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 *);
452 * Called after the request has been handled.
454 void (*hpreq_fini)(struct ptlrpc_request *);
458 * Represents remote procedure call.
460 * This is a staple structure used by everybody wanting to send a request
463 struct ptlrpc_request {
464 /* Request type: one of PTL_RPC_MSG_* */
467 * Linkage item through which this request is included into
468 * sending/delayed lists on client and into rqbd list on server
472 * Server side list of incoming unserved requests sorted by arrival
473 * time. Traversed from time to time to notice about to expire
474 * requests and sent back "early replies" to clients to let them
475 * know server is alive and well, just very busy to service their
478 cfs_list_t rq_timed_list;
479 /** server-side history, used for debuging purposes. */
480 cfs_list_t rq_history_list;
481 /** server-side per-export list */
482 cfs_list_t rq_exp_list;
483 /** server-side hp handlers */
484 struct ptlrpc_hpreq_ops *rq_ops;
485 /** history sequence # */
486 __u64 rq_history_seq;
487 /** the index of service's srv_at_array into which request is linked */
489 /** Result of request processing */
491 /** Lock to protect request flags and some other important bits, like
494 cfs_spinlock_t rq_lock;
495 /** client-side flags are serialized by rq_lock */
496 unsigned long rq_intr:1, rq_replied:1, rq_err:1,
497 rq_timedout:1, rq_resend:1, rq_restart:1,
499 * when ->rq_replay is set, request is kept by the client even
500 * after server commits corresponding transaction. This is
501 * used for operations that require sequence of multiple
502 * requests to be replayed. The only example currently is file
503 * open/close. When last request in such a sequence is
504 * committed, ->rq_replay is cleared on all requests in the
508 rq_no_resend:1, rq_waiting:1, rq_receiving_reply:1,
509 rq_no_delay:1, rq_net_err:1, rq_wait_ctx:1,
510 rq_early:1, rq_must_unlink:1,
511 rq_fake:1, /* this fake req */
512 rq_memalloc:1, /* req originated from "kswapd" */
513 /* server-side flags */
514 rq_packed_final:1, /* packed final reply */
515 rq_hp:1, /* high priority RPC */
516 rq_at_linked:1, /* link into service's srv_at_array */
519 /* whether the "rq_set" is a valid one */
522 enum rq_phase rq_phase; /* one of RQ_PHASE_* */
523 enum rq_phase rq_next_phase; /* one of RQ_PHASE_* to be used next */
524 cfs_atomic_t rq_refcount;/* client-side refcount for SENT race,
525 server-side refcounf for multiple replies */
527 /** initial thread servicing this request */
528 struct ptlrpc_thread *rq_svc_thread;
530 /** Portal to which this request would be sent */
531 int rq_request_portal; /* XXX FIXME bug 249 */
532 /** Portal where to wait for reply and where reply would be sent */
533 int rq_reply_portal; /* XXX FIXME bug 249 */
537 * !rq_truncate : # reply bytes actually received,
538 * rq_truncate : required repbuf_len for resend
541 /** Request length */
543 /** Request message - what client sent */
544 struct lustre_msg *rq_reqmsg;
548 /** Reply message - server response */
549 struct lustre_msg *rq_repmsg;
550 /** Transaction number */
555 * List item to for replay list. Not yet commited requests get linked
557 * Also see \a rq_replay comment above.
559 cfs_list_t rq_replay_list;
562 * security and encryption data
564 struct ptlrpc_cli_ctx *rq_cli_ctx; /**< client's half ctx */
565 struct ptlrpc_svc_ctx *rq_svc_ctx; /**< server's half ctx */
566 cfs_list_t rq_ctx_chain; /**< link to waited ctx */
568 struct sptlrpc_flavor rq_flvr; /**< for client & server */
569 enum lustre_sec_part rq_sp_from;
571 unsigned long /* client/server security flags */
572 rq_ctx_init:1, /* context initiation */
573 rq_ctx_fini:1, /* context destroy */
574 rq_bulk_read:1, /* request bulk read */
575 rq_bulk_write:1, /* request bulk write */
576 /* server authentication flags */
577 rq_auth_gss:1, /* authenticated by gss */
578 rq_auth_remote:1, /* authed as remote user */
579 rq_auth_usr_root:1, /* authed as root */
580 rq_auth_usr_mdt:1, /* authed as mdt */
581 rq_auth_usr_ost:1, /* authed as ost */
582 /* security tfm flags */
585 /* doesn't expect reply FIXME */
587 rq_pill_init:1; /* pill initialized */
589 uid_t rq_auth_uid; /* authed uid */
590 uid_t rq_auth_mapped_uid; /* authed uid mapped to */
592 /* (server side), pointed directly into req buffer */
593 struct ptlrpc_user_desc *rq_user_desc;
595 /** early replies go to offset 0, regular replies go after that */
596 unsigned int rq_reply_off;
598 /* various buffer pointers */
599 struct lustre_msg *rq_reqbuf; /* req wrapper */
600 int rq_reqbuf_len; /* req wrapper buf len */
601 int rq_reqdata_len; /* req wrapper msg len */
602 char *rq_repbuf; /* rep buffer */
603 int rq_repbuf_len; /* rep buffer len */
604 struct lustre_msg *rq_repdata; /* rep wrapper msg */
605 int rq_repdata_len; /* rep wrapper msg len */
606 struct lustre_msg *rq_clrbuf; /* only in priv mode */
607 int rq_clrbuf_len; /* only in priv mode */
608 int rq_clrdata_len; /* only in priv mode */
612 /** Fields that help to see if request and reply were swabbed or not */
613 __u32 rq_req_swab_mask;
614 __u32 rq_rep_swab_mask;
616 /** What was import generation when this request was sent */
617 int rq_import_generation;
618 enum lustre_imp_state rq_send_state;
620 /** how many early replies (for stats) */
623 /** client+server request */
624 lnet_handle_md_t rq_req_md_h;
625 struct ptlrpc_cb_id rq_req_cbid;
626 /** optional time limit for send attempts */
627 cfs_duration_t rq_delay_limit;
628 /** time request was first queued */
629 cfs_time_t rq_queued_time;
632 /** request arrival time */
633 struct timeval rq_arrival_time;
634 /** separated reply state */
635 struct ptlrpc_reply_state *rq_reply_state;
636 /** incoming request buffer */
637 struct ptlrpc_request_buffer_desc *rq_rqbd;
639 __u32 rq_uid; /* peer uid, used in MDS only */
642 /** client-only incoming reply */
643 lnet_handle_md_t rq_reply_md_h;
644 cfs_waitq_t rq_reply_waitq;
645 struct ptlrpc_cb_id rq_reply_cbid;
649 /** Peer description (the other side) */
650 lnet_process_id_t rq_peer;
651 /** Server-side, export on which request was received */
652 struct obd_export *rq_export;
653 /** Client side, import where request is being sent */
654 struct obd_import *rq_import;
656 /** Replay callback, called after request is replayed at recovery */
657 void (*rq_replay_cb)(struct ptlrpc_request *);
659 * Commit callback, called when request is committed and about to be
662 void (*rq_commit_cb)(struct ptlrpc_request *);
663 /** Opaq data for replay and commit callbacks. */
666 /** For bulk requests on client only: bulk descriptor */
667 struct ptlrpc_bulk_desc *rq_bulk;
669 /** client outgoing req */
671 * when request/reply sent (secs), or time when request should be sent
674 /** time for request really sent out */
677 /** when request must finish. volatile
678 * so that servers' early reply updates to the deadline aren't
679 * kept in per-cpu cache */
680 volatile time_t rq_deadline;
681 /** when req reply unlink must finish. */
682 time_t rq_reply_deadline;
683 /** when req bulk unlink must finish. */
684 time_t rq_bulk_deadline;
686 * service time estimate (secs)
687 * If the requestsis not served by this time, it is marked as timed out.
