drivers/block/aoe/aoecmd.c

   1 /* Copyright (c) 2012 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 aoetgt *t;
 288         struct sk_buff *skb;
 289         struct sk_buff_head queue;
 290         ulong bcnt, fbcnt;
 291         char writebit, extbit;
 292
 293         writebit = 0x10;
 294         extbit = 0x4;
 295
 296         buf = nextbuf(d);
 297         if (buf == NULL)
 298                 return 0;
 299         f = newframe(d);
 300         if (f == NULL)
 301                 return 0;
 302         t = *d->tgt;
 303         bcnt = d->maxbcnt;
 304         if (bcnt == 0)
 305                 bcnt = DEFAULTBCNT;
 306         if (bcnt > buf->resid)
 307                 bcnt = buf->resid;
 308         fbcnt = bcnt;
 309         f->bv = buf->bv;
 310         f->bv_off = f->bv->bv_offset + (f->bv->bv_len - buf->bv_resid);
 311         do {
 312                 if (fbcnt < buf->bv_resid) {
 313                         buf->bv_resid -= fbcnt;
 314                         buf->resid -= fbcnt;
 315                         break;
 316                 }
 317                 fbcnt -= buf->bv_resid;
 318                 buf->resid -= buf->bv_resid;
 319                 if (buf->resid == 0) {
 320                         d->ip.buf = NULL;
 321                         break;
 322                 }
 323                 buf->bv++;
 324                 buf->bv_resid = buf->bv->bv_len;
 325                 WARN_ON(buf->bv_resid == 0);
 326         } while (fbcnt);
 327
 328         /* initialize the headers & frame */
 329         skb = f->skb;
 330         h = (struct aoe_hdr *) skb_mac_header(skb);
 331         ah = (struct aoe_atahdr *) (h+1);
 332         skb_put(skb, sizeof *h + sizeof *ah);
 333         memset(h, 0, skb->len);
 334         f->tag = aoehdr_atainit(d, t, h);
 335         fhash(f);
 336         t->nout++;
 337         f->waited = 0;
 338         f->buf = buf;
 339         f->bcnt = bcnt;
 340         f->lba = buf->sector;
 341
 342         /* set up ata header */
 343         ah->scnt = bcnt >> 9;
 344         put_lba(ah, buf->sector);
 345         if (d->flags & DEVFL_EXT) {
 346                 ah->aflags |= AOEAFL_EXT;
 347         } else {
 348                 extbit = 0;
 349                 ah->lba3 &= 0x0f;
 350                 ah->lba3 |= 0xe0;       /* LBA bit + obsolete 0xa0 */
 351         }
 352         if (bio_data_dir(buf->bio) == WRITE) {
 353                 skb_fillup(skb, f->bv, f->bv_off, bcnt);
 354                 ah->aflags |= AOEAFL_WRITE;
 355                 skb->len += bcnt;
 356                 skb->data_len = bcnt;
 357                 skb->truesize += bcnt;
 358                 t->wpkts++;
 359         } else {
 360                 t->rpkts++;
 361                 writebit = 0;
 362         }
 363
 364         ah->cmdstat = ATA_CMD_PIO_READ | writebit | extbit;
 365
 366         /* mark all tracking fields and load out */
 367         buf->nframesout += 1;
 368         buf->sector += bcnt >> 9;
 369
 370         skb->dev = t->ifp->nd;
 371         skb = skb_clone(skb, GFP_ATOMIC);
 372         if (skb) {
 373                 __skb_queue_head_init(&queue);
 374                 __skb_queue_tail(&queue, skb);
 375                 aoenet_xmit(&queue);
 376         }
 377         return 1;
 378 }
 379
 380 /* some callers cannot sleep, and they can call this function,
 381  * transmitting the packets later, when interrupts are on
 382  */
 383 static void
 384 aoecmd_cfg_pkts(ushort aoemajor, unsigned char aoeminor, struct sk_buff_head *queue)
 385 {
 386         struct aoe_hdr *h;
 387         struct aoe_cfghdr *ch;
 388         struct sk_buff *skb;
 389         struct net_device *ifp;
 390
 391         rcu_read_lock();
 392         for_each_netdev_rcu(&init_net, ifp) {
 393                 dev_hold(ifp);
 394                 if (!is_aoe_netif(ifp))
 395                         goto cont;
 396
 397                 skb = new_skb(sizeof *h + sizeof *ch);
 398                 if (skb == NULL) {
 399                         printk(KERN_INFO "aoe: skb alloc failure\n");
 400                         goto cont;
 401                 }
 402                 skb_put(skb, sizeof *h + sizeof *ch);
 403                 skb->dev = ifp;
 404                 __skb_queue_tail(queue, skb);
 405                 h = (struct aoe_hdr *) skb_mac_header(skb);
 406                 memset(h, 0, sizeof *h + sizeof *ch);
 407
 408                 memset(h->dst, 0xff, sizeof h->dst);
 409                 memcpy(h->src, ifp->dev_addr, sizeof h->src);
