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 struct ksock_sched_info;
93 struct ksock_sched { /* per scheduler state */
94 spinlock_t kss_lock; /* serialise */
95 struct list_head kss_rx_conns; /* conn waiting to be read */
96 /* conn waiting to be written */
97 struct list_head kss_tx_conns;
98 /* zombie noop tx list */
99 struct list_head kss_zombie_noop_txs;
100 wait_queue_head_t kss_waitq; /* where scheduler sleeps */
101 /* # connections assigned to this scheduler */
103 struct ksock_sched_info *kss_info; /* owner of it */
104 #if !SOCKNAL_SINGLE_FRAG_RX
105 struct page *kss_rx_scratch_pgs[LNET_MAX_IOV];
107 #if !SOCKNAL_SINGLE_FRAG_TX || !SOCKNAL_SINGLE_FRAG_RX
108 struct kvec kss_scratch_iov[LNET_MAX_IOV];
112 struct ksock_sched_info {
113 int ksi_nthreads_max; /* max allowed threads */
114 int ksi_nthreads; /* number of threads */
115 int ksi_cpt; /* CPT id */
116 struct ksock_sched *ksi_scheds; /* array of schedulers */
119 #define KSOCK_CPT_SHIFT 16
120 #define KSOCK_THREAD_ID(cpt, sid) (((cpt) << KSOCK_CPT_SHIFT) | (sid))
121 #define KSOCK_THREAD_CPT(id) ((id) >> KSOCK_CPT_SHIFT)
122 #define KSOCK_THREAD_SID(id) ((id) & ((1UL << KSOCK_CPT_SHIFT) - 1))
124 struct ksock_interface { /* in-use interface */
125 __u32 ksni_ipaddr; /* interface's IP address */
126 __u32 ksni_netmask; /* interface's network mask */
127 int ksni_nroutes; /* # routes using (active) */
128 int ksni_npeers; /* # peers using (passive) */
129 char ksni_name[IFNAMSIZ]; /* interface name */
132 struct ksock_tunables {
133 /* "stuck" socket timeout (seconds) */
135 /* # scheduler threads in each pool while starting */
137 int *ksnd_nconnds; /* # connection daemons */
138 int *ksnd_nconnds_max; /* max # connection daemons */
139 int *ksnd_min_reconnectms; /* first connection retry after (ms)... */
140 int *ksnd_max_reconnectms; /* ...exponentially increasing to this */
141 int *ksnd_eager_ack; /* make TCP ack eagerly? */
142 int *ksnd_typed_conns; /* drive sockets by type? */
143 int *ksnd_min_bulk; /* smallest "large" message */
144 int *ksnd_tx_buffer_size; /* socket tx buffer size */
145 int *ksnd_rx_buffer_size; /* socket rx buffer size */
146 int *ksnd_nagle; /* enable NAGLE? */
147 int *ksnd_round_robin; /* round robin for multiple interfaces */
148 int *ksnd_keepalive; /* # secs for sending keepalive NOOP */
149 int *ksnd_keepalive_idle; /* # idle secs before 1st probe */
150 int *ksnd_keepalive_count; /* # probes */
151 int *ksnd_keepalive_intvl; /* time between probes */
152 int *ksnd_credits; /* # concurrent sends */
153 int *ksnd_peertxcredits; /* # concurrent sends to 1 peer_ni */
154 int *ksnd_peerrtrcredits; /* # per-peer_ni router buffer credits */
155 int *ksnd_peertimeout; /* seconds to consider peer_ni dead */
156 int *ksnd_enable_csum; /* enable check sum */
157 int *ksnd_inject_csum_error; /* set non-zero to inject checksum error */
158 int *ksnd_nonblk_zcack; /* always send zc-ack on non-blocking connection */
159 unsigned int *ksnd_zc_min_payload; /* minimum zero copy payload size */
