linux/drivers/scsi/xen-scsifront.c
<<
>>
Prefs
   1/*
   2 * Xen SCSI frontend driver
   3 *
   4 * Copyright (c) 2008, FUJITSU Limited
   5 *
   6 * This program is free software; you can redistribute it and/or
   7 * modify it under the terms of the GNU General Public License version 2
   8 * as published by the Free Software Foundation; or, when distributed
   9 * separately from the Linux kernel or incorporated into other
  10 * software packages, subject to the following license:
  11 *
  12 * Permission is hereby granted, free of charge, to any person obtaining a copy
  13 * of this source file (the "Software"), to deal in the Software without
  14 * restriction, including without limitation the rights to use, copy, modify,
  15 * merge, publish, distribute, sublicense, and/or sell copies of the Software,
  16 * and to permit persons to whom the Software is furnished to do so, subject to
  17 * the following conditions:
  18 *
  19 * The above copyright notice and this permission notice shall be included in
  20 * all copies or substantial portions of the Software.
  21 *
  22 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  23 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  24 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  25 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  26 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  27 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  28 * IN THE SOFTWARE.
  29 */
  30
  31#include <linux/module.h>
  32#include <linux/kernel.h>
  33#include <linux/device.h>
  34#include <linux/wait.h>
  35#include <linux/interrupt.h>
  36#include <linux/mutex.h>
  37#include <linux/spinlock.h>
  38#include <linux/sched.h>
  39#include <linux/blkdev.h>
  40#include <linux/pfn.h>
  41#include <linux/slab.h>
  42#include <linux/bitops.h>
  43
  44#include <scsi/scsi_cmnd.h>
  45#include <scsi/scsi_device.h>
  46#include <scsi/scsi.h>
  47#include <scsi/scsi_host.h>
  48
  49#include <xen/xen.h>
  50#include <xen/xenbus.h>
  51#include <xen/grant_table.h>
  52#include <xen/events.h>
  53#include <xen/page.h>
  54
  55#include <xen/interface/grant_table.h>
  56#include <xen/interface/io/vscsiif.h>
  57#include <xen/interface/io/protocols.h>
  58
  59#include <asm/xen/hypervisor.h>
  60
  61
  62#define GRANT_INVALID_REF       0
  63
  64#define VSCSIFRONT_OP_ADD_LUN   1
  65#define VSCSIFRONT_OP_DEL_LUN   2
  66#define VSCSIFRONT_OP_READD_LUN 3
  67
  68/* Tuning point. */
  69#define VSCSIIF_DEFAULT_CMD_PER_LUN 10
  70#define VSCSIIF_MAX_TARGET          64
  71#define VSCSIIF_MAX_LUN             255
  72
  73#define VSCSIIF_RING_SIZE       __CONST_RING_SIZE(vscsiif, PAGE_SIZE)
  74#define VSCSIIF_MAX_REQS        VSCSIIF_RING_SIZE
  75
  76#define vscsiif_grants_sg(_sg)  (PFN_UP((_sg) *         \
  77                                sizeof(struct scsiif_request_segment)))
  78
  79struct vscsifrnt_shadow {
  80        /* command between backend and frontend */
  81        unsigned char act;
  82        uint16_t rqid;
  83
  84        unsigned int nr_grants;         /* number of grants in gref[] */
  85        struct scsiif_request_segment *sg;      /* scatter/gather elements */
  86
  87        /* Do reset or abort function. */
  88        wait_queue_head_t wq_reset;     /* reset work queue           */
  89        int wait_reset;                 /* reset work queue condition */
  90        int32_t rslt_reset;             /* reset response status:     */
  91                                        /* SUCCESS or FAILED or:      */
  92#define RSLT_RESET_WAITING      0
  93#define RSLT_RESET_ERR          -1
  94
  95        /* Requested struct scsi_cmnd is stored from kernel. */
  96        struct scsi_cmnd *sc;
  97        int gref[vscsiif_grants_sg(SG_ALL) + SG_ALL];
  98};
  99
 100struct vscsifrnt_info {
 101        struct xenbus_device *dev;
 102
 103        struct Scsi_Host *host;
 104        int host_active;
 105
 106        unsigned int evtchn;
 107        unsigned int irq;
 108
 109        grant_ref_t ring_ref;
 110        struct vscsiif_front_ring ring;
 111        struct vscsiif_response ring_rsp;
 112
 113        spinlock_t shadow_lock;
 114        DECLARE_BITMAP(shadow_free_bitmap, VSCSIIF_MAX_REQS);
 115        struct vscsifrnt_shadow *shadow[VSCSIIF_MAX_REQS];
 116
 117        /* Following items are protected by the host lock. */
 118        wait_queue_head_t wq_sync;
 119        wait_queue_head_t wq_pause;
 120        unsigned int wait_ring_available:1;
 121        unsigned int waiting_pause:1;
 122        unsigned int pause:1;
 123        unsigned callers;
 124
 125        char dev_state_path[64];
 126        struct task_struct *curr;
 127};
 128
 129static DEFINE_MUTEX(scsifront_mutex);
 130
 131static void scsifront_wake_up(struct vscsifrnt_info *info)
 132{
 133        info->wait_ring_available = 0;
 134        wake_up(&info->wq_sync);
 135}
 136
 137static int scsifront_get_rqid(struct vscsifrnt_info *info)
 138{
 139        unsigned long flags;
 140        int free;
 141
 142        spin_lock_irqsave(&info->shadow_lock, flags);
 143
 144        free = find_first_bit(info->shadow_free_bitmap, VSCSIIF_MAX_REQS);
 145        __clear_bit(free, info->shadow_free_bitmap);
 146
 147        spin_unlock_irqrestore(&info->shadow_lock, flags);
 148
 149        return free;
 150}
 151
 152static int _scsifront_put_rqid(struct vscsifrnt_info *info, uint32_t id)
