1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2001, 2002 Cluster File Systems, Inc.
5 * Author: Zach Brown <zab@zabbo.net>
6 * Author: Peter J. Braam <braam@clusterfs.com>
7 * Author: Phil Schwan <phil@clusterfs.com>
8 * Author: Eric Barton <eric@bartonsoftware.com>
10 * This file is part of Portals, http://www.sf.net/projects/lustre/
12 * Portals is free software; you can redistribute it and/or
13 * modify it under the terms of version 2 of the GNU General Public
14 * License as published by the Free Software Foundation.
16 * Portals is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
19 * GNU General Public License for more details.
21 * You should have received a copy of the GNU General Public License
22 * along with Portals; if not, write to the Free Software
23 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
27 #define DEBUG_PORTAL_ALLOC
29 # define EXPORT_SYMTAB
32 #include <linux/config.h>
33 #include <linux/module.h>
34 #include <linux/kernel.h>
36 #include <linux/string.h>
37 #include <linux/stat.h>
38 #include <linux/errno.h>
39 #include <linux/smp_lock.h>
40 #include <linux/unistd.h>
43 #include <linux/uio.h>
45 #include <asm/system.h>
46 #include <asm/uaccess.h>
49 #include <linux/init.h>
51 #include <linux/file.h>
52 #include <linux/stat.h>
53 #include <linux/list.h>
54 #include <linux/kmod.h>
55 #include <linux/sysctl.h>
56 #include <asm/uaccess.h>
57 #include <asm/segment.h>
58 #include <asm/div64.h>
60 #define DEBUG_SUBSYSTEM S_SOCKNAL
62 #include <linux/kp30.h>
63 #include <linux/portals_compat25.h>
64 #include <linux/kpr.h>
65 #include <portals/p30.h>
66 #include <portals/lib-p30.h>
67 #include <portals/nal.h>
68 #include <portals/socknal.h>
70 #define SOCKNAL_N_AUTOCONNECTD 4 /* # socknal autoconnect daemons */
72 #define SOCKNAL_MIN_RECONNECT_INTERVAL HZ /* first failed connection retry... */
73 #define SOCKNAL_MAX_RECONNECT_INTERVAL (60*HZ) /* ...exponentially increasing to this */
75 /* default vals for runtime tunables */
76 #define SOCKNAL_IO_TIMEOUT 50 /* default comms timeout (seconds) */
77 #define SOCKNAL_EAGER_ACK 0 /* default eager ack (boolean) */
78 #define SOCKNAL_TYPED_CONNS 1 /* unidirectional large, bidirectional small? */
79 #define SOCKNAL_ZC_MIN_FRAG (2<<10) /* default smallest zerocopy fragment */
80 #define SOCKNAL_MIN_BULK (1<<10) /* smallest "large" message */
81 #define SOCKNAL_BUFFER_SIZE (8<<20) /* default socket buffer size */
82 #define SOCKNAL_NAGLE 0 /* enable/disable NAGLE? */
83 #define SOCKNAL_IRQ_AFFINITY 1 /* enable/disable IRQ affinity? */
84 #define SOCKNAL_KEEPALIVE_IDLE 0 /* # seconds idle before 1st probe */
85 #define SOCKNAL_KEEPALIVE_COUNT 10 /* # unanswered probes to determine peer death */
86 #define SOCKNAL_KEEPALIVE_INTVL 1 /* seconds between probes */
88 #define SOCKNAL_PEER_HASH_SIZE 101 /* # peer lists */
90 #define SOCKNAL_SMALL_FWD_NMSGS 128 /* # small messages I can be forwarding at any time */
91 #define SOCKNAL_LARGE_FWD_NMSGS 64 /* # large messages I can be forwarding at any time */
93 #define SOCKNAL_SMALL_FWD_PAGES 1 /* # pages in a small message fwd buffer */
95 #define SOCKNAL_LARGE_FWD_PAGES (PAGE_ALIGN(PTL_MTU) >> PAGE_SHIFT)
96 /* # pages in a large message fwd buffer */
98 #define SOCKNAL_RESCHED 100 /* # scheduler loops before reschedule */
99 #define SOCKNAL_ENOMEM_RETRY 1 /* jiffies between retries */
101 #define SOCKNAL_MAX_INTERFACES 16 /* Largest number of interfaces we bind */
103 #define SOCKNAL_ROUND_ROBIN 0 /* round robin / load balance */
105 #define SOCKNAL_TX_LOW_WATER(sk) (((sk)->sk_sndbuf*8)/10)
107 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,72))
108 # define sk_allocation allocation
109 # define sk_data_ready data_ready
110 # define sk_write_space write_space
111 # define sk_user_data user_data
112 # define sk_prot prot
113 # define sk_sndbuf sndbuf
114 # define sk_socket socket
117 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0))
