2 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
4 * Copyright (c) 2011, 2016, 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/stat.h>
45 #include <linux/string.h>
46 #include <linux/syscalls.h>
47 #include <linux/sysctl.h>
48 #include <linux/uio.h>
49 #include <linux/unistd.h>
53 #include <libcfs/libcfs.h>
54 #include <lnet/lnet.h>
55 #include <lnet/lib-lnet.h>
56 #include <lnet/socklnd.h>
58 #ifdef HAVE_TCP_SENDPAGE_USE_SOCKET
59 # define cfs_tcp_sendpage(sk, page, offset, size, flags) \
60 tcp_sendpage((sk)->sk_socket, page, offset, size, flags)
61 #else /* !HAVE_TCP_SENDPAGE_USE_SOCKET */
62 # define cfs_tcp_sendpage(sk, page, offset, size, flags) \
63 tcp_sendpage(sk, page, offset, size, flags)
64 #endif /* HAVE_TCP_SENDPAGE_USE_SOCKET */
66 #ifndef NETIF_F_CSUM_MASK
67 # define NETIF_F_CSUM_MASK NETIF_F_ALL_CSUM
70 /* assume one thread for each connection type */
71 #define SOCKNAL_NSCHEDS 3
72 #define SOCKNAL_NSCHEDS_HIGH (SOCKNAL_NSCHEDS << 1)
74 #define SOCKNAL_PEER_HASH_SIZE 101 /* # peer lists */
75 #define SOCKNAL_RESCHED 100 /* # scheduler loops before reschedule */
76 #define SOCKNAL_INSANITY_RECONN 5000 /* connd is trying on reconn infinitely */
77 #define SOCKNAL_ENOMEM_RETRY CFS_TICK /* jiffies between retries */
79 #define SOCKNAL_SINGLE_FRAG_TX 0 /* disable multi-fragment sends */
80 #define SOCKNAL_SINGLE_FRAG_RX 0 /* disable multi-fragment receives */
82 #define SOCKNAL_VERSION_DEBUG 0 /* enable protocol version debugging */
84 /* risk kmap deadlock on multi-frag I/O (backs off to single-frag if disabled).
85 * no risk if we're not running on a CONFIG_HIGHMEM platform. */
87 # define SOCKNAL_RISK_KMAP_DEADLOCK 0
89 # define SOCKNAL_RISK_KMAP_DEADLOCK 1
92 struct ksock_sched_info;
94 typedef struct /* per scheduler state */
96 spinlock_t kss_lock; /* serialise */
97 struct list_head kss_rx_conns; /* conn waiting to be read */
98 /* conn waiting to be written */
99 struct list_head kss_tx_conns;
100 /* zombie noop tx list */
101 struct list_head kss_zombie_noop_txs;
102 wait_queue_head_t kss_waitq; /* where scheduler sleeps */
103 /* # connections assigned to this scheduler */
105 struct ksock_sched_info *kss_info; /* owner of it */
106 #if !SOCKNAL_SINGLE_FRAG_RX
107 struct page *kss_rx_scratch_pgs[LNET_MAX_IOV];
109 #if !SOCKNAL_SINGLE_FRAG_TX || !SOCKNAL_SINGLE_FRAG_RX
110 struct kvec kss_scratch_iov[LNET_MAX_IOV];
114 struct ksock_sched_info {
115 int ksi_nthreads_max; /* max allowed threads */
116 int ksi_nthreads; /* number of threads */
117 int ksi_cpt; /* CPT id */
118 ksock_sched_t *ksi_scheds; /* array of schedulers */
121 #define KSOCK_CPT_SHIFT 16
122 #define KSOCK_THREAD_ID(cpt, sid) (((cpt) << KSOCK_CPT_SHIFT) | (sid))
123 #define KSOCK_THREAD_CPT(id) ((id) >> KSOCK_CPT_SHIFT)
124 #define KSOCK_THREAD_SID(id) ((id) & ((1UL << KSOCK_CPT_SHIFT) - 1))
126 typedef struct /* in-use interface */
128 __u32 ksni_ipaddr; /* interface's IP address */
129 __u32 ksni_netmask; /* interface's network mask */
