1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Author: Zach Brown <zab@zabbo.net>
5 * Author: Peter J. Braam <braam@clusterfs.com>
6 * Author: Phil Schwan <phil@clusterfs.com>
7 * Author: Eric Barton <eric@bartonsoftware.com>
9 * This file is part of Lustre, http://www.lustre.org
11 * Portals is free software; you can redistribute it and/or
12 * modify it under the terms of version 2 of the GNU General Public
13 * License as published by the Free Software Foundation.
15 * Portals is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 * GNU General Public License for more details.
20 * You should have received a copy of the GNU General Public License
21 * along with Portals; if not, write to the Free Software
22 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
26 #define DEBUG_PORTAL_ALLOC
28 # define EXPORT_SYMTAB
31 #define DEBUG_SUBSYSTEM S_NAL
33 #if defined(__linux__)
34 #include "socknal_lib-linux.h"
35 #elif defined(__APPLE__)
36 #include "socknal_lib-darwin.h"
38 #error Unsupported Operating System
41 #include <libcfs/kp30.h>
42 #include <portals/kpr.h>
43 #include <portals/p30.h>
44 #include <portals/lib-p30.h>
45 #include <portals/nal.h>
46 #include <portals/socknal.h>
48 #define SOCKNAL_N_AUTOCONNECTD 4 /* # socknal autoconnect daemons */
50 #define SOCKNAL_MIN_RECONNECT_INTERVAL cfs_time_seconds(1) /* first failed connection retry... */
51 #define SOCKNAL_MAX_RECONNECT_INTERVAL cfs_time_seconds(60) /* ...exponentially increasing to this */
53 /* default vals for runtime tunables */
54 #define SOCKNAL_IO_TIMEOUT 50 /* default comms timeout (seconds) */
55 #define SOCKNAL_EAGER_ACK SOCKNAL_ARCH_EAGER_ACK /* default eager ack (boolean) */
56 #define SOCKNAL_TYPED_CONNS 1 /* unidirectional large, bidirectional small? */
57 #define SOCKNAL_ZC_MIN_FRAG (2<<10) /* default smallest zerocopy fragment */
58 #define SOCKNAL_MIN_BULK (1<<10) /* smallest "large" message */
59 #define SOCKNAL_BUFFER_SIZE (8<<20) /* default socket buffer size */
60 #define SOCKNAL_NAGLE 0 /* enable/disable NAGLE? */
61 #define SOCKNAL_IRQ_AFFINITY 1 /* enable/disable IRQ affinity? */
62 #define SOCKNAL_KEEPALIVE_IDLE 30 /* # seconds idle before 1st probe */
64 #define SOCKNAL_KEEPALIVE_COUNT 10 /* # unanswered probes to determine peer death */
65 #define SOCKNAL_KEEPALIVE_INTVL 2 /* seconds between probes */
67 #define SOCKNAL_PEER_HASH_SIZE 101 /* # peer lists */
69 #define SOCKNAL_SMALL_FWD_NMSGS 128 /* # small messages I can be forwarding at any time */
70 #define SOCKNAL_LARGE_FWD_NMSGS 64 /* # large messages I can be forwarding at any time */
72 #define SOCKNAL_SMALL_FWD_PAGES 1 /* # pages in a small message fwd buffer */
74 #define SOCKNAL_LARGE_FWD_PAGES (PAGE_ALIGN(PTL_MTU) >> PAGE_SHIFT)
75 /* # pages in a large message fwd buffer */
77 #define SOCKNAL_RESCHED 100 /* # scheduler loops before reschedule */
78 #define SOCKNAL_ENOMEM_RETRY CFS_MIN_DELAY /* jiffies between retries */
80 #define SOCKNAL_MAX_INTERFACES 16 /* Largest number of interfaces we bind */
82 #define SOCKNAL_ROUND_ROBIN 0 /* round robin / load balance */
84 #define SOCKNAL_SINGLE_FRAG_TX 0 /* disable multi-fragment sends */
85 #define SOCKNAL_SINGLE_FRAG_RX 0 /* disable multi-fragment receives */
87 /* risk kmap deadlock on multi-frag I/O (backs off to single-frag if disabled).
