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 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 /** \defgroup PtlRPC Portal RPC and networking module.
38 * PortalRPC is the layer used by rest of lustre code to achieve network
39 * communications: establish connections with corresponding export and import
40 * states, listen for a service, send and receive RPCs.
41 * PortalRPC also includes base recovery framework: packet resending and
42 * replaying, reconnections, pinger.
44 * PortalRPC utilizes LNet as its transport layer.
58 #if defined(__linux__)
59 #include <linux/lustre_net.h>
60 #elif defined(__APPLE__)
61 #include <darwin/lustre_net.h>
62 #elif defined(__WINNT__)
63 #include <winnt/lustre_net.h>
65 #error Unsupported operating system.
68 #include <libcfs/libcfs.h>
70 #include <lnet/lnet.h>
71 #include <lustre/lustre_idl.h>
72 #include <lustre_ha.h>
73 #include <lustre_sec.h>
74 #include <lustre_import.h>
75 #include <lprocfs_status.h>
76 #include <lu_object.h>
77 #include <lustre_req_layout.h>
79 #include <obd_support.h>
80 #include <lustre_ver.h>
82 /* MD flags we _always_ use */
83 #define PTLRPC_MD_OPTIONS 0
86 * Define maxima for bulk I/O
87 * CAVEAT EMPTOR, with multinet (i.e. routers forwarding between networks)
88 * these limits are system wide and not interface-local. */
89 #define PTLRPC_MAX_BRW_BITS LNET_MTU_BITS
90 #define PTLRPC_MAX_BRW_SIZE (1<<LNET_MTU_BITS)
91 #define PTLRPC_MAX_BRW_PAGES (PTLRPC_MAX_BRW_SIZE >> CFS_PAGE_SHIFT)
93 /* When PAGE_SIZE is a constant, we can check our arithmetic here with cpp! */
95 # if ((PTLRPC_MAX_BRW_PAGES & (PTLRPC_MAX_BRW_PAGES - 1)) != 0)
96 # error "PTLRPC_MAX_BRW_PAGES isn't a power of two"
98 # if (PTLRPC_MAX_BRW_SIZE != (PTLRPC_MAX_BRW_PAGES * CFS_PAGE_SIZE))
99 # error "PTLRPC_MAX_BRW_SIZE isn't PTLRPC_MAX_BRW_PAGES * CFS_PAGE_SIZE"
101 # if (PTLRPC_MAX_BRW_SIZE > LNET_MTU)
102 # error "PTLRPC_MAX_BRW_SIZE too big"
104 # if (PTLRPC_MAX_BRW_PAGES > LNET_MAX_IOV)
105 # error "PTLRPC_MAX_BRW_PAGES too big"
107 #endif /* __KERNEL__ */
110 * The following constants determine how memory is used to buffer incoming
113 * ?_NBUFS # buffers to allocate when growing the pool
114 * ?_BUFSIZE # bytes in a single request buffer
115 * ?_MAXREQSIZE # maximum request service will receive
117 * When fewer than ?_NBUFS/2 buffers are posted for receive, another chunk
118 * of ?_NBUFS is added to the pool.
120 * Messages larger than ?_MAXREQSIZE are dropped. Request buffers are
121 * considered full when less than ?_MAXREQSIZE is left in them.
123 #define LDLM_THREADS_AUTO_MIN (2)
124 #define LDLM_THREADS_AUTO_MAX min_t(unsigned, cfs_num_online_cpus() * \
125 cfs_num_online_cpus() * 32, 128)
126 #define LDLM_BL_THREADS LDLM_THREADS_AUTO_MIN
127 #define LDLM_NBUFS (64 * cfs_num_online_cpus())
128 #define LDLM_BUFSIZE (8 * 1024)
129 #define LDLM_MAXREQSIZE (5 * 1024)
130 #define LDLM_MAXREPSIZE (1024)
132 /** Absolute limits */
133 #define MDT_MIN_THREADS 2UL
134 #ifndef MDT_MAX_THREADS
135 #define MDT_MAX_THREADS 512UL
137 #define MDS_NBUFS (64 * cfs_num_online_cpus())
138 #define MDS_BUFSIZE (8 * 1024)
140 * Assume file name length = FNAME_MAX = 256 (true for ext3).
141 * path name length = PATH_MAX = 4096
142 * LOV MD size max = EA_MAX = 4000
143 * symlink: FNAME_MAX + PATH_MAX <- largest
144 * link: FNAME_MAX + PATH_MAX (mds_rec_link < mds_rec_create)
145 * rename: FNAME_MAX + FNAME_MAX
146 * open: FNAME_MAX + EA_MAX
148 * MDS_MAXREQSIZE ~= 4736 bytes =
149 * lustre_msg + ldlm_request + mds_body + mds_rec_create + FNAME_MAX + PATH_MAX
150 * MDS_MAXREPSIZE ~= 8300 bytes = lustre_msg + llog_header
151 * or, for mds_close() and mds_reint_unlink() on a many-OST filesystem:
152 * = 9210 bytes = lustre_msg + mds_body + 160 * (easize + cookiesize)
154 * Realistic size is about 512 bytes (20 character name + 128 char symlink),
155 * except in the open case where there are a large number of OSTs in a LOV.
157 #define MDS_MAXREQSIZE (5 * 1024)
158 #define MDS_MAXREPSIZE max(9 * 1024, 362 + LOV_MAX_STRIPE_COUNT * 56)
160 /** FLD_MAXREQSIZE == lustre_msg + __u32 padding + ptlrpc_body + opc + md_fld */
161 #define FLD_MAXREQSIZE (160)
163 /** FLD_MAXREPSIZE == lustre_msg + ptlrpc_body + md_fld */
164 #define FLD_MAXREPSIZE (152)
167 * SEQ_MAXREQSIZE == lustre_msg + __u32 padding + ptlrpc_body + opc + lu_range +
169 #define SEQ_MAXREQSIZE (160)
171 /** SEQ_MAXREPSIZE == lustre_msg + ptlrpc_body + lu_range */
172 #define SEQ_MAXREPSIZE (152)
174 /** MGS threads must be >= 3, see bug 22458 comment #28 */
175 #define MGS_THREADS_AUTO_MIN 3
176 #define MGS_THREADS_AUTO_MAX 32
177 #define MGS_NBUFS (64 * cfs_num_online_cpus())
178 #define MGS_BUFSIZE (8 * 1024)
179 #define MGS_MAXREQSIZE (7 * 1024)
180 #define MGS_MAXREPSIZE (9 * 1024)
182 /** Absolute OSS limits */
183 #define OSS_THREADS_MIN 3 /* difficult replies, HPQ, others */
184 #define OSS_THREADS_MAX 512
185 #define OST_NBUFS (64 * cfs_num_online_cpus())
186 #define OST_BUFSIZE (8 * 1024)
189 * OST_MAXREQSIZE ~= 4768 bytes =
190 * lustre_msg + obdo + 16 * obd_ioobj + 256 * niobuf_remote
192 * - single object with 16 pages is 512 bytes
193 * - OST_MAXREQSIZE must be at least 1 page of cookies plus some spillover
195 #define OST_MAXREQSIZE (5 * 1024)
196 #define OST_MAXREPSIZE (9 * 1024)
198 /* Macro to hide a typecast. */
199 #define ptlrpc_req_async_args(req) ((void *)&req->rq_async_args)
202 * Structure to single define portal connection.
204 struct ptlrpc_connection {
205 /** linkage for connections hash table */
206 cfs_hlist_node_t c_hash;
207 /** Our own lnet nid for this connection */
209 /** Remote side nid for this connection */
210 lnet_process_id_t c_peer;
211 /** UUID of the other side */
212 struct obd_uuid c_remote_uuid;
213 /** reference counter for this connection */
214 cfs_atomic_t c_refcount;
217 /** Client definition for PortalRPC */
218 struct ptlrpc_client {
219 /** What lnet portal does this client send messages to by default */
220 __u32 cli_request_portal;
221 /** What portal do we expect replies on */
222 __u32 cli_reply_portal;
223 /** Name of the client */
227 /** state flags of requests */
228 /* XXX only ones left are those used by the bulk descs as well! */
229 #define PTL_RPC_FL_INTR (1 << 0) /* reply wait was interrupted by user */
230 #define PTL_RPC_FL_TIMEOUT (1 << 7) /* request timed out waiting for reply */
232 #define REQ_MAX_ACK_LOCKS 8
234 union ptlrpc_async_args {
236 * Scratchpad for passing args to completion interpreter. Users
237 * cast to the struct of their choosing, and CLASSERT that this is
238 * big enough. For _tons_ of context, OBD_ALLOC a struct and store
239 * a pointer to it here. The pointer_arg ensures this struct is at
240 * least big enough for that.
