1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
6 * Copyright (c) 2011, Whamcloud, Inc.
8 * Author: Zach Brown <zab@zabbo.net>
9 * Author: Peter J. Braam <braam@clusterfs.com>
10 * Author: Phil Schwan <phil@clusterfs.com>
11 * Author: Eric Barton <eric@bartonsoftware.com>
13 * This file is part of Lustre, http://www.lustre.org
15 * Portals is free software; you can redistribute it and/or
16 * modify it under the terms of version 2 of the GNU General Public
17 * License as published by the Free Software Foundation.
19 * Portals is distributed in the hope that it will be useful,
20 * but WITHOUT ANY WARRANTY; without even the implied warranty of
21 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
22 * GNU General Public License for more details.
24 * You should have received a copy of the GNU General Public License
25 * along with Portals; if not, write to the Free Software
26 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
30 #define DEBUG_PORTAL_ALLOC
32 # define EXPORT_SYMTAB
35 #define DEBUG_SUBSYSTEM S_LND
37 #if defined(__linux__)
38 #include "socklnd_lib-linux.h"
39 #elif defined(__APPLE__)
40 #include "socklnd_lib-darwin.h"
41 #elif defined(__WINNT__)
42 #include "socklnd_lib-winnt.h"
44 #error Unsupported Operating System
47 #include <libcfs/libcfs.h>
48 #include <lnet/lnet.h>
49 #include <lnet/lib-lnet.h>
50 #include <lnet/socklnd.h>
51 #include <lnet/lnet-sysctl.h>
53 #define SOCKNAL_PEER_HASH_SIZE 101 /* # peer lists */
54 #define SOCKNAL_RESCHED 100 /* # scheduler loops before reschedule */
55 #define SOCKNAL_INSANITY_RECONN 5000 /* connd is trying on reconn infinitely */
56 #define SOCKNAL_ENOMEM_RETRY CFS_TICK /* jiffies between retries */
58 #define SOCKNAL_SINGLE_FRAG_TX 0 /* disable multi-fragment sends */
59 #define SOCKNAL_SINGLE_FRAG_RX 0 /* disable multi-fragment receives */
61 #define SOCKNAL_VERSION_DEBUG 0 /* enable protocol version debugging */
63 /* risk kmap deadlock on multi-frag I/O (backs off to single-frag if disabled).
64 * no risk if we're not running on a CONFIG_HIGHMEM platform. */
66 # define SOCKNAL_RISK_KMAP_DEADLOCK 0
68 # define SOCKNAL_RISK_KMAP_DEADLOCK 1
71 typedef struct /* per scheduler state */
73 cfs_spinlock_t kss_lock; /* serialise */
74 cfs_list_t kss_rx_conns; /* conn waiting to be read */
75 cfs_list_t kss_tx_conns; /* conn waiting to be written */
76 cfs_list_t kss_zombie_noop_txs; /* zombie noop tx list */
77 cfs_waitq_t kss_waitq; /* where scheduler sleeps */
78 int kss_nconns; /* # connections assigned to this scheduler */
79 #if !SOCKNAL_SINGLE_FRAG_RX
80 struct page *kss_rx_scratch_pgs[LNET_MAX_IOV];
82 #if !SOCKNAL_SINGLE_FRAG_TX || !SOCKNAL_SINGLE_FRAG_RX
83 struct iovec kss_scratch_iov[LNET_MAX_IOV];
89 unsigned int ksni_valid:1; /* been set yet? */
90 unsigned int ksni_bound:1; /* bound to a cpu yet? */
91 unsigned int ksni_sched:6; /* which scheduler (assumes < 64) */
94 typedef struct /* in-use interface */
96 __u32 ksni_ipaddr; /* interface's IP address */
