1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2002 Cluster File Systems, Inc.
6 * This file is part of Lustre, http://www.lustre.org.
8 * Lustre is free software; you can redistribute it and/or
9 * modify it under the terms of version 2 of the GNU General Public
10 * License as published by the Free Software Foundation.
12 * Lustre is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with Lustre; if not, write to the Free Software
19 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
26 #include <linux/tqueue.h>
27 #include <linux/kp30.h>
28 // #include <linux/obd.h>
29 #include <portals/p30.h>
30 #include <linux/lustre_idl.h>
31 #include <linux/lustre_ha.h>
32 #include <linux/lustre_import.h>
34 /* The following constants determine how much memory is devoted to
35 * buffering in the lustre services.
37 * ?_NEVENTS # event queue entries
39 * ?_NBUFS # request buffers
40 * ?_BUFSIZE # bytes in a single request buffer
41 * total memory = ?_NBUFS * ?_BUFSIZE
43 * ?_MAXREQSIZE # maximum request service will receive
44 * larger messages will get dropped.
45 * request buffers are auto-unlinked when less than ?_MAXREQSIZE
49 #define LDLM_NUM_THREADS 4
50 #define LDLM_NEVENTS 1024
52 #define LDLM_BUFSIZE (64 * 1024)
53 #define LDLM_MAXREQSIZE 1024
55 #define MDT_NUM_THREADS 8
56 #define MDS_NEVENTS 1024
58 #define MDS_BUFSIZE (64 * 1024)
59 #define MDS_MAXREQSIZE 1024
61 #define OST_NUM_THREADS 6
62 #define OST_NEVENTS min(num_physpages / 16, 32768UL)
63 #define OST_NBUFS min(OST_NEVENTS / 128, 256UL)
64 #define OST_BUFSIZE ((OST_NEVENTS > 4096UL ? 128 : 64) * 1024)
65 #define OST_MAXREQSIZE (8 * 1024)
67 #define CONN_INVALID 1
69 struct ptlrpc_connection {
70 struct list_head c_link;
71 struct lustre_peer c_peer;
72 __u8 c_local_uuid[37]; /* XXX do we need this? */
73 __u8 c_remote_uuid[37];
76 __u32 c_generation; /* changes upon new connection */
77 __u32 c_epoch; /* changes when peer changes */
78 __u32 c_bootcount; /* peer's boot count */
80 spinlock_t c_lock; /* also protects req->rq_list */
89 __u64 c_last_xid; /* protected by c_lock */
90 __u64 c_last_committed;/* protected by c_lock */
91 struct list_head c_delayed_head;/* delayed until post-recovery */
92 struct list_head c_sending_head;/* protected by c_lock */
93 struct list_head c_dying_head; /* protected by c_lock */
94 struct recovd_data c_recovd_data;
96 struct list_head c_imports;
97 struct list_head c_exports;
98 struct list_head c_sb_chain;
99 __u32 c_flags; /* can we indicate INVALID elsewhere? */
102 struct ptlrpc_client {
103 __u32 cli_request_portal;
104 __u32 cli_reply_portal;
106 __u32 cli_target_devno;
109 // struct semaphore cli_rpc_sem; /* limits outstanding requests */
114 /* state flags of requests */
115 #define PTL_RPC_FL_INTR (1 << 0)
116 #define PTL_RPC_FL_REPLIED (1 << 1) /* reply was received */
117 #define PTL_RPC_FL_SENT (1 << 2)
118 #define PTL_BULK_FL_SENT (1 << 3)
119 #define PTL_BULK_FL_RCVD (1 << 4)
120 #define PTL_RPC_FL_ERR (1 << 5)
121 #define PTL_RPC_FL_TIMEOUT (1 << 6)
122 #define PTL_RPC_FL_RESEND (1 << 7)
123 #define PTL_RPC_FL_RECOVERY (1 << 8) /* retransmission for recovery */
124 #define PTL_RPC_FL_FINISHED (1 << 9)
125 #define PTL_RPC_FL_RETAIN (1 << 10) /* retain for replay after reply */
126 #define PTL_RPC_FL_REPLAY (1 << 11) /* replay upon recovery */
127 #define PTL_RPC_FL_ALLOCREP (1 << 12) /* reply buffer allocated */
129 struct ptlrpc_request {
