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, 2014, 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
459 #define OSS_NTHRS_MAX 512
461 /* threads for handling "create" request */
462 #define OSS_CR_THR_FACTOR 1
463 #define OSS_CR_NTHRS_INIT PTLRPC_NTHRS_INIT
464 #define OSS_CR_NTHRS_BASE 8
465 #define OSS_CR_NTHRS_MAX 64
468 * OST_IO_MAXREQSIZE ~=
469 * lustre_msg + ptlrpc_body + obdo + obd_ioobj +
470 * DT_MAX_BRW_PAGES * niobuf_remote
472 * - single object with 16 pages is 512 bytes
473 * - OST_IO_MAXREQSIZE must be at least 1 page of cookies plus some spillover
474 * - Must be a multiple of 1024
475 * - actual size is about 18K
477 #define _OST_MAXREQSIZE_SUM (sizeof(struct lustre_msg) + \
478 sizeof(struct ptlrpc_body) + \
479 sizeof(struct obdo) + \
480 sizeof(struct obd_ioobj) + \
481 sizeof(struct niobuf_remote) * DT_MAX_BRW_PAGES)
483 * FIEMAP request can be 4K+ for now
485 #define OST_MAXREQSIZE (16 * 1024)
486 #define OST_IO_MAXREQSIZE max_t(int, OST_MAXREQSIZE, \
487 (((_OST_MAXREQSIZE_SUM - 1) | (1024 - 1)) + 1))
489 #define OST_MAXREPSIZE (9 * 1024)
490 #define OST_IO_MAXREPSIZE OST_MAXREPSIZE
493 /** OST_BUFSIZE = max_reqsize + max sptlrpc payload size */
494 #define OST_BUFSIZE max_t(int, OST_MAXREQSIZE + 1024, 16 * 1024)
496 * OST_IO_MAXREQSIZE is 18K, giving extra 46K can increase buffer utilization
497 * rate of request buffer, please check comment of MDS_LOV_BUFSIZE for details.
499 #define OST_IO_BUFSIZE max_t(int, OST_IO_MAXREQSIZE + 1024, 64 * 1024)
501 /* Macro to hide a typecast. */
502 #define ptlrpc_req_async_args(req) ((void *)&req->rq_async_args)
504 struct ptlrpc_replay_async_args {
510 * Structure to single define portal connection.
512 struct ptlrpc_connection {
513 /** linkage for connections hash table */
514 struct hlist_node c_hash;
515 /** Our own lnet nid for this connection */
517 /** Remote side nid for this connection */
518 lnet_process_id_t c_peer;
519 /** UUID of the other side */
520 struct obd_uuid c_remote_uuid;
521 /** reference counter for this connection */
525 /** Client definition for PortalRPC */
526 struct ptlrpc_client {
527 /** What lnet portal does this client send messages to by default */
528 __u32 cli_request_portal;
529 /** What portal do we expect replies on */
530 __u32 cli_reply_portal;
531 /** Name of the client */
535 /** state flags of requests */
536 /* XXX only ones left are those used by the bulk descs as well! */
537 #define PTL_RPC_FL_INTR (1 << 0) /* reply wait was interrupted by user */
538 #define PTL_RPC_FL_TIMEOUT (1 << 7) /* request timed out waiting for reply */
540 #define REQ_MAX_ACK_LOCKS 8
542 union ptlrpc_async_args {
544 * Scratchpad for passing args to completion interpreter. Users
545 * cast to the struct of their choosing, and CLASSERT that this is
546 * big enough. For _tons_ of context, OBD_ALLOC a struct and store
547 * a pointer to it here. The pointer_arg ensures this struct is at
548 * least big enough for that.
550 void *pointer_arg[11];
554 struct ptlrpc_request_set;
555 typedef int (*set_interpreter_func)(struct ptlrpc_request_set *, void *, int);
556 typedef int (*set_producer_func)(struct ptlrpc_request_set *, void *);
559 * Definition of request set structure.
560 * Request set is a list of requests (not necessary to the same target) that
561 * once populated with RPCs could be sent in parallel.
562 * There are two kinds of request sets. General purpose and with dedicated
563 * serving thread. Example of the latter is ptlrpcd set.
564 * For general purpose sets once request set started sending it is impossible
565 * to add new requests to such set.
566 * Provides a way to call "completion callbacks" when all requests in the set
569 struct ptlrpc_request_set {
570 atomic_t set_refcount;
571 /** number of in queue requests */
572 atomic_t set_new_count;
573 /** number of uncompleted requests */
574 atomic_t set_remaining;
575 /** wait queue to wait on for request events */
576 wait_queue_head_t set_waitq;
577 wait_queue_head_t *set_wakeup_ptr;
578 /** List of requests in the set */
579 struct list_head set_requests;
581 * List of completion callbacks to be called when the set is completed
582 * This is only used if \a set_interpret is NULL.
583 * Links struct ptlrpc_set_cbdata.
585 struct list_head set_cblist;
586 /** Completion callback, if only one. */
587 set_interpreter_func set_interpret;
588 /** opaq argument passed to completion \a set_interpret callback. */
591 * Lock for \a set_new_requests manipulations
592 * locked so that any old caller can communicate requests to
593 * the set holder who can then fold them into the lock-free set
595 spinlock_t set_new_req_lock;
596 /** List of new yet unsent requests. Only used with ptlrpcd now. */
597 struct list_head set_new_requests;
599 /** rq_status of requests that have been freed already */
601 /** Additional fields used by the flow control extension */
602 /** Maximum number of RPCs in flight */
603 int set_max_inflight;
604 /** Callback function used to generate RPCs */
605 set_producer_func set_producer;
606 /** opaq argument passed to the producer callback */
607 void *set_producer_arg;
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 /** Size of the state */
675 /** Transaction number */
679 struct obd_export *rs_export;
680 struct ptlrpc_service_part *rs_svcpt;
681 /** Lnet metadata handle for the reply */
682 lnet_handle_md_t rs_md_h;
683 atomic_t rs_refcount;
685 /** Context for the sevice thread */
686 struct ptlrpc_svc_ctx *rs_svc_ctx;
687 /** Reply buffer (actually sent to the client), encoded if needed */
688 struct lustre_msg *rs_repbuf; /* wrapper */
689 /** Size of the reply buffer */
690 int rs_repbuf_len; /* wrapper buf length */
691 /** Size of the reply message */
692 int rs_repdata_len; /* wrapper msg length */
694 * Actual reply message. Its content is encrupted (if needed) to
695 * produce reply buffer for actual sending. In simple case
696 * of no network encryption we jus set \a rs_repbuf to \a rs_msg
698 struct lustre_msg *rs_msg; /* reply message */
700 /** Number of locks awaiting client ACK */
702 /** Handles of locks awaiting client reply ACK */
703 struct lustre_handle rs_locks[RS_MAX_LOCKS];
704 /** Lock modes of locks in \a rs_locks */
705 enum ldlm_mode rs_modes[RS_MAX_LOCKS];
708 struct ptlrpc_thread;
712 RQ_PHASE_NEW = 0xebc0de00,
713 RQ_PHASE_RPC = 0xebc0de01,
714 RQ_PHASE_BULK = 0xebc0de02,
715 RQ_PHASE_INTERPRET = 0xebc0de03,
716 RQ_PHASE_COMPLETE = 0xebc0de04,
717 RQ_PHASE_UNREGISTERING = 0xebc0de05,
718 RQ_PHASE_UNDEFINED = 0xebc0de06
721 /** Type of request interpreter call-back */
722 typedef int (*ptlrpc_interpterer_t)(const struct lu_env *env,
723 struct ptlrpc_request *req,
725 /** Type of request resend call-back */
726 typedef void (*ptlrpc_resend_cb_t)(struct ptlrpc_request *req,
730 * Definition of request pool structure.
731 * The pool is used to store empty preallocated requests for the case
732 * when we would actually need to send something without performing
733 * any allocations (to avoid e.g. OOM).
735 struct ptlrpc_request_pool {
736 /** Locks the list */
738 /** list of ptlrpc_request structs */
739 struct list_head prp_req_list;
740 /** Maximum message size that would fit into a rquest from this pool */
742 /** Function to allocate more requests for this pool */
743 int (*prp_populate)(struct ptlrpc_request_pool *, int);
751 #include <lustre_nrs.h>
754 * Basic request prioritization operations structure.
755 * The whole idea is centered around locks and RPCs that might affect locks.
756 * When a lock is contended we try to give priority to RPCs that might lead
757 * to fastest release of that lock.
758 * Currently only implemented for OSTs only in a way that makes all
759 * IO and truncate RPCs that are coming from a locked region where a lock is
760 * contended a priority over other requests.
762 struct ptlrpc_hpreq_ops {
764 * Check if the lock handle of the given lock is the same as
765 * taken from the request.
767 int (*hpreq_lock_match)(struct ptlrpc_request *, struct ldlm_lock *);
769 * Check if the request is a high priority one.
771 int (*hpreq_check)(struct ptlrpc_request *);
773 * Called after the request has been handled.
