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/hdreg.h>
   8 #include <linux/blkdev.h>
   9 #include <linux/skbuff.h>
  10 #include <linux/netdevice.h>
  11 #include <linux/genhd.h>
  12 #include <linux/moduleparam.h>
  13 #include <net/net_namespace.h>
  14 #include <asm/unaligned.h>
  15 #include "aoe.h"
  16
  17 static int aoe_deadsecs = 60 * 3;
  18 module_param(aoe_deadsecs, int, 0644);
  19 MODULE_PARM_DESC(aoe_deadsecs, "After aoe_deadsecs seconds, give up and fail dev.");
  20
  21 static int aoe_maxout = 16;
  22 module_param(aoe_maxout, int, 0644);
  23 MODULE_PARM_DESC(aoe_maxout,
  24         "Only aoe_maxout outstanding packets for every MAC on eX.Y.");
  25
  26 static struct sk_buff *
  27 new_skb(ulong len)
  28 {
  29         struct sk_buff *skb;
  30
  31         skb = alloc_skb(len, GFP_ATOMIC);
  32         if (skb) {
  33                 skb_reset_mac_header(skb);
  34                 skb_reset_network_header(skb);
  35                 skb->protocol = __constant_htons(ETH_P_AOE);
  36                 skb->priority = 0;
  37                 skb->next = skb->prev = NULL;
  38
  39                 /* tell the network layer not to perform IP checksums
  40                  * or to get the NIC to do it
  41                  */
  42                 skb->ip_summed = CHECKSUM_NONE;
  43         }
  44         return skb;
  45 }
  46
  47 static struct frame *
  48 getframe(struct aoetgt *t, int tag)
  49 {
  50         struct frame *f, *e;
  51
  52         f = t->frames;
  53         e = f + t->nframes;
  54         for (; f<e; f++)
  55                 if (f->tag == tag)
  56                         return f;
  57         return NULL;
  58 }
  59
  60 /*
  61  * Leave the top bit clear so we have tagspace for userland.
  62  * The bottom 16 bits are the xmit tick for rexmit/rttavg processing.
  63  * This driver reserves tag -1 to mean "unused frame."
  64  */
  65 static int
  66 newtag(struct aoetgt *t)
  67 {
  68         register ulong n;
  69
  70         n = jiffies & 0xffff;
  71         return n |= (++t->lasttag & 0x7fff) << 16;
  72 }
  73
  74 static int
  75 aoehdr_atainit(struct aoedev *d, struct aoetgt *t, struct aoe_hdr *h)
  76 {
  77         u32 host_tag = newtag(t);
  78
  79         memcpy(h->src, t->ifp->nd->dev_addr, sizeof h->src);
  80         memcpy(h->dst, t->addr, sizeof h->dst);
  81         h->type = __constant_cpu_to_be16(ETH_P_AOE);
  82         h->verfl = AOE_HVER;
  83         h->major = cpu_to_be16(d->aoemajor);
  84         h->minor = d->aoeminor;
  85         h->cmd = AOECMD_ATA;
  86         h->tag = cpu_to_be32(host_tag);
  87
  88         return host_tag;
  89 }
  90
  91 static inline void
  92 put_lba(struct aoe_atahdr *ah, sector_t lba)
  93 {
  94         ah->lba0 = lba;
  95         ah->lba1 = lba >>= 8;
  96         ah->lba2 = lba >>= 8;
  97         ah->lba3 = lba >>= 8;
  98         ah->lba4 = lba >>= 8;
  99         ah->lba5 = lba >>= 8;
 100 }
 101
 102 static void
 103 ifrotate(struct aoetgt *t)
 104 {
 105         t->ifp++;
 106         if (t->ifp >= &t->ifs[NAOEIFS] || t->ifp->nd == NULL)
 107                 t->ifp = t->ifs;
 108         if (t->ifp->nd == NULL) {
 109                 printk(KERN_INFO "aoe: no interface to rotate to\n");
 110                 BUG();
 111         }
 112 }
 113
 114 static void
 115 skb_pool_put(struct aoedev *d, struct sk_buff *skb)
 116 {
 117         if (!d->skbpool_hd)
 118                 d->skbpool_hd = skb;
 119         else
 120                 d->skbpool_tl->next = skb;
 121         d->skbpool_tl = skb;
 122 }
 123
 124 static struct sk_buff *
 125 skb_pool_get(struct aoedev *d)
 126 {
 127         struct sk_buff *skb;
 128
 129         skb = d->skbpool_hd;
 130         if (skb && atomic_read(&skb_shinfo(skb)->dataref) == 1) {
 131                 d->skbpool_hd = skb->next;
 132                 skb->next = NULL;
 133                 return skb;
 134         }
 135         if (d->nskbpool < NSKBPOOLMAX
 136         && (skb = new_skb(ETH_ZLEN))) {
 137                 d->nskbpool++;
 138                 return skb;
 139         }
 140         return NULL;
 141 }
 142
 143 /* freeframe is where we do our load balancing so it's a little hairy. */
 144 static struct frame *
 145 freeframe(struct aoedev *d)
 146 {
 147         struct frame *f, *e, *rf;
 148         struct aoetgt **t;
 149         struct sk_buff *skb;
 150
 151         if (d->targets[0] == NULL) {    /* shouldn't happen, but I'm paranoid */
 152                 printk(KERN_ERR "aoe: NULL TARGETS!\n");
 153                 return NULL;
 154         }
 155         t = d->tgt;
 156         t++;
 157         if (t >= &d->targets[NTARGETS] || !*t)
 158                 t = d->targets;
 159         for (;;) {
 160                 if ((*t)->nout < (*t)->maxout
 161                 && t != d->htgt
 162                 && (*t)->ifp->nd) {
 163                         rf = NULL;
 164                         f = (*t)->frames;
 165                         e = f + (*t)->nframes;
 166                         for (; f < e; f++) {