97 __u32 ksni_netmask; /* interface's network mask */
98 int ksni_nroutes; /* # routes using (active) */
99 int ksni_npeers; /* # peers using (passive) */
100 char ksni_name[16]; /* interface name */
105 int *ksnd_timeout; /* "stuck" socket timeout (seconds) */
106 int *ksnd_nconnds; /* # connection daemons */
107 int *ksnd_nconnds_max; /* max # connection daemons */
108 int *ksnd_min_reconnectms; /* first connection retry after (ms)... */
109 int *ksnd_max_reconnectms; /* ...exponentially increasing to this */
110 int *ksnd_eager_ack; /* make TCP ack eagerly? */
111 int *ksnd_typed_conns; /* drive sockets by type? */
112 int *ksnd_min_bulk; /* smallest "large" message */
113 int *ksnd_tx_buffer_size; /* socket tx buffer size */
114 int *ksnd_rx_buffer_size; /* socket rx buffer size */
115 int *ksnd_nagle; /* enable NAGLE? */
116 int *ksnd_round_robin; /* round robin for multiple interfaces */
117 int *ksnd_keepalive; /* # secs for sending keepalive NOOP */
118 int *ksnd_keepalive_idle; /* # idle secs before 1st probe */
119 int *ksnd_keepalive_count; /* # probes */
120 int *ksnd_keepalive_intvl; /* time between probes */
121 int *ksnd_credits; /* # concurrent sends */
122 int *ksnd_peertxcredits; /* # concurrent sends to 1 peer */
123 int *ksnd_peerrtrcredits; /* # per-peer router buffer credits */
124 int *ksnd_peertimeout; /* seconds to consider peer dead */
125 int *ksnd_enable_csum; /* enable check sum */
126 int *ksnd_inject_csum_error; /* set non-zero to inject checksum error */
127 int *ksnd_nonblk_zcack; /* always send zc-ack on non-blocking connection */
128 unsigned int *ksnd_zc_min_payload; /* minimum zero copy payload size */
129 int *ksnd_zc_recv; /* enable ZC receive (for Chelsio TOE) */
130 int *ksnd_zc_recv_min_nfrags; /* minimum # of fragments to enable ZC receive */
132 int *ksnd_irq_affinity; /* enable IRQ affinity? */
134 #ifdef SOCKNAL_BACKOFF
135 int *ksnd_backoff_init; /* initial TCP backoff */
136 int *ksnd_backoff_max; /* maximum TCP backoff */
138 #if SOCKNAL_VERSION_DEBUG
139 int *ksnd_protocol; /* protocol version */
141 #if defined(CONFIG_SYSCTL) && !CFS_SYSFS_MODULE_PARM
142 cfs_sysctl_table_header_t *ksnd_sysctl; /* sysctl interface */
148 __u64 ksnn_incarnation; /* my epoch */
149 cfs_spinlock_t ksnn_lock; /* serialise */
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 */
166 cfs_rwlock_t ksnd_global_lock; /* stabilize peer/conn ops */
167 cfs_list_t *ksnd_peers; /* hash table of all my known peers */
168 int ksnd_peer_hash_size; /* size of ksnd_peers */
170 int ksnd_nthreads; /* # live threads */
171 int ksnd_shuttingdown; /* tell threads to exit */
172 int ksnd_nschedulers; /* # schedulers */
173 ksock_sched_t *ksnd_schedulers; /* their state */
175 cfs_atomic_t ksnd_nactive_txs; /* #active txs */
177 cfs_list_t ksnd_deathrow_conns; /* conns to close: reaper_lock*/
178 cfs_list_t ksnd_zombie_conns; /* conns to free: reaper_lock */
179 cfs_list_t ksnd_enomem_conns; /* conns to retry: reaper_lock*/
