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 #ifdef HAVE_TCP_SENDPAGE_USE_SOCKET
58 # define cfs_tcp_sendpage(sk, page, offset, size, flags) \
59 tcp_sendpage((sk)->sk_socket, page, offset, size, flags)
60 #else /* !HAVE_TCP_SENDPAGE_USE_SOCKET */
61 # define cfs_tcp_sendpage(sk, page, offset, size, flags) \
62 tcp_sendpage(sk, page, offset, size, flags)
63 #endif /* HAVE_TCP_SENDPAGE_USE_SOCKET */
65 #ifndef NETIF_F_CSUM_MASK
66 # define NETIF_F_CSUM_MASK NETIF_F_ALL_CSUM
69 /* assume one thread for each connection type */
70 #define SOCKNAL_NSCHEDS 3
71 #define SOCKNAL_NSCHEDS_HIGH (SOCKNAL_NSCHEDS << 1)
73 #define SOCKNAL_PEER_HASH_SIZE 101 /* # peer_ni lists */
74 #define SOCKNAL_RESCHED 100 /* # scheduler loops before reschedule */
75 #define SOCKNAL_INSANITY_RECONN 5000 /* connd is trying on reconn infinitely */
76 #define SOCKNAL_ENOMEM_RETRY 1 /* seconds between retries */
78 #define SOCKNAL_SINGLE_FRAG_TX 0 /* disable multi-fragment sends */
79 #define SOCKNAL_SINGLE_FRAG_RX 0 /* disable multi-fragment receives */
81 #define SOCKNAL_VERSION_DEBUG 0 /* enable protocol version debugging */
83 /* risk kmap deadlock on multi-frag I/O (backs off to single-frag if disabled).
84 * no risk if we're not running on a CONFIG_HIGHMEM platform. */
86 # define SOCKNAL_RISK_KMAP_DEADLOCK 0
88 # define SOCKNAL_RISK_KMAP_DEADLOCK 1
91 /* per scheduler state */
95 /* conn waiting to be written */
96 struct list_head kss_rx_conns;
97 struct list_head kss_tx_conns;
98 /* zombie noop tx list */
99 struct list_head kss_zombie_noop_txs;
100 /* where scheduler sleeps */
101 wait_queue_head_t kss_waitq;
102 /* # connections assigned to this scheduler */
104 /* max allowed threads */
105 int kss_nthreads_max;
106 /* number of threads */
112 #define KSOCK_CPT_SHIFT 16
113 #define KSOCK_THREAD_ID(cpt, sid) (((cpt) << KSOCK_CPT_SHIFT) | (sid))
114 #define KSOCK_THREAD_CPT(id) ((id) >> KSOCK_CPT_SHIFT)
115 #define KSOCK_THREAD_SID(id) ((id) & ((1UL << KSOCK_CPT_SHIFT) - 1))
117 struct ksock_interface { /* in-use interface */
118 __u32 ksni_ipaddr; /* interface's IP address */
119 __u32 ksni_netmask; /* interface's network mask */
120 int ksni_nroutes; /* # routes using (active) */
121 int ksni_npeers; /* # peers using (passive) */
122 char ksni_name[IFNAMSIZ]; /* interface name */
125 struct ksock_tunables {
126 /* "stuck" socket timeout (seconds) */
128 /* # scheduler threads in each pool while starting */
130 int *ksnd_nconnds; /* # connection daemons */
131 int *ksnd_nconnds_max; /* max # connection daemons */
132 int *ksnd_min_reconnectms; /* first connection retry after (ms)... */
133 int *ksnd_max_reconnectms; /* ...exponentially increasing to this */
134 int *ksnd_eager_ack; /* make TCP ack eagerly? */
135 int *ksnd_typed_conns; /* drive sockets by type? */
136 int *ksnd_min_bulk; /* smallest "large" message */
137 int *ksnd_tx_buffer_size; /* socket tx buffer size */
138 int *ksnd_rx_buffer_size; /* socket rx buffer size */
139 int *ksnd_nagle; /* enable NAGLE? */
