drivers/nvme/target/fc.c

   1 /*
   2  * Copyright (c) 2016 Avago Technologies.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of version 2 of the GNU General Public License as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful.
   9  * ALL EXPRESS OR IMPLIED CONDITIONS, REPRESENTATIONS AND WARRANTIES,
  10  * INCLUDING ANY IMPLIED WARRANTY OF MERCHANTABILITY, FITNESS FOR A
  11  * PARTICULAR PURPOSE, OR NON-INFRINGEMENT, ARE DISCLAIMED, EXCEPT TO
  12  * THE EXTENT THAT SUCH DISCLAIMERS ARE HELD TO BE LEGALLY INVALID.
  13  * See the GNU General Public License for more details, a copy of which
  14  * can be found in the file COPYING included with this package
  15  *
  16  */
  17 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  18 #include <linux/module.h>
  19 #include <linux/slab.h>
  20 #include <linux/blk-mq.h>
  21 #include <linux/parser.h>
  22 #include <linux/random.h>
  23 #include <uapi/scsi/fc/fc_fs.h>
  24 #include <uapi/scsi/fc/fc_els.h>
  25
  26 #include "nvmet.h"
  27 #include <linux/nvme-fc-driver.h>
  28 #include <linux/nvme-fc.h>
  29
  30
  31 /* *************************** Data Structures/Defines ****************** */
  32
  33
  34 #define NVMET_LS_CTX_COUNT              4
  35
  36 /* for this implementation, assume small single frame rqst/rsp */
  37 #define NVME_FC_MAX_LS_BUFFER_SIZE              2048
  38
  39 struct nvmet_fc_tgtport;
  40 struct nvmet_fc_tgt_assoc;
  41
  42 struct nvmet_fc_ls_iod {
  43         struct nvmefc_tgt_ls_req        *lsreq;
  44         struct nvmefc_tgt_fcp_req       *fcpreq;        /* only if RS */
  45
  46         struct list_head                ls_list;        /* tgtport->ls_list */
  47
  48         struct nvmet_fc_tgtport         *tgtport;
  49         struct nvmet_fc_tgt_assoc       *assoc;
  50
  51         u8                              *rqstbuf;
  52         u8                              *rspbuf;
  53         u16                             rqstdatalen;
  54         dma_addr_t                      rspdma;
  55
  56         struct scatterlist              sg[2];
  57
  58         struct work_struct              work;
  59 } __aligned(sizeof(unsigned long long));
  60
  61 /* desired maximum for a single sequence - if sg list allows it */
  62 #define NVMET_FC_MAX_SEQ_LENGTH         (256 * 1024)
  63
  64 enum nvmet_fcp_datadir {
  65         NVMET_FCP_NODATA,
  66         NVMET_FCP_WRITE,
  67         NVMET_FCP_READ,
  68         NVMET_FCP_ABORTED,
  69 };
  70
  71 struct nvmet_fc_fcp_iod {
  72         struct nvmefc_tgt_fcp_req       *fcpreq;
  73
  74         struct nvme_fc_cmd_iu           cmdiubuf;
  75         struct nvme_fc_ersp_iu          rspiubuf;
  76         dma_addr_t                      rspdma;
  77         struct scatterlist              *next_sg;
  78         struct scatterlist              *data_sg;
  79         int                             data_sg_cnt;
  80         u32                             total_length;
  81         u32                             offset;
  82         enum nvmet_fcp_datadir          io_dir;
  83         bool                            active;
  84         bool                            abort;
  85         bool                            aborted;
  86         bool                            writedataactive;
  87         spinlock_t                      flock;
  88
  89         struct nvmet_req                req;
  90         struct work_struct              work;
  91         struct work_struct              done_work;
  92
  93         struct nvmet_fc_tgtport         *tgtport;
  94         struct nvmet_fc_tgt_queue       *queue;
  95
  96         struct list_head                fcp_list;       /* tgtport->fcp_list */
  97 };
  98
  99 struct nvmet_fc_tgtport {
 100
 101         struct nvmet_fc_target_port     fc_target_port;
 102
 103         struct list_head                tgt_list; /* nvmet_fc_target_list */
 104         struct device                   *dev;   /* dev for dma mapping */
 105         struct nvmet_fc_target_template *ops;
 106
 107         struct nvmet_fc_ls_iod          *iod;
 108         spinlock_t                      lock;
 109         struct list_head                ls_list;
 110         struct list_head                ls_busylist;
 111         struct list_head                assoc_list;
 112         struct ida                      assoc_cnt;
 113         struct nvmet_port               *port;
 114         struct kref                     ref;
 115         u32                             max_sg_cnt;
 116 };
 117
 118 struct nvmet_fc_defer_fcp_req {
 119         struct list_head                req_list;
 120         struct nvmefc_tgt_fcp_req       *fcp_req;
 121 };
 122
 123 struct nvmet_fc_tgt_queue {
 124         bool                            ninetypercent;
 125         u16                             qid;
 126         u16                             sqsize;
 127         u16                             ersp_ratio;
 128         __le16                          sqhd;
 129         int                             cpu;
 130         atomic_t                        connected;
 131         atomic_t                        sqtail;
 132         atomic_t                        zrspcnt;
 133         atomic_t                        rsn;
 134         spinlock_t                      qlock;
 135         struct nvmet_port               *port;
 136         struct nvmet_cq                 nvme_cq;
 137         struct nvmet_sq                 nvme_sq;
 138         struct nvmet_fc_tgt_assoc       *assoc;
 139         struct nvmet_fc_fcp_iod         *fod;           /* array of fcp_iods */
 140         struct list_head                fod_list;
 141         struct list_head                pending_cmd_list;
 142         struct list_head                avail_defer_list;
 143         struct workqueue_struct         *work_q;
 144         struct kref                     ref;
 145 } __aligned(sizeof(unsigned long long));
 146
 147 struct nvmet_fc_tgt_assoc {
 148         u64                             association_id;
 149         u32                             a_id;
 150         struct nvmet_fc_tgtport         *tgtport;
 151         struct list_head                a_list;
 152         struct nvmet_fc_tgt_queue       *queues[NVMET_NR_QUEUES + 1];
 153         struct kref                     ref;
 154 };
 155
 156
 157 static inline int
 158 nvmet_fc_iodnum(struct nvmet_fc_ls_iod *iodptr)
 159 {
 160         return (iodptr - iodptr->tgtport->iod);
 161 }
 162
 163 static inline int
 164 nvmet_fc_fodnum(struct nvmet_fc_fcp_iod *fodptr)
 165 {
 166         return (fodptr - fodptr->queue->fod);
 167 }
 168
 169
 170 /*
 171  * Association and Connection IDs:
 172  *
 173  * Association ID will have random number in upper 6 bytes and zero
 174  *   in lower 2 bytes
 175  *
 176  * Connection IDs will be Association ID with QID or'd in lower 2 bytes
 177  *
 178  * note: Association ID = Connection ID for queue 0
 179  */
 180 #define BYTES_FOR_QID                   sizeof(u16)
 181 #define BYTES_FOR_QID_SHIFT             (BYTES_FOR_QID * 8)
 182 #define NVMET_FC_QUEUEID_MASK           ((u64)((1 << BYTES_FOR_QID_SHIFT) - 1))
 183
 184 static inline u64
 185 nvmet_fc_makeconnid(struct nvmet_fc_tgt_assoc *assoc, u16 qid)
 186 {
 187         return (assoc->association_id | qid);
 188 }
 189
 190 static inline u64
 191 nvmet_fc_getassociationid(u64 connectionid)
 192 {
 193         return connectionid & ~NVMET_FC_QUEUEID_MASK;
 194 }
 195
 196 static inline u16
 197 nvmet_fc_getqueueid(u64 connectionid)
 198 {
 199         return (u16)(connectionid & NVMET_FC_QUEUEID_MASK);
 200 }
 201
 202 static inline struct nvmet_fc_tgtport *
 203 targetport_to_tgtport(struct nvmet_fc_target_port *targetport)
 204 {
 205         return container_of(targetport, struct nvmet_fc_tgtport,
 206                                  fc_target_port);
 207 }
 208
 209 static inline struct nvmet_fc_fcp_iod *
 210 nvmet_req_to_fod(struct nvmet_req *nvme_req)
 211 {
 212         return container_of(nvme_req, struct nvmet_fc_fcp_iod, req);
 213 }
 214
 215
 216 /* *************************** Globals **************************** */
 217
 218
 219 static DEFINE_SPINLOCK(nvmet_fc_tgtlock);
 220
 221 static LIST_HEAD(nvmet_fc_target_list);
 222 static DEFINE_IDA(nvmet_fc_tgtport_cnt);
 223
 224
 225 static void nvmet_fc_handle_ls_rqst_work(struct work_struct *work);
 226 static void nvmet_fc_handle_fcp_rqst_work(struct work_struct *work);
 227 static void nvmet_fc_fcp_rqst_op_done_work(struct work_struct *work);
 228 static void nvmet_fc_tgt_a_put(struct nvmet_fc_tgt_assoc *assoc);
 229 static int nvmet_fc_tgt_a_get(struct nvmet_fc_tgt_assoc *assoc);
 230 static void nvmet_fc_tgt_q_put(struct nvmet_fc_tgt_queue *queue);
 231 static int nvmet_fc_tgt_q_get(struct nvmet_fc_tgt_queue *queue);
 232 static void nvmet_fc_tgtport_put(struct nvmet_fc_tgtport *tgtport);
 233 static int nvmet_fc_tgtport_get(struct nvmet_fc_tgtport *tgtport);
 234 static void nvmet_fc_handle_fcp_rqst(struct nvmet_fc_tgtport *tgtport,
 235                                         struct nvmet_fc_fcp_iod *fod);
 236
 237
 238 /* *********************** FC-NVME DMA Handling **************************** */
 239
 240 /*
 241  * The fcloop device passes in a NULL device pointer. Real LLD's will
 242  * pass in a valid device pointer. If NULL is passed to the dma mapping
 243  * routines, depending on the platform, it may or may not succeed, and
 244  * may crash.
 245  *
 246  * As such:
 247  * Wrapper all the dma routines and check the dev pointer.
 248  *
 249  * If simple mappings (return just a dma address, we'll noop them,
 250  * returning a dma address of 0.
 251  *
 252  * On more complex mappings (dma_map_sg), a pseudo routine fills
 253  * in the scatter list, setting all dma addresses to 0.
