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, 2017, 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 #ifdef HAVE_RHASHTABLE_LOOKUP_GET_INSERT_FAST
55 #include <linux/rhashtable.h>
57 #include <libcfs/linux/linux-hash.h>
59 #include <linux/uio.h>
60 #include <libcfs/libcfs.h>
62 #include <lnet/lib-types.h>
63 #include <uapi/linux/lnet/nidstr.h>
64 #include <uapi/linux/lustre/lustre_idl.h>
65 #include <lustre_ha.h>
66 #include <lustre_sec.h>
67 #include <lustre_import.h>
68 #include <lprocfs_status.h>
69 #include <lu_object.h>
70 #include <lustre_req_layout.h>
71 #include <obd_support.h>
72 #include <uapi/linux/lustre/lustre_ver.h>
74 /* MD flags we _always_ use */
75 #define PTLRPC_MD_OPTIONS 0
78 * log2 max # of bulk operations in one request: 2=4MB/RPC, 5=32MB/RPC, ...
79 * In order for the client and server to properly negotiate the maximum
80 * possible transfer size, PTLRPC_BULK_OPS_COUNT must be a power-of-two
81 * value. The client is free to limit the actual RPC size for any bulk
82 * transfer via cl_max_pages_per_rpc to some non-power-of-two value.
83 * NOTE: This is limited to 16 (=64GB RPCs) by IOOBJ_MAX_BRW_BITS. */
84 #define PTLRPC_BULK_OPS_BITS 4
85 #if PTLRPC_BULK_OPS_BITS > 16
86 #error "More than 65536 BRW RPCs not allowed by IOOBJ_MAX_BRW_BITS."
88 #define PTLRPC_BULK_OPS_COUNT (1U << PTLRPC_BULK_OPS_BITS)
90 * PTLRPC_BULK_OPS_MASK is for the convenience of the client only, and
91 * should not be used on the server at all. Otherwise, it imposes a
92 * protocol limitation on the maximum RPC size that can be used by any
93 * RPC sent to that server in the future. Instead, the server should
94 * use the negotiated per-client ocd_brw_size to determine the bulk
96 #define PTLRPC_BULK_OPS_MASK (~((__u64)PTLRPC_BULK_OPS_COUNT - 1))
99 * Define maxima for bulk I/O.
101 * A single PTLRPC BRW request is sent via up to PTLRPC_BULK_OPS_COUNT
102 * of LNET_MTU sized RDMA transfers. Clients and servers negotiate the
103 * currently supported maximum between peers at connect via ocd_brw_size.
105 #define PTLRPC_MAX_BRW_BITS (LNET_MTU_BITS + PTLRPC_BULK_OPS_BITS)
106 #define PTLRPC_MAX_BRW_SIZE (1U << PTLRPC_MAX_BRW_BITS)
107 #define PTLRPC_MAX_BRW_PAGES (PTLRPC_MAX_BRW_SIZE >> PAGE_SHIFT)
109 #define ONE_MB_BRW_SIZE (1U << LNET_MTU_BITS)
110 #define MD_MAX_BRW_SIZE (1U << LNET_MTU_BITS)
111 #define MD_MAX_BRW_PAGES (MD_MAX_BRW_SIZE >> PAGE_SHIFT)
112 #define DT_MAX_BRW_SIZE PTLRPC_MAX_BRW_SIZE
113 #define DT_DEF_BRW_SIZE (4 * ONE_MB_BRW_SIZE)
114 #define DT_MAX_BRW_PAGES (DT_MAX_BRW_SIZE >> PAGE_SHIFT)
115 #define OFD_MAX_BRW_SIZE (1U << LNET_MTU_BITS)
117 /* When PAGE_SIZE is a constant, we can check our arithmetic here with cpp! */
118 #if ((PTLRPC_MAX_BRW_PAGES & (PTLRPC_MAX_BRW_PAGES - 1)) != 0)
119 # error "PTLRPC_MAX_BRW_PAGES isn't a power of two"
121 #if (PTLRPC_MAX_BRW_SIZE != (PTLRPC_MAX_BRW_PAGES * PAGE_SIZE))
122 # error "PTLRPC_MAX_BRW_SIZE isn't PTLRPC_MAX_BRW_PAGES * PAGE_SIZE"
124 #if (PTLRPC_MAX_BRW_SIZE > LNET_MTU * PTLRPC_BULK_OPS_COUNT)
125 # error "PTLRPC_MAX_BRW_SIZE too big"
127 #if (PTLRPC_MAX_BRW_PAGES > LNET_MAX_IOV * PTLRPC_BULK_OPS_COUNT)
128 # error "PTLRPC_MAX_BRW_PAGES too big"
131 #define PTLRPC_NTHRS_INIT 2
136 * Constants determine how memory is used to buffer incoming service requests.
138 * ?_NBUFS # buffers to allocate when growing the pool
139 * ?_BUFSIZE # bytes in a single request buffer
140 * ?_MAXREQSIZE # maximum request service will receive
142 * When fewer than ?_NBUFS/2 buffers are posted for receive, another chunk
143 * of ?_NBUFS is added to the pool.
145 * Messages larger than ?_MAXREQSIZE are dropped. Request buffers are
146 * considered full when less than ?_MAXREQSIZE is left in them.
151 * Constants determine how threads are created for ptlrpc service.
153 * ?_NTHRS_INIT # threads to create for each service partition on
154 * initializing. If it's non-affinity service and
155 * there is only one partition, it's the overall #
156 * threads for the service while initializing.
157 * ?_NTHRS_BASE # threads should be created at least for each
158 * ptlrpc partition to keep the service healthy.
159 * It's the low-water mark of threads upper-limit
160 * for each partition.
161 * ?_THR_FACTOR # threads can be added on threads upper-limit for
162 * each CPU core. This factor is only for reference,
163 * we might decrease value of factor if number of cores
164 * per CPT is above a limit.
165 * ?_NTHRS_MAX # overall threads can be created for a service,
166 * it's a soft limit because if service is running
167 * on machine with hundreds of cores and tens of
168 * CPU partitions, we need to guarantee each partition
169 * has ?_NTHRS_BASE threads, which means total threads
170 * will be ?_NTHRS_BASE * number_of_cpts which can
171 * exceed ?_NTHRS_MAX.
175 * #define MDS_NTHRS_INIT 2
176 * #define MDS_NTHRS_BASE 64
177 * #define MDS_NTHRS_FACTOR 8
178 * #define MDS_NTHRS_MAX 1024
181 * ---------------------------------------------------------------------
182 * Server(A) has 16 cores, user configured it to 4 partitions so each
183 * partition has 4 cores, then actual number of service threads on each
185 * MDS_NTHRS_BASE(64) + cores(4) * MDS_NTHRS_FACTOR(8) = 96
187 * Total number of threads for the service is:
188 * 96 * partitions(4) = 384
191 * ---------------------------------------------------------------------
192 * Server(B) has 32 cores, user configured it to 4 partitions so each
193 * partition has 8 cores, then actual number of service threads on each
195 * MDS_NTHRS_BASE(64) + cores(8) * MDS_NTHRS_FACTOR(8) = 128
197 * Total number of threads for the service is:
198 * 128 * partitions(4) = 512
201 * ---------------------------------------------------------------------
202 * Server(B) has 96 cores, user configured it to 8 partitions so each
203 * partition has 12 cores, then actual number of service threads on each
205 * MDS_NTHRS_BASE(64) + cores(12) * MDS_NTHRS_FACTOR(8) = 160
207 * Total number of threads for the service is:
208 * 160 * partitions(8) = 1280
210 * However, it's above the soft limit MDS_NTHRS_MAX, so we choose this number
211 * as upper limit of threads number for each partition:
212 * MDS_NTHRS_MAX(1024) / partitions(8) = 128
215 * ---------------------------------------------------------------------
216 * Server(C) have a thousand of cores and user configured it to 32 partitions
217 * MDS_NTHRS_BASE(64) * 32 = 2048
219 * which is already above soft limit MDS_NTHRS_MAX(1024), but we still need
220 * to guarantee that each partition has at least MDS_NTHRS_BASE(64) threads
221 * to keep service healthy, so total number of threads will just be 2048.
223 * NB: we don't suggest to choose server with that many cores because backend
224 * filesystem itself, buffer cache, or underlying network stack might
225 * have some SMP scalability issues at that large scale.
227 * If user already has a fat machine with hundreds or thousands of cores,
228 * there are two choices for configuration:
229 * a) create CPU table from subset of all CPUs and run Lustre on
231 * b) bind service threads on a few partitions, see modparameters of
232 * MDS and OSS for details
234 * NB: these calculations (and examples below) are simplified to help
235 * understanding, the real implementation is a little more complex,
236 * please see ptlrpc_server_nthreads_check() for details.
241 * LDLM threads constants:
243 * Given 8 as factor and 24 as base threads number
246 * On 4-core machine we will have 24 + 8 * 4 = 56 threads.
249 * On 8-core machine with 2 partitions we will have 24 + 4 * 8 = 56
250 * threads for each partition and total threads number will be 112.
253 * On 64-core machine with 8 partitions we will need LDLM_NTHRS_BASE(24)
254 * threads for each partition to keep service healthy, so total threads
255 * number should be 24 * 8 = 192.