 153{
 154        int empty = bitmap_empty(info->shadow_free_bitmap, VSCSIIF_MAX_REQS);
 155
 156        __set_bit(id, info->shadow_free_bitmap);
 157        info->shadow[id] = NULL;
 158
 159        return empty || info->wait_ring_available;
 160}
 161
 162static void scsifront_put_rqid(struct vscsifrnt_info *info, uint32_t id)
 163{
 164        unsigned long flags;
 165        int kick;
 166
 167        spin_lock_irqsave(&info->shadow_lock, flags);
 168        kick = _scsifront_put_rqid(info, id);
 169        spin_unlock_irqrestore(&info->shadow_lock, flags);
 170
 171        if (kick)
 172                scsifront_wake_up(info);
 173}
 174
 175static struct vscsiif_request *scsifront_pre_req(struct vscsifrnt_info *info)
 176{
 177        struct vscsiif_front_ring *ring = &(info->ring);
 178        struct vscsiif_request *ring_req;
 179        uint32_t id;
 180
 181        id = scsifront_get_rqid(info);  /* use id in response */
 182        if (id >= VSCSIIF_MAX_REQS)
 183                return NULL;
 184
 185        ring_req = RING_GET_REQUEST(&(info->ring), ring->req_prod_pvt);
 186
 187        ring->req_prod_pvt++;
 188
 189        ring_req->rqid = (uint16_t)id;
 190
 191        return ring_req;
 192}
 193
 194static void scsifront_do_request(struct vscsifrnt_info *info)
 195{
 196        struct vscsiif_front_ring *ring = &(info->ring);
 197        int notify;
 198
 199        RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(ring, notify);
 200        if (notify)
 201                notify_remote_via_irq(info->irq);
 202}
 203
 204static void scsifront_gnttab_done(struct vscsifrnt_info *info, uint32_t id)
 205{
 206        struct vscsifrnt_shadow *s = info->shadow[id];
 207        int i;
 208
 209        if (s->sc->sc_data_direction == DMA_NONE)
 210                return;
 211
 212        for (i = 0; i < s->nr_grants; i++) {
 213                if (unlikely(gnttab_query_foreign_access(s->gref[i]) != 0)) {
 214                        shost_printk(KERN_ALERT, info->host, KBUILD_MODNAME
 215                                     "grant still in use by backend\n");
 216                        BUG();
 217                }
 218                gnttab_end_foreign_access(s->gref[i], 0, 0UL);
 219        }
 220
 221        kfree(s->sg);
 222}
 223
 224static void scsifront_cdb_cmd_done(struct vscsifrnt_info *info,
 225                                   struct vscsiif_response *ring_rsp)
 226{
 227        struct scsi_cmnd *sc;
 228        uint32_t id;
 229        uint8_t sense_len;
 230
 231        id = ring_rsp->rqid;
 232        sc = info->shadow[id]->sc;
 233
 234        BUG_ON(sc == NULL);
 235
 236        scsifront_gnttab_done(info, id);
 237        scsifront_put_rqid(info, id);
 238
 239        sc->result = ring_rsp->rslt;
 240        scsi_set_resid(sc, ring_rsp->residual_len);
 241
 242        sense_len = min_t(uint8_t, VSCSIIF_SENSE_BUFFERSIZE,
 243                          ring_rsp->sense_len);
 244
 245        if (sense_len)
 246                memcpy(sc->sense_buffer, ring_rsp->sense_buffer, sense_len);
 247
 248        sc->scsi_done(sc);
 249}
 250
 251static void scsifront_sync_cmd_done(struct vscsifrnt_info *info,
 252                                    struct vscsiif_response *ring_rsp)
 253{
 254        uint16_t id = ring_rsp->rqid;
 255        unsigned long flags;
 256        struct vscsifrnt_shadow *shadow = info->shadow[id];
 257        int kick;
 258
 259        spin_lock_irqsave(&info->shadow_lock, flags);
 260        shadow->wait_reset = 1;
 261        switch (shadow->rslt_reset) {
 262        case RSLT_RESET_WAITING:
 263                shadow->rslt_reset = ring_rsp->rslt;
 264                break;
 265        case RSLT_RESET_ERR:
 266                kick = _scsifront_put_rqid(info, id);
 267                spin_unlock_irqrestore(&info->shadow_lock, flags);
 268                kfree(shadow);
 269                if (kick)
 270                        scsifront_wake_up(info);
 271                return;
 272        default:
 273                shost_printk(KERN_ERR, info->host, KBUILD_MODNAME
 274                             "bad reset state %d, possibly leaking %u\n",
 275                             shadow->rslt_reset, id);
 276                break;
 277        }
 278        spin_unlock_irqrestore(&info->shadow_lock, flags);
 279
 280        wake_up(&shadow->wq_reset);
 281}
 282
 283static void scsifront_do_response(struct vscsifrnt_info *info,
 284                                  struct vscsiif_response *ring_rsp)
 285{
 286        if (WARN(ring_rsp->rqid >= VSCSIIF_MAX_REQS ||
 287                 test_bit(ring_rsp->rqid, info->shadow_free_bitmap),
 288                 "illegal rqid %u returned by backend!\n", ring_rsp->rqid))
 289                return;
 290
 291        if (info->shadow[ring_rsp->rqid]->act == VSCSIIF_ACT_SCSI_CDB)
 292                scsifront_cdb_cmd_done(info, ring_rsp);
 293        else
 294                scsifront_sync_cmd_done(info, ring_rsp);
 295}
 296
 297static int scsifront_ring_drain(struct vscsifrnt_info *info)
 298{
 299        struct vscsiif_response *ring_rsp;
 300        RING_IDX i, rp;
 301        int more_to_do = 0;
 302
 303        rp = info->ring.sring->rsp_prod;
 304        rmb();  /* ordering required respective to dom0 */
 305        for (i = info->ring.rsp_cons; i != rp; i++) {
 306                ring_rsp = RING_GET_RESPONSE(&info->ring, i);
