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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2010, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 /** \defgroup PtlRPC Portal RPC and networking module.
34 * PortalRPC is the layer used by rest of lustre code to achieve network
35 * communications: establish connections with corresponding export and import
36 * states, listen for a service, send and receive RPCs.
37 * PortalRPC also includes base recovery framework: packet resending and
38 * replaying, reconnections, pinger.
40 * PortalRPC utilizes LNet as its transport layer.
54 #include <linux/uio.h>
55 #include <libcfs/libcfs.h>
56 #include <lnet/nidstr.h>
58 #include <lustre/lustre_idl.h>
59 #include <lustre_ha.h>
60 #include <lustre_sec.h>
61 #include <lustre_import.h>
62 #include <lprocfs_status.h>
63 #include <lu_object.h>
64 #include <lustre_req_layout.h>
65 #include <obd_support.h>
66 #include <lustre_ver.h>
68 /* MD flags we _always_ use */
69 #define PTLRPC_MD_OPTIONS 0
72 * log2 max # of bulk operations in one request: 2=4MB/RPC, 5=32MB/RPC, ...
73 * In order for the client and server to properly negotiate the maximum
74 * possible transfer size, PTLRPC_BULK_OPS_COUNT must be a power-of-two
75 * value. The client is free to limit the actual RPC size for any bulk
76 * transfer via cl_max_pages_per_rpc to some non-power-of-two value.
77 * NOTE: This is limited to 16 (=64GB RPCs) by IOOBJ_MAX_BRW_BITS. */
78 #define PTLRPC_BULK_OPS_BITS 4
79 #if PTLRPC_BULK_OPS_BITS > 16
80 #error "More than 65536 BRW RPCs not allowed by IOOBJ_MAX_BRW_BITS."
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_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_SHIFT)
106 #define DT_MAX_BRW_SIZE PTLRPC_MAX_BRW_SIZE
107 #define DT_MAX_BRW_PAGES (DT_MAX_BRW_SIZE >> PAGE_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_SIZE))
115 # error "PTLRPC_MAX_BRW_SIZE isn't PTLRPC_MAX_BRW_PAGES * PAGE_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_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 struct lnet_process_id 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)(struct lnet_event *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 struct lnet_handle_md 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_UNREG_RPC = 0xebc0de05,
719 RQ_PHASE_UNREG_BULK = 0xebc0de06,
720 RQ_PHASE_UNDEFINED = 0xebc0de07
723 /** Type of request interpreter call-back */
724 typedef int (*ptlrpc_interpterer_t)(const struct lu_env *env,
725 struct ptlrpc_request *req,
727 /** Type of request resend call-back */
728 typedef void (*ptlrpc_resend_cb_t)(struct ptlrpc_request *req,
732 * Definition of request pool structure.
733 * The pool is used to store empty preallocated requests for the case
734 * when we would actually need to send something without performing
735 * any allocations (to avoid e.g. OOM).
737 struct ptlrpc_request_pool {
738 /** Locks the list */
740 /** list of ptlrpc_request structs */
741 struct list_head prp_req_list;
742 /** Maximum message size that would fit into a rquest from this pool */
744 /** Function to allocate more requests for this pool */
745 int (*prp_populate)(struct ptlrpc_request_pool *, int);
753 #include <lustre_nrs.h>
756 * Basic request prioritization operations structure.
757 * The whole idea is centered around locks and RPCs that might affect locks.
758 * When a lock is contended we try to give priority to RPCs that might lead
759 * to fastest release of that lock.
760 * Currently only implemented for OSTs only in a way that makes all
761 * IO and truncate RPCs that are coming from a locked region where a lock is
762 * contended a priority over other requests.
764 struct ptlrpc_hpreq_ops {
766 * Check if the lock handle of the given lock is the same as
767 * taken from the request.
769 int (*hpreq_lock_match)(struct ptlrpc_request *, struct ldlm_lock *);
771 * Check if the request is a high priority one.
773 int (*hpreq_check)(struct ptlrpc_request *);
775 * Called after the request has been handled.
777 void (*hpreq_fini)(struct ptlrpc_request *);
780 struct ptlrpc_cli_req {
781 /** For bulk requests on client only: bulk descriptor */
782 struct ptlrpc_bulk_desc *cr_bulk;
783 /** optional time limit for send attempts */
784 cfs_duration_t cr_delay_limit;
785 /** time request was first queued */
786 cfs_time_t cr_queued_time;
787 /** request sent timeval */
788 struct timeval cr_sent_tv;
789 /** time for request really sent out */
791 /** when req reply unlink must finish. */
792 time_t cr_reply_deadline;
793 /** when req bulk unlink must finish. */
794 time_t cr_bulk_deadline;
795 /** when req unlink must finish. */
796 time_t cr_req_deadline;
797 /** Portal to which this request would be sent */
799 /** Portal where to wait for reply and where reply would be sent */
801 /** request resending number */
802 unsigned int cr_resend_nr;
803 /** What was import generation when this request was sent */
805 enum lustre_imp_state cr_send_state;
806 /** Per-request waitq introduced by bug 21938 for recovery waiting */
807 wait_queue_head_t cr_set_waitq;
808 /** Link item for request set lists */
809 struct list_head cr_set_chain;
810 /** link to waited ctx */
811 struct list_head cr_ctx_chain;
813 /** client's half ctx */
814 struct ptlrpc_cli_ctx *cr_cli_ctx;
815 /** Link back to the request set */
816 struct ptlrpc_request_set *cr_set;
817 /** outgoing request MD handle */
818 struct lnet_handle_md cr_req_md_h;
819 /** request-out callback parameter */
820 struct ptlrpc_cb_id cr_req_cbid;
821 /** incoming reply MD handle */
822 struct lnet_handle_md cr_reply_md_h;
823 wait_queue_head_t cr_reply_waitq;
824 /** reply callback parameter */
825 struct ptlrpc_cb_id cr_reply_cbid;
826 /** Async completion handler, called when reply is received */
827 ptlrpc_interpterer_t cr_reply_interp;
828 /** Resend handler, called when request is resend to update RPC data */
829 ptlrpc_resend_cb_t cr_resend_cb;
830 /** Async completion context */
831 union ptlrpc_async_args cr_async_args;
832 /** Opaq data for replay and commit callbacks. */
834 /** Link to the imp->imp_unreplied_list */
835 struct list_head cr_unreplied_list;
837 * Commit callback, called when request is committed and about to be
840 void (*cr_commit_cb)(struct ptlrpc_request *);
841 /** Replay callback, called after request is replayed at recovery */
842 void (*cr_replay_cb)(struct ptlrpc_request *);
845 /** client request member alias */
846 /* NB: these alias should NOT be used by any new code, instead they should
847 * be removed step by step to avoid potential abuse */
848 #define rq_bulk rq_cli.cr_bulk
849 #define rq_delay_limit rq_cli.cr_delay_limit
850 #define rq_queued_time rq_cli.cr_queued_time
851 #define rq_sent_tv rq_cli.cr_sent_tv
852 #define rq_real_sent rq_cli.cr_sent_out
853 #define rq_reply_deadline rq_cli.cr_reply_deadline
854 #define rq_bulk_deadline rq_cli.cr_bulk_deadline
855 #define rq_req_deadline rq_cli.cr_req_deadline
856 #define rq_nr_resend rq_cli.cr_resend_nr
857 #define rq_request_portal rq_cli.cr_req_ptl
858 #define rq_reply_portal rq_cli.cr_rep_ptl
859 #define rq_import_generation rq_cli.cr_imp_gen
860 #define rq_send_state rq_cli.cr_send_state
861 #define rq_set_chain rq_cli.cr_set_chain
862 #define rq_ctx_chain rq_cli.cr_ctx_chain
863 #define rq_set rq_cli.cr_set
864 #define rq_set_waitq rq_cli.cr_set_waitq
865 #define rq_cli_ctx rq_cli.cr_cli_ctx
866 #define rq_req_md_h rq_cli.cr_req_md_h
867 #define rq_req_cbid rq_cli.cr_req_cbid
868 #define rq_reply_md_h rq_cli.cr_reply_md_h
869 #define rq_reply_waitq rq_cli.cr_reply_waitq
870 #define rq_reply_cbid rq_cli.cr_reply_cbid
871 #define rq_interpret_reply rq_cli.cr_reply_interp
872 #define rq_resend_cb rq_cli.cr_resend_cb
873 #define rq_async_args rq_cli.cr_async_args
874 #define rq_cb_data rq_cli.cr_cb_data
875 #define rq_unreplied_list rq_cli.cr_unreplied_list
876 #define rq_commit_cb rq_cli.cr_commit_cb
877 #define rq_replay_cb rq_cli.cr_replay_cb
879 struct ptlrpc_srv_req {
880 /** initial thread servicing this request */
881 struct ptlrpc_thread *sr_svc_thread;
883 * Server side list of incoming unserved requests sorted by arrival
884 * time. Traversed from time to time to notice about to expire
885 * requests and sent back "early replies" to clients to let them
886 * know server is alive and well, just very busy to service their
889 struct list_head sr_timed_list;
890 /** server-side per-export list */
891 struct list_head sr_exp_list;
892 /** server-side history, used for debuging purposes. */
893 struct list_head sr_hist_list;
894 /** history sequence # */
896 /** the index of service's srv_at_array into which request is linked */
900 /** authed uid mapped to */
901 uid_t sr_auth_mapped_uid;
902 /** RPC is generated from what part of Lustre */
903 enum lustre_sec_part sr_sp_from;
904 /** request session context */
905 struct lu_context sr_ses;
909 /** stub for NRS request */
910 struct ptlrpc_nrs_request sr_nrq;
912 /** request arrival time */
913 struct timeval sr_arrival_time;
914 /** server's half ctx */
915 struct ptlrpc_svc_ctx *sr_svc_ctx;
916 /** (server side), pointed directly into req buffer */
917 struct ptlrpc_user_desc *sr_user_desc;
918 /** separated reply state, may be vmalloc'd */
919 struct ptlrpc_reply_state *sr_reply_state;
920 /** server-side hp handlers */
921 struct ptlrpc_hpreq_ops *sr_ops;
922 /** incoming request buffer */
923 struct ptlrpc_request_buffer_desc *sr_rqbd;
926 /** server request member alias */
927 /* NB: these alias should NOT be used by any new code, instead they should
928 * be removed step by step to avoid potential abuse */
929 #define rq_svc_thread rq_srv.sr_svc_thread
930 #define rq_timed_list rq_srv.sr_timed_list
931 #define rq_exp_list rq_srv.sr_exp_list
932 #define rq_history_list rq_srv.sr_hist_list
933 #define rq_history_seq rq_srv.sr_hist_seq
934 #define rq_at_index rq_srv.sr_at_index
935 #define rq_auth_uid rq_srv.sr_auth_uid
936 #define rq_auth_mapped_uid rq_srv.sr_auth_mapped_uid
937 #define rq_sp_from rq_srv.sr_sp_from
938 #define rq_session rq_srv.sr_ses
939 #define rq_nrq rq_srv.sr_nrq
940 #define rq_arrival_time rq_srv.sr_arrival_time
941 #define rq_reply_state rq_srv.sr_reply_state
942 #define rq_svc_ctx rq_srv.sr_svc_ctx
943 #define rq_user_desc rq_srv.sr_user_desc
944 #define rq_ops rq_srv.sr_ops
945 #define rq_rqbd rq_srv.sr_rqbd
948 * Represents remote procedure call.
