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.
53 #include <linux/kobject.h>
54 #include <linux/uio.h>
55 #include <libcfs/libcfs.h>
57 #include <uapi/linux/lnet/nidstr.h>
58 #include <uapi/linux/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 <uapi/linux/lustre/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 (1U << PTLRPC_MAX_BRW_BITS)
101 #define PTLRPC_MAX_BRW_PAGES (PTLRPC_MAX_BRW_SIZE >> PAGE_SHIFT)
103 #define ONE_MB_BRW_SIZE (1U << LNET_MTU_BITS)
104 #define MD_MAX_BRW_SIZE (1U << 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_DEF_BRW_SIZE (4 * ONE_MB_BRW_SIZE)
108 #define DT_MAX_BRW_PAGES (DT_MAX_BRW_SIZE >> PAGE_SHIFT)
109 #define OFD_MAX_BRW_SIZE (1U << LNET_MTU_BITS)
111 /* When PAGE_SIZE is a constant, we can check our arithmetic here with cpp! */
112 #if ((PTLRPC_MAX_BRW_PAGES & (PTLRPC_MAX_BRW_PAGES - 1)) != 0)
113 # error "PTLRPC_MAX_BRW_PAGES isn't a power of two"
115 #if (PTLRPC_MAX_BRW_SIZE != (PTLRPC_MAX_BRW_PAGES * PAGE_SIZE))
116 # error "PTLRPC_MAX_BRW_SIZE isn't PTLRPC_MAX_BRW_PAGES * PAGE_SIZE"
118 #if (PTLRPC_MAX_BRW_SIZE > LNET_MTU * PTLRPC_BULK_OPS_COUNT)
119 # error "PTLRPC_MAX_BRW_SIZE too big"
121 #if (PTLRPC_MAX_BRW_PAGES > LNET_MAX_IOV * PTLRPC_BULK_OPS_COUNT)
122 # error "PTLRPC_MAX_BRW_PAGES too big"
125 #define PTLRPC_NTHRS_INIT 2
130 * Constants determine how memory is used to buffer incoming service requests.
132 * ?_NBUFS # buffers to allocate when growing the pool
133 * ?_BUFSIZE # bytes in a single request buffer
134 * ?_MAXREQSIZE # maximum request service will receive
136 * When fewer than ?_NBUFS/2 buffers are posted for receive, another chunk
137 * of ?_NBUFS is added to the pool.
139 * Messages larger than ?_MAXREQSIZE are dropped. Request buffers are
140 * considered full when less than ?_MAXREQSIZE is left in them.
145 * Constants determine how threads are created for ptlrpc service.
147 * ?_NTHRS_INIT # threads to create for each service partition on
148 * initializing. If it's non-affinity service and
149 * there is only one partition, it's the overall #
150 * threads for the service while initializing.
151 * ?_NTHRS_BASE # threads should be created at least for each
152 * ptlrpc partition to keep the service healthy.
153 * It's the low-water mark of threads upper-limit
154 * for each partition.
155 * ?_THR_FACTOR # threads can be added on threads upper-limit for
156 * each CPU core. This factor is only for reference,
157 * we might decrease value of factor if number of cores
158 * per CPT is above a limit.
159 * ?_NTHRS_MAX # overall threads can be created for a service,
160 * it's a soft limit because if service is running
161 * on machine with hundreds of cores and tens of
162 * CPU partitions, we need to guarantee each partition
163 * has ?_NTHRS_BASE threads, which means total threads
164 * will be ?_NTHRS_BASE * number_of_cpts which can
165 * exceed ?_NTHRS_MAX.
169 * #define MDS_NTHRS_INIT 2
170 * #define MDS_NTHRS_BASE 64
171 * #define MDS_NTHRS_FACTOR 8
172 * #define MDS_NTHRS_MAX 1024
175 * ---------------------------------------------------------------------
176 * Server(A) has 16 cores, user configured it to 4 partitions so each
177 * partition has 4 cores, then actual number of service threads on each
179 * MDS_NTHRS_BASE(64) + cores(4) * MDS_NTHRS_FACTOR(8) = 96
181 * Total number of threads for the service is:
182 * 96 * partitions(4) = 384
185 * ---------------------------------------------------------------------
186 * Server(B) has 32 cores, user configured it to 4 partitions so each
187 * partition has 8 cores, then actual number of service threads on each
189 * MDS_NTHRS_BASE(64) + cores(8) * MDS_NTHRS_FACTOR(8) = 128
191 * Total number of threads for the service is:
192 * 128 * partitions(4) = 512
195 * ---------------------------------------------------------------------
196 * Server(B) has 96 cores, user configured it to 8 partitions so each
197 * partition has 12 cores, then actual number of service threads on each
199 * MDS_NTHRS_BASE(64) + cores(12) * MDS_NTHRS_FACTOR(8) = 160
201 * Total number of threads for the service is:
202 * 160 * partitions(8) = 1280
204 * However, it's above the soft limit MDS_NTHRS_MAX, so we choose this number
205 * as upper limit of threads number for each partition:
206 * MDS_NTHRS_MAX(1024) / partitions(8) = 128
209 * ---------------------------------------------------------------------
210 * Server(C) have a thousand of cores and user configured it to 32 partitions
211 * MDS_NTHRS_BASE(64) * 32 = 2048
213 * which is already above soft limit MDS_NTHRS_MAX(1024), but we still need
214 * to guarantee that each partition has at least MDS_NTHRS_BASE(64) threads
215 * to keep service healthy, so total number of threads will just be 2048.
217 * NB: we don't suggest to choose server with that many cores because backend
218 * filesystem itself, buffer cache, or underlying network stack might
219 * have some SMP scalability issues at that large scale.
221 * If user already has a fat machine with hundreds or thousands of cores,
222 * there are two choices for configuration:
223 * a) create CPU table from subset of all CPUs and run Lustre on
225 * b) bind service threads on a few partitions, see modparameters of
226 * MDS and OSS for details
228 * NB: these calculations (and examples below) are simplified to help
229 * understanding, the real implementation is a little more complex,
230 * please see ptlrpc_server_nthreads_check() for details.
235 * LDLM threads constants:
237 * Given 8 as factor and 24 as base threads number
240 * On 4-core machine we will have 24 + 8 * 4 = 56 threads.
243 * On 8-core machine with 2 partitions we will have 24 + 4 * 8 = 56
244 * threads for each partition and total threads number will be 112.
247 * On 64-core machine with 8 partitions we will need LDLM_NTHRS_BASE(24)
248 * threads for each partition to keep service healthy, so total threads
249 * number should be 24 * 8 = 192.
251 * So with these constants, threads number will be at the similar level
252 * of old versions, unless target machine has over a hundred cores
254 #define LDLM_THR_FACTOR 8
255 #define LDLM_NTHRS_INIT PTLRPC_NTHRS_INIT
256 #define LDLM_NTHRS_BASE 24
257 #define LDLM_NTHRS_MAX (num_online_cpus() == 1 ? 64 : 128)
259 #define LDLM_BL_THREADS LDLM_NTHRS_AUTO_INIT
260 #define LDLM_CLIENT_NBUFS 1
261 #define LDLM_SERVER_NBUFS 64
262 #define LDLM_BUFSIZE (8 * 1024)
263 #define LDLM_MAXREQSIZE (5 * 1024)
264 #define LDLM_MAXREPSIZE (1024)
267 * MDS threads constants:
269 * Please see examples in "Thread Constants", MDS threads number will be at
270 * the comparable level of old versions, unless the server has many cores.
272 #ifndef MDS_MAX_THREADS
273 #define MDS_MAX_THREADS 1024
274 #define MDS_MAX_OTHR_THREADS 256
276 #else /* MDS_MAX_THREADS */
277 #if MDS_MAX_THREADS < PTLRPC_NTHRS_INIT
278 #undef MDS_MAX_THREADS
279 #define MDS_MAX_THREADS PTLRPC_NTHRS_INIT
281 #define MDS_MAX_OTHR_THREADS max(PTLRPC_NTHRS_INIT, MDS_MAX_THREADS / 2)
284 /* default service */
285 #define MDS_THR_FACTOR 8
286 #define MDS_NTHRS_INIT PTLRPC_NTHRS_INIT
287 #define MDS_NTHRS_MAX MDS_MAX_THREADS
288 #define MDS_NTHRS_BASE min(64, MDS_NTHRS_MAX)
290 /* read-page service */
291 #define MDS_RDPG_THR_FACTOR 4
292 #define MDS_RDPG_NTHRS_INIT PTLRPC_NTHRS_INIT
293 #define MDS_RDPG_NTHRS_MAX MDS_MAX_OTHR_THREADS
294 #define MDS_RDPG_NTHRS_BASE min(48, MDS_RDPG_NTHRS_MAX)
296 /* these should be removed when we remove setattr service in the future */
297 #define MDS_SETA_THR_FACTOR 4
298 #define MDS_SETA_NTHRS_INIT PTLRPC_NTHRS_INIT
299 #define MDS_SETA_NTHRS_MAX MDS_MAX_OTHR_THREADS
300 #define MDS_SETA_NTHRS_BASE min(48, MDS_SETA_NTHRS_MAX)
302 /* non-affinity threads */
303 #define MDS_OTHR_NTHRS_INIT PTLRPC_NTHRS_INIT
304 #define MDS_OTHR_NTHRS_MAX MDS_MAX_OTHR_THREADS
309 * Assume file name length = FNAME_MAX = 256 (true for ext3).
310 * path name length = PATH_MAX = 4096
311 * LOV MD size max = EA_MAX = 24 * 2000
312 * (NB: 24 is size of lov_ost_data)
313 * LOV LOGCOOKIE size max = 32 * 2000
314 * (NB: 32 is size of llog_cookie)
315 * symlink: FNAME_MAX + PATH_MAX <- largest
316 * link: FNAME_MAX + PATH_MAX (mds_rec_link < mds_rec_create)
317 * rename: FNAME_MAX + FNAME_MAX
318 * open: FNAME_MAX + EA_MAX
320 * MDS_MAXREQSIZE ~= 4736 bytes =
321 * lustre_msg + ldlm_request + mdt_body + mds_rec_create + FNAME_MAX + PATH_MAX
322 * MDS_MAXREPSIZE ~= 8300 bytes = lustre_msg + llog_header
324 * Realistic size is about 512 bytes (20 character name + 128 char symlink),
325 * except in the open case where there are a large number of OSTs in a LOV.
327 #define MDS_MAXREQSIZE (5 * 1024) /* >= 4736 */
328 #define MDS_MAXREPSIZE (9 * 1024) /* >= 8300 */
331 * MDS incoming request with LOV EA
332 * 24 = sizeof(struct lov_ost_data), i.e: replay of opencreate
334 #define MDS_LOV_MAXREQSIZE max(MDS_MAXREQSIZE, \
335 362 + LOV_MAX_STRIPE_COUNT * 24)
337 * MDS outgoing reply with LOV EA
339 * NB: max reply size Lustre 2.4+ client can get from old MDS is:
340 * LOV_MAX_STRIPE_COUNT * (llog_cookie + lov_ost_data) + extra bytes
342 * but 2.4 or later MDS will never send reply with llog_cookie to any
343 * version client. This macro is defined for server side reply buffer size.
345 #define MDS_LOV_MAXREPSIZE MDS_LOV_MAXREQSIZE
348 * This is the size of a maximum REINT_SETXATTR request:
350 * lustre_msg 56 (32 + 4 x 5 + 4)
352 * mdt_rec_setxattr 136
354 * name 256 (XATTR_NAME_MAX)
355 * value 65536 (XATTR_SIZE_MAX)
357 #define MDS_EA_MAXREQSIZE 66288
360 * These are the maximum request and reply sizes (rounded up to 1 KB
361 * boundaries) for the "regular" MDS_REQUEST_PORTAL and MDS_REPLY_PORTAL.
