-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
- *
+/*
* GPL HEADER START
*
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
/*
* Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
- */
-/*
- * Copyright (c) 2011 Whamcloud, Inc.
+ *
+ * Copyright (c) 2010, 2014, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
* @{
*/
-#if defined(__linux__)
-#include <linux/lustre_net.h>
-#elif defined(__APPLE__)
-#include <darwin/lustre_net.h>
-#elif defined(__WINNT__)
-#include <winnt/lustre_net.h>
-#else
-#error Unsupported operating system.
-#endif
-
#include <libcfs/libcfs.h>
-// #include <obd.h>
-#include <lnet/lnet.h>
+#include <lnet/nidstr.h>
+#include <lnet/api.h>
#include <lustre/lustre_idl.h>
#include <lustre_ha.h>
#include <lustre_sec.h>
#include <lprocfs_status.h>
#include <lu_object.h>
#include <lustre_req_layout.h>
-
#include <obd_support.h>
#include <lustre_ver.h>
#define PTLRPC_MD_OPTIONS 0
/**
- * Define maxima for bulk I/O
- * CAVEAT EMPTOR, with multinet (i.e. routers forwarding between networks)
- * these limits are system wide and not interface-local. */
-#define PTLRPC_MAX_BRW_BITS LNET_MTU_BITS
-#define PTLRPC_MAX_BRW_SIZE (1<<LNET_MTU_BITS)
-#define PTLRPC_MAX_BRW_PAGES (PTLRPC_MAX_BRW_SIZE >> CFS_PAGE_SHIFT)
+ * Max # of bulk operations in one request.
+ * In order for the client and server to properly negotiate the maximum
+ * possible transfer size, PTLRPC_BULK_OPS_COUNT must be a power-of-two
+ * value. The client is free to limit the actual RPC size for any bulk
+ * transfer via cl_max_pages_per_rpc to some non-power-of-two value. */
+#define PTLRPC_BULK_OPS_BITS 2
+#define PTLRPC_BULK_OPS_COUNT (1U << PTLRPC_BULK_OPS_BITS)
+/**
+ * PTLRPC_BULK_OPS_MASK is for the convenience of the client only, and
+ * should not be used on the server at all. Otherwise, it imposes a
+ * protocol limitation on the maximum RPC size that can be used by any
+ * RPC sent to that server in the future. Instead, the server should
+ * use the negotiated per-client ocd_brw_size to determine the bulk
+ * RPC count. */
+#define PTLRPC_BULK_OPS_MASK (~((__u64)PTLRPC_BULK_OPS_COUNT - 1))
+
+/**
+ * Define maxima for bulk I/O.
+ *
+ * A single PTLRPC BRW request is sent via up to PTLRPC_BULK_OPS_COUNT
+ * of LNET_MTU sized RDMA transfers. Clients and servers negotiate the
+ * currently supported maximum between peers at connect via ocd_brw_size.
+ */
+#define PTLRPC_MAX_BRW_BITS (LNET_MTU_BITS + PTLRPC_BULK_OPS_BITS)
+#define PTLRPC_MAX_BRW_SIZE (1 << PTLRPC_MAX_BRW_BITS)
+#define PTLRPC_MAX_BRW_PAGES (PTLRPC_MAX_BRW_SIZE >> PAGE_CACHE_SHIFT)
+
+#define ONE_MB_BRW_SIZE (1 << LNET_MTU_BITS)
+#define MD_MAX_BRW_SIZE (1 << LNET_MTU_BITS)
+#define MD_MAX_BRW_PAGES (MD_MAX_BRW_SIZE >> PAGE_CACHE_SHIFT)
+#define DT_MAX_BRW_SIZE PTLRPC_MAX_BRW_SIZE
+#define DT_MAX_BRW_PAGES (DT_MAX_BRW_SIZE >> PAGE_CACHE_SHIFT)
+#define OFD_MAX_BRW_SIZE (1 << LNET_MTU_BITS)
/* When PAGE_SIZE is a constant, we can check our arithmetic here with cpp! */
-#ifdef __KERNEL__
-# if ((PTLRPC_MAX_BRW_PAGES & (PTLRPC_MAX_BRW_PAGES - 1)) != 0)
-# error "PTLRPC_MAX_BRW_PAGES isn't a power of two"
-# endif
-# if (PTLRPC_MAX_BRW_SIZE != (PTLRPC_MAX_BRW_PAGES * CFS_PAGE_SIZE))
-# error "PTLRPC_MAX_BRW_SIZE isn't PTLRPC_MAX_BRW_PAGES * CFS_PAGE_SIZE"
-# endif
-# if (PTLRPC_MAX_BRW_SIZE > LNET_MTU)
-# error "PTLRPC_MAX_BRW_SIZE too big"
-# endif
-# if (PTLRPC_MAX_BRW_PAGES > LNET_MAX_IOV)
-# error "PTLRPC_MAX_BRW_PAGES too big"
-# endif
-#endif /* __KERNEL__ */
-
-/**
- * The following constants determine how memory is used to buffer incoming
- * service requests.
+#if ((PTLRPC_MAX_BRW_PAGES & (PTLRPC_MAX_BRW_PAGES - 1)) != 0)
+# error "PTLRPC_MAX_BRW_PAGES isn't a power of two"
+#endif
+#if (PTLRPC_MAX_BRW_SIZE != (PTLRPC_MAX_BRW_PAGES * PAGE_CACHE_SIZE))
+# error "PTLRPC_MAX_BRW_SIZE isn't PTLRPC_MAX_BRW_PAGES * PAGE_CACHE_SIZE"
+#endif
+#if (PTLRPC_MAX_BRW_SIZE > LNET_MTU * PTLRPC_BULK_OPS_COUNT)
+# error "PTLRPC_MAX_BRW_SIZE too big"
+#endif
+#if (PTLRPC_MAX_BRW_PAGES > LNET_MAX_IOV * PTLRPC_BULK_OPS_COUNT)
+# error "PTLRPC_MAX_BRW_PAGES too big"
+#endif
+
+#define PTLRPC_NTHRS_INIT 2
+
+/**
+ * Buffer Constants
+ *
+ * Constants determine how memory is used to buffer incoming service requests.
*
* ?_NBUFS # buffers to allocate when growing the pool
* ?_BUFSIZE # bytes in a single request buffer
* Messages larger than ?_MAXREQSIZE are dropped. Request buffers are
* considered full when less than ?_MAXREQSIZE is left in them.
*/
-#define LDLM_THREADS_AUTO_MIN (2)
-#define LDLM_THREADS_AUTO_MAX min_t(unsigned, cfs_num_online_cpus() * \
- cfs_num_online_cpus() * 32, 128)
-#define LDLM_BL_THREADS LDLM_THREADS_AUTO_MIN
-#define LDLM_NBUFS (64 * cfs_num_online_cpus())
-#define LDLM_BUFSIZE (8 * 1024)
-#define LDLM_MAXREQSIZE (5 * 1024)
-#define LDLM_MAXREPSIZE (1024)
-
-/** Absolute limits */
-#define MDT_MIN_THREADS 2UL
-#ifndef MDT_MAX_THREADS
-#define MDT_MAX_THREADS 512UL
+/**
+ * Thread Constants
+ *
+ * Constants determine how threads are created for ptlrpc service.
+ *
+ * ?_NTHRS_INIT # threads to create for each service partition on
+ * initializing. If it's non-affinity service and
+ * there is only one partition, it's the overall #
+ * threads for the service while initializing.
+ * ?_NTHRS_BASE # threads should be created at least for each
+ * ptlrpc partition to keep the service healthy.
+ * It's the low-water mark of threads upper-limit
+ * for each partition.
+ * ?_THR_FACTOR # threads can be added on threads upper-limit for
+ * each CPU core. This factor is only for reference,
+ * we might decrease value of factor if number of cores
+ * per CPT is above a limit.
+ * ?_NTHRS_MAX # overall threads can be created for a service,
+ * it's a soft limit because if service is running
+ * on machine with hundreds of cores and tens of
+ * CPU partitions, we need to guarantee each partition
+ * has ?_NTHRS_BASE threads, which means total threads
+ * will be ?_NTHRS_BASE * number_of_cpts which can
+ * exceed ?_NTHRS_MAX.
+ *
+ * Examples
+ *
+ * #define MDS_NTHRS_INIT 2
+ * #define MDS_NTHRS_BASE 64
+ * #define MDS_NTHRS_FACTOR 8
+ * #define MDS_NTHRS_MAX 1024
+ *
+ * Example 1):
+ * ---------------------------------------------------------------------
+ * Server(A) has 16 cores, user configured it to 4 partitions so each
+ * partition has 4 cores, then actual number of service threads on each
+ * partition is:
+ * MDS_NTHRS_BASE(64) + cores(4) * MDS_NTHRS_FACTOR(8) = 96
+ *
+ * Total number of threads for the service is:
+ * 96 * partitions(4) = 384
+ *
+ * Example 2):
+ * ---------------------------------------------------------------------
+ * Server(B) has 32 cores, user configured it to 4 partitions so each
+ * partition has 8 cores, then actual number of service threads on each
+ * partition is:
+ * MDS_NTHRS_BASE(64) + cores(8) * MDS_NTHRS_FACTOR(8) = 128
+ *
+ * Total number of threads for the service is:
+ * 128 * partitions(4) = 512
+ *
+ * Example 3):
+ * ---------------------------------------------------------------------
+ * Server(B) has 96 cores, user configured it to 8 partitions so each
+ * partition has 12 cores, then actual number of service threads on each
+ * partition is:
+ * MDS_NTHRS_BASE(64) + cores(12) * MDS_NTHRS_FACTOR(8) = 160
+ *
+ * Total number of threads for the service is:
+ * 160 * partitions(8) = 1280
+ *
+ * However, it's above the soft limit MDS_NTHRS_MAX, so we choose this number
+ * as upper limit of threads number for each partition:
+ * MDS_NTHRS_MAX(1024) / partitions(8) = 128
+ *
+ * Example 4):
+ * ---------------------------------------------------------------------
+ * Server(C) have a thousand of cores and user configured it to 32 partitions
+ * MDS_NTHRS_BASE(64) * 32 = 2048
+ *
+ * which is already above soft limit MDS_NTHRS_MAX(1024), but we still need
+ * to guarantee that each partition has at least MDS_NTHRS_BASE(64) threads
+ * to keep service healthy, so total number of threads will just be 2048.
+ *
+ * NB: we don't suggest to choose server with that many cores because backend
+ * filesystem itself, buffer cache, or underlying network stack might
+ * have some SMP scalability issues at that large scale.
+ *
+ * If user already has a fat machine with hundreds or thousands of cores,
+ * there are two choices for configuration:
+ * a) create CPU table from subset of all CPUs and run Lustre on
+ * top of this subset
+ * b) bind service threads on a few partitions, see modparameters of
+ * MDS and OSS for details
+*
+ * NB: these calculations (and examples below) are simplified to help
+ * understanding, the real implementation is a little more complex,
+ * please see ptlrpc_server_nthreads_check() for details.
+ *
+ */
+
+ /*
+ * LDLM threads constants:
+ *
+ * Given 8 as factor and 24 as base threads number
+ *
+ * example 1)
+ * On 4-core machine we will have 24 + 8 * 4 = 56 threads.
+ *
+ * example 2)
+ * On 8-core machine with 2 partitions we will have 24 + 4 * 8 = 56
+ * threads for each partition and total threads number will be 112.
+ *
+ * example 3)
+ * On 64-core machine with 8 partitions we will need LDLM_NTHRS_BASE(24)
+ * threads for each partition to keep service healthy, so total threads
+ * number should be 24 * 8 = 192.
+ *
+ * So with these constants, threads number will be at the similar level
+ * of old versions, unless target machine has over a hundred cores
+ */
+#define LDLM_THR_FACTOR 8
+#define LDLM_NTHRS_INIT PTLRPC_NTHRS_INIT
+#define LDLM_NTHRS_BASE 24
+#define LDLM_NTHRS_MAX (num_online_cpus() == 1 ? 64 : 128)
+
+#define LDLM_BL_THREADS LDLM_NTHRS_AUTO_INIT
+#define LDLM_CLIENT_NBUFS 1
+#define LDLM_SERVER_NBUFS 64
+#define LDLM_BUFSIZE (8 * 1024)
+#define LDLM_MAXREQSIZE (5 * 1024)
+#define LDLM_MAXREPSIZE (1024)
+
+ /*
+ * MDS threads constants:
+ *
+ * Please see examples in "Thread Constants", MDS threads number will be at
+ * the comparable level of old versions, unless the server has many cores.
+ */
+#ifndef MDS_MAX_THREADS
+#define MDS_MAX_THREADS 1024
+#define MDS_MAX_OTHR_THREADS 256
+
+#else /* MDS_MAX_THREADS */
+#if MDS_MAX_THREADS < PTLRPC_NTHRS_INIT
+#undef MDS_MAX_THREADS
+#define MDS_MAX_THREADS PTLRPC_NTHRS_INIT
#endif
-#define MDS_NBUFS (64 * cfs_num_online_cpus())
+#define MDS_MAX_OTHR_THREADS max(PTLRPC_NTHRS_INIT, MDS_MAX_THREADS / 2)
+#endif
+
+/* default service */
+#define MDS_THR_FACTOR 8
+#define MDS_NTHRS_INIT PTLRPC_NTHRS_INIT
+#define MDS_NTHRS_MAX MDS_MAX_THREADS
+#define MDS_NTHRS_BASE min(64, MDS_NTHRS_MAX)
+
+/* read-page service */
+#define MDS_RDPG_THR_FACTOR 4
+#define MDS_RDPG_NTHRS_INIT PTLRPC_NTHRS_INIT
+#define MDS_RDPG_NTHRS_MAX MDS_MAX_OTHR_THREADS
+#define MDS_RDPG_NTHRS_BASE min(48, MDS_RDPG_NTHRS_MAX)
+
+/* these should be removed when we remove setattr service in the future */
+#define MDS_SETA_THR_FACTOR 4
+#define MDS_SETA_NTHRS_INIT PTLRPC_NTHRS_INIT
+#define MDS_SETA_NTHRS_MAX MDS_MAX_OTHR_THREADS
+#define MDS_SETA_NTHRS_BASE min(48, MDS_SETA_NTHRS_MAX)
+
+/* non-affinity threads */
+#define MDS_OTHR_NTHRS_INIT PTLRPC_NTHRS_INIT
+#define MDS_OTHR_NTHRS_MAX MDS_MAX_OTHR_THREADS
+
+#define MDS_NBUFS 64
+
/**
* Assume file name length = FNAME_MAX = 256 (true for ext3).
- * path name length = PATH_MAX = 4096
- * LOV MD size max = EA_MAX = 48000 (2000 stripes)
+ * path name length = PATH_MAX = 4096
+ * LOV MD size max = EA_MAX = 24 * 2000
+ * (NB: 24 is size of lov_ost_data)
+ * LOV LOGCOOKIE size max = 32 * 2000
+ * (NB: 32 is size of llog_cookie)
* symlink: FNAME_MAX + PATH_MAX <- largest
* link: FNAME_MAX + PATH_MAX (mds_rec_link < mds_rec_create)
* rename: FNAME_MAX + FNAME_MAX
* MDS_MAXREQSIZE ~= 4736 bytes =
* lustre_msg + ldlm_request + mdt_body + mds_rec_create + FNAME_MAX + PATH_MAX
* MDS_MAXREPSIZE ~= 8300 bytes = lustre_msg + llog_header
- * or, for mds_close() and mds_reint_unlink() on a many-OST filesystem:
- * = 9210 bytes = lustre_msg + mdt_body + 160 * (easize + cookiesize)
*
* Realistic size is about 512 bytes (20 character name + 128 char symlink),
* except in the open case where there are a large number of OSTs in a LOV.
*/
-#define MDS_MAXREPSIZE max(10 * 1024, 362 + LOV_MAX_STRIPE_COUNT * 56)
-#define MDS_MAXREQSIZE MDS_MAXREPSIZE
+#define MDS_MAXREQSIZE (5 * 1024) /* >= 4736 */
+#define MDS_MAXREPSIZE (9 * 1024) /* >= 8300 */
-/** MDS_BUFSIZE = max_reqsize + max sptlrpc payload size */
-#define MDS_BUFSIZE (MDS_MAXREQSIZE + 1024)
+/**
+ * MDS incoming request with LOV EA
+ * 24 = sizeof(struct lov_ost_data), i.e: replay of opencreate
+ */
+#define MDS_LOV_MAXREQSIZE max(MDS_MAXREQSIZE, \
+ 362 + LOV_MAX_STRIPE_COUNT * 24)
+/**
+ * MDS outgoing reply with LOV EA
+ *
+ * NB: max reply size Lustre 2.4+ client can get from old MDS is:
+ * LOV_MAX_STRIPE_COUNT * (llog_cookie + lov_ost_data) + extra bytes
+ *
+ * but 2.4 or later MDS will never send reply with llog_cookie to any
+ * version client. This macro is defined for server side reply buffer size.
+ */
+#define MDS_LOV_MAXREPSIZE MDS_LOV_MAXREQSIZE
+
+/**
+ * This is the size of a maximum REINT_SETXATTR request:
+ *
+ * lustre_msg 56 (32 + 4 x 5 + 4)
+ * ptlrpc_body 184
+ * mdt_rec_setxattr 136
+ * lustre_capa 120
+ * name 256 (XATTR_NAME_MAX)
+ * value 65536 (XATTR_SIZE_MAX)
+ */
+#define MDS_EA_MAXREQSIZE 66288
+
+/**
+ * These are the maximum request and reply sizes (rounded up to 1 KB
+ * boundaries) for the "regular" MDS_REQUEST_PORTAL and MDS_REPLY_PORTAL.