691 /** Multi-rpc bits */
692 /** Link item for request set lists */
693 cfs_list_t rq_set_chain;
694 /** Per-request waitq introduced by bug 21938 for recovery waiting */
695 cfs_waitq_t rq_set_waitq;
696 /** Link back to the request set */
697 struct ptlrpc_request_set *rq_set;
698 /** Async completion handler, called when reply is received */
699 ptlrpc_interpterer_t rq_interpret_reply;
700 /** Async completion context */
701 union ptlrpc_async_args rq_async_args;
703 /** Pool if request is from preallocated list */
704 struct ptlrpc_request_pool *rq_pool;
706 struct lu_context rq_session;
707 struct lu_context rq_recov_session;
709 /** request format description */
710 struct req_capsule rq_pill;
714 * Call completion handler for rpc if any, return it's status or original
715 * rc if there was no handler defined for this request.
717 static inline int ptlrpc_req_interpret(const struct lu_env *env,
718 struct ptlrpc_request *req, int rc)
720 if (req->rq_interpret_reply != NULL) {
721 req->rq_status = req->rq_interpret_reply(env, req,
724 return req->rq_status;
730 * Returns 1 if request buffer at offset \a index was already swabbed
732 static inline int lustre_req_swabbed(struct ptlrpc_request *req, int index)
734 LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
735 return req->rq_req_swab_mask & (1 << index);
739 * Returns 1 if request reply buffer at offset \a index was already swabbed
741 static inline int lustre_rep_swabbed(struct ptlrpc_request *req, int index)
743 LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
744 return req->rq_rep_swab_mask & (1 << index);
748 * Returns 1 if request needs to be swabbed into local cpu byteorder
750 static inline int ptlrpc_req_need_swab(struct ptlrpc_request *req)
752 return lustre_req_swabbed(req, MSG_PTLRPC_HEADER_OFF);
756 * Returns 1 if request reply needs to be swabbed into local cpu byteorder
758 static inline int ptlrpc_rep_need_swab(struct ptlrpc_request *req)
760 return lustre_rep_swabbed(req, MSG_PTLRPC_HEADER_OFF);
764 * Mark request buffer at offset \a index that it was already swabbed
766 static inline void lustre_set_req_swabbed(struct ptlrpc_request *req, int index)
768 LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
769 LASSERT((req->rq_req_swab_mask & (1 << index)) == 0);
770 req->rq_req_swab_mask |= 1 << index;
774 * Mark request reply buffer at offset \a index that it was already swabbed
776 static inline void lustre_set_rep_swabbed(struct ptlrpc_request *req, int index)
778 LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
779 LASSERT((req->rq_rep_swab_mask & (1 << index)) == 0);
780 req->rq_rep_swab_mask |= 1 << index;
784 * Convert numerical request phase value \a phase into text string description
786 static inline const char *
787 ptlrpc_phase2str(enum rq_phase phase)
796 case RQ_PHASE_INTERPRET:
798 case RQ_PHASE_COMPLETE:
800 case RQ_PHASE_UNREGISTERING:
801 return "Unregistering";
808 * Convert numerical request phase of the request \a req into text stringi
811 static inline const char *
812 ptlrpc_rqphase2str(struct ptlrpc_request *req)
814 return ptlrpc_phase2str(req->rq_phase);
818 * Debugging functions and helpers to print request structure into debug log
821 /* Spare the preprocessor, spoil the bugs. */
822 #define FLAG(field, str) (field ? str : "")
824 /** Convert bit flags into a string */
825 #define DEBUG_REQ_FLAGS(req) \
826 ptlrpc_rqphase2str(req), \
827 FLAG(req->rq_intr, "I"), FLAG(req->rq_replied, "R"), \
828 FLAG(req->rq_err, "E"), \
829 FLAG(req->rq_timedout, "X") /* eXpired */, FLAG(req->rq_resend, "S"), \
830 FLAG(req->rq_restart, "T"), FLAG(req->rq_replay, "P"), \
831 FLAG(req->rq_no_resend, "N"), \
832 FLAG(req->rq_waiting, "W"), \
833 FLAG(req->rq_wait_ctx, "C"), FLAG(req->rq_hp, "H"), \
834 FLAG(req->rq_committed, "M")
836 #define REQ_FLAGS_FMT "%s:%s%s%s%s%s%s%s%s%s%s%s%s"
838 void _debug_req(struct ptlrpc_request *req, __u32 mask,
839 struct libcfs_debug_msg_data *data, const char *fmt, ...)
840 __attribute__ ((format (printf, 4, 5)));
843 * Helper that decides if we need to print request accordig to current debug
846 #define debug_req(cdls, level, req, file, func, line, fmt, a...) \
850 if (((level) & D_CANTMASK) != 0 || \
851 ((libcfs_debug & (level)) != 0 && \
852 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) { \
853 static struct libcfs_debug_msg_data _req_dbg_data = \
854 DEBUG_MSG_DATA_INIT(cdls, DEBUG_SUBSYSTEM, file, func, line); \
855 _debug_req((req), (level), &_req_dbg_data, fmt, ##a); \
860 * This is the debug print function you need to use to print request sturucture
861 * content into lustre debug log.
862 * for most callers (level is a constant) this is resolved at compile time */
863 #define DEBUG_REQ(level, req, fmt, args...) \
865 if ((level) & (D_ERROR | D_WARNING)) { \
866 static cfs_debug_limit_state_t cdls; \
867 debug_req(&cdls, level, req, __FILE__, __func__, __LINE__, \
868 "@@@ "fmt" ", ## args); \
870 debug_req(NULL, level, req, __FILE__, __func__, __LINE__, \
871 "@@@ "fmt" ", ## args); \
876 * Structure that defines a single page of a bulk transfer
878 struct ptlrpc_bulk_page {
879 /** Linkage to list of pages in a bulk */
882 * Number of bytes in a page to transfer starting from \a bp_pageoffset
885 /** offset within a page */
887 /** The page itself */
888 struct page *bp_page;
891 #define BULK_GET_SOURCE 0
892 #define BULK_PUT_SINK 1
893 #define BULK_GET_SINK 2
894 #define BULK_PUT_SOURCE 3
897 * Definition of buk descriptor.