 167                                 if (f->tag != FREETAG)
 168                                         continue;
 169                                 skb = f->skb;
 170                                 if (!skb
 171                                 && !(f->skb = skb = new_skb(ETH_ZLEN)))
 172                                         continue;
 173                                 if (atomic_read(&skb_shinfo(skb)->dataref)
 174                                         != 1) {
 175                                         if (!rf)
 176                                                 rf = f;
 177                                         continue;
 178                                 }
 179 gotone:                         skb_shinfo(skb)->nr_frags = skb->data_len = 0;
 180                                 skb_trim(skb, 0);
 181                                 d->tgt = t;
 182                                 ifrotate(*t);
 183                                 return f;
 184                         }
 185                         /* Work can be done, but the network layer is
 186                            holding our precious packets.  Try to grab
 187                            one from the pool. */
 188                         f = rf;
 189                         if (f == NULL) {        /* more paranoia */
 190                                 printk(KERN_ERR
 191                                         "aoe: freeframe: %s.\n",
 192                                         "unexpected null rf");
 193                                 d->flags |= DEVFL_KICKME;
 194                                 return NULL;
 195                         }
 196                         skb = skb_pool_get(d);
 197                         if (skb) {
 198                                 skb_pool_put(d, f->skb);
 199                                 f->skb = skb;
 200                                 goto gotone;
 201                         }
 202                         (*t)->dataref++;
 203                         if ((*t)->nout == 0)
 204                                 d->flags |= DEVFL_KICKME;
 205                 }
 206                 if (t == d->tgt)        /* we've looped and found nada */
 207                         break;
 208                 t++;
 209                 if (t >= &d->targets[NTARGETS] || !*t)
 210                         t = d->targets;
 211         }
 212         return NULL;
 213 }
 214
 215 static int
 216 aoecmd_ata_rw(struct aoedev *d)
 217 {
 218         struct frame *f;
 219         struct aoe_hdr *h;
 220         struct aoe_atahdr *ah;
 221         struct buf *buf;
 222         struct bio_vec *bv;
 223         struct aoetgt *t;
 224         struct sk_buff *skb;
 225         ulong bcnt;
 226         char writebit, extbit;
 227
 228         writebit = 0x10;
 229         extbit = 0x4;
 230
 231         f = freeframe(d);
 232         if (f == NULL)
 233                 return 0;
 234         t = *d->tgt;
 235         buf = d->inprocess;
 236         bv = buf->bv;
 237         bcnt = t->ifp->maxbcnt;
 238         if (bcnt == 0)
 239                 bcnt = DEFAULTBCNT;
 240         if (bcnt > buf->bv_resid)
 241                 bcnt = buf->bv_resid;
 242         /* initialize the headers & frame */
 243         skb = f->skb;
 244         h = (struct aoe_hdr *) skb_mac_header(skb);
 245         ah = (struct aoe_atahdr *) (h+1);
 246         skb_put(skb, sizeof *h + sizeof *ah);
 247         memset(h, 0, skb->len);
 248         f->tag = aoehdr_atainit(d, t, h);
 249         t->nout++;
 250         f->waited = 0;
 251         f->buf = buf;
 252         f->bufaddr = page_address(bv->bv_page) + buf->bv_off;
 253         f->bcnt = bcnt;
 254         f->lba = buf->sector;
 255
 256         /* set up ata header */
 257         ah->scnt = bcnt >> 9;
 258         put_lba(ah, buf->sector);
 259         if (d->flags & DEVFL_EXT) {
 260                 ah->aflags |= AOEAFL_EXT;
 261         } else {
 262                 extbit = 0;
 263                 ah->lba3 &= 0x0f;
 264                 ah->lba3 |= 0xe0;       /* LBA bit + obsolete 0xa0 */
 265         }
 266         if (bio_data_dir(buf->bio) == WRITE) {
 267                 skb_fill_page_desc(skb, 0, bv->bv_page, buf->bv_off, bcnt);
 268                 ah->aflags |= AOEAFL_WRITE;
 269                 skb->len += bcnt;
 270                 skb->data_len = bcnt;
 271                 t->wpkts++;
 272         } else {
 273                 t->rpkts++;
 274                 writebit = 0;
 275         }
 276
 277         ah->cmdstat = WIN_READ | writebit | extbit;
 278
 279         /* mark all tracking fields and load out */
 280         buf->nframesout += 1;
 281         buf->bv_off += bcnt;
 282         buf->bv_resid -= bcnt;
 283         buf->resid -= bcnt;
 284         buf->sector += bcnt >> 9;
 285         if (buf->resid == 0) {
 286                 d->inprocess = NULL;
 287         } else if (buf->bv_resid == 0) {
 288                 buf->bv = ++bv;
 289                 buf->bv_resid = bv->bv_len;
 290                 WARN_ON(buf->bv_resid == 0);
