LXR linux/drivers/infiniband/ulp/iser/iscsi

   1/*
   2 * iSER transport for the Open iSCSI Initiator & iSER transport internals
   3 *
   4 * Copyright (C) 2004 Dmitry Yusupov
   5 * Copyright (C) 2004 Alex Aizman
   6 * Copyright (C) 2005 Mike Christie
   7 * based on code maintained by open-iscsi@googlegroups.com
   8 *
   9 * Copyright (c) 2004, 2005, 2006 Voltaire, Inc. All rights reserved.
  10 * Copyright (c) 2005, 2006 Cisco Systems.  All rights reserved.
  11 *
  12 * This software is available to you under a choice of one of two
  13 * licenses.  You may choose to be licensed under the terms of the GNU
  14 * General Public License (GPL) Version 2, available from the file
  15 * COPYING in the main directory of this source tree, or the
  16 * OpenIB.org BSD license below:
  17 *
  18 *     Redistribution and use in source and binary forms, with or
  19 *     without modification, are permitted provided that the following
  20 *     conditions are met:
  21 *
  22 *      - Redistributions of source code must retain the above
  23 *        copyright notice, this list of conditions and the following
  24 *        disclaimer.
  25 *
  26 *      - Redistributions in binary form must reproduce the above
  27 *        copyright notice, this list of conditions and the following
  28 *        disclaimer in the documentation and/or other materials
  29 *        provided with the distribution.
  30 *
  31 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  32 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  33 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  34 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  35 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  36 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  37 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  38 * SOFTWARE.
  39 */
  40#ifndef __ISCSI_ISER_H__
  41#define __ISCSI_ISER_H__
  42
  43#include <linux/types.h>
  44#include <linux/net.h>
  45#include <scsi/libiscsi.h>
  46#include <scsi/scsi_transport_iscsi.h>
  47
  48#include <linux/interrupt.h>
  49#include <linux/wait.h>
  50#include <linux/sched.h>
  51#include <linux/list.h>
  52#include <linux/slab.h>
  53#include <linux/dma-mapping.h>
  54#include <linux/mutex.h>
  55#include <linux/mempool.h>
  56#include <linux/uio.h>
  57
  58#include <linux/socket.h>
  59#include <linux/in.h>
  60#include <linux/in6.h>
  61
  62#include <rdma/ib_verbs.h>
  63#include <rdma/ib_fmr_pool.h>
  64#include <rdma/rdma_cm.h>
  65
  66#define DRV_NAME        "iser"
  67#define PFX             DRV_NAME ": "
  68#define DRV_VER         "0.1"
  69#define DRV_DATE        "May 7th, 2006"
  70
  71#define iser_dbg(fmt, arg...)                           \
  72        do {                                            \
  73                if (iser_debug_level > 1)               \
  74                        printk(KERN_DEBUG PFX "%s:" fmt,\
  75                                __func__ , ## arg);     \
  76        } while (0)
  77
  78#define iser_warn(fmt, arg...)                          \
  79        do {                                            \
  80                if (iser_debug_level > 0)               \
  81                        printk(KERN_DEBUG PFX "%s:" fmt,\
  82                                __func__ , ## arg);     \
  83        } while (0)
  84
  85#define iser_err(fmt, arg...)                           \
  86        do {                                            \
  87                printk(KERN_ERR PFX "%s:" fmt,          \
  88                       __func__ , ## arg);              \
  89        } while (0)
  90
  91#define SHIFT_4K        12
  92#define SIZE_4K (1ULL << SHIFT_4K)
  93#define MASK_4K (~(SIZE_4K-1))
  94
  95                                        /* support up to 512KB in one RDMA */
  96#define ISCSI_ISER_SG_TABLESIZE         (0x80000 >> SHIFT_4K)
  97#define ISER_DEF_CMD_PER_LUN            128
  98
  99/* QP settings */
 100/* Maximal bounds on received asynchronous PDUs */
 101#define ISER_MAX_RX_MISC_PDUS           4 /* NOOP_IN(2) , ASYNC_EVENT(2)   */
 102