898 * Bulks are special "Two phase" RPCs where initial request message
899 * is sent first and it is followed bt a transfer (o receiving) of a large
900 * amount of data to be settled into pages referenced from the bulk descriptors.
901 * Bulks transfers (the actual data following the small requests) are done
902 * on separate LNet portals.
903 * In lustre we use bulk transfers for READ and WRITE transfers from/to OSTs.
904 * Another user is readpage for MDT.
906 struct ptlrpc_bulk_desc {
907 /** completed successfully */
908 unsigned long bd_success:1;
909 /** accessible to the network (network io potentially in progress) */
910 unsigned long bd_network_rw:1;
911 /** {put,get}{source,sink} */
912 unsigned long bd_type:2;
914 unsigned long bd_registered:1;
915 /** For serialization with callback */
916 cfs_spinlock_t bd_lock;
917 /** Import generation when request for this bulk was sent */
918 int bd_import_generation;
919 /** Server side - export this bulk created for */
920 struct obd_export *bd_export;
921 /** Client side - import this bulk was sent on */
922 struct obd_import *bd_import;
923 /** LNet portal for this bulk */
925 /** Back pointer to the request */
926 struct ptlrpc_request *bd_req;
927 cfs_waitq_t bd_waitq; /* server side only WQ */
928 int bd_iov_count; /* # entries in bd_iov */
929 int bd_max_iov; /* allocated size of bd_iov */
930 int bd_nob; /* # bytes covered */
931 int bd_nob_transferred; /* # bytes GOT/PUT */
935 struct ptlrpc_cb_id bd_cbid; /* network callback info */
936 lnet_handle_md_t bd_md_h; /* associated MD */
937 lnet_nid_t bd_sender; /* stash event::sender */
939 #if defined(__KERNEL__)
941 * encrypt iov, size is either 0 or bd_iov_count.
943 lnet_kiov_t *bd_enc_iov;
945 lnet_kiov_t bd_iov[0];
947 lnet_md_iovec_t bd_iov[0];
952 SVC_STOPPED = 1 << 0,
953 SVC_STOPPING = 1 << 1,
954 SVC_STARTING = 1 << 2,
955 SVC_RUNNING = 1 << 3,
961 * Definition of server service thread structure
963 struct ptlrpc_thread {
965 * List of active threads in svc->srv_threads
969 * thread-private data (preallocated memory)
974 * service thread index, from ptlrpc_start_threads
982 * put watchdog in the structure per thread b=14840
984 struct lc_watchdog *t_watchdog;
986 * the svc this thread belonged to b=18582
988 struct ptlrpc_service *t_svc;
989 cfs_waitq_t t_ctl_waitq;
990 struct lu_env *t_env;
993 static inline int thread_is_stopped(struct ptlrpc_thread *thread)
995 return !!(thread->t_flags & SVC_STOPPED);
998 static inline int thread_is_stopping(struct ptlrpc_thread *thread)
1000 return !!(thread->t_flags & SVC_STOPPING);
1003 static inline int thread_is_starting(struct ptlrpc_thread *thread)
1005 return !!(thread->t_flags & SVC_STARTING);
1008 static inline int thread_is_running(struct ptlrpc_thread *thread)
1010 return !!(thread->t_flags & SVC_RUNNING);
1013 static inline int thread_is_event(struct ptlrpc_thread *thread)
1015 return !!(thread->t_flags & SVC_EVENT);
1018 static inline int thread_is_signal(struct ptlrpc_thread *thread)
1020 return !!(thread->t_flags & SVC_SIGNAL);
1023 static inline void thread_clear_flags(struct ptlrpc_thread *thread, __u32 flags)
1025 thread->t_flags &= ~flags;
1028 static inline void thread_set_flags(struct ptlrpc_thread *thread, __u32 flags)
1030 thread->t_flags = flags;
1033 static inline void thread_add_flags(struct ptlrpc_thread *thread, __u32 flags)
1035 thread->t_flags |= flags;
1038 static inline int thread_test_and_clear_flags(struct ptlrpc_thread *thread,
1041 if (thread->t_flags & flags) {
1042 thread->t_flags &= ~flags;
1049 * Request buffer descriptor structure.
1050 * This is a structure that contains one posted request buffer for service.
1051 * Once data land into a buffer, event callback creates actual request and
1052 * notifies wakes one of the service threads to process new incoming request.
1053 * More than one request can fit into the buffer.
1055 struct ptlrpc_request_buffer_desc {
1056 /** Link item for rqbds on a service */
1057 cfs_list_t rqbd_list;
1058 /** History of requests for this buffer */
1059 cfs_list_t rqbd_reqs;
1060 /** Back pointer to service for which this buffer is registered */
1061 struct ptlrpc_service *rqbd_service;
1062 /** LNet descriptor */
1063 lnet_handle_md_t rqbd_md_h;
1065 /** The buffer itself */
1067 struct ptlrpc_cb_id rqbd_cbid;
1069 * This "embedded" request structure is only used for the
1070 * last request to fit into the buffer
1072 struct ptlrpc_request rqbd_req;
1075 typedef int (*svc_thr_init_t)(struct ptlrpc_thread *thread);
1076 typedef void (*svc_thr_done_t)(struct ptlrpc_thread *thread);
1077 typedef int (*svc_handler_t)(struct ptlrpc_request *req);
1078 typedef int (*svc_hpreq_handler_t)(struct ptlrpc_request *);
1079 typedef void (*svc_req_printfn_t)(void *, struct ptlrpc_request *);
1081 #ifndef __cfs_cacheline_aligned
1082 /* NB: put it here for reducing patche dependence */
1083 # define __cfs_cacheline_aligned
1087 * How many high priority requests to serve before serving one normal
1090 #define PTLRPC_SVC_HP_RATIO 10
1093 * Definition of PortalRPC service.
1094 * The service is listening on a particular portal (like tcp port)
1095 * and perform actions for a specific server like IO service for OST
1096 * or general metadata service for MDS.
1098 * ptlrpc service has four locks:
1100 * serialize operations on rqbd and requests waiting for preprocess
1102 * serialize operations active requests sent to this portal
1104 * serialize adaptive timeout stuff
1106 * serialize operations on RS list (reply states)
1108 * We don't have any use-case to take two or more locks at the same time
1109 * for now, so there is no lock order issue.
1111 struct ptlrpc_service {
1112 /** most often accessed fields */
1113 /** chain thru all services */
1114 cfs_list_t srv_list;
1115 /** only statically allocated strings here; we don't clean them */
1117 /** only statically allocated strings here; we don't clean them */
1118 char *srv_thread_name;
1119 /** service thread list */
1120 cfs_list_t srv_threads;
1121 /** threads to start at beginning of service */
1122 int srv_threads_min;
1123 /** thread upper limit */
1124 int srv_threads_max;
1125 /** always increasing number */
1126 unsigned srv_threads_next_id;
1127 /** # of starting threads */
1128 int srv_threads_starting;
1129 /** # running threads */
1130 int srv_threads_running;
1132 /** service operations, move to ptlrpc_svc_ops_t in the future */
1135 * if non-NULL called during thread creation (ptlrpc_start_thread())
1136 * to initialize service specific per-thread state.
