1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
40 #if defined(__linux__)
41 #include <linux/lustre_net.h>
42 #elif defined(__APPLE__)
43 #include <darwin/lustre_net.h>
44 #elif defined(__WINNT__)
45 #include <winnt/lustre_net.h>
47 #error Unsupported operating system.
50 #include <libcfs/kp30.h>
52 #include <lnet/lnet.h>
53 #include <lustre/lustre_idl.h>
54 #include <lustre_ha.h>
55 #include <lustre_import.h>
56 #include <lprocfs_status.h>
58 #include <obd_support.h>
60 /* MD flags we _always_ use */
61 #define PTLRPC_MD_OPTIONS 0
63 /* Define maxima for bulk I/O
64 * CAVEAT EMPTOR, with multinet (i.e. routers forwarding between networks)
65 * these limits are system wide and not interface-local. */
66 #define PTLRPC_MAX_BRW_BITS LNET_MTU_BITS
67 #define PTLRPC_MAX_BRW_SIZE (1<<LNET_MTU_BITS)
68 #define PTLRPC_MAX_BRW_PAGES (PTLRPC_MAX_BRW_SIZE >> CFS_PAGE_SHIFT)
70 /* When PAGE_SIZE is a constant, we can check our arithmetic here with cpp! */
72 # if ((PTLRPC_MAX_BRW_PAGES & (PTLRPC_MAX_BRW_PAGES - 1)) != 0)
73 # error "PTLRPC_MAX_BRW_PAGES isn't a power of two"
75 # if (PTLRPC_MAX_BRW_SIZE != (PTLRPC_MAX_BRW_PAGES * CFS_PAGE_SIZE))
76 # error "PTLRPC_MAX_BRW_SIZE isn't PTLRPC_MAX_BRW_PAGES * CFS_PAGE_SIZE"
78 # if (PTLRPC_MAX_BRW_SIZE > LNET_MTU)
79 # error "PTLRPC_MAX_BRW_SIZE too big"
81 # if (PTLRPC_MAX_BRW_PAGES > LNET_MAX_IOV)
82 # error "PTLRPC_MAX_BRW_PAGES too big"
84 #endif /* __KERNEL__ */
86 /* Size over which to OBD_VMALLOC() rather than OBD_ALLOC() service request
88 #define SVC_BUF_VMALLOC_THRESHOLD (2 * CFS_PAGE_SIZE)
90 /* The following constants determine how memory is used to buffer incoming
93 * ?_NBUFS # buffers to allocate when growing the pool
94 * ?_BUFSIZE # bytes in a single request buffer
95 * ?_MAXREQSIZE # maximum request service will receive
97 * When fewer than ?_NBUFS/2 buffers are posted for receive, another chunk
98 * of ?_NBUFS is added to the pool.
100 * Messages larger than ?_MAXREQSIZE are dropped. Request buffers are
101 * considered full when less than ?_MAXREQSIZE is left in them.
104 #define LDLM_THREADS_AUTO_MIN \
105 min((int)(num_online_cpus() * num_online_cpus() * 2), 8)
106 #define LDLM_THREADS_AUTO_MAX (LDLM_THREADS_AUTO_MIN * 16)
107 #define LDLM_BL_THREADS LDLM_THREADS_AUTO_MIN
108 #define LDLM_NBUFS (64 * num_online_cpus())
109 #define LDLM_BUFSIZE (8 * 1024)
110 #define LDLM_MAXREQSIZE (5 * 1024)
111 #define LDLM_MAXREPSIZE (1024)
113 /* Absolute limits */
114 #define MDS_THREADS_MIN 2
115 #define MDS_THREADS_MAX 512
116 #define MDS_THREADS_MIN_READPAGE 2
117 #define MDS_NBUFS (64 * num_online_cpus())
118 #define MDS_BUFSIZE (8 * 1024)
119 /* Assume file name length = FNAME_MAX = 256 (true for ext3).
120 * path name length = PATH_MAX = 4096
121 * LOV MD size max = EA_MAX = 4000
122 * symlink: FNAME_MAX + PATH_MAX <- largest
123 * link: FNAME_MAX + PATH_MAX (mds_rec_link < mds_rec_create)
124 * rename: FNAME_MAX + FNAME_MAX
125 * open: FNAME_MAX + EA_MAX
127 * MDS_MAXREQSIZE ~= 4736 bytes =
128 * lustre_msg + ldlm_request + mds_body + mds_rec_create + FNAME_MAX + PATH_MAX
129 * MDS_MAXREPSIZE ~= 8300 bytes = lustre_msg + llog_header
130 * or, for mds_close() and mds_reint_unlink() on a many-OST filesystem:
131 * = 9210 bytes = lustre_msg + mds_body + 160 * (easize + cookiesize)
133 * Realistic size is about 512 bytes (20 character name + 128 char symlink),
134 * except in the open case where there are a large number of OSTs in a LOV.