363 #define MDS_REG_MAXREQSIZE (((max(MDS_EA_MAXREQSIZE, \
364 MDS_LOV_MAXREQSIZE) + 1023) >> 10) << 10)
365 #define MDS_REG_MAXREPSIZE MDS_REG_MAXREQSIZE
368 * The update request includes all of updates from the create, which might
369 * include linkea (4K maxim), together with other updates, we set it to 1000K:
370 * lustre_msg + ptlrpc_body + OUT_UPDATE_BUFFER_SIZE_MAX
372 #define OUT_MAXREQSIZE (1000 * 1024)
373 #define OUT_MAXREPSIZE MDS_MAXREPSIZE
375 /** MDS_BUFSIZE = max_reqsize (w/o LOV EA) + max sptlrpc payload size */
376 #define MDS_BUFSIZE max(MDS_MAXREQSIZE + SPTLRPC_MAX_PAYLOAD, \
380 * MDS_REG_BUFSIZE should at least be MDS_REG_MAXREQSIZE + SPTLRPC_MAX_PAYLOAD.
381 * However, we need to allocate a much larger buffer for it because LNet
382 * requires each MD(rqbd) has at least MDS_REQ_MAXREQSIZE bytes left to avoid
383 * dropping of maximum-sized incoming request. So if MDS_REG_BUFSIZE is only a
384 * little larger than MDS_REG_MAXREQSIZE, then it can only fit in one request
385 * even there are about MDS_REG_MAX_REQSIZE bytes left in a rqbd, and memory
386 * utilization is very low.
388 * In the meanwhile, size of rqbd can't be too large, because rqbd can't be
389 * reused until all requests fit in it have been processed and released,
390 * which means one long blocked request can prevent the rqbd be reused.
391 * Now we set request buffer size to 160 KB, so even each rqbd is unlinked
392 * from LNet with unused 65 KB, buffer utilization will be about 59%.
393 * Please check LU-2432 for details.
395 #define MDS_REG_BUFSIZE max(MDS_REG_MAXREQSIZE + SPTLRPC_MAX_PAYLOAD, \
399 * OUT_BUFSIZE = max_out_reqsize + max sptlrpc payload (~1K) which is
400 * about 10K, for the same reason as MDS_REG_BUFSIZE, we also give some
401 * extra bytes to each request buffer to improve buffer utilization rate.
403 #define OUT_BUFSIZE max(OUT_MAXREQSIZE + SPTLRPC_MAX_PAYLOAD, \
406 /** FLD_MAXREQSIZE == lustre_msg + __u32 padding + ptlrpc_body + opc */
407 #define FLD_MAXREQSIZE (160)
409 /** FLD_MAXREPSIZE == lustre_msg + ptlrpc_body */
410 #define FLD_MAXREPSIZE (152)
411 #define FLD_BUFSIZE (1 << 12)
414 * SEQ_MAXREQSIZE == lustre_msg + __u32 padding + ptlrpc_body + opc + lu_range +
416 #define SEQ_MAXREQSIZE (160)
418 /** SEQ_MAXREPSIZE == lustre_msg + ptlrpc_body + lu_range */
419 #define SEQ_MAXREPSIZE (152)
420 #define SEQ_BUFSIZE (1 << 12)
422 /** MGS threads must be >= 3, see bug 22458 comment #28 */
423 #define MGS_NTHRS_INIT (PTLRPC_NTHRS_INIT + 1)
424 #define MGS_NTHRS_MAX 32
427 #define MGS_BUFSIZE (8 * 1024)
428 #define MGS_MAXREQSIZE (7 * 1024)
429 #define MGS_MAXREPSIZE (9 * 1024)
432 * OSS threads constants:
434 * Given 8 as factor and 64 as base threads number
437 * On 8-core server configured to 2 partitions, we will have
438 * 64 + 8 * 4 = 96 threads for each partition, 192 total threads.
441 * On 32-core machine configured to 4 partitions, we will have
442 * 64 + 8 * 8 = 112 threads for each partition, so total threads number
443 * will be 112 * 4 = 448.
446 * On 64-core machine configured to 4 partitions, we will have
447 * 64 + 16 * 8 = 192 threads for each partition, so total threads number
448 * will be 192 * 4 = 768 which is above limit OSS_NTHRS_MAX(512), so we
449 * cut off the value to OSS_NTHRS_MAX(512) / 4 which is 128 threads
450 * for each partition.
452 * So we can see that with these constants, threads number wil be at the
453 * similar level of old versions, unless the server has many cores.
455 /* depress threads factor for VM with small memory size */
456 #define OSS_THR_FACTOR min_t(int, 8, \
457 NUM_CACHEPAGES >> (28 - PAGE_SHIFT))
458 #define OSS_NTHRS_INIT (PTLRPC_NTHRS_INIT + 1)
459 #define OSS_NTHRS_BASE 64
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 struct lnet_process_id 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;
608 unsigned int set_allow_intr:1;
612 * Description of a single ptrlrpc_set callback
614 struct ptlrpc_set_cbdata {
615 /** List linkage item */
616 struct list_head psc_item;
617 /** Pointer to interpreting function */
618 set_interpreter_func psc_interpret;
619 /** Opaq argument to pass to the callback */
623 struct ptlrpc_bulk_desc;
624 struct ptlrpc_service_part;
625 struct ptlrpc_service;
628 * ptlrpc callback & work item stuff
630 struct ptlrpc_cb_id {
631 void (*cbid_fn)(struct lnet_event *ev); /* specific callback fn */
632 void *cbid_arg; /* additional arg */
635 /** Maximum number of locks to fit into reply state */
636 #define RS_MAX_LOCKS 8
640 * Structure to define reply state on the server
641 * Reply state holds various reply message information. Also for "difficult"
642 * replies (rep-ack case) we store the state after sending reply and wait
643 * for the client to acknowledge the reception. In these cases locks could be
644 * added to the state for replay/failover consistency guarantees.
646 struct ptlrpc_reply_state {
647 /** Callback description */
648 struct ptlrpc_cb_id rs_cb_id;
649 /** Linkage for list of all reply states in a system */
650 struct list_head rs_list;
651 /** Linkage for list of all reply states on same export */
652 struct list_head rs_exp_list;
653 /** Linkage for list of all reply states for same obd */
654 struct list_head rs_obd_list;
656 struct list_head rs_debug_list;
658 /** A spinlock to protect the reply state flags */
660 /** Reply state flags */
661 unsigned long rs_difficult:1; /* ACK/commit stuff */
662 unsigned long rs_no_ack:1; /* no ACK, even for
663 difficult requests */
664 unsigned long rs_scheduled:1; /* being handled? */
665 unsigned long rs_scheduled_ever:1;/* any schedule attempts? */
666 unsigned long rs_handled:1; /* been handled yet? */
667 unsigned long rs_on_net:1; /* reply_out_callback pending? */
668 unsigned long rs_prealloc:1; /* rs from prealloc list */
669 unsigned long rs_committed:1;/* the transaction was committed
670 and the rs was dispatched
671 by ptlrpc_commit_replies */
672 unsigned long rs_convert_lock:1; /* need to convert saved
673 * locks to COS mode */
674 atomic_t rs_refcount; /* number of users */
675 /** Number of locks awaiting client ACK */
678 /** Size of the state */
682 /** Transaction number */
686 struct obd_export *rs_export;
687 struct ptlrpc_service_part *rs_svcpt;
688 /** Lnet metadata handle for the reply */
689 struct lnet_handle_md rs_md_h;
691 /** Context for the sevice thread */
692 struct ptlrpc_svc_ctx *rs_svc_ctx;
693 /** Reply buffer (actually sent to the client), encoded if needed */
694 struct lustre_msg *rs_repbuf; /* wrapper */
695 /** Size of the reply buffer */
696 int rs_repbuf_len; /* wrapper buf length */
697 /** Size of the reply message */
698 int rs_repdata_len; /* wrapper msg length */
700 * Actual reply message. Its content is encrupted (if needed) to
701 * produce reply buffer for actual sending. In simple case
702 * of no network encryption we jus set \a rs_repbuf to \a rs_msg
704 struct lustre_msg *rs_msg; /* reply message */
706 /** Handles of locks awaiting client reply ACK */
707 struct lustre_handle rs_locks[RS_MAX_LOCKS];
708 /** Lock modes of locks in \a rs_locks */
709 enum ldlm_mode rs_modes[RS_MAX_LOCKS];
712 struct ptlrpc_thread;
716 RQ_PHASE_NEW = 0xebc0de00,
717 RQ_PHASE_RPC = 0xebc0de01,
718 RQ_PHASE_BULK = 0xebc0de02,
719 RQ_PHASE_INTERPRET = 0xebc0de03,
720 RQ_PHASE_COMPLETE = 0xebc0de04,
721 RQ_PHASE_UNREG_RPC = 0xebc0de05,
722 RQ_PHASE_UNREG_BULK = 0xebc0de06,
723 RQ_PHASE_UNDEFINED = 0xebc0de07
726 /** Type of request interpreter call-back */
727 typedef int (*ptlrpc_interpterer_t)(const struct lu_env *env,
728 struct ptlrpc_request *req,
730 /** Type of request resend call-back */
731 typedef void (*ptlrpc_resend_cb_t)(struct ptlrpc_request *req,
735 * Definition of request pool structure.
736 * The pool is used to store empty preallocated requests for the case
737 * when we would actually need to send something without performing
738 * any allocations (to avoid e.g. OOM).
740 struct ptlrpc_request_pool {
741 /** Locks the list */
743 /** list of ptlrpc_request structs */
744 struct list_head prp_req_list;
745 /** Maximum message size that would fit into a rquest from this pool */
747 /** Function to allocate more requests for this pool */
748 int (*prp_populate)(struct ptlrpc_request_pool *, int);
756 #include <lustre_nrs.h>
759 * Basic request prioritization operations structure.
760 * The whole idea is centered around locks and RPCs that might affect locks.
761 * When a lock is contended we try to give priority to RPCs that might lead
762 * to fastest release of that lock.
763 * Currently only implemented for OSTs only in a way that makes all
764 * IO and truncate RPCs that are coming from a locked region where a lock is
765 * contended a priority over other requests.
767 struct ptlrpc_hpreq_ops {
769 * Check if the lock handle of the given lock is the same as
770 * taken from the request.
772 int (*hpreq_lock_match)(struct ptlrpc_request *, struct ldlm_lock *);
774 * Check if the request is a high priority one.
776 int (*hpreq_check)(struct ptlrpc_request *);
778 * Called after the request has been handled.