 291                 buf->bv_off = bv->bv_offset;
 292         }
 293
 294         skb->dev = t->ifp->nd;
 295         skb = skb_clone(skb, GFP_ATOMIC);
 296         if (skb) {
 297                 if (d->sendq_hd)
 298                         d->sendq_tl->next = skb;
 299                 else
 300                         d->sendq_hd = skb;
 301                 d->sendq_tl = skb;
 302         }
 303         return 1;
 304 }
 305
 306 /* some callers cannot sleep, and they can call this function,
 307  * transmitting the packets later, when interrupts are on
 308  */
 309 static struct sk_buff *
 310 aoecmd_cfg_pkts(ushort aoemajor, unsigned char aoeminor, struct sk_buff **tail)
 311 {
 312         struct aoe_hdr *h;
 313         struct aoe_cfghdr *ch;
 314         struct sk_buff *skb, *sl, *sl_tail;
 315         struct net_device *ifp;
 316
 317         sl = sl_tail = NULL;
 318
 319         read_lock(&dev_base_lock);
 320         for_each_netdev(&init_net, ifp) {
 321                 dev_hold(ifp);
 322                 if (!is_aoe_netif(ifp))
 323                         goto cont;
 324
 325                 skb = new_skb(sizeof *h + sizeof *ch);
 326                 if (skb == NULL) {
 327                         printk(KERN_INFO "aoe: skb alloc failure\n");
 328                         goto cont;
 329                 }
 330                 skb_put(skb, sizeof *h + sizeof *ch);
 331                 skb->dev = ifp;
 332                 if (sl_tail == NULL)
 333                         sl_tail = skb;
 334                 h = (struct aoe_hdr *) skb_mac_header(skb);
 335                 memset(h, 0, sizeof *h + sizeof *ch);
 336
 337                 memset(h->dst, 0xff, sizeof h->dst);
 338                 memcpy(h->src, ifp->dev_addr, sizeof h->src);
 339                 h->type = __constant_cpu_to_be16(ETH_P_AOE);
 340                 h->verfl = AOE_HVER;
 341                 h->major = cpu_to_be16(aoemajor);
 342                 h->minor = aoeminor;
 343                 h->cmd = AOECMD_CFG;
 344
 345                 skb->next = sl;
 346                 sl = skb;
 347 cont:
 348                 dev_put(ifp);
 349         }
 350         read_unlock(&dev_base_lock);
 351
 352         if (tail != NULL)
 353                 *tail = sl_tail;
 354         return sl;
 355 }
 356
 357 static void
 358 resend(struct aoedev *d, struct aoetgt *t, struct frame *f)
 359 {
 360         struct sk_buff *skb;
 361         struct aoe_hdr *h;
 362         struct aoe_atahdr *ah;
 363         char buf[128];
 364         u32 n;
 365
 366         ifrotate(t);
 367         n = newtag(t);
 368         skb = f->skb;
 369         h = (struct aoe_hdr *) skb_mac_header(skb);
 370         ah = (struct aoe_atahdr *) (h+1);
 371
 372         snprintf(buf, sizeof buf,
 373                 "%15s e%ld.%d oldtag=%08x@%08lx newtag=%08x "
 374                 "s=%012llx d=%012llx nout=%d\n",
 375                 "retransmit", d->aoemajor, d->aoeminor, f->tag, jiffies, n,
 376                 mac_addr(h->src),
 377                 mac_addr(h->dst), t->nout);
 378         aoechr_error(buf);
 379
 380         f->tag = n;
 381         h->tag = cpu_to_be32(n);
 382         memcpy(h->dst, t->addr, sizeof h->dst);
 383         memcpy(h->src, t->ifp->nd->dev_addr, sizeof h->src);
 384
 385         switch (ah->cmdstat) {
 386         default:
 387                 break;
 388         case WIN_READ:
 389         case WIN_READ_EXT:
 390         case WIN_WRITE:
 391         case WIN_WRITE_EXT:
 392                 put_lba(ah, f->lba);
 393
 394                 n = f->bcnt;
 395                 if (n > DEFAULTBCNT)
 396                         n = DEFAULTBCNT;
 397                 ah->scnt = n >> 9;
 398                 if (ah->aflags & AOEAFL_WRITE) {
 399                         skb_fill_page_desc(skb, 0, virt_to_page(f->bufaddr),
 400                                 offset_in_page(f->bufaddr), n);
 401                         skb->len = sizeof *h + sizeof *ah + n;
 402                         skb->data_len = n;
 403                 }
 404         }
 405         skb->dev = t->ifp->nd;
 406         skb = skb_clone(skb, GFP_ATOMIC);
 407         if (skb == NULL)
 408                 return;
 409         if (d->sendq_hd)
 410                 d->sendq_tl->next = skb;
 411         else
 412                 d->sendq_hd = skb;
 413         d->sendq_tl = skb;
 414 }
 415
 416 static int
 417 tsince(int tag)
 418 {
 419         int n;
 420
 421         n = jiffies & 0xffff;
 422         n -= tag & 0xffff;
 423         if (n < 0)
 424                 n += 1<<16;
 425         return n;
 426 }
 427
 428 static struct aoeif *
 429 getif(struct aoetgt *t, struct net_device *nd)
 430 {
 431         struct aoeif *p, *e;
 432
 433         p = t->ifs;
 434         e = p + NAOEIFS;
 435         for (; p < e; p++)
 436                 if (p->nd == nd)
 437                         return p;
 438         return NULL;
 439 }
 440
 441 static struct aoeif *
 442 addif(struct aoetgt *t, struct net_device *nd)
 443 {
 444         struct aoeif *p;
 445
 446         p = getif(t, NULL);