136 #define MDS_MAXREQSIZE (5 * 1024)
137 #define MDS_MAXREPSIZE max(9 * 1024, 280 + LOV_MAX_STRIPE_COUNT * 56)
139 #define MGS_THREADS_AUTO_MIN 2
140 #define MGS_THREADS_AUTO_MAX 32
141 #define MGS_NBUFS (64 * num_online_cpus())
142 #define MGS_BUFSIZE (8 * 1024)
143 #define MGS_MAXREQSIZE (8 * 1024)
144 #define MGS_MAXREPSIZE (9 * 1024)
146 /* Absolute limits */
147 #define OSS_THREADS_MIN 3 /* difficult replies, HPQ, others */
148 #define OSS_THREADS_MAX 512
149 #define OST_NBUFS (64 * num_online_cpus())
150 #define OST_BUFSIZE (8 * 1024)
151 /* OST_MAXREQSIZE ~= 4768 bytes =
152 * lustre_msg + obdo + 16 * obd_ioobj + 256 * niobuf_remote
154 * - single object with 16 pages is 512 bytes
155 * - OST_MAXREQSIZE must be at least 1 page of cookies plus some spillover
157 #define OST_MAXREQSIZE (5 * 1024)
158 #define OST_MAXREPSIZE (9 * 1024)
160 /* Macro to hide a typecast. */
161 #define ptlrpc_req_async_args(req) ((void *)&req->rq_async_args)
163 struct ptlrpc_connection {
164 struct hlist_node c_hash;
166 lnet_process_id_t c_peer;
167 struct obd_uuid c_remote_uuid;
171 struct ptlrpc_client {
172 __u32 cli_request_portal;
173 __u32 cli_reply_portal;
177 /* state flags of requests */
178 /* XXX only ones left are those used by the bulk descs as well! */
179 #define PTL_RPC_FL_INTR (1 << 0) /* reply wait was interrupted by user */
180 #define PTL_RPC_FL_TIMEOUT (1 << 7) /* request timed out waiting for reply */
182 #define REQ_MAX_ACK_LOCKS 8
184 union ptlrpc_async_args {
185 /* Scratchpad for passing args to completion interpreter. Users
186 * cast to the struct of their choosing, and LASSERT that this is
187 * big enough. For _tons_ of context, OBD_ALLOC a struct and store
188 * a pointer to it here. The pointer_arg ensures this struct is at
189 * least big enough for that. */
190 void *pointer_arg[9];
194 struct ptlrpc_request_set;
195 typedef int (*set_interpreter_func)(struct ptlrpc_request_set *, void *, int);
197 struct ptlrpc_request_set {
198 int set_remaining; /* # uncompleted requests */
199 cfs_waitq_t set_waitq;
200 cfs_waitq_t *set_wakeup_ptr;
201 struct list_head set_requests;
202 struct list_head set_cblist; /* list of completion callbacks */
203 set_interpreter_func set_interpret; /* completion callback */
204 void *set_arg; /* completion context */
205 /* locked so that any old caller can communicate requests to
206 * the set holder who can then fold them into the lock-free set */
207 spinlock_t set_new_req_lock;
208 struct list_head set_new_requests;
211 struct ptlrpc_set_cbdata {
212 struct list_head psc_item;
213 set_interpreter_func psc_interpret;
217 struct ptlrpc_bulk_desc;
220 * ptlrpc callback & work item stuff
222 struct ptlrpc_cb_id {
223 void (*cbid_fn)(lnet_event_t *ev); /* specific callback fn */
224 void *cbid_arg; /* additional arg */
227 #define RS_MAX_LOCKS 4
230 struct ptlrpc_reply_state {
231 struct ptlrpc_cb_id rs_cb_id;
232 struct list_head rs_list;
233 struct list_head rs_exp_list;
234 struct list_head rs_obd_list;
236 struct list_head rs_debug_list;
238 /* updates to following flag serialised by srv_request_lock */
239 unsigned long rs_difficult:1; /* ACK/commit stuff */
240 unsigned long rs_scheduled:1; /* being handled? */
241 unsigned long rs_scheduled_ever:1;/* any schedule attempts? */
242 unsigned long rs_handled:1; /* been handled yet? */
243 unsigned long rs_on_net:1; /* reply_out_callback pending? */
244 unsigned long rs_prealloc:1; /* rs from prealloc list */
249 struct obd_export *rs_export;
250 struct ptlrpc_service *rs_service;
251 lnet_handle_md_t rs_md_h;
252 atomic_t rs_refcount;
254 /* locks awaiting client reply ACK */
256 struct lustre_handle rs_locks[RS_MAX_LOCKS];
257 ldlm_mode_t rs_modes[RS_MAX_LOCKS];
258 /* last member: variable sized reply message */
259 struct lustre_msg *rs_msg;
262 struct ptlrpc_thread;
265 RQ_PHASE_NEW = 0xebc0de00,
266 RQ_PHASE_RPC = 0xebc0de01,
267 RQ_PHASE_BULK = 0xebc0de02,
268 RQ_PHASE_INTERPRET = 0xebc0de03,
269 RQ_PHASE_COMPLETE = 0xebc0de04,
270 RQ_PHASE_UNREGISTERING = 0xebc0de05,
271 RQ_PHASE_UNDEFINED = 0xebc0de06
274 struct ptlrpc_request_pool {
276 struct list_head prp_req_list; /* list of ptlrpc_request structs */
278 void (*prp_populate)(struct ptlrpc_request_pool *, int);
283 struct ptlrpc_hpreq_ops {
285 * Check if the lock handle of the given lock is the same as
286 * taken from the request.
288 int (*hpreq_lock_match)(struct ptlrpc_request *, struct ldlm_lock *);
290 * Check if the request is a high priority one.
