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 * $Id: iscsi_iser.h 7051 2006-05-10 12:29:11Z ogerlitz $
  41 */
  42#ifndef __ISCSI_ISER_H__
  43#define __ISCSI_ISER_H__
  44
  45#include <linux/types.h>
  46#include <linux/net.h>
  47#include <scsi/libiscsi.h>
  48#include <scsi/scsi_transport_iscsi.h>
  49
  50#include <linux/wait.h>
  51#include <linux/sched.h>
  52#include <linux/list.h>
  53#include <linux/slab.h>
  54#include <linux/dma-mapping.h>
  55#include <linux/mutex.h>
  56#include <linux/mempool.h>
  57#include <linux/uio.h>
  58
  59#include <linux/socket.h>
  60#include <linux/in.h>
  61#include <linux/in6.h>
  62
  63#include <rdma/ib_verbs.h>
  64#include <rdma/ib_fmr_pool.h>
  65#include <rdma/rdma_cm.h>
  66
  67#define DRV_NAME        "iser"
  68#define PFX             DRV_NAME ": "
  69#define DRV_VER         "0.1"
  70#define DRV_DATE        "May 7th, 2006"
  71
  72#define iser_dbg(fmt, arg...)                           \
  73        do {                                            \
  74                if (iser_debug_level > 0)               \
  75                        printk(KERN_DEBUG PFX "%s:" fmt,\
  76                                __func__ , ## arg);     \
  77        } while (0)
  78
  79#define iser_err(fmt, arg...)                           \
  80        do {                                            \
  81                printk(KERN_ERR PFX "%s:" fmt,          \
  82                       __func__ , ## arg);              \
  83        } while (0)
  84
  85                                        /* support upto 512KB in one RDMA */
  86#define ISCSI_ISER_SG_TABLESIZE         (0x80000 >> PAGE_SHIFT)
  87#define ISCSI_ISER_MAX_LUN              256
  88#define ISCSI_ISER_MAX_CMD_LEN          16
  89
  90/* QP settings */
  91/* Maximal bounds on received asynchronous PDUs */
  92#define ISER_MAX_RX_MISC_PDUS           4 /* NOOP_IN(2) , ASYNC_EVENT(2)   */
  93
  94#define ISER_MAX_TX_MISC_PDUS           6 /* NOOP_OUT(2), TEXT(1),         *
  95                                           * SCSI_TMFUNC(2), LOGOUT(1) */
  96
  97#define ISER_QP_MAX_RECV_DTOS           (ISCSI_XMIT_CMDS_MAX + \
  98                                        ISER_MAX_RX_MISC_PDUS    +  \
  99                                        ISER_MAX_TX_MISC_PDUS)
 100
 101/* the max TX (send) WR supported by the iSER QP is defined by                 *
 102 * max_send_wr = T * (1 + D) + C ; D is how many inflight dataouts we expect   *
 103 * to have at max for SCSI command. The tx posting & completion handling code  *
 104 * supports -EAGAIN scheme where tx is suspended till the QP has room for more *
 105 * send WR. D=8 comes from 64K/8K                                              */
 106
 107#define ISER_INFLIGHT_DATAOUTS          8
 108
 109#define ISER_QP_MAX_REQ_DTOS            (ISCSI_XMIT_CMDS_MAX *    \
 110                                        (1 + ISER_INFLIGHT_DATAOUTS) + \
 111                                        ISER_MAX_TX_MISC_PDUS        + \
 112                                        ISER_MAX_RX_MISC_PDUS)
 113
 114#define ISER_VER                        0x10
 115#define ISER_WSV                        0x08
 116#define ISER_RSV                        0x04
 117
 118struct iser_hdr {
 119        u8      flags;
 120        u8      rsvd[3];
 121        __be32  write_stag; /* write rkey */
 122        __be64  write_va;
 123        __be32  read_stag;  /* read rkey */
 124        __be64  read_va;
 125} __attribute__((packed));
 126
 127
 128/* Length of an object name string */
 129#define ISER_OBJECT_NAME_SIZE               64
 130
 131enum iser_ib_conn_state {
 132        ISER_CONN_INIT,            /* descriptor allocd, no conn          */
 133        ISER_CONN_PENDING,         /* in the process of being established */
 134        ISER_CONN_UP,              /* up and running                      */
 135        ISER_CONN_TERMINATING,     /* in the process of being terminated  */
 136        ISER_CONN_DOWN,            /* shut down                           */
 137        ISER_CONN_STATES_NUM
 138};
 139
 140enum iser_task_status {
 141        ISER_TASK_STATUS_INIT = 0,
 142        ISER_TASK_STATUS_STARTED,
 143        ISER_TASK_STATUS_COMPLETED
 144};
 145
 146enum iser_data_dir {
 147        ISER_DIR_IN = 0,           /* to initiator */
 148        ISER_DIR_OUT,              /* from initiator */
 149        ISER_DIRS_NUM
 150};
 151
 152struct iser_data_buf {