+ */
+#define MDS_REG_MAXREQSIZE (((max(MDS_EA_MAXREQSIZE, \
+ MDS_LOV_MAXREQSIZE) + 1023) >> 10) << 10)
+#define MDS_REG_MAXREPSIZE MDS_REG_MAXREQSIZE
+
+/**
+ * The update request includes all of updates from the create, which might
+ * include linkea (4K maxim), together with other updates, we set it to 9K:
+ * lustre_msg + ptlrpc_body + UPDATE_BUF_SIZE (8K)
+ */
+#define OUT_MAXREQSIZE (9 * 1024)
+#define OUT_MAXREPSIZE MDS_MAXREPSIZE
+
+/** MDS_BUFSIZE = max_reqsize (w/o LOV EA) + max sptlrpc payload size */
+#define MDS_BUFSIZE max(MDS_MAXREQSIZE + SPTLRPC_MAX_PAYLOAD, \
+ 8 * 1024)
+
+/**
+ * MDS_REG_BUFSIZE should at least be MDS_REG_MAXREQSIZE + SPTLRPC_MAX_PAYLOAD.
+ * However, we need to allocate a much larger buffer for it because LNet
+ * requires each MD(rqbd) has at least MDS_REQ_MAXREQSIZE bytes left to avoid
+ * dropping of maximum-sized incoming request. So if MDS_REG_BUFSIZE is only a
+ * little larger than MDS_REG_MAXREQSIZE, then it can only fit in one request
+ * even there are about MDS_REG_MAX_REQSIZE bytes left in a rqbd, and memory
+ * utilization is very low.
+ *
+ * In the meanwhile, size of rqbd can't be too large, because rqbd can't be
+ * reused until all requests fit in it have been processed and released,
+ * which means one long blocked request can prevent the rqbd be reused.
+ * Now we set request buffer size to 160 KB, so even each rqbd is unlinked
+ * from LNet with unused 65 KB, buffer utilization will be about 59%.
+ * Please check LU-2432 for details.
+ */
+#define MDS_REG_BUFSIZE max(MDS_REG_MAXREQSIZE + SPTLRPC_MAX_PAYLOAD, \
+ 160 * 1024)
+
+/**
+ * OUT_BUFSIZE = max_out_reqsize + max sptlrpc payload (~1K) which is
+ * about 10K, for the same reason as MDS_REG_BUFSIZE, we also give some
+ * extra bytes to each request buffer to improve buffer utilization rate.
+ */
+#define OUT_BUFSIZE max(OUT_MAXREQSIZE + SPTLRPC_MAX_PAYLOAD, \
+ 24 * 1024)
/** FLD_MAXREQSIZE == lustre_msg + __u32 padding + ptlrpc_body + opc */
#define FLD_MAXREQSIZE (160)
/** FLD_MAXREPSIZE == lustre_msg + ptlrpc_body */
#define FLD_MAXREPSIZE (152)
+#define FLD_BUFSIZE (1 << 12)
/**
* SEQ_MAXREQSIZE == lustre_msg + __u32 padding + ptlrpc_body + opc + lu_range +
/** SEQ_MAXREPSIZE == lustre_msg + ptlrpc_body + lu_range */
#define SEQ_MAXREPSIZE (152)
+#define SEQ_BUFSIZE (1 << 12)
/** MGS threads must be >= 3, see bug 22458 comment #28 */
-#define MGS_THREADS_AUTO_MIN 3
-#define MGS_THREADS_AUTO_MAX 32
-#define MGS_NBUFS (64 * cfs_num_online_cpus())
+#define MGS_NTHRS_INIT (PTLRPC_NTHRS_INIT + 1)
+#define MGS_NTHRS_MAX 32
+
+#define MGS_NBUFS 64
#define MGS_BUFSIZE (8 * 1024)
#define MGS_MAXREQSIZE (7 * 1024)
#define MGS_MAXREPSIZE (9 * 1024)
-/** Absolute OSS limits */
-#define OSS_THREADS_MIN 3 /* difficult replies, HPQ, others */
-#define OSS_THREADS_MAX 512
-#define OST_NBUFS (64 * cfs_num_online_cpus())
-#define OST_BUFSIZE (8 * 1024)
+ /*
+ * OSS threads constants:
+ *
+ * Given 8 as factor and 64 as base threads number
+ *
+ * example 1):
+ * On 8-core server configured to 2 partitions, we will have
+ * 64 + 8 * 4 = 96 threads for each partition, 192 total threads.
+ *
+ * example 2):
+ * On 32-core machine configured to 4 partitions, we will have
+ * 64 + 8 * 8 = 112 threads for each partition, so total threads number
+ * will be 112 * 4 = 448.
+ *
+ * example 3):
+ * On 64-core machine configured to 4 partitions, we will have
+ * 64 + 16 * 8 = 192 threads for each partition, so total threads number
+ * will be 192 * 4 = 768 which is above limit OSS_NTHRS_MAX(512), so we
+ * cut off the value to OSS_NTHRS_MAX(512) / 4 which is 128 threads
+ * for each partition.
+ *
+ * So we can see that with these constants, threads number wil be at the
+ * similar level of old versions, unless the server has many cores.
+ */
+ /* depress threads factor for VM with small memory size */
+#define OSS_THR_FACTOR min_t(int, 8, \
+ NUM_CACHEPAGES >> (28 - PAGE_CACHE_SHIFT))
+#define OSS_NTHRS_INIT (PTLRPC_NTHRS_INIT + 1)
+#define OSS_NTHRS_BASE 64
+#define OSS_NTHRS_MAX 512
+
+/* threads for handling "create" request */
+#define OSS_CR_THR_FACTOR 1
+#define OSS_CR_NTHRS_INIT PTLRPC_NTHRS_INIT
+#define OSS_CR_NTHRS_BASE 8
+#define OSS_CR_NTHRS_MAX 64
/**
- * OST_MAXREQSIZE ~= 4768 bytes =
- * lustre_msg + obdo + 16 * obd_ioobj + 256 * niobuf_remote
+ * OST_IO_MAXREQSIZE ~=
+ * lustre_msg + ptlrpc_body + obdo + obd_ioobj +
+ * DT_MAX_BRW_PAGES * niobuf_remote
*
* - single object with 16 pages is 512 bytes
- * - OST_MAXREQSIZE must be at least 1 page of cookies plus some spillover
+ * - OST_IO_MAXREQSIZE must be at least 1 page of cookies plus some spillover
+ * - Must be a multiple of 1024
+ * - actual size is about 18K
+ */
+#define _OST_MAXREQSIZE_SUM (sizeof(struct lustre_msg) + \
+ sizeof(struct ptlrpc_body) + \
+ sizeof(struct obdo) + \
+ sizeof(struct obd_ioobj) + \
+ sizeof(struct niobuf_remote) * DT_MAX_BRW_PAGES)
+/**
+ * FIEMAP request can be 4K+ for now
+ */
+#define OST_MAXREQSIZE (16 * 1024)
+#define OST_IO_MAXREQSIZE max_t(int, OST_MAXREQSIZE, \
+ (((_OST_MAXREQSIZE_SUM - 1) | (1024 - 1)) + 1))
+
+#define OST_MAXREPSIZE (9 * 1024)
+#define OST_IO_MAXREPSIZE OST_MAXREPSIZE
+
+#define OST_NBUFS 64
+/** OST_BUFSIZE = max_reqsize + max sptlrpc payload size */
+#define OST_BUFSIZE max_t(int, OST_MAXREQSIZE + 1024, 16 * 1024)
+/**
+ * OST_IO_MAXREQSIZE is 18K, giving extra 46K can increase buffer utilization
+ * rate of request buffer, please check comment of MDS_LOV_BUFSIZE for details.
*/
-#define OST_MAXREQSIZE (5 * 1024)
-#define OST_MAXREPSIZE (9 * 1024)
+#define OST_IO_BUFSIZE max_t(int, OST_IO_MAXREQSIZE + 1024, 64 * 1024)
/* Macro to hide a typecast. */
#define ptlrpc_req_async_args(req) ((void *)&req->rq_async_args)
+struct ptlrpc_replay_async_args {
+ int praa_old_state;
+ int praa_old_status;
+};
+
/**
* Structure to single define portal connection.
*/
struct ptlrpc_connection {
- /** linkage for connections hash table */
- cfs_hlist_node_t c_hash;
- /** Our own lnet nid for this connection */
- lnet_nid_t c_self;
- /** Remote side nid for this connection */
- lnet_process_id_t c_peer;
- /** UUID of the other side */
- struct obd_uuid c_remote_uuid;
- /** reference counter for this connection */
- cfs_atomic_t c_refcount;
+ /** linkage for connections hash table */
+ struct hlist_node c_hash;
+ /** Our own lnet nid for this connection */
+ lnet_nid_t c_self;
+ /** Remote side nid for this connection */
+ lnet_process_id_t c_peer;
+ /** UUID of the other side */
+ struct obd_uuid c_remote_uuid;
+ /** reference counter for this connection */
+ atomic_t c_refcount;
};
/** Client definition for PortalRPC */
* least big enough for that.
*/
void *pointer_arg[11];
- __u64 space[6];
+ __u64 space[7];
};
struct ptlrpc_request_set;
typedef int (*set_interpreter_func)(struct ptlrpc_request_set *, void *, int);
+typedef int (*set_producer_func)(struct ptlrpc_request_set *, void *);
/**
* Definition of request set structure.
* returned.
*/
struct ptlrpc_request_set {
- cfs_atomic_t set_refcount;
- /** number of in queue requests */
- cfs_atomic_t set_new_count;
- /** number of uncompleted requests */
- cfs_atomic_t set_remaining;
- /** wait queue to wait on for request events */
- cfs_waitq_t set_waitq;
- cfs_waitq_t *set_wakeup_ptr;
- /** List of requests in the set */
- cfs_list_t set_requests;
- /**
- * List of completion callbacks to be called when the set is completed
- * This is only used if \a set_interpret is NULL.
- * Links struct ptlrpc_set_cbdata.
- */
- cfs_list_t set_cblist;
- /** Completion callback, if only one. */
- set_interpreter_func set_interpret;
- /** opaq argument passed to completion \a set_interpret callback. */
- void *set_arg;
- /**
- * Lock for \a set_new_requests manipulations
- * locked so that any old caller can communicate requests to
- * the set holder who can then fold them into the lock-free set
- */
- cfs_spinlock_t set_new_req_lock;
- /** List of new yet unsent requests. Only used with ptlrpcd now. */
- cfs_list_t set_new_requests;
+ atomic_t set_refcount;
+ /** number of in queue requests */
+ atomic_t set_new_count;
+ /** number of uncompleted requests */
+ atomic_t set_remaining;
+ /** wait queue to wait on for request events */
+ wait_queue_head_t set_waitq;
+ wait_queue_head_t *set_wakeup_ptr;
+ /** List of requests in the set */
+ struct list_head set_requests;
+ /**
+ * List of completion callbacks to be called when the set is completed
+ * This is only used if \a set_interpret is NULL.
+ * Links struct ptlrpc_set_cbdata.
+ */
+ struct list_head set_cblist;
+ /** Completion callback, if only one. */
+ set_interpreter_func set_interpret;
+ /** opaq argument passed to completion \a set_interpret callback. */
+ void *set_arg;
+ /**
+ * Lock for \a set_new_requests manipulations
+ * locked so that any old caller can communicate requests to
+ * the set holder who can then fold them into the lock-free set
+ */
+ spinlock_t set_new_req_lock;
+ /** List of new yet unsent requests. Only used with ptlrpcd now. */
+ struct list_head set_new_requests;
+
+ /** rq_status of requests that have been freed already */
+ int set_rc;
+ /** Additional fields used by the flow control extension */
+ /** Maximum number of RPCs in flight */
+ int set_max_inflight;
+ /** Callback function used to generate RPCs */
+ set_producer_func set_producer;
+ /** opaq argument passed to the producer callback */
+ void *set_producer_arg;
};
/**
* Description of a single ptrlrpc_set callback
*/
struct ptlrpc_set_cbdata {
- /** List linkage item */
- cfs_list_t psc_item;
- /** Pointer to interpreting function */
- set_interpreter_func psc_interpret;
- /** Opaq argument to pass to the callback */
- void *psc_data;
+ /** List linkage item */
+ struct list_head psc_item;
+ /** Pointer to interpreting function */
+ set_interpreter_func psc_interpret;
+ /** Opaq argument to pass to the callback */
+ void *psc_data;
};
struct ptlrpc_bulk_desc;
+struct ptlrpc_service_part;
+struct ptlrpc_service;
/**
* ptlrpc callback & work item stuff
* added to the state for replay/failover consistency guarantees.
*/
struct ptlrpc_reply_state {
- /** Callback description */
- struct ptlrpc_cb_id rs_cb_id;
- /** Linkage for list of all reply states in a system */
- cfs_list_t rs_list;
- /** Linkage for list of all reply states on same export */
- cfs_list_t rs_exp_list;
- /** Linkage for list of all reply states for same obd */
- cfs_list_t rs_obd_list;
+ /** Callback description */
+ struct ptlrpc_cb_id rs_cb_id;
+ /** Linkage for list of all reply states in a system */
+ struct list_head rs_list;
+ /** Linkage for list of all reply states on same export */
+ struct list_head rs_exp_list;
+ /** Linkage for list of all reply states for same obd */
+ struct list_head rs_obd_list;
#if RS_DEBUG
- cfs_list_t rs_debug_list;
+ struct list_head rs_debug_list;
#endif
- /** A spinlock to protect the reply state flags */
- cfs_spinlock_t rs_lock;
- /** Reply state flags */
+ /** A spinlock to protect the reply state flags */
+ spinlock_t rs_lock;
+ /** Reply state flags */
unsigned long rs_difficult:1; /* ACK/commit stuff */
unsigned long rs_no_ack:1; /* no ACK, even for
difficult requests */
__u64 rs_transno;
/** xid */
__u64 rs_xid;
- struct obd_export *rs_export;
- struct ptlrpc_service *rs_service;
- /** Lnet metadata handle for the reply */
- lnet_handle_md_t rs_md_h;
- cfs_atomic_t rs_refcount;
-
- /** Context for the sevice thread */
- struct ptlrpc_svc_ctx *rs_svc_ctx;
- /** Reply buffer (actually sent to the client), encoded if needed */
- struct lustre_msg *rs_repbuf; /* wrapper */
+ struct obd_export *rs_export;
+ struct ptlrpc_service_part *rs_svcpt;
+ /** Lnet metadata handle for the reply */
+ lnet_handle_md_t rs_md_h;
+ atomic_t rs_refcount;
+
+ /** Context for the sevice thread */
+ struct ptlrpc_svc_ctx *rs_svc_ctx;
+ /** Reply buffer (actually sent to the client), encoded if needed */
+ struct lustre_msg *rs_repbuf; /* wrapper */
/** Size of the reply buffer */
int rs_repbuf_len; /* wrapper buf length */
/** Size of the reply message */
typedef int (*ptlrpc_interpterer_t)(const struct lu_env *env,
struct ptlrpc_request *req,
void *arg, int rc);
+/** Type of request resend call-back */
+typedef void (*ptlrpc_resend_cb_t)(struct ptlrpc_request *req,
+ void *arg);
/**
* Definition of request pool structure.
* any allocations (to avoid e.g. OOM).
*/
struct ptlrpc_request_pool {
- /** Locks the list */
- cfs_spinlock_t prp_lock;
- /** list of ptlrpc_request structs */
- cfs_list_t prp_req_list;
- /** Maximum message size that would fit into a rquest from this pool */
- int prp_rq_size;
- /** Function to allocate more requests for this pool */
- void (*prp_populate)(struct ptlrpc_request_pool *, int);
+ /** Locks the list */
+ spinlock_t prp_lock;
+ /** list of ptlrpc_request structs */
+ struct list_head prp_req_list;
+ /** Maximum message size that would fit into a rquest from this pool */
+ int prp_rq_size;
+ /** Function to allocate more requests for this pool */
+ void (*prp_populate)(struct ptlrpc_request_pool *, int);
};
struct lu_context;
struct ldlm_lock;
/**
+ * \defgroup nrs Network Request Scheduler
+ * @{
+ */
+struct ptlrpc_nrs_policy;
+struct ptlrpc_nrs_resource;
+struct ptlrpc_nrs_request;
+
+/**
+ * NRS control operations.
+ *
+ * These are common for all policies.
+ */
+enum ptlrpc_nrs_ctl {
+ /**
+ * Not a valid opcode.
+ */
+ PTLRPC_NRS_CTL_INVALID,
+ /**
+ * Activate the policy.
+ */
+ PTLRPC_NRS_CTL_START,
+ /**
+ * Reserved for multiple primary policies, which may be a possibility
+ * in the future.
+ */
+ PTLRPC_NRS_CTL_STOP,
+ /**
+ * Policies can start using opcodes from this value and onwards for
+ * their own purposes; the assigned value itself is arbitrary.
+ */
+ PTLRPC_NRS_CTL_1ST_POL_SPEC = 0x20,
+};
+
+/**
+ * ORR policy operations
+ */
+enum nrs_ctl_orr {
+ NRS_CTL_ORR_RD_QUANTUM = PTLRPC_NRS_CTL_1ST_POL_SPEC,
+ NRS_CTL_ORR_WR_QUANTUM,
+ NRS_CTL_ORR_RD_OFF_TYPE,
+ NRS_CTL_ORR_WR_OFF_TYPE,
+ NRS_CTL_ORR_RD_SUPP_REQ,
+ NRS_CTL_ORR_WR_SUPP_REQ,
+};
+
+/**
+ * NRS policy operations.
+ *
+ * These determine the behaviour of a policy, and are called in response to
+ * NRS core events.
+ */
+struct ptlrpc_nrs_pol_ops {
+ /**
+ * Called during policy registration; this operation is optional.
+ *
+ * \param[in,out] policy The policy being initialized
+ */
+ int (*op_policy_init) (struct ptlrpc_nrs_policy *policy);
+ /**
+ * Called during policy unregistration; this operation is optional.
+ *
+ * \param[in,out] policy The policy being unregistered/finalized
+ */
+ void (*op_policy_fini) (struct ptlrpc_nrs_policy *policy);
+ /**
+ * Called when activating a policy via lprocfs; policies allocate and
+ * initialize their resources here; this operation is optional.
