2 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
4 * Copyright (c) 2011, 2013, Intel Corporation.
6 * Author: Zach Brown <zab@zabbo.net>
7 * Author: Peter J. Braam <braam@clusterfs.com>
8 * Author: Phil Schwan <phil@clusterfs.com>
9 * Author: Eric Barton <eric@bartonsoftware.com>
11 * This file is part of Lustre, http://www.lustre.org
13 * Portals is free software; you can redistribute it and/or
14 * modify it under the terms of version 2 of the GNU General Public
15 * License as published by the Free Software Foundation.
17 * Portals is distributed in the hope that it will be useful,
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
20 * GNU General Public License for more details.
22 * You should have received a copy of the GNU General Public License
23 * along with Portals; if not, write to the Free Software
24 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
28 #ifndef _SOCKLND_SOCKLND_H_
29 #define _SOCKLND_SOCKLND_H_
31 #define DEBUG_PORTAL_ALLOC
32 #define DEBUG_SUBSYSTEM S_LND
34 #ifdef HAVE_COMPAT_RDMA
35 #include <linux/compat-2.6.h>
37 #include <linux/crc32.h>
38 #include <linux/errno.h>
40 #include <linux/init.h>
41 #include <linux/kernel.h>
42 #include <linux/kthread.h>
43 #include <linux/kmod.h>
44 #include <linux/list.h>
46 #include <linux/module.h>
47 #include <linux/stat.h>
48 #include <linux/string.h>
49 #include <linux/syscalls.h>
50 #include <linux/sysctl.h>
51 #include <linux/uio.h>
52 #include <linux/unistd.h>
56 #include <libcfs/libcfs.h>
57 #include <lnet/lnet.h>
58 #include <lnet/lib-lnet.h>
59 #include <lnet/socklnd.h>
61 #ifdef HAVE_TCP_SENDPAGE_USE_SOCKET
62 # define cfs_tcp_sendpage(sk, page, offset, size, flags) \
63 tcp_sendpage((sk)->sk_socket, page, offset, size, flags)
64 #else /* !HAVE_TCP_SENDPAGE_USE_SOCKET */
65 # define cfs_tcp_sendpage(sk, page, offset, size, flags) \
66 tcp_sendpage(sk, page, offset, size, flags)
67 #endif /* HAVE_TCP_SENDPAGE_USE_SOCKET */
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 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 CFS_TICK /* jiffies 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 typedef struct /* per scheduler state */
95 spinlock_t kss_lock; /* serialise */
96 struct list_head kss_rx_conns; /* conn waiting to be read */
97 /* conn waiting to be written */
98 struct list_head kss_tx_conns;
99 /* zombie noop tx list */
100 struct list_head kss_zombie_noop_txs;
101 wait_queue_head_t kss_waitq; /* where scheduler sleeps */
102 /* # connections assigned to this scheduler */
104 struct ksock_sched_info *kss_info; /* owner of it */
105 #if !SOCKNAL_SINGLE_FRAG_RX
106 struct page *kss_rx_scratch_pgs[LNET_MAX_IOV];
108 #if !SOCKNAL_SINGLE_FRAG_TX || !SOCKNAL_SINGLE_FRAG_RX
109 struct kvec kss_scratch_iov[LNET_MAX_IOV];
113 struct ksock_sched_info {
114 int ksi_nthreads_max; /* max allowed threads */
115 int ksi_nthreads; /* number of threads */
116 int ksi_cpt; /* CPT id */
117 ksock_sched_t *ksi_scheds; /* array of schedulers */
120 #define KSOCK_CPT_SHIFT 16
121 #define KSOCK_THREAD_ID(cpt, sid) (((cpt) << KSOCK_CPT_SHIFT) | (sid))
