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 \
91 min((int)(num_online_cpus() * num_online_cpus() * 2), 8)
92 #define LDLM_THREADS_AUTO_MAX (LDLM_THREADS_AUTO_MIN * 16)
93 #define LDLM_BL_THREADS LDLM_THREADS_AUTO_MIN
94 #define LDLM_NBUFS (64 * num_online_cpus())
95 #define LDLM_BUFSIZE (8 * 1024)
96 #define LDLM_MAXREQSIZE (5 * 1024)
97 #define LDLM_MAXREPSIZE (1024)
99 #define MDT_MIN_THREADS 2UL
100 #define MDT_MAX_THREADS 512UL
101 #define MDT_NUM_THREADS max(min_t(unsigned long, MDT_MAX_THREADS, \
102 num_physpages >> (25 - PAGE_SHIFT)), 2UL)
103 #define FLD_NUM_THREADS max(min_t(unsigned long, MDT_MAX_THREADS, \
104 num_physpages >> (25 - PAGE_SHIFT)), 2UL)
105 #define SEQ_NUM_THREADS max(min_t(unsigned long, MDT_MAX_THREADS, \
106 num_physpages >> (25 - PAGE_SHIFT)), 2UL)
108 /* Absolute limits */
109 #define MDS_THREADS_MIN 2
110 #define MDS_THREADS_MAX 512
111 #define MDS_THREADS_MIN_READPAGE 2
112 #define MDS_NBUFS (64 * num_online_cpus())
113 #define MDS_BUFSIZE (8 * 1024)
114 /* Assume file name length = FNAME_MAX = 256 (true for ext3).
115 * path name length = PATH_MAX = 4096
116 * LOV MD size max = EA_MAX = 4000
117 * symlink: FNAME_MAX + PATH_MAX <- largest
118 * link: FNAME_MAX + PATH_MAX (mds_rec_link < mds_rec_create)
119 * rename: FNAME_MAX + FNAME_MAX
120 * open: FNAME_MAX + EA_MAX
122 * MDS_MAXREQSIZE ~= 4736 bytes =
123 * lustre_msg + ldlm_request + mds_body + mds_rec_create + FNAME_MAX + PATH_MAX
124 * MDS_MAXREPSIZE ~= 8300 bytes = lustre_msg + llog_header
125 * or, for mds_close() and mds_reint_unlink() on a many-OST filesystem:
126 * = 9210 bytes = lustre_msg + mds_body + 160 * (easize + cookiesize)
128 * Realistic size is about 512 bytes (20 character name + 128 char symlink),
129 * except in the open case where there are a large number of OSTs in a LOV.
131 #define MDS_MAXREQSIZE (5 * 1024)
132 #define MDS_MAXREPSIZE max(9 * 1024, 280 + LOV_MAX_STRIPE_COUNT * 56)
134 /* FLD_MAXREQSIZE == lustre_msg + __u32 padding + ptlrpc_body + opc + md_fld */
135 #define FLD_MAXREQSIZE (160)
137 /* FLD_MAXREPSIZE == lustre_msg + ptlrpc_body + md_fld */
138 #define FLD_MAXREPSIZE (152)
140 /* SEQ_MAXREQSIZE == lustre_msg + __u32 padding + ptlrpc_body + opc + lu_range +
142 #define SEQ_MAXREQSIZE (160)
144 /* SEQ_MAXREPSIZE == lustre_msg + ptlrpc_body + lu_range */
145 #define SEQ_MAXREPSIZE (152)
147 #define MGS_THREADS_AUTO_MIN 2
148 #define MGS_THREADS_AUTO_MAX 32
149 #define MGS_NBUFS (64 * num_online_cpus())
150 #define MGS_BUFSIZE (8 * 1024)
151 #define MGS_MAXREQSIZE (7 * 1024)
152 #define MGS_MAXREPSIZE (9 * 1024)
154 /* Absolute limits */
155 #define OSS_THREADS_MIN 2
156 #define OSS_THREADS_MAX 512
157 #define OST_NBUFS (64 * num_online_cpus())
158 #define OST_BUFSIZE (8 * 1024)
159 /* OST_MAXREQSIZE ~= 4768 bytes =
160 * lustre_msg + obdo + 16 * obd_ioobj + 256 * niobuf_remote
162 * - single object with 16 pages is 512 bytes
163 * - OST_MAXREQSIZE must be at least 1 page of cookies plus some spillover
165 #define OST_MAXREQSIZE (5 * 1024)
166 #define OST_MAXREPSIZE (9 * 1024)
168 struct ptlrpc_connection {
169 struct list_head c_link;
170 struct hlist_node c_hash;
172 lnet_process_id_t c_peer;
173 struct obd_uuid c_remote_uuid;
177 struct ptlrpc_client {
178 __u32 cli_request_portal;
179 __u32 cli_reply_portal;
183 /* state flags of requests */
184 /* XXX only ones left are those used by the bulk descs as well! */
185 #define PTL_RPC_FL_INTR (1 << 0) /* reply wait was interrupted by user */
186 #define PTL_RPC_FL_TIMEOUT (1 << 7) /* request timed out waiting for reply */
188 #define REQ_MAX_ACK_LOCKS 8
190 union ptlrpc_async_args {
191 /* Scratchpad for passing args to completion interpreter. Users
192 * cast to the struct of their choosing, and LASSERT that this is
193 * big enough. For _tons_ of context, OBD_ALLOC a struct and store
194 * a pointer to it here. The pointer_arg ensures this struct is at
195 * least big enough for that. */
196 void *pointer_arg[9];
200 struct ptlrpc_request_set;
201 typedef int (*set_interpreter_func)(struct ptlrpc_request_set *, void *, int);
203 struct ptlrpc_request_set {
