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 (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
33 * Copyright (c) 2011 Whamcloud, Inc.
36 * This file is part of Lustre, http://www.lustre.org/
37 * Lustre is a trademark of Sun Microsystems, Inc.
39 /** \defgroup PtlRPC Portal RPC and networking module.
41 * PortalRPC is the layer used by rest of lustre code to achieve network
42 * communications: establish connections with corresponding export and import
43 * states, listen for a service, send and receive RPCs.
44 * PortalRPC also includes base recovery framework: packet resending and
45 * replaying, reconnections, pinger.
47 * PortalRPC utilizes LNet as its transport layer.
61 #if defined(__linux__)
62 #include <linux/lustre_net.h>
63 #elif defined(__APPLE__)
64 #include <darwin/lustre_net.h>
65 #elif defined(__WINNT__)
66 #include <winnt/lustre_net.h>
68 #error Unsupported operating system.
71 #include <libcfs/libcfs.h>
73 #include <lnet/lnet.h>
74 #include <lustre/lustre_idl.h>
75 #include <lustre_ha.h>
76 #include <lustre_sec.h>
77 #include <lustre_import.h>
78 #include <lprocfs_status.h>
79 #include <lu_object.h>
80 #include <lustre_req_layout.h>
82 #include <obd_support.h>
83 #include <lustre_ver.h>
85 /* MD flags we _always_ use */
86 #define PTLRPC_MD_OPTIONS 0
89 * Define maxima for bulk I/O
90 * CAVEAT EMPTOR, with multinet (i.e. routers forwarding between networks)
91 * these limits are system wide and not interface-local. */
92 #define PTLRPC_MAX_BRW_BITS LNET_MTU_BITS
93 #define PTLRPC_MAX_BRW_SIZE (1<<LNET_MTU_BITS)
94 #define PTLRPC_MAX_BRW_PAGES (PTLRPC_MAX_BRW_SIZE >> CFS_PAGE_SHIFT)
96 /* When PAGE_SIZE is a constant, we can check our arithmetic here with cpp! */
98 # if ((PTLRPC_MAX_BRW_PAGES & (PTLRPC_MAX_BRW_PAGES - 1)) != 0)
99 # error "PTLRPC_MAX_BRW_PAGES isn't a power of two"
101 # if (PTLRPC_MAX_BRW_SIZE != (PTLRPC_MAX_BRW_PAGES * CFS_PAGE_SIZE))
102 # error "PTLRPC_MAX_BRW_SIZE isn't PTLRPC_MAX_BRW_PAGES * CFS_PAGE_SIZE"
104 # if (PTLRPC_MAX_BRW_SIZE > LNET_MTU)
105 # error "PTLRPC_MAX_BRW_SIZE too big"
107 # if (PTLRPC_MAX_BRW_PAGES > LNET_MAX_IOV)
108 # error "PTLRPC_MAX_BRW_PAGES too big"
110 #endif /* __KERNEL__ */
113 * The following constants determine how memory is used to buffer incoming
116 * ?_NBUFS # buffers to allocate when growing the pool
117 * ?_BUFSIZE # bytes in a single request buffer
118 * ?_MAXREQSIZE # maximum request service will receive
120 * When fewer than ?_NBUFS/2 buffers are posted for receive, another chunk
121 * of ?_NBUFS is added to the pool.
123 * Messages larger than ?_MAXREQSIZE are dropped. Request buffers are
124 * considered full when less than ?_MAXREQSIZE is left in them.
126 #define LDLM_THREADS_AUTO_MIN (2)
127 #define LDLM_THREADS_AUTO_MAX min_t(unsigned, cfs_num_online_cpus() * \
128 cfs_num_online_cpus() * 32, 128)
129 #define LDLM_BL_THREADS LDLM_THREADS_AUTO_MIN
130 #define LDLM_NBUFS (64 * cfs_num_online_cpus())
131 #define LDLM_BUFSIZE (8 * 1024)
132 #define LDLM_MAXREQSIZE (5 * 1024)
133 #define LDLM_MAXREPSIZE (1024)
135 /** Absolute limits */
136 #define MDT_MIN_THREADS 2UL
137 #ifndef MDT_MAX_THREADS
138 #define MDT_MAX_THREADS 512UL
140 #define MDS_NBUFS (64 * cfs_num_online_cpus())
142 * Assume file name length = FNAME_MAX = 256 (true for ext3).
143 * path name length = PATH_MAX = 4096
144 * LOV MD size max = EA_MAX = 48000 (2000 stripes)
145 * symlink: FNAME_MAX + PATH_MAX <- largest
146 * link: FNAME_MAX + PATH_MAX (mds_rec_link < mds_rec_create)
147 * rename: FNAME_MAX + FNAME_MAX
148 * open: FNAME_MAX + EA_MAX
150 * MDS_MAXREQSIZE ~= 4736 bytes =
151 * lustre_msg + ldlm_request + mdt_body + mds_rec_create + FNAME_MAX + PATH_MAX
152 * MDS_MAXREPSIZE ~= 8300 bytes = lustre_msg + llog_header
153 * or, for mds_close() and mds_reint_unlink() on a many-OST filesystem:
154 * = 9210 bytes = lustre_msg + mdt_body + 160 * (easize + cookiesize)
156 * Realistic size is about 512 bytes (20 character name + 128 char symlink),
157 * except in the open case where there are a large number of OSTs in a LOV.
159 #define MDS_MAXREPSIZE max(10 * 1024, 362 + LOV_MAX_STRIPE_COUNT * 56)
160 #define MDS_MAXREQSIZE MDS_MAXREPSIZE
162 /** MDS_BUFSIZE = max_reqsize + max sptlrpc payload size */
163 #define MDS_BUFSIZE (MDS_MAXREQSIZE + 1024)
165 /** FLD_MAXREQSIZE == lustre_msg + __u32 padding + ptlrpc_body + opc */
166 #define FLD_MAXREQSIZE (160)
168 /** FLD_MAXREPSIZE == lustre_msg + ptlrpc_body */
169 #define FLD_MAXREPSIZE (152)
172 * SEQ_MAXREQSIZE == lustre_msg + __u32 padding + ptlrpc_body + opc + lu_range +
174 #define SEQ_MAXREQSIZE (160)
176 /** SEQ_MAXREPSIZE == lustre_msg + ptlrpc_body + lu_range */
177 #define SEQ_MAXREPSIZE (152)
179 /** MGS threads must be >= 3, see bug 22458 comment #28 */
180 #define MGS_THREADS_AUTO_MIN 3
181 #define MGS_THREADS_AUTO_MAX 32
182 #define MGS_NBUFS (64 * cfs_num_online_cpus())
183 #define MGS_BUFSIZE (8 * 1024)
184 #define MGS_MAXREQSIZE (7 * 1024)
185 #define MGS_MAXREPSIZE (9 * 1024)
187 /** Absolute OSS limits */
188 #define OSS_THREADS_MIN 3 /* difficult replies, HPQ, others */
189 #define OSS_THREADS_MAX 512
190 #define OST_NBUFS (64 * cfs_num_online_cpus())
191 #define OST_BUFSIZE (8 * 1024)
194 * OST_MAXREQSIZE ~= 4768 bytes =
195 * lustre_msg + obdo + 16 * obd_ioobj + 256 * niobuf_remote
197 * - single object with 16 pages is 512 bytes
198 * - OST_MAXREQSIZE must be at least 1 page of cookies plus some spillover
200 #define OST_MAXREQSIZE (5 * 1024)
201 #define OST_MAXREPSIZE (9 * 1024)
203 /* Macro to hide a typecast. */
204 #define ptlrpc_req_async_args(req) ((void *)&req->rq_async_args)
207 * Structure to single define portal connection.
209 struct ptlrpc_connection {
210 /** linkage for connections hash table */
211 cfs_hlist_node_t c_hash;
212 /** Our own lnet nid for this connection */
214 /** Remote side nid for this connection */
215 lnet_process_id_t c_peer;
216 /** UUID of the other side */
217 struct obd_uuid c_remote_uuid;
218 /** reference counter for this connection */
219 cfs_atomic_t c_refcount;
222 /** Client definition for PortalRPC */
223 struct ptlrpc_client {
224 /** What lnet portal does this client send messages to by default */
225 __u32 cli_request_portal;
226 /** What portal do we expect replies on */
227 __u32 cli_reply_portal;
228 /** Name of the client */
232 /** state flags of requests */
233 /* XXX only ones left are those used by the bulk descs as well! */
234 #define PTL_RPC_FL_INTR (1 << 0) /* reply wait was interrupted by user */
235 #define PTL_RPC_FL_TIMEOUT (1 << 7) /* request timed out waiting for reply */
237 #define REQ_MAX_ACK_LOCKS 8
239 union ptlrpc_async_args {
241 * Scratchpad for passing args to completion interpreter. Users
242 * cast to the struct of their choosing, and CLASSERT that this is
243 * big enough. For _tons_ of context, OBD_ALLOC a struct and store
244 * a pointer to it here. The pointer_arg ensures this struct is at
245 * least big enough for that.