1138 svc_thr_init_t srv_init;
1140 * if non-NULL called during thread shutdown (ptlrpc_main()) to
1141 * destruct state created by ->srv_init().
1143 svc_thr_done_t srv_done;
1144 /** Handler function for incoming requests for this service */
1145 svc_handler_t srv_handler;
1146 /** hp request handler */
1147 svc_hpreq_handler_t srv_hpreq_handler;
1148 /** service-specific print fn */
1149 svc_req_printfn_t srv_req_printfn;
1152 /** Root of /proc dir tree for this service */
1153 cfs_proc_dir_entry_t *srv_procroot;
1154 /** Pointer to statistic data for this service */
1155 struct lprocfs_stats *srv_stats;
1156 /** # hp per lp reqs to handle */
1157 int srv_hpreq_ratio;
1158 /** biggest request to receive */
1159 int srv_max_req_size;
1160 /** biggest reply to send */
1161 int srv_max_reply_size;
1162 /** size of individual buffers */
1164 /** # buffers to allocate in 1 group */
1165 int srv_nbuf_per_group;
1166 /** Local portal on which to receive requests */
1167 __u32 srv_req_portal;
1168 /** Portal on the client to send replies to */
1169 __u32 srv_rep_portal;
1171 * Tags for lu_context associated with this thread, see struct
1175 /** soft watchdog timeout multiplier */
1176 int srv_watchdog_factor;
1177 /** bind threads to CPUs */
1178 unsigned srv_cpu_affinity:1;
1179 /** under unregister_service */
1180 unsigned srv_is_stopping:1;
1183 * serialize the following fields, used for protecting
1184 * rqbd list and incoming requests waiting for preprocess
1186 cfs_spinlock_t srv_lock __cfs_cacheline_aligned;
1187 /** incoming reqs */
1188 cfs_list_t srv_req_in_queue;
1189 /** total # req buffer descs allocated */
1191 /** # posted request buffers */
1192 int srv_nrqbd_receiving;
1193 /** timeout before re-posting reqs, in tick */
1194 cfs_duration_t srv_rqbd_timeout;
1195 /** request buffers to be reposted */
1196 cfs_list_t srv_idle_rqbds;
1197 /** req buffers receiving */
1198 cfs_list_t srv_active_rqbds;
1199 /** request buffer history */
1200 cfs_list_t srv_history_rqbds;
1201 /** # request buffers in history */
1202 int srv_n_history_rqbds;
1203 /** max # request buffers in history */
1204 int srv_max_history_rqbds;
1205 /** request history */
1206 cfs_list_t srv_request_history;
1207 /** next request sequence # */
1208 __u64 srv_request_seq;
1209 /** highest seq culled from history */
1210 __u64 srv_request_max_cull_seq;
1212 * all threads sleep on this. This wait-queue is signalled when new
1213 * incoming request arrives and when difficult reply has to be handled.
1215 cfs_waitq_t srv_waitq;
1218 * serialize the following fields, used for processing requests
1219 * sent to this portal
1221 cfs_spinlock_t srv_rq_lock __cfs_cacheline_aligned;
1222 /** # reqs in either of the queues below */
1223 /** reqs waiting for service */
1224 cfs_list_t srv_request_queue;
1225 /** high priority queue */
1226 cfs_list_t srv_request_hpq;
1227 /** # incoming reqs */
1228 int srv_n_queued_reqs;
1229 /** # reqs being served */
1230 int srv_n_active_reqs;
1231 /** # HPreqs being served */
1232 int srv_n_active_hpreq;
1233 /** # hp requests handled */
1234 int srv_hpreq_count;
1239 * serialize the following fields, used for changes on
1242 cfs_spinlock_t srv_at_lock __cfs_cacheline_aligned;
1243 /** estimated rpc service time */
1244 struct adaptive_timeout srv_at_estimate;
1245 /** reqs waiting for replies */
1246 struct ptlrpc_at_array srv_at_array;
1247 /** early reply timer */
1248 cfs_timer_t srv_at_timer;
1249 /** check early replies */
1250 unsigned srv_at_check;
1252 cfs_time_t srv_at_checktime;
1256 * serialize the following fields, used for processing
1257 * replies for this portal
1259 cfs_spinlock_t srv_rs_lock __cfs_cacheline_aligned;
1260 /** all the active replies */
1261 cfs_list_t srv_active_replies;
1263 /** replies waiting for service */
1264 cfs_list_t srv_reply_queue;
1266 /** List of free reply_states */
1267 cfs_list_t srv_free_rs_list;
1268 /** waitq to run, when adding stuff to srv_free_rs_list */
1269 cfs_waitq_t srv_free_rs_waitq;
1270 /** # 'difficult' replies */
1271 cfs_atomic_t srv_n_difficult_replies;
1272 //struct ptlrpc_srv_ni srv_interfaces[0];
1276 * Declaration of ptlrpcd control structure
1278 struct ptlrpcd_ctl {
1280 * Ptlrpc thread control flags (LIOD_START, LIOD_STOP, LIOD_FORCE)
1282 unsigned long pc_flags;
1284 * Thread lock protecting structure fields.
1286 cfs_spinlock_t pc_lock;
1290 cfs_completion_t pc_starting;
1294 cfs_completion_t pc_finishing;
1296 * Thread requests set.
1298 struct ptlrpc_request_set *pc_set;
1300 * Thread name used in cfs_daemonize()
1304 * Environment for request interpreters to run in.
1306 struct lu_env pc_env;
1308 * Index of ptlrpcd thread in the array.
1312 * Number of the ptlrpcd's partners.
1316 * Pointer to the array of partners' ptlrpcd_ctl structure.
1318 struct ptlrpcd_ctl **pc_partners;
1320 * Record the partner index to be processed next.
1325 * Async rpcs flag to make sure that ptlrpcd_check() is called only
1330 * Currently not used.
1334 * User-space async rpcs callback.
1336 void *pc_wait_callback;
1338 * User-space check idle rpcs callback.
1340 void *pc_idle_callback;
1344 /* Bits for pc_flags */
1345 enum ptlrpcd_ctl_flags {
1347 * Ptlrpc thread start flag.
1349 LIOD_START = 1 << 0,
1351 * Ptlrpc thread stop flag.
1355 * Ptlrpc thread force flag (only stop force so far).
1356 * This will cause aborting any inflight rpcs handled
1357 * by thread if LIOD_STOP is specified.
1359 LIOD_FORCE = 1 << 2,
1361 * This is a recovery ptlrpc thread.
1363 LIOD_RECOVERY = 1 << 3,
1365 * The ptlrpcd is bound to some CPU core.
