1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2002, 2003 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 #if defined(__linux__)
27 #include <linux/lustre_net.h>
28 #elif defined(__APPLE__)
29 #include <darwin/lustre_net.h>
30 #elif defined(__WINNT__)
31 #include <winnt/lustre_net.h>
33 #error Unsupported operating system.
36 #include <libcfs/kp30.h>
38 #include <lnet/lnet.h>
39 #include <lustre/lustre_idl.h>
40 #include <lustre_ha.h>
41 #include <lustre_import.h>
42 #include <lprocfs_status.h>
44 /* MD flags we _always_ use */
45 #define PTLRPC_MD_OPTIONS 0
47 /* Define maxima for bulk I/O
48 * CAVEAT EMPTOR, with multinet (i.e. routers forwarding between networks)
49 * these limits are system wide and not interface-local. */
50 #define PTLRPC_MAX_BRW_BITS LNET_MTU_BITS
51 #define PTLRPC_MAX_BRW_SIZE (1<<LNET_MTU_BITS)
52 #define PTLRPC_MAX_BRW_PAGES (PTLRPC_MAX_BRW_SIZE >> CFS_PAGE_SHIFT)
54 /* When PAGE_SIZE is a constant, we can check our arithmetic here with cpp! */
56 # if ((PTLRPC_MAX_BRW_PAGES & (PTLRPC_MAX_BRW_PAGES - 1)) != 0)
57 # error "PTLRPC_MAX_BRW_PAGES isn't a power of two"
59 # if (PTLRPC_MAX_BRW_SIZE != (PTLRPC_MAX_BRW_PAGES * CFS_PAGE_SIZE))
60 # error "PTLRPC_MAX_BRW_SIZE isn't PTLRPC_MAX_BRW_PAGES * CFS_PAGE_SIZE"
62 # if (PTLRPC_MAX_BRW_SIZE > LNET_MTU)
63 # error "PTLRPC_MAX_BRW_SIZE too big"
65 # if (PTLRPC_MAX_BRW_PAGES > LNET_MAX_IOV)
66 # error "PTLRPC_MAX_BRW_PAGES too big"
68 #endif /* __KERNEL__ */
70 /* Size over which to OBD_VMALLOC() rather than OBD_ALLOC() service request
72 #define SVC_BUF_VMALLOC_THRESHOLD (2 * CFS_PAGE_SIZE)
74 /* The following constants determine how memory is used to buffer incoming
77 * ?_NBUFS # buffers to allocate when growing the pool
78 * ?_BUFSIZE # bytes in a single request buffer
79 * ?_MAXREQSIZE # maximum request service will receive
81 * When fewer than ?_NBUFS/2 buffers are posted for receive, another chunk
82 * of ?_NBUFS is added to the pool.
84 * Messages larger than ?_MAXREQSIZE are dropped. Request buffers are
85 * considered full when less than ?_MAXREQSIZE is left in them.
88 #define LDLM_THREADS_AUTO_MIN \
89 min((int)(num_online_cpus() * num_online_cpus() * 2), 8)
90 #define LDLM_THREADS_AUTO_MAX (LDLM_THREADS_AUTO_MIN * 16)
91 #define LDLM_BL_THREADS LDLM_THREADS_AUTO_MIN
92 #define LDLM_NBUFS (64 * num_online_cpus())
93 #define LDLM_BUFSIZE (8 * 1024)
94 #define LDLM_MAXREQSIZE (5 * 1024)
95 #define LDLM_MAXREPSIZE (1024)
98 #define MDS_THREADS_MIN 2
99 #define MDS_THREADS_MAX 512
100 #define MDS_THREADS_MIN_READPAGE 2
101 #define MDS_NBUFS (64 * num_online_cpus())
102 #define MDS_BUFSIZE (8 * 1024)
103 /* Assume file name length = FNAME_MAX = 256 (true for ext3).
104 * path name length = PATH_MAX = 4096
105 * LOV MD size max = EA_MAX = 4000
106 * symlink: FNAME_MAX + PATH_MAX <- largest
107 * link: FNAME_MAX + PATH_MAX (mds_rec_link < mds_rec_create)
108 * rename: FNAME_MAX + FNAME_MAX
109 * open: FNAME_MAX + EA_MAX
111 * MDS_MAXREQSIZE ~= 4736 bytes =
112 * lustre_msg + ldlm_request + mds_body + mds_rec_create + FNAME_MAX + PATH_MAX
113 * MDS_MAXREPSIZE ~= 8300 bytes = lustre_msg + llog_header
114 * or, for mds_close() and mds_reint_unlink() on a many-OST filesystem:
115 * = 9210 bytes = lustre_msg + mds_body + 160 * (easize + cookiesize)
117 * Realistic size is about 512 bytes (20 character name + 128 char symlink),
118 * except in the open case where there are a large number of OSTs in a LOV.