130 int rq_type; /* one of PTL_RPC_MSG_* */
131 struct list_head rq_list;
132 struct obd_device *rq_obd;
135 atomic_t rq_refcount;
137 int rq_request_portal; /* XXX FIXME bug 625069 */
138 int rq_reply_portal; /* XXX FIXME bug 625069 */
141 struct lustre_msg *rq_reqmsg;
144 struct lustre_msg *rq_repmsg;
150 // void * rq_reply_handle;
151 wait_queue_head_t rq_wait_for_rep;
154 ptl_md_t rq_reply_md;
155 ptl_handle_me_t rq_reply_me_h;
157 /* outgoing req/rep */
160 struct lustre_peer rq_peer; /* XXX see service.c can this be factored away? */
161 struct obd_export *rq_export;
162 struct ptlrpc_connection *rq_connection;
163 struct obd_import *rq_import;
164 struct ptlrpc_service *rq_svc;
166 void (*rq_replay_cb)(struct ptlrpc_request *);
169 #define DEBUG_REQ(level, req, fmt, args...) \
172 "@@@ " fmt " req x"LPD64"/t"LPD64" o%d->%s:%d lens %d/%d fl %x\n" , ## args, \
173 req->rq_xid, req->rq_transno, \
174 req->rq_reqmsg ? req->rq_reqmsg->opc : -1, \
175 req->rq_connection ? (char *)req->rq_connection->c_remote_uuid : "<?>", \
176 (req->rq_import && req->rq_import->imp_client) ? \
177 req->rq_import->imp_client->cli_request_portal : -1, \
178 req->rq_reqlen, req->rq_replen, req->rq_flags); \
181 struct ptlrpc_bulk_page {
182 struct ptlrpc_bulk_desc *bp_desc;
183 struct list_head bp_link;
186 struct page *bp_page;
189 struct dentry *bp_dentry;
190 int (*bp_cb)(struct ptlrpc_bulk_page *);
194 struct ptlrpc_bulk_desc {
196 struct ptlrpc_connection *bd_connection;
197 struct ptlrpc_client *bd_client;
199 struct lustre_handle bd_conn;
200 void (*bd_ptl_ev_hdlr)(struct ptlrpc_bulk_desc *, void *);
201 void *bd_ptl_ev_data;
203 wait_queue_head_t bd_waitq;
204 struct list_head bd_page_list;
206 atomic_t bd_refcount;
207 void *bd_desc_private;
209 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
210 struct work_struct bd_queue;
212 struct tq_struct bd_queue;
216 ptl_handle_md_t bd_md_h;
217 ptl_handle_me_t bd_me_h;
219 atomic_t bd_source_callback_count;
221 struct iovec bd_iov[16]; /* self-sized pre-allocated iov */
224 struct ptlrpc_thread {
225 struct list_head t_link;
228 wait_queue_head_t t_ctl_waitq;
231 struct ptlrpc_request_buffer_desc {
232 struct list_head rqbd_list;
233 struct ptlrpc_service *rqbd_service;
234 ptl_handle_me_t rqbd_me_h;
235 atomic_t rqbd_refcount;
239 struct ptlrpc_service {
243 /* incoming request buffers */
244 /* FIXME: perhaps a list of EQs, if multiple NIs are used? */
246 __u32 srv_max_req_size; /* biggest request to receive */
247 __u32 srv_buf_size; /* # bytes in a request buffer */
248 struct list_head srv_rqbds; /* all the request buffer descriptors */
249 __u32 srv_nrqbds; /* # request buffers */
250 atomic_t srv_nrqbds_receiving; /* # request buffers posted for input */
252 __u32 srv_req_portal;
253 __u32 srv_rep_portal;
258 ptl_handle_eq_t srv_eq_h;
260 struct lustre_peer srv_self;
262 wait_queue_head_t srv_waitq; /* all threads sleep on this */
265 struct list_head srv_threads;
266 int (*srv_handler)(struct ptlrpc_request *req);
267 char *srv_name; /* only statically allocated strings here; we don't clean them */
270 static inline void ptlrpc_hdl2req(struct ptlrpc_request *req, struct lustre_handle *h)
272 req->rq_reqmsg->addr = h->addr;
273 req->rq_reqmsg->cookie = h->cookie;
276 typedef void (*bulk_callback_t)(struct ptlrpc_bulk_desc *, void *);
278 typedef int (*svc_handler_t)(struct ptlrpc_request *req);
280 /* rpc/connection.c */