180 cfs_waitq_t ksnd_reaper_waitq; /* reaper sleeps here */
181 cfs_time_t ksnd_reaper_waketime;/* when reaper will wake */
182 cfs_spinlock_t ksnd_reaper_lock; /* serialise */
184 int ksnd_enomem_tx; /* test ENOMEM sender */
185 int ksnd_stall_tx; /* test sluggish sender */
186 int ksnd_stall_rx; /* test sluggish receiver */
188 cfs_list_t ksnd_connd_connreqs; /* incoming connection requests */
189 cfs_list_t ksnd_connd_routes; /* routes waiting to be connected */
190 cfs_waitq_t ksnd_connd_waitq; /* connds sleep here */
191 int ksnd_connd_connecting;/* # connds connecting */
192 /** time stamp of the last failed connecting attempt */
193 long ksnd_connd_failed_stamp;
194 /** # starting connd */
195 unsigned ksnd_connd_starting;
196 /** time stamp of the last starting connd */
197 long ksnd_connd_starting_stamp;
198 /** # running connd */
199 unsigned ksnd_connd_running;
200 cfs_spinlock_t ksnd_connd_lock; /* serialise */
202 cfs_list_t ksnd_idle_noop_txs; /* list head for freed noop tx */
203 cfs_spinlock_t ksnd_tx_lock; /* serialise, NOT safe in g_lock */
205 ksock_irqinfo_t ksnd_irqinfo[CFS_NR_IRQS];/* irq->scheduler lookup */
209 #define SOCKNAL_INIT_NOTHING 0
210 #define SOCKNAL_INIT_DATA 1
211 #define SOCKNAL_INIT_ALL 2
213 /* A packet just assembled for transmission is represented by 1 or more
214 * struct iovec fragments (the first frag contains the portals header),
215 * followed by 0 or more lnet_kiov_t fragments.
217 * On the receive side, initially 1 struct iovec fragment is posted for
218 * receive (the header). Once the header has been received, the payload is
219 * received into either struct iovec or lnet_kiov_t fragments, depending on
220 * what the header matched or whether the message needs forwarding. */
222 struct ksock_conn; /* forward ref */
223 struct ksock_peer; /* forward ref */
224 struct ksock_route; /* forward ref */
225 struct ksock_proto; /* forward ref */
227 typedef struct /* transmit packet */
229 cfs_list_t tx_list; /* queue on conn for transmission etc */
230 cfs_list_t tx_zc_list; /* queue on peer for ZC request */
231 cfs_atomic_t tx_refcount; /* tx reference count */
232 int tx_nob; /* # packet bytes */
233 int tx_resid; /* residual bytes */
234 int tx_niov; /* # packet iovec frags */
235 struct iovec *tx_iov; /* packet iovec frags */
236 int tx_nkiov; /* # packet page frags */
237 unsigned short tx_zc_aborted; /* aborted ZC request */
238 unsigned short tx_zc_capable:1; /* payload is large enough for ZC */
239 unsigned short tx_zc_checked:1; /* Have I checked if I should ZC? */
240 unsigned short tx_nonblk:1; /* it's a non-blocking ACK */
241 lnet_kiov_t *tx_kiov; /* packet page frags */
242 struct ksock_conn *tx_conn; /* owning conn */
243 lnet_msg_t *tx_lnetmsg; /* lnet message for lnet_finalize() */
244 cfs_time_t tx_deadline; /* when (in jiffies) tx times out */
245 ksock_msg_t tx_msg; /* socklnd message buffer */
246 int tx_desc_size; /* size of this descriptor */
