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 MODULE_DESCRIPTION("iSER (iSCSI Extensions for RDMA) Datamover"); 78 MODULE_LICENSE("Dual BSD/GPL"); 79 MODULE_AUTHOR("Alex Nezhinsky, Dan Bar Dov, Or Gerlitz"); 80 MODULE_VERSION(DRV_VER); 81 82 static struct scsi_host_template iscsi_iser_sht; 83 static struct iscsi_transport iscsi_iser_transport; 84 static struct scsi_transport_template *iscsi_iser_scsi_transport; 85 static struct workqueue_struct *release_wq; 86 struct iser_global ig; 87 88 int iser_debug_level = 0; 89 module_param_named(debug_level, iser_debug_level, int, S_IRUGO | S_IWUSR); 90 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0 (default:disabled)"); 91 92 static unsigned int iscsi_max_lun = 512; 93 module_param_named(max_lun, iscsi_max_lun, uint, S_IRUGO); 94 MODULE_PARM_DESC(max_lun, "Max LUNs to allow per session (default:512"); 95 96 unsigned int iser_max_sectors = ISER_DEF_MAX_SECTORS; 97 module_param_named(max_sectors, iser_max_sectors, uint, S_IRUGO | S_IWUSR); 98 MODULE_PARM_DESC(max_sectors, "Max number of sectors in a single scsi command (default:1024"); 99 100 bool iser_pi_enable = false; 101 module_param_named(pi_enable, iser_pi_enable, bool, S_IRUGO); 102 MODULE_PARM_DESC(pi_enable, "Enable T10-PI offload support (default:disabled)"); 103 104 int iser_pi_guard; 105 module_param_named(pi_guard, iser_pi_guard, int, S_IRUGO); 106 MODULE_PARM_DESC(pi_guard, "T10-PI guard_type [deprecated]"); 107 108 /* 109 * iscsi_iser_recv() - Process a successfull recv completion 110 * @conn: iscsi connection 111 * @hdr: iscsi header 112 * @rx_data: buffer containing receive data payload 113 * @rx_data_len: length of rx_data 114 * 115 * Notes: In case of data length errors or iscsi PDU completion failures 116 * this routine will signal iscsi layer of connection failure. 117 */ 118 void 119 iscsi_iser_recv(struct iscsi_conn *conn, struct iscsi_hdr *hdr, 120 char *rx_data, int rx_data_len) 121 { 122 int rc = 0; 123 int datalen; 124 int ahslen; 125 126 /* verify PDU length */ 127 datalen = ntoh24(hdr->dlength); 128 if (datalen > rx_data_len || (datalen + 4) < rx_data_len) { 129 iser_err("wrong datalen %d (hdr), %d (IB)\n", 130 datalen, rx_data_len); 131 rc = ISCSI_ERR_DATALEN; 132 goto error; 133 } 134 135 if (datalen != rx_data_len) 136 iser_dbg("aligned datalen (%d) hdr, %d (IB)\n", 137 datalen, rx_data_len); 138 139 /* read AHS */ 140 ahslen = hdr->hlength * 4; 141 142 rc = iscsi_complete_pdu(conn, hdr, rx_data, rx_data_len); 143 if (rc && rc != ISCSI_ERR_NO_SCSI_CMD) 144 goto error; 145 146 return; 147 error: 148 iscsi_conn_failure(conn, rc); 149 } 150 151 /** 152 * iscsi_iser_pdu_alloc() - allocate an iscsi-iser PDU 153 * @task: iscsi task 154 * @opcode: iscsi command opcode 155 * 156 * Netes: This routine can't fail, just assign iscsi task 157 * hdr and max hdr size. 158 */ 159 static int 160 iscsi_iser_pdu_alloc(struct iscsi_task *task, uint8_t opcode) 161 { 162 struct iscsi_iser_task *iser_task = task->dd_data; 163 164 task->hdr = (struct iscsi_hdr *)&iser_task->desc.iscsi_header; 165 task->hdr_max = sizeof(iser_task->desc.iscsi_header); 166 167 return 0; 168 } 169 170 /** 171 * iser_initialize_task_headers() - Initialize task headers 172 * @task: iscsi task 173 * @tx_desc: iser tx descriptor 174 * 175 * Notes: 176 * This routine may race with iser teardown flow for scsi 177 * error handling TMFs. So for TMF we should acquire the 178 * state mutex to avoid dereferencing the IB device which 179 * may have already been terminated. 