140 int *ksnd_round_robin; /* round robin for multiple interfaces */
141 int *ksnd_keepalive; /* # secs for sending keepalive NOOP */
142 int *ksnd_keepalive_idle; /* # idle secs before 1st probe */
143 int *ksnd_keepalive_count; /* # probes */
144 int *ksnd_keepalive_intvl; /* time between probes */
145 int *ksnd_credits; /* # concurrent sends */
146 int *ksnd_peertxcredits; /* # concurrent sends to 1 peer_ni */
147 int *ksnd_peerrtrcredits; /* # per-peer_ni router buffer credits */
148 int *ksnd_peertimeout; /* seconds to consider peer_ni dead */
149 int *ksnd_enable_csum; /* enable check sum */
150 int *ksnd_inject_csum_error; /* set non-zero to inject checksum error */
151 int *ksnd_nonblk_zcack; /* always send zc-ack on non-blocking connection */
152 unsigned int *ksnd_zc_min_payload; /* minimum zero copy payload size */
153 int *ksnd_zc_recv; /* enable ZC receive (for Chelsio TOE) */
154 int *ksnd_zc_recv_min_nfrags; /* minimum # of fragments to enable ZC receive */
156 int *ksnd_irq_affinity; /* enable IRQ affinity? */
158 #ifdef SOCKNAL_BACKOFF
159 int *ksnd_backoff_init; /* initial TCP backoff */
160 int *ksnd_backoff_max; /* maximum TCP backoff */
162 #if SOCKNAL_VERSION_DEBUG
163 int *ksnd_protocol; /* protocol version */
165 int *ksnd_conns_per_peer; /* for typed mode, yields:
166 * 1 + 2*conns_per_peer total
168 * conns_per_peer total
173 __u64 ksnn_incarnation; /* my epoch */
174 spinlock_t ksnn_lock; /* serialise */
175 struct list_head ksnn_list; /* chain on global list */
176 int ksnn_npeers; /* # peers */
177 int ksnn_shutdown; /* shutting down? */
178 int ksnn_ninterfaces; /* IP interfaces */
179 struct ksock_interface ksnn_interfaces[LNET_INTERFACES_NUM];
183 #define SOCKNAL_CONND_TIMEOUT 120
184 /** reserved thread for accepting & creating new connd */
185 #define SOCKNAL_CONND_RESV 1
187 struct ksock_nal_data {
188 int ksnd_init; /* initialisation state */
189 int ksnd_nnets; /* # networks set up */
190 struct list_head ksnd_nets; /* list of nets */
191 /* stabilize peer_ni/conn ops */
192 rwlock_t ksnd_global_lock;
193 /* hash table of all my known peers */
194 struct list_head *ksnd_peers;
195 int ksnd_peer_hash_size; /* size of ksnd_peers */
197 int ksnd_nthreads; /* # live threads */
198 int ksnd_shuttingdown; /* tell threads to exit */
199 /* schedulers information */
200 struct ksock_sched **ksnd_schedulers;
202 atomic_t ksnd_nactive_txs; /* #active txs */
204 /* conns to close: reaper_lock*/
205 struct list_head ksnd_deathrow_conns;
206 /* conns to free: reaper_lock */
207 struct list_head ksnd_zombie_conns;
208 /* conns to retry: reaper_lock*/
209 struct list_head ksnd_enomem_conns;
210 /* reaper sleeps here */
211 wait_queue_head_t ksnd_reaper_waitq;
212 /* when reaper will wake */
213 time64_t ksnd_reaper_waketime;
215 spinlock_t ksnd_reaper_lock;
217 int ksnd_enomem_tx; /* test ENOMEM sender */
218 int ksnd_stall_tx; /* test sluggish sender */
219 int ksnd_stall_rx; /* test sluggish receiver */
221 /* incoming connection requests */
222 struct list_head ksnd_connd_connreqs;