249 struct iovec iov; /* virt hdr */
250 lnet_kiov_t kiov[0]; /* paged payload */
253 struct iovec iov[1]; /* virt hdr + payload */
258 #define KSOCK_NOOP_TX_SIZE ((int)offsetof(ksock_tx_t, tx_frags.paged.kiov[0]))
260 /* network zero copy callback descriptor embedded in ksock_tx_t */
262 /* space for the rx frag descriptors; we either read a single contiguous
263 * header, or up to LNET_MAX_IOV frags of payload of either type. */
265 struct iovec iov[LNET_MAX_IOV];
266 lnet_kiov_t kiov[LNET_MAX_IOV];
267 } ksock_rxiovspace_t;
269 #define SOCKNAL_RX_KSM_HEADER 1 /* reading ksock message header */
270 #define SOCKNAL_RX_LNET_HEADER 2 /* reading lnet message header */
271 #define SOCKNAL_RX_PARSE 3 /* Calling lnet_parse() */
272 #define SOCKNAL_RX_PARSE_WAIT 4 /* waiting to be told to read the body */
273 #define SOCKNAL_RX_LNET_PAYLOAD 5 /* reading lnet payload (to deliver here) */
274 #define SOCKNAL_RX_SLOP 6 /* skipping body */
276 typedef struct ksock_conn
278 struct ksock_peer *ksnc_peer; /* owning peer */
279 struct ksock_route *ksnc_route; /* owning route */
280 cfs_list_t ksnc_list; /* stash on peer's conn list */
281 cfs_socket_t *ksnc_sock; /* actual socket */
282 void *ksnc_saved_data_ready; /* socket's original data_ready() callback */
283 void *ksnc_saved_write_space; /* socket's original write_space() callback */
284 cfs_atomic_t ksnc_conn_refcount; /* conn refcount */
285 cfs_atomic_t ksnc_sock_refcount; /* sock refcount */
286 ksock_sched_t *ksnc_scheduler; /* who schedules this connection */
287 __u32 ksnc_myipaddr; /* my IP */
288 __u32 ksnc_ipaddr; /* peer's IP */
289 int ksnc_port; /* peer's port */
290 int ksnc_type:3; /* type of connection, should be signed value */
291 int ksnc_closing:1; /* being shut down */
292 int ksnc_flip:1; /* flip or not, only for V2.x */
293 int ksnc_zc_capable:1; /* enable to ZC */
294 struct ksock_proto *ksnc_proto; /* protocol for the connection */
297 cfs_list_t ksnc_rx_list; /* where I enq waiting input or a forwarding descriptor */
298 cfs_time_t ksnc_rx_deadline; /* when (in jiffies) receive times out */
299 __u8 ksnc_rx_started; /* started receiving a message */
300 __u8 ksnc_rx_ready; /* data ready to read */
301 __u8 ksnc_rx_scheduled;/* being progressed */
302 __u8 ksnc_rx_state; /* what is being read */
303 int ksnc_rx_nob_left; /* # bytes to next hdr/body */
304 int ksnc_rx_nob_wanted; /* bytes actually wanted */
305 int ksnc_rx_niov; /* # iovec frags */
306 struct iovec *ksnc_rx_iov; /* the iovec frags */
307 int ksnc_rx_nkiov; /* # page frags */
308 lnet_kiov_t *ksnc_rx_kiov; /* the page frags */
309 ksock_rxiovspace_t ksnc_rx_iov_space;/* space for frag descriptors */
310 __u32 ksnc_rx_csum; /* partial checksum for incoming data */
311 void *ksnc_cookie; /* rx lnet_finalize passthru arg */
312 ksock_msg_t ksnc_msg; /* incoming message buffer:
313 * V2.x message takes the
315 * V1.x message is a bare
316 * lnet_hdr_t, it's stored in
