1 /* 2 * libcxgbi.c: Chelsio common library for T3/T4 iSCSI driver. 3 * 4 * Copyright (c) 2010 Chelsio Communications, Inc. 5 * 6 * This program is free software; you can redistribute it and/or modify 7 * it under the terms of the GNU General Public License as published by 8 * the Free Software Foundation. 9 * 10 * Written by: Karen Xie (kxie@chelsio.com) 11 * Written by: Rakesh Ranjan (rranjan@chelsio.com) 12 */ 13 14 #define pr_fmt(fmt) KBUILD_MODNAME ":%s: " fmt, __func__ 15 16 #include <linux/skbuff.h> 17 #include <linux/crypto.h> 18 #include <linux/scatterlist.h> 19 #include <linux/pci.h> 20 #include <scsi/scsi.h> 21 #include <scsi/scsi_cmnd.h> 22 #include <scsi/scsi_host.h> 23 #include <linux/if_vlan.h> 24 #include <linux/inet.h> 25 #include <net/dst.h> 26 #include <net/route.h> 27 #include <net/ipv6.h> 28 #include <net/ip6_route.h> 29 #include <net/addrconf.h> 30 31 #include <linux/inetdevice.h> /* ip_dev_find */ 32 #include <linux/module.h> 33 #include <net/tcp.h> 34 35 static unsigned int dbg_level; 36 37 #include "libcxgbi.h" 38 39 #define DRV_MODULE_NAME "libcxgbi" 40 #define DRV_MODULE_DESC "Chelsio iSCSI driver library" 41 #define DRV_MODULE_VERSION "0.9.0" 42 #define DRV_MODULE_RELDATE "Jun. 2010" 43 44 MODULE_AUTHOR("Chelsio Communications, Inc."); 45 MODULE_DESCRIPTION(DRV_MODULE_DESC); 46 MODULE_VERSION(DRV_MODULE_VERSION); 47 MODULE_LICENSE("GPL"); 48 49 module_param(dbg_level, uint, 0644); 50 MODULE_PARM_DESC(dbg_level, "libiscsi debug level (default=0)"); 51 52 53 /* 54 * cxgbi device management 55 * maintains a list of the cxgbi devices 56 */ 57 static LIST_HEAD(cdev_list); 58 static DEFINE_MUTEX(cdev_mutex); 59 60 static LIST_HEAD(cdev_rcu_list); 61 static DEFINE_SPINLOCK(cdev_rcu_lock); 62 63 int cxgbi_device_portmap_create(struct cxgbi_device *cdev, unsigned int base, 64 unsigned int max_conn) 65 { 66 struct cxgbi_ports_map *pmap = &cdev->pmap; 67 68 pmap->port_csk = cxgbi_alloc_big_mem(max_conn * 69 sizeof(struct cxgbi_sock *), 70 GFP_KERNEL); 71 if (!pmap->port_csk) { 72 pr_warn("cdev 0x%p, portmap OOM %u.\n", cdev, max_conn); 73 return -ENOMEM; 74 } 75 76 pmap->max_connect = max_conn; 77 pmap->sport_base = base; 78 spin_lock_init(&pmap->lock); 79 return 0; 80 } 81 EXPORT_SYMBOL_GPL(cxgbi_device_portmap_create); 82 83 void cxgbi_device_portmap_cleanup(struct cxgbi_device *cdev) 84 { 85 struct cxgbi_ports_map *pmap = &cdev->pmap; 86 struct cxgbi_sock *csk; 87 int i; 88 89 for (i = 0; i < pmap->max_connect; i++) { 90 if (pmap->port_csk[i]) { 91 csk = pmap->port_csk[i]; 92 pmap->port_csk[i] = NULL; 93 log_debug(1 << CXGBI_DBG_SOCK, 94 "csk 0x%p, cdev 0x%p, offload down.\n", 95 csk, cdev); 96 spin_lock_bh(&csk->lock); 97 cxgbi_sock_set_flag(csk, CTPF_OFFLOAD_DOWN); 98 cxgbi_sock_closed(csk); 99 spin_unlock_bh(&csk->lock); 100 cxgbi_sock_put(csk); 101 } 102 } 103 } 104 EXPORT_SYMBOL_GPL(cxgbi_device_portmap_cleanup); 105 106 static inline void cxgbi_device_destroy(struct cxgbi_device *cdev) 107 { 108 log_debug(1 << CXGBI_DBG_DEV, 109 "cdev 0x%p, p# %u.\n", cdev, cdev->nports); 110 cxgbi_hbas_remove(cdev); 111 cxgbi_device_portmap_cleanup(cdev); 112 if (cdev->dev_ddp_cleanup) 113 cdev->dev_ddp_cleanup(cdev); 114 else 115 cxgbi_ddp_cleanup(cdev); 116 if (cdev->ddp) 117 cxgbi_ddp_cleanup(cdev); 118 if (cdev->pmap.max_connect) 119 cxgbi_free_big_mem(cdev->pmap.port_csk); 120 kfree(cdev); 121 } 122 123 struct cxgbi_device *cxgbi_device_register(unsigned int extra, 124 unsigned int nports) 125 { 126 struct cxgbi_device *cdev; 127 128 cdev = kzalloc(sizeof(*cdev) + extra + nports * 129 (sizeof(struct cxgbi_hba *) + 130 sizeof(struct net_device *)), 131 GFP_KERNEL); 132 if (!cdev) { 133 pr_warn("nport %d, OOM.\n", nports); 134 return NULL; 135 } 136 cdev->ports = (struct net_device **)(cdev + 1); 137 cdev->hbas = (struct cxgbi_hba **)(((char*)cdev->ports) + nports * 138 sizeof(struct net_device *)); 139 if (extra) 140 cdev->dd_data = ((char *)cdev->hbas) + 141 nports * sizeof(struct cxgbi_hba *); 142 spin_lock_init(&cdev->pmap.lock); 143 144 mutex_lock(&cdev_mutex); 145 list_add_tail(&cdev->list_head, &cdev_list); 146 mutex_unlock(&cdev_mutex); 147 148 spin_lock(&cdev_rcu_lock); 149 list_add_tail_rcu(&cdev->rcu_node, &cdev_rcu_list); 150 spin_unlock(&cdev_rcu_lock); 151 152 log_debug(1 << CXGBI_DBG_DEV, 153 "cdev 0x%p, p# %u.\n", cdev, nports); 154 return cdev; 155 } 156 EXPORT_SYMBOL_GPL(cxgbi_device_register); 157 158 void cxgbi_device_unregister(struct cxgbi_device *cdev) 159 { 160 log_debug(1 << CXGBI_DBG_DEV, 161 "cdev 0x%p, p# %u,%s.\n", 162 cdev, cdev->nports, cdev->nports ? cdev->ports[0]->name : ""); 163 164 mutex_lock(&cdev_mutex); 165 list_del(&cdev->list_head); 166 mutex_unlock(&cdev_mutex); 167 168 spin_lock(&cdev_rcu_lock); 169 list_del_rcu(&cdev->rcu_node); 170 spin_unlock(&cdev_rcu_lock); 171 synchronize_rcu(); 172 173 cxgbi_device_destroy(cdev); 174 } 175 EXPORT_SYMBOL_GPL(cxgbi_device_unregister); 176 177 void cxgbi_device_unregister_all(unsigned int flag) 178 { 179 struct cxgbi_device *cdev, *tmp; 180 181 mutex_lock(&cdev_mutex); 182 list_for_each_entry_safe(cdev, tmp, &cdev_list, list_head) { 183 if ((cdev->flags & flag) == flag) { 184 mutex_unlock(&cdev_mutex); 185 cxgbi_device_unregister(cdev); 186 mutex_lock(&cdev_mutex); 187 } 188 } 189 mutex_unlock(&cdev_mutex); 190 } 191 EXPORT_SYMBOL_GPL(cxgbi_device_unregister_all); 192 193 struct cxgbi_device *cxgbi_device_find_by_lldev(void *lldev) 194 { 195 struct cxgbi_device *cdev, *tmp; 196 197 mutex_lock(&cdev_mutex); 198 list_for_each_entry_safe(cdev, tmp, &cdev_list, list_head) { 199 if (cdev->lldev == lldev) { 200 mutex_unlock(&cdev_mutex); 201 return cdev; 202 } 203 } 204 mutex_unlock(&cdev_mutex); 205 206 log_debug(1 << CXGBI_DBG_DEV, 207 "lldev 0x%p, NO match found.\n", lldev); 208 return NULL; 209 } 210 EXPORT_SYMBOL_GPL(cxgbi_device_find_by_lldev); 211 212 struct cxgbi_device *cxgbi_device_find_by_netdev(struct net_device *ndev, 213 int *port) 214 { 215 struct net_device *vdev = NULL; 216 struct cxgbi_device *cdev, *tmp; 217 int i; 218 219 if (ndev->priv_flags & IFF_802_1Q_VLAN) { 220 vdev = ndev; 221 ndev = vlan_dev_real_dev(ndev); 222 log_debug(1 << CXGBI_DBG_DEV, 223 "vlan dev %s -> %s.\n", vdev->name, ndev->name); 224 } 225 226 mutex_lock(&cdev_mutex); 227 list_for_each_entry_safe(cdev, tmp, &cdev_list, list_head) { 228 for (i = 0; i < cdev->nports; i++) { 229 if (ndev == cdev->ports[i]) { 230 cdev->hbas[i]->vdev = vdev; 231 mutex_unlock(&cdev_mutex); 232 if (port) 233 *port = i; 234 return cdev; 235 } 236 } 237 } 238 mutex_unlock(&cdev_mutex); 239 log_debug(1 << CXGBI_DBG_DEV, 240 "ndev 0x%p, %s, NO match found.\n", ndev, ndev->name); 241 return NULL; 242 } 243 EXPORT_SYMBOL_GPL(cxgbi_device_find_by_netdev); 244 245 struct cxgbi_device *cxgbi_device_find_by_netdev_rcu(struct net_device *ndev, 246 int *port) 247 { 248 struct net_device *vdev = NULL; 249 struct cxgbi_device *cdev; 250 int i; 251 252 if (ndev->priv_flags & IFF_802_1Q_VLAN) { 253 vdev = ndev; 254 ndev = vlan_dev_real_dev(ndev); 255 pr_info("vlan dev %s -> %s.\n", vdev->name, ndev->name); 256 } 257 258 rcu_read_lock(); 259 list_for_each_entry_rcu(cdev, &cdev_rcu_list, rcu_node) { 260 for (i = 0; i < cdev->nports; i++) { 261 if (ndev == cdev->ports[i]) { 262 cdev->hbas[i]->vdev = vdev; 263 rcu_read_unlock(); 264 if (port) 265 *port = i; 266 return cdev; 267 } 268 } 269 } 270 rcu_read_unlock(); 271 272 log_debug(1 << CXGBI_DBG_DEV, 273 "ndev 0x%p, %s, NO match found.\n", ndev, ndev->name); 274 return NULL; 275 } 276 EXPORT_SYMBOL_GPL(cxgbi_device_find_by_netdev_rcu); 277 278 #if IS_ENABLED(CONFIG_IPV6) 279 static struct cxgbi_device *cxgbi_device_find_by_mac(struct net_device *ndev, 280 int *port) 281 { 282 struct net_device *vdev = NULL; 283 struct cxgbi_device *cdev, *tmp; 284 int i; 285 286 if (ndev->priv_flags & IFF_802_1Q_VLAN) { 287 vdev = ndev; 288 ndev = vlan_dev_real_dev(ndev); 289 pr_info("vlan dev %s -> %s.\n", vdev->name, ndev->name); 290 } 291 292 mutex_lock(&cdev_mutex); 293 list_for_each_entry_safe(cdev, tmp, &cdev_list, list_head) { 294 for (i = 0; i < cdev->nports; i++) { 295 if (!memcmp(ndev->dev_addr, cdev->ports[i]->dev_addr, 296 MAX_ADDR_LEN)) { 297 cdev->hbas[i]->vdev = vdev; 298 mutex_unlock(&cdev_mutex); 299 if (port) 300 *port = i; 301 return cdev; 302 } 303 } 304 } 305 mutex_unlock(&cdev_mutex); 306 log_debug(1 << CXGBI_DBG_DEV, 307 "ndev 0x%p, %s, NO match mac found.\n", 308 ndev, ndev->name); 309 return NULL; 310 } 311 #endif 312 313 void cxgbi_hbas_remove(struct cxgbi_device *cdev) 314 { 315 int i; 316 struct cxgbi_hba *chba; 317 318 log_debug(1 << CXGBI_DBG_DEV, 319 "cdev 0x%p, p#%u.\n", cdev, cdev->nports); 320 321 for (i = 0; i < cdev->nports; i++) { 322 chba = cdev->hbas[i]; 323 if (chba) { 324 cdev->hbas[i] = NULL; 325 iscsi_host_remove(chba->shost); 326 pci_dev_put(cdev->pdev); 327 iscsi_host_free(chba->shost); 328 } 329 } 330 } 331 EXPORT_SYMBOL_GPL(cxgbi_hbas_remove); 332 333 int cxgbi_hbas_add(struct cxgbi_device *cdev, u64 max_lun, 334 unsigned int max_id, struct scsi_host_template *sht, 335 struct scsi_transport_template *stt) 336 { 337 struct cxgbi_hba *chba; 338 struct Scsi_Host *shost; 339 int i, err; 340 341 log_debug(1 << CXGBI_DBG_DEV, "cdev 0x%p, p#%u.\n", cdev, cdev->nports); 342 343 for (i = 0; i < cdev->nports; i++) { 344 shost = iscsi_host_alloc(sht, sizeof(*chba), 1); 345 if (!shost) { 346 pr_info("0x%p, p%d, %s, host alloc failed.\n", 347 cdev, i, cdev->ports[i]->name); 348 err = -ENOMEM; 349 goto err_out; 350 } 351 352 shost->transportt = stt; 353 shost->max_lun = max_lun; 354 shost->max_id = max_id; 355 shost->max_channel = 0; 356 shost->max_cmd_len = 16; 357 358 chba = iscsi_host_priv(shost); 359 chba->cdev = cdev; 360 chba->ndev = cdev->ports[i]; 361 chba->shost = shost; 362 363 log_debug(1 << CXGBI_DBG_DEV, 364 "cdev 0x%p, p#%d %s: chba 0x%p.\n", 365 cdev, i, cdev->ports[i]->name, chba); 366 367 pci_dev_get(cdev->pdev); 368 err = iscsi_host_add(shost, &cdev->pdev->dev); 369 if (err) { 370 pr_info("cdev 0x%p, p#%d %s, host add failed.