775 void (*hpreq_fini)(struct ptlrpc_request *);
778 struct ptlrpc_cli_req {
779 /** For bulk requests on client only: bulk descriptor */
780 struct ptlrpc_bulk_desc *cr_bulk;
781 /** optional time limit for send attempts */
782 cfs_duration_t cr_delay_limit;
783 /** time request was first queued */
784 cfs_time_t cr_queued_time;
785 /** request sent timeval */
786 struct timeval cr_sent_tv;
787 /** time for request really sent out */
789 /** when req reply unlink must finish. */
790 time_t cr_reply_deadline;
791 /** when req bulk unlink must finish. */
792 time_t cr_bulk_deadline;
793 /** Portal to which this request would be sent */
795 /** Portal where to wait for reply and where reply would be sent */
797 /** request resending number */
798 unsigned int cr_resend_nr;
799 /** What was import generation when this request was sent */
801 enum lustre_imp_state cr_send_state;
802 /** Per-request waitq introduced by bug 21938 for recovery waiting */
803 wait_queue_head_t cr_set_waitq;
804 /** Link item for request set lists */
805 struct list_head cr_set_chain;
806 /** link to waited ctx */
807 struct list_head cr_ctx_chain;
809 /** client's half ctx */
810 struct ptlrpc_cli_ctx *cr_cli_ctx;
811 /** Link back to the request set */
812 struct ptlrpc_request_set *cr_set;
813 /** outgoing request MD handle */
814 lnet_handle_md_t cr_req_md_h;
815 /** request-out callback parameter */
816 struct ptlrpc_cb_id cr_req_cbid;
817 /** incoming reply MD handle */
818 lnet_handle_md_t cr_reply_md_h;
819 wait_queue_head_t cr_reply_waitq;
820 /** reply callback parameter */
821 struct ptlrpc_cb_id cr_reply_cbid;
822 /** Async completion handler, called when reply is received */
823 ptlrpc_interpterer_t cr_reply_interp;
824 /** Resend handler, called when request is resend to update RPC data */
825 ptlrpc_resend_cb_t cr_resend_cb;
826 /** Async completion context */
827 union ptlrpc_async_args cr_async_args;
828 /** Opaq data for replay and commit callbacks. */
831 * Commit callback, called when request is committed and about to be
834 void (*cr_commit_cb)(struct ptlrpc_request *);
835 /** Replay callback, called after request is replayed at recovery */
836 void (*cr_replay_cb)(struct ptlrpc_request *);
839 /** client request member alias */
840 /* NB: these alias should NOT be used by any new code, instead they should
841 * be removed step by step to avoid potential abuse */
842 #define rq_bulk rq_cli.cr_bulk
843 #define rq_delay_limit rq_cli.cr_delay_limit
844 #define rq_queued_time rq_cli.cr_queued_time
845 #define rq_sent_tv rq_cli.cr_sent_tv
846 #define rq_real_sent rq_cli.cr_sent_out
847 #define rq_reply_deadline rq_cli.cr_reply_deadline
848 #define rq_bulk_deadline rq_cli.cr_bulk_deadline
849 #define rq_nr_resend rq_cli.cr_resend_nr
850 #define rq_request_portal rq_cli.cr_req_ptl
851 #define rq_reply_portal rq_cli.cr_rep_ptl
852 #define rq_import_generation rq_cli.cr_imp_gen
853 #define rq_send_state rq_cli.cr_send_state
854 #define rq_set_chain rq_cli.cr_set_chain
855 #define rq_ctx_chain rq_cli.cr_ctx_chain
856 #define rq_set rq_cli.cr_set
857 #define rq_set_waitq rq_cli.cr_set_waitq
858 #define rq_cli_ctx rq_cli.cr_cli_ctx
859 #define rq_req_md_h rq_cli.cr_req_md_h
860 #define rq_req_cbid rq_cli.cr_req_cbid
861 #define rq_reply_md_h rq_cli.cr_reply_md_h
862 #define rq_reply_waitq rq_cli.cr_reply_waitq
863 #define rq_reply_cbid rq_cli.cr_reply_cbid
864 #define rq_interpret_reply rq_cli.cr_reply_interp
865 #define rq_resend_cb rq_cli.cr_resend_cb
866 #define rq_async_args rq_cli.cr_async_args
867 #define rq_cb_data rq_cli.cr_cb_data
868 #define rq_commit_cb rq_cli.cr_commit_cb
869 #define rq_replay_cb rq_cli.cr_replay_cb
871 struct ptlrpc_srv_req {
872 /** initial thread servicing this request */
873 struct ptlrpc_thread *sr_svc_thread;
875 * Server side list of incoming unserved requests sorted by arrival
876 * time. Traversed from time to time to notice about to expire
877 * requests and sent back "early replies" to clients to let them
878 * know server is alive and well, just very busy to service their
881 struct list_head sr_timed_list;
882 /** server-side per-export list */
883 struct list_head sr_exp_list;
884 /** server-side history, used for debuging purposes. */
885 struct list_head sr_hist_list;
886 /** history sequence # */
888 /** the index of service's srv_at_array into which request is linked */
892 /** authed uid mapped to */
893 uid_t sr_auth_mapped_uid;
894 /** RPC is generated from what part of Lustre */
895 enum lustre_sec_part sr_sp_from;
896 /** request session context */
897 struct lu_context sr_ses;
901 /** stub for NRS request */
902 struct ptlrpc_nrs_request sr_nrq;
904 /** request arrival time */
905 struct timeval sr_arrival_time;
906 /** server's half ctx */
907 struct ptlrpc_svc_ctx *sr_svc_ctx;
908 /** (server side), pointed directly into req buffer */
909 struct ptlrpc_user_desc *sr_user_desc;
910 /** separated reply state */
911 struct ptlrpc_reply_state *sr_reply_state;
912 /** server-side hp handlers */
913 struct ptlrpc_hpreq_ops *sr_ops;
914 /** incoming request buffer */
915 struct ptlrpc_request_buffer_desc *sr_rqbd;
918 /** server request member alias */
919 /* NB: these alias should NOT be used by any new code, instead they should
920 * be removed step by step to avoid potential abuse */
921 #define rq_svc_thread rq_srv.sr_svc_thread
922 #define rq_timed_list rq_srv.sr_timed_list
923 #define rq_exp_list rq_srv.sr_exp_list
924 #define rq_history_list rq_srv.sr_hist_list
925 #define rq_history_seq rq_srv.sr_hist_seq
926 #define rq_at_index rq_srv.sr_at_index
927 #define rq_auth_uid rq_srv.sr_auth_uid
928 #define rq_auth_mapped_uid rq_srv.sr_auth_mapped_uid
929 #define rq_sp_from rq_srv.sr_sp_from
930 #define rq_session rq_srv.sr_ses
931 #define rq_nrq rq_srv.sr_nrq
932 #define rq_arrival_time rq_srv.sr_arrival_time
933 #define rq_reply_state rq_srv.sr_reply_state
934 #define rq_svc_ctx rq_srv.sr_svc_ctx
935 #define rq_user_desc rq_srv.sr_user_desc
936 #define rq_ops rq_srv.sr_ops
937 #define rq_rqbd rq_srv.sr_rqbd
940 * Represents remote procedure call.
942 * This is a staple structure used by everybody wanting to send a request
945 struct ptlrpc_request {
946 /* Request type: one of PTL_RPC_MSG_* */
948 /** Result of request processing */
951 * Linkage item through which this request is included into
952 * sending/delayed lists on client and into rqbd list on server
954 struct list_head rq_list;
955 /** Lock to protect request flags and some other important bits, like
959 /** client-side flags are serialized by rq_lock @{ */
960 unsigned int rq_intr:1, rq_replied:1, rq_err:1,
961 rq_timedout:1, rq_resend:1, rq_restart:1,
963 * when ->rq_replay is set, request is kept by the client even
964 * after server commits corresponding transaction. This is
965 * used for operations that require sequence of multiple
966 * requests to be replayed. The only example currently is file
967 * open/close. When last request in such a sequence is
968 * committed, ->rq_replay is cleared on all requests in the
972 rq_no_resend:1, rq_waiting:1, rq_receiving_reply:1,
973 rq_no_delay:1, rq_net_err:1, rq_wait_ctx:1,
975 rq_req_unlinked:1, /* unlinked request buffer from lnet */
976 rq_reply_unlinked:1, /* unlinked reply buffer from lnet */
977 rq_memalloc:1, /* req originated from "kswapd" */
979 rq_reply_truncated:1,
980 /** whether the "rq_set" is a valid one */
983 /** do not resend request on -EINPROGRESS */
984 rq_no_retry_einprogress:1,
985 /* allow the req to be sent if the import is in recovery
988 /* bulk request, sent to server, but uncommitted */
992 /** server-side flags @{ */
994 rq_hp:1, /**< high priority RPC */
995 rq_at_linked:1, /**< link into service's srv_at_array */
996 rq_packed_final:1; /**< packed final reply */
999 /** one of RQ_PHASE_* */
1000 enum rq_phase rq_phase;
1001 /** one of RQ_PHASE_* to be used next */
1002 enum rq_phase rq_next_phase;
1004 * client-side refcount for SENT race, server-side refcounf
1005 * for multiple replies
1007 atomic_t rq_refcount;
1010 * !rq_truncate : # reply bytes actually received,
1011 * rq_truncate : required repbuf_len for resend
1013 int rq_nob_received;
1014 /** Request length */
1018 /** Pool if request is from preallocated list */
1019 struct ptlrpc_request_pool *rq_pool;
1020 /** Request message - what client sent */
1021 struct lustre_msg *rq_reqmsg;
1022 /** Reply message - server response */
1023 struct lustre_msg *rq_repmsg;
1024 /** Transaction number */
1028 /** bulk match bits */
1031 * List item to for replay list. Not yet commited requests get linked
1033 * Also see \a rq_replay comment above.
