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 * Copyright (c) 2013 Mellanox Technologies. All rights reserved. 12 * 13 * This software is available to you under a choice of one of two 14 * licenses. You may choose to be licensed under the terms of the GNU 15 * General Public License (GPL) Version 2, available from the file 16 * COPYING in the main directory of this source tree, or the 17 * OpenIB.org BSD license below: 18 * 19 * Redistribution and use in source and binary forms, with or 20 * without modification, are permitted provided that the following 21 * conditions are met: 22 * 23 * - Redistributions of source code must retain the above 24 * copyright notice, this list of conditions and the following 25 * disclaimer. 26 * 27 * - Redistributions in binary form must reproduce the above 28 * copyright notice, this list of conditions and the following 29 * disclaimer in the documentation and/or other materials 30 * provided with the distribution. 31 * 32 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 33 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 34 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 35 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 36 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 37 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 38 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 39 * SOFTWARE. 40 */ 41 #ifndef __ISCSI_ISER_H__ 42 #define __ISCSI_ISER_H__ 43 44 #include <linux/types.h> 45 #include <linux/net.h> 46 #include <linux/printk.h> 47 #include <scsi/libiscsi.h> 48 #include <scsi/scsi_transport_iscsi.h> 49 50 #include <linux/interrupt.h> 51 #include <linux/wait.h> 52 #include <linux/sched.h> 53 #include <linux/list.h> 54 #include <linux/slab.h> 55 #include <linux/dma-mapping.h> 56 #include <linux/mutex.h> 57 #include <linux/mempool.h> 58 #include <linux/uio.h> 59 60 #include <linux/socket.h> 61 #include <linux/in.h> 62 #include <linux/in6.h> 63 64 #include <rdma/ib_verbs.h> 65 #include <rdma/ib_fmr_pool.h> 66 #include <rdma/rdma_cm.h> 67 68 #define DRV_NAME "iser" 69 #define PFX DRV_NAME ": " 70 #define DRV_VER "1.1" 71 72 #define iser_dbg(fmt, arg...) \ 73 do { \ 74 if (iser_debug_level > 2) \ 75 printk(KERN_DEBUG PFX "%s:" fmt,\ 76 __func__ , ## arg); \ 77 } while (0) 78 79 #define iser_warn(fmt, arg...) \ 80 do { \ 81 if (iser_debug_level > 1) \ 82 pr_warn(PFX "%s:" fmt, \ 83 __func__ , ## arg); \ 84 } while (0) 85 86 #define iser_info(fmt, arg...) \ 87 do { \ 88 if (iser_debug_level > 0) \ 89 pr_info(PFX "%s:" fmt, \ 90 __func__ , ## arg); \ 91 } while (0) 92 93 #define iser_err(fmt, arg...) \ 94 do { \ 95 printk(KERN_ERR PFX "%s:" fmt, \ 96 __func__ , ## arg); \ 97 } while (0) 98 99 #define SHIFT_4K 12 100 #define SIZE_4K (1ULL << SHIFT_4K) 101 #define MASK_4K (~(SIZE_4K-1)) 102 103 /* support up to 512KB in one RDMA */ 104 #define ISCSI_ISER_SG_TABLESIZE (0x80000 >> SHIFT_4K) 105 #define ISER_DEF_CMD_PER_LUN ISCSI_DEF_XMIT_CMDS_MAX 106 107 /* QP settings */ 108 /* Maximal bounds on received asynchronous