1 /* 2 * iSCSI Initiator over iSER Data-Path 3 * 4 * Copyright (C) 2004 Dmitry Yusupov 5 * Copyright (C) 2004 Alex Aizman 6 * Copyright (C) 2005 Mike Christie 7 * Copyright (c) 2005, 2006 Voltaire, Inc. All rights reserved. 8 * Copyright (c) 2013-2014 Mellanox Technologies. All rights reserved. 9 * maintained by openib-general@openib.org 10 * 11 * This software is available to you under a choice of one of two 12 * licenses. You may choose to be licensed under the terms of the GNU 13 * General Public License (GPL) Version 2, available from the file 14 * COPYING in the main directory of this source tree, or the 15 * OpenIB.org BSD license below: 16 * 17 * Redistribution and use in source and binary forms, with or 18 * without modification, are permitted provided that the following 19 * conditions are met: 20 * 21 * - Redistributions of source code must retain the above 22 * copyright notice, this list of conditions and the following 23 * disclaimer. 24 * 25 * - Redistributions in binary form must reproduce the above 26 * copyright notice, this list of conditions and the following 27 * disclaimer in the documentation and/or other materials 28 * provided with the distribution. 29 * 30 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 31 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 32 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 33 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 34 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 35 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 36 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 37 * SOFTWARE. 38 * 39 * Credits: 40 * Christoph Hellwig 41 * FUJITA Tomonori 42 * Arne Redlich 43 * Zhenyu Wang 44 * Modified by: 45 * Erez Zilber 46 */ 47 48 #include <linux/types.h> 49 #include <linux/list.h> 50 #include <linux/hardirq.h> 51 #include <linux/kfifo.h> 52 #include <linux/blkdev.h> 53 #include <linux/init.h> 54 #include <linux/ioctl.h> 55 #include <linux/cdev.h> 56 #include <linux/in.h> 57 #include <linux/net.h> 58 #include <linux/scatterlist.h> 59 #include <linux/delay.h> 60 #include <linux/slab.h> 61 #include <linux/module.h> 62 63 #include <net/sock.h> 64 65 #include <asm/uaccess.h> 66 67 #include <scsi/scsi_cmnd.h> 68 #include <scsi/scsi_device.h> 69 #include <scsi/scsi_eh.h> 70 #include <scsi/scsi_tcq.h> 71 #include <scsi/scsi_host.h> 72 #include <scsi/scsi.h> 73 #include <scsi/scsi_transport_iscsi.h> 74 75 #include "iscsi_iser.h" 76 77 static struct scsi_host_template iscsi_iser_sht; 78 static struct iscsi_transport iscsi_iser_transport; 79 static struct scsi_transport_template *iscsi_iser_scsi_transport; 80 81 static unsigned int iscsi_max_lun = 512; 82 module_param_named(max_lun, iscsi_max_lun, uint, S_IRUGO); 83 84 int iser_debug_level = 0; 85 bool iser_pi_enable = false; 86 int iser_pi_guard = 1; 87 88 MODULE_DESCRIPTION("iSER (iSCSI Extensions for RDMA) Datamover"); 89 MODULE_LICENSE("Dual BSD/GPL"); 90 MODULE_AUTHOR("Alex Nezhinsky, Dan Bar Dov, Or Gerlitz"); 91 MODULE_VERSION(DRV_VER); 92 93 module_param_named(debug_level, iser_debug_level, int, 0644); 94 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0 (default:disabled)"); 95 96 module_param_named(pi_enable, iser_pi_enable, bool, 0644); 97 MODULE_PARM_DESC(pi_enable, "Enable T10-PI offload support (default:disabled)"); 98 99 module_param_named(pi_guard, iser_pi_guard, int, 0644); 100 MODULE_PARM_DESC(pi_guard, "T10-PI guard_type [deprecated]"); 101 102 static struct workqueue_struct *release_wq; 103 struct iser_global ig; 104 105 /* 106 * iscsi_iser_recv() - Process a successfull recv completion 107 * @conn: iscsi connection 108 * @hdr: iscsi header 109 * @rx_data: buffer containing receive data payload 110 * @rx_data_len: length of rx_data 111 * 112 * Notes: In case of data length errors or iscsi PDU completion failures 113 * this routine will signal iscsi layer of connection failure. 