drivers/nvme/host/fc.c

   1 /*
   2  * Copyright (c) 2016 Avago Technologies.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of version 2 of the GNU General Public License as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful.
   9  * ALL EXPRESS OR IMPLIED CONDITIONS, REPRESENTATIONS AND WARRANTIES,
  10  * INCLUDING ANY IMPLIED WARRANTY OF MERCHANTABILITY, FITNESS FOR A
  11  * PARTICULAR PURPOSE, OR NON-INFRINGEMENT, ARE DISCLAIMED, EXCEPT TO
  12  * THE EXTENT THAT SUCH DISCLAIMERS ARE HELD TO BE LEGALLY INVALID.
  13  * See the GNU General Public License for more details, a copy of which
  14  * can be found in the file COPYING included with this package
  15  *
  16  */
  17 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  18 #include <linux/module.h>
  19 #include <linux/parser.h>
  20 #include <uapi/scsi/fc/fc_fs.h>
  21 #include <uapi/scsi/fc/fc_els.h>
  22
  23 #include "nvme.h"
  24 #include "fabrics.h"
  25 #include <linux/nvme-fc-driver.h>
  26 #include <linux/nvme-fc.h>
  27
  28
  29 /* *************************** Data Structures/Defines ****************** */
  30
  31
  32 /*
  33  * We handle AEN commands ourselves and don't even let the
  34  * block layer know about them.
  35  */
  36 #define NVME_FC_NR_AEN_COMMANDS 1
  37 #define NVME_FC_AQ_BLKMQ_DEPTH  \
  38         (NVMF_AQ_DEPTH - NVME_FC_NR_AEN_COMMANDS)
  39 #define AEN_CMDID_BASE          (NVME_FC_AQ_BLKMQ_DEPTH + 1)
  40
  41 enum nvme_fc_queue_flags {
  42         NVME_FC_Q_CONNECTED = (1 << 0),
  43 };
  44
  45 #define NVMEFC_QUEUE_DELAY      3               /* ms units */
  46
  47 struct nvme_fc_queue {
  48         struct nvme_fc_ctrl     *ctrl;
  49         struct device           *dev;
  50         struct blk_mq_hw_ctx    *hctx;
  51         void                    *lldd_handle;
  52         int                     queue_size;
  53         size_t                  cmnd_capsule_len;
  54         u32                     qnum;
  55         u32                     rqcnt;
  56         u32                     seqno;
  57
  58         u64                     connection_id;
  59         atomic_t                csn;
  60
  61         unsigned long           flags;
  62 } __aligned(sizeof(u64));       /* alignment for other things alloc'd with */
  63
  64 struct nvmefc_ls_req_op {
  65         struct nvmefc_ls_req    ls_req;
  66
  67         struct nvme_fc_ctrl     *ctrl;
  68         struct nvme_fc_queue    *queue;
  69         struct request          *rq;
  70
  71         int                     ls_error;
  72         struct completion       ls_done;
  73         struct list_head        lsreq_list;     /* ctrl->ls_req_list */
  74         bool                    req_queued;
  75 };
  76
  77 enum nvme_fcpop_state {
  78         FCPOP_STATE_UNINIT      = 0,
  79         FCPOP_STATE_IDLE        = 1,
  80         FCPOP_STATE_ACTIVE      = 2,
  81         FCPOP_STATE_ABORTED     = 3,
  82 };
  83
  84 struct nvme_fc_fcp_op {
  85         struct nvme_request     nreq;           /*
  86                                                  * nvme/host/core.c
  87                                                  * requires this to be
  88                                                  * the 1st element in the
  89                                                  * private structure
  90                                                  * associated with the
  91                                                  * request.
  92                                                  */
  93         struct nvmefc_fcp_req   fcp_req;
  94
  95         struct nvme_fc_ctrl     *ctrl;
  96         struct nvme_fc_queue    *queue;
  97         struct request          *rq;
  98
  99         atomic_t                state;
 100         u32                     rqno;
 101         u32                     nents;
 102
 103         struct nvme_fc_cmd_iu   cmd_iu;
 104         struct nvme_fc_ersp_iu  rsp_iu;
 105 };
 106
 107 struct nvme_fc_lport {
 108         struct nvme_fc_local_port       localport;
 109
 110         struct ida                      endp_cnt;
 111         struct list_head                port_list;      /* nvme_fc_port_list */
 112         struct list_head                endp_list;
 113         struct device                   *dev;   /* physical device for dma */
 114         struct nvme_fc_port_template    *ops;
 115         struct kref                     ref;
 116 } __aligned(sizeof(u64));       /* alignment for other things alloc'd with */
 117
 118 struct nvme_fc_rport {
 119         struct nvme_fc_remote_port      remoteport;
 120
 121         struct list_head                endp_list; /* for lport->endp_list */
 122         struct list_head                ctrl_list;
 123         spinlock_t                      lock;
 124         struct kref                     ref;
 125 } __aligned(sizeof(u64));       /* alignment for other things alloc'd with */
 126
 127 enum nvme_fcctrl_state {
 128         FCCTRL_INIT             = 0,
 129         FCCTRL_ACTIVE           = 1,
 130 };
 131
 132 struct nvme_fc_ctrl {
 133         spinlock_t              lock;
 134         struct nvme_fc_queue    *queues;
 135         u32                     queue_count;
 136
 137         struct device           *dev;
 138         struct nvme_fc_lport    *lport;
 139         struct nvme_fc_rport    *rport;
 140         u32                     cnum;
 141
 142         u64                     association_id;
 143
 144         u64                     cap;
 145
 146         struct list_head        ctrl_list;      /* rport->ctrl_list */
 147         struct list_head        ls_req_list;
 148
 149         struct blk_mq_tag_set   admin_tag_set;
 150         struct blk_mq_tag_set   tag_set;
 151
 152         struct work_struct      delete_work;
 153         struct kref             ref;
 154         int                     state;
 155
 156         struct nvme_fc_fcp_op   aen_ops[NVME_FC_NR_AEN_COMMANDS];
 157
 158         struct nvme_ctrl        ctrl;
 159 };
 160
 161 static inline struct nvme_fc_ctrl *
 162 to_fc_ctrl(struct nvme_ctrl *ctrl)
 163 {
 164         return container_of(ctrl, struct nvme_fc_ctrl, ctrl);
 165 }
 166
 167 static inline struct nvme_fc_lport *
 168 localport_to_lport(struct nvme_fc_local_port *portptr)
 169 {
 170         return container_of(portptr, struct nvme_fc_lport, localport);
 171 }
 172
 173 static inline struct nvme_fc_rport *
 174 remoteport_to_rport(struct nvme_fc_remote_port *portptr)
 175 {
 176         return container_of(portptr, struct nvme_fc_rport, remoteport);
 177 }
 178
 179 static inline struct nvmefc_ls_req_op *
 180 ls_req_to_lsop(struct nvmefc_ls_req *lsreq)
 181 {
 182         return container_of(lsreq, struct nvmefc_ls_req_op, ls_req);
 183 }
 184
 185 static inline struct nvme_fc_fcp_op *
 186 fcp_req_to_fcp_op(struct nvmefc_fcp_req *fcpreq)
 187 {
 188         return container_of(fcpreq, struct nvme_fc_fcp_op, fcp_req);
 189 }
 190
 191
 192
 193 /* *************************** Globals **************************** */
 194
 195
 196 static DEFINE_SPINLOCK(nvme_fc_lock);
 197
 198 static LIST_HEAD(nvme_fc_lport_list);
 199 static DEFINE_IDA(nvme_fc_local_port_cnt);
 200 static DEFINE_IDA(nvme_fc_ctrl_cnt);
 201
 202 static struct workqueue_struct *nvme_fc_wq;
 203
 204
 205
 206 /* *********************** FC-NVME Port Management ************************ */
 207
 208 static int __nvme_fc_del_ctrl(struct nvme_fc_ctrl *);
 209 static void __nvme_fc_delete_hw_queue(struct nvme_fc_ctrl *,
 210                         struct nvme_fc_queue *, unsigned int);
 211
 212
 213 /**
 214  * nvme_fc_register_localport - transport entry point called by an
 215  *                              LLDD to register the existence of a NVME
 216  *                              host FC port.
 217  * @pinfo:     pointer to information about the port to be registered
 218  * @template:  LLDD entrypoints and operational parameters for the port
 219  * @dev:       physical hardware device node port corresponds to. Will be
 220  *             used for DMA mappings
 221  * @lport_p:   pointer to a local port pointer. Upon success, the routine
 222  *             will allocate a nvme_fc_local_port structure and place its
 223  *             address in the local port pointer. Upon failure, local port
 224  *             pointer will be set to 0.
 225  *
 226  * Returns:
 227  * a completion status. Must be 0 upon success; a negative errno
 228  * (ex: -ENXIO) upon failure.
 229  */
 230 int
 231 nvme_fc_register_localport(struct nvme_fc_port_info *pinfo,
 232                         struct nvme_fc_port_template *template,
 233                         struct device *dev,
 234                         struct nvme_fc_local_port **portptr)
 235 {
 236         struct nvme_fc_lport *newrec;
 237         unsigned long flags;
 238         int ret, idx;
 239
 240         if (!template->localport_delete || !template->remoteport_delete ||
 241             !template->ls_req || !template->fcp_io ||
 242             !template->ls_abort || !template->fcp_abort ||
 243             !template->max_hw_queues || !template->max_sgl_segments ||
 244             !template->max_dif_sgl_segments || !template->dma_boundary) {
 245                 ret = -EINVAL;
 246                 goto out_reghost_failed;
 247         }
 248
 249         newrec = kmalloc((sizeof(*newrec) + template->local_priv_sz),
 250                          GFP_KERNEL);
 251         if (!newrec) {
 252                 ret = -ENOMEM;
 253                 goto out_reghost_failed;
 254         }
 255
 256         idx = ida_simple_get(&nvme_fc_local_port_cnt, 0, 0, GFP_KERNEL);
 257         if (idx < 0) {
 258                 ret = -ENOSPC;
 259                 goto out_fail_kfree;
 260         }
 261
 262         if (!get_device(dev) && dev) {
 263                 ret = -ENODEV;
 264                 goto out_ida_put;
 265         }
 266
 267         INIT_LIST_HEAD(&newrec->port_list);
 268         INIT_LIST_HEAD(&newrec->endp_list);
 269         kref_init(&newrec->ref);
 270         newrec->ops = template;
 271         newrec->dev = dev;
 272         ida_init(&newrec->endp_cnt);
 273         newrec->localport.private = &newrec[1];
 274         newrec->localport.node_name = pinfo->node_name;
 275         newrec->localport.port_name = pinfo->port_name;
 276         newrec->localport.port_role = pinfo->port_role;
 277         newrec->localport.port_id = pinfo->port_id;
 278         newrec->localport.port_state = FC_OBJSTATE_ONLINE;
 279         newrec->localport.port_num = idx;
 280
 281         spin_lock_irqsave(&nvme_fc_lock, flags);
 282         list_add_tail(&newrec->port_list, &nvme_fc_lport_list);
 283         spin_unlock_irqrestore(&nvme_fc_lock, flags);
 284
 285         if (dev)
 286                 dma_set_seg_boundary(dev, template->dma_boundary);
 287
 288         *portptr = &newrec->localport;
 289         return 0;
 290
 291 out_ida_put:
 292         ida_simple_remove(&nvme_fc_local_port_cnt, idx);
 293 out_fail_kfree:
 294         kfree(newrec);
 295 out_reghost_failed:
 296         *portptr = NULL;
 297
 298         return ret;
 299 }
 300 EXPORT_SYMBOL_GPL(nvme_fc_register_localport);
 301
 302 static void
 303 nvme_fc_free_lport(struct kref *ref)
 304 {
 305         struct nvme_fc_lport *lport =
 306                 container_of(ref, struct nvme_fc_lport, ref);
 307         unsigned long flags;
 308
 309         WARN_ON(lport->localport.port_state != FC_OBJSTATE_DELETED);
 310         WARN_ON(!list_empty(&lport->endp_list));
 311
 312         /* remove from transport list */
 313         spin_lock_irqsave(&nvme_fc_lock, flags);
 314         list_del(&lport->port_list);
 315         spin_unlock_irqrestore(&nvme_fc_lock, flags);
 316
 317         /* let the LLDD know we've finished tearing it down */
 318         lport->ops->localport_delete(&lport->localport);
 319
 320         ida_simple_remove(&nvme_fc_local_port_cnt, lport->localport.port_num);
 321         ida_destroy(&lport->endp_cnt);
 322
 323         put_device(lport->dev);
 324
 325         kfree(lport);
 326 }
 327
 328 static void
 329 nvme_fc_lport_put(struct nvme_fc_lport *lport)
 330 {
 331         kref_put(&lport->ref, nvme_fc_free_lport);
 332 }
 333
 334 static int
 335 nvme_fc_lport_get(struct nvme_fc_lport *lport)
 336 {
 337         return kref_get_unless_zero(&lport->ref);
 338 }
 339
 340 /**
 341  * nvme_fc_unregister_localport - transport entry point called by an
 342  *                              LLDD to deregister/remove a previously
 343  *                              registered a NVME host FC port.
 344  * @localport: pointer to the (registered) local port that is to be
 345  *             deregistered.
 346  *
 347  * Returns:
 348  * a completion status. Must be 0 upon success; a negative errno
 349  * (ex: -ENXIO) upon failure.