242 void *pointer_arg[11];
246 struct ptlrpc_request_set;
247 typedef int (*set_interpreter_func)(struct ptlrpc_request_set *, void *, int);
250 * Definition of request set structure.
251 * Request set is a list of requests (not necessary to the same target) that
252 * once populated with RPCs could be sent in parallel.
253 * There are two kinds of request sets. General purpose and with dedicated
254 * serving thread. Example of the latter is ptlrpcd set.
255 * For general purpose sets once request set started sending it is impossible
256 * to add new requests to such set.
257 * Provides a way to call "completion callbacks" when all requests in the set
260 struct ptlrpc_request_set {
261 /** number of uncompleted requests */
262 cfs_atomic_t set_remaining;
263 /** wait queue to wait on for request events */
264 cfs_waitq_t set_waitq;
265 cfs_waitq_t *set_wakeup_ptr;
266 /** List of requests in the set */
267 cfs_list_t set_requests;
269 * List of completion callbacks to be called when the set is completed
270 * This is only used if \a set_interpret is NULL.
271 * Links struct ptlrpc_set_cbdata.
273 cfs_list_t set_cblist;
274 /** Completion callback, if only one. */
275 set_interpreter_func set_interpret;
276 /** opaq argument passed to completion \a set_interpret callback. */
279 * Lock for \a set_new_requests manipulations
280 * locked so that any old caller can communicate requests to
281 * the set holder who can then fold them into the lock-free set
283 cfs_spinlock_t set_new_req_lock;
284 /** List of new yet unsent requests. Only used with ptlrpcd now. */
285 cfs_list_t set_new_requests;
289 * Description of a single ptrlrpc_set callback
291 struct ptlrpc_set_cbdata {
292 /** List linkage item */
294 /** Pointer to interpreting function */
295 set_interpreter_func psc_interpret;
296 /** Opaq argument to pass to the callback */
300 struct ptlrpc_bulk_desc;
303 * ptlrpc callback & work item stuff
305 struct ptlrpc_cb_id {
306 void (*cbid_fn)(lnet_event_t *ev); /* specific callback fn */
307 void *cbid_arg; /* additional arg */
310 /** Maximum number of locks to fit into reply state */
311 #define RS_MAX_LOCKS 8
315 * Structure to define reply state on the server
316 * Reply state holds various reply message information. Also for "difficult"
317 * replies (rep-ack case) we store the state after sending reply and wait
318 * for the client to acknowledge the reception. In these cases locks could be
319 * added to the state for replay/failover consistency guarantees.
321 struct ptlrpc_reply_state {
322 /** Callback description */
323 struct ptlrpc_cb_id rs_cb_id;
324 /** Linkage for list of all reply states in a system */
326 /** Linkage for list of all reply states on same export */
327 cfs_list_t rs_exp_list;
328 /** Linkage for list of all reply states for same obd */
329 cfs_list_t rs_obd_list;
331 cfs_list_t rs_debug_list;
333 /** A spinlock to protect the reply state flags */
334 cfs_spinlock_t rs_lock;
335 /** Reply state flags */
336 unsigned long rs_difficult:1; /* ACK/commit stuff */
337 unsigned long rs_no_ack:1; /* no ACK, even for
338 difficult requests */
339 unsigned long rs_scheduled:1; /* being handled? */
340 unsigned long rs_scheduled_ever:1;/* any schedule attempts? */
341 unsigned long rs_handled:1; /* been handled yet? */
342 unsigned long rs_on_net:1; /* reply_out_callback pending? */
343 unsigned long rs_prealloc:1; /* rs from prealloc list */
344 unsigned long rs_committed:1;/* the transaction was committed
345 and the rs was dispatched
346 by ptlrpc_commit_replies */
347 /** Size of the state */
351 /** Transaction number */
355 struct obd_export *rs_export;
356 struct ptlrpc_service *rs_service;
357 /** Lnet metadata handle for the reply */
358 lnet_handle_md_t rs_md_h;
359 cfs_atomic_t rs_refcount;
361 /** Context for the sevice thread */
362 struct ptlrpc_svc_ctx *rs_svc_ctx;
363 /** Reply buffer (actually sent to the client), encoded if needed */
364 struct lustre_msg *rs_repbuf; /* wrapper */
365 /** Size of the reply buffer */
366 int rs_repbuf_len; /* wrapper buf length */
367 /** Size of the reply message */
368 int rs_repdata_len; /* wrapper msg length */
370 * Actual reply message. Its content is encrupted (if needed) to
371 * produce reply buffer for actual sending. In simple case
372 * of no network encryption we jus set \a rs_repbuf to \a rs_msg
374 struct lustre_msg *rs_msg; /* reply message */
376 /** Number of locks awaiting client ACK */
378 /** Handles of locks awaiting client reply ACK */
379 struct lustre_handle rs_locks[RS_MAX_LOCKS];
380 /** Lock modes of locks in \a rs_locks */
381 ldlm_mode_t rs_modes[RS_MAX_LOCKS];
384 struct ptlrpc_thread;
388 RQ_PHASE_NEW = 0xebc0de00,
389 RQ_PHASE_RPC = 0xebc0de01,
390 RQ_PHASE_BULK = 0xebc0de02,
391 RQ_PHASE_INTERPRET = 0xebc0de03,
392 RQ_PHASE_COMPLETE = 0xebc0de04,
393 RQ_PHASE_UNREGISTERING = 0xebc0de05,
394 RQ_PHASE_UNDEFINED = 0xebc0de06
397 /** Type of request interpreter call-back */
398 typedef int (*ptlrpc_interpterer_t)(const struct lu_env *env,
399 struct ptlrpc_request *req,
403 * Definition of request pool structure.
404 * The pool is used to store empty preallocated requests for the case
405 * when we would actually need to send something without performing
406 * any allocations (to avoid e.g. OOM).
408 struct ptlrpc_request_pool {
409 /** Locks the list */
410 cfs_spinlock_t prp_lock;
411 /** list of ptlrpc_request structs */
412 cfs_list_t prp_req_list;
413 /** Maximum message size that would fit into a rquest from this pool */
415 /** Function to allocate more requests for this pool */
416 void (*prp_populate)(struct ptlrpc_request_pool *, int);
425 * Basic request prioritization operations structure.
426 * The whole idea is centered around locks and RPCs that might affect locks.
427 * When a lock is contended we try to give priority to RPCs that might lead
428 * to fastest release of that lock.
429 * Currently only implemented for OSTs only in a way that makes all
430 * IO and truncate RPCs that are coming from a locked region where a lock is
431 * contended a priority over other requests.
433 struct ptlrpc_hpreq_ops {
435 * Check if the lock handle of the given lock is the same as
436 * taken from the request.
438 int (*hpreq_lock_match)(struct ptlrpc_request *, struct ldlm_lock *);
440 * Check if the request is a high priority one.
442 int (*hpreq_check)(struct ptlrpc_request *);
446 * Represents remote procedure call.