317 * ksnc_msg.ksm_u.lnetmsg */
320 cfs_list_t ksnc_tx_list; /* where I enq waiting for output space */
321 cfs_list_t ksnc_tx_queue; /* packets waiting to be sent */
322 ksock_tx_t *ksnc_tx_carrier; /* next TX that can carry a LNet message or ZC-ACK */
323 cfs_time_t ksnc_tx_deadline; /* when (in jiffies) tx times out */
324 int ksnc_tx_bufnob; /* send buffer marker */
325 cfs_atomic_t ksnc_tx_nob; /* # bytes queued */
326 int ksnc_tx_ready; /* write space */
327 int ksnc_tx_scheduled; /* being progressed */
328 cfs_time_t ksnc_tx_last_post; /* time stamp of the last posted TX */
331 typedef struct ksock_route
333 cfs_list_t ksnr_list; /* chain on peer route list */
334 cfs_list_t ksnr_connd_list; /* chain on ksnr_connd_routes */
335 struct ksock_peer *ksnr_peer; /* owning peer */
336 cfs_atomic_t ksnr_refcount; /* # users */
337 cfs_time_t ksnr_timeout; /* when (in jiffies) reconnection can happen next */
338 cfs_duration_t ksnr_retry_interval; /* how long between retries */
339 __u32 ksnr_myipaddr; /* my IP */
340 __u32 ksnr_ipaddr; /* IP address to connect to */
341 int ksnr_port; /* port to connect to */
342 unsigned int ksnr_scheduled:1; /* scheduled for attention */
343 unsigned int ksnr_connecting:1;/* connection establishment in progress */
344 unsigned int ksnr_connected:4; /* connections established by type */
345 unsigned int ksnr_deleted:1; /* been removed from peer? */
346 unsigned int ksnr_share_count; /* created explicitly? */
347 int ksnr_conn_count; /* # conns established by this route */
350 #define SOCKNAL_KEEPALIVE_PING 1 /* cookie for keepalive ping */
352 typedef struct ksock_peer
354 cfs_list_t ksnp_list; /* stash on global peer list */
355 cfs_time_t ksnp_last_alive; /* when (in jiffies) I was last alive */
356 lnet_process_id_t ksnp_id; /* who's on the other end(s) */
357 cfs_atomic_t ksnp_refcount; /* # users */
358 int ksnp_sharecount; /* lconf usage counter */
359 int ksnp_closing; /* being closed */
360 int ksnp_accepting;/* # passive connections pending */
361 int ksnp_error; /* errno on closing last conn */
362 __u64 ksnp_zc_next_cookie;/* ZC completion cookie */
363 __u64 ksnp_incarnation; /* latest known peer incarnation */
364 struct ksock_proto *ksnp_proto; /* latest known peer protocol */
365 cfs_list_t ksnp_conns; /* all active connections */
366 cfs_list_t ksnp_routes; /* routes */
367 cfs_list_t ksnp_tx_queue; /* waiting packets */
368 cfs_spinlock_t ksnp_lock; /* serialize, NOT safe in g_lock */
369 cfs_list_t ksnp_zc_req_list; /* zero copy requests wait for ACK */
370 cfs_time_t ksnp_send_keepalive; /* time to send keepalive */
371 lnet_ni_t *ksnp_ni; /* which network */
372 int ksnp_n_passive_ips; /* # of... */
373 __u32 ksnp_passive_ips[LNET_MAX_INTERFACES]; /* preferred local interfaces */
376 typedef struct ksock_connreq
378 cfs_list_t ksncr_list; /* stash on ksnd_connd_connreqs */
379 lnet_ni_t *ksncr_ni; /* chosen NI */
380 cfs_socket_t *ksncr_sock; /* accepted socket */