780 void (*hpreq_fini)(struct ptlrpc_request *);
783 struct ptlrpc_cli_req {
784 /** For bulk requests on client only: bulk descriptor */
785 struct ptlrpc_bulk_desc *cr_bulk;
786 /** optional time limit for send attempts */
787 time64_t cr_delay_limit;
788 /** time request was first queued */
789 time64_t cr_queued_time;
790 /** request sent in nanoseconds */
792 /** time for request really sent out */
793 time64_t cr_sent_out;
794 /** when req reply unlink must finish. */
795 time64_t cr_reply_deadline;
796 /** when req bulk unlink must finish. */
797 time64_t cr_bulk_deadline;
798 /** when req unlink must finish. */
799 time64_t cr_req_deadline;
800 /** Portal to which this request would be sent */
802 /** Portal where to wait for reply and where reply would be sent */
804 /** request resending number */
805 unsigned int cr_resend_nr;
806 /** What was import generation when this request was sent */
808 enum lustre_imp_state cr_send_state;
809 /** Per-request waitq introduced by bug 21938 for recovery waiting */
810 wait_queue_head_t cr_set_waitq;
811 /** Link item for request set lists */
812 struct list_head cr_set_chain;
813 /** link to waited ctx */
814 struct list_head cr_ctx_chain;
816 /** client's half ctx */
817 struct ptlrpc_cli_ctx *cr_cli_ctx;
818 /** Link back to the request set */
819 struct ptlrpc_request_set *cr_set;
820 /** outgoing request MD handle */
821 struct lnet_handle_md cr_req_md_h;
822 /** request-out callback parameter */
823 struct ptlrpc_cb_id cr_req_cbid;
824 /** incoming reply MD handle */
825 struct lnet_handle_md cr_reply_md_h;
826 wait_queue_head_t cr_reply_waitq;
827 /** reply callback parameter */
828 struct ptlrpc_cb_id cr_reply_cbid;
829 /** Async completion handler, called when reply is received */
830 ptlrpc_interpterer_t cr_reply_interp;
831 /** Resend handler, called when request is resend to update RPC data */
832 ptlrpc_resend_cb_t cr_resend_cb;
833 /** Async completion context */
834 union ptlrpc_async_args cr_async_args;
835 /** Opaq data for replay and commit callbacks. */
837 /** Link to the imp->imp_unreplied_list */
838 struct list_head cr_unreplied_list;
840 * Commit callback, called when request is committed and about to be
843 void (*cr_commit_cb)(struct ptlrpc_request *);
844 /** Replay callback, called after request is replayed at recovery */
845 void (*cr_replay_cb)(struct ptlrpc_request *);
848 /** client request member alias */
849 /* NB: these alias should NOT be used by any new code, instead they should
850 * be removed step by step to avoid potential abuse */
851 #define rq_bulk rq_cli.cr_bulk
852 #define rq_delay_limit rq_cli.cr_delay_limit
853 #define rq_queued_time rq_cli.cr_queued_time
854 #define rq_sent_ns rq_cli.cr_sent_ns
855 #define rq_real_sent rq_cli.cr_sent_out
856 #define rq_reply_deadline rq_cli.cr_reply_deadline
857 #define rq_bulk_deadline rq_cli.cr_bulk_deadline
858 #define rq_req_deadline rq_cli.cr_req_deadline
859 #define rq_nr_resend rq_cli.cr_resend_nr
860 #define rq_request_portal rq_cli.cr_req_ptl
861 #define rq_reply_portal rq_cli.cr_rep_ptl
862 #define rq_import_generation rq_cli.cr_imp_gen
863 #define rq_send_state rq_cli.cr_send_state
864 #define rq_set_chain rq_cli.cr_set_chain
865 #define rq_ctx_chain rq_cli.cr_ctx_chain
866 #define rq_set rq_cli.cr_set
867 #define rq_set_waitq rq_cli.cr_set_waitq
868 #define rq_cli_ctx rq_cli.cr_cli_ctx
869 #define rq_req_md_h rq_cli.cr_req_md_h
870 #define rq_req_cbid rq_cli.cr_req_cbid
871 #define rq_reply_md_h rq_cli.cr_reply_md_h
872 #define rq_reply_waitq rq_cli.cr_reply_waitq
873 #define rq_reply_cbid rq_cli.cr_reply_cbid
874 #define rq_interpret_reply rq_cli.cr_reply_interp
875 #define rq_resend_cb rq_cli.cr_resend_cb
876 #define rq_async_args rq_cli.cr_async_args
877 #define rq_cb_data rq_cli.cr_cb_data
878 #define rq_unreplied_list rq_cli.cr_unreplied_list
879 #define rq_commit_cb rq_cli.cr_commit_cb
880 #define rq_replay_cb rq_cli.cr_replay_cb
882 struct ptlrpc_srv_req {
883 /** initial thread servicing this request */
884 struct ptlrpc_thread *sr_svc_thread;
886 * Server side list of incoming unserved requests sorted by arrival
887 * time. Traversed from time to time to notice about to expire
888 * requests and sent back "early replies" to clients to let them
889 * know server is alive and well, just very busy to service their
892 struct list_head sr_timed_list;
893 /** server-side per-export list */
894 struct list_head sr_exp_list;
895 /** server-side history, used for debuging purposes. */
896 struct list_head sr_hist_list;
897 /** history sequence # */
899 /** the index of service's srv_at_array into which request is linked */
903 /** authed uid mapped to */
904 uid_t sr_auth_mapped_uid;
905 /** RPC is generated from what part of Lustre */
906 enum lustre_sec_part sr_sp_from;
907 /** request session context */
908 struct lu_context sr_ses;
912 /** stub for NRS request */
913 struct ptlrpc_nrs_request sr_nrq;
915 /** request arrival time */
916 struct timespec64 sr_arrival_time;
917 /** server's half ctx */
918 struct ptlrpc_svc_ctx *sr_svc_ctx;
919 /** (server side), pointed directly into req buffer */
920 struct ptlrpc_user_desc *sr_user_desc;
921 /** separated reply state, may be vmalloc'd */
922 struct ptlrpc_reply_state *sr_reply_state;
923 /** server-side hp handlers */
924 struct ptlrpc_hpreq_ops *sr_ops;
925 /** incoming request buffer */
926 struct ptlrpc_request_buffer_desc *sr_rqbd;
929 /** server request member alias */
930 /* NB: these alias should NOT be used by any new code, instead they should
931 * be removed step by step to avoid potential abuse */
932 #define rq_svc_thread rq_srv.sr_svc_thread
933 #define rq_timed_list rq_srv.sr_timed_list
934 #define rq_exp_list rq_srv.sr_exp_list
935 #define rq_history_list rq_srv.sr_hist_list
936 #define rq_history_seq rq_srv.sr_hist_seq
937 #define rq_at_index rq_srv.sr_at_index
938 #define rq_auth_uid rq_srv.sr_auth_uid
939 #define rq_auth_mapped_uid rq_srv.sr_auth_mapped_uid
940 #define rq_sp_from rq_srv.sr_sp_from
941 #define rq_session rq_srv.sr_ses
942 #define rq_nrq rq_srv.sr_nrq
943 #define rq_arrival_time rq_srv.sr_arrival_time
944 #define rq_reply_state rq_srv.sr_reply_state
945 #define rq_svc_ctx rq_srv.sr_svc_ctx
946 #define rq_user_desc rq_srv.sr_user_desc
947 #define rq_ops rq_srv.sr_ops
948 #define rq_rqbd rq_srv.sr_rqbd
951 * Represents remote procedure call.
953 * This is a staple structure used by everybody wanting to send a request
956 struct ptlrpc_request {
957 /* Request type: one of PTL_RPC_MSG_* */
959 /** Result of request processing */
962 * Linkage item through which this request is included into
963 * sending/delayed lists on client and into rqbd list on server
965 struct list_head rq_list;
966 /** Lock to protect request flags and some other important bits, like
970 spinlock_t rq_early_free_lock;
971 /** client-side flags are serialized by rq_lock @{ */
972 unsigned int rq_intr:1, rq_replied:1, rq_err:1,
973 rq_timedout:1, rq_resend:1, rq_restart:1,
975 * when ->rq_replay is set, request is kept by the client even
976 * after server commits corresponding transaction. This is
977 * used for operations that require sequence of multiple
978 * requests to be replayed. The only example currently is file
979 * open/close. When last request in such a sequence is
980 * committed, ->rq_replay is cleared on all requests in the
984 rq_no_resend:1, rq_waiting:1, rq_receiving_reply:1,
985 rq_no_delay:1, rq_net_err:1, rq_wait_ctx:1,
987 rq_req_unlinked:1, /* unlinked request buffer from lnet */
988 rq_reply_unlinked:1, /* unlinked reply buffer from lnet */
989 rq_memalloc:1, /* req originated from "kswapd" */
991 rq_reply_truncated:1,
992 /** whether the "rq_set" is a valid one */
995 /** do not resend request on -EINPROGRESS */
996 rq_no_retry_einprogress:1,
997 /* allow the req to be sent if the import is in recovery
1000 /* bulk request, sent to server, but uncommitted */
1002 rq_early_free_repbuf:1, /* free reply buffer in advance */
1006 /** server-side flags @{ */
1008 rq_hp:1, /**< high priority RPC */
1009 rq_at_linked:1, /**< link into service's srv_at_array */
1010 rq_packed_final:1; /**< packed final reply */
1013 /** one of RQ_PHASE_* */
1014 enum rq_phase rq_phase;
1015 /** one of RQ_PHASE_* to be used next */
1016 enum rq_phase rq_next_phase;
1018 * client-side refcount for SENT race, server-side refcounf
1019 * for multiple replies
1021 atomic_t rq_refcount;
1024 * !rq_truncate : # reply bytes actually received,
1025 * rq_truncate : required repbuf_len for resend
1027 int rq_nob_received;
1028 /** Request length */
1032 /** Pool if request is from preallocated list */
1033 struct ptlrpc_request_pool *rq_pool;
1034 /** Request message - what client sent */
1035 struct lustre_msg *rq_reqmsg;
1036 /** Reply message - server response */
1037 struct lustre_msg *rq_repmsg;
1038 /** Transaction number */
1042 /** bulk match bits */
1045 * List item to for replay list. Not yet committed requests get linked
1047 * Also see \a rq_replay comment above.
1048 * It's also link chain on obd_export::exp_req_replay_queue
1050 struct list_head rq_replay_list;
1051 /** non-shared members for client & server request*/
1053 struct ptlrpc_cli_req rq_cli;
1054 struct ptlrpc_srv_req rq_srv;
1057 * security and encryption data
1059 /** description of flavors for client & server */
1060 struct sptlrpc_flavor rq_flvr;
1062 /* client/server security flags */
1064 rq_ctx_init:1, /* context initiation */
1065 rq_ctx_fini:1, /* context destroy */
1066 rq_bulk_read:1, /* request bulk read */
1067 rq_bulk_write:1, /* request bulk write */
1068 /* server authentication flags */
1069 rq_auth_gss:1, /* authenticated by gss */
1070 rq_auth_usr_root:1, /* authed as root */
1071 rq_auth_usr_mdt:1, /* authed as mdt */
1072 rq_auth_usr_ost:1, /* authed as ost */
1073 /* security tfm flags */
1076 /* doesn't expect reply FIXME */
1078 rq_pill_init:1, /* pill initialized */
1079 rq_srv_req:1; /* server request */
1082 /** various buffer pointers */
1083 struct lustre_msg *rq_reqbuf; /**< req wrapper, vmalloc*/
1084 char *rq_repbuf; /**< rep buffer, vmalloc */
1085 struct lustre_msg *rq_repdata; /**< rep wrapper msg */
1086 /** only in priv mode */
1087 struct lustre_msg *rq_clrbuf;
1088 int rq_reqbuf_len; /* req wrapper buf len */
1089 int rq_reqdata_len; /* req wrapper msg len */
1090 int rq_repbuf_len; /* rep buffer len */
1091 int rq_repdata_len; /* rep wrapper msg len */
1092 int rq_clrbuf_len; /* only in priv mode */
1093 int rq_clrdata_len; /* only in priv mode */
1095 /** early replies go to offset 0, regular replies go after that */
1096 unsigned int rq_reply_off;
1099 /** Fields that help to see if request and reply were swabbed or not */
1100 __u32 rq_req_swab_mask;
1101 __u32 rq_rep_swab_mask;
1103 /** how many early replies (for stats) */
1105 /** Server-side, export on which request was received */
1106 struct obd_export *rq_export;
1107 /** import where request is being sent */
1108 struct obd_import *rq_import;
1111 /** Peer description (the other side) */
1112 struct lnet_process_id rq_peer;
1113 /** Descriptor for the NID from which the peer sent the request. */
1114 struct lnet_process_id rq_source;
1116 * service time estimate (secs)
1117 * If the request is not served by this time, it is marked as timed out.
