LXR linux/include/linux/sunrpc/svc

   1/*
   2 * Copyright (c) 2005-2006 Network Appliance, Inc. All rights reserved.
   3 *
   4 * This software is available to you under a choice of one of two
   5 * licenses.  You may choose to be licensed under the terms of the GNU
   6 * General Public License (GPL) Version 2, available from the file
   7 * COPYING in the main directory of this source tree, or the BSD-type
   8 * license below:
   9 *
  10 * Redistribution and use in source and binary forms, with or without
  11 * modification, are permitted provided that the following conditions
  12 * are met:
  13 *
  14 *      Redistributions of source code must retain the above copyright
  15 *      notice, this list of conditions and the following disclaimer.
  16 *
  17 *      Redistributions in binary form must reproduce the above
  18 *      copyright notice, this list of conditions and the following
  19 *      disclaimer in the documentation and/or other materials provided
  20 *      with the distribution.
  21 *
  22 *      Neither the name of the Network Appliance, Inc. nor the names of
  23 *      its contributors may be used to endorse or promote products
  24 *      derived from this software without specific prior written
  25 *      permission.
  26 *
  27 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  28 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  29 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  30 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  31 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  32 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  33 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  34 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  35 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  36 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  37 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  38 *
  39 * Author: Tom Tucker <tom@opengridcomputing.com>
  40 */
  41
  42#ifndef SVC_RDMA_H
  43#define SVC_RDMA_H
  44#include <linux/sunrpc/xdr.h>
  45#include <linux/sunrpc/svcsock.h>
  46#include <linux/sunrpc/rpc_rdma.h>
  47#include <rdma/ib_verbs.h>
  48#include <rdma/rdma_cm.h>
  49#define SVCRDMA_DEBUG
  50
  51/* RPC/RDMA parameters and stats */
  52extern unsigned int svcrdma_ord;
  53extern unsigned int svcrdma_max_requests;
  54extern unsigned int svcrdma_max_req_size;
  55
  56extern atomic_t rdma_stat_recv;
  57extern atomic_t rdma_stat_read;
  58extern atomic_t rdma_stat_write;
  59extern atomic_t rdma_stat_sq_starve;
  60extern atomic_t rdma_stat_rq_starve;
  61extern atomic_t rdma_stat_rq_poll;
  62extern atomic_t rdma_stat_rq_prod;
  63extern atomic_t rdma_stat_sq_poll;
  64extern atomic_t rdma_stat_sq_prod;
  65
  66#define RPCRDMA_VERSION 1
  67
  68/*
  69 * Contexts are built when an RDMA request is created and are a
  70 * record of the resources that can be recovered when the request
  71 * completes.
  72 */
  73struct svc_rdma_op_ctxt {
  74        struct svc_rdma_op_ctxt *read_hdr;
  75        int hdr_count;
  76        struct xdr_buf arg;
  77        struct list_head dto_q;
  78        enum ib_wr_opcode wr_op;
  79        enum ib_wc_status wc_status;
  80        u32 byte_len;
  81        struct svcxprt_rdma *xprt;
  82        unsigned long flags;
  83        enum dma_data_direction direction;
  84        int count;
  85        struct ib_sge sge[RPCSVC_MAXPAGES];
  86        struct page *pages[RPCSVC_MAXPAGES];
  87};
  88
  89/*
  90 * NFS_ requests are mapped on the client side by the chunk lists in
  91 * the RPCRDMA header. During the fetching of the RPC from the client
  92 * and the writing of the reply to the client, the memory in the
  93 * client and the memory in the server must be mapped as contiguous
  94 * vaddr/len for access by the hardware. These data strucures keep
  95 * these mappings.
  96 *
  97 * For an RDMA_WRITE, the 'sge' maps the RPC REPLY. For RDMA_READ, the
  98 * 'sge' in the svc_rdma_req_map maps the server side RPC reply and the
  99 * 'ch' field maps the read-list of the RPCRDMA header to the 'sge'
 100 * mapping of the reply.
