2 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
4 * Copyright (c) 2011, 2017, 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 #ifndef _SOCKLND_SOCKLND_H_
29 #define _SOCKLND_SOCKLND_H_
31 #define DEBUG_PORTAL_ALLOC
32 #define DEBUG_SUBSYSTEM S_LND
34 #include <linux/crc32.h>
35 #include <linux/errno.h>
37 #include <linux/init.h>
38 #include <linux/kernel.h>
39 #include <linux/kthread.h>
40 #include <linux/kmod.h>
41 #include <linux/list.h>
43 #include <linux/module.h>
44 #include <linux/pagemap.h>
45 #include <linux/stat.h>
46 #include <linux/string.h>
47 #include <linux/syscalls.h>
48 #include <linux/sysctl.h>
49 #include <linux/uio.h>
50 #include <linux/unistd.h>
54 #include <lnet/lib-lnet.h>
55 #include <lnet/socklnd.h>
57 #ifndef NETIF_F_CSUM_MASK
58 # define NETIF_F_CSUM_MASK NETIF_F_ALL_CSUM
61 /* assume one thread for each connection type */
62 #define SOCKNAL_NSCHEDS 3
63 #define SOCKNAL_NSCHEDS_HIGH (SOCKNAL_NSCHEDS << 1)
65 #define SOCKNAL_PEER_HASH_SIZE 101 /* # peer_ni lists */
66 #define SOCKNAL_RESCHED 100 /* # scheduler loops before reschedule */
67 #define SOCKNAL_INSANITY_RECONN 5000 /* connd is trying on reconn infinitely */
68 #define SOCKNAL_ENOMEM_RETRY 1 /* seconds between retries */
70 #define SOCKNAL_SINGLE_FRAG_TX 0 /* disable multi-fragment sends */
71 #define SOCKNAL_SINGLE_FRAG_RX 0 /* disable multi-fragment receives */
73 #define SOCKNAL_VERSION_DEBUG 0 /* enable protocol version debugging */
75 /* risk kmap deadlock on multi-frag I/O (backs off to single-frag if disabled).
76 * no risk if we're not running on a CONFIG_HIGHMEM platform. */
78 # define SOCKNAL_RISK_KMAP_DEADLOCK 0
80 # define SOCKNAL_RISK_KMAP_DEADLOCK 1
83 /* per scheduler state */
87 /* conn waiting to be written */
88 struct list_head kss_rx_conns;
89 struct list_head kss_tx_conns;
90 /* zombie noop tx list */
91 struct list_head kss_zombie_noop_txs;
92 /* where scheduler sleeps */
93 wait_queue_head_t kss_waitq;
94 /* # connections assigned to this scheduler */
96 /* max allowed threads */
98 /* number of threads */
104 #define KSOCK_CPT_SHIFT 16
105 #define KSOCK_THREAD_ID(cpt, sid) (((cpt) << KSOCK_CPT_SHIFT) | (sid))
106 #define KSOCK_THREAD_CPT(id) ((id) >> KSOCK_CPT_SHIFT)
107 #define KSOCK_THREAD_SID(id) ((id) & ((1UL << KSOCK_CPT_SHIFT) - 1))
109 struct ksock_interface { /* in-use interface */
110 __u32 ksni_ipaddr; /* interface's IP address */
111 __u32 ksni_netmask; /* interface's network mask */
112 int ksni_nroutes; /* # routes using (active) */
113 int ksni_npeers; /* # peers using (passive) */
114 char ksni_name[IFNAMSIZ]; /* interface name */
117 struct ksock_tunables {
118 /* "stuck" socket timeout (seconds) */
120 /* # scheduler threads in each pool while starting */
122 int *ksnd_nconnds; /* # connection daemons */
123 int *ksnd_nconnds_max; /* max # connection daemons */
124 int *ksnd_min_reconnectms; /* first connection retry after (ms)... */
125 int *ksnd_max_reconnectms; /* ...exponentially increasing to this */
126 int *ksnd_eager_ack; /* make TCP ack eagerly? */
127 int *ksnd_typed_conns; /* drive sockets by type? */
128 int *ksnd_min_bulk; /* smallest "large" message */
129 int *ksnd_tx_buffer_size; /* socket tx buffer size */
130 int *ksnd_rx_buffer_size; /* socket rx buffer size */
131 int *ksnd_nagle; /* enable NAGLE? */
132 int *ksnd_round_robin; /* round robin for multiple interfaces */
133 int *ksnd_keepalive; /* # secs for sending keepalive NOOP */
134 int *ksnd_keepalive_idle; /* # idle secs before 1st probe */
135 int *ksnd_keepalive_count; /* # probes */
136 int *ksnd_keepalive_intvl; /* time between probes */
137 int *ksnd_credits; /* # concurrent sends */
