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/
31 /** \defgroup PtlRPC Portal RPC and networking module.
33 * PortalRPC is the layer used by rest of lustre code to achieve network
34 * communications: establish connections with corresponding export and import
35 * states, listen for a service, send and receive RPCs.
36 * PortalRPC also includes base recovery framework: packet resending and
37 * replaying, reconnections, pinger.
39 * PortalRPC utilizes LNet as its transport layer.
52 #include <linux/kobject.h>
53 #include <linux/rhashtable.h>
54 #include <linux/uio.h>
55 #include <libcfs/libcfs.h>
57 #include <lnet/lib-types.h>
58 #include <uapi/linux/lnet/nidstr.h>
59 #include <uapi/linux/lustre/lustre_idl.h>
60 #include <lustre_ha.h>
61 #include <lustre_sec.h>
62 #include <lustre_import.h>
63 #include <lprocfs_status.h>
64 #include <lu_object.h>
65 #include <lustre_req_layout.h>
66 #include <obd_support.h>
67 #include <uapi/linux/lustre/lustre_ver.h>
69 /* MD flags we _always_ use */
70 #define PTLRPC_MD_OPTIONS 0
73 * log2 max # of bulk operations in one request: 2=4MB/RPC, 5=32MB/RPC, ...
74 * In order for the client and server to properly negotiate the maximum
75 * possible transfer size, PTLRPC_BULK_OPS_COUNT must be a power-of-two
76 * value. The client is free to limit the actual RPC size for any bulk
77 * transfer via cl_max_pages_per_rpc to some non-power-of-two value.
78 * NOTE: This is limited to 16 (=64GB RPCs) by IOOBJ_MAX_BRW_BITS.
80 #define PTLRPC_BULK_OPS_BITS 6
81 #if PTLRPC_BULK_OPS_BITS > 16
82 #error "More than 65536 BRW RPCs not allowed by IOOBJ_MAX_BRW_BITS."
84 #define PTLRPC_BULK_OPS_COUNT (1U << PTLRPC_BULK_OPS_BITS)
86 * PTLRPC_BULK_OPS_MASK is for the convenience of the client only, and
87 * should not be used on the server at all. Otherwise, it imposes a
88 * protocol limitation on the maximum RPC size that can be used by any
89 * RPC sent to that server in the future. Instead, the server should
90 * use the negotiated per-client ocd_brw_size to determine the bulk
93 #define PTLRPC_BULK_OPS_MASK (~((__u64)PTLRPC_BULK_OPS_COUNT - 1))
96 * Define maxima for bulk I/O.
98 * A single PTLRPC BRW request is sent via up to PTLRPC_BULK_OPS_COUNT
99 * of LNET_MTU sized RDMA transfers. Clients and servers negotiate the
100 * currently supported maximum between peers at connect via ocd_brw_size.
102 #define PTLRPC_MAX_BRW_BITS (LNET_MTU_BITS + PTLRPC_BULK_OPS_BITS)
103 #define PTLRPC_MAX_BRW_SIZE (1U << PTLRPC_MAX_BRW_BITS)
104 #define PTLRPC_MAX_BRW_PAGES (PTLRPC_MAX_BRW_SIZE >> PAGE_SHIFT)
106 #define ONE_MB_BRW_SIZE (1U << LNET_MTU_BITS)
107 #define MD_MAX_BRW_SIZE (1U << LNET_MTU_BITS)
108 #define MD_MAX_BRW_PAGES (MD_MAX_BRW_SIZE >> PAGE_SHIFT)
109 #define DT_MAX_BRW_SIZE PTLRPC_MAX_BRW_SIZE
110 #define DT_DEF_BRW_SIZE (4 * ONE_MB_BRW_SIZE)
111 #define DT_MAX_BRW_PAGES (DT_MAX_BRW_SIZE >> PAGE_SHIFT)
112 #define OFD_MAX_BRW_SIZE (1U << LNET_MTU_BITS)
114 /* When PAGE_SIZE is a constant, we can check our arithmetic here with cpp! */
115 #if ((PTLRPC_MAX_BRW_PAGES & (PTLRPC_MAX_BRW_PAGES - 1)) != 0)
116 # error "PTLRPC_MAX_BRW_PAGES isn't a power of two"
118 #if (PTLRPC_MAX_BRW_SIZE != (PTLRPC_MAX_BRW_PAGES * PAGE_SIZE))
119 # error "PTLRPC_MAX_BRW_SIZE isn't PTLRPC_MAX_BRW_PAGES * PAGE_SIZE"
121 #if (PTLRPC_MAX_BRW_SIZE > LNET_MTU * PTLRPC_BULK_OPS_COUNT)
122 # error "PTLRPC_MAX_BRW_SIZE too big"
124 #if (PTLRPC_MAX_BRW_PAGES > LNET_MAX_IOV * PTLRPC_BULK_OPS_COUNT)
125 # error "PTLRPC_MAX_BRW_PAGES too big"
128 #define PTLRPC_NTHRS_INIT 2
133 * Constants determine how memory is used to buffer incoming service requests.
135 * ?_NBUFS # buffers to allocate when growing the pool
136 * ?_BUFSIZE # bytes in a single request buffer
137 * ?_MAXREQSIZE # maximum request service will receive
139 * When fewer than ?_NBUFS/2 buffers are posted for receive, another chunk
140 * of ?_NBUFS is added to the pool.
142 * Messages larger than ?_MAXREQSIZE are dropped. Request buffers are
143 * considered full when less than ?_MAXREQSIZE is left in them.
148 * Constants determine how threads are created for ptlrpc service.
150 * ?_NTHRS_INIT # threads to create for each service partition on
151 * initializing. If it's non-affinity service and
152 * there is only one partition, it's the overall #
153 * threads for the service while initializing.
154 * ?_NTHRS_BASE # threads should be created at least for each
155 * ptlrpc partition to keep the service healthy.
156 * It's the low-water mark of threads upper-limit
157 * for each partition.
158 * ?_THR_FACTOR # threads can be added on threads upper-limit for
159 * each CPU core. This factor is only for reference,
160 * we might decrease value of factor if number of cores
161 * per CPT is above a limit.
162 * ?_NTHRS_MAX # overall threads can be created for a service,
163 * it's a soft limit because if service is running
164 * on machine with hundreds of cores and tens of
165 * CPU partitions, we need to guarantee each partition
166 * has ?_NTHRS_BASE threads, which means total threads
167 * will be ?_NTHRS_BASE * number_of_cpts which can
168 * exceed ?_NTHRS_MAX.
172 * #define MDS_NTHRS_INIT 2
173 * #define MDS_NTHRS_BASE 64
174 * #define MDS_NTHRS_FACTOR 8
175 * #define MDS_NTHRS_MAX 1024
178 * ---------------------------------------------------------------------
179 * Server(A) has 16 cores, user configured it to 4 partitions so each
180 * partition has 4 cores, then actual number of service threads on each
182 * MDS_NTHRS_BASE(64) + cores(4) * MDS_NTHRS_FACTOR(8) = 96
184 * Total number of threads for the service is:
185 * 96 * partitions(4) = 384
188 * ---------------------------------------------------------------------
189 * Server(B) has 32 cores, user configured it to 4 partitions so each
190 * partition has 8 cores, then actual number of service threads on each
192 * MDS_NTHRS_BASE(64) + cores(8) * MDS_NTHRS_FACTOR(8) = 128
194 * Total number of threads for the service is:
195 * 128 * partitions(4) = 512
198 * ---------------------------------------------------------------------
199 * Server(B) has 96 cores, user configured it to 8 partitions so each
200 * partition has 12 cores, then actual number of service threads on each
202 * MDS_NTHRS_BASE(64) + cores(12) * MDS_NTHRS_FACTOR(8) = 160
204 * Total number of threads for the service is:
205 * 160 * partitions(8) = 1280
207 * However, it's above the soft limit MDS_NTHRS_MAX, so we choose this number
208 * as upper limit of threads number for each partition:
209 * MDS_NTHRS_MAX(1024) / partitions(8) = 128
212 * ---------------------------------------------------------------------
213 * Server(C) have a thousand of cores and user configured it to 32 partitions
214 * MDS_NTHRS_BASE(64) * 32 = 2048
216 * which is already above soft limit MDS_NTHRS_MAX(1024), but we still need
217 * to guarantee that each partition has at least MDS_NTHRS_BASE(64) threads
218 * to keep service healthy, so total number of threads will just be 2048.
220 * NB: we don't suggest to choose server with that many cores because backend
221 * filesystem itself, buffer cache, or underlying network stack might
222 * have some SMP scalability issues at that large scale.
224 * If user already has a fat machine with hundreds or thousands of cores,
225 * there are two choices for configuration:
226 * a) create CPU table from subset of all CPUs and run Lustre on
228 * b) bind service threads on a few partitions, see modparameters of
229 * MDS and OSS for details
231 * NB: these calculations (and examples below) are simplified to help
232 * understanding, the real implementation is a little more complex,
233 * please see ptlrpc_server_nthreads_check() for details.
238 * LDLM threads constants:
240 * Given 8 as factor and 24 as base threads number
243 * On 4-core machine we will have 24 + 8 * 4 = 56 threads.
246 * On 8-core machine with 2 partitions we will have 24 + 4 * 8 = 56
247 * threads for each partition and total threads number will be 112.
250 * On 64-core machine with 8 partitions we will need LDLM_NTHRS_BASE(24)
251 * threads for each partition to keep service healthy, so total threads
252 * number should be 24 * 8 = 192.
254 * So with these constants, threads number will be at the similar level
255 * of old versions, unless target machine has over a hundred cores
257 #define LDLM_THR_FACTOR 8
258 #define LDLM_NTHRS_INIT PTLRPC_NTHRS_INIT
259 #define LDLM_NTHRS_BASE 24
260 #define LDLM_NTHRS_MAX (num_online_cpus() == 1 ? 64 : 128)
262 #define LDLM_BL_THREADS LDLM_NTHRS_AUTO_INIT
263 #define LDLM_CLIENT_NBUFS 1
264 #define LDLM_SERVER_NBUFS 64
265 #define LDLM_BUFSIZE (8 * 1024)
266 #define LDLM_MAXREQSIZE (5 * 1024)
267 #define LDLM_MAXREPSIZE (1024)
270 * MDS threads constants:
272 * Please see examples in "Thread Constants", MDS threads number will be at
273 * the comparable level of old versions, unless the server has many cores.
275 #ifndef MDS_MAX_THREADS
276 #define MDS_MAX_THREADS 1024
277 #define MDS_MAX_OTHR_THREADS 256
279 #else /* MDS_MAX_THREADS */
280 #if MDS_MAX_THREADS < PTLRPC_NTHRS_INIT
281 #undef MDS_MAX_THREADS
282 #define MDS_MAX_THREADS PTLRPC_NTHRS_INIT
284 #define MDS_MAX_OTHR_THREADS max(PTLRPC_NTHRS_INIT, MDS_MAX_THREADS / 2)
287 /* default service */
288 #define MDS_THR_FACTOR 8
289 #define MDS_NTHRS_INIT PTLRPC_NTHRS_INIT
290 #define MDS_NTHRS_MAX MDS_MAX_THREADS
291 #define MDS_NTHRS_BASE min(64, MDS_NTHRS_MAX)
293 /* read-page service */
294 #define MDS_RDPG_THR_FACTOR 4
295 #define MDS_RDPG_NTHRS_INIT PTLRPC_NTHRS_INIT
296 #define MDS_RDPG_NTHRS_MAX MDS_MAX_OTHR_THREADS
297 #define MDS_RDPG_NTHRS_BASE min(48, MDS_RDPG_NTHRS_MAX)
299 /* these should be removed when we remove setattr service in the future */
300 #define MDS_SETA_THR_FACTOR 4
301 #define MDS_SETA_NTHRS_INIT PTLRPC_NTHRS_INIT
302 #define MDS_SETA_NTHRS_MAX MDS_MAX_OTHR_THREADS
303 #define MDS_SETA_NTHRS_BASE min(48, MDS_SETA_NTHRS_MAX)
305 /* non-affinity threads */
306 #define MDS_OTHR_NTHRS_INIT PTLRPC_NTHRS_INIT
307 #define MDS_OTHR_NTHRS_MAX MDS_MAX_OTHR_THREADS
312 * Assume file name length = FNAME_MAX = 256 (true for ext3).
