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>
48 #include <linux/init.h>
50 #include <linux/file.h>
51 #include <linux/stat.h>
52 #include <linux/list.h>
53 #include <linux/kmod.h>
54 #include <linux/sysctl.h>
55 #include <asm/uaccess.h>
56 #include <asm/segment.h>
57 #include <asm/div64.h>
59 #define DEBUG_SUBSYSTEM S_SOCKNAL
61 #include <linux/kp30.h>
62 #include <portals/p30.h>
63 #include <portals/lib-p30.h>
64 #include <portals/socknal.h>
67 # define SOCKNAL_N_SCHED num_online_cpus() /* # socknal schedulers */
69 # define SOCKNAL_N_SCHED 1 /* # socknal schedulers */
71 #define SOCKNAL_N_AUTOCONNECTD 4 /* # socknal autoconnect daemons */
73 #define SOCKNAL_MIN_RECONNECT_INTERVAL HZ /* first failed connection retry... */
74 #define SOCKNAL_MAX_RECONNECT_INTERVAL (60*HZ) /* ...exponentially increasing to this */
76 /* default vals for runtime tunables */
77 #define SOCKNAL_IO_TIMEOUT 50 /* default comms timeout (seconds) */
78 #define SOCKNAL_EAGER_ACK 0 /* default eager ack (boolean) */
79 #define SOCKNAL_TYPED_CONNS 1 /* unidirectional large, bidirectional small? */
80 #define SOCKNAL_ZC_MIN_FRAG (2<<10) /* default smallest zerocopy fragment */
81 #define SOCKNAL_MIN_BULK (1<<10) /* smallest "large" message */
82 #define SOCKNAL_USE_KEEPALIVES 0 /* use tcp/ip keepalive? */
84 #define SOCKNAL_PEER_HASH_SIZE 101 /* # peer lists */
86 #define SOCKNAL_SMALL_FWD_NMSGS 128 /* # small messages I can be forwarding at any time */
87 #define SOCKNAL_LARGE_FWD_NMSGS 64 /* # large messages I can be forwarding at any time */
89 #define SOCKNAL_SMALL_FWD_PAGES 1 /* # pages in a small message fwd buffer */
91 #define SOCKNAL_LARGE_FWD_PAGES (PAGE_ALIGN (sizeof (ptl_hdr_t) + PTL_MTU) >> PAGE_SHIFT)
92 /* # pages in a large message fwd buffer */
94 #define SOCKNAL_RESCHED 100 /* # scheduler loops before reschedule */
95 #define SOCKNAL_ENOMEM_RETRY 1 /* jiffies between retries */
97 #define SOCKNAL_TX_LOW_WATER(sk) (((sk)->sk_sndbuf*8)/10)
99 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,72))
100 # define sk_data_ready data_ready
101 # define sk_write_space write_space
102 # define sk_user_data user_data
103 # define sk_prot prot
104 # define sk_sndbuf sndbuf
105 # define sk_socket socket
108 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0))
109 # define sk_wmem_queued wmem_queued
112 typedef struct /* pool of forwarding buffers */
114 spinlock_t fmp_lock; /* serialise */
115 struct list_head fmp_idle_fmbs; /* free buffers */
116 struct list_head fmp_blocked_conns; /* connections waiting for a buffer */
117 int fmp_nactive_fmbs; /* # buffers in use */
121 typedef struct /* per scheduler state */
123 spinlock_t kss_lock; /* serialise */
124 struct list_head kss_rx_conns; /* conn waiting to be read */
125 struct list_head kss_tx_conns; /* conn waiting to be written */
127 struct list_head kss_zctxdone_list; /* completed ZC transmits */
129 wait_queue_head_t kss_waitq; /* where scheduler sleeps */
130 int kss_nconns; /* # connections assigned to this scheduler */
134 int ksni_valid:1; /* been set yet? */
135 int ksni_bound:1; /* bound to a cpu yet? */
136 int ksni_sched:6; /* which scheduler (assumes < 64) */
140 int ksnd_init; /* initialisation state */
141 int ksnd_io_timeout; /* "stuck" socket timeout (seconds) */
142 int ksnd_eager_ack; /* make TCP ack eagerly? */
143 int ksnd_typed_conns; /* drive sockets by type? */
144 int ksnd_min_bulk; /* smallest "large" message */
146 unsigned int ksnd_zc_min_frag; /* minimum zero copy frag size */
148 struct ctl_table_header *ksnd_sysctl; /* sysctl interface */