\n", 371 cdev, i, cdev->ports[i]->name); 372 pci_dev_put(cdev->pdev); 373 scsi_host_put(shost); 374 goto err_out; 375 } 376 377 cdev->hbas[i] = chba; 378 } 379 380 return 0; 381 382 err_out: 383 cxgbi_hbas_remove(cdev); 384 return err; 385 } 386 EXPORT_SYMBOL_GPL(cxgbi_hbas_add); 387 388 /* 389 * iSCSI offload 390 * 391 * - source port management 392 * To find a free source port in the port allocation map we use a very simple 393 * rotor scheme to look for the next free port. 394 * 395 * If a source port has been specified make sure that it doesn't collide with 396 * our normal source port allocation map. If it's outside the range of our 397 * allocation/deallocation scheme just let them use it. 398 * 399 * If the source port is outside our allocation range, the caller is 400 * responsible for keeping track of their port usage. 401 */ 402 403 static struct cxgbi_sock *find_sock_on_port(struct cxgbi_device *cdev, 404 unsigned char port_id) 405 { 406 struct cxgbi_ports_map *pmap = &cdev->pmap; 407 unsigned int i; 408 unsigned int used; 409 410 if (!pmap->max_connect || !pmap->used) 411 return NULL; 412 413 spin_lock_bh(&pmap->lock); 414 used = pmap->used; 415 for (i = 0; used && i < pmap->max_connect; i++) { 416 struct cxgbi_sock *csk = pmap->port_csk[i]; 417 418 if (csk) { 419 if (csk->port_id == port_id) { 420 spin_unlock_bh(&pmap->lock); 421 return csk; 422 } 423 used--; 424 } 425 } 426 spin_unlock_bh(&pmap->lock); 427 428 return NULL; 429 } 430 431 static int sock_get_port(struct cxgbi_sock *csk) 432 { 433 struct cxgbi_device *cdev = csk->cdev; 434 struct cxgbi_ports_map *pmap = &cdev->pmap; 435 unsigned int start; 436 int idx; 437 __be16 *port; 438 439 if (!pmap->max_connect) { 440 pr_err("cdev 0x%p, p#%u %s, NO port map.\n", 441 cdev, csk->port_id, cdev->ports[csk->port_id]->name); 442 return -EADDRNOTAVAIL; 443 } 444 445 if (csk->csk_family == AF_INET) 446 port = &csk->saddr.sin_port; 447 else /* ipv6 */ 448 port = &csk->saddr6.sin6_port; 449 450 if (*port) { 451 pr_err("source port NON-ZERO %u.\n", 452 ntohs(*port)); 453 return -EADDRINUSE; 454 } 455 456 spin_lock_bh(&pmap->lock); 457 if (pmap->used >= pmap->max_connect) { 458 spin_unlock_bh(&pmap->lock); 459 pr_info("cdev 0x%p, p#%u %s, ALL ports used.\n", 460 cdev, csk->port_id, cdev->ports[csk->port_id]->name); 461 return -EADDRNOTAVAIL; 462 } 463 464 start = idx = pmap->next; 465 do { 466 if (++idx >= pmap->max_connect) 467 idx = 0; 468 if (!pmap->port_csk[idx]) { 469 pmap->used++; 470 *port = htons(pmap->sport_base + idx); 471 pmap->next = idx; 472 pmap->port_csk[idx] = csk; 473 spin_unlock_bh(&pmap->lock); 474 cxgbi_sock_get(csk); 475 log_debug(1 << CXGBI_DBG_SOCK, 476 "cdev 0x%p, p#%u %s, p %u, %u.\n", 477 cdev, csk->port_id, 478 cdev->ports[csk->port_id]->name, 479 pmap->sport_base + idx, pmap->next); 480 return 0; 481 } 482 } while (idx != start); 483 spin_unlock_bh(&pmap->lock); 484 485 /* should not happen */ 486 pr_warn("cdev 0x%p, p#%u %s, next %u?\n", 487 cdev, csk->port_id, cdev->ports[csk->port_id]->name, 488 pmap->next); 489 return -EADDRNOTAVAIL; 490 } 491 492 static void sock_put_port(struct cxgbi_sock *csk) 493 { 494 struct cxgbi_device *cdev = csk->cdev; 495 struct cxgbi_ports_map *pmap = &cdev->pmap; 496 __be16 *port; 497 498 if (csk->csk_family == AF_INET) 499 port = &csk->saddr.sin_port; 500 else /* ipv6 */ 501 port = &csk->saddr6.sin6_port; 502 503 if (*port) { 504 int idx = ntohs(*port) - pmap->sport_base; 505 506 *port = 0; 507 if (idx < 0 || idx >= pmap->max_connect) { 508 pr_err("cdev 0x%p, p#%u %s, port %u OOR.\n", 509 cdev, csk->port_id, 510 cdev->ports[csk->port_id]->name, 511 ntohs(*port)); 512 return; 513 } 514 515 spin_lock_bh(&pmap->lock); 516 pmap->port_csk[idx] = NULL; 517 pmap->used--; 518 spin_unlock_bh(&pmap->lock); 519 520 log_debug(1 << CXGBI_DBG_SOCK, 521 "cdev 0x%p, p#%u %s, release %u.\n", 522 cdev, csk->port_id, cdev->ports[csk->port_id]->name, 523 pmap->sport_base + idx); 524 525 cxgbi_sock_put(csk); 526 } 527 } 528 529 /* 530 * iscsi tcp connection 531 */ 532 void cxgbi_sock_free_cpl_skbs(struct cxgbi_sock *csk) 533 { 534 if (csk->cpl_close) { 535 kfree_skb(csk->cpl_close); 536 csk->cpl_close = NULL; 537 } 538 if (csk->cpl_abort_req) { 539 kfree_skb(csk->cpl_abort_req); 540 csk->cpl_abort_req = NULL; 541 } 542 if (csk->cpl_abort_rpl) { 543 kfree_skb(csk->cpl_abort_rpl); 544 csk->cpl_abort_rpl = NULL; 545 } 546 } 547 EXPORT_SYMBOL_GPL(cxgbi_sock_free_cpl_skbs); 548 549 static struct cxgbi_sock *cxgbi_sock_create(struct cxgbi_device *cdev) 550 { 551 struct cxgbi_sock *csk = kzalloc(sizeof(*csk), GFP_NOIO); 552 553 if (!csk) { 554 pr_info("alloc csk %zu failed.\n", sizeof(*csk)); 555 return NULL; 556 } 557 558 if (cdev->csk_alloc_cpls(csk) < 0) { 559 pr_info("csk 0x%p, alloc cpls failed.\n", csk); 560 kfree(csk); 561 return NULL; 562 } 563 564 spin_lock_init(&csk->lock); 565 kref_init(&csk->refcnt); 566 skb_queue_head_init(&csk->receive_queue); 567 skb_queue_head_init(&csk->write_queue); 568 setup_timer(&csk->retry_timer, NULL, (unsigned long)csk); 569 rwlock_init(&csk->callback_lock); 570 csk->cdev = cdev; 571 csk->flags = 0; 572 cxgbi_sock_set_state(csk, CTP_CLOSED); 573 574 log_debug(1 << CXGBI_DBG_SOCK, "cdev 0x%p, new csk 0x%p.\n", cdev, csk); 575 576 return csk; 577 } 578 579 static struct rtable *find_route_ipv4(struct flowi4 *fl4, 580 __be32 saddr, __be32 daddr, 581 __be16 sport, __be16 dport, u8 tos) 582 { 583 struct rtable *rt; 584 585 rt = ip_route_output_ports(&init_net, fl4, NULL, daddr, saddr, 586 dport, sport, IPPROTO_TCP, tos, 0); 587 if (IS_ERR(rt)) 588 return NULL; 589 590 return rt; 591 } 592 593 static struct cxgbi_sock *cxgbi_check_route(struct sockaddr *dst_addr) 594 { 595 struct sockaddr_in *daddr = (struct sockaddr_in *)dst_addr; 596 struct dst_entry *dst; 597 struct net_device *ndev; 598 struct cxgbi_device *cdev; 599 struct rtable *rt = NULL; 600 struct neighbour *n; 601 struct flowi4 fl4; 602 struct cxgbi_sock *csk = NULL; 603 unsigned int mtu = 0; 604 int port = 0xFFFF; 605 int err = 0; 606 607 rt = find_route_ipv4(&fl4, 0, daddr->sin_addr.s_addr, 0, daddr->sin_port, 0); 608 if (!rt) { 609 pr_info("no route to ipv4 0x%x, port %u.\n", 610 be32_to_cpu(daddr->sin_addr.s_addr), 611 be16_to_cpu(daddr->sin_port)); 612 err = -ENETUNREACH; 613 goto err_out; 614 } 615 dst = &rt->dst; 616 n = dst_neigh_lookup(dst, &daddr->sin_addr.s_addr); 617 if (!n) { 618 err = -ENODEV; 619 goto rel_rt; 620 } 621 ndev = n->dev; 622 623 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) { 624 pr_info("multi-cast route %pI4, port %u, dev %s.\n", 625 &daddr->sin_addr.s_addr, ntohs(daddr->sin_port), 626 ndev->name); 627 err = -ENETUNREACH; 628 goto rel_neigh; 629 } 630 631 if (ndev->flags & IFF_LOOPBACK) { 632 ndev = ip_dev_find(&init_net, daddr->sin_addr.s_addr); 633 mtu = ndev->mtu; 634 pr_info("rt dev %s, loopback -> %s, mtu %u.\n", 635 n->dev->name, ndev->name, mtu); 636 } 637 638 cdev = cxgbi_device_find_by_netdev(ndev, &port); 639 if (!cdev) { 640 pr_info("dst %pI4, %s, NOT cxgbi device.\n", 641 &daddr->sin_addr.s_addr, ndev->name); 642 err = -ENETUNREACH; 643 goto rel_neigh; 644 } 645 log_debug(1 << CXGBI_DBG_SOCK, 646 "route to %pI4 :%u, ndev p#%d,%s, cdev 0x%p.\n", 647 &daddr->sin_addr.s_addr, ntohs(daddr->sin_port), 648 port, ndev->name, cdev); 649 650 csk = cxgbi_sock_create(cdev); 651 if (!csk) { 652 err = -ENOMEM; 653 goto rel_neigh; 654 } 655 csk->cdev = cdev; 656 csk->port_id = port; 657 csk->mtu = mtu; 658 csk->dst = dst; 659 660 csk->csk_family = AF_INET; 661 csk->daddr.sin_addr.s_addr = daddr->sin_addr.s_addr; 662 csk->daddr.sin_port = daddr->sin_port; 663 csk->daddr.sin_family = daddr->sin_family; 664 csk->saddr.sin_family = daddr->sin_family; 665 csk->saddr.sin_addr.s_addr = fl4.saddr; 666 neigh_release(n); 667 668 return csk; 669 670 rel_neigh: 671 neigh_release(n); 672 673 rel_rt: 674 ip_rt_put(rt); 675 if (csk) 676 cxgbi_sock_closed(csk); 677 err_out: 678 return ERR_PTR(err); 679 } 680 681 #if IS_ENABLED(CONFIG_IPV6) 682 static struct rt6_info *find_route_ipv6(const struct in6_addr *saddr, 683 const struct in6_addr *daddr) 684 { 685 struct flowi6 fl; 686 687 if (saddr) 688 memcpy(&fl.saddr, saddr, sizeof(struct in6_addr)); 689 if (daddr) 690 memcpy(&fl.daddr, daddr, sizeof(struct in6_addr)); 691 return (struct rt6_info *)ip6_route_output(&init_net, NULL, &fl); 692 } 693 694 static struct cxgbi_sock *cxgbi_check_route6(struct sockaddr *dst_addr) 695 { 696 struct sockaddr_in6 *daddr6 = (struct sockaddr_in6 *)dst_addr; 697 struct dst_entry *dst; 698 struct net_device *ndev; 699 struct cxgbi_device *cdev; 700 struct rt6_info *rt = NULL; 701 struct neighbour *n; 702 struct in6_addr pref_saddr; 703 struct cxgbi_sock *csk = NULL; 704 unsigned int mtu = 0; 705 int port = 0xFFFF; 706 int err = 0; 707 708 rt = find_route_ipv6(NULL, &daddr6->sin6_addr); 709 710 if (!rt) { 711 pr_info("no route to ipv6 %pI6 port %u\n", 712 daddr6->sin6_addr.s6_addr, 713 be16_to_cpu(daddr6->sin6_port)); 714 err = -ENETUNREACH; 715 goto err_out; 716 } 717 718 dst = &rt->dst; 719 720 n = dst_neigh_lookup(dst, &daddr6->sin6_addr); 721 722 if (!n) { 723 pr_info("%pI6, port %u, dst no neighbour.\n", 724 daddr6->sin6_addr.s6_addr, 725 be16_to_cpu(daddr6->sin6_port)); 726 err = -ENETUNREACH; 727 goto rel_rt; 728 } 729 ndev = n->dev; 730 731 if (ipv6_addr_is_multicast(&rt->rt6i_dst.addr)) { 732 pr_info("multi-cast route %pI6 port %u, dev %s.\n", 733 daddr6->sin6_addr.s6_addr, 734 ntohs(daddr6->sin6_port), ndev->name); 735 err = -ENETUNREACH; 736 goto rel_rt; 737 } 738 739 cdev = cxgbi_device_find_by_netdev(ndev, &port); 740 if (!cdev) 741 cdev = cxgbi_device_find_by_mac(ndev, &port); 742 if (!cdev) { 743 pr_info("dst %pI6 %s, NOT cxgbi device.\n", 744 daddr6->sin6_addr.s6_addr, ndev->name); 745 err = -ENETUNREACH; 746 goto rel_rt; 747 } 748 log_debug(1 << CXGBI_DBG_SOCK, 749 "route to %pI6 :%u, ndev p#%d,%s, cdev 0x%p.\n", 750 daddr6->sin6_addr.s6_addr, ntohs(daddr6->sin6_port), port, 751 ndev->name, cdev); 752 753 csk = cxgbi_sock_create(cdev); 754 if (!csk) { 755 err = -ENOMEM; 756 goto rel_rt; 757 } 758 csk->cdev = cdev; 759 csk->port_id = port; 760 csk->mtu = mtu; 761 csk->dst = dst; 762 763 if (ipv6_addr_any(&rt->rt6i_prefsrc.addr)) { 764 struct inet6_dev *idev = ip6_dst_idev((struct dst_entry *)rt); 765 766 err = ipv6_dev_get_saddr(&init_net, idev ? idev->dev : NULL, 767 &daddr6->sin6_addr, 0, &pref_saddr); 768 if (err) { 769 pr_info("failed to get source address to reach %pI6\n", 770 &daddr6->sin6_addr); 771 goto rel_rt; 772 } 773 } else { 774 pref_saddr = rt->rt6i_prefsrc.addr; 775 } 776 777 csk->csk_family = AF_INET6; 778 csk->daddr6.sin6_addr = daddr6->sin6_addr; 779 csk->daddr6.sin6_port = daddr6->sin6_port; 780 csk->daddr6.sin6_family = daddr6->sin6_family; 781 csk->saddr6.sin6_family = daddr6->sin6_family; 782 csk->saddr6.sin6_addr = pref_saddr; 783 784 neigh_release(n); 785 return csk; 786 787 rel_rt: 788 if (n) 789 neigh_release(n); 790 791 ip6_rt_put(rt); 792 if (csk) 793 cxgbi_sock_closed(csk); 794 err_out: 795 return ERR_PTR(err); 796 } 797 #endif /* IS_ENABLED(CONFIG_IPV6) */ 798 799 void cxgbi_sock_established(struct cxgbi_sock *csk, unsigned int snd_isn, 800 unsigned int opt) 801 { 802 csk->write_seq = csk->snd_nxt = csk->snd_una = snd_isn; 803 dst_confirm(csk->dst); 804 smp_mb(); 805 cxgbi_sock_set_state(csk, CTP_ESTABLISHED); 806 } 807 EXPORT_SYMBOL_GPL(cxgbi_sock_established); 808 809 static void cxgbi_inform_iscsi_conn_closing(struct cxgbi_sock *csk) 810 { 811 log_debug(1 << CXGBI_DBG_SOCK, 812 "csk 0x%p, state %u, flags 0x%lx, conn 0x%p.