292 int (*hpreq_check)(struct ptlrpc_request *);
295 struct ptlrpc_request {
296 int rq_type; /* one of PTL_RPC_MSG_* */
297 struct list_head rq_list;
298 struct list_head rq_timed_list; /* server-side early replies */
299 struct list_head rq_history_list; /* server-side history */
300 struct list_head rq_exp_list; /* server-side per-export list */
301 struct ptlrpc_hpreq_ops *rq_ops; /* server-side hp handlers */
303 __u64 rq_history_seq; /* history sequence # */
306 /* client-side flags are serialized by rq_lock */
307 unsigned long rq_intr:1, rq_replied:1, rq_err:1,
308 rq_timedout:1, rq_resend:1, rq_restart:1,
310 * when ->rq_replay is set, request is kept by the client even
311 * after server commits corresponding transaction. This is
312 * used for operations that require sequence of multiple
313 * requests to be replayed. The only example currently is file
314 * open/close. When last request in such a sequence is
315 * committed, ->rq_replay is cleared on all requests in the
319 rq_no_resend:1, rq_waiting:1, rq_receiving_reply:1,
320 rq_no_delay:1, rq_net_err:1, rq_early:1, rq_must_unlink:1,
321 /* server-side flags */
322 rq_packed_final:1, /* packed final reply */
323 rq_sent_final:1, /* stop sending early replies */
324 rq_hp:1; /* high priority RPC */
325 enum rq_phase rq_phase; /* one of RQ_PHASE_* */
326 enum rq_phase rq_next_phase; /* one of RQ_PHASE_* to be used next */
327 atomic_t rq_refcount; /* client-side refcount for SENT race,
328 server-side refcounf for multiple replies */
330 struct ptlrpc_thread *rq_svc_thread; /* initial thread servicing req */
332 int rq_request_portal; /* XXX FIXME bug 249 */
333 int rq_reply_portal; /* XXX FIXME bug 249 */
335 int rq_nob_received; /* client-side # reply bytes actually received */
338 struct lustre_msg *rq_reqmsg;
341 struct lustre_msg *rq_repbuf; /* client only, buf may be bigger than msg */
342 struct lustre_msg *rq_repmsg;
345 struct list_head rq_replay_list;
347 __u32 rq_req_swab_mask;
348 __u32 rq_rep_swab_mask;
350 int rq_import_generation;
351 enum lustre_imp_state rq_send_state;
353 int rq_early_count; /* how many early replies (for stats) */
355 /* client+server request */
356 lnet_handle_md_t rq_req_md_h;
357 struct ptlrpc_cb_id rq_req_cbid;
360 struct timeval rq_arrival_time; /* request arrival time */
361 struct ptlrpc_reply_state *rq_reply_state; /* separated reply state */
362 struct ptlrpc_request_buffer_desc *rq_rqbd; /* incoming request buffer*/
364 __u32 rq_uid; /* peer uid, used in MDS only */
367 /* client-only incoming reply */
368 lnet_handle_md_t rq_reply_md_h;
369 cfs_waitq_t rq_reply_waitq;
370 struct ptlrpc_cb_id rq_reply_cbid;
373 lnet_process_id_t rq_peer;
374 struct obd_export *rq_export;
375 struct obd_import *rq_import;
377 void (*rq_replay_cb)(struct ptlrpc_request *);
378 void (*rq_commit_cb)(struct ptlrpc_request *);
381 struct ptlrpc_bulk_desc *rq_bulk; /* client side bulk */
382 /* client outgoing req */
383 time_t rq_sent; /* when request sent, seconds,
384 * or time when request should
386 volatile time_t rq_deadline; /* when request must finish. volatile
387 so that servers' early reply updates to the deadline aren't
388 kept in per-cpu cache */
389 time_t rq_reply_deadline; /* when req reply unlink must finish. */
390 time_t rq_bulk_deadline; /* when req bulk unlink must finish. */
391 int rq_timeout; /* service time estimate (secs) */
394 struct list_head rq_set_chain;
395 struct ptlrpc_request_set *rq_set;
396 void *rq_interpret_reply; /* Async completion handler */
397 union ptlrpc_async_args rq_async_args; /* Async completion context */
398 struct ptlrpc_request_pool *rq_pool; /* Pool if request from
402 static inline void lustre_set_req_swabbed(struct ptlrpc_request *req, int index)
404 LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
405 LASSERT((req->rq_req_swab_mask & (1 << index)) == 0);
406 req->rq_req_swab_mask |= 1 << index;
409 static inline void lustre_set_rep_swabbed(struct ptlrpc_request *req, int index)
411 LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
412 LASSERT((req->rq_rep_swab_mask & (1 << index)) == 0);
413 req->rq_rep_swab_mask |= 1 << index;
416 static inline int lustre_req_swabbed(struct ptlrpc_request *req, int index)
418 LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
419 return req->rq_req_swab_mask & (1 << index);
422 static inline int lustre_rep_swabbed(struct ptlrpc_request *req, int index)
424 LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
425 return req->rq_rep_swab_mask & (1 << index);
428 static inline int lustre_req_need_swab(struct ptlrpc_request *req)
430 return req->rq_req_swab_mask & (1 << MSG_PTLRPC_HEADER_OFF);
433 static inline int lustre_rep_need_swab(struct ptlrpc_request *req)
435 return req->rq_rep_swab_mask & (1 << MSG_PTLRPC_HEADER_OFF);
438 static inline const char *
439 ptlrpc_phase2str(enum rq_phase phase)
448 case RQ_PHASE_INTERPRET:
450 case RQ_PHASE_COMPLETE:
452 case RQ_PHASE_UNREGISTERING:
453 return "Unregistering";
459 static inline const char *
460 ptlrpc_rqphase2str(struct ptlrpc_request *req)
462 return ptlrpc_phase2str(req->rq_phase);
465 /* Spare the preprocessor, spoil the bugs. */
466 #define FLAG(field, str) (field ? str : "")
468 #define DEBUG_REQ_FLAGS(req) \
469 ptlrpc_rqphase2str(req), \
470 FLAG(req->rq_intr, "I"), FLAG(req->rq_replied, "R"), \
471 FLAG(req->rq_err, "E"), \
472 FLAG(req->rq_timedout, "X") /* eXpired */, FLAG(req->rq_resend, "S"), \
473 FLAG(req->rq_restart, "T"), FLAG(req->rq_replay, "P"), \
474 FLAG(req->rq_no_resend, "N"), \
475 FLAG(req->rq_waiting, "W"), FLAG(req->rq_hp, "H")
477 #define REQ_FLAGS_FMT "%s:%s%s%s%s%s%s%s%s%s%s"
479 void _debug_req(struct ptlrpc_request *req, __u32 mask,
480 struct libcfs_debug_msg_data *data, const char *fmt, ...)