247 void *pointer_arg[11];
251 struct ptlrpc_request_set;
252 typedef int (*set_interpreter_func)(struct ptlrpc_request_set *, void *, int);
255 * Definition of request set structure.
256 * Request set is a list of requests (not necessary to the same target) that
257 * once populated with RPCs could be sent in parallel.
258 * There are two kinds of request sets. General purpose and with dedicated
259 * serving thread. Example of the latter is ptlrpcd set.
260 * For general purpose sets once request set started sending it is impossible
261 * to add new requests to such set.
262 * Provides a way to call "completion callbacks" when all requests in the set
265 struct ptlrpc_request_set {
266 cfs_atomic_t set_refcount;
267 /** number of in queue requests */
268 cfs_atomic_t set_new_count;
269 /** number of uncompleted requests */
270 cfs_atomic_t set_remaining;
271 /** wait queue to wait on for request events */
272 cfs_waitq_t set_waitq;
273 cfs_waitq_t *set_wakeup_ptr;
274 /** List of requests in the set */
275 cfs_list_t set_requests;
277 * List of completion callbacks to be called when the set is completed
278 * This is only used if \a set_interpret is NULL.
279 * Links struct ptlrpc_set_cbdata.
281 cfs_list_t set_cblist;
282 /** Completion callback, if only one. */
283 set_interpreter_func set_interpret;
284 /** opaq argument passed to completion \a set_interpret callback. */
287 * Lock for \a set_new_requests manipulations
288 * locked so that any old caller can communicate requests to
289 * the set holder who can then fold them into the lock-free set
291 cfs_spinlock_t set_new_req_lock;
292 /** List of new yet unsent requests. Only used with ptlrpcd now. */
293 cfs_list_t set_new_requests;
297 * Description of a single ptrlrpc_set callback
299 struct ptlrpc_set_cbdata {
300 /** List linkage item */
302 /** Pointer to interpreting function */
303 set_interpreter_func psc_interpret;
304 /** Opaq argument to pass to the callback */
308 struct ptlrpc_bulk_desc;
311 * ptlrpc callback & work item stuff
313 struct ptlrpc_cb_id {
314 void (*cbid_fn)(lnet_event_t *ev); /* specific callback fn */
315 void *cbid_arg; /* additional arg */
318 /** Maximum number of locks to fit into reply state */
319 #define RS_MAX_LOCKS 8
323 * Structure to define reply state on the server
324 * Reply state holds various reply message information. Also for "difficult"
325 * replies (rep-ack case) we store the state after sending reply and wait
326 * for the client to acknowledge the reception. In these cases locks could be
327 * added to the state for replay/failover consistency guarantees.
329 struct ptlrpc_reply_state {
330 /** Callback description */
331 struct ptlrpc_cb_id rs_cb_id;
332 /** Linkage for list of all reply states in a system */
334 /** Linkage for list of all reply states on same export */
335 cfs_list_t rs_exp_list;
336 /** Linkage for list of all reply states for same obd */
337 cfs_list_t rs_obd_list;
339 cfs_list_t rs_debug_list;
341 /** A spinlock to protect the reply state flags */
342 cfs_spinlock_t rs_lock;
343 /** Reply state flags */
344 unsigned long rs_difficult:1; /* ACK/commit stuff */
345 unsigned long rs_no_ack:1; /* no ACK, even for
346 difficult requests */
347 unsigned long rs_scheduled:1; /* being handled? */
348 unsigned long rs_scheduled_ever:1;/* any schedule attempts? */
349 unsigned long rs_handled:1; /* been handled yet? */
350 unsigned long rs_on_net:1; /* reply_out_callback pending? */
351 unsigned long rs_prealloc:1; /* rs from prealloc list */
352 unsigned long rs_committed:1;/* the transaction was committed
353 and the rs was dispatched
354 by ptlrpc_commit_replies */
355 /** Size of the state */
359 /** Transaction number */
363 struct obd_export *rs_export;
364 struct ptlrpc_service *rs_service;
365 /** Lnet metadata handle for the reply */
366 lnet_handle_md_t rs_md_h;
367 cfs_atomic_t rs_refcount;
369 /** Context for the sevice thread */
370 struct ptlrpc_svc_ctx *rs_svc_ctx;
371 /** Reply buffer (actually sent to the client), encoded if needed */
372 struct lustre_msg *rs_repbuf; /* wrapper */
373 /** Size of the reply buffer */
374 int rs_repbuf_len; /* wrapper buf length */
375 /** Size of the reply message */
376 int rs_repdata_len; /* wrapper msg length */
378 * Actual reply message. Its content is encrupted (if needed) to
379 * produce reply buffer for actual sending. In simple case
380 * of no network encryption we jus set \a rs_repbuf to \a rs_msg
382 struct lustre_msg *rs_msg; /* reply message */
384 /** Number of locks awaiting client ACK */
386 /** Handles of locks awaiting client reply ACK */
387 struct lustre_handle rs_locks[RS_MAX_LOCKS];
388 /** Lock modes of locks in \a rs_locks */
389 ldlm_mode_t rs_modes[RS_MAX_LOCKS];
392 struct ptlrpc_thread;
396 RQ_PHASE_NEW = 0xebc0de00,
397 RQ_PHASE_RPC = 0xebc0de01,
398 RQ_PHASE_BULK = 0xebc0de02,
399 RQ_PHASE_INTERPRET = 0xebc0de03,
400 RQ_PHASE_COMPLETE = 0xebc0de04,
401 RQ_PHASE_UNREGISTERING = 0xebc0de05,
402 RQ_PHASE_UNDEFINED = 0xebc0de06
405 /** Type of request interpreter call-back */
406 typedef int (*ptlrpc_interpterer_t)(const struct lu_env *env,
407 struct ptlrpc_request *req,
411 * Definition of request pool structure.
412 * The pool is used to store empty preallocated requests for the case
413 * when we would actually need to send something without performing
414 * any allocations (to avoid e.g. OOM).
416 struct ptlrpc_request_pool {
417 /** Locks the list */
418 cfs_spinlock_t prp_lock;
419 /** list of ptlrpc_request structs */
420 cfs_list_t prp_req_list;
421 /** Maximum message size that would fit into a rquest from this pool */
423 /** Function to allocate more requests for this pool */
424 void (*prp_populate)(struct ptlrpc_request_pool *, int);
433 * Basic request prioritization operations structure.
434 * The whole idea is centered around locks and RPCs that might affect locks.
435 * When a lock is contended we try to give priority to RPCs that might lead
436 * to fastest release of that lock.
437 * Currently only implemented for OSTs only in a way that makes all
438 * IO and truncate RPCs that are coming from a locked region where a lock is
439 * contended a priority over other requests.
441 struct ptlrpc_hpreq_ops {
443 * Check if the lock handle of the given lock is the same as
444 * taken from the request.
446 int (*hpreq_lock_match)(struct ptlrpc_request *, struct ldlm_lock *);
448 * Check if the request is a high priority one.
450 int (*hpreq_check)(struct ptlrpc_request *);
454 * Represents remote procedure call.
456 * This is a staple structure used by everybody wanting to send a request
459 struct ptlrpc_request {
460 /* Request type: one of PTL_RPC_MSG_* */
463 * Linkage item through which this request is included into
464 * sending/delayed lists on client and into rqbd list on server
468 * Server side list of incoming unserved requests sorted by arrival
469 * time. Traversed from time to time to notice about to expire
470 * requests and sent back "early replies" to clients to let them
471 * know server is alive and well, just very busy to service their
474 cfs_list_t rq_timed_list;
475 /** server-side history, used for debuging purposes. */
476 cfs_list_t rq_history_list;
477 /** server-side per-export list */
478 cfs_list_t rq_exp_list;
479 /** server-side hp handlers */
480 struct ptlrpc_hpreq_ops *rq_ops;
481 /** history sequence # */
482 __u64 rq_history_seq;
483 /** the index of service's srv_at_array into which request is linked */
485 /** Result of request processing */
487 /** Lock to protect request flags and some other important bits, like
490 cfs_spinlock_t rq_lock;
491 /** client-side flags are serialized by rq_lock */
492 unsigned long rq_intr:1, rq_replied:1, rq_err:1,
493 rq_timedout:1, rq_resend:1, rq_restart:1,
495 * when ->rq_replay is set, request is kept by the client even
496 * after server commits corresponding transaction. This is
497 * used for operations that require sequence of multiple
498 * requests to be replayed. The only example currently is file
499 * open/close. When last request in such a sequence is
500 * committed, ->rq_replay is cleared on all requests in the
504 rq_no_resend:1, rq_waiting:1, rq_receiving_reply:1,
505 rq_no_delay:1, rq_net_err:1, rq_wait_ctx:1,
506 rq_early:1, rq_must_unlink:1,
507 rq_fake:1, /* this fake req */
508 rq_memalloc:1, /* req originated from "kswapd" */
509 /* server-side flags */
510 rq_packed_final:1, /* packed final reply */
511 rq_hp:1, /* high priority RPC */
512 rq_at_linked:1, /* link into service's srv_at_array */
515 /* whether the "rq_set" is a valid one */
518 enum rq_phase rq_phase; /* one of RQ_PHASE_* */
519 enum rq_phase rq_next_phase; /* one of RQ_PHASE_* to be used next */
520 cfs_atomic_t rq_refcount;/* client-side refcount for SENT race,
521 server-side refcounf for multiple replies */
523 /** initial thread servicing this request */
524 struct ptlrpc_thread *rq_svc_thread;
526 /** Portal to which this request would be sent */
527 int rq_request_portal; /* XXX FIXME bug 249 */
528 /** Portal where to wait for reply and where reply would be sent */
529 int rq_reply_portal; /* XXX FIXME bug 249 */
533 * !rq_truncate : # reply bytes actually received,
534 * rq_truncate : required repbuf_len for resend
537 /** Request length */
539 /** Request message - what client sent */
540 struct lustre_msg *rq_reqmsg;
544 /** Reply message - server response */
545 struct lustre_msg *rq_repmsg;
546 /** Transaction number */
551 * List item to for replay list. Not yet commited requests get linked
553 * Also see \a rq_replay comment above.