 254  */
 255
 256 static inline dma_addr_t
 257 fc_dma_map_single(struct device *dev, void *ptr, size_t size,
 258                 enum dma_data_direction dir)
 259 {
 260         return dev ? dma_map_single(dev, ptr, size, dir) : (dma_addr_t)0L;
 261 }
 262
 263 static inline int
 264 fc_dma_mapping_error(struct device *dev, dma_addr_t dma_addr)
 265 {
 266         return dev ? dma_mapping_error(dev, dma_addr) : 0;
 267 }
 268
 269 static inline void
 270 fc_dma_unmap_single(struct device *dev, dma_addr_t addr, size_t size,
 271         enum dma_data_direction dir)
 272 {
 273         if (dev)
 274                 dma_unmap_single(dev, addr, size, dir);
 275 }
 276
 277 static inline void
 278 fc_dma_sync_single_for_cpu(struct device *dev, dma_addr_t addr, size_t size,
 279                 enum dma_data_direction dir)
 280 {
 281         if (dev)
 282                 dma_sync_single_for_cpu(dev, addr, size, dir);
 283 }
 284
 285 static inline void
 286 fc_dma_sync_single_for_device(struct device *dev, dma_addr_t addr, size_t size,
 287                 enum dma_data_direction dir)
 288 {
 289         if (dev)
 290                 dma_sync_single_for_device(dev, addr, size, dir);
 291 }
 292
 293 /* pseudo dma_map_sg call */
 294 static int
 295 fc_map_sg(struct scatterlist *sg, int nents)
 296 {
 297         struct scatterlist *s;
 298         int i;
 299
 300         WARN_ON(nents == 0 || sg[0].length == 0);
 301
 302         for_each_sg(sg, s, nents, i) {
 303                 s->dma_address = 0L;
 304 #ifdef CONFIG_NEED_SG_DMA_LENGTH
 305                 s->dma_length = s->length;
 306 #endif
 307         }
 308         return nents;
 309 }
 310
 311 static inline int
 312 fc_dma_map_sg(struct device *dev, struct scatterlist *sg, int nents,
 313                 enum dma_data_direction dir)
 314 {
 315         return dev ? dma_map_sg(dev, sg, nents, dir) : fc_map_sg(sg, nents);
 316 }
 317
 318 static inline void
 319 fc_dma_unmap_sg(struct device *dev, struct scatterlist *sg, int nents,
 320                 enum dma_data_direction dir)
 321 {
 322         if (dev)
 323                 dma_unmap_sg(dev, sg, nents, dir);
 324 }
 325
 326
 327 /* *********************** FC-NVME Port Management ************************ */
 328
 329
 330 static int
 331 nvmet_fc_alloc_ls_iodlist(struct nvmet_fc_tgtport *tgtport)
 332 {
 333         struct nvmet_fc_ls_iod *iod;
 334         int i;
 335
 336         iod = kcalloc(NVMET_LS_CTX_COUNT, sizeof(struct nvmet_fc_ls_iod),
 337                         GFP_KERNEL);
 338         if (!iod)
 339                 return -ENOMEM;
 340
 341         tgtport->iod = iod;
 342
 343         for (i = 0; i < NVMET_LS_CTX_COUNT; iod++, i++) {
 344                 INIT_WORK(&iod->work, nvmet_fc_handle_ls_rqst_work);
 345                 iod->tgtport = tgtport;
 346                 list_add_tail(&iod->ls_list, &tgtport->ls_list);
 347
 348                 iod->rqstbuf = kcalloc(2, NVME_FC_MAX_LS_BUFFER_SIZE,
 349                         GFP_KERNEL);
 350                 if (!iod->rqstbuf)
 351                         goto out_fail;
 352
 353                 iod->rspbuf = iod->rqstbuf + NVME_FC_MAX_LS_BUFFER_SIZE;
 354
 355                 iod->rspdma = fc_dma_map_single(tgtport->dev, iod->rspbuf,
 356                                                 NVME_FC_MAX_LS_BUFFER_SIZE,
 357                                                 DMA_TO_DEVICE);
 358                 if (fc_dma_mapping_error(tgtport->dev, iod->rspdma))
 359                         goto out_fail;
 360         }
 361
 362         return 0;
 363
 364 out_fail:
 365         kfree(iod->rqstbuf);
 366         list_del(&iod->ls_list);
 367         for (iod--, i--; i >= 0; iod--, i--) {
 368                 fc_dma_unmap_single(tgtport->dev, iod->rspdma,
 369                                 NVME_FC_MAX_LS_BUFFER_SIZE, DMA_TO_DEVICE);
 370                 kfree(iod->rqstbuf);
 371                 list_del(&iod->ls_list);
 372         }
 373
 374         kfree(iod);
 375
 376         return -EFAULT;
 377 }
 378
 379 static void
 380 nvmet_fc_free_ls_iodlist(struct nvmet_fc_tgtport *tgtport)
 381 {
 382         struct nvmet_fc_ls_iod *iod = tgtport->iod;
 383         int i;
 384
 385         for (i = 0; i < NVMET_LS_CTX_COUNT; iod++, i++) {
 386                 fc_dma_unmap_single(tgtport->dev,
 387                                 iod->rspdma, NVME_FC_MAX_LS_BUFFER_SIZE,
 388                                 DMA_TO_DEVICE);
 389                 kfree(iod->rqstbuf);
 390                 list_del(&iod->ls_list);
 391         }
 392         kfree(tgtport->iod);
 393 }
 394
 395 static struct nvmet_fc_ls_iod *
 396 nvmet_fc_alloc_ls_iod(struct nvmet_fc_tgtport *tgtport)
 397 {
 398         struct nvmet_fc_ls_iod *iod;
 399         unsigned long flags;
 400
 401         spin_lock_irqsave(&tgtport->lock, flags);
 402         iod = list_first_entry_or_null(&tgtport->ls_list,
 403                                         struct nvmet_fc_ls_iod, ls_list);
 404         if (iod)
 405                 list_move_tail(&iod->ls_list, &tgtport->ls_busylist);
 406         spin_unlock_irqrestore(&tgtport->lock, flags);
 407         return iod;
 408 }
 409
 410
 411 static void
 412 nvmet_fc_free_ls_iod(struct nvmet_fc_tgtport *tgtport,
 413                         struct nvmet_fc_ls_iod *iod)
 414 {
 415         unsigned long flags;
 416
 417         spin_lock_irqsave(&tgtport->lock, flags);
 418         list_move(&iod->ls_list, &tgtport->ls_list);
 419         spin_unlock_irqrestore(&tgtport->lock, flags);
 420 }
 421
 422 static void
 423 nvmet_fc_prep_fcp_iodlist(struct nvmet_fc_tgtport *tgtport,
 424                                 struct nvmet_fc_tgt_queue *queue)
 425 {
 426         struct nvmet_fc_fcp_iod *fod = queue->fod;
 427         int i;
 428
 429         for (i = 0; i < queue->sqsize; fod++, i++) {
 430                 INIT_WORK(&fod->work, nvmet_fc_handle_fcp_rqst_work);
 431                 INIT_WORK(&fod->done_work, nvmet_fc_fcp_rqst_op_done_work);
 432                 fod->tgtport = tgtport;
 433                 fod->queue = queue;
 434                 fod->active = false;
 435                 fod->abort = false;
 436                 fod->aborted = false;
 437                 fod->fcpreq = NULL;
 438                 list_add_tail(&fod->fcp_list, &queue->fod_list);
 439                 spin_lock_init(&fod->flock);
 440
 441                 fod->rspdma = fc_dma_map_single(tgtport->dev, &fod->rspiubuf,
 442                                         sizeof(fod->rspiubuf), DMA_TO_DEVICE);
 443                 if (fc_dma_mapping_error(tgtport->dev, fod->rspdma)) {
 444                         list_del(&fod->fcp_list);
 445                         for (fod--, i--; i >= 0; fod--, i--) {
 446                                 fc_dma_unmap_single(tgtport->dev, fod->rspdma,
 447                                                 sizeof(fod->rspiubuf),
 448                                                 DMA_TO_DEVICE);
 449                                 fod->rspdma = 0L;
 450                                 list_del(&fod->fcp_list);
 451                         }
 452
 453                         return;
 454                 }
 455         }
 456 }
 457
 458 static void
 459 nvmet_fc_destroy_fcp_iodlist(struct nvmet_fc_tgtport *tgtport,
 460                                 struct nvmet_fc_tgt_queue *queue)
 461 {
 462         struct nvmet_fc_fcp_iod *fod = queue->fod;
 463         int i;
 464
 465         for (i = 0; i < queue->sqsize; fod++, i++) {
 466                 if (fod->rspdma)
 467                         fc_dma_unmap_single(tgtport->dev, fod->rspdma,
 468                                 sizeof(fod->rspiubuf), DMA_TO_DEVICE);
 469         }
 470 }
 471
 472 static struct nvmet_fc_fcp_iod *
 473 nvmet_fc_alloc_fcp_iod(struct nvmet_fc_tgt_queue *queue)
 474 {
 475         struct nvmet_fc_fcp_iod *fod;
 476
 477         lockdep_assert_held(&queue->qlock);
 478
 479         fod = list_first_entry_or_null(&queue->fod_list,
 480                                         struct nvmet_fc_fcp_iod, fcp_list);
 481         if (fod) {
 482                 list_del(&fod->fcp_list);
 483                 fod->active = true;
 484                 /*
 485                  * no queue reference is taken, as it was taken by the
 486                  * queue lookup just prior to the allocation. The iod
 487                  * will "inherit" that reference.
 488                  */
 489         }
 490         return fod;
 491 }
 492
 493
 494 static void
 495 nvmet_fc_queue_fcp_req(struct nvmet_fc_tgtport *tgtport,
 496                        struct nvmet_fc_tgt_queue *queue,
 497                        struct nvmefc_tgt_fcp_req *fcpreq)
 498 {
 499         struct nvmet_fc_fcp_iod *fod = fcpreq->nvmet_fc_private;
 500
 501         /*
 502          * put all admin cmds on hw queue id 0. All io commands go to
 503          * the respective hw queue based on a modulo basis
 504          */
 505         fcpreq->hwqid = queue->qid ?
 506                         ((queue->qid - 1) % tgtport->ops->max_hw_queues) : 0;
 507
 508         if (tgtport->ops->target_features & NVMET_FCTGTFEAT_CMD_IN_ISR)
 509                 queue_work_on(queue->cpu, queue->work_q, &fod->work);
 510         else
 511                 nvmet_fc_handle_fcp_rqst(tgtport, fod);
 512 }
 513
 514 static void
 515 nvmet_fc_free_fcp_iod(struct nvmet_fc_tgt_queue *queue,
 516                         struct nvmet_fc_fcp_iod *fod)
 517 {
 518         struct nvmefc_tgt_fcp_req *fcpreq = fod->fcpreq;
 519         struct nvmet_fc_tgtport *tgtport = fod->tgtport;
 520         struct nvmet_fc_defer_fcp_req *deferfcp;
 521         unsigned long flags;
 522
 523         fc_dma_sync_single_for_cpu(tgtport->dev, fod->rspdma,
 524                                 sizeof(fod->rspiubuf), DMA_TO_DEVICE);
 525
 526         fcpreq->nvmet_fc_private = NULL;
 527
 528         fod->active = false;
 529         fod->abort = false;
 530         fod->aborted = false;
 531         fod->writedataactive = false;
 532         fod->fcpreq = NULL;
 533
 534         tgtport->ops->fcp_req_release(&tgtport->fc_target_port, fcpreq);
 535
 536         /* release the queue lookup reference on the completed IO */
 537         nvmet_fc_tgt_q_put(queue);
 538
 539         spin_lock_irqsave(&queue->qlock, flags);
 540         deferfcp = list_first_entry_or_null(&queue->pending_cmd_list,
 541                                 struct nvmet_fc_defer_fcp_req, req_list);
 542         if (!deferfcp) {
 543                 list_add_tail(&fod->fcp_list, &fod->queue->fod_list);
 544                 spin_unlock_irqrestore(&queue->qlock, flags);
 545                 return;
 546         }
 547
 548         /* Re-use the fod for the next pending cmd that was deferred */
 549         list_del(&deferfcp->req_list);
 550
 551         fcpreq = deferfcp->fcp_req;
 552
 553         /* deferfcp can be reused for another IO at a later date */
 554         list_add_tail(&deferfcp->req_list, &queue->avail_defer_list);
 555
 556         spin_unlock_irqrestore(&queue->qlock, flags);
 557
 558         /* Save NVME CMD IO in fod */
 559         memcpy(&fod->cmdiubuf, fcpreq->rspaddr, fcpreq->rsplen);
 560
 561         /* Setup new fcpreq to be processed */
 562         fcpreq->rspaddr = NULL;
 563         fcpreq->rsplen  = 0;
 564         fcpreq->nvmet_fc_private = fod;
 565         fod->fcpreq = fcpreq;
 566         fod->active = true;
 567
 568         /* inform LLDD IO is now being processed */
 569         tgtport->ops->defer_rcv(&tgtport->fc_target_port, fcpreq);
 570
 571         /* Submit deferred IO for processing */
 572         nvmet_fc_queue_fcp_req(tgtport, queue, fcpreq);
 573
 574         /*
 575          * Leave the queue lookup get reference taken when
 576          * fod was originally allocated.
 577          */
 578 }
 579
 580 static int
 581 nvmet_fc_queue_to_cpu(struct nvmet_fc_tgtport *tgtport, int qid)
 582 {
 583         int cpu, idx, cnt;
 584
 585         if (tgtport->ops->max_hw_queues == 1)
 586                 return WORK_CPU_UNBOUND;
 587
 588         /* Simple cpu selection based on qid modulo active cpu count */
 589         idx = !qid ? 0 : (qid - 1) % num_active_cpus();
 590
 591         /* find the n'th active cpu */
 592         for (cpu = 0, cnt = 0; ; ) {
 593                 if (cpu_active(cpu)) {
 594                         if (cnt == idx)
 595                                 break;
 596                         cnt++;
 597                 }
 598                 cpu = (cpu + 1) % num_possible_cpus();
 599         }
 600
 601         return cpu;
 602 }
 603
 604 static struct nvmet_fc_tgt_queue *
 605 nvmet_fc_alloc_target_queue(struct nvmet_fc_tgt_assoc *assoc,
 606                         u16 qid, u16 sqsize)
 607 {
 608         struct nvmet_fc_tgt_queue *queue;
 609         unsigned long flags;
 610         int ret;
 611
 612         if (qid > NVMET_NR_QUEUES)
 613                 return NULL;
 614
 615         queue = kzalloc((sizeof(*queue) +
 616                                 (sizeof(struct nvmet_fc_fcp_iod) * sqsize)),
 617                                 GFP_KERNEL);
 618         if (!queue)
 619                 return NULL;
 620
 621         if (!nvmet_fc_tgt_a_get(assoc))
 622                 goto out_free_queue;
 623
 624         queue->work_q = alloc_workqueue("ntfc%d.%d.%d", 0, 0,
 625                                 assoc->tgtport->fc_target_port.port_num,
 626                                 assoc->a_id, qid);
 627         if (!queue->work_q)
 628                 goto out_a_put;
 629
 630         queue->fod = (struct nvmet_fc_fcp_iod *)&queue[1];
 631         queue->qid = qid;
 632         queue->sqsize = sqsize;
 633         queue->assoc = assoc;
 634         queue->port = assoc->tgtport->port;
 635         queue->cpu = nvmet_fc_queue_to_cpu(assoc->tgtport, qid);
 636         INIT_LIST_HEAD(&queue->fod_list);
 637         INIT_LIST_HEAD(&queue->avail_defer_list);
 638         INIT_LIST_HEAD(&queue->pending_cmd_list);
 639         atomic_set(&queue->connected, 0);
 640         atomic_set(&queue->sqtail, 0);
 641         atomic_set(&queue->rsn, 1);
 642         atomic_set(&queue->zrspcnt, 0);
 643         spin_lock_init(&queue->qlock);
 644         kref_init(&queue->ref);
 645
 646         nvmet_fc_prep_fcp_iodlist(assoc->tgtport, queue);
 647
 648         ret = nvmet_sq_init(&queue->nvme_sq);
 649         if (ret)
 650                 goto out_fail_iodlist;
 651
 652         WARN_ON(assoc->queues[qid]);
 653         spin_lock_irqsave(&assoc->tgtport->lock, flags);
 654         assoc->queues[qid] = queue;
 655         spin_unlock_irqrestore(&assoc->tgtport->lock, flags);
 656
 657         return queue;
 658
 659 out_fail_iodlist:
 660         nvmet_fc_destroy_fcp_iodlist(assoc->tgtport, queue);
 661         destroy_workqueue(queue->work_q);
 662 out_a_put:
 663         nvmet_fc_tgt_a_put(assoc);
 664 out_free_queue:
 665         kfree(queue);
 666         return NULL;
 667 }
 668
 669
 670 static void
 671 nvmet_fc_tgt_queue_free(struct kref *ref)
 672 {
 673         struct nvmet_fc_tgt_queue *queue =
 674                 container_of(ref, struct nvmet_fc_tgt_queue, ref);
 675         unsigned long flags;
 676
 677         spin_lock_irqsave(&queue->assoc->tgtport->lock, flags);
 678         queue->assoc->queues[queue->qid] = NULL;
 679         spin_unlock_irqrestore(&queue->assoc->tgtport->lock, flags);
 680
 681         nvmet_fc_destroy_fcp_iodlist(queue->assoc->tgtport, queue);
 682
 683         nvmet_fc_tgt_a_put(queue->assoc);
 684
 685         destroy_workqueue(queue->work_q);
 686
 687         kfree(queue);
 688 }
 689
 690 static void
 691 nvmet_fc_tgt_q_put(struct nvmet_fc_tgt_queue *queue)
 692 {
 693         kref_put(&queue->ref, nvmet_fc_tgt_queue_free);
 694 }
 695
 696 static int
 697 nvmet_fc_tgt_q_get(struct nvmet_fc_tgt_queue *queue)
 698 {
 699         return kref_get_unless_zero(&queue->ref);
 700 }
 701
 702
 703 static void
 704 nvmet_fc_delete_target_queue(struct nvmet_fc_tgt_queue *queue)
 705 {
 706         struct nvmet_fc_tgtport *tgtport = queue->assoc->tgtport;
 707         struct nvmet_fc_fcp_iod *fod = queue->fod;
 708         struct nvmet_fc_defer_fcp_req *deferfcp, *tempptr;
 709         unsigned long flags;
 710         int i, writedataactive;
 711         bool disconnect;
 712
 713         disconnect = atomic_xchg(&queue->connected, 0);
 714
 715         spin_lock_irqsave(&queue->qlock, flags);
 716         /* about outstanding io's */
 717         for (i = 0; i < queue->sqsize; fod++, i++) {
 718                 if (fod->active) {
 719                         spin_lock(&fod->flock);
 720                         fod->abort = true;
 721                         writedataactive = fod->writedataactive;
 722                         spin_unlock(&fod->flock);
 723                         /*
 724                          * only call lldd abort routine if waiting for
 725                          * writedata. other outstanding ops should finish
 726                          * on their own.
