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.
32 * Copyright (c) 2011, 2012, Whamcloud, Inc.
35 * This file is part of Lustre, http://www.lustre.org/
36 * Lustre is a trademark of Sun Microsystems, Inc.
38 /** \defgroup PtlRPC Portal RPC and networking module.
40 * PortalRPC is the layer used by rest of lustre code to achieve network
41 * communications: establish connections with corresponding export and import
42 * states, listen for a service, send and receive RPCs.
43 * PortalRPC also includes base recovery framework: packet resending and
44 * replaying, reconnections, pinger.
46 * PortalRPC utilizes LNet as its transport layer.
60 #if defined(__linux__)
61 #include <linux/lustre_net.h>
62 #elif defined(__APPLE__)
63 #include <darwin/lustre_net.h>
64 #elif defined(__WINNT__)
65 #include <winnt/lustre_net.h>
67 #error Unsupported operating system.
70 #include <libcfs/libcfs.h>
72 #include <lnet/lnet.h>
73 #include <lustre/lustre_idl.h>
74 #include <lustre_ha.h>
75 #include <lustre_sec.h>
76 #include <lustre_import.h>
77 #include <lprocfs_status.h>
78 #include <lu_object.h>
79 #include <lustre_req_layout.h>
81 #include <obd_support.h>
82 #include <lustre_ver.h>
84 /* MD flags we _always_ use */
85 #define PTLRPC_MD_OPTIONS 0
88 * Define maxima for bulk I/O
89 * CAVEAT EMPTOR, with multinet (i.e. routers forwarding between networks)
90 * these limits are system wide and not interface-local. */
91 #define PTLRPC_MAX_BRW_BITS LNET_MTU_BITS
92 #define PTLRPC_MAX_BRW_SIZE (1<<LNET_MTU_BITS)
93 #define PTLRPC_MAX_BRW_PAGES (PTLRPC_MAX_BRW_SIZE >> CFS_PAGE_SHIFT)
95 /* When PAGE_SIZE is a constant, we can check our arithmetic here with cpp! */
97 # if ((PTLRPC_MAX_BRW_PAGES & (PTLRPC_MAX_BRW_PAGES - 1)) != 0)
98 # error "PTLRPC_MAX_BRW_PAGES isn't a power of two"
100 # if (PTLRPC_MAX_BRW_SIZE != (PTLRPC_MAX_BRW_PAGES * CFS_PAGE_SIZE))
101 # error "PTLRPC_MAX_BRW_SIZE isn't PTLRPC_MAX_BRW_PAGES * CFS_PAGE_SIZE"
103 # if (PTLRPC_MAX_BRW_SIZE > LNET_MTU)
104 # error "PTLRPC_MAX_BRW_SIZE too big"
106 # if (PTLRPC_MAX_BRW_PAGES > LNET_MAX_IOV)
107 # error "PTLRPC_MAX_BRW_PAGES too big"
109 #endif /* __KERNEL__ */
112 * The following constants determine how memory is used to buffer incoming
115 * ?_NBUFS # buffers to allocate when growing the pool
116 * ?_BUFSIZE # bytes in a single request buffer
117 * ?_MAXREQSIZE # maximum request service will receive
119 * When fewer than ?_NBUFS/2 buffers are posted for receive, another chunk
120 * of ?_NBUFS is added to the pool.
122 * Messages larger than ?_MAXREQSIZE are dropped. Request buffers are
123 * considered full when less than ?_MAXREQSIZE is left in them.
125 #define LDLM_THREADS_AUTO_MIN (2)
126 #define LDLM_THREADS_AUTO_MAX min_t(unsigned, cfs_num_online_cpus() * \
127 cfs_num_online_cpus() * 32, 128)
128 #define LDLM_BL_THREADS LDLM_THREADS_AUTO_MIN
129 #define LDLM_NBUFS (64 * cfs_num_online_cpus())
130 #define LDLM_BUFSIZE (8 * 1024)
131 #define LDLM_MAXREQSIZE (5 * 1024)
132 #define LDLM_MAXREPSIZE (1024)
134 /** Absolute limits */
135 #define MDT_MIN_THREADS 2UL
136 #ifndef MDT_MAX_THREADS
137 #define MDT_MAX_THREADS 512UL
139 #define MDS_NBUFS (64 * cfs_num_online_cpus())
141 * Assume file name length = FNAME_MAX = 256 (true for ext3).
142 * path name length = PATH_MAX = 4096
143 * LOV MD size max = EA_MAX = 48000 (2000 stripes)
144 * symlink: FNAME_MAX + PATH_MAX <- largest
145 * link: FNAME_MAX + PATH_MAX (mds_rec_link < mds_rec_create)
146 * rename: FNAME_MAX + FNAME_MAX
147 * open: FNAME_MAX + EA_MAX
149 * MDS_MAXREQSIZE ~= 4736 bytes =
150 * lustre_msg + ldlm_request + mdt_body + mds_rec_create + FNAME_MAX + PATH_MAX
151 * MDS_MAXREPSIZE ~= 8300 bytes = lustre_msg + llog_header
152 * or, for mds_close() and mds_reint_unlink() on a many-OST filesystem:
153 * = 9210 bytes = lustre_msg + mdt_body + 160 * (easize + cookiesize)
155 * Realistic size is about 512 bytes (20 character name + 128 char symlink),
156 * except in the open case where there are a large number of OSTs in a LOV.
158 #define MDS_MAXREPSIZE max(10 * 1024, 362 + LOV_MAX_STRIPE_COUNT * 56)
159 #define MDS_MAXREQSIZE MDS_MAXREPSIZE
161 /** MDS_BUFSIZE = max_reqsize + max sptlrpc payload size */
162 #define MDS_BUFSIZE (MDS_MAXREQSIZE + 1024)
164 /** FLD_MAXREQSIZE == lustre_msg + __u32 padding + ptlrpc_body + opc */
165 #define FLD_MAXREQSIZE (160)
167 /** FLD_MAXREPSIZE == lustre_msg + ptlrpc_body */
168 #define FLD_MAXREPSIZE (152)
171 * SEQ_MAXREQSIZE == lustre_msg + __u32 padding + ptlrpc_body + opc + lu_range +
173 #define SEQ_MAXREQSIZE (160)
175 /** SEQ_MAXREPSIZE == lustre_msg + ptlrpc_body + lu_range */
176 #define SEQ_MAXREPSIZE (152)
178 /** MGS threads must be >= 3, see bug 22458 comment #28 */
179 #define MGS_THREADS_AUTO_MIN 3
180 #define MGS_THREADS_AUTO_MAX 32
181 #define MGS_NBUFS (64 * cfs_num_online_cpus())
182 #define MGS_BUFSIZE (8 * 1024)
183 #define MGS_MAXREQSIZE (7 * 1024)
184 #define MGS_MAXREPSIZE (9 * 1024)
186 /** Absolute OSS limits */
187 #define OSS_THREADS_MIN 3 /* difficult replies, HPQ, others */
188 #define OSS_THREADS_MAX 512
189 #define OST_NBUFS (64 * cfs_num_online_cpus())
190 #define OST_BUFSIZE (8 * 1024)
193 * OST_MAXREQSIZE ~= 4768 bytes =
194 * lustre_msg + obdo + 16 * obd_ioobj + 256 * niobuf_remote
196 * - single object with 16 pages is 512 bytes
197 * - OST_MAXREQSIZE must be at least 1 page of cookies plus some spillover
199 #define OST_MAXREQSIZE (5 * 1024)
200 #define OST_MAXREPSIZE (9 * 1024)
202 /* Macro to hide a typecast. */
203 #define ptlrpc_req_async_args(req) ((void *)&req->rq_async_args)
206 * Structure to single define portal connection.
208 struct ptlrpc_connection {
209 /** linkage for connections hash table */
210 cfs_hlist_node_t c_hash;
211 /** Our own lnet nid for this connection */
213 /** Remote side nid for this connection */
214 lnet_process_id_t c_peer;
215 /** UUID of the other side */
216 struct obd_uuid c_remote_uuid;
217 /** reference counter for this connection */
218 cfs_atomic_t c_refcount;
221 /** Client definition for PortalRPC */
222 struct ptlrpc_client {
223 /** What lnet portal does this client send messages to by default */
224 __u32 cli_request_portal;
225 /** What portal do we expect replies on */
226 __u32 cli_reply_portal;
227 /** Name of the client */
231 /** state flags of requests */
232 /* XXX only ones left are those used by the bulk descs as well! */
233 #define PTL_RPC_FL_INTR (1 << 0) /* reply wait was interrupted by user */
234 #define PTL_RPC_FL_TIMEOUT (1 << 7) /* request timed out waiting for reply */
236 #define REQ_MAX_ACK_LOCKS 8
238 union ptlrpc_async_args {
240 * Scratchpad for passing args to completion interpreter. Users
241 * cast to the struct of their choosing, and CLASSERT that this is
242 * big enough. For _tons_ of context, OBD_ALLOC a struct and store
243 * a pointer to it here. The pointer_arg ensures this struct is at
244 * least big enough for that.
