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 ISER_MAX_RX_MISC_PDUS + \ 107 ISER_MAX_TX_MISC_PDUS) 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 125 126 struct 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)); 134 135 136 /* Length of an object name string */ 137 #define ISER_OBJECT_NAME_SIZE 64 138 139 enum iser_ib_conn_state { 140 ISER_CONN_INIT, /* descriptor allocd, no conn */ 141 ISER_CONN_PENDING, /* in the process of being established */ 142 ISER_CONN_UP, /* up and running */ 143 ISER_CONN_TERMINATING, /* in the process of being terminated */ 144 ISER_CONN_DOWN, /* shut down */ 145 ISER_CONN_STATES_NUM 146 }; 147 148 enum iser_task_status { 149 ISER_TASK_STATUS_INIT = 0, 150 ISER_TASK_STATUS_STARTED, 151 ISER_TASK_STATUS_COMPLETED 152 }; 153 154 enum iser_data_dir { 155 ISER_DIR_IN = 0, /* to initiator */ 156 ISER_DIR_OUT, /* from initiator */ 157 ISER_DIRS_NUM 158 }; 159 160 struct iser_data_buf { 161 void *buf; /* pointer to the sg list */ 162 unsigned int size; /* num entries of this sg */ 163 unsigned long data_len; /* total data len */ 164 unsigned int dma_nents; /* returned by dma_map_sg */ 165 char *copy_buf; /* allocated copy buf for SGs unaligned * 166 * for rdma which are copied */ 167 struct scatterlist sg_single; /* SG-ified clone of a non SG SC or * 168 * unaligned SG */ 169 }; 170 171 /* fwd declarations */ 172 struct iser_device; 173 struct iscsi_iser_conn; 174 struct iscsi_iser_task; 175 struct iscsi_endpoint; 176 177 struct iser_mem_reg { 178 u32 lkey; 179 u32 rkey; 180 u64 va; 181 u64 len; 182 void *mem_h; 183 int is_fmr; 184 }; 185 186 struct iser_regd_buf { 187 struct iser_mem_reg reg; /* memory registration info */ 188 void *virt_addr; 189 struct iser_device *device; /* device->device for dma_unmap */ 190 u64 dma_addr; /* if non zero, addr for dma_unmap */ 191 enum dma_data_direction direction; /* direction for dma_unmap */ 192 unsigned int data_size; 193 atomic_t ref_count; /* refcount, freed when dec to 0 */ 194 }; 195 196 #define MAX_REGD_BUF_VECTOR_LEN 2 197 198 struct iser_dto { 199 struct iscsi_iser_task *task; 200 struct iser_conn *ib_conn; 201 int notify_enable; 202 203 /* vector of registered buffers */ 204 unsigned int regd_vector_len; 205 struct iser_regd_buf *regd[MAX_REGD_BUF_VECTOR_LEN]; 206 207 /* offset into the registered buffer may be specified */ 208 unsigned int offset[MAX_REGD_BUF_VECTOR_LEN]; 209 210 /* a smaller size may be specified, if 0, then full size is used */ 211 unsigned int used_sz[MAX_REGD_BUF_VECTOR_LEN]; 212 }; 213 214 enum iser_desc_type { 215 ISCSI_RX, 216 ISCSI_TX_CONTROL , 217 ISCSI_TX_SCSI_COMMAND, 218 ISCSI_TX_DATAOUT 219 }; 220 221 struct iser_desc { 222 struct iser_hdr iser_header; 223 struct iscsi_hdr iscsi_header; 224 struct iser_regd_buf hdr_regd_buf; 225 void *data; /* used by RX & TX_CONTROL */ 226 struct iser_regd_buf data_regd_buf; /* used by RX & TX_CONTROL */ 227 enum iser_desc_type type; 228 struct iser_dto dto; 229 }; 230 231 struct iser_device { 232 struct ib_device *ib_device; 233 struct ib_pd *pd; 234 struct ib_cq *cq; 235 struct ib_mr *mr; 236 struct tasklet_struct cq_tasklet; 237 struct list_head ig_list; /* entry in ig devices list */ 238 int refcount; 239 }; 240 241 struct iser_conn { 242 struct iscsi_iser_conn *iser_conn; /* iser conn for upcalls */ 243 struct iscsi_endpoint *ep; 244 enum iser_ib_conn_state state; /* rdma connection state */ 245 atomic_t refcount; 246 spinlock_t lock; /* used for state changes */ 247 struct iser_device *device; /* device context */ 248 struct rdma_cm_id *cma_id; /* CMA ID */ 249 struct ib_qp *qp; /* QP */ 250 struct ib_fmr_pool *fmr_pool; /* pool of IB FMRs */ 251 int disc_evt_flag; /* disconn event delivered */ 252 wait_queue_head_t wait; /* waitq for