180 */ 181 int 182 iser_initialize_task_headers(struct iscsi_task *task, 183 struct iser_tx_desc *tx_desc) 184 { 185 struct iser_conn *iser_conn = task->conn->dd_data; 186 struct iser_device *device = iser_conn->ib_conn.device; 187 struct iscsi_iser_task *iser_task = task->dd_data; 188 u64 dma_addr; 189 const bool mgmt_task = !task->sc && !in_interrupt(); 190 int ret = 0; 191 192 if (unlikely(mgmt_task)) 193 mutex_lock(&iser_conn->state_mutex); 194 195 if (unlikely(iser_conn->state != ISER_CONN_UP)) { 196 ret = -ENODEV; 197 goto out; 198 } 199 200 dma_addr = ib_dma_map_single(device->ib_device, (void *)tx_desc, 201 ISER_HEADERS_LEN, DMA_TO_DEVICE); 202 if (ib_dma_mapping_error(device->ib_device, dma_addr)) { 203 ret = -ENOMEM; 204 goto out; 205 } 206 207 tx_desc->wr_idx = 0; 208 tx_desc->mapped = true; 209 tx_desc->dma_addr = dma_addr; 210 tx_desc->tx_sg[0].addr = tx_desc->dma_addr; 211 tx_desc->tx_sg[0].length = ISER_HEADERS_LEN; 212 tx_desc->tx_sg[0].lkey = device->pd->local_dma_lkey; 213 214 iser_task->iser_conn = iser_conn; 215 out: 216 if (unlikely(mgmt_task)) 217 mutex_unlock(&iser_conn->state_mutex); 218 219 return ret; 220 } 221 222 /** 223 * iscsi_iser_task_init() - Initialize iscsi-iser task 224 * @task: iscsi task 225 * 226 * Initialize the task for the scsi command or mgmt command. 227 * 228 * Return: Returns zero on success or -ENOMEM when failing 229 * to init task headers (dma mapping error). 230 */ 231 static int 232 iscsi_iser_task_init(struct iscsi_task *task) 233 { 234 struct iscsi_iser_task *iser_task = task->dd_data; 235 int ret; 236 237 ret = iser_initialize_task_headers(task, &iser_task->desc); 238 if (ret) { 239 iser_err("Failed to init task %p, err = %d\n", 240 iser_task, ret); 241 return ret; 242 } 243 244 /* mgmt task */ 245 if (!task->sc) 246 return 0; 247 248 iser_task->command_sent = 0; 249 iser_task_rdma_init(iser_task); 250 iser_task->sc = task->sc; 251 252 return 0; 253 } 254 255 /** 256 * iscsi_iser_mtask_xmit() - xmit management (immediate) task 257 * @conn: iscsi connection 258 * @task: task management task 259 * 260 * Notes: 261 * The function can return -EAGAIN in which case caller must 262 * call it again later, or recover. '0' return code means successful 263 * xmit. 264 * 265 **/ 266 static int 267 iscsi_iser_mtask_xmit(struct iscsi_conn *conn, struct iscsi_task *task) 268 { 269 int error = 0; 270 271 iser_dbg("mtask xmit [cid %d itt 0x%x]\n", conn->id, task->itt); 272 273 error = iser_send_control(conn, task); 274 275 /* since iser xmits control with zero copy, tasks can not be recycled 276 * right after sending them. 277 * The recycling scheme is based on whether a response is expected 278 * - if yes, the task is recycled at iscsi_complete_pdu 279 * - if no, the task is recycled at iser_snd_completion 280 */ 281 return error; 282 } 283 284 static int 285 iscsi_iser_task_xmit_unsol_data(struct iscsi_conn *conn, 286 struct iscsi_task *task) 287 { 288 struct iscsi_r2t_info *r2t = &task->unsol_r2t; 289 struct iscsi_data hdr; 290 int error = 0; 291 292 /* Send data-out PDUs while there's still unsolicited data to send */ 293 while (iscsi_task_has_unsol_data(task)) { 294 iscsi_prep_data_out_pdu(task, r2t, &hdr); 295 iser_dbg("Sending data-out: itt 0x%x, data count %d\n", 296 hdr.itt, r2t->data_count); 297 298 /* the buffer description has been passed with the command */ 299 /* Send the command */ 300 error = iser_send_data_out(conn, task, &hdr); 301 if (error) { 302 r2t->datasn--; 303 goto iscsi_iser_task_xmit_unsol_data_exit; 304 } 305 r2t->sent += r2t->data_count; 306 iser_dbg("Need to send %d more as data-out PDUs\n", 307 r2t->data_length - r2t->sent); 308 } 309 310 iscsi_iser_task_xmit_unsol_data_exit: 311 return error; 312 } 313 314 /** 315 * iscsi_iser_task_xmit() - xmit iscsi-iser task 316 * @task: iscsi task 317 * 318 * Return: zero on success or escalates $error on failure. 