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 <linux/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 81 static struct scsi_host_template iscsi_iser_sht; 82 static struct iscsi_transport iscsi_iser_transport; 83 static struct scsi_transport_template *iscsi_iser_scsi_transport; 84 static struct workqueue_struct *release_wq; 85 static DEFINE_MUTEX(unbind_iser_conn_mutex); 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_always_reg = true; 101 module_param_named(always_register, iser_always_reg, bool, S_IRUGO); 102 MODULE_PARM_DESC(always_register, 103 "Always register memory, even for continuous memory regions (default:true)"); 104 105 bool iser_pi_enable = false; 106 module_param_named(pi_enable, iser_pi_enable, bool, S_IRUGO); 107 MODULE_PARM_DESC(pi_enable, "Enable T10-PI offload support (default:disabled)"); 108 109 int iser_pi_guard; 110 module_param_named(pi_guard, iser_pi_guard, int, S_IRUGO); 111 MODULE_PARM_DESC(pi_guard, "T10-PI guard_type [deprecated]"); 112 113 /* 114 * iscsi_iser_recv() - Process a successful recv completion 115 * @conn: iscsi connection 116 * @hdr: iscsi header 117 * @rx_data: buffer containing receive data payload 118 * @rx_data_len: length of rx_data 119 * 120 * Notes: In case of data length errors or iscsi PDU completion failures 121 * this routine will signal iscsi layer of connection failure. 122 */ 123 void 124 iscsi_iser_recv(struct iscsi_conn *conn, struct iscsi_hdr *hdr, 125 char *rx_data, int rx_data_len) 126 { 127 int rc = 0; 128 int datalen; 129 130 /* verify PDU length */ 131 datalen = ntoh24(hdr->dlength); 132 if (datalen > rx_data_len || (datalen + 4) < rx_data_len) { 133 iser_err("wrong datalen %d (hdr), %d (IB)\n", 134 datalen, rx_data_len); 135 rc = ISCSI_ERR_DATALEN; 136 goto error; 137 } 138 139 if (datalen != rx_data_len) 140 iser_dbg("aligned datalen (%d) hdr, %d (IB)\n", 141 datalen, rx_data_len); 142 143 rc = iscsi_complete_pdu(conn, hdr, rx_data, rx_data_len); 144 if (rc && rc != ISCSI_ERR_NO_SCSI_CMD) 145 goto error; 146 147 return; 148 error: 149 iscsi_conn_failure(conn, rc); 150 } 151 152 /** 153 * iscsi_iser_pdu_alloc() - allocate an iscsi-iser PDU 154 * @task: iscsi task 155 * @opcode: iscsi command opcode 156 * 157 * Netes: This routine can't fail, just assign iscsi task 158 * hdr and max hdr size. 159 */ 160 static int 161 iscsi_iser_pdu_alloc(struct iscsi_task *task, uint8_t opcode) 162 { 163 struct iscsi_iser_task *iser_task = task->dd_data; 164 165 task->hdr = (struct iscsi_hdr *)&iser_task->desc.iscsi_header; 166 task->hdr_max = sizeof(iser_task->desc.iscsi_header); 167 168 return 0; 169 } 170 171 /** 172 * iser_initialize_task_headers() - Initialize task headers 173 * @task: iscsi task 174 * @tx_desc: iser tx descriptor 175 * 176 * Notes: 177 * This routine may race with iser teardown flow for scsi 178 * error handling TMFs. So for TMF we should acquire the 179 * state mutex to avoid dereferencing the IB device which 180 * may have already been terminated. 181 */ 182 int 183 iser_initialize_task_headers(struct iscsi_task *task, 184 struct iser_tx_desc *tx_desc) 185 { 186 struct iser_conn *iser_conn = task->conn->dd_data; 187 struct iser_device *device = iser_conn->ib_conn.device; 188 struct iscsi_iser_task *iser_task = task->dd_data; 189 u64 dma_addr; 190 191 if (unlikely(iser_conn->state != ISER_CONN_UP)) 192 return -ENODEV; 193 194 dma_addr = ib_dma_map_single(device->ib_device, (void *)tx_desc, 195 ISER_HEADERS_LEN, DMA_TO_DEVICE); 196 if (ib_dma_mapping_error(device->ib_device, dma_addr)) 197 return -ENOMEM; 198 199 tx_desc->inv_wr.next = NULL; 200 tx_desc->reg_wr.wr.next = NULL; 201 tx_desc->mapped = true; 202 tx_desc->dma_addr = dma_addr; 203 tx_desc->tx_sg[0].addr = tx_desc->dma_addr; 204 tx_desc->tx_sg[0].length = ISER_HEADERS_LEN; 205 tx_desc->tx_sg[0].lkey = device->pd->local_dma_lkey; 206 207 iser_task->iser_conn = iser_conn; 208 209 return 0; 210 } 211 212 /** 213 * iscsi_iser_task_init() - Initialize iscsi-iser task 214 * @task: iscsi task 215 * 216 * Initialize the task for the scsi command or mgmt command. 