 101 */
 102struct svc_rdma_chunk_sge {
 103        int start;              /* sge no for this chunk */
 104        int count;              /* sge count for this chunk */
 105};
 106struct svc_rdma_req_map {
 107        unsigned long count;
 108        union {
 109                struct kvec sge[RPCSVC_MAXPAGES];
 110                struct svc_rdma_chunk_sge ch[RPCSVC_MAXPAGES];
 111        };
 112};
 113
 114#define RDMACTXT_F_LAST_CTXT    2
 115
 116struct svcxprt_rdma {
 117        struct svc_xprt      sc_xprt;           /* SVC transport structure */
 118        struct rdma_cm_id    *sc_cm_id;         /* RDMA connection id */
 119        struct list_head     sc_accept_q;       /* Conn. waiting accept */
 120        int                  sc_ord;            /* RDMA read limit */
 121        int                  sc_max_sge;
 122
 123        int                  sc_sq_depth;       /* Depth of SQ */
 124        atomic_t             sc_sq_count;       /* Number of SQ WR on queue */
 125
 126        int                  sc_max_requests;   /* Depth of RQ */
 127        int                  sc_max_req_size;   /* Size of each RQ WR buf */
 128
 129        struct ib_pd         *sc_pd;
 130
 131        atomic_t             sc_dma_used;
 132        atomic_t             sc_ctxt_used;
 133        struct list_head     sc_rq_dto_q;
 134        spinlock_t           sc_rq_dto_lock;
 135        struct ib_qp         *sc_qp;
 136        struct ib_cq         *sc_rq_cq;
 137        struct ib_cq         *sc_sq_cq;
 138        struct ib_mr         *sc_phys_mr;       /* MR for server memory */
 139
 140        spinlock_t           sc_lock;           /* transport lock */
 141
 142        wait_queue_head_t    sc_send_wait;      /* SQ exhaustion waitlist */
 143        unsigned long        sc_flags;
 144        struct list_head     sc_dto_q;          /* DTO tasklet I/O pending Q */
 145        struct list_head     sc_read_complete_q;
 146        struct work_struct   sc_work;
 147};
 148/* sc_flags */
 149#define RDMAXPRT_RQ_PENDING     1
 150#define RDMAXPRT_SQ_PENDING     2
 151#define RDMAXPRT_CONN_PENDING   3
 152
 153#define RPCRDMA_LISTEN_BACKLOG  10
 154/* The default ORD value is based on two outstanding full-size writes with a
 155 * page size of 4k, or 32k * 2 ops / 4k = 16 outstanding RDMA_READ.  */
 156#define RPCRDMA_ORD             (64/4)
 157#define RPCRDMA_SQ_DEPTH_MULT   8
 158#define RPCRDMA_MAX_THREADS     16
 159#define RPCRDMA_MAX_REQUESTS    16
 160#define RPCRDMA_MAX_REQ_SIZE    4096
 161
 162/* svc_rdma_marshal.c */
 163extern void svc_rdma_rcl_chunk_counts(struct rpcrdma_read_chunk *,
 164                                      int *, int *);
 165extern int svc_rdma_xdr_decode_req(struct rpcrdma_msg **, struct svc_rqst *);
 166extern int svc_rdma_xdr_decode_deferred_req(struct svc_rqst *);
 167extern int svc_rdma_xdr_encode_error(struct svcxprt_rdma *,
 168                                     struct rpcrdma_msg *,
 169                                     enum rpcrdma_errcode, u32 *);
 170extern void svc_rdma_xdr_encode_write_list(struct rpcrdma_msg *, int);
 171extern void svc_rdma_xdr_encode_reply_array(struct rpcrdma_write_array *, int);
 172extern void svc_rdma_xdr_encode_array_chunk(struct rpcrdma_write_array *, int,
 173                                            u32, u64, u32);
 174extern void svc_rdma_xdr_encode_reply_header(struct svcxprt_rdma *,
 175                                             struct rpcrdma_msg *,
 176                                             struct rpcrdma_msg *,
 177                                             enum rpcrdma_proc);
 178extern int svc_rdma_xdr_get_reply_hdr_len(struct rpcrdma_msg *);
 179
 180/* svc_rdma_recvfrom.c */
 181extern int svc_rdma_recvfrom(struct svc_rqst *);
