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/libcfs.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>
45 #include <lustre_req_layout.h>
47 /* MD flags we _always_ use */
48 #define PTLRPC_MD_OPTIONS 0
50 /* Define maxima for bulk I/O
51 * CAVEAT EMPTOR, with multinet (i.e. routers forwarding between networks)
52 * these limits are system wide and not interface-local. */
53 #define PTLRPC_MAX_BRW_BITS LNET_MTU_BITS
54 #define PTLRPC_MAX_BRW_SIZE (1<<LNET_MTU_BITS)
55 #define PTLRPC_MAX_BRW_PAGES (PTLRPC_MAX_BRW_SIZE >> CFS_PAGE_SHIFT)
57 /* When PAGE_SIZE is a constant, we can check our arithmetic here with cpp! */
59 # if ((PTLRPC_MAX_BRW_PAGES & (PTLRPC_MAX_BRW_PAGES - 1)) != 0)
60 # error "PTLRPC_MAX_BRW_PAGES isn't a power of two"
62 # if (PTLRPC_MAX_BRW_SIZE != (PTLRPC_MAX_BRW_PAGES * CFS_PAGE_SIZE))
63 # error "PTLRPC_MAX_BRW_SIZE isn't PTLRPC_MAX_BRW_PAGES * CFS_PAGE_SIZE"
65 # if (PTLRPC_MAX_BRW_SIZE > LNET_MTU)
66 # error "PTLRPC_MAX_BRW_SIZE too big"
68 # if (PTLRPC_MAX_BRW_PAGES > LNET_MAX_IOV)
69 # error "PTLRPC_MAX_BRW_PAGES too big"
71 #endif /* __KERNEL__ */
73 /* Size over which to OBD_VMALLOC() rather than OBD_ALLOC() service request
75 #define SVC_BUF_VMALLOC_THRESHOLD (2 * CFS_PAGE_SIZE)
77 /* The following constants determine how memory is used to buffer incoming
80 * ?_NBUFS # buffers to allocate when growing the pool
81 * ?_BUFSIZE # bytes in a single request buffer
82 * ?_MAXREQSIZE # maximum request service will receive
84 * When fewer than ?_NBUFS/2 buffers are posted for receive, another chunk
85 * of ?_NBUFS is added to the pool.
87 * Messages larger than ?_MAXREQSIZE are dropped. Request buffers are
88 * considered full when less than ?_MAXREQSIZE is left in them.
91 #define LDLM_THREADS_AUTO_MIN \
92 min((int)(num_online_cpus() * num_online_cpus() * 2), 8)
93 #define LDLM_THREADS_AUTO_MAX (LDLM_THREADS_AUTO_MIN * 16)
94 #define LDLM_BL_THREADS LDLM_THREADS_AUTO_MIN
95 #define LDLM_NBUFS (64 * num_online_cpus())
96 #define LDLM_BUFSIZE (8 * 1024)
97 #define LDLM_MAXREQSIZE (5 * 1024)
98 #define LDLM_MAXREPSIZE (1024)
100 #define MDT_MIN_THREADS 2UL
101 #define MDT_MAX_THREADS 512UL
102 #define MDT_NUM_THREADS max(min_t(unsigned long, MDT_MAX_THREADS, \
103 num_physpages >> (25 - CFS_PAGE_SHIFT)), 2UL)
104 #define FLD_NUM_THREADS max(min_t(unsigned long, MDT_MAX_THREADS, \
105 num_physpages >> (25 - CFS_PAGE_SHIFT)), 2UL)
106 #define SEQ_NUM_THREADS max(min_t(unsigned long, MDT_MAX_THREADS, \
107 num_physpages >> (25 - CFS_PAGE_SHIFT)), 2UL)
109 /* Absolute limits */
110 #define MDS_THREADS_MIN 2
111 #define MDS_THREADS_MAX 512
112 #define MDS_THREADS_MIN_READPAGE 2
113 #define MDS_NBUFS (64 * num_online_cpus())
114 #define MDS_BUFSIZE (8 * 1024)
115 /* Assume file name length = FNAME_MAX = 256 (true for ext3).
116 * path name length = PATH_MAX = 4096
117 * LOV MD size max = EA_MAX = 4000
118 * symlink: FNAME_MAX + PATH_MAX <- largest
119 * link: FNAME_MAX + PATH_MAX (mds_rec_link < mds_rec_create)
120 * rename: FNAME_MAX + FNAME_MAX
121 * open: FNAME_MAX + EA_MAX
123 * MDS_MAXREQSIZE ~= 4736 bytes =
124 * lustre_msg + ldlm_request + mds_body + mds_rec_create + FNAME_MAX + PATH_MAX
125 * MDS_MAXREPSIZE ~= 8300 bytes = lustre_msg + llog_header
126 * or, for mds_close() and mds_reint_unlink() on a many-OST filesystem:
127 * = 9210 bytes = lustre_msg + mds_body + 160 * (easize + cookiesize)
129 * Realistic size is about 512 bytes (20 character name + 128 char symlink),
130 * except in the open case where there are a large number of OSTs in a LOV.
