1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
40 #if defined(__linux__)
41 #include <linux/lustre_net.h>
42 #elif defined(__APPLE__)
43 #include <darwin/lustre_net.h>
44 #elif defined(__WINNT__)
45 #include <winnt/lustre_net.h>
47 #error Unsupported operating system.
50 #include <libcfs/libcfs.h>
52 #include <lnet/lnet.h>
53 #include <lustre/lustre_idl.h>
54 #include <lustre_ha.h>
55 #include <lustre_sec.h>
56 #include <lustre_import.h>
57 #include <lprocfs_status.h>
58 #include <lu_object.h>
59 #include <lustre_req_layout.h>
61 /* MD flags we _always_ use */
62 #define PTLRPC_MD_OPTIONS 0
64 /* Define maxima for bulk I/O
65 * CAVEAT EMPTOR, with multinet (i.e. routers forwarding between networks)
66 * these limits are system wide and not interface-local. */
67 #define PTLRPC_MAX_BRW_BITS LNET_MTU_BITS
68 #define PTLRPC_MAX_BRW_SIZE (1<<LNET_MTU_BITS)
69 #define PTLRPC_MAX_BRW_PAGES (PTLRPC_MAX_BRW_SIZE >> CFS_PAGE_SHIFT)
71 /* When PAGE_SIZE is a constant, we can check our arithmetic here with cpp! */
73 # if ((PTLRPC_MAX_BRW_PAGES & (PTLRPC_MAX_BRW_PAGES - 1)) != 0)
74 # error "PTLRPC_MAX_BRW_PAGES isn't a power of two"
76 # if (PTLRPC_MAX_BRW_SIZE != (PTLRPC_MAX_BRW_PAGES * CFS_PAGE_SIZE))
77 # error "PTLRPC_MAX_BRW_SIZE isn't PTLRPC_MAX_BRW_PAGES * CFS_PAGE_SIZE"
79 # if (PTLRPC_MAX_BRW_SIZE > LNET_MTU)
80 # error "PTLRPC_MAX_BRW_SIZE too big"
82 # if (PTLRPC_MAX_BRW_PAGES > LNET_MAX_IOV)
83 # error "PTLRPC_MAX_BRW_PAGES too big"
85 #endif /* __KERNEL__ */
87 /* Size over which to OBD_VMALLOC() rather than OBD_ALLOC() service request
89 #define SVC_BUF_VMALLOC_THRESHOLD (2 * CFS_PAGE_SIZE)
91 /* The following constants determine how memory is used to buffer incoming
94 * ?_NBUFS # buffers to allocate when growing the pool
95 * ?_BUFSIZE # bytes in a single request buffer
96 * ?_MAXREQSIZE # maximum request service will receive
98 * When fewer than ?_NBUFS/2 buffers are posted for receive, another chunk
99 * of ?_NBUFS is added to the pool.
101 * Messages larger than ?_MAXREQSIZE are dropped. Request buffers are
102 * considered full when less than ?_MAXREQSIZE is left in them.
105 #define LDLM_THREADS_AUTO_MIN \
106 min((int)(num_online_cpus() * num_online_cpus() * 2), 8)
107 #define LDLM_THREADS_AUTO_MAX (LDLM_THREADS_AUTO_MIN * 16)
108 #define LDLM_BL_THREADS LDLM_THREADS_AUTO_MIN
109 #define LDLM_NBUFS (64 * num_online_cpus())
110 #define LDLM_BUFSIZE (8 * 1024)
111 #define LDLM_MAXREQSIZE (5 * 1024)
112 #define LDLM_MAXREPSIZE (1024)
114 #define MDT_MIN_THREADS 2UL
115 #define MDT_MAX_THREADS 512UL
116 #define MDT_NUM_THREADS max(min_t(unsigned long, MDT_MAX_THREADS, \
117 num_physpages >> (25 - CFS_PAGE_SHIFT)), 2UL)
118 #define FLD_NUM_THREADS max(min_t(unsigned long, MDT_MAX_THREADS, \
119 num_physpages >> (25 - CFS_PAGE_SHIFT)), 2UL)
120 #define SEQ_NUM_THREADS max(min_t(unsigned long, MDT_MAX_THREADS, \
121 num_physpages >> (25 - CFS_PAGE_SHIFT)), 2UL)
123 /* Absolute limits */
124 #define MDS_THREADS_MIN 2
125 #define MDS_THREADS_MAX 512
126 #define MDS_THREADS_MIN_READPAGE 2
127 #define MDS_NBUFS (64 * num_online_cpus())
128 #define MDS_BUFSIZE (8 * 1024)
129 /* Assume file name length = FNAME_MAX = 256 (true for ext3).
130 * path name length = PATH_MAX = 4096
131 * LOV MD size max = EA_MAX = 4000
132 * symlink: FNAME_MAX + PATH_MAX <- largest
133 * link: FNAME_MAX + PATH_MAX (mds_rec_link < mds_rec_create)
134 * rename: FNAME_MAX + FNAME_MAX
135 * open: FNAME_MAX + EA_MAX
137 * MDS_MAXREQSIZE ~= 4736 bytes =
138 * lustre_msg + ldlm_request + mds_body + mds_rec_create + FNAME_MAX + PATH_MAX
139 * MDS_MAXREPSIZE ~= 8300 bytes = lustre_msg + llog_header
140 * or, for mds_close() and mds_reint_unlink() on a many-OST filesystem:
141 * = 9210 bytes = lustre_msg + mds_body + 160 * (easize + cookiesize)
143 * Realistic size is about 512 bytes (20 character name + 128 char symlink),
144 * except in the open case where there are a large number of OSTs in a LOV.
