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) 2013-2014 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 #include <scsi/scsi_cmnd.h> 50 #include <scsi/scsi_device.h> 51 #include <scsi/iser.h> 52 53 #include <linux/interrupt.h> 54 #include <linux/wait.h> 55 #include <linux/sched.h> 56 #include <linux/list.h> 57 #include <linux/slab.h> 58 #include <linux/dma-mapping.h> 59 #include <linux/mutex.h> 60 #include <linux/mempool.h> 61 #include <linux/uio.h> 62 63 #include <linux/socket.h> 64 #include <linux/in.h> 65 #include <linux/in6.h> 66 67 #include <rdma/ib_verbs.h> 68 #include <rdma/rdma_cm.h> 69 70 #define DRV_NAME "iser" 71 #define PFX DRV_NAME ": " 72 #define DRV_VER "1.6" 73 74 #define iser_dbg(fmt, arg...) \ 75 do { \ 76 if (unlikely(iser_debug_level > 2)) \ 77 printk(KERN_DEBUG PFX "%s: " fmt,\ 78 __func__ , ## arg); \ 79 } while (0) 80 81 #define iser_warn(fmt, arg...) \ 82 do { \ 83 if (unlikely(iser_debug_level > 0)) \ 84 pr_warn(PFX "%s: " fmt, \ 85 __func__ , ## arg); \ 86 } while (0) 87 88 #define iser_info(fmt, arg...) \ 89 do { \ 90 if (unlikely(iser_debug_level > 1)) \ 91 pr_info(PFX "%s: " fmt, \ 92 __func__ , ## arg); \ 93 } while (0) 94 95 #define iser_err(fmt, arg...) \ 96 pr_err(PFX "%s: " fmt, __func__ , ## arg) 97 98 /* Default support is 512KB I/O size */ 99 #define ISER_DEF_MAX_SECTORS 1024 100 #define ISCSI_ISER_DEF_SG_TABLESIZE \ 101 ((ISER_DEF_MAX_SECTORS * SECTOR_SIZE) >> ilog2(SZ_4K)) 102 /* Maximum support is 16MB I/O size */ 103 #define ISCSI_ISER_MAX_SG_TABLESIZE ((32768 * SECTOR_SIZE) >> ilog2(SZ_4K)) 104 105 #define ISER_DEF_XMIT_CMDS_DEFAULT 512 106 #if ISCSI_DEF_XMIT_CMDS_MAX > ISER_DEF_XMIT_CMDS_DEFAULT 107 #define ISER_DEF_XMIT_CMDS_MAX ISCSI_DEF_XMIT_CMDS_MAX 108 #else 109 #define ISER_DEF_XMIT_CMDS_MAX ISER_DEF_XMIT_CMDS_DEFAULT 110 #endif 111 #define ISER_DEF_CMD_PER_LUN ISER_DEF_XMIT_CMDS_MAX 112 113 /* QP settings */ 114 /* Maximal bounds on received asynchronous PDUs */ 115 #define ISER_MAX_RX_MISC_PDUS 4 /* NOOP_IN(2) , ASYNC_EVENT(2) */ 116 117 #define ISER_MAX_TX_MISC_PDUS 6 /* NOOP_OUT(2), TEXT(1), * 118 * SCSI_TMFUNC(2), LOGOUT(1) */ 119 120 #define ISER_QP_MAX_RECV_DTOS (ISER_DEF_XMIT_CMDS_MAX) 121 122 /* the max TX (send) WR supported by the iSER QP is defined by * 123 * max_send_wr = T * (1 + D) + C ; D is how many inflight dataouts we expect * 124 * to have at max for SCSI command. The tx posting & completion handling code * 125 * supports -EAGAIN scheme where tx is suspended till the QP has room for more * 126 * send WR. D=8 comes from 64K/8K */ 127 128 #define ISER_INFLIGHT_DATAOUTS 8 129 130 #define ISER_QP_MAX_REQ_DTOS (ISER_DEF_XMIT_CMDS_MAX * \ 131 (1 + ISER_INFLIGHT_DATAOUTS) + \ 132 ISER_MAX_TX_MISC_PDUS + \ 133 ISER_MAX_RX_MISC_PDUS) 134 135 /* Max registration work requests per command */ 136 #define ISER_MAX_REG_WR_PER_CMD 5 137 138 /* For Signature we don't support DATAOUTs so no need to make room for them */ 139 #define ISER_QP_SIG_MAX_REQ_DTOS (ISER_DEF_XMIT_CMDS_MAX * \ 140 (1 + ISER_MAX_REG_WR_PER_CMD) + \ 141 ISER_MAX_TX_MISC_PDUS + \ 142 ISER_MAX_RX_MISC_PDUS) 143 144 #define ISER_GET_MAX_XMIT_CMDS(send_wr) ((send_wr \ 145 - ISER_MAX_TX_MISC_PDUS \ 146 - ISER_MAX_RX_MISC_PDUS) / \ 147 (1 + ISER_INFLIGHT_DATAOUTS)) 