1034 * It's also link chain on obd_export::exp_req_replay_queue
1036 struct list_head rq_replay_list;
1037 /** non-shared members for client & server request*/
1039 struct ptlrpc_cli_req rq_cli;
1040 struct ptlrpc_srv_req rq_srv;
1043 * security and encryption data
1045 /** description of flavors for client & server */
1046 struct sptlrpc_flavor rq_flvr;
1048 /* client/server security flags */
1050 rq_ctx_init:1, /* context initiation */
1051 rq_ctx_fini:1, /* context destroy */
1052 rq_bulk_read:1, /* request bulk read */
1053 rq_bulk_write:1, /* request bulk write */
1054 /* server authentication flags */
1055 rq_auth_gss:1, /* authenticated by gss */
1056 rq_auth_remote:1, /* authed as remote user */
1057 rq_auth_usr_root:1, /* authed as root */
1058 rq_auth_usr_mdt:1, /* authed as mdt */
1059 rq_auth_usr_ost:1, /* authed as ost */
1060 /* security tfm flags */
1063 /* doesn't expect reply FIXME */
1065 rq_pill_init:1, /* pill initialized */
1066 rq_srv_req:1; /* server request */
1069 /** various buffer pointers */
1070 struct lustre_msg *rq_reqbuf; /**< req wrapper */
1071 char *rq_repbuf; /**< rep buffer */
1072 struct lustre_msg *rq_repdata; /**< rep wrapper msg */
1073 /** only in priv mode */
1074 struct lustre_msg *rq_clrbuf;
1075 int rq_reqbuf_len; /* req wrapper buf len */
1076 int rq_reqdata_len; /* req wrapper msg len */
1077 int rq_repbuf_len; /* rep buffer len */
1078 int rq_repdata_len; /* rep wrapper msg len */
1079 int rq_clrbuf_len; /* only in priv mode */
1080 int rq_clrdata_len; /* only in priv mode */
1082 /** early replies go to offset 0, regular replies go after that */
1083 unsigned int rq_reply_off;
1086 /** Fields that help to see if request and reply were swabbed or not */
1087 __u32 rq_req_swab_mask;
1088 __u32 rq_rep_swab_mask;
1090 /** how many early replies (for stats) */
1092 /** Server-side, export on which request was received */
1093 struct obd_export *rq_export;
1094 /** import where request is being sent */
1095 struct obd_import *rq_import;
1098 /** Peer description (the other side) */
1099 lnet_process_id_t rq_peer;
1101 * service time estimate (secs)
1102 * If the request is not served by this time, it is marked as timed out.
1106 * when request/reply sent (secs), or time when request should be sent
1109 /** when request must finish. */
1111 /** request format description */
1112 struct req_capsule rq_pill;
1116 * Call completion handler for rpc if any, return it's status or original
1117 * rc if there was no handler defined for this request.
1119 static inline int ptlrpc_req_interpret(const struct lu_env *env,
1120 struct ptlrpc_request *req, int rc)
1122 if (req->rq_interpret_reply != NULL) {
1123 req->rq_status = req->rq_interpret_reply(env, req,
1124 &req->rq_async_args,
1126 return req->rq_status;
1134 int ptlrpc_nrs_policy_register(struct ptlrpc_nrs_pol_conf *conf);
1135 int ptlrpc_nrs_policy_unregister(struct ptlrpc_nrs_pol_conf *conf);
1136 void ptlrpc_nrs_req_hp_move(struct ptlrpc_request *req);
1137 void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
1138 struct ptlrpc_nrs_pol_info *info);
1141 * Can the request be moved from the regular NRS head to the high-priority NRS
1142 * head (of the same PTLRPC service partition), if any?
1144 * For a reliable result, this should be checked under svcpt->scp_req lock.
1146 static inline bool ptlrpc_nrs_req_can_move(struct ptlrpc_request *req)
1148 struct ptlrpc_nrs_request *nrq = &req->rq_nrq;
1151 * LU-898: Check ptlrpc_nrs_request::nr_enqueued to make sure the
1152 * request has been enqueued first, and ptlrpc_nrs_request::nr_started
1153 * to make sure it has not been scheduled yet (analogous to previous
1154 * (non-NRS) checking of !list_empty(&ptlrpc_request::rq_list).
1156 return nrq->nr_enqueued && !nrq->nr_started && !req->rq_hp;
1161 * Returns 1 if request buffer at offset \a index was already swabbed
1163 static inline int lustre_req_swabbed(struct ptlrpc_request *req, size_t index)
1165 LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
1166 return req->rq_req_swab_mask & (1 << index);
1170 * Returns 1 if request reply buffer at offset \a index was already swabbed
1172 static inline int lustre_rep_swabbed(struct ptlrpc_request *req, size_t index)
1174 LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
1175 return req->rq_rep_swab_mask & (1 << index);
1179 * Returns 1 if request needs to be swabbed into local cpu byteorder
1181 static inline int ptlrpc_req_need_swab(struct ptlrpc_request *req)
1183 return lustre_req_swabbed(req, MSG_PTLRPC_HEADER_OFF);
1187 * Returns 1 if request reply needs to be swabbed into local cpu byteorder
1189 static inline int ptlrpc_rep_need_swab(struct ptlrpc_request *req)
1191 return lustre_rep_swabbed(req, MSG_PTLRPC_HEADER_OFF);
1195 * Mark request buffer at offset \a index that it was already swabbed
1197 static inline void lustre_set_req_swabbed(struct ptlrpc_request *req,
1200 LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
1201 LASSERT((req->rq_req_swab_mask & (1 << index)) == 0);
1202 req->rq_req_swab_mask |= 1 << index;
1206 * Mark request reply buffer at offset \a index that it was already swabbed
1208 static inline void lustre_set_rep_swabbed(struct ptlrpc_request *req,
1211 LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
1212 LASSERT((req->rq_rep_swab_mask & (1 << index)) == 0);
1213 req->rq_rep_swab_mask |= 1 << index;
1217 * Convert numerical request phase value \a phase into text string description
1219 static inline const char *
1220 ptlrpc_phase2str(enum rq_phase phase)
1229 case RQ_PHASE_INTERPRET:
1231 case RQ_PHASE_COMPLETE:
1233 case RQ_PHASE_UNREGISTERING:
1234 return "Unregistering";
1241 * Convert numerical request phase of the request \a req into text stringi
1244 static inline const char *
1245 ptlrpc_rqphase2str(struct ptlrpc_request *req)
1247 return ptlrpc_phase2str(req->rq_phase);
1251 * Debugging functions and helpers to print request structure into debug log
1254 /* Spare the preprocessor, spoil the bugs. */
1255 #define FLAG(field, str) (field ? str : "")
1257 /** Convert bit flags into a string */
1258 #define DEBUG_REQ_FLAGS(req) \
1259 ptlrpc_rqphase2str(req), \
1260 FLAG(req->rq_intr, "I"), FLAG(req->rq_replied, "R"), \
1261 FLAG(req->rq_err, "E"), \
1262 FLAG(req->rq_timedout, "X") /* eXpired */, FLAG(req->rq_resend, "S"), \
1263 FLAG(req->rq_restart, "T"), FLAG(req->rq_replay, "P"), \
1264 FLAG(req->rq_no_resend, "N"), \
1265 FLAG(req->rq_waiting, "W"), \
1266 FLAG(req->rq_wait_ctx, "C"), FLAG(req->rq_hp, "H"), \
1267 FLAG(req->rq_committed, "M")
1269 #define REQ_FLAGS_FMT "%s:%s%s%s%s%s%s%s%s%s%s%s%s"
1271 void _debug_req(struct ptlrpc_request *req,
1272 struct libcfs_debug_msg_data *data, const char *fmt, ...)
1273 __attribute__ ((format (printf, 3, 4)));
1276 * Helper that decides if we need to print request accordig to current debug
1279 #define debug_req(msgdata, mask, cdls, req, fmt, a...) \
1281 CFS_CHECK_STACK(msgdata, mask, cdls); \
1283 if (((mask) & D_CANTMASK) != 0 || \
1284 ((libcfs_debug & (mask)) != 0 && \
1285 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) \
1286 _debug_req((req), msgdata, fmt, ##a); \
1290 * This is the debug print function you need to use to print request sturucture
1291 * content into lustre debug log.
1292 * for most callers (level is a constant) this is resolved at compile time */
1293 #define DEBUG_REQ(level, req, fmt, args...) \
1295 if ((level) & (D_ERROR | D_WARNING)) { \
1296 static cfs_debug_limit_state_t cdls; \
1297 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, level, &cdls); \
1298 debug_req(&msgdata, level, &cdls, req, "@@@ "fmt" ", ## args);\
1300 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, level, NULL); \
1301 debug_req(&msgdata, level, NULL, req, "@@@ "fmt" ", ## args); \
1307 * Structure that defines a single page of a bulk transfer
1309 struct ptlrpc_bulk_page {
1310 /** Linkage to list of pages in a bulk */
1311 struct list_head bp_link;
1313 * Number of bytes in a page to transfer starting from \a bp_pageoffset
1316 /** offset within a page */
1318 /** The page itself */
1319 struct page *bp_page;
1322 enum ptlrpc_bulk_op_type {
1323 PTLRPC_BULK_OP_ACTIVE = 0x00000001,
1324 PTLRPC_BULK_OP_PASSIVE = 0x00000002,
1325 PTLRPC_BULK_OP_PUT = 0x00000004,
1326 PTLRPC_BULK_OP_GET = 0x00000008,
1327 PTLRPC_BULK_BUF_KVEC = 0x00000010,
1328 PTLRPC_BULK_BUF_KIOV = 0x00000020,
1329 PTLRPC_BULK_GET_SOURCE = PTLRPC_BULK_OP_PASSIVE | PTLRPC_BULK_OP_GET,
1330 PTLRPC_BULK_PUT_SINK = PTLRPC_BULK_OP_PASSIVE | PTLRPC_BULK_OP_PUT,
1331 PTLRPC_BULK_GET_SINK = PTLRPC_BULK_OP_ACTIVE | PTLRPC_BULK_OP_GET,
1332 PTLRPC_BULK_PUT_SOURCE = PTLRPC_BULK_OP_ACTIVE | PTLRPC_BULK_OP_PUT,
1335 static inline bool ptlrpc_is_bulk_op_get(enum ptlrpc_bulk_op_type type)
1337 return (type & PTLRPC_BULK_OP_GET) == PTLRPC_BULK_OP_GET;
1340 static inline bool ptlrpc_is_bulk_get_source(enum ptlrpc_bulk_op_type type)
1342 return (type & PTLRPC_BULK_GET_SOURCE) == PTLRPC_BULK_GET_SOURCE;
1345 static inline bool ptlrpc_is_bulk_put_sink(enum ptlrpc_bulk_op_type type)
1347 return (type & PTLRPC_BULK_PUT_SINK) == PTLRPC_BULK_PUT_SINK;
1350 static inline bool ptlrpc_is_bulk_get_sink(enum ptlrpc_bulk_op_type type)
1352 return (type & PTLRPC_BULK_GET_SINK) == PTLRPC_BULK_GET_SINK;
1355 static inline bool ptlrpc_is_bulk_put_source(enum ptlrpc_bulk_op_type type)
1357 return (type & PTLRPC_BULK_PUT_SOURCE) == PTLRPC_BULK_PUT_SOURCE;
1360 static inline bool ptlrpc_is_bulk_desc_kvec(enum ptlrpc_bulk_op_type type)
1362 return ((type & PTLRPC_BULK_BUF_KVEC) | (type & PTLRPC_BULK_BUF_KIOV))
1363 == PTLRPC_BULK_BUF_KVEC;
1366 static inline bool ptlrpc_is_bulk_desc_kiov(enum ptlrpc_bulk_op_type type)
1368 return ((type & PTLRPC_BULK_BUF_KVEC) | (type & PTLRPC_BULK_BUF_KIOV))
1369 == PTLRPC_BULK_BUF_KIOV;
1372 static inline bool ptlrpc_is_bulk_op_active(enum ptlrpc_bulk_op_type type)
1374 return ((type & PTLRPC_BULK_OP_ACTIVE) |
1375 (type & PTLRPC_BULK_OP_PASSIVE))
1376 == PTLRPC_BULK_OP_ACTIVE;
1379 static inline bool ptlrpc_is_bulk_op_passive(enum ptlrpc_bulk_op_type type)
1381 return ((type & PTLRPC_BULK_OP_ACTIVE) |
1382 (type & PTLRPC_BULK_OP_PASSIVE))
1383 == PTLRPC_BULK_OP_PASSIVE;
1386 struct ptlrpc_bulk_frag_ops {
1388 * Add a page \a page to the bulk descriptor \a desc
1389 * Data to transfer in the page starts at offset \a pageoffset and
1390 * amount of data to transfer from the page is \a len
1392 void (*add_kiov_frag)(struct ptlrpc_bulk_desc *desc,
1393 struct page *page, int pageoffset, int len);
1396 * Add a \a fragment to the bulk descriptor \a desc.