313 * path name length = PATH_MAX = 4096
314 * LOV MD size max = EA_MAX = 24 * 2000
315 * (NB: 24 is size of lov_ost_data)
316 * LOV LOGCOOKIE size max = 32 * 2000
317 * (NB: 32 is size of llog_cookie)
318 * symlink: FNAME_MAX + PATH_MAX <- largest
319 * link: FNAME_MAX + PATH_MAX (mds_rec_link < mds_rec_create)
320 * rename: FNAME_MAX + FNAME_MAX
321 * open: FNAME_MAX + EA_MAX
323 * MDS_MAXREQSIZE ~= 4736 bytes =
324 * lustre_msg + ldlm_request + mdt_body + mds_rec_create + FNAME_MAX + PATH_MAX
325 * MDS_MAXREPSIZE ~= 8300 bytes = lustre_msg + llog_header
327 * Realistic size is about 512 bytes (20 character name + 128 char symlink),
328 * except in the open case where there are a large number of OSTs in a LOV.
330 #define MDS_MAXREQSIZE (5 * 1024) /* >= 4736 */
331 #define MDS_MAXREPSIZE (9 * 1024) /* >= 8300 */
334 * MDS incoming request with LOV EA
335 * 24 = sizeof(struct lov_ost_data), i.e: replay of opencreate
337 #define MDS_LOV_MAXREQSIZE max(MDS_MAXREQSIZE, \
338 362 + LOV_MAX_STRIPE_COUNT * 24)
340 * MDS outgoing reply with LOV EA
342 * NB: max reply size Lustre 2.4+ client can get from old MDS is:
343 * LOV_MAX_STRIPE_COUNT * (llog_cookie + lov_ost_data) + extra bytes
345 * but 2.4 or later MDS will never send reply with llog_cookie to any
346 * version client. This macro is defined for server side reply buffer size.
348 #define MDS_LOV_MAXREPSIZE MDS_LOV_MAXREQSIZE
351 * This is the size of a maximum REINT_SETXATTR request:
353 * lustre_msg 56 (32 + 4 x 5 + 4)
355 * mdt_rec_setxattr 136
357 * name 256 (XATTR_NAME_MAX)
358 * value 65536 (XATTR_SIZE_MAX)
360 #define MDS_EA_MAXREQSIZE 66288
363 * These are the maximum request and reply sizes (rounded up to 1 KB
364 * boundaries) for the "regular" MDS_REQUEST_PORTAL and MDS_REPLY_PORTAL.
366 #define MDS_REG_MAXREQSIZE (((max(MDS_EA_MAXREQSIZE, \
367 MDS_LOV_MAXREQSIZE) + 1023) >> 10) << 10)
368 #define MDS_REG_MAXREPSIZE MDS_REG_MAXREQSIZE
371 * The update request includes all of updates from the create, which might
372 * include linkea (4K maxim), together with other updates, we set it to 1000K:
373 * lustre_msg + ptlrpc_body + OUT_UPDATE_BUFFER_SIZE_MAX
375 #define OUT_MAXREQSIZE (1000 * 1024)
376 #define OUT_MAXREPSIZE MDS_MAXREPSIZE
378 #define BUT_MAXREQSIZE OUT_MAXREQSIZE
379 #define BUT_MAXREPSIZE BUT_MAXREQSIZE
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 +
423 #define SEQ_MAXREQSIZE (160)
425 /** SEQ_MAXREPSIZE == lustre_msg + ptlrpc_body + lu_range */
426 #define SEQ_MAXREPSIZE (152)
427 #define SEQ_BUFSIZE (1 << 12)
429 /** MGS threads must be >= 3, see bug 22458 comment #28 */
430 #define MGS_NTHRS_INIT (PTLRPC_NTHRS_INIT + 1)
431 #define MGS_NTHRS_MAX 32
434 #define MGS_BUFSIZE (8 * 1024)
435 #define MGS_MAXREQSIZE (7 * 1024)
436 #define MGS_MAXREPSIZE (9 * 1024)
439 * OSS threads constants:
441 * Given 8 as factor and 64 as base threads number
444 * On 8-core server configured to 2 partitions, we will have
445 * 64 + 8 * 4 = 96 threads for each partition, 192 total threads.
448 * On 32-core machine configured to 4 partitions, we will have
449 * 64 + 8 * 8 = 112 threads for each partition, so total threads number
450 * will be 112 * 4 = 448.
453 * On 64-core machine configured to 4 partitions, we will have
454 * 64 + 16 * 8 = 192 threads for each partition, so total threads number
455 * will be 192 * 4 = 768 which is above limit OSS_NTHRS_MAX(512), so we
456 * cut off the value to OSS_NTHRS_MAX(512) / 4 which is 128 threads
457 * for each partition.
459 * So we can see that with these constants, threads number wil be at the
460 * similar level of old versions, unless the server has many cores.
462 /* depress threads factor for VM with small memory size */
463 #define OSS_THR_FACTOR min_t(int, 8, \
464 NUM_CACHEPAGES >> (28 - PAGE_SHIFT))
465 #define OSS_NTHRS_INIT (PTLRPC_NTHRS_INIT + 1)
466 #define OSS_NTHRS_BASE 64
468 /* threads for handling "create" request */
469 #define OSS_CR_THR_FACTOR 1
470 #define OSS_CR_NTHRS_INIT PTLRPC_NTHRS_INIT
471 #define OSS_CR_NTHRS_BASE 8
472 #define OSS_CR_NTHRS_MAX 64
475 * OST_IO_MAXREQSIZE ~=
476 * lustre_msg + ptlrpc_body + obdo + obd_ioobj +
477 * DT_MAX_BRW_PAGES * niobuf_remote
479 * - single object with 16 pages is 512 bytes
480 * - OST_IO_MAXREQSIZE must be at least 1 niobuf per page of data
481 * - Must be a multiple of 1024
482 * - should allow a reasonably large SHORT_IO_BYTES size (64KB)
484 #define _OST_MAXREQSIZE_BASE ((unsigned long)(sizeof(struct lustre_msg) + \
485 /* lm_buflens */ sizeof(__u32) * 4 + \
486 sizeof(struct ptlrpc_body) + \
487 sizeof(struct obdo) + \
488 sizeof(struct obd_ioobj) + \
489 sizeof(struct niobuf_remote)))
490 #define _OST_MAXREQSIZE_SUM ((unsigned long)(_OST_MAXREQSIZE_BASE + \
491 sizeof(struct niobuf_remote) * \
494 * FIEMAP request can be 4K+ for now
496 #define OST_MAXREQSIZE (16UL * 1024UL)
497 #define OST_IO_MAXREQSIZE max(OST_MAXREQSIZE, \
498 ((_OST_MAXREQSIZE_SUM - 1) | \
500 /* Safe estimate of free space in standard RPC, provides upper limit for # of
501 * bytes of i/o to pack in RPC (skipping bulk transfer).
503 #define OST_MAX_SHORT_IO_BYTES ((OST_IO_MAXREQSIZE - _OST_MAXREQSIZE_BASE) & \
506 /* Actual size used for short i/o buffer. Calculation means this:
507 * At least one page (for large PAGE_SIZE), or 16 KiB, but not more
508 * than the available space aligned to a page boundary.
510 #define OBD_DEF_SHORT_IO_BYTES min(max(PAGE_SIZE, 16UL * 1024UL), \
511 OST_MAX_SHORT_IO_BYTES)
513 #define OST_MAXREPSIZE (9 * 1024)
514 #define OST_IO_MAXREPSIZE OST_MAXREPSIZE
517 /** OST_BUFSIZE = max_reqsize + max sptlrpc payload size */
518 #define OST_BUFSIZE max_t(int, OST_MAXREQSIZE + 1024, 32 * 1024)
520 * OST_IO_MAXREQSIZE is 18K, giving extra 46K can increase buffer utilization
521 * rate of request buffer, please check comment of MDS_LOV_BUFSIZE for details.
523 #define OST_IO_BUFSIZE max_t(int, OST_IO_MAXREQSIZE + 1024, 64 * 1024)
525 /* Macro to hide a typecast and BUILD_BUG. */
526 #define ptlrpc_req_async_args(_var, req) ({ \
527 BUILD_BUG_ON(sizeof(*_var) > sizeof(req->rq_async_args)); \
528 (typeof(_var))&req->rq_async_args; \
531 struct ptlrpc_replay_async_args {
537 * Structure to single define portal connection.
539 struct ptlrpc_connection {
540 /** linkage for connections hash table */
541 struct rhash_head c_hash;
542 /** Our own lnet nid for this connection */
543 struct lnet_nid c_self;
544 /** Remote side nid for this connection */
545 struct lnet_processid c_peer;
546 /** UUID of the other side */
547 struct obd_uuid c_remote_uuid;
548 /** reference counter for this connection */
552 /** Client definition for PortalRPC */
553 struct ptlrpc_client {
554 /** What lnet portal does this client send messages to by default */
555 __u32 cli_request_portal;
556 /** What portal do we expect replies on */
557 __u32 cli_reply_portal;
558 /** Name of the client */
559 const char *cli_name;
562 /** state flags of requests */
563 /* XXX only ones left are those used by the bulk descs as well! */
564 #define PTL_RPC_FL_INTR BIT(0) /* reply wait was interrupted by user */
565 #define PTL_RPC_FL_TIMEOUT BIT(7) /* req timed out waiting for reply */
567 #define REQ_MAX_ACK_LOCKS 8
569 union ptlrpc_async_args {
571 * Scratchpad for passing args to completion interpreter. Users
572 * cast to the struct of their choosing, and BUILD_BUG_ON that this is
573 * big enough. For _tons_ of context, OBD_ALLOC a struct and store
574 * a pointer to it here. The pointer_arg ensures this struct is at
575 * least big enough for that.
577 void *pointer_arg[11];
581 struct ptlrpc_request_set;
582 typedef int (*set_producer_func)(struct ptlrpc_request_set *, void *);
585 * Definition of request set structure.
586 * Request set is a list of requests (not necessary to the same target) that
587 * once populated with RPCs could be sent in parallel.
588 * There are two kinds of request sets. General purpose and with dedicated
589 * serving thread. Example of the latter is ptlrpcd set.
590 * For general purpose sets once request set started sending it is impossible
591 * to add new requests to such set.
592 * Provides a way to call "completion callbacks" when all requests in the set
595 struct ptlrpc_request_set {
596 atomic_t set_refcount;
597 /** number of in queue requests */
598 atomic_t set_new_count;
599 /** number of uncompleted requests */
600 atomic_t set_remaining;
601 /** wait queue to wait on for request events */
602 wait_queue_head_t set_waitq;
603 /** List of requests in the set */
604 struct list_head set_requests;
606 * Lock for \a set_new_requests manipulations
607 * locked so that any old caller can communicate requests to
608 * the set holder who can then fold them into the lock-free set
610 spinlock_t set_new_req_lock;
611 /** List of new yet unsent requests. Only used with ptlrpcd now. */
612 struct list_head set_new_requests;
614 /** rq_status of requests that have been freed already */
616 /** Additional fields used by the flow control extension */
617 /** Maximum number of RPCs in flight */
618 int set_max_inflight;
619 /** Callback function used to generate RPCs */
620 set_producer_func set_producer;
621 /** opaq argument passed to the producer callback */
622 void *set_producer_arg;
623 unsigned int set_allow_intr:1;
626 struct ptlrpc_bulk_desc;
627 struct ptlrpc_service_part;
628 struct ptlrpc_service;
631 * ptlrpc callback & work item stuff
633 struct ptlrpc_cb_id {
634 void (*cbid_fn)(struct lnet_event *ev); /* specific callback fn */
635 void *cbid_arg; /* additional arg */
638 /** Maximum number of locks to fit into reply state, migrating directory max
639 * stripe count is 2 * LMV_MAX_STRIPES_PER_MDT, plus source parent, target
640 * parent, source and target master object:
641 * 2 * LMV_MAX_STRIPES_PER_MDT + 4
643 #define RS_MAX_LOCKS 14
647 * Structure to define reply state on the server
648 * Reply state holds various reply message information. Also for "difficult"
649 * replies (rep-ack case) we store the state after sending reply and wait
650 * for the client to acknowledge the reception. In these cases locks could be
651 * added to the state for replay/failover consistency guarantees.
