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/dw/setattr. When last request in such a sequence
301 * is committed, ->rq_replay is cleared on all requests in the
305 /* this is the last request in the sequence. */
307 rq_no_resend:1, rq_waiting:1, rq_receiving_reply:1,
308 rq_no_delay:1, rq_net_err:1, rq_wait_ctx:1;
309 enum rq_phase rq_phase; /* one of RQ_PHASE_* */
310 atomic_t rq_refcount; /* client-side refcount for SENT race */
312 struct ptlrpc_thread *rq_svc_thread; /* initial thread servicing req */
314 int rq_request_portal; /* XXX FIXME bug 249 */
315 int rq_reply_portal; /* XXX FIXME bug 249 */
317 int rq_nob_received; /* client-side # reply bytes actually received */
320 struct lustre_msg *rq_reqmsg;
322 int rq_timeout; /* time to wait for reply (seconds) */
324 struct lustre_msg *rq_repmsg;
327 struct list_head rq_replay_list;
328 struct list_head rq_mod_list;
330 struct ptlrpc_cli_ctx *rq_cli_ctx; /* client's half ctx */
331 struct ptlrpc_svc_ctx *rq_svc_ctx; /* server's half ctx */
332 struct list_head rq_ctx_chain; /* link to waited ctx */
333 ptlrpc_sec_flavor_t rq_sec_flavor; /* client & server */
334 /* client security flags */
335 unsigned int rq_ctx_init:1, /* context initiation */
336 rq_ctx_fini:1, /* context destroy */
337 rq_bulk_read:1, /* request bulk read */
338 rq_bulk_write:1, /* request bulk write */
339 /* server authentication flags */
340 rq_auth_gss:1, /* authenticated by gss */
341 rq_auth_remote:1, /* authed as remote user */
342 rq_auth_usr_root:1, /* authed as root */
343 rq_auth_usr_mdt:1, /* authed as mdt */
344 /* doesn't expect reply FIXME */
347 uid_t rq_auth_uid; /* authed uid */
348 uid_t rq_auth_mapped_uid; /* authed uid mapped to */
350 /* (server side), pointed directly into req buffer */
351 struct ptlrpc_user_desc *rq_user_desc;
353 /* various buffer pointers */
354 struct lustre_msg *rq_reqbuf; /* req wrapper */
355 int rq_reqbuf_len; /* req wrapper buf len */
356 int rq_reqdata_len; /* req wrapper msg len */
357 struct lustre_msg *rq_repbuf; /* rep wrapper */
358 int rq_repbuf_len; /* rep wrapper buf len */
359 int rq_repdata_len; /* rep wrapper msg len */
360 struct lustre_msg *rq_clrbuf; /* only in priv mode */
361 int rq_clrbuf_len; /* only in priv mode */
362 int rq_clrdata_len; /* only in priv mode */
364 __u32 rq_req_swab_mask;
365 __u32 rq_rep_swab_mask;
367 int rq_import_generation;
368 enum lustre_imp_state rq_send_state;
370 /* client+server request */
371 lnet_handle_md_t rq_req_md_h;
372 struct ptlrpc_cb_id rq_req_cbid;
375 struct timeval rq_arrival_time; /* request arrival time */
376 struct ptlrpc_reply_state *rq_reply_state; /* separated reply state */
377 struct ptlrpc_request_buffer_desc *rq_rqbd; /* incoming request buffer*/
379 __u32 rq_uid; /* peer uid, used in MDS only */
382 /* client-only incoming reply */
383 lnet_handle_md_t rq_reply_md_h;
384 cfs_waitq_t rq_reply_waitq;
385 struct ptlrpc_cb_id rq_reply_cbid;
388 lnet_process_id_t rq_peer;
389 struct obd_export *rq_export;