138 int *ksnd_peertxcredits; /* # concurrent sends to 1 peer_ni */
139 int *ksnd_peerrtrcredits; /* # per-peer_ni router buffer credits */
140 int *ksnd_peertimeout; /* seconds to consider peer_ni dead */
141 int *ksnd_enable_csum; /* enable check sum */
142 int *ksnd_inject_csum_error; /* set non-zero to inject checksum error */
143 int *ksnd_nonblk_zcack; /* always send zc-ack on non-blocking connection */
144 unsigned int *ksnd_zc_min_payload; /* minimum zero copy payload size */
145 int *ksnd_zc_recv; /* enable ZC receive (for Chelsio TOE) */
146 int *ksnd_zc_recv_min_nfrags; /* minimum # of fragments to enable ZC receive */
147 int *ksnd_irq_affinity; /* enable IRQ affinity? */
148 #ifdef SOCKNAL_BACKOFF
149 int *ksnd_backoff_init; /* initial TCP backoff */
150 int *ksnd_backoff_max; /* maximum TCP backoff */
152 #if SOCKNAL_VERSION_DEBUG
153 int *ksnd_protocol; /* protocol version */
158 __u64 ksnn_incarnation; /* my epoch */
159 struct list_head ksnn_list; /* chain on global list */
160 atomic_t ksnn_npeers; /* # peers */
161 int ksnn_ninterfaces; /* IP interfaces */
162 struct ksock_interface ksnn_interfaces[LNET_INTERFACES_NUM];
164 /* When the ksock_net is shut down, this (negative) bias is added to
165 * ksnn_npeers, which prevents new peers from being added.
167 #define SOCKNAL_SHUTDOWN_BIAS (INT_MIN+1)
170 #define SOCKNAL_CONND_TIMEOUT 120
171 /** reserved thread for accepting & creating new connd */
172 #define SOCKNAL_CONND_RESV 1
174 struct ksock_nal_data {
175 int ksnd_init; /* initialisation state */
176 int ksnd_nnets; /* # networks set up */
177 struct list_head ksnd_nets; /* list of nets */
178 /* stabilize peer_ni/conn ops */
179 rwlock_t ksnd_global_lock;
180 /* hash table of all my known peers */
181 struct list_head *ksnd_peers;
182 int ksnd_peer_hash_size; /* size of ksnd_peers */
184 int ksnd_nthreads; /* # live threads */
185 int ksnd_shuttingdown; /* tell threads to exit */
186 /* schedulers information */
187 struct ksock_sched **ksnd_schedulers;
189 atomic_t ksnd_nactive_txs; /* #active txs */
191 /* conns to close: reaper_lock*/
192 struct list_head ksnd_deathrow_conns;
193 /* conns to free: reaper_lock */
194 struct list_head ksnd_zombie_conns;
195 /* conns to retry: reaper_lock*/
196 struct list_head ksnd_enomem_conns;
197 /* reaper sleeps here */
198 wait_queue_head_t ksnd_reaper_waitq;
199 /* when reaper will wake */
200 time64_t ksnd_reaper_waketime;
202 spinlock_t ksnd_reaper_lock;
204 int ksnd_enomem_tx; /* test ENOMEM sender */
205 int ksnd_stall_tx; /* test sluggish sender */
206 int ksnd_stall_rx; /* test sluggish receiver */
208 /* incoming connection requests */
209 struct list_head ksnd_connd_connreqs;
210 /* routes waiting to be connected */
211 struct list_head ksnd_connd_routes;
212 /* connds sleep here */
213 wait_queue_head_t ksnd_connd_waitq;
214 /* # connds connecting */
215 int ksnd_connd_connecting;
216 /** time stamp of the last failed connecting attempt */
217 time64_t ksnd_connd_failed_stamp;
218 /** # starting connd */
219 unsigned ksnd_connd_starting;
220 /** time stamp of the last starting connd */
221 time64_t ksnd_connd_starting_stamp;
222 /** # running connd */
223 unsigned ksnd_connd_running;
225 spinlock_t ksnd_connd_lock;
227 /* list head for freed noop tx */
228 struct list_head ksnd_idle_noop_txs;
229 /* serialise, g_lock unsafe */
230 spinlock_t ksnd_tx_lock;
233 #define SOCKNAL_INIT_NOTHING 0
234 #define SOCKNAL_INIT_DATA 1
235 #define SOCKNAL_INIT_ALL 2
237 /* A packet just assembled for transmission is represented by 1 or more
238 * struct kvec fragments (the first frag contains the portals header),
239 * followed by 0 or more lnet_kiov_t fragments.