120 #define MDS_MAXREQSIZE (5 * 1024)
121 #define MDS_MAXREPSIZE max(9 * 1024, 280 + LOV_MAX_STRIPE_COUNT * 56)
123 #define MGS_THREADS_AUTO_MIN 2
124 #define MGS_THREADS_AUTO_MAX 32
125 #define MGS_NBUFS (64 * num_online_cpus())
126 #define MGS_BUFSIZE (8 * 1024)
127 #define MGS_MAXREQSIZE (8 * 1024)
128 #define MGS_MAXREPSIZE (9 * 1024)
130 /* Absolute limits */
131 #define OSS_THREADS_MIN 2
132 #define OSS_THREADS_MAX 512
133 #define OST_NBUFS (64 * num_online_cpus())
134 #define OST_BUFSIZE (8 * 1024)
135 /* OST_MAXREQSIZE ~= 4768 bytes =
136 * lustre_msg + obdo + 16 * obd_ioobj + 256 * niobuf_remote
138 * - single object with 16 pages is 512 bytes
139 * - OST_MAXREQSIZE must be at least 1 page of cookies plus some spillover
141 #define OST_MAXREQSIZE (5 * 1024)
142 #define OST_MAXREPSIZE (9 * 1024)
144 struct ptlrpc_connection {
145 struct list_head c_link;
146 struct hlist_node c_hash;
148 lnet_process_id_t c_peer;
149 struct obd_uuid c_remote_uuid;
153 struct ptlrpc_client {
154 __u32 cli_request_portal;
155 __u32 cli_reply_portal;
159 /* state flags of requests */
160 /* XXX only ones left are those used by the bulk descs as well! */
161 #define PTL_RPC_FL_INTR (1 << 0) /* reply wait was interrupted by user */
162 #define PTL_RPC_FL_TIMEOUT (1 << 7) /* request timed out waiting for reply */
164 #define REQ_MAX_ACK_LOCKS 8
166 union ptlrpc_async_args {
167 /* Scratchpad for passing args to completion interpreter. Users
168 * cast to the struct of their choosing, and LASSERT that this is
169 * big enough. For _tons_ of context, OBD_ALLOC a struct and store
170 * a pointer to it here. The pointer_arg ensures this struct is at
171 * least big enough for that. */
172 void *pointer_arg[9];
176 struct ptlrpc_request_set;
177 typedef int (*set_interpreter_func)(struct ptlrpc_request_set *, void *, int);
179 struct ptlrpc_request_set {
180 int set_remaining; /* # uncompleted requests */
181 cfs_waitq_t set_waitq;
182 cfs_waitq_t *set_wakeup_ptr;
183 struct list_head set_requests;
184 struct list_head set_cblist; /* list of completion callbacks */
185 set_interpreter_func set_interpret; /* completion callback */
186 void *set_arg; /* completion context */
187 void *set_countp; /* pointer to NOB counter in case
188 * of directIO (bug11737) */
189 /* locked so that any old caller can communicate requests to
190 * the set holder who can then fold them into the lock-free set */
191 spinlock_t set_new_req_lock;
192 struct list_head set_new_requests;
195 struct ptlrpc_set_cbdata {
196 struct list_head psc_item;
197 set_interpreter_func psc_interpret;
201 struct ptlrpc_bulk_desc;
204 * ptlrpc callback & work item stuff
206 struct ptlrpc_cb_id {
207 void (*cbid_fn)(lnet_event_t *ev); /* specific callback fn */
208 void *cbid_arg; /* additional arg */
211 #define RS_MAX_LOCKS 4
214 struct ptlrpc_reply_state {
215 struct ptlrpc_cb_id rs_cb_id;
216 struct list_head rs_list;
217 struct list_head rs_exp_list;
218 struct list_head rs_obd_list;
220 struct list_head rs_debug_list;
222 /* updates to following flag serialised by srv_request_lock */
223 unsigned int rs_difficult:1; /* ACK/commit stuff */
224 unsigned int rs_scheduled:1; /* being handled? */
225 unsigned int rs_scheduled_ever:1;/* any schedule attempts? */
226 unsigned int rs_handled:1; /* been handled yet? */
227 unsigned int rs_on_net:1; /* reply_out_callback pending? */
228 unsigned int rs_prealloc:1; /* rs from prealloc list */
233 struct obd_export *rs_export;
234 struct ptlrpc_service *rs_service;
235 lnet_handle_md_t rs_md_h;
236 atomic_t rs_refcount;
238 /* locks awaiting client reply ACK */
240 struct lustre_handle rs_locks[RS_MAX_LOCKS];
241 ldlm_mode_t rs_modes[RS_MAX_LOCKS];
242 /* last member: variable sized reply message */
243 struct lustre_msg *rs_msg;
246 struct ptlrpc_thread;
249 RQ_PHASE_NEW = 0xebc0de00,
250 RQ_PHASE_RPC = 0xebc0de01,
251 RQ_PHASE_BULK = 0xebc0de02,
252 RQ_PHASE_INTERPRET = 0xebc0de03,
253 RQ_PHASE_COMPLETE = 0xebc0de04,
256 struct ptlrpc_request_pool {
258 struct list_head prp_req_list; /* list of ptlrpc_request structs */
260 void (*prp_populate)(struct ptlrpc_request_pool *, int);
263 struct ptlrpc_request {
264 int rq_type; /* one of PTL_RPC_MSG_* */
265 struct list_head rq_list;
266 struct list_head rq_timed_list; /* server-side early replies */
267 struct list_head rq_history_list; /* server-side history */
268 __u64 rq_history_seq; /* history sequence # */
271 /* client-side flags are serialized by rq_lock */