950 * This is a staple structure used by everybody wanting to send a request
953 struct ptlrpc_request {
954 /* Request type: one of PTL_RPC_MSG_* */
956 /** Result of request processing */
959 * Linkage item through which this request is included into
960 * sending/delayed lists on client and into rqbd list on server
962 struct list_head rq_list;
963 /** Lock to protect request flags and some other important bits, like
967 /** client-side flags are serialized by rq_lock @{ */
968 unsigned int rq_intr:1, rq_replied:1, rq_err:1,
969 rq_timedout:1, rq_resend:1, rq_restart:1,
971 * when ->rq_replay is set, request is kept by the client even
972 * after server commits corresponding transaction. This is
973 * used for operations that require sequence of multiple
974 * requests to be replayed. The only example currently is file
975 * open/close. When last request in such a sequence is
976 * committed, ->rq_replay is cleared on all requests in the
980 rq_no_resend:1, rq_waiting:1, rq_receiving_reply:1,
981 rq_no_delay:1, rq_net_err:1, rq_wait_ctx:1,
983 rq_req_unlinked:1, /* unlinked request buffer from lnet */
984 rq_reply_unlinked:1, /* unlinked reply buffer from lnet */
985 rq_memalloc:1, /* req originated from "kswapd" */
987 rq_reply_truncated:1,
988 /** whether the "rq_set" is a valid one */
991 /** do not resend request on -EINPROGRESS */
992 rq_no_retry_einprogress:1,
993 /* allow the req to be sent if the import is in recovery
996 /* bulk request, sent to server, but uncommitted */
1001 /** server-side flags @{ */
1003 rq_hp:1, /**< high priority RPC */
1004 rq_at_linked:1, /**< link into service's srv_at_array */
1005 rq_packed_final:1; /**< packed final reply */
1008 /** one of RQ_PHASE_* */
1009 enum rq_phase rq_phase;
1010 /** one of RQ_PHASE_* to be used next */
1011 enum rq_phase rq_next_phase;
1013 * client-side refcount for SENT race, server-side refcounf
1014 * for multiple replies
1016 atomic_t rq_refcount;
1019 * !rq_truncate : # reply bytes actually received,
1020 * rq_truncate : required repbuf_len for resend
1022 int rq_nob_received;
1023 /** Request length */
1027 /** Pool if request is from preallocated list */
1028 struct ptlrpc_request_pool *rq_pool;
1029 /** Request message - what client sent */
1030 struct lustre_msg *rq_reqmsg;
1031 /** Reply message - server response */
1032 struct lustre_msg *rq_repmsg;
1033 /** Transaction number */
1037 /** bulk match bits */
1040 * List item to for replay list. Not yet committed requests get linked
1042 * Also see \a rq_replay comment above.
1043 * It's also link chain on obd_export::exp_req_replay_queue
1045 struct list_head rq_replay_list;
1046 /** non-shared members for client & server request*/
1048 struct ptlrpc_cli_req rq_cli;
1049 struct ptlrpc_srv_req rq_srv;
1052 * security and encryption data
1054 /** description of flavors for client & server */
1055 struct sptlrpc_flavor rq_flvr;
1057 /* client/server security flags */
1059 rq_ctx_init:1, /* context initiation */
1060 rq_ctx_fini:1, /* context destroy */
1061 rq_bulk_read:1, /* request bulk read */
1062 rq_bulk_write:1, /* request bulk write */
1063 /* server authentication flags */
1064 rq_auth_gss:1, /* authenticated by gss */
1065 rq_auth_usr_root:1, /* authed as root */
1066 rq_auth_usr_mdt:1, /* authed as mdt */
1067 rq_auth_usr_ost:1, /* authed as ost */
1068 /* security tfm flags */
1071 /* doesn't expect reply FIXME */
1073 rq_pill_init:1, /* pill initialized */
1074 rq_srv_req:1; /* server request */
1077 /** various buffer pointers */
1078 struct lustre_msg *rq_reqbuf; /**< req wrapper, vmalloc*/
1079 char *rq_repbuf; /**< rep buffer, vmalloc */
1080 struct lustre_msg *rq_repdata; /**< rep wrapper msg */
1081 /** only in priv mode */
1082 struct lustre_msg *rq_clrbuf;
1083 int rq_reqbuf_len; /* req wrapper buf len */
1084 int rq_reqdata_len; /* req wrapper msg len */
1085 int rq_repbuf_len; /* rep buffer len */
1086 int rq_repdata_len; /* rep wrapper msg len */
1087 int rq_clrbuf_len; /* only in priv mode */
1088 int rq_clrdata_len; /* only in priv mode */
1090 /** early replies go to offset 0, regular replies go after that */
1091 unsigned int rq_reply_off;
1094 /** Fields that help to see if request and reply were swabbed or not */
1095 __u32 rq_req_swab_mask;
1096 __u32 rq_rep_swab_mask;
1098 /** how many early replies (for stats) */
1100 /** Server-side, export on which request was received */
1101 struct obd_export *rq_export;
1102 /** import where request is being sent */
1103 struct obd_import *rq_import;
1106 /** Peer description (the other side) */
1107 struct lnet_process_id rq_peer;
1108 /** Descriptor for the NID from which the peer sent the request. */
1109 struct lnet_process_id rq_source;
1111 * service time estimate (secs)
1112 * If the request is not served by this time, it is marked as timed out.
1116 * when request/reply sent (secs), or time when request should be sent
1119 /** when request must finish. */
1121 /** request format description */
1122 struct req_capsule rq_pill;
1126 * Call completion handler for rpc if any, return it's status or original
1127 * rc if there was no handler defined for this request.
1129 static inline int ptlrpc_req_interpret(const struct lu_env *env,
1130 struct ptlrpc_request *req, int rc)
1132 if (req->rq_interpret_reply != NULL) {
1133 req->rq_status = req->rq_interpret_reply(env, req,
1134 &req->rq_async_args,
1136 return req->rq_status;
1144 int ptlrpc_nrs_policy_register(struct ptlrpc_nrs_pol_conf *conf);
1145 int ptlrpc_nrs_policy_unregister(struct ptlrpc_nrs_pol_conf *conf);
1146 void ptlrpc_nrs_req_hp_move(struct ptlrpc_request *req);
1147 void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
1148 struct ptlrpc_nrs_pol_info *info);
1151 * Can the request be moved from the regular NRS head to the high-priority NRS
1152 * head (of the same PTLRPC service partition), if any?
1154 * For a reliable result, this should be checked under svcpt->scp_req lock.
1156 static inline bool ptlrpc_nrs_req_can_move(struct ptlrpc_request *req)
1158 struct ptlrpc_nrs_request *nrq = &req->rq_nrq;
1161 * LU-898: Check ptlrpc_nrs_request::nr_enqueued to make sure the
1162 * request has been enqueued first, and ptlrpc_nrs_request::nr_started
1163 * to make sure it has not been scheduled yet (analogous to previous
1164 * (non-NRS) checking of !list_empty(&ptlrpc_request::rq_list).