132 #define MDS_MAXREQSIZE (5 * 1024)
133 #define MDS_MAXREPSIZE max(9 * 1024, 280 + LOV_MAX_STRIPE_COUNT * 56)
135 /* FLD_MAXREQSIZE == lustre_msg + __u32 padding + ptlrpc_body + opc + md_fld */
136 #define FLD_MAXREQSIZE (160)
138 /* FLD_MAXREPSIZE == lustre_msg + ptlrpc_body + md_fld */
139 #define FLD_MAXREPSIZE (152)
141 /* SEQ_MAXREQSIZE == lustre_msg + __u32 padding + ptlrpc_body + opc + lu_range +
143 #define SEQ_MAXREQSIZE (160)
145 /* SEQ_MAXREPSIZE == lustre_msg + ptlrpc_body + lu_range */
146 #define SEQ_MAXREPSIZE (152)
148 #define MGS_THREADS_AUTO_MIN 2
149 #define MGS_THREADS_AUTO_MAX 32
150 #define MGS_NBUFS (64 * num_online_cpus())
151 #define MGS_BUFSIZE (8 * 1024)
152 #define MGS_MAXREQSIZE (7 * 1024)
153 #define MGS_MAXREPSIZE (9 * 1024)
155 /* Absolute limits */
156 #define OSS_THREADS_MIN 2
157 #define OSS_THREADS_MAX 512
158 #define OST_NBUFS (64 * num_online_cpus())
159 #define OST_BUFSIZE (8 * 1024)
160 /* OST_MAXREQSIZE ~= 4768 bytes =
161 * lustre_msg + obdo + 16 * obd_ioobj + 256 * niobuf_remote
163 * - single object with 16 pages is 512 bytes
164 * - OST_MAXREQSIZE must be at least 1 page of cookies plus some spillover
166 #define OST_MAXREQSIZE (5 * 1024)
167 #define OST_MAXREPSIZE (9 * 1024)
169 /* Macro to hide a typecast. */
170 #define ptlrpc_req_async_args(req) ((void *)&req->rq_async_args)
172 struct ptlrpc_connection {
173 struct list_head c_link;
174 struct hlist_node c_hash;
176 lnet_process_id_t c_peer;
177 struct obd_uuid c_remote_uuid;
181 struct ptlrpc_client {
182 __u32 cli_request_portal;
183 __u32 cli_reply_portal;
187 /* state flags of requests */
188 /* XXX only ones left are those used by the bulk descs as well! */
189 #define PTL_RPC_FL_INTR (1 << 0) /* reply wait was interrupted by user */
190 #define PTL_RPC_FL_TIMEOUT (1 << 7) /* request timed out waiting for reply */
192 #define REQ_MAX_ACK_LOCKS 8
194 union ptlrpc_async_args {
195 /* Scratchpad for passing args to completion interpreter. Users
196 * cast to the struct of their choosing, and LASSERT that this is
197 * big enough. For _tons_ of context, OBD_ALLOC a struct and store
198 * a pointer to it here. The pointer_arg ensures this struct is at
199 * least big enough for that. */
200 void *pointer_arg[9];
204 struct ptlrpc_request_set;
205 typedef int (*set_interpreter_func)(struct ptlrpc_request_set *, void *, int);
207 struct ptlrpc_request_set {
208 int set_remaining; /* # uncompleted requests */
209 cfs_waitq_t set_waitq;
210 cfs_waitq_t *set_wakeup_ptr;
211 struct list_head set_requests;
212 struct list_head set_cblist; /* list of completion callbacks */
213 set_interpreter_func set_interpret; /* completion callback */
214 void *set_arg; /* completion context */
215 /* locked so that any old caller can communicate requests to
216 * the set holder who can then fold them into the lock-free set */
217 spinlock_t set_new_req_lock;
218 struct list_head set_new_requests;
221 struct ptlrpc_set_cbdata {
222 struct list_head psc_item;
223 set_interpreter_func psc_interpret;
227 struct ptlrpc_bulk_desc;
230 * ptlrpc callback & work item stuff
232 struct ptlrpc_cb_id {
233 void (*cbid_fn)(lnet_event_t *ev); /* specific callback fn */
234 void *cbid_arg; /* additional arg */
237 #define RS_MAX_LOCKS 4
240 struct ptlrpc_reply_state {
241 struct ptlrpc_cb_id rs_cb_id;
242 struct list_head rs_list;
243 struct list_head rs_exp_list;
244 struct list_head rs_obd_list;
246 struct list_head rs_debug_list;
248 /* updates to following flag serialised by srv_request_lock */
249 unsigned long rs_difficult:1; /* ACK/commit stuff */
250 unsigned long rs_scheduled:1; /* being handled? */
251 unsigned long rs_scheduled_ever:1;/* any schedule attempts? */
252 unsigned long rs_handled:1; /* been handled yet? */
253 unsigned long rs_on_net:1; /* reply_out_callback pending? */
254 unsigned long rs_prealloc:1; /* rs from prealloc list */
259 struct obd_export *rs_export;
260 struct ptlrpc_service *rs_service;
261 lnet_handle_md_t rs_md_h;
262 atomic_t rs_refcount;
264 struct ptlrpc_svc_ctx *rs_svc_ctx;
265 struct lustre_msg *rs_repbuf; /* wrapper */
266 int rs_repbuf_len; /* wrapper buf length */
267 int rs_repdata_len; /* wrapper msg length */
268 struct lustre_msg *rs_msg; /* reply message */
270 /* locks awaiting client reply ACK */
272 struct lustre_handle rs_locks[RS_MAX_LOCKS];
273 ldlm_mode_t rs_modes[RS_MAX_LOCKS];
276 struct ptlrpc_thread;
279 RQ_PHASE_NEW = 0xebc0de00,
280 RQ_PHASE_RPC = 0xebc0de01,
281 RQ_PHASE_BULK = 0xebc0de02,
282 RQ_PHASE_INTERPRET = 0xebc0de03,
283 RQ_PHASE_COMPLETE = 0xebc0de04,
286 struct ptlrpc_request_pool {
288 struct list_head prp_req_list; /* list of ptlrpc_request structs */
290 void (*prp_populate)(struct ptlrpc_request_pool *, int);
296 struct ptlrpc_request {
297 int rq_type; /* one of PTL_RPC_MSG_* */