146 #define MDS_MAXREQSIZE (5 * 1024)
147 #define MDS_MAXREPSIZE max(9 * 1024, 280 + LOV_MAX_STRIPE_COUNT * 56)
149 /* FLD_MAXREQSIZE == lustre_msg + __u32 padding + ptlrpc_body + opc + md_fld */
150 #define FLD_MAXREQSIZE (160)
152 /* FLD_MAXREPSIZE == lustre_msg + ptlrpc_body + md_fld */
153 #define FLD_MAXREPSIZE (152)
155 /* SEQ_MAXREQSIZE == lustre_msg + __u32 padding + ptlrpc_body + opc + lu_range +
157 #define SEQ_MAXREQSIZE (160)
159 /* SEQ_MAXREPSIZE == lustre_msg + ptlrpc_body + lu_range */
160 #define SEQ_MAXREPSIZE (152)
162 #define MGS_THREADS_AUTO_MIN 2
163 #define MGS_THREADS_AUTO_MAX 32
164 #define MGS_NBUFS (64 * num_online_cpus())
165 #define MGS_BUFSIZE (8 * 1024)
166 #define MGS_MAXREQSIZE (7 * 1024)
167 #define MGS_MAXREPSIZE (9 * 1024)
169 /* Absolute limits */
170 #define OSS_THREADS_MIN 2
171 #define OSS_THREADS_MAX 512
172 #define OST_NBUFS (64 * num_online_cpus())
173 #define OST_BUFSIZE (8 * 1024)
174 /* OST_MAXREQSIZE ~= 4768 bytes =
175 * lustre_msg + obdo + 16 * obd_ioobj + 256 * niobuf_remote
177 * - single object with 16 pages is 512 bytes
178 * - OST_MAXREQSIZE must be at least 1 page of cookies plus some spillover
180 #define OST_MAXREQSIZE (5 * 1024)
181 #define OST_MAXREPSIZE (9 * 1024)
183 /* Macro to hide a typecast. */
184 #define ptlrpc_req_async_args(req) ((void *)&req->rq_async_args)
186 struct ptlrpc_connection {
187 struct hlist_node c_hash;
189 lnet_process_id_t c_peer;
190 struct obd_uuid c_remote_uuid;
194 struct ptlrpc_client {
195 __u32 cli_request_portal;
196 __u32 cli_reply_portal;
200 /* state flags of requests */
201 /* XXX only ones left are those used by the bulk descs as well! */
202 #define PTL_RPC_FL_INTR (1 << 0) /* reply wait was interrupted by user */
203 #define PTL_RPC_FL_TIMEOUT (1 << 7) /* request timed out waiting for reply */
205 #define REQ_MAX_ACK_LOCKS 8
207 union ptlrpc_async_args {
208 /* Scratchpad for passing args to completion interpreter. Users
209 * cast to the struct of their choosing, and LASSERT that this is
210 * big enough. For _tons_ of context, OBD_ALLOC a struct and store
211 * a pointer to it here. The pointer_arg ensures this struct is at
212 * least big enough for that. */
213 void *pointer_arg[9];
217 struct ptlrpc_request_set;
218 typedef int (*set_interpreter_func)(struct ptlrpc_request_set *, void *, int);
220 struct ptlrpc_request_set {
221 int set_remaining; /* # uncompleted requests */
222 cfs_waitq_t set_waitq;
223 cfs_waitq_t *set_wakeup_ptr;
224 struct list_head set_requests;
225 struct list_head set_cblist; /* list of completion callbacks */
226 set_interpreter_func set_interpret; /* completion callback */
227 void *set_arg; /* completion context */
228 /* locked so that any old caller can communicate requests to
229 * the set holder who can then fold them into the lock-free set */
230 spinlock_t set_new_req_lock;
231 struct list_head set_new_requests;
234 struct ptlrpc_set_cbdata {
235 struct list_head psc_item;
236 set_interpreter_func psc_interpret;
240 struct ptlrpc_bulk_desc;
243 * ptlrpc callback & work item stuff
245 struct ptlrpc_cb_id {
246 void (*cbid_fn)(lnet_event_t *ev); /* specific callback fn */
247 void *cbid_arg; /* additional arg */
250 #define RS_MAX_LOCKS 8
253 struct ptlrpc_reply_state {
254 struct ptlrpc_cb_id rs_cb_id;
255 struct list_head rs_list;
256 struct list_head rs_exp_list;
257 struct list_head rs_obd_list;
259 struct list_head rs_debug_list;
261 /* updates to following flag serialised by srv_request_lock */
262 unsigned long rs_difficult:1; /* ACK/commit stuff */
263 unsigned long rs_scheduled:1; /* being handled? */
264 unsigned long rs_scheduled_ever:1;/* any schedule attempts? */
265 unsigned long rs_handled:1; /* been handled yet? */
266 unsigned long rs_on_net:1; /* reply_out_callback pending? */
267 unsigned long rs_prealloc:1; /* rs from prealloc list */
272 struct obd_export *rs_export;
273 struct ptlrpc_service *rs_service;
274 lnet_handle_md_t rs_md_h;
275 atomic_t rs_refcount;
277 struct ptlrpc_svc_ctx *rs_svc_ctx;
278 struct lustre_msg *rs_repbuf; /* wrapper */
279 int rs_repbuf_len; /* wrapper buf length */
280 int rs_repdata_len; /* wrapper msg length */
281 struct lustre_msg *rs_msg; /* reply message */
283 /* locks awaiting client reply ACK */
285 struct lustre_handle rs_locks[RS_MAX_LOCKS];
286 ldlm_mode_t rs_modes[RS_MAX_LOCKS];
289 struct ptlrpc_thread;
292 RQ_PHASE_NEW = 0xebc0de00,
293 RQ_PHASE_RPC = 0xebc0de01,
294 RQ_PHASE_BULK = 0xebc0de02,
295 RQ_PHASE_INTERPRET = 0xebc0de03,
296 RQ_PHASE_COMPLETE = 0xebc0de04,
299 struct ptlrpc_request_pool {
301 struct list_head prp_req_list; /* list of ptlrpc_request structs */
303 void (*prp_populate)(struct ptlrpc_request_pool *, int);
309 struct ptlrpc_request {
310 int rq_type; /* one of PTL_RPC_MSG_* */
311 struct list_head rq_list;
312 struct list_head rq_timed_list; /* server-side early replies */
313 struct list_head rq_history_list; /* server-side history */