160 int *ksnd_zc_recv; /* enable ZC receive (for Chelsio TOE) */
161 int *ksnd_zc_recv_min_nfrags; /* minimum # of fragments to enable ZC receive */
163 int *ksnd_irq_affinity; /* enable IRQ affinity? */
165 #ifdef SOCKNAL_BACKOFF
166 int *ksnd_backoff_init; /* initial TCP backoff */
167 int *ksnd_backoff_max; /* maximum TCP backoff */
169 #if SOCKNAL_VERSION_DEBUG
170 int *ksnd_protocol; /* protocol version */
175 __u64 ksnn_incarnation; /* my epoch */
176 spinlock_t ksnn_lock; /* serialise */
177 struct list_head ksnn_list; /* chain on global list */
178 int ksnn_npeers; /* # peers */
179 int ksnn_shutdown; /* shutting down? */
180 int ksnn_ninterfaces; /* IP interfaces */
181 struct ksock_interface ksnn_interfaces[LNET_INTERFACES_NUM];
185 #define SOCKNAL_CONND_TIMEOUT 120
186 /** reserved thread for accepting & creating new connd */
187 #define SOCKNAL_CONND_RESV 1
189 struct ksock_nal_data {
190 int ksnd_init; /* initialisation state */
191 int ksnd_nnets; /* # networks set up */
192 struct list_head ksnd_nets; /* list of nets */
193 /* stabilize peer_ni/conn ops */
194 rwlock_t ksnd_global_lock;
195 /* hash table of all my known peers */
196 struct list_head *ksnd_peers;
197 int ksnd_peer_hash_size; /* size of ksnd_peers */
199 int ksnd_nthreads; /* # live threads */
200 int ksnd_shuttingdown; /* tell threads to exit */
201 /* schedulers information */
202 struct ksock_sched_info **ksnd_sched_info;
204 atomic_t ksnd_nactive_txs; /* #active txs */
206 /* conns to close: reaper_lock*/
207 struct list_head ksnd_deathrow_conns;
208 /* conns to free: reaper_lock */
209 struct list_head ksnd_zombie_conns;
210 /* conns to retry: reaper_lock*/
211 struct list_head ksnd_enomem_conns;
212 /* reaper sleeps here */
213 wait_queue_head_t ksnd_reaper_waitq;
214 /* when reaper will wake */
215 time64_t ksnd_reaper_waketime;
217 spinlock_t ksnd_reaper_lock;
219 int ksnd_enomem_tx; /* test ENOMEM sender */
220 int ksnd_stall_tx; /* test sluggish sender */
221 int ksnd_stall_rx; /* test sluggish receiver */
223 /* incoming connection requests */
224 struct list_head ksnd_connd_connreqs;
225 /* routes waiting to be connected */
226 struct list_head ksnd_connd_routes;
227 /* connds sleep here */
228 wait_queue_head_t ksnd_connd_waitq;
229 /* # connds connecting */
230 int ksnd_connd_connecting;
231 /** time stamp of the last failed connecting attempt */
232 time64_t ksnd_connd_failed_stamp;
233 /** # starting connd */
234 unsigned ksnd_connd_starting;
235 /** time stamp of the last starting connd */
236 time64_t ksnd_connd_starting_stamp;
237 /** # running connd */
238 unsigned ksnd_connd_running;
240 spinlock_t ksnd_connd_lock;
242 /* list head for freed noop tx */
243 struct list_head ksnd_idle_noop_txs;
244 /* serialise, g_lock unsafe */
245 spinlock_t ksnd_tx_lock;
248 #define SOCKNAL_INIT_NOTHING 0
249 #define SOCKNAL_INIT_DATA 1
250 #define SOCKNAL_INIT_ALL 2
252 /* A packet just assembled for transmission is represented by 1 or more
253 * struct kvec fragments (the first frag contains the portals header),
254 * followed by 0 or more lnet_kiov_t fragments.
