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_sec.h>
42 #include <lustre_import.h>
43 #include <lprocfs_status.h>
44 #include <lu_object.h>
46 /* MD flags we _always_ use */
47 #define PTLRPC_MD_OPTIONS 0
49 /* Define maxima for bulk I/O
50 * CAVEAT EMPTOR, with multinet (i.e. routers forwarding between networks)
51 * these limits are system wide and not interface-local. */
52 #define PTLRPC_MAX_BRW_BITS LNET_MTU_BITS
53 #define PTLRPC_MAX_BRW_SIZE (1<<LNET_MTU_BITS)
54 #define PTLRPC_MAX_BRW_PAGES (PTLRPC_MAX_BRW_SIZE >> CFS_PAGE_SHIFT)
56 /* When PAGE_SIZE is a constant, we can check our arithmetic here with cpp! */
58 # if ((PTLRPC_MAX_BRW_PAGES & (PTLRPC_MAX_BRW_PAGES - 1)) != 0)
59 # error "PTLRPC_MAX_BRW_PAGES isn't a power of two"
61 # if (PTLRPC_MAX_BRW_SIZE != (PTLRPC_MAX_BRW_PAGES * CFS_PAGE_SIZE))
62 # error "PTLRPC_MAX_BRW_SIZE isn't PTLRPC_MAX_BRW_PAGES * CFS_PAGE_SIZE"
64 # if (PTLRPC_MAX_BRW_SIZE > LNET_MTU)
65 # error "PTLRPC_MAX_BRW_SIZE too big"
67 # if (PTLRPC_MAX_BRW_PAGES > LNET_MAX_IOV)
68 # error "PTLRPC_MAX_BRW_PAGES too big"
70 #endif /* __KERNEL__ */
72 /* Size over which to OBD_VMALLOC() rather than OBD_ALLOC() service request
74 #define SVC_BUF_VMALLOC_THRESHOLD (2 * CFS_PAGE_SIZE)
76 /* The following constants determine how memory is used to buffer incoming
79 * ?_NBUFS # buffers to allocate when growing the pool
80 * ?_BUFSIZE # bytes in a single request buffer
81 * ?_MAXREQSIZE # maximum request service will receive
83 * When fewer than ?_NBUFS/2 buffers are posted for receive, another chunk
84 * of ?_NBUFS is added to the pool.
86 * Messages larger than ?_MAXREQSIZE are dropped. Request buffers are
87 * considered full when less than ?_MAXREQSIZE is left in them.
90 #define LDLM_THREADS_AUTO_MIN min((int)(smp_num_cpus * smp_num_cpus * 2), 8)
91 #define LDLM_THREADS_AUTO_MAX (LDLM_THREADS_AUTO_MIN * 16)
92 #define LDLM_BL_THREADS LDLM_THREADS_AUTO_MIN
93 #define LDLM_NBUFS (64 * smp_num_cpus)
94 #define LDLM_BUFSIZE (8 * 1024)
95 #define LDLM_MAXREQSIZE (5 * 1024)
96 #define LDLM_MAXREPSIZE (1024)
98 #define MDT_MIN_THREADS 2UL
99 #define MDT_MAX_THREADS 512UL
100 #define MDT_NUM_THREADS max(min_t(unsigned long, MDT_MAX_THREADS, \
101 num_physpages >> (25 - PAGE_SHIFT)), 2UL)
102 #define FLD_NUM_THREADS max(min_t(unsigned long, MDT_MAX_THREADS, \
103 num_physpages >> (25 - PAGE_SHIFT)), 2UL)
104 #define SEQ_NUM_THREADS max(min_t(unsigned long, MDT_MAX_THREADS, \
105 num_physpages >> (25 - PAGE_SHIFT)), 2UL)
107 /* Absolute limits */
108 #define MDS_THREADS_MIN 2
109 #define MDS_THREADS_MAX 512
110 #define MDS_THREADS_MIN_READPAGE 2
111 #define MDS_NBUFS (64 * smp_num_cpus)
112 #define MDS_BUFSIZE (8 * 1024)
113 /* Assume file name length = FNAME_MAX = 256 (true for ext3).
114 * path name length = PATH_MAX = 4096
115 * LOV MD size max = EA_MAX = 4000
116 * symlink: FNAME_MAX + PATH_MAX <- largest
117 * link: FNAME_MAX + PATH_MAX (mds_rec_link < mds_rec_create)
118 * rename: FNAME_MAX + FNAME_MAX
119 * open: FNAME_MAX + EA_MAX
121 * MDS_MAXREQSIZE ~= 4736 bytes =
122 * lustre_msg + ldlm_request + mds_body + mds_rec_create + FNAME_MAX + PATH_MAX
123 * MDS_MAXREPSIZE ~= 8300 bytes = lustre_msg + llog_header
124 * or, for mds_close() and mds_reint_unlink() on a many-OST filesystem:
125 * = 9210 bytes = lustre_msg + mds_body + 160 * (easize + cookiesize)
127 * Realistic size is about 512 bytes (20 character name + 128 char symlink),
128 * except in the open case where there are a large number of OSTs in a LOV.