1397 * Data to transfer in the fragment is pointed to by \a frag
1398 * The size of the fragment is \a len
1400 int (*add_iov_frag)(struct ptlrpc_bulk_desc *desc, void *frag, int len);
1403 * Uninitialize and free bulk descriptor \a desc.
1404 * Works on bulk descriptors both from server and client side.
1406 void (*release_frags)(struct ptlrpc_bulk_desc *desc);
1409 extern const struct ptlrpc_bulk_frag_ops ptlrpc_bulk_kiov_pin_ops;
1410 extern const struct ptlrpc_bulk_frag_ops ptlrpc_bulk_kiov_nopin_ops;
1411 extern const struct ptlrpc_bulk_frag_ops ptlrpc_bulk_kvec_ops;
1414 * Definition of bulk descriptor.
1415 * Bulks are special "Two phase" RPCs where initial request message
1416 * is sent first and it is followed bt a transfer (o receiving) of a large
1417 * amount of data to be settled into pages referenced from the bulk descriptors.
1418 * Bulks transfers (the actual data following the small requests) are done
1419 * on separate LNet portals.
1420 * In lustre we use bulk transfers for READ and WRITE transfers from/to OSTs.
1421 * Another user is readpage for MDT.
1423 struct ptlrpc_bulk_desc {
1424 /** completed with failure */
1425 unsigned long bd_failure:1;
1427 unsigned long bd_registered:1;
1428 /** For serialization with callback */
1430 /** Import generation when request for this bulk was sent */
1431 int bd_import_generation;
1432 /** {put,get}{source,sink}{kvec,kiov} */
1433 enum ptlrpc_bulk_op_type bd_type;
1434 /** LNet portal for this bulk */
1436 /** Server side - export this bulk created for */
1437 struct obd_export *bd_export;
1438 /** Client side - import this bulk was sent on */
1439 struct obd_import *bd_import;
1440 /** Back pointer to the request */
1441 struct ptlrpc_request *bd_req;
1442 struct ptlrpc_bulk_frag_ops *bd_frag_ops;
1443 wait_queue_head_t bd_waitq; /* server side only WQ */
1444 int bd_iov_count; /* # entries in bd_iov */
1445 int bd_max_iov; /* allocated size of bd_iov */
1446 int bd_nob; /* # bytes covered */
1447 int bd_nob_transferred; /* # bytes GOT/PUT */
1449 __u64 bd_last_mbits;
1451 struct ptlrpc_cb_id bd_cbid; /* network callback info */
1452 lnet_nid_t bd_sender; /* stash event::sender */
1453 int bd_md_count; /* # valid entries in bd_mds */
1454 int bd_md_max_brw; /* max entries in bd_mds */
1455 /** array of associated MDs */
1456 lnet_handle_md_t bd_mds[PTLRPC_BULK_OPS_COUNT];
1461 * encrypt iov, size is either 0 or bd_iov_count.
1463 lnet_kiov_t *bd_enc_vec;
1464 lnet_kiov_t bd_vec[0];
1468 struct kvec *bd_enc_kvec;
1469 struct kvec bd_kvec[0];
1475 #define GET_KIOV(desc) ((desc)->bd_u.bd_kiov.bd_vec)
1476 #define BD_GET_KIOV(desc, i) ((desc)->bd_u.bd_kiov.bd_vec[i])
1477 #define GET_ENC_KIOV(desc) ((desc)->bd_u.bd_kiov.bd_enc_vec)
1478 #define BD_GET_ENC_KIOV(desc, i) ((desc)->bd_u.bd_kiov.bd_enc_vec[i])
1479 #define GET_KVEC(desc) ((desc)->bd_u.bd_kvec.bd_kvec)
1480 #define BD_GET_KVEC(desc, i) ((desc)->bd_u.bd_kvec.bd_kvec[i])
1481 #define GET_ENC_KVEC(desc) ((desc)->bd_u.bd_kvec.bd_enc_kvec)
1482 #define BD_GET_ENC_KVEC(desc, i) ((desc)->bd_u.bd_kvec.bd_enc_kvec[i])
1485 SVC_STOPPED = 1 << 0,
1486 SVC_STOPPING = 1 << 1,
1487 SVC_STARTING = 1 << 2,
1488 SVC_RUNNING = 1 << 3,
1490 SVC_SIGNAL = 1 << 5,
1493 #define PTLRPC_THR_NAME_LEN 32
1495 * Definition of server service thread structure
1497 struct ptlrpc_thread {
1499 * List of active threads in svc->srv_threads
1501 struct list_head t_link;
1503 * thread-private data (preallocated memory)
1508 * service thread index, from ptlrpc_start_threads
1512 * service thread pid
1516 * put watchdog in the structure per thread b=14840
1518 struct lc_watchdog *t_watchdog;
1520 * the svc this thread belonged to b=18582
1522 struct ptlrpc_service_part *t_svcpt;
1523 wait_queue_head_t t_ctl_waitq;
1524 struct lu_env *t_env;
1525 char t_name[PTLRPC_THR_NAME_LEN];
1528 static inline int thread_is_init(struct ptlrpc_thread *thread)
1530 return thread->t_flags == 0;
1533 static inline int thread_is_stopped(struct ptlrpc_thread *thread)
1535 return !!(thread->t_flags & SVC_STOPPED);
1538 static inline int thread_is_stopping(struct ptlrpc_thread *thread)
1540 return !!(thread->t_flags & SVC_STOPPING);
1543 static inline int thread_is_starting(struct ptlrpc_thread *thread)
1545 return !!(thread->t_flags & SVC_STARTING);
1548 static inline int thread_is_running(struct ptlrpc_thread *thread)
1550 return !!(thread->t_flags & SVC_RUNNING);
1553 static inline int thread_is_event(struct ptlrpc_thread *thread)
1555 return !!(thread->t_flags & SVC_EVENT);
1558 static inline int thread_is_signal(struct ptlrpc_thread *thread)
1560 return !!(thread->t_flags & SVC_SIGNAL);
1563 static inline void thread_clear_flags(struct ptlrpc_thread *thread, __u32 flags)
1565 thread->t_flags &= ~flags;
1568 static inline void thread_set_flags(struct ptlrpc_thread *thread, __u32 flags)
1570 thread->t_flags = flags;
1573 static inline void thread_add_flags(struct ptlrpc_thread *thread, __u32 flags)
1575 thread->t_flags |= flags;
1578 static inline int thread_test_and_clear_flags(struct ptlrpc_thread *thread,
1581 if (thread->t_flags & flags) {
1582 thread->t_flags &= ~flags;
1589 * Request buffer descriptor structure.
1590 * This is a structure that contains one posted request buffer for service.
1591 * Once data land into a buffer, event callback creates actual request and
1592 * notifies wakes one of the service threads to process new incoming request.
1593 * More than one request can fit into the buffer.
1595 struct ptlrpc_request_buffer_desc {
1596 /** Link item for rqbds on a service */
1597 struct list_head rqbd_list;
1598 /** History of requests for this buffer */
1599 struct list_head rqbd_reqs;
1600 /** Back pointer to service for which this buffer is registered */
1601 struct ptlrpc_service_part *rqbd_svcpt;
1602 /** LNet descriptor */
1603 lnet_handle_md_t rqbd_md_h;
1605 /** The buffer itself */
1607 struct ptlrpc_cb_id rqbd_cbid;
1609 * This "embedded" request structure is only used for the
1610 * last request to fit into the buffer
1612 struct ptlrpc_request rqbd_req;
1615 typedef int (*svc_handler_t)(struct ptlrpc_request *req);
1617 struct ptlrpc_service_ops {
1619 * if non-NULL called during thread creation (ptlrpc_start_thread())
1620 * to initialize service specific per-thread state.
1622 int (*so_thr_init)(struct ptlrpc_thread *thr);
1624 * if non-NULL called during thread shutdown (ptlrpc_main()) to
1625 * destruct state created by ->srv_init().