256 * On the receive side, initially 1 struct kvec fragment is posted for
257 * receive (the header). Once the header has been received, the payload is
258 * received into either struct kvec or lnet_kiov_t fragments, depending on
259 * what the header matched or whether the message needs forwarding. */
261 struct ksock_conn; /* forward ref */
262 struct ksock_peer; /* forward ref */
263 struct ksock_route; /* forward ref */
264 struct ksock_proto; /* forward ref */
266 struct ksock_tx { /* transmit packet */
267 struct list_head tx_list; /* queue on conn for transmission etc */
268 struct list_head tx_zc_list; /* queue on peer_ni for ZC request */
269 atomic_t tx_refcount; /* tx reference count */
270 int tx_nob; /* # packet bytes */
271 int tx_resid; /* residual bytes */
272 int tx_niov; /* # packet kvec frags */
273 struct kvec *tx_iov; /* packet kvec frags */
274 int tx_nkiov; /* # packet page frags */
275 unsigned short tx_zc_aborted; /* aborted ZC request */
276 unsigned short tx_zc_capable:1; /* payload is large enough for ZC */
277 unsigned short tx_zc_checked:1; /* Have I checked if I should ZC? */
278 unsigned short tx_nonblk:1; /* it's a non-blocking ACK */
279 lnet_kiov_t *tx_kiov; /* packet page frags */
280 struct ksock_conn *tx_conn; /* owning conn */
281 struct lnet_msg *tx_lnetmsg; /* lnet message for lnet_finalize() */
282 time64_t tx_deadline; /* when (in secs) tx times out */
283 struct ksock_msg tx_msg; /* socklnd message buffer */
284 int tx_desc_size; /* size of this descriptor */
285 enum lnet_msg_hstatus tx_hstatus; /* health status of tx */
288 struct kvec iov; /* virt hdr */
289 lnet_kiov_t kiov[0]; /* paged payload */
292 struct kvec iov[1]; /* virt hdr + payload */
297 #define KSOCK_NOOP_TX_SIZE ((int)offsetof(struct ksock_tx, tx_frags.paged.kiov[0]))
299 /* network zero copy callback descriptor embedded in struct ksock_tx */
301 /* space for the rx frag descriptors; we either read a single contiguous
302 * header, or up to LNET_MAX_IOV frags of payload of either type. */
303 union ksock_rxiovspace {
304 struct kvec iov[LNET_MAX_IOV];
305 lnet_kiov_t kiov[LNET_MAX_IOV];
308 #define SOCKNAL_RX_KSM_HEADER 1 /* reading ksock message header */
309 #define SOCKNAL_RX_LNET_HEADER 2 /* reading lnet message header */
310 #define SOCKNAL_RX_PARSE 3 /* Calling lnet_parse() */
311 #define SOCKNAL_RX_PARSE_WAIT 4 /* waiting to be told to read the body */
312 #define SOCKNAL_RX_LNET_PAYLOAD 5 /* reading lnet payload (to deliver here) */
313 #define SOCKNAL_RX_SLOP 6 /* skipping body */
316 struct ksock_peer_ni *ksnc_peer; /* owning peer_ni */
317 struct ksock_route *ksnc_route; /* owning route */
318 struct list_head ksnc_list; /* stash on peer_ni's conn list */
319 struct socket *ksnc_sock; /* actual socket */
320 void *ksnc_saved_data_ready; /* socket's original data_ready() callback */
321 void *ksnc_saved_write_space; /* socket's original write_space() callback */
322 atomic_t ksnc_conn_refcount; /* conn refcount */
323 atomic_t ksnc_sock_refcount; /* sock refcount */
324 struct ksock_sched *ksnc_scheduler; /* who schedules this connection */
325 __u32 ksnc_myipaddr; /* my IP */
326 __u32 ksnc_ipaddr; /* peer_ni's IP */