448 * This is a staple structure used by everybody wanting to send a request
451 struct ptlrpc_request {
452 /* Request type: one of PTL_RPC_MSG_* */
455 * Linkage item through which this request is included into
456 * sending/delayed lists on client and into rqbd list on server
460 * Server side list of incoming unserved requests sorted by arrival
461 * time. Traversed from time to time to notice about to expire
462 * requests and sent back "early replies" to clients to let them
463 * know server is alive and well, just very busy to service their
466 cfs_list_t rq_timed_list;
467 /** server-side history, used for debuging purposes. */
468 cfs_list_t rq_history_list;
469 /** server-side per-export list */
470 cfs_list_t rq_exp_list;
471 /** server-side hp handlers */
472 struct ptlrpc_hpreq_ops *rq_ops;
473 /** history sequence # */
474 __u64 rq_history_seq;
475 /** the index of service's srv_at_array into which request is linked */
477 /** Result of request processing */
479 /** Lock to protect request flags and some other important bits, like
482 cfs_spinlock_t rq_lock;
483 /** client-side flags are serialized by rq_lock */
484 unsigned long rq_intr:1, rq_replied:1, rq_err:1,
485 rq_timedout:1, rq_resend:1, rq_restart:1,
487 * when ->rq_replay is set, request is kept by the client even
488 * after server commits corresponding transaction. This is
489 * used for operations that require sequence of multiple
490 * requests to be replayed. The only example currently is file
491 * open/close. When last request in such a sequence is
492 * committed, ->rq_replay is cleared on all requests in the
496 rq_no_resend:1, rq_waiting:1, rq_receiving_reply:1,
497 rq_no_delay:1, rq_net_err:1, rq_wait_ctx:1,
498 rq_early:1, rq_must_unlink:1,
499 rq_fake:1, /* this fake req */
500 rq_memalloc:1, /* req originated from "kswapd" */
501 /* server-side flags */
502 rq_packed_final:1, /* packed final reply */
503 rq_hp:1, /* high priority RPC */
504 rq_at_linked:1, /* link into service's srv_at_array */
507 /* whether the "rq_set" is a valid one */
510 enum rq_phase rq_phase; /* one of RQ_PHASE_* */
511 enum rq_phase rq_next_phase; /* one of RQ_PHASE_* to be used next */
512 cfs_atomic_t rq_refcount;/* client-side refcount for SENT race,
513 server-side refcounf for multiple replies */
515 /** initial thread servicing this request */
516 struct ptlrpc_thread *rq_svc_thread;
518 /** Portal to which this request would be sent */
519 int rq_request_portal; /* XXX FIXME bug 249 */
520 /** Portal where to wait for reply and where reply would be sent */
521 int rq_reply_portal; /* XXX FIXME bug 249 */
525 * !rq_truncate : # reply bytes actually received,
526 * rq_truncate : required repbuf_len for resend
529 /** Request length */
531 /** Request message - what client sent */
532 struct lustre_msg *rq_reqmsg;
536 /** Reply message - server response */
537 struct lustre_msg *rq_repmsg;
538 /** Transaction number */
543 * List item to for replay list. Not yet commited requests get linked
545 * Also see \a rq_replay comment above.
547 cfs_list_t rq_replay_list;
550 * security and encryption data
552 struct ptlrpc_cli_ctx *rq_cli_ctx; /**< client's half ctx */
553 struct ptlrpc_svc_ctx *rq_svc_ctx; /**< server's half ctx */
554 cfs_list_t rq_ctx_chain; /**< link to waited ctx */
556 struct sptlrpc_flavor rq_flvr; /**< for client & server */
557 enum lustre_sec_part rq_sp_from;
559 unsigned long /* client/server security flags */
560 rq_ctx_init:1, /* context initiation */
561 rq_ctx_fini:1, /* context destroy */
562 rq_bulk_read:1, /* request bulk read */
563 rq_bulk_write:1, /* request bulk write */
564 /* server authentication flags */
565 rq_auth_gss:1, /* authenticated by gss */
566 rq_auth_remote:1, /* authed as remote user */
567 rq_auth_usr_root:1, /* authed as root */
568 rq_auth_usr_mdt:1, /* authed as mdt */
569 rq_auth_usr_ost:1, /* authed as ost */
570 /* security tfm flags */
573 /* doesn't expect reply FIXME */
575 rq_pill_init:1; /* pill initialized */
577 uid_t rq_auth_uid; /* authed uid */
578 uid_t rq_auth_mapped_uid; /* authed uid mapped to */
580 /* (server side), pointed directly into req buffer */
581 struct ptlrpc_user_desc *rq_user_desc;
583 /** early replies go to offset 0, regular replies go after that */
584 unsigned int rq_reply_off;
586 /* various buffer pointers */
587 struct lustre_msg *rq_reqbuf; /* req wrapper */
588 int rq_reqbuf_len; /* req wrapper buf len */
589 int rq_reqdata_len; /* req wrapper msg len */
590 char *rq_repbuf; /* rep buffer */
591 int rq_repbuf_len; /* rep buffer len */
592 struct lustre_msg *rq_repdata; /* rep wrapper msg */
593 int rq_repdata_len; /* rep wrapper msg len */
594 struct lustre_msg *rq_clrbuf; /* only in priv mode */
595 int rq_clrbuf_len; /* only in priv mode */
596 int rq_clrdata_len; /* only in priv mode */
600 /** Fields that help to see if request and reply were swabbed or not */
601 __u32 rq_req_swab_mask;
602 __u32 rq_rep_swab_mask;
604 /** What was import generation when this request was sent */
605 int rq_import_generation;
606 enum lustre_imp_state rq_send_state;
608 /** how many early replies (for stats) */
611 /** client+server request */
612 lnet_handle_md_t rq_req_md_h;
613 struct ptlrpc_cb_id rq_req_cbid;
614 /** optional time limit for send attempts */
615 cfs_duration_t rq_delay_limit;
616 /** time request was first queued */
617 cfs_time_t rq_queued_time;
620 /** request arrival time */
621 struct timeval rq_arrival_time;
622 /** separated reply state */
623 struct ptlrpc_reply_state *rq_reply_state;
624 /** incoming request buffer */
625 struct ptlrpc_request_buffer_desc *rq_rqbd;
627 __u32 rq_uid; /* peer uid, used in MDS only */
630 /** client-only incoming reply */
631 lnet_handle_md_t rq_reply_md_h;
632 cfs_waitq_t rq_reply_waitq;
633 struct ptlrpc_cb_id rq_reply_cbid;
637 /** Peer description (the other side) */
638 lnet_process_id_t rq_peer;
639 /** Server-side, export on which request was received */
640 struct obd_export *rq_export;
641 /** Client side, import where request is being sent */
642 struct obd_import *rq_import;
644 /** Replay callback, called after request is replayed at recovery */
645 void (*rq_replay_cb)(struct ptlrpc_request *);
647 * Commit callback, called when request is committed and about to be
650 void (*rq_commit_cb)(struct ptlrpc_request *);
651 /** Opaq data for replay and commit callbacks. */
654 /** For bulk requests on client only: bulk descriptor */
655 struct ptlrpc_bulk_desc *rq_bulk;
657 /** client outgoing req */
659 * when request/reply sent (secs), or time when request should be sent
662 /** time for request really sent out */
665 /** when request must finish. volatile
666 * so that servers' early reply updates to the deadline aren't
667 * kept in per-cpu cache */
668 volatile time_t rq_deadline;
669 /** when req reply unlink must finish. */
670 time_t rq_reply_deadline;
671 /** when req bulk unlink must finish. */
672 time_t rq_bulk_deadline;
674 * service time estimate (secs)
675 * If the requestsis not served by this time, it is marked as timed out.
679 /** Multi-rpc bits */
680 /** Link item for request set lists */
681 cfs_list_t rq_set_chain;
682 /** Per-request waitq introduced by bug 21938 for recovery waiting */
683 cfs_waitq_t rq_set_waitq;
684 /** Link back to the request set */
685 struct ptlrpc_request_set *rq_set;
686 /** Async completion handler, called when reply is received */
687 ptlrpc_interpterer_t rq_interpret_reply;
688 /** Async completion context */
689 union ptlrpc_async_args rq_async_args;
691 /** Pool if request is from preallocated list */
692 struct ptlrpc_request_pool *rq_pool;
694 struct lu_context rq_session;
695 struct lu_context rq_recov_session;
697 /** request format description */
698 struct req_capsule rq_pill;
702 * Call completion handler for rpc if any, return it's status or original
703 * rc if there was no handler defined for this request.