298 struct list_head rq_list;
299 struct list_head rq_timed_list; /* server-side early replies */
300 struct list_head rq_history_list; /* server-side history */
301 __u64 rq_history_seq; /* history sequence # */
304 /* client-side flags are serialized by rq_lock */
305 unsigned long rq_intr:1, rq_replied:1, rq_err:1,
306 rq_timedout:1, rq_resend:1, rq_restart:1,
308 * when ->rq_replay is set, request is kept by the client even
309 * after server commits corresponding transaction. This is
310 * used for operations that require sequence of multiple
311 * requests to be replayed. The only example currently is file
312 * open/close/dw/setattr. When last request in such a sequence
313 * is committed, ->rq_replay is cleared on all requests in the
317 /* this is the last request in the sequence. */
319 rq_no_resend:1, rq_waiting:1, rq_receiving_reply:1,
320 rq_no_delay:1, rq_net_err:1, rq_wait_ctx:1,
321 rq_early:1, rq_must_unlink:1,
322 /* server-side flags */
323 rq_packed_final:1, /* packed final reply */
324 rq_sent_final:1; /* stop sending early replies */
326 enum rq_phase rq_phase; /* one of RQ_PHASE_* */
327 atomic_t rq_refcount; /* client-side refcount for SENT race,
328 server-side refcounf for multiple replies */
330 struct ptlrpc_thread *rq_svc_thread; /* initial thread servicing req */
332 int rq_request_portal; /* XXX FIXME bug 249 */
333 int rq_reply_portal; /* XXX FIXME bug 249 */
335 int rq_nob_received; /* client-side # reply bytes actually received */
338 struct lustre_msg *rq_reqmsg;
341 struct lustre_msg *rq_repmsg;
344 struct list_head rq_replay_list;
345 struct list_head rq_mod_list;
347 struct ptlrpc_cli_ctx *rq_cli_ctx; /* client's half ctx */
348 struct ptlrpc_svc_ctx *rq_svc_ctx; /* server's half ctx */
349 struct list_head rq_ctx_chain; /* link to waited ctx */
351 struct sptlrpc_flavor rq_flvr; /* client & server */
352 enum lustre_sec_part rq_sp_from;
354 unsigned long /* client/server security flags */
355 rq_ctx_init:1, /* context initiation */
356 rq_ctx_fini:1, /* context destroy */
357 rq_bulk_read:1, /* request bulk read */
358 rq_bulk_write:1, /* request bulk write */
359 /* server authentication flags */
360 rq_auth_gss:1, /* authenticated by gss */
361 rq_auth_remote:1, /* authed as remote user */
362 rq_auth_usr_root:1, /* authed as root */
363 rq_auth_usr_mdt:1, /* authed as mdt */
364 /* security tfm flags */
367 /* doesn't expect reply FIXME */
370 uid_t rq_auth_uid; /* authed uid */
371 uid_t rq_auth_mapped_uid; /* authed uid mapped to */
373 /* (server side), pointed directly into req buffer */
374 struct ptlrpc_user_desc *rq_user_desc;
376 /* early replies go to offset 0, regular replies go after that */
377 unsigned int rq_reply_off;
379 /* various buffer pointers */
380 struct lustre_msg *rq_reqbuf; /* req wrapper */
381 int rq_reqbuf_len; /* req wrapper buf len */
382 int rq_reqdata_len; /* req wrapper msg len */
383 char *rq_repbuf; /* rep buffer */
384 int rq_repbuf_len; /* rep buffer len */
385 struct lustre_msg *rq_repdata; /* rep wrapper msg */
386 int rq_repdata_len; /* rep wrapper msg len */
387 struct lustre_msg *rq_clrbuf; /* only in priv mode */
388 int rq_clrbuf_len; /* only in priv mode */
389 int rq_clrdata_len; /* only in priv mode */
391 __u32 rq_req_swab_mask;
392 __u32 rq_rep_swab_mask;
394 int rq_import_generation;
395 enum lustre_imp_state rq_send_state;
397 int rq_early_count; /* how many early replies (for stats) */
399 /* client+server request */
400 lnet_handle_md_t rq_req_md_h;
401 struct ptlrpc_cb_id rq_req_cbid;
404 struct timeval rq_arrival_time; /* request arrival time */
405 struct ptlrpc_reply_state *rq_reply_state; /* separated reply state */
406 struct ptlrpc_request_buffer_desc *rq_rqbd; /* incoming request buffer*/
408 __u32 rq_uid; /* peer uid, used in MDS only */
411 /* client-only incoming reply */
412 lnet_handle_md_t rq_reply_md_h;
413 cfs_waitq_t rq_reply_waitq;
414 struct ptlrpc_cb_id rq_reply_cbid;
417 lnet_process_id_t rq_peer;
418 struct obd_export *rq_export;
419 struct obd_import *rq_import;
421 void (*rq_replay_cb)(struct ptlrpc_request *);
422 void (*rq_commit_cb)(struct ptlrpc_request *);
425 struct ptlrpc_bulk_desc *rq_bulk;/* client side bulk */
427 /* client outgoing req */
428 time_t rq_sent; /* when request/reply sent (secs), or
429 * time when request should be sent */
431 volatile time_t rq_deadline; /* when request must finish. volatile
432 so that servers' early reply updates to the deadline aren't
433 kept in per-cpu cache */
434 int rq_timeout; /* service time estimate (secs) */
437 struct list_head rq_set_chain;
438 struct ptlrpc_request_set *rq_set;
439 void *rq_interpret_reply; /* Async completion handler */
440 union ptlrpc_async_args rq_async_args; /* Async completion context */
441 struct ptlrpc_request_pool *rq_pool; /* Pool if request from