130 int ksni_nroutes; /* # routes using (active) */
131 int ksni_npeers; /* # peers using (passive) */
132 char ksni_name[IFNAMSIZ]; /* interface name */
137 /* "stuck" socket timeout (seconds) */
139 /* # scheduler threads in each pool while starting */
141 int *ksnd_nconnds; /* # connection daemons */
142 int *ksnd_nconnds_max; /* max # connection daemons */
143 int *ksnd_min_reconnectms; /* first connection retry after (ms)... */
144 int *ksnd_max_reconnectms; /* ...exponentially increasing to this */
145 int *ksnd_eager_ack; /* make TCP ack eagerly? */
146 int *ksnd_typed_conns; /* drive sockets by type? */
147 int *ksnd_min_bulk; /* smallest "large" message */
148 int *ksnd_tx_buffer_size; /* socket tx buffer size */
149 int *ksnd_rx_buffer_size; /* socket rx buffer size */
150 int *ksnd_nagle; /* enable NAGLE? */
151 int *ksnd_round_robin; /* round robin for multiple interfaces */
152 int *ksnd_keepalive; /* # secs for sending keepalive NOOP */
153 int *ksnd_keepalive_idle; /* # idle secs before 1st probe */
154 int *ksnd_keepalive_count; /* # probes */
155 int *ksnd_keepalive_intvl; /* time between probes */
156 int *ksnd_credits; /* # concurrent sends */
157 int *ksnd_peertxcredits; /* # concurrent sends to 1 peer */
158 int *ksnd_peerrtrcredits; /* # per-peer router buffer credits */
159 int *ksnd_peertimeout; /* seconds to consider peer dead */
160 int *ksnd_enable_csum; /* enable check sum */
161 int *ksnd_inject_csum_error; /* set non-zero to inject checksum error */
162 int *ksnd_nonblk_zcack; /* always send zc-ack on non-blocking connection */
163 unsigned int *ksnd_zc_min_payload; /* minimum zero copy payload size */
164 int *ksnd_zc_recv; /* enable ZC receive (for Chelsio TOE) */
165 int *ksnd_zc_recv_min_nfrags; /* minimum # of fragments to enable ZC receive */
167 int *ksnd_irq_affinity; /* enable IRQ affinity? */
169 #ifdef SOCKNAL_BACKOFF
170 int *ksnd_backoff_init; /* initial TCP backoff */
171 int *ksnd_backoff_max; /* maximum TCP backoff */
173 #if SOCKNAL_VERSION_DEBUG
174 int *ksnd_protocol; /* protocol version */
180 __u64 ksnn_incarnation; /* my epoch */
181 spinlock_t ksnn_lock; /* serialise */
182 struct list_head ksnn_list; /* chain on global list */
183 int ksnn_npeers; /* # peers */
184 int ksnn_shutdown; /* shutting down? */
185 int ksnn_ninterfaces; /* IP interfaces */
186 ksock_interface_t ksnn_interfaces[LNET_MAX_INTERFACES];
190 #define SOCKNAL_CONND_TIMEOUT 120
191 /** reserved thread for accepting & creating new connd */
192 #define SOCKNAL_CONND_RESV 1
196 int ksnd_init; /* initialisation state */
197 int ksnd_nnets; /* # networks set up */
198 struct list_head ksnd_nets; /* list of nets */
199 /* stabilize peer/conn ops */
200 rwlock_t ksnd_global_lock;
201 /* hash table of all my known peers */
202 struct list_head *ksnd_peers;
203 int ksnd_peer_hash_size; /* size of ksnd_peers */
205 int ksnd_nthreads; /* # live threads */
206 int ksnd_shuttingdown; /* tell threads to exit */
207 /* schedulers information */