705 static inline int ptlrpc_req_interpret(const struct lu_env *env,
706 struct ptlrpc_request *req, int rc)
708 if (req->rq_interpret_reply != NULL) {
709 req->rq_status = req->rq_interpret_reply(env, req,
712 return req->rq_status;
718 * Returns 1 if request buffer at offset \a index was already swabbed
720 static inline int lustre_req_swabbed(struct ptlrpc_request *req, int index)
722 LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
723 return req->rq_req_swab_mask & (1 << index);
727 * Returns 1 if request reply buffer at offset \a index was already swabbed
729 static inline int lustre_rep_swabbed(struct ptlrpc_request *req, int index)
731 LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
732 return req->rq_rep_swab_mask & (1 << index);
736 * Returns 1 if request needs to be swabbed into local cpu byteorder
738 static inline int ptlrpc_req_need_swab(struct ptlrpc_request *req)
740 return lustre_req_swabbed(req, MSG_PTLRPC_HEADER_OFF);
744 * Returns 1 if request reply needs to be swabbed into local cpu byteorder
746 static inline int ptlrpc_rep_need_swab(struct ptlrpc_request *req)
748 return lustre_rep_swabbed(req, MSG_PTLRPC_HEADER_OFF);
752 * Mark request buffer at offset \a index that it was already swabbed
754 static inline void lustre_set_req_swabbed(struct ptlrpc_request *req, int index)
756 LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
757 LASSERT((req->rq_req_swab_mask & (1 << index)) == 0);
758 req->rq_req_swab_mask |= 1 << index;
762 * Mark request reply buffer at offset \a index that it was already swabbed
764 static inline void lustre_set_rep_swabbed(struct ptlrpc_request *req, int index)
766 LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
767 LASSERT((req->rq_rep_swab_mask & (1 << index)) == 0);
768 req->rq_rep_swab_mask |= 1 << index;
772 * Convert numerical request phase value \a phase into text string description
774 static inline const char *
775 ptlrpc_phase2str(enum rq_phase phase)
784 case RQ_PHASE_INTERPRET:
786 case RQ_PHASE_COMPLETE:
788 case RQ_PHASE_UNREGISTERING:
789 return "Unregistering";
796 * Convert numerical request phase of the request \a req into text stringi
799 static inline const char *
800 ptlrpc_rqphase2str(struct ptlrpc_request *req)
802 return ptlrpc_phase2str(req->rq_phase);
806 * Debugging functions and helpers to print request structure into debug log
809 /* Spare the preprocessor, spoil the bugs. */
810 #define FLAG(field, str) (field ? str : "")
812 /** Convert bit flags into a string */
813 #define DEBUG_REQ_FLAGS(req) \
814 ptlrpc_rqphase2str(req), \
815 FLAG(req->rq_intr, "I"), FLAG(req->rq_replied, "R"), \
816 FLAG(req->rq_err, "E"), \
817 FLAG(req->rq_timedout, "X") /* eXpired */, FLAG(req->rq_resend, "S"), \
818 FLAG(req->rq_restart, "T"), FLAG(req->rq_replay, "P"), \
819 FLAG(req->rq_no_resend, "N"), \
820 FLAG(req->rq_waiting, "W"), \
821 FLAG(req->rq_wait_ctx, "C"), FLAG(req->rq_hp, "H"), \
822 FLAG(req->rq_committed, "M")
824 #define REQ_FLAGS_FMT "%s:%s%s%s%s%s%s%s%s%s%s%s%s"
826 void _debug_req(struct ptlrpc_request *req, __u32 mask,
827 struct libcfs_debug_msg_data *data, const char *fmt, ...)
828 __attribute__ ((format (printf, 4, 5)));
831 * Helper that decides if we need to print request accordig to current debug
834 #define debug_req(cdls, level, req, file, func, line, fmt, a...) \
838 if (((level) & D_CANTMASK) != 0 || \
839 ((libcfs_debug & (level)) != 0 && \
840 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) { \
841 static struct libcfs_debug_msg_data _req_dbg_data = \
842 DEBUG_MSG_DATA_INIT(cdls, DEBUG_SUBSYSTEM, file, func, line); \
843 _debug_req((req), (level), &_req_dbg_data, fmt, ##a); \
848 * This is the debug print function you need to use to print request sturucture
849 * content into lustre debug log.
850 * for most callers (level is a constant) this is resolved at compile time */
851 #define DEBUG_REQ(level, req, fmt, args...) \
853 if ((level) & (D_ERROR | D_WARNING)) { \
854 static cfs_debug_limit_state_t cdls; \
855 debug_req(&cdls, level, req, __FILE__, __func__, __LINE__, \
856 "@@@ "fmt" ", ## args); \
858 debug_req(NULL, level, req, __FILE__, __func__, __LINE__, \
859 "@@@ "fmt" ", ## args); \
864 * Structure that defines a single page of a bulk transfer
866 struct ptlrpc_bulk_page {
867 /** Linkage to list of pages in a bulk */
870 * Number of bytes in a page to transfer starting from \a bp_pageoffset
873 /** offset within a page */
875 /** The page itself */
876 struct page *bp_page;
879 #define BULK_GET_SOURCE 0
880 #define BULK_PUT_SINK 1
881 #define BULK_GET_SINK 2
882 #define BULK_PUT_SOURCE 3
885 * Definition of buk descriptor.
886 * Bulks are special "Two phase" RPCs where initial request message
887 * is sent first and it is followed bt a transfer (o receiving) of a large
888 * amount of data to be settled into pages referenced from the bulk descriptors.
889 * Bulks transfers (the actual data following the small requests) are done
890 * on separate LNet portals.
891 * In lustre we use bulk transfers for READ and WRITE transfers from/to OSTs.
892 * Another user is readpage for MDT.
894 struct ptlrpc_bulk_desc {
895 /** completed successfully */
896 unsigned long bd_success:1;
897 /** accessible to the network (network io potentially in progress) */
898 unsigned long bd_network_rw:1;
899 /** {put,get}{source,sink} */
900 unsigned long bd_type:2;
902 unsigned long bd_registered:1;
903 /** For serialization with callback */
904 cfs_spinlock_t bd_lock;
905 /** Import generation when request for this bulk was sent */
906 int bd_import_generation;
907 /** Server side - export this bulk created for */
908 struct obd_export *bd_export;
909 /** Client side - import this bulk was sent on */
910 struct obd_import *bd_import;
911 /** LNet portal for this bulk */
913 /** Back pointer to the request */
914 struct ptlrpc_request *bd_req;
915 cfs_waitq_t bd_waitq; /* server side only WQ */
916 int bd_iov_count; /* # entries in bd_iov */
917 int bd_max_iov; /* allocated size of bd_iov */
918 int bd_nob; /* # bytes covered */
919 int bd_nob_transferred; /* # bytes GOT/PUT */
923 struct ptlrpc_cb_id bd_cbid; /* network callback info */
924 lnet_handle_md_t bd_md_h; /* associated MD */
925 lnet_nid_t bd_sender; /* stash event::sender */
927 #if defined(__KERNEL__)
929 * encrypt iov, size is either 0 or bd_iov_count.
931 lnet_kiov_t *bd_enc_iov;
933 lnet_kiov_t bd_iov[0];
935 lnet_md_iovec_t bd_iov[0];
940 SVC_STOPPED = 1 << 0,
941 SVC_STOPPING = 1 << 1,
942 SVC_STARTING = 1 << 2,
943 SVC_RUNNING = 1 << 3,
949 * Definition of server service thread structure
951 struct ptlrpc_thread {
953 * List of active threads in svc->srv_threads
957 * thread-private data (preallocated memory)
962 * service thread index, from ptlrpc_start_threads
970 * put watchdog in the structure per thread b=14840
972 struct lc_watchdog *t_watchdog;
974 * the svc this thread belonged to b=18582
976 struct ptlrpc_service *t_svc;
977 cfs_waitq_t t_ctl_waitq;
978 struct lu_env *t_env;
982 * Request buffer descriptor structure.
983 * This is a structure that contains one posted request buffer for service.
984 * Once data land into a buffer, event callback creates actual request and
985 * notifies wakes one of the service threads to process new incoming request.
986 * More than one request can fit into the buffer.
988 struct ptlrpc_request_buffer_desc {
989 /** Link item for rqbds on a service */
990 cfs_list_t rqbd_list;
991 /** History of requests for this buffer */
992 cfs_list_t rqbd_reqs;
993 /** Back pointer to service for which this buffer is registered */
994 struct ptlrpc_service *rqbd_service;
995 /** LNet descriptor */
996 lnet_handle_md_t rqbd_md_h;
998 /** The buffer itself */
1000 struct ptlrpc_cb_id rqbd_cbid;
1002 * This "embedded" request structure is only used for the
1003 * last request to fit into the buffer
1005 struct ptlrpc_request rqbd_req;
1008 typedef int (*svc_thr_init_t)(struct ptlrpc_thread *thread);
1009 typedef void (*svc_thr_done_t)(struct ptlrpc_thread *thread);
1010 typedef int (*svc_handler_t)(struct ptlrpc_request *req);
1011 typedef int (*svc_hpreq_handler_t)(struct ptlrpc_request *);
1012 typedef void (*svc_req_printfn_t)(void *, struct ptlrpc_request *);
1014 #ifndef __cfs_cacheline_aligned
1015 /* NB: put it here for reducing patche dependence */
1016 # define __cfs_cacheline_aligned
1020 * How many high priority requests to serve before serving one normal
1023 #define PTLRPC_SVC_HP_RATIO 10
1026 * Definition of PortalRPC service.
1027 * The service is listening on a particular portal (like tcp port)
1028 * and perform actions for a specific server like IO service for OST
1029 * or general metadata service for MDS.