314 __u64 rq_history_seq; /* history sequence # */
317 /* client-side flags are serialized by rq_lock */
318 unsigned long rq_intr:1, rq_replied:1, rq_err:1,
319 rq_timedout:1, rq_resend:1, rq_restart:1,
321 * when ->rq_replay is set, request is kept by the client even
322 * after server commits corresponding transaction. This is
323 * used for operations that require sequence of multiple
324 * requests to be replayed. The only example currently is file
325 * open/close/dw/setattr. When last request in such a sequence
326 * is committed, ->rq_replay is cleared on all requests in the
330 /* this is the last request in the sequence. */
332 rq_no_resend:1, rq_waiting:1, rq_receiving_reply:1,
333 rq_no_delay:1, rq_net_err:1, rq_wait_ctx:1,
334 rq_early:1, rq_must_unlink:1,
335 /* server-side flags */
336 rq_packed_final:1, /* packed final reply */
337 rq_sent_final:1; /* stop sending early replies */
339 enum rq_phase rq_phase; /* one of RQ_PHASE_* */
340 atomic_t rq_refcount; /* client-side refcount for SENT race,
341 server-side refcounf for multiple replies */
343 struct ptlrpc_thread *rq_svc_thread; /* initial thread servicing req */
345 int rq_request_portal; /* XXX FIXME bug 249 */
346 int rq_reply_portal; /* XXX FIXME bug 249 */
348 int rq_nob_received; /* client-side # reply bytes actually received */
351 struct lustre_msg *rq_reqmsg;
354 struct lustre_msg *rq_repmsg;
357 struct list_head rq_replay_list;
358 struct list_head rq_mod_list;
360 struct ptlrpc_cli_ctx *rq_cli_ctx; /* client's half ctx */
361 struct ptlrpc_svc_ctx *rq_svc_ctx; /* server's half ctx */
362 struct list_head rq_ctx_chain; /* link to waited ctx */
364 struct sptlrpc_flavor rq_flvr; /* client & server */
365 enum lustre_sec_part rq_sp_from;
367 unsigned long /* client/server security flags */
368 rq_ctx_init:1, /* context initiation */
369 rq_ctx_fini:1, /* context destroy */
370 rq_bulk_read:1, /* request bulk read */
371 rq_bulk_write:1, /* request bulk write */
372 /* server authentication flags */
373 rq_auth_gss:1, /* authenticated by gss */
374 rq_auth_remote:1, /* authed as remote user */
375 rq_auth_usr_root:1, /* authed as root */
376 rq_auth_usr_mdt:1, /* authed as mdt */
377 /* security tfm flags */
380 /* doesn't expect reply FIXME */
383 uid_t rq_auth_uid; /* authed uid */
384 uid_t rq_auth_mapped_uid; /* authed uid mapped to */
386 /* (server side), pointed directly into req buffer */
387 struct ptlrpc_user_desc *rq_user_desc;
389 /* early replies go to offset 0, regular replies go after that */
390 unsigned int rq_reply_off;
392 /* various buffer pointers */
393 struct lustre_msg *rq_reqbuf; /* req wrapper */
394 int rq_reqbuf_len; /* req wrapper buf len */
395 int rq_reqdata_len; /* req wrapper msg len */
396 char *rq_repbuf; /* rep buffer */
397 int rq_repbuf_len; /* rep buffer len */
398 struct lustre_msg *rq_repdata; /* rep wrapper msg */
399 int rq_repdata_len; /* rep wrapper msg len */
400 struct lustre_msg *rq_clrbuf; /* only in priv mode */
401 int rq_clrbuf_len; /* only in priv mode */
402 int rq_clrdata_len; /* only in priv mode */
404 __u32 rq_req_swab_mask;
405 __u32 rq_rep_swab_mask;
407 int rq_import_generation;
408 enum lustre_imp_state rq_send_state;
410 int rq_early_count; /* how many early replies (for stats) */
412 /* client+server request */
413 lnet_handle_md_t rq_req_md_h;
414 struct ptlrpc_cb_id rq_req_cbid;
417 struct timeval rq_arrival_time; /* request arrival time */
418 struct ptlrpc_reply_state *rq_reply_state; /* separated reply state */
419 struct ptlrpc_request_buffer_desc *rq_rqbd; /* incoming request buffer*/
421 __u32 rq_uid; /* peer uid, used in MDS only */
424 /* client-only incoming reply */
425 lnet_handle_md_t rq_reply_md_h;
426 cfs_waitq_t rq_reply_waitq;
427 struct ptlrpc_cb_id rq_reply_cbid;
430 lnet_process_id_t rq_peer;
431 struct obd_export *rq_export;
432 struct obd_import *rq_import;
434 void (*rq_replay_cb)(struct ptlrpc_request *);
435 void (*rq_commit_cb)(struct ptlrpc_request *);
438 struct ptlrpc_bulk_desc *rq_bulk;/* client side bulk */
440 /* client outgoing req */
441 time_t rq_sent; /* when request/reply sent (secs), or
442 * time when request should be sent */
444 volatile time_t rq_deadline; /* when request must finish. volatile
445 so that servers' early reply updates to the deadline aren't
446 kept in per-cpu cache */
447 int rq_timeout; /* service time estimate (secs) */
450 struct list_head rq_set_chain;
451 struct ptlrpc_request_set *rq_set;
452 void *rq_interpret_reply; /* Async completion handler */
453 union ptlrpc_async_args rq_async_args; /* Async completion context */
454 struct ptlrpc_request_pool *rq_pool; /* Pool if request from
456 struct lu_context rq_session;
459 struct req_capsule rq_pill;
462 static inline void ptlrpc_close_replay_seq(struct ptlrpc_request *req)
464 spin_lock(&req->rq_lock);
466 req->rq_sequence = 1;
467 spin_unlock(&req->rq_lock);