481 __attribute__ ((format (printf, 4, 5)));
483 #define debug_req(cdls, level, req, file, func, line, fmt, a...) \
487 if (((level) & D_CANTMASK) != 0 || \
488 ((libcfs_debug & (level)) != 0 && \
489 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) { \
490 static struct libcfs_debug_msg_data _req_dbg_data = \
491 DEBUG_MSG_DATA_INIT(cdls, DEBUG_SUBSYSTEM, file, func, line); \
492 _debug_req((req), (level), &_req_dbg_data, fmt, ##a); \
496 /* for most callers (level is a constant) this is resolved at compile time */
497 #define DEBUG_REQ(level, req, fmt, args...) \
499 if ((level) & (D_ERROR | D_WARNING)) { \
500 static cfs_debug_limit_state_t cdls; \
501 debug_req(&cdls, level, req, __FILE__, __func__, __LINE__, \
502 "@@@ "fmt" ", ## args); \
504 debug_req(NULL, level, req, __FILE__, __func__, __LINE__, \
505 "@@@ "fmt" ", ## args); \
508 struct ptlrpc_bulk_page {
509 struct list_head bp_link;
511 int bp_pageoffset; /* offset within a page */
512 struct page *bp_page;
515 #define BULK_GET_SOURCE 0
516 #define BULK_PUT_SINK 1
517 #define BULK_GET_SINK 2
518 #define BULK_PUT_SOURCE 3
520 struct ptlrpc_bulk_desc {
521 unsigned long bd_success:1; /* completed successfully */
522 unsigned long bd_network_rw:1; /* accessible to the network */
523 unsigned long bd_type:2; /* {put,get}{source,sink} */
524 unsigned long bd_registered:1; /* client side */
525 spinlock_t bd_lock; /* serialise with callback */
526 int bd_import_generation;
527 struct obd_export *bd_export;
528 struct obd_import *bd_import;
530 struct ptlrpc_request *bd_req; /* associated request */
531 cfs_waitq_t bd_waitq; /* server side only WQ */
532 int bd_iov_count; /* # entries in bd_iov */
533 int bd_max_iov; /* allocated size of bd_iov */
534 int bd_nob; /* # bytes covered */
535 int bd_nob_transferred; /* # bytes GOT/PUT */
539 struct ptlrpc_cb_id bd_cbid; /* network callback info */
540 lnet_handle_md_t bd_md_h; /* associated MD */
541 lnet_nid_t bd_sender; /* stash event::sender */
543 #if defined(__KERNEL__)
544 lnet_kiov_t bd_iov[0];
546 lnet_md_iovec_t bd_iov[0];
550 struct ptlrpc_thread {
552 struct list_head t_link; /* active threads in svc->srv_threads */
554 void *t_data; /* thread-private data (preallocated memory) */
557 unsigned int t_id; /* service thread index, from ptlrpc_start_threads */
558 struct lc_watchdog *t_watchdog; /* put watchdog in the structure per
560 cfs_waitq_t t_ctl_waitq;
563 struct ptlrpc_request_buffer_desc {
564 struct list_head rqbd_list;
565 struct list_head rqbd_reqs;
566 struct ptlrpc_service *rqbd_service;
567 lnet_handle_md_t rqbd_md_h;
570 struct ptlrpc_cb_id rqbd_cbid;
571 struct ptlrpc_request rqbd_req;
574 typedef int (*svc_handler_t)(struct ptlrpc_request *req);
575 typedef void (*svcreq_printfn_t)(void *, struct ptlrpc_request *);
576 typedef int (*svc_hpreq_handler_t)(struct ptlrpc_request *);
578 #define PTLRPC_SVC_HP_RATIO 10
580 struct ptlrpc_service {
581 struct list_head srv_list; /* chain thru all services */
582 int srv_max_req_size; /* biggest request to receive */
583 int srv_max_reply_size; /* biggest reply to send */
584 int srv_buf_size; /* size of individual buffers */
585 int srv_nbuf_per_group; /* # buffers to allocate in 1 group */
586 int srv_nbufs; /* total # req buffer descs allocated */
587 int srv_threads_min; /* threads to start at SOW */
588 int srv_threads_max; /* thread upper limit */
589 int srv_threads_started; /* index of last started thread */
590 int srv_threads_running; /* # running threads */
591 int srv_n_difficult_replies; /* # 'difficult' replies */
592 int srv_n_active_reqs; /* # reqs being served */
593 int srv_n_hpreq; /* # HPreqs being served */
594 cfs_duration_t srv_rqbd_timeout; /* timeout before re-posting reqs, in tick */
595 int srv_watchdog_factor; /* soft watchdog timeout mutiplier */
596 unsigned srv_cpu_affinity:1; /* bind threads to CPUs */
597 unsigned srv_at_check:1; /* check early replies */
598 cfs_time_t srv_at_checktime; /* debug */
600 __u32 srv_req_portal;
601 __u32 srv_rep_portal;
604 struct adaptive_timeout srv_at_estimate;/* estimated rpc service time */
605 spinlock_t srv_at_lock;
606 struct list_head srv_at_list; /* reqs waiting for replies */
607 cfs_timer_t srv_at_timer; /* early reply timer */
609 int srv_n_queued_reqs; /* # reqs in either of the queues below */
610 int srv_hpreq_count; /* # hp requests handled */
611 int srv_hpreq_ratio; /* # hp per lp reqs to handle */
612 struct list_head srv_req_in_queue; /* incoming reqs */
613 struct list_head srv_request_queue; /* reqs waiting for service */
614 struct list_head srv_request_hpq; /* high priority queue */
616 struct list_head srv_request_history; /* request history */
617 __u64 srv_request_seq; /* next request sequence # */
618 __u64 srv_request_max_cull_seq; /* highest seq culled from history */
619 svcreq_printfn_t srv_request_history_print_fn; /* service-specific print fn */
621 struct list_head srv_idle_rqbds; /* request buffers to be reposted */
622 struct list_head srv_active_rqbds; /* req buffers receiving */
623 struct list_head srv_history_rqbds; /* request buffer history */
624 int srv_nrqbd_receiving; /* # posted request buffers */
625 int srv_n_history_rqbds; /* # request buffers in history */
626 int srv_max_history_rqbds;/* max # request buffers in history */
628 atomic_t srv_outstanding_replies;
629 struct list_head srv_active_replies; /* all the active replies */
630 struct list_head srv_reply_queue; /* replies waiting for service */
632 cfs_waitq_t srv_waitq; /* all threads sleep on this. This
633 * wait-queue is signalled when new
634 * incoming request arrives and when
635 * difficult reply has to be handled. */
637 struct list_head srv_threads; /* service thread list */
638 svc_handler_t srv_handler;
639 svc_hpreq_handler_t srv_hpreq_handler; /* hp request handler */
641 char *srv_name; /* only statically allocated strings here; we don't clean them */
642 char *srv_thread_name; /* only statically allocated strings here; we don't clean them */
646 cfs_proc_dir_entry_t *srv_procroot;
647 struct lprocfs_stats *srv_stats;
649 /* List of free reply_states */
650 struct list_head srv_free_rs_list;
651 /* waitq to run, when adding stuff to srv_free_rs_list */
652 cfs_waitq_t srv_free_rs_waitq;
655 * if non-NULL called during thread creation (ptlrpc_start_thread())
656 * to initialize service specific per-thread state.