+ *
+ * \param[in,out] policy The policy being started
+ * \param[in,out] arg A generic char buffer
+ *
+ * \see nrs_policy_start_locked()
+ */
+ int (*op_policy_start) (struct ptlrpc_nrs_policy *policy,
+ char *arg);
+ /**
+ * Called when deactivating a policy via lprocfs; policies deallocate
+ * their resources here; this operation is optional
+ *
+ * \param[in,out] policy The policy being stopped
+ *
+ * \see nrs_policy_stop0()
+ */
+ void (*op_policy_stop) (struct ptlrpc_nrs_policy *policy);
+ /**
+ * Used for policy-specific operations; i.e. not generic ones like
+ * \e PTLRPC_NRS_CTL_START and \e PTLRPC_NRS_CTL_GET_INFO; analogous
+ * to an ioctl; this operation is optional.
+ *
+ * \param[in,out] policy The policy carrying out operation \a opc
+ * \param[in] opc The command operation being carried out
+ * \param[in,out] arg An generic buffer for communication between the
+ * user and the control operation
+ *
+ * \retval -ve error
+ * \retval 0 success
+ *
+ * \see ptlrpc_nrs_policy_control()
+ */
+ int (*op_policy_ctl) (struct ptlrpc_nrs_policy *policy,
+ enum ptlrpc_nrs_ctl opc, void *arg);
+
+ /**
+ * Called when obtaining references to the resources of the resource
+ * hierarchy for a request that has arrived for handling at the PTLRPC
+ * service. Policies should return -ve for requests they do not wish
+ * to handle. This operation is mandatory.
+ *
+ * \param[in,out] policy The policy we're getting resources for.
+ * \param[in,out] nrq The request we are getting resources for.
+ * \param[in] parent The parent resource of the resource being
+ * requested; set to NULL if none.
+ * \param[out] resp The resource is to be returned here; the
+ * fallback policy in an NRS head should
+ * \e always return a non-NULL pointer value.
+ * \param[in] moving_req When set, signifies that this is an attempt
+ * to obtain resources for a request being moved
+ * to the high-priority NRS head by
+ * ldlm_lock_reorder_req().
+ * This implies two things:
+ * 1. We are under obd_export::exp_rpc_lock and
+ * so should not sleep.
+ * 2. We should not perform non-idempotent or can
+ * skip performing idempotent operations that
+ * were carried out when resources were first
+ * taken for the request when it was initialized
+ * in ptlrpc_nrs_req_initialize().
+ *
+ * \retval 0, +ve The level of the returned resource in the resource
+ * hierarchy; currently only 0 (for a non-leaf resource)
+ * and 1 (for a leaf resource) are supported by the
+ * framework.
+ * \retval -ve error
+ *
+ * \see ptlrpc_nrs_req_initialize()
+ * \see ptlrpc_nrs_hpreq_add_nolock()
+ * \see ptlrpc_nrs_req_hp_move()
+ */
+ int (*op_res_get) (struct ptlrpc_nrs_policy *policy,
+ struct ptlrpc_nrs_request *nrq,
+ const struct ptlrpc_nrs_resource *parent,
+ struct ptlrpc_nrs_resource **resp,
+ bool moving_req);
+ /**
+ * Called when releasing references taken for resources in the resource
+ * hierarchy for the request; this operation is optional.
+ *
+ * \param[in,out] policy The policy the resource belongs to
+ * \param[in] res The resource to be freed
+ *
+ * \see ptlrpc_nrs_req_finalize()
+ * \see ptlrpc_nrs_hpreq_add_nolock()
+ * \see ptlrpc_nrs_req_hp_move()
+ */
+ void (*op_res_put) (struct ptlrpc_nrs_policy *policy,
+ const struct ptlrpc_nrs_resource *res);
+
+ /**
+ * Obtains a request for handling from the policy, and optionally
+ * removes the request from the policy; this operation is mandatory.
+ *
+ * \param[in,out] policy The policy to poll
+ * \param[in] peek When set, signifies that we just want to
+ * examine the request, and not handle it, so the
+ * request is not removed from the policy.
+ * \param[in] force When set, it will force a policy to return a
+ * request if it has one queued.
+ *
+ * \retval NULL No request available for handling
+ * \retval valid-pointer The request polled for handling
+ *
+ * \see ptlrpc_nrs_req_get_nolock()
+ */
+ struct ptlrpc_nrs_request *
+ (*op_req_get) (struct ptlrpc_nrs_policy *policy, bool peek,
+ bool force);
+ /**
+ * Called when attempting to add a request to a policy for later
+ * handling; this operation is mandatory.
+ *
+ * \param[in,out] policy The policy on which to enqueue \a nrq
+ * \param[in,out] nrq The request to enqueue
+ *
+ * \retval 0 success
+ * \retval != 0 error
+ *
+ * \see ptlrpc_nrs_req_add_nolock()
+ */
+ int (*op_req_enqueue) (struct ptlrpc_nrs_policy *policy,
+ struct ptlrpc_nrs_request *nrq);
+ /**
+ * Removes a request from the policy's set of pending requests. Normally
+ * called after a request has been polled successfully from the policy
+ * for handling; this operation is mandatory.
+ *
+ * \param[in,out] policy The policy the request \a nrq belongs to
+ * \param[in,out] nrq The request to dequeue
+ *
+ * \see ptlrpc_nrs_req_del_nolock()
+ */
+ void (*op_req_dequeue) (struct ptlrpc_nrs_policy *policy,
+ struct ptlrpc_nrs_request *nrq);
+ /**
+ * Called after the request being carried out. Could be used for
+ * job/resource control; this operation is optional.
+ *
+ * \param[in,out] policy The policy which is stopping to handle request
+ * \a nrq
+ * \param[in,out] nrq The request
+ *
+ * \pre assert_spin_locked(&svcpt->scp_req_lock)
+ *
+ * \see ptlrpc_nrs_req_stop_nolock()
+ */
+ void (*op_req_stop) (struct ptlrpc_nrs_policy *policy,
+ struct ptlrpc_nrs_request *nrq);
+ /**
+ * Registers the policy's lprocfs interface with a PTLRPC service.
+ *
+ * \param[in] svc The service
+ *
+ * \retval 0 success
+ * \retval != 0 error
+ */
+ int (*op_lprocfs_init) (struct ptlrpc_service *svc);
+ /**
+ * Unegisters the policy's lprocfs interface with a PTLRPC service.
+ *
+ * In cases of failed policy registration in
+ * \e ptlrpc_nrs_policy_register(), this function may be called for a
+ * service which has not registered the policy successfully, so
+ * implementations of this method should make sure their operations are
+ * safe in such cases.
+ *
+ * \param[in] svc The service
+ */
+ void (*op_lprocfs_fini) (struct ptlrpc_service *svc);
+};
+
+/**
+ * Policy flags
+ */
+enum nrs_policy_flags {
+ /**
+ * Fallback policy, use this flag only on a single supported policy per
+ * service. The flag cannot be used on policies that use
+ * \e PTLRPC_NRS_FL_REG_EXTERN
+ */
+ PTLRPC_NRS_FL_FALLBACK = (1 << 0),
+ /**
+ * Start policy immediately after registering.
+ */
+ PTLRPC_NRS_FL_REG_START = (1 << 1),
+ /**
+ * This is a policy registering from a module different to the one NRS
+ * core ships in (currently ptlrpc).
+ */
+ PTLRPC_NRS_FL_REG_EXTERN = (1 << 2),
+};
+
+/**
+ * NRS queue type.
+ *
+ * Denotes whether an NRS instance is for handling normal or high-priority
+ * RPCs, or whether an operation pertains to one or both of the NRS instances
+ * in a service.
+ */
+enum ptlrpc_nrs_queue_type {
+ PTLRPC_NRS_QUEUE_REG = (1 << 0),
+ PTLRPC_NRS_QUEUE_HP = (1 << 1),
+ PTLRPC_NRS_QUEUE_BOTH = (PTLRPC_NRS_QUEUE_REG | PTLRPC_NRS_QUEUE_HP)
+};
+
+/**
+ * NRS head
+ *
+ * A PTLRPC service has at least one NRS head instance for handling normal
+ * priority RPCs, and may optionally have a second NRS head instance for
+ * handling high-priority RPCs. Each NRS head maintains a list of available
+ * policies, of which one and only one policy is acting as the fallback policy,
+ * and optionally a different policy may be acting as the primary policy. For
+ * all RPCs handled by this NRS head instance, NRS core will first attempt to
+ * enqueue the RPC using the primary policy (if any). The fallback policy is
+ * used in the following cases:
+ * - when there was no primary policy in the
+ * ptlrpc_nrs_pol_state::NRS_POL_STATE_STARTED state at the time the request
+ * was initialized.
+ * - when the primary policy that was at the
+ * ptlrpc_nrs_pol_state::PTLRPC_NRS_POL_STATE_STARTED state at the time the
+ * RPC was initialized, denoted it did not wish, or for some other reason was
+ * not able to handle the request, by returning a non-valid NRS resource
+ * reference.
+ * - when the primary policy that was at the
+ * ptlrpc_nrs_pol_state::PTLRPC_NRS_POL_STATE_STARTED state at the time the
+ * RPC was initialized, fails later during the request enqueueing stage.
+ *
+ * \see nrs_resource_get_safe()
+ * \see nrs_request_enqueue()
+ */
+struct ptlrpc_nrs {
+ spinlock_t nrs_lock;
+ /** XXX Possibly replace svcpt->scp_req_lock with another lock here. */
+ /**
+ * List of registered policies
+ */
+ struct list_head nrs_policy_list;
+ /**
+ * List of policies with queued requests. Policies that have any
+ * outstanding requests are queued here, and this list is queried
+ * in a round-robin manner from NRS core when obtaining a request
+ * for handling. This ensures that requests from policies that at some
+ * point transition away from the
+ * ptlrpc_nrs_pol_state::NRS_POL_STATE_STARTED state are drained.
+ */
+ struct list_head nrs_policy_queued;
+ /**
+ * Service partition for this NRS head
+ */
+ struct ptlrpc_service_part *nrs_svcpt;
+ /**
+ * Primary policy, which is the preferred policy for handling RPCs
+ */
+ struct ptlrpc_nrs_policy *nrs_policy_primary;
+ /**
+ * Fallback policy, which is the backup policy for handling RPCs
+ */
+ struct ptlrpc_nrs_policy *nrs_policy_fallback;
+ /**
+ * This NRS head handles either HP or regular requests
+ */
+ enum ptlrpc_nrs_queue_type nrs_queue_type;
+ /**
+ * # queued requests from all policies in this NRS head
+ */
+ unsigned long nrs_req_queued;
+ /**
+ * # scheduled requests from all policies in this NRS head
+ */
+ unsigned long nrs_req_started;
+ /**
+ * # policies on this NRS
+ */
+ unsigned nrs_num_pols;
+ /**
+ * This NRS head is in progress of starting a policy
+ */
+ unsigned nrs_policy_starting:1;
+ /**
+ * In progress of shutting down the whole NRS head; used during
+ * unregistration
+ */
+ unsigned nrs_stopping:1;
+ /**
+ * NRS policy is throttling reqeust
+ */
+ unsigned nrs_throttling:1;
+};
+
+#define NRS_POL_NAME_MAX 16
+
+struct ptlrpc_nrs_pol_desc;
+
+/**
+ * Service compatibility predicate; this determines whether a policy is adequate
+ * for handling RPCs of a particular PTLRPC service.
+ *
+ * XXX:This should give the same result during policy registration and
+ * unregistration, and for all partitions of a service; so the result should not
+ * depend on temporal service or other properties, that may influence the
+ * result.
+ */
+typedef bool (*nrs_pol_desc_compat_t) (const struct ptlrpc_service *svc,
+ const struct ptlrpc_nrs_pol_desc *desc);
+
+struct ptlrpc_nrs_pol_conf {
+ /**
+ * Human-readable policy name
+ */
+ char nc_name[NRS_POL_NAME_MAX];
+ /**
+ * NRS operations for this policy
+ */
+ const struct ptlrpc_nrs_pol_ops *nc_ops;
+ /**
+ * Service compatibility predicate
+ */
+ nrs_pol_desc_compat_t nc_compat;
+ /**
+ * Set for policies that support a single ptlrpc service, i.e. ones that
+ * have \a pd_compat set to nrs_policy_compat_one(). The variable value
+ * depicts the name of the single service that such policies are
+ * compatible with.
+ */
+ const char *nc_compat_svc_name;
+ /**
+ * Owner module for this policy descriptor; policies registering from a
+ * different module to the one the NRS framework is held within
+ * (currently ptlrpc), should set this field to THIS_MODULE.
+ */
+ struct module *nc_owner;
+ /**
+ * Policy registration flags; a bitmast of \e nrs_policy_flags
+ */
+ unsigned nc_flags;
+};
+
+/**
+ * NRS policy registering descriptor
+ *
+ * Is used to hold a description of a policy that can be passed to NRS core in
+ * order to register the policy with NRS heads in different PTLRPC services.
+ */
+struct ptlrpc_nrs_pol_desc {
+ /**
+ * Human-readable policy name
+ */
+ char pd_name[NRS_POL_NAME_MAX];
+ /**
+ * Link into nrs_core::nrs_policies
+ */
+ struct list_head pd_list;
+ /**
+ * NRS operations for this policy
+ */
+ const struct ptlrpc_nrs_pol_ops *pd_ops;
+ /**
+ * Service compatibility predicate
+ */
+ nrs_pol_desc_compat_t pd_compat;
+ /**
+ * Set for policies that are compatible with only one PTLRPC service.
+ *
+ * \see ptlrpc_nrs_pol_conf::nc_compat_svc_name
+ */
+ const char *pd_compat_svc_name;
+ /**
+ * Owner module for this policy descriptor.
+ *
+ * We need to hold a reference to the module whenever we might make use
+ * of any of the module's contents, i.e.
+ * - If one or more instances of the policy are at a state where they
+ * might be handling a request, i.e.
+ * ptlrpc_nrs_pol_state::NRS_POL_STATE_STARTED or
+ * ptlrpc_nrs_pol_state::NRS_POL_STATE_STOPPING as we will have to
+ * call into the policy's ptlrpc_nrs_pol_ops() handlers. A reference
+ * is taken on the module when
+ * \e ptlrpc_nrs_pol_desc::pd_refs becomes 1, and released when it
+ * becomes 0, so that we hold only one reference to the module maximum
+ * at any time.
+ *
+ * We do not need to hold a reference to the module, even though we
+ * might use code and data from the module, in the following cases:
+ * - During external policy registration, because this should happen in
+ * the module's init() function, in which case the module is safe from
+ * removal because a reference is being held on the module by the
+ * kernel, and iirc kmod (and I guess module-init-tools also) will
+ * serialize any racing processes properly anyway.
+ * - During external policy unregistration, because this should happen
+ * in a module's exit() function, and any attempts to start a policy
+ * instance would need to take a reference on the module, and this is
+ * not possible once we have reached the point where the exit()
+ * handler is called.
+ * - During service registration and unregistration, as service setup
+ * and cleanup, and policy registration, unregistration and policy
+ * instance starting, are serialized by \e nrs_core::nrs_mutex, so
+ * as long as users adhere to the convention of registering policies
+ * in init() and unregistering them in module exit() functions, there
+ * should not be a race between these operations.
+ * - During any policy-specific lprocfs operations, because a reference
+ * is held by the kernel on a proc entry that has been entered by a
+ * syscall, so as long as proc entries are removed during unregistration time,
+ * then unregistration and lprocfs operations will be properly
+ * serialized.
+ */
+ struct module *pd_owner;
+ /**
+ * Bitmask of \e nrs_policy_flags
+ */
+ unsigned pd_flags;
+ /**
+ * # of references on this descriptor
+ */
+ atomic_t pd_refs;
+};
+
+/**
+ * NRS policy state
+ *
+ * Policies transition from one state to the other during their lifetime
+ */
+enum ptlrpc_nrs_pol_state {
+ /**
+ * Not a valid policy state.
+ */
+ NRS_POL_STATE_INVALID,
+ /**
+ * Policies are at this state either at the start of their life, or
+ * transition here when the user selects a different policy to act
+ * as the primary one.
+ */
+ NRS_POL_STATE_STOPPED,
+ /**
+ * Policy is progress of stopping
+ */
+ NRS_POL_STATE_STOPPING,
+ /**
+ * Policy is in progress of starting
+ */
+ NRS_POL_STATE_STARTING,
+ /**
+ * A policy is in this state in two cases:
+ * - it is the fallback policy, which is always in this state.
+ * - it has been activated by the user; i.e. it is the primary policy,
+ */
+ NRS_POL_STATE_STARTED,
+};
+
+/**
+ * NRS policy information
+ *
+ * Used for obtaining information for the status of a policy via lprocfs
+ */
+struct ptlrpc_nrs_pol_info {
+ /**
+ * Policy name
+ */
+ char pi_name[NRS_POL_NAME_MAX];
+ /**
+ * Current policy state
+ */
+ enum ptlrpc_nrs_pol_state pi_state;
+ /**
+ * # RPCs enqueued for later dispatching by the policy
+ */
+ long pi_req_queued;
+ /**
+ * # RPCs started for dispatch by the policy
+ */
+ long pi_req_started;
+ /**
+ * Is this a fallback policy?
+ */
+ unsigned pi_fallback:1;
+};
+
+/**
+ * NRS policy
+ *
+ * There is one instance of this for each policy in each NRS head of each
+ * PTLRPC service partition.
