drivers/block/aoe/aoecmd.c

   1 /* Copyright (c) 2007 Coraid, Inc.  See COPYING for GPL terms. */
   2 /*
   3  * aoecmd.c
   4  * Filesystem request handling methods
   5  */
   6
   7 #include <linux/ata.h>
   8 #include <linux/slab.h>
   9 #include <linux/hdreg.h>
  10 #include <linux/blkdev.h>
  11 #include <linux/skbuff.h>
  12 #include <linux/netdevice.h>
  13 #include <linux/genhd.h>
  14 #include <linux/moduleparam.h>
  15 #include <linux/workqueue.h>
  16 #include <linux/kthread.h>
  17 #include <net/net_namespace.h>
  18 #include <asm/unaligned.h>
  19 #include <linux/uio.h>
  20 #include "aoe.h"
  21
  22 #define MAXIOC (8192)   /* default meant to avoid most soft lockups */
  23
  24 static void ktcomplete(struct frame *, struct sk_buff *);
  25
  26 static struct buf *nextbuf(struct aoedev *);
  27
  28 static int aoe_deadsecs = 60 * 3;
  29 module_param(aoe_deadsecs, int, 0644);
  30 MODULE_PARM_DESC(aoe_deadsecs, "After aoe_deadsecs seconds, give up and fail dev.");
  31
  32 static int aoe_maxout = 16;
  33 module_param(aoe_maxout, int, 0644);
  34 MODULE_PARM_DESC(aoe_maxout,
  35         "Only aoe_maxout outstanding packets for every MAC on eX.Y.");
  36
  37 static wait_queue_head_t ktiowq;
  38 static struct ktstate kts;
  39
  40 /* io completion queue */
  41 static struct {
  42         struct list_head head;
  43         spinlock_t lock;
  44 } iocq;
  45
  46 static struct sk_buff *
  47 new_skb(ulong len)
  48 {
  49         struct sk_buff *skb;
  50
  51         skb = alloc_skb(len, GFP_ATOMIC);
  52         if (skb) {
  53                 skb_reset_mac_header(skb);
  54                 skb_reset_network_header(skb);
  55                 skb->protocol = __constant_htons(ETH_P_AOE);
  56                 skb_checksum_none_assert(skb);
  57         }
  58         return skb;
  59 }
  60
  61 static struct frame *
  62 getframe(struct aoedev *d, u32 tag)
  63 {
  64         struct frame *f;
  65         struct list_head *head, *pos, *nx;
  66         u32 n;
  67
  68         n = tag % NFACTIVE;
  69         head = &d->factive[n];
  70         list_for_each_safe(pos, nx, head) {
  71                 f = list_entry(pos, struct frame, head);
  72                 if (f->tag == tag) {
  73                         list_del(pos);
  74                         return f;
  75                 }
  76         }
  77         return NULL;
  78 }
  79
  80 /*
  81  * Leave the top bit clear so we have tagspace for userland.
  82  * The bottom 16 bits are the xmit tick for rexmit/rttavg processing.
  83  * This driver reserves tag -1 to mean "unused frame."
  84  */
  85 static int
  86 newtag(struct aoedev *d)
  87 {
  88         register ulong n;
  89
  90         n = jiffies & 0xffff;
  91         return n |= (++d->lasttag & 0x7fff) << 16;
  92 }
  93
  94 static u32
  95 aoehdr_atainit(struct aoedev *d, struct aoetgt *t, struct aoe_hdr *h)
  96 {
  97         u32 host_tag = newtag(d);
  98
  99         memcpy(h->src, t->ifp->nd->dev_addr, sizeof h->src);
 100         memcpy(h->dst, t->addr, sizeof h->dst);
 101         h->type = __constant_cpu_to_be16(ETH_P_AOE);
 102         h->verfl = AOE_HVER;
 103         h->major = cpu_to_be16(d->aoemajor);
 104         h->minor = d->aoeminor;
 105         h->cmd = AOECMD_ATA;
 106         h->tag = cpu_to_be32(host_tag);
 107
 108         return host_tag;
 109 }
 110
 111 static inline void
 112 put_lba(struct aoe_atahdr *ah, sector_t lba)
 113 {
 114         ah->lba0 = lba;
 115         ah->lba1 = lba >>= 8;
 116         ah->lba2 = lba >>= 8;
 117         ah->lba3 = lba >>= 8;
 118         ah->lba4 = lba >>= 8;
 119         ah->lba5 = lba >>= 8;
 120 }
 121
 122 static struct aoeif *
 123 ifrotate(struct aoetgt *t)
 124 {
 125         struct aoeif *ifp;
 126
 127         ifp = t->ifp;
 128         ifp++;
 129         if (ifp >= &t->ifs[NAOEIFS] || ifp->nd == NULL)
 130                 ifp = t->ifs;
 131         if (ifp->nd == NULL)
 132                 return NULL;
 133         return t->ifp = ifp;
 134 }
 135
 136 static void
 137 skb_pool_put(struct aoedev *d, struct sk_buff *skb)
 138 {
 139         __skb_queue_tail(&d->skbpool, skb);
 140 }
 141
 142 static struct sk_buff *
 143 skb_pool_get(struct aoedev *d)
 144 {
 145         struct sk_buff *skb = skb_peek(&d->skbpool);
 146
 147         if (skb && atomic_read(&skb_shinfo(skb)->dataref) == 1) {
 148                 __skb_unlink(skb, &d->skbpool);
 149                 return skb;
 150         }
 151         if (skb_queue_len(&d->skbpool) < NSKBPOOLMAX &&
 152             (skb = new_skb(ETH_ZLEN)))
 153                 return skb;
 154
 155         return NULL;
 156 }
 157
 158 void
 159 aoe_freetframe(struct frame *f)
 160 {
 161         struct aoetgt *t;
 162
 163         t = f->t;
 164         f->buf = NULL;
 165         f->bv = NULL;
 166         f->r_skb = NULL;
 167         list_add(&f->head, &t->ffree);
 168 }
 169
 170 static struct frame *
 171 newtframe(struct aoedev *d, struct aoetgt *t)
 172 {
 173         struct frame *f;
 174         struct sk_buff *skb;
 175         struct list_head *pos;
 176
 177         if (list_empty(&t->ffree)) {
 178                 if (t->falloc >= NSKBPOOLMAX*2)
 179                         return NULL;
 180                 f = kcalloc(1, sizeof(*f), GFP_ATOMIC);
 181                 if (f == NULL)
 182                         return NULL;
 183                 t->falloc++;
 184                 f->t = t;
 185         } else {
 186                 pos = t->ffree.next;
 187                 list_del(pos);
 188                 f = list_entry(pos, struct frame, head);
 189         }
 190
 191         skb = f->skb;
 192         if (skb == NULL) {
 193                 f->skb = skb = new_skb(ETH_ZLEN);
 194                 if (!skb) {
 195 bail:                   aoe_freetframe(f);
 196                         return NULL;
 197                 }
