4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2010, 2015, Intel Corporation.
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 #include <linux/uio.h>
59 #include <libcfs/libcfs.h>
60 #include <lnet/nidstr.h>
62 #include <lustre/lustre_idl.h>
63 #include <lustre_ha.h>
64 #include <lustre_sec.h>
65 #include <lustre_import.h>
66 #include <lprocfs_status.h>
67 #include <lu_object.h>
68 #include <lustre_req_layout.h>
69 #include <obd_support.h>
70 #include <lustre_ver.h>
72 /* MD flags we _always_ use */
73 #define PTLRPC_MD_OPTIONS 0
76 * Max # of bulk operations in one request.
77 * In order for the client and server to properly negotiate the maximum
78 * possible transfer size, PTLRPC_BULK_OPS_COUNT must be a power-of-two
79 * value. The client is free to limit the actual RPC size for any bulk
80 * transfer via cl_max_pages_per_rpc to some non-power-of-two value. */
81 #define PTLRPC_BULK_OPS_BITS 2
82 #define PTLRPC_BULK_OPS_COUNT (1U << PTLRPC_BULK_OPS_BITS)
84 * PTLRPC_BULK_OPS_MASK is for the convenience of the client only, and
85 * should not be used on the server at all. Otherwise, it imposes a
86 * protocol limitation on the maximum RPC size that can be used by any
87 * RPC sent to that server in the future. Instead, the server should
88 * use the negotiated per-client ocd_brw_size to determine the bulk
90 #define PTLRPC_BULK_OPS_MASK (~((__u64)PTLRPC_BULK_OPS_COUNT - 1))
93 * Define maxima for bulk I/O.
95 * A single PTLRPC BRW request is sent via up to PTLRPC_BULK_OPS_COUNT
96 * of LNET_MTU sized RDMA transfers. Clients and servers negotiate the
97 * currently supported maximum between peers at connect via ocd_brw_size.
99 #define PTLRPC_MAX_BRW_BITS (LNET_MTU_BITS + PTLRPC_BULK_OPS_BITS)
100 #define PTLRPC_MAX_BRW_SIZE (1 << PTLRPC_MAX_BRW_BITS)
101 #define PTLRPC_MAX_BRW_PAGES (PTLRPC_MAX_BRW_SIZE >> PAGE_CACHE_SHIFT)
103 #define ONE_MB_BRW_SIZE (1 << LNET_MTU_BITS)
104 #define MD_MAX_BRW_SIZE (1 << LNET_MTU_BITS)
105 #define MD_MAX_BRW_PAGES (MD_MAX_BRW_SIZE >> PAGE_CACHE_SHIFT)
106 #define DT_MAX_BRW_SIZE PTLRPC_MAX_BRW_SIZE
107 #define DT_MAX_BRW_PAGES (DT_MAX_BRW_SIZE >> PAGE_CACHE_SHIFT)
108 #define OFD_MAX_BRW_SIZE (1 << LNET_MTU_BITS)
110 /* When PAGE_SIZE is a constant, we can check our arithmetic here with cpp! */
111 #if ((PTLRPC_MAX_BRW_PAGES & (PTLRPC_MAX_BRW_PAGES - 1)) != 0)
112 # error "PTLRPC_MAX_BRW_PAGES isn't a power of two"
114 #if (PTLRPC_MAX_BRW_SIZE != (PTLRPC_MAX_BRW_PAGES * PAGE_CACHE_SIZE))
115 # error "PTLRPC_MAX_BRW_SIZE isn't PTLRPC_MAX_BRW_PAGES * PAGE_CACHE_SIZE"
117 #if (PTLRPC_MAX_BRW_SIZE > LNET_MTU * PTLRPC_BULK_OPS_COUNT)
118 # error "PTLRPC_MAX_BRW_SIZE too big"
120 #if (PTLRPC_MAX_BRW_PAGES > LNET_MAX_IOV * PTLRPC_BULK_OPS_COUNT)
121 # error "PTLRPC_MAX_BRW_PAGES too big"
124 #define PTLRPC_NTHRS_INIT 2
129 * Constants determine how memory is used to buffer incoming service requests.
131 * ?_NBUFS # buffers to allocate when growing the pool
132 * ?_BUFSIZE # bytes in a single request buffer
133 * ?_MAXREQSIZE # maximum request service will receive
135 * When fewer than ?_NBUFS/2 buffers are posted for receive, another chunk
136 * of ?_NBUFS is added to the pool.
138 * Messages larger than ?_MAXREQSIZE are dropped. Request buffers are
139 * considered full when less than ?_MAXREQSIZE is left in them.
144 * Constants determine how threads are created for ptlrpc service.
146 * ?_NTHRS_INIT # threads to create for each service partition on
147 * initializing. If it's non-affinity service and
148 * there is only one partition, it's the overall #
149 * threads for the service while initializing.
150 * ?_NTHRS_BASE # threads should be created at least for each
151 * ptlrpc partition to keep the service healthy.
152 * It's the low-water mark of threads upper-limit
153 * for each partition.
154 * ?_THR_FACTOR # threads can be added on threads upper-limit for
155 * each CPU core. This factor is only for reference,
156 * we might decrease value of factor if number of cores
157 * per CPT is above a limit.
158 * ?_NTHRS_MAX # overall threads can be created for a service,
159 * it's a soft limit because if service is running
160 * on machine with hundreds of cores and tens of
161 * CPU partitions, we need to guarantee each partition
162 * has ?_NTHRS_BASE threads, which means total threads
163 * will be ?_NTHRS_BASE * number_of_cpts which can
164 * exceed ?_NTHRS_MAX.
168 * #define MDS_NTHRS_INIT 2
169 * #define MDS_NTHRS_BASE 64
170 * #define MDS_NTHRS_FACTOR 8
171 * #define MDS_NTHRS_MAX 1024
174 * ---------------------------------------------------------------------
175 * Server(A) has 16 cores, user configured it to 4 partitions so each
176 * partition has 4 cores, then actual number of service threads on each
178 * MDS_NTHRS_BASE(64) + cores(4) * MDS_NTHRS_FACTOR(8) = 96
180 * Total number of threads for the service is:
181 * 96 * partitions(4) = 384
184 * ---------------------------------------------------------------------
185 * Server(B) has 32 cores, user configured it to 4 partitions so each
186 * partition has 8 cores, then actual number of service threads on each
188 * MDS_NTHRS_BASE(64) + cores(8) * MDS_NTHRS_FACTOR(8) = 128
190 * Total number of threads for the service is:
191 * 128 * partitions(4) = 512
194 * ---------------------------------------------------------------------
195 * Server(B) has 96 cores, user configured it to 8 partitions so each
196 * partition has 12 cores, then actual number of service threads on each
198 * MDS_NTHRS_BASE(64) + cores(12) * MDS_NTHRS_FACTOR(8) = 160
200 * Total number of threads for the service is:
201 * 160 * partitions(8) = 1280
203 * However, it's above the soft limit MDS_NTHRS_MAX, so we choose this number
204 * as upper limit of threads number for each partition:
205 * MDS_NTHRS_MAX(1024) / partitions(8) = 128
208 * ---------------------------------------------------------------------
209 * Server(C) have a thousand of cores and user configured it to 32 partitions
210 * MDS_NTHRS_BASE(64) * 32 = 2048
212 * which is already above soft limit MDS_NTHRS_MAX(1024), but we still need
213 * to guarantee that each partition has at least MDS_NTHRS_BASE(64) threads
214 * to keep service healthy, so total number of threads will just be 2048.
216 * NB: we don't suggest to choose server with that many cores because backend
217 * filesystem itself, buffer cache, or underlying network stack might
218 * have some SMP scalability issues at that large scale.
220 * If user already has a fat machine with hundreds or thousands of cores,
221 * there are two choices for configuration:
222 * a) create CPU table from subset of all CPUs and run Lustre on
224 * b) bind service threads on a few partitions, see modparameters of
225 * MDS and OSS for details
227 * NB: these calculations (and examples below) are simplified to help
228 * understanding, the real implementation is a little more complex,
229 * please see ptlrpc_server_nthreads_check() for details.
234 * LDLM threads constants:
236 * Given 8 as factor and 24 as base threads number
239 * On 4-core machine we will have 24 + 8 * 4 = 56 threads.
242 * On 8-core machine with 2 partitions we will have 24 + 4 * 8 = 56
243 * threads for each partition and total threads number will be 112.
246 * On 64-core machine with 8 partitions we will need LDLM_NTHRS_BASE(24)
247 * threads for each partition to keep service healthy, so total threads
248 * number should be 24 * 8 = 192.
250 * So with these constants, threads number will be at the similar level
251 * of old versions, unless target machine has over a hundred cores
253 #define LDLM_THR_FACTOR 8
254 #define LDLM_NTHRS_INIT PTLRPC_NTHRS_INIT
255 #define LDLM_NTHRS_BASE 24
256 #define LDLM_NTHRS_MAX (num_online_cpus() == 1 ? 64 : 128)
258 #define LDLM_BL_THREADS LDLM_NTHRS_AUTO_INIT
259 #define LDLM_CLIENT_NBUFS 1
260 #define LDLM_SERVER_NBUFS 64
261 #define LDLM_BUFSIZE (8 * 1024)
262 #define LDLM_MAXREQSIZE (5 * 1024)
263 #define LDLM_MAXREPSIZE (1024)
266 * MDS threads constants:
268 * Please see examples in "Thread Constants", MDS threads number will be at
269 * the comparable level of old versions, unless the server has many cores.
271 #ifndef MDS_MAX_THREADS
272 #define MDS_MAX_THREADS 1024
273 #define MDS_MAX_OTHR_THREADS 256
275 #else /* MDS_MAX_THREADS */
276 #if MDS_MAX_THREADS < PTLRPC_NTHRS_INIT
277 #undef MDS_MAX_THREADS
278 #define MDS_MAX_THREADS PTLRPC_NTHRS_INIT
280 #define MDS_MAX_OTHR_THREADS max(PTLRPC_NTHRS_INIT, MDS_MAX_THREADS / 2)
283 /* default service */
284 #define MDS_THR_FACTOR 8
285 #define MDS_NTHRS_INIT PTLRPC_NTHRS_INIT
286 #define MDS_NTHRS_MAX MDS_MAX_THREADS
287 #define MDS_NTHRS_BASE min(64, MDS_NTHRS_MAX)
289 /* read-page service */
290 #define MDS_RDPG_THR_FACTOR 4
291 #define MDS_RDPG_NTHRS_INIT PTLRPC_NTHRS_INIT
292 #define MDS_RDPG_NTHRS_MAX MDS_MAX_OTHR_THREADS
293 #define MDS_RDPG_NTHRS_BASE min(48, MDS_RDPG_NTHRS_MAX)
295 /* these should be removed when we remove setattr service in the future */
296 #define MDS_SETA_THR_FACTOR 4
297 #define MDS_SETA_NTHRS_INIT PTLRPC_NTHRS_INIT
298 #define MDS_SETA_NTHRS_MAX MDS_MAX_OTHR_THREADS
299 #define MDS_SETA_NTHRS_BASE min(48, MDS_SETA_NTHRS_MAX)
301 /* non-affinity threads */
302 #define MDS_OTHR_NTHRS_INIT PTLRPC_NTHRS_INIT
303 #define MDS_OTHR_NTHRS_MAX MDS_MAX_OTHR_THREADS
308 * Assume file name length = FNAME_MAX = 256 (true for ext3).
309 * path name length = PATH_MAX = 4096
310 * LOV MD size max = EA_MAX = 24 * 2000
311 * (NB: 24 is size of lov_ost_data)
312 * LOV LOGCOOKIE size max = 32 * 2000
313 * (NB: 32 is size of llog_cookie)
314 * symlink: FNAME_MAX + PATH_MAX <- largest
315 * link: FNAME_MAX + PATH_MAX (mds_rec_link < mds_rec_create)
316 * rename: FNAME_MAX + FNAME_MAX
317 * open: FNAME_MAX + EA_MAX
319 * MDS_MAXREQSIZE ~= 4736 bytes =
320 * lustre_msg + ldlm_request + mdt_body + mds_rec_create + FNAME_MAX + PATH_MAX
321 * MDS_MAXREPSIZE ~= 8300 bytes = lustre_msg + llog_header
323 * Realistic size is about 512 bytes (20 character name + 128 char symlink),
324 * except in the open case where there are a large number of OSTs in a LOV.