130 #define MDS_MAXREQSIZE (5 * 1024)
131 #define MDS_MAXREPSIZE max(9 * 1024, 280 + LOV_MAX_STRIPE_COUNT * 56)
133 /* FLD_MAXREQSIZE == lustre_msg + __u32 padding + ptlrpc_body + opc + md_fld */
134 #define FLD_MAXREQSIZE (160)
136 /* FLD_MAXREPSIZE == lustre_msg + ptlrpc_body + md_fld */
137 #define FLD_MAXREPSIZE (152)
139 /* SEQ_MAXREQSIZE == lustre_msg + __u32 padding + ptlrpc_body + opc + lu_range +
141 #define SEQ_MAXREQSIZE (160)
143 /* SEQ_MAXREPSIZE == lustre_msg + ptlrpc_body + lu_range */
144 #define SEQ_MAXREPSIZE (152)
146 #define MGS_THREADS_AUTO_MIN 2
147 #define MGS_THREADS_AUTO_MAX 32
148 #define MGS_NBUFS (64 * smp_num_cpus)
149 #define MGS_BUFSIZE (8 * 1024)
150 #define MGS_MAXREQSIZE (7 * 1024)
151 #define MGS_MAXREPSIZE (9 * 1024)
153 /* Absolute limits */
154 #define OSS_THREADS_MIN 2
155 #define OSS_THREADS_MAX 512
156 #define OST_NBUFS (64 * smp_num_cpus)
157 #define OST_BUFSIZE (8 * 1024)
158 /* OST_MAXREQSIZE ~= 4768 bytes =
159 * lustre_msg + obdo + 16 * obd_ioobj + 256 * niobuf_remote
161 * - single object with 16 pages is 512 bytes
162 * - OST_MAXREQSIZE must be at least 1 page of cookies plus some spillover
164 #define OST_MAXREQSIZE (5 * 1024)
165 #define OST_MAXREPSIZE (9 * 1024)
167 struct ptlrpc_connection {
168 struct list_head c_link;
170 lnet_process_id_t c_peer;
171 struct obd_uuid c_remote_uuid;
175 struct ptlrpc_client {
176 __u32 cli_request_portal;
177 __u32 cli_reply_portal;
181 /* state flags of requests */
182 /* XXX only ones left are those used by the bulk descs as well! */
183 #define PTL_RPC_FL_INTR (1 << 0) /* reply wait was interrupted by user */
184 #define PTL_RPC_FL_TIMEOUT (1 << 7) /* request timed out waiting for reply */
186 #define REQ_MAX_ACK_LOCKS 8
188 union ptlrpc_async_args {
189 /* Scratchpad for passing args to completion interpreter. Users
190 * cast to the struct of their choosing, and LASSERT that this is
191 * big enough. For _tons_ of context, OBD_ALLOC a struct and store
192 * a pointer to it here. The pointer_arg ensures this struct is at
193 * least big enough for that. */
194 void *pointer_arg[9];
198 struct ptlrpc_request_set;
199 typedef int (*set_interpreter_func)(struct ptlrpc_request_set *, void *, int);
201 struct ptlrpc_request_set {
202 int set_remaining; /* # uncompleted requests */
203 cfs_waitq_t set_waitq;
204 cfs_waitq_t *set_wakeup_ptr;
205 struct list_head set_requests;
206 set_interpreter_func set_interpret; /* completion callback */
207 void *set_arg; /* completion context */
208 void *set_countp; /* pointer to NOB counter in case
209 * of directIO (bug11737) */
210 /* locked so that any old caller can communicate requests to
211 * the set holder who can then fold them into the lock-free set */
212 spinlock_t set_new_req_lock;
213 struct list_head set_new_requests;
216 struct ptlrpc_bulk_desc;
219 * ptlrpc callback & work item stuff
221 struct ptlrpc_cb_id {
222 void (*cbid_fn)(lnet_event_t *ev); /* specific callback fn */
223 void *cbid_arg; /* additional arg */
226 #define RS_MAX_LOCKS 4
229 struct ptlrpc_reply_state {
230 struct ptlrpc_cb_id rs_cb_id;
231 struct list_head rs_list;
232 struct list_head rs_exp_list;
233 struct list_head rs_obd_list;
235 struct list_head rs_debug_list;
237 /* updates to following flag serialised by srv_request_lock */
238 unsigned int rs_difficult:1; /* ACK/commit stuff */
239 unsigned int rs_scheduled:1; /* being handled? */
240 unsigned int rs_scheduled_ever:1;/* any schedule attempts? */
241 unsigned int rs_handled:1; /* been handled yet? */
242 unsigned int rs_on_net:1; /* reply_out_callback pending? */
243 unsigned int rs_prealloc:1; /* rs from prealloc list */
248 struct obd_export *rs_export;
249 struct ptlrpc_service *rs_service;
250 lnet_handle_md_t rs_md_h;
251 atomic_t rs_refcount;
253 struct ptlrpc_svc_ctx *rs_svc_ctx;
254 struct lustre_msg *rs_repbuf; /* wrapper */
255 int rs_repbuf_len; /* wrapper buf length */
256 int rs_repdata_len; /* wrapper msg length */
257 struct lustre_msg *rs_msg; /* reply message */
259 /* locks awaiting client reply ACK */
261 struct lustre_handle rs_locks[RS_MAX_LOCKS];
262 ldlm_mode_t rs_modes[RS_MAX_LOCKS];