246 void *pointer_arg[11];
250 struct ptlrpc_request_set;
251 typedef int (*set_interpreter_func)(struct ptlrpc_request_set *, void *, int);
254 * Definition of request set structure.
255 * Request set is a list of requests (not necessary to the same target) that
256 * once populated with RPCs could be sent in parallel.
257 * There are two kinds of request sets. General purpose and with dedicated
258 * serving thread. Example of the latter is ptlrpcd set.
259 * For general purpose sets once request set started sending it is impossible
260 * to add new requests to such set.
261 * Provides a way to call "completion callbacks" when all requests in the set
264 struct ptlrpc_request_set {
265 cfs_atomic_t set_refcount;
266 /** number of in queue requests */
267 cfs_atomic_t set_new_count;
268 /** number of uncompleted requests */
269 cfs_atomic_t set_remaining;
270 /** wait queue to wait on for request events */
271 cfs_waitq_t set_waitq;
272 cfs_waitq_t *set_wakeup_ptr;
273 /** List of requests in the set */
274 cfs_list_t set_requests;
276 * List of completion callbacks to be called when the set is completed
277 * This is only used if \a set_interpret is NULL.
278 * Links struct ptlrpc_set_cbdata.
280 cfs_list_t set_cblist;
281 /** Completion callback, if only one. */
282 set_interpreter_func set_interpret;
283 /** opaq argument passed to completion \a set_interpret callback. */
286 * Lock for \a set_new_requests manipulations
287 * locked so that any old caller can communicate requests to
288 * the set holder who can then fold them into the lock-free set
290 cfs_spinlock_t set_new_req_lock;
291 /** List of new yet unsent requests. Only used with ptlrpcd now. */
292 cfs_list_t set_new_requests;
296 * Description of a single ptrlrpc_set callback
298 struct ptlrpc_set_cbdata {
299 /** List linkage item */
301 /** Pointer to interpreting function */
302 set_interpreter_func psc_interpret;
303 /** Opaq argument to pass to the callback */
307 struct ptlrpc_bulk_desc;
310 * ptlrpc callback & work item stuff
312 struct ptlrpc_cb_id {
313 void (*cbid_fn)(lnet_event_t *ev); /* specific callback fn */
314 void *cbid_arg; /* additional arg */
317 /** Maximum number of locks to fit into reply state */
318 #define RS_MAX_LOCKS 8
322 * Structure to define reply state on the server
323 * Reply state holds various reply message information. Also for "difficult"
324 * replies (rep-ack case) we store the state after sending reply and wait
325 * for the client to acknowledge the reception. In these cases locks could be
326 * added to the state for replay/failover consistency guarantees.
328 struct ptlrpc_reply_state {
329 /** Callback description */
330 struct ptlrpc_cb_id rs_cb_id;
331 /** Linkage for list of all reply states in a system */
333 /** Linkage for list of all reply states on same export */
334 cfs_list_t rs_exp_list;
335 /** Linkage for list of all reply states for same obd */
336 cfs_list_t rs_obd_list;
338 cfs_list_t rs_debug_list;
340 /** A spinlock to protect the reply state flags */
341 cfs_spinlock_t rs_lock;
342 /** Reply state flags */
343 unsigned long rs_difficult:1; /* ACK/commit stuff */
344 unsigned long rs_no_ack:1; /* no ACK, even for
345 difficult requests */
346 unsigned long rs_scheduled:1; /* being handled? */
347 unsigned long rs_scheduled_ever:1;/* any schedule attempts? */
348 unsigned long rs_handled:1; /* been handled yet? */
349 unsigned long rs_on_net:1; /* reply_out_callback pending? */
350 unsigned long rs_prealloc:1; /* rs from prealloc list */
351 unsigned long rs_committed:1;/* the transaction was committed
352 and the rs was dispatched
353 by ptlrpc_commit_replies */
354 /** Size of the state */
358 /** Transaction number */
362 struct obd_export *rs_export;
363 struct ptlrpc_service *rs_service;
364 /** Lnet metadata handle for the reply */
365 lnet_handle_md_t rs_md_h;
366 cfs_atomic_t rs_refcount;
368 /** Context for the sevice thread */
369 struct ptlrpc_svc_ctx *rs_svc_ctx;
370 /** Reply buffer (actually sent to the client), encoded if needed */
371 struct lustre_msg *rs_repbuf; /* wrapper */
372 /** Size of the reply buffer */
373 int rs_repbuf_len; /* wrapper buf length */
374 /** Size of the reply message */
375 int rs_repdata_len; /* wrapper msg length */
377 * Actual reply message. Its content is encrupted (if needed) to
378 * produce reply buffer for actual sending. In simple case
379 * of no network encryption we jus set \a rs_repbuf to \a rs_msg
381 struct lustre_msg *rs_msg; /* reply message */
383 /** Number of locks awaiting client ACK */
385 /** Handles of locks awaiting client reply ACK */
386 struct lustre_handle rs_locks[RS_MAX_LOCKS];
387 /** Lock modes of locks in \a rs_locks */
388 ldlm_mode_t rs_modes[RS_MAX_LOCKS];
391 struct ptlrpc_thread;
395 RQ_PHASE_NEW = 0xebc0de00,
396 RQ_PHASE_RPC = 0xebc0de01,
397 RQ_PHASE_BULK = 0xebc0de02,
398 RQ_PHASE_INTERPRET = 0xebc0de03,
399 RQ_PHASE_COMPLETE = 0xebc0de04,
400 RQ_PHASE_UNREGISTERING = 0xebc0de05,
401 RQ_PHASE_UNDEFINED = 0xebc0de06
404 /** Type of request interpreter call-back */
405 typedef int (*ptlrpc_interpterer_t)(const struct lu_env *env,
406 struct ptlrpc_request *req,
410 * Definition of request pool structure.
411 * The pool is used to store empty preallocated requests for the case
412 * when we would actually need to send something without performing
413 * any allocations (to avoid e.g. OOM).
415 struct ptlrpc_request_pool {
416 /** Locks the list */
417 cfs_spinlock_t prp_lock;
418 /** list of ptlrpc_request structs */
419 cfs_list_t prp_req_list;
420 /** Maximum message size that would fit into a rquest from this pool */
422 /** Function to allocate more requests for this pool */
423 void (*prp_populate)(struct ptlrpc_request_pool *, int);
432 * Basic request prioritization operations structure.
433 * The whole idea is centered around locks and RPCs that might affect locks.
434 * When a lock is contended we try to give priority to RPCs that might lead
435 * to fastest release of that lock.
436 * Currently only implemented for OSTs only in a way that makes all
437 * IO and truncate RPCs that are coming from a locked region where a lock is
438 * contended a priority over other requests.
440 struct ptlrpc_hpreq_ops {
442 * Check if the lock handle of the given lock is the same as
443 * taken from the request.
445 int (*hpreq_lock_match)(struct ptlrpc_request *, struct ldlm_lock *);
447 * Check if the request is a high priority one.
449 int (*hpreq_check)(struct ptlrpc_request *);
451 * Called after the request has been handled.
453 void (*hpreq_fini)(struct ptlrpc_request *);
457 * Represents remote procedure call.