118 # define sk_wmem_queued wmem_queued
122 typedef struct /* pool of forwarding buffers */
124 spinlock_t fmp_lock; /* serialise */
125 struct list_head fmp_idle_fmbs; /* free buffers */
126 struct list_head fmp_blocked_conns; /* connections waiting for a buffer */
127 int fmp_nactive_fmbs; /* # buffers in use */
128 int fmp_buff_pages; /* # pages per buffer */
132 typedef struct /* per scheduler state */
134 spinlock_t kss_lock; /* serialise */
135 struct list_head kss_rx_conns; /* conn waiting to be read */
136 struct list_head kss_tx_conns; /* conn waiting to be written */
138 struct list_head kss_zctxdone_list; /* completed ZC transmits */
140 wait_queue_head_t kss_waitq; /* where scheduler sleeps */
141 int kss_nconns; /* # connections assigned to this scheduler */
146 int ksni_valid:1; /* been set yet? */
147 int ksni_bound:1; /* bound to a cpu yet? */
148 int ksni_sched:6; /* which scheduler (assumes < 64) */
153 __u32 ksni_ipaddr; /* interface's IP address */
154 __u32 ksni_netmask; /* interface's network mask */
155 int ksni_nroutes; /* # routes using (active) */
156 int ksni_npeers; /* # peers using (passive) */
161 int ksnd_io_timeout; /* "stuck" socket timeout (seconds) */
162 int ksnd_eager_ack; /* make TCP ack eagerly? */
163 int ksnd_typed_conns; /* drive sockets by type? */
164 int ksnd_min_bulk; /* smallest "large" message */
165 int ksnd_buffer_size; /* socket buffer size */
166 int ksnd_nagle; /* enable NAGLE? */
167 int ksnd_irq_affinity; /* enable IRQ affinity? */
168 int ksnd_keepalive_idle; /* # idle secs before 1st probe */
169 int ksnd_keepalive_count; /* # probes */
170 int ksnd_keepalive_intvl; /* time between probes */
172 unsigned int ksnd_zc_min_frag; /* minimum zero copy frag size */
174 struct ctl_table_header *ksnd_sysctl; /* sysctl interface */
179 int ksnd_init; /* initialisation state */
180 __u64 ksnd_incarnation; /* my epoch */
182 rwlock_t ksnd_global_lock; /* stabilize peer/conn ops */
183 struct list_head *ksnd_peers; /* hash table of all my known peers */
184 int ksnd_peer_hash_size; /* size of ksnd_peers */
186 int ksnd_nthreads; /* # live threads */
187 int ksnd_shuttingdown; /* tell threads to exit */
188 int ksnd_nschedulers; /* # schedulers */
189 ksock_sched_t *ksnd_schedulers; /* their state */
191 atomic_t ksnd_npeers; /* total # peers extant */
192 atomic_t ksnd_nclosing_conns; /* # closed conns extant */
194 kpr_router_t ksnd_router; /* THE router */
196 ksock_fmb_pool_t ksnd_small_fmp; /* small message forwarding buffers */
197 ksock_fmb_pool_t ksnd_large_fmp; /* large message forwarding buffers */
199 atomic_t ksnd_nactive_ltxs; /* #active ltxs */
201 struct list_head ksnd_deathrow_conns; /* conns to be closed */
202 struct list_head ksnd_zombie_conns; /* conns to be freed */
203 struct list_head ksnd_enomem_conns; /* conns to be retried */
204 wait_queue_head_t ksnd_reaper_waitq; /* reaper sleeps here */
205 unsigned long ksnd_reaper_waketime; /* when reaper will wake */
206 spinlock_t ksnd_reaper_lock; /* serialise */
208 int ksnd_enomem_tx; /* test ENOMEM sender */
209 int ksnd_stall_tx; /* test sluggish sender */
210 int ksnd_stall_rx; /* test sluggish receiver */
212 struct list_head ksnd_autoconnectd_routes; /* routes waiting to be connected */
213 wait_queue_head_t ksnd_autoconnectd_waitq; /* autoconnectds sleep here */
214 spinlock_t ksnd_autoconnectd_lock; /* serialise */
216 ksock_irqinfo_t ksnd_irqinfo[NR_IRQS];/* irq->scheduler lookup */
218 int ksnd_ninterfaces;
219 ksock_interface_t ksnd_interfaces[SOCKNAL_MAX_INTERFACES]; /* published interfaces */
222 #define SOCKNAL_INIT_NOTHING 0
223 #define SOCKNAL_INIT_DATA 1
224 #define SOCKNAL_INIT_LIB 2
225 #define SOCKNAL_INIT_ALL 3
227 /* A packet just assembled for transmission is represented by 1 or more
228 * struct iovec fragments (the first frag contains the portals header),
229 * followed by 0 or more ptl_kiov_t fragments.