272 unsigned int rq_intr:1, rq_replied:1, rq_err:1,
273 rq_timedout:1, rq_resend:1, rq_restart:1,
275 * when ->rq_replay is set, request is kept by the client even
276 * after server commits corresponding transaction. This is
277 * used for operations that require sequence of multiple
278 * requests to be replayed. The only example currently is file
279 * open/close. When last request in such a sequence is
280 * committed, ->rq_replay is cleared on all requests in the
284 rq_no_resend:1, rq_waiting:1, rq_receiving_reply:1,
285 rq_no_delay:1, rq_net_err:1, rq_early:1, rq_must_unlink:1,
286 /* server-side flags */
287 rq_packed_final:1; /* packed final reply */
288 enum rq_phase rq_phase; /* one of RQ_PHASE_* */
289 atomic_t rq_refcount; /* client-side refcount for SENT race,
290 server-side refcounf for multiple replies */
292 struct ptlrpc_thread *rq_svc_thread; /* initial thread servicing req */
294 int rq_request_portal; /* XXX FIXME bug 249 */
295 int rq_reply_portal; /* XXX FIXME bug 249 */
297 int rq_nob_received; /* client-side # reply bytes actually received */
300 struct lustre_msg *rq_reqmsg;
303 struct lustre_msg *rq_repbuf; /* client only, buf may be bigger than msg */
304 struct lustre_msg *rq_repmsg;
307 struct list_head rq_replay_list;
309 __u32 rq_req_swab_mask;
310 __u32 rq_rep_swab_mask;
312 int rq_import_generation;
313 enum lustre_imp_state rq_send_state;
315 int rq_early_count; /* how many early replies (for stats) */
317 /* client+server request */
318 lnet_handle_md_t rq_req_md_h;
319 struct ptlrpc_cb_id rq_req_cbid;
322 struct timeval rq_arrival_time; /* request arrival time */
323 struct ptlrpc_reply_state *rq_reply_state; /* separated reply state */
324 struct ptlrpc_request_buffer_desc *rq_rqbd; /* incoming request buffer*/
326 __u32 rq_uid; /* peer uid, used in MDS only */
329 /* client-only incoming reply */
330 lnet_handle_md_t rq_reply_md_h;
331 cfs_waitq_t rq_reply_waitq;
332 struct ptlrpc_cb_id rq_reply_cbid;
335 lnet_process_id_t rq_peer;
336 struct obd_export *rq_export;
337 struct obd_import *rq_import;
339 void (*rq_replay_cb)(struct ptlrpc_request *);
340 void (*rq_commit_cb)(struct ptlrpc_request *);
343 struct ptlrpc_bulk_desc *rq_bulk; /* client side bulk */
344 /* client outgoing req */
345 time_t rq_sent; /* when request sent, seconds,
346 * or time when request should
348 volatile time_t rq_deadline; /* when request must finish. volatile
349 so that servers' early reply updates to the deadline aren't
350 kept in per-cpu cache */
351 int rq_timeout; /* service time estimate (secs) */
354 struct list_head rq_set_chain;
355 struct ptlrpc_request_set *rq_set;
356 void *rq_interpret_reply; /* Async completion handler */
357 union ptlrpc_async_args rq_async_args; /* Async completion context */
358 struct ptlrpc_request_pool *rq_pool; /* Pool if request from
362 static inline void lustre_set_req_swabbed(struct ptlrpc_request *req, int index)
364 LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
365 LASSERT((req->rq_req_swab_mask & (1 << index)) == 0);
366 req->rq_req_swab_mask |= 1 << index;
369 static inline void lustre_set_rep_swabbed(struct ptlrpc_request *req, int index)
371 LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
372 LASSERT((req->rq_rep_swab_mask & (1 << index)) == 0);
373 req->rq_rep_swab_mask |= 1 << index;
376 static inline int lustre_req_swabbed(struct ptlrpc_request *req, int index)
378 LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
379 return req->rq_req_swab_mask & (1 << index);
382 static inline int lustre_rep_swabbed(struct ptlrpc_request *req, int index)
384 LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
385 return req->rq_rep_swab_mask & (1 << index);
388 static inline const char *
389 ptlrpc_rqphase2str(struct ptlrpc_request *req)
391 switch (req->rq_phase) {
398 case RQ_PHASE_INTERPRET:
400 case RQ_PHASE_COMPLETE:
407 /* Spare the preprocessor, spoil the bugs. */
408 #define FLAG(field, str) (field ? str : "")
410 #define DEBUG_REQ_FLAGS(req) \
411 ptlrpc_rqphase2str(req), \
412 FLAG(req->rq_intr, "I"), FLAG(req->rq_replied, "R"), \
413 FLAG(req->rq_err, "E"), \
414 FLAG(req->rq_timedout, "X") /* eXpired */, FLAG(req->rq_resend, "S"), \
415 FLAG(req->rq_restart, "T"), FLAG(req->rq_replay, "P"), \
416 FLAG(req->rq_no_resend, "N"), \
417 FLAG(req->rq_waiting, "W")
419 #define REQ_FLAGS_FMT "%s:%s%s%s%s%s%s%s%s%s"
421 void _debug_req(struct ptlrpc_request *req, __u32 mask,
422 struct libcfs_debug_msg_data *data, const char *fmt, ...)