204 int set_remaining; /* # uncompleted requests */
205 cfs_waitq_t set_waitq;
206 cfs_waitq_t *set_wakeup_ptr;
207 struct list_head set_requests;
208 set_interpreter_func set_interpret; /* completion callback */
209 void *set_arg; /* completion context */
210 void *set_countp; /* pointer to NOB counter in case
211 * of directIO (bug11737) */
212 /* locked so that any old caller can communicate requests to
213 * the set holder who can then fold them into the lock-free set */
214 spinlock_t set_new_req_lock;
215 struct list_head set_new_requests;
218 struct ptlrpc_bulk_desc;
221 * ptlrpc callback & work item stuff
223 struct ptlrpc_cb_id {
224 void (*cbid_fn)(lnet_event_t *ev); /* specific callback fn */
225 void *cbid_arg; /* additional arg */
228 #define RS_MAX_LOCKS 4
231 struct ptlrpc_reply_state {
232 struct ptlrpc_cb_id rs_cb_id;
233 struct list_head rs_list;
234 struct list_head rs_exp_list;
235 struct list_head rs_obd_list;
237 struct list_head rs_debug_list;
239 /* updates to following flag serialised by srv_request_lock */
240 unsigned int rs_difficult:1; /* ACK/commit stuff */
241 unsigned int rs_scheduled:1; /* being handled? */
242 unsigned int rs_scheduled_ever:1;/* any schedule attempts? */
243 unsigned int rs_handled:1; /* been handled yet? */
244 unsigned int rs_on_net:1; /* reply_out_callback pending? */
245 unsigned int rs_prealloc:1; /* rs from prealloc list */
250 struct obd_export *rs_export;
251 struct ptlrpc_service *rs_service;
252 lnet_handle_md_t rs_md_h;
253 atomic_t rs_refcount;
255 struct ptlrpc_svc_ctx *rs_svc_ctx;
256 struct lustre_msg *rs_repbuf; /* wrapper */
257 int rs_repbuf_len; /* wrapper buf length */
258 int rs_repdata_len; /* wrapper msg length */
259 struct lustre_msg *rs_msg; /* reply message */
261 /* locks awaiting client reply ACK */
263 struct lustre_handle rs_locks[RS_MAX_LOCKS];
264 ldlm_mode_t rs_modes[RS_MAX_LOCKS];
267 struct ptlrpc_thread;
270 RQ_PHASE_NEW = 0xebc0de00,
271 RQ_PHASE_RPC = 0xebc0de01,
272 RQ_PHASE_BULK = 0xebc0de02,
273 RQ_PHASE_INTERPRET = 0xebc0de03,
274 RQ_PHASE_COMPLETE = 0xebc0de04,
277 struct ptlrpc_request_pool {
279 struct list_head prp_req_list; /* list of ptlrpc_request structs */
281 void (*prp_populate)(struct ptlrpc_request_pool *, int);
287 struct ptlrpc_request {
288 int rq_type; /* one of PTL_RPC_MSG_* */
289 struct list_head rq_list;
290 struct list_head rq_history_list; /* server-side history */
291 __u64 rq_history_seq; /* history sequence # */
294 /* client-side flags */
295 unsigned int rq_intr:1, rq_replied:1, rq_err:1,
296 rq_timedout:1, rq_resend:1, rq_restart:1,
298 * when ->rq_replay is set, request is kept by the client even
299 * after server commits corresponding transaction. This is
300 * used for operations that require sequence of multiple
301 * requests to be replayed. The only example currently is file
302 * open/close/dw/setattr. When last request in such a sequence
303 * is committed, ->rq_replay is cleared on all requests in the
307 /* this is the last request in the sequence. */
309 rq_no_resend:1, rq_waiting:1, rq_receiving_reply:1,
310 rq_no_delay:1, rq_net_err:1, rq_wait_ctx:1;
311 enum rq_phase rq_phase; /* one of RQ_PHASE_* */
312 atomic_t rq_refcount; /* client-side refcount for SENT race */
314 struct ptlrpc_thread *rq_svc_thread; /* initial thread servicing req */
316 int rq_request_portal; /* XXX FIXME bug 249 */
317 int rq_reply_portal; /* XXX FIXME bug 249 */
319 int rq_nob_received; /* client-side # reply bytes actually received */
322 struct lustre_msg *rq_reqmsg;
324 int rq_timeout; /* time to wait for reply (seconds) */
326 struct lustre_msg *rq_repmsg;
329 struct list_head rq_replay_list;
330 struct list_head rq_mod_list;
332 struct ptlrpc_cli_ctx *rq_cli_ctx; /* client's half ctx */
333 struct ptlrpc_svc_ctx *rq_svc_ctx; /* server's half ctx */
334 struct list_head rq_ctx_chain; /* link to waited ctx */
335 ptlrpc_sec_flavor_t rq_sec_flavor; /* client & server */
336 /* client security flags */
337 unsigned int rq_ctx_init:1, /* context initiation */
338 rq_ctx_fini:1, /* context destroy */
339 rq_bulk_read:1, /* request bulk read */