1166 return nrq->nr_enqueued && !nrq->nr_started && !req->rq_hp;
1171 * Returns 1 if request buffer at offset \a index was already swabbed
1173 static inline int lustre_req_swabbed(struct ptlrpc_request *req, size_t index)
1175 LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
1176 return req->rq_req_swab_mask & (1 << index);
1180 * Returns 1 if request reply buffer at offset \a index was already swabbed
1182 static inline int lustre_rep_swabbed(struct ptlrpc_request *req, size_t index)
1184 LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
1185 return req->rq_rep_swab_mask & (1 << index);
1189 * Returns 1 if request needs to be swabbed into local cpu byteorder
1191 static inline int ptlrpc_req_need_swab(struct ptlrpc_request *req)
1193 return lustre_req_swabbed(req, MSG_PTLRPC_HEADER_OFF);
1197 * Returns 1 if request reply needs to be swabbed into local cpu byteorder
1199 static inline int ptlrpc_rep_need_swab(struct ptlrpc_request *req)
1201 return lustre_rep_swabbed(req, MSG_PTLRPC_HEADER_OFF);
1205 * Mark request buffer at offset \a index that it was already swabbed
1207 static inline void lustre_set_req_swabbed(struct ptlrpc_request *req,
1210 LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
1211 LASSERT((req->rq_req_swab_mask & (1 << index)) == 0);
1212 req->rq_req_swab_mask |= 1 << index;
1216 * Mark request reply buffer at offset \a index that it was already swabbed
1218 static inline void lustre_set_rep_swabbed(struct ptlrpc_request *req,
1221 LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
1222 LASSERT((req->rq_rep_swab_mask & (1 << index)) == 0);
1223 req->rq_rep_swab_mask |= 1 << index;
1227 * Convert numerical request phase value \a phase into text string description
1229 static inline const char *
1230 ptlrpc_phase2str(enum rq_phase phase)
1239 case RQ_PHASE_INTERPRET:
1241 case RQ_PHASE_COMPLETE:
1243 case RQ_PHASE_UNREG_RPC:
1245 case RQ_PHASE_UNREG_BULK:
1253 * Convert numerical request phase of the request \a req into text stringi
1256 static inline const char *
1257 ptlrpc_rqphase2str(struct ptlrpc_request *req)
1259 return ptlrpc_phase2str(req->rq_phase);
1263 * Debugging functions and helpers to print request structure into debug log
1266 /* Spare the preprocessor, spoil the bugs. */
1267 #define FLAG(field, str) (field ? str : "")
1269 /** Convert bit flags into a string */
1270 #define DEBUG_REQ_FLAGS(req) \
1271 ptlrpc_rqphase2str(req), \
1272 FLAG(req->rq_intr, "I"), FLAG(req->rq_replied, "R"), \
1273 FLAG(req->rq_err, "E"), FLAG(req->rq_net_err, "e"), \
1274 FLAG(req->rq_timedout, "X") /* eXpired */, FLAG(req->rq_resend, "S"), \
1275 FLAG(req->rq_restart, "T"), FLAG(req->rq_replay, "P"), \
1276 FLAG(req->rq_no_resend, "N"), \
1277 FLAG(req->rq_waiting, "W"), \
1278 FLAG(req->rq_wait_ctx, "C"), FLAG(req->rq_hp, "H"), \
1279 FLAG(req->rq_committed, "M")
1281 #define REQ_FLAGS_FMT "%s:%s%s%s%s%s%s%s%s%s%s%s%s%s"
1283 void _debug_req(struct ptlrpc_request *req,
1284 struct libcfs_debug_msg_data *data, const char *fmt, ...)
1285 __attribute__ ((format (printf, 3, 4)));
1288 * Helper that decides if we need to print request accordig to current debug
1291 #define debug_req(msgdata, mask, cdls, req, fmt, a...) \
1293 CFS_CHECK_STACK(msgdata, mask, cdls); \
1295 if (((mask) & D_CANTMASK) != 0 || \
1296 ((libcfs_debug & (mask)) != 0 && \
1297 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) \
1298 _debug_req((req), msgdata, fmt, ##a); \
1302 * This is the debug print function you need to use to print request sturucture
1303 * content into lustre debug log.
1304 * for most callers (level is a constant) this is resolved at compile time */
1305 #define DEBUG_REQ(level, req, fmt, args...) \
1307 if ((level) & (D_ERROR | D_WARNING)) { \
1308 static struct cfs_debug_limit_state cdls; \
1309 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, level, &cdls); \
1310 debug_req(&msgdata, level, &cdls, req, "@@@ "fmt" ", ## args);\
1312 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, level, NULL); \
1313 debug_req(&msgdata, level, NULL, req, "@@@ "fmt" ", ## args); \
1319 * Structure that defines a single page of a bulk transfer
1321 struct ptlrpc_bulk_page {
1322 /** Linkage to list of pages in a bulk */
1323 struct list_head bp_link;
1325 * Number of bytes in a page to transfer starting from \a bp_pageoffset
1328 /** offset within a page */
1330 /** The page itself */
1331 struct page *bp_page;
1334 enum ptlrpc_bulk_op_type {
1335 PTLRPC_BULK_OP_ACTIVE = 0x00000001,
1336 PTLRPC_BULK_OP_PASSIVE = 0x00000002,
1337 PTLRPC_BULK_OP_PUT = 0x00000004,
1338 PTLRPC_BULK_OP_GET = 0x00000008,
1339 PTLRPC_BULK_BUF_KVEC = 0x00000010,
1340 PTLRPC_BULK_BUF_KIOV = 0x00000020,
1341 PTLRPC_BULK_GET_SOURCE = PTLRPC_BULK_OP_PASSIVE | PTLRPC_BULK_OP_GET,
1342 PTLRPC_BULK_PUT_SINK = PTLRPC_BULK_OP_PASSIVE | PTLRPC_BULK_OP_PUT,
1343 PTLRPC_BULK_GET_SINK = PTLRPC_BULK_OP_ACTIVE | PTLRPC_BULK_OP_GET,
1344 PTLRPC_BULK_PUT_SOURCE = PTLRPC_BULK_OP_ACTIVE | PTLRPC_BULK_OP_PUT,
1347 static inline bool ptlrpc_is_bulk_op_get(enum ptlrpc_bulk_op_type type)
1349 return (type & PTLRPC_BULK_OP_GET) == PTLRPC_BULK_OP_GET;
1352 static inline bool ptlrpc_is_bulk_get_source(enum ptlrpc_bulk_op_type type)
1354 return (type & PTLRPC_BULK_GET_SOURCE) == PTLRPC_BULK_GET_SOURCE;
1357 static inline bool ptlrpc_is_bulk_put_sink(enum ptlrpc_bulk_op_type type)
1359 return (type & PTLRPC_BULK_PUT_SINK) == PTLRPC_BULK_PUT_SINK;
1362 static inline bool ptlrpc_is_bulk_get_sink(enum ptlrpc_bulk_op_type type)
1364 return (type & PTLRPC_BULK_GET_SINK) == PTLRPC_BULK_GET_SINK;
1367 static inline bool ptlrpc_is_bulk_put_source(enum ptlrpc_bulk_op_type type)
1369 return (type & PTLRPC_BULK_PUT_SOURCE) == PTLRPC_BULK_PUT_SOURCE;
1372 static inline bool ptlrpc_is_bulk_desc_kvec(enum ptlrpc_bulk_op_type type)
1374 return ((type & PTLRPC_BULK_BUF_KVEC) | (type & PTLRPC_BULK_BUF_KIOV))
1375 == PTLRPC_BULK_BUF_KVEC;
1378 static inline bool ptlrpc_is_bulk_desc_kiov(enum ptlrpc_bulk_op_type type)
1380 return ((type & PTLRPC_BULK_BUF_KVEC) | (type & PTLRPC_BULK_BUF_KIOV))
1381 == PTLRPC_BULK_BUF_KIOV;
1384 static inline bool ptlrpc_is_bulk_op_active(enum ptlrpc_bulk_op_type type)
1386 return ((type & PTLRPC_BULK_OP_ACTIVE) |
1387 (type & PTLRPC_BULK_OP_PASSIVE))
1388 == PTLRPC_BULK_OP_ACTIVE;
1391 static inline bool ptlrpc_is_bulk_op_passive(enum ptlrpc_bulk_op_type type)
1393 return ((type & PTLRPC_BULK_OP_ACTIVE) |
1394 (type & PTLRPC_BULK_OP_PASSIVE))
1395 == PTLRPC_BULK_OP_PASSIVE;
1398 struct ptlrpc_bulk_frag_ops {
1400 * Add a page \a page to the bulk descriptor \a desc
1401 * Data to transfer in the page starts at offset \a pageoffset and
1402 * amount of data to transfer from the page is \a len
1404 void (*add_kiov_frag)(struct ptlrpc_bulk_desc *desc,
1405 struct page *page, int pageoffset, int len);
1408 * Add a \a fragment to the bulk descriptor \a desc.
1409 * Data to transfer in the fragment is pointed to by \a frag
1410 * The size of the fragment is \a len
1412 int (*add_iov_frag)(struct ptlrpc_bulk_desc *desc, void *frag, int len);
1415 * Uninitialize and free bulk descriptor \a desc.
1416 * Works on bulk descriptors both from server and client side.
1418 void (*release_frags)(struct ptlrpc_bulk_desc *desc);
1421 extern const struct ptlrpc_bulk_frag_ops ptlrpc_bulk_kiov_pin_ops;
1422 extern const struct ptlrpc_bulk_frag_ops ptlrpc_bulk_kiov_nopin_ops;
1423 extern const struct ptlrpc_bulk_frag_ops ptlrpc_bulk_kvec_ops;
1426 * Definition of bulk descriptor.