 198         }
 199
 200         if (atomic_read(&skb_shinfo(skb)->dataref) != 1) {
 201                 skb = skb_pool_get(d);
 202                 if (skb == NULL)
 203                         goto bail;
 204                 skb_pool_put(d, f->skb);
 205                 f->skb = skb;
 206         }
 207
 208         skb->truesize -= skb->data_len;
 209         skb_shinfo(skb)->nr_frags = skb->data_len = 0;
 210         skb_trim(skb, 0);
 211         return f;
 212 }
 213
 214 static struct frame *
 215 newframe(struct aoedev *d)
 216 {
 217         struct frame *f;
 218         struct aoetgt *t, **tt;
 219         int totout = 0;
 220
 221         if (d->targets[0] == NULL) {    /* shouldn't happen, but I'm paranoid */
 222                 printk(KERN_ERR "aoe: NULL TARGETS!\n");
 223                 return NULL;
 224         }
 225         tt = d->tgt;    /* last used target */
 226         for (;;) {
 227                 tt++;
 228                 if (tt >= &d->targets[NTARGETS] || !*tt)
 229                         tt = d->targets;
 230                 t = *tt;
 231                 totout += t->nout;
 232                 if (t->nout < t->maxout
 233                 && t != d->htgt
 234                 && t->ifp->nd) {
 235                         f = newtframe(d, t);
 236                         if (f) {
 237                                 ifrotate(t);
 238                                 d->tgt = tt;
 239                                 return f;
 240                         }
 241                 }
 242                 if (tt == d->tgt)       /* we've looped and found nada */
 243                         break;
 244         }
 245         if (totout == 0) {
 246                 d->kicked++;
 247                 d->flags |= DEVFL_KICKME;
 248         }
 249         return NULL;
 250 }
 251
 252 static void
 253 skb_fillup(struct sk_buff *skb, struct bio_vec *bv, ulong off, ulong cnt)
 254 {
 255         int frag = 0;
 256         ulong fcnt;
 257 loop:
 258         fcnt = bv->bv_len - (off - bv->bv_offset);
 259         if (fcnt > cnt)
 260                 fcnt = cnt;
 261         skb_fill_page_desc(skb, frag++, bv->bv_page, off, fcnt);
 262         cnt -= fcnt;
 263         if (cnt <= 0)
 264                 return;
 265         bv++;
 266         off = bv->bv_offset;
 267         goto loop;
 268 }
 269
 270 static void
 271 fhash(struct frame *f)
 272 {
 273         struct aoedev *d = f->t->d;
 274         u32 n;
 275
 276         n = f->tag % NFACTIVE;
 277         list_add_tail(&f->head, &d->factive[n]);
 278 }
 279
 280 static int
 281 aoecmd_ata_rw(struct aoedev *d)
 282 {
 283         struct frame *f;
 284         struct aoe_hdr *h;
 285         struct aoe_atahdr *ah;
 286         struct buf *buf;
 287         struct bio_vec *bv;
 288         struct aoetgt *t;
 289         struct sk_buff *skb;
 290         struct sk_buff_head queue;
 291         ulong bcnt, fbcnt;
 292         char writebit, extbit;
 293
 294         writebit = 0x10;
 295         extbit = 0x4;
 296
 297         buf = nextbuf(d);
 298         if (buf == NULL)
 299                 return 0;
 300         f = newframe(d);
 301         if (f == NULL)
 302                 return 0;
 303         t = *d->tgt;
 304         bv = buf->bv;
 305         bcnt = d->maxbcnt;
 306         if (bcnt == 0)
 307                 bcnt = DEFAULTBCNT;
 308         if (bcnt > buf->resid)
 309                 bcnt = buf->resid;
 310         fbcnt = bcnt;
 311         f->bv = buf->bv;
 312         f->bv_off = f->bv->bv_offset + (f->bv->bv_len - buf->bv_resid);
 313         do {
 314                 if (fbcnt < buf->bv_resid) {
 315                         buf->bv_resid -= fbcnt;
 316                         buf->resid -= fbcnt;
 317                         break;
 318                 }
 319                 fbcnt -= buf->bv_resid;
 320                 buf->resid -= buf->bv_resid;
 321                 if (buf->resid == 0) {
 322                         d->ip.buf = NULL;
 323                         break;
 324                 }
 325                 buf->bv++;
 326                 buf->bv_resid = buf->bv->bv_len;
 327                 WARN_ON(buf->bv_resid == 0);
 328         } while (fbcnt);
 329
 330         /* initialize the headers & frame */
 331         skb = f->skb;
 332         h = (struct aoe_hdr *) skb_mac_header(skb);
 333         ah = (struct aoe_atahdr *) (h+1);
 334         skb_put(skb, sizeof *h + sizeof *ah);
 335         memset(h, 0, skb->len);
 336         f->tag = aoehdr_atainit(d, t, h);
 337         fhash(f);
 338         t->nout++;
 339         f->waited = 0;
 340         f->buf = buf;
 341         f->bcnt = bcnt;
 342         f->lba = buf->sector;
 343
 344         /* set up ata header */
 345         ah->scnt = bcnt >> 9;
 346         put_lba(ah, buf->sector);
 347         if (d->flags & DEVFL_EXT) {
 348                 ah->aflags |= AOEAFL_EXT;
 349         } else {
 350                 extbit = 0;
 351                 ah->lba3 &= 0x0f;
 352                 ah->lba3 |= 0xe0;       /* LBA bit + obsolete 0xa0 */
 353         }
 354         if (bio_data_dir(buf->bio) == WRITE) {
 355                 skb_fillup(skb, f->bv, f->bv_off, bcnt);
 356                 ah->aflags |= AOEAFL_WRITE;
 357                 skb->len += bcnt;
 358                 skb->data_len = bcnt;
 359                 skb->truesize += bcnt;
 360                 t->wpkts++;
 361         } else {
 362                 t->rpkts++;
 363                 writebit = 0;
 364         }
 365
 366         ah->cmdstat = ATA_CMD_PIO_READ | writebit | extbit;
 367
 368         /* mark all tracking fields and load out */
 369         buf->nframesout += 1;
 370         buf->sector += bcnt >> 9;
 371
 372         skb->dev = t->ifp->nd;
 373         skb = skb_clone(skb, GFP_ATOMIC);
 374         if (skb) {
 375                 __skb_queue_head_init(&queue);