122 #define KSOCK_THREAD_CPT(id) ((id) >> KSOCK_CPT_SHIFT)
123 #define KSOCK_THREAD_SID(id) ((id) & ((1UL << KSOCK_CPT_SHIFT) - 1))
125 typedef struct /* in-use interface */
127 __u32 ksni_ipaddr; /* interface's IP address */
128 __u32 ksni_netmask; /* interface's network mask */
129 int ksni_nroutes; /* # routes using (active) */
130 int ksni_npeers; /* # peers using (passive) */
131 char ksni_name[IFNAMSIZ]; /* interface name */
136 /* "stuck" socket timeout (seconds) */
138 /* # scheduler threads in each pool while starting */
140 int *ksnd_nconnds; /* # connection daemons */
141 int *ksnd_nconnds_max; /* max # connection daemons */
142 int *ksnd_min_reconnectms; /* first connection retry after (ms)... */
143 int *ksnd_max_reconnectms; /* ...exponentially increasing to this */
144 int *ksnd_eager_ack; /* make TCP ack eagerly? */
145 int *ksnd_typed_conns; /* drive sockets by type? */
146 int *ksnd_min_bulk; /* smallest "large" message */
147 int *ksnd_tx_buffer_size; /* socket tx buffer size */
148 int *ksnd_rx_buffer_size; /* socket rx buffer size */
149 int *ksnd_nagle; /* enable NAGLE? */
150 int *ksnd_round_robin; /* round robin for multiple interfaces */
151 int *ksnd_keepalive; /* # secs for sending keepalive NOOP */
152 int *ksnd_keepalive_idle; /* # idle secs before 1st probe */
153 int *ksnd_keepalive_count; /* # probes */
154 int *ksnd_keepalive_intvl; /* time between probes */
155 int *ksnd_credits; /* # concurrent sends */
156 int *ksnd_peertxcredits; /* # concurrent sends to 1 peer */
157 int *ksnd_peerrtrcredits; /* # per-peer router buffer credits */
158 int *ksnd_peertimeout; /* seconds to consider peer dead */
159 int *ksnd_enable_csum; /* enable check sum */
160 int *ksnd_inject_csum_error; /* set non-zero to inject checksum error */
161 int *ksnd_nonblk_zcack; /* always send zc-ack on non-blocking connection */
162 unsigned int *ksnd_zc_min_payload; /* minimum zero copy payload size */
163 int *ksnd_zc_recv; /* enable ZC receive (for Chelsio TOE) */
164 int *ksnd_zc_recv_min_nfrags; /* minimum # of fragments to enable ZC receive */
166 int *ksnd_irq_affinity; /* enable IRQ affinity? */
168 #ifdef SOCKNAL_BACKOFF
169 int *ksnd_backoff_init; /* initial TCP backoff */
170 int *ksnd_backoff_max; /* maximum TCP backoff */
172 #if SOCKNAL_VERSION_DEBUG
173 int *ksnd_protocol; /* protocol version */
175 #if defined(CONFIG_SYSCTL) && !CFS_SYSFS_MODULE_PARM
176 struct ctl_table_header *ksnd_sysctl; /* sysctl interface */
182 __u64 ksnn_incarnation; /* my epoch */
183 spinlock_t ksnn_lock; /* serialise */
184 struct list_head ksnn_list; /* chain on global list */
185 int ksnn_npeers; /* # peers */
186 int ksnn_shutdown; /* shutting down? */
187 int ksnn_ninterfaces; /* IP interfaces */
188 ksock_interface_t ksnn_interfaces[LNET_MAX_INTERFACES];
192 #define SOCKNAL_CONND_TIMEOUT 120
193 /** reserved thread for accepting & creating new connd */
194 #define SOCKNAL_CONND_RESV 1
198 int ksnd_init; /* initialisation state */
199 int ksnd_nnets; /* # networks set up */