1427 * Bulks are special "Two phase" RPCs where initial request message
1428 * is sent first and it is followed bt a transfer (o receiving) of a large
1429 * amount of data to be settled into pages referenced from the bulk descriptors.
1430 * Bulks transfers (the actual data following the small requests) are done
1431 * on separate LNet portals.
1432 * In lustre we use bulk transfers for READ and WRITE transfers from/to OSTs.
1433 * Another user is readpage for MDT.
1435 struct ptlrpc_bulk_desc {
1436 /** completed with failure */
1437 unsigned long bd_failure:1;
1439 unsigned long bd_registered:1;
1440 /** For serialization with callback */
1442 /** Import generation when request for this bulk was sent */
1443 int bd_import_generation;
1444 /** {put,get}{source,sink}{kvec,kiov} */
1445 enum ptlrpc_bulk_op_type bd_type;
1446 /** LNet portal for this bulk */
1448 /** Server side - export this bulk created for */
1449 struct obd_export *bd_export;
1450 /** Client side - import this bulk was sent on */
1451 struct obd_import *bd_import;
1452 /** Back pointer to the request */
1453 struct ptlrpc_request *bd_req;
1454 struct ptlrpc_bulk_frag_ops *bd_frag_ops;
1455 wait_queue_head_t bd_waitq; /* server side only WQ */
1456 int bd_iov_count; /* # entries in bd_iov */
1457 int bd_max_iov; /* allocated size of bd_iov */
1458 int bd_nob; /* # bytes covered */
1459 int bd_nob_transferred; /* # bytes GOT/PUT */
1461 __u64 bd_last_mbits;
1463 struct ptlrpc_cb_id bd_cbid; /* network callback info */
1464 lnet_nid_t bd_sender; /* stash event::sender */
1465 int bd_md_count; /* # valid entries in bd_mds */
1466 int bd_md_max_brw; /* max entries in bd_mds */
1467 /** array of associated MDs */
1468 struct lnet_handle_md bd_mds[PTLRPC_BULK_OPS_COUNT];
1473 * encrypt iov, size is either 0 or bd_iov_count.
1475 lnet_kiov_t *bd_enc_vec;
1476 lnet_kiov_t *bd_vec;
1480 struct kvec *bd_enc_kvec;
1481 struct kvec *bd_kvec;
1487 #define GET_KIOV(desc) ((desc)->bd_u.bd_kiov.bd_vec)
1488 #define BD_GET_KIOV(desc, i) ((desc)->bd_u.bd_kiov.bd_vec[i])
1489 #define GET_ENC_KIOV(desc) ((desc)->bd_u.bd_kiov.bd_enc_vec)
1490 #define BD_GET_ENC_KIOV(desc, i) ((desc)->bd_u.bd_kiov.bd_enc_vec[i])
1491 #define GET_KVEC(desc) ((desc)->bd_u.bd_kvec.bd_kvec)
1492 #define BD_GET_KVEC(desc, i) ((desc)->bd_u.bd_kvec.bd_kvec[i])
1493 #define GET_ENC_KVEC(desc) ((desc)->bd_u.bd_kvec.bd_enc_kvec)
1494 #define BD_GET_ENC_KVEC(desc, i) ((desc)->bd_u.bd_kvec.bd_enc_kvec[i])
1498 SVC_STOPPED = 1 << 0,
1499 SVC_STOPPING = 1 << 1,
1500 SVC_STARTING = 1 << 2,
1501 SVC_RUNNING = 1 << 3,
1503 SVC_SIGNAL = 1 << 5,
1506 #define PTLRPC_THR_NAME_LEN 32
1508 * Definition of server service thread structure
1510 struct ptlrpc_thread {
1512 * List of active threads in svc->srv_threads
1514 struct list_head t_link;
1516 * thread-private data (preallocated vmalloc'd memory)
1521 * service thread index, from ptlrpc_start_threads
1525 * service thread pid
1529 * put watchdog in the structure per thread b=14840
1531 struct lc_watchdog *t_watchdog;
1533 * the svc this thread belonged to b=18582
1535 struct ptlrpc_service_part *t_svcpt;
1536 wait_queue_head_t t_ctl_waitq;
1537 struct lu_env *t_env;
1538 char t_name[PTLRPC_THR_NAME_LEN];
1541 static inline int thread_is_init(struct ptlrpc_thread *thread)
1543 return thread->t_flags == 0;
1546 static inline int thread_is_stopped(struct ptlrpc_thread *thread)
1548 return !!(thread->t_flags & SVC_STOPPED);
1551 static inline int thread_is_stopping(struct ptlrpc_thread *thread)
1553 return !!(thread->t_flags & SVC_STOPPING);
1556 static inline int thread_is_starting(struct ptlrpc_thread *thread)
1558 return !!(thread->t_flags & SVC_STARTING);
1561 static inline int thread_is_running(struct ptlrpc_thread *thread)
1563 return !!(thread->t_flags & SVC_RUNNING);
1566 static inline int thread_is_event(struct ptlrpc_thread *thread)
1568 return !!(thread->t_flags & SVC_EVENT);
1571 static inline int thread_is_signal(struct ptlrpc_thread *thread)
1573 return !!(thread->t_flags & SVC_SIGNAL);
1576 static inline void thread_clear_flags(struct ptlrpc_thread *thread, __u32 flags)
1578 thread->t_flags &= ~flags;
1581 static inline void thread_set_flags(struct ptlrpc_thread *thread, __u32 flags)
1583 thread->t_flags = flags;
1586 static inline void thread_add_flags(struct ptlrpc_thread *thread, __u32 flags)
1588 thread->t_flags |= flags;
1591 static inline int thread_test_and_clear_flags(struct ptlrpc_thread *thread,
1594 if (thread->t_flags & flags) {
1595 thread->t_flags &= ~flags;
1602 * Request buffer descriptor structure.
1603 * This is a structure that contains one posted request buffer for service.
1604 * Once data land into a buffer, event callback creates actual request and
1605 * notifies wakes one of the service threads to process new incoming request.
1606 * More than one request can fit into the buffer.
1608 struct ptlrpc_request_buffer_desc {
1609 /** Link item for rqbds on a service */
1610 struct list_head rqbd_list;
1611 /** History of requests for this buffer */
1612 struct list_head rqbd_reqs;
1613 /** Back pointer to service for which this buffer is registered */
1614 struct ptlrpc_service_part *rqbd_svcpt;
1615 /** LNet descriptor */
1616 struct lnet_handle_md rqbd_md_h;
1618 /** The buffer itself */
1620 struct ptlrpc_cb_id rqbd_cbid;
1622 * This "embedded" request structure is only used for the
1623 * last request to fit into the buffer
1625 struct ptlrpc_request rqbd_req;
1628 typedef int (*svc_handler_t)(struct ptlrpc_request *req);
1630 struct ptlrpc_service_ops {
1632 * if non-NULL called during thread creation (ptlrpc_start_thread())
1633 * to initialize service specific per-thread state.
1635 int (*so_thr_init)(struct ptlrpc_thread *thr);
1637 * if non-NULL called during thread shutdown (ptlrpc_main()) to
1638 * destruct state created by ->srv_init().
1640 void (*so_thr_done)(struct ptlrpc_thread *thr);
1642 * Handler function for incoming requests for this service
1644 int (*so_req_handler)(struct ptlrpc_request *req);
1646 * function to determine priority of the request, it's called
1647 * on every new request
1649 int (*so_hpreq_handler)(struct ptlrpc_request *);
1651 * service-specific print fn
1653 void (*so_req_printer)(void *, struct ptlrpc_request *);
1656 #ifndef __cfs_cacheline_aligned
1657 /* NB: put it here for reducing patche dependence */
1658 # define __cfs_cacheline_aligned
1662 * How many high priority requests to serve before serving one normal
1665 #define PTLRPC_SVC_HP_RATIO 10
1668 * Definition of PortalRPC service.
1669 * The service is listening on a particular portal (like tcp port)
1670 * and perform actions for a specific server like IO service for OST
1671 * or general metadata service for MDS.
1673 struct ptlrpc_service {
1674 /** serialize /proc operations */
1675 spinlock_t srv_lock;
1676 /** most often accessed fields */
1677 /** chain thru all services */
1678 struct list_head srv_list;
1679 /** service operations table */
1680 struct ptlrpc_service_ops srv_ops;
1681 /** only statically allocated strings here; we don't clean them */
1683 /** only statically allocated strings here; we don't clean them */
1684 char *srv_thread_name;
1685 /** service thread list */
1686 struct list_head srv_threads;
1687 /** threads # should be created for each partition on initializing */
1688 int srv_nthrs_cpt_init;
1689 /** limit of threads number for each partition */
1690 int srv_nthrs_cpt_limit;
1691 /** Root of /proc dir tree for this service */
1692 struct proc_dir_entry *srv_procroot;
1693 /** Pointer to statistic data for this service */
1694 struct lprocfs_stats *srv_stats;
1695 /** # hp per lp reqs to handle */
1696 int srv_hpreq_ratio;
1697 /** biggest request to receive */
1698 int srv_max_req_size;
1699 /** biggest reply to send */
1700 int srv_max_reply_size;
1701 /** size of individual buffers */
1703 /** # buffers to allocate in 1 group */
1704 int srv_nbuf_per_group;
1705 /** Local portal on which to receive requests */
1706 __u32 srv_req_portal;
1707 /** Portal on the client to send replies to */
1708 __u32 srv_rep_portal;
1710 * Tags for lu_context associated with this thread, see struct
1714 /** soft watchdog timeout multiplier */
1715 int srv_watchdog_factor;
1716 /** under unregister_service */
1717 unsigned srv_is_stopping:1;
1719 /** max # request buffers in history per partition */
1720 int srv_hist_nrqbds_cpt_max;
1721 /** number of CPTs this service bound on */
1723 /** CPTs array this service bound on */
1725 /** 2^srv_cptab_bits >= cfs_cpt_numbert(srv_cptable) */
1727 /** CPT table this service is running over */
1728 struct cfs_cpt_table *srv_cptable;
1730 * partition data for ptlrpc service
1732 struct ptlrpc_service_part *srv_parts[0];
1736 * Definition of PortalRPC service partition data.