241 * On the receive side, initially 1 struct kvec fragment is posted for
242 * receive (the header). Once the header has been received, the payload is
243 * received into either struct kvec or lnet_kiov_t fragments, depending on
244 * what the header matched or whether the message needs forwarding. */
246 struct ksock_conn; /* forward ref */
247 struct ksock_route; /* forward ref */
248 struct ksock_proto; /* forward ref */
250 struct ksock_tx { /* transmit packet */
251 struct list_head tx_list; /* queue on conn for transmission etc */
252 struct list_head tx_zc_list; /* queue on peer_ni for ZC request */
253 atomic_t tx_refcount; /* tx reference count */
254 int tx_nob; /* # packet bytes */
255 int tx_resid; /* residual bytes */
256 int tx_niov; /* # packet kvec frags */
257 struct kvec *tx_iov; /* packet kvec frags */
258 int tx_nkiov; /* # packet page frags */
259 unsigned short tx_zc_aborted; /* aborted ZC request */
260 unsigned short tx_zc_capable:1; /* payload is large enough for ZC */
261 unsigned short tx_zc_checked:1; /* Have I checked if I should ZC? */
262 unsigned short tx_nonblk:1; /* it's a non-blocking ACK */
263 lnet_kiov_t *tx_kiov; /* packet page frags */
264 struct ksock_conn *tx_conn; /* owning conn */
265 struct lnet_msg *tx_lnetmsg; /* lnet message for lnet_finalize() */
266 time64_t tx_deadline; /* when (in secs) tx times out */
267 struct ksock_msg tx_msg; /* socklnd message buffer */
268 int tx_desc_size; /* size of this descriptor */
269 enum lnet_msg_hstatus tx_hstatus; /* health status of tx */
272 struct kvec iov; /* virt hdr */
273 lnet_kiov_t kiov[0]; /* paged payload */
276 struct kvec iov[1]; /* virt hdr + payload */
281 #define KSOCK_NOOP_TX_SIZE ((int)offsetof(struct ksock_tx, tx_frags.paged.kiov[0]))
283 /* network zero copy callback descriptor embedded in struct ksock_tx */
285 /* space for the rx frag descriptors; we either read a single contiguous
286 * header, or up to LNET_MAX_IOV frags of payload of either type. */
287 union ksock_rxiovspace {
288 struct kvec iov[LNET_MAX_IOV];
289 lnet_kiov_t kiov[LNET_MAX_IOV];
292 #define SOCKNAL_RX_KSM_HEADER 1 /* reading ksock message header */
293 #define SOCKNAL_RX_LNET_HEADER 2 /* reading lnet message header */
294 #define SOCKNAL_RX_PARSE 3 /* Calling lnet_parse() */
295 #define SOCKNAL_RX_PARSE_WAIT 4 /* waiting to be told to read the body */
296 #define SOCKNAL_RX_LNET_PAYLOAD 5 /* reading lnet payload (to deliver here) */
297 #define SOCKNAL_RX_SLOP 6 /* skipping body */
300 struct ksock_peer_ni *ksnc_peer; /* owning peer_ni */
301 struct ksock_route *ksnc_route; /* owning route */
302 struct list_head ksnc_list; /* stash on peer_ni's conn list */
303 struct socket *ksnc_sock; /* actual socket */
304 void *ksnc_saved_data_ready; /* socket's original data_ready() callback */
305 void *ksnc_saved_write_space; /* socket's original write_space() callback */
306 atomic_t ksnc_conn_refcount; /* conn refcount */
307 atomic_t ksnc_sock_refcount; /* sock refcount */
308 struct ksock_sched *ksnc_scheduler; /* who schedules this connection */
309 __u32 ksnc_myipaddr; /* my IP */
310 __u32 ksnc_ipaddr; /* peer_ni's IP */