423 __attribute__ ((format (printf, 4, 5)));
425 #define debug_req(cdls, level, req, file, func, line, fmt, a...) \
429 if (((level) & D_CANTMASK) != 0 || \
430 ((libcfs_debug & (level)) != 0 && \
431 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) { \
432 static struct libcfs_debug_msg_data _req_dbg_data = \
433 DEBUG_MSG_DATA_INIT(cdls, DEBUG_SUBSYSTEM, file, func, line); \
434 _debug_req((req), (level), &_req_dbg_data, fmt, ##a); \
438 /* for most callers (level is a constant) this is resolved at compile time */
439 #define DEBUG_REQ(level, req, fmt, args...) \
441 if ((level) & (D_ERROR | D_WARNING)) { \
442 static cfs_debug_limit_state_t cdls; \
443 debug_req(&cdls, level, req, __FILE__, __func__, __LINE__, \
444 "@@@ "fmt" ", ## args); \
446 debug_req(NULL, level, req, __FILE__, __func__, __LINE__, \
447 "@@@ "fmt" ", ## args); \
450 struct ptlrpc_bulk_page {
451 struct list_head bp_link;
453 int bp_pageoffset; /* offset within a page */
454 struct page *bp_page;
457 #define BULK_GET_SOURCE 0
458 #define BULK_PUT_SINK 1
459 #define BULK_GET_SINK 2
460 #define BULK_PUT_SOURCE 3
462 struct ptlrpc_bulk_desc {
463 unsigned int bd_success:1; /* completed successfully */
464 unsigned int bd_network_rw:1; /* accessible to the network */
465 unsigned int bd_type:2; /* {put,get}{source,sink} */
466 unsigned int bd_registered:1; /* client side */
467 spinlock_t bd_lock; /* serialise with callback */
468 int bd_import_generation;
469 struct obd_export *bd_export;
470 struct obd_import *bd_import;
472 struct ptlrpc_request *bd_req; /* associated request */
473 cfs_waitq_t bd_waitq; /* server side only WQ */
474 int bd_iov_count; /* # entries in bd_iov */
475 int bd_max_iov; /* allocated size of bd_iov */
476 int bd_nob; /* # bytes covered */
477 int bd_nob_transferred; /* # bytes GOT/PUT */
481 struct ptlrpc_cb_id bd_cbid; /* network callback info */
482 lnet_handle_md_t bd_md_h; /* associated MD */
483 lnet_nid_t bd_sender; /* stash event::sender */
485 #if defined(__KERNEL__)
486 lnet_kiov_t bd_iov[0];
488 lnet_md_iovec_t bd_iov[0];
492 struct ptlrpc_thread {
494 struct list_head t_link; /* active threads for service, from svc->srv_threads */
496 void *t_data; /* thread-private data (preallocated memory) */
499 unsigned int t_id; /* service thread index, from ptlrpc_start_threads */
500 cfs_waitq_t t_ctl_waitq;
503 struct ptlrpc_request_buffer_desc {
504 struct list_head rqbd_list;
505 struct list_head rqbd_reqs;
506 struct ptlrpc_service *rqbd_service;
507 lnet_handle_md_t rqbd_md_h;
510 struct ptlrpc_cb_id rqbd_cbid;
511 struct ptlrpc_request rqbd_req;
514 typedef int (*svc_handler_t)(struct ptlrpc_request *req);
515 typedef void (*svcreq_printfn_t)(void *, struct ptlrpc_request *);
517 struct ptlrpc_service {
518 struct list_head srv_list; /* chain thru all services */
519 int srv_max_req_size; /* biggest request to receive */
520 int srv_max_reply_size; /* biggest reply to send */
521 int srv_buf_size; /* size of individual buffers */
522 int srv_nbuf_per_group; /* # buffers to allocate in 1 group */
523 int srv_nbufs; /* total # req buffer descs allocated */
524 int srv_threads_min; /* threads to start at SOW */
525 int srv_threads_max; /* thread upper limit */
526 int srv_threads_started; /* index of last started thread */
527 int srv_threads_running; /* # running threads */
528 int srv_n_difficult_replies; /* # 'difficult' replies */
529 int srv_n_active_reqs; /* # reqs being served */
530 cfs_duration_t srv_rqbd_timeout; /* timeout before re-posting reqs, in tick */
531 int srv_watchdog_factor; /* soft watchdog timeout mutiplier */
532 unsigned srv_cpu_affinity:1; /* bind threads to CPUs */
533 unsigned srv_at_check:1; /* check early replies */
535 __u32 srv_req_portal;
536 __u32 srv_rep_portal;
539 struct adaptive_timeout srv_at_estimate;/* estimated service time */
540 spinlock_t srv_at_lock;
541 struct list_head srv_at_list; /* reqs waiting for replies */
542 cfs_timer_t srv_at_timer; /* early reply timer */