208 struct ksock_sched_info **ksnd_sched_info;
210 atomic_t ksnd_nactive_txs; /* #active txs */
212 /* conns to close: reaper_lock*/
213 struct list_head ksnd_deathrow_conns;
214 /* conns to free: reaper_lock */
215 struct list_head ksnd_zombie_conns;
216 /* conns to retry: reaper_lock*/
217 struct list_head ksnd_enomem_conns;
218 /* reaper sleeps here */
219 wait_queue_head_t ksnd_reaper_waitq;
220 /* when reaper will wake */
221 cfs_time_t ksnd_reaper_waketime;
223 spinlock_t ksnd_reaper_lock;
225 int ksnd_enomem_tx; /* test ENOMEM sender */
226 int ksnd_stall_tx; /* test sluggish sender */
227 int ksnd_stall_rx; /* test sluggish receiver */
229 /* incoming connection requests */
230 struct list_head ksnd_connd_connreqs;
231 /* routes waiting to be connected */
232 struct list_head ksnd_connd_routes;
233 /* connds sleep here */
234 wait_queue_head_t ksnd_connd_waitq;
235 /* # connds connecting */
236 int ksnd_connd_connecting;
237 /** time stamp of the last failed connecting attempt */
238 time64_t ksnd_connd_failed_stamp;
239 /** # starting connd */
240 unsigned ksnd_connd_starting;
241 /** time stamp of the last starting connd */
242 time64_t ksnd_connd_starting_stamp;
243 /** # running connd */
244 unsigned ksnd_connd_running;
246 spinlock_t ksnd_connd_lock;
248 /* list head for freed noop tx */
249 struct list_head ksnd_idle_noop_txs;
250 /* serialise, g_lock unsafe */
251 spinlock_t ksnd_tx_lock;
255 #define SOCKNAL_INIT_NOTHING 0
256 #define SOCKNAL_INIT_DATA 1
257 #define SOCKNAL_INIT_ALL 2
259 /* A packet just assembled for transmission is represented by 1 or more
260 * struct kvec fragments (the first frag contains the portals header),
261 * followed by 0 or more lnet_kiov_t fragments.
263 * On the receive side, initially 1 struct kvec fragment is posted for
264 * receive (the header). Once the header has been received, the payload is
265 * received into either struct kvec or lnet_kiov_t fragments, depending on
266 * what the header matched or whether the message needs forwarding. */
268 struct ksock_conn; /* forward ref */
269 struct ksock_peer; /* forward ref */
270 struct ksock_route; /* forward ref */
271 struct ksock_proto; /* forward ref */
273 typedef struct /* transmit packet */
275 struct list_head tx_list; /* queue on conn for transmission etc */
276 struct list_head tx_zc_list; /* queue on peer for ZC request */
277 atomic_t tx_refcount; /* tx reference count */
278 int tx_nob; /* # packet bytes */
279 int tx_resid; /* residual bytes */
280 int tx_niov; /* # packet kvec frags */
281 struct kvec *tx_iov; /* packet kvec frags */
282 int tx_nkiov; /* # packet page frags */
283 unsigned short tx_zc_aborted; /* aborted ZC request */
284 unsigned short tx_zc_capable:1; /* payload is large enough for ZC */
285 unsigned short tx_zc_checked:1; /* Have I checked if I should ZC? */
286 unsigned short tx_nonblk:1; /* it's a non-blocking ACK */
287 lnet_kiov_t *tx_kiov; /* packet page frags */
288 struct ksock_conn *tx_conn; /* owning conn */