459 * This is a staple structure used by everybody wanting to send a request
462 struct ptlrpc_request {
463 /* Request type: one of PTL_RPC_MSG_* */
466 * Linkage item through which this request is included into
467 * sending/delayed lists on client and into rqbd list on server
471 * Server side list of incoming unserved requests sorted by arrival
472 * time. Traversed from time to time to notice about to expire
473 * requests and sent back "early replies" to clients to let them
474 * know server is alive and well, just very busy to service their
477 cfs_list_t rq_timed_list;
478 /** server-side history, used for debuging purposes. */
479 cfs_list_t rq_history_list;
480 /** server-side per-export list */
481 cfs_list_t rq_exp_list;
482 /** server-side hp handlers */
483 struct ptlrpc_hpreq_ops *rq_ops;
484 /** history sequence # */
485 __u64 rq_history_seq;
486 /** the index of service's srv_at_array into which request is linked */
488 /** Result of request processing */
490 /** Lock to protect request flags and some other important bits, like
493 cfs_spinlock_t rq_lock;
494 /** client-side flags are serialized by rq_lock */
495 unsigned long rq_intr:1, rq_replied:1, rq_err:1,
496 rq_timedout:1, rq_resend:1, rq_restart:1,
498 * when ->rq_replay is set, request is kept by the client even
499 * after server commits corresponding transaction. This is
500 * used for operations that require sequence of multiple
501 * requests to be replayed. The only example currently is file
502 * open/close. When last request in such a sequence is
503 * committed, ->rq_replay is cleared on all requests in the
507 rq_no_resend:1, rq_waiting:1, rq_receiving_reply:1,
508 rq_no_delay:1, rq_net_err:1, rq_wait_ctx:1,
509 rq_early:1, rq_must_unlink:1,
510 rq_fake:1, /* this fake req */
511 rq_memalloc:1, /* req originated from "kswapd" */
512 /* server-side flags */
513 rq_packed_final:1, /* packed final reply */
514 rq_hp:1, /* high priority RPC */
515 rq_at_linked:1, /* link into service's srv_at_array */
518 /* whether the "rq_set" is a valid one */
521 enum rq_phase rq_phase; /* one of RQ_PHASE_* */
522 enum rq_phase rq_next_phase; /* one of RQ_PHASE_* to be used next */
523 cfs_atomic_t rq_refcount;/* client-side refcount for SENT race,
524 server-side refcounf for multiple replies */
526 /** initial thread servicing this request */
527 struct ptlrpc_thread *rq_svc_thread;
529 /** Portal to which this request would be sent */
530 int rq_request_portal; /* XXX FIXME bug 249 */
531 /** Portal where to wait for reply and where reply would be sent */
532 int rq_reply_portal; /* XXX FIXME bug 249 */
536 * !rq_truncate : # reply bytes actually received,
537 * rq_truncate : required repbuf_len for resend
540 /** Request length */
542 /** Request message - what client sent */
543 struct lustre_msg *rq_reqmsg;
547 /** Reply message - server response */
548 struct lustre_msg *rq_repmsg;
549 /** Transaction number */
554 * List item to for replay list. Not yet commited requests get linked
556 * Also see \a rq_replay comment above.
558 cfs_list_t rq_replay_list;
561 * security and encryption data
563 struct ptlrpc_cli_ctx *rq_cli_ctx; /**< client's half ctx */
564 struct ptlrpc_svc_ctx *rq_svc_ctx; /**< server's half ctx */
565 cfs_list_t rq_ctx_chain; /**< link to waited ctx */
567 struct sptlrpc_flavor rq_flvr; /**< for client & server */
568 enum lustre_sec_part rq_sp_from;
570 unsigned long /* client/server security flags */
571 rq_ctx_init:1, /* context initiation */
572 rq_ctx_fini:1, /* context destroy */
573 rq_bulk_read:1, /* request bulk read */
574 rq_bulk_write:1, /* request bulk write */
575 /* server authentication flags */
576 rq_auth_gss:1, /* authenticated by gss */
577 rq_auth_remote:1, /* authed as remote user */
578 rq_auth_usr_root:1, /* authed as root */
579 rq_auth_usr_mdt:1, /* authed as mdt */
580 rq_auth_usr_ost:1, /* authed as ost */
581 /* security tfm flags */
584 /* doesn't expect reply FIXME */
586 rq_pill_init:1; /* pill initialized */
588 uid_t rq_auth_uid; /* authed uid */
589 uid_t rq_auth_mapped_uid; /* authed uid mapped to */
591 /* (server side), pointed directly into req buffer */
592 struct ptlrpc_user_desc *rq_user_desc;
594 /** early replies go to offset 0, regular replies go after that */
595 unsigned int rq_reply_off;
597 /* various buffer pointers */
598 struct lustre_msg *rq_reqbuf; /* req wrapper */
599 int rq_reqbuf_len; /* req wrapper buf len */
600 int rq_reqdata_len; /* req wrapper msg len */
601 char *rq_repbuf; /* rep buffer */
602 int rq_repbuf_len; /* rep buffer len */
603 struct lustre_msg *rq_repdata; /* rep wrapper msg */
604 int rq_repdata_len; /* rep wrapper msg len */
605 struct lustre_msg *rq_clrbuf; /* only in priv mode */
606 int rq_clrbuf_len; /* only in priv mode */
607 int rq_clrdata_len; /* only in priv mode */
611 /** Fields that help to see if request and reply were swabbed or not */
612 __u32 rq_req_swab_mask;
613 __u32 rq_rep_swab_mask;
615 /** What was import generation when this request was sent */
616 int rq_import_generation;
617 enum lustre_imp_state rq_send_state;
619 /** how many early replies (for stats) */
622 /** client+server request */
623 lnet_handle_md_t rq_req_md_h;
624 struct ptlrpc_cb_id rq_req_cbid;
625 /** optional time limit for send attempts */
626 cfs_duration_t rq_delay_limit;
627 /** time request was first queued */
628 cfs_time_t rq_queued_time;
631 /** request arrival time */
632 struct timeval rq_arrival_time;
633 /** separated reply state */
634 struct ptlrpc_reply_state *rq_reply_state;
635 /** incoming request buffer */
636 struct ptlrpc_request_buffer_desc *rq_rqbd;
638 __u32 rq_uid; /* peer uid, used in MDS only */
641 /** client-only incoming reply */
642 lnet_handle_md_t rq_reply_md_h;
643 cfs_waitq_t rq_reply_waitq;
644 struct ptlrpc_cb_id rq_reply_cbid;
648 /** Peer description (the other side) */
649 lnet_process_id_t rq_peer;
650 /** Server-side, export on which request was received */
651 struct obd_export *rq_export;
652 /** Client side, import where request is being sent */
653 struct obd_import *rq_import;
655 /** Replay callback, called after request is replayed at recovery */
656 void (*rq_replay_cb)(struct ptlrpc_request *);
658 * Commit callback, called when request is committed and about to be
661 void (*rq_commit_cb)(struct ptlrpc_request *);
662 /** Opaq data for replay and commit callbacks. */
665 /** For bulk requests on client only: bulk descriptor */
666 struct ptlrpc_bulk_desc *rq_bulk;
668 /** client outgoing req */
670 * when request/reply sent (secs), or time when request should be sent
673 /** time for request really sent out */
676 /** when request must finish. volatile
677 * so that servers' early reply updates to the deadline aren't
678 * kept in per-cpu cache */
679 volatile time_t rq_deadline;
680 /** when req reply unlink must finish. */
681 time_t rq_reply_deadline;
682 /** when req bulk unlink must finish. */
683 time_t rq_bulk_deadline;
685 * service time estimate (secs)
686 * If the requestsis not served by this time, it is marked as timed out.
690 /** Multi-rpc bits */
691 /** Link item for request set lists */
692 cfs_list_t rq_set_chain;
693 /** Per-request waitq introduced by bug 21938 for recovery waiting */
694 cfs_waitq_t rq_set_waitq;
695 /** Link back to the request set */
696 struct ptlrpc_request_set *rq_set;
697 /** Async completion handler, called when reply is received */
698 ptlrpc_interpterer_t rq_interpret_reply;
699 /** Async completion context */
700 union ptlrpc_async_args rq_async_args;
702 /** Pool if request is from preallocated list */
703 struct ptlrpc_request_pool *rq_pool;
705 struct lu_context rq_session;
706 struct lu_context rq_recov_session;
708 /** request format description */
709 struct req_capsule rq_pill;
713 * Call completion handler for rpc if any, return it's status or original
714 * rc if there was no handler defined for this request.
716 static inline int ptlrpc_req_interpret(const struct lu_env *env,
717 struct ptlrpc_request *req, int rc)
719 if (req->rq_interpret_reply != NULL) {
720 req->rq_status = req->rq_interpret_reply(env, req,
723 return req->rq_status;
729 * Returns 1 if request buffer at offset \a index was already swabbed
731 static inline int lustre_req_swabbed(struct ptlrpc_request *req, int index)
733 LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
734 return req->rq_req_swab_mask & (1 << index);
738 * Returns 1 if request reply buffer at offset \a index was already swabbed
740 static inline int lustre_rep_swabbed(struct ptlrpc_request *req, int index)
742 LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
743 return req->rq_rep_swab_mask & (1 << index);
747 * Returns 1 if request needs to be swabbed into local cpu byteorder
749 static inline int ptlrpc_req_need_swab(struct ptlrpc_request *req)
751 return lustre_req_swabbed(req, MSG_PTLRPC_HEADER_OFF);
755 * Returns 1 if request reply needs to be swabbed into local cpu byteorder
757 static inline int ptlrpc_rep_need_swab(struct ptlrpc_request *req)
759 return lustre_rep_swabbed(req, MSG_PTLRPC_HEADER_OFF);
763 * Mark request buffer at offset \a index that it was already swabbed
765 static inline void lustre_set_req_swabbed(struct ptlrpc_request *req, int index)
767 LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
768 LASSERT((req->rq_req_swab_mask & (1 << index)) == 0);
769 req->rq_req_swab_mask |= 1 << index;
773 * Mark request reply buffer at offset \a index that it was already swabbed
775 static inline void lustre_set_rep_swabbed(struct ptlrpc_request *req, int index)
777 LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
778 LASSERT((req->rq_rep_swab_mask & (1 << index)) == 0);
779 req->rq_rep_swab_mask |= 1 << index;
783 * Convert numerical request phase value \a phase into text string description
785 static inline const char *
786 ptlrpc_phase2str(enum rq_phase phase)
795 case RQ_PHASE_INTERPRET:
797 case RQ_PHASE_COMPLETE:
799 case RQ_PHASE_UNREGISTERING:
800 return "Unregistering";
807 * Convert numerical request phase of the request \a req into text stringi
810 static inline const char *
811 ptlrpc_rqphase2str(struct ptlrpc_request *req)
813 return ptlrpc_phase2str(req->rq_phase);
817 * Debugging functions and helpers to print request structure into debug log
820 /* Spare the preprocessor, spoil the bugs. */
821 #define FLAG(field, str) (field ? str : "")
823 /** Convert bit flags into a string */
824 #define DEBUG_REQ_FLAGS(req) \
825 ptlrpc_rqphase2str(req), \
826 FLAG(req->rq_intr, "I"), FLAG(req->rq_replied, "R"), \
827 FLAG(req->rq_err, "E"), \
828 FLAG(req->rq_timedout, "X") /* eXpired */, FLAG(req->rq_resend, "S"), \
829 FLAG(req->rq_restart, "T"), FLAG(req->rq_replay, "P"), \
830 FLAG(req->rq_no_resend, "N"), \
831 FLAG(req->rq_waiting, "W"), \
832 FLAG(req->rq_wait_ctx, "C"), FLAG(req->rq_hp, "H"), \
833 FLAG(req->rq_committed, "M")
835 #define REQ_FLAGS_FMT "%s:%s%s%s%s%s%s%s%s%s%s%s%s"
837 void _debug_req(struct ptlrpc_request *req,
838 struct libcfs_debug_msg_data *data, const char *fmt, ...)