\n", 813 csk, csk->state, csk->flags, csk->user_data); 814 815 if (csk->state != CTP_ESTABLISHED) { 816 read_lock_bh(&csk->callback_lock); 817 if (csk->user_data) 818 iscsi_conn_failure(csk->user_data, 819 ISCSI_ERR_TCP_CONN_CLOSE); 820 read_unlock_bh(&csk->callback_lock); 821 } 822 } 823 824 void cxgbi_sock_closed(struct cxgbi_sock *csk) 825 { 826 log_debug(1 << CXGBI_DBG_SOCK, "csk 0x%p,%u,0x%lx,%u.\n", 827 csk, (csk)->state, (csk)->flags, (csk)->tid); 828 cxgbi_sock_set_flag(csk, CTPF_ACTIVE_CLOSE_NEEDED); 829 if (csk->state == CTP_ACTIVE_OPEN || csk->state == CTP_CLOSED) 830 return; 831 if (csk->saddr.sin_port) 832 sock_put_port(csk); 833 if (csk->dst) 834 dst_release(csk->dst); 835 csk->cdev->csk_release_offload_resources(csk); 836 cxgbi_sock_set_state(csk, CTP_CLOSED); 837 cxgbi_inform_iscsi_conn_closing(csk); 838 cxgbi_sock_put(csk); 839 } 840 EXPORT_SYMBOL_GPL(cxgbi_sock_closed); 841 842 static void need_active_close(struct cxgbi_sock *csk) 843 { 844 int data_lost; 845 int close_req = 0; 846 847 log_debug(1 << CXGBI_DBG_SOCK, "csk 0x%p,%u,0x%lx,%u.\n", 848 csk, (csk)->state, (csk)->flags, (csk)->tid); 849 spin_lock_bh(&csk->lock); 850 dst_confirm(csk->dst); 851 data_lost = skb_queue_len(&csk->receive_queue); 852 __skb_queue_purge(&csk->receive_queue); 853 854 if (csk->state == CTP_ACTIVE_OPEN) 855 cxgbi_sock_set_flag(csk, CTPF_ACTIVE_CLOSE_NEEDED); 856 else if (csk->state == CTP_ESTABLISHED) { 857 close_req = 1; 858 cxgbi_sock_set_state(csk, CTP_ACTIVE_CLOSE); 859 } else if (csk->state == CTP_PASSIVE_CLOSE) { 860 close_req = 1; 861 cxgbi_sock_set_state(csk, CTP_CLOSE_WAIT_2); 862 } 863 864 if (close_req) { 865 if (data_lost) 866 csk->cdev->csk_send_abort_req(csk); 867 else 868 csk->cdev->csk_send_close_req(csk); 869 } 870 871 spin_unlock_bh(&csk->lock); 872 } 873 874 void cxgbi_sock_fail_act_open(struct cxgbi_sock *csk, int errno) 875 { 876 pr_info("csk 0x%p,%u,%lx, %pI4:%u-%pI4:%u, err %d.\n", 877 csk, csk->state, csk->flags, 878 &csk->saddr.sin_addr.s_addr, csk->saddr.sin_port, 879 &csk->daddr.sin_addr.s_addr, csk->daddr.sin_port, 880 errno); 881 882 cxgbi_sock_set_state(csk, CTP_CONNECTING); 883 csk->err = errno; 884 cxgbi_sock_closed(csk); 885 } 886 EXPORT_SYMBOL_GPL(cxgbi_sock_fail_act_open); 887 888 void cxgbi_sock_act_open_req_arp_failure(void *handle, struct sk_buff *skb) 889 { 890 struct cxgbi_sock *csk = (struct cxgbi_sock *)skb->sk; 891 892 log_debug(1 << CXGBI_DBG_SOCK, "csk 0x%p,%u,0x%lx,%u.\n", 893 csk, (csk)->state, (csk)->flags, (csk)->tid); 894 cxgbi_sock_get(csk); 895 spin_lock_bh(&csk->lock); 896 if (csk->state == CTP_ACTIVE_OPEN) 897 cxgbi_sock_fail_act_open(csk, -EHOSTUNREACH); 898 spin_unlock_bh(&csk->lock); 899 cxgbi_sock_put(csk); 900 __kfree_skb(skb); 901 } 902 EXPORT_SYMBOL_GPL(cxgbi_sock_act_open_req_arp_failure); 903 904 void cxgbi_sock_rcv_abort_rpl(struct cxgbi_sock *csk) 905 { 906 cxgbi_sock_get(csk); 907 spin_lock_bh(&csk->lock); 908 909 cxgbi_sock_set_flag(csk, CTPF_ABORT_RPL_RCVD); 910 if (cxgbi_sock_flag(csk, CTPF_ABORT_RPL_PENDING)) { 911 cxgbi_sock_clear_flag(csk, CTPF_ABORT_RPL_PENDING); 912 if (cxgbi_sock_flag(csk, CTPF_ABORT_REQ_RCVD)) 913 pr_err("csk 0x%p,%u,0x%lx,%u,ABT_RPL_RSS.\n", 914 csk, csk->state, csk->flags, csk->tid); 915 cxgbi_sock_closed(csk); 916 } 917 918 spin_unlock_bh(&csk->lock); 919 cxgbi_sock_put(csk); 920 } 921 EXPORT_SYMBOL_GPL(cxgbi_sock_rcv_abort_rpl); 922 923 void cxgbi_sock_rcv_peer_close(struct cxgbi_sock *csk) 924 { 925 log_debug(1 << CXGBI_DBG_SOCK, "csk 0x%p,%u,0x%lx,%u.\n", 926 csk, (csk)->state, (csk)->flags, (csk)->tid); 927 cxgbi_sock_get(csk); 928 spin_lock_bh(&csk->lock); 929 930 if (cxgbi_sock_flag(csk, CTPF_ABORT_RPL_PENDING)) 931 goto done; 932 933 switch (csk->state) { 934 case CTP_ESTABLISHED: 935 cxgbi_sock_set_state(csk, CTP_PASSIVE_CLOSE); 936 break; 937 case CTP_ACTIVE_CLOSE: 938 cxgbi_sock_set_state(csk, CTP_CLOSE_WAIT_2); 939 break; 940 case CTP_CLOSE_WAIT_1: 941 cxgbi_sock_closed(csk); 942 break; 943 case CTP_ABORTING: 944 break; 945 default: 946 pr_err("csk 0x%p,%u,0x%lx,%u, bad state.\n", 947 csk, csk->state, csk->flags, csk->tid); 948 } 949 cxgbi_inform_iscsi_conn_closing(csk); 950 done: 951 spin_unlock_bh(&csk->lock); 952 cxgbi_sock_put(csk); 953 } 954 EXPORT_SYMBOL_GPL(cxgbi_sock_rcv_peer_close); 955 956 void cxgbi_sock_rcv_close_conn_rpl(struct cxgbi_sock *csk, u32 snd_nxt) 957 { 958 log_debug(1 << CXGBI_DBG_SOCK, "csk 0x%p,%u,0x%lx,%u.\n", 959 csk, (csk)->state, (csk)->flags, (csk)->tid); 960 cxgbi_sock_get(csk); 961 spin_lock_bh(&csk->lock); 962 963 csk->snd_una = snd_nxt - 1; 964 if (cxgbi_sock_flag(csk, CTPF_ABORT_RPL_PENDING)) 965 goto done; 966 967 switch (csk->state) { 968 case CTP_ACTIVE_CLOSE: 969 cxgbi_sock_set_state(csk, CTP_CLOSE_WAIT_1); 970 break; 971 case CTP_CLOSE_WAIT_1: 972 case CTP_CLOSE_WAIT_2: 973 cxgbi_sock_closed(csk); 974 break; 975 case CTP_ABORTING: 976 break; 977 default: 978 pr_err("csk 0x%p,%u,0x%lx,%u, bad state.\n", 979 csk, csk->state, csk->flags, csk->tid); 980 } 981 done: 982 spin_unlock_bh(&csk->lock); 983 cxgbi_sock_put(csk); 984 } 985 EXPORT_SYMBOL_GPL(cxgbi_sock_rcv_close_conn_rpl); 986 987 void cxgbi_sock_rcv_wr_ack(struct cxgbi_sock *csk, unsigned int credits, 988 unsigned int snd_una, int seq_chk) 989 { 990 log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK, 991 "csk 0x%p,%u,0x%lx,%u, cr %u,%u+%u, snd_una %u,%d.\n", 992 csk, csk->state, csk->flags, csk->tid, credits, 993 csk->wr_cred, csk->wr_una_cred, snd_una, seq_chk); 994 995 spin_lock_bh(&csk->lock); 996 997 csk->wr_cred += credits; 998 if (csk->wr_una_cred > csk->wr_max_cred - csk->wr_cred) 999 csk->wr_una_cred = csk->wr_max_cred - csk->wr_cred; 1000 1001 while (credits) { 1002 struct sk_buff *p = cxgbi_sock_peek_wr(csk); 1003 1004 if (unlikely(!p)) { 1005 pr_err("csk 0x%p,%u,0x%lx,%u, cr %u,%u+%u, empty.\n", 1006 csk, csk->state, csk->flags, csk->tid, credits, 1007 csk->wr_cred, csk->wr_una_cred); 1008 break; 1009 } 1010 1011 if (unlikely(credits < p->csum)) { 1012 pr_warn("csk 0x%p,%u,0x%lx,%u, cr %u,%u+%u, < %u.\n", 1013 csk, csk->state, csk->flags, csk->tid, 1014 credits, csk->wr_cred, csk->wr_una_cred, 1015 p->csum); 1016 p->csum -= credits; 1017 break; 1018 } else { 1019 cxgbi_sock_dequeue_wr(csk); 1020 credits -= p->csum; 1021 kfree_skb(p); 1022 } 1023 } 1024 1025 cxgbi_sock_check_wr_invariants(csk); 1026 1027 if (seq_chk) { 1028 if (unlikely(before(snd_una, csk->snd_una))) { 1029 pr_warn("csk 0x%p,%u,0x%lx,%u, snd_una %u/%u.", 1030 csk, csk->state, csk->flags, csk->tid, snd_una, 1031 csk->snd_una); 1032 goto done; 1033 } 1034 1035 if (csk->snd_una != snd_una) { 1036 csk->snd_una = snd_una; 1037 dst_confirm(csk->dst); 1038 } 1039 } 1040 1041 if (skb_queue_len(&csk->write_queue)) { 1042 if (csk->cdev->csk_push_tx_frames(csk, 0)) 1043 cxgbi_conn_tx_open(csk); 1044 } else 1045 cxgbi_conn_tx_open(csk); 1046 done: 1047 spin_unlock_bh(&csk->lock); 1048 } 1049 EXPORT_SYMBOL_GPL(cxgbi_sock_rcv_wr_ack); 1050 1051 static unsigned int cxgbi_sock_find_best_mtu(struct cxgbi_sock *csk, 1052 unsigned short mtu) 1053 { 1054 int i = 0; 1055 1056 while (i < csk->cdev->nmtus - 1 && csk->cdev->mtus[i + 1] <= mtu) 1057 ++i; 1058 1059 return i; 1060 } 1061 1062 unsigned int cxgbi_sock_select_mss(struct cxgbi_sock *csk, unsigned int pmtu) 1063 { 1064 unsigned int idx; 1065 struct dst_entry *dst = csk->dst; 1066 1067 csk->advmss = dst_metric_advmss(dst); 1068 1069 if (csk->advmss > pmtu - 40) 1070 csk->advmss = pmtu - 40; 1071 if (csk->advmss < csk->cdev->mtus[0] - 40) 1072 csk->advmss = csk->cdev->mtus[0] - 40; 1073 idx = cxgbi_sock_find_best_mtu(csk, csk->advmss + 40); 1074 1075 return idx; 1076 } 1077 EXPORT_SYMBOL_GPL(cxgbi_sock_select_mss); 1078 1079 void cxgbi_sock_skb_entail(struct cxgbi_sock *csk, struct sk_buff *skb) 1080 { 1081 cxgbi_skcb_tcp_seq(skb) = csk->write_seq; 1082 __skb_queue_tail(&csk->write_queue, skb); 1083 } 1084 EXPORT_SYMBOL_GPL(cxgbi_sock_skb_entail); 1085 1086 void cxgbi_sock_purge_wr_queue(struct cxgbi_sock *csk) 1087 { 1088 struct sk_buff *skb; 1089 1090 while ((skb = cxgbi_sock_dequeue_wr(csk)) != NULL) 1091 kfree_skb(skb); 1092 } 1093 EXPORT_SYMBOL_GPL(cxgbi_sock_purge_wr_queue); 1094 1095 void cxgbi_sock_check_wr_invariants(const struct cxgbi_sock *csk) 1096 { 1097 int pending = cxgbi_sock_count_pending_wrs(csk); 1098 1099 if (unlikely(csk->wr_cred + pending != csk->wr_max_cred)) 1100 pr_err("csk 0x%p, tid %u, credit %u + %u != %u.\n", 1101 csk, csk->tid, csk->wr_cred, pending, csk->wr_max_cred); 1102 } 1103 EXPORT_SYMBOL_GPL(cxgbi_sock_check_wr_invariants); 1104 1105 static int cxgbi_sock_send_pdus(struct cxgbi_sock *csk, struct sk_buff *skb) 1106 { 1107 struct cxgbi_device *cdev = csk->cdev; 1108 struct sk_buff *next; 1109 int err, copied = 0; 1110 1111 spin_lock_bh(&csk->lock); 1112 1113 if (csk->state != CTP_ESTABLISHED) { 1114 log_debug(1 << CXGBI_DBG_PDU_TX, 1115 "csk 0x%p,%u,0x%lx,%u, EAGAIN.\n", 1116 csk, csk->state, csk->flags, csk->tid); 1117 err = -EAGAIN; 1118 goto out_err; 1119 } 1120 1121 if (csk->err) { 1122 log_debug(1 << CXGBI_DBG_PDU_TX, 1123 "csk 0x%p,%u,0x%lx,%u, EPIPE %d.\n", 1124 csk, csk->state, csk->flags, csk->tid, csk->err); 1125 err = -EPIPE; 1126 goto out_err; 1127 } 1128 1129 if (csk->write_seq - csk->snd_una >= cdev->snd_win) { 1130 log_debug(1 << CXGBI_DBG_PDU_TX, 1131 "csk 0x%p,%u,0x%lx,%u, FULL %u-%u >= %u.\n", 1132 csk, csk->state, csk->flags, csk->tid, csk->write_seq, 1133 csk->snd_una, cdev->snd_win); 1134 err = -ENOBUFS; 1135 goto out_err; 1136 } 1137 1138 while (skb) { 1139 int frags = skb_shinfo(skb)->nr_frags + 1140 (skb->len != skb->data_len); 1141 1142 if (unlikely(skb_headroom(skb) < cdev->skb_tx_rsvd)) { 1143 pr_err("csk 0x%p, skb head %u < %u.