 410                 h->type = __constant_cpu_to_be16(ETH_P_AOE);
 411                 h->verfl = AOE_HVER;
 412                 h->major = cpu_to_be16(aoemajor);
 413                 h->minor = aoeminor;
 414                 h->cmd = AOECMD_CFG;
 415
 416 cont:
 417                 dev_put(ifp);
 418         }
 419         rcu_read_unlock();
 420 }
 421
 422 static void
 423 resend(struct aoedev *d, struct frame *f)
 424 {
 425         struct sk_buff *skb;
 426         struct sk_buff_head queue;
 427         struct aoe_hdr *h;
 428         struct aoe_atahdr *ah;
 429         struct aoetgt *t;
 430         char buf[128];
 431         u32 n;
 432
 433         t = f->t;
 434         n = newtag(d);
 435         skb = f->skb;
 436         if (ifrotate(t) == NULL) {
 437                 /* probably can't happen, but set it up to fail anyway */
 438                 pr_info("aoe: resend: no interfaces to rotate to.\n");
 439                 ktcomplete(f, NULL);
 440                 return;
 441         }
 442         h = (struct aoe_hdr *) skb_mac_header(skb);
 443         ah = (struct aoe_atahdr *) (h+1);
 444
 445         snprintf(buf, sizeof buf,
 446                 "%15s e%ld.%d oldtag=%08x@%08lx newtag=%08x s=%pm d=%pm nout=%d\n",
 447                 "retransmit", d->aoemajor, d->aoeminor, f->tag, jiffies, n,
 448                 h->src, h->dst, t->nout);
 449         aoechr_error(buf);
 450
 451         f->tag = n;
 452         fhash(f);
 453         h->tag = cpu_to_be32(n);
 454         memcpy(h->dst, t->addr, sizeof h->dst);
 455         memcpy(h->src, t->ifp->nd->dev_addr, sizeof h->src);
 456
 457         skb->dev = t->ifp->nd;
 458         skb = skb_clone(skb, GFP_ATOMIC);
 459         if (skb == NULL)
 460                 return;
 461         __skb_queue_head_init(&queue);
 462         __skb_queue_tail(&queue, skb);
 463         aoenet_xmit(&queue);
 464 }
 465
 466 static int
 467 tsince(u32 tag)
 468 {
 469         int n;
 470
 471         n = jiffies & 0xffff;
 472         n -= tag & 0xffff;
 473         if (n < 0)
 474                 n += 1<<16;
 475         return n;
 476 }
 477
 478 static struct aoeif *
 479 getif(struct aoetgt *t, struct net_device *nd)
 480 {
 481         struct aoeif *p, *e;
 482
 483         p = t->ifs;
 484         e = p + NAOEIFS;
 485         for (; p < e; p++)
 486                 if (p->nd == nd)
 487                         return p;
 488         return NULL;
 489 }
 490
 491 static void
 492 ejectif(struct aoetgt *t, struct aoeif *ifp)
 493 {
 494         struct aoeif *e;
 495         struct net_device *nd;
 496         ulong n;
 497
 498         nd = ifp->nd;
 499         e = t->ifs + NAOEIFS - 1;
 500         n = (e - ifp) * sizeof *ifp;
 501         memmove(ifp, ifp+1, n);
 502         e->nd = NULL;
 503         dev_put(nd);
 504 }
 505
 506 static int
 507 sthtith(struct aoedev *d)
 508 {
 509         struct frame *f, *nf;
 510         struct list_head *nx, *pos, *head;
 511         struct sk_buff *skb;
 512         struct aoetgt *ht = d->htgt;
 513         int i;
 514
 515         for (i = 0; i < NFACTIVE; i++) {
 516                 head = &d->factive[i];
 517                 list_for_each_safe(pos, nx, head) {
 518                         f = list_entry(pos, struct frame, head);
 519                         if (f->t != ht)
 520                                 continue;
 521
 522                         nf = newframe(d);
 523                         if (!nf)
 524                                 return 0;
 525
 526                         /* remove frame from active list */
 527                         list_del(pos);
 528
 529                         /* reassign all pertinent bits to new outbound frame */
 530                         skb = nf->skb;
 531                         nf->skb = f->skb;
 532                         nf->buf = f->buf;
 533                         nf->bcnt = f->bcnt;
 534                         nf->lba = f->lba;
 535                         nf->bv = f->bv;
 536                         nf->bv_off = f->bv_off;
 537                         nf->waited = 0;
 538                         f->skb = skb;
 539                         aoe_freetframe(f);
 540                         ht->nout--;
 541                         nf->t->nout++;
 542                         resend(d, nf);
 543                 }
 544         }
 545         /* We've cleaned up the outstanding so take away his
 546          * interfaces so he won't be used.  We should remove him from
 547          * the target array here, but cleaning up a target is
 548          * involved.  PUNT!