390 struct obd_import *rq_import;
392 void (*rq_replay_cb)(struct ptlrpc_request *);
393 void (*rq_commit_cb)(struct ptlrpc_request *);
396 struct ptlrpc_bulk_desc *rq_bulk; /* client side bulk */
397 time_t rq_sent; /* when request sent, seconds */
400 struct list_head rq_set_chain;
401 struct ptlrpc_request_set *rq_set;
402 void *rq_interpret_reply; /* Async completion handler */
403 union ptlrpc_async_args rq_async_args; /* Async completion context */
404 void *rq_ptlrpcd_data;
405 struct ptlrpc_request_pool *rq_pool; /* Pool if request from
407 struct lu_context rq_session;
410 static inline void ptlrpc_close_replay_seq(struct ptlrpc_request *req)
412 spin_lock(&req->rq_lock);
414 req->rq_sequence = 1;
415 spin_unlock(&req->rq_lock);
418 static inline void lustre_set_req_swabbed(struct ptlrpc_request *req, int index)
420 LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
421 LASSERT((req->rq_req_swab_mask & (1 << index)) == 0);
422 req->rq_req_swab_mask |= 1 << index;
425 static inline void lustre_set_rep_swabbed(struct ptlrpc_request *req, int index)
427 LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
428 LASSERT((req->rq_rep_swab_mask & (1 << index)) == 0);
429 req->rq_rep_swab_mask |= 1 << index;
432 static inline int lustre_req_swabbed(struct ptlrpc_request *req, int index)
434 LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
435 return req->rq_req_swab_mask & (1 << index);
438 static inline int lustre_rep_swabbed(struct ptlrpc_request *req, int index)
440 LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
441 return req->rq_rep_swab_mask & (1 << index);
444 #define SWAB_PARANOIA 1
447 /* unpacking: assert idx not unpacked already */
448 #define LASSERT_REQSWAB(rq, idx) lustre_set_req_swabbed(rq, idx)
449 #define LASSERT_REPSWAB(rq, idx) lustre_set_rep_swabbed(rq, idx)
451 /* just looking: assert idx already unpacked */
452 #define LASSERT_REQSWABBED(rq, idx) LASSERT(lustre_req_swabbed(rq, idx))
453 #define LASSERT_REPSWABBED(rq, idx) LASSERT(lustre_rep_swabbed(rq, idx))
456 #define LASSERT_REQSWAB(rq, idx)
457 #define LASSERT_REPSWAB(rq, idx)
458 #define LASSERT_REQSWABBED(rq, idx)
459 #define LASSERT_REPSWABBED(rq, idx)
462 static inline const char *
463 ptlrpc_rqphase2str(const struct ptlrpc_request *req)
465 switch (req->rq_phase) {
472 case RQ_PHASE_INTERPRET:
474 case RQ_PHASE_COMPLETE:
481 /* Spare the preprocessor, spoil the bugs. */
482 #define FLAG(field, str) (field ? str : "")
484 #define DEBUG_REQ_FLAGS(req) \
485 ptlrpc_rqphase2str(req), \
486 FLAG(req->rq_intr, "I"), FLAG(req->rq_replied, "R"), \
487 FLAG(req->rq_err, "E"), \
488 FLAG(req->rq_timedout, "X") /* eXpired */, FLAG(req->rq_resend, "S"), \
489 FLAG(req->rq_restart, "T"), FLAG(req->rq_replay, "P"), \
490 FLAG(req->rq_no_resend, "N"), \
491 FLAG(req->rq_waiting, "W"), \
492 FLAG(req->rq_wait_ctx, "C")
494 #define REQ_FLAGS_FMT "%s:%s%s%s%s%s%s%s%s%s%s"
496 void _debug_req(struct ptlrpc_request *req, __u32 mask,
497 struct libcfs_debug_msg_data *data, const char *fmt, ...)