 376                 __skb_queue_tail(&queue, skb);
 377                 aoenet_xmit(&queue);
 378         }
 379         return 1;
 380 }
 381
 382 /* some callers cannot sleep, and they can call this function,
 383  * transmitting the packets later, when interrupts are on
 384  */
 385 static void
 386 aoecmd_cfg_pkts(ushort aoemajor, unsigned char aoeminor, struct sk_buff_head *queue)
 387 {
 388         struct aoe_hdr *h;
 389         struct aoe_cfghdr *ch;
 390         struct sk_buff *skb;
 391         struct net_device *ifp;
 392
 393         rcu_read_lock();
 394         for_each_netdev_rcu(&init_net, ifp) {
 395                 dev_hold(ifp);
 396                 if (!is_aoe_netif(ifp))
 397                         goto cont;
 398
 399                 skb = new_skb(sizeof *h + sizeof *ch);
 400                 if (skb == NULL) {
 401                         printk(KERN_INFO "aoe: skb alloc failure\n");
 402                         goto cont;
 403                 }
 404                 skb_put(skb, sizeof *h + sizeof *ch);
 405                 skb->dev = ifp;
 406                 __skb_queue_tail(queue, skb);
 407                 h = (struct aoe_hdr *) skb_mac_header(skb);
 408                 memset(h, 0, sizeof *h + sizeof *ch);
 409
 410                 memset(h->dst, 0xff, sizeof h->dst);
 411                 memcpy(h->src, ifp->dev_addr, sizeof h->src);
 412                 h->type = __constant_cpu_to_be16(ETH_P_AOE);
 413                 h->verfl = AOE_HVER;
 414                 h->major = cpu_to_be16(aoemajor);
 415                 h->minor = aoeminor;
 416                 h->cmd = AOECMD_CFG;
 417
 418 cont:
 419                 dev_put(ifp);
 420         }
 421         rcu_read_unlock();
 422 }
 423
 424 static void
 425 resend(struct aoedev *d, struct frame *f)
 426 {
 427         struct sk_buff *skb;
 428         struct sk_buff_head queue;
 429         struct aoe_hdr *h;
 430         struct aoe_atahdr *ah;
 431         struct aoetgt *t;
 432         char buf[128];
 433         u32 n;
 434
 435         t = f->t;
 436         n = newtag(d);
 437         skb = f->skb;
 438         if (ifrotate(t) == NULL) {
 439                 /* probably can't happen, but set it up to fail anyway */
 440                 pr_info("aoe: resend: no interfaces to rotate to.\n");
 441                 ktcomplete(f, NULL);
 442                 return;
 443         }
 444         h = (struct aoe_hdr *) skb_mac_header(skb);
 445         ah = (struct aoe_atahdr *) (h+1);
 446
 447         snprintf(buf, sizeof buf,
 448                 "%15s e%ld.%d oldtag=%08x@%08lx newtag=%08x s=%pm d=%pm nout=%d\n",
 449                 "retransmit", d->aoemajor, d->aoeminor, f->tag, jiffies, n,
 450                 h->src, h->dst, t->nout);
 451         aoechr_error(buf);
 452
 453         f->tag = n;
 454         fhash(f);
 455         h->tag = cpu_to_be32(n);
 456         memcpy(h->dst, t->addr, sizeof h->dst);
 457         memcpy(h->src, t->ifp->nd->dev_addr, sizeof h->src);
 458
 459         skb->dev = t->ifp->nd;
 460         skb = skb_clone(skb, GFP_ATOMIC);
 461         if (skb == NULL)
 462                 return;
 463         __skb_queue_head_init(&queue);
 464         __skb_queue_tail(&queue, skb);
 465         aoenet_xmit(&queue);
 466 }
 467
 468 static int
 469 tsince(u32 tag)
 470 {
 471         int n;
 472
 473         n = jiffies & 0xffff;
 474         n -= tag & 0xffff;
 475         if (n < 0)
 476                 n += 1<<16;
 477         return n;
 478 }
 479
 480 static struct aoeif *
 481 getif(struct aoetgt *t, struct net_device *nd)
 482 {
 483         struct aoeif *p, *e;
 484
 485         p = t->ifs;
 486         e = p + NAOEIFS;
 487         for (; p < e; p++)
 488                 if (p->nd == nd)
 489                         return p;
 490         return NULL;
 491 }
 492
 493 static void
 494 ejectif(struct aoetgt *t, struct aoeif *ifp)
 495 {
 496         struct aoeif *e;
 497         ulong n;
 498
 499         e = t->ifs + NAOEIFS - 1;
 500         n = (e - ifp) * sizeof *ifp;
 501         memmove(ifp, ifp+1, n);
 502         e->nd = NULL;
 503 }
 504
 505 static int
 506 sthtith(struct aoedev *d)
 507 {
 508         struct frame *f, *nf;
 509         struct list_head *nx, *pos, *head;
 510         struct sk_buff *skb;
 511         struct aoetgt *ht = d->htgt;
 512         int i;
 513
 514         for (i = 0; i < NFACTIVE; i++) {
 515                 head = &d->factive[i];
 516                 list_for_each_safe(pos, nx, head) {
 517                         f = list_entry(pos, struct frame, head);
 518                         if (f->t != ht)
 519                                 continue;
 520
 521                         nf = newframe(d);
 522                         if (!nf)
 523                                 return 0;
 524
 525                         /* remove frame from active list */
 526                         list_del(pos);
 527
 528                         /* reassign all pertinent bits to new outbound frame */
 529                         skb = nf->skb;
 530                         nf->skb = f->skb;
 531                         nf->buf = f->buf;
 532                         nf->bcnt = f->bcnt;
 533                         nf->lba = f->lba;
 534                         nf->bv = f->bv;
 535                         nf->bv_off = f->bv_off;
 536                         nf->waited = 0;
 537                         f->skb = skb;
 538                         aoe_freetframe(f);
 539                         ht->nout--;
 540                         nf->t->nout++;
 541                         resend(d, nf);
 542                 }
 543         }
 544         /* We've cleaned up the outstanding so take away his
 545          * interfaces so he won't be used.  We should remove him from
 546          * the target array here, but cleaning up a target is
 547          * involved.  PUNT!