114 */ 115 void 116 iscsi_iser_recv(struct iscsi_conn *conn, struct iscsi_hdr *hdr, 117 char *rx_data, int rx_data_len) 118 { 119 int rc = 0; 120 int datalen; 121 int ahslen; 122 123 /* verify PDU length */ 124 datalen = ntoh24(hdr->dlength); 125 if (datalen > rx_data_len || (datalen + 4) < rx_data_len) { 126 iser_err("wrong datalen %d (hdr), %d (IB)\n", 127 datalen, rx_data_len); 128 rc = ISCSI_ERR_DATALEN; 129 goto error; 130 } 131 132 if (datalen != rx_data_len) 133 iser_dbg("aligned datalen (%d) hdr, %d (IB)\n", 134 datalen, rx_data_len); 135 136 /* read AHS */ 137 ahslen = hdr->hlength * 4; 138 139 rc = iscsi_complete_pdu(conn, hdr, rx_data, rx_data_len); 140 if (rc && rc != ISCSI_ERR_NO_SCSI_CMD) 141 goto error; 142 143 return; 144 error: 145 iscsi_conn_failure(conn, rc); 146 } 147 148 /** 149 * iscsi_iser_pdu_alloc() - allocate an iscsi-iser PDU 150 * @task: iscsi task 151 * @opcode: iscsi command opcode 152 * 153 * Netes: This routine can't fail, just assign iscsi task 154 * hdr and max hdr size. 155 */ 156 static int 157 iscsi_iser_pdu_alloc(struct iscsi_task *task, uint8_t opcode) 158 { 159 struct iscsi_iser_task *iser_task = task->dd_data; 160 161 task->hdr = (struct iscsi_hdr *)&iser_task->desc.iscsi_header; 162 task->hdr_max = sizeof(iser_task->desc.iscsi_header); 163 164 return 0; 165 } 166 167 /** 168 * iser_initialize_task_headers() - Initialize task headers 169 * @task: iscsi task 170 * @tx_desc: iser tx descriptor 171 * 172 * Notes: 173 * This routine may race with iser teardown flow for scsi 174 * error handling TMFs. So for TMF we should acquire the 175 * state mutex to avoid dereferencing the IB device which 176 * may have already been terminated. 177 */ 178 int 179 iser_initialize_task_headers(struct iscsi_task *task, 180 struct iser_tx_desc *tx_desc) 181 { 182 struct iser_conn *iser_conn = task->conn->dd_data; 183 struct iser_device *device = iser_conn->ib_conn.device; 184 struct iscsi_iser_task *iser_task = task->dd_data; 185 u64 dma_addr; 186 const bool mgmt_task = !task->sc && !in_interrupt(); 187 int ret = 0; 188 189 if (unlikely(mgmt_task)) 190 mutex_lock(&iser_conn->state_mutex); 191 192 if (unlikely(iser_conn->state != ISER_CONN_UP)) { 193 ret = -ENODEV; 194 goto out; 195 } 196 197 dma_addr = ib_dma_map_single(device->ib_device, (void *)tx_desc, 198 ISER_HEADERS_LEN, DMA_TO_DEVICE); 199 if (ib_dma_mapping_error(device->ib_device, dma_addr)) { 200 ret = -ENOMEM; 201 goto out; 202 } 203 204 tx_desc->dma_addr = dma_addr; 205 tx_desc->tx_sg[0].addr = tx_desc->dma_addr; 206 tx_desc->tx_sg[0].length = ISER_HEADERS_LEN; 207 tx_desc->tx_sg[0].lkey = device->mr->lkey; 208 209 iser_task->iser_conn = iser_conn; 210 out: 211 if (unlikely(mgmt_task)) 212 mutex_unlock(&iser_conn->state_mutex); 213 214 return ret; 215 } 216 217 /** 218 * iscsi_iser_task_init() - Initialize iscsi-iser task 219 * @task: iscsi task 220 * 221 * Initialize the task for the scsi command or mgmt command. 222 * 223 * Return: Returns zero on success or -ENOMEM when failing 224 * to init task headers (dma mapping error). 225 */ 226 static int 227 iscsi_iser_task_init(struct iscsi_task *task) 228 { 229 struct iscsi_iser_task *iser_task = task->dd_data; 230 int ret; 231 232 ret = iser_initialize_task_headers(task, &iser_task->desc); 233 if (ret) { 234 iser_err("Failed to init task %p, err = %d\n", 235 iser_task, ret); 236 return ret; 237 } 238 239 /* mgmt task */ 240 if (!task->sc) 241 return 0; 242 243 iser_task->command_sent = 0; 244 iser_task_rdma_init(iser_task); 245 iser_task->sc = task->sc; 246 247 return 0; 248 } 249 250 /** 251 * iscsi_iser_mtask_xmit() - xmit management (immediate) task 252 * @conn: iscsi connection 253 * @task: task management task 254 * 255 * Notes: 256 * The function can return -EAGAIN in which case caller must 257 * call it again later, or recover. '0' return code means successful 258 * xmit. 259 * 260 **/ 261 static int 262 iscsi_iser_mtask_xmit(struct iscsi_conn *conn, struct iscsi_task *task) 263 { 264 int error = 0; 265 266 iser_dbg("mtask xmit [cid %d itt 0x%x]\n", conn->id, task->itt); 267 268 error = iser_send_control(conn, task); 269 270 /* since iser xmits control with zero copy, tasks can not be recycled 271 * right after sending them. 272 * The recycling scheme is based on whether a response is expected 273 * - if yes, the task is recycled at iscsi_complete_pdu 274 * - if no, the task is recycled at iser_snd_completion 