498 __attribute__ ((format (printf, 4, 5)));
500 #define debug_req(cdls, level, req, file, func, line, fmt, a...) \
504 if (((level) & D_CANTMASK) != 0 || \
505 ((libcfs_debug & (level)) != 0 && \
506 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) { \
507 static struct libcfs_debug_msg_data _req_dbg_data = \
508 DEBUG_MSG_DATA_INIT(cdls, DEBUG_SUBSYSTEM, file, func, line); \
509 _debug_req((req), (level), &_req_dbg_data, fmt, ##a); \
513 /* for most callers (level is a constant) this is resolved at compile time */
514 #define DEBUG_REQ(level, req, fmt, args...) \
516 if ((level) & (D_ERROR | D_WARNING)) { \
517 static cfs_debug_limit_state_t cdls; \
518 debug_req(&cdls, level, req, __FILE__, __func__, __LINE__, \
519 "@@@ "fmt" ", ## args); \
521 debug_req(NULL, level, req, __FILE__, __func__, __LINE__, \
522 "@@@ "fmt" ", ## args); \
525 struct ptlrpc_bulk_page {
526 struct list_head bp_link;
528 int bp_pageoffset; /* offset within a page */
529 struct page *bp_page;
532 #define BULK_GET_SOURCE 0
533 #define BULK_PUT_SINK 1
534 #define BULK_GET_SINK 2
535 #define BULK_PUT_SOURCE 3
537 struct ptlrpc_bulk_desc {
538 unsigned int bd_success:1; /* completed successfully */
539 unsigned int bd_network_rw:1; /* accessible to the network */
540 unsigned int bd_type:2; /* {put,get}{source,sink} */
541 unsigned int bd_registered:1; /* client side */
542 spinlock_t bd_lock; /* serialise with callback */
543 int bd_import_generation;
544 struct obd_export *bd_export;
545 struct obd_import *bd_import;
547 struct ptlrpc_request *bd_req; /* associated request */
548 cfs_waitq_t bd_waitq; /* server side only WQ */
549 int bd_iov_count; /* # entries in bd_iov */
550 int bd_max_iov; /* allocated size of bd_iov */
551 int bd_nob; /* # bytes covered */
552 int bd_nob_transferred; /* # bytes GOT/PUT */
556 struct ptlrpc_cb_id bd_cbid; /* network callback info */
557 lnet_handle_md_t bd_md_h; /* associated MD */
558 lnet_nid_t bd_sender; /* stash event::sender */
560 cfs_page_t **bd_enc_pages;
561 #if defined(__KERNEL__)
562 lnet_kiov_t bd_iov[0];
564 lnet_md_iovec_t bd_iov[0];
568 struct ptlrpc_thread {
570 struct list_head t_link; /* active threads for service, from svc->srv_threads */
572 void *t_data; /* thread-private data (preallocated memory) */
575 unsigned int t_id; /* service thread index, from ptlrpc_start_threads */
576 cfs_waitq_t t_ctl_waitq;
577 struct lu_env *t_env;
580 struct ptlrpc_request_buffer_desc {
581 struct list_head rqbd_list;
582 struct list_head rqbd_reqs;
583 struct ptlrpc_service *rqbd_service;
584 lnet_handle_md_t rqbd_md_h;
587 struct ptlrpc_cb_id rqbd_cbid;
588 struct ptlrpc_request rqbd_req;
591 typedef int (*svc_handler_t)(struct ptlrpc_request *req);
592 typedef void (*svcreq_printfn_t)(void *, struct ptlrpc_request *);
594 struct ptlrpc_service {
595 struct list_head srv_list; /* chain thru all services */
596 int srv_max_req_size; /* biggest request to receive */
597 int srv_max_reply_size; /* biggest reply to send */
598 int srv_buf_size; /* size of individual buffers */
599 int srv_nbuf_per_group; /* # buffers to allocate in 1 group */
600 int srv_nbufs; /* total # req buffer descs allocated */
601 int srv_threads_min; /* threads to start at SOW */
602 int srv_threads_max; /* thread upper limit */