289 lnet_msg_t *tx_lnetmsg; /* lnet message for lnet_finalize() */
290 cfs_time_t tx_deadline; /* when (in jiffies) tx times out */
291 struct ksock_msg tx_msg; /* socklnd message buffer */
292 int tx_desc_size; /* size of this descriptor */
295 struct kvec iov; /* virt hdr */
296 lnet_kiov_t kiov[0]; /* paged payload */
299 struct kvec iov[1]; /* virt hdr + payload */
304 #define KSOCK_NOOP_TX_SIZE ((int)offsetof(ksock_tx_t, tx_frags.paged.kiov[0]))
306 /* network zero copy callback descriptor embedded in ksock_tx_t */
308 /* space for the rx frag descriptors; we either read a single contiguous
309 * header, or up to LNET_MAX_IOV frags of payload of either type. */
311 struct kvec iov[LNET_MAX_IOV];
312 lnet_kiov_t kiov[LNET_MAX_IOV];
313 } ksock_rxiovspace_t;
315 #define SOCKNAL_RX_KSM_HEADER 1 /* reading ksock message header */
316 #define SOCKNAL_RX_LNET_HEADER 2 /* reading lnet message header */
317 #define SOCKNAL_RX_PARSE 3 /* Calling lnet_parse() */
318 #define SOCKNAL_RX_PARSE_WAIT 4 /* waiting to be told to read the body */
319 #define SOCKNAL_RX_LNET_PAYLOAD 5 /* reading lnet payload (to deliver here) */
320 #define SOCKNAL_RX_SLOP 6 /* skipping body */
322 typedef struct ksock_conn
324 struct ksock_peer *ksnc_peer; /* owning peer */
325 struct ksock_route *ksnc_route; /* owning route */
326 struct list_head ksnc_list; /* stash on peer's conn list */
327 struct socket *ksnc_sock; /* actual socket */
328 void *ksnc_saved_data_ready; /* socket's original data_ready() callback */
329 void *ksnc_saved_write_space; /* socket's original write_space() callback */
330 atomic_t ksnc_conn_refcount; /* conn refcount */
331 atomic_t ksnc_sock_refcount; /* sock refcount */
332 ksock_sched_t *ksnc_scheduler; /* who schedules this connection */
333 __u32 ksnc_myipaddr; /* my IP */
334 __u32 ksnc_ipaddr; /* peer's IP */
335 int ksnc_port; /* peer's port */
336 signed int ksnc_type:3; /* type of connection,
337 * should be signed value */
338 unsigned int ksnc_closing:1; /* being shut down */
339 unsigned int ksnc_flip:1; /* flip or not, only for V2.x */
340 unsigned int ksnc_zc_capable:1; /* enable to ZC */
341 struct ksock_proto *ksnc_proto; /* protocol for the connection */
345 /* where I enq waiting input or a forwarding descriptor */
346 struct list_head ksnc_rx_list;
347 cfs_time_t ksnc_rx_deadline; /* when (in jiffies) receive times out */
348 __u8 ksnc_rx_started; /* started receiving a message */
349 __u8 ksnc_rx_ready; /* data ready to read */
350 __u8 ksnc_rx_scheduled;/* being progressed */
351 __u8 ksnc_rx_state; /* what is being read */
352 int ksnc_rx_nob_left; /* # bytes to next hdr/body */
353 int ksnc_rx_nob_wanted; /* bytes actually wanted */
354 int ksnc_rx_niov; /* # kvec frags */
355 struct kvec *ksnc_rx_iov; /* the kvec frags */
356 int ksnc_rx_nkiov; /* # page frags */
357 lnet_kiov_t *ksnc_rx_kiov; /* the page frags */
358 ksock_rxiovspace_t ksnc_rx_iov_space;/* space for frag descriptors */