88 * no risk if we're not running on a CONFIG_HIGHMEM platform. */
90 # define SOCKNAL_RISK_KMAP_DEADLOCK 0
92 # define SOCKNAL_RISK_KMAP_DEADLOCK 1
95 typedef struct /* pool of forwarding buffers */
97 spinlock_t fmp_lock; /* serialise */
98 struct list_head fmp_idle_fmbs; /* free buffers */
99 struct list_head fmp_blocked_conns; /* connections waiting for a buffer */
100 int fmp_nactive_fmbs; /* # buffers in use */
101 int fmp_buff_pages; /* # pages per buffer */
105 typedef struct /* per scheduler state */
107 spinlock_t kss_lock; /* serialise */
108 struct list_head kss_rx_conns; /* conn waiting to be read */
109 struct list_head kss_tx_conns; /* conn waiting to be written */
111 struct list_head kss_zctxdone_list; /* completed ZC transmits */
113 cfs_waitq_t kss_waitq; /* where scheduler sleeps */
114 int kss_nconns; /* # connections assigned to this scheduler */
119 int ksni_valid:1; /* been set yet? */
120 int ksni_bound:1; /* bound to a cpu yet? */
121 int ksni_sched:6; /* which scheduler (assumes < 64) */
126 __u32 ksni_ipaddr; /* interface's IP address */
127 __u32 ksni_netmask; /* interface's network mask */
128 int ksni_nroutes; /* # routes using (active) */
129 int ksni_npeers; /* # peers using (passive) */
134 int ksnd_io_timeout; /* "stuck" socket timeout (seconds) */
135 int ksnd_eager_ack; /* make TCP ack eagerly? */
136 int ksnd_typed_conns; /* drive sockets by type? */
137 int ksnd_min_bulk; /* smallest "large" message */
138 int ksnd_buffer_size; /* socket buffer size */
139 int ksnd_nagle; /* enable NAGLE? */
140 int ksnd_irq_affinity; /* enable IRQ affinity? */
141 int ksnd_keepalive_idle; /* # idle secs before 1st probe */
142 int ksnd_keepalive_count; /* # probes */
143 int ksnd_keepalive_intvl; /* time between probes */
145 unsigned int ksnd_zc_min_frag; /* minimum zero copy frag size */
147 cfs_sysctl_table_header_t *ksnd_sysctl; /* sysctl interface */
152 int ksnd_init; /* initialisation state */
153 __u64 ksnd_incarnation; /* my epoch */
155 rwlock_t ksnd_global_lock; /* stabilize peer/conn ops */
156 struct list_head *ksnd_peers; /* hash table of all my known peers */
157 int ksnd_peer_hash_size; /* size of ksnd_peers */
159 int ksnd_nthreads; /* # live threads */
160 int ksnd_shuttingdown; /* tell threads to exit */
161 int ksnd_nschedulers; /* # schedulers */
162 ksock_sched_t *ksnd_schedulers; /* their state */
164 atomic_t ksnd_npeers; /* total # peers extant */
165 atomic_t ksnd_nclosing_conns; /* # closed conns extant */
167 kpr_router_t ksnd_router; /* THE router */
169 ksock_fmb_pool_t ksnd_small_fmp; /* small message forwarding buffers */
170 ksock_fmb_pool_t ksnd_large_fmp; /* large message forwarding buffers */
172 atomic_t ksnd_nactive_ltxs; /* #active ltxs */
174 struct list_head ksnd_deathrow_conns; /* conns to close: reaper_lock*/
175 struct list_head ksnd_zombie_conns; /* conns to free: reaper_lock */
176 struct list_head ksnd_enomem_conns; /* conns to retry: reaper_lock*/
177 cfs_waitq_t ksnd_reaper_waitq; /* reaper sleeps here */