 549          */
 550         memset(ht->ifs, 0, sizeof ht->ifs);
 551         d->htgt = NULL;
 552         return 1;
 553 }
 554
 555 static void
 556 rexmit_timer(ulong vp)
 557 {
 558         struct aoedev *d;
 559         struct aoetgt *t, **tt, **te;
 560         struct aoeif *ifp;
 561         struct frame *f;
 562         struct list_head *head, *pos, *nx;
 563         LIST_HEAD(flist);
 564         register long timeout;
 565         ulong flags, n;
 566         int i;
 567
 568         d = (struct aoedev *) vp;
 569
 570         /* timeout is always ~150% of the moving average */
 571         timeout = d->rttavg;
 572         timeout += timeout >> 1;
 573
 574         spin_lock_irqsave(&d->lock, flags);
 575
 576         if (d->flags & DEVFL_TKILL) {
 577                 spin_unlock_irqrestore(&d->lock, flags);
 578                 return;
 579         }
 580
 581         /* collect all frames to rexmit into flist */
 582         for (i = 0; i < NFACTIVE; i++) {
 583                 head = &d->factive[i];
 584                 list_for_each_safe(pos, nx, head) {
 585                         f = list_entry(pos, struct frame, head);
 586                         if (tsince(f->tag) < timeout)
 587                                 break;  /* end of expired frames */
 588                         /* move to flist for later processing */
 589                         list_move_tail(pos, &flist);
 590                 }
 591         }
 592         /* window check */
 593         tt = d->targets;
 594         te = tt + d->ntargets;
 595         for (; tt < te && (t = *tt); tt++) {
 596                 if (t->nout == t->maxout
 597                 && t->maxout < t->nframes
 598                 && (jiffies - t->lastwadj)/HZ > 10) {
 599                         t->maxout++;
 600                         t->lastwadj = jiffies;
 601                 }
 602         }
 603
 604         if (!list_empty(&flist)) {      /* retransmissions necessary */
 605                 n = d->rttavg <<= 1;
 606                 if (n > MAXTIMER)
 607                         d->rttavg = MAXTIMER;
 608         }
 609
 610         /* process expired frames */
 611         while (!list_empty(&flist)) {
 612                 pos = flist.next;
 613                 f = list_entry(pos, struct frame, head);
 614                 n = f->waited += timeout;
 615                 n /= HZ;
 616                 if (n > aoe_deadsecs) {
 617                         /* Waited too long.  Device failure.
 618                          * Hang all frames on first hash bucket for downdev
 619                          * to clean up.
 620                          */
 621                         list_splice(&flist, &d->factive[0]);
 622                         aoedev_downdev(d);
 623                         break;
 624                 }
 625                 list_del(pos);
 626
 627                 t = f->t;
 628                 if (n > aoe_deadsecs/2)
 629                         d->htgt = t; /* see if another target can help */
 630
 631                 if (t->nout == t->maxout) {
 632                         if (t->maxout > 1)
 633                                 t->maxout--;
 634                         t->lastwadj = jiffies;
 635                 }
 636
 637                 ifp = getif(t, f->skb->dev);
 638                 if (ifp && ++ifp->lost > (t->nframes << 1)
 639                 && (ifp != t->ifs || t->ifs[1].nd)) {
 640                         ejectif(t, ifp);
 641                         ifp = NULL;
 642                 }
 643                 resend(d, f);
 644         }
 645
 646         if ((d->flags & DEVFL_KICKME || d->htgt) && d->blkq) {
 647                 d->flags &= ~DEVFL_KICKME;
 648                 d->blkq->request_fn(d->blkq);
 649         }
 650
 651         d->timer.expires = jiffies + TIMERTICK;
 652         add_timer(&d->timer);
 653
 654         spin_unlock_irqrestore(&d->lock, flags);
 655 }
 656
 657 static unsigned long
 658 rqbiocnt(struct request *r)
 659 {
 660         struct bio *bio;
 661         unsigned long n = 0;
 662
 663         __rq_for_each_bio(bio, r)
 664                 n++;
 665         return n;
 666 }
 667
 668 /* This can be removed if we are certain that no users of the block
 669  * layer will ever use zero-count pages in bios.  Otherwise we have to
 670  * protect against the put_page sometimes done by the network layer.
 671  *
 672  * See http://oss.sgi.com/archives/xfs/2007-01/msg00594.html for
 673  * discussion.
 674  *
 675  * We cannot use get_page in the workaround, because it insists on a
 676  * positive page count as a precondition.  So we use _count directly.
 677  */
 678 static void
 679 bio_pageinc(struct bio *bio)
 680 {
 681         struct bio_vec *bv;
 682         struct page *page;
 683         int i;
 684
 685         bio_for_each_segment(bv, bio, i) {
 686                 page = bv->bv_page;
 687                 /* Non-zero page count for non-head members of
 688                  * compound pages is no longer allowed by the kernel,
 689                  * but this has never been seen here.
