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/refcount.h>
46 #include <linux/stat.h>
47 #include <linux/string.h>
48 #include <linux/syscalls.h>
49 #include <linux/sysctl.h>
50 #include <linux/uio.h>
51 #include <linux/unistd.h>
52 #include <linux/hashtable.h>
56 #include <lnet/lib-lnet.h>
57 #include <lnet/socklnd.h>
59 #include <libcfs/linux/linux-net.h>
61 #ifndef NETIF_F_CSUM_MASK
62 # define NETIF_F_CSUM_MASK NETIF_F_ALL_CSUM
65 /* assume one thread for each connection type */
66 #define SOCKNAL_NSCHEDS 3
67 #define SOCKNAL_NSCHEDS_HIGH (SOCKNAL_NSCHEDS << 1)
69 #define SOCKNAL_PEER_HASH_BITS 7 /* log2 of # peer_ni lists */
70 #define SOCKNAL_INSANITY_RECONN 5000 /* connd is trying on reconn infinitely */
71 #define SOCKNAL_ENOMEM_RETRY 1 /* seconds between retries */
73 #define SOCKNAL_SINGLE_FRAG_TX 0 /* disable multi-fragment sends */
74 #define SOCKNAL_SINGLE_FRAG_RX 0 /* disable multi-fragment receives */
76 #define SOCKNAL_VERSION_DEBUG 0 /* enable protocol version debugging */
78 /* risk kmap deadlock on multi-frag I/O (backs off to single-frag if disabled).
79 * no risk if we're not running on a CONFIG_HIGHMEM platform. */
81 # define SOCKNAL_RISK_KMAP_DEADLOCK 0
83 # define SOCKNAL_RISK_KMAP_DEADLOCK 1
86 enum ksocklnd_ni_lnd_tunables_attr {
87 LNET_NET_SOCKLND_TUNABLES_ATTR_UNSPEC = 0,
89 LNET_NET_SOCKLND_TUNABLES_ATTR_CONNS_PER_PEER,
90 LNET_NET_SOCKLND_TUNABLES_ATTR_LND_TIMEOUT,
91 LNET_NET_SOCKLND_TUNABLES_ATTR_LND_TOS,
92 __LNET_NET_SOCKLND_TUNABLES_ATTR_MAX_PLUS_ONE,
95 #define LNET_NET_SOCKLND_TUNABLES_ATTR_MAX (__LNET_NET_SOCKLND_TUNABLES_ATTR_MAX_PLUS_ONE - 1)
97 /* per scheduler state */
101 /* conn waiting to be written */
102 struct list_head kss_rx_conns;
103 struct list_head kss_tx_conns;
104 /* zombie noop tx list */
105 struct list_head kss_zombie_noop_txs;
106 /* where scheduler sleeps */
107 wait_queue_head_t kss_waitq;
108 /* # connections assigned to this scheduler */
110 /* max allowed threads */
111 int kss_nthreads_max;
112 /* number of threads */
118 #define KSOCK_CPT_SHIFT 16
119 #define KSOCK_THREAD_ID(cpt, sid) (((cpt) << KSOCK_CPT_SHIFT) | (sid))
120 #define KSOCK_THREAD_CPT(id) ((id) >> KSOCK_CPT_SHIFT)
121 #define KSOCK_THREAD_SID(id) ((id) & ((1UL << KSOCK_CPT_SHIFT) - 1))
123 struct ksock_interface { /* in-use interface */
124 int ksni_index; /* Linux interface index */
125 struct sockaddr_storage ksni_addr; /* interface's 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 */
162 int *ksnd_irq_affinity; /* enable IRQ affinity? */
163 #ifdef SOCKNAL_BACKOFF
164 int *ksnd_backoff_init; /* initial TCP backoff */
165 int *ksnd_backoff_max; /* maximum TCP backoff */
167 #if SOCKNAL_VERSION_DEBUG
168 int *ksnd_protocol; /* protocol version */
170 int *ksnd_conns_per_peer; /* for typed mode, yields:
171 * 1 + 2*conns_per_peer total
173 * conns_per_peer total
178 __u64 ksnn_incarnation; /* my epoch */
179 struct list_head ksnn_list; /* chain on global list */
180 atomic_t ksnn_npeers; /* # peers */
181 struct ksock_interface ksnn_interface; /* IP interface */
182 struct lnet_ni *ksnn_ni;
184 /* When the ksock_net is shut down, this (negative) bias is added to
185 * ksnn_npeers, which prevents new peers from being added.