265 struct ptlrpc_thread;
268 RQ_PHASE_NEW = 0xebc0de00,
269 RQ_PHASE_RPC = 0xebc0de01,
270 RQ_PHASE_BULK = 0xebc0de02,
271 RQ_PHASE_INTERPRET = 0xebc0de03,
272 RQ_PHASE_COMPLETE = 0xebc0de04,
275 struct ptlrpc_request_pool {
277 struct list_head prp_req_list; /* list of ptlrpc_request structs */
279 void (*prp_populate)(struct ptlrpc_request_pool *, int);
285 struct ptlrpc_request {
286 int rq_type; /* one of PTL_RPC_MSG_* */
287 struct list_head rq_list;
288 struct list_head rq_history_list; /* server-side history */
289 __u64 rq_history_seq; /* history sequence # */
292 /* client-side flags */
293 unsigned int rq_intr:1, rq_replied:1, rq_err:1,
294 rq_timedout:1, rq_resend:1, rq_restart:1,
296 * when ->rq_replay is set, request is kept by the client even
297 * after server commits corresponding transaction. This is
298 * used for operations that require sequence of multiple
299 * requests to be replayed. The only example currently is file
300 * open/close. When last request in such a sequence is
301 * committed, ->rq_replay is cleared on all requests in the
305 rq_no_resend:1, rq_waiting:1, rq_receiving_reply:1,
306 rq_no_delay:1, rq_net_err:1, rq_wait_ctx:1;
307 enum rq_phase rq_phase; /* one of RQ_PHASE_* */
308 atomic_t rq_refcount; /* client-side refcount for SENT race */
310 struct ptlrpc_thread *rq_svc_thread; /* initial thread servicing req */
312 int rq_request_portal; /* XXX FIXME bug 249 */
313 int rq_reply_portal; /* XXX FIXME bug 249 */
315 int rq_nob_received; /* client-side # reply bytes actually received */
318 struct lustre_msg *rq_reqmsg;
320 int rq_timeout; /* time to wait for reply (seconds) */
322 struct lustre_msg *rq_repmsg;
325 struct list_head rq_replay_list;
327 struct ptlrpc_cli_ctx *rq_cli_ctx; /* client's half ctx */
328 struct ptlrpc_svc_ctx *rq_svc_ctx; /* server's half ctx */
329 struct list_head rq_ctx_chain; /* link to waited ctx */
330 ptlrpc_sec_flavor_t rq_sec_flavor; /* client & server */
331 /* client security flags */
332 unsigned int rq_ctx_init:1, /* context initiation */
333 rq_ctx_fini:1, /* context destroy */
334 rq_bulk_read:1, /* request bulk read */
335 rq_bulk_write:1, /* request bulk write */
336 /* server authentication flags */
337 rq_auth_gss:1, /* authenticated by gss */
338 rq_auth_remote:1, /* authed as remote user */
339 rq_auth_usr_root:1, /* authed as root */
340 rq_auth_usr_mdt:1; /* authed as mdt */
342 uid_t rq_auth_uid; /* authed uid */
343 uid_t rq_auth_mapped_uid; /* authed uid mapped to */
345 /* (server side), pointed directly into req buffer */
346 struct ptlrpc_user_desc *rq_user_desc;
348 /* various buffer pointers */
349 struct lustre_msg *rq_reqbuf; /* req wrapper */
350 int rq_reqbuf_len; /* req wrapper buf len */
351 int rq_reqdata_len; /* req wrapper msg len */
352 struct lustre_msg *rq_repbuf; /* rep wrapper */
353 int rq_repbuf_len; /* rep wrapper buf len */
354 int rq_repdata_len; /* rep wrapper msg len */
355 struct lustre_msg *rq_clrbuf; /* only in priv mode */
356 int rq_clrbuf_len; /* only in priv mode */
357 int rq_clrdata_len; /* only in priv mode */
359 __u32 rq_req_swab_mask;
360 __u32 rq_rep_swab_mask;
362 int rq_import_generation;
363 enum lustre_imp_state rq_send_state;
365 /* client+server request */
366 lnet_handle_md_t rq_req_md_h;
367 struct ptlrpc_cb_id rq_req_cbid;
370 struct timeval rq_arrival_time; /* request arrival time */
371 struct ptlrpc_reply_state *rq_reply_state; /* separated reply state */
372 struct ptlrpc_request_buffer_desc *rq_rqbd; /* incoming request buffer*/
374 __u32 rq_uid; /* peer uid, used in MDS only */
377 /* client-only incoming reply */
378 lnet_handle_md_t rq_reply_md_h;
379 cfs_waitq_t rq_reply_waitq;
380 struct ptlrpc_cb_id rq_reply_cbid;
383 lnet_process_id_t rq_peer;
384 struct obd_export *rq_export;
385 struct obd_import *rq_import;
387 void (*rq_replay_cb)(struct ptlrpc_request *);
388 void (*rq_commit_cb)(struct ptlrpc_request *);
391 struct ptlrpc_bulk_desc *rq_bulk; /* client side bulk */
392 time_t rq_sent; /* when request sent, seconds */
395 struct list_head rq_set_chain;
396 struct ptlrpc_request_set *rq_set;
397 void *rq_interpret_reply; /* Async completion handler */