1370 /* ptlrpc/events.c */
1371 extern lnet_handle_eq_t ptlrpc_eq_h;
1372 extern int ptlrpc_uuid_to_peer(struct obd_uuid *uuid,
1373 lnet_process_id_t *peer, lnet_nid_t *self);
1375 * These callbacks are invoked by LNet when something happened to
1379 extern void request_out_callback (lnet_event_t *ev);
1380 extern void reply_in_callback(lnet_event_t *ev);
1381 extern void client_bulk_callback (lnet_event_t *ev);
1382 extern void request_in_callback(lnet_event_t *ev);
1383 extern void reply_out_callback(lnet_event_t *ev);
1384 extern void server_bulk_callback (lnet_event_t *ev);
1387 /* ptlrpc/connection.c */
1388 struct ptlrpc_connection *ptlrpc_connection_get(lnet_process_id_t peer,
1390 struct obd_uuid *uuid);
1391 int ptlrpc_connection_put(struct ptlrpc_connection *c);
1392 struct ptlrpc_connection *ptlrpc_connection_addref(struct ptlrpc_connection *);
1393 int ptlrpc_connection_init(void);
1394 void ptlrpc_connection_fini(void);
1395 extern lnet_pid_t ptl_get_pid(void);
1397 /* ptlrpc/niobuf.c */
1399 * Actual interfacing with LNet to put/get/register/unregister stuff
1402 int ptlrpc_start_bulk_transfer(struct ptlrpc_bulk_desc *desc);
1403 void ptlrpc_abort_bulk(struct ptlrpc_bulk_desc *desc);
1404 int ptlrpc_register_bulk(struct ptlrpc_request *req);
1405 int ptlrpc_unregister_bulk(struct ptlrpc_request *req, int async);
1407 static inline int ptlrpc_server_bulk_active(struct ptlrpc_bulk_desc *desc)
1411 LASSERT(desc != NULL);
1413 cfs_spin_lock(&desc->bd_lock);
1414 rc = desc->bd_network_rw;
1415 cfs_spin_unlock(&desc->bd_lock);
1419 static inline int ptlrpc_client_bulk_active(struct ptlrpc_request *req)
1421 struct ptlrpc_bulk_desc *desc = req->rq_bulk;
1424 LASSERT(req != NULL);
1426 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_BULK_UNLINK) &&
1427 req->rq_bulk_deadline > cfs_time_current_sec())
1433 cfs_spin_lock(&desc->bd_lock);
1434 rc = desc->bd_network_rw;
1435 cfs_spin_unlock(&desc->bd_lock);
1439 #define PTLRPC_REPLY_MAYBE_DIFFICULT 0x01
1440 #define PTLRPC_REPLY_EARLY 0x02
1441 int ptlrpc_send_reply(struct ptlrpc_request *req, int flags);
1442 int ptlrpc_reply(struct ptlrpc_request *req);
1443 int ptlrpc_send_error(struct ptlrpc_request *req, int difficult);
1444 int ptlrpc_error(struct ptlrpc_request *req);
1445 void ptlrpc_resend_req(struct ptlrpc_request *request);
1446 int ptlrpc_at_get_net_latency(struct ptlrpc_request *req);
1447 int ptl_send_rpc(struct ptlrpc_request *request, int noreply);
1448 int ptlrpc_register_rqbd (struct ptlrpc_request_buffer_desc *rqbd);
1451 /* ptlrpc/client.c */
1453 * Client-side portals API. Everything to send requests, receive replies,
1454 * request queues, request management, etc.
1457 void ptlrpc_init_client(int req_portal, int rep_portal, char *name,
1458 struct ptlrpc_client *);
1459 void ptlrpc_cleanup_client(struct obd_import *imp);
1460 struct ptlrpc_connection *ptlrpc_uuid_to_connection(struct obd_uuid *uuid);
1462 int ptlrpc_queue_wait(struct ptlrpc_request *req);
1463 int ptlrpc_replay_req(struct ptlrpc_request *req);
1464 int ptlrpc_unregister_reply(struct ptlrpc_request *req, int async);
1465 void ptlrpc_restart_req(struct ptlrpc_request *req);
1466 void ptlrpc_abort_inflight(struct obd_import *imp);
1467 void ptlrpc_cleanup_imp(struct obd_import *imp);
1468 void ptlrpc_abort_set(struct ptlrpc_request_set *set);
1470 struct ptlrpc_request_set *ptlrpc_prep_set(void);
1471 int ptlrpc_set_add_cb(struct ptlrpc_request_set *set,
1472 set_interpreter_func fn, void *data);
1473 int ptlrpc_set_next_timeout(struct ptlrpc_request_set *);
1474 int ptlrpc_check_set(const struct lu_env *env, struct ptlrpc_request_set *set);
1475 int ptlrpc_set_wait(struct ptlrpc_request_set *);
1476 int ptlrpc_expired_set(void *data);
1477 void ptlrpc_interrupted_set(void *data);
1478 void ptlrpc_mark_interrupted(struct ptlrpc_request *req);
1479 void ptlrpc_set_destroy(struct ptlrpc_request_set *);
1480 void ptlrpc_set_add_req(struct ptlrpc_request_set *, struct ptlrpc_request *);
1481 void ptlrpc_set_add_new_req(struct ptlrpcd_ctl *pc,
1482 struct ptlrpc_request *req);
1484 void ptlrpc_free_rq_pool(struct ptlrpc_request_pool *pool);
1485 void ptlrpc_add_rqs_to_pool(struct ptlrpc_request_pool *pool, int num_rq);
1487 struct ptlrpc_request_pool *
1488 ptlrpc_init_rq_pool(int, int,
1489 void (*populate_pool)(struct ptlrpc_request_pool *, int));
1491 void ptlrpc_at_set_req_timeout(struct ptlrpc_request *req);
1492 struct ptlrpc_request *ptlrpc_request_alloc(struct obd_import *imp,
1493 const struct req_format *format);
1494 struct ptlrpc_request *ptlrpc_request_alloc_pool(struct obd_import *imp,
1495 struct ptlrpc_request_pool *,
1496 const struct req_format *format);
1497 void ptlrpc_request_free(struct ptlrpc_request *request);
1498 int ptlrpc_request_pack(struct ptlrpc_request *request,
1499 __u32 version, int opcode);
1500 struct ptlrpc_request *ptlrpc_request_alloc_pack(struct obd_import *imp,
1501 const struct req_format *format,
1502 __u32 version, int opcode);
1503 int ptlrpc_request_bufs_pack(struct ptlrpc_request *request,
1504 __u32 version, int opcode, char **bufs,
1505 struct ptlrpc_cli_ctx *ctx);
1506 struct ptlrpc_request *ptlrpc_prep_fakereq(struct obd_import *imp,
1507 unsigned int timeout,
1508 ptlrpc_interpterer_t interpreter);
1509 void ptlrpc_fakereq_finished(struct ptlrpc_request *req);
1511 struct ptlrpc_request *ptlrpc_prep_req(struct obd_import *imp, __u32 version,
1512 int opcode, int count, __u32 *lengths,
1514 struct ptlrpc_request *ptlrpc_prep_req_pool(struct obd_import *imp,