839 __attribute__ ((format (printf, 3, 4)));
842 * Helper that decides if we need to print request accordig to current debug
845 #define debug_req(msgdata, mask, cdls, req, fmt, a...) \
847 CFS_CHECK_STACK(msgdata, mask, cdls); \
849 if (((mask) & D_CANTMASK) != 0 || \
850 ((libcfs_debug & (mask)) != 0 && \
851 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) \
852 _debug_req((req), msgdata, 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 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, level, &cdls); \
864 debug_req(&msgdata, level, &cdls, req, "@@@ "fmt" ", ## args);\
866 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, level, NULL); \
867 debug_req(&msgdata, level, NULL, req, "@@@ "fmt" ", ## args); \
873 * Structure that defines a single page of a bulk transfer
875 struct ptlrpc_bulk_page {
876 /** Linkage to list of pages in a bulk */
879 * Number of bytes in a page to transfer starting from \a bp_pageoffset
882 /** offset within a page */
884 /** The page itself */
885 struct page *bp_page;
888 #define BULK_GET_SOURCE 0
889 #define BULK_PUT_SINK 1
890 #define BULK_GET_SINK 2
891 #define BULK_PUT_SOURCE 3
894 * Definition of buk descriptor.
895 * Bulks are special "Two phase" RPCs where initial request message
896 * is sent first and it is followed bt a transfer (o receiving) of a large
897 * amount of data to be settled into pages referenced from the bulk descriptors.
898 * Bulks transfers (the actual data following the small requests) are done
899 * on separate LNet portals.
900 * In lustre we use bulk transfers for READ and WRITE transfers from/to OSTs.
901 * Another user is readpage for MDT.
903 struct ptlrpc_bulk_desc {
904 /** completed successfully */
905 unsigned long bd_success:1;
906 /** accessible to the network (network io potentially in progress) */
907 unsigned long bd_network_rw:1;
908 /** {put,get}{source,sink} */
909 unsigned long bd_type:2;
911 unsigned long bd_registered:1;
912 /** For serialization with callback */
913 cfs_spinlock_t bd_lock;
914 /** Import generation when request for this bulk was sent */
915 int bd_import_generation;
916 /** Server side - export this bulk created for */
917 struct obd_export *bd_export;
918 /** Client side - import this bulk was sent on */
919 struct obd_import *bd_import;
920 /** LNet portal for this bulk */
922 /** Back pointer to the request */
923 struct ptlrpc_request *bd_req;
924 cfs_waitq_t bd_waitq; /* server side only WQ */
925 int bd_iov_count; /* # entries in bd_iov */
926 int bd_max_iov; /* allocated size of bd_iov */
927 int bd_nob; /* # bytes covered */
928 int bd_nob_transferred; /* # bytes GOT/PUT */
932 struct ptlrpc_cb_id bd_cbid; /* network callback info */
933 lnet_handle_md_t bd_md_h; /* associated MD */
934 lnet_nid_t bd_sender; /* stash event::sender */
936 #if defined(__KERNEL__)
938 * encrypt iov, size is either 0 or bd_iov_count.
940 lnet_kiov_t *bd_enc_iov;
942 lnet_kiov_t bd_iov[0];
944 lnet_md_iovec_t bd_iov[0];
949 SVC_STOPPED = 1 << 0,
950 SVC_STOPPING = 1 << 1,
951 SVC_STARTING = 1 << 2,
952 SVC_RUNNING = 1 << 3,
958 * Definition of server service thread structure
960 struct ptlrpc_thread {
962 * List of active threads in svc->srv_threads
966 * thread-private data (preallocated memory)
971 * service thread index, from ptlrpc_start_threads
979 * put watchdog in the structure per thread b=14840
981 struct lc_watchdog *t_watchdog;
983 * the svc this thread belonged to b=18582
985 struct ptlrpc_service *t_svc;
986 cfs_waitq_t t_ctl_waitq;
987 struct lu_env *t_env;
990 static inline int thread_is_stopped(struct ptlrpc_thread *thread)
992 return !!(thread->t_flags & SVC_STOPPED);
995 static inline int thread_is_stopping(struct ptlrpc_thread *thread)
997 return !!(thread->t_flags & SVC_STOPPING);
1000 static inline int thread_is_starting(struct ptlrpc_thread *thread)
1002 return !!(thread->t_flags & SVC_STARTING);
1005 static inline int thread_is_running(struct ptlrpc_thread *thread)
1007 return !!(thread->t_flags & SVC_RUNNING);
1010 static inline int thread_is_event(struct ptlrpc_thread *thread)
1012 return !!(thread->t_flags & SVC_EVENT);
1015 static inline int thread_is_signal(struct ptlrpc_thread *thread)
1017 return !!(thread->t_flags & SVC_SIGNAL);
1020 static inline void thread_clear_flags(struct ptlrpc_thread *thread, __u32 flags)
1022 thread->t_flags &= ~flags;
1025 static inline void thread_set_flags(struct ptlrpc_thread *thread, __u32 flags)
1027 thread->t_flags = flags;
1030 static inline void thread_add_flags(struct ptlrpc_thread *thread, __u32 flags)
1032 thread->t_flags |= flags;
1035 static inline int thread_test_and_clear_flags(struct ptlrpc_thread *thread,
1038 if (thread->t_flags & flags) {
1039 thread->t_flags &= ~flags;
1046 * Request buffer descriptor structure.
1047 * This is a structure that contains one posted request buffer for service.
1048 * Once data land into a buffer, event callback creates actual request and
1049 * notifies wakes one of the service threads to process new incoming request.
1050 * More than one request can fit into the buffer.
1052 struct ptlrpc_request_buffer_desc {
1053 /** Link item for rqbds on a service */
1054 cfs_list_t rqbd_list;
1055 /** History of requests for this buffer */
1056 cfs_list_t rqbd_reqs;
1057 /** Back pointer to service for which this buffer is registered */
1058 struct ptlrpc_service *rqbd_service;
1059 /** LNet descriptor */
1060 lnet_handle_md_t rqbd_md_h;
1062 /** The buffer itself */
1064 struct ptlrpc_cb_id rqbd_cbid;
1066 * This "embedded" request structure is only used for the
1067 * last request to fit into the buffer
1069 struct ptlrpc_request rqbd_req;
1072 typedef int (*svc_thr_init_t)(struct ptlrpc_thread *thread);
1073 typedef void (*svc_thr_done_t)(struct ptlrpc_thread *thread);
1074 typedef int (*svc_handler_t)(struct ptlrpc_request *req);
1075 typedef int (*svc_hpreq_handler_t)(struct ptlrpc_request *);
1076 typedef void (*svc_req_printfn_t)(void *, struct ptlrpc_request *);
1078 #ifndef __cfs_cacheline_aligned
1079 /* NB: put it here for reducing patche dependence */
1080 # define __cfs_cacheline_aligned
1084 * How many high priority requests to serve before serving one normal
1087 #define PTLRPC_SVC_HP_RATIO 10
1090 * Definition of PortalRPC service.
1091 * The service is listening on a particular portal (like tcp port)
1092 * and perform actions for a specific server like IO service for OST
1093 * or general metadata service for MDS.