327 int ksnc_port; /* peer_ni's port */
328 signed int ksnc_type:3; /* type of connection,
329 * should be signed value */
330 unsigned int ksnc_closing:1; /* being shut down */
331 unsigned int ksnc_flip:1; /* flip or not, only for V2.x */
332 unsigned int ksnc_zc_capable:1; /* enable to ZC */
333 struct ksock_proto *ksnc_proto; /* protocol for the connection */
337 /* where I enq waiting input or a forwarding descriptor */
338 struct list_head ksnc_rx_list;
339 time64_t ksnc_rx_deadline; /* when (in seconds) receive times out */
340 __u8 ksnc_rx_started; /* started receiving a message */
341 __u8 ksnc_rx_ready; /* data ready to read */
342 __u8 ksnc_rx_scheduled;/* being progressed */
343 __u8 ksnc_rx_state; /* what is being read */
344 int ksnc_rx_nob_left; /* # bytes to next hdr/body */
345 int ksnc_rx_nob_wanted; /* bytes actually wanted */
346 int ksnc_rx_niov; /* # kvec frags */
347 struct kvec *ksnc_rx_iov; /* the kvec frags */
348 int ksnc_rx_nkiov; /* # page frags */
349 lnet_kiov_t *ksnc_rx_kiov; /* the page frags */
350 union ksock_rxiovspace ksnc_rx_iov_space;/* space for frag descriptors */
351 __u32 ksnc_rx_csum; /* partial checksum for incoming data */
352 struct lnet_msg *ksnc_lnet_msg; /* rx lnet_finalize arg*/
353 struct ksock_msg ksnc_msg; /* incoming message buffer:
354 * V2.x message takes the
356 * V1.x message is a bare
357 * struct lnet_hdr, it's stored
358 * in ksnc_msg.ksm_u.lnetmsg
361 /* where I enq waiting for output space */
362 struct list_head ksnc_tx_list;
363 /* packets waiting to be sent */
364 struct list_head ksnc_tx_queue;
365 /* next TX that can carry a LNet message or ZC-ACK */
366 struct ksock_tx *ksnc_tx_carrier;
367 /* when (in seconds) tx times out */
368 time64_t ksnc_tx_deadline;
369 /* send buffer marker */
372 atomic_t ksnc_tx_nob;
375 /* being progressed */
376 int ksnc_tx_scheduled;
377 /* time stamp of the last posted TX */
378 time64_t ksnc_tx_last_post;
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_share_count; /* created explicitly? */
396 int ksnr_conn_count; /* # conns established by this route */
399 #define SOCKNAL_KEEPALIVE_PING 1 /* cookie for keepalive ping */
401 struct ksock_peer_ni {
402 struct list_head ksnp_list; /* stash on global peer_ni list */
403 time64_t ksnp_last_alive;/* when (in seconds) I was last alive */
404 struct lnet_process_id ksnp_id; /* who's on the other end(s) */
405 atomic_t ksnp_refcount; /* # users */
406 int ksnp_sharecount; /* lconf usage counter */
407 int ksnp_closing; /* being closed */
408 int ksnp_accepting;/* # passive connections pending */
409 int ksnp_error; /* errno on closing last conn */
410 __u64 ksnp_zc_next_cookie;/* ZC completion cookie */
411 __u64 ksnp_incarnation; /* latest known peer_ni incarnation */
412 struct ksock_proto *ksnp_proto; /* latest known peer_ni protocol */
413 struct list_head ksnp_conns; /* all active connections */
414 struct list_head ksnp_routes; /* routes */
415 struct list_head ksnp_tx_queue; /* waiting packets */
416 spinlock_t ksnp_lock; /* serialize, g_lock unsafe */
417 /* zero copy requests wait for ACK */
418 struct list_head ksnp_zc_req_list;
419 time64_t ksnp_send_keepalive; /* time to send keepalive */