217 * 218 * Return: Returns zero on success or -ENOMEM when failing 219 * to init task headers (dma mapping error). 220 */ 221 static int 222 iscsi_iser_task_init(struct iscsi_task *task) 223 { 224 struct iscsi_iser_task *iser_task = task->dd_data; 225 int ret; 226 227 ret = iser_initialize_task_headers(task, &iser_task->desc); 228 if (ret) { 229 iser_err("Failed to init task %p, err = %d\n", 230 iser_task, ret); 231 return ret; 232 } 233 234 /* mgmt task */ 235 if (!task->sc) 236 return 0; 237 238 iser_task->command_sent = 0; 239 iser_task_rdma_init(iser_task); 240 iser_task->sc = task->sc; 241 242 return 0; 243 } 244 245 /** 246 * iscsi_iser_mtask_xmit() - xmit management (immediate) task 247 * @conn: iscsi connection 248 * @task: task management task 249 * 250 * Notes: 251 * The function can return -EAGAIN in which case caller must 252 * call it again later, or recover. '0' return code means successful 253 * xmit. 254 * 255 **/ 256 static int 257 iscsi_iser_mtask_xmit(struct iscsi_conn *conn, struct iscsi_task *task) 258 { 259 int error = 0; 260 261 iser_dbg("mtask xmit [cid %d itt 0x%x]\n", conn->id, task->itt); 262 263 error = iser_send_control(conn, task); 264 265 /* since iser xmits control with zero copy, tasks can not be recycled 266 * right after sending them. 267 * The recycling scheme is based on whether a response is expected 268 * - if yes, the task is recycled at iscsi_complete_pdu 269 * - if no, the task is recycled at iser_snd_completion 270 */ 271 return error; 272 } 273 274 static int 275 iscsi_iser_task_xmit_unsol_data(struct iscsi_conn *conn, 276 struct iscsi_task *task) 277 { 278 struct iscsi_r2t_info *r2t = &task->unsol_r2t; 279 struct iscsi_data hdr; 280 int error = 0; 281 282 /* Send data-out PDUs while there's still unsolicited data to send */ 283 while (iscsi_task_has_unsol_data(task)) { 284 iscsi_prep_data_out_pdu(task, r2t, &hdr); 285 iser_dbg("Sending data-out: itt 0x%x, data count %d\n", 286 hdr.itt, r2t->data_count); 287 288 /* the buffer description has been passed with the command */ 289 /* Send the command */ 290 error = iser_send_data_out(conn, task, &hdr); 291 if (error) { 292 r2t->datasn--; 293 goto iscsi_iser_task_xmit_unsol_data_exit; 294 } 295 r2t->sent += r2t->data_count; 296 iser_dbg("Need to send %d more as data-out PDUs\n", 297 r2t->data_length - r2t->sent); 298 } 299 300 iscsi_iser_task_xmit_unsol_data_exit: 301 return error; 302 } 303 304 /** 305 * iscsi_iser_task_xmit() - xmit iscsi-iser task 306 * @task: iscsi task 307 * 308 * Return: zero on success or escalates $error on failure. 309 */ 310 static int 311 iscsi_iser_task_xmit(struct iscsi_task *task) 312 { 313 struct iscsi_conn *conn = task->conn; 314 struct iscsi_iser_task *iser_task = task->dd_data; 315 int error = 0; 316 317 if (!task->sc) 318 return iscsi_iser_mtask_xmit(conn, task); 319 320 if (task->sc->sc_data_direction == DMA_TO_DEVICE) { 321 BUG_ON(scsi_bufflen(task->sc) == 0); 322 323 iser_dbg("cmd [itt %x total %d imm %d unsol_data %d\n", 324 task->itt, scsi_bufflen(task->sc), 325 task->imm_count, task->unsol_r2t.data_length); 326 } 327 328 iser_dbg("ctask xmit [cid %d itt 0x%x]\n", 329 conn->id, task->itt); 330 331 /* Send the cmd PDU */ 332 if (!iser_task->command_sent) { 333 error = iser_send_command(conn, task); 334 if (error) 335 goto iscsi_iser_task_xmit_exit; 336 iser_task->command_sent = 1; 337 } 338 339 /* Send unsolicited data-out PDU(s) if necessary */ 340 if (iscsi_task_has_unsol_data(task)) 341 error = iscsi_iser_task_xmit_unsol_data(conn, task); 342 343 iscsi_iser_task_xmit_exit: 344 return error; 345 } 346 347 /** 348 * iscsi_iser_cleanup_task() - cleanup an iscsi-iser task 349 * @task: iscsi task 350 * 351 * Notes: In case the RDMA device is already NULL (might have 352 * been removed in DEVICE_REMOVAL CM event it will bail-out 353 * without doing dma unmapping. 