 350  */
 351 int
 352 nvme_fc_unregister_localport(struct nvme_fc_local_port *portptr)
 353 {
 354         struct nvme_fc_lport *lport = localport_to_lport(portptr);
 355         unsigned long flags;
 356
 357         if (!portptr)
 358                 return -EINVAL;
 359
 360         spin_lock_irqsave(&nvme_fc_lock, flags);
 361
 362         if (portptr->port_state != FC_OBJSTATE_ONLINE) {
 363                 spin_unlock_irqrestore(&nvme_fc_lock, flags);
 364                 return -EINVAL;
 365         }
 366         portptr->port_state = FC_OBJSTATE_DELETED;
 367
 368         spin_unlock_irqrestore(&nvme_fc_lock, flags);
 369
 370         nvme_fc_lport_put(lport);
 371
 372         return 0;
 373 }
 374 EXPORT_SYMBOL_GPL(nvme_fc_unregister_localport);
 375
 376 /**
 377  * nvme_fc_register_remoteport - transport entry point called by an
 378  *                              LLDD to register the existence of a NVME
 379  *                              subsystem FC port on its fabric.
 380  * @localport: pointer to the (registered) local port that the remote
 381  *             subsystem port is connected to.
 382  * @pinfo:     pointer to information about the port to be registered
 383  * @rport_p:   pointer to a remote port pointer. Upon success, the routine
 384  *             will allocate a nvme_fc_remote_port structure and place its
 385  *             address in the remote port pointer. Upon failure, remote port
 386  *             pointer will be set to 0.
 387  *
 388  * Returns:
 389  * a completion status. Must be 0 upon success; a negative errno
 390  * (ex: -ENXIO) upon failure.
 391  */
 392 int
 393 nvme_fc_register_remoteport(struct nvme_fc_local_port *localport,
 394                                 struct nvme_fc_port_info *pinfo,
 395                                 struct nvme_fc_remote_port **portptr)
 396 {
 397         struct nvme_fc_lport *lport = localport_to_lport(localport);
 398         struct nvme_fc_rport *newrec;
 399         unsigned long flags;
 400         int ret, idx;
 401
 402         newrec = kmalloc((sizeof(*newrec) + lport->ops->remote_priv_sz),
 403                          GFP_KERNEL);
 404         if (!newrec) {
 405                 ret = -ENOMEM;
 406                 goto out_reghost_failed;
 407         }
 408
 409         if (!nvme_fc_lport_get(lport)) {
 410                 ret = -ESHUTDOWN;
 411                 goto out_kfree_rport;
 412         }
 413
 414         idx = ida_simple_get(&lport->endp_cnt, 0, 0, GFP_KERNEL);
 415         if (idx < 0) {
 416                 ret = -ENOSPC;
 417                 goto out_lport_put;
 418         }
 419
 420         INIT_LIST_HEAD(&newrec->endp_list);
 421         INIT_LIST_HEAD(&newrec->ctrl_list);
 422         kref_init(&newrec->ref);
 423         spin_lock_init(&newrec->lock);
 424         newrec->remoteport.localport = &lport->localport;
 425         newrec->remoteport.private = &newrec[1];
 426         newrec->remoteport.port_role = pinfo->port_role;
 427         newrec->remoteport.node_name = pinfo->node_name;
 428         newrec->remoteport.port_name = pinfo->port_name;
 429         newrec->remoteport.port_id = pinfo->port_id;
 430         newrec->remoteport.port_state = FC_OBJSTATE_ONLINE;
 431         newrec->remoteport.port_num = idx;
 432
 433         spin_lock_irqsave(&nvme_fc_lock, flags);
 434         list_add_tail(&newrec->endp_list, &lport->endp_list);
 435         spin_unlock_irqrestore(&nvme_fc_lock, flags);
 436
 437         *portptr = &newrec->remoteport;
 438         return 0;
 439
 440 out_lport_put:
 441         nvme_fc_lport_put(lport);
 442 out_kfree_rport:
 443         kfree(newrec);
 444 out_reghost_failed:
 445         *portptr = NULL;
 446         return ret;
 447
 448 }
 449 EXPORT_SYMBOL_GPL(nvme_fc_register_remoteport);
 450
 451 static void
 452 nvme_fc_free_rport(struct kref *ref)
 453 {
 454         struct nvme_fc_rport *rport =
 455                 container_of(ref, struct nvme_fc_rport, ref);
 456         struct nvme_fc_lport *lport =
 457                         localport_to_lport(rport->remoteport.localport);
 458         unsigned long flags;
 459
 460         WARN_ON(rport->remoteport.port_state != FC_OBJSTATE_DELETED);
 461         WARN_ON(!list_empty(&rport->ctrl_list));
 462
 463         /* remove from lport list */
 464         spin_lock_irqsave(&nvme_fc_lock, flags);
 465         list_del(&rport->endp_list);
 466         spin_unlock_irqrestore(&nvme_fc_lock, flags);
 467
 468         /* let the LLDD know we've finished tearing it down */
 469         lport->ops->remoteport_delete(&rport->remoteport);
 470
 471         ida_simple_remove(&lport->endp_cnt, rport->remoteport.port_num);
 472
 473         kfree(rport);
 474
 475         nvme_fc_lport_put(lport);
 476 }
 477
 478 static void
 479 nvme_fc_rport_put(struct nvme_fc_rport *rport)
 480 {
 481         kref_put(&rport->ref, nvme_fc_free_rport);
 482 }
 483
 484 static int
 485 nvme_fc_rport_get(struct nvme_fc_rport *rport)
 486 {
 487         return kref_get_unless_zero(&rport->ref);
 488 }
 489
 490 /**
 491  * nvme_fc_unregister_remoteport - transport entry point called by an
 492  *                              LLDD to deregister/remove a previously
 493  *                              registered a NVME subsystem FC port.
 494  * @remoteport: pointer to the (registered) remote port that is to be
 495  *              deregistered.
 496  *
 497  * Returns:
 498  * a completion status. Must be 0 upon success; a negative errno
 499  * (ex: -ENXIO) upon failure.
 500  */
 501 int
 502 nvme_fc_unregister_remoteport(struct nvme_fc_remote_port *portptr)
 503 {
 504         struct nvme_fc_rport *rport = remoteport_to_rport(portptr);
 505         struct nvme_fc_ctrl *ctrl;
 506         unsigned long flags;
 507
 508         if (!portptr)
 509                 return -EINVAL;
 510
 511         spin_lock_irqsave(&rport->lock, flags);
 512
 513         if (portptr->port_state != FC_OBJSTATE_ONLINE) {
 514                 spin_unlock_irqrestore(&rport->lock, flags);
 515                 return -EINVAL;
 516         }
 517         portptr->port_state = FC_OBJSTATE_DELETED;
 518
 519         /* tear down all associations to the remote port */
 520         list_for_each_entry(ctrl, &rport->ctrl_list, ctrl_list)
 521                 __nvme_fc_del_ctrl(ctrl);
 522
 523         spin_unlock_irqrestore(&rport->lock, flags);
 524
 525         nvme_fc_rport_put(rport);
 526         return 0;
 527 }
 528 EXPORT_SYMBOL_GPL(nvme_fc_unregister_remoteport);
 529
 530
 531 /* *********************** FC-NVME DMA Handling **************************** */
 532
 533 /*
 534  * The fcloop device passes in a NULL device pointer. Real LLD's will
 535  * pass in a valid device pointer. If NULL is passed to the dma mapping
 536  * routines, depending on the platform, it may or may not succeed, and
 537  * may crash.
 538  *
 539  * As such:
 540  * Wrapper all the dma routines and check the dev pointer.
 541  *
 542  * If simple mappings (return just a dma address, we'll noop them,
 543  * returning a dma address of 0.
 544  *
 545  * On more complex mappings (dma_map_sg), a pseudo routine fills
 546  * in the scatter list, setting all dma addresses to 0.
 547  */
 548
 549 static inline dma_addr_t
 550 fc_dma_map_single(struct device *dev, void *ptr, size_t size,
 551                 enum dma_data_direction dir)
 552 {
 553         return dev ? dma_map_single(dev, ptr, size, dir) : (dma_addr_t)0L;
 554 }
 555
 556 static inline int
 557 fc_dma_mapping_error(struct device *dev, dma_addr_t dma_addr)
 558 {
 559         return dev ? dma_mapping_error(dev, dma_addr) : 0;
 560 }
 561
 562 static inline void
 563 fc_dma_unmap_single(struct device *dev, dma_addr_t addr, size_t size,
 564         enum dma_data_direction dir)
 565 {
 566         if (dev)
 567                 dma_unmap_single(dev, addr, size, dir);
 568 }
 569
 570 static inline void
 571 fc_dma_sync_single_for_cpu(struct device *dev, dma_addr_t addr, size_t size,
 572                 enum dma_data_direction dir)
 573 {
 574         if (dev)
 575                 dma_sync_single_for_cpu(dev, addr, size, dir);
 576 }
 577
 578 static inline void
 579 fc_dma_sync_single_for_device(struct device *dev, dma_addr_t addr, size_t size,
 580                 enum dma_data_direction dir)
 581 {
 582         if (dev)
 583                 dma_sync_single_for_device(dev, addr, size, dir);
 584 }
 585
 586 /* pseudo dma_map_sg call */
 587 static int
 588 fc_map_sg(struct scatterlist *sg, int nents)
 589 {
 590         struct scatterlist *s;
 591         int i;
 592
 593         WARN_ON(nents == 0 || sg[0].length == 0);
 594
 595         for_each_sg(sg, s, nents, i) {
 596                 s->dma_address = 0L;
 597 #ifdef CONFIG_NEED_SG_DMA_LENGTH
 598                 s->dma_length = s->length;
 599 #endif
 600         }
 601         return nents;
 602 }
 603
 604 static inline int
 605 fc_dma_map_sg(struct device *dev, struct scatterlist *sg, int nents,
 606                 enum dma_data_direction dir)
 607 {
 608         return dev ? dma_map_sg(dev, sg, nents, dir) : fc_map_sg(sg, nents);
 609 }
 610
 611 static inline void
 612 fc_dma_unmap_sg(struct device *dev, struct scatterlist *sg, int nents,
 613                 enum dma_data_direction dir)
 614 {
 615         if (dev)
 616                 dma_unmap_sg(dev, sg, nents, dir);
 617 }
 618
 619
 620 /* *********************** FC-NVME LS Handling **************************** */
 621
 622 static void nvme_fc_ctrl_put(struct nvme_fc_ctrl *);
 623 static int nvme_fc_ctrl_get(struct nvme_fc_ctrl *);
 624
 625
 626 static void
 627 __nvme_fc_finish_ls_req(struct nvme_fc_ctrl *ctrl,
 628                 struct nvmefc_ls_req_op *lsop)
 629 {
 630         struct nvmefc_ls_req *lsreq = &lsop->ls_req;
 631         unsigned long flags;
 632
 633         spin_lock_irqsave(&ctrl->lock, flags);
 634
 635         if (!lsop->req_queued) {
 636                 spin_unlock_irqrestore(&ctrl->lock, flags);
 637                 return;
 638         }
 639
 640         list_del(&lsop->lsreq_list);
 641
 642         lsop->req_queued = false;
 643
 644         spin_unlock_irqrestore(&ctrl->lock, flags);
 645
 646         fc_dma_unmap_single(ctrl->dev, lsreq->rqstdma,
 647                                   (lsreq->rqstlen + lsreq->rsplen),
 648                                   DMA_BIDIRECTIONAL);
 649
 650         nvme_fc_ctrl_put(ctrl);
 651 }
 652
 653 static int
 654 __nvme_fc_send_ls_req(struct nvme_fc_ctrl *ctrl,
 655                 struct nvmefc_ls_req_op *lsop,
 656                 void (*done)(struct nvmefc_ls_req *req, int status))
 657 {
 658         struct nvmefc_ls_req *lsreq = &lsop->ls_req;
 659         unsigned long flags;
 660         int ret;
 661
 662         if (!nvme_fc_ctrl_get(ctrl))
 663                 return -ESHUTDOWN;
 664
 665         lsreq->done = done;
 666         lsop->ctrl = ctrl;
 667         lsop->req_queued = false;
 668         INIT_LIST_HEAD(&lsop->lsreq_list);
 669         init_completion(&lsop->ls_done);
 670
 671         lsreq->rqstdma = fc_dma_map_single(ctrl->dev, lsreq->rqstaddr,
 672                                   lsreq->rqstlen + lsreq->rsplen,
 673                                   DMA_BIDIRECTIONAL);
 674         if (fc_dma_mapping_error(ctrl->dev, lsreq->rqstdma)) {
 675                 nvme_fc_ctrl_put(ctrl);
 676                 dev_err(ctrl->dev,
 677                         "els request command failed EFAULT.\n");
 678                 return -EFAULT;
 679         }
 680         lsreq->rspdma = lsreq->rqstdma + lsreq->rqstlen;
 681
 682         spin_lock_irqsave(&ctrl->lock, flags);
 683
 684         list_add_tail(&lsop->lsreq_list, &ctrl->ls_req_list);
 685
 686         lsop->req_queued = true;
 687
 688         spin_unlock_irqrestore(&ctrl->lock, flags);
 689
 690         ret = ctrl->lport->ops->ls_req(&ctrl->lport->localport,
 691                                         &ctrl->rport->remoteport, lsreq);
 692         if (ret)
 693                 lsop->ls_error = ret;
 694
 695         return ret;
 696 }
 697
 698 static void
 699 nvme_fc_send_ls_req_done(struct nvmefc_ls_req *lsreq, int status)
 700 {
 701         struct nvmefc_ls_req_op *lsop = ls_req_to_lsop(lsreq);
 702
 703         lsop->ls_error = status;
 704         complete(&lsop->ls_done);
 705 }
 706
 707 static int
 708 nvme_fc_send_ls_req(struct nvme_fc_ctrl *ctrl, struct nvmefc_ls_req_op *lsop)
 709 {
 710         struct nvmefc_ls_req *lsreq = &lsop->ls_req;
 711         struct fcnvme_ls_rjt *rjt = lsreq->rspaddr;
 712         int ret;
 713
 714         ret = __nvme_fc_send_ls_req(ctrl, lsop, nvme_fc_send_ls_req_done);
 715
 716         if (!ret)
 717                 /*
 718                  * No timeout/not interruptible as we need the struct
