1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * iSCSI Initiator over TCP/IP Data-Path 4 * 5 * Copyright (C) 2004 Dmitry Yusupov 6 * Copyright (C) 2004 Alex Aizman 7 * Copyright (C) 2005 - 2006 Mike Christie 8 * Copyright (C) 2006 Red Hat, Inc. All rights reserved. 9 * maintained by open-iscsi@googlegroups.com 10 * 11 * See the file COPYING included with this distribution for more details. 12 * 13 * Credits: 14 * Christoph Hellwig 15 * FUJITA Tomonori 16 * Arne Redlich 17 * Zhenyu Wang 18 */ 19 20 #include <crypto/hash.h> 21 #include <linux/types.h> 22 #include <linux/inet.h> 23 #include <linux/slab.h> 24 #include <linux/sched/mm.h> 25 #include <linux/file.h> 26 #include <linux/blkdev.h> 27 #include <linux/delay.h> 28 #include <linux/kfifo.h> 29 #include <linux/scatterlist.h> 30 #include <linux/module.h> 31 #include <linux/backing-dev.h> 32 #include <net/tcp.h> 33 #include <scsi/scsi_cmnd.h> 34 #include <scsi/scsi_device.h> 35 #include <scsi/scsi_host.h> 36 #include <scsi/scsi.h> 37 #include <scsi/scsi_transport_iscsi.h> 38 #include <trace/events/iscsi.h> 39 #include <trace/events/sock.h> 40 41 #include "iscsi_tcp.h" 42 43 MODULE_AUTHOR("Mike Christie <michaelc@cs.wisc.edu>, " 44 "Dmitry Yusupov <dmitry_yus@yahoo.com>, " 45 "Alex Aizman <itn780@yahoo.com>"); 46 MODULE_DESCRIPTION("iSCSI/TCP data-path"); 47 MODULE_LICENSE("GPL"); 48 49 static struct scsi_transport_template *iscsi_sw_tcp_scsi_transport; 50 static const struct scsi_host_template iscsi_sw_tcp_sht; 51 static struct iscsi_transport iscsi_sw_tcp_transport; 52 53 static unsigned int iscsi_max_lun = ~0; 54 module_param_named(max_lun, iscsi_max_lun, uint, S_IRUGO); 55 56 static bool iscsi_recv_from_iscsi_q; 57 module_param_named(recv_from_iscsi_q, iscsi_recv_from_iscsi_q, bool, 0644); 58 MODULE_PARM_DESC(recv_from_iscsi_q, "Set to true to read iSCSI data/headers from the iscsi_q workqueue. The default is false which will perform reads from the network softirq context."); 59 60 static int iscsi_sw_tcp_dbg; 61 module_param_named(debug_iscsi_tcp, iscsi_sw_tcp_dbg, int, 62 S_IRUGO | S_IWUSR); 63 MODULE_PARM_DESC(debug_iscsi_tcp, "Turn on debugging for iscsi_tcp module " 64 "Set to 1 to turn on, and zero to turn off. Default is off."); 65 66 #define ISCSI_SW_TCP_DBG(_conn, dbg_fmt, arg...) \ 67 do { \ 68 if (iscsi_sw_tcp_dbg) \ 69 iscsi_conn_printk(KERN_INFO, _conn, \ 70 "%s " dbg_fmt, \ 71 __func__, ##arg); \ 72 iscsi_dbg_trace(trace_iscsi_dbg_sw_tcp, \ 73 &(_conn)->cls_conn->dev, \ 74 "%s " dbg_fmt, __func__, ##arg);\ 75 } while (0); 76 77 78 /** 79 * iscsi_sw_tcp_recv - TCP receive in sendfile fashion 80 * @rd_desc: read descriptor 81 * @skb: socket buffer 82 * @offset: offset in skb 83 * @len: skb->len - offset 84 */ 85 static int iscsi_sw_tcp_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, 86 unsigned int offset, size_t len) 87 { 88 struct iscsi_conn *conn = rd_desc->arg.data; 89 unsigned int consumed, total_consumed = 0; 90 int status; 91 92 ISCSI_SW_TCP_DBG(conn, "in %d bytes\n", skb->len - offset); 93 94 do { 95 status = 0; 96 consumed = iscsi_tcp_recv_skb(conn, skb, offset, 0, &status); 97 offset += consumed; 98 total_consumed += consumed; 99 } while (consumed != 0 && status != ISCSI_TCP_SKB_DONE); 100 101 ISCSI_SW_TCP_DBG(conn, "read %d bytes status %d\n", 102 skb->len - offset, status); 103 return total_consumed; 104 } 105 106 /** 107 * iscsi_sw_sk_state_check - check socket state 108 * @sk: socket 109 * 110 * If the socket is in CLOSE or CLOSE_WAIT we should 111 * not close the connection if there is still some 112 * data pending. 113 * 114 * Must be called with sk_callback_lock. 115 */ 116 static inline int iscsi_sw_sk_state_check(struct sock *sk) 117 { 118 struct iscsi_conn *conn = sk->sk_user_data; 119 120 if ((sk->sk_state == TCP_CLOSE_WAIT || sk->sk_state == TCP_CLOSE) && 121 (conn->session->state != ISCSI_STATE_LOGGING_OUT) && 122 !atomic_read(&sk->sk_rmem_alloc)) { 123 ISCSI_SW_TCP_DBG(conn, "TCP_CLOSE|TCP_CLOSE_WAIT\n"); 124 iscsi_conn_failure(conn, ISCSI_ERR_TCP_CONN_CLOSE); 125 return -ECONNRESET; 126 } 127 return 0; 128 } 129 130 static void iscsi_sw_tcp_recv_data(struct iscsi_conn *conn) 131 { 132 struct iscsi_tcp_conn *tcp_conn = conn->dd_data; 133 struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data; 134 struct sock *sk = tcp_sw_conn->sock->sk; 135 read_descriptor_t