354 */ 355 static void iscsi_iser_cleanup_task(struct iscsi_task *task) 356 { 357 struct iscsi_iser_task *iser_task = task->dd_data; 358 struct iser_tx_desc *tx_desc = &iser_task->desc; 359 struct iser_conn *iser_conn = task->conn->dd_data; 360 struct iser_device *device = iser_conn->ib_conn.device; 361 362 /* DEVICE_REMOVAL event might have already released the device */ 363 if (!device) 364 return; 365 366 if (likely(tx_desc->mapped)) { 367 ib_dma_unmap_single(device->ib_device, tx_desc->dma_addr, 368 ISER_HEADERS_LEN, DMA_TO_DEVICE); 369 tx_desc->mapped = false; 370 } 371 372 /* mgmt tasks do not need special cleanup */ 373 if (!task->sc) 374 return; 375 376 if (iser_task->status == ISER_TASK_STATUS_STARTED) { 377 iser_task->status = ISER_TASK_STATUS_COMPLETED; 378 iser_task_rdma_finalize(iser_task); 379 } 380 } 381 382 /** 383 * iscsi_iser_check_protection() - check protection information status of task. 384 * @task: iscsi task 385 * @sector: error sector if exsists (output) 386 * 387 * Return: zero if no data-integrity errors have occured 388 * 0x1: data-integrity error occured in the guard-block 389 * 0x2: data-integrity error occured in the reference tag 390 * 0x3: data-integrity error occured in the application tag 391 * 392 * In addition the error sector is marked. 393 */ 394 static u8 395 iscsi_iser_check_protection(struct iscsi_task *task, sector_t *sector) 396 { 397 struct iscsi_iser_task *iser_task = task->dd_data; 398 enum iser_data_dir dir = iser_task->dir[ISER_DIR_IN] ? 399 ISER_DIR_IN : ISER_DIR_OUT; 400 401 return iser_check_task_pi_status(iser_task, dir, sector); 402 } 403 404 /** 405 * iscsi_iser_conn_create() - create a new iscsi-iser connection 406 * @cls_session: iscsi class connection 407 * @conn_idx: connection index within the session (for MCS) 408 * 409 * Return: iscsi_cls_conn when iscsi_conn_setup succeeds or NULL 410 * otherwise. 411 */ 412 static struct iscsi_cls_conn * 413 iscsi_iser_conn_create(struct iscsi_cls_session *cls_session, 414 uint32_t conn_idx) 415 { 416 struct iscsi_conn *conn; 417 struct iscsi_cls_conn *cls_conn; 418 419 cls_conn = iscsi_conn_setup(cls_session, 0, conn_idx); 420 if (!cls_conn) 421 return NULL; 422 conn = cls_conn->dd_data; 423 424 /* 425 * due to issues with the login code re iser sematics 426 * this not set in iscsi_conn_setup - FIXME 427 */ 428 conn->max_recv_dlength = ISER_RECV_DATA_SEG_LEN; 429 430 return cls_conn; 431 } 432 433 /** 434 * iscsi_iser_conn_bind() - bind iscsi and iser connection structures 435 * @cls_session: iscsi class session 436 * @cls_conn: iscsi class connection 437 * @transport_eph: transport end-point handle 438 * @is_leading: indicate if this is the session leading connection (MCS) 439 * 440 * Return: zero on success, $error if iscsi_conn_bind fails and 441 * -EINVAL in case end-point doesn't exsits anymore or iser connection 442 * state is not UP (teardown already started). 