319 */ 320 static int 321 iscsi_iser_task_xmit(struct iscsi_task *task) 322 { 323 struct iscsi_conn *conn = task->conn; 324 struct iscsi_iser_task *iser_task = task->dd_data; 325 int error = 0; 326 327 if (!task->sc) 328 return iscsi_iser_mtask_xmit(conn, task); 329 330 if (task->sc->sc_data_direction == DMA_TO_DEVICE) { 331 BUG_ON(scsi_bufflen(task->sc) == 0); 332 333 iser_dbg("cmd [itt %x total %d imm %d unsol_data %d\n", 334 task->itt, scsi_bufflen(task->sc), 335 task->imm_count, task->unsol_r2t.data_length); 336 } 337 338 iser_dbg("ctask xmit [cid %d itt 0x%x]\n", 339 conn->id, task->itt); 340 341 /* Send the cmd PDU */ 342 if (!iser_task->command_sent) { 343 error = iser_send_command(conn, task); 344 if (error) 345 goto iscsi_iser_task_xmit_exit; 346 iser_task->command_sent = 1; 347 } 348 349 /* Send unsolicited data-out PDU(s) if necessary */ 350 if (iscsi_task_has_unsol_data(task)) 351 error = iscsi_iser_task_xmit_unsol_data(conn, task); 352 353 iscsi_iser_task_xmit_exit: 354 return error; 355 } 356 357 /** 358 * iscsi_iser_cleanup_task() - cleanup an iscsi-iser task 359 * @task: iscsi task 360 * 361 * Notes: In case the RDMA device is already NULL (might have 362 * been removed in DEVICE_REMOVAL CM event it will bail-out 363 * without doing dma unmapping. 364 */ 365 static void iscsi_iser_cleanup_task(struct iscsi_task *task) 366 { 367 struct iscsi_iser_task *iser_task = task->dd_data; 368 struct iser_tx_desc *tx_desc = &iser_task->desc; 369 struct iser_conn *iser_conn = task->conn->dd_data; 370 struct iser_device *device = iser_conn->ib_conn.device; 371 372 /* DEVICE_REMOVAL event might have already released the device */ 373 if (!device) 374 return; 375 376 if (likely(tx_desc->mapped)) { 377 ib_dma_unmap_single(device->ib_device, tx_desc->dma_addr, 378 ISER_HEADERS_LEN, DMA_TO_DEVICE); 379 tx_desc->mapped = false; 380 } 381 382 /* mgmt tasks do not need special cleanup */ 383 if (!task->sc) 384 return; 385 386 if (iser_task->status == ISER_TASK_STATUS_STARTED) { 387 iser_task->status = ISER_TASK_STATUS_COMPLETED; 388 iser_task_rdma_finalize(iser_task); 389 } 390 } 391 392 /** 393 * iscsi_iser_check_protection() - check protection information status of task. 394 * @task: iscsi task 395 * @sector: error sector if exsists (output) 396 * 397 * Return: zero if no data-integrity errors have occured 398 * 0x1: data-integrity error occured in the guard-block 399 * 0x2: data-integrity error occured in the reference tag 400 * 0x3: data-integrity error occured in the application tag 401 * 402 * In addition the error sector is marked. 403 */ 404 static u8 405 iscsi_iser_check_protection(struct iscsi_task *task, sector_t *sector) 406 { 407 struct iscsi_iser_task *iser_task = task->dd_data; 408 409 if (iser_task->dir[ISER_DIR_IN]) 410 return iser_check_task_pi_status(iser_task, ISER_DIR_IN, 411 sector); 412 else 413 return iser_check_task_pi_status(iser_task, ISER_DIR_OUT, 414 sector); 415 } 416 417 /** 418 * iscsi_iser_conn_create() - create a new iscsi-iser connection 419 * @cls_session: iscsi class connection 420 * @conn_idx: connection index within the session (for MCS) 421 * 422 * Return: iscsi_cls_conn when iscsi_conn_setup succeeds or NULL 423 * otherwise. 424 */ 425 static struct iscsi_cls_conn * 426 iscsi_iser_conn_create(struct iscsi_cls_session *cls_session, 427 uint32_t conn_idx) 428 { 429 struct iscsi_conn *conn; 430 struct iscsi_cls_conn *cls_conn; 431 432 cls_conn = iscsi_conn_setup(cls_session, 0, conn_idx); 433 if (!cls_conn) 434 return NULL; 435 conn = cls_conn->dd_data; 436 437 /* 438 * due to issues with the login code re iser sematics 439 * this not set in iscsi_conn_setup - FIXME 440 */ 441 conn->max_recv_dlength = ISER_RECV_DATA_SEG_LEN; 442 443 return cls_conn; 444 } 445 446 /** 447 * iscsi_iser_conn_bind() - bind iscsi and iser connection structures 448 * @cls_session: iscsi class session 449 * @cls_conn: iscsi class connection 450 * @transport_eph: transport end-point handle 451 * @is_leading: indicate if this is the session leading connection (MCS) 452 * 453 * Return: zero on success, $error if iscsi_conn_bind fails and 454 * -EINVAL in case end-point doesn't exsits anymore or iser connection 455 * state is not UP (teardown already started). 