653 struct ptlrpc_reply_state {
654 /** Callback description */
655 struct ptlrpc_cb_id rs_cb_id;
656 /** Linkage for list of all reply states in a system */
657 struct list_head rs_list;
658 /** Linkage for list of all reply states on same export */
659 struct list_head rs_exp_list;
660 /** Linkage for list of all reply states for same obd */
661 struct list_head rs_obd_list;
663 struct list_head rs_debug_list;
665 /** A spinlock to protect the reply state flags */
667 /** Reply state flags */
668 unsigned long rs_difficult:1; /* ACK/commit stuff */
669 unsigned long rs_no_ack:1; /* no ACK, (incl difficult reqs) */
670 unsigned long rs_scheduled:1; /* being handled? */
671 unsigned long rs_scheduled_ever:1;/* any schedule attempts? */
672 unsigned long rs_handled:1; /* been handled yet? */
673 unsigned long rs_sent:1; /* Got LNET_EVENT_SEND? */
674 unsigned long rs_unlinked:1; /* Reply MD unlinked? */
675 unsigned long rs_prealloc:1; /* rs from prealloc list */
676 /* transaction committed and rs dispatched by ptlrpc_commit_replies */
677 unsigned long rs_committed:1;
678 atomic_t rs_refcount; /* number of users */
679 /** Number of locks awaiting client ACK */
682 /** Size of the state */
686 /** Transaction number */
690 struct obd_export *rs_export;
691 struct ptlrpc_service_part *rs_svcpt;
692 /** Lnet metadata handle for the reply */
693 struct lnet_handle_md rs_md_h;
695 /** Context for the sevice thread */
696 struct ptlrpc_svc_ctx *rs_svc_ctx;
697 /** Reply buffer (actually sent to the client), encoded if needed */
698 struct lustre_msg *rs_repbuf; /* wrapper */
699 /** Size of the reply buffer */
700 int rs_repbuf_len; /* wrapper buf length */
701 /** Size of the reply message */
702 int rs_repdata_len; /* wrapper msg length */
704 * Actual reply message. Its content is encrupted (if needed) to
705 * produce reply buffer for actual sending. In simple case
706 * of no network encryption we jus set \a rs_repbuf to \a rs_msg
708 struct lustre_msg *rs_msg; /* reply message */
710 /** Handles of locks awaiting client reply ACK */
711 struct lustre_handle rs_locks[RS_MAX_LOCKS];
714 struct ptlrpc_thread;
718 RQ_PHASE_NEW = 0xebc0de00,
719 RQ_PHASE_RPC = 0xebc0de01,
720 RQ_PHASE_BULK = 0xebc0de02,
721 RQ_PHASE_INTERPRET = 0xebc0de03,
722 RQ_PHASE_COMPLETE = 0xebc0de04,
723 RQ_PHASE_UNREG_RPC = 0xebc0de05,
724 RQ_PHASE_UNREG_BULK = 0xebc0de06,
725 RQ_PHASE_UNDEFINED = 0xebc0de07
728 /** Type of request interpreter call-back */
729 typedef int (*ptlrpc_interpterer_t)(const struct lu_env *env,
730 struct ptlrpc_request *req,
732 /** Type of request resend call-back */
733 typedef void (*ptlrpc_resend_cb_t)(struct ptlrpc_request *req,
737 * Definition of request pool structure.
738 * The pool is used to store empty preallocated requests for the case
739 * when we would actually need to send something without performing
740 * any allocations (to avoid e.g. OOM).
742 struct ptlrpc_request_pool {
743 /** Locks the list */
745 /** list of ptlrpc_request structs */
746 struct list_head prp_req_list;
747 /** Maximum message size that would fit into a rquest from this pool */
749 /** Function to allocate more requests for this pool */
750 int (*prp_populate)(struct ptlrpc_request_pool *, int);
758 #include <lustre_nrs.h>
761 * Basic request prioritization operations structure.
762 * The whole idea is centered around locks and RPCs that might affect locks.
763 * When a lock is contended we try to give priority to RPCs that might lead
764 * to fastest release of that lock.
765 * Currently only implemented for OSTs only in a way that makes all
766 * IO and truncate RPCs that are coming from a locked region where a lock is
767 * contended a priority over other requests.
769 struct ptlrpc_hpreq_ops {
771 * Check if the lock handle of the given lock is the same as
772 * taken from the request.
774 int (*hpreq_lock_match)(struct ptlrpc_request *, struct ldlm_lock *);
776 * Check if the request is a high priority one.
778 int (*hpreq_check)(struct ptlrpc_request *);
780 * Called after the request has been handled.
782 void (*hpreq_fini)(struct ptlrpc_request *);
785 struct ptlrpc_cli_req {
786 /** For bulk requests on client only: bulk descriptor */
787 struct ptlrpc_bulk_desc *cr_bulk;
788 /** optional time limit for send attempts. This is a timeout
789 * not a timestamp so timeout_t (s32) is used instead of time64_t
791 timeout_t cr_delay_limit;
792 /** time request was first queued */
793 time64_t cr_queued_time;
794 /** request sent in nanoseconds */
796 /** time for request really sent out */
797 time64_t cr_sent_out;
798 /** when req reply unlink must finish. */
799 time64_t cr_reply_deadline;
800 /** when req bulk unlink must finish. */
801 time64_t cr_bulk_deadline;
802 /** when req unlink must finish. */
803 time64_t cr_req_deadline;
804 /** Portal to which this request would be sent */
806 /** Portal where to wait for reply and where reply would be sent */
808 /** request resending number */
809 unsigned int cr_resend_nr;
810 /** What was import generation when this request was sent */
812 enum lustre_imp_state cr_send_state;
813 /** Per-request waitq introduced by bug 21938 for recovery waiting */
814 wait_queue_head_t cr_set_waitq;
815 /** Link item for request set lists */
816 struct list_head cr_set_chain;
817 /** link to waited ctx */
818 struct list_head cr_ctx_chain;
820 /** client's half ctx */
821 struct ptlrpc_cli_ctx *cr_cli_ctx;
822 /** Link back to the request set */
823 struct ptlrpc_request_set *cr_set;
824 /** outgoing request MD handle */
825 struct lnet_handle_md cr_req_md_h;
826 /** request-out callback parameter */
827 struct ptlrpc_cb_id cr_req_cbid;
828 /** incoming reply MD handle */
829 struct lnet_handle_md cr_reply_md_h;
830 wait_queue_head_t cr_reply_waitq;
831 /** reply callback parameter */
832 struct ptlrpc_cb_id cr_reply_cbid;
833 /** Async completion handler, called when reply is received */
834 ptlrpc_interpterer_t cr_reply_interp;
835 /** Resend handler, called when request is resend to update RPC data */
836 ptlrpc_resend_cb_t cr_resend_cb;
837 /** Async completion context */
838 union ptlrpc_async_args cr_async_args;
839 /** Opaq data for replay and commit callbacks. */
841 /** Link to the imp->imp_unreplied_list */
842 struct list_head cr_unreplied_list;
844 * Commit callback, called when request is committed and about to be
847 void (*cr_commit_cb)(struct ptlrpc_request *);
848 /** Replay callback, called after request is replayed at recovery */
849 void (*cr_replay_cb)(struct ptlrpc_request *);
852 /** client request member alias */
853 /* NB: these alias should NOT be used by any new code, instead they should
854 * be removed step by step to avoid potential abuse
856 #define rq_bulk rq_cli.cr_bulk
857 #define rq_delay_limit rq_cli.cr_delay_limit
858 #define rq_queued_time rq_cli.cr_queued_time
859 #define rq_sent_ns rq_cli.cr_sent_ns
860 #define rq_real_sent rq_cli.cr_sent_out
861 #define rq_reply_deadline rq_cli.cr_reply_deadline
862 #define rq_bulk_deadline rq_cli.cr_bulk_deadline
863 #define rq_req_deadline rq_cli.cr_req_deadline
864 #define rq_nr_resend rq_cli.cr_resend_nr
865 #define rq_request_portal rq_cli.cr_req_ptl
866 #define rq_reply_portal rq_cli.cr_rep_ptl
867 #define rq_import_generation rq_cli.cr_imp_gen
868 #define rq_send_state rq_cli.cr_send_state
869 #define rq_set_chain rq_cli.cr_set_chain
870 #define rq_ctx_chain rq_cli.cr_ctx_chain
871 #define rq_set rq_cli.cr_set
872 #define rq_set_waitq rq_cli.cr_set_waitq
873 #define rq_cli_ctx rq_cli.cr_cli_ctx
874 #define rq_req_md_h rq_cli.cr_req_md_h
875 #define rq_req_cbid rq_cli.cr_req_cbid
876 #define rq_reply_md_h rq_cli.cr_reply_md_h
877 #define rq_reply_waitq rq_cli.cr_reply_waitq
878 #define rq_reply_cbid rq_cli.cr_reply_cbid
879 #define rq_interpret_reply rq_cli.cr_reply_interp
880 #define rq_resend_cb rq_cli.cr_resend_cb
881 #define rq_async_args rq_cli.cr_async_args
882 #define rq_cb_data rq_cli.cr_cb_data
883 #define rq_unreplied_list rq_cli.cr_unreplied_list
884 #define rq_commit_cb rq_cli.cr_commit_cb
885 #define rq_replay_cb rq_cli.cr_replay_cb
887 struct ptlrpc_srv_req {
888 /** initial thread servicing this request */
889 struct ptlrpc_thread *sr_svc_thread;
891 * Server side list of incoming unserved requests sorted by arrival
892 * time. Traversed from time to time to notice about to expire
893 * requests and sent back "early replies" to clients to let them
894 * know server is alive and well, just very busy to service their
897 struct list_head sr_timed_list;
898 /** server-side per-export list */
899 struct list_head sr_exp_list;
900 /** server-side history, used for debuging purposes. */
901 struct list_head sr_hist_list;
902 /** history sequence # */
904 /** the index of service's srv_at_array into which request is linked */
908 /** authed uid mapped to */
909 uid_t sr_auth_mapped_uid;
910 /** RPC is generated from what part of Lustre */
911 enum lustre_sec_part sr_sp_from;
912 /** request session context */
913 struct lu_context sr_ses;
917 /** stub for NRS request */
918 struct ptlrpc_nrs_request sr_nrq;
920 /** request arrival time */
921 struct timespec64 sr_arrival_time;
922 /** server's half ctx */
923 struct ptlrpc_svc_ctx *sr_svc_ctx;
924 /** (server side), pointed directly into req buffer */
925 struct ptlrpc_user_desc *sr_user_desc;
926 /** separated reply state, may be vmalloc'd */
927 struct ptlrpc_reply_state *sr_reply_state;
928 /** server-side hp handlers */
929 struct ptlrpc_hpreq_ops *sr_ops;
930 /** incoming request buffer */
931 struct ptlrpc_request_buffer_desc *sr_rqbd;
934 /** server request member alias */
935 /* NB: these alias should NOT be used by any new code, instead they should
936 * be removed step by step to avoid potential abuse
938 #define rq_svc_thread rq_srv.sr_svc_thread
939 #define rq_timed_list rq_srv.sr_timed_list
940 #define rq_exp_list rq_srv.sr_exp_list
941 #define rq_history_list rq_srv.sr_hist_list
942 #define rq_history_seq rq_srv.sr_hist_seq
943 #define rq_at_index rq_srv.sr_at_index
944 #define rq_auth_uid rq_srv.sr_auth_uid
945 #define rq_auth_mapped_uid rq_srv.sr_auth_mapped_uid
946 #define rq_sp_from rq_srv.sr_sp_from
947 #define rq_session rq_srv.sr_ses
948 #define rq_nrq rq_srv.sr_nrq
949 #define rq_arrival_time rq_srv.sr_arrival_time
950 #define rq_reply_state rq_srv.sr_reply_state
951 #define rq_svc_ctx rq_srv.sr_svc_ctx
952 #define rq_user_desc rq_srv.sr_user_desc
953 #define rq_ops rq_srv.sr_ops
954 #define rq_rqbd rq_srv.sr_rqbd
955 #define rq_reqmsg rq_pill.rc_reqmsg
956 #define rq_repmsg rq_pill.rc_repmsg
957 #define rq_req_swab_mask rq_pill.rc_req_swab_mask
958 #define rq_rep_swab_mask rq_pill.rc_rep_swab_mask
961 * Represents remote procedure call.