rd_desc; 136 137 /* 138 * Use rd_desc to pass 'conn' to iscsi_tcp_recv. 139 * We set count to 1 because we want the network layer to 140 * hand us all the skbs that are available. iscsi_tcp_recv 141 * handled pdus that cross buffers or pdus that still need data. 142 */ 143 rd_desc.arg.data = conn; 144 rd_desc.count = 1; 145 146 tcp_read_sock(sk, &rd_desc, iscsi_sw_tcp_recv); 147 148 /* If we had to (atomically) map a highmem page, 149 * unmap it now. */ 150 iscsi_tcp_segment_unmap(&tcp_conn->in.segment); 151 152 iscsi_sw_sk_state_check(sk); 153 } 154 155 static void iscsi_sw_tcp_recv_data_work(struct work_struct *work) 156 { 157 struct iscsi_conn *conn = container_of(work, struct iscsi_conn, 158 recvwork); 159 struct iscsi_tcp_conn *tcp_conn = conn->dd_data; 160 struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data; 161 struct sock *sk = tcp_sw_conn->sock->sk; 162 163 lock_sock(sk); 164 iscsi_sw_tcp_recv_data(conn); 165 release_sock(sk); 166 } 167 168 static void iscsi_sw_tcp_data_ready(struct sock *sk) 169 { 170 struct iscsi_sw_tcp_conn *tcp_sw_conn; 171 struct iscsi_tcp_conn *tcp_conn; 172 struct iscsi_conn *conn; 173 174 trace_sk_data_ready(sk); 175 176 read_lock_bh(&sk->sk_callback_lock); 177 conn = sk->sk_user_data; 178 if (!conn) { 179 read_unlock_bh(&sk->sk_callback_lock); 180 return; 181 } 182 tcp_conn = conn->dd_data; 183 tcp_sw_conn = tcp_conn->dd_data; 184 185 if (tcp_sw_conn->queue_recv) 186 iscsi_conn_queue_recv(conn); 187 else 188 iscsi_sw_tcp_recv_data(conn); 189 read_unlock_bh(&sk->sk_callback_lock); 190 } 191 192 static void iscsi_sw_tcp_state_change(struct sock *sk) 193 { 194 struct iscsi_tcp_conn *tcp_conn; 195 struct iscsi_sw_tcp_conn *tcp_sw_conn; 196 struct iscsi_conn *conn; 197 void (*old_state_change)(struct sock *); 198 199 read_lock_bh(&sk->sk_callback_lock); 200 conn = sk->sk_user_data; 201 if (!conn) { 202 read_unlock_bh(&sk->sk_callback_lock); 203 return; 204 } 205 206 iscsi_sw_sk_state_check(sk); 207 208 tcp_conn = conn->dd_data; 209 tcp_sw_conn = tcp_conn->dd_data; 210 old_state_change = tcp_sw_conn->old_state_change; 211 212 read_unlock_bh(&sk->sk_callback_lock); 213 214 old_state_change(sk); 215 } 216 217 /** 218 * iscsi_sw_tcp_write_space - Called when more output buffer space is available 219 * @sk: socket space is available for 220 **/ 221 static void iscsi_sw_tcp_write_space(struct sock *sk) 222 { 223 struct iscsi_conn *conn; 224 struct iscsi_tcp_conn *tcp_conn; 225 struct iscsi_sw_tcp_conn *tcp_sw_conn; 226 void (*old_write_space)(struct sock *); 227 228 read_lock_bh(&sk->sk_callback_lock); 229 conn = sk->sk_user_data; 230 if (!conn) { 231 read_unlock_bh(&sk->sk_callback_lock); 232 return; 233 } 234 235 tcp_conn = conn->dd_data; 236 tcp_sw_conn = tcp_conn->dd_data; 237 old_write_space = tcp_sw_conn->old_write_space; 238 read_unlock_bh(&sk->sk_callback_lock); 239 240 old_write_space(sk); 241 242 ISCSI_SW_TCP_DBG(conn, "iscsi_write_space\n"); 243 iscsi_conn_queue_xmit(conn); 244 } 245 246 static void iscsi_sw_tcp_conn_set_callbacks(struct iscsi_conn *conn) 247 { 248 struct iscsi_tcp_conn *tcp_conn = conn->dd_data; 249 struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data; 250 struct sock *sk = tcp_sw_conn->sock->sk; 251 252 /* assign new callbacks */ 253 write_lock_bh(&sk->sk_callback_lock); 254 sk->sk_user_data = conn; 255 tcp_sw_conn->old_data_ready = sk->sk_data_ready; 256 tcp_sw_conn->old_state_change = sk->sk_state_change; 257 tcp_sw_conn->old_write_space = sk->sk_write_space; 258 sk->sk_data_ready = iscsi_sw_tcp_data_ready; 259 sk->sk_state_change = iscsi_sw_tcp_state_change; 260 sk->sk_write_space = iscsi_sw_tcp_write_space; 261 write_unlock_bh(&sk->sk_callback_lock); 262 } 263 264 static void 265 iscsi_sw_tcp_conn_restore_callbacks(struct iscsi_conn *conn) 266 { 267 struct iscsi_tcp_conn *tcp_conn = conn->dd_data; 268 struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data; 269 struct sock *sk = tcp_sw_conn->sock->sk; 270 271 /* restore socket callbacks, see also: iscsi_conn_set_callbacks() */ 272 write_lock_bh(&sk->sk_callback_lock); 273 sk->sk_user_data = NULL; 274 sk->sk_data_ready = tcp_sw_conn->old_data_ready; 275 sk->sk_state_change = tcp_sw_conn->old_state_change; 276 sk->sk_write_space = tcp_sw_conn->old_write_space; 277 sk->sk_no_check_tx = 0; 278 write_unlock_bh(&sk->sk_callback_lock); 279 } 280 281 /** 282 * iscsi_sw_tcp_xmit_segment - transmit segment 283 * @tcp_conn: the iSCSI TCP connection 284 * @segment: the buffer to transmnit 285 * 286 * This function transmits as much of the buffer as 287 * the network layer will accept, and returns the number of 288 * bytes transmitted. 