555 cfs_list_t rq_replay_list;
558 * security and encryption data
560 struct ptlrpc_cli_ctx *rq_cli_ctx; /**< client's half ctx */
561 struct ptlrpc_svc_ctx *rq_svc_ctx; /**< server's half ctx */
562 cfs_list_t rq_ctx_chain; /**< link to waited ctx */
564 struct sptlrpc_flavor rq_flvr; /**< for client & server */
565 enum lustre_sec_part rq_sp_from;
567 unsigned long /* client/server security flags */
568 rq_ctx_init:1, /* context initiation */
569 rq_ctx_fini:1, /* context destroy */
570 rq_bulk_read:1, /* request bulk read */
571 rq_bulk_write:1, /* request bulk write */
572 /* server authentication flags */
573 rq_auth_gss:1, /* authenticated by gss */
574 rq_auth_remote:1, /* authed as remote user */
575 rq_auth_usr_root:1, /* authed as root */
576 rq_auth_usr_mdt:1, /* authed as mdt */
577 rq_auth_usr_ost:1, /* authed as ost */
578 /* security tfm flags */
581 /* doesn't expect reply FIXME */
583 rq_pill_init:1; /* pill initialized */
585 uid_t rq_auth_uid; /* authed uid */
586 uid_t rq_auth_mapped_uid; /* authed uid mapped to */
588 /* (server side), pointed directly into req buffer */
589 struct ptlrpc_user_desc *rq_user_desc;
591 /** early replies go to offset 0, regular replies go after that */
592 unsigned int rq_reply_off;
594 /* various buffer pointers */
595 struct lustre_msg *rq_reqbuf; /* req wrapper */
596 int rq_reqbuf_len; /* req wrapper buf len */
597 int rq_reqdata_len; /* req wrapper msg len */
598 char *rq_repbuf; /* rep buffer */
599 int rq_repbuf_len; /* rep buffer len */
600 struct lustre_msg *rq_repdata; /* rep wrapper msg */
601 int rq_repdata_len; /* rep wrapper msg len */
602 struct lustre_msg *rq_clrbuf; /* only in priv mode */
603 int rq_clrbuf_len; /* only in priv mode */
604 int rq_clrdata_len; /* only in priv mode */
608 /** Fields that help to see if request and reply were swabbed or not */
609 __u32 rq_req_swab_mask;
610 __u32 rq_rep_swab_mask;
612 /** What was import generation when this request was sent */
613 int rq_import_generation;
614 enum lustre_imp_state rq_send_state;
616 /** how many early replies (for stats) */
619 /** client+server request */
620 lnet_handle_md_t rq_req_md_h;
621 struct ptlrpc_cb_id rq_req_cbid;
622 /** optional time limit for send attempts */
623 cfs_duration_t rq_delay_limit;
624 /** time request was first queued */
625 cfs_time_t rq_queued_time;
628 /** request arrival time */
629 struct timeval rq_arrival_time;
630 /** separated reply state */
631 struct ptlrpc_reply_state *rq_reply_state;
632 /** incoming request buffer */
633 struct ptlrpc_request_buffer_desc *rq_rqbd;
635 __u32 rq_uid; /* peer uid, used in MDS only */
638 /** client-only incoming reply */
639 lnet_handle_md_t rq_reply_md_h;
640 cfs_waitq_t rq_reply_waitq;
641 struct ptlrpc_cb_id rq_reply_cbid;
645 /** Peer description (the other side) */
646 lnet_process_id_t rq_peer;
647 /** Server-side, export on which request was received */
648 struct obd_export *rq_export;
649 /** Client side, import where request is being sent */
650 struct obd_import *rq_import;
652 /** Replay callback, called after request is replayed at recovery */
653 void (*rq_replay_cb)(struct ptlrpc_request *);
655 * Commit callback, called when request is committed and about to be
658 void (*rq_commit_cb)(struct ptlrpc_request *);
659 /** Opaq data for replay and commit callbacks. */
662 /** For bulk requests on client only: bulk descriptor */
663 struct ptlrpc_bulk_desc *rq_bulk;
665 /** client outgoing req */
667 * when request/reply sent (secs), or time when request should be sent
670 /** time for request really sent out */
673 /** when request must finish. volatile
674 * so that servers' early reply updates to the deadline aren't
675 * kept in per-cpu cache */
676 volatile time_t rq_deadline;
677 /** when req reply unlink must finish. */
678 time_t rq_reply_deadline;
679 /** when req bulk unlink must finish. */
680 time_t rq_bulk_deadline;
682 * service time estimate (secs)
683 * If the requestsis not served by this time, it is marked as timed out.
687 /** Multi-rpc bits */
688 /** Link item for request set lists */
689 cfs_list_t rq_set_chain;
690 /** Per-request waitq introduced by bug 21938 for recovery waiting */
691 cfs_waitq_t rq_set_waitq;
692 /** Link back to the request set */
693 struct ptlrpc_request_set *rq_set;
694 /** Async completion handler, called when reply is received */
695 ptlrpc_interpterer_t rq_interpret_reply;
696 /** Async completion context */
697 union ptlrpc_async_args rq_async_args;
699 /** Pool if request is from preallocated list */
700 struct ptlrpc_request_pool *rq_pool;
702 struct lu_context rq_session;
703 struct lu_context rq_recov_session;
705 /** request format description */
706 struct req_capsule rq_pill;
710 * Call completion handler for rpc if any, return it's status or original
711 * rc if there was no handler defined for this request.
713 static inline int ptlrpc_req_interpret(const struct lu_env *env,
714 struct ptlrpc_request *req, int rc)
716 if (req->rq_interpret_reply != NULL) {
717 req->rq_status = req->rq_interpret_reply(env, req,
720 return req->rq_status;
726 * Returns 1 if request buffer at offset \a index was already swabbed
728 static inline int lustre_req_swabbed(struct ptlrpc_request *req, int index)
730 LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
731 return req->rq_req_swab_mask & (1 << index);
735 * Returns 1 if request reply buffer at offset \a index was already swabbed
737 static inline int lustre_rep_swabbed(struct ptlrpc_request *req, int index)
739 LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
740 return req->rq_rep_swab_mask & (1 << index);
744 * Returns 1 if request needs to be swabbed into local cpu byteorder
746 static inline int ptlrpc_req_need_swab(struct ptlrpc_request *req)
748 return lustre_req_swabbed(req, MSG_PTLRPC_HEADER_OFF);
752 * Returns 1 if request reply needs to be swabbed into local cpu byteorder
754 static inline int ptlrpc_rep_need_swab(struct ptlrpc_request *req)
756 return lustre_rep_swabbed(req, MSG_PTLRPC_HEADER_OFF);
760 * Mark request buffer at offset \a index that it was already swabbed
762 static inline void lustre_set_req_swabbed(struct ptlrpc_request *req, int index)
764 LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
765 LASSERT((req->rq_req_swab_mask & (1 << index)) == 0);
766 req->rq_req_swab_mask |= 1 << index;
770 * Mark request reply buffer at offset \a index that it was already swabbed
772 static inline void lustre_set_rep_swabbed(struct ptlrpc_request *req, int index)
774 LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
775 LASSERT((req->rq_rep_swab_mask & (1 << index)) == 0);
776 req->rq_rep_swab_mask |= 1 << index;
780 * Convert numerical request phase value \a phase into text string description
782 static inline const char *
783 ptlrpc_phase2str(enum rq_phase phase)
792 case RQ_PHASE_INTERPRET:
794 case RQ_PHASE_COMPLETE:
796 case RQ_PHASE_UNREGISTERING:
797 return "Unregistering";
804 * Convert numerical request phase of the request \a req into text stringi
807 static inline const char *
808 ptlrpc_rqphase2str(struct ptlrpc_request *req)
810 return ptlrpc_phase2str(req->rq_phase);
814 * Debugging functions and helpers to print request structure into debug log
817 /* Spare the preprocessor, spoil the bugs. */
818 #define FLAG(field, str) (field ? str : "")
820 /** Convert bit flags into a string */
821 #define DEBUG_REQ_FLAGS(req) \
822 ptlrpc_rqphase2str(req), \
823 FLAG(req->rq_intr, "I"), FLAG(req->rq_replied, "R"), \
824 FLAG(req->rq_err, "E"), \
825 FLAG(req->rq_timedout, "X") /* eXpired */, FLAG(req->rq_resend, "S"), \
826 FLAG(req->rq_restart, "T"), FLAG(req->rq_replay, "P"), \
827 FLAG(req->rq_no_resend, "N"), \
828 FLAG(req->rq_waiting, "W"), \
829 FLAG(req->rq_wait_ctx, "C"), FLAG(req->rq_hp, "H"), \
830 FLAG(req->rq_committed, "M")
832 #define REQ_FLAGS_FMT "%s:%s%s%s%s%s%s%s%s%s%s%s%s"
834 void _debug_req(struct ptlrpc_request *req, __u32 mask,
835 struct libcfs_debug_msg_data *data, const char *fmt, ...)