275 */ 276 return error; 277 } 278 279 static int 280 iscsi_iser_task_xmit_unsol_data(struct iscsi_conn *conn, 281 struct iscsi_task *task) 282 { 283 struct iscsi_r2t_info *r2t = &task->unsol_r2t; 284 struct iscsi_data hdr; 285 int error = 0; 286 287 /* Send data-out PDUs while there's still unsolicited data to send */ 288 while (iscsi_task_has_unsol_data(task)) { 289 iscsi_prep_data_out_pdu(task, r2t, &hdr); 290 iser_dbg("Sending data-out: itt 0x%x, data count %d\n", 291 hdr.itt, r2t->data_count); 292 293 /* the buffer description has been passed with the command */ 294 /* Send the command */ 295 error = iser_send_data_out(conn, task, &hdr); 296 if (error) { 297 r2t->datasn--; 298 goto iscsi_iser_task_xmit_unsol_data_exit; 299 } 300 r2t->sent += r2t->data_count; 301 iser_dbg("Need to send %d more as data-out PDUs\n", 302 r2t->data_length - r2t->sent); 303 } 304 305 iscsi_iser_task_xmit_unsol_data_exit: 306 return error; 307 } 308 309 /** 310 * iscsi_iser_task_xmit() - xmit iscsi-iser task 311 * @task: iscsi task 312 * 313 * Return: zero on success or escalates $error on failure. 314 */ 315 static int 316 iscsi_iser_task_xmit(struct iscsi_task *task) 317 { 318 struct iscsi_conn *conn = task->conn; 319 struct iscsi_iser_task *iser_task = task->dd_data; 320 int error = 0; 321 322 if (!task->sc) 323 return iscsi_iser_mtask_xmit(conn, task); 324 325 if (task->sc->sc_data_direction == DMA_TO_DEVICE) { 326 BUG_ON(scsi_bufflen(task->sc) == 0); 327 328 iser_dbg("cmd [itt %x total %d imm %d unsol_data %d\n", 329 task->itt, scsi_bufflen(task->sc), 330 task->imm_count, task->unsol_r2t.data_length); 331 } 332 333 iser_dbg("ctask xmit [cid %d itt 0x%x]\n", 334 conn->id, task->itt); 335 336 /* Send the cmd PDU */ 337 if (!iser_task->command_sent) { 338 error = iser_send_command(conn, task); 339 if (error) 340 goto iscsi_iser_task_xmit_exit; 341 iser_task->command_sent = 1; 342 } 343 344 /* Send unsolicited data-out PDU(s) if necessary */ 345 if (iscsi_task_has_unsol_data(task)) 346 error = iscsi_iser_task_xmit_unsol_data(conn, task); 347 348 iscsi_iser_task_xmit_exit: 349 return error; 350 } 351 352 /** 353 * iscsi_iser_cleanup_task() - cleanup an iscsi-iser task 354 * @task: iscsi task 355 * 356 * Notes: In case the RDMA device is already NULL (might have 357 * been removed in DEVICE_REMOVAL CM event it will bail-out 358 * without doing dma unmapping. 359 */ 360 static void iscsi_iser_cleanup_task(struct iscsi_task *task) 361 { 362 struct iscsi_iser_task *iser_task = task->dd_data; 363 struct iser_tx_desc *tx_desc = &iser_task->desc; 364 struct iser_conn *iser_conn = task->conn->dd_data; 365 struct iser_device *device = iser_conn->ib_conn.device; 366 367 /* DEVICE_REMOVAL event might have already released the device */ 368 if (!device) 369 return; 370 371 ib_dma_unmap_single(device->ib_device, 372 tx_desc->dma_addr, ISER_HEADERS_LEN, DMA_TO_DEVICE); 373 374 /* mgmt tasks do not need special cleanup */ 375 if (!task->sc) 376 return; 377 378 if (iser_task->status == ISER_TASK_STATUS_STARTED) { 379 iser_task->status = ISER_TASK_STATUS_COMPLETED; 380 iser_task_rdma_finalize(iser_task); 381 } 382 } 383 384 /** 385 * iscsi_iser_check_protection() - check protection information status of task. 386 * @task: iscsi task 387 * @sector: error sector if exsists (output) 388 * 389 * Return: zero if no data-integrity errors have occured 390 * 0x1: data-integrity error occured in the guard-block 391 * 0x2: data-integrity error occured in the reference tag 392 * 0x3: data-integrity error occured in the application tag 393 * 394 * In addition the error sector is marked. 395 */ 396 static u8 397 iscsi_iser_check_protection(struct iscsi_task *task, sector_t *sector) 398 { 399 struct iscsi_iser_task *iser_task = task->dd_data; 400 401 if (iser_task->dir[ISER_DIR_IN]) 402 return iser_check_task_pi_status(iser_task, ISER_DIR_IN, 403 sector); 404 else 405 return iser_check_task_pi_status(iser_task, ISER_DIR_OUT, 406 sector); 407 } 408 409 /** 410 * iscsi_iser_conn_create() - create a new iscsi-iser connection 411 * @cls_session: iscsi class connection 412 * @conn_idx: connection index within the session (for MCS) 413 * 414 * Return: iscsi_cls_conn when iscsi_conn_setup succeeds or NULL 415 * otherwise. 