\n", 1144 csk, skb_headroom(skb), cdev->skb_tx_rsvd); 1145 err = -EINVAL; 1146 goto out_err; 1147 } 1148 1149 if (frags >= SKB_WR_LIST_SIZE) { 1150 pr_err("csk 0x%p, frags %d, %u,%u >%u.\n", 1151 csk, skb_shinfo(skb)->nr_frags, skb->len, 1152 skb->data_len, (uint)(SKB_WR_LIST_SIZE)); 1153 err = -EINVAL; 1154 goto out_err; 1155 } 1156 1157 next = skb->next; 1158 skb->next = NULL; 1159 cxgbi_skcb_set_flag(skb, SKCBF_TX_NEED_HDR); 1160 cxgbi_sock_skb_entail(csk, skb); 1161 copied += skb->len; 1162 csk->write_seq += skb->len + 1163 cxgbi_ulp_extra_len(cxgbi_skcb_ulp_mode(skb)); 1164 skb = next; 1165 } 1166 done: 1167 if (likely(skb_queue_len(&csk->write_queue))) 1168 cdev->csk_push_tx_frames(csk, 1); 1169 spin_unlock_bh(&csk->lock); 1170 return copied; 1171 1172 out_err: 1173 if (copied == 0 && err == -EPIPE) 1174 copied = csk->err ? csk->err : -EPIPE; 1175 else 1176 copied = err; 1177 goto done; 1178 } 1179 1180 /* 1181 * Direct Data Placement - 1182 * Directly place the iSCSI Data-In or Data-Out PDU's payload into pre-posted 1183 * final destination host-memory buffers based on the Initiator Task Tag (ITT) 1184 * in Data-In or Target Task Tag (TTT) in Data-Out PDUs. 1185 * The host memory address is programmed into h/w in the format of pagepod 1186 * entries. 1187 * The location of the pagepod entry is encoded into ddp tag which is used as 1188 * the base for ITT/TTT. 1189 */ 1190 1191 static unsigned char ddp_page_order[DDP_PGIDX_MAX] = {0, 1, 2, 4}; 1192 static unsigned char ddp_page_shift[DDP_PGIDX_MAX] = {12, 13, 14, 16}; 1193 static unsigned char page_idx = DDP_PGIDX_MAX; 1194 1195 static unsigned char sw_tag_idx_bits; 1196 static unsigned char sw_tag_age_bits; 1197 1198 /* 1199 * Direct-Data Placement page size adjustment 1200 */ 1201 static int ddp_adjust_page_table(void) 1202 { 1203 int i; 1204 unsigned int base_order, order; 1205 1206 if (PAGE_SIZE < (1UL << ddp_page_shift[0])) { 1207 pr_info("PAGE_SIZE 0x%lx too small, min 0x%lx\n", 1208 PAGE_SIZE, 1UL << ddp_page_shift[0]); 1209 return -EINVAL; 1210 } 1211 1212 base_order = get_order(1UL << ddp_page_shift[0]); 1213 order = get_order(1UL << PAGE_SHIFT); 1214 1215 for (i = 0; i < DDP_PGIDX_MAX; i++) { 1216 /* first is the kernel page size, then just doubling */ 1217 ddp_page_order[i] = order - base_order + i; 1218 ddp_page_shift[i] = PAGE_SHIFT + i; 1219 } 1220 return 0; 1221 } 1222 1223 static int ddp_find_page_index(unsigned long pgsz) 1224 { 1225 int i; 1226 1227 for (i = 0; i < DDP_PGIDX_MAX; i++) { 1228 if (pgsz == (1UL << ddp_page_shift[i])) 1229 return i; 1230 } 1231 pr_info("ddp page size %lu not supported.\n", pgsz); 1232 return DDP_PGIDX_MAX; 1233 } 1234 1235 static void ddp_setup_host_page_size(void) 1236 { 1237 if (page_idx == DDP_PGIDX_MAX) { 1238 page_idx = ddp_find_page_index(PAGE_SIZE); 1239 1240 if (page_idx == DDP_PGIDX_MAX) { 1241 pr_info("system PAGE %lu, update hw.\n", PAGE_SIZE); 1242 if (ddp_adjust_page_table() < 0) { 1243 pr_info("PAGE %lu, disable ddp.\n", PAGE_SIZE); 1244 return; 1245 } 1246 page_idx = ddp_find_page_index(PAGE_SIZE); 1247 } 1248 pr_info("system PAGE %lu, ddp idx %u.\n", PAGE_SIZE, page_idx); 1249 } 1250 } 1251 1252 void cxgbi_ddp_page_size_factor(int *pgsz_factor) 1253 { 1254 int i; 1255 1256 for (i = 0; i < DDP_PGIDX_MAX; i++) 1257 pgsz_factor[i] = ddp_page_order[i]; 1258 } 1259 EXPORT_SYMBOL_GPL(cxgbi_ddp_page_size_factor); 1260 1261 /* 1262 * DDP setup & teardown 1263 */ 1264 1265 void cxgbi_ddp_ppod_set(struct cxgbi_pagepod *ppod, 1266 struct cxgbi_pagepod_hdr *hdr, 1267 struct cxgbi_gather_list *gl, unsigned int gidx) 1268 { 1269 int i; 1270 1271 memcpy(ppod, hdr, sizeof(*hdr)); 1272 for (i = 0; i < (PPOD_PAGES_MAX + 1); i++, gidx++) { 1273 ppod->addr[i] = gidx < gl->nelem ? 1274 cpu_to_be64(gl->phys_addr[gidx]) : 0ULL; 1275 } 1276 } 1277 EXPORT_SYMBOL_GPL(cxgbi_ddp_ppod_set); 1278 1279 void cxgbi_ddp_ppod_clear(struct cxgbi_pagepod *ppod) 1280 { 1281 memset(ppod, 0, sizeof(*ppod)); 1282 } 1283 EXPORT_SYMBOL_GPL(cxgbi_ddp_ppod_clear); 1284 1285 static inline int ddp_find_unused_entries(struct cxgbi_ddp_info *ddp, 1286 unsigned int start, unsigned int max, 1287 unsigned int count, 1288 struct cxgbi_gather_list *gl) 1289 { 1290 unsigned int i, j, k; 1291 1292 /* not enough entries */ 1293 if ((max - start) < count) { 1294 log_debug(1 << CXGBI_DBG_DDP, 1295 "NOT enough entries %u+%u < %u.\n", start, count, max); 1296 return -EBUSY; 1297 } 1298 1299 max -= count; 1300 spin_lock(&ddp->map_lock); 1301 for (i = start; i < max;) { 1302 for (j = 0, k = i; j < count; j++, k++) { 1303 if (ddp->gl_map[k]) 1304 break; 1305 } 1306 if (j == count) { 1307 for (j = 0, k = i; j < count; j++, k++) 1308 ddp->gl_map[k] = gl; 1309 spin_unlock(&ddp->map_lock); 1310 return i; 1311 } 1312 i += j + 1; 1313 } 1314 spin_unlock(&ddp->map_lock); 1315 log_debug(1 << CXGBI_DBG_DDP, 1316 "NO suitable entries %u available.\n", count); 1317 return -EBUSY; 1318 } 1319 1320 static inline void ddp_unmark_entries(struct cxgbi_ddp_info *ddp, 1321 int start, int count) 1322 { 1323 spin_lock(&ddp->map_lock); 1324 memset(&ddp->gl_map[start], 0, 1325 count * sizeof(struct cxgbi_gather_list *)); 1326 spin_unlock(&ddp->map_lock); 1327 } 1328 1329 static inline void ddp_gl_unmap(struct pci_dev *pdev, 1330 struct cxgbi_gather_list *gl) 1331 { 1332 int i; 1333 1334 for (i = 0; i < gl->nelem; i++) 1335 dma_unmap_page(&pdev->dev, gl->phys_addr[i], PAGE_SIZE, 1336 PCI_DMA_FROMDEVICE); 1337 } 1338 1339 static inline int ddp_gl_map(struct pci_dev *pdev, 1340 struct cxgbi_gather_list *gl) 1341 { 1342 int i; 1343 1344 for (i = 0; i < gl->nelem; i++) { 1345 gl->phys_addr[i] = dma_map_page(&pdev->dev, gl->pages[i], 0, 1346 PAGE_SIZE, 1347 PCI_DMA_FROMDEVICE); 1348 if (unlikely(dma_mapping_error(&pdev->dev, gl->phys_addr[i]))) { 1349 log_debug(1 << CXGBI_DBG_DDP, 1350 "page %d 0x%p, 0x%p dma mapping err.\n", 1351 i, gl->pages[i], pdev); 1352 goto unmap; 1353 } 1354 } 1355 return i; 1356 unmap: 1357 if (i) { 1358 unsigned int nelem = gl->nelem; 1359 1360 gl->nelem = i; 1361 ddp_gl_unmap(pdev, gl); 1362 gl->nelem = nelem; 1363 } 1364 return -EINVAL; 1365 } 1366 1367 static void ddp_release_gl(struct cxgbi_gather_list *gl, 1368 struct pci_dev *pdev) 1369 { 1370 ddp_gl_unmap(pdev, gl); 1371 kfree(gl); 1372 } 1373 1374 static struct cxgbi_gather_list *ddp_make_gl(unsigned int xferlen, 1375 struct scatterlist *sgl, 1376 unsigned int sgcnt, 1377 struct pci_dev *pdev, 1378 gfp_t gfp) 1379 { 1380 struct cxgbi_gather_list *gl; 1381 struct scatterlist *sg = sgl; 1382 struct page *sgpage = sg_page(sg); 1383 unsigned int sglen = sg->length; 1384 unsigned int sgoffset = sg->offset; 1385 unsigned int npages = (xferlen + sgoffset + PAGE_SIZE - 1) >> 1386 PAGE_SHIFT; 1387 int i = 1, j = 0; 1388 1389 if (xferlen < DDP_THRESHOLD) { 1390 log_debug(1 << CXGBI_DBG_DDP, 1391 "xfer %u < threshold %u, no ddp.\n", 1392 xferlen, DDP_THRESHOLD); 1393 return NULL; 1394 } 1395 1396 gl = kzalloc(sizeof(struct cxgbi_gather_list) + 1397 npages * (sizeof(dma_addr_t) + 1398 sizeof(struct page *)), gfp); 1399 if (!gl) { 1400 log_debug(1 << CXGBI_DBG_DDP, 1401 "xfer %u, %u pages, OOM.\n", xferlen, npages); 1402 return NULL; 1403 } 1404 1405 log_debug(1 << CXGBI_DBG_DDP, 1406 "xfer %u, sgl %u, gl max %u.\n", xferlen, sgcnt, npages); 1407 1408 gl->pages = (struct page **)&gl->phys_addr[npages]; 1409 gl->nelem = npages; 1410 gl->length = xferlen; 1411 gl->offset = sgoffset; 1412 gl->pages[0] = sgpage; 1413 1414 for (i = 1, sg = sg_next(sgl), j = 0; i < sgcnt; 1415 i++, sg = sg_next(sg)) { 1416 struct page *page = sg_page(sg); 1417 1418 if (sgpage == page && sg->offset == sgoffset + sglen) 1419 sglen += sg->length; 1420 else { 1421 /* make sure the sgl is fit for ddp: 1422 * each has the same page size, and 1423 * all of the middle pages are used completely 1424 */ 1425 if ((j && sgoffset) || ((i != sgcnt - 1) && 1426 ((sglen + sgoffset) & ~PAGE_MASK))) { 1427 log_debug(1 << CXGBI_DBG_DDP, 1428 "page %d/%u, %u + %u.\n", 1429 i, sgcnt, sgoffset, sglen); 1430 goto error_out; 1431 } 1432 1433 j++; 1434 if (j == gl->nelem || sg->offset) { 1435 log_debug(1 << CXGBI_DBG_DDP, 1436 "page %d/%u, offset %u.\n", 1437 j, gl->nelem, sg->offset); 1438 goto error_out; 1439 } 1440 gl->pages[j] = page; 1441 sglen = sg->length; 1442 sgoffset = sg->offset; 1443 sgpage = page; 1444 } 1445 } 1446 gl->nelem = ++j; 1447 1448 if (ddp_gl_map(pdev, gl) < 0) 1449 goto error_out; 1450 1451 return gl; 1452 1453 error_out: 1454 kfree(gl); 1455 return NULL; 1456 } 1457 1458 static void ddp_tag_release(struct cxgbi_hba *chba, u32 tag) 1459 { 1460 struct cxgbi_device *cdev = chba->cdev; 1461 struct cxgbi_ddp_info *ddp = cdev->ddp; 1462 u32 idx; 1463 1464 idx = (tag >> PPOD_IDX_SHIFT) & ddp->idx_mask; 1465 if (idx < ddp->nppods) { 1466 struct cxgbi_gather_list *gl = ddp->gl_map[idx]; 1467 unsigned int npods; 1468 1469 if (!gl || !gl->nelem) { 1470 pr_warn("tag 0x%x, idx %u, gl 0x%p, %u.\n", 1471 tag, idx, gl, gl ? gl->nelem : 0); 1472 return; 1473 } 1474 npods = (gl->nelem + PPOD_PAGES_MAX - 1) >> PPOD_PAGES_SHIFT; 1475 log_debug(1 << CXGBI_DBG_DDP, 1476 "tag 0x%x, release idx %u, npods %u.\n", 1477 tag, idx, npods); 1478 cdev->csk_ddp_clear(chba, tag, idx, npods); 1479 ddp_unmark_entries(ddp, idx, npods); 1480 ddp_release_gl(gl, ddp->pdev); 1481 } else 1482 pr_warn("tag 0x%x, idx %u > max %u.\n", tag, idx, ddp->nppods); 1483 } 1484 1485 static int ddp_tag_reserve(struct cxgbi_sock *csk, unsigned int tid, 1486 u32 sw_tag, u32 *tagp, struct cxgbi_gather_list *gl, 1487 gfp_t gfp) 1488 { 1489 struct cxgbi_device *cdev = csk->cdev; 1490 struct cxgbi_ddp_info *ddp = cdev->ddp; 1491 struct cxgbi_tag_format *tformat = &cdev->tag_format; 1492 struct cxgbi_pagepod_hdr hdr; 1493 unsigned int npods; 1494 int idx = -1; 1495 int err = -ENOMEM; 1496 u32 tag; 1497 1498 npods = (gl->nelem + PPOD_PAGES_MAX - 1) >> PPOD_PAGES_SHIFT; 1499 if (ddp->idx_last == ddp->nppods) 1500 idx = ddp_find_unused_entries(ddp, 0, ddp->nppods, 1501 npods, gl); 1502 else { 1503 idx = ddp_find_unused_entries(ddp, ddp->idx_last + 1, 1504 ddp->nppods, npods, 1505 gl); 1506 if (idx < 0 && ddp->idx_last >= npods) { 1507 idx = ddp_find_unused_entries(ddp, 0, 1508 min(ddp->idx_last + npods, ddp->nppods), 1509 npods, gl); 1510 } 1511 } 1512 if (idx < 0) { 1513 log_debug(1 << CXGBI_DBG_DDP, 1514 "xferlen %u, gl %u, npods %u NO DDP.