326 #define MDS_MAXREQSIZE (5 * 1024) /* >= 4736 */
327 #define MDS_MAXREPSIZE (9 * 1024) /* >= 8300 */
330 * MDS incoming request with LOV EA
331 * 24 = sizeof(struct lov_ost_data), i.e: replay of opencreate
333 #define MDS_LOV_MAXREQSIZE max(MDS_MAXREQSIZE, \
334 362 + LOV_MAX_STRIPE_COUNT * 24)
336 * MDS outgoing reply with LOV EA
338 * NB: max reply size Lustre 2.4+ client can get from old MDS is:
339 * LOV_MAX_STRIPE_COUNT * (llog_cookie + lov_ost_data) + extra bytes
341 * but 2.4 or later MDS will never send reply with llog_cookie to any
342 * version client. This macro is defined for server side reply buffer size.
344 #define MDS_LOV_MAXREPSIZE MDS_LOV_MAXREQSIZE
347 * This is the size of a maximum REINT_SETXATTR request:
349 * lustre_msg 56 (32 + 4 x 5 + 4)
351 * mdt_rec_setxattr 136
353 * name 256 (XATTR_NAME_MAX)
354 * value 65536 (XATTR_SIZE_MAX)
356 #define MDS_EA_MAXREQSIZE 66288
359 * These are the maximum request and reply sizes (rounded up to 1 KB
360 * boundaries) for the "regular" MDS_REQUEST_PORTAL and MDS_REPLY_PORTAL.
362 #define MDS_REG_MAXREQSIZE (((max(MDS_EA_MAXREQSIZE, \
363 MDS_LOV_MAXREQSIZE) + 1023) >> 10) << 10)
364 #define MDS_REG_MAXREPSIZE MDS_REG_MAXREQSIZE
367 * The update request includes all of updates from the create, which might
368 * include linkea (4K maxim), together with other updates, we set it to 1000K:
369 * lustre_msg + ptlrpc_body + OUT_UPDATE_BUFFER_SIZE_MAX
371 #define OUT_MAXREQSIZE (1000 * 1024)
372 #define OUT_MAXREPSIZE MDS_MAXREPSIZE
374 /** MDS_BUFSIZE = max_reqsize (w/o LOV EA) + max sptlrpc payload size */
375 #define MDS_BUFSIZE max(MDS_MAXREQSIZE + SPTLRPC_MAX_PAYLOAD, \
379 * MDS_REG_BUFSIZE should at least be MDS_REG_MAXREQSIZE + SPTLRPC_MAX_PAYLOAD.
380 * However, we need to allocate a much larger buffer for it because LNet
381 * requires each MD(rqbd) has at least MDS_REQ_MAXREQSIZE bytes left to avoid
382 * dropping of maximum-sized incoming request. So if MDS_REG_BUFSIZE is only a
383 * little larger than MDS_REG_MAXREQSIZE, then it can only fit in one request
384 * even there are about MDS_REG_MAX_REQSIZE bytes left in a rqbd, and memory
385 * utilization is very low.
387 * In the meanwhile, size of rqbd can't be too large, because rqbd can't be
388 * reused until all requests fit in it have been processed and released,
389 * which means one long blocked request can prevent the rqbd be reused.
390 * Now we set request buffer size to 160 KB, so even each rqbd is unlinked
391 * from LNet with unused 65 KB, buffer utilization will be about 59%.
392 * Please check LU-2432 for details.
394 #define MDS_REG_BUFSIZE max(MDS_REG_MAXREQSIZE + SPTLRPC_MAX_PAYLOAD, \
398 * OUT_BUFSIZE = max_out_reqsize + max sptlrpc payload (~1K) which is
399 * about 10K, for the same reason as MDS_REG_BUFSIZE, we also give some
400 * extra bytes to each request buffer to improve buffer utilization rate.
402 #define OUT_BUFSIZE max(OUT_MAXREQSIZE + SPTLRPC_MAX_PAYLOAD, \
405 /** FLD_MAXREQSIZE == lustre_msg + __u32 padding + ptlrpc_body + opc */
406 #define FLD_MAXREQSIZE (160)
408 /** FLD_MAXREPSIZE == lustre_msg + ptlrpc_body */
409 #define FLD_MAXREPSIZE (152)
410 #define FLD_BUFSIZE (1 << 12)
413 * SEQ_MAXREQSIZE == lustre_msg + __u32 padding + ptlrpc_body + opc + lu_range +
415 #define SEQ_MAXREQSIZE (160)
417 /** SEQ_MAXREPSIZE == lustre_msg + ptlrpc_body + lu_range */
418 #define SEQ_MAXREPSIZE (152)
419 #define SEQ_BUFSIZE (1 << 12)
421 /** MGS threads must be >= 3, see bug 22458 comment #28 */
422 #define MGS_NTHRS_INIT (PTLRPC_NTHRS_INIT + 1)
423 #define MGS_NTHRS_MAX 32
426 #define MGS_BUFSIZE (8 * 1024)
427 #define MGS_MAXREQSIZE (7 * 1024)
428 #define MGS_MAXREPSIZE (9 * 1024)
431 * OSS threads constants:
433 * Given 8 as factor and 64 as base threads number
436 * On 8-core server configured to 2 partitions, we will have
437 * 64 + 8 * 4 = 96 threads for each partition, 192 total threads.
440 * On 32-core machine configured to 4 partitions, we will have
441 * 64 + 8 * 8 = 112 threads for each partition, so total threads number
442 * will be 112 * 4 = 448.
445 * On 64-core machine configured to 4 partitions, we will have
446 * 64 + 16 * 8 = 192 threads for each partition, so total threads number
447 * will be 192 * 4 = 768 which is above limit OSS_NTHRS_MAX(512), so we
448 * cut off the value to OSS_NTHRS_MAX(512) / 4 which is 128 threads
449 * for each partition.
451 * So we can see that with these constants, threads number wil be at the
452 * similar level of old versions, unless the server has many cores.
454 /* depress threads factor for VM with small memory size */
455 #define OSS_THR_FACTOR min_t(int, 8, \
456 NUM_CACHEPAGES >> (28 - PAGE_CACHE_SHIFT))
457 #define OSS_NTHRS_INIT (PTLRPC_NTHRS_INIT + 1)
458 #define OSS_NTHRS_BASE 64
460 /* threads for handling "create" request */
461 #define OSS_CR_THR_FACTOR 1
462 #define OSS_CR_NTHRS_INIT PTLRPC_NTHRS_INIT
463 #define OSS_CR_NTHRS_BASE 8
464 #define OSS_CR_NTHRS_MAX 64
467 * OST_IO_MAXREQSIZE ~=
468 * lustre_msg + ptlrpc_body + obdo + obd_ioobj +
469 * DT_MAX_BRW_PAGES * niobuf_remote
471 * - single object with 16 pages is 512 bytes
472 * - OST_IO_MAXREQSIZE must be at least 1 page of cookies plus some spillover
473 * - Must be a multiple of 1024
474 * - actual size is about 18K
476 #define _OST_MAXREQSIZE_SUM (sizeof(struct lustre_msg) + \
477 sizeof(struct ptlrpc_body) + \
478 sizeof(struct obdo) + \
479 sizeof(struct obd_ioobj) + \
480 sizeof(struct niobuf_remote) * DT_MAX_BRW_PAGES)
482 * FIEMAP request can be 4K+ for now
484 #define OST_MAXREQSIZE (16 * 1024)
485 #define OST_IO_MAXREQSIZE max_t(int, OST_MAXREQSIZE, \
486 (((_OST_MAXREQSIZE_SUM - 1) | (1024 - 1)) + 1))
488 #define OST_MAXREPSIZE (9 * 1024)
489 #define OST_IO_MAXREPSIZE OST_MAXREPSIZE
492 /** OST_BUFSIZE = max_reqsize + max sptlrpc payload size */
493 #define OST_BUFSIZE max_t(int, OST_MAXREQSIZE + 1024, 16 * 1024)
495 * OST_IO_MAXREQSIZE is 18K, giving extra 46K can increase buffer utilization
496 * rate of request buffer, please check comment of MDS_LOV_BUFSIZE for details.
498 #define OST_IO_BUFSIZE max_t(int, OST_IO_MAXREQSIZE + 1024, 64 * 1024)
500 /* Macro to hide a typecast. */
501 #define ptlrpc_req_async_args(req) ((void *)&req->rq_async_args)
503 struct ptlrpc_replay_async_args {
509 * Structure to single define portal connection.
511 struct ptlrpc_connection {
512 /** linkage for connections hash table */
513 struct hlist_node c_hash;
514 /** Our own lnet nid for this connection */
516 /** Remote side nid for this connection */
517 lnet_process_id_t c_peer;
518 /** UUID of the other side */
519 struct obd_uuid c_remote_uuid;
520 /** reference counter for this connection */
524 /** Client definition for PortalRPC */
525 struct ptlrpc_client {
526 /** What lnet portal does this client send messages to by default */
527 __u32 cli_request_portal;
528 /** What portal do we expect replies on */
529 __u32 cli_reply_portal;
530 /** Name of the client */
534 /** state flags of requests */
535 /* XXX only ones left are those used by the bulk descs as well! */
536 #define PTL_RPC_FL_INTR (1 << 0) /* reply wait was interrupted by user */
537 #define PTL_RPC_FL_TIMEOUT (1 << 7) /* request timed out waiting for reply */
539 #define REQ_MAX_ACK_LOCKS 8
541 union ptlrpc_async_args {
543 * Scratchpad for passing args to completion interpreter. Users
544 * cast to the struct of their choosing, and CLASSERT that this is
545 * big enough. For _tons_ of context, OBD_ALLOC a struct and store
546 * a pointer to it here. The pointer_arg ensures this struct is at
547 * least big enough for that.
549 void *pointer_arg[11];
553 struct ptlrpc_request_set;
554 typedef int (*set_interpreter_func)(struct ptlrpc_request_set *, void *, int);
555 typedef int (*set_producer_func)(struct ptlrpc_request_set *, void *);
558 * Definition of request set structure.
559 * Request set is a list of requests (not necessary to the same target) that
560 * once populated with RPCs could be sent in parallel.
561 * There are two kinds of request sets. General purpose and with dedicated
562 * serving thread. Example of the latter is ptlrpcd set.
563 * For general purpose sets once request set started sending it is impossible
564 * to add new requests to such set.
565 * Provides a way to call "completion callbacks" when all requests in the set
568 struct ptlrpc_request_set {
569 atomic_t set_refcount;
570 /** number of in queue requests */
571 atomic_t set_new_count;
572 /** number of uncompleted requests */
573 atomic_t set_remaining;
574 /** wait queue to wait on for request events */
575 wait_queue_head_t set_waitq;
576 wait_queue_head_t *set_wakeup_ptr;
577 /** List of requests in the set */
578 struct list_head set_requests;
580 * List of completion callbacks to be called when the set is completed
581 * This is only used if \a set_interpret is NULL.
582 * Links struct ptlrpc_set_cbdata.
584 struct list_head set_cblist;
585 /** Completion callback, if only one. */
586 set_interpreter_func set_interpret;
587 /** opaq argument passed to completion \a set_interpret callback. */
590 * Lock for \a set_new_requests manipulations
591 * locked so that any old caller can communicate requests to
592 * the set holder who can then fold them into the lock-free set
594 spinlock_t set_new_req_lock;
595 /** List of new yet unsent requests. Only used with ptlrpcd now. */
596 struct list_head set_new_requests;
598 /** rq_status of requests that have been freed already */
600 /** Additional fields used by the flow control extension */
601 /** Maximum number of RPCs in flight */
602 int set_max_inflight;
603 /** Callback function used to generate RPCs */
604 set_producer_func set_producer;
605 /** opaq argument passed to the producer callback */
606 void *set_producer_arg;
607 unsigned int set_allow_intr:1;
611 * Description of a single ptrlrpc_set callback
613 struct ptlrpc_set_cbdata {
614 /** List linkage item */
615 struct list_head psc_item;
616 /** Pointer to interpreting function */
617 set_interpreter_func psc_interpret;
618 /** Opaq argument to pass to the callback */
622 struct ptlrpc_bulk_desc;
623 struct ptlrpc_service_part;
624 struct ptlrpc_service;
627 * ptlrpc callback & work item stuff
629 struct ptlrpc_cb_id {
630 void (*cbid_fn)(lnet_event_t *ev); /* specific callback fn */
631 void *cbid_arg; /* additional arg */
634 /** Maximum number of locks to fit into reply state */
635 #define RS_MAX_LOCKS 8
639 * Structure to define reply state on the server
640 * Reply state holds various reply message information. Also for "difficult"
641 * replies (rep-ack case) we store the state after sending reply and wait
642 * for the client to acknowledge the reception. In these cases locks could be
643 * added to the state for replay/failover consistency guarantees.