420 struct lnet_ni *ksnp_ni; /* which network */
421 int ksnp_n_passive_ips; /* # of... */
422 __u32 ksnp_passive_ips[LNET_INTERFACES_NUM]; /* preferred local interfaces */
425 struct ksock_connreq {
426 /* stash on ksnd_connd_connreqs */
427 struct list_head ksncr_list;
429 struct lnet_ni *ksncr_ni;
430 /* accepted socket */
431 struct socket *ksncr_sock;
434 extern struct ksock_nal_data ksocknal_data;
435 extern struct ksock_tunables ksocknal_tunables;
437 #define SOCKNAL_MATCH_NO 0 /* TX can't match type of connection */
438 #define SOCKNAL_MATCH_YES 1 /* TX matches type of connection */
439 #define SOCKNAL_MATCH_MAY 2 /* TX can be sent on the connection, but not preferred */
442 int pro_version; /* version number of protocol */
443 int (*pro_send_hello)(struct ksock_conn *, struct ksock_hello_msg *); /* handshake function */
444 int (*pro_recv_hello)(struct ksock_conn *, struct ksock_hello_msg *, int);/* handshake function */
445 void (*pro_pack)(struct ksock_tx *); /* message pack */
446 void (*pro_unpack)(struct ksock_msg *); /* message unpack */
447 struct ksock_tx *(*pro_queue_tx_msg)(struct ksock_conn *, struct ksock_tx *); /* queue tx on the connection */
448 int (*pro_queue_tx_zcack)(struct ksock_conn *, struct ksock_tx *, __u64); /* queue ZC ack on the connection */
449 int (*pro_handle_zcreq)(struct ksock_conn *, __u64, int); /* handle ZC request */
450 int (*pro_handle_zcack)(struct ksock_conn *, __u64, __u64); /* handle ZC ACK */
451 int (*pro_match_tx)(struct ksock_conn *, struct ksock_tx *, int); /* msg type matches the connection type:
453 * return MATCH_NO : no
454 * return MATCH_YES : matching type
455 * return MATCH_MAY : can be backup */
458 extern struct ksock_proto ksocknal_protocol_v1x;
459 extern struct ksock_proto ksocknal_protocol_v2x;
460 extern struct ksock_proto ksocknal_protocol_v3x;
462 #define KSOCK_PROTO_V1_MAJOR LNET_PROTO_TCP_VERSION_MAJOR
463 #define KSOCK_PROTO_V1_MINOR LNET_PROTO_TCP_VERSION_MINOR
464 #define KSOCK_PROTO_V1 KSOCK_PROTO_V1_MAJOR
466 #ifndef CPU_MASK_NONE
467 #define CPU_MASK_NONE 0UL
470 static inline __u32 ksocknal_csum(__u32 crc, unsigned char const *p, size_t len)
473 return crc32_le(crc, p, len);
476 crc = ((crc + 0x100) & ~0xff) | ((crc + *p++) & 0xff) ;
483 ksocknal_route_mask(void)
485 if (!*ksocknal_tunables.ksnd_typed_conns)
486 return (1 << SOCKLND_CONN_ANY);
488 return ((1 << SOCKLND_CONN_CONTROL) |
489 (1 << SOCKLND_CONN_BULK_IN) |
490 (1 << SOCKLND_CONN_BULK_OUT));
493 static inline struct list_head *
494 ksocknal_nid2peerlist (lnet_nid_t nid)
496 unsigned int hash = ((unsigned int)nid) % ksocknal_data.ksnd_peer_hash_size;
498 return (&ksocknal_data.ksnd_peers [hash]);
502 ksocknal_conn_addref(struct ksock_conn *conn)
504 LASSERT(atomic_read(&conn->ksnc_conn_refcount) > 0);
505 atomic_inc(&conn->ksnc_conn_refcount);
508 extern void ksocknal_queue_zombie_conn(struct ksock_conn *conn);
509 extern void ksocknal_finalize_zcreq(struct ksock_conn *conn);
512 ksocknal_conn_decref(struct ksock_conn *conn)
514 LASSERT(atomic_read(&conn->ksnc_conn_refcount) > 0);
515 if (atomic_dec_and_test(&conn->ksnc_conn_refcount))
516 ksocknal_queue_zombie_conn(conn);
520 ksocknal_connsock_addref(struct ksock_conn *conn)