187 #define SOCKNAL_SHUTDOWN_BIAS (INT_MIN+1)
190 #define SOCKNAL_CONND_TIMEOUT 120
191 /** reserved thread for accepting & creating new connd */
192 #define SOCKNAL_CONND_RESV 1
194 struct ksock_nal_data {
195 int ksnd_init; /* initialisation state */
196 int ksnd_nnets; /* # networks set up */
197 struct list_head ksnd_nets; /* list of nets */
198 /* stabilize peer_ni/conn ops */
199 rwlock_t ksnd_global_lock;
200 /* hash table of all my known peers */
201 DECLARE_HASHTABLE(ksnd_peers, SOCKNAL_PEER_HASH_BITS);
203 atomic_t ksnd_nthreads; /* # live threads */
204 int ksnd_shuttingdown; /* tell threads to exit */
205 /* schedulers information */
206 struct ksock_sched **ksnd_schedulers;
208 atomic_t ksnd_nactive_txs; /* #active txs */
210 /* conns to close: reaper_lock*/
211 struct list_head ksnd_deathrow_conns;
212 /* conns to free: reaper_lock */
213 struct list_head ksnd_zombie_conns;
214 /* conns to retry: reaper_lock*/
215 struct list_head ksnd_enomem_conns;
216 /* reaper sleeps here */
217 wait_queue_head_t ksnd_reaper_waitq;
218 /* when reaper will wake */
219 time64_t ksnd_reaper_waketime;
221 spinlock_t ksnd_reaper_lock;
223 int ksnd_enomem_tx; /* test ENOMEM sender */
224 int ksnd_stall_tx; /* test sluggish sender */
225 int ksnd_stall_rx; /* test sluggish receiver */
227 /* incoming connection requests */
228 struct list_head ksnd_connd_connreqs;
229 /* routes waiting to be connected */
230 struct list_head ksnd_connd_routes;
231 /* connds sleep here */
232 wait_queue_head_t ksnd_connd_waitq;
233 /* # connds connecting */
234 int ksnd_connd_connecting;
235 /** time stamp of the last failed connecting attempt */
236 time64_t ksnd_connd_failed_stamp;
237 /** # starting connd */
238 unsigned ksnd_connd_starting;
239 /** time stamp of the last starting connd */
240 time64_t ksnd_connd_starting_stamp;
241 /** # running connd */
242 unsigned ksnd_connd_running;
244 spinlock_t ksnd_connd_lock;
246 /* list head for freed noop tx */
247 struct list_head ksnd_idle_noop_txs;
248 /* serialise, g_lock unsafe */
249 spinlock_t ksnd_tx_lock;
252 #define SOCKNAL_INIT_NOTHING 0
253 #define SOCKNAL_INIT_DATA 1
254 #define SOCKNAL_INIT_ALL 2
256 /* A packet just assembled for transmission is represented by 1
257 * struct iovec fragment - the portals header - followed by 0
258 * or more struct bio_vec fragments.
260 * On the receive side, initially 1 struct kvec fragment is posted for
261 * receive (the header). Once the header has been received, the payload is
262 * received into struct bio_vec fragments.