 307                scsifront_do_response(info, ring_rsp);
 308        }
 309
 310        info->ring.rsp_cons = i;
 311
 312        if (i != info->ring.req_prod_pvt)
 313                RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
 314        else
 315                info->ring.sring->rsp_event = i + 1;
 316
 317        return more_to_do;
 318}
 319
 320static int scsifront_cmd_done(struct vscsifrnt_info *info)
 321{
 322        int more_to_do;
 323        unsigned long flags;
 324
 325        spin_lock_irqsave(info->host->host_lock, flags);
 326
 327        more_to_do = scsifront_ring_drain(info);
 328
 329        info->wait_ring_available = 0;
 330
 331        spin_unlock_irqrestore(info->host->host_lock, flags);
 332
 333        wake_up(&info->wq_sync);
 334
 335        return more_to_do;
 336}
 337
 338static irqreturn_t scsifront_irq_fn(int irq, void *dev_id)
 339{
 340        struct vscsifrnt_info *info = dev_id;
 341
 342        while (scsifront_cmd_done(info))
 343                /* Yield point for this unbounded loop. */
 344                cond_resched();
 345
 346        return IRQ_HANDLED;
 347}
 348
 349static void scsifront_finish_all(struct vscsifrnt_info *info)
 350{
 351        unsigned i;
 352        struct vscsiif_response resp;
 353
 354        scsifront_ring_drain(info);
 355
 356        for (i = 0; i < VSCSIIF_MAX_REQS; i++) {
 357                if (test_bit(i, info->shadow_free_bitmap))
 358                        continue;
 359                resp.rqid = i;
 360                resp.sense_len = 0;
 361                resp.rslt = DID_RESET << 16;
 362                resp.residual_len = 0;
 363                scsifront_do_response(info, &resp);
 364        }
 365}
 366
 367static int map_data_for_request(struct vscsifrnt_info *info,
 368                                struct scsi_cmnd *sc,
 369                                struct vscsiif_request *ring_req,
 370                                struct vscsifrnt_shadow *shadow)
 371{
 372        grant_ref_t gref_head;
 373        struct page *page;
 374        int err, ref, ref_cnt = 0;
 375        int grant_ro = (sc->sc_data_direction == DMA_TO_DEVICE);
 376        unsigned int i, off, len, bytes;
 377        unsigned int data_len = scsi_bufflen(sc);
 378        unsigned int data_grants = 0, seg_grants = 0;
 379        struct scatterlist *sg;
 380        struct scsiif_request_segment *seg;
 381
 382        ring_req->nr_segments = 0;
 383        if (sc->sc_data_direction == DMA_NONE || !data_len)
 384                return 0;
 385
 386        scsi_for_each_sg(sc, sg, scsi_sg_count(sc), i)
 387                data_grants += PFN_UP(sg->offset + sg->length);
 388
 389        if (data_grants > VSCSIIF_SG_TABLESIZE) {
 390                if (data_grants > info->host->sg_tablesize) {
 391                        shost_printk(KERN_ERR, info->host, KBUILD_MODNAME
 392                             "Unable to map request_buffer for command!\n");
 393                        return -E2BIG;
 394                }
 395                seg_grants = vscsiif_grants_sg(data_grants);
 396                shadow->sg = kcalloc(data_grants,
 397                        sizeof(struct scsiif_request_segment), GFP_ATOMIC);
 398                if (!shadow->sg)
 399                        return -ENOMEM;
 400        }
 401        seg = shadow->sg ? : ring_req->seg;
 402
 403        err = gnttab_alloc_grant_references(seg_grants + data_grants,
 404                                            &gref_head);
 405        if (err) {
 406                kfree(shadow->sg);
 407                shost_printk(KERN_ERR, info->host, KBUILD_MODNAME
 408                             "gnttab_alloc_grant_references() error\n");
 409                return -ENOMEM;
 410        }
 411
 412        if (seg_grants) {
 413                page = virt_to_page(seg);
 414                off = (unsigned long)seg & ~PAGE_MASK;
 415                len = sizeof(struct scsiif_request_segment) * data_grants;
 416                while (len > 0) {
 417                        bytes = min_t(unsigned int, len, PAGE_SIZE - off);
 418
 419                        ref = gnttab_claim_grant_reference(&gref_head);
 420                        BUG_ON(ref == -ENOSPC);
 421
 422                        gnttab_grant_foreign_access_ref(ref,
 423                                info->dev->otherend_id,
 424                                xen_page_to_gfn(page), 1);
 425                        shadow->gref[ref_cnt] = ref;
 426                        ring_req->seg[ref_cnt].gref   = ref;
 427                        ring_req->seg[ref_cnt].offset = (uint16_t)off;
 428                        ring_req->seg[ref_cnt].length = (uint16_t)bytes;
 429
 430                        page++;
 431                        len -= bytes;
 432                        off = 0;
 433                        ref_cnt++;
 434                }
 435                BUG_ON(seg_grants < ref_cnt);
 436                seg_grants = ref_cnt;
 437        }
 438
 439        scsi_for_each_sg(sc, sg, scsi_sg_count(sc), i) {
 440                page = sg_page(sg);
 441                off = sg->offset;
 442                len = sg->length;
 443
 444                while (len > 0 && data_len > 0) {
 445                        /*
 446                         * sg sends a scatterlist that is larger than
 447                         * the data_len it wants transferred for certain
 448                         * IO sizes.