289 * 290 * If CRC hashing is enabled, the function will compute the 291 * hash as it goes. When the entire segment has been transmitted, 292 * it will retrieve the hash value and send it as well. 293 */ 294 static int iscsi_sw_tcp_xmit_segment(struct iscsi_tcp_conn *tcp_conn, 295 struct iscsi_segment *segment) 296 { 297 struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data; 298 struct socket *sk = tcp_sw_conn->sock; 299 unsigned int copied = 0; 300 int r = 0; 301 302 while (!iscsi_tcp_segment_done(tcp_conn, segment, 0, r)) { 303 struct scatterlist *sg; 304 unsigned int offset, copy; 305 int flags = 0; 306 307 r = 0; 308 offset = segment->copied; 309 copy = segment->size - offset; 310 311 if (segment->total_copied + segment->size < segment->total_size) 312 flags |= MSG_MORE | MSG_SENDPAGE_NOTLAST; 313 314 if (tcp_sw_conn->queue_recv) 315 flags |= MSG_DONTWAIT; 316 317 /* Use sendpage if we can; else fall back to sendmsg */ 318 if (!segment->data) { 319 sg = segment->sg; 320 offset += segment->sg_offset + sg->offset; 321 r = tcp_sw_conn->sendpage(sk, sg_page(sg), offset, 322 copy, flags); 323 } else { 324 struct msghdr msg = { .msg_flags = flags }; 325 struct kvec iov = { 326 .iov_base = segment->data + offset, 327 .iov_len = copy 328 }; 329 330 r = kernel_sendmsg(sk, &msg, &iov, 1, copy); 331 } 332 333 if (r < 0) { 334 iscsi_tcp_segment_unmap(segment); 335 return r; 336 } 337 copied += r; 338 } 339 return copied; 340 } 341 342 /** 343 * iscsi_sw_tcp_xmit - TCP transmit 344 * @conn: iscsi connection 345 **/ 346 static int iscsi_sw_tcp_xmit(struct iscsi_conn *conn) 347 { 348 struct iscsi_tcp_conn *tcp_conn = conn->dd_data; 349 struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data; 350 struct iscsi_segment *segment = &tcp_sw_conn->out.segment; 351 unsigned int consumed = 0; 352 int rc = 0; 353 354 while (1) { 355 rc = iscsi_sw_tcp_xmit_segment(tcp_conn, segment); 356 /* 357 * We may not have been able to send data because the conn 358 * is getting stopped. libiscsi will know so propagate err 359 * for it to do the right thing. 360 */ 361 if (rc == -EAGAIN) 362 return rc; 363 else if (rc < 0) { 364 rc = ISCSI_ERR_XMIT_FAILED; 365 goto error; 366 } else if (rc == 0) 367 break; 368 369 consumed += rc; 370 371 if (segment->total_copied >= segment->total_size) { 372 if (segment->done != NULL) { 373 rc = segment->done(tcp_conn, segment); 374 if (rc != 0) 375 goto error; 376 } 377 } 378 } 379 380 ISCSI_SW_TCP_DBG(conn, "xmit %d bytes\n", consumed); 381 382 conn->txdata_octets += consumed; 383 return consumed; 384 385 error: 386 /* Transmit error. We could initiate error recovery 387 * here. */ 388 ISCSI_SW_TCP_DBG(conn, "Error sending PDU, errno=%d\n", rc); 389 iscsi_conn_failure(conn, rc); 390 return -EIO; 391 } 392 393 /** 394 * iscsi_sw_tcp_xmit_qlen - return the number of bytes queued for xmit 395 * @conn: iscsi connection 396 */ 397 static inline int iscsi_sw_tcp_xmit_qlen(struct iscsi_conn *conn) 398 { 399 struct iscsi_tcp_conn *tcp_conn = conn->dd_data; 400 struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data; 401 struct iscsi_segment *segment = &tcp_sw_conn->out.segment; 402 403 return segment->total_copied - segment->total_size; 404 } 405 406 static int iscsi_sw_tcp_pdu_xmit(struct iscsi_task *task) 407 { 408 struct iscsi_conn *conn = task->conn; 409 unsigned int noreclaim_flag; 410 struct iscsi_tcp_conn *tcp_conn = conn->dd_data; 411 struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data; 412 int rc = 0; 413 414 if (!tcp_sw_conn->sock) { 415 iscsi_conn_printk(KERN_ERR, conn, 416 "Transport not bound to socket!\n"); 417 return -EINVAL; 418 } 419 420 noreclaim_flag = memalloc_noreclaim_save(); 421 422 while (iscsi_sw_tcp_xmit_qlen(conn)) { 423 rc = iscsi_sw_tcp_xmit(conn); 424 if (rc == 0) { 425 rc = -EAGAIN; 426 break; 427 } 428 if (rc < 0) 429 break; 430 rc = 0; 431 } 432 433 memalloc_noreclaim_restore(noreclaim_flag); 434 return rc; 435 } 436 437 /* 438 * This is called when we're done sending the header. 