 727                          */
 728                         if (writedataactive) {
 729                                 spin_lock(&fod->flock);
 730                                 fod->aborted = true;
 731                                 spin_unlock(&fod->flock);
 732                                 tgtport->ops->fcp_abort(
 733                                         &tgtport->fc_target_port, fod->fcpreq);
 734                         }
 735                 }
 736         }
 737
 738         /* Cleanup defer'ed IOs in queue */
 739         list_for_each_entry_safe(deferfcp, tempptr, &queue->avail_defer_list,
 740                                 req_list) {
 741                 list_del(&deferfcp->req_list);
 742                 kfree(deferfcp);
 743         }
 744
 745         for (;;) {
 746                 deferfcp = list_first_entry_or_null(&queue->pending_cmd_list,
 747                                 struct nvmet_fc_defer_fcp_req, req_list);
 748                 if (!deferfcp)
 749                         break;
 750
 751                 list_del(&deferfcp->req_list);
 752                 spin_unlock_irqrestore(&queue->qlock, flags);
 753
 754                 tgtport->ops->defer_rcv(&tgtport->fc_target_port,
 755                                 deferfcp->fcp_req);
 756
 757                 tgtport->ops->fcp_abort(&tgtport->fc_target_port,
 758                                 deferfcp->fcp_req);
 759
 760                 tgtport->ops->fcp_req_release(&tgtport->fc_target_port,
 761                                 deferfcp->fcp_req);
 762
 763                 /* release the queue lookup reference */
 764                 nvmet_fc_tgt_q_put(queue);
 765
 766                 kfree(deferfcp);
 767
 768                 spin_lock_irqsave(&queue->qlock, flags);
 769         }
 770         spin_unlock_irqrestore(&queue->qlock, flags);
 771
 772         flush_workqueue(queue->work_q);
 773
 774         if (disconnect)
 775                 nvmet_sq_destroy(&queue->nvme_sq);
 776
 777         nvmet_fc_tgt_q_put(queue);
 778 }
 779
 780 static struct nvmet_fc_tgt_queue *
 781 nvmet_fc_find_target_queue(struct nvmet_fc_tgtport *tgtport,
 782                                 u64 connection_id)
 783 {
 784         struct nvmet_fc_tgt_assoc *assoc;
 785         struct nvmet_fc_tgt_queue *queue;
 786         u64 association_id = nvmet_fc_getassociationid(connection_id);
 787         u16 qid = nvmet_fc_getqueueid(connection_id);
 788         unsigned long flags;
 789
 790         if (qid > NVMET_NR_QUEUES)
 791                 return NULL;
 792
 793         spin_lock_irqsave(&tgtport->lock, flags);
 794         list_for_each_entry(assoc, &tgtport->assoc_list, a_list) {
 795                 if (association_id == assoc->association_id) {
 796                         queue = assoc->queues[qid];
 797                         if (queue &&
 798                             (!atomic_read(&queue->connected) ||
 799                              !nvmet_fc_tgt_q_get(queue)))
 800                                 queue = NULL;
 801                         spin_unlock_irqrestore(&tgtport->lock, flags);
 802                         return queue;
 803                 }
 804         }
 805         spin_unlock_irqrestore(&tgtport->lock, flags);
 806         return NULL;
 807 }
 808
 809 static struct nvmet_fc_tgt_assoc *
 810 nvmet_fc_alloc_target_assoc(struct nvmet_fc_tgtport *tgtport)
 811 {
 812         struct nvmet_fc_tgt_assoc *assoc, *tmpassoc;
 813         unsigned long flags;
 814         u64 ran;
 815         int idx;
 816         bool needrandom = true;
 817
 818         assoc = kzalloc(sizeof(*assoc), GFP_KERNEL);
 819         if (!assoc)
 820                 return NULL;
 821
 822         idx = ida_simple_get(&tgtport->assoc_cnt, 0, 0, GFP_KERNEL);
 823         if (idx < 0)
 824                 goto out_free_assoc;
 825
 826         if (!nvmet_fc_tgtport_get(tgtport))
 827                 goto out_ida_put;
 828
 829         assoc->tgtport = tgtport;
 830         assoc->a_id = idx;
 831         INIT_LIST_HEAD(&assoc->a_list);
 832         kref_init(&assoc->ref);
 833
 834         while (needrandom) {
 835                 get_random_bytes(&ran, sizeof(ran) - BYTES_FOR_QID);
 836                 ran = ran << BYTES_FOR_QID_SHIFT;
 837
 838                 spin_lock_irqsave(&tgtport->lock, flags);
 839                 needrandom = false;
 840                 list_for_each_entry(tmpassoc, &tgtport->assoc_list, a_list)
 841                         if (ran == tmpassoc->association_id) {
 842                                 needrandom = true;
 843                                 break;
 844                         }
 845                 if (!needrandom) {
 846                         assoc->association_id = ran;
 847                         list_add_tail(&assoc->a_list, &tgtport->assoc_list);
 848                 }
 849                 spin_unlock_irqrestore(&tgtport->lock, flags);
 850         }
 851
 852         return assoc;
 853
 854 out_ida_put:
 855         ida_simple_remove(&tgtport->assoc_cnt, idx);
 856 out_free_assoc:
 857         kfree(assoc);
 858         return NULL;
 859 }
 860
 861 static void
 862 nvmet_fc_target_assoc_free(struct kref *ref)
 863 {
 864         struct nvmet_fc_tgt_assoc *assoc =
 865                 container_of(ref, struct nvmet_fc_tgt_assoc, ref);
 866         struct nvmet_fc_tgtport *tgtport = assoc->tgtport;
 867         unsigned long flags;
 868
 869         spin_lock_irqsave(&tgtport->lock, flags);
 870         list_del(&assoc->a_list);
 871         spin_unlock_irqrestore(&tgtport->lock, flags);
 872         ida_simple_remove(&tgtport->assoc_cnt, assoc->a_id);
 873         kfree(assoc);
 874         nvmet_fc_tgtport_put(tgtport);
 875 }
 876
 877 static void
 878 nvmet_fc_tgt_a_put(struct nvmet_fc_tgt_assoc *assoc)
 879 {
 880         kref_put(&assoc->ref, nvmet_fc_target_assoc_free);
 881 }
 882
 883 static int
 884 nvmet_fc_tgt_a_get(struct nvmet_fc_tgt_assoc *assoc)
 885 {
 886         return kref_get_unless_zero(&assoc->ref);
 887 }
 888
 889 static void
 890 nvmet_fc_delete_target_assoc(struct nvmet_fc_tgt_assoc *assoc)
 891 {
 892         struct nvmet_fc_tgtport *tgtport = assoc->tgtport;
 893         struct nvmet_fc_tgt_queue *queue;
 894         unsigned long flags;
 895         int i;
 896
 897         spin_lock_irqsave(&tgtport->lock, flags);
 898         for (i = NVMET_NR_QUEUES; i >= 0; i--) {
 899                 queue = assoc->queues[i];
 900                 if (queue) {
 901                         if (!nvmet_fc_tgt_q_get(queue))
 902                                 continue;
 903                         spin_unlock_irqrestore(&tgtport->lock, flags);
 904                         nvmet_fc_delete_target_queue(queue);
 905                         nvmet_fc_tgt_q_put(queue);
 906                         spin_lock_irqsave(&tgtport->lock, flags);
 907                 }
 908         }
 909         spin_unlock_irqrestore(&tgtport->lock, flags);
 910
 911         nvmet_fc_tgt_a_put(assoc);
 912 }
 913
 914 static struct nvmet_fc_tgt_assoc *
 915 nvmet_fc_find_target_assoc(struct nvmet_fc_tgtport *tgtport,
 916                                 u64 association_id)
 917 {
 918         struct nvmet_fc_tgt_assoc *assoc;
 919         struct nvmet_fc_tgt_assoc *ret = NULL;
 920         unsigned long flags;
 921
 922         spin_lock_irqsave(&tgtport->lock, flags);
 923         list_for_each_entry(assoc, &tgtport->assoc_list, a_list) {
 924                 if (association_id == assoc->association_id) {
 925                         ret = assoc;
 926                         nvmet_fc_tgt_a_get(assoc);
 927                         break;
 928                 }
 929         }
 930         spin_unlock_irqrestore(&tgtport->lock, flags);
 931
 932         return ret;
 933 }
 934
 935
 936 /**
 937  * nvme_fc_register_targetport - transport entry point called by an
 938  *                              LLDD to register the existence of a local
 939  *                              NVME subystem FC port.
 940  * @pinfo:     pointer to information about the port to be registered
 941  * @template:  LLDD entrypoints and operational parameters for the port
 942  * @dev:       physical hardware device node port corresponds to. Will be
 943  *             used for DMA mappings
 944  * @portptr:   pointer to a local port pointer. Upon success, the routine
 945  *             will allocate a nvme_fc_local_port structure and place its
 946  *             address in the local port pointer. Upon failure, local port
 947  *             pointer will be set to NULL.
 948  *
 949  * Returns:
 950  * a completion status. Must be 0 upon success; a negative errno
 951  * (ex: -ENXIO) upon failure.