836 __attribute__ ((format (printf, 4, 5)));
839 * Helper that decides if we need to print request accordig to current debug
842 #define debug_req(cdls, level, req, file, func, line, fmt, a...) \
846 if (((level) & D_CANTMASK) != 0 || \
847 ((libcfs_debug & (level)) != 0 && \
848 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) { \
849 static struct libcfs_debug_msg_data _req_dbg_data = \
850 DEBUG_MSG_DATA_INIT(cdls, DEBUG_SUBSYSTEM, file, func, line); \
851 _debug_req((req), (level), &_req_dbg_data, fmt, ##a); \
856 * This is the debug print function you need to use to print request sturucture
857 * content into lustre debug log.
858 * for most callers (level is a constant) this is resolved at compile time */
859 #define DEBUG_REQ(level, req, fmt, args...) \
861 if ((level) & (D_ERROR | D_WARNING)) { \
862 static cfs_debug_limit_state_t cdls; \
863 debug_req(&cdls, level, req, __FILE__, __func__, __LINE__, \
864 "@@@ "fmt" ", ## args); \
866 debug_req(NULL, level, req, __FILE__, __func__, __LINE__, \
867 "@@@ "fmt" ", ## args); \
872 * Structure that defines a single page of a bulk transfer
874 struct ptlrpc_bulk_page {
875 /** Linkage to list of pages in a bulk */
878 * Number of bytes in a page to transfer starting from \a bp_pageoffset
881 /** offset within a page */
883 /** The page itself */
884 struct page *bp_page;
887 #define BULK_GET_SOURCE 0
888 #define BULK_PUT_SINK 1
889 #define BULK_GET_SINK 2
890 #define BULK_PUT_SOURCE 3
893 * Definition of buk descriptor.
894 * Bulks are special "Two phase" RPCs where initial request message
895 * is sent first and it is followed bt a transfer (o receiving) of a large
896 * amount of data to be settled into pages referenced from the bulk descriptors.
897 * Bulks transfers (the actual data following the small requests) are done
898 * on separate LNet portals.
899 * In lustre we use bulk transfers for READ and WRITE transfers from/to OSTs.
900 * Another user is readpage for MDT.
902 struct ptlrpc_bulk_desc {
903 /** completed successfully */
904 unsigned long bd_success:1;
905 /** accessible to the network (network io potentially in progress) */
906 unsigned long bd_network_rw:1;
907 /** {put,get}{source,sink} */
908 unsigned long bd_type:2;
910 unsigned long bd_registered:1;
911 /** For serialization with callback */
912 cfs_spinlock_t bd_lock;
913 /** Import generation when request for this bulk was sent */
914 int bd_import_generation;
915 /** Server side - export this bulk created for */
916 struct obd_export *bd_export;
917 /** Client side - import this bulk was sent on */
918 struct obd_import *bd_import;
919 /** LNet portal for this bulk */
921 /** Back pointer to the request */
922 struct ptlrpc_request *bd_req;
923 cfs_waitq_t bd_waitq; /* server side only WQ */
924 int bd_iov_count; /* # entries in bd_iov */
925 int bd_max_iov; /* allocated size of bd_iov */
926 int bd_nob; /* # bytes covered */
927 int bd_nob_transferred; /* # bytes GOT/PUT */
931 struct ptlrpc_cb_id bd_cbid; /* network callback info */
932 lnet_handle_md_t bd_md_h; /* associated MD */
933 lnet_nid_t bd_sender; /* stash event::sender */
935 #if defined(__KERNEL__)
937 * encrypt iov, size is either 0 or bd_iov_count.
939 lnet_kiov_t *bd_enc_iov;
941 lnet_kiov_t bd_iov[0];
943 lnet_md_iovec_t bd_iov[0];
948 SVC_STOPPED = 1 << 0,
949 SVC_STOPPING = 1 << 1,
950 SVC_STARTING = 1 << 2,
951 SVC_RUNNING = 1 << 3,
957 * Definition of server service thread structure
959 struct ptlrpc_thread {
961 * List of active threads in svc->srv_threads
965 * thread-private data (preallocated memory)
970 * service thread index, from ptlrpc_start_threads
978 * put watchdog in the structure per thread b=14840
980 struct lc_watchdog *t_watchdog;
982 * the svc this thread belonged to b=18582
984 struct ptlrpc_service *t_svc;
985 cfs_waitq_t t_ctl_waitq;
986 struct lu_env *t_env;
989 static inline int thread_is_stopped(struct ptlrpc_thread *thread)
991 return !!(thread->t_flags & SVC_STOPPED);
994 static inline int thread_is_stopping(struct ptlrpc_thread *thread)
996 return !!(thread->t_flags & SVC_STOPPING);
999 static inline int thread_is_starting(struct ptlrpc_thread *thread)
1001 return !!(thread->t_flags & SVC_STARTING);
1004 static inline int thread_is_running(struct ptlrpc_thread *thread)
1006 return !!(thread->t_flags & SVC_RUNNING);
1009 static inline int thread_is_event(struct ptlrpc_thread *thread)
1011 return !!(thread->t_flags & SVC_EVENT);
1014 static inline int thread_is_signal(struct ptlrpc_thread *thread)
1016 return !!(thread->t_flags & SVC_SIGNAL);
1019 static inline void thread_clear_flags(struct ptlrpc_thread *thread, __u32 flags)
1021 thread->t_flags &= ~flags;
1024 static inline void thread_set_flags(struct ptlrpc_thread *thread, __u32 flags)
1026 thread->t_flags = flags;
1029 static inline void thread_add_flags(struct ptlrpc_thread *thread, __u32 flags)
1031 thread->t_flags |= flags;
1034 static inline int thread_test_and_clear_flags(struct ptlrpc_thread *thread,
1037 if (thread->t_flags & flags) {
1038 thread->t_flags &= ~flags;
1045 * Request buffer descriptor structure.
1046 * This is a structure that contains one posted request buffer for service.
1047 * Once data land into a buffer, event callback creates actual request and
1048 * notifies wakes one of the service threads to process new incoming request.
1049 * More than one request can fit into the buffer.
1051 struct ptlrpc_request_buffer_desc {
1052 /** Link item for rqbds on a service */
1053 cfs_list_t rqbd_list;
1054 /** History of requests for this buffer */
1055 cfs_list_t rqbd_reqs;
1056 /** Back pointer to service for which this buffer is registered */
1057 struct ptlrpc_service *rqbd_service;
1058 /** LNet descriptor */
1059 lnet_handle_md_t rqbd_md_h;
1061 /** The buffer itself */
1063 struct ptlrpc_cb_id rqbd_cbid;
1065 * This "embedded" request structure is only used for the
1066 * last request to fit into the buffer
1068 struct ptlrpc_request rqbd_req;
1071 typedef int (*svc_thr_init_t)(struct ptlrpc_thread *thread);
1072 typedef void (*svc_thr_done_t)(struct ptlrpc_thread *thread);
1073 typedef int (*svc_handler_t)(struct ptlrpc_request *req);
1074 typedef int (*svc_hpreq_handler_t)(struct ptlrpc_request *);
1075 typedef void (*svc_req_printfn_t)(void *, struct ptlrpc_request *);
1077 #ifndef __cfs_cacheline_aligned
1078 /* NB: put it here for reducing patche dependence */
1079 # define __cfs_cacheline_aligned
1083 * How many high priority requests to serve before serving one normal
1086 #define PTLRPC_SVC_HP_RATIO 10
1089 * Definition of PortalRPC service.