+ */
+struct ptlrpc_nrs_policy {
+ /**
+ * Linkage into the NRS head's list of policies,
+ * ptlrpc_nrs:nrs_policy_list
+ */
+ struct list_head pol_list;
+ /**
+ * Linkage into the NRS head's list of policies with enqueued
+ * requests ptlrpc_nrs:nrs_policy_queued
+ */
+ struct list_head pol_list_queued;
+ /**
+ * Current state of this policy
+ */
+ enum ptlrpc_nrs_pol_state pol_state;
+ /**
+ * Bitmask of nrs_policy_flags
+ */
+ unsigned pol_flags;
+ /**
+ * # RPCs enqueued for later dispatching by the policy
+ */
+ long pol_req_queued;
+ /**
+ * # RPCs started for dispatch by the policy
+ */
+ long pol_req_started;
+ /**
+ * Usage Reference count taken on the policy instance
+ */
+ long pol_ref;
+ /**
+ * The NRS head this policy has been created at
+ */
+ struct ptlrpc_nrs *pol_nrs;
+ /**
+ * Private policy data; varies by policy type
+ */
+ void *pol_private;
+ /**
+ * Policy descriptor for this policy instance.
+ */
+ struct ptlrpc_nrs_pol_desc *pol_desc;
+};
+
+/**
+ * NRS resource
+ *
+ * Resources are embedded into two types of NRS entities:
+ * - Inside NRS policies, in the policy's private data in
+ * ptlrpc_nrs_policy::pol_private
+ * - In objects that act as prime-level scheduling entities in different NRS
+ * policies; e.g. on a policy that performs round robin or similar order
+ * scheduling across client NIDs, there would be one NRS resource per unique
+ * client NID. On a policy which performs round robin scheduling across
+ * backend filesystem objects, there would be one resource associated with
+ * each of the backend filesystem objects partaking in the scheduling
+ * performed by the policy.
+ *
+ * NRS resources share a parent-child relationship, in which resources embedded
+ * in policy instances are the parent entities, with all scheduling entities
+ * a policy schedules across being the children, thus forming a simple resource
+ * hierarchy. This hierarchy may be extended with one or more levels in the
+ * future if the ability to have more than one primary policy is added.
+ *
+ * Upon request initialization, references to the then active NRS policies are
+ * taken and used to later handle the dispatching of the request with one of
+ * these policies.
+ *
+ * \see nrs_resource_get_safe()
+ * \see ptlrpc_nrs_req_add()
+ */
+struct ptlrpc_nrs_resource {
+ /**
+ * This NRS resource's parent; is NULL for resources embedded in NRS
+ * policy instances; i.e. those are top-level ones.
+ */
+ struct ptlrpc_nrs_resource *res_parent;
+ /**
+ * The policy associated with this resource.
+ */
+ struct ptlrpc_nrs_policy *res_policy;
+};
+
+enum {
+ NRS_RES_FALLBACK,
+ NRS_RES_PRIMARY,
+ NRS_RES_MAX
+};
+
+/* \name fifo
+ *
+ * FIFO policy
+ *
+ * This policy is a logical wrapper around previous, non-NRS functionality.
+ * It dispatches RPCs in the same order as they arrive from the network. This
+ * policy is currently used as the fallback policy, and the only enabled policy
+ * on all NRS heads of all PTLRPC service partitions.
+ * @{
+ */
+
+/**
+ * Private data structure for the FIFO policy
+ */
+struct nrs_fifo_head {
+ /**
+ * Resource object for policy instance.
+ */
+ struct ptlrpc_nrs_resource fh_res;
+ /**
+ * List of queued requests.
+ */
+ struct list_head fh_list;
+ /**
+ * For debugging purposes.
+ */
+ __u64 fh_sequence;
+};
+
+struct nrs_fifo_req {
+ struct list_head fr_list;
+ __u64 fr_sequence;
+};
+
+/** @} fifo */
+
+/**
+ * \name CRR-N
+ *
+ * CRR-N, Client Round Robin over NIDs
+ * @{
+ */
+
+/**
+ * private data structure for CRR-N NRS
+ */
+struct nrs_crrn_net {
+ struct ptlrpc_nrs_resource cn_res;
+ cfs_binheap_t *cn_binheap;
+ cfs_hash_t *cn_cli_hash;
+ /**
+ * Used when a new scheduling round commences, in order to synchronize
+ * all clients with the new round number.
+ */
+ __u64 cn_round;
+ /**
+ * Determines the relevant ordering amongst request batches within a
+ * scheduling round.
+ */
+ __u64 cn_sequence;
+ /**
+ * Round Robin quantum; the maximum number of RPCs that each request
+ * batch for each client can have in a scheduling round.
+ */
+ __u16 cn_quantum;
+};
+
+/**
+ * Object representing a client in CRR-N, as identified by its NID
+ */
+struct nrs_crrn_client {
+ struct ptlrpc_nrs_resource cc_res;
+ struct hlist_node cc_hnode;
+ lnet_nid_t cc_nid;
+ /**
+ * The round number against which this client is currently scheduling
+ * requests.
+ */
+ __u64 cc_round;
+ /**
+ * The sequence number used for requests scheduled by this client during
+ * the current round number.
+ */
+ __u64 cc_sequence;
+ atomic_t cc_ref;
+ /**
+ * Round Robin quantum; the maximum number of RPCs the client is allowed
+ * to schedule in a single batch of each round.
+ */
+ __u16 cc_quantum;
+ /**
+ * # of pending requests for this client, on all existing rounds
+ */
+ __u16 cc_active;
+};
+
+/**
+ * CRR-N NRS request definition
+ */
+struct nrs_crrn_req {
+ /**
+ * Round number for this request; shared with all other requests in the
+ * same batch.
+ */
+ __u64 cr_round;
+ /**
+ * Sequence number for this request; shared with all other requests in
+ * the same batch.
+ */
+ __u64 cr_sequence;
+};
+
+/**
+ * CRR-N policy operations.
+ */
+enum nrs_ctl_crr {
+ /**
+ * Read the RR quantum size of a CRR-N policy.
+ */
+ NRS_CTL_CRRN_RD_QUANTUM = PTLRPC_NRS_CTL_1ST_POL_SPEC,
+ /**
+ * Write the RR quantum size of a CRR-N policy.
+ */
+ NRS_CTL_CRRN_WR_QUANTUM,
+};
+
+/** @} CRR-N */
+
+/**
+ * \name ORR/TRR
+ *
+ * ORR/TRR (Object-based Round Robin/Target-based Round Robin) NRS policies
+ * @{
+ */
+
+/**
+ * Lower and upper byte offsets of a brw RPC
+ */
+struct nrs_orr_req_range {
+ __u64 or_start;
+ __u64 or_end;
+};
+
+/**
+ * RPC types supported by the ORR/TRR policies
+ */
+enum nrs_orr_supp {
+ NOS_OST_READ = (1 << 0),
+ NOS_OST_WRITE = (1 << 1),
+ NOS_OST_RW = (NOS_OST_READ | NOS_OST_WRITE),
+ /**
+ * Default value for policies.
+ */
+ NOS_DFLT = NOS_OST_READ
+};
+
+/**
+ * As unique keys for grouping RPCs together, we use the object's OST FID for
+ * the ORR policy, and the OST index for the TRR policy.
+ *
+ * XXX: We waste some space for TRR policy instances by using a union, but it
+ * allows to consolidate some of the code between ORR and TRR, and these
+ * policies will probably eventually merge into one anyway.
+ */
+struct nrs_orr_key {
+ union {
+ /** object FID for ORR */
+ struct lu_fid ok_fid;
+ /** OST index for TRR */
+ __u32 ok_idx;
+ };
+};
+
+/**
+ * The largest base string for unique hash/slab object names is
+ * "nrs_orr_reg_", so 13 characters. We add 3 to this to be used for the CPT
+ * id number, so this _should_ be more than enough for the maximum number of
+ * CPTs on any system. If it does happen that this statement is incorrect,
+ * nrs_orr_genobjname() will inevitably yield a non-unique name and cause
+ * kmem_cache_create() to complain (on Linux), so the erroneous situation
+ * will hopefully not go unnoticed.
+ */
+#define NRS_ORR_OBJ_NAME_MAX (sizeof("nrs_orr_reg_") + 3)
+
+/**
+ * private data structure for ORR and TRR NRS
+ */
+struct nrs_orr_data {
+ struct ptlrpc_nrs_resource od_res;
+ cfs_binheap_t *od_binheap;
+ cfs_hash_t *od_obj_hash;
+ struct kmem_cache *od_cache;
+ /**
+ * Used when a new scheduling round commences, in order to synchronize
+ * all object or OST batches with the new round number.
+ */
+ __u64 od_round;
+ /**
+ * Determines the relevant ordering amongst request batches within a
+ * scheduling round.
+ */
+ __u64 od_sequence;
+ /**
+ * RPC types that are currently supported.
+ */
+ enum nrs_orr_supp od_supp;
+ /**
+ * Round Robin quantum; the maxium number of RPCs that each request
+ * batch for each object or OST can have in a scheduling round.
+ */
+ __u16 od_quantum;
+ /**
+ * Whether to use physical disk offsets or logical file offsets.
+ */
+ bool od_physical;
+ /**
+ * XXX: We need to provide a persistently allocated string to hold
+ * unique object names for this policy, since in currently supported
+ * versions of Linux by Lustre, kmem_cache_create() just sets a pointer
+ * to the name string provided. kstrdup() is used in the version of
+ * kmeme_cache_create() in current Linux mainline, so we may be able to
+ * remove this in the future.
+ */
+ char od_objname[NRS_ORR_OBJ_NAME_MAX];
+};
+
+/**
+ * Represents a backend-fs object or OST in the ORR and TRR policies
+ * respectively
+ */
+struct nrs_orr_object {
+ struct ptlrpc_nrs_resource oo_res;
+ struct hlist_node oo_hnode;
+ /**
+ * The round number against which requests are being scheduled for this
+ * object or OST
+ */
+ __u64 oo_round;
+ /**
+ * The sequence number used for requests scheduled for this object or
+ * OST during the current round number.
+ */
+ __u64 oo_sequence;
+ /**
+ * The key of the object or OST for which this structure instance is
+ * scheduling RPCs
+ */
+ struct nrs_orr_key oo_key;
+ long oo_ref;
+ /**
+ * Round Robin quantum; the maximum number of RPCs that are allowed to
+ * be scheduled for the object or OST in a single batch of each round.
+ */
+ __u16 oo_quantum;
+ /**
+ * # of pending requests for this object or OST, on all existing rounds
+ */
+ __u16 oo_active;
+};
+
+/**
+ * ORR/TRR NRS request definition
+ */
+struct nrs_orr_req {
+ /**
+ * The offset range this request covers
+ */
+ struct nrs_orr_req_range or_range;
+ /**
+ * Round number for this request; shared with all other requests in the
+ * same batch.
+ */
+ __u64 or_round;
+ /**
+ * Sequence number for this request; shared with all other requests in
+ * the same batch.
+ */
+ __u64 or_sequence;
+ /**
+ * For debugging purposes.
+ */
+ struct nrs_orr_key or_key;
+ /**
+ * An ORR policy instance has filled in request information while
+ * enqueueing the request on the service partition's regular NRS head.
+ */
+ unsigned int or_orr_set:1;
+ /**
+ * A TRR policy instance has filled in request information while
+ * enqueueing the request on the service partition's regular NRS head.
+ */
+ unsigned int or_trr_set:1;
+ /**
+ * Request offset ranges have been filled in with logical offset
+ * values.
+ */
+ unsigned int or_logical_set:1;
+ /**
+ * Request offset ranges have been filled in with physical offset
+ * values.
+ */
+ unsigned int or_physical_set:1;
+};
+
+/** @} ORR/TRR */
+
+#include <lustre_nrs_tbf.h>
+
+/**
+ * NRS request
+ *
+ * Instances of this object exist embedded within ptlrpc_request; the main
+ * purpose of this object is to hold references to the request's resources
+ * for the lifetime of the request, and to hold properties that policies use
+ * use for determining the request's scheduling priority.
+ * */
+struct ptlrpc_nrs_request {
+ /**
+ * The request's resource hierarchy.
+ */
+ struct ptlrpc_nrs_resource *nr_res_ptrs[NRS_RES_MAX];
+ /**
+ * Index into ptlrpc_nrs_request::nr_res_ptrs of the resource of the
+ * policy that was used to enqueue the request.
+ *
+ * \see nrs_request_enqueue()
+ */
+ unsigned nr_res_idx;
+ unsigned nr_initialized:1;
+ unsigned nr_enqueued:1;
+ unsigned nr_started:1;
+ unsigned nr_finalized:1;
+ cfs_binheap_node_t nr_node;
+
+ /**
+ * Policy-specific fields, used for determining a request's scheduling
+ * priority, and other supporting functionality.
+ */
+ union {
+ /**
+ * Fields for the FIFO policy
+ */
+ struct nrs_fifo_req fifo;
+ /**
+ * CRR-N request defintion
+ */
+ struct nrs_crrn_req crr;
+ /** ORR and TRR share the same request definition */
+ struct nrs_orr_req orr;
+ /**
+ * TBF request definition
+ */
+ struct nrs_tbf_req tbf;
+ } nr_u;
+ /**
+ * Externally-registering policies may want to use this to allocate
+ * their own request properties.
+ */
+ void *ext;
+};
+
+/** @} nrs */
+
+/**
* Basic request prioritization operations structure.
* The whole idea is centered around locks and RPCs that might affect locks.
* When a lock is contended we try to give priority to RPCs that might lead
void (*hpreq_fini)(struct ptlrpc_request *);
};
+struct ptlrpc_cli_req {
+ /** For bulk requests on client only: bulk descriptor */
+ struct ptlrpc_bulk_desc *cr_bulk;
+ /** optional time limit for send attempts */
+ cfs_duration_t cr_delay_limit;
+ /** time request was first queued */
+ cfs_time_t cr_queued_time;
+ /** request sent timeval */
+ struct timeval cr_sent_tv;
+ /** time for request really sent out */
+ time_t cr_sent_out;
+ /** when req reply unlink must finish. */
+ time_t cr_reply_deadline;
+ /** when req bulk unlink must finish. */
+ time_t cr_bulk_deadline;
+ /** Portal to which this request would be sent */
+ short cr_req_ptl;
+ /** Portal where to wait for reply and where reply would be sent */
+ short cr_rep_ptl;
+ /** request resending number */
+ unsigned int cr_resend_nr;
+ /** What was import generation when this request was sent */
+ int cr_imp_gen;
+ enum lustre_imp_state cr_send_state;
+ /** Per-request waitq introduced by bug 21938 for recovery waiting */
+ wait_queue_head_t cr_set_waitq;
+ /** Link item for request set lists */
+ struct list_head cr_set_chain;
+ /** link to waited ctx */
+ struct list_head cr_ctx_chain;
+
+ /** client's half ctx */
+ struct ptlrpc_cli_ctx *cr_cli_ctx;
+ /** Link back to the request set */
+ struct ptlrpc_request_set *cr_set;
+ /** outgoing request MD handle */
+ lnet_handle_md_t cr_req_md_h;
+ /** request-out callback parameter */
+ struct ptlrpc_cb_id cr_req_cbid;
+ /** incoming reply MD handle */
+ lnet_handle_md_t cr_reply_md_h;
+ wait_queue_head_t cr_reply_waitq;
+ /** reply callback parameter */
+ struct ptlrpc_cb_id cr_reply_cbid;
+ /** Async completion handler, called when reply is received */
+ ptlrpc_interpterer_t cr_reply_interp;
+ /** Resend handler, called when request is resend to update RPC data */
+ ptlrpc_resend_cb_t cr_resend_cb;
+ /** Async completion context */
+ union ptlrpc_async_args cr_async_args;
+ /** Opaq data for replay and commit callbacks. */
+ void *cr_cb_data;
+ /**
+ * Commit callback, called when request is committed and about to be
+ * freed.
+ */
+ void (*cr_commit_cb)(struct ptlrpc_request *);
+ /** Replay callback, called after request is replayed at recovery */
+ void (*cr_replay_cb)(struct ptlrpc_request *);
+};
+
+/** client request member alias */
+/* NB: these alias should NOT be used by any new code, instead they should
+ * be removed step by step to avoid potential abuse */
+#define rq_bulk rq_cli.cr_bulk
+#define rq_delay_limit rq_cli.cr_delay_limit
+#define rq_queued_time rq_cli.cr_queued_time
+#define rq_sent_tv rq_cli.cr_sent_tv
+#define rq_real_sent rq_cli.cr_sent_out
+#define rq_reply_deadline rq_cli.cr_reply_deadline
+#define rq_bulk_deadline rq_cli.cr_bulk_deadline
+#define rq_nr_resend rq_cli.cr_resend_nr
+#define rq_request_portal rq_cli.cr_req_ptl
+#define rq_reply_portal rq_cli.cr_rep_ptl
+#define rq_import_generation rq_cli.cr_imp_gen
+#define rq_send_state rq_cli.cr_send_state
+#define rq_set_chain rq_cli.cr_set_chain
+#define rq_ctx_chain rq_cli.cr_ctx_chain
+#define rq_set rq_cli.cr_set
+#define rq_set_waitq rq_cli.cr_set_waitq
+#define rq_cli_ctx rq_cli.cr_cli_ctx
+#define rq_req_md_h rq_cli.cr_req_md_h
+#define rq_req_cbid rq_cli.cr_req_cbid
+#define rq_reply_md_h rq_cli.cr_reply_md_h
+#define rq_reply_waitq rq_cli.cr_reply_waitq
+#define rq_reply_cbid rq_cli.cr_reply_cbid
+#define rq_interpret_reply rq_cli.cr_reply_interp
+#define rq_resend_cb rq_cli.cr_resend_cb
+#define rq_async_args rq_cli.cr_async_args
+#define rq_cb_data rq_cli.cr_cb_data
+#define rq_commit_cb rq_cli.cr_commit_cb
+#define rq_replay_cb rq_cli.cr_replay_cb
+
+struct ptlrpc_srv_req {
+ /** initial thread servicing this request */
+ struct ptlrpc_thread *sr_svc_thread;
+ /**
+ * Server side list of incoming unserved requests sorted by arrival
+ * time. Traversed from time to time to notice about to expire
+ * requests and sent back "early replies" to clients to let them
+ * know server is alive and well, just very busy to service their
+ * requests in time
+ */
+ struct list_head sr_timed_list;
+ /** server-side per-export list */
+ struct list_head sr_exp_list;
+ /** server-side history, used for debuging purposes. */
+ struct list_head sr_hist_list;
+ /** history sequence # */
+ __u64 sr_hist_seq;
+ /** the index of service's srv_at_array into which request is linked */
+ time_t sr_at_index;
+ /** authed uid */
+ uid_t sr_auth_uid;
+ /** authed uid mapped to */
+ uid_t sr_auth_mapped_uid;
+ /** RPC is generated from what part of Lustre */
+ enum lustre_sec_part sr_sp_from;
+ /** request session context */
+ struct lu_context sr_ses;
+ /** \addtogroup nrs
+ * @{
+ */
+ /** stub for NRS request */
+ struct ptlrpc_nrs_request sr_nrq;
+ /** @} nrs */
+ /** request arrival time */
+ struct timeval sr_arrival_time;
+ /** server's half ctx */
+ struct ptlrpc_svc_ctx *sr_svc_ctx;
+ /** (server side), pointed directly into req buffer */
+ struct ptlrpc_user_desc *sr_user_desc;
+ /** separated reply state */
+ struct ptlrpc_reply_state *sr_reply_state;
+ /** server-side hp handlers */
+ struct ptlrpc_hpreq_ops *sr_ops;
+ /** incoming request buffer */
+ struct ptlrpc_request_buffer_desc *sr_rqbd;
+};
+
+/** server request member alias */
+/* NB: these alias should NOT be used by any new code, instead they should
+ * be removed step by step to avoid potential abuse */
+#define rq_svc_thread rq_srv.sr_svc_thread
+#define rq_timed_list rq_srv.sr_timed_list
+#define rq_exp_list rq_srv.sr_exp_list
+#define rq_history_list rq_srv.sr_hist_list
+#define rq_history_seq rq_srv.sr_hist_seq
+#define rq_at_index rq_srv.sr_at_index
+#define rq_auth_uid rq_srv.sr_auth_uid
+#define rq_auth_mapped_uid rq_srv.sr_auth_mapped_uid
+#define rq_sp_from rq_srv.sr_sp_from
+#define rq_session rq_srv.sr_ses
+#define rq_nrq rq_srv.sr_nrq
+#define rq_arrival_time rq_srv.sr_arrival_time
+#define rq_reply_state rq_srv.sr_reply_state
+#define rq_svc_ctx rq_srv.sr_svc_ctx
+#define rq_user_desc rq_srv.sr_user_desc
+#define rq_ops rq_srv.sr_ops
+#define rq_rqbd rq_srv.sr_rqbd
+
/**
* Represents remote procedure call.