398 union ptlrpc_async_args rq_async_args; /* Async completion context */
399 void *rq_ptlrpcd_data;
400 struct ptlrpc_request_pool *rq_pool; /* Pool if request from
402 struct lu_context rq_session;
405 static inline void lustre_set_req_swabbed(struct ptlrpc_request *req, int index)
407 LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
408 LASSERT((req->rq_req_swab_mask & (1 << index)) == 0);
409 req->rq_req_swab_mask |= 1 << index;
412 static inline void lustre_set_rep_swabbed(struct ptlrpc_request *req, int index)
414 LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
415 LASSERT((req->rq_rep_swab_mask & (1 << index)) == 0);
416 req->rq_rep_swab_mask |= 1 << index;
419 static inline int lustre_req_swabbed(struct ptlrpc_request *req, int index)
421 LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
422 return req->rq_req_swab_mask & (1 << index);
425 static inline int lustre_rep_swabbed(struct ptlrpc_request *req, int index)
427 LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
428 return req->rq_rep_swab_mask & (1 << index);
431 #define SWAB_PARANOIA 1
434 /* unpacking: assert idx not unpacked already */
435 #define LASSERT_REQSWAB(rq, idx) lustre_set_req_swabbed(rq, idx)
436 #define LASSERT_REPSWAB(rq, idx) lustre_set_rep_swabbed(rq, idx)
438 /* just looking: assert idx already unpacked */
439 #define LASSERT_REQSWABBED(rq, idx) LASSERT(lustre_req_swabbed(rq, idx))
440 #define LASSERT_REPSWABBED(rq, idx) LASSERT(lustre_rep_swabbed(rq, idx))
443 #define LASSERT_REQSWAB(rq, idx)
444 #define LASSERT_REPSWAB(rq, idx)
445 #define LASSERT_REQSWABBED(rq, idx)
446 #define LASSERT_REPSWABBED(rq, idx)
449 static inline const char *
450 ptlrpc_rqphase2str(const struct ptlrpc_request *req)
452 switch (req->rq_phase) {
459 case RQ_PHASE_INTERPRET:
461 case RQ_PHASE_COMPLETE:
468 /* Spare the preprocessor, spoil the bugs. */
469 #define FLAG(field, str) (field ? str : "")
471 #define DEBUG_REQ_FLAGS(req) \
472 ptlrpc_rqphase2str(req), \
473 FLAG(req->rq_intr, "I"), FLAG(req->rq_replied, "R"), \
474 FLAG(req->rq_err, "E"), \
475 FLAG(req->rq_timedout, "X") /* eXpired */, FLAG(req->rq_resend, "S"), \
476 FLAG(req->rq_restart, "T"), FLAG(req->rq_replay, "P"), \
477 FLAG(req->rq_no_resend, "N"), \
478 FLAG(req->rq_waiting, "W"), \
479 FLAG(req->rq_wait_ctx, "C")
481 #define REQ_FLAGS_FMT "%s:%s%s%s%s%s%s%s%s%s%s"
483 void _debug_req(struct ptlrpc_request *req, __u32 mask,
484 struct libcfs_debug_msg_data *data, const char *fmt, ...)
485 __attribute__ ((format (printf, 4, 5)));
487 #define debug_req(cdls, level, req, file, func, line, fmt, a...) \
491 if (((level) & D_CANTMASK) != 0 || \
492 ((libcfs_debug & (level)) != 0 && \
493 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) { \
494 static struct libcfs_debug_msg_data _req_dbg_data = \
495 DEBUG_MSG_DATA_INIT(cdls, DEBUG_SUBSYSTEM, file, func, line); \
496 _debug_req((req), (level), &_req_dbg_data, fmt, ##a); \
500 /* for most callers (level is a constant) this is resolved at compile time */
501 #define DEBUG_REQ(level, req, fmt, args...) \
503 if ((level) & (D_ERROR | D_WARNING)) { \
504 static cfs_debug_limit_state_t cdls; \
505 debug_req(&cdls, level, req, __FILE__, __func__, __LINE__, \
506 "@@@ "fmt" ", ## args); \
508 debug_req(NULL, level, req, __FILE__, __func__, __LINE__, \
509 "@@@ "fmt" ", ## args); \
512 struct ptlrpc_bulk_page {
513 struct list_head bp_link;
515 int bp_pageoffset; /* offset within a page */
516 struct page *bp_page;
519 #define BULK_GET_SOURCE 0
520 #define BULK_PUT_SINK 1
521 #define BULK_GET_SINK 2
522 #define BULK_PUT_SOURCE 3
524 struct ptlrpc_bulk_desc {
525 unsigned int bd_success:1; /* completed successfully */
526 unsigned int bd_network_rw:1; /* accessible to the network */
527 unsigned int bd_type:2; /* {put,get}{source,sink} */
528 unsigned int bd_registered:1; /* client side */
529 spinlock_t bd_lock; /* serialise with callback */
530 int bd_import_generation;
531 struct obd_export *bd_export;
532 struct obd_import *bd_import;
534 struct ptlrpc_request *bd_req; /* associated request */
535 cfs_waitq_t bd_waitq; /* server side only WQ */
536 int bd_iov_count; /* # entries in bd_iov */