 719                  * to exist until the lldd calls us back. Thus mandate
 720                  * wait until driver calls back. lldd responsible for
 721                  * the timeout action
 722                  */
 723                 wait_for_completion(&lsop->ls_done);
 724
 725         __nvme_fc_finish_ls_req(ctrl, lsop);
 726
 727         if (ret) {
 728                 dev_err(ctrl->dev,
 729                         "ls request command failed (%d).\n", ret);
 730                 return ret;
 731         }
 732
 733         /* ACC or RJT payload ? */
 734         if (rjt->w0.ls_cmd == FCNVME_LS_RJT)
 735                 return -ENXIO;
 736
 737         return 0;
 738 }
 739
 740 static void
 741 nvme_fc_send_ls_req_async(struct nvme_fc_ctrl *ctrl,
 742                 struct nvmefc_ls_req_op *lsop,
 743                 void (*done)(struct nvmefc_ls_req *req, int status))
 744 {
 745         int ret;
 746
 747         ret = __nvme_fc_send_ls_req(ctrl, lsop, done);
 748
 749         /* don't wait for completion */
 750
 751         if (ret)
 752                 done(&lsop->ls_req, ret);
 753 }
 754
 755 /* Validation Error indexes into the string table below */
 756 enum {
 757         VERR_NO_ERROR           = 0,
 758         VERR_LSACC              = 1,
 759         VERR_LSDESC_RQST        = 2,
 760         VERR_LSDESC_RQST_LEN    = 3,
 761         VERR_ASSOC_ID           = 4,
 762         VERR_ASSOC_ID_LEN       = 5,
 763         VERR_CONN_ID            = 6,
 764         VERR_CONN_ID_LEN        = 7,
 765         VERR_CR_ASSOC           = 8,
 766         VERR_CR_ASSOC_ACC_LEN   = 9,
 767         VERR_CR_CONN            = 10,
 768         VERR_CR_CONN_ACC_LEN    = 11,
 769         VERR_DISCONN            = 12,
 770         VERR_DISCONN_ACC_LEN    = 13,
 771 };
 772
 773 static char *validation_errors[] = {
 774         "OK",
 775         "Not LS_ACC",
 776         "Not LSDESC_RQST",
 777         "Bad LSDESC_RQST Length",
 778         "Not Association ID",
 779         "Bad Association ID Length",
 780         "Not Connection ID",
 781         "Bad Connection ID Length",
 782         "Not CR_ASSOC Rqst",
 783         "Bad CR_ASSOC ACC Length",
 784         "Not CR_CONN Rqst",
 785         "Bad CR_CONN ACC Length",
 786         "Not Disconnect Rqst",
 787         "Bad Disconnect ACC Length",
 788 };
 789
 790 static int
 791 nvme_fc_connect_admin_queue(struct nvme_fc_ctrl *ctrl,
 792         struct nvme_fc_queue *queue, u16 qsize, u16 ersp_ratio)
 793 {
 794         struct nvmefc_ls_req_op *lsop;
 795         struct nvmefc_ls_req *lsreq;
 796         struct fcnvme_ls_cr_assoc_rqst *assoc_rqst;
 797         struct fcnvme_ls_cr_assoc_acc *assoc_acc;
 798         int ret, fcret = 0;
 799
 800         lsop = kzalloc((sizeof(*lsop) +
 801                          ctrl->lport->ops->lsrqst_priv_sz +
 802                          sizeof(*assoc_rqst) + sizeof(*assoc_acc)), GFP_KERNEL);
 803         if (!lsop) {
 804                 ret = -ENOMEM;
 805                 goto out_no_memory;
 806         }
 807         lsreq = &lsop->ls_req;
 808
 809         lsreq->private = (void *)&lsop[1];
 810         assoc_rqst = (struct fcnvme_ls_cr_assoc_rqst *)
 811                         (lsreq->private + ctrl->lport->ops->lsrqst_priv_sz);
 812         assoc_acc = (struct fcnvme_ls_cr_assoc_acc *)&assoc_rqst[1];
 813
 814         assoc_rqst->w0.ls_cmd = FCNVME_LS_CREATE_ASSOCIATION;
 815         assoc_rqst->desc_list_len =
 816                         cpu_to_be32(sizeof(struct fcnvme_lsdesc_cr_assoc_cmd));
 817
 818         assoc_rqst->assoc_cmd.desc_tag =
 819                         cpu_to_be32(FCNVME_LSDESC_CREATE_ASSOC_CMD);
 820         assoc_rqst->assoc_cmd.desc_len =
 821                         fcnvme_lsdesc_len(
 822                                 sizeof(struct fcnvme_lsdesc_cr_assoc_cmd));
 823
 824         assoc_rqst->assoc_cmd.ersp_ratio = cpu_to_be16(ersp_ratio);
 825         assoc_rqst->assoc_cmd.sqsize = cpu_to_be16(qsize);
 826         /* Linux supports only Dynamic controllers */
 827         assoc_rqst->assoc_cmd.cntlid = cpu_to_be16(0xffff);
 828         memcpy(&assoc_rqst->assoc_cmd.hostid, &ctrl->ctrl.opts->host->id,
 829                 min_t(size_t, FCNVME_ASSOC_HOSTID_LEN, sizeof(uuid_be)));
 830         strncpy(assoc_rqst->assoc_cmd.hostnqn, ctrl->ctrl.opts->host->nqn,
 831                 min(FCNVME_ASSOC_HOSTNQN_LEN, NVMF_NQN_SIZE));
 832         strncpy(assoc_rqst->assoc_cmd.subnqn, ctrl->ctrl.opts->subsysnqn,
 833                 min(FCNVME_ASSOC_SUBNQN_LEN, NVMF_NQN_SIZE));
 834
 835         lsop->queue = queue;
 836         lsreq->rqstaddr = assoc_rqst;
 837         lsreq->rqstlen = sizeof(*assoc_rqst);
 838         lsreq->rspaddr = assoc_acc;
 839         lsreq->rsplen = sizeof(*assoc_acc);
 840         lsreq->timeout = NVME_FC_CONNECT_TIMEOUT_SEC;
 841
 842         ret = nvme_fc_send_ls_req(ctrl, lsop);
 843         if (ret)
 844                 goto out_free_buffer;
 845
 846         /* process connect LS completion */
 847
 848         /* validate the ACC response */
 849         if (assoc_acc->hdr.w0.ls_cmd != FCNVME_LS_ACC)
 850                 fcret = VERR_LSACC;
 851         if (assoc_acc->hdr.desc_list_len !=
 852                         fcnvme_lsdesc_len(
 853                                 sizeof(struct fcnvme_ls_cr_assoc_acc)))
 854                 fcret = VERR_CR_ASSOC_ACC_LEN;
 855         if (assoc_acc->hdr.rqst.desc_tag != cpu_to_be32(FCNVME_LSDESC_RQST))
 856                 fcret = VERR_LSDESC_RQST;
 857         else if (assoc_acc->hdr.rqst.desc_len !=
 858                         fcnvme_lsdesc_len(sizeof(struct fcnvme_lsdesc_rqst)))
 859                 fcret = VERR_LSDESC_RQST_LEN;
 860         else if (assoc_acc->hdr.rqst.w0.ls_cmd != FCNVME_LS_CREATE_ASSOCIATION)
 861                 fcret = VERR_CR_ASSOC;
 862         else if (assoc_acc->associd.desc_tag !=
 863                         cpu_to_be32(FCNVME_LSDESC_ASSOC_ID))
 864                 fcret = VERR_ASSOC_ID;
 865         else if (assoc_acc->associd.desc_len !=
 866                         fcnvme_lsdesc_len(
 867                                 sizeof(struct fcnvme_lsdesc_assoc_id)))
 868                 fcret = VERR_ASSOC_ID_LEN;
 869         else if (assoc_acc->connectid.desc_tag !=
 870                         cpu_to_be32(FCNVME_LSDESC_CONN_ID))
 871                 fcret = VERR_CONN_ID;
 872         else if (assoc_acc->connectid.desc_len !=
 873                         fcnvme_lsdesc_len(sizeof(struct fcnvme_lsdesc_conn_id)))
 874                 fcret = VERR_CONN_ID_LEN;
 875
 876         if (fcret) {
 877                 ret = -EBADF;
 878                 dev_err(ctrl->dev,
 879                         "q %d connect failed: %s\n",
 880                         queue->qnum, validation_errors[fcret]);
 881         } else {
 882                 ctrl->association_id =
 883                         be64_to_cpu(assoc_acc->associd.association_id);
 884                 queue->connection_id =
 885                         be64_to_cpu(assoc_acc->connectid.connection_id);
 886                 set_bit(NVME_FC_Q_CONNECTED, &queue->flags);
 887         }
 888
 889 out_free_buffer:
 890         kfree(lsop);
 891 out_no_memory:
 892         if (ret)
 893                 dev_err(ctrl->dev,
 894                         "queue %d connect admin queue failed (%d).\n",
 895                         queue->qnum, ret);
 896         return ret;
 897 }
 898
 899 static int
 900 nvme_fc_connect_queue(struct nvme_fc_ctrl *ctrl, struct nvme_fc_queue *queue,
 901                         u16 qsize, u16 ersp_ratio)
 902 {
 903         struct nvmefc_ls_req_op *lsop;
 904         struct nvmefc_ls_req *lsreq;
 905         struct fcnvme_ls_cr_conn_rqst *conn_rqst;
 906         struct fcnvme_ls_cr_conn_acc *conn_acc;
 907         int ret, fcret = 0;
 908
 909         lsop = kzalloc((sizeof(*lsop) +
 910                          ctrl->lport->ops->lsrqst_priv_sz +
 911                          sizeof(*conn_rqst) + sizeof(*conn_acc)), GFP_KERNEL);
 912         if (!lsop) {
 913                 ret = -ENOMEM;
 914                 goto out_no_memory;
 915         }
 916         lsreq = &lsop->ls_req;
 917
 918         lsreq->private = (void *)&lsop[1];
 919         conn_rqst = (struct fcnvme_ls_cr_conn_rqst *)
 920                         (lsreq->private + ctrl->lport->ops->lsrqst_priv_sz);
 921         conn_acc = (struct fcnvme_ls_cr_conn_acc *)&conn_rqst[1];
 922
 923         conn_rqst->w0.ls_cmd = FCNVME_LS_CREATE_CONNECTION;
 924         conn_rqst->desc_list_len = cpu_to_be32(
 925                                 sizeof(struct fcnvme_lsdesc_assoc_id) +
 926                                 sizeof(struct fcnvme_lsdesc_cr_conn_cmd));
 927
 928         conn_rqst->associd.desc_tag = cpu_to_be32(FCNVME_LSDESC_ASSOC_ID);
 929         conn_rqst->associd.desc_len =
 930                         fcnvme_lsdesc_len(
 931                                 sizeof(struct fcnvme_lsdesc_assoc_id));
 932         conn_rqst->associd.association_id = cpu_to_be64(ctrl->association_id);
 933         conn_rqst->connect_cmd.desc_tag =
 934                         cpu_to_be32(FCNVME_LSDESC_CREATE_CONN_CMD);
 935         conn_rqst->connect_cmd.desc_len =
 936                         fcnvme_lsdesc_len(
 937                                 sizeof(struct fcnvme_lsdesc_cr_conn_cmd));
 938         conn_rqst->connect_cmd.ersp_ratio = cpu_to_be16(ersp_ratio);
 939         conn_rqst->connect_cmd.qid  = cpu_to_be16(queue->qnum);
 940         conn_rqst->connect_cmd.sqsize = cpu_to_be16(qsize);
 941
 942         lsop->queue = queue;
 943         lsreq->rqstaddr = conn_rqst;
 944         lsreq->rqstlen = sizeof(*conn_rqst);
 945         lsreq->rspaddr = conn_acc;
 946         lsreq->rsplen = sizeof(*conn_acc);
 947         lsreq->timeout = NVME_FC_CONNECT_TIMEOUT_SEC;
 948
 949         ret = nvme_fc_send_ls_req(ctrl, lsop);
 950         if (ret)
 951                 goto out_free_buffer;
 952
 953         /* process connect LS completion */
 954
 955         /* validate the ACC response */
 956         if (conn_acc->hdr.w0.ls_cmd != FCNVME_LS_ACC)
 957                 fcret = VERR_LSACC;
 958         if (conn_acc->hdr.desc_list_len !=
 959                         fcnvme_lsdesc_len(sizeof(struct fcnvme_ls_cr_conn_acc)))
 960                 fcret = VERR_CR_CONN_ACC_LEN;
 961         if (conn_acc->hdr.rqst.desc_tag != cpu_to_be32(FCNVME_LSDESC_RQST))
 962                 fcret = VERR_LSDESC_RQST;
 963         else if (conn_acc->hdr.rqst.desc_len !=
 964                         fcnvme_lsdesc_len(sizeof(struct fcnvme_lsdesc_rqst)))
 965                 fcret = VERR_LSDESC_RQST_LEN;
 966         else if (conn_acc->hdr.rqst.w0.ls_cmd != FCNVME_LS_CREATE_CONNECTION)
 967                 fcret = VERR_CR_CONN;
 968         else if (conn_acc->connectid.desc_tag !=
 969                         cpu_to_be32(FCNVME_LSDESC_CONN_ID))
 970                 fcret = VERR_CONN_ID;
 971         else if (conn_acc->connectid.desc_len !=
 972                         fcnvme_lsdesc_len(sizeof(struct fcnvme_lsdesc_conn_id)))
 973                 fcret = VERR_CONN_ID_LEN;
 974
 975         if (fcret) {
 976                 ret = -EBADF;
 977                 dev_err(ctrl->dev,
 978                         "q %d connect failed: %s\n",
 979                         queue->qnum, validation_errors[fcret]);
 980         } else {
 981                 queue->connection_id =
 982                         be64_to_cpu(conn_acc->connectid.connection_id);
 983                 set_bit(NVME_FC_Q_CONNECTED, &queue->flags);
 984         }
 985
 986 out_free_buffer:
 987         kfree(lsop);
 988 out_no_memory:
 989         if (ret)
 990                 dev_err(ctrl->dev,
 991                         "queue %d connect command failed (%d).\n",
 992                         queue->qnum, ret);
 993         return ret;
 994 }
 995
 996 static void
 997 nvme_fc_disconnect_assoc_done(struct nvmefc_ls_req *lsreq, int status)
 998 {
 999         struct nvmefc_ls_req_op *lsop = ls_req_to_lsop(lsreq);
1000         struct nvme_fc_ctrl *ctrl = lsop->ctrl;
1001
1002         __nvme_fc_finish_ls_req(ctrl, lsop);
1003
1004         if (status)
1005                 dev_err(ctrl->dev,
1006                         "disconnect assoc ls request command failed (%d).\n",
1007                         status);
1008
1009         /* fc-nvme iniator doesn't care about success or failure of cmd */
1010
1011         kfree(lsop);
1012 }
1013
1014 /*
1015  * This routine sends a FC-NVME LS to disconnect (aka terminate)
1016  * the FC-NVME Association.  Terminating the association also
1017  * terminates the FC-NVME connections (per queue, both admin and io
1018  * queues) that are part of the association. E.g. things are torn
1019  * down, and the related FC-NVME Association ID and Connection IDs
1020  * become invalid.