456 */ 457 static int 458 iscsi_iser_conn_bind(struct iscsi_cls_session *cls_session, 459 struct iscsi_cls_conn *cls_conn, 460 uint64_t transport_eph, 461 int is_leading) 462 { 463 struct iscsi_conn *conn = cls_conn->dd_data; 464 struct iser_conn *iser_conn; 465 struct iscsi_endpoint *ep; 466 int error; 467 468 error = iscsi_conn_bind(cls_session, cls_conn, is_leading); 469 if (error) 470 return error; 471 472 /* the transport ep handle comes from user space so it must be 473 * verified against the global ib connections list */ 474 ep = iscsi_lookup_endpoint(transport_eph); 475 if (!ep) { 476 iser_err("can't bind eph %llx\n", 477 (unsigned long long)transport_eph); 478 return -EINVAL; 479 } 480 iser_conn = ep->dd_data; 481 482 mutex_lock(&iser_conn->state_mutex); 483 if (iser_conn->state != ISER_CONN_UP) { 484 error = -EINVAL; 485 iser_err("iser_conn %p state is %d, teardown started\n", 486 iser_conn, iser_conn->state); 487 goto out; 488 } 489 490 error = iser_alloc_rx_descriptors(iser_conn, conn->session); 491 if (error) 492 goto out; 493 494 /* binds the iSER connection retrieved from the previously 495 * connected ep_handle to the iSCSI layer connection. exchanges 496 * connection pointers */ 497 iser_info("binding iscsi conn %p to iser_conn %p\n", conn, iser_conn); 498 499 conn->dd_data = iser_conn; 500 iser_conn->iscsi_conn = conn; 501 502 out: 503 mutex_unlock(&iser_conn->state_mutex); 504 return error; 505 } 506 507 /** 508 * iscsi_iser_conn_start() - start iscsi-iser connection 509 * @cls_conn: iscsi class connection 510 * 511 * Notes: Here iser intialize (or re-initialize) stop_completion as 512 * from this point iscsi must call conn_stop in session/connection 513 * teardown so iser transport must wait for it. 514 */ 515 static int 516 iscsi_iser_conn_start(struct iscsi_cls_conn *cls_conn) 517 { 518 struct iscsi_conn *iscsi_conn; 519 struct iser_conn *iser_conn; 520 521 iscsi_conn = cls_conn->dd_data; 522 iser_conn = iscsi_conn->dd_data; 523 reinit_completion(&iser_conn->stop_completion); 524 525 return iscsi_conn_start(cls_conn); 526 } 527 528 /** 529 * iscsi_iser_conn_stop() - stop iscsi-iser connection 530 * @cls_conn: iscsi class connection 531 * @flag: indicate if recover or terminate (passed as is) 532 * 533 * Notes: Calling iscsi_conn_stop might theoretically race with 534 * DEVICE_REMOVAL event and dereference a previously freed RDMA device 535 * handle, so we call it under iser the state lock to protect against 536 * this kind of race. 537 */ 538 static void 539 iscsi_iser_conn_stop(struct iscsi_cls_conn *cls_conn, int flag) 540 { 541 struct iscsi_conn *conn = cls_conn->dd_data; 542 struct iser_conn *iser_conn = conn->dd_data; 543 544 iser_info("stopping iscsi_conn: %p, iser_conn: %p\n", conn, iser_conn); 545 546 /* 547 * Userspace may have goofed up and not bound the connection or 548 * might have only partially setup the connection. 549 */ 550 if (iser_conn) { 551 mutex_lock(&iser_conn->state_mutex); 552 iser_conn_terminate(iser_conn); 553 iscsi_conn_stop(cls_conn, flag); 554 555 /* unbind */ 556 iser_conn->iscsi_conn = NULL; 557 conn->dd_data = NULL; 558 559 complete(&iser_conn->stop_completion); 560 mutex_unlock(&iser_conn->state_mutex); 561 } else { 562 iscsi_conn_stop(cls_conn, flag); 563 } 564 } 565 566 /** 567 * iscsi_iser_session_destroy() - destroy iscsi-iser session 568 * @cls_session: iscsi class session 569 * 570 * Removes and free iscsi host. 571 */ 572 static void 573 iscsi_iser_session_destroy(struct iscsi_cls_session *cls_session) 574 { 575 struct Scsi_Host *shost = iscsi_session_to_shost(cls_session); 576 577 iscsi_session_teardown(cls_session); 578 iscsi_host_remove(shost); 579 iscsi_host_free(shost); 580 } 581 582 static inline unsigned int 583 iser_dif_prot_caps(int prot_caps) 584 { 585 return ((prot_caps & IB_PROT_T10DIF_TYPE_1) ? 