470 static inline void lustre_set_req_swabbed(struct ptlrpc_request *req, int index)
472 LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
473 LASSERT((req->rq_req_swab_mask & (1 << index)) == 0);
474 req->rq_req_swab_mask |= 1 << index;
477 static inline void lustre_set_rep_swabbed(struct ptlrpc_request *req, int index)
479 LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
480 LASSERT((req->rq_rep_swab_mask & (1 << index)) == 0);
481 req->rq_rep_swab_mask |= 1 << index;
484 static inline int lustre_req_swabbed(struct ptlrpc_request *req, int index)
486 LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
487 return req->rq_req_swab_mask & (1 << index);
490 static inline int lustre_rep_swabbed(struct ptlrpc_request *req, int index)
492 LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
493 return req->rq_rep_swab_mask & (1 << index);
496 static inline const char *
497 ptlrpc_rqphase2str(const struct ptlrpc_request *req)
499 switch (req->rq_phase) {
506 case RQ_PHASE_INTERPRET:
508 case RQ_PHASE_COMPLETE:
515 /* Spare the preprocessor, spoil the bugs. */
516 #define FLAG(field, str) (field ? str : "")
518 #define DEBUG_REQ_FLAGS(req) \
519 ptlrpc_rqphase2str(req), \
520 FLAG(req->rq_intr, "I"), FLAG(req->rq_replied, "R"), \
521 FLAG(req->rq_err, "E"), \
522 FLAG(req->rq_timedout, "X") /* eXpired */, FLAG(req->rq_resend, "S"), \
523 FLAG(req->rq_restart, "T"), FLAG(req->rq_replay, "P"), \
524 FLAG(req->rq_no_resend, "N"), \
525 FLAG(req->rq_waiting, "W"), \
526 FLAG(req->rq_wait_ctx, "C")
528 #define REQ_FLAGS_FMT "%s:%s%s%s%s%s%s%s%s%s%s"
530 void _debug_req(struct ptlrpc_request *req, __u32 mask,
531 struct libcfs_debug_msg_data *data, const char *fmt, ...)
532 __attribute__ ((format (printf, 4, 5)));
534 #define debug_req(cdls, level, req, file, func, line, fmt, a...) \
538 if (((level) & D_CANTMASK) != 0 || \
539 ((libcfs_debug & (level)) != 0 && \
540 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) { \
541 static struct libcfs_debug_msg_data _req_dbg_data = \
542 DEBUG_MSG_DATA_INIT(cdls, DEBUG_SUBSYSTEM, file, func, line); \
543 _debug_req((req), (level), &_req_dbg_data, fmt, ##a); \
547 /* for most callers (level is a constant) this is resolved at compile time */
548 #define DEBUG_REQ(level, req, fmt, args...) \
550 if ((level) & (D_ERROR | D_WARNING)) { \
551 static cfs_debug_limit_state_t cdls; \
552 debug_req(&cdls, level, req, __FILE__, __func__, __LINE__, \
553 "@@@ "fmt" ", ## args); \
555 debug_req(NULL, level, req, __FILE__, __func__, __LINE__, \
556 "@@@ "fmt" ", ## args); \
559 struct ptlrpc_bulk_page {
560 struct list_head bp_link;
562 int bp_pageoffset; /* offset within a page */
563 struct page *bp_page;
566 #define BULK_GET_SOURCE 0
567 #define BULK_PUT_SINK 1
568 #define BULK_GET_SINK 2
569 #define BULK_PUT_SOURCE 3
571 struct ptlrpc_bulk_desc {
572 unsigned long bd_success:1; /* completed successfully */
573 unsigned long bd_network_rw:1; /* accessible to the network */
574 unsigned long bd_type:2; /* {put,get}{source,sink} */
575 unsigned long bd_registered:1; /* client side */
576 spinlock_t bd_lock; /* serialise with callback */
577 int bd_import_generation;
578 struct obd_export *bd_export;
579 struct obd_import *bd_import;
581 struct ptlrpc_request *bd_req; /* associated request */
582 cfs_waitq_t bd_waitq; /* server side only WQ */
583 int bd_iov_count; /* # entries in bd_iov */
584 int bd_max_iov; /* allocated size of bd_iov */
585 int bd_nob; /* # bytes covered */
586 int bd_nob_transferred; /* # bytes GOT/PUT */
590 struct ptlrpc_cb_id bd_cbid; /* network callback info */
591 lnet_handle_md_t bd_md_h; /* associated MD */
592 lnet_nid_t bd_sender; /* stash event::sender */
594 cfs_page_t **bd_enc_pages;
595 #if defined(__KERNEL__)
596 lnet_kiov_t bd_iov[0];
598 lnet_md_iovec_t bd_iov[0];
602 struct ptlrpc_thread {
604 struct list_head t_link; /* active threads in svc->srv_threads */
606 void *t_data; /* thread-private data (preallocated memory) */
609 unsigned int t_id; /* service thread index, from ptlrpc_start_threads */
610 cfs_waitq_t t_ctl_waitq;
611 struct lu_env *t_env;
614 struct ptlrpc_request_buffer_desc {
615 struct list_head rqbd_list;
616 struct list_head rqbd_reqs;
617 struct ptlrpc_service *rqbd_service;
618 lnet_handle_md_t rqbd_md_h;
621 struct ptlrpc_cb_id rqbd_cbid;
622 struct ptlrpc_request rqbd_req;
625 typedef int (*svc_handler_t)(struct ptlrpc_request *req);
626 typedef void (*svcreq_printfn_t)(void *, struct ptlrpc_request *);
628 struct ptlrpc_service {
629 struct list_head srv_list; /* chain thru all services */
630 int srv_max_req_size; /* biggest request to receive */
631 int srv_max_reply_size; /* biggest reply to send */
632 int srv_buf_size; /* size of individual buffers */
633 int srv_nbuf_per_group; /* # buffers to allocate in 1 group */
634 int srv_nbufs; /* total # req buffer descs allocated */
635 int srv_threads_min; /* threads to start at SOW */
636 int srv_threads_max; /* thread upper limit */