1737 * Although a service only has one instance of it right now, but we
1738 * will have multiple instances very soon (instance per CPT).
1740 * it has four locks:
1742 * serialize operations on rqbd and requests waiting for preprocess
1744 * serialize operations active requests sent to this portal
1746 * serialize adaptive timeout stuff
1748 * serialize operations on RS list (reply states)
1750 * We don't have any use-case to take two or more locks at the same time
1751 * for now, so there is no lock order issue.
1753 struct ptlrpc_service_part {
1754 /** back reference to owner */
1755 struct ptlrpc_service *scp_service __cfs_cacheline_aligned;
1756 /* CPT id, reserved */
1758 /** always increasing number */
1760 /** # of starting threads */
1761 int scp_nthrs_starting;
1762 /** # of stopping threads, reserved for shrinking threads */
1763 int scp_nthrs_stopping;
1764 /** # running threads */
1765 int scp_nthrs_running;
1766 /** service threads list */
1767 struct list_head scp_threads;
1770 * serialize the following fields, used for protecting
1771 * rqbd list and incoming requests waiting for preprocess,
1772 * threads starting & stopping are also protected by this lock.
1774 spinlock_t scp_lock __cfs_cacheline_aligned;
1775 /** total # req buffer descs allocated */
1776 int scp_nrqbds_total;
1777 /** # posted request buffers for receiving */
1778 int scp_nrqbds_posted;
1779 /** in progress of allocating rqbd */
1780 int scp_rqbd_allocating;
1781 /** # incoming reqs */
1782 int scp_nreqs_incoming;
1783 /** request buffers to be reposted */
1784 struct list_head scp_rqbd_idle;
1785 /** req buffers receiving */
1786 struct list_head scp_rqbd_posted;
1787 /** incoming reqs */
1788 struct list_head scp_req_incoming;
1789 /** timeout before re-posting reqs, in tick */
1790 cfs_duration_t scp_rqbd_timeout;
1792 * all threads sleep on this. This wait-queue is signalled when new
1793 * incoming request arrives and when difficult reply has to be handled.
1795 wait_queue_head_t scp_waitq;
1797 /** request history */
1798 struct list_head scp_hist_reqs;
1799 /** request buffer history */
1800 struct list_head scp_hist_rqbds;
1801 /** # request buffers in history */
1802 int scp_hist_nrqbds;
1803 /** sequence number for request */
1805 /** highest seq culled from history */
1806 __u64 scp_hist_seq_culled;
1809 * serialize the following fields, used for processing requests
1810 * sent to this portal
1812 spinlock_t scp_req_lock __cfs_cacheline_aligned;
1813 /** # reqs in either of the NRS heads below */
1814 /** # reqs being served */
1815 int scp_nreqs_active;
1816 /** # HPreqs being served */
1817 int scp_nhreqs_active;
1818 /** # hp requests handled */
1821 /** NRS head for regular requests */
1822 struct ptlrpc_nrs scp_nrs_reg;
1823 /** NRS head for HP requests; this is only valid for services that can
1824 * handle HP requests */
1825 struct ptlrpc_nrs *scp_nrs_hp;
1830 * serialize the following fields, used for changes on
1833 spinlock_t scp_at_lock __cfs_cacheline_aligned;
1834 /** estimated rpc service time */
1835 struct adaptive_timeout scp_at_estimate;
1836 /** reqs waiting for replies */
1837 struct ptlrpc_at_array scp_at_array;
1838 /** early reply timer */
1839 struct timer_list scp_at_timer;
1841 cfs_time_t scp_at_checktime;
1842 /** check early replies */
1843 unsigned scp_at_check;
1847 * serialize the following fields, used for processing
1848 * replies for this portal
1850 spinlock_t scp_rep_lock __cfs_cacheline_aligned;
1851 /** all the active replies */
1852 struct list_head scp_rep_active;
1853 /** List of free reply_states */
1854 struct list_head scp_rep_idle;
1855 /** waitq to run, when adding stuff to srv_free_rs_list */
1856 wait_queue_head_t scp_rep_waitq;
1857 /** # 'difficult' replies */
1858 atomic_t scp_nreps_difficult;
1861 #define ptlrpc_service_for_each_part(part, i, svc) \
1863 i < (svc)->srv_ncpts && \
1864 (svc)->srv_parts != NULL && \
1865 ((part) = (svc)->srv_parts[i]) != NULL; i++)
1868 * Declaration of ptlrpcd control structure
1870 struct ptlrpcd_ctl {
1872 * Ptlrpc thread control flags (LIOD_START, LIOD_STOP, LIOD_FORCE)
1874 unsigned long pc_flags;
1876 * Thread lock protecting structure fields.
1882 struct completion pc_starting;
1886 struct completion pc_finishing;
1888 * Thread requests set.
1890 struct ptlrpc_request_set *pc_set;
1892 * Thread name used in kthread_run()
1896 * CPT the thread is bound on.
1900 * Index of ptlrpcd thread in the array.
1904 * Pointer to the array of partners' ptlrpcd_ctl structure.
1906 struct ptlrpcd_ctl **pc_partners;
1908 * Number of the ptlrpcd's partners.
1912 * Record the partner index to be processed next.
1916 * Error code if the thread failed to fully start.
1921 /* Bits for pc_flags */
1922 enum ptlrpcd_ctl_flags {
1924 * Ptlrpc thread start flag.
1926 LIOD_START = 1 << 0,
1928 * Ptlrpc thread stop flag.
1932 * Ptlrpc thread force flag (only stop force so far).
1933 * This will cause aborting any inflight rpcs handled
1934 * by thread if LIOD_STOP is specified.
1936 LIOD_FORCE = 1 << 2,
1938 * This is a recovery ptlrpc thread.
1940 LIOD_RECOVERY = 1 << 3,
1947 * Service compatibility function; the policy is compatible with all services.
1949 * \param[in] svc The service the policy is attempting to register with.
1950 * \param[in] desc The policy descriptor
1952 * \retval true The policy is compatible with the service
1954 * \see ptlrpc_nrs_pol_desc::pd_compat()
1956 static inline bool nrs_policy_compat_all(const struct ptlrpc_service *svc,
1957 const struct ptlrpc_nrs_pol_desc *desc)
1963 * Service compatibility function; the policy is compatible with only a specific
1964 * service which is identified by its human-readable name at
1965 * ptlrpc_service::srv_name.
1967 * \param[in] svc The service the policy is attempting to register with.
1968 * \param[in] desc The policy descriptor
1970 * \retval false The policy is not compatible with the service
1971 * \retval true The policy is compatible with the service
1973 * \see ptlrpc_nrs_pol_desc::pd_compat()
1975 static inline bool nrs_policy_compat_one(const struct ptlrpc_service *svc,
1976 const struct ptlrpc_nrs_pol_desc *desc)
1978 LASSERT(desc->pd_compat_svc_name != NULL);
1979 return strcmp(svc->srv_name, desc->pd_compat_svc_name) == 0;
1984 /* ptlrpc/events.c */
1985 extern struct lnet_handle_eq ptlrpc_eq_h;
1986 extern int ptlrpc_uuid_to_peer(struct obd_uuid *uuid,
1987 struct lnet_process_id *peer, lnet_nid_t *self);
1989 * These callbacks are invoked by LNet when something happened to
1993 extern void request_out_callback(struct lnet_event *ev);
1994 extern void reply_in_callback(struct lnet_event *ev);
1995 extern void client_bulk_callback(struct lnet_event *ev);
1996 extern void request_in_callback(struct lnet_event *ev);
1997 extern void reply_out_callback(struct lnet_event *ev);
1998 #ifdef HAVE_SERVER_SUPPORT
1999 extern void server_bulk_callback(struct lnet_event *ev);
2003 /* ptlrpc/connection.c */
2004 struct ptlrpc_connection *ptlrpc_connection_get(struct lnet_process_id peer,
2006 struct obd_uuid *uuid);
2007 int ptlrpc_connection_put(struct ptlrpc_connection *c);
2008 struct ptlrpc_connection *ptlrpc_connection_addref(struct ptlrpc_connection *);
2009 int ptlrpc_connection_init(void);
2010 void ptlrpc_connection_fini(void);
2011 extern lnet_pid_t ptl_get_pid(void);
2013 /* ptlrpc/niobuf.c */
2015 * Actual interfacing with LNet to put/get/register/unregister stuff
2018 #ifdef HAVE_SERVER_SUPPORT
2019 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_exp(struct ptlrpc_request *req,
2020 unsigned nfrags, unsigned max_brw,
2023 const struct ptlrpc_bulk_frag_ops
2025 int ptlrpc_start_bulk_transfer(struct ptlrpc_bulk_desc *desc);
2026 void ptlrpc_abort_bulk(struct ptlrpc_bulk_desc *desc);
2028 static inline int ptlrpc_server_bulk_active(struct ptlrpc_bulk_desc *desc)
2032 LASSERT(desc != NULL);
2034 spin_lock(&desc->bd_lock);
2035 rc = desc->bd_md_count;
2036 spin_unlock(&desc->bd_lock);
2041 int ptlrpc_register_bulk(struct ptlrpc_request *req);
2042 int ptlrpc_unregister_bulk(struct ptlrpc_request *req, int async);
2044 static inline int ptlrpc_client_bulk_active(struct ptlrpc_request *req)
2046 struct ptlrpc_bulk_desc *desc;
2049 LASSERT(req != NULL);
2050 desc = req->rq_bulk;
2052 if (req->rq_bulk_deadline > cfs_time_current_sec())
2058 spin_lock(&desc->bd_lock);
2059 rc = desc->bd_md_count;
2060 spin_unlock(&desc->bd_lock);
2064 #define PTLRPC_REPLY_MAYBE_DIFFICULT 0x01
2065 #define PTLRPC_REPLY_EARLY 0x02
2066 int ptlrpc_send_reply(struct ptlrpc_request *req, int flags);
2067 int ptlrpc_reply(struct ptlrpc_request *req);
2068 int ptlrpc_send_error(struct ptlrpc_request *req, int difficult);
2069 int ptlrpc_error(struct ptlrpc_request *req);
2070 int ptlrpc_at_get_net_latency(struct ptlrpc_request *req);
2071 int ptl_send_rpc(struct ptlrpc_request *request, int noreply);
2072 int ptlrpc_register_rqbd(struct ptlrpc_request_buffer_desc *rqbd);
2075 /* ptlrpc/client.c */
2077 * Client-side portals API. Everything to send requests, receive replies,
2078 * request queues, request management, etc.