 690                  */
 691                 if (unlikely(PageCompound(page)))
 692                         if (compound_trans_head(page) != page) {
 693                                 pr_crit("page tail used for block I/O\n");
 694                                 BUG();
 695                         }
 696                 atomic_inc(&page->_count);
 697         }
 698 }
 699
 700 static void
 701 bio_pagedec(struct bio *bio)
 702 {
 703         struct bio_vec *bv;
 704         int i;
 705
 706         bio_for_each_segment(bv, bio, i)
 707                 atomic_dec(&bv->bv_page->_count);
 708 }
 709
 710 static void
 711 bufinit(struct buf *buf, struct request *rq, struct bio *bio)
 712 {
 713         struct bio_vec *bv;
 714
 715         memset(buf, 0, sizeof(*buf));
 716         buf->rq = rq;
 717         buf->bio = bio;
 718         buf->resid = bio->bi_size;
 719         buf->sector = bio->bi_sector;
 720         bio_pageinc(bio);
 721         buf->bv = bv = &bio->bi_io_vec[bio->bi_idx];
 722         buf->bv_resid = bv->bv_len;
 723         WARN_ON(buf->bv_resid == 0);
 724 }
 725
 726 static struct buf *
 727 nextbuf(struct aoedev *d)
 728 {
 729         struct request *rq;
 730         struct request_queue *q;
 731         struct buf *buf;
 732         struct bio *bio;
 733
 734         q = d->blkq;
 735         if (q == NULL)
 736                 return NULL;    /* initializing */
 737         if (d->ip.buf)
 738                 return d->ip.buf;
 739         rq = d->ip.rq;
 740         if (rq == NULL) {
 741                 rq = blk_peek_request(q);
 742                 if (rq == NULL)
 743                         return NULL;
 744                 blk_start_request(rq);
 745                 d->ip.rq = rq;
 746                 d->ip.nxbio = rq->bio;
 747                 rq->special = (void *) rqbiocnt(rq);
 748         }
 749         buf = mempool_alloc(d->bufpool, GFP_ATOMIC);
 750         if (buf == NULL) {
 751                 pr_err("aoe: nextbuf: unable to mempool_alloc!\n");
 752                 return NULL;
 753         }
 754         bio = d->ip.nxbio;
 755         bufinit(buf, rq, bio);
 756         bio = bio->bi_next;
 757         d->ip.nxbio = bio;
 758         if (bio == NULL)
 759                 d->ip.rq = NULL;
 760         return d->ip.buf = buf;
 761 }
 762
 763 /* enters with d->lock held */
 764 void
 765 aoecmd_work(struct aoedev *d)
 766 {
 767         if (d->htgt && !sthtith(d))
 768                 return;
 769         while (aoecmd_ata_rw(d))
 770                 ;
 771 }
 772
 773 /* this function performs work that has been deferred until sleeping is OK
 774  */
 775 void
 776 aoecmd_sleepwork(struct work_struct *work)
 777 {
 778         struct aoedev *d = container_of(work, struct aoedev, work);
 779         struct block_device *bd;
 780         u64 ssize;
 781
 782         if (d->flags & DEVFL_GDALLOC)
 783                 aoeblk_gdalloc(d);
 784
 785         if (d->flags & DEVFL_NEWSIZE) {
 786                 ssize = get_capacity(d->gd);
 787                 bd = bdget_disk(d->gd, 0);
 788                 if (bd) {
 789                         mutex_lock(&bd->bd_inode->i_mutex);
 790                         i_size_write(bd->bd_inode, (loff_t)ssize<<9);
 791                         mutex_unlock(&bd->bd_inode->i_mutex);
 792                         bdput(bd);
 793                 }
 794                 spin_lock_irq(&d->lock);
 795                 d->flags |= DEVFL_UP;
 796                 d->flags &= ~DEVFL_NEWSIZE;
 797                 spin_unlock_irq(&d->lock);
 798         }
 799 }
 800
 801 static void
 802 ataid_complete(struct aoedev *d, struct aoetgt *t, unsigned char *id)
 803 {
 804         u64 ssize;
 805         u16 n;
 806
 807         /* word 83: command set supported */
 808         n = get_unaligned_le16(&id[83 << 1]);
 809
 810         /* word 86: command set/feature enabled */
 811         n |= get_unaligned_le16(&id[86 << 1]);
 812
 813         if (n & (1<<10)) {      /* bit 10: LBA 48 */
 814                 d->flags |= DEVFL_EXT;
 815
 816                 /* word 100: number lba48 sectors */
 817                 ssize = get_unaligned_le64(&id[100 << 1]);
 818
 819                 /* set as in ide-disk.c:init_idedisk_capacity */
 820                 d->geo.cylinders = ssize;
 821                 d->geo.cylinders /= (255 * 63);
 822                 d->geo.heads = 255;
 823                 d->geo.sectors = 63;
 824         } else {
 825                 d->flags &= ~DEVFL_EXT;
 826
 827                 /* number lba28 sectors */
 828                 ssize = get_unaligned_le32(&id[60 << 1]);
 829
 830                 /* NOTE: obsolete in ATA 6 */
 831                 d->geo.cylinders = get_unaligned_le16(&id[54 << 1]);
 832                 d->geo.heads = get_unaligned_le16(&id[55 << 1]);
 833                 d->geo.sectors = get_unaligned_le16(&id[56 << 1]);
 834         }
 835
 836         if (d->ssize != ssize)
 837                 printk(KERN_INFO
 838                         "aoe: %pm e%ld.%d v%04x has %llu sectors\n",
 839                         t->addr,
 840                         d->aoemajor, d->aoeminor,
 841                         d->fw_ver, (long long)ssize);
 842         d->ssize = ssize;
 843         d->geo.start = 0;
 844         if (d->flags & (DEVFL_GDALLOC|DEVFL_NEWSIZE))
 845                 return;
 846         if (d->gd != NULL) {
 847                 set_capacity(d->gd, ssize);
 848                 d->flags |= DEVFL_NEWSIZE;
 849         } else
 850                 d->flags |= DEVFL_GDALLOC;
 851         schedule_work(&d->work);
 852 }
 853
 854 static void
 855 calc_rttavg(struct aoedev *d, int rtt)
 856 {
 857         register long n;
 858
 859         n = rtt;
 860         if (n < 0) {
 861                 n = -rtt;
 862                 if (n < MINTIMER)
 863                         n = MINTIMER;
 864                 else if (n > MAXTIMER)
 865                         n = MAXTIMER;