603 int srv_threads_started; /* index of last started thread */
604 int srv_threads_running; /* # running threads */
605 int srv_n_difficult_replies; /* # 'difficult' replies */
606 int srv_n_active_reqs; /* # reqs being served */
607 cfs_duration_t srv_rqbd_timeout; /* timeout before re-posting reqs, in tick */
608 int srv_watchdog_timeout; /* soft watchdog timeout, in ms */
609 unsigned srv_cpu_affinity:1; /* bind threads to CPUs */
611 __u32 srv_req_portal;
612 __u32 srv_rep_portal;
614 int srv_n_queued_reqs; /* # reqs waiting to be served */
615 struct list_head srv_request_queue; /* reqs waiting for service */
617 struct list_head srv_request_history; /* request history */
618 __u64 srv_request_seq; /* next request sequence # */
619 __u64 srv_request_max_cull_seq; /* highest seq culled from history */
620 svcreq_printfn_t srv_request_history_print_fn; /* service-specific print fn */
622 struct list_head srv_idle_rqbds; /* request buffers to be reposted */
623 struct list_head srv_active_rqbds; /* req buffers receiving */
624 struct list_head srv_history_rqbds; /* request buffer history */
625 int srv_nrqbd_receiving; /* # posted request buffers */
626 int srv_n_history_rqbds; /* # request buffers in history */
627 int srv_max_history_rqbds;/* max # request buffers in history */
629 atomic_t srv_outstanding_replies;
630 struct list_head srv_active_replies; /* all the active replies */
631 struct list_head srv_reply_queue; /* replies waiting for service */
633 cfs_waitq_t srv_waitq; /* all threads sleep on this. This
634 * wait-queue is signalled when new
635 * incoming request arrives and when
636 * difficult reply has to be handled. */
638 struct list_head srv_threads; /* service thread list */
639 svc_handler_t srv_handler;
641 char *srv_name; /* only statically allocated strings here; we don't clean them */
642 char *srv_thread_name; /* only statically allocated strings here; we don't clean them */
646 cfs_proc_dir_entry_t *srv_procroot;
647 struct lprocfs_stats *srv_stats;
649 /* List of free reply_states */
650 struct list_head srv_free_rs_list;
651 /* waitq to run, when adding stuff to srv_free_rs_list */
652 cfs_waitq_t srv_free_rs_waitq;
655 * Tags for lu_context associated with this thread, see struct
660 * if non-NULL called during thread creation (ptlrpc_start_thread())
661 * to initialize service specific per-thread state.
663 int (*srv_init)(struct ptlrpc_thread *thread);
665 * if non-NULL called during thread shutdown (ptlrpc_main()) to
666 * destruct state created by ->srv_init().
668 void (*srv_done)(struct ptlrpc_thread *thread);
670 //struct ptlrpc_srv_ni srv_interfaces[0];
673 /* ptlrpc/events.c */
674 extern lnet_handle_eq_t ptlrpc_eq_h;
675 extern int ptlrpc_uuid_to_peer(struct obd_uuid *uuid,
676 lnet_process_id_t *peer, lnet_nid_t *self);
677 extern void request_out_callback (lnet_event_t *ev);
678 extern void reply_in_callback(lnet_event_t *ev);
679 extern void client_bulk_callback (lnet_event_t *ev);
680 extern void request_in_callback(lnet_event_t *ev);
681 extern void reply_out_callback(lnet_event_t *ev);
682 extern void server_bulk_callback (lnet_event_t *ev);
684 /* ptlrpc/connection.c */
685 void ptlrpc_dump_connections(void);
686 void ptlrpc_readdress_connection(struct ptlrpc_connection *, struct obd_uuid *);