 952  */
 953 int
 954 nvmet_fc_register_targetport(struct nvmet_fc_port_info *pinfo,
 955                         struct nvmet_fc_target_template *template,
 956                         struct device *dev,
 957                         struct nvmet_fc_target_port **portptr)
 958 {
 959         struct nvmet_fc_tgtport *newrec;
 960         unsigned long flags;
 961         int ret, idx;
 962
 963         if (!template->xmt_ls_rsp || !template->fcp_op ||
 964             !template->fcp_abort ||
 965             !template->fcp_req_release || !template->targetport_delete ||
 966             !template->max_hw_queues || !template->max_sgl_segments ||
 967             !template->max_dif_sgl_segments || !template->dma_boundary) {
 968                 ret = -EINVAL;
 969                 goto out_regtgt_failed;
 970         }
 971
 972         newrec = kzalloc((sizeof(*newrec) + template->target_priv_sz),
 973                          GFP_KERNEL);
 974         if (!newrec) {
 975                 ret = -ENOMEM;
 976                 goto out_regtgt_failed;
 977         }
 978
 979         idx = ida_simple_get(&nvmet_fc_tgtport_cnt, 0, 0, GFP_KERNEL);
 980         if (idx < 0) {
 981                 ret = -ENOSPC;
 982                 goto out_fail_kfree;
 983         }
 984
 985         if (!get_device(dev) && dev) {
 986                 ret = -ENODEV;
 987                 goto out_ida_put;
 988         }
 989
 990         newrec->fc_target_port.node_name = pinfo->node_name;
 991         newrec->fc_target_port.port_name = pinfo->port_name;
 992         newrec->fc_target_port.private = &newrec[1];
 993         newrec->fc_target_port.port_id = pinfo->port_id;
 994         newrec->fc_target_port.port_num = idx;
 995         INIT_LIST_HEAD(&newrec->tgt_list);
 996         newrec->dev = dev;
 997         newrec->ops = template;
 998         spin_lock_init(&newrec->lock);
 999         INIT_LIST_HEAD(&newrec->ls_list);
1000         INIT_LIST_HEAD(&newrec->ls_busylist);
1001         INIT_LIST_HEAD(&newrec->assoc_list);
1002         kref_init(&newrec->ref);
1003         ida_init(&newrec->assoc_cnt);
1004         newrec->max_sg_cnt = template->max_sgl_segments;
1005
1006         ret = nvmet_fc_alloc_ls_iodlist(newrec);
1007         if (ret) {
1008                 ret = -ENOMEM;
1009                 goto out_free_newrec;
1010         }
1011
1012         spin_lock_irqsave(&nvmet_fc_tgtlock, flags);
1013         list_add_tail(&newrec->tgt_list, &nvmet_fc_target_list);
1014         spin_unlock_irqrestore(&nvmet_fc_tgtlock, flags);
1015
1016         *portptr = &newrec->fc_target_port;
1017         return 0;
1018
1019 out_free_newrec:
1020         put_device(dev);
1021 out_ida_put:
1022         ida_simple_remove(&nvmet_fc_tgtport_cnt, idx);
1023 out_fail_kfree:
1024         kfree(newrec);
1025 out_regtgt_failed:
1026         *portptr = NULL;
1027         return ret;
1028 }
1029 EXPORT_SYMBOL_GPL(nvmet_fc_register_targetport);
1030
1031
1032 static void
1033 nvmet_fc_free_tgtport(struct kref *ref)
1034 {
1035         struct nvmet_fc_tgtport *tgtport =
1036                 container_of(ref, struct nvmet_fc_tgtport, ref);
1037         struct device *dev = tgtport->dev;
1038         unsigned long flags;
1039
1040         spin_lock_irqsave(&nvmet_fc_tgtlock, flags);
1041         list_del(&tgtport->tgt_list);
1042         spin_unlock_irqrestore(&nvmet_fc_tgtlock, flags);
1043
1044         nvmet_fc_free_ls_iodlist(tgtport);
1045
1046         /* let the LLDD know we've finished tearing it down */
1047         tgtport->ops->targetport_delete(&tgtport->fc_target_port);
1048
1049         ida_simple_remove(&nvmet_fc_tgtport_cnt,
1050                         tgtport->fc_target_port.port_num);
1051
1052         ida_destroy(&tgtport->assoc_cnt);
1053
1054         kfree(tgtport);
1055
1056         put_device(dev);
1057 }
1058
1059 static void
1060 nvmet_fc_tgtport_put(struct nvmet_fc_tgtport *tgtport)
1061 {
1062         kref_put(&tgtport->ref, nvmet_fc_free_tgtport);
1063 }
1064
1065 static int
1066 nvmet_fc_tgtport_get(struct nvmet_fc_tgtport *tgtport)
1067 {
1068         return kref_get_unless_zero(&tgtport->ref);
1069 }
1070
1071 static void
1072 __nvmet_fc_free_assocs(struct nvmet_fc_tgtport *tgtport)
1073 {
1074         struct nvmet_fc_tgt_assoc *assoc, *next;
1075         unsigned long flags;
1076
1077         spin_lock_irqsave(&tgtport->lock, flags);
1078         list_for_each_entry_safe(assoc, next,
1079                                 &tgtport->assoc_list, a_list) {
1080                 if (!nvmet_fc_tgt_a_get(assoc))
1081                         continue;
1082                 spin_unlock_irqrestore(&tgtport->lock, flags);
1083                 nvmet_fc_delete_target_assoc(assoc);
1084                 nvmet_fc_tgt_a_put(assoc);
1085                 spin_lock_irqsave(&tgtport->lock, flags);
1086         }
1087         spin_unlock_irqrestore(&tgtport->lock, flags);
1088 }
1089
1090 /*
1091  * nvmet layer has called to terminate an association
1092  */
1093 static void
1094 nvmet_fc_delete_ctrl(struct nvmet_ctrl *ctrl)
1095 {
1096         struct nvmet_fc_tgtport *tgtport, *next;
1097         struct nvmet_fc_tgt_assoc *assoc;
1098         struct nvmet_fc_tgt_queue *queue;
1099         unsigned long flags;
1100         bool found_ctrl = false;
1101
1102         /* this is a bit ugly, but don't want to make locks layered */
1103         spin_lock_irqsave(&nvmet_fc_tgtlock, flags);
1104         list_for_each_entry_safe(tgtport, next, &nvmet_fc_target_list,
1105                         tgt_list) {
1106                 if (!nvmet_fc_tgtport_get(tgtport))
1107                         continue;
1108                 spin_unlock_irqrestore(&nvmet_fc_tgtlock, flags);
1109
1110                 spin_lock_irqsave(&tgtport->lock, flags);
1111                 list_for_each_entry(assoc, &tgtport->assoc_list, a_list) {
1112                         queue = assoc->queues[0];
1113                         if (queue && queue->nvme_sq.ctrl == ctrl) {
1114                                 if (nvmet_fc_tgt_a_get(assoc))
1115                                         found_ctrl = true;
1116                                 break;
1117                         }
1118                 }
1119                 spin_unlock_irqrestore(&tgtport->lock, flags);
1120
1121                 nvmet_fc_tgtport_put(tgtport);
1122
1123                 if (found_ctrl) {
1124                         nvmet_fc_delete_target_assoc(assoc);
1125                         nvmet_fc_tgt_a_put(assoc);
1126                         return;
1127                 }
1128
1129                 spin_lock_irqsave(&nvmet_fc_tgtlock, flags);
1130         }
1131         spin_unlock_irqrestore(&nvmet_fc_tgtlock, flags);
1132 }
1133
1134 /**
1135  * nvme_fc_unregister_targetport - transport entry point called by an
1136  *                              LLDD to deregister/remove a previously
1137  *                              registered a local NVME subsystem FC port.
1138  * @tgtport: pointer to the (registered) target port that is to be
1139  *           deregistered.
1140  *
1141  * Returns:
1142  * a completion status. Must be 0 upon success; a negative errno
1143  * (ex: -ENXIO) upon failure.
1144  */
1145 int
1146 nvmet_fc_unregister_targetport(struct nvmet_fc_target_port *target_port)
1147 {
1148         struct nvmet_fc_tgtport *tgtport = targetport_to_tgtport(target_port);
1149
1150         /* terminate any outstanding associations */
1151         __nvmet_fc_free_assocs(tgtport);
1152
1153         nvmet_fc_tgtport_put(tgtport);
1154
1155         return 0;
1156 }
1157 EXPORT_SYMBOL_GPL(nvmet_fc_unregister_targetport);
1158
1159
1160 /* *********************** FC-NVME LS Handling **************************** */
1161
1162
1163 static void
1164 nvmet_fc_format_rsp_hdr(void *buf, u8 ls_cmd, __be32 desc_len, u8 rqst_ls_cmd)
1165 {
1166         struct fcnvme_ls_acc_hdr *acc = buf;
1167
1168         acc->w0.ls_cmd = ls_cmd;
1169         acc->desc_list_len = desc_len;
1170         acc->rqst.desc_tag = cpu_to_be32(FCNVME_LSDESC_RQST);
1171         acc->rqst.desc_len =
1172                         fcnvme_lsdesc_len(sizeof(struct fcnvme_lsdesc_rqst));
1173         acc->rqst.w0.ls_cmd = rqst_ls_cmd;
1174 }
1175
1176 static int
1177 nvmet_fc_format_rjt(void *buf, u16 buflen, u8 ls_cmd,
1178                         u8 reason, u8 explanation, u8 vendor)
1179 {
1180         struct fcnvme_ls_rjt *rjt = buf;
1181
1182         nvmet_fc_format_rsp_hdr(buf, FCNVME_LSDESC_RQST,
1183                         fcnvme_lsdesc_len(sizeof(struct fcnvme_ls_rjt)),
1184                         ls_cmd);
1185         rjt->rjt.desc_tag = cpu_to_be32(FCNVME_LSDESC_RJT);
1186         rjt->rjt.desc_len = fcnvme_lsdesc_len(sizeof(struct fcnvme_lsdesc_rjt));
1187         rjt->rjt.reason_code = reason;
1188         rjt->rjt.reason_explanation = explanation;
1189         rjt->rjt.vendor = vendor;
1190
1191         return sizeof(struct fcnvme_ls_rjt);
1192 }
1193
1194 /* Validation Error indexes into the string table below */
1195 enum {
1196         VERR_NO_ERROR           = 0,
1197         VERR_CR_ASSOC_LEN       = 1,
1198         VERR_CR_ASSOC_RQST_LEN  = 2,
1199         VERR_CR_ASSOC_CMD       = 3,
1200         VERR_CR_ASSOC_CMD_LEN   = 4,
1201         VERR_ERSP_RATIO         = 5,
1202         VERR_ASSOC_ALLOC_FAIL   = 6,
1203         VERR_QUEUE_ALLOC_FAIL   = 7,
1204         VERR_CR_CONN_LEN        = 8,
1205         VERR_CR_CONN_RQST_LEN   = 9,
1206         VERR_ASSOC_ID           = 10,
1207         VERR_ASSOC_ID_LEN       = 11,
1208         VERR_NO_ASSOC           = 12,
1209         VERR_CONN_ID            = 13,
1210         VERR_CONN_ID_LEN        = 14,
1211         VERR_NO_CONN            = 15,
1212         VERR_CR_CONN_CMD        = 16,
1213         VERR_CR_CONN_CMD_LEN    = 17,
1214         VERR_DISCONN_LEN        = 18,
1215         VERR_DISCONN_RQST_LEN   = 19,
1216         VERR_DISCONN_CMD        = 20,
1217         VERR_DISCONN_CMD_LEN    = 21,
1218         VERR_DISCONN_SCOPE      = 22,
1219         VERR_RS_LEN             = 23,
1220         VERR_RS_RQST_LEN        = 24,
1221         VERR_RS_CMD             = 25,
1222         VERR_RS_CMD_LEN         = 26,
1223         VERR_RS_RCTL            = 27,
1224         VERR_RS_RO              = 28,
1225 };
1226
1227 static char *validation_errors[] = {
1228         "OK",
1229         "Bad CR_ASSOC Length",
1230         "Bad CR_ASSOC Rqst Length",
1231         "Not CR_ASSOC Cmd",
1232         "Bad CR_ASSOC Cmd Length",
1233         "Bad Ersp Ratio",
1234         "Association Allocation Failed",
1235         "Queue Allocation Failed",
1236         "Bad CR_CONN Length",
1237         "Bad CR_CONN Rqst Length",
1238         "Not Association ID",
1239         "Bad Association ID Length",
1240         "No Association",
1241         "Not Connection ID",
1242         "Bad Connection ID Length",
1243         "No Connection",
1244         "Not CR_CONN Cmd",
1245         "Bad CR_CONN Cmd Length",
1246         "Bad DISCONN Length",
1247         "Bad DISCONN Rqst Length",
1248         "Not DISCONN Cmd",
1249         "Bad DISCONN Cmd Length",
1250         "Bad Disconnect Scope",
1251         "Bad RS Length",
1252         "Bad RS Rqst Length",
1253         "Not RS Cmd",
1254         "Bad RS Cmd Length",
1255         "Bad RS R_CTL",
1256         "Bad RS Relative Offset",
1257 };
1258
1259 static void
1260 nvmet_fc_ls_create_association(struct nvmet_fc_tgtport *tgtport,
1261                         struct nvmet_fc_ls_iod *iod)
1262 {
1263         struct fcnvme_ls_cr_assoc_rqst *rqst =
1264                                 (struct fcnvme_ls_cr_assoc_rqst *)iod->rqstbuf;
1265         struct fcnvme_ls_cr_assoc_acc *acc =
1266                                 (struct fcnvme_ls_cr_assoc_acc *)iod->rspbuf;
1267         struct nvmet_fc_tgt_queue *queue;
1268         int ret = 0;
1269
1270         memset(acc, 0, sizeof(*acc));
1271
1272         /*
1273          * FC-NVME spec changes. There are initiators sending different
1274          * lengths as padding sizes for Create Association Cmd descriptor
1275          * was incorrect.
1276          * Accept anything of "minimum" length. Assume format per 1.15
1277          * spec (with HOSTID reduced to 16 bytes), ignore how long the
1278          * trailing pad length is.