645 struct ptlrpc_reply_state {
646 /** Callback description */
647 struct ptlrpc_cb_id rs_cb_id;
648 /** Linkage for list of all reply states in a system */
649 struct list_head rs_list;
650 /** Linkage for list of all reply states on same export */
651 struct list_head rs_exp_list;
652 /** Linkage for list of all reply states for same obd */
653 struct list_head rs_obd_list;
655 struct list_head rs_debug_list;
657 /** A spinlock to protect the reply state flags */
659 /** Reply state flags */
660 unsigned long rs_difficult:1; /* ACK/commit stuff */
661 unsigned long rs_no_ack:1; /* no ACK, even for
662 difficult requests */
663 unsigned long rs_scheduled:1; /* being handled? */
664 unsigned long rs_scheduled_ever:1;/* any schedule attempts? */
665 unsigned long rs_handled:1; /* been handled yet? */
666 unsigned long rs_on_net:1; /* reply_out_callback pending? */
667 unsigned long rs_prealloc:1; /* rs from prealloc list */
668 unsigned long rs_committed:1;/* the transaction was committed
669 and the rs was dispatched
670 by ptlrpc_commit_replies */
671 atomic_t rs_refcount; /* number of users */
672 /** Number of locks awaiting client ACK */
675 /** Size of the state */
679 /** Transaction number */
683 struct obd_export *rs_export;
684 struct ptlrpc_service_part *rs_svcpt;
685 /** Lnet metadata handle for the reply */
686 lnet_handle_md_t rs_md_h;
688 /** Context for the sevice thread */
689 struct ptlrpc_svc_ctx *rs_svc_ctx;
690 /** Reply buffer (actually sent to the client), encoded if needed */
691 struct lustre_msg *rs_repbuf; /* wrapper */
692 /** Size of the reply buffer */
693 int rs_repbuf_len; /* wrapper buf length */
694 /** Size of the reply message */
695 int rs_repdata_len; /* wrapper msg length */
697 * Actual reply message. Its content is encrupted (if needed) to
698 * produce reply buffer for actual sending. In simple case
699 * of no network encryption we jus set \a rs_repbuf to \a rs_msg
701 struct lustre_msg *rs_msg; /* reply message */
703 /** Handles of locks awaiting client reply ACK */
704 struct lustre_handle rs_locks[RS_MAX_LOCKS];
705 /** Lock modes of locks in \a rs_locks */
706 enum ldlm_mode rs_modes[RS_MAX_LOCKS];
709 struct ptlrpc_thread;
713 RQ_PHASE_NEW = 0xebc0de00,
714 RQ_PHASE_RPC = 0xebc0de01,
715 RQ_PHASE_BULK = 0xebc0de02,
716 RQ_PHASE_INTERPRET = 0xebc0de03,
717 RQ_PHASE_COMPLETE = 0xebc0de04,
718 RQ_PHASE_UNREGISTERING = 0xebc0de05,
719 RQ_PHASE_UNDEFINED = 0xebc0de06
722 /** Type of request interpreter call-back */
723 typedef int (*ptlrpc_interpterer_t)(const struct lu_env *env,
724 struct ptlrpc_request *req,
726 /** Type of request resend call-back */
727 typedef void (*ptlrpc_resend_cb_t)(struct ptlrpc_request *req,
731 * Definition of request pool structure.
732 * The pool is used to store empty preallocated requests for the case
733 * when we would actually need to send something without performing
734 * any allocations (to avoid e.g. OOM).
736 struct ptlrpc_request_pool {
737 /** Locks the list */
739 /** list of ptlrpc_request structs */
740 struct list_head prp_req_list;
741 /** Maximum message size that would fit into a rquest from this pool */
743 /** Function to allocate more requests for this pool */
744 int (*prp_populate)(struct ptlrpc_request_pool *, int);
752 #include <lustre_nrs.h>
755 * Basic request prioritization operations structure.
756 * The whole idea is centered around locks and RPCs that might affect locks.
757 * When a lock is contended we try to give priority to RPCs that might lead
758 * to fastest release of that lock.
759 * Currently only implemented for OSTs only in a way that makes all
760 * IO and truncate RPCs that are coming from a locked region where a lock is
761 * contended a priority over other requests.
763 struct ptlrpc_hpreq_ops {
765 * Check if the lock handle of the given lock is the same as
766 * taken from the request.
768 int (*hpreq_lock_match)(struct ptlrpc_request *, struct ldlm_lock *);
770 * Check if the request is a high priority one.
772 int (*hpreq_check)(struct ptlrpc_request *);
774 * Called after the request has been handled.
776 void (*hpreq_fini)(struct ptlrpc_request *);
779 struct ptlrpc_cli_req {
780 /** For bulk requests on client only: bulk descriptor */
781 struct ptlrpc_bulk_desc *cr_bulk;
782 /** optional time limit for send attempts */
783 cfs_duration_t cr_delay_limit;
784 /** time request was first queued */
785 cfs_time_t cr_queued_time;
786 /** request sent timeval */
787 struct timeval cr_sent_tv;
788 /** time for request really sent out */
790 /** when req reply unlink must finish. */
791 time_t cr_reply_deadline;
792 /** when req bulk unlink must finish. */
793 time_t cr_bulk_deadline;
794 /** Portal to which this request would be sent */
796 /** Portal where to wait for reply and where reply would be sent */
798 /** request resending number */
799 unsigned int cr_resend_nr;
800 /** What was import generation when this request was sent */
802 enum lustre_imp_state cr_send_state;
803 /** Per-request waitq introduced by bug 21938 for recovery waiting */
804 wait_queue_head_t cr_set_waitq;
805 /** Link item for request set lists */
806 struct list_head cr_set_chain;
807 /** link to waited ctx */
808 struct list_head cr_ctx_chain;
810 /** client's half ctx */
811 struct ptlrpc_cli_ctx *cr_cli_ctx;
812 /** Link back to the request set */
813 struct ptlrpc_request_set *cr_set;
814 /** outgoing request MD handle */
815 lnet_handle_md_t cr_req_md_h;
816 /** request-out callback parameter */
817 struct ptlrpc_cb_id cr_req_cbid;
818 /** incoming reply MD handle */
819 lnet_handle_md_t cr_reply_md_h;
820 wait_queue_head_t cr_reply_waitq;
821 /** reply callback parameter */
822 struct ptlrpc_cb_id cr_reply_cbid;
823 /** Async completion handler, called when reply is received */
824 ptlrpc_interpterer_t cr_reply_interp;
825 /** Resend handler, called when request is resend to update RPC data */
826 ptlrpc_resend_cb_t cr_resend_cb;
827 /** Async completion context */
828 union ptlrpc_async_args cr_async_args;
829 /** Opaq data for replay and commit callbacks. */
831 /** Link to the imp->imp_unreplied_list */
832 struct list_head cr_unreplied_list;
834 * Commit callback, called when request is committed and about to be
837 void (*cr_commit_cb)(struct ptlrpc_request *);
838 /** Replay callback, called after request is replayed at recovery */
839 void (*cr_replay_cb)(struct ptlrpc_request *);
842 /** client request member alias */
843 /* NB: these alias should NOT be used by any new code, instead they should
844 * be removed step by step to avoid potential abuse */
845 #define rq_bulk rq_cli.cr_bulk
846 #define rq_delay_limit rq_cli.cr_delay_limit
847 #define rq_queued_time rq_cli.cr_queued_time
848 #define rq_sent_tv rq_cli.cr_sent_tv
849 #define rq_real_sent rq_cli.cr_sent_out
850 #define rq_reply_deadline rq_cli.cr_reply_deadline
851 #define rq_bulk_deadline rq_cli.cr_bulk_deadline
852 #define rq_nr_resend rq_cli.cr_resend_nr
853 #define rq_request_portal rq_cli.cr_req_ptl
854 #define rq_reply_portal rq_cli.cr_rep_ptl
855 #define rq_import_generation rq_cli.cr_imp_gen
856 #define rq_send_state rq_cli.cr_send_state
857 #define rq_set_chain rq_cli.cr_set_chain
858 #define rq_ctx_chain rq_cli.cr_ctx_chain
859 #define rq_set rq_cli.cr_set
860 #define rq_set_waitq rq_cli.cr_set_waitq
861 #define rq_cli_ctx rq_cli.cr_cli_ctx
862 #define rq_req_md_h rq_cli.cr_req_md_h
863 #define rq_req_cbid rq_cli.cr_req_cbid
864 #define rq_reply_md_h rq_cli.cr_reply_md_h
865 #define rq_reply_waitq rq_cli.cr_reply_waitq
866 #define rq_reply_cbid rq_cli.cr_reply_cbid
867 #define rq_interpret_reply rq_cli.cr_reply_interp
868 #define rq_resend_cb rq_cli.cr_resend_cb
869 #define rq_async_args rq_cli.cr_async_args
870 #define rq_cb_data rq_cli.cr_cb_data
871 #define rq_unreplied_list rq_cli.cr_unreplied_list
872 #define rq_commit_cb rq_cli.cr_commit_cb
873 #define rq_replay_cb rq_cli.cr_replay_cb
875 struct ptlrpc_srv_req {
876 /** initial thread servicing this request */
877 struct ptlrpc_thread *sr_svc_thread;
879 * Server side list of incoming unserved requests sorted by arrival
880 * time. Traversed from time to time to notice about to expire
881 * requests and sent back "early replies" to clients to let them
882 * know server is alive and well, just very busy to service their
885 struct list_head sr_timed_list;
886 /** server-side per-export list */
887 struct list_head sr_exp_list;
888 /** server-side history, used for debuging purposes. */
889 struct list_head sr_hist_list;
890 /** history sequence # */
892 /** the index of service's srv_at_array into which request is linked */
896 /** authed uid mapped to */
897 uid_t sr_auth_mapped_uid;
898 /** RPC is generated from what part of Lustre */
899 enum lustre_sec_part sr_sp_from;
900 /** request session context */
901 struct lu_context sr_ses;
905 /** stub for NRS request */
906 struct ptlrpc_nrs_request sr_nrq;
908 /** request arrival time */
909 struct timeval sr_arrival_time;
910 /** server's half ctx */
911 struct ptlrpc_svc_ctx *sr_svc_ctx;
912 /** (server side), pointed directly into req buffer */
913 struct ptlrpc_user_desc *sr_user_desc;
914 /** separated reply state, may be vmalloc'd */
915 struct ptlrpc_reply_state *sr_reply_state;
916 /** server-side hp handlers */
917 struct ptlrpc_hpreq_ops *sr_ops;
918 /** incoming request buffer */
919 struct ptlrpc_request_buffer_desc *sr_rqbd;
922 /** server request member alias */
923 /* NB: these alias should NOT be used by any new code, instead they should
924 * be removed step by step to avoid potential abuse */
925 #define rq_svc_thread rq_srv.sr_svc_thread
926 #define rq_timed_list rq_srv.sr_timed_list
927 #define rq_exp_list rq_srv.sr_exp_list
928 #define rq_history_list rq_srv.sr_hist_list
929 #define rq_history_seq rq_srv.sr_hist_seq
930 #define rq_at_index rq_srv.sr_at_index
931 #define rq_auth_uid rq_srv.sr_auth_uid
932 #define rq_auth_mapped_uid rq_srv.sr_auth_mapped_uid
933 #define rq_sp_from rq_srv.sr_sp_from
934 #define rq_session rq_srv.sr_ses
935 #define rq_nrq rq_srv.sr_nrq
936 #define rq_arrival_time rq_srv.sr_arrival_time
937 #define rq_reply_state rq_srv.sr_reply_state
938 #define rq_svc_ctx rq_srv.sr_svc_ctx
939 #define rq_user_desc rq_srv.sr_user_desc
940 #define rq_ops rq_srv.sr_ops
941 #define rq_rqbd rq_srv.sr_rqbd
944 * Represents remote procedure call.