 447         if (!p)
 448                 return NULL;
 449         p->nd = nd;
 450         p->maxbcnt = DEFAULTBCNT;
 451         p->lost = 0;
 452         p->lostjumbo = 0;
 453         return p;
 454 }
 455
 456 static void
 457 ejectif(struct aoetgt *t, struct aoeif *ifp)
 458 {
 459         struct aoeif *e;
 460         ulong n;
 461
 462         e = t->ifs + NAOEIFS - 1;
 463         n = (e - ifp) * sizeof *ifp;
 464         memmove(ifp, ifp+1, n);
 465         e->nd = NULL;
 466 }
 467
 468 static int
 469 sthtith(struct aoedev *d)
 470 {
 471         struct frame *f, *e, *nf;
 472         struct sk_buff *skb;
 473         struct aoetgt *ht = *d->htgt;
 474
 475         f = ht->frames;
 476         e = f + ht->nframes;
 477         for (; f < e; f++) {
 478                 if (f->tag == FREETAG)
 479                         continue;
 480                 nf = freeframe(d);
 481                 if (!nf)
 482                         return 0;
 483                 skb = nf->skb;
 484                 *nf = *f;
 485                 f->skb = skb;
 486                 f->tag = FREETAG;
 487                 nf->waited = 0;
 488                 ht->nout--;
 489                 (*d->tgt)->nout++;
 490                 resend(d, *d->tgt, nf);
 491         }
 492         /* he's clean, he's useless.  take away his interfaces */
 493         memset(ht->ifs, 0, sizeof ht->ifs);
 494         d->htgt = NULL;
 495         return 1;
 496 }
 497
 498 static inline unsigned char
 499 ata_scnt(unsigned char *packet) {
 500         struct aoe_hdr *h;
 501         struct aoe_atahdr *ah;
 502
 503         h = (struct aoe_hdr *) packet;
 504         ah = (struct aoe_atahdr *) (h+1);
 505         return ah->scnt;
 506 }
 507
 508 static void
 509 rexmit_timer(ulong vp)
 510 {
 511         struct aoedev *d;
 512         struct aoetgt *t, **tt, **te;
 513         struct aoeif *ifp;
 514         struct frame *f, *e;
 515         struct sk_buff *sl;
 516         register long timeout;
 517         ulong flags, n;
 518
 519         d = (struct aoedev *) vp;
 520         sl = NULL;
 521
 522         /* timeout is always ~150% of the moving average */
 523         timeout = d->rttavg;
 524         timeout += timeout >> 1;
 525
 526         spin_lock_irqsave(&d->lock, flags);
 527
 528         if (d->flags & DEVFL_TKILL) {
 529                 spin_unlock_irqrestore(&d->lock, flags);
 530                 return;
 531         }
 532         tt = d->targets;
 533         te = tt + NTARGETS;
 534         for (; tt < te && *tt; tt++) {
 535                 t = *tt;
 536                 f = t->frames;
 537                 e = f + t->nframes;
 538                 for (; f < e; f++) {
 539                         if (f->tag == FREETAG
 540                         || tsince(f->tag) < timeout)
 541                                 continue;
 542                         n = f->waited += timeout;
 543                         n /= HZ;
 544                         if (n > aoe_deadsecs) {
 545                                 /* waited too long.  device failure. */
 546                                 aoedev_downdev(d);
 547                                 break;
 548                         }
 549
 550                         if (n > HELPWAIT /* see if another target can help */
 551                         && (tt != d->targets || d->targets[1]))
 552                                 d->htgt = tt;
 553
 554                         if (t->nout == t->maxout) {
 555                                 if (t->maxout > 1)
 556                                         t->maxout--;
 557                                 t->lastwadj = jiffies;
 558                         }
 559
 560                         ifp = getif(t, f->skb->dev);
 561                         if (ifp && ++ifp->lost > (t->nframes << 1)
 562                         && (ifp != t->ifs || t->ifs[1].nd)) {
 563                                 ejectif(t, ifp);
 564                                 ifp = NULL;
 565                         }
 566
 567                         if (ata_scnt(skb_mac_header(f->skb)) > DEFAULTBCNT / 512
 568                         && ifp && ++ifp->lostjumbo > (t->nframes << 1)
 569                         && ifp->maxbcnt != DEFAULTBCNT) {
 570                                 printk(KERN_INFO
 571                                         "aoe: e%ld.%d: "
 572                                         "too many lost jumbo on "
 573                                         "%s:%012llx - "
 574                                         "falling back to %d frames.\n",
 575                                         d->aoemajor, d->aoeminor,
 576                                         ifp->nd->name, mac_addr(t->addr),
 577                                         DEFAULTBCNT);
 578                                 ifp->maxbcnt = 0;
 579                         }
 580                         resend(d, t, f);
 581                 }
 582
 583                 /* window check */
 584                 if (t->nout == t->maxout
 585                 && t->maxout < t->nframes