 103#define ISER_MAX_TX_MISC_PDUS           6 /* NOOP_OUT(2), TEXT(1),         *
 104                                           * SCSI_TMFUNC(2), LOGOUT(1) */
 105
 106#define ISER_QP_MAX_RECV_DTOS           (ISCSI_DEF_XMIT_CMDS_MAX)
 107
 108#define ISER_MIN_POSTED_RX              (ISCSI_DEF_XMIT_CMDS_MAX >> 2)
 109
 110/* the max TX (send) WR supported by the iSER QP is defined by                 *
 111 * max_send_wr = T * (1 + D) + C ; D is how many inflight dataouts we expect   *
 112 * to have at max for SCSI command. The tx posting & completion handling code  *
 113 * supports -EAGAIN scheme where tx is suspended till the QP has room for more *
 114 * send WR. D=8 comes from 64K/8K                                              */
 115
 116#define ISER_INFLIGHT_DATAOUTS          8
 117
 118#define ISER_QP_MAX_REQ_DTOS            (ISCSI_DEF_XMIT_CMDS_MAX *    \
 119                                        (1 + ISER_INFLIGHT_DATAOUTS) + \
 120                                        ISER_MAX_TX_MISC_PDUS        + \
 121                                        ISER_MAX_RX_MISC_PDUS)
 122
 123#define ISER_VER                        0x10
 124#define ISER_WSV                        0x08
 125#define ISER_RSV                        0x04
 126
 127struct iser_hdr {
 128        u8      flags;
 129        u8      rsvd[3];
 130        __be32  write_stag; /* write rkey */
 131        __be64  write_va;
 132        __be32  read_stag;  /* read rkey */
 133        __be64  read_va;
 134} __attribute__((packed));
 135
 136/* Constant PDU lengths calculations */
 137#define ISER_HEADERS_LEN  (sizeof(struct iser_hdr) + sizeof(struct iscsi_hdr))
 138
 139#define ISER_RECV_DATA_SEG_LEN  128
 140#define ISER_RX_PAYLOAD_SIZE    (ISER_HEADERS_LEN + ISER_RECV_DATA_SEG_LEN)
 141#define ISER_RX_LOGIN_SIZE      (ISER_HEADERS_LEN + ISCSI_DEF_MAX_RECV_SEG_LEN)
 142
 143/* Length of an object name string */
 144#define ISER_OBJECT_NAME_SIZE               64
 145
 146enum iser_ib_conn_state {
 147        ISER_CONN_INIT,            /* descriptor allocd, no conn          */
 148        ISER_CONN_PENDING,         /* in the process of being established */
 149        ISER_CONN_UP,              /* up and running                      */
 150        ISER_CONN_TERMINATING,     /* in the process of being terminated  */
 151        ISER_CONN_DOWN,            /* shut down                           */
 152        ISER_CONN_STATES_NUM
 153};
 154
 155enum iser_task_status {
 156        ISER_TASK_STATUS_INIT = 0,
 157        ISER_TASK_STATUS_STARTED,
 158        ISER_TASK_STATUS_COMPLETED
 159};
 160
 161enum iser_data_dir {
 162        ISER_DIR_IN = 0,           /* to initiator */
 163        ISER_DIR_OUT,              /* from initiator */
 164        ISER_DIRS_NUM
 165};
 166
 167struct iser_data_buf {
 168        void               *buf;      /* pointer to the sg list               */
 169        unsigned int       size;      /* num entries of this sg               */
 170        unsigned long      data_len;  /* total data len                       */
 171        unsigned int       dma_nents; /* returned by dma_map_sg               */
 172        char               *copy_buf; /* allocated copy buf for SGs unaligned *
 173                                       * for rdma which are copied            */
 174        struct scatterlist sg_single; /* SG-ified clone of a non SG SC or     *
 175                                       * unaligned SG                         */
 176  };
 177
 178/* fwd declarations */
 179struct iser_device;
 180struct iser_cq_desc;
 181struct iscsi_iser_conn;
 182struct iscsi_iser_task;
 183struct iscsi_endpoint;
 184
 185struct iser_mem_reg {
 186        u32  lkey;
 187        u32  rkey;
 188        u64  va;
 189        u64  len;
 190        void *mem_h;
 191        int  is_fmr;
 192};
 193
 194struct iser_regd_buf {
 195        struct iser_mem_reg     reg;        /* memory registration info        */
 196        void                    *virt_addr;
 197        struct iser_device      *device;    /* device->device for dma_unmap    */