311 int ksnc_port; /* peer_ni's port */
312 signed int ksnc_type:3; /* type of connection,
313 * should be signed value */
314 unsigned int ksnc_closing:1; /* being shut down */
315 unsigned int ksnc_flip:1; /* flip or not, only for V2.x */
316 unsigned int ksnc_zc_capable:1; /* enable to ZC */
317 struct ksock_proto *ksnc_proto; /* protocol for the connection */
321 /* where I enq waiting input or a forwarding descriptor */
322 struct list_head ksnc_rx_list;
323 time64_t ksnc_rx_deadline; /* when (in seconds) receive times out */
324 __u8 ksnc_rx_started; /* started receiving a message */
325 __u8 ksnc_rx_ready; /* data ready to read */
326 __u8 ksnc_rx_scheduled;/* being progressed */
327 __u8 ksnc_rx_state; /* what is being read */
328 int ksnc_rx_nob_left; /* # bytes to next hdr/body */
329 int ksnc_rx_nob_wanted; /* bytes actually wanted */
330 int ksnc_rx_niov; /* # kvec frags */
331 struct kvec *ksnc_rx_iov; /* the kvec frags */
332 int ksnc_rx_nkiov; /* # page frags */
333 lnet_kiov_t *ksnc_rx_kiov; /* the page frags */
334 union ksock_rxiovspace ksnc_rx_iov_space;/* space for frag descriptors */
335 __u32 ksnc_rx_csum; /* partial checksum for incoming data */
336 struct lnet_msg *ksnc_lnet_msg; /* rx lnet_finalize arg*/
337 struct ksock_msg ksnc_msg; /* incoming message buffer:
338 * V2.x message takes the
340 * V1.x message is a bare
341 * struct lnet_hdr, it's stored
342 * in ksnc_msg.ksm_u.lnetmsg
345 /* where I enq waiting for output space */
346 struct list_head ksnc_tx_list;
347 /* packets waiting to be sent */
348 struct list_head ksnc_tx_queue;
349 /* next TX that can carry a LNet message or ZC-ACK */
350 struct ksock_tx *ksnc_tx_carrier;
351 /* when (in seconds) tx times out */
352 time64_t ksnc_tx_deadline;
353 /* send buffer marker */
356 atomic_t ksnc_tx_nob;
359 /* being progressed */
360 int ksnc_tx_scheduled;
361 /* time stamp of the last posted TX */
362 time64_t ksnc_tx_last_post;
366 struct list_head ksnr_list; /* chain on peer_ni route list */
367 struct list_head ksnr_connd_list; /* chain on ksnr_connd_routes */
368 struct ksock_peer_ni *ksnr_peer; /* owning peer_ni */
369 atomic_t ksnr_refcount; /* # users */
370 time64_t ksnr_timeout; /* when (in secs) reconnection can happen next */
371 time64_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_ni? */
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 struct ksock_peer_ni {
386 struct list_head ksnp_list; /* stash on global peer_ni list */
387 time64_t ksnp_last_alive;/* when (in seconds) I was last alive */
388 struct lnet_process_id ksnp_id; /* who's on the other end(s) */
389 atomic_t ksnp_refcount; /* # users */
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_ni incarnation */
395 struct ksock_proto *ksnp_proto; /* latest known peer_ni 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 time64_t ksnp_send_keepalive; /* time to send keepalive */
403 struct lnet_ni *ksnp_ni; /* which network */
404 int ksnp_n_passive_ips; /* # of... */
405 __u32 ksnp_passive_ips[LNET_INTERFACES_NUM]; /* preferred local interfaces */
408 struct ksock_connreq {
409 /* stash on ksnd_connd_connreqs */
410 struct list_head ksncr_list;