537 int bd_max_iov; /* allocated size of bd_iov */
538 int bd_nob; /* # bytes covered */
539 int bd_nob_transferred; /* # bytes GOT/PUT */
543 struct ptlrpc_cb_id bd_cbid; /* network callback info */
544 lnet_handle_md_t bd_md_h; /* associated MD */
545 lnet_nid_t bd_sender; /* stash event::sender */
547 cfs_page_t **bd_enc_pages;
548 #if defined(__KERNEL__)
549 lnet_kiov_t bd_iov[0];
551 lnet_md_iovec_t bd_iov[0];
555 struct ptlrpc_thread {
557 struct list_head t_link; /* active threads for service, from svc->srv_threads */
559 void *t_data; /* thread-private data (preallocated memory) */
562 unsigned int t_id; /* service thread index, from ptlrpc_start_threads */
563 cfs_waitq_t t_ctl_waitq;
564 struct lu_env *t_env;
567 struct ptlrpc_request_buffer_desc {
568 struct list_head rqbd_list;
569 struct list_head rqbd_reqs;
570 struct ptlrpc_service *rqbd_service;
571 lnet_handle_md_t rqbd_md_h;
574 struct ptlrpc_cb_id rqbd_cbid;
575 struct ptlrpc_request rqbd_req;
578 typedef int (*svc_handler_t)(struct ptlrpc_request *req);
579 typedef void (*svcreq_printfn_t)(void *, struct ptlrpc_request *);
581 struct ptlrpc_service {
582 struct list_head srv_list; /* chain thru all services */
583 int srv_max_req_size; /* biggest request to receive */
584 int srv_max_reply_size; /* biggest reply to send */
585 int srv_buf_size; /* size of individual buffers */
586 int srv_nbuf_per_group; /* # buffers to allocate in 1 group */
587 int srv_nbufs; /* total # req buffer descs allocated */
588 int srv_threads_min; /* threads to start at SOW */
589 int srv_threads_max; /* thread upper limit */
590 int srv_threads_started; /* index of last started thread */
591 int srv_threads_running; /* # running threads */
592 int srv_n_difficult_replies; /* # 'difficult' replies */
593 int srv_n_active_reqs; /* # reqs being served */
594 cfs_duration_t srv_rqbd_timeout; /* timeout before re-posting reqs, in tick */
595 int srv_watchdog_timeout; /* soft watchdog timeout, in ms */
596 unsigned srv_cpu_affinity:1; /* bind threads to CPUs */
598 __u32 srv_req_portal;
599 __u32 srv_rep_portal;
601 int srv_n_queued_reqs; /* # reqs waiting to be served */
602 struct list_head srv_request_queue; /* reqs waiting for service */
604 struct list_head srv_request_history; /* request history */
605 __u64 srv_request_seq; /* next request sequence # */
606 __u64 srv_request_max_cull_seq; /* highest seq culled from history */
607 svcreq_printfn_t srv_request_history_print_fn; /* service-specific print fn */
609 struct list_head srv_idle_rqbds; /* request buffers to be reposted */
610 struct list_head srv_active_rqbds; /* req buffers receiving */
611 struct list_head srv_history_rqbds; /* request buffer history */
612 int srv_nrqbd_receiving; /* # posted request buffers */
613 int srv_n_history_rqbds; /* # request buffers in history */
614 int srv_max_history_rqbds;/* max # request buffers in history */
616 atomic_t srv_outstanding_replies;
617 struct list_head srv_active_replies; /* all the active replies */
618 struct list_head srv_reply_queue; /* replies waiting for service */
620 cfs_waitq_t srv_waitq; /* all threads sleep on this. This
621 * wait-queue is signalled when new
622 * incoming request arrives and when
623 * difficult reply has to be handled. */
625 struct list_head srv_threads; /* service thread list */
626 svc_handler_t srv_handler;
628 char *srv_name; /* only statically allocated strings here; we don't clean them */
629 char *srv_thread_name; /* only statically allocated strings here; we don't clean them */
633 cfs_proc_dir_entry_t *srv_procroot;
634 struct lprocfs_stats *srv_stats;
636 /* List of free reply_states */
637 struct list_head srv_free_rs_list;
638 /* waitq to run, when adding stuff to srv_free_rs_list */
639 cfs_waitq_t srv_free_rs_waitq;
642 * Tags for lu_context associated with this thread, see struct
647 * if non-NULL called during thread creation (ptlrpc_start_thread())
648 * to initialize service specific per-thread state.
650 int (*srv_init)(struct ptlrpc_thread *thread);
652 * if non-NULL called during thread shutdown (ptlrpc_main()) to
653 * destruct state created by ->srv_init().