 866                 d->mintimer += (n - d->mintimer) >> 1;
 867         } else if (n < d->mintimer)
 868                 n = d->mintimer;
 869         else if (n > MAXTIMER)
 870                 n = MAXTIMER;
 871
 872         /* g == .25; cf. Congestion Avoidance and Control, Jacobson & Karels; 1988 */
 873         n -= d->rttavg;
 874         d->rttavg += n >> 2;
 875 }
 876
 877 static struct aoetgt *
 878 gettgt(struct aoedev *d, char *addr)
 879 {
 880         struct aoetgt **t, **e;
 881
 882         t = d->targets;
 883         e = t + NTARGETS;
 884         for (; t < e && *t; t++)
 885                 if (memcmp((*t)->addr, addr, sizeof((*t)->addr)) == 0)
 886                         return *t;
 887         return NULL;
 888 }
 889
 890 static void
 891 bvcpy(struct bio_vec *bv, ulong off, struct sk_buff *skb, long cnt)
 892 {
 893         ulong fcnt;
 894         char *p;
 895         int soff = 0;
 896 loop:
 897         fcnt = bv->bv_len - (off - bv->bv_offset);
 898         if (fcnt > cnt)
 899                 fcnt = cnt;
 900         p = page_address(bv->bv_page) + off;
 901         skb_copy_bits(skb, soff, p, fcnt);
 902         soff += fcnt;
 903         cnt -= fcnt;
 904         if (cnt <= 0)
 905                 return;
 906         bv++;
 907         off = bv->bv_offset;
 908         goto loop;
 909 }
 910
 911 void
 912 aoe_end_request(struct aoedev *d, struct request *rq, int fastfail)
 913 {
 914         struct bio *bio;
 915         int bok;
 916         struct request_queue *q;
 917
 918         q = d->blkq;
 919         if (rq == d->ip.rq)
 920                 d->ip.rq = NULL;
 921         do {
 922                 bio = rq->bio;
 923                 bok = !fastfail && test_bit(BIO_UPTODATE, &bio->bi_flags);
 924         } while (__blk_end_request(rq, bok ? 0 : -EIO, bio->bi_size));
 925
 926         /* cf. http://lkml.org/lkml/2006/10/31/28 */
 927         if (!fastfail)
 928                 __blk_run_queue(q);
 929 }
 930
 931 static void
 932 aoe_end_buf(struct aoedev *d, struct buf *buf)
 933 {
 934         struct request *rq;
 935         unsigned long n;
 936
 937         if (buf == d->ip.buf)
 938                 d->ip.buf = NULL;
 939         rq = buf->rq;
 940         bio_pagedec(buf->bio);
 941         mempool_free(buf, d->bufpool);
 942         n = (unsigned long) rq->special;
 943         rq->special = (void *) --n;
 944         if (n == 0)
 945                 aoe_end_request(d, rq, 0);
 946 }
 947
 948 static void
 949 ktiocomplete(struct frame *f)
 950 {
 951         struct aoe_hdr *hin, *hout;
 952         struct aoe_atahdr *ahin, *ahout;
 953         struct buf *buf;
 954         struct sk_buff *skb;
 955         struct aoetgt *t;
 956         struct aoeif *ifp;
 957         struct aoedev *d;
 958         long n;
 959
 960         if (f == NULL)
 961                 return;
 962
 963         t = f->t;
 964         d = t->d;
 965
 966         hout = (struct aoe_hdr *) skb_mac_header(f->skb);
 967         ahout = (struct aoe_atahdr *) (hout+1);
 968         buf = f->buf;
 969         skb = f->r_skb;
 970         if (skb == NULL)
 971                 goto noskb;     /* just fail the buf. */
 972
 973         hin = (struct aoe_hdr *) skb->data;
 974         skb_pull(skb, sizeof(*hin));
 975         ahin = (struct aoe_atahdr *) skb->data;
 976         skb_pull(skb, sizeof(*ahin));
 977         if (ahin->cmdstat & 0xa9) {     /* these bits cleared on success */
 978                 pr_err("aoe: ata error cmd=%2.2Xh stat=%2.2Xh from e%ld.%d\n",
 979                         ahout->cmdstat, ahin->cmdstat,
 980                         d->aoemajor, d->aoeminor);
 981 noskb:          if (buf)
 982                         clear_bit(BIO_UPTODATE, &buf->bio->bi_flags);
 983                 goto badrsp;
 984         }
 985
 986         n = ahout->scnt << 9;
 987         switch (ahout->cmdstat) {
 988         case ATA_CMD_PIO_READ:
 989         case ATA_CMD_PIO_READ_EXT:
 990                 if (skb->len < n) {
 991                         pr_err("aoe: runt data size in read.  skb->len=%d need=%ld\n",
 992                                 skb->len, n);
 993                         clear_bit(BIO_UPTODATE, &buf->bio->bi_flags);
 994                         break;
 995                 }
 996                 bvcpy(f->bv, f->bv_off, skb, n);
 997         case ATA_CMD_PIO_WRITE:
 998         case ATA_CMD_PIO_WRITE_EXT:
 999                 spin_lock_irq(&d->lock);
1000                 ifp = getif(t, skb->dev);
1001                 if (ifp)
1002                         ifp->lost = 0;
1003                 if (d->htgt == t) /* I'll help myself, thank you. */
1004                         d->htgt = NULL;
1005                 spin_unlock_irq(&d->lock);
1006                 break;
1007         case ATA_CMD_ID_ATA:
1008                 if (skb->len < 512) {
1009                         pr_info("aoe: runt data size in ataid.  skb->len=%d\n",
1010                                 skb->len);
1011                         break;
1012                 }
1013                 if (skb_linearize(skb))
1014                         break;
1015                 spin_lock_irq(&d->lock);
1016                 ataid_complete(d, t, skb->data);
1017                 spin_unlock_irq(&d->lock);
1018                 break;
1019         default:
1020                 pr_info("aoe: unrecognized ata command %2.2Xh for %d.%d\n",
1021                         ahout->cmdstat,
1022                         be16_to_cpu(get_unaligned(&hin->major)),
1023                         hin->minor);
1024         }
1025 badrsp:
1026         spin_lock_irq(&d->lock);
1027
1028         aoe_freetframe(f);
1029
1030         if (buf && --buf->nframesout == 0 && buf->resid == 0)
1031                 aoe_end_buf(d, buf);
1032
1033         aoecmd_work(d);
1034
1035         spin_unlock_irq(&d->lock);
1036         aoedev_put(d);
1037         dev_kfree_skb(skb);
1038 }
1039
1040 /* Enters with iocq.lock held.