1118 * Do not change to time64_t since this is transmitted over the wire.
1122 * when request/reply sent (secs), or time when request should be sent
1125 /** when request must finish. */
1126 time64_t rq_deadline;
1127 /** request format description */
1128 struct req_capsule rq_pill;
1132 * Call completion handler for rpc if any, return it's status or original
1133 * rc if there was no handler defined for this request.
1135 static inline int ptlrpc_req_interpret(const struct lu_env *env,
1136 struct ptlrpc_request *req, int rc)
1138 if (req->rq_interpret_reply != NULL) {
1139 req->rq_status = req->rq_interpret_reply(env, req,
1140 &req->rq_async_args,
1142 return req->rq_status;
1150 int ptlrpc_nrs_policy_register(struct ptlrpc_nrs_pol_conf *conf);
1151 int ptlrpc_nrs_policy_unregister(struct ptlrpc_nrs_pol_conf *conf);
1152 void ptlrpc_nrs_req_hp_move(struct ptlrpc_request *req);
1153 void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
1154 struct ptlrpc_nrs_pol_info *info);
1157 * Can the request be moved from the regular NRS head to the high-priority NRS
1158 * head (of the same PTLRPC service partition), if any?
1160 * For a reliable result, this should be checked under svcpt->scp_req lock.
1162 static inline bool ptlrpc_nrs_req_can_move(struct ptlrpc_request *req)
1164 struct ptlrpc_nrs_request *nrq = &req->rq_nrq;
1167 * LU-898: Check ptlrpc_nrs_request::nr_enqueued to make sure the
1168 * request has been enqueued first, and ptlrpc_nrs_request::nr_started
1169 * to make sure it has not been scheduled yet (analogous to previous
1170 * (non-NRS) checking of !list_empty(&ptlrpc_request::rq_list).
1172 return nrq->nr_enqueued && !nrq->nr_started && !req->rq_hp;
1177 * Returns 1 if request buffer at offset \a index was already swabbed
1179 static inline int lustre_req_swabbed(struct ptlrpc_request *req, size_t index)
1181 LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
1182 return req->rq_req_swab_mask & (1 << index);
1186 * Returns 1 if request reply buffer at offset \a index was already swabbed
1188 static inline int lustre_rep_swabbed(struct ptlrpc_request *req, size_t index)
1190 LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
1191 return req->rq_rep_swab_mask & (1 << index);
1195 * Returns 1 if request needs to be swabbed into local cpu byteorder
1197 static inline int ptlrpc_req_need_swab(struct ptlrpc_request *req)
1199 return lustre_req_swabbed(req, MSG_PTLRPC_HEADER_OFF);
1203 * Returns 1 if request reply needs to be swabbed into local cpu byteorder
1205 static inline int ptlrpc_rep_need_swab(struct ptlrpc_request *req)
1207 return lustre_rep_swabbed(req, MSG_PTLRPC_HEADER_OFF);
1211 * Mark request buffer at offset \a index that it was already swabbed
1213 static inline void lustre_set_req_swabbed(struct ptlrpc_request *req,
1216 LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
1217 LASSERT((req->rq_req_swab_mask & (1 << index)) == 0);
1218 req->rq_req_swab_mask |= 1 << index;
1222 * Mark request reply buffer at offset \a index that it was already swabbed
1224 static inline void lustre_set_rep_swabbed(struct ptlrpc_request *req,
1227 LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
1228 LASSERT((req->rq_rep_swab_mask & (1 << index)) == 0);
1229 req->rq_rep_swab_mask |= 1 << index;
1233 * Convert numerical request phase value \a phase into text string description
1235 static inline const char *
1236 ptlrpc_phase2str(enum rq_phase phase)
1245 case RQ_PHASE_INTERPRET:
1247 case RQ_PHASE_COMPLETE:
1249 case RQ_PHASE_UNREG_RPC:
1251 case RQ_PHASE_UNREG_BULK:
1259 * Convert numerical request phase of the request \a req into text stringi
1262 static inline const char *
1263 ptlrpc_rqphase2str(struct ptlrpc_request *req)
1265 return ptlrpc_phase2str(req->rq_phase);
1269 * Debugging functions and helpers to print request structure into debug log
1272 /* Spare the preprocessor, spoil the bugs. */
1273 #define FLAG(field, str) (field ? str : "")
1275 /** Convert bit flags into a string */
1276 #define DEBUG_REQ_FLAGS(req) \
1277 ptlrpc_rqphase2str(req), \
1278 FLAG(req->rq_intr, "I"), FLAG(req->rq_replied, "R"), \
1279 FLAG(req->rq_err, "E"), FLAG(req->rq_net_err, "e"), \
1280 FLAG(req->rq_timedout, "X") /* eXpired */, FLAG(req->rq_resend, "S"), \
1281 FLAG(req->rq_restart, "T"), FLAG(req->rq_replay, "P"), \
1282 FLAG(req->rq_no_resend, "N"), \
1283 FLAG(req->rq_waiting, "W"), \
1284 FLAG(req->rq_wait_ctx, "C"), FLAG(req->rq_hp, "H"), \
1285 FLAG(req->rq_committed, "M")
1287 #define REQ_FLAGS_FMT "%s:%s%s%s%s%s%s%s%s%s%s%s%s%s"
1289 void _debug_req(struct ptlrpc_request *req,
1290 struct libcfs_debug_msg_data *data, const char *fmt, ...)
1291 __attribute__ ((format (printf, 3, 4)));
1294 * Helper that decides if we need to print request accordig to current debug
1297 #define debug_req(msgdata, mask, cdls, req, fmt, a...) \
1299 CFS_CHECK_STACK(msgdata, mask, cdls); \
1301 if (((mask) & D_CANTMASK) != 0 || \
1302 ((libcfs_debug & (mask)) != 0 && \
1303 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) \
1304 _debug_req((req), msgdata, fmt, ##a); \
1308 * This is the debug print function you need to use to print request sturucture
1309 * content into lustre debug log.
1310 * for most callers (level is a constant) this is resolved at compile time */
1311 #define DEBUG_REQ(level, req, fmt, args...) \
1313 if ((level) & (D_ERROR | D_WARNING)) { \
1314 static struct cfs_debug_limit_state cdls; \
1315 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, level, &cdls); \
1316 debug_req(&msgdata, level, &cdls, req, "@@@ "fmt" ", ## args);\
1318 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, level, NULL); \
1319 debug_req(&msgdata, level, NULL, req, "@@@ "fmt" ", ## args); \
1325 * Structure that defines a single page of a bulk transfer
1327 struct ptlrpc_bulk_page {
1328 /** Linkage to list of pages in a bulk */
1329 struct list_head bp_link;
1331 * Number of bytes in a page to transfer starting from \a bp_pageoffset
1334 /** offset within a page */
1336 /** The page itself */
1337 struct page *bp_page;
1340 enum ptlrpc_bulk_op_type {
1341 PTLRPC_BULK_OP_ACTIVE = 0x00000001,
1342 PTLRPC_BULK_OP_PASSIVE = 0x00000002,
1343 PTLRPC_BULK_OP_PUT = 0x00000004,
1344 PTLRPC_BULK_OP_GET = 0x00000008,
1345 PTLRPC_BULK_BUF_KVEC = 0x00000010,
1346 PTLRPC_BULK_BUF_KIOV = 0x00000020,
1347 PTLRPC_BULK_GET_SOURCE = PTLRPC_BULK_OP_PASSIVE | PTLRPC_BULK_OP_GET,
1348 PTLRPC_BULK_PUT_SINK = PTLRPC_BULK_OP_PASSIVE | PTLRPC_BULK_OP_PUT,
1349 PTLRPC_BULK_GET_SINK = PTLRPC_BULK_OP_ACTIVE | PTLRPC_BULK_OP_GET,
1350 PTLRPC_BULK_PUT_SOURCE = PTLRPC_BULK_OP_ACTIVE | PTLRPC_BULK_OP_PUT,
1353 static inline bool ptlrpc_is_bulk_op_get(enum ptlrpc_bulk_op_type type)
1355 return (type & PTLRPC_BULK_OP_GET) == PTLRPC_BULK_OP_GET;
1358 static inline bool ptlrpc_is_bulk_get_source(enum ptlrpc_bulk_op_type type)
1360 return (type & PTLRPC_BULK_GET_SOURCE) == PTLRPC_BULK_GET_SOURCE;
1363 static inline bool ptlrpc_is_bulk_put_sink(enum ptlrpc_bulk_op_type type)
1365 return (type & PTLRPC_BULK_PUT_SINK) == PTLRPC_BULK_PUT_SINK;
1368 static inline bool ptlrpc_is_bulk_get_sink(enum ptlrpc_bulk_op_type type)
1370 return (type & PTLRPC_BULK_GET_SINK) == PTLRPC_BULK_GET_SINK;
1373 static inline bool ptlrpc_is_bulk_put_source(enum ptlrpc_bulk_op_type type)
1375 return (type & PTLRPC_BULK_PUT_SOURCE) == PTLRPC_BULK_PUT_SOURCE;
1378 static inline bool ptlrpc_is_bulk_desc_kvec(enum ptlrpc_bulk_op_type type)
1380 return ((type & PTLRPC_BULK_BUF_KVEC) | (type & PTLRPC_BULK_BUF_KIOV))
1381 == PTLRPC_BULK_BUF_KVEC;
1384 static inline bool ptlrpc_is_bulk_desc_kiov(enum ptlrpc_bulk_op_type type)
1386 return ((type & PTLRPC_BULK_BUF_KVEC) | (type & PTLRPC_BULK_BUF_KIOV))
1387 == PTLRPC_BULK_BUF_KIOV;
1390 static inline bool ptlrpc_is_bulk_op_active(enum ptlrpc_bulk_op_type type)
1392 return ((type & PTLRPC_BULK_OP_ACTIVE) |
1393 (type & PTLRPC_BULK_OP_PASSIVE))
1394 == PTLRPC_BULK_OP_ACTIVE;
1397 static inline bool ptlrpc_is_bulk_op_passive(enum ptlrpc_bulk_op_type type)
1399 return ((type & PTLRPC_BULK_OP_ACTIVE) |
1400 (type & PTLRPC_BULK_OP_PASSIVE))
1401 == PTLRPC_BULK_OP_PASSIVE;
1404 struct ptlrpc_bulk_frag_ops {
1406 * Add a page \a page to the bulk descriptor \a desc
1407 * Data to transfer in the page starts at offset \a pageoffset and
1408 * amount of data to transfer from the page is \a len
1410 void (*add_kiov_frag)(struct ptlrpc_bulk_desc *desc,
1411 struct page *page, int pageoffset, int len);
1414 * Add a \a fragment to the bulk descriptor \a desc.
1415 * Data to transfer in the fragment is pointed to by \a frag
1416 * The size of the fragment is \a len
1418 int (*add_iov_frag)(struct ptlrpc_bulk_desc *desc, void *frag, int len);
1421 * Uninitialize and free bulk descriptor \a desc.
1422 * Works on bulk descriptors both from server and client side.
1424 void (*release_frags)(struct ptlrpc_bulk_desc *desc);
1427 extern const struct ptlrpc_bulk_frag_ops ptlrpc_bulk_kiov_pin_ops;
1428 extern const struct ptlrpc_bulk_frag_ops ptlrpc_bulk_kiov_nopin_ops;
1429 extern const struct ptlrpc_bulk_frag_ops ptlrpc_bulk_kvec_ops;
1432 * Definition of bulk descriptor.
1433 * Bulks are special "Two phase" RPCs where initial request message
1434 * is sent first and it is followed bt a transfer (o receiving) of a large
1435 * amount of data to be settled into pages referenced from the bulk descriptors.
1436 * Bulks transfers (the actual data following the small requests) are done
1437 * on separate LNet portals.