 586                 && (jiffies - t->lastwadj)/HZ > 10) {
 587                         t->maxout++;
 588                         t->lastwadj = jiffies;
 589                 }
 590         }
 591
 592         if (d->sendq_hd) {
 593                 n = d->rttavg <<= 1;
 594                 if (n > MAXTIMER)
 595                         d->rttavg = MAXTIMER;
 596         }
 597
 598         if (d->flags & DEVFL_KICKME || d->htgt) {
 599                 d->flags &= ~DEVFL_KICKME;
 600                 aoecmd_work(d);
 601         }
 602
 603         sl = d->sendq_hd;
 604         d->sendq_hd = d->sendq_tl = NULL;
 605
 606         d->timer.expires = jiffies + TIMERTICK;
 607         add_timer(&d->timer);
 608
 609         spin_unlock_irqrestore(&d->lock, flags);
 610
 611         aoenet_xmit(sl);
 612 }
 613
 614 /* enters with d->lock held */
 615 void
 616 aoecmd_work(struct aoedev *d)
 617 {
 618         struct buf *buf;
 619 loop:
 620         if (d->htgt && !sthtith(d))
 621                 return;
 622         if (d->inprocess == NULL) {
 623                 if (list_empty(&d->bufq))
 624                         return;
 625                 buf = container_of(d->bufq.next, struct buf, bufs);
 626                 list_del(d->bufq.next);
 627                 d->inprocess = buf;
 628         }
 629         if (aoecmd_ata_rw(d))
 630                 goto loop;
 631 }
 632
 633 /* this function performs work that has been deferred until sleeping is OK
 634  */
 635 void
 636 aoecmd_sleepwork(struct work_struct *work)
 637 {
 638         struct aoedev *d = container_of(work, struct aoedev, work);
 639
 640         if (d->flags & DEVFL_GDALLOC)
 641                 aoeblk_gdalloc(d);
 642
 643         if (d->flags & DEVFL_NEWSIZE) {
 644                 struct block_device *bd;
 645                 unsigned long flags;
 646                 u64 ssize;
 647
 648                 ssize = get_capacity(d->gd);
 649                 bd = bdget_disk(d->gd, 0);
 650
 651                 if (bd) {
 652                         mutex_lock(&bd->bd_inode->i_mutex);
 653                         i_size_write(bd->bd_inode, (loff_t)ssize<<9);
 654                         mutex_unlock(&bd->bd_inode->i_mutex);
 655                         bdput(bd);
 656                 }
 657                 spin_lock_irqsave(&d->lock, flags);
 658                 d->flags |= DEVFL_UP;
 659                 d->flags &= ~DEVFL_NEWSIZE;
 660                 spin_unlock_irqrestore(&d->lock, flags);
 661         }
 662 }
 663
 664 static void
 665 ataid_complete(struct aoedev *d, struct aoetgt *t, unsigned char *id)
 666 {
 667         u64 ssize;
 668         u16 n;
 669
 670         /* word 83: command set supported */
 671         n = get_unaligned_le16(&id[83 << 1]);
 672
 673         /* word 86: command set/feature enabled */
 674         n |= get_unaligned_le16(&id[86 << 1]);
 675
 676         if (n & (1<<10)) {      /* bit 10: LBA 48 */
 677                 d->flags |= DEVFL_EXT;
 678
 679                 /* word 100: number lba48 sectors */
 680                 ssize = get_unaligned_le64(&id[100 << 1]);
 681
 682                 /* set as in ide-disk.c:init_idedisk_capacity */
 683                 d->geo.cylinders = ssize;
 684                 d->geo.cylinders /= (255 * 63);
 685                 d->geo.heads = 255;
 686                 d->geo.sectors = 63;
 687         } else {
 688                 d->flags &= ~DEVFL_EXT;
 689
 690                 /* number lba28 sectors */
 691                 ssize = get_unaligned_le32(&id[60 << 1]);
 692
 693                 /* NOTE: obsolete in ATA 6 */
 694                 d->geo.cylinders = get_unaligned_le16(&id[54 << 1]);
 695                 d->geo.heads = get_unaligned_le16(&id[55 << 1]);
 696                 d->geo.sectors = get_unaligned_le16(&id[56 << 1]);
 697         }
 698
 699         if (d->ssize != ssize)
 700                 printk(KERN_INFO
 701                         "aoe: %012llx e%ld.%d v%04x has %llu sectors\n",
 702                         mac_addr(t->addr),
 703                         d->aoemajor, d->aoeminor,
 704                         d->fw_ver, (long long)ssize);
 705         d->ssize = ssize;
 706         d->geo.start = 0;
 707         if (d->flags & (DEVFL_GDALLOC|DEVFL_NEWSIZE))
 708                 return;
 709         if (d->gd != NULL) {
 710                 set_capacity(d->gd, ssize);
 711                 d->flags |= DEVFL_NEWSIZE;
 712         } else
 713                 d->flags |= DEVFL_GDALLOC;
 714         schedule_work(&d->work);
 715 }
 716
 717 static void
 718 calc_rttavg(struct aoedev *d, int rtt)
 719 {
 720         register long n;
 721
 722         n = rtt;
 723         if (n < 0) {
 724                 n = -rtt;
 725                 if (n < MINTIMER)
 726                         n = MINTIMER;
 727                 else if (n > MAXTIMER)
 728                         n = MAXTIMER;
 729                 d->mintimer += (n - d->mintimer) >> 1;
 730         } else if (n < d->mintimer)
 731                 n = d->mintimer;
 732         else if (n > MAXTIMER)