443 */ 444 static int 445 iscsi_iser_conn_bind(struct iscsi_cls_session *cls_session, 446 struct iscsi_cls_conn *cls_conn, 447 uint64_t transport_eph, 448 int is_leading) 449 { 450 struct iscsi_conn *conn = cls_conn->dd_data; 451 struct iser_conn *iser_conn; 452 struct iscsi_endpoint *ep; 453 int error; 454 455 error = iscsi_conn_bind(cls_session, cls_conn, is_leading); 456 if (error) 457 return error; 458 459 /* the transport ep handle comes from user space so it must be 460 * verified against the global ib connections list */ 461 ep = iscsi_lookup_endpoint(transport_eph); 462 if (!ep) { 463 iser_err("can't bind eph %llx\n", 464 (unsigned long long)transport_eph); 465 return -EINVAL; 466 } 467 iser_conn = ep->dd_data; 468 469 mutex_lock(&iser_conn->state_mutex); 470 if (iser_conn->state != ISER_CONN_UP) { 471 error = -EINVAL; 472 iser_err("iser_conn %p state is %d, teardown started\n", 473 iser_conn, iser_conn->state); 474 goto out; 475 } 476 477 error = iser_alloc_rx_descriptors(iser_conn, conn->session); 478 if (error) 479 goto out; 480 481 /* binds the iSER connection retrieved from the previously 482 * connected ep_handle to the iSCSI layer connection. exchanges 483 * connection pointers */ 484 iser_info("binding iscsi conn %p to iser_conn %p\n", conn, iser_conn); 485 486 conn->dd_data = iser_conn; 487 iser_conn->iscsi_conn = conn; 488 489 out: 490 mutex_unlock(&iser_conn->state_mutex); 491 return error; 492 } 493 494 /** 495 * iscsi_iser_conn_start() - start iscsi-iser connection 496 * @cls_conn: iscsi class connection 497 * 498 * Notes: Here iser intialize (or re-initialize) stop_completion as 499 * from this point iscsi must call conn_stop in session/connection 500 * teardown so iser transport must wait for it. 501 */ 502 static int 503 iscsi_iser_conn_start(struct iscsi_cls_conn *cls_conn) 504 { 505 struct iscsi_conn *iscsi_conn; 506 struct iser_conn *iser_conn; 507 508 iscsi_conn = cls_conn->dd_data; 509 iser_conn = iscsi_conn->dd_data; 510 reinit_completion(&iser_conn->stop_completion); 511 512 return iscsi_conn_start(cls_conn); 513 } 514 515 /** 516 * iscsi_iser_conn_stop() - stop iscsi-iser connection 517 * @cls_conn: iscsi class connection 518 * @flag: indicate if recover or terminate (passed as is) 519 * 520 * Notes: Calling iscsi_conn_stop might theoretically race with 521 * DEVICE_REMOVAL event and dereference a previously freed RDMA device 522 * handle, so we call it under iser the state lock to protect against 523 * this kind of race. 524 */ 525 static void 526 iscsi_iser_conn_stop(struct iscsi_cls_conn *cls_conn, int flag) 527 { 528 struct iscsi_conn *conn = cls_conn->dd_data; 529 struct iser_conn *iser_conn = conn->dd_data; 530 531 iser_info("stopping iscsi_conn: %p, iser_conn: %p\n", conn, iser_conn); 532 533 /* 534 * Userspace may have goofed up and not bound the connection or 535 * might have only partially setup the connection. 536 */ 537 if (iser_conn) { 538 mutex_lock(&iser_conn->state_mutex); 539 mutex_lock(&unbind_iser_conn_mutex); 540 iser_conn_terminate(iser_conn); 541 iscsi_conn_stop(cls_conn, flag); 542 543 /* unbind */ 544 iser_conn->iscsi_conn = NULL; 545 conn->dd_data = NULL; 546 mutex_unlock(&unbind_iser_conn_mutex); 547 548 complete(&iser_conn->stop_completion); 549 mutex_unlock(&iser_conn->state_mutex); 550 } else { 551 iscsi_conn_stop(cls_conn, flag); 552 } 553 } 554 555 /** 556 * iscsi_iser_session_destroy() - destroy iscsi-iser session 557 * @cls_session: iscsi class session 558 * 559 * Removes and free iscsi host. 560 */ 561 static void 562 iscsi_iser_session_destroy(struct iscsi_cls_session *cls_session) 563 { 564 struct Scsi_Host *shost = iscsi_session_to_shost(cls_session); 565 566 iscsi_session_teardown(cls_session); 567 iscsi_host_remove(shost); 568 iscsi_host_free(shost); 569 } 570 571 static inline unsigned int 572 iser_dif_prot_caps(int prot_caps) 573 { 574 return ((prot_caps & IB_PROT_T10DIF_TYPE_1) ? 575 SHOST_DIF_TYPE1_PROTECTION | SHOST_DIX_TYPE0_PROTECTION | 576 SHOST_DIX_TYPE1_PROTECTION : 0) | 577 ((prot_caps & IB_PROT_T10DIF_TYPE_2) ? 578 SHOST_DIF_TYPE2_PROTECTION | SHOST_DIX_TYPE2_PROTECTION : 0) | 579 ((prot_caps & IB_PROT_T10DIF_TYPE_3) ? 