1095 * ptlrpc service has four locks:
1097 * serialize operations on rqbd and requests waiting for preprocess
1099 * serialize operations active requests sent to this portal
1101 * serialize adaptive timeout stuff
1103 * serialize operations on RS list (reply states)
1105 * We don't have any use-case to take two or more locks at the same time
1106 * for now, so there is no lock order issue.
1108 struct ptlrpc_service {
1109 /** most often accessed fields */
1110 /** chain thru all services */
1111 cfs_list_t srv_list;
1112 /** only statically allocated strings here; we don't clean them */
1114 /** only statically allocated strings here; we don't clean them */
1115 char *srv_thread_name;
1116 /** service thread list */
1117 cfs_list_t srv_threads;
1118 /** threads to start at beginning of service */
1119 int srv_threads_min;
1120 /** thread upper limit */
1121 int srv_threads_max;
1122 /** always increasing number */
1123 unsigned srv_threads_next_id;
1124 /** # of starting threads */
1125 int srv_threads_starting;
1126 /** # running threads */
1127 int srv_threads_running;
1129 /** service operations, move to ptlrpc_svc_ops_t in the future */
1132 * if non-NULL called during thread creation (ptlrpc_start_thread())
1133 * to initialize service specific per-thread state.
1135 svc_thr_init_t srv_init;
1137 * if non-NULL called during thread shutdown (ptlrpc_main()) to
1138 * destruct state created by ->srv_init().
1140 svc_thr_done_t srv_done;
1141 /** Handler function for incoming requests for this service */
1142 svc_handler_t srv_handler;
1143 /** hp request handler */
1144 svc_hpreq_handler_t srv_hpreq_handler;
1145 /** service-specific print fn */
1146 svc_req_printfn_t srv_req_printfn;
1149 /** Root of /proc dir tree for this service */
1150 cfs_proc_dir_entry_t *srv_procroot;
1151 /** Pointer to statistic data for this service */
1152 struct lprocfs_stats *srv_stats;
1153 /** # hp per lp reqs to handle */
1154 int srv_hpreq_ratio;
1155 /** biggest request to receive */
1156 int srv_max_req_size;
1157 /** biggest reply to send */
1158 int srv_max_reply_size;
1159 /** size of individual buffers */
1161 /** # buffers to allocate in 1 group */
1162 int srv_nbuf_per_group;
1163 /** Local portal on which to receive requests */
1164 __u32 srv_req_portal;
1165 /** Portal on the client to send replies to */
1166 __u32 srv_rep_portal;
1168 * Tags for lu_context associated with this thread, see struct
1172 /** soft watchdog timeout multiplier */
1173 int srv_watchdog_factor;
1174 /** bind threads to CPUs */
1175 unsigned srv_cpu_affinity:1;
1176 /** under unregister_service */
1177 unsigned srv_is_stopping:1;
1180 * serialize the following fields, used for protecting
1181 * rqbd list and incoming requests waiting for preprocess
1183 cfs_spinlock_t srv_lock __cfs_cacheline_aligned;
1184 /** incoming reqs */
1185 cfs_list_t srv_req_in_queue;
1186 /** total # req buffer descs allocated */
1188 /** # posted request buffers */
1189 int srv_nrqbd_receiving;
1190 /** timeout before re-posting reqs, in tick */
1191 cfs_duration_t srv_rqbd_timeout;
1192 /** request buffers to be reposted */
1193 cfs_list_t srv_idle_rqbds;
1194 /** req buffers receiving */
1195 cfs_list_t srv_active_rqbds;
1196 /** request buffer history */
1197 cfs_list_t srv_history_rqbds;
1198 /** # request buffers in history */
1199 int srv_n_history_rqbds;
1200 /** max # request buffers in history */
1201 int srv_max_history_rqbds;
1202 /** request history */
1203 cfs_list_t srv_request_history;
1204 /** next request sequence # */
1205 __u64 srv_request_seq;
1206 /** highest seq culled from history */
1207 __u64 srv_request_max_cull_seq;
1209 * all threads sleep on this. This wait-queue is signalled when new
1210 * incoming request arrives and when difficult reply has to be handled.
1212 cfs_waitq_t srv_waitq;
1215 * serialize the following fields, used for processing requests
1216 * sent to this portal
1218 cfs_spinlock_t srv_rq_lock __cfs_cacheline_aligned;
1219 /** # reqs in either of the queues below */
1220 /** reqs waiting for service */
1221 cfs_list_t srv_request_queue;
1222 /** high priority queue */
1223 cfs_list_t srv_request_hpq;
1224 /** # incoming reqs */
1225 int srv_n_queued_reqs;
1226 /** # reqs being served */
1227 int srv_n_active_reqs;
1228 /** # HPreqs being served */
1229 int srv_n_active_hpreq;
1230 /** # hp requests handled */
1231 int srv_hpreq_count;
1236 * serialize the following fields, used for changes on
1239 cfs_spinlock_t srv_at_lock __cfs_cacheline_aligned;
1240 /** estimated rpc service time */
1241 struct adaptive_timeout srv_at_estimate;
1242 /** reqs waiting for replies */
1243 struct ptlrpc_at_array srv_at_array;
1244 /** early reply timer */
1245 cfs_timer_t srv_at_timer;
1246 /** check early replies */
1247 unsigned srv_at_check;
1249 cfs_time_t srv_at_checktime;
1253 * serialize the following fields, used for processing
1254 * replies for this portal
1256 cfs_spinlock_t srv_rs_lock __cfs_cacheline_aligned;
1257 /** all the active replies */
1258 cfs_list_t srv_active_replies;
1260 /** replies waiting for service */
1261 cfs_list_t srv_reply_queue;
1263 /** List of free reply_states */
1264 cfs_list_t srv_free_rs_list;
1265 /** waitq to run, when adding stuff to srv_free_rs_list */
1266 cfs_waitq_t srv_free_rs_waitq;
1267 /** # 'difficult' replies */
1268 cfs_atomic_t srv_n_difficult_replies;
1269 //struct ptlrpc_srv_ni srv_interfaces[0];
1273 * Declaration of ptlrpcd control structure
1275 struct ptlrpcd_ctl {
1277 * Ptlrpc thread control flags (LIOD_START, LIOD_STOP, LIOD_FORCE)
1279 unsigned long pc_flags;
1281 * Thread lock protecting structure fields.
1283 cfs_spinlock_t pc_lock;
1287 cfs_completion_t pc_starting;
1291 cfs_completion_t pc_finishing;
1293 * Thread requests set.
1295 struct ptlrpc_request_set *pc_set;
1297 * Thread name used in cfs_daemonize()
1301 * Environment for request interpreters to run in.
1303 struct lu_env pc_env;
1305 * Index of ptlrpcd thread in the array.
1309 * Number of the ptlrpcd's partners.
1313 * Pointer to the array of partners' ptlrpcd_ctl structure.
1315 struct ptlrpcd_ctl **pc_partners;
1317 * Record the partner index to be processed next.
1322 * Async rpcs flag to make sure that ptlrpcd_check() is called only
1327 * Currently not used.
1331 * User-space async rpcs callback.
1333 void *pc_wait_callback;
1335 * User-space check idle rpcs callback.
1337 void *pc_idle_callback;
1341 /* Bits for pc_flags */
1342 enum ptlrpcd_ctl_flags {
1344 * Ptlrpc thread start flag.
1346 LIOD_START = 1 << 0,
1348 * Ptlrpc thread stop flag.
1352 * Ptlrpc thread force flag (only stop force so far).
1353 * This will cause aborting any inflight rpcs handled
1354 * by thread if LIOD_STOP is specified.
1356 LIOD_FORCE = 1 << 2,
1358 * This is a recovery ptlrpc thread.
1360 LIOD_RECOVERY = 1 << 3,
1362 * The ptlrpcd is bound to some CPU core.
