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 min((int)(smp_num_cpus * smp_num_cpus * 2), 8)
89 #define LDLM_THREADS_AUTO_MAX (LDLM_THREADS_AUTO_MIN * 16)
90 #define LDLM_BL_THREADS LDLM_THREADS_AUTO_MIN
91 #define LDLM_NBUFS (64 * smp_num_cpus)
92 #define LDLM_BUFSIZE (8 * 1024)
93 #define LDLM_MAXREQSIZE (5 * 1024)
94 #define LDLM_MAXREPSIZE (1024)
97 #define MDS_THREADS_MIN 2
98 #define MDS_THREADS_MAX 512
99 #define MDS_THREADS_MIN_READPAGE 2
100 #define MDS_NBUFS (64 * smp_num_cpus)
101 #define MDS_BUFSIZE (8 * 1024)
102 /* Assume file name length = FNAME_MAX = 256 (true for ext3).
103 * path name length = PATH_MAX = 4096
104 * LOV MD size max = EA_MAX = 4000
105 * symlink: FNAME_MAX + PATH_MAX <- largest
106 * link: FNAME_MAX + PATH_MAX (mds_rec_link < mds_rec_create)
107 * rename: FNAME_MAX + FNAME_MAX
108 * open: FNAME_MAX + EA_MAX
110 * MDS_MAXREQSIZE ~= 4736 bytes =
111 * lustre_msg + ldlm_request + mds_body + mds_rec_create + FNAME_MAX + PATH_MAX
112 * MDS_MAXREPSIZE ~= 8300 bytes = lustre_msg + llog_header
113 * or, for mds_close() and mds_reint_unlink() on a many-OST filesystem:
114 * = 9210 bytes = lustre_msg + mds_body + 160 * (easize + cookiesize)
116 * Realistic size is about 512 bytes (20 character name + 128 char symlink),
117 * except in the open case where there are a large number of OSTs in a LOV.
119 #define MDS_MAXREQSIZE (5 * 1024)
120 #define MDS_MAXREPSIZE max(9 * 1024, 280 + LOV_MAX_STRIPE_COUNT * 56)
122 #define MGS_THREADS_AUTO_MIN 2
123 #define MGS_THREADS_AUTO_MAX 32
124 #define MGS_NBUFS (64 * smp_num_cpus)
125 #define MGS_BUFSIZE (8 * 1024)
126 #define MGS_MAXREQSIZE (8 * 1024)
127 #define MGS_MAXREPSIZE (9 * 1024)
129 /* Absolute limits */
130 #define OSS_THREADS_MIN 2
131 #define OSS_THREADS_MAX 512
132 #define OST_NBUFS (64 * smp_num_cpus)
133 #define OST_BUFSIZE (8 * 1024)
134 /* OST_MAXREQSIZE ~= 4768 bytes =
135 * lustre_msg + obdo + 16 * obd_ioobj + 256 * niobuf_remote
137 * - single object with 16 pages is 512 bytes
138 * - OST_MAXREQSIZE must be at least 1 page of cookies plus some spillover
140 #define OST_MAXREQSIZE (5 * 1024)
141 #define OST_MAXREPSIZE (9 * 1024)
143 struct ptlrpc_connection {
144 struct list_head c_link;
145 struct hlist_node c_hash;
147 lnet_process_id_t c_peer;
148 struct obd_uuid c_remote_uuid;
152 struct ptlrpc_client {
153 __u32 cli_request_portal;
154 __u32 cli_reply_portal;
158 /* state flags of requests */
159 /* XXX only ones left are those used by the bulk descs as well! */
160 #define PTL_RPC_FL_INTR (1 << 0) /* reply wait was interrupted by user */
161 #define PTL_RPC_FL_TIMEOUT (1 << 7) /* request timed out waiting for reply */
163 #define REQ_MAX_ACK_LOCKS 8
165 union ptlrpc_async_args {
166 /* Scratchpad for passing args to completion interpreter. Users
167 * cast to the struct of their choosing, and LASSERT that this is
168 * big enough. For _tons_ of context, OBD_ALLOC a struct and store
169 * a pointer to it here. The pointer_arg ensures this struct is at
170 * least big enough for that. */
171 void *pointer_arg[9];
175 struct ptlrpc_request_set;
176 typedef int (*set_interpreter_func)(struct ptlrpc_request_set *, void *, int);
178 struct ptlrpc_request_set {
179 int set_remaining; /* # uncompleted requests */
180 cfs_waitq_t set_waitq;
181 cfs_waitq_t *set_wakeup_ptr;
182 struct list_head set_requests;
183 set_interpreter_func set_interpret; /* completion callback */
184 void *set_arg; /* completion context */
185 void *set_countp; /* pointer to NOB counter in case
186 * of directIO (bug11737) */
