linux/net/sunrpc/xprtrdma/xprt_rdma.h
<<
>>
Prefs
   1/*
   2 * Copyright (c) 2003-2007 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
  40#ifndef _LINUX_SUNRPC_XPRT_RDMA_H
  41#define _LINUX_SUNRPC_XPRT_RDMA_H
  42
  43#include <linux/wait.h>                 /* wait_queue_head_t, etc */
  44#include <linux/spinlock.h>             /* spinlock_t, etc */
  45#include <linux/atomic.h>                       /* atomic_t, etc */
  46#include <linux/workqueue.h>            /* struct work_struct */
  47
  48#include <rdma/rdma_cm.h>               /* RDMA connection api */
  49#include <rdma/ib_verbs.h>              /* RDMA verbs api */
  50
  51#include <linux/sunrpc/clnt.h>          /* rpc_xprt */
  52#include <linux/sunrpc/rpc_rdma.h>      /* RPC/RDMA protocol */
  53#include <linux/sunrpc/xprtrdma.h>      /* xprt parameters */
  54
  55#define RDMA_RESOLVE_TIMEOUT    (5000)  /* 5 seconds */
  56#define RDMA_CONNECT_RETRY_MAX  (2)     /* retries if no listener backlog */
  57
  58#define RPCRDMA_BIND_TO         (60U * HZ)
  59#define RPCRDMA_INIT_REEST_TO   (5U * HZ)
  60#define RPCRDMA_MAX_REEST_TO    (30U * HZ)
  61#define RPCRDMA_IDLE_DISC_TO    (5U * 60 * HZ)
  62
  63/*
  64 * Interface Adapter -- one per transport instance
  65 */
  66struct rpcrdma_ia {
  67        const struct rpcrdma_memreg_ops *ri_ops;
  68        struct ib_device        *ri_device;
  69        struct rdma_cm_id       *ri_id;
  70        struct ib_pd            *ri_pd;
  71        struct completion       ri_done;
  72        struct completion       ri_remove_done;
  73        int                     ri_async_rc;
  74        unsigned int            ri_max_segs;
  75        unsigned int            ri_max_frmr_depth;
  76        unsigned int            ri_max_inline_write;
  77        unsigned int            ri_max_inline_read;
  78        unsigned int            ri_max_send_sges;
  79        bool                    ri_reminv_expected;
  80        bool                    ri_implicit_roundup;
  81        enum ib_mr_type         ri_mrtype;
  82        unsigned long           ri_flags;
  83        struct ib_qp_attr       ri_qp_attr;
  84        struct ib_qp_init_attr  ri_qp_init_attr;
  85};
  86
  87enum {
  88        RPCRDMA_IAF_REMOVING = 0,
  89};
  90
  91/*
  92 * RDMA Endpoint -- one per transport instance
  93 */
  94
  95struct rpcrdma_ep {
  96        atomic_t                rep_cqcount;
  97        int                     rep_cqinit;
  98        int                     rep_connected;
  99        struct ib_qp_init_attr  rep_attr;
 100        wait_queue_head_t       rep_connect_wait;
 101        struct rpcrdma_connect_private  rep_cm_private;
 102        struct rdma_conn_param  rep_remote_cma;
 103        struct sockaddr_storage rep_remote_addr;
 104        struct delayed_work     rep_connect_worker;
 105};
 106
 107static inline void
 108rpcrdma_init_cqcount(struct rpcrdma_ep *ep, int count)
 109{
 110        atomic_set(&ep->rep_cqcount, ep->rep_cqinit - count);
 111}
 112
 113/* To update send queue accounting, provider must take a
 114 * send completion every now and then.
 115 */
 116static inline void
 117rpcrdma_set_signaled(struct rpcrdma_ep *ep, struct ib_send_wr *send_wr)
 118{
 119        send_wr->send_flags = 0;
 120        if (unlikely(atomic_sub_return(1, &ep->rep_cqcount) <= 0)) {
 121                rpcrdma_init_cqcount(ep, 0);
 122                send_wr->send_flags = IB_SEND_SIGNALED;
 123        }
 124}
 125
 126/* Pre-allocate extra Work Requests for handling backward receives
 127 * and sends. This is a fixed value because the Work Queues are
 128 * allocated when the forward channel is set up.