439 * Simply copy the data_segment to the send segment, and return. 440 */ 441 static int iscsi_sw_tcp_send_hdr_done(struct iscsi_tcp_conn *tcp_conn, 442 struct iscsi_segment *segment) 443 { 444 struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data; 445 446 tcp_sw_conn->out.segment = tcp_sw_conn->out.data_segment; 447 ISCSI_SW_TCP_DBG(tcp_conn->iscsi_conn, 448 "Header done. Next segment size %u total_size %u\n", 449 tcp_sw_conn->out.segment.size, 450 tcp_sw_conn->out.segment.total_size); 451 return 0; 452 } 453 454 static void iscsi_sw_tcp_send_hdr_prep(struct iscsi_conn *conn, void *hdr, 455 size_t hdrlen) 456 { 457 struct iscsi_tcp_conn *tcp_conn = conn->dd_data; 458 struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data; 459 460 ISCSI_SW_TCP_DBG(conn, "%s\n", conn->hdrdgst_en ? 461 "digest enabled" : "digest disabled"); 462 463 /* Clear the data segment - needs to be filled in by the 464 * caller using iscsi_tcp_send_data_prep() */ 465 memset(&tcp_sw_conn->out.data_segment, 0, 466 sizeof(struct iscsi_segment)); 467 468 /* If header digest is enabled, compute the CRC and 469 * place the digest into the same buffer. We make 470 * sure that both iscsi_tcp_task and mtask have 471 * sufficient room. 472 */ 473 if (conn->hdrdgst_en) { 474 iscsi_tcp_dgst_header(tcp_sw_conn->tx_hash, hdr, hdrlen, 475 hdr + hdrlen); 476 hdrlen += ISCSI_DIGEST_SIZE; 477 } 478 479 /* Remember header pointer for later, when we need 480 * to decide whether there's a payload to go along 481 * with the header. */ 482 tcp_sw_conn->out.hdr = hdr; 483 484 iscsi_segment_init_linear(&tcp_sw_conn->out.segment, hdr, hdrlen, 485 iscsi_sw_tcp_send_hdr_done, NULL); 486 } 487 488 /* 489 * Prepare the send buffer for the payload data. 490 * Padding and checksumming will all be taken care 491 * of by the iscsi_segment routines. 492 */ 493 static int 494 iscsi_sw_tcp_send_data_prep(struct iscsi_conn *conn, struct scatterlist *sg, 495 unsigned int count, unsigned int offset, 496 unsigned int len) 497 { 498 struct iscsi_tcp_conn *tcp_conn = conn->dd_data; 499 struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data; 500 struct ahash_request *tx_hash = NULL; 501 unsigned int hdr_spec_len; 502 503 ISCSI_SW_TCP_DBG(conn, "offset=%d, datalen=%d %s\n", offset, len, 504 conn->datadgst_en ? 505 "digest enabled" : "digest disabled"); 506 507 /* Make sure the datalen matches what the caller 508 said he would send. */ 509 hdr_spec_len = ntoh24(tcp_sw_conn->out.hdr->dlength); 510 WARN_ON(iscsi_padded(len) != iscsi_padded(hdr_spec_len)); 511 512 if (conn->datadgst_en) 513 tx_hash = tcp_sw_conn->tx_hash; 514 515 return iscsi_segment_seek_sg(&tcp_sw_conn->out.data_segment, 516 sg, count, offset, len, 517 NULL, tx_hash); 518 } 519 520 static void 521 iscsi_sw_tcp_send_linear_data_prep(struct iscsi_conn *conn, void *data, 522 size_t len) 523 { 524 struct iscsi_tcp_conn *tcp_conn = conn->dd_data; 525 struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data; 526 struct ahash_request *tx_hash = NULL; 527 unsigned int hdr_spec_len; 528 529 ISCSI_SW_TCP_DBG(conn, "datalen=%zd %s\n", len, conn->datadgst_en ? 530 "digest enabled" : "digest disabled"); 531 532 /* Make sure the datalen matches what the caller 533 said he would send. */ 534 hdr_spec_len = ntoh24(tcp_sw_conn->out.hdr->dlength); 535 WARN_ON(iscsi_padded(len) != iscsi_padded(hdr_spec_len)); 536 537 if (conn->datadgst_en) 538 tx_hash = tcp_sw_conn->tx_hash; 539 540 iscsi_segment_init_linear(&tcp_sw_conn->out.data_segment, 541 data, len, NULL, tx_hash); 542 } 543 544 static int iscsi_sw_tcp_pdu_init(struct iscsi_task *task, 545 unsigned int offset, unsigned int count) 546 { 547 struct iscsi_conn *conn = task->conn; 548 int err = 0; 549 550 iscsi_sw_tcp_send_hdr_prep(conn, task->hdr, task->hdr_len); 551 552 if (!count) 553 return 0; 554 555 if (!task->sc) 556 iscsi_sw_tcp_send_linear_data_prep(conn, task->data, count); 557 else { 558 struct scsi_data_buffer *sdb = &task->sc->sdb; 559 560 err = iscsi_sw_tcp_send_data_prep(conn, sdb->table.sgl, 561 sdb->table.nents, offset, 562 count); 563 } 564 565 if (err) { 566 /* got invalid offset/len */ 567 return -EIO; 568 } 569 return 0; 570 } 571 572 static int iscsi_sw_tcp_pdu_alloc(struct iscsi_task *task, uint8_t opcode) 573 { 574 struct iscsi_tcp_task *tcp_task = task->dd_data; 575 576 task->hdr = task->dd_data + sizeof(*tcp_task); 