340 rq_bulk_write:1, /* request bulk write */
341 /* server authentication flags */
342 rq_auth_gss:1, /* authenticated by gss */
343 rq_auth_remote:1, /* authed as remote user */
344 rq_auth_usr_root:1, /* authed as root */
345 rq_auth_usr_mdt:1, /* authed as mdt */
346 /* doesn't expect reply FIXME */
349 uid_t rq_auth_uid; /* authed uid */
350 uid_t rq_auth_mapped_uid; /* authed uid mapped to */
352 /* (server side), pointed directly into req buffer */
353 struct ptlrpc_user_desc *rq_user_desc;
355 /* various buffer pointers */
356 struct lustre_msg *rq_reqbuf; /* req wrapper */
357 int rq_reqbuf_len; /* req wrapper buf len */
358 int rq_reqdata_len; /* req wrapper msg len */
359 struct lustre_msg *rq_repbuf; /* rep wrapper */
360 int rq_repbuf_len; /* rep wrapper buf len */
361 int rq_repdata_len; /* rep wrapper msg len */
362 struct lustre_msg *rq_clrbuf; /* only in priv mode */
363 int rq_clrbuf_len; /* only in priv mode */
364 int rq_clrdata_len; /* only in priv mode */
366 __u32 rq_req_swab_mask;
367 __u32 rq_rep_swab_mask;
369 int rq_import_generation;
370 enum lustre_imp_state rq_send_state;
372 /* client+server request */
373 lnet_handle_md_t rq_req_md_h;
374 struct ptlrpc_cb_id rq_req_cbid;
377 struct timeval rq_arrival_time; /* request arrival time */
378 struct ptlrpc_reply_state *rq_reply_state; /* separated reply state */
379 struct ptlrpc_request_buffer_desc *rq_rqbd; /* incoming request buffer*/
381 __u32 rq_uid; /* peer uid, used in MDS only */
384 /* client-only incoming reply */
385 lnet_handle_md_t rq_reply_md_h;
386 cfs_waitq_t rq_reply_waitq;
387 struct ptlrpc_cb_id rq_reply_cbid;
390 lnet_process_id_t rq_peer;
391 struct obd_export *rq_export;
392 struct obd_import *rq_import;
394 void (*rq_replay_cb)(struct ptlrpc_request *);
395 void (*rq_commit_cb)(struct ptlrpc_request *);
398 struct ptlrpc_bulk_desc *rq_bulk; /* client side bulk */
399 time_t rq_sent; /* when request sent, seconds,
400 * or time when request should
403 struct list_head rq_set_chain;
404 struct ptlrpc_request_set *rq_set;
405 void *rq_interpret_reply; /* Async completion handler */
406 union ptlrpc_async_args rq_async_args; /* Async completion context */
407 struct ptlrpc_request_pool *rq_pool; /* Pool if request from
409 struct lu_context rq_session;
412 static inline void ptlrpc_close_replay_seq(struct ptlrpc_request *req)
414 spin_lock(&req->rq_lock);
416 req->rq_sequence = 1;
417 spin_unlock(&req->rq_lock);
420 static inline void lustre_set_req_swabbed(struct ptlrpc_request *req, int index)
422 LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
423 LASSERT((req->rq_req_swab_mask & (1 << index)) == 0);
424 req->rq_req_swab_mask |= 1 << index;
427 static inline void lustre_set_rep_swabbed(struct ptlrpc_request *req, int index)
429 LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
430 LASSERT((req->rq_rep_swab_mask & (1 << index)) == 0);
431 req->rq_rep_swab_mask |= 1 << index;
434 static inline int lustre_req_swabbed(struct ptlrpc_request *req, int index)
436 LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
437 return req->rq_req_swab_mask & (1 << index);
440 static inline int lustre_rep_swabbed(struct ptlrpc_request *req, int index)
442 LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
443 return req->rq_rep_swab_mask & (1 << index);
446 static inline const char *
447 ptlrpc_rqphase2str(const struct ptlrpc_request *req)
449 switch (req->rq_phase) {
456 case RQ_PHASE_INTERPRET:
458 case RQ_PHASE_COMPLETE:
465 /* Spare the preprocessor, spoil the bugs. */
466 #define FLAG(field, str) (field ? str : "")
468 #define DEBUG_REQ_FLAGS(req) \
469 ptlrpc_rqphase2str(req), \
470 FLAG(req->rq_intr, "I"), FLAG(req->rq_replied, "R"), \
471 FLAG(req->rq_err, "E"), \
472 FLAG(req->rq_timedout, "X") /* eXpired */, FLAG(req->rq_resend, "S"), \
473 FLAG(req->rq_restart, "T"), FLAG(req->rq_replay, "P"), \
474 FLAG(req->rq_no_resend, "N"), \
475 FLAG(req->rq_waiting, "W"), \
476 FLAG(req->rq_wait_ctx, "C")
478 #define REQ_FLAGS_FMT "%s:%s%s%s%s%s%s%s%s%s%s"
480 void _debug_req(struct ptlrpc_request *req, __u32 mask,
481 struct libcfs_debug_msg_data *data, const char *fmt, ...)