359 __u32 ksnc_rx_csum; /* partial checksum for incoming data */
360 void *ksnc_cookie; /* rx lnet_finalize passthru arg */
361 struct ksock_msg ksnc_msg; /* incoming message buffer:
362 * V2.x message takes the
364 * V1.x message is a bare
365 * lnet_hdr_t, it's stored in
366 * ksnc_msg.ksm_u.lnetmsg */
369 /* where I enq waiting for output space */
370 struct list_head ksnc_tx_list;
371 /* packets waiting to be sent */
372 struct list_head ksnc_tx_queue;
373 /* next TX that can carry a LNet message or ZC-ACK */
374 ksock_tx_t *ksnc_tx_carrier;
375 /* when (in jiffies) tx times out */
376 cfs_time_t ksnc_tx_deadline;
377 /* send buffer marker */
380 atomic_t ksnc_tx_nob;
383 /* being progressed */
384 int ksnc_tx_scheduled;
385 /* time stamp of the last posted TX */
386 cfs_time_t ksnc_tx_last_post;
389 typedef struct ksock_route
391 struct list_head ksnr_list; /* chain on peer route list */
392 struct list_head ksnr_connd_list; /* chain on ksnr_connd_routes */
393 struct ksock_peer *ksnr_peer; /* owning peer */
394 atomic_t ksnr_refcount; /* # users */
395 cfs_time_t ksnr_timeout; /* when (in jiffies) reconnection can happen next */
396 cfs_duration_t ksnr_retry_interval; /* how long between retries */
397 __u32 ksnr_myipaddr; /* my IP */
398 __u32 ksnr_ipaddr; /* IP address to connect to */
399 int ksnr_port; /* port to connect to */
400 unsigned int ksnr_scheduled:1; /* scheduled for attention */
401 unsigned int ksnr_connecting:1;/* connection establishment in progress */
402 unsigned int ksnr_connected:4; /* connections established by type */
403 unsigned int ksnr_deleted:1; /* been removed from peer? */
404 unsigned int ksnr_share_count; /* created explicitly? */
405 int ksnr_conn_count; /* # conns established by this route */
408 #define SOCKNAL_KEEPALIVE_PING 1 /* cookie for keepalive ping */
410 typedef struct ksock_peer
412 struct list_head ksnp_list; /* stash on global peer list */
413 cfs_time_t ksnp_last_alive; /* when (in jiffies) I was last alive */
414 lnet_process_id_t ksnp_id; /* who's on the other end(s) */
415 atomic_t ksnp_refcount; /* # users */
416 int ksnp_sharecount; /* lconf usage counter */
417 int ksnp_closing; /* being closed */
418 int ksnp_accepting;/* # passive connections pending */
419 int ksnp_error; /* errno on closing last conn */
420 __u64 ksnp_zc_next_cookie;/* ZC completion cookie */
421 __u64 ksnp_incarnation; /* latest known peer incarnation */
422 struct ksock_proto *ksnp_proto; /* latest known peer protocol */
423 struct list_head ksnp_conns; /* all active connections */
424 struct list_head ksnp_routes; /* routes */
425 struct list_head ksnp_tx_queue; /* waiting packets */
426 spinlock_t ksnp_lock; /* serialize, g_lock unsafe */
427 /* zero copy requests wait for ACK */
428 struct list_head ksnp_zc_req_list;
429 cfs_time_t ksnp_send_keepalive; /* time to send keepalive */
430 lnet_ni_t *ksnp_ni; /* which network */
431 int ksnp_n_passive_ips; /* # of... */
432 __u32 ksnp_passive_ips[LNET_MAX_INTERFACES]; /* preferred local interfaces */