149 __u64 ksnd_incarnation; /* my epoch */
151 rwlock_t ksnd_global_lock; /* stabilize peer/conn ops */
152 struct list_head *ksnd_peers; /* hash table of all my known peers */
153 int ksnd_peer_hash_size; /* size of ksnd_peers */
155 nal_cb_t *ksnd_nal_cb;
156 spinlock_t ksnd_nal_cb_lock; /* lib cli/sti lock */
158 atomic_t ksnd_nthreads; /* # live threads */
159 int ksnd_shuttingdown; /* tell threads to exit */
160 ksock_sched_t *ksnd_schedulers; /* scheduler state */
162 atomic_t ksnd_npeers; /* total # peers extant */
163 atomic_t ksnd_nclosing_conns; /* # closed conns extant */
165 kpr_router_t ksnd_router; /* THE router */
167 ksock_fmb_pool_t ksnd_small_fmp; /* small message forwarding buffers */
168 ksock_fmb_pool_t ksnd_large_fmp; /* large message forwarding buffers */
170 atomic_t ksnd_nactive_ltxs; /* #active ltxs */
172 struct list_head ksnd_deathrow_conns; /* conns to be closed */
173 struct list_head ksnd_zombie_conns; /* conns to be freed */
174 struct list_head ksnd_enomem_conns; /* conns to be retried */
175 wait_queue_head_t ksnd_reaper_waitq; /* reaper sleeps here */
176 unsigned long ksnd_reaper_waketime; /* when reaper will wake */
177 spinlock_t ksnd_reaper_lock; /* serialise */
179 int ksnd_enomem_tx; /* test ENOMEM sender */
180 int ksnd_stall_tx; /* test sluggish sender */
181 int ksnd_stall_rx; /* test sluggish receiver */
183 struct list_head ksnd_autoconnectd_routes; /* routes waiting to be connected */
184 wait_queue_head_t ksnd_autoconnectd_waitq; /* autoconnectds sleep here */
185 spinlock_t ksnd_autoconnectd_lock; /* serialise */
187 ksock_irqinfo_t ksnd_irqinfo[NR_IRQS];/* irq->scheduler lookup */
190 #define SOCKNAL_INIT_NOTHING 0
191 #define SOCKNAL_INIT_DATA 1
192 #define SOCKNAL_INIT_PTL 2
193 #define SOCKNAL_INIT_ALL 3
195 /* A packet just assembled for transmission is represented by 1 or more
196 * struct iovec fragments and 0 or more ptl_kiov_t fragments. Forwarded
197 * messages, or messages from an MD with PTL_MD_KIOV _not_ set have 0
198 * ptl_kiov_t fragments. Messages from an MD with PTL_MD_KIOV set, have 1
199 * struct iovec fragment (the header) and up to PTL_MD_MAX_IOV ptl_kiov_t
202 * On the receive side, initially 1 struct iovec fragment is posted for
203 * receive (the header). Once the header has been received, if the message
204 * requires forwarding or will be received into mapped memory, up to
205 * PTL_MD_MAX_IOV struct iovec fragments describe the target memory.
206 * Otherwise up to PTL_MD_MAX_IOV ptl_kiov_t fragments are used.
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 #define KSOCK_ZCCD_2_TX(ptr) list_entry (ptr, ksock_tx_t, tx_zccd)
231 /* network zero copy callback descriptor embedded in ksock_tx_t */
233 typedef struct /* locally transmitted packet */
235 ksock_tx_t ltx_tx; /* send info */
236 void *ltx_private; /* lib_finalize() callback arg */
237 void *ltx_cookie; /* lib_finalize() callback arg */
238 ptl_hdr_t ltx_hdr; /* buffer for packet header */
239 int ltx_desc_size; /* bytes allocated for this desc */
240 struct iovec ltx_iov[1]; /* iov for hdr + payload */
241 ptl_kiov_t ltx_kiov[0]; /* kiov for payload */
244 #define KSOCK_TX_2_KPR_FWD_DESC(ptr) list_entry ((kprfd_scratch_t *)ptr, kpr_fwd_desc_t, kprfd_scratch)
245 /* forwarded packets (router->socknal) embedded in kpr_fwd_desc_t::kprfd_scratch */
247 #define KSOCK_TX_2_KSOCK_LTX(ptr) list_entry (ptr, ksock_ltx_t, ltx_tx)
248 /* local packets (lib->socknal) embedded in ksock_ltx_t::ltx_tx */
250 /* NB list_entry() is used here as convenient macro for calculating a
251 * pointer to a struct from the address of a member. */