 129 */
 130#if defined(CONFIG_SUNRPC_BACKCHANNEL)
 131#define RPCRDMA_BACKWARD_WRS            (8)
 132#else
 133#define RPCRDMA_BACKWARD_WRS            (0)
 134#endif
 135
 136/* Registered buffer -- registered kmalloc'd memory for RDMA SEND/RECV
 137 *
 138 * The below structure appears at the front of a large region of kmalloc'd
 139 * memory, which always starts on a good alignment boundary.
 140 */
 141
 142struct rpcrdma_regbuf {
 143        struct ib_sge           rg_iov;
 144        struct ib_device        *rg_device;
 145        enum dma_data_direction rg_direction;
 146        __be32                  rg_base[0] __attribute__ ((aligned(256)));
 147};
 148
 149static inline u64
 150rdmab_addr(struct rpcrdma_regbuf *rb)
 151{
 152        return rb->rg_iov.addr;
 153}
 154
 155static inline u32
 156rdmab_length(struct rpcrdma_regbuf *rb)
 157{
 158        return rb->rg_iov.length;
 159}
 160
 161static inline u32
 162rdmab_lkey(struct rpcrdma_regbuf *rb)
 163{
 164        return rb->rg_iov.lkey;
 165}
 166
 167static inline struct rpcrdma_msg *
 168rdmab_to_msg(struct rpcrdma_regbuf *rb)
 169{
 170        return (struct rpcrdma_msg *)rb->rg_base;
 171}
 172
 173static inline struct ib_device *
 174rdmab_device(struct rpcrdma_regbuf *rb)
 175{
 176        return rb->rg_device;
 177}
 178
 179#define RPCRDMA_DEF_GFP         (GFP_NOIO | __GFP_NOWARN)
 180
 181/* To ensure a transport can always make forward progress,
 182 * the number of RDMA segments allowed in header chunk lists
 183 * is capped at 8. This prevents less-capable devices and
 184 * memory registrations from overrunning the Send buffer
 185 * while building chunk lists.
 186 *
 187 * Elements of the Read list take up more room than the
 188 * Write list or Reply chunk. 8 read segments means the Read
 189 * list (or Write list or Reply chunk) cannot consume more
 190 * than
 191 *
 192 * ((8 + 2) * read segment size) + 1 XDR words, or 244 bytes.
 193 *
 194 * And the fixed part of the header is another 24 bytes.
 195 *
 196 * The smallest inline threshold is 1024 bytes, ensuring that
 197 * at least 750 bytes are available for RPC messages.
 198 */
 199enum {
 200        RPCRDMA_MAX_HDR_SEGS = 8,
 201        RPCRDMA_HDRBUF_SIZE = 256,
 202};
 203
 204/*
 205 * struct rpcrdma_rep -- this structure encapsulates state required to recv
 206 * and complete a reply, asychronously. It needs several pieces of
 207 * state:
 208 *   o recv buffer (posted to provider)
 209 *   o ib_sge (also donated to provider)
 210 *   o status of reply (length, success or not)
 211 *   o bookkeeping state to get run by reply handler (list, etc)
 212 *
 213 * These are allocated during initialization, per-transport instance.
 214 *
 215 * N of these are associated with a transport instance, and stored in
 216 * struct rpcrdma_buffer. N is the max number of outstanding requests.
 217 */
 218
 219struct rpcrdma_rep {
 220        struct ib_cqe           rr_cqe;
 221        int                     rr_wc_flags;
 222        u32                     rr_inv_rkey;
 223        struct rpcrdma_regbuf   *rr_rdmabuf;
 224        struct rpcrdma_xprt     *rr_rxprt;
 225        struct work_struct      rr_work;
 226        struct xdr_buf          rr_hdrbuf;
 227        struct xdr_stream       rr_stream;
 228        struct list_head        rr_list;
 229        struct ib_recv_wr       rr_recv_wr;
 230};
 231
 232/*
 233 * struct rpcrdma_mw - external memory region metadata
 234 *
 235 * An external memory region is any buffer or page that is registered
 236 * on the fly (ie, not pre-registered).