963 * This is a staple structure used by everybody wanting to send a request
966 struct ptlrpc_request {
967 /* Request type: one of PTL_RPC_MSG_* */
969 /** Result of request processing */
972 * Linkage item through which this request is included into
973 * sending/delayed lists on client and into rqbd list on server
975 struct list_head rq_list;
976 /** Lock to protect request flags and some other important bits, like
980 spinlock_t rq_early_free_lock;
981 /** client-side flags are serialized by rq_lock @{ */
982 unsigned int rq_intr:1, rq_replied:1, rq_err:1,
983 rq_timedout:1, rq_resend:1, rq_restart:1,
985 * when ->rq_replay is set, request is kept by the client even
986 * after server commits corresponding transaction. This is
987 * used for operations that require sequence of multiple
988 * requests to be replayed. The only example currently is file
989 * open/close. When last request in such a sequence is
990 * committed, ->rq_replay is cleared on all requests in the
994 rq_no_resend:1, rq_waiting:1, rq_receiving_reply:1,
995 rq_no_delay:1, rq_net_err:1, rq_wait_ctx:1,
997 rq_req_unlinked:1, /* unlinked request buffer from lnet */
998 rq_reply_unlinked:1, /* unlinked reply buffer from lnet */
999 rq_memalloc:1, /* req originated from "kswapd" */
1001 rq_reply_truncated:1,
1002 /** whether the "rq_set" is a valid one */
1004 rq_generation_set:1,
1005 /** do not resend request on -EINPROGRESS */
1006 rq_no_retry_einprogress:1,
1007 /* allow req to be sent if the import is in recovery status */
1009 /* bulk request, sent to server, but uncommitted */
1011 rq_early_free_repbuf:1, /* free reply buffer in advance */
1015 /** server-side flags are serialized by rq_lock @{ */
1017 rq_hp:1, /**< high priority RPC */
1018 rq_at_linked:1, /**< link into service's srv_at_array */
1019 rq_packed_final:1, /**< packed final reply */
1020 rq_obsolete:1; /* aborted by a signal on a client */
1023 /** one of RQ_PHASE_* */
1024 enum rq_phase rq_phase;
1025 /** one of RQ_PHASE_* to be used next */
1026 enum rq_phase rq_next_phase;
1028 * client-side refcount for SENT race, server-side refcounf
1029 * for multiple replies
1031 atomic_t rq_refcount;
1034 * !rq_truncate : # reply bytes actually received,
1035 * rq_truncate : required repbuf_len for resend
1037 int rq_nob_received;
1038 /** Request length */
1042 /** Pool if request is from preallocated list */
1043 struct ptlrpc_request_pool *rq_pool;
1044 /** Transaction number */
1048 /** bulk match bits */
1050 /** reply match bits */
1053 * List item to for replay list. Not yet committed requests get linked
1055 * Also see \a rq_replay comment above.
1056 * It's also link chain on obd_export::exp_req_replay_queue
1058 struct list_head rq_replay_list;
1059 /** non-shared members for client & server request*/
1061 struct ptlrpc_cli_req rq_cli;
1062 struct ptlrpc_srv_req rq_srv;
1064 /* security and encryption data */
1065 /* description of flavors for client & server */
1066 struct sptlrpc_flavor rq_flvr;
1068 /* client/server security flags */
1070 rq_ctx_init:1, /* context initiation */
1071 rq_ctx_fini:1, /* context destroy */
1072 rq_bulk_read:1, /* request bulk read */
1073 rq_bulk_write:1, /* request bulk write */
1074 /* server authentication flags */
1075 rq_auth_gss:1, /* authenticated by gss */
1076 rq_auth_usr_root:1, /* authed as root */
1077 rq_auth_usr_mdt:1, /* authed as mdt */
1078 rq_auth_usr_ost:1, /* authed as ost */
1079 /* security tfm flags */
1082 /* doesn't expect reply FIXME */
1084 rq_pill_init:1, /* pill initialized */
1085 rq_srv_req:1; /* server request */
1087 /** various buffer pointers */
1088 struct lustre_msg *rq_reqbuf; /**< req wrapper, vmalloc*/
1089 char *rq_repbuf; /**< rep buffer, vmalloc */
1090 struct lustre_msg *rq_repdata; /**< rep wrapper msg */
1091 /** only in priv mode */
1092 struct lustre_msg *rq_clrbuf;
1093 int rq_reqbuf_len; /* req wrapper buf len */
1094 int rq_reqdata_len; /* req wrapper msg len */
1095 int rq_repbuf_len; /* rep buffer len */
1096 int rq_repdata_len; /* rep wrapper msg len */
1097 int rq_clrbuf_len; /* only in priv mode */
1098 int rq_clrdata_len; /* only in priv mode */
1100 /** early replies go to offset 0, regular replies go after that */
1101 unsigned int rq_reply_off;
1104 /** how many early replies (for stats) */
1106 /** Server-side, export on which request was received */
1107 struct obd_export *rq_export;
1108 /** import where request is being sent */
1109 struct obd_import *rq_import;
1111 struct lnet_nid rq_self;
1112 /** Peer description (the other side) */
1113 struct lnet_processid rq_peer;
1114 /** Descriptor for the NID from which the peer sent the request. */
1115 struct lnet_processid rq_source;
1117 * service time estimate (secs)
1118 * If the request is not served by this time, it is marked as timed out.
1119 * Do not change to time64_t since this is transmitted over the wire.
1121 * The linux kernel handles timestamps with time64_t and timeouts
1122 * are normally done with jiffies. Lustre shares the rq_timeout between
1123 * nodes. Since jiffies can vary from node to node Lustre instead
1124 * will express the timeout value in seconds. To avoid confusion with
1125 * timestamps (time64_t) and jiffy timeouts (long) Lustre timeouts
1126 * are expressed in s32 (timeout_t). Also what is transmitted over
1127 * the wire is 32 bits.
1129 timeout_t rq_timeout;
1131 * when request/reply sent (secs), or time when request should be sent
1134 /** when request must finish. */
1135 time64_t rq_deadline;
1136 /** request format description */
1137 struct req_capsule rq_pill;
1141 * Call completion handler for rpc if any, return it's status or original
1142 * rc if there was no handler defined for this request.
1144 static inline int ptlrpc_req_interpret(const struct lu_env *env,
1145 struct ptlrpc_request *req, int rc)
1147 if (req->rq_interpret_reply != NULL) {
1148 req->rq_status = req->rq_interpret_reply(env, req,
1149 &req->rq_async_args,
1151 return req->rq_status;
1160 void ptlrpc_nrs_req_hp_move(struct ptlrpc_request *req);
1163 * Can the request be moved from the regular NRS head to the high-priority NRS
1164 * head (of the same PTLRPC service partition), if any?
1166 * For a reliable result, this should be checked under svcpt->scp_req lock.
1168 static inline bool ptlrpc_nrs_req_can_move(struct ptlrpc_request *req)
1170 struct ptlrpc_nrs_request *nrq = &req->rq_nrq;
1173 * LU-898: Check ptlrpc_nrs_request::nr_enqueued to make sure the
1174 * request has been enqueued first, and ptlrpc_nrs_request::nr_started
1175 * to make sure it has not been scheduled yet (analogous to previous
1176 * (non-NRS) checking of !list_empty(&ptlrpc_request::rq_list).
1178 return nrq->nr_enqueued && !nrq->nr_started && !req->rq_hp;
1182 static inline bool req_capsule_ptlreq(struct req_capsule *pill)
1184 struct ptlrpc_request *req = pill->rc_req;
1186 return req != NULL && pill == &req->rq_pill;
1189 static inline bool req_capsule_subreq(struct req_capsule *pill)
1191 struct ptlrpc_request *req = pill->rc_req;
1193 return req == NULL || pill != &req->rq_pill;
1197 * Returns true if request needs to be swabbed into local cpu byteorder
1199 static inline bool req_capsule_req_need_swab(struct req_capsule *pill)
1201 struct ptlrpc_request *req = pill->rc_req;
1203 return req && req_capsule_req_swabbed(&req->rq_pill,
1204 MSG_PTLRPC_HEADER_OFF);
1208 * Returns true if request reply needs to be swabbed into local cpu byteorder
1210 static inline bool req_capsule_rep_need_swab(struct req_capsule *pill)
1212 struct ptlrpc_request *req = pill->rc_req;
1214 return req && req_capsule_rep_swabbed(&req->rq_pill,
1215 MSG_PTLRPC_HEADER_OFF);
1219 * Convert numerical request phase value \a phase into text string description
1221 static inline const char *
1222 ptlrpc_phase2str(enum rq_phase phase)
1231 case RQ_PHASE_INTERPRET:
1233 case RQ_PHASE_COMPLETE:
1235 case RQ_PHASE_UNREG_RPC:
1237 case RQ_PHASE_UNREG_BULK:
1245 * Convert numerical request phase of the request \a req into text stringi
1248 static inline const char *
1249 ptlrpc_rqphase2str(struct ptlrpc_request *req)
1251 return ptlrpc_phase2str(req->rq_phase);
1255 * Debugging functions and helpers to print request structure into debug log
1258 /* Spare the preprocessor, spoil the bugs. */
1259 #define FLAG(field, str) (field ? str : "")
1261 /** Convert bit flags into a string */
1262 #define DEBUG_REQ_FLAGS(req) \
1263 ptlrpc_rqphase2str(req), \
1264 FLAG(req->rq_intr, "I"), FLAG(req->rq_replied, "R"), \
1265 FLAG(req->rq_err, "E"), FLAG(req->rq_net_err, "e"), \
1266 FLAG(req->rq_timedout, "X") /* eXpired */, FLAG(req->rq_resend, "S"), \
1267 FLAG(req->rq_restart, "T"), FLAG(req->rq_replay, "P"), \
1268 FLAG(req->rq_no_resend, "N"), FLAG(req->rq_no_reply, "n"), \
1269 FLAG(req->rq_waiting, "W"), \
1270 FLAG(req->rq_wait_ctx, "C"), FLAG(req->rq_hp, "H"), \
1271 FLAG(req->rq_committed, "M"), \
1272 FLAG(req->rq_req_unlinked, "Q"), \
1273 FLAG(req->rq_reply_unlinked, "U"), \
1274 FLAG(req->rq_receiving_reply, "r")
1276 #define REQ_FLAGS_FMT "%s:%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s"
1278 __printf(3, 4) /* __attribute__ */
1279 void _debug_req(struct ptlrpc_request *req,
1280 struct libcfs_debug_msg_data *data, const char *fmt, ...);
1283 * Helper that decides if we need to print request accordig to current debug
1286 #define debug_req(msgdata, mask, cdls, req, fmt, a...) \
1288 if (((mask) & D_CANTMASK) != 0 || \
1289 ((libcfs_debug & (mask)) != 0 && \
1290 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) \
1291 _debug_req((req), msgdata, fmt, ##a); \
1295 * This is the debug print function you need to use to print request sturucture
1296 * content into lustre debug log.