435 typedef struct ksock_connreq
437 /* stash on ksnd_connd_connreqs */
438 struct list_head ksncr_list;
441 /* accepted socket */
442 struct socket *ksncr_sock;
445 extern ksock_nal_data_t ksocknal_data;
446 extern ksock_tunables_t ksocknal_tunables;
448 #define SOCKNAL_MATCH_NO 0 /* TX can't match type of connection */
449 #define SOCKNAL_MATCH_YES 1 /* TX matches type of connection */
450 #define SOCKNAL_MATCH_MAY 2 /* TX can be sent on the connection, but not preferred */
452 typedef struct ksock_proto
454 int pro_version; /* version number of protocol */
455 int (*pro_send_hello)(ksock_conn_t *, struct ksock_hello_msg *); /* handshake function */
456 int (*pro_recv_hello)(ksock_conn_t *, struct ksock_hello_msg *, int);/* handshake function */
457 void (*pro_pack)(ksock_tx_t *); /* message pack */
458 void (*pro_unpack)(struct ksock_msg *); /* message unpack */
459 ksock_tx_t *(*pro_queue_tx_msg)(ksock_conn_t *, ksock_tx_t *); /* queue tx on the connection */
460 int (*pro_queue_tx_zcack)(ksock_conn_t *, ksock_tx_t *, __u64); /* queue ZC ack on the connection */
461 int (*pro_handle_zcreq)(ksock_conn_t *, __u64, int); /* handle ZC request */
462 int (*pro_handle_zcack)(ksock_conn_t *, __u64, __u64); /* handle ZC ACK */
463 int (*pro_match_tx)(ksock_conn_t *, ksock_tx_t *, int); /* msg type matches the connection type:
465 * return MATCH_NO : no
466 * return MATCH_YES : matching type
467 * return MATCH_MAY : can be backup */
470 extern ksock_proto_t ksocknal_protocol_v1x;
471 extern ksock_proto_t ksocknal_protocol_v2x;
472 extern ksock_proto_t ksocknal_protocol_v3x;
474 #define KSOCK_PROTO_V1_MAJOR LNET_PROTO_TCP_VERSION_MAJOR
475 #define KSOCK_PROTO_V1_MINOR LNET_PROTO_TCP_VERSION_MINOR
476 #define KSOCK_PROTO_V1 KSOCK_PROTO_V1_MAJOR
478 #ifndef CPU_MASK_NONE
479 #define CPU_MASK_NONE 0UL
482 static inline __u32 ksocknal_csum(__u32 crc, unsigned char const *p, size_t len)
485 return crc32_le(crc, p, len);
488 crc = ((crc + 0x100) & ~0xff) | ((crc + *p++) & 0xff) ;
495 ksocknal_route_mask(void)
497 if (!*ksocknal_tunables.ksnd_typed_conns)
498 return (1 << SOCKLND_CONN_ANY);
500 return ((1 << SOCKLND_CONN_CONTROL) |
501 (1 << SOCKLND_CONN_BULK_IN) |
502 (1 << SOCKLND_CONN_BULK_OUT));
505 static inline struct list_head *
506 ksocknal_nid2peerlist (lnet_nid_t nid)
508 unsigned int hash = ((unsigned int)nid) % ksocknal_data.ksnd_peer_hash_size;
510 return (&ksocknal_data.ksnd_peers [hash]);
514 ksocknal_conn_addref (ksock_conn_t *conn)
516 LASSERT (atomic_read(&conn->ksnc_conn_refcount) > 0);
517 atomic_inc(&conn->ksnc_conn_refcount);
520 extern void ksocknal_queue_zombie_conn (ksock_conn_t *conn);
521 extern void ksocknal_finalize_zcreq(ksock_conn_t *conn);
524 ksocknal_conn_decref (ksock_conn_t *conn)
526 LASSERT (atomic_read(&conn->ksnc_conn_refcount) > 0);
527 if (atomic_dec_and_test(&conn->ksnc_conn_refcount))
528 ksocknal_queue_zombie_conn(conn);
532 ksocknal_connsock_addref (ksock_conn_t *conn)
536 read_lock(&ksocknal_data.ksnd_global_lock);