482 __attribute__ ((format (printf, 4, 5)));
484 #define debug_req(cdls, level, req, file, func, line, fmt, a...) \
488 if (((level) & D_CANTMASK) != 0 || \
489 ((libcfs_debug & (level)) != 0 && \
490 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) { \
491 static struct libcfs_debug_msg_data _req_dbg_data = \
492 DEBUG_MSG_DATA_INIT(cdls, DEBUG_SUBSYSTEM, file, func, line); \
493 _debug_req((req), (level), &_req_dbg_data, fmt, ##a); \
497 /* for most callers (level is a constant) this is resolved at compile time */
498 #define DEBUG_REQ(level, req, fmt, args...) \
500 if ((level) & (D_ERROR | D_WARNING)) { \
501 static cfs_debug_limit_state_t cdls; \
502 debug_req(&cdls, level, req, __FILE__, __func__, __LINE__, \
503 "@@@ "fmt" ", ## args); \
505 debug_req(NULL, level, req, __FILE__, __func__, __LINE__, \
506 "@@@ "fmt" ", ## args); \
509 struct ptlrpc_bulk_page {
510 struct list_head bp_link;
512 int bp_pageoffset; /* offset within a page */
513 struct page *bp_page;
516 #define BULK_GET_SOURCE 0
517 #define BULK_PUT_SINK 1
518 #define BULK_GET_SINK 2
519 #define BULK_PUT_SOURCE 3
521 struct ptlrpc_bulk_desc {
522 unsigned int bd_success:1; /* completed successfully */
523 unsigned int bd_network_rw:1; /* accessible to the network */
524 unsigned int bd_type:2; /* {put,get}{source,sink} */
525 unsigned int bd_registered:1; /* client side */
526 spinlock_t bd_lock; /* serialise with callback */
527 int bd_import_generation;
528 struct obd_export *bd_export;
529 struct obd_import *bd_import;
531 struct ptlrpc_request *bd_req; /* associated request */
532 cfs_waitq_t bd_waitq; /* server side only WQ */
533 int bd_iov_count; /* # entries in bd_iov */
534 int bd_max_iov; /* allocated size of bd_iov */
535 int bd_nob; /* # bytes covered */
536 int bd_nob_transferred; /* # bytes GOT/PUT */
540 struct ptlrpc_cb_id bd_cbid; /* network callback info */
541 lnet_handle_md_t bd_md_h; /* associated MD */
542 lnet_nid_t bd_sender; /* stash event::sender */
544 cfs_page_t **bd_enc_pages;
545 #if defined(__KERNEL__)
546 lnet_kiov_t bd_iov[0];
548 lnet_md_iovec_t bd_iov[0];
552 struct ptlrpc_thread {
554 struct list_head t_link; /* active threads for service, from svc->srv_threads */
556 void *t_data; /* thread-private data (preallocated memory) */
559 unsigned int t_id; /* service thread index, from ptlrpc_start_threads */
560 cfs_waitq_t t_ctl_waitq;
561 struct lu_env *t_env;
564 struct ptlrpc_request_buffer_desc {
565 struct list_head rqbd_list;
566 struct list_head rqbd_reqs;
567 struct ptlrpc_service *rqbd_service;
568 lnet_handle_md_t rqbd_md_h;
571 struct ptlrpc_cb_id rqbd_cbid;
572 struct ptlrpc_request rqbd_req;
575 typedef int (*svc_handler_t)(struct ptlrpc_request *req);
576 typedef void (*svcreq_printfn_t)(void *, struct ptlrpc_request *);
578 struct ptlrpc_service {
579 struct list_head srv_list; /* chain thru all services */
580 int srv_max_req_size; /* biggest request to receive */
581 int srv_max_reply_size; /* biggest reply to send */
582 int srv_buf_size; /* size of individual buffers */
583 int srv_nbuf_per_group; /* # buffers to allocate in 1 group */
584 int srv_nbufs; /* total # req buffer descs allocated */
585 int srv_threads_min; /* threads to start at SOW */
586 int srv_threads_max; /* thread upper limit */
587 int srv_threads_started; /* index of last started thread */
588 int srv_threads_running; /* # running threads */
589 int srv_n_difficult_replies; /* # 'difficult' replies */
590 int srv_n_active_reqs; /* # reqs being served */
591 cfs_duration_t srv_rqbd_timeout; /* timeout before re-posting reqs, in tick */
592 int srv_watchdog_timeout; /* soft watchdog timeout, in ms */
593 unsigned srv_cpu_affinity:1; /* bind threads to CPUs */
595 __u32 srv_req_portal;
596 __u32 srv_rep_portal;
598 int srv_n_queued_reqs; /* # reqs waiting to be served */
599 struct list_head srv_request_queue; /* reqs waiting for service */
601 struct list_head srv_request_history; /* request history */