655 void (*srv_done)(struct ptlrpc_thread *thread);
657 //struct ptlrpc_srv_ni srv_interfaces[0];
660 /* ptlrpc/events.c */
661 extern lnet_handle_eq_t ptlrpc_eq_h;
662 extern int ptlrpc_uuid_to_peer(struct obd_uuid *uuid,
663 lnet_process_id_t *peer, lnet_nid_t *self);
664 extern void request_out_callback (lnet_event_t *ev);
665 extern void reply_in_callback(lnet_event_t *ev);
666 extern void client_bulk_callback (lnet_event_t *ev);
667 extern void request_in_callback(lnet_event_t *ev);
668 extern void reply_out_callback(lnet_event_t *ev);
669 extern void server_bulk_callback (lnet_event_t *ev);
671 /* ptlrpc/connection.c */
672 void ptlrpc_dump_connections(void);
673 void ptlrpc_readdress_connection(struct ptlrpc_connection *, struct obd_uuid *);
674 struct ptlrpc_connection *ptlrpc_get_connection(lnet_process_id_t peer,
675 lnet_nid_t self, struct obd_uuid *uuid);
676 int ptlrpc_put_connection(struct ptlrpc_connection *c);
677 struct ptlrpc_connection *ptlrpc_connection_addref(struct ptlrpc_connection *);
678 void ptlrpc_init_connection(void);
679 void ptlrpc_cleanup_connection(void);
680 extern lnet_pid_t ptl_get_pid(void);
682 /* ptlrpc/niobuf.c */
683 int ptlrpc_start_bulk_transfer(struct ptlrpc_bulk_desc *desc);
684 void ptlrpc_abort_bulk(struct ptlrpc_bulk_desc *desc);
685 int ptlrpc_register_bulk(struct ptlrpc_request *req);
686 void ptlrpc_unregister_bulk (struct ptlrpc_request *req);
688 static inline int ptlrpc_bulk_active (struct ptlrpc_bulk_desc *desc)
692 spin_lock(&desc->bd_lock);
693 rc = desc->bd_network_rw;
694 spin_unlock(&desc->bd_lock);
698 int ptlrpc_send_reply(struct ptlrpc_request *req, int);
699 int ptlrpc_reply(struct ptlrpc_request *req);
700 int ptlrpc_error(struct ptlrpc_request *req);
701 void ptlrpc_resend_req(struct ptlrpc_request *request);
702 int ptl_send_rpc(struct ptlrpc_request *request, int noreply);
703 int ptlrpc_register_rqbd (struct ptlrpc_request_buffer_desc *rqbd);
705 /* ptlrpc/client.c */
706 void ptlrpc_init_client(int req_portal, int rep_portal, char *name,
707 struct ptlrpc_client *);
708 void ptlrpc_cleanup_client(struct obd_import *imp);
709 struct ptlrpc_connection *ptlrpc_uuid_to_connection(struct obd_uuid *uuid);
712 ptlrpc_client_receiving_reply (struct ptlrpc_request *req)
716 spin_lock(&req->rq_lock);
717 rc = req->rq_receiving_reply;
718 spin_unlock(&req->rq_lock);
723 ptlrpc_client_replied (struct ptlrpc_request *req)
727 spin_lock(&req->rq_lock);
728 rc = req->rq_replied;
729 spin_unlock(&req->rq_lock);
734 ptlrpc_wake_client_req (struct ptlrpc_request *req)
736 if (req->rq_set == NULL)
737 cfs_waitq_signal(&req->rq_reply_waitq);
739 cfs_waitq_signal(&req->rq_set->set_waitq);
742 int ptlrpc_queue_wait(struct ptlrpc_request *req);
743 int ptlrpc_replay_req(struct ptlrpc_request *req);
744 void ptlrpc_unregister_reply(struct ptlrpc_request *req);
745 void ptlrpc_restart_req(struct ptlrpc_request *req);
746 void ptlrpc_abort_inflight(struct obd_import *imp);
748 struct ptlrpc_request_set *ptlrpc_prep_set(void);
749 int ptlrpc_set_next_timeout(struct ptlrpc_request_set *);
750 int ptlrpc_check_set(struct ptlrpc_request_set *set);
751 int ptlrpc_set_wait(struct ptlrpc_request_set *);
752 int ptlrpc_expired_set(void *data);
753 void ptlrpc_interrupted_set(void *data);
754 void ptlrpc_mark_interrupted(struct ptlrpc_request *req);
755 void ptlrpc_set_destroy(struct ptlrpc_request_set *);
756 void ptlrpc_set_add_req(struct ptlrpc_request_set *, struct ptlrpc_request *);
757 void ptlrpc_set_add_new_req(struct ptlrpc_request_set *,
758 struct ptlrpc_request *);
760 void ptlrpc_free_rq_pool(struct ptlrpc_request_pool *pool);
761 void ptlrpc_add_rqs_to_pool(struct ptlrpc_request_pool *pool, int num_rq);
762 struct ptlrpc_request_pool *ptlrpc_init_rq_pool(int, int,
763 void (*populate_pool)(struct ptlrpc_request_pool *, int));
764 struct ptlrpc_request *ptlrpc_prep_req(struct obd_import *imp, __u32 version,
765 int opcode, int count, int *lengths,
767 struct ptlrpc_request *ptlrpc_prep_req_pool(struct obd_import *imp,
768 __u32 version, int opcode,
769 int count, int *lengths, char **bufs,
770 struct ptlrpc_request_pool *pool,
771 struct ptlrpc_cli_ctx *ctx);
772 void ptlrpc_free_req(struct ptlrpc_request *request);
773 void ptlrpc_req_finished(struct ptlrpc_request *request);
774 void ptlrpc_req_finished_with_imp_lock(struct ptlrpc_request *request);
775 struct ptlrpc_request *ptlrpc_request_addref(struct ptlrpc_request *req);
776 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_imp (struct ptlrpc_request *req,
777 int npages, int type, int portal);