 237 *
 238 * Each rpcrdma_buffer has a list of free MWs anchored in rb_mws. During
 239 * call_allocate, rpcrdma_buffer_get() assigns one to each segment in
 240 * an rpcrdma_req. Then rpcrdma_register_external() grabs these to keep
 241 * track of registration metadata while each RPC is pending.
 242 * rpcrdma_deregister_external() uses this metadata to unmap and
 243 * release these resources when an RPC is complete.
 244 */
 245enum rpcrdma_frmr_state {
 246        FRMR_IS_INVALID,        /* ready to be used */
 247        FRMR_IS_VALID,          /* in use */
 248        FRMR_FLUSHED_FR,        /* flushed FASTREG WR */
 249        FRMR_FLUSHED_LI,        /* flushed LOCALINV WR */
 250};
 251
 252struct rpcrdma_frmr {
 253        struct ib_mr                    *fr_mr;
 254        struct ib_cqe                   fr_cqe;
 255        enum rpcrdma_frmr_state         fr_state;
 256        struct completion               fr_linv_done;
 257        union {
 258                struct ib_reg_wr        fr_regwr;
 259                struct ib_send_wr       fr_invwr;
 260        };
 261};
 262
 263struct rpcrdma_fmr {
 264        struct ib_fmr           *fm_mr;
 265        u64                     *fm_physaddrs;
 266};
 267
 268struct rpcrdma_mw {
 269        struct list_head        mw_list;
 270        struct scatterlist      *mw_sg;
 271        int                     mw_nents;
 272        enum dma_data_direction mw_dir;
 273        unsigned long           mw_flags;
 274        union {
 275                struct rpcrdma_fmr      fmr;
 276                struct rpcrdma_frmr     frmr;
 277        };
 278        struct rpcrdma_xprt     *mw_xprt;
 279        u32                     mw_handle;
 280        u32                     mw_length;
 281        u64                     mw_offset;
 282        struct list_head        mw_all;
 283};
 284
 285/* mw_flags */
 286enum {
 287        RPCRDMA_MW_F_RI         = 1,
 288};
 289
 290/*
 291 * struct rpcrdma_req -- structure central to the request/reply sequence.
 292 *
 293 * N of these are associated with a transport instance, and stored in
 294 * struct rpcrdma_buffer. N is the max number of outstanding requests.
 295 *
 296 * It includes pre-registered buffer memory for send AND recv.
 297 * The recv buffer, however, is not owned by this structure, and
 298 * is "donated" to the hardware when a recv is posted. When a
 299 * reply is handled, the recv buffer used is given back to the
 300 * struct rpcrdma_req associated with the request.
 301 *
 302 * In addition to the basic memory, this structure includes an array
 303 * of iovs for send operations. The reason is that the iovs passed to
 304 * ib_post_{send,recv} must not be modified until the work request
 305 * completes.
 306 */
 307
 308/* Maximum number of page-sized "segments" per chunk list to be
 309 * registered or invalidated. Must handle a Reply chunk:
 310 */
 311enum {
 312        RPCRDMA_MAX_IOV_SEGS    = 3,
 313        RPCRDMA_MAX_DATA_SEGS   = ((1 * 1024 * 1024) / PAGE_SIZE) + 1,
 314        RPCRDMA_MAX_SEGS        = RPCRDMA_MAX_DATA_SEGS +
 315                                  RPCRDMA_MAX_IOV_SEGS,
 316};
 317
 318struct rpcrdma_mr_seg {         /* chunk descriptors */
 319        u32             mr_len;         /* length of chunk or segment */
 320        struct page     *mr_page;       /* owning page, if any */
 321        char            *mr_offset;     /* kva if no page, else offset */
 322};
 323
 324/* The Send SGE array is provisioned to send a maximum size
 325 * inline request:
 326 * - RPC-over-RDMA header
 327 * - xdr_buf head iovec
 328 * - RPCRDMA_MAX_INLINE bytes, in pages
 329 * - xdr_buf tail iovec
 330 *
 331 * The actual number of array elements consumed by each RPC
 332 * depends on the device's max_sge limit.