264 struct ksock_conn; /* forward ref */
265 struct ksock_conn_cb; /* forward ref */
266 struct ksock_proto; /* forward ref */
268 struct ksock_tx { /* transmit packet */
269 struct list_head tx_list; /* queue on conn for transmission etc */
270 struct list_head tx_zc_list; /* queue on peer_ni for ZC request */
271 refcount_t tx_refcount; /* tx reference count */
272 int tx_nob; /* # packet bytes */
273 int tx_resid; /* residual bytes */
274 int tx_niov; /* # packet kvec frags */
275 int tx_nkiov; /* # packet page frags */
276 unsigned short tx_zc_aborted; /* aborted ZC request */
277 unsigned short tx_zc_capable:1; /* payload is large enough for ZC */
278 unsigned short tx_zc_checked:1; /* Have I checked if I should ZC? */
279 unsigned short tx_nonblk:1; /* it's a non-blocking ACK */
280 struct bio_vec *tx_kiov; /* packet page frags */
281 struct ksock_conn *tx_conn; /* owning conn */
282 struct lnet_msg *tx_lnetmsg; /* lnet message for lnet_finalize() */
283 time64_t tx_deadline; /* when (in secs) tx times out */
284 struct ksock_msg tx_msg; /* socklnd message buffer */
285 int tx_desc_size; /* size of this descriptor */
286 enum lnet_msg_hstatus tx_hstatus; /* health status of tx */
287 struct kvec tx_hdr; /* virt hdr */
288 struct bio_vec tx_payload[0]; /* paged payload */
291 #define KSOCK_NOOP_TX_SIZE ((int)offsetof(struct ksock_tx, tx_payload[0]))
293 /* space for the rx frag descriptors; we either read a single contiguous
294 * header, or up to LNET_MAX_IOV frags of payload of either type. */
295 union ksock_rxiovspace {
296 struct kvec iov[LNET_MAX_IOV];
297 struct bio_vec kiov[LNET_MAX_IOV];
300 #define SOCKNAL_RX_KSM_HEADER 1 /* reading ksock message header */
301 #define SOCKNAL_RX_LNET_HEADER 2 /* reading lnet message header */
302 #define SOCKNAL_RX_PARSE 3 /* Calling lnet_parse() */
303 #define SOCKNAL_RX_PARSE_WAIT 4 /* waiting to be told to read the body */
304 #define SOCKNAL_RX_LNET_PAYLOAD 5 /* reading lnet payload (to deliver here) */
305 #define SOCKNAL_RX_SLOP 6 /* skipping body */
308 struct ksock_peer_ni *ksnc_peer; /* owning peer_ni */
309 struct ksock_conn_cb *ksnc_conn_cb; /* owning conn control block */
310 struct list_head ksnc_list; /* on peer_ni's conn list */
311 struct socket *ksnc_sock; /* actual socket */
312 void *ksnc_saved_data_ready; /* socket's original
314 void *ksnc_saved_write_space; /* socket's original
315 * write_space() cb */
316 refcount_t ksnc_conn_refcount; /* conn refcount */
317 refcount_t ksnc_sock_refcount; /* sock refcount */
318 struct ksock_sched *ksnc_scheduler; /* who schedules this
320 struct sockaddr_storage ksnc_myaddr; /* my address */
321 struct sockaddr_storage ksnc_peeraddr; /* peer_ni's address */
322 signed int ksnc_type:3; /* type of connection,
325 unsigned int ksnc_closing:1; /* being shut down */
326 unsigned int ksnc_flip:1; /* flip or not, only for V2.x */
327 unsigned int ksnc_zc_capable:1; /* enable to ZC */
328 const struct ksock_proto *ksnc_proto; /* protocol for the connection */
332 /* where I enq waiting input or a forwarding descriptor */
333 struct list_head ksnc_rx_list;
334 time64_t ksnc_rx_deadline; /* when (in seconds) receive times out */
335 __u8 ksnc_rx_started; /* started receiving a message */
336 __u8 ksnc_rx_ready; /* data ready to read */