412 struct lnet_ni *ksncr_ni;
413 /* accepted socket */
414 struct socket *ksncr_sock;
417 extern struct ksock_nal_data ksocknal_data;
418 extern struct ksock_tunables ksocknal_tunables;
420 #define SOCKNAL_MATCH_NO 0 /* TX can't match type of connection */
421 #define SOCKNAL_MATCH_YES 1 /* TX matches type of connection */
422 #define SOCKNAL_MATCH_MAY 2 /* TX can be sent on the connection, but not preferred */
425 int pro_version; /* version number of protocol */
426 int (*pro_send_hello)(struct ksock_conn *, struct ksock_hello_msg *); /* handshake function */
427 int (*pro_recv_hello)(struct ksock_conn *, struct ksock_hello_msg *, int);/* handshake function */
428 void (*pro_pack)(struct ksock_tx *); /* message pack */
429 void (*pro_unpack)(struct ksock_msg *); /* message unpack */
430 struct ksock_tx *(*pro_queue_tx_msg)(struct ksock_conn *, struct ksock_tx *); /* queue tx on the connection */
431 int (*pro_queue_tx_zcack)(struct ksock_conn *, struct ksock_tx *, __u64); /* queue ZC ack on the connection */
432 int (*pro_handle_zcreq)(struct ksock_conn *, __u64, int); /* handle ZC request */
433 int (*pro_handle_zcack)(struct ksock_conn *, __u64, __u64); /* handle ZC ACK */
434 int (*pro_match_tx)(struct ksock_conn *, struct ksock_tx *, int); /* msg type matches the connection type:
436 * return MATCH_NO : no
437 * return MATCH_YES : matching type
438 * return MATCH_MAY : can be backup */
441 extern struct ksock_proto ksocknal_protocol_v1x;
442 extern struct ksock_proto ksocknal_protocol_v2x;
443 extern struct ksock_proto ksocknal_protocol_v3x;
445 #define KSOCK_PROTO_V1_MAJOR LNET_PROTO_TCP_VERSION_MAJOR
446 #define KSOCK_PROTO_V1_MINOR LNET_PROTO_TCP_VERSION_MINOR
447 #define KSOCK_PROTO_V1 KSOCK_PROTO_V1_MAJOR
449 #ifndef CPU_MASK_NONE
450 #define CPU_MASK_NONE 0UL
453 static inline __u32 ksocknal_csum(__u32 crc, unsigned char const *p, size_t len)
456 return crc32_le(crc, p, len);
459 crc = ((crc + 0x100) & ~0xff) | ((crc + *p++) & 0xff) ;
466 ksocknal_route_mask(void)
468 if (!*ksocknal_tunables.ksnd_typed_conns)
469 return (1 << SOCKLND_CONN_ANY);
471 return ((1 << SOCKLND_CONN_CONTROL) |
472 (1 << SOCKLND_CONN_BULK_IN) |
473 (1 << SOCKLND_CONN_BULK_OUT));
476 static inline struct list_head *
477 ksocknal_nid2peerlist (lnet_nid_t nid)
479 unsigned int hash = ((unsigned int)nid) % ksocknal_data.ksnd_peer_hash_size;
481 return (&ksocknal_data.ksnd_peers [hash]);
485 ksocknal_conn_addref(struct ksock_conn *conn)
487 LASSERT(atomic_read(&conn->ksnc_conn_refcount) > 0);
488 atomic_inc(&conn->ksnc_conn_refcount);
491 extern void ksocknal_queue_zombie_conn(struct ksock_conn *conn);
492 extern void ksocknal_finalize_zcreq(struct ksock_conn *conn);
495 ksocknal_conn_decref(struct ksock_conn *conn)
497 LASSERT(atomic_read(&conn->ksnc_conn_refcount) > 0);
498 if (atomic_dec_and_test(&conn->ksnc_conn_refcount))
499 ksocknal_queue_zombie_conn(conn);
503 ksocknal_connsock_addref(struct ksock_conn *conn)
507 read_lock(&ksocknal_data.ksnd_global_lock);
508 if (!conn->ksnc_closing) {
509 LASSERT(atomic_read(&conn->ksnc_sock_refcount) > 0);
510 atomic_inc(&conn->ksnc_sock_refcount);
513 read_unlock(&ksocknal_data.ksnd_global_lock);