187 /* locked so that any old caller can communicate requests to
188 * the set holder who can then fold them into the lock-free set */
189 spinlock_t set_new_req_lock;
190 struct list_head set_new_requests;
193 struct ptlrpc_bulk_desc;
196 * ptlrpc callback & work item stuff
198 struct ptlrpc_cb_id {
199 void (*cbid_fn)(lnet_event_t *ev); /* specific callback fn */
200 void *cbid_arg; /* additional arg */
203 #define RS_MAX_LOCKS 4
206 struct ptlrpc_reply_state {
207 struct ptlrpc_cb_id rs_cb_id;
208 struct list_head rs_list;
209 struct list_head rs_exp_list;
210 struct list_head rs_obd_list;
212 struct list_head rs_debug_list;
214 /* updates to following flag serialised by srv_request_lock */
215 unsigned int rs_difficult:1; /* ACK/commit stuff */
216 unsigned int rs_scheduled:1; /* being handled? */
217 unsigned int rs_scheduled_ever:1;/* any schedule attempts? */
218 unsigned int rs_handled:1; /* been handled yet? */
219 unsigned int rs_on_net:1; /* reply_out_callback pending? */
220 unsigned int rs_prealloc:1; /* rs from prealloc list */
225 struct obd_export *rs_export;
226 struct ptlrpc_service *rs_service;
227 lnet_handle_md_t rs_md_h;
228 atomic_t rs_refcount;
230 /* locks awaiting client reply ACK */
232 struct lustre_handle rs_locks[RS_MAX_LOCKS];
233 ldlm_mode_t rs_modes[RS_MAX_LOCKS];
234 /* last member: variable sized reply message */
235 struct lustre_msg *rs_msg;
238 struct ptlrpc_thread;
241 RQ_PHASE_NEW = 0xebc0de00,
242 RQ_PHASE_RPC = 0xebc0de01,
243 RQ_PHASE_BULK = 0xebc0de02,
244 RQ_PHASE_INTERPRET = 0xebc0de03,
245 RQ_PHASE_COMPLETE = 0xebc0de04,
248 struct ptlrpc_request_pool {
250 struct list_head prp_req_list; /* list of ptlrpc_request structs */
252 void (*prp_populate)(struct ptlrpc_request_pool *, int);
255 struct ptlrpc_request {
256 int rq_type; /* one of PTL_RPC_MSG_* */
257 struct list_head rq_list;
258 struct list_head rq_timed_list; /* server-side early replies */
259 struct list_head rq_history_list; /* server-side history */
260 __u64 rq_history_seq; /* history sequence # */
263 /* client-side flags are serialized by rq_lock */
264 unsigned int rq_intr:1, rq_replied:1, rq_err:1,
265 rq_timedout:1, rq_resend:1, rq_restart:1,
267 * when ->rq_replay is set, request is kept by the client even
268 * after server commits corresponding transaction. This is
269 * used for operations that require sequence of multiple
270 * requests to be replayed. The only example currently is file
271 * open/close. When last request in such a sequence is
272 * committed, ->rq_replay is cleared on all requests in the
276 rq_no_resend:1, rq_waiting:1, rq_receiving_reply:1,
277 rq_no_delay:1, rq_net_err:1, rq_early:1, rq_must_unlink:1,
278 /* server-side flags */
279 rq_packed_final:1; /* packed final reply */
280 enum rq_phase rq_phase; /* one of RQ_PHASE_* */
281 atomic_t rq_refcount; /* client-side refcount for SENT race,
282 server-side refcounf for multiple replies */
284 struct ptlrpc_thread *rq_svc_thread; /* initial thread servicing req */
286 int rq_request_portal; /* XXX FIXME bug 249 */
287 int rq_reply_portal; /* XXX FIXME bug 249 */
289 int rq_nob_received; /* client-side # reply bytes actually received */
292 struct lustre_msg *rq_reqmsg;
295 struct lustre_msg *rq_repbuf; /* client only, buf may be bigger than msg */
296 struct lustre_msg *rq_repmsg;
299 struct list_head rq_replay_list;
301 __u32 rq_req_swab_mask;
302 __u32 rq_rep_swab_mask;
304 int rq_import_generation;
305 enum lustre_imp_state rq_send_state;
307 int rq_early_count; /* how many early replies (for stats) */
309 /* client+server request */