1438 * In lustre we use bulk transfers for READ and WRITE transfers from/to OSTs.
1439 * Another user is readpage for MDT.
1441 struct ptlrpc_bulk_desc {
1442 /** completed with failure */
1443 unsigned long bd_failure:1;
1445 unsigned long bd_registered:1;
1446 /** For serialization with callback */
1448 /** Import generation when request for this bulk was sent */
1449 int bd_import_generation;
1450 /** {put,get}{source,sink}{kvec,kiov} */
1451 enum ptlrpc_bulk_op_type bd_type;
1452 /** LNet portal for this bulk */
1454 /** Server side - export this bulk created for */
1455 struct obd_export *bd_export;
1456 /** Client side - import this bulk was sent on */
1457 struct obd_import *bd_import;
1458 /** Back pointer to the request */
1459 struct ptlrpc_request *bd_req;
1460 struct ptlrpc_bulk_frag_ops *bd_frag_ops;
1461 wait_queue_head_t bd_waitq; /* server side only WQ */
1462 int bd_iov_count; /* # entries in bd_iov */
1463 int bd_max_iov; /* allocated size of bd_iov */
1464 int bd_nob; /* # bytes covered */
1465 int bd_nob_transferred; /* # bytes GOT/PUT */
1467 __u64 bd_last_mbits;
1469 struct ptlrpc_cb_id bd_cbid; /* network callback info */
1470 lnet_nid_t bd_sender; /* stash event::sender */
1471 int bd_md_count; /* # valid entries in bd_mds */
1472 int bd_md_max_brw; /* max entries in bd_mds */
1473 /** array of associated MDs */
1474 struct lnet_handle_md bd_mds[PTLRPC_BULK_OPS_COUNT];
1479 * encrypt iov, size is either 0 or bd_iov_count.
1481 lnet_kiov_t *bd_enc_vec;
1482 lnet_kiov_t *bd_vec;
1486 struct kvec *bd_enc_kvec;
1487 struct kvec *bd_kvec;
1493 #define GET_KIOV(desc) ((desc)->bd_u.bd_kiov.bd_vec)
1494 #define BD_GET_KIOV(desc, i) ((desc)->bd_u.bd_kiov.bd_vec[i])
1495 #define GET_ENC_KIOV(desc) ((desc)->bd_u.bd_kiov.bd_enc_vec)
1496 #define BD_GET_ENC_KIOV(desc, i) ((desc)->bd_u.bd_kiov.bd_enc_vec[i])
1497 #define GET_KVEC(desc) ((desc)->bd_u.bd_kvec.bd_kvec)
1498 #define BD_GET_KVEC(desc, i) ((desc)->bd_u.bd_kvec.bd_kvec[i])
1499 #define GET_ENC_KVEC(desc) ((desc)->bd_u.bd_kvec.bd_enc_kvec)
1500 #define BD_GET_ENC_KVEC(desc, i) ((desc)->bd_u.bd_kvec.bd_enc_kvec[i])
1504 SVC_STOPPED = 1 << 0,
1505 SVC_STOPPING = 1 << 1,
1506 SVC_STARTING = 1 << 2,
1507 SVC_RUNNING = 1 << 3,
1509 SVC_SIGNAL = 1 << 5,
1512 #define PTLRPC_THR_NAME_LEN 32
1514 * Definition of server service thread structure
1516 struct ptlrpc_thread {
1518 * List of active threads in svc->srv_threads
1520 struct list_head t_link;
1522 * thread-private data (preallocated vmalloc'd memory)
1527 * service thread index, from ptlrpc_start_threads
1531 * service thread pid
1535 * put watchdog in the structure per thread b=14840
1537 struct lc_watchdog *t_watchdog;
1539 * the svc this thread belonged to b=18582
1541 struct ptlrpc_service_part *t_svcpt;
1542 wait_queue_head_t t_ctl_waitq;
1543 struct lu_env *t_env;
1544 char t_name[PTLRPC_THR_NAME_LEN];
1547 static inline int thread_is_init(struct ptlrpc_thread *thread)
1549 return thread->t_flags == 0;
1552 static inline int thread_is_stopped(struct ptlrpc_thread *thread)
1554 return !!(thread->t_flags & SVC_STOPPED);
1557 static inline int thread_is_stopping(struct ptlrpc_thread *thread)
1559 return !!(thread->t_flags & SVC_STOPPING);
1562 static inline int thread_is_starting(struct ptlrpc_thread *thread)
1564 return !!(thread->t_flags & SVC_STARTING);
1567 static inline int thread_is_running(struct ptlrpc_thread *thread)
1569 return !!(thread->t_flags & SVC_RUNNING);
1572 static inline int thread_is_event(struct ptlrpc_thread *thread)
1574 return !!(thread->t_flags & SVC_EVENT);
1577 static inline int thread_is_signal(struct ptlrpc_thread *thread)
1579 return !!(thread->t_flags & SVC_SIGNAL);
1582 static inline void thread_clear_flags(struct ptlrpc_thread *thread, __u32 flags)
1584 thread->t_flags &= ~flags;
1587 static inline void thread_set_flags(struct ptlrpc_thread *thread, __u32 flags)
1589 thread->t_flags = flags;
1592 static inline void thread_add_flags(struct ptlrpc_thread *thread, __u32 flags)
1594 thread->t_flags |= flags;
1597 static inline int thread_test_and_clear_flags(struct ptlrpc_thread *thread,
1600 if (thread->t_flags & flags) {
1601 thread->t_flags &= ~flags;
1608 * Request buffer descriptor structure.
1609 * This is a structure that contains one posted request buffer for service.
1610 * Once data land into a buffer, event callback creates actual request and
1611 * notifies wakes one of the service threads to process new incoming request.
1612 * More than one request can fit into the buffer.
1614 struct ptlrpc_request_buffer_desc {
1615 /** Link item for rqbds on a service */
1616 struct list_head rqbd_list;
1617 /** History of requests for this buffer */
1618 struct list_head rqbd_reqs;
1619 /** Back pointer to service for which this buffer is registered */
1620 struct ptlrpc_service_part *rqbd_svcpt;
1621 /** LNet descriptor */
1622 struct lnet_handle_md rqbd_md_h;
1624 /** The buffer itself */
1626 struct ptlrpc_cb_id rqbd_cbid;
1628 * This "embedded" request structure is only used for the
1629 * last request to fit into the buffer
1631 struct ptlrpc_request rqbd_req;
1634 typedef int (*svc_handler_t)(struct ptlrpc_request *req);
1636 struct ptlrpc_service_ops {
1638 * if non-NULL called during thread creation (ptlrpc_start_thread())
1639 * to initialize service specific per-thread state.
1641 int (*so_thr_init)(struct ptlrpc_thread *thr);
1643 * if non-NULL called during thread shutdown (ptlrpc_main()) to
1644 * destruct state created by ->srv_init().
1646 void (*so_thr_done)(struct ptlrpc_thread *thr);
1648 * Handler function for incoming requests for this service
1650 int (*so_req_handler)(struct ptlrpc_request *req);
1652 * function to determine priority of the request, it's called
1653 * on every new request
1655 int (*so_hpreq_handler)(struct ptlrpc_request *);
1657 * service-specific print fn
1659 void (*so_req_printer)(void *, struct ptlrpc_request *);
1662 #ifndef __cfs_cacheline_aligned
1663 /* NB: put it here for reducing patche dependence */
1664 # define __cfs_cacheline_aligned
1668 * How many high priority requests to serve before serving one normal
1671 #define PTLRPC_SVC_HP_RATIO 10
1674 * Definition of PortalRPC service.
1675 * The service is listening on a particular portal (like tcp port)
1676 * and perform actions for a specific server like IO service for OST
1677 * or general metadata service for MDS.
1679 struct ptlrpc_service {
1680 /** serialize /proc operations */
1681 spinlock_t srv_lock;
1682 /** most often accessed fields */
1683 /** chain thru all services */
1684 struct list_head srv_list;
1685 /** service operations table */
1686 struct ptlrpc_service_ops srv_ops;
1687 /** only statically allocated strings here; we don't clean them */
1689 /** only statically allocated strings here; we don't clean them */
1690 char *srv_thread_name;
1691 /** service thread list */
1692 struct list_head srv_threads;
1693 /** threads # should be created for each partition on initializing */
1694 int srv_nthrs_cpt_init;
1695 /** limit of threads number for each partition */
1696 int srv_nthrs_cpt_limit;
1697 /** Root of /proc dir tree for this service */
1698 struct proc_dir_entry *srv_procroot;
1699 /** Pointer to statistic data for this service */
1700 struct lprocfs_stats *srv_stats;
1701 /** # hp per lp reqs to handle */
1702 int srv_hpreq_ratio;
1703 /** biggest request to receive */
1704 int srv_max_req_size;
1705 /** biggest reply to send */
1706 int srv_max_reply_size;
1707 /** size of individual buffers */
1709 /** # buffers to allocate in 1 group */
1710 int srv_nbuf_per_group;
1711 /** Local portal on which to receive requests */
1712 __u32 srv_req_portal;
1713 /** Portal on the client to send replies to */
1714 __u32 srv_rep_portal;
1716 * Tags for lu_context associated with this thread, see struct
1720 /** soft watchdog timeout multiplier */
1721 int srv_watchdog_factor;
1722 /** under unregister_service */
1723 unsigned srv_is_stopping:1;
1725 /** max # request buffers in history per partition */
1726 int srv_hist_nrqbds_cpt_max;
1727 /** number of CPTs this service bound on */
1729 /** CPTs array this service bound on */
1731 /** 2^srv_cptab_bits >= cfs_cpt_numbert(srv_cptable) */
1733 /** CPT table this service is running over */
1734 struct cfs_cpt_table *srv_cptable;
1737 struct kobject srv_kobj;
1738 struct completion srv_kobj_unregister;
1740 * partition data for ptlrpc service
1742 struct ptlrpc_service_part *srv_parts[0];
1746 * Definition of PortalRPC service partition data.
1747 * Although a service only has one instance of it right now, but we
1748 * will have multiple instances very soon (instance per CPT).
1750 * it has four locks:
1752 * serialize operations on rqbd and requests waiting for preprocess
1754 * serialize operations active requests sent to this portal
1756 * serialize adaptive timeout stuff
1758 * serialize operations on RS list (reply states)
1760 * We don't have any use-case to take two or more locks at the same time
1761 * for now, so there is no lock order issue.
1763 struct ptlrpc_service_part {
1764 /** back reference to owner */
1765 struct ptlrpc_service *scp_service __cfs_cacheline_aligned;
1766 /* CPT id, reserved */
1768 /** always increasing number */
1770 /** # of starting threads */
1771 int scp_nthrs_starting;
1772 /** # of stopping threads, reserved for shrinking threads */
1773 int scp_nthrs_stopping;
1774 /** # running threads */
1775 int scp_nthrs_running;
1776 /** service threads list */
1777 struct list_head scp_threads;
1780 * serialize the following fields, used for protecting
1781 * rqbd list and incoming requests waiting for preprocess,
1782 * threads starting & stopping are also protected by this lock.
1784 spinlock_t scp_lock __cfs_cacheline_aligned;
1785 /** total # req buffer descs allocated */
1786 int scp_nrqbds_total;
1787 /** # posted request buffers for receiving */
1788 int scp_nrqbds_posted;
1789 /** in progress of allocating rqbd */
1790 int scp_rqbd_allocating;
1791 /** # incoming reqs */
1792 int scp_nreqs_incoming;
1793 /** request buffers to be reposted */
1794 struct list_head scp_rqbd_idle;
1795 /** req buffers receiving */
1796 struct list_head scp_rqbd_posted;
1797 /** incoming reqs */
1798 struct list_head scp_req_incoming;
1799 /** timeout before re-posting reqs, in tick */
1800 cfs_duration_t scp_rqbd_timeout;
1802 * all threads sleep on this. This wait-queue is signalled when new
1803 * incoming request arrives and when difficult reply has to be handled.