1279          */
1280         if (iod->rqstdatalen < FCNVME_LSDESC_CRA_RQST_MINLEN)
1281                 ret = VERR_CR_ASSOC_LEN;
1282         else if (be32_to_cpu(rqst->desc_list_len) <
1283                         FCNVME_LSDESC_CRA_RQST_MIN_LISTLEN)
1284                 ret = VERR_CR_ASSOC_RQST_LEN;
1285         else if (rqst->assoc_cmd.desc_tag !=
1286                         cpu_to_be32(FCNVME_LSDESC_CREATE_ASSOC_CMD))
1287                 ret = VERR_CR_ASSOC_CMD;
1288         else if (be32_to_cpu(rqst->assoc_cmd.desc_len) <
1289                         FCNVME_LSDESC_CRA_CMD_DESC_MIN_DESCLEN)
1290                 ret = VERR_CR_ASSOC_CMD_LEN;
1291         else if (!rqst->assoc_cmd.ersp_ratio ||
1292                  (be16_to_cpu(rqst->assoc_cmd.ersp_ratio) >=
1293                                 be16_to_cpu(rqst->assoc_cmd.sqsize)))
1294                 ret = VERR_ERSP_RATIO;
1295
1296         else {
1297                 /* new association w/ admin queue */
1298                 iod->assoc = nvmet_fc_alloc_target_assoc(tgtport);
1299                 if (!iod->assoc)
1300                         ret = VERR_ASSOC_ALLOC_FAIL;
1301                 else {
1302                         queue = nvmet_fc_alloc_target_queue(iod->assoc, 0,
1303                                         be16_to_cpu(rqst->assoc_cmd.sqsize));
1304                         if (!queue)
1305                                 ret = VERR_QUEUE_ALLOC_FAIL;
1306                 }
1307         }
1308
1309         if (ret) {
1310                 dev_err(tgtport->dev,
1311                         "Create Association LS failed: %s\n",
1312                         validation_errors[ret]);
1313                 iod->lsreq->rsplen = nvmet_fc_format_rjt(acc,
1314                                 NVME_FC_MAX_LS_BUFFER_SIZE, rqst->w0.ls_cmd,
1315                                 FCNVME_RJT_RC_LOGIC,
1316                                 FCNVME_RJT_EXP_NONE, 0);
1317                 return;
1318         }
1319
1320         queue->ersp_ratio = be16_to_cpu(rqst->assoc_cmd.ersp_ratio);
1321         atomic_set(&queue->connected, 1);
1322         queue->sqhd = 0;        /* best place to init value */
1323
1324         /* format a response */
1325
1326         iod->lsreq->rsplen = sizeof(*acc);
1327
1328         nvmet_fc_format_rsp_hdr(acc, FCNVME_LS_ACC,
1329                         fcnvme_lsdesc_len(
1330                                 sizeof(struct fcnvme_ls_cr_assoc_acc)),
1331                         FCNVME_LS_CREATE_ASSOCIATION);
1332         acc->associd.desc_tag = cpu_to_be32(FCNVME_LSDESC_ASSOC_ID);
1333         acc->associd.desc_len =
1334                         fcnvme_lsdesc_len(
1335                                 sizeof(struct fcnvme_lsdesc_assoc_id));
1336         acc->associd.association_id =
1337                         cpu_to_be64(nvmet_fc_makeconnid(iod->assoc, 0));
1338         acc->connectid.desc_tag = cpu_to_be32(FCNVME_LSDESC_CONN_ID);
1339         acc->connectid.desc_len =
1340                         fcnvme_lsdesc_len(
1341                                 sizeof(struct fcnvme_lsdesc_conn_id));
1342         acc->connectid.connection_id = acc->associd.association_id;
1343 }
1344
1345 static void
1346 nvmet_fc_ls_create_connection(struct nvmet_fc_tgtport *tgtport,
1347                         struct nvmet_fc_ls_iod *iod)
1348 {
1349         struct fcnvme_ls_cr_conn_rqst *rqst =
1350                                 (struct fcnvme_ls_cr_conn_rqst *)iod->rqstbuf;
1351         struct fcnvme_ls_cr_conn_acc *acc =
1352                                 (struct fcnvme_ls_cr_conn_acc *)iod->rspbuf;
1353         struct nvmet_fc_tgt_queue *queue;
1354         int ret = 0;
1355
1356         memset(acc, 0, sizeof(*acc));
1357
1358         if (iod->rqstdatalen < sizeof(struct fcnvme_ls_cr_conn_rqst))
1359                 ret = VERR_CR_CONN_LEN;
1360         else if (rqst->desc_list_len !=
1361                         fcnvme_lsdesc_len(
1362                                 sizeof(struct fcnvme_ls_cr_conn_rqst)))
1363                 ret = VERR_CR_CONN_RQST_LEN;
1364         else if (rqst->associd.desc_tag != cpu_to_be32(FCNVME_LSDESC_ASSOC_ID))
1365                 ret = VERR_ASSOC_ID;
1366         else if (rqst->associd.desc_len !=
1367                         fcnvme_lsdesc_len(
1368                                 sizeof(struct fcnvme_lsdesc_assoc_id)))
1369                 ret = VERR_ASSOC_ID_LEN;
1370         else if (rqst->connect_cmd.desc_tag !=
1371                         cpu_to_be32(FCNVME_LSDESC_CREATE_CONN_CMD))
1372                 ret = VERR_CR_CONN_CMD;
1373         else if (rqst->connect_cmd.desc_len !=
1374                         fcnvme_lsdesc_len(
1375                                 sizeof(struct fcnvme_lsdesc_cr_conn_cmd)))
1376                 ret = VERR_CR_CONN_CMD_LEN;
1377         else if (!rqst->connect_cmd.ersp_ratio ||
1378                  (be16_to_cpu(rqst->connect_cmd.ersp_ratio) >=
1379                                 be16_to_cpu(rqst->connect_cmd.sqsize)))
1380                 ret = VERR_ERSP_RATIO;
1381
1382         else {
1383                 /* new io queue */
1384                 iod->assoc = nvmet_fc_find_target_assoc(tgtport,
1385                                 be64_to_cpu(rqst->associd.association_id));
1386                 if (!iod->assoc)
1387                         ret = VERR_NO_ASSOC;
1388                 else {
1389                         queue = nvmet_fc_alloc_target_queue(iod->assoc,
1390                                         be16_to_cpu(rqst->connect_cmd.qid),
1391                                         be16_to_cpu(rqst->connect_cmd.sqsize));
1392                         if (!queue)
1393                                 ret = VERR_QUEUE_ALLOC_FAIL;
1394
1395                         /* release get taken in nvmet_fc_find_target_assoc */
1396                         nvmet_fc_tgt_a_put(iod->assoc);
1397                 }
1398         }
1399
1400         if (ret) {
1401                 dev_err(tgtport->dev,
1402                         "Create Connection LS failed: %s\n",
1403                         validation_errors[ret]);
1404                 iod->lsreq->rsplen = nvmet_fc_format_rjt(acc,
1405                                 NVME_FC_MAX_LS_BUFFER_SIZE, rqst->w0.ls_cmd,
1406                                 (ret == VERR_NO_ASSOC) ?
1407                                         FCNVME_RJT_RC_INV_ASSOC :
1408                                         FCNVME_RJT_RC_LOGIC,
1409                                 FCNVME_RJT_EXP_NONE, 0);
1410                 return;
1411         }
1412
1413         queue->ersp_ratio = be16_to_cpu(rqst->connect_cmd.ersp_ratio);
1414         atomic_set(&queue->connected, 1);
1415         queue->sqhd = 0;        /* best place to init value */
1416
1417         /* format a response */
1418
1419         iod->lsreq->rsplen = sizeof(*acc);
1420
1421         nvmet_fc_format_rsp_hdr(acc, FCNVME_LS_ACC,
1422                         fcnvme_lsdesc_len(sizeof(struct fcnvme_ls_cr_conn_acc)),
1423                         FCNVME_LS_CREATE_CONNECTION);
1424         acc->connectid.desc_tag = cpu_to_be32(FCNVME_LSDESC_CONN_ID);
1425         acc->connectid.desc_len =
1426                         fcnvme_lsdesc_len(
1427                                 sizeof(struct fcnvme_lsdesc_conn_id));
1428         acc->connectid.connection_id =
1429                         cpu_to_be64(nvmet_fc_makeconnid(iod->assoc,
1430                                 be16_to_cpu(rqst->connect_cmd.qid)));
1431 }
1432
1433 static void
1434 nvmet_fc_ls_disconnect(struct nvmet_fc_tgtport *tgtport,
1435                         struct nvmet_fc_ls_iod *iod)
1436 {
1437         struct fcnvme_ls_disconnect_rqst *rqst =
1438                         (struct fcnvme_ls_disconnect_rqst *)iod->rqstbuf;
1439         struct fcnvme_ls_disconnect_acc *acc =
1440                         (struct fcnvme_ls_disconnect_acc *)iod->rspbuf;
1441         struct nvmet_fc_tgt_queue *queue = NULL;
1442         struct nvmet_fc_tgt_assoc *assoc;
1443         int ret = 0;
1444         bool del_assoc = false;
1445
1446         memset(acc, 0, sizeof(*acc));
1447
1448         if (iod->rqstdatalen < sizeof(struct fcnvme_ls_disconnect_rqst))
1449                 ret = VERR_DISCONN_LEN;
1450         else if (rqst->desc_list_len !=
1451                         fcnvme_lsdesc_len(
1452                                 sizeof(struct fcnvme_ls_disconnect_rqst)))
1453                 ret = VERR_DISCONN_RQST_LEN;
1454         else if (rqst->associd.desc_tag != cpu_to_be32(FCNVME_LSDESC_ASSOC_ID))
1455                 ret = VERR_ASSOC_ID;
1456         else if (rqst->associd.desc_len !=
1457                         fcnvme_lsdesc_len(
1458                                 sizeof(struct fcnvme_lsdesc_assoc_id)))
1459                 ret = VERR_ASSOC_ID_LEN;
1460         else if (rqst->discon_cmd.desc_tag !=
1461                         cpu_to_be32(FCNVME_LSDESC_DISCONN_CMD))
1462                 ret = VERR_DISCONN_CMD;
1463         else if (rqst->discon_cmd.desc_len !=
1464                         fcnvme_lsdesc_len(
1465                                 sizeof(struct fcnvme_lsdesc_disconn_cmd)))
1466                 ret = VERR_DISCONN_CMD_LEN;
1467         else if ((rqst->discon_cmd.scope != FCNVME_DISCONN_ASSOCIATION) &&
1468                         (rqst->discon_cmd.scope != FCNVME_DISCONN_CONNECTION))
1469                 ret = VERR_DISCONN_SCOPE;
1470         else {
1471                 /* match an active association */
1472                 assoc = nvmet_fc_find_target_assoc(tgtport,
1473                                 be64_to_cpu(rqst->associd.association_id));
1474                 iod->assoc = assoc;
1475                 if (assoc) {
1476                         if (rqst->discon_cmd.scope ==
1477                                         FCNVME_DISCONN_CONNECTION) {
1478                                 queue = nvmet_fc_find_target_queue(tgtport,
1479                                                 be64_to_cpu(
1480                                                         rqst->discon_cmd.id));
1481                                 if (!queue) {
1482                                         nvmet_fc_tgt_a_put(assoc);
1483                                         ret = VERR_NO_CONN;
1484                                 }
1485                         }
1486                 } else
1487                         ret = VERR_NO_ASSOC;
1488         }
1489
1490         if (ret) {
1491                 dev_err(tgtport->dev,
1492                         "Disconnect LS failed: %s\n",
1493                         validation_errors[ret]);
1494                 iod->lsreq->rsplen = nvmet_fc_format_rjt(acc,
1495                                 NVME_FC_MAX_LS_BUFFER_SIZE, rqst->w0.ls_cmd,
1496                                 (ret == VERR_NO_ASSOC) ?
1497                                         FCNVME_RJT_RC_INV_ASSOC :
1498                                         (ret == VERR_NO_CONN) ?
1499                                                 FCNVME_RJT_RC_INV_CONN :
1500                                                 FCNVME_RJT_RC_LOGIC,
1501                                 FCNVME_RJT_EXP_NONE, 0);
1502                 return;
1503         }
1504
1505         /* format a response */
1506
1507         iod->lsreq->rsplen = sizeof(*acc);
1508
1509         nvmet_fc_format_rsp_hdr(acc, FCNVME_LS_ACC,
1510                         fcnvme_lsdesc_len(
1511                                 sizeof(struct fcnvme_ls_disconnect_acc)),
1512                         FCNVME_LS_DISCONNECT);
1513
1514
1515         /* are we to delete a Connection ID (queue) */
1516         if (queue) {
1517                 int qid = queue->qid;
1518
1519                 nvmet_fc_delete_target_queue(queue);
1520
1521                 /* release the get taken by find_target_queue */
1522                 nvmet_fc_tgt_q_put(queue);
1523
1524                 /* tear association down if io queue terminated */
1525                 if (!qid)
1526                         del_assoc = true;
1527         }
1528
1529         /* release get taken in nvmet_fc_find_target_assoc */
1530         nvmet_fc_tgt_a_put(iod->assoc);
1531
1532         if (del_assoc)
1533                 nvmet_fc_delete_target_assoc(iod->assoc);
1534 }
1535
1536
1537 /* *********************** NVME Ctrl Routines **************************** */
1538
1539
1540 static void nvmet_fc_fcp_nvme_cmd_done(struct nvmet_req *nvme_req);
1541
1542 static struct nvmet_fabrics_ops nvmet_fc_tgt_fcp_ops;
1543
1544 static void
1545 nvmet_fc_xmt_ls_rsp_done(struct nvmefc_tgt_ls_req *lsreq)
1546 {
1547         struct nvmet_fc_ls_iod *iod = lsreq->nvmet_fc_private;
1548         struct nvmet_fc_tgtport *tgtport = iod->tgtport;
1549
1550         fc_dma_sync_single_for_cpu(tgtport->dev, iod->rspdma,
1551                                 NVME_FC_MAX_LS_BUFFER_SIZE, DMA_TO_DEVICE);
1552         nvmet_fc_free_ls_iod(tgtport, iod);
1553         nvmet_fc_tgtport_put(tgtport);
1554 }
1555
1556 static void
1557 nvmet_fc_xmt_ls_rsp(struct nvmet_fc_tgtport *tgtport,
1558                                 struct nvmet_fc_ls_iod *iod)
1559 {
1560         int ret;
1561
1562         fc_dma_sync_single_for_device(tgtport->dev, iod->rspdma,
1563                                   NVME_FC_MAX_LS_BUFFER_SIZE, DMA_TO_DEVICE);
1564
1565         ret = tgtport->ops->xmt_ls_rsp(&tgtport->fc_target_port, iod->lsreq);
1566         if (ret)
1567                 nvmet_fc_xmt_ls_rsp_done(iod->lsreq);
1568 }
1569
1570 /*
1571  * Actual processing routine for received FC-NVME LS Requests from the LLD
1572  */
1573 static void
1574 nvmet_fc_handle_ls_rqst(struct nvmet_fc_tgtport *tgtport,
1575                         struct nvmet_fc_ls_iod *iod)
1576 {
1577         struct fcnvme_ls_rqst_w0 *w0 =
1578                         (struct fcnvme_ls_rqst_w0 *)iod->rqstbuf;
1579
1580         iod->lsreq->nvmet_fc_private = iod;
1581         iod->lsreq->rspbuf = iod->rspbuf;
1582         iod->lsreq->rspdma = iod->rspdma;
1583         iod->lsreq->done = nvmet_fc_xmt_ls_rsp_done;
1584         /* Be preventative. handlers will later set to valid length */
1585         iod->lsreq->rsplen = 0;
1586
1587         iod->assoc = NULL;
1588
1589         /*
1590          * handlers:
1591          *   parse request input, execute the request, and format the
1592          *   LS response
1593          */
1594         switch (w0->ls_cmd) {
1595         case FCNVME_LS_CREATE_ASSOCIATION:
1596                 /* Creates Association and initial Admin Queue/Connection */
1597                 nvmet_fc_ls_create_association(tgtport, iod);
1598                 break;
1599         case FCNVME_LS_CREATE_CONNECTION:
1600                 /* Creates an IO Queue/Connection */
1601                 nvmet_fc_ls_create_connection(tgtport, iod);
1602                 break;
1603         case FCNVME_LS_DISCONNECT:
1604                 /* Terminate a Queue/Connection or the Association */
1605                 nvmet_fc_ls_disconnect(tgtport, iod);
1606                 break;
1607         default:
1608                 iod->lsreq->rsplen = nvmet_fc_format_rjt(iod->rspbuf,
1609                                 NVME_FC_MAX_LS_BUFFER_SIZE, w0->ls_cmd,
1610                                 FCNVME_RJT_RC_INVAL, FCNVME_RJT_EXP_NONE, 0);
1611         }
1612
1613         nvmet_fc_xmt_ls_rsp(tgtport, iod);
1614 }
1615
1616 /*
1617  * Actual processing routine for received FC-NVME LS Requests from the LLD
1618  */
1619 static void
1620 nvmet_fc_handle_ls_rqst_work(struct work_struct *work)
1621 {
1622         struct nvmet_fc_ls_iod *iod =
1623                 container_of(work, struct nvmet_fc_ls_iod, work);
1624         struct nvmet_fc_tgtport *tgtport = iod->tgtport;
1625
1626         nvmet_fc_handle_ls_rqst(tgtport, iod);
1627 }
1628
1629
1630 /**
1631  * nvmet_fc_rcv_ls_req - transport entry point called by an LLDD
1632  *                       upon the reception of a NVME LS request.