200 struct list_head ksnd_nets; /* list of nets */
201 /* stabilize peer/conn ops */
202 rwlock_t ksnd_global_lock;
203 /* hash table of all my known peers */
204 struct list_head *ksnd_peers;
205 int ksnd_peer_hash_size; /* size of ksnd_peers */
207 int ksnd_nthreads; /* # live threads */
208 int ksnd_shuttingdown; /* tell threads to exit */
209 /* schedulers information */
210 struct ksock_sched_info **ksnd_sched_info;
212 atomic_t ksnd_nactive_txs; /* #active txs */
214 /* conns to close: reaper_lock*/
215 struct list_head ksnd_deathrow_conns;
216 /* conns to free: reaper_lock */
217 struct list_head ksnd_zombie_conns;
218 /* conns to retry: reaper_lock*/
219 struct list_head ksnd_enomem_conns;
220 /* reaper sleeps here */
221 wait_queue_head_t ksnd_reaper_waitq;
222 /* when reaper will wake */
223 cfs_time_t ksnd_reaper_waketime;
225 spinlock_t ksnd_reaper_lock;
227 int ksnd_enomem_tx; /* test ENOMEM sender */
228 int ksnd_stall_tx; /* test sluggish sender */
229 int ksnd_stall_rx; /* test sluggish receiver */
231 /* incoming connection requests */
232 struct list_head ksnd_connd_connreqs;
233 /* routes waiting to be connected */
234 struct list_head ksnd_connd_routes;
235 /* connds sleep here */
236 wait_queue_head_t ksnd_connd_waitq;
237 /* # connds connecting */
238 int ksnd_connd_connecting;
239 /** time stamp of the last failed connecting attempt */
240 long ksnd_connd_failed_stamp;
241 /** # starting connd */
242 unsigned ksnd_connd_starting;
243 /** time stamp of the last starting connd */
244 long ksnd_connd_starting_stamp;
245 /** # running connd */
246 unsigned ksnd_connd_running;
248 spinlock_t ksnd_connd_lock;
250 /* list head for freed noop tx */
251 struct list_head ksnd_idle_noop_txs;
252 /* serialise, g_lock unsafe */
253 spinlock_t ksnd_tx_lock;
257 #define SOCKNAL_INIT_NOTHING 0
258 #define SOCKNAL_INIT_DATA 1
259 #define SOCKNAL_INIT_ALL 2
261 /* A packet just assembled for transmission is represented by 1 or more
262 * struct kvec fragments (the first frag contains the portals header),
263 * followed by 0 or more lnet_kiov_t fragments.
265 * On the receive side, initially 1 struct kvec fragment is posted for
266 * receive (the header). Once the header has been received, the payload is
267 * received into either struct kvec or lnet_kiov_t fragments, depending on
268 * what the header matched or whether the message needs forwarding. */
270 struct ksock_conn; /* forward ref */
271 struct ksock_peer; /* forward ref */
272 struct ksock_route; /* forward ref */
273 struct ksock_proto; /* forward ref */
275 typedef struct /* transmit packet */
277 struct list_head tx_list; /* queue on conn for transmission etc */
278 struct list_head tx_zc_list; /* queue on peer for ZC request */
279 atomic_t tx_refcount; /* tx reference count */
280 int tx_nob; /* # packet bytes */
281 int tx_resid; /* residual bytes */
282 int tx_niov; /* # packet kvec frags */
283 struct kvec *tx_iov; /* packet kvec frags */
284 int tx_nkiov; /* # packet page frags */
285 unsigned short tx_zc_aborted; /* aborted ZC request */