337 __u8 ksnc_rx_scheduled;/* being progressed */
338 __u8 ksnc_rx_state; /* what is being read */
339 int ksnc_rx_nob_left; /* # bytes to next hdr/body */
340 int ksnc_rx_nob_wanted; /* bytes actually wanted */
341 int ksnc_rx_niov; /* # kvec frags */
342 struct kvec *ksnc_rx_iov; /* the kvec frags */
343 int ksnc_rx_nkiov; /* # page frags */
344 struct bio_vec *ksnc_rx_kiov; /* the page frags */
345 union ksock_rxiovspace ksnc_rx_iov_space;/* space for frag descriptors */
346 __u32 ksnc_rx_csum; /* partial checksum for incoming
348 struct lnet_msg *ksnc_lnet_msg; /* rx lnet_finalize arg*/
349 struct ksock_msg ksnc_msg; /* incoming message buffer:
350 * V2.x message takes the
352 * V1.x message is a bare
353 * struct lnet_hdr_nid4, it's
355 * ksnc_msg.ksm_u.lnetmsg
358 /* where I enq waiting for output space */
359 struct list_head ksnc_tx_list;
360 /* packets waiting to be sent */
361 struct list_head ksnc_tx_queue;
362 /* next TX that can carry a LNet message or ZC-ACK */
363 struct ksock_tx *ksnc_tx_carrier;
364 /* when (in seconds) tx times out */
365 time64_t ksnc_tx_deadline;
366 /* send buffer marker */
369 atomic_t ksnc_tx_nob;
372 /* being progressed */
373 int ksnc_tx_scheduled;
374 /* time stamp of the last posted TX */
375 time64_t ksnc_tx_last_post;
378 #define SOCKNAL_CONN_COUNT_MAX_BITS 8 /* max conn count bits */
379 #define SOCKNAL_MAX_BUSY_RETRIES 3
381 struct ksock_conn_cb {
382 struct list_head ksnr_connd_list;/* chain on ksnr_connd_routes */
383 struct ksock_peer_ni *ksnr_peer; /* owning peer_ni */
384 refcount_t ksnr_refcount; /* # users */
385 time64_t ksnr_timeout; /* when (in secs) reconnection
388 time64_t ksnr_retry_interval;/* secs between retries */
389 struct sockaddr_storage ksnr_addr; /* IP address to connect to */
390 unsigned int ksnr_scheduled:1;/* scheduled for attention */
391 unsigned int ksnr_connecting:1;/* connection in progress */
392 unsigned int ksnr_connected:4;/* connections by type */
393 unsigned int ksnr_deleted:1; /* been removed from peer_ni? */
394 unsigned int ksnr_ctrl_conn_count:2; /* # conns by type */
395 unsigned int ksnr_blki_conn_count:8;
396 unsigned int ksnr_blko_conn_count:8;
397 int ksnr_conn_count;/* total # conns for this cb */
398 unsigned int ksnr_max_conns; /* conns_per_peer at peer
401 unsigned int ksnr_busy_retry_count;/* counts retry attempts
406 #define SOCKNAL_KEEPALIVE_PING 1 /* cookie for keepalive ping */
408 struct ksock_peer_ni {
409 struct hlist_node ksnp_list; /* stash on global peer_ni list */
410 time64_t ksnp_last_alive;/* when (in seconds) I was last alive */
411 struct lnet_processid ksnp_id; /* who's on the other end(s) */
412 refcount_t ksnp_refcount; /* # users */
413 int ksnp_closing; /* being closed */
414 int ksnp_accepting; /* # passive connections pending */
415 int ksnp_error; /* errno on closing last conn */
416 __u64 ksnp_zc_next_cookie;/* ZC completion cookie */
417 __u64 ksnp_incarnation; /* latest known peer_ni incarnation */
418 const struct ksock_proto *ksnp_proto; /* latest known protocol */
419 struct list_head ksnp_conns; /* all active connections */
420 struct ksock_conn_cb *ksnp_conn_cb; /* conn control block */
421 struct list_head ksnp_tx_queue; /* waiting packets */