586 SHOST_DIF_TYPE1_PROTECTION | SHOST_DIX_TYPE0_PROTECTION | 587 SHOST_DIX_TYPE1_PROTECTION : 0) | 588 ((prot_caps & IB_PROT_T10DIF_TYPE_2) ? 589 SHOST_DIF_TYPE2_PROTECTION | SHOST_DIX_TYPE2_PROTECTION : 0) | 590 ((prot_caps & IB_PROT_T10DIF_TYPE_3) ? 591 SHOST_DIF_TYPE3_PROTECTION | SHOST_DIX_TYPE3_PROTECTION : 0); 592 } 593 594 /** 595 * iscsi_iser_session_create() - create an iscsi-iser session 596 * @ep: iscsi end-point handle 597 * @cmds_max: maximum commands in this session 598 * @qdepth: session command queue depth 599 * @initial_cmdsn: initiator command sequnce number 600 * 601 * Allocates and adds a scsi host, expose DIF supprot if 602 * exists, and sets up an iscsi session. 603 */ 604 static struct iscsi_cls_session * 605 iscsi_iser_session_create(struct iscsi_endpoint *ep, 606 uint16_t cmds_max, uint16_t qdepth, 607 uint32_t initial_cmdsn) 608 { 609 struct iscsi_cls_session *cls_session; 610 struct iscsi_session *session; 611 struct Scsi_Host *shost; 612 struct iser_conn *iser_conn = NULL; 613 struct ib_conn *ib_conn; 614 u16 max_cmds; 615 616 shost = iscsi_host_alloc(&iscsi_iser_sht, 0, 0); 617 if (!shost) 618 return NULL; 619 shost->transportt = iscsi_iser_scsi_transport; 620 shost->cmd_per_lun = qdepth; 621 shost->max_lun = iscsi_max_lun; 622 shost->max_id = 0; 623 shost->max_channel = 0; 624 shost->max_cmd_len = 16; 625 626 /* 627 * older userspace tools (before 2.0-870) did not pass us 628 * the leading conn's ep so this will be NULL; 629 */ 630 if (ep) { 631 iser_conn = ep->dd_data; 632 max_cmds = iser_conn->max_cmds; 633 shost->sg_tablesize = iser_conn->scsi_sg_tablesize; 634 shost->max_sectors = iser_conn->scsi_max_sectors; 635 636 mutex_lock(&iser_conn->state_mutex); 637 if (iser_conn->state != ISER_CONN_UP) { 638 iser_err("iser conn %p already started teardown\n", 639 iser_conn); 640 mutex_unlock(&iser_conn->state_mutex); 641 goto free_host; 642 } 643 644 ib_conn = &iser_conn->ib_conn; 645 if (ib_conn->pi_support) { 646 u32 sig_caps = ib_conn->device->dev_attr.sig_prot_cap; 647 648 scsi_host_set_prot(shost, iser_dif_prot_caps(sig_caps)); 649 scsi_host_set_guard(shost, SHOST_DIX_GUARD_IP | 650 SHOST_DIX_GUARD_CRC); 651 } 652 653 /* 654 * Limit the sg_tablesize and max_sectors based on the device 655 * max fastreg page list length. 656 */ 657 shost->sg_tablesize = min_t(unsigned short, shost->sg_tablesize, 658 ib_conn->device->dev_attr.max_fast_reg_page_list_len); 659 shost->max_sectors = min_t(unsigned int, 660 1024, (shost->sg_tablesize * PAGE_SIZE) >> 9); 661 662 if (iscsi_host_add(shost, 663 ib_conn->device->ib_device->dma_device)) { 664 mutex_unlock(&iser_conn->state_mutex); 665 goto free_host; 666 } 667 mutex_unlock(&iser_conn->state_mutex); 668 } else { 669 max_cmds = ISER_DEF_XMIT_CMDS_MAX; 670 if (iscsi_host_add(shost, NULL)) 671 goto free_host; 672 } 673 674 if (cmds_max > max_cmds) { 675 iser_info("cmds_max changed from %u to %u\n", 676 cmds_max, max_cmds); 677 cmds_max = max_cmds; 678 } 679 680 cls_session = iscsi_session_setup(&iscsi_iser_transport, shost, 681 cmds_max, 0, 682 sizeof(struct iscsi_iser_task), 683 initial_cmdsn, 0); 684 if (!cls_session) 685 goto remove_host; 686 session = cls_session->dd_data; 687 688 shost->can_queue = session->scsi_cmds_max; 689 return cls_session; 690 691 remove_host: 692 iscsi_host_remove(shost); 693 free_host: 694 iscsi_host_free(shost); 695 return NULL; 696 } 697 698 static int 699 iscsi_iser_set_param(struct iscsi_cls_conn *cls_conn, 700 enum iscsi_param param, char *buf, int buflen) 701 { 702 int value; 703 704 switch (param) { 705 case ISCSI_PARAM_MAX_RECV_DLENGTH: 706 /* TBD */ 707 break; 708 case ISCSI_PARAM_HDRDGST_EN: 709 sscanf(buf, "%d", &value); 710 if (value) { 711 iser_err("DataDigest wasn't negotiated to None\n"); 712 return -EPROTO; 713 } 714 break; 715 case ISCSI_PARAM_DATADGST_EN: 716 sscanf(buf, "%d", &value); 717 if (value) { 718 iser_err("DataDigest wasn't negotiated to None\n"); 