286 unsigned short tx_zc_capable:1; /* payload is large enough for ZC */
287 unsigned short tx_zc_checked:1; /* Have I checked if I should ZC? */
288 unsigned short tx_nonblk:1; /* it's a non-blocking ACK */
289 lnet_kiov_t *tx_kiov; /* packet page frags */
290 struct ksock_conn *tx_conn; /* owning conn */
291 lnet_msg_t *tx_lnetmsg; /* lnet message for lnet_finalize() */
292 cfs_time_t tx_deadline; /* when (in jiffies) tx times out */
293 ksock_msg_t tx_msg; /* socklnd message buffer */
294 int tx_desc_size; /* size of this descriptor */
297 struct kvec iov; /* virt hdr */
298 lnet_kiov_t kiov[0]; /* paged payload */
301 struct kvec iov[1]; /* virt hdr + payload */
306 #define KSOCK_NOOP_TX_SIZE ((int)offsetof(ksock_tx_t, tx_frags.paged.kiov[0]))
308 /* network zero copy callback descriptor embedded in ksock_tx_t */
310 /* space for the rx frag descriptors; we either read a single contiguous
311 * header, or up to LNET_MAX_IOV frags of payload of either type. */
313 struct kvec iov[LNET_MAX_IOV];
314 lnet_kiov_t kiov[LNET_MAX_IOV];
315 } ksock_rxiovspace_t;
317 #define SOCKNAL_RX_KSM_HEADER 1 /* reading ksock message header */
318 #define SOCKNAL_RX_LNET_HEADER 2 /* reading lnet message header */
319 #define SOCKNAL_RX_PARSE 3 /* Calling lnet_parse() */
320 #define SOCKNAL_RX_PARSE_WAIT 4 /* waiting to be told to read the body */
321 #define SOCKNAL_RX_LNET_PAYLOAD 5 /* reading lnet payload (to deliver here) */
322 #define SOCKNAL_RX_SLOP 6 /* skipping body */
324 typedef struct ksock_conn
326 struct ksock_peer *ksnc_peer; /* owning peer */
327 struct ksock_route *ksnc_route; /* owning route */
328 struct list_head ksnc_list; /* stash on peer's conn list */
329 struct socket *ksnc_sock; /* actual socket */
330 void *ksnc_saved_data_ready; /* socket's original data_ready() callback */
331 void *ksnc_saved_write_space; /* socket's original write_space() callback */
332 atomic_t ksnc_conn_refcount; /* conn refcount */
333 atomic_t ksnc_sock_refcount; /* sock refcount */
334 ksock_sched_t *ksnc_scheduler; /* who schedules this connection */
335 __u32 ksnc_myipaddr; /* my IP */
336 __u32 ksnc_ipaddr; /* peer's IP */
337 int ksnc_port; /* peer's port */
338 signed int ksnc_type:3; /* type of connection,
339 * should be signed value */
340 unsigned int ksnc_closing:1; /* being shut down */
341 unsigned int ksnc_flip:1; /* flip or not, only for V2.x */
342 unsigned int ksnc_zc_capable:1; /* enable to ZC */
343 struct ksock_proto *ksnc_proto; /* protocol for the connection */
347 /* where I enq waiting input or a forwarding descriptor */
348 struct list_head ksnc_rx_list;
349 cfs_time_t ksnc_rx_deadline; /* when (in jiffies) receive times out */
350 __u8 ksnc_rx_started; /* started receiving a message */
351 __u8 ksnc_rx_ready; /* data ready to read */
352 __u8 ksnc_rx_scheduled;/* being progressed */
353 __u8 ksnc_rx_state; /* what is being read */
354 int ksnc_rx_nob_left; /* # bytes to next hdr/body */
355 int ksnc_rx_nob_wanted; /* bytes actually wanted */
356 int ksnc_rx_niov; /* # kvec frags */
357 struct kvec *ksnc_rx_iov; /* the kvec frags */