1297 * for most callers (level is a constant) this is resolved at compile time
1299 #define DEBUG_REQ(level, req, fmt, args...) \
1301 if ((level) & (D_ERROR | D_WARNING)) { \
1302 static struct cfs_debug_limit_state cdls; \
1303 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, level, &cdls); \
1304 debug_req(&msgdata, level, &cdls, req, "@@@ "fmt" ", ## args);\
1306 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, level, NULL); \
1307 debug_req(&msgdata, level, NULL, req, "@@@ "fmt" ", ## args); \
1312 enum ptlrpc_bulk_op_type {
1313 PTLRPC_BULK_OP_ACTIVE = 0x00000001,
1314 PTLRPC_BULK_OP_PASSIVE = 0x00000002,
1315 PTLRPC_BULK_OP_PUT = 0x00000004,
1316 PTLRPC_BULK_OP_GET = 0x00000008,
1317 PTLRPC_BULK_GET_SOURCE = PTLRPC_BULK_OP_PASSIVE | PTLRPC_BULK_OP_GET,
1318 PTLRPC_BULK_PUT_SINK = PTLRPC_BULK_OP_PASSIVE | PTLRPC_BULK_OP_PUT,
1319 PTLRPC_BULK_GET_SINK = PTLRPC_BULK_OP_ACTIVE | PTLRPC_BULK_OP_GET,
1320 PTLRPC_BULK_PUT_SOURCE = PTLRPC_BULK_OP_ACTIVE | PTLRPC_BULK_OP_PUT,
1323 static inline bool ptlrpc_is_bulk_op_get(enum ptlrpc_bulk_op_type type)
1325 return (type & PTLRPC_BULK_OP_GET) == PTLRPC_BULK_OP_GET;
1328 static inline bool ptlrpc_is_bulk_get_source(enum ptlrpc_bulk_op_type type)
1330 return (type & PTLRPC_BULK_GET_SOURCE) == PTLRPC_BULK_GET_SOURCE;
1333 static inline bool ptlrpc_is_bulk_put_sink(enum ptlrpc_bulk_op_type type)
1335 return (type & PTLRPC_BULK_PUT_SINK) == PTLRPC_BULK_PUT_SINK;
1338 static inline bool ptlrpc_is_bulk_get_sink(enum ptlrpc_bulk_op_type type)
1340 return (type & PTLRPC_BULK_GET_SINK) == PTLRPC_BULK_GET_SINK;
1343 static inline bool ptlrpc_is_bulk_put_source(enum ptlrpc_bulk_op_type type)
1345 return (type & PTLRPC_BULK_PUT_SOURCE) == PTLRPC_BULK_PUT_SOURCE;
1348 static inline bool ptlrpc_is_bulk_op_active(enum ptlrpc_bulk_op_type type)
1350 return ((type & PTLRPC_BULK_OP_ACTIVE) |
1351 (type & PTLRPC_BULK_OP_PASSIVE))
1352 == PTLRPC_BULK_OP_ACTIVE;
1355 static inline bool ptlrpc_is_bulk_op_passive(enum ptlrpc_bulk_op_type type)
1357 return ((type & PTLRPC_BULK_OP_ACTIVE) |
1358 (type & PTLRPC_BULK_OP_PASSIVE))
1359 == PTLRPC_BULK_OP_PASSIVE;
1362 struct ptlrpc_bulk_frag_ops {
1364 * Add a page \a page to the bulk descriptor \a desc
1365 * Data to transfer in the page starts at offset \a pageoffset and
1366 * amount of data to transfer from the page is \a len
1368 void (*add_kiov_frag)(struct ptlrpc_bulk_desc *desc,
1369 struct page *page, int pageoffset, int len);
1372 * Add a \a fragment to the bulk descriptor \a desc.
1373 * Data to transfer in the fragment is pointed to by \a frag
1374 * The size of the fragment is \a len
1376 int (*add_iov_frag)(struct ptlrpc_bulk_desc *desc, void *frag, int len);
1379 * Uninitialize and free bulk descriptor \a desc.
1380 * Works on bulk descriptors both from server and client side.
1382 void (*release_frags)(struct ptlrpc_bulk_desc *desc);
1385 extern const struct ptlrpc_bulk_frag_ops ptlrpc_bulk_kiov_pin_ops;
1386 extern const struct ptlrpc_bulk_frag_ops ptlrpc_bulk_kiov_nopin_ops;
1389 * Definition of bulk descriptor.
1390 * Bulks are special "Two phase" RPCs where initial request message
1391 * is sent first and it is followed bt a transfer (o receiving) of a large
1392 * amount of data to be settled into pages referenced from the bulk descriptors.
1393 * Bulks transfers (the actual data following the small requests) are done
1394 * on separate LNet portals.
1395 * In lustre we use bulk transfers for READ and WRITE transfers from/to OSTs.
1396 * Another user is readpage for MDT.
1398 struct ptlrpc_bulk_desc {
1399 unsigned int bd_refs; /* number MD's assigned including zero-sends */
1400 /** completed with failure */
1401 unsigned long bd_failure:1;
1403 unsigned long bd_registered:1,
1404 /* bulk request is RDMA transfer, use page->host as real address */
1406 /** For serialization with callback */
1408 /** {put,get}{source,sink}{kvec,kiov} */
1409 enum ptlrpc_bulk_op_type bd_type;
1410 /** LNet portal for this bulk */
1412 /** Server side - export this bulk created for */
1413 struct obd_export *bd_export;
1414 /** Client side - import this bulk was sent on */
1415 struct obd_import *bd_import;
1416 /** Back pointer to the request */
1417 struct ptlrpc_request *bd_req;
1418 const struct ptlrpc_bulk_frag_ops *bd_frag_ops;
1419 wait_queue_head_t bd_waitq; /* server side only WQ */
1420 int bd_iov_count; /* # entries in bd_iov */
1421 int bd_max_iov; /* allocated size of bd_iov */
1422 int bd_nob; /* # bytes covered */
1423 int bd_nob_transferred; /* # bytes GOT/PUT */
1424 unsigned int bd_nob_last; /* # bytes in last MD */
1426 __u64 bd_last_mbits;
1428 struct ptlrpc_cb_id bd_cbid; /* network callback info */
1429 struct lnet_nid bd_sender; /* stash event::sender */
1430 int bd_md_count; /* # valid entries in bd_mds */
1431 int bd_md_max_brw; /* max entries in bd_mds */
1433 /** array of offsets for each MD */
1434 unsigned int bd_mds_off[PTLRPC_BULK_OPS_COUNT];
1435 /** array of associated MDs */
1436 struct lnet_handle_md bd_mds[PTLRPC_BULK_OPS_COUNT];
1438 /* encrypted iov, size is either 0 or bd_iov_count. */
1439 struct bio_vec *bd_enc_vec;
1440 struct bio_vec *bd_vec;
1445 SVC_STOPPED = BIT(0),
1446 SVC_STOPPING = BIT(1),
1447 SVC_STARTING = BIT(2),
1448 SVC_RUNNING = BIT(3),
1451 #define PTLRPC_THR_NAME_LEN 32
1453 * Definition of server service thread structure
1455 struct ptlrpc_thread {
1457 * List of active threads in svcpt->scp_threads
1459 struct list_head t_link;
1461 * thread-private data (preallocated vmalloc'd memory)
1466 * service thread index, from ptlrpc_start_threads
1472 struct task_struct *t_task;
1476 * put watchdog in the structure per thread b=14840
1478 struct delayed_work t_watchdog;
1480 * the svc this thread belonged to b=18582
1482 struct ptlrpc_service_part *t_svcpt;
1483 wait_queue_head_t t_ctl_waitq;
1484 struct lu_env *t_env;
1485 char t_name[PTLRPC_THR_NAME_LEN];
1488 static inline int thread_is_init(struct ptlrpc_thread *thread)
1490 return thread->t_flags == 0;
1493 static inline int thread_is_stopped(struct ptlrpc_thread *thread)
1495 return !!(thread->t_flags & SVC_STOPPED);
1498 static inline int thread_is_stopping(struct ptlrpc_thread *thread)
1500 return !!(thread->t_flags & SVC_STOPPING);
1503 static inline int thread_is_starting(struct ptlrpc_thread *thread)
1505 return !!(thread->t_flags & SVC_STARTING);
1508 static inline int thread_is_running(struct ptlrpc_thread *thread)
1510 return !!(thread->t_flags & SVC_RUNNING);
1513 static inline void thread_clear_flags(struct ptlrpc_thread *thread, __u32 flags)
1515 thread->t_flags &= ~flags;
1518 static inline void thread_set_flags(struct ptlrpc_thread *thread, __u32 flags)
1520 thread->t_flags = flags;
1523 static inline void thread_add_flags(struct ptlrpc_thread *thread, __u32 flags)
1525 thread->t_flags |= flags;
1528 static inline int thread_test_and_clear_flags(struct ptlrpc_thread *thread,
1531 if (thread->t_flags & flags) {
1532 thread->t_flags &= ~flags;
1539 * Request buffer descriptor structure.
1540 * This is a structure that contains one posted request buffer for service.
1541 * Once data land into a buffer, event callback creates actual request and
1542 * notifies wakes one of the service threads to process new incoming request.
1543 * More than one request can fit into the buffer.
1545 struct ptlrpc_request_buffer_desc {
1546 /** Link item for rqbds on a service */
1547 struct list_head rqbd_list;
1548 /** History of requests for this buffer */
1549 struct list_head rqbd_reqs;
1550 /** Back pointer to service for which this buffer is registered */
1551 struct ptlrpc_service_part *rqbd_svcpt;
1552 /** LNet descriptor */
1553 struct lnet_handle_md rqbd_md_h;
1555 /** The buffer itself */
1557 struct ptlrpc_cb_id rqbd_cbid;
1559 * This "embedded" request structure is only used for the
1560 * last request to fit into the buffer
1562 struct ptlrpc_request rqbd_req;
1565 typedef int (*svc_handler_t)(struct ptlrpc_request *req);
1567 struct ptlrpc_service_ops {
1569 * if non-NULL called during thread creation (ptlrpc_start_thread())
1570 * to initialize service specific per-thread state.
1572 int (*so_thr_init)(struct ptlrpc_thread *thr);
1574 * if non-NULL called during thread shutdown (ptlrpc_main()) to
1575 * destruct state created by ->srv_init().
1577 void (*so_thr_done)(struct ptlrpc_thread *thr);
1579 * Handler function for incoming requests for this service
1581 int (*so_req_handler)(struct ptlrpc_request *req);
1583 * function to determine priority of the request, it's called
1584 * on every new request
1586 int (*so_hpreq_handler)(struct ptlrpc_request *);
1588 * service-specific print fn
1590 void (*so_req_printer)(void *, struct ptlrpc_request *);
1593 #ifndef __cfs_cacheline_aligned
1594 /* NB: put it here for reducing patche dependence */
1595 # define __cfs_cacheline_aligned
1599 * How many high priority requests to serve before serving one normal
1602 #define PTLRPC_SVC_HP_RATIO 10
1605 * Definition of PortalRPC service.
1606 * The service is listening on a particular portal (like tcp port)
1607 * and perform actions for a specific server like IO service for OST
1608 * or general metadata service for MDS.
1610 struct ptlrpc_service {
1611 /** serialize /proc operations */
1612 spinlock_t srv_lock;
1613 /** most often accessed fields */
1614 /** chain thru all services */
1615 struct list_head srv_list;
1616 /** service operations table */
1617 struct ptlrpc_service_ops srv_ops;
1618 /** only statically allocated strings here; we don't clean them */
1620 /** only statically allocated strings here; we don't clean them */
1621 char *srv_thread_name;
1622 /** threads # should be created for each partition on initializing */
1623 int srv_nthrs_cpt_init;
1624 /** limit of threads number for each partition */
1625 int srv_nthrs_cpt_limit;
1626 /** Root of debugfs dir tree for this service */
1627 struct dentry *srv_debugfs_entry;
1628 /** Pointer to statistic data for this service */
1629 struct lprocfs_stats *srv_stats;
1630 /** # hp per lp reqs to handle */
1631 int srv_hpreq_ratio;
1632 /** biggest request to receive */
1633 int srv_max_req_size;
1634 /** biggest reply to send */
1635 int srv_max_reply_size;
1636 /** size of individual buffers */
1638 /** # buffers to allocate in 1 group */
1639 int srv_nbuf_per_group;
1640 /** Local portal on which to receive requests */
1641 __u32 srv_req_portal;
1642 /** Portal on the client to send replies to */
1643 __u32 srv_rep_portal;
1645 * Tags for lu_context associated with this thread, see struct
1649 /** soft watchdog timeout multiplier */
1650 int srv_watchdog_factor;
1651 /** under unregister_service */
1652 unsigned srv_is_stopping:1;
1653 /** Whether or not to restrict service threads to CPUs in this CPT */
1654 unsigned srv_cpt_bind:1;
1656 /** max # request buffers */
1658 /** max # request buffers in history per partition */
1659 int srv_hist_nrqbds_cpt_max;
1660 /** number of CPTs this service associated with */
1662 /** CPTs array this service associated with */
1664 /** 2^srv_cptab_bits >= cfs_cpt_numbert(srv_cptable) */
1666 /** CPT table this service is running over */
1667 struct cfs_cpt_table *srv_cptable;
1670 struct kobject srv_kobj;
1671 struct completion srv_kobj_unregister;
1673 * partition data for ptlrpc service
1675 struct ptlrpc_service_part *srv_parts[0];
1679 * Definition of PortalRPC service partition data.
1680 * Although a service only has one instance of it right now, but we
1681 * will have multiple instances very soon (instance per CPT).
1683 * it has four locks:
1685 * serialize operations on rqbd and requests waiting for preprocess
1687 * serialize operations active requests sent to this portal
1689 * serialize adaptive timeout stuff
1691 * serialize operations on RS list (reply states)
1693 * We don't have any use-case to take two or more locks at the same time
1694 * for now, so there is no lock order issue.