416 */ 417 static struct iscsi_cls_conn * 418 iscsi_iser_conn_create(struct iscsi_cls_session *cls_session, 419 uint32_t conn_idx) 420 { 421 struct iscsi_conn *conn; 422 struct iscsi_cls_conn *cls_conn; 423 424 cls_conn = iscsi_conn_setup(cls_session, 0, conn_idx); 425 if (!cls_conn) 426 return NULL; 427 conn = cls_conn->dd_data; 428 429 /* 430 * due to issues with the login code re iser sematics 431 * this not set in iscsi_conn_setup - FIXME 432 */ 433 conn->max_recv_dlength = ISER_RECV_DATA_SEG_LEN; 434 435 return cls_conn; 436 } 437 438 /** 439 * iscsi_iser_conn_bind() - bind iscsi and iser connection structures 440 * @cls_session: iscsi class session 441 * @cls_conn: iscsi class connection 442 * @transport_eph: transport end-point handle 443 * @is_leading: indicate if this is the session leading connection (MCS) 444 * 445 * Return: zero on success, $error if iscsi_conn_bind fails and 446 * -EINVAL in case end-point doesn't exsits anymore or iser connection 447 * state is not UP (teardown already started). 448 */ 449 static int 450 iscsi_iser_conn_bind(struct iscsi_cls_session *cls_session, 451 struct iscsi_cls_conn *cls_conn, 452 uint64_t transport_eph, 453 int is_leading) 454 { 455 struct iscsi_conn *conn = cls_conn->dd_data; 456 struct iser_conn *iser_conn; 457 struct iscsi_endpoint *ep; 458 int error; 459 460 error = iscsi_conn_bind(cls_session, cls_conn, is_leading); 461 if (error) 462 return error; 463 464 /* the transport ep handle comes from user space so it must be 465 * verified against the global ib connections list */ 466 ep = iscsi_lookup_endpoint(transport_eph); 467 if (!ep) { 468 iser_err("can't bind eph %llx\n", 469 (unsigned long long)transport_eph); 470 return -EINVAL; 471 } 472 iser_conn = ep->dd_data; 473 474 mutex_lock(&iser_conn->state_mutex); 475 if (iser_conn->state != ISER_CONN_UP) { 476 error = -EINVAL; 477 iser_err("iser_conn %p state is %d, teardown started\n", 478 iser_conn, iser_conn->state); 479 goto out; 480 } 481 482 error = iser_alloc_rx_descriptors(iser_conn, conn->session); 483 if (error) 484 goto out; 485 486 /* binds the iSER connection retrieved from the previously 487 * connected ep_handle to the iSCSI layer connection. exchanges 488 * connection pointers */ 489 iser_info("binding iscsi conn %p to iser_conn %p\n", conn, iser_conn); 490 491 conn->dd_data = iser_conn; 492 iser_conn->iscsi_conn = conn; 493 494 out: 495 mutex_unlock(&iser_conn->state_mutex); 496 return error; 497 } 498 499 /** 500 * iscsi_iser_conn_start() - start iscsi-iser connection 501 * @cls_conn: iscsi class connection 502 * 503 * Notes: Here iser intialize (or re-initialize) stop_completion as 504 * from this point iscsi must call conn_stop in session/connection 505 * teardown so iser transport must wait for it. 506 */ 507 static int 508 iscsi_iser_conn_start(struct iscsi_cls_conn *cls_conn) 509 { 510 struct iscsi_conn *iscsi_conn; 511 struct iser_conn *iser_conn; 512 513 iscsi_conn = cls_conn->dd_data; 514 iser_conn = iscsi_conn->dd_data; 515 reinit_completion(&iser_conn->stop_completion); 516 517 return iscsi_conn_start(cls_conn); 518 } 519 520 /** 521 * iscsi_iser_conn_stop() - stop iscsi-iser connection 522 * @cls_conn: iscsi class connection 523 * @flag: indicate if recover or terminate (passed as is) 524 * 525 * Notes: Calling iscsi_conn_stop might theoretically race with 526 * DEVICE_REMOVAL event and dereference a previously freed RDMA device 527 * handle, so we call it under iser the state lock to protect against 528 * this kind of race. 529 */ 530 static void 531 iscsi_iser_conn_stop(struct iscsi_cls_conn *cls_conn, int flag) 532 { 533 struct iscsi_conn *conn = cls_conn->dd_data; 534 struct iser_conn *iser_conn = conn->dd_data; 535 536 iser_info("stopping iscsi_conn: %p, iser_conn: %p\n", conn, iser_conn); 537 538 /* 539 * Userspace may have goofed up and not bound the connection or 540 * might have only partially setup the connection. 