223 /* routes waiting to be connected */
224 struct list_head ksnd_connd_routes;
225 /* connds sleep here */
226 wait_queue_head_t ksnd_connd_waitq;
227 /* # connds connecting */
228 int ksnd_connd_connecting;
229 /** time stamp of the last failed connecting attempt */
230 time64_t ksnd_connd_failed_stamp;
231 /** # starting connd */
232 unsigned ksnd_connd_starting;
233 /** time stamp of the last starting connd */
234 time64_t ksnd_connd_starting_stamp;
235 /** # running connd */
236 unsigned ksnd_connd_running;
238 spinlock_t ksnd_connd_lock;
240 /* list head for freed noop tx */
241 struct list_head ksnd_idle_noop_txs;
242 /* serialise, g_lock unsafe */
243 spinlock_t ksnd_tx_lock;
246 #define SOCKNAL_INIT_NOTHING 0
247 #define SOCKNAL_INIT_DATA 1
248 #define SOCKNAL_INIT_ALL 2
250 /* A packet just assembled for transmission is represented by 1 or more
251 * struct kvec fragments (the first frag contains the portals header),
252 * followed by 0 or more lnet_kiov_t fragments.
254 * On the receive side, initially 1 struct kvec fragment is posted for
255 * receive (the header). Once the header has been received, the payload is
256 * received into either struct kvec or lnet_kiov_t fragments, depending on
257 * what the header matched or whether the message needs forwarding. */
259 struct ksock_conn; /* forward ref */
260 struct ksock_peer; /* forward ref */
261 struct ksock_route; /* forward ref */
262 struct ksock_proto; /* forward ref */
264 struct ksock_tx { /* transmit packet */
265 struct list_head tx_list; /* queue on conn for transmission etc */
266 struct list_head tx_zc_list; /* queue on peer_ni for ZC request */
267 atomic_t tx_refcount; /* tx reference count */
268 int tx_nob; /* # packet bytes */
269 int tx_resid; /* residual bytes */
270 int tx_niov; /* # packet kvec frags */
271 struct kvec *tx_iov; /* packet kvec frags */
272 int tx_nkiov; /* # packet page frags */
273 unsigned short tx_zc_aborted; /* aborted ZC request */
274 unsigned short tx_zc_capable:1; /* payload is large enough for ZC */
275 unsigned short tx_zc_checked:1; /* Have I checked if I should ZC? */
276 unsigned short tx_nonblk:1; /* it's a non-blocking ACK */
277 lnet_kiov_t *tx_kiov; /* packet page frags */
278 struct ksock_conn *tx_conn; /* owning conn */
279 struct lnet_msg *tx_lnetmsg; /* lnet message for lnet_finalize() */
280 time64_t tx_deadline; /* when (in secs) tx times out */
281 struct ksock_msg tx_msg; /* socklnd message buffer */
282 int tx_desc_size; /* size of this descriptor */
283 enum lnet_msg_hstatus tx_hstatus; /* health status of tx */
286 struct kvec iov; /* virt hdr */
287 lnet_kiov_t kiov[0]; /* paged payload */
290 struct kvec iov[1]; /* virt hdr + payload */
295 #define KSOCK_NOOP_TX_SIZE ((int)offsetof(struct ksock_tx, tx_frags.paged.kiov[0]))
297 /* network zero copy callback descriptor embedded in struct ksock_tx */
299 /* space for the rx frag descriptors; we either read a single contiguous
300 * header, or up to LNET_MAX_IOV frags of payload of either type. */
301 union ksock_rxiovspace {