 449                         */
 450                        bytes = min_t(unsigned int, len, PAGE_SIZE - off);
 451                        bytes = min(bytes, data_len);
 452
 453                        ref = gnttab_claim_grant_reference(&gref_head);
 454                        BUG_ON(ref == -ENOSPC);
 455
 456                        gnttab_grant_foreign_access_ref(ref,
 457                                info->dev->otherend_id,
 458                                xen_page_to_gfn(page),
 459                                grant_ro);
 460
 461                        shadow->gref[ref_cnt] = ref;
 462                        seg->gref   = ref;
 463                        seg->offset = (uint16_t)off;
 464                        seg->length = (uint16_t)bytes;
 465
 466                        page++;
 467                        seg++;
 468                        len -= bytes;
 469                        data_len -= bytes;
 470                        off = 0;
 471                        ref_cnt++;
 472                }
 473        }
 474
 475        if (seg_grants)
 476                ring_req->nr_segments = VSCSIIF_SG_GRANT | seg_grants;
 477        else
 478                ring_req->nr_segments = (uint8_t)ref_cnt;
 479        shadow->nr_grants = ref_cnt;
 480
 481        return 0;
 482}
 483
 484static struct vscsiif_request *scsifront_command2ring(
 485                struct vscsifrnt_info *info, struct scsi_cmnd *sc,
 486                struct vscsifrnt_shadow *shadow)
 487{
 488        struct vscsiif_request *ring_req;
 489
 490        memset(shadow, 0, sizeof(*shadow));
 491
 492        ring_req = scsifront_pre_req(info);
 493        if (!ring_req)
 494                return NULL;
 495
 496        info->shadow[ring_req->rqid] = shadow;
 497        shadow->rqid = ring_req->rqid;
 498
 499        ring_req->id      = sc->device->id;
 500        ring_req->lun     = sc->device->lun;
 501        ring_req->channel = sc->device->channel;
 502        ring_req->cmd_len = sc->cmd_len;
 503
 504        BUG_ON(sc->cmd_len > VSCSIIF_MAX_COMMAND_SIZE);
 505
 506        memcpy(ring_req->cmnd, sc->cmnd, sc->cmd_len);
 507
 508        ring_req->sc_data_direction   = (uint8_t)sc->sc_data_direction;
 509        ring_req->timeout_per_command = sc->request->timeout / HZ;
 510
 511        return ring_req;
 512}
 513
 514static int scsifront_enter(struct vscsifrnt_info *info)
 515{
 516        if (info->pause)
 517                return 1;
 518        info->callers++;
 519        return 0;
 520}
 521
 522static void scsifront_return(struct vscsifrnt_info *info)
 523{
 524        info->callers--;
 525        if (info->callers)
 526                return;
 527
 528        if (!info->waiting_pause)
 529                return;
 530
 531        info->waiting_pause = 0;
 532        wake_up(&info->wq_pause);
 533}
 534
 535static int scsifront_queuecommand(struct Scsi_Host *shost,
 536                                  struct scsi_cmnd *sc)
 537{
 538        struct vscsifrnt_info *info = shost_priv(shost);
 539        struct vscsiif_request *ring_req;
 540        struct vscsifrnt_shadow *shadow = scsi_cmd_priv(sc);
 541        unsigned long flags;
 542        int err;
 543        uint16_t rqid;
 544
 545        spin_lock_irqsave(shost->host_lock, flags);
 546        if (scsifront_enter(info)) {
 547                spin_unlock_irqrestore(shost->host_lock, flags);
 548                return SCSI_MLQUEUE_HOST_BUSY;
 549        }
 550        if (RING_FULL(&info->ring))
 551                goto busy;
 552
 553        ring_req = scsifront_command2ring(info, sc, shadow);
 554        if (!ring_req)
 555                goto busy;
 556
 557        sc->result = 0;
 558
 559        rqid = ring_req->rqid;
 560        ring_req->act = VSCSIIF_ACT_SCSI_CDB;
 561
 562        shadow->sc  = sc;
 563        shadow->act = VSCSIIF_ACT_SCSI_CDB;
 564
 565        err = map_data_for_request(info, sc, ring_req, shadow);
 566        if (err < 0) {
 567                pr_debug("%s: err %d\n", __func__, err);
 568                scsifront_put_rqid(info, rqid);
 569                scsifront_return(info);
 570                spin_unlock_irqrestore(shost->host_lock, flags);
 571                if (err == -ENOMEM)
 572                        return SCSI_MLQUEUE_HOST_BUSY;
 573                sc->result = DID_ERROR << 16;
 574                sc->scsi_done(sc);
 575                return 0;
 576        }
 577
 578        scsifront_do_request(info);
 579        scsifront_return(info);
 580        spin_unlock_irqrestore(shost->host_lock, flags);
 581
 582        return 0;
 583
 584busy:
 585        scsifront_return(info);
 586        spin_unlock_irqrestore(shost->host_lock, flags);
 587        pr_debug("%s: busy\n", __func__);
 588        return SCSI_MLQUEUE_HOST_BUSY;
 589}
 590
 591/*
 592 * Any exception handling (reset or abort) must be forwarded to the backend.