1031 * ptlrpc service has four locks:
1033 * serialize operations on rqbd and requests waiting for preprocess
1035 * serialize operations active requests sent to this portal
1037 * serialize adaptive timeout stuff
1039 * serialize operations on RS list (reply states)
1041 * We don't have any use-case to take two or more locks at the same time
1042 * for now, so there is no lock order issue.
1044 struct ptlrpc_service {
1045 /** most often accessed fields */
1046 /** chain thru all services */
1047 cfs_list_t srv_list;
1048 /** only statically allocated strings here; we don't clean them */
1050 /** only statically allocated strings here; we don't clean them */
1051 char *srv_thread_name;
1052 /** service thread list */
1053 cfs_list_t srv_threads;
1054 /** threads to start at beginning of service */
1055 int srv_threads_min;
1056 /** thread upper limit */
1057 int srv_threads_max;
1058 /** always increasing number */
1059 unsigned srv_threads_next_id;
1060 /** # of starting threads */
1061 int srv_threads_starting;
1062 /** # running threads */
1063 int srv_threads_running;
1065 /** service operations, move to ptlrpc_svc_ops_t in the future */
1068 * if non-NULL called during thread creation (ptlrpc_start_thread())
1069 * to initialize service specific per-thread state.
1071 svc_thr_init_t srv_init;
1073 * if non-NULL called during thread shutdown (ptlrpc_main()) to
1074 * destruct state created by ->srv_init().
1076 svc_thr_done_t srv_done;
1077 /** Handler function for incoming requests for this service */
1078 svc_handler_t srv_handler;
1079 /** hp request handler */
1080 svc_hpreq_handler_t srv_hpreq_handler;
1081 /** service-specific print fn */
1082 svc_req_printfn_t srv_req_printfn;
1085 /** Root of /proc dir tree for this service */
1086 cfs_proc_dir_entry_t *srv_procroot;
1087 /** Pointer to statistic data for this service */
1088 struct lprocfs_stats *srv_stats;
1089 /** # hp per lp reqs to handle */
1090 int srv_hpreq_ratio;
1091 /** biggest request to receive */
1092 int srv_max_req_size;
1093 /** biggest reply to send */
1094 int srv_max_reply_size;
1095 /** size of individual buffers */
1097 /** # buffers to allocate in 1 group */
1098 int srv_nbuf_per_group;
1099 /** Local portal on which to receive requests */
1100 __u32 srv_req_portal;
1101 /** Portal on the client to send replies to */
1102 __u32 srv_rep_portal;
1104 * Tags for lu_context associated with this thread, see struct
1108 /** soft watchdog timeout multiplier */
1109 int srv_watchdog_factor;
1110 /** bind threads to CPUs */
1111 unsigned srv_cpu_affinity:1;
1112 /** under unregister_service */
1113 unsigned srv_is_stopping:1;
1116 * serialize the following fields, used for protecting
1117 * rqbd list and incoming requests waiting for preprocess
1119 cfs_spinlock_t srv_lock __cfs_cacheline_aligned;
1120 /** incoming reqs */
1121 cfs_list_t srv_req_in_queue;
1122 /** total # req buffer descs allocated */
1124 /** # posted request buffers */
1125 int srv_nrqbd_receiving;
1126 /** timeout before re-posting reqs, in tick */
1127 cfs_duration_t srv_rqbd_timeout;
1128 /** request buffers to be reposted */
1129 cfs_list_t srv_idle_rqbds;
1130 /** req buffers receiving */
1131 cfs_list_t srv_active_rqbds;
1132 /** request buffer history */
1133 cfs_list_t srv_history_rqbds;
1134 /** # request buffers in history */
1135 int srv_n_history_rqbds;
1136 /** max # request buffers in history */
1137 int srv_max_history_rqbds;
1138 /** request history */
1139 cfs_list_t srv_request_history;
1140 /** next request sequence # */
1141 __u64 srv_request_seq;
1142 /** highest seq culled from history */
1143 __u64 srv_request_max_cull_seq;
1145 * all threads sleep on this. This wait-queue is signalled when new
1146 * incoming request arrives and when difficult reply has to be handled.
1148 cfs_waitq_t srv_waitq;
1151 * serialize the following fields, used for processing requests
1152 * sent to this portal
1154 cfs_spinlock_t srv_rq_lock __cfs_cacheline_aligned;
1155 /** # reqs in either of the queues below */
1156 /** reqs waiting for service */
1157 cfs_list_t srv_request_queue;
1158 /** high priority queue */
1159 cfs_list_t srv_request_hpq;
1160 /** # incoming reqs */
1161 int srv_n_queued_reqs;
1162 /** # reqs being served */
1163 int srv_n_active_reqs;
1164 /** # HPreqs being served */
1165 int srv_n_active_hpreq;
1166 /** # hp requests handled */
1167 int srv_hpreq_count;
1172 * serialize the following fields, used for changes on
1175 cfs_spinlock_t srv_at_lock __cfs_cacheline_aligned;
1176 /** estimated rpc service time */
1177 struct adaptive_timeout srv_at_estimate;
1178 /** reqs waiting for replies */
1179 struct ptlrpc_at_array srv_at_array;
1180 /** early reply timer */
1181 cfs_timer_t srv_at_timer;
1182 /** check early replies */
1183 unsigned srv_at_check;
1185 cfs_time_t srv_at_checktime;
1189 * serialize the following fields, used for processing
1190 * replies for this portal
1192 cfs_spinlock_t srv_rs_lock __cfs_cacheline_aligned;
1193 /** all the active replies */
1194 cfs_list_t srv_active_replies;
1196 /** replies waiting for service */
1197 cfs_list_t srv_reply_queue;
1199 /** List of free reply_states */
1200 cfs_list_t srv_free_rs_list;
1201 /** waitq to run, when adding stuff to srv_free_rs_list */
1202 cfs_waitq_t srv_free_rs_waitq;
1203 /** # 'difficult' replies */
1204 cfs_atomic_t srv_n_difficult_replies;
1205 //struct ptlrpc_srv_ni srv_interfaces[0];
1209 * Declaration of ptlrpcd control structure
1211 struct ptlrpcd_ctl {
1213 * Ptlrpc thread control flags (LIOD_START, LIOD_STOP, LIOD_FORCE)
1215 unsigned long pc_flags;
1217 * Thread lock protecting structure fields.
1219 cfs_spinlock_t pc_lock;
1223 cfs_completion_t pc_starting;
1227 cfs_completion_t pc_finishing;
1229 * Thread requests set.
1231 struct ptlrpc_request_set *pc_set;
1233 * Thread name used in cfs_daemonize()
1237 * Environment for request interpreters to run in.
1239 struct lu_env pc_env;
1242 * Async rpcs flag to make sure that ptlrpcd_check() is called only
1247 * Currently not used.
1251 * User-space async rpcs callback.
1253 void *pc_wait_callback;
1255 * User-space check idle rpcs callback.
1257 void *pc_idle_callback;
1261 /* Bits for pc_flags */
1262 enum ptlrpcd_ctl_flags {
1264 * Ptlrpc thread start flag.
1266 LIOD_START = 1 << 0,
1268 * Ptlrpc thread stop flag.
1272 * Ptlrpc thread force flag (only stop force so far).
1273 * This will cause aborting any inflight rpcs handled
1274 * by thread if LIOD_STOP is specified.
1276 LIOD_FORCE = 1 << 2,
1278 * This is a recovery ptlrpc thread.