946 * This is a staple structure used by everybody wanting to send a request
949 struct ptlrpc_request {
950 /* Request type: one of PTL_RPC_MSG_* */
952 /** Result of request processing */
955 * Linkage item through which this request is included into
956 * sending/delayed lists on client and into rqbd list on server
958 struct list_head rq_list;
959 /** Lock to protect request flags and some other important bits, like
963 /** client-side flags are serialized by rq_lock @{ */
964 unsigned int rq_intr:1, rq_replied:1, rq_err:1,
965 rq_timedout:1, rq_resend:1, rq_restart:1,
967 * when ->rq_replay is set, request is kept by the client even
968 * after server commits corresponding transaction. This is
969 * used for operations that require sequence of multiple
970 * requests to be replayed. The only example currently is file
971 * open/close. When last request in such a sequence is
972 * committed, ->rq_replay is cleared on all requests in the
976 rq_no_resend:1, rq_waiting:1, rq_receiving_reply:1,
977 rq_no_delay:1, rq_net_err:1, rq_wait_ctx:1,
979 rq_req_unlinked:1, /* unlinked request buffer from lnet */
980 rq_reply_unlinked:1, /* unlinked reply buffer from lnet */
981 rq_memalloc:1, /* req originated from "kswapd" */
983 rq_reply_truncated:1,
984 /** whether the "rq_set" is a valid one */
987 /** do not resend request on -EINPROGRESS */
988 rq_no_retry_einprogress:1,
989 /* allow the req to be sent if the import is in recovery
992 /* bulk request, sent to server, but uncommitted */
997 /** server-side flags @{ */
999 rq_hp:1, /**< high priority RPC */
1000 rq_at_linked:1, /**< link into service's srv_at_array */
1001 rq_packed_final:1; /**< packed final reply */
1004 /** one of RQ_PHASE_* */
1005 enum rq_phase rq_phase;
1006 /** one of RQ_PHASE_* to be used next */
1007 enum rq_phase rq_next_phase;
1009 * client-side refcount for SENT race, server-side refcounf
1010 * for multiple replies
1012 atomic_t rq_refcount;
1015 * !rq_truncate : # reply bytes actually received,
1016 * rq_truncate : required repbuf_len for resend
1018 int rq_nob_received;
1019 /** Request length */
1023 /** Pool if request is from preallocated list */
1024 struct ptlrpc_request_pool *rq_pool;
1025 /** Request message - what client sent */
1026 struct lustre_msg *rq_reqmsg;
1027 /** Reply message - server response */
1028 struct lustre_msg *rq_repmsg;
1029 /** Transaction number */
1033 /** bulk match bits */
1036 * List item to for replay list. Not yet committed requests get linked
1038 * Also see \a rq_replay comment above.
1039 * It's also link chain on obd_export::exp_req_replay_queue
1041 struct list_head rq_replay_list;
1042 /** non-shared members for client & server request*/
1044 struct ptlrpc_cli_req rq_cli;
1045 struct ptlrpc_srv_req rq_srv;
1048 * security and encryption data
1050 /** description of flavors for client & server */
1051 struct sptlrpc_flavor rq_flvr;
1053 /* client/server security flags */
1055 rq_ctx_init:1, /* context initiation */
1056 rq_ctx_fini:1, /* context destroy */
1057 rq_bulk_read:1, /* request bulk read */
1058 rq_bulk_write:1, /* request bulk write */
1059 /* server authentication flags */
1060 rq_auth_gss:1, /* authenticated by gss */
1061 rq_auth_remote:1, /* authed as remote user */
1062 rq_auth_usr_root:1, /* authed as root */
1063 rq_auth_usr_mdt:1, /* authed as mdt */
1064 rq_auth_usr_ost:1, /* authed as ost */
1065 /* security tfm flags */
1068 /* doesn't expect reply FIXME */
1070 rq_pill_init:1, /* pill initialized */
1071 rq_srv_req:1; /* server request */
1074 /** various buffer pointers */
1075 struct lustre_msg *rq_reqbuf; /**< req wrapper, vmalloc*/
1076 char *rq_repbuf; /**< rep buffer, vmalloc */
1077 struct lustre_msg *rq_repdata; /**< rep wrapper msg */
1078 /** only in priv mode */
1079 struct lustre_msg *rq_clrbuf;
1080 int rq_reqbuf_len; /* req wrapper buf len */
1081 int rq_reqdata_len; /* req wrapper msg len */
1082 int rq_repbuf_len; /* rep buffer len */
1083 int rq_repdata_len; /* rep wrapper msg len */
1084 int rq_clrbuf_len; /* only in priv mode */
1085 int rq_clrdata_len; /* only in priv mode */
1087 /** early replies go to offset 0, regular replies go after that */
1088 unsigned int rq_reply_off;
1091 /** Fields that help to see if request and reply were swabbed or not */
1092 __u32 rq_req_swab_mask;
1093 __u32 rq_rep_swab_mask;
1095 /** how many early replies (for stats) */
1097 /** Server-side, export on which request was received */
1098 struct obd_export *rq_export;
1099 /** import where request is being sent */
1100 struct obd_import *rq_import;
1103 /** Peer description (the other side) */
1104 lnet_process_id_t rq_peer;
1106 * service time estimate (secs)
1107 * If the request is not served by this time, it is marked as timed out.
1111 * when request/reply sent (secs), or time when request should be sent
1114 /** when request must finish. */
1116 /** request format description */
1117 struct req_capsule rq_pill;
1121 * Call completion handler for rpc if any, return it's status or original
1122 * rc if there was no handler defined for this request.
1124 static inline int ptlrpc_req_interpret(const struct lu_env *env,
1125 struct ptlrpc_request *req, int rc)
1127 if (req->rq_interpret_reply != NULL) {
1128 req->rq_status = req->rq_interpret_reply(env, req,
1129 &req->rq_async_args,
1131 return req->rq_status;
1139 int ptlrpc_nrs_policy_register(struct ptlrpc_nrs_pol_conf *conf);
1140 int ptlrpc_nrs_policy_unregister(struct ptlrpc_nrs_pol_conf *conf);
1141 void ptlrpc_nrs_req_hp_move(struct ptlrpc_request *req);
1142 void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
1143 struct ptlrpc_nrs_pol_info *info);
1146 * Can the request be moved from the regular NRS head to the high-priority NRS
1147 * head (of the same PTLRPC service partition), if any?
1149 * For a reliable result, this should be checked under svcpt->scp_req lock.
1151 static inline bool ptlrpc_nrs_req_can_move(struct ptlrpc_request *req)
1153 struct ptlrpc_nrs_request *nrq = &req->rq_nrq;
1156 * LU-898: Check ptlrpc_nrs_request::nr_enqueued to make sure the
1157 * request has been enqueued first, and ptlrpc_nrs_request::nr_started
1158 * to make sure it has not been scheduled yet (analogous to previous
1159 * (non-NRS) checking of !list_empty(&ptlrpc_request::rq_list).
1161 return nrq->nr_enqueued && !nrq->nr_started && !req->rq_hp;
1166 * Returns 1 if request buffer at offset \a index was already swabbed
1168 static inline int lustre_req_swabbed(struct ptlrpc_request *req, size_t index)
1170 LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
1171 return req->rq_req_swab_mask & (1 << index);
1175 * Returns 1 if request reply buffer at offset \a index was already swabbed
1177 static inline int lustre_rep_swabbed(struct ptlrpc_request *req, size_t index)
1179 LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
1180 return req->rq_rep_swab_mask & (1 << index);
1184 * Returns 1 if request needs to be swabbed into local cpu byteorder
1186 static inline int ptlrpc_req_need_swab(struct ptlrpc_request *req)
1188 return lustre_req_swabbed(req, MSG_PTLRPC_HEADER_OFF);
1192 * Returns 1 if request reply needs to be swabbed into local cpu byteorder
1194 static inline int ptlrpc_rep_need_swab(struct ptlrpc_request *req)
1196 return lustre_rep_swabbed(req, MSG_PTLRPC_HEADER_OFF);
1200 * Mark request buffer at offset \a index that it was already swabbed
1202 static inline void lustre_set_req_swabbed(struct ptlrpc_request *req,
1205 LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
1206 LASSERT((req->rq_req_swab_mask & (1 << index)) == 0);
1207 req->rq_req_swab_mask |= 1 << index;
1211 * Mark request reply buffer at offset \a index that it was already swabbed
1213 static inline void lustre_set_rep_swabbed(struct ptlrpc_request *req,
1216 LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
1217 LASSERT((req->rq_rep_swab_mask & (1 << index)) == 0);
1218 req->rq_rep_swab_mask |= 1 << index;
1222 * Convert numerical request phase value \a phase into text string description
1224 static inline const char *
1225 ptlrpc_phase2str(enum rq_phase phase)
1234 case RQ_PHASE_INTERPRET:
1236 case RQ_PHASE_COMPLETE:
1238 case RQ_PHASE_UNREGISTERING:
1239 return "Unregistering";
1246 * Convert numerical request phase of the request \a req into text stringi
1249 static inline const char *
1250 ptlrpc_rqphase2str(struct ptlrpc_request *req)
1252 return ptlrpc_phase2str(req->rq_phase);
1256 * Debugging functions and helpers to print request structure into debug log
1259 /* Spare the preprocessor, spoil the bugs. */
1260 #define FLAG(field, str) (field ? str : "")
1262 /** Convert bit flags into a string */
1263 #define DEBUG_REQ_FLAGS(req) \
1264 ptlrpc_rqphase2str(req), \
1265 FLAG(req->rq_intr, "I"), FLAG(req->rq_replied, "R"), \
1266 FLAG(req->rq_err, "E"), \
1267 FLAG(req->rq_timedout, "X") /* eXpired */, FLAG(req->rq_resend, "S"), \
1268 FLAG(req->rq_restart, "T"), FLAG(req->rq_replay, "P"), \
1269 FLAG(req->rq_no_resend, "N"), \
1270 FLAG(req->rq_waiting, "W"), \
1271 FLAG(req->rq_wait_ctx, "C"), FLAG(req->rq_hp, "H"), \
1272 FLAG(req->rq_committed, "M")
1274 #define REQ_FLAGS_FMT "%s:%s%s%s%s%s%s%s%s%s%s%s%s"
1276 void _debug_req(struct ptlrpc_request *req,
1277 struct libcfs_debug_msg_data *data, const char *fmt, ...)
1278 __attribute__ ((format (printf, 3, 4)));
1281 * Helper that decides if we need to print request accordig to current debug
1284 #define debug_req(msgdata, mask, cdls, req, fmt, a...) \
1286 CFS_CHECK_STACK(msgdata, mask, cdls); \
1288 if (((mask) & D_CANTMASK) != 0 || \
1289 ((libcfs_debug & (mask)) != 0 && \
1290 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) \
1291 _debug_req((req), msgdata, fmt, ##a); \
1295 * This is the debug print function you need to use to print request sturucture
1296 * content into lustre debug log.