1021  *
1022  * The behavior of the fc-nvme initiator is such that it's
1023  * understanding of the association and connections will implicitly
1024  * be torn down. The action is implicit as it may be due to a loss of
1025  * connectivity with the fc-nvme target, so you may never get a
1026  * response even if you tried.  As such, the action of this routine
1027  * is to asynchronously send the LS, ignore any results of the LS, and
1028  * continue on with terminating the association. If the fc-nvme target
1029  * is present and receives the LS, it too can tear down.
1030  */
1031 static void
1032 nvme_fc_xmt_disconnect_assoc(struct nvme_fc_ctrl *ctrl)
1033 {
1034         struct fcnvme_ls_disconnect_rqst *discon_rqst;
1035         struct fcnvme_ls_disconnect_acc *discon_acc;
1036         struct nvmefc_ls_req_op *lsop;
1037         struct nvmefc_ls_req *lsreq;
1038
1039         lsop = kzalloc((sizeof(*lsop) +
1040                          ctrl->lport->ops->lsrqst_priv_sz +
1041                          sizeof(*discon_rqst) + sizeof(*discon_acc)),
1042                         GFP_KERNEL);
1043         if (!lsop)
1044                 /* couldn't sent it... too bad */
1045                 return;
1046
1047         lsreq = &lsop->ls_req;
1048
1049         lsreq->private = (void *)&lsop[1];
1050         discon_rqst = (struct fcnvme_ls_disconnect_rqst *)
1051                         (lsreq->private + ctrl->lport->ops->lsrqst_priv_sz);
1052         discon_acc = (struct fcnvme_ls_disconnect_acc *)&discon_rqst[1];
1053
1054         discon_rqst->w0.ls_cmd = FCNVME_LS_DISCONNECT;
1055         discon_rqst->desc_list_len = cpu_to_be32(
1056                                 sizeof(struct fcnvme_lsdesc_assoc_id) +
1057                                 sizeof(struct fcnvme_lsdesc_disconn_cmd));
1058
1059         discon_rqst->associd.desc_tag = cpu_to_be32(FCNVME_LSDESC_ASSOC_ID);
1060         discon_rqst->associd.desc_len =
1061                         fcnvme_lsdesc_len(
1062                                 sizeof(struct fcnvme_lsdesc_assoc_id));
1063
1064         discon_rqst->associd.association_id = cpu_to_be64(ctrl->association_id);
1065
1066         discon_rqst->discon_cmd.desc_tag = cpu_to_be32(
1067                                                 FCNVME_LSDESC_DISCONN_CMD);
1068         discon_rqst->discon_cmd.desc_len =
1069                         fcnvme_lsdesc_len(
1070                                 sizeof(struct fcnvme_lsdesc_disconn_cmd));
1071         discon_rqst->discon_cmd.scope = FCNVME_DISCONN_ASSOCIATION;
1072         discon_rqst->discon_cmd.id = cpu_to_be64(ctrl->association_id);
1073
1074         lsreq->rqstaddr = discon_rqst;
1075         lsreq->rqstlen = sizeof(*discon_rqst);
1076         lsreq->rspaddr = discon_acc;
1077         lsreq->rsplen = sizeof(*discon_acc);
1078         lsreq->timeout = NVME_FC_CONNECT_TIMEOUT_SEC;
1079
1080         nvme_fc_send_ls_req_async(ctrl, lsop, nvme_fc_disconnect_assoc_done);
1081
1082         /* only meaningful part to terminating the association */
1083         ctrl->association_id = 0;
1084 }
1085
1086
1087 /* *********************** NVME Ctrl Routines **************************** */
1088
1089
1090 static int
1091 nvme_fc_reinit_request(void *data, struct request *rq)
1092 {
1093         struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
1094         struct nvme_fc_cmd_iu *cmdiu = &op->cmd_iu;
1095
1096         memset(cmdiu, 0, sizeof(*cmdiu));
1097         cmdiu->scsi_id = NVME_CMD_SCSI_ID;
1098         cmdiu->fc_id = NVME_CMD_FC_ID;
1099         cmdiu->iu_len = cpu_to_be16(sizeof(*cmdiu) / sizeof(u32));
1100         memset(&op->rsp_iu, 0, sizeof(op->rsp_iu));
1101
1102         return 0;
1103 }
1104
1105 static void
1106 __nvme_fc_exit_request(struct nvme_fc_ctrl *ctrl,
1107                 struct nvme_fc_fcp_op *op)
1108 {
1109         fc_dma_unmap_single(ctrl->lport->dev, op->fcp_req.rspdma,
1110                                 sizeof(op->rsp_iu), DMA_FROM_DEVICE);
1111         fc_dma_unmap_single(ctrl->lport->dev, op->fcp_req.cmddma,
1112                                 sizeof(op->cmd_iu), DMA_TO_DEVICE);
1113
1114         atomic_set(&op->state, FCPOP_STATE_UNINIT);
1115 }
1116
1117 static void
1118 nvme_fc_exit_request(void *data, struct request *rq,
1119                                 unsigned int hctx_idx, unsigned int rq_idx)
1120 {
1121         struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
1122
1123         return __nvme_fc_exit_request(data, op);
1124 }
1125
1126 static void
1127 nvme_fc_exit_aen_ops(struct nvme_fc_ctrl *ctrl)
1128 {
1129         struct nvme_fc_fcp_op *aen_op = ctrl->aen_ops;
1130         int i;
1131
1132         for (i = 0; i < NVME_FC_NR_AEN_COMMANDS; i++, aen_op++) {
1133                 if (atomic_read(&aen_op->state) == FCPOP_STATE_UNINIT)
1134                         continue;
1135                 __nvme_fc_exit_request(ctrl, aen_op);
1136                 nvme_fc_ctrl_put(ctrl);
1137         }
1138 }
1139
1140 void
1141 nvme_fc_fcpio_done(struct nvmefc_fcp_req *req)
1142 {
1143         struct nvme_fc_fcp_op *op = fcp_req_to_fcp_op(req);
1144         struct request *rq = op->rq;
1145         struct nvmefc_fcp_req *freq = &op->fcp_req;
1146         struct nvme_fc_ctrl *ctrl = op->ctrl;
1147         struct nvme_fc_queue *queue = op->queue;
1148         struct nvme_completion *cqe = &op->rsp_iu.cqe;
1149         u16 status;
1150
1151         /*
1152          * WARNING:
1153          * The current linux implementation of a nvme controller
1154          * allocates a single tag set for all io queues and sizes
1155          * the io queues to fully hold all possible tags. Thus, the
1156          * implementation does not reference or care about the sqhd
1157          * value as it never needs to use the sqhd/sqtail pointers
1158          * for submission pacing.
1159          *
1160          * This affects the FC-NVME implementation in two ways:
1161          * 1) As the value doesn't matter, we don't need to waste
1162          *    cycles extracting it from ERSPs and stamping it in the
1163          *    cases where the transport fabricates CQEs on successful
1164          *    completions.
1165          * 2) The FC-NVME implementation requires that delivery of
1166          *    ERSP completions are to go back to the nvme layer in order
1167          *    relative to the rsn, such that the sqhd value will always
1168          *    be "in order" for the nvme layer. As the nvme layer in
1169          *    linux doesn't care about sqhd, there's no need to return
1170          *    them in order.
1171          *
1172          * Additionally:
1173          * As the core nvme layer in linux currently does not look at
1174          * every field in the cqe - in cases where the FC transport must
1175          * fabricate a CQE, the following fields will not be set as they
1176          * are not referenced:
1177          *      cqe.sqid,  cqe.sqhd,  cqe.command_id
1178          */
1179
1180         fc_dma_sync_single_for_cpu(ctrl->lport->dev, op->fcp_req.rspdma,
1181                                 sizeof(op->rsp_iu), DMA_FROM_DEVICE);
1182
1183         if (atomic_read(&op->state) == FCPOP_STATE_ABORTED)
1184                 status = NVME_SC_ABORT_REQ | NVME_SC_DNR;
1185         else
1186                 status = freq->status;
1187
1188         /*
1189          * For the linux implementation, if we have an unsuccesful
1190          * status, they blk-mq layer can typically be called with the
1191          * non-zero status and the content of the cqe isn't important.
1192          */
1193         if (status)
1194                 goto done;
1195
1196         /*
1197          * command completed successfully relative to the wire
1198          * protocol. However, validate anything received and
1199          * extract the status and result from the cqe (create it
1200          * where necessary).
1201          */
1202
1203         switch (freq->rcv_rsplen) {
1204
1205         case 0:
1206         case NVME_FC_SIZEOF_ZEROS_RSP:
1207                 /*
1208                  * No response payload or 12 bytes of payload (which
1209                  * should all be zeros) are considered successful and
1210                  * no payload in the CQE by the transport.
1211                  */
1212                 if (freq->transferred_length !=
1213                         be32_to_cpu(op->cmd_iu.data_len)) {
1214                         status = -EIO;
1215                         goto done;
1216                 }
1217                 op->nreq.result.u64 = 0;
1218                 break;
1219
1220         case sizeof(struct nvme_fc_ersp_iu):
1221                 /*
1222                  * The ERSP IU contains a full completion with CQE.
1223                  * Validate ERSP IU and look at cqe.