178 cfs_time_t ksnd_reaper_waketime; /* when reaper will wake */
179 spinlock_t ksnd_reaper_lock; /* serialise */
181 int ksnd_enomem_tx; /* test ENOMEM sender */
182 int ksnd_stall_tx; /* test sluggish sender */
183 int ksnd_stall_rx; /* test sluggish receiver */
185 struct list_head ksnd_autoconnectd_routes; /* routes waiting to be connected */
186 cfs_waitq_t ksnd_autoconnectd_waitq; /* autoconnectds sleep here */
187 spinlock_t ksnd_autoconnectd_lock; /* serialise */
189 ksock_irqinfo_t ksnd_irqinfo[NR_IRQS];/* irq->scheduler lookup */
191 int ksnd_ninterfaces;
192 ksock_interface_t ksnd_interfaces[SOCKNAL_MAX_INTERFACES]; /* published interfaces */
195 #define SOCKNAL_INIT_NOTHING 0
196 #define SOCKNAL_INIT_DATA 1
197 #define SOCKNAL_INIT_LIB 2
198 #define SOCKNAL_INIT_ALL 3
200 /* A packet just assembled for transmission is represented by 1 or more
201 * struct iovec fragments (the first frag contains the portals header),
202 * followed by 0 or more ptl_kiov_t fragments.
204 * On the receive side, initially 1 struct iovec fragment is posted for
205 * receive (the header). Once the header has been received, the payload is
206 * received into either struct iovec or ptl_kiov_t fragments, depending on
207 * what the header matched or whether the message needs forwarding. */
209 struct ksock_conn; /* forward ref */
210 struct ksock_peer; /* forward ref */
211 struct ksock_route; /* forward ref */
213 typedef struct /* transmit packet */
215 struct list_head tx_list; /* queue on conn for transmission etc */
216 char tx_isfwd; /* forwarding / sourced here */
217 int tx_nob; /* # packet bytes */
218 int tx_resid; /* residual bytes */
219 int tx_niov; /* # packet iovec frags */
220 struct iovec *tx_iov; /* packet iovec frags */
221 int tx_nkiov; /* # packet page frags */
222 ptl_kiov_t *tx_kiov; /* packet page frags */
223 struct ksock_conn *tx_conn; /* owning conn */
224 ptl_hdr_t *tx_hdr; /* packet header (for debug only) */
226 zccd_t tx_zccd; /* zero copy callback descriptor */
230 typedef struct /* forwarded packet */
232 ksock_tx_t ftx_tx; /* send info */
233 struct iovec ftx_iov; /* hdr iovec */
236 #define KSOCK_ZCCD_2_TX(ptr) list_entry (ptr, ksock_tx_t, tx_zccd)
237 /* network zero copy callback descriptor embedded in ksock_tx_t */
239 typedef struct /* locally transmitted packet */
241 ksock_tx_t ltx_tx; /* send info */
242 void *ltx_private; /* lib_finalize() callback arg */
243 void *ltx_cookie; /* lib_finalize() callback arg */
244 ptl_hdr_t ltx_hdr; /* buffer for packet header */
245 int ltx_desc_size; /* bytes allocated for this desc */
246 struct iovec ltx_iov[1]; /* iov for hdr + payload */
247 ptl_kiov_t ltx_kiov[0]; /* kiov for payload */
250 #define KSOCK_TX_2_KPR_FWD_DESC(ptr) list_entry ((kprfd_scratch_t *)ptr, kpr_fwd_desc_t, kprfd_scratch)
251 /* forwarded packets (router->socknal) embedded in kpr_fwd_desc_t::kprfd_scratch */
253 #define KSOCK_TX_2_KSOCK_LTX(ptr) list_entry (ptr, ksock_ltx_t, ltx_tx)