 182
 183/* svc_rdma_sendto.c */
 184extern int svc_rdma_sendto(struct svc_rqst *);
 185
 186/* svc_rdma_transport.c */
 187extern int svc_rdma_send(struct svcxprt_rdma *, struct ib_send_wr *);
 188extern void svc_rdma_send_error(struct svcxprt_rdma *, struct rpcrdma_msg *,
 189                                enum rpcrdma_errcode);
 190struct page *svc_rdma_get_page(void);
 191extern int svc_rdma_post_recv(struct svcxprt_rdma *);
 192extern int svc_rdma_create_listen(struct svc_serv *, int, struct sockaddr *);
 193extern struct svc_rdma_op_ctxt *svc_rdma_get_context(struct svcxprt_rdma *);
 194extern void svc_rdma_put_context(struct svc_rdma_op_ctxt *, int);
 195extern struct svc_rdma_req_map *svc_rdma_get_req_map(void);
 196extern void svc_rdma_put_req_map(struct svc_rdma_req_map *);
 197extern void svc_sq_reap(struct svcxprt_rdma *);
 198extern void svc_rq_reap(struct svcxprt_rdma *);
 199extern struct svc_xprt_class svc_rdma_class;
 200extern void svc_rdma_prep_reply_hdr(struct svc_rqst *);
 201
 202/* svc_rdma.c */
 203extern int svc_rdma_init(void);
 204extern void svc_rdma_cleanup(void);
 205
 206/*
 207 * Returns the address of the first read chunk or <nul> if no read chunk is
 208 * present
 209 */
 210static inline struct rpcrdma_read_chunk *
 211svc_rdma_get_read_chunk(struct rpcrdma_msg *rmsgp)
 212{
 213        struct rpcrdma_read_chunk *ch =
 214                (struct rpcrdma_read_chunk *)&rmsgp->rm_body.rm_chunks[0];
 215
 216        if (ch->rc_discrim == 0)
 217                return NULL;
 218
 219        return ch;
 220}
 221
 222/*
 223 * Returns the address of the first read write array element or <nul> if no
 224 * write array list is present
 225 */
 226static inline struct rpcrdma_write_array *
 227svc_rdma_get_write_array(struct rpcrdma_msg *rmsgp)
 228{
 229        if (rmsgp->rm_body.rm_chunks[0] != 0
 230            || rmsgp->rm_body.rm_chunks[1] == 0)
 231                return NULL;
 232
 233        return (struct rpcrdma_write_array *)&rmsgp->rm_body.rm_chunks[1];
 234}
 235
 236/*
 237 * Returns the address of the first reply array element or <nul> if no
 238 * reply array is present
 239 */
 240static inline struct rpcrdma_write_array *
 241svc_rdma_get_reply_array(struct rpcrdma_msg *rmsgp)
 242{
 243        struct rpcrdma_read_chunk *rch;
 244        struct rpcrdma_write_array *wr_ary;
 245        struct rpcrdma_write_array *rp_ary;
 246
 247        /* XXX: Need to fix when reply list may occur with read-list and/or
 248         * write list */
 249        if (rmsgp->rm_body.rm_chunks[0] != 0 ||
 250            rmsgp->rm_body.rm_chunks[1] != 0)
 251                return NULL;
 252
 253        rch = svc_rdma_get_read_chunk(rmsgp);
 254        if (rch) {
 255                while (rch->rc_discrim)
 256                        rch++;
 257
 258                /* The reply list follows an empty write array located
 259                 * at 'rc_position' here. The reply array is at rc_target.
 260                 */
 261                rp_ary = (struct rpcrdma_write_array *)&rch->rc_target;
 262
 263                goto found_it;
 264        }
 265
 266        wr_ary = svc_rdma_get_write_array(rmsgp);
 267        if (wr_ary) {
 268                rp_ary = (struct rpcrdma_write_array *)
 269                        &wr_ary->
 270                        wc_array[wr_ary->wc_nchunks].wc_target.rs_length;
 271
 272                goto found_it;
 273        }
 274
 275        /* No read list, no write list */
 276        rp_ary = (struct rpcrdma_write_array *)
 277                &rmsgp->rm_body.rm_chunks[2];
 278
 279 found_it:
 280        if (rp_ary->wc_discrim == 0)
 281                return NULL;
 282
 283        return rp_ary;
 284}
 285#endif
 286