778 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_exp(struct ptlrpc_request *req,
779 int npages, int type, int portal);
780 void ptlrpc_free_bulk(struct ptlrpc_bulk_desc *bulk);
781 void ptlrpc_prep_bulk_page(struct ptlrpc_bulk_desc *desc,
782 cfs_page_t *page, int pageoffset, int len);
783 void ptlrpc_retain_replayable_request(struct ptlrpc_request *req,
784 struct obd_import *imp);
785 __u64 ptlrpc_next_xid(void);
786 __u64 ptlrpc_sample_next_xid(void);
787 __u64 ptlrpc_req_xid(struct ptlrpc_request *request);
789 struct ptlrpc_service_conf {
792 int psc_max_req_size;
793 int psc_max_reply_size;
796 int psc_watchdog_timeout; /* in ms */
802 /* ptlrpc/service.c */
803 void ptlrpc_save_lock (struct ptlrpc_request *req,
804 struct lustre_handle *lock, int mode);
805 void ptlrpc_commit_replies (struct obd_device *obd);
806 void ptlrpc_schedule_difficult_reply (struct ptlrpc_reply_state *rs);
807 struct ptlrpc_service *ptlrpc_init_svc_conf(struct ptlrpc_service_conf *c,
808 svc_handler_t h, char *name,
809 struct proc_dir_entry *proc_entry,
810 svcreq_printfn_t prntfn,
813 struct ptlrpc_service *ptlrpc_init_svc(int nbufs, int bufsize, int max_req_size,
815 int req_portal, int rep_portal,
816 int watchdog_timeout, /* in ms */
817 svc_handler_t, char *name,
818 cfs_proc_dir_entry_t *proc_entry,
820 int min_threads, int max_threads,
821 char *threadname, __u32 ctx_tags);
822 void ptlrpc_stop_all_threads(struct ptlrpc_service *svc);
824 int ptlrpc_start_threads(struct obd_device *dev, struct ptlrpc_service *svc);
825 int ptlrpc_start_thread(struct obd_device *dev, struct ptlrpc_service *svc);
826 int ptlrpc_unregister_service(struct ptlrpc_service *service);
827 int liblustre_check_services (void *arg);
828 void ptlrpc_daemonize(char *name);
829 int ptlrpc_service_health_check(struct ptlrpc_service *);
832 struct ptlrpc_svc_data {
834 struct ptlrpc_service *svc;
835 struct ptlrpc_thread *thread;
836 struct obd_device *dev;
839 /* ptlrpc/import.c */
840 int ptlrpc_connect_import(struct obd_import *imp, char * new_uuid);
841 int ptlrpc_init_import(struct obd_import *imp);
842 int ptlrpc_disconnect_import(struct obd_import *imp, int noclose);
843 int ptlrpc_import_recovery_state_machine(struct obd_import *imp);
845 /* ptlrpc/pack_generic.c */
846 int lustre_msg_swabbed(struct lustre_msg *msg);
847 int lustre_msg_check_version(struct lustre_msg *msg, __u32 version);
848 void lustre_init_msg_v2(struct lustre_msg_v2 *msg, int count, int *lens,
850 int lustre_pack_request(struct ptlrpc_request *, __u32 magic, int count,
851 int *lens, char **bufs);
852 int lustre_pack_reply(struct ptlrpc_request *, int count, int *lens,
854 int lustre_pack_reply_v2(struct ptlrpc_request *req, int count,
855 int *lens, char **bufs);
856 int lustre_shrink_msg(struct lustre_msg *msg, int segment,
857 unsigned int newlen, int move_data);
858 void lustre_free_reply_state(struct ptlrpc_reply_state *rs);
859 int lustre_msg_size(__u32 magic, int count, int *lengths);
860 int lustre_msg_size_v2(int count, int *lengths);
861 int lustre_unpack_msg(struct lustre_msg *m, int len);
862 void *lustre_msg_buf_v1(void *msg, int n, int min_size);
863 void *lustre_msg_buf_v2(struct lustre_msg_v2 *m, int n, int min_size);
864 void *lustre_msg_buf(struct lustre_msg *m, int n, int minlen);
865 int lustre_msg_buflen(struct lustre_msg *m, int n);
866 void lustre_msg_set_buflen(struct lustre_msg *m, int n, int len);
867 int lustre_msg_bufcount(struct lustre_msg *m);
868 char *lustre_msg_string (struct lustre_msg *m, int n, int max_len);
869 void *lustre_swab_buf(struct lustre_msg *, int n, int minlen, void *swabber);
870 void *lustre_swab_reqbuf(struct ptlrpc_request *req, int n, int minlen,
872 void *lustre_swab_repbuf(struct ptlrpc_request *req, int n, int minlen,
874 __u32 lustre_msg_get_flags(struct lustre_msg *msg);
875 void lustre_msg_add_flags(struct lustre_msg *msg, int flags);
876 void lustre_msg_set_flags(struct lustre_msg *msg, int flags);
877 void lustre_msg_clear_flags(struct lustre_msg *msg, int flags);
878 __u32 lustre_msg_get_op_flags(struct lustre_msg *msg);
879 void lustre_msg_add_op_flags(struct lustre_msg *msg, int flags);
880 void lustre_msg_set_op_flags(struct lustre_msg *msg, int flags);
881 struct lustre_handle *lustre_msg_get_handle(struct lustre_msg *msg);
882 __u32 lustre_msg_get_type(struct lustre_msg *msg);
883 __u32 lustre_msg_get_version(struct lustre_msg *msg);
884 void lustre_msg_add_version(struct lustre_msg *msg, int version);
885 __u32 lustre_msg_get_opc(struct lustre_msg *msg);