 548          */
 549         memset(ht->ifs, 0, sizeof ht->ifs);
 550         d->htgt = NULL;
 551         return 1;
 552 }
 553
 554 static inline unsigned char
 555 ata_scnt(unsigned char *packet) {
 556         struct aoe_hdr *h;
 557         struct aoe_atahdr *ah;
 558
 559         h = (struct aoe_hdr *) packet;
 560         ah = (struct aoe_atahdr *) (h+1);
 561         return ah->scnt;
 562 }
 563
 564 static void
 565 rexmit_timer(ulong vp)
 566 {
 567         struct aoedev *d;
 568         struct aoetgt *t, **tt, **te;
 569         struct aoeif *ifp;
 570         struct frame *f;
 571         struct list_head *head, *pos, *nx;
 572         LIST_HEAD(flist);
 573         register long timeout;
 574         ulong flags, n;
 575         int i;
 576
 577         d = (struct aoedev *) vp;
 578
 579         /* timeout is always ~150% of the moving average */
 580         timeout = d->rttavg;
 581         timeout += timeout >> 1;
 582
 583         spin_lock_irqsave(&d->lock, flags);
 584
 585         if (d->flags & DEVFL_TKILL) {
 586                 spin_unlock_irqrestore(&d->lock, flags);
 587                 return;
 588         }
 589
 590         /* collect all frames to rexmit into flist */
 591         for (i = 0; i < NFACTIVE; i++) {
 592                 head = &d->factive[i];
 593                 list_for_each_safe(pos, nx, head) {
 594                         f = list_entry(pos, struct frame, head);
 595                         if (tsince(f->tag) < timeout)
 596                                 break;  /* end of expired frames */
 597                         /* move to flist for later processing */
 598                         list_move_tail(pos, &flist);
 599                 }
 600         }
 601         /* window check */
 602         tt = d->targets;
 603         te = tt + d->ntargets;
 604         for (; tt < te && (t = *tt); tt++) {
 605                 if (t->nout == t->maxout
 606                 && t->maxout < t->nframes
 607                 && (jiffies - t->lastwadj)/HZ > 10) {
 608                         t->maxout++;
 609                         t->lastwadj = jiffies;
 610                 }
 611         }
 612
 613         if (!list_empty(&flist)) {      /* retransmissions necessary */
 614                 n = d->rttavg <<= 1;
 615                 if (n > MAXTIMER)
 616                         d->rttavg = MAXTIMER;
 617         }
 618
 619         /* process expired frames */
 620         while (!list_empty(&flist)) {
 621                 pos = flist.next;
 622                 f = list_entry(pos, struct frame, head);
 623                 n = f->waited += timeout;
 624                 n /= HZ;
 625                 if (n > aoe_deadsecs) {
 626                         /* Waited too long.  Device failure.
 627                          * Hang all frames on first hash bucket for downdev
 628                          * to clean up.
 629                          */
 630                         list_splice(&flist, &d->factive[0]);
 631                         aoedev_downdev(d);
 632                         break;
 633                 }
 634                 list_del(pos);
 635
 636                 t = f->t;
 637                 if (n > aoe_deadsecs/2)
 638                         d->htgt = t; /* see if another target can help */
 639
 640                 if (t->nout == t->maxout) {
 641                         if (t->maxout > 1)
 642                                 t->maxout--;
 643                         t->lastwadj = jiffies;
 644                 }
 645
 646                 ifp = getif(t, f->skb->dev);
 647                 if (ifp && ++ifp->lost > (t->nframes << 1)
 648                 && (ifp != t->ifs || t->ifs[1].nd)) {
 649                         ejectif(t, ifp);
 650                         ifp = NULL;
 651                 }
 652                 resend(d, f);
 653         }
 654
 655         if ((d->flags & DEVFL_KICKME || d->htgt) && d->blkq) {
 656                 d->flags &= ~DEVFL_KICKME;
 657                 d->blkq->request_fn(d->blkq);
 658         }
 659
 660         d->timer.expires = jiffies + TIMERTICK;
 661         add_timer(&d->timer);
 662
 663         spin_unlock_irqrestore(&d->lock, flags);
 664 }
 665
 666 static unsigned long
 667 rqbiocnt(struct request *r)
 668 {
 669         struct bio *bio;
 670         unsigned long n = 0;
 671
 672         __rq_for_each_bio(bio, r)
 673                 n++;
 674         return n;
 675 }
 676
 677 /* This can be removed if we are certain that no users of the block
 678  * layer will ever use zero-count pages in bios.  Otherwise we have to
 679  * protect against the put_page sometimes done by the network layer.
 680  *
 681  * See http://oss.sgi.com/archives/xfs/2007-01/msg00594.html for
 682  * discussion.
 683  *
 684  * We cannot use get_page in the workaround, because it insists on a
 685  * positive page count as a precondition.  So we use _count directly.
 686  */
 687 static void
 688 bio_pageinc(struct bio *bio)
 689 {
 690         struct bio_vec *bv;
 691         struct page *page;
 692         int i;
 693
 694         bio_for_each_segment(bv, bio, i) {
 695                 page = bv->bv_page;
 696                 /* Non-zero page count for non-head members of
 697                  * compound pages is no longer allowed by the kernel,
 698                  * but this has never been seen here.