519 ksocknal_connsock_decref(struct ksock_conn *conn)
521 LASSERT(atomic_read(&conn->ksnc_sock_refcount) > 0);
522 if (atomic_dec_and_test(&conn->ksnc_sock_refcount)) {
523 LASSERT (conn->ksnc_closing);
524 sock_release(conn->ksnc_sock);
525 conn->ksnc_sock = NULL;
526 ksocknal_finalize_zcreq(conn);
531 ksocknal_tx_addref(struct ksock_tx *tx)
533 LASSERT(atomic_read(&tx->tx_refcount) > 0);
534 atomic_inc(&tx->tx_refcount);
537 extern void ksocknal_tx_prep(struct ksock_conn *, struct ksock_tx *tx);
538 extern void ksocknal_tx_done(struct lnet_ni *ni, struct ksock_tx *tx, int error);
541 ksocknal_tx_decref(struct ksock_tx *tx)
543 LASSERT(atomic_read(&tx->tx_refcount) > 0);
544 if (atomic_dec_and_test(&tx->tx_refcount))
545 ksocknal_tx_done(NULL, tx, 0);
549 ksocknal_route_addref(struct ksock_route *route)
551 LASSERT(atomic_read(&route->ksnr_refcount) > 0);
552 atomic_inc(&route->ksnr_refcount);
555 extern void ksocknal_destroy_route(struct ksock_route *route);
558 ksocknal_route_decref(struct ksock_route *route)
560 LASSERT(atomic_read(&route->ksnr_refcount) > 0);
561 if (atomic_dec_and_test(&route->ksnr_refcount))
562 ksocknal_destroy_route (route);
566 ksocknal_peer_addref(struct ksock_peer_ni *peer_ni)
568 LASSERT(atomic_read(&peer_ni->ksnp_refcount) > 0);
569 atomic_inc(&peer_ni->ksnp_refcount);
572 extern void ksocknal_destroy_peer(struct ksock_peer_ni *peer_ni);
575 ksocknal_peer_decref(struct ksock_peer_ni *peer_ni)
577 LASSERT (atomic_read (&peer_ni->ksnp_refcount) > 0);
578 if (atomic_dec_and_test(&peer_ni->ksnp_refcount))
579 ksocknal_destroy_peer(peer_ni);
582 int ksocknal_startup(struct lnet_ni *ni);
583 void ksocknal_shutdown(struct lnet_ni *ni);
584 int ksocknal_ctl(struct lnet_ni *ni, unsigned int cmd, void *arg);
585 int ksocknal_send(struct lnet_ni *ni, void *private, struct lnet_msg *lntmsg);
586 int ksocknal_recv(struct lnet_ni *ni, void *private, struct lnet_msg *lntmsg,
587 int delayed, unsigned int niov,
588 struct kvec *iov, lnet_kiov_t *kiov,
589 unsigned int offset, unsigned int mlen, unsigned int rlen);
590 int ksocknal_accept(struct lnet_ni *ni, struct socket *sock);
592 int ksocknal_add_peer(struct lnet_ni *ni, struct lnet_process_id id, __u32 ip,
594 struct ksock_peer_ni *ksocknal_find_peer_locked(struct lnet_ni *ni,
595 struct lnet_process_id id);
596 struct ksock_peer_ni *ksocknal_find_peer(struct lnet_ni *ni,
597 struct lnet_process_id id);
598 extern void ksocknal_peer_failed(struct ksock_peer_ni *peer_ni);
599 extern int ksocknal_create_conn(struct lnet_ni *ni, struct ksock_route *route,
600 struct socket *sock, int type);
601 extern void ksocknal_close_conn_locked(struct ksock_conn *conn, int why);
602 extern void ksocknal_terminate_conn(struct ksock_conn *conn);
603 extern void ksocknal_destroy_conn(struct ksock_conn *conn);
604 extern int ksocknal_close_peer_conns_locked(struct ksock_peer_ni *peer_ni,
605 __u32 ipaddr, int why);
606 extern int ksocknal_close_conn_and_siblings(struct ksock_conn *conn, int why);
607 int ksocknal_close_matching_conns(struct lnet_process_id id, __u32 ipaddr);
608 extern struct ksock_conn *ksocknal_find_conn_locked(struct ksock_peer_ni *peer_ni,