2081 void ptlrpc_request_committed(struct ptlrpc_request *req, int force);
2083 void ptlrpc_init_client(int req_portal, int rep_portal, char *name,
2084 struct ptlrpc_client *);
2085 void ptlrpc_cleanup_client(struct obd_import *imp);
2086 struct ptlrpc_connection *ptlrpc_uuid_to_connection(struct obd_uuid *uuid,
2087 lnet_nid_t nid4refnet);
2089 int ptlrpc_queue_wait(struct ptlrpc_request *req);
2090 int ptlrpc_replay_req(struct ptlrpc_request *req);
2091 void ptlrpc_restart_req(struct ptlrpc_request *req);
2092 void ptlrpc_abort_inflight(struct obd_import *imp);
2093 void ptlrpc_cleanup_imp(struct obd_import *imp);
2094 void ptlrpc_abort_set(struct ptlrpc_request_set *set);
2096 struct ptlrpc_request_set *ptlrpc_prep_set(void);
2097 struct ptlrpc_request_set *ptlrpc_prep_fcset(int max, set_producer_func func,
2099 int ptlrpc_set_add_cb(struct ptlrpc_request_set *set,
2100 set_interpreter_func fn, void *data);
2101 int ptlrpc_check_set(const struct lu_env *env, struct ptlrpc_request_set *set);
2102 int ptlrpc_set_wait(struct ptlrpc_request_set *);
2103 void ptlrpc_mark_interrupted(struct ptlrpc_request *req);
2104 void ptlrpc_set_destroy(struct ptlrpc_request_set *);
2105 void ptlrpc_set_add_req(struct ptlrpc_request_set *, struct ptlrpc_request *);
2107 void ptlrpc_free_rq_pool(struct ptlrpc_request_pool *pool);
2108 int ptlrpc_add_rqs_to_pool(struct ptlrpc_request_pool *pool, int num_rq);
2110 struct ptlrpc_request_pool *
2111 ptlrpc_init_rq_pool(int, int,
2112 int (*populate_pool)(struct ptlrpc_request_pool *, int));
2114 void ptlrpc_at_set_req_timeout(struct ptlrpc_request *req);
2115 struct ptlrpc_request *ptlrpc_request_alloc(struct obd_import *imp,
2116 const struct req_format *format);
2117 struct ptlrpc_request *ptlrpc_request_alloc_pool(struct obd_import *imp,
2118 struct ptlrpc_request_pool *,
2119 const struct req_format *format);
2120 void ptlrpc_request_free(struct ptlrpc_request *request);
2121 int ptlrpc_request_pack(struct ptlrpc_request *request,
2122 __u32 version, int opcode);
2123 struct ptlrpc_request *ptlrpc_request_alloc_pack(struct obd_import *imp,
2124 const struct req_format *format,
2125 __u32 version, int opcode);
2126 int ptlrpc_request_bufs_pack(struct ptlrpc_request *request,
2127 __u32 version, int opcode, char **bufs,
2128 struct ptlrpc_cli_ctx *ctx);
2129 void ptlrpc_req_finished(struct ptlrpc_request *request);
2130 void ptlrpc_req_finished_with_imp_lock(struct ptlrpc_request *request);
2131 struct ptlrpc_request *ptlrpc_request_addref(struct ptlrpc_request *req);
2132 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_imp(struct ptlrpc_request *req,
2133 unsigned nfrags, unsigned max_brw,
2136 const struct ptlrpc_bulk_frag_ops
2139 int ptlrpc_prep_bulk_frag(struct ptlrpc_bulk_desc *desc,
2140 void *frag, int len);
2141 void __ptlrpc_prep_bulk_page(struct ptlrpc_bulk_desc *desc,
2142 struct page *page, int pageoffset, int len,
2144 static inline void ptlrpc_prep_bulk_page_pin(struct ptlrpc_bulk_desc *desc,
2145 struct page *page, int pageoffset,
2148 __ptlrpc_prep_bulk_page(desc, page, pageoffset, len, 1);
2151 static inline void ptlrpc_prep_bulk_page_nopin(struct ptlrpc_bulk_desc *desc,
2152 struct page *page, int pageoffset,
2155 __ptlrpc_prep_bulk_page(desc, page, pageoffset, len, 0);
2158 void ptlrpc_free_bulk(struct ptlrpc_bulk_desc *bulk);
2160 static inline void ptlrpc_release_bulk_page_pin(struct ptlrpc_bulk_desc *desc)
2164 for (i = 0; i < desc->bd_iov_count ; i++)
2165 put_page(BD_GET_KIOV(desc, i).kiov_page);
2168 static inline void ptlrpc_release_bulk_noop(struct ptlrpc_bulk_desc *desc)
2172 void ptlrpc_retain_replayable_request(struct ptlrpc_request *req,
2173 struct obd_import *imp);
2174 __u64 ptlrpc_next_xid(void);
2175 __u64 ptlrpc_sample_next_xid(void);
2176 __u64 ptlrpc_req_xid(struct ptlrpc_request *request);
2178 /* Set of routines to run a function in ptlrpcd context */
2179 void *ptlrpcd_alloc_work(struct obd_import *imp,
2180 int (*cb)(const struct lu_env *, void *), void *data);
2181 void ptlrpcd_destroy_work(void *handler);
2182 int ptlrpcd_queue_work(void *handler);
2185 struct ptlrpc_service_buf_conf {
2186 /* nbufs is buffers # to allocate when growing the pool */
2187 unsigned int bc_nbufs;
2188 /* buffer size to post */
2189 unsigned int bc_buf_size;
2190 /* portal to listed for requests on */
2191 unsigned int bc_req_portal;
2192 /* portal of where to send replies to */
2193 unsigned int bc_rep_portal;
2194 /* maximum request size to be accepted for this service */
2195 unsigned int bc_req_max_size;
2196 /* maximum reply size this service can ever send */
2197 unsigned int bc_rep_max_size;
2200 struct ptlrpc_service_thr_conf {
2201 /* threadname should be 8 characters or less - 6 will be added on */
2203 /* threads increasing factor for each CPU */
2204 unsigned int tc_thr_factor;
2205 /* service threads # to start on each partition while initializing */
2206 unsigned int tc_nthrs_init;
2208 * low water of threads # upper-limit on each partition while running,
2209 * service availability may be impacted if threads number is lower
2210 * than this value. It can be ZERO if the service doesn't require
2211 * CPU affinity or there is only one partition.