 699                  */
 700                 if (unlikely(PageCompound(page)))
 701                         if (compound_trans_head(page) != page) {
 702                                 pr_crit("page tail used for block I/O\n");
 703                                 BUG();
 704                         }
 705                 atomic_inc(&page->_count);
 706         }
 707 }
 708
 709 static void
 710 bio_pagedec(struct bio *bio)
 711 {
 712         struct bio_vec *bv;
 713         int i;
 714
 715         bio_for_each_segment(bv, bio, i)
 716                 atomic_dec(&bv->bv_page->_count);
 717 }
 718
 719 static void
 720 bufinit(struct buf *buf, struct request *rq, struct bio *bio)
 721 {
 722         struct bio_vec *bv;
 723
 724         memset(buf, 0, sizeof(*buf));
 725         buf->rq = rq;
 726         buf->bio = bio;
 727         buf->resid = bio->bi_size;
 728         buf->sector = bio->bi_sector;
 729         bio_pageinc(bio);
 730         buf->bv = bv = &bio->bi_io_vec[bio->bi_idx];
 731         buf->bv_resid = bv->bv_len;
 732         WARN_ON(buf->bv_resid == 0);
 733 }
 734
 735 static struct buf *
 736 nextbuf(struct aoedev *d)
 737 {
 738         struct request *rq;
 739         struct request_queue *q;
 740         struct buf *buf;
 741         struct bio *bio;
 742
 743         q = d->blkq;
 744         if (q == NULL)
 745                 return NULL;    /* initializing */
 746         if (d->ip.buf)
 747                 return d->ip.buf;
 748         rq = d->ip.rq;
 749         if (rq == NULL) {
 750                 rq = blk_peek_request(q);
 751                 if (rq == NULL)
 752                         return NULL;
 753                 blk_start_request(rq);
 754                 d->ip.rq = rq;
 755                 d->ip.nxbio = rq->bio;
 756                 rq->special = (void *) rqbiocnt(rq);
 757         }
 758         buf = mempool_alloc(d->bufpool, GFP_ATOMIC);
 759         if (buf == NULL) {
 760                 pr_err("aoe: nextbuf: unable to mempool_alloc!\n");
 761                 return NULL;
 762         }
 763         bio = d->ip.nxbio;
 764         bufinit(buf, rq, bio);
 765         bio = bio->bi_next;
 766         d->ip.nxbio = bio;
 767         if (bio == NULL)
 768                 d->ip.rq = NULL;
 769         return d->ip.buf = buf;
 770 }
 771
 772 /* enters with d->lock held */
 773 void
 774 aoecmd_work(struct aoedev *d)
 775 {
 776         if (d->htgt && !sthtith(d))
 777                 return;
 778         while (aoecmd_ata_rw(d))
 779                 ;
 780 }
 781
 782 /* this function performs work that has been deferred until sleeping is OK
 783  */
 784 void
 785 aoecmd_sleepwork(struct work_struct *work)
 786 {
 787         struct aoedev *d = container_of(work, struct aoedev, work);
 788
 789         if (d->flags & DEVFL_GDALLOC)
 790                 aoeblk_gdalloc(d);
 791
 792         if (d->flags & DEVFL_NEWSIZE) {
 793                 struct block_device *bd;
 794                 unsigned long flags;
 795                 u64 ssize;
 796
 797                 ssize = get_capacity(d->gd);
 798                 bd = bdget_disk(d->gd, 0);
 799
 800                 if (bd) {
 801                         mutex_lock(&bd->bd_inode->i_mutex);
 802                         i_size_write(bd->bd_inode, (loff_t)ssize<<9);
 803                         mutex_unlock(&bd->bd_inode->i_mutex);
 804                         bdput(bd);
 805                 }
 806                 spin_lock_irqsave(&d->lock, flags);
 807                 d->flags |= DEVFL_UP;
 808                 d->flags &= ~DEVFL_NEWSIZE;
 809                 spin_unlock_irqrestore(&d->lock, flags);
 810         }
 811 }
 812
 813 static void
 814 ataid_complete(struct aoedev *d, struct aoetgt *t, unsigned char *id)
 815 {
 816         u64 ssize;
 817         u16 n;
 818
 819         /* word 83: command set supported */
 820         n = get_unaligned_le16(&id[83 << 1]);
 821
 822         /* word 86: command set/feature enabled */
 823         n |= get_unaligned_le16(&id[86 << 1]);
 824
 825         if (n & (1<<10)) {      /* bit 10: LBA 48 */
 826                 d->flags |= DEVFL_EXT;
 827
 828                 /* word 100: number lba48 sectors */
 829                 ssize = get_unaligned_le64(&id[100 << 1]);
 830
 831                 /* set as in ide-disk.c:init_idedisk_capacity */
 832                 d->geo.cylinders = ssize;
 833                 d->geo.cylinders /= (255 * 63);
 834                 d->geo.heads = 255;
 835                 d->geo.sectors = 63;
 836         } else {
 837                 d->flags &= ~DEVFL_EXT;
 838
 839                 /* number lba28 sectors */
 840                 ssize = get_unaligned_le32(&id[60 << 1]);
 841
 842                 /* NOTE: obsolete in ATA 6 */
 843                 d->geo.cylinders = get_unaligned_le16(&id[54 << 1]);
 844                 d->geo.heads = get_unaligned_le16(&id[55 << 1]);
 845                 d->geo.sectors = get_unaligned_le16(&id[56 << 1]);
 846         }
 847
 848         if (d->ssize != ssize)
 849                 printk(KERN_INFO
 850                         "aoe: %pm e%ld.%d v%04x has %llu sectors\n",
 851                         t->addr,
 852                         d->aoemajor, d->aoeminor,
 853                         d->fw_ver, (long long)ssize);
 854         d->ssize = ssize;
 855         d->geo.start = 0;
 856         if (d->flags & (DEVFL_GDALLOC|DEVFL_NEWSIZE))
 857                 return;
 858         if (d->gd != NULL) {
 859                 set_capacity(d->gd, ssize);
 860                 d->flags |= DEVFL_NEWSIZE;
 861         } else
 862                 d->flags |= DEVFL_GDALLOC;
 863         schedule_work(&d->work);
 864 }
 865
 866 static void
 867 calc_rttavg(struct aoedev *d, int rtt)
 868 {
 869         register long n;
 870
 871         n = rtt;
 872         if (n < 0) {
 873                 n = -rtt;
 874                 if (n < MINTIMER)
 875                         n = MINTIMER;
 876                 else if (n > MAXTIMER)
 877                         n = MAXTIMER;