 593 * We have to wait until an answer is returned. This answer contains the
 594 * result to be returned to the requestor.
 595 */
 596static int scsifront_action_handler(struct scsi_cmnd *sc, uint8_t act)
 597{
 598        struct Scsi_Host *host = sc->device->host;
 599        struct vscsifrnt_info *info = shost_priv(host);
 600        struct vscsifrnt_shadow *shadow, *s = scsi_cmd_priv(sc);
 601        struct vscsiif_request *ring_req;
 602        int err = 0;
 603
 604        shadow = kmalloc(sizeof(*shadow), GFP_NOIO);
 605        if (!shadow)
 606                return FAILED;
 607
 608        spin_lock_irq(host->host_lock);
 609
 610        for (;;) {
 611                if (!RING_FULL(&info->ring)) {
 612                        ring_req = scsifront_command2ring(info, sc, shadow);
 613                        if (ring_req)
 614                                break;
 615                }
 616                if (err || info->pause) {
 617                        spin_unlock_irq(host->host_lock);
 618                        kfree(shadow);
 619                        return FAILED;
 620                }
 621                info->wait_ring_available = 1;
 622                spin_unlock_irq(host->host_lock);
 623                err = wait_event_interruptible(info->wq_sync,
 624                                               !info->wait_ring_available);
 625                spin_lock_irq(host->host_lock);
 626        }
 627
 628        if (scsifront_enter(info)) {
 629                spin_unlock_irq(host->host_lock);
 630                return FAILED;
 631        }
 632
 633        ring_req->act = act;
 634        ring_req->ref_rqid = s->rqid;
 635
 636        shadow->act = act;
 637        shadow->rslt_reset = RSLT_RESET_WAITING;
 638        init_waitqueue_head(&shadow->wq_reset);
 639
 640        ring_req->nr_segments = 0;
 641
 642        scsifront_do_request(info);
 643
 644        spin_unlock_irq(host->host_lock);
 645        err = wait_event_interruptible(shadow->wq_reset, shadow->wait_reset);
 646        spin_lock_irq(host->host_lock);
 647
 648        if (!err) {
 649                err = shadow->rslt_reset;
 650                scsifront_put_rqid(info, shadow->rqid);
 651                kfree(shadow);
 652        } else {
 653                spin_lock(&info->shadow_lock);
 654                shadow->rslt_reset = RSLT_RESET_ERR;
 655                spin_unlock(&info->shadow_lock);
 656                err = FAILED;
 657        }
 658
 659        scsifront_return(info);
 660        spin_unlock_irq(host->host_lock);
 661        return err;
 662}
 663
 664static int scsifront_eh_abort_handler(struct scsi_cmnd *sc)
 665{
 666        pr_debug("%s\n", __func__);
 667        return scsifront_action_handler(sc, VSCSIIF_ACT_SCSI_ABORT);
 668}
 669
 670static int scsifront_dev_reset_handler(struct scsi_cmnd *sc)
 671{
 672        pr_debug("%s\n", __func__);
 673        return scsifront_action_handler(sc, VSCSIIF_ACT_SCSI_RESET);
 674}
 675
 676static int scsifront_sdev_configure(struct scsi_device *sdev)
 677{
 678        struct vscsifrnt_info *info = shost_priv(sdev->host);
 679
 680        if (info && current == info->curr)
 681                xenbus_printf(XBT_NIL, info->dev->nodename,
 682                              info->dev_state_path, "%d", XenbusStateConnected);
 683
 684        return 0;
 685}
 686
 687static void scsifront_sdev_destroy(struct scsi_device *sdev)
 688{
 689        struct vscsifrnt_info *info = shost_priv(sdev->host);
 690
 691        if (info && current == info->curr)
 692                xenbus_printf(XBT_NIL, info->dev->nodename,
 693                              info->dev_state_path, "%d", XenbusStateClosed);
 694}
 695
 696static struct scsi_host_template scsifront_sht = {
 697        .module                 = THIS_MODULE,
 698        .name                   = "Xen SCSI frontend driver",
 699        .queuecommand           = scsifront_queuecommand,
 700        .eh_abort_handler       = scsifront_eh_abort_handler,
 701        .eh_device_reset_handler = scsifront_dev_reset_handler,
 702        .slave_configure        = scsifront_sdev_configure,
 703        .slave_destroy          = scsifront_sdev_destroy,
 704        .cmd_per_lun            = VSCSIIF_DEFAULT_CMD_PER_LUN,
 705        .can_queue              = VSCSIIF_MAX_REQS,
 706        .this_id                = -1,
 707        .cmd_size               = sizeof(struct vscsifrnt_shadow),
 708        .sg_tablesize           = VSCSIIF_SG_TABLESIZE,
 709        .use_clustering         = DISABLE_CLUSTERING,
 710        .proc_name              = "scsifront",
 711};
 712
 713static int scsifront_alloc_ring(struct vscsifrnt_info *info)
 714{
 715        struct xenbus_device *dev = info->dev;
 716        struct vscsiif_sring *sring;
 717        grant_ref_t gref;
 718        int err = -ENOMEM;
 719
 720        /***** Frontend to Backend ring start *****/
 721        sring = (struct vscsiif_sring *)__get_free_page(GFP_KERNEL);
 722        if (!sring) {