\n", 1515 gl->length, gl->nelem, npods); 1516 return idx; 1517 } 1518 1519 tag = cxgbi_ddp_tag_base(tformat, sw_tag); 1520 tag |= idx << PPOD_IDX_SHIFT; 1521 1522 hdr.rsvd = 0; 1523 hdr.vld_tid = htonl(PPOD_VALID_FLAG | PPOD_TID(tid)); 1524 hdr.pgsz_tag_clr = htonl(tag & ddp->rsvd_tag_mask); 1525 hdr.max_offset = htonl(gl->length); 1526 hdr.page_offset = htonl(gl->offset); 1527 1528 err = cdev->csk_ddp_set(csk, &hdr, idx, npods, gl); 1529 if (err < 0) 1530 goto unmark_entries; 1531 1532 ddp->idx_last = idx; 1533 log_debug(1 << CXGBI_DBG_DDP, 1534 "xfer %u, gl %u,%u, tid 0x%x, tag 0x%x->0x%x(%u,%u).\n", 1535 gl->length, gl->nelem, gl->offset, tid, sw_tag, tag, idx, 1536 npods); 1537 *tagp = tag; 1538 return 0; 1539 1540 unmark_entries: 1541 ddp_unmark_entries(ddp, idx, npods); 1542 return err; 1543 } 1544 1545 int cxgbi_ddp_reserve(struct cxgbi_sock *csk, unsigned int *tagp, 1546 unsigned int sw_tag, unsigned int xferlen, 1547 struct scatterlist *sgl, unsigned int sgcnt, gfp_t gfp) 1548 { 1549 struct cxgbi_device *cdev = csk->cdev; 1550 struct cxgbi_tag_format *tformat = &cdev->tag_format; 1551 struct cxgbi_gather_list *gl; 1552 int err; 1553 1554 if (page_idx >= DDP_PGIDX_MAX || !cdev->ddp || 1555 xferlen < DDP_THRESHOLD) { 1556 log_debug(1 << CXGBI_DBG_DDP, 1557 "pgidx %u, xfer %u, NO ddp.\n", page_idx, xferlen); 1558 return -EINVAL; 1559 } 1560 1561 if (!cxgbi_sw_tag_usable(tformat, sw_tag)) { 1562 log_debug(1 << CXGBI_DBG_DDP, 1563 "sw_tag 0x%x NOT usable.\n", sw_tag); 1564 return -EINVAL; 1565 } 1566 1567 gl = ddp_make_gl(xferlen, sgl, sgcnt, cdev->pdev, gfp); 1568 if (!gl) 1569 return -ENOMEM; 1570 1571 err = ddp_tag_reserve(csk, csk->tid, sw_tag, tagp, gl, gfp); 1572 if (err < 0) 1573 ddp_release_gl(gl, cdev->pdev); 1574 1575 return err; 1576 } 1577 1578 static void ddp_destroy(struct kref *kref) 1579 { 1580 struct cxgbi_ddp_info *ddp = container_of(kref, 1581 struct cxgbi_ddp_info, 1582 refcnt); 1583 struct cxgbi_device *cdev = ddp->cdev; 1584 int i = 0; 1585 1586 pr_info("kref 0, destroy ddp 0x%p, cdev 0x%p.\n", ddp, cdev); 1587 1588 while (i < ddp->nppods) { 1589 struct cxgbi_gather_list *gl = ddp->gl_map[i]; 1590 1591 if (gl) { 1592 int npods = (gl->nelem + PPOD_PAGES_MAX - 1) 1593 >> PPOD_PAGES_SHIFT; 1594 pr_info("cdev 0x%p, ddp %d + %d.\n", cdev, i, npods); 1595 kfree(gl); 1596 i += npods; 1597 } else 1598 i++; 1599 } 1600 cxgbi_free_big_mem(ddp); 1601 } 1602 1603 int cxgbi_ddp_cleanup(struct cxgbi_device *cdev) 1604 { 1605 struct cxgbi_ddp_info *ddp = cdev->ddp; 1606 1607 log_debug(1 << CXGBI_DBG_DDP, 1608 "cdev 0x%p, release ddp 0x%p.\n", cdev, ddp); 1609 cdev->ddp = NULL; 1610 if (ddp) 1611 return kref_put(&ddp->refcnt, ddp_destroy); 1612 return 0; 1613 } 1614 EXPORT_SYMBOL_GPL(cxgbi_ddp_cleanup); 1615 1616 int cxgbi_ddp_init(struct cxgbi_device *cdev, 1617 unsigned int llimit, unsigned int ulimit, 1618 unsigned int max_txsz, unsigned int max_rxsz) 1619 { 1620 struct cxgbi_ddp_info *ddp; 1621 unsigned int ppmax, bits; 1622 1623 ppmax = (ulimit - llimit + 1) >> PPOD_SIZE_SHIFT; 1624 bits = __ilog2_u32(ppmax) + 1; 1625 if (bits > PPOD_IDX_MAX_SIZE) 1626 bits = PPOD_IDX_MAX_SIZE; 1627 ppmax = (1 << (bits - 1)) - 1; 1628 1629 ddp = cxgbi_alloc_big_mem(sizeof(struct cxgbi_ddp_info) + 1630 ppmax * (sizeof(struct cxgbi_gather_list *) + 1631 sizeof(struct sk_buff *)), 1632 GFP_KERNEL); 1633 if (!ddp) { 1634 pr_warn("cdev 0x%p, ddp ppmax %u OOM.\n", cdev, ppmax); 1635 return -ENOMEM; 1636 } 1637 ddp->gl_map = (struct cxgbi_gather_list **)(ddp + 1); 1638 cdev->ddp = ddp; 1639 1640 spin_lock_init(&ddp->map_lock); 1641 kref_init(&ddp->refcnt); 1642 1643 ddp->cdev = cdev; 1644 ddp->pdev = cdev->pdev; 1645 ddp->llimit = llimit; 1646 ddp->ulimit = ulimit; 1647 ddp->max_txsz = min_t(unsigned int, max_txsz, ULP2_MAX_PKT_SIZE); 1648 ddp->max_rxsz = min_t(unsigned int, max_rxsz, ULP2_MAX_PKT_SIZE); 1649 ddp->nppods = ppmax; 1650 ddp->idx_last = ppmax; 1651 ddp->idx_bits = bits; 1652 ddp->idx_mask = (1 << bits) - 1; 1653 ddp->rsvd_tag_mask = (1 << (bits + PPOD_IDX_SHIFT)) - 1; 1654 1655 cdev->tag_format.sw_bits = sw_tag_idx_bits + sw_tag_age_bits; 1656 cdev->tag_format.rsvd_bits = ddp->idx_bits; 1657 cdev->tag_format.rsvd_shift = PPOD_IDX_SHIFT; 1658 cdev->tag_format.rsvd_mask = (1 << cdev->tag_format.rsvd_bits) - 1; 1659 1660 pr_info("%s tag format, sw %u, rsvd %u,%u, mask 0x%x.\n", 1661 cdev->ports[0]->name, cdev->tag_format.sw_bits, 1662 cdev->tag_format.rsvd_bits, cdev->tag_format.rsvd_shift, 1663 cdev->tag_format.rsvd_mask); 1664 1665 cdev->tx_max_size = min_t(unsigned int, ULP2_MAX_PDU_PAYLOAD, 1666 ddp->max_txsz - ISCSI_PDU_NONPAYLOAD_LEN); 1667 cdev->rx_max_size = min_t(unsigned int, ULP2_MAX_PDU_PAYLOAD, 1668 ddp->max_rxsz - ISCSI_PDU_NONPAYLOAD_LEN); 1669 1670 log_debug(1 << CXGBI_DBG_DDP, 1671 "%s max payload size: %u/%u, %u/%u.\n", 1672 cdev->ports[0]->name, cdev->tx_max_size, ddp->max_txsz, 1673 cdev->rx_max_size, ddp->max_rxsz); 1674 return 0; 1675 } 1676 EXPORT_SYMBOL_GPL(cxgbi_ddp_init); 1677 1678 /* 1679 * APIs interacting with open-iscsi libraries 1680 */ 1681 1682 static unsigned char padding[4]; 1683 1684 static void task_release_itt(struct iscsi_task *task, itt_t hdr_itt) 1685 { 1686 struct scsi_cmnd *sc = task->sc; 1687 struct iscsi_tcp_conn *tcp_conn = task->conn->dd_data; 1688 struct cxgbi_conn *cconn = tcp_conn->dd_data; 1689 struct cxgbi_hba *chba = cconn->chba; 1690 struct cxgbi_tag_format *tformat = &chba->cdev->tag_format; 1691 u32 tag = ntohl((__force u32)hdr_itt); 1692 1693 log_debug(1 << CXGBI_DBG_DDP, 1694 "cdev 0x%p, release tag 0x%x.\n", chba->cdev, tag); 1695 if (sc && 1696 (scsi_bidi_cmnd(sc) || sc->sc_data_direction == DMA_FROM_DEVICE) && 1697 cxgbi_is_ddp_tag(tformat, tag)) 1698 ddp_tag_release(chba, tag); 1699 } 1700 1701 static int task_reserve_itt(struct iscsi_task *task, itt_t *hdr_itt) 1702 { 1703 struct scsi_cmnd *sc = task->sc; 1704 struct iscsi_conn *conn = task->conn; 1705 struct iscsi_session *sess = conn->session; 1706 struct iscsi_tcp_conn *tcp_conn = conn->dd_data; 1707 struct cxgbi_conn *cconn = tcp_conn->dd_data; 1708 struct cxgbi_hba *chba = cconn->chba; 1709 struct cxgbi_tag_format *tformat = &chba->cdev->tag_format; 1710 u32 sw_tag = (sess->age << cconn->task_idx_bits) | task->itt; 1711 u32 tag = 0; 1712 int err = -EINVAL; 1713 1714 if (sc && 1715 (scsi_bidi_cmnd(sc) || sc->sc_data_direction == DMA_FROM_DEVICE)) { 1716 err = cxgbi_ddp_reserve(cconn->cep->csk, &tag, sw_tag, 1717 scsi_in(sc)->length, 1718 scsi_in(sc)->table.sgl, 1719 scsi_in(sc)->table.nents, 1720 GFP_ATOMIC); 1721 if (err < 0) 1722 log_debug(1 << CXGBI_DBG_DDP, 1723 "csk 0x%p, R task 0x%p, %u,%u, no ddp.\n", 1724 cconn->cep->csk, task, scsi_in(sc)->length, 1725 scsi_in(sc)->table.nents); 1726 } 1727 1728 if (err < 0) 1729 tag = cxgbi_set_non_ddp_tag(tformat, sw_tag); 1730 /* the itt need to sent in big-endian order */ 1731 *hdr_itt = (__force itt_t)htonl(tag); 1732 1733 log_debug(1 << CXGBI_DBG_DDP, 1734 "cdev 0x%p, task 0x%p, 0x%x(0x%x,0x%x)->0x%x/0x%x.\n", 1735 chba->cdev, task, sw_tag, task->itt, sess->age, tag, *hdr_itt); 1736 return 0; 1737 } 1738 1739 void cxgbi_parse_pdu_itt(struct iscsi_conn *conn, itt_t itt, int *idx, int *age) 1740 { 1741 struct iscsi_tcp_conn *tcp_conn = conn->dd_data; 1742 struct cxgbi_conn *cconn = tcp_conn->dd_data; 1743 struct cxgbi_device *cdev = cconn->chba->cdev; 1744 u32 tag = ntohl((__force u32) itt); 1745 u32 sw_bits; 1746 1747 sw_bits = cxgbi_tag_nonrsvd_bits(&cdev->tag_format, tag); 1748 if (idx) 1749 *idx = sw_bits & ((1 << cconn->task_idx_bits) - 1); 1750 if (age) 1751 *age = (sw_bits >> cconn->task_idx_bits) & ISCSI_AGE_MASK; 1752 1753 log_debug(1 << CXGBI_DBG_DDP, 1754 "cdev 0x%p, tag 0x%x/0x%x, -> 0x%x(0x%x,0x%x).\n", 1755 cdev, tag, itt, sw_bits, idx ? *idx : 0xFFFFF, 1756 age ? *age : 0xFF); 1757 } 1758 EXPORT_SYMBOL_GPL(cxgbi_parse_pdu_itt); 1759 1760 void cxgbi_conn_tx_open(struct cxgbi_sock *csk) 1761 { 1762 struct iscsi_conn *conn = csk->user_data; 1763 1764 if (conn) { 1765 log_debug(1 << CXGBI_DBG_SOCK, 1766 "csk 0x%p, cid %d.\n", csk, conn->id); 1767 iscsi_conn_queue_work(conn); 1768 } 1769 } 1770 EXPORT_SYMBOL_GPL(cxgbi_conn_tx_open); 1771 1772 /* 1773 * pdu receive, interact with libiscsi_tcp 1774 */ 1775 static inline int read_pdu_skb(struct iscsi_conn *conn, 1776 struct sk_buff *skb, 1777 unsigned int offset, 1778 int offloaded) 1779 { 1780 int status = 0; 1781 int bytes_read; 1782 1783 bytes_read = iscsi_tcp_recv_skb(conn, skb, offset, offloaded, &status); 1784 switch (status) { 1785 case ISCSI_TCP_CONN_ERR: 1786 pr_info("skb 0x%p, off %u, %d, TCP_ERR.\n", 1787 skb, offset, offloaded); 1788 return -EIO; 1789 case ISCSI_TCP_SUSPENDED: 1790 log_debug(1 << CXGBI_DBG_PDU_RX, 1791 "skb 0x%p, off %u, %d, TCP_SUSPEND, rc %d.\n", 1792 skb, offset, offloaded, bytes_read); 1793 /* no transfer - just have caller flush queue */ 1794 return bytes_read; 1795 case ISCSI_TCP_SKB_DONE: 1796 pr_info("skb 0x%p, off %u, %d, TCP_SKB_DONE.\n", 1797 skb, offset, offloaded); 1798 /* 1799 * pdus should always fit in the skb and we should get 1800 * segment done notifcation. 1801 */ 1802 iscsi_conn_printk(KERN_ERR, conn, "Invalid pdu or skb."); 1803 return -EFAULT; 1804 case ISCSI_TCP_SEGMENT_DONE: 1805 log_debug(1 << CXGBI_DBG_PDU_RX, 1806 "skb 0x%p, off %u, %d, TCP_SEG_DONE, rc %d.\n", 1807 skb, offset, offloaded, bytes_read); 1808 return bytes_read; 1809 default: 1810 pr_info("skb 0x%p, off %u, %d, invalid status %d.\n", 1811 skb, offset, offloaded, status); 1812 return -EINVAL; 1813 } 1814 } 1815 1816 static int skb_read_pdu_bhs(struct iscsi_conn *conn, struct sk_buff *skb) 1817 { 1818 struct iscsi_tcp_conn *tcp_conn = conn->dd_data; 1819 1820 log_debug(1 << CXGBI_DBG_PDU_RX, 1821 "conn 0x%p, skb 0x%p, len %u, flag 0x%lx.\n", 1822 conn, skb, skb->len, cxgbi_skcb_flags(skb)); 1823 1824 if (!iscsi_tcp_recv_segment_is_hdr(tcp_conn)) { 1825 pr_info("conn 0x%p, skb 0x%p, not hdr.\n", conn, skb); 1826 iscsi_conn_failure(conn, ISCSI_ERR_PROTO); 1827 return -EIO; 1828 } 1829 1830 if (conn->hdrdgst_en && 1831 cxgbi_skcb_test_flag(skb, SKCBF_RX_HCRC_ERR)) { 1832 pr_info("conn 0x%p, skb 0x%p, hcrc.\n", conn, skb); 1833 iscsi_conn_failure(conn, ISCSI_ERR_HDR_DGST); 1834 return -EIO; 1835 } 1836 1837 return read_pdu_skb(conn, skb, 0, 0); 1838 } 1839 1840 static int skb_read_pdu_data(struct iscsi_conn *conn, struct sk_buff *lskb, 1841 struct sk_buff *skb, unsigned int offset) 1842 { 1843 struct iscsi_tcp_conn *tcp_conn = conn->dd_data; 1844 bool offloaded = 0; 1845 int opcode = tcp_conn->in.hdr->opcode & ISCSI_OPCODE_MASK; 1846 1847 log_debug(1 << CXGBI_DBG_PDU_RX, 1848 "conn 0x%p, skb 0x%p, len %u, flag 0x%lx.