443 struct lu_context rq_session;
446 struct req_capsule rq_pill;
449 static inline void ptlrpc_close_replay_seq(struct ptlrpc_request *req)
451 spin_lock(&req->rq_lock);
453 req->rq_sequence = 1;
454 spin_unlock(&req->rq_lock);
457 static inline void lustre_set_req_swabbed(struct ptlrpc_request *req, int index)
459 LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
460 LASSERT((req->rq_req_swab_mask & (1 << index)) == 0);
461 req->rq_req_swab_mask |= 1 << index;
464 static inline void lustre_set_rep_swabbed(struct ptlrpc_request *req, int index)
466 LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
467 LASSERT((req->rq_rep_swab_mask & (1 << index)) == 0);
468 req->rq_rep_swab_mask |= 1 << index;
471 static inline int lustre_req_swabbed(struct ptlrpc_request *req, int index)
473 LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
474 return req->rq_req_swab_mask & (1 << index);
477 static inline int lustre_rep_swabbed(struct ptlrpc_request *req, int index)
479 LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
480 return req->rq_rep_swab_mask & (1 << index);
483 static inline const char *
484 ptlrpc_rqphase2str(const struct ptlrpc_request *req)
486 switch (req->rq_phase) {
493 case RQ_PHASE_INTERPRET:
495 case RQ_PHASE_COMPLETE:
502 /* Spare the preprocessor, spoil the bugs. */
503 #define FLAG(field, str) (field ? str : "")
505 #define DEBUG_REQ_FLAGS(req) \
506 ptlrpc_rqphase2str(req), \
507 FLAG(req->rq_intr, "I"), FLAG(req->rq_replied, "R"), \
508 FLAG(req->rq_err, "E"), \
509 FLAG(req->rq_timedout, "X") /* eXpired */, FLAG(req->rq_resend, "S"), \
510 FLAG(req->rq_restart, "T"), FLAG(req->rq_replay, "P"), \
511 FLAG(req->rq_no_resend, "N"), \
512 FLAG(req->rq_waiting, "W"), \
513 FLAG(req->rq_wait_ctx, "C")
515 #define REQ_FLAGS_FMT "%s:%s%s%s%s%s%s%s%s%s%s"
517 void _debug_req(struct ptlrpc_request *req, __u32 mask,
518 struct libcfs_debug_msg_data *data, const char *fmt, ...)
519 __attribute__ ((format (printf, 4, 5)));
521 #define debug_req(cdls, level, req, file, func, line, fmt, a...) \
525 if (((level) & D_CANTMASK) != 0 || \
526 ((libcfs_debug & (level)) != 0 && \
527 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) { \
528 static struct libcfs_debug_msg_data _req_dbg_data = \
529 DEBUG_MSG_DATA_INIT(cdls, DEBUG_SUBSYSTEM, file, func, line); \
530 _debug_req((req), (level), &_req_dbg_data, fmt, ##a); \
534 /* for most callers (level is a constant) this is resolved at compile time */
535 #define DEBUG_REQ(level, req, fmt, args...) \
537 if ((level) & (D_ERROR | D_WARNING)) { \
538 static cfs_debug_limit_state_t cdls; \
539 debug_req(&cdls, level, req, __FILE__, __func__, __LINE__, \
540 "@@@ "fmt" ", ## args); \
542 debug_req(NULL, level, req, __FILE__, __func__, __LINE__, \
543 "@@@ "fmt" ", ## args); \
546 struct ptlrpc_bulk_page {
547 struct list_head bp_link;
549 int bp_pageoffset; /* offset within a page */
550 struct page *bp_page;
553 #define BULK_GET_SOURCE 0
554 #define BULK_PUT_SINK 1
555 #define BULK_GET_SINK 2
556 #define BULK_PUT_SOURCE 3
558 struct ptlrpc_bulk_desc {
559 unsigned long bd_success:1; /* completed successfully */
560 unsigned long bd_network_rw:1; /* accessible to the network */
561 unsigned long bd_type:2; /* {put,get}{source,sink} */
562 unsigned long bd_registered:1; /* client side */
563 spinlock_t bd_lock; /* serialise with callback */
564 int bd_import_generation;
565 struct obd_export *bd_export;
566 struct obd_import *bd_import;
568 struct ptlrpc_request *bd_req; /* associated request */
569 cfs_waitq_t bd_waitq; /* server side only WQ */
570 int bd_iov_count; /* # entries in bd_iov */
571 int bd_max_iov; /* allocated size of bd_iov */
572 int bd_nob; /* # bytes covered */
573 int bd_nob_transferred; /* # bytes GOT/PUT */
577 struct ptlrpc_cb_id bd_cbid; /* network callback info */
578 lnet_handle_md_t bd_md_h; /* associated MD */
579 lnet_nid_t bd_sender; /* stash event::sender */
581 cfs_page_t **bd_enc_pages;
582 #if defined(__KERNEL__)
583 lnet_kiov_t bd_iov[0];
585 lnet_md_iovec_t bd_iov[0];
589 struct ptlrpc_thread {
591 struct list_head t_link; /* active threads for service, from svc->srv_threads */
593 void *t_data; /* thread-private data (preallocated memory) */
596 unsigned int t_id; /* service thread index, from ptlrpc_start_threads */
597 cfs_waitq_t t_ctl_waitq;
598 struct lu_env *t_env;
601 struct ptlrpc_request_buffer_desc {
602 struct list_head rqbd_list;
603 struct list_head rqbd_reqs;
604 struct ptlrpc_service *rqbd_service;
605 lnet_handle_md_t rqbd_md_h;
608 struct ptlrpc_cb_id rqbd_cbid;
609 struct ptlrpc_request rqbd_req;
612 typedef int (*svc_handler_t)(struct ptlrpc_request *req);
613 typedef void (*svcreq_printfn_t)(void *, struct ptlrpc_request *);
615 struct ptlrpc_service {
616 struct list_head srv_list; /* chain thru all services */