687 struct ptlrpc_connection *ptlrpc_get_connection(lnet_process_id_t peer,
688 lnet_nid_t self, struct obd_uuid *uuid);
689 int ptlrpc_put_connection(struct ptlrpc_connection *c);
690 struct ptlrpc_connection *ptlrpc_connection_addref(struct ptlrpc_connection *);
691 void ptlrpc_init_connection(void);
692 void ptlrpc_cleanup_connection(void);
693 extern lnet_pid_t ptl_get_pid(void);
695 /* ptlrpc/niobuf.c */
696 int ptlrpc_start_bulk_transfer(struct ptlrpc_bulk_desc *desc);
697 void ptlrpc_abort_bulk(struct ptlrpc_bulk_desc *desc);
698 int ptlrpc_register_bulk(struct ptlrpc_request *req);
699 void ptlrpc_unregister_bulk (struct ptlrpc_request *req);
701 static inline int ptlrpc_bulk_active (struct ptlrpc_bulk_desc *desc)
705 spin_lock(&desc->bd_lock);
706 rc = desc->bd_network_rw;
707 spin_unlock(&desc->bd_lock);
711 int ptlrpc_send_reply(struct ptlrpc_request *req, int);
712 int ptlrpc_reply(struct ptlrpc_request *req);
713 int ptlrpc_error(struct ptlrpc_request *req);
714 void ptlrpc_resend_req(struct ptlrpc_request *request);
715 int ptl_send_rpc(struct ptlrpc_request *request, int noreply);
716 int ptlrpc_register_rqbd (struct ptlrpc_request_buffer_desc *rqbd);
718 /* ptlrpc/client.c */
719 void ptlrpc_init_client(int req_portal, int rep_portal, char *name,
720 struct ptlrpc_client *);
721 void ptlrpc_cleanup_client(struct obd_import *imp);
722 struct ptlrpc_connection *ptlrpc_uuid_to_connection(struct obd_uuid *uuid);
725 ptlrpc_client_receiving_reply (struct ptlrpc_request *req)
729 spin_lock(&req->rq_lock);
730 rc = req->rq_receiving_reply;
731 spin_unlock(&req->rq_lock);
736 ptlrpc_client_replied (struct ptlrpc_request *req)
740 spin_lock(&req->rq_lock);
741 rc = req->rq_replied;
742 spin_unlock(&req->rq_lock);
747 ptlrpc_wake_client_req (struct ptlrpc_request *req)
749 if (req->rq_set == NULL)
750 cfs_waitq_signal(&req->rq_reply_waitq);
752 cfs_waitq_signal(&req->rq_set->set_waitq);
755 int ptlrpc_queue_wait(struct ptlrpc_request *req);
756 int ptlrpc_replay_req(struct ptlrpc_request *req);
757 void ptlrpc_unregister_reply(struct ptlrpc_request *req);
758 void ptlrpc_restart_req(struct ptlrpc_request *req);
759 void ptlrpc_abort_inflight(struct obd_import *imp);
761 struct ptlrpc_request_set *ptlrpc_prep_set(void);
762 int ptlrpc_set_next_timeout(struct ptlrpc_request_set *);
763 int ptlrpc_check_set(struct ptlrpc_request_set *set);
764 int ptlrpc_set_wait(struct ptlrpc_request_set *);
765 int ptlrpc_expired_set(void *data);
766 void ptlrpc_interrupted_set(void *data);
767 void ptlrpc_mark_interrupted(struct ptlrpc_request *req);
768 void ptlrpc_set_destroy(struct ptlrpc_request_set *);
769 void ptlrpc_set_add_req(struct ptlrpc_request_set *, struct ptlrpc_request *);
770 void ptlrpc_set_add_new_req(struct ptlrpc_request_set *,
771 struct ptlrpc_request *);
773 void ptlrpc_free_rq_pool(struct ptlrpc_request_pool *pool);
774 void ptlrpc_add_rqs_to_pool(struct ptlrpc_request_pool *pool, int num_rq);
775 struct ptlrpc_request_pool *ptlrpc_init_rq_pool(int, int,
776 void (*populate_pool)(struct ptlrpc_request_pool *, int));
777 struct ptlrpc_request *ptlrpc_prep_req(struct obd_import *imp, __u32 version,
778 int opcode, int count, int *lengths,
780 struct ptlrpc_request *ptlrpc_prep_req_pool(struct obd_import *imp,
781 __u32 version, int opcode,