1805 wait_queue_head_t scp_waitq;
1807 /** request history */
1808 struct list_head scp_hist_reqs;
1809 /** request buffer history */
1810 struct list_head scp_hist_rqbds;
1811 /** # request buffers in history */
1812 int scp_hist_nrqbds;
1813 /** sequence number for request */
1815 /** highest seq culled from history */
1816 __u64 scp_hist_seq_culled;
1819 * serialize the following fields, used for processing requests
1820 * sent to this portal
1822 spinlock_t scp_req_lock __cfs_cacheline_aligned;
1823 /** # reqs in either of the NRS heads below */
1824 /** # reqs being served */
1825 int scp_nreqs_active;
1826 /** # HPreqs being served */
1827 int scp_nhreqs_active;
1828 /** # hp requests handled */
1831 /** NRS head for regular requests */
1832 struct ptlrpc_nrs scp_nrs_reg;
1833 /** NRS head for HP requests; this is only valid for services that can
1834 * handle HP requests */
1835 struct ptlrpc_nrs *scp_nrs_hp;
1840 * serialize the following fields, used for changes on
1843 spinlock_t scp_at_lock __cfs_cacheline_aligned;
1844 /** estimated rpc service time */
1845 struct adaptive_timeout scp_at_estimate;
1846 /** reqs waiting for replies */
1847 struct ptlrpc_at_array scp_at_array;
1848 /** early reply timer */
1849 struct timer_list scp_at_timer;
1851 cfs_time_t scp_at_checktime;
1852 /** check early replies */
1853 unsigned scp_at_check;
1857 * serialize the following fields, used for processing
1858 * replies for this portal
1860 spinlock_t scp_rep_lock __cfs_cacheline_aligned;
1861 /** all the active replies */
1862 struct list_head scp_rep_active;
1863 /** List of free reply_states */
1864 struct list_head scp_rep_idle;
1865 /** waitq to run, when adding stuff to srv_free_rs_list */
1866 wait_queue_head_t scp_rep_waitq;
1867 /** # 'difficult' replies */
1868 atomic_t scp_nreps_difficult;
1871 #define ptlrpc_service_for_each_part(part, i, svc) \
1873 i < (svc)->srv_ncpts && \
1874 (svc)->srv_parts != NULL && \
1875 ((part) = (svc)->srv_parts[i]) != NULL; i++)
1878 * Declaration of ptlrpcd control structure
1880 struct ptlrpcd_ctl {
1882 * Ptlrpc thread control flags (LIOD_START, LIOD_STOP, LIOD_FORCE)
1884 unsigned long pc_flags;
1886 * Thread lock protecting structure fields.
1892 struct completion pc_starting;
1896 struct completion pc_finishing;
1898 * Thread requests set.
1900 struct ptlrpc_request_set *pc_set;
1902 * Thread name used in kthread_run()
1906 * CPT the thread is bound on.
1910 * Index of ptlrpcd thread in the array.
1914 * Pointer to the array of partners' ptlrpcd_ctl structure.
1916 struct ptlrpcd_ctl **pc_partners;
1918 * Number of the ptlrpcd's partners.
1922 * Record the partner index to be processed next.
1926 * Error code if the thread failed to fully start.
1931 /* Bits for pc_flags */
1932 enum ptlrpcd_ctl_flags {
1934 * Ptlrpc thread start flag.
1936 LIOD_START = 1 << 0,
1938 * Ptlrpc thread stop flag.
1942 * Ptlrpc thread force flag (only stop force so far).
1943 * This will cause aborting any inflight rpcs handled
1944 * by thread if LIOD_STOP is specified.
1946 LIOD_FORCE = 1 << 2,
1948 * This is a recovery ptlrpc thread.
1950 LIOD_RECOVERY = 1 << 3,
1957 * Service compatibility function; the policy is compatible with all services.
1959 * \param[in] svc The service the policy is attempting to register with.
1960 * \param[in] desc The policy descriptor
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_all(const struct ptlrpc_service *svc,
1967 const struct ptlrpc_nrs_pol_desc *desc)
1973 * Service compatibility function; the policy is compatible with only a specific
1974 * service which is identified by its human-readable name at
1975 * ptlrpc_service::srv_name.
1977 * \param[in] svc The service the policy is attempting to register with.
1978 * \param[in] desc The policy descriptor
1980 * \retval false The policy is not compatible with the service
1981 * \retval true The policy is compatible with the service
1983 * \see ptlrpc_nrs_pol_desc::pd_compat()
1985 static inline bool nrs_policy_compat_one(const struct ptlrpc_service *svc,
1986 const struct ptlrpc_nrs_pol_desc *desc)
1988 LASSERT(desc->pd_compat_svc_name != NULL);
1989 return strcmp(svc->srv_name, desc->pd_compat_svc_name) == 0;
1994 /* ptlrpc/events.c */
1995 extern struct lnet_handle_eq ptlrpc_eq_h;
1996 extern int ptlrpc_uuid_to_peer(struct obd_uuid *uuid,
1997 struct lnet_process_id *peer, lnet_nid_t *self);
1999 * These callbacks are invoked by LNet when something happened to
2003 extern void request_out_callback(struct lnet_event *ev);
2004 extern void reply_in_callback(struct lnet_event *ev);
2005 extern void client_bulk_callback(struct lnet_event *ev);
2006 extern void request_in_callback(struct lnet_event *ev);
2007 extern void reply_out_callback(struct lnet_event *ev);
2008 #ifdef HAVE_SERVER_SUPPORT
2009 extern void server_bulk_callback(struct lnet_event *ev);
2013 /* ptlrpc/connection.c */
2014 struct ptlrpc_connection *ptlrpc_connection_get(struct lnet_process_id peer,
2016 struct obd_uuid *uuid);
2017 int ptlrpc_connection_put(struct ptlrpc_connection *c);
2018 struct ptlrpc_connection *ptlrpc_connection_addref(struct ptlrpc_connection *);
2019 int ptlrpc_connection_init(void);
2020 void ptlrpc_connection_fini(void);
2021 extern lnet_pid_t ptl_get_pid(void);
2024 * Check if the peer connection is on the local node. We need to use GFP_NOFS
2025 * for requests from a local client to avoid recursing into the filesystem
2026 * as we might end up waiting on a page sent in the request we're serving.
2028 * Use __GFP_HIGHMEM so that the pages can use all of the available memory
2029 * on 32-bit machines. Use more aggressive GFP_HIGHUSER flags from non-local
2030 * clients to be able to generate more memory pressure on the OSS and allow
2031 * inactive pages to be reclaimed, since it doesn't have any other processes
2032 * or allocations that generate memory reclaim pressure.
2034 * See b=17576 (bdf50dc9) and b=19529 (3dcf18d3) for details.
2036 static inline bool ptlrpc_connection_is_local(struct ptlrpc_connection *conn)
2041 if (conn->c_peer.nid == conn->c_self)
2044 RETURN(LNetIsPeerLocal(conn->c_peer.nid));
2047 /* ptlrpc/niobuf.c */
2049 * Actual interfacing with LNet to put/get/register/unregister stuff
2052 #ifdef HAVE_SERVER_SUPPORT
2053 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_exp(struct ptlrpc_request *req,
2054 unsigned nfrags, unsigned max_brw,
2057 const struct ptlrpc_bulk_frag_ops
2059 int ptlrpc_start_bulk_transfer(struct ptlrpc_bulk_desc *desc);
2060 void ptlrpc_abort_bulk(struct ptlrpc_bulk_desc *desc);
2062 static inline int ptlrpc_server_bulk_active(struct ptlrpc_bulk_desc *desc)
2066 LASSERT(desc != NULL);
2068 spin_lock(&desc->bd_lock);
2069 rc = desc->bd_md_count;
2070 spin_unlock(&desc->bd_lock);
2075 int ptlrpc_register_bulk(struct ptlrpc_request *req);
2076 int ptlrpc_unregister_bulk(struct ptlrpc_request *req, int async);
2078 static inline int ptlrpc_client_bulk_active(struct ptlrpc_request *req)
2080 struct ptlrpc_bulk_desc *desc;
2083 LASSERT(req != NULL);
2084 desc = req->rq_bulk;
2086 if (req->rq_bulk_deadline > ktime_get_real_seconds())
2092 spin_lock(&desc->bd_lock);
2093 rc = desc->bd_md_count;
2094 spin_unlock(&desc->bd_lock);
2098 #define PTLRPC_REPLY_MAYBE_DIFFICULT 0x01
2099 #define PTLRPC_REPLY_EARLY 0x02
2100 int ptlrpc_send_reply(struct ptlrpc_request *req, int flags);
2101 int ptlrpc_reply(struct ptlrpc_request *req);
2102 int ptlrpc_send_error(struct ptlrpc_request *req, int difficult);
2103 int ptlrpc_error(struct ptlrpc_request *req);
2104 int ptlrpc_at_get_net_latency(struct ptlrpc_request *req);
2105 int ptl_send_rpc(struct ptlrpc_request *request, int noreply);
2106 int ptlrpc_register_rqbd(struct ptlrpc_request_buffer_desc *rqbd);
2109 /* ptlrpc/client.c */
2111 * Client-side portals API. Everything to send requests, receive replies,
2112 * request queues, request management, etc.
