root/drivers/nvme/host/fc.c

DEFINITIONS

1. to_fc_ctrl
2. localport_to_lport
3. remoteport_to_rport
4. ls_req_to_lsop
5. fcp_req_to_fcp_op
6. nvme_fc_free_lport
7. nvme_fc_lport_put
8. nvme_fc_lport_get
9. nvme_fc_attach_to_unreg_lport
10. nvme_fc_register_localport
11. nvme_fc_unregister_localport
12. nvme_fc_signal_discovery_scan
13. nvme_fc_free_rport
14. nvme_fc_rport_put
15. nvme_fc_rport_get
16. nvme_fc_resume_controller
17. nvme_fc_attach_to_suspended_rport
18. __nvme_fc_set_dev_loss_tmo
19. nvme_fc_register_remoteport
20. nvme_fc_abort_lsops
21. nvme_fc_ctrl_connectivity_loss
22. nvme_fc_unregister_remoteport
23. nvme_fc_rescan_remoteport
24. nvme_fc_set_remoteport_devloss
25. fc_dma_map_single
26. fc_dma_mapping_error
27. fc_dma_unmap_single
28. fc_dma_sync_single_for_cpu
29. fc_dma_sync_single_for_device
30. fc_map_sg
31. fc_dma_map_sg
32. fc_dma_unmap_sg
33. __nvme_fc_finish_ls_req
34. __nvme_fc_send_ls_req
35. nvme_fc_send_ls_req_done
36. nvme_fc_send_ls_req
37. nvme_fc_send_ls_req_async
38. nvme_fc_connect_admin_queue
39. nvme_fc_connect_queue
40. nvme_fc_disconnect_assoc_done
41. nvme_fc_xmt_disconnect_assoc
42. __nvme_fc_exit_request
43. nvme_fc_exit_request
44. __nvme_fc_abort_op
45. nvme_fc_abort_aen_ops
46. __nvme_fc_fcpop_chk_teardowns
47. nvme_fc_fcpio_done
48. __nvme_fc_init_request
49. nvme_fc_init_request
50. nvme_fc_init_aen_ops
51. nvme_fc_term_aen_ops
52. __nvme_fc_init_hctx
53. nvme_fc_init_hctx
54. nvme_fc_init_admin_hctx
55. nvme_fc_init_queue
56. nvme_fc_free_queue
57. __nvme_fc_delete_hw_queue
58. nvme_fc_free_io_queues
59. __nvme_fc_create_hw_queue
60. nvme_fc_delete_hw_io_queues
61. nvme_fc_create_hw_io_queues
62. nvme_fc_connect_io_queues
63. nvme_fc_init_io_queues
64. nvme_fc_ctrl_free
65. nvme_fc_ctrl_put
66. nvme_fc_ctrl_get
67. nvme_fc_nvme_ctrl_freed
68. nvme_fc_error_recovery
69. nvme_fc_timeout
70. nvme_fc_map_data
71. nvme_fc_unmap_data
72. nvme_fc_start_fcp_op
73. nvme_fc_queue_rq
74. nvme_fc_submit_async_event
75. nvme_fc_complete_rq
76. nvme_fc_terminate_exchange
77. nvme_fc_create_io_queues
78. nvme_fc_recreate_io_queues
79. nvme_fc_rport_active_on_lport
80. nvme_fc_rport_inactive_on_lport
81. nvme_fc_ctlr_active_on_rport
82. nvme_fc_ctlr_inactive_on_rport
83. nvme_fc_create_association
84. nvme_fc_delete_association
85. nvme_fc_delete_ctrl
86. nvme_fc_reconnect_or_delete
87. __nvme_fc_terminate_io
88. nvme_fc_reset_ctrl_work
89. nvme_fc_connect_err_work
90. nvme_fc_connect_ctrl_work
91. nvme_fc_existing_controller
92. nvme_fc_init_ctrl
93. __nvme_fc_parse_u64
94. nvme_fc_parse_traddr
95. nvme_fc_create_ctrl
96. nvme_fc_nvme_discovery_store
97. nvme_fc_init_module
98. nvme_fc_delete_controllers
99. nvme_fc_cleanup_for_unload
100. nvme_fc_exit_module

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (c) 2016 Avago Technologies.  All rights reserved.
   4  */
   5 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   6 #include <linux/module.h>
   7 #include <linux/parser.h>
   8 #include <uapi/scsi/fc/fc_fs.h>
   9 #include <uapi/scsi/fc/fc_els.h>
  10 #include <linux/delay.h>
  11 #include <linux/overflow.h>
  12 
  13 #include "nvme.h"
  14 #include "fabrics.h"
  15 #include <linux/nvme-fc-driver.h>
  16 #include <linux/nvme-fc.h>
  17 #include <scsi/scsi_transport_fc.h>
  18 
  19 /* *************************** Data Structures/Defines ****************** */
  20 
  21 
  22 enum nvme_fc_queue_flags {
  23         NVME_FC_Q_CONNECTED = 0,
  24         NVME_FC_Q_LIVE,
  25 };
  26 
  27 #define NVME_FC_DEFAULT_DEV_LOSS_TMO    60      /* seconds */
  28 
  29 struct nvme_fc_queue {
  30         struct nvme_fc_ctrl     *ctrl;
  31         struct device           *dev;
  32         struct blk_mq_hw_ctx    *hctx;
  33         void                    *lldd_handle;
  34         size_t                  cmnd_capsule_len;
  35         u32                     qnum;
  36         u32                     rqcnt;
  37         u32                     seqno;
  38 
  39         u64                     connection_id;
  40         atomic_t                csn;
  41 
  42         unsigned long           flags;
  43 } __aligned(sizeof(u64));       /* alignment for other things alloc'd with */
  44 
  45 enum nvme_fcop_flags {
  46         FCOP_FLAGS_TERMIO       = (1 << 0),
  47         FCOP_FLAGS_AEN          = (1 << 1),
  48 };
  49 
  50 struct nvmefc_ls_req_op {
  51         struct nvmefc_ls_req    ls_req;
  52 
  53         struct nvme_fc_rport    *rport;
  54         struct nvme_fc_queue    *queue;
  55         struct request          *rq;
  56         u32                     flags;
  57 
  58         int                     ls_error;
  59         struct completion       ls_done;
  60         struct list_head        lsreq_list;     /* rport->ls_req_list */
  61         bool                    req_queued;
  62 };
  63 
  64 enum nvme_fcpop_state {
  65         FCPOP_STATE_UNINIT      = 0,
  66         FCPOP_STATE_IDLE        = 1,
  67         FCPOP_STATE_ACTIVE      = 2,
  68         FCPOP_STATE_ABORTED     = 3,
  69         FCPOP_STATE_COMPLETE    = 4,
  70 };
  71 
  72 struct nvme_fc_fcp_op {
  73         struct nvme_request     nreq;           /*
  74                                                  * nvme/host/core.c
  75                                                  * requires this to be
  76                                                  * the 1st element in the
  77                                                  * private structure
  78                                                  * associated with the
  79                                                  * request.
  80                                                  */
  81         struct nvmefc_fcp_req   fcp_req;
  82 
  83         struct nvme_fc_ctrl     *ctrl;
  84         struct nvme_fc_queue    *queue;
  85         struct request          *rq;
  86 
  87         atomic_t                state;
  88         u32                     flags;
  89         u32                     rqno;
  90         u32                     nents;
  91 
  92         struct nvme_fc_cmd_iu   cmd_iu;
  93         struct nvme_fc_ersp_iu  rsp_iu;
  94 };
  95 
  96 struct nvme_fcp_op_w_sgl {
  97         struct nvme_fc_fcp_op   op;
  98         struct scatterlist      sgl[SG_CHUNK_SIZE];
  99         uint8_t                 priv[0];
 100 };
 101 
 102 struct nvme_fc_lport {
 103         struct nvme_fc_local_port       localport;
 104 
 105         struct ida                      endp_cnt;
 106         struct list_head                port_list;      /* nvme_fc_port_list */
 107         struct list_head                endp_list;
 108         struct device                   *dev;   /* physical device for dma */
 109         struct nvme_fc_port_template    *ops;
 110         struct kref                     ref;
 111         atomic_t                        act_rport_cnt;
 112 } __aligned(sizeof(u64));       /* alignment for other things alloc'd with */
 113 
 114 struct nvme_fc_rport {
 115         struct nvme_fc_remote_port      remoteport;
 116 
 117         struct list_head                endp_list; /* for lport->endp_list */
 118         struct list_head                ctrl_list;
 119         struct list_head                ls_req_list;
 120         struct list_head                disc_list;
 121         struct device                   *dev;   /* physical device for dma */
 122         struct nvme_fc_lport            *lport;
 123         spinlock_t                      lock;
 124         struct kref                     ref;
 125         atomic_t                        act_ctrl_cnt;
 126         unsigned long                   dev_loss_end;
 127 } __aligned(sizeof(u64));       /* alignment for other things alloc'd with */
 128 
 129 enum nvme_fcctrl_flags {
 130         FCCTRL_TERMIO           = (1 << 0),
 131 };
 132 
 133 struct nvme_fc_ctrl {
 134         spinlock_t              lock;
 135         struct nvme_fc_queue    *queues;
 136         struct device           *dev;
 137         struct nvme_fc_lport    *lport;
 138         struct nvme_fc_rport    *rport;
 139         u32                     cnum;
 140 
 141         bool                    ioq_live;
 142         bool                    assoc_active;
 143         atomic_t                err_work_active;
 144         u64                     association_id;
 145 
 146         struct list_head        ctrl_list;      /* rport->ctrl_list */
 147 
 148         struct blk_mq_tag_set   admin_tag_set;
 149         struct blk_mq_tag_set   tag_set;
 150 
 151         struct delayed_work     connect_work;
 152         struct work_struct      err_work;
 153 
 154         struct kref             ref;
 155         u32                     flags;
 156         u32                     iocnt;
 157         wait_queue_head_t       ioabort_wait;
 158 
 159         struct nvme_fc_fcp_op   aen_ops[NVME_NR_AEN_COMMANDS];
 160 
 161         struct nvme_ctrl        ctrl;
 162 };
 163 
 164 static inline struct nvme_fc_ctrl *
 165 to_fc_ctrl(struct nvme_ctrl *ctrl)
 166 {
 167         return container_of(ctrl, struct nvme_fc_ctrl, ctrl);
 168 }
 169 
 170 static inline struct nvme_fc_lport *
 171 localport_to_lport(struct nvme_fc_local_port *portptr)
 172 {
 173         return container_of(portptr, struct nvme_fc_lport, localport);
 174 }
 175 
 176 static inline struct nvme_fc_rport *
 177 remoteport_to_rport(struct nvme_fc_remote_port *portptr)
 178 {
 179         return container_of(portptr, struct nvme_fc_rport, remoteport);
 180 }
 181 
 182 static inline struct nvmefc_ls_req_op *
 183 ls_req_to_lsop(struct nvmefc_ls_req *lsreq)
 184 {
 185         return container_of(lsreq, struct nvmefc_ls_req_op, ls_req);
 186 }
 187 
 188 static inline struct nvme_fc_fcp_op *
 189 fcp_req_to_fcp_op(struct nvmefc_fcp_req *fcpreq)
 190 {
 191         return container_of(fcpreq, struct nvme_fc_fcp_op, fcp_req);
 192 }
 193 
 194 
 195 
 196 /* *************************** Globals **************************** */
 197 
 198 
 199 static DEFINE_SPINLOCK(nvme_fc_lock);
 200 
 201 static LIST_HEAD(nvme_fc_lport_list);
 202 static DEFINE_IDA(nvme_fc_local_port_cnt);
 203 static DEFINE_IDA(nvme_fc_ctrl_cnt);
 204 
 205 static struct workqueue_struct *nvme_fc_wq;
 206 
 207 static bool nvme_fc_waiting_to_unload;
 208 static DECLARE_COMPLETION(nvme_fc_unload_proceed);
 209 
 210 /*
 211  * These items are short-term. They will eventually be moved into
 212  * a generic FC class. See comments in module init.
 213  */
 214 static struct device *fc_udev_device;
 215 
 216 
 217 /* *********************** FC-NVME Port Management ************************ */
 218 
 219 static void __nvme_fc_delete_hw_queue(struct nvme_fc_ctrl *,
 220                         struct nvme_fc_queue *, unsigned int);
 221 
 222 static void
 223 nvme_fc_free_lport(struct kref *ref)
 224 {
 225         struct nvme_fc_lport *lport =
 226                 container_of(ref, struct nvme_fc_lport, ref);
 227         unsigned long flags;
 228 
 229         WARN_ON(lport->localport.port_state != FC_OBJSTATE_DELETED);
 230         WARN_ON(!list_empty(&lport->endp_list));
 231 
 232         /* remove from transport list */
 233         spin_lock_irqsave(&nvme_fc_lock, flags);
 234         list_del(&lport->port_list);
 235         if (nvme_fc_waiting_to_unload && list_empty(&nvme_fc_lport_list))
 236                 complete(&nvme_fc_unload_proceed);
 237         spin_unlock_irqrestore(&nvme_fc_lock, flags);
 238 
 239         ida_simple_remove(&nvme_fc_local_port_cnt, lport->localport.port_num);
 240         ida_destroy(&lport->endp_cnt);
 241 
 242         put_device(lport->dev);
 243 
 244         kfree(lport);
 245 }
 246 
 247 static void
 248 nvme_fc_lport_put(struct nvme_fc_lport *lport)
 249 {
 250         kref_put(&lport->ref, nvme_fc_free_lport);
 251 }
 252 
 253 static int
 254 nvme_fc_lport_get(struct nvme_fc_lport *lport)
 255 {
 256         return kref_get_unless_zero(&lport->ref);
 257 }
 258 
 259 
 260 static struct nvme_fc_lport *
 261 nvme_fc_attach_to_unreg_lport(struct nvme_fc_port_info *pinfo,
 262                         struct nvme_fc_port_template *ops,
 263                         struct device *dev)
 264 {
 265         struct nvme_fc_lport *lport;
 266         unsigned long flags;
 267 
 268         spin_lock_irqsave(&nvme_fc_lock, flags);
 269 
 270         list_for_each_entry(lport, &nvme_fc_lport_list, port_list) {
 271                 if (lport->localport.node_name != pinfo->node_name ||
 272                     lport->localport.port_name != pinfo->port_name)
 273                         continue;
 274 
 275                 if (lport->dev != dev) {
 276                         lport = ERR_PTR(-EXDEV);
 277                         goto out_done;
 278                 }
 279 
 280                 if (lport->localport.port_state != FC_OBJSTATE_DELETED) {
 281                         lport = ERR_PTR(-EEXIST);
 282                         goto out_done;
 283                 }
 284 
 285                 if (!nvme_fc_lport_get(lport)) {
 286                         /*
 287                          * fails if ref cnt already 0. If so,
 288                          * act as if lport already deleted
 289                          */
 290                         lport = NULL;
 291                         goto out_done;
 292                 }
 293 
 294                 /* resume the lport */
 295 
 296                 lport->ops = ops;
 297                 lport->localport.port_role = pinfo->port_role;
 298                 lport->localport.port_id = pinfo->port_id;
 299                 lport->localport.port_state = FC_OBJSTATE_ONLINE;
 300 
 301                 spin_unlock_irqrestore(&nvme_fc_lock, flags);
 302 
 303                 return lport;
 304         }
 305 
 306         lport = NULL;
 307 
 308 out_done:
 309         spin_unlock_irqrestore(&nvme_fc_lock, flags);
 310 
 311         return lport;
 312 }
 313 
 314 /**
 315  * nvme_fc_register_localport - transport entry point called by an
 316  *                              LLDD to register the existence of a NVME
 317  *                              host FC port.
 318  * @pinfo:     pointer to information about the port to be registered
 319  * @template:  LLDD entrypoints and operational parameters for the port
 320  * @dev:       physical hardware device node port corresponds to. Will be
 321  *             used for DMA mappings
 322  * @portptr:   pointer to a local port pointer. Upon success, the routine
 323  *             will allocate a nvme_fc_local_port structure and place its
 324  *             address in the local port pointer. Upon failure, local port
 325  *             pointer will be set to 0.
 326  *
 327  * Returns:
 328  * a completion status. Must be 0 upon success; a negative errno
 329  * (ex: -ENXIO) upon failure.
 330  */
 331 int
 332 nvme_fc_register_localport(struct nvme_fc_port_info *pinfo,
 333                         struct nvme_fc_port_template *template,
 334                         struct device *dev,
 335                         struct nvme_fc_local_port **portptr)
 336 {
 337         struct nvme_fc_lport *newrec;
 338         unsigned long flags;
 339         int ret, idx;
 340 
 341         if (!template->localport_delete || !template->remoteport_delete ||
 342             !template->ls_req || !template->fcp_io ||
 343             !template->ls_abort || !template->fcp_abort ||
 344             !template->max_hw_queues || !template->max_sgl_segments ||
 345             !template->max_dif_sgl_segments || !template->dma_boundary) {
 346                 ret = -EINVAL;
 347                 goto out_reghost_failed;
 348         }
 349 
 350         /*
 351          * look to see if there is already a localport that had been
 352          * deregistered and in the process of waiting for all the
 353          * references to fully be removed.  If the references haven't
 354          * expired, we can simply re-enable the localport. Remoteports
 355          * and controller reconnections should resume naturally.
 356          */
 357         newrec = nvme_fc_attach_to_unreg_lport(pinfo, template, dev);
 358 
 359         /* found an lport, but something about its state is bad */
 360         if (IS_ERR(newrec)) {
 361                 ret = PTR_ERR(newrec);
 362                 goto out_reghost_failed;
 363 
 364         /* found existing lport, which was resumed */
 365         } else if (newrec) {
 366                 *portptr = &newrec->localport;
 367                 return 0;
 368         }
 369 
 370         /* nothing found - allocate a new localport struct */
 371 
 372         newrec = kmalloc((sizeof(*newrec) + template->local_priv_sz),
 373                          GFP_KERNEL);
 374         if (!newrec) {
 375                 ret = -ENOMEM;
 376                 goto out_reghost_failed;
 377         }
 378 
 379         idx = ida_simple_get(&nvme_fc_local_port_cnt, 0, 0, GFP_KERNEL);
 380         if (idx < 0) {
 381                 ret = -ENOSPC;
 382                 goto out_fail_kfree;
 383         }
 384 
 385         if (!get_device(dev) && dev) {
 386                 ret = -ENODEV;
 387                 goto out_ida_put;
 388         }
 389 
 390         INIT_LIST_HEAD(&newrec->port_list);
 391         INIT_LIST_HEAD(&newrec->endp_list);
 392         kref_init(&newrec->ref);
 393         atomic_set(&newrec->act_rport_cnt, 0);
 394         newrec->ops = template;
 395         newrec->dev = dev;
 396         ida_init(&newrec->endp_cnt);
 397         newrec->localport.private = &newrec[1];
 398         newrec->localport.node_name = pinfo->node_name;
 399         newrec->localport.port_name = pinfo->port_name;
 400         newrec->localport.port_role = pinfo->port_role;
 401         newrec->localport.port_id = pinfo->port_id;
 402         newrec->localport.port_state = FC_OBJSTATE_ONLINE;
 403         newrec->localport.port_num = idx;
 404 
 405         spin_lock_irqsave(&nvme_fc_lock, flags);
 406         list_add_tail(&newrec->port_list, &nvme_fc_lport_list);
 407         spin_unlock_irqrestore(&nvme_fc_lock, flags);
 408 
 409         if (dev)
 410                 dma_set_seg_boundary(dev, template->dma_boundary);
 411 
 412         *portptr = &newrec->localport;
 413         return 0;
 414 
 415 out_ida_put:
 416         ida_simple_remove(&nvme_fc_local_port_cnt, idx);
 417 out_fail_kfree:
 418         kfree(newrec);
 419 out_reghost_failed:
 420         *portptr = NULL;
 421 
 422         return ret;
 423 }
 424 EXPORT_SYMBOL_GPL(nvme_fc_register_localport);
 425 
 426 /**
 427  * nvme_fc_unregister_localport - transport entry point called by an
 428  *                              LLDD to deregister/remove a previously
 429  *                              registered a NVME host FC port.