254 /* local packets (lib->socknal) embedded in ksock_ltx_t::ltx_tx */
256 /* NB list_entry() is used here as convenient macro for calculating a
257 * pointer to a struct from the address of a member. */
259 typedef struct /* Kernel portals Socket Forwarding message buffer */
260 { /* (socknal->router) */
261 struct list_head fmb_list; /* queue idle */
262 kpr_fwd_desc_t fmb_fwd; /* router's descriptor */
263 ksock_fmb_pool_t *fmb_pool; /* owning pool */
264 struct ksock_peer *fmb_peer; /* peer received from */
265 ptl_hdr_t fmb_hdr; /* message header */
266 ptl_kiov_t fmb_kiov[0]; /* payload frags */
269 /* space for the rx frag descriptors; we either read a single contiguous
270 * header, or up to PTL_MD_MAX_IOV frags of payload of either type. */
272 struct iovec iov[PTL_MD_MAX_IOV];
273 ptl_kiov_t kiov[PTL_MD_MAX_IOV];
274 } ksock_rxiovspace_t;
276 #define SOCKNAL_RX_HEADER 1 /* reading header */
277 #define SOCKNAL_RX_BODY 2 /* reading body (to deliver here) */
278 #define SOCKNAL_RX_BODY_FWD 3 /* reading body (to forward) */
279 #define SOCKNAL_RX_SLOP 4 /* skipping body */
280 #define SOCKNAL_RX_GET_FMB 5 /* scheduled for forwarding */
281 #define SOCKNAL_RX_FMB_SLEEP 6 /* blocked waiting for a fwd desc */
283 typedef struct ksock_conn
285 struct ksock_peer *ksnc_peer; /* owning peer */
286 struct ksock_route *ksnc_route; /* owning route */
287 struct list_head ksnc_list; /* stash on peer's conn list */
288 struct socket *ksnc_sock; /* actual socket */
289 void *ksnc_saved_data_ready; /* socket's original data_ready() callback */
290 void *ksnc_saved_write_space; /* socket's original write_space() callback */
291 atomic_t ksnc_refcount; /* # users */
292 ksock_sched_t *ksnc_scheduler; /* who schedules this connection */
293 __u32 ksnc_myipaddr; /* my IP */
294 __u32 ksnc_ipaddr; /* peer's IP */
295 int ksnc_port; /* peer's port */
296 int ksnc_closing; /* being shut down */
297 int ksnc_type; /* type of connection */
298 __u64 ksnc_incarnation; /* peer's incarnation */
301 struct list_head ksnc_rx_list; /* where I enq waiting input or a forwarding descriptor */
302 cfs_time_t ksnc_rx_deadline; /* when (in jiffies) receive times out */
303 int ksnc_rx_started; /* started receiving a message */
304 int ksnc_rx_ready; /* data ready to read */
305 int ksnc_rx_scheduled; /* being progressed */
306 int ksnc_rx_state; /* what is being read */
307 int ksnc_rx_nob_left; /* # bytes to next hdr/body */
308 int ksnc_rx_nob_wanted; /* bytes actually wanted */
309 int ksnc_rx_niov; /* # iovec frags */
310 struct iovec *ksnc_rx_iov; /* the iovec frags */
311 int ksnc_rx_nkiov; /* # page frags */
312 ptl_kiov_t *ksnc_rx_kiov; /* the page frags */
313 ksock_rxiovspace_t ksnc_rx_iov_space; /* space for frag descriptors */
314 void *ksnc_cookie; /* rx lib_finalize passthru arg */
315 ptl_hdr_t ksnc_hdr; /* where I read headers into */
318 struct list_head ksnc_tx_list; /* where I enq waiting for output space */
319 struct list_head ksnc_tx_queue; /* packets waiting to be sent */