1696 struct ptlrpc_service_part {
1697 /** back reference to owner */
1698 struct ptlrpc_service *scp_service __cfs_cacheline_aligned;
1699 /* CPT id, reserved */
1701 /** always increasing number */
1703 /** # of starting threads */
1704 int scp_nthrs_starting;
1705 /** # running threads */
1706 int scp_nthrs_running;
1707 /** service threads list */
1708 struct list_head scp_threads;
1711 * serialize the following fields, used for protecting
1712 * rqbd list and incoming requests waiting for preprocess,
1713 * threads starting & stopping are also protected by this lock.
1715 spinlock_t scp_lock __cfs_cacheline_aligned;
1716 /** userland serialization */
1717 struct mutex scp_mutex;
1718 /** total # req buffer descs allocated */
1719 int scp_nrqbds_total;
1720 /** # posted request buffers for receiving */
1721 int scp_nrqbds_posted;
1722 /** in progress of allocating rqbd */
1723 int scp_rqbd_allocating;
1724 /** # incoming reqs */
1725 int scp_nreqs_incoming;
1726 /** request buffers to be reposted */
1727 struct list_head scp_rqbd_idle;
1728 /** req buffers receiving */
1729 struct list_head scp_rqbd_posted;
1730 /** incoming reqs */
1731 struct list_head scp_req_incoming;
1732 /** timeout before re-posting reqs, in jiffies */
1733 long scp_rqbd_timeout;
1735 * all threads sleep on this. This wait-queue is signalled when new
1736 * incoming request arrives and when difficult reply has to be handled.
1738 wait_queue_head_t scp_waitq;
1740 /** request history */
1741 struct list_head scp_hist_reqs;
1742 /** request buffer history */
1743 struct list_head scp_hist_rqbds;
1744 /** # request buffers in history */
1745 int scp_hist_nrqbds;
1746 /** sequence number for request */
1748 /** highest seq culled from history */
1749 __u64 scp_hist_seq_culled;
1752 * serialize the following fields, used for processing requests
1753 * sent to this portal
1755 spinlock_t scp_req_lock __cfs_cacheline_aligned;
1756 /** # reqs in either of the NRS heads below */
1757 /** # reqs being served */
1758 int scp_nreqs_active;
1759 /** # HPreqs being served */
1760 int scp_nhreqs_active;
1761 /** # hp requests handled */
1764 /** NRS head for regular requests */
1765 struct ptlrpc_nrs scp_nrs_reg;
1766 /** NRS head for HP reqs; valid for services that can handle HP reqs */
1767 struct ptlrpc_nrs *scp_nrs_hp;
1772 * serialize the following fields, used for changes on
1775 spinlock_t scp_at_lock __cfs_cacheline_aligned;
1776 /** estimated rpc service time */
1777 struct adaptive_timeout scp_at_estimate;
1778 /** reqs waiting for replies */
1779 struct ptlrpc_at_array scp_at_array;
1780 /** early reply timer */
1781 struct timer_list scp_at_timer;
1783 ktime_t scp_at_checktime;
1784 /** check early replies */
1785 unsigned int scp_at_check;
1789 * serialize the following fields, used for processing
1790 * replies for this portal
1792 spinlock_t scp_rep_lock __cfs_cacheline_aligned;
1793 /** all the active replies */
1794 struct list_head scp_rep_active;
1795 /** List of free reply_states */
1796 struct list_head scp_rep_idle;
1797 /** waitq to run, when adding stuff to srv_free_rs_list */
1798 wait_queue_head_t scp_rep_waitq;
1799 /** # 'difficult' replies */
1800 atomic_t scp_nreps_difficult;
1803 #define ptlrpc_service_for_each_part(part, i, svc) \
1805 i < (svc)->srv_ncpts && \
1806 (svc)->srv_parts != NULL && \
1807 ((part) = (svc)->srv_parts[i]) != NULL; i++)
1810 * Declaration of ptlrpcd control structure
1812 struct ptlrpcd_ctl {
1814 * Ptlrpc thread control flags (LIOD_START, LIOD_STOP, LIOD_FORCE)
1816 unsigned long pc_flags;
1818 * Thread lock protecting structure fields.
1824 struct completion pc_starting;
1828 struct completion pc_finishing;
1830 * Thread requests set.
1832 struct ptlrpc_request_set *pc_set;
1834 * Thread name used in kthread_run()
1838 * CPT the thread is bound on.
1842 * Index of ptlrpcd thread in the array.
1846 * Pointer to the array of partners' ptlrpcd_ctl structure.
1848 struct ptlrpcd_ctl **pc_partners;
1850 * Number of the ptlrpcd's partners.
1854 * Record the partner index to be processed next.
1858 * Error code if the thread failed to fully start.
1863 /* Bits for pc_flags */
1864 enum ptlrpcd_ctl_flags {
1866 * Ptlrpc thread start flag.
1868 LIOD_START = BIT(0),
1870 * Ptlrpc thread stop flag.
1874 * Ptlrpc thread force flag (only stop force so far).
1875 * This will cause aborting any inflight rpcs handled
1876 * by thread if LIOD_STOP is specified.
1878 LIOD_FORCE = BIT(2),
1880 * This is a recovery ptlrpc thread.
1882 LIOD_RECOVERY = BIT(3),
1889 * Service compatibility function; the policy is compatible with all services.
1891 * \param[in] svc The service the policy is attempting to register with.
1892 * \param[in] desc The policy descriptor
1894 * \retval true The policy is compatible with the service
1896 * \see ptlrpc_nrs_pol_desc::pd_compat()
1898 static inline bool nrs_policy_compat_all(const struct ptlrpc_service *svc,
1899 const struct ptlrpc_nrs_pol_desc *desc)
1905 * Service compatibility function; the policy is compatible with only a specific
1906 * service which is identified by its human-readable name at
1907 * ptlrpc_service::srv_name.
1909 * \param[in] svc The service the policy is attempting to register with.
1910 * \param[in] desc The policy descriptor
1912 * \retval false The policy is not compatible with the service
1913 * \retval true The policy is compatible with the service
1915 * \see ptlrpc_nrs_pol_desc::pd_compat()
1917 static inline bool nrs_policy_compat_one(const struct ptlrpc_service *svc,
1918 const struct ptlrpc_nrs_pol_desc *desc)
1920 LASSERT(desc->pd_compat_svc_name != NULL);
1921 return strcmp(svc->srv_name, desc->pd_compat_svc_name) == 0;
1926 /* ptlrpc/events.c */
1927 extern int ptlrpc_uuid_to_peer(struct obd_uuid *uuid,
1928 struct lnet_processid *peer,
1929 struct lnet_nid *self,
1932 * These callbacks are invoked by LNet when something happened to
1936 extern void request_out_callback(struct lnet_event *ev);
1937 extern void reply_in_callback(struct lnet_event *ev);
1938 extern void client_bulk_callback(struct lnet_event *ev);
1939 extern void request_in_callback(struct lnet_event *ev);
1940 extern void reply_out_callback(struct lnet_event *ev);
1941 #ifdef HAVE_SERVER_SUPPORT
1942 extern void server_bulk_callback(struct lnet_event *ev);
1946 /* ptlrpc/connection.c */
1947 struct ptlrpc_connection *ptlrpc_connection_get(struct lnet_processid *peer,
1948 struct lnet_nid *self,
1949 struct obd_uuid *uuid);
1951 static inline void ptlrpc_connection_put(struct ptlrpc_connection *conn)
1956 LASSERT(atomic_read(&conn->c_refcount) > 0);
1959 * We do not remove connection from hashtable and
1960 * do not free it even if last caller released ref,
1961 * as we want to have it cached for the case it is
1964 * Deallocating it and later creating new connection
1965 * again would be wastful. This way we also avoid
1966 * expensive locking to protect things from get/put
1967 * race when found cached connection is freed by
1968 * ptlrpc_connection_put().
1970 * It will be freed later in module unload time,
1971 * when ptlrpc_connection_fini()->lh_exit->conn_exit()
1974 atomic_dec(&conn->c_refcount);
1976 CDEBUG(D_INFO, "PUT conn=%p refcount %d to %s\n",
1977 conn, atomic_read(&conn->c_refcount),
1978 libcfs_nidstr(&conn->c_peer.nid));
1981 struct ptlrpc_connection *ptlrpc_connection_addref(struct ptlrpc_connection *);
1982 int ptlrpc_connection_init(void);
1983 void ptlrpc_connection_fini(void);
1984 extern lnet_pid_t ptl_get_pid(void);
1987 * Check if the peer connection is on the local node. We need to use GFP_NOFS
1988 * for requests from a local client to avoid recursing into the filesystem
1989 * as we might end up waiting on a page sent in the request we're serving.
1991 * Use __GFP_HIGHMEM so that the pages can use all of the available memory
1992 * on 32-bit machines. Use more aggressive GFP_HIGHUSER flags from non-local
1993 * clients to be able to generate more memory pressure on the OSS and allow
1994 * inactive pages to be reclaimed, since it doesn't have any other processes
1995 * or allocations that generate memory reclaim pressure.
1997 * See b=17576 (bdf50dc9) and b=19529 (3dcf18d3) for details.
1999 static inline bool ptlrpc_connection_is_local(struct ptlrpc_connection *conn)
2004 if (nid_same(&conn->c_peer.nid, &conn->c_self))
2007 RETURN(LNetIsPeerLocal(&conn->c_peer.nid));
2010 /* ptlrpc/niobuf.c */
2012 * Actual interfacing with LNet to put/get/register/unregister stuff
2015 #ifdef HAVE_SERVER_SUPPORT
2016 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_exp(struct ptlrpc_request *req,
2017 unsigned int nfrags,
2018 unsigned int max_brw,
2020 unsigned int portal,
2021 const struct ptlrpc_bulk_frag_ops
2023 int ptlrpc_start_bulk_transfer(struct ptlrpc_bulk_desc *desc);
2024 void ptlrpc_abort_bulk(struct ptlrpc_bulk_desc *desc);
2026 static inline int ptlrpc_server_bulk_active(struct ptlrpc_bulk_desc *desc)
2030 LASSERT(desc != NULL);
2032 spin_lock(&desc->bd_lock);
2034 spin_unlock(&desc->bd_lock);
2039 int ptlrpc_register_bulk(struct ptlrpc_request *req);
2040 int ptlrpc_unregister_bulk(struct ptlrpc_request *req, int async);
2042 static inline int ptlrpc_client_bulk_active(struct ptlrpc_request *req)
2044 struct ptlrpc_bulk_desc *desc;
2047 LASSERT(req != NULL);
2048 desc = req->rq_bulk;
2053 if (req->rq_bulk_deadline > ktime_get_real_seconds())
2057 spin_lock(&desc->bd_lock);
2059 spin_unlock(&desc->bd_lock);
2063 #define PTLRPC_REPLY_MAYBE_DIFFICULT 0x01
2064 #define PTLRPC_REPLY_EARLY 0x02
2065 int ptlrpc_send_reply(struct ptlrpc_request *req, int flags);
2066 int ptlrpc_reply(struct ptlrpc_request *req);
2067 int ptlrpc_send_error(struct ptlrpc_request *req, int difficult);
2068 int ptlrpc_error(struct ptlrpc_request *req);
2069 int ptlrpc_at_get_net_latency(struct ptlrpc_request *req);
2070 int ptl_send_rpc(struct ptlrpc_request *request, int noreply);
2071 int ptlrpc_register_rqbd(struct ptlrpc_request_buffer_desc *rqbd);
2074 /* ptlrpc/client.c */
2076 * Client-side portals API. Everything to send requests, receive replies,
2077 * request queues, request management, etc.