358 int ksnc_rx_nkiov; /* # page frags */
359 lnet_kiov_t *ksnc_rx_kiov; /* the page frags */
360 ksock_rxiovspace_t ksnc_rx_iov_space;/* space for frag descriptors */
361 __u32 ksnc_rx_csum; /* partial checksum for incoming data */
362 void *ksnc_cookie; /* rx lnet_finalize passthru arg */
363 ksock_msg_t ksnc_msg; /* incoming message buffer:
364 * V2.x message takes the
366 * V1.x message is a bare
367 * lnet_hdr_t, it's stored in
368 * ksnc_msg.ksm_u.lnetmsg */
371 /* where I enq waiting for output space */
372 struct list_head ksnc_tx_list;
373 /* packets waiting to be sent */
374 struct list_head ksnc_tx_queue;
375 /* next TX that can carry a LNet message or ZC-ACK */
376 ksock_tx_t *ksnc_tx_carrier;
377 /* when (in jiffies) tx times out */
378 cfs_time_t ksnc_tx_deadline;
379 /* send buffer marker */
382 atomic_t ksnc_tx_nob;
385 /* being progressed */
386 int ksnc_tx_scheduled;
387 /* time stamp of the last posted TX */
388 cfs_time_t ksnc_tx_last_post;
391 typedef struct ksock_route
393 struct list_head ksnr_list; /* chain on peer route list */
394 struct list_head ksnr_connd_list; /* chain on ksnr_connd_routes */
395 struct ksock_peer *ksnr_peer; /* owning peer */
396 atomic_t ksnr_refcount; /* # users */
397 cfs_time_t ksnr_timeout; /* when (in jiffies) reconnection can happen next */
398 cfs_duration_t ksnr_retry_interval; /* how long between retries */
399 __u32 ksnr_myipaddr; /* my IP */
400 __u32 ksnr_ipaddr; /* IP address to connect to */
401 int ksnr_port; /* port to connect to */
402 unsigned int ksnr_scheduled:1; /* scheduled for attention */
403 unsigned int ksnr_connecting:1;/* connection establishment in progress */
404 unsigned int ksnr_connected:4; /* connections established by type */
405 unsigned int ksnr_deleted:1; /* been removed from peer? */
406 unsigned int ksnr_share_count; /* created explicitly? */
407 int ksnr_conn_count; /* # conns established by this route */
410 #define SOCKNAL_KEEPALIVE_PING 1 /* cookie for keepalive ping */
412 typedef struct ksock_peer
414 struct list_head ksnp_list; /* stash on global peer list */
415 cfs_time_t ksnp_last_alive; /* when (in jiffies) I was last alive */
416 lnet_process_id_t ksnp_id; /* who's on the other end(s) */
417 atomic_t ksnp_refcount; /* # users */
418 int ksnp_sharecount; /* lconf usage counter */
419 int ksnp_closing; /* being closed */
420 int ksnp_accepting;/* # passive connections pending */
421 int ksnp_error; /* errno on closing last conn */
422 __u64 ksnp_zc_next_cookie;/* ZC completion cookie */
423 __u64 ksnp_incarnation; /* latest known peer incarnation */
424 struct ksock_proto *ksnp_proto; /* latest known peer protocol */
425 struct list_head ksnp_conns; /* all active connections */
426 struct list_head ksnp_routes; /* routes */
427 struct list_head ksnp_tx_queue; /* waiting packets */
428 spinlock_t ksnp_lock; /* serialize, g_lock unsafe */
429 /* zero copy requests wait for ACK */
430 struct list_head ksnp_zc_req_list;
431 cfs_time_t ksnp_send_keepalive; /* time to send keepalive */
432 lnet_ni_t *ksnp_ni; /* which network */