580 SHOST_DIF_TYPE3_PROTECTION | SHOST_DIX_TYPE3_PROTECTION : 0); 581 } 582 583 /** 584 * iscsi_iser_session_create() - create an iscsi-iser session 585 * @ep: iscsi end-point handle 586 * @cmds_max: maximum commands in this session 587 * @qdepth: session command queue depth 588 * @initial_cmdsn: initiator command sequnce number 589 * 590 * Allocates and adds a scsi host, expose DIF supprot if 591 * exists, and sets up an iscsi session. 592 */ 593 static struct iscsi_cls_session * 594 iscsi_iser_session_create(struct iscsi_endpoint *ep, 595 uint16_t cmds_max, uint16_t qdepth, 596 uint32_t initial_cmdsn) 597 { 598 struct iscsi_cls_session *cls_session; 599 struct Scsi_Host *shost; 600 struct iser_conn *iser_conn = NULL; 601 struct ib_conn *ib_conn; 602 struct ib_device *ib_dev; 603 u32 max_fr_sectors; 604 605 shost = iscsi_host_alloc(&iscsi_iser_sht, 0, 0); 606 if (!shost) 607 return NULL; 608 shost->transportt = iscsi_iser_scsi_transport; 609 shost->cmd_per_lun = qdepth; 610 shost->max_lun = iscsi_max_lun; 611 shost->max_id = 0; 612 shost->max_channel = 0; 613 shost->max_cmd_len = 16; 614 615 /* 616 * older userspace tools (before 2.0-870) did not pass us 617 * the leading conn's ep so this will be NULL; 618 */ 619 if (ep) { 620 iser_conn = ep->dd_data; 621 shost->sg_tablesize = iser_conn->scsi_sg_tablesize; 622 shost->can_queue = min_t(u16, cmds_max, iser_conn->max_cmds); 623 624 mutex_lock(&iser_conn->state_mutex); 625 if (iser_conn->state != ISER_CONN_UP) { 626 iser_err("iser conn %p already started teardown\n", 627 iser_conn); 628 mutex_unlock(&iser_conn->state_mutex); 629 goto free_host; 630 } 631 632 ib_conn = &iser_conn->ib_conn; 633 ib_dev = ib_conn->device->ib_device; 634 if (ib_conn->pi_support) { 635 u32 sig_caps = ib_dev->attrs.sig_prot_cap; 636 637 shost->sg_prot_tablesize = shost->sg_tablesize; 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 if (!(ib_dev->attrs.device_cap_flags & IB_DEVICE_SG_GAPS_REG)) 644 shost->virt_boundary_mask = SZ_4K - 1; 645 646 if (iscsi_host_add(shost, ib_dev->dev.parent)) { 647 mutex_unlock(&iser_conn->state_mutex); 648 goto free_host; 649 } 650 mutex_unlock(&iser_conn->state_mutex); 651 } else { 652 shost->can_queue = min_t(u16, cmds_max, ISER_DEF_XMIT_CMDS_MAX); 653 if (iscsi_host_add(shost, NULL)) 654 goto free_host; 655 } 656 657 max_fr_sectors = (shost->sg_tablesize * PAGE_SIZE) >> 9; 658 shost->max_sectors = min(iser_max_sectors, max_fr_sectors); 659 660 iser_dbg("iser_conn %p, sg_tablesize %u, max_sectors %u\n", 661 iser_conn, shost->sg_tablesize, 662 shost->max_sectors); 663 664 if (shost->max_sectors < iser_max_sectors) 665 iser_warn("max_sectors was reduced from %u to %u\n", 666 iser_max_sectors, shost->max_sectors); 667 668 cls_session = iscsi_session_setup(&iscsi_iser_transport, shost, 669 shost->can_queue, 0, 670 sizeof(struct iscsi_iser_task), 671 initial_cmdsn, 0); 672 if (!cls_session) 673 goto remove_host; 674 675 return cls_session; 676 677 remove_host: 678 iscsi_host_remove(shost); 679 free_host: 680 iscsi_host_free(shost); 681 return NULL; 682 } 683 684 static int 685 iscsi_iser_set_param(struct iscsi_cls_conn *cls_conn, 686 enum iscsi_param param, char *buf, int buflen) 687 { 688 int value; 689 690 switch (param) { 691 case ISCSI_PARAM_MAX_RECV_DLENGTH: 692 /* TBD */ 693 break; 694 case ISCSI_PARAM_HDRDGST_EN: 695 sscanf(buf, "%d", &value); 696 if (value) { 697 iser_err("DataDigest wasn't negotiated to None\n"); 698 return -EPROTO; 699 } 700 break; 701 case ISCSI_PARAM_DATADGST_EN: 702 sscanf(buf, "%d", &value); 703 if (value) { 704 iser_err("DataDigest wasn't negotiated to None\n"); 705 return -EPROTO; 706 } 707 break; 708 case ISCSI_PARAM_IFMARKER_EN: 709 sscanf(buf, "%d", &value); 710 if (value) { 711 iser_err("IFMarker wasn't negotiated to No\n"); 712 return -EPROTO; 713 } 714 break; 715 case ISCSI_PARAM_OFMARKER_EN: 716 sscanf(buf, "%d", &value); 717 if (value) { 718 iser_err("OFMarker wasn't negotiated to No\n"); 719 return -EPROTO; 720 } 721 break; 722 default: 723 return iscsi_set_param(cls_conn, param, buf, buflen); 724 } 725 726 return 0; 727 } 728 729 /** 730 * iscsi_iser_conn_get_stats() - get iscsi connection statistics 731 * @cls_conn: iscsi class connection 732 * @stats: iscsi stats to output 733 * 734 * Output connection statistics. 