 198        enum dma_data_direction direction;  /* direction for dma_unmap         */
 199        unsigned int            data_size;
 200};
 201
 202enum iser_desc_type {
 203        ISCSI_TX_CONTROL ,
 204        ISCSI_TX_SCSI_COMMAND,
 205        ISCSI_TX_DATAOUT
 206};
 207
 208struct iser_tx_desc {
 209        struct iser_hdr              iser_header;
 210        struct iscsi_hdr             iscsi_header;
 211        enum   iser_desc_type        type;
 212        u64                          dma_addr;
 213        /* sg[0] points to iser/iscsi headers, sg[1] optionally points to either
 214        of immediate data, unsolicited data-out or control (login,text) */
 215        struct ib_sge                tx_sg[2];
 216        int                          num_sge;
 217};
 218
 219#define ISER_RX_PAD_SIZE        (256 - (ISER_RX_PAYLOAD_SIZE + \
 220                                        sizeof(u64) + sizeof(struct ib_sge)))
 221struct iser_rx_desc {
 222        struct iser_hdr              iser_header;
 223        struct iscsi_hdr             iscsi_header;
 224        char                         data[ISER_RECV_DATA_SEG_LEN];
 225        u64                          dma_addr;
 226        struct ib_sge                rx_sg;
 227        char                         pad[ISER_RX_PAD_SIZE];
 228} __attribute__((packed));
 229
 230#define ISER_MAX_CQ 4
 231
 232struct iser_device {
 233        struct ib_device             *ib_device;
 234        struct ib_pd                 *pd;
 235        struct ib_cq                 *rx_cq[ISER_MAX_CQ];
 236        struct ib_cq                 *tx_cq[ISER_MAX_CQ];
 237        struct ib_mr                 *mr;
 238        struct tasklet_struct        cq_tasklet[ISER_MAX_CQ];
 239        struct ib_event_handler      event_handler;
 240        struct list_head             ig_list; /* entry in ig devices list */
 241        int                          refcount;
 242        int                          cq_active_qps[ISER_MAX_CQ];
 243        int                          cqs_used;
 244        struct iser_cq_desc          *cq_desc;
 245};
 246
 247struct iser_conn {
 248        struct iscsi_iser_conn       *iser_conn; /* iser conn for upcalls  */
 249        struct iscsi_endpoint        *ep;
 250        enum iser_ib_conn_state      state;         /* rdma connection state   */
 251        atomic_t                     refcount;
 252        spinlock_t                   lock;          /* used for state changes  */
 253        struct iser_device           *device;       /* device context          */
 254        struct rdma_cm_id            *cma_id;       /* CMA ID                  */
 255        struct ib_qp                 *qp;           /* QP                      */
 256        struct ib_fmr_pool           *fmr_pool;     /* pool of IB FMRs         */
 257        wait_queue_head_t            wait;          /* waitq for conn/disconn  */
 258        int                          post_recv_buf_count; /* posted rx count  */
 259        atomic_t                     post_send_buf_count; /* posted tx count   */
 260        char                         name[ISER_OBJECT_NAME_SIZE];
 261        struct iser_page_vec         *page_vec;     /* represents SG to fmr maps*
 262                                                     * maps serialized as tx is*/
 263        struct list_head             conn_list;       /* entry in ig conn list */
 264
 265        char                         *login_buf;
 266        char                         *login_req_buf, *login_resp_buf;
 267        u64                          login_req_dma, login_resp_dma;
 268        unsigned int                 rx_desc_head;
 269        struct iser_rx_desc          *rx_descs;
 270        struct ib_recv_wr            rx_wr[ISER_MIN_POSTED_RX];
 271};
 272
 273struct iscsi_iser_conn {
 274        struct iscsi_conn            *iscsi_conn;/* ptr to iscsi conn */
 275        struct iser_conn             *ib_conn;   /* iSER IB conn      */
 276};
 277
 278struct iscsi_iser_task {
 279        struct iser_tx_desc          desc;
 280        struct iscsi_iser_conn       *iser_conn;
 281        enum iser_task_status        status;