637 int srv_threads_started; /* index of last started thread */
638 int srv_threads_running; /* # running threads */
639 int srv_n_difficult_replies; /* # 'difficult' replies */
640 int srv_n_active_reqs; /* # reqs being served */
641 cfs_duration_t srv_rqbd_timeout; /* timeout before re-posting reqs, in tick */
642 int srv_watchdog_factor; /* soft watchdog timeout mutiplier */
643 unsigned srv_cpu_affinity:1; /* bind threads to CPUs */
644 unsigned srv_at_check:1; /* check early replies */
645 cfs_time_t srv_at_checktime; /* debug */
647 __u32 srv_req_portal;
648 __u32 srv_rep_portal;
651 struct adaptive_timeout srv_at_estimate;/* estimated rpc service time */
652 spinlock_t srv_at_lock;
653 struct list_head srv_at_list; /* reqs waiting for replies */
654 cfs_timer_t srv_at_timer; /* early reply timer */
656 int srv_n_queued_reqs; /* # reqs in either of the queues below */
657 struct list_head srv_req_in_queue; /* incoming reqs */
658 struct list_head srv_request_queue; /* reqs waiting for service */
660 struct list_head srv_request_history; /* request history */
661 __u64 srv_request_seq; /* next request sequence # */
662 __u64 srv_request_max_cull_seq; /* highest seq culled from history */
663 svcreq_printfn_t srv_request_history_print_fn; /* service-specific print fn */
665 struct list_head srv_idle_rqbds; /* request buffers to be reposted */
666 struct list_head srv_active_rqbds; /* req buffers receiving */
667 struct list_head srv_history_rqbds; /* request buffer history */
668 int srv_nrqbd_receiving; /* # posted request buffers */
669 int srv_n_history_rqbds; /* # request buffers in history */
670 int srv_max_history_rqbds;/* max # request buffers in history */
672 atomic_t srv_outstanding_replies;
673 struct list_head srv_active_replies; /* all the active replies */
674 struct list_head srv_reply_queue; /* replies waiting for service */
676 cfs_waitq_t srv_waitq; /* all threads sleep on this. This
677 * wait-queue is signalled when new
678 * incoming request arrives and when
679 * difficult reply has to be handled. */
681 struct list_head srv_threads; /* service thread list */
682 svc_handler_t srv_handler;
684 char *srv_name; /* only statically allocated strings here; we don't clean them */
685 char *srv_thread_name; /* only statically allocated strings here; we don't clean them */
689 cfs_proc_dir_entry_t *srv_procroot;
690 struct lprocfs_stats *srv_stats;
692 /* List of free reply_states */
693 struct list_head srv_free_rs_list;
694 /* waitq to run, when adding stuff to srv_free_rs_list */
695 cfs_waitq_t srv_free_rs_waitq;
698 * Tags for lu_context associated with this thread, see struct
703 * if non-NULL called during thread creation (ptlrpc_start_thread())
704 * to initialize service specific per-thread state.
706 int (*srv_init)(struct ptlrpc_thread *thread);
708 * if non-NULL called during thread shutdown (ptlrpc_main()) to
709 * destruct state created by ->srv_init().
711 void (*srv_done)(struct ptlrpc_thread *thread);
713 //struct ptlrpc_srv_ni srv_interfaces[0];
718 * Ptlrpc thread control flags (LIOD_START, LIOD_STOP, LIOD_STOP_FORCE)
720 unsigned long pc_flags;
722 * Thread lock protecting structure fields.
728 struct completion pc_starting;
732 struct completion pc_finishing;
734 * Thread requests set.
736 struct ptlrpc_request_set *pc_set;
738 * Thread name used in cfs_daemonize()
743 * Async rpcs flag to make sure that ptlrpcd_check() is called only
748 * Currently not used.
752 * User-space async rpcs callback.
754 void *pc_wait_callback;
756 * User-space check idle rpcs callback.
758 void *pc_idle_callback;
762 /* Bits for pc_flags */
763 enum ptlrpcd_ctl_flags {
765 * Ptlrpc thread start flag.
769 * Ptlrpc thread stop flag.
773 * Ptlrpc thread stop force flag. This will cause also
774 * aborting any inflight rpcs handled by thread.
776 LIOD_STOP_FORCE = 1 << 2
779 /* ptlrpc/events.c */
780 extern lnet_handle_eq_t ptlrpc_eq_h;
781 extern int ptlrpc_uuid_to_peer(struct obd_uuid *uuid,
782 lnet_process_id_t *peer, lnet_nid_t *self);
783 extern void request_out_callback (lnet_event_t *ev);
784 extern void reply_in_callback(lnet_event_t *ev);
785 extern void client_bulk_callback (lnet_event_t *ev);
786 extern void request_in_callback(lnet_event_t *ev);
787 extern void reply_out_callback(lnet_event_t *ev);
788 extern void server_bulk_callback (lnet_event_t *ev);
790 /* ptlrpc/connection.c */
791 struct ptlrpc_connection *ptlrpc_connection_get(lnet_process_id_t peer,
793 struct obd_uuid *uuid);
794 int ptlrpc_connection_put(struct ptlrpc_connection *c);
795 struct ptlrpc_connection *ptlrpc_connection_addref(struct ptlrpc_connection *);
796 int ptlrpc_connection_init(void);
797 void ptlrpc_connection_fini(void);
798 extern lnet_pid_t ptl_get_pid(void);
800 /* ptlrpc/niobuf.c */
801 int ptlrpc_start_bulk_transfer(struct ptlrpc_bulk_desc *desc);
802 void ptlrpc_abort_bulk(struct ptlrpc_bulk_desc *desc);