1090 * The service is listening on a particular portal (like tcp port)
1091 * and perform actions for a specific server like IO service for OST
1092 * or general metadata service for MDS.
1094 * ptlrpc service has four locks:
1096 * serialize operations on rqbd and requests waiting for preprocess
1098 * serialize operations active requests sent to this portal
1100 * serialize adaptive timeout stuff
1102 * serialize operations on RS list (reply states)
1104 * We don't have any use-case to take two or more locks at the same time
1105 * for now, so there is no lock order issue.
1107 struct ptlrpc_service {
1108 /** most often accessed fields */
1109 /** chain thru all services */
1110 cfs_list_t srv_list;
1111 /** only statically allocated strings here; we don't clean them */
1113 /** only statically allocated strings here; we don't clean them */
1114 char *srv_thread_name;
1115 /** service thread list */
1116 cfs_list_t srv_threads;
1117 /** threads to start at beginning of service */
1118 int srv_threads_min;
1119 /** thread upper limit */
1120 int srv_threads_max;
1121 /** always increasing number */
1122 unsigned srv_threads_next_id;
1123 /** # of starting threads */
1124 int srv_threads_starting;
1125 /** # running threads */
1126 int srv_threads_running;
1128 /** service operations, move to ptlrpc_svc_ops_t in the future */
1131 * if non-NULL called during thread creation (ptlrpc_start_thread())
1132 * to initialize service specific per-thread state.
1134 svc_thr_init_t srv_init;
1136 * if non-NULL called during thread shutdown (ptlrpc_main()) to
1137 * destruct state created by ->srv_init().
1139 svc_thr_done_t srv_done;
1140 /** Handler function for incoming requests for this service */
1141 svc_handler_t srv_handler;
1142 /** hp request handler */
1143 svc_hpreq_handler_t srv_hpreq_handler;
1144 /** service-specific print fn */
1145 svc_req_printfn_t srv_req_printfn;
1148 /** Root of /proc dir tree for this service */
1149 cfs_proc_dir_entry_t *srv_procroot;
1150 /** Pointer to statistic data for this service */
1151 struct lprocfs_stats *srv_stats;
1152 /** # hp per lp reqs to handle */
1153 int srv_hpreq_ratio;
1154 /** biggest request to receive */
1155 int srv_max_req_size;
1156 /** biggest reply to send */
1157 int srv_max_reply_size;
1158 /** size of individual buffers */
1160 /** # buffers to allocate in 1 group */
1161 int srv_nbuf_per_group;
1162 /** Local portal on which to receive requests */
1163 __u32 srv_req_portal;
1164 /** Portal on the client to send replies to */
1165 __u32 srv_rep_portal;
1167 * Tags for lu_context associated with this thread, see struct
1171 /** soft watchdog timeout multiplier */
1172 int srv_watchdog_factor;
1173 /** bind threads to CPUs */
1174 unsigned srv_cpu_affinity:1;
1175 /** under unregister_service */
1176 unsigned srv_is_stopping:1;
1179 * serialize the following fields, used for protecting
1180 * rqbd list and incoming requests waiting for preprocess
1182 cfs_spinlock_t srv_lock __cfs_cacheline_aligned;
1183 /** incoming reqs */
1184 cfs_list_t srv_req_in_queue;
1185 /** total # req buffer descs allocated */
1187 /** # posted request buffers */
1188 int srv_nrqbd_receiving;
1189 /** timeout before re-posting reqs, in tick */
1190 cfs_duration_t srv_rqbd_timeout;
1191 /** request buffers to be reposted */
1192 cfs_list_t srv_idle_rqbds;
1193 /** req buffers receiving */
1194 cfs_list_t srv_active_rqbds;
1195 /** request buffer history */
1196 cfs_list_t srv_history_rqbds;
1197 /** # request buffers in history */
1198 int srv_n_history_rqbds;
1199 /** max # request buffers in history */
1200 int srv_max_history_rqbds;
1201 /** request history */
1202 cfs_list_t srv_request_history;
1203 /** next request sequence # */
1204 __u64 srv_request_seq;
1205 /** highest seq culled from history */
1206 __u64 srv_request_max_cull_seq;
1208 * all threads sleep on this. This wait-queue is signalled when new
1209 * incoming request arrives and when difficult reply has to be handled.
1211 cfs_waitq_t srv_waitq;
1214 * serialize the following fields, used for processing requests
1215 * sent to this portal
1217 cfs_spinlock_t srv_rq_lock __cfs_cacheline_aligned;
1218 /** # reqs in either of the queues below */
1219 /** reqs waiting for service */
1220 cfs_list_t srv_request_queue;
1221 /** high priority queue */
1222 cfs_list_t srv_request_hpq;
1223 /** # incoming reqs */
1224 int srv_n_queued_reqs;
1225 /** # reqs being served */
1226 int srv_n_active_reqs;
1227 /** # HPreqs being served */
1228 int srv_n_active_hpreq;
1229 /** # hp requests handled */
1230 int srv_hpreq_count;
1235 * serialize the following fields, used for changes on
1238 cfs_spinlock_t srv_at_lock __cfs_cacheline_aligned;
1239 /** estimated rpc service time */
1240 struct adaptive_timeout srv_at_estimate;
1241 /** reqs waiting for replies */
1242 struct ptlrpc_at_array srv_at_array;
1243 /** early reply timer */
1244 cfs_timer_t srv_at_timer;
1245 /** check early replies */
1246 unsigned srv_at_check;
1248 cfs_time_t srv_at_checktime;
1252 * serialize the following fields, used for processing
1253 * replies for this portal
1255 cfs_spinlock_t srv_rs_lock __cfs_cacheline_aligned;
1256 /** all the active replies */
1257 cfs_list_t srv_active_replies;
1259 /** replies waiting for service */
1260 cfs_list_t srv_reply_queue;
1262 /** List of free reply_states */
1263 cfs_list_t srv_free_rs_list;
1264 /** waitq to run, when adding stuff to srv_free_rs_list */
1265 cfs_waitq_t srv_free_rs_waitq;
1266 /** # 'difficult' replies */
1267 cfs_atomic_t srv_n_difficult_replies;
1268 //struct ptlrpc_srv_ni srv_interfaces[0];
1272 * Declaration of ptlrpcd control structure
1274 struct ptlrpcd_ctl {
1276 * Ptlrpc thread control flags (LIOD_START, LIOD_STOP, LIOD_FORCE)
1278 unsigned long pc_flags;
1280 * Thread lock protecting structure fields.
1282 cfs_spinlock_t pc_lock;
1286 cfs_completion_t pc_starting;
1290 cfs_completion_t pc_finishing;
1292 * Thread requests set.
1294 struct ptlrpc_request_set *pc_set;
1296 * Thread name used in cfs_daemonize()
1300 * Environment for request interpreters to run in.
1302 struct lu_env pc_env;
1304 * Index of ptlrpcd thread in the array.
1308 * Number of the ptlrpcd's partners.
1312 * Pointer to the array of partners' ptlrpcd_ctl structure.
1314 struct ptlrpcd_ctl **pc_partners;
1316 * Record the partner index to be processed next.
1321 * Async rpcs flag to make sure that ptlrpcd_check() is called only
1326 * Currently not used.
1330 * User-space async rpcs callback.
1332 void *pc_wait_callback;
1334 * User-space check idle rpcs callback.
1336 void *pc_idle_callback;
1340 /* Bits for pc_flags */
1341 enum ptlrpcd_ctl_flags {
1343 * Ptlrpc thread start flag.
1345 LIOD_START = 1 << 0,
1347 * Ptlrpc thread stop flag.
1351 * Ptlrpc thread force flag (only stop force so far).
1352 * This will cause aborting any inflight rpcs handled
1353 * by thread if LIOD_STOP is specified.
1355 LIOD_FORCE = 1 << 2,
1357 * This is a recovery ptlrpc thread.
1359 LIOD_RECOVERY = 1 << 3,
1361 * The ptlrpcd is bound to some CPU core.