 430  * @portptr: pointer to the (registered) local port that is to be deregistered.
 431  *
 432  * Returns:
 433  * a completion status. Must be 0 upon success; a negative errno
 434  * (ex: -ENXIO) upon failure.
 435  */
 436 int
 437 nvme_fc_unregister_localport(struct nvme_fc_local_port *portptr)
 438 {
 439         struct nvme_fc_lport *lport = localport_to_lport(portptr);
 440         unsigned long flags;
 441 
 442         if (!portptr)
 443                 return -EINVAL;
 444 
 445         spin_lock_irqsave(&nvme_fc_lock, flags);
 446 
 447         if (portptr->port_state != FC_OBJSTATE_ONLINE) {
 448                 spin_unlock_irqrestore(&nvme_fc_lock, flags);
 449                 return -EINVAL;
 450         }
 451         portptr->port_state = FC_OBJSTATE_DELETED;
 452 
 453         spin_unlock_irqrestore(&nvme_fc_lock, flags);
 454 
 455         if (atomic_read(&lport->act_rport_cnt) == 0)
 456                 lport->ops->localport_delete(&lport->localport);
 457 
 458         nvme_fc_lport_put(lport);
 459 
 460         return 0;
 461 }
 462 EXPORT_SYMBOL_GPL(nvme_fc_unregister_localport);
 463 
 464 /*
 465  * TRADDR strings, per FC-NVME are fixed format:
 466  *   "nn-0x<16hexdigits>:pn-0x<16hexdigits>" - 43 characters
 467  * udev event will only differ by prefix of what field is
 468  * being specified:
 469  *    "NVMEFC_HOST_TRADDR=" or "NVMEFC_TRADDR=" - 19 max characters
 470  *  19 + 43 + null_fudge = 64 characters
 471  */
 472 #define FCNVME_TRADDR_LENGTH            64
 473 
 474 static void
 475 nvme_fc_signal_discovery_scan(struct nvme_fc_lport *lport,
 476                 struct nvme_fc_rport *rport)
 477 {
 478         char hostaddr[FCNVME_TRADDR_LENGTH];    /* NVMEFC_HOST_TRADDR=...*/
 479         char tgtaddr[FCNVME_TRADDR_LENGTH];     /* NVMEFC_TRADDR=...*/
 480         char *envp[4] = { "FC_EVENT=nvmediscovery", hostaddr, tgtaddr, NULL };
 481 
 482         if (!(rport->remoteport.port_role & FC_PORT_ROLE_NVME_DISCOVERY))
 483                 return;
 484 
 485         snprintf(hostaddr, sizeof(hostaddr),
 486                 "NVMEFC_HOST_TRADDR=nn-0x%016llx:pn-0x%016llx",
 487                 lport->localport.node_name, lport->localport.port_name);
 488         snprintf(tgtaddr, sizeof(tgtaddr),
 489                 "NVMEFC_TRADDR=nn-0x%016llx:pn-0x%016llx",
 490                 rport->remoteport.node_name, rport->remoteport.port_name);
 491         kobject_uevent_env(&fc_udev_device->kobj, KOBJ_CHANGE, envp);
 492 }
 493 
 494 static void
 495 nvme_fc_free_rport(struct kref *ref)
 496 {
 497         struct nvme_fc_rport *rport =
 498                 container_of(ref, struct nvme_fc_rport, ref);
 499         struct nvme_fc_lport *lport =
 500                         localport_to_lport(rport->remoteport.localport);
 501         unsigned long flags;
 502 
 503         WARN_ON(rport->remoteport.port_state != FC_OBJSTATE_DELETED);
 504         WARN_ON(!list_empty(&rport->ctrl_list));
 505 
 506         /* remove from lport list */
 507         spin_lock_irqsave(&nvme_fc_lock, flags);
 508         list_del(&rport->endp_list);
 509         spin_unlock_irqrestore(&nvme_fc_lock, flags);
 510 
 511         WARN_ON(!list_empty(&rport->disc_list));
 512         ida_simple_remove(&lport->endp_cnt, rport->remoteport.port_num);
 513 
 514         kfree(rport);
 515 
 516         nvme_fc_lport_put(lport);
 517 }
 518 
 519 static void
 520 nvme_fc_rport_put(struct nvme_fc_rport *rport)
 521 {
 522         kref_put(&rport->ref, nvme_fc_free_rport);
 523 }
 524 
 525 static int
 526 nvme_fc_rport_get(struct nvme_fc_rport *rport)
 527 {
 528         return kref_get_unless_zero(&rport->ref);
 529 }
 530 
 531 static void
 532 nvme_fc_resume_controller(struct nvme_fc_ctrl *ctrl)
 533 {
 534         switch (ctrl->ctrl.state) {
 535         case NVME_CTRL_NEW:
 536         case NVME_CTRL_CONNECTING:
 537                 /*
 538                  * As all reconnects were suppressed, schedule a
 539                  * connect.
 540                  */
 541                 dev_info(ctrl->ctrl.device,
 542                         "NVME-FC{%d}: connectivity re-established. "
 543                         "Attempting reconnect\n", ctrl->cnum);
 544 
 545                 queue_delayed_work(nvme_wq, &ctrl->connect_work, 0);
 546                 break;
 547 
 548         case NVME_CTRL_RESETTING:
 549                 /*
 550                  * Controller is already in the process of terminating the
 551                  * association. No need to do anything further. The reconnect
 552                  * step will naturally occur after the reset completes.
 553                  */
 554                 break;
 555 
 556         default:
 557                 /* no action to take - let it delete */
 558                 break;
 559         }
 560 }
 561 
 562 static struct nvme_fc_rport *
 563 nvme_fc_attach_to_suspended_rport(struct nvme_fc_lport *lport,
 564                                 struct nvme_fc_port_info *pinfo)
 565 {
 566         struct nvme_fc_rport *rport;
 567         struct nvme_fc_ctrl *ctrl;
 568         unsigned long flags;
 569 
 570         spin_lock_irqsave(&nvme_fc_lock, flags);
 571 
 572         list_for_each_entry(rport, &lport->endp_list, endp_list) {
 573                 if (rport->remoteport.node_name != pinfo->node_name ||
 574                     rport->remoteport.port_name != pinfo->port_name)
 575                         continue;
 576 
 577                 if (!nvme_fc_rport_get(rport)) {
 578                         rport = ERR_PTR(-ENOLCK);
 579                         goto out_done;
 580                 }
 581 
 582                 spin_unlock_irqrestore(&nvme_fc_lock, flags);
 583 
 584                 spin_lock_irqsave(&rport->lock, flags);
 585 
 586                 /* has it been unregistered */
 587                 if (rport->remoteport.port_state != FC_OBJSTATE_DELETED) {
 588                         /* means lldd called us twice */
 589                         spin_unlock_irqrestore(&rport->lock, flags);
 590                         nvme_fc_rport_put(rport);
 591                         return ERR_PTR(-ESTALE);
 592                 }
 593 
 594                 rport->remoteport.port_role = pinfo->port_role;
 595                 rport->remoteport.port_id = pinfo->port_id;
 596                 rport->remoteport.port_state = FC_OBJSTATE_ONLINE;
 597                 rport->dev_loss_end = 0;
 598 
 599                 /*
 600                  * kick off a reconnect attempt on all associations to the
 601                  * remote port. A successful reconnects will resume i/o.
 602                  */
 603                 list_for_each_entry(ctrl, &rport->ctrl_list, ctrl_list)
 604                         nvme_fc_resume_controller(ctrl);
 605 
 606                 spin_unlock_irqrestore(&rport->lock, flags);
 607 
 608                 return rport;
 609         }
 610 
 611         rport = NULL;
 612 
 613 out_done:
 614         spin_unlock_irqrestore(&nvme_fc_lock, flags);
 615 
 616         return rport;
 617 }
 618 
 619 static inline void
 620 __nvme_fc_set_dev_loss_tmo(struct nvme_fc_rport *rport,
 621                         struct nvme_fc_port_info *pinfo)
 622 {
 623         if (pinfo->dev_loss_tmo)
 624                 rport->remoteport.dev_loss_tmo = pinfo->dev_loss_tmo;
 625         else
 626                 rport->remoteport.dev_loss_tmo = NVME_FC_DEFAULT_DEV_LOSS_TMO;
 627 }
 628 
 629 /**
 630  * nvme_fc_register_remoteport - transport entry point called by an
 631  *                              LLDD to register the existence of a NVME
 632  *                              subsystem FC port on its fabric.
 633  * @localport: pointer to the (registered) local port that the remote
 634  *             subsystem port is connected to.
 635  * @pinfo:     pointer to information about the port to be registered
 636  * @portptr:   pointer to a remote port pointer. Upon success, the routine
 637  *             will allocate a nvme_fc_remote_port structure and place its
 638  *             address in the remote port pointer. Upon failure, remote port
 639  *             pointer will be set to 0.
 640  *
 641  * Returns:
 642  * a completion status. Must be 0 upon success; a negative errno
 643  * (ex: -ENXIO) upon failure.
 644  */
 645 int
 646 nvme_fc_register_remoteport(struct nvme_fc_local_port *localport,
 647                                 struct nvme_fc_port_info *pinfo,
 648                                 struct nvme_fc_remote_port **portptr)
 649 {
 650         struct nvme_fc_lport *lport = localport_to_lport(localport);
 651         struct nvme_fc_rport *newrec;
 652         unsigned long flags;
 653         int ret, idx;
 654 
 655         if (!nvme_fc_lport_get(lport)) {
 656                 ret = -ESHUTDOWN;
 657                 goto out_reghost_failed;
 658         }
 659 
 660         /*
 661          * look to see if there is already a remoteport that is waiting
 662          * for a reconnect (within dev_loss_tmo) with the same WWN's.
 663          * If so, transition to it and reconnect.
 664          */
 665         newrec = nvme_fc_attach_to_suspended_rport(lport, pinfo);
 666 
 667         /* found an rport, but something about its state is bad */
 668         if (IS_ERR(newrec)) {
 669                 ret = PTR_ERR(newrec);
 670                 goto out_lport_put;
 671 
 672         /* found existing rport, which was resumed */
 673         } else if (newrec) {
 674                 nvme_fc_lport_put(lport);
 675                 __nvme_fc_set_dev_loss_tmo(newrec, pinfo);
 676                 nvme_fc_signal_discovery_scan(lport, newrec);
 677                 *portptr = &newrec->remoteport;
 678                 return 0;
 679         }
 680 
 681         /* nothing found - allocate a new remoteport struct */
 682 
 683         newrec = kmalloc((sizeof(*newrec) + lport->ops->remote_priv_sz),
 684                          GFP_KERNEL);
 685         if (!newrec) {
 686                 ret = -ENOMEM;
 687                 goto out_lport_put;
 688         }
 689 
 690         idx = ida_simple_get(&lport->endp_cnt, 0, 0, GFP_KERNEL);
 691         if (idx < 0) {
 692                 ret = -ENOSPC;
 693                 goto out_kfree_rport;
 694         }
 695 
 696         INIT_LIST_HEAD(&newrec->endp_list);
 697         INIT_LIST_HEAD(&newrec->ctrl_list);
 698         INIT_LIST_HEAD(&newrec->ls_req_list);
 699         INIT_LIST_HEAD(&newrec->disc_list);
 700         kref_init(&newrec->ref);
 701         atomic_set(&newrec->act_ctrl_cnt, 0);
 702         spin_lock_init(&newrec->lock);
 703         newrec->remoteport.localport = &lport->localport;
 704         newrec->dev = lport->dev;
 705         newrec->lport = lport;
 706         newrec->remoteport.private = &newrec[1];
 707         newrec->remoteport.port_role = pinfo->port_role;
 708         newrec->remoteport.node_name = pinfo->node_name;
 709         newrec->remoteport.port_name = pinfo->port_name;
 710         newrec->remoteport.port_id = pinfo->port_id;
 711         newrec->remoteport.port_state = FC_OBJSTATE_ONLINE;
 712         newrec->remoteport.port_num = idx;
 713         __nvme_fc_set_dev_loss_tmo(newrec, pinfo);
 714 
 715         spin_lock_irqsave(&nvme_fc_lock, flags);
 716         list_add_tail(&newrec->endp_list, &lport->endp_list);
 717         spin_unlock_irqrestore(&nvme_fc_lock, flags);
 718 
 719         nvme_fc_signal_discovery_scan(lport, newrec);
 720 
 721         *portptr = &newrec->remoteport;
 722         return 0;
 723 
 724 out_kfree_rport:
 725         kfree(newrec);
 726 out_lport_put:
 727         nvme_fc_lport_put(lport);
 728 out_reghost_failed:
 729         *portptr = NULL;
 730         return ret;
 731 }
 732 EXPORT_SYMBOL_GPL(nvme_fc_register_remoteport);
 733 
 734 static int
 735 nvme_fc_abort_lsops(struct nvme_fc_rport *rport)
 736 {
 737         struct nvmefc_ls_req_op *lsop;
 738         unsigned long flags;
 739 
 740 restart:
 741         spin_lock_irqsave(&rport->lock, flags);
 742 
 743         list_for_each_entry(lsop, &rport->ls_req_list, lsreq_list) {
 744                 if (!(lsop->flags & FCOP_FLAGS_TERMIO)) {
 745                         lsop->flags |= FCOP_FLAGS_TERMIO;
 746                         spin_unlock_irqrestore(&rport->lock, flags);
 747                         rport->lport->ops->ls_abort(&rport->lport->localport,
 748                                                 &rport->remoteport,
 749                                                 &lsop->ls_req);
 750                         goto restart;
 751                 }
 752         }
 753         spin_unlock_irqrestore(&rport->lock, flags);
 754 
 755         return 0;
 756 }
 757 
 758 static void
 759 nvme_fc_ctrl_connectivity_loss(struct nvme_fc_ctrl *ctrl)
 760 {
 761         dev_info(ctrl->ctrl.device,
 762                 "NVME-FC{%d}: controller connectivity lost. Awaiting "
 763                 "Reconnect", ctrl->cnum);
 764 
 765         switch (ctrl->ctrl.state) {
 766         case NVME_CTRL_NEW:
 767         case NVME_CTRL_LIVE:
 768                 /*
 769                  * Schedule a controller reset. The reset will terminate the
 770                  * association and schedule the reconnect timer.  Reconnects
 771                  * will be attempted until either the ctlr_loss_tmo
 772                  * (max_retries * connect_delay) expires or the remoteport's
 773                  * dev_loss_tmo expires.
 774                  */
 775                 if (nvme_reset_ctrl(&ctrl->ctrl)) {
 776                         dev_warn(ctrl->ctrl.device,
 777                                 "NVME-FC{%d}: Couldn't schedule reset.\n",
 778                                 ctrl->cnum);
 779                         nvme_delete_ctrl(&ctrl->ctrl);
 780                 }
 781                 break;
 782 
 783         case NVME_CTRL_CONNECTING:
 784                 /*
 785                  * The association has already been terminated and the
 786                  * controller is attempting reconnects.  No need to do anything
 787                  * futher.  Reconnects will be attempted until either the
 788                  * ctlr_loss_tmo (max_retries * connect_delay) expires or the
 789                  * remoteport's dev_loss_tmo expires.
 790                  */
 791                 break;
 792 
 793         case NVME_CTRL_RESETTING:
 794                 /*
 795                  * Controller is already in the process of terminating the
 796                  * association.  No need to do anything further. The reconnect
 797                  * step will kick in naturally after the association is
 798                  * terminated.
 799                  */
 800                 break;
 801 
 802         case NVME_CTRL_DELETING:
 803         default:
 804                 /* no action to take - let it delete */
 805                 break;
 806         }
 807 }
 808 
 809 /**
 810  * nvme_fc_unregister_remoteport - transport entry point called by an
 811  *                              LLDD to deregister/remove a previously
 812  *                              registered a NVME subsystem FC port.
 813  * @portptr: pointer to the (registered) remote port that is to be
 814  *           deregistered.
 815  *
 816  * Returns:
 817  * a completion status. Must be 0 upon success; a negative errno
 818  * (ex: -ENXIO) upon failure.
 819  */
 820 int
 821 nvme_fc_unregister_remoteport(struct nvme_fc_remote_port *portptr)
 822 {
 823         struct nvme_fc_rport *rport = remoteport_to_rport(portptr);
 824         struct nvme_fc_ctrl *ctrl;
 825         unsigned long flags;
 826 
 827         if (!portptr)
 828                 return -EINVAL;
 829 
 830         spin_lock_irqsave(&rport->lock, flags);
 831 
 832         if (portptr->port_state != FC_OBJSTATE_ONLINE) {
 833                 spin_unlock_irqrestore(&rport->lock, flags);
 834                 return -EINVAL;
 835         }
 836         portptr->port_state = FC_OBJSTATE_DELETED;
 837 
 838         rport->dev_loss_end = jiffies + (portptr->dev_loss_tmo * HZ);
 839 
 840         list_for_each_entry(ctrl, &rport->ctrl_list, ctrl_list) {
 841                 /* if dev_loss_tmo==0, dev loss is immediate */
 842                 if (!portptr->dev_loss_tmo) {
 843                         dev_warn(ctrl->ctrl.device,
 844                                 "NVME-FC{%d}: controller connectivity lost.\n",
 845                                 ctrl->cnum);
 846                         nvme_delete_ctrl(&ctrl->ctrl);
 847                 } else
 848                         nvme_fc_ctrl_connectivity_loss(ctrl);
 849         }
 850 
 851         spin_unlock_irqrestore(&rport->lock, flags);
 852 
 853         nvme_fc_abort_lsops(rport);
 854 
 855         if (atomic_read(&rport->act_ctrl_cnt) == 0)
 856                 rport->lport->ops->remoteport_delete(portptr);
 857 
 858         /*
 859          * release the reference, which will allow, if all controllers
 860          * go away, which should only occur after dev_loss_tmo occurs,
 861          * for the rport to be torn down.
 862          */
 863         nvme_fc_rport_put(rport);
 864 
 865         return 0;
 866 }
 867 EXPORT_SYMBOL_GPL(nvme_fc_unregister_remoteport);
 868 
 869 /**
 870  * nvme_fc_rescan_remoteport - transport entry point called by an
 871  *                              LLDD to request a nvme device rescan.
 872  * @remoteport: pointer to the (registered) remote port that is to be
 873  *              rescanned.