310 lnet_handle_md_t rq_req_md_h;
311 struct ptlrpc_cb_id rq_req_cbid;
314 struct timeval rq_arrival_time; /* request arrival time */
315 struct ptlrpc_reply_state *rq_reply_state; /* separated reply state */
316 struct ptlrpc_request_buffer_desc *rq_rqbd; /* incoming request buffer*/
318 __u32 rq_uid; /* peer uid, used in MDS only */
321 /* client-only incoming reply */
322 lnet_handle_md_t rq_reply_md_h;
323 cfs_waitq_t rq_reply_waitq;
324 struct ptlrpc_cb_id rq_reply_cbid;
327 lnet_process_id_t rq_peer;
328 struct obd_export *rq_export;
329 struct obd_import *rq_import;
331 void (*rq_replay_cb)(struct ptlrpc_request *);
332 void (*rq_commit_cb)(struct ptlrpc_request *);
335 struct ptlrpc_bulk_desc *rq_bulk; /* client side bulk */
337 /* client outgoing req */
338 time_t rq_sent; /* when request/reply sent (secs) */
339 volatile time_t rq_deadline; /* when request must finish. volatile
340 so that servers' early reply updates to the deadline aren't
341 kept in per-cpu cache */
342 int rq_timeout; /* service time estimate (secs) */
345 struct list_head rq_set_chain;
346 struct ptlrpc_request_set *rq_set;
347 void *rq_interpret_reply; /* Async completion handler */
348 union ptlrpc_async_args rq_async_args; /* Async completion context */
349 void *rq_ptlrpcd_data;
350 struct ptlrpc_request_pool *rq_pool; /* Pool if request from
354 static inline void lustre_set_req_swabbed(struct ptlrpc_request *req, int index)
356 LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
357 LASSERT((req->rq_req_swab_mask & (1 << index)) == 0);
358 req->rq_req_swab_mask |= 1 << index;
361 static inline void lustre_set_rep_swabbed(struct ptlrpc_request *req, int index)
363 LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
364 LASSERT((req->rq_rep_swab_mask & (1 << index)) == 0);
365 req->rq_rep_swab_mask |= 1 << index;
368 static inline int lustre_req_swabbed(struct ptlrpc_request *req, int index)
370 LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
371 return req->rq_req_swab_mask & (1 << index);
374 static inline int lustre_rep_swabbed(struct ptlrpc_request *req, int index)
376 LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
377 return req->rq_rep_swab_mask & (1 << index);
380 #define SWAB_PARANOIA 1
383 /* unpacking: assert idx not unpacked already */
384 #define LASSERT_REQSWAB(rq, idx) lustre_set_req_swabbed(rq, idx)
385 #define LASSERT_REPSWAB(rq, idx) lustre_set_rep_swabbed(rq, idx)
387 /* just looking: assert idx already unpacked */
388 #define LASSERT_REQSWABBED(rq, idx) LASSERT(lustre_req_swabbed(rq, idx))
389 #define LASSERT_REPSWABBED(rq, idx) LASSERT(lustre_rep_swabbed(rq, idx))
392 #define LASSERT_REQSWAB(rq, idx)
393 #define LASSERT_REPSWAB(rq, idx)
394 #define LASSERT_REQSWABBED(rq, idx)
395 #define LASSERT_REPSWABBED(rq, idx)
398 static inline const char *
399 ptlrpc_rqphase2str(struct ptlrpc_request *req)
401 switch (req->rq_phase) {
408 case RQ_PHASE_INTERPRET:
410 case RQ_PHASE_COMPLETE:
417 /* Spare the preprocessor, spoil the bugs. */
418 #define FLAG(field, str) (field ? str : "")
420 #define DEBUG_REQ_FLAGS(req) \
421 ptlrpc_rqphase2str(req), \
422 FLAG(req->rq_intr, "I"), FLAG(req->rq_replied, "R"), \
423 FLAG(req->rq_err, "E"), \
424 FLAG(req->rq_timedout, "X") /* eXpired */, FLAG(req->rq_resend, "S"), \
425 FLAG(req->rq_restart, "T"), FLAG(req->rq_replay, "P"), \
426 FLAG(req->rq_no_resend, "N"), \
427 FLAG(req->rq_waiting, "W")
429 #define REQ_FLAGS_FMT "%s:%s%s%s%s%s%s%s%s%s"
431 void _debug_req(struct ptlrpc_request *req, __u32 mask,
432 struct libcfs_debug_msg_data *data, const char *fmt, ...)