1515 __u32 version, int opcode,
1516 int count, __u32 *lengths, char **bufs,
1517 struct ptlrpc_request_pool *pool);
1518 void ptlrpc_req_finished(struct ptlrpc_request *request);
1519 void ptlrpc_req_finished_with_imp_lock(struct ptlrpc_request *request);
1520 struct ptlrpc_request *ptlrpc_request_addref(struct ptlrpc_request *req);
1521 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_imp (struct ptlrpc_request *req,
1522 int npages, int type, int portal);
1523 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_exp(struct ptlrpc_request *req,
1524 int npages, int type, int portal);
1525 void ptlrpc_free_bulk(struct ptlrpc_bulk_desc *bulk);
1526 void ptlrpc_prep_bulk_page(struct ptlrpc_bulk_desc *desc,
1527 cfs_page_t *page, int pageoffset, int len);
1528 void ptlrpc_retain_replayable_request(struct ptlrpc_request *req,
1529 struct obd_import *imp);
1530 __u64 ptlrpc_next_xid(void);
1531 __u64 ptlrpc_sample_next_xid(void);
1532 __u64 ptlrpc_req_xid(struct ptlrpc_request *request);
1534 /* Set of routines to run a function in ptlrpcd context */
1535 void *ptlrpcd_alloc_work(struct obd_import *imp,
1536 int (*cb)(const struct lu_env *, void *), void *data);
1537 void ptlrpcd_destroy_work(void *handler);
1538 int ptlrpcd_queue_work(void *handler);
1542 struct ptlrpc_service_conf {
1545 int psc_max_req_size;
1546 int psc_max_reply_size;
1549 int psc_watchdog_factor;
1550 int psc_min_threads;
1551 int psc_max_threads;
1555 /* ptlrpc/service.c */
1557 * Server-side services API. Register/unregister service, request state
1558 * management, service thread management
1562 void ptlrpc_save_lock (struct ptlrpc_request *req,
1563 struct lustre_handle *lock, int mode, int no_ack);
1564 void ptlrpc_commit_replies(struct obd_export *exp);
1565 void ptlrpc_dispatch_difficult_reply (struct ptlrpc_reply_state *rs);
1566 void ptlrpc_schedule_difficult_reply (struct ptlrpc_reply_state *rs);
1567 struct ptlrpc_service *ptlrpc_init_svc_conf(struct ptlrpc_service_conf *c,
1568 svc_handler_t h, char *name,
1569 struct proc_dir_entry *proc_entry,
1570 svc_req_printfn_t prntfn,
1573 struct ptlrpc_service *ptlrpc_init_svc(int nbufs, int bufsize, int max_req_size,
1575 int req_portal, int rep_portal,
1576 int watchdog_factor,
1577 svc_handler_t, char *name,
1578 cfs_proc_dir_entry_t *proc_entry,
1580 int min_threads, int max_threads,
1581 char *threadname, __u32 ctx_tags,
1582 svc_hpreq_handler_t);
1583 void ptlrpc_stop_all_threads(struct ptlrpc_service *svc);
1585 int ptlrpc_start_threads(struct ptlrpc_service *svc);
1586 int ptlrpc_start_thread(struct ptlrpc_service *svc);
1587 int ptlrpc_unregister_service(struct ptlrpc_service *service);
1588 int liblustre_check_services (void *arg);
1589 void ptlrpc_daemonize(char *name);
1590 int ptlrpc_service_health_check(struct ptlrpc_service *);
1591 void ptlrpc_hpreq_reorder(struct ptlrpc_request *req);
1592 void ptlrpc_server_drop_request(struct ptlrpc_request *req);
1595 int ptlrpc_hr_init(void);
1596 void ptlrpc_hr_fini(void);
1598 # define ptlrpc_hr_init() (0)
1599 # define ptlrpc_hr_fini() do {} while(0)
1602 struct ptlrpc_svc_data {
1604 struct ptlrpc_service *svc;
1605 struct ptlrpc_thread *thread;
1609 /* ptlrpc/import.c */
1614 int ptlrpc_connect_import(struct obd_import *imp);
1615 int ptlrpc_init_import(struct obd_import *imp);
1616 int ptlrpc_disconnect_import(struct obd_import *imp, int noclose);
1617 int ptlrpc_import_recovery_state_machine(struct obd_import *imp);
1619 /* ptlrpc/pack_generic.c */
1620 int ptlrpc_reconnect_import(struct obd_import *imp);
1624 * ptlrpc msg buffer and swab interface
1628 int ptlrpc_buf_need_swab(struct ptlrpc_request *req, const int inout,
1630 void ptlrpc_buf_set_swabbed(struct ptlrpc_request *req, const int inout,
1632 int ptlrpc_unpack_rep_msg(struct ptlrpc_request *req, int len);
1633 int ptlrpc_unpack_req_msg(struct ptlrpc_request *req, int len);
1635 int lustre_msg_check_version(struct lustre_msg *msg, __u32 version);
1636 void lustre_init_msg_v2(struct lustre_msg_v2 *msg, int count, __u32 *lens,
1638 int lustre_pack_request(struct ptlrpc_request *, __u32 magic, int count,
1639 __u32 *lens, char **bufs);
1640 int lustre_pack_reply(struct ptlrpc_request *, int count, __u32 *lens,
1642 int lustre_pack_reply_v2(struct ptlrpc_request *req, int count,
1643 __u32 *lens, char **bufs, int flags);
1644 #define LPRFL_EARLY_REPLY 1
1645 int lustre_pack_reply_flags(struct ptlrpc_request *, int count, __u32 *lens,
1646 char **bufs, int flags);
1647 int lustre_shrink_msg(struct lustre_msg *msg, int segment,
1648 unsigned int newlen, int move_data);
1649 void lustre_free_reply_state(struct ptlrpc_reply_state *rs);
1650 int __lustre_unpack_msg(struct lustre_msg *m, int len);
1651 int lustre_msg_hdr_size(__u32 magic, int count);
1652 int lustre_msg_size(__u32 magic, int count, __u32 *lengths);
1653 int lustre_msg_size_v2(int count, __u32 *lengths);
1654 int lustre_packed_msg_size(struct lustre_msg *msg);
1655 int lustre_msg_early_size(void);
1656 void *lustre_msg_buf_v2(struct lustre_msg_v2 *m, int n, int min_size);
1657 void *lustre_msg_buf(struct lustre_msg *m, int n, int minlen);
1658 int lustre_msg_buflen(struct lustre_msg *m, int n);
1659 void lustre_msg_set_buflen(struct lustre_msg *m, int n, int len);
1660 int lustre_msg_bufcount(struct lustre_msg *m);
1661 char *lustre_msg_string (struct lustre_msg *m, int n, int max_len);
1662 __u32 lustre_msghdr_get_flags(struct lustre_msg *msg);
1663 void lustre_msghdr_set_flags(struct lustre_msg *msg, __u32 flags);
1664 __u32 lustre_msg_get_flags(struct lustre_msg *msg);