602 __u64 srv_request_seq; /* next request sequence # */
603 __u64 srv_request_max_cull_seq; /* highest seq culled from history */
604 svcreq_printfn_t srv_request_history_print_fn; /* service-specific print fn */
606 struct list_head srv_idle_rqbds; /* request buffers to be reposted */
607 struct list_head srv_active_rqbds; /* req buffers receiving */
608 struct list_head srv_history_rqbds; /* request buffer history */
609 int srv_nrqbd_receiving; /* # posted request buffers */
610 int srv_n_history_rqbds; /* # request buffers in history */
611 int srv_max_history_rqbds;/* max # request buffers in history */
613 atomic_t srv_outstanding_replies;
614 struct list_head srv_active_replies; /* all the active replies */
615 struct list_head srv_reply_queue; /* replies waiting for service */
617 cfs_waitq_t srv_waitq; /* all threads sleep on this. This
618 * wait-queue is signalled when new
619 * incoming request arrives and when
620 * difficult reply has to be handled. */
622 struct list_head srv_threads; /* service thread list */
623 svc_handler_t srv_handler;
625 char *srv_name; /* only statically allocated strings here; we don't clean them */
626 char *srv_thread_name; /* only statically allocated strings here; we don't clean them */
630 cfs_proc_dir_entry_t *srv_procroot;
631 struct lprocfs_stats *srv_stats;
633 /* List of free reply_states */
634 struct list_head srv_free_rs_list;
635 /* waitq to run, when adding stuff to srv_free_rs_list */
636 cfs_waitq_t srv_free_rs_waitq;
639 * Tags for lu_context associated with this thread, see struct
644 * if non-NULL called during thread creation (ptlrpc_start_thread())
645 * to initialize service specific per-thread state.
647 int (*srv_init)(struct ptlrpc_thread *thread);
649 * if non-NULL called during thread shutdown (ptlrpc_main()) to
650 * destruct state created by ->srv_init().
652 void (*srv_done)(struct ptlrpc_thread *thread);
654 //struct ptlrpc_srv_ni srv_interfaces[0];
657 /* ptlrpc/events.c */
658 extern lnet_handle_eq_t ptlrpc_eq_h;
659 extern int ptlrpc_uuid_to_peer(struct obd_uuid *uuid,
660 lnet_process_id_t *peer, lnet_nid_t *self);
661 extern void request_out_callback (lnet_event_t *ev);
662 extern void reply_in_callback(lnet_event_t *ev);
663 extern void client_bulk_callback (lnet_event_t *ev);
664 extern void request_in_callback(lnet_event_t *ev);
665 extern void reply_out_callback(lnet_event_t *ev);
666 extern void server_bulk_callback (lnet_event_t *ev);
668 /* ptlrpc/connection.c */
669 void ptlrpc_dump_connections(void);
670 void ptlrpc_readdress_connection(struct ptlrpc_connection *, struct obd_uuid *);
671 struct ptlrpc_connection *ptlrpc_get_connection(lnet_process_id_t peer,
672 lnet_nid_t self, struct obd_uuid *uuid);
673 int ptlrpc_put_connection(struct ptlrpc_connection *c);
674 struct ptlrpc_connection *ptlrpc_connection_addref(struct ptlrpc_connection *);
675 int ptlrpc_init_connection(void);
676 void ptlrpc_cleanup_connection(void);
677 extern lnet_pid_t ptl_get_pid(void);
679 /* ptlrpc/niobuf.c */
680 int ptlrpc_start_bulk_transfer(struct ptlrpc_bulk_desc *desc);
681 void ptlrpc_abort_bulk(struct ptlrpc_bulk_desc *desc);
682 int ptlrpc_register_bulk(struct ptlrpc_request *req);
683 void ptlrpc_unregister_bulk (struct ptlrpc_request *req);
685 static inline int ptlrpc_bulk_active (struct ptlrpc_bulk_desc *desc)
689 spin_lock(&desc->bd_lock);
690 rc = desc->bd_network_rw;
691 spin_unlock(&desc->bd_lock);
695 int ptlrpc_send_reply(struct ptlrpc_request *req, int);
696 int ptlrpc_reply(struct ptlrpc_request *req);
697 int ptlrpc_error(struct ptlrpc_request *req);
698 void ptlrpc_resend_req(struct ptlrpc_request *request);
699 int ptl_send_rpc(struct ptlrpc_request *request, int noreply);
700 int ptlrpc_register_rqbd (struct ptlrpc_request_buffer_desc *rqbd);
702 /* ptlrpc/client.c */
703 void ptlrpc_init_client(int req_portal, int rep_portal, char *name,
704 struct ptlrpc_client *);
705 void ptlrpc_cleanup_client(struct obd_import *imp);