*
* in Lustre.
*/
struct ptlrpc_request {
- /* Request type: one of PTL_RPC_MSG_* */
- int rq_type;
- /**
- * Linkage item through which this request is included into
- * sending/delayed lists on client and into rqbd list on server
- */
- cfs_list_t rq_list;
- /**
- * Server side list of incoming unserved requests sorted by arrival
- * time. Traversed from time to time to notice about to expire
- * requests and sent back "early replies" to clients to let them
- * know server is alive and well, just very busy to service their
- * requests in time
- */
- cfs_list_t rq_timed_list;
- /** server-side history, used for debuging purposes. */
- cfs_list_t rq_history_list;
- /** server-side per-export list */
- cfs_list_t rq_exp_list;
- /** server-side hp handlers */
- struct ptlrpc_hpreq_ops *rq_ops;
- /** history sequence # */
- __u64 rq_history_seq;
- /** the index of service's srv_at_array into which request is linked */
- time_t rq_at_index;
- /** Result of request processing */
- int rq_status;
- /** Lock to protect request flags and some other important bits, like
- * rq_list
- */
- cfs_spinlock_t rq_lock;
- /** client-side flags are serialized by rq_lock */
- unsigned long rq_intr:1, rq_replied:1, rq_err:1,
+ /* Request type: one of PTL_RPC_MSG_* */
+ int rq_type;
+ /** Result of request processing */
+ int rq_status;
+ /**
+ * Linkage item through which this request is included into
+ * sending/delayed lists on client and into rqbd list on server
+ */
+ struct list_head rq_list;
+ /** Lock to protect request flags and some other important bits, like
+ * rq_list
+ */
+ spinlock_t rq_lock;
+ /** client-side flags are serialized by rq_lock @{ */
+ unsigned int rq_intr:1, rq_replied:1, rq_err:1,
rq_timedout:1, rq_resend:1, rq_restart:1,
/**
* when ->rq_replay is set, request is kept by the client even
rq_replay:1,
rq_no_resend:1, rq_waiting:1, rq_receiving_reply:1,
rq_no_delay:1, rq_net_err:1, rq_wait_ctx:1,
- rq_early:1, rq_must_unlink:1,
- rq_fake:1, /* this fake req */
- rq_memalloc:1, /* req originated from "kswapd" */
- /* server-side flags */
- rq_packed_final:1, /* packed final reply */
- rq_hp:1, /* high priority RPC */
- rq_at_linked:1, /* link into service's srv_at_array */
- rq_reply_truncate:1,
- rq_committed:1,
- /* whether the "rq_set" is a valid one */
- rq_invalid_rqset:1;
-
- enum rq_phase rq_phase; /* one of RQ_PHASE_* */
- enum rq_phase rq_next_phase; /* one of RQ_PHASE_* to be used next */
- cfs_atomic_t rq_refcount;/* client-side refcount for SENT race,
- server-side refcounf for multiple replies */
-
- /** initial thread servicing this request */
- struct ptlrpc_thread *rq_svc_thread;
-
- /** Portal to which this request would be sent */
- int rq_request_portal; /* XXX FIXME bug 249 */
- /** Portal where to wait for reply and where reply would be sent */
- int rq_reply_portal; /* XXX FIXME bug 249 */
-
+ rq_early:1,
+ rq_req_unlinked:1, /* unlinked request buffer from lnet */
+ rq_reply_unlinked:1, /* unlinked reply buffer from lnet */
+ rq_memalloc:1, /* req originated from "kswapd" */
+ rq_committed:1,
+ rq_reply_truncated:1,
+ /** whether the "rq_set" is a valid one */
+ rq_invalid_rqset:1,
+ rq_generation_set:1,
+ /** do not resend request on -EINPROGRESS */
+ rq_no_retry_einprogress:1,
+ /* allow the req to be sent if the import is in recovery
+ * status */
+ rq_allow_replay:1,
+ /* bulk request, sent to server, but uncommitted */
+ rq_unstable:1;
+ /** @} */
+
+ /** server-side flags @{ */
+ unsigned int
+ rq_hp:1, /**< high priority RPC */
+ rq_at_linked:1, /**< link into service's srv_at_array */
+ rq_packed_final:1; /**< packed final reply */
+ /** @} */
+
+ /** one of RQ_PHASE_* */
+ enum rq_phase rq_phase;
+ /** one of RQ_PHASE_* to be used next */
+ enum rq_phase rq_next_phase;
+ /**
+ * client-side refcount for SENT race, server-side refcounf
+ * for multiple replies
+ */
+ atomic_t rq_refcount;
/**
* client-side:
* !rq_truncate : # reply bytes actually received,
int rq_nob_received;
/** Request length */
int rq_reqlen;
- /** Request message - what client sent */
- struct lustre_msg *rq_reqmsg;
-
/** Reply length */
int rq_replen;
+ /** Pool if request is from preallocated list */
+ struct ptlrpc_request_pool *rq_pool;
+ /** Request message - what client sent */
+ struct lustre_msg *rq_reqmsg;
/** Reply message - server response */
struct lustre_msg *rq_repmsg;
/** Transaction number */
__u64 rq_transno;
/** xid */
__u64 rq_xid;
- /**
- * List item to for replay list. Not yet commited requests get linked
- * there.
- * Also see \a rq_replay comment above.
- */
- cfs_list_t rq_replay_list;
-
- /**
- * security and encryption data
- * @{ */
- struct ptlrpc_cli_ctx *rq_cli_ctx; /**< client's half ctx */
- struct ptlrpc_svc_ctx *rq_svc_ctx; /**< server's half ctx */
- cfs_list_t rq_ctx_chain; /**< link to waited ctx */
-
- struct sptlrpc_flavor rq_flvr; /**< for client & server */
- enum lustre_sec_part rq_sp_from;
-
- unsigned long /* client/server security flags */
+ /**
+ * List item to for replay list. Not yet commited requests get linked
+ * there.
+ * Also see \a rq_replay comment above.
+ * It's also link chain on obd_export::exp_req_replay_queue
+ */
+ struct list_head rq_replay_list;
+ /** non-shared members for client & server request*/
+ union {
+ struct ptlrpc_cli_req rq_cli;
+ struct ptlrpc_srv_req rq_srv;
+ };
+ /**
+ * security and encryption data
+ * @{ */
+ /** description of flavors for client & server */
+ struct sptlrpc_flavor rq_flvr;
+
+ /* client/server security flags */
+ unsigned int
rq_ctx_init:1, /* context initiation */
rq_ctx_fini:1, /* context destroy */
rq_bulk_read:1, /* request bulk read */
rq_pack_bulk:1,
/* doesn't expect reply FIXME */
rq_no_reply:1,
- rq_pill_init:1; /* pill initialized */
-
- uid_t rq_auth_uid; /* authed uid */
- uid_t rq_auth_mapped_uid; /* authed uid mapped to */
-
- /* (server side), pointed directly into req buffer */
- struct ptlrpc_user_desc *rq_user_desc;
+ rq_pill_init:1, /* pill initialized */
+ rq_srv_req:1; /* server request */
- /** early replies go to offset 0, regular replies go after that */
- unsigned int rq_reply_off;
- /* various buffer pointers */
- struct lustre_msg *rq_reqbuf; /* req wrapper */
+ /** various buffer pointers */
+ struct lustre_msg *rq_reqbuf; /**< req wrapper */
+ char *rq_repbuf; /**< rep buffer */
+ struct lustre_msg *rq_repdata; /**< rep wrapper msg */
+ /** only in priv mode */
+ struct lustre_msg *rq_clrbuf;
int rq_reqbuf_len; /* req wrapper buf len */
int rq_reqdata_len; /* req wrapper msg len */
- char *rq_repbuf; /* rep buffer */
int rq_repbuf_len; /* rep buffer len */
- struct lustre_msg *rq_repdata; /* rep wrapper msg */
int rq_repdata_len; /* rep wrapper msg len */
- struct lustre_msg *rq_clrbuf; /* only in priv mode */
int rq_clrbuf_len; /* only in priv mode */
int rq_clrdata_len; /* only in priv mode */
- /** @} */
-
- /** Fields that help to see if request and reply were swabbed or not */
- __u32 rq_req_swab_mask;
- __u32 rq_rep_swab_mask;
-
- /** What was import generation when this request was sent */
- int rq_import_generation;
- enum lustre_imp_state rq_send_state;
-
- /** how many early replies (for stats) */
- int rq_early_count;
-
- /** client+server request */
- lnet_handle_md_t rq_req_md_h;
- struct ptlrpc_cb_id rq_req_cbid;
- /** optional time limit for send attempts */
- cfs_duration_t rq_delay_limit;
- /** time request was first queued */
- cfs_time_t rq_queued_time;
-
- /* server-side... */
- /** request arrival time */
- struct timeval rq_arrival_time;
- /** separated reply state */
- struct ptlrpc_reply_state *rq_reply_state;
- /** incoming request buffer */
- struct ptlrpc_request_buffer_desc *rq_rqbd;
-#ifdef CRAY_XT3
- __u32 rq_uid; /* peer uid, used in MDS only */
-#endif
-
- /** client-only incoming reply */
- lnet_handle_md_t rq_reply_md_h;
- cfs_waitq_t rq_reply_waitq;
- struct ptlrpc_cb_id rq_reply_cbid;
-
- /** our LNet NID */
- lnet_nid_t rq_self;
- /** Peer description (the other side) */
- lnet_process_id_t rq_peer;
- /** Server-side, export on which request was received */
- struct obd_export *rq_export;
- /** Client side, import where request is being sent */
- struct obd_import *rq_import;
-
- /** Replay callback, called after request is replayed at recovery */
- void (*rq_replay_cb)(struct ptlrpc_request *);
- /**
- * Commit callback, called when request is committed and about to be
- * freed.
- */
- void (*rq_commit_cb)(struct ptlrpc_request *);
- /** Opaq data for replay and commit callbacks. */
- void *rq_cb_data;
-
- /** For bulk requests on client only: bulk descriptor */
- struct ptlrpc_bulk_desc *rq_bulk;
-
- /** client outgoing req */
- /**
- * when request/reply sent (secs), or time when request should be sent
- */
- time_t rq_sent;
- /** time for request really sent out */
- time_t rq_real_sent;
-
- /** when request must finish. volatile
- * so that servers' early reply updates to the deadline aren't
- * kept in per-cpu cache */
- volatile time_t rq_deadline;
- /** when req reply unlink must finish. */
- time_t rq_reply_deadline;
- /** when req bulk unlink must finish. */
- time_t rq_bulk_deadline;
- /**
- * service time estimate (secs)
- * If the requestsis not served by this time, it is marked as timed out.
- */
- int rq_timeout;
-
- /** Multi-rpc bits */
- /** Link item for request set lists */
- cfs_list_t rq_set_chain;
- /** Per-request waitq introduced by bug 21938 for recovery waiting */
- cfs_waitq_t rq_set_waitq;
- /** Link back to the request set */
- struct ptlrpc_request_set *rq_set;
- /** Async completion handler, called when reply is received */
- ptlrpc_interpterer_t rq_interpret_reply;
- /** Async completion context */
- union ptlrpc_async_args rq_async_args;
-
- /** Pool if request is from preallocated list */
- struct ptlrpc_request_pool *rq_pool;
-
- struct lu_context rq_session;
- struct lu_context rq_recov_session;
-
- /** request format description */
- struct req_capsule rq_pill;
+ /** early replies go to offset 0, regular replies go after that */
+ unsigned int rq_reply_off;
+
+ /** @} */
+
+ /** Fields that help to see if request and reply were swabbed or not */
+ __u32 rq_req_swab_mask;
+ __u32 rq_rep_swab_mask;
+
+ /** how many early replies (for stats) */
+ int rq_early_count;
+ /** Server-side, export on which request was received */
+ struct obd_export *rq_export;
+ /** import where request is being sent */
+ struct obd_import *rq_import;
+ /** our LNet NID */
+ lnet_nid_t rq_self;
+ /** Peer description (the other side) */
+ lnet_process_id_t rq_peer;
+ /**
+ * service time estimate (secs)
+ * If the request is not served by this time, it is marked as timed out.
+ */
+ int rq_timeout;
+ /**
+ * when request/reply sent (secs), or time when request should be sent
+ */
+ time_t rq_sent;
+ /** when request must finish. */
+ time_t rq_deadline;
+ /** request format description */
+ struct req_capsule rq_pill;
};
/**
return rc;
}
+/** \addtogroup nrs
+ * @{
+ */
+int ptlrpc_nrs_policy_register(struct ptlrpc_nrs_pol_conf *conf);
+int ptlrpc_nrs_policy_unregister(struct ptlrpc_nrs_pol_conf *conf);
+void ptlrpc_nrs_req_hp_move(struct ptlrpc_request *req);
+void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
+ struct ptlrpc_nrs_pol_info *info);
+
+/*
+ * Can the request be moved from the regular NRS head to the high-priority NRS
+ * head (of the same PTLRPC service partition), if any?
+ *
+ * For a reliable result, this should be checked under svcpt->scp_req lock.
+ */
+static inline bool ptlrpc_nrs_req_can_move(struct ptlrpc_request *req)
+{
+ struct ptlrpc_nrs_request *nrq = &req->rq_nrq;
+
+ /**
+ * LU-898: Check ptlrpc_nrs_request::nr_enqueued to make sure the
+ * request has been enqueued first, and ptlrpc_nrs_request::nr_started
+ * to make sure it has not been scheduled yet (analogous to previous
+ * (non-NRS) checking of !list_empty(&ptlrpc_request::rq_list).
+ */
+ return nrq->nr_enqueued && !nrq->nr_started && !req->rq_hp;
+}
+/** @} nrs */
+
/**
* Returns 1 if request buffer at offset \a index was already swabbed
*/
-static inline int lustre_req_swabbed(struct ptlrpc_request *req, int index)
+static inline int lustre_req_swabbed(struct ptlrpc_request *req, size_t index)
{
LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
return req->rq_req_swab_mask & (1 << index);
/**
* Returns 1 if request reply buffer at offset \a index was already swabbed
*/
-static inline int lustre_rep_swabbed(struct ptlrpc_request *req, int index)
+static inline int lustre_rep_swabbed(struct ptlrpc_request *req, size_t index)
{
LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
return req->rq_rep_swab_mask & (1 << index);
/**
* Mark request buffer at offset \a index that it was already swabbed
*/
-static inline void lustre_set_req_swabbed(struct ptlrpc_request *req, int index)
+static inline void lustre_set_req_swabbed(struct ptlrpc_request *req,
+ size_t index)
{
LASSERT(index < sizeof(req->rq_req_swab_mask) * 8);
LASSERT((req->rq_req_swab_mask & (1 << index)) == 0);
/**
* Mark request reply buffer at offset \a index that it was already swabbed
*/
-static inline void lustre_set_rep_swabbed(struct ptlrpc_request *req, int index)
+static inline void lustre_set_rep_swabbed(struct ptlrpc_request *req,
+ size_t index)
{
LASSERT(index < sizeof(req->rq_rep_swab_mask) * 8);
LASSERT((req->rq_rep_swab_mask & (1 << index)) == 0);
/**
* Debugging functions and helpers to print request structure into debug log
* @{
- */
+ */
/* Spare the preprocessor, spoil the bugs. */
#define FLAG(field, str) (field ? str : "")
#define REQ_FLAGS_FMT "%s:%s%s%s%s%s%s%s%s%s%s%s%s"
-void _debug_req(struct ptlrpc_request *req, __u32 mask,
+void _debug_req(struct ptlrpc_request *req,
struct libcfs_debug_msg_data *data, const char *fmt, ...)
