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/wait.h>
  49#include <linux/sched.h>
  50#include <linux/list.h>
  51#include <linux/slab.h>
  52#include <linux/dma-mapping.h>
  53#include <linux/mutex.h>
  54#include <linux/mempool.h>
  55#include <linux/uio.h>
  56
  57#include <linux/socket.h>
  58#include <linux/in.h>
  59#include <linux/in6.h>
  60
  61#include <rdma/ib_verbs.h>
  62#include <rdma/ib_fmr_pool.h>
  63#include <rdma/rdma_cm.h>
  64
  65#define DRV_NAME        "iser"
  66#define PFX             DRV_NAME ": "
  67#define DRV_VER         "0.1"
  68#define DRV_DATE        "May 7th, 2006"
  69
  70#define iser_dbg(fmt, arg...)                           \
  71        do {                                            \
  72                if (iser_debug_level > 1)               \
  73                        printk(KERN_DEBUG PFX "%s:" fmt,\
  74                                __func__ , ## arg);     \
  75        } while (0)
  76
  77#define iser_warn(fmt, arg...)                          \
  78        do {                                            \
  79                if (iser_debug_level > 0)               \
  80                        printk(KERN_DEBUG PFX "%s:" fmt,\
  81                                __func__ , ## arg);     \
  82        } while (0)
  83
  84#define iser_err(fmt, arg...)                           \
  85        do {                                            \
  86                printk(KERN_ERR PFX "%s:" fmt,          \
  87                       __func__ , ## arg);              \
  88        } while (0)
  89
  90#define SHIFT_4K        12
  91#define SIZE_4K (1UL << SHIFT_4K)
  92#define MASK_4K (~(SIZE_4K-1))
  93
  94                                        /* support upto 512KB in one RDMA */
  95#define ISCSI_ISER_SG_TABLESIZE         (0x80000 >> SHIFT_4K)
  96#define ISER_DEF_CMD_PER_LUN            128
  97
  98/* QP settings */
  99/* Maximal bounds on received asynchronous PDUs */
 100#define ISER_MAX_RX_MISC_PDUS           4 /* NOOP_IN(2) , ASYNC_EVENT(2)   */
 101
 102#define ISER_MAX_TX_MISC_PDUS           6 /* NOOP_OUT(2), TEXT(1),         *
 103                                           * SCSI_TMFUNC(2), LOGOUT(1) */
 104
 105#define ISER_QP_MAX_RECV_DTOS           (ISCSI_DEF_XMIT_CMDS_MAX)
 106
 107#define ISER_MIN_POSTED_RX              (ISCSI_DEF_XMIT_CMDS_MAX >> 2)
 108
 109/* the max TX (send) WR supported by the iSER QP is defined by                 *
 110 * max_send_wr = T * (1 + D) + C ; D is how many inflight dataouts we expect   *
 111 * to have at max for SCSI command. The tx posting & completion handling code  *
 112 * supports -EAGAIN scheme where tx is suspended till the QP has room for more *
 113 * send WR. D=8 comes from 64K/8K                                              */
 114
 115#define ISER_INFLIGHT_DATAOUTS          8
 116
 117#define ISER_QP_MAX_REQ_DTOS            (ISCSI_DEF_XMIT_CMDS_MAX *    \
 118                                        (1 + ISER_INFLIGHT_DATAOUTS) + \
 119                                        ISER_MAX_TX_MISC_PDUS        + \
 120                                        ISER_MAX_RX_MISC_PDUS)
 121
 122#define ISER_VER                        0x10
 123#define ISER_WSV                        0x08
 124#define ISER_RSV                        0x04
 125
 126struct iser_hdr {
 127        u8      flags;
 128        u8      rsvd[3];
 129        __be32  write_stag; /* write rkey */
 130        __be64  write_va;
 131        __be32  read_stag;  /* read rkey */
 132        __be64  read_va;
 133} __attribute__((packed));
 134
 135/* Constant PDU lengths calculations */
 136#define ISER_HEADERS_LEN  (sizeof(struct iser_hdr) + sizeof(struct iscsi_hdr))
 137
 138#define ISER_RECV_DATA_SEG_LEN  128
 139#define ISER_RX_PAYLOAD_SIZE    (ISER_HEADERS_LEN + ISER_RECV_DATA_SEG_LEN)
 140#define ISER_RX_LOGIN_SIZE      (ISER_HEADERS_LEN + ISCSI_DEF_MAX_RECV_SEG_LEN)
 141
 142/* Length of an object name string */
 143#define ISER_OBJECT_NAME_SIZE               64
 144
 145enum iser_ib_conn_state {
 146        ISER_CONN_INIT,            /* descriptor allocd, no conn          */
 147        ISER_CONN_PENDING,         /* in the process of being established */
 148        ISER_CONN_UP,              /* up and running                      */
 149        ISER_CONN_TERMINATING,     /* in the process of being terminated  */
 150        ISER_CONN_DOWN,            /* shut down                           */
 151        ISER_CONN_STATES_NUM
 152};
 153
 154enum iser_task_status {
 155        ISER_TASK_STATUS_INIT = 0,
 156        ISER_TASK_STATUS_STARTED,
 157        ISER_TASK_STATUS_COMPLETED