803 int ptlrpc_register_bulk(struct ptlrpc_request *req);
804 void ptlrpc_unregister_bulk (struct ptlrpc_request *req);
806 static inline int ptlrpc_bulk_active (struct ptlrpc_bulk_desc *desc)
810 spin_lock(&desc->bd_lock);
811 rc = desc->bd_network_rw;
812 spin_unlock(&desc->bd_lock);
816 #define PTLRPC_REPLY_MAYBE_DIFFICULT 0x01
817 #define PTLRPC_REPLY_EARLY 0x02
818 int ptlrpc_send_reply(struct ptlrpc_request *req, int flags);
819 int ptlrpc_reply(struct ptlrpc_request *req);
820 int ptlrpc_send_error(struct ptlrpc_request *req, int difficult);
821 int ptlrpc_error(struct ptlrpc_request *req);
822 void ptlrpc_resend_req(struct ptlrpc_request *request);
823 int ptlrpc_at_get_net_latency(struct ptlrpc_request *req);
824 int ptl_send_rpc(struct ptlrpc_request *request, int noreply);
825 int ptlrpc_register_rqbd (struct ptlrpc_request_buffer_desc *rqbd);
827 /* ptlrpc/client.c */
828 void ptlrpc_init_client(int req_portal, int rep_portal, char *name,
829 struct ptlrpc_client *);
830 void ptlrpc_cleanup_client(struct obd_import *imp);
831 struct ptlrpc_connection *ptlrpc_uuid_to_connection(struct obd_uuid *uuid);
834 ptlrpc_client_recv_or_unlink (struct ptlrpc_request *req)
838 spin_lock(&req->rq_lock);
839 rc = req->rq_receiving_reply || req->rq_must_unlink;
840 spin_unlock(&req->rq_lock);
845 ptlrpc_wake_client_req (struct ptlrpc_request *req)
847 if (req->rq_set == NULL)
848 cfs_waitq_signal(&req->rq_reply_waitq);
850 cfs_waitq_signal(&req->rq_set->set_waitq);
853 int ptlrpc_queue_wait(struct ptlrpc_request *req);
854 int ptlrpc_replay_req(struct ptlrpc_request *req);
855 void ptlrpc_unregister_reply(struct ptlrpc_request *req);
856 void ptlrpc_restart_req(struct ptlrpc_request *req);
857 void ptlrpc_abort_inflight(struct obd_import *imp);
858 void ptlrpc_abort_set(struct ptlrpc_request_set *set);
860 struct ptlrpc_request_set *ptlrpc_prep_set(void);
861 int ptlrpc_set_add_cb(struct ptlrpc_request_set *set,
862 set_interpreter_func fn, void *data);
863 int ptlrpc_set_next_timeout(struct ptlrpc_request_set *);
864 int ptlrpc_check_set(struct ptlrpc_request_set *set);
865 int ptlrpc_set_wait(struct ptlrpc_request_set *);
866 int ptlrpc_expired_set(void *data);
867 void ptlrpc_interrupted_set(void *data);
868 void ptlrpc_mark_interrupted(struct ptlrpc_request *req);
869 void ptlrpc_set_destroy(struct ptlrpc_request_set *);
870 void ptlrpc_set_add_req(struct ptlrpc_request_set *, struct ptlrpc_request *);
871 int ptlrpc_set_add_new_req(struct ptlrpcd_ctl *pc,
872 struct ptlrpc_request *req);
874 void ptlrpc_free_rq_pool(struct ptlrpc_request_pool *pool);
875 void ptlrpc_add_rqs_to_pool(struct ptlrpc_request_pool *pool, int num_rq);
877 struct ptlrpc_request_pool *
878 ptlrpc_init_rq_pool(int, int,
879 void (*populate_pool)(struct ptlrpc_request_pool *, int));
881 void ptlrpc_at_set_req_timeout(struct ptlrpc_request *req);
882 struct ptlrpc_request *ptlrpc_request_alloc(struct obd_import *imp,
883 const struct req_format *format);
884 struct ptlrpc_request *ptlrpc_request_alloc_pool(struct obd_import *imp,
885 struct ptlrpc_request_pool *,
886 const struct req_format *format);
887 void ptlrpc_request_free(struct ptlrpc_request *request);
888 int ptlrpc_request_pack(struct ptlrpc_request *request,
889 __u32 version, int opcode);
890 struct ptlrpc_request *ptlrpc_request_alloc_pack(struct obd_import *imp,
891 const struct req_format *format,
892 __u32 version, int opcode);
893 int ptlrpc_request_bufs_pack(struct ptlrpc_request *request,
894 __u32 version, int opcode, char **bufs,
895 struct ptlrpc_cli_ctx *ctx);
896 struct ptlrpc_request *ptlrpc_prep_req(struct obd_import *imp, __u32 version,
897 int opcode, int count, __u32 *lengths,
899 struct ptlrpc_request *ptlrpc_prep_req_pool(struct obd_import *imp,
900 __u32 version, int opcode,
901 int count, __u32 *lengths, char **bufs,
902 struct ptlrpc_request_pool *pool);
903 void ptlrpc_req_finished(struct ptlrpc_request *request);
904 void ptlrpc_req_finished_with_imp_lock(struct ptlrpc_request *request);
905 struct ptlrpc_request *ptlrpc_request_addref(struct ptlrpc_request *req);
906 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_imp (struct ptlrpc_request *req,
907 int npages, int type, int portal);
908 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_exp(struct ptlrpc_request *req,
909 int npages, int type, int portal);
910 void ptlrpc_free_bulk(struct ptlrpc_bulk_desc *bulk);
911 void ptlrpc_prep_bulk_page(struct ptlrpc_bulk_desc *desc,
912 cfs_page_t *page, int pageoffset, int len);
913 void ptlrpc_retain_replayable_request(struct ptlrpc_request *req,
914 struct obd_import *imp);
915 __u64 ptlrpc_next_xid(void);
916 __u64 ptlrpc_sample_next_xid(void);
917 __u64 ptlrpc_req_xid(struct ptlrpc_request *request);
919 struct ptlrpc_service_conf {
922 int psc_max_req_size;
923 int psc_max_reply_size;
926 int psc_watchdog_factor;
932 /* ptlrpc/service.c */
933 void ptlrpc_save_lock (struct ptlrpc_request *req,