544 int srv_n_queued_reqs; /* # reqs in either of the queues below */
545 struct list_head srv_req_in_queue; /* incoming reqs */
546 struct list_head srv_request_queue; /* reqs waiting for service */
548 struct list_head srv_request_history; /* request history */
549 __u64 srv_request_seq; /* next request sequence # */
550 __u64 srv_request_max_cull_seq; /* highest seq culled from history */
551 svcreq_printfn_t srv_request_history_print_fn; /* service-specific print fn */
553 struct list_head srv_idle_rqbds; /* request buffers to be reposted */
554 struct list_head srv_active_rqbds; /* req buffers receiving */
555 struct list_head srv_history_rqbds; /* request buffer history */
556 int srv_nrqbd_receiving; /* # posted request buffers */
557 int srv_n_history_rqbds; /* # request buffers in history */
558 int srv_max_history_rqbds;/* max # request buffers in history */
560 atomic_t srv_outstanding_replies;
561 struct list_head srv_active_replies; /* all the active replies */
562 struct list_head srv_reply_queue; /* replies waiting for service */
564 cfs_waitq_t srv_waitq; /* all threads sleep on this. This
565 * wait-queue is signalled when new
566 * incoming request arrives and when
567 * difficult reply has to be handled. */
569 struct list_head srv_threads; /* service thread list */
570 svc_handler_t srv_handler;
572 char *srv_name; /* only statically allocated strings here; we don't clean them */
573 char *srv_thread_name; /* only statically allocated strings here; we don't clean them */
577 cfs_proc_dir_entry_t *srv_procroot;
578 struct lprocfs_stats *srv_stats;
580 /* List of free reply_states */
581 struct list_head srv_free_rs_list;
582 /* waitq to run, when adding stuff to srv_free_rs_list */
583 cfs_waitq_t srv_free_rs_waitq;
586 * if non-NULL called during thread creation (ptlrpc_start_thread())
587 * to initialize service specific per-thread state.
589 int (*srv_init)(struct ptlrpc_thread *thread);
591 * if non-NULL called during thread shutdown (ptlrpc_main()) to
592 * destruct state created by ->srv_init().
594 void (*srv_done)(struct ptlrpc_thread *thread);
596 //struct ptlrpc_srv_ni srv_interfaces[0];
599 /* ptlrpc/events.c */
600 extern lnet_handle_eq_t ptlrpc_eq_h;
601 extern int ptlrpc_uuid_to_peer(struct obd_uuid *uuid,
602 lnet_process_id_t *peer, lnet_nid_t *self);
603 extern void request_out_callback (lnet_event_t *ev);
604 extern void reply_in_callback(lnet_event_t *ev);
605 extern void client_bulk_callback (lnet_event_t *ev);
606 extern void request_in_callback(lnet_event_t *ev);
607 extern void reply_out_callback(lnet_event_t *ev);
608 extern void server_bulk_callback (lnet_event_t *ev);
610 /* ptlrpc/connection.c */
611 void ptlrpc_dump_connections(void);
612 void ptlrpc_readdress_connection(struct ptlrpc_connection *, struct obd_uuid *);
613 struct ptlrpc_connection *ptlrpc_get_connection(lnet_process_id_t peer,
614 lnet_nid_t self, struct obd_uuid *uuid);
615 int ptlrpc_put_connection(struct ptlrpc_connection *c);
616 struct ptlrpc_connection *ptlrpc_connection_addref(struct ptlrpc_connection *);
617 int ptlrpc_init_connection(void);
618 void ptlrpc_cleanup_connection(void);
619 extern lnet_pid_t ptl_get_pid(void);
621 /* ptlrpc/niobuf.c */
622 int ptlrpc_start_bulk_transfer(struct ptlrpc_bulk_desc *desc);
623 void ptlrpc_abort_bulk(struct ptlrpc_bulk_desc *desc);
624 int ptlrpc_register_bulk(struct ptlrpc_request *req);
625 void ptlrpc_unregister_bulk (struct ptlrpc_request *req);
627 static inline int ptlrpc_bulk_active (struct ptlrpc_bulk_desc *desc)
631 spin_lock(&desc->bd_lock);
632 rc = desc->bd_network_rw;
633 spin_unlock(&desc->bd_lock);
637 #define PTLRPC_REPLY_MAYBE_DIFFICULT 0x01
638 #define PTLRPC_REPLY_EARLY 0x02
639 int ptlrpc_send_reply(struct ptlrpc_request *req, int flags);
640 int ptlrpc_reply(struct ptlrpc_request *req);
641 int ptlrpc_error(struct ptlrpc_request *req);
642 void ptlrpc_resend_req(struct ptlrpc_request *request);
643 int ptlrpc_at_get_net_latency(struct ptlrpc_request *req);