577 task->hdr_max = sizeof(struct iscsi_sw_tcp_hdrbuf) - ISCSI_DIGEST_SIZE; 578 return 0; 579 } 580 581 static struct iscsi_cls_conn * 582 iscsi_sw_tcp_conn_create(struct iscsi_cls_session *cls_session, 583 uint32_t conn_idx) 584 { 585 struct iscsi_conn *conn; 586 struct iscsi_cls_conn *cls_conn; 587 struct iscsi_tcp_conn *tcp_conn; 588 struct iscsi_sw_tcp_conn *tcp_sw_conn; 589 struct crypto_ahash *tfm; 590 591 cls_conn = iscsi_tcp_conn_setup(cls_session, sizeof(*tcp_sw_conn), 592 conn_idx); 593 if (!cls_conn) 594 return NULL; 595 conn = cls_conn->dd_data; 596 tcp_conn = conn->dd_data; 597 tcp_sw_conn = tcp_conn->dd_data; 598 INIT_WORK(&conn->recvwork, iscsi_sw_tcp_recv_data_work); 599 tcp_sw_conn->queue_recv = iscsi_recv_from_iscsi_q; 600 601 mutex_init(&tcp_sw_conn->sock_lock); 602 603 tfm = crypto_alloc_ahash("crc32c", 0, CRYPTO_ALG_ASYNC); 604 if (IS_ERR(tfm)) 605 goto free_conn; 606 607 tcp_sw_conn->tx_hash = ahash_request_alloc(tfm, GFP_KERNEL); 608 if (!tcp_sw_conn->tx_hash) 609 goto free_tfm; 610 ahash_request_set_callback(tcp_sw_conn->tx_hash, 0, NULL, NULL); 611 612 tcp_sw_conn->rx_hash = ahash_request_alloc(tfm, GFP_KERNEL); 613 if (!tcp_sw_conn->rx_hash) 614 goto free_tx_hash; 615 ahash_request_set_callback(tcp_sw_conn->rx_hash, 0, NULL, NULL); 616 617 tcp_conn->rx_hash = tcp_sw_conn->rx_hash; 618 619 return cls_conn; 620 621 free_tx_hash: 622 ahash_request_free(tcp_sw_conn->tx_hash); 623 free_tfm: 624 crypto_free_ahash(tfm); 625 free_conn: 626 iscsi_conn_printk(KERN_ERR, conn, 627 "Could not create connection due to crc32c " 628 "loading error. Make sure the crc32c " 629 "module is built as a module or into the " 630 "kernel\n"); 631 iscsi_tcp_conn_teardown(cls_conn); 632 return NULL; 633 } 634 635 static void iscsi_sw_tcp_release_conn(struct iscsi_conn *conn) 636 { 637 struct iscsi_tcp_conn *tcp_conn = conn->dd_data; 638 struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data; 639 struct socket *sock = tcp_sw_conn->sock; 640 641 /* 642 * The iscsi transport class will make sure we are not called in 643 * parallel with start, stop, bind and destroys. However, this can be 644 * called twice if userspace does a stop then a destroy. 645 */ 646 if (!sock) 647 return; 648 649 /* 650 * Make sure we start socket shutdown now in case userspace is up 651 * but delayed in releasing the socket. 652 */ 653 kernel_sock_shutdown(sock, SHUT_RDWR); 654 655 sock_hold(sock->sk); 656 iscsi_sw_tcp_conn_restore_callbacks(conn); 657 sock_put(sock->sk); 658 659 iscsi_suspend_rx(conn); 660 661 mutex_lock(&tcp_sw_conn->sock_lock); 662 tcp_sw_conn->sock = NULL; 663 mutex_unlock(&tcp_sw_conn->sock_lock); 664 sockfd_put(sock); 665 } 666 667 static void iscsi_sw_tcp_conn_destroy(struct iscsi_cls_conn *cls_conn) 668 { 669 struct iscsi_conn *conn = cls_conn->dd_data; 670 struct iscsi_tcp_conn *tcp_conn = conn->dd_data; 671 struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data; 672 673 iscsi_sw_tcp_release_conn(conn); 674 675 ahash_request_free(tcp_sw_conn->rx_hash); 676 if (tcp_sw_conn->tx_hash) { 677 struct crypto_ahash *tfm; 678 679 tfm = crypto_ahash_reqtfm(tcp_sw_conn->tx_hash); 680 ahash_request_free(tcp_sw_conn->tx_hash); 681 crypto_free_ahash(tfm); 682 } 683 684 iscsi_tcp_conn_teardown(cls_conn); 685 } 686 687 static void iscsi_sw_tcp_conn_stop(struct iscsi_cls_conn *cls_conn, int flag) 688 { 689 struct iscsi_conn *conn = cls_conn->dd_data; 690 struct iscsi_tcp_conn *tcp_conn = conn->dd_data; 691 struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data; 692 struct socket *sock = tcp_sw_conn->sock; 693 694 /* userspace may have goofed up and not bound us */ 695 if (!sock) 696 return; 697 698 sock->sk->sk_err = EIO; 699 wake_up_interruptible(sk_sleep(sock->sk)); 700 701 /* stop xmit side */ 702 iscsi_suspend_tx(conn); 703 704 /* stop recv side and release socket */ 705 iscsi_sw_tcp_release_conn(conn); 706 707 iscsi_conn_stop(cls_conn, flag); 708 } 709 710 static int 711 iscsi_sw_tcp_conn_bind(struct iscsi_cls_session *cls_session, 712 struct iscsi_cls_conn *cls_conn, uint64_t transport_eph, 713 int is_leading) 714 { 715 struct iscsi_conn *conn = cls_conn->dd_data; 716 struct iscsi_tcp_conn *tcp_conn = conn->dd_data; 717 struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data; 718 struct sock *sk; 719 struct socket *sock; 720 int err; 721 722 /* lookup for existing socket */ 723 sock = sockfd_lookup((int)transport_eph, &err); 724 if (!sock) { 725 iscsi_conn_printk(KERN_ERR, conn, 726 "sockfd_lookup failed %d\n", err); 727 return -EEXIST; 728 } 729 730 err = iscsi_conn_bind(cls_session, cls_conn, is_leading); 731 if (err) 732 goto free_socket; 733 734 mutex_lock(&tcp_sw_conn->sock_lock); 735 /* bind iSCSI connection and socket */ 736 tcp_sw_conn->sock = sock; 737 mutex_unlock(&tcp_sw_conn->sock_lock); 738 739 /* setup Socket parameters */ 740 sk = sock->sk; 741 sk->sk_reuse = SK_CAN_REUSE; 742 sk->sk_sndtimeo = 15 * HZ; /* FIXME: make it configurable */ 743 sk->sk_allocation = GFP_ATOMIC; 744 sk->sk_use_task_frag = false; 745 sk_set_memalloc(sk); 746 sock_no_linger(sk); 747 748 iscsi_sw_tcp_conn_set_callbacks(conn); 749 tcp_sw_conn->sendpage = tcp_sw_conn->sock->ops->sendpage; 750 /* 751 * set receive state machine into initial state 752 */ 753 iscsi_tcp_hdr_recv_prep(tcp_conn); 754 return 0; 755 756 free_socket: 757 sockfd_put(sock); 758 return err; 759 } 760 761 static int iscsi_sw_tcp_conn_set_param(struct iscsi_cls_conn *cls_conn, 762 enum iscsi_param param, char *buf, 763 int buflen) 764 { 765 struct iscsi_conn *conn = cls_conn->dd_data; 766 struct iscsi_tcp_conn *tcp_conn = conn->dd_data; 767 struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data; 768 769 switch(param) { 770 case ISCSI_PARAM_HDRDGST_EN: 771 iscsi_set_param(cls_conn, param, buf, buflen); 772 break; 773 case ISCSI_PARAM_DATADGST_EN: 774 iscsi_set_param(cls_conn, param, buf, buflen); 775 776 mutex_lock(&tcp_sw_conn->sock_lock); 777 if (!tcp_sw_conn->sock) { 778 mutex_unlock(&tcp_sw_conn->sock_lock); 779 return -ENOTCONN; 780 } 781 tcp_sw_conn->sendpage = conn->datadgst_en ? 782 sock_no_sendpage : tcp_sw_conn->sock->ops->sendpage; 783 mutex_unlock(&tcp_sw_conn->sock_lock); 784 break; 785 case ISCSI_PARAM_MAX_R2T: 786 return iscsi_tcp_set_max_r2t(conn, buf); 787 default: 788 return iscsi_set_param(cls_conn, param, buf, buflen); 789 } 790 791 return 0; 792 } 793 794 static int iscsi_sw_tcp_conn_get_param(struct iscsi_cls_conn *cls_conn, 795 enum iscsi_param param, char *buf) 796 { 797 struct iscsi_conn *conn = cls_conn->dd_data; 798 struct iscsi_sw_tcp_conn *tcp_sw_conn; 799 struct iscsi_tcp_conn *tcp_conn; 800 struct sockaddr_in6 addr; 801 struct socket *sock; 802 int rc; 803 804 switch(param) { 805 case ISCSI_PARAM_CONN_PORT: 806 case ISCSI_PARAM_CONN_ADDRESS: 807 case ISCSI_PARAM_LOCAL_PORT: 808 spin_lock_bh(&conn->session->frwd_lock); 809 if (!conn->session->leadconn) { 810 spin_unlock_bh(&conn->session->frwd_lock); 811 return -ENOTCONN; 812 } 813 /* 814 * The conn has been setup and bound, so just grab a ref 815 * incase a destroy runs while we are in the net layer. 816 */ 817 iscsi_get_conn(conn->cls_conn); 818 spin_unlock_bh(&conn->session->frwd_lock); 819 820 tcp_conn = conn->dd_data; 821 tcp_sw_conn = tcp_conn->dd_data; 822 823 mutex_lock(&tcp_sw_conn->sock_lock); 824 sock = tcp_sw_conn->sock; 825 if (!sock) { 826 rc = -ENOTCONN; 827 goto sock_unlock; 828 } 829 830 if (param == ISCSI_PARAM_LOCAL_PORT) 831 rc = kernel_getsockname(sock, 832 (struct sockaddr *)&addr); 833 else 834 rc = kernel_getpeername(sock, 835 (struct sockaddr *)&addr); 836 sock_unlock: 837 mutex_unlock(&tcp_sw_conn->sock_lock); 838 iscsi_put_conn(conn->cls_conn); 839 if (rc < 0) 840 return rc; 841 842 return iscsi_conn_get_addr_param((struct sockaddr_storage *) 843 &addr, param, buf); 844 default: 845 return iscsi_conn_get_param(cls_conn, param, buf); 846 } 847 848 return 0; 849 } 850 851 static int iscsi_sw_tcp_host_get_param(struct Scsi_Host *shost, 852 enum iscsi_host_param param, char *buf) 853 { 854 struct iscsi_sw_tcp_host *tcp_sw_host = iscsi_host_priv(shost); 855 struct iscsi_session *session; 856 struct iscsi_conn *conn; 857 struct iscsi_tcp_conn *tcp_conn; 858 struct iscsi_sw_tcp_conn *tcp_sw_conn; 859 struct sockaddr_in6 addr; 860 struct socket *sock; 861 int rc; 862 863 switch (param) { 864 case ISCSI_HOST_PARAM_IPADDRESS: 865 session = tcp_sw_host->session; 866 if (!session) 867 return -ENOTCONN; 868 869 spin_lock_bh(&session->frwd_lock); 870 conn = session->leadconn; 871 if (!conn) { 872 spin_unlock_bh(&session->frwd_lock); 873 return -ENOTCONN; 874 } 875 tcp_conn = conn->dd_data; 876 tcp_sw_conn = tcp_conn->dd_data; 877 /* 878 * The conn has been setup and bound, so just grab a ref 879 * incase a destroy runs while we are in the net layer. 