 733                 n = MAXTIMER;
 734
 735         /* g == .25; cf. Congestion Avoidance and Control, Jacobson & Karels; 1988 */
 736         n -= d->rttavg;
 737         d->rttavg += n >> 2;
 738 }
 739
 740 static struct aoetgt *
 741 gettgt(struct aoedev *d, char *addr)
 742 {
 743         struct aoetgt **t, **e;
 744
 745         t = d->targets;
 746         e = t + NTARGETS;
 747         for (; t < e && *t; t++)
 748                 if (memcmp((*t)->addr, addr, sizeof((*t)->addr)) == 0)
 749                         return *t;
 750         return NULL;
 751 }
 752
 753 static inline void
 754 diskstats(struct gendisk *disk, struct bio *bio, ulong duration, sector_t sector)
 755 {
 756         unsigned long n_sect = bio->bi_size >> 9;
 757         const int rw = bio_data_dir(bio);
 758         struct hd_struct *part;
 759         int cpu;
 760
 761         cpu = disk_stat_lock();
 762         part = disk_map_sector_rcu(disk, sector);
 763
 764         all_stat_inc(cpu, disk, part, ios[rw], sector);
 765         all_stat_add(cpu, disk, part, ticks[rw], duration, sector);
 766         all_stat_add(cpu, disk, part, sectors[rw], n_sect, sector);
 767         all_stat_add(cpu, disk, part, io_ticks, duration, sector);
 768
 769         disk_stat_unlock();
 770 }
 771
 772 void
 773 aoecmd_ata_rsp(struct sk_buff *skb)
 774 {
 775         struct aoedev *d;
 776         struct aoe_hdr *hin, *hout;
 777         struct aoe_atahdr *ahin, *ahout;
 778         struct frame *f;
 779         struct buf *buf;
 780         struct sk_buff *sl;
 781         struct aoetgt *t;
 782         struct aoeif *ifp;
 783         register long n;
 784         ulong flags;
 785         char ebuf[128];
 786         u16 aoemajor;
 787
 788         hin = (struct aoe_hdr *) skb_mac_header(skb);
 789         aoemajor = get_unaligned_be16(&hin->major);
 790         d = aoedev_by_aoeaddr(aoemajor, hin->minor);
 791         if (d == NULL) {
 792                 snprintf(ebuf, sizeof ebuf, "aoecmd_ata_rsp: ata response "
 793                         "for unknown device %d.%d\n",
 794                          aoemajor, hin->minor);
 795                 aoechr_error(ebuf);
 796                 return;
 797         }
 798
 799         spin_lock_irqsave(&d->lock, flags);
 800
 801         n = get_unaligned_be32(&hin->tag);
 802         t = gettgt(d, hin->src);
 803         if (t == NULL) {
 804                 printk(KERN_INFO "aoe: can't find target e%ld.%d:%012llx\n",
 805                         d->aoemajor, d->aoeminor, mac_addr(hin->src));
 806                 spin_unlock_irqrestore(&d->lock, flags);
 807                 return;
 808         }
 809         f = getframe(t, n);
 810         if (f == NULL) {
 811                 calc_rttavg(d, -tsince(n));
 812                 spin_unlock_irqrestore(&d->lock, flags);
 813                 snprintf(ebuf, sizeof ebuf,
 814                         "%15s e%d.%d    tag=%08x@%08lx\n",
 815                         "unexpected rsp",
 816                         get_unaligned_be16(&hin->major),
 817                         hin->minor,
 818                         get_unaligned_be32(&hin->tag),
 819                         jiffies);
 820                 aoechr_error(ebuf);
 821                 return;
 822         }
 823
 824         calc_rttavg(d, tsince(f->tag));
 825
 826         ahin = (struct aoe_atahdr *) (hin+1);
 827         hout = (struct aoe_hdr *) skb_mac_header(f->skb);
 828         ahout = (struct aoe_atahdr *) (hout+1);
 829         buf = f->buf;
 830
 831         if (ahin->cmdstat & 0xa9) {     /* these bits cleared on success */
 832                 printk(KERN_ERR
 833                         "aoe: ata error cmd=%2.2Xh stat=%2.2Xh from e%ld.%d\n",
 834                         ahout->cmdstat, ahin->cmdstat,
 835                         d->aoemajor, d->aoeminor);
 836                 if (buf)
 837                         buf->flags |= BUFFL_FAIL;
 838         } else {
 839                 if (d->htgt && t == *d->htgt) /* I'll help myself, thank you. */
 840                         d->htgt = NULL;
 841                 n = ahout->scnt << 9;
 842                 switch (ahout->cmdstat) {
 843                 case WIN_READ:
 844                 case WIN_READ_EXT:
 845                         if (skb->len - sizeof *hin - sizeof *ahin < n) {
 846                                 printk(KERN_ERR
 847                                         "aoe: %s.  skb->len=%d need=%ld\n",
 848                                         "runt data size in read", skb->len, n);
 849                                 /* fail frame f?  just returning will rexmit. */
 850                                 spin_unlock_irqrestore(&d->lock, flags);
 851                                 return;
 852                         }
 853                         memcpy(f->bufaddr, ahin+1, n);
 854                 case WIN_WRITE:
 855                 case WIN_WRITE_EXT:
 856                         ifp = getif(t, skb->dev);
 857                         if (ifp) {
 858                                 ifp->lost = 0;
 859                                 if (n > DEFAULTBCNT)