1366 /* ptlrpc/events.c */
1367 extern lnet_handle_eq_t ptlrpc_eq_h;
1368 extern int ptlrpc_uuid_to_peer(struct obd_uuid *uuid,
1369 lnet_process_id_t *peer, lnet_nid_t *self);
1371 * These callbacks are invoked by LNet when something happened to
1375 extern void request_out_callback (lnet_event_t *ev);
1376 extern void reply_in_callback(lnet_event_t *ev);
1377 extern void client_bulk_callback (lnet_event_t *ev);
1378 extern void request_in_callback(lnet_event_t *ev);
1379 extern void reply_out_callback(lnet_event_t *ev);
1380 extern void server_bulk_callback (lnet_event_t *ev);
1383 /* ptlrpc/connection.c */
1384 struct ptlrpc_connection *ptlrpc_connection_get(lnet_process_id_t peer,
1386 struct obd_uuid *uuid);
1387 int ptlrpc_connection_put(struct ptlrpc_connection *c);
1388 struct ptlrpc_connection *ptlrpc_connection_addref(struct ptlrpc_connection *);
1389 int ptlrpc_connection_init(void);
1390 void ptlrpc_connection_fini(void);
1391 extern lnet_pid_t ptl_get_pid(void);
1393 /* ptlrpc/niobuf.c */
1395 * Actual interfacing with LNet to put/get/register/unregister stuff
1398 int ptlrpc_start_bulk_transfer(struct ptlrpc_bulk_desc *desc);
1399 void ptlrpc_abort_bulk(struct ptlrpc_bulk_desc *desc);
1400 int ptlrpc_register_bulk(struct ptlrpc_request *req);
1401 int ptlrpc_unregister_bulk(struct ptlrpc_request *req, int async);
1403 static inline int ptlrpc_server_bulk_active(struct ptlrpc_bulk_desc *desc)
1407 LASSERT(desc != NULL);
1409 cfs_spin_lock(&desc->bd_lock);
1410 rc = desc->bd_network_rw;
1411 cfs_spin_unlock(&desc->bd_lock);
1415 static inline int ptlrpc_client_bulk_active(struct ptlrpc_request *req)
1417 struct ptlrpc_bulk_desc *desc = req->rq_bulk;
1420 LASSERT(req != NULL);
1422 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_BULK_UNLINK) &&
1423 req->rq_bulk_deadline > cfs_time_current_sec())
1429 cfs_spin_lock(&desc->bd_lock);
1430 rc = desc->bd_network_rw;
1431 cfs_spin_unlock(&desc->bd_lock);
1435 #define PTLRPC_REPLY_MAYBE_DIFFICULT 0x01
1436 #define PTLRPC_REPLY_EARLY 0x02
1437 int ptlrpc_send_reply(struct ptlrpc_request *req, int flags);
1438 int ptlrpc_reply(struct ptlrpc_request *req);
1439 int ptlrpc_send_error(struct ptlrpc_request *req, int difficult);
1440 int ptlrpc_error(struct ptlrpc_request *req);
1441 void ptlrpc_resend_req(struct ptlrpc_request *request);
1442 int ptlrpc_at_get_net_latency(struct ptlrpc_request *req);
1443 int ptl_send_rpc(struct ptlrpc_request *request, int noreply);
1444 int ptlrpc_register_rqbd (struct ptlrpc_request_buffer_desc *rqbd);
1447 /* ptlrpc/client.c */
1449 * Client-side portals API. Everything to send requests, receive replies,
1450 * request queues, request management, etc.
1453 void ptlrpc_init_client(int req_portal, int rep_portal, char *name,
1454 struct ptlrpc_client *);
1455 void ptlrpc_cleanup_client(struct obd_import *imp);
1456 struct ptlrpc_connection *ptlrpc_uuid_to_connection(struct obd_uuid *uuid);
1458 int ptlrpc_queue_wait(struct ptlrpc_request *req);
1459 int ptlrpc_replay_req(struct ptlrpc_request *req);
1460 int ptlrpc_unregister_reply(struct ptlrpc_request *req, int async);
1461 void ptlrpc_restart_req(struct ptlrpc_request *req);
1462 void ptlrpc_abort_inflight(struct obd_import *imp);
1463 void ptlrpc_cleanup_imp(struct obd_import *imp);
1464 void ptlrpc_abort_set(struct ptlrpc_request_set *set);
1466 struct ptlrpc_request_set *ptlrpc_prep_set(void);
1467 int ptlrpc_set_add_cb(struct ptlrpc_request_set *set,
1468 set_interpreter_func fn, void *data);
1469 int ptlrpc_set_next_timeout(struct ptlrpc_request_set *);
1470 int ptlrpc_check_set(const struct lu_env *env, struct ptlrpc_request_set *set);
1471 int ptlrpc_set_wait(struct ptlrpc_request_set *);
1472 int ptlrpc_expired_set(void *data);
1473 void ptlrpc_interrupted_set(void *data);
1474 void ptlrpc_mark_interrupted(struct ptlrpc_request *req);
1475 void ptlrpc_set_destroy(struct ptlrpc_request_set *);
1476 void ptlrpc_set_add_req(struct ptlrpc_request_set *, struct ptlrpc_request *);
1477 void ptlrpc_set_add_new_req(struct ptlrpcd_ctl *pc,
1478 struct ptlrpc_request *req);
1480 void ptlrpc_free_rq_pool(struct ptlrpc_request_pool *pool);
1481 void ptlrpc_add_rqs_to_pool(struct ptlrpc_request_pool *pool, int num_rq);
1483 struct ptlrpc_request_pool *
1484 ptlrpc_init_rq_pool(int, int,
1485 void (*populate_pool)(struct ptlrpc_request_pool *, int));
1487 void ptlrpc_at_set_req_timeout(struct ptlrpc_request *req);
1488 struct ptlrpc_request *ptlrpc_request_alloc(struct obd_import *imp,
1489 const struct req_format *format);
1490 struct ptlrpc_request *ptlrpc_request_alloc_pool(struct obd_import *imp,
1491 struct ptlrpc_request_pool *,
1492 const struct req_format *format);
1493 void ptlrpc_request_free(struct ptlrpc_request *request);
1494 int ptlrpc_request_pack(struct ptlrpc_request *request,
1495 __u32 version, int opcode);
1496 struct ptlrpc_request *ptlrpc_request_alloc_pack(struct obd_import *imp,
1497 const struct req_format *format,
1498 __u32 version, int opcode);
1499 int ptlrpc_request_bufs_pack(struct ptlrpc_request *request,
1500 __u32 version, int opcode, char **bufs,
1501 struct ptlrpc_cli_ctx *ctx);
1502 struct ptlrpc_request *ptlrpc_prep_fakereq(struct obd_import *imp,
1503 unsigned int timeout,
1504 ptlrpc_interpterer_t interpreter);
1505 void ptlrpc_fakereq_finished(struct ptlrpc_request *req);
1507 struct ptlrpc_request *ptlrpc_prep_req(struct obd_import *imp, __u32 version,
1508 int opcode, int count, __u32 *lengths,
1510 struct ptlrpc_request *ptlrpc_prep_req_pool(struct obd_import *imp,
1511 __u32 version, int opcode,
1512 int count, __u32 *lengths, char **bufs,
1513 struct ptlrpc_request_pool *pool);
1514 void ptlrpc_req_finished(struct ptlrpc_request *request);
1515 void ptlrpc_req_finished_with_imp_lock(struct ptlrpc_request *request);
1516 struct ptlrpc_request *ptlrpc_request_addref(struct ptlrpc_request *req);
1517 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_imp (struct ptlrpc_request *req,
1518 int npages, int type, int portal);
1519 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_exp(struct ptlrpc_request *req,
1520 int npages, int type, int portal);
1521 void ptlrpc_free_bulk(struct ptlrpc_bulk_desc *bulk);
1522 void ptlrpc_prep_bulk_page(struct ptlrpc_bulk_desc *desc,
1523 cfs_page_t *page, int pageoffset, int len);
1524 void ptlrpc_retain_replayable_request(struct ptlrpc_request *req,
1525 struct obd_import *imp);
1526 __u64 ptlrpc_next_xid(void);
1527 __u64 ptlrpc_sample_next_xid(void);
1528 __u64 ptlrpc_req_xid(struct ptlrpc_request *request);
1532 struct ptlrpc_service_conf {
1535 int psc_max_req_size;
1536 int psc_max_reply_size;
1539 int psc_watchdog_factor;
1540 int psc_min_threads;
1541 int psc_max_threads;
1545 /* ptlrpc/service.c */
1547 * Server-side services API. Register/unregister service, request state
1548 * management, service thread management
1552 void ptlrpc_save_lock (struct ptlrpc_request *req,
1553 struct lustre_handle *lock, int mode, int no_ack);
1554 void ptlrpc_commit_replies(struct obd_export *exp);
1555 void ptlrpc_dispatch_difficult_reply (struct ptlrpc_reply_state *rs);
1556 void ptlrpc_schedule_difficult_reply (struct ptlrpc_reply_state *rs);
1557 struct ptlrpc_service *ptlrpc_init_svc_conf(struct ptlrpc_service_conf *c,
1558 svc_handler_t h, char *name,
1559 struct proc_dir_entry *proc_entry,