\n", 1849 conn, skb, skb->len, cxgbi_skcb_flags(skb)); 1850 1851 if (conn->datadgst_en && 1852 cxgbi_skcb_test_flag(lskb, SKCBF_RX_DCRC_ERR)) { 1853 pr_info("conn 0x%p, skb 0x%p, dcrc 0x%lx.\n", 1854 conn, lskb, cxgbi_skcb_flags(lskb)); 1855 iscsi_conn_failure(conn, ISCSI_ERR_DATA_DGST); 1856 return -EIO; 1857 } 1858 1859 if (iscsi_tcp_recv_segment_is_hdr(tcp_conn)) 1860 return 0; 1861 1862 /* coalesced, add header digest length */ 1863 if (lskb == skb && conn->hdrdgst_en) 1864 offset += ISCSI_DIGEST_SIZE; 1865 1866 if (cxgbi_skcb_test_flag(lskb, SKCBF_RX_DATA_DDPD)) 1867 offloaded = 1; 1868 1869 if (opcode == ISCSI_OP_SCSI_DATA_IN) 1870 log_debug(1 << CXGBI_DBG_PDU_RX, 1871 "skb 0x%p, op 0x%x, itt 0x%x, %u %s ddp'ed.\n", 1872 skb, opcode, ntohl(tcp_conn->in.hdr->itt), 1873 tcp_conn->in.datalen, offloaded ? "is" : "not"); 1874 1875 return read_pdu_skb(conn, skb, offset, offloaded); 1876 } 1877 1878 static void csk_return_rx_credits(struct cxgbi_sock *csk, int copied) 1879 { 1880 struct cxgbi_device *cdev = csk->cdev; 1881 int must_send; 1882 u32 credits; 1883 1884 log_debug(1 << CXGBI_DBG_PDU_RX, 1885 "csk 0x%p,%u,0x%lx,%u, seq %u, wup %u, thre %u, %u.\n", 1886 csk, csk->state, csk->flags, csk->tid, csk->copied_seq, 1887 csk->rcv_wup, cdev->rx_credit_thres, 1888 cdev->rcv_win); 1889 1890 if (csk->state != CTP_ESTABLISHED) 1891 return; 1892 1893 credits = csk->copied_seq - csk->rcv_wup; 1894 if (unlikely(!credits)) 1895 return; 1896 if (unlikely(cdev->rx_credit_thres == 0)) 1897 return; 1898 1899 must_send = credits + 16384 >= cdev->rcv_win; 1900 if (must_send || credits >= cdev->rx_credit_thres) 1901 csk->rcv_wup += cdev->csk_send_rx_credits(csk, credits); 1902 } 1903 1904 void cxgbi_conn_pdu_ready(struct cxgbi_sock *csk) 1905 { 1906 struct cxgbi_device *cdev = csk->cdev; 1907 struct iscsi_conn *conn = csk->user_data; 1908 struct sk_buff *skb; 1909 unsigned int read = 0; 1910 int err = 0; 1911 1912 log_debug(1 << CXGBI_DBG_PDU_RX, 1913 "csk 0x%p, conn 0x%p.\n", csk, conn); 1914 1915 if (unlikely(!conn || conn->suspend_rx)) { 1916 log_debug(1 << CXGBI_DBG_PDU_RX, 1917 "csk 0x%p, conn 0x%p, id %d, suspend_rx %lu!\n", 1918 csk, conn, conn ? conn->id : 0xFF, 1919 conn ? conn->suspend_rx : 0xFF); 1920 return; 1921 } 1922 1923 while (!err) { 1924 skb = skb_peek(&csk->receive_queue); 1925 if (!skb || 1926 !(cxgbi_skcb_test_flag(skb, SKCBF_RX_STATUS))) { 1927 if (skb) 1928 log_debug(1 << CXGBI_DBG_PDU_RX, 1929 "skb 0x%p, NOT ready 0x%lx.\n", 1930 skb, cxgbi_skcb_flags(skb)); 1931 break; 1932 } 1933 __skb_unlink(skb, &csk->receive_queue); 1934 1935 read += cxgbi_skcb_rx_pdulen(skb); 1936 log_debug(1 << CXGBI_DBG_PDU_RX, 1937 "csk 0x%p, skb 0x%p,%u,f 0x%lx, pdu len %u.\n", 1938 csk, skb, skb->len, cxgbi_skcb_flags(skb), 1939 cxgbi_skcb_rx_pdulen(skb)); 1940 1941 if (cxgbi_skcb_test_flag(skb, SKCBF_RX_COALESCED)) { 1942 err = skb_read_pdu_bhs(conn, skb); 1943 if (err < 0) { 1944 pr_err("coalesced bhs, csk 0x%p, skb 0x%p,%u, " 1945 "f 0x%lx, plen %u.\n", 1946 csk, skb, skb->len, 1947 cxgbi_skcb_flags(skb), 1948 cxgbi_skcb_rx_pdulen(skb)); 1949 goto skb_done; 1950 } 1951 err = skb_read_pdu_data(conn, skb, skb, 1952 err + cdev->skb_rx_extra); 1953 if (err < 0) 1954 pr_err("coalesced data, csk 0x%p, skb 0x%p,%u, " 1955 "f 0x%lx, plen %u.\n", 1956 csk, skb, skb->len, 1957 cxgbi_skcb_flags(skb), 1958 cxgbi_skcb_rx_pdulen(skb)); 1959 } else { 1960 err = skb_read_pdu_bhs(conn, skb); 1961 if (err < 0) { 1962 pr_err("bhs, csk 0x%p, skb 0x%p,%u, " 1963 "f 0x%lx, plen %u.\n", 1964 csk, skb, skb->len, 1965 cxgbi_skcb_flags(skb), 1966 cxgbi_skcb_rx_pdulen(skb)); 1967 goto skb_done; 1968 } 1969 1970 if (cxgbi_skcb_test_flag(skb, SKCBF_RX_DATA)) { 1971 struct sk_buff *dskb; 1972 1973 dskb = skb_peek(&csk->receive_queue); 1974 if (!dskb) { 1975 pr_err("csk 0x%p, skb 0x%p,%u, f 0x%lx," 1976 " plen %u, NO data.\n", 1977 csk, skb, skb->len, 1978 cxgbi_skcb_flags(skb), 1979 cxgbi_skcb_rx_pdulen(skb)); 1980 err = -EIO; 1981 goto skb_done; 1982 } 1983 __skb_unlink(dskb, &csk->receive_queue); 1984 1985 err = skb_read_pdu_data(conn, skb, dskb, 0); 1986 if (err < 0) 1987 pr_err("data, csk 0x%p, skb 0x%p,%u, " 1988 "f 0x%lx, plen %u, dskb 0x%p," 1989 "%u.\n", 1990 csk, skb, skb->len, 1991 cxgbi_skcb_flags(skb), 1992 cxgbi_skcb_rx_pdulen(skb), 1993 dskb, dskb->len); 1994 __kfree_skb(dskb); 1995 } else 1996 err = skb_read_pdu_data(conn, skb, skb, 0); 1997 } 1998 skb_done: 1999 __kfree_skb(skb); 2000 2001 if (err < 0) 2002 break; 2003 } 2004 2005 log_debug(1 << CXGBI_DBG_PDU_RX, "csk 0x%p, read %u.\n", csk, read); 2006 if (read) { 2007 csk->copied_seq += read; 2008 csk_return_rx_credits(csk, read); 2009 conn->rxdata_octets += read; 2010 } 2011 2012 if (err < 0) { 2013 pr_info("csk 0x%p, 0x%p, rx failed %d, read %u.\n", 2014 csk, conn, err, read); 2015 iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED); 2016 } 2017 } 2018 EXPORT_SYMBOL_GPL(cxgbi_conn_pdu_ready); 2019 2020 static int sgl_seek_offset(struct scatterlist *sgl, unsigned int sgcnt, 2021 unsigned int offset, unsigned int *off, 2022 struct scatterlist **sgp) 2023 { 2024 int i; 2025 struct scatterlist *sg; 2026 2027 for_each_sg(sgl, sg, sgcnt, i) { 2028 if (offset < sg->length) { 2029 *off = offset; 2030 *sgp = sg; 2031 return 0; 2032 } 2033 offset -= sg->length; 2034 } 2035 return -EFAULT; 2036 } 2037 2038 static int sgl_read_to_frags(struct scatterlist *sg, unsigned int sgoffset, 2039 unsigned int dlen, struct page_frag *frags, 2040 int frag_max) 2041 { 2042 unsigned int datalen = dlen; 2043 unsigned int sglen = sg->length - sgoffset; 2044 struct page *page = sg_page(sg); 2045 int i; 2046 2047 i = 0; 2048 do { 2049 unsigned int copy; 2050 2051 if (!sglen) { 2052 sg = sg_next(sg); 2053 if (!sg) { 2054 pr_warn("sg %d NULL, len %u/%u.\n", 2055 i, datalen, dlen); 2056 return -EINVAL; 2057 } 2058 sgoffset = 0; 2059 sglen = sg->length; 2060 page = sg_page(sg); 2061 2062 } 2063 copy = min(datalen, sglen); 2064 if (i && page == frags[i - 1].page && 2065 sgoffset + sg->offset == 2066 frags[i - 1].offset + frags[i - 1].size) { 2067 frags[i - 1].size += copy; 2068 } else { 2069 if (i >= frag_max) { 2070 pr_warn("too many pages %u, dlen %u.\n", 2071 frag_max, dlen); 2072 return -EINVAL; 2073 } 2074 2075 frags[i].page = page; 2076 frags[i].offset = sg->offset + sgoffset; 2077 frags[i].size = copy; 2078 i++; 2079 } 2080 datalen -= copy; 2081 sgoffset += copy; 2082 sglen -= copy; 2083 } while (datalen); 2084 2085 return i; 2086 } 2087 2088 int cxgbi_conn_alloc_pdu(struct iscsi_task *task, u8 opcode) 2089 { 2090 struct iscsi_tcp_conn *tcp_conn = task->conn->dd_data; 2091 struct cxgbi_conn *cconn = tcp_conn->dd_data; 2092 struct cxgbi_device *cdev = cconn->chba->cdev; 2093 struct iscsi_conn *conn = task->conn; 2094 struct iscsi_tcp_task *tcp_task = task->dd_data; 2095 struct cxgbi_task_data *tdata = iscsi_task_cxgbi_data(task); 2096 struct scsi_cmnd *sc = task->sc; 2097 int headroom = SKB_TX_ISCSI_PDU_HEADER_MAX; 2098 2099 tcp_task->dd_data = tdata; 2100 task->hdr = NULL; 2101 2102 if (SKB_MAX_HEAD(cdev->skb_tx_rsvd) > (512 * MAX_SKB_FRAGS) && 2103 (opcode == ISCSI_OP_SCSI_DATA_OUT || 2104 (opcode == ISCSI_OP_SCSI_CMD && 2105 (scsi_bidi_cmnd(sc) || sc->sc_data_direction == DMA_TO_DEVICE)))) 2106 /* data could goes into skb head */ 2107 headroom += min_t(unsigned int, 2108 SKB_MAX_HEAD(cdev->skb_tx_rsvd), 2109 conn->max_xmit_dlength); 2110 2111 tdata->skb = alloc_skb(cdev->skb_tx_rsvd + headroom, GFP_ATOMIC); 2112 if (!tdata->skb) { 2113 struct cxgbi_sock *csk = cconn->cep->csk; 2114 struct net_device *ndev = cdev->ports[csk->port_id]; 2115 ndev->stats.tx_dropped++; 2116 return -ENOMEM; 2117 } 2118 2119 skb_reserve(tdata->skb, cdev->skb_tx_rsvd); 2120 task->hdr = (struct iscsi_hdr *)tdata->skb->data; 2121 task->hdr_max = SKB_TX_ISCSI_PDU_HEADER_MAX; /* BHS + AHS */ 2122 2123 /* data_out uses scsi_cmd's itt */ 2124 if (opcode != ISCSI_OP_SCSI_DATA_OUT) 2125 task_reserve_itt(task, &task->hdr->itt); 2126 2127 log_debug(1 << CXGBI_DBG_ISCSI | 1 << CXGBI_DBG_PDU_TX, 2128 "task 0x%p, op 0x%x, skb 0x%p,%u+%u/%u, itt 0x%x.\n", 2129 task, opcode, tdata->skb, cdev->skb_tx_rsvd, headroom, 2130 conn->max_xmit_dlength, ntohl(task->hdr->itt)); 2131 2132 return 0; 2133 } 2134 EXPORT_SYMBOL_GPL(cxgbi_conn_alloc_pdu); 2135 2136 static inline void tx_skb_setmode(struct sk_buff *skb, int hcrc, int dcrc) 2137 { 2138 if (hcrc || dcrc) { 2139 u8 submode = 0; 2140 2141 if (hcrc) 2142 submode |= 1; 2143 if (dcrc) 2144 submode |= 2; 2145 cxgbi_skcb_ulp_mode(skb) = (ULP2_MODE_ISCSI << 4) | submode; 2146 } else 2147 cxgbi_skcb_ulp_mode(skb) = 0; 2148 } 2149 2150 int cxgbi_conn_init_pdu(struct iscsi_task *task, unsigned int offset, 2151 unsigned int count) 2152 { 2153 struct iscsi_conn *conn = task->conn; 2154 struct cxgbi_task_data *tdata = iscsi_task_cxgbi_data(task); 2155 struct sk_buff *skb = tdata->skb; 2156 unsigned int datalen = count; 2157 int i, padlen = iscsi_padding(count); 2158 struct page *pg; 2159 2160 log_debug(1 << CXGBI_DBG_ISCSI | 1 << CXGBI_DBG_PDU_TX, 2161 "task 0x%p,0x%p, skb 0x%p, 0x%x,0x%x,0x%x, %u+%u.\n", 2162 task, task->sc, skb, (*skb->data) & ISCSI_OPCODE_MASK, 2163 ntohl(task->cmdsn), ntohl(task->hdr->itt), offset, count); 2164 2165 skb_put(skb, task->hdr_len); 2166 tx_skb_setmode(skb, conn->hdrdgst_en, datalen ? conn->datadgst_en : 0); 2167 if (!count) 2168 return 0; 2169 2170 if (task->sc) { 2171 struct scsi_data_buffer *sdb = scsi_out(task->sc); 2172 struct scatterlist *sg = NULL; 2173 int err; 2174 2175 tdata->offset = offset; 2176 tdata->count = count; 2177 err = sgl_seek_offset( 2178 sdb->table.sgl, sdb->table.nents, 2179 tdata->offset, &tdata->sgoffset, &sg); 2180 if (err < 0) { 2181 pr_warn("tpdu, sgl %u, bad offset %u/%u.\n", 2182 sdb->table.nents, tdata->offset, sdb->length); 2183 return err; 2184 } 2185 err = sgl_read_to_frags(sg, tdata->sgoffset, tdata->count, 2186 tdata->frags, MAX_PDU_FRAGS); 2187 if (err < 0) { 2188 pr_warn("tpdu, sgl %u, bad offset %u + %u.