886 __u64 lustre_msg_get_last_xid(struct lustre_msg *msg);
887 __u64 lustre_msg_get_last_committed(struct lustre_msg *msg);
888 __u64 lustre_msg_get_transno(struct lustre_msg *msg);
889 int lustre_msg_get_status(struct lustre_msg *msg);
890 __u32 lustre_msg_get_conn_cnt(struct lustre_msg *msg);
891 __u32 lustre_msg_get_magic(struct lustre_msg *msg);
892 void lustre_msg_set_handle(struct lustre_msg *msg,struct lustre_handle *handle);
893 void lustre_msg_set_type(struct lustre_msg *msg, __u32 type);
894 void lustre_msg_set_opc(struct lustre_msg *msg, __u32 opc);
895 void lustre_msg_set_last_xid(struct lustre_msg *msg, __u64 last_xid);
896 void lustre_msg_set_last_committed(struct lustre_msg *msg,__u64 last_committed);
897 void lustre_msg_set_transno(struct lustre_msg *msg, __u64 transno);
898 void lustre_msg_set_status(struct lustre_msg *msg, __u32 status);
899 void lustre_msg_set_conn_cnt(struct lustre_msg *msg, __u32 conn_cnt);
902 lustre_shrink_reply(struct ptlrpc_request *req, int segment,
903 unsigned int newlen, int move_data)
905 LASSERT(req->rq_reply_state);
906 LASSERT(req->rq_repmsg);
907 req->rq_replen = lustre_shrink_msg(req->rq_repmsg, segment,
912 ptlrpc_rs_addref(struct ptlrpc_reply_state *rs)
914 LASSERT(atomic_read(&rs->rs_refcount) > 0);
915 atomic_inc(&rs->rs_refcount);
919 ptlrpc_rs_decref(struct ptlrpc_reply_state *rs)
921 LASSERT(atomic_read(&rs->rs_refcount) > 0);
922 if (atomic_dec_and_test(&rs->rs_refcount))
923 lustre_free_reply_state(rs);
926 static inline __u32 lustre_request_magic(struct ptlrpc_request *req)
928 return lustre_msg_get_magic(req->rq_reqmsg);
931 static inline int ptlrpc_req_get_repsize(struct ptlrpc_request *req)
933 switch (req->rq_reqmsg->lm_magic) {
934 case LUSTRE_MSG_MAGIC_V1:
935 CERROR("function not supported for lustre_msg V1!\n");
937 case LUSTRE_MSG_MAGIC_V2:
938 return req->rq_reqmsg->lm_repsize;
940 LASSERTF(0, "incorrect message magic: %08x\n",
941 req->rq_reqmsg->lm_magic);
947 ptlrpc_req_set_repsize(struct ptlrpc_request *req, int count, int *lens)
949 req->rq_replen = lustre_msg_size(req->rq_reqmsg->lm_magic, count, lens);
950 if (req->rq_reqmsg->lm_magic == LUSTRE_MSG_MAGIC_V2)
951 req->rq_reqmsg->lm_repsize = req->rq_replen;
954 /* ldlm/ldlm_lib.c */
955 int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg);
956 int client_obd_cleanup(struct obd_device *obddev);
957 int client_connect_import(const struct lu_env *env,
958 struct lustre_handle *conn, struct obd_device *obd,
959 struct obd_uuid *cluuid, struct obd_connect_data *);
960 int client_disconnect_export(struct obd_export *exp);
961 int client_import_add_conn(struct obd_import *imp, struct obd_uuid *uuid,
963 int client_import_del_conn(struct obd_import *imp, struct obd_uuid *uuid);
964 int import_set_conn_priority(struct obd_import *imp, struct obd_uuid *uuid);
966 /* ptlrpc/pinger.c */
967 int ptlrpc_pinger_add_import(struct obd_import *imp);
968 int ptlrpc_pinger_del_import(struct obd_import *imp);
969 cfs_time_t ptlrpc_suspend_wakeup_time(void);
971 void ping_evictor_start(void);
972 void ping_evictor_stop(void);
974 #define ping_evictor_start() do {} while (0)
975 #define ping_evictor_stop() do {} while (0)
977 int ptlrpc_check_and_wait_suspend(struct ptlrpc_request *req);
979 /* ptlrpc/ptlrpcd.c */
980 void ptlrpcd_wake(struct ptlrpc_request *req);
981 void ptlrpcd_add_req(struct ptlrpc_request *req);
982 int ptlrpcd_addref(void);
983 void ptlrpcd_decref(void);
985 /* ptlrpc/lproc_ptlrpc.c */
986 const char* ll_opcode2str(__u32 opcode);
988 void ptlrpc_lprocfs_register_obd(struct obd_device *obd);
989 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd);
990 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int opc, int bytes);
992 static inline void ptlrpc_lprocfs_register_obd(struct obd_device *obd) {}
993 static inline void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd) {}
994 static inline void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int opc,
998 /* ptlrpc/llog_server.c */
999 int llog_origin_handle_create(struct ptlrpc_request *req);
1000 int llog_origin_handle_destroy(struct ptlrpc_request *req);
1001 int llog_origin_handle_prev_block(struct ptlrpc_request *req);
1002 int llog_origin_handle_next_block(struct ptlrpc_request *req);
1003 int llog_origin_handle_read_header(struct ptlrpc_request *req);
1004 int llog_origin_handle_close(struct ptlrpc_request *req);
1005 int llog_origin_handle_cancel(struct ptlrpc_request *req);
1006 int llog_catinfo(struct ptlrpc_request *req);
1008 /* ptlrpc/llog_client.c */
1009 extern struct llog_operations llog_client_ops;