1297 * for most callers (level is a constant) this is resolved at compile time */
1298 #define DEBUG_REQ(level, req, fmt, args...) \
1300 if ((level) & (D_ERROR | D_WARNING)) { \
1301 static struct cfs_debug_limit_state cdls; \
1302 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, level, &cdls); \
1303 debug_req(&msgdata, level, &cdls, req, "@@@ "fmt" ", ## args);\
1305 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, level, NULL); \
1306 debug_req(&msgdata, level, NULL, req, "@@@ "fmt" ", ## args); \
1312 * Structure that defines a single page of a bulk transfer
1314 struct ptlrpc_bulk_page {
1315 /** Linkage to list of pages in a bulk */
1316 struct list_head bp_link;
1318 * Number of bytes in a page to transfer starting from \a bp_pageoffset
1321 /** offset within a page */
1323 /** The page itself */
1324 struct page *bp_page;
1327 enum ptlrpc_bulk_op_type {
1328 PTLRPC_BULK_OP_ACTIVE = 0x00000001,
1329 PTLRPC_BULK_OP_PASSIVE = 0x00000002,
1330 PTLRPC_BULK_OP_PUT = 0x00000004,
1331 PTLRPC_BULK_OP_GET = 0x00000008,
1332 PTLRPC_BULK_BUF_KVEC = 0x00000010,
1333 PTLRPC_BULK_BUF_KIOV = 0x00000020,
1334 PTLRPC_BULK_GET_SOURCE = PTLRPC_BULK_OP_PASSIVE | PTLRPC_BULK_OP_GET,
1335 PTLRPC_BULK_PUT_SINK = PTLRPC_BULK_OP_PASSIVE | PTLRPC_BULK_OP_PUT,
1336 PTLRPC_BULK_GET_SINK = PTLRPC_BULK_OP_ACTIVE | PTLRPC_BULK_OP_GET,
1337 PTLRPC_BULK_PUT_SOURCE = PTLRPC_BULK_OP_ACTIVE | PTLRPC_BULK_OP_PUT,
1340 static inline bool ptlrpc_is_bulk_op_get(enum ptlrpc_bulk_op_type type)
1342 return (type & PTLRPC_BULK_OP_GET) == PTLRPC_BULK_OP_GET;
1345 static inline bool ptlrpc_is_bulk_get_source(enum ptlrpc_bulk_op_type type)
1347 return (type & PTLRPC_BULK_GET_SOURCE) == PTLRPC_BULK_GET_SOURCE;
1350 static inline bool ptlrpc_is_bulk_put_sink(enum ptlrpc_bulk_op_type type)
1352 return (type & PTLRPC_BULK_PUT_SINK) == PTLRPC_BULK_PUT_SINK;
1355 static inline bool ptlrpc_is_bulk_get_sink(enum ptlrpc_bulk_op_type type)
1357 return (type & PTLRPC_BULK_GET_SINK) == PTLRPC_BULK_GET_SINK;
1360 static inline bool ptlrpc_is_bulk_put_source(enum ptlrpc_bulk_op_type type)
1362 return (type & PTLRPC_BULK_PUT_SOURCE) == PTLRPC_BULK_PUT_SOURCE;
1365 static inline bool ptlrpc_is_bulk_desc_kvec(enum ptlrpc_bulk_op_type type)
1367 return ((type & PTLRPC_BULK_BUF_KVEC) | (type & PTLRPC_BULK_BUF_KIOV))
1368 == PTLRPC_BULK_BUF_KVEC;
1371 static inline bool ptlrpc_is_bulk_desc_kiov(enum ptlrpc_bulk_op_type type)
1373 return ((type & PTLRPC_BULK_BUF_KVEC) | (type & PTLRPC_BULK_BUF_KIOV))
1374 == PTLRPC_BULK_BUF_KIOV;
1377 static inline bool ptlrpc_is_bulk_op_active(enum ptlrpc_bulk_op_type type)
1379 return ((type & PTLRPC_BULK_OP_ACTIVE) |
1380 (type & PTLRPC_BULK_OP_PASSIVE))
1381 == PTLRPC_BULK_OP_ACTIVE;
1384 static inline bool ptlrpc_is_bulk_op_passive(enum ptlrpc_bulk_op_type type)
1386 return ((type & PTLRPC_BULK_OP_ACTIVE) |
1387 (type & PTLRPC_BULK_OP_PASSIVE))
1388 == PTLRPC_BULK_OP_PASSIVE;
1391 struct ptlrpc_bulk_frag_ops {
1393 * Add a page \a page to the bulk descriptor \a desc
1394 * Data to transfer in the page starts at offset \a pageoffset and
1395 * amount of data to transfer from the page is \a len
1397 void (*add_kiov_frag)(struct ptlrpc_bulk_desc *desc,
1398 struct page *page, int pageoffset, int len);
1401 * Add a \a fragment to the bulk descriptor \a desc.
1402 * Data to transfer in the fragment is pointed to by \a frag
1403 * The size of the fragment is \a len
1405 int (*add_iov_frag)(struct ptlrpc_bulk_desc *desc, void *frag, int len);
1408 * Uninitialize and free bulk descriptor \a desc.
1409 * Works on bulk descriptors both from server and client side.
1411 void (*release_frags)(struct ptlrpc_bulk_desc *desc);
1414 extern const struct ptlrpc_bulk_frag_ops ptlrpc_bulk_kiov_pin_ops;
1415 extern const struct ptlrpc_bulk_frag_ops ptlrpc_bulk_kiov_nopin_ops;
1416 extern const struct ptlrpc_bulk_frag_ops ptlrpc_bulk_kvec_ops;
1419 * Definition of bulk descriptor.
1420 * Bulks are special "Two phase" RPCs where initial request message
1421 * is sent first and it is followed bt a transfer (o receiving) of a large
1422 * amount of data to be settled into pages referenced from the bulk descriptors.
1423 * Bulks transfers (the actual data following the small requests) are done
1424 * on separate LNet portals.
1425 * In lustre we use bulk transfers for READ and WRITE transfers from/to OSTs.
1426 * Another user is readpage for MDT.
1428 struct ptlrpc_bulk_desc {
1429 /** completed with failure */
1430 unsigned long bd_failure:1;
1432 unsigned long bd_registered:1;
1433 /** For serialization with callback */
1435 /** Import generation when request for this bulk was sent */
1436 int bd_import_generation;
1437 /** {put,get}{source,sink}{kvec,kiov} */
1438 enum ptlrpc_bulk_op_type bd_type;
1439 /** LNet portal for this bulk */
1441 /** Server side - export this bulk created for */
1442 struct obd_export *bd_export;
1443 /** Client side - import this bulk was sent on */
1444 struct obd_import *bd_import;
1445 /** Back pointer to the request */
1446 struct ptlrpc_request *bd_req;
1447 struct ptlrpc_bulk_frag_ops *bd_frag_ops;
1448 wait_queue_head_t bd_waitq; /* server side only WQ */
1449 int bd_iov_count; /* # entries in bd_iov */
1450 int bd_max_iov; /* allocated size of bd_iov */
1451 int bd_nob; /* # bytes covered */
1452 int bd_nob_transferred; /* # bytes GOT/PUT */
1454 __u64 bd_last_mbits;
1456 struct ptlrpc_cb_id bd_cbid; /* network callback info */
1457 lnet_nid_t bd_sender; /* stash event::sender */
1458 int bd_md_count; /* # valid entries in bd_mds */
1459 int bd_md_max_brw; /* max entries in bd_mds */
1460 /** array of associated MDs */
1461 lnet_handle_md_t bd_mds[PTLRPC_BULK_OPS_COUNT];
1466 * encrypt iov, size is either 0 or bd_iov_count.
1468 lnet_kiov_t *bd_enc_vec;
1469 lnet_kiov_t bd_vec[0];
1473 struct kvec *bd_enc_kvec;
1474 struct kvec bd_kvec[0];
1480 #define GET_KIOV(desc) ((desc)->bd_u.bd_kiov.bd_vec)
1481 #define BD_GET_KIOV(desc, i) ((desc)->bd_u.bd_kiov.bd_vec[i])
1482 #define GET_ENC_KIOV(desc) ((desc)->bd_u.bd_kiov.bd_enc_vec)
1483 #define BD_GET_ENC_KIOV(desc, i) ((desc)->bd_u.bd_kiov.bd_enc_vec[i])
1484 #define GET_KVEC(desc) ((desc)->bd_u.bd_kvec.bd_kvec)
1485 #define BD_GET_KVEC(desc, i) ((desc)->bd_u.bd_kvec.bd_kvec[i])
1486 #define GET_ENC_KVEC(desc) ((desc)->bd_u.bd_kvec.bd_enc_kvec)
1487 #define BD_GET_ENC_KVEC(desc, i) ((desc)->bd_u.bd_kvec.bd_enc_kvec[i])
1490 SVC_STOPPED = 1 << 0,
1491 SVC_STOPPING = 1 << 1,
1492 SVC_STARTING = 1 << 2,
1493 SVC_RUNNING = 1 << 3,
1495 SVC_SIGNAL = 1 << 5,
1498 #define PTLRPC_THR_NAME_LEN 32
1500 * Definition of server service thread structure
1502 struct ptlrpc_thread {
1504 * List of active threads in svc->srv_threads
1506 struct list_head t_link;
1508 * thread-private data (preallocated vmalloc'd memory)
1513 * service thread index, from ptlrpc_start_threads
1517 * service thread pid
1521 * put watchdog in the structure per thread b=14840
1523 struct lc_watchdog *t_watchdog;
1525 * the svc this thread belonged to b=18582
1527 struct ptlrpc_service_part *t_svcpt;
1528 wait_queue_head_t t_ctl_waitq;
1529 struct lu_env *t_env;
1530 char t_name[PTLRPC_THR_NAME_LEN];
1533 static inline int thread_is_init(struct ptlrpc_thread *thread)
1535 return thread->t_flags == 0;
1538 static inline int thread_is_stopped(struct ptlrpc_thread *thread)
1540 return !!(thread->t_flags & SVC_STOPPED);
1543 static inline int thread_is_stopping(struct ptlrpc_thread *thread)
1545 return !!(thread->t_flags & SVC_STOPPING);
1548 static inline int thread_is_starting(struct ptlrpc_thread *thread)
1550 return !!(thread->t_flags & SVC_STARTING);
1553 static inline int thread_is_running(struct ptlrpc_thread *thread)
1555 return !!(thread->t_flags & SVC_RUNNING);
1558 static inline int thread_is_event(struct ptlrpc_thread *thread)
1560 return !!(thread->t_flags & SVC_EVENT);
1563 static inline int thread_is_signal(struct ptlrpc_thread *thread)
1565 return !!(thread->t_flags & SVC_SIGNAL);
1568 static inline void thread_clear_flags(struct ptlrpc_thread *thread, __u32 flags)
1570 thread->t_flags &= ~flags;
1573 static inline void thread_set_flags(struct ptlrpc_thread *thread, __u32 flags)
1575 thread->t_flags = flags;
1578 static inline void thread_add_flags(struct ptlrpc_thread *thread, __u32 flags)
1580 thread->t_flags |= flags;
1583 static inline int thread_test_and_clear_flags(struct ptlrpc_thread *thread,
1586 if (thread->t_flags & flags) {
1587 thread->t_flags &= ~flags;
1594 * Request buffer descriptor structure.
1595 * This is a structure that contains one posted request buffer for service.
1596 * Once data land into a buffer, event callback creates actual request and
1597 * notifies wakes one of the service threads to process new incoming request.
1598 * More than one request can fit into the buffer.
1600 struct ptlrpc_request_buffer_desc {
1601 /** Link item for rqbds on a service */
1602 struct list_head rqbd_list;
1603 /** History of requests for this buffer */
1604 struct list_head rqbd_reqs;
1605 /** Back pointer to service for which this buffer is registered */
1606 struct ptlrpc_service_part *rqbd_svcpt;
1607 /** LNet descriptor */
1608 lnet_handle_md_t rqbd_md_h;
1610 /** The buffer itself */
1612 struct ptlrpc_cb_id rqbd_cbid;
1614 * This "embedded" request structure is only used for the
1615 * last request to fit into the buffer
1617 struct ptlrpc_request rqbd_req;
1620 typedef int (*svc_handler_t)(struct ptlrpc_request *req);
1622 struct ptlrpc_service_ops {
1624 * if non-NULL called during thread creation (ptlrpc_start_thread())
1625 * to initialize service specific per-thread state.
1627 int (*so_thr_init)(struct ptlrpc_thread *thr);
1629 * if non-NULL called during thread shutdown (ptlrpc_main()) to
1630 * destruct state created by ->srv_init().
1632 void (*so_thr_done)(struct ptlrpc_thread *thr);
1634 * Handler function for incoming requests for this service
1636 int (*so_req_handler)(struct ptlrpc_request *req);
1638 * function to determine priority of the request, it's called
1639 * on every new request
1641 int (*so_hpreq_handler)(struct ptlrpc_request *);
1643 * service-specific print fn
1645 void (*so_req_printer)(void *, struct ptlrpc_request *);
1648 #ifndef __cfs_cacheline_aligned
1649 /* NB: put it here for reducing patche dependence */
1650 # define __cfs_cacheline_aligned
1654 * How many high priority requests to serve before serving one normal
1657 #define PTLRPC_SVC_HP_RATIO 10
1660 * Definition of PortalRPC service.
1661 * The service is listening on a particular portal (like tcp port)
1662 * and perform actions for a specific server like IO service for OST
1663 * or general metadata service for MDS.