706 struct ptlrpc_connection *ptlrpc_uuid_to_connection(struct obd_uuid *uuid);
709 ptlrpc_client_receiving_reply (struct ptlrpc_request *req)
713 spin_lock(&req->rq_lock);
714 rc = req->rq_receiving_reply;
715 spin_unlock(&req->rq_lock);
720 ptlrpc_client_replied (struct ptlrpc_request *req)
724 spin_lock(&req->rq_lock);
725 rc = req->rq_replied;
726 spin_unlock(&req->rq_lock);
731 ptlrpc_wake_client_req (struct ptlrpc_request *req)
733 if (req->rq_set == NULL)
734 cfs_waitq_signal(&req->rq_reply_waitq);
736 cfs_waitq_signal(&req->rq_set->set_waitq);
739 int ptlrpc_queue_wait(struct ptlrpc_request *req);
740 int ptlrpc_replay_req(struct ptlrpc_request *req);
741 void ptlrpc_unregister_reply(struct ptlrpc_request *req);
742 void ptlrpc_restart_req(struct ptlrpc_request *req);
743 void ptlrpc_abort_inflight(struct obd_import *imp);
745 struct ptlrpc_request_set *ptlrpc_prep_set(void);
746 int ptlrpc_set_next_timeout(struct ptlrpc_request_set *);
747 int ptlrpc_check_set(struct ptlrpc_request_set *set);
748 int ptlrpc_set_wait(struct ptlrpc_request_set *);
749 int ptlrpc_expired_set(void *data);
750 void ptlrpc_interrupted_set(void *data);
751 void ptlrpc_mark_interrupted(struct ptlrpc_request *req);
752 void ptlrpc_set_destroy(struct ptlrpc_request_set *);
753 void ptlrpc_set_add_req(struct ptlrpc_request_set *, struct ptlrpc_request *);
754 void ptlrpc_set_add_new_req(struct ptlrpc_request_set *,
755 struct ptlrpc_request *);
757 void ptlrpc_free_rq_pool(struct ptlrpc_request_pool *pool);
758 void ptlrpc_add_rqs_to_pool(struct ptlrpc_request_pool *pool, int num_rq);
759 struct ptlrpc_request_pool *ptlrpc_init_rq_pool(int, int,
760 void (*populate_pool)(struct ptlrpc_request_pool *, int));
761 struct ptlrpc_request *ptlrpc_prep_req(struct obd_import *imp, __u32 version,
762 int opcode, int count, int *lengths,
764 struct ptlrpc_request *ptlrpc_prep_req_pool(struct obd_import *imp,
765 __u32 version, int opcode,
766 int count, int *lengths, char **bufs,
767 struct ptlrpc_request_pool *pool,
768 struct ptlrpc_cli_ctx *ctx);
769 void ptlrpc_free_req(struct ptlrpc_request *request);
770 void ptlrpc_req_finished(struct ptlrpc_request *request);
771 void ptlrpc_req_finished_with_imp_lock(struct ptlrpc_request *request);
772 struct ptlrpc_request *ptlrpc_request_addref(struct ptlrpc_request *req);
773 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_imp (struct ptlrpc_request *req,
774 int npages, int type, int portal);
775 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_exp(struct ptlrpc_request *req,
776 int npages, int type, int portal);
777 void ptlrpc_free_bulk(struct ptlrpc_bulk_desc *bulk);
778 void ptlrpc_prep_bulk_page(struct ptlrpc_bulk_desc *desc,
779 cfs_page_t *page, int pageoffset, int len);
780 void ptlrpc_retain_replayable_request(struct ptlrpc_request *req,
781 struct obd_import *imp);
782 __u64 ptlrpc_next_xid(void);
783 __u64 ptlrpc_sample_next_xid(void);
784 __u64 ptlrpc_req_xid(struct ptlrpc_request *request);
786 struct ptlrpc_service_conf {
789 int psc_max_req_size;
790 int psc_max_reply_size;
793 int psc_watchdog_timeout; /* in ms */
799 /* ptlrpc/service.c */
800 void ptlrpc_save_lock (struct ptlrpc_request *req,
801 struct lustre_handle *lock, int mode);
802 void ptlrpc_commit_replies (struct obd_device *obd);
803 void ptlrpc_schedule_difficult_reply (struct ptlrpc_reply_state *rs);
804 struct ptlrpc_service *ptlrpc_init_svc_conf(struct ptlrpc_service_conf *c,
805 svc_handler_t h, char *name,
806 struct proc_dir_entry *proc_entry,
807 svcreq_printfn_t prntfn,
810 struct ptlrpc_service *ptlrpc_init_svc(int nbufs, int bufsize, int max_req_size,
812 int req_portal, int rep_portal,
813 int watchdog_timeout, /* in ms */
814 svc_handler_t, char *name,
815 cfs_proc_dir_entry_t *proc_entry,
817 int min_threads, int max_threads,
818 char *threadname, __u32 ctx_tags);
819 void ptlrpc_stop_all_threads(struct ptlrpc_service *svc);
821 int ptlrpc_start_threads(struct obd_device *dev, struct ptlrpc_service *svc);