422 spinlock_t ksnp_lock; /* serialize, g_lock unsafe */
423 /* zero copy requests wait for ACK */
424 struct list_head ksnp_zc_req_list;
425 time64_t ksnp_send_keepalive; /* time to send keepalive */
426 struct lnet_ni *ksnp_ni; /* which network */
429 struct ksock_connreq {
430 /* stash on ksnd_connd_connreqs */
431 struct list_head ksncr_list;
433 struct lnet_ni *ksncr_ni;
434 /* accepted socket */
435 struct socket *ksncr_sock;
438 extern struct ksock_nal_data ksocknal_data;
439 extern struct ksock_tunables ksocknal_tunables;
440 extern struct lnet_ioctl_config_socklnd_tunables ksock_default_tunables;
442 #define SOCKNAL_MATCH_NO 0 /* TX can't match type of connection */
443 #define SOCKNAL_MATCH_YES 1 /* TX matches type of connection */
444 #define SOCKNAL_MATCH_MAY 2 /* TX can be sent on the connection, but not preferred */
447 int pro_version; /* version number of protocol */
448 int (*pro_send_hello)(struct ksock_conn *, struct ksock_hello_msg *); /* handshake function */
449 int (*pro_recv_hello)(struct ksock_conn *, struct ksock_hello_msg *, int);/* handshake function */
450 void (*pro_pack)(struct ksock_tx *); /* message pack */
451 void (*pro_unpack)(struct ksock_msg *, struct lnet_hdr *); /* message unpack */
452 struct ksock_tx *(*pro_queue_tx_msg)(struct ksock_conn *, struct ksock_tx *); /* queue tx on the connection */
453 int (*pro_queue_tx_zcack)(struct ksock_conn *, struct ksock_tx *, __u64); /* queue ZC ack on the connection */
454 int (*pro_handle_zcreq)(struct ksock_conn *, __u64, int); /* handle ZC request */
455 int (*pro_handle_zcack)(struct ksock_conn *, __u64, __u64); /* handle ZC ACK */
456 int (*pro_match_tx)(struct ksock_conn *, struct ksock_tx *, int); /* msg type matches the connection type:
458 * return MATCH_NO : no
459 * return MATCH_YES : matching type
460 * return MATCH_MAY : can be backup */
463 extern const struct ksock_proto ksocknal_protocol_v1x;
464 extern const struct ksock_proto ksocknal_protocol_v2x;
465 extern const struct ksock_proto ksocknal_protocol_v3x;
466 extern const struct ksock_proto ksocknal_protocol_v4x;
468 #define KSOCK_PROTO_V1_MAJOR LNET_PROTO_TCP_VERSION_MAJOR
469 #define KSOCK_PROTO_V1_MINOR LNET_PROTO_TCP_VERSION_MINOR
470 #define KSOCK_PROTO_V1 KSOCK_PROTO_V1_MAJOR
472 #ifndef CPU_MASK_NONE
473 #define CPU_MASK_NONE 0UL
476 #if LINUX_VERSION_CODE < KERNEL_VERSION(3, 11, 0)
477 #undef netdev_notifier_info_to_dev
478 #define netdev_notifier_info_to_dev(ndev) ndev
481 static inline __u32 ksocknal_csum(__u32 crc, unsigned char const *p, size_t len)
484 return crc32_le(crc, p, len);
487 crc = ((crc + 0x100) & ~0xff) | ((crc + *p++) & 0xff) ;
494 ksocknal_conn_cb_mask(void)
496 if (!*ksocknal_tunables.ksnd_typed_conns)
497 return BIT(SOCKLND_CONN_ANY);
499 return (BIT(SOCKLND_CONN_CONTROL) |
500 BIT(SOCKLND_CONN_BULK_IN) |
501 BIT(SOCKLND_CONN_BULK_OUT));
505 ksocknal_conn_addref(struct ksock_conn *conn)
507 refcount_inc(&conn->ksnc_conn_refcount);
510 extern void ksocknal_queue_zombie_conn(struct ksock_conn *conn);
511 extern void ksocknal_finalize_zcreq(struct ksock_conn *conn);
514 ksocknal_conn_decref(struct ksock_conn *conn)
516 if (refcount_dec_and_test(&conn->ksnc_conn_refcount))
517 ksocknal_queue_zombie_conn(conn);
521 ksocknal_connsock_addref(struct ksock_conn *conn)