782 int count, int *lengths, char **bufs,
783 struct ptlrpc_request_pool *pool,
784 struct ptlrpc_cli_ctx *ctx);
785 void ptlrpc_free_req(struct ptlrpc_request *request);
786 void ptlrpc_req_finished(struct ptlrpc_request *request);
787 void ptlrpc_req_finished_with_imp_lock(struct ptlrpc_request *request);
788 struct ptlrpc_request *ptlrpc_request_addref(struct ptlrpc_request *req);
789 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_imp (struct ptlrpc_request *req,
790 int npages, int type, int portal);
791 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_exp(struct ptlrpc_request *req,
792 int npages, int type, int portal);
793 void ptlrpc_free_bulk(struct ptlrpc_bulk_desc *bulk);
794 void ptlrpc_prep_bulk_page(struct ptlrpc_bulk_desc *desc,
795 cfs_page_t *page, int pageoffset, int len);
796 void ptlrpc_retain_replayable_request(struct ptlrpc_request *req,
797 struct obd_import *imp);
798 __u64 ptlrpc_next_xid(void);
799 __u64 ptlrpc_sample_next_xid(void);
800 __u64 ptlrpc_req_xid(struct ptlrpc_request *request);
802 struct ptlrpc_service_conf {
805 int psc_max_req_size;
806 int psc_max_reply_size;
809 int psc_watchdog_timeout; /* in ms */
815 /* ptlrpc/service.c */
816 void ptlrpc_save_lock (struct ptlrpc_request *req,
817 struct lustre_handle *lock, int mode);
818 void ptlrpc_commit_replies (struct obd_device *obd);
819 void ptlrpc_schedule_difficult_reply (struct ptlrpc_reply_state *rs);
820 struct ptlrpc_service *ptlrpc_init_svc_conf(struct ptlrpc_service_conf *c,
821 svc_handler_t h, char *name,
822 struct proc_dir_entry *proc_entry,
823 svcreq_printfn_t prntfn,
826 struct ptlrpc_service *ptlrpc_init_svc(int nbufs, int bufsize, int max_req_size,
828 int req_portal, int rep_portal,
829 int watchdog_timeout, /* in ms */
830 svc_handler_t, char *name,
831 cfs_proc_dir_entry_t *proc_entry,
833 int min_threads, int max_threads,
834 char *threadname, __u32 ctx_tags);
835 void ptlrpc_stop_all_threads(struct ptlrpc_service *svc);
837 int ptlrpc_start_threads(struct obd_device *dev, struct ptlrpc_service *svc);
838 int ptlrpc_start_thread(struct obd_device *dev, struct ptlrpc_service *svc);
839 int ptlrpc_unregister_service(struct ptlrpc_service *service);
840 int liblustre_check_services (void *arg);
841 void ptlrpc_daemonize(char *name);
842 int ptlrpc_service_health_check(struct ptlrpc_service *);
845 struct ptlrpc_svc_data {
847 struct ptlrpc_service *svc;
848 struct ptlrpc_thread *thread;
849 struct obd_device *dev;
852 /* ptlrpc/import.c */
853 int ptlrpc_connect_import(struct obd_import *imp, char * new_uuid);
854 int ptlrpc_init_import(struct obd_import *imp);
855 int ptlrpc_disconnect_import(struct obd_import *imp, int noclose);
856 int ptlrpc_import_recovery_state_machine(struct obd_import *imp);
858 /* ptlrpc/pack_generic.c */
859 int lustre_msg_swabbed(struct lustre_msg *msg);
860 int lustre_msg_check_version(struct lustre_msg *msg, __u32 version);
861 void lustre_init_msg_v2(struct lustre_msg_v2 *msg, int count, int *lens,
863 int lustre_pack_request(struct ptlrpc_request *, __u32 magic, int count,
864 int *lens, char **bufs);
865 int lustre_pack_reply(struct ptlrpc_request *, int count, int *lens,
867 int lustre_pack_reply_v2(struct ptlrpc_request *req, int count,
868 int *lens, char **bufs);
869 int lustre_shrink_msg(struct lustre_msg *msg, int segment,