644 int ptl_send_rpc(struct ptlrpc_request *request, int noreply);
645 int ptlrpc_register_rqbd (struct ptlrpc_request_buffer_desc *rqbd);
647 /* ptlrpc/client.c */
648 void ptlrpc_init_client(int req_portal, int rep_portal, char *name,
649 struct ptlrpc_client *);
650 void ptlrpc_cleanup_client(struct obd_import *imp);
651 struct ptlrpc_connection *ptlrpc_uuid_to_connection(struct obd_uuid *uuid);
654 ptlrpc_client_recv_or_unlink (struct ptlrpc_request *req)
658 spin_lock(&req->rq_lock);
659 rc = req->rq_receiving_reply || req->rq_must_unlink;
660 spin_unlock(&req->rq_lock);
665 ptlrpc_wake_client_req (struct ptlrpc_request *req)
667 if (req->rq_set == NULL)
668 cfs_waitq_signal(&req->rq_reply_waitq);
670 cfs_waitq_signal(&req->rq_set->set_waitq);
673 int ptlrpc_queue_wait(struct ptlrpc_request *req);
674 int ptlrpc_replay_req(struct ptlrpc_request *req);
675 void ptlrpc_unregister_reply(struct ptlrpc_request *req);
676 void ptlrpc_restart_req(struct ptlrpc_request *req);
677 void ptlrpc_abort_inflight(struct obd_import *imp);
679 struct ptlrpc_request_set *ptlrpc_prep_set(void);
680 int ptlrpc_set_add_cb(struct ptlrpc_request_set *set,
681 set_interpreter_func fn, void *data);
682 int ptlrpc_set_next_timeout(struct ptlrpc_request_set *);
683 int ptlrpc_check_set(struct ptlrpc_request_set *set);
684 int ptlrpc_set_wait(struct ptlrpc_request_set *);
685 int ptlrpc_expired_set(void *data);
686 void ptlrpc_interrupted_set(void *data);
687 void ptlrpc_mark_interrupted(struct ptlrpc_request *req);
688 void ptlrpc_set_destroy(struct ptlrpc_request_set *);
689 void ptlrpc_set_add_req(struct ptlrpc_request_set *, struct ptlrpc_request *);
690 void ptlrpc_set_add_new_req(struct ptlrpc_request_set *,
691 struct ptlrpc_request *);
693 void ptlrpc_free_rq_pool(struct ptlrpc_request_pool *pool);
694 void ptlrpc_add_rqs_to_pool(struct ptlrpc_request_pool *pool, int num_rq);
695 struct ptlrpc_request_pool *ptlrpc_init_rq_pool(int, int,
696 void (*populate_pool)(struct ptlrpc_request_pool *, int));
697 struct ptlrpc_request *ptlrpc_prep_req(struct obd_import *imp, __u32 version,
698 int opcode, int count, int *lengths,
700 struct ptlrpc_request *ptlrpc_prep_req_pool(struct obd_import *imp,
701 __u32 version, int opcode,
702 int count, int *lengths, char **bufs,
703 struct ptlrpc_request_pool *pool);
704 void ptlrpc_free_req(struct ptlrpc_request *request);
705 void ptlrpc_req_finished(struct ptlrpc_request *request);
706 void ptlrpc_req_finished_with_imp_lock(struct ptlrpc_request *request);
707 struct ptlrpc_request *ptlrpc_request_addref(struct ptlrpc_request *req);
708 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_imp (struct ptlrpc_request *req,
709 int npages, int type, int portal);
710 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_exp(struct ptlrpc_request *req,
711 int npages, int type, int portal);
712 void ptlrpc_free_bulk(struct ptlrpc_bulk_desc *bulk);
713 void ptlrpc_prep_bulk_page(struct ptlrpc_bulk_desc *desc,
714 cfs_page_t *page, int pageoffset, int len);
715 void ptlrpc_retain_replayable_request(struct ptlrpc_request *req,
716 struct obd_import *imp);
717 __u64 ptlrpc_next_xid(void);
718 __u64 ptlrpc_sample_next_xid(void);
719 __u64 ptlrpc_req_xid(struct ptlrpc_request *request);
721 /* ptlrpc/service.c */
722 void ptlrpc_save_lock (struct ptlrpc_request *req,
723 struct lustre_handle *lock, int mode);
724 void ptlrpc_commit_replies (struct obd_device *obd);
725 void ptlrpc_schedule_difficult_reply (struct ptlrpc_reply_state *rs);
726 struct ptlrpc_service *ptlrpc_init_svc(int nbufs, int bufsize, int max_req_size,
728 int req_portal, int rep_portal,
730 svc_handler_t, char *name,
731 cfs_proc_dir_entry_t *proc_entry,
733 int min_threads, int max_threads,
735 void ptlrpc_stop_all_threads(struct ptlrpc_service *svc);
737 int ptlrpc_start_threads(struct obd_device *dev, struct ptlrpc_service *svc);
738 int ptlrpc_start_thread(struct obd_device *dev, struct ptlrpc_service *svc);
739 int ptlrpc_unregister_service(struct ptlrpc_service *service);