 158};
 159
 160enum iser_data_dir {
 161        ISER_DIR_IN = 0,           /* to initiator */
 162        ISER_DIR_OUT,              /* from initiator */
 163        ISER_DIRS_NUM
 164};
 165
 166struct iser_data_buf {
 167        void               *buf;      /* pointer to the sg list               */
 168        unsigned int       size;      /* num entries of this sg               */
 169        unsigned long      data_len;  /* total data len                       */
 170        unsigned int       dma_nents; /* returned by dma_map_sg               */
 171        char               *copy_buf; /* allocated copy buf for SGs unaligned *
 172                                       * for rdma which are copied            */
 173        struct scatterlist sg_single; /* SG-ified clone of a non SG SC or     *
 174                                       * unaligned SG                         */
 175  };
 176
 177/* fwd declarations */
 178struct iser_device;
 179struct iscsi_iser_conn;
 180struct iscsi_iser_task;
 181struct iscsi_endpoint;
 182
 183struct iser_mem_reg {
 184        u32  lkey;
 185        u32  rkey;
 186        u64  va;
 187        u64  len;
 188        void *mem_h;
 189        int  is_fmr;
 190};
 191
 192struct iser_regd_buf {
 193        struct iser_mem_reg     reg;        /* memory registration info        */
 194        void                    *virt_addr;
 195        struct iser_device      *device;    /* device->device for dma_unmap    */
 196        enum dma_data_direction direction;  /* direction for dma_unmap         */
 197        unsigned int            data_size;
 198};
 199
 200enum iser_desc_type {
 201        ISCSI_TX_CONTROL ,
 202        ISCSI_TX_SCSI_COMMAND,
 203        ISCSI_TX_DATAOUT
 204};
 205
 206struct iser_tx_desc {
 207        struct iser_hdr              iser_header;
 208        struct iscsi_hdr             iscsi_header;
 209        enum   iser_desc_type        type;
 210        u64                          dma_addr;
 211        /* sg[0] points to iser/iscsi headers, sg[1] optionally points to either
 212        of immediate data, unsolicited data-out or control (login,text) */
 213        struct ib_sge                tx_sg[2];
 214        int                          num_sge;
 215};
 216
 217#define ISER_RX_PAD_SIZE        (256 - (ISER_RX_PAYLOAD_SIZE + \
 218                                        sizeof(u64) + sizeof(struct ib_sge)))
 219struct iser_rx_desc {
 220        struct iser_hdr              iser_header;
 221        struct iscsi_hdr             iscsi_header;
 222        char                         data[ISER_RECV_DATA_SEG_LEN];
 223        u64                          dma_addr;
 224        struct ib_sge                rx_sg;
 225        char                         pad[ISER_RX_PAD_SIZE];
 226} __attribute__((packed));
 227
 228struct iser_device {
 229        struct ib_device             *ib_device;
 230        struct ib_pd                 *pd;
 231        struct ib_cq                 *rx_cq;
 232        struct ib_cq                 *tx_cq;
 233        struct ib_mr                 *mr;
 234        struct tasklet_struct        cq_tasklet;
 235        struct ib_event_handler      event_handler;
 236        struct list_head             ig_list; /* entry in ig devices list */
 237        int                          refcount;
 238};
 239
 240struct iser_conn {
 241        struct iscsi_iser_conn       *iser_conn; /* iser conn for upcalls  */
 242        struct iscsi_endpoint        *ep;
 243        enum iser_ib_conn_state      state;         /* rdma connection state   */
 244        atomic_t                     refcount;
 245        spinlock_t                   lock;          /* used for state changes  */
 246        struct iser_device           *device;       /* device context          */
 247        struct rdma_cm_id            *cma_id;       /* CMA ID                  */
 248        struct ib_qp                 *qp;           /* QP                      */
 249        struct ib_fmr_pool           *fmr_pool;     /* pool of IB FMRs         */
 250        wait_queue_head_t            wait;          /* waitq for conn/disconn  */
 251        int                          post_recv_buf_count; /* posted rx count  */
 252        atomic_t                     post_send_buf_count; /* posted tx count   */
 253        char                         name[ISER_OBJECT_NAME_SIZE];
 254        struct iser_page_vec         *page_vec;     /* represents SG to fmr maps*
 255                                                     * maps serialized as tx is*/
 256        struct list_head             conn_list;       /* entry in ig conn list */
