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 #include <obd_support.h>
63 /* MD flags we _always_ use */
64 #define PTLRPC_MD_OPTIONS 0
66 /* Define maxima for bulk I/O
67 * CAVEAT EMPTOR, with multinet (i.e. routers forwarding between networks)
68 * these limits are system wide and not interface-local. */
69 #define PTLRPC_MAX_BRW_BITS LNET_MTU_BITS
70 #define PTLRPC_MAX_BRW_SIZE (1<<LNET_MTU_BITS)
71 #define PTLRPC_MAX_BRW_PAGES (PTLRPC_MAX_BRW_SIZE >> CFS_PAGE_SHIFT)
73 /* When PAGE_SIZE is a constant, we can check our arithmetic here with cpp! */
75 # if ((PTLRPC_MAX_BRW_PAGES & (PTLRPC_MAX_BRW_PAGES - 1)) != 0)
76 # error "PTLRPC_MAX_BRW_PAGES isn't a power of two"
78 # if (PTLRPC_MAX_BRW_SIZE != (PTLRPC_MAX_BRW_PAGES * CFS_PAGE_SIZE))
79 # error "PTLRPC_MAX_BRW_SIZE isn't PTLRPC_MAX_BRW_PAGES * CFS_PAGE_SIZE"
81 # if (PTLRPC_MAX_BRW_SIZE > LNET_MTU)
82 # error "PTLRPC_MAX_BRW_SIZE too big"
84 # if (PTLRPC_MAX_BRW_PAGES > LNET_MAX_IOV)
85 # error "PTLRPC_MAX_BRW_PAGES too big"
87 #endif /* __KERNEL__ */
89 /* Size over which to OBD_VMALLOC() rather than OBD_ALLOC() service request
91 #define SVC_BUF_VMALLOC_THRESHOLD (2 * CFS_PAGE_SIZE)
93 /* The following constants determine how memory is used to buffer incoming
96 * ?_NBUFS # buffers to allocate when growing the pool
97 * ?_BUFSIZE # bytes in a single request buffer
98 * ?_MAXREQSIZE # maximum request service will receive
100 * When fewer than ?_NBUFS/2 buffers are posted for receive, another chunk
101 * of ?_NBUFS is added to the pool.
103 * Messages larger than ?_MAXREQSIZE are dropped. Request buffers are
104 * considered full when less than ?_MAXREQSIZE is left in them.
107 #define LDLM_THREADS_AUTO_MIN (2)
108 #define LDLM_THREADS_AUTO_MAX min(num_online_cpus()*num_online_cpus()*32, 128)
109 #define LDLM_BL_THREADS LDLM_THREADS_AUTO_MIN
110 #define LDLM_NBUFS (64 * num_online_cpus())
111 #define LDLM_BUFSIZE (8 * 1024)
112 #define LDLM_MAXREQSIZE (5 * 1024)
113 #define LDLM_MAXREPSIZE (1024)
115 #define MDT_MIN_THREADS 2UL
116 #define MDT_MAX_THREADS 512UL
117 #define MDT_NUM_THREADS max(min_t(unsigned long, MDT_MAX_THREADS, \
118 num_physpages >> (25 - CFS_PAGE_SHIFT)), 2UL)
119 #define FLD_NUM_THREADS max(min_t(unsigned long, MDT_MAX_THREADS, \
120 num_physpages >> (25 - CFS_PAGE_SHIFT)), 2UL)
121 #define SEQ_NUM_THREADS max(min_t(unsigned long, MDT_MAX_THREADS, \
122 num_physpages >> (25 - CFS_PAGE_SHIFT)), 2UL)
124 /* Absolute limits */
125 #define MDS_THREADS_MIN 2
126 #define MDS_THREADS_MAX 512
127 #define MDS_THREADS_MIN_READPAGE 2
128 #define MDS_NBUFS (64 * num_online_cpus())
129 #define MDS_BUFSIZE (8 * 1024)
130 /* Assume file name length = FNAME_MAX = 256 (true for ext3).
131 * path name length = PATH_MAX = 4096
132 * LOV MD size max = EA_MAX = 4000
133 * symlink: FNAME_MAX + PATH_MAX <- largest
134 * link: FNAME_MAX + PATH_MAX (mds_rec_link < mds_rec_create)
135 * rename: FNAME_MAX + FNAME_MAX
136 * open: FNAME_MAX + EA_MAX
138 * MDS_MAXREQSIZE ~= 4736 bytes =
139 * lustre_msg + ldlm_request + mds_body + mds_rec_create + FNAME_MAX + PATH_MAX
140 * MDS_MAXREPSIZE ~= 8300 bytes = lustre_msg + llog_header
141 * or, for mds_close() and mds_reint_unlink() on a many-OST filesystem:
142 * = 9210 bytes = lustre_msg + mds_body + 160 * (easize + cookiesize)
144 * Realistic size is about 512 bytes (20 character name + 128 char symlink),
145 * except in the open case where there are a large number of OSTs in a LOV.