 723                xenbus_dev_fatal(dev, err,
 724                        "fail to allocate shared ring (Front to Back)");
 725                return err;
 726        }
 727        SHARED_RING_INIT(sring);
 728        FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
 729
 730        err = xenbus_grant_ring(dev, sring, 1, &gref);
 731        if (err < 0) {
 732                free_page((unsigned long)sring);
 733                xenbus_dev_fatal(dev, err,
 734                        "fail to grant shared ring (Front to Back)");
 735                return err;
 736        }
 737        info->ring_ref = gref;
 738
 739        err = xenbus_alloc_evtchn(dev, &info->evtchn);
 740        if (err) {
 741                xenbus_dev_fatal(dev, err, "xenbus_alloc_evtchn");
 742                goto free_gnttab;
 743        }
 744
 745        err = bind_evtchn_to_irq(info->evtchn);
 746        if (err <= 0) {
 747                xenbus_dev_fatal(dev, err, "bind_evtchn_to_irq");
 748                goto free_gnttab;
 749        }
 750
 751        info->irq = err;
 752
 753        err = request_threaded_irq(info->irq, NULL, scsifront_irq_fn,
 754                                   IRQF_ONESHOT, "scsifront", info);
 755        if (err) {
 756                xenbus_dev_fatal(dev, err, "request_threaded_irq");
 757                goto free_irq;
 758        }
 759
 760        return 0;
 761
 762/* free resource */
 763free_irq:
 764        unbind_from_irqhandler(info->irq, info);
 765free_gnttab:
 766        gnttab_end_foreign_access(info->ring_ref, 0,
 767                                  (unsigned long)info->ring.sring);
 768
 769        return err;
 770}
 771
 772static void scsifront_free_ring(struct vscsifrnt_info *info)
 773{
 774        unbind_from_irqhandler(info->irq, info);
 775        gnttab_end_foreign_access(info->ring_ref, 0,
 776                                  (unsigned long)info->ring.sring);
 777}
 778
 779static int scsifront_init_ring(struct vscsifrnt_info *info)
 780{
 781        struct xenbus_device *dev = info->dev;
 782        struct xenbus_transaction xbt;
 783        int err;
 784
 785        pr_debug("%s\n", __func__);
 786
 787        err = scsifront_alloc_ring(info);
 788        if (err)
 789                return err;
 790        pr_debug("%s: %u %u\n", __func__, info->ring_ref, info->evtchn);
 791
 792again:
 793        err = xenbus_transaction_start(&xbt);
 794        if (err)
 795                xenbus_dev_fatal(dev, err, "starting transaction");
 796
 797        err = xenbus_printf(xbt, dev->nodename, "ring-ref", "%u",
 798                            info->ring_ref);
 799        if (err) {
 800                xenbus_dev_fatal(dev, err, "%s", "writing ring-ref");
 801                goto fail;
 802        }
 803
 804        err = xenbus_printf(xbt, dev->nodename, "event-channel", "%u",
 805                            info->evtchn);
 806
 807        if (err) {
 808                xenbus_dev_fatal(dev, err, "%s", "writing event-channel");
 809                goto fail;
 810        }
 811
 812        err = xenbus_transaction_end(xbt, 0);
 813        if (err) {
 814                if (err == -EAGAIN)
 815                        goto again;
 816                xenbus_dev_fatal(dev, err, "completing transaction");
 817                goto free_sring;
 818        }
 819
 820        return 0;
 821
 822fail:
 823        xenbus_transaction_end(xbt, 1);
 824free_sring:
 825        scsifront_free_ring(info);
 826
 827        return err;
 828}
 829
 830
 831static int scsifront_probe(struct xenbus_device *dev,
 832                           const struct xenbus_device_id *id)
 833{
 834        struct vscsifrnt_info *info;
 835        struct Scsi_Host *host;
 836        int err = -ENOMEM;
 837        char name[TASK_COMM_LEN];
 838
 839        host = scsi_host_alloc(&scsifront_sht, sizeof(*info));
 840        if (!host) {
 841                xenbus_dev_fatal(dev, err, "fail to allocate scsi host");
 842                return err;
 843        }
 844        info = (struct vscsifrnt_info *)host->hostdata;
 845
 846        dev_set_drvdata(&dev->dev, info);
 847        info->dev = dev;
 848
 849        bitmap_fill(info->shadow_free_bitmap, VSCSIIF_MAX_REQS);
 850
 851        err = scsifront_init_ring(info);
 852        if (err) {
 853                scsi_host_put(host);
 854                return err;
 855        }
 856
 857        init_waitqueue_head(&info->wq_sync);
 858        init_waitqueue_head(&info->wq_pause);
 859        spin_lock_init(&info->shadow_lock);
 860
 861        snprintf(name, TASK_COMM_LEN, "vscsiif.%d", host->host_no);
 862
 863        host->max_id      = VSCSIIF_MAX_TARGET;
 864        host->max_channel = 0;
 865        host->max_lun     = VSCSIIF_MAX_LUN;
 866        host->max_sectors = (host->sg_tablesize - 1) * PAGE_SIZE / 512;
 867        host->max_cmd_len = VSCSIIF_MAX_COMMAND_SIZE;
 868
 869        err = scsi_add_host(host, &dev->dev);
 870        if (err) {
 871                dev_err(&dev->dev, "fail to add scsi host %d\n", err);