 878                 d->mintimer += (n - d->mintimer) >> 1;
 879         } else if (n < d->mintimer)
 880                 n = d->mintimer;
 881         else if (n > MAXTIMER)
 882                 n = MAXTIMER;
 883
 884         /* g == .25; cf. Congestion Avoidance and Control, Jacobson & Karels; 1988 */
 885         n -= d->rttavg;
 886         d->rttavg += n >> 2;
 887 }
 888
 889 static struct aoetgt *
 890 gettgt(struct aoedev *d, char *addr)
 891 {
 892         struct aoetgt **t, **e;
 893
 894         t = d->targets;
 895         e = t + NTARGETS;
 896         for (; t < e && *t; t++)
 897                 if (memcmp((*t)->addr, addr, sizeof((*t)->addr)) == 0)
 898                         return *t;
 899         return NULL;
 900 }
 901
 902 static void
 903 bvcpy(struct bio_vec *bv, ulong off, struct sk_buff *skb, long cnt)
 904 {
 905         ulong fcnt;
 906         char *p;
 907         int soff = 0;
 908 loop:
 909         fcnt = bv->bv_len - (off - bv->bv_offset);
 910         if (fcnt > cnt)
 911                 fcnt = cnt;
 912         p = page_address(bv->bv_page) + off;
 913         skb_copy_bits(skb, soff, p, fcnt);
 914         soff += fcnt;
 915         cnt -= fcnt;
 916         if (cnt <= 0)
 917                 return;
 918         bv++;
 919         off = bv->bv_offset;
 920         goto loop;
 921 }
 922
 923 void
 924 aoe_end_request(struct aoedev *d, struct request *rq, int fastfail)
 925 {
 926         struct bio *bio;
 927         int bok;
 928         struct request_queue *q;
 929
 930         q = d->blkq;
 931         if (rq == d->ip.rq)
 932                 d->ip.rq = NULL;
 933         do {
 934                 bio = rq->bio;
 935                 bok = !fastfail && test_bit(BIO_UPTODATE, &bio->bi_flags);
 936         } while (__blk_end_request(rq, bok ? 0 : -EIO, bio->bi_size));
 937
 938         /* cf. http://lkml.org/lkml/2006/10/31/28 */
 939         if (!fastfail)
 940                 q->request_fn(q);
 941 }
 942
 943 static void
 944 aoe_end_buf(struct aoedev *d, struct buf *buf)
 945 {
 946         struct request *rq;
 947         unsigned long n;
 948
 949         if (buf == d->ip.buf)
 950                 d->ip.buf = NULL;
 951         rq = buf->rq;
 952         bio_pagedec(buf->bio);
 953         mempool_free(buf, d->bufpool);
 954         n = (unsigned long) rq->special;
 955         rq->special = (void *) --n;
 956         if (n == 0)
 957                 aoe_end_request(d, rq, 0);
 958 }
 959
 960 static void
 961 ktiocomplete(struct frame *f)
 962 {
 963         struct aoe_hdr *hin, *hout;
 964         struct aoe_atahdr *ahin, *ahout;
 965         struct buf *buf;
 966         struct sk_buff *skb;
 967         struct aoetgt *t;
 968         struct aoeif *ifp;
 969         struct aoedev *d;
 970         long n;
 971
 972         if (f == NULL)
 973                 return;
 974
 975         t = f->t;
 976         d = t->d;
 977
 978         hout = (struct aoe_hdr *) skb_mac_header(f->skb);
 979         ahout = (struct aoe_atahdr *) (hout+1);
 980         buf = f->buf;
 981         skb = f->r_skb;
 982         if (skb == NULL)
 983                 goto noskb;     /* just fail the buf. */
 984
 985         hin = (struct aoe_hdr *) skb->data;
 986         skb_pull(skb, sizeof(*hin));
 987         ahin = (struct aoe_atahdr *) skb->data;
 988         skb_pull(skb, sizeof(*ahin));
 989         if (ahin->cmdstat & 0xa9) {     /* these bits cleared on success */
 990                 pr_err("aoe: ata error cmd=%2.2Xh stat=%2.2Xh from e%ld.%d\n",
 991                         ahout->cmdstat, ahin->cmdstat,
 992                         d->aoemajor, d->aoeminor);
 993 noskb:  if (buf)
 994                         clear_bit(BIO_UPTODATE, &buf->bio->bi_flags);
 995                 goto badrsp;
 996         }
 997
 998         n = ahout->scnt << 9;
 999         switch (ahout->cmdstat) {
1000         case ATA_CMD_PIO_READ:
1001         case ATA_CMD_PIO_READ_EXT:
1002                 if (skb->len < n) {
1003                         pr_err("aoe: runt data size in read.  skb->len=%d need=%ld\n",
1004                                 skb->len, n);
1005                         clear_bit(BIO_UPTODATE, &buf->bio->bi_flags);
1006                         break;
1007                 }
1008                 bvcpy(f->bv, f->bv_off, skb, n);
1009         case ATA_CMD_PIO_WRITE:
1010         case ATA_CMD_PIO_WRITE_EXT:
1011                 spin_lock_irq(&d->lock);
1012                 ifp = getif(t, skb->dev);
1013                 if (ifp)
1014                         ifp->lost = 0;
1015                 if (d->htgt == t) /* I'll help myself, thank you. */
1016                         d->htgt = NULL;
1017                 spin_unlock_irq(&d->lock);
1018                 break;
1019         case ATA_CMD_ID_ATA:
1020                 if (skb->len < 512) {
1021                         pr_info("aoe: runt data size in ataid.  skb->len=%d\n",
1022                                 skb->len);
1023                         break;
1024                 }
1025                 if (skb_linearize(skb))
1026                         break;
1027                 spin_lock_irq(&d->lock);
1028                 ataid_complete(d, t, skb->data);
1029                 spin_unlock_irq(&d->lock);
1030                 break;
1031         default:
1032                 pr_info("aoe: unrecognized ata command %2.2Xh for %d.%d\n",
1033                         ahout->cmdstat,
1034                         be16_to_cpu(get_unaligned(&hin->major)),
1035                         hin->minor);
1036         }
1037 badrsp:
1038         spin_lock_irq(&d->lock);
1039
1040         aoe_freetframe(f);
1041
1042         if (buf && --buf->nframesout == 0 && buf->resid == 0)
1043                 aoe_end_buf(d, buf);
1044
1045         aoecmd_work(d);
1046
1047         spin_unlock_irq(&d->lock);
1048         aoedev_put(d);
1049         dev_kfree_skb(skb);
1050 }
1051
1052 /* Enters with iocq.lock held.
1053  * Returns true iff responses needing processing remain.