147 #define MDS_MAXREQSIZE (5 * 1024)
148 #define MDS_MAXREPSIZE max(9 * 1024, 362 + LOV_MAX_STRIPE_COUNT * 56)
150 /* FLD_MAXREQSIZE == lustre_msg + __u32 padding + ptlrpc_body + opc + md_fld */
151 #define FLD_MAXREQSIZE (160)
153 /* FLD_MAXREPSIZE == lustre_msg + ptlrpc_body + md_fld */
154 #define FLD_MAXREPSIZE (152)
156 /* SEQ_MAXREQSIZE == lustre_msg + __u32 padding + ptlrpc_body + opc + lu_range +
158 #define SEQ_MAXREQSIZE (160)
160 /* SEQ_MAXREPSIZE == lustre_msg + ptlrpc_body + lu_range */
161 #define SEQ_MAXREPSIZE (152)
163 #define MGS_THREADS_AUTO_MIN 2
164 #define MGS_THREADS_AUTO_MAX 32
165 #define MGS_NBUFS (64 * num_online_cpus())
166 #define MGS_BUFSIZE (8 * 1024)
167 #define MGS_MAXREQSIZE (7 * 1024)
168 #define MGS_MAXREPSIZE (9 * 1024)
170 /* Absolute limits */
171 #define OSS_THREADS_MIN 3 /* difficult replies, HPQ, others */
172 #define OSS_THREADS_MAX 512
173 #define OST_NBUFS (64 * num_online_cpus())
174 #define OST_BUFSIZE (8 * 1024)
175 /* OST_MAXREQSIZE ~= 4768 bytes =
176 * lustre_msg + obdo + 16 * obd_ioobj + 256 * niobuf_remote
178 * - single object with 16 pages is 512 bytes
179 * - OST_MAXREQSIZE must be at least 1 page of cookies plus some spillover
181 #define OST_MAXREQSIZE (5 * 1024)
182 #define OST_MAXREPSIZE (9 * 1024)
184 /* Macro to hide a typecast. */
185 #define ptlrpc_req_async_args(req) ((void *)&req->rq_async_args)
187 struct ptlrpc_connection {
188 struct hlist_node c_hash;
190 lnet_process_id_t c_peer;
191 struct obd_uuid c_remote_uuid;
195 struct ptlrpc_client {
196 __u32 cli_request_portal;
197 __u32 cli_reply_portal;
201 /* state flags of requests */
202 /* XXX only ones left are those used by the bulk descs as well! */
203 #define PTL_RPC_FL_INTR (1 << 0) /* reply wait was interrupted by user */
204 #define PTL_RPC_FL_TIMEOUT (1 << 7) /* request timed out waiting for reply */
206 #define REQ_MAX_ACK_LOCKS 8
208 union ptlrpc_async_args {
209 /* Scratchpad for passing args to completion interpreter. Users
210 * cast to the struct of their choosing, and LASSERT that this is
211 * big enough. For _tons_ of context, OBD_ALLOC a struct and store
212 * a pointer to it here. The pointer_arg ensures this struct is at
213 * least big enough for that. */
214 void *pointer_arg[11];
218 struct ptlrpc_request_set;
219 typedef int (*set_interpreter_func)(struct ptlrpc_request_set *, void *, int);
221 struct ptlrpc_request_set {
222 int set_remaining; /* # uncompleted requests */
223 cfs_waitq_t set_waitq;
224 cfs_waitq_t *set_wakeup_ptr;
225 struct list_head set_requests;
226 struct list_head set_cblist; /* list of completion callbacks */
227 set_interpreter_func set_interpret; /* completion callback */
228 void *set_arg; /* completion context */
229 /* locked so that any old caller can communicate requests to
230 * the set holder who can then fold them into the lock-free set */
231 spinlock_t set_new_req_lock;
232 struct list_head set_new_requests;
235 struct ptlrpc_set_cbdata {
236 struct list_head psc_item;
237 set_interpreter_func psc_interpret;
241 struct ptlrpc_bulk_desc;
244 * ptlrpc callback & work item stuff
246 struct ptlrpc_cb_id {
247 void (*cbid_fn)(lnet_event_t *ev); /* specific callback fn */
248 void *cbid_arg; /* additional arg */
251 #define RS_MAX_LOCKS 8
254 struct ptlrpc_reply_state {
255 struct ptlrpc_cb_id rs_cb_id;
256 struct list_head rs_list;
257 struct list_head rs_exp_list;
258 struct list_head rs_obd_list;
260 struct list_head rs_debug_list;
262 /* A spinlock to protect the reply state flags */
264 /* Reply state flags */
265 unsigned long rs_difficult:1; /* ACK/commit stuff */
266 unsigned long rs_no_ack:1; /* no ACK, even for
267 difficult requests */
268 unsigned long rs_scheduled:1; /* being handled? */
269 unsigned long rs_scheduled_ever:1;/* any schedule attempts? */
270 unsigned long rs_handled:1; /* been handled yet? */
271 unsigned long rs_on_net:1; /* reply_out_callback pending? */
272 unsigned long rs_prealloc:1; /* rs from prealloc list */
278 struct obd_export *rs_export;
279 struct ptlrpc_service *rs_service;
280 lnet_handle_md_t rs_md_h;
281 atomic_t rs_refcount;
283 struct ptlrpc_svc_ctx *rs_svc_ctx;
284 struct lustre_msg *rs_repbuf; /* wrapper */
285 int rs_repbuf_len; /* wrapper buf length */
286 int rs_repdata_len; /* wrapper msg length */
287 struct lustre_msg *rs_msg; /* reply message */
289 /* locks awaiting client reply ACK */
291 struct lustre_handle rs_locks[RS_MAX_LOCKS];
292 ldlm_mode_t rs_modes[RS_MAX_LOCKS];
295 struct ptlrpc_thread;
298 RQ_PHASE_NEW = 0xebc0de00,
299 RQ_PHASE_RPC = 0xebc0de01,
300 RQ_PHASE_BULK = 0xebc0de02,
301 RQ_PHASE_INTERPRET = 0xebc0de03,
302 RQ_PHASE_COMPLETE = 0xebc0de04,
303 RQ_PHASE_UNREGISTERING = 0xebc0de05,
304 RQ_PHASE_UNDEFINED = 0xebc0de06
307 /** Type of request interpreter call-back */
308 typedef int (*ptlrpc_interpterer_t)(const struct lu_env *env,
309 struct ptlrpc_request *req,
312 struct ptlrpc_request_pool {
314 struct list_head prp_req_list; /* list of ptlrpc_request structs */
316 void (*prp_populate)(struct ptlrpc_request_pool *, int);
324 struct ptlrpc_hpreq_ops {
326 * Check if the lock handle of the given lock is the same as
327 * taken from the request.
329 int (*hpreq_lock_match)(struct ptlrpc_request *, struct ldlm_lock *);
331 * Check if the request is a high priority one.
333 int (*hpreq_check)(struct ptlrpc_request *);
337 * Represents remote procedure call.