822 int ptlrpc_start_thread(struct obd_device *dev, struct ptlrpc_service *svc);
823 int ptlrpc_unregister_service(struct ptlrpc_service *service);
824 int liblustre_check_services (void *arg);
825 void ptlrpc_daemonize(char *name);
826 int ptlrpc_service_health_check(struct ptlrpc_service *);
829 struct ptlrpc_svc_data {
831 struct ptlrpc_service *svc;
832 struct ptlrpc_thread *thread;
833 struct obd_device *dev;
836 /* ptlrpc/import.c */
837 int ptlrpc_connect_import(struct obd_import *imp, char * new_uuid);
838 int ptlrpc_init_import(struct obd_import *imp);
839 int ptlrpc_disconnect_import(struct obd_import *imp, int noclose);
840 int ptlrpc_import_recovery_state_machine(struct obd_import *imp);
842 /* ptlrpc/pack_generic.c */
843 int lustre_msg_swabbed(struct lustre_msg *msg);
844 int lustre_msg_check_version(struct lustre_msg *msg, __u32 version);
845 void lustre_init_msg_v2(struct lustre_msg_v2 *msg, int count, int *lens,
847 int lustre_pack_request(struct ptlrpc_request *, __u32 magic, int count,
848 int *lens, char **bufs);
849 int lustre_pack_reply(struct ptlrpc_request *, int count, int *lens,
851 int lustre_pack_reply_v2(struct ptlrpc_request *req, int count,
852 int *lens, char **bufs);
853 int lustre_shrink_msg(struct lustre_msg *msg, int segment,
854 unsigned int newlen, int move_data);
855 void lustre_free_reply_state(struct ptlrpc_reply_state *rs);
856 int lustre_msg_size(__u32 magic, int count, int *lengths);
857 int lustre_msg_size_v2(int count, int *lengths);
858 int lustre_packed_msg_size(struct lustre_msg *msg);
859 int lustre_unpack_msg(struct lustre_msg *m, int len);
860 void *lustre_msg_buf_v1(void *msg, int n, int min_size);
861 void *lustre_msg_buf_v2(struct lustre_msg_v2 *m, int n, int min_size);
862 void *lustre_msg_buf(struct lustre_msg *m, int n, int minlen);
863 int lustre_msg_buflen(struct lustre_msg *m, int n);
864 void lustre_msg_set_buflen(struct lustre_msg *m, int n, int len);
865 int lustre_msg_bufcount(struct lustre_msg *m);
866 char *lustre_msg_string (struct lustre_msg *m, int n, int max_len);
867 void *lustre_swab_buf(struct lustre_msg *, int n, int minlen, void *swabber);
868 void *lustre_swab_reqbuf(struct ptlrpc_request *req, int n, int minlen,
870 void *lustre_swab_repbuf(struct ptlrpc_request *req, int n, int minlen,
872 __u32 lustre_msg_get_flags(struct lustre_msg *msg);
873 void lustre_msg_add_flags(struct lustre_msg *msg, int flags);
874 void lustre_msg_set_flags(struct lustre_msg *msg, int flags);
875 void lustre_msg_clear_flags(struct lustre_msg *msg, int flags);
876 __u32 lustre_msg_get_op_flags(struct lustre_msg *msg);
877 void lustre_msg_add_op_flags(struct lustre_msg *msg, int flags);
878 void lustre_msg_set_op_flags(struct lustre_msg *msg, int flags);
879 struct lustre_handle *lustre_msg_get_handle(struct lustre_msg *msg);
880 __u32 lustre_msg_get_type(struct lustre_msg *msg);
881 __u32 lustre_msg_get_version(struct lustre_msg *msg);
882 void lustre_msg_add_version(struct lustre_msg *msg, int version);
883 __u32 lustre_msg_get_opc(struct lustre_msg *msg);
884 __u64 lustre_msg_get_last_xid(struct lustre_msg *msg);
885 __u64 lustre_msg_get_last_committed(struct lustre_msg *msg);
886 __u64 lustre_msg_get_transno(struct lustre_msg *msg);
887 __u64 lustre_msg_get_slv(struct lustre_msg *msg);
888 __u32 lustre_msg_get_limit(struct lustre_msg *msg);
889 void lustre_msg_set_slv(struct lustre_msg *msg, __u64 slv);
890 void lustre_msg_set_limit(struct lustre_msg *msg, __u64 limit);
891 int lustre_msg_get_status(struct lustre_msg *msg);
892 __u32 lustre_msg_get_conn_cnt(struct lustre_msg *msg);
893 __u32 lustre_msg_get_magic(struct lustre_msg *msg);
894 void lustre_msg_set_handle(struct lustre_msg *msg,struct lustre_handle *handle);
895 void lustre_msg_set_type(struct lustre_msg *msg, __u32 type);
896 void lustre_msg_set_opc(struct lustre_msg *msg, __u32 opc);
897 void lustre_msg_set_last_xid(struct lustre_msg *msg, __u64 last_xid);