719 return -EPROTO; 720 } 721 break; 722 case ISCSI_PARAM_IFMARKER_EN: 723 sscanf(buf, "%d", &value); 724 if (value) { 725 iser_err("IFMarker wasn't negotiated to No\n"); 726 return -EPROTO; 727 } 728 break; 729 case ISCSI_PARAM_OFMARKER_EN: 730 sscanf(buf, "%d", &value); 731 if (value) { 732 iser_err("OFMarker wasn't negotiated to No\n"); 733 return -EPROTO; 734 } 735 break; 736 default: 737 return iscsi_set_param(cls_conn, param, buf, buflen); 738 } 739 740 return 0; 741 } 742 743 /** 744 * iscsi_iser_set_param() - set class connection parameter 745 * @cls_conn: iscsi class connection 746 * @stats: iscsi stats to output 747 * 748 * Output connection statistics. 749 */ 750 static void 751 iscsi_iser_conn_get_stats(struct iscsi_cls_conn *cls_conn, struct iscsi_stats *stats) 752 { 753 struct iscsi_conn *conn = cls_conn->dd_data; 754 755 stats->txdata_octets = conn->txdata_octets; 756 stats->rxdata_octets = conn->rxdata_octets; 757 stats->scsicmd_pdus = conn->scsicmd_pdus_cnt; 758 stats->dataout_pdus = conn->dataout_pdus_cnt; 759 stats->scsirsp_pdus = conn->scsirsp_pdus_cnt; 760 stats->datain_pdus = conn->datain_pdus_cnt; /* always 0 */ 761 stats->r2t_pdus = conn->r2t_pdus_cnt; /* always 0 */ 762 stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt; 763 stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt; 764 stats->custom_length = 1; 765 strcpy(stats->custom[0].desc, "fmr_unalign_cnt"); 766 stats->custom[0].value = conn->fmr_unalign_cnt; 767 } 768 769 static int iscsi_iser_get_ep_param(struct iscsi_endpoint *ep, 770 enum iscsi_param param, char *buf) 771 { 772 struct iser_conn *iser_conn = ep->dd_data; 773 int len; 774 775 switch (param) { 776 case ISCSI_PARAM_CONN_PORT: 777 case ISCSI_PARAM_CONN_ADDRESS: 778 if (!iser_conn || !iser_conn->ib_conn.cma_id) 779 return -ENOTCONN; 780 781 return iscsi_conn_get_addr_param((struct sockaddr_storage *) 782 &iser_conn->ib_conn.cma_id->route.addr.dst_addr, 783 param, buf); 784 break; 785 default: 786 return -ENOSYS; 787 } 788 789 return len; 790 } 791 792 /** 793 * iscsi_iser_ep_connect() - Initiate iSER connection establishment 794 * @shost: scsi_host 795 * @dst_addr: destination address 796 * @non-blocking: indicate if routine can block 797 * 798 * Allocate an iscsi endpoint, an iser_conn structure and bind them. 799 * After that start RDMA connection establishment via rdma_cm. We 800 * don't allocate iser_conn embedded in iscsi_endpoint since in teardown 801 * the endpoint will be destroyed at ep_disconnect while iser_conn will 802 * cleanup its resources asynchronuously. 803 * 804 * Return: iscsi_endpoint created by iscsi layer or ERR_PTR(error) 805 * if fails. 806 */ 807 static struct iscsi_endpoint * 808 iscsi_iser_ep_connect(struct Scsi_Host *shost, struct sockaddr *dst_addr, 809 int non_blocking) 810 { 811 int err; 812 struct iser_conn *iser_conn; 813 struct iscsi_endpoint *ep; 814 815 ep = iscsi_create_endpoint(0); 816 if (!ep) 817 return ERR_PTR(-ENOMEM); 818 819 iser_conn = kzalloc(sizeof(*iser_conn), GFP_KERNEL); 820 if (!iser_conn) { 821 err = -ENOMEM; 822 goto failure; 823 } 824 825 ep->dd_data = iser_conn; 826 iser_conn->ep = ep; 827 iser_conn_init(iser_conn); 828 829 err = iser_connect(iser_conn, NULL, dst_addr, non_blocking); 830 if (err) 831 goto failure; 832 833 return ep; 834 failure: 835 iscsi_destroy_endpoint(ep); 836 return ERR_PTR(err); 837 } 838 839 /** 840 * iscsi_iser_ep_poll() - poll for iser connection establishment to complete 841 * @ep: iscsi endpoint (created at ep_connect) 842 * @timeout_ms: polling timeout allowed in ms. 843 * 844 * This routine boils down to waiting for up_completion signaling 845 * that cma_id got CONNECTED event. 846 * 847 * Return: 1 if succeeded in connection establishment, 0 if timeout expired 848 * (libiscsi will retry will kick in) or -1 if interrupted by signal 849 * or more likely iser connection state transitioned to TEMINATING or 850 * DOWN during the wait period. 