339 struct ptlrpc_request {
340 int rq_type; /* one of PTL_RPC_MSG_* */
341 struct list_head rq_list;
342 struct list_head rq_timed_list; /* server-side early replies */
343 struct list_head rq_history_list; /* server-side history */
344 struct list_head rq_exp_list; /* server-side per-export list */
345 struct ptlrpc_hpreq_ops *rq_ops; /* server-side hp handlers */
346 __u64 rq_history_seq; /* history sequence # */
347 /* the index of service's srv_at_array into which request is linked */
351 /* client-side flags are serialized by rq_lock */
352 unsigned long rq_intr:1, rq_replied:1, rq_err:1,
353 rq_timedout:1, rq_resend:1, rq_restart:1,
355 * when ->rq_replay is set, request is kept by the client even
356 * after server commits corresponding transaction. This is
357 * used for operations that require sequence of multiple
358 * requests to be replayed. The only example currently is file
359 * open/close. When last request in such a sequence is
360 * committed, ->rq_replay is cleared on all requests in the
364 rq_no_resend:1, rq_waiting:1, rq_receiving_reply:1,
365 rq_no_delay:1, rq_net_err:1, rq_wait_ctx:1,
366 rq_early:1, rq_must_unlink:1,
367 rq_fake:1, /* this fake req */
368 /* server-side flags */
369 rq_packed_final:1, /* packed final reply */
370 rq_hp:1, /* high priority RPC */
371 rq_at_linked:1, /* link into service's srv_at_array */
374 enum rq_phase rq_phase; /* one of RQ_PHASE_* */
375 enum rq_phase rq_next_phase; /* one of RQ_PHASE_* to be used next */
376 atomic_t rq_refcount; /* client-side refcount for SENT race,
377 server-side refcounf for multiple replies */
379 struct ptlrpc_thread *rq_svc_thread; /* initial thread servicing req */
381 int rq_request_portal; /* XXX FIXME bug 249 */
382 int rq_reply_portal; /* XXX FIXME bug 249 */
384 int rq_nob_received; /* client-side # reply bytes actually received */
387 struct lustre_msg *rq_reqmsg;
390 struct lustre_msg *rq_repmsg;
393 struct list_head rq_replay_list;
395 struct ptlrpc_cli_ctx *rq_cli_ctx; /* client's half ctx */
396 struct ptlrpc_svc_ctx *rq_svc_ctx; /* server's half ctx */
397 struct list_head rq_ctx_chain; /* link to waited ctx */
399 struct sptlrpc_flavor rq_flvr; /* client & server */
400 enum lustre_sec_part rq_sp_from;
402 unsigned long /* client/server security flags */
403 rq_ctx_init:1, /* context initiation */
404 rq_ctx_fini:1, /* context destroy */
405 rq_bulk_read:1, /* request bulk read */
406 rq_bulk_write:1, /* request bulk write */
407 /* server authentication flags */
408 rq_auth_gss:1, /* authenticated by gss */
409 rq_auth_remote:1, /* authed as remote user */
410 rq_auth_usr_root:1, /* authed as root */
411 rq_auth_usr_mdt:1, /* authed as mdt */
412 /* security tfm flags */
415 /* doesn't expect reply FIXME */
418 uid_t rq_auth_uid; /* authed uid */
419 uid_t rq_auth_mapped_uid; /* authed uid mapped to */
421 /* (server side), pointed directly into req buffer */
422 struct ptlrpc_user_desc *rq_user_desc;
424 /* early replies go to offset 0, regular replies go after that */
425 unsigned int rq_reply_off;
427 /* various buffer pointers */
428 struct lustre_msg *rq_reqbuf; /* req wrapper */
429 int rq_reqbuf_len; /* req wrapper buf len */
430 int rq_reqdata_len; /* req wrapper msg len */
431 char *rq_repbuf; /* rep buffer */
432 int rq_repbuf_len; /* rep buffer len */
433 struct lustre_msg *rq_repdata; /* rep wrapper msg */
434 int rq_repdata_len; /* rep wrapper msg len */
435 struct lustre_msg *rq_clrbuf; /* only in priv mode */
436 int rq_clrbuf_len; /* only in priv mode */
437 int rq_clrdata_len; /* only in priv mode */
439 __u32 rq_req_swab_mask;
440 __u32 rq_rep_swab_mask;
442 int rq_import_generation;
443 enum lustre_imp_state rq_send_state;
445 int rq_early_count; /* how many early replies (for stats) */
447 /* client+server request */
448 lnet_handle_md_t rq_req_md_h;
449 struct ptlrpc_cb_id rq_req_cbid;
452 struct timeval rq_arrival_time; /* request arrival time */
453 struct ptlrpc_reply_state *rq_reply_state; /* separated reply state */
454 struct ptlrpc_request_buffer_desc *rq_rqbd; /* incoming request buffer*/
456 __u32 rq_uid; /* peer uid, used in MDS only */
459 /* client-only incoming reply */
460 lnet_handle_md_t rq_reply_md_h;
461 cfs_waitq_t rq_reply_waitq;
462 struct ptlrpc_cb_id rq_reply_cbid;
465 lnet_process_id_t rq_peer;
466 struct obd_export *rq_export;
467 struct obd_import *rq_import;
469 void (*rq_replay_cb)(struct ptlrpc_request *);
470 void (*rq_commit_cb)(struct ptlrpc_request *);
473 struct ptlrpc_bulk_desc *rq_bulk;/* client side bulk */
475 /* client outgoing req */
476 time_t rq_sent; /* when request/reply sent (secs), or
477 * time when request should be sent */
479 volatile time_t rq_deadline; /* when request must finish. volatile
480 so that servers' early reply updates to the deadline aren't
481 kept in per-cpu cache */
482 time_t rq_reply_deadline; /* when req reply unlink must finish. */
483 time_t rq_bulk_deadline; /* when req bulk unlink must finish. */
484 int rq_timeout; /* service time estimate (secs) */
487 struct list_head rq_set_chain;
488 struct ptlrpc_request_set *rq_set;
489 /** Async completion handler */
490 ptlrpc_interpterer_t rq_interpret_reply;
491 union ptlrpc_async_args rq_async_args; /* Async completion context */
492 struct ptlrpc_request_pool *rq_pool; /* Pool if request from
494 struct lu_context rq_session;
495 struct lu_context rq_recov_session;
498 struct req_capsule rq_pill;
501 static inline int ptlrpc_req_interpret(const struct lu_env *env,
502 struct ptlrpc_request *req, int rc)
504 if (req->rq_interpret_reply != NULL) {
505 req->rq_status = req->rq_interpret_reply(env, req,
508 return req->rq_status;
513 static inline int lustre_req_swabbed(struct ptlrpc_request *req, int index)
515 LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
516 return req->rq_req_swab_mask & (1 << index);
519 static inline int lustre_rep_swabbed(struct ptlrpc_request *req, int index)
521 LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
522 return req->rq_rep_swab_mask & (1 << index);
525 static inline int ptlrpc_req_need_swab(struct ptlrpc_request *req)
527 return lustre_req_swabbed(req, MSG_PTLRPC_HEADER_OFF);
530 static inline int ptlrpc_rep_need_swab(struct ptlrpc_request *req)
532 return lustre_rep_swabbed(req, MSG_PTLRPC_HEADER_OFF);
535 static inline void lustre_set_req_swabbed(struct ptlrpc_request *req, int index)
537 LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
538 LASSERT((req->rq_req_swab_mask & (1 << index)) == 0);
539 req->rq_req_swab_mask |= 1 << index;
542 static inline void lustre_set_rep_swabbed(struct ptlrpc_request *req, int index)
544 LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
545 LASSERT((req->rq_rep_swab_mask & (1 << index)) == 0);
546 req->rq_rep_swab_mask |= 1 << index;
549 static inline const char *
550 ptlrpc_phase2str(enum rq_phase phase)
559 case RQ_PHASE_INTERPRET:
561 case RQ_PHASE_COMPLETE:
563 case RQ_PHASE_UNREGISTERING:
564 return "Unregistering";
570 static inline const char *
571 ptlrpc_rqphase2str(struct ptlrpc_request *req)
573 return ptlrpc_phase2str(req->rq_phase);
576 /* Spare the preprocessor, spoil the bugs. */
577 #define FLAG(field, str) (field ? str : "")
579 #define DEBUG_REQ_FLAGS(req) \
580 ptlrpc_rqphase2str(req), \
581 FLAG(req->rq_intr, "I"), FLAG(req->rq_replied, "R"), \
582 FLAG(req->rq_err, "E"), \
583 FLAG(req->rq_timedout, "X") /* eXpired */, FLAG(req->rq_resend, "S"), \
584 FLAG(req->rq_restart, "T"), FLAG(req->rq_replay, "P"), \
585 FLAG(req->rq_no_resend, "N"), \
586 FLAG(req->rq_waiting, "W"), \
587 FLAG(req->rq_wait_ctx, "C"), FLAG(req->rq_hp, "H")
589 #define REQ_FLAGS_FMT "%s:%s%s%s%s%s%s%s%s%s%s%s"
591 void _debug_req(struct ptlrpc_request *req, __u32 mask,
592 struct libcfs_debug_msg_data *data, const char *fmt, ...)