- __attribute__ ((format (printf, 4, 5)));
+ __attribute__ ((format (printf, 3, 4)));
/**
* Helper that decides if we need to print request accordig to current debug
* level settings
*/
-#define debug_req(cdls, level, req, file, func, line, fmt, a...) \
+#define debug_req(msgdata, mask, cdls, req, fmt, a...) \
do { \
- CFS_CHECK_STACK(); \
+ CFS_CHECK_STACK(msgdata, mask, cdls); \
\
- if (((level) & D_CANTMASK) != 0 || \
- ((libcfs_debug & (level)) != 0 && \
- (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) { \
- static struct libcfs_debug_msg_data _req_dbg_data = \
- DEBUG_MSG_DATA_INIT(cdls, DEBUG_SUBSYSTEM, file, func, line); \
- _debug_req((req), (level), &_req_dbg_data, fmt, ##a); \
- } \
+ if (((mask) & D_CANTMASK) != 0 || \
+ ((libcfs_debug & (mask)) != 0 && \
+ (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) \
+ _debug_req((req), msgdata, fmt, ##a); \
} while(0)
/**
do { \
if ((level) & (D_ERROR | D_WARNING)) { \
static cfs_debug_limit_state_t cdls; \
- debug_req(&cdls, level, req, __FILE__, __func__, __LINE__, \
- "@@@ "fmt" ", ## args); \
- } else \
- debug_req(NULL, level, req, __FILE__, __func__, __LINE__, \
- "@@@ "fmt" ", ## args); \
+ LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, level, &cdls); \
+ debug_req(&msgdata, level, &cdls, req, "@@@ "fmt" ", ## args);\
+ } else { \
+ LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, level, NULL); \
+ debug_req(&msgdata, level, NULL, req, "@@@ "fmt" ", ## args); \
+ } \
} while (0)
/** @} */
* Structure that defines a single page of a bulk transfer
*/
struct ptlrpc_bulk_page {
- /** Linkage to list of pages in a bulk */
- cfs_list_t bp_link;
- /**
- * Number of bytes in a page to transfer starting from \a bp_pageoffset
- */
- int bp_buflen;
- /** offset within a page */
- int bp_pageoffset;
- /** The page itself */
- struct page *bp_page;
+ /** Linkage to list of pages in a bulk */
+ struct list_head bp_link;
+ /**
+ * Number of bytes in a page to transfer starting from \a bp_pageoffset
+ */
+ int bp_buflen;
+ /** offset within a page */
+ int bp_pageoffset;
+ /** The page itself */
+ struct page *bp_page;
};
#define BULK_GET_SOURCE 0
#define BULK_PUT_SOURCE 3
/**
- * Definition of buk descriptor.
+ * Definition of bulk descriptor.
* Bulks are special "Two phase" RPCs where initial request message
* is sent first and it is followed bt a transfer (o receiving) of a large
* amount of data to be settled into pages referenced from the bulk descriptors.
* Another user is readpage for MDT.
*/
struct ptlrpc_bulk_desc {
- /** completed successfully */
- unsigned long bd_success:1;
- /** accessible to the network (network io potentially in progress) */
- unsigned long bd_network_rw:1;
- /** {put,get}{source,sink} */
- unsigned long bd_type:2;
- /** client side */
- unsigned long bd_registered:1;
- /** For serialization with callback */
- cfs_spinlock_t bd_lock;
- /** Import generation when request for this bulk was sent */
- int bd_import_generation;
- /** Server side - export this bulk created for */
- struct obd_export *bd_export;
- /** Client side - import this bulk was sent on */
- struct obd_import *bd_import;
- /** LNet portal for this bulk */
- __u32 bd_portal;
- /** Back pointer to the request */
- struct ptlrpc_request *bd_req;
- cfs_waitq_t bd_waitq; /* server side only WQ */
- int bd_iov_count; /* # entries in bd_iov */
- int bd_max_iov; /* allocated size of bd_iov */
- int bd_nob; /* # bytes covered */
- int bd_nob_transferred; /* # bytes GOT/PUT */
-
- __u64 bd_last_xid;
-
- struct ptlrpc_cb_id bd_cbid; /* network callback info */
- lnet_handle_md_t bd_md_h; /* associated MD */
- lnet_nid_t bd_sender; /* stash event::sender */
-
-#if defined(__KERNEL__)
- /*
- * encrypt iov, size is either 0 or bd_iov_count.
- */
- lnet_kiov_t *bd_enc_iov;
-
- lnet_kiov_t bd_iov[0];
-#else
- lnet_md_iovec_t bd_iov[0];
-#endif
+ /** completed with failure */
+ unsigned long bd_failure:1;
+ /** {put,get}{source,sink} */
+ unsigned long bd_type:2;
+ /** client side */
+ unsigned long bd_registered:1;
+ /** For serialization with callback */
+ spinlock_t bd_lock;
+ /** Import generation when request for this bulk was sent */
+ int bd_import_generation;
+ /** LNet portal for this bulk */
+ __u32 bd_portal;
+ /** Server side - export this bulk created for */
+ struct obd_export *bd_export;
+ /** Client side - import this bulk was sent on */
+ struct obd_import *bd_import;
+ /** Back pointer to the request */
+ struct ptlrpc_request *bd_req;
+ wait_queue_head_t bd_waitq; /* server side only WQ */
+ int bd_iov_count; /* # entries in bd_iov */
+ int bd_max_iov; /* allocated size of bd_iov */
+ int bd_nob; /* # bytes covered */
+ int bd_nob_transferred; /* # bytes GOT/PUT */
+
+ __u64 bd_last_xid;
+
+ struct ptlrpc_cb_id bd_cbid; /* network callback info */
+ lnet_nid_t bd_sender; /* stash event::sender */
+ int bd_md_count; /* # valid entries in bd_mds */
+ int bd_md_max_brw; /* max entries in bd_mds */
+ /** array of associated MDs */
+ lnet_handle_md_t bd_mds[PTLRPC_BULK_OPS_COUNT];
+
+ /*
+ * encrypt iov, size is either 0 or bd_iov_count.
+ */
+ lnet_kiov_t *bd_enc_iov;
+
+ lnet_kiov_t bd_iov[0];
};
enum {
SVC_SIGNAL = 1 << 5,
};
+#define PTLRPC_THR_NAME_LEN 32
/**
* Definition of server service thread structure
*/
/**
* List of active threads in svc->srv_threads
*/
- cfs_list_t t_link;
+ struct list_head t_link;
/**
* thread-private data (preallocated memory)
*/
/**
* service thread pid
*/
- pid_t t_pid;
+ pid_t t_pid;
/**
* put watchdog in the structure per thread b=14840
*/
/**
* the svc this thread belonged to b=18582
*/
- struct ptlrpc_service *t_svc;
- cfs_waitq_t t_ctl_waitq;
- struct lu_env *t_env;
+ struct ptlrpc_service_part *t_svcpt;
+ wait_queue_head_t t_ctl_waitq;
+ struct lu_env *t_env;
+ char t_name[PTLRPC_THR_NAME_LEN];
};
+static inline int thread_is_init(struct ptlrpc_thread *thread)
+{
+ return thread->t_flags == 0;
+}
+
static inline int thread_is_stopped(struct ptlrpc_thread *thread)
{
return !!(thread->t_flags & SVC_STOPPED);
* More than one request can fit into the buffer.
*/
struct ptlrpc_request_buffer_desc {
- /** Link item for rqbds on a service */
- cfs_list_t rqbd_list;
- /** History of requests for this buffer */
- cfs_list_t rqbd_reqs;
- /** Back pointer to service for which this buffer is registered */
- struct ptlrpc_service *rqbd_service;
- /** LNet descriptor */
- lnet_handle_md_t rqbd_md_h;
- int rqbd_refcount;
- /** The buffer itself */
- char *rqbd_buffer;
- struct ptlrpc_cb_id rqbd_cbid;
- /**
- * This "embedded" request structure is only used for the
- * last request to fit into the buffer
- */
- struct ptlrpc_request rqbd_req;
+ /** Link item for rqbds on a service */
+ struct list_head rqbd_list;
+ /** History of requests for this buffer */
+ struct list_head rqbd_reqs;
+ /** Back pointer to service for which this buffer is registered */
+ struct ptlrpc_service_part *rqbd_svcpt;
+ /** LNet descriptor */
+ lnet_handle_md_t rqbd_md_h;
+ int rqbd_refcount;
+ /** The buffer itself */
+ char *rqbd_buffer;
+ struct ptlrpc_cb_id rqbd_cbid;
+ /**
+ * This "embedded" request structure is only used for the
+ * last request to fit into the buffer
+ */
+ struct ptlrpc_request rqbd_req;
};
-typedef int (*svc_thr_init_t)(struct ptlrpc_thread *thread);
-typedef void (*svc_thr_done_t)(struct ptlrpc_thread *thread);
typedef int (*svc_handler_t)(struct ptlrpc_request *req);
-typedef int (*svc_hpreq_handler_t)(struct ptlrpc_request *);
-typedef void (*svc_req_printfn_t)(void *, struct ptlrpc_request *);
+
+struct ptlrpc_service_ops {
+ /**
+ * if non-NULL called during thread creation (ptlrpc_start_thread())
+ * to initialize service specific per-thread state.
+ */
+ int (*so_thr_init)(struct ptlrpc_thread *thr);
+ /**
+ * if non-NULL called during thread shutdown (ptlrpc_main()) to
+ * destruct state created by ->srv_init().
+ */
+ void (*so_thr_done)(struct ptlrpc_thread *thr);
+ /**
+ * Handler function for incoming requests for this service
+ */
+ int (*so_req_handler)(struct ptlrpc_request *req);
+ /**
+ * function to determine priority of the request, it's called
+ * on every new request
+ */
+ int (*so_hpreq_handler)(struct ptlrpc_request *);
+ /**
+ * service-specific print fn
+ */
+ void (*so_req_printer)(void *, struct ptlrpc_request *);
+};
#ifndef __cfs_cacheline_aligned
/* NB: put it here for reducing patche dependence */
* The service is listening on a particular portal (like tcp port)
* and perform actions for a specific server like IO service for OST
* or general metadata service for MDS.
- *
- * ptlrpc service has four locks:
- * \a srv_lock
- * serialize operations on rqbd and requests waiting for preprocess
- * \a srv_rq_lock
- * serialize operations active requests sent to this portal
- * \a srv_at_lock
- * serialize adaptive timeout stuff
- * \a srv_rs_lock
- * serialize operations on RS list (reply states)
- *
- * We don't have any use-case to take two or more locks at the same time
- * for now, so there is no lock order issue.
*/
struct ptlrpc_service {
- /** most often accessed fields */
- /** chain thru all services */
- cfs_list_t srv_list;
+ /** serialize /proc operations */
+ spinlock_t srv_lock;
+ /** most often accessed fields */
+ /** chain thru all services */
+ struct list_head srv_list;
+ /** service operations table */
+ struct ptlrpc_service_ops srv_ops;
/** only statically allocated strings here; we don't clean them */
char *srv_name;
/** only statically allocated strings here; we don't clean them */
char *srv_thread_name;
/** service thread list */
- cfs_list_t srv_threads;
- /** threads to start at beginning of service */
- int srv_threads_min;
- /** thread upper limit */
- int srv_threads_max;
- /** always increasing number */
- unsigned srv_threads_next_id;
- /** # of starting threads */
- int srv_threads_starting;
- /** # running threads */
- int srv_threads_running;
-
- /** service operations, move to ptlrpc_svc_ops_t in the future */
- /** @{ */
- /**
- * if non-NULL called during thread creation (ptlrpc_start_thread())
- * to initialize service specific per-thread state.
- */
- svc_thr_init_t srv_init;
- /**
- * if non-NULL called during thread shutdown (ptlrpc_main()) to
- * destruct state created by ->srv_init().
- */
- svc_thr_done_t srv_done;
- /** Handler function for incoming requests for this service */
- svc_handler_t srv_handler;
- /** hp request handler */
- svc_hpreq_handler_t srv_hpreq_handler;
- /** service-specific print fn */
- svc_req_printfn_t srv_req_printfn;
- /** @} */
-
+ struct list_head srv_threads;
+ /** threads # should be created for each partition on initializing */
+ int srv_nthrs_cpt_init;
+ /** limit of threads number for each partition */
+ int srv_nthrs_cpt_limit;
/** Root of /proc dir tree for this service */
- cfs_proc_dir_entry_t *srv_procroot;
+ struct proc_dir_entry *srv_procroot;
/** Pointer to statistic data for this service */
struct lprocfs_stats *srv_stats;
/** # hp per lp reqs to handle */
__u32 srv_ctx_tags;
/** soft watchdog timeout multiplier */
int srv_watchdog_factor;
- /** bind threads to CPUs */
- unsigned srv_cpu_affinity:1;
/** under unregister_service */
unsigned srv_is_stopping:1;
- /**
- * serialize the following fields, used for protecting
- * rqbd list and incoming requests waiting for preprocess
- */
- cfs_spinlock_t srv_lock __cfs_cacheline_aligned;
- /** incoming reqs */
- cfs_list_t srv_req_in_queue;
- /** total # req buffer descs allocated */
- int srv_nbufs;
- /** # posted request buffers */
- int srv_nrqbd_receiving;
- /** timeout before re-posting reqs, in tick */
- cfs_duration_t srv_rqbd_timeout;
- /** request buffers to be reposted */
- cfs_list_t srv_idle_rqbds;
- /** req buffers receiving */
- cfs_list_t srv_active_rqbds;
- /** request buffer history */
- cfs_list_t srv_history_rqbds;
- /** # request buffers in history */
- int srv_n_history_rqbds;
- /** max # request buffers in history */
- int srv_max_history_rqbds;
- /** request history */
- cfs_list_t srv_request_history;
- /** next request sequence # */
- __u64 srv_request_seq;
- /** highest seq culled from history */
- __u64 srv_request_max_cull_seq;
- /**
- * all threads sleep on this. This wait-queue is signalled when new
- * incoming request arrives and when difficult reply has to be handled.
- */
- cfs_waitq_t srv_waitq;
-
- /**
- * serialize the following fields, used for processing requests
- * sent to this portal
- */
- cfs_spinlock_t srv_rq_lock __cfs_cacheline_aligned;
- /** # reqs in either of the queues below */
- /** reqs waiting for service */
- cfs_list_t srv_request_queue;
- /** high priority queue */
- cfs_list_t srv_request_hpq;
- /** # incoming reqs */
- int srv_n_queued_reqs;
- /** # reqs being served */
- int srv_n_active_reqs;
- /** # HPreqs being served */
- int srv_n_active_hpreq;
- /** # hp requests handled */
- int srv_hpreq_count;
-
- /** AT stuff */
- /** @{ */
- /**
- * serialize the following fields, used for changes on
- * adaptive timeout
- */
- cfs_spinlock_t srv_at_lock __cfs_cacheline_aligned;
- /** estimated rpc service time */
- struct adaptive_timeout srv_at_estimate;
- /** reqs waiting for replies */
- struct ptlrpc_at_array srv_at_array;
- /** early reply timer */
- cfs_timer_t srv_at_timer;
- /** check early replies */
- unsigned srv_at_check;
- /** debug */
- cfs_time_t srv_at_checktime;
- /** @} */
+ /** max # request buffers in history per partition */
+ int srv_hist_nrqbds_cpt_max;
+ /** number of CPTs this service bound on */
+ int srv_ncpts;
+ /** CPTs array this service bound on */
+ __u32 *srv_cpts;
+ /** 2^srv_cptab_bits >= cfs_cpt_numbert(srv_cptable) */
+ int srv_cpt_bits;
+ /** CPT table this service is running over */
+ struct cfs_cpt_table *srv_cptable;
+ /**
+ * partition data for ptlrpc service
+ */
+ struct ptlrpc_service_part *srv_parts[0];
+};
- /**
- * serialize the following fields, used for processing
- * replies for this portal
- */
- cfs_spinlock_t srv_rs_lock __cfs_cacheline_aligned;
- /** all the active replies */
- cfs_list_t srv_active_replies;
-#ifndef __KERNEL__
- /** replies waiting for service */
- cfs_list_t srv_reply_queue;
-#endif
- /** List of free reply_states */
- cfs_list_t srv_free_rs_list;
- /** waitq to run, when adding stuff to srv_free_rs_list */
- cfs_waitq_t srv_free_rs_waitq;
- /** # 'difficult' replies */
- cfs_atomic_t srv_n_difficult_replies;
- //struct ptlrpc_srv_ni srv_interfaces[0];
+/**
+ * Definition of PortalRPC service partition data.
+ * Although a service only has one instance of it right now, but we
+ * will have multiple instances very soon (instance per CPT).
+ *
+ * it has four locks:
+ * \a scp_lock
+ * serialize operations on rqbd and requests waiting for preprocess
+ * \a scp_req_lock
+ * serialize operations active requests sent to this portal
+ * \a scp_at_lock
+ * serialize adaptive timeout stuff
+ * \a scp_rep_lock
+ * serialize operations on RS list (reply states)
+ *
+ * We don't have any use-case to take two or more locks at the same time
+ * for now, so there is no lock order issue.
+ */
+struct ptlrpc_service_part {
+ /** back reference to owner */
+ struct ptlrpc_service *scp_service __cfs_cacheline_aligned;
+ /* CPT id, reserved */
+ int scp_cpt;
+ /** always increasing number */
+ int scp_thr_nextid;
+ /** # of starting threads */
+ int scp_nthrs_starting;
+ /** # of stopping threads, reserved for shrinking threads */
+ int scp_nthrs_stopping;
+ /** # running threads */
+ int scp_nthrs_running;
+ /** service threads list */
+ struct list_head scp_threads;
+
+ /**
+ * serialize the following fields, used for protecting
+ * rqbd list and incoming requests waiting for preprocess,
+ * threads starting & stopping are also protected by this lock.