PDUs */ 109 #define ISER_MAX_RX_MISC_PDUS 4 /* NOOP_IN(2) , ASYNC_EVENT(2) */ 110 111 #define ISER_MAX_TX_MISC_PDUS 6 /* NOOP_OUT(2), TEXT(1), * 112 * SCSI_TMFUNC(2), LOGOUT(1) */ 113 114 #define ISER_QP_MAX_RECV_DTOS (ISCSI_DEF_XMIT_CMDS_MAX) 115 116 #define ISER_MIN_POSTED_RX (ISCSI_DEF_XMIT_CMDS_MAX >> 2) 117 118 /* the max TX (send) WR supported by the iSER QP is defined by * 119 * max_send_wr = T * (1 + D) + C ; D is how many inflight dataouts we expect * 120 * to have at max for SCSI command. The tx posting & completion handling code * 121 * supports -EAGAIN scheme where tx is suspended till the QP has room for more * 122 * send WR. D=8 comes from 64K/8K */ 123 124 #define ISER_INFLIGHT_DATAOUTS 8 125 126 #define ISER_QP_MAX_REQ_DTOS (ISCSI_DEF_XMIT_CMDS_MAX * \ 127 (1 + ISER_INFLIGHT_DATAOUTS) + \ 128 ISER_MAX_TX_MISC_PDUS + \ 129 ISER_MAX_RX_MISC_PDUS) 130 131 #define ISER_VER 0x10 132 #define ISER_WSV 0x08 133 #define ISER_RSV 0x04 134 135 struct iser_hdr { 136 u8 flags; 137 u8 rsvd[3]; 138 __be32 write_stag; /* write rkey */ 139 __be64 write_va; 140 __be32 read_stag; /* read rkey */ 141 __be64 read_va; 142 } __attribute__((packed)); 143 144 145 #define ISER_ZBVA_NOT_SUPPORTED 0x80 146 #define ISER_SEND_W_INV_NOT_SUPPORTED 0x40 147 148 struct iser_cm_hdr { 149 u8 flags; 150 u8 rsvd[3]; 151 } __packed; 152 153 /* Constant PDU lengths calculations */ 154 #define ISER_HEADERS_LEN (sizeof(struct iser_hdr) + sizeof(struct iscsi_hdr)) 155 156 #define ISER_RECV_DATA_SEG_LEN 128 157 #define ISER_RX_PAYLOAD_SIZE (ISER_HEADERS_LEN + ISER_RECV_DATA_SEG_LEN) 158 #define ISER_RX_LOGIN_SIZE (ISER_HEADERS_LEN + ISCSI_DEF_MAX_RECV_SEG_LEN) 159 160 /* Length of an object name string */ 161 #define ISER_OBJECT_NAME_SIZE 64 162 163 enum iser_ib_conn_state { 164 ISER_CONN_INIT, /* descriptor allocd, no conn */ 165 ISER_CONN_PENDING, /* in the process of being established */ 166 ISER_CONN_UP, /* up and running */ 167 ISER_CONN_TERMINATING, /* in the process of being terminated */ 168 ISER_CONN_DOWN, /* shut down */ 169 ISER_CONN_STATES_NUM 170 }; 171 172 enum iser_task_status { 173 ISER_TASK_STATUS_INIT = 0, 174 ISER_TASK_STATUS_STARTED, 175 ISER_TASK_STATUS_COMPLETED 176 }; 177 178 enum iser_data_dir { 179 ISER_DIR_IN = 0, /* to initiator */ 180 ISER_DIR_OUT, /* from initiator */ 181 ISER_DIRS_NUM 182 }; 183 184 struct iser_data_buf { 185 void *buf; /* pointer to the sg list */ 186 unsigned int size; /* num entries of this sg */ 187 unsigned long data_len; /* total data len */ 188 unsigned int dma_nents; /* returned by dma_map_sg */ 189 char *copy_buf; /* allocated copy buf for SGs unaligned * 190 * for rdma which are copied */ 191 struct scatterlist sg_single; /* SG-ified clone of a non SG SC or * 192 * unaligned SG */ 193 }; 194 195 /* fwd declarations */ 196 struct iser_device; 197 struct iser_cq_desc; 198 struct iscsi_iser_conn; 199 struct iscsi_iser_task; 200 struct iscsi_endpoint; 201 202 struct iser_mem_reg { 203 u32 lkey; 204 u32 rkey; 205 u64 va; 206 u64 len; 207 void *mem_h; 208 int is_fmr; 