2115 void ptlrpc_request_committed(struct ptlrpc_request *req, int force);
2117 void ptlrpc_init_client(int req_portal, int rep_portal, char *name,
2118 struct ptlrpc_client *);
2119 void ptlrpc_cleanup_client(struct obd_import *imp);
2120 struct ptlrpc_connection *ptlrpc_uuid_to_connection(struct obd_uuid *uuid,
2121 lnet_nid_t nid4refnet);
2123 int ptlrpc_queue_wait(struct ptlrpc_request *req);
2124 int ptlrpc_replay_req(struct ptlrpc_request *req);
2125 void ptlrpc_restart_req(struct ptlrpc_request *req);
2126 void ptlrpc_abort_inflight(struct obd_import *imp);
2127 void ptlrpc_cleanup_imp(struct obd_import *imp);
2128 void ptlrpc_abort_set(struct ptlrpc_request_set *set);
2130 struct ptlrpc_request_set *ptlrpc_prep_set(void);
2131 struct ptlrpc_request_set *ptlrpc_prep_fcset(int max, set_producer_func func,
2133 int ptlrpc_set_add_cb(struct ptlrpc_request_set *set,
2134 set_interpreter_func fn, void *data);
2135 int ptlrpc_check_set(const struct lu_env *env, struct ptlrpc_request_set *set);
2136 int ptlrpc_set_wait(struct ptlrpc_request_set *);
2137 void ptlrpc_mark_interrupted(struct ptlrpc_request *req);
2138 void ptlrpc_set_destroy(struct ptlrpc_request_set *);
2139 void ptlrpc_set_add_req(struct ptlrpc_request_set *, struct ptlrpc_request *);
2141 void ptlrpc_free_rq_pool(struct ptlrpc_request_pool *pool);
2142 int ptlrpc_add_rqs_to_pool(struct ptlrpc_request_pool *pool, int num_rq);
2144 struct ptlrpc_request_pool *
2145 ptlrpc_init_rq_pool(int, int,
2146 int (*populate_pool)(struct ptlrpc_request_pool *, int));
2148 void ptlrpc_at_set_req_timeout(struct ptlrpc_request *req);
2149 struct ptlrpc_request *ptlrpc_request_alloc(struct obd_import *imp,
2150 const struct req_format *format);
2151 struct ptlrpc_request *ptlrpc_request_alloc_pool(struct obd_import *imp,
2152 struct ptlrpc_request_pool *,
2153 const struct req_format *format);
2154 void ptlrpc_request_free(struct ptlrpc_request *request);
2155 int ptlrpc_request_pack(struct ptlrpc_request *request,
2156 __u32 version, int opcode);
2157 struct ptlrpc_request *ptlrpc_request_alloc_pack(struct obd_import *imp,
2158 const struct req_format *format,
2159 __u32 version, int opcode);
2160 int ptlrpc_request_bufs_pack(struct ptlrpc_request *request,
2161 __u32 version, int opcode, char **bufs,
2162 struct ptlrpc_cli_ctx *ctx);
2163 void ptlrpc_req_finished(struct ptlrpc_request *request);
2164 void ptlrpc_req_finished_with_imp_lock(struct ptlrpc_request *request);
2165 struct ptlrpc_request *ptlrpc_request_addref(struct ptlrpc_request *req);
2166 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_imp(struct ptlrpc_request *req,
2167 unsigned nfrags, unsigned max_brw,
2170 const struct ptlrpc_bulk_frag_ops
2173 int ptlrpc_prep_bulk_frag(struct ptlrpc_bulk_desc *desc,
2174 void *frag, int len);
2175 void __ptlrpc_prep_bulk_page(struct ptlrpc_bulk_desc *desc,
2176 struct page *page, int pageoffset, int len,
2178 static inline void ptlrpc_prep_bulk_page_pin(struct ptlrpc_bulk_desc *desc,
2179 struct page *page, int pageoffset,
2182 __ptlrpc_prep_bulk_page(desc, page, pageoffset, len, 1);
2185 static inline void ptlrpc_prep_bulk_page_nopin(struct ptlrpc_bulk_desc *desc,
2186 struct page *page, int pageoffset,
2189 __ptlrpc_prep_bulk_page(desc, page, pageoffset, len, 0);
2192 void ptlrpc_free_bulk(struct ptlrpc_bulk_desc *bulk);
2194 static inline void ptlrpc_release_bulk_page_pin(struct ptlrpc_bulk_desc *desc)
2198 for (i = 0; i < desc->bd_iov_count ; i++)
2199 put_page(BD_GET_KIOV(desc, i).kiov_page);
2202 static inline void ptlrpc_release_bulk_noop(struct ptlrpc_bulk_desc *desc)
2206 void ptlrpc_retain_replayable_request(struct ptlrpc_request *req,
2207 struct obd_import *imp);
2208 __u64 ptlrpc_next_xid(void);
2209 __u64 ptlrpc_sample_next_xid(void);
2210 __u64 ptlrpc_req_xid(struct ptlrpc_request *request);
2212 /* Set of routines to run a function in ptlrpcd context */
2213 void *ptlrpcd_alloc_work(struct obd_import *imp,
2214 int (*cb)(const struct lu_env *, void *), void *data);
2215 void ptlrpcd_destroy_work(void *handler);
2216 int ptlrpcd_queue_work(void *handler);
2219 struct ptlrpc_service_buf_conf {
2220 /* nbufs is buffers # to allocate when growing the pool */
2221 unsigned int bc_nbufs;
2222 /* buffer size to post */
2223 unsigned int bc_buf_size;
2224 /* portal to listed for requests on */
2225 unsigned int bc_req_portal;
2226 /* portal of where to send replies to */
2227 unsigned int bc_rep_portal;
2228 /* maximum request size to be accepted for this service */
2229 unsigned int bc_req_max_size;
2230 /* maximum reply size this service can ever send */
2231 unsigned int bc_rep_max_size;
2234 struct ptlrpc_service_thr_conf {
2235 /* threadname should be 8 characters or less - 6 will be added on */
2237 /* threads increasing factor for each CPU */
2238 unsigned int tc_thr_factor;
2239 /* service threads # to start on each partition while initializing */
2240 unsigned int tc_nthrs_init;
2242 * low water of threads # upper-limit on each partition while running,
2243 * service availability may be impacted if threads number is lower
2244 * than this value. It can be ZERO if the service doesn't require
2245 * CPU affinity or there is only one partition.
2247 unsigned int tc_nthrs_base;
2248 /* "soft" limit for total threads number */
2249 unsigned int tc_nthrs_max;
2250 /* user specified threads number, it will be validated due to
2251 * other members of this structure. */
2252 unsigned int tc_nthrs_user;
2253 /* set NUMA node affinity for service threads */
2254 unsigned int tc_cpu_affinity;
2255 /* Tags for lu_context associated with service thread */
2259 struct ptlrpc_service_cpt_conf {
2260 struct cfs_cpt_table *cc_cptable;
2261 /* string pattern to describe CPTs for a service */
2265 struct ptlrpc_service_conf {
2268 /* soft watchdog timeout multiplifier to print stuck service traces */
2269 unsigned int psc_watchdog_factor;
2270 /* buffer information */
2271 struct ptlrpc_service_buf_conf psc_buf;
2272 /* thread information */
2273 struct ptlrpc_service_thr_conf psc_thr;
2274 /* CPU partition information */
2275 struct ptlrpc_service_cpt_conf psc_cpt;
2276 /* function table */
2277 struct ptlrpc_service_ops psc_ops;
2280 /* ptlrpc/service.c */
2282 * Server-side services API. Register/unregister service, request state
2283 * management, service thread management
2287 void ptlrpc_save_lock(struct ptlrpc_request *req, struct lustre_handle *lock,
2288 int mode, bool no_ack, bool convert_lock);
2289 void ptlrpc_commit_replies(struct obd_export *exp);
2290 void ptlrpc_dispatch_difficult_reply(struct ptlrpc_reply_state *rs);
2291 void ptlrpc_schedule_difficult_reply(struct ptlrpc_reply_state *rs);
2292 int ptlrpc_hpreq_handler(struct ptlrpc_request *req);
2293 struct ptlrpc_service *ptlrpc_register_service(
2294 struct ptlrpc_service_conf *conf,
2295 struct kset *parent,
2296 struct proc_dir_entry *proc_entry);
2297 void ptlrpc_stop_all_threads(struct ptlrpc_service *svc);
2299 int ptlrpc_start_threads(struct ptlrpc_service *svc);
2300 int ptlrpc_unregister_service(struct ptlrpc_service *service);
2301 int ptlrpc_service_health_check(struct ptlrpc_service *);
2302 void ptlrpc_server_drop_request(struct ptlrpc_request *req);
2303 void ptlrpc_request_change_export(struct ptlrpc_request *req,
2304 struct obd_export *export);
2305 void ptlrpc_update_export_timer(struct obd_export *exp,
2306 time64_t extra_delay);
2308 int ptlrpc_hr_init(void);
2309 void ptlrpc_hr_fini(void);
2313 /* ptlrpc/import.c */
2318 int ptlrpc_connect_import(struct obd_import *imp);
2319 int ptlrpc_init_import(struct obd_import *imp);
2320 int ptlrpc_disconnect_import(struct obd_import *imp, int noclose);
2321 int ptlrpc_import_recovery_state_machine(struct obd_import *imp);
2322 void deuuidify(char *uuid, const char *prefix, char **uuid_start,
2324 void ptlrpc_import_enter_resend(struct obd_import *imp);
2325 /* ptlrpc/pack_generic.c */
2326 int ptlrpc_reconnect_import(struct obd_import *imp);
2330 * ptlrpc msg buffer and swab interface
2334 int ptlrpc_buf_need_swab(struct ptlrpc_request *req, const int inout,
2336 void ptlrpc_buf_set_swabbed(struct ptlrpc_request *req, const int inout,
2338 int ptlrpc_unpack_rep_msg(struct ptlrpc_request *req, int len);
2339 int ptlrpc_unpack_req_msg(struct ptlrpc_request *req, int len);
2341 int lustre_msg_check_version(struct lustre_msg *msg, __u32 version);
2342 void lustre_init_msg_v2(struct lustre_msg_v2 *msg, int count, __u32 *lens,
2344 int lustre_pack_request(struct ptlrpc_request *, __u32 magic, int count,
2345 __u32 *lens, char **bufs);
2346 int lustre_pack_reply(struct ptlrpc_request *, int count, __u32 *lens,
2348 int lustre_pack_reply_v2(struct ptlrpc_request *req, int count,
2349 __u32 *lens, char **bufs, int flags);
2350 #define LPRFL_EARLY_REPLY 1
2351 int lustre_pack_reply_flags(struct ptlrpc_request *, int count, __u32 *lens,
2352 char **bufs, int flags);
2353 int lustre_shrink_msg(struct lustre_msg *msg, int segment,
2354 unsigned int newlen, int move_data);
2355 void lustre_free_reply_state(struct ptlrpc_reply_state *rs);
2356 int __lustre_unpack_msg(struct lustre_msg *m, int len);
2357 __u32 lustre_msg_hdr_size(__u32 magic, __u32 count);
2358 __u32 lustre_msg_size(__u32 magic, int count, __u32 *lengths);
2359 __u32 lustre_msg_size_v2(int count, __u32 *lengths);
2360 __u32 lustre_packed_msg_size(struct lustre_msg *msg);
2361 __u32 lustre_msg_early_size(void);
2362 void *lustre_msg_buf_v2(struct lustre_msg_v2 *m, __u32 n, __u32 min_size);
2363 void *lustre_msg_buf(struct lustre_msg *m, __u32 n, __u32 minlen);
2364 __u32 lustre_msg_buflen(struct lustre_msg *m, __u32 n);
2365 void lustre_msg_set_buflen(struct lustre_msg *m, __u32 n, __u32 len);
2366 __u32 lustre_msg_bufcount(struct lustre_msg *m);
2367 char *lustre_msg_string(struct lustre_msg *m, __u32 n, __u32 max_len);
2368 __u32 lustre_msghdr_get_flags(struct lustre_msg *msg);
2369 void lustre_msghdr_set_flags(struct lustre_msg *msg, __u32 flags);
2370 __u32 lustre_msg_get_flags(struct lustre_msg *msg);
2371 void lustre_msg_add_flags(struct lustre_msg *msg, __u32 flags);
2372 void lustre_msg_set_flags(struct lustre_msg *msg, __u32 flags);
2373 void lustre_msg_clear_flags(struct lustre_msg *msg, __u32 flags);
2374 __u32 lustre_msg_get_op_flags(struct lustre_msg *msg);
2375 void lustre_msg_add_op_flags(struct lustre_msg *msg, __u32 flags);
2376 struct lustre_handle *lustre_msg_get_handle(struct lustre_msg *msg);
2377 __u32 lustre_msg_get_type(struct lustre_msg *msg);
2378 __u32 lustre_msg_get_version(struct lustre_msg *msg);
2379 void lustre_msg_add_version(struct lustre_msg *msg, __u32 version);
2380 __u32 lustre_msg_get_opc(struct lustre_msg *msg);
2381 __u64 lustre_msg_get_last_xid(struct lustre_msg *msg);
2382 __u16 lustre_msg_get_tag(struct lustre_msg *msg);
2383 __u64 lustre_msg_get_last_committed(struct lustre_msg *msg);
2384 __u64 *lustre_msg_get_versions(struct lustre_msg *msg);
2385 __u64 lustre_msg_get_transno(struct lustre_msg *msg);
2386 __u64 lustre_msg_get_slv(struct lustre_msg *msg);
2387 __u32 lustre_msg_get_limit(struct lustre_msg *msg);
2388 void lustre_msg_set_slv(struct lustre_msg *msg, __u64 slv);
2389 void lustre_msg_set_limit(struct lustre_msg *msg, __u64 limit);
2390 int lustre_msg_get_status(struct lustre_msg *msg);
2391 __u32 lustre_msg_get_conn_cnt(struct lustre_msg *msg);
2392 __u32 lustre_msg_get_magic(struct lustre_msg *msg);
2393 __u32 lustre_msg_get_timeout(struct lustre_msg *msg);
2394 __u32 lustre_msg_get_service_time(struct lustre_msg *msg);
2395 char *lustre_msg_get_jobid(struct lustre_msg *msg);
2396 __u32 lustre_msg_get_cksum(struct lustre_msg *msg);
2397 __u64 lustre_msg_get_mbits(struct lustre_msg *msg);
2398 __u32 lustre_msg_calc_cksum(struct lustre_msg *msg);
2399 void lustre_msg_set_handle(struct lustre_msg *msg,struct lustre_handle *handle);
2400 void lustre_msg_set_type(struct lustre_msg *msg, __u32 type);
2401 void lustre_msg_set_opc(struct lustre_msg *msg, __u32 opc);
2402 void lustre_msg_set_last_xid(struct lustre_msg *msg, __u64 last_xid);
2403 void lustre_msg_set_tag(struct lustre_msg *msg, __u16 tag);
2404 void lustre_msg_set_last_committed(struct lustre_msg *msg,__u64 last_committed);
2405 void lustre_msg_set_versions(struct lustre_msg *msg, __u64 *versions);
2406 void lustre_msg_set_transno(struct lustre_msg *msg, __u64 transno);
2407 void lustre_msg_set_status(struct lustre_msg *msg, __u32 status);
2408 void lustre_msg_set_conn_cnt(struct lustre_msg *msg, __u32 conn_cnt);
2409 void ptlrpc_req_set_repsize(struct ptlrpc_request *req, int count, __u32 *sizes);
2410 void ptlrpc_request_set_replen(struct ptlrpc_request *req);
2411 void lustre_msg_set_timeout(struct lustre_msg *msg, __u32 timeout);
2412 void lustre_msg_set_service_time(struct lustre_msg *msg, __u32 service_time);
2413 void lustre_msg_set_jobid(struct lustre_msg *msg, char *jobid);
2414 void lustre_msg_set_cksum(struct lustre_msg *msg, __u32 cksum);
2415 void lustre_msg_set_mbits(struct lustre_msg *msg, __u64 mbits);
2418 lustre_shrink_reply(struct ptlrpc_request *req, int segment,
2419 unsigned int newlen, int move_data)
2421 LASSERT(req->rq_reply_state);
2422 LASSERT(req->rq_repmsg);
2423 req->rq_replen = lustre_shrink_msg(req->rq_repmsg, segment,
2427 #ifdef LUSTRE_TRANSLATE_ERRNOS
2429 static inline int ptlrpc_status_hton(int h)
2432 * Positive errnos must be network errnos, such as LUSTRE_EDEADLK,
2433 * ELDLM_LOCK_ABORTED, etc.