740 int liblustre_check_services (void *arg);
741 void ptlrpc_daemonize(char *name);
742 int ptlrpc_service_health_check(struct ptlrpc_service *);
745 struct ptlrpc_svc_data {
747 struct ptlrpc_service *svc;
748 struct ptlrpc_thread *thread;
749 struct obd_device *dev;
752 /* ptlrpc/import.c */
753 int ptlrpc_connect_import(struct obd_import *imp, char * new_uuid);
754 int ptlrpc_init_import(struct obd_import *imp);
755 int ptlrpc_disconnect_import(struct obd_import *imp, int noclose);
756 int ptlrpc_import_recovery_state_machine(struct obd_import *imp);
757 void ptlrpc_import_setasync(struct obd_import *imp, int count);
759 /* ptlrpc/pack_generic.c */
760 int lustre_msg_swabbed(struct lustre_msg *msg);
761 int lustre_msg_check_version(struct lustre_msg *msg, __u32 version);
762 int lustre_pack_request(struct ptlrpc_request *, __u32 magic, int count,
763 int *lens, char **bufs);
764 int lustre_pack_reply(struct ptlrpc_request *, int count, int *lens,
766 #define LPRFL_EARLY_REPLY 1
767 int lustre_pack_reply_flags(struct ptlrpc_request *, int count, int *lens,
768 char **bufs, int flags);
769 void lustre_shrink_reply(struct ptlrpc_request *req, int segment,
770 unsigned int newlen, int move_data);
771 void lustre_free_reply_state(struct ptlrpc_reply_state *rs);
772 int lustre_msg_size(__u32 magic, int count, int *lengths);
773 int lustre_msg_early_size(void);
774 int lustre_unpack_msg(struct lustre_msg *m, int len);
775 void *lustre_msg_buf(struct lustre_msg *m, int n, int minlen);
776 int lustre_msg_buflen(struct lustre_msg *m, int n);
777 void lustre_msg_set_buflen(struct lustre_msg *m, int n, int len);
778 int lustre_msg_bufcount(struct lustre_msg *m);
779 char *lustre_msg_string (struct lustre_msg *m, int n, int max_len);
780 void *lustre_swab_buf(struct lustre_msg *, int n, int minlen, void *swabber);
781 void *lustre_swab_reqbuf(struct ptlrpc_request *req, int n, int minlen,
783 void *lustre_swab_repbuf(struct ptlrpc_request *req, int n, int minlen,
785 __u32 lustre_msg_get_flags(struct lustre_msg *msg);
786 void lustre_msg_add_flags(struct lustre_msg *msg, int flags);
787 void lustre_msg_set_flags(struct lustre_msg *msg, int flags);
788 void lustre_msg_clear_flags(struct lustre_msg *msg, int flags);
789 __u32 lustre_msg_get_op_flags(struct lustre_msg *msg);
790 void lustre_msg_add_op_flags(struct lustre_msg *msg, int flags);
791 void lustre_msg_set_op_flags(struct lustre_msg *msg, int flags);
792 struct lustre_handle *lustre_msg_get_handle(struct lustre_msg *msg);
793 __u32 lustre_msg_get_type(struct lustre_msg *msg);
794 __u32 lustre_msg_get_version(struct lustre_msg *msg);
795 void lustre_msg_add_version(struct lustre_msg *msg, int version);
796 __u32 lustre_msg_get_opc(struct lustre_msg *msg);
797 __u64 lustre_msg_get_last_xid(struct lustre_msg *msg);
798 __u64 lustre_msg_get_last_committed(struct lustre_msg *msg);
799 __u64 lustre_msg_get_transno(struct lustre_msg *msg);
800 __u64 lustre_msg_get_slv(struct lustre_msg *msg);
801 __u32 lustre_msg_get_limit(struct lustre_msg *msg);
802 void lustre_msg_set_slv(struct lustre_msg *msg, __u64 slv);
803 void lustre_msg_set_limit(struct lustre_msg *msg, __u64 limit);
804 int lustre_msg_get_status(struct lustre_msg *msg);
805 __u32 lustre_msg_get_conn_cnt(struct lustre_msg *msg);
806 int lustre_msg_is_v1(struct lustre_msg *msg);
807 __u32 lustre_msg_get_magic(struct lustre_msg *msg);
808 __u32 lustre_msg_get_timeout(struct lustre_msg *msg);
809 __u32 lustre_msg_get_service_time(struct lustre_msg *msg);
810 __u32 lustre_msg_get_cksum(struct lustre_msg *msg);
811 __u32 lustre_msg_calc_cksum(struct lustre_msg *msg);
812 void lustre_msg_set_handle(struct lustre_msg *msg,struct lustre_handle *handle);
813 void lustre_msg_set_type(struct lustre_msg *msg, __u32 type);
814 void lustre_msg_set_opc(struct lustre_msg *msg, __u32 opc);
815 void lustre_msg_set_last_xid(struct lustre_msg *msg, __u64 last_xid);
816 void lustre_msg_set_last_committed(struct lustre_msg *msg,__u64 last_committed);