1665 struct ptlrpc_service {
1666 /** serialize /proc operations */
1667 spinlock_t srv_lock;
1668 /** most often accessed fields */
1669 /** chain thru all services */
1670 struct list_head srv_list;
1671 /** service operations table */
1672 struct ptlrpc_service_ops srv_ops;
1673 /** only statically allocated strings here; we don't clean them */
1675 /** only statically allocated strings here; we don't clean them */
1676 char *srv_thread_name;
1677 /** service thread list */
1678 struct list_head srv_threads;
1679 /** threads # should be created for each partition on initializing */
1680 int srv_nthrs_cpt_init;
1681 /** limit of threads number for each partition */
1682 int srv_nthrs_cpt_limit;
1683 /** Root of /proc dir tree for this service */
1684 struct proc_dir_entry *srv_procroot;
1685 /** Pointer to statistic data for this service */
1686 struct lprocfs_stats *srv_stats;
1687 /** # hp per lp reqs to handle */
1688 int srv_hpreq_ratio;
1689 /** biggest request to receive */
1690 int srv_max_req_size;
1691 /** biggest reply to send */
1692 int srv_max_reply_size;
1693 /** size of individual buffers */
1695 /** # buffers to allocate in 1 group */
1696 int srv_nbuf_per_group;
1697 /** Local portal on which to receive requests */
1698 __u32 srv_req_portal;
1699 /** Portal on the client to send replies to */
1700 __u32 srv_rep_portal;
1702 * Tags for lu_context associated with this thread, see struct
1706 /** soft watchdog timeout multiplier */
1707 int srv_watchdog_factor;
1708 /** under unregister_service */
1709 unsigned srv_is_stopping:1;
1711 /** max # request buffers in history per partition */
1712 int srv_hist_nrqbds_cpt_max;
1713 /** number of CPTs this service bound on */
1715 /** CPTs array this service bound on */
1717 /** 2^srv_cptab_bits >= cfs_cpt_numbert(srv_cptable) */
1719 /** CPT table this service is running over */
1720 struct cfs_cpt_table *srv_cptable;
1722 * partition data for ptlrpc service
1724 struct ptlrpc_service_part *srv_parts[0];
1728 * Definition of PortalRPC service partition data.
1729 * Although a service only has one instance of it right now, but we
1730 * will have multiple instances very soon (instance per CPT).
1732 * it has four locks:
1734 * serialize operations on rqbd and requests waiting for preprocess
1736 * serialize operations active requests sent to this portal
1738 * serialize adaptive timeout stuff
1740 * serialize operations on RS list (reply states)
1742 * We don't have any use-case to take two or more locks at the same time
1743 * for now, so there is no lock order issue.
1745 struct ptlrpc_service_part {
1746 /** back reference to owner */
1747 struct ptlrpc_service *scp_service __cfs_cacheline_aligned;
1748 /* CPT id, reserved */
1750 /** always increasing number */
1752 /** # of starting threads */
1753 int scp_nthrs_starting;
1754 /** # of stopping threads, reserved for shrinking threads */
1755 int scp_nthrs_stopping;
1756 /** # running threads */
1757 int scp_nthrs_running;
1758 /** service threads list */
1759 struct list_head scp_threads;
1762 * serialize the following fields, used for protecting
1763 * rqbd list and incoming requests waiting for preprocess,
1764 * threads starting & stopping are also protected by this lock.
1766 spinlock_t scp_lock __cfs_cacheline_aligned;
1767 /** total # req buffer descs allocated */
1768 int scp_nrqbds_total;
1769 /** # posted request buffers for receiving */
1770 int scp_nrqbds_posted;
1771 /** in progress of allocating rqbd */
1772 int scp_rqbd_allocating;
1773 /** # incoming reqs */
1774 int scp_nreqs_incoming;
1775 /** request buffers to be reposted */
1776 struct list_head scp_rqbd_idle;
1777 /** req buffers receiving */
1778 struct list_head scp_rqbd_posted;
1779 /** incoming reqs */
1780 struct list_head scp_req_incoming;
1781 /** timeout before re-posting reqs, in tick */
1782 cfs_duration_t scp_rqbd_timeout;
1784 * all threads sleep on this. This wait-queue is signalled when new
1785 * incoming request arrives and when difficult reply has to be handled.
1787 wait_queue_head_t scp_waitq;
1789 /** request history */
1790 struct list_head scp_hist_reqs;
1791 /** request buffer history */
1792 struct list_head scp_hist_rqbds;
1793 /** # request buffers in history */
1794 int scp_hist_nrqbds;
1795 /** sequence number for request */
1797 /** highest seq culled from history */
1798 __u64 scp_hist_seq_culled;
1801 * serialize the following fields, used for processing requests
1802 * sent to this portal
1804 spinlock_t scp_req_lock __cfs_cacheline_aligned;
1805 /** # reqs in either of the NRS heads below */
1806 /** # reqs being served */
1807 int scp_nreqs_active;
1808 /** # HPreqs being served */
1809 int scp_nhreqs_active;
1810 /** # hp requests handled */
1813 /** NRS head for regular requests */
1814 struct ptlrpc_nrs scp_nrs_reg;
1815 /** NRS head for HP requests; this is only valid for services that can
1816 * handle HP requests */
1817 struct ptlrpc_nrs *scp_nrs_hp;
1822 * serialize the following fields, used for changes on
1825 spinlock_t scp_at_lock __cfs_cacheline_aligned;
1826 /** estimated rpc service time */
1827 struct adaptive_timeout scp_at_estimate;
1828 /** reqs waiting for replies */
1829 struct ptlrpc_at_array scp_at_array;
1830 /** early reply timer */
1831 struct timer_list scp_at_timer;
1833 cfs_time_t scp_at_checktime;
1834 /** check early replies */
1835 unsigned scp_at_check;
1839 * serialize the following fields, used for processing
1840 * replies for this portal
1842 spinlock_t scp_rep_lock __cfs_cacheline_aligned;
1843 /** all the active replies */
1844 struct list_head scp_rep_active;
1845 /** List of free reply_states */
1846 struct list_head scp_rep_idle;
1847 /** waitq to run, when adding stuff to srv_free_rs_list */
1848 wait_queue_head_t scp_rep_waitq;
1849 /** # 'difficult' replies */
1850 atomic_t scp_nreps_difficult;
1853 #define ptlrpc_service_for_each_part(part, i, svc) \
1855 i < (svc)->srv_ncpts && \
1856 (svc)->srv_parts != NULL && \
1857 ((part) = (svc)->srv_parts[i]) != NULL; i++)
1860 * Declaration of ptlrpcd control structure
1862 struct ptlrpcd_ctl {
1864 * Ptlrpc thread control flags (LIOD_START, LIOD_STOP, LIOD_FORCE)
1866 unsigned long pc_flags;
1868 * Thread lock protecting structure fields.
1874 struct completion pc_starting;
1878 struct completion pc_finishing;
1880 * Thread requests set.
1882 struct ptlrpc_request_set *pc_set;
1884 * Thread name used in kthread_run()
1888 * Environment for request interpreters to run in.
1890 struct lu_env pc_env;
1892 * CPT the thread is bound on.
1896 * Index of ptlrpcd thread in the array.
1900 * Pointer to the array of partners' ptlrpcd_ctl structure.
1902 struct ptlrpcd_ctl **pc_partners;
1904 * Number of the ptlrpcd's partners.
1908 * Record the partner index to be processed next.
1912 * Error code if the thread failed to fully start.
1917 /* Bits for pc_flags */
1918 enum ptlrpcd_ctl_flags {
1920 * Ptlrpc thread start flag.
1922 LIOD_START = 1 << 0,
1924 * Ptlrpc thread stop flag.
1928 * Ptlrpc thread force flag (only stop force so far).
1929 * This will cause aborting any inflight rpcs handled
1930 * by thread if LIOD_STOP is specified.
1932 LIOD_FORCE = 1 << 2,
1934 * This is a recovery ptlrpc thread.
1936 LIOD_RECOVERY = 1 << 3,
1943 * Service compatibility function; the policy is compatible with all services.
1945 * \param[in] svc The service the policy is attempting to register with.
1946 * \param[in] desc The policy descriptor
1948 * \retval true The policy is compatible with the service
1950 * \see ptlrpc_nrs_pol_desc::pd_compat()
1952 static inline bool nrs_policy_compat_all(const struct ptlrpc_service *svc,
1953 const struct ptlrpc_nrs_pol_desc *desc)
1959 * Service compatibility function; the policy is compatible with only a specific
1960 * service which is identified by its human-readable name at
1961 * ptlrpc_service::srv_name.
1963 * \param[in] svc The service the policy is attempting to register with.
1964 * \param[in] desc The policy descriptor
1966 * \retval false The policy is not compatible with the service
1967 * \retval true The policy is compatible with the service
1969 * \see ptlrpc_nrs_pol_desc::pd_compat()
1971 static inline bool nrs_policy_compat_one(const struct ptlrpc_service *svc,
1972 const struct ptlrpc_nrs_pol_desc *desc)
1974 LASSERT(desc->pd_compat_svc_name != NULL);
1975 return strcmp(svc->srv_name, desc->pd_compat_svc_name) == 0;
1980 /* ptlrpc/events.c */
1981 extern lnet_handle_eq_t ptlrpc_eq_h;
1982 extern int ptlrpc_uuid_to_peer(struct obd_uuid *uuid,
1983 lnet_process_id_t *peer, lnet_nid_t *self);
1985 * These callbacks are invoked by LNet when something happened to
1989 extern void request_out_callback(lnet_event_t *ev);
1990 extern void reply_in_callback(lnet_event_t *ev);
1991 extern void client_bulk_callback(lnet_event_t *ev);
1992 extern void request_in_callback(lnet_event_t *ev);
1993 extern void reply_out_callback(lnet_event_t *ev);
1994 #ifdef HAVE_SERVER_SUPPORT
1995 extern void server_bulk_callback(lnet_event_t *ev);
1999 /* ptlrpc/connection.c */
2000 struct ptlrpc_connection *ptlrpc_connection_get(lnet_process_id_t peer,
2002 struct obd_uuid *uuid);
2003 int ptlrpc_connection_put(struct ptlrpc_connection *c);
2004 struct ptlrpc_connection *ptlrpc_connection_addref(struct ptlrpc_connection *);
2005 int ptlrpc_connection_init(void);
2006 void ptlrpc_connection_fini(void);
2007 extern lnet_pid_t ptl_get_pid(void);
2009 /* ptlrpc/niobuf.c */
2011 * Actual interfacing with LNet to put/get/register/unregister stuff
2014 #ifdef HAVE_SERVER_SUPPORT
2015 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_exp(struct ptlrpc_request *req,
2016 unsigned nfrags, unsigned max_brw,
2019 const struct ptlrpc_bulk_frag_ops
2021 int ptlrpc_start_bulk_transfer(struct ptlrpc_bulk_desc *desc);
2022 void ptlrpc_abort_bulk(struct ptlrpc_bulk_desc *desc);
2024 static inline int ptlrpc_server_bulk_active(struct ptlrpc_bulk_desc *desc)
2028 LASSERT(desc != NULL);
2030 spin_lock(&desc->bd_lock);
2031 rc = desc->bd_md_count;
2032 spin_unlock(&desc->bd_lock);
2037 int ptlrpc_register_bulk(struct ptlrpc_request *req);
2038 int ptlrpc_unregister_bulk(struct ptlrpc_request *req, int async);
2040 static inline int ptlrpc_client_bulk_active(struct ptlrpc_request *req)
2042 struct ptlrpc_bulk_desc *desc;
2045 LASSERT(req != NULL);
2046 desc = req->rq_bulk;
2048 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_BULK_UNLINK) &&
2049 req->rq_bulk_deadline > cfs_time_current_sec())
2055 spin_lock(&desc->bd_lock);
2056 rc = desc->bd_md_count;
2057 spin_unlock(&desc->bd_lock);
2061 #define PTLRPC_REPLY_MAYBE_DIFFICULT 0x01
2062 #define PTLRPC_REPLY_EARLY 0x02
2063 int ptlrpc_send_reply(struct ptlrpc_request *req, int flags);
2064 int ptlrpc_reply(struct ptlrpc_request *req);
2065 int ptlrpc_send_error(struct ptlrpc_request *req, int difficult);
2066 int ptlrpc_error(struct ptlrpc_request *req);
2067 int ptlrpc_at_get_net_latency(struct ptlrpc_request *req);
2068 int ptl_send_rpc(struct ptlrpc_request *request, int noreply);
2069 int ptlrpc_register_rqbd(struct ptlrpc_request_buffer_desc *rqbd);
2072 /* ptlrpc/client.c */
2074 * Client-side portals API. Everything to send requests, receive replies,
2075 * request queues, request management, etc.