1367 /* ptlrpc/events.c */
1368 extern lnet_handle_eq_t ptlrpc_eq_h;
1369 extern int ptlrpc_uuid_to_peer(struct obd_uuid *uuid,
1370 lnet_process_id_t *peer, lnet_nid_t *self);
1372 * These callbacks are invoked by LNet when something happened to
1376 extern void request_out_callback (lnet_event_t *ev);
1377 extern void reply_in_callback(lnet_event_t *ev);
1378 extern void client_bulk_callback (lnet_event_t *ev);
1379 extern void request_in_callback(lnet_event_t *ev);
1380 extern void reply_out_callback(lnet_event_t *ev);
1381 extern void server_bulk_callback (lnet_event_t *ev);
1384 /* ptlrpc/connection.c */
1385 struct ptlrpc_connection *ptlrpc_connection_get(lnet_process_id_t peer,
1387 struct obd_uuid *uuid);
1388 int ptlrpc_connection_put(struct ptlrpc_connection *c);
1389 struct ptlrpc_connection *ptlrpc_connection_addref(struct ptlrpc_connection *);
1390 int ptlrpc_connection_init(void);
1391 void ptlrpc_connection_fini(void);
1392 extern lnet_pid_t ptl_get_pid(void);
1394 /* ptlrpc/niobuf.c */
1396 * Actual interfacing with LNet to put/get/register/unregister stuff
1399 int ptlrpc_start_bulk_transfer(struct ptlrpc_bulk_desc *desc);
1400 void ptlrpc_abort_bulk(struct ptlrpc_bulk_desc *desc);
1401 int ptlrpc_register_bulk(struct ptlrpc_request *req);
1402 int ptlrpc_unregister_bulk(struct ptlrpc_request *req, int async);
1404 static inline int ptlrpc_server_bulk_active(struct ptlrpc_bulk_desc *desc)
1408 LASSERT(desc != NULL);
1410 cfs_spin_lock(&desc->bd_lock);
1411 rc = desc->bd_network_rw;
1412 cfs_spin_unlock(&desc->bd_lock);
1416 static inline int ptlrpc_client_bulk_active(struct ptlrpc_request *req)
1418 struct ptlrpc_bulk_desc *desc = req->rq_bulk;
1421 LASSERT(req != NULL);
1423 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_BULK_UNLINK) &&
1424 req->rq_bulk_deadline > cfs_time_current_sec())
1430 cfs_spin_lock(&desc->bd_lock);
1431 rc = desc->bd_network_rw;
1432 cfs_spin_unlock(&desc->bd_lock);
1436 #define PTLRPC_REPLY_MAYBE_DIFFICULT 0x01
1437 #define PTLRPC_REPLY_EARLY 0x02
1438 int ptlrpc_send_reply(struct ptlrpc_request *req, int flags);
1439 int ptlrpc_reply(struct ptlrpc_request *req);
1440 int ptlrpc_send_error(struct ptlrpc_request *req, int difficult);
1441 int ptlrpc_error(struct ptlrpc_request *req);
1442 void ptlrpc_resend_req(struct ptlrpc_request *request);
1443 int ptlrpc_at_get_net_latency(struct ptlrpc_request *req);
1444 int ptl_send_rpc(struct ptlrpc_request *request, int noreply);
1445 int ptlrpc_register_rqbd (struct ptlrpc_request_buffer_desc *rqbd);
1448 /* ptlrpc/client.c */
1450 * Client-side portals API. Everything to send requests, receive replies,
1451 * request queues, request management, etc.
1454 void ptlrpc_init_client(int req_portal, int rep_portal, char *name,
1455 struct ptlrpc_client *);
1456 void ptlrpc_cleanup_client(struct obd_import *imp);
1457 struct ptlrpc_connection *ptlrpc_uuid_to_connection(struct obd_uuid *uuid);
1459 int ptlrpc_queue_wait(struct ptlrpc_request *req);
1460 int ptlrpc_replay_req(struct ptlrpc_request *req);
1461 int ptlrpc_unregister_reply(struct ptlrpc_request *req, int async);
1462 void ptlrpc_restart_req(struct ptlrpc_request *req);
1463 void ptlrpc_abort_inflight(struct obd_import *imp);
1464 void ptlrpc_cleanup_imp(struct obd_import *imp);
1465 void ptlrpc_abort_set(struct ptlrpc_request_set *set);
1467 struct ptlrpc_request_set *ptlrpc_prep_set(void);
1468 int ptlrpc_set_add_cb(struct ptlrpc_request_set *set,
1469 set_interpreter_func fn, void *data);
1470 int ptlrpc_set_next_timeout(struct ptlrpc_request_set *);
1471 int ptlrpc_check_set(const struct lu_env *env, struct ptlrpc_request_set *set);
1472 int ptlrpc_set_wait(struct ptlrpc_request_set *);
1473 int ptlrpc_expired_set(void *data);
1474 void ptlrpc_interrupted_set(void *data);
1475 void ptlrpc_mark_interrupted(struct ptlrpc_request *req);
1476 void ptlrpc_set_destroy(struct ptlrpc_request_set *);
1477 void ptlrpc_set_add_req(struct ptlrpc_request_set *, struct ptlrpc_request *);
1478 void ptlrpc_set_add_new_req(struct ptlrpcd_ctl *pc,
1479 struct ptlrpc_request *req);
1481 void ptlrpc_free_rq_pool(struct ptlrpc_request_pool *pool);
1482 void ptlrpc_add_rqs_to_pool(struct ptlrpc_request_pool *pool, int num_rq);
1484 struct ptlrpc_request_pool *
1485 ptlrpc_init_rq_pool(int, int,
1486 void (*populate_pool)(struct ptlrpc_request_pool *, int));
1488 void ptlrpc_at_set_req_timeout(struct ptlrpc_request *req);
1489 struct ptlrpc_request *ptlrpc_request_alloc(struct obd_import *imp,
1490 const struct req_format *format);
1491 struct ptlrpc_request *ptlrpc_request_alloc_pool(struct obd_import *imp,
1492 struct ptlrpc_request_pool *,
1493 const struct req_format *format);
1494 void ptlrpc_request_free(struct ptlrpc_request *request);
1495 int ptlrpc_request_pack(struct ptlrpc_request *request,
1496 __u32 version, int opcode);
1497 struct ptlrpc_request *ptlrpc_request_alloc_pack(struct obd_import *imp,
1498 const struct req_format *format,
1499 __u32 version, int opcode);
1500 int ptlrpc_request_bufs_pack(struct ptlrpc_request *request,
1501 __u32 version, int opcode, char **bufs,
1502 struct ptlrpc_cli_ctx *ctx);
1503 struct ptlrpc_request *ptlrpc_prep_fakereq(struct obd_import *imp,
1504 unsigned int timeout,
1505 ptlrpc_interpterer_t interpreter);
1506 void ptlrpc_fakereq_finished(struct ptlrpc_request *req);
1508 struct ptlrpc_request *ptlrpc_prep_req(struct obd_import *imp, __u32 version,
1509 int opcode, int count, __u32 *lengths,
1511 struct ptlrpc_request *ptlrpc_prep_req_pool(struct obd_import *imp,
1512 __u32 version, int opcode,
1513 int count, __u32 *lengths, char **bufs,
1514 struct ptlrpc_request_pool *pool);
1515 void ptlrpc_req_finished(struct ptlrpc_request *request);
1516 void ptlrpc_req_finished_with_imp_lock(struct ptlrpc_request *request);
1517 struct ptlrpc_request *ptlrpc_request_addref(struct ptlrpc_request *req);
1518 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_imp (struct ptlrpc_request *req,
1519 int npages, int type, int portal);
1520 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_exp(struct ptlrpc_request *req,
1521 int npages, int type, int portal);
1522 void ptlrpc_free_bulk(struct ptlrpc_bulk_desc *bulk);
1523 void ptlrpc_prep_bulk_page(struct ptlrpc_bulk_desc *desc,
1524 cfs_page_t *page, int pageoffset, int len);
1525 void ptlrpc_retain_replayable_request(struct ptlrpc_request *req,
1526 struct obd_import *imp);
1527 __u64 ptlrpc_next_xid(void);
1528 __u64 ptlrpc_sample_next_xid(void);
1529 __u64 ptlrpc_req_xid(struct ptlrpc_request *request);
1531 /* Set of routines to run a function in ptlrpcd context */
1532 void *ptlrpcd_alloc_work(struct obd_import *imp,
1533 int (*cb)(const struct lu_env *, void *), void *data);
1534 void ptlrpcd_destroy_work(void *handler);
1535 int ptlrpcd_queue_work(void *handler);
1539 struct ptlrpc_service_conf {
1542 int psc_max_req_size;
1543 int psc_max_reply_size;
1546 int psc_watchdog_factor;
1547 int psc_min_threads;
1548 int psc_max_threads;
1552 /* ptlrpc/service.c */
1554 * Server-side services API. Register/unregister service, request state
1555 * management, service thread management
1559 void ptlrpc_save_lock (struct ptlrpc_request *req,
1560 struct lustre_handle *lock, int mode, int no_ack);
1561 void ptlrpc_commit_replies(struct obd_export *exp);
1562 void ptlrpc_dispatch_difficult_reply (struct ptlrpc_reply_state *rs);
1563 void ptlrpc_schedule_difficult_reply (struct ptlrpc_reply_state *rs);