 153        void               *buf;      /* pointer to the sg list               */
 154        unsigned int       size;      /* num entries of this sg               */
 155        unsigned long      data_len;  /* total data len                       */
 156        unsigned int       dma_nents; /* returned by dma_map_sg               */
 157        char               *copy_buf; /* allocated copy buf for SGs unaligned *
 158                                       * for rdma which are copied            */
 159        struct scatterlist sg_single; /* SG-ified clone of a non SG SC or     *
 160                                       * unaligned SG                         */
 161  };
 162
 163/* fwd declarations */
 164struct iser_device;
 165struct iscsi_iser_conn;
 166struct iscsi_iser_cmd_task;
 167
 168struct iser_mem_reg {
 169        u32  lkey;
 170        u32  rkey;
 171        u64  va;
 172        u64  len;
 173        void *mem_h;
 174};
 175
 176struct iser_regd_buf {
 177        struct iser_mem_reg     reg;        /* memory registration info        */
 178        void                    *virt_addr;
 179        struct iser_device      *device;    /* device->device for dma_unmap    */
 180        dma_addr_t              dma_addr;   /* if non zero, addr for dma_unmap */
 181        enum dma_data_direction direction;  /* direction for dma_unmap         */
 182        unsigned int            data_size;
 183        atomic_t                ref_count;  /* refcount, freed when dec to 0   */
 184};
 185
 186#define MAX_REGD_BUF_VECTOR_LEN 2
 187
 188struct iser_dto {
 189        struct iscsi_iser_cmd_task *ctask;
 190        struct iscsi_iser_conn     *conn;
 191        int                        notify_enable;
 192
 193        /* vector of registered buffers */
 194        unsigned int               regd_vector_len;
 195        struct iser_regd_buf       *regd[MAX_REGD_BUF_VECTOR_LEN];
 196
 197        /* offset into the registered buffer may be specified */
 198        unsigned int               offset[MAX_REGD_BUF_VECTOR_LEN];
 199
 200        /* a smaller size may be specified, if 0, then full size is used */
 201        unsigned int               used_sz[MAX_REGD_BUF_VECTOR_LEN];
 202};
 203
 204enum iser_desc_type {
 205        ISCSI_RX,
 206        ISCSI_TX_CONTROL ,
 207        ISCSI_TX_SCSI_COMMAND,
 208        ISCSI_TX_DATAOUT
 209};
 210
 211struct iser_desc {
 212        struct iser_hdr              iser_header;
 213        struct iscsi_hdr             iscsi_header;
 214        struct iser_regd_buf         hdr_regd_buf;
 215        void                         *data;         /* used by RX & TX_CONTROL */
 216        struct iser_regd_buf         data_regd_buf; /* used by RX & TX_CONTROL */
 217        enum   iser_desc_type        type;
 218        struct iser_dto              dto;
 219};
 220
 221struct iser_device {
 222        struct ib_device             *ib_device;
 223        struct ib_pd                 *pd;
 224        struct ib_cq                 *cq;
 225        struct ib_mr                 *mr;
 226        struct tasklet_struct        cq_tasklet;
 227        struct list_head             ig_list; /* entry in ig devices list */
 228        int                          refcount;
 229};
 230
 231struct iser_conn {
 232        struct iscsi_iser_conn       *iser_conn; /* iser conn for upcalls  */
 233        enum iser_ib_conn_state      state;         /* rdma connection state   */
 234        spinlock_t                   lock;          /* used for state changes  */
 235        struct iser_device           *device;       /* device context          */
 236        struct rdma_cm_id            *cma_id;       /* CMA ID                  */
 237        struct ib_qp                 *qp;           /* QP                      */
 238        struct ib_fmr_pool           *fmr_pool;     /* pool of IB FMRs         */
 239        int                          disc_evt_flag; /* disconn event delivered */
 240        wait_queue_head_t            wait;          /* waitq for conn/disconn  */
 241        atomic_t                     post_recv_buf_count; /* posted rx count   */
 242        atomic_t                     post_send_buf_count; /* posted tx count   */
 243        struct work_struct           comperror_work; /* conn term sleepable ctx*/
 244        char                         name[ISER_OBJECT_NAME_SIZE];
 245        struct iser_page_vec         *page_vec;     /* represents SG to fmr maps*