 333 */
 334enum {
 335        RPCRDMA_MIN_SEND_SGES = 3,
 336        RPCRDMA_MAX_PAGE_SGES = RPCRDMA_MAX_INLINE >> PAGE_SHIFT,
 337        RPCRDMA_MAX_SEND_SGES = 1 + 1 + RPCRDMA_MAX_PAGE_SGES + 1,
 338};
 339
 340struct rpcrdma_buffer;
 341struct rpcrdma_req {
 342        struct list_head        rl_list;
 343        unsigned int            rl_mapped_sges;
 344        unsigned int            rl_connect_cookie;
 345        struct rpcrdma_buffer   *rl_buffer;
 346        struct rpcrdma_rep      *rl_reply;
 347        struct xdr_stream       rl_stream;
 348        struct xdr_buf          rl_hdrbuf;
 349        struct ib_send_wr       rl_send_wr;
 350        struct ib_sge           rl_send_sge[RPCRDMA_MAX_SEND_SGES];
 351        struct rpcrdma_regbuf   *rl_rdmabuf;    /* xprt header */
 352        struct rpcrdma_regbuf   *rl_sendbuf;    /* rq_snd_buf */
 353        struct rpcrdma_regbuf   *rl_recvbuf;    /* rq_rcv_buf */
 354
 355        struct ib_cqe           rl_cqe;
 356        struct list_head        rl_all;
 357        bool                    rl_backchannel;
 358
 359        struct list_head        rl_registered;  /* registered segments */
 360        struct rpcrdma_mr_seg   rl_segments[RPCRDMA_MAX_SEGS];
 361};
 362
 363static inline void
 364rpcrdma_set_xprtdata(struct rpc_rqst *rqst, struct rpcrdma_req *req)
 365{
 366        rqst->rq_xprtdata = req;
 367}
 368
 369static inline struct rpcrdma_req *
 370rpcr_to_rdmar(struct rpc_rqst *rqst)
 371{
 372        return rqst->rq_xprtdata;
 373}
 374
 375static inline void
 376rpcrdma_push_mw(struct rpcrdma_mw *mw, struct list_head *list)
 377{
 378        list_add_tail(&mw->mw_list, list);
 379}
 380
 381static inline struct rpcrdma_mw *
 382rpcrdma_pop_mw(struct list_head *list)
 383{
 384        struct rpcrdma_mw *mw;
 385
 386        mw = list_first_entry(list, struct rpcrdma_mw, mw_list);
 387        list_del(&mw->mw_list);
 388        return mw;
 389}
 390
 391/*
 392 * struct rpcrdma_buffer -- holds list/queue of pre-registered memory for
 393 * inline requests/replies, and client/server credits.
 394 *
 395 * One of these is associated with a transport instance
 396 */
 397struct rpcrdma_buffer {
 398        spinlock_t              rb_mwlock;      /* protect rb_mws list */
 399        struct list_head        rb_mws;
 400        struct list_head        rb_all;
 401
 402        spinlock_t              rb_lock;        /* protect buf lists */
 403        int                     rb_send_count, rb_recv_count;
 404        struct list_head        rb_send_bufs;
 405        struct list_head        rb_recv_bufs;
 406        u32                     rb_max_requests;
 407        atomic_t                rb_credits;     /* most recent credit grant */
 408
 409        u32                     rb_bc_srv_max_requests;
 410        spinlock_t              rb_reqslock;    /* protect rb_allreqs */
 411        struct list_head        rb_allreqs;
 412
 413        u32                     rb_bc_max_requests;
 414
 415        spinlock_t              rb_recovery_lock; /* protect rb_stale_mrs */
 416        struct list_head        rb_stale_mrs;
 417        struct delayed_work     rb_recovery_worker;
 418        struct delayed_work     rb_refresh_worker;
 419};
 420#define rdmab_to_ia(b) (&container_of((b), struct rpcrdma_xprt, rx_buf)->rx_ia)
 421
 422/*
 423 * Internal structure for transport instance creation. This
 424 * exists primarily for modularity.