593 __attribute__ ((format (printf, 4, 5)));
595 #define debug_req(cdls, level, req, file, func, line, fmt, a...) \
599 if (((level) & D_CANTMASK) != 0 || \
600 ((libcfs_debug & (level)) != 0 && \
601 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) { \
602 static struct libcfs_debug_msg_data _req_dbg_data = \
603 DEBUG_MSG_DATA_INIT(cdls, DEBUG_SUBSYSTEM, file, func, line); \
604 _debug_req((req), (level), &_req_dbg_data, fmt, ##a); \
608 /* for most callers (level is a constant) this is resolved at compile time */
609 #define DEBUG_REQ(level, req, fmt, args...) \
611 if ((level) & (D_ERROR | D_WARNING)) { \
612 static cfs_debug_limit_state_t cdls; \
613 debug_req(&cdls, level, req, __FILE__, __func__, __LINE__, \
614 "@@@ "fmt" ", ## args); \
616 debug_req(NULL, level, req, __FILE__, __func__, __LINE__, \
617 "@@@ "fmt" ", ## args); \
620 struct ptlrpc_bulk_page {
621 struct list_head bp_link;
623 int bp_pageoffset; /* offset within a page */
624 struct page *bp_page;
627 #define BULK_GET_SOURCE 0
628 #define BULK_PUT_SINK 1
629 #define BULK_GET_SINK 2
630 #define BULK_PUT_SOURCE 3
632 struct ptlrpc_bulk_desc {
633 unsigned long bd_success:1; /* completed successfully */
634 unsigned long bd_network_rw:1; /* accessible to the network */
635 unsigned long bd_type:2; /* {put,get}{source,sink} */
636 unsigned long bd_registered:1; /* client side */
637 spinlock_t bd_lock; /* serialise with callback */
638 int bd_import_generation;
639 struct obd_export *bd_export;
640 struct obd_import *bd_import;
642 struct ptlrpc_request *bd_req; /* associated request */
643 cfs_waitq_t bd_waitq; /* server side only WQ */
644 int bd_iov_count; /* # entries in bd_iov */
645 int bd_max_iov; /* allocated size of bd_iov */
646 int bd_nob; /* # bytes covered */
647 int bd_nob_transferred; /* # bytes GOT/PUT */
651 struct ptlrpc_cb_id bd_cbid; /* network callback info */
652 lnet_handle_md_t bd_md_h; /* associated MD */
653 lnet_nid_t bd_sender; /* stash event::sender */
655 #if defined(__KERNEL__)
657 * encrypt iov, size is either 0 or bd_iov_count.
659 lnet_kiov_t *bd_enc_iov;
661 lnet_kiov_t bd_iov[0];
663 lnet_md_iovec_t bd_iov[0];
667 struct ptlrpc_thread {
669 * active threads in svc->srv_threads
671 struct list_head t_link;
673 * thread-private data (preallocated memory)
678 * service thread index, from ptlrpc_start_threads
682 * put watchdog in the structure per thread b=14840
684 struct lc_watchdog *t_watchdog;
686 * the svc this thread belonged to b=18582
688 struct ptlrpc_service *t_svc;
689 cfs_waitq_t t_ctl_waitq;
690 struct lu_env *t_env;
693 struct ptlrpc_request_buffer_desc {
694 struct list_head rqbd_list;
695 struct list_head rqbd_reqs;
696 struct ptlrpc_service *rqbd_service;
697 lnet_handle_md_t rqbd_md_h;
700 struct ptlrpc_cb_id rqbd_cbid;
701 struct ptlrpc_request rqbd_req;
704 typedef int (*svc_handler_t)(struct ptlrpc_request *req);
705 typedef void (*svcreq_printfn_t)(void *, struct ptlrpc_request *);
706 typedef int (*svc_hpreq_handler_t)(struct ptlrpc_request *);
708 #define PTLRPC_SVC_HP_RATIO 10
710 struct ptlrpc_service {
711 struct list_head srv_list; /* chain thru all services */
712 int srv_max_req_size; /* biggest request to receive */
713 int srv_max_reply_size; /* biggest reply to send */
714 int srv_buf_size; /* size of individual buffers */
715 int srv_nbuf_per_group; /* # buffers to allocate in 1 group */
716 int srv_nbufs; /* total # req buffer descs allocated */
717 int srv_threads_min; /* threads to start at SOW */
718 int srv_threads_max; /* thread upper limit */
719 int srv_threads_started; /* index of last started thread */
720 int srv_threads_running; /* # running threads */
721 atomic_t srv_n_difficult_replies; /* # 'difficult' replies */
722 int srv_n_active_reqs; /* # reqs being served */
723 int srv_n_hpreq; /* # HPreqs being served */
724 cfs_duration_t srv_rqbd_timeout; /* timeout before re-posting reqs, in tick */
725 int srv_watchdog_factor; /* soft watchdog timeout multiplier */
726 unsigned srv_cpu_affinity:1; /* bind threads to CPUs */
727 unsigned srv_at_check:1; /* check early replies */
728 unsigned srv_is_stopping:1; /* under unregister_service */
729 cfs_time_t srv_at_checktime; /* debug */
731 __u32 srv_req_portal;
732 __u32 srv_rep_portal;
735 struct adaptive_timeout srv_at_estimate;/* estimated rpc service time */
736 spinlock_t srv_at_lock;
737 struct ptlrpc_at_array srv_at_array; /* reqs waiting for replies */
738 cfs_timer_t srv_at_timer; /* early reply timer */
740 int srv_n_queued_reqs; /* # reqs in either of the queues below */
741 int srv_hpreq_count; /* # hp requests handled */
742 int srv_hpreq_ratio; /* # hp per lp reqs to handle */
743 struct list_head srv_req_in_queue; /* incoming reqs */
744 struct list_head srv_request_queue; /* reqs waiting for service */
745 struct list_head srv_request_hpq; /* high priority queue */
747 struct list_head srv_request_history; /* request history */
748 __u64 srv_request_seq; /* next request sequence # */
749 __u64 srv_request_max_cull_seq; /* highest seq culled from history */
750 svcreq_printfn_t srv_request_history_print_fn; /* service-specific print fn */
752 struct list_head srv_idle_rqbds; /* request buffers to be reposted */
753 struct list_head srv_active_rqbds; /* req buffers receiving */
754 struct list_head srv_history_rqbds; /* request buffer history */
755 int srv_nrqbd_receiving; /* # posted request buffers */
756 int srv_n_history_rqbds; /* # request buffers in history */
757 int srv_max_history_rqbds;/* max # request buffers in history */
759 atomic_t srv_outstanding_replies;
760 struct list_head srv_active_replies; /* all the active replies */
762 struct list_head srv_reply_queue; /* replies waiting for service */
764 cfs_waitq_t srv_waitq; /* all threads sleep on this. This
765 * wait-queue is signalled when new
766 * incoming request arrives and when
767 * difficult reply has to be handled. */
769 struct list_head srv_threads; /* service thread list */
770 svc_handler_t srv_handler;
771 svc_hpreq_handler_t srv_hpreq_handler; /* hp request handler */
773 char *srv_name; /* only statically allocated strings here; we don't clean them */
774 char *srv_thread_name; /* only statically allocated strings here; we don't clean them */
778 cfs_proc_dir_entry_t *srv_procroot;
779 struct lprocfs_stats *srv_stats;
781 /* List of free reply_states */
782 struct list_head srv_free_rs_list;
783 /* waitq to run, when adding stuff to srv_free_rs_list */
784 cfs_waitq_t srv_free_rs_waitq;
787 * Tags for lu_context associated with this thread, see struct
792 * if non-NULL called during thread creation (ptlrpc_start_thread())
793 * to initialize service specific per-thread state.