658 int (*srv_init)(struct ptlrpc_thread *thread);
660 * if non-NULL called during thread shutdown (ptlrpc_main()) to
661 * destruct state created by ->srv_init().
663 void (*srv_done)(struct ptlrpc_thread *thread);
665 //struct ptlrpc_srv_ni srv_interfaces[0];
670 * Ptlrpc thread control flags (LIOD_START, LIOD_STOP, LIOD_FORCE)
672 unsigned long pc_flags;
674 * Thread lock protecting structure fields.
680 struct completion pc_starting;
684 struct completion pc_finishing;
686 * Thread requests set.
688 struct ptlrpc_request_set *pc_set;
690 * Thread name used in cfs_daemonize()
695 * Async rpcs flag to make sure that ptlrpcd_check() is called only
700 * Currently not used.
704 * User-space async rpcs callback.
706 void *pc_wait_callback;
708 * User-space check idle rpcs callback.
710 void *pc_idle_callback;
714 /* Bits for pc_flags */
715 enum ptlrpcd_ctl_flags {
717 * Ptlrpc thread start flag.
721 * Ptlrpc thread stop flag.
725 * Ptlrpc thread force flag (only stop force so far).
726 * This will cause aborting any inflight rpcs handled
727 * by thread if LIOD_STOP is specified.
732 /* ptlrpc/events.c */
733 extern lnet_handle_eq_t ptlrpc_eq_h;
734 extern int ptlrpc_uuid_to_peer(struct obd_uuid *uuid,
735 lnet_process_id_t *peer, lnet_nid_t *self);
736 extern void request_out_callback (lnet_event_t *ev);
737 extern void reply_in_callback(lnet_event_t *ev);
738 extern void client_bulk_callback (lnet_event_t *ev);
739 extern void request_in_callback(lnet_event_t *ev);
740 extern void reply_out_callback(lnet_event_t *ev);
741 extern void server_bulk_callback (lnet_event_t *ev);
743 /* ptlrpc/connection.c */
744 struct ptlrpc_connection *ptlrpc_connection_get(lnet_process_id_t peer,
746 struct obd_uuid *uuid);
747 int ptlrpc_connection_put(struct ptlrpc_connection *c);
748 struct ptlrpc_connection *ptlrpc_connection_addref(struct ptlrpc_connection *);
749 int ptlrpc_connection_init(void);
750 void ptlrpc_connection_fini(void);
751 extern lnet_pid_t ptl_get_pid(void);
753 /* ptlrpc/niobuf.c */
754 int ptlrpc_start_bulk_transfer(struct ptlrpc_bulk_desc *desc);
755 void ptlrpc_abort_bulk(struct ptlrpc_bulk_desc *desc);
756 int ptlrpc_register_bulk(struct ptlrpc_request *req);
757 int ptlrpc_unregister_bulk(struct ptlrpc_request *req, int async);
759 static inline int ptlrpc_server_bulk_active(struct ptlrpc_bulk_desc *desc)
763 LASSERT(desc != NULL);
765 spin_lock(&desc->bd_lock);
766 rc = desc->bd_network_rw;
767 spin_unlock(&desc->bd_lock);
771 static inline int ptlrpc_client_bulk_active(struct ptlrpc_request *req)
773 struct ptlrpc_bulk_desc *desc = req->rq_bulk;
776 LASSERT(req != NULL);
778 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_BULK_UNLINK) &&
779 req->rq_bulk_deadline > cfs_time_current_sec())
785 spin_lock(&desc->bd_lock);
786 rc = desc->bd_network_rw;
787 spin_unlock(&desc->bd_lock);
791 #define PTLRPC_REPLY_MAYBE_DIFFICULT 0x01
792 #define PTLRPC_REPLY_EARLY 0x02
793 int ptlrpc_send_reply(struct ptlrpc_request *req, int flags);
794 int ptlrpc_reply(struct ptlrpc_request *req);
795 int ptlrpc_send_error(struct ptlrpc_request *req, int difficult);
796 int ptlrpc_error(struct ptlrpc_request *req);
797 void ptlrpc_resend_req(struct ptlrpc_request *request);
798 int ptlrpc_at_get_net_latency(struct ptlrpc_request *req);
799 int ptl_send_rpc(struct ptlrpc_request *request, int noreply);
800 int ptlrpc_register_rqbd (struct ptlrpc_request_buffer_desc *rqbd);
802 /* ptlrpc/client.c */
803 void ptlrpc_init_client(int req_portal, int rep_portal, char *name,
804 struct ptlrpc_client *);
805 void ptlrpc_cleanup_client(struct obd_import *imp);
806 struct ptlrpc_connection *ptlrpc_uuid_to_connection(struct obd_uuid *uuid);
808 int ptlrpc_queue_wait(struct ptlrpc_request *req);
809 int ptlrpc_replay_req(struct ptlrpc_request *req);
810 int ptlrpc_unregister_reply(struct ptlrpc_request *req, int async);
811 void ptlrpc_restart_req(struct ptlrpc_request *req);
812 void ptlrpc_abort_inflight(struct obd_import *imp);
813 void ptlrpc_abort_set(struct ptlrpc_request_set *set);
815 struct ptlrpc_request_set *ptlrpc_prep_set(void);
816 int ptlrpc_set_add_cb(struct ptlrpc_request_set *set,
817 set_interpreter_func fn, void *data);
818 int ptlrpc_set_next_timeout(struct ptlrpc_request_set *);
819 int ptlrpc_check_set(struct ptlrpc_request_set *set);
820 int ptlrpc_set_wait(struct ptlrpc_request_set *);