735 */ 736 static void 737 iscsi_iser_conn_get_stats(struct iscsi_cls_conn *cls_conn, struct iscsi_stats *stats) 738 { 739 struct iscsi_conn *conn = cls_conn->dd_data; 740 741 stats->txdata_octets = conn->txdata_octets; 742 stats->rxdata_octets = conn->rxdata_octets; 743 stats->scsicmd_pdus = conn->scsicmd_pdus_cnt; 744 stats->dataout_pdus = conn->dataout_pdus_cnt; 745 stats->scsirsp_pdus = conn->scsirsp_pdus_cnt; 746 stats->datain_pdus = conn->datain_pdus_cnt; /* always 0 */ 747 stats->r2t_pdus = conn->r2t_pdus_cnt; /* always 0 */ 748 stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt; 749 stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt; 750 stats->custom_length = 0; 751 } 752 753 static int iscsi_iser_get_ep_param(struct iscsi_endpoint *ep, 754 enum iscsi_param param, char *buf) 755 { 756 struct iser_conn *iser_conn = ep->dd_data; 757 758 switch (param) { 759 case ISCSI_PARAM_CONN_PORT: 760 case ISCSI_PARAM_CONN_ADDRESS: 761 if (!iser_conn || !iser_conn->ib_conn.cma_id) 762 return -ENOTCONN; 763 764 return iscsi_conn_get_addr_param((struct sockaddr_storage *) 765 &iser_conn->ib_conn.cma_id->route.addr.dst_addr, 766 param, buf); 767 default: 768 break; 769 } 770 return -ENOSYS; 771 } 772 773 /** 774 * iscsi_iser_ep_connect() - Initiate iSER connection establishment 775 * @shost: scsi_host 776 * @dst_addr: destination address 777 * @non_blocking: indicate if routine can block 778 * 779 * Allocate an iscsi endpoint, an iser_conn structure and bind them. 780 * After that start RDMA connection establishment via rdma_cm. We 781 * don't allocate iser_conn embedded in iscsi_endpoint since in teardown 782 * the endpoint will be destroyed at ep_disconnect while iser_conn will 783 * cleanup its resources asynchronuously. 784 * 785 * Return: iscsi_endpoint created by iscsi layer or ERR_PTR(error) 786 * if fails. 787 */ 788 static struct iscsi_endpoint * 789 iscsi_iser_ep_connect(struct Scsi_Host *shost, struct sockaddr *dst_addr, 790 int non_blocking) 791 { 792 int err; 793 struct iser_conn *iser_conn; 794 struct iscsi_endpoint *ep; 795 796 ep = iscsi_create_endpoint(0); 797 if (!ep) 798 return ERR_PTR(-ENOMEM); 799 800 iser_conn = kzalloc(sizeof(*iser_conn), GFP_KERNEL); 801 if (!iser_conn) { 802 err = -ENOMEM; 803 goto failure; 804 } 805 806 ep->dd_data = iser_conn; 807 iser_conn->ep = ep; 808 iser_conn_init(iser_conn); 809 810 err = iser_connect(iser_conn, NULL, dst_addr, non_blocking); 811 if (err) 812 goto failure; 813 814 return ep; 815 failure: 816 iscsi_destroy_endpoint(ep); 817 return ERR_PTR(err); 818 } 819 820 /** 821 * iscsi_iser_ep_poll() - poll for iser connection establishment to complete 822 * @ep: iscsi endpoint (created at ep_connect) 823 * @timeout_ms: polling timeout allowed in ms. 824 * 825 * This routine boils down to waiting for up_completion signaling 826 * that cma_id got CONNECTED event. 827 * 828 * Return: 1 if succeeded in connection establishment, 0 if timeout expired 829 * (libiscsi will retry will kick in) or -1 if interrupted by signal 830 * or more likely iser connection state transitioned to TEMINATING or 831 * DOWN during the wait period. 