209 }; 210 211 struct iser_regd_buf { 212 struct iser_mem_reg reg; /* memory registration info */ 213 void *virt_addr; 214 struct iser_device *device; /* device->device for dma_unmap */ 215 enum dma_data_direction direction; /* direction for dma_unmap */ 216 unsigned int data_size; 217 }; 218 219 enum iser_desc_type { 220 ISCSI_TX_CONTROL , 221 ISCSI_TX_SCSI_COMMAND, 222 ISCSI_TX_DATAOUT 223 }; 224 225 struct iser_tx_desc { 226 struct iser_hdr iser_header; 227 struct iscsi_hdr iscsi_header; 228 enum iser_desc_type type; 229 u64 dma_addr; 230 /* sg[0] points to iser/iscsi headers, sg[1] optionally points to either 231 of immediate data, unsolicited data-out or control (login,text) */ 232 struct ib_sge tx_sg[2]; 233 int num_sge; 234 }; 235 236 #define ISER_RX_PAD_SIZE (256 - (ISER_RX_PAYLOAD_SIZE + \ 237 sizeof(u64) + sizeof(struct ib_sge))) 238 struct iser_rx_desc { 239 struct iser_hdr iser_header; 240 struct iscsi_hdr iscsi_header; 241 char data[ISER_RECV_DATA_SEG_LEN]; 242 u64 dma_addr; 243 struct ib_sge rx_sg; 244 char pad[ISER_RX_PAD_SIZE]; 245 } __attribute__((packed)); 246 247 #define ISER_MAX_CQ 4 248 249 struct iser_device { 250 struct ib_device *ib_device; 251 struct ib_pd *pd; 252 struct ib_cq *rx_cq[ISER_MAX_CQ]; 253 struct ib_cq *tx_cq[ISER_MAX_CQ]; 254 struct ib_mr *mr; 255 struct tasklet_struct cq_tasklet[ISER_MAX_CQ]; 256 struct ib_event_handler event_handler; 257 struct list_head ig_list; /* entry in ig devices list */ 258 int refcount; 259 int cq_active_qps[ISER_MAX_CQ]; 260 int cqs_used; 261 struct iser_cq_desc *cq_desc; 262 }; 263 264 struct iser_conn { 265 struct iscsi_iser_conn *iser_conn; /* iser conn for upcalls */ 266 struct iscsi_endpoint *ep; 267 enum iser_ib_conn_state state; /* rdma connection state */ 268 atomic_t refcount; 269 spinlock_t lock; /* used for state changes */ 270 struct iser_device *device; /* device context */ 271 struct rdma_cm_id *cma_id; /* CMA ID */ 272 struct ib_qp *qp; /* QP */ 273 struct ib_fmr_pool *fmr_pool; /* pool of IB FMRs */ 274 wait_queue_head_t wait; /* waitq for conn/disconn */ 275 int post_recv_buf_count; /* posted rx count */ 276 atomic_t post_send_buf_count; /* posted tx count */ 277 char name[ISER_OBJECT_NAME_SIZE]; 278 struct iser_page_vec *page_vec; /* represents SG to fmr maps* 279 * maps serialized as tx is*/ 280 struct list_head conn_list; /* entry in ig conn list */ 281 282 char *login_buf; 283 char *login_req_buf, *login_resp_buf; 284 u64 login_req_dma, login_resp_dma; 285 unsigned int rx_desc_head; 286 struct iser_rx_desc *rx_descs; 287 struct ib_recv_wr rx_wr[ISER_MIN_POSTED_RX]; 288 }; 289 290 struct iscsi_iser_conn { 291 struct iscsi_conn *iscsi_conn;/* ptr to iscsi conn */ 292 struct iser_conn *ib_conn; /* iSER IB conn */ 293 }; 294 295 struct iscsi_iser_task { 296 struct iser_tx_desc desc; 297 struct iscsi_iser_conn *iser_conn; 298 enum iser_task_status status; 299 int command_sent; /* set if command sent */ 300 int dir[ISER_DIRS_NUM]; /* set if dir use*/ 301 struct iser_regd_buf rdma_regd[ISER_DIRS_NUM];/* regd