880 */ 881 iscsi_get_conn(conn->cls_conn); 882 spin_unlock_bh(&session->frwd_lock); 883 884 mutex_lock(&tcp_sw_conn->sock_lock); 885 sock = tcp_sw_conn->sock; 886 if (!sock) 887 rc = -ENOTCONN; 888 else 889 rc = kernel_getsockname(sock, (struct sockaddr *)&addr); 890 mutex_unlock(&tcp_sw_conn->sock_lock); 891 iscsi_put_conn(conn->cls_conn); 892 if (rc < 0) 893 return rc; 894 895 return iscsi_conn_get_addr_param((struct sockaddr_storage *) 896 &addr, 897 (enum iscsi_param)param, buf); 898 default: 899 return iscsi_host_get_param(shost, param, buf); 900 } 901 902 return 0; 903 } 904 905 static void 906 iscsi_sw_tcp_conn_get_stats(struct iscsi_cls_conn *cls_conn, 907 struct iscsi_stats *stats) 908 { 909 struct iscsi_conn *conn = cls_conn->dd_data; 910 struct iscsi_tcp_conn *tcp_conn = conn->dd_data; 911 struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data; 912 913 stats->custom_length = 3; 914 strcpy(stats->custom[0].desc, "tx_sendpage_failures"); 915 stats->custom[0].value = tcp_sw_conn->sendpage_failures_cnt; 916 strcpy(stats->custom[1].desc, "rx_discontiguous_hdr"); 917 stats->custom[1].value = tcp_sw_conn->discontiguous_hdr_cnt; 918 strcpy(stats->custom[2].desc, "eh_abort_cnt"); 919 stats->custom[2].value = conn->eh_abort_cnt; 920 921 iscsi_tcp_conn_get_stats(cls_conn, stats); 922 } 923 924 static struct iscsi_cls_session * 925 iscsi_sw_tcp_session_create(struct iscsi_endpoint *ep, uint16_t cmds_max, 926 uint16_t qdepth, uint32_t initial_cmdsn) 927 { 928 struct iscsi_cls_session *cls_session; 929 struct iscsi_session *session; 930 struct iscsi_sw_tcp_host *tcp_sw_host; 931 struct Scsi_Host *shost; 932 int rc; 933 934 if (ep) { 935 printk(KERN_ERR "iscsi_tcp: invalid ep %p.\n", ep); 936 return NULL; 937 } 938 939 shost = iscsi_host_alloc(&iscsi_sw_tcp_sht, 940 sizeof(struct iscsi_sw_tcp_host), 1); 941 if (!shost) 942 return NULL; 943 shost->transportt = iscsi_sw_tcp_scsi_transport; 944 shost->cmd_per_lun = qdepth; 945 shost->max_lun = iscsi_max_lun; 946 shost->max_id = 0; 947 shost->max_channel = 0; 948 shost->max_cmd_len = SCSI_MAX_VARLEN_CDB_SIZE; 949 950 rc = iscsi_host_get_max_scsi_cmds(shost, cmds_max); 951 if (rc < 0) 952 goto free_host; 953 shost->can_queue = rc; 954 955 if (iscsi_host_add(shost, NULL)) 956 goto free_host; 957 958 cls_session = iscsi_session_setup(&iscsi_sw_tcp_transport, shost, 959 cmds_max, 0, 960 sizeof(struct iscsi_tcp_task) + 961 sizeof(struct iscsi_sw_tcp_hdrbuf), 962 initial_cmdsn, 0); 963 if (!cls_session) 964 goto remove_host; 965 session = cls_session->dd_data; 966 967 if (iscsi_tcp_r2tpool_alloc(session)) 968 goto remove_session; 969 970 /* We are now fully setup so expose the session to sysfs. */ 971 tcp_sw_host = iscsi_host_priv(shost); 972 tcp_sw_host->session = session; 973 return cls_session; 974 975 remove_session: 976 iscsi_session_teardown(cls_session); 977 remove_host: 978 iscsi_host_remove(shost, false); 979 free_host: 980 iscsi_host_free(shost); 981 return NULL; 982 } 983 984 static void iscsi_sw_tcp_session_destroy(struct iscsi_cls_session *cls_session) 985 { 986 struct Scsi_Host *shost = iscsi_session_to_shost(cls_session); 987 struct iscsi_session *session = cls_session->dd_data; 988 989 if (WARN_ON_ONCE(session->leadconn)) 990 return; 991 992 iscsi_session_remove(cls_session); 993 /* 994 * Our get_host_param needs to access the session, so remove the 995 * host from sysfs before freeing the session to make sure userspace 996 * is no longer accessing the callout. 997 */ 998 iscsi_host_remove(shost, false); 999 1000 iscsi_tcp_r2tpool_free(cls_session->dd_data); 1001 1002 iscsi_session_free(cls_session); 1003 iscsi_host_free(shost); 1004 } 1005 1006 static umode_t iscsi_sw_tcp_attr_is_visible(int param_type, int param) 1007 { 1008 switch (param_type) { 1009 case ISCSI_HOST_PARAM: 1010 switch (param) { 1011 case ISCSI_HOST_PARAM_NETDEV_NAME: 1012 case ISCSI_HOST_PARAM_HWADDRESS: 1013 case ISCSI_HOST_PARAM_IPADDRESS: 1014 case ISCSI_HOST_PARAM_INITIATOR_NAME: 1015 return S_IRUGO; 1016 default: 1017 return 0; 1018 } 1019 case ISCSI_PARAM: 1020 switch (param) { 1021 case ISCSI_PARAM_MAX_RECV_DLENGTH: 1022 case ISCSI_PARAM_MAX_XMIT_DLENGTH: 1023 case ISCSI_PARAM_HDRDGST_EN: 1024 case ISCSI_PARAM_DATADGST_EN: 