821 int ptlrpc_expired_set(void *data);
822 void ptlrpc_interrupted_set(void *data);
823 void ptlrpc_mark_interrupted(struct ptlrpc_request *req);
824 void ptlrpc_set_destroy(struct ptlrpc_request_set *);
825 void ptlrpc_set_add_req(struct ptlrpc_request_set *, struct ptlrpc_request *);
826 int ptlrpc_set_add_new_req(struct ptlrpcd_ctl *pc,
827 struct ptlrpc_request *req);
829 void ptlrpc_free_rq_pool(struct ptlrpc_request_pool *pool);
830 void ptlrpc_add_rqs_to_pool(struct ptlrpc_request_pool *pool, int num_rq);
832 struct ptlrpc_request_pool *
833 ptlrpc_init_rq_pool(int, int,
834 void (*populate_pool)(struct ptlrpc_request_pool *, int));
836 void ptlrpc_at_set_req_timeout(struct ptlrpc_request *req);
837 struct ptlrpc_request *ptlrpc_prep_req(struct obd_import *imp, __u32 version,
838 int opcode, int count, __u32 *lengths,
840 struct ptlrpc_request *ptlrpc_prep_req_pool(struct obd_import *imp,
841 __u32 version, int opcode,
842 int count, __u32 *lengths, char **bufs,
843 struct ptlrpc_request_pool *pool);
844 void ptlrpc_req_finished(struct ptlrpc_request *request);
845 void ptlrpc_req_finished_with_imp_lock(struct ptlrpc_request *request);
846 struct ptlrpc_request *ptlrpc_request_addref(struct ptlrpc_request *req);
847 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_imp (struct ptlrpc_request *req,
848 int npages, int type, int portal);
849 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_exp(struct ptlrpc_request *req,
850 int npages, int type, int portal);
851 void ptlrpc_free_bulk(struct ptlrpc_bulk_desc *bulk);
852 void ptlrpc_prep_bulk_page(struct ptlrpc_bulk_desc *desc,
853 cfs_page_t *page, int pageoffset, int len);
854 void ptlrpc_retain_replayable_request(struct ptlrpc_request *req,
855 struct obd_import *imp);
856 __u64 ptlrpc_next_xid(void);
857 __u64 ptlrpc_sample_next_xid(void);
858 __u64 ptlrpc_req_xid(struct ptlrpc_request *request);
860 /* ptlrpc/service.c */
861 void ptlrpc_save_lock (struct ptlrpc_request *req,
862 struct lustre_handle *lock, int mode);
863 void ptlrpc_commit_replies (struct obd_export *exp);
864 void ptlrpc_schedule_difficult_reply (struct ptlrpc_reply_state *rs);
865 struct ptlrpc_service *ptlrpc_init_svc(int nbufs, int bufsize, int max_req_size,
867 int req_portal, int rep_portal,
869 svc_handler_t, char *name,
870 cfs_proc_dir_entry_t *proc_entry,
872 int min_threads, int max_threads,
873 char *threadname, svc_hpreq_handler_t);
874 void ptlrpc_stop_all_threads(struct ptlrpc_service *svc);
876 int ptlrpc_start_threads(struct obd_device *dev, struct ptlrpc_service *svc);
877 int ptlrpc_start_thread(struct obd_device *dev, struct ptlrpc_service *svc);
878 int ptlrpc_unregister_service(struct ptlrpc_service *service);
879 int liblustre_check_services (void *arg);
880 void ptlrpc_daemonize(char *name);
881 int ptlrpc_service_health_check(struct ptlrpc_service *);
882 void ptlrpc_hpreq_reorder(struct ptlrpc_request *req);
885 struct ptlrpc_svc_data {
887 struct ptlrpc_service *svc;
888 struct ptlrpc_thread *thread;
889 struct obd_device *dev;
892 /* ptlrpc/import.c */
893 int ptlrpc_connect_import(struct obd_import *imp, char * new_uuid);
894 int ptlrpc_init_import(struct obd_import *imp);
895 int ptlrpc_disconnect_import(struct obd_import *imp, int noclose);
896 int ptlrpc_import_recovery_state_machine(struct obd_import *imp);
897 void ptlrpc_import_setasync(struct obd_import *imp, int count);
898 int ptlrpc_reconnect_import(struct obd_import *imp);
900 /* ptlrpc/pack_generic.c */
901 int lustre_msg_swabbed(struct lustre_msg *msg);
902 int lustre_msg_check_version(struct lustre_msg *msg, __u32 version);
903 int lustre_pack_request(struct ptlrpc_request *, __u32 magic, int count,
904 __u32 *lens, char **bufs);
905 int lustre_pack_reply(struct ptlrpc_request *, int count, __u32 *lens,
907 #define LPRFL_EARLY_REPLY 1
908 int lustre_pack_reply_flags(struct ptlrpc_request *, int count, __u32 *lens,
909 char **bufs, int flags);
910 void lustre_shrink_reply(struct ptlrpc_request *req, int segment,
911 unsigned int newlen, int move_data);
912 void lustre_free_reply_state(struct ptlrpc_reply_state *rs);
913 int lustre_msg_size(__u32 magic, int count, __u32 *lengths);
914 int lustre_packed_msg_size(struct lustre_msg *msg);
915 int lustre_msg_early_size(struct ptlrpc_request *req);
916 int lustre_unpack_msg(struct lustre_msg *m, int len);