541 */ 542 if (iser_conn) { 543 mutex_lock(&iser_conn->state_mutex); 544 iser_conn_terminate(iser_conn); 545 iscsi_conn_stop(cls_conn, flag); 546 547 /* unbind */ 548 iser_conn->iscsi_conn = NULL; 549 conn->dd_data = NULL; 550 551 complete(&iser_conn->stop_completion); 552 mutex_unlock(&iser_conn->state_mutex); 553 } else { 554 iscsi_conn_stop(cls_conn, flag); 555 } 556 } 557 558 /** 559 * iscsi_iser_session_destroy() - destroy iscsi-iser session 560 * @cls_session: iscsi class session 561 * 562 * Removes and free iscsi host. 563 */ 564 static void 565 iscsi_iser_session_destroy(struct iscsi_cls_session *cls_session) 566 { 567 struct Scsi_Host *shost = iscsi_session_to_shost(cls_session); 568 569 iscsi_session_teardown(cls_session); 570 iscsi_host_remove(shost); 571 iscsi_host_free(shost); 572 } 573 574 static inline unsigned int 575 iser_dif_prot_caps(int prot_caps) 576 { 577 return ((prot_caps & IB_PROT_T10DIF_TYPE_1) ? 578 SHOST_DIF_TYPE1_PROTECTION | SHOST_DIX_TYPE0_PROTECTION | 579 SHOST_DIX_TYPE1_PROTECTION : 0) | 580 ((prot_caps & IB_PROT_T10DIF_TYPE_2) ? 581 SHOST_DIF_TYPE2_PROTECTION | SHOST_DIX_TYPE2_PROTECTION : 0) | 582 ((prot_caps & IB_PROT_T10DIF_TYPE_3) ? 583 SHOST_DIF_TYPE3_PROTECTION | SHOST_DIX_TYPE3_PROTECTION : 0); 584 } 585 586 /** 587 * iscsi_iser_session_create() - create an iscsi-iser session 588 * @ep: iscsi end-point handle 589 * @cmds_max: maximum commands in this session 590 * @qdepth: session command queue depth 591 * @initial_cmdsn: initiator command sequnce number 592 * 593 * Allocates and adds a scsi host, expose DIF supprot if 594 * exists, and sets up an iscsi session. 595 */ 596 static struct iscsi_cls_session * 597 iscsi_iser_session_create(struct iscsi_endpoint *ep, 598 uint16_t cmds_max, uint16_t qdepth, 599 uint32_t initial_cmdsn) 600 { 601 struct iscsi_cls_session *cls_session; 602 struct iscsi_session *session; 603 struct Scsi_Host *shost; 604 struct iser_conn *iser_conn = NULL; 605 struct ib_conn *ib_conn; 606 u16 max_cmds; 607 608 shost = iscsi_host_alloc(&iscsi_iser_sht, 0, 0); 609 if (!shost) 610 return NULL; 611 shost->transportt = iscsi_iser_scsi_transport; 612 shost->cmd_per_lun = qdepth; 613 shost->max_lun = iscsi_max_lun; 614 shost->max_id = 0; 615 shost->max_channel = 0; 616 shost->max_cmd_len = 16; 617 618 /* 619 * older userspace tools (before 2.0-870) did not pass us 620 * the leading conn's ep so this will be NULL; 621 */ 622 if (ep) { 623 iser_conn = ep->dd_data; 624 max_cmds = iser_conn->max_cmds; 625 626 mutex_lock(&iser_conn->state_mutex); 627 if (iser_conn->state != ISER_CONN_UP) { 628 iser_err("iser conn %p already started teardown\n", 629 iser_conn); 630 mutex_unlock(&iser_conn->state_mutex); 631 goto free_host; 632 } 633 634 ib_conn = &iser_conn->ib_conn; 635 if (ib_conn->pi_support) { 636 u32 sig_caps = ib_conn->device->dev_attr.sig_prot_cap; 637 638 scsi_host_set_prot(shost, iser_dif_prot_caps(sig_caps)); 639 scsi_host_set_guard(shost, SHOST_DIX_GUARD_IP | 640 SHOST_DIX_GUARD_CRC); 641 } 642 643 /* 644 * Limit the sg_tablesize and max_sectors based on the device 645 * max fastreg page list length. 646 */ 647 shost->sg_tablesize = min_t(unsigned short, shost->sg_tablesize, 648 ib_conn->device->dev_attr.max_fast_reg_page_list_len); 649 shost->max_sectors = min_t(unsigned int, 650 1024, (shost->sg_tablesize * PAGE_SIZE) >> 9); 651 652 if (iscsi_host_add(shost, 653 ib_conn->device->ib_device->dma_device)) { 654 mutex_unlock(&iser_conn->state_mutex); 655 goto free_host; 656 } 657 mutex_unlock(&iser_conn->state_mutex); 658 } else { 659 max_cmds = ISER_DEF_XMIT_CMDS_MAX; 660 if (iscsi_host_add(shost, NULL)) 661 goto free_host; 662 } 663 664 if (cmds_max > max_cmds) { 665 iser_info("cmds_max changed from %u to %u\n", 666 cmds_max, max_cmds); 667 cmds_max = max_cmds; 668 } 669 670 cls_session = iscsi_session_setup(&iscsi_iser_transport, shost, 671 cmds_max, 0, 672 sizeof(struct iscsi_iser_task), 673 initial_cmdsn, 0); 674 if (!cls_session) 675 goto remove_host; 676 session = cls_session->dd_data; 677 678 shost->can_queue = session->scsi_cmds_max; 679 return cls_session; 680 681 remove_host: 682 iscsi_host_remove(shost); 683 free_host: 684 iscsi_host_free(shost); 685 return NULL; 686 } 687 688 static int 689 iscsi_iser_set_param(struct iscsi_cls_conn *cls_conn, 690 enum iscsi_param