870 unsigned int newlen, int move_data);
871 void lustre_free_reply_state(struct ptlrpc_reply_state *rs);
872 int lustre_msg_size(__u32 magic, int count, int *lengths);
873 int lustre_msg_size_v2(int count, int *lengths);
874 int lustre_unpack_msg(struct lustre_msg *m, int len);
875 void *lustre_msg_buf_v1(void *msg, int n, int min_size);
876 void *lustre_msg_buf_v2(struct lustre_msg_v2 *m, int n, int min_size);
877 void *lustre_msg_buf(struct lustre_msg *m, int n, int minlen);
878 int lustre_msg_buflen(struct lustre_msg *m, int n);
879 void lustre_msg_set_buflen(struct lustre_msg *m, int n, int len);
880 int lustre_msg_bufcount(struct lustre_msg *m);
881 char *lustre_msg_string (struct lustre_msg *m, int n, int max_len);
882 void *lustre_swab_buf(struct lustre_msg *, int n, int minlen, void *swabber);
883 void *lustre_swab_reqbuf(struct ptlrpc_request *req, int n, int minlen,
885 void *lustre_swab_repbuf(struct ptlrpc_request *req, int n, int minlen,
887 __u32 lustre_msg_get_flags(struct lustre_msg *msg);
888 void lustre_msg_add_flags(struct lustre_msg *msg, int flags);
889 void lustre_msg_set_flags(struct lustre_msg *msg, int flags);
890 void lustre_msg_clear_flags(struct lustre_msg *msg, int flags);
891 __u32 lustre_msg_get_op_flags(struct lustre_msg *msg);
892 void lustre_msg_add_op_flags(struct lustre_msg *msg, int flags);
893 void lustre_msg_set_op_flags(struct lustre_msg *msg, int flags);
894 struct lustre_handle *lustre_msg_get_handle(struct lustre_msg *msg);
895 __u32 lustre_msg_get_type(struct lustre_msg *msg);
896 __u32 lustre_msg_get_version(struct lustre_msg *msg);
897 void lustre_msg_add_version(struct lustre_msg *msg, int version);
898 __u32 lustre_msg_get_opc(struct lustre_msg *msg);
899 __u64 lustre_msg_get_last_xid(struct lustre_msg *msg);
900 __u64 lustre_msg_get_last_committed(struct lustre_msg *msg);
901 __u64 lustre_msg_get_transno(struct lustre_msg *msg);
902 __u64 lustre_msg_get_slv(struct lustre_msg *msg);
903 __u32 lustre_msg_get_limit(struct lustre_msg *msg);
904 void lustre_msg_set_slv(struct lustre_msg *msg, __u64 slv);
905 void lustre_msg_set_limit(struct lustre_msg *msg, __u64 limit);
906 int lustre_msg_get_status(struct lustre_msg *msg);
907 __u32 lustre_msg_get_conn_cnt(struct lustre_msg *msg);
908 __u32 lustre_msg_get_magic(struct lustre_msg *msg);
909 void lustre_msg_set_handle(struct lustre_msg *msg,struct lustre_handle *handle);
910 void lustre_msg_set_type(struct lustre_msg *msg, __u32 type);
911 void lustre_msg_set_opc(struct lustre_msg *msg, __u32 opc);
912 void lustre_msg_set_last_xid(struct lustre_msg *msg, __u64 last_xid);
913 void lustre_msg_set_last_committed(struct lustre_msg *msg,__u64 last_committed);
914 void lustre_msg_set_transno(struct lustre_msg *msg, __u64 transno);
915 void lustre_msg_set_status(struct lustre_msg *msg, __u32 status);
916 void lustre_msg_set_conn_cnt(struct lustre_msg *msg, __u32 conn_cnt);
919 lustre_shrink_reply(struct ptlrpc_request *req, int segment,
920 unsigned int newlen, int move_data)
922 LASSERT(req->rq_reply_state);
923 LASSERT(req->rq_repmsg);
924 req->rq_replen = lustre_shrink_msg(req->rq_repmsg, segment,
929 ptlrpc_rs_addref(struct ptlrpc_reply_state *rs)
931 LASSERT(atomic_read(&rs->rs_refcount) > 0);
932 atomic_inc(&rs->rs_refcount);