1665 void lustre_msg_add_flags(struct lustre_msg *msg, int flags);
1666 void lustre_msg_set_flags(struct lustre_msg *msg, int flags);
1667 void lustre_msg_clear_flags(struct lustre_msg *msg, int flags);
1668 __u32 lustre_msg_get_op_flags(struct lustre_msg *msg);
1669 void lustre_msg_add_op_flags(struct lustre_msg *msg, int flags);
1670 void lustre_msg_set_op_flags(struct lustre_msg *msg, int flags);
1671 struct lustre_handle *lustre_msg_get_handle(struct lustre_msg *msg);
1672 __u32 lustre_msg_get_type(struct lustre_msg *msg);
1673 __u32 lustre_msg_get_version(struct lustre_msg *msg);
1674 void lustre_msg_add_version(struct lustre_msg *msg, int version);
1675 __u32 lustre_msg_get_opc(struct lustre_msg *msg);
1676 __u64 lustre_msg_get_last_xid(struct lustre_msg *msg);
1677 __u64 lustre_msg_get_last_committed(struct lustre_msg *msg);
1678 __u64 *lustre_msg_get_versions(struct lustre_msg *msg);
1679 __u64 lustre_msg_get_transno(struct lustre_msg *msg);
1680 __u64 lustre_msg_get_slv(struct lustre_msg *msg);
1681 __u32 lustre_msg_get_limit(struct lustre_msg *msg);
1682 void lustre_msg_set_slv(struct lustre_msg *msg, __u64 slv);
1683 void lustre_msg_set_limit(struct lustre_msg *msg, __u64 limit);
1684 int lustre_msg_get_status(struct lustre_msg *msg);
1685 __u32 lustre_msg_get_conn_cnt(struct lustre_msg *msg);
1686 int lustre_msg_is_v1(struct lustre_msg *msg);
1687 __u32 lustre_msg_get_magic(struct lustre_msg *msg);
1688 __u32 lustre_msg_get_timeout(struct lustre_msg *msg);
1689 __u32 lustre_msg_get_service_time(struct lustre_msg *msg);
1690 __u32 lustre_msg_get_cksum(struct lustre_msg *msg);
1691 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 0, 0)
1692 __u32 lustre_msg_calc_cksum(struct lustre_msg *msg, int compat18);
1694 # warning "remove checksum compatibility support for b1_8"
1695 __u32 lustre_msg_calc_cksum(struct lustre_msg *msg);
1697 void lustre_msg_set_handle(struct lustre_msg *msg,struct lustre_handle *handle);
1698 void lustre_msg_set_type(struct lustre_msg *msg, __u32 type);
1699 void lustre_msg_set_opc(struct lustre_msg *msg, __u32 opc);
1700 void lustre_msg_set_last_xid(struct lustre_msg *msg, __u64 last_xid);
1701 void lustre_msg_set_last_committed(struct lustre_msg *msg,__u64 last_committed);
1702 void lustre_msg_set_versions(struct lustre_msg *msg, __u64 *versions);
1703 void lustre_msg_set_transno(struct lustre_msg *msg, __u64 transno);
1704 void lustre_msg_set_status(struct lustre_msg *msg, __u32 status);
1705 void lustre_msg_set_conn_cnt(struct lustre_msg *msg, __u32 conn_cnt);
1706 void ptlrpc_req_set_repsize(struct ptlrpc_request *req, int count, __u32 *sizes);
1707 void ptlrpc_request_set_replen(struct ptlrpc_request *req);
1708 void lustre_msg_set_timeout(struct lustre_msg *msg, __u32 timeout);
1709 void lustre_msg_set_service_time(struct lustre_msg *msg, __u32 service_time);
1710 void lustre_msg_set_cksum(struct lustre_msg *msg, __u32 cksum);
1713 lustre_shrink_reply(struct ptlrpc_request *req, int segment,
1714 unsigned int newlen, int move_data)
1716 LASSERT(req->rq_reply_state);
1717 LASSERT(req->rq_repmsg);
1718 req->rq_replen = lustre_shrink_msg(req->rq_repmsg, segment,
1723 /** Change request phase of \a req to \a new_phase */
1725 ptlrpc_rqphase_move(struct ptlrpc_request *req, enum rq_phase new_phase)
1727 if (req->rq_phase == new_phase)
1730 if (new_phase == RQ_PHASE_UNREGISTERING) {
1731 req->rq_next_phase = req->rq_phase;
1733 cfs_atomic_inc(&req->rq_import->imp_unregistering);
1736 if (req->rq_phase == RQ_PHASE_UNREGISTERING) {
1738 cfs_atomic_dec(&req->rq_import->imp_unregistering);
1741 DEBUG_REQ(D_INFO, req, "move req \"%s\" -> \"%s\"",
1742 ptlrpc_rqphase2str(req), ptlrpc_phase2str(new_phase));
1744 req->rq_phase = new_phase;
1748 * Returns true if request \a req got early reply and hard deadline is not met
1751 ptlrpc_client_early(struct ptlrpc_request *req)
1753 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
1754 req->rq_reply_deadline > cfs_time_current_sec())
1756 return req->rq_early;
1760 * Returns true if we got real reply from server for this request
1763 ptlrpc_client_replied(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_replied;
1771 /** Returns true if request \a req is in process of receiving server reply */
1773 ptlrpc_client_recv(struct ptlrpc_request *req)
1775 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
1776 req->rq_reply_deadline > cfs_time_current_sec())
1778 return req->rq_receiving_reply;
1782 ptlrpc_client_recv_or_unlink(struct ptlrpc_request *req)
1786 cfs_spin_lock(&req->rq_lock);
1787 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
1788 req->rq_reply_deadline > cfs_time_current_sec()) {
1789 cfs_spin_unlock(&req->rq_lock);
1792 rc = req->rq_receiving_reply || req->rq_must_unlink;
1793 cfs_spin_unlock(&req->rq_lock);
1798 ptlrpc_client_wake_req(struct ptlrpc_request *req)
1800 if (req->rq_set == NULL)
1801 cfs_waitq_signal(&req->rq_reply_waitq);
1803 cfs_waitq_signal(&req->rq_set->set_waitq);
1807 ptlrpc_rs_addref(struct ptlrpc_reply_state *rs)
1809 LASSERT(cfs_atomic_read(&rs->rs_refcount) > 0);
1810 cfs_atomic_inc(&rs->rs_refcount);
1814 ptlrpc_rs_decref(struct ptlrpc_reply_state *rs)
1816 LASSERT(cfs_atomic_read(&rs->rs_refcount) > 0);
1817 if (cfs_atomic_dec_and_test(&rs->rs_refcount))
1818 lustre_free_reply_state(rs);
1821 /* Should only be called once per req */
1822 static inline void ptlrpc_req_drop_rs(struct ptlrpc_request *req)
1824 if (req->rq_reply_state == NULL)
1825 return; /* shouldn't occur */
1826 ptlrpc_rs_decref(req->rq_reply_state);