rdma buf */ 302 struct iser_data_buf data[ISER_DIRS_NUM]; /* orig. data des*/ 303 struct iser_data_buf data_copy[ISER_DIRS_NUM];/* contig. copy */ 304 }; 305 306 struct iser_page_vec { 307 u64 *pages; 308 int length; 309 int offset; 310 int data_size; 311 }; 312 313 struct iser_cq_desc { 314 struct iser_device *device; 315 int cq_index; 316 }; 317 318 struct iser_global { 319 struct mutex device_list_mutex;/* */ 320 struct list_head device_list; /* all iSER devices */ 321 struct mutex connlist_mutex; 322 struct list_head connlist; /* all iSER IB connections */ 323 324 struct kmem_cache *desc_cache; 325 }; 326 327 extern struct iser_global ig; 328 extern int iser_debug_level; 329 330 /* allocate connection resources needed for rdma functionality */ 331 int iser_conn_set_full_featured_mode(struct iscsi_conn *conn); 332 333 int iser_send_control(struct iscsi_conn *conn, 334 struct iscsi_task *task); 335 336 int iser_send_command(struct iscsi_conn *conn, 337 struct iscsi_task *task); 338 339 int iser_send_data_out(struct iscsi_conn *conn, 340 struct iscsi_task *task, 341 struct iscsi_data *hdr); 342 343 void iscsi_iser_recv(struct iscsi_conn *conn, 344 struct iscsi_hdr *hdr, 345 char *rx_data, 346 int rx_data_len); 347 348 void iser_conn_init(struct iser_conn *ib_conn); 349 350 void iser_conn_get(struct iser_conn *ib_conn); 351 352 int iser_conn_put(struct iser_conn *ib_conn, int destroy_cma_id_allowed); 353 354 void iser_conn_terminate(struct iser_conn *ib_conn); 355 356 void iser_rcv_completion(struct iser_rx_desc *desc, 357 unsigned long dto_xfer_len, 358 struct iser_conn *ib_conn); 359 360 void iser_snd_completion(struct iser_tx_desc *desc, struct iser_conn *ib_conn); 361 362 void iser_task_rdma_init(struct iscsi_iser_task *task); 363 364 void iser_task_rdma_finalize(struct iscsi_iser_task *task); 365 366 void iser_free_rx_descriptors(struct iser_conn *ib_conn); 367 368 void iser_finalize_rdma_unaligned_sg(struct iscsi_iser_task *task, 369 enum iser_data_dir cmd_dir); 370 371 int iser_reg_rdma_mem(struct iscsi_iser_task *task, 372 enum iser_data_dir cmd_dir); 373 374 int iser_connect(struct iser_conn *ib_conn, 375 struct sockaddr_in *src_addr, 376 struct sockaddr_in *dst_addr, 377 int non_blocking); 378 379 int iser_reg_page_vec(struct iser_conn *ib_conn, 380 struct iser_page_vec *page_vec, 381 struct iser_mem_reg *mem_reg); 382 383 void iser_unreg_mem(struct iser_mem_reg *mem_reg); 384 385 int iser_post_recvl(struct iser_conn *ib_conn); 386 int iser_post_recvm(struct iser_conn *ib_conn, int count); 387 int iser_post_send(struct iser_conn *ib_conn, struct iser_tx_desc *tx_desc); 388 389 int iser_dma_map_task_data(struct iscsi_iser_task *iser_task, 390 struct iser_data_buf *data, 391 enum iser_data_dir iser_dir, 392 enum dma_data_direction dma_dir); 393 394 void iser_dma_unmap_task_data(struct iscsi_iser_task *iser_task); 395 int iser_initialize_task_headers(struct iscsi_task *task, 396 struct iser_tx_desc *tx_desc); 397 int iser_alloc_rx_descriptors(struct iser_conn *ib_conn); 398 #endif 399