433 __attribute__ ((format (printf, 4, 5)));
435 #define debug_req(cdls, level, req, file, func, line, fmt, a...) \
439 if (((level) & D_CANTMASK) != 0 || \
440 ((libcfs_debug & (level)) != 0 && \
441 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) { \
442 static struct libcfs_debug_msg_data _req_dbg_data = \
443 DEBUG_MSG_DATA_INIT(cdls, DEBUG_SUBSYSTEM, file, func, line); \
444 _debug_req((req), (level), &_req_dbg_data, fmt, ##a); \
448 /* for most callers (level is a constant) this is resolved at compile time */
449 #define DEBUG_REQ(level, req, fmt, args...) \
451 if ((level) & (D_ERROR | D_WARNING)) { \
452 static cfs_debug_limit_state_t cdls; \
453 debug_req(&cdls, level, req, __FILE__, __func__, __LINE__, \
454 "@@@ "fmt" ", ## args); \
456 debug_req(NULL, level, req, __FILE__, __func__, __LINE__, \
457 "@@@ "fmt" ", ## args); \
460 struct ptlrpc_bulk_page {
461 struct list_head bp_link;
463 int bp_pageoffset; /* offset within a page */
464 struct page *bp_page;
467 #define BULK_GET_SOURCE 0
468 #define BULK_PUT_SINK 1
469 #define BULK_GET_SINK 2
470 #define BULK_PUT_SOURCE 3
472 struct ptlrpc_bulk_desc {
473 unsigned int bd_success:1; /* completed successfully */
474 unsigned int bd_network_rw:1; /* accessible to the network */
475 unsigned int bd_type:2; /* {put,get}{source,sink} */
476 unsigned int bd_registered:1; /* client side */
477 spinlock_t bd_lock; /* serialise with callback */
478 int bd_import_generation;
479 struct obd_export *bd_export;
480 struct obd_import *bd_import;
482 struct ptlrpc_request *bd_req; /* associated request */
483 cfs_waitq_t bd_waitq; /* server side only WQ */
484 int bd_iov_count; /* # entries in bd_iov */
485 int bd_max_iov; /* allocated size of bd_iov */
486 int bd_nob; /* # bytes covered */
487 int bd_nob_transferred; /* # bytes GOT/PUT */
491 struct ptlrpc_cb_id bd_cbid; /* network callback info */
492 lnet_handle_md_t bd_md_h; /* associated MD */
493 lnet_nid_t bd_sender; /* stash event::sender */
495 #if defined(__KERNEL__)
496 lnet_kiov_t bd_iov[0];
498 lnet_md_iovec_t bd_iov[0];
502 struct ptlrpc_thread {
504 struct list_head t_link; /* active threads for service, from svc->srv_threads */
506 void *t_data; /* thread-private data (preallocated memory) */
509 unsigned int t_id; /* service thread index, from ptlrpc_start_threads */
510 cfs_waitq_t t_ctl_waitq;
513 struct ptlrpc_request_buffer_desc {
514 struct list_head rqbd_list;
515 struct list_head rqbd_reqs;
516 struct ptlrpc_service *rqbd_service;
517 lnet_handle_md_t rqbd_md_h;
520 struct ptlrpc_cb_id rqbd_cbid;
521 struct ptlrpc_request rqbd_req;
524 typedef int (*svc_handler_t)(struct ptlrpc_request *req);
525 typedef void (*svcreq_printfn_t)(void *, struct ptlrpc_request *);
527 struct ptlrpc_service {
528 struct list_head srv_list; /* chain thru all services */
529 int srv_max_req_size; /* biggest request to receive */
530 int srv_max_reply_size; /* biggest reply to send */
531 int srv_buf_size; /* size of individual buffers */
532 int srv_nbuf_per_group; /* # buffers to allocate in 1 group */
533 int srv_nbufs; /* total # req buffer descs allocated */
534 int srv_threads_min; /* threads to start at SOW */
535 int srv_threads_max; /* thread upper limit */
536 int srv_threads_started; /* index of last started thread */
537 int srv_threads_running; /* # running threads */
538 int srv_n_difficult_replies; /* # 'difficult' replies */
539 int srv_n_active_reqs; /* # reqs being served */
540 cfs_duration_t srv_rqbd_timeout; /* timeout before re-posting reqs, in tick */
541 int srv_watchdog_factor; /* soft watchdog timeout mutiplier */
542 unsigned srv_cpu_affinity:1; /* bind threads to CPUs */
543 unsigned srv_at_check:1; /* check early replies */
545 __u32 srv_req_portal;
546 __u32 srv_rep_portal;
549 struct adaptive_timeout srv_at_estimate;/* estimated service time */
550 spinlock_t srv_at_lock;
551 struct list_head srv_at_list; /* reqs waiting for replies */
552 cfs_timer_t srv_at_timer; /* early reply timer */
554 int srv_n_queued_reqs; /* # reqs in either of the queues below */
555 struct list_head srv_req_in_queue; /* incoming reqs */
556 struct list_head srv_request_queue; /* reqs waiting for service */
558 struct list_head srv_request_history; /* request history */
559 __u64 srv_request_seq; /* next request sequence # */
560 __u64 srv_request_max_cull_seq; /* highest seq culled from history */
561 svcreq_printfn_t srv_request_history_print_fn; /* service-specific print fn */
563 struct list_head srv_idle_rqbds; /* request buffers to be reposted */
564 struct list_head srv_active_rqbds; /* req buffers receiving */
565 struct list_head srv_history_rqbds; /* request buffer history */
566 int srv_nrqbd_receiving; /* # posted request buffers */
567 int srv_n_history_rqbds; /* # request buffers in history */
568 int srv_max_history_rqbds;/* max # request buffers in history */
570 atomic_t srv_outstanding_replies;
571 struct list_head srv_active_replies; /* all the active replies */
572 struct list_head srv_reply_queue; /* replies waiting for service */
574 cfs_waitq_t srv_waitq; /* all threads sleep on this. This
575 * wait-queue is signalled when new
576 * incoming request arrives and when
577 * difficult reply has to be handled. */
579 struct list_head srv_threads; /* service thread list */
580 svc_handler_t srv_handler;
582 char *srv_name; /* only statically allocated strings here; we don't clean them */
583 char *srv_thread_name; /* only statically allocated strings here; we don't clean them */
587 cfs_proc_dir_entry_t *srv_procroot;
588 struct lprocfs_stats *srv_stats;
590 /* List of free reply_states */
591 struct list_head srv_free_rs_list;
592 /* waitq to run, when adding stuff to srv_free_rs_list */
593 cfs_waitq_t srv_free_rs_waitq;
596 * if non-NULL called during thread creation (ptlrpc_start_thread())
597 * to initialize service specific per-thread state.