 246                                                     * maps serialized as tx is*/
 247        struct list_head             conn_list;       /* entry in ig conn list */
 248};
 249
 250struct iscsi_iser_conn {
 251        struct iscsi_conn            *iscsi_conn;/* ptr to iscsi conn */
 252        struct iser_conn             *ib_conn;   /* iSER IB conn      */
 253
 254        rwlock_t                     lock;
 255};
 256
 257struct iscsi_iser_cmd_task {
 258        struct iser_desc             desc;
 259        struct iscsi_iser_conn       *iser_conn;
 260        int                          rdma_data_count;/* RDMA bytes           */
 261        enum iser_task_status        status;
 262        int                          command_sent;  /* set if command  sent  */
 263        int                          dir[ISER_DIRS_NUM];      /* set if dir use*/
 264        struct iser_regd_buf         rdma_regd[ISER_DIRS_NUM];/* regd rdma buf */
 265        struct iser_data_buf         data[ISER_DIRS_NUM];     /* orig. data des*/
 266        struct iser_data_buf         data_copy[ISER_DIRS_NUM];/* contig. copy  */
 267};
 268
 269struct iser_page_vec {
 270        u64 *pages;
 271        int length;
 272        int offset;
 273        int data_size;
 274};
 275
 276struct iser_global {
 277        struct mutex      device_list_mutex;/*                   */
 278        struct list_head  device_list;       /* all iSER devices */
 279        struct mutex      connlist_mutex;
 280        struct list_head  connlist;             /* all iSER IB connections */
 281
 282        kmem_cache_t *desc_cache;
 283};
 284
 285extern struct iser_global ig;
 286extern int iser_debug_level;
 287
 288/* allocate connection resources needed for rdma functionality */
 289int iser_conn_set_full_featured_mode(struct iscsi_conn *conn);
 290
 291int iser_send_control(struct iscsi_conn      *conn,
 292                      struct iscsi_mgmt_task *mtask);
 293
 294int iser_send_command(struct iscsi_conn      *conn,
 295                      struct iscsi_cmd_task  *ctask);
 296
 297int iser_send_data_out(struct iscsi_conn     *conn,
 298                       struct iscsi_cmd_task *ctask,
 299                       struct iscsi_data          *hdr);
 300
 301void iscsi_iser_recv(struct iscsi_conn *conn,
 302                     struct iscsi_hdr       *hdr,
 303                     char                   *rx_data,
 304                     int                    rx_data_len);
 305
 306int  iser_conn_init(struct iser_conn **ib_conn);
 307
 308void iser_conn_terminate(struct iser_conn *ib_conn);
 309
 310void iser_conn_release(struct iser_conn *ib_conn);
 311
 312void iser_rcv_completion(struct iser_desc *desc,
 313                         unsigned long    dto_xfer_len);
 314
 315void iser_snd_completion(struct iser_desc *desc);
 316
 317void iser_ctask_rdma_init(struct iscsi_iser_cmd_task     *ctask);
 318
 319void iser_ctask_rdma_finalize(struct iscsi_iser_cmd_task *ctask);
 320
 321void iser_dto_buffs_release(struct iser_dto *dto);
 322
 323int  iser_regd_buff_release(struct iser_regd_buf *regd_buf);
 324
 325void iser_reg_single(struct iser_device      *device,
 326                     struct iser_regd_buf    *regd_buf,
 327                     enum dma_data_direction direction);
 328
 329int  iser_start_rdma_unaligned_sg(struct iscsi_iser_cmd_task    *ctask,
 330                                  enum iser_data_dir            cmd_dir);
 331
 332void iser_finalize_rdma_unaligned_sg(struct iscsi_iser_cmd_task *ctask,
 333                                     enum iser_data_dir         cmd_dir);
 334
 335int  iser_reg_rdma_mem(struct iscsi_iser_cmd_task *ctask,
 336                       enum   iser_data_dir        cmd_dir);
 337
 338int  iser_connect(struct iser_conn   *ib_conn,
 339                  struct sockaddr_in *src_addr,
 340                  struct sockaddr_in *dst_addr,
 341                  int                non_blocking);
 342
 343int  iser_reg_page_vec(struct iser_conn     *ib_conn,
 344                       struct iser_page_vec *page_vec,
 345                       struct iser_mem_reg  *mem_reg);
 346
 347void iser_unreg_mem(struct iser_mem_reg *mem_reg);
 348
 349int  iser_post_recv(struct iser_desc *rx_desc);
 350int  iser_post_send(struct iser_desc *tx_desc);
 351
 352int iser_conn_state_comp(struct iser_conn *ib_conn,
 353                         enum iser_ib_conn_state comp);
 354#endif
 355