917 void *lustre_msg_buf(struct lustre_msg *m, int n, int minlen);
918 int lustre_msg_buflen(struct lustre_msg *m, int n);
919 void lustre_msg_set_buflen(struct lustre_msg *m, int n, int len);
920 int lustre_msg_bufcount(struct lustre_msg *m);
921 char *lustre_msg_string (struct lustre_msg *m, int n, int max_len);
922 void *lustre_swab_reqbuf(struct ptlrpc_request *req, int n, int minlen,
924 void *lustre_swab_repbuf(struct ptlrpc_request *req, int n, int minlen,
926 __u32 lustre_msghdr_get_flags(struct lustre_msg *msg);
927 void lustre_msghdr_set_flags(struct lustre_msg *msg, __u32 flags);
928 __u32 lustre_msg_get_flags(struct lustre_msg *msg);
929 void lustre_msg_add_flags(struct lustre_msg *msg, int flags);
930 void lustre_msg_set_flags(struct lustre_msg *msg, int flags);
931 void lustre_msg_clear_flags(struct lustre_msg *msg, int flags);
932 __u32 lustre_msg_get_op_flags(struct lustre_msg *msg);
933 void lustre_msg_add_op_flags(struct lustre_msg *msg, int flags);
934 void lustre_msg_set_op_flags(struct lustre_msg *msg, int flags);
935 struct lustre_handle *lustre_msg_get_handle(struct lustre_msg *msg);
936 __u32 lustre_msg_get_type(struct lustre_msg *msg);
937 __u32 lustre_msg_get_version(struct lustre_msg *msg);
938 void lustre_msg_add_version(struct lustre_msg *msg, int version);
939 __u32 lustre_msg_get_opc(struct lustre_msg *msg);
940 __u64 lustre_msg_get_last_xid(struct lustre_msg *msg);
941 __u64 lustre_msg_get_last_committed(struct lustre_msg *msg);
942 __u64 *lustre_msg_get_versions(struct lustre_msg *msg);
943 __u64 lustre_msg_get_transno(struct lustre_msg *msg);
944 __u64 lustre_msg_get_slv(struct lustre_msg *msg);
945 __u32 lustre_msg_get_limit(struct lustre_msg *msg);
946 void lustre_msg_set_slv(struct lustre_msg *msg, __u64 slv);
947 void lustre_msg_set_limit(struct lustre_msg *msg, __u64 limit);
948 int lustre_msg_get_status(struct lustre_msg *msg);
949 __u32 lustre_msg_get_conn_cnt(struct lustre_msg *msg);
950 int lustre_msg_is_v1(struct lustre_msg *msg);
951 __u32 lustre_msg_get_magic(struct lustre_msg *msg);
952 __u32 lustre_msg_get_timeout(struct lustre_msg *msg);
953 __u32 lustre_msg_get_service_time(struct lustre_msg *msg);
954 __u32 lustre_msg_get_cksum(struct lustre_msg *msg);
955 __u32 lustre_msg_calc_cksum(struct lustre_msg *msg);
956 void lustre_msg_set_handle(struct lustre_msg *msg,struct lustre_handle *handle);
957 void lustre_msg_set_type(struct lustre_msg *msg, __u32 type);
958 void lustre_msg_set_opc(struct lustre_msg *msg, __u32 opc);
959 void lustre_msg_set_last_xid(struct lustre_msg *msg, __u64 last_xid);
960 void lustre_msg_set_last_committed(struct lustre_msg *msg,__u64 last_committed);
961 void lustre_msg_set_versions(struct lustre_msg *msg, __u64 *versions);
962 void lustre_msg_set_transno(struct lustre_msg *msg, __u64 transno);
963 void lustre_msg_set_status(struct lustre_msg *msg, __u32 status);
964 void lustre_msg_set_conn_cnt(struct lustre_msg *msg, __u32 conn_cnt);
965 void lustre_msg_set_timeout(struct lustre_msg *msg, __u32 timeout);
966 void lustre_msg_set_service_time(struct lustre_msg *msg, __u32 service_time);
967 void lustre_msg_set_cksum(struct lustre_msg *msg, __u32 cksum);
970 ptlrpc_rqphase_move(struct ptlrpc_request *req, enum rq_phase new_phase)
972 if (req->rq_phase == new_phase)
975 if (new_phase == RQ_PHASE_UNREGISTERING) {
976 req->rq_next_phase = req->rq_phase;
978 atomic_inc(&req->rq_import->imp_unregistering);
981 if (req->rq_phase == RQ_PHASE_UNREGISTERING) {
983 atomic_dec(&req->rq_import->imp_unregistering);
986 DEBUG_REQ(D_RPCTRACE, req, "move req \"%s\" -> \"%s\"",
987 ptlrpc_rqphase2str(req), ptlrpc_phase2str(new_phase));
989 req->rq_phase = new_phase;
993 ptlrpc_client_early(struct ptlrpc_request *req)
995 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
996 req->rq_reply_deadline > cfs_time_current_sec())
998 return req->rq_early;
1002 ptlrpc_client_replied(struct ptlrpc_request *req)
1004 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
1005 req->rq_reply_deadline > cfs_time_current_sec())
1007 return req->rq_replied;
1011 ptlrpc_client_recv(struct ptlrpc_request *req)
1013 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
1014 req->rq_reply_deadline > cfs_time_current_sec())
1016 return req->rq_receiving_reply;
1020 ptlrpc_client_recv_or_unlink(struct ptlrpc_request *req)