 874  *
 875  * Returns: N/A
 876  */
 877 void
 878 nvme_fc_rescan_remoteport(struct nvme_fc_remote_port *remoteport)
 879 {
 880         struct nvme_fc_rport *rport = remoteport_to_rport(remoteport);
 881 
 882         nvme_fc_signal_discovery_scan(rport->lport, rport);
 883 }
 884 EXPORT_SYMBOL_GPL(nvme_fc_rescan_remoteport);
 885 
 886 int
 887 nvme_fc_set_remoteport_devloss(struct nvme_fc_remote_port *portptr,
 888                         u32 dev_loss_tmo)
 889 {
 890         struct nvme_fc_rport *rport = remoteport_to_rport(portptr);
 891         unsigned long flags;
 892 
 893         spin_lock_irqsave(&rport->lock, flags);
 894 
 895         if (portptr->port_state != FC_OBJSTATE_ONLINE) {
 896                 spin_unlock_irqrestore(&rport->lock, flags);
 897                 return -EINVAL;
 898         }
 899 
 900         /* a dev_loss_tmo of 0 (immediate) is allowed to be set */
 901         rport->remoteport.dev_loss_tmo = dev_loss_tmo;
 902 
 903         spin_unlock_irqrestore(&rport->lock, flags);
 904 
 905         return 0;
 906 }
 907 EXPORT_SYMBOL_GPL(nvme_fc_set_remoteport_devloss);
 908 
 909 
 910 /* *********************** FC-NVME DMA Handling **************************** */
 911 
 912 /*
 913  * The fcloop device passes in a NULL device pointer. Real LLD's will
 914  * pass in a valid device pointer. If NULL is passed to the dma mapping
 915  * routines, depending on the platform, it may or may not succeed, and
 916  * may crash.
 917  *
 918  * As such:
 919  * Wrapper all the dma routines and check the dev pointer.
 920  *
 921  * If simple mappings (return just a dma address, we'll noop them,
 922  * returning a dma address of 0.
 923  *
 924  * On more complex mappings (dma_map_sg), a pseudo routine fills
 925  * in the scatter list, setting all dma addresses to 0.
 926  */
 927 
 928 static inline dma_addr_t
 929 fc_dma_map_single(struct device *dev, void *ptr, size_t size,
 930                 enum dma_data_direction dir)
 931 {
 932         return dev ? dma_map_single(dev, ptr, size, dir) : (dma_addr_t)0L;
 933 }
 934 
 935 static inline int
 936 fc_dma_mapping_error(struct device *dev, dma_addr_t dma_addr)
 937 {
 938         return dev ? dma_mapping_error(dev, dma_addr) : 0;
 939 }
 940 
 941 static inline void
 942 fc_dma_unmap_single(struct device *dev, dma_addr_t addr, size_t size,
 943         enum dma_data_direction dir)
 944 {
 945         if (dev)
 946                 dma_unmap_single(dev, addr, size, dir);
 947 }
 948 
 949 static inline void
 950 fc_dma_sync_single_for_cpu(struct device *dev, dma_addr_t addr, size_t size,
 951                 enum dma_data_direction dir)
 952 {
 953         if (dev)
 954                 dma_sync_single_for_cpu(dev, addr, size, dir);
 955 }
 956 
 957 static inline void
 958 fc_dma_sync_single_for_device(struct device *dev, dma_addr_t addr, size_t size,
 959                 enum dma_data_direction dir)
 960 {
 961         if (dev)
 962                 dma_sync_single_for_device(dev, addr, size, dir);
 963 }
 964 
 965 /* pseudo dma_map_sg call */
 966 static int
 967 fc_map_sg(struct scatterlist *sg, int nents)
 968 {
 969         struct scatterlist *s;
 970         int i;
 971 
 972         WARN_ON(nents == 0 || sg[0].length == 0);
 973 
 974         for_each_sg(sg, s, nents, i) {
 975                 s->dma_address = 0L;
 976 #ifdef CONFIG_NEED_SG_DMA_LENGTH
 977                 s->dma_length = s->length;
 978 #endif
 979         }
 980         return nents;
 981 }
 982 
 983 static inline int
 984 fc_dma_map_sg(struct device *dev, struct scatterlist *sg, int nents,
 985                 enum dma_data_direction dir)
 986 {
 987         return dev ? dma_map_sg(dev, sg, nents, dir) : fc_map_sg(sg, nents);
 988 }
 989 
 990 static inline void
 991 fc_dma_unmap_sg(struct device *dev, struct scatterlist *sg, int nents,
 992                 enum dma_data_direction dir)
 993 {
 994         if (dev)
 995                 dma_unmap_sg(dev, sg, nents, dir);
 996 }
 997 
 998 /* *********************** FC-NVME LS Handling **************************** */
 999 
1000 static void nvme_fc_ctrl_put(struct nvme_fc_ctrl *);
1001 static int nvme_fc_ctrl_get(struct nvme_fc_ctrl *);
1002 
1003 
1004 static void
1005 __nvme_fc_finish_ls_req(struct nvmefc_ls_req_op *lsop)
1006 {
1007         struct nvme_fc_rport *rport = lsop->rport;
1008         struct nvmefc_ls_req *lsreq = &lsop->ls_req;
1009         unsigned long flags;
1010 
1011         spin_lock_irqsave(&rport->lock, flags);
1012 
1013         if (!lsop->req_queued) {
1014                 spin_unlock_irqrestore(&rport->lock, flags);
1015                 return;
1016         }
1017 
1018         list_del(&lsop->lsreq_list);
1019 
1020         lsop->req_queued = false;
1021 
1022         spin_unlock_irqrestore(&rport->lock, flags);
1023 
1024         fc_dma_unmap_single(rport->dev, lsreq->rqstdma,
1025                                   (lsreq->rqstlen + lsreq->rsplen),
1026                                   DMA_BIDIRECTIONAL);
1027 
1028         nvme_fc_rport_put(rport);
1029 }
1030 
1031 static int
1032 __nvme_fc_send_ls_req(struct nvme_fc_rport *rport,
1033                 struct nvmefc_ls_req_op *lsop,
1034                 void (*done)(struct nvmefc_ls_req *req, int status))
1035 {
1036         struct nvmefc_ls_req *lsreq = &lsop->ls_req;
1037         unsigned long flags;
1038         int ret = 0;
1039 
1040         if (rport->remoteport.port_state != FC_OBJSTATE_ONLINE)
1041                 return -ECONNREFUSED;
1042 
1043         if (!nvme_fc_rport_get(rport))
1044                 return -ESHUTDOWN;
1045 
1046         lsreq->done = done;
1047         lsop->rport = rport;
1048         lsop->req_queued = false;
1049         INIT_LIST_HEAD(&lsop->lsreq_list);
1050         init_completion(&lsop->ls_done);
1051 
1052         lsreq->rqstdma = fc_dma_map_single(rport->dev, lsreq->rqstaddr,
1053                                   lsreq->rqstlen + lsreq->rsplen,
1054                                   DMA_BIDIRECTIONAL);
1055         if (fc_dma_mapping_error(rport->dev, lsreq->rqstdma)) {
1056                 ret = -EFAULT;
1057                 goto out_putrport;
1058         }
1059         lsreq->rspdma = lsreq->rqstdma + lsreq->rqstlen;
1060 
1061         spin_lock_irqsave(&rport->lock, flags);
1062 
1063         list_add_tail(&lsop->lsreq_list, &rport->ls_req_list);
1064 
1065         lsop->req_queued = true;
1066 
1067         spin_unlock_irqrestore(&rport->lock, flags);
1068 
1069         ret = rport->lport->ops->ls_req(&rport->lport->localport,
1070                                         &rport->remoteport, lsreq);
1071         if (ret)
1072                 goto out_unlink;
1073 
1074         return 0;
1075 
1076 out_unlink:
1077         lsop->ls_error = ret;
1078         spin_lock_irqsave(&rport->lock, flags);
1079         lsop->req_queued = false;
1080         list_del(&lsop->lsreq_list);
1081         spin_unlock_irqrestore(&rport->lock, flags);
1082         fc_dma_unmap_single(rport->dev, lsreq->rqstdma,
1083                                   (lsreq->rqstlen + lsreq->rsplen),
1084                                   DMA_BIDIRECTIONAL);
1085 out_putrport:
1086         nvme_fc_rport_put(rport);
1087 
1088         return ret;
1089 }
1090 
1091 static void
1092 nvme_fc_send_ls_req_done(struct nvmefc_ls_req *lsreq, int status)
1093 {
1094         struct nvmefc_ls_req_op *lsop = ls_req_to_lsop(lsreq);
1095 
1096         lsop->ls_error = status;
1097         complete(&lsop->ls_done);
1098 }
1099 
1100 static int
1101 nvme_fc_send_ls_req(struct nvme_fc_rport *rport, struct nvmefc_ls_req_op *lsop)
1102 {
1103         struct nvmefc_ls_req *lsreq = &lsop->ls_req;
1104         struct fcnvme_ls_rjt *rjt = lsreq->rspaddr;
1105         int ret;
1106 
1107         ret = __nvme_fc_send_ls_req(rport, lsop, nvme_fc_send_ls_req_done);
1108 
1109         if (!ret) {
1110                 /*
1111                  * No timeout/not interruptible as we need the struct
1112                  * to exist until the lldd calls us back. Thus mandate
1113                  * wait until driver calls back. lldd responsible for
1114                  * the timeout action
1115                  */
1116                 wait_for_completion(&lsop->ls_done);
1117 
1118                 __nvme_fc_finish_ls_req(lsop);
1119 
1120                 ret = lsop->ls_error;
1121         }
1122 
1123         if (ret)
1124                 return ret;
1125 
1126         /* ACC or RJT payload ? */
1127         if (rjt->w0.ls_cmd == FCNVME_LS_RJT)
1128                 return -ENXIO;
1129 
1130         return 0;
1131 }
1132 
1133 static int
1134 nvme_fc_send_ls_req_async(struct nvme_fc_rport *rport,
1135                 struct nvmefc_ls_req_op *lsop,
1136                 void (*done)(struct nvmefc_ls_req *req, int status))
1137 {
1138         /* don't wait for completion */
1139 
1140         return __nvme_fc_send_ls_req(rport, lsop, done);
1141 }
1142 
1143 /* Validation Error indexes into the string table below */
1144 enum {
1145         VERR_NO_ERROR           = 0,
1146         VERR_LSACC              = 1,
1147         VERR_LSDESC_RQST        = 2,
1148         VERR_LSDESC_RQST_LEN    = 3,
1149         VERR_ASSOC_ID           = 4,
1150         VERR_ASSOC_ID_LEN       = 5,
1151         VERR_CONN_ID            = 6,
1152         VERR_CONN_ID_LEN        = 7,
1153         VERR_CR_ASSOC           = 8,
1154         VERR_CR_ASSOC_ACC_LEN   = 9,
1155         VERR_CR_CONN            = 10,
1156         VERR_CR_CONN_ACC_LEN    = 11,
1157         VERR_DISCONN            = 12,
1158         VERR_DISCONN_ACC_LEN    = 13,
1159 };
1160 
1161 static char *validation_errors[] = {
1162         "OK",
1163         "Not LS_ACC",
1164         "Not LSDESC_RQST",
1165         "Bad LSDESC_RQST Length",
1166         "Not Association ID",
1167         "Bad Association ID Length",
1168         "Not Connection ID",
1169         "Bad Connection ID Length",
1170         "Not CR_ASSOC Rqst",
1171         "Bad CR_ASSOC ACC Length",
1172         "Not CR_CONN Rqst",
1173         "Bad CR_CONN ACC Length",
1174         "Not Disconnect Rqst",
1175         "Bad Disconnect ACC Length",
1176 };
1177 
1178 static int
1179 nvme_fc_connect_admin_queue(struct nvme_fc_ctrl *ctrl,
1180         struct nvme_fc_queue *queue, u16 qsize, u16 ersp_ratio)
1181 {
1182         struct nvmefc_ls_req_op *lsop;
1183         struct nvmefc_ls_req *lsreq;
1184         struct fcnvme_ls_cr_assoc_rqst *assoc_rqst;
1185         struct fcnvme_ls_cr_assoc_acc *assoc_acc;
1186         int ret, fcret = 0;
1187 
1188         lsop = kzalloc((sizeof(*lsop) +
1189                          ctrl->lport->ops->lsrqst_priv_sz +
1190                          sizeof(*assoc_rqst) + sizeof(*assoc_acc)), GFP_KERNEL);
1191         if (!lsop) {
1192                 ret = -ENOMEM;
1193                 goto out_no_memory;
1194         }
1195         lsreq = &lsop->ls_req;
1196 
1197         lsreq->private = (void *)&lsop[1];
1198         assoc_rqst = (struct fcnvme_ls_cr_assoc_rqst *)
1199                         (lsreq->private + ctrl->lport->ops->lsrqst_priv_sz);
1200         assoc_acc = (struct fcnvme_ls_cr_assoc_acc *)&assoc_rqst[1];
1201 
1202         assoc_rqst->w0.ls_cmd = FCNVME_LS_CREATE_ASSOCIATION;
1203         assoc_rqst->desc_list_len =
1204                         cpu_to_be32(sizeof(struct fcnvme_lsdesc_cr_assoc_cmd));
1205 
1206         assoc_rqst->assoc_cmd.desc_tag =
1207                         cpu_to_be32(FCNVME_LSDESC_CREATE_ASSOC_CMD);
1208         assoc_rqst->assoc_cmd.desc_len =
1209                         fcnvme_lsdesc_len(
1210                                 sizeof(struct fcnvme_lsdesc_cr_assoc_cmd));
1211 
1212         assoc_rqst->assoc_cmd.ersp_ratio = cpu_to_be16(ersp_ratio);
1213         assoc_rqst->assoc_cmd.sqsize = cpu_to_be16(qsize - 1);
1214         /* Linux supports only Dynamic controllers */
1215         assoc_rqst->assoc_cmd.cntlid = cpu_to_be16(0xffff);
1216         uuid_copy(&assoc_rqst->assoc_cmd.hostid, &ctrl->ctrl.opts->host->id);
1217         strncpy(assoc_rqst->assoc_cmd.hostnqn, ctrl->ctrl.opts->host->nqn,
1218                 min(FCNVME_ASSOC_HOSTNQN_LEN, NVMF_NQN_SIZE));
1219         strncpy(assoc_rqst->assoc_cmd.subnqn, ctrl->ctrl.opts->subsysnqn,
1220                 min(FCNVME_ASSOC_SUBNQN_LEN, NVMF_NQN_SIZE));
1221 
1222         lsop->queue = queue;
1223         lsreq->rqstaddr = assoc_rqst;
1224         lsreq->rqstlen = sizeof(*assoc_rqst);
1225         lsreq->rspaddr = assoc_acc;
1226         lsreq->rsplen = sizeof(*assoc_acc);
1227         lsreq->timeout = NVME_FC_CONNECT_TIMEOUT_SEC;
1228 
1229         ret = nvme_fc_send_ls_req(ctrl->rport, lsop);
1230         if (ret)
1231                 goto out_free_buffer;
1232 
1233         /* process connect LS completion */
1234 
1235         /* validate the ACC response */
1236         if (assoc_acc->hdr.w0.ls_cmd != FCNVME_LS_ACC)
1237                 fcret = VERR_LSACC;
1238         else if (assoc_acc->hdr.desc_list_len !=
1239                         fcnvme_lsdesc_len(
1240                                 sizeof(struct fcnvme_ls_cr_assoc_acc)))
1241                 fcret = VERR_CR_ASSOC_ACC_LEN;
1242         else if (assoc_acc->hdr.rqst.desc_tag !=
1243                         cpu_to_be32(FCNVME_LSDESC_RQST))
1244                 fcret = VERR_LSDESC_RQST;
1245         else if (assoc_acc->hdr.rqst.desc_len !=
1246                         fcnvme_lsdesc_len(sizeof(struct fcnvme_lsdesc_rqst)))
1247                 fcret = VERR_LSDESC_RQST_LEN;
1248         else if (assoc_acc->hdr.rqst.w0.ls_cmd != FCNVME_LS_CREATE_ASSOCIATION)
1249                 fcret = VERR_CR_ASSOC;
1250         else if (assoc_acc->associd.desc_tag !=
1251                         cpu_to_be32(FCNVME_LSDESC_ASSOC_ID))
1252                 fcret = VERR_ASSOC_ID;
1253         else if (assoc_acc->associd.desc_len !=
1254                         fcnvme_lsdesc_len(
1255                                 sizeof(struct fcnvme_lsdesc_assoc_id)))
1256                 fcret = VERR_ASSOC_ID_LEN;
1257         else if (assoc_acc->connectid.desc_tag !=
1258                         cpu_to_be32(FCNVME_LSDESC_CONN_ID))
1259                 fcret = VERR_CONN_ID;
1260         else if (assoc_acc->connectid.desc_len !=
1261                         fcnvme_lsdesc_len(sizeof(struct fcnvme_lsdesc_conn_id)))
1262                 fcret = VERR_CONN_ID_LEN;
1263 
1264         if (fcret) {
1265                 ret = -EBADF;
1266                 dev_err(ctrl->dev,
1267                         "q %d connect failed: %s\n",
1268                         queue->qnum, validation_errors[fcret]);
1269         } else {
1270                 ctrl->association_id =
1271                         be64_to_cpu(assoc_acc->associd.association_id);
1272                 queue->connection_id =
1273                         be64_to_cpu(assoc_acc->connectid.connection_id);
1274                 set_bit(NVME_FC_Q_CONNECTED, &queue->flags);
1275         }
1276 
1277 out_free_buffer:
1278         kfree(lsop);
1279 out_no_memory:
1280         if (ret)
1281                 dev_err(ctrl->dev,
1282                         "queue %d connect admin queue failed (%d).\n",
1283                         queue->qnum, ret);
1284         return ret;
1285 }
1286 
1287 static int
1288 nvme_fc_connect_queue(struct nvme_fc_ctrl *ctrl, struct nvme_fc_queue *queue,
1289                         u16 qsize, u16 ersp_ratio)
1290 {
1291         struct nvmefc_ls_req_op *lsop;
1292         struct nvmefc_ls_req *lsreq;
1293         struct fcnvme_ls_cr_conn_rqst *conn_rqst;
1294         struct fcnvme_ls_cr_conn_acc *conn_acc;
1295         int ret, fcret = 0;
1296 
1297         lsop = kzalloc((sizeof(*lsop) +
1298                          ctrl->lport->ops->lsrqst_priv_sz +
1299                          sizeof(*conn_rqst) + sizeof(*conn_acc)), GFP_KERNEL);
1300         if (!lsop) {
1301                 ret = -ENOMEM;
1302                 goto out_no_memory;
1303         }
1304         lsreq = &lsop->ls_req;
1305 
1306         lsreq->private = (void *)&lsop[1];
1307         conn_rqst = (struct fcnvme_ls_cr_conn_rqst *)
1308                         (lsreq->private + ctrl->lport->ops->lsrqst_priv_sz);
1309         conn_acc = (struct fcnvme_ls_cr_conn_acc *)&conn_rqst[1];
1310 
1311         conn_rqst->w0.ls_cmd = FCNVME_LS_CREATE_CONNECTION;
1312         conn_rqst->desc_list_len = cpu_to_be32(
1313                                 sizeof(struct fcnvme_lsdesc_assoc_id) +
1314                                 sizeof(struct fcnvme_lsdesc_cr_conn_cmd));
1315 
1316         conn_rqst->associd.desc_tag = cpu_to_be32(FCNVME_LSDESC_ASSOC_ID);
1317         conn_rqst->associd.desc_len =
1318                         fcnvme_lsdesc_len(
1319                                 sizeof(struct fcnvme_lsdesc_assoc_id));
1320         conn_rqst->associd.association_id = cpu_to_be64(ctrl->association_id);
1321         conn_rqst->connect_cmd.desc_tag =
1322                         cpu_to_be32(FCNVME_LSDESC_CREATE_CONN_CMD);
1323         conn_rqst->connect_cmd.desc_len =
1324                         fcnvme_lsdesc_len(
1325                                 sizeof(struct fcnvme_lsdesc_cr_conn_cmd));
1326         conn_rqst->connect_cmd.ersp_ratio = cpu_to_be16(ersp_ratio);
1327         conn_rqst->connect_cmd.qid  = cpu_to_be16(queue->qnum);
1328         conn_rqst->connect_cmd.sqsize = cpu_to_be16(qsize - 1);
1329 
1330         lsop->queue = queue;
1331         lsreq->rqstaddr = conn_rqst;
1332         lsreq->rqstlen = sizeof(*conn_rqst);
1333         lsreq->rspaddr = conn_acc;
1334         lsreq->rsplen = sizeof(*conn_acc);
1335         lsreq->timeout = NVME_FC_CONNECT_TIMEOUT_SEC;
1336 
1337         ret = nvme_fc_send_ls_req(ctrl->rport, lsop);
1338         if (ret)
1339                 goto out_free_buffer;
1340 
1341         /* process connect LS completion */
1342 
1343         /* validate the ACC response */
1344         if (conn_acc->hdr.w0.ls_cmd != FCNVME_LS_ACC)
1345                 fcret = VERR_LSACC;
1346         else if (conn_acc->hdr.desc_list_len !=
1347                         fcnvme_lsdesc_len(sizeof(struct fcnvme_ls_cr_conn_acc)))
1348                 fcret = VERR_CR_CONN_ACC_LEN;
1349         else if (conn_acc->hdr.rqst.desc_tag != cpu_to_be32(FCNVME_LSDESC_RQST))
1350                 fcret = VERR_LSDESC_RQST;
1351         else if (conn_acc->hdr.rqst.desc_len !=
1352                         fcnvme_lsdesc_len(sizeof(struct fcnvme_lsdesc_rqst)))
1353                 fcret = VERR_LSDESC_RQST_LEN;
1354         else if (conn_acc->hdr.rqst.w0.ls_cmd != FCNVME_LS_CREATE_CONNECTION)
1355                 fcret = VERR_CR_CONN;
1356         else if (conn_acc->connectid.desc_tag !=
1357                         cpu_to_be32(FCNVME_LSDESC_CONN_ID))
1358                 fcret = VERR_CONN_ID;
1359         else if (conn_acc->connectid.desc_len !=
1360                         fcnvme_lsdesc_len(sizeof(struct fcnvme_lsdesc_conn_id)))
1361                 fcret = VERR_CONN_ID_LEN;
1362 
1363         if (fcret) {
1364                 ret = -EBADF;
1365                 dev_err(ctrl->dev,
1366                         "q %d connect failed: %s\n",
1367                         queue->qnum, validation_errors[fcret]);
1368         } else {
1369                 queue->connection_id =
1370                         be64_to_cpu(conn_acc->connectid.connection_id);
1371                 set_bit(NVME_FC_Q_CONNECTED, &queue->flags);
1372         }
1373 
1374 out_free_buffer:
1375         kfree(lsop);
1376 out_no_memory:
1377         if (ret)
1378                 dev_err(ctrl->dev,
1379                         "queue %d connect command failed (%d).\n",
1380                         queue->qnum, ret);
1381         return ret;
1382 }
1383 
1384 static void
1385 nvme_fc_disconnect_assoc_done(struct nvmefc_ls_req *lsreq, int status)
1386 {
1387         struct nvmefc_ls_req_op *lsop = ls_req_to_lsop(lsreq);
1388 
1389         __nvme_fc_finish_ls_req(lsop);
1390 
1391         /* fc-nvme initiator doesn't care about success or failure of cmd */
1392 
1393         kfree(lsop);
1394 }
1395 
1396 /*
1397  * This routine sends a FC-NVME LS to disconnect (aka terminate)
1398  * the FC-NVME Association.  Terminating the association also
1399  * terminates the FC-NVME connections (per queue, both admin and io
1400  * queues) that are part of the association. E.g. things are torn
1401  * down, and the related FC-NVME Association ID and Connection IDs
1402  * become invalid.