param, char *buf, int buflen) 691 { 692 int value; 693 694 switch (param) { 695 case ISCSI_PARAM_MAX_RECV_DLENGTH: 696 /* TBD */ 697 break; 698 case ISCSI_PARAM_HDRDGST_EN: 699 sscanf(buf, "%d", &value); 700 if (value) { 701 iser_err("DataDigest wasn't negotiated to None\n"); 702 return -EPROTO; 703 } 704 break; 705 case ISCSI_PARAM_DATADGST_EN: 706 sscanf(buf, "%d", &value); 707 if (value) { 708 iser_err("DataDigest wasn't negotiated to None\n"); 709 return -EPROTO; 710 } 711 break; 712 case ISCSI_PARAM_IFMARKER_EN: 713 sscanf(buf, "%d", &value); 714 if (value) { 715 iser_err("IFMarker wasn't negotiated to No\n"); 716 return -EPROTO; 717 } 718 break; 719 case ISCSI_PARAM_OFMARKER_EN: 720 sscanf(buf, "%d", &value); 721 if (value) { 722 iser_err("OFMarker wasn't negotiated to No\n"); 723 return -EPROTO; 724 } 725 break; 726 default: 727 return iscsi_set_param(cls_conn, param, buf, buflen); 728 } 729 730 return 0; 731 } 732 733 /** 734 * iscsi_iser_set_param() - set class connection parameter 735 * @cls_conn: iscsi class connection 736 * @stats: iscsi stats to output 737 * 738 * Output connection statistics. 739 */ 740 static void 741 iscsi_iser_conn_get_stats(struct iscsi_cls_conn *cls_conn, struct iscsi_stats *stats) 742 { 743 struct iscsi_conn *conn = cls_conn->dd_data; 744 745 stats->txdata_octets = conn->txdata_octets; 746 stats->rxdata_octets = conn->rxdata_octets; 747 stats->scsicmd_pdus = conn->scsicmd_pdus_cnt; 748 stats->dataout_pdus = conn->dataout_pdus_cnt; 749 stats->scsirsp_pdus = conn->scsirsp_pdus_cnt; 750 stats->datain_pdus = conn->datain_pdus_cnt; /* always 0 */ 751 stats->r2t_pdus = conn->r2t_pdus_cnt; /* always 0 */ 752 stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt; 753 stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt; 754 stats->custom_length = 4; 755 strcpy(stats->custom[0].desc, "qp_tx_queue_full"); 756 stats->custom[0].value = 0; /* TB iser_conn->qp_tx_queue_full; */ 757 strcpy(stats->custom[1].desc, "fmr_map_not_avail"); 758 stats->custom[1].value = 0; /* TB iser_conn->fmr_map_not_avail */; 759 strcpy(stats->custom[2].desc, "eh_abort_cnt"); 760 stats->custom[2].value = conn->eh_abort_cnt; 761 strcpy(stats->custom[3].desc, "fmr_unalign_cnt"); 762 stats->custom[3].value = conn->fmr_unalign_cnt; 763 } 764 765 static int iscsi_iser_get_ep_param(struct iscsi_endpoint *ep, 766 enum iscsi_param param, char *buf) 767 { 768 struct iser_conn *iser_conn = ep->dd_data; 769 int len; 770 771 switch (param) { 772 case ISCSI_PARAM_CONN_PORT: 773 case ISCSI_PARAM_CONN_ADDRESS: 774 if (!iser_conn || !iser_conn->ib_conn.cma_id) 775 return -ENOTCONN; 776 777 return iscsi_conn_get_addr_param((struct sockaddr_storage *) 778 &iser_conn->ib_conn.cma_id->route.addr.dst_addr, 779 param, buf); 780 break; 781 default: 782 return -ENOSYS; 783 } 784 785 return len; 786 } 787 788 /** 789 * iscsi_iser_ep_connect() - Initiate iSER connection establishment 790 * @shost: scsi_host 791 * @dst_addr: destination address 792 * @non-blocking: indicate if routine can block 793 * 794 * Allocate an iscsi endpoint, an iser_conn structure and bind them. 795 * After that start RDMA connection establishment via rdma_cm. We 796 * don't allocate iser_conn embedded in iscsi_endpoint since in teardown 797 * the endpoint will be destroyed at ep_disconnect while iser_conn will 798 * cleanup its resources asynchronuously. 799 * 800 * Return: iscsi_endpoint created by iscsi layer or ERR_PTR(error) 801 * if fails. 802 */ 803 static struct iscsi_endpoint * 804 iscsi_iser_ep_connect(struct Scsi_Host *shost, struct sockaddr *dst_addr, 805 int non_blocking) 806 { 807 int err; 808 struct iser_conn *iser_conn; 809 struct iscsi_endpoint *ep; 810 811 ep = iscsi_create_endpoint(0); 812 if (!ep) 813 return ERR_PTR(-ENOMEM); 814 815 iser_conn = kzalloc(sizeof(*iser_conn), GFP_KERNEL); 816 if (!iser_conn) { 817 err = -ENOMEM; 818 goto failure; 819 } 820 821 ep->dd_data = iser_conn; 822 iser_conn->ep = ep; 823 iser_conn_init(iser_conn); 824 825 err = iser_connect(iser_conn, NULL, dst_addr, non_blocking); 826 if (err) 827 goto failure; 828 829 return ep; 830 failure: 831 iscsi_destroy_endpoint(ep); 832 return ERR_PTR(err); 833 } 834 835 /** 836 * iscsi_iser_ep_poll() - poll for iser connection establishment to complete 837 * @ep: iscsi endpoint (created at ep_connect) 838 * @timeout_ms: polling timeout allowed in ms. 839 * 840 * This routine boils down to waiting for up_completion signaling 841 * that cma_id got CONNECTED event. 