383 extern ksock_nal_data_t ksocknal_data;
384 extern ksock_tunables_t ksocknal_tunables;
386 #define SOCKNAL_MATCH_NO 0 /* TX can't match type of connection */
387 #define SOCKNAL_MATCH_YES 1 /* TX matches type of connection */
388 #define SOCKNAL_MATCH_MAY 2 /* TX can be sent on the connection, but not preferred */
390 typedef struct ksock_proto
392 int pro_version; /* version number of protocol */
393 int (*pro_send_hello)(ksock_conn_t *, ksock_hello_msg_t *); /* handshake function */
394 int (*pro_recv_hello)(ksock_conn_t *, ksock_hello_msg_t *, int);/* handshake function */
395 void (*pro_pack)(ksock_tx_t *); /* message pack */
396 void (*pro_unpack)(ksock_msg_t *); /* message unpack */
397 ksock_tx_t *(*pro_queue_tx_msg)(ksock_conn_t *, ksock_tx_t *); /* queue tx on the connection */
398 int (*pro_queue_tx_zcack)(ksock_conn_t *, ksock_tx_t *, __u64); /* queue ZC ack on the connection */
399 int (*pro_handle_zcreq)(ksock_conn_t *, __u64, int); /* handle ZC request */
400 int (*pro_handle_zcack)(ksock_conn_t *, __u64, __u64); /* handle ZC ACK */
401 int (*pro_match_tx)(ksock_conn_t *, ksock_tx_t *, int); /* msg type matches the connection type:
403 * return MATCH_NO : no
404 * return MATCH_YES : matching type
405 * return MATCH_MAY : can be backup */
408 extern ksock_proto_t ksocknal_protocol_v1x;
409 extern ksock_proto_t ksocknal_protocol_v2x;
410 extern ksock_proto_t ksocknal_protocol_v3x;
412 #define KSOCK_PROTO_V1_MAJOR LNET_PROTO_TCP_VERSION_MAJOR
413 #define KSOCK_PROTO_V1_MINOR LNET_PROTO_TCP_VERSION_MINOR
414 #define KSOCK_PROTO_V1 KSOCK_PROTO_V1_MAJOR
416 #ifndef CPU_MASK_NONE
417 #define CPU_MASK_NONE 0UL
421 ksocknal_route_mask(void)
423 if (!*ksocknal_tunables.ksnd_typed_conns)
424 return (1 << SOCKLND_CONN_ANY);
426 return ((1 << SOCKLND_CONN_CONTROL) |
427 (1 << SOCKLND_CONN_BULK_IN) |
428 (1 << SOCKLND_CONN_BULK_OUT));
431 static inline cfs_list_t *
432 ksocknal_nid2peerlist (lnet_nid_t nid)
434 unsigned int hash = ((unsigned int)nid) % ksocknal_data.ksnd_peer_hash_size;
436 return (&ksocknal_data.ksnd_peers [hash]);
440 ksocknal_conn_addref (ksock_conn_t *conn)
442 LASSERT (cfs_atomic_read(&conn->ksnc_conn_refcount) > 0);
443 cfs_atomic_inc(&conn->ksnc_conn_refcount);
446 extern void ksocknal_queue_zombie_conn (ksock_conn_t *conn);
447 extern void ksocknal_finalize_zcreq(ksock_conn_t *conn);
450 ksocknal_conn_decref (ksock_conn_t *conn)
452 LASSERT (cfs_atomic_read(&conn->ksnc_conn_refcount) > 0);
453 if (cfs_atomic_dec_and_test(&conn->ksnc_conn_refcount))
454 ksocknal_queue_zombie_conn(conn);
458 ksocknal_connsock_addref (ksock_conn_t *conn)
462 cfs_read_lock (&ksocknal_data.ksnd_global_lock);
463 if (!conn->ksnc_closing) {
464 LASSERT (cfs_atomic_read(&conn->ksnc_sock_refcount) > 0);
465 cfs_atomic_inc(&conn->ksnc_sock_refcount);
468 cfs_read_unlock (&ksocknal_data.ksnd_global_lock);
474 ksocknal_connsock_decref (ksock_conn_t *conn)
476 LASSERT (cfs_atomic_read(&conn->ksnc_sock_refcount) > 0);