 860                                         ifp->lostjumbo = 0;
 861                         }
 862                         if (f->bcnt -= n) {
 863                                 f->lba += n >> 9;
 864                                 f->bufaddr += n;
 865                                 resend(d, t, f);
 866                                 goto xmit;
 867                         }
 868                         break;
 869                 case WIN_IDENTIFY:
 870                         if (skb->len - sizeof *hin - sizeof *ahin < 512) {
 871                                 printk(KERN_INFO
 872                                         "aoe: runt data size in ataid.  skb->len=%d\n",
 873                                         skb->len);
 874                                 spin_unlock_irqrestore(&d->lock, flags);
 875                                 return;
 876                         }
 877                         ataid_complete(d, t, (char *) (ahin+1));
 878                         break;
 879                 default:
 880                         printk(KERN_INFO
 881                                 "aoe: unrecognized ata command %2.2Xh for %d.%d\n",
 882                                 ahout->cmdstat,
 883                                 get_unaligned_be16(&hin->major),
 884                                 hin->minor);
 885                 }
 886         }
 887
 888         if (buf && --buf->nframesout == 0 && buf->resid == 0) {
 889                 diskstats(d->gd, buf->bio, jiffies - buf->stime, buf->sector);
 890                 n = (buf->flags & BUFFL_FAIL) ? -EIO : 0;
 891                 bio_endio(buf->bio, n);
 892                 mempool_free(buf, d->bufpool);
 893         }
 894
 895         f->buf = NULL;
 896         f->tag = FREETAG;
 897         t->nout--;
 898
 899         aoecmd_work(d);
 900 xmit:
 901         sl = d->sendq_hd;
 902         d->sendq_hd = d->sendq_tl = NULL;
 903
 904         spin_unlock_irqrestore(&d->lock, flags);
 905         aoenet_xmit(sl);
 906 }
 907
 908 void
 909 aoecmd_cfg(ushort aoemajor, unsigned char aoeminor)
 910 {
 911         struct sk_buff *sl;
 912
 913         sl = aoecmd_cfg_pkts(aoemajor, aoeminor, NULL);
 914
 915         aoenet_xmit(sl);
 916 }
 917
 918 struct sk_buff *
 919 aoecmd_ata_id(struct aoedev *d)
 920 {
 921         struct aoe_hdr *h;
 922         struct aoe_atahdr *ah;
 923         struct frame *f;
 924         struct sk_buff *skb;
 925         struct aoetgt *t;
 926
 927         f = freeframe(d);
 928         if (f == NULL)
 929                 return NULL;
 930
 931         t = *d->tgt;
 932
 933         /* initialize the headers & frame */
 934         skb = f->skb;
 935         h = (struct aoe_hdr *) skb_mac_header(skb);
 936         ah = (struct aoe_atahdr *) (h+1);
 937         skb_put(skb, sizeof *h + sizeof *ah);
 938         memset(h, 0, skb->len);
 939         f->tag = aoehdr_atainit(d, t, h);
 940         t->nout++;
 941         f->waited = 0;
 942
 943         /* set up ata header */
 944         ah->scnt = 1;
 945         ah->cmdstat = WIN_IDENTIFY;
 946         ah->lba3 = 0xa0;
 947
 948         skb->dev = t->ifp->nd;
 949
 950         d->rttavg = MAXTIMER;
 951         d->timer.function = rexmit_timer;
 952
 953         return skb_clone(skb, GFP_ATOMIC);
 954 }
 955
 956 static struct aoetgt *
 957 addtgt(struct aoedev *d, char *addr, ulong nframes)
 958 {
 959         struct aoetgt *t, **tt, **te;
 960         struct frame *f, *e;
 961
 962         tt = d->targets;
 963         te = tt + NTARGETS;
 964         for (; tt < te && *tt; tt++)
 965                 ;
 966
 967         if (tt == te) {
 968                 printk(KERN_INFO
 969                         "aoe: device addtgt failure; too many targets\n");
 970                 return NULL;
 971         }
 972         t = kcalloc(1, sizeof *t, GFP_ATOMIC);
 973         f = kcalloc(nframes, sizeof *f, GFP_ATOMIC);
 974         if (!t || !f) {
 975                 kfree(f);
 976                 kfree(t);
 977                 printk(KERN_INFO "aoe: cannot allocate memory to add target\n");
 978                 return NULL;
 979         }
 980
 981         t->nframes = nframes;
 982         t->frames = f;
 983         e = f + nframes;
 984         for (; f < e; f++)
 985                 f->tag = FREETAG;
 986         memcpy(t->addr, addr, sizeof t->addr);
 987         t->ifp = t->ifs;
 988         t->maxout = t->nframes;
 989         return *tt = t;
 990 }
 991
 992 void
 993 aoecmd_cfg_rsp(struct sk_buff *skb)
 994 {
 995         struct aoedev *d;
 996         struct aoe_hdr *h;
 997         struct aoe_cfghdr *ch;
 998         struct aoetgt *t;
 999         struct aoeif *ifp;
1000         ulong flags, sysminor, aoemajor;
1001         struct sk_buff *sl;
1002         u16 n;
1003
1004         h = (struct aoe_hdr *) skb_mac_header(skb);
1005         ch = (struct aoe_cfghdr *) (h+1);
1006
1007         /*
1008          * Enough people have their dip switches set backwards to
1009          * warrant a loud message for this special case.