 282        int                          command_sent;  /* set if command  sent  */
 283        int                          dir[ISER_DIRS_NUM];      /* set if dir use*/
 284        struct iser_regd_buf         rdma_regd[ISER_DIRS_NUM];/* regd rdma buf */
 285        struct iser_data_buf         data[ISER_DIRS_NUM];     /* orig. data des*/
 286        struct iser_data_buf         data_copy[ISER_DIRS_NUM];/* contig. copy  */
 287};
 288
 289struct iser_page_vec {
 290        u64 *pages;
 291        int length;
 292        int offset;
 293        int data_size;
 294};
 295
 296struct iser_cq_desc {
 297        struct iser_device           *device;
 298        int                          cq_index;
 299};
 300
 301struct iser_global {
 302        struct mutex      device_list_mutex;/*                   */
 303        struct list_head  device_list;       /* all iSER devices */
 304        struct mutex      connlist_mutex;
 305        struct list_head  connlist;             /* all iSER IB connections */
 306
 307        struct kmem_cache *desc_cache;
 308};
 309
 310extern struct iser_global ig;
 311extern int iser_debug_level;
 312
 313/* allocate connection resources needed for rdma functionality */
 314int iser_conn_set_full_featured_mode(struct iscsi_conn *conn);
 315
 316int iser_send_control(struct iscsi_conn *conn,
 317                      struct iscsi_task *task);
 318
 319int iser_send_command(struct iscsi_conn *conn,
 320                      struct iscsi_task *task);
 321
 322int iser_send_data_out(struct iscsi_conn *conn,
 323                       struct iscsi_task *task,
 324                       struct iscsi_data *hdr);
 325
 326void iscsi_iser_recv(struct iscsi_conn *conn,
 327                     struct iscsi_hdr       *hdr,
 328                     char                   *rx_data,
 329                     int                    rx_data_len);
 330
 331void iser_conn_init(struct iser_conn *ib_conn);
 332
 333void iser_conn_get(struct iser_conn *ib_conn);
 334
 335int iser_conn_put(struct iser_conn *ib_conn, int destroy_cma_id_allowed);
 336
 337void iser_conn_terminate(struct iser_conn *ib_conn);
 338
 339void iser_rcv_completion(struct iser_rx_desc *desc,
 340                         unsigned long    dto_xfer_len,
 341                        struct iser_conn *ib_conn);
 342
 343void iser_snd_completion(struct iser_tx_desc *desc, struct iser_conn *ib_conn);
 344
 345void iser_task_rdma_init(struct iscsi_iser_task *task);
 346
 347void iser_task_rdma_finalize(struct iscsi_iser_task *task);
 348
 349void iser_free_rx_descriptors(struct iser_conn *ib_conn);
 350
 351void iser_finalize_rdma_unaligned_sg(struct iscsi_iser_task *task,
 352                                     enum iser_data_dir         cmd_dir);
 353
 354int  iser_reg_rdma_mem(struct iscsi_iser_task *task,
 355                       enum   iser_data_dir        cmd_dir);
 356
 357int  iser_connect(struct iser_conn   *ib_conn,
 358                  struct sockaddr_in *src_addr,
 359                  struct sockaddr_in *dst_addr,
 360                  int                non_blocking);
 361
 362int  iser_reg_page_vec(struct iser_conn     *ib_conn,
 363                       struct iser_page_vec *page_vec,
 364                       struct iser_mem_reg  *mem_reg);
 365
 366void iser_unreg_mem(struct iser_mem_reg *mem_reg);
 367
 368int  iser_post_recvl(struct iser_conn *ib_conn);
 369int  iser_post_recvm(struct iser_conn *ib_conn, int count);
 370int  iser_post_send(struct iser_conn *ib_conn, struct iser_tx_desc *tx_desc);
 371
 372int iser_dma_map_task_data(struct iscsi_iser_task *iser_task,
 373                            struct iser_data_buf       *data,
 374                            enum   iser_data_dir       iser_dir,
 375                            enum   dma_data_direction  dma_dir);
 376
 377void iser_dma_unmap_task_data(struct iscsi_iser_task *iser_task);
 378int  iser_initialize_task_headers(struct iscsi_task *task,
 379                        struct iser_tx_desc *tx_desc);
 380int iser_alloc_rx_descriptors(struct iser_conn *ib_conn);
 381#endif
 382