617 int srv_max_req_size; /* biggest request to receive */
618 int srv_max_reply_size; /* biggest reply to send */
619 int srv_buf_size; /* size of individual buffers */
620 int srv_nbuf_per_group; /* # buffers to allocate in 1 group */
621 int srv_nbufs; /* total # req buffer descs allocated */
622 int srv_threads_min; /* threads to start at SOW */
623 int srv_threads_max; /* thread upper limit */
624 int srv_threads_started; /* index of last started thread */
625 int srv_threads_running; /* # running threads */
626 int srv_n_difficult_replies; /* # 'difficult' replies */
627 int srv_n_active_reqs; /* # reqs being served */
628 cfs_duration_t srv_rqbd_timeout; /* timeout before re-posting reqs, in tick */
629 int srv_watchdog_factor; /* soft watchdog timeout mutiplier */
630 unsigned srv_cpu_affinity:1; /* bind threads to CPUs */
631 unsigned srv_at_check:1; /* check early replies */
632 cfs_time_t srv_at_checktime; /* debug */
634 __u32 srv_req_portal;
635 __u32 srv_rep_portal;
638 struct adaptive_timeout srv_at_estimate;/* estimated rpc service time */
639 spinlock_t srv_at_lock;
640 struct list_head srv_at_list; /* reqs waiting for replies */
641 cfs_timer_t srv_at_timer; /* early reply timer */
643 int srv_n_queued_reqs; /* # reqs in either of the queues below */
644 struct list_head srv_req_in_queue; /* incoming reqs */
645 struct list_head srv_request_queue; /* reqs waiting for service */
647 struct list_head srv_request_history; /* request history */
648 __u64 srv_request_seq; /* next request sequence # */
649 __u64 srv_request_max_cull_seq; /* highest seq culled from history */
650 svcreq_printfn_t srv_request_history_print_fn; /* service-specific print fn */
652 struct list_head srv_idle_rqbds; /* request buffers to be reposted */
653 struct list_head srv_active_rqbds; /* req buffers receiving */
654 struct list_head srv_history_rqbds; /* request buffer history */
655 int srv_nrqbd_receiving; /* # posted request buffers */
656 int srv_n_history_rqbds; /* # request buffers in history */
657 int srv_max_history_rqbds;/* max # request buffers in history */
659 atomic_t srv_outstanding_replies;
660 struct list_head srv_active_replies; /* all the active replies */
661 struct list_head srv_reply_queue; /* replies waiting for service */
663 cfs_waitq_t srv_waitq; /* all threads sleep on this. This
664 * wait-queue is signalled when new
665 * incoming request arrives and when
666 * difficult reply has to be handled. */
668 struct list_head srv_threads; /* service thread list */
669 svc_handler_t srv_handler;
671 char *srv_name; /* only statically allocated strings here; we don't clean them */
672 char *srv_thread_name; /* only statically allocated strings here; we don't clean them */
676 cfs_proc_dir_entry_t *srv_procroot;
677 struct lprocfs_stats *srv_stats;
679 /* List of free reply_states */
680 struct list_head srv_free_rs_list;
681 /* waitq to run, when adding stuff to srv_free_rs_list */
682 cfs_waitq_t srv_free_rs_waitq;
685 * Tags for lu_context associated with this thread, see struct
690 * if non-NULL called during thread creation (ptlrpc_start_thread())
691 * to initialize service specific per-thread state.
693 int (*srv_init)(struct ptlrpc_thread *thread);
695 * if non-NULL called during thread shutdown (ptlrpc_main()) to
696 * destruct state created by ->srv_init().
698 void (*srv_done)(struct ptlrpc_thread *thread);
700 //struct ptlrpc_srv_ni srv_interfaces[0];
703 /* ptlrpc/events.c */
704 extern lnet_handle_eq_t ptlrpc_eq_h;
705 extern int ptlrpc_uuid_to_peer(struct obd_uuid *uuid,
706 lnet_process_id_t *peer, lnet_nid_t *self);
707 extern void request_out_callback (lnet_event_t *ev);
708 extern void reply_in_callback(lnet_event_t *ev);
709 extern void client_bulk_callback (lnet_event_t *ev);
710 extern void request_in_callback(lnet_event_t *ev);
711 extern void reply_out_callback(lnet_event_t *ev);
712 extern void server_bulk_callback (lnet_event_t *ev);
714 /* ptlrpc/connection.c */
715 void ptlrpc_dump_connections(void);
716 void ptlrpc_readdress_connection(struct ptlrpc_connection *, struct obd_uuid *);
717 struct ptlrpc_connection *ptlrpc_get_connection(lnet_process_id_t peer,
718 lnet_nid_t self, struct obd_uuid *uuid);
719 int ptlrpc_put_connection(struct ptlrpc_connection *c);
720 struct ptlrpc_connection *ptlrpc_connection_addref(struct ptlrpc_connection *);
721 int ptlrpc_init_connection(void);
722 void ptlrpc_cleanup_connection(void);
723 extern lnet_pid_t ptl_get_pid(void);
725 /* ptlrpc/niobuf.c */
726 int ptlrpc_start_bulk_transfer(struct ptlrpc_bulk_desc *desc);
727 void ptlrpc_abort_bulk(struct ptlrpc_bulk_desc *desc);
728 int ptlrpc_register_bulk(struct ptlrpc_request *req);
729 void ptlrpc_unregister_bulk (struct ptlrpc_request *req);
731 static inline int ptlrpc_bulk_active (struct ptlrpc_bulk_desc *desc)