1633  *
1634  * The nvmet-fc layer will copy payload to an internal structure for
1635  * processing.  As such, upon completion of the routine, the LLDD may
1636  * immediately free/reuse the LS request buffer passed in the call.
1637  *
1638  * If this routine returns error, the LLDD should abort the exchange.
1639  *
1640  * @tgtport:    pointer to the (registered) target port the LS was
1641  *              received on.
1642  * @lsreq:      pointer to a lsreq request structure to be used to reference
1643  *              the exchange corresponding to the LS.
1644  * @lsreqbuf:   pointer to the buffer containing the LS Request
1645  * @lsreqbuf_len: length, in bytes, of the received LS request
1646  */
1647 int
1648 nvmet_fc_rcv_ls_req(struct nvmet_fc_target_port *target_port,
1649                         struct nvmefc_tgt_ls_req *lsreq,
1650                         void *lsreqbuf, u32 lsreqbuf_len)
1651 {
1652         struct nvmet_fc_tgtport *tgtport = targetport_to_tgtport(target_port);
1653         struct nvmet_fc_ls_iod *iod;
1654
1655         if (lsreqbuf_len > NVME_FC_MAX_LS_BUFFER_SIZE)
1656                 return -E2BIG;
1657
1658         if (!nvmet_fc_tgtport_get(tgtport))
1659                 return -ESHUTDOWN;
1660
1661         iod = nvmet_fc_alloc_ls_iod(tgtport);
1662         if (!iod) {
1663                 nvmet_fc_tgtport_put(tgtport);
1664                 return -ENOENT;
1665         }
1666
1667         iod->lsreq = lsreq;
1668         iod->fcpreq = NULL;
1669         memcpy(iod->rqstbuf, lsreqbuf, lsreqbuf_len);
1670         iod->rqstdatalen = lsreqbuf_len;
1671
1672         schedule_work(&iod->work);
1673
1674         return 0;
1675 }
1676 EXPORT_SYMBOL_GPL(nvmet_fc_rcv_ls_req);
1677
1678
1679 /*
1680  * **********************
1681  * Start of FCP handling
1682  * **********************
1683  */
1684
1685 static int
1686 nvmet_fc_alloc_tgt_pgs(struct nvmet_fc_fcp_iod *fod)
1687 {
1688         struct scatterlist *sg;
1689         struct page *page;
1690         unsigned int nent;
1691         u32 page_len, length;
1692         int i = 0;
1693
1694         length = fod->total_length;
1695         nent = DIV_ROUND_UP(length, PAGE_SIZE);
1696         sg = kmalloc_array(nent, sizeof(struct scatterlist), GFP_KERNEL);
1697         if (!sg)
1698                 goto out;
1699
1700         sg_init_table(sg, nent);
1701
1702         while (length) {
1703                 page_len = min_t(u32, length, PAGE_SIZE);
1704
1705                 page = alloc_page(GFP_KERNEL);
1706                 if (!page)
1707                         goto out_free_pages;
1708
1709                 sg_set_page(&sg[i], page, page_len, 0);
1710                 length -= page_len;
1711                 i++;
1712         }
1713
1714         fod->data_sg = sg;
1715         fod->data_sg_cnt = nent;
1716         fod->data_sg_cnt = fc_dma_map_sg(fod->tgtport->dev, sg, nent,
1717                                 ((fod->io_dir == NVMET_FCP_WRITE) ?
1718                                         DMA_FROM_DEVICE : DMA_TO_DEVICE));
1719                                 /* note: write from initiator perspective */
1720         fod->next_sg = fod->data_sg;
1721
1722         return 0;
1723
1724 out_free_pages:
1725         while (i > 0) {
1726                 i--;
1727                 __free_page(sg_page(&sg[i]));
1728         }
1729         kfree(sg);
1730         fod->data_sg = NULL;
1731         fod->data_sg_cnt = 0;
1732 out:
1733         return NVME_SC_INTERNAL;
1734 }
1735
1736 static void
1737 nvmet_fc_free_tgt_pgs(struct nvmet_fc_fcp_iod *fod)
1738 {
1739         struct scatterlist *sg;
1740         int count;
1741
1742         if (!fod->data_sg || !fod->data_sg_cnt)
1743                 return;
1744
1745         fc_dma_unmap_sg(fod->tgtport->dev, fod->data_sg, fod->data_sg_cnt,
1746                                 ((fod->io_dir == NVMET_FCP_WRITE) ?
1747                                         DMA_FROM_DEVICE : DMA_TO_DEVICE));
1748         for_each_sg(fod->data_sg, sg, fod->data_sg_cnt, count)
1749                 __free_page(sg_page(sg));
1750         kfree(fod->data_sg);
1751         fod->data_sg = NULL;
1752         fod->data_sg_cnt = 0;
1753 }
1754
1755
1756 static bool
1757 queue_90percent_full(struct nvmet_fc_tgt_queue *q, u32 sqhd)
1758 {
1759         u32 sqtail, used;
1760
1761         /* egad, this is ugly. And sqtail is just a best guess */
1762         sqtail = atomic_read(&q->sqtail) % q->sqsize;
1763
1764         used = (sqtail < sqhd) ? (sqtail + q->sqsize - sqhd) : (sqtail - sqhd);
1765         return ((used * 10) >= (((u32)(q->sqsize - 1) * 9)));
1766 }
1767
1768 /*
1769  * Prep RSP payload.
1770  * May be a NVMET_FCOP_RSP or NVMET_FCOP_READDATA_RSP op
1771  */
1772 static void
1773 nvmet_fc_prep_fcp_rsp(struct nvmet_fc_tgtport *tgtport,
1774                                 struct nvmet_fc_fcp_iod *fod)
1775 {
1776         struct nvme_fc_ersp_iu *ersp = &fod->rspiubuf;
1777         struct nvme_common_command *sqe = &fod->cmdiubuf.sqe.common;
1778         struct nvme_completion *cqe = &ersp->cqe;
1779         u32 *cqewd = (u32 *)cqe;
1780         bool send_ersp = false;
1781         u32 rsn, rspcnt, xfr_length;
1782
1783         if (fod->fcpreq->op == NVMET_FCOP_READDATA_RSP)
1784                 xfr_length = fod->total_length;
1785         else
1786                 xfr_length = fod->offset;
1787
1788         /*
1789          * check to see if we can send a 0's rsp.
1790          *   Note: to send a 0's response, the NVME-FC host transport will
1791          *   recreate the CQE. The host transport knows: sq id, SQHD (last
1792          *   seen in an ersp), and command_id. Thus it will create a
1793          *   zero-filled CQE with those known fields filled in. Transport
1794          *   must send an ersp for any condition where the cqe won't match
1795          *   this.
1796          *
1797          * Here are the FC-NVME mandated cases where we must send an ersp:
1798          *  every N responses, where N=ersp_ratio
1799          *  force fabric commands to send ersp's (not in FC-NVME but good
1800          *    practice)
1801          *  normal cmds: any time status is non-zero, or status is zero
1802          *     but words 0 or 1 are non-zero.
1803          *  the SQ is 90% or more full
1804          *  the cmd is a fused command
1805          *  transferred data length not equal to cmd iu length
1806          */
1807         rspcnt = atomic_inc_return(&fod->queue->zrspcnt);
1808         if (!(rspcnt % fod->queue->ersp_ratio) ||
1809             sqe->opcode == nvme_fabrics_command ||
1810             xfr_length != fod->total_length ||
1811             (le16_to_cpu(cqe->status) & 0xFFFE) || cqewd[0] || cqewd[1] ||
1812             (sqe->flags & (NVME_CMD_FUSE_FIRST | NVME_CMD_FUSE_SECOND)) ||
1813             queue_90percent_full(fod->queue, le16_to_cpu(cqe->sq_head)))
1814                 send_ersp = true;
1815
1816         /* re-set the fields */
1817         fod->fcpreq->rspaddr = ersp;
1818         fod->fcpreq->rspdma = fod->rspdma;
1819
1820         if (!send_ersp) {
1821                 memset(ersp, 0, NVME_FC_SIZEOF_ZEROS_RSP);
1822                 fod->fcpreq->rsplen = NVME_FC_SIZEOF_ZEROS_RSP;
1823         } else {
1824                 ersp->iu_len = cpu_to_be16(sizeof(*ersp)/sizeof(u32));
1825                 rsn = atomic_inc_return(&fod->queue->rsn);
1826                 ersp->rsn = cpu_to_be32(rsn);
1827                 ersp->xfrd_len = cpu_to_be32(xfr_length);
1828                 fod->fcpreq->rsplen = sizeof(*ersp);
1829         }
1830
1831         fc_dma_sync_single_for_device(tgtport->dev, fod->rspdma,
1832                                   sizeof(fod->rspiubuf), DMA_TO_DEVICE);
1833 }
1834
1835 static void nvmet_fc_xmt_fcp_op_done(struct nvmefc_tgt_fcp_req *fcpreq);
1836
1837 static void
1838 nvmet_fc_abort_op(struct nvmet_fc_tgtport *tgtport,
1839                                 struct nvmet_fc_fcp_iod *fod)
1840 {
1841         struct nvmefc_tgt_fcp_req *fcpreq = fod->fcpreq;
1842
1843         /* data no longer needed */
1844         nvmet_fc_free_tgt_pgs(fod);
1845
1846         /*
1847          * if an ABTS was received or we issued the fcp_abort early
1848          * don't call abort routine again.
1849          */
1850         /* no need to take lock - lock was taken earlier to get here */
1851         if (!fod->aborted)
1852                 tgtport->ops->fcp_abort(&tgtport->fc_target_port, fcpreq);
1853
1854         nvmet_fc_free_fcp_iod(fod->queue, fod);
1855 }
1856
1857 static void
1858 nvmet_fc_xmt_fcp_rsp(struct nvmet_fc_tgtport *tgtport,
1859                                 struct nvmet_fc_fcp_iod *fod)
1860 {
1861         int ret;
1862
1863         fod->fcpreq->op = NVMET_FCOP_RSP;
1864         fod->fcpreq->timeout = 0;
1865
1866         nvmet_fc_prep_fcp_rsp(tgtport, fod);
1867
1868         ret = tgtport->ops->fcp_op(&tgtport->fc_target_port, fod->fcpreq);
1869         if (ret)
1870                 nvmet_fc_abort_op(tgtport, fod);
1871 }
1872
1873 static void
1874 nvmet_fc_transfer_fcp_data(struct nvmet_fc_tgtport *tgtport,
1875                                 struct nvmet_fc_fcp_iod *fod, u8 op)
1876 {
1877         struct nvmefc_tgt_fcp_req *fcpreq = fod->fcpreq;
1878         struct scatterlist *sg = fod->next_sg;
1879         unsigned long flags;
1880         u32 remaininglen = fod->total_length - fod->offset;
1881         u32 tlen = 0;
1882         int ret;
1883
1884         fcpreq->op = op;
1885         fcpreq->offset = fod->offset;
1886         fcpreq->timeout = NVME_FC_TGTOP_TIMEOUT_SEC;
1887
1888         /*
1889          * for next sequence:
1890          *  break at a sg element boundary
1891          *  attempt to keep sequence length capped at
1892          *    NVMET_FC_MAX_SEQ_LENGTH but allow sequence to
1893          *    be longer if a single sg element is larger
1894          *    than that amount. This is done to avoid creating
1895          *    a new sg list to use for the tgtport api.
1896          */
1897         fcpreq->sg = sg;
1898         fcpreq->sg_cnt = 0;
1899         while (tlen < remaininglen &&
1900                fcpreq->sg_cnt < tgtport->max_sg_cnt &&
1901                tlen + sg_dma_len(sg) < NVMET_FC_MAX_SEQ_LENGTH) {
1902                 fcpreq->sg_cnt++;
1903                 tlen += sg_dma_len(sg);
1904                 sg = sg_next(sg);
1905         }
1906         if (tlen < remaininglen && fcpreq->sg_cnt == 0) {
1907                 fcpreq->sg_cnt++;
1908                 tlen += min_t(u32, sg_dma_len(sg), remaininglen);
1909                 sg = sg_next(sg);
1910         }
1911         if (tlen < remaininglen)
1912                 fod->next_sg = sg;
1913         else
1914                 fod->next_sg = NULL;
1915
1916         fcpreq->transfer_length = tlen;
1917         fcpreq->transferred_length = 0;
1918         fcpreq->fcp_error = 0;
1919         fcpreq->rsplen = 0;
1920
1921         /*
1922          * If the last READDATA request: check if LLDD supports
1923          * combined xfr with response.