525 read_lock(&ksocknal_data.ksnd_global_lock);
526 if (!conn->ksnc_closing) {
527 refcount_inc(&conn->ksnc_sock_refcount);
530 read_unlock(&ksocknal_data.ksnd_global_lock);
536 ksocknal_connsock_decref(struct ksock_conn *conn)
538 if (refcount_dec_and_test(&conn->ksnc_sock_refcount)) {
539 LASSERT (conn->ksnc_closing);
540 sock_release(conn->ksnc_sock);
541 conn->ksnc_sock = NULL;
542 ksocknal_finalize_zcreq(conn);
547 ksocknal_tx_addref(struct ksock_tx *tx)
549 refcount_inc(&tx->tx_refcount);
552 extern void ksocknal_tx_prep(struct ksock_conn *, struct ksock_tx *tx);
553 extern void ksocknal_tx_done(struct lnet_ni *ni, struct ksock_tx *tx, int error);
556 ksocknal_tx_decref(struct ksock_tx *tx)
558 if (refcount_dec_and_test(&tx->tx_refcount))
559 ksocknal_tx_done(NULL, tx, 0);
563 ksocknal_conn_cb_addref(struct ksock_conn_cb *conn_cb)
565 refcount_inc(&conn_cb->ksnr_refcount);
568 extern void ksocknal_destroy_conn_cb(struct ksock_conn_cb *conn_cb);
571 ksocknal_conn_cb_decref(struct ksock_conn_cb *conn_cb)
573 if (refcount_dec_and_test(&conn_cb->ksnr_refcount))
574 ksocknal_destroy_conn_cb(conn_cb);
578 ksocknal_peer_addref(struct ksock_peer_ni *peer_ni)
580 refcount_inc(&peer_ni->ksnp_refcount);
583 extern void ksocknal_destroy_peer(struct ksock_peer_ni *peer_ni);
586 ksocknal_peer_decref(struct ksock_peer_ni *peer_ni)
588 if (refcount_dec_and_test(&peer_ni->ksnp_refcount))
589 ksocknal_destroy_peer(peer_ni);
592 static inline int ksocknal_timeout(void)
594 return *ksocknal_tunables.ksnd_timeout ?: lnet_get_lnd_timeout();
597 static inline int ksocknal_conns_per_peer(void)
599 return *ksocknal_tunables.ksnd_conns_per_peer ?: 1;
602 int ksocknal_startup(struct lnet_ni *ni);
603 void ksocknal_shutdown(struct lnet_ni *ni);
604 int ksocknal_ctl(struct lnet_ni *ni, unsigned int cmd, void *arg);
605 int ksocknal_send(struct lnet_ni *ni, void *private, struct lnet_msg *lntmsg);
606 int ksocknal_recv(struct lnet_ni *ni, void *private, struct lnet_msg *lntmsg,
607 int delayed, unsigned int niov,
608 struct bio_vec *kiov,
609 unsigned int offset, unsigned int mlen, unsigned int rlen);
610 int ksocknal_accept(struct lnet_ni *ni, struct socket *sock);
612 unsigned int ksocknal_get_conn_count_by_type(struct ksock_conn_cb *conn_cb,
614 int ksocknal_add_peer(struct lnet_ni *ni, struct lnet_processid *id,
615 struct sockaddr *addr);
616 struct ksock_peer_ni *ksocknal_find_peer_locked(struct lnet_ni *ni,
617 struct lnet_processid *id);
618 struct ksock_peer_ni *ksocknal_find_peer(struct lnet_ni *ni,
619 struct lnet_processid *id);
620 extern void ksocknal_peer_failed(struct ksock_peer_ni *peer_ni);
621 extern int ksocknal_create_conn(struct lnet_ni *ni,
622 struct ksock_conn_cb *conn_cb,
623 struct socket *sock, int type);
624 extern void ksocknal_close_conn_locked(struct ksock_conn *conn, int why);
625 extern void ksocknal_terminate_conn(struct ksock_conn *conn);
626 extern void ksocknal_destroy_conn(struct ksock_conn *conn);
627 extern int ksocknal_close_peer_conns_locked(struct ksock_peer_ni *peer_ni,
628 struct sockaddr *peer, int why);
629 extern int ksocknal_close_conn_and_siblings(struct ksock_conn *conn, int why);
630 int ksocknal_close_matching_conns(struct lnet_processid *id, __u32 ipaddr);