1627 void (*so_thr_done)(struct ptlrpc_thread *thr);
1629 * Handler function for incoming requests for this service
1631 int (*so_req_handler)(struct ptlrpc_request *req);
1633 * function to determine priority of the request, it's called
1634 * on every new request
1636 int (*so_hpreq_handler)(struct ptlrpc_request *);
1638 * service-specific print fn
1640 void (*so_req_printer)(void *, struct ptlrpc_request *);
1643 #ifndef __cfs_cacheline_aligned
1644 /* NB: put it here for reducing patche dependence */
1645 # define __cfs_cacheline_aligned
1649 * How many high priority requests to serve before serving one normal
1652 #define PTLRPC_SVC_HP_RATIO 10
1655 * Definition of PortalRPC service.
1656 * The service is listening on a particular portal (like tcp port)
1657 * and perform actions for a specific server like IO service for OST
1658 * or general metadata service for MDS.
1660 struct ptlrpc_service {
1661 /** serialize /proc operations */
1662 spinlock_t srv_lock;
1663 /** most often accessed fields */
1664 /** chain thru all services */
1665 struct list_head srv_list;
1666 /** service operations table */
1667 struct ptlrpc_service_ops srv_ops;
1668 /** only statically allocated strings here; we don't clean them */
1670 /** only statically allocated strings here; we don't clean them */
1671 char *srv_thread_name;
1672 /** service thread list */
1673 struct list_head srv_threads;
1674 /** threads # should be created for each partition on initializing */
1675 int srv_nthrs_cpt_init;
1676 /** limit of threads number for each partition */
1677 int srv_nthrs_cpt_limit;
1678 /** Root of /proc dir tree for this service */
1679 struct proc_dir_entry *srv_procroot;
1680 /** Pointer to statistic data for this service */
1681 struct lprocfs_stats *srv_stats;
1682 /** # hp per lp reqs to handle */
1683 int srv_hpreq_ratio;
1684 /** biggest request to receive */
1685 int srv_max_req_size;
1686 /** biggest reply to send */
1687 int srv_max_reply_size;
1688 /** size of individual buffers */
1690 /** # buffers to allocate in 1 group */
1691 int srv_nbuf_per_group;
1692 /** Local portal on which to receive requests */
1693 __u32 srv_req_portal;
1694 /** Portal on the client to send replies to */
1695 __u32 srv_rep_portal;
1697 * Tags for lu_context associated with this thread, see struct
1701 /** soft watchdog timeout multiplier */
1702 int srv_watchdog_factor;
1703 /** under unregister_service */
1704 unsigned srv_is_stopping:1;
1706 /** max # request buffers in history per partition */
1707 int srv_hist_nrqbds_cpt_max;
1708 /** number of CPTs this service bound on */
1710 /** CPTs array this service bound on */
1712 /** 2^srv_cptab_bits >= cfs_cpt_numbert(srv_cptable) */
1714 /** CPT table this service is running over */
1715 struct cfs_cpt_table *srv_cptable;
1717 * partition data for ptlrpc service
1719 struct ptlrpc_service_part *srv_parts[0];
1723 * Definition of PortalRPC service partition data.
1724 * Although a service only has one instance of it right now, but we
1725 * will have multiple instances very soon (instance per CPT).
1727 * it has four locks:
1729 * serialize operations on rqbd and requests waiting for preprocess
1731 * serialize operations active requests sent to this portal
1733 * serialize adaptive timeout stuff
1735 * serialize operations on RS list (reply states)
1737 * We don't have any use-case to take two or more locks at the same time
1738 * for now, so there is no lock order issue.
1740 struct ptlrpc_service_part {
1741 /** back reference to owner */
1742 struct ptlrpc_service *scp_service __cfs_cacheline_aligned;
1743 /* CPT id, reserved */
1745 /** always increasing number */
1747 /** # of starting threads */
1748 int scp_nthrs_starting;
1749 /** # of stopping threads, reserved for shrinking threads */
1750 int scp_nthrs_stopping;
1751 /** # running threads */
1752 int scp_nthrs_running;
1753 /** service threads list */
1754 struct list_head scp_threads;
1757 * serialize the following fields, used for protecting
1758 * rqbd list and incoming requests waiting for preprocess,
1759 * threads starting & stopping are also protected by this lock.
1761 spinlock_t scp_lock __cfs_cacheline_aligned;
1762 /** total # req buffer descs allocated */
1763 int scp_nrqbds_total;
1764 /** # posted request buffers for receiving */
1765 int scp_nrqbds_posted;
1766 /** in progress of allocating rqbd */
1767 int scp_rqbd_allocating;
1768 /** # incoming reqs */
1769 int scp_nreqs_incoming;
1770 /** request buffers to be reposted */
1771 struct list_head scp_rqbd_idle;
1772 /** req buffers receiving */
1773 struct list_head scp_rqbd_posted;
1774 /** incoming reqs */
1775 struct list_head scp_req_incoming;
1776 /** timeout before re-posting reqs, in tick */
1777 cfs_duration_t scp_rqbd_timeout;
1779 * all threads sleep on this. This wait-queue is signalled when new
1780 * incoming request arrives and when difficult reply has to be handled.
1782 wait_queue_head_t scp_waitq;
1784 /** request history */
1785 struct list_head scp_hist_reqs;
1786 /** request buffer history */
1787 struct list_head scp_hist_rqbds;
1788 /** # request buffers in history */
1789 int scp_hist_nrqbds;
1790 /** sequence number for request */
1792 /** highest seq culled from history */
1793 __u64 scp_hist_seq_culled;
1796 * serialize the following fields, used for processing requests
1797 * sent to this portal
1799 spinlock_t scp_req_lock __cfs_cacheline_aligned;
1800 /** # reqs in either of the NRS heads below */
1801 /** # reqs being served */
1802 int scp_nreqs_active;
1803 /** # HPreqs being served */
1804 int scp_nhreqs_active;
1805 /** # hp requests handled */
1808 /** NRS head for regular requests */
1809 struct ptlrpc_nrs scp_nrs_reg;
1810 /** NRS head for HP requests; this is only valid for services that can
1811 * handle HP requests */
1812 struct ptlrpc_nrs *scp_nrs_hp;
1817 * serialize the following fields, used for changes on
1820 spinlock_t scp_at_lock __cfs_cacheline_aligned;
1821 /** estimated rpc service time */
1822 struct adaptive_timeout scp_at_estimate;
1823 /** reqs waiting for replies */
1824 struct ptlrpc_at_array scp_at_array;
1825 /** early reply timer */
1826 struct timer_list scp_at_timer;
1828 cfs_time_t scp_at_checktime;
1829 /** check early replies */
1830 unsigned scp_at_check;
1834 * serialize the following fields, used for processing
1835 * replies for this portal
1837 spinlock_t scp_rep_lock __cfs_cacheline_aligned;
1838 /** all the active replies */
1839 struct list_head scp_rep_active;
1840 /** List of free reply_states */
1841 struct list_head scp_rep_idle;
1842 /** waitq to run, when adding stuff to srv_free_rs_list */
1843 wait_queue_head_t scp_rep_waitq;
1844 /** # 'difficult' replies */
1845 atomic_t scp_nreps_difficult;
1848 #define ptlrpc_service_for_each_part(part, i, svc) \
1850 i < (svc)->srv_ncpts && \
1851 (svc)->srv_parts != NULL && \
1852 ((part) = (svc)->srv_parts[i]) != NULL; i++)
1855 * Declaration of ptlrpcd control structure
1857 struct ptlrpcd_ctl {
1859 * Ptlrpc thread control flags (LIOD_START, LIOD_STOP, LIOD_FORCE)
1861 unsigned long pc_flags;
1863 * Thread lock protecting structure fields.
1869 struct completion pc_starting;
1873 struct completion pc_finishing;
1875 * Thread requests set.
1877 struct ptlrpc_request_set *pc_set;
1879 * Thread name used in kthread_run()
1883 * Environment for request interpreters to run in.
1885 struct lu_env pc_env;
1887 * CPT the thread is bound on.
1891 * Index of ptlrpcd thread in the array.
1895 * Pointer to the array of partners' ptlrpcd_ctl structure.
1897 struct ptlrpcd_ctl **pc_partners;
1899 * Number of the ptlrpcd's partners.
1903 * Record the partner index to be processed next.
1907 * Error code if the thread failed to fully start.
1912 /* Bits for pc_flags */
1913 enum ptlrpcd_ctl_flags {
1915 * Ptlrpc thread start flag.
1917 LIOD_START = 1 << 0,
1919 * Ptlrpc thread stop flag.
1923 * Ptlrpc thread force flag (only stop force so far).
1924 * This will cause aborting any inflight rpcs handled
1925 * by thread if LIOD_STOP is specified.
1927 LIOD_FORCE = 1 << 2,
1929 * This is a recovery ptlrpc thread.
1931 LIOD_RECOVERY = 1 << 3,
1938 * Service compatibility function; the policy is compatible with all services.
1940 * \param[in] svc The service the policy is attempting to register with.
1941 * \param[in] desc The policy descriptor
1943 * \retval true The policy is compatible with the service
1945 * \see ptlrpc_nrs_pol_desc::pd_compat()
1947 static inline bool nrs_policy_compat_all(const struct ptlrpc_service *svc,
1948 const struct ptlrpc_nrs_pol_desc *desc)
1954 * Service compatibility function; the policy is compatible with only a specific
1955 * service which is identified by its human-readable name at
1956 * ptlrpc_service::srv_name.
1958 * \param[in] svc The service the policy is attempting to register with.