1564 struct ptlrpc_service *ptlrpc_init_svc_conf(struct ptlrpc_service_conf *c,
1565 svc_handler_t h, char *name,
1566 struct proc_dir_entry *proc_entry,
1567 svc_req_printfn_t prntfn,
1570 struct ptlrpc_service *ptlrpc_init_svc(int nbufs, int bufsize, int max_req_size,
1572 int req_portal, int rep_portal,
1573 int watchdog_factor,
1574 svc_handler_t, char *name,
1575 cfs_proc_dir_entry_t *proc_entry,
1577 int min_threads, int max_threads,
1578 char *threadname, __u32 ctx_tags,
1579 svc_hpreq_handler_t);
1580 void ptlrpc_stop_all_threads(struct ptlrpc_service *svc);
1582 int ptlrpc_start_threads(struct ptlrpc_service *svc);
1583 int ptlrpc_start_thread(struct ptlrpc_service *svc);
1584 int ptlrpc_unregister_service(struct ptlrpc_service *service);
1585 int liblustre_check_services (void *arg);
1586 void ptlrpc_daemonize(char *name);
1587 int ptlrpc_service_health_check(struct ptlrpc_service *);
1588 void ptlrpc_hpreq_reorder(struct ptlrpc_request *req);
1589 void ptlrpc_server_drop_request(struct ptlrpc_request *req);
1592 int ptlrpc_hr_init(void);
1593 void ptlrpc_hr_fini(void);
1595 # define ptlrpc_hr_init() (0)
1596 # define ptlrpc_hr_fini() do {} while(0)
1599 struct ptlrpc_svc_data {
1601 struct ptlrpc_service *svc;
1602 struct ptlrpc_thread *thread;
1606 /* ptlrpc/import.c */
1611 int ptlrpc_connect_import(struct obd_import *imp);
1612 int ptlrpc_init_import(struct obd_import *imp);
1613 int ptlrpc_disconnect_import(struct obd_import *imp, int noclose);
1614 int ptlrpc_import_recovery_state_machine(struct obd_import *imp);
1616 /* ptlrpc/pack_generic.c */
1617 int ptlrpc_reconnect_import(struct obd_import *imp);
1621 * ptlrpc msg buffer and swab interface
1625 int ptlrpc_buf_need_swab(struct ptlrpc_request *req, const int inout,
1627 void ptlrpc_buf_set_swabbed(struct ptlrpc_request *req, const int inout,
1629 int ptlrpc_unpack_rep_msg(struct ptlrpc_request *req, int len);
1630 int ptlrpc_unpack_req_msg(struct ptlrpc_request *req, int len);
1632 int lustre_msg_check_version(struct lustre_msg *msg, __u32 version);
1633 void lustre_init_msg_v2(struct lustre_msg_v2 *msg, int count, __u32 *lens,
1635 int lustre_pack_request(struct ptlrpc_request *, __u32 magic, int count,
1636 __u32 *lens, char **bufs);
1637 int lustre_pack_reply(struct ptlrpc_request *, int count, __u32 *lens,
1639 int lustre_pack_reply_v2(struct ptlrpc_request *req, int count,
1640 __u32 *lens, char **bufs, int flags);
1641 #define LPRFL_EARLY_REPLY 1
1642 int lustre_pack_reply_flags(struct ptlrpc_request *, int count, __u32 *lens,
1643 char **bufs, int flags);
1644 int lustre_shrink_msg(struct lustre_msg *msg, int segment,
1645 unsigned int newlen, int move_data);
1646 void lustre_free_reply_state(struct ptlrpc_reply_state *rs);
1647 int __lustre_unpack_msg(struct lustre_msg *m, int len);
1648 int lustre_msg_hdr_size(__u32 magic, int count);
1649 int lustre_msg_size(__u32 magic, int count, __u32 *lengths);
1650 int lustre_msg_size_v2(int count, __u32 *lengths);
1651 int lustre_packed_msg_size(struct lustre_msg *msg);
1652 int lustre_msg_early_size(void);
1653 void *lustre_msg_buf_v2(struct lustre_msg_v2 *m, int n, int min_size);
1654 void *lustre_msg_buf(struct lustre_msg *m, int n, int minlen);
1655 int lustre_msg_buflen(struct lustre_msg *m, int n);
1656 void lustre_msg_set_buflen(struct lustre_msg *m, int n, int len);
1657 int lustre_msg_bufcount(struct lustre_msg *m);
1658 char *lustre_msg_string (struct lustre_msg *m, int n, int max_len);
1659 __u32 lustre_msghdr_get_flags(struct lustre_msg *msg);
1660 void lustre_msghdr_set_flags(struct lustre_msg *msg, __u32 flags);
1661 __u32 lustre_msg_get_flags(struct lustre_msg *msg);
1662 void lustre_msg_add_flags(struct lustre_msg *msg, int flags);
1663 void lustre_msg_set_flags(struct lustre_msg *msg, int flags);
1664 void lustre_msg_clear_flags(struct lustre_msg *msg, int flags);
1665 __u32 lustre_msg_get_op_flags(struct lustre_msg *msg);
1666 void lustre_msg_add_op_flags(struct lustre_msg *msg, int flags);
1667 void lustre_msg_set_op_flags(struct lustre_msg *msg, int flags);
1668 struct lustre_handle *lustre_msg_get_handle(struct lustre_msg *msg);
1669 __u32 lustre_msg_get_type(struct lustre_msg *msg);
1670 __u32 lustre_msg_get_version(struct lustre_msg *msg);
1671 void lustre_msg_add_version(struct lustre_msg *msg, int version);
1672 __u32 lustre_msg_get_opc(struct lustre_msg *msg);
1673 __u64 lustre_msg_get_last_xid(struct lustre_msg *msg);
1674 __u64 lustre_msg_get_last_committed(struct lustre_msg *msg);
1675 __u64 *lustre_msg_get_versions(struct lustre_msg *msg);
1676 __u64 lustre_msg_get_transno(struct lustre_msg *msg);
1677 __u64 lustre_msg_get_slv(struct lustre_msg *msg);
1678 __u32 lustre_msg_get_limit(struct lustre_msg *msg);
1679 void lustre_msg_set_slv(struct lustre_msg *msg, __u64 slv);
1680 void lustre_msg_set_limit(struct lustre_msg *msg, __u64 limit);
1681 int lustre_msg_get_status(struct lustre_msg *msg);
1682 __u32 lustre_msg_get_conn_cnt(struct lustre_msg *msg);
1683 int lustre_msg_is_v1(struct lustre_msg *msg);
1684 __u32 lustre_msg_get_magic(struct lustre_msg *msg);
1685 __u32 lustre_msg_get_timeout(struct lustre_msg *msg);
1686 __u32 lustre_msg_get_service_time(struct lustre_msg *msg);
1687 __u32 lustre_msg_get_cksum(struct lustre_msg *msg);
1688 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 0, 0)
1689 __u32 lustre_msg_calc_cksum(struct lustre_msg *msg, int compat18);
1691 # warning "remove checksum compatibility support for b1_8"
1692 __u32 lustre_msg_calc_cksum(struct lustre_msg *msg);
1694 void lustre_msg_set_handle(struct lustre_msg *msg,struct lustre_handle *handle);
1695 void lustre_msg_set_type(struct lustre_msg *msg, __u32 type);
1696 void lustre_msg_set_opc(struct lustre_msg *msg, __u32 opc);
1697 void lustre_msg_set_last_xid(struct lustre_msg *msg, __u64 last_xid);
1698 void lustre_msg_set_last_committed(struct lustre_msg *msg,__u64 last_committed);
1699 void lustre_msg_set_versions(struct lustre_msg *msg, __u64 *versions);
1700 void lustre_msg_set_transno(struct lustre_msg *msg, __u64 transno);
1701 void lustre_msg_set_status(struct lustre_msg *msg, __u32 status);
1702 void lustre_msg_set_conn_cnt(struct lustre_msg *msg, __u32 conn_cnt);
1703 void ptlrpc_req_set_repsize(struct ptlrpc_request *req, int count, __u32 *sizes);
1704 void ptlrpc_request_set_replen(struct ptlrpc_request *req);
1705 void lustre_msg_set_timeout(struct lustre_msg *msg, __u32 timeout);
1706 void lustre_msg_set_service_time(struct lustre_msg *msg, __u32 service_time);
1707 void lustre_msg_set_cksum(struct lustre_msg *msg, __u32 cksum);
1710 lustre_shrink_reply(struct ptlrpc_request *req, int segment,
1711 unsigned int newlen, int move_data)
1713 LASSERT(req->rq_reply_state);
1714 LASSERT(req->rq_repmsg);
1715 req->rq_replen = lustre_shrink_msg(req->rq_repmsg, segment,
1720 /** Change request phase of \a req to \a new_phase */
1722 ptlrpc_rqphase_move(struct ptlrpc_request *req, enum rq_phase new_phase)
1724 if (req->rq_phase == new_phase)
1727 if (new_phase == RQ_PHASE_UNREGISTERING) {
1728 req->rq_next_phase = req->rq_phase;
1730 cfs_atomic_inc(&req->rq_import->imp_unregistering);
1733 if (req->rq_phase == RQ_PHASE_UNREGISTERING) {
1735 cfs_atomic_dec(&req->rq_import->imp_unregistering);
1738 DEBUG_REQ(D_INFO, req, "move req \"%s\" -> \"%s\"",
1739 ptlrpc_rqphase2str(req), ptlrpc_phase2str(new_phase));
1741 req->rq_phase = new_phase;