1403  *
1404  * The behavior of the fc-nvme initiator is such that it's
1405  * understanding of the association and connections will implicitly
1406  * be torn down. The action is implicit as it may be due to a loss of
1407  * connectivity with the fc-nvme target, so you may never get a
1408  * response even if you tried.  As such, the action of this routine
1409  * is to asynchronously send the LS, ignore any results of the LS, and
1410  * continue on with terminating the association. If the fc-nvme target
1411  * is present and receives the LS, it too can tear down.
1412  */
1413 static void
1414 nvme_fc_xmt_disconnect_assoc(struct nvme_fc_ctrl *ctrl)
1415 {
1416         struct fcnvme_ls_disconnect_rqst *discon_rqst;
1417         struct fcnvme_ls_disconnect_acc *discon_acc;
1418         struct nvmefc_ls_req_op *lsop;
1419         struct nvmefc_ls_req *lsreq;
1420         int ret;
1421 
1422         lsop = kzalloc((sizeof(*lsop) +
1423                          ctrl->lport->ops->lsrqst_priv_sz +
1424                          sizeof(*discon_rqst) + sizeof(*discon_acc)),
1425                         GFP_KERNEL);
1426         if (!lsop)
1427                 /* couldn't sent it... too bad */
1428                 return;
1429 
1430         lsreq = &lsop->ls_req;
1431 
1432         lsreq->private = (void *)&lsop[1];
1433         discon_rqst = (struct fcnvme_ls_disconnect_rqst *)
1434                         (lsreq->private + ctrl->lport->ops->lsrqst_priv_sz);
1435         discon_acc = (struct fcnvme_ls_disconnect_acc *)&discon_rqst[1];
1436 
1437         discon_rqst->w0.ls_cmd = FCNVME_LS_DISCONNECT;
1438         discon_rqst->desc_list_len = cpu_to_be32(
1439                                 sizeof(struct fcnvme_lsdesc_assoc_id) +
1440                                 sizeof(struct fcnvme_lsdesc_disconn_cmd));
1441 
1442         discon_rqst->associd.desc_tag = cpu_to_be32(FCNVME_LSDESC_ASSOC_ID);
1443         discon_rqst->associd.desc_len =
1444                         fcnvme_lsdesc_len(
1445                                 sizeof(struct fcnvme_lsdesc_assoc_id));
1446 
1447         discon_rqst->associd.association_id = cpu_to_be64(ctrl->association_id);
1448 
1449         discon_rqst->discon_cmd.desc_tag = cpu_to_be32(
1450                                                 FCNVME_LSDESC_DISCONN_CMD);
1451         discon_rqst->discon_cmd.desc_len =
1452                         fcnvme_lsdesc_len(
1453                                 sizeof(struct fcnvme_lsdesc_disconn_cmd));
1454         discon_rqst->discon_cmd.scope = FCNVME_DISCONN_ASSOCIATION;
1455         discon_rqst->discon_cmd.id = cpu_to_be64(ctrl->association_id);
1456 
1457         lsreq->rqstaddr = discon_rqst;
1458         lsreq->rqstlen = sizeof(*discon_rqst);
1459         lsreq->rspaddr = discon_acc;
1460         lsreq->rsplen = sizeof(*discon_acc);
1461         lsreq->timeout = NVME_FC_CONNECT_TIMEOUT_SEC;
1462 
1463         ret = nvme_fc_send_ls_req_async(ctrl->rport, lsop,
1464                                 nvme_fc_disconnect_assoc_done);
1465         if (ret)
1466                 kfree(lsop);
1467 
1468         /* only meaningful part to terminating the association */
1469         ctrl->association_id = 0;
1470 }
1471 
1472 
1473 /* *********************** NVME Ctrl Routines **************************** */
1474 
1475 static void nvme_fc_error_recovery(struct nvme_fc_ctrl *ctrl, char *errmsg);
1476 
1477 static void
1478 __nvme_fc_exit_request(struct nvme_fc_ctrl *ctrl,
1479                 struct nvme_fc_fcp_op *op)
1480 {
1481         fc_dma_unmap_single(ctrl->lport->dev, op->fcp_req.rspdma,
1482                                 sizeof(op->rsp_iu), DMA_FROM_DEVICE);
1483         fc_dma_unmap_single(ctrl->lport->dev, op->fcp_req.cmddma,
1484                                 sizeof(op->cmd_iu), DMA_TO_DEVICE);
1485 
1486         atomic_set(&op->state, FCPOP_STATE_UNINIT);
1487 }
1488 
1489 static void
1490 nvme_fc_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1491                 unsigned int hctx_idx)
1492 {
1493         struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
1494 
1495         return __nvme_fc_exit_request(set->driver_data, op);
1496 }
1497 
1498 static int
1499 __nvme_fc_abort_op(struct nvme_fc_ctrl *ctrl, struct nvme_fc_fcp_op *op)
1500 {
1501         unsigned long flags;
1502         int opstate;
1503 
1504         spin_lock_irqsave(&ctrl->lock, flags);
1505         opstate = atomic_xchg(&op->state, FCPOP_STATE_ABORTED);
1506         if (opstate != FCPOP_STATE_ACTIVE)
1507                 atomic_set(&op->state, opstate);
1508         else if (ctrl->flags & FCCTRL_TERMIO)
1509                 ctrl->iocnt++;
1510         spin_unlock_irqrestore(&ctrl->lock, flags);
1511 
1512         if (opstate != FCPOP_STATE_ACTIVE)
1513                 return -ECANCELED;
1514 
1515         ctrl->lport->ops->fcp_abort(&ctrl->lport->localport,
1516                                         &ctrl->rport->remoteport,
1517                                         op->queue->lldd_handle,
1518                                         &op->fcp_req);
1519 
1520         return 0;
1521 }
1522 
1523 static void
1524 nvme_fc_abort_aen_ops(struct nvme_fc_ctrl *ctrl)
1525 {
1526         struct nvme_fc_fcp_op *aen_op = ctrl->aen_ops;
1527         int i;
1528 
1529         /* ensure we've initialized the ops once */
1530         if (!(aen_op->flags & FCOP_FLAGS_AEN))
1531                 return;
1532 
1533         for (i = 0; i < NVME_NR_AEN_COMMANDS; i++, aen_op++)
1534                 __nvme_fc_abort_op(ctrl, aen_op);
1535 }
1536 
1537 static inline void
1538 __nvme_fc_fcpop_chk_teardowns(struct nvme_fc_ctrl *ctrl,
1539                 struct nvme_fc_fcp_op *op, int opstate)
1540 {
1541         unsigned long flags;
1542 
1543         if (opstate == FCPOP_STATE_ABORTED) {
1544                 spin_lock_irqsave(&ctrl->lock, flags);
1545                 if (ctrl->flags & FCCTRL_TERMIO) {
1546                         if (!--ctrl->iocnt)
1547                                 wake_up(&ctrl->ioabort_wait);
1548                 }
1549                 spin_unlock_irqrestore(&ctrl->lock, flags);
1550         }
1551 }
1552 
1553 static void
1554 nvme_fc_fcpio_done(struct nvmefc_fcp_req *req)
1555 {
1556         struct nvme_fc_fcp_op *op = fcp_req_to_fcp_op(req);
1557         struct request *rq = op->rq;
1558         struct nvmefc_fcp_req *freq = &op->fcp_req;
1559         struct nvme_fc_ctrl *ctrl = op->ctrl;
1560         struct nvme_fc_queue *queue = op->queue;
1561         struct nvme_completion *cqe = &op->rsp_iu.cqe;
1562         struct nvme_command *sqe = &op->cmd_iu.sqe;
1563         __le16 status = cpu_to_le16(NVME_SC_SUCCESS << 1);
1564         union nvme_result result;
1565         bool terminate_assoc = true;
1566         int opstate;
1567 
1568         /*
1569          * WARNING:
1570          * The current linux implementation of a nvme controller
1571          * allocates a single tag set for all io queues and sizes
1572          * the io queues to fully hold all possible tags. Thus, the
1573          * implementation does not reference or care about the sqhd
1574          * value as it never needs to use the sqhd/sqtail pointers
1575          * for submission pacing.
1576          *
1577          * This affects the FC-NVME implementation in two ways:
1578          * 1) As the value doesn't matter, we don't need to waste
1579          *    cycles extracting it from ERSPs and stamping it in the
1580          *    cases where the transport fabricates CQEs on successful
1581          *    completions.
1582          * 2) The FC-NVME implementation requires that delivery of
1583          *    ERSP completions are to go back to the nvme layer in order
1584          *    relative to the rsn, such that the sqhd value will always
1585          *    be "in order" for the nvme layer. As the nvme layer in
1586          *    linux doesn't care about sqhd, there's no need to return
1587          *    them in order.
1588          *
1589          * Additionally:
1590          * As the core nvme layer in linux currently does not look at
1591          * every field in the cqe - in cases where the FC transport must
1592          * fabricate a CQE, the following fields will not be set as they
1593          * are not referenced:
1594          *      cqe.sqid,  cqe.sqhd,  cqe.command_id
1595          *
1596          * Failure or error of an individual i/o, in a transport
1597          * detected fashion unrelated to the nvme completion status,
1598          * potentially cause the initiator and target sides to get out
1599          * of sync on SQ head/tail (aka outstanding io count allowed).
1600          * Per FC-NVME spec, failure of an individual command requires
1601          * the connection to be terminated, which in turn requires the
1602          * association to be terminated.
1603          */
1604 
1605         opstate = atomic_xchg(&op->state, FCPOP_STATE_COMPLETE);
1606 
1607         fc_dma_sync_single_for_cpu(ctrl->lport->dev, op->fcp_req.rspdma,
1608                                 sizeof(op->rsp_iu), DMA_FROM_DEVICE);
1609 
1610         if (opstate == FCPOP_STATE_ABORTED)
1611                 status = cpu_to_le16(NVME_SC_HOST_PATH_ERROR << 1);
1612         else if (freq->status) {
1613                 status = cpu_to_le16(NVME_SC_HOST_PATH_ERROR << 1);
1614                 dev_info(ctrl->ctrl.device,
1615                         "NVME-FC{%d}: io failed due to lldd error %d\n",
1616                         ctrl->cnum, freq->status);
1617         }
1618 
1619         /*
1620          * For the linux implementation, if we have an unsuccesful
1621          * status, they blk-mq layer can typically be called with the
1622          * non-zero status and the content of the cqe isn't important.
1623          */
1624         if (status)
1625                 goto done;
1626 
1627         /*
1628          * command completed successfully relative to the wire
1629          * protocol. However, validate anything received and
1630          * extract the status and result from the cqe (create it
1631          * where necessary).
1632          */
1633 
1634         switch (freq->rcv_rsplen) {
1635 
1636         case 0:
1637         case NVME_FC_SIZEOF_ZEROS_RSP:
1638                 /*
1639                  * No response payload or 12 bytes of payload (which
1640                  * should all be zeros) are considered successful and
1641                  * no payload in the CQE by the transport.
1642                  */
1643                 if (freq->transferred_length !=
1644                     be32_to_cpu(op->cmd_iu.data_len)) {
1645                         status = cpu_to_le16(NVME_SC_HOST_PATH_ERROR << 1);
1646                         dev_info(ctrl->ctrl.device,
1647                                 "NVME-FC{%d}: io failed due to bad transfer "
1648                                 "length: %d vs expected %d\n",
1649                                 ctrl->cnum, freq->transferred_length,
1650                                 be32_to_cpu(op->cmd_iu.data_len));
1651                         goto done;
1652                 }
1653                 result.u64 = 0;
1654                 break;
1655 
1656         case sizeof(struct nvme_fc_ersp_iu):
1657                 /*
1658                  * The ERSP IU contains a full completion with CQE.
1659                  * Validate ERSP IU and look at cqe.