735 spin_lock(&desc->bd_lock);
736 rc = desc->bd_network_rw;
737 spin_unlock(&desc->bd_lock);
741 #define PTLRPC_REPLY_MAYBE_DIFFICULT 0x01
742 #define PTLRPC_REPLY_EARLY 0x02
743 int ptlrpc_send_reply(struct ptlrpc_request *req, int flags);
744 int ptlrpc_reply(struct ptlrpc_request *req);
745 int ptlrpc_send_error(struct ptlrpc_request *req, int difficult);
746 int ptlrpc_error(struct ptlrpc_request *req);
747 void ptlrpc_resend_req(struct ptlrpc_request *request);
748 int ptlrpc_at_get_net_latency(struct ptlrpc_request *req);
749 int ptl_send_rpc(struct ptlrpc_request *request, int noreply);
750 int ptlrpc_register_rqbd (struct ptlrpc_request_buffer_desc *rqbd);
752 /* ptlrpc/client.c */
753 void ptlrpc_init_client(int req_portal, int rep_portal, char *name,
754 struct ptlrpc_client *);
755 void ptlrpc_cleanup_client(struct obd_import *imp);
756 struct ptlrpc_connection *ptlrpc_uuid_to_connection(struct obd_uuid *uuid);
759 ptlrpc_client_recv_or_unlink (struct ptlrpc_request *req)
763 spin_lock(&req->rq_lock);
764 rc = req->rq_receiving_reply || req->rq_must_unlink;
765 spin_unlock(&req->rq_lock);
770 ptlrpc_wake_client_req (struct ptlrpc_request *req)
772 if (req->rq_set == NULL)
773 cfs_waitq_signal(&req->rq_reply_waitq);
775 cfs_waitq_signal(&req->rq_set->set_waitq);
778 int ptlrpc_queue_wait(struct ptlrpc_request *req);
779 int ptlrpc_replay_req(struct ptlrpc_request *req);
780 void ptlrpc_unregister_reply(struct ptlrpc_request *req);
781 void ptlrpc_restart_req(struct ptlrpc_request *req);
782 void ptlrpc_abort_inflight(struct obd_import *imp);
784 struct ptlrpc_request_set *ptlrpc_prep_set(void);
785 int ptlrpc_set_add_cb(struct ptlrpc_request_set *set,
786 set_interpreter_func fn, void *data);
787 int ptlrpc_set_next_timeout(struct ptlrpc_request_set *);
788 int ptlrpc_check_set(struct ptlrpc_request_set *set);
789 int ptlrpc_set_wait(struct ptlrpc_request_set *);
790 int ptlrpc_expired_set(void *data);
791 void ptlrpc_interrupted_set(void *data);
792 void ptlrpc_mark_interrupted(struct ptlrpc_request *req);
793 void ptlrpc_set_destroy(struct ptlrpc_request_set *);
794 void ptlrpc_set_add_req(struct ptlrpc_request_set *, struct ptlrpc_request *);
795 void ptlrpc_set_add_new_req(struct ptlrpc_request_set *,
796 struct ptlrpc_request *);
798 void ptlrpc_free_rq_pool(struct ptlrpc_request_pool *pool);
799 void ptlrpc_add_rqs_to_pool(struct ptlrpc_request_pool *pool, int num_rq);
800 struct ptlrpc_request_pool *ptlrpc_init_rq_pool(int, int,
801 void (*populate_pool)(struct ptlrpc_request_pool *, int));
802 void ptlrpc_at_set_req_timeout(struct ptlrpc_request *req);
803 struct ptlrpc_request *ptlrpc_request_alloc(struct obd_import *imp,
804 const struct req_format *format);
805 struct ptlrpc_request *ptlrpc_request_alloc_pool(struct obd_import *imp,
806 struct ptlrpc_request_pool *,
807 const struct req_format *format);
808 void ptlrpc_request_free(struct ptlrpc_request *request);
809 int ptlrpc_request_pack(struct ptlrpc_request *request,
810 __u32 version, int opcode);
811 struct ptlrpc_request *ptlrpc_request_alloc_pack(struct obd_import *imp,
812 const struct req_format *format,
813 __u32 version, int opcode);
814 int ptlrpc_request_bufs_pack(struct ptlrpc_request *request,
815 __u32 version, int opcode, char **bufs,
816 struct ptlrpc_cli_ctx *ctx);
817 struct ptlrpc_request *ptlrpc_prep_req(struct obd_import *imp, __u32 version,
818 int opcode, int count, int *lengths,
820 struct ptlrpc_request *ptlrpc_prep_req_pool(struct obd_import *imp,
821 __u32 version, int opcode,
822 int count, int *lengths, char **bufs,
823 struct ptlrpc_request_pool *pool);
824 void ptlrpc_free_req(struct ptlrpc_request *request);
825 void ptlrpc_req_finished(struct ptlrpc_request *request);
826 void ptlrpc_req_finished_with_imp_lock(struct ptlrpc_request *request);
827 struct ptlrpc_request *ptlrpc_request_addref(struct ptlrpc_request *req);
828 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_imp (struct ptlrpc_request *req,
829 int npages, int type, int portal);
830 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_exp(struct ptlrpc_request *req,
831 int npages, int type, int portal);
832 void ptlrpc_free_bulk(struct ptlrpc_bulk_desc *bulk);
833 void ptlrpc_prep_bulk_page(struct ptlrpc_bulk_desc *desc,
834 cfs_page_t *page, int pageoffset, int len);
835 void ptlrpc_retain_replayable_request(struct ptlrpc_request *req,
836 struct obd_import *imp);
837 __u64 ptlrpc_next_xid(void);
838 __u64 ptlrpc_sample_next_xid(void);
839 __u64 ptlrpc_req_xid(struct ptlrpc_request *request);
841 struct ptlrpc_service_conf {
844 int psc_max_req_size;
845 int psc_max_reply_size;
848 int psc_watchdog_factor;
854 /* ptlrpc/service.c */
855 void ptlrpc_save_lock (struct ptlrpc_request *req,
856 struct lustre_handle *lock, int mode);