1560 svc_req_printfn_t prntfn,
1563 struct ptlrpc_service *ptlrpc_init_svc(int nbufs, int bufsize, int max_req_size,
1565 int req_portal, int rep_portal,
1566 int watchdog_factor,
1567 svc_handler_t, char *name,
1568 cfs_proc_dir_entry_t *proc_entry,
1570 int min_threads, int max_threads,
1571 char *threadname, __u32 ctx_tags,
1572 svc_hpreq_handler_t);
1573 void ptlrpc_stop_all_threads(struct ptlrpc_service *svc);
1575 int ptlrpc_start_threads(struct ptlrpc_service *svc);
1576 int ptlrpc_start_thread(struct ptlrpc_service *svc);
1577 int ptlrpc_unregister_service(struct ptlrpc_service *service);
1578 int liblustre_check_services (void *arg);
1579 void ptlrpc_daemonize(char *name);
1580 int ptlrpc_service_health_check(struct ptlrpc_service *);
1581 void ptlrpc_hpreq_reorder(struct ptlrpc_request *req);
1582 void ptlrpc_server_drop_request(struct ptlrpc_request *req);
1585 int ptlrpc_hr_init(void);
1586 void ptlrpc_hr_fini(void);
1588 # define ptlrpc_hr_init() (0)
1589 # define ptlrpc_hr_fini() do {} while(0)
1592 struct ptlrpc_svc_data {
1594 struct ptlrpc_service *svc;
1595 struct ptlrpc_thread *thread;
1599 /* ptlrpc/import.c */
1604 int ptlrpc_connect_import(struct obd_import *imp);
1605 int ptlrpc_init_import(struct obd_import *imp);
1606 int ptlrpc_disconnect_import(struct obd_import *imp, int noclose);
1607 int ptlrpc_import_recovery_state_machine(struct obd_import *imp);
1609 /* ptlrpc/pack_generic.c */
1610 int ptlrpc_reconnect_import(struct obd_import *imp);
1614 * ptlrpc msg buffer and swab interface
1618 int ptlrpc_buf_need_swab(struct ptlrpc_request *req, const int inout,
1620 void ptlrpc_buf_set_swabbed(struct ptlrpc_request *req, const int inout,
1622 int ptlrpc_unpack_rep_msg(struct ptlrpc_request *req, int len);
1623 int ptlrpc_unpack_req_msg(struct ptlrpc_request *req, int len);
1625 int lustre_msg_check_version(struct lustre_msg *msg, __u32 version);
1626 void lustre_init_msg_v2(struct lustre_msg_v2 *msg, int count, __u32 *lens,
1628 int lustre_pack_request(struct ptlrpc_request *, __u32 magic, int count,
1629 __u32 *lens, char **bufs);
1630 int lustre_pack_reply(struct ptlrpc_request *, int count, __u32 *lens,
1632 int lustre_pack_reply_v2(struct ptlrpc_request *req, int count,
1633 __u32 *lens, char **bufs, int flags);
1634 #define LPRFL_EARLY_REPLY 1
1635 int lustre_pack_reply_flags(struct ptlrpc_request *, int count, __u32 *lens,
1636 char **bufs, int flags);
1637 int lustre_shrink_msg(struct lustre_msg *msg, int segment,
1638 unsigned int newlen, int move_data);
1639 void lustre_free_reply_state(struct ptlrpc_reply_state *rs);
1640 int __lustre_unpack_msg(struct lustre_msg *m, int len);
1641 int lustre_msg_hdr_size(__u32 magic, int count);
1642 int lustre_msg_size(__u32 magic, int count, __u32 *lengths);
1643 int lustre_msg_size_v2(int count, __u32 *lengths);
1644 int lustre_packed_msg_size(struct lustre_msg *msg);
1645 int lustre_msg_early_size(void);
1646 void *lustre_msg_buf_v2(struct lustre_msg_v2 *m, int n, int min_size);
1647 void *lustre_msg_buf(struct lustre_msg *m, int n, int minlen);
1648 int lustre_msg_buflen(struct lustre_msg *m, int n);
1649 void lustre_msg_set_buflen(struct lustre_msg *m, int n, int len);
1650 int lustre_msg_bufcount(struct lustre_msg *m);
1651 char *lustre_msg_string (struct lustre_msg *m, int n, int max_len);
1652 __u32 lustre_msghdr_get_flags(struct lustre_msg *msg);
1653 void lustre_msghdr_set_flags(struct lustre_msg *msg, __u32 flags);
1654 __u32 lustre_msg_get_flags(struct lustre_msg *msg);
1655 void lustre_msg_add_flags(struct lustre_msg *msg, int flags);
1656 void lustre_msg_set_flags(struct lustre_msg *msg, int flags);
1657 void lustre_msg_clear_flags(struct lustre_msg *msg, int flags);
1658 __u32 lustre_msg_get_op_flags(struct lustre_msg *msg);
1659 void lustre_msg_add_op_flags(struct lustre_msg *msg, int flags);
1660 void lustre_msg_set_op_flags(struct lustre_msg *msg, int flags);
1661 struct lustre_handle *lustre_msg_get_handle(struct lustre_msg *msg);
1662 __u32 lustre_msg_get_type(struct lustre_msg *msg);
1663 __u32 lustre_msg_get_version(struct lustre_msg *msg);
1664 void lustre_msg_add_version(struct lustre_msg *msg, int version);
1665 __u32 lustre_msg_get_opc(struct lustre_msg *msg);
1666 __u64 lustre_msg_get_last_xid(struct lustre_msg *msg);
1667 __u64 lustre_msg_get_last_committed(struct lustre_msg *msg);
1668 __u64 *lustre_msg_get_versions(struct lustre_msg *msg);
1669 __u64 lustre_msg_get_transno(struct lustre_msg *msg);
1670 __u64 lustre_msg_get_slv(struct lustre_msg *msg);
1671 __u32 lustre_msg_get_limit(struct lustre_msg *msg);
1672 void lustre_msg_set_slv(struct lustre_msg *msg, __u64 slv);
1673 void lustre_msg_set_limit(struct lustre_msg *msg, __u64 limit);
1674 int lustre_msg_get_status(struct lustre_msg *msg);
1675 __u32 lustre_msg_get_conn_cnt(struct lustre_msg *msg);
1676 int lustre_msg_is_v1(struct lustre_msg *msg);
1677 __u32 lustre_msg_get_magic(struct lustre_msg *msg);
1678 __u32 lustre_msg_get_timeout(struct lustre_msg *msg);
1679 __u32 lustre_msg_get_service_time(struct lustre_msg *msg);
1680 __u32 lustre_msg_get_cksum(struct lustre_msg *msg);
1681 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 0, 0)
1682 __u32 lustre_msg_calc_cksum(struct lustre_msg *msg, int compat18);
1684 # warning "remove checksum compatibility support for b1_8"
1685 __u32 lustre_msg_calc_cksum(struct lustre_msg *msg);
1687 void lustre_msg_set_handle(struct lustre_msg *msg,struct lustre_handle *handle);
1688 void lustre_msg_set_type(struct lustre_msg *msg, __u32 type);
1689 void lustre_msg_set_opc(struct lustre_msg *msg, __u32 opc);
1690 void lustre_msg_set_last_xid(struct lustre_msg *msg, __u64 last_xid);
1691 void lustre_msg_set_last_committed(struct lustre_msg *msg,__u64 last_committed);
1692 void lustre_msg_set_versions(struct lustre_msg *msg, __u64 *versions);
1693 void lustre_msg_set_transno(struct lustre_msg *msg, __u64 transno);
1694 void lustre_msg_set_status(struct lustre_msg *msg, __u32 status);
1695 void lustre_msg_set_conn_cnt(struct lustre_msg *msg, __u32 conn_cnt);
1696 void ptlrpc_req_set_repsize(struct ptlrpc_request *req, int count, __u32 *sizes);
1697 void ptlrpc_request_set_replen(struct ptlrpc_request *req);
1698 void lustre_msg_set_timeout(struct lustre_msg *msg, __u32 timeout);
1699 void lustre_msg_set_service_time(struct lustre_msg *msg, __u32 service_time);
1700 void lustre_msg_set_cksum(struct lustre_msg *msg, __u32 cksum);
1703 lustre_shrink_reply(struct ptlrpc_request *req, int segment,
1704 unsigned int newlen, int move_data)
1706 LASSERT(req->rq_reply_state);
1707 LASSERT(req->rq_repmsg);
1708 req->rq_replen = lustre_shrink_msg(req->rq_repmsg, segment,
1713 /** Change request phase of \a req to \a new_phase */
1715 ptlrpc_rqphase_move(struct ptlrpc_request *req, enum rq_phase new_phase)
1717 if (req->rq_phase == new_phase)
1720 if (new_phase == RQ_PHASE_UNREGISTERING) {
1721 req->rq_next_phase = req->rq_phase;
1723 cfs_atomic_inc(&req->rq_import->imp_unregistering);
1726 if (req->rq_phase == RQ_PHASE_UNREGISTERING) {
1728 cfs_atomic_dec(&req->rq_import->imp_unregistering);
1731 DEBUG_REQ(D_INFO, req, "move req \"%s\" -> \"%s\"",
1732 ptlrpc_rqphase2str(req), ptlrpc_phase2str(new_phase));
1734 req->rq_phase = new_phase;
1738 * Returns true if request \a req got early reply and hard deadline is not met
1741 ptlrpc_client_early(struct ptlrpc_request *req)