1660                  */
1661                 if (unlikely(be16_to_cpu(op->rsp_iu.iu_len) !=
1662                                         (freq->rcv_rsplen / 4) ||
1663                              be32_to_cpu(op->rsp_iu.xfrd_len) !=
1664                                         freq->transferred_length ||
1665                              op->rsp_iu.status_code ||
1666                              sqe->common.command_id != cqe->command_id)) {
1667                         status = cpu_to_le16(NVME_SC_HOST_PATH_ERROR << 1);
1668                         dev_info(ctrl->ctrl.device,
1669                                 "NVME-FC{%d}: io failed due to bad NVMe_ERSP: "
1670                                 "iu len %d, xfr len %d vs %d, status code "
1671                                 "%d, cmdid %d vs %d\n",
1672                                 ctrl->cnum, be16_to_cpu(op->rsp_iu.iu_len),
1673                                 be32_to_cpu(op->rsp_iu.xfrd_len),
1674                                 freq->transferred_length,
1675                                 op->rsp_iu.status_code,
1676                                 sqe->common.command_id,
1677                                 cqe->command_id);
1678                         goto done;
1679                 }
1680                 result = cqe->result;
1681                 status = cqe->status;
1682                 break;
1683 
1684         default:
1685                 status = cpu_to_le16(NVME_SC_HOST_PATH_ERROR << 1);
1686                 dev_info(ctrl->ctrl.device,
1687                         "NVME-FC{%d}: io failed due to odd NVMe_xRSP iu "
1688                         "len %d\n",
1689                         ctrl->cnum, freq->rcv_rsplen);
1690                 goto done;
1691         }
1692 
1693         terminate_assoc = false;
1694 
1695 done:
1696         if (op->flags & FCOP_FLAGS_AEN) {
1697                 nvme_complete_async_event(&queue->ctrl->ctrl, status, &result);
1698                 __nvme_fc_fcpop_chk_teardowns(ctrl, op, opstate);
1699                 atomic_set(&op->state, FCPOP_STATE_IDLE);
1700                 op->flags = FCOP_FLAGS_AEN;     /* clear other flags */
1701                 nvme_fc_ctrl_put(ctrl);
1702                 goto check_error;
1703         }
1704 
1705         __nvme_fc_fcpop_chk_teardowns(ctrl, op, opstate);
1706         nvme_end_request(rq, status, result);
1707 
1708 check_error:
1709         if (terminate_assoc)
1710                 nvme_fc_error_recovery(ctrl, "transport detected io error");
1711 }
1712 
1713 static int
1714 __nvme_fc_init_request(struct nvme_fc_ctrl *ctrl,
1715                 struct nvme_fc_queue *queue, struct nvme_fc_fcp_op *op,
1716                 struct request *rq, u32 rqno)
1717 {
1718         struct nvme_fcp_op_w_sgl *op_w_sgl =
1719                 container_of(op, typeof(*op_w_sgl), op);
1720         struct nvme_fc_cmd_iu *cmdiu = &op->cmd_iu;
1721         int ret = 0;
1722 
1723         memset(op, 0, sizeof(*op));
1724         op->fcp_req.cmdaddr = &op->cmd_iu;
1725         op->fcp_req.cmdlen = sizeof(op->cmd_iu);
1726         op->fcp_req.rspaddr = &op->rsp_iu;
1727         op->fcp_req.rsplen = sizeof(op->rsp_iu);
1728         op->fcp_req.done = nvme_fc_fcpio_done;
1729         op->ctrl = ctrl;
1730         op->queue = queue;
1731         op->rq = rq;
1732         op->rqno = rqno;
1733 
1734         cmdiu->scsi_id = NVME_CMD_SCSI_ID;
1735         cmdiu->fc_id = NVME_CMD_FC_ID;
1736         cmdiu->iu_len = cpu_to_be16(sizeof(*cmdiu) / sizeof(u32));
1737 
1738         op->fcp_req.cmddma = fc_dma_map_single(ctrl->lport->dev,
1739                                 &op->cmd_iu, sizeof(op->cmd_iu), DMA_TO_DEVICE);
1740         if (fc_dma_mapping_error(ctrl->lport->dev, op->fcp_req.cmddma)) {
1741                 dev_err(ctrl->dev,
1742                         "FCP Op failed - cmdiu dma mapping failed.\n");
1743                 ret = EFAULT;
1744                 goto out_on_error;
1745         }
1746 
1747         op->fcp_req.rspdma = fc_dma_map_single(ctrl->lport->dev,
1748                                 &op->rsp_iu, sizeof(op->rsp_iu),
1749                                 DMA_FROM_DEVICE);
1750         if (fc_dma_mapping_error(ctrl->lport->dev, op->fcp_req.rspdma)) {
1751                 dev_err(ctrl->dev,
1752                         "FCP Op failed - rspiu dma mapping failed.\n");
1753                 ret = EFAULT;
1754         }
1755 
1756         atomic_set(&op->state, FCPOP_STATE_IDLE);
1757 out_on_error:
1758         return ret;
1759 }
1760 
1761 static int
1762 nvme_fc_init_request(struct blk_mq_tag_set *set, struct request *rq,
1763                 unsigned int hctx_idx, unsigned int numa_node)
1764 {
1765         struct nvme_fc_ctrl *ctrl = set->driver_data;
1766         struct nvme_fcp_op_w_sgl *op = blk_mq_rq_to_pdu(rq);
1767         int queue_idx = (set == &ctrl->tag_set) ? hctx_idx + 1 : 0;
1768         struct nvme_fc_queue *queue = &ctrl->queues[queue_idx];
1769         int res;
1770 
1771         res = __nvme_fc_init_request(ctrl, queue, &op->op, rq, queue->rqcnt++);
1772         if (res)
1773                 return res;
1774         op->op.fcp_req.first_sgl = &op->sgl[0];
1775         op->op.fcp_req.private = &op->priv[0];
1776         nvme_req(rq)->ctrl = &ctrl->ctrl;
1777         return res;
1778 }
1779 
1780 static int
1781 nvme_fc_init_aen_ops(struct nvme_fc_ctrl *ctrl)
1782 {
1783         struct nvme_fc_fcp_op *aen_op;
1784         struct nvme_fc_cmd_iu *cmdiu;
1785         struct nvme_command *sqe;
1786         void *private;
1787         int i, ret;
1788 
1789         aen_op = ctrl->aen_ops;
1790         for (i = 0; i < NVME_NR_AEN_COMMANDS; i++, aen_op++) {
1791                 private = kzalloc(ctrl->lport->ops->fcprqst_priv_sz,
1792                                                 GFP_KERNEL);
1793                 if (!private)
1794                         return -ENOMEM;
1795 
1796                 cmdiu = &aen_op->cmd_iu;
1797                 sqe = &cmdiu->sqe;
1798                 ret = __nvme_fc_init_request(ctrl, &ctrl->queues[0],
1799                                 aen_op, (struct request *)NULL,
1800                                 (NVME_AQ_BLK_MQ_DEPTH + i));
1801                 if (ret) {
1802                         kfree(private);
1803                         return ret;
1804                 }
1805 
1806                 aen_op->flags = FCOP_FLAGS_AEN;
1807                 aen_op->fcp_req.private = private;
1808 
1809                 memset(sqe, 0, sizeof(*sqe));
1810                 sqe->common.opcode = nvme_admin_async_event;
1811                 /* Note: core layer may overwrite the sqe.command_id value */
1812                 sqe->common.command_id = NVME_AQ_BLK_MQ_DEPTH + i;
1813         }
1814         return 0;
1815 }
1816 
1817 static void
1818 nvme_fc_term_aen_ops(struct nvme_fc_ctrl *ctrl)
1819 {
1820         struct nvme_fc_fcp_op *aen_op;
1821         int i;
1822 
1823         aen_op = ctrl->aen_ops;
1824         for (i = 0; i < NVME_NR_AEN_COMMANDS; i++, aen_op++) {
1825                 if (!aen_op->fcp_req.private)
1826                         continue;
1827 
1828                 __nvme_fc_exit_request(ctrl, aen_op);
1829 
1830                 kfree(aen_op->fcp_req.private);
1831                 aen_op->fcp_req.private = NULL;
1832         }
1833 }
1834 
1835 static inline void
1836 __nvme_fc_init_hctx(struct blk_mq_hw_ctx *hctx, struct nvme_fc_ctrl *ctrl,
1837                 unsigned int qidx)
1838 {
1839         struct nvme_fc_queue *queue = &ctrl->queues[qidx];
1840 
1841         hctx->driver_data = queue;
1842         queue->hctx = hctx;
1843 }
1844 
1845 static int
1846 nvme_fc_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
1847                 unsigned int hctx_idx)
1848 {
1849         struct nvme_fc_ctrl *ctrl = data;
1850 
1851         __nvme_fc_init_hctx(hctx, ctrl, hctx_idx + 1);
1852 
1853         return 0;
1854 }
1855 
1856 static int
1857 nvme_fc_init_admin_hctx(struct blk_mq_hw_ctx *hctx, void *data,
1858                 unsigned int hctx_idx)
1859 {
1860         struct nvme_fc_ctrl *ctrl = data;
1861 
1862         __nvme_fc_init_hctx(hctx, ctrl, hctx_idx);
1863 
1864         return 0;
1865 }
1866 
1867 static void
1868 nvme_fc_init_queue(struct nvme_fc_ctrl *ctrl, int idx)
1869 {
1870         struct nvme_fc_queue *queue;
1871 
1872         queue = &ctrl->queues[idx];
1873         memset(queue, 0, sizeof(*queue));
1874         queue->ctrl = ctrl;
1875         queue->qnum = idx;
1876         atomic_set(&queue->csn, 0);
1877         queue->dev = ctrl->dev;
1878 
1879         if (idx > 0)
1880                 queue->cmnd_capsule_len = ctrl->ctrl.ioccsz * 16;
1881         else
1882                 queue->cmnd_capsule_len = sizeof(struct nvme_command);
1883 
1884         /*
1885          * Considered whether we should allocate buffers for all SQEs
1886          * and CQEs and dma map them - mapping their respective entries
1887          * into the request structures (kernel vm addr and dma address)
1888          * thus the driver could use the buffers/mappings directly.
1889          * It only makes sense if the LLDD would use them for its
1890          * messaging api. It's very unlikely most adapter api's would use
1891          * a native NVME sqe/cqe. More reasonable if FC-NVME IU payload
1892          * structures were used instead.
1893          */
1894 }
1895 
1896 /*
1897  * This routine terminates a queue at the transport level.
1898  * The transport has already ensured that all outstanding ios on
1899  * the queue have been terminated.
1900  * The transport will send a Disconnect LS request to terminate
1901  * the queue's connection. Termination of the admin queue will also
1902  * terminate the association at the target.
1903  */
1904 static void
1905 nvme_fc_free_queue(struct nvme_fc_queue *queue)
1906 {
1907         if (!test_and_clear_bit(NVME_FC_Q_CONNECTED, &queue->flags))
1908                 return;
1909 
1910         clear_bit(NVME_FC_Q_LIVE, &queue->flags);
1911         /*
1912          * Current implementation never disconnects a single queue.
1913          * It always terminates a whole association. So there is never
1914          * a disconnect(queue) LS sent to the target.
1915          */
1916 
1917         queue->connection_id = 0;
1918         atomic_set(&queue->csn, 0);
1919 }
1920 
1921 static void
1922 __nvme_fc_delete_hw_queue(struct nvme_fc_ctrl *ctrl,
1923         struct nvme_fc_queue *queue, unsigned int qidx)
1924 {
1925         if (ctrl->lport->ops->delete_queue)
1926                 ctrl->lport->ops->delete_queue(&ctrl->lport->localport, qidx,
1927                                 queue->lldd_handle);
1928         queue->lldd_handle = NULL;
1929 }
1930 
1931 static void
1932 nvme_fc_free_io_queues(struct nvme_fc_ctrl *ctrl)
1933 {
1934         int i;
1935 
1936         for (i = 1; i < ctrl->ctrl.queue_count; i++)
1937                 nvme_fc_free_queue(&ctrl->queues[i]);
1938 }
1939 
1940 static int
1941 __nvme_fc_create_hw_queue(struct nvme_fc_ctrl *ctrl,
1942         struct nvme_fc_queue *queue, unsigned int qidx, u16 qsize)
1943 {
1944         int ret = 0;
1945 
1946         queue->lldd_handle = NULL;
1947         if (ctrl->lport->ops->create_queue)
1948                 ret = ctrl->lport->ops->create_queue(&ctrl->lport->localport,
1949                                 qidx, qsize, &queue->lldd_handle);
1950 
1951         return ret;
1952 }
1953 
1954 static void
1955 nvme_fc_delete_hw_io_queues(struct nvme_fc_ctrl *ctrl)
1956 {
1957         struct nvme_fc_queue *queue = &ctrl->queues[ctrl->ctrl.queue_count - 1];
1958         int i;
1959 
1960         for (i = ctrl->ctrl.queue_count - 1; i >= 1; i--, queue--)
1961                 __nvme_fc_delete_hw_queue(ctrl, queue, i);
1962 }
1963 
1964 static int
1965 nvme_fc_create_hw_io_queues(struct nvme_fc_ctrl *ctrl, u16 qsize)
1966 {
1967         struct nvme_fc_queue *queue = &ctrl->queues[1];
1968         int i, ret;
1969 
1970         for (i = 1; i < ctrl->ctrl.queue_count; i++, queue++) {
1971                 ret = __nvme_fc_create_hw_queue(ctrl, queue, i, qsize);
1972                 if (ret)
1973                         goto delete_queues;
1974         }
1975 
1976         return 0;
1977 
1978 delete_queues:
1979         for (; i >= 0; i--)
1980                 __nvme_fc_delete_hw_queue(ctrl, &ctrl->queues[i], i);
1981         return ret;
1982 }
1983 
1984 static int
1985 nvme_fc_connect_io_queues(struct nvme_fc_ctrl *ctrl, u16 qsize)
1986 {
1987         int i, ret = 0;
1988 
1989         for (i = 1; i < ctrl->ctrl.queue_count; i++) {
1990                 ret = nvme_fc_connect_queue(ctrl, &ctrl->queues[i], qsize,
1991                                         (qsize / 5));
1992                 if (ret)
1993                         break;
1994                 ret = nvmf_connect_io_queue(&ctrl->ctrl, i, false);
1995                 if (ret)
1996                         break;
1997 
1998                 set_bit(NVME_FC_Q_LIVE, &ctrl->queues[i].flags);
1999         }
2000 
2001         return ret;
2002 }
2003 
2004 static void
2005 nvme_fc_init_io_queues(struct nvme_fc_ctrl *ctrl)
2006 {
2007         int i;
2008 
2009         for (i = 1; i < ctrl->ctrl.queue_count; i++)
2010                 nvme_fc_init_queue(ctrl, i);
2011 }
2012 
2013 static void
2014 nvme_fc_ctrl_free(struct kref *ref)
2015 {
2016         struct nvme_fc_ctrl *ctrl =
2017                 container_of(ref, struct nvme_fc_ctrl, ref);
2018         unsigned long flags;
2019 
2020         if (ctrl->ctrl.tagset) {
2021                 blk_cleanup_queue(ctrl->ctrl.connect_q);
2022                 blk_mq_free_tag_set(&ctrl->tag_set);
2023         }
2024 
2025         /* remove from rport list */
2026         spin_lock_irqsave(&ctrl->rport->lock, flags);
2027         list_del(&ctrl->ctrl_list);
2028         spin_unlock_irqrestore(&ctrl->rport->lock, flags);
2029 
2030         blk_mq_unquiesce_queue(ctrl->ctrl.admin_q);
2031         blk_cleanup_queue(ctrl->ctrl.admin_q);
2032         blk_cleanup_queue(ctrl->ctrl.fabrics_q);
2033         blk_mq_free_tag_set(&ctrl->admin_tag_set);
2034 
2035         kfree(ctrl->queues);
2036 
2037         put_device(ctrl->dev);
2038         nvme_fc_rport_put(ctrl->rport);
2039 
2040         ida_simple_remove(&nvme_fc_ctrl_cnt, ctrl->cnum);
2041         if (ctrl->ctrl.opts)
2042                 nvmf_free_options(ctrl->ctrl.opts);
2043         kfree(ctrl);
2044 }
2045 
2046 static void
2047 nvme_fc_ctrl_put(struct nvme_fc_ctrl *ctrl)
2048 {
2049         kref_put(&ctrl->ref, nvme_fc_ctrl_free);
2050 }
2051 
2052 static int
2053 nvme_fc_ctrl_get(struct nvme_fc_ctrl *ctrl)
2054 {
2055         return kref_get_unless_zero(&ctrl->ref);
2056 }
2057 
2058 /*
2059  * All accesses from nvme core layer done - can now free the
2060  * controller. Called after last nvme_put_ctrl() call
2061  */
2062 static void
2063 nvme_fc_nvme_ctrl_freed(struct nvme_ctrl *nctrl)
2064 {
2065         struct nvme_fc_ctrl *ctrl = to_fc_ctrl(nctrl);
2066 
2067         WARN_ON(nctrl != &ctrl->ctrl);
2068 
2069         nvme_fc_ctrl_put(ctrl);
2070 }
2071 
2072 static void
2073 nvme_fc_error_recovery(struct nvme_fc_ctrl *ctrl, char *errmsg)
2074 {
2075         int active;
2076 
2077         /*
2078          * if an error (io timeout, etc) while (re)connecting,
2079          * it's an error on creating the new association.
2080          * Start the error recovery thread if it hasn't already
2081          * been started. It is expected there could be multiple
2082          * ios hitting this path before things are cleaned up.
2083          */
2084         if (ctrl->ctrl.state == NVME_CTRL_CONNECTING) {
2085                 active = atomic_xchg(&ctrl->err_work_active, 1);
2086                 if (!active && !queue_work(nvme_fc_wq, &ctrl->err_work)) {
2087                         atomic_set(&ctrl->err_work_active, 0);
2088                         WARN_ON(1);
2089                 }
2090                 return;
2091         }
2092 
2093         /* Otherwise, only proceed if in LIVE state - e.g. on first error */
2094         if (ctrl->ctrl.state != NVME_CTRL_LIVE)
2095                 return;
2096 
2097         dev_warn(ctrl->ctrl.device,
2098                 "NVME-FC{%d}: transport association error detected: %s\n",
2099                 ctrl->cnum, errmsg);
2100         dev_warn(ctrl->ctrl.device,
2101                 "NVME-FC{%d}: resetting controller\n", ctrl->cnum);
2102 
2103         nvme_reset_ctrl(&ctrl->ctrl);
2104 }
2105 
2106 static enum blk_eh_timer_return
2107 nvme_fc_timeout(struct request *rq, bool reserved)
2108 {
2109         struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
2110         struct nvme_fc_ctrl *ctrl = op->ctrl;
2111 
2112         /*
2113          * we can't individually ABTS an io without affecting the queue,
2114          * thus killing the queue, and thus the association.
2115          * So resolve by performing a controller reset, which will stop
2116          * the host/io stack, terminate the association on the link,
2117          * and recreate an association on the link.
2118          */
2119         nvme_fc_error_recovery(ctrl, "io timeout error");
2120 
2121         /*
2122          * the io abort has been initiated. Have the reset timer
2123          * restarted and the abort completion will complete the io
2124          * shortly. Avoids a synchronous wait while the abort finishes.
2125          */
2126         return BLK_EH_RESET_TIMER;
2127 }
2128 
2129 static int
2130 nvme_fc_map_data(struct nvme_fc_ctrl *ctrl, struct request *rq,
2131                 struct nvme_fc_fcp_op *op)
2132 {
2133         struct nvmefc_fcp_req *freq = &op->fcp_req;
2134         int ret;
2135 
2136         freq->sg_cnt = 0;
2137 
2138         if (!blk_rq_nr_phys_segments(rq))
2139                 return 0;
2140 
2141         freq->sg_table.sgl = freq->first_sgl;
2142         ret = sg_alloc_table_chained(&freq->sg_table,
2143                         blk_rq_nr_phys_segments(rq), freq->sg_table.sgl,
2144                         SG_CHUNK_SIZE);
2145         if (ret)
2146                 return -ENOMEM;
2147 
2148         op->nents = blk_rq_map_sg(rq->q, rq, freq->sg_table.sgl);
2149         WARN_ON(op->nents > blk_rq_nr_phys_segments(rq));
2150         freq->sg_cnt = fc_dma_map_sg(ctrl->lport->dev, freq->sg_table.sgl,
2151                                 op->nents, rq_dma_dir(rq));
2152         if (unlikely(freq->sg_cnt <= 0)) {
2153                 sg_free_table_chained(&freq->sg_table, SG_CHUNK_SIZE);
2154                 freq->sg_cnt = 0;
2155                 return -EFAULT;
2156         }
2157 
2158         /*
2159          * TODO: blk_integrity_rq(rq)  for DIF
2160          */
2161         return 0;
2162 }
2163 
2164 static void
2165 nvme_fc_unmap_data(struct nvme_fc_ctrl *ctrl, struct request *rq,
2166                 struct nvme_fc_fcp_op *op)
2167 {
2168         struct nvmefc_fcp_req *freq = &op->fcp_req;
2169 
2170         if (!freq->sg_cnt)
2171                 return;
2172 
2173         fc_dma_unmap_sg(ctrl->lport->dev, freq->sg_table.sgl, op->nents,
2174                         rq_dma_dir(rq));
2175 
2176         nvme_cleanup_cmd(rq);
2177 
2178         sg_free_table_chained(&freq->sg_table, SG_CHUNK_SIZE);
2179 
2180         freq->sg_cnt = 0;
2181 }
2182 
2183 /*
2184  * In FC, the queue is a logical thing. At transport connect, the target
2185  * creates its "queue" and returns a handle that is to be given to the
2186  * target whenever it posts something to the corresponding SQ.  When an
2187  * SQE is sent on a SQ, FC effectively considers the SQE, or rather the
2188  * command contained within the SQE, an io, and assigns a FC exchange
2189  * to it. The SQE and the associated SQ handle are sent in the initial
2190  * CMD IU sents on the exchange. All transfers relative to the io occur
2191  * as part of the exchange.  The CQE is the last thing for the io,
2192  * which is transferred (explicitly or implicitly) with the RSP IU
2193  * sent on the exchange. After the CQE is received, the FC exchange is
2194  * terminaed and the Exchange may be used on a different io.
2195  *
2196  * The transport to LLDD api has the transport making a request for a
2197  * new fcp io request to the LLDD. The LLDD then allocates a FC exchange
2198  * resource and transfers the command. The LLDD will then process all
2199  * steps to complete the io. Upon completion, the transport done routine
2200  * is called.
2201  *
2202  * So - while the operation is outstanding to the LLDD, there is a link
2203  * level FC exchange resource that is also outstanding. This must be
2204  * considered in all cleanup operations.