1924          */
1925         if ((op == NVMET_FCOP_READDATA) &&
1926             ((fod->offset + fcpreq->transfer_length) == fod->total_length) &&
1927             (tgtport->ops->target_features & NVMET_FCTGTFEAT_READDATA_RSP)) {
1928                 fcpreq->op = NVMET_FCOP_READDATA_RSP;
1929                 nvmet_fc_prep_fcp_rsp(tgtport, fod);
1930         }
1931
1932         ret = tgtport->ops->fcp_op(&tgtport->fc_target_port, fod->fcpreq);
1933         if (ret) {
1934                 /*
1935                  * should be ok to set w/o lock as its in the thread of
1936                  * execution (not an async timer routine) and doesn't
1937                  * contend with any clearing action
1938                  */
1939                 fod->abort = true;
1940
1941                 if (op == NVMET_FCOP_WRITEDATA) {
1942                         spin_lock_irqsave(&fod->flock, flags);
1943                         fod->writedataactive = false;
1944                         spin_unlock_irqrestore(&fod->flock, flags);
1945                         nvmet_req_complete(&fod->req, NVME_SC_INTERNAL);
1946                 } else /* NVMET_FCOP_READDATA or NVMET_FCOP_READDATA_RSP */ {
1947                         fcpreq->fcp_error = ret;
1948                         fcpreq->transferred_length = 0;
1949                         nvmet_fc_xmt_fcp_op_done(fod->fcpreq);
1950                 }
1951         }
1952 }
1953
1954 static inline bool
1955 __nvmet_fc_fod_op_abort(struct nvmet_fc_fcp_iod *fod, bool abort)
1956 {
1957         struct nvmefc_tgt_fcp_req *fcpreq = fod->fcpreq;
1958         struct nvmet_fc_tgtport *tgtport = fod->tgtport;
1959
1960         /* if in the middle of an io and we need to tear down */
1961         if (abort) {
1962                 if (fcpreq->op == NVMET_FCOP_WRITEDATA) {
1963                         nvmet_req_complete(&fod->req, NVME_SC_INTERNAL);
1964                         return true;
1965                 }
1966
1967                 nvmet_fc_abort_op(tgtport, fod);
1968                 return true;
1969         }
1970
1971         return false;
1972 }
1973
1974 /*
1975  * actual done handler for FCP operations when completed by the lldd
1976  */
1977 static void
1978 nvmet_fc_fod_op_done(struct nvmet_fc_fcp_iod *fod)
1979 {
1980         struct nvmefc_tgt_fcp_req *fcpreq = fod->fcpreq;
1981         struct nvmet_fc_tgtport *tgtport = fod->tgtport;
1982         unsigned long flags;
1983         bool abort;
1984
1985         spin_lock_irqsave(&fod->flock, flags);
1986         abort = fod->abort;
1987         fod->writedataactive = false;
1988         spin_unlock_irqrestore(&fod->flock, flags);
1989
1990         switch (fcpreq->op) {
1991
1992         case NVMET_FCOP_WRITEDATA:
1993                 if (__nvmet_fc_fod_op_abort(fod, abort))
1994                         return;
1995                 if (fcpreq->fcp_error ||
1996                     fcpreq->transferred_length != fcpreq->transfer_length) {
1997                         spin_lock_irqsave(&fod->flock, flags);
1998                         fod->abort = true;
1999                         spin_unlock_irqrestore(&fod->flock, flags);
2000
2001                         nvmet_req_complete(&fod->req, NVME_SC_INTERNAL);
2002                         return;
2003                 }
2004
2005                 fod->offset += fcpreq->transferred_length;
2006                 if (fod->offset != fod->total_length) {
2007                         spin_lock_irqsave(&fod->flock, flags);
2008                         fod->writedataactive = true;
2009                         spin_unlock_irqrestore(&fod->flock, flags);
2010
2011                         /* transfer the next chunk */
2012                         nvmet_fc_transfer_fcp_data(tgtport, fod,
2013                                                 NVMET_FCOP_WRITEDATA);
2014                         return;
2015                 }
2016
2017                 /* data transfer complete, resume with nvmet layer */
2018
2019                 fod->req.execute(&fod->req);
2020
2021                 break;
2022
2023         case NVMET_FCOP_READDATA:
2024         case NVMET_FCOP_READDATA_RSP:
2025                 if (__nvmet_fc_fod_op_abort(fod, abort))
2026                         return;
2027                 if (fcpreq->fcp_error ||
2028                     fcpreq->transferred_length != fcpreq->transfer_length) {
2029                         nvmet_fc_abort_op(tgtport, fod);
2030                         return;
2031                 }
2032
2033                 /* success */
2034
2035                 if (fcpreq->op == NVMET_FCOP_READDATA_RSP) {
2036                         /* data no longer needed */
2037                         nvmet_fc_free_tgt_pgs(fod);
2038                         nvmet_fc_free_fcp_iod(fod->queue, fod);
2039                         return;
2040                 }
2041
2042                 fod->offset += fcpreq->transferred_length;
2043                 if (fod->offset != fod->total_length) {
2044                         /* transfer the next chunk */
2045                         nvmet_fc_transfer_fcp_data(tgtport, fod,
2046                                                 NVMET_FCOP_READDATA);
2047                         return;
2048                 }
2049
2050                 /* data transfer complete, send response */
2051
2052                 /* data no longer needed */
2053                 nvmet_fc_free_tgt_pgs(fod);
2054
2055                 nvmet_fc_xmt_fcp_rsp(tgtport, fod);
2056
2057                 break;
2058
2059         case NVMET_FCOP_RSP:
2060                 if (__nvmet_fc_fod_op_abort(fod, abort))
2061                         return;
2062                 nvmet_fc_free_fcp_iod(fod->queue, fod);
2063                 break;
2064
2065         default:
2066                 break;
2067         }
2068 }
2069
2070 static void
2071 nvmet_fc_fcp_rqst_op_done_work(struct work_struct *work)
2072 {
2073         struct nvmet_fc_fcp_iod *fod =
2074                 container_of(work, struct nvmet_fc_fcp_iod, done_work);
2075
2076         nvmet_fc_fod_op_done(fod);
2077 }
2078
2079 static void
2080 nvmet_fc_xmt_fcp_op_done(struct nvmefc_tgt_fcp_req *fcpreq)
2081 {
2082         struct nvmet_fc_fcp_iod *fod = fcpreq->nvmet_fc_private;
2083         struct nvmet_fc_tgt_queue *queue = fod->queue;
2084
2085         if (fod->tgtport->ops->target_features & NVMET_FCTGTFEAT_OPDONE_IN_ISR)
2086                 /* context switch so completion is not in ISR context */
2087                 queue_work_on(queue->cpu, queue->work_q, &fod->done_work);
2088         else
2089                 nvmet_fc_fod_op_done(fod);
2090 }
2091
2092 /*
2093  * actual completion handler after execution by the nvmet layer
2094  */
2095 static void
2096 __nvmet_fc_fcp_nvme_cmd_done(struct nvmet_fc_tgtport *tgtport,
2097                         struct nvmet_fc_fcp_iod *fod, int status)
2098 {
2099         struct nvme_common_command *sqe = &fod->cmdiubuf.sqe.common;
2100         struct nvme_completion *cqe = &fod->rspiubuf.cqe;
2101         unsigned long flags;
2102         bool abort;
2103
2104         spin_lock_irqsave(&fod->flock, flags);
2105         abort = fod->abort;
2106         spin_unlock_irqrestore(&fod->flock, flags);
2107
2108         /* if we have a CQE, snoop the last sq_head value */
2109         if (!status)
2110                 fod->queue->sqhd = cqe->sq_head;
2111
2112         if (abort) {
2113                 nvmet_fc_abort_op(tgtport, fod);
2114                 return;
2115         }
2116
2117         /* if an error handling the cmd post initial parsing */
2118         if (status) {
2119                 /* fudge up a failed CQE status for our transport error */
2120                 memset(cqe, 0, sizeof(*cqe));
2121                 cqe->sq_head = fod->queue->sqhd;        /* echo last cqe sqhd */
2122                 cqe->sq_id = cpu_to_le16(fod->queue->qid);
2123                 cqe->command_id = sqe->command_id;
2124                 cqe->status = cpu_to_le16(status);
2125         } else {
2126
2127                 /*
2128                  * try to push the data even if the SQE status is non-zero.
2129                  * There may be a status where data still was intended to
2130                  * be moved
2131                  */
2132                 if ((fod->io_dir == NVMET_FCP_READ) && (fod->data_sg_cnt)) {
2133                         /* push the data over before sending rsp */
2134                         nvmet_fc_transfer_fcp_data(tgtport, fod,
2135                                                 NVMET_FCOP_READDATA);
2136                         return;
2137                 }
2138
2139                 /* writes & no data - fall thru */
2140         }
2141
2142         /* data no longer needed */
2143         nvmet_fc_free_tgt_pgs(fod);
2144
2145         nvmet_fc_xmt_fcp_rsp(tgtport, fod);
2146 }
2147
2148
2149 static void
2150 nvmet_fc_fcp_nvme_cmd_done(struct nvmet_req *nvme_req)
2151 {
2152         struct nvmet_fc_fcp_iod *fod = nvmet_req_to_fod(nvme_req);
2153         struct nvmet_fc_tgtport *tgtport = fod->tgtport;
2154
2155         __nvmet_fc_fcp_nvme_cmd_done(tgtport, fod, 0);
2156 }
2157
2158
2159 /*
2160  * Actual processing routine for received FC-NVME LS Requests from the LLD
2161  */
2162 static void
2163 nvmet_fc_handle_fcp_rqst(struct nvmet_fc_tgtport *tgtport,
2164                         struct nvmet_fc_fcp_iod *fod)
2165 {
2166         struct nvme_fc_cmd_iu *cmdiu = &fod->cmdiubuf;
2167         int ret;
2168
2169         /*
2170          * Fused commands are currently not supported in the linux
2171          * implementation.
2172          *
2173          * As such, the implementation of the FC transport does not
2174          * look at the fused commands and order delivery to the upper
2175          * layer until we have both based on csn.
2176          */
2177
2178         fod->fcpreq->done = nvmet_fc_xmt_fcp_op_done;
2179
2180         fod->total_length = be32_to_cpu(cmdiu->data_len);
2181         if (cmdiu->flags & FCNVME_CMD_FLAGS_WRITE) {
2182                 fod->io_dir = NVMET_FCP_WRITE;
2183                 if (!nvme_is_write(&cmdiu->sqe))
2184                         goto transport_error;
2185         } else if (cmdiu->flags & FCNVME_CMD_FLAGS_READ) {
2186                 fod->io_dir = NVMET_FCP_READ;
2187                 if (nvme_is_write(&cmdiu->sqe))
2188                         goto transport_error;
2189         } else {
2190                 fod->io_dir = NVMET_FCP_NODATA;
2191                 if (fod->total_length)
2192                         goto transport_error;
2193         }
2194
2195         fod->req.cmd = &fod->cmdiubuf.sqe;
2196         fod->req.rsp = &fod->rspiubuf.cqe;
2197         fod->req.port = fod->queue->port;
2198
2199         /* ensure nvmet handlers will set cmd handler callback */
2200         fod->req.execute = NULL;
2201
2202         /* clear any response payload */
2203         memset(&fod->rspiubuf, 0, sizeof(fod->rspiubuf));
2204
2205         fod->data_sg = NULL;
2206         fod->data_sg_cnt = 0;
2207
2208         ret = nvmet_req_init(&fod->req,
2209                                 &fod->queue->nvme_cq,
2210                                 &fod->queue->nvme_sq,
2211                                 &nvmet_fc_tgt_fcp_ops);
2212         if (!ret) {
2213                 /* bad SQE content or invalid ctrl state */
2214                 /* nvmet layer has already called op done to send rsp. */
2215                 return;
2216         }
2217
2218         /* keep a running counter of tail position */
2219         atomic_inc(&fod->queue->sqtail);
2220
2221         if (fod->total_length) {
2222                 ret = nvmet_fc_alloc_tgt_pgs(fod);
2223                 if (ret) {
2224                         nvmet_req_complete(&fod->req, ret);
2225                         return;
2226                 }
2227         }
2228         fod->req.sg = fod->data_sg;
2229         fod->req.sg_cnt = fod->data_sg_cnt;
2230         fod->offset = 0;
2231
2232         if (fod->io_dir == NVMET_FCP_WRITE) {
2233                 /* pull the data over before invoking nvmet layer */
2234                 nvmet_fc_transfer_fcp_data(tgtport, fod, NVMET_FCOP_WRITEDATA);
2235                 return;
2236         }
2237
2238         /*
2239          * Reads or no data:
2240          *
2241          * can invoke the nvmet_layer now. If read data, cmd completion will
2242          * push the data
2243          */
2244
2245         fod->req.execute(&fod->req);
2246
2247         return;
2248
2249 transport_error:
2250         nvmet_fc_abort_op(tgtport, fod);
2251 }
2252
2253 /*
2254  * Actual processing routine for received FC-NVME LS Requests from the LLD
2255  */
2256 static void
2257 nvmet_fc_handle_fcp_rqst_work(struct work_struct *work)
2258 {
2259         struct nvmet_fc_fcp_iod *fod =
2260                 container_of(work, struct nvmet_fc_fcp_iod, work);
2261         struct nvmet_fc_tgtport *tgtport = fod->tgtport;
2262
2263         nvmet_fc_handle_fcp_rqst(tgtport, fod);
2264 }
2265
2266 /**
2267  * nvmet_fc_rcv_fcp_req - transport entry point called by an LLDD
2268  *                       upon the reception of a NVME FCP CMD IU.
2269  *
2270  * Pass a FC-NVME FCP CMD IU received from the FC link to the nvmet-fc
2271  * layer for processing.
2272  *
2273  * The nvmet_fc layer allocates a local job structure (struct
2274  * nvmet_fc_fcp_iod) from the queue for the io and copies the
2275  * CMD IU buffer to the job structure. As such, on a successful
2276  * completion (returns 0), the LLDD may immediately free/reuse
2277  * the CMD IU buffer passed in the call.
2278  *
2279  * However, in some circumstances, due to the packetized nature of FC
2280  * and the api of the FC LLDD which may issue a hw command to send the
2281  * response, but the LLDD may not get the hw completion for that command
2282  * and upcall the nvmet_fc layer before a new command may be
2283  * asynchronously received - its possible for a command to be received
2284  * before the LLDD and nvmet_fc have recycled the job structure. It gives
2285  * the appearance of more commands received than fits in the sq.
2286  * To alleviate this scenario, a temporary queue is maintained in the
2287  * transport for pending LLDD requests waiting for a queue job structure.
2288  * In these "overrun" cases, a temporary queue element is allocated
2289  * the LLDD request and CMD iu buffer information remembered, and the
2290  * routine returns a -EOVERFLOW status. Subsequently, when a queue job
2291  * structure is freed, it is immediately reallocated for anything on the
2292  * pending request list. The LLDDs defer_rcv() callback is called,
2293  * informing the LLDD that it may reuse the CMD IU buffer, and the io
2294  * is then started normally with the transport.