320 cfs_time_t ksnc_tx_deadline; /* when (in jiffies) tx times out */
321 int ksnc_tx_bufnob; /* send buffer marker */
322 atomic_t ksnc_tx_nob; /* # bytes queued */
323 int ksnc_tx_ready; /* write space */
324 int ksnc_tx_scheduled; /* being progressed */
326 #if !SOCKNAL_SINGLE_FRAG_RX
327 struct iovec ksnc_rx_scratch_iov[PTL_MD_MAX_IOV];
329 #if !SOCKNAL_SINGLE_FRAG_TX
330 struct iovec ksnc_tx_scratch_iov[PTL_MD_MAX_IOV];
334 #define KSNR_TYPED_ROUTES ((1 << SOCKNAL_CONN_CONTROL) | \
335 (1 << SOCKNAL_CONN_BULK_IN) | \
336 (1 << SOCKNAL_CONN_BULK_OUT))
338 typedef struct ksock_route
340 struct list_head ksnr_list; /* chain on peer route list */
341 struct list_head ksnr_connect_list; /* chain on autoconnect list */
342 struct ksock_peer *ksnr_peer; /* owning peer */
343 atomic_t ksnr_refcount; /* # users */
344 cfs_time_t ksnr_timeout; /* when (in jiffies) reconnection can happen next */
345 cfs_duration_t ksnr_retry_interval; /* how long between retries */
346 __u32 ksnr_myipaddr; /* my IP */
347 __u32 ksnr_ipaddr; /* IP address to connect to */
348 int ksnr_port; /* port to connect to */
349 unsigned int ksnr_connecting:1; /* autoconnect in progress */
350 unsigned int ksnr_connected:4; /* connections established by type */
351 unsigned int ksnr_deleted:1; /* been removed from peer? */
352 unsigned int ksnr_share_count; /* created explicitly? */
353 int ksnr_conn_count; /* # conns established by this route */
356 typedef struct ksock_peer
358 struct list_head ksnp_list; /* stash on global peer list */
359 ptl_nid_t ksnp_nid; /* who's on the other end(s) */
360 atomic_t ksnp_refcount; /* # users */
361 int ksnp_sharecount; /* lconf usage counter */
362 int ksnp_closing; /* being closed */
363 int ksnp_error; /* errno on closing last conn */
364 struct list_head ksnp_conns; /* all active connections */
365 struct list_head ksnp_routes; /* routes */
366 struct list_head ksnp_tx_queue; /* waiting packets */
367 cfs_time_t ksnp_last_alive; /* when (in jiffies) I was last alive */
368 int ksnp_n_passive_ips; /* # of... */
369 __u32 ksnp_passive_ips[SOCKNAL_MAX_INTERFACES]; /* preferred local interfaces */
373 extern lib_nal_t ksocknal_lib;
374 extern ksock_nal_data_t ksocknal_data;
375 extern ksock_tunables_t ksocknal_tunables;
377 static inline struct list_head *
378 ksocknal_nid2peerlist (ptl_nid_t nid)
380 unsigned int hash = ((unsigned int)nid) % ksocknal_data.ksnd_peer_hash_size;
382 return (&ksocknal_data.ksnd_peers [hash]);
386 ksocknal_getconnsock (ksock_conn_t *conn)
390 read_lock (&ksocknal_data.ksnd_global_lock);
391 if (!conn->ksnc_closing) {
393 cfs_get_file (KSN_CONN2FILE(conn));
395 read_unlock (&ksocknal_data.ksnd_global_lock);
401 ksocknal_putconnsock (ksock_conn_t *conn)
403 cfs_put_file (KSN_CONN2FILE(conn));
406 extern void ksocknal_put_route (ksock_route_t *route);
407 extern void ksocknal_put_peer (ksock_peer_t *peer);
408 extern ksock_peer_t *ksocknal_find_peer_locked (ptl_nid_t nid);