253 typedef struct /* Kernel portals Socket Forwarding message buffer */
254 { /* (socknal->router) */
255 struct list_head fmb_list; /* queue idle */
256 kpr_fwd_desc_t fmb_fwd; /* router's descriptor */
257 int fmb_npages; /* # pages allocated */
258 ksock_fmb_pool_t *fmb_pool; /* owning pool */
259 struct ksock_peer *fmb_peer; /* peer received from */
260 struct page *fmb_pages[SOCKNAL_LARGE_FWD_PAGES];
261 struct iovec fmb_iov[SOCKNAL_LARGE_FWD_PAGES];
264 /* space for the rx frag descriptors; we either read a single contiguous
265 * header, or PTL_MD_MAX_IOV frags of payload of either type. */
267 struct iovec iov[PTL_MD_MAX_IOV];
268 ptl_kiov_t kiov[PTL_MD_MAX_IOV];
269 } ksock_rxiovspace_t;
271 #define SOCKNAL_RX_HEADER 1 /* reading header */
272 #define SOCKNAL_RX_BODY 2 /* reading body (to deliver here) */
273 #define SOCKNAL_RX_BODY_FWD 3 /* reading body (to forward) */
274 #define SOCKNAL_RX_SLOP 4 /* skipping body */
275 #define SOCKNAL_RX_GET_FMB 5 /* scheduled for forwarding */
276 #define SOCKNAL_RX_FMB_SLEEP 6 /* blocked waiting for a fwd desc */
278 typedef struct ksock_conn
280 struct ksock_peer *ksnc_peer; /* owning peer */
281 struct ksock_route *ksnc_route; /* owning route */
282 struct list_head ksnc_list; /* stash on peer's conn list */
283 struct socket *ksnc_sock; /* actual socket */
284 void *ksnc_saved_data_ready; /* socket's original data_ready() callback */
285 void *ksnc_saved_write_space; /* socket's original write_space() callback */
286 atomic_t ksnc_refcount; /* # users */
287 ksock_sched_t *ksnc_scheduler; /* who schedules this connection */
288 __u32 ksnc_ipaddr; /* peer's IP */
289 int ksnc_port; /* peer's port */
290 int ksnc_closing; /* being shut down */
291 int ksnc_type; /* type of connection */
292 __u64 ksnc_incarnation; /* peer's incarnation */
295 struct list_head ksnc_rx_list; /* where I enq waiting input or a forwarding descriptor */
296 unsigned long ksnc_rx_deadline; /* when (in jiffies) receive times out */
297 int ksnc_rx_started; /* started receiving a message */
298 int ksnc_rx_ready; /* data ready to read */
299 int ksnc_rx_scheduled; /* being progressed */
300 int ksnc_rx_state; /* what is being read */
301 int ksnc_rx_nob_left; /* # bytes to next hdr/body */
302 int ksnc_rx_nob_wanted; /* bytes actually wanted */
303 int ksnc_rx_niov; /* # iovec frags */
304 struct iovec *ksnc_rx_iov; /* the iovec frags */
305 int ksnc_rx_nkiov; /* # page frags */
306 ptl_kiov_t *ksnc_rx_kiov; /* the page frags */
307 ksock_rxiovspace_t ksnc_rx_iov_space; /* space for frag descriptors */
308 void *ksnc_cookie; /* rx lib_finalize passthru arg */
309 ptl_hdr_t ksnc_hdr; /* where I read headers into */
312 struct list_head ksnc_tx_list; /* where I enq waiting for output space */
313 struct list_head ksnc_tx_queue; /* packets waiting to be sent */
314 unsigned long ksnc_tx_deadline; /* when (in jiffies) tx times out */
315 atomic_t ksnc_tx_nob; /* # bytes queued */
316 int ksnc_tx_ready; /* write space */
317 int ksnc_tx_scheduled; /* being progressed */
320 #define KSNR_TYPED_ROUTES ((1 << SOCKNAL_CONN_CONTROL) | \
321 (1 << SOCKNAL_CONN_BULK_IN) | \
322 (1 << SOCKNAL_CONN_BULK_OUT))
324 typedef struct ksock_route
326 struct list_head ksnr_list; /* chain on peer route list */
327 struct list_head ksnr_connect_list; /* chain on autoconnect list */
328 struct ksock_peer *ksnr_peer; /* owning peer */
329 atomic_t ksnr_refcount; /* # users */
330 int ksnr_sharecount; /* lconf usage counter */
331 unsigned long ksnr_timeout; /* when (in jiffies) reconnection can happen next */