1054  */
1055 static int
1056 ktio(void)
1057 {
1058         struct frame *f;
1059         struct list_head *pos;
1060         int i;
1061
1062         for (i = 0; ; ++i) {
1063                 if (i == MAXIOC)
1064                         return 1;
1065                 if (list_empty(&iocq.head))
1066                         return 0;
1067                 pos = iocq.head.next;
1068                 list_del(pos);
1069                 spin_unlock_irq(&iocq.lock);
1070                 f = list_entry(pos, struct frame, head);
1071                 ktiocomplete(f);
1072                 spin_lock_irq(&iocq.lock);
1073         }
1074 }
1075
1076 static int
1077 kthread(void *vp)
1078 {
1079         struct ktstate *k;
1080         DECLARE_WAITQUEUE(wait, current);
1081         int more;
1082
1083         k = vp;
1084         current->flags |= PF_NOFREEZE;
1085         set_user_nice(current, -10);
1086         complete(&k->rendez);   /* tell spawner we're running */
1087         do {
1088                 spin_lock_irq(k->lock);
1089                 more = k->fn();
1090                 if (!more) {
1091                         add_wait_queue(k->waitq, &wait);
1092                         __set_current_state(TASK_INTERRUPTIBLE);
1093                 }
1094                 spin_unlock_irq(k->lock);
1095                 if (!more) {
1096                         schedule();
1097                         remove_wait_queue(k->waitq, &wait);
1098                 } else
1099                         cond_resched();
1100         } while (!kthread_should_stop());
1101         complete(&k->rendez);   /* tell spawner we're stopping */
1102         return 0;
1103 }
1104
1105 void
1106 aoe_ktstop(struct ktstate *k)
1107 {
1108         kthread_stop(k->task);
1109         wait_for_completion(&k->rendez);
1110 }
1111
1112 int
1113 aoe_ktstart(struct ktstate *k)
1114 {
1115         struct task_struct *task;
1116
1117         init_completion(&k->rendez);
1118         task = kthread_run(kthread, k, k->name);
1119         if (task == NULL || IS_ERR(task))
1120                 return -ENOMEM;
1121         k->task = task;
1122         wait_for_completion(&k->rendez); /* allow kthread to start */
1123         init_completion(&k->rendez);    /* for waiting for exit later */
1124         return 0;
1125 }
1126
1127 /* pass it off to kthreads for processing */
1128 static void
1129 ktcomplete(struct frame *f, struct sk_buff *skb)
1130 {
1131         ulong flags;
1132
1133         f->r_skb = skb;
1134         spin_lock_irqsave(&iocq.lock, flags);
1135         list_add_tail(&f->head, &iocq.head);
1136         spin_unlock_irqrestore(&iocq.lock, flags);
1137         wake_up(&ktiowq);
1138 }
1139
1140 struct sk_buff *
1141 aoecmd_ata_rsp(struct sk_buff *skb)
1142 {
1143         struct aoedev *d;
1144         struct aoe_hdr *h;
1145         struct frame *f;
1146         struct aoetgt *t;
1147         u32 n;
1148         ulong flags;
1149         char ebuf[128];
1150         u16 aoemajor;
1151
1152         h = (struct aoe_hdr *) skb->data;
1153         aoemajor = be16_to_cpu(get_unaligned(&h->major));
1154         d = aoedev_by_aoeaddr(aoemajor, h->minor);
1155         if (d == NULL) {
1156                 snprintf(ebuf, sizeof ebuf, "aoecmd_ata_rsp: ata response "
1157                         "for unknown device %d.%d\n",
1158                         aoemajor, h->minor);
1159                 aoechr_error(ebuf);
1160                 return skb;
1161         }
1162
1163         spin_lock_irqsave(&d->lock, flags);
1164
1165         n = be32_to_cpu(get_unaligned(&h->tag));
1166         f = getframe(d, n);
1167         if (f == NULL) {
1168                 calc_rttavg(d, -tsince(n));
1169                 spin_unlock_irqrestore(&d->lock, flags);
1170                 aoedev_put(d);
1171                 snprintf(ebuf, sizeof ebuf,
1172                         "%15s e%d.%d    tag=%08x@%08lx\n",
1173                         "unexpected rsp",
1174                         get_unaligned_be16(&h->major),
1175                         h->minor,
1176                         get_unaligned_be32(&h->tag),
1177                         jiffies);
1178                 aoechr_error(ebuf);
1179                 return skb;
1180         }
1181         t = f->t;
1182         calc_rttavg(d, tsince(f->tag));
1183         t->nout--;
1184         aoecmd_work(d);
1185
1186         spin_unlock_irqrestore(&d->lock, flags);
1187
1188         ktcomplete(f, skb);
1189
1190         /*
1191          * Note here that we do not perform an aoedev_put, as we are
1192          * leaving this reference for the ktio to release.
1193          */
1194         return NULL;
1195 }
1196
1197 void
1198 aoecmd_cfg(ushort aoemajor, unsigned char aoeminor)
1199 {
1200         struct sk_buff_head queue;
1201
1202         __skb_queue_head_init(&queue);
1203         aoecmd_cfg_pkts(aoemajor, aoeminor, &queue);
1204         aoenet_xmit(&queue);
1205 }
1206
1207 struct sk_buff *
1208 aoecmd_ata_id(struct aoedev *d)
1209 {
1210         struct aoe_hdr *h;
1211         struct aoe_atahdr *ah;
1212         struct frame *f;
1213         struct sk_buff *skb;
1214         struct aoetgt *t;
1215
1216         f = newframe(d);
1217         if (f == NULL)
1218                 return NULL;
1219
1220         t = *d->tgt;
1221
1222         /* initialize the headers & frame */
1223         skb = f->skb;
1224         h = (struct aoe_hdr *) skb_mac_header(skb);
1225         ah = (struct aoe_atahdr *) (h+1);
1226         skb_put(skb, sizeof *h + sizeof *ah);
1227         memset(h, 0, skb->len);
1228         f->tag = aoehdr_atainit(d, t, h);
1229         fhash(f);
1230         t->nout++;
1231         f->waited = 0;
1232
1233         /* set up ata header */
1234         ah->scnt = 1;
1235         ah->cmdstat = ATA_CMD_ID_ATA;
1236         ah->lba3 = 0xa0;
1237
1238         skb->dev = t->ifp->nd;
1239
1240         d->rttavg = MAXTIMER;
1241         d->timer.function = rexmit_timer;
1242
1243         return skb_clone(skb, GFP_ATOMIC);
1244 }
1245
1246 static struct aoetgt *
1247 addtgt(struct aoedev *d, char *addr, ulong nframes)
1248 {
1249         struct aoetgt *t, **tt, **te;
1250
1251         tt = d->targets;
1252         te = tt + NTARGETS;
1253         for (; tt < te && *tt; tt++)
1254                 ;
1255
1256         if (tt == te) {
1257                 printk(KERN_INFO
1258                         "aoe: device addtgt failure; too many targets\n");
1259                 return NULL;
1260         }
1261         t = kzalloc(sizeof(*t), GFP_ATOMIC);
1262         if (!t) {
1263                 printk(KERN_INFO "aoe: cannot allocate memory to add target\n");
1264                 return NULL;
1265         }
1266
1267         d->ntargets++;
1268         t->nframes = nframes;
1269         t->d = d;
1270         memcpy(t->addr, addr, sizeof t->addr);
1271         t->ifp = t->ifs;
1272         t->maxout = t->nframes;
1273         INIT_LIST_HEAD(&t->ffree);
1274         return *tt = t;
1275 }
1276
1277 static void
1278 setdbcnt(struct aoedev *d)
1279 {
1280         struct aoetgt **t, **e;
1281         int bcnt = 0;
1282
1283         t = d->targets;
1284         e = t + NTARGETS;
1285         for (; t < e && *t; t++)
1286                 if (bcnt == 0 || bcnt > (*t)->minbcnt)
1287                         bcnt = (*t)->minbcnt;
1288         if (bcnt != d->maxbcnt) {
1289                 d->maxbcnt = bcnt;
1290                 pr_info("aoe: e%ld.%d: setting %d byte data frames\n",
1291                         d->aoemajor, d->aoeminor, bcnt);
1292         }
1293 }
1294
1295 static void
1296 setifbcnt(struct aoetgt *t, struct net_device *nd, int bcnt)
1297 {
1298         struct aoedev *d;
1299         struct aoeif *p, *e;
1300         int minbcnt;
1301
1302         d = t->d;
1303         minbcnt = bcnt;
1304         p = t->ifs;
1305         e = p + NAOEIFS;
1306         for (; p < e; p++) {
1307                 if (p->nd == NULL)
1308                         break;          /* end of the valid interfaces */
1309                 if (p->nd == nd) {
1310                         p->bcnt = bcnt; /* we're updating */
1311                         nd = NULL;
1312                 } else if (minbcnt > p->bcnt)
1313                         minbcnt = p->bcnt; /* find the min interface */
1314         }
1315         if (nd) {
1316                 if (p == e) {
1317                         pr_err("aoe: device setifbcnt failure; too many interfaces.\n");
1318                         return;
1319                 }
1320                 p->nd = nd;
1321                 p->bcnt = bcnt;
1322         }
1323         t->minbcnt = minbcnt;
1324         setdbcnt(d);
1325 }
1326
1327 void
1328 aoecmd_cfg_rsp(struct sk_buff *skb)
1329 {
1330         struct aoedev *d;
1331         struct aoe_hdr *h;
1332         struct aoe_cfghdr *ch;
1333         struct aoetgt *t;
1334         ulong flags, sysminor, aoemajor;
1335         struct sk_buff *sl;
1336         struct sk_buff_head queue;
1337         u16 n;
1338
1339         sl = NULL;
1340         h = (struct aoe_hdr *) skb_mac_header(skb);
1341         ch = (struct aoe_cfghdr *) (h+1);
1342
1343         /*
1344          * Enough people have their dip switches set backwards to
1345          * warrant a loud message for this special case.