1224                  */
1225                 if (unlikely(be16_to_cpu(op->rsp_iu.iu_len) !=
1226                                         (freq->rcv_rsplen / 4) ||
1227                              be32_to_cpu(op->rsp_iu.xfrd_len) !=
1228                                         freq->transferred_length ||
1229                              op->rqno != le16_to_cpu(cqe->command_id))) {
1230                         status = -EIO;
1231                         goto done;
1232                 }
1233                 op->nreq.result = cqe->result;
1234                 status = le16_to_cpu(cqe->status) >> 1;
1235                 break;
1236
1237         default:
1238                 status = -EIO;
1239                 goto done;
1240         }
1241
1242 done:
1243         if (!queue->qnum && op->rqno >= AEN_CMDID_BASE) {
1244                 nvme_complete_async_event(&queue->ctrl->ctrl, status,
1245                                         &op->nreq.result);
1246                 nvme_fc_ctrl_put(ctrl);
1247                 return;
1248         }
1249
1250         blk_mq_complete_request(rq, status);
1251 }
1252
1253 static int
1254 __nvme_fc_init_request(struct nvme_fc_ctrl *ctrl,
1255                 struct nvme_fc_queue *queue, struct nvme_fc_fcp_op *op,
1256                 struct request *rq, u32 rqno)
1257 {
1258         struct nvme_fc_cmd_iu *cmdiu = &op->cmd_iu;
1259         int ret = 0;
1260
1261         memset(op, 0, sizeof(*op));
1262         op->fcp_req.cmdaddr = &op->cmd_iu;
1263         op->fcp_req.cmdlen = sizeof(op->cmd_iu);
1264         op->fcp_req.rspaddr = &op->rsp_iu;
1265         op->fcp_req.rsplen = sizeof(op->rsp_iu);
1266         op->fcp_req.done = nvme_fc_fcpio_done;
1267         op->fcp_req.first_sgl = (struct scatterlist *)&op[1];
1268         op->fcp_req.private = &op->fcp_req.first_sgl[SG_CHUNK_SIZE];
1269         op->ctrl = ctrl;
1270         op->queue = queue;
1271         op->rq = rq;
1272         op->rqno = rqno;
1273
1274         cmdiu->scsi_id = NVME_CMD_SCSI_ID;
1275         cmdiu->fc_id = NVME_CMD_FC_ID;
1276         cmdiu->iu_len = cpu_to_be16(sizeof(*cmdiu) / sizeof(u32));
1277
1278         op->fcp_req.cmddma = fc_dma_map_single(ctrl->lport->dev,
1279                                 &op->cmd_iu, sizeof(op->cmd_iu), DMA_TO_DEVICE);
1280         if (fc_dma_mapping_error(ctrl->lport->dev, op->fcp_req.cmddma)) {
1281                 dev_err(ctrl->dev,
1282                         "FCP Op failed - cmdiu dma mapping failed.\n");
1283                 ret = EFAULT;
1284                 goto out_on_error;
1285         }
1286
1287         op->fcp_req.rspdma = fc_dma_map_single(ctrl->lport->dev,
1288                                 &op->rsp_iu, sizeof(op->rsp_iu),
1289                                 DMA_FROM_DEVICE);
1290         if (fc_dma_mapping_error(ctrl->lport->dev, op->fcp_req.rspdma)) {
1291                 dev_err(ctrl->dev,
1292                         "FCP Op failed - rspiu dma mapping failed.\n");
1293                 ret = EFAULT;
1294         }
1295
1296         atomic_set(&op->state, FCPOP_STATE_IDLE);
1297 out_on_error:
1298         return ret;
1299 }
1300
1301 static int
1302 nvme_fc_init_request(void *data, struct request *rq,
1303                                 unsigned int hctx_idx, unsigned int rq_idx,
1304                                 unsigned int numa_node)
1305 {
1306         struct nvme_fc_ctrl *ctrl = data;
1307         struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
1308         struct nvme_fc_queue *queue = &ctrl->queues[hctx_idx+1];
1309
1310         return __nvme_fc_init_request(ctrl, queue, op, rq, queue->rqcnt++);
1311 }
1312
1313 static int
1314 nvme_fc_init_admin_request(void *data, struct request *rq,
1315                                 unsigned int hctx_idx, unsigned int rq_idx,
1316                                 unsigned int numa_node)
1317 {
1318         struct nvme_fc_ctrl *ctrl = data;
1319         struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
1320         struct nvme_fc_queue *queue = &ctrl->queues[0];
1321
1322         return __nvme_fc_init_request(ctrl, queue, op, rq, queue->rqcnt++);
1323 }
1324
1325 static int
1326 nvme_fc_init_aen_ops(struct nvme_fc_ctrl *ctrl)
1327 {
1328         struct nvme_fc_fcp_op *aen_op;
1329         struct nvme_fc_cmd_iu *cmdiu;
1330         struct nvme_command *sqe;
1331         int i, ret;
1332
1333         aen_op = ctrl->aen_ops;
1334         for (i = 0; i < NVME_FC_NR_AEN_COMMANDS; i++, aen_op++) {
1335                 cmdiu = &aen_op->cmd_iu;
1336                 sqe = &cmdiu->sqe;
1337                 ret = __nvme_fc_init_request(ctrl, &ctrl->queues[0],
1338                                 aen_op, (struct request *)NULL,
1339                                 (AEN_CMDID_BASE + i));
1340                 if (ret)
1341                         return ret;
1342
1343                 memset(sqe, 0, sizeof(*sqe));
1344                 sqe->common.opcode = nvme_admin_async_event;
1345                 sqe->common.command_id = AEN_CMDID_BASE + i;
1346         }
1347         return 0;
1348 }
1349
1350
1351 static inline void
1352 __nvme_fc_init_hctx(struct blk_mq_hw_ctx *hctx, struct nvme_fc_ctrl *ctrl,
1353                 unsigned int qidx)
1354 {
1355         struct nvme_fc_queue *queue = &ctrl->queues[qidx];
1356
1357         hctx->driver_data = queue;
1358         queue->hctx = hctx;
1359 }
1360
1361 static int
1362 nvme_fc_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
1363                 unsigned int hctx_idx)
1364 {
1365         struct nvme_fc_ctrl *ctrl = data;
1366
1367         __nvme_fc_init_hctx(hctx, ctrl, hctx_idx + 1);
1368
1369         return 0;
1370 }
1371
1372 static int
1373 nvme_fc_init_admin_hctx(struct blk_mq_hw_ctx *hctx, void *data,
1374                 unsigned int hctx_idx)
1375 {
1376         struct nvme_fc_ctrl *ctrl = data;
1377
1378         __nvme_fc_init_hctx(hctx, ctrl, hctx_idx);
1379
1380         return 0;
1381 }
1382
1383 static void
1384 nvme_fc_init_queue(struct nvme_fc_ctrl *ctrl, int idx, size_t queue_size)
1385 {
1386         struct nvme_fc_queue *queue;
1387
1388         queue = &ctrl->queues[idx];
1389         memset(queue, 0, sizeof(*queue));
1390         queue->ctrl = ctrl;
1391         queue->qnum = idx;
1392         atomic_set(&queue->csn, 1);
1393         queue->dev = ctrl->dev;
1394
1395         if (idx > 0)
1396                 queue->cmnd_capsule_len = ctrl->ctrl.ioccsz * 16;
1397         else
1398                 queue->cmnd_capsule_len = sizeof(struct nvme_command);
1399
1400         queue->queue_size = queue_size;
1401
1402         /*
1403          * Considered whether we should allocate buffers for all SQEs
1404          * and CQEs and dma map them - mapping their respective entries
1405          * into the request structures (kernel vm addr and dma address)
1406          * thus the driver could use the buffers/mappings directly.
1407          * It only makes sense if the LLDD would use them for its
1408          * messaging api. It's very unlikely most adapter api's would use
1409          * a native NVME sqe/cqe. More reasonable if FC-NVME IU payload
1410          * structures were used instead.
1411          */
1412 }
1413
1414 /*
1415  * This routine terminates a queue at the transport level.
1416  * The transport has already ensured that all outstanding ios on
1417  * the queue have been terminated.
1418  * The transport will send a Disconnect LS request to terminate
1419  * the queue's connection. Termination of the admin queue will also
1420  * terminate the association at the target.
1421  */
1422 static void
1423 nvme_fc_free_queue(struct nvme_fc_queue *queue)
1424 {
1425         if (!test_and_clear_bit(NVME_FC_Q_CONNECTED, &queue->flags))
1426                 return;
1427
1428         /*
1429          * Current implementation never disconnects a single queue.
1430          * It always terminates a whole association. So there is never
1431          * a disconnect(queue) LS sent to the target.
1432          */
1433
1434         queue->connection_id = 0;
1435         clear_bit(NVME_FC_Q_CONNECTED, &queue->flags);
1436 }
1437
1438 static void
1439 __nvme_fc_delete_hw_queue(struct nvme_fc_ctrl *ctrl,
1440         struct nvme_fc_queue *queue, unsigned int qidx)
1441 {
1442         if (ctrl->lport->ops->delete_queue)
1443                 ctrl->lport->ops->delete_queue(&ctrl->lport->localport, qidx,
1444                                 queue->lldd_handle);
1445         queue->lldd_handle = NULL;
1446 }
1447
1448 static void
1449 nvme_fc_destroy_admin_queue(struct nvme_fc_ctrl *ctrl)
1450 {
1451         __nvme_fc_delete_hw_queue(ctrl, &ctrl->queues[0], 0);
1452         blk_cleanup_queue(ctrl->ctrl.admin_q);
1453         blk_mq_free_tag_set(&ctrl->admin_tag_set);
1454         nvme_fc_free_queue(&ctrl->queues[0]);
1455 }
1456
1457 static void
1458 nvme_fc_free_io_queues(struct nvme_fc_ctrl *ctrl)
1459 {
1460         int i;
1461
1462         for (i = 1; i < ctrl->queue_count; i++)
1463                 nvme_fc_free_queue(&ctrl->queues[i]);
1464 }
1465
1466 static int
1467 __nvme_fc_create_hw_queue(struct nvme_fc_ctrl *ctrl,
1468         struct nvme_fc_queue *queue, unsigned int qidx, u16 qsize)
1469 {
1470         int ret = 0;
1471
1472         queue->lldd_handle = NULL;
1473         if (ctrl->lport->ops->create_queue)
1474                 ret = ctrl->lport->ops->create_queue(&ctrl->lport->localport,
1475                                 qidx, qsize, &queue->lldd_handle);
1476
1477         return ret;
1478 }
1479
1480 static void
1481 nvme_fc_delete_hw_io_queues(struct nvme_fc_ctrl *ctrl)
1482 {
1483         struct nvme_fc_queue *queue = &ctrl->queues[ctrl->queue_count - 1];
1484         int i;
1485
1486         for (i = ctrl->queue_count - 1; i >= 1; i--, queue--)
1487                 __nvme_fc_delete_hw_queue(ctrl, queue, i);
1488 }
1489
1490 static int
1491 nvme_fc_create_hw_io_queues(struct nvme_fc_ctrl *ctrl, u16 qsize)
1492 {
1493         struct nvme_fc_queue *queue = &ctrl->queues[1];
1494         int i, ret;
1495
1496         for (i = 1; i < ctrl->queue_count; i++, queue++) {
1497                 ret = __nvme_fc_create_hw_queue(ctrl, queue, i, qsize);
1498                 if (ret)
1499                         goto delete_queues;
1500         }
1501
1502         return 0;
1503
1504 delete_queues:
1505         for (; i >= 0; i--)
1506                 __nvme_fc_delete_hw_queue(ctrl, &ctrl->queues[i], i);
1507         return ret;
1508 }
1509
1510 static int
1511 nvme_fc_connect_io_queues(struct nvme_fc_ctrl *ctrl, u16 qsize)
1512 {
1513         int i, ret = 0;
1514
1515         for (i = 1; i < ctrl->queue_count; i++) {
1516                 ret = nvme_fc_connect_queue(ctrl, &ctrl->queues[i], qsize,
1517                                         (qsize / 5));
1518                 if (ret)
1519                         break;
1520                 ret = nvmf_connect_io_queue(&ctrl->ctrl, i);
1521                 if (ret)
1522                         break;
1523         }
1524
1525         return ret;
1526 }
1527
1528 static void
1529 nvme_fc_init_io_queues(struct nvme_fc_ctrl *ctrl)
1530 {
1531         int i;
1532
1533         for (i = 1; i < ctrl->queue_count; i++)
1534                 nvme_fc_init_queue(ctrl, i, ctrl->ctrl.sqsize);
1535 }
1536
1537 static void
1538 nvme_fc_ctrl_free(struct kref *ref)
1539 {
1540         struct nvme_fc_ctrl *ctrl =
1541                 container_of(ref, struct nvme_fc_ctrl, ref);
1542         unsigned long flags;
1543
1544         if (ctrl->state != FCCTRL_INIT) {
1545                 /* remove from rport list */
1546                 spin_lock_irqsave(&ctrl->rport->lock, flags);
1547                 list_del(&ctrl->ctrl_list);
1548                 spin_unlock_irqrestore(&ctrl->rport->lock, flags);
1549         }
1550
1551         put_device(ctrl->dev);
1552         nvme_fc_rport_put(ctrl->rport);
1553
1554         kfree(ctrl->queues);
1555         ida_simple_remove(&nvme_fc_ctrl_cnt, ctrl->cnum);
1556         nvmf_free_options(ctrl->ctrl.opts);
1557         kfree(ctrl);
1558 }
1559
1560 static void
1561 nvme_fc_ctrl_put(struct nvme_fc_ctrl *ctrl)
1562 {
1563         kref_put(&ctrl->ref, nvme_fc_ctrl_free);
1564 }
1565
1566 static int
1567 nvme_fc_ctrl_get(struct nvme_fc_ctrl *ctrl)
1568 {
1569         return kref_get_unless_zero(&ctrl->ref);
1570 }
1571
1572 /*
1573  * All accesses from nvme core layer done - can now free the
1574  * controller. Called after last nvme_put_ctrl() call
1575  */
1576 static void
1577 nvme_fc_free_nvme_ctrl(struct nvme_ctrl *nctrl)
1578 {
1579         struct nvme_fc_ctrl *ctrl = to_fc_ctrl(nctrl);
1580
1581         WARN_ON(nctrl != &ctrl->ctrl);
1582
1583         /*
1584          * Tear down the association, which will generate link
1585          * traffic to terminate connections
1586          */
1587
1588         if (ctrl->state != FCCTRL_INIT) {
1589                 /* send a Disconnect(association) LS to fc-nvme target */
1590                 nvme_fc_xmt_disconnect_assoc(ctrl);
1591
1592                 if (ctrl->ctrl.tagset) {
1593                         blk_cleanup_queue(ctrl->ctrl.connect_q);
1594                         blk_mq_free_tag_set(&ctrl->tag_set);
1595                         nvme_fc_delete_hw_io_queues(ctrl);
1596                         nvme_fc_free_io_queues(ctrl);
1597                 }
1598
1599                 nvme_fc_exit_aen_ops(ctrl);
1600
1601                 nvme_fc_destroy_admin_queue(ctrl);
1602         }
1603
1604         nvme_fc_ctrl_put(ctrl);
1605 }
1606
1607
1608 static int
1609 __nvme_fc_abort_op(struct nvme_fc_ctrl *ctrl, struct nvme_fc_fcp_op *op)
1610 {
1611         int state;
1612
1613         state = atomic_xchg(&op->state, FCPOP_STATE_ABORTED);
1614         if (state != FCPOP_STATE_ACTIVE) {
1615                 atomic_set(&op->state, state);
1616                 return -ECANCELED; /* fail */
1617         }
1618
1619         ctrl->lport->ops->fcp_abort(&ctrl->lport->localport,
1620                                         &ctrl->rport->remoteport,
1621                                         op->queue->lldd_handle,
1622                                         &op->fcp_req);
1623
1624         return 0;
1625 }
1626
1627 enum blk_eh_timer_return
1628 nvme_fc_timeout(struct request *rq, bool reserved)
1629 {
1630         struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
1631         struct nvme_fc_ctrl *ctrl = op->ctrl;
1632         int ret;
1633
1634         if (reserved)
1635                 return BLK_EH_RESET_TIMER;
1636
1637         ret = __nvme_fc_abort_op(ctrl, op);
1638         if (ret)
1639                 /* io wasn't active to abort consider it done */
1640                 return BLK_EH_HANDLED;
1641
1642         /*
1643          * TODO: force a controller reset
1644          *   when that happens, queues will be torn down and outstanding
1645          *   ios will be terminated, and the above abort, on a single io
1646          *   will no longer be needed.