2078 void ptlrpc_request_committed(struct ptlrpc_request *req, int force);
2080 void ptlrpc_init_client(int req_portal, int rep_portal, char *name,
2081 struct ptlrpc_client *);
2082 void ptlrpc_cleanup_client(struct obd_import *imp);
2083 struct ptlrpc_connection *ptlrpc_uuid_to_connection(struct obd_uuid *uuid);
2085 int ptlrpc_queue_wait(struct ptlrpc_request *req);
2086 int ptlrpc_replay_req(struct ptlrpc_request *req);
2087 void ptlrpc_restart_req(struct ptlrpc_request *req);
2088 void ptlrpc_abort_inflight(struct obd_import *imp);
2089 void ptlrpc_cleanup_imp(struct obd_import *imp);
2090 void ptlrpc_abort_set(struct ptlrpc_request_set *set);
2092 struct ptlrpc_request_set *ptlrpc_prep_set(void);
2093 struct ptlrpc_request_set *ptlrpc_prep_fcset(int max, set_producer_func func,
2095 int ptlrpc_set_add_cb(struct ptlrpc_request_set *set,
2096 set_interpreter_func fn, void *data);
2097 int ptlrpc_check_set(const struct lu_env *env, struct ptlrpc_request_set *set);
2098 int ptlrpc_set_wait(struct ptlrpc_request_set *);
2099 void ptlrpc_mark_interrupted(struct ptlrpc_request *req);
2100 void ptlrpc_set_destroy(struct ptlrpc_request_set *);
2101 void ptlrpc_set_add_req(struct ptlrpc_request_set *, struct ptlrpc_request *);
2103 void ptlrpc_free_rq_pool(struct ptlrpc_request_pool *pool);
2104 int ptlrpc_add_rqs_to_pool(struct ptlrpc_request_pool *pool, int num_rq);
2106 struct ptlrpc_request_pool *
2107 ptlrpc_init_rq_pool(int, int,
2108 int (*populate_pool)(struct ptlrpc_request_pool *, int));
2110 void ptlrpc_at_set_req_timeout(struct ptlrpc_request *req);
2111 struct ptlrpc_request *ptlrpc_request_alloc(struct obd_import *imp,
2112 const struct req_format *format);
2113 struct ptlrpc_request *ptlrpc_request_alloc_pool(struct obd_import *imp,
2114 struct ptlrpc_request_pool *,
2115 const struct req_format *format);
2116 void ptlrpc_request_free(struct ptlrpc_request *request);
2117 int ptlrpc_request_pack(struct ptlrpc_request *request,
2118 __u32 version, int opcode);
2119 struct ptlrpc_request *ptlrpc_request_alloc_pack(struct obd_import *imp,
2120 const struct req_format *format,
2121 __u32 version, int opcode);
2122 int ptlrpc_request_bufs_pack(struct ptlrpc_request *request,
2123 __u32 version, int opcode, char **bufs,
2124 struct ptlrpc_cli_ctx *ctx);
2125 void ptlrpc_req_finished(struct ptlrpc_request *request);
2126 void ptlrpc_req_finished_with_imp_lock(struct ptlrpc_request *request);
2127 struct ptlrpc_request *ptlrpc_request_addref(struct ptlrpc_request *req);
2128 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_imp(struct ptlrpc_request *req,
2129 unsigned nfrags, unsigned max_brw,
2132 const struct ptlrpc_bulk_frag_ops
2135 int ptlrpc_prep_bulk_frag(struct ptlrpc_bulk_desc *desc,
2136 void *frag, int len);
2137 void __ptlrpc_prep_bulk_page(struct ptlrpc_bulk_desc *desc,
2138 struct page *page, int pageoffset, int len,
2140 static inline void ptlrpc_prep_bulk_page_pin(struct ptlrpc_bulk_desc *desc,
2141 struct page *page, int pageoffset,
2144 __ptlrpc_prep_bulk_page(desc, page, pageoffset, len, 1);
2147 static inline void ptlrpc_prep_bulk_page_nopin(struct ptlrpc_bulk_desc *desc,
2148 struct page *page, int pageoffset,
2151 __ptlrpc_prep_bulk_page(desc, page, pageoffset, len, 0);
2154 void ptlrpc_free_bulk(struct ptlrpc_bulk_desc *bulk);
2156 static inline void ptlrpc_release_bulk_page_pin(struct ptlrpc_bulk_desc *desc)
2160 for (i = 0; i < desc->bd_iov_count ; i++)
2161 page_cache_release(BD_GET_KIOV(desc, i).kiov_page);
2164 static inline void ptlrpc_release_bulk_noop(struct ptlrpc_bulk_desc *desc)
2168 void ptlrpc_retain_replayable_request(struct ptlrpc_request *req,
2169 struct obd_import *imp);
2170 __u64 ptlrpc_next_xid(void);
2171 __u64 ptlrpc_sample_next_xid(void);
2172 __u64 ptlrpc_req_xid(struct ptlrpc_request *request);
2174 /* Set of routines to run a function in ptlrpcd context */
2175 void *ptlrpcd_alloc_work(struct obd_import *imp,
2176 int (*cb)(const struct lu_env *, void *), void *data);
2177 void ptlrpcd_destroy_work(void *handler);
2178 int ptlrpcd_queue_work(void *handler);
2181 struct ptlrpc_service_buf_conf {
2182 /* nbufs is buffers # to allocate when growing the pool */
2183 unsigned int bc_nbufs;
2184 /* buffer size to post */
2185 unsigned int bc_buf_size;
2186 /* portal to listed for requests on */
2187 unsigned int bc_req_portal;
2188 /* portal of where to send replies to */
2189 unsigned int bc_rep_portal;
2190 /* maximum request size to be accepted for this service */
2191 unsigned int bc_req_max_size;
2192 /* maximum reply size this service can ever send */
2193 unsigned int bc_rep_max_size;
2196 struct ptlrpc_service_thr_conf {
2197 /* threadname should be 8 characters or less - 6 will be added on */
2199 /* threads increasing factor for each CPU */
2200 unsigned int tc_thr_factor;
2201 /* service threads # to start on each partition while initializing */
2202 unsigned int tc_nthrs_init;
2204 * low water of threads # upper-limit on each partition while running,
2205 * service availability may be impacted if threads number is lower
2206 * than this value. It can be ZERO if the service doesn't require
2207 * CPU affinity or there is only one partition.
2209 unsigned int tc_nthrs_base;
2210 /* "soft" limit for total threads number */
2211 unsigned int tc_nthrs_max;
2212 /* user specified threads number, it will be validated due to
2213 * other members of this structure. */
2214 unsigned int tc_nthrs_user;
2215 /* set NUMA node affinity for service threads */
2216 unsigned int tc_cpu_affinity;
2217 /* Tags for lu_context associated with service thread */
2221 struct ptlrpc_service_cpt_conf {
2222 struct cfs_cpt_table *cc_cptable;
2223 /* string pattern to describe CPTs for a service */
2227 struct ptlrpc_service_conf {
2230 /* soft watchdog timeout multiplifier to print stuck service traces */
2231 unsigned int psc_watchdog_factor;
2232 /* buffer information */
2233 struct ptlrpc_service_buf_conf psc_buf;
2234 /* thread information */
2235 struct ptlrpc_service_thr_conf psc_thr;
2236 /* CPU partition information */
2237 struct ptlrpc_service_cpt_conf psc_cpt;
2238 /* function table */
2239 struct ptlrpc_service_ops psc_ops;
2242 /* ptlrpc/service.c */
2244 * Server-side services API. Register/unregister service, request state
2245 * management, service thread management
2249 void ptlrpc_save_lock(struct ptlrpc_request *req,
2250 struct lustre_handle *lock, int mode, int no_ack);
2251 void ptlrpc_commit_replies(struct obd_export *exp);
2252 void ptlrpc_dispatch_difficult_reply(struct ptlrpc_reply_state *rs);
2253 void ptlrpc_schedule_difficult_reply(struct ptlrpc_reply_state *rs);
2254 int ptlrpc_hpreq_handler(struct ptlrpc_request *req);
2255 struct ptlrpc_service *ptlrpc_register_service(
2256 struct ptlrpc_service_conf *conf,
2257 struct proc_dir_entry *proc_entry);
2258 void ptlrpc_stop_all_threads(struct ptlrpc_service *svc);
2260 int ptlrpc_start_threads(struct ptlrpc_service *svc);
2261 int ptlrpc_unregister_service(struct ptlrpc_service *service);
2262 int liblustre_check_services(void *arg);
2263 void ptlrpc_daemonize(char *name);
2264 int ptlrpc_service_health_check(struct ptlrpc_service *);
2265 void ptlrpc_server_drop_request(struct ptlrpc_request *req);
2266 void ptlrpc_request_change_export(struct ptlrpc_request *req,
2267 struct obd_export *export);
2268 void ptlrpc_update_export_timer(struct obd_export *exp, long extra_delay);
2270 int ptlrpc_hr_init(void);
2271 void ptlrpc_hr_fini(void);
2275 /* ptlrpc/import.c */
2280 int ptlrpc_connect_import(struct obd_import *imp);
2281 int ptlrpc_init_import(struct obd_import *imp);
2282 int ptlrpc_disconnect_import(struct obd_import *imp, int noclose);
2283 int ptlrpc_import_recovery_state_machine(struct obd_import *imp);
2284 void deuuidify(char *uuid, const char *prefix, char **uuid_start,
2286 void ptlrpc_import_enter_resend(struct obd_import *imp);
2287 /* ptlrpc/pack_generic.c */
2288 int ptlrpc_reconnect_import(struct obd_import *imp);
2292 * ptlrpc msg buffer and swab interface
2296 int ptlrpc_buf_need_swab(struct ptlrpc_request *req, const int inout,
2298 void ptlrpc_buf_set_swabbed(struct ptlrpc_request *req, const int inout,
2300 int ptlrpc_unpack_rep_msg(struct ptlrpc_request *req, int len);
2301 int ptlrpc_unpack_req_msg(struct ptlrpc_request *req, int len);
2303 int lustre_msg_check_version(struct lustre_msg *msg, __u32 version);
2304 void lustre_init_msg_v2(struct lustre_msg_v2 *msg, int count, __u32 *lens,
2306 int lustre_pack_request(struct ptlrpc_request *, __u32 magic, int count,
2307 __u32 *lens, char **bufs);
2308 int lustre_pack_reply(struct ptlrpc_request *, int count, __u32 *lens,
2310 int lustre_pack_reply_v2(struct ptlrpc_request *req, int count,
2311 __u32 *lens, char **bufs, int flags);
2312 #define LPRFL_EARLY_REPLY 1
2313 int lustre_pack_reply_flags(struct ptlrpc_request *, int count, __u32 *lens,
2314 char **bufs, int flags);
2315 int lustre_shrink_msg(struct lustre_msg *msg, int segment,
2316 unsigned int newlen, int move_data);
2317 void lustre_free_reply_state(struct ptlrpc_reply_state *rs);
2318 int __lustre_unpack_msg(struct lustre_msg *m, int len);
2319 __u32 lustre_msg_hdr_size(__u32 magic, __u32 count);
2320 __u32 lustre_msg_size(__u32 magic, int count, __u32 *lengths);
2321 __u32 lustre_msg_size_v2(int count, __u32 *lengths);
2322 __u32 lustre_packed_msg_size(struct lustre_msg *msg);
2323 __u32 lustre_msg_early_size(void);
2324 void *lustre_msg_buf_v2(struct lustre_msg_v2 *m, __u32 n, __u32 min_size);
2325 void *lustre_msg_buf(struct lustre_msg *m, __u32 n, __u32 minlen);
2326 __u32 lustre_msg_buflen(struct lustre_msg *m, __u32 n);
2327 void lustre_msg_set_buflen(struct lustre_msg *m, __u32 n, __u32 len);
2328 __u32 lustre_msg_bufcount(struct lustre_msg *m);
2329 char *lustre_msg_string(struct lustre_msg *m, __u32 n, __u32 max_len);
2330 __u32 lustre_msghdr_get_flags(struct lustre_msg *msg);
2331 void lustre_msghdr_set_flags(struct lustre_msg *msg, __u32 flags);
2332 __u32 lustre_msg_get_flags(struct lustre_msg *msg);
2333 void lustre_msg_add_flags(struct lustre_msg *msg, __u32 flags);
2334 void lustre_msg_set_flags(struct lustre_msg *msg, __u32 flags);
2335 void lustre_msg_clear_flags(struct lustre_msg *msg, __u32 flags);
2336 __u32 lustre_msg_get_op_flags(struct lustre_msg *msg);
2337 void lustre_msg_add_op_flags(struct lustre_msg *msg, __u32 flags);
2338 struct lustre_handle *lustre_msg_get_handle(struct lustre_msg *msg);
2339 __u32 lustre_msg_get_type(struct lustre_msg *msg);
2340 __u32 lustre_msg_get_version(struct lustre_msg *msg);
2341 void lustre_msg_add_version(struct lustre_msg *msg, __u32 version);
2342 __u32 lustre_msg_get_opc(struct lustre_msg *msg);
2343 __u64 lustre_msg_get_last_xid(struct lustre_msg *msg);
2344 __u16 lustre_msg_get_tag(struct lustre_msg *msg);
2345 __u64 lustre_msg_get_last_committed(struct lustre_msg *msg);
2346 __u64 *lustre_msg_get_versions(struct lustre_msg *msg);
2347 __u64 lustre_msg_get_transno(struct lustre_msg *msg);
2348 __u64 lustre_msg_get_slv(struct lustre_msg *msg);
2349 __u32 lustre_msg_get_limit(struct lustre_msg *msg);
2350 void lustre_msg_set_slv(struct lustre_msg *msg, __u64 slv);
2351 void lustre_msg_set_limit(struct lustre_msg *msg, __u64 limit);
2352 int lustre_msg_get_status(struct lustre_msg *msg);
2353 __u32 lustre_msg_get_conn_cnt(struct lustre_msg *msg);
2354 __u32 lustre_msg_get_magic(struct lustre_msg *msg);
2355 __u32 lustre_msg_get_timeout(struct lustre_msg *msg);
2356 __u32 lustre_msg_get_service_time(struct lustre_msg *msg);
2357 char *lustre_msg_get_jobid(struct lustre_msg *msg);
2358 __u32 lustre_msg_get_cksum(struct lustre_msg *msg);
2359 __u64 lustre_msg_get_mbits(struct lustre_msg *msg);
2360 __u32 lustre_msg_calc_cksum(struct lustre_msg *msg);
2361 void lustre_msg_set_handle(struct lustre_msg *msg,struct lustre_handle *handle);
2362 void lustre_msg_set_type(struct lustre_msg *msg, __u32 type);
2363 void lustre_msg_set_opc(struct lustre_msg *msg, __u32 opc);
2364 void lustre_msg_set_last_xid(struct lustre_msg *msg, __u64 last_xid);
2365 void lustre_msg_set_tag(struct lustre_msg *msg, __u16 tag);
2366 void lustre_msg_set_last_committed(struct lustre_msg *msg,__u64 last_committed);
2367 void lustre_msg_set_versions(struct lustre_msg *msg, __u64 *versions);
2368 void lustre_msg_set_transno(struct lustre_msg *msg, __u64 transno);
2369 void lustre_msg_set_status(struct lustre_msg *msg, __u32 status);
2370 void lustre_msg_set_conn_cnt(struct lustre_msg *msg, __u32 conn_cnt);
2371 void ptlrpc_req_set_repsize(struct ptlrpc_request *req, int count, __u32 *sizes);
2372 void ptlrpc_request_set_replen(struct ptlrpc_request *req);
2373 void lustre_msg_set_timeout(struct lustre_msg *msg, __u32 timeout);
2374 void lustre_msg_set_service_time(struct lustre_msg *msg, __u32 service_time);
2375 void lustre_msg_set_jobid(struct lustre_msg *msg, char *jobid);
2376 void lustre_msg_set_cksum(struct lustre_msg *msg, __u32 cksum);
2377 void lustre_msg_set_mbits(struct lustre_msg *msg, __u64 mbits);
2380 lustre_shrink_reply(struct ptlrpc_request *req, int segment,
2381 unsigned int newlen, int move_data)
2383 LASSERT(req->rq_reply_state);
2384 LASSERT(req->rq_repmsg);
2385 req->rq_replen = lustre_shrink_msg(req->rq_repmsg, segment,
2389 #ifdef LUSTRE_TRANSLATE_ERRNOS
2391 static inline int ptlrpc_status_hton(int h)
2394 * Positive errnos must be network errnos, such as LUSTRE_EDEADLK,
2395 * ELDLM_LOCK_ABORTED, etc.