1010          */
1011         aoemajor = get_unaligned_be16(&h->major);
1012         if (aoemajor == 0xfff) {
1013                 printk(KERN_ERR "aoe: Warning: shelf address is all ones.  "
1014                         "Check shelf dip switches.\n");
1015                 return;
1016         }
1017
1018         sysminor = SYSMINOR(aoemajor, h->minor);
1019         if (sysminor * AOE_PARTITIONS + AOE_PARTITIONS > MINORMASK) {
1020                 printk(KERN_INFO "aoe: e%ld.%d: minor number too large\n",
1021                         aoemajor, (int) h->minor);
1022                 return;
1023         }
1024
1025         n = be16_to_cpu(ch->bufcnt);
1026         if (n > aoe_maxout)     /* keep it reasonable */
1027                 n = aoe_maxout;
1028
1029         d = aoedev_by_sysminor_m(sysminor);
1030         if (d == NULL) {
1031                 printk(KERN_INFO "aoe: device sysminor_m failure\n");
1032                 return;
1033         }
1034
1035         spin_lock_irqsave(&d->lock, flags);
1036
1037         t = gettgt(d, h->src);
1038         if (!t) {
1039                 t = addtgt(d, h->src, n);
1040                 if (!t) {
1041                         spin_unlock_irqrestore(&d->lock, flags);
1042                         return;
1043                 }
1044         }
1045         ifp = getif(t, skb->dev);
1046         if (!ifp) {
1047                 ifp = addif(t, skb->dev);
1048                 if (!ifp) {
1049                         printk(KERN_INFO
1050                                 "aoe: device addif failure; "
1051                                 "too many interfaces?\n");
1052                         spin_unlock_irqrestore(&d->lock, flags);
1053                         return;
1054                 }
1055         }
1056         if (ifp->maxbcnt) {
1057                 n = ifp->nd->mtu;
1058                 n -= sizeof (struct aoe_hdr) + sizeof (struct aoe_atahdr);
1059                 n /= 512;
1060                 if (n > ch->scnt)
1061                         n = ch->scnt;
1062                 n = n ? n * 512 : DEFAULTBCNT;
1063                 if (n != ifp->maxbcnt) {
1064                         printk(KERN_INFO
1065                                 "aoe: e%ld.%d: setting %d%s%s:%012llx\n",
1066                                 d->aoemajor, d->aoeminor, n,
1067                                 " byte data frames on ", ifp->nd->name,
1068                                 mac_addr(t->addr));
1069                         ifp->maxbcnt = n;
1070                 }
1071         }
1072
1073         /* don't change users' perspective */
1074         if (d->nopen) {
1075                 spin_unlock_irqrestore(&d->lock, flags);
1076                 return;
1077         }
1078         d->fw_ver = be16_to_cpu(ch->fwver);
1079
1080         sl = aoecmd_ata_id(d);
1081
1082         spin_unlock_irqrestore(&d->lock, flags);
1083
1084         aoenet_xmit(sl);
1085 }
1086
1087 void
1088 aoecmd_cleanslate(struct aoedev *d)
1089 {
1090         struct aoetgt **t, **te;
1091         struct aoeif *p, *e;
1092
1093         d->mintimer = MINTIMER;
1094
1095         t = d->targets;
1096         te = t + NTARGETS;
1097         for (; t < te && *t; t++) {
1098                 (*t)->maxout = (*t)->nframes;
1099                 p = (*t)->ifs;
1100                 e = p + NAOEIFS;
1101                 for (; p < e; p++) {
1102                         p->lostjumbo = 0;
1103                         p->lost = 0;
1104                         p->maxbcnt = DEFAULTBCNT;
1105                 }
1106         }
1107 }