599 int (*srv_init)(struct ptlrpc_thread *thread);
601 * if non-NULL called during thread shutdown (ptlrpc_main()) to
602 * destruct state created by ->srv_init().
604 void (*srv_done)(struct ptlrpc_thread *thread);
606 //struct ptlrpc_srv_ni srv_interfaces[0];
609 /* ptlrpc/events.c */
610 extern lnet_handle_eq_t ptlrpc_eq_h;
611 extern int ptlrpc_uuid_to_peer(struct obd_uuid *uuid,
612 lnet_process_id_t *peer, lnet_nid_t *self);
613 extern void request_out_callback (lnet_event_t *ev);
614 extern void reply_in_callback(lnet_event_t *ev);
615 extern void client_bulk_callback (lnet_event_t *ev);
616 extern void request_in_callback(lnet_event_t *ev);
617 extern void reply_out_callback(lnet_event_t *ev);
618 extern void server_bulk_callback (lnet_event_t *ev);
620 /* ptlrpc/connection.c */
621 void ptlrpc_dump_connections(void);
622 void ptlrpc_readdress_connection(struct ptlrpc_connection *, struct obd_uuid *);
623 struct ptlrpc_connection *ptlrpc_get_connection(lnet_process_id_t peer,
624 lnet_nid_t self, struct obd_uuid *uuid);
625 int ptlrpc_put_connection(struct ptlrpc_connection *c);
626 struct ptlrpc_connection *ptlrpc_connection_addref(struct ptlrpc_connection *);
627 int ptlrpc_init_connection(void);
628 void ptlrpc_cleanup_connection(void);
629 extern lnet_pid_t ptl_get_pid(void);
631 /* ptlrpc/niobuf.c */
632 int ptlrpc_start_bulk_transfer(struct ptlrpc_bulk_desc *desc);
633 void ptlrpc_abort_bulk(struct ptlrpc_bulk_desc *desc);
634 int ptlrpc_register_bulk(struct ptlrpc_request *req);
635 void ptlrpc_unregister_bulk (struct ptlrpc_request *req);
637 static inline int ptlrpc_bulk_active (struct ptlrpc_bulk_desc *desc)
641 spin_lock(&desc->bd_lock);
642 rc = desc->bd_network_rw;
643 spin_unlock(&desc->bd_lock);
647 #define PTLRPC_REPLY_MAYBE_DIFFICULT 0x01
648 #define PTLRPC_REPLY_EARLY 0x02
649 int ptlrpc_send_reply(struct ptlrpc_request *req, int flags);
650 int ptlrpc_reply(struct ptlrpc_request *req);
651 int ptlrpc_error(struct ptlrpc_request *req);
652 void ptlrpc_resend_req(struct ptlrpc_request *request);
653 int ptlrpc_at_get_net_latency(struct ptlrpc_request *req);
654 int ptl_send_rpc(struct ptlrpc_request *request, int noreply);
655 int ptlrpc_register_rqbd (struct ptlrpc_request_buffer_desc *rqbd);
657 /* ptlrpc/client.c */
658 void ptlrpc_init_client(int req_portal, int rep_portal, char *name,
659 struct ptlrpc_client *);
660 void ptlrpc_cleanup_client(struct obd_import *imp);
661 struct ptlrpc_connection *ptlrpc_uuid_to_connection(struct obd_uuid *uuid);
664 ptlrpc_client_recv_or_unlink (struct ptlrpc_request *req)
668 spin_lock(&req->rq_lock);
669 rc = req->rq_receiving_reply || req->rq_must_unlink;
670 spin_unlock(&req->rq_lock);
675 ptlrpc_wake_client_req (struct ptlrpc_request *req)
677 if (req->rq_set == NULL)
678 cfs_waitq_signal(&req->rq_reply_waitq);
680 cfs_waitq_signal(&req->rq_set->set_waitq);
683 int ptlrpc_queue_wait(struct ptlrpc_request *req);
684 int ptlrpc_replay_req(struct ptlrpc_request *req);
685 void ptlrpc_unregister_reply(struct ptlrpc_request *req);
686 void ptlrpc_restart_req(struct ptlrpc_request *req);
687 void ptlrpc_abort_inflight(struct obd_import *imp);
689 struct ptlrpc_request_set *ptlrpc_prep_set(void);
690 int ptlrpc_set_next_timeout(struct ptlrpc_request_set *);
691 int ptlrpc_check_set(struct ptlrpc_request_set *set);
692 int ptlrpc_set_wait(struct ptlrpc_request_set *);
693 int ptlrpc_expired_set(void *data);
694 void ptlrpc_interrupted_set(void *data);
695 void ptlrpc_mark_interrupted(struct ptlrpc_request *req);