257 * So with these constants, threads number will be at the similar level
258 * of old versions, unless target machine has over a hundred cores
260 #define LDLM_THR_FACTOR 8
261 #define LDLM_NTHRS_INIT PTLRPC_NTHRS_INIT
262 #define LDLM_NTHRS_BASE 24
263 #define LDLM_NTHRS_MAX (num_online_cpus() == 1 ? 64 : 128)
265 #define LDLM_BL_THREADS LDLM_NTHRS_AUTO_INIT
266 #define LDLM_CLIENT_NBUFS 1
267 #define LDLM_SERVER_NBUFS 64
268 #define LDLM_BUFSIZE (8 * 1024)
269 #define LDLM_MAXREQSIZE (5 * 1024)
270 #define LDLM_MAXREPSIZE (1024)
273 * MDS threads constants:
275 * Please see examples in "Thread Constants", MDS threads number will be at
276 * the comparable level of old versions, unless the server has many cores.
278 #ifndef MDS_MAX_THREADS
279 #define MDS_MAX_THREADS 1024
280 #define MDS_MAX_OTHR_THREADS 256
282 #else /* MDS_MAX_THREADS */
283 #if MDS_MAX_THREADS < PTLRPC_NTHRS_INIT
284 #undef MDS_MAX_THREADS
285 #define MDS_MAX_THREADS PTLRPC_NTHRS_INIT
287 #define MDS_MAX_OTHR_THREADS max(PTLRPC_NTHRS_INIT, MDS_MAX_THREADS / 2)
290 /* default service */
291 #define MDS_THR_FACTOR 8
292 #define MDS_NTHRS_INIT PTLRPC_NTHRS_INIT
293 #define MDS_NTHRS_MAX MDS_MAX_THREADS
294 #define MDS_NTHRS_BASE min(64, MDS_NTHRS_MAX)
296 /* read-page service */
297 #define MDS_RDPG_THR_FACTOR 4
298 #define MDS_RDPG_NTHRS_INIT PTLRPC_NTHRS_INIT
299 #define MDS_RDPG_NTHRS_MAX MDS_MAX_OTHR_THREADS
300 #define MDS_RDPG_NTHRS_BASE min(48, MDS_RDPG_NTHRS_MAX)
302 /* these should be removed when we remove setattr service in the future */
303 #define MDS_SETA_THR_FACTOR 4
304 #define MDS_SETA_NTHRS_INIT PTLRPC_NTHRS_INIT
305 #define MDS_SETA_NTHRS_MAX MDS_MAX_OTHR_THREADS
306 #define MDS_SETA_NTHRS_BASE min(48, MDS_SETA_NTHRS_MAX)
308 /* non-affinity threads */
309 #define MDS_OTHR_NTHRS_INIT PTLRPC_NTHRS_INIT
310 #define MDS_OTHR_NTHRS_MAX MDS_MAX_OTHR_THREADS
315 * Assume file name length = FNAME_MAX = 256 (true for ext3).
316 * path name length = PATH_MAX = 4096
317 * LOV MD size max = EA_MAX = 24 * 2000
318 * (NB: 24 is size of lov_ost_data)
319 * LOV LOGCOOKIE size max = 32 * 2000
320 * (NB: 32 is size of llog_cookie)
321 * symlink: FNAME_MAX + PATH_MAX <- largest
322 * link: FNAME_MAX + PATH_MAX (mds_rec_link < mds_rec_create)
323 * rename: FNAME_MAX + FNAME_MAX
324 * open: FNAME_MAX + EA_MAX
326 * MDS_MAXREQSIZE ~= 4736 bytes =
327 * lustre_msg + ldlm_request + mdt_body + mds_rec_create + FNAME_MAX + PATH_MAX
328 * MDS_MAXREPSIZE ~= 8300 bytes = lustre_msg + llog_header
330 * Realistic size is about 512 bytes (20 character name + 128 char symlink),
331 * except in the open case where there are a large number of OSTs in a LOV.
333 #define MDS_MAXREQSIZE (5 * 1024) /* >= 4736 */
334 #define MDS_MAXREPSIZE (9 * 1024) /* >= 8300 */
337 * MDS incoming request with LOV EA
338 * 24 = sizeof(struct lov_ost_data), i.e: replay of opencreate
340 #define MDS_LOV_MAXREQSIZE max(MDS_MAXREQSIZE, \
341 362 + LOV_MAX_STRIPE_COUNT * 24)
343 * MDS outgoing reply with LOV EA
345 * NB: max reply size Lustre 2.4+ client can get from old MDS is:
346 * LOV_MAX_STRIPE_COUNT * (llog_cookie + lov_ost_data) + extra bytes
348 * but 2.4 or later MDS will never send reply with llog_cookie to any
349 * version client. This macro is defined for server side reply buffer size.
351 #define MDS_LOV_MAXREPSIZE MDS_LOV_MAXREQSIZE
354 * This is the size of a maximum REINT_SETXATTR request:
356 * lustre_msg 56 (32 + 4 x 5 + 4)
358 * mdt_rec_setxattr 136
360 * name 256 (XATTR_NAME_MAX)
361 * value 65536 (XATTR_SIZE_MAX)
363 #define MDS_EA_MAXREQSIZE 66288
366 * These are the maximum request and reply sizes (rounded up to 1 KB
367 * boundaries) for the "regular" MDS_REQUEST_PORTAL and MDS_REPLY_PORTAL.
369 #define MDS_REG_MAXREQSIZE (((max(MDS_EA_MAXREQSIZE, \
370 MDS_LOV_MAXREQSIZE) + 1023) >> 10) << 10)
371 #define MDS_REG_MAXREPSIZE MDS_REG_MAXREQSIZE
374 * The update request includes all of updates from the create, which might
375 * include linkea (4K maxim), together with other updates, we set it to 1000K:
376 * lustre_msg + ptlrpc_body + OUT_UPDATE_BUFFER_SIZE_MAX
378 #define OUT_MAXREQSIZE (1000 * 1024)
379 #define OUT_MAXREPSIZE MDS_MAXREPSIZE
381 /** MDS_BUFSIZE = max_reqsize (w/o LOV EA) + max sptlrpc payload size */
382 #define MDS_BUFSIZE max(MDS_MAXREQSIZE + SPTLRPC_MAX_PAYLOAD, \
386 * MDS_REG_BUFSIZE should at least be MDS_REG_MAXREQSIZE + SPTLRPC_MAX_PAYLOAD.
387 * However, we need to allocate a much larger buffer for it because LNet
388 * requires each MD(rqbd) has at least MDS_REQ_MAXREQSIZE bytes left to avoid
389 * dropping of maximum-sized incoming request. So if MDS_REG_BUFSIZE is only a
390 * little larger than MDS_REG_MAXREQSIZE, then it can only fit in one request
391 * even there are about MDS_REG_MAX_REQSIZE bytes left in a rqbd, and memory
392 * utilization is very low.
394 * In the meanwhile, size of rqbd can't be too large, because rqbd can't be
395 * reused until all requests fit in it have been processed and released,
396 * which means one long blocked request can prevent the rqbd be reused.
397 * Now we set request buffer size to 160 KB, so even each rqbd is unlinked
398 * from LNet with unused 65 KB, buffer utilization will be about 59%.
399 * Please check LU-2432 for details.
401 #define MDS_REG_BUFSIZE max(MDS_REG_MAXREQSIZE + SPTLRPC_MAX_PAYLOAD, \
405 * OUT_BUFSIZE = max_out_reqsize + max sptlrpc payload (~1K) which is
406 * about 10K, for the same reason as MDS_REG_BUFSIZE, we also give some
407 * extra bytes to each request buffer to improve buffer utilization rate.
409 #define OUT_BUFSIZE max(OUT_MAXREQSIZE + SPTLRPC_MAX_PAYLOAD, \
412 /** FLD_MAXREQSIZE == lustre_msg + __u32 padding + ptlrpc_body + opc */
413 #define FLD_MAXREQSIZE (160)
415 /** FLD_MAXREPSIZE == lustre_msg + ptlrpc_body */
416 #define FLD_MAXREPSIZE (152)
417 #define FLD_BUFSIZE (1 << 12)
420 * SEQ_MAXREQSIZE == lustre_msg + __u32 padding + ptlrpc_body + opc + lu_range +
422 #define SEQ_MAXREQSIZE (160)
424 /** SEQ_MAXREPSIZE == lustre_msg + ptlrpc_body + lu_range */
425 #define SEQ_MAXREPSIZE (152)
426 #define SEQ_BUFSIZE (1 << 12)
428 /** MGS threads must be >= 3, see bug 22458 comment #28 */
429 #define MGS_NTHRS_INIT (PTLRPC_NTHRS_INIT + 1)
430 #define MGS_NTHRS_MAX 32
433 #define MGS_BUFSIZE (8 * 1024)
434 #define MGS_MAXREQSIZE (7 * 1024)
435 #define MGS_MAXREPSIZE (9 * 1024)
438 * OSS threads constants:
440 * Given 8 as factor and 64 as base threads number
443 * On 8-core server configured to 2 partitions, we will have
444 * 64 + 8 * 4 = 96 threads for each partition, 192 total threads.