 257
 258        char                         *login_buf;
 259        u64                          login_dma;
 260        unsigned int                 rx_desc_head;
 261        struct iser_rx_desc          *rx_descs;
 262        struct ib_recv_wr            rx_wr[ISER_MIN_POSTED_RX];
 263};
 264
 265struct iscsi_iser_conn {
 266        struct iscsi_conn            *iscsi_conn;/* ptr to iscsi conn */
 267        struct iser_conn             *ib_conn;   /* iSER IB conn      */
 268};
 269
 270struct iscsi_iser_task {
 271        struct iser_tx_desc          desc;
 272        struct iscsi_iser_conn       *iser_conn;
 273        enum iser_task_status        status;
 274        int                          command_sent;  /* set if command  sent  */
 275        int                          dir[ISER_DIRS_NUM];      /* set if dir use*/
 276        struct iser_regd_buf         rdma_regd[ISER_DIRS_NUM];/* regd rdma buf */
 277        struct iser_data_buf         data[ISER_DIRS_NUM];     /* orig. data des*/
 278        struct iser_data_buf         data_copy[ISER_DIRS_NUM];/* contig. copy  */
 279        int                          headers_initialized;
 280};
 281
 282struct iser_page_vec {
 283        u64 *pages;
 284        int length;
 285        int offset;
 286        int data_size;
 287};
 288
 289struct iser_global {
 290        struct mutex      device_list_mutex;/*                   */
 291        struct list_head  device_list;       /* all iSER devices */
 292        struct mutex      connlist_mutex;
 293        struct list_head  connlist;             /* all iSER IB connections */
 294
 295        struct kmem_cache *desc_cache;
 296};
 297
 298extern struct iser_global ig;
 299extern int iser_debug_level;
 300
 301/* allocate connection resources needed for rdma functionality */
 302int iser_conn_set_full_featured_mode(struct iscsi_conn *conn);
 303
 304int iser_send_control(struct iscsi_conn *conn,
 305                      struct iscsi_task *task);
 306
 307int iser_send_command(struct iscsi_conn *conn,
 308                      struct iscsi_task *task);
 309
 310int iser_send_data_out(struct iscsi_conn *conn,
 311                       struct iscsi_task *task,
 312                       struct iscsi_data *hdr);
 313
 314void iscsi_iser_recv(struct iscsi_conn *conn,
 315                     struct iscsi_hdr       *hdr,
 316                     char                   *rx_data,
 317                     int                    rx_data_len);
 318
 319void iser_conn_init(struct iser_conn *ib_conn);
 320
 321void iser_conn_get(struct iser_conn *ib_conn);
 322
 323int iser_conn_put(struct iser_conn *ib_conn, int destroy_cma_id_allowed);
 324
 325void iser_conn_terminate(struct iser_conn *ib_conn);
 326
 327void iser_rcv_completion(struct iser_rx_desc *desc,
 328                         unsigned long    dto_xfer_len,
 329                        struct iser_conn *ib_conn);
 330
 331void iser_snd_completion(struct iser_tx_desc *desc, struct iser_conn *ib_conn);
 332
 333void iser_task_rdma_init(struct iscsi_iser_task *task);
 334
 335void iser_task_rdma_finalize(struct iscsi_iser_task *task);
 336
 337void iser_free_rx_descriptors(struct iser_conn *ib_conn);
 338
 339void iser_finalize_rdma_unaligned_sg(struct iscsi_iser_task *task,
 340                                     enum iser_data_dir         cmd_dir);
 341
 342int  iser_reg_rdma_mem(struct iscsi_iser_task *task,
 343                       enum   iser_data_dir        cmd_dir);
 344
 345int  iser_connect(struct iser_conn   *ib_conn,
 346                  struct sockaddr_in *src_addr,
 347                  struct sockaddr_in *dst_addr,
 348                  int                non_blocking);
 349
 350int  iser_reg_page_vec(struct iser_conn     *ib_conn,
 351                       struct iser_page_vec *page_vec,
 352                       struct iser_mem_reg  *mem_reg);
 353
 354void iser_unreg_mem(struct iser_mem_reg *mem_reg);
 355
 356int  iser_post_recvl(struct iser_conn *ib_conn);
 357int  iser_post_recvm(struct iser_conn *ib_conn, int count);
 358int  iser_post_send(struct iser_conn *ib_conn, struct iser_tx_desc *tx_desc);
 359
 360int iser_dma_map_task_data(struct iscsi_iser_task *iser_task,
 361                            struct iser_data_buf       *data,
 362                            enum   iser_data_dir       iser_dir,
 363                            enum   dma_data_direction  dma_dir);
 364
 365void iser_dma_unmap_task_data(struct iscsi_iser_task *iser_task);
 366int  iser_initialize_task_headers(struct iscsi_task *task,
 367                        struct iser_tx_desc *tx_desc);
 368#endif
 369