696 void ptlrpc_set_destroy(struct ptlrpc_request_set *);
697 void ptlrpc_set_add_req(struct ptlrpc_request_set *, struct ptlrpc_request *);
698 void ptlrpc_set_add_new_req(struct ptlrpc_request_set *,
699 struct ptlrpc_request *);
701 void ptlrpc_free_rq_pool(struct ptlrpc_request_pool *pool);
702 void ptlrpc_add_rqs_to_pool(struct ptlrpc_request_pool *pool, int num_rq);
703 struct ptlrpc_request_pool *ptlrpc_init_rq_pool(int, int,
704 void (*populate_pool)(struct ptlrpc_request_pool *, int));
705 struct ptlrpc_request *ptlrpc_prep_req(struct obd_import *imp, __u32 version,
706 int opcode, int count, int *lengths,
708 struct ptlrpc_request *ptlrpc_prep_req_pool(struct obd_import *imp,
709 __u32 version, int opcode,
710 int count, int *lengths, char **bufs,
711 struct ptlrpc_request_pool *pool);
712 void ptlrpc_free_req(struct ptlrpc_request *request);
713 void ptlrpc_req_finished(struct ptlrpc_request *request);
714 void ptlrpc_req_finished_with_imp_lock(struct ptlrpc_request *request);
715 struct ptlrpc_request *ptlrpc_request_addref(struct ptlrpc_request *req);
716 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_imp (struct ptlrpc_request *req,
717 int npages, int type, int portal);
718 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_exp(struct ptlrpc_request *req,
719 int npages, int type, int portal);
720 void ptlrpc_free_bulk(struct ptlrpc_bulk_desc *bulk);
721 void ptlrpc_prep_bulk_page(struct ptlrpc_bulk_desc *desc,
722 cfs_page_t *page, int pageoffset, int len);
723 void ptlrpc_retain_replayable_request(struct ptlrpc_request *req,
724 struct obd_import *imp);
725 __u64 ptlrpc_next_xid(void);
726 __u64 ptlrpc_sample_next_xid(void);
727 __u64 ptlrpc_req_xid(struct ptlrpc_request *request);
729 /* ptlrpc/service.c */
730 void ptlrpc_save_lock (struct ptlrpc_request *req,
731 struct lustre_handle *lock, int mode);
732 void ptlrpc_commit_replies (struct obd_device *obd);
733 void ptlrpc_schedule_difficult_reply (struct ptlrpc_reply_state *rs);
734 struct ptlrpc_service *ptlrpc_init_svc(int nbufs, int bufsize, int max_req_size,
736 int req_portal, int rep_portal,
738 svc_handler_t, char *name,
739 cfs_proc_dir_entry_t *proc_entry,
741 int min_threads, int max_threads,
743 void ptlrpc_stop_all_threads(struct ptlrpc_service *svc);
745 int ptlrpc_start_threads(struct obd_device *dev, struct ptlrpc_service *svc);
746 int ptlrpc_start_thread(struct obd_device *dev, struct ptlrpc_service *svc);
747 int ptlrpc_unregister_service(struct ptlrpc_service *service);
748 int liblustre_check_services (void *arg);
749 void ptlrpc_daemonize(char *name);
750 int ptlrpc_service_health_check(struct ptlrpc_service *);
753 struct ptlrpc_svc_data {
755 struct ptlrpc_service *svc;
756 struct ptlrpc_thread *thread;
757 struct obd_device *dev;
760 /* ptlrpc/import.c */
761 int ptlrpc_connect_import(struct obd_import *imp, char * new_uuid);
762 int ptlrpc_init_import(struct obd_import *imp);
763 int ptlrpc_disconnect_import(struct obd_import *imp, int noclose);
764 int ptlrpc_import_recovery_state_machine(struct obd_import *imp);
765 void ptlrpc_import_setasync(struct obd_import *imp, int count);
767 /* ptlrpc/pack_generic.c */
768 int lustre_msg_swabbed(struct lustre_msg *msg);
769 int lustre_msg_check_version(struct lustre_msg *msg, __u32 version);
770 int lustre_pack_request(struct ptlrpc_request *, __u32 magic, int count,
771 int *lens, char **bufs);
772 int lustre_pack_reply(struct ptlrpc_request *, int count, int *lens,
774 #define LPRFL_EARLY_REPLY 1