2213 unsigned int tc_nthrs_base;
2214 /* "soft" limit for total threads number */
2215 unsigned int tc_nthrs_max;
2216 /* user specified threads number, it will be validated due to
2217 * other members of this structure. */
2218 unsigned int tc_nthrs_user;
2219 /* set NUMA node affinity for service threads */
2220 unsigned int tc_cpu_affinity;
2221 /* Tags for lu_context associated with service thread */
2225 struct ptlrpc_service_cpt_conf {
2226 struct cfs_cpt_table *cc_cptable;
2227 /* string pattern to describe CPTs for a service */
2231 struct ptlrpc_service_conf {
2234 /* soft watchdog timeout multiplifier to print stuck service traces */
2235 unsigned int psc_watchdog_factor;
2236 /* buffer information */
2237 struct ptlrpc_service_buf_conf psc_buf;
2238 /* thread information */
2239 struct ptlrpc_service_thr_conf psc_thr;
2240 /* CPU partition information */
2241 struct ptlrpc_service_cpt_conf psc_cpt;
2242 /* function table */
2243 struct ptlrpc_service_ops psc_ops;
2246 /* ptlrpc/service.c */
2248 * Server-side services API. Register/unregister service, request state
2249 * management, service thread management
2253 void ptlrpc_save_lock(struct ptlrpc_request *req,
2254 struct lustre_handle *lock, int mode, int no_ack);
2255 void ptlrpc_commit_replies(struct obd_export *exp);
2256 void ptlrpc_dispatch_difficult_reply(struct ptlrpc_reply_state *rs);
2257 void ptlrpc_schedule_difficult_reply(struct ptlrpc_reply_state *rs);
2258 int ptlrpc_hpreq_handler(struct ptlrpc_request *req);
2259 struct ptlrpc_service *ptlrpc_register_service(
2260 struct ptlrpc_service_conf *conf,
2261 struct proc_dir_entry *proc_entry);
2262 void ptlrpc_stop_all_threads(struct ptlrpc_service *svc);
2264 int ptlrpc_start_threads(struct ptlrpc_service *svc);
2265 int ptlrpc_unregister_service(struct ptlrpc_service *service);
2266 int liblustre_check_services(void *arg);
2267 void ptlrpc_daemonize(char *name);
2268 int ptlrpc_service_health_check(struct ptlrpc_service *);
2269 void ptlrpc_server_drop_request(struct ptlrpc_request *req);
2270 void ptlrpc_request_change_export(struct ptlrpc_request *req,
2271 struct obd_export *export);
2272 void ptlrpc_update_export_timer(struct obd_export *exp, long extra_delay);
2274 int ptlrpc_hr_init(void);
2275 void ptlrpc_hr_fini(void);
2279 /* ptlrpc/import.c */
2284 int ptlrpc_connect_import(struct obd_import *imp);
2285 int ptlrpc_init_import(struct obd_import *imp);
2286 int ptlrpc_disconnect_import(struct obd_import *imp, int noclose);
2287 int ptlrpc_import_recovery_state_machine(struct obd_import *imp);
2288 void deuuidify(char *uuid, const char *prefix, char **uuid_start,
2290 void ptlrpc_import_enter_resend(struct obd_import *imp);
2291 /* ptlrpc/pack_generic.c */
2292 int ptlrpc_reconnect_import(struct obd_import *imp);
2296 * ptlrpc msg buffer and swab interface
2300 int ptlrpc_buf_need_swab(struct ptlrpc_request *req, const int inout,
2302 void ptlrpc_buf_set_swabbed(struct ptlrpc_request *req, const int inout,
2304 int ptlrpc_unpack_rep_msg(struct ptlrpc_request *req, int len);
2305 int ptlrpc_unpack_req_msg(struct ptlrpc_request *req, int len);
2307 int lustre_msg_check_version(struct lustre_msg *msg, __u32 version);
2308 void lustre_init_msg_v2(struct lustre_msg_v2 *msg, int count, __u32 *lens,
2310 int lustre_pack_request(struct ptlrpc_request *, __u32 magic, int count,
2311 __u32 *lens, char **bufs);
2312 int lustre_pack_reply(struct ptlrpc_request *, int count, __u32 *lens,
2314 int lustre_pack_reply_v2(struct ptlrpc_request *req, int count,
2315 __u32 *lens, char **bufs, int flags);
2316 #define LPRFL_EARLY_REPLY 1
2317 int lustre_pack_reply_flags(struct ptlrpc_request *, int count, __u32 *lens,
2318 char **bufs, int flags);
2319 int lustre_shrink_msg(struct lustre_msg *msg, int segment,
2320 unsigned int newlen, int move_data);
2321 void lustre_free_reply_state(struct ptlrpc_reply_state *rs);
2322 int __lustre_unpack_msg(struct lustre_msg *m, int len);
2323 __u32 lustre_msg_hdr_size(__u32 magic, __u32 count);
2324 __u32 lustre_msg_size(__u32 magic, int count, __u32 *lengths);
2325 __u32 lustre_msg_size_v2(int count, __u32 *lengths);
2326 __u32 lustre_packed_msg_size(struct lustre_msg *msg);
2327 __u32 lustre_msg_early_size(void);
2328 void *lustre_msg_buf_v2(struct lustre_msg_v2 *m, __u32 n, __u32 min_size);
2329 void *lustre_msg_buf(struct lustre_msg *m, __u32 n, __u32 minlen);
2330 __u32 lustre_msg_buflen(struct lustre_msg *m, __u32 n);
2331 void lustre_msg_set_buflen(struct lustre_msg *m, __u32 n, __u32 len);
2332 __u32 lustre_msg_bufcount(struct lustre_msg *m);
2333 char *lustre_msg_string(struct lustre_msg *m, __u32 n, __u32 max_len);
2334 __u32 lustre_msghdr_get_flags(struct lustre_msg *msg);
2335 void lustre_msghdr_set_flags(struct lustre_msg *msg, __u32 flags);
2336 __u32 lustre_msg_get_flags(struct lustre_msg *msg);
2337 void lustre_msg_add_flags(struct lustre_msg *msg, __u32 flags);
2338 void lustre_msg_set_flags(struct lustre_msg *msg, __u32 flags);
2339 void lustre_msg_clear_flags(struct lustre_msg *msg, __u32 flags);
2340 __u32 lustre_msg_get_op_flags(struct lustre_msg *msg);
2341 void lustre_msg_add_op_flags(struct lustre_msg *msg, __u32 flags);
2342 struct lustre_handle *lustre_msg_get_handle(struct lustre_msg *msg);
2343 __u32 lustre_msg_get_type(struct lustre_msg *msg);
2344 __u32 lustre_msg_get_version(struct lustre_msg *msg);
2345 void lustre_msg_add_version(struct lustre_msg *msg, __u32 version);
2346 __u32 lustre_msg_get_opc(struct lustre_msg *msg);
2347 __u64 lustre_msg_get_last_xid(struct lustre_msg *msg);
2348 __u16 lustre_msg_get_tag(struct lustre_msg *msg);
2349 __u64 lustre_msg_get_last_committed(struct lustre_msg *msg);
2350 __u64 *lustre_msg_get_versions(struct lustre_msg *msg);
2351 __u64 lustre_msg_get_transno(struct lustre_msg *msg);
2352 __u64 lustre_msg_get_slv(struct lustre_msg *msg);
2353 __u32 lustre_msg_get_limit(struct lustre_msg *msg);
2354 void lustre_msg_set_slv(struct lustre_msg *msg, __u64 slv);
2355 void lustre_msg_set_limit(struct lustre_msg *msg, __u64 limit);
2356 int lustre_msg_get_status(struct lustre_msg *msg);
2357 __u32 lustre_msg_get_conn_cnt(struct lustre_msg *msg);
2358 __u32 lustre_msg_get_magic(struct lustre_msg *msg);
2359 __u32 lustre_msg_get_timeout(struct lustre_msg *msg);
2360 __u32 lustre_msg_get_service_time(struct lustre_msg *msg);
2361 char *lustre_msg_get_jobid(struct lustre_msg *msg);
2362 __u32 lustre_msg_get_cksum(struct lustre_msg *msg);
2363 __u64 lustre_msg_get_mbits(struct lustre_msg *msg);
2364 __u32 lustre_msg_calc_cksum(struct lustre_msg *msg);
2365 void lustre_msg_set_handle(struct lustre_msg *msg,struct lustre_handle *handle);
2366 void lustre_msg_set_type(struct lustre_msg *msg, __u32 type);
2367 void lustre_msg_set_opc(struct lustre_msg *msg, __u32 opc);
2368 void lustre_msg_set_last_xid(struct lustre_msg *msg, __u64 last_xid);
2369 void lustre_msg_set_tag(struct lustre_msg *msg, __u16 tag);
2370 void lustre_msg_set_last_committed(struct lustre_msg *msg,__u64 last_committed);
2371 void lustre_msg_set_versions(struct lustre_msg *msg, __u64 *versions);
2372 void lustre_msg_set_transno(struct lustre_msg *msg, __u64 transno);
2373 void lustre_msg_set_status(struct lustre_msg *msg, __u32 status);
2374 void lustre_msg_set_conn_cnt(struct lustre_msg *msg, __u32 conn_cnt);
2375 void ptlrpc_req_set_repsize(struct ptlrpc_request *req, int count, __u32 *sizes);
2376 void ptlrpc_request_set_replen(struct ptlrpc_request *req);
2377 void lustre_msg_set_timeout(struct lustre_msg *msg, __u32 timeout);
2378 void lustre_msg_set_service_time(struct lustre_msg *msg, __u32 service_time);
2379 void lustre_msg_set_jobid(struct lustre_msg *msg, char *jobid);
2380 void lustre_msg_set_cksum(struct lustre_msg *msg, __u32 cksum);
2381 void lustre_msg_set_mbits(struct lustre_msg *msg, __u64 mbits);
2384 lustre_shrink_reply(struct ptlrpc_request *req, int segment,
2385 unsigned int newlen, int move_data)
2387 LASSERT(req->rq_reply_state);
2388 LASSERT(req->rq_repmsg);
2389 req->rq_replen = lustre_shrink_msg(req->rq_repmsg, segment,
2393 #ifdef LUSTRE_TRANSLATE_ERRNOS
2395 static inline int ptlrpc_status_hton(int h)
2398 * Positive errnos must be network errnos, such as LUSTRE_EDEADLK,
2399 * ELDLM_LOCK_ABORTED, etc.