281 void ptlrpc_readdress_connection(struct ptlrpc_connection *conn, obd_uuid_t uuid);
282 struct ptlrpc_connection *ptlrpc_get_connection(struct lustre_peer *peer,
284 int ptlrpc_put_connection(struct ptlrpc_connection *c);
285 struct ptlrpc_connection *ptlrpc_connection_addref(struct ptlrpc_connection *);
286 void ptlrpc_init_connection(void);
287 void ptlrpc_cleanup_connection(void);
290 int ptlrpc_check_bulk_sent(struct ptlrpc_bulk_desc *bulk);
291 int ptlrpc_check_bulk_received(struct ptlrpc_bulk_desc *bulk);
292 int ptlrpc_send_bulk(struct ptlrpc_bulk_desc *);
293 int ptlrpc_register_bulk(struct ptlrpc_bulk_desc *);
294 int ptlrpc_abort_bulk(struct ptlrpc_bulk_desc *bulk);
295 int ptlrpc_reply(struct ptlrpc_service *svc, struct ptlrpc_request *req);
296 int ptlrpc_error(struct ptlrpc_service *svc, struct ptlrpc_request *req);
297 void ptlrpc_resend_req(struct ptlrpc_request *request);
298 int ptl_send_rpc(struct ptlrpc_request *request);
299 void ptlrpc_link_svc_me(struct ptlrpc_request_buffer_desc *rqbd);
302 void ptlrpc_init_client(int req_portal, int rep_portal, char *name,
303 struct ptlrpc_client *);
304 void ptlrpc_cleanup_client(struct obd_import *imp);
305 __u8 *ptlrpc_req_to_uuid(struct ptlrpc_request *req);
306 struct ptlrpc_connection *ptlrpc_uuid_to_connection(obd_uuid_t uuid);
308 int ptlrpc_queue_wait(struct ptlrpc_request *req);
309 void ptlrpc_continue_req(struct ptlrpc_request *req);
310 int ptlrpc_replay_req(struct ptlrpc_request *req);
311 void ptlrpc_restart_req(struct ptlrpc_request *req);
313 struct ptlrpc_request *ptlrpc_prep_req(struct obd_import *imp, int opcode,
314 int count, int *lengths, char **bufs);
315 void ptlrpc_free_req(struct ptlrpc_request *request);
316 void ptlrpc_req_finished(struct ptlrpc_request *request);
317 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk(struct ptlrpc_connection *);
318 void ptlrpc_free_bulk(struct ptlrpc_bulk_desc *bulk);
319 struct ptlrpc_bulk_page *ptlrpc_prep_bulk_page(struct ptlrpc_bulk_desc *desc);
320 void ptlrpc_free_bulk_page(struct ptlrpc_bulk_page *page);
321 int ptlrpc_check_status(struct ptlrpc_request *req, int err);
324 struct ptlrpc_service *
325 ptlrpc_init_svc(__u32 nevents, __u32 nbufs, __u32 bufsize, __u32 max_req_size,
326 int req_portal, int rep_portal,
327 obd_uuid_t uuid, svc_handler_t, char *name);
328 void ptlrpc_stop_all_threads(struct ptlrpc_service *svc);
329 int ptlrpc_start_thread(struct obd_device *dev, struct ptlrpc_service *svc,
331 int ptlrpc_unregister_service(struct ptlrpc_service *service);
333 struct ptlrpc_svc_data {
335 struct ptlrpc_service *svc;
336 struct ptlrpc_thread *thread;
337 struct obd_device *dev;
340 /* rpc/pack_generic.c */
341 int lustre_pack_msg(int count, int *lens, char **bufs, int *len,
342 struct lustre_msg **msg);
343 int lustre_msg_size(int count, int *lengths);
344 int lustre_unpack_msg(struct lustre_msg *m, int len);
345 void *lustre_msg_buf(struct lustre_msg *m, int n);
347 static inline void ptlrpc_bulk_decref(struct ptlrpc_bulk_desc *desc)
349 if (atomic_dec_and_test(&desc->bd_refcount)) {
350 CDEBUG(D_PAGE, "Released last ref on %p, freeing\n", desc);
351 ptlrpc_free_bulk(desc);
353 CDEBUG(D_PAGE, "%p -> %d\n", desc,
354 atomic_read(&desc->bd_refcount));
358 static inline void ptlrpc_bulk_addref(struct ptlrpc_bulk_desc *desc)
360 atomic_inc(&desc->bd_refcount);
361 CDEBUG(D_PAGE, "Set refcount of %p to %d\n", desc,
362 atomic_read(&desc->bd_refcount));