795 int (*srv_init)(struct ptlrpc_thread *thread);
797 * if non-NULL called during thread shutdown (ptlrpc_main()) to
798 * destruct state created by ->srv_init().
800 void (*srv_done)(struct ptlrpc_thread *thread);
802 //struct ptlrpc_srv_ni srv_interfaces[0];
807 * Ptlrpc thread control flags (LIOD_START, LIOD_STOP, LIOD_FORCE)
809 unsigned long pc_flags;
811 * Thread lock protecting structure fields.
817 struct completion pc_starting;
821 struct completion pc_finishing;
823 * Thread requests set.
825 struct ptlrpc_request_set *pc_set;
827 * Thread name used in cfs_daemonize()
831 * Environment for request interpreters to run in.
833 struct lu_env pc_env;
836 * Async rpcs flag to make sure that ptlrpcd_check() is called only
841 * Currently not used.
845 * User-space async rpcs callback.
847 void *pc_wait_callback;
849 * User-space check idle rpcs callback.
851 void *pc_idle_callback;
855 /* Bits for pc_flags */
856 enum ptlrpcd_ctl_flags {
858 * Ptlrpc thread start flag.
862 * Ptlrpc thread stop flag.
866 * Ptlrpc thread force flag (only stop force so far).
867 * This will cause aborting any inflight rpcs handled
868 * by thread if LIOD_STOP is specified.
872 * This is a recovery ptlrpc thread.
874 LIOD_RECOVERY = 1 << 3
877 /* ptlrpc/events.c */
878 extern lnet_handle_eq_t ptlrpc_eq_h;
879 extern int ptlrpc_uuid_to_peer(struct obd_uuid *uuid,
880 lnet_process_id_t *peer, lnet_nid_t *self);
881 extern void request_out_callback (lnet_event_t *ev);
882 extern void reply_in_callback(lnet_event_t *ev);
883 extern void client_bulk_callback (lnet_event_t *ev);
884 extern void request_in_callback(lnet_event_t *ev);
885 extern void reply_out_callback(lnet_event_t *ev);
886 extern void server_bulk_callback (lnet_event_t *ev);
888 /* ptlrpc/connection.c */
889 struct ptlrpc_connection *ptlrpc_connection_get(lnet_process_id_t peer,
891 struct obd_uuid *uuid);
892 int ptlrpc_connection_put(struct ptlrpc_connection *c);
893 struct ptlrpc_connection *ptlrpc_connection_addref(struct ptlrpc_connection *);
894 int ptlrpc_connection_init(void);
895 void ptlrpc_connection_fini(void);
896 extern lnet_pid_t ptl_get_pid(void);
898 /* ptlrpc/niobuf.c */
899 int ptlrpc_start_bulk_transfer(struct ptlrpc_bulk_desc *desc);
900 void ptlrpc_abort_bulk(struct ptlrpc_bulk_desc *desc);
901 int ptlrpc_register_bulk(struct ptlrpc_request *req);
902 int ptlrpc_unregister_bulk(struct ptlrpc_request *req, int async);
904 static inline int ptlrpc_server_bulk_active(struct ptlrpc_bulk_desc *desc)
908 LASSERT(desc != NULL);
910 spin_lock(&desc->bd_lock);
911 rc = desc->bd_network_rw;
912 spin_unlock(&desc->bd_lock);
916 static inline int ptlrpc_client_bulk_active(struct ptlrpc_request *req)
918 struct ptlrpc_bulk_desc *desc = req->rq_bulk;
921 LASSERT(req != NULL);
923 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_BULK_UNLINK) &&
924 req->rq_bulk_deadline > cfs_time_current_sec())
930 spin_lock(&desc->bd_lock);
931 rc = desc->bd_network_rw;
932 spin_unlock(&desc->bd_lock);
936 #define PTLRPC_REPLY_MAYBE_DIFFICULT 0x01
937 #define PTLRPC_REPLY_EARLY 0x02
938 int ptlrpc_send_reply(struct ptlrpc_request *req, int flags);
939 int ptlrpc_reply(struct ptlrpc_request *req);
940 int ptlrpc_send_error(struct ptlrpc_request *req, int difficult);
941 int ptlrpc_error(struct ptlrpc_request *req);
942 void ptlrpc_resend_req(struct ptlrpc_request *request);
943 int ptlrpc_at_get_net_latency(struct ptlrpc_request *req);
944 int ptl_send_rpc(struct ptlrpc_request *request, int noreply);
945 int ptlrpc_register_rqbd (struct ptlrpc_request_buffer_desc *rqbd);
947 /* ptlrpc/client.c */
948 void ptlrpc_init_client(int req_portal, int rep_portal, char *name,
949 struct ptlrpc_client *);
950 void ptlrpc_cleanup_client(struct obd_import *imp);
951 struct ptlrpc_connection *ptlrpc_uuid_to_connection(struct obd_uuid *uuid);
953 int ptlrpc_queue_wait(struct ptlrpc_request *req);
954 int ptlrpc_replay_req(struct ptlrpc_request *req);
955 int ptlrpc_unregister_reply(struct ptlrpc_request *req, int async);
956 void ptlrpc_restart_req(struct ptlrpc_request *req);
957 void ptlrpc_abort_inflight(struct obd_import *imp);
958 void ptlrpc_cleanup_imp(struct obd_import *imp);
959 void ptlrpc_abort_set(struct ptlrpc_request_set *set);
961 struct ptlrpc_request_set *ptlrpc_prep_set(void);
962 int ptlrpc_set_add_cb(struct ptlrpc_request_set *set,
963 set_interpreter_func fn, void *data);
964 int ptlrpc_set_next_timeout(struct ptlrpc_request_set *);
965 int ptlrpc_check_set(const struct lu_env *env, struct ptlrpc_request_set *set);
966 int ptlrpc_set_wait(struct ptlrpc_request_set *);
967 int ptlrpc_expired_set(void *data);
968 void ptlrpc_interrupted_set(void *data);
969 void ptlrpc_mark_interrupted(struct ptlrpc_request *req);
970 void ptlrpc_set_destroy(struct ptlrpc_request_set *);
971 void ptlrpc_set_add_req(struct ptlrpc_request_set *, struct ptlrpc_request *);
972 int ptlrpc_set_add_new_req(struct ptlrpcd_ctl *pc,
973 struct ptlrpc_request *req);
975 void ptlrpc_free_rq_pool(struct ptlrpc_request_pool *pool);
976 void ptlrpc_add_rqs_to_pool(struct ptlrpc_request_pool *pool, int num_rq);
978 struct ptlrpc_request_pool *
979 ptlrpc_init_rq_pool(int, int,
980 void (*populate_pool)(struct ptlrpc_request_pool *, int));