302 struct kvec iov[LNET_MAX_IOV];
303 lnet_kiov_t kiov[LNET_MAX_IOV];
306 #define SOCKNAL_RX_KSM_HEADER 1 /* reading ksock message header */
307 #define SOCKNAL_RX_LNET_HEADER 2 /* reading lnet message header */
308 #define SOCKNAL_RX_PARSE 3 /* Calling lnet_parse() */
309 #define SOCKNAL_RX_PARSE_WAIT 4 /* waiting to be told to read the body */
310 #define SOCKNAL_RX_LNET_PAYLOAD 5 /* reading lnet payload (to deliver here) */
311 #define SOCKNAL_RX_SLOP 6 /* skipping body */
314 struct ksock_peer_ni *ksnc_peer; /* owning peer_ni */
315 struct ksock_route *ksnc_route; /* owning route */
316 struct list_head ksnc_list; /* stash on peer_ni's conn list */
317 struct socket *ksnc_sock; /* actual socket */
318 void *ksnc_saved_data_ready; /* socket's original data_ready() callback */
319 void *ksnc_saved_write_space; /* socket's original write_space() callback */
320 atomic_t ksnc_conn_refcount; /* conn refcount */
321 atomic_t ksnc_sock_refcount; /* sock refcount */
322 struct ksock_sched *ksnc_scheduler; /* who schedules this connection */
323 __u32 ksnc_myipaddr; /* my IP */
324 __u32 ksnc_ipaddr; /* peer_ni's IP */
325 int ksnc_port; /* peer_ni's port */
326 signed int ksnc_type:3; /* type of connection,
327 * should be signed value */
328 unsigned int ksnc_closing:1; /* being shut down */
329 unsigned int ksnc_flip:1; /* flip or not, only for V2.x */
330 unsigned int ksnc_zc_capable:1; /* enable to ZC */
331 struct ksock_proto *ksnc_proto; /* protocol for the connection */
335 /* where I enq waiting input or a forwarding descriptor */
336 struct list_head ksnc_rx_list;
337 time64_t ksnc_rx_deadline; /* when (in seconds) receive times out */
338 __u8 ksnc_rx_started; /* started receiving a message */
339 __u8 ksnc_rx_ready; /* data ready to read */
340 __u8 ksnc_rx_scheduled;/* being progressed */
341 __u8 ksnc_rx_state; /* what is being read */
342 int ksnc_rx_nob_left; /* # bytes to next hdr/body */
343 int ksnc_rx_nob_wanted; /* bytes actually wanted */
344 int ksnc_rx_niov; /* # kvec frags */
345 struct kvec *ksnc_rx_iov; /* the kvec frags */
346 int ksnc_rx_nkiov; /* # page frags */
347 lnet_kiov_t *ksnc_rx_kiov; /* the page frags */
348 union ksock_rxiovspace ksnc_rx_iov_space;/* space for frag descriptors */
349 __u32 ksnc_rx_csum; /* partial checksum for incoming data */
350 struct lnet_msg *ksnc_lnet_msg; /* rx lnet_finalize arg*/
351 struct ksock_msg ksnc_msg; /* incoming message buffer:
352 * V2.x message takes the
354 * V1.x message is a bare
355 * struct lnet_hdr, it's stored
356 * in ksnc_msg.ksm_u.lnetmsg
359 /* where I enq waiting for output space */
360 struct list_head ksnc_tx_list;
361 /* packets waiting to be sent */
362 struct list_head ksnc_tx_queue;
363 /* next TX that can carry a LNet message or ZC-ACK */
364 struct ksock_tx *ksnc_tx_carrier;
365 /* when (in seconds) tx times out */
366 time64_t ksnc_tx_deadline;
367 /* send buffer marker */
370 atomic_t ksnc_tx_nob;
373 /* being progressed */
374 int ksnc_tx_scheduled;
375 /* time stamp of the last posted TX */