\n", 2189 sdb->table.nents, tdata->offset, tdata->count); 2190 return err; 2191 } 2192 tdata->nr_frags = err; 2193 2194 if (tdata->nr_frags > MAX_SKB_FRAGS || 2195 (padlen && tdata->nr_frags == MAX_SKB_FRAGS)) { 2196 char *dst = skb->data + task->hdr_len; 2197 struct page_frag *frag = tdata->frags; 2198 2199 /* data fits in the skb's headroom */ 2200 for (i = 0; i < tdata->nr_frags; i++, frag++) { 2201 char *src = kmap_atomic(frag->page); 2202 2203 memcpy(dst, src+frag->offset, frag->size); 2204 dst += frag->size; 2205 kunmap_atomic(src); 2206 } 2207 if (padlen) { 2208 memset(dst, 0, padlen); 2209 padlen = 0; 2210 } 2211 skb_put(skb, count + padlen); 2212 } else { 2213 /* data fit into frag_list */ 2214 for (i = 0; i < tdata->nr_frags; i++) { 2215 __skb_fill_page_desc(skb, i, 2216 tdata->frags[i].page, 2217 tdata->frags[i].offset, 2218 tdata->frags[i].size); 2219 skb_frag_ref(skb, i); 2220 } 2221 skb_shinfo(skb)->nr_frags = tdata->nr_frags; 2222 skb->len += count; 2223 skb->data_len += count; 2224 skb->truesize += count; 2225 } 2226 2227 } else { 2228 pg = virt_to_page(task->data); 2229 2230 get_page(pg); 2231 skb_fill_page_desc(skb, 0, pg, offset_in_page(task->data), 2232 count); 2233 skb->len += count; 2234 skb->data_len += count; 2235 skb->truesize += count; 2236 } 2237 2238 if (padlen) { 2239 i = skb_shinfo(skb)->nr_frags; 2240 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags, 2241 virt_to_page(padding), offset_in_page(padding), 2242 padlen); 2243 2244 skb->data_len += padlen; 2245 skb->truesize += padlen; 2246 skb->len += padlen; 2247 } 2248 2249 return 0; 2250 } 2251 EXPORT_SYMBOL_GPL(cxgbi_conn_init_pdu); 2252 2253 int cxgbi_conn_xmit_pdu(struct iscsi_task *task) 2254 { 2255 struct iscsi_tcp_conn *tcp_conn = task->conn->dd_data; 2256 struct cxgbi_conn *cconn = tcp_conn->dd_data; 2257 struct cxgbi_task_data *tdata = iscsi_task_cxgbi_data(task); 2258 struct sk_buff *skb = tdata->skb; 2259 unsigned int datalen; 2260 int err; 2261 2262 if (!skb) { 2263 log_debug(1 << CXGBI_DBG_ISCSI | 1 << CXGBI_DBG_PDU_TX, 2264 "task 0x%p, skb NULL.\n", task); 2265 return 0; 2266 } 2267 2268 datalen = skb->data_len; 2269 tdata->skb = NULL; 2270 err = cxgbi_sock_send_pdus(cconn->cep->csk, skb); 2271 if (err > 0) { 2272 int pdulen = err; 2273 2274 log_debug(1 << CXGBI_DBG_PDU_TX, 2275 "task 0x%p,0x%p, skb 0x%p, len %u/%u, rv %d.\n", 2276 task, task->sc, skb, skb->len, skb->data_len, err); 2277 2278 if (task->conn->hdrdgst_en) 2279 pdulen += ISCSI_DIGEST_SIZE; 2280 2281 if (datalen && task->conn->datadgst_en) 2282 pdulen += ISCSI_DIGEST_SIZE; 2283 2284 task->conn->txdata_octets += pdulen; 2285 return 0; 2286 } 2287 2288 if (err == -EAGAIN || err == -ENOBUFS) { 2289 log_debug(1 << CXGBI_DBG_PDU_TX, 2290 "task 0x%p, skb 0x%p, len %u/%u, %d EAGAIN.\n", 2291 task, skb, skb->len, skb->data_len, err); 2292 /* reset skb to send when we are called again */ 2293 tdata->skb = skb; 2294 return err; 2295 } 2296 2297 log_debug(1 << CXGBI_DBG_ISCSI | 1 << CXGBI_DBG_PDU_TX, 2298 "itt 0x%x, skb 0x%p, len %u/%u, xmit err %d.\n", 2299 task->itt, skb, skb->len, skb->data_len, err); 2300 2301 kfree_skb(skb); 2302 2303 iscsi_conn_printk(KERN_ERR, task->conn, "xmit err %d.\n", err); 2304 iscsi_conn_failure(task->conn, ISCSI_ERR_XMIT_FAILED); 2305 return err; 2306 } 2307 EXPORT_SYMBOL_GPL(cxgbi_conn_xmit_pdu); 2308 2309 void cxgbi_cleanup_task(struct iscsi_task *task) 2310 { 2311 struct cxgbi_task_data *tdata = iscsi_task_cxgbi_data(task); 2312 2313 log_debug(1 << CXGBI_DBG_ISCSI, 2314 "task 0x%p, skb 0x%p, itt 0x%x.\n", 2315 task, tdata->skb, task->hdr_itt); 2316 2317 /* never reached the xmit task callout */ 2318 if (tdata->skb) 2319 __kfree_skb(tdata->skb); 2320 memset(tdata, 0, sizeof(*tdata)); 2321 2322 task_release_itt(task, task->hdr_itt); 2323 iscsi_tcp_cleanup_task(task); 2324 } 2325 EXPORT_SYMBOL_GPL(cxgbi_cleanup_task); 2326 2327 void cxgbi_get_conn_stats(struct iscsi_cls_conn *cls_conn, 2328 struct iscsi_stats *stats) 2329 { 2330 struct iscsi_conn *conn = cls_conn->dd_data; 2331 2332 stats->txdata_octets = conn->txdata_octets; 2333 stats->rxdata_octets = conn->rxdata_octets; 2334 stats->scsicmd_pdus = conn->scsicmd_pdus_cnt; 2335 stats->dataout_pdus = conn->dataout_pdus_cnt; 2336 stats->scsirsp_pdus = conn->scsirsp_pdus_cnt; 2337 stats->datain_pdus = conn->datain_pdus_cnt; 2338 stats->r2t_pdus = conn->r2t_pdus_cnt; 2339 stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt; 2340 stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt; 2341 stats->digest_err = 0; 2342 stats->timeout_err = 0; 2343 stats->custom_length = 1; 2344 strcpy(stats->custom[0].desc, "eh_abort_cnt"); 2345 stats->custom[0].value = conn->eh_abort_cnt; 2346 } 2347 EXPORT_SYMBOL_GPL(cxgbi_get_conn_stats); 2348 2349 static int cxgbi_conn_max_xmit_dlength(struct iscsi_conn *conn) 2350 { 2351 struct iscsi_tcp_conn *tcp_conn = conn->dd_data; 2352 struct cxgbi_conn *cconn = tcp_conn->dd_data; 2353 struct cxgbi_device *cdev = cconn->chba->cdev; 2354 unsigned int headroom = SKB_MAX_HEAD(cdev->skb_tx_rsvd); 2355 unsigned int max_def = 512 * MAX_SKB_FRAGS; 2356 unsigned int max = max(max_def, headroom); 2357 2358 max = min(cconn->chba->cdev->tx_max_size, max); 2359 if (conn->max_xmit_dlength) 2360 conn->max_xmit_dlength = min(conn->max_xmit_dlength, max); 2361 else 2362 conn->max_xmit_dlength = max; 2363 cxgbi_align_pdu_size(conn->max_xmit_dlength); 2364 2365 return 0; 2366 } 2367 2368 static int cxgbi_conn_max_recv_dlength(struct iscsi_conn *conn) 2369 { 2370 struct iscsi_tcp_conn *tcp_conn = conn->dd_data; 2371 struct cxgbi_conn *cconn = tcp_conn->dd_data; 2372 unsigned int max = cconn->chba->cdev->rx_max_size; 2373 2374 cxgbi_align_pdu_size(max); 2375 2376 if (conn->max_recv_dlength) { 2377 if (conn->max_recv_dlength > max) { 2378 pr_err("MaxRecvDataSegmentLength %u > %u.\n", 2379 conn->max_recv_dlength, max); 2380 return -EINVAL; 2381 } 2382 conn->max_recv_dlength = min(conn->max_recv_dlength, max); 2383 cxgbi_align_pdu_size(conn->max_recv_dlength); 2384 } else 2385 conn->max_recv_dlength = max; 2386 2387 return 0; 2388 } 2389 2390 int cxgbi_set_conn_param(struct iscsi_cls_conn *cls_conn, 2391 enum iscsi_param param, char *buf, int buflen) 2392 { 2393 struct iscsi_conn *conn = cls_conn->dd_data; 2394 struct iscsi_tcp_conn *tcp_conn = conn->dd_data; 2395 struct cxgbi_conn *cconn = tcp_conn->dd_data; 2396 struct cxgbi_sock *csk = cconn->cep->csk; 2397 int err; 2398 2399 log_debug(1 << CXGBI_DBG_ISCSI, 2400 "cls_conn 0x%p, param %d, buf(%d) %s.\n", 2401 cls_conn, param, buflen, buf); 2402 2403 switch (param) { 2404 case ISCSI_PARAM_HDRDGST_EN: 2405 err = iscsi_set_param(cls_conn, param, buf, buflen); 2406 if (!err && conn->hdrdgst_en) 2407 err = csk->cdev->csk_ddp_setup_digest(csk, csk->tid, 2408 conn->hdrdgst_en, 2409 conn->datadgst_en, 0); 2410 break; 2411 case ISCSI_PARAM_DATADGST_EN: 2412 err = iscsi_set_param(cls_conn, param, buf, buflen); 2413 if (!err && conn->datadgst_en) 2414 err = csk->cdev->csk_ddp_setup_digest(csk, csk->tid, 2415 conn->hdrdgst_en, 2416 conn->datadgst_en, 0); 2417 break; 2418 case ISCSI_PARAM_MAX_R2T: 2419 return iscsi_tcp_set_max_r2t(conn, buf); 2420 case ISCSI_PARAM_MAX_RECV_DLENGTH: 2421 err = iscsi_set_param(cls_conn, param, buf, buflen); 2422 if (!err) 2423 err = cxgbi_conn_max_recv_dlength(conn); 2424 break; 2425 case ISCSI_PARAM_MAX_XMIT_DLENGTH: 2426 err = iscsi_set_param(cls_conn, param, buf, buflen); 2427 if (!err) 2428 err = cxgbi_conn_max_xmit_dlength(conn); 2429 break; 2430 default: 2431 return iscsi_set_param(cls_conn, param, buf, buflen); 2432 } 2433 return err; 2434 } 2435 EXPORT_SYMBOL_GPL(cxgbi_set_conn_param); 2436 2437 static inline int csk_print_port(struct cxgbi_sock *csk, char *buf) 2438 { 2439 int len; 2440 2441 cxgbi_sock_get(csk); 2442 len = sprintf(buf, "%hu\n", ntohs(csk->daddr.sin_port)); 2443 cxgbi_sock_put(csk); 2444 2445 return len; 2446 } 2447 2448 static inline int csk_print_ip(struct cxgbi_sock *csk, char *buf) 2449 { 2450 int len; 2451 2452 cxgbi_sock_get(csk); 2453 if (csk->csk_family == AF_INET) 2454 len = sprintf(buf, "%pI4", 2455 &csk->daddr.sin_addr.s_addr); 2456 else 2457 len = sprintf(buf, "%pI6", 2458 &csk->daddr6.sin6_addr); 2459 2460 cxgbi_sock_put(csk); 2461 2462 return len; 2463 } 2464 2465 int cxgbi_get_ep_param(struct iscsi_endpoint *ep, enum iscsi_param param, 2466 char *buf) 2467 { 2468 struct cxgbi_endpoint *cep = ep->dd_data; 2469 struct cxgbi_sock *csk; 2470 int len; 2471 2472 log_debug(1 << CXGBI_DBG_ISCSI, 2473 "cls_conn 0x%p, param %d.\n", ep, param); 2474 2475 switch (param) { 2476 case ISCSI_PARAM_CONN_PORT: 2477 case ISCSI_PARAM_CONN_ADDRESS: 2478 if (!cep) 2479 return -ENOTCONN; 2480 2481 csk = cep->csk; 2482 if (!csk) 2483 return -ENOTCONN; 2484 2485 return iscsi_conn_get_addr_param((struct sockaddr_storage *) 2486 &csk->daddr, param, buf); 2487 default: 2488 return -ENOSYS; 2489 } 2490 return len; 2491 } 2492 EXPORT_SYMBOL_GPL(cxgbi_get_ep_param); 2493 2494 struct iscsi_cls_conn * 2495 cxgbi_create_conn(struct iscsi_cls_session *cls_session, u32 cid) 2496 { 2497 struct iscsi_cls_conn *cls_conn; 2498 struct iscsi_conn *conn; 2499 struct iscsi_tcp_conn *tcp_conn; 2500 struct cxgbi_conn *cconn; 2501 2502 cls_conn = iscsi_tcp_conn_setup(cls_session, sizeof(*cconn), cid); 2503 if (!cls_conn) 2504 return NULL; 2505 2506 conn = cls_conn->dd_data; 2507 tcp_conn = conn->dd_data; 2508 cconn = tcp_conn->dd_data; 2509 cconn->iconn = conn; 2510 2511 log_debug(1 << CXGBI_DBG_ISCSI, 2512 "cid %u(0x%x), cls 0x%p,0x%p, conn 0x%p,0x%p,0x%p.\n", 2513 cid, cid, cls_session, cls_conn, conn, tcp_conn, cconn); 2514 2515 return cls_conn; 2516 } 2517 EXPORT_SYMBOL_GPL(cxgbi_create_conn); 2518 2519 int cxgbi_bind_conn(struct iscsi_cls_session *cls_session, 2520 struct iscsi_cls_conn *cls_conn, 2521 u64 transport_eph, int is_leading) 2522 { 2523 struct iscsi_conn *conn = cls_conn->dd_data; 2524 struct iscsi_tcp_conn *tcp_conn = conn->dd_data; 2525 struct cxgbi_conn *cconn = tcp_conn->dd_data; 2526 struct iscsi_endpoint *ep; 2527 struct cxgbi_endpoint *cep; 2528 struct cxgbi_sock *csk; 2529 int err; 2530 2531 ep = iscsi_lookup_endpoint(transport_eph); 2532 if (!ep) 2533 return -EINVAL; 2534 2535 /* setup ddp pagesize */ 2536 cep = ep->dd_data; 2537 csk = cep->csk; 2538 err = csk->cdev->csk_ddp_setup_pgidx(csk, csk->tid, page_idx, 0); 2539 if (err < 0) 2540 return err; 2541 2542 err = iscsi_conn_bind(cls_session, cls_conn, is_leading); 2543 if (err) 2544 return -EINVAL; 2545 2546 /* calculate the tag idx bits needed for this conn based on cmds_max */ 2547 cconn->task_idx_bits = (__ilog2_u32(conn->session->cmds_max - 1)) + 1; 2548 2549 write_lock_bh(&csk->callback_lock); 2550 csk->user_data = conn; 2551 cconn->chba = cep->chba; 2552 cconn->cep = cep; 2553 cep->cconn = cconn; 2554 write_unlock_bh(&csk->callback_lock); 2555 2556 cxgbi_conn_max_xmit_dlength(conn); 2557 cxgbi_conn_max_recv_dlength(conn); 2558 2559 log_debug(1 << CXGBI_DBG_ISCSI, 2560 "cls 0x%p,0x%p, ep 0x%p, cconn 0x%p, csk 0x%p.