898 void lustre_msg_set_last_committed(struct lustre_msg *msg,__u64 last_committed);
899 void lustre_msg_set_transno(struct lustre_msg *msg, __u64 transno);
900 void lustre_msg_set_status(struct lustre_msg *msg, __u32 status);
901 void lustre_msg_set_conn_cnt(struct lustre_msg *msg, __u32 conn_cnt);
904 lustre_shrink_reply(struct ptlrpc_request *req, int segment,
905 unsigned int newlen, int move_data)
907 LASSERT(req->rq_reply_state);
908 LASSERT(req->rq_repmsg);
909 req->rq_replen = lustre_shrink_msg(req->rq_repmsg, segment,
914 ptlrpc_rs_addref(struct ptlrpc_reply_state *rs)
916 LASSERT(atomic_read(&rs->rs_refcount) > 0);
917 atomic_inc(&rs->rs_refcount);
921 ptlrpc_rs_decref(struct ptlrpc_reply_state *rs)
923 LASSERT(atomic_read(&rs->rs_refcount) > 0);
924 if (atomic_dec_and_test(&rs->rs_refcount))
925 lustre_free_reply_state(rs);
928 static inline __u32 lustre_request_magic(struct ptlrpc_request *req)
930 return lustre_msg_get_magic(req->rq_reqmsg);
933 static inline int ptlrpc_req_get_repsize(struct ptlrpc_request *req)
935 switch (req->rq_reqmsg->lm_magic) {
936 case LUSTRE_MSG_MAGIC_V1:
937 CERROR("function not supported for lustre_msg V1!\n");
939 case LUSTRE_MSG_MAGIC_V2:
940 return req->rq_reqmsg->lm_repsize;
942 LASSERTF(0, "incorrect message magic: %08x\n",
943 req->rq_reqmsg->lm_magic);
949 ptlrpc_req_set_repsize(struct ptlrpc_request *req, int count, int *lens)
951 req->rq_replen = lustre_msg_size(req->rq_reqmsg->lm_magic, count, lens);
952 if (req->rq_reqmsg->lm_magic == LUSTRE_MSG_MAGIC_V2)
953 req->rq_reqmsg->lm_repsize = req->rq_replen;
956 /* ldlm/ldlm_lib.c */
957 int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg);
958 int client_obd_cleanup(struct obd_device *obddev);
959 int client_connect_import(const struct lu_env *env,
960 struct lustre_handle *conn, struct obd_device *obd,
961 struct obd_uuid *cluuid, struct obd_connect_data *);
962 int client_disconnect_export(struct obd_export *exp);
963 int client_import_add_conn(struct obd_import *imp, struct obd_uuid *uuid,
965 int client_import_del_conn(struct obd_import *imp, struct obd_uuid *uuid);
966 int import_set_conn_priority(struct obd_import *imp, struct obd_uuid *uuid);
968 /* ptlrpc/pinger.c */
969 int ptlrpc_pinger_add_import(struct obd_import *imp);
970 int ptlrpc_pinger_del_import(struct obd_import *imp);
971 cfs_time_t ptlrpc_suspend_wakeup_time(void);
973 void ping_evictor_start(void);
974 void ping_evictor_stop(void);
976 #define ping_evictor_start() do {} while (0)
977 #define ping_evictor_stop() do {} while (0)
979 int ptlrpc_check_and_wait_suspend(struct ptlrpc_request *req);
981 /* ptlrpc/ptlrpcd.c */
982 void ptlrpcd_wake(struct ptlrpc_request *req);
983 void ptlrpcd_add_req(struct ptlrpc_request *req);
984 int ptlrpcd_addref(void);
985 void ptlrpcd_decref(void);
987 /* ptlrpc/lproc_ptlrpc.c */
988 const char* ll_opcode2str(__u32 opcode);
990 void ptlrpc_lprocfs_register_obd(struct obd_device *obd);
991 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd);
992 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int opc, int bytes);
994 static inline void ptlrpc_lprocfs_register_obd(struct obd_device *obd) {}
995 static inline void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd) {}
996 static inline void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int opc,
1000 /* ptlrpc/llog_server.c */
1001 int llog_origin_handle_create(struct ptlrpc_request *req);
1002 int llog_origin_handle_destroy(struct ptlrpc_request *req);
1003 int llog_origin_handle_prev_block(struct ptlrpc_request *req);
1004 int llog_origin_handle_next_block(struct ptlrpc_request *req);
1005 int llog_origin_handle_read_header(struct ptlrpc_request *req);
1006 int llog_origin_handle_close(struct ptlrpc_request *req);
1007 int llog_origin_handle_cancel(struct ptlrpc_request *req);
1008 int llog_catinfo(struct ptlrpc_request *req);
1010 /* ptlrpc/llog_client.c */
1011 extern struct llog_operations llog_client_ops;