1280 LIOD_RECOVERY = 1 << 3
1283 /* ptlrpc/events.c */
1284 extern lnet_handle_eq_t ptlrpc_eq_h;
1285 extern int ptlrpc_uuid_to_peer(struct obd_uuid *uuid,
1286 lnet_process_id_t *peer, lnet_nid_t *self);
1288 * These callbacks are invoked by LNet when something happened to
1292 extern void request_out_callback (lnet_event_t *ev);
1293 extern void reply_in_callback(lnet_event_t *ev);
1294 extern void client_bulk_callback (lnet_event_t *ev);
1295 extern void request_in_callback(lnet_event_t *ev);
1296 extern void reply_out_callback(lnet_event_t *ev);
1297 extern void server_bulk_callback (lnet_event_t *ev);
1300 /* ptlrpc/connection.c */
1301 struct ptlrpc_connection *ptlrpc_connection_get(lnet_process_id_t peer,
1303 struct obd_uuid *uuid);
1304 int ptlrpc_connection_put(struct ptlrpc_connection *c);
1305 struct ptlrpc_connection *ptlrpc_connection_addref(struct ptlrpc_connection *);
1306 int ptlrpc_connection_init(void);
1307 void ptlrpc_connection_fini(void);
1308 extern lnet_pid_t ptl_get_pid(void);
1310 /* ptlrpc/niobuf.c */
1312 * Actual interfacing with LNet to put/get/register/unregister stuff
1315 int ptlrpc_start_bulk_transfer(struct ptlrpc_bulk_desc *desc);
1316 void ptlrpc_abort_bulk(struct ptlrpc_bulk_desc *desc);
1317 int ptlrpc_register_bulk(struct ptlrpc_request *req);
1318 int ptlrpc_unregister_bulk(struct ptlrpc_request *req, int async);
1320 static inline int ptlrpc_server_bulk_active(struct ptlrpc_bulk_desc *desc)
1324 LASSERT(desc != NULL);
1326 cfs_spin_lock(&desc->bd_lock);
1327 rc = desc->bd_network_rw;
1328 cfs_spin_unlock(&desc->bd_lock);
1332 static inline int ptlrpc_client_bulk_active(struct ptlrpc_request *req)
1334 struct ptlrpc_bulk_desc *desc = req->rq_bulk;
1337 LASSERT(req != NULL);
1339 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_BULK_UNLINK) &&
1340 req->rq_bulk_deadline > cfs_time_current_sec())
1346 cfs_spin_lock(&desc->bd_lock);
1347 rc = desc->bd_network_rw;
1348 cfs_spin_unlock(&desc->bd_lock);
1352 #define PTLRPC_REPLY_MAYBE_DIFFICULT 0x01
1353 #define PTLRPC_REPLY_EARLY 0x02
1354 int ptlrpc_send_reply(struct ptlrpc_request *req, int flags);
1355 int ptlrpc_reply(struct ptlrpc_request *req);
1356 int ptlrpc_send_error(struct ptlrpc_request *req, int difficult);
1357 int ptlrpc_error(struct ptlrpc_request *req);
1358 void ptlrpc_resend_req(struct ptlrpc_request *request);
1359 int ptlrpc_at_get_net_latency(struct ptlrpc_request *req);
1360 int ptl_send_rpc(struct ptlrpc_request *request, int noreply);
1361 int ptlrpc_register_rqbd (struct ptlrpc_request_buffer_desc *rqbd);
1364 /* ptlrpc/client.c */
1366 * Client-side portals API. Everything to send requests, receive replies,
1367 * request queues, request management, etc.
1370 void ptlrpc_init_client(int req_portal, int rep_portal, char *name,
1371 struct ptlrpc_client *);
1372 void ptlrpc_cleanup_client(struct obd_import *imp);
1373 struct ptlrpc_connection *ptlrpc_uuid_to_connection(struct obd_uuid *uuid);
1375 int ptlrpc_queue_wait(struct ptlrpc_request *req);
1376 int ptlrpc_replay_req(struct ptlrpc_request *req);
1377 int ptlrpc_unregister_reply(struct ptlrpc_request *req, int async);
1378 void ptlrpc_restart_req(struct ptlrpc_request *req);
1379 void ptlrpc_abort_inflight(struct obd_import *imp);
1380 void ptlrpc_cleanup_imp(struct obd_import *imp);
1381 void ptlrpc_abort_set(struct ptlrpc_request_set *set);
1383 struct ptlrpc_request_set *ptlrpc_prep_set(void);
1384 int ptlrpc_set_add_cb(struct ptlrpc_request_set *set,
1385 set_interpreter_func fn, void *data);
1386 int ptlrpc_set_next_timeout(struct ptlrpc_request_set *);
1387 int ptlrpc_check_set(const struct lu_env *env, struct ptlrpc_request_set *set);
1388 int ptlrpc_set_wait(struct ptlrpc_request_set *);
1389 int ptlrpc_expired_set(void *data);
1390 void ptlrpc_interrupted_set(void *data);
1391 void ptlrpc_mark_interrupted(struct ptlrpc_request *req);
1392 void ptlrpc_set_destroy(struct ptlrpc_request_set *);
1393 void ptlrpc_set_add_req(struct ptlrpc_request_set *, struct ptlrpc_request *);
1394 int ptlrpc_set_add_new_req(struct ptlrpcd_ctl *pc,
1395 struct ptlrpc_request *req);
1397 void ptlrpc_free_rq_pool(struct ptlrpc_request_pool *pool);
1398 void ptlrpc_add_rqs_to_pool(struct ptlrpc_request_pool *pool, int num_rq);
1400 struct ptlrpc_request_pool *
1401 ptlrpc_init_rq_pool(int, int,
1402 void (*populate_pool)(struct ptlrpc_request_pool *, int));
1404 void ptlrpc_at_set_req_timeout(struct ptlrpc_request *req);
1405 struct ptlrpc_request *ptlrpc_request_alloc(struct obd_import *imp,
1406 const struct req_format *format);
1407 struct ptlrpc_request *ptlrpc_request_alloc_pool(struct obd_import *imp,
1408 struct ptlrpc_request_pool *,
1409 const struct req_format *format);
1410 void ptlrpc_request_free(struct ptlrpc_request *request);
1411 int ptlrpc_request_pack(struct ptlrpc_request *request,
1412 __u32 version, int opcode);
1413 struct ptlrpc_request *ptlrpc_request_alloc_pack(struct obd_import *imp,
1414 const struct req_format *format,
1415 __u32 version, int opcode);
1416 int ptlrpc_request_bufs_pack(struct ptlrpc_request *request,
1417 __u32 version, int opcode, char **bufs,
1418 struct ptlrpc_cli_ctx *ctx);
1419 struct ptlrpc_request *ptlrpc_prep_fakereq(struct obd_import *imp,
1420 unsigned int timeout,
1421 ptlrpc_interpterer_t interpreter);
1422 void ptlrpc_fakereq_finished(struct ptlrpc_request *req);
1424 struct ptlrpc_request *ptlrpc_prep_req(struct obd_import *imp, __u32 version,
1425 int opcode, int count, __u32 *lengths,
1427 struct ptlrpc_request *ptlrpc_prep_req_pool(struct obd_import *imp,
1428 __u32 version, int opcode,
1429 int count, __u32 *lengths, char **bufs,
1430 struct ptlrpc_request_pool *pool);
1431 void ptlrpc_req_finished(struct ptlrpc_request *request);
1432 void ptlrpc_req_finished_with_imp_lock(struct ptlrpc_request *request);
1433 struct ptlrpc_request *ptlrpc_request_addref(struct ptlrpc_request *req);
1434 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_imp (struct ptlrpc_request *req,
1435 int npages, int type, int portal);
1436 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_exp(struct ptlrpc_request *req,
1437 int npages, int type, int portal);
1438 void ptlrpc_free_bulk(struct ptlrpc_bulk_desc *bulk);
1439 void ptlrpc_prep_bulk_page(struct ptlrpc_bulk_desc *desc,
1440 cfs_page_t *page, int pageoffset, int len);
1441 void ptlrpc_retain_replayable_request(struct ptlrpc_request *req,
1442 struct obd_import *imp);
1443 __u64 ptlrpc_next_xid(void);
1444 __u64 ptlrpc_sample_next_xid(void);
1445 __u64 ptlrpc_req_xid(struct ptlrpc_request *request);
1449 struct ptlrpc_service_conf {
1452 int psc_max_req_size;
1453 int psc_max_reply_size;
1456 int psc_watchdog_factor;
1457 int psc_min_threads;
1458 int psc_max_threads;
1462 /* ptlrpc/service.c */
1464 * Server-side services API. Register/unregister service, request state
1465 * management, service thread management
1469 void ptlrpc_save_lock (struct ptlrpc_request *req,
1470 struct lustre_handle *lock, int mode, int no_ack);
1471 void ptlrpc_commit_replies(struct obd_export *exp);