609 struct ksock_tx *tx, int nonblk);
611 extern int ksocknal_launch_packet(struct lnet_ni *ni, struct ksock_tx *tx,
612 struct lnet_process_id id);
613 extern struct ksock_tx *ksocknal_alloc_tx(int type, int size);
614 extern void ksocknal_free_tx(struct ksock_tx *tx);
615 extern struct ksock_tx *ksocknal_alloc_tx_noop(__u64 cookie, int nonblk);
616 extern void ksocknal_next_tx_carrier(struct ksock_conn *conn);
617 extern void ksocknal_queue_tx_locked(struct ksock_tx *tx, struct ksock_conn *conn);
618 extern void ksocknal_txlist_done(struct lnet_ni *ni, struct list_head *txlist,
620 extern void ksocknal_notify(lnet_nid_t gw_nid);
621 extern void ksocknal_query(struct lnet_ni *ni, lnet_nid_t nid, time64_t *when);
622 extern int ksocknal_thread_start(int (*fn)(void *arg), void *arg, char *name);
623 extern void ksocknal_thread_fini(void);
624 extern void ksocknal_launch_all_connections_locked(struct ksock_peer_ni *peer_ni);
625 extern struct ksock_route *ksocknal_find_connectable_route_locked(struct ksock_peer_ni *peer_ni);
626 extern struct ksock_route *ksocknal_find_connecting_route_locked(struct ksock_peer_ni *peer_ni);
627 extern int ksocknal_new_packet(struct ksock_conn *conn, int skip);
628 extern int ksocknal_scheduler(void *arg);
629 extern int ksocknal_connd(void *arg);
630 extern int ksocknal_reaper(void *arg);
631 int ksocknal_send_hello(struct lnet_ni *ni, struct ksock_conn *conn,
632 lnet_nid_t peer_nid, struct ksock_hello_msg *hello);
633 int ksocknal_recv_hello(struct lnet_ni *ni, struct ksock_conn *conn,
634 struct ksock_hello_msg *hello,
635 struct lnet_process_id *id,
637 extern void ksocknal_read_callback(struct ksock_conn *conn);
638 extern void ksocknal_write_callback(struct ksock_conn *conn);
640 extern int ksocknal_lib_zc_capable(struct ksock_conn *conn);
641 extern void ksocknal_lib_save_callback(struct socket *sock, struct ksock_conn *conn);
642 extern void ksocknal_lib_set_callback(struct socket *sock, struct ksock_conn *conn);
643 extern void ksocknal_lib_reset_callback(struct socket *sock,
644 struct ksock_conn *conn);
645 extern void ksocknal_lib_push_conn(struct ksock_conn *conn);
646 extern int ksocknal_lib_get_conn_addrs(struct ksock_conn *conn);
647 extern int ksocknal_lib_setup_sock(struct socket *so);
648 extern int ksocknal_lib_send_iov(struct ksock_conn *conn, struct ksock_tx *tx,
649 struct kvec *scratch_iov);
650 extern int ksocknal_lib_send_kiov(struct ksock_conn *conn, struct ksock_tx *tx,
651 struct kvec *scratch_iov);
652 extern void ksocknal_lib_eager_ack(struct ksock_conn *conn);
653 extern int ksocknal_lib_recv_iov(struct ksock_conn *conn,
654 struct kvec *scratchiov);
655 extern int ksocknal_lib_recv_kiov(struct ksock_conn *conn, struct page **pages,
656 struct kvec *scratchiov);
657 extern int ksocknal_lib_get_conn_tunables(struct ksock_conn *conn, int *txmem,
658 int *rxmem, int *nagle);
660 extern int ksocknal_tunables_init(void);
662 extern void ksocknal_lib_csum_tx(struct ksock_tx *tx);
664 extern int ksocknal_lib_memory_pressure(struct ksock_conn *conn);
665 extern int ksocknal_lib_bind_thread_to_cpu(int id);
667 #endif /* _SOCKLND_SOCKLND_H_ */