631 extern struct ksock_conn *ksocknal_find_conn_locked(struct ksock_peer_ni *peer_ni,
632 struct ksock_tx *tx, int nonblk);
634 extern int ksocknal_launch_packet(struct lnet_ni *ni, struct ksock_tx *tx,
635 struct lnet_processid *id);
636 extern struct ksock_tx *ksocknal_alloc_tx(int type, int size);
637 extern void ksocknal_free_tx(struct ksock_tx *tx);
638 extern struct ksock_tx *ksocknal_alloc_tx_noop(__u64 cookie, int nonblk);
639 extern void ksocknal_next_tx_carrier(struct ksock_conn *conn);
640 extern void ksocknal_queue_tx_locked(struct ksock_tx *tx, struct ksock_conn *conn);
641 extern void ksocknal_txlist_done(struct lnet_ni *ni, struct list_head *txlist,
643 #define ksocknal_thread_start(fn, data, namefmt, arg...) \
645 struct task_struct *__task = kthread_run(fn, data, \
647 if (!IS_ERR(__task)) \
648 atomic_inc(&ksocknal_data.ksnd_nthreads); \
649 PTR_ERR_OR_ZERO(__task); \
652 extern void ksocknal_thread_fini(void);
653 extern void ksocknal_launch_all_connections_locked(struct ksock_peer_ni *peer_ni);
654 extern struct ksock_conn_cb *ksocknal_find_connectable_conn_cb_locked(struct ksock_peer_ni *peer_ni);
655 extern struct ksock_conn_cb *ksocknal_find_connecting_conn_cb_locked(struct ksock_peer_ni *peer_ni);
656 extern int ksocknal_new_packet(struct ksock_conn *conn, int skip);
657 extern int ksocknal_scheduler(void *arg);
658 extern int ksocknal_connd(void *arg);
659 extern int ksocknal_reaper(void *arg);
660 int ksocknal_send_hello(struct lnet_ni *ni, struct ksock_conn *conn,
661 struct lnet_nid *peer_nid,
662 struct ksock_hello_msg *hello);
663 int ksocknal_recv_hello(struct lnet_ni *ni, struct ksock_conn *conn,
664 struct ksock_hello_msg *hello,
665 struct lnet_processid *id,
667 extern void ksocknal_read_callback(struct ksock_conn *conn);
668 extern void ksocknal_write_callback(struct ksock_conn *conn);
670 extern int ksocknal_lib_zc_capable(struct ksock_conn *conn);
671 extern void ksocknal_lib_save_callback(struct socket *sock, struct ksock_conn *conn);
672 extern void ksocknal_lib_set_callback(struct socket *sock, struct ksock_conn *conn);
673 extern void ksocknal_lib_reset_callback(struct socket *sock,
674 struct ksock_conn *conn);
675 extern void ksocknal_lib_push_conn(struct ksock_conn *conn);
676 extern int ksocknal_lib_get_conn_addrs(struct ksock_conn *conn);
677 extern int ksocknal_lib_setup_sock(struct socket *sock, struct lnet_ni *ni);
678 extern int ksocknal_lib_send_hdr(struct ksock_conn *conn, struct ksock_tx *tx,
679 struct kvec *scratch_iov);
680 extern int ksocknal_lib_send_kiov(struct ksock_conn *conn, struct ksock_tx *tx,
681 struct kvec *scratch_iov);
682 extern void ksocknal_lib_eager_ack(struct ksock_conn *conn);
683 extern int ksocknal_lib_recv_iov(struct ksock_conn *conn,
684 struct kvec *scratchiov);
685 extern int ksocknal_lib_recv_kiov(struct ksock_conn *conn, struct page **pages,
686 struct kvec *scratchiov);
687 extern int ksocknal_lib_get_conn_tunables(struct ksock_conn *conn, int *txmem,
688 int *rxmem, int *nagle);
690 extern int ksocknal_tunables_init(void);
691 extern void ksocknal_tunables_setup(struct lnet_ni *ni);
693 extern void ksocknal_lib_csum_tx(struct ksock_tx *tx);
695 extern int ksocknal_lib_memory_pressure(struct ksock_conn *conn);
697 #endif /* _SOCKLND_SOCKLND_H_ */