1041  * Returns true iff responses needing processing remain.
1042  */
1043 static int
1044 ktio(void)
1045 {
1046         struct frame *f;
1047         struct list_head *pos;
1048         int i;
1049
1050         for (i = 0; ; ++i) {
1051                 if (i == MAXIOC)
1052                         return 1;
1053                 if (list_empty(&iocq.head))
1054                         return 0;
1055                 pos = iocq.head.next;
1056                 list_del(pos);
1057                 spin_unlock_irq(&iocq.lock);
1058                 f = list_entry(pos, struct frame, head);
1059                 ktiocomplete(f);
1060                 spin_lock_irq(&iocq.lock);
1061         }
1062 }
1063
1064 static int
1065 kthread(void *vp)
1066 {
1067         struct ktstate *k;
1068         DECLARE_WAITQUEUE(wait, current);
1069         int more;
1070
1071         k = vp;
1072         current->flags |= PF_NOFREEZE;
1073         set_user_nice(current, -10);
1074         complete(&k->rendez);   /* tell spawner we're running */
1075         do {
1076                 spin_lock_irq(k->lock);
1077                 more = k->fn();
1078                 if (!more) {
1079                         add_wait_queue(k->waitq, &wait);
1080                         __set_current_state(TASK_INTERRUPTIBLE);
1081                 }
1082                 spin_unlock_irq(k->lock);
1083                 if (!more) {
1084                         schedule();
1085                         remove_wait_queue(k->waitq, &wait);
1086                 } else
1087                         cond_resched();
1088         } while (!kthread_should_stop());
1089         complete(&k->rendez);   /* tell spawner we're stopping */
1090         return 0;
1091 }
1092
1093 void
1094 aoe_ktstop(struct ktstate *k)
1095 {
1096         kthread_stop(k->task);
1097         wait_for_completion(&k->rendez);
1098 }
1099
1100 int
1101 aoe_ktstart(struct ktstate *k)
1102 {
1103         struct task_struct *task;
1104
1105         init_completion(&k->rendez);
1106         task = kthread_run(kthread, k, k->name);
1107         if (task == NULL || IS_ERR(task))
1108                 return -ENOMEM;
1109         k->task = task;
1110         wait_for_completion(&k->rendez); /* allow kthread to start */
1111         init_completion(&k->rendez);    /* for waiting for exit later */
1112         return 0;
1113 }
1114
1115 /* pass it off to kthreads for processing */
1116 static void
1117 ktcomplete(struct frame *f, struct sk_buff *skb)
1118 {
1119         ulong flags;
1120
1121         f->r_skb = skb;
1122         spin_lock_irqsave(&iocq.lock, flags);
1123         list_add_tail(&f->head, &iocq.head);
1124         spin_unlock_irqrestore(&iocq.lock, flags);
1125         wake_up(&ktiowq);
1126 }
1127
1128 struct sk_buff *
1129 aoecmd_ata_rsp(struct sk_buff *skb)
1130 {
1131         struct aoedev *d;
1132         struct aoe_hdr *h;
1133         struct frame *f;
1134         struct aoetgt *t;
1135         u32 n;
1136         ulong flags;
1137         char ebuf[128];
1138         u16 aoemajor;
1139
1140         h = (struct aoe_hdr *) skb->data;
1141         aoemajor = be16_to_cpu(get_unaligned(&h->major));
1142         d = aoedev_by_aoeaddr(aoemajor, h->minor, 0);
1143         if (d == NULL) {
1144                 snprintf(ebuf, sizeof ebuf, "aoecmd_ata_rsp: ata response "
1145                         "for unknown device %d.%d\n",
1146                         aoemajor, h->minor);
1147                 aoechr_error(ebuf);
1148                 return skb;
1149         }
1150
1151         spin_lock_irqsave(&d->lock, flags);
1152
1153         n = be32_to_cpu(get_unaligned(&h->tag));
1154         f = getframe(d, n);
1155         if (f == NULL) {
1156                 calc_rttavg(d, -tsince(n));
1157                 spin_unlock_irqrestore(&d->lock, flags);
1158                 aoedev_put(d);
1159                 snprintf(ebuf, sizeof ebuf,
1160                         "%15s e%d.%d    tag=%08x@%08lx\n",
1161                         "unexpected rsp",
1162                         get_unaligned_be16(&h->major),
1163                         h->minor,
1164                         get_unaligned_be32(&h->tag),
1165                         jiffies);
1166                 aoechr_error(ebuf);
1167                 return skb;
1168         }
1169         t = f->t;
1170         calc_rttavg(d, tsince(f->tag));
1171         t->nout--;
1172         aoecmd_work(d);
1173
1174         spin_unlock_irqrestore(&d->lock, flags);
1175
1176         ktcomplete(f, skb);
1177
1178         /*
1179          * Note here that we do not perform an aoedev_put, as we are
1180          * leaving this reference for the ktio to release.