851 */ 852 static int 853 iscsi_iser_ep_poll(struct iscsi_endpoint *ep, int timeout_ms) 854 { 855 struct iser_conn *iser_conn = ep->dd_data; 856 int rc; 857 858 rc = wait_for_completion_interruptible_timeout(&iser_conn->up_completion, 859 msecs_to_jiffies(timeout_ms)); 860 /* if conn establishment failed, return error code to iscsi */ 861 if (rc == 0) { 862 mutex_lock(&iser_conn->state_mutex); 863 if (iser_conn->state == ISER_CONN_TERMINATING || 864 iser_conn->state == ISER_CONN_DOWN) 865 rc = -1; 866 mutex_unlock(&iser_conn->state_mutex); 867 } 868 869 iser_info("iser conn %p rc = %d\n", iser_conn, rc); 870 871 if (rc > 0) 872 return 1; /* success, this is the equivalent of POLLOUT */ 873 else if (!rc) 874 return 0; /* timeout */ 875 else 876 return rc; /* signal */ 877 } 878 879 /** 880 * iscsi_iser_ep_disconnect() - Initiate connection teardown process 881 * @ep: iscsi endpoint handle 882 * 883 * This routine is not blocked by iser and RDMA termination process 884 * completion as we queue a deffered work for iser/RDMA destruction 885 * and cleanup or actually call it immediately in case we didn't pass 886 * iscsi conn bind/start stage, thus it is safe. 887 */ 888 static void 889 iscsi_iser_ep_disconnect(struct iscsi_endpoint *ep) 890 { 891 struct iser_conn *iser_conn = ep->dd_data; 892 893 iser_info("ep %p iser conn %p\n", ep, iser_conn); 894 895 mutex_lock(&iser_conn->state_mutex); 896 iser_conn_terminate(iser_conn); 897 898 /* 899 * if iser_conn and iscsi_conn are bound, we must wait for 900 * iscsi_conn_stop and flush errors completion before freeing 901 * the iser resources. Otherwise we are safe to free resources 902 * immediately. 903 */ 904 if (iser_conn->iscsi_conn) { 905 INIT_WORK(&iser_conn->release_work, iser_release_work); 906 queue_work(release_wq, &iser_conn->release_work); 907 mutex_unlock(&iser_conn->state_mutex); 908 } else { 909 iser_conn->state = ISER_CONN_DOWN; 910 mutex_unlock(&iser_conn->state_mutex); 911 iser_conn_release(iser_conn); 912 } 913 914 iscsi_destroy_endpoint(ep); 915 } 916 917 static umode_t iser_attr_is_visible(int param_type, int param) 918 { 919 switch (param_type) { 920 case ISCSI_HOST_PARAM: 921 switch (param) { 922 case ISCSI_HOST_PARAM_NETDEV_NAME: 923 case ISCSI_HOST_PARAM_HWADDRESS: 924 case ISCSI_HOST_PARAM_INITIATOR_NAME: 925 return S_IRUGO; 926 default: 927 return 0; 928 } 929 case ISCSI_PARAM: 930 switch (param) { 931 case ISCSI_PARAM_MAX_RECV_DLENGTH: 932 case ISCSI_PARAM_MAX_XMIT_DLENGTH: 933 case ISCSI_PARAM_HDRDGST_EN: 934 case ISCSI_PARAM_DATADGST_EN: 935 case ISCSI_PARAM_CONN_ADDRESS: 936 case ISCSI_PARAM_CONN_PORT: 937 case ISCSI_PARAM_EXP_STATSN: 938 case ISCSI_PARAM_PERSISTENT_ADDRESS: 939 case ISCSI_PARAM_PERSISTENT_PORT: 940 case ISCSI_PARAM_PING_TMO: 941 case ISCSI_PARAM_RECV_TMO: 942 case ISCSI_PARAM_INITIAL_R2T_EN: 943 case ISCSI_PARAM_MAX_R2T: 944 case ISCSI_PARAM_IMM_DATA_EN: 945 case ISCSI_PARAM_FIRST_BURST: 946 case ISCSI_PARAM_MAX_BURST: 947 case ISCSI_PARAM_PDU_INORDER_EN: 948 case ISCSI_PARAM_DATASEQ_INORDER_EN: 949 case ISCSI_PARAM_TARGET_NAME: 950 case ISCSI_PARAM_TPGT: 951 case ISCSI_PARAM_USERNAME: 952 case ISCSI_PARAM_PASSWORD: 953 case ISCSI_PARAM_USERNAME_IN: 954 case ISCSI_PARAM_PASSWORD_IN: 955 case ISCSI_PARAM_FAST_ABORT: 956 case ISCSI_PARAM_ABORT_TMO: 957 case ISCSI_PARAM_LU_RESET_TMO: 958 case ISCSI_PARAM_TGT_RESET_TMO: 959 case ISCSI_PARAM_IFACE_NAME: 960 case ISCSI_PARAM_INITIATOR_NAME: 961 case ISCSI_PARAM_DISCOVERY_SESS: 962 return S_IRUGO; 963 default: 964 return 0; 965 } 966 } 967 968 return 0; 969 } 970 971 static struct scsi_host_template iscsi_iser_sht = { 972 .module = THIS_MODULE, 973 .name = "iSCSI Initiator over iSER", 974 .queuecommand = iscsi_queuecommand, 975 .change_queue_depth = scsi_change_queue_depth, 976 .sg_tablesize = ISCSI_ISER_DEF_SG_TABLESIZE, 