2295  *
2296  * The LLDD, when receiving an -EOVERFLOW completion status, is to treat
2297  * the completion as successful but must not reuse the CMD IU buffer
2298  * until the LLDD's defer_rcv() callback has been called for the
2299  * corresponding struct nvmefc_tgt_fcp_req pointer.
2300  *
2301  * If there is any other condition in which an error occurs, the
2302  * transport will return a non-zero status indicating the error.
2303  * In all cases other than -EOVERFLOW, the transport has not accepted the
2304  * request and the LLDD should abort the exchange.
2305  *
2306  * @target_port: pointer to the (registered) target port the FCP CMD IU
2307  *              was received on.
2308  * @fcpreq:     pointer to a fcpreq request structure to be used to reference
2309  *              the exchange corresponding to the FCP Exchange.
2310  * @cmdiubuf:   pointer to the buffer containing the FCP CMD IU
2311  * @cmdiubuf_len: length, in bytes, of the received FCP CMD IU
2312  */
2313 int
2314 nvmet_fc_rcv_fcp_req(struct nvmet_fc_target_port *target_port,
2315                         struct nvmefc_tgt_fcp_req *fcpreq,
2316                         void *cmdiubuf, u32 cmdiubuf_len)
2317 {
2318         struct nvmet_fc_tgtport *tgtport = targetport_to_tgtport(target_port);
2319         struct nvme_fc_cmd_iu *cmdiu = cmdiubuf;
2320         struct nvmet_fc_tgt_queue *queue;
2321         struct nvmet_fc_fcp_iod *fod;
2322         struct nvmet_fc_defer_fcp_req *deferfcp;
2323         unsigned long flags;
2324
2325         /* validate iu, so the connection id can be used to find the queue */
2326         if ((cmdiubuf_len != sizeof(*cmdiu)) ||
2327                         (cmdiu->scsi_id != NVME_CMD_SCSI_ID) ||
2328                         (cmdiu->fc_id != NVME_CMD_FC_ID) ||
2329                         (be16_to_cpu(cmdiu->iu_len) != (sizeof(*cmdiu)/4)))
2330                 return -EIO;
2331
2332         queue = nvmet_fc_find_target_queue(tgtport,
2333                                 be64_to_cpu(cmdiu->connection_id));
2334         if (!queue)
2335                 return -ENOTCONN;
2336
2337         /*
2338          * note: reference taken by find_target_queue
2339          * After successful fod allocation, the fod will inherit the
2340          * ownership of that reference and will remove the reference
2341          * when the fod is freed.
2342          */
2343
2344         spin_lock_irqsave(&queue->qlock, flags);
2345
2346         fod = nvmet_fc_alloc_fcp_iod(queue);
2347         if (fod) {
2348                 spin_unlock_irqrestore(&queue->qlock, flags);
2349
2350                 fcpreq->nvmet_fc_private = fod;
2351                 fod->fcpreq = fcpreq;
2352
2353                 memcpy(&fod->cmdiubuf, cmdiubuf, cmdiubuf_len);
2354
2355                 nvmet_fc_queue_fcp_req(tgtport, queue, fcpreq);
2356
2357                 return 0;
2358         }
2359
2360         if (!tgtport->ops->defer_rcv) {
2361                 spin_unlock_irqrestore(&queue->qlock, flags);
2362                 /* release the queue lookup reference */
2363                 nvmet_fc_tgt_q_put(queue);
2364                 return -ENOENT;
2365         }
2366
2367         deferfcp = list_first_entry_or_null(&queue->avail_defer_list,
2368                         struct nvmet_fc_defer_fcp_req, req_list);
2369         if (deferfcp) {
2370                 /* Just re-use one that was previously allocated */
2371                 list_del(&deferfcp->req_list);
2372         } else {
2373                 spin_unlock_irqrestore(&queue->qlock, flags);
2374
2375                 /* Now we need to dynamically allocate one */
2376                 deferfcp = kmalloc(sizeof(*deferfcp), GFP_KERNEL);
2377                 if (!deferfcp) {
2378                         /* release the queue lookup reference */
2379                         nvmet_fc_tgt_q_put(queue);
2380                         return -ENOMEM;
2381                 }
2382                 spin_lock_irqsave(&queue->qlock, flags);
2383         }
2384
2385         /* For now, use rspaddr / rsplen to save payload information */
2386         fcpreq->rspaddr = cmdiubuf;
2387         fcpreq->rsplen  = cmdiubuf_len;
2388         deferfcp->fcp_req = fcpreq;
2389
2390         /* defer processing till a fod becomes available */
2391         list_add_tail(&deferfcp->req_list, &queue->pending_cmd_list);
2392
2393         /* NOTE: the queue lookup reference is still valid */
2394
2395         spin_unlock_irqrestore(&queue->qlock, flags);
2396
2397         return -EOVERFLOW;
2398 }
2399 EXPORT_SYMBOL_GPL(nvmet_fc_rcv_fcp_req);
2400
2401 /**
2402  * nvmet_fc_rcv_fcp_abort - transport entry point called by an LLDD
2403  *                       upon the reception of an ABTS for a FCP command
2404  *
2405  * Notify the transport that an ABTS has been received for a FCP command
2406  * that had been given to the transport via nvmet_fc_rcv_fcp_req(). The
2407  * LLDD believes the command is still being worked on
2408  * (template_ops->fcp_req_release() has not been called).
2409  *
2410  * The transport will wait for any outstanding work (an op to the LLDD,
2411  * which the lldd should complete with error due to the ABTS; or the
2412  * completion from the nvmet layer of the nvme command), then will
2413  * stop processing and call the nvmet_fc_rcv_fcp_req() callback to
2414  * return the i/o context to the LLDD.  The LLDD may send the BA_ACC
2415  * to the ABTS either after return from this function (assuming any
2416  * outstanding op work has been terminated) or upon the callback being
2417  * called.
2418  *
2419  * @target_port: pointer to the (registered) target port the FCP CMD IU
2420  *              was received on.
2421  * @fcpreq:     pointer to the fcpreq request structure that corresponds
2422  *              to the exchange that received the ABTS.
2423  */
2424 void
2425 nvmet_fc_rcv_fcp_abort(struct nvmet_fc_target_port *target_port,
2426                         struct nvmefc_tgt_fcp_req *fcpreq)
2427 {
2428         struct nvmet_fc_fcp_iod *fod = fcpreq->nvmet_fc_private;
2429         struct nvmet_fc_tgt_queue *queue;
2430         unsigned long flags;
2431
2432         if (!fod || fod->fcpreq != fcpreq)
2433                 /* job appears to have already completed, ignore abort */
2434                 return;
2435
2436         queue = fod->queue;
2437
2438         spin_lock_irqsave(&queue->qlock, flags);
2439         if (fod->active) {
2440                 /*
2441                  * mark as abort. The abort handler, invoked upon completion
2442                  * of any work, will detect the aborted status and do the
2443                  * callback.
2444                  */
2445                 spin_lock(&fod->flock);
2446                 fod->abort = true;
2447                 fod->aborted = true;
2448                 spin_unlock(&fod->flock);
2449         }
2450         spin_unlock_irqrestore(&queue->qlock, flags);
2451 }
2452 EXPORT_SYMBOL_GPL(nvmet_fc_rcv_fcp_abort);
2453
2454
2455 struct nvmet_fc_traddr {
2456         u64     nn;
2457         u64     pn;
2458 };
2459
2460 static int
2461 __nvme_fc_parse_u64(substring_t *sstr, u64 *val)
2462 {
2463         u64 token64;
2464
2465         if (match_u64(sstr, &token64))
2466                 return -EINVAL;
2467         *val = token64;
2468
2469         return 0;
2470 }
2471
2472 /*
2473  * This routine validates and extracts the WWN's from the TRADDR string.
2474  * As kernel parsers need the 0x to determine number base, universally
2475  * build string to parse with 0x prefix before parsing name strings.
2476  */
2477 static int
2478 nvme_fc_parse_traddr(struct nvmet_fc_traddr *traddr, char *buf, size_t blen)
2479 {
2480         char name[2 + NVME_FC_TRADDR_HEXNAMELEN + 1];
2481         substring_t wwn = { name, &name[sizeof(name)-1] };
2482         int nnoffset, pnoffset;
2483
2484         /* validate it string one of the 2 allowed formats */
2485         if (strnlen(buf, blen) == NVME_FC_TRADDR_MAXLENGTH &&
2486                         !strncmp(buf, "nn-0x", NVME_FC_TRADDR_OXNNLEN) &&
2487                         !strncmp(&buf[NVME_FC_TRADDR_MAX_PN_OFFSET],
2488                                 "pn-0x", NVME_FC_TRADDR_OXNNLEN)) {
2489                 nnoffset = NVME_FC_TRADDR_OXNNLEN;
2490                 pnoffset = NVME_FC_TRADDR_MAX_PN_OFFSET +
2491                                                 NVME_FC_TRADDR_OXNNLEN;
2492         } else if ((strnlen(buf, blen) == NVME_FC_TRADDR_MINLENGTH &&
2493                         !strncmp(buf, "nn-", NVME_FC_TRADDR_NNLEN) &&
2494                         !strncmp(&buf[NVME_FC_TRADDR_MIN_PN_OFFSET],
2495                                 "pn-", NVME_FC_TRADDR_NNLEN))) {
2496                 nnoffset = NVME_FC_TRADDR_NNLEN;
2497                 pnoffset = NVME_FC_TRADDR_MIN_PN_OFFSET + NVME_FC_TRADDR_NNLEN;
2498         } else
2499                 goto out_einval;
2500
2501         name[0] = '0';
2502         name[1] = 'x';
2503         name[2 + NVME_FC_TRADDR_HEXNAMELEN] = 0;
2504
2505         memcpy(&name[2], &buf[nnoffset], NVME_FC_TRADDR_HEXNAMELEN);
2506         if (__nvme_fc_parse_u64(&wwn, &traddr->nn))
2507                 goto out_einval;
2508
2509         memcpy(&name[2], &buf[pnoffset], NVME_FC_TRADDR_HEXNAMELEN);
2510         if (__nvme_fc_parse_u64(&wwn, &traddr->pn))
2511                 goto out_einval;
2512
2513         return 0;
2514
2515 out_einval:
2516         pr_warn("%s: bad traddr string\n", __func__);
2517         return -EINVAL;
2518 }
2519
2520 static int
2521 nvmet_fc_add_port(struct nvmet_port *port)
2522 {
2523         struct nvmet_fc_tgtport *tgtport;
2524         struct nvmet_fc_traddr traddr = { 0L, 0L };
2525         unsigned long flags;
2526         int ret;
2527
2528         /* validate the address info */
2529         if ((port->disc_addr.trtype != NVMF_TRTYPE_FC) ||
2530             (port->disc_addr.adrfam != NVMF_ADDR_FAMILY_FC))
2531                 return -EINVAL;
2532
2533         /* map the traddr address info to a target port */
2534
2535         ret = nvme_fc_parse_traddr(&traddr, port->disc_addr.traddr,
2536                         sizeof(port->disc_addr.traddr));
2537         if (ret)
2538                 return ret;
2539
2540         ret = -ENXIO;
2541         spin_lock_irqsave(&nvmet_fc_tgtlock, flags);
2542         list_for_each_entry(tgtport, &nvmet_fc_target_list, tgt_list) {
2543                 if ((tgtport->fc_target_port.node_name == traddr.nn) &&
2544                     (tgtport->fc_target_port.port_name == traddr.pn)) {
2545                         /* a FC port can only be 1 nvmet port id */
2546                         if (!tgtport->port) {
2547                                 tgtport->port = port;
2548                                 port->priv = tgtport;
2549                                 nvmet_fc_tgtport_get(tgtport);
2550                                 ret = 0;
2551                         } else
2552                                 ret = -EALREADY;
2553                         break;
2554                 }
2555         }
2556         spin_unlock_irqrestore(&nvmet_fc_tgtlock, flags);
2557         return ret;
2558 }
2559
2560 static void
2561 nvmet_fc_remove_port(struct nvmet_port *port)
2562 {
2563         struct nvmet_fc_tgtport *tgtport = port->priv;
2564         unsigned long flags;
2565         bool matched = false;
2566
2567         spin_lock_irqsave(&nvmet_fc_tgtlock, flags);
2568         if (tgtport->port == port) {
2569                 matched = true;
2570                 tgtport->port = NULL;
2571         }
2572         spin_unlock_irqrestore(&nvmet_fc_tgtlock, flags);
2573
2574         if (matched)
2575                 nvmet_fc_tgtport_put(tgtport);
2576 }
2577
2578 static struct nvmet_fabrics_ops nvmet_fc_tgt_fcp_ops = {
2579         .owner                  = THIS_MODULE,
2580         .type                   = NVMF_TRTYPE_FC,
2581         .msdbd                  = 1,
2582         .add_port               = nvmet_fc_add_port,
2583         .remove_port            = nvmet_fc_remove_port,
2584         .queue_response         = nvmet_fc_fcp_nvme_cmd_done,
2585         .delete_ctrl            = nvmet_fc_delete_ctrl,
2586 };
2587
2588 static int __init nvmet_fc_init_module(void)
2589 {
2590         return nvmet_register_transport(&nvmet_fc_tgt_fcp_ops);
2591 }
2592
2593 static void __exit nvmet_fc_exit_module(void)
2594 {
2595         /* sanity check - all lports should be removed */
2596         if (!list_empty(&nvmet_fc_target_list))
2597                 pr_warn("%s: targetport list not empty\n", __func__);
2598
2599         nvmet_unregister_transport(&nvmet_fc_tgt_fcp_ops);
2600
2601         ida_destroy(&nvmet_fc_tgtport_cnt);
2602 }
2603
2604 module_init(nvmet_fc_init_module);
2605 module_exit(nvmet_fc_exit_module);
2606
2607 MODULE_LICENSE("GPL v2");