ksock_tx_t *tx);
525 ksocknal_tx_decref (ksock_tx_t *tx)
527 LASSERT (atomic_read(&tx->tx_refcount) > 0);
528 if (atomic_dec_and_test(&tx->tx_refcount))
529 ksocknal_tx_done(NULL, tx);
533 ksocknal_route_addref (ksock_route_t *route)
535 LASSERT (atomic_read(&route->ksnr_refcount) > 0);
536 atomic_inc(&route->ksnr_refcount);
539 extern void ksocknal_destroy_route (ksock_route_t *route);
542 ksocknal_route_decref (ksock_route_t *route)
544 LASSERT (atomic_read (&route->ksnr_refcount) > 0);
545 if (atomic_dec_and_test(&route->ksnr_refcount))
546 ksocknal_destroy_route (route);
550 ksocknal_peer_addref (ksock_peer_t *peer)
552 LASSERT (atomic_read (&peer->ksnp_refcount) > 0);
553 atomic_inc(&peer->ksnp_refcount);
556 extern void ksocknal_destroy_peer (ksock_peer_t *peer);
559 ksocknal_peer_decref (ksock_peer_t *peer)
561 LASSERT (atomic_read (&peer->ksnp_refcount) > 0);
562 if (atomic_dec_and_test(&peer->ksnp_refcount))
563 ksocknal_destroy_peer (peer);
566 int ksocknal_startup (lnet_ni_t *ni);
567 void ksocknal_shutdown (lnet_ni_t *ni);
568 int ksocknal_ctl(lnet_ni_t *ni, unsigned int cmd, void *arg);
569 int ksocknal_send (lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg);
570 int ksocknal_recv(lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg,
571 int delayed, unsigned int niov,
572 struct iovec *iov, lnet_kiov_t *kiov,
573 unsigned int offset, unsigned int mlen, unsigned int rlen);
574 int ksocknal_accept(lnet_ni_t *ni, cfs_socket_t *sock);
576 extern int ksocknal_add_peer(lnet_ni_t *ni, lnet_process_id_t id, __u32 ip, int port);
577 extern ksock_peer_t *ksocknal_find_peer_locked (lnet_ni_t *ni, lnet_process_id_t id);
578 extern ksock_peer_t *ksocknal_find_peer (lnet_ni_t *ni, lnet_process_id_t id);
579 extern void ksocknal_peer_failed (ksock_peer_t *peer);
580 extern int ksocknal_create_conn (lnet_ni_t *ni, ksock_route_t *route,
581 cfs_socket_t *sock, int type);
582 extern void ksocknal_close_conn_locked (ksock_conn_t *conn, int why);
583 extern void ksocknal_terminate_conn (ksock_conn_t *conn);
584 extern void ksocknal_destroy_conn (ksock_conn_t *conn);
585 extern int ksocknal_close_peer_conns_locked (ksock_peer_t *peer,
586 __u32 ipaddr, int why);
587 extern int ksocknal_close_conn_and_siblings (ksock_conn_t *conn, int why);
588 extern int ksocknal_close_matching_conns(lnet_process_id_t id, __u32 ipaddr);
589 extern ksock_conn_t *ksocknal_find_conn_locked(ksock_peer_t *peer,
590 ksock_tx_t *tx, int nonblk);
592 extern int ksocknal_launch_packet(lnet_ni_t *ni, ksock_tx_t *tx,
593 lnet_process_id_t id);
594 extern ksock_tx_t *ksocknal_alloc_tx(int type, int size);
595 extern void ksocknal_free_tx (ksock_tx_t *tx);
596 extern ksock_tx_t *ksocknal_alloc_tx_noop(__u64 cookie, int nonblk);
597 extern void ksocknal_next_tx_carrier(ksock_conn_t *conn);
598 extern void ksocknal_queue_tx_locked (ksock_tx_t *tx, ksock_conn_t *conn);
599 extern void ksocknal_txlist_done(lnet_ni_t *ni, struct list_head *txlist,
601 extern void ksocknal_notify (lnet_ni_t *ni, lnet_nid_t gw_nid, int alive);
602 extern void ksocknal_query (struct lnet_ni *ni, lnet_nid_t nid, cfs_time_t *when);
603 extern int ksocknal_thread_start(int (*fn)(void *arg), void *arg, char *name);
604 extern void ksocknal_thread_fini (void);
605 extern void ksocknal_launch_all_connections_locked (ksock_peer_t *peer);
606 extern ksock_route_t *ksocknal_find_connectable_route_locked (ksock_peer_t *peer);
607 extern ksock_route_t *ksocknal_find_connecting_route_locked (ksock_peer_t *peer);
608 extern int ksocknal_new_packet (ksock_conn_t *conn, int skip);
609 extern int ksocknal_scheduler (void *arg);
610 extern int ksocknal_connd (void *arg);
611 extern int ksocknal_reaper (void *arg);
612 extern int ksocknal_send_hello (lnet_ni_t *ni, ksock_conn_t *conn,
613 lnet_nid_t peer_nid, ksock_hello_msg_t *hello);
614 extern int ksocknal_recv_hello (lnet_ni_t *ni, ksock_conn_t *conn,
615 ksock_hello_msg_t *hello, lnet_process_id_t *id,
617 extern void ksocknal_read_callback(ksock_conn_t *conn);
618 extern void ksocknal_write_callback(ksock_conn_t *conn);
620 extern int ksocknal_lib_zc_capable(ksock_conn_t *conn);
621 extern void ksocknal_lib_save_callback(cfs_socket_t *sock, ksock_conn_t *conn);
622 extern void ksocknal_lib_set_callback(cfs_socket_t *sock, ksock_conn_t *conn);
623 extern void ksocknal_lib_reset_callback(cfs_socket_t *sock, ksock_conn_t *conn);
624 extern void ksocknal_lib_push_conn (ksock_conn_t *conn);
625 extern int ksocknal_lib_get_conn_addrs (ksock_conn_t *conn);
626 extern int ksocknal_lib_setup_sock (cfs_socket_t *so);
627 extern int ksocknal_lib_send_iov (ksock_conn_t *conn, ksock_tx_t *tx);
628 extern int ksocknal_lib_send_kiov (ksock_conn_t *conn, ksock_tx_t *tx);
629 extern void ksocknal_lib_eager_ack (ksock_conn_t *conn);
630 extern int ksocknal_lib_recv_iov (ksock_conn_t *conn);
631 extern int ksocknal_lib_recv_kiov (ksock_conn_t *conn);
632 extern int ksocknal_lib_get_conn_tunables (ksock_conn_t *conn, int *txmem,
633 int *rxmem, int *nagle);
635 extern int ksocknal_tunables_init(void);
636 extern void ksocknal_tunables_fini(void);
637 extern int ksocknal_lib_tunables_init(void);
638 extern void ksocknal_lib_tunables_fini(void);
640 extern void ksocknal_lib_csum_tx(ksock_tx_t *tx);
642 extern int ksocknal_lib_memory_pressure(ksock_conn_t *conn);
643 extern int ksocknal_lib_bind_thread_to_cpu(int id);