1827 req->rq_reply_state = NULL;
1828 req->rq_repmsg = NULL;
1831 static inline __u32 lustre_request_magic(struct ptlrpc_request *req)
1833 return lustre_msg_get_magic(req->rq_reqmsg);
1836 static inline int ptlrpc_req_get_repsize(struct ptlrpc_request *req)
1838 switch (req->rq_reqmsg->lm_magic) {
1839 case LUSTRE_MSG_MAGIC_V2:
1840 return req->rq_reqmsg->lm_repsize;
1842 LASSERTF(0, "incorrect message magic: %08x\n",
1843 req->rq_reqmsg->lm_magic);
1848 static inline int ptlrpc_send_limit_expired(struct ptlrpc_request *req)
1850 if (req->rq_delay_limit != 0 &&
1851 cfs_time_before(cfs_time_add(req->rq_queued_time,
1852 cfs_time_seconds(req->rq_delay_limit)),
1853 cfs_time_current())) {
1859 static inline int ptlrpc_no_resend(struct ptlrpc_request *req)
1861 if (!req->rq_no_resend && ptlrpc_send_limit_expired(req)) {
1862 cfs_spin_lock(&req->rq_lock);
1863 req->rq_no_resend = 1;
1864 cfs_spin_unlock(&req->rq_lock);
1866 return req->rq_no_resend;
1869 /* ldlm/ldlm_lib.c */
1871 * Target client logic
1874 int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg);
1875 int client_obd_cleanup(struct obd_device *obddev);
1876 int client_connect_import(const struct lu_env *env,
1877 struct obd_export **exp, struct obd_device *obd,
1878 struct obd_uuid *cluuid, struct obd_connect_data *,
1880 int client_disconnect_export(struct obd_export *exp);
1881 int client_import_add_conn(struct obd_import *imp, struct obd_uuid *uuid,
1883 int client_import_del_conn(struct obd_import *imp, struct obd_uuid *uuid);
1884 int client_import_find_conn(struct obd_import *imp, lnet_nid_t peer,
1885 struct obd_uuid *uuid);
1886 int import_set_conn_priority(struct obd_import *imp, struct obd_uuid *uuid);
1887 void client_destroy_import(struct obd_import *imp);
1890 int server_disconnect_export(struct obd_export *exp);
1892 /* ptlrpc/pinger.c */
1894 * Pinger API (client side only)
1897 enum timeout_event {
1900 struct timeout_item;
1901 typedef int (*timeout_cb_t)(struct timeout_item *, void *);
1902 int ptlrpc_pinger_add_import(struct obd_import *imp);
1903 int ptlrpc_pinger_del_import(struct obd_import *imp);
1904 int ptlrpc_add_timeout_client(int time, enum timeout_event event,
1905 timeout_cb_t cb, void *data,
1906 cfs_list_t *obd_list);
1907 int ptlrpc_del_timeout_client(cfs_list_t *obd_list,
1908 enum timeout_event event);
1909 struct ptlrpc_request * ptlrpc_prep_ping(struct obd_import *imp);
1910 int ptlrpc_obd_ping(struct obd_device *obd);
1911 cfs_time_t ptlrpc_suspend_wakeup_time(void);
1913 void ping_evictor_start(void);
1914 void ping_evictor_stop(void);
1916 #define ping_evictor_start() do {} while (0)
1917 #define ping_evictor_stop() do {} while (0)
1919 int ptlrpc_check_and_wait_suspend(struct ptlrpc_request *req);
1922 /* ptlrpc daemon bind policy */
1924 /* all ptlrpcd threads are free mode */
1925 PDB_POLICY_NONE = 1,
1926 /* all ptlrpcd threads are bound mode */
1927 PDB_POLICY_FULL = 2,
1928 /* <free1 bound1> <free2 bound2> ... <freeN boundN> */
1929 PDB_POLICY_PAIR = 3,
1930 /* <free1 bound1> <bound1 free2> ... <freeN boundN> <boundN free1>,
1931 * means each ptlrpcd[X] has two partners: thread[X-1] and thread[X+1]*/
1932 PDB_POLICY_NEIGHBOR = 4,
1935 /* ptlrpc daemon load policy
1936 * It is caller's duty to specify how to push the async RPC into some ptlrpcd
1937 * queue, but it is not enforced, affected by "ptlrpcd_bind_policy". If it is
1938 * "PDB_POLICY_FULL", then the RPC will be processed by the selected ptlrpcd,
1939 * Otherwise, the RPC may be processed by the selected ptlrpcd or its partner,
1940 * depends on which is scheduled firstly, to accelerate the RPC processing. */
1942 /* on the same CPU core as the caller */
1943 PDL_POLICY_SAME = 1,
1944 /* within the same CPU partition, but not the same core as the caller */
1945 PDL_POLICY_LOCAL = 2,
1946 /* round-robin on all CPU cores, but not the same core as the caller */
1947 PDL_POLICY_ROUND = 3,
1948 /* the specified CPU core is preferred, but not enforced */
1949 PDL_POLICY_PREFERRED = 4,
1952 /* ptlrpc/ptlrpcd.c */
1953 void ptlrpcd_stop(struct ptlrpcd_ctl *pc, int force);
1954 void ptlrpcd_wake(struct ptlrpc_request *req);
1955 void ptlrpcd_add_req(struct ptlrpc_request *req, pdl_policy_t policy, int idx);
1956 void ptlrpcd_add_rqset(struct ptlrpc_request_set *set);
1957 int ptlrpcd_addref(void);
1958 void ptlrpcd_decref(void);
1960 /* ptlrpc/lproc_ptlrpc.c */
1962 * procfs output related functions
1965 const char* ll_opcode2str(__u32 opcode);
1967 void ptlrpc_lprocfs_register_obd(struct obd_device *obd);
1968 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd);
1969 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes);
1971 static inline void ptlrpc_lprocfs_register_obd(struct obd_device *obd) {}
1972 static inline void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd) {}
1973 static inline void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes) {}
1977 /* ptlrpc/llog_server.c */
1978 int llog_origin_handle_create(struct ptlrpc_request *req);
1979 int llog_origin_handle_destroy(struct ptlrpc_request *req);
1980 int llog_origin_handle_prev_block(struct ptlrpc_request *req);
1981 int llog_origin_handle_next_block(struct ptlrpc_request *req);
1982 int llog_origin_handle_read_header(struct ptlrpc_request *req);
1983 int llog_origin_handle_close(struct ptlrpc_request *req);
1984 int llog_origin_handle_cancel(struct ptlrpc_request *req);
1985 int llog_catinfo(struct ptlrpc_request *req);
1987 /* ptlrpc/llog_client.c */
1988 extern struct llog_operations llog_client_ops;