 872                goto free_sring;
 873        }
 874        info->host = host;
 875        info->host_active = 1;
 876
 877        xenbus_switch_state(dev, XenbusStateInitialised);
 878
 879        return 0;
 880
 881free_sring:
 882        scsifront_free_ring(info);
 883        scsi_host_put(host);
 884        return err;
 885}
 886
 887static int scsifront_resume(struct xenbus_device *dev)
 888{
 889        struct vscsifrnt_info *info = dev_get_drvdata(&dev->dev);
 890        struct Scsi_Host *host = info->host;
 891        int err;
 892
 893        spin_lock_irq(host->host_lock);
 894
 895        /* Finish all still pending commands. */
 896        scsifront_finish_all(info);
 897
 898        spin_unlock_irq(host->host_lock);
 899
 900        /* Reconnect to dom0. */
 901        scsifront_free_ring(info);
 902        err = scsifront_init_ring(info);
 903        if (err) {
 904                dev_err(&dev->dev, "fail to resume %d\n", err);
 905                scsi_host_put(host);
 906                return err;
 907        }
 908
 909        xenbus_switch_state(dev, XenbusStateInitialised);
 910
 911        return 0;
 912}
 913
 914static int scsifront_suspend(struct xenbus_device *dev)
 915{
 916        struct vscsifrnt_info *info = dev_get_drvdata(&dev->dev);
 917        struct Scsi_Host *host = info->host;
 918        int err = 0;
 919
 920        /* No new commands for the backend. */
 921        spin_lock_irq(host->host_lock);
 922        info->pause = 1;
 923        while (info->callers && !err) {
 924                info->waiting_pause = 1;
 925                info->wait_ring_available = 0;
 926                spin_unlock_irq(host->host_lock);
 927                wake_up(&info->wq_sync);
 928                err = wait_event_interruptible(info->wq_pause,
 929                                               !info->waiting_pause);
 930                spin_lock_irq(host->host_lock);
 931        }
 932        spin_unlock_irq(host->host_lock);
 933        return err;
 934}
 935
 936static int scsifront_remove(struct xenbus_device *dev)
 937{
 938        struct vscsifrnt_info *info = dev_get_drvdata(&dev->dev);
 939
 940        pr_debug("%s: %s removed\n", __func__, dev->nodename);
 941
 942        mutex_lock(&scsifront_mutex);
 943        if (info->host_active) {
 944                /* Scsi_host not yet removed */
 945                scsi_remove_host(info->host);
 946                info->host_active = 0;
 947        }
 948        mutex_unlock(&scsifront_mutex);
 949
 950        scsifront_free_ring(info);
 951        scsi_host_put(info->host);
 952
 953        return 0;
 954}
 955
 956static void scsifront_disconnect(struct vscsifrnt_info *info)
 957{
 958        struct xenbus_device *dev = info->dev;
 959        struct Scsi_Host *host = info->host;
 960
 961        pr_debug("%s: %s disconnect\n", __func__, dev->nodename);
 962
 963        /*
 964         * When this function is executed, all devices of
 965         * Frontend have been deleted.
 966         * Therefore, it need not block I/O before remove_host.
 967         */
 968
 969        mutex_lock(&scsifront_mutex);
 970        if (info->host_active) {
 971                scsi_remove_host(host);
 972                info->host_active = 0;
 973        }
 974        mutex_unlock(&scsifront_mutex);
 975
 976        xenbus_frontend_closed(dev);
 977}
 978
 979static void scsifront_do_lun_hotplug(struct vscsifrnt_info *info, int op)
 980{
 981        struct xenbus_device *dev = info->dev;
 982        int i, err = 0;
 983        char str[64];
 984        char **dir;
 985        unsigned int dir_n = 0;
 986        unsigned int device_state;
 987        unsigned int hst, chn, tgt, lun;
 988        struct scsi_device *sdev;
 989
 990        dir = xenbus_directory(XBT_NIL, dev->otherend, "vscsi-devs", &dir_n);
 991        if (IS_ERR(dir))
 992                return;
 993
 994        /* mark current task as the one allowed to modify device states */
 995        BUG_ON(info->curr);
 996        info->curr = current;
 997
 998        for (i = 0; i < dir_n; i++) {
 999                /* read status */
1000                snprintf(str, sizeof(str), "vscsi-devs/%s/state", dir[i]);
1001                err = xenbus_scanf(XBT_NIL, dev->otherend, str, "%u",
1002                                   &device_state);
1003                if (XENBUS_EXIST_ERR(err))
1004                        continue;
1005
1006                /* virtual SCSI device */
1007                snprintf(str, sizeof(str), "vscsi-devs/%s/v-dev", dir[i]);
1008                err = xenbus_scanf(XBT_NIL, dev->otherend, str,
1009                                   "%u:%u:%u:%u", &hst, &chn, &tgt, &lun);
1010                if (XENBUS_EXIST_ERR(err))
1011                        continue;
1012
1013                /*
1014                 * Front device state path, used in slave_configure called
1015                 * on successfull scsi_add_device, and in slave_destroy called
1016                 * on remove of a device.