1346          */
1347         aoemajor = get_unaligned_be16(&h->major);
1348         if (aoemajor == 0xfff) {
1349                 printk(KERN_ERR "aoe: Warning: shelf address is all ones.  "
1350                         "Check shelf dip switches.\n");
1351                 return;
1352         }
1353         if (h->minor >= NPERSHELF) {
1354                 pr_err("aoe: e%ld.%d %s, %d\n",
1355                         aoemajor, h->minor,
1356                         "slot number larger than the maximum",
1357                         NPERSHELF-1);
1358                 return;
1359         }
1360
1361         sysminor = SYSMINOR(aoemajor, h->minor);
1362         if (sysminor * AOE_PARTITIONS + AOE_PARTITIONS > MINORMASK) {
1363                 printk(KERN_INFO "aoe: e%ld.%d: minor number too large\n",
1364                         aoemajor, (int) h->minor);
1365                 return;
1366         }
1367
1368         n = be16_to_cpu(ch->bufcnt);
1369         if (n > aoe_maxout)     /* keep it reasonable */
1370                 n = aoe_maxout;
1371
1372         d = aoedev_by_sysminor_m(sysminor);
1373         if (d == NULL) {
1374                 printk(KERN_INFO "aoe: device sysminor_m failure\n");
1375                 return;
1376         }
1377
1378         spin_lock_irqsave(&d->lock, flags);
1379
1380         t = gettgt(d, h->src);
1381         if (!t) {
1382                 t = addtgt(d, h->src, n);
1383                 if (!t)
1384                         goto bail;
1385         }
1386         n = skb->dev->mtu;
1387         n -= sizeof(struct aoe_hdr) + sizeof(struct aoe_atahdr);
1388         n /= 512;
1389         if (n > ch->scnt)
1390                 n = ch->scnt;
1391         n = n ? n * 512 : DEFAULTBCNT;
1392         setifbcnt(t, skb->dev, n);
1393
1394         /* don't change users' perspective */
1395         if (d->nopen == 0) {
1396                 d->fw_ver = be16_to_cpu(ch->fwver);
1397                 sl = aoecmd_ata_id(d);
1398         }
1399 bail:
1400         spin_unlock_irqrestore(&d->lock, flags);
1401         aoedev_put(d);
1402         if (sl) {
1403                 __skb_queue_head_init(&queue);
1404                 __skb_queue_tail(&queue, sl);
1405                 aoenet_xmit(&queue);
1406         }
1407 }
1408
1409 void
1410 aoecmd_cleanslate(struct aoedev *d)
1411 {
1412         struct aoetgt **t, **te;
1413
1414         d->mintimer = MINTIMER;
1415         d->maxbcnt = 0;
1416
1417         t = d->targets;
1418         te = t + NTARGETS;
1419         for (; t < te && *t; t++)
1420                 (*t)->maxout = (*t)->nframes;
1421 }
1422
1423 void
1424 aoe_failbuf(struct aoedev *d, struct buf *buf)
1425 {
1426         if (buf == NULL)
1427                 return;
1428         buf->resid = 0;
1429         clear_bit(BIO_UPTODATE, &buf->bio->bi_flags);
1430         if (buf->nframesout == 0)
1431                 aoe_end_buf(d, buf);
1432 }
1433
1434 void
1435 aoe_flush_iocq(void)
1436 {
1437         struct frame *f;
1438         struct aoedev *d;
1439         LIST_HEAD(flist);
1440         struct list_head *pos;
1441         struct sk_buff *skb;
1442         ulong flags;
1443
1444         spin_lock_irqsave(&iocq.lock, flags);
1445         list_splice_init(&iocq.head, &flist);
1446         spin_unlock_irqrestore(&iocq.lock, flags);
1447         while (!list_empty(&flist)) {
1448                 pos = flist.next;
1449                 list_del(pos);
1450                 f = list_entry(pos, struct frame, head);
1451                 d = f->t->d;
1452                 skb = f->r_skb;
1453                 spin_lock_irqsave(&d->lock, flags);
1454                 if (f->buf) {
1455                         f->buf->nframesout--;
1456                         aoe_failbuf(d, f->buf);
1457                 }
1458                 aoe_freetframe(f);
1459                 spin_unlock_irqrestore(&d->lock, flags);
1460                 dev_kfree_skb(skb);
1461                 aoedev_put(d);
1462         }
1463 }
1464
1465 int __init
1466 aoecmd_init(void)
1467 {
1468         INIT_LIST_HEAD(&iocq.head);
1469         spin_lock_init(&iocq.lock);
1470         init_waitqueue_head(&ktiowq);
1471         kts.name = "aoe_ktio";
1472         kts.fn = ktio;
1473         kts.waitq = &ktiowq;
1474         kts.lock = &iocq.lock;
1475         return aoe_ktstart(&kts);
1476 }
1477
1478 void
1479 aoecmd_exit(void)
1480 {
1481         aoe_ktstop(&kts);
1482         aoe_flush_iocq();
1483 }