231 * On the receive side, initially 1 struct iovec fragment is posted for
232 * receive (the header). Once the header has been received, the payload is
233 * received into either struct iovec or ptl_kiov_t fragments, depending on
234 * what the header matched or whether the message needs forwarding. */
236 struct ksock_conn; /* forward ref */
237 struct ksock_peer; /* forward ref */
238 struct ksock_route; /* forward ref */
240 typedef struct /* transmit packet */
242 struct list_head tx_list; /* queue on conn for transmission etc */
243 char tx_isfwd; /* forwarding / sourced here */
244 int tx_nob; /* # packet bytes */
245 int tx_resid; /* residual bytes */
246 int tx_niov; /* # packet iovec frags */
247 struct iovec *tx_iov; /* packet iovec frags */
248 int tx_nkiov; /* # packet page frags */
249 ptl_kiov_t *tx_kiov; /* packet page frags */
250 struct ksock_conn *tx_conn; /* owning conn */
251 ptl_hdr_t *tx_hdr; /* packet header (for debug only) */
253 zccd_t tx_zccd; /* zero copy callback descriptor */
257 typedef struct /* forwarded packet */
259 ksock_tx_t ftx_tx; /* send info */
260 struct iovec ftx_iov; /* hdr iovec */
263 #define KSOCK_ZCCD_2_TX(ptr) list_entry (ptr, ksock_tx_t, tx_zccd)
264 /* network zero copy callback descriptor embedded in ksock_tx_t */
266 typedef struct /* locally transmitted packet */
268 ksock_tx_t ltx_tx; /* send info */
269 void *ltx_private; /* lib_finalize() callback arg */
270 void *ltx_cookie; /* lib_finalize() callback arg */
271 ptl_hdr_t ltx_hdr; /* buffer for packet header */
272 int ltx_desc_size; /* bytes allocated for this desc */
273 struct iovec ltx_iov[1]; /* iov for hdr + payload */
274 ptl_kiov_t ltx_kiov[0]; /* kiov for payload */
277 #define KSOCK_TX_2_KPR_FWD_DESC(ptr) list_entry ((kprfd_scratch_t *)ptr, kpr_fwd_desc_t, kprfd_scratch)
278 /* forwarded packets (router->socknal) embedded in kpr_fwd_desc_t::kprfd_scratch */
280 #define KSOCK_TX_2_KSOCK_LTX(ptr) list_entry (ptr, ksock_ltx_t, ltx_tx)
281 /* local packets (lib->socknal) embedded in ksock_ltx_t::ltx_tx */
283 /* NB list_entry() is used here as convenient macro for calculating a
284 * pointer to a struct from the address of a member. */
286 typedef struct /* Kernel portals Socket Forwarding message buffer */
287 { /* (socknal->router) */
288 struct list_head fmb_list; /* queue idle */
289 kpr_fwd_desc_t fmb_fwd; /* router's descriptor */
290 ksock_fmb_pool_t *fmb_pool; /* owning pool */
291 struct ksock_peer *fmb_peer; /* peer received from */
292 ptl_hdr_t fmb_hdr; /* message header */
293 ptl_kiov_t fmb_kiov[0]; /* payload frags */
296 /* space for the rx frag descriptors; we either read a single contiguous
297 * header, or up to PTL_MD_MAX_IOV frags of payload of either type. */
299 struct iovec iov[PTL_MD_MAX_IOV];
300 ptl_kiov_t kiov[PTL_MD_MAX_IOV];
301 } ksock_rxiovspace_t;
303 #define SOCKNAL_RX_HEADER 1 /* reading header */
304 #define SOCKNAL_RX_BODY 2 /* reading body (to deliver here) */
305 #define SOCKNAL_RX_BODY_FWD 3 /* reading body (to forward) */
306 #define SOCKNAL_RX_SLOP 4 /* skipping body */
307 #define SOCKNAL_RX_GET_FMB 5 /* scheduled for forwarding */
308 #define SOCKNAL_RX_FMB_SLEEP 6 /* blocked waiting for a fwd desc */
310 typedef struct ksock_conn
312 struct ksock_peer *ksnc_peer; /* owning peer */
313 struct ksock_route *ksnc_route; /* owning route */
314 struct list_head ksnc_list; /* stash on peer's conn list */
315 struct socket *ksnc_sock; /* actual socket */