1025 case ISCSI_PARAM_CONN_ADDRESS: 1026 case ISCSI_PARAM_CONN_PORT: 1027 case ISCSI_PARAM_LOCAL_PORT: 1028 case ISCSI_PARAM_EXP_STATSN: 1029 case ISCSI_PARAM_PERSISTENT_ADDRESS: 1030 case ISCSI_PARAM_PERSISTENT_PORT: 1031 case ISCSI_PARAM_PING_TMO: 1032 case ISCSI_PARAM_RECV_TMO: 1033 case ISCSI_PARAM_INITIAL_R2T_EN: 1034 case ISCSI_PARAM_MAX_R2T: 1035 case ISCSI_PARAM_IMM_DATA_EN: 1036 case ISCSI_PARAM_FIRST_BURST: 1037 case ISCSI_PARAM_MAX_BURST: 1038 case ISCSI_PARAM_PDU_INORDER_EN: 1039 case ISCSI_PARAM_DATASEQ_INORDER_EN: 1040 case ISCSI_PARAM_ERL: 1041 case ISCSI_PARAM_TARGET_NAME: 1042 case ISCSI_PARAM_TPGT: 1043 case ISCSI_PARAM_USERNAME: 1044 case ISCSI_PARAM_PASSWORD: 1045 case ISCSI_PARAM_USERNAME_IN: 1046 case ISCSI_PARAM_PASSWORD_IN: 1047 case ISCSI_PARAM_FAST_ABORT: 1048 case ISCSI_PARAM_ABORT_TMO: 1049 case ISCSI_PARAM_LU_RESET_TMO: 1050 case ISCSI_PARAM_TGT_RESET_TMO: 1051 case ISCSI_PARAM_IFACE_NAME: 1052 case ISCSI_PARAM_INITIATOR_NAME: 1053 return S_IRUGO; 1054 default: 1055 return 0; 1056 } 1057 } 1058 1059 return 0; 1060 } 1061 1062 static int iscsi_sw_tcp_slave_configure(struct scsi_device *sdev) 1063 { 1064 struct iscsi_sw_tcp_host *tcp_sw_host = iscsi_host_priv(sdev->host); 1065 struct iscsi_session *session = tcp_sw_host->session; 1066 struct iscsi_conn *conn = session->leadconn; 1067 1068 if (conn->datadgst_en) 1069 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, 1070 sdev->request_queue); 1071 blk_queue_dma_alignment(sdev->request_queue, 0); 1072 return 0; 1073 } 1074 1075 static const struct scsi_host_template iscsi_sw_tcp_sht = { 1076 .module = THIS_MODULE, 1077 .name = "iSCSI Initiator over TCP/IP", 1078 .queuecommand = iscsi_queuecommand, 1079 .change_queue_depth = scsi_change_queue_depth, 1080 .can_queue = ISCSI_TOTAL_CMDS_MAX, 1081 .sg_tablesize = 4096, 1082 .max_sectors = 0xFFFF, 1083 .cmd_per_lun = ISCSI_DEF_CMD_PER_LUN, 1084 .eh_timed_out = iscsi_eh_cmd_timed_out, 1085 .eh_abort_handler = iscsi_eh_abort, 1086 .eh_device_reset_handler= iscsi_eh_device_reset, 1087 .eh_target_reset_handler = iscsi_eh_recover_target, 1088 .dma_boundary = PAGE_SIZE - 1, 1089 .slave_configure = iscsi_sw_tcp_slave_configure, 1090 .proc_name = "iscsi_tcp", 1091 .this_id = -1, 1092 .track_queue_depth = 1, 1093 .cmd_size = sizeof(struct iscsi_cmd), 1094 }; 1095 1096 static struct iscsi_transport iscsi_sw_tcp_transport = { 1097 .owner = THIS_MODULE, 1098 .name = "tcp", 1099 .caps = CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_HDRDGST 1100 | CAP_DATADGST, 1101 /* session management */ 1102 .create_session = iscsi_sw_tcp_session_create, 1103 .destroy_session = iscsi_sw_tcp_session_destroy, 1104 /* connection management */ 1105 .create_conn = iscsi_sw_tcp_conn_create, 1106 .bind_conn = iscsi_sw_tcp_conn_bind, 1107 .destroy_conn = iscsi_sw_tcp_conn_destroy, 1108 .attr_is_visible = iscsi_sw_tcp_attr_is_visible, 1109 .set_param = iscsi_sw_tcp_conn_set_param, 1110 .get_conn_param = iscsi_sw_tcp_conn_get_param, 1111 .get_session_param = iscsi_session_get_param, 1112 .start_conn = iscsi_conn_start, 1113 .stop_conn = iscsi_sw_tcp_conn_stop, 1114 /* iscsi host params */ 1115 .get_host_param = iscsi_sw_tcp_host_get_param, 1116 .set_host_param = iscsi_host_set_param, 1117 /* IO */ 1118 .send_pdu = iscsi_conn_send_pdu, 1119 .get_stats = iscsi_sw_tcp_conn_get_stats, 1120 /* iscsi task/cmd helpers */ 1121 .init_task = iscsi_tcp_task_init, 1122 .xmit_task = iscsi_tcp_task_xmit, 1123 .cleanup_task = iscsi_tcp_cleanup_task, 1124 /* low level pdu helpers */ 1125 .xmit_pdu = iscsi_sw_tcp_pdu_xmit, 1126 .init_pdu = iscsi_sw_tcp_pdu_init, 1127 .alloc_pdu = iscsi_sw_tcp_pdu_alloc, 1128 /* recovery */ 1129 .session_recovery_timedout = iscsi_session_recovery_timedout, 1130 }; 1131 1132 static int __init iscsi_sw_tcp_init(void) 1133 { 1134 if (iscsi_max_lun < 1) { 1135 printk(KERN_ERR "iscsi_tcp: Invalid max_lun value of %u\n", 1136 iscsi_max_lun); 1137 return -EINVAL; 1138 } 1139 1140 iscsi_sw_tcp_scsi_transport = iscsi_register_transport( 1141 &iscsi_sw_tcp_transport); 1142 if (!iscsi_sw_tcp_scsi_transport) 1143 return -ENODEV; 1144 1145 return 0; 1146 } 1147 1148 static void __exit iscsi_sw_tcp_exit(void) 1149 { 1150 iscsi_unregister_transport(&iscsi_sw_tcp_transport); 1151 } 1152 1153 module_init(iscsi_sw_tcp_init); 1154 module_exit(iscsi_sw_tcp_exit); 1155