2402 return -lustre_errno_hton(-h);
2407 static inline int ptlrpc_status_ntoh(int n)
2410 * See the comment in ptlrpc_status_hton().
2413 return -lustre_errno_ntoh(-n);
2420 #define ptlrpc_status_hton(h) (h)
2421 #define ptlrpc_status_ntoh(n) (n)
2426 /** Change request phase of \a req to \a new_phase */
2428 ptlrpc_rqphase_move(struct ptlrpc_request *req, enum rq_phase new_phase)
2430 if (req->rq_phase == new_phase)
2433 if (new_phase == RQ_PHASE_UNREG_RPC ||
2434 new_phase == RQ_PHASE_UNREG_BULK) {
2435 /* No embedded unregistering phases */
2436 if (req->rq_phase == RQ_PHASE_UNREG_RPC ||
2437 req->rq_phase == RQ_PHASE_UNREG_BULK)
2440 req->rq_next_phase = req->rq_phase;
2442 atomic_inc(&req->rq_import->imp_unregistering);
2445 if (req->rq_phase == RQ_PHASE_UNREG_RPC ||
2446 req->rq_phase == RQ_PHASE_UNREG_BULK) {
2448 atomic_dec(&req->rq_import->imp_unregistering);
2451 DEBUG_REQ(D_INFO, req, "move req \"%s\" -> \"%s\"",
2452 ptlrpc_rqphase2str(req), ptlrpc_phase2str(new_phase));
2454 req->rq_phase = new_phase;
2458 * Returns true if request \a req got early reply and hard deadline is not met
2461 ptlrpc_client_early(struct ptlrpc_request *req)
2463 return req->rq_early;
2467 * Returns true if we got real reply from server for this request
2470 ptlrpc_client_replied(struct ptlrpc_request *req)
2472 if (req->rq_reply_deadline > cfs_time_current_sec())
2474 return req->rq_replied;
2477 /** Returns true if request \a req is in process of receiving server reply */
2479 ptlrpc_client_recv(struct ptlrpc_request *req)
2481 if (req->rq_reply_deadline > cfs_time_current_sec())
2483 return req->rq_receiving_reply;
2487 ptlrpc_client_recv_or_unlink(struct ptlrpc_request *req)
2491 spin_lock(&req->rq_lock);
2492 if (req->rq_reply_deadline > cfs_time_current_sec()) {
2493 spin_unlock(&req->rq_lock);
2496 if (req->rq_req_deadline > cfs_time_current_sec()) {
2497 spin_unlock(&req->rq_lock);
2501 rc = !req->rq_req_unlinked || !req->rq_reply_unlinked ||
2502 req->rq_receiving_reply;
2503 spin_unlock(&req->rq_lock);
2508 ptlrpc_client_wake_req(struct ptlrpc_request *req)
2510 if (req->rq_set == NULL)
2511 wake_up(&req->rq_reply_waitq);
2513 wake_up(&req->rq_set->set_waitq);
2517 ptlrpc_rs_addref(struct ptlrpc_reply_state *rs)
2519 LASSERT(atomic_read(&rs->rs_refcount) > 0);
2520 atomic_inc(&rs->rs_refcount);
2524 ptlrpc_rs_decref(struct ptlrpc_reply_state *rs)
2526 LASSERT(atomic_read(&rs->rs_refcount) > 0);
2527 if (atomic_dec_and_test(&rs->rs_refcount))
2528 lustre_free_reply_state(rs);
2531 /* Should only be called once per req */
2532 static inline void ptlrpc_req_drop_rs(struct ptlrpc_request *req)
2534 if (req->rq_reply_state == NULL)
2535 return; /* shouldn't occur */
2536 ptlrpc_rs_decref(req->rq_reply_state);
2537 req->rq_reply_state = NULL;
2538 req->rq_repmsg = NULL;
2541 static inline __u32 lustre_request_magic(struct ptlrpc_request *req)
2543 return lustre_msg_get_magic(req->rq_reqmsg);
2546 static inline int ptlrpc_req_get_repsize(struct ptlrpc_request *req)
2548 switch (req->rq_reqmsg->lm_magic) {
2549 case LUSTRE_MSG_MAGIC_V2:
2550 return req->rq_reqmsg->lm_repsize;
2552 LASSERTF(0, "incorrect message magic: %08x\n",
2553 req->rq_reqmsg->lm_magic);
2558 static inline int ptlrpc_send_limit_expired(struct ptlrpc_request *req)
2560 if (req->rq_delay_limit != 0 &&
2561 cfs_time_before(cfs_time_add(req->rq_queued_time,
2562 cfs_time_seconds(req->rq_delay_limit)),
2563 cfs_time_current())) {
2569 static inline int ptlrpc_no_resend(struct ptlrpc_request *req)
2571 if (!req->rq_no_resend && ptlrpc_send_limit_expired(req)) {
2572 spin_lock(&req->rq_lock);
2573 req->rq_no_resend = 1;
2574 spin_unlock(&req->rq_lock);
2576 return req->rq_no_resend;
2580 ptlrpc_server_get_timeout(struct ptlrpc_service_part *svcpt)
2582 int at = AT_OFF ? 0 : at_get(&svcpt->scp_at_estimate);
2584 return svcpt->scp_service->srv_watchdog_factor *
2585 max_t(int, at, obd_timeout);
2588 static inline struct ptlrpc_service *
2589 ptlrpc_req2svc(struct ptlrpc_request *req)
2591 LASSERT(req->rq_rqbd != NULL);
2592 return req->rq_rqbd->rqbd_svcpt->scp_service;
2595 /* ldlm/ldlm_lib.c */
2597 * Target client logic
2600 int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg);
2601 int client_obd_cleanup(struct obd_device *obddev);
2602 int client_connect_import(const struct lu_env *env,
2603 struct obd_export **exp, struct obd_device *obd,
2604 struct obd_uuid *cluuid, struct obd_connect_data *,
2606 int client_disconnect_export(struct obd_export *exp);
2607 int client_import_add_conn(struct obd_import *imp, struct obd_uuid *uuid,
2609 int client_import_del_conn(struct obd_import *imp, struct obd_uuid *uuid);
2610 int client_import_find_conn(struct obd_import *imp, lnet_nid_t peer,
2611 struct obd_uuid *uuid);
2612 int import_set_conn_priority(struct obd_import *imp, struct obd_uuid *uuid);
2613 void client_destroy_import(struct obd_import *imp);
2616 #ifdef HAVE_SERVER_SUPPORT
2617 int server_disconnect_export(struct obd_export *exp);
2620 /* ptlrpc/pinger.c */
2622 * Pinger API (client side only)
2625 enum timeout_event {
2628 struct timeout_item;
2629 typedef int (*timeout_cb_t)(struct timeout_item *, void *);
2630 int ptlrpc_pinger_add_import(struct obd_import *imp);
2631 int ptlrpc_pinger_del_import(struct obd_import *imp);
2632 int ptlrpc_add_timeout_client(int time, enum timeout_event event,
2633 timeout_cb_t cb, void *data,
2634 struct list_head *obd_list);
2635 int ptlrpc_del_timeout_client(struct list_head *obd_list,
2636 enum timeout_event event);
2637 struct ptlrpc_request * ptlrpc_prep_ping(struct obd_import *imp);
2638 int ptlrpc_obd_ping(struct obd_device *obd);
2639 void ping_evictor_start(void);
2640 void ping_evictor_stop(void);
2641 void ptlrpc_pinger_ir_up(void);
2642 void ptlrpc_pinger_ir_down(void);
2644 int ptlrpc_pinger_suppress_pings(void);
2646 /* ptlrpc/ptlrpcd.c */
2647 void ptlrpcd_stop(struct ptlrpcd_ctl *pc, int force);
2648 void ptlrpcd_free(struct ptlrpcd_ctl *pc);
2649 void ptlrpcd_wake(struct ptlrpc_request *req);
2650 void ptlrpcd_add_req(struct ptlrpc_request *req);
2651 void ptlrpcd_add_rqset(struct ptlrpc_request_set *set);
2652 int ptlrpcd_addref(void);
2653 void ptlrpcd_decref(void);
2655 /* ptlrpc/lproc_ptlrpc.c */
2657 * procfs output related functions
2660 const char* ll_opcode2str(__u32 opcode);
2661 const int ll_str2opcode(const char *ops);
2662 #ifdef CONFIG_PROC_FS
2663 void ptlrpc_lprocfs_register_obd(struct obd_device *obd);
2664 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd);
2665 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes);
2667 static inline void ptlrpc_lprocfs_register_obd(struct obd_device *obd) {}
2668 static inline void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd) {}
2669 static inline void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes) {}
2673 /* ptlrpc/llog_server.c */
2674 int llog_origin_handle_open(struct ptlrpc_request *req);
2675 int llog_origin_handle_destroy(struct ptlrpc_request *req);
2676 int llog_origin_handle_prev_block(struct ptlrpc_request *req);
2677 int llog_origin_handle_next_block(struct ptlrpc_request *req);
2678 int llog_origin_handle_read_header(struct ptlrpc_request *req);
2679 int llog_origin_handle_close(struct ptlrpc_request *req);
2681 /* ptlrpc/llog_client.c */
2682 extern struct llog_operations llog_client_ops;