1959 * \param[in] desc The policy descriptor
1961 * \retval false The policy is not compatible with the service
1962 * \retval true The policy is compatible with the service
1964 * \see ptlrpc_nrs_pol_desc::pd_compat()
1966 static inline bool nrs_policy_compat_one(const struct ptlrpc_service *svc,
1967 const struct ptlrpc_nrs_pol_desc *desc)
1969 LASSERT(desc->pd_compat_svc_name != NULL);
1970 return strcmp(svc->srv_name, desc->pd_compat_svc_name) == 0;
1975 /* ptlrpc/events.c */
1976 extern lnet_handle_eq_t ptlrpc_eq_h;
1977 extern int ptlrpc_uuid_to_peer(struct obd_uuid *uuid,
1978 lnet_process_id_t *peer, lnet_nid_t *self);
1980 * These callbacks are invoked by LNet when something happened to
1984 extern void request_out_callback(lnet_event_t *ev);
1985 extern void reply_in_callback(lnet_event_t *ev);
1986 extern void client_bulk_callback(lnet_event_t *ev);
1987 extern void request_in_callback(lnet_event_t *ev);
1988 extern void reply_out_callback(lnet_event_t *ev);
1989 #ifdef HAVE_SERVER_SUPPORT
1990 extern void server_bulk_callback(lnet_event_t *ev);
1994 /* ptlrpc/connection.c */
1995 struct ptlrpc_connection *ptlrpc_connection_get(lnet_process_id_t peer,
1997 struct obd_uuid *uuid);
1998 int ptlrpc_connection_put(struct ptlrpc_connection *c);
1999 struct ptlrpc_connection *ptlrpc_connection_addref(struct ptlrpc_connection *);
2000 int ptlrpc_connection_init(void);
2001 void ptlrpc_connection_fini(void);
2002 extern lnet_pid_t ptl_get_pid(void);
2004 /* ptlrpc/niobuf.c */
2006 * Actual interfacing with LNet to put/get/register/unregister stuff
2009 #ifdef HAVE_SERVER_SUPPORT
2010 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_exp(struct ptlrpc_request *req,
2011 unsigned nfrags, unsigned max_brw,
2014 const struct ptlrpc_bulk_frag_ops
2016 int ptlrpc_start_bulk_transfer(struct ptlrpc_bulk_desc *desc);
2017 void ptlrpc_abort_bulk(struct ptlrpc_bulk_desc *desc);
2019 static inline int ptlrpc_server_bulk_active(struct ptlrpc_bulk_desc *desc)
2023 LASSERT(desc != NULL);
2025 spin_lock(&desc->bd_lock);
2026 rc = desc->bd_md_count;
2027 spin_unlock(&desc->bd_lock);
2032 int ptlrpc_register_bulk(struct ptlrpc_request *req);
2033 int ptlrpc_unregister_bulk(struct ptlrpc_request *req, int async);
2035 static inline int ptlrpc_client_bulk_active(struct ptlrpc_request *req)
2037 struct ptlrpc_bulk_desc *desc;
2040 LASSERT(req != NULL);
2041 desc = req->rq_bulk;
2043 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_BULK_UNLINK) &&
2044 req->rq_bulk_deadline > cfs_time_current_sec())
2050 spin_lock(&desc->bd_lock);
2051 rc = desc->bd_md_count;
2052 spin_unlock(&desc->bd_lock);
2056 #define PTLRPC_REPLY_MAYBE_DIFFICULT 0x01
2057 #define PTLRPC_REPLY_EARLY 0x02
2058 int ptlrpc_send_reply(struct ptlrpc_request *req, int flags);
2059 int ptlrpc_reply(struct ptlrpc_request *req);
2060 int ptlrpc_send_error(struct ptlrpc_request *req, int difficult);
2061 int ptlrpc_error(struct ptlrpc_request *req);
2062 int ptlrpc_at_get_net_latency(struct ptlrpc_request *req);
2063 int ptl_send_rpc(struct ptlrpc_request *request, int noreply);
2064 int ptlrpc_register_rqbd(struct ptlrpc_request_buffer_desc *rqbd);
2067 /* ptlrpc/client.c */
2069 * Client-side portals API. Everything to send requests, receive replies,
2070 * request queues, request management, etc.
2073 void ptlrpc_request_committed(struct ptlrpc_request *req, int force);
2075 void ptlrpc_init_client(int req_portal, int rep_portal, char *name,
2076 struct ptlrpc_client *);
2077 void ptlrpc_cleanup_client(struct obd_import *imp);
2078 struct ptlrpc_connection *ptlrpc_uuid_to_connection(struct obd_uuid *uuid);
2080 int ptlrpc_queue_wait(struct ptlrpc_request *req);
2081 int ptlrpc_replay_req(struct ptlrpc_request *req);
2082 void ptlrpc_restart_req(struct ptlrpc_request *req);
2083 void ptlrpc_abort_inflight(struct obd_import *imp);
2084 void ptlrpc_cleanup_imp(struct obd_import *imp);
2085 void ptlrpc_abort_set(struct ptlrpc_request_set *set);
2087 struct ptlrpc_request_set *ptlrpc_prep_set(void);
2088 struct ptlrpc_request_set *ptlrpc_prep_fcset(int max, set_producer_func func,
2090 int ptlrpc_set_add_cb(struct ptlrpc_request_set *set,
2091 set_interpreter_func fn, void *data);
2092 int ptlrpc_check_set(const struct lu_env *env, struct ptlrpc_request_set *set);
2093 int ptlrpc_set_wait(struct ptlrpc_request_set *);
2094 void ptlrpc_mark_interrupted(struct ptlrpc_request *req);
2095 void ptlrpc_set_destroy(struct ptlrpc_request_set *);
2096 void ptlrpc_set_add_req(struct ptlrpc_request_set *, struct ptlrpc_request *);
2098 void ptlrpc_free_rq_pool(struct ptlrpc_request_pool *pool);
2099 int ptlrpc_add_rqs_to_pool(struct ptlrpc_request_pool *pool, int num_rq);
2101 struct ptlrpc_request_pool *
2102 ptlrpc_init_rq_pool(int, int,
2103 int (*populate_pool)(struct ptlrpc_request_pool *, int));
2105 void ptlrpc_at_set_req_timeout(struct ptlrpc_request *req);
2106 struct ptlrpc_request *ptlrpc_request_alloc(struct obd_import *imp,
2107 const struct req_format *format);
2108 struct ptlrpc_request *ptlrpc_request_alloc_pool(struct obd_import *imp,
2109 struct ptlrpc_request_pool *,
2110 const struct req_format *format);
2111 void ptlrpc_request_free(struct ptlrpc_request *request);
2112 int ptlrpc_request_pack(struct ptlrpc_request *request,
2113 __u32 version, int opcode);
2114 struct ptlrpc_request *ptlrpc_request_alloc_pack(struct obd_import *imp,
2115 const struct req_format *format,
2116 __u32 version, int opcode);
2117 int ptlrpc_request_bufs_pack(struct ptlrpc_request *request,
2118 __u32 version, int opcode, char **bufs,
2119 struct ptlrpc_cli_ctx *ctx);
2120 struct ptlrpc_request *ptlrpc_prep_req(struct obd_import *imp, __u32 version,
2121 int opcode, int count, __u32 *lengths,
2123 struct ptlrpc_request *ptlrpc_prep_req_pool(struct obd_import *imp,
2124 __u32 version, int opcode,
2125 int count, __u32 *lengths, char **bufs,
2126 struct ptlrpc_request_pool *pool);
2127 void ptlrpc_req_finished(struct ptlrpc_request *request);
2128 void ptlrpc_req_finished_with_imp_lock(struct ptlrpc_request *request);
2129 struct ptlrpc_request *ptlrpc_request_addref(struct ptlrpc_request *req);
2130 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_imp(struct ptlrpc_request *req,
2131 unsigned nfrags, unsigned max_brw,
2134 const struct ptlrpc_bulk_frag_ops
2137 int ptlrpc_prep_bulk_frag(struct ptlrpc_bulk_desc *desc,
2138 void *frag, int len);
2139 void __ptlrpc_prep_bulk_page(struct ptlrpc_bulk_desc *desc,
2140 struct page *page, int pageoffset, int len,
2142 static inline void ptlrpc_prep_bulk_page_pin(struct ptlrpc_bulk_desc *desc,
2143 struct page *page, int pageoffset,
2146 __ptlrpc_prep_bulk_page(desc, page, pageoffset, len, 1);
2149 static inline void ptlrpc_prep_bulk_page_nopin(struct ptlrpc_bulk_desc *desc,
2150 struct page *page, int pageoffset,
2153 __ptlrpc_prep_bulk_page(desc, page, pageoffset, len, 0);
2156 void ptlrpc_free_bulk(struct ptlrpc_bulk_desc *bulk);
2158 static inline void ptlrpc_release_bulk_page_pin(struct ptlrpc_bulk_desc *desc)
2162 for (i = 0; i < desc->bd_iov_count ; i++)
2163 page_cache_release(BD_GET_KIOV(desc, i).kiov_page);
2166 static inline void ptlrpc_release_bulk_noop(struct ptlrpc_bulk_desc *desc)
2170 void ptlrpc_retain_replayable_request(struct ptlrpc_request *req,
2171 struct obd_import *imp);
2172 __u64 ptlrpc_next_xid(void);
2173 __u64 ptlrpc_sample_next_xid(void);
2174 __u64 ptlrpc_req_xid(struct ptlrpc_request *request);
2176 /* Set of routines to run a function in ptlrpcd context */
2177 void *ptlrpcd_alloc_work(struct obd_import *imp,
2178 int (*cb)(const struct lu_env *, void *), void *data);
2179 void ptlrpcd_destroy_work(void *handler);
2180 int ptlrpcd_queue_work(void *handler);
2183 struct ptlrpc_service_buf_conf {
2184 /* nbufs is buffers # to allocate when growing the pool */
2185 unsigned int bc_nbufs;
2186 /* buffer size to post */
2187 unsigned int bc_buf_size;
2188 /* portal to listed for requests on */
2189 unsigned int bc_req_portal;
2190 /* portal of where to send replies to */
2191 unsigned int bc_rep_portal;
2192 /* maximum request size to be accepted for this service */
2193 unsigned int bc_req_max_size;
2194 /* maximum reply size this service can ever send */
2195 unsigned int bc_rep_max_size;
2198 struct ptlrpc_service_thr_conf {
2199 /* threadname should be 8 characters or less - 6 will be added on */
2201 /* threads increasing factor for each CPU */
2202 unsigned int tc_thr_factor;
2203 /* service threads # to start on each partition while initializing */
2204 unsigned int tc_nthrs_init;
2206 * low water of threads # upper-limit on each partition while running,
2207 * service availability may be impacted if threads number is lower
2208 * than this value. It can be ZERO if the service doesn't require
2209 * CPU affinity or there is only one partition.