conn/disconn */ 253 atomic_t post_recv_buf_count; /* posted rx count */ 254 atomic_t post_send_buf_count; /* posted tx count */ 255 atomic_t unexpected_pdu_count;/* count of received * 256 * unexpected pdus * 257 * not yet retired */ 258 char name[ISER_OBJECT_NAME_SIZE]; 259 struct iser_page_vec *page_vec; /* represents SG to fmr maps* 260 * maps serialized as tx is*/ 261 struct list_head conn_list; /* entry in ig conn list */ 262 }; 263 264 struct iscsi_iser_conn { 265 struct iscsi_conn *iscsi_conn;/* ptr to iscsi conn */ 266 struct iser_conn *ib_conn; /* iSER IB conn */ 267 }; 268 269 struct iscsi_iser_task { 270 struct iser_desc desc; 271 struct iscsi_iser_conn *iser_conn; 272 enum iser_task_status status; 273 int command_sent; /* set if command sent */ 274 int dir[ISER_DIRS_NUM]; /* set if dir use*/ 275 struct iser_regd_buf rdma_regd[ISER_DIRS_NUM];/* regd rdma buf */ 276 struct iser_data_buf data[ISER_DIRS_NUM]; /* orig. data des*/ 277 struct iser_data_buf data_copy[ISER_DIRS_NUM];/* contig. copy */ 278 }; 279 280 struct iser_page_vec { 281 u64 *pages; 282 int length; 283 int offset; 284 int data_size; 285 }; 286 287 struct iser_global { 288 struct mutex device_list_mutex;/* */ 289 struct list_head device_list; /* all iSER devices */ 290 struct mutex connlist_mutex; 291 struct list_head connlist; /* all iSER IB connections */ 292 293 struct kmem_cache *desc_cache; 294 }; 295 296 extern struct iser_global ig; 297 extern int iser_debug_level; 298 299 /* allocate connection resources needed for rdma functionality */ 300 int iser_conn_set_full_featured_mode(struct iscsi_conn *conn); 301 302 int iser_send_control(struct iscsi_conn *conn, 303 struct iscsi_task *task); 304 305 int iser_send_command(struct iscsi_conn *conn, 306 struct iscsi_task *task); 307 308 int iser_send_data_out(struct iscsi_conn *conn, 309 struct iscsi_task *task, 310 struct iscsi_data *hdr); 311 312 void iscsi_iser_recv(struct iscsi_conn *conn, 313 struct iscsi_hdr *hdr, 314 char *rx_data, 315 int rx_data_len); 316 317 void iser_conn_init(struct iser_conn *ib_conn); 318 319 void iser_conn_get(struct iser_conn *ib_conn); 320 321 void iser_conn_put(struct iser_conn *ib_conn); 322 323 void iser_conn_terminate(struct iser_conn *ib_conn); 324 325 void iser_rcv_completion(struct iser_desc *desc, 326 unsigned long dto_xfer_len); 327 328 void iser_snd_completion(struct iser_desc *desc); 329 330 void iser_task_rdma_init(struct iscsi_iser_task *task); 331 332 void iser_task_rdma_finalize(struct iscsi_iser_task *task); 333 334 void iser_dto_buffs_release(struct iser_dto *dto); 335 336 int iser_regd_buff_release(struct iser_regd_buf *regd_buf); 337 338 void iser_reg_single(struct iser_device *device, 339 struct iser_regd_buf *regd_buf, 340 enum dma_data_direction direction); 341 342 void iser_finalize_rdma_unaligned_sg(struct iscsi_iser_task *task, 343 enum iser_data_dir cmd_dir); 344 345 int iser_reg_rdma_mem(struct iscsi_iser_task *task, 346 enum iser_data_dir cmd_dir); 347 348 int iser_connect(struct iser_conn *ib_conn, 349 struct sockaddr_in *src_addr, 350 struct sockaddr_in *dst_addr, 351 int non_blocking); 352 353 int iser_reg_page_vec(struct iser_conn *ib_conn, 354 struct iser_page_vec *page_vec, 355 struct iser_mem_reg *mem_reg); 356 357 void iser_unreg_mem(struct iser_mem_reg *mem_reg); 358 359 int iser_post_recv(struct iser_desc *rx_desc); 360 int iser_post_send(struct iser_desc *tx_desc); 361 362 int iser_conn_state_comp(struct iser_conn *ib_conn, 363 enum iser_ib_conn_state comp); 364 365 int iser_dma_map_task_data(struct iscsi_iser_task *iser_task, 366 struct iser_data_buf *data, 367 enum iser_data_dir iser_dir, 368 enum dma_data_direction dma_dir); 369 370 void iser_dma_unmap_task_data(struct iscsi_iser_task *iser_task); 371 #endif 372