376 time64_t ksnc_tx_last_post;
379 #define SOCKNAL_CONN_COUNT_MAX_BITS 8 /* max conn count bits */
382 struct list_head ksnr_list; /* chain on peer_ni route list */
383 struct list_head ksnr_connd_list; /* chain on ksnr_connd_routes */
384 struct ksock_peer_ni *ksnr_peer; /* owning peer_ni */
385 atomic_t ksnr_refcount; /* # users */
386 time64_t ksnr_timeout; /* when (in secs) reconnection can happen next */
387 time64_t ksnr_retry_interval; /* how long between retries */
388 __u32 ksnr_myipaddr; /* my IP */
389 __u32 ksnr_ipaddr; /* IP address to connect to */
390 int ksnr_port; /* port to connect to */
391 unsigned int ksnr_scheduled:1; /* scheduled for attention */
392 unsigned int ksnr_connecting:1;/* connection establishment in progress */
393 unsigned int ksnr_connected:4; /* connections established by type */
394 unsigned int ksnr_deleted:1; /* been removed from peer_ni? */
395 unsigned int ksnr_ctrl_conn_count:1; /* # conns by type */
396 unsigned int ksnr_blki_conn_count:8;
397 unsigned int ksnr_blko_conn_count:8;
398 unsigned int ksnr_share_count; /* created explicitly? */
399 int ksnr_conn_count; /* # conns established by this route */
402 #define SOCKNAL_KEEPALIVE_PING 1 /* cookie for keepalive ping */
404 struct ksock_peer_ni {
405 struct list_head ksnp_list; /* stash on global peer_ni list */
406 time64_t ksnp_last_alive;/* when (in seconds) I was last alive */
407 struct lnet_process_id ksnp_id; /* who's on the other end(s) */
408 atomic_t ksnp_refcount; /* # users */
409 int ksnp_sharecount; /* lconf usage counter */
410 int ksnp_closing; /* being closed */
411 int ksnp_accepting;/* # passive connections pending */
412 int ksnp_error; /* errno on closing last conn */
413 __u64 ksnp_zc_next_cookie;/* ZC completion cookie */
414 __u64 ksnp_incarnation; /* latest known peer_ni incarnation */
415 struct ksock_proto *ksnp_proto; /* latest known peer_ni protocol */
416 struct list_head ksnp_conns; /* all active connections */
417 struct list_head ksnp_routes; /* routes */
418 struct list_head ksnp_tx_queue; /* waiting packets */
419 spinlock_t ksnp_lock; /* serialize, g_lock unsafe */
420 /* zero copy requests wait for ACK */
421 struct list_head ksnp_zc_req_list;
422 time64_t ksnp_send_keepalive; /* time to send keepalive */
423 struct lnet_ni *ksnp_ni; /* which network */
424 int ksnp_n_passive_ips; /* # of... */
425 __u32 ksnp_passive_ips[LNET_INTERFACES_NUM]; /* preferred local interfaces */
428 struct ksock_connreq {
429 /* stash on ksnd_connd_connreqs */
430 struct list_head ksncr_list;
432 struct lnet_ni *ksncr_ni;
433 /* accepted socket */
434 struct socket *ksncr_sock;
437 extern struct ksock_nal_data ksocknal_data;
438 extern struct ksock_tunables ksocknal_tunables;
440 #define SOCKNAL_MATCH_NO 0 /* TX can't match type of connection */
441 #define SOCKNAL_MATCH_YES 1 /* TX matches type of connection */
442 #define SOCKNAL_MATCH_MAY 2 /* TX can be sent on the connection, but not preferred */
445 int pro_version; /* version number of protocol */