2205  */
2206 static blk_status_t
2207 nvme_fc_start_fcp_op(struct nvme_fc_ctrl *ctrl, struct nvme_fc_queue *queue,
2208         struct nvme_fc_fcp_op *op, u32 data_len,
2209         enum nvmefc_fcp_datadir io_dir)
2210 {
2211         struct nvme_fc_cmd_iu *cmdiu = &op->cmd_iu;
2212         struct nvme_command *sqe = &cmdiu->sqe;
2213         int ret, opstate;
2214 
2215         /*
2216          * before attempting to send the io, check to see if we believe
2217          * the target device is present
2218          */
2219         if (ctrl->rport->remoteport.port_state != FC_OBJSTATE_ONLINE)
2220                 return BLK_STS_RESOURCE;
2221 
2222         if (!nvme_fc_ctrl_get(ctrl))
2223                 return BLK_STS_IOERR;
2224 
2225         /* format the FC-NVME CMD IU and fcp_req */
2226         cmdiu->connection_id = cpu_to_be64(queue->connection_id);
2227         cmdiu->data_len = cpu_to_be32(data_len);
2228         switch (io_dir) {
2229         case NVMEFC_FCP_WRITE:
2230                 cmdiu->flags = FCNVME_CMD_FLAGS_WRITE;
2231                 break;
2232         case NVMEFC_FCP_READ:
2233                 cmdiu->flags = FCNVME_CMD_FLAGS_READ;
2234                 break;
2235         case NVMEFC_FCP_NODATA:
2236                 cmdiu->flags = 0;
2237                 break;
2238         }
2239         op->fcp_req.payload_length = data_len;
2240         op->fcp_req.io_dir = io_dir;
2241         op->fcp_req.transferred_length = 0;
2242         op->fcp_req.rcv_rsplen = 0;
2243         op->fcp_req.status = NVME_SC_SUCCESS;
2244         op->fcp_req.sqid = cpu_to_le16(queue->qnum);
2245 
2246         /*
2247          * validate per fabric rules, set fields mandated by fabric spec
2248          * as well as those by FC-NVME spec.
2249          */
2250         WARN_ON_ONCE(sqe->common.metadata);
2251         sqe->common.flags |= NVME_CMD_SGL_METABUF;
2252 
2253         /*
2254          * format SQE DPTR field per FC-NVME rules:
2255          *    type=0x5     Transport SGL Data Block Descriptor
2256          *    subtype=0xA  Transport-specific value
2257          *    address=0
2258          *    length=length of the data series
2259          */
2260         sqe->rw.dptr.sgl.type = (NVME_TRANSPORT_SGL_DATA_DESC << 4) |
2261                                         NVME_SGL_FMT_TRANSPORT_A;
2262         sqe->rw.dptr.sgl.length = cpu_to_le32(data_len);
2263         sqe->rw.dptr.sgl.addr = 0;
2264 
2265         if (!(op->flags & FCOP_FLAGS_AEN)) {
2266                 ret = nvme_fc_map_data(ctrl, op->rq, op);
2267                 if (ret < 0) {
2268                         nvme_cleanup_cmd(op->rq);
2269                         nvme_fc_ctrl_put(ctrl);
2270                         if (ret == -ENOMEM || ret == -EAGAIN)
2271                                 return BLK_STS_RESOURCE;
2272                         return BLK_STS_IOERR;
2273                 }
2274         }
2275 
2276         fc_dma_sync_single_for_device(ctrl->lport->dev, op->fcp_req.cmddma,
2277                                   sizeof(op->cmd_iu), DMA_TO_DEVICE);
2278 
2279         atomic_set(&op->state, FCPOP_STATE_ACTIVE);
2280 
2281         if (!(op->flags & FCOP_FLAGS_AEN))
2282                 blk_mq_start_request(op->rq);
2283 
2284         cmdiu->csn = cpu_to_be32(atomic_inc_return(&queue->csn));
2285         ret = ctrl->lport->ops->fcp_io(&ctrl->lport->localport,
2286                                         &ctrl->rport->remoteport,
2287                                         queue->lldd_handle, &op->fcp_req);
2288 
2289         if (ret) {
2290                 /*
2291                  * If the lld fails to send the command is there an issue with
2292                  * the csn value?  If the command that fails is the Connect,
2293                  * no - as the connection won't be live.  If it is a command
2294                  * post-connect, it's possible a gap in csn may be created.
2295                  * Does this matter?  As Linux initiators don't send fused
2296                  * commands, no.  The gap would exist, but as there's nothing
2297                  * that depends on csn order to be delivered on the target
2298                  * side, it shouldn't hurt.  It would be difficult for a
2299                  * target to even detect the csn gap as it has no idea when the
2300                  * cmd with the csn was supposed to arrive.
2301                  */
2302                 opstate = atomic_xchg(&op->state, FCPOP_STATE_COMPLETE);
2303                 __nvme_fc_fcpop_chk_teardowns(ctrl, op, opstate);
2304 
2305                 if (!(op->flags & FCOP_FLAGS_AEN))
2306                         nvme_fc_unmap_data(ctrl, op->rq, op);
2307 
2308                 nvme_fc_ctrl_put(ctrl);
2309 
2310                 if (ctrl->rport->remoteport.port_state == FC_OBJSTATE_ONLINE &&
2311                                 ret != -EBUSY)
2312                         return BLK_STS_IOERR;
2313 
2314                 return BLK_STS_RESOURCE;
2315         }
2316 
2317         return BLK_STS_OK;
2318 }
2319 
2320 static blk_status_t
2321 nvme_fc_queue_rq(struct blk_mq_hw_ctx *hctx,
2322                         const struct blk_mq_queue_data *bd)
2323 {
2324         struct nvme_ns *ns = hctx->queue->queuedata;
2325         struct nvme_fc_queue *queue = hctx->driver_data;
2326         struct nvme_fc_ctrl *ctrl = queue->ctrl;
2327         struct request *rq = bd->rq;
2328         struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
2329         struct nvme_fc_cmd_iu *cmdiu = &op->cmd_iu;
2330         struct nvme_command *sqe = &cmdiu->sqe;
2331         enum nvmefc_fcp_datadir io_dir;
2332         bool queue_ready = test_bit(NVME_FC_Q_LIVE, &queue->flags);
2333         u32 data_len;
2334         blk_status_t ret;
2335 
2336         if (ctrl->rport->remoteport.port_state != FC_OBJSTATE_ONLINE ||
2337             !nvmf_check_ready(&queue->ctrl->ctrl, rq, queue_ready))
2338                 return nvmf_fail_nonready_command(&queue->ctrl->ctrl, rq);
2339 
2340         ret = nvme_setup_cmd(ns, rq, sqe);
2341         if (ret)
2342                 return ret;
2343 
2344         /*
2345          * nvme core doesn't quite treat the rq opaquely. Commands such
2346          * as WRITE ZEROES will return a non-zero rq payload_bytes yet
2347          * there is no actual payload to be transferred.
2348          * To get it right, key data transmission on there being 1 or
2349          * more physical segments in the sg list. If there is no
2350          * physical segments, there is no payload.
2351          */
2352         if (blk_rq_nr_phys_segments(rq)) {
2353                 data_len = blk_rq_payload_bytes(rq);
2354                 io_dir = ((rq_data_dir(rq) == WRITE) ?
2355                                         NVMEFC_FCP_WRITE : NVMEFC_FCP_READ);
2356         } else {
2357                 data_len = 0;
2358                 io_dir = NVMEFC_FCP_NODATA;
2359         }
2360 
2361 
2362         return nvme_fc_start_fcp_op(ctrl, queue, op, data_len, io_dir);
2363 }
2364 
2365 static void
2366 nvme_fc_submit_async_event(struct nvme_ctrl *arg)
2367 {
2368         struct nvme_fc_ctrl *ctrl = to_fc_ctrl(arg);
2369         struct nvme_fc_fcp_op *aen_op;
2370         unsigned long flags;
2371         bool terminating = false;
2372         blk_status_t ret;
2373 
2374         spin_lock_irqsave(&ctrl->lock, flags);
2375         if (ctrl->flags & FCCTRL_TERMIO)
2376                 terminating = true;
2377         spin_unlock_irqrestore(&ctrl->lock, flags);
2378 
2379         if (terminating)
2380                 return;
2381 
2382         aen_op = &ctrl->aen_ops[0];
2383 
2384         ret = nvme_fc_start_fcp_op(ctrl, aen_op->queue, aen_op, 0,
2385                                         NVMEFC_FCP_NODATA);
2386         if (ret)
2387                 dev_err(ctrl->ctrl.device,
2388                         "failed async event work\n");
2389 }
2390 
2391 static void
2392 nvme_fc_complete_rq(struct request *rq)
2393 {
2394         struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
2395         struct nvme_fc_ctrl *ctrl = op->ctrl;
2396 
2397         atomic_set(&op->state, FCPOP_STATE_IDLE);
2398 
2399         nvme_fc_unmap_data(ctrl, rq, op);
2400         nvme_complete_rq(rq);
2401         nvme_fc_ctrl_put(ctrl);
2402 }
2403 
2404 /*
2405  * This routine is used by the transport when it needs to find active
2406  * io on a queue that is to be terminated. The transport uses
2407  * blk_mq_tagset_busy_itr() to find the busy requests, which then invoke
2408  * this routine to kill them on a 1 by 1 basis.
2409  *
2410  * As FC allocates FC exchange for each io, the transport must contact
2411  * the LLDD to terminate the exchange, thus releasing the FC exchange.
2412  * After terminating the exchange the LLDD will call the transport's
2413  * normal io done path for the request, but it will have an aborted
2414  * status. The done path will return the io request back to the block
2415  * layer with an error status.
2416  */
2417 static bool
2418 nvme_fc_terminate_exchange(struct request *req, void *data, bool reserved)
2419 {
2420         struct nvme_ctrl *nctrl = data;
2421         struct nvme_fc_ctrl *ctrl = to_fc_ctrl(nctrl);
2422         struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(req);
2423 
2424         __nvme_fc_abort_op(ctrl, op);
2425         return true;
2426 }
2427 
2428 
2429 static const struct blk_mq_ops nvme_fc_mq_ops = {
2430         .queue_rq       = nvme_fc_queue_rq,
2431         .complete       = nvme_fc_complete_rq,
2432         .init_request   = nvme_fc_init_request,
2433         .exit_request   = nvme_fc_exit_request,
2434         .init_hctx      = nvme_fc_init_hctx,
2435         .timeout        = nvme_fc_timeout,
2436 };
2437 
2438 static int
2439 nvme_fc_create_io_queues(struct nvme_fc_ctrl *ctrl)
2440 {
2441         struct nvmf_ctrl_options *opts = ctrl->ctrl.opts;
2442         unsigned int nr_io_queues;
2443         int ret;
2444 
2445         nr_io_queues = min(min(opts->nr_io_queues, num_online_cpus()),
2446                                 ctrl->lport->ops->max_hw_queues);
2447         ret = nvme_set_queue_count(&ctrl->ctrl, &nr_io_queues);
2448         if (ret) {
2449                 dev_info(ctrl->ctrl.device,
2450                         "set_queue_count failed: %d\n", ret);
2451                 return ret;
2452         }
2453 
2454         ctrl->ctrl.queue_count = nr_io_queues + 1;
2455         if (!nr_io_queues)
2456                 return 0;
2457 
2458         nvme_fc_init_io_queues(ctrl);
2459 
2460         memset(&ctrl->tag_set, 0, sizeof(ctrl->tag_set));
2461         ctrl->tag_set.ops = &nvme_fc_mq_ops;
2462         ctrl->tag_set.queue_depth = ctrl->ctrl.opts->queue_size;
2463         ctrl->tag_set.reserved_tags = 1; /* fabric connect */
2464         ctrl->tag_set.numa_node = ctrl->ctrl.numa_node;
2465         ctrl->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
2466         ctrl->tag_set.cmd_size =
2467                 struct_size((struct nvme_fcp_op_w_sgl *)NULL, priv,
2468                             ctrl->lport->ops->fcprqst_priv_sz);
2469         ctrl->tag_set.driver_data = ctrl;
2470         ctrl->tag_set.nr_hw_queues = ctrl->ctrl.queue_count - 1;
2471         ctrl->tag_set.timeout = NVME_IO_TIMEOUT;
2472 
2473         ret = blk_mq_alloc_tag_set(&ctrl->tag_set);
2474         if (ret)
2475                 return ret;
2476 
2477         ctrl->ctrl.tagset = &ctrl->tag_set;
2478 
2479         ctrl->ctrl.connect_q = blk_mq_init_queue(&ctrl->tag_set);
2480         if (IS_ERR(ctrl->ctrl.connect_q)) {
2481                 ret = PTR_ERR(ctrl->ctrl.connect_q);
2482                 goto out_free_tag_set;
2483         }
2484 
2485         ret = nvme_fc_create_hw_io_queues(ctrl, ctrl->ctrl.sqsize + 1);
2486         if (ret)
2487                 goto out_cleanup_blk_queue;
2488 
2489         ret = nvme_fc_connect_io_queues(ctrl, ctrl->ctrl.sqsize + 1);
2490         if (ret)
2491                 goto out_delete_hw_queues;
2492 
2493         ctrl->ioq_live = true;
2494 
2495         return 0;
2496 
2497 out_delete_hw_queues:
2498         nvme_fc_delete_hw_io_queues(ctrl);
2499 out_cleanup_blk_queue:
2500         blk_cleanup_queue(ctrl->ctrl.connect_q);
2501 out_free_tag_set:
2502         blk_mq_free_tag_set(&ctrl->tag_set);
2503         nvme_fc_free_io_queues(ctrl);
2504 
2505         /* force put free routine to ignore io queues */
2506         ctrl->ctrl.tagset = NULL;
2507 
2508         return ret;
2509 }
2510 
2511 static int
2512 nvme_fc_recreate_io_queues(struct nvme_fc_ctrl *ctrl)
2513 {
2514         struct nvmf_ctrl_options *opts = ctrl->ctrl.opts;
2515         u32 prior_ioq_cnt = ctrl->ctrl.queue_count - 1;
2516         unsigned int nr_io_queues;
2517         int ret;
2518 
2519         nr_io_queues = min(min(opts->nr_io_queues, num_online_cpus()),
2520                                 ctrl->lport->ops->max_hw_queues);
2521         ret = nvme_set_queue_count(&ctrl->ctrl, &nr_io_queues);
2522         if (ret) {
2523                 dev_info(ctrl->ctrl.device,
2524                         "set_queue_count failed: %d\n", ret);
2525                 return ret;
2526         }
2527 
2528         if (!nr_io_queues && prior_ioq_cnt) {
2529                 dev_info(ctrl->ctrl.device,
2530                         "Fail Reconnect: At least 1 io queue "
2531                         "required (was %d)\n", prior_ioq_cnt);
2532                 return -ENOSPC;
2533         }
2534 
2535         ctrl->ctrl.queue_count = nr_io_queues + 1;
2536         /* check for io queues existing */
2537         if (ctrl->ctrl.queue_count == 1)
2538                 return 0;
2539 
2540         ret = nvme_fc_create_hw_io_queues(ctrl, ctrl->ctrl.sqsize + 1);
2541         if (ret)
2542                 goto out_free_io_queues;
2543 
2544         ret = nvme_fc_connect_io_queues(ctrl, ctrl->ctrl.sqsize + 1);
2545         if (ret)
2546                 goto out_delete_hw_queues;
2547 
2548         if (prior_ioq_cnt != nr_io_queues)
2549                 dev_info(ctrl->ctrl.device,
2550                         "reconnect: revising io queue count from %d to %d\n",
2551                         prior_ioq_cnt, nr_io_queues);
2552         blk_mq_update_nr_hw_queues(&ctrl->tag_set, nr_io_queues);
2553 
2554         return 0;
2555 
2556 out_delete_hw_queues:
2557         nvme_fc_delete_hw_io_queues(ctrl);
2558 out_free_io_queues:
2559         nvme_fc_free_io_queues(ctrl);
2560         return ret;
2561 }
2562 
2563 static void
2564 nvme_fc_rport_active_on_lport(struct nvme_fc_rport *rport)
2565 {
2566         struct nvme_fc_lport *lport = rport->lport;
2567 
2568         atomic_inc(&lport->act_rport_cnt);
2569 }
2570 
2571 static void
2572 nvme_fc_rport_inactive_on_lport(struct nvme_fc_rport *rport)
2573 {
2574         struct nvme_fc_lport *lport = rport->lport;
2575         u32 cnt;
2576 
2577         cnt = atomic_dec_return(&lport->act_rport_cnt);
2578         if (cnt == 0 && lport->localport.port_state == FC_OBJSTATE_DELETED)
2579                 lport->ops->localport_delete(&lport->localport);
2580 }
2581 
2582 static int
2583 nvme_fc_ctlr_active_on_rport(struct nvme_fc_ctrl *ctrl)
2584 {
2585         struct nvme_fc_rport *rport = ctrl->rport;
2586         u32 cnt;
2587 
2588         if (ctrl->assoc_active)
2589                 return 1;
2590 
2591         ctrl->assoc_active = true;
2592         cnt = atomic_inc_return(&rport->act_ctrl_cnt);
2593         if (cnt == 1)
2594                 nvme_fc_rport_active_on_lport(rport);
2595 
2596         return 0;
2597 }
2598 
2599 static int
2600 nvme_fc_ctlr_inactive_on_rport(struct nvme_fc_ctrl *ctrl)
2601 {
2602         struct nvme_fc_rport *rport = ctrl->rport;
2603         struct nvme_fc_lport *lport = rport->lport;
2604         u32 cnt;
2605 
2606         /* ctrl->assoc_active=false will be set independently */
2607 
2608         cnt = atomic_dec_return(&rport->act_ctrl_cnt);
2609         if (cnt == 0) {
2610                 if (rport->remoteport.port_state == FC_OBJSTATE_DELETED)
2611                         lport->ops->remoteport_delete(&rport->remoteport);
2612                 nvme_fc_rport_inactive_on_lport(rport);
2613         }
2614 
2615         return 0;
2616 }
2617 
2618 /*
2619  * This routine restarts the controller on the host side, and
2620  * on the link side, recreates the controller association.