1743 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
1744 req->rq_reply_deadline > cfs_time_current_sec())
1746 return req->rq_early;
1750 * Returns true if we got real reply from server for this request
1753 ptlrpc_client_replied(struct ptlrpc_request *req)
1755 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
1756 req->rq_reply_deadline > cfs_time_current_sec())
1758 return req->rq_replied;
1761 /** Returns true if request \a req is in process of receiving server reply */
1763 ptlrpc_client_recv(struct ptlrpc_request *req)
1765 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
1766 req->rq_reply_deadline > cfs_time_current_sec())
1768 return req->rq_receiving_reply;
1772 ptlrpc_client_recv_or_unlink(struct ptlrpc_request *req)
1776 cfs_spin_lock(&req->rq_lock);
1777 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
1778 req->rq_reply_deadline > cfs_time_current_sec()) {
1779 cfs_spin_unlock(&req->rq_lock);
1782 rc = req->rq_receiving_reply || req->rq_must_unlink;
1783 cfs_spin_unlock(&req->rq_lock);
1788 ptlrpc_client_wake_req(struct ptlrpc_request *req)
1790 if (req->rq_set == NULL)
1791 cfs_waitq_signal(&req->rq_reply_waitq);
1793 cfs_waitq_signal(&req->rq_set->set_waitq);
1797 ptlrpc_rs_addref(struct ptlrpc_reply_state *rs)
1799 LASSERT(cfs_atomic_read(&rs->rs_refcount) > 0);
1800 cfs_atomic_inc(&rs->rs_refcount);
1804 ptlrpc_rs_decref(struct ptlrpc_reply_state *rs)
1806 LASSERT(cfs_atomic_read(&rs->rs_refcount) > 0);
1807 if (cfs_atomic_dec_and_test(&rs->rs_refcount))
1808 lustre_free_reply_state(rs);
1811 /* Should only be called once per req */
1812 static inline void ptlrpc_req_drop_rs(struct ptlrpc_request *req)
1814 if (req->rq_reply_state == NULL)
1815 return; /* shouldn't occur */
1816 ptlrpc_rs_decref(req->rq_reply_state);
1817 req->rq_reply_state = NULL;
1818 req->rq_repmsg = NULL;
1821 static inline __u32 lustre_request_magic(struct ptlrpc_request *req)
1823 return lustre_msg_get_magic(req->rq_reqmsg);
1826 static inline int ptlrpc_req_get_repsize(struct ptlrpc_request *req)
1828 switch (req->rq_reqmsg->lm_magic) {
1829 case LUSTRE_MSG_MAGIC_V2:
1830 return req->rq_reqmsg->lm_repsize;
1832 LASSERTF(0, "incorrect message magic: %08x\n",
1833 req->rq_reqmsg->lm_magic);
1838 static inline int ptlrpc_send_limit_expired(struct ptlrpc_request *req)
1840 if (req->rq_delay_limit != 0 &&
1841 cfs_time_before(cfs_time_add(req->rq_queued_time,
1842 cfs_time_seconds(req->rq_delay_limit)),
1843 cfs_time_current())) {
1849 static inline int ptlrpc_no_resend(struct ptlrpc_request *req)
1851 if (!req->rq_no_resend && ptlrpc_send_limit_expired(req)) {
1852 cfs_spin_lock(&req->rq_lock);
1853 req->rq_no_resend = 1;
1854 cfs_spin_unlock(&req->rq_lock);
1856 return req->rq_no_resend;
1859 /* ldlm/ldlm_lib.c */
1861 * Target client logic
1864 int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg);
1865 int client_obd_cleanup(struct obd_device *obddev);
1866 int client_connect_import(const struct lu_env *env,
1867 struct obd_export **exp, struct obd_device *obd,
1868 struct obd_uuid *cluuid, struct obd_connect_data *,
1870 int client_disconnect_export(struct obd_export *exp);
1871 int client_import_add_conn(struct obd_import *imp, struct obd_uuid *uuid,
1873 int client_import_del_conn(struct obd_import *imp, struct obd_uuid *uuid);
1874 int client_import_find_conn(struct obd_import *imp, lnet_nid_t peer,
1875 struct obd_uuid *uuid);
1876 int import_set_conn_priority(struct obd_import *imp, struct obd_uuid *uuid);
1877 void client_destroy_import(struct obd_import *imp);
1880 int server_disconnect_export(struct obd_export *exp);
1882 /* ptlrpc/pinger.c */
1884 * Pinger API (client side only)
1887 enum timeout_event {
1890 struct timeout_item;
1891 typedef int (*timeout_cb_t)(struct timeout_item *, void *);
1892 int ptlrpc_pinger_add_import(struct obd_import *imp);
1893 int ptlrpc_pinger_del_import(struct obd_import *imp);
1894 int ptlrpc_add_timeout_client(int time, enum timeout_event event,
1895 timeout_cb_t cb, void *data,
1896 cfs_list_t *obd_list);
1897 int ptlrpc_del_timeout_client(cfs_list_t *obd_list,
1898 enum timeout_event event);
1899 struct ptlrpc_request * ptlrpc_prep_ping(struct obd_import *imp);
1900 int ptlrpc_obd_ping(struct obd_device *obd);
1901 cfs_time_t ptlrpc_suspend_wakeup_time(void);
1903 void ping_evictor_start(void);
1904 void ping_evictor_stop(void);
1906 #define ping_evictor_start() do {} while (0)
1907 #define ping_evictor_stop() do {} while (0)
1909 int ptlrpc_check_and_wait_suspend(struct ptlrpc_request *req);
1912 /* ptlrpc daemon bind policy */
1914 /* all ptlrpcd threads are free mode */
1915 PDB_POLICY_NONE = 1,
1916 /* all ptlrpcd threads are bound mode */
1917 PDB_POLICY_FULL = 2,
1918 /* <free1 bound1> <free2 bound2> ... <freeN boundN> */
1919 PDB_POLICY_PAIR = 3,
1920 /* <free1 bound1> <bound1 free2> ... <freeN boundN> <boundN free1>,
1921 * means each ptlrpcd[X] has two partners: thread[X-1] and thread[X+1]*/
1922 PDB_POLICY_NEIGHBOR = 4,
1925 /* ptlrpc daemon load policy
1926 * It is caller's duty to specify how to push the async RPC into some ptlrpcd
1927 * queue, but it is not enforced, affected by "ptlrpcd_bind_policy". If it is
1928 * "PDB_POLICY_FULL", then the RPC will be processed by the selected ptlrpcd,
1929 * Otherwise, the RPC may be processed by the selected ptlrpcd or its partner,
1930 * depends on which is scheduled firstly, to accelerate the RPC processing. */
1932 /* on the same CPU core as the caller */
1933 PDL_POLICY_SAME = 1,
1934 /* within the same CPU partition, but not the same core as the caller */
1935 PDL_POLICY_LOCAL = 2,
1936 /* round-robin on all CPU cores, but not the same core as the caller */
1937 PDL_POLICY_ROUND = 3,
1938 /* the specified CPU core is preferred, but not enforced */
1939 PDL_POLICY_PREFERRED = 4,
1942 /* ptlrpc/ptlrpcd.c */
1943 void ptlrpcd_stop(struct ptlrpcd_ctl *pc, int force);
1944 void ptlrpcd_wake(struct ptlrpc_request *req);
1945 void ptlrpcd_add_req(struct ptlrpc_request *req, pdl_policy_t policy, int idx);
1946 void ptlrpcd_add_rqset(struct ptlrpc_request_set *set);
1947 int ptlrpcd_addref(void);
1948 void ptlrpcd_decref(void);
1950 /* ptlrpc/lproc_ptlrpc.c */
1952 * procfs output related functions
1955 const char* ll_opcode2str(__u32 opcode);
1957 void ptlrpc_lprocfs_register_obd(struct obd_device *obd);
1958 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd);
1959 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes);
1961 static inline void ptlrpc_lprocfs_register_obd(struct obd_device *obd) {}
1962 static inline void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd) {}
1963 static inline void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes) {}
1967 /* ptlrpc/llog_server.c */
1968 int llog_origin_handle_create(struct ptlrpc_request *req);
1969 int llog_origin_handle_destroy(struct ptlrpc_request *req);
1970 int llog_origin_handle_prev_block(struct ptlrpc_request *req);
1971 int llog_origin_handle_next_block(struct ptlrpc_request *req);
1972 int llog_origin_handle_read_header(struct ptlrpc_request *req);
1973 int llog_origin_handle_close(struct ptlrpc_request *req);
1974 int llog_origin_handle_cancel(struct ptlrpc_request *req);
1975 int llog_catinfo(struct ptlrpc_request *req);
1977 /* ptlrpc/llog_client.c */
1978 extern struct llog_operations llog_client_ops;