842 * 843 * Return: 1 if succeeded in connection establishment, 0 if timeout expired 844 * (libiscsi will retry will kick in) or -1 if interrupted by signal 845 * or more likely iser connection state transitioned to TEMINATING or 846 * DOWN during the wait period. 847 */ 848 static int 849 iscsi_iser_ep_poll(struct iscsi_endpoint *ep, int timeout_ms) 850 { 851 struct iser_conn *iser_conn; 852 int rc; 853 854 iser_conn = ep->dd_data; 855 rc = wait_for_completion_interruptible_timeout(&iser_conn->up_completion, 856 msecs_to_jiffies(timeout_ms)); 857 /* if conn establishment failed, return error code to iscsi */ 858 if (rc == 0) { 859 mutex_lock(&iser_conn->state_mutex); 860 if (iser_conn->state == ISER_CONN_TERMINATING || 861 iser_conn->state == ISER_CONN_DOWN) 862 rc = -1; 863 mutex_unlock(&iser_conn->state_mutex); 864 } 865 866 iser_info("ib conn %p rc = %d\n", iser_conn, rc); 867 868 if (rc > 0) 869 return 1; /* success, this is the equivalent of POLLOUT */ 870 else if (!rc) 871 return 0; /* timeout */ 872 else 873 return rc; /* signal */ 874 } 875 876 /** 877 * iscsi_iser_ep_disconnect() - Initiate connection teardown process 878 * @ep: iscsi endpoint handle 879 * 880 * This routine is not blocked by iser and RDMA termination process 881 * completion as we queue a deffered work for iser/RDMA destruction 882 * and cleanup or actually call it immediately in case we didn't pass 883 * iscsi conn bind/start stage, thus it is safe. 884 */ 885 static void 886 iscsi_iser_ep_disconnect(struct iscsi_endpoint *ep) 887 { 888 struct iser_conn *iser_conn; 889 890 iser_conn = ep->dd_data; 891 iser_info("ep %p iser conn %p state %d\n", 892 ep, iser_conn, iser_conn->state); 893 894 mutex_lock(&iser_conn->state_mutex); 895 iser_conn_terminate(iser_conn); 896 897 /* 898 * if iser_conn and iscsi_conn are bound, we must wait for 899 * iscsi_conn_stop and flush errors completion before freeing 900 * the iser resources. Otherwise we are safe to free resources 901 * immediately. 902 */ 903 if (iser_conn->iscsi_conn) { 904 INIT_WORK(&iser_conn->release_work, iser_release_work); 905 queue_work(release_wq, &iser_conn->release_work); 906 mutex_unlock(&iser_conn->state_mutex); 907 } else { 908 iser_conn->state = ISER_CONN_DOWN; 909 mutex_unlock(&iser_conn->state_mutex); 910 iser_conn_release(iser_conn); 911 } 912 iscsi_destroy_endpoint(ep); 913 } 914 915 static umode_t iser_attr_is_visible(int param_type, int param) 916 { 917 switch (param_type) { 918 case ISCSI_HOST_PARAM: 919 switch (param) { 920 case ISCSI_HOST_PARAM_NETDEV_NAME: 921 case ISCSI_HOST_PARAM_HWADDRESS: 922 case ISCSI_HOST_PARAM_INITIATOR_NAME: 923 return S_IRUGO; 924 default: 925 return 0; 926 } 927 case ISCSI_PARAM: 928 switch (param) { 929 case ISCSI_PARAM_MAX_RECV_DLENGTH: 930 case ISCSI_PARAM_MAX_XMIT_DLENGTH: 931 case ISCSI_PARAM_HDRDGST_EN: 932 case ISCSI_PARAM_DATADGST_EN: 933 case ISCSI_PARAM_CONN_ADDRESS: 934 case ISCSI_PARAM_CONN_PORT: 935 case ISCSI_PARAM_EXP_STATSN: 936 case ISCSI_PARAM_PERSISTENT_ADDRESS: 937 case ISCSI_PARAM_PERSISTENT_PORT: 938 case ISCSI_PARAM_PING_TMO: 939 case ISCSI_PARAM_RECV_TMO: 940 case ISCSI_PARAM_INITIAL_R2T_EN: 941 case ISCSI_PARAM_MAX_R2T: 942 case ISCSI_PARAM_IMM_DATA_EN: 943 case ISCSI_PARAM_FIRST_BURST: 944 case ISCSI_PARAM_MAX_BURST: 945 case ISCSI_PARAM_PDU_INORDER_EN: 946 case ISCSI_PARAM_DATASEQ_INORDER_EN: 947 case ISCSI_PARAM_TARGET_NAME: 948 case ISCSI_PARAM_TPGT: 949 case ISCSI_PARAM_USERNAME: 950 case ISCSI_PARAM_PASSWORD: 951 case ISCSI_PARAM_USERNAME_IN: 952 case ISCSI_PARAM_PASSWORD_IN: 953 case ISCSI_PARAM_FAST_ABORT: 954 case ISCSI_PARAM_ABORT_TMO: 955 case ISCSI_PARAM_LU_RESET_TMO: 956 case ISCSI_PARAM_TGT_RESET_TMO: 957 case ISCSI_PARAM_IFACE_NAME: 958 case ISCSI_PARAM_INITIATOR_NAME: 959 case ISCSI_PARAM_DISCOVERY_SESS: 960 return S_IRUGO; 961 default: 962 return 0; 963 } 964 } 965 966 return 0; 967 } 968 969 static struct scsi_host_template iscsi_iser_sht = { 970 .module = THIS_MODULE, 971 .name = "iSCSI Initiator over iSER", 972 .queuecommand = iscsi_queuecommand, 973 .change_queue_depth = scsi_change_queue_depth, 974 .sg_tablesize = ISCSI_ISER_SG_TABLESIZE, 975 .max_sectors = 1024, 976 .cmd_per_lun = ISER_DEF_CMD_PER_LUN, 977 .eh_abort_handler = iscsi_eh_abort, 978 .eh_device_reset_handler= iscsi_eh_device_reset, 979 .eh_target_reset_handler = iscsi_eh_recover_target, 980 .target_alloc = iscsi_target_alloc, 981 .use_clustering = DISABLE_CLUSTERING, 982 .proc_name = "iscsi_iser", 983 .this_id = -1, 984 .track_queue_depth = 1, 985 }; 986 987 static struct iscsi_transport iscsi_iser_transport = { 988 .owner = THIS_MODULE, 989 .name = "iser", 990 .caps = CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_TEXT_NEGO, 991 /* session management */ 992 .create_session = iscsi_iser_session_create, 993 .destroy_session = iscsi_iser_session_destroy, 994 /* connection management */ 995 .create_conn = iscsi_iser_conn_create, 996 .bind_conn = iscsi_iser_conn_bind, 997 .destroy_conn = iscsi_conn_teardown, 998 .attr_is_visible = iser_attr_is_visible, 999 .set_param = iscsi_iser_set_param, 1000 .get_conn_param = iscsi_conn_get_param, 1001 .get_ep_param = iscsi_iser_get_ep_param, 1002 .get_session_param = iscsi_session_get_param, 1003 .start_conn = iscsi_iser_conn_start, 1004 .stop_conn = iscsi_iser_conn_stop, 1005 /* iscsi host params */ 1006 .get_host_param = iscsi_host_get_param, 1007 .set_host_param = iscsi_host_set_param, 1008 /* IO */ 1009 .send_pdu = iscsi_conn_send_pdu, 1010 .get_stats = iscsi_iser_conn_get_stats, 1011 .init_task = iscsi_iser_task_init, 1012 .xmit_task = iscsi_iser_task_xmit, 1013 .cleanup_task = iscsi_iser_cleanup_task, 1014 .alloc_pdu = iscsi_iser_pdu_alloc, 1015 .check_protection = iscsi_iser_check_protection, 1016 /* recovery */ 1017 .session_recovery_timedout = iscsi_session_recovery_timedout, 1018 1019 .ep_connect = iscsi_iser_ep_connect, 1020 .ep_poll = iscsi_iser_ep_poll, 1021 .ep_disconnect = iscsi_iser_ep_disconnect 1022 }; 1023 1024 static int __init iser_init(void) 1025 { 1026 int err; 1027 1028 iser_dbg("Starting iSER datamover...\n"); 1029 1030 if (iscsi_max_lun < 1) { 1031 iser_err("Invalid max_lun value of %u\n", iscsi_max_lun); 1032 return -EINVAL; 1033 } 1034 1035 memset(&ig, 0, sizeof(struct iser_global)); 1036 1037 ig.desc_cache = kmem_cache_create("iser_descriptors", 1038 sizeof(struct iser_tx_desc), 1039 0, SLAB_HWCACHE_ALIGN, 1040 NULL); 1041 if (ig.desc_cache == NULL) 1042 return -ENOMEM; 1043 1044 /* device init is called only after the first addr resolution */ 1045 mutex_init(&ig.device_list_mutex); 1046 INIT_LIST_HEAD(&ig.device_list); 1047 mutex_init(&ig.connlist_mutex); 1048 INIT_LIST_HEAD(&ig.connlist); 1049 1050 release_wq = alloc_workqueue("release workqueue", 0, 0); 1051 if (!release_wq) { 1052 iser_err("failed to allocate release workqueue\n"); 1053 return -ENOMEM; 1054 } 1055 1056 iscsi_iser_scsi_transport = iscsi_register_transport( 1057 &iscsi_iser_transport); 1058 if (!iscsi_iser_scsi_transport) { 1059 iser_err("iscsi_register_transport failed\n"); 1060 err = -EINVAL; 1061 goto register_transport_failure; 1062 } 1063 1064 return 0; 1065 1066 register_transport_failure: 1067 kmem_cache_destroy(ig.desc_cache); 1068 1069 return err; 1070 } 1071 1072 static void __exit iser_exit(void) 1073 { 1074 struct iser_conn *iser_conn, *n; 1075 int connlist_empty; 1076 1077 iser_dbg("Removing iSER datamover...\n"); 1078 destroy_workqueue(release_wq); 1079 1080 mutex_lock(&ig.connlist_mutex); 1081 connlist_empty = list_empty(&ig.connlist); 1082 mutex_unlock(&ig.connlist_mutex); 1083 1084 if (!connlist_empty) { 1085 iser_err("Error cleanup stage completed but we still have iser " 1086 "connections, destroying them anyway.\n"); 1087 list_for_each_entry_safe(iser_conn, n, &ig.connlist, 1088 conn_list) { 1089 iser_conn_release(iser_conn); 1090 } 1091 } 1092 1093 iscsi_unregister_transport(&iscsi_iser_transport); 1094 kmem_cache_destroy(ig.desc_cache); 1095 } 1096 1097 module_init(iser_init); 1098 module_exit(iser_exit); 1099