1181          */
1182         return NULL;
1183 }
1184
1185 void
1186 aoecmd_cfg(ushort aoemajor, unsigned char aoeminor)
1187 {
1188         struct sk_buff_head queue;
1189
1190         __skb_queue_head_init(&queue);
1191         aoecmd_cfg_pkts(aoemajor, aoeminor, &queue);
1192         aoenet_xmit(&queue);
1193 }
1194
1195 struct sk_buff *
1196 aoecmd_ata_id(struct aoedev *d)
1197 {
1198         struct aoe_hdr *h;
1199         struct aoe_atahdr *ah;
1200         struct frame *f;
1201         struct sk_buff *skb;
1202         struct aoetgt *t;
1203
1204         f = newframe(d);
1205         if (f == NULL)
1206                 return NULL;
1207
1208         t = *d->tgt;
1209
1210         /* initialize the headers & frame */
1211         skb = f->skb;
1212         h = (struct aoe_hdr *) skb_mac_header(skb);
1213         ah = (struct aoe_atahdr *) (h+1);
1214         skb_put(skb, sizeof *h + sizeof *ah);
1215         memset(h, 0, skb->len);
1216         f->tag = aoehdr_atainit(d, t, h);
1217         fhash(f);
1218         t->nout++;
1219         f->waited = 0;
1220
1221         /* set up ata header */
1222         ah->scnt = 1;
1223         ah->cmdstat = ATA_CMD_ID_ATA;
1224         ah->lba3 = 0xa0;
1225
1226         skb->dev = t->ifp->nd;
1227
1228         d->rttavg = MAXTIMER;
1229         d->timer.function = rexmit_timer;
1230
1231         return skb_clone(skb, GFP_ATOMIC);
1232 }
1233
1234 static struct aoetgt *
1235 addtgt(struct aoedev *d, char *addr, ulong nframes)
1236 {
1237         struct aoetgt *t, **tt, **te;
1238
1239         tt = d->targets;
1240         te = tt + NTARGETS;
1241         for (; tt < te && *tt; tt++)
1242                 ;
1243
1244         if (tt == te) {
1245                 printk(KERN_INFO
1246                         "aoe: device addtgt failure; too many targets\n");
1247                 return NULL;
1248         }
1249         t = kzalloc(sizeof(*t), GFP_ATOMIC);
1250         if (!t) {
1251                 printk(KERN_INFO "aoe: cannot allocate memory to add target\n");
1252                 return NULL;
1253         }
1254
1255         d->ntargets++;
1256         t->nframes = nframes;
1257         t->d = d;
1258         memcpy(t->addr, addr, sizeof t->addr);
1259         t->ifp = t->ifs;
1260         t->maxout = t->nframes;
1261         INIT_LIST_HEAD(&t->ffree);
1262         return *tt = t;
1263 }
1264
1265 static void
1266 setdbcnt(struct aoedev *d)
1267 {
1268         struct aoetgt **t, **e;
1269         int bcnt = 0;
1270
1271         t = d->targets;
1272         e = t + NTARGETS;
1273         for (; t < e && *t; t++)
1274                 if (bcnt == 0 || bcnt > (*t)->minbcnt)
1275                         bcnt = (*t)->minbcnt;
1276         if (bcnt != d->maxbcnt) {
1277                 d->maxbcnt = bcnt;
1278                 pr_info("aoe: e%ld.%d: setting %d byte data frames\n",
1279                         d->aoemajor, d->aoeminor, bcnt);
1280         }
1281 }
1282
1283 static void
1284 setifbcnt(struct aoetgt *t, struct net_device *nd, int bcnt)
1285 {
1286         struct aoedev *d;
1287         struct aoeif *p, *e;
1288         int minbcnt;
1289
1290         d = t->d;
1291         minbcnt = bcnt;
1292         p = t->ifs;
1293         e = p + NAOEIFS;
1294         for (; p < e; p++) {
1295                 if (p->nd == NULL)
1296                         break;          /* end of the valid interfaces */
1297                 if (p->nd == nd) {
1298                         p->bcnt = bcnt; /* we're updating */
1299                         nd = NULL;
1300                 } else if (minbcnt > p->bcnt)
1301                         minbcnt = p->bcnt; /* find the min interface */
1302         }
1303         if (nd) {
1304                 if (p == e) {
1305                         pr_err("aoe: device setifbcnt failure; too many interfaces.\n");
1306                         return;
1307                 }
1308                 dev_hold(nd);
1309                 p->nd = nd;
1310                 p->bcnt = bcnt;
1311         }
1312         t->minbcnt = minbcnt;
1313         setdbcnt(d);
1314 }
1315
1316 void
1317 aoecmd_cfg_rsp(struct sk_buff *skb)
1318 {
1319         struct aoedev *d;
1320         struct aoe_hdr *h;
1321         struct aoe_cfghdr *ch;
1322         struct aoetgt *t;
1323         ulong flags, aoemajor;
1324         struct sk_buff *sl;
1325         struct sk_buff_head queue;
1326         u16 n;
1327
1328         sl = NULL;
1329         h = (struct aoe_hdr *) skb_mac_header(skb);
1330         ch = (struct aoe_cfghdr *) (h+1);
1331
1332         /*
1333          * Enough people have their dip switches set backwards to
1334          * warrant a loud message for this special case.