832 */ 833 static int 834 iscsi_iser_ep_poll(struct iscsi_endpoint *ep, int timeout_ms) 835 { 836 struct iser_conn *iser_conn = ep->dd_data; 837 int rc; 838 839 rc = wait_for_completion_interruptible_timeout(&iser_conn->up_completion, 840 msecs_to_jiffies(timeout_ms)); 841 /* if conn establishment failed, return error code to iscsi */ 842 if (rc == 0) { 843 mutex_lock(&iser_conn->state_mutex); 844 if (iser_conn->state == ISER_CONN_TERMINATING || 845 iser_conn->state == ISER_CONN_DOWN) 846 rc = -1; 847 mutex_unlock(&iser_conn->state_mutex); 848 } 849 850 iser_info("iser conn %p rc = %d\n", iser_conn, rc); 851 852 if (rc > 0) 853 return 1; /* success, this is the equivalent of EPOLLOUT */ 854 else if (!rc) 855 return 0; /* timeout */ 856 else 857 return rc; /* signal */ 858 } 859 860 /** 861 * iscsi_iser_ep_disconnect() - Initiate connection teardown process 862 * @ep: iscsi endpoint handle 863 * 864 * This routine is not blocked by iser and RDMA termination process 865 * completion as we queue a deffered work for iser/RDMA destruction 866 * and cleanup or actually call it immediately in case we didn't pass 867 * iscsi conn bind/start stage, thus it is safe. 868 */ 869 static void 870 iscsi_iser_ep_disconnect(struct iscsi_endpoint *ep) 871 { 872 struct iser_conn *iser_conn = ep->dd_data; 873 874 iser_info("ep %p iser conn %p\n", ep, iser_conn); 875 876 mutex_lock(&iser_conn->state_mutex); 877 iser_conn_terminate(iser_conn); 878 879 /* 880 * if iser_conn and iscsi_conn are bound, we must wait for 881 * iscsi_conn_stop and flush errors completion before freeing 882 * the iser resources. Otherwise we are safe to free resources 883 * immediately. 884 */ 885 if (iser_conn->iscsi_conn) { 886 INIT_WORK(&iser_conn->release_work, iser_release_work); 887 queue_work(release_wq, &iser_conn->release_work); 888 mutex_unlock(&iser_conn->state_mutex); 889 } else { 890 iser_conn->state = ISER_CONN_DOWN; 891 mutex_unlock(&iser_conn->state_mutex); 892 iser_conn_release(iser_conn); 893 } 894 895 iscsi_destroy_endpoint(ep); 896 } 897 898 static umode_t iser_attr_is_visible(int param_type, int param) 899 { 900 switch (param_type) { 901 case ISCSI_HOST_PARAM: 902 switch (param) { 903 case ISCSI_HOST_PARAM_NETDEV_NAME: 904 case ISCSI_HOST_PARAM_HWADDRESS: 905 case ISCSI_HOST_PARAM_INITIATOR_NAME: 906 return S_IRUGO; 907 default: 908 return 0; 909 } 910 case ISCSI_PARAM: 911 switch (param) { 912 case ISCSI_PARAM_MAX_RECV_DLENGTH: 913 case ISCSI_PARAM_MAX_XMIT_DLENGTH: 914 case ISCSI_PARAM_HDRDGST_EN: 915 case ISCSI_PARAM_DATADGST_EN: 916 case ISCSI_PARAM_CONN_ADDRESS: 917 case ISCSI_PARAM_CONN_PORT: 918 case ISCSI_PARAM_EXP_STATSN: 919 case ISCSI_PARAM_PERSISTENT_ADDRESS: 920 case ISCSI_PARAM_PERSISTENT_PORT: 921 case ISCSI_PARAM_PING_TMO: 922 case ISCSI_PARAM_RECV_TMO: 923 case ISCSI_PARAM_INITIAL_R2T_EN: 924 case ISCSI_PARAM_MAX_R2T: 925 case ISCSI_PARAM_IMM_DATA_EN: 926 case ISCSI_PARAM_FIRST_BURST: 927 case ISCSI_PARAM_MAX_BURST: 928 case ISCSI_PARAM_PDU_INORDER_EN: 929 case ISCSI_PARAM_DATASEQ_INORDER_EN: 930 case ISCSI_PARAM_TARGET_NAME: 931 case ISCSI_PARAM_TPGT: 932 case ISCSI_PARAM_USERNAME: 933 case ISCSI_PARAM_PASSWORD: 934 case ISCSI_PARAM_USERNAME_IN: 935 case ISCSI_PARAM_PASSWORD_IN: 936 case ISCSI_PARAM_FAST_ABORT: 937 case ISCSI_PARAM_ABORT_TMO: 938 case ISCSI_PARAM_LU_RESET_TMO: 939 case ISCSI_PARAM_TGT_RESET_TMO: 940 case ISCSI_PARAM_IFACE_NAME: 941 case ISCSI_PARAM_INITIATOR_NAME: 942 case ISCSI_PARAM_DISCOVERY_SESS: 943 return S_IRUGO; 944 default: 945 return 0; 946 } 947 } 948 949 return 0; 950 } 951 952 static struct scsi_host_template iscsi_iser_sht = { 953 .module = THIS_MODULE, 954 .name = "iSCSI Initiator over iSER", 955 .queuecommand = iscsi_queuecommand, 956 .change_queue_depth = scsi_change_queue_depth, 957 .sg_tablesize = ISCSI_ISER_DEF_SG_TABLESIZE, 958 .cmd_per_lun = ISER_DEF_CMD_PER_LUN, 959 .eh_timed_out = iscsi_eh_cmd_timed_out, 960 .eh_abort_handler = iscsi_eh_abort, 961 .eh_device_reset_handler= iscsi_eh_device_reset, 962 .eh_target_reset_handler = iscsi_eh_recover_target, 963 .target_alloc = iscsi_target_alloc, 964 .proc_name = "iscsi_iser", 965 .this_id = -1, 966 .track_queue_depth = 1, 967 }; 968 969 static struct iscsi_transport iscsi_iser_transport = { 970 .owner = THIS_MODULE, 971 .name = "iser", 972 .caps = CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_TEXT_NEGO, 973 /* session management */ 974 .create_session = iscsi_iser_session_create, 975 .destroy_session = iscsi_iser_session_destroy, 976 /* connection management */ 977 .create_conn = iscsi_iser_conn_create, 978 .bind_conn = iscsi_iser_conn_bind, 979 .destroy_conn = iscsi_conn_teardown, 980 .attr_is_visible = iser_attr_is_visible, 981 .set_param = iscsi_iser_set_param, 982 .get_conn_param = iscsi_conn_get_param, 983 .get_ep_param = iscsi_iser_get_ep_param, 984 .get_session_param = iscsi_session_get_param, 985 .start_conn = iscsi_iser_conn_start, 986 .stop_conn = iscsi_iser_conn_stop, 987 /* iscsi host params */ 988 .get_host_param = iscsi_host_get_param, 989 .set_host_param = iscsi_host_set_param, 990 /* IO */ 991 .send_pdu = iscsi_conn_send_pdu, 992 .get_stats = iscsi_iser_conn_get_stats, 993 .init_task = iscsi_iser_task_init, 994 .xmit_task = iscsi_iser_task_xmit, 995 .cleanup_task = iscsi_iser_cleanup_task, 996 .alloc_pdu = iscsi_iser_pdu_alloc, 997 .check_protection = iscsi_iser_check_protection, 998 /* recovery */ 999 .session_recovery_timedout = iscsi_session_recovery_timedout, 1000 1001 .ep_connect = iscsi_iser_ep_connect, 1002 .ep_poll = iscsi_iser_ep_poll, 1003 .ep_disconnect = iscsi_iser_ep_disconnect 1004 }; 1005 1006 static int __init iser_init(void) 1007 { 1008 int err; 1009 1010 iser_dbg("Starting iSER datamover...\n"); 1011 1012 if (iscsi_max_lun < 1) { 1013 iser_err("Invalid max_lun value of %u\n", iscsi_max_lun); 1014 return -EINVAL; 1015 } 1016 1017 memset(&ig, 0, sizeof(struct iser_global)); 1018 1019 ig.desc_cache = kmem_cache_create("iser_descriptors", 1020 sizeof(struct iser_tx_desc), 1021 0, SLAB_HWCACHE_ALIGN, 1022 NULL); 1023 if (ig.desc_cache == NULL) 1024 return -ENOMEM; 1025 1026 /* device init is called only after the first addr resolution */ 1027 mutex_init(&ig.device_list_mutex); 1028 INIT_LIST_HEAD(&ig.device_list); 1029 mutex_init(&ig.connlist_mutex); 1030 INIT_LIST_HEAD(&ig.connlist); 1031 1032 release_wq = alloc_workqueue("release workqueue", 0, 0); 1033 if (!release_wq) { 1034 iser_err("failed to allocate release workqueue\n"); 1035 err = -ENOMEM; 1036 goto err_alloc_wq; 1037 } 1038 1039 iscsi_iser_scsi_transport = iscsi_register_transport( 1040 &iscsi_iser_transport); 1041 if (!iscsi_iser_scsi_transport) { 1042 iser_err("iscsi_register_transport failed\n"); 1043 err = -EINVAL; 1044 goto err_reg; 1045 } 1046 1047 return 0; 1048 1049 err_reg: 1050 destroy_workqueue(release_wq); 1051 err_alloc_wq: 1052 kmem_cache_destroy(ig.desc_cache); 1053 1054 return err; 1055 } 1056 1057 static void __exit iser_exit(void) 1058 { 1059 struct iser_conn *iser_conn, *n; 1060 int connlist_empty; 1061 1062 iser_dbg("Removing iSER datamover...\n"); 1063 destroy_workqueue(release_wq); 1064 1065 mutex_lock(&ig.connlist_mutex); 1066 connlist_empty = list_empty(&ig.connlist); 1067 mutex_unlock(&ig.connlist_mutex); 1068 1069 if (!connlist_empty) { 1070 iser_err("Error cleanup stage completed but we still have iser " 1071 "connections, destroying them anyway\n"); 1072 list_for_each_entry_safe(iser_conn, n, &ig.connlist, 1073 conn_list) { 1074 iser_conn_release(iser_conn); 1075 } 1076 } 1077 1078 iscsi_unregister_transport(&iscsi_iser_transport); 1079 kmem_cache_destroy(ig.desc_cache); 1080 } 1081 1082 module_init(iser_init); 1083 module_exit(iser_exit); 1084