934 struct lustre_handle *lock, int mode);
935 void ptlrpc_commit_replies (struct obd_device *obd);
936 void ptlrpc_schedule_difficult_reply (struct ptlrpc_reply_state *rs);
937 struct ptlrpc_service *ptlrpc_init_svc_conf(struct ptlrpc_service_conf *c,
938 svc_handler_t h, char *name,
939 struct proc_dir_entry *proc_entry,
940 svcreq_printfn_t prntfn,
943 struct ptlrpc_service *ptlrpc_init_svc(int nbufs, int bufsize, int max_req_size,
945 int req_portal, int rep_portal,
947 svc_handler_t, char *name,
948 cfs_proc_dir_entry_t *proc_entry,
950 int min_threads, int max_threads,
951 char *threadname, __u32 ctx_tags);
952 void ptlrpc_stop_all_threads(struct ptlrpc_service *svc);
954 int ptlrpc_start_threads(struct obd_device *dev, struct ptlrpc_service *svc);
955 int ptlrpc_start_thread(struct obd_device *dev, struct ptlrpc_service *svc);
956 int ptlrpc_unregister_service(struct ptlrpc_service *service);
957 int liblustre_check_services (void *arg);
958 void ptlrpc_daemonize(char *name);
959 int ptlrpc_service_health_check(struct ptlrpc_service *);
962 struct ptlrpc_svc_data {
964 struct ptlrpc_service *svc;
965 struct ptlrpc_thread *thread;
966 struct obd_device *dev;
969 /* ptlrpc/import.c */
970 int ptlrpc_connect_import(struct obd_import *imp, char * new_uuid);
971 int ptlrpc_init_import(struct obd_import *imp);
972 int ptlrpc_disconnect_import(struct obd_import *imp, int noclose);
973 int ptlrpc_import_recovery_state_machine(struct obd_import *imp);
975 /* ptlrpc/pack_generic.c */
976 int ptlrpc_reconnect_import(struct obd_import *imp);
977 int lustre_msg_swabbed(struct lustre_msg *msg);
978 int lustre_msg_check_version(struct lustre_msg *msg, __u32 version);
979 void lustre_init_msg_v2(struct lustre_msg_v2 *msg, int count, __u32 *lens,
981 int lustre_pack_request(struct ptlrpc_request *, __u32 magic, int count,
982 __u32 *lens, char **bufs);
983 int lustre_pack_reply(struct ptlrpc_request *, int count, __u32 *lens,
985 int lustre_pack_reply_v2(struct ptlrpc_request *req, int count,
986 __u32 *lens, char **bufs, int flags);
987 #define LPRFL_EARLY_REPLY 1
988 int lustre_pack_reply_flags(struct ptlrpc_request *, int count, __u32 *lens,
989 char **bufs, int flags);
990 int lustre_shrink_msg(struct lustre_msg *msg, int segment,
991 unsigned int newlen, int move_data);
992 void lustre_free_reply_state(struct ptlrpc_reply_state *rs);
993 int lustre_msg_hdr_size(__u32 magic, int count);
994 int lustre_msg_size(__u32 magic, int count, __u32 *lengths);
995 int lustre_msg_size_v2(int count, __u32 *lengths);
996 int lustre_packed_msg_size(struct lustre_msg *msg);
997 int lustre_msg_early_size(void);
998 int lustre_unpack_msg(struct lustre_msg *m, int len);
999 void *lustre_msg_buf_v2(struct lustre_msg_v2 *m, int n, int min_size);
1000 void *lustre_msg_buf(struct lustre_msg *m, int n, int minlen);
1001 int lustre_msg_buflen(struct lustre_msg *m, int n);
1002 void lustre_msg_set_buflen(struct lustre_msg *m, int n, int len);
1003 int lustre_msg_bufcount(struct lustre_msg *m);
1004 char *lustre_msg_string (struct lustre_msg *m, int n, int max_len);
1005 void *lustre_swab_buf(struct lustre_msg *, int n, int minlen, void *swabber);
1006 void *lustre_swab_reqbuf(struct ptlrpc_request *req, int n, int minlen,
1008 void *lustre_swab_repbuf(struct ptlrpc_request *req, int n, int minlen,
1010 __u32 lustre_msghdr_get_flags(struct lustre_msg *msg);
1011 void lustre_msghdr_set_flags(struct lustre_msg *msg, __u32 flags);
1012 __u32 lustre_msg_get_flags(struct lustre_msg *msg);
1013 void lustre_msg_add_flags(struct lustre_msg *msg, int flags);
1014 void lustre_msg_set_flags(struct lustre_msg *msg, int flags);
1015 void lustre_msg_clear_flags(struct lustre_msg *msg, int flags);
1016 __u32 lustre_msg_get_op_flags(struct lustre_msg *msg);
1017 void lustre_msg_add_op_flags(struct lustre_msg *msg, int flags);
1018 void lustre_msg_set_op_flags(struct lustre_msg *msg, int flags);
1019 struct lustre_handle *lustre_msg_get_handle(struct lustre_msg *msg);
1020 __u32 lustre_msg_get_type(struct lustre_msg *msg);
1021 __u32 lustre_msg_get_version(struct lustre_msg *msg);
1022 void lustre_msg_add_version(struct lustre_msg *msg, int version);
1023 __u32 lustre_msg_get_opc(struct lustre_msg *msg);
1024 __u64 lustre_msg_get_last_xid(struct lustre_msg *msg);
1025 __u64 lustre_msg_get_last_committed(struct lustre_msg *msg);
1026 __u64 lustre_msg_get_transno(struct lustre_msg *msg);
1027 __u64 lustre_msg_get_slv(struct lustre_msg *msg);
1028 __u32 lustre_msg_get_limit(struct lustre_msg *msg);
1029 void lustre_msg_set_slv(struct lustre_msg *msg, __u64 slv);
1030 void lustre_msg_set_limit(struct lustre_msg *msg, __u64 limit);
1031 int lustre_msg_get_status(struct lustre_msg *msg);
1032 __u32 lustre_msg_get_conn_cnt(struct lustre_msg *msg);
1033 int lustre_msg_is_v1(struct lustre_msg *msg);
1034 __u32 lustre_msg_get_magic(struct lustre_msg *msg);
1035 __u32 lustre_msg_get_timeout(struct lustre_msg *msg);