447 * On 32-core machine configured to 4 partitions, we will have
448 * 64 + 8 * 8 = 112 threads for each partition, so total threads number
449 * will be 112 * 4 = 448.
452 * On 64-core machine configured to 4 partitions, we will have
453 * 64 + 16 * 8 = 192 threads for each partition, so total threads number
454 * will be 192 * 4 = 768 which is above limit OSS_NTHRS_MAX(512), so we
455 * cut off the value to OSS_NTHRS_MAX(512) / 4 which is 128 threads
456 * for each partition.
458 * So we can see that with these constants, threads number wil be at the
459 * similar level of old versions, unless the server has many cores.
461 /* depress threads factor for VM with small memory size */
462 #define OSS_THR_FACTOR min_t(int, 8, \
463 NUM_CACHEPAGES >> (28 - PAGE_SHIFT))
464 #define OSS_NTHRS_INIT (PTLRPC_NTHRS_INIT + 1)
465 #define OSS_NTHRS_BASE 64
467 /* threads for handling "create" request */
468 #define OSS_CR_THR_FACTOR 1
469 #define OSS_CR_NTHRS_INIT PTLRPC_NTHRS_INIT
470 #define OSS_CR_NTHRS_BASE 8
471 #define OSS_CR_NTHRS_MAX 64
474 * OST_IO_MAXREQSIZE ~=
475 * lustre_msg + ptlrpc_body + obdo + obd_ioobj +
476 * DT_MAX_BRW_PAGES * niobuf_remote
478 * - single object with 16 pages is 512 bytes
479 * - OST_IO_MAXREQSIZE must be at least 1 page of cookies plus some spillover
480 * - Must be a multiple of 1024
482 #define _OST_MAXREQSIZE_BASE (sizeof(struct lustre_msg) + \
483 sizeof(struct ptlrpc_body) + \
484 sizeof(struct obdo) + \
485 sizeof(struct obd_ioobj) + \
486 sizeof(struct niobuf_remote))
487 #define _OST_MAXREQSIZE_SUM (_OST_MAXREQSIZE_BASE + \
488 sizeof(struct niobuf_remote) * \
489 (DT_MAX_BRW_PAGES - 1))
491 * FIEMAP request can be 4K+ for now
493 #define OST_MAXREQSIZE (16 * 1024)
494 #define OST_IO_MAXREQSIZE max_t(int, OST_MAXREQSIZE, \
495 (((_OST_MAXREQSIZE_SUM - 1) | (1024 - 1)) + 1))
496 /* Safe estimate of free space in standard RPC, provides upper limit for # of
497 * bytes of i/o to pack in RPC (skipping bulk transfer). */
498 #define OST_SHORT_IO_SPACE (OST_IO_MAXREQSIZE - _OST_MAXREQSIZE_BASE)
500 /* Actual size used for short i/o buffer. Calculation means this:
501 * At least one page (for large PAGE_SIZE), or 16 KiB, but not more
502 * than the available space aligned to a page boundary. */
503 #define OBD_MAX_SHORT_IO_BYTES (min(max(PAGE_SIZE, 16UL * 1024UL), \
504 OST_SHORT_IO_SPACE & PAGE_MASK))
506 #define OST_MAXREPSIZE (9 * 1024)
507 #define OST_IO_MAXREPSIZE OST_MAXREPSIZE
510 /** OST_BUFSIZE = max_reqsize + max sptlrpc payload size */
511 #define OST_BUFSIZE max_t(int, OST_MAXREQSIZE + 1024, 16 * 1024)
513 * OST_IO_MAXREQSIZE is 18K, giving extra 46K can increase buffer utilization
514 * rate of request buffer, please check comment of MDS_LOV_BUFSIZE for details.
516 #define OST_IO_BUFSIZE max_t(int, OST_IO_MAXREQSIZE + 1024, 64 * 1024)
519 /* Macro to hide a typecast. */
520 #define ptlrpc_req_async_args(req) ((void *)&req->rq_async_args)
522 struct ptlrpc_replay_async_args {
528 * Structure to single define portal connection.
530 struct ptlrpc_connection {
531 /** linkage for connections hash table */
532 struct rhash_head c_hash;
533 /** Our own lnet nid for this connection */
535 /** Remote side nid for this connection */
536 struct lnet_process_id c_peer;
537 /** UUID of the other side */
538 struct obd_uuid c_remote_uuid;
539 /** reference counter for this connection */
543 /** Client definition for PortalRPC */
544 struct ptlrpc_client {
545 /** What lnet portal does this client send messages to by default */
546 __u32 cli_request_portal;
547 /** What portal do we expect replies on */
548 __u32 cli_reply_portal;
549 /** Name of the client */
553 /** state flags of requests */
554 /* XXX only ones left are those used by the bulk descs as well! */
555 #define PTL_RPC_FL_INTR (1 << 0) /* reply wait was interrupted by user */
556 #define PTL_RPC_FL_TIMEOUT (1 << 7) /* request timed out waiting for reply */
558 #define REQ_MAX_ACK_LOCKS 8
560 union ptlrpc_async_args {
562 * Scratchpad for passing args to completion interpreter. Users
563 * cast to the struct of their choosing, and CLASSERT that this is
564 * big enough. For _tons_ of context, OBD_ALLOC a struct and store
565 * a pointer to it here. The pointer_arg ensures this struct is at
566 * least big enough for that.
568 void *pointer_arg[11];
572 struct ptlrpc_request_set;
573 typedef int (*set_producer_func)(struct ptlrpc_request_set *, void *);
576 * Definition of request set structure.
577 * Request set is a list of requests (not necessary to the same target) that
578 * once populated with RPCs could be sent in parallel.
579 * There are two kinds of request sets. General purpose and with dedicated
580 * serving thread. Example of the latter is ptlrpcd set.
581 * For general purpose sets once request set started sending it is impossible
582 * to add new requests to such set.
583 * Provides a way to call "completion callbacks" when all requests in the set
586 struct ptlrpc_request_set {
587 atomic_t set_refcount;
588 /** number of in queue requests */
589 atomic_t set_new_count;
590 /** number of uncompleted requests */
591 atomic_t set_remaining;
592 /** wait queue to wait on for request events */
593 wait_queue_head_t set_waitq;
594 /** List of requests in the set */
595 struct list_head set_requests;
597 * Lock for \a set_new_requests manipulations
598 * locked so that any old caller can communicate requests to
599 * the set holder who can then fold them into the lock-free set
601 spinlock_t set_new_req_lock;
602 /** List of new yet unsent requests. Only used with ptlrpcd now. */
603 struct list_head set_new_requests;
605 /** rq_status of requests that have been freed already */
607 /** Additional fields used by the flow control extension */
608 /** Maximum number of RPCs in flight */
609 int set_max_inflight;
610 /** Callback function used to generate RPCs */
611 set_producer_func set_producer;
612 /** opaq argument passed to the producer callback */
613 void *set_producer_arg;
614 unsigned int set_allow_intr:1;
617 struct ptlrpc_bulk_desc;
618 struct ptlrpc_service_part;
619 struct ptlrpc_service;
622 * ptlrpc callback & work item stuff
624 struct ptlrpc_cb_id {
625 void (*cbid_fn)(struct lnet_event *ev); /* specific callback fn */
626 void *cbid_arg; /* additional arg */
629 /** Maximum number of locks to fit into reply state */
630 #define RS_MAX_LOCKS 8
634 * Structure to define reply state on the server
635 * Reply state holds various reply message information. Also for "difficult"
636 * replies (rep-ack case) we store the state after sending reply and wait
637 * for the client to acknowledge the reception. In these cases locks could be
638 * added to the state for replay/failover consistency guarantees.