446 int (*pro_send_hello)(struct ksock_conn *, struct ksock_hello_msg *); /* handshake function */
447 int (*pro_recv_hello)(struct ksock_conn *, struct ksock_hello_msg *, int);/* handshake function */
448 void (*pro_pack)(struct ksock_tx *); /* message pack */
449 void (*pro_unpack)(struct ksock_msg *); /* message unpack */
450 struct ksock_tx *(*pro_queue_tx_msg)(struct ksock_conn *, struct ksock_tx *); /* queue tx on the connection */
451 int (*pro_queue_tx_zcack)(struct ksock_conn *, struct ksock_tx *, __u64); /* queue ZC ack on the connection */
452 int (*pro_handle_zcreq)(struct ksock_conn *, __u64, int); /* handle ZC request */
453 int (*pro_handle_zcack)(struct ksock_conn *, __u64, __u64); /* handle ZC ACK */
454 int (*pro_match_tx)(struct ksock_conn *, struct ksock_tx *, int); /* msg type matches the connection type:
456 * return MATCH_NO : no
457 * return MATCH_YES : matching type
458 * return MATCH_MAY : can be backup */
461 extern struct ksock_proto ksocknal_protocol_v1x;
462 extern struct ksock_proto ksocknal_protocol_v2x;
463 extern struct ksock_proto ksocknal_protocol_v3x;
465 #define KSOCK_PROTO_V1_MAJOR LNET_PROTO_TCP_VERSION_MAJOR
466 #define KSOCK_PROTO_V1_MINOR LNET_PROTO_TCP_VERSION_MINOR
467 #define KSOCK_PROTO_V1 KSOCK_PROTO_V1_MAJOR
469 #ifndef CPU_MASK_NONE
470 #define CPU_MASK_NONE 0UL
473 static inline __u32 ksocknal_csum(__u32 crc, unsigned char const *p, size_t len)
476 return crc32_le(crc, p, len);
479 crc = ((crc + 0x100) & ~0xff) | ((crc + *p++) & 0xff) ;
486 ksocknal_route_mask(void)
488 if (!*ksocknal_tunables.ksnd_typed_conns)
489 return (1 << SOCKLND_CONN_ANY);
491 return ((1 << SOCKLND_CONN_CONTROL) |
492 (1 << SOCKLND_CONN_BULK_IN) |
493 (1 << SOCKLND_CONN_BULK_OUT));
496 static inline struct list_head *
497 ksocknal_nid2peerlist (lnet_nid_t nid)
499 unsigned int hash = ((unsigned int)nid) % ksocknal_data.ksnd_peer_hash_size;
501 return (&ksocknal_data.ksnd_peers [hash]);
505 ksocknal_conn_addref(struct ksock_conn *conn)
507 LASSERT(atomic_read(&conn->ksnc_conn_refcount) > 0);
508 atomic_inc(&conn->ksnc_conn_refcount);
511 extern void ksocknal_queue_zombie_conn(struct ksock_conn *conn);
512 extern void ksocknal_finalize_zcreq(struct ksock_conn *conn);
515 ksocknal_conn_decref(struct ksock_conn *conn)
517 LASSERT(atomic_read(&conn->ksnc_conn_refcount) > 0);
518 if (atomic_dec_and_test(&conn->ksnc_conn_refcount))
519 ksocknal_queue_zombie_conn(conn);
523 ksocknal_connsock_addref(struct ksock_conn *conn)
527 read_lock(&ksocknal_data.ksnd_global_lock);
528 if (!conn->ksnc_closing) {
529 LASSERT(atomic_read(&conn->ksnc_sock_refcount) > 0);
530 atomic_inc(&conn->ksnc_sock_refcount);
533 read_unlock(&ksocknal_data.ksnd_global_lock);
539 ksocknal_connsock_decref(struct ksock_conn *conn)
541 LASSERT(atomic_read(&conn->ksnc_sock_refcount) > 0);
542 if (atomic_dec_and_test(&conn->ksnc_sock_refcount)) {
543 LASSERT (conn->ksnc_closing);
544 sock_release(conn->ksnc_sock);
545 conn->ksnc_sock = NULL;
546 ksocknal_finalize_zcreq(conn);
551 ksocknal_tx_addref(struct ksock_tx *tx)