1647          */
1648
1649         return BLK_EH_HANDLED;
1650 }
1651
1652 static int
1653 nvme_fc_map_data(struct nvme_fc_ctrl *ctrl, struct request *rq,
1654                 struct nvme_fc_fcp_op *op)
1655 {
1656         struct nvmefc_fcp_req *freq = &op->fcp_req;
1657         enum dma_data_direction dir;
1658         int ret;
1659
1660         freq->sg_cnt = 0;
1661
1662         if (!blk_rq_payload_bytes(rq))
1663                 return 0;
1664
1665         freq->sg_table.sgl = freq->first_sgl;
1666         ret = sg_alloc_table_chained(&freq->sg_table,
1667                         blk_rq_nr_phys_segments(rq), freq->sg_table.sgl);
1668         if (ret)
1669                 return -ENOMEM;
1670
1671         op->nents = blk_rq_map_sg(rq->q, rq, freq->sg_table.sgl);
1672         WARN_ON(op->nents > blk_rq_nr_phys_segments(rq));
1673         dir = (rq_data_dir(rq) == WRITE) ? DMA_TO_DEVICE : DMA_FROM_DEVICE;
1674         freq->sg_cnt = fc_dma_map_sg(ctrl->lport->dev, freq->sg_table.sgl,
1675                                 op->nents, dir);
1676         if (unlikely(freq->sg_cnt <= 0)) {
1677                 sg_free_table_chained(&freq->sg_table, true);
1678                 freq->sg_cnt = 0;
1679                 return -EFAULT;
1680         }
1681
1682         /*
1683          * TODO: blk_integrity_rq(rq)  for DIF
1684          */
1685         return 0;
1686 }
1687
1688 static void
1689 nvme_fc_unmap_data(struct nvme_fc_ctrl *ctrl, struct request *rq,
1690                 struct nvme_fc_fcp_op *op)
1691 {
1692         struct nvmefc_fcp_req *freq = &op->fcp_req;
1693
1694         if (!freq->sg_cnt)
1695                 return;
1696
1697         fc_dma_unmap_sg(ctrl->lport->dev, freq->sg_table.sgl, op->nents,
1698                                 ((rq_data_dir(rq) == WRITE) ?
1699                                         DMA_TO_DEVICE : DMA_FROM_DEVICE));
1700
1701         nvme_cleanup_cmd(rq);
1702
1703         sg_free_table_chained(&freq->sg_table, true);
1704
1705         freq->sg_cnt = 0;
1706 }
1707
1708 /*
1709  * In FC, the queue is a logical thing. At transport connect, the target
1710  * creates its "queue" and returns a handle that is to be given to the
1711  * target whenever it posts something to the corresponding SQ.  When an
1712  * SQE is sent on a SQ, FC effectively considers the SQE, or rather the
1713  * command contained within the SQE, an io, and assigns a FC exchange
1714  * to it. The SQE and the associated SQ handle are sent in the initial
1715  * CMD IU sents on the exchange. All transfers relative to the io occur
1716  * as part of the exchange.  The CQE is the last thing for the io,
1717  * which is transferred (explicitly or implicitly) with the RSP IU
1718  * sent on the exchange. After the CQE is received, the FC exchange is
1719  * terminaed and the Exchange may be used on a different io.
1720  *
1721  * The transport to LLDD api has the transport making a request for a
1722  * new fcp io request to the LLDD. The LLDD then allocates a FC exchange
1723  * resource and transfers the command. The LLDD will then process all
1724  * steps to complete the io. Upon completion, the transport done routine
1725  * is called.
1726  *
1727  * So - while the operation is outstanding to the LLDD, there is a link
1728  * level FC exchange resource that is also outstanding. This must be
1729  * considered in all cleanup operations.
1730  */
1731 static int
1732 nvme_fc_start_fcp_op(struct nvme_fc_ctrl *ctrl, struct nvme_fc_queue *queue,
1733         struct nvme_fc_fcp_op *op, u32 data_len,
1734         enum nvmefc_fcp_datadir io_dir)
1735 {
1736         struct nvme_fc_cmd_iu *cmdiu = &op->cmd_iu;
1737         struct nvme_command *sqe = &cmdiu->sqe;
1738         u32 csn;
1739         int ret;
1740
1741         if (!nvme_fc_ctrl_get(ctrl))
1742                 return BLK_MQ_RQ_QUEUE_ERROR;
1743
1744         /* format the FC-NVME CMD IU and fcp_req */
1745         cmdiu->connection_id = cpu_to_be64(queue->connection_id);
1746         csn = atomic_inc_return(&queue->csn);
1747         cmdiu->csn = cpu_to_be32(csn);
1748         cmdiu->data_len = cpu_to_be32(data_len);
1749         switch (io_dir) {
1750         case NVMEFC_FCP_WRITE:
1751                 cmdiu->flags = FCNVME_CMD_FLAGS_WRITE;
1752                 break;
1753         case NVMEFC_FCP_READ:
1754                 cmdiu->flags = FCNVME_CMD_FLAGS_READ;
1755                 break;
1756         case NVMEFC_FCP_NODATA:
1757                 cmdiu->flags = 0;
1758                 break;
1759         }
1760         op->fcp_req.payload_length = data_len;
1761         op->fcp_req.io_dir = io_dir;
1762         op->fcp_req.transferred_length = 0;
1763         op->fcp_req.rcv_rsplen = 0;
1764         op->fcp_req.status = 0;
1765         op->fcp_req.sqid = cpu_to_le16(queue->qnum);
1766
1767         /*
1768          * validate per fabric rules, set fields mandated by fabric spec
1769          * as well as those by FC-NVME spec.
1770          */
1771         WARN_ON_ONCE(sqe->common.metadata);
1772         WARN_ON_ONCE(sqe->common.dptr.prp1);
1773         WARN_ON_ONCE(sqe->common.dptr.prp2);
1774         sqe->common.flags |= NVME_CMD_SGL_METABUF;
1775
1776         /*
1777          * format SQE DPTR field per FC-NVME rules
1778          *    type=data block descr; subtype=offset;
1779          *    offset is currently 0.
1780          */
1781         sqe->rw.dptr.sgl.type = NVME_SGL_FMT_OFFSET;
1782         sqe->rw.dptr.sgl.length = cpu_to_le32(data_len);
1783         sqe->rw.dptr.sgl.addr = 0;
1784
1785         /* odd that we set the command_id - should come from nvme-fabrics */
1786         WARN_ON_ONCE(sqe->common.command_id != cpu_to_le16(op->rqno));
1787
1788         if (op->rq) {                           /* skipped on aens */
1789                 ret = nvme_fc_map_data(ctrl, op->rq, op);
1790                 if (ret < 0) {
1791                         dev_err(queue->ctrl->ctrl.device,
1792                              "Failed to map data (%d)\n", ret);
1793                         nvme_cleanup_cmd(op->rq);
1794                         nvme_fc_ctrl_put(ctrl);
1795                         return (ret == -ENOMEM || ret == -EAGAIN) ?
1796                                 BLK_MQ_RQ_QUEUE_BUSY : BLK_MQ_RQ_QUEUE_ERROR;
1797                 }
1798         }
1799
1800         fc_dma_sync_single_for_device(ctrl->lport->dev, op->fcp_req.cmddma,
1801                                   sizeof(op->cmd_iu), DMA_TO_DEVICE);
1802
1803         atomic_set(&op->state, FCPOP_STATE_ACTIVE);
1804
1805         if (op->rq)
1806                 blk_mq_start_request(op->rq);
1807
1808         ret = ctrl->lport->ops->fcp_io(&ctrl->lport->localport,
1809                                         &ctrl->rport->remoteport,
1810                                         queue->lldd_handle, &op->fcp_req);
1811
1812         if (ret) {
1813                 dev_err(ctrl->dev,
1814                         "Send nvme command failed - lldd returned %d.\n", ret);
1815
1816                 if (op->rq) {                   /* normal request */
1817                         nvme_fc_unmap_data(ctrl, op->rq, op);
1818                         nvme_cleanup_cmd(op->rq);
1819                 }
1820                 /* else - aen. no cleanup needed */
1821
1822                 nvme_fc_ctrl_put(ctrl);
1823
1824                 if (ret != -EBUSY)
1825                         return BLK_MQ_RQ_QUEUE_ERROR;
1826
1827                 if (op->rq) {
1828                         blk_mq_stop_hw_queues(op->rq->q);
1829                         blk_mq_delay_queue(queue->hctx, NVMEFC_QUEUE_DELAY);
1830                 }
1831                 return BLK_MQ_RQ_QUEUE_BUSY;
1832         }
1833
1834         return BLK_MQ_RQ_QUEUE_OK;
1835 }
1836
1837 static int
1838 nvme_fc_queue_rq(struct blk_mq_hw_ctx *hctx,
1839                         const struct blk_mq_queue_data *bd)
1840 {
1841         struct nvme_ns *ns = hctx->queue->queuedata;
1842         struct nvme_fc_queue *queue = hctx->driver_data;
1843         struct nvme_fc_ctrl *ctrl = queue->ctrl;
1844         struct request *rq = bd->rq;
1845         struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
1846         struct nvme_fc_cmd_iu *cmdiu = &op->cmd_iu;
1847         struct nvme_command *sqe = &cmdiu->sqe;
1848         enum nvmefc_fcp_datadir io_dir;
1849         u32 data_len;
1850         int ret;
1851
1852         ret = nvme_setup_cmd(ns, rq, sqe);
1853         if (ret)
1854                 return ret;
1855
1856         data_len = blk_rq_payload_bytes(rq);
1857         if (data_len)
1858                 io_dir = ((rq_data_dir(rq) == WRITE) ?