433 int ksnp_n_passive_ips; /* # of... */
434 __u32 ksnp_passive_ips[LNET_MAX_INTERFACES]; /* preferred local interfaces */
437 typedef struct ksock_connreq
439 /* stash on ksnd_connd_connreqs */
440 struct list_head ksncr_list;
443 /* accepted socket */
444 struct socket *ksncr_sock;
447 extern ksock_nal_data_t ksocknal_data;
448 extern ksock_tunables_t ksocknal_tunables;
450 #define SOCKNAL_MATCH_NO 0 /* TX can't match type of connection */
451 #define SOCKNAL_MATCH_YES 1 /* TX matches type of connection */
452 #define SOCKNAL_MATCH_MAY 2 /* TX can be sent on the connection, but not preferred */
454 typedef struct ksock_proto
456 int pro_version; /* version number of protocol */
457 int (*pro_send_hello)(ksock_conn_t *, ksock_hello_msg_t *); /* handshake function */
458 int (*pro_recv_hello)(ksock_conn_t *, ksock_hello_msg_t *, int);/* handshake function */
459 void (*pro_pack)(ksock_tx_t *); /* message pack */
460 void (*pro_unpack)(ksock_msg_t *); /* message unpack */
461 ksock_tx_t *(*pro_queue_tx_msg)(ksock_conn_t *, ksock_tx_t *); /* queue tx on the connection */
462 int (*pro_queue_tx_zcack)(ksock_conn_t *, ksock_tx_t *, __u64); /* queue ZC ack on the connection */
463 int (*pro_handle_zcreq)(ksock_conn_t *, __u64, int); /* handle ZC request */
464 int (*pro_handle_zcack)(ksock_conn_t *, __u64, __u64); /* handle ZC ACK */
465 int (*pro_match_tx)(ksock_conn_t *, ksock_tx_t *, int); /* msg type matches the connection type:
467 * return MATCH_NO : no
468 * return MATCH_YES : matching type
469 * return MATCH_MAY : can be backup */
472 extern ksock_proto_t ksocknal_protocol_v1x;
473 extern ksock_proto_t ksocknal_protocol_v2x;
474 extern ksock_proto_t ksocknal_protocol_v3x;
476 #define KSOCK_PROTO_V1_MAJOR LNET_PROTO_TCP_VERSION_MAJOR
477 #define KSOCK_PROTO_V1_MINOR LNET_PROTO_TCP_VERSION_MINOR
478 #define KSOCK_PROTO_V1 KSOCK_PROTO_V1_MAJOR
480 #ifndef CPU_MASK_NONE
481 #define CPU_MASK_NONE 0UL
484 static inline __u32 ksocknal_csum(__u32 crc, unsigned char const *p, size_t len)
487 return crc32_le(crc, p, len);
490 crc = ((crc + 0x100) & ~0xff) | ((crc + *p++) & 0xff) ;
497 ksocknal_route_mask(void)
499 if (!*ksocknal_tunables.ksnd_typed_conns)
500 return (1 << SOCKLND_CONN_ANY);
502 return ((1 << SOCKLND_CONN_CONTROL) |
503 (1 << SOCKLND_CONN_BULK_IN) |
504 (1 << SOCKLND_CONN_BULK_OUT));
507 static inline struct list_head *
508 ksocknal_nid2peerlist (lnet_nid_t nid)
510 unsigned int hash = ((unsigned int)nid) % ksocknal_data.ksnd_peer_hash_size;
512 return (&ksocknal_data.ksnd_peers [hash]);
516 ksocknal_conn_addref (ksock_conn_t *conn)
518 LASSERT (atomic_read(&conn->ksnc_conn_refcount) > 0);
519 atomic_inc(&conn->ksnc_conn_refcount);
522 extern void ksocknal_queue_zombie_conn (ksock_conn_t *conn);
523 extern void ksocknal_finalize_zcreq(ksock_conn_t *conn);
526 ksocknal_conn_decref (ksock_conn_t *conn)
528 LASSERT (atomic_read(&conn->ksnc_conn_refcount) > 0);
529 if (atomic_dec_and_test(&conn->ksnc_conn_refcount))
530 ksocknal_queue_zombie_conn(conn);
534 ksocknal_connsock_addref (ksock_conn_t *conn)