477 if (cfs_atomic_dec_and_test(&conn->ksnc_sock_refcount)) {
478 LASSERT (conn->ksnc_closing);
479 libcfs_sock_release(conn->ksnc_sock);
480 conn->ksnc_sock = NULL;
481 ksocknal_finalize_zcreq(conn);
486 ksocknal_tx_addref (ksock_tx_t *tx)
488 LASSERT (cfs_atomic_read(&tx->tx_refcount) > 0);
489 cfs_atomic_inc(&tx->tx_refcount);
492 extern void ksocknal_tx_prep (ksock_conn_t *, ksock_tx_t *tx);
493 extern void ksocknal_tx_done (lnet_ni_t *ni, ksock_tx_t *tx);
496 ksocknal_tx_decref (ksock_tx_t *tx)
498 LASSERT (cfs_atomic_read(&tx->tx_refcount) > 0);
499 if (cfs_atomic_dec_and_test(&tx->tx_refcount))
500 ksocknal_tx_done(NULL, tx);
504 ksocknal_route_addref (ksock_route_t *route)
506 LASSERT (cfs_atomic_read(&route->ksnr_refcount) > 0);
507 cfs_atomic_inc(&route->ksnr_refcount);
510 extern void ksocknal_destroy_route (ksock_route_t *route);
513 ksocknal_route_decref (ksock_route_t *route)
515 LASSERT (cfs_atomic_read (&route->ksnr_refcount) > 0);
516 if (cfs_atomic_dec_and_test(&route->ksnr_refcount))
517 ksocknal_destroy_route (route);
521 ksocknal_peer_addref (ksock_peer_t *peer)
523 LASSERT (cfs_atomic_read (&peer->ksnp_refcount) > 0);
524 cfs_atomic_inc(&peer->ksnp_refcount);
527 extern void ksocknal_destroy_peer (ksock_peer_t *peer);
530 ksocknal_peer_decref (ksock_peer_t *peer)
532 LASSERT (cfs_atomic_read (&peer->ksnp_refcount) > 0);
533 if (cfs_atomic_dec_and_test(&peer->ksnp_refcount))
534 ksocknal_destroy_peer (peer);
537 int ksocknal_startup (lnet_ni_t *ni);
538 void ksocknal_shutdown (lnet_ni_t *ni);
539 int ksocknal_ctl(lnet_ni_t *ni, unsigned int cmd, void *arg);
540 int ksocknal_send (lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg);
541 int ksocknal_recv(lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg,
542 int delayed, unsigned int niov,
543 struct iovec *iov, lnet_kiov_t *kiov,
544 unsigned int offset, unsigned int mlen, unsigned int rlen);
545 int ksocknal_accept(lnet_ni_t *ni, cfs_socket_t *sock);
547 extern int ksocknal_add_peer(lnet_ni_t *ni, lnet_process_id_t id, __u32 ip, int port);
548 extern ksock_peer_t *ksocknal_find_peer_locked (lnet_ni_t *ni, lnet_process_id_t id);
549 extern ksock_peer_t *ksocknal_find_peer (lnet_ni_t *ni, lnet_process_id_t id);
550 extern void ksocknal_peer_failed (ksock_peer_t *peer);
551 extern int ksocknal_create_conn (lnet_ni_t *ni, ksock_route_t *route,
552 cfs_socket_t *sock, int type);
553 extern void ksocknal_close_conn_locked (ksock_conn_t *conn, int why);
554 extern void ksocknal_terminate_conn (ksock_conn_t *conn);
555 extern void ksocknal_destroy_conn (ksock_conn_t *conn);
556 extern int ksocknal_close_peer_conns_locked (ksock_peer_t *peer,
557 __u32 ipaddr, int why);
558 extern int ksocknal_close_conn_and_siblings (ksock_conn_t *conn, int why);
559 extern int ksocknal_close_matching_conns (lnet_process_id_t id, __u32 ipaddr);
560 extern ksock_conn_t *ksocknal_find_conn_locked(ksock_peer_t *peer,
561 ksock_tx_t *tx, int nonblk);
563 extern int ksocknal_launch_packet(lnet_ni_t *ni, ksock_tx_t *tx,