775 int lustre_pack_reply_flags(struct ptlrpc_request *, int count, int *lens,
776 char **bufs, int flags);
777 void lustre_shrink_reply(struct ptlrpc_request *req, int segment,
778 unsigned int newlen, int move_data);
779 void lustre_free_reply_state(struct ptlrpc_reply_state *rs);
780 int lustre_msg_size(__u32 magic, int count, int *lengths);
781 int lustre_msg_early_size(void);
782 int lustre_unpack_msg(struct lustre_msg *m, int len);
783 void *lustre_msg_buf(struct lustre_msg *m, int n, int minlen);
784 int lustre_msg_buflen(struct lustre_msg *m, int n);
785 void lustre_msg_set_buflen(struct lustre_msg *m, int n, int len);
786 int lustre_msg_bufcount(struct lustre_msg *m);
787 char *lustre_msg_string (struct lustre_msg *m, int n, int max_len);
788 void *lustre_swab_buf(struct lustre_msg *, int n, int minlen, void *swabber);
789 void *lustre_swab_reqbuf(struct ptlrpc_request *req, int n, int minlen,
791 void *lustre_swab_repbuf(struct ptlrpc_request *req, int n, int minlen,
793 __u32 lustre_msg_get_flags(struct lustre_msg *msg);
794 void lustre_msg_add_flags(struct lustre_msg *msg, int flags);
795 void lustre_msg_set_flags(struct lustre_msg *msg, int flags);
796 void lustre_msg_clear_flags(struct lustre_msg *msg, int flags);
797 __u32 lustre_msg_get_op_flags(struct lustre_msg *msg);
798 void lustre_msg_add_op_flags(struct lustre_msg *msg, int flags);
799 void lustre_msg_set_op_flags(struct lustre_msg *msg, int flags);
800 struct lustre_handle *lustre_msg_get_handle(struct lustre_msg *msg);
801 __u32 lustre_msg_get_type(struct lustre_msg *msg);
802 __u32 lustre_msg_get_version(struct lustre_msg *msg);
803 void lustre_msg_add_version(struct lustre_msg *msg, int version);
804 __u32 lustre_msg_get_opc(struct lustre_msg *msg);
805 __u64 lustre_msg_get_last_xid(struct lustre_msg *msg);
806 __u64 lustre_msg_get_last_committed(struct lustre_msg *msg);
807 __u64 lustre_msg_get_transno(struct lustre_msg *msg);
808 __u64 lustre_msg_get_slv(struct lustre_msg *msg);
809 __u32 lustre_msg_get_limit(struct lustre_msg *msg);
810 void lustre_msg_set_slv(struct lustre_msg *msg, __u64 slv);
811 void lustre_msg_set_limit(struct lustre_msg *msg, __u64 limit);
812 int lustre_msg_get_status(struct lustre_msg *msg);
813 __u32 lustre_msg_get_conn_cnt(struct lustre_msg *msg);
814 int lustre_msg_is_v1(struct lustre_msg *msg);
815 __u32 lustre_msg_get_magic(struct lustre_msg *msg);
816 __u32 lustre_msg_get_timeout(struct lustre_msg *msg);
817 __u32 lustre_msg_get_service_time(struct lustre_msg *msg);
818 __u32 lustre_msg_get_cksum(struct lustre_msg *msg);
819 __u32 lustre_msg_calc_cksum(struct lustre_msg *msg);
820 void lustre_msg_set_handle(struct lustre_msg *msg,struct lustre_handle *handle);
821 void lustre_msg_set_type(struct lustre_msg *msg, __u32 type);
822 void lustre_msg_set_opc(struct lustre_msg *msg, __u32 opc);
823 void lustre_msg_set_last_xid(struct lustre_msg *msg, __u64 last_xid);
824 void lustre_msg_set_last_committed(struct lustre_msg *msg,__u64 last_committed);
825 void lustre_msg_set_transno(struct lustre_msg *msg, __u64 transno);
826 void lustre_msg_set_status(struct lustre_msg *msg, __u32 status);
827 void lustre_msg_set_conn_cnt(struct lustre_msg *msg, __u32 conn_cnt);
828 void lustre_msg_set_timeout(struct lustre_msg *msg, __u32 timeout);
829 void lustre_msg_set_service_time(struct lustre_msg *msg, __u32 service_time);
830 void lustre_msg_set_cksum(struct lustre_msg *msg, __u32 cksum);
833 ptlrpc_rs_addref(struct ptlrpc_reply_state *rs)
835 LASSERT(atomic_read(&rs->rs_refcount) > 0);