524 read_lock(&ksocknal_data.ksnd_global_lock);
525 if (!conn->ksnc_closing) {
526 LASSERT(atomic_read(&conn->ksnc_sock_refcount) > 0);
527 atomic_inc(&conn->ksnc_sock_refcount);
530 read_unlock(&ksocknal_data.ksnd_global_lock);
536 ksocknal_connsock_decref(struct ksock_conn *conn)
538 LASSERT(atomic_read(&conn->ksnc_sock_refcount) > 0);
539 if (atomic_dec_and_test(&conn->ksnc_sock_refcount)) {
540 LASSERT (conn->ksnc_closing);
541 sock_release(conn->ksnc_sock);
542 conn->ksnc_sock = NULL;
543 ksocknal_finalize_zcreq(conn);
548 ksocknal_tx_addref(struct ksock_tx *tx)
550 LASSERT(atomic_read(&tx->tx_refcount) > 0);
551 atomic_inc(&tx->tx_refcount);
554 extern void ksocknal_tx_prep(struct ksock_conn *, struct ksock_tx *tx);
555 extern void ksocknal_tx_done(struct lnet_ni *ni, struct ksock_tx *tx, int error);
558 ksocknal_tx_decref(struct ksock_tx *tx)
560 LASSERT(atomic_read(&tx->tx_refcount) > 0);
561 if (atomic_dec_and_test(&tx->tx_refcount))
562 ksocknal_tx_done(NULL, tx, 0);
566 ksocknal_route_addref(struct ksock_route *route)
568 LASSERT(atomic_read(&route->ksnr_refcount) > 0);
569 atomic_inc(&route->ksnr_refcount);
572 extern void ksocknal_destroy_route(struct ksock_route *route);
575 ksocknal_route_decref(struct ksock_route *route)
577 LASSERT(atomic_read(&route->ksnr_refcount) > 0);
578 if (atomic_dec_and_test(&route->ksnr_refcount))
579 ksocknal_destroy_route (route);
583 ksocknal_peer_addref(struct ksock_peer_ni *peer_ni)
585 LASSERT(atomic_read(&peer_ni->ksnp_refcount) > 0);
586 atomic_inc(&peer_ni->ksnp_refcount);
589 extern void ksocknal_destroy_peer(struct ksock_peer_ni *peer_ni);
592 ksocknal_peer_decref(struct ksock_peer_ni *peer_ni)
594 LASSERT (atomic_read (&peer_ni->ksnp_refcount) > 0);
595 if (atomic_dec_and_test(&peer_ni->ksnp_refcount))
596 ksocknal_destroy_peer(peer_ni);
599 int ksocknal_startup(struct lnet_ni *ni);
600 void ksocknal_shutdown(struct lnet_ni *ni);
601 int ksocknal_ctl(struct lnet_ni *ni, unsigned int cmd, void *arg);
602 int ksocknal_send(struct lnet_ni *ni, void *private, struct lnet_msg *lntmsg);
603 int ksocknal_recv(struct lnet_ni *ni, void *private, struct lnet_msg *lntmsg,
604 int delayed, unsigned int niov,
605 struct kvec *iov, lnet_kiov_t *kiov,
606 unsigned int offset, unsigned int mlen, unsigned int rlen);
607 int ksocknal_accept(struct lnet_ni *ni, struct socket *sock);
609 int ksocknal_add_peer(struct lnet_ni *ni, struct lnet_process_id id, __u32 ip,
611 struct ksock_peer_ni *ksocknal_find_peer_locked(struct lnet_ni *ni,
612 struct lnet_process_id id);
613 struct ksock_peer_ni *ksocknal_find_peer(struct lnet_ni *ni,
614 struct lnet_process_id id);
615 extern void ksocknal_peer_failed(struct ksock_peer_ni *peer_ni);
616 extern int ksocknal_create_conn(struct lnet_ni *ni, struct ksock_route *route,
617 struct socket *sock, int type);
618 extern void ksocknal_close_conn_locked(struct ksock_conn *conn, int why);
619 extern void ksocknal_terminate_conn(struct ksock_conn *conn);
620 extern void ksocknal_destroy_conn(struct ksock_conn *conn);
621 extern int ksocknal_close_peer_conns_locked(struct ksock_peer_ni *peer_ni,
622 __u32 ipaddr, int why);
623 extern int ksocknal_close_conn_and_siblings(struct ksock_conn *conn, int why);