982 void ptlrpc_at_set_req_timeout(struct ptlrpc_request *req);
983 struct ptlrpc_request *ptlrpc_request_alloc(struct obd_import *imp,
984 const struct req_format *format);
985 struct ptlrpc_request *ptlrpc_request_alloc_pool(struct obd_import *imp,
986 struct ptlrpc_request_pool *,
987 const struct req_format *format);
988 void ptlrpc_request_free(struct ptlrpc_request *request);
989 int ptlrpc_request_pack(struct ptlrpc_request *request,
990 __u32 version, int opcode);
991 struct ptlrpc_request *ptlrpc_request_alloc_pack(struct obd_import *imp,
992 const struct req_format *format,
993 __u32 version, int opcode);
994 int ptlrpc_request_bufs_pack(struct ptlrpc_request *request,
995 __u32 version, int opcode, char **bufs,
996 struct ptlrpc_cli_ctx *ctx);
997 struct ptlrpc_request *ptlrpc_prep_fakereq(struct obd_import *imp,
998 unsigned int timeout,
999 ptlrpc_interpterer_t interpreter);
1000 void ptlrpc_fakereq_finished(struct ptlrpc_request *req);
1002 struct ptlrpc_request *ptlrpc_prep_req(struct obd_import *imp, __u32 version,
1003 int opcode, int count, __u32 *lengths,
1005 struct ptlrpc_request *ptlrpc_prep_req_pool(struct obd_import *imp,
1006 __u32 version, int opcode,
1007 int count, __u32 *lengths, char **bufs,
1008 struct ptlrpc_request_pool *pool);
1009 void ptlrpc_req_finished(struct ptlrpc_request *request);
1010 void ptlrpc_req_finished_with_imp_lock(struct ptlrpc_request *request);
1011 struct ptlrpc_request *ptlrpc_request_addref(struct ptlrpc_request *req);
1012 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_imp (struct ptlrpc_request *req,
1013 int npages, int type, int portal);
1014 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_exp(struct ptlrpc_request *req,
1015 int npages, int type, int portal);
1016 void ptlrpc_free_bulk(struct ptlrpc_bulk_desc *bulk);
1017 void ptlrpc_prep_bulk_page(struct ptlrpc_bulk_desc *desc,
1018 cfs_page_t *page, int pageoffset, int len);
1019 void ptlrpc_retain_replayable_request(struct ptlrpc_request *req,
1020 struct obd_import *imp);
1021 __u64 ptlrpc_next_xid(void);
1022 __u64 ptlrpc_sample_next_xid(void);
1023 __u64 ptlrpc_req_xid(struct ptlrpc_request *request);
1025 struct ptlrpc_service_conf {
1028 int psc_max_req_size;
1029 int psc_max_reply_size;
1032 int psc_watchdog_factor;
1033 int psc_min_threads;
1034 int psc_max_threads;
1038 /* ptlrpc/service.c */
1039 void ptlrpc_save_lock (struct ptlrpc_request *req,
1040 struct lustre_handle *lock, int mode, int no_ack);
1041 void ptlrpc_commit_replies(struct obd_export *exp);
1042 void ptlrpc_dispatch_difficult_reply (struct ptlrpc_reply_state *rs);
1043 void ptlrpc_schedule_difficult_reply (struct ptlrpc_reply_state *rs);
1044 struct ptlrpc_service *ptlrpc_init_svc_conf(struct ptlrpc_service_conf *c,
1045 svc_handler_t h, char *name,
1046 struct proc_dir_entry *proc_entry,
1047 svcreq_printfn_t prntfn,
1050 struct ptlrpc_service *ptlrpc_init_svc(int nbufs, int bufsize, int max_req_size,
1052 int req_portal, int rep_portal,
1053 int watchdog_factor,
1054 svc_handler_t, char *name,
1055 cfs_proc_dir_entry_t *proc_entry,
1057 int min_threads, int max_threads,
1058 char *threadname, __u32 ctx_tags,
1059 svc_hpreq_handler_t);
1060 void ptlrpc_stop_all_threads(struct ptlrpc_service *svc);
1062 int ptlrpc_start_threads(struct obd_device *dev, struct ptlrpc_service *svc);
1063 int ptlrpc_start_thread(struct obd_device *dev, struct ptlrpc_service *svc);
1064 int ptlrpc_unregister_service(struct ptlrpc_service *service);
1065 int liblustre_check_services (void *arg);
1066 void ptlrpc_daemonize(char *name);
1067 int ptlrpc_service_health_check(struct ptlrpc_service *);
1068 void ptlrpc_hpreq_reorder(struct ptlrpc_request *req);
1069 void ptlrpc_server_drop_request(struct ptlrpc_request *req);
1072 int ptlrpc_hr_init(void);
1073 void ptlrpc_hr_fini(void);
1075 # define ptlrpc_hr_init() (0)
1076 # define ptlrpc_hr_fini() do {} while(0)
1079 struct ptlrpc_svc_data {
1081 struct ptlrpc_service *svc;
1082 struct ptlrpc_thread *thread;
1083 struct obd_device *dev;
1086 /* ptlrpc/import.c */
1087 int ptlrpc_connect_import(struct obd_import *imp, char * new_uuid);
1088 int ptlrpc_init_import(struct obd_import *imp);
1089 int ptlrpc_disconnect_import(struct obd_import *imp, int noclose);
1090 int ptlrpc_import_recovery_state_machine(struct obd_import *imp);
1092 /* ptlrpc/pack_generic.c */
1093 int ptlrpc_reconnect_import(struct obd_import *imp);
1095 /** ptlrpc mgs buffer swab interface */
1096 int ptlrpc_buf_need_swab(struct ptlrpc_request *req, const int inout,
1098 void ptlrpc_buf_set_swabbed(struct ptlrpc_request *req, const int inout,
1100 int ptlrpc_unpack_rep_msg(struct ptlrpc_request *req, int len);
1101 int ptlrpc_unpack_req_msg(struct ptlrpc_request *req, int len);
1103 int lustre_msg_check_version(struct lustre_msg *msg, __u32 version);
1104 void lustre_init_msg_v2(struct lustre_msg_v2 *msg, int count, __u32 *lens,
1106 int lustre_pack_request(struct ptlrpc_request *, __u32 magic, int count,
1107 __u32 *lens, char **bufs);
1108 int lustre_pack_reply(struct ptlrpc_request *, int count, __u32 *lens,
1110 int lustre_pack_reply_v2(struct ptlrpc_request *req, int count,
1111 __u32 *lens, char **bufs, int flags);
1112 #define LPRFL_EARLY_REPLY 1