1036 __u32 lustre_msg_get_service_time(struct lustre_msg *msg);
1037 __u32 lustre_msg_get_cksum(struct lustre_msg *msg);
1038 __u32 lustre_msg_calc_cksum(struct lustre_msg *msg);
1039 void lustre_msg_set_handle(struct lustre_msg *msg,struct lustre_handle *handle);
1040 void lustre_msg_set_type(struct lustre_msg *msg, __u32 type);
1041 void lustre_msg_set_opc(struct lustre_msg *msg, __u32 opc);
1042 void lustre_msg_set_last_xid(struct lustre_msg *msg, __u64 last_xid);
1043 void lustre_msg_set_last_committed(struct lustre_msg *msg,__u64 last_committed);
1044 void lustre_msg_set_transno(struct lustre_msg *msg, __u64 transno);
1045 void lustre_msg_set_status(struct lustre_msg *msg, __u32 status);
1046 void lustre_msg_set_conn_cnt(struct lustre_msg *msg, __u32 conn_cnt);
1047 void ptlrpc_req_set_repsize(struct ptlrpc_request *req, int count, __u32 *sizes);
1048 void ptlrpc_request_set_replen(struct ptlrpc_request *req);
1049 void lustre_msg_set_timeout(struct lustre_msg *msg, __u32 timeout);
1050 void lustre_msg_set_service_time(struct lustre_msg *msg, __u32 service_time);
1051 void lustre_msg_set_cksum(struct lustre_msg *msg, __u32 cksum);
1054 lustre_shrink_reply(struct ptlrpc_request *req, int segment,
1055 unsigned int newlen, int move_data)
1057 LASSERT(req->rq_reply_state);
1058 LASSERT(req->rq_repmsg);
1059 req->rq_replen = lustre_shrink_msg(req->rq_repmsg, segment,
1064 ptlrpc_rs_addref(struct ptlrpc_reply_state *rs)
1066 LASSERT(atomic_read(&rs->rs_refcount) > 0);
1067 atomic_inc(&rs->rs_refcount);
1071 ptlrpc_rs_decref(struct ptlrpc_reply_state *rs)
1073 LASSERT(atomic_read(&rs->rs_refcount) > 0);
1074 if (atomic_dec_and_test(&rs->rs_refcount))
1075 lustre_free_reply_state(rs);
1078 /* Should only be called once per req */
1079 static inline void ptlrpc_req_drop_rs(struct ptlrpc_request *req)
1081 if (req->rq_reply_state == NULL)
1082 return; /* shouldn't occur */
1083 ptlrpc_rs_decref(req->rq_reply_state);
1084 req->rq_reply_state = NULL;
1085 req->rq_repmsg = NULL;
1088 static inline __u32 lustre_request_magic(struct ptlrpc_request *req)
1090 return lustre_msg_get_magic(req->rq_reqmsg);
1093 static inline int ptlrpc_req_get_repsize(struct ptlrpc_request *req)
1095 switch (req->rq_reqmsg->lm_magic) {
1096 case LUSTRE_MSG_MAGIC_V2:
1097 return req->rq_reqmsg->lm_repsize;
1099 LASSERTF(0, "incorrect message magic: %08x\n",
1100 req->rq_reqmsg->lm_magic);
1105 /* ldlm/ldlm_lib.c */
1106 int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg);
1107 int client_obd_cleanup(struct obd_device *obddev);
1108 int client_connect_import(const struct lu_env *env,
1109 struct lustre_handle *conn, struct obd_device *obd,
1110 struct obd_uuid *cluuid, struct obd_connect_data *,
1112 int client_disconnect_export(struct obd_export *exp);
1113 int client_import_add_conn(struct obd_import *imp, struct obd_uuid *uuid,
1115 int client_import_del_conn(struct obd_import *imp, struct obd_uuid *uuid);
1116 int import_set_conn_priority(struct obd_import *imp, struct obd_uuid *uuid);
1118 /* ptlrpc/pinger.c */
1119 int ptlrpc_pinger_add_import(struct obd_import *imp);
1120 int ptlrpc_pinger_del_import(struct obd_import *imp);
1121 cfs_time_t ptlrpc_suspend_wakeup_time(void);
1123 void ping_evictor_start(void);
1124 void ping_evictor_stop(void);
1126 #define ping_evictor_start() do {} while (0)
1127 #define ping_evictor_stop() do {} while (0)
1129 int ptlrpc_check_and_wait_suspend(struct ptlrpc_request *req);
1131 /* ptlrpc/ptlrpcd.c */
1132 int ptlrpcd_start(char *name, struct ptlrpcd_ctl *pc);
1133 void ptlrpcd_stop(struct ptlrpcd_ctl *pc, int force);
1134 void ptlrpcd_wake(struct ptlrpc_request *req);
1135 void ptlrpcd_add_req(struct ptlrpc_request *req);
1136 int ptlrpcd_addref(void);
1137 void ptlrpcd_decref(void);
1139 /* ptlrpc/lproc_ptlrpc.c */
1140 const char* ll_opcode2str(__u32 opcode);
1142 void ptlrpc_lprocfs_register_obd(struct obd_device *obd);
1143 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd);
1144 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes);
1146 static inline void ptlrpc_lprocfs_register_obd(struct obd_device *obd) {}
1147 static inline void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd) {}
1148 static inline void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes) {}
1151 /* ptlrpc/llog_server.c */
1152 int llog_origin_handle_create(struct ptlrpc_request *req);
1153 int llog_origin_handle_destroy(struct ptlrpc_request *req);
1154 int llog_origin_handle_prev_block(struct ptlrpc_request *req);
1155 int llog_origin_handle_next_block(struct ptlrpc_request *req);
1156 int llog_origin_handle_read_header(struct ptlrpc_request *req);
1157 int llog_origin_handle_close(struct ptlrpc_request *req);
1158 int llog_origin_handle_cancel(struct ptlrpc_request *req);
1159 int llog_catinfo(struct ptlrpc_request *req);
1161 /* ptlrpc/llog_client.c */
1162 extern struct llog_operations llog_client_ops;