2080 void ptlrpc_request_committed(struct ptlrpc_request *req, int force);
2082 void ptlrpc_init_client(int req_portal, int rep_portal, const char *name,
2083 struct ptlrpc_client *pc);
2084 void ptlrpc_cleanup_client(struct obd_import *imp);
2085 struct ptlrpc_connection *ptlrpc_uuid_to_connection(struct obd_uuid *uuid,
2088 int ptlrpc_queue_wait(struct ptlrpc_request *req);
2089 int ptlrpc_replay_req(struct ptlrpc_request *req);
2090 void ptlrpc_restart_req(struct ptlrpc_request *req);
2091 void ptlrpc_abort_inflight(struct obd_import *imp);
2092 void ptlrpc_cleanup_imp(struct obd_import *imp);
2093 void ptlrpc_abort_set(struct ptlrpc_request_set *set);
2095 struct ptlrpc_request_set *ptlrpc_prep_set(void);
2096 struct ptlrpc_request_set *ptlrpc_prep_fcset(int max, set_producer_func func,
2098 int ptlrpc_check_set(const struct lu_env *env, struct ptlrpc_request_set *set);
2099 int ptlrpc_set_wait(const struct lu_env *env, struct ptlrpc_request_set *set);
2100 void ptlrpc_set_destroy(struct ptlrpc_request_set *set);
2101 void ptlrpc_set_add_req(struct ptlrpc_request_set *set,
2102 struct ptlrpc_request *req);
2103 #define PTLRPCD_SET ((struct ptlrpc_request_set *)1)
2105 void ptlrpc_free_rq_pool(struct ptlrpc_request_pool *pool);
2106 int ptlrpc_add_rqs_to_pool(struct ptlrpc_request_pool *pool, int num_rq);
2108 struct ptlrpc_request_pool *
2109 ptlrpc_init_rq_pool(int num_rq, int msgsiz,
2110 int (*populate_pool)(struct ptlrpc_request_pool *, int));
2112 void ptlrpc_at_set_req_timeout(struct ptlrpc_request *req);
2113 struct ptlrpc_request *ptlrpc_request_alloc(struct obd_import *imp,
2114 const struct req_format *format);
2115 struct ptlrpc_request *ptlrpc_request_alloc_pool(struct obd_import *imp,
2116 struct ptlrpc_request_pool *pr,
2117 const struct req_format *fmat);
2118 void ptlrpc_request_free(struct ptlrpc_request *request);
2119 int ptlrpc_request_pack(struct ptlrpc_request *request, __u32 version,
2121 struct ptlrpc_request *
2122 ptlrpc_request_alloc_pack(struct obd_import *imp,
2123 const struct req_format *format,
2124 __u32 version, int opcode);
2125 int ptlrpc_request_bufs_pack(struct ptlrpc_request *request,
2126 __u32 version, int opcode, char **bufs,
2127 struct ptlrpc_cli_ctx *ctx);
2128 #define ptlrpc_req_finished(rq) ptlrpc_req_put(rq)
2129 void ptlrpc_req_put(struct ptlrpc_request *request);
2130 void ptlrpc_req_put_with_imp_lock(struct ptlrpc_request *request);
2131 struct ptlrpc_request *ptlrpc_request_addref(struct ptlrpc_request *req);
2132 struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_imp(struct ptlrpc_request *req,
2133 unsigned int nfrags,
2134 unsigned int max_brw, unsigned int type,
2135 unsigned int portal,
2136 const struct ptlrpc_bulk_frag_ops *ops);
2138 void __ptlrpc_prep_bulk_page(struct ptlrpc_bulk_desc *desc,
2139 struct page *page, int pageoffset, int len,
2142 void ptlrpc_free_bulk(struct ptlrpc_bulk_desc *bulk);
2144 static inline void ptlrpc_release_bulk_noop(struct ptlrpc_bulk_desc *desc)
2148 void ptlrpc_retain_replayable_request(struct ptlrpc_request *req,
2149 struct obd_import *imp);
2150 __u64 ptlrpc_next_xid(void);
2151 __u64 ptlrpc_sample_next_xid(void);
2152 __u64 ptlrpc_req_xid(struct ptlrpc_request *request);
2153 void ptlrpc_get_mod_rpc_slot(struct ptlrpc_request *req);
2154 void ptlrpc_put_mod_rpc_slot(struct ptlrpc_request *req);
2156 /* Set of routines to run a function in ptlrpcd context */
2157 void *ptlrpcd_alloc_work(struct obd_import *imp,
2158 int (*cb)(const struct lu_env *, void *), void *data);
2159 void ptlrpcd_destroy_work(void *handler);
2160 int ptlrpcd_queue_work(void *handler);
2163 struct ptlrpc_service_buf_conf {
2164 /* nbufs is buffers # to allocate when growing the pool */
2165 unsigned int bc_nbufs;
2166 /* buffer size to post */
2167 unsigned int bc_buf_size;
2168 /* portal to listed for requests on */
2169 unsigned int bc_req_portal;
2170 /* portal of where to send replies to */
2171 unsigned int bc_rep_portal;
2172 /* maximum request size to be accepted for this service */
2173 unsigned int bc_req_max_size;
2174 /* maximum reply size this service can ever send */
2175 unsigned int bc_rep_max_size;
2178 struct ptlrpc_service_thr_conf {
2179 /* threadname should be 8 characters or less - 6 will be added on */
2181 /* threads increasing factor for each CPU */
2182 unsigned int tc_thr_factor;
2183 /* service threads # to start on each partition while initializing */
2184 unsigned int tc_nthrs_init;
2186 * low water of threads # upper-limit on each partition while running,
2187 * service availability may be impacted if threads number is lower
2188 * than this value. It can be ZERO if the service doesn't require
2189 * CPU affinity or there is only one partition.
2191 unsigned int tc_nthrs_base;
2192 /* "soft" limit for total threads number */
2193 unsigned int tc_nthrs_max;
2194 /* user specified threads number, it will be validated due to
2195 * other members of this structure.
2197 unsigned int tc_nthrs_user;
2198 /* bind service threads to only CPUs in their associated CPT */
2199 unsigned int tc_cpu_bind;
2200 /* Tags for lu_context associated with service thread */
2204 struct ptlrpc_service_cpt_conf {
2205 struct cfs_cpt_table *cc_cptable;
2206 /* string pattern to describe CPTs for a service */
2208 /* whether or not to have per-CPT service partitions */
2212 struct ptlrpc_service_conf {
2215 /* soft watchdog timeout multiplifier to print stuck service traces */
2216 unsigned int psc_watchdog_factor;
2217 /* buffer information */
2218 struct ptlrpc_service_buf_conf psc_buf;
2219 /* thread information */
2220 struct ptlrpc_service_thr_conf psc_thr;
2221 /* CPU partition information */
2222 struct ptlrpc_service_cpt_conf psc_cpt;
2223 /* function table */
2224 struct ptlrpc_service_ops psc_ops;
2227 /* ptlrpc/service.c */
2229 * Server-side services API. Register/unregister service, request state
2230 * management, service thread management
2234 void ptlrpc_save_lock(struct ptlrpc_request *req, struct lustre_handle *lock,
2236 void ptlrpc_commit_replies(struct obd_export *exp);
2237 void ptlrpc_dispatch_difficult_reply(struct ptlrpc_reply_state *rs);
2238 void ptlrpc_schedule_difficult_reply(struct ptlrpc_reply_state *rs);
2239 int ptlrpc_hpreq_handler(struct ptlrpc_request *req);
2240 struct ptlrpc_service *ptlrpc_register_service(
2241 struct ptlrpc_service_conf *conf,
2242 struct kset *parent,
2243 struct dentry *debugfs_entry);
2245 int ptlrpc_unregister_service(struct ptlrpc_service *service);
2246 int ptlrpc_service_health_check(struct ptlrpc_service *service);
2247 void ptlrpc_server_drop_request(struct ptlrpc_request *req);
2248 void ptlrpc_request_change_export(struct ptlrpc_request *req,
2249 struct obd_export *export);
2250 void ptlrpc_update_export_timer(struct obd_export *exp,
2251 time64_t extra_delay);
2253 int ptlrpc_hr_init(void);
2254 void ptlrpc_hr_fini(void);
2256 void ptlrpc_watchdog_init(struct delayed_work *work, timeout_t timeout);
2257 void ptlrpc_watchdog_disable(struct delayed_work *work);
2258 void ptlrpc_watchdog_touch(struct delayed_work *work, timeout_t timeout);
2262 /* ptlrpc/import.c */
2267 int ptlrpc_connect_import(struct obd_import *imp);
2268 int ptlrpc_connect_import_locked(struct obd_import *imp);
2269 int ptlrpc_init_import(struct obd_import *imp);
2270 int ptlrpc_disconnect_import(struct obd_import *imp, int noclose);
2271 int ptlrpc_disconnect_and_idle_import(struct obd_import *imp);
2272 int ptlrpc_import_recovery_state_machine(struct obd_import *imp);
2273 void deuuidify(char *uuid, const char *prefix, char **uuid_start,
2275 void ptlrpc_import_enter_resend(struct obd_import *imp);
2276 /* ptlrpc/pack_generic.c */
2277 int ptlrpc_reconnect_import(struct obd_import *imp);
2281 * ptlrpc msg buffer and swab interface
2285 #define PTLRPC_MAX_BUFCOUNT \
2286 (sizeof(((struct ptlrpc_request *)0)->rq_req_swab_mask) * 8)
2287 #define MD_MAX_BUFLEN (MDS_REG_MAXREQSIZE > OUT_MAXREQSIZE ? \
2288 MDS_REG_MAXREQSIZE : OUT_MAXREQSIZE)
2289 #define PTLRPC_MAX_BUFLEN (OST_IO_MAXREQSIZE > MD_MAX_BUFLEN ? \
2290 OST_IO_MAXREQSIZE : MD_MAX_BUFLEN)
2291 int ptlrpc_unpack_rep_msg(struct ptlrpc_request *req, int len);
2292 int ptlrpc_unpack_req_msg(struct ptlrpc_request *req, int len);
2294 int lustre_msg_check_version(struct lustre_msg *msg, __u32 version);
2295 void lustre_init_msg_v2(struct lustre_msg_v2 *msg, int count, __u32 *lens,
2297 int lustre_pack_request(struct ptlrpc_request *req, __u32 magic, int count,
2298 __u32 *lens, char **bufs);
2299 int lustre_pack_reply(struct ptlrpc_request *req, int count, __u32 *lens,
2301 int lustre_pack_reply_v2(struct ptlrpc_request *req, int count,
2302 __u32 *lens, char **bufs, int flags);
2303 #define LPRFL_EARLY_REPLY 1
2304 int lustre_pack_reply_flags(struct ptlrpc_request *, int count, __u32 *lens,
2305 char **bufs, int flags);
2306 int lustre_shrink_msg(struct lustre_msg *msg, int segment,
2307 unsigned int newlen, int move_data);
2308 int lustre_grow_msg(struct lustre_msg *msg, int segment, unsigned int newlen);
2309 void lustre_free_reply_state(struct ptlrpc_reply_state *rs);
2310 int __lustre_unpack_msg(struct lustre_msg *m, int len);
2311 __u32 lustre_msg_hdr_size(__u32 magic, __u32 count);
2312 __u32 lustre_msg_size(__u32 magic, int count, __u32 *lengths);
2313 __u32 lustre_msg_size_v2(int count, __u32 *lengths);
2314 __u32 lustre_packed_msg_size(struct lustre_msg *msg);
2315 extern __u32 lustre_msg_early_size;
2316 void *lustre_msg_buf_v2(struct lustre_msg_v2 *m, __u32 n, __u32 min_size);
2317 void *lustre_msg_buf(struct lustre_msg *m, __u32 n, __u32 minlen);
2318 __u32 lustre_msg_buflen(struct lustre_msg *m, __u32 n);
2319 void lustre_msg_set_buflen(struct lustre_msg *m, __u32 n, __u32 len);
2320 __u32 lustre_msg_bufcount(struct lustre_msg *m);
2321 char *lustre_msg_string(struct lustre_msg *m, __u32 n, __u32 max_len);
2322 __u32 lustre_msghdr_get_flags(struct lustre_msg *msg);
2323 void lustre_msghdr_set_flags(struct lustre_msg *msg, __u32 flags);
2324 __u32 lustre_msg_get_flags(struct lustre_msg *msg);
2325 void lustre_msg_add_flags(struct lustre_msg *msg, __u32 flags);
2326 void lustre_msg_set_flags(struct lustre_msg *msg, __u32 flags);
2327 void lustre_msg_clear_flags(struct lustre_msg *msg, __u32 flags);