624 int ksocknal_close_matching_conns(struct lnet_process_id id, __u32 ipaddr);
625 extern struct ksock_conn *ksocknal_find_conn_locked(struct ksock_peer_ni *peer_ni,
626 struct ksock_tx *tx, int nonblk);
628 extern int ksocknal_launch_packet(struct lnet_ni *ni, struct ksock_tx *tx,
629 struct lnet_process_id id);
630 extern struct ksock_tx *ksocknal_alloc_tx(int type, int size);
631 extern void ksocknal_free_tx(struct ksock_tx *tx);
632 extern struct ksock_tx *ksocknal_alloc_tx_noop(__u64 cookie, int nonblk);
633 extern void ksocknal_next_tx_carrier(struct ksock_conn *conn);
634 extern void ksocknal_queue_tx_locked(struct ksock_tx *tx, struct ksock_conn *conn);
635 extern void ksocknal_txlist_done(struct lnet_ni *ni, struct list_head *txlist,
637 extern void ksocknal_notify(struct lnet_ni *ni, lnet_nid_t gw_nid, int alive);
638 extern void ksocknal_query(struct lnet_ni *ni, lnet_nid_t nid, time64_t *when);
639 extern int ksocknal_thread_start(int (*fn)(void *arg), void *arg, char *name);
640 extern void ksocknal_thread_fini(void);
641 extern void ksocknal_launch_all_connections_locked(struct ksock_peer_ni *peer_ni);
642 extern struct ksock_route *ksocknal_find_connectable_route_locked(struct ksock_peer_ni *peer_ni);
643 extern struct ksock_route *ksocknal_find_connecting_route_locked(struct ksock_peer_ni *peer_ni);
644 extern int ksocknal_new_packet(struct ksock_conn *conn, int skip);
645 extern int ksocknal_scheduler(void *arg);
646 extern int ksocknal_connd(void *arg);
647 extern int ksocknal_reaper(void *arg);
648 int ksocknal_send_hello(struct lnet_ni *ni, struct ksock_conn *conn,
649 lnet_nid_t peer_nid, struct ksock_hello_msg *hello);
650 int ksocknal_recv_hello(struct lnet_ni *ni, struct ksock_conn *conn,
651 struct ksock_hello_msg *hello,
652 struct lnet_process_id *id,
654 extern void ksocknal_read_callback(struct ksock_conn *conn);
655 extern void ksocknal_write_callback(struct ksock_conn *conn);
657 extern int ksocknal_lib_zc_capable(struct ksock_conn *conn);
658 extern void ksocknal_lib_save_callback(struct socket *sock, struct ksock_conn *conn);
659 extern void ksocknal_lib_set_callback(struct socket *sock, struct ksock_conn *conn);
660 extern void ksocknal_lib_reset_callback(struct socket *sock,
661 struct ksock_conn *conn);
662 extern void ksocknal_lib_push_conn(struct ksock_conn *conn);
663 extern int ksocknal_lib_get_conn_addrs(struct ksock_conn *conn);
664 extern int ksocknal_lib_setup_sock(struct socket *so);
665 extern int ksocknal_lib_send_iov(struct ksock_conn *conn, struct ksock_tx *tx);
666 extern int ksocknal_lib_send_kiov(struct ksock_conn *conn, struct ksock_tx *tx);
667 extern void ksocknal_lib_eager_ack(struct ksock_conn *conn);
668 extern int ksocknal_lib_recv_iov(struct ksock_conn *conn);
669 extern int ksocknal_lib_recv_kiov(struct ksock_conn *conn);
670 extern int ksocknal_lib_get_conn_tunables(struct ksock_conn *conn, int *txmem,
671 int *rxmem, int *nagle);
673 extern int ksocknal_tunables_init(void);
675 extern void ksocknal_lib_csum_tx(struct ksock_tx *tx);
677 extern int ksocknal_lib_memory_pressure(struct ksock_conn *conn);
678 extern int ksocknal_lib_bind_thread_to_cpu(int id);
680 #endif /* _SOCKLND_SOCKLND_H_ */