1024 spin_lock(&req->rq_lock);
1025 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
1026 req->rq_reply_deadline > cfs_time_current_sec()) {
1027 spin_unlock(&req->rq_lock);
1030 rc = req->rq_receiving_reply || req->rq_must_unlink;
1031 spin_unlock(&req->rq_lock);
1036 ptlrpc_client_wake_req(struct ptlrpc_request *req)
1038 if (req->rq_set == NULL)
1039 cfs_waitq_signal(&req->rq_reply_waitq);
1041 cfs_waitq_signal(&req->rq_set->set_waitq);
1045 ptlrpc_rs_addref(struct ptlrpc_reply_state *rs)
1047 LASSERT(atomic_read(&rs->rs_refcount) > 0);
1048 atomic_inc(&rs->rs_refcount);
1052 ptlrpc_rs_decref(struct ptlrpc_reply_state *rs)
1054 LASSERT(atomic_read(&rs->rs_refcount) > 0);
1055 if (atomic_dec_and_test(&rs->rs_refcount))
1056 lustre_free_reply_state(rs);
1059 /* Should only be called once per req */
1060 static inline void ptlrpc_req_drop_rs(struct ptlrpc_request *req)
1062 if (req->rq_reply_state == NULL)
1063 return; /* shouldn't occur */
1064 ptlrpc_rs_decref(req->rq_reply_state);
1065 req->rq_reply_state = NULL;
1066 req->rq_repmsg = NULL;
1069 static inline __u32 lustre_request_magic(struct ptlrpc_request *req)
1071 return lustre_msg_get_magic(req->rq_reqmsg);
1074 static inline int ptlrpc_req_get_repsize(struct ptlrpc_request *req)
1076 switch (req->rq_reqmsg->lm_magic) {
1077 case LUSTRE_MSG_MAGIC_V1:
1078 CERROR("function not supported for lustre_msg V1!\n");
1080 case LUSTRE_MSG_MAGIC_V2:
1081 return req->rq_reqmsg->lm_repsize;
1083 LASSERTF(0, "incorrect message magic: %08x\n",
1084 req->rq_reqmsg->lm_magic);
1090 ptlrpc_req_set_repsize(struct ptlrpc_request *req, int count, __u32 *lens)
1092 int size = lustre_msg_size(req->rq_reqmsg->lm_magic, count, lens);
1094 req->rq_replen = size + lustre_msg_early_size(req);
1095 if (req->rq_reqmsg->lm_magic == LUSTRE_MSG_MAGIC_V2)
1096 req->rq_reqmsg->lm_repsize = size;
1099 /* ldlm/ldlm_lib.c */
1100 int client_obd_setup(struct obd_device *obddev, obd_count len, void *buf);
1101 int client_obd_cleanup(struct obd_device * obddev);
1102 int client_connect_import(struct lustre_handle *conn, struct obd_device *obd,
1103 struct obd_uuid *cluuid, struct obd_connect_data *,
1105 int client_disconnect_export(struct obd_export *exp);
1106 int client_import_add_conn(struct obd_import *imp, struct obd_uuid *uuid,
1108 int client_import_del_conn(struct obd_import *imp, struct obd_uuid *uuid);
1109 int import_set_conn_priority(struct obd_import *imp, struct obd_uuid *uuid);
1111 /* ptlrpc/pinger.c */
1112 int ptlrpc_pinger_add_import(struct obd_import *imp);
1113 int ptlrpc_pinger_del_import(struct obd_import *imp);
1114 struct ptlrpc_request * ptlrpc_prep_ping(struct obd_import *imp);
1115 int ptlrpc_obd_ping(struct obd_device *obd);
1117 void ping_evictor_start(void);
1118 void ping_evictor_stop(void);
1120 #define ping_evictor_start() do {} while (0)
1121 #define ping_evictor_stop() do {} while (0)
1124 /* ptlrpc/ptlrpcd.c */
1125 int ptlrpcd_start(char *name, struct ptlrpcd_ctl *pc);
1126 void ptlrpcd_stop(struct ptlrpcd_ctl *pc, int force);
1127 void ptlrpcd_wake(struct ptlrpc_request *req);
1128 void ptlrpcd_add_req(struct ptlrpc_request *req);
1129 int ptlrpcd_addref(void);
1130 void ptlrpcd_decref(void);
1132 /* ptlrpc/lproc_ptlrpc.c */
1133 const char* ll_opcode2str(__u32 opcode);
1135 void ptlrpc_lprocfs_register_obd(struct obd_device *obd);
1136 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd);
1137 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes);
1139 static inline void ptlrpc_lprocfs_register_obd(struct obd_device *obd) {}
1140 static inline void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd) {}
1141 static inline void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes) {}
1144 /* ptlrpc/llog_server.c */
1145 int llog_origin_handle_create(struct ptlrpc_request *req);
1146 int llog_origin_handle_destroy(struct ptlrpc_request *req);
1147 int llog_origin_handle_prev_block(struct ptlrpc_request *req);
1148 int llog_origin_handle_next_block(struct ptlrpc_request *req);
1149 int llog_origin_handle_read_header(struct ptlrpc_request *req);
1150 int llog_origin_handle_close(struct ptlrpc_request *req);
1151 int llog_origin_handle_cancel(struct ptlrpc_request *req);
1152 int llog_catinfo(struct ptlrpc_request *req);
1154 /* ptlrpc/llog_client.c */
1155 extern struct llog_operations llog_client_ops;