1472 void ptlrpc_dispatch_difficult_reply (struct ptlrpc_reply_state *rs);
1473 void ptlrpc_schedule_difficult_reply (struct ptlrpc_reply_state *rs);
1474 struct ptlrpc_service *ptlrpc_init_svc_conf(struct ptlrpc_service_conf *c,
1475 svc_handler_t h, char *name,
1476 struct proc_dir_entry *proc_entry,
1477 svc_req_printfn_t prntfn,
1480 struct ptlrpc_service *ptlrpc_init_svc(int nbufs, int bufsize, int max_req_size,
1482 int req_portal, int rep_portal,
1483 int watchdog_factor,
1484 svc_handler_t, char *name,
1485 cfs_proc_dir_entry_t *proc_entry,
1487 int min_threads, int max_threads,
1488 char *threadname, __u32 ctx_tags,
1489 svc_hpreq_handler_t);
1490 void ptlrpc_stop_all_threads(struct ptlrpc_service *svc);
1492 int ptlrpc_start_threads(struct ptlrpc_service *svc);
1493 int ptlrpc_start_thread(struct ptlrpc_service *svc);
1494 int ptlrpc_unregister_service(struct ptlrpc_service *service);
1495 int liblustre_check_services (void *arg);
1496 void ptlrpc_daemonize(char *name);
1497 int ptlrpc_service_health_check(struct ptlrpc_service *);
1498 void ptlrpc_hpreq_reorder(struct ptlrpc_request *req);
1499 void ptlrpc_server_drop_request(struct ptlrpc_request *req);
1502 int ptlrpc_hr_init(void);
1503 void ptlrpc_hr_fini(void);
1505 # define ptlrpc_hr_init() (0)
1506 # define ptlrpc_hr_fini() do {} while(0)
1509 struct ptlrpc_svc_data {
1511 struct ptlrpc_service *svc;
1512 struct ptlrpc_thread *thread;
1516 /* ptlrpc/import.c */
1521 int ptlrpc_connect_import(struct obd_import *imp, char * new_uuid);
1522 int ptlrpc_init_import(struct obd_import *imp);
1523 int ptlrpc_disconnect_import(struct obd_import *imp, int noclose);
1524 int ptlrpc_import_recovery_state_machine(struct obd_import *imp);
1526 /* ptlrpc/pack_generic.c */
1527 int ptlrpc_reconnect_import(struct obd_import *imp);
1531 * ptlrpc msg buffer and swab interface
1535 int ptlrpc_buf_need_swab(struct ptlrpc_request *req, const int inout,
1537 void ptlrpc_buf_set_swabbed(struct ptlrpc_request *req, const int inout,
1539 int ptlrpc_unpack_rep_msg(struct ptlrpc_request *req, int len);
1540 int ptlrpc_unpack_req_msg(struct ptlrpc_request *req, int len);
1542 int lustre_msg_check_version(struct lustre_msg *msg, __u32 version);
1543 void lustre_init_msg_v2(struct lustre_msg_v2 *msg, int count, __u32 *lens,
1545 int lustre_pack_request(struct ptlrpc_request *, __u32 magic, int count,
1546 __u32 *lens, char **bufs);
1547 int lustre_pack_reply(struct ptlrpc_request *, int count, __u32 *lens,
1549 int lustre_pack_reply_v2(struct ptlrpc_request *req, int count,
1550 __u32 *lens, char **bufs, int flags);
1551 #define LPRFL_EARLY_REPLY 1
1552 int lustre_pack_reply_flags(struct ptlrpc_request *, int count, __u32 *lens,
1553 char **bufs, int flags);
1554 int lustre_shrink_msg(struct lustre_msg *msg, int segment,
1555 unsigned int newlen, int move_data);
1556 void lustre_free_reply_state(struct ptlrpc_reply_state *rs);
1557 int __lustre_unpack_msg(struct lustre_msg *m, int len);
1558 int lustre_msg_hdr_size(__u32 magic, int count);
1559 int lustre_msg_size(__u32 magic, int count, __u32 *lengths);
1560 int lustre_msg_size_v2(int count, __u32 *lengths);
1561 int lustre_packed_msg_size(struct lustre_msg *msg);
1562 int lustre_msg_early_size(void);
1563 void *lustre_msg_buf_v2(struct lustre_msg_v2 *m, int n, int min_size);
1564 void *lustre_msg_buf(struct lustre_msg *m, int n, int minlen);
1565 int lustre_msg_buflen(struct lustre_msg *m, int n);
1566 void lustre_msg_set_buflen(struct lustre_msg *m, int n, int len);
1567 int lustre_msg_bufcount(struct lustre_msg *m);
1568 char *lustre_msg_string (struct lustre_msg *m, int n, int max_len);
1569 __u32 lustre_msghdr_get_flags(struct lustre_msg *msg);
1570 void lustre_msghdr_set_flags(struct lustre_msg *msg, __u32 flags);
1571 __u32 lustre_msg_get_flags(struct lustre_msg *msg);
1572 void lustre_msg_add_flags(struct lustre_msg *msg, int flags);
1573 void lustre_msg_set_flags(struct lustre_msg *msg, int flags);
1574 void lustre_msg_clear_flags(struct lustre_msg *msg, int flags);
1575 __u32 lustre_msg_get_op_flags(struct lustre_msg *msg);
1576 void lustre_msg_add_op_flags(struct lustre_msg *msg, int flags);
1577 void lustre_msg_set_op_flags(struct lustre_msg *msg, int flags);
1578 struct lustre_handle *lustre_msg_get_handle(struct lustre_msg *msg);
1579 __u32 lustre_msg_get_type(struct lustre_msg *msg);
1580 __u32 lustre_msg_get_version(struct lustre_msg *msg);
1581 void lustre_msg_add_version(struct lustre_msg *msg, int version);
1582 __u32 lustre_msg_get_opc(struct lustre_msg *msg);
1583 __u64 lustre_msg_get_last_xid(struct lustre_msg *msg);
1584 __u64 lustre_msg_get_last_committed(struct lustre_msg *msg);
1585 __u64 *lustre_msg_get_versions(struct lustre_msg *msg);
1586 __u64 lustre_msg_get_transno(struct lustre_msg *msg);
1587 __u64 lustre_msg_get_slv(struct lustre_msg *msg);
1588 __u32 lustre_msg_get_limit(struct lustre_msg *msg);
1589 void lustre_msg_set_slv(struct lustre_msg *msg, __u64 slv);
1590 void lustre_msg_set_limit(struct lustre_msg *msg, __u64 limit);
1591 int lustre_msg_get_status(struct lustre_msg *msg);
1592 __u32 lustre_msg_get_conn_cnt(struct lustre_msg *msg);
1593 int lustre_msg_is_v1(struct lustre_msg *msg);
1594 __u32 lustre_msg_get_magic(struct lustre_msg *msg);
1595 __u32 lustre_msg_get_timeout(struct lustre_msg *msg);
1596 __u32 lustre_msg_get_service_time(struct lustre_msg *msg);
1597 __u32 lustre_msg_get_cksum(struct lustre_msg *msg);
1598 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 0, 0)
1599 __u32 lustre_msg_calc_cksum(struct lustre_msg *msg, int compat18);
1601 # warning "remove checksum compatibility support for b1_8"
1602 __u32 lustre_msg_calc_cksum(struct lustre_msg *msg);
1604 void lustre_msg_set_handle(struct lustre_msg *msg,struct lustre_handle *handle);
1605 void lustre_msg_set_type(struct lustre_msg *msg, __u32 type);
1606 void lustre_msg_set_opc(struct lustre_msg *msg, __u32 opc);
1607 void lustre_msg_set_last_xid(struct lustre_msg *msg, __u64 last_xid);
1608 void lustre_msg_set_last_committed(struct lustre_msg *msg,__u64 last_committed);
1609 void lustre_msg_set_versions(struct lustre_msg *msg, __u64 *versions);
1610 void lustre_msg_set_transno(struct lustre_msg *msg, __u64 transno);
1611 void lustre_msg_set_status(struct lustre_msg *msg, __u32 status);
1612 void lustre_msg_set_conn_cnt(struct lustre_msg *msg, __u32 conn_cnt);
1613 void ptlrpc_req_set_repsize(struct ptlrpc_request *req, int count, __u32 *sizes);
1614 void ptlrpc_request_set_replen(struct ptlrpc_request *req);
1615 void lustre_msg_set_timeout(struct lustre_msg *msg, __u32 timeout);
1616 void lustre_msg_set_service_time(struct lustre_msg *msg, __u32 service_time);
1617 void lustre_msg_set_cksum(struct lustre_msg *msg, __u32 cksum);
1620 lustre_shrink_reply(struct ptlrpc_request *req, int segment,
1621 unsigned int newlen, int move_data)
1623 LASSERT(req->rq_reply_state);
1624 LASSERT(req->rq_repmsg);
1625 req->rq_replen = lustre_shrink_msg(req->rq_repmsg, segment,
1630 /** Change request phase of \a req to \a new_phase */
1632 ptlrpc_rqphase_move(struct ptlrpc_request *req, enum rq_phase new_phase)