1859                                         NVMEFC_FCP_WRITE : NVMEFC_FCP_READ);
1860         else
1861                 io_dir = NVMEFC_FCP_NODATA;
1862
1863         return nvme_fc_start_fcp_op(ctrl, queue, op, data_len, io_dir);
1864 }
1865
1866 static struct blk_mq_tags *
1867 nvme_fc_tagset(struct nvme_fc_queue *queue)
1868 {
1869         if (queue->qnum == 0)
1870                 return queue->ctrl->admin_tag_set.tags[queue->qnum];
1871
1872         return queue->ctrl->tag_set.tags[queue->qnum - 1];
1873 }
1874
1875 static int
1876 nvme_fc_poll(struct blk_mq_hw_ctx *hctx, unsigned int tag)
1877
1878 {
1879         struct nvme_fc_queue *queue = hctx->driver_data;
1880         struct nvme_fc_ctrl *ctrl = queue->ctrl;
1881         struct request *req;
1882         struct nvme_fc_fcp_op *op;
1883
1884         req = blk_mq_tag_to_rq(nvme_fc_tagset(queue), tag);
1885         if (!req) {
1886                 dev_err(queue->ctrl->ctrl.device,
1887                          "tag 0x%x on QNum %#x not found\n",
1888                         tag, queue->qnum);
1889                 return 0;
1890         }
1891
1892         op = blk_mq_rq_to_pdu(req);
1893
1894         if ((atomic_read(&op->state) == FCPOP_STATE_ACTIVE) &&
1895                  (ctrl->lport->ops->poll_queue))
1896                 ctrl->lport->ops->poll_queue(&ctrl->lport->localport,
1897                                                  queue->lldd_handle);
1898
1899         return ((atomic_read(&op->state) != FCPOP_STATE_ACTIVE));
1900 }
1901
1902 static void
1903 nvme_fc_submit_async_event(struct nvme_ctrl *arg, int aer_idx)
1904 {
1905         struct nvme_fc_ctrl *ctrl = to_fc_ctrl(arg);
1906         struct nvme_fc_fcp_op *aen_op;
1907         int ret;
1908
1909         if (aer_idx > NVME_FC_NR_AEN_COMMANDS)
1910                 return;
1911
1912         aen_op = &ctrl->aen_ops[aer_idx];
1913
1914         ret = nvme_fc_start_fcp_op(ctrl, aen_op->queue, aen_op, 0,
1915                                         NVMEFC_FCP_NODATA);
1916         if (ret)
1917                 dev_err(ctrl->ctrl.device,
1918                         "failed async event work [%d]\n", aer_idx);
1919 }
1920
1921 static void
1922 nvme_fc_complete_rq(struct request *rq)
1923 {
1924         struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
1925         struct nvme_fc_ctrl *ctrl = op->ctrl;
1926         int error = 0, state;
1927
1928         state = atomic_xchg(&op->state, FCPOP_STATE_IDLE);
1929
1930         nvme_cleanup_cmd(rq);
1931
1932         nvme_fc_unmap_data(ctrl, rq, op);
1933
1934         if (unlikely(rq->errors)) {
1935                 if (nvme_req_needs_retry(rq, rq->errors)) {
1936                         nvme_requeue_req(rq);
1937                         return;
1938                 }
1939
1940                 if (blk_rq_is_passthrough(rq))
1941                         error = rq->errors;
1942                 else
1943                         error = nvme_error_status(rq->errors);
1944         }
1945
1946         nvme_fc_ctrl_put(ctrl);
1947
1948         blk_mq_end_request(rq, error);
1949 }
1950
1951 static struct blk_mq_ops nvme_fc_mq_ops = {
1952         .queue_rq       = nvme_fc_queue_rq,
1953         .complete       = nvme_fc_complete_rq,
1954         .init_request   = nvme_fc_init_request,
1955         .exit_request   = nvme_fc_exit_request,
1956         .reinit_request = nvme_fc_reinit_request,
1957         .init_hctx      = nvme_fc_init_hctx,
1958         .poll           = nvme_fc_poll,
1959         .timeout        = nvme_fc_timeout,
1960 };
1961
1962 static struct blk_mq_ops nvme_fc_admin_mq_ops = {
1963         .queue_rq       = nvme_fc_queue_rq,
1964         .complete       = nvme_fc_complete_rq,
1965         .init_request   = nvme_fc_init_admin_request,
1966         .exit_request   = nvme_fc_exit_request,
1967         .reinit_request = nvme_fc_reinit_request,
1968         .init_hctx      = nvme_fc_init_admin_hctx,
1969         .timeout        = nvme_fc_timeout,
1970 };
1971
1972 static int
1973 nvme_fc_configure_admin_queue(struct nvme_fc_ctrl *ctrl)
1974 {
1975         u32 segs;
1976         int error;
1977
1978         nvme_fc_init_queue(ctrl, 0, NVME_FC_AQ_BLKMQ_DEPTH);
1979
1980         error = nvme_fc_connect_admin_queue(ctrl, &ctrl->queues[0],
1981                                 NVME_FC_AQ_BLKMQ_DEPTH,
1982                                 (NVME_FC_AQ_BLKMQ_DEPTH / 4));
1983         if (error)
1984                 return error;
1985
1986         memset(&ctrl->admin_tag_set, 0, sizeof(ctrl->admin_tag_set));
1987         ctrl->admin_tag_set.ops = &nvme_fc_admin_mq_ops;
1988         ctrl->admin_tag_set.queue_depth = NVME_FC_AQ_BLKMQ_DEPTH;
1989         ctrl->admin_tag_set.reserved_tags = 2; /* fabric connect + Keep-Alive */
1990         ctrl->admin_tag_set.numa_node = NUMA_NO_NODE;
1991         ctrl->admin_tag_set.cmd_size = sizeof(struct nvme_fc_fcp_op) +
1992                                         (SG_CHUNK_SIZE *
1993                                                 sizeof(struct scatterlist)) +
1994                                         ctrl->lport->ops->fcprqst_priv_sz;
1995         ctrl->admin_tag_set.driver_data = ctrl;
1996         ctrl->admin_tag_set.nr_hw_queues = 1;
1997         ctrl->admin_tag_set.timeout = ADMIN_TIMEOUT;
1998
1999         error = blk_mq_alloc_tag_set(&ctrl->admin_tag_set);
2000         if (error)
2001                 goto out_free_queue;
2002
2003         ctrl->ctrl.admin_q = blk_mq_init_queue(&ctrl->admin_tag_set);
2004         if (IS_ERR(ctrl->ctrl.admin_q)) {
2005                 error = PTR_ERR(ctrl->ctrl.admin_q);
2006                 goto out_free_tagset;
2007         }
2008
2009         error = __nvme_fc_create_hw_queue(ctrl, &ctrl->queues[0], 0,
2010                                 NVME_FC_AQ_BLKMQ_DEPTH);
2011         if (error)
2012                 goto out_cleanup_queue;
2013
2014         error = nvmf_connect_admin_queue(&ctrl->ctrl);
2015         if (error)
2016                 goto out_delete_hw_queue;
2017
2018         error = nvmf_reg_read64(&ctrl->ctrl, NVME_REG_CAP, &ctrl->cap);
2019         if (error) {
2020                 dev_err(ctrl->ctrl.device,
2021                         "prop_get NVME_REG_CAP failed\n");
2022                 goto out_delete_hw_queue;
2023         }
2024
2025         ctrl->ctrl.sqsize =
2026                 min_t(int, NVME_CAP_MQES(ctrl->cap), ctrl->ctrl.sqsize);
2027
2028         error = nvme_enable_ctrl(&ctrl->ctrl, ctrl->cap);
2029         if (error)
2030                 goto out_delete_hw_queue;
2031
2032         segs = min_t(u32, NVME_FC_MAX_SEGMENTS,
2033                         ctrl->lport->ops->max_sgl_segments);
2034         ctrl->ctrl.max_hw_sectors = (segs - 1) << (PAGE_SHIFT - 9);
2035
2036         error = nvme_init_identify(&ctrl->ctrl);
2037         if (error)
2038                 goto out_delete_hw_queue;
2039
2040         nvme_start_keep_alive(&ctrl->ctrl);
2041
2042         return 0;
2043
2044 out_delete_hw_queue:
2045         __nvme_fc_delete_hw_queue(ctrl, &ctrl->queues[0], 0);
2046 out_cleanup_queue:
2047         blk_cleanup_queue(ctrl->ctrl.admin_q);
2048 out_free_tagset:
2049         blk_mq_free_tag_set(&ctrl->admin_tag_set);
2050 out_free_queue:
2051         nvme_fc_free_queue(&ctrl->queues[0]);
2052         return error;
2053 }
2054
2055 /*
2056  * This routine is used by the transport when it needs to find active
2057  * io on a queue that is to be terminated. The transport uses
2058  * blk_mq_tagset_busy_itr() to find the busy requests, which then invoke
2059  * this routine to kill them on a 1 by 1 basis.
2060  *
2061  * As FC allocates FC exchange for each io, the transport must contact
2062  * the LLDD to terminate the exchange, thus releasing the FC exchange.
2063  * After terminating the exchange the LLDD will call the transport's
2064  * normal io done path for the request, but it will have an aborted
2065  * status. The done path will return the io request back to the block
2066  * layer with an error status.
2067  */
2068 static void
2069 nvme_fc_terminate_exchange(struct request *req, void *data, bool reserved)
2070 {
2071         struct nvme_ctrl *nctrl = data;
2072         struct nvme_fc_ctrl *ctrl = to_fc_ctrl(nctrl);
2073         struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(req);
2074 int status;
2075
2076         if (!blk_mq_request_started(req))
2077                 return;
2078
2079         /* this performs an ABTS-LS on the FC exchange for the io */
2080         status = __nvme_fc_abort_op(ctrl, op);
2081         /*
2082          * if __nvme_fc_abort_op failed: io wasn't active to abort
2083          * consider it done. Assume completion path already completing
2084          * in parallel
2085          */
2086         if (status)
2087                 /* io wasn't active to abort consider it done */
2088                 /* assume completion path already completing in parallel */
2089                 return;
2090 }
2091
2092
2093 /*
2094  * This routine stops operation of the controller. Admin and IO queues
2095  * are stopped, outstanding ios on them terminated, and the nvme ctrl
2096  * is shutdown.
2097  */
2098 static void
2099 nvme_fc_shutdown_ctrl(struct nvme_fc_ctrl *ctrl)
2100 {
2101         /*
2102          * If io queues are present, stop them and terminate all outstanding
2103          * ios on them. As FC allocates FC exchange for each io, the
2104          * transport must contact the LLDD to terminate the exchange,
2105          * thus releasing the FC exchange. We use blk_mq_tagset_busy_itr()
2106          * to tell us what io's are busy and invoke a transport routine
2107          * to kill them with the LLDD.  After terminating the exchange
2108          * the LLDD will call the transport's normal io done path, but it
2109          * will have an aborted status. The done path will return the
2110          * io requests back to the block layer as part of normal completions
2111          * (but with error status).
2112          */
2113         if (ctrl->queue_count > 1) {
2114                 nvme_stop_queues(&ctrl->ctrl);
2115                 blk_mq_tagset_busy_iter(&ctrl->tag_set,
2116                                 nvme_fc_terminate_exchange, &ctrl->ctrl);
2117         }
2118
2119         if (ctrl->ctrl.state == NVME_CTRL_LIVE)
2120                 nvme_shutdown_ctrl(&ctrl->ctrl);
2121
2122         /*
2123          * now clean up the admin queue. Same thing as above.
2124          * use blk_mq_tagset_busy_itr() and the transport routine to
2125          * terminate the exchanges.
2126          */
2127         blk_mq_stop_hw_queues(ctrl->ctrl.admin_q);
2128         blk_mq_tagset_busy_iter(&ctrl->admin_tag_set,
2129                                 nvme_fc_terminate_exchange, &ctrl->ctrl);
2130 }
2131
2132 /*
2133  * Called to teardown an association.
2134  * May be called with association fully in place or partially in place.
2135  */
2136 static void
2137 __nvme_fc_remove_ctrl(struct nvme_fc_ctrl *ctrl)
2138 {
2139         nvme_stop_keep_alive(&ctrl->ctrl);
2140
2141         /* stop and terminate ios on admin and io queues */
2142         nvme_fc_shutdown_ctrl(ctrl);
2143
2144         /*
2145          * tear down the controller
2146          * This will result in the last reference on the nvme ctrl to
2147          * expire, calling the transport nvme_fc_free_nvme_ctrl() callback.
2148          * From there, the transport will tear down it's logical queues and
2149          * association.