2328 __u32 lustre_msg_get_op_flags(struct lustre_msg *msg);
2329 void lustre_msg_add_op_flags(struct lustre_msg *msg, __u32 flags);
2330 struct lustre_handle *lustre_msg_get_handle(struct lustre_msg *msg);
2331 __u32 lustre_msg_get_type(struct lustre_msg *msg);
2332 enum lustre_msg_version lustre_msg_get_version(struct lustre_msg *msg);
2333 void lustre_msg_add_version(struct lustre_msg *msg, __u32 version);
2334 __u32 lustre_msg_get_opc(struct lustre_msg *msg);
2335 __u64 lustre_msg_get_last_xid(struct lustre_msg *msg);
2336 __u16 lustre_msg_get_tag(struct lustre_msg *msg);
2337 __u64 lustre_msg_get_last_committed(struct lustre_msg *msg);
2338 __u64 *lustre_msg_get_versions(struct lustre_msg *msg);
2339 __u64 lustre_msg_get_transno(struct lustre_msg *msg);
2340 __u64 lustre_msg_get_slv(struct lustre_msg *msg);
2341 __u32 lustre_msg_get_limit(struct lustre_msg *msg);
2342 void lustre_msg_set_slv(struct lustre_msg *msg, __u64 slv);
2343 void lustre_msg_set_limit(struct lustre_msg *msg, __u64 limit);
2344 int lustre_msg_get_status(struct lustre_msg *msg);
2345 __u32 lustre_msg_get_conn_cnt(struct lustre_msg *msg);
2346 __u32 lustre_msg_get_magic(struct lustre_msg *msg);
2347 timeout_t lustre_msg_get_timeout(struct lustre_msg *msg);
2348 timeout_t lustre_msg_get_service_timeout(struct lustre_msg *msg);
2349 int lustre_msg_get_uid_gid(struct lustre_msg *msg, __u32 *uid, __u32 *gid);
2350 char *lustre_msg_get_jobid(struct lustre_msg *msg);
2351 __u32 lustre_msg_get_cksum(struct lustre_msg *msg);
2352 __u64 lustre_msg_get_mbits(struct lustre_msg *msg);
2353 __u32 lustre_msg_calc_cksum(struct lustre_msg *msg, __u32 buf);
2354 void lustre_msg_set_handle(struct lustre_msg *msg,
2355 struct lustre_handle *handle);
2356 void lustre_msg_set_type(struct lustre_msg *msg, __u32 type);
2357 void lustre_msg_set_opc(struct lustre_msg *msg, __u32 opc);
2358 void lustre_msg_set_last_xid(struct lustre_msg *msg, __u64 last_xid);
2359 void lustre_msg_set_tag(struct lustre_msg *msg, __u16 tag);
2360 void lustre_msg_set_last_committed(struct lustre_msg *msg,
2361 __u64 last_committed);
2362 void lustre_msg_set_versions(struct lustre_msg *msg, __u64 *versions);
2363 void lustre_msg_set_transno(struct lustre_msg *msg, __u64 transno);
2364 void lustre_msg_set_status(struct lustre_msg *msg, __u32 status);
2365 void lustre_msg_set_conn_cnt(struct lustre_msg *msg, __u32 conn_cnt);
2366 void ptlrpc_request_set_replen(struct ptlrpc_request *req);
2367 void lustre_msg_set_timeout(struct lustre_msg *msg, timeout_t timeout);
2368 void lustre_msg_set_service_timeout(struct lustre_msg *msg,
2369 timeout_t service_timeout);
2370 void lustre_msg_set_uid_gid(struct lustre_msg *msg, __u32 *uid, __u32 *gid);
2371 void lustre_msg_set_jobid(struct lustre_msg *msg, char *jobid);
2372 void lustre_msg_set_cksum(struct lustre_msg *msg, __u32 cksum);
2373 void lustre_msg_set_mbits(struct lustre_msg *msg, __u64 mbits);
2376 lustre_shrink_reply(struct ptlrpc_request *req, int segment,
2377 unsigned int newlen, int move_data)
2379 LASSERT(req->rq_reply_state);
2380 LASSERT(req->rq_repmsg);
2381 req->rq_replen = lustre_shrink_msg(req->rq_repmsg, segment,
2385 #ifdef LUSTRE_TRANSLATE_ERRNOS
2387 static inline int ptlrpc_status_hton(int h)
2390 * Positive errnos must be network errnos, such as LUSTRE_EDEADLK,
2391 * ELDLM_LOCK_ABORTED, etc.
2394 return -lustre_errno_hton(-h);
2399 static inline int ptlrpc_status_ntoh(int n)
2402 * See the comment in ptlrpc_status_hton().
2405 return -lustre_errno_ntoh(-n);
2412 #define ptlrpc_status_hton(h) (h)
2413 #define ptlrpc_status_ntoh(n) (n)
2418 /** Change request phase of \a req to \a new_phase */
2420 ptlrpc_rqphase_move(struct ptlrpc_request *req, enum rq_phase new_phase)
2422 if (req->rq_phase == new_phase)
2425 if (new_phase == RQ_PHASE_UNREG_RPC ||
2426 new_phase == RQ_PHASE_UNREG_BULK) {
2427 /* No embedded unregistering phases */
2428 if (req->rq_phase == RQ_PHASE_UNREG_RPC ||
2429 req->rq_phase == RQ_PHASE_UNREG_BULK)
2432 req->rq_next_phase = req->rq_phase;
2434 atomic_inc(&req->rq_import->imp_unregistering);
2437 if (req->rq_phase == RQ_PHASE_UNREG_RPC ||
2438 req->rq_phase == RQ_PHASE_UNREG_BULK) {
2440 atomic_dec(&req->rq_import->imp_unregistering);
2443 DEBUG_REQ(D_INFO, req, "move request phase from %s to %s",
2444 ptlrpc_rqphase2str(req), ptlrpc_phase2str(new_phase));
2446 req->rq_phase = new_phase;
2450 * Returns true if request \a req got early reply and hard deadline is not met
2453 ptlrpc_client_early(struct ptlrpc_request *req)
2455 return req->rq_early;
2459 * Returns true if we got real reply from server for this request
2462 ptlrpc_client_replied(struct ptlrpc_request *req)
2464 if (req->rq_reply_deadline > ktime_get_real_seconds())
2466 return req->rq_replied;
2469 /** Returns true if request \a req is in process of receiving server reply */
2471 ptlrpc_client_recv(struct ptlrpc_request *req)
2473 if (req->rq_reply_deadline > ktime_get_real_seconds())
2475 return req->rq_receiving_reply;
2479 ptlrpc_cli_wait_unlink(struct ptlrpc_request *req)
2483 spin_lock(&req->rq_lock);
2484 if (req->rq_reply_deadline > ktime_get_real_seconds()) {
2485 spin_unlock(&req->rq_lock);
2488 if (req->rq_req_deadline > ktime_get_real_seconds()) {
2489 spin_unlock(&req->rq_lock);
2493 rc = !req->rq_req_unlinked || !req->rq_reply_unlinked ||
2494 req->rq_receiving_reply;
2495 spin_unlock(&req->rq_lock);
2500 ptlrpc_client_wake_req(struct ptlrpc_request *req)
2502 smp_mb(); /* ensure client context is deleted before wakeing up */
2503 if (req->rq_set == NULL)
2504 wake_up(&req->rq_reply_waitq);
2506 wake_up(&req->rq_set->set_waitq);
2510 ptlrpc_rs_addref(struct ptlrpc_reply_state *rs)
2512 LASSERT(atomic_read(&rs->rs_refcount) > 0);
2513 atomic_inc(&rs->rs_refcount);
2517 ptlrpc_rs_decref(struct ptlrpc_reply_state *rs)
2519 LASSERT(atomic_read(&rs->rs_refcount) > 0);
2520 if (atomic_dec_and_test(&rs->rs_refcount))
2521 lustre_free_reply_state(rs);
2524 /* Should only be called once per req */
2525 static inline void ptlrpc_req_drop_rs(struct ptlrpc_request *req)
2527 if (req->rq_reply_state == NULL)
2528 return; /* shouldn't occur */
2530 /* req_repmsg equals rq_reply_state->rs_msg,
2531 * so set it to NULL before rq_reply_state is possibly freed
2533 spin_lock(&req->rq_early_free_lock);
2534 req->rq_repmsg = NULL;
2535 spin_unlock(&req->rq_early_free_lock);
2537 ptlrpc_rs_decref(req->rq_reply_state);
2538 req->rq_reply_state = NULL;
2541 static inline __u32 lustre_request_magic(struct ptlrpc_request *req)
2543 return lustre_msg_get_magic(req->rq_reqmsg);
2546 static inline int ptlrpc_send_limit_expired(struct ptlrpc_request *req)
2548 if (req->rq_delay_limit != 0 &&
2549 req->rq_queued_time + req->rq_delay_limit < ktime_get_seconds())
2554 static inline int ptlrpc_no_resend(struct ptlrpc_request *req)
2556 if (!req->rq_no_resend && ptlrpc_send_limit_expired(req)) {
2557 spin_lock(&req->rq_lock);
2558 req->rq_no_resend = 1;
2559 spin_unlock(&req->rq_lock);
2561 return req->rq_no_resend;
2564 int ptlrpc_server_get_timeout(struct ptlrpc_service_part *svcpt);
2566 static inline struct ptlrpc_service *
2567 ptlrpc_req2svc(struct ptlrpc_request *req)
2569 LASSERT(req->rq_rqbd != NULL);
2570 return req->rq_rqbd->rqbd_svcpt->scp_service;
2573 /* ldlm/ldlm_lib.c */
2575 * Target client logic
2578 int client_obd_setup(struct obd_device *obd, struct lustre_cfg *lcfg);
2579 int client_obd_cleanup(struct obd_device *obd);
2580 int client_connect_import(const struct lu_env *env,
2581 struct obd_export **exp, struct obd_device *obd,
2582 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
2584 int client_disconnect_export(struct obd_export *exp);
2585 int client_import_add_conn(struct obd_import *imp, struct obd_uuid *uuid,
2587 int client_import_dyn_add_conn(struct obd_import *imp, struct obd_uuid *uuid,
2588 struct lnet_nid *prim_nid, int priority);
2589 int client_import_add_nids_to_conn(struct obd_import *imp,
2590 struct lnet_nid *nidlist,
2591 int nid_count, int nid_size,
2592 struct obd_uuid *uuid);
2593 int client_import_del_conn(struct obd_import *imp, struct obd_uuid *uuid);
2594 int import_set_conn_priority(struct obd_import *imp, struct obd_uuid *uuid);
2595 void client_destroy_import(struct obd_import *imp);
2598 #ifdef HAVE_SERVER_SUPPORT
2599 int server_disconnect_export(struct obd_export *exp);
2602 /* ptlrpc/pinger.c */
2604 * Pinger API (client side only)
2607 enum timeout_event {
2610 struct timeout_item;
2611 typedef int (*timeout_cb_t)(struct timeout_item *, void *);
2612 int ptlrpc_pinger_add_import(struct obd_import *imp);
2613 int ptlrpc_pinger_del_import(struct obd_import *imp);
2614 struct ptlrpc_request *ptlrpc_prep_ping(struct obd_import *imp);
2615 int ptlrpc_obd_ping(struct obd_device *obd);
2616 void ping_evictor_start(void);
2617 void ping_evictor_stop(void);
2618 void ptlrpc_pinger_ir_up(void);
2619 void ptlrpc_pinger_ir_down(void);
2621 int ptlrpc_pinger_suppress_pings(void);
2623 /* ptlrpc/ptlrpcd.c */
2624 void ptlrpcd_stop(struct ptlrpcd_ctl *pc, int force);
2625 void ptlrpcd_free(struct ptlrpcd_ctl *pc);
2626 void ptlrpcd_wake(struct ptlrpc_request *req);
2627 void ptlrpcd_add_req(struct ptlrpc_request *req);
2628 void ptlrpcd_add_rqset(struct ptlrpc_request_set *set);
2629 int ptlrpcd_addref(void);
2630 void ptlrpcd_decref(void);
2632 /* ptlrpc/lproc_ptlrpc.c */
2634 * procfs output related functions
2637 const char *ll_opcode2str(__u32 opcode);
2638 const int ll_str2opcode(const char *ops);
2639 #ifdef CONFIG_PROC_FS
2640 void ptlrpc_lprocfs_register_obd(struct obd_device *obd);
2641 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd);
2642 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes);
2644 static inline void ptlrpc_lprocfs_register_obd(struct obd_device *obd) {}
2645 static inline void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd) {}
2646 static inline void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes) {}
2650 /* ptlrpc/llog_server.c */
2651 int llog_origin_handle_open(struct ptlrpc_request *req);
2652 int llog_origin_handle_prev_block(struct ptlrpc_request *req);
2653 int llog_origin_handle_next_block(struct ptlrpc_request *req);
2654 int llog_origin_handle_read_header(struct ptlrpc_request *req);
2656 /* ptlrpc/llog_client.c */
2657 extern const struct llog_operations llog_client_ops;