836 atomic_inc(&rs->rs_refcount);
840 ptlrpc_rs_decref(struct ptlrpc_reply_state *rs)
842 LASSERT(atomic_read(&rs->rs_refcount) > 0);
843 if (atomic_dec_and_test(&rs->rs_refcount))
844 lustre_free_reply_state(rs);
847 /* Should only be called once per req */
848 static inline void ptlrpc_req_drop_rs(struct ptlrpc_request *req)
850 if (req->rq_reply_state == NULL)
851 return; /* shouldn't occur */
852 ptlrpc_rs_decref(req->rq_reply_state);
853 req->rq_reply_state = NULL;
854 req->rq_repmsg = NULL;
857 static inline __u32 lustre_request_magic(struct ptlrpc_request *req)
859 return lustre_msg_get_magic(req->rq_reqmsg);
862 static inline int ptlrpc_req_get_repsize(struct ptlrpc_request *req)
864 switch (req->rq_reqmsg->lm_magic) {
865 case LUSTRE_MSG_MAGIC_V1:
866 CERROR("function not supported for lustre_msg V1!\n");
868 case LUSTRE_MSG_MAGIC_V2:
869 return req->rq_reqmsg->lm_repsize;
871 LASSERTF(0, "incorrect message magic: %08x\n",
872 req->rq_reqmsg->lm_magic);
878 ptlrpc_req_set_repsize(struct ptlrpc_request *req, int count, int *lens)
880 int size = lustre_msg_size(req->rq_reqmsg->lm_magic, count, lens);
882 req->rq_replen = size + lustre_msg_early_size();
883 if (req->rq_reqmsg->lm_magic == LUSTRE_MSG_MAGIC_V2)
884 req->rq_reqmsg->lm_repsize = size;
887 /* ldlm/ldlm_lib.c */
888 int client_obd_setup(struct obd_device *obddev, obd_count len, void *buf);
889 int client_obd_cleanup(struct obd_device * obddev);
890 int client_connect_import(struct lustre_handle *conn, struct obd_device *obd,
891 struct obd_uuid *cluuid, struct obd_connect_data *);
892 int client_disconnect_export(struct obd_export *exp);
893 int client_import_add_conn(struct obd_import *imp, struct obd_uuid *uuid,
895 int client_import_del_conn(struct obd_import *imp, struct obd_uuid *uuid);
896 int import_set_conn_priority(struct obd_import *imp, struct obd_uuid *uuid);
898 /* ptlrpc/pinger.c */
899 int ptlrpc_pinger_add_import(struct obd_import *imp);
900 int ptlrpc_pinger_del_import(struct obd_import *imp);
902 void ping_evictor_start(void);
903 void ping_evictor_stop(void);
905 #define ping_evictor_start() do {} while (0)
906 #define ping_evictor_stop() do {} while (0)
909 /* ptlrpc/ptlrpcd.c */
910 void ptlrpcd_wake(struct ptlrpc_request *req);
911 void ptlrpcd_add_req(struct ptlrpc_request *req);
912 int ptlrpcd_addref(void);
913 void ptlrpcd_decref(void);
915 /* ptlrpc/lproc_ptlrpc.c */
916 const char* ll_opcode2str(__u32 opcode);
918 void ptlrpc_lprocfs_register_obd(struct obd_device *obd);
919 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd);
920 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int opc, int bytes);
922 static inline void ptlrpc_lprocfs_register_obd(struct obd_device *obd) {}
923 static inline void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd) {}
924 static inline void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int opc,
928 /* ptlrpc/llog_server.c */
929 int llog_origin_handle_create(struct ptlrpc_request *req);
930 int llog_origin_handle_destroy(struct ptlrpc_request *req);
931 int llog_origin_handle_prev_block(struct ptlrpc_request *req);
932 int llog_origin_handle_next_block(struct ptlrpc_request *req);
933 int llog_origin_handle_read_header(struct ptlrpc_request *req);
934 int llog_origin_handle_close(struct ptlrpc_request *req);
935 int llog_origin_handle_cancel(struct ptlrpc_request *req);
936 int llog_catinfo(struct ptlrpc_request *req);
938 /* ptlrpc/llog_client.c */
939 extern struct llog_operations llog_client_ops;