1113 int lustre_pack_reply_flags(struct ptlrpc_request *, int count, __u32 *lens,
1114 char **bufs, int flags);
1115 int lustre_shrink_msg(struct lustre_msg *msg, int segment,
1116 unsigned int newlen, int move_data);
1117 void lustre_free_reply_state(struct ptlrpc_reply_state *rs);
1118 int __lustre_unpack_msg(struct lustre_msg *m, int len);
1119 int lustre_msg_hdr_size(__u32 magic, int count);
1120 int lustre_msg_size(__u32 magic, int count, __u32 *lengths);
1121 int lustre_msg_size_v2(int count, __u32 *lengths);
1122 int lustre_packed_msg_size(struct lustre_msg *msg);
1123 int lustre_msg_early_size(void);
1124 void *lustre_msg_buf_v2(struct lustre_msg_v2 *m, int n, int min_size);
1125 void *lustre_msg_buf(struct lustre_msg *m, int n, int minlen);
1126 int lustre_msg_buflen(struct lustre_msg *m, int n);
1127 void lustre_msg_set_buflen(struct lustre_msg *m, int n, int len);
1128 int lustre_msg_bufcount(struct lustre_msg *m);
1129 char *lustre_msg_string (struct lustre_msg *m, int n, int max_len);
1130 void *lustre_swab_reqbuf(struct ptlrpc_request *req, int n, int minlen,
1132 void *lustre_swab_repbuf(struct ptlrpc_request *req, int n, int minlen,
1134 __u32 lustre_msghdr_get_flags(struct lustre_msg *msg);
1135 void lustre_msghdr_set_flags(struct lustre_msg *msg, __u32 flags);
1136 __u32 lustre_msg_get_flags(struct lustre_msg *msg);
1137 void lustre_msg_add_flags(struct lustre_msg *msg, int flags);
1138 void lustre_msg_set_flags(struct lustre_msg *msg, int flags);
1139 void lustre_msg_clear_flags(struct lustre_msg *msg, int flags);
1140 __u32 lustre_msg_get_op_flags(struct lustre_msg *msg);
1141 void lustre_msg_add_op_flags(struct lustre_msg *msg, int flags);
1142 void lustre_msg_set_op_flags(struct lustre_msg *msg, int flags);
1143 struct lustre_handle *lustre_msg_get_handle(struct lustre_msg *msg);
1144 __u32 lustre_msg_get_type(struct lustre_msg *msg);
1145 __u32 lustre_msg_get_version(struct lustre_msg *msg);
1146 void lustre_msg_add_version(struct lustre_msg *msg, int version);
1147 __u32 lustre_msg_get_opc(struct lustre_msg *msg);
1148 __u64 lustre_msg_get_last_xid(struct lustre_msg *msg);
1149 __u64 lustre_msg_get_last_committed(struct lustre_msg *msg);
1150 __u64 *lustre_msg_get_versions(struct lustre_msg *msg);
1151 __u64 lustre_msg_get_transno(struct lustre_msg *msg);
1152 __u64 lustre_msg_get_slv(struct lustre_msg *msg);
1153 __u32 lustre_msg_get_limit(struct lustre_msg *msg);
1154 void lustre_msg_set_slv(struct lustre_msg *msg, __u64 slv);
1155 void lustre_msg_set_limit(struct lustre_msg *msg, __u64 limit);
1156 int lustre_msg_get_status(struct lustre_msg *msg);
1157 __u32 lustre_msg_get_conn_cnt(struct lustre_msg *msg);
1158 int lustre_msg_is_v1(struct lustre_msg *msg);
1159 __u32 lustre_msg_get_magic(struct lustre_msg *msg);
1160 __u32 lustre_msg_get_timeout(struct lustre_msg *msg);
1161 __u32 lustre_msg_get_service_time(struct lustre_msg *msg);
1162 __u32 lustre_msg_get_cksum(struct lustre_msg *msg);
1163 __u32 lustre_msg_calc_cksum(struct lustre_msg *msg);
1164 void lustre_msg_set_handle(struct lustre_msg *msg,struct lustre_handle *handle);
1165 void lustre_msg_set_type(struct lustre_msg *msg, __u32 type);
1166 void lustre_msg_set_opc(struct lustre_msg *msg, __u32 opc);
1167 void lustre_msg_set_last_xid(struct lustre_msg *msg, __u64 last_xid);
1168 void lustre_msg_set_last_committed(struct lustre_msg *msg,__u64 last_committed);
1169 void lustre_msg_set_versions(struct lustre_msg *msg, __u64 *versions);
1170 void lustre_msg_set_transno(struct lustre_msg *msg, __u64 transno);
1171 void lustre_msg_set_status(struct lustre_msg *msg, __u32 status);
1172 void lustre_msg_set_conn_cnt(struct lustre_msg *msg, __u32 conn_cnt);
1173 void ptlrpc_req_set_repsize(struct ptlrpc_request *req, int count, __u32 *sizes);
1174 void ptlrpc_request_set_replen(struct ptlrpc_request *req);
1175 void lustre_msg_set_timeout(struct lustre_msg *msg, __u32 timeout);
1176 void lustre_msg_set_service_time(struct lustre_msg *msg, __u32 service_time);
1177 void lustre_msg_set_cksum(struct lustre_msg *msg, __u32 cksum);
1180 lustre_shrink_reply(struct ptlrpc_request *req, int segment,
1181 unsigned int newlen, int move_data)
1183 LASSERT(req->rq_reply_state);
1184 LASSERT(req->rq_repmsg);
1185 req->rq_replen = lustre_shrink_msg(req->rq_repmsg, segment,
1190 ptlrpc_rqphase_move(struct ptlrpc_request *req, enum rq_phase new_phase)
1192 if (req->rq_phase == new_phase)
1195 if (new_phase == RQ_PHASE_UNREGISTERING) {
1196 req->rq_next_phase = req->rq_phase;
1198 atomic_inc(&req->rq_import->imp_unregistering);
1201 if (req->rq_phase == RQ_PHASE_UNREGISTERING) {
1203 atomic_dec(&req->rq_import->imp_unregistering);
1206 DEBUG_REQ(D_RPCTRACE, req, "move req \"%s\" -> \"%s\"",
1207 ptlrpc_rqphase2str(req), ptlrpc_phase2str(new_phase));
1209 req->rq_phase = new_phase;
1213 ptlrpc_client_early(struct ptlrpc_request *req)
1215 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
1216 req->rq_reply_deadline > cfs_time_current_sec())
1218 return req->rq_early;
1222 ptlrpc_client_replied(struct ptlrpc_request *req)
1224 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
1225 req->rq_reply_deadline > cfs_time_current_sec())
1227 return req->rq_replied;