+ */
+ spinlock_t scp_lock __cfs_cacheline_aligned;
+ /** total # req buffer descs allocated */
+ int scp_nrqbds_total;
+ /** # posted request buffers for receiving */
+ int scp_nrqbds_posted;
+ /** in progress of allocating rqbd */
+ int scp_rqbd_allocating;
+ /** # incoming reqs */
+ int scp_nreqs_incoming;
+ /** request buffers to be reposted */
+ struct list_head scp_rqbd_idle;
+ /** req buffers receiving */
+ struct list_head scp_rqbd_posted;
+ /** incoming reqs */
+ struct list_head scp_req_incoming;
+ /** timeout before re-posting reqs, in tick */
+ cfs_duration_t scp_rqbd_timeout;
+ /**
+ * all threads sleep on this. This wait-queue is signalled when new
+ * incoming request arrives and when difficult reply has to be handled.
+ */
+ wait_queue_head_t scp_waitq;
+
+ /** request history */
+ struct list_head scp_hist_reqs;
+ /** request buffer history */
+ struct list_head scp_hist_rqbds;
+ /** # request buffers in history */
+ int scp_hist_nrqbds;
+ /** sequence number for request */
+ __u64 scp_hist_seq;
+ /** highest seq culled from history */
+ __u64 scp_hist_seq_culled;
+
+ /**
+ * serialize the following fields, used for processing requests
+ * sent to this portal
+ */
+ spinlock_t scp_req_lock __cfs_cacheline_aligned;
+ /** # reqs in either of the NRS heads below */
+ /** # reqs being served */
+ int scp_nreqs_active;
+ /** # HPreqs being served */
+ int scp_nhreqs_active;
+ /** # hp requests handled */
+ int scp_hreq_count;
+
+ /** NRS head for regular requests */
+ struct ptlrpc_nrs scp_nrs_reg;
+ /** NRS head for HP requests; this is only valid for services that can
+ * handle HP requests */
+ struct ptlrpc_nrs *scp_nrs_hp;
+
+ /** AT stuff */
+ /** @{ */
+ /**
+ * serialize the following fields, used for changes on
+ * adaptive timeout
+ */
+ spinlock_t scp_at_lock __cfs_cacheline_aligned;
+ /** estimated rpc service time */
+ struct adaptive_timeout scp_at_estimate;
+ /** reqs waiting for replies */
+ struct ptlrpc_at_array scp_at_array;
+ /** early reply timer */
+ struct timer_list scp_at_timer;
+ /** debug */
+ cfs_time_t scp_at_checktime;
+ /** check early replies */
+ unsigned scp_at_check;
+ /** @} */
+
+ /**
+ * serialize the following fields, used for processing
+ * replies for this portal
+ */
+ spinlock_t scp_rep_lock __cfs_cacheline_aligned;
+ /** all the active replies */
+ struct list_head scp_rep_active;
+ /** List of free reply_states */
+ struct list_head scp_rep_idle;
+ /** waitq to run, when adding stuff to srv_free_rs_list */
+ wait_queue_head_t scp_rep_waitq;
+ /** # 'difficult' replies */
+ atomic_t scp_nreps_difficult;
};
+#define ptlrpc_service_for_each_part(part, i, svc) \
+ for (i = 0; \
+ i < (svc)->srv_ncpts && \
+ (svc)->srv_parts != NULL && \
+ ((part) = (svc)->srv_parts[i]) != NULL; i++)
+
/**
* Declaration of ptlrpcd control structure
*/
struct ptlrpcd_ctl {
- /**
- * Ptlrpc thread control flags (LIOD_START, LIOD_STOP, LIOD_FORCE)
- */
- unsigned long pc_flags;
- /**
- * Thread lock protecting structure fields.
- */
- cfs_spinlock_t pc_lock;
- /**
- * Start completion.
- */
- cfs_completion_t pc_starting;
- /**
- * Stop completion.
- */
- cfs_completion_t pc_finishing;
+ /**
+ * Ptlrpc thread control flags (LIOD_START, LIOD_STOP, LIOD_FORCE)
+ */
+ unsigned long pc_flags;
+ /**
+ * Thread lock protecting structure fields.
+ */
+ spinlock_t pc_lock;
+ /**
+ * Start completion.
+ */
+ struct completion pc_starting;
+ /**
+ * Stop completion.
+ */
+ struct completion pc_finishing;
/**
* Thread requests set.
*/
struct ptlrpc_request_set *pc_set;
/**
- * Thread name used in cfs_daemonize()
+ * Thread name used in kthread_run()
*/
char pc_name[16];
/**
* Record the partner index to be processed next.
*/
int pc_cursor;
-#ifndef __KERNEL__
- /**
- * Async rpcs flag to make sure that ptlrpcd_check() is called only
- * once.
- */
- int pc_recurred;
- /**
- * Currently not used.
- */
- void *pc_callback;
- /**
- * User-space async rpcs callback.
- */
- void *pc_wait_callback;
- /**
- * User-space check idle rpcs callback.
- */
- void *pc_idle_callback;
-#endif
};
/* Bits for pc_flags */
LIOD_BIND = 1 << 4,
};
+/**
+ * \addtogroup nrs
+ * @{
+ *
+ * Service compatibility function; the policy is compatible with all services.
+ *
+ * \param[in] svc The service the policy is attempting to register with.
+ * \param[in] desc The policy descriptor
+ *
+ * \retval true The policy is compatible with the service
+ *
+ * \see ptlrpc_nrs_pol_desc::pd_compat()
+ */
+static inline bool nrs_policy_compat_all(const struct ptlrpc_service *svc,
+ const struct ptlrpc_nrs_pol_desc *desc)
+{
+ return true;
+}
+
+/**
+ * Service compatibility function; the policy is compatible with only a specific
+ * service which is identified by its human-readable name at
+ * ptlrpc_service::srv_name.
+ *
+ * \param[in] svc The service the policy is attempting to register with.
+ * \param[in] desc The policy descriptor
+ *
+ * \retval false The policy is not compatible with the service
+ * \retval true The policy is compatible with the service
+ *
+ * \see ptlrpc_nrs_pol_desc::pd_compat()
+ */
+static inline bool nrs_policy_compat_one(const struct ptlrpc_service *svc,
+ const struct ptlrpc_nrs_pol_desc *desc)
+{
+ LASSERT(desc->pd_compat_svc_name != NULL);
+ return strcmp(svc->srv_name, desc->pd_compat_svc_name) == 0;
+}
+
+/** @} nrs */
+
/* ptlrpc/events.c */
extern lnet_handle_eq_t ptlrpc_eq_h;
extern int ptlrpc_uuid_to_peer(struct obd_uuid *uuid,
* underlying buffer
* @{
*/
-extern void request_out_callback (lnet_event_t *ev);
+extern void request_out_callback(lnet_event_t *ev);
extern void reply_in_callback(lnet_event_t *ev);
-extern void client_bulk_callback (lnet_event_t *ev);
+extern void client_bulk_callback(lnet_event_t *ev);
extern void request_in_callback(lnet_event_t *ev);
extern void reply_out_callback(lnet_event_t *ev);
-extern void server_bulk_callback (lnet_event_t *ev);
+#ifdef HAVE_SERVER_SUPPORT
+extern void server_bulk_callback(lnet_event_t *ev);
+#endif
/** @} */
/* ptlrpc/connection.c */
* Actual interfacing with LNet to put/get/register/unregister stuff
* @{
*/
+#ifdef HAVE_SERVER_SUPPORT
+struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_exp(struct ptlrpc_request *req,
+ unsigned npages, unsigned max_brw,
+ unsigned type, unsigned portal);
int ptlrpc_start_bulk_transfer(struct ptlrpc_bulk_desc *desc);
void ptlrpc_abort_bulk(struct ptlrpc_bulk_desc *desc);
-int ptlrpc_register_bulk(struct ptlrpc_request *req);
-int ptlrpc_unregister_bulk(struct ptlrpc_request *req, int async);
static inline int ptlrpc_server_bulk_active(struct ptlrpc_bulk_desc *desc)
{
- int rc;
+ int rc;
- LASSERT(desc != NULL);
+ LASSERT(desc != NULL);
- cfs_spin_lock(&desc->bd_lock);
- rc = desc->bd_network_rw;
- cfs_spin_unlock(&desc->bd_lock);
- return rc;
+ spin_lock(&desc->bd_lock);
+ rc = desc->bd_md_count;
+ spin_unlock(&desc->bd_lock);
+ return rc;
}
+#endif
+
+int ptlrpc_register_bulk(struct ptlrpc_request *req);
+int ptlrpc_unregister_bulk(struct ptlrpc_request *req, int async);
static inline int ptlrpc_client_bulk_active(struct ptlrpc_request *req)
{
- struct ptlrpc_bulk_desc *desc = req->rq_bulk;
+ struct ptlrpc_bulk_desc *desc;
int rc;
LASSERT(req != NULL);
+ desc = req->rq_bulk;
if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_BULK_UNLINK) &&
req->rq_bulk_deadline > cfs_time_current_sec())
if (!desc)
return 0;
- cfs_spin_lock(&desc->bd_lock);
- rc = desc->bd_network_rw;
- cfs_spin_unlock(&desc->bd_lock);
- return rc;
+ spin_lock(&desc->bd_lock);
+ rc = desc->bd_md_count;
+ spin_unlock(&desc->bd_lock);
+ return rc;
}
#define PTLRPC_REPLY_MAYBE_DIFFICULT 0x01
int ptlrpc_reply(struct ptlrpc_request *req);
int ptlrpc_send_error(struct ptlrpc_request *req, int difficult);
int ptlrpc_error(struct ptlrpc_request *req);
-void ptlrpc_resend_req(struct ptlrpc_request *request);
int ptlrpc_at_get_net_latency(struct ptlrpc_request *req);
int ptl_send_rpc(struct ptlrpc_request *request, int noreply);
-int ptlrpc_register_rqbd (struct ptlrpc_request_buffer_desc *rqbd);
+int ptlrpc_register_rqbd(struct ptlrpc_request_buffer_desc *rqbd);
/** @} */
/* ptlrpc/client.c */
* request queues, request management, etc.
* @{
*/
+void ptlrpc_request_committed(struct ptlrpc_request *req, int force);
+
void ptlrpc_init_client(int req_portal, int rep_portal, char *name,
struct ptlrpc_client *);
void ptlrpc_cleanup_client(struct obd_import *imp);
int ptlrpc_queue_wait(struct ptlrpc_request *req);
int ptlrpc_replay_req(struct ptlrpc_request *req);
-int ptlrpc_unregister_reply(struct ptlrpc_request *req, int async);
void ptlrpc_restart_req(struct ptlrpc_request *req);
void ptlrpc_abort_inflight(struct obd_import *imp);
void ptlrpc_cleanup_imp(struct obd_import *imp);
void ptlrpc_abort_set(struct ptlrpc_request_set *set);
struct ptlrpc_request_set *ptlrpc_prep_set(void);
+struct ptlrpc_request_set *ptlrpc_prep_fcset(int max, set_producer_func func,
+ void *arg);
int ptlrpc_set_add_cb(struct ptlrpc_request_set *set,
set_interpreter_func fn, void *data);
-int ptlrpc_set_next_timeout(struct ptlrpc_request_set *);
int ptlrpc_check_set(const struct lu_env *env, struct ptlrpc_request_set *set);
int ptlrpc_set_wait(struct ptlrpc_request_set *);
-int ptlrpc_expired_set(void *data);
-void ptlrpc_interrupted_set(void *data);
void ptlrpc_mark_interrupted(struct ptlrpc_request *req);
void ptlrpc_set_destroy(struct ptlrpc_request_set *);
void ptlrpc_set_add_req(struct ptlrpc_request_set *, struct ptlrpc_request *);
-void ptlrpc_set_add_new_req(struct ptlrpcd_ctl *pc,
- struct ptlrpc_request *req);
void ptlrpc_free_rq_pool(struct ptlrpc_request_pool *pool);
void ptlrpc_add_rqs_to_pool(struct ptlrpc_request_pool *pool, int num_rq);
int ptlrpc_request_bufs_pack(struct ptlrpc_request *request,
__u32 version, int opcode, char **bufs,
struct ptlrpc_cli_ctx *ctx);
-struct ptlrpc_request *ptlrpc_prep_fakereq(struct obd_import *imp,
- unsigned int timeout,
- ptlrpc_interpterer_t interpreter);
-void ptlrpc_fakereq_finished(struct ptlrpc_request *req);
-
struct ptlrpc_request *ptlrpc_prep_req(struct obd_import *imp, __u32 version,
int opcode, int count, __u32 *lengths,
char **bufs);
void ptlrpc_req_finished(struct ptlrpc_request *request);
void ptlrpc_req_finished_with_imp_lock(struct ptlrpc_request *request);
struct ptlrpc_request *ptlrpc_request_addref(struct ptlrpc_request *req);
-struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_imp (struct ptlrpc_request *req,
- int npages, int type, int portal);
-struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_exp(struct ptlrpc_request *req,
- int npages, int type, int portal);
-void ptlrpc_free_bulk(struct ptlrpc_bulk_desc *bulk);
-void ptlrpc_prep_bulk_page(struct ptlrpc_bulk_desc *desc,
- cfs_page_t *page, int pageoffset, int len);
+struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_imp(struct ptlrpc_request *req,
+ unsigned npages, unsigned max_brw,
+ unsigned type, unsigned portal);
+void __ptlrpc_free_bulk(struct ptlrpc_bulk_desc *bulk, int pin);
+static inline void ptlrpc_free_bulk_pin(struct ptlrpc_bulk_desc *bulk)
+{
+ __ptlrpc_free_bulk(bulk, 1);
+}
+static inline void ptlrpc_free_bulk_nopin(struct ptlrpc_bulk_desc *bulk)
+{
+ __ptlrpc_free_bulk(bulk, 0);
+}
+void __ptlrpc_prep_bulk_page(struct ptlrpc_bulk_desc *desc,
+ struct page *page, int pageoffset, int len, int);
+static inline void ptlrpc_prep_bulk_page_pin(struct ptlrpc_bulk_desc *desc,
+ struct page *page, int pageoffset,
+ int len)
+{
+ __ptlrpc_prep_bulk_page(desc, page, pageoffset, len, 1);
+}
+
+static inline void ptlrpc_prep_bulk_page_nopin(struct ptlrpc_bulk_desc *desc,
+ struct page *page, int pageoffset,
+ int len)
+{
+ __ptlrpc_prep_bulk_page(desc, page, pageoffset, len, 0);
+}
+
void ptlrpc_retain_replayable_request(struct ptlrpc_request *req,
struct obd_import *imp);
__u64 ptlrpc_next_xid(void);
int ptlrpcd_queue_work(void *handler);
/** @} */
+struct ptlrpc_service_buf_conf {
+ /* nbufs is buffers # to allocate when growing the pool */
+ unsigned int bc_nbufs;
+ /* buffer size to post */
+ unsigned int bc_buf_size;
+ /* portal to listed for requests on */
+ unsigned int bc_req_portal;
+ /* portal of where to send replies to */
+ unsigned int bc_rep_portal;
+ /* maximum request size to be accepted for this service */
+ unsigned int bc_req_max_size;
+ /* maximum reply size this service can ever send */
+ unsigned int bc_rep_max_size;
+};
+
+struct ptlrpc_service_thr_conf {
+ /* threadname should be 8 characters or less - 6 will be added on */
+ char *tc_thr_name;
+ /* threads increasing factor for each CPU */
+ unsigned int tc_thr_factor;
+ /* service threads # to start on each partition while initializing */
+ unsigned int tc_nthrs_init;
+ /*
+ * low water of threads # upper-limit on each partition while running,
+ * service availability may be impacted if threads number is lower
+ * than this value. It can be ZERO if the service doesn't require
+ * CPU affinity or there is only one partition.