936 ptlrpc_rs_decref(struct ptlrpc_reply_state *rs)
938 LASSERT(atomic_read(&rs->rs_refcount) > 0);
939 if (atomic_dec_and_test(&rs->rs_refcount))
940 lustre_free_reply_state(rs);
943 static inline __u32 lustre_request_magic(struct ptlrpc_request *req)
945 return lustre_msg_get_magic(req->rq_reqmsg);
948 static inline int ptlrpc_req_get_repsize(struct ptlrpc_request *req)
950 switch (req->rq_reqmsg->lm_magic) {
951 case LUSTRE_MSG_MAGIC_V1:
952 CERROR("function not supported for lustre_msg V1!\n");
954 case LUSTRE_MSG_MAGIC_V2:
955 return req->rq_reqmsg->lm_repsize;
957 LASSERTF(0, "incorrect message magic: %08x\n",
958 req->rq_reqmsg->lm_magic);
964 ptlrpc_req_set_repsize(struct ptlrpc_request *req, int count, int *lens)
966 req->rq_replen = lustre_msg_size(req->rq_reqmsg->lm_magic, count, lens);
967 if (req->rq_reqmsg->lm_magic == LUSTRE_MSG_MAGIC_V2)
968 req->rq_reqmsg->lm_repsize = req->rq_replen;
971 /* ldlm/ldlm_lib.c */
972 int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg);
973 int client_obd_cleanup(struct obd_device *obddev);
974 int client_connect_import(const struct lu_env *env,
975 struct lustre_handle *conn, struct obd_device *obd,
976 struct obd_uuid *cluuid, struct obd_connect_data *);
977 int client_disconnect_export(struct obd_export *exp);
978 int client_import_add_conn(struct obd_import *imp, struct obd_uuid *uuid,
980 int client_import_del_conn(struct obd_import *imp, struct obd_uuid *uuid);
981 int import_set_conn_priority(struct obd_import *imp, struct obd_uuid *uuid);
983 /* ptlrpc/pinger.c */
984 int ptlrpc_pinger_add_import(struct obd_import *imp);
985 int ptlrpc_pinger_del_import(struct obd_import *imp);
986 cfs_time_t ptlrpc_suspend_wakeup_time(void);
988 void ping_evictor_start(void);
989 void ping_evictor_stop(void);
991 #define ping_evictor_start() do {} while (0)
992 #define ping_evictor_stop() do {} while (0)
994 int ptlrpc_check_and_wait_suspend(struct ptlrpc_request *req);
996 /* ptlrpc/ptlrpcd.c */
997 void ptlrpcd_wake(struct ptlrpc_request *req);
998 void ptlrpcd_add_req(struct ptlrpc_request *req);
999 int ptlrpcd_addref(void);
1000 void ptlrpcd_decref(void);
1002 /* ptlrpc/lproc_ptlrpc.c */
1003 const char* ll_opcode2str(__u32 opcode);
1005 void ptlrpc_lprocfs_register_obd(struct obd_device *obd);
1006 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd);
1007 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int opc, int bytes);
1009 static inline void ptlrpc_lprocfs_register_obd(struct obd_device *obd) {}
1010 static inline void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd) {}
1011 static inline void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int opc,
1015 /* ptlrpc/llog_server.c */
1016 int llog_origin_handle_create(struct ptlrpc_request *req);
1017 int llog_origin_handle_destroy(struct ptlrpc_request *req);
1018 int llog_origin_handle_prev_block(struct ptlrpc_request *req);
1019 int llog_origin_handle_next_block(struct ptlrpc_request *req);
1020 int llog_origin_handle_read_header(struct ptlrpc_request *req);
1021 int llog_origin_handle_close(struct ptlrpc_request *req);
1022 int llog_origin_handle_cancel(struct ptlrpc_request *req);
1023 int llog_catinfo(struct ptlrpc_request *req);
1025 /* ptlrpc/llog_client.c */
1026 extern struct llog_operations llog_client_ops;