553 LASSERT(atomic_read(&tx->tx_refcount) > 0);
554 atomic_inc(&tx->tx_refcount);
557 extern void ksocknal_tx_prep(struct ksock_conn *, struct ksock_tx *tx);
558 extern void ksocknal_tx_done(struct lnet_ni *ni, struct ksock_tx *tx, int error);
561 ksocknal_tx_decref(struct ksock_tx *tx)
563 LASSERT(atomic_read(&tx->tx_refcount) > 0);
564 if (atomic_dec_and_test(&tx->tx_refcount))
565 ksocknal_tx_done(NULL, tx, 0);
569 ksocknal_route_addref(struct ksock_route *route)
571 LASSERT(atomic_read(&route->ksnr_refcount) > 0);
572 atomic_inc(&route->ksnr_refcount);
575 extern void ksocknal_destroy_route(struct ksock_route *route);
578 ksocknal_route_decref(struct ksock_route *route)
580 LASSERT(atomic_read(&route->ksnr_refcount) > 0);
581 if (atomic_dec_and_test(&route->ksnr_refcount))
582 ksocknal_destroy_route (route);
586 ksocknal_peer_addref(struct ksock_peer_ni *peer_ni)
588 LASSERT(atomic_read(&peer_ni->ksnp_refcount) > 0);
589 atomic_inc(&peer_ni->ksnp_refcount);
592 extern void ksocknal_destroy_peer(struct ksock_peer_ni *peer_ni);
595 ksocknal_peer_decref(struct ksock_peer_ni *peer_ni)
597 LASSERT (atomic_read (&peer_ni->ksnp_refcount) > 0);
598 if (atomic_dec_and_test(&peer_ni->ksnp_refcount))
599 ksocknal_destroy_peer(peer_ni);
602 static inline int ksocknal_conns_per_peer(void)
604 return *ksocknal_tunables.ksnd_conns_per_peer ?: 1;
607 int ksocknal_startup(struct lnet_ni *ni);
608 void ksocknal_shutdown(struct lnet_ni *ni);
609 int ksocknal_ctl(struct lnet_ni *ni, unsigned int cmd, void *arg);
610 int ksocknal_send(struct lnet_ni *ni, void *private, struct lnet_msg *lntmsg);
611 int ksocknal_recv(struct lnet_ni *ni, void *private, struct lnet_msg *lntmsg,
612 int delayed, unsigned int niov,
613 struct kvec *iov, lnet_kiov_t *kiov,
614 unsigned int offset, unsigned int mlen, unsigned int rlen);
615 int ksocknal_accept(struct lnet_ni *ni, struct socket *sock);
617 int ksocknal_add_peer(struct lnet_ni *ni, struct lnet_process_id id, __u32 ip,
619 struct ksock_peer_ni *ksocknal_find_peer_locked(struct lnet_ni *ni,
620 struct lnet_process_id id);
621 struct ksock_peer_ni *ksocknal_find_peer(struct lnet_ni *ni,
622 struct lnet_process_id id);
623 extern void ksocknal_peer_failed(struct ksock_peer_ni *peer_ni);
624 extern int ksocknal_create_conn(struct lnet_ni *ni, struct ksock_route *route,
625 struct socket *sock, int type);
626 extern void ksocknal_close_conn_locked(struct ksock_conn *conn, int why);
627 extern void ksocknal_terminate_conn(struct ksock_conn *conn);
628 extern void ksocknal_destroy_conn(struct ksock_conn *conn);
629 extern int ksocknal_close_peer_conns_locked(struct ksock_peer_ni *peer_ni,
630 __u32 ipaddr, int why);
631 extern int ksocknal_close_conn_and_siblings(struct ksock_conn *conn, int why);
632 int ksocknal_close_matching_conns(struct lnet_process_id id, __u32 ipaddr);
633 extern struct ksock_conn *ksocknal_find_conn_locked(struct ksock_peer_ni *peer_ni,
634 struct ksock_tx *tx, int nonblk);
636 extern int ksocknal_launch_packet(struct lnet_ni *ni, struct ksock_tx *tx,
637 struct lnet_process_id id);