2621  */
2622 static int
2623 nvme_fc_create_association(struct nvme_fc_ctrl *ctrl)
2624 {
2625         struct nvmf_ctrl_options *opts = ctrl->ctrl.opts;
2626         int ret;
2627         bool changed;
2628 
2629         ++ctrl->ctrl.nr_reconnects;
2630 
2631         if (ctrl->rport->remoteport.port_state != FC_OBJSTATE_ONLINE)
2632                 return -ENODEV;
2633 
2634         if (nvme_fc_ctlr_active_on_rport(ctrl))
2635                 return -ENOTUNIQ;
2636 
2637         dev_info(ctrl->ctrl.device,
2638                 "NVME-FC{%d}: create association : host wwpn 0x%016llx "
2639                 " rport wwpn 0x%016llx: NQN \"%s\"\n",
2640                 ctrl->cnum, ctrl->lport->localport.port_name,
2641                 ctrl->rport->remoteport.port_name, ctrl->ctrl.opts->subsysnqn);
2642 
2643         /*
2644          * Create the admin queue
2645          */
2646 
2647         ret = __nvme_fc_create_hw_queue(ctrl, &ctrl->queues[0], 0,
2648                                 NVME_AQ_DEPTH);
2649         if (ret)
2650                 goto out_free_queue;
2651 
2652         ret = nvme_fc_connect_admin_queue(ctrl, &ctrl->queues[0],
2653                                 NVME_AQ_DEPTH, (NVME_AQ_DEPTH / 4));
2654         if (ret)
2655                 goto out_delete_hw_queue;
2656 
2657         ret = nvmf_connect_admin_queue(&ctrl->ctrl);
2658         if (ret)
2659                 goto out_disconnect_admin_queue;
2660 
2661         set_bit(NVME_FC_Q_LIVE, &ctrl->queues[0].flags);
2662 
2663         /*
2664          * Check controller capabilities
2665          *
2666          * todo:- add code to check if ctrl attributes changed from
2667          * prior connection values
2668          */
2669 
2670         ret = nvme_enable_ctrl(&ctrl->ctrl);
2671         if (ret)
2672                 goto out_disconnect_admin_queue;
2673 
2674         ctrl->ctrl.max_hw_sectors =
2675                 (ctrl->lport->ops->max_sgl_segments - 1) << (PAGE_SHIFT - 9);
2676 
2677         blk_mq_unquiesce_queue(ctrl->ctrl.admin_q);
2678 
2679         ret = nvme_init_identify(&ctrl->ctrl);
2680         if (ret)
2681                 goto out_disconnect_admin_queue;
2682 
2683         /* sanity checks */
2684 
2685         /* FC-NVME does not have other data in the capsule */
2686         if (ctrl->ctrl.icdoff) {
2687                 dev_err(ctrl->ctrl.device, "icdoff %d is not supported!\n",
2688                                 ctrl->ctrl.icdoff);
2689                 goto out_disconnect_admin_queue;
2690         }
2691 
2692         /* FC-NVME supports normal SGL Data Block Descriptors */
2693 
2694         if (opts->queue_size > ctrl->ctrl.maxcmd) {
2695                 /* warn if maxcmd is lower than queue_size */
2696                 dev_warn(ctrl->ctrl.device,
2697                         "queue_size %zu > ctrl maxcmd %u, reducing "
2698                         "to queue_size\n",
2699                         opts->queue_size, ctrl->ctrl.maxcmd);
2700                 opts->queue_size = ctrl->ctrl.maxcmd;
2701         }
2702 
2703         if (opts->queue_size > ctrl->ctrl.sqsize + 1) {
2704                 /* warn if sqsize is lower than queue_size */
2705                 dev_warn(ctrl->ctrl.device,
2706                         "queue_size %zu > ctrl sqsize %u, clamping down\n",
2707                         opts->queue_size, ctrl->ctrl.sqsize + 1);
2708                 opts->queue_size = ctrl->ctrl.sqsize + 1;
2709         }
2710 
2711         ret = nvme_fc_init_aen_ops(ctrl);
2712         if (ret)
2713                 goto out_term_aen_ops;
2714 
2715         /*
2716          * Create the io queues
2717          */
2718 
2719         if (ctrl->ctrl.queue_count > 1) {
2720                 if (!ctrl->ioq_live)
2721                         ret = nvme_fc_create_io_queues(ctrl);
2722                 else
2723                         ret = nvme_fc_recreate_io_queues(ctrl);
2724                 if (ret)
2725                         goto out_term_aen_ops;
2726         }
2727 
2728         changed = nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_LIVE);
2729 
2730         ctrl->ctrl.nr_reconnects = 0;
2731 
2732         if (changed)
2733                 nvme_start_ctrl(&ctrl->ctrl);
2734 
2735         return 0;       /* Success */
2736 
2737 out_term_aen_ops:
2738         nvme_fc_term_aen_ops(ctrl);
2739 out_disconnect_admin_queue:
2740         /* send a Disconnect(association) LS to fc-nvme target */
2741         nvme_fc_xmt_disconnect_assoc(ctrl);
2742 out_delete_hw_queue:
2743         __nvme_fc_delete_hw_queue(ctrl, &ctrl->queues[0], 0);
2744 out_free_queue:
2745         nvme_fc_free_queue(&ctrl->queues[0]);
2746         ctrl->assoc_active = false;
2747         nvme_fc_ctlr_inactive_on_rport(ctrl);
2748 
2749         return ret;
2750 }
2751 
2752 /*
2753  * This routine stops operation of the controller on the host side.
2754  * On the host os stack side: Admin and IO queues are stopped,
2755  *   outstanding ios on them terminated via FC ABTS.
2756  * On the link side: the association is terminated.
2757  */
2758 static void
2759 nvme_fc_delete_association(struct nvme_fc_ctrl *ctrl)
2760 {
2761         unsigned long flags;
2762 
2763         if (!ctrl->assoc_active)
2764                 return;
2765         ctrl->assoc_active = false;
2766 
2767         spin_lock_irqsave(&ctrl->lock, flags);
2768         ctrl->flags |= FCCTRL_TERMIO;
2769         ctrl->iocnt = 0;
2770         spin_unlock_irqrestore(&ctrl->lock, flags);
2771 
2772         /*
2773          * If io queues are present, stop them and terminate all outstanding
2774          * ios on them. As FC allocates FC exchange for each io, the
2775          * transport must contact the LLDD to terminate the exchange,
2776          * thus releasing the FC exchange. We use blk_mq_tagset_busy_itr()
2777          * to tell us what io's are busy and invoke a transport routine
2778          * to kill them with the LLDD.  After terminating the exchange
2779          * the LLDD will call the transport's normal io done path, but it
2780          * will have an aborted status. The done path will return the
2781          * io requests back to the block layer as part of normal completions
2782          * (but with error status).
2783          */
2784         if (ctrl->ctrl.queue_count > 1) {
2785                 nvme_stop_queues(&ctrl->ctrl);
2786                 blk_mq_tagset_busy_iter(&ctrl->tag_set,
2787                                 nvme_fc_terminate_exchange, &ctrl->ctrl);
2788                 blk_mq_tagset_wait_completed_request(&ctrl->tag_set);
2789         }
2790 
2791         /*
2792          * Other transports, which don't have link-level contexts bound
2793          * to sqe's, would try to gracefully shutdown the controller by
2794          * writing the registers for shutdown and polling (call
2795          * nvme_shutdown_ctrl()). Given a bunch of i/o was potentially
2796          * just aborted and we will wait on those contexts, and given
2797          * there was no indication of how live the controlelr is on the
2798          * link, don't send more io to create more contexts for the
2799          * shutdown. Let the controller fail via keepalive failure if
2800          * its still present.
2801          */
2802 
2803         /*
2804          * clean up the admin queue. Same thing as above.
2805          * use blk_mq_tagset_busy_itr() and the transport routine to
2806          * terminate the exchanges.
2807          */
2808         blk_mq_quiesce_queue(ctrl->ctrl.admin_q);
2809         blk_mq_tagset_busy_iter(&ctrl->admin_tag_set,
2810                                 nvme_fc_terminate_exchange, &ctrl->ctrl);
2811         blk_mq_tagset_wait_completed_request(&ctrl->admin_tag_set);
2812 
2813         /* kill the aens as they are a separate path */
2814         nvme_fc_abort_aen_ops(ctrl);
2815 
2816         /* wait for all io that had to be aborted */
2817         spin_lock_irq(&ctrl->lock);
2818         wait_event_lock_irq(ctrl->ioabort_wait, ctrl->iocnt == 0, ctrl->lock);
2819         ctrl->flags &= ~FCCTRL_TERMIO;
2820         spin_unlock_irq(&ctrl->lock);
2821 
2822         nvme_fc_term_aen_ops(ctrl);
2823 
2824         /*
2825          * send a Disconnect(association) LS to fc-nvme target
2826          * Note: could have been sent at top of process, but
2827          * cleaner on link traffic if after the aborts complete.
2828          * Note: if association doesn't exist, association_id will be 0
2829          */
2830         if (ctrl->association_id)
2831                 nvme_fc_xmt_disconnect_assoc(ctrl);
2832 
2833         if (ctrl->ctrl.tagset) {
2834                 nvme_fc_delete_hw_io_queues(ctrl);
2835                 nvme_fc_free_io_queues(ctrl);
2836         }
2837 
2838         __nvme_fc_delete_hw_queue(ctrl, &ctrl->queues[0], 0);
2839         nvme_fc_free_queue(&ctrl->queues[0]);
2840 
2841         /* re-enable the admin_q so anything new can fast fail */
2842         blk_mq_unquiesce_queue(ctrl->ctrl.admin_q);
2843 
2844         /* resume the io queues so that things will fast fail */
2845         nvme_start_queues(&ctrl->ctrl);
2846 
2847         nvme_fc_ctlr_inactive_on_rport(ctrl);
2848 }
2849 
2850 static void
2851 nvme_fc_delete_ctrl(struct nvme_ctrl *nctrl)
2852 {
2853         struct nvme_fc_ctrl *ctrl = to_fc_ctrl(nctrl);
2854 
2855         cancel_work_sync(&ctrl->err_work);
2856         cancel_delayed_work_sync(&ctrl->connect_work);
2857         /*
2858          * kill the association on the link side.  this will block
2859          * waiting for io to terminate
2860          */
2861         nvme_fc_delete_association(ctrl);
2862 }
2863 
2864 static void
2865 nvme_fc_reconnect_or_delete(struct nvme_fc_ctrl *ctrl, int status)
2866 {
2867         struct nvme_fc_rport *rport = ctrl->rport;
2868         struct nvme_fc_remote_port *portptr = &rport->remoteport;
2869         unsigned long recon_delay = ctrl->ctrl.opts->reconnect_delay * HZ;
2870         bool recon = true;
2871 
2872         if (ctrl->ctrl.state != NVME_CTRL_CONNECTING)
2873                 return;
2874 
2875         if (portptr->port_state == FC_OBJSTATE_ONLINE)
2876                 dev_info(ctrl->ctrl.device,
2877                         "NVME-FC{%d}: reset: Reconnect attempt failed (%d)\n",
2878                         ctrl->cnum, status);
2879         else if (time_after_eq(jiffies, rport->dev_loss_end))
2880                 recon = false;
2881 
2882         if (recon && nvmf_should_reconnect(&ctrl->ctrl)) {
2883                 if (portptr->port_state == FC_OBJSTATE_ONLINE)
2884                         dev_info(ctrl->ctrl.device,
2885                                 "NVME-FC{%d}: Reconnect attempt in %ld "
2886                                 "seconds\n",
2887                                 ctrl->cnum, recon_delay / HZ);
2888                 else if (time_after(jiffies + recon_delay, rport->dev_loss_end))
2889                         recon_delay = rport->dev_loss_end - jiffies;
2890 
2891                 queue_delayed_work(nvme_wq, &ctrl->connect_work, recon_delay);
2892         } else {
2893                 if (portptr->port_state == FC_OBJSTATE_ONLINE)
2894                         dev_warn(ctrl->ctrl.device,
2895                                 "NVME-FC{%d}: Max reconnect attempts (%d) "
2896                                 "reached.\n",
2897                                 ctrl->cnum, ctrl->ctrl.nr_reconnects);
2898                 else
2899                         dev_warn(ctrl->ctrl.device,
2900                                 "NVME-FC{%d}: dev_loss_tmo (%d) expired "
2901                                 "while waiting for remoteport connectivity.\n",
2902                                 ctrl->cnum, portptr->dev_loss_tmo);
2903                 WARN_ON(nvme_delete_ctrl(&ctrl->ctrl));
2904         }
2905 }
2906 
2907 static void
2908 __nvme_fc_terminate_io(struct nvme_fc_ctrl *ctrl)
2909 {
2910         /*
2911          * if state is connecting - the error occurred as part of a
2912          * reconnect attempt. The create_association error paths will
2913          * clean up any outstanding io.
2914          *
2915          * if it's a different state - ensure all pending io is
2916          * terminated. Given this can delay while waiting for the
2917          * aborted io to return, we recheck adapter state below
2918          * before changing state.
2919          */
2920         if (ctrl->ctrl.state != NVME_CTRL_CONNECTING) {
2921                 nvme_stop_keep_alive(&ctrl->ctrl);
2922 
2923                 /* will block will waiting for io to terminate */
2924                 nvme_fc_delete_association(ctrl);
2925         }
2926 
2927         if (ctrl->ctrl.state != NVME_CTRL_CONNECTING &&
2928             !nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_CONNECTING))
2929                 dev_err(ctrl->ctrl.device,
2930                         "NVME-FC{%d}: error_recovery: Couldn't change state "
2931                         "to CONNECTING\n", ctrl->cnum);
2932 }
2933 
2934 static void
2935 nvme_fc_reset_ctrl_work(struct work_struct *work)
2936 {
2937         struct nvme_fc_ctrl *ctrl =
2938                 container_of(work, struct nvme_fc_ctrl, ctrl.reset_work);
2939         int ret;
2940 
2941         __nvme_fc_terminate_io(ctrl);
2942 
2943         nvme_stop_ctrl(&ctrl->ctrl);
2944 
2945         if (ctrl->rport->remoteport.port_state == FC_OBJSTATE_ONLINE)
2946                 ret = nvme_fc_create_association(ctrl);
2947         else
2948                 ret = -ENOTCONN;
2949 
2950         if (ret)
2951                 nvme_fc_reconnect_or_delete(ctrl, ret);
2952         else
2953                 dev_info(ctrl->ctrl.device,
2954                         "NVME-FC{%d}: controller reset complete\n",
2955                         ctrl->cnum);
2956 }
2957 
2958 static void
2959 nvme_fc_connect_err_work(struct work_struct *work)
2960 {
2961         struct nvme_fc_ctrl *ctrl =
2962                         container_of(work, struct nvme_fc_ctrl, err_work);
2963 
2964         __nvme_fc_terminate_io(ctrl);
2965 
2966         atomic_set(&ctrl->err_work_active, 0);
2967 
2968         /*
2969          * Rescheduling the connection after recovering
2970          * from the io error is left to the reconnect work
2971          * item, which is what should have stalled waiting on
2972          * the io that had the error that scheduled this work.
2973          */
2974 }
2975 
2976 static const struct nvme_ctrl_ops nvme_fc_ctrl_ops = {
2977         .name                   = "fc",
2978         .module                 = THIS_MODULE,
2979         .flags                  = NVME_F_FABRICS,
2980         .reg_read32             = nvmf_reg_read32,
2981         .reg_read64             = nvmf_reg_read64,
2982         .reg_write32            = nvmf_reg_write32,
2983         .free_ctrl              = nvme_fc_nvme_ctrl_freed,
2984         .submit_async_event     = nvme_fc_submit_async_event,
2985         .delete_ctrl            = nvme_fc_delete_ctrl,
2986         .get_address            = nvmf_get_address,
2987 };
2988 
2989 static void
2990 nvme_fc_connect_ctrl_work(struct work_struct *work)
2991 {
2992         int ret;
2993 
2994         struct nvme_fc_ctrl *ctrl =
2995                         container_of(to_delayed_work(work),
2996                                 struct nvme_fc_ctrl, connect_work);
2997 
2998         ret = nvme_fc_create_association(ctrl);
2999         if (ret)
3000                 nvme_fc_reconnect_or_delete(ctrl, ret);
3001         else
3002                 dev_info(ctrl->ctrl.device,
3003                         "NVME-FC{%d}: controller connect complete\n",
3004                         ctrl->cnum);
3005 }
3006 
3007 
3008 static const struct blk_mq_ops nvme_fc_admin_mq_ops = {
3009         .queue_rq       = nvme_fc_queue_rq,
3010         .complete       = nvme_fc_complete_rq,
3011         .init_request   = nvme_fc_init_request,
3012         .exit_request   = nvme_fc_exit_request,
3013         .init_hctx      = nvme_fc_init_admin_hctx,
3014         .timeout        = nvme_fc_timeout,
3015 };
3016 
3017 
3018 /*
3019  * Fails a controller request if it matches an existing controller
3020  * (association) with the same tuple:
3021  * <Host NQN, Host ID, local FC port, remote FC port, SUBSYS NQN>
3022  *
3023  * The ports don't need to be compared as they are intrinsically
3024  * already matched by the port pointers supplied.
3025  */
3026 static bool
3027 nvme_fc_existing_controller(struct nvme_fc_rport *rport,
3028                 struct nvmf_ctrl_options *opts)
3029 {
3030         struct nvme_fc_ctrl *ctrl;
3031         unsigned long flags;
3032         bool found = false;
3033 
3034         spin_lock_irqsave(&rport->lock, flags);
3035         list_for_each_entry(ctrl, &rport->ctrl_list, ctrl_list) {
3036                 found = nvmf_ctlr_matches_baseopts(&ctrl->ctrl, opts);
3037                 if (found)
3038                         break;
3039         }
3040         spin_unlock_irqrestore(&rport->lock, flags);
3041 
3042         return found;
3043 }
3044 
3045 static struct nvme_ctrl *
3046 nvme_fc_init_ctrl(struct device *dev, struct nvmf_ctrl_options *opts,
3047         struct nvme_fc_lport *lport, struct nvme_fc_rport *rport)
3048 {
3049         struct nvme_fc_ctrl *ctrl;
3050         unsigned long flags;
3051         int ret, idx;
3052 
3053         if (!(rport->remoteport.port_role &
3054             (FC_PORT_ROLE_NVME_DISCOVERY | FC_PORT_ROLE_NVME_TARGET))) {
3055                 ret = -EBADR;
3056                 goto out_fail;
3057         }
3058 
3059         if (!opts->duplicate_connect &&
3060             nvme_fc_existing_controller(rport, opts)) {
3061                 ret = -EALREADY;
3062                 goto out_fail;
3063         }
3064 
3065         ctrl = kzalloc(sizeof(*ctrl), GFP_KERNEL);
3066         if (!ctrl) {
3067                 ret = -ENOMEM;
3068                 goto out_fail;
3069         }
3070 
3071         idx = ida_simple_get(&nvme_fc_ctrl_cnt, 0, 0, GFP_KERNEL);
3072         if (idx < 0) {
3073                 ret = -ENOSPC;
3074                 goto out_free_ctrl;
3075         }
3076 
3077         ctrl->ctrl.opts = opts;
3078         ctrl->ctrl.nr_reconnects = 0;
3079         if (lport->dev)
3080                 ctrl->ctrl.numa_node = dev_to_node(lport->dev);
3081         else
3082                 ctrl->ctrl.numa_node = NUMA_NO_NODE;
3083         INIT_LIST_HEAD(&ctrl->ctrl_list);
3084         ctrl->lport = lport;
3085         ctrl->rport = rport;
3086         ctrl->dev = lport->dev;
3087         ctrl->cnum = idx;
3088         ctrl->ioq_live = false;
3089         ctrl->assoc_active = false;
3090         atomic_set(&ctrl->err_work_active, 0);
3091         init_waitqueue_head(&ctrl->ioabort_wait);
3092 
3093         get_device(ctrl->dev);
3094         kref_init(&ctrl->ref);
3095 
3096         INIT_WORK(&ctrl->ctrl.reset_work, nvme_fc_reset_ctrl_work);
3097         INIT_DELAYED_WORK(&ctrl->connect_work, nvme_fc_connect_ctrl_work);
3098         INIT_WORK(&ctrl->err_work, nvme_fc_connect_err_work);
3099         spin_lock_init(&ctrl->lock);
3100 
3101         /* io queue count */
3102         ctrl->ctrl.queue_count = min_t(unsigned int,
3103                                 opts->nr_io_queues,
3104                                 lport->ops->max_hw_queues);
3105         ctrl->ctrl.queue_count++;       /* +1 for admin queue */
3106 
3107         ctrl->ctrl.sqsize = opts->queue_size - 1;
3108         ctrl->ctrl.kato = opts->kato;
3109         ctrl->ctrl.cntlid = 0xffff;
3110 
3111         ret = -ENOMEM;
3112         ctrl->queues = kcalloc(ctrl->ctrl.queue_count,
3113                                 sizeof(struct nvme_fc_queue), GFP_KERNEL);
3114         if (!ctrl->queues)
3115                 goto out_free_ida;
3116 
3117         nvme_fc_init_queue(ctrl, 0);
3118 
3119         memset(&ctrl->admin_tag_set, 0, sizeof(ctrl->admin_tag_set));
3120         ctrl->admin_tag_set.ops = &nvme_fc_admin_mq_ops;
3121         ctrl->admin_tag_set.queue_depth = NVME_AQ_MQ_TAG_DEPTH;
3122         ctrl->admin_tag_set.reserved_tags = 2; /* fabric connect + Keep-Alive */
3123         ctrl->admin_tag_set.numa_node = ctrl->ctrl.numa_node;
3124         ctrl->admin_tag_set.cmd_size =
3125                 struct_size((struct nvme_fcp_op_w_sgl *)NULL, priv,
3126                             ctrl->lport->ops->fcprqst_priv_sz);
3127         ctrl->admin_tag_set.driver_data = ctrl;
3128         ctrl->admin_tag_set.nr_hw_queues = 1;
3129         ctrl->admin_tag_set.timeout = ADMIN_TIMEOUT;
3130         ctrl->admin_tag_set.flags = BLK_MQ_F_NO_SCHED;
3131 
3132         ret = blk_mq_alloc_tag_set(&ctrl->admin_tag_set);
3133         if (ret)
3134                 goto out_free_queues;
3135         ctrl->ctrl.admin_tagset = &ctrl->admin_tag_set;
3136 
3137         ctrl->ctrl.fabrics_q = blk_mq_init_queue(&ctrl->admin_tag_set);
3138         if (IS_ERR(ctrl->ctrl.fabrics_q)) {
3139                 ret = PTR_ERR(ctrl->ctrl.fabrics_q);
3140                 goto out_free_admin_tag_set;
3141         }
3142 
3143         ctrl->ctrl.admin_q = blk_mq_init_queue(&ctrl->admin_tag_set);
3144         if (IS_ERR(ctrl->ctrl.admin_q)) {
3145                 ret = PTR_ERR(ctrl->ctrl.admin_q);
3146                 goto out_cleanup_fabrics_q;
3147         }
3148 
3149         /*
3150          * Would have been nice to init io queues tag set as well.