640 struct ptlrpc_reply_state {
641 /** Callback description */
642 struct ptlrpc_cb_id rs_cb_id;
643 /** Linkage for list of all reply states in a system */
644 struct list_head rs_list;
645 /** Linkage for list of all reply states on same export */
646 struct list_head rs_exp_list;
647 /** Linkage for list of all reply states for same obd */
648 struct list_head rs_obd_list;
650 struct list_head rs_debug_list;
652 /** A spinlock to protect the reply state flags */
654 /** Reply state flags */
655 unsigned long rs_difficult:1; /* ACK/commit stuff */
656 unsigned long rs_no_ack:1; /* no ACK, even for
657 difficult requests */
658 unsigned long rs_scheduled:1; /* being handled? */
659 unsigned long rs_scheduled_ever:1;/* any schedule attempts? */
660 unsigned long rs_handled:1; /* been handled yet? */
661 unsigned long rs_on_net:1; /* reply_out_callback pending? */
662 unsigned long rs_prealloc:1; /* rs from prealloc list */
663 unsigned long rs_committed:1;/* the transaction was committed
664 and the rs was dispatched
665 by ptlrpc_commit_replies */
666 unsigned long rs_convert_lock:1; /* need to convert saved
667 * locks to COS mode */
668 atomic_t rs_refcount; /* number of users */
669 /** Number of locks awaiting client ACK */
672 /** Size of the state */
676 /** Transaction number */
680 struct obd_export *rs_export;
681 struct ptlrpc_service_part *rs_svcpt;
682 /** Lnet metadata handle for the reply */
683 struct lnet_handle_md rs_md_h;
685 /** Context for the sevice thread */
686 struct ptlrpc_svc_ctx *rs_svc_ctx;
687 /** Reply buffer (actually sent to the client), encoded if needed */
688 struct lustre_msg *rs_repbuf; /* wrapper */
689 /** Size of the reply buffer */
690 int rs_repbuf_len; /* wrapper buf length */
691 /** Size of the reply message */
692 int rs_repdata_len; /* wrapper msg length */
694 * Actual reply message. Its content is encrupted (if needed) to
695 * produce reply buffer for actual sending. In simple case
696 * of no network encryption we jus set \a rs_repbuf to \a rs_msg
698 struct lustre_msg *rs_msg; /* reply message */
700 /** Handles of locks awaiting client reply ACK */
701 struct lustre_handle rs_locks[RS_MAX_LOCKS];
702 /** Lock modes of locks in \a rs_locks */
703 enum ldlm_mode rs_modes[RS_MAX_LOCKS];
706 struct ptlrpc_thread;
710 RQ_PHASE_NEW = 0xebc0de00,
711 RQ_PHASE_RPC = 0xebc0de01,
712 RQ_PHASE_BULK = 0xebc0de02,
713 RQ_PHASE_INTERPRET = 0xebc0de03,
714 RQ_PHASE_COMPLETE = 0xebc0de04,
715 RQ_PHASE_UNREG_RPC = 0xebc0de05,
716 RQ_PHASE_UNREG_BULK = 0xebc0de06,
717 RQ_PHASE_UNDEFINED = 0xebc0de07
720 /** Type of request interpreter call-back */
721 typedef int (*ptlrpc_interpterer_t)(const struct lu_env *env,
722 struct ptlrpc_request *req,
724 /** Type of request resend call-back */
725 typedef void (*ptlrpc_resend_cb_t)(struct ptlrpc_request *req,
729 * Definition of request pool structure.
730 * The pool is used to store empty preallocated requests for the case
731 * when we would actually need to send something without performing
732 * any allocations (to avoid e.g. OOM).
734 struct ptlrpc_request_pool {
735 /** Locks the list */
737 /** list of ptlrpc_request structs */
738 struct list_head prp_req_list;
739 /** Maximum message size that would fit into a rquest from this pool */
741 /** Function to allocate more requests for this pool */
742 int (*prp_populate)(struct ptlrpc_request_pool *, int);
750 #include <lustre_nrs.h>
753 * Basic request prioritization operations structure.
754 * The whole idea is centered around locks and RPCs that might affect locks.
755 * When a lock is contended we try to give priority to RPCs that might lead
756 * to fastest release of that lock.
757 * Currently only implemented for OSTs only in a way that makes all
758 * IO and truncate RPCs that are coming from a locked region where a lock is
759 * contended a priority over other requests.
761 struct ptlrpc_hpreq_ops {
763 * Check if the lock handle of the given lock is the same as
764 * taken from the request.
766 int (*hpreq_lock_match)(struct ptlrpc_request *, struct ldlm_lock *);
768 * Check if the request is a high priority one.
770 int (*hpreq_check)(struct ptlrpc_request *);
772 * Called after the request has been handled.
774 void (*hpreq_fini)(struct ptlrpc_request *);
777 struct ptlrpc_cli_req {
778 /** For bulk requests on client only: bulk descriptor */
779 struct ptlrpc_bulk_desc *cr_bulk;
780 /** optional time limit for send attempts */
781 time64_t cr_delay_limit;
782 /** time request was first queued */
783 time64_t cr_queued_time;
784 /** request sent in nanoseconds */
786 /** time for request really sent out */
787 time64_t cr_sent_out;
788 /** when req reply unlink must finish. */
789 time64_t cr_reply_deadline;
790 /** when req bulk unlink must finish. */
791 time64_t cr_bulk_deadline;
792 /** when req unlink must finish. */
793 time64_t cr_req_deadline;
794 /** Portal to which this request would be sent */
796 /** Portal where to wait for reply and where reply would be sent */
798 /** request resending number */
799 unsigned int cr_resend_nr;
800 /** What was import generation when this request was sent */
802 enum lustre_imp_state cr_send_state;
803 /** Per-request waitq introduced by bug 21938 for recovery waiting */
804 wait_queue_head_t cr_set_waitq;
805 /** Link item for request set lists */
806 struct list_head cr_set_chain;
807 /** link to waited ctx */
808 struct list_head cr_ctx_chain;
810 /** client's half ctx */
811 struct ptlrpc_cli_ctx *cr_cli_ctx;
812 /** Link back to the request set */
813 struct ptlrpc_request_set *cr_set;
814 /** outgoing request MD handle */
815 struct lnet_handle_md cr_req_md_h;
816 /** request-out callback parameter */
817 struct ptlrpc_cb_id cr_req_cbid;
818 /** incoming reply MD handle */
819 struct lnet_handle_md cr_reply_md_h;
820 wait_queue_head_t cr_reply_waitq;
821 /** reply callback parameter */
822 struct ptlrpc_cb_id cr_reply_cbid;
823 /** Async completion handler, called when reply is received */
824 ptlrpc_interpterer_t cr_reply_interp;
825 /** Resend handler, called when request is resend to update RPC data */
826 ptlrpc_resend_cb_t cr_resend_cb;
827 /** Async completion context */
828 union ptlrpc_async_args cr_async_args;
829 /** Opaq data for replay and commit callbacks. */
831 /** Link to the imp->imp_unreplied_list */
832 struct list_head cr_unreplied_list;
834 * Commit callback, called when request is committed and about to be
837 void (*cr_commit_cb)(struct ptlrpc_request *);
838 /** Replay callback, called after request is replayed at recovery */
839 void (*cr_replay_cb)(struct ptlrpc_request *);
842 /** client request member alias */
843 /* NB: these alias should NOT be used by any new code, instead they should
844 * be removed step by step to avoid potential abuse */
845 #define rq_bulk rq_cli.cr_bulk
846 #define rq_delay_limit rq_cli.cr_delay_limit
847 #define rq_queued_time rq_cli.cr_queued_time
848 #define rq_sent_ns rq_cli.cr_sent_ns
849 #define rq_real_sent rq_cli.cr_sent_out
850 #define rq_reply_deadline rq_cli.cr_reply_deadline
851 #define rq_bulk_deadline rq_cli.cr_bulk_deadline
852 #define rq_req_deadline rq_cli.cr_req_deadline
853 #define rq_nr_resend rq_cli.cr_resend_nr
854 #define rq_request_portal rq_cli.cr_req_ptl
855 #define rq_reply_portal rq_cli.cr_rep_ptl
856 #define rq_import_generation rq_cli.cr_imp_gen
857 #define rq_send_state rq_cli.cr_send_state
858 #define rq_set_chain rq_cli.cr_set_chain
859 #define rq_ctx_chain rq_cli.cr_ctx_chain
860 #define rq_set rq_cli.cr_set
861 #define rq_set_waitq rq_cli.cr_set_waitq
862 #define rq_cli_ctx rq_cli.cr_cli_ctx
863 #define rq_req_md_h rq_cli.cr_req_md_h
864 #define rq_req_cbid rq_cli.cr_req_cbid
865 #define rq_reply_md_h rq_cli.cr_reply_md_h
866 #define rq_reply_waitq rq_cli.cr_reply_waitq
867 #define rq_reply_cbid rq_cli.cr_reply_cbid
868 #define rq_interpret_reply rq_cli.cr_reply_interp
869 #define rq_resend_cb rq_cli.cr_resend_cb
870 #define rq_async_args rq_cli.cr_async_args
871 #define rq_cb_data rq_cli.cr_cb_data
872 #define rq_unreplied_list rq_cli.cr_unreplied_list
873 #define rq_commit_cb rq_cli.cr_commit_cb
874 #define rq_replay_cb rq_cli.cr_replay_cb
876 struct ptlrpc_srv_req {
877 /** initial thread servicing this request */
878 struct ptlrpc_thread *sr_svc_thread;
880 * Server side list of incoming unserved requests sorted by arrival
881 * time. Traversed from time to time to notice about to expire
882 * requests and sent back "early replies" to clients to let them
883 * know server is alive and well, just very busy to service their
886 struct list_head sr_timed_list;
887 /** server-side per-export list */
888 struct list_head sr_exp_list;
889 /** server-side history, used for debuging purposes. */
890 struct list_head sr_hist_list;
891 /** history sequence # */
893 /** the index of service's srv_at_array into which request is linked */
897 /** authed uid mapped to */
898 uid_t sr_auth_mapped_uid;
899 /** RPC is generated from what part of Lustre */
900 enum lustre_sec_part sr_sp_from;
901 /** request session context */
902 struct lu_context sr_ses;
906 /** stub for NRS request */
907 struct ptlrpc_nrs_request sr_nrq;
909 /** request arrival time */
910 struct timespec64 sr_arrival_time;
911 /** server's half ctx */
912 struct ptlrpc_svc_ctx *sr_svc_ctx;
913 /** (server side), pointed directly into req buffer */
914 struct ptlrpc_user_desc *sr_user_desc;
915 /** separated reply state, may be vmalloc'd */
916 struct ptlrpc_reply_state *sr_reply_state;
917 /** server-side hp handlers */
918 struct ptlrpc_hpreq_ops *sr_ops;
919 /** incoming request buffer */
920 struct ptlrpc_request_buffer_desc *sr_rqbd;
923 /** server request member alias */
924 /* NB: these alias should NOT be used by any new code, instead they should
925 * be removed step by step to avoid potential abuse */
926 #define rq_svc_thread rq_srv.sr_svc_thread
927 #define rq_timed_list rq_srv.sr_timed_list
928 #define rq_exp_list rq_srv.sr_exp_list
929 #define rq_history_list rq_srv.sr_hist_list
930 #define rq_history_seq rq_srv.sr_hist_seq
931 #define rq_at_index rq_srv.sr_at_index
932 #define rq_auth_uid rq_srv.sr_auth_uid
933 #define rq_auth_mapped_uid rq_srv.sr_auth_mapped_uid
934 #define rq_sp_from rq_srv.sr_sp_from
935 #define rq_session rq_srv.sr_ses
936 #define rq_nrq rq_srv.sr_nrq
937 #define rq_arrival_time rq_srv.sr_arrival_time
938 #define rq_reply_state rq_srv.sr_reply_state
939 #define rq_svc_ctx rq_srv.sr_svc_ctx
940 #define rq_user_desc rq_srv.sr_user_desc
941 #define rq_ops rq_srv.sr_ops
942 #define rq_rqbd rq_srv.sr_rqbd
945 * Represents remote procedure call.