\n", 2561 cls_session, cls_conn, ep, cconn, csk); 2562 /* init recv engine */ 2563 iscsi_tcp_hdr_recv_prep(tcp_conn); 2564 2565 return 0; 2566 } 2567 EXPORT_SYMBOL_GPL(cxgbi_bind_conn); 2568 2569 struct iscsi_cls_session *cxgbi_create_session(struct iscsi_endpoint *ep, 2570 u16 cmds_max, u16 qdepth, 2571 u32 initial_cmdsn) 2572 { 2573 struct cxgbi_endpoint *cep; 2574 struct cxgbi_hba *chba; 2575 struct Scsi_Host *shost; 2576 struct iscsi_cls_session *cls_session; 2577 struct iscsi_session *session; 2578 2579 if (!ep) { 2580 pr_err("missing endpoint.\n"); 2581 return NULL; 2582 } 2583 2584 cep = ep->dd_data; 2585 chba = cep->chba; 2586 shost = chba->shost; 2587 2588 BUG_ON(chba != iscsi_host_priv(shost)); 2589 2590 cls_session = iscsi_session_setup(chba->cdev->itp, shost, 2591 cmds_max, 0, 2592 sizeof(struct iscsi_tcp_task) + 2593 sizeof(struct cxgbi_task_data), 2594 initial_cmdsn, ISCSI_MAX_TARGET); 2595 if (!cls_session) 2596 return NULL; 2597 2598 session = cls_session->dd_data; 2599 if (iscsi_tcp_r2tpool_alloc(session)) 2600 goto remove_session; 2601 2602 log_debug(1 << CXGBI_DBG_ISCSI, 2603 "ep 0x%p, cls sess 0x%p.\n", ep, cls_session); 2604 return cls_session; 2605 2606 remove_session: 2607 iscsi_session_teardown(cls_session); 2608 return NULL; 2609 } 2610 EXPORT_SYMBOL_GPL(cxgbi_create_session); 2611 2612 void cxgbi_destroy_session(struct iscsi_cls_session *cls_session) 2613 { 2614 log_debug(1 << CXGBI_DBG_ISCSI, 2615 "cls sess 0x%p.\n", cls_session); 2616 2617 iscsi_tcp_r2tpool_free(cls_session->dd_data); 2618 iscsi_session_teardown(cls_session); 2619 } 2620 EXPORT_SYMBOL_GPL(cxgbi_destroy_session); 2621 2622 int cxgbi_set_host_param(struct Scsi_Host *shost, enum iscsi_host_param param, 2623 char *buf, int buflen) 2624 { 2625 struct cxgbi_hba *chba = iscsi_host_priv(shost); 2626 2627 if (!chba->ndev) { 2628 shost_printk(KERN_ERR, shost, "Could not get host param. " 2629 "netdev for host not set.\n"); 2630 return -ENODEV; 2631 } 2632 2633 log_debug(1 << CXGBI_DBG_ISCSI, 2634 "shost 0x%p, hba 0x%p,%s, param %d, buf(%d) %s.\n", 2635 shost, chba, chba->ndev->name, param, buflen, buf); 2636 2637 switch (param) { 2638 case ISCSI_HOST_PARAM_IPADDRESS: 2639 { 2640 __be32 addr = in_aton(buf); 2641 log_debug(1 << CXGBI_DBG_ISCSI, 2642 "hba %s, req. ipv4 %pI4.\n", chba->ndev->name, &addr); 2643 cxgbi_set_iscsi_ipv4(chba, addr); 2644 return 0; 2645 } 2646 case ISCSI_HOST_PARAM_HWADDRESS: 2647 case ISCSI_HOST_PARAM_NETDEV_NAME: 2648 return 0; 2649 default: 2650 return iscsi_host_set_param(shost, param, buf, buflen); 2651 } 2652 } 2653 EXPORT_SYMBOL_GPL(cxgbi_set_host_param); 2654 2655 int cxgbi_get_host_param(struct Scsi_Host *shost, enum iscsi_host_param param, 2656 char *buf) 2657 { 2658 struct cxgbi_hba *chba = iscsi_host_priv(shost); 2659 int len = 0; 2660 2661 if (!chba->ndev) { 2662 shost_printk(KERN_ERR, shost, "Could not get host param. " 2663 "netdev for host not set.\n"); 2664 return -ENODEV; 2665 } 2666 2667 log_debug(1 << CXGBI_DBG_ISCSI, 2668 "shost 0x%p, hba 0x%p,%s, param %d.\n", 2669 shost, chba, chba->ndev->name, param); 2670 2671 switch (param) { 2672 case ISCSI_HOST_PARAM_HWADDRESS: 2673 len = sysfs_format_mac(buf, chba->ndev->dev_addr, 6); 2674 break; 2675 case ISCSI_HOST_PARAM_NETDEV_NAME: 2676 len = sprintf(buf, "%s\n", chba->ndev->name); 2677 break; 2678 case ISCSI_HOST_PARAM_IPADDRESS: 2679 { 2680 struct cxgbi_sock *csk = find_sock_on_port(chba->cdev, 2681 chba->port_id); 2682 if (csk) { 2683 len = sprintf(buf, "%pIS", 2684 (struct sockaddr *)&csk->saddr); 2685 } 2686 log_debug(1 << CXGBI_DBG_ISCSI, 2687 "hba %s, addr %s.\n", chba->ndev->name, buf); 2688 break; 2689 } 2690 default: 2691 return iscsi_host_get_param(shost, param, buf); 2692 } 2693 2694 return len; 2695 } 2696 EXPORT_SYMBOL_GPL(cxgbi_get_host_param); 2697 2698 struct iscsi_endpoint *cxgbi_ep_connect(struct Scsi_Host *shost, 2699 struct sockaddr *dst_addr, 2700 int non_blocking) 2701 { 2702 struct iscsi_endpoint *ep; 2703 struct cxgbi_endpoint *cep; 2704 struct cxgbi_hba *hba = NULL; 2705 struct cxgbi_sock *csk; 2706 int err = -EINVAL; 2707 2708 log_debug(1 << CXGBI_DBG_ISCSI | 1 << CXGBI_DBG_SOCK, 2709 "shost 0x%p, non_blocking %d, dst_addr 0x%p.\n", 2710 shost, non_blocking, dst_addr); 2711 2712 if (shost) { 2713 hba = iscsi_host_priv(shost); 2714 if (!hba) { 2715 pr_info("shost 0x%p, priv NULL.\n", shost); 2716 goto err_out; 2717 } 2718 } 2719 2720 if (dst_addr->sa_family == AF_INET) { 2721 csk = cxgbi_check_route(dst_addr); 2722 #if IS_ENABLED(CONFIG_IPV6) 2723 } else if (dst_addr->sa_family == AF_INET6) { 2724 csk = cxgbi_check_route6(dst_addr); 2725 #endif 2726 } else { 2727 pr_info("address family 0x%x NOT supported.\n", 2728 dst_addr->sa_family); 2729 err = -EAFNOSUPPORT; 2730 return (struct iscsi_endpoint *)ERR_PTR(err); 2731 } 2732 2733 if (IS_ERR(csk)) 2734 return (struct iscsi_endpoint *)csk; 2735 cxgbi_sock_get(csk); 2736 2737 if (!hba) 2738 hba = csk->cdev->hbas[csk->port_id]; 2739 else if (hba != csk->cdev->hbas[csk->port_id]) { 2740 pr_info("Could not connect through requested host %u" 2741 "hba 0x%p != 0x%p (%u).\n", 2742 shost->host_no, hba, 2743 csk->cdev->hbas[csk->port_id], csk->port_id); 2744 err = -ENOSPC; 2745 goto release_conn; 2746 } 2747 2748 err = sock_get_port(csk); 2749 if (err) 2750 goto release_conn; 2751 2752 cxgbi_sock_set_state(csk, CTP_CONNECTING); 2753 err = csk->cdev->csk_init_act_open(csk); 2754 if (err) 2755 goto release_conn; 2756 2757 if (cxgbi_sock_is_closing(csk)) { 2758 err = -ENOSPC; 2759 pr_info("csk 0x%p is closing.\n", csk); 2760 goto release_conn; 2761 } 2762 2763 ep = iscsi_create_endpoint(sizeof(*cep)); 2764 if (!ep) { 2765 err = -ENOMEM; 2766 pr_info("iscsi alloc ep, OOM.\n"); 2767 goto release_conn; 2768 } 2769 2770 cep = ep->dd_data; 2771 cep->csk = csk; 2772 cep->chba = hba; 2773 2774 log_debug(1 << CXGBI_DBG_ISCSI | 1 << CXGBI_DBG_SOCK, 2775 "ep 0x%p, cep 0x%p, csk 0x%p, hba 0x%p,%s.\n", 2776 ep, cep, csk, hba, hba->ndev->name); 2777 return ep; 2778 2779 release_conn: 2780 cxgbi_sock_put(csk); 2781 cxgbi_sock_closed(csk); 2782 err_out: 2783 return ERR_PTR(err); 2784 } 2785 EXPORT_SYMBOL_GPL(cxgbi_ep_connect); 2786 2787 int cxgbi_ep_poll(struct iscsi_endpoint *ep, int timeout_ms) 2788 { 2789 struct cxgbi_endpoint *cep = ep->dd_data; 2790 struct cxgbi_sock *csk = cep->csk; 2791 2792 if (!cxgbi_sock_is_established(csk)) 2793 return 0; 2794 return 1; 2795 } 2796 EXPORT_SYMBOL_GPL(cxgbi_ep_poll); 2797 2798 void cxgbi_ep_disconnect(struct iscsi_endpoint *ep) 2799 { 2800 struct cxgbi_endpoint *cep = ep->dd_data; 2801 struct cxgbi_conn *cconn = cep->cconn; 2802 struct cxgbi_sock *csk = cep->csk; 2803 2804 log_debug(1 << CXGBI_DBG_ISCSI | 1 << CXGBI_DBG_SOCK, 2805 "ep 0x%p, cep 0x%p, cconn 0x%p, csk 0x%p,%u,0x%lx.\n", 2806 ep, cep, cconn, csk, csk->state, csk->flags); 2807 2808 if (cconn && cconn->iconn) { 2809 iscsi_suspend_tx(cconn->iconn); 2810 write_lock_bh(&csk->callback_lock); 2811 cep->csk->user_data = NULL; 2812 cconn->cep = NULL; 2813 write_unlock_bh(&csk->callback_lock); 2814 } 2815 iscsi_destroy_endpoint(ep); 2816 2817 if (likely(csk->state >= CTP_ESTABLISHED)) 2818 need_active_close(csk); 2819 else 2820 cxgbi_sock_closed(csk); 2821 2822 cxgbi_sock_put(csk); 2823 } 2824 EXPORT_SYMBOL_GPL(cxgbi_ep_disconnect); 2825 2826 int cxgbi_iscsi_init(struct iscsi_transport *itp, 2827 struct scsi_transport_template **stt) 2828 { 2829 *stt = iscsi_register_transport(itp); 2830 if (*stt == NULL) { 2831 pr_err("unable to register %s transport 0x%p.\n", 2832 itp->name, itp); 2833 return -ENODEV; 2834 } 2835 log_debug(1 << CXGBI_DBG_ISCSI, 2836 "%s, registered iscsi transport 0x%p.\n", 2837 itp->name, stt); 2838 return 0; 2839 } 2840 EXPORT_SYMBOL_GPL(cxgbi_iscsi_init); 2841 2842 void cxgbi_iscsi_cleanup(struct iscsi_transport *itp, 2843 struct scsi_transport_template **stt) 2844 { 2845 if (*stt) { 2846 log_debug(1 << CXGBI_DBG_ISCSI, 2847 "de-register transport 0x%p, %s, stt 0x%p.\n", 2848 itp, itp->name, *stt); 2849 *stt = NULL; 2850 iscsi_unregister_transport(itp); 2851 } 2852 } 2853 EXPORT_SYMBOL_GPL(cxgbi_iscsi_cleanup); 2854 2855 umode_t cxgbi_attr_is_visible(int param_type, int param) 2856 { 2857 switch (param_type) { 2858 case ISCSI_HOST_PARAM: 2859 switch (param) { 2860 case ISCSI_HOST_PARAM_NETDEV_NAME: 2861 case ISCSI_HOST_PARAM_HWADDRESS: 2862 case ISCSI_HOST_PARAM_IPADDRESS: 2863 case ISCSI_HOST_PARAM_INITIATOR_NAME: 2864 return S_IRUGO; 2865 default: 2866 return 0; 2867 } 2868 case ISCSI_PARAM: 2869 switch (param) { 2870 case ISCSI_PARAM_MAX_RECV_DLENGTH: 2871 case ISCSI_PARAM_MAX_XMIT_DLENGTH: 2872 case ISCSI_PARAM_HDRDGST_EN: 2873 case ISCSI_PARAM_DATADGST_EN: 2874 case ISCSI_PARAM_CONN_ADDRESS: 2875 case ISCSI_PARAM_CONN_PORT: 2876 case ISCSI_PARAM_EXP_STATSN: 2877 case ISCSI_PARAM_PERSISTENT_ADDRESS: 2878 case ISCSI_PARAM_PERSISTENT_PORT: 2879 case ISCSI_PARAM_PING_TMO: 2880 case ISCSI_PARAM_RECV_TMO: 2881 case ISCSI_PARAM_INITIAL_R2T_EN: 2882 case ISCSI_PARAM_MAX_R2T: 2883 case ISCSI_PARAM_IMM_DATA_EN: 2884 case ISCSI_PARAM_FIRST_BURST: 2885 case ISCSI_PARAM_MAX_BURST: 2886 case ISCSI_PARAM_PDU_INORDER_EN: 2887 case ISCSI_PARAM_DATASEQ_INORDER_EN: 2888 case ISCSI_PARAM_ERL: 2889 case ISCSI_PARAM_TARGET_NAME: 2890 case ISCSI_PARAM_TPGT: 2891 case ISCSI_PARAM_USERNAME: 2892 case ISCSI_PARAM_PASSWORD: 2893 case ISCSI_PARAM_USERNAME_IN: 2894 case ISCSI_PARAM_PASSWORD_IN: 2895 case ISCSI_PARAM_FAST_ABORT: 2896 case ISCSI_PARAM_ABORT_TMO: 2897 case ISCSI_PARAM_LU_RESET_TMO: 2898 case ISCSI_PARAM_TGT_RESET_TMO: 2899 case ISCSI_PARAM_IFACE_NAME: 2900 case ISCSI_PARAM_INITIATOR_NAME: 2901 return S_IRUGO; 2902 default: 2903 return 0; 2904 } 2905 } 2906 2907 return 0; 2908 } 2909 EXPORT_SYMBOL_GPL(cxgbi_attr_is_visible); 2910 2911 static int __init libcxgbi_init_module(void) 2912 { 2913 sw_tag_idx_bits = (__ilog2_u32(ISCSI_ITT_MASK)) + 1; 2914 sw_tag_age_bits = (__ilog2_u32(ISCSI_AGE_MASK)) + 1; 2915 2916 pr_info("tag itt 0x%x, %u bits, age 0x%x, %u bits.\n", 2917 ISCSI_ITT_MASK, sw_tag_idx_bits, 2918 ISCSI_AGE_MASK, sw_tag_age_bits); 2919 2920 ddp_setup_host_page_size(); 2921 return 0; 2922 } 2923 2924 static void __exit libcxgbi_exit_module(void) 2925 { 2926 cxgbi_device_unregister_all(0xFF); 2927 return; 2928 } 2929 2930 module_init(libcxgbi_init_module); 2931 module_exit(libcxgbi_exit_module); 2932