2150          */
2151         nvme_uninit_ctrl(&ctrl->ctrl);
2152
2153         nvme_put_ctrl(&ctrl->ctrl);
2154 }
2155
2156 static void
2157 nvme_fc_del_ctrl_work(struct work_struct *work)
2158 {
2159         struct nvme_fc_ctrl *ctrl =
2160                         container_of(work, struct nvme_fc_ctrl, delete_work);
2161
2162         __nvme_fc_remove_ctrl(ctrl);
2163 }
2164
2165 static int
2166 __nvme_fc_del_ctrl(struct nvme_fc_ctrl *ctrl)
2167 {
2168         if (!nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_DELETING))
2169                 return -EBUSY;
2170
2171         if (!queue_work(nvme_fc_wq, &ctrl->delete_work))
2172                 return -EBUSY;
2173
2174         return 0;
2175 }
2176
2177 /*
2178  * Request from nvme core layer to delete the controller
2179  */
2180 static int
2181 nvme_fc_del_nvme_ctrl(struct nvme_ctrl *nctrl)
2182 {
2183         struct nvme_fc_ctrl *ctrl = to_fc_ctrl(nctrl);
2184         struct nvme_fc_rport *rport = ctrl->rport;
2185         unsigned long flags;
2186         int ret;
2187
2188         spin_lock_irqsave(&rport->lock, flags);
2189         ret = __nvme_fc_del_ctrl(ctrl);
2190         spin_unlock_irqrestore(&rport->lock, flags);
2191         if (ret)
2192                 return ret;
2193
2194         flush_work(&ctrl->delete_work);
2195
2196         return 0;
2197 }
2198
2199 static int
2200 nvme_fc_reset_nvme_ctrl(struct nvme_ctrl *nctrl)
2201 {
2202         return -EIO;
2203 }
2204
2205 static const struct nvme_ctrl_ops nvme_fc_ctrl_ops = {
2206         .name                   = "fc",
2207         .module                 = THIS_MODULE,
2208         .is_fabrics             = true,
2209         .reg_read32             = nvmf_reg_read32,
2210         .reg_read64             = nvmf_reg_read64,
2211         .reg_write32            = nvmf_reg_write32,
2212         .reset_ctrl             = nvme_fc_reset_nvme_ctrl,
2213         .free_ctrl              = nvme_fc_free_nvme_ctrl,
2214         .submit_async_event     = nvme_fc_submit_async_event,
2215         .delete_ctrl            = nvme_fc_del_nvme_ctrl,
2216         .get_subsysnqn          = nvmf_get_subsysnqn,
2217         .get_address            = nvmf_get_address,
2218 };
2219
2220 static int
2221 nvme_fc_create_io_queues(struct nvme_fc_ctrl *ctrl)
2222 {
2223         struct nvmf_ctrl_options *opts = ctrl->ctrl.opts;
2224         int ret;
2225
2226         ret = nvme_set_queue_count(&ctrl->ctrl, &opts->nr_io_queues);
2227         if (ret) {
2228                 dev_info(ctrl->ctrl.device,
2229                         "set_queue_count failed: %d\n", ret);
2230                 return ret;
2231         }
2232
2233         ctrl->queue_count = opts->nr_io_queues + 1;
2234         if (!opts->nr_io_queues)
2235                 return 0;
2236
2237         dev_info(ctrl->ctrl.device, "creating %d I/O queues.\n",
2238                         opts->nr_io_queues);
2239
2240         nvme_fc_init_io_queues(ctrl);
2241
2242         memset(&ctrl->tag_set, 0, sizeof(ctrl->tag_set));
2243         ctrl->tag_set.ops = &nvme_fc_mq_ops;
2244         ctrl->tag_set.queue_depth = ctrl->ctrl.opts->queue_size;
2245         ctrl->tag_set.reserved_tags = 1; /* fabric connect */
2246         ctrl->tag_set.numa_node = NUMA_NO_NODE;
2247         ctrl->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
2248         ctrl->tag_set.cmd_size = sizeof(struct nvme_fc_fcp_op) +
2249                                         (SG_CHUNK_SIZE *
2250                                                 sizeof(struct scatterlist)) +
2251                                         ctrl->lport->ops->fcprqst_priv_sz;
2252         ctrl->tag_set.driver_data = ctrl;
2253         ctrl->tag_set.nr_hw_queues = ctrl->queue_count - 1;
2254         ctrl->tag_set.timeout = NVME_IO_TIMEOUT;
2255
2256         ret = blk_mq_alloc_tag_set(&ctrl->tag_set);
2257         if (ret)
2258                 return ret;
2259
2260         ctrl->ctrl.tagset = &ctrl->tag_set;
2261
2262         ctrl->ctrl.connect_q = blk_mq_init_queue(&ctrl->tag_set);
2263         if (IS_ERR(ctrl->ctrl.connect_q)) {
2264                 ret = PTR_ERR(ctrl->ctrl.connect_q);
2265                 goto out_free_tag_set;
2266         }
2267
2268         ret = nvme_fc_create_hw_io_queues(ctrl, ctrl->ctrl.opts->queue_size);
2269         if (ret)
2270                 goto out_cleanup_blk_queue;
2271
2272         ret = nvme_fc_connect_io_queues(ctrl, ctrl->ctrl.opts->queue_size);
2273         if (ret)
2274                 goto out_delete_hw_queues;
2275
2276         return 0;
2277
2278 out_delete_hw_queues:
2279         nvme_fc_delete_hw_io_queues(ctrl);
2280 out_cleanup_blk_queue:
2281         nvme_stop_keep_alive(&ctrl->ctrl);
2282         blk_cleanup_queue(ctrl->ctrl.connect_q);
2283 out_free_tag_set:
2284         blk_mq_free_tag_set(&ctrl->tag_set);
2285         nvme_fc_free_io_queues(ctrl);
2286
2287         /* force put free routine to ignore io queues */
2288         ctrl->ctrl.tagset = NULL;
2289
2290         return ret;
2291 }
2292
2293
2294 static struct nvme_ctrl *
2295 __nvme_fc_create_ctrl(struct device *dev, struct nvmf_ctrl_options *opts,
2296         struct nvme_fc_lport *lport, struct nvme_fc_rport *rport)
2297 {
2298         struct nvme_fc_ctrl *ctrl;
2299         unsigned long flags;
2300         int ret, idx;
2301         bool changed;
2302
2303         ctrl = kzalloc(sizeof(*ctrl), GFP_KERNEL);
2304         if (!ctrl) {
2305                 ret = -ENOMEM;
2306                 goto out_fail;
2307         }
2308
2309         idx = ida_simple_get(&nvme_fc_ctrl_cnt, 0, 0, GFP_KERNEL);
2310         if (idx < 0) {
2311                 ret = -ENOSPC;
2312                 goto out_free_ctrl;
2313         }
2314
2315         ctrl->ctrl.opts = opts;
2316         INIT_LIST_HEAD(&ctrl->ctrl_list);
2317         INIT_LIST_HEAD(&ctrl->ls_req_list);
2318         ctrl->lport = lport;
2319         ctrl->rport = rport;
2320         ctrl->dev = lport->dev;
2321         ctrl->state = FCCTRL_INIT;
2322         ctrl->cnum = idx;
2323
2324         ret = nvme_init_ctrl(&ctrl->ctrl, dev, &nvme_fc_ctrl_ops, 0);
2325         if (ret)
2326                 goto out_free_ida;
2327
2328         get_device(ctrl->dev);
2329         kref_init(&ctrl->ref);
2330
2331         INIT_WORK(&ctrl->delete_work, nvme_fc_del_ctrl_work);
2332         spin_lock_init(&ctrl->lock);
2333
2334         /* io queue count */
2335         ctrl->queue_count = min_t(unsigned int,
2336                                 opts->nr_io_queues,
2337                                 lport->ops->max_hw_queues);
2338         opts->nr_io_queues = ctrl->queue_count; /* so opts has valid value */
2339         ctrl->queue_count++;    /* +1 for admin queue */
2340
2341         ctrl->ctrl.sqsize = opts->queue_size - 1;
2342         ctrl->ctrl.kato = opts->kato;
2343
2344         ret = -ENOMEM;
2345         ctrl->queues = kcalloc(ctrl->queue_count, sizeof(struct nvme_fc_queue),
2346                                 GFP_KERNEL);
2347         if (!ctrl->queues)
2348                 goto out_uninit_ctrl;
2349
2350         ret = nvme_fc_configure_admin_queue(ctrl);
2351         if (ret)
2352                 goto out_uninit_ctrl;
2353
2354         /* sanity checks */
2355
2356         /* FC-NVME does not have other data in the capsule */
2357         if (ctrl->ctrl.icdoff) {
2358                 dev_err(ctrl->ctrl.device, "icdoff %d is not supported!\n",
2359                                 ctrl->ctrl.icdoff);
2360                 goto out_remove_admin_queue;
2361         }
2362
2363         /* FC-NVME supports normal SGL Data Block Descriptors */
2364
2365         if (opts->queue_size > ctrl->ctrl.maxcmd) {
2366                 /* warn if maxcmd is lower than queue_size */
2367                 dev_warn(ctrl->ctrl.device,
2368                         "queue_size %zu > ctrl maxcmd %u, reducing "
2369                         "to queue_size\n",
2370                         opts->queue_size, ctrl->ctrl.maxcmd);
2371                 opts->queue_size = ctrl->ctrl.maxcmd;
2372         }
2373
2374         ret = nvme_fc_init_aen_ops(ctrl);
2375         if (ret)
2376                 goto out_exit_aen_ops;
2377
2378         if (ctrl->queue_count > 1) {
2379                 ret = nvme_fc_create_io_queues(ctrl);
2380                 if (ret)
2381                         goto out_exit_aen_ops;
2382         }
2383
2384         spin_lock_irqsave(&ctrl->lock, flags);
2385         ctrl->state = FCCTRL_ACTIVE;
2386         spin_unlock_irqrestore(&ctrl->lock, flags);
2387
2388         changed = nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_LIVE);
2389         WARN_ON_ONCE(!changed);
2390
2391         dev_info(ctrl->ctrl.device,
2392                 "NVME-FC{%d}: new ctrl: NQN \"%s\"\n",
2393                 ctrl->cnum, ctrl->ctrl.opts->subsysnqn);
2394
2395         kref_get(&ctrl->ctrl.kref);
2396
2397         spin_lock_irqsave(&rport->lock, flags);
2398         list_add_tail(&ctrl->ctrl_list, &rport->ctrl_list);
2399         spin_unlock_irqrestore(&rport->lock, flags);
2400
2401         if (opts->nr_io_queues) {
2402                 nvme_queue_scan(&ctrl->ctrl);
2403                 nvme_queue_async_events(&ctrl->ctrl);
2404         }
2405
2406         return &ctrl->ctrl;
2407
2408 out_exit_aen_ops:
2409         nvme_fc_exit_aen_ops(ctrl);
2410 out_remove_admin_queue:
2411         /* send a Disconnect(association) LS to fc-nvme target */
2412         nvme_fc_xmt_disconnect_assoc(ctrl);
2413         nvme_stop_keep_alive(&ctrl->ctrl);
2414         nvme_fc_destroy_admin_queue(ctrl);
2415 out_uninit_ctrl:
2416         nvme_uninit_ctrl(&ctrl->ctrl);
2417         nvme_put_ctrl(&ctrl->ctrl);
2418         if (ret > 0)
2419                 ret = -EIO;
2420         /* exit via here will follow ctlr ref point callbacks to free */
2421         return ERR_PTR(ret);
2422
2423 out_free_ida:
2424         ida_simple_remove(&nvme_fc_ctrl_cnt, ctrl->cnum);
2425 out_free_ctrl:
2426         kfree(ctrl);
2427 out_fail:
2428         nvme_fc_rport_put(rport);
2429         /* exit via here doesn't follow ctlr ref points */
2430         return ERR_PTR(ret);
2431 }
2432
2433 enum {
2434         FCT_TRADDR_ERR          = 0,
2435         FCT_TRADDR_WWNN         = 1 << 0,
2436         FCT_TRADDR_WWPN         = 1 << 1,
2437 };
2438
2439 struct nvmet_fc_traddr {
2440         u64     nn;
2441         u64     pn;
2442 };
2443
2444 static const match_table_t traddr_opt_tokens = {
2445         { FCT_TRADDR_WWNN,      "nn-%s"         },
2446         { FCT_TRADDR_WWPN,      "pn-%s"         },
2447         { FCT_TRADDR_ERR,       NULL            }
2448 };
2449
2450 static int
2451 nvme_fc_parse_address(struct nvmet_fc_traddr *traddr, char *buf)
2452 {
2453         substring_t args[MAX_OPT_ARGS];
2454         char *options, *o, *p;
2455         int token, ret = 0;
2456         u64 token64;
2457
2458         options = o = kstrdup(buf, GFP_KERNEL);
2459         if (!options)
2460                 return -ENOMEM;
2461
2462         while ((p = strsep(&o, ":\n")) != NULL) {
2463                 if (!*p)
2464                         continue;
2465
2466                 token = match_token(p, traddr_opt_tokens, args);
2467                 switch (token) {
2468                 case FCT_TRADDR_WWNN:
2469                         if (match_u64(args, &token64)) {
2470                                 ret = -EINVAL;
2471                                 goto out;
2472                         }
2473                         traddr->nn = token64;
2474                         break;
2475                 case FCT_TRADDR_WWPN:
2476                         if (match_u64(args, &token64)) {
2477                                 ret = -EINVAL;
2478                                 goto out;
2479                         }
2480                         traddr->pn = token64;
2481                         break;
2482                 default:
2483                         pr_warn("unknown traddr token or missing value '%s'\n",
2484                                         p);
2485                         ret = -EINVAL;
2486                         goto out;
2487                 }
2488         }
2489
2490 out:
2491         kfree(options);
2492         return ret;
2493 }
2494
2495 static struct nvme_ctrl *
2496 nvme_fc_create_ctrl(struct device *dev, struct nvmf_ctrl_options *opts)
2497 {
2498         struct nvme_fc_lport *lport;
2499         struct nvme_fc_rport *rport;
2500         struct nvmet_fc_traddr laddr = { 0L, 0L };
2501         struct nvmet_fc_traddr raddr = { 0L, 0L };
2502         unsigned long flags;
2503         int ret;
2504
2505         ret = nvme_fc_parse_address(&raddr, opts->traddr);
2506         if (ret || !raddr.nn || !raddr.pn)
2507                 return ERR_PTR(-EINVAL);
2508
2509         ret = nvme_fc_parse_address(&laddr, opts->host_traddr);
2510         if (ret || !laddr.nn || !laddr.pn)
2511                 return ERR_PTR(-EINVAL);
2512
2513         /* find the host and remote ports to connect together */
2514         spin_lock_irqsave(&nvme_fc_lock, flags);
2515         list_for_each_entry(lport, &nvme_fc_lport_list, port_list) {
2516                 if (lport->localport.node_name != laddr.nn ||
2517                     lport->localport.port_name != laddr.pn)
2518                         continue;
2519
2520                 list_for_each_entry(rport, &lport->endp_list, endp_list) {
2521                         if (rport->remoteport.node_name != raddr.nn ||
2522                             rport->remoteport.port_name != raddr.pn)
2523                                 continue;
2524
2525                         /* if fail to get reference fall through. Will error */
2526                         if (!nvme_fc_rport_get(rport))
2527                                 break;
2528
2529                         spin_unlock_irqrestore(&nvme_fc_lock, flags);
2530
2531                         return __nvme_fc_create_ctrl(dev, opts, lport, rport);
2532                 }
2533         }
2534         spin_unlock_irqrestore(&nvme_fc_lock, flags);
2535
2536         return ERR_PTR(-ENOENT);
2537 }
2538
2539
2540 static struct nvmf_transport_ops nvme_fc_transport = {
2541         .name           = "fc",
2542         .required_opts  = NVMF_OPT_TRADDR | NVMF_OPT_HOST_TRADDR,
2543         .allowed_opts   = NVMF_OPT_RECONNECT_DELAY,
2544         .create_ctrl    = nvme_fc_create_ctrl,
2545 };
2546
2547 static int __init nvme_fc_init_module(void)
2548 {
2549         nvme_fc_wq = create_workqueue("nvme_fc_wq");
2550         if (!nvme_fc_wq)
2551                 return -ENOMEM;
2552
2553         return nvmf_register_transport(&nvme_fc_transport);
2554 }
2555
2556 static void __exit nvme_fc_exit_module(void)
2557 {
2558         /* sanity check - all lports should be removed */
2559         if (!list_empty(&nvme_fc_lport_list))
2560                 pr_warn("%s: localport list not empty\n", __func__);
2561
2562         nvmf_unregister_transport(&nvme_fc_transport);
2563
2564         destroy_workqueue(nvme_fc_wq);
2565
2566         ida_destroy(&nvme_fc_local_port_cnt);
2567         ida_destroy(&nvme_fc_ctrl_cnt);
2568 }
2569
2570 module_init(nvme_fc_init_module);
2571 module_exit(nvme_fc_exit_module);
2572
2573 MODULE_LICENSE("GPL v2");