947 * This is a staple structure used by everybody wanting to send a request
950 struct ptlrpc_request {
951 /* Request type: one of PTL_RPC_MSG_* */
953 /** Result of request processing */
956 * Linkage item through which this request is included into
957 * sending/delayed lists on client and into rqbd list on server
959 struct list_head rq_list;
960 /** Lock to protect request flags and some other important bits, like
964 spinlock_t rq_early_free_lock;
965 /** client-side flags are serialized by rq_lock @{ */
966 unsigned int rq_intr:1, rq_replied:1, rq_err:1,
967 rq_timedout:1, rq_resend:1, rq_restart:1,
969 * when ->rq_replay is set, request is kept by the client even
970 * after server commits corresponding transaction. This is
971 * used for operations that require sequence of multiple
972 * requests to be replayed. The only example currently is file
973 * open/close. When last request in such a sequence is
974 * committed, ->rq_replay is cleared on all requests in the
978 rq_no_resend:1, rq_waiting:1, rq_receiving_reply:1,
979 rq_no_delay:1, rq_net_err:1, rq_wait_ctx:1,
981 rq_req_unlinked:1, /* unlinked request buffer from lnet */
982 rq_reply_unlinked:1, /* unlinked reply buffer from lnet */
983 rq_memalloc:1, /* req originated from "kswapd" */
985 rq_reply_truncated:1,
986 /** whether the "rq_set" is a valid one */
989 /** do not resend request on -EINPROGRESS */
990 rq_no_retry_einprogress:1,
991 /* allow the req to be sent if the import is in recovery
994 /* bulk request, sent to server, but uncommitted */
996 rq_early_free_repbuf:1, /* free reply buffer in advance */
1000 /** server-side flags @{ */
1002 rq_hp:1, /**< high priority RPC */
1003 rq_at_linked:1, /**< link into service's srv_at_array */
1004 rq_packed_final:1; /**< packed final reply */
1007 /** one of RQ_PHASE_* */
1008 enum rq_phase rq_phase;
1009 /** one of RQ_PHASE_* to be used next */
1010 enum rq_phase rq_next_phase;
1012 * client-side refcount for SENT race, server-side refcounf
1013 * for multiple replies
1015 atomic_t rq_refcount;
1018 * !rq_truncate : # reply bytes actually received,
1019 * rq_truncate : required repbuf_len for resend
1021 int rq_nob_received;
1022 /** Request length */
1026 /** Pool if request is from preallocated list */
1027 struct ptlrpc_request_pool *rq_pool;
1028 /** Request message - what client sent */
1029 struct lustre_msg *rq_reqmsg;
1030 /** Reply message - server response */
1031 struct lustre_msg *rq_repmsg;
1032 /** Transaction number */
1036 /** bulk match bits */
1039 * List item to for replay list. Not yet committed requests get linked
1041 * Also see \a rq_replay comment above.
1042 * It's also link chain on obd_export::exp_req_replay_queue
1044 struct list_head rq_replay_list;
1045 /** non-shared members for client & server request*/
1047 struct ptlrpc_cli_req rq_cli;
1048 struct ptlrpc_srv_req rq_srv;
1051 * security and encryption data
1053 /** description of flavors for client & server */
1054 struct sptlrpc_flavor rq_flvr;
1056 /* client/server security flags */
1058 rq_ctx_init:1, /* context initiation */
1059 rq_ctx_fini:1, /* context destroy */
1060 rq_bulk_read:1, /* request bulk read */
1061 rq_bulk_write:1, /* request bulk write */
1062 /* server authentication flags */
1063 rq_auth_gss:1, /* authenticated by gss */
1064 rq_auth_usr_root:1, /* authed as root */
1065 rq_auth_usr_mdt:1, /* authed as mdt */
1066 rq_auth_usr_ost:1, /* authed as ost */
1067 /* security tfm flags */
1070 /* doesn't expect reply FIXME */
1072 rq_pill_init:1, /* pill initialized */
1073 rq_srv_req:1; /* server request */
1076 /** various buffer pointers */
1077 struct lustre_msg *rq_reqbuf; /**< req wrapper, vmalloc*/
1078 char *rq_repbuf; /**< rep buffer, vmalloc */
1079 struct lustre_msg *rq_repdata; /**< rep wrapper msg */
1080 /** only in priv mode */
1081 struct lustre_msg *rq_clrbuf;
1082 int rq_reqbuf_len; /* req wrapper buf len */
1083 int rq_reqdata_len; /* req wrapper msg len */
1084 int rq_repbuf_len; /* rep buffer len */
1085 int rq_repdata_len; /* rep wrapper msg len */
1086 int rq_clrbuf_len; /* only in priv mode */
1087 int rq_clrdata_len; /* only in priv mode */
1089 /** early replies go to offset 0, regular replies go after that */
1090 unsigned int rq_reply_off;
1093 /** Fields that help to see if request and reply were swabbed or not */
1094 __u32 rq_req_swab_mask;
1095 __u32 rq_rep_swab_mask;
1097 /** how many early replies (for stats) */
1099 /** Server-side, export on which request was received */
1100 struct obd_export *rq_export;
1101 /** import where request is being sent */
1102 struct obd_import *rq_import;
1105 /** Peer description (the other side) */
1106 struct lnet_process_id rq_peer;
1107 /** Descriptor for the NID from which the peer sent the request. */
1108 struct lnet_process_id rq_source;
1110 * service time estimate (secs)
1111 * If the request is not served by this time, it is marked as timed out.
1112 * Do not change to time64_t since this is transmitted over the wire.
1116 * when request/reply sent (secs), or time when request should be sent
1119 /** when request must finish. */
1120 time64_t rq_deadline;
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 true if request buffer at offset \a index was already swabbed
1173 static inline bool 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 true if request reply buffer at offset \a index was already swabbed
1182 static inline bool 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 true if request needs to be swabbed into local cpu byteorder
1191 static inline bool ptlrpc_req_need_swab(struct ptlrpc_request *req)
1193 return lustre_req_swabbed(req, MSG_PTLRPC_HEADER_OFF);
1197 * Returns true if request reply needs to be swabbed into local cpu byteorder
1199 static inline bool 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 debugfs dir tree for this service */
1692 struct dentry *srv_debugfs_entry;
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;
1718 /** Whether or not to restrict service threads to CPUs in this CPT */
1719 unsigned srv_cpt_bind:1;
1721 /** max # request buffers */
1723 /** max # request buffers in history per partition */
1724 int srv_hist_nrqbds_cpt_max;
1725 /** number of CPTs this service associated with */
1727 /** CPTs array this service associated with */
1729 /** 2^srv_cptab_bits >= cfs_cpt_numbert(srv_cptable) */
1731 /** CPT table this service is running over */
1732 struct cfs_cpt_table *srv_cptable;
1735 struct kobject srv_kobj;
1736 struct completion srv_kobj_unregister;
1738 * partition data for ptlrpc service
1740 struct ptlrpc_service_part *srv_parts[0];
1744 * Definition of PortalRPC service partition data.
1745 * Although a service only has one instance of it right now, but we
1746 * will have multiple instances very soon (instance per CPT).
1748 * it has four locks:
1750 * serialize operations on rqbd and requests waiting for preprocess
1752 * serialize operations active requests sent to this portal
1754 * serialize adaptive timeout stuff
1756 * serialize operations on RS list (reply states)
1758 * We don't have any use-case to take two or more locks at the same time
1759 * for now, so there is no lock order issue.