148 149 #define ISER_SIGNAL_CMD_COUNT 32 150 151 /* Constant PDU lengths calculations */ 152 #define ISER_HEADERS_LEN (sizeof(struct iser_ctrl) + sizeof(struct iscsi_hdr)) 153 154 #define ISER_RECV_DATA_SEG_LEN 128 155 #define ISER_RX_PAYLOAD_SIZE (ISER_HEADERS_LEN + ISER_RECV_DATA_SEG_LEN) 156 #define ISER_RX_LOGIN_SIZE (ISER_HEADERS_LEN + ISCSI_DEF_MAX_RECV_SEG_LEN) 157 158 /* Length of an object name string */ 159 #define ISER_OBJECT_NAME_SIZE 64 160 161 enum iser_conn_state { 162 ISER_CONN_INIT, /* descriptor allocd, no conn */ 163 ISER_CONN_PENDING, /* in the process of being established */ 164 ISER_CONN_UP, /* up and running */ 165 ISER_CONN_TERMINATING, /* in the process of being terminated */ 166 ISER_CONN_DOWN, /* shut down */ 167 ISER_CONN_STATES_NUM 168 }; 169 170 enum iser_task_status { 171 ISER_TASK_STATUS_INIT = 0, 172 ISER_TASK_STATUS_STARTED, 173 ISER_TASK_STATUS_COMPLETED 174 }; 175 176 enum iser_data_dir { 177 ISER_DIR_IN = 0, /* to initiator */ 178 ISER_DIR_OUT, /* from initiator */ 179 ISER_DIRS_NUM 180 }; 181 182 /** 183 * struct iser_data_buf - iSER data buffer 184 * 185 * @sg: pointer to the sg list 186 * @size: num entries of this sg 187 * @data_len: total beffer byte len 188 * @dma_nents: returned by dma_map_sg 189 */ 190 struct iser_data_buf { 191 struct scatterlist *sg; 192 int size; 193 unsigned long data_len; 194 int dma_nents; 195 }; 196 197 /* fwd declarations */ 198 struct iser_device; 199 struct iscsi_iser_task; 200 struct iscsi_endpoint; 201 struct iser_reg_resources; 202 203 /** 204 * struct iser_mem_reg - iSER memory registration info 205 * 206 * @sge: memory region sg element 207 * @rkey: memory region remote key 208 * @mem_h: pointer to registration context (FMR/Fastreg) 209 */ 210 struct iser_mem_reg { 211 struct ib_sge sge; 212 u32 rkey; 213 void *mem_h; 214 }; 215 216 enum iser_desc_type { 217 ISCSI_TX_CONTROL , 218 ISCSI_TX_SCSI_COMMAND, 219 ISCSI_TX_DATAOUT 220 }; 221 222 /** 223 * struct iser_tx_desc - iSER TX descriptor 224 * 225 * @iser_header: iser header 226 * @iscsi_header: iscsi header 227 * @type: command/control/dataout 228 * @dma_addr: header buffer dma_address 229 * @tx_sg: sg[0] points to iser/iscsi headers 230 * sg[1] optionally points to either of immediate data 231 * unsolicited data-out or control 232 * @num_sge: number sges used on this TX task 233 * @cqe: completion handler 234 * @mapped: Is the task header mapped 235 * @reg_wr: registration WR 236 * @send_wr: send WR 237 * @inv_wr: invalidate WR 238 */ 239 struct iser_tx_desc { 240 struct iser_ctrl iser_header; 241 struct iscsi_hdr iscsi_header; 242 enum iser_desc_type type; 243 u64 dma_addr; 244 struct ib_sge tx_sg[2]; 245 int num_sge; 246 struct ib_cqe cqe; 247 bool mapped; 248 struct ib_reg_wr reg_wr; 249 struct ib_send_wr send_wr; 250 struct ib_send_wr inv_wr; 251 }; 252 253 #define ISER_RX_PAD_SIZE (256 - (ISER_RX_PAYLOAD_SIZE + \ 254 sizeof(u64) + sizeof(struct ib_sge) + \ 255 sizeof(struct ib_cqe))) 256 /** 257 * struct iser_rx_desc - iSER RX descriptor 258 * 259 * @iser_header: iser header 260 * @iscsi_header: iscsi header 261 * @data: received data segment 262 * @dma_addr: receive buffer dma address 263 * @rx_sg: ib_sge of receive buffer 264 * @cqe: completion