537 if (!conn->ksnc_closing) {
538 LASSERT(atomic_read(&conn->ksnc_sock_refcount) > 0);
539 atomic_inc(&conn->ksnc_sock_refcount);
542 read_unlock(&ksocknal_data.ksnd_global_lock);
548 ksocknal_connsock_decref (ksock_conn_t *conn)
550 LASSERT (atomic_read(&conn->ksnc_sock_refcount) > 0);
551 if (atomic_dec_and_test(&conn->ksnc_sock_refcount)) {
552 LASSERT (conn->ksnc_closing);
553 sock_release(conn->ksnc_sock);
554 conn->ksnc_sock = NULL;
555 ksocknal_finalize_zcreq(conn);
560 ksocknal_tx_addref (ksock_tx_t *tx)
562 LASSERT (atomic_read(&tx->tx_refcount) > 0);
563 atomic_inc(&tx->tx_refcount);
566 extern void ksocknal_tx_prep (ksock_conn_t *, ksock_tx_t *tx);
567 extern void ksocknal_tx_done (lnet_ni_t *ni, ksock_tx_t *tx);
570 ksocknal_tx_decref (ksock_tx_t *tx)
572 LASSERT (atomic_read(&tx->tx_refcount) > 0);
573 if (atomic_dec_and_test(&tx->tx_refcount))
574 ksocknal_tx_done(NULL, tx);
578 ksocknal_route_addref (ksock_route_t *route)
580 LASSERT (atomic_read(&route->ksnr_refcount) > 0);
581 atomic_inc(&route->ksnr_refcount);
584 extern void ksocknal_destroy_route (ksock_route_t *route);
587 ksocknal_route_decref (ksock_route_t *route)
589 LASSERT (atomic_read (&route->ksnr_refcount) > 0);
590 if (atomic_dec_and_test(&route->ksnr_refcount))
591 ksocknal_destroy_route (route);
595 ksocknal_peer_addref (ksock_peer_t *peer)
597 LASSERT (atomic_read (&peer->ksnp_refcount) > 0);
598 atomic_inc(&peer->ksnp_refcount);
601 extern void ksocknal_destroy_peer (ksock_peer_t *peer);
604 ksocknal_peer_decref (ksock_peer_t *peer)
606 LASSERT (atomic_read (&peer->ksnp_refcount) > 0);
607 if (atomic_dec_and_test(&peer->ksnp_refcount))
608 ksocknal_destroy_peer (peer);
611 int ksocknal_startup (lnet_ni_t *ni);
612 void ksocknal_shutdown (lnet_ni_t *ni);
613 int ksocknal_ctl(lnet_ni_t *ni, unsigned int cmd, void *arg);
614 int ksocknal_send (lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg);
615 int ksocknal_recv(lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg,
616 int delayed, unsigned int niov,
617 struct kvec *iov, lnet_kiov_t *kiov,
618 unsigned int offset, unsigned int mlen, unsigned int rlen);
619 int ksocknal_accept(lnet_ni_t *ni, struct socket *sock);
621 extern int ksocknal_add_peer(lnet_ni_t *ni, lnet_process_id_t id, __u32 ip, int port);
622 extern ksock_peer_t *ksocknal_find_peer_locked (lnet_ni_t *ni, lnet_process_id_t id);
623 extern ksock_peer_t *ksocknal_find_peer (lnet_ni_t *ni, lnet_process_id_t id);
624 extern void ksocknal_peer_failed (ksock_peer_t *peer);
625 extern int ksocknal_create_conn (lnet_ni_t *ni, ksock_route_t *route,
626 struct socket *sock, int type);
627 extern void ksocknal_close_conn_locked (ksock_conn_t *conn, int why);
628 extern void ksocknal_terminate_conn (ksock_conn_t *conn);
629 extern void ksocknal_destroy_conn (ksock_conn_t *conn);
630 extern int ksocknal_close_peer_conns_locked (ksock_peer_t *peer,
631 __u32 ipaddr, int why);
632 extern int ksocknal_close_conn_and_siblings (ksock_conn_t *conn, int why);