1761 struct ptlrpc_service_part {
1762 /** back reference to owner */
1763 struct ptlrpc_service *scp_service __cfs_cacheline_aligned;
1764 /* CPT id, reserved */
1766 /** always increasing number */
1768 /** # of starting threads */
1769 int scp_nthrs_starting;
1770 /** # of stopping threads, reserved for shrinking threads */
1771 int scp_nthrs_stopping;
1772 /** # running threads */
1773 int scp_nthrs_running;
1774 /** service threads list */
1775 struct list_head scp_threads;
1778 * serialize the following fields, used for protecting
1779 * rqbd list and incoming requests waiting for preprocess,
1780 * threads starting & stopping are also protected by this lock.
1782 spinlock_t scp_lock __cfs_cacheline_aligned;
1783 /** userland serialization */
1784 struct mutex scp_mutex;
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 jiffies */
1800 long 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 ktime_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;
2089 if (req->rq_bulk_deadline > ktime_get_real_seconds())
2093 spin_lock(&desc->bd_lock);
2094 rc = desc->bd_md_count;
2095 spin_unlock(&desc->bd_lock);
2099 #define PTLRPC_REPLY_MAYBE_DIFFICULT 0x01
2100 #define PTLRPC_REPLY_EARLY 0x02
2101 int ptlrpc_send_reply(struct ptlrpc_request *req, int flags);
2102 int ptlrpc_reply(struct ptlrpc_request *req);
2103 int ptlrpc_send_error(struct ptlrpc_request *req, int difficult);
2104 int ptlrpc_error(struct ptlrpc_request *req);
2105 int ptlrpc_at_get_net_latency(struct ptlrpc_request *req);
2106 int ptl_send_rpc(struct ptlrpc_request *request, int noreply);
2107 int ptlrpc_register_rqbd(struct ptlrpc_request_buffer_desc *rqbd);
2110 /* ptlrpc/client.c */
2112 * Client-side portals API. Everything to send requests, receive replies,
2113 * request queues, request management, etc.
2116 void ptlrpc_request_committed(struct ptlrpc_request *req, int force);
2118 void ptlrpc_init_client(int req_portal, int rep_portal, char *name,
2119 struct ptlrpc_client *);
2120 void ptlrpc_cleanup_client(struct obd_import *imp);
2121 struct ptlrpc_connection *ptlrpc_uuid_to_connection(struct obd_uuid *uuid,
2122 lnet_nid_t nid4refnet);
2124 int ptlrpc_queue_wait(struct ptlrpc_request *req);
2125 int ptlrpc_replay_req(struct ptlrpc_request *req);
2126 void ptlrpc_restart_req(struct ptlrpc_request *req);
2127 void ptlrpc_abort_inflight(struct obd_import *imp);
2128 void ptlrpc_cleanup_imp(struct obd_import *imp);
2129 void ptlrpc_abort_set(struct ptlrpc_request_set *set);
2131 struct ptlrpc_request_set *ptlrpc_prep_set(void);
2132 struct ptlrpc_request_set *ptlrpc_prep_fcset(int max, set_producer_func func,
2134 int ptlrpc_check_set(const struct lu_env *env, struct ptlrpc_request_set *set);
2135 int ptlrpc_set_wait(const struct lu_env *env, struct ptlrpc_request_set *);
2136 void ptlrpc_mark_interrupted(struct ptlrpc_request *req);
2137 void ptlrpc_set_destroy(struct ptlrpc_request_set *);
2138 void ptlrpc_set_add_req(struct ptlrpc_request_set *, struct ptlrpc_request *);
2140 void ptlrpc_free_rq_pool(struct ptlrpc_request_pool *pool);
2141 int ptlrpc_add_rqs_to_pool(struct ptlrpc_request_pool *pool, int num_rq);
2143 struct ptlrpc_request_pool *
2144 ptlrpc_init_rq_pool(int, int,
2145 int (*populate_pool)(struct ptlrpc_request_pool *, int));
2147 void ptlrpc_at_set_req_timeout(struct ptlrpc_request *req);
2148 struct ptlrpc_request *ptlrpc_request_alloc(struct obd_import *imp,
2149 const struct req_format *format);
2150 struct ptlrpc_request *ptlrpc_request_alloc_pool(struct obd_import *imp,
2151 struct ptlrpc_request_pool *,
2152 const struct req_format *format);
2153 void ptlrpc_request_free(struct ptlrpc_request *request);
2154 int ptlrpc_request_pack(struct ptlrpc_request *request,
2155 __u32 version, int opcode);
2156 struct ptlrpc_request *ptlrpc_request_alloc_pack(struct obd_import *imp,
2157 const struct req_format *format,
2158 __u32 version, int opcode);
2159 int ptlrpc_request_bufs_pack(struct ptlrpc_request *request,
2160 __u32 version, int opcode, char **bufs,
2161 struct ptlrpc_cli_ctx *ctx);
2162 void ptlrpc_req_finished(struct ptlrpc_request *request);
2163 void ptlrpc_req_finished_with_imp_lock(struct ptlrpc_request *request);
2164 struct ptlrpc_request *ptlrpc_request_addref(struct ptlrpc_request *req);
2165 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_imp(struct ptlrpc_request *req,
2166 unsigned nfrags, unsigned max_brw,
2169 const struct ptlrpc_bulk_frag_ops
2172 int ptlrpc_prep_bulk_frag(struct ptlrpc_bulk_desc *desc,
2173 void *frag, int len);
2174 void __ptlrpc_prep_bulk_page(struct ptlrpc_bulk_desc *desc,
2175 struct page *page, int pageoffset, int len,
2177 static inline void ptlrpc_prep_bulk_page_pin(struct ptlrpc_bulk_desc *desc,
2178 struct page *page, int pageoffset,
2181 __ptlrpc_prep_bulk_page(desc, page, pageoffset, len, 1);
2184 static inline void ptlrpc_prep_bulk_page_nopin(struct ptlrpc_bulk_desc *desc,
2185 struct page *page, int pageoffset,
2188 __ptlrpc_prep_bulk_page(desc, page, pageoffset, len, 0);
2191 void ptlrpc_free_bulk(struct ptlrpc_bulk_desc *bulk);
2193 static inline void ptlrpc_release_bulk_page_pin(struct ptlrpc_bulk_desc *desc)
2197 for (i = 0; i < desc->bd_iov_count ; i++)
2198 put_page(BD_GET_KIOV(desc, i).kiov_page);
2201 static inline void ptlrpc_release_bulk_noop(struct ptlrpc_bulk_desc *desc)
2205 void ptlrpc_retain_replayable_request(struct ptlrpc_request *req,
2206 struct obd_import *imp);
2207 __u64 ptlrpc_next_xid(void);
2208 __u64 ptlrpc_sample_next_xid(void);
2209 __u64 ptlrpc_req_xid(struct ptlrpc_request *request);
2211 /* Set of routines to run a function in ptlrpcd context */
2212 void *ptlrpcd_alloc_work(struct obd_import *imp,
2213 int (*cb)(const struct lu_env *, void *), void *data);
2214 void ptlrpcd_destroy_work(void *handler);
2215 int ptlrpcd_queue_work(void *handler);
2218 struct ptlrpc_service_buf_conf {
2219 /* nbufs is buffers # to allocate when growing the pool */
2220 unsigned int bc_nbufs;
2221 /* buffer size to post */
2222 unsigned int bc_buf_size;
2223 /* portal to listed for requests on */
2224 unsigned int bc_req_portal;
2225 /* portal of where to send replies to */
2226 unsigned int bc_rep_portal;
2227 /* maximum request size to be accepted for this service */
2228 unsigned int bc_req_max_size;
2229 /* maximum reply size this service can ever send */
2230 unsigned int bc_rep_max_size;
2233 struct ptlrpc_service_thr_conf {
2234 /* threadname should be 8 characters or less - 6 will be added on */
2236 /* threads increasing factor for each CPU */
2237 unsigned int tc_thr_factor;
2238 /* service threads # to start on each partition while initializing */
2239 unsigned int tc_nthrs_init;
2241 * low water of threads # upper-limit on each partition while running,
2242 * service availability may be impacted if threads number is lower
2243 * than this value. It can be ZERO if the service doesn't require
2244 * CPU affinity or there is only one partition.