 425 *
 426 * This data should be set with mount options
 427 */
 428struct rpcrdma_create_data_internal {
 429        struct sockaddr_storage addr;   /* RDMA server address */
 430        unsigned int    max_requests;   /* max requests (slots) in flight */
 431        unsigned int    rsize;          /* mount rsize - max read hdr+data */
 432        unsigned int    wsize;          /* mount wsize - max write hdr+data */
 433        unsigned int    inline_rsize;   /* max non-rdma read data payload */
 434        unsigned int    inline_wsize;   /* max non-rdma write data payload */
 435        unsigned int    padding;        /* non-rdma write header padding */
 436};
 437
 438/*
 439 * Statistics for RPCRDMA
 440 */
 441struct rpcrdma_stats {
 442        /* accessed when sending a call */
 443        unsigned long           read_chunk_count;
 444        unsigned long           write_chunk_count;
 445        unsigned long           reply_chunk_count;
 446        unsigned long long      total_rdma_request;
 447
 448        /* rarely accessed error counters */
 449        unsigned long long      pullup_copy_count;
 450        unsigned long           hardway_register_count;
 451        unsigned long           failed_marshal_count;
 452        unsigned long           bad_reply_count;
 453        unsigned long           mrs_recovered;
 454        unsigned long           mrs_orphaned;
 455        unsigned long           mrs_allocated;
 456
 457        /* accessed when receiving a reply */
 458        unsigned long long      total_rdma_reply;
 459        unsigned long long      fixup_copy_count;
 460        unsigned long           local_inv_needed;
 461        unsigned long           nomsg_call_count;
 462        unsigned long           bcall_count;
 463};
 464
 465/*
 466 * Per-registration mode operations
 467 */
 468struct rpcrdma_xprt;
 469struct rpcrdma_memreg_ops {
 470        struct rpcrdma_mr_seg *
 471                        (*ro_map)(struct rpcrdma_xprt *,
 472                                  struct rpcrdma_mr_seg *, int, bool,
 473                                  struct rpcrdma_mw **);
 474        void            (*ro_unmap_sync)(struct rpcrdma_xprt *,
 475                                         struct list_head *);
 476        void            (*ro_unmap_safe)(struct rpcrdma_xprt *,
 477                                         struct rpcrdma_req *, bool);
 478        void            (*ro_recover_mr)(struct rpcrdma_mw *);
 479        int             (*ro_open)(struct rpcrdma_ia *,
 480                                   struct rpcrdma_ep *,
 481                                   struct rpcrdma_create_data_internal *);
 482        size_t          (*ro_maxpages)(struct rpcrdma_xprt *);
 483        int             (*ro_init_mr)(struct rpcrdma_ia *,
 484                                      struct rpcrdma_mw *);
 485        void            (*ro_release_mr)(struct rpcrdma_mw *);
 486        const char      *ro_displayname;
 487        const int       ro_send_w_inv_ok;
 488};
 489
 490extern const struct rpcrdma_memreg_ops rpcrdma_fmr_memreg_ops;
 491extern const struct rpcrdma_memreg_ops rpcrdma_frwr_memreg_ops;
 492
 493/*
 494 * RPCRDMA transport -- encapsulates the structures above for
 495 * integration with RPC.
 496 *
 497 * The contained structures are embedded, not pointers,
 498 * for convenience. This structure need not be visible externally.
 499 *
 500 * It is allocated and initialized during mount, and released
 501 * during unmount.