633 extern int ksocknal_close_matching_conns(lnet_process_id_t id, __u32 ipaddr);
634 extern ksock_conn_t *ksocknal_find_conn_locked(ksock_peer_t *peer,
635 ksock_tx_t *tx, int nonblk);
637 extern int ksocknal_launch_packet(lnet_ni_t *ni, ksock_tx_t *tx,
638 lnet_process_id_t id);
639 extern ksock_tx_t *ksocknal_alloc_tx(int type, int size);
640 extern void ksocknal_free_tx (ksock_tx_t *tx);
641 extern ksock_tx_t *ksocknal_alloc_tx_noop(__u64 cookie, int nonblk);
642 extern void ksocknal_next_tx_carrier(ksock_conn_t *conn);
643 extern void ksocknal_queue_tx_locked (ksock_tx_t *tx, ksock_conn_t *conn);
644 extern void ksocknal_txlist_done(lnet_ni_t *ni, struct list_head *txlist,
646 extern void ksocknal_notify (lnet_ni_t *ni, lnet_nid_t gw_nid, int alive);
647 extern void ksocknal_query (struct lnet_ni *ni, lnet_nid_t nid, cfs_time_t *when);
648 extern int ksocknal_thread_start(int (*fn)(void *arg), void *arg, char *name);
649 extern void ksocknal_thread_fini (void);
650 extern void ksocknal_launch_all_connections_locked (ksock_peer_t *peer);
651 extern ksock_route_t *ksocknal_find_connectable_route_locked (ksock_peer_t *peer);
652 extern ksock_route_t *ksocknal_find_connecting_route_locked (ksock_peer_t *peer);
653 extern int ksocknal_new_packet (ksock_conn_t *conn, int skip);
654 extern int ksocknal_scheduler (void *arg);
655 extern int ksocknal_connd (void *arg);
656 extern int ksocknal_reaper (void *arg);
657 extern int ksocknal_send_hello(lnet_ni_t *ni, ksock_conn_t *conn,
658 lnet_nid_t peer_nid, struct ksock_hello_msg *hello);
659 extern int ksocknal_recv_hello(lnet_ni_t *ni, ksock_conn_t *conn,
660 struct ksock_hello_msg *hello, lnet_process_id_t *id,
662 extern void ksocknal_read_callback(ksock_conn_t *conn);
663 extern void ksocknal_write_callback(ksock_conn_t *conn);
665 extern int ksocknal_lib_zc_capable(ksock_conn_t *conn);
666 extern void ksocknal_lib_save_callback(struct socket *sock, ksock_conn_t *conn);
667 extern void ksocknal_lib_set_callback(struct socket *sock, ksock_conn_t *conn);
668 extern void ksocknal_lib_reset_callback(struct socket *sock,
670 extern void ksocknal_lib_push_conn(ksock_conn_t *conn);
671 extern int ksocknal_lib_get_conn_addrs(ksock_conn_t *conn);
672 extern int ksocknal_lib_setup_sock(struct socket *so);
673 extern int ksocknal_lib_send_iov(ksock_conn_t *conn, ksock_tx_t *tx);
674 extern int ksocknal_lib_send_kiov(ksock_conn_t *conn, ksock_tx_t *tx);
675 extern void ksocknal_lib_eager_ack(ksock_conn_t *conn);
676 extern int ksocknal_lib_recv_iov(ksock_conn_t *conn);
677 extern int ksocknal_lib_recv_kiov(ksock_conn_t *conn);
678 extern int ksocknal_lib_get_conn_tunables(ksock_conn_t *conn, int *txmem,
679 int *rxmem, int *nagle);
681 extern int ksocknal_tunables_init(void);
683 extern void ksocknal_lib_csum_tx(ksock_tx_t *tx);
685 extern int ksocknal_lib_memory_pressure(ksock_conn_t *conn);
686 extern int ksocknal_lib_bind_thread_to_cpu(int id);
688 #endif /* _SOCKLND_SOCKLND_H_ */