2211 unsigned int tc_nthrs_base;
2212 /* "soft" limit for total threads number */
2213 unsigned int tc_nthrs_max;
2214 /* user specified threads number, it will be validated due to
2215 * other members of this structure. */
2216 unsigned int tc_nthrs_user;
2217 /* set NUMA node affinity for service threads */
2218 unsigned int tc_cpu_affinity;
2219 /* Tags for lu_context associated with service thread */
2223 struct ptlrpc_service_cpt_conf {
2224 struct cfs_cpt_table *cc_cptable;
2225 /* string pattern to describe CPTs for a service */
2229 struct ptlrpc_service_conf {
2232 /* soft watchdog timeout multiplifier to print stuck service traces */
2233 unsigned int psc_watchdog_factor;
2234 /* buffer information */
2235 struct ptlrpc_service_buf_conf psc_buf;
2236 /* thread information */
2237 struct ptlrpc_service_thr_conf psc_thr;
2238 /* CPU partition information */
2239 struct ptlrpc_service_cpt_conf psc_cpt;
2240 /* function table */
2241 struct ptlrpc_service_ops psc_ops;
2244 /* ptlrpc/service.c */
2246 * Server-side services API. Register/unregister service, request state
2247 * management, service thread management
2251 void ptlrpc_save_lock(struct ptlrpc_request *req,
2252 struct lustre_handle *lock, int mode, int no_ack);
2253 void ptlrpc_commit_replies(struct obd_export *exp);
2254 void ptlrpc_dispatch_difficult_reply(struct ptlrpc_reply_state *rs);
2255 void ptlrpc_schedule_difficult_reply(struct ptlrpc_reply_state *rs);
2256 int ptlrpc_hpreq_handler(struct ptlrpc_request *req);
2257 struct ptlrpc_service *ptlrpc_register_service(
2258 struct ptlrpc_service_conf *conf,
2259 struct proc_dir_entry *proc_entry);
2260 void ptlrpc_stop_all_threads(struct ptlrpc_service *svc);
2262 int ptlrpc_start_threads(struct ptlrpc_service *svc);
2263 int ptlrpc_unregister_service(struct ptlrpc_service *service);
2264 int liblustre_check_services(void *arg);
2265 void ptlrpc_daemonize(char *name);
2266 int ptlrpc_service_health_check(struct ptlrpc_service *);
2267 void ptlrpc_server_drop_request(struct ptlrpc_request *req);
2268 void ptlrpc_request_change_export(struct ptlrpc_request *req,
2269 struct obd_export *export);
2270 void ptlrpc_update_export_timer(struct obd_export *exp, long extra_delay);
2272 int ptlrpc_hr_init(void);
2273 void ptlrpc_hr_fini(void);
2277 /* ptlrpc/import.c */
2282 int ptlrpc_connect_import(struct obd_import *imp);
2283 int ptlrpc_init_import(struct obd_import *imp);
2284 int ptlrpc_disconnect_import(struct obd_import *imp, int noclose);
2285 int ptlrpc_import_recovery_state_machine(struct obd_import *imp);
2286 void deuuidify(char *uuid, const char *prefix, char **uuid_start,
2288 void ptlrpc_import_enter_resend(struct obd_import *imp);
2289 /* ptlrpc/pack_generic.c */
2290 int ptlrpc_reconnect_import(struct obd_import *imp);
2294 * ptlrpc msg buffer and swab interface
2298 int ptlrpc_buf_need_swab(struct ptlrpc_request *req, const int inout,
2300 void ptlrpc_buf_set_swabbed(struct ptlrpc_request *req, const int inout,
2302 int ptlrpc_unpack_rep_msg(struct ptlrpc_request *req, int len);
2303 int ptlrpc_unpack_req_msg(struct ptlrpc_request *req, int len);
2305 int lustre_msg_check_version(struct lustre_msg *msg, __u32 version);
2306 void lustre_init_msg_v2(struct lustre_msg_v2 *msg, int count, __u32 *lens,
2308 int lustre_pack_request(struct ptlrpc_request *, __u32 magic, int count,
2309 __u32 *lens, char **bufs);
2310 int lustre_pack_reply(struct ptlrpc_request *, int count, __u32 *lens,
2312 int lustre_pack_reply_v2(struct ptlrpc_request *req, int count,
2313 __u32 *lens, char **bufs, int flags);
2314 #define LPRFL_EARLY_REPLY 1
2315 int lustre_pack_reply_flags(struct ptlrpc_request *, int count, __u32 *lens,
2316 char **bufs, int flags);
2317 int lustre_shrink_msg(struct lustre_msg *msg, int segment,
2318 unsigned int newlen, int move_data);
2319 void lustre_free_reply_state(struct ptlrpc_reply_state *rs);
2320 int __lustre_unpack_msg(struct lustre_msg *m, int len);
2321 __u32 lustre_msg_hdr_size(__u32 magic, __u32 count);
2322 __u32 lustre_msg_size(__u32 magic, int count, __u32 *lengths);
2323 __u32 lustre_msg_size_v2(int count, __u32 *lengths);
2324 __u32 lustre_packed_msg_size(struct lustre_msg *msg);
2325 __u32 lustre_msg_early_size(void);
2326 void *lustre_msg_buf_v2(struct lustre_msg_v2 *m, __u32 n, __u32 min_size);
2327 void *lustre_msg_buf(struct lustre_msg *m, __u32 n, __u32 minlen);
2328 __u32 lustre_msg_buflen(struct lustre_msg *m, __u32 n);
2329 void lustre_msg_set_buflen(struct lustre_msg *m, __u32 n, __u32 len);
2330 __u32 lustre_msg_bufcount(struct lustre_msg *m);
2331 char *lustre_msg_string(struct lustre_msg *m, __u32 n, __u32 max_len);
2332 __u32 lustre_msghdr_get_flags(struct lustre_msg *msg);
2333 void lustre_msghdr_set_flags(struct lustre_msg *msg, __u32 flags);
2334 __u32 lustre_msg_get_flags(struct lustre_msg *msg);
2335 void lustre_msg_add_flags(struct lustre_msg *msg, __u32 flags);
2336 void lustre_msg_set_flags(struct lustre_msg *msg, __u32 flags);
2337 void lustre_msg_clear_flags(struct lustre_msg *msg, __u32 flags);
2338 __u32 lustre_msg_get_op_flags(struct lustre_msg *msg);
2339 void lustre_msg_add_op_flags(struct lustre_msg *msg, __u32 flags);
2340 struct lustre_handle *lustre_msg_get_handle(struct lustre_msg *msg);
2341 __u32 lustre_msg_get_type(struct lustre_msg *msg);
2342 __u32 lustre_msg_get_version(struct lustre_msg *msg);
2343 void lustre_msg_add_version(struct lustre_msg *msg, __u32 version);
2344 __u32 lustre_msg_get_opc(struct lustre_msg *msg);
2345 __u64 lustre_msg_get_last_xid(struct lustre_msg *msg);
2346 __u16 lustre_msg_get_tag(struct lustre_msg *msg);
2347 __u64 lustre_msg_get_last_committed(struct lustre_msg *msg);
2348 __u64 *lustre_msg_get_versions(struct lustre_msg *msg);
2349 __u64 lustre_msg_get_transno(struct lustre_msg *msg);
2350 __u64 lustre_msg_get_slv(struct lustre_msg *msg);
2351 __u32 lustre_msg_get_limit(struct lustre_msg *msg);
2352 void lustre_msg_set_slv(struct lustre_msg *msg, __u64 slv);
2353 void lustre_msg_set_limit(struct lustre_msg *msg, __u64 limit);
2354 int lustre_msg_get_status(struct lustre_msg *msg);
2355 __u32 lustre_msg_get_conn_cnt(struct lustre_msg *msg);
2356 __u32 lustre_msg_get_magic(struct lustre_msg *msg);
2357 __u32 lustre_msg_get_timeout(struct lustre_msg *msg);
2358 __u32 lustre_msg_get_service_time(struct lustre_msg *msg);
2359 char *lustre_msg_get_jobid(struct lustre_msg *msg);
2360 __u32 lustre_msg_get_cksum(struct lustre_msg *msg);
2361 __u64 lustre_msg_get_mbits(struct lustre_msg *msg);
2362 __u32 lustre_msg_calc_cksum(struct lustre_msg *msg);
2363 void lustre_msg_set_handle(struct lustre_msg *msg,struct lustre_handle *handle);
2364 void lustre_msg_set_type(struct lustre_msg *msg, __u32 type);
2365 void lustre_msg_set_opc(struct lustre_msg *msg, __u32 opc);
2366 void lustre_msg_set_last_xid(struct lustre_msg *msg, __u64 last_xid);
2367 void lustre_msg_set_tag(struct lustre_msg *msg, __u16 tag);
2368 void lustre_msg_set_last_committed(struct lustre_msg *msg,__u64 last_committed);
2369 void lustre_msg_set_versions(struct lustre_msg *msg, __u64 *versions);
2370 void lustre_msg_set_transno(struct lustre_msg *msg, __u64 transno);
2371 void lustre_msg_set_status(struct lustre_msg *msg, __u32 status);
2372 void lustre_msg_set_conn_cnt(struct lustre_msg *msg, __u32 conn_cnt);
2373 void ptlrpc_req_set_repsize(struct ptlrpc_request *req, int count, __u32 *sizes);
2374 void ptlrpc_request_set_replen(struct ptlrpc_request *req);
2375 void lustre_msg_set_timeout(struct lustre_msg *msg, __u32 timeout);
2376 void lustre_msg_set_service_time(struct lustre_msg *msg, __u32 service_time);
2377 void lustre_msg_set_jobid(struct lustre_msg *msg, char *jobid);
2378 void lustre_msg_set_cksum(struct lustre_msg *msg, __u32 cksum);
2379 void lustre_msg_set_mbits(struct lustre_msg *msg, __u64 mbits);
2382 lustre_shrink_reply(struct ptlrpc_request *req, int segment,
2383 unsigned int newlen, int move_data)
2385 LASSERT(req->rq_reply_state);
2386 LASSERT(req->rq_repmsg);
2387 req->rq_replen = lustre_shrink_msg(req->rq_repmsg, segment,
2391 #ifdef LUSTRE_TRANSLATE_ERRNOS
2393 static inline int ptlrpc_status_hton(int h)
2396 * Positive errnos must be network errnos, such as LUSTRE_EDEADLK,
2397 * ELDLM_LOCK_ABORTED, etc.