 502 */
 503struct rpcrdma_xprt {
 504        struct rpc_xprt         rx_xprt;
 505        struct rpcrdma_ia       rx_ia;
 506        struct rpcrdma_ep       rx_ep;
 507        struct rpcrdma_buffer   rx_buf;
 508        struct rpcrdma_create_data_internal rx_data;
 509        struct delayed_work     rx_connect_worker;
 510        struct rpcrdma_stats    rx_stats;
 511};
 512
 513#define rpcx_to_rdmax(x) container_of(x, struct rpcrdma_xprt, rx_xprt)
 514#define rpcx_to_rdmad(x) (rpcx_to_rdmax(x)->rx_data)
 515
 516/* Setting this to 0 ensures interoperability with early servers.
 517 * Setting this to 1 enhances certain unaligned read/write performance.
 518 * Default is 0, see sysctl entry and rpc_rdma.c rpcrdma_convert_iovs() */
 519extern int xprt_rdma_pad_optimize;
 520
 521/* This setting controls the hunt for a supported memory
 522 * registration strategy.
 523 */
 524extern unsigned int xprt_rdma_memreg_strategy;
 525
 526/*
 527 * Interface Adapter calls - xprtrdma/verbs.c
 528 */
 529int rpcrdma_ia_open(struct rpcrdma_xprt *xprt, struct sockaddr *addr);
 530void rpcrdma_ia_remove(struct rpcrdma_ia *ia);
 531void rpcrdma_ia_close(struct rpcrdma_ia *);
 532bool frwr_is_supported(struct rpcrdma_ia *);
 533bool fmr_is_supported(struct rpcrdma_ia *);
 534
 535/*
 536 * Endpoint calls - xprtrdma/verbs.c
 537 */
 538int rpcrdma_ep_create(struct rpcrdma_ep *, struct rpcrdma_ia *,
 539                                struct rpcrdma_create_data_internal *);
 540void rpcrdma_ep_destroy(struct rpcrdma_ep *, struct rpcrdma_ia *);
 541int rpcrdma_ep_connect(struct rpcrdma_ep *, struct rpcrdma_ia *);
 542void rpcrdma_conn_func(struct rpcrdma_ep *ep);
 543void rpcrdma_ep_disconnect(struct rpcrdma_ep *, struct rpcrdma_ia *);
 544
 545int rpcrdma_ep_post(struct rpcrdma_ia *, struct rpcrdma_ep *,
 546                                struct rpcrdma_req *);
 547int rpcrdma_ep_post_recv(struct rpcrdma_ia *, struct rpcrdma_rep *);
 548
 549/*
 550 * Buffer calls - xprtrdma/verbs.c
 551 */
 552struct rpcrdma_req *rpcrdma_create_req(struct rpcrdma_xprt *);
 553struct rpcrdma_rep *rpcrdma_create_rep(struct rpcrdma_xprt *);
 554void rpcrdma_destroy_req(struct rpcrdma_req *);
 555int rpcrdma_buffer_create(struct rpcrdma_xprt *);
 556void rpcrdma_buffer_destroy(struct rpcrdma_buffer *);
 557
 558struct rpcrdma_mw *rpcrdma_get_mw(struct rpcrdma_xprt *);
 559void rpcrdma_put_mw(struct rpcrdma_xprt *, struct rpcrdma_mw *);
 560struct rpcrdma_req *rpcrdma_buffer_get(struct rpcrdma_buffer *);
 561void rpcrdma_buffer_put(struct rpcrdma_req *);
 562void rpcrdma_recv_buffer_get(struct rpcrdma_req *);
 563void rpcrdma_recv_buffer_put(struct rpcrdma_rep *);
 564
 565void rpcrdma_defer_mr_recovery(struct rpcrdma_mw *);
 566
 567struct rpcrdma_regbuf *rpcrdma_alloc_regbuf(size_t, enum dma_data_direction,
 568                                            gfp_t);
 569bool __rpcrdma_dma_map_regbuf(struct rpcrdma_ia *, struct rpcrdma_regbuf *);
 570void rpcrdma_free_regbuf(struct rpcrdma_regbuf *);
 571
 572static inline bool
 573rpcrdma_regbuf_is_mapped(struct rpcrdma_regbuf *rb)
 574{
 575        return rb->rg_device != NULL;
 576}
 577
 578static inline bool
 579rpcrdma_dma_map_regbuf(struct rpcrdma_ia *ia, struct rpcrdma_regbuf *rb)
 580{
 581        if (likely(rpcrdma_regbuf_is_mapped(rb)))
 582                return true;
 583        return __rpcrdma_dma_map_regbuf(ia, rb);
 584}
 585
 586int rpcrdma_ep_post_extra_recv(struct rpcrdma_xprt *, unsigned int);
 587
 588int rpcrdma_alloc_wq(void);
 589void rpcrdma_destroy_wq(void);
 590
 591/*
 592 * Wrappers for chunk registration, shared by read/write chunk code.