1745 * Returns true if request \a req got early reply and hard deadline is not met
1748 ptlrpc_client_early(struct ptlrpc_request *req)
1750 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
1751 req->rq_reply_deadline > cfs_time_current_sec())
1753 return req->rq_early;
1757 * Returns true if we got real reply from server for this request
1760 ptlrpc_client_replied(struct ptlrpc_request *req)
1762 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
1763 req->rq_reply_deadline > cfs_time_current_sec())
1765 return req->rq_replied;
1768 /** Returns true if request \a req is in process of receiving server reply */
1770 ptlrpc_client_recv(struct ptlrpc_request *req)
1772 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
1773 req->rq_reply_deadline > cfs_time_current_sec())
1775 return req->rq_receiving_reply;
1779 ptlrpc_client_recv_or_unlink(struct ptlrpc_request *req)
1783 cfs_spin_lock(&req->rq_lock);
1784 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
1785 req->rq_reply_deadline > cfs_time_current_sec()) {
1786 cfs_spin_unlock(&req->rq_lock);
1789 rc = req->rq_receiving_reply || req->rq_must_unlink;
1790 cfs_spin_unlock(&req->rq_lock);
1795 ptlrpc_client_wake_req(struct ptlrpc_request *req)
1797 if (req->rq_set == NULL)
1798 cfs_waitq_signal(&req->rq_reply_waitq);
1800 cfs_waitq_signal(&req->rq_set->set_waitq);
1804 ptlrpc_rs_addref(struct ptlrpc_reply_state *rs)
1806 LASSERT(cfs_atomic_read(&rs->rs_refcount) > 0);
1807 cfs_atomic_inc(&rs->rs_refcount);
1811 ptlrpc_rs_decref(struct ptlrpc_reply_state *rs)
1813 LASSERT(cfs_atomic_read(&rs->rs_refcount) > 0);
1814 if (cfs_atomic_dec_and_test(&rs->rs_refcount))
1815 lustre_free_reply_state(rs);
1818 /* Should only be called once per req */
1819 static inline void ptlrpc_req_drop_rs(struct ptlrpc_request *req)
1821 if (req->rq_reply_state == NULL)
1822 return; /* shouldn't occur */
1823 ptlrpc_rs_decref(req->rq_reply_state);
1824 req->rq_reply_state = NULL;
1825 req->rq_repmsg = NULL;
1828 static inline __u32 lustre_request_magic(struct ptlrpc_request *req)
1830 return lustre_msg_get_magic(req->rq_reqmsg);
1833 static inline int ptlrpc_req_get_repsize(struct ptlrpc_request *req)
1835 switch (req->rq_reqmsg->lm_magic) {
1836 case LUSTRE_MSG_MAGIC_V2:
1837 return req->rq_reqmsg->lm_repsize;
1839 LASSERTF(0, "incorrect message magic: %08x\n",
1840 req->rq_reqmsg->lm_magic);
1845 static inline int ptlrpc_send_limit_expired(struct ptlrpc_request *req)
1847 if (req->rq_delay_limit != 0 &&
1848 cfs_time_before(cfs_time_add(req->rq_queued_time,
1849 cfs_time_seconds(req->rq_delay_limit)),
1850 cfs_time_current())) {
1856 static inline int ptlrpc_no_resend(struct ptlrpc_request *req)
1858 if (!req->rq_no_resend && ptlrpc_send_limit_expired(req)) {
1859 cfs_spin_lock(&req->rq_lock);
1860 req->rq_no_resend = 1;
1861 cfs_spin_unlock(&req->rq_lock);
1863 return req->rq_no_resend;
1866 /* ldlm/ldlm_lib.c */
1868 * Target client logic
1871 int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg);
1872 int client_obd_cleanup(struct obd_device *obddev);
1873 int client_connect_import(const struct lu_env *env,
1874 struct obd_export **exp, struct obd_device *obd,
1875 struct obd_uuid *cluuid, struct obd_connect_data *,
1877 int client_disconnect_export(struct obd_export *exp);
1878 int client_import_add_conn(struct obd_import *imp, struct obd_uuid *uuid,
1880 int client_import_del_conn(struct obd_import *imp, struct obd_uuid *uuid);
1881 int client_import_find_conn(struct obd_import *imp, lnet_nid_t peer,
1882 struct obd_uuid *uuid);
1883 int import_set_conn_priority(struct obd_import *imp, struct obd_uuid *uuid);
1884 void client_destroy_import(struct obd_import *imp);
1887 int server_disconnect_export(struct obd_export *exp);
1889 /* ptlrpc/pinger.c */
1891 * Pinger API (client side only)
1894 enum timeout_event {
1897 struct timeout_item;
1898 typedef int (*timeout_cb_t)(struct timeout_item *, void *);
1899 int ptlrpc_pinger_add_import(struct obd_import *imp);
1900 int ptlrpc_pinger_del_import(struct obd_import *imp);
1901 int ptlrpc_add_timeout_client(int time, enum timeout_event event,
1902 timeout_cb_t cb, void *data,
1903 cfs_list_t *obd_list);
1904 int ptlrpc_del_timeout_client(cfs_list_t *obd_list,
1905 enum timeout_event event);
1906 struct ptlrpc_request * ptlrpc_prep_ping(struct obd_import *imp);
1907 int ptlrpc_obd_ping(struct obd_device *obd);
1908 cfs_time_t ptlrpc_suspend_wakeup_time(void);
1910 void ping_evictor_start(void);
1911 void ping_evictor_stop(void);
1913 #define ping_evictor_start() do {} while (0)
1914 #define ping_evictor_stop() do {} while (0)
1916 int ptlrpc_check_and_wait_suspend(struct ptlrpc_request *req);
1919 /* ptlrpc daemon bind policy */
1921 /* all ptlrpcd threads are free mode */
1922 PDB_POLICY_NONE = 1,
1923 /* all ptlrpcd threads are bound mode */
1924 PDB_POLICY_FULL = 2,
1925 /* <free1 bound1> <free2 bound2> ... <freeN boundN> */
1926 PDB_POLICY_PAIR = 3,
1927 /* <free1 bound1> <bound1 free2> ... <freeN boundN> <boundN free1>,
1928 * means each ptlrpcd[X] has two partners: thread[X-1] and thread[X+1]*/
1929 PDB_POLICY_NEIGHBOR = 4,
1932 /* ptlrpc daemon load policy
1933 * It is caller's duty to specify how to push the async RPC into some ptlrpcd
1934 * queue, but it is not enforced, affected by "ptlrpcd_bind_policy". If it is
1935 * "PDB_POLICY_FULL", then the RPC will be processed by the selected ptlrpcd,
1936 * Otherwise, the RPC may be processed by the selected ptlrpcd or its partner,
1937 * depends on which is scheduled firstly, to accelerate the RPC processing. */
1939 /* on the same CPU core as the caller */
1940 PDL_POLICY_SAME = 1,
1941 /* within the same CPU partition, but not the same core as the caller */
1942 PDL_POLICY_LOCAL = 2,
1943 /* round-robin on all CPU cores, but not the same core as the caller */
1944 PDL_POLICY_ROUND = 3,
1945 /* the specified CPU core is preferred, but not enforced */
1946 PDL_POLICY_PREFERRED = 4,
1949 /* ptlrpc/ptlrpcd.c */
1950 void ptlrpcd_stop(struct ptlrpcd_ctl *pc, int force);
1951 void ptlrpcd_wake(struct ptlrpc_request *req);
1952 void ptlrpcd_add_req(struct ptlrpc_request *req, pdl_policy_t policy, int idx);
1953 void ptlrpcd_add_rqset(struct ptlrpc_request_set *set);
1954 int ptlrpcd_addref(void);
1955 void ptlrpcd_decref(void);
1957 /* ptlrpc/lproc_ptlrpc.c */
1959 * procfs output related functions
1962 const char* ll_opcode2str(__u32 opcode);
1964 void ptlrpc_lprocfs_register_obd(struct obd_device *obd);
1965 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd);
1966 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes);
1968 static inline void ptlrpc_lprocfs_register_obd(struct obd_device *obd) {}
1969 static inline void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd) {}
1970 static inline void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes) {}
1974 /* ptlrpc/llog_server.c */
1975 int llog_origin_handle_create(struct ptlrpc_request *req);
1976 int llog_origin_handle_destroy(struct ptlrpc_request *req);
1977 int llog_origin_handle_prev_block(struct ptlrpc_request *req);
1978 int llog_origin_handle_next_block(struct ptlrpc_request *req);
1979 int llog_origin_handle_read_header(struct ptlrpc_request *req);
1980 int llog_origin_handle_close(struct ptlrpc_request *req);
1981 int llog_origin_handle_cancel(struct ptlrpc_request *req);
1982 int llog_catinfo(struct ptlrpc_request *req);
1984 /* ptlrpc/llog_client.c */
1985 extern struct llog_operations llog_client_ops;