1231 ptlrpc_client_recv(struct ptlrpc_request *req)
1233 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
1234 req->rq_reply_deadline > cfs_time_current_sec())
1236 return req->rq_receiving_reply;
1240 ptlrpc_client_recv_or_unlink(struct ptlrpc_request *req)
1244 spin_lock(&req->rq_lock);
1245 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
1246 req->rq_reply_deadline > cfs_time_current_sec()) {
1247 spin_unlock(&req->rq_lock);
1250 rc = req->rq_receiving_reply || req->rq_must_unlink;
1251 spin_unlock(&req->rq_lock);
1256 ptlrpc_client_wake_req(struct ptlrpc_request *req)
1258 if (req->rq_set == NULL)
1259 cfs_waitq_signal(&req->rq_reply_waitq);
1261 cfs_waitq_signal(&req->rq_set->set_waitq);
1265 ptlrpc_rs_addref(struct ptlrpc_reply_state *rs)
1267 LASSERT(atomic_read(&rs->rs_refcount) > 0);
1268 atomic_inc(&rs->rs_refcount);
1272 ptlrpc_rs_decref(struct ptlrpc_reply_state *rs)
1274 LASSERT(atomic_read(&rs->rs_refcount) > 0);
1275 if (atomic_dec_and_test(&rs->rs_refcount))
1276 lustre_free_reply_state(rs);
1279 /* Should only be called once per req */
1280 static inline void ptlrpc_req_drop_rs(struct ptlrpc_request *req)
1282 if (req->rq_reply_state == NULL)
1283 return; /* shouldn't occur */
1284 ptlrpc_rs_decref(req->rq_reply_state);
1285 req->rq_reply_state = NULL;
1286 req->rq_repmsg = NULL;
1289 static inline __u32 lustre_request_magic(struct ptlrpc_request *req)
1291 return lustre_msg_get_magic(req->rq_reqmsg);
1294 static inline int ptlrpc_req_get_repsize(struct ptlrpc_request *req)
1296 switch (req->rq_reqmsg->lm_magic) {
1297 case LUSTRE_MSG_MAGIC_V2:
1298 return req->rq_reqmsg->lm_repsize;
1300 LASSERTF(0, "incorrect message magic: %08x\n",
1301 req->rq_reqmsg->lm_magic);
1306 /* ldlm/ldlm_lib.c */
1307 int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg);
1308 int client_obd_cleanup(struct obd_device *obddev);
1309 int client_connect_import(const struct lu_env *env,
1310 struct obd_export **exp, struct obd_device *obd,
1311 struct obd_uuid *cluuid, struct obd_connect_data *,
1313 int client_disconnect_export(struct obd_export *exp);
1314 int client_import_add_conn(struct obd_import *imp, struct obd_uuid *uuid,
1316 int client_import_del_conn(struct obd_import *imp, struct obd_uuid *uuid);
1317 int import_set_conn_priority(struct obd_import *imp, struct obd_uuid *uuid);
1318 void client_destroy_import(struct obd_import *imp);
1320 /* ptlrpc/pinger.c */
1321 enum timeout_event {
1324 struct timeout_item;
1325 typedef int (*timeout_cb_t)(struct timeout_item *, void *);
1326 int ptlrpc_pinger_add_import(struct obd_import *imp);
1327 int ptlrpc_pinger_del_import(struct obd_import *imp);
1328 int ptlrpc_add_timeout_client(int time, enum timeout_event event,
1329 timeout_cb_t cb, void *data,
1330 struct list_head *obd_list);
1331 int ptlrpc_del_timeout_client(struct list_head *obd_list,
1332 enum timeout_event event);
1333 struct ptlrpc_request * ptlrpc_prep_ping(struct obd_import *imp);
1334 int ptlrpc_obd_ping(struct obd_device *obd);
1335 cfs_time_t ptlrpc_suspend_wakeup_time(void);
1337 void ping_evictor_start(void);
1338 void ping_evictor_stop(void);
1340 #define ping_evictor_start() do {} while (0)
1341 #define ping_evictor_stop() do {} while (0)
1343 int ptlrpc_check_and_wait_suspend(struct ptlrpc_request *req);
1345 /* ptlrpc/ptlrpcd.c */
1348 * Ptlrpcd scope is a set of two threads: ptlrpcd-foo and ptlrpcd-foo-rcv,
1349 * these threads are used to asynchronously send requests queued with
1350 * ptlrpcd_add_req(req, PCSOPE_FOO), and to handle completion call-backs for
1351 * such requests. Multiple scopes are needed to avoid dead-locks.
1353 enum ptlrpcd_scope {
1354 /** Scope of bulk read-write rpcs. */
1356 /** Everything else. */
1361 int ptlrpcd_start(const char *name, struct ptlrpcd_ctl *pc);
1362 void ptlrpcd_stop(struct ptlrpcd_ctl *pc, int force);
1363 void ptlrpcd_wake(struct ptlrpc_request *req);
1364 int ptlrpcd_add_req(struct ptlrpc_request *req, enum ptlrpcd_scope scope);
1365 void ptlrpcd_add_rqset(struct ptlrpc_request_set *set);
1366 int ptlrpcd_addref(void);
1367 void ptlrpcd_decref(void);
1369 /* ptlrpc/lproc_ptlrpc.c */
1370 const char* ll_opcode2str(__u32 opcode);
1372 void ptlrpc_lprocfs_register_obd(struct obd_device *obd);
1373 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd);
1374 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes);
1376 static inline void ptlrpc_lprocfs_register_obd(struct obd_device *obd) {}
1377 static inline void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd) {}
1378 static inline void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes) {}
1381 /* ptlrpc/llog_server.c */
1382 int llog_origin_handle_create(struct ptlrpc_request *req);
1383 int llog_origin_handle_destroy(struct ptlrpc_request *req);
1384 int llog_origin_handle_prev_block(struct ptlrpc_request *req);
1385 int llog_origin_handle_next_block(struct ptlrpc_request *req);
1386 int llog_origin_handle_read_header(struct ptlrpc_request *req);
1387 int llog_origin_handle_close(struct ptlrpc_request *req);
1388 int llog_origin_handle_cancel(struct ptlrpc_request *req);
1389 int llog_catinfo(struct ptlrpc_request *req);
1391 /* ptlrpc/llog_client.c */
1392 extern struct llog_operations llog_client_ops;