2400 return -lustre_errno_hton(-h);
2405 static inline int ptlrpc_status_ntoh(int n)
2408 * See the comment in ptlrpc_status_hton().
2411 return -lustre_errno_ntoh(-n);
2418 #define ptlrpc_status_hton(h) (h)
2419 #define ptlrpc_status_ntoh(n) (n)
2424 /** Change request phase of \a req to \a new_phase */
2426 ptlrpc_rqphase_move(struct ptlrpc_request *req, enum rq_phase new_phase)
2428 if (req->rq_phase == new_phase)
2431 if (new_phase == RQ_PHASE_UNREGISTERING) {
2432 req->rq_next_phase = req->rq_phase;
2434 atomic_inc(&req->rq_import->imp_unregistering);
2437 if (req->rq_phase == RQ_PHASE_UNREGISTERING) {
2439 atomic_dec(&req->rq_import->imp_unregistering);
2442 DEBUG_REQ(D_INFO, req, "move req \"%s\" -> \"%s\"",
2443 ptlrpc_rqphase2str(req), ptlrpc_phase2str(new_phase));
2445 req->rq_phase = new_phase;
2449 * Returns true if request \a req got early reply and hard deadline is not met
2452 ptlrpc_client_early(struct ptlrpc_request *req)
2454 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
2455 req->rq_reply_deadline > cfs_time_current_sec())
2457 return req->rq_early;
2461 * Returns true if we got real reply from server for this request
2464 ptlrpc_client_replied(struct ptlrpc_request *req)
2466 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
2467 req->rq_reply_deadline > cfs_time_current_sec())
2469 return req->rq_replied;
2472 /** Returns true if request \a req is in process of receiving server reply */
2474 ptlrpc_client_recv(struct ptlrpc_request *req)
2476 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
2477 req->rq_reply_deadline > cfs_time_current_sec())
2479 return req->rq_receiving_reply;
2483 ptlrpc_client_recv_or_unlink(struct ptlrpc_request *req)
2487 spin_lock(&req->rq_lock);
2488 if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
2489 req->rq_reply_deadline > cfs_time_current_sec()) {
2490 spin_unlock(&req->rq_lock);
2493 rc = !req->rq_req_unlinked || !req->rq_reply_unlinked ||
2494 req->rq_receiving_reply;
2495 spin_unlock(&req->rq_lock);
2500 ptlrpc_client_wake_req(struct ptlrpc_request *req)
2502 if (req->rq_set == NULL)
2503 wake_up(&req->rq_reply_waitq);
2505 wake_up(&req->rq_set->set_waitq);
2509 ptlrpc_rs_addref(struct ptlrpc_reply_state *rs)
2511 LASSERT(atomic_read(&rs->rs_refcount) > 0);
2512 atomic_inc(&rs->rs_refcount);
2516 ptlrpc_rs_decref(struct ptlrpc_reply_state *rs)
2518 LASSERT(atomic_read(&rs->rs_refcount) > 0);
2519 if (atomic_dec_and_test(&rs->rs_refcount))
2520 lustre_free_reply_state(rs);
2523 /* Should only be called once per req */
2524 static inline void ptlrpc_req_drop_rs(struct ptlrpc_request *req)
2526 if (req->rq_reply_state == NULL)
2527 return; /* shouldn't occur */
2528 ptlrpc_rs_decref(req->rq_reply_state);
2529 req->rq_reply_state = NULL;
2530 req->rq_repmsg = NULL;
2533 static inline __u32 lustre_request_magic(struct ptlrpc_request *req)
2535 return lustre_msg_get_magic(req->rq_reqmsg);
2538 static inline int ptlrpc_req_get_repsize(struct ptlrpc_request *req)
2540 switch (req->rq_reqmsg->lm_magic) {
2541 case LUSTRE_MSG_MAGIC_V2:
2542 return req->rq_reqmsg->lm_repsize;
2544 LASSERTF(0, "incorrect message magic: %08x\n",
2545 req->rq_reqmsg->lm_magic);
2550 static inline int ptlrpc_send_limit_expired(struct ptlrpc_request *req)
2552 if (req->rq_delay_limit != 0 &&
2553 cfs_time_before(cfs_time_add(req->rq_queued_time,
2554 cfs_time_seconds(req->rq_delay_limit)),
2555 cfs_time_current())) {
2561 static inline int ptlrpc_no_resend(struct ptlrpc_request *req)
2563 if (!req->rq_no_resend && ptlrpc_send_limit_expired(req)) {
2564 spin_lock(&req->rq_lock);
2565 req->rq_no_resend = 1;
2566 spin_unlock(&req->rq_lock);
2568 return req->rq_no_resend;
2572 ptlrpc_server_get_timeout(struct ptlrpc_service_part *svcpt)
2574 int at = AT_OFF ? 0 : at_get(&svcpt->scp_at_estimate);
2576 return svcpt->scp_service->srv_watchdog_factor *
2577 max_t(int, at, obd_timeout);
2580 static inline struct ptlrpc_service *
2581 ptlrpc_req2svc(struct ptlrpc_request *req)
2583 LASSERT(req->rq_rqbd != NULL);
2584 return req->rq_rqbd->rqbd_svcpt->scp_service;
2587 /* ldlm/ldlm_lib.c */
2589 * Target client logic
2592 int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg);
2593 int client_obd_cleanup(struct obd_device *obddev);
2594 int client_connect_import(const struct lu_env *env,
2595 struct obd_export **exp, struct obd_device *obd,
2596 struct obd_uuid *cluuid, struct obd_connect_data *,
2598 int client_disconnect_export(struct obd_export *exp);
2599 int client_import_add_conn(struct obd_import *imp, struct obd_uuid *uuid,
2601 int client_import_del_conn(struct obd_import *imp, struct obd_uuid *uuid);
2602 int client_import_find_conn(struct obd_import *imp, lnet_nid_t peer,
2603 struct obd_uuid *uuid);
2604 int import_set_conn_priority(struct obd_import *imp, struct obd_uuid *uuid);
2605 void client_destroy_import(struct obd_import *imp);
2608 #ifdef HAVE_SERVER_SUPPORT
2609 int server_disconnect_export(struct obd_export *exp);
2612 /* ptlrpc/pinger.c */
2614 * Pinger API (client side only)
2617 enum timeout_event {
2620 struct timeout_item;
2621 typedef int (*timeout_cb_t)(struct timeout_item *, void *);
2622 int ptlrpc_pinger_add_import(struct obd_import *imp);
2623 int ptlrpc_pinger_del_import(struct obd_import *imp);
2624 int ptlrpc_add_timeout_client(int time, enum timeout_event event,
2625 timeout_cb_t cb, void *data,
2626 struct list_head *obd_list);
2627 int ptlrpc_del_timeout_client(struct list_head *obd_list,
2628 enum timeout_event event);
2629 struct ptlrpc_request * ptlrpc_prep_ping(struct obd_import *imp);
2630 int ptlrpc_obd_ping(struct obd_device *obd);
2631 void ping_evictor_start(void);
2632 void ping_evictor_stop(void);
2633 void ptlrpc_pinger_ir_up(void);
2634 void ptlrpc_pinger_ir_down(void);
2636 int ptlrpc_pinger_suppress_pings(void);
2638 /* ptlrpc/ptlrpcd.c */
2639 void ptlrpcd_stop(struct ptlrpcd_ctl *pc, int force);
2640 void ptlrpcd_free(struct ptlrpcd_ctl *pc);
2641 void ptlrpcd_wake(struct ptlrpc_request *req);
2642 void ptlrpcd_add_req(struct ptlrpc_request *req);
2643 void ptlrpcd_add_rqset(struct ptlrpc_request_set *set);
2644 int ptlrpcd_addref(void);
2645 void ptlrpcd_decref(void);
2647 /* ptlrpc/lproc_ptlrpc.c */
2649 * procfs output related functions
2652 const char* ll_opcode2str(__u32 opcode);
2653 #ifdef CONFIG_PROC_FS
2654 void ptlrpc_lprocfs_register_obd(struct obd_device *obd);
2655 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd);
2656 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes);
2658 static inline void ptlrpc_lprocfs_register_obd(struct obd_device *obd) {}
2659 static inline void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd) {}
2660 static inline void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes) {}
2664 /* ptlrpc/llog_server.c */
2665 int llog_origin_handle_open(struct ptlrpc_request *req);
2666 int llog_origin_handle_destroy(struct ptlrpc_request *req);
2667 int llog_origin_handle_prev_block(struct ptlrpc_request *req);
2668 int llog_origin_handle_next_block(struct ptlrpc_request *req);
2669 int llog_origin_handle_read_header(struct ptlrpc_request *req);
2670 int llog_origin_handle_close(struct ptlrpc_request *req);
2672 /* ptlrpc/llog_client.c */
2673 extern struct llog_operations llog_client_ops;