564 lnet_process_id_t id);
565 extern ksock_tx_t *ksocknal_alloc_tx(int type, int size);
566 extern void ksocknal_free_tx (ksock_tx_t *tx);
567 extern ksock_tx_t *ksocknal_alloc_tx_noop(__u64 cookie, int nonblk);
568 extern void ksocknal_next_tx_carrier(ksock_conn_t *conn);
569 extern void ksocknal_queue_tx_locked (ksock_tx_t *tx, ksock_conn_t *conn);
570 extern void ksocknal_txlist_done (lnet_ni_t *ni, cfs_list_t *txlist,
572 extern void ksocknal_notify (lnet_ni_t *ni, lnet_nid_t gw_nid, int alive);
573 extern void ksocknal_query (struct lnet_ni *ni, lnet_nid_t nid, cfs_time_t *when);
574 extern int ksocknal_thread_start (int (*fn)(void *arg), void *arg);
575 extern void ksocknal_thread_fini (void);
576 extern void ksocknal_launch_all_connections_locked (ksock_peer_t *peer);
577 extern ksock_route_t *ksocknal_find_connectable_route_locked (ksock_peer_t *peer);
578 extern ksock_route_t *ksocknal_find_connecting_route_locked (ksock_peer_t *peer);
579 extern int ksocknal_new_packet (ksock_conn_t *conn, int skip);
580 extern int ksocknal_scheduler (void *arg);
581 extern int ksocknal_connd (void *arg);
582 extern int ksocknal_reaper (void *arg);
583 extern int ksocknal_send_hello (lnet_ni_t *ni, ksock_conn_t *conn,
584 lnet_nid_t peer_nid, ksock_hello_msg_t *hello);
585 extern int ksocknal_recv_hello (lnet_ni_t *ni, ksock_conn_t *conn,
586 ksock_hello_msg_t *hello, lnet_process_id_t *id,
588 extern void ksocknal_read_callback(ksock_conn_t *conn);
589 extern void ksocknal_write_callback(ksock_conn_t *conn);
591 extern int ksocknal_lib_zc_capable(ksock_conn_t *conn);
592 extern void ksocknal_lib_save_callback(cfs_socket_t *sock, ksock_conn_t *conn);
593 extern void ksocknal_lib_set_callback(cfs_socket_t *sock, ksock_conn_t *conn);
594 extern void ksocknal_lib_reset_callback(cfs_socket_t *sock, ksock_conn_t *conn);
595 extern void ksocknal_lib_push_conn (ksock_conn_t *conn);
596 extern void ksocknal_lib_bind_irq (unsigned int irq);
597 extern int ksocknal_lib_get_conn_addrs (ksock_conn_t *conn);
598 extern unsigned int ksocknal_lib_sock_irq (cfs_socket_t *sock);
599 extern int ksocknal_lib_setup_sock (cfs_socket_t *so);
600 extern int ksocknal_lib_send_iov (ksock_conn_t *conn, ksock_tx_t *tx);
601 extern int ksocknal_lib_send_kiov (ksock_conn_t *conn, ksock_tx_t *tx);
602 extern void ksocknal_lib_eager_ack (ksock_conn_t *conn);
603 extern int ksocknal_lib_recv_iov (ksock_conn_t *conn);
604 extern int ksocknal_lib_recv_kiov (ksock_conn_t *conn);
605 extern int ksocknal_lib_get_conn_tunables (ksock_conn_t *conn, int *txmem,
606 int *rxmem, int *nagle);
608 extern int ksocknal_tunables_init(void);
609 extern void ksocknal_tunables_fini(void);
610 extern int ksocknal_lib_tunables_init(void);
611 extern void ksocknal_lib_tunables_fini(void);
613 extern void ksocknal_lib_csum_tx(ksock_tx_t *tx);
615 extern int ksocknal_lib_memory_pressure(ksock_conn_t *conn);
616 extern int ksocknal_lib_bind_thread_to_cpu(int id);