2398 return -lustre_errno_hton(-h);
2403 static inline int ptlrpc_status_ntoh(int n)
2406 * See the comment in ptlrpc_status_hton().
2409 return -lustre_errno_ntoh(-n);
2416 #define ptlrpc_status_hton(h) (h)
2417 #define ptlrpc_status_ntoh(n) (n)
2422 /** Change request phase of \a req to \a new_phase */
2424 ptlrpc_rqphase_move(struct ptlrpc_request *req, enum rq_phase new_phase)
2426 if (req->rq_phase == new_phase)
2429 if (new_phase == RQ_PHASE_UNREGISTERING) {
2430 req->rq_next_phase = req->rq_phase;
2432 atomic_inc(&req->rq_import->imp_unregistering);
2435 if (req->rq_phase == RQ_PHASE_UNREGISTERING) {
2437 atomic_dec(&req->rq_import->imp_unregistering);
2440 DEBUG_REQ(D_INFO, req, "move req \"%s\" -> \"%s\"",
2441 ptlrpc_rqphase2str(req), ptlrpc_phase2str(new_phase));
2443 req->rq_phase = new_phase;
2447 * Returns true if request \a req got early reply and hard deadline is not met
2450 ptlrpc_client_early(struct ptlrpc_request *req)
2452 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
2453 req->rq_reply_deadline > cfs_time_current_sec())
2455 return req->rq_early;
2459 * Returns true if we got real reply from server for this request
2462 ptlrpc_client_replied(struct ptlrpc_request *req)
2464 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
2465 req->rq_reply_deadline > cfs_time_current_sec())
2467 return req->rq_replied;
2470 /** Returns true if request \a req is in process of receiving server reply */
2472 ptlrpc_client_recv(struct ptlrpc_request *req)
2474 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
2475 req->rq_reply_deadline > cfs_time_current_sec())
2477 return req->rq_receiving_reply;
2481 ptlrpc_client_recv_or_unlink(struct ptlrpc_request *req)
2485 spin_lock(&req->rq_lock);
2486 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
2487 req->rq_reply_deadline > cfs_time_current_sec()) {
2488 spin_unlock(&req->rq_lock);
2491 rc = !req->rq_req_unlinked || !req->rq_reply_unlinked ||
2492 req->rq_receiving_reply;
2493 spin_unlock(&req->rq_lock);
2498 ptlrpc_client_wake_req(struct ptlrpc_request *req)
2500 if (req->rq_set == NULL)
2501 wake_up(&req->rq_reply_waitq);
2503 wake_up(&req->rq_set->set_waitq);
2507 ptlrpc_rs_addref(struct ptlrpc_reply_state *rs)
2509 LASSERT(atomic_read(&rs->rs_refcount) > 0);
2510 atomic_inc(&rs->rs_refcount);
2514 ptlrpc_rs_decref(struct ptlrpc_reply_state *rs)
2516 LASSERT(atomic_read(&rs->rs_refcount) > 0);
2517 if (atomic_dec_and_test(&rs->rs_refcount))
2518 lustre_free_reply_state(rs);
2521 /* Should only be called once per req */
2522 static inline void ptlrpc_req_drop_rs(struct ptlrpc_request *req)
2524 if (req->rq_reply_state == NULL)
2525 return; /* shouldn't occur */
2526 ptlrpc_rs_decref(req->rq_reply_state);
2527 req->rq_reply_state = NULL;
2528 req->rq_repmsg = NULL;
2531 static inline __u32 lustre_request_magic(struct ptlrpc_request *req)
2533 return lustre_msg_get_magic(req->rq_reqmsg);
2536 static inline int ptlrpc_req_get_repsize(struct ptlrpc_request *req)
2538 switch (req->rq_reqmsg->lm_magic) {
2539 case LUSTRE_MSG_MAGIC_V2:
2540 return req->rq_reqmsg->lm_repsize;
2542 LASSERTF(0, "incorrect message magic: %08x\n",
2543 req->rq_reqmsg->lm_magic);
2548 static inline int ptlrpc_send_limit_expired(struct ptlrpc_request *req)
2550 if (req->rq_delay_limit != 0 &&
2551 cfs_time_before(cfs_time_add(req->rq_queued_time,
2552 cfs_time_seconds(req->rq_delay_limit)),
2553 cfs_time_current())) {
2559 static inline int ptlrpc_no_resend(struct ptlrpc_request *req)
2561 if (!req->rq_no_resend && ptlrpc_send_limit_expired(req)) {
2562 spin_lock(&req->rq_lock);
2563 req->rq_no_resend = 1;
2564 spin_unlock(&req->rq_lock);
2566 return req->rq_no_resend;
2570 ptlrpc_server_get_timeout(struct ptlrpc_service_part *svcpt)
2572 int at = AT_OFF ? 0 : at_get(&svcpt->scp_at_estimate);
2574 return svcpt->scp_service->srv_watchdog_factor *
2575 max_t(int, at, obd_timeout);
2578 static inline struct ptlrpc_service *
2579 ptlrpc_req2svc(struct ptlrpc_request *req)
2581 LASSERT(req->rq_rqbd != NULL);
2582 return req->rq_rqbd->rqbd_svcpt->scp_service;
2585 /* ldlm/ldlm_lib.c */
2587 * Target client logic
2590 int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg);
2591 int client_obd_cleanup(struct obd_device *obddev);
2592 int client_connect_import(const struct lu_env *env,
2593 struct obd_export **exp, struct obd_device *obd,
2594 struct obd_uuid *cluuid, struct obd_connect_data *,
2596 int client_disconnect_export(struct obd_export *exp);
2597 int client_import_add_conn(struct obd_import *imp, struct obd_uuid *uuid,
2599 int client_import_del_conn(struct obd_import *imp, struct obd_uuid *uuid);
2600 int client_import_find_conn(struct obd_import *imp, lnet_nid_t peer,
2601 struct obd_uuid *uuid);
2602 int import_set_conn_priority(struct obd_import *imp, struct obd_uuid *uuid);
2603 void client_destroy_import(struct obd_import *imp);
2606 #ifdef HAVE_SERVER_SUPPORT
2607 int server_disconnect_export(struct obd_export *exp);
2610 /* ptlrpc/pinger.c */
2612 * Pinger API (client side only)
2615 enum timeout_event {
2618 struct timeout_item;
2619 typedef int (*timeout_cb_t)(struct timeout_item *, void *);
2620 int ptlrpc_pinger_add_import(struct obd_import *imp);
2621 int ptlrpc_pinger_del_import(struct obd_import *imp);
2622 int ptlrpc_add_timeout_client(int time, enum timeout_event event,
2623 timeout_cb_t cb, void *data,
2624 struct list_head *obd_list);
2625 int ptlrpc_del_timeout_client(struct list_head *obd_list,
2626 enum timeout_event event);
2627 struct ptlrpc_request * ptlrpc_prep_ping(struct obd_import *imp);
2628 int ptlrpc_obd_ping(struct obd_device *obd);
2629 void ping_evictor_start(void);
2630 void ping_evictor_stop(void);
2631 void ptlrpc_pinger_ir_up(void);
2632 void ptlrpc_pinger_ir_down(void);
2634 int ptlrpc_pinger_suppress_pings(void);
2636 /* ptlrpc/ptlrpcd.c */
2637 void ptlrpcd_stop(struct ptlrpcd_ctl *pc, int force);
2638 void ptlrpcd_free(struct ptlrpcd_ctl *pc);
2639 void ptlrpcd_wake(struct ptlrpc_request *req);
2640 void ptlrpcd_add_req(struct ptlrpc_request *req);
2641 void ptlrpcd_add_rqset(struct ptlrpc_request_set *set);
2642 int ptlrpcd_addref(void);
2643 void ptlrpcd_decref(void);
2645 /* ptlrpc/lproc_ptlrpc.c */
2647 * procfs output related functions
2650 const char* ll_opcode2str(__u32 opcode);
2651 #ifdef CONFIG_PROC_FS
2652 void ptlrpc_lprocfs_register_obd(struct obd_device *obd);
2653 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd);
2654 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes);
2656 static inline void ptlrpc_lprocfs_register_obd(struct obd_device *obd) {}
2657 static inline void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd) {}
2658 static inline void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes) {}
2662 /* ptlrpc/llog_server.c */
2663 int llog_origin_handle_open(struct ptlrpc_request *req);
2664 int llog_origin_handle_destroy(struct ptlrpc_request *req);
2665 int llog_origin_handle_prev_block(struct ptlrpc_request *req);
2666 int llog_origin_handle_next_block(struct ptlrpc_request *req);
2667 int llog_origin_handle_read_header(struct ptlrpc_request *req);
2668 int llog_origin_handle_close(struct ptlrpc_request *req);
2670 /* ptlrpc/llog_client.c */
2671 extern struct llog_operations llog_client_ops;