handler 265 * @pad: for sense data TODO: Modify to maximum sense length supported 266 */ 267 struct iser_rx_desc { 268 struct iser_ctrl iser_header; 269 struct iscsi_hdr iscsi_header; 270 char data[ISER_RECV_DATA_SEG_LEN]; 271 u64 dma_addr; 272 struct ib_sge rx_sg; 273 struct ib_cqe cqe; 274 char pad[ISER_RX_PAD_SIZE]; 275 } __packed; 276 277 /** 278 * struct iser_login_desc - iSER login descriptor 279 * 280 * @req: pointer to login request buffer 281 * @rsp: pointer to login response buffer 282 * @req_dma: DMA address of login request buffer 283 * @rsp_dma: DMA address of login response buffer 284 * @sge: IB sge for login post recv 285 * @cqe: completion handler 286 */ 287 struct iser_login_desc { 288 void *req; 289 void *rsp; 290 u64 req_dma; 291 u64 rsp_dma; 292 struct ib_sge sge; 293 struct ib_cqe cqe; 294 } __packed; 295 296 struct iser_conn; 297 struct ib_conn; 298 299 /** 300 * struct iser_device - iSER device handle 301 * 302 * @ib_device: RDMA device 303 * @pd: Protection Domain for this device 304 * @mr: Global DMA memory region 305 * @event_handler: IB events handle routine 306 * @ig_list: entry in devices list 307 * @refcount: Reference counter, dominated by open iser connections 308 */ 309 struct iser_device { 310 struct ib_device *ib_device; 311 struct ib_pd *pd; 312 struct ib_event_handler event_handler; 313 struct list_head ig_list; 314 int refcount; 315 }; 316 317 /** 318 * struct iser_reg_resources - Fast registration resources 319 * 320 * @mr: memory region 321 * @sig_mr: signature memory region 322 * @mr_valid: is mr valid indicator 323 */ 324 struct iser_reg_resources { 325 struct ib_mr *mr; 326 struct ib_mr *sig_mr; 327 u8 mr_valid:1; 328 }; 329 330 /** 331 * struct iser_fr_desc - Fast registration descriptor 332 * 333 * @list: entry in connection fastreg pool 334 * @rsc: data buffer registration resources 335 * @sig_protected: is region protected indicator 336 * @all_list: first and last list members 337 */ 338 struct iser_fr_desc { 339 struct list_head list; 340 struct iser_reg_resources rsc; 341 bool sig_protected; 342 struct list_head all_list; 343 }; 344 345 /** 346 * struct iser_fr_pool - connection fast registration pool 347 * 348 * @list: list of fastreg descriptors 349 * @lock: protects fastreg pool 350 * @size: size of the pool 351 * @all_list: first and last list members 352 */ 353 struct iser_fr_pool { 354 struct list_head list; 355 spinlock_t lock; 356 int size; 357 struct list_head all_list; 358 }; 359 360 /** 361 * struct ib_conn - Infiniband related objects 362 * 363 * @cma_id: rdma_cm connection maneger handle 364 * @qp: Connection Queue-pair 365 * @cq: Connection completion queue 366 * @cq_size: The number of max outstanding completions 367 * @sig_count: send work request signal count 368 * @device: reference to iser device 369 * @fr_pool: connection fast registration poool 370 * @pi_support: Indicate device T10-PI support 371 * @reg_cqe: completion handler 372 */ 373 struct ib_conn { 374 struct rdma_cm_id *cma_id; 375 struct ib_qp *qp; 376 struct ib_cq *cq; 377 u32 cq_size; 378 u8 sig_count; 379 struct iser_device *device; 380 struct iser_fr_pool fr_pool; 381 bool pi_support; 382 struct ib_cqe reg_cqe; 383 }; 384 385 /** 386 * struct iser_conn - iSER