817 void lustre_msg_set_transno(struct lustre_msg *msg, __u64 transno);
818 void lustre_msg_set_status(struct lustre_msg *msg, __u32 status);
819 void lustre_msg_set_conn_cnt(struct lustre_msg *msg, __u32 conn_cnt);
820 void lustre_msg_set_timeout(struct lustre_msg *msg, __u32 timeout);
821 void lustre_msg_set_service_time(struct lustre_msg *msg, __u32 service_time);
822 void lustre_msg_set_cksum(struct lustre_msg *msg, __u32 cksum);
825 ptlrpc_rs_addref(struct ptlrpc_reply_state *rs)
827 LASSERT(atomic_read(&rs->rs_refcount) > 0);
828 atomic_inc(&rs->rs_refcount);
832 ptlrpc_rs_decref(struct ptlrpc_reply_state *rs)
834 LASSERT(atomic_read(&rs->rs_refcount) > 0);
835 if (atomic_dec_and_test(&rs->rs_refcount))
836 lustre_free_reply_state(rs);
839 /* Should only be called once per req */
840 static inline void ptlrpc_req_drop_rs(struct ptlrpc_request *req)
842 if (req->rq_reply_state == NULL)
843 return; /* shouldn't occur */
844 ptlrpc_rs_decref(req->rq_reply_state);
845 req->rq_reply_state = NULL;
846 req->rq_repmsg = NULL;
849 static inline __u32 lustre_request_magic(struct ptlrpc_request *req)
851 return lustre_msg_get_magic(req->rq_reqmsg);
854 static inline int ptlrpc_req_get_repsize(struct ptlrpc_request *req)
856 switch (req->rq_reqmsg->lm_magic) {
857 case LUSTRE_MSG_MAGIC_V1:
858 CERROR("function not supported for lustre_msg V1!\n");
860 case LUSTRE_MSG_MAGIC_V2:
861 return req->rq_reqmsg->lm_repsize;
863 LASSERTF(0, "incorrect message magic: %08x\n",
864 req->rq_reqmsg->lm_magic);
870 ptlrpc_req_set_repsize(struct ptlrpc_request *req, int count, int *lens)
872 int size = lustre_msg_size(req->rq_reqmsg->lm_magic, count, lens);
874 req->rq_replen = size + lustre_msg_early_size();
875 if (req->rq_reqmsg->lm_magic == LUSTRE_MSG_MAGIC_V2)
876 req->rq_reqmsg->lm_repsize = size;
879 /* ldlm/ldlm_lib.c */
880 int client_obd_setup(struct obd_device *obddev, obd_count len, void *buf);
881 int client_obd_cleanup(struct obd_device * obddev);
882 int client_connect_import(struct lustre_handle *conn, struct obd_device *obd,
883 struct obd_uuid *cluuid, struct obd_connect_data *,
885 int client_disconnect_export(struct obd_export *exp);
886 int client_import_add_conn(struct obd_import *imp, struct obd_uuid *uuid,
888 int client_import_del_conn(struct obd_import *imp, struct obd_uuid *uuid);
889 int import_set_conn_priority(struct obd_import *imp, struct obd_uuid *uuid);
891 /* ptlrpc/pinger.c */
892 int ptlrpc_pinger_add_import(struct obd_import *imp);
893 int ptlrpc_pinger_del_import(struct obd_import *imp);
895 void ping_evictor_start(void);
896 void ping_evictor_stop(void);
898 #define ping_evictor_start() do {} while (0)
899 #define ping_evictor_stop() do {} while (0)
902 /* ptlrpc/ptlrpcd.c */
903 void ptlrpcd_wake(struct ptlrpc_request *req);
904 void ptlrpcd_add_req(struct ptlrpc_request *req);
905 int ptlrpcd_addref(void);
906 void ptlrpcd_decref(void);
908 /* ptlrpc/lproc_ptlrpc.c */
909 const char* ll_opcode2str(__u32 opcode);
911 void ptlrpc_lprocfs_register_obd(struct obd_device *obd);
912 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd);
913 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int opc, int bytes);
915 static inline void ptlrpc_lprocfs_register_obd(struct obd_device *obd) {}
916 static inline void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd) {}
917 static inline void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int opc,
921 /* ptlrpc/llog_server.c */
922 int llog_origin_handle_create(struct ptlrpc_request *req);
923 int llog_origin_handle_destroy(struct ptlrpc_request *req);
924 int llog_origin_handle_prev_block(struct ptlrpc_request *req);
925 int llog_origin_handle_next_block(struct ptlrpc_request *req);
926 int llog_origin_handle_read_header(struct ptlrpc_request *req);
927 int llog_origin_handle_close(struct ptlrpc_request *req);
928 int llog_origin_handle_cancel(struct ptlrpc_request *req);
929 int llog_catinfo(struct ptlrpc_request *req);
931 /* ptlrpc/llog_client.c */
932 extern struct llog_operations llog_client_ops;