638 extern struct ksock_tx *ksocknal_alloc_tx(int type, int size);
639 extern void ksocknal_free_tx(struct ksock_tx *tx);
640 extern struct ksock_tx *ksocknal_alloc_tx_noop(__u64 cookie, int nonblk);
641 extern void ksocknal_next_tx_carrier(struct ksock_conn *conn);
642 extern void ksocknal_queue_tx_locked(struct ksock_tx *tx, struct ksock_conn *conn);
643 extern void ksocknal_txlist_done(struct lnet_ni *ni, struct list_head *txlist,
645 extern void ksocknal_notify(struct lnet_ni *ni, lnet_nid_t gw_nid, int alive);
646 extern void ksocknal_query(struct lnet_ni *ni, lnet_nid_t nid, time64_t *when);
647 extern int ksocknal_thread_start(int (*fn)(void *arg), void *arg, char *name);
648 extern void ksocknal_thread_fini(void);
649 extern void ksocknal_launch_all_connections_locked(struct ksock_peer_ni *peer_ni);
650 extern struct ksock_route *ksocknal_find_connectable_route_locked(struct ksock_peer_ni *peer_ni);
651 extern struct ksock_route *ksocknal_find_connecting_route_locked(struct ksock_peer_ni *peer_ni);
652 extern int ksocknal_new_packet(struct ksock_conn *conn, int skip);
653 extern int ksocknal_scheduler(void *arg);
654 extern int ksocknal_connd(void *arg);
655 extern int ksocknal_reaper(void *arg);
656 int ksocknal_send_hello(struct lnet_ni *ni, struct ksock_conn *conn,
657 lnet_nid_t peer_nid, struct ksock_hello_msg *hello);
658 int ksocknal_recv_hello(struct lnet_ni *ni, struct ksock_conn *conn,
659 struct ksock_hello_msg *hello,
660 struct lnet_process_id *id,
662 extern void ksocknal_read_callback(struct ksock_conn *conn);
663 extern void ksocknal_write_callback(struct ksock_conn *conn);
665 extern int ksocknal_lib_zc_capable(struct ksock_conn *conn);
666 extern void ksocknal_lib_save_callback(struct socket *sock, struct ksock_conn *conn);
667 extern void ksocknal_lib_set_callback(struct socket *sock, struct ksock_conn *conn);
668 extern void ksocknal_lib_reset_callback(struct socket *sock,
669 struct ksock_conn *conn);
670 extern void ksocknal_lib_push_conn(struct ksock_conn *conn);
671 extern int ksocknal_lib_get_conn_addrs(struct ksock_conn *conn);
672 extern int ksocknal_lib_setup_sock(struct socket *so);
673 extern int ksocknal_lib_send_iov(struct ksock_conn *conn, struct ksock_tx *tx,
674 struct kvec *scratch_iov);
675 extern int ksocknal_lib_send_kiov(struct ksock_conn *conn, struct ksock_tx *tx,
676 struct kvec *scratch_iov);
677 extern void ksocknal_lib_eager_ack(struct ksock_conn *conn);
678 extern int ksocknal_lib_recv_iov(struct ksock_conn *conn,
679 struct kvec *scratchiov);
680 extern int ksocknal_lib_recv_kiov(struct ksock_conn *conn, struct page **pages,
681 struct kvec *scratchiov);
682 extern int ksocknal_lib_get_conn_tunables(struct ksock_conn *conn, int *txmem,
683 int *rxmem, int *nagle);
685 extern int ksocknal_tunables_init(void);
687 extern void ksocknal_lib_csum_tx(struct ksock_tx *tx);
689 extern int ksocknal_lib_memory_pressure(struct ksock_conn *conn);
690 extern int ksocknal_lib_bind_thread_to_cpu(int id);
692 #endif /* _SOCKLND_SOCKLND_H_ */