connection context 387 * 388 * @ib_conn: connection RDMA resources 389 * @iscsi_conn: link to matching iscsi connection 390 * @ep: transport handle 391 * @state: connection logical state 392 * @qp_max_recv_dtos: maximum number of data outs, corresponds 393 * to max number of post recvs 394 * @max_cmds: maximum cmds allowed for this connection 395 * @name: connection peer portal 396 * @release_work: deffered work for release job 397 * @state_mutex: protects iser onnection state 398 * @stop_completion: conn_stop completion 399 * @ib_completion: RDMA cleanup completion 400 * @up_completion: connection establishment completed 401 * (state is ISER_CONN_UP) 402 * @conn_list: entry in ig conn list 403 * @login_desc: login descriptor 404 * @rx_descs: rx buffers array (cyclic buffer) 405 * @num_rx_descs: number of rx descriptors 406 * @scsi_sg_tablesize: scsi host sg_tablesize 407 * @pages_per_mr: maximum pages available for registration 408 * @snd_w_inv: connection uses remote invalidation 409 */ 410 struct iser_conn { 411 struct ib_conn ib_conn; 412 struct iscsi_conn *iscsi_conn; 413 struct iscsi_endpoint *ep; 414 enum iser_conn_state state; 415 unsigned qp_max_recv_dtos; 416 u16 max_cmds; 417 char name[ISER_OBJECT_NAME_SIZE]; 418 struct work_struct release_work; 419 struct mutex state_mutex; 420 struct completion stop_completion; 421 struct completion ib_completion; 422 struct completion up_completion; 423 struct list_head conn_list; 424 struct iser_login_desc login_desc; 425 struct iser_rx_desc *rx_descs; 426 u32 num_rx_descs; 427 unsigned short scsi_sg_tablesize; 428 unsigned short pages_per_mr; 429 bool snd_w_inv; 430 }; 431 432 /** 433 * struct iscsi_iser_task - iser task context 434 * 435 * @desc: TX descriptor 436 * @iser_conn: link to iser connection 437 * @status: current task status 438 * @sc: link to scsi command 439 * @command_sent: indicate if command was sent 440 * @dir: iser data direction 441 * @rdma_reg: task rdma registration desc 442 * @data: iser data buffer desc 443 * @prot: iser protection buffer desc 444 */ 445 struct iscsi_iser_task { 446 struct iser_tx_desc desc; 447 struct iser_conn *iser_conn; 448 enum iser_task_status status; 449 struct scsi_cmnd *sc; 450 int command_sent; 451 int dir[ISER_DIRS_NUM]; 452 struct iser_mem_reg rdma_reg[ISER_DIRS_NUM]; 453 struct iser_data_buf data[ISER_DIRS_NUM]; 454 struct iser_data_buf prot[ISER_DIRS_NUM]; 455 }; 456 457 /** 458 * struct iser_global - iSER global context 459 * 460 * @device_list_mutex: protects device_list 461 * @device_list: iser devices global list 462 * @connlist_mutex: protects connlist 463 * @connlist: iser connections global list 464 * @desc_cache: kmem cache for tx dataout 465 */ 466 struct iser_global { 467 struct mutex device_list_mutex; 468 struct list_head device_list; 469 struct mutex connlist_mutex; 470 struct list_head connlist; 471 struct kmem_cache *desc_cache; 472 }; 473 474 extern struct iser_global ig; 475 extern int iser_debug_level; 476 extern bool iser_pi_enable; 477 extern unsigned int iser_max_sectors; 478 extern bool iser_always_reg; 479 480 int iser_send_control(struct iscsi_conn *conn, 481 struct iscsi_task *task); 482 483 int iser_send_command(struct iscsi_conn *conn, 484 struct iscsi_task *task); 485 486 int iser_send_data_out(struct iscsi_conn *conn, 487 struct iscsi_task *task, 488 struct iscsi_data *hdr); 489 490 void iscsi_iser_recv(struct iscsi_conn *conn, 491 struct iscsi_hdr *hdr, 492 char *rx_data, 493 int rx_data_len); 494 495 void iser_conn_init(struct iser_conn *iser_conn); 496 497 void iser_conn_release(struct iser_conn *iser_conn); 498 499 int iser_conn_terminate(struct iser_conn *iser_conn); 500 501 void iser_release_work(struct work_struct *work); 502 503 void iser_err_comp(struct ib_wc *wc, const char *type); 504 void iser_login_rsp(struct ib_cq *cq, struct ib_wc *wc); 505 void iser_task_rsp(struct ib_cq *cq, struct ib_wc *wc); 506 void iser_cmd_comp(struct ib_cq *cq, struct ib_wc *wc); 507 void iser_ctrl_comp(struct ib_cq *cq, struct ib_wc *wc); 508 void iser_dataout_comp(struct ib_cq *cq, struct ib_wc *wc); 509 void iser_reg_comp(struct ib_cq *cq, struct ib_wc *wc); 510 511 void iser_task_rdma_init(struct iscsi_iser_task *task); 512 513 void iser_task_rdma_finalize(struct iscsi_iser_task *task); 514 515 void iser_free_rx_descriptors(struct iser_conn *iser_conn); 516 517 void iser_finalize_rdma_unaligned_sg(struct iscsi_iser_task *iser_task, 518 struct iser_data_buf *mem, 519 enum iser_data_dir cmd_dir); 520 521 int iser_reg_mem_fastreg(struct iscsi_iser_task *task, 522 enum iser_data_dir dir, 523 bool all_imm); 524 void iser_unreg_mem_fastreg(struct iscsi_iser_task *task, 525 enum iser_data_dir dir); 526 527 int iser_connect(struct iser_conn *iser_conn, 528 struct sockaddr *src_addr, 529 struct sockaddr *dst_addr, 530 int non_blocking); 531 532 int iser_post_recvl(struct iser_conn *iser_conn); 533 int iser_post_recvm(struct iser_conn *iser_conn, 534 struct iser_rx_desc *rx_desc); 535 int iser_post_send(struct ib_conn *ib_conn, struct iser_tx_desc *tx_desc, 536 bool signal); 537 538 int iser_dma_map_task_data(struct iscsi_iser_task *iser_task, 539 struct iser_data_buf *data, 540 enum iser_data_dir iser_dir, 541 enum dma_data_direction dma_dir); 542 543 void iser_dma_unmap_task_data(struct iscsi_iser_task *iser_task, 544 struct iser_data_buf *data, 545 enum dma_data_direction dir); 546 547 int iser_initialize_task_headers(struct iscsi_task *task, 548 struct iser_tx_desc *tx_desc); 549 int iser_alloc_rx_descriptors(struct iser_conn *iser_conn, 550 struct iscsi_session *session); 551 int iser_alloc_fastreg_pool(struct ib_conn *ib_conn, 552 unsigned cmds_max, 553 unsigned int size); 554 void iser_free_fastreg_pool(struct ib_conn *ib_conn); 555 u8 iser_check_task_pi_status(struct iscsi_iser_task *iser_task, 556 enum iser_data_dir cmd_dir, sector_t *sector); 557 558 static inline struct iser_conn * 559 to_iser_conn(struct ib_conn *ib_conn) 560 { 561 return container_of(ib_conn, struct iser_conn, ib_conn); 562 } 563 564 static inline struct iser_rx_desc * 565 iser_rx(struct ib_cqe *cqe) 566 { 567 return container_of(cqe, struct iser_rx_desc, cqe); 568 } 569 570 static inline struct iser_tx_desc * 571 iser_tx(struct ib_cqe *cqe) 572 { 573 return container_of(cqe, struct iser_tx_desc, cqe); 574 } 575 576 static inline struct iser_login_desc * 577 iser_login(struct ib_cqe *cqe) 578 { 579 return container_of(cqe, struct iser_login_desc, cqe); 580 } 581 582 #endif 583