2246 unsigned int tc_nthrs_base;
2247 /* "soft" limit for total threads number */
2248 unsigned int tc_nthrs_max;
2249 /* user specified threads number, it will be validated due to
2250 * other members of this structure. */
2251 unsigned int tc_nthrs_user;
2252 /* bind service threads to only CPUs in their associated CPT */
2253 unsigned int tc_cpu_bind;
2254 /* Tags for lu_context associated with service thread */
2258 struct ptlrpc_service_cpt_conf {
2259 struct cfs_cpt_table *cc_cptable;
2260 /* string pattern to describe CPTs for a service */
2262 /* whether or not to have per-CPT service partitions */
2266 struct ptlrpc_service_conf {
2269 /* soft watchdog timeout multiplifier to print stuck service traces */
2270 unsigned int psc_watchdog_factor;
2271 /* buffer information */
2272 struct ptlrpc_service_buf_conf psc_buf;
2273 /* thread information */
2274 struct ptlrpc_service_thr_conf psc_thr;
2275 /* CPU partition information */
2276 struct ptlrpc_service_cpt_conf psc_cpt;
2277 /* function table */
2278 struct ptlrpc_service_ops psc_ops;
2281 /* ptlrpc/service.c */
2283 * Server-side services API. Register/unregister service, request state
2284 * management, service thread management
2288 void ptlrpc_save_lock(struct ptlrpc_request *req, struct lustre_handle *lock,
2289 int mode, bool no_ack, bool convert_lock);
2290 void ptlrpc_commit_replies(struct obd_export *exp);
2291 void ptlrpc_dispatch_difficult_reply(struct ptlrpc_reply_state *rs);
2292 void ptlrpc_schedule_difficult_reply(struct ptlrpc_reply_state *rs);
2293 int ptlrpc_hpreq_handler(struct ptlrpc_request *req);
2294 struct ptlrpc_service *ptlrpc_register_service(
2295 struct ptlrpc_service_conf *conf,
2296 struct kset *parent,
2297 struct dentry *debugfs_entry);
2298 void ptlrpc_stop_all_threads(struct ptlrpc_service *svc);
2300 int ptlrpc_start_threads(struct ptlrpc_service *svc);
2301 int ptlrpc_unregister_service(struct ptlrpc_service *service);
2302 int ptlrpc_service_health_check(struct ptlrpc_service *);
2303 void ptlrpc_server_drop_request(struct ptlrpc_request *req);
2304 void ptlrpc_request_change_export(struct ptlrpc_request *req,
2305 struct obd_export *export);
2306 void ptlrpc_update_export_timer(struct obd_export *exp,
2307 time64_t extra_delay);
2309 int ptlrpc_hr_init(void);
2310 void ptlrpc_hr_fini(void);
2314 /* ptlrpc/import.c */
2319 int ptlrpc_connect_import(struct obd_import *imp);
2320 int ptlrpc_init_import(struct obd_import *imp);
2321 int ptlrpc_disconnect_import(struct obd_import *imp, int noclose);
2322 int ptlrpc_disconnect_and_idle_import(struct obd_import *imp);
2323 int ptlrpc_import_recovery_state_machine(struct obd_import *imp);
2324 void deuuidify(char *uuid, const char *prefix, char **uuid_start,
2326 void ptlrpc_import_enter_resend(struct obd_import *imp);
2327 /* ptlrpc/pack_generic.c */
2328 int ptlrpc_reconnect_import(struct obd_import *imp);
2332 * ptlrpc msg buffer and swab interface
2336 bool ptlrpc_buf_need_swab(struct ptlrpc_request *req, const int inout,
2338 void ptlrpc_buf_set_swabbed(struct ptlrpc_request *req, const int inout,
2340 int ptlrpc_unpack_rep_msg(struct ptlrpc_request *req, int len);
2341 int ptlrpc_unpack_req_msg(struct ptlrpc_request *req, int len);
2343 int lustre_msg_check_version(struct lustre_msg *msg, __u32 version);
2344 void lustre_init_msg_v2(struct lustre_msg_v2 *msg, int count, __u32 *lens,
2346 int lustre_pack_request(struct ptlrpc_request *, __u32 magic, int count,
2347 __u32 *lens, char **bufs);
2348 int lustre_pack_reply(struct ptlrpc_request *, int count, __u32 *lens,
2350 int lustre_pack_reply_v2(struct ptlrpc_request *req, int count,
2351 __u32 *lens, char **bufs, int flags);
2352 #define LPRFL_EARLY_REPLY 1
2353 int lustre_pack_reply_flags(struct ptlrpc_request *, int count, __u32 *lens,
2354 char **bufs, int flags);
2355 int lustre_shrink_msg(struct lustre_msg *msg, int segment,
2356 unsigned int newlen, int move_data);
2357 void lustre_free_reply_state(struct ptlrpc_reply_state *rs);
2358 int __lustre_unpack_msg(struct lustre_msg *m, int len);
2359 __u32 lustre_msg_hdr_size(__u32 magic, __u32 count);
2360 __u32 lustre_msg_size(__u32 magic, int count, __u32 *lengths);
2361 __u32 lustre_msg_size_v2(int count, __u32 *lengths);
2362 __u32 lustre_packed_msg_size(struct lustre_msg *msg);
2363 __u32 lustre_msg_early_size(void);
2364 void *lustre_msg_buf_v2(struct lustre_msg_v2 *m, __u32 n, __u32 min_size);
2365 void *lustre_msg_buf(struct lustre_msg *m, __u32 n, __u32 minlen);
2366 __u32 lustre_msg_buflen(struct lustre_msg *m, __u32 n);
2367 void lustre_msg_set_buflen(struct lustre_msg *m, __u32 n, __u32 len);
2368 __u32 lustre_msg_bufcount(struct lustre_msg *m);
2369 char *lustre_msg_string(struct lustre_msg *m, __u32 n, __u32 max_len);
2370 __u32 lustre_msghdr_get_flags(struct lustre_msg *msg);
2371 void lustre_msghdr_set_flags(struct lustre_msg *msg, __u32 flags);
2372 __u32 lustre_msg_get_flags(struct lustre_msg *msg);
2373 void lustre_msg_add_flags(struct lustre_msg *msg, __u32 flags);
2374 void lustre_msg_set_flags(struct lustre_msg *msg, __u32 flags);
2375 void lustre_msg_clear_flags(struct lustre_msg *msg, __u32 flags);
2376 __u32 lustre_msg_get_op_flags(struct lustre_msg *msg);
2377 void lustre_msg_add_op_flags(struct lustre_msg *msg, __u32 flags);
2378 struct lustre_handle *lustre_msg_get_handle(struct lustre_msg *msg);
2379 __u32 lustre_msg_get_type(struct lustre_msg *msg);
2380 enum lustre_msg_version lustre_msg_get_version(struct lustre_msg *msg);
2381 void lustre_msg_add_version(struct lustre_msg *msg, __u32 version);
2382 __u32 lustre_msg_get_opc(struct lustre_msg *msg);
2383 __u64 lustre_msg_get_last_xid(struct lustre_msg *msg);
2384 __u16 lustre_msg_get_tag(struct lustre_msg *msg);
2385 __u64 lustre_msg_get_last_committed(struct lustre_msg *msg);
2386 __u64 *lustre_msg_get_versions(struct lustre_msg *msg);
2387 __u64 lustre_msg_get_transno(struct lustre_msg *msg);
2388 __u64 lustre_msg_get_slv(struct lustre_msg *msg);
2389 __u32 lustre_msg_get_limit(struct lustre_msg *msg);
2390 void lustre_msg_set_slv(struct lustre_msg *msg, __u64 slv);
2391 void lustre_msg_set_limit(struct lustre_msg *msg, __u64 limit);
2392 int lustre_msg_get_status(struct lustre_msg *msg);
2393 __u32 lustre_msg_get_conn_cnt(struct lustre_msg *msg);
2394 __u32 lustre_msg_get_magic(struct lustre_msg *msg);
2395 __u32 lustre_msg_get_timeout(struct lustre_msg *msg);
2396 __u32 lustre_msg_get_service_time(struct lustre_msg *msg);
2397 char *lustre_msg_get_jobid(struct lustre_msg *msg);
2398 __u32 lustre_msg_get_cksum(struct lustre_msg *msg);
2399 __u64 lustre_msg_get_mbits(struct lustre_msg *msg);
2400 __u32 lustre_msg_calc_cksum(struct lustre_msg *msg);
2401 void lustre_msg_set_handle(struct lustre_msg *msg,struct lustre_handle *handle);
2402 void lustre_msg_set_type(struct lustre_msg *msg, __u32 type);
2403 void lustre_msg_set_opc(struct lustre_msg *msg, __u32 opc);
2404 void lustre_msg_set_last_xid(struct lustre_msg *msg, __u64 last_xid);
2405 void lustre_msg_set_tag(struct lustre_msg *msg, __u16 tag);
2406 void lustre_msg_set_last_committed(struct lustre_msg *msg,__u64 last_committed);
2407 void lustre_msg_set_versions(struct lustre_msg *msg, __u64 *versions);
2408 void lustre_msg_set_transno(struct lustre_msg *msg, __u64 transno);
2409 void lustre_msg_set_status(struct lustre_msg *msg, __u32 status);
2410 void lustre_msg_set_conn_cnt(struct lustre_msg *msg, __u32 conn_cnt);
2411 void ptlrpc_req_set_repsize(struct ptlrpc_request *req, int count, __u32 *sizes);
2412 void ptlrpc_request_set_replen(struct ptlrpc_request *req);
2413 void lustre_msg_set_timeout(struct lustre_msg *msg, __u32 timeout);
2414 void lustre_msg_set_service_time(struct lustre_msg *msg, __u32 service_time);
2415 void lustre_msg_set_jobid(struct lustre_msg *msg, char *jobid);
2416 void lustre_msg_set_cksum(struct lustre_msg *msg, __u32 cksum);
2417 void lustre_msg_set_mbits(struct lustre_msg *msg, __u64 mbits);
2420 lustre_shrink_reply(struct ptlrpc_request *req, int segment,
2421 unsigned int newlen, int move_data)
2423 LASSERT(req->rq_reply_state);
2424 LASSERT(req->rq_repmsg);
2425 req->rq_replen = lustre_shrink_msg(req->rq_repmsg, segment,
2429 #ifdef LUSTRE_TRANSLATE_ERRNOS
2431 static inline int ptlrpc_status_hton(int h)
2434 * Positive errnos must be network errnos, such as LUSTRE_EDEADLK,
2435 * ELDLM_LOCK_ABORTED, etc.