316 void *ksnc_saved_data_ready; /* socket's original data_ready() callback */
317 void *ksnc_saved_write_space; /* socket's original write_space() callback */
318 atomic_t ksnc_refcount; /* # users */
319 ksock_sched_t *ksnc_scheduler; /* who schedules this connection */
320 __u32 ksnc_myipaddr; /* my IP */
321 __u32 ksnc_ipaddr; /* peer's IP */
322 int ksnc_port; /* peer's port */
323 int ksnc_closing; /* being shut down */
324 int ksnc_type; /* type of connection */
325 __u64 ksnc_incarnation; /* peer's incarnation */
328 struct list_head ksnc_rx_list; /* where I enq waiting input or a forwarding descriptor */
329 unsigned long ksnc_rx_deadline; /* when (in jiffies) receive times out */
330 int ksnc_rx_started; /* started receiving a message */
331 int ksnc_rx_ready; /* data ready to read */
332 int ksnc_rx_scheduled; /* being progressed */
333 int ksnc_rx_state; /* what is being read */
334 int ksnc_rx_nob_left; /* # bytes to next hdr/body */
335 int ksnc_rx_nob_wanted; /* bytes actually wanted */
336 int ksnc_rx_niov; /* # iovec frags */
337 struct iovec *ksnc_rx_iov; /* the iovec frags */
338 int ksnc_rx_nkiov; /* # page frags */
339 ptl_kiov_t *ksnc_rx_kiov; /* the page frags */
340 ksock_rxiovspace_t ksnc_rx_iov_space; /* space for frag descriptors */
341 void *ksnc_cookie; /* rx lib_finalize passthru arg */
342 ptl_hdr_t ksnc_hdr; /* where I read headers into */
345 struct list_head ksnc_tx_list; /* where I enq waiting for output space */
346 struct list_head ksnc_tx_queue; /* packets waiting to be sent */
347 unsigned long ksnc_tx_deadline; /* when (in jiffies) tx times out */
348 int ksnc_tx_bufnob; /* send buffer marker */
349 atomic_t ksnc_tx_nob; /* # bytes queued */
350 int ksnc_tx_ready; /* write space */
351 int ksnc_tx_scheduled; /* being progressed */
354 #define KSNR_TYPED_ROUTES ((1 << SOCKNAL_CONN_CONTROL) | \
355 (1 << SOCKNAL_CONN_BULK_IN) | \
356 (1 << SOCKNAL_CONN_BULK_OUT))
358 typedef struct ksock_route
360 struct list_head ksnr_list; /* chain on peer route list */
361 struct list_head ksnr_connect_list; /* chain on autoconnect list */
362 struct ksock_peer *ksnr_peer; /* owning peer */
363 atomic_t ksnr_refcount; /* # users */
364 unsigned long ksnr_timeout; /* when (in jiffies) reconnection can happen next */
365 unsigned int ksnr_retry_interval; /* how long between retries */
366 __u32 ksnr_myipaddr; /* my IP */
367 __u32 ksnr_ipaddr; /* IP address to connect to */
368 int ksnr_port; /* port to connect to */
369 unsigned int ksnr_connecting:4; /* autoconnects in progress by type */
370 unsigned int ksnr_connected:4; /* connections established by type */
371 unsigned int ksnr_deleted:1; /* been removed from peer? */
372 unsigned int ksnr_share_count; /* created explicitly? */
373 int ksnr_conn_count; /* # conns established by this route */
376 typedef struct ksock_peer
378 struct list_head ksnp_list; /* stash on global peer list */
379 ptl_nid_t ksnp_nid; /* who's on the other end(s) */
380 atomic_t ksnp_refcount; /* # users */
381 int ksnp_sharecount; /* lconf usage counter */
382 int ksnp_closing; /* being closed */
383 int ksnp_error; /* errno on closing last conn */
384 struct list_head ksnp_conns; /* all active connections */
385 struct list_head ksnp_routes; /* routes */
386 struct list_head ksnp_tx_queue; /* waiting packets */
387 unsigned long ksnp_last_alive; /* when (in jiffies) I was last alive */
388 int ksnp_n_passive_ips; /* # of... */
389 __u32 ksnp_passive_ips[SOCKNAL_MAX_INTERFACES]; /* preferred local interfaces */