1017                 */
1018                snprintf(info->dev_state_path, sizeof(info->dev_state_path),
1019                         "vscsi-devs/%s/state", dir[i]);
1020
1021                switch (op) {
1022                case VSCSIFRONT_OP_ADD_LUN:
1023                        if (device_state != XenbusStateInitialised)
1024                                break;
1025
1026                        if (scsi_add_device(info->host, chn, tgt, lun)) {
1027                                dev_err(&dev->dev, "scsi_add_device\n");
1028                                xenbus_printf(XBT_NIL, dev->nodename,
1029                                              info->dev_state_path,
1030                                              "%d", XenbusStateClosed);
1031                        }
1032                        break;
1033                case VSCSIFRONT_OP_DEL_LUN:
1034                        if (device_state != XenbusStateClosing)
1035                                break;
1036
1037                        sdev = scsi_device_lookup(info->host, chn, tgt, lun);
1038                        if (sdev) {
1039                                scsi_remove_device(sdev);
1040                                scsi_device_put(sdev);
1041                        }
1042                        break;
1043                case VSCSIFRONT_OP_READD_LUN:
1044                        if (device_state == XenbusStateConnected)
1045                                xenbus_printf(XBT_NIL, dev->nodename,
1046                                              info->dev_state_path,
1047                                              "%d", XenbusStateConnected);
1048                        break;
1049                default:
1050                        break;
1051                }
1052        }
1053
1054        info->curr = NULL;
1055
1056        kfree(dir);
1057}
1058
1059static void scsifront_read_backend_params(struct xenbus_device *dev,
1060                                          struct vscsifrnt_info *info)
1061{
1062        unsigned int sg_grant, nr_segs;
1063        int ret;
1064        struct Scsi_Host *host = info->host;
1065
1066        ret = xenbus_scanf(XBT_NIL, dev->otherend, "feature-sg-grant", "%u",
1067                           &sg_grant);
1068        if (ret != 1)
1069                sg_grant = 0;
1070        nr_segs = min_t(unsigned int, sg_grant, SG_ALL);
1071        nr_segs = max_t(unsigned int, nr_segs, VSCSIIF_SG_TABLESIZE);
1072        nr_segs = min_t(unsigned int, nr_segs,
1073                        VSCSIIF_SG_TABLESIZE * PAGE_SIZE /
1074                        sizeof(struct scsiif_request_segment));
1075
1076        if (!info->pause && sg_grant)
1077                dev_info(&dev->dev, "using up to %d SG entries\n", nr_segs);
1078        else if (info->pause && nr_segs < host->sg_tablesize)
1079                dev_warn(&dev->dev,
1080                         "SG entries decreased from %d to %u - device may not work properly anymore\n",
1081                         host->sg_tablesize, nr_segs);
1082
1083        host->sg_tablesize = nr_segs;
1084        host->max_sectors = (nr_segs - 1) * PAGE_SIZE / 512;
1085}
1086
1087static void scsifront_backend_changed(struct xenbus_device *dev,
1088                                      enum xenbus_state backend_state)
1089{
1090        struct vscsifrnt_info *info = dev_get_drvdata(&dev->dev);
1091
1092        pr_debug("%s: %p %u %u\n", __func__, dev, dev->state, backend_state);
1093
1094        switch (backend_state) {
1095        case XenbusStateUnknown:
1096        case XenbusStateInitialising:
1097        case XenbusStateInitWait:
1098        case XenbusStateInitialised:
1099                break;
1100
1101        case XenbusStateConnected:
1102                scsifront_read_backend_params(dev, info);
1103
1104                if (info->pause) {
1105                        scsifront_do_lun_hotplug(info, VSCSIFRONT_OP_READD_LUN);
1106                        xenbus_switch_state(dev, XenbusStateConnected);
1107                        info->pause = 0;
1108                        return;
1109                }
1110
1111                if (xenbus_read_driver_state(dev->nodename) ==
1112                    XenbusStateInitialised)
1113                        scsifront_do_lun_hotplug(info, VSCSIFRONT_OP_ADD_LUN);
1114
1115                if (dev->state != XenbusStateConnected)
1116                        xenbus_switch_state(dev, XenbusStateConnected);
1117                break;
1118
1119        case XenbusStateClosed:
1120                if (dev->state == XenbusStateClosed)
1121                        break;
1122                /* Missed the backend's Closing state -- fallthrough */
1123        case XenbusStateClosing:
1124                scsifront_disconnect(info);
1125                break;
1126
1127        case XenbusStateReconfiguring:
1128                scsifront_do_lun_hotplug(info, VSCSIFRONT_OP_DEL_LUN);
1129                xenbus_switch_state(dev, XenbusStateReconfiguring);
1130                break;
1131
1132        case XenbusStateReconfigured:
1133                scsifront_do_lun_hotplug(info, VSCSIFRONT_OP_ADD_LUN);
1134                xenbus_switch_state(dev, XenbusStateConnected);
1135                break;
1136        }
1137}
1138
1139static const struct xenbus_device_id scsifront_ids[] = {
1140        { "vscsi" },
1141        { "" }
1142};
1143
1144static struct xenbus_driver scsifront_driver = {
1145        .ids                    = scsifront_ids,
1146        .probe                  = scsifront_probe,
1147        .remove                 = scsifront_remove,
1148        .resume                 = scsifront_resume,
1149        .suspend                = scsifront_suspend,
1150        .otherend_changed       = scsifront_backend_changed,
1151};
1152
1153static int __init scsifront_init(void)
1154{
1155        if (!xen_domain())
1156                return -ENODEV;
1157
1158        return xenbus_register_frontend(&scsifront_driver);
1159}
1160module_init(scsifront_init);
1161
1162static void __exit scsifront_exit(void)
1163{
1164        xenbus_unregister_driver(&scsifront_driver);
1165}
1166module_exit(scsifront_exit);
1167
1168MODULE_DESCRIPTION("Xen SCSI frontend driver");
1169MODULE_LICENSE("GPL");
1170MODULE_ALIAS("xen:vscsi");
1171MODULE_AUTHOR("Juergen Gross <jgross@suse.com>");
1172