2436 return -lustre_errno_hton(-h);
2441 static inline int ptlrpc_status_ntoh(int n)
2444 * See the comment in ptlrpc_status_hton().
2447 return -lustre_errno_ntoh(-n);
2454 #define ptlrpc_status_hton(h) (h)
2455 #define ptlrpc_status_ntoh(n) (n)
2460 /** Change request phase of \a req to \a new_phase */
2462 ptlrpc_rqphase_move(struct ptlrpc_request *req, enum rq_phase new_phase)
2464 if (req->rq_phase == new_phase)
2467 if (new_phase == RQ_PHASE_UNREG_RPC ||
2468 new_phase == RQ_PHASE_UNREG_BULK) {
2469 /* No embedded unregistering phases */
2470 if (req->rq_phase == RQ_PHASE_UNREG_RPC ||
2471 req->rq_phase == RQ_PHASE_UNREG_BULK)
2474 req->rq_next_phase = req->rq_phase;
2476 atomic_inc(&req->rq_import->imp_unregistering);
2479 if (req->rq_phase == RQ_PHASE_UNREG_RPC ||
2480 req->rq_phase == RQ_PHASE_UNREG_BULK) {
2482 atomic_dec(&req->rq_import->imp_unregistering);
2485 DEBUG_REQ(D_INFO, req, "move req \"%s\" -> \"%s\"",
2486 ptlrpc_rqphase2str(req), ptlrpc_phase2str(new_phase));
2488 req->rq_phase = new_phase;
2492 * Returns true if request \a req got early reply and hard deadline is not met
2495 ptlrpc_client_early(struct ptlrpc_request *req)
2497 return req->rq_early;
2501 * Returns true if we got real reply from server for this request
2504 ptlrpc_client_replied(struct ptlrpc_request *req)
2506 if (req->rq_reply_deadline > ktime_get_real_seconds())
2508 return req->rq_replied;
2511 /** Returns true if request \a req is in process of receiving server reply */
2513 ptlrpc_client_recv(struct ptlrpc_request *req)
2515 if (req->rq_reply_deadline > ktime_get_real_seconds())
2517 return req->rq_receiving_reply;
2521 ptlrpc_client_recv_or_unlink(struct ptlrpc_request *req)
2525 spin_lock(&req->rq_lock);
2526 if (req->rq_reply_deadline > ktime_get_real_seconds()) {
2527 spin_unlock(&req->rq_lock);
2530 if (req->rq_req_deadline > ktime_get_real_seconds()) {
2531 spin_unlock(&req->rq_lock);
2535 rc = !req->rq_req_unlinked || !req->rq_reply_unlinked ||
2536 req->rq_receiving_reply;
2537 spin_unlock(&req->rq_lock);
2542 ptlrpc_client_wake_req(struct ptlrpc_request *req)
2545 if (req->rq_set == NULL)
2546 wake_up(&req->rq_reply_waitq);
2548 wake_up(&req->rq_set->set_waitq);
2552 ptlrpc_rs_addref(struct ptlrpc_reply_state *rs)
2554 LASSERT(atomic_read(&rs->rs_refcount) > 0);
2555 atomic_inc(&rs->rs_refcount);
2559 ptlrpc_rs_decref(struct ptlrpc_reply_state *rs)
2561 LASSERT(atomic_read(&rs->rs_refcount) > 0);
2562 if (atomic_dec_and_test(&rs->rs_refcount))
2563 lustre_free_reply_state(rs);
2566 /* Should only be called once per req */
2567 static inline void ptlrpc_req_drop_rs(struct ptlrpc_request *req)
2569 if (req->rq_reply_state == NULL)
2570 return; /* shouldn't occur */
2571 ptlrpc_rs_decref(req->rq_reply_state);
2572 req->rq_reply_state = NULL;
2573 req->rq_repmsg = NULL;
2576 static inline __u32 lustre_request_magic(struct ptlrpc_request *req)
2578 return lustre_msg_get_magic(req->rq_reqmsg);
2581 static inline int ptlrpc_req_get_repsize(struct ptlrpc_request *req)
2583 switch (req->rq_reqmsg->lm_magic) {
2584 case LUSTRE_MSG_MAGIC_V2:
2585 return req->rq_reqmsg->lm_repsize;
2587 LASSERTF(0, "incorrect message magic: %08x\n",
2588 req->rq_reqmsg->lm_magic);
2593 static inline int ptlrpc_send_limit_expired(struct ptlrpc_request *req)
2595 if (req->rq_delay_limit != 0 &&
2596 req->rq_queued_time + req->rq_delay_limit < ktime_get_seconds())
2601 static inline int ptlrpc_no_resend(struct ptlrpc_request *req)
2603 if (!req->rq_no_resend && ptlrpc_send_limit_expired(req)) {
2604 spin_lock(&req->rq_lock);
2605 req->rq_no_resend = 1;
2606 spin_unlock(&req->rq_lock);
2608 return req->rq_no_resend;
2612 ptlrpc_server_get_timeout(struct ptlrpc_service_part *svcpt)
2614 int at = AT_OFF ? 0 : at_get(&svcpt->scp_at_estimate);
2616 return svcpt->scp_service->srv_watchdog_factor *
2617 max_t(int, at, obd_timeout);
2620 static inline struct ptlrpc_service *
2621 ptlrpc_req2svc(struct ptlrpc_request *req)
2623 LASSERT(req->rq_rqbd != NULL);
2624 return req->rq_rqbd->rqbd_svcpt->scp_service;
2627 /* ldlm/ldlm_lib.c */
2629 * Target client logic
2632 int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg);
2633 int client_obd_cleanup(struct obd_device *obddev);
2634 int client_connect_import(const struct lu_env *env,
2635 struct obd_export **exp, struct obd_device *obd,
2636 struct obd_uuid *cluuid, struct obd_connect_data *,
2638 int client_disconnect_export(struct obd_export *exp);
2639 int client_import_add_conn(struct obd_import *imp, struct obd_uuid *uuid,
2641 int client_import_del_conn(struct obd_import *imp, struct obd_uuid *uuid);
2642 int client_import_find_conn(struct obd_import *imp, lnet_nid_t peer,
2643 struct obd_uuid *uuid);
2644 int import_set_conn_priority(struct obd_import *imp, struct obd_uuid *uuid);
2645 void client_destroy_import(struct obd_import *imp);
2648 #ifdef HAVE_SERVER_SUPPORT
2649 int server_disconnect_export(struct obd_export *exp);
2652 /* ptlrpc/pinger.c */
2654 * Pinger API (client side only)
2657 enum timeout_event {
2660 struct timeout_item;
2661 typedef int (*timeout_cb_t)(struct timeout_item *, void *);
2662 int ptlrpc_pinger_add_import(struct obd_import *imp);
2663 int ptlrpc_pinger_del_import(struct obd_import *imp);
2664 int ptlrpc_add_timeout_client(time64_t time, enum timeout_event event,
2665 timeout_cb_t cb, void *data,
2666 struct list_head *obd_list);
2667 int ptlrpc_del_timeout_client(struct list_head *obd_list,
2668 enum timeout_event event);
2669 struct ptlrpc_request * ptlrpc_prep_ping(struct obd_import *imp);
2670 int ptlrpc_obd_ping(struct obd_device *obd);
2671 void ping_evictor_start(void);
2672 void ping_evictor_stop(void);
2673 void ptlrpc_pinger_ir_up(void);
2674 void ptlrpc_pinger_ir_down(void);
2676 int ptlrpc_pinger_suppress_pings(void);
2678 /* ptlrpc/ptlrpcd.c */
2679 void ptlrpcd_stop(struct ptlrpcd_ctl *pc, int force);
2680 void ptlrpcd_free(struct ptlrpcd_ctl *pc);
2681 void ptlrpcd_wake(struct ptlrpc_request *req);
2682 void ptlrpcd_add_req(struct ptlrpc_request *req);
2683 void ptlrpcd_add_rqset(struct ptlrpc_request_set *set);
2684 int ptlrpcd_addref(void);
2685 void ptlrpcd_decref(void);
2687 /* ptlrpc/lproc_ptlrpc.c */
2689 * procfs output related functions
2692 const char* ll_opcode2str(__u32 opcode);
2693 const int ll_str2opcode(const char *ops);
2694 #ifdef CONFIG_PROC_FS
2695 void ptlrpc_lprocfs_register_obd(struct obd_device *obd);
2696 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd);
2697 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes);
2699 static inline void ptlrpc_lprocfs_register_obd(struct obd_device *obd) {}
2700 static inline void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd) {}
2701 static inline void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes) {}
2705 /* ptlrpc/llog_server.c */
2706 int llog_origin_handle_open(struct ptlrpc_request *req);
2707 int llog_origin_handle_destroy(struct ptlrpc_request *req);
2708 int llog_origin_handle_prev_block(struct ptlrpc_request *req);
2709 int llog_origin_handle_next_block(struct ptlrpc_request *req);
2710 int llog_origin_handle_read_header(struct ptlrpc_request *req);
2711 int llog_origin_handle_close(struct ptlrpc_request *req);
2713 /* ptlrpc/llog_client.c */
2714 extern struct llog_operations llog_client_ops;