393 extern lib_nal_t ksocknal_lib;
394 extern ksock_nal_data_t ksocknal_data;
395 extern ksock_tunables_t ksocknal_tunables;
397 static inline struct list_head *
398 ksocknal_nid2peerlist (ptl_nid_t nid)
400 unsigned int hash = ((unsigned int)nid) % ksocknal_data.ksnd_peer_hash_size;
402 return (&ksocknal_data.ksnd_peers [hash]);
406 ksocknal_getconnsock (ksock_conn_t *conn)
410 read_lock (&ksocknal_data.ksnd_global_lock);
411 if (!conn->ksnc_closing) {
413 get_file (conn->ksnc_sock->file);
415 read_unlock (&ksocknal_data.ksnd_global_lock);
421 ksocknal_putconnsock (ksock_conn_t *conn)
423 fput (conn->ksnc_sock->file);
428 int ksocknal_nsched(void)
433 #include <linux/lustre_version.h>
434 # if !(defined(CONFIG_X86) && (LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,21))) || defined(CONFIG_X86_64) || (LUSTRE_KERNEL_VERSION < 39) || ((LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)) && !defined(CONFIG_X86_HT))
436 ksocknal_nsched(void)
438 return num_online_cpus();
442 ksocknal_sched2cpu(int i)
448 ksocknal_irqsched2cpu(int i)
454 ksocknal_nsched(void)
456 if (smp_num_siblings == 1)
457 return (num_online_cpus());
459 /* We need to know if this assumption is crap */
460 LASSERT (smp_num_siblings == 2);
461 return (num_online_cpus()/2);
465 ksocknal_sched2cpu(int i)
467 if (smp_num_siblings == 1)
474 ksocknal_irqsched2cpu(int i)
476 return (ksocknal_sched2cpu(i) + 1);
481 extern void ksocknal_put_route (ksock_route_t *route);
482 extern void ksocknal_put_peer (ksock_peer_t *peer);
483 extern ksock_peer_t *ksocknal_find_peer_locked (ptl_nid_t nid);
484 extern ksock_peer_t *ksocknal_get_peer (ptl_nid_t nid);
485 extern int ksocknal_del_route (ptl_nid_t nid, __u32 ipaddr,
486 int single, int keep_conn);
487 extern int ksocknal_create_conn (ksock_route_t *route,
488 struct socket *sock, int type);
489 extern void ksocknal_close_conn_locked (ksock_conn_t *conn, int why);
490 extern void ksocknal_terminate_conn (ksock_conn_t *conn);
491 extern void ksocknal_destroy_conn (ksock_conn_t *conn);
492 extern void ksocknal_put_conn (ksock_conn_t *conn);
493 extern int ksocknal_close_stale_conns_locked (ksock_peer_t *peer, __u64 incarnation);
494 extern int ksocknal_close_conn_and_siblings (ksock_conn_t *conn, int why);
495 extern int ksocknal_close_matching_conns (ptl_nid_t nid, __u32 ipaddr);
497 extern void ksocknal_queue_tx_locked (ksock_tx_t *tx, ksock_conn_t *conn);
498 extern void ksocknal_tx_done (ksock_tx_t *tx, int asynch);
499 extern void ksocknal_fwd_packet (void *arg, kpr_fwd_desc_t *fwd);
500 extern void ksocknal_fmb_callback (void *arg, int error);
501 extern void ksocknal_notify (void *arg, ptl_nid_t gw_nid, int alive);
502 extern int ksocknal_thread_start (int (*fn)(void *arg), void *arg);
503 extern int ksocknal_new_packet (ksock_conn_t *conn, int skip);
504 extern int ksocknal_scheduler (void *arg);
505 extern void ksocknal_data_ready(struct sock *sk, int n);
506 extern void ksocknal_write_space(struct sock *sk);
507 extern int ksocknal_autoconnectd (void *arg);
508 extern int ksocknal_reaper (void *arg);
509 extern int ksocknal_get_conn_tunables (ksock_conn_t *conn, int *txmem,
510 int *rxmem, int *nagle);
511 extern int ksocknal_setup_sock (struct socket *sock);
512 extern int ksocknal_send_hello (ksock_conn_t *conn, __u32 *ipaddrs, int nipaddrs);
513 extern int ksocknal_recv_hello (ksock_conn_t *conn,
514 ptl_nid_t *nid, __u64 *incarnation, __u32 *ipaddrs);