1335          */
1336         aoemajor = get_unaligned_be16(&h->major);
1337         if (aoemajor == 0xfff) {
1338                 printk(KERN_ERR "aoe: Warning: shelf address is all ones.  "
1339                         "Check shelf dip switches.\n");
1340                 return;
1341         }
1342         if (aoemajor == 0xffff) {
1343                 pr_info("aoe: e%ld.%d: broadcast shelf number invalid\n",
1344                         aoemajor, (int) h->minor);
1345                 return;
1346         }
1347         if (h->minor == 0xff) {
1348                 pr_info("aoe: e%ld.%d: broadcast slot number invalid\n",
1349                         aoemajor, (int) h->minor);
1350                 return;
1351         }
1352
1353         n = be16_to_cpu(ch->bufcnt);
1354         if (n > aoe_maxout)     /* keep it reasonable */
1355                 n = aoe_maxout;
1356
1357         d = aoedev_by_aoeaddr(aoemajor, h->minor, 1);
1358         if (d == NULL) {
1359                 pr_info("aoe: device allocation failure\n");
1360                 return;
1361         }
1362
1363         spin_lock_irqsave(&d->lock, flags);
1364
1365         t = gettgt(d, h->src);
1366         if (t) {
1367                 t->nframes = n;
1368                 if (n < t->maxout)
1369                         t->maxout = n;
1370         } else {
1371                 t = addtgt(d, h->src, n);
1372                 if (!t)
1373                         goto bail;
1374         }
1375         n = skb->dev->mtu;
1376         n -= sizeof(struct aoe_hdr) + sizeof(struct aoe_atahdr);
1377         n /= 512;
1378         if (n > ch->scnt)
1379                 n = ch->scnt;
1380         n = n ? n * 512 : DEFAULTBCNT;
1381         setifbcnt(t, skb->dev, n);
1382
1383         /* don't change users' perspective */
1384         if (d->nopen == 0) {
1385                 d->fw_ver = be16_to_cpu(ch->fwver);
1386                 sl = aoecmd_ata_id(d);
1387         }
1388 bail:
1389         spin_unlock_irqrestore(&d->lock, flags);
1390         aoedev_put(d);
1391         if (sl) {
1392                 __skb_queue_head_init(&queue);
1393                 __skb_queue_tail(&queue, sl);
1394                 aoenet_xmit(&queue);
1395         }
1396 }
1397
1398 void
1399 aoecmd_cleanslate(struct aoedev *d)
1400 {
1401         struct aoetgt **t, **te;
1402
1403         d->mintimer = MINTIMER;
1404         d->maxbcnt = 0;
1405
1406         t = d->targets;
1407         te = t + NTARGETS;
1408         for (; t < te && *t; t++)
1409                 (*t)->maxout = (*t)->nframes;
1410 }
1411
1412 void
1413 aoe_failbuf(struct aoedev *d, struct buf *buf)
1414 {
1415         if (buf == NULL)
1416                 return;
1417         buf->resid = 0;
1418         clear_bit(BIO_UPTODATE, &buf->bio->bi_flags);
1419         if (buf->nframesout == 0)
1420                 aoe_end_buf(d, buf);
1421 }
1422
1423 void
1424 aoe_flush_iocq(void)
1425 {
1426         struct frame *f;
1427         struct aoedev *d;
1428         LIST_HEAD(flist);
1429         struct list_head *pos;
1430         struct sk_buff *skb;
1431         ulong flags;
1432
1433         spin_lock_irqsave(&iocq.lock, flags);
1434         list_splice_init(&iocq.head, &flist);
1435         spin_unlock_irqrestore(&iocq.lock, flags);
1436         while (!list_empty(&flist)) {
1437                 pos = flist.next;
1438                 list_del(pos);
1439                 f = list_entry(pos, struct frame, head);
1440                 d = f->t->d;
1441                 skb = f->r_skb;
1442                 spin_lock_irqsave(&d->lock, flags);
1443                 if (f->buf) {
1444                         f->buf->nframesout--;
1445                         aoe_failbuf(d, f->buf);
1446                 }
1447                 aoe_freetframe(f);
1448                 spin_unlock_irqrestore(&d->lock, flags);
1449                 dev_kfree_skb(skb);
1450                 aoedev_put(d);
1451         }
1452 }
1453
1454 int __init
1455 aoecmd_init(void)
1456 {
1457         INIT_LIST_HEAD(&iocq.head);
1458         spin_lock_init(&iocq.lock);
1459         init_waitqueue_head(&ktiowq);
1460         kts.name = "aoe_ktio";
1461         kts.fn = ktio;
1462         kts.waitq = &ktiowq;
1463         kts.lock = &iocq.lock;
1464         return aoe_ktstart(&kts);
1465 }
1466
1467 void
1468 aoecmd_exit(void)
1469 {
1470         aoe_ktstop(&kts);
1471         aoe_flush_iocq();
1472 }