538 read_lock(&ksocknal_data.ksnd_global_lock);
539 if (!conn->ksnc_closing) {
540 LASSERT(atomic_read(&conn->ksnc_sock_refcount) > 0);
541 atomic_inc(&conn->ksnc_sock_refcount);
544 read_unlock(&ksocknal_data.ksnd_global_lock);
550 ksocknal_connsock_decref (ksock_conn_t *conn)
552 LASSERT (atomic_read(&conn->ksnc_sock_refcount) > 0);
553 if (atomic_dec_and_test(&conn->ksnc_sock_refcount)) {
554 LASSERT (conn->ksnc_closing);
555 sock_release(conn->ksnc_sock);
556 conn->ksnc_sock = NULL;
557 ksocknal_finalize_zcreq(conn);
562 ksocknal_tx_addref (ksock_tx_t *tx)
564 LASSERT (atomic_read(&tx->tx_refcount) > 0);
565 atomic_inc(&tx->tx_refcount);
568 extern void ksocknal_tx_prep (ksock_conn_t *, ksock_tx_t *tx);
569 extern void ksocknal_tx_done (lnet_ni_t *ni, ksock_tx_t *tx);
572 ksocknal_tx_decref (ksock_tx_t *tx)
574 LASSERT (atomic_read(&tx->tx_refcount) > 0);
575 if (atomic_dec_and_test(&tx->tx_refcount))
576 ksocknal_tx_done(NULL, tx);
580 ksocknal_route_addref (ksock_route_t *route)
582 LASSERT (atomic_read(&route->ksnr_refcount) > 0);
583 atomic_inc(&route->ksnr_refcount);
586 extern void ksocknal_destroy_route (ksock_route_t *route);
589 ksocknal_route_decref (ksock_route_t *route)
591 LASSERT (atomic_read (&route->ksnr_refcount) > 0);
592 if (atomic_dec_and_test(&route->ksnr_refcount))
593 ksocknal_destroy_route (route);
597 ksocknal_peer_addref (ksock_peer_t *peer)
599 LASSERT (atomic_read (&peer->ksnp_refcount) > 0);
600 atomic_inc(&peer->ksnp_refcount);
603 extern void ksocknal_destroy_peer (ksock_peer_t *peer);
606 ksocknal_peer_decref (ksock_peer_t *peer)
608 LASSERT (atomic_read (&peer->ksnp_refcount) > 0);
609 if (atomic_dec_and_test(&peer->ksnp_refcount))
610 ksocknal_destroy_peer (peer);
613 int ksocknal_startup (lnet_ni_t *ni);
614 void ksocknal_shutdown (lnet_ni_t *ni);
615 int ksocknal_ctl(lnet_ni_t *ni, unsigned int cmd, void *arg);
616 int ksocknal_send (lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg);
617 int ksocknal_recv(lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg,
618 int delayed, unsigned int niov,
619 struct kvec *iov, lnet_kiov_t *kiov,
620 unsigned int offset, unsigned int mlen, unsigned int rlen);
621 int ksocknal_accept(lnet_ni_t *ni, struct socket *sock);
623 extern int ksocknal_add_peer(lnet_ni_t *ni, lnet_process_id_t id, __u32 ip, int port);
624 extern ksock_peer_t *ksocknal_find_peer_locked (lnet_ni_t *ni, lnet_process_id_t id);
625 extern ksock_peer_t *ksocknal_find_peer (lnet_ni_t *ni, lnet_process_id_t id);
626 extern void ksocknal_peer_failed (ksock_peer_t *peer);
627 extern int ksocknal_create_conn (lnet_ni_t *ni, ksock_route_t *route,
628 struct socket *sock, int type);
629 extern void ksocknal_close_conn_locked (ksock_conn_t *conn, int why);
630 extern void ksocknal_terminate_conn (ksock_conn_t *conn);
631 extern void ksocknal_destroy_conn (ksock_conn_t *conn);
632 extern int ksocknal_close_peer_conns_locked (ksock_peer_t *peer,
633 __u32 ipaddr, int why);
634 extern int ksocknal_close_conn_and_siblings (ksock_conn_t *conn, int why);