409 extern ksock_peer_t *ksocknal_get_peer (ptl_nid_t nid);
410 extern int ksocknal_del_route (ptl_nid_t nid, __u32 ipaddr,
411 int single, int keep_conn);
412 extern int ksocknal_create_conn (ksock_route_t *route,
413 struct socket *sock, int type);
414 extern void ksocknal_close_conn_locked (ksock_conn_t *conn, int why);
415 extern void ksocknal_terminate_conn (ksock_conn_t *conn);
416 extern void ksocknal_destroy_conn (ksock_conn_t *conn);
417 extern void ksocknal_put_conn (ksock_conn_t *conn);
418 extern int ksocknal_close_stale_conns_locked (ksock_peer_t *peer, __u64 incarnation);
419 extern int ksocknal_close_conn_and_siblings (ksock_conn_t *conn, int why);
420 extern int ksocknal_close_matching_conns (ptl_nid_t nid, __u32 ipaddr);
422 extern void ksocknal_queue_tx_locked (ksock_tx_t *tx, ksock_conn_t *conn);
423 extern void ksocknal_tx_done (ksock_tx_t *tx, int asynch);
424 extern void ksocknal_fwd_packet (void *arg, kpr_fwd_desc_t *fwd);
425 extern void ksocknal_fmb_callback (void *arg, int error);
426 extern void ksocknal_notify (void *arg, ptl_nid_t gw_nid, int alive);
427 extern int ksocknal_thread_start (int (*fn)(void *arg), void *arg);
428 extern int ksocknal_new_packet (ksock_conn_t *conn, int skip);
429 extern int ksocknal_scheduler (void *arg);
430 extern int ksocknal_autoconnectd (void *arg);
431 extern int ksocknal_reaper (void *arg);
432 extern int ksocknal_setup_sock (struct socket *sock);
433 extern int ksocknal_send_hello (ksock_conn_t *conn, __u32 *ipaddrs, int nipaddrs);
434 extern int ksocknal_recv_hello (ksock_conn_t *conn,
435 ptl_nid_t *nid, __u64 *incarnation, __u32 *ipaddrs);
437 extern void ksocknal_lib_save_callback(struct socket *sock, ksock_conn_t *conn);
438 extern void ksocknal_lib_set_callback(struct socket *sock, ksock_conn_t *conn);
439 extern void ksocknal_lib_act_callback(struct socket *sock, ksock_conn_t *conn);
440 extern void ksocknal_lib_reset_callback(struct socket *sock, ksock_conn_t *conn);
441 extern void ksocknal_lib_push_conn (ksock_conn_t *conn);
442 extern void ksocknal_lib_bind_irq (unsigned int irq);
443 extern int ksocknal_lib_get_conn_addrs (ksock_conn_t *conn);
444 extern unsigned int ksocknal_lib_sock_irq (struct socket *sock);
445 extern int ksocknal_lib_setup_sock (struct socket *so);
446 extern int ksocknal_lib_send_iov (ksock_conn_t *conn, ksock_tx_t *tx);
447 extern int ksocknal_lib_send_kiov (ksock_conn_t *conn, ksock_tx_t *tx);
448 extern void ksocknal_lib_eager_ack (ksock_conn_t *conn);
449 extern int ksocknal_lib_recv_iov (ksock_conn_t *conn);
450 extern int ksocknal_lib_recv_kiov (ksock_conn_t *conn);
451 extern int ksocknal_lib_sock_write (struct socket *sock,
452 void *buffer, int nob);
453 extern int ksocknal_lib_sock_read (struct socket *sock,
454 void *buffer, int nob);
455 extern int ksocknal_lib_get_conn_tunables (ksock_conn_t *conn, int *txmem,
456 int *rxmem, int *nagle);
457 extern int ksocknal_lib_connect_sock(struct socket **sockp, int *may_retry,
458 ksock_route_t *route, int local_port);