+ */
+ unsigned int tc_nthrs_base;
+ /* "soft" limit for total threads number */
+ unsigned int tc_nthrs_max;
+ /* user specified threads number, it will be validated due to
+ * other members of this structure. */
+ unsigned int tc_nthrs_user;
+ /* set NUMA node affinity for service threads */
+ unsigned int tc_cpu_affinity;
+ /* Tags for lu_context associated with service thread */
+ __u32 tc_ctx_tags;
+};
+
+struct ptlrpc_service_cpt_conf {
+ struct cfs_cpt_table *cc_cptable;
+ /* string pattern to describe CPTs for a service */
+ char *cc_pattern;
+};
struct ptlrpc_service_conf {
- int psc_nbufs;
- int psc_bufsize;
- int psc_max_req_size;
- int psc_max_reply_size;
- int psc_req_portal;
- int psc_rep_portal;
- int psc_watchdog_factor;
- int psc_min_threads;
- int psc_max_threads;
- __u32 psc_ctx_tags;
+ /* service name */
+ char *psc_name;
+ /* soft watchdog timeout multiplifier to print stuck service traces */
+ unsigned int psc_watchdog_factor;
+ /* buffer information */
+ struct ptlrpc_service_buf_conf psc_buf;
+ /* thread information */
+ struct ptlrpc_service_thr_conf psc_thr;
+ /* CPU partition information */
+ struct ptlrpc_service_cpt_conf psc_cpt;
+ /* function table */
+ struct ptlrpc_service_ops psc_ops;
};
/* ptlrpc/service.c */
*
* @{
*/
-void ptlrpc_save_lock (struct ptlrpc_request *req,
- struct lustre_handle *lock, int mode, int no_ack);
+void ptlrpc_save_lock(struct ptlrpc_request *req,
+ struct lustre_handle *lock, int mode, int no_ack);
void ptlrpc_commit_replies(struct obd_export *exp);
-void ptlrpc_dispatch_difficult_reply (struct ptlrpc_reply_state *rs);
-void ptlrpc_schedule_difficult_reply (struct ptlrpc_reply_state *rs);
-struct ptlrpc_service *ptlrpc_init_svc_conf(struct ptlrpc_service_conf *c,
- svc_handler_t h, char *name,
- struct proc_dir_entry *proc_entry,
- svc_req_printfn_t prntfn,
- char *threadname);
-
-struct ptlrpc_service *ptlrpc_init_svc(int nbufs, int bufsize, int max_req_size,
- int max_reply_size,
- int req_portal, int rep_portal,
- int watchdog_factor,
- svc_handler_t, char *name,
- cfs_proc_dir_entry_t *proc_entry,
- svc_req_printfn_t,
- int min_threads, int max_threads,
- char *threadname, __u32 ctx_tags,
- svc_hpreq_handler_t);
+void ptlrpc_dispatch_difficult_reply(struct ptlrpc_reply_state *rs);
+void ptlrpc_schedule_difficult_reply(struct ptlrpc_reply_state *rs);
+int ptlrpc_hpreq_handler(struct ptlrpc_request *req);
+struct ptlrpc_service *ptlrpc_register_service(
+ struct ptlrpc_service_conf *conf,
+ struct proc_dir_entry *proc_entry);
void ptlrpc_stop_all_threads(struct ptlrpc_service *svc);
int ptlrpc_start_threads(struct ptlrpc_service *svc);
-int ptlrpc_start_thread(struct ptlrpc_service *svc);
int ptlrpc_unregister_service(struct ptlrpc_service *service);
-int liblustre_check_services (void *arg);
+int liblustre_check_services(void *arg);
void ptlrpc_daemonize(char *name);
int ptlrpc_service_health_check(struct ptlrpc_service *);
-void ptlrpc_hpreq_reorder(struct ptlrpc_request *req);
void ptlrpc_server_drop_request(struct ptlrpc_request *req);
+void ptlrpc_request_change_export(struct ptlrpc_request *req,
+ struct obd_export *export);
+void ptlrpc_update_export_timer(struct obd_export *exp, long extra_delay);
-#ifdef __KERNEL__
int ptlrpc_hr_init(void);
void ptlrpc_hr_fini(void);
-#else
-# define ptlrpc_hr_init() (0)
-# define ptlrpc_hr_fini() do {} while(0)
-#endif
-struct ptlrpc_svc_data {
- char *name;
- struct ptlrpc_service *svc;
- struct ptlrpc_thread *thread;
-};
/** @} */
/* ptlrpc/import.c */
int ptlrpc_init_import(struct obd_import *imp);
int ptlrpc_disconnect_import(struct obd_import *imp, int noclose);
int ptlrpc_import_recovery_state_machine(struct obd_import *imp);
+void deuuidify(char *uuid, const char *prefix, char **uuid_start,
+ int *uuid_len);
/* ptlrpc/pack_generic.c */
int ptlrpc_reconnect_import(struct obd_import *imp);
/** @} */
/**
- * ptlrpc msg buffer and swab interface
+ * ptlrpc msg buffer and swab interface
*
* @{
*/
int ptlrpc_buf_need_swab(struct ptlrpc_request *req, const int inout,
- int index);
+ __u32 index);
void ptlrpc_buf_set_swabbed(struct ptlrpc_request *req, const int inout,
- int index);
+ __u32 index);
int ptlrpc_unpack_rep_msg(struct ptlrpc_request *req, int len);
int ptlrpc_unpack_req_msg(struct ptlrpc_request *req, int len);
unsigned int newlen, int move_data);
void lustre_free_reply_state(struct ptlrpc_reply_state *rs);
int __lustre_unpack_msg(struct lustre_msg *m, int len);
-int lustre_msg_hdr_size(__u32 magic, int count);
-int lustre_msg_size(__u32 magic, int count, __u32 *lengths);
-int lustre_msg_size_v2(int count, __u32 *lengths);
-int lustre_packed_msg_size(struct lustre_msg *msg);
-int lustre_msg_early_size(void);
-void *lustre_msg_buf_v2(struct lustre_msg_v2 *m, int n, int min_size);
-void *lustre_msg_buf(struct lustre_msg *m, int n, int minlen);
-int lustre_msg_buflen(struct lustre_msg *m, int n);
-void lustre_msg_set_buflen(struct lustre_msg *m, int n, int len);
-int lustre_msg_bufcount(struct lustre_msg *m);
-char *lustre_msg_string (struct lustre_msg *m, int n, int max_len);
+__u32 lustre_msg_hdr_size(__u32 magic, __u32 count);
+__u32 lustre_msg_size(__u32 magic, int count, __u32 *lengths);
+__u32 lustre_msg_size_v2(int count, __u32 *lengths);
+__u32 lustre_packed_msg_size(struct lustre_msg *msg);
+__u32 lustre_msg_early_size(void);
+void *lustre_msg_buf_v2(struct lustre_msg_v2 *m, __u32 n, __u32 min_size);
+void *lustre_msg_buf(struct lustre_msg *m, __u32 n, __u32 minlen);
+__u32 lustre_msg_buflen(struct lustre_msg *m, __u32 n);
+void lustre_msg_set_buflen(struct lustre_msg *m, __u32 n, __u32 len);
+__u32 lustre_msg_bufcount(struct lustre_msg *m);
+char *lustre_msg_string(struct lustre_msg *m, __u32 n, __u32 max_len);
__u32 lustre_msghdr_get_flags(struct lustre_msg *msg);
void lustre_msghdr_set_flags(struct lustre_msg *msg, __u32 flags);
__u32 lustre_msg_get_flags(struct lustre_msg *msg);
-void lustre_msg_add_flags(struct lustre_msg *msg, int flags);
-void lustre_msg_set_flags(struct lustre_msg *msg, int flags);
-void lustre_msg_clear_flags(struct lustre_msg *msg, int flags);
+void lustre_msg_add_flags(struct lustre_msg *msg, __u32 flags);
+void lustre_msg_set_flags(struct lustre_msg *msg, __u32 flags);
+void lustre_msg_clear_flags(struct lustre_msg *msg, __u32 flags);
__u32 lustre_msg_get_op_flags(struct lustre_msg *msg);
-void lustre_msg_add_op_flags(struct lustre_msg *msg, int flags);
-void lustre_msg_set_op_flags(struct lustre_msg *msg, int flags);
+void lustre_msg_add_op_flags(struct lustre_msg *msg, __u32 flags);
+void lustre_msg_set_op_flags(struct lustre_msg *msg, __u32 flags);
struct lustre_handle *lustre_msg_get_handle(struct lustre_msg *msg);
__u32 lustre_msg_get_type(struct lustre_msg *msg);
__u32 lustre_msg_get_version(struct lustre_msg *msg);
-void lustre_msg_add_version(struct lustre_msg *msg, int version);
+void lustre_msg_add_version(struct lustre_msg *msg, __u32 version);
__u32 lustre_msg_get_opc(struct lustre_msg *msg);
__u64 lustre_msg_get_last_xid(struct lustre_msg *msg);
+__u16 lustre_msg_get_tag(struct lustre_msg *msg);
__u64 lustre_msg_get_last_committed(struct lustre_msg *msg);
__u64 *lustre_msg_get_versions(struct lustre_msg *msg);
__u64 lustre_msg_get_transno(struct lustre_msg *msg);
__u32 lustre_msg_get_magic(struct lustre_msg *msg);
__u32 lustre_msg_get_timeout(struct lustre_msg *msg);
__u32 lustre_msg_get_service_time(struct lustre_msg *msg);
+char *lustre_msg_get_jobid(struct lustre_msg *msg);
__u32 lustre_msg_get_cksum(struct lustre_msg *msg);
-#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 0, 0)
+#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 53, 0)
__u32 lustre_msg_calc_cksum(struct lustre_msg *msg, int compat18);
#else
-# warning "remove checksum compatibility support for b1_8"
__u32 lustre_msg_calc_cksum(struct lustre_msg *msg);
#endif
void lustre_msg_set_handle(struct lustre_msg *msg,struct lustre_handle *handle);
void lustre_msg_set_type(struct lustre_msg *msg, __u32 type);
void lustre_msg_set_opc(struct lustre_msg *msg, __u32 opc);
void lustre_msg_set_last_xid(struct lustre_msg *msg, __u64 last_xid);
+void lustre_msg_set_tag(struct lustre_msg *msg, __u16 tag);
void lustre_msg_set_last_committed(struct lustre_msg *msg,__u64 last_committed);
void lustre_msg_set_versions(struct lustre_msg *msg, __u64 *versions);
void lustre_msg_set_transno(struct lustre_msg *msg, __u64 transno);
void ptlrpc_request_set_replen(struct ptlrpc_request *req);
void lustre_msg_set_timeout(struct lustre_msg *msg, __u32 timeout);
void lustre_msg_set_service_time(struct lustre_msg *msg, __u32 service_time);
+void lustre_msg_set_jobid(struct lustre_msg *msg, char *jobid);
void lustre_msg_set_cksum(struct lustre_msg *msg, __u32 cksum);
static inline void
req->rq_replen = lustre_shrink_msg(req->rq_repmsg, segment,
newlen, move_data);
}
+
+#ifdef LUSTRE_TRANSLATE_ERRNOS
+
+static inline int ptlrpc_status_hton(int h)
+{
+ /*
+ * Positive errnos must be network errnos, such as LUSTRE_EDEADLK,
+ * ELDLM_LOCK_ABORTED, etc.
+ */
+ if (h < 0)
+ return -lustre_errno_hton(-h);
+ else
+ return h;
+}
+
+static inline int ptlrpc_status_ntoh(int n)
+{
+ /*
+ * See the comment in ptlrpc_status_hton().
+ */
+ if (n < 0)
+ return -lustre_errno_ntoh(-n);
+ else
+ return n;
+}
+
+#else
+
+#define ptlrpc_status_hton(h) (h)
+#define ptlrpc_status_ntoh(n) (n)
+
+#endif
/** @} */
/** Change request phase of \a req to \a new_phase */
static inline void
ptlrpc_rqphase_move(struct ptlrpc_request *req, enum rq_phase new_phase)
{
- if (req->rq_phase == new_phase)
- return;
+ if (req->rq_phase == new_phase)
+ return;
- if (new_phase == RQ_PHASE_UNREGISTERING) {
- req->rq_next_phase = req->rq_phase;
- if (req->rq_import)
- cfs_atomic_inc(&req->rq_import->imp_unregistering);
- }
+ if (new_phase == RQ_PHASE_UNREGISTERING) {
+ req->rq_next_phase = req->rq_phase;
+ if (req->rq_import)
+ atomic_inc(&req->rq_import->imp_unregistering);
+ }
- if (req->rq_phase == RQ_PHASE_UNREGISTERING) {
- if (req->rq_import)
- cfs_atomic_dec(&req->rq_import->imp_unregistering);
- }
+ if (req->rq_phase == RQ_PHASE_UNREGISTERING) {
+ if (req->rq_import)
+ atomic_dec(&req->rq_import->imp_unregistering);
+ }
- DEBUG_REQ(D_INFO, req, "move req \"%s\" -> \"%s\"",
- ptlrpc_rqphase2str(req), ptlrpc_phase2str(new_phase));
+ DEBUG_REQ(D_INFO, req, "move req \"%s\" -> \"%s\"",
+ ptlrpc_rqphase2str(req), ptlrpc_phase2str(new_phase));
- req->rq_phase = new_phase;
+ req->rq_phase = new_phase;
}
/**
- * Returns true if request \a req got early reply and hard deadline is not met
+ * Returns true if request \a req got early reply and hard deadline is not met
*/
static inline int
ptlrpc_client_early(struct ptlrpc_request *req)
static inline int
ptlrpc_client_recv_or_unlink(struct ptlrpc_request *req)
{
- int rc;
-
- cfs_spin_lock(&req->rq_lock);
- if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
- req->rq_reply_deadline > cfs_time_current_sec()) {
- cfs_spin_unlock(&req->rq_lock);
- return 1;
- }
- rc = req->rq_receiving_reply || req->rq_must_unlink;
- cfs_spin_unlock(&req->rq_lock);
- return rc;
+ int rc;
+
+ spin_lock(&req->rq_lock);
+ if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
+ req->rq_reply_deadline > cfs_time_current_sec()) {
+ spin_unlock(&req->rq_lock);
+ return 1;
+ }
+ rc = !req->rq_req_unlinked || !req->rq_reply_unlinked ||
+ req->rq_receiving_reply;
+ spin_unlock(&req->rq_lock);
+ return rc;
}
static inline void
ptlrpc_client_wake_req(struct ptlrpc_request *req)
{
- if (req->rq_set == NULL)
- cfs_waitq_signal(&req->rq_reply_waitq);
- else
- cfs_waitq_signal(&req->rq_set->set_waitq);
+ if (req->rq_set == NULL)
+ wake_up(&req->rq_reply_waitq);
+ else
+ wake_up(&req->rq_set->set_waitq);
}
static inline void
ptlrpc_rs_addref(struct ptlrpc_reply_state *rs)
{
- LASSERT(cfs_atomic_read(&rs->rs_refcount) > 0);
- cfs_atomic_inc(&rs->rs_refcount);
+ LASSERT(atomic_read(&rs->rs_refcount) > 0);
+ atomic_inc(&rs->rs_refcount);
}
static inline void
ptlrpc_rs_decref(struct ptlrpc_reply_state *rs)
{
- LASSERT(cfs_atomic_read(&rs->rs_refcount) > 0);
- if (cfs_atomic_dec_and_test(&rs->rs_refcount))
- lustre_free_reply_state(rs);
+ LASSERT(atomic_read(&rs->rs_refcount) > 0);
+ if (atomic_dec_and_test(&rs->rs_refcount))
+ lustre_free_reply_state(rs);
}
/* Should only be called once per req */
static inline int ptlrpc_no_resend(struct ptlrpc_request *req)
{
- if (!req->rq_no_resend && ptlrpc_send_limit_expired(req)) {
- cfs_spin_lock(&req->rq_lock);
- req->rq_no_resend = 1;
- cfs_spin_unlock(&req->rq_lock);
- }
- return req->rq_no_resend;
+ if (!req->rq_no_resend && ptlrpc_send_limit_expired(req)) {
+ spin_lock(&req->rq_lock);
+ req->rq_no_resend = 1;
+ spin_unlock(&req->rq_lock);
+ }
+ return req->rq_no_resend;
+}
+
+static inline int
+ptlrpc_server_get_timeout(struct ptlrpc_service_part *svcpt)
+{
+ int at = AT_OFF ? 0 : at_get(&svcpt->scp_at_estimate);
+
+ return svcpt->scp_service->srv_watchdog_factor *
+ max_t(int, at, obd_timeout);
+}
+
+static inline struct ptlrpc_service *
+ptlrpc_req2svc(struct ptlrpc_request *req)
+{
+ LASSERT(req->rq_rqbd != NULL);
+ return req->rq_rqbd->rqbd_svcpt->scp_service;
}
/* ldlm/ldlm_lib.c */
void client_destroy_import(struct obd_import *imp);
/** @} */
+#ifdef HAVE_SERVER_SUPPORT
int server_disconnect_export(struct obd_export *exp);
+#endif
/* ptlrpc/pinger.c */
/**
int ptlrpc_pinger_add_import(struct obd_import *imp);
int ptlrpc_pinger_del_import(struct obd_import *imp);
int ptlrpc_add_timeout_client(int time, enum timeout_event event,
- timeout_cb_t cb, void *data,
- cfs_list_t *obd_list);
-int ptlrpc_del_timeout_client(cfs_list_t *obd_list,
+ timeout_cb_t cb, void *data,
+ struct list_head *obd_list);
+int ptlrpc_del_timeout_client(struct list_head *obd_list,
enum timeout_event event);
struct ptlrpc_request * ptlrpc_prep_ping(struct obd_import *imp);
int ptlrpc_obd_ping(struct obd_device *obd);
-cfs_time_t ptlrpc_suspend_wakeup_time(void);
-#ifdef __KERNEL__
void ping_evictor_start(void);
void ping_evictor_stop(void);
-#else
-#define ping_evictor_start() do {} while (0)
-#define ping_evictor_stop() do {} while (0)
-#endif
-int ptlrpc_check_and_wait_suspend(struct ptlrpc_request *req);
+void ptlrpc_pinger_ir_up(void);
+void ptlrpc_pinger_ir_down(void);
/** @} */
+int ptlrpc_pinger_suppress_pings(void);
/* ptlrpc daemon bind policy */
typedef enum {
/* <free1 bound1> <free2 bound2> ... <freeN boundN> */
PDB_POLICY_PAIR = 3,
/* <free1 bound1> <bound1 free2> ... <freeN boundN> <boundN free1>,
- * means each ptlrpcd[X] has two partners: thread[X-1] and thread[X+1]*/
+ * means each ptlrpcd[X] has two partners: thread[X-1] and thread[X+1].
+ * If kernel supports NUMA, pthrpcd threads are binded and
+ * grouped by NUMA node */
PDB_POLICY_NEIGHBOR = 4,
} pdb_policy_t;
/* ptlrpc/ptlrpcd.c */
void ptlrpcd_stop(struct ptlrpcd_ctl *pc, int force);
+void ptlrpcd_free(struct ptlrpcd_ctl *pc);
void ptlrpcd_wake(struct ptlrpc_request *req);
void ptlrpcd_add_req(struct ptlrpc_request *req, pdl_policy_t policy, int idx);
void ptlrpcd_add_rqset(struct ptlrpc_request_set *set);
* @{
*/
const char* ll_opcode2str(__u32 opcode);
-#ifdef LPROCFS
+#ifdef CONFIG_PROC_FS
void ptlrpc_lprocfs_register_obd(struct obd_device *obd);
void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd);
void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes);
/** @} */
/* ptlrpc/llog_server.c */
-int llog_origin_handle_create(struct ptlrpc_request *req);
+int llog_origin_handle_open(struct ptlrpc_request *req);
int llog_origin_handle_destroy(struct ptlrpc_request *req);
int llog_origin_handle_prev_block(struct ptlrpc_request *req);
int llog_origin_handle_next_block(struct ptlrpc_request *req);
int llog_origin_handle_read_header(struct ptlrpc_request *req);
int llog_origin_handle_close(struct ptlrpc_request *req);
-int llog_origin_handle_cancel(struct ptlrpc_request *req);
-int llog_catinfo(struct ptlrpc_request *req);
/* ptlrpc/llog_client.c */
extern struct llog_operations llog_client_ops;
git://git.whamcloud.com / fs / lustre-release.git / blobdiff