635 extern int ksocknal_close_matching_conns(lnet_process_id_t id, __u32 ipaddr);
636 extern ksock_conn_t *ksocknal_find_conn_locked(ksock_peer_t *peer,
637 ksock_tx_t *tx, int nonblk);
639 extern int ksocknal_launch_packet(lnet_ni_t *ni, ksock_tx_t *tx,
640 lnet_process_id_t id);
641 extern ksock_tx_t *ksocknal_alloc_tx(int type, int size);
642 extern void ksocknal_free_tx (ksock_tx_t *tx);
643 extern ksock_tx_t *ksocknal_alloc_tx_noop(__u64 cookie, int nonblk);
644 extern void ksocknal_next_tx_carrier(ksock_conn_t *conn);
645 extern void ksocknal_queue_tx_locked (ksock_tx_t *tx, ksock_conn_t *conn);
646 extern void ksocknal_txlist_done(lnet_ni_t *ni, struct list_head *txlist,
648 extern void ksocknal_notify (lnet_ni_t *ni, lnet_nid_t gw_nid, int alive);
649 extern void ksocknal_query (struct lnet_ni *ni, lnet_nid_t nid, cfs_time_t *when);
650 extern int ksocknal_thread_start(int (*fn)(void *arg), void *arg, char *name);
651 extern void ksocknal_thread_fini (void);
652 extern void ksocknal_launch_all_connections_locked (ksock_peer_t *peer);
653 extern ksock_route_t *ksocknal_find_connectable_route_locked (ksock_peer_t *peer);
654 extern ksock_route_t *ksocknal_find_connecting_route_locked (ksock_peer_t *peer);
655 extern int ksocknal_new_packet (ksock_conn_t *conn, int skip);
656 extern int ksocknal_scheduler (void *arg);
657 extern int ksocknal_connd (void *arg);
658 extern int ksocknal_reaper (void *arg);
659 extern int ksocknal_send_hello (lnet_ni_t *ni, ksock_conn_t *conn,
660 lnet_nid_t peer_nid, ksock_hello_msg_t *hello);
661 extern int ksocknal_recv_hello (lnet_ni_t *ni, ksock_conn_t *conn,
662 ksock_hello_msg_t *hello, lnet_process_id_t *id,
664 extern void ksocknal_read_callback(ksock_conn_t *conn);
665 extern void ksocknal_write_callback(ksock_conn_t *conn);
667 extern int ksocknal_lib_zc_capable(ksock_conn_t *conn);
668 extern void ksocknal_lib_save_callback(struct socket *sock, ksock_conn_t *conn);
669 extern void ksocknal_lib_set_callback(struct socket *sock, ksock_conn_t *conn);
670 extern void ksocknal_lib_reset_callback(struct socket *sock,
672 extern void ksocknal_lib_push_conn(ksock_conn_t *conn);
673 extern int ksocknal_lib_get_conn_addrs(ksock_conn_t *conn);
674 extern int ksocknal_lib_setup_sock(struct socket *so);
675 extern int ksocknal_lib_send_iov(ksock_conn_t *conn, ksock_tx_t *tx);
676 extern int ksocknal_lib_send_kiov(ksock_conn_t *conn, ksock_tx_t *tx);
677 extern void ksocknal_lib_eager_ack(ksock_conn_t *conn);
678 extern int ksocknal_lib_recv_iov(ksock_conn_t *conn);
679 extern int ksocknal_lib_recv_kiov(ksock_conn_t *conn);
680 extern int ksocknal_lib_get_conn_tunables(ksock_conn_t *conn, int *txmem,
681 int *rxmem, int *nagle);
683 extern int ksocknal_tunables_init(void);
684 extern void ksocknal_tunables_fini(void);
685 extern int ksocknal_lib_tunables_init(void);
686 extern void ksocknal_lib_tunables_fini(void);
688 extern void ksocknal_lib_csum_tx(ksock_tx_t *tx);
690 extern int ksocknal_lib_memory_pressure(ksock_conn_t *conn);
691 extern int ksocknal_lib_bind_thread_to_cpu(int id);
693 #endif /* _SOCKLND_SOCKLND_H_ */