977 .max_sectors = ISER_DEF_MAX_SECTORS, 978 .cmd_per_lun = ISER_DEF_CMD_PER_LUN, 979 .eh_abort_handler = iscsi_eh_abort, 980 .eh_device_reset_handler= iscsi_eh_device_reset, 981 .eh_target_reset_handler = iscsi_eh_recover_target, 982 .target_alloc = iscsi_target_alloc, 983 .use_clustering = DISABLE_CLUSTERING, 984 .proc_name = "iscsi_iser", 985 .this_id = -1, 986 .track_queue_depth = 1, 987 }; 988 989 static struct iscsi_transport iscsi_iser_transport = { 990 .owner = THIS_MODULE, 991 .name = "iser", 992 .caps = CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_TEXT_NEGO, 993 /* session management */ 994 .create_session = iscsi_iser_session_create, 995 .destroy_session = iscsi_iser_session_destroy, 996 /* connection management */ 997 .create_conn = iscsi_iser_conn_create, 998 .bind_conn = iscsi_iser_conn_bind, 999 .destroy_conn = iscsi_conn_teardown, 1000 .attr_is_visible = iser_attr_is_visible, 1001 .set_param = iscsi_iser_set_param, 1002 .get_conn_param = iscsi_conn_get_param, 1003 .get_ep_param = iscsi_iser_get_ep_param, 1004 .get_session_param = iscsi_session_get_param, 1005 .start_conn = iscsi_iser_conn_start, 1006 .stop_conn = iscsi_iser_conn_stop, 1007 /* iscsi host params */ 1008 .get_host_param = iscsi_host_get_param, 1009 .set_host_param = iscsi_host_set_param, 1010 /* IO */ 1011 .send_pdu = iscsi_conn_send_pdu, 1012 .get_stats = iscsi_iser_conn_get_stats, 1013 .init_task = iscsi_iser_task_init, 1014 .xmit_task = iscsi_iser_task_xmit, 1015 .cleanup_task = iscsi_iser_cleanup_task, 1016 .alloc_pdu = iscsi_iser_pdu_alloc, 1017 .check_protection = iscsi_iser_check_protection, 1018 /* recovery */ 1019 .session_recovery_timedout = iscsi_session_recovery_timedout, 1020 1021 .ep_connect = iscsi_iser_ep_connect, 1022 .ep_poll = iscsi_iser_ep_poll, 1023 .ep_disconnect = iscsi_iser_ep_disconnect 1024 }; 1025 1026 static int __init iser_init(void) 1027 { 1028 int err; 1029 1030 iser_dbg("Starting iSER datamover...\n"); 1031 1032 if (iscsi_max_lun < 1) { 1033 iser_err("Invalid max_lun value of %u\n", iscsi_max_lun); 1034 return -EINVAL; 1035 } 1036 1037 memset(&ig, 0, sizeof(struct iser_global)); 1038 1039 ig.desc_cache = kmem_cache_create("iser_descriptors", 1040 sizeof(struct iser_tx_desc), 1041 0, SLAB_HWCACHE_ALIGN, 1042 NULL); 1043 if (ig.desc_cache == NULL) 1044 return -ENOMEM; 1045 1046 /* device init is called only after the first addr resolution */ 1047 mutex_init(&ig.device_list_mutex); 1048 INIT_LIST_HEAD(&ig.device_list); 1049 mutex_init(&ig.connlist_mutex); 1050 INIT_LIST_HEAD(&ig.connlist); 1051 1052 release_wq = alloc_workqueue("release workqueue", 0, 0); 1053 if (!release_wq) { 1054 iser_err("failed to allocate release workqueue\n"); 1055 return -ENOMEM; 1056 } 1057 1058 iscsi_iser_scsi_transport = iscsi_register_transport( 1059 &iscsi_iser_transport); 1060 if (!iscsi_iser_scsi_transport) { 1061 iser_err("iscsi_register_transport failed\n"); 1062 err = -EINVAL; 1063 goto register_transport_failure; 1064 } 1065 1066 return 0; 1067 1068 register_transport_failure: 1069 kmem_cache_destroy(ig.desc_cache); 1070 1071 return err; 1072 } 1073 1074 static void __exit iser_exit(void) 1075 { 1076 struct iser_conn *iser_conn, *n; 1077 int connlist_empty; 1078 1079 iser_dbg("Removing iSER datamover...\n"); 1080 destroy_workqueue(release_wq); 1081 1082 mutex_lock(&ig.connlist_mutex); 1083 connlist_empty = list_empty(&ig.connlist); 1084 mutex_unlock(&ig.connlist_mutex); 1085 1086 if (!connlist_empty) { 1087 iser_err("Error cleanup stage completed but we still have iser " 1088 "connections, destroying them anyway\n"); 1089 list_for_each_entry_safe(iser_conn, n, &ig.connlist, 1090 conn_list) { 1091 iser_conn_release(iser_conn); 1092 } 1093 } 1094 1095 iscsi_unregister_transport(&iscsi_iser_transport); 1096 kmem_cache_destroy(ig.desc_cache); 1097 } 1098 1099 module_init(iser_init); 1100 module_exit(iser_exit); 1101