2438 return -lustre_errno_hton(-h);
2443 static inline int ptlrpc_status_ntoh(int n)
2446 * See the comment in ptlrpc_status_hton().
2449 return -lustre_errno_ntoh(-n);
2456 #define ptlrpc_status_hton(h) (h)
2457 #define ptlrpc_status_ntoh(n) (n)
2462 /** Change request phase of \a req to \a new_phase */
2464 ptlrpc_rqphase_move(struct ptlrpc_request *req, enum rq_phase new_phase)
2466 if (req->rq_phase == new_phase)
2469 if (new_phase == RQ_PHASE_UNREG_RPC ||
2470 new_phase == RQ_PHASE_UNREG_BULK) {
2471 /* No embedded unregistering phases */
2472 if (req->rq_phase == RQ_PHASE_UNREG_RPC ||
2473 req->rq_phase == RQ_PHASE_UNREG_BULK)
2476 req->rq_next_phase = req->rq_phase;
2478 atomic_inc(&req->rq_import->imp_unregistering);
2481 if (req->rq_phase == RQ_PHASE_UNREG_RPC ||
2482 req->rq_phase == RQ_PHASE_UNREG_BULK) {
2484 atomic_dec(&req->rq_import->imp_unregistering);
2487 DEBUG_REQ(D_INFO, req, "move req \"%s\" -> \"%s\"",
2488 ptlrpc_rqphase2str(req), ptlrpc_phase2str(new_phase));
2490 req->rq_phase = new_phase;
2494 * Returns true if request \a req got early reply and hard deadline is not met
2497 ptlrpc_client_early(struct ptlrpc_request *req)
2499 return req->rq_early;
2503 * Returns true if we got real reply from server for this request
2506 ptlrpc_client_replied(struct ptlrpc_request *req)
2508 if (req->rq_reply_deadline > ktime_get_real_seconds())
2510 return req->rq_replied;
2513 /** Returns true if request \a req is in process of receiving server reply */
2515 ptlrpc_client_recv(struct ptlrpc_request *req)
2517 if (req->rq_reply_deadline > ktime_get_real_seconds())
2519 return req->rq_receiving_reply;
2523 ptlrpc_client_recv_or_unlink(struct ptlrpc_request *req)
2527 spin_lock(&req->rq_lock);
2528 if (req->rq_reply_deadline > ktime_get_real_seconds()) {
2529 spin_unlock(&req->rq_lock);
2532 if (req->rq_req_deadline > ktime_get_real_seconds()) {
2533 spin_unlock(&req->rq_lock);
2537 rc = !req->rq_req_unlinked || !req->rq_reply_unlinked ||
2538 req->rq_receiving_reply;
2539 spin_unlock(&req->rq_lock);
2544 ptlrpc_client_wake_req(struct ptlrpc_request *req)
2547 if (req->rq_set == NULL)
2548 wake_up(&req->rq_reply_waitq);
2550 wake_up(&req->rq_set->set_waitq);
2554 ptlrpc_rs_addref(struct ptlrpc_reply_state *rs)
2556 LASSERT(atomic_read(&rs->rs_refcount) > 0);
2557 atomic_inc(&rs->rs_refcount);
2561 ptlrpc_rs_decref(struct ptlrpc_reply_state *rs)
2563 LASSERT(atomic_read(&rs->rs_refcount) > 0);
2564 if (atomic_dec_and_test(&rs->rs_refcount))
2565 lustre_free_reply_state(rs);
2568 /* Should only be called once per req */
2569 static inline void ptlrpc_req_drop_rs(struct ptlrpc_request *req)
2571 if (req->rq_reply_state == NULL)
2572 return; /* shouldn't occur */
2573 ptlrpc_rs_decref(req->rq_reply_state);
2574 req->rq_reply_state = NULL;
2575 req->rq_repmsg = NULL;
2578 static inline __u32 lustre_request_magic(struct ptlrpc_request *req)
2580 return lustre_msg_get_magic(req->rq_reqmsg);
2583 static inline int ptlrpc_req_get_repsize(struct ptlrpc_request *req)
2585 switch (req->rq_reqmsg->lm_magic) {
2586 case LUSTRE_MSG_MAGIC_V2:
2587 return req->rq_reqmsg->lm_repsize;
2589 LASSERTF(0, "incorrect message magic: %08x\n",
2590 req->rq_reqmsg->lm_magic);
2595 static inline int ptlrpc_send_limit_expired(struct ptlrpc_request *req)
2597 if (req->rq_delay_limit != 0 &&
2598 req->rq_queued_time + req->rq_delay_limit < ktime_get_seconds())
2603 static inline int ptlrpc_no_resend(struct ptlrpc_request *req)
2605 if (!req->rq_no_resend && ptlrpc_send_limit_expired(req)) {
2606 spin_lock(&req->rq_lock);
2607 req->rq_no_resend = 1;
2608 spin_unlock(&req->rq_lock);
2610 return req->rq_no_resend;
2614 ptlrpc_server_get_timeout(struct ptlrpc_service_part *svcpt)
2616 int at = AT_OFF ? 0 : at_get(&svcpt->scp_at_estimate);
2618 return svcpt->scp_service->srv_watchdog_factor *
2619 max_t(int, at, obd_timeout);
2622 static inline struct ptlrpc_service *
2623 ptlrpc_req2svc(struct ptlrpc_request *req)
2625 LASSERT(req->rq_rqbd != NULL);
2626 return req->rq_rqbd->rqbd_svcpt->scp_service;
2629 /* ldlm/ldlm_lib.c */
2631 * Target client logic
2634 int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg);
2635 int client_obd_cleanup(struct obd_device *obddev);
2636 int client_connect_import(const struct lu_env *env,
2637 struct obd_export **exp, struct obd_device *obd,
2638 struct obd_uuid *cluuid, struct obd_connect_data *,
2640 int client_disconnect_export(struct obd_export *exp);
2641 int client_import_add_conn(struct obd_import *imp, struct obd_uuid *uuid,
2643 int client_import_del_conn(struct obd_import *imp, struct obd_uuid *uuid);
2644 int client_import_find_conn(struct obd_import *imp, lnet_nid_t peer,
2645 struct obd_uuid *uuid);
2646 int import_set_conn_priority(struct obd_import *imp, struct obd_uuid *uuid);
2647 void client_destroy_import(struct obd_import *imp);
2650 #ifdef HAVE_SERVER_SUPPORT
2651 int server_disconnect_export(struct obd_export *exp);
2654 /* ptlrpc/pinger.c */
2656 * Pinger API (client side only)
2659 enum timeout_event {
2662 struct timeout_item;
2663 typedef int (*timeout_cb_t)(struct timeout_item *, void *);
2664 int ptlrpc_pinger_add_import(struct obd_import *imp);
2665 int ptlrpc_pinger_del_import(struct obd_import *imp);
2666 int ptlrpc_add_timeout_client(time64_t time, enum timeout_event event,
2667 timeout_cb_t cb, void *data,
2668 struct list_head *obd_list);
2669 int ptlrpc_del_timeout_client(struct list_head *obd_list,
2670 enum timeout_event event);
2671 struct ptlrpc_request * ptlrpc_prep_ping(struct obd_import *imp);
2672 int ptlrpc_obd_ping(struct obd_device *obd);
2673 void ping_evictor_start(void);
2674 void ping_evictor_stop(void);
2675 void ptlrpc_pinger_ir_up(void);
2676 void ptlrpc_pinger_ir_down(void);
2678 int ptlrpc_pinger_suppress_pings(void);
2680 /* ptlrpc/ptlrpcd.c */
2681 void ptlrpcd_stop(struct ptlrpcd_ctl *pc, int force);
2682 void ptlrpcd_free(struct ptlrpcd_ctl *pc);
2683 void ptlrpcd_wake(struct ptlrpc_request *req);
2684 void ptlrpcd_add_req(struct ptlrpc_request *req);
2685 void ptlrpcd_add_rqset(struct ptlrpc_request_set *set);
2686 int ptlrpcd_addref(void);
2687 void ptlrpcd_decref(void);
2689 /* ptlrpc/lproc_ptlrpc.c */
2691 * procfs output related functions
2694 const char* ll_opcode2str(__u32 opcode);
2695 const int ll_str2opcode(const char *ops);
2696 #ifdef CONFIG_PROC_FS
2697 void ptlrpc_lprocfs_register_obd(struct obd_device *obd);
2698 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd);
2699 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes);
2701 static inline void ptlrpc_lprocfs_register_obd(struct obd_device *obd) {}
2702 static inline void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd) {}
2703 static inline void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes) {}
2707 /* ptlrpc/llog_server.c */
2708 int llog_origin_handle_open(struct ptlrpc_request *req);
2709 int llog_origin_handle_prev_block(struct ptlrpc_request *req);
2710 int llog_origin_handle_next_block(struct ptlrpc_request *req);
2711 int llog_origin_handle_read_header(struct ptlrpc_request *req);
2713 /* ptlrpc/llog_client.c */
2714 extern struct llog_operations llog_client_ops;