857 void ptlrpc_commit_replies (struct obd_device *obd);
858 void ptlrpc_schedule_difficult_reply (struct ptlrpc_reply_state *rs);
859 struct ptlrpc_service *ptlrpc_init_svc_conf(struct ptlrpc_service_conf *c,
860 svc_handler_t h, char *name,
861 struct proc_dir_entry *proc_entry,
862 svcreq_printfn_t prntfn,
865 struct ptlrpc_service *ptlrpc_init_svc(int nbufs, int bufsize, int max_req_size,
867 int req_portal, int rep_portal,
869 svc_handler_t, char *name,
870 cfs_proc_dir_entry_t *proc_entry,
872 int min_threads, int max_threads,
873 char *threadname, __u32 ctx_tags);
874 void ptlrpc_stop_all_threads(struct ptlrpc_service *svc);
876 int ptlrpc_start_threads(struct obd_device *dev, struct ptlrpc_service *svc);
877 int ptlrpc_start_thread(struct obd_device *dev, struct ptlrpc_service *svc);
878 int ptlrpc_unregister_service(struct ptlrpc_service *service);
879 int liblustre_check_services (void *arg);
880 void ptlrpc_daemonize(char *name);
881 int ptlrpc_service_health_check(struct ptlrpc_service *);
884 struct ptlrpc_svc_data {
886 struct ptlrpc_service *svc;
887 struct ptlrpc_thread *thread;
888 struct obd_device *dev;
891 /* ptlrpc/import.c */
892 int ptlrpc_connect_import(struct obd_import *imp, char * new_uuid);
893 int ptlrpc_init_import(struct obd_import *imp);
894 int ptlrpc_disconnect_import(struct obd_import *imp, int noclose);
895 int ptlrpc_import_recovery_state_machine(struct obd_import *imp);
897 /* ptlrpc/pack_generic.c */
898 int ptlrpc_reconnect_import(struct obd_import *imp);
899 int lustre_msg_swabbed(struct lustre_msg *msg);
900 int lustre_msg_check_version(struct lustre_msg *msg, __u32 version);
901 void lustre_init_msg_v2(struct lustre_msg_v2 *msg, int count, int *lens,
903 int lustre_pack_request(struct ptlrpc_request *, __u32 magic, int count,
904 int *lens, char **bufs);
905 int lustre_pack_reply(struct ptlrpc_request *, int count, int *lens,
907 int lustre_pack_reply_v2(struct ptlrpc_request *req, int count,
908 int *lens, char **bufs, int flags);
909 #define LPRFL_EARLY_REPLY 1
910 int lustre_pack_reply_flags(struct ptlrpc_request *, int count, int *lens,
911 char **bufs, int flags);
912 int lustre_shrink_msg(struct lustre_msg *msg, int segment,
913 unsigned int newlen, int move_data);
914 void lustre_free_reply_state(struct ptlrpc_reply_state *rs);
915 int lustre_msg_hdr_size(__u32 magic, int count);
916 int lustre_msg_size(__u32 magic, int count, int *lengths);
917 int lustre_msg_size_v2(int count, int *lengths);
918 int lustre_packed_msg_size(struct lustre_msg *msg);
919 int lustre_msg_early_size(void);
920 int lustre_unpack_msg(struct lustre_msg *m, int len);
921 void *lustre_msg_buf_v2(struct lustre_msg_v2 *m, int n, int min_size);
922 void *lustre_msg_buf(struct lustre_msg *m, int n, int minlen);
923 int lustre_msg_buflen(struct lustre_msg *m, int n);
924 void lustre_msg_set_buflen(struct lustre_msg *m, int n, int len);
925 int lustre_msg_bufcount(struct lustre_msg *m);
926 char *lustre_msg_string (struct lustre_msg *m, int n, int max_len);
927 void *lustre_swab_buf(struct lustre_msg *, int n, int minlen, void *swabber);
928 void *lustre_swab_reqbuf(struct ptlrpc_request *req, int n, int minlen,
930 void *lustre_swab_repbuf(struct ptlrpc_request *req, int n, int minlen,
932 __u32 lustre_msghdr_get_flags(struct lustre_msg *msg);
933 void lustre_msghdr_set_flags(struct lustre_msg *msg, __u32 flags);
934 __u32 lustre_msg_get_flags(struct lustre_msg *msg);
935 void lustre_msg_add_flags(struct lustre_msg *msg, int flags);
936 void lustre_msg_set_flags(struct lustre_msg *msg, int flags);
937 void lustre_msg_clear_flags(struct lustre_msg *msg, int flags);
938 __u32 lustre_msg_get_op_flags(struct lustre_msg *msg);
939 void lustre_msg_add_op_flags(struct lustre_msg *msg, int flags);
940 void lustre_msg_set_op_flags(struct lustre_msg *msg, int flags);
941 struct lustre_handle *lustre_msg_get_handle(struct lustre_msg *msg);
942 __u32 lustre_msg_get_type(struct lustre_msg *msg);
943 __u32 lustre_msg_get_version(struct lustre_msg *msg);
944 void lustre_msg_add_version(struct lustre_msg *msg, int version);
945 __u32 lustre_msg_get_opc(struct lustre_msg *msg);
946 __u64 lustre_msg_get_last_xid(struct lustre_msg *msg);
947 __u64 lustre_msg_get_last_committed(struct lustre_msg *msg);
948 __u64 lustre_msg_get_transno(struct lustre_msg *msg);
949 __u64 lustre_msg_get_slv(struct lustre_msg *msg);
950 __u32 lustre_msg_get_limit(struct lustre_msg *msg);
951 void lustre_msg_set_slv(struct lustre_msg *msg, __u64 slv);
952 void lustre_msg_set_limit(struct lustre_msg *msg, __u64 limit);
953 int lustre_msg_get_status(struct lustre_msg *msg);
954 __u32 lustre_msg_get_conn_cnt(struct lustre_msg *msg);
955 int lustre_msg_is_v1(struct lustre_msg *msg);
956 __u32 lustre_msg_get_magic(struct lustre_msg *msg);