332 unsigned int ksnr_retry_interval; /* how long between retries */
333 __u32 ksnr_ipaddr; /* an IP address for this peer */
334 int ksnr_port; /* port to connect to */
335 int ksnr_buffer_size; /* size of socket buffers */
336 unsigned int ksnr_irq_affinity:1; /* set affinity? */
337 unsigned int ksnr_nonagel:1; /* disable nagle? */
338 unsigned int ksnr_eager:1; /* connect eagery? */
339 unsigned int ksnr_connecting:4; /* autoconnects in progress by type */
340 unsigned int ksnr_connected:4; /* connections established by type */
341 unsigned int ksnr_deleted:1; /* been removed from peer? */
342 int ksnr_conn_count; /* # conns established by this route */
345 typedef struct ksock_peer
347 struct list_head ksnp_list; /* stash on global peer list */
348 ptl_nid_t ksnp_nid; /* who's on the other end(s) */
349 atomic_t ksnp_refcount; /* # users */
350 int ksnp_closing; /* being closed */
351 int ksnp_error; /* errno on closing last conn */
352 struct list_head ksnp_conns; /* all active connections */
353 struct list_head ksnp_routes; /* routes */
354 struct list_head ksnp_tx_queue; /* waiting packets */
355 unsigned long ksnp_last_alive; /* when (in jiffies) I was last alive */
359 extern nal_cb_t ksocknal_lib;
360 extern ksock_nal_data_t ksocknal_data;
362 static inline struct list_head *
363 ksocknal_nid2peerlist (ptl_nid_t nid)
365 unsigned int hash = ((unsigned int)nid) % ksocknal_data.ksnd_peer_hash_size;
367 return (&ksocknal_data.ksnd_peers [hash]);
371 ksocknal_getconnsock (ksock_conn_t *conn)
375 read_lock (&ksocknal_data.ksnd_global_lock);
376 if (!conn->ksnc_closing) {
378 get_file (conn->ksnc_sock->file);
380 read_unlock (&ksocknal_data.ksnd_global_lock);
386 ksocknal_putconnsock (ksock_conn_t *conn)
388 fput (conn->ksnc_sock->file);
391 extern void ksocknal_put_route (ksock_route_t *route);
392 extern void ksocknal_put_peer (ksock_peer_t *peer);
393 extern ksock_peer_t *ksocknal_find_peer_locked (ptl_nid_t nid);
394 extern ksock_peer_t *ksocknal_get_peer (ptl_nid_t nid);
395 extern int ksocknal_del_route (ptl_nid_t nid, __u32 ipaddr,
396 int single, int keep_conn);
397 extern int ksocknal_create_conn (ksock_route_t *route,
398 struct socket *sock, int bind_irq, int type);
399 extern void ksocknal_close_conn_locked (ksock_conn_t *conn, int why);
400 extern void ksocknal_terminate_conn (ksock_conn_t *conn);
401 extern void ksocknal_destroy_conn (ksock_conn_t *conn);
402 extern void ksocknal_put_conn (ksock_conn_t *conn);
403 extern int ksocknal_close_stale_conns_locked (ksock_peer_t *peer, __u64 incarnation);
404 extern int ksocknal_close_conn_and_siblings (ksock_conn_t *conn, int why);
405 extern int ksocknal_close_matching_conns (ptl_nid_t nid, __u32 ipaddr);
407 extern void ksocknal_queue_tx_locked (ksock_tx_t *tx, ksock_conn_t *conn);
408 extern void ksocknal_tx_done (ksock_tx_t *tx, int asynch);
409 extern void ksocknal_fwd_packet (void *arg, kpr_fwd_desc_t *fwd);
410 extern void ksocknal_fmb_callback (void *arg, int error);
411 extern void ksocknal_notify (void *arg, ptl_nid_t gw_nid, int alive);
412 extern int ksocknal_thread_start (int (*fn)(void *arg), void *arg);
413 extern int ksocknal_new_packet (ksock_conn_t *conn, int skip);
414 extern int ksocknal_scheduler (void *arg);
415 extern void ksocknal_data_ready(struct sock *sk, int n);
416 extern void ksocknal_write_space(struct sock *sk);
417 extern int ksocknal_autoconnectd (void *arg);
418 extern int ksocknal_reaper (void *arg);
419 extern int ksocknal_setup_sock (struct socket *sock);
420 extern int ksocknal_hello (struct socket *sock,
421 ptl_nid_t *nid, int *type, __u64 *incarnation);