3151          * However, we require interaction from the controller
3152          * for max io queue count before we can do so.
3153          * Defer this to the connect path.
3154          */
3155 
3156         ret = nvme_init_ctrl(&ctrl->ctrl, dev, &nvme_fc_ctrl_ops, 0);
3157         if (ret)
3158                 goto out_cleanup_admin_q;
3159 
3160         /* at this point, teardown path changes to ref counting on nvme ctrl */
3161 
3162         spin_lock_irqsave(&rport->lock, flags);
3163         list_add_tail(&ctrl->ctrl_list, &rport->ctrl_list);
3164         spin_unlock_irqrestore(&rport->lock, flags);
3165 
3166         if (!nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_RESETTING) ||
3167             !nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_CONNECTING)) {
3168                 dev_err(ctrl->ctrl.device,
3169                         "NVME-FC{%d}: failed to init ctrl state\n", ctrl->cnum);
3170                 goto fail_ctrl;
3171         }
3172 
3173         nvme_get_ctrl(&ctrl->ctrl);
3174 
3175         if (!queue_delayed_work(nvme_wq, &ctrl->connect_work, 0)) {
3176                 nvme_put_ctrl(&ctrl->ctrl);
3177                 dev_err(ctrl->ctrl.device,
3178                         "NVME-FC{%d}: failed to schedule initial connect\n",
3179                         ctrl->cnum);
3180                 goto fail_ctrl;
3181         }
3182 
3183         flush_delayed_work(&ctrl->connect_work);
3184 
3185         dev_info(ctrl->ctrl.device,
3186                 "NVME-FC{%d}: new ctrl: NQN \"%s\"\n",
3187                 ctrl->cnum, ctrl->ctrl.opts->subsysnqn);
3188 
3189         return &ctrl->ctrl;
3190 
3191 fail_ctrl:
3192         nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_DELETING);
3193         cancel_work_sync(&ctrl->ctrl.reset_work);
3194         cancel_work_sync(&ctrl->err_work);
3195         cancel_delayed_work_sync(&ctrl->connect_work);
3196 
3197         ctrl->ctrl.opts = NULL;
3198 
3199         /* initiate nvme ctrl ref counting teardown */
3200         nvme_uninit_ctrl(&ctrl->ctrl);
3201 
3202         /* Remove core ctrl ref. */
3203         nvme_put_ctrl(&ctrl->ctrl);
3204 
3205         /* as we're past the point where we transition to the ref
3206          * counting teardown path, if we return a bad pointer here,
3207          * the calling routine, thinking it's prior to the
3208          * transition, will do an rport put. Since the teardown
3209          * path also does a rport put, we do an extra get here to
3210          * so proper order/teardown happens.
3211          */
3212         nvme_fc_rport_get(rport);
3213 
3214         return ERR_PTR(-EIO);
3215 
3216 out_cleanup_admin_q:
3217         blk_cleanup_queue(ctrl->ctrl.admin_q);
3218 out_cleanup_fabrics_q:
3219         blk_cleanup_queue(ctrl->ctrl.fabrics_q);
3220 out_free_admin_tag_set:
3221         blk_mq_free_tag_set(&ctrl->admin_tag_set);
3222 out_free_queues:
3223         kfree(ctrl->queues);
3224 out_free_ida:
3225         put_device(ctrl->dev);
3226         ida_simple_remove(&nvme_fc_ctrl_cnt, ctrl->cnum);
3227 out_free_ctrl:
3228         kfree(ctrl);
3229 out_fail:
3230         /* exit via here doesn't follow ctlr ref points */
3231         return ERR_PTR(ret);
3232 }
3233 
3234 
3235 struct nvmet_fc_traddr {
3236         u64     nn;
3237         u64     pn;
3238 };
3239 
3240 static int
3241 __nvme_fc_parse_u64(substring_t *sstr, u64 *val)
3242 {
3243         u64 token64;
3244 
3245         if (match_u64(sstr, &token64))
3246                 return -EINVAL;
3247         *val = token64;
3248 
3249         return 0;
3250 }
3251 
3252 /*
3253  * This routine validates and extracts the WWN's from the TRADDR string.
3254  * As kernel parsers need the 0x to determine number base, universally
3255  * build string to parse with 0x prefix before parsing name strings.
3256  */
3257 static int
3258 nvme_fc_parse_traddr(struct nvmet_fc_traddr *traddr, char *buf, size_t blen)
3259 {
3260         char name[2 + NVME_FC_TRADDR_HEXNAMELEN + 1];
3261         substring_t wwn = { name, &name[sizeof(name)-1] };
3262         int nnoffset, pnoffset;
3263 
3264         /* validate if string is one of the 2 allowed formats */
3265         if (strnlen(buf, blen) == NVME_FC_TRADDR_MAXLENGTH &&
3266                         !strncmp(buf, "nn-0x", NVME_FC_TRADDR_OXNNLEN) &&
3267                         !strncmp(&buf[NVME_FC_TRADDR_MAX_PN_OFFSET],
3268                                 "pn-0x", NVME_FC_TRADDR_OXNNLEN)) {
3269                 nnoffset = NVME_FC_TRADDR_OXNNLEN;
3270                 pnoffset = NVME_FC_TRADDR_MAX_PN_OFFSET +
3271                                                 NVME_FC_TRADDR_OXNNLEN;
3272         } else if ((strnlen(buf, blen) == NVME_FC_TRADDR_MINLENGTH &&
3273                         !strncmp(buf, "nn-", NVME_FC_TRADDR_NNLEN) &&
3274                         !strncmp(&buf[NVME_FC_TRADDR_MIN_PN_OFFSET],
3275                                 "pn-", NVME_FC_TRADDR_NNLEN))) {
3276                 nnoffset = NVME_FC_TRADDR_NNLEN;
3277                 pnoffset = NVME_FC_TRADDR_MIN_PN_OFFSET + NVME_FC_TRADDR_NNLEN;
3278         } else
3279                 goto out_einval;
3280 
3281         name[0] = '0';
3282         name[1] = 'x';
3283         name[2 + NVME_FC_TRADDR_HEXNAMELEN] = 0;
3284 
3285         memcpy(&name[2], &buf[nnoffset], NVME_FC_TRADDR_HEXNAMELEN);
3286         if (__nvme_fc_parse_u64(&wwn, &traddr->nn))
3287                 goto out_einval;
3288 
3289         memcpy(&name[2], &buf[pnoffset], NVME_FC_TRADDR_HEXNAMELEN);
3290         if (__nvme_fc_parse_u64(&wwn, &traddr->pn))
3291                 goto out_einval;
3292 
3293         return 0;
3294 
3295 out_einval:
3296         pr_warn("%s: bad traddr string\n", __func__);
3297         return -EINVAL;
3298 }
3299 
3300 static struct nvme_ctrl *
3301 nvme_fc_create_ctrl(struct device *dev, struct nvmf_ctrl_options *opts)
3302 {
3303         struct nvme_fc_lport *lport;
3304         struct nvme_fc_rport *rport;
3305         struct nvme_ctrl *ctrl;
3306         struct nvmet_fc_traddr laddr = { 0L, 0L };
3307         struct nvmet_fc_traddr raddr = { 0L, 0L };
3308         unsigned long flags;
3309         int ret;
3310 
3311         ret = nvme_fc_parse_traddr(&raddr, opts->traddr, NVMF_TRADDR_SIZE);
3312         if (ret || !raddr.nn || !raddr.pn)
3313                 return ERR_PTR(-EINVAL);
3314 
3315         ret = nvme_fc_parse_traddr(&laddr, opts->host_traddr, NVMF_TRADDR_SIZE);
3316         if (ret || !laddr.nn || !laddr.pn)
3317                 return ERR_PTR(-EINVAL);
3318 
3319         /* find the host and remote ports to connect together */
3320         spin_lock_irqsave(&nvme_fc_lock, flags);
3321         list_for_each_entry(lport, &nvme_fc_lport_list, port_list) {
3322                 if (lport->localport.node_name != laddr.nn ||
3323                     lport->localport.port_name != laddr.pn)
3324                         continue;
3325 
3326                 list_for_each_entry(rport, &lport->endp_list, endp_list) {
3327                         if (rport->remoteport.node_name != raddr.nn ||
3328                             rport->remoteport.port_name != raddr.pn)
3329                                 continue;
3330 
3331                         /* if fail to get reference fall through. Will error */
3332                         if (!nvme_fc_rport_get(rport))
3333                                 break;
3334 
3335                         spin_unlock_irqrestore(&nvme_fc_lock, flags);
3336 
3337                         ctrl = nvme_fc_init_ctrl(dev, opts, lport, rport);
3338                         if (IS_ERR(ctrl))
3339                                 nvme_fc_rport_put(rport);
3340                         return ctrl;
3341                 }
3342         }
3343         spin_unlock_irqrestore(&nvme_fc_lock, flags);
3344 
3345         pr_warn("%s: %s - %s combination not found\n",
3346                 __func__, opts->traddr, opts->host_traddr);
3347         return ERR_PTR(-ENOENT);
3348 }
3349 
3350 
3351 static struct nvmf_transport_ops nvme_fc_transport = {
3352         .name           = "fc",
3353         .module         = THIS_MODULE,
3354         .required_opts  = NVMF_OPT_TRADDR | NVMF_OPT_HOST_TRADDR,
3355         .allowed_opts   = NVMF_OPT_RECONNECT_DELAY | NVMF_OPT_CTRL_LOSS_TMO,
3356         .create_ctrl    = nvme_fc_create_ctrl,
3357 };
3358 
3359 /* Arbitrary successive failures max. With lots of subsystems could be high */
3360 #define DISCOVERY_MAX_FAIL      20
3361 
3362 static ssize_t nvme_fc_nvme_discovery_store(struct device *dev,
3363                 struct device_attribute *attr, const char *buf, size_t count)
3364 {
3365         unsigned long flags;
3366         LIST_HEAD(local_disc_list);
3367         struct nvme_fc_lport *lport;
3368         struct nvme_fc_rport *rport;
3369         int failcnt = 0;
3370 
3371         spin_lock_irqsave(&nvme_fc_lock, flags);
3372 restart:
3373         list_for_each_entry(lport, &nvme_fc_lport_list, port_list) {
3374                 list_for_each_entry(rport, &lport->endp_list, endp_list) {
3375                         if (!nvme_fc_lport_get(lport))
3376                                 continue;
3377                         if (!nvme_fc_rport_get(rport)) {
3378                                 /*
3379                                  * This is a temporary condition. Upon restart
3380                                  * this rport will be gone from the list.
3381                                  *
3382                                  * Revert the lport put and retry.  Anything
3383                                  * added to the list already will be skipped (as
3384                                  * they are no longer list_empty).  Loops should
3385                                  * resume at rports that were not yet seen.
3386                                  */
3387                                 nvme_fc_lport_put(lport);
3388 
3389                                 if (failcnt++ < DISCOVERY_MAX_FAIL)
3390                                         goto restart;
3391 
3392                                 pr_err("nvme_discovery: too many reference "
3393                                        "failures\n");
3394                                 goto process_local_list;
3395                         }
3396                         if (list_empty(&rport->disc_list))
3397                                 list_add_tail(&rport->disc_list,
3398                                               &local_disc_list);
3399                 }
3400         }
3401 
3402 process_local_list:
3403         while (!list_empty(&local_disc_list)) {
3404                 rport = list_first_entry(&local_disc_list,
3405                                          struct nvme_fc_rport, disc_list);
3406                 list_del_init(&rport->disc_list);
3407                 spin_unlock_irqrestore(&nvme_fc_lock, flags);
3408 
3409                 lport = rport->lport;
3410                 /* signal discovery. Won't hurt if it repeats */
3411                 nvme_fc_signal_discovery_scan(lport, rport);
3412                 nvme_fc_rport_put(rport);
3413                 nvme_fc_lport_put(lport);
3414 
3415                 spin_lock_irqsave(&nvme_fc_lock, flags);
3416         }
3417         spin_unlock_irqrestore(&nvme_fc_lock, flags);
3418 
3419         return count;
3420 }
3421 static DEVICE_ATTR(nvme_discovery, 0200, NULL, nvme_fc_nvme_discovery_store);
3422 
3423 static struct attribute *nvme_fc_attrs[] = {
3424         &dev_attr_nvme_discovery.attr,
3425         NULL
3426 };
3427 
3428 static struct attribute_group nvme_fc_attr_group = {
3429         .attrs = nvme_fc_attrs,
3430 };
3431 
3432 static const struct attribute_group *nvme_fc_attr_groups[] = {
3433         &nvme_fc_attr_group,
3434         NULL
3435 };
3436 
3437 static struct class fc_class = {
3438         .name = "fc",
3439         .dev_groups = nvme_fc_attr_groups,
3440         .owner = THIS_MODULE,
3441 };
3442 
3443 static int __init nvme_fc_init_module(void)
3444 {
3445         int ret;
3446 
3447         nvme_fc_wq = alloc_workqueue("nvme_fc_wq", WQ_MEM_RECLAIM, 0);
3448         if (!nvme_fc_wq)
3449                 return -ENOMEM;
3450 
3451         /*
3452          * NOTE:
3453          * It is expected that in the future the kernel will combine
3454          * the FC-isms that are currently under scsi and now being
3455          * added to by NVME into a new standalone FC class. The SCSI
3456          * and NVME protocols and their devices would be under this
3457          * new FC class.
3458          *
3459          * As we need something to post FC-specific udev events to,
3460          * specifically for nvme probe events, start by creating the
3461          * new device class.  When the new standalone FC class is
3462          * put in place, this code will move to a more generic
3463          * location for the class.
3464          */
3465         ret = class_register(&fc_class);
3466         if (ret) {
3467                 pr_err("couldn't register class fc\n");
3468                 goto out_destroy_wq;
3469         }
3470 
3471         /*
3472          * Create a device for the FC-centric udev events
3473          */
3474         fc_udev_device = device_create(&fc_class, NULL, MKDEV(0, 0), NULL,
3475                                 "fc_udev_device");
3476         if (IS_ERR(fc_udev_device)) {
3477                 pr_err("couldn't create fc_udev device!\n");
3478                 ret = PTR_ERR(fc_udev_device);
3479                 goto out_destroy_class;
3480         }
3481 
3482         ret = nvmf_register_transport(&nvme_fc_transport);
3483         if (ret)
3484                 goto out_destroy_device;
3485 
3486         return 0;
3487 
3488 out_destroy_device:
3489         device_destroy(&fc_class, MKDEV(0, 0));
3490 out_destroy_class:
3491         class_unregister(&fc_class);
3492 out_destroy_wq:
3493         destroy_workqueue(nvme_fc_wq);
3494 
3495         return ret;
3496 }
3497 
3498 static void
3499 nvme_fc_delete_controllers(struct nvme_fc_rport *rport)
3500 {
3501         struct nvme_fc_ctrl *ctrl;
3502 
3503         spin_lock(&rport->lock);
3504         list_for_each_entry(ctrl, &rport->ctrl_list, ctrl_list) {
3505                 dev_warn(ctrl->ctrl.device,
3506                         "NVME-FC{%d}: transport unloading: deleting ctrl\n",
3507                         ctrl->cnum);
3508                 nvme_delete_ctrl(&ctrl->ctrl);
3509         }
3510         spin_unlock(&rport->lock);
3511 }
3512 
3513 static void
3514 nvme_fc_cleanup_for_unload(void)
3515 {
3516         struct nvme_fc_lport *lport;
3517         struct nvme_fc_rport *rport;
3518 
3519         list_for_each_entry(lport, &nvme_fc_lport_list, port_list) {
3520                 list_for_each_entry(rport, &lport->endp_list, endp_list) {
3521                         nvme_fc_delete_controllers(rport);
3522                 }
3523         }
3524 }
3525 
3526 static void __exit nvme_fc_exit_module(void)
3527 {
3528         unsigned long flags;
3529         bool need_cleanup = false;
3530 
3531         spin_lock_irqsave(&nvme_fc_lock, flags);
3532         nvme_fc_waiting_to_unload = true;
3533         if (!list_empty(&nvme_fc_lport_list)) {
3534                 need_cleanup = true;
3535                 nvme_fc_cleanup_for_unload();
3536         }
3537         spin_unlock_irqrestore(&nvme_fc_lock, flags);
3538         if (need_cleanup) {
3539                 pr_info("%s: waiting for ctlr deletes\n", __func__);
3540                 wait_for_completion(&nvme_fc_unload_proceed);
3541                 pr_info("%s: ctrl deletes complete\n", __func__);
3542         }
3543 
3544         nvmf_unregister_transport(&nvme_fc_transport);
3545 
3546         ida_destroy(&nvme_fc_local_port_cnt);
3547         ida_destroy(&nvme_fc_ctrl_cnt);
3548 
3549         device_destroy(&fc_class, MKDEV(0, 0));
3550         class_unregister(&fc_class);
3551         destroy_workqueue(nvme_fc_wq);
3552 }
3553 
3554 module_init(nvme_fc_init_module);
3555 module_exit(nvme_fc_exit_module);
3556 
3557 MODULE_LICENSE("GPL v2");