957 __u32 lustre_msg_get_timeout(struct lustre_msg *msg);
958 __u32 lustre_msg_get_service_time(struct lustre_msg *msg);
959 __u32 lustre_msg_get_cksum(struct lustre_msg *msg);
960 __u32 lustre_msg_calc_cksum(struct lustre_msg *msg);
961 void lustre_msg_set_handle(struct lustre_msg *msg,struct lustre_handle *handle);
962 void lustre_msg_set_type(struct lustre_msg *msg, __u32 type);
963 void lustre_msg_set_opc(struct lustre_msg *msg, __u32 opc);
964 void lustre_msg_set_last_xid(struct lustre_msg *msg, __u64 last_xid);
965 void lustre_msg_set_last_committed(struct lustre_msg *msg,__u64 last_committed);
966 void lustre_msg_set_transno(struct lustre_msg *msg, __u64 transno);
967 void lustre_msg_set_status(struct lustre_msg *msg, __u32 status);
968 void lustre_msg_set_conn_cnt(struct lustre_msg *msg, __u32 conn_cnt);
969 void ptlrpc_req_set_repsize(struct ptlrpc_request *req, int count, int *sizes);
970 void ptlrpc_request_set_replen(struct ptlrpc_request *req);
971 void lustre_msg_set_timeout(struct lustre_msg *msg, __u32 timeout);
972 void lustre_msg_set_service_time(struct lustre_msg *msg, __u32 service_time);
973 void lustre_msg_set_cksum(struct lustre_msg *msg, __u32 cksum);
976 lustre_shrink_reply(struct ptlrpc_request *req, int segment,
977 unsigned int newlen, int move_data)
979 LASSERT(req->rq_reply_state);
980 LASSERT(req->rq_repmsg);
981 req->rq_replen = lustre_shrink_msg(req->rq_repmsg, segment,
986 ptlrpc_rs_addref(struct ptlrpc_reply_state *rs)
988 LASSERT(atomic_read(&rs->rs_refcount) > 0);
989 atomic_inc(&rs->rs_refcount);
993 ptlrpc_rs_decref(struct ptlrpc_reply_state *rs)
995 LASSERT(atomic_read(&rs->rs_refcount) > 0);
996 if (atomic_dec_and_test(&rs->rs_refcount))
997 lustre_free_reply_state(rs);
1000 /* Should only be called once per req */
1001 static inline void ptlrpc_req_drop_rs(struct ptlrpc_request *req)
1003 if (req->rq_reply_state == NULL)
1004 return; /* shouldn't occur */
1005 ptlrpc_rs_decref(req->rq_reply_state);
1006 req->rq_reply_state = NULL;
1007 req->rq_repmsg = NULL;
1010 static inline __u32 lustre_request_magic(struct ptlrpc_request *req)
1012 return lustre_msg_get_magic(req->rq_reqmsg);
1015 static inline int ptlrpc_req_get_repsize(struct ptlrpc_request *req)
1017 switch (req->rq_reqmsg->lm_magic) {
1018 case LUSTRE_MSG_MAGIC_V2:
1019 return req->rq_reqmsg->lm_repsize;
1021 LASSERTF(0, "incorrect message magic: %08x\n",
1022 req->rq_reqmsg->lm_magic);
1027 /* ldlm/ldlm_lib.c */
1028 int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg);
1029 int client_obd_cleanup(struct obd_device *obddev);
1030 int client_connect_import(const struct lu_env *env,
1031 struct lustre_handle *conn, struct obd_device *obd,
1032 struct obd_uuid *cluuid, struct obd_connect_data *,
1034 int client_disconnect_export(struct obd_export *exp);
1035 int client_import_add_conn(struct obd_import *imp, struct obd_uuid *uuid,
1037 int client_import_del_conn(struct obd_import *imp, struct obd_uuid *uuid);
1038 int import_set_conn_priority(struct obd_import *imp, struct obd_uuid *uuid);
1040 /* ptlrpc/pinger.c */
1041 int ptlrpc_pinger_add_import(struct obd_import *imp);
1042 int ptlrpc_pinger_del_import(struct obd_import *imp);
1043 cfs_time_t ptlrpc_suspend_wakeup_time(void);
1045 void ping_evictor_start(void);
1046 void ping_evictor_stop(void);
1048 #define ping_evictor_start() do {} while (0)
1049 #define ping_evictor_stop() do {} while (0)
1051 int ptlrpc_check_and_wait_suspend(struct ptlrpc_request *req);
1053 /* ptlrpc/ptlrpcd.c */
1054 void ptlrpcd_wake(struct ptlrpc_request *req);
1055 void ptlrpcd_add_req(struct ptlrpc_request *req);
1056 int ptlrpcd_addref(void);
1057 void ptlrpcd_decref(void);
1059 /* ptlrpc/lproc_ptlrpc.c */
1060 const char* ll_opcode2str(__u32 opcode);
1062 void ptlrpc_lprocfs_register_obd(struct obd_device *obd);
1063 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd);
1064 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int opc, int bytes);
1066 static inline void ptlrpc_lprocfs_register_obd(struct obd_device *obd) {}
1067 static inline void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd) {}
1068 static inline void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int opc,
1072 /* ptlrpc/llog_server.c */
1073 int llog_origin_handle_create(struct ptlrpc_request *req);
1074 int llog_origin_handle_destroy(struct ptlrpc_request *req);
1075 int llog_origin_handle_prev_block(struct ptlrpc_request *req);
1076 int llog_origin_handle_next_block(struct ptlrpc_request *req);
1077 int llog_origin_handle_read_header(struct ptlrpc_request *req);
1078 int llog_origin_handle_close(struct ptlrpc_request *req);
1079 int llog_origin_handle_cancel(struct ptlrpc_request *req);
1080 int llog_catinfo(struct ptlrpc_request *req);
1082 /* ptlrpc/llog_client.c */
1083 extern struct llog_operations llog_client_ops;