 593 */
 594
 595static inline enum dma_data_direction
 596rpcrdma_data_dir(bool writing)
 597{
 598        return writing ? DMA_FROM_DEVICE : DMA_TO_DEVICE;
 599}
 600
 601/*
 602 * RPC/RDMA protocol calls - xprtrdma/rpc_rdma.c
 603 */
 604
 605enum rpcrdma_chunktype {
 606        rpcrdma_noch = 0,
 607        rpcrdma_readch,
 608        rpcrdma_areadch,
 609        rpcrdma_writech,
 610        rpcrdma_replych
 611};
 612
 613bool rpcrdma_prepare_send_sges(struct rpcrdma_ia *, struct rpcrdma_req *,
 614                               u32, struct xdr_buf *, enum rpcrdma_chunktype);
 615void rpcrdma_unmap_sges(struct rpcrdma_ia *, struct rpcrdma_req *);
 616int rpcrdma_marshal_req(struct rpcrdma_xprt *r_xprt, struct rpc_rqst *rqst);
 617void rpcrdma_set_max_header_sizes(struct rpcrdma_xprt *);
 618void rpcrdma_reply_handler(struct work_struct *work);
 619
 620static inline void rpcrdma_set_xdrlen(struct xdr_buf *xdr, size_t len)
 621{
 622        xdr->head[0].iov_len = len;
 623        xdr->len = len;
 624}
 625
 626/* RPC/RDMA module init - xprtrdma/transport.c
 627 */
 628extern unsigned int xprt_rdma_max_inline_read;
 629void xprt_rdma_format_addresses(struct rpc_xprt *xprt, struct sockaddr *sap);
 630void xprt_rdma_free_addresses(struct rpc_xprt *xprt);
 631void rpcrdma_connect_worker(struct work_struct *work);
 632void xprt_rdma_print_stats(struct rpc_xprt *xprt, struct seq_file *seq);
 633int xprt_rdma_init(void);
 634void xprt_rdma_cleanup(void);
 635
 636/* Backchannel calls - xprtrdma/backchannel.c
 637 */
 638#if defined(CONFIG_SUNRPC_BACKCHANNEL)
 639int xprt_rdma_bc_setup(struct rpc_xprt *, unsigned int);
 640int xprt_rdma_bc_up(struct svc_serv *, struct net *);
 641size_t xprt_rdma_bc_maxpayload(struct rpc_xprt *);
 642int rpcrdma_bc_post_recv(struct rpcrdma_xprt *, unsigned int);
 643void rpcrdma_bc_receive_call(struct rpcrdma_xprt *, struct rpcrdma_rep *);
 644int rpcrdma_bc_marshal_reply(struct rpc_rqst *);
 645void xprt_rdma_bc_free_rqst(struct rpc_rqst *);
 646void xprt_rdma_bc_destroy(struct rpc_xprt *, unsigned int);
 647#endif  /* CONFIG_SUNRPC_BACKCHANNEL */
 648
 649extern struct xprt_class xprt_rdma_bc;
 650
 651#endif                          /* _LINUX_SUNRPC_XPRT_RDMA_H */
 652