1634 if (req->rq_phase == new_phase)
1637 if (new_phase == RQ_PHASE_UNREGISTERING) {
1638 req->rq_next_phase = req->rq_phase;
1640 cfs_atomic_inc(&req->rq_import->imp_unregistering);
1643 if (req->rq_phase == RQ_PHASE_UNREGISTERING) {
1645 cfs_atomic_dec(&req->rq_import->imp_unregistering);
1648 DEBUG_REQ(D_INFO, req, "move req \"%s\" -> \"%s\"",
1649 ptlrpc_rqphase2str(req), ptlrpc_phase2str(new_phase));
1651 req->rq_phase = new_phase;
1655 * Returns true if request \a req got early reply and hard deadline is not met
1658 ptlrpc_client_early(struct ptlrpc_request *req)
1660 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
1661 req->rq_reply_deadline > cfs_time_current_sec())
1663 return req->rq_early;
1667 * Returns true if we got real reply from server for this request
1670 ptlrpc_client_replied(struct ptlrpc_request *req)
1672 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
1673 req->rq_reply_deadline > cfs_time_current_sec())
1675 return req->rq_replied;
1678 /** Returns true if request \a req is in process of receiving server reply */
1680 ptlrpc_client_recv(struct ptlrpc_request *req)
1682 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
1683 req->rq_reply_deadline > cfs_time_current_sec())
1685 return req->rq_receiving_reply;
1689 ptlrpc_client_recv_or_unlink(struct ptlrpc_request *req)
1693 cfs_spin_lock(&req->rq_lock);
1694 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
1695 req->rq_reply_deadline > cfs_time_current_sec()) {
1696 cfs_spin_unlock(&req->rq_lock);
1699 rc = req->rq_receiving_reply || req->rq_must_unlink;
1700 cfs_spin_unlock(&req->rq_lock);
1705 ptlrpc_client_wake_req(struct ptlrpc_request *req)
1707 if (req->rq_set == NULL)
1708 cfs_waitq_signal(&req->rq_reply_waitq);
1710 cfs_waitq_signal(&req->rq_set->set_waitq);
1714 ptlrpc_rs_addref(struct ptlrpc_reply_state *rs)
1716 LASSERT(cfs_atomic_read(&rs->rs_refcount) > 0);
1717 cfs_atomic_inc(&rs->rs_refcount);
1721 ptlrpc_rs_decref(struct ptlrpc_reply_state *rs)
1723 LASSERT(cfs_atomic_read(&rs->rs_refcount) > 0);
1724 if (cfs_atomic_dec_and_test(&rs->rs_refcount))
1725 lustre_free_reply_state(rs);
1728 /* Should only be called once per req */
1729 static inline void ptlrpc_req_drop_rs(struct ptlrpc_request *req)
1731 if (req->rq_reply_state == NULL)
1732 return; /* shouldn't occur */
1733 ptlrpc_rs_decref(req->rq_reply_state);
1734 req->rq_reply_state = NULL;
1735 req->rq_repmsg = NULL;
1738 static inline __u32 lustre_request_magic(struct ptlrpc_request *req)
1740 return lustre_msg_get_magic(req->rq_reqmsg);
1743 static inline int ptlrpc_req_get_repsize(struct ptlrpc_request *req)
1745 switch (req->rq_reqmsg->lm_magic) {
1746 case LUSTRE_MSG_MAGIC_V2:
1747 return req->rq_reqmsg->lm_repsize;
1749 LASSERTF(0, "incorrect message magic: %08x\n",
1750 req->rq_reqmsg->lm_magic);
1755 static inline int ptlrpc_send_limit_expired(struct ptlrpc_request *req)
1757 if (req->rq_delay_limit != 0 &&
1758 cfs_time_before(cfs_time_add(req->rq_queued_time,
1759 cfs_time_seconds(req->rq_delay_limit)),
1760 cfs_time_current())) {
1766 static inline int ptlrpc_no_resend(struct ptlrpc_request *req)
1768 if (!req->rq_no_resend && ptlrpc_send_limit_expired(req)) {
1769 cfs_spin_lock(&req->rq_lock);
1770 req->rq_no_resend = 1;
1771 cfs_spin_unlock(&req->rq_lock);
1773 return req->rq_no_resend;
1776 /* ldlm/ldlm_lib.c */
1778 * Target client logic
1781 int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg);
1782 int client_obd_cleanup(struct obd_device *obddev);
1783 int client_connect_import(const struct lu_env *env,
1784 struct obd_export **exp, struct obd_device *obd,
1785 struct obd_uuid *cluuid, struct obd_connect_data *,
1787 int client_disconnect_export(struct obd_export *exp);
1788 int client_import_add_conn(struct obd_import *imp, struct obd_uuid *uuid,
1790 int client_import_del_conn(struct obd_import *imp, struct obd_uuid *uuid);
1791 int client_import_find_conn(struct obd_import *imp, lnet_nid_t peer,
1792 struct obd_uuid *uuid);
1793 int import_set_conn_priority(struct obd_import *imp, struct obd_uuid *uuid);
1794 void client_destroy_import(struct obd_import *imp);
1797 int server_disconnect_export(struct obd_export *exp);
1799 /* ptlrpc/pinger.c */
1801 * Pinger API (client side only)
1804 enum timeout_event {
1807 struct timeout_item;
1808 typedef int (*timeout_cb_t)(struct timeout_item *, void *);
1809 int ptlrpc_pinger_add_import(struct obd_import *imp);
1810 int ptlrpc_pinger_del_import(struct obd_import *imp);
1811 int ptlrpc_add_timeout_client(int time, enum timeout_event event,
1812 timeout_cb_t cb, void *data,
1813 cfs_list_t *obd_list);
1814 int ptlrpc_del_timeout_client(cfs_list_t *obd_list,
1815 enum timeout_event event);
1816 struct ptlrpc_request * ptlrpc_prep_ping(struct obd_import *imp);
1817 int ptlrpc_obd_ping(struct obd_device *obd);
1818 cfs_time_t ptlrpc_suspend_wakeup_time(void);
1820 void ping_evictor_start(void);
1821 void ping_evictor_stop(void);
1823 #define ping_evictor_start() do {} while (0)
1824 #define ping_evictor_stop() do {} while (0)
1826 int ptlrpc_check_and_wait_suspend(struct ptlrpc_request *req);
1829 /* ptlrpc/ptlrpcd.c */
1832 * Ptlrpcd scope is a set of two threads: ptlrpcd-foo and ptlrpcd-foo-rcv,
1833 * these threads are used to asynchronously send requests queued with
1834 * ptlrpcd_add_req(req, PCSOPE_FOO), and to handle completion call-backs for
1835 * such requests. Multiple scopes are needed to avoid dead-locks.
1837 enum ptlrpcd_scope {
1838 /** Scope of bulk read-write rpcs. */
1840 /** Everything else. */
1845 int ptlrpcd_start(const char *name, struct ptlrpcd_ctl *pc);
1846 void ptlrpcd_stop(struct ptlrpcd_ctl *pc, int force);
1847 void ptlrpcd_wake(struct ptlrpc_request *req);
1848 int ptlrpcd_add_req(struct ptlrpc_request *req, enum ptlrpcd_scope scope);
1849 void ptlrpcd_add_rqset(struct ptlrpc_request_set *set);
1850 int ptlrpcd_addref(void);
1851 void ptlrpcd_decref(void);
1853 /* ptlrpc/lproc_ptlrpc.c */
1855 * procfs output related functions
1858 const char* ll_opcode2str(__u32 opcode);
1860 void ptlrpc_lprocfs_register_obd(struct obd_device *obd);
1861 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd);
1862 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes);
1864 static inline void ptlrpc_lprocfs_register_obd(struct obd_device *obd) {}
1865 static inline void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd) {}
1866 static inline void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes) {}
1870 /* ptlrpc/llog_server.c */
1871 int llog_origin_handle_create(struct ptlrpc_request *req);
1872 int llog_origin_handle_destroy(struct ptlrpc_request *req);
1873 int llog_origin_handle_prev_block(struct ptlrpc_request *req);
1874 int llog_origin_handle_next_block(struct ptlrpc_request *req);
1875 int llog_origin_handle_read_header(struct ptlrpc_request *req);
1876 int llog_origin_handle_close(struct ptlrpc_request *req);
1877 int llog_origin_handle_cancel(struct ptlrpc_request *req);
1878 int llog_catinfo(struct ptlrpc_request *req);
1880 /* ptlrpc/llog_client.c */
1881 extern struct llog_operations llog_client_ops;