drivers/block/aoe/aoecmd.c

   1 /* Copyright (c) 2006 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 <asm/unaligned.h>
  13 #include "aoe.h"
  14
  15 #define TIMERTICK (HZ / 10)
  16 #define MINTIMER (2 * TIMERTICK)
  17 #define MAXTIMER (HZ << 1)
  18
  19 static int aoe_deadsecs = 60 * 3;
  20 module_param(aoe_deadsecs, int, 0644);
  21 MODULE_PARM_DESC(aoe_deadsecs, "After aoe_deadsecs seconds, give up and fail dev.");
  22
  23 struct sk_buff *
  24 new_skb(ulong len)
  25 {
  26         struct sk_buff *skb;
  27
  28         skb = alloc_skb(len, GFP_ATOMIC);
  29         if (skb) {
  30                 skb->nh.raw = skb->mac.raw = skb->data;
  31                 skb->protocol = __constant_htons(ETH_P_AOE);
  32                 skb->priority = 0;
  33                 skb_put(skb, len);
  34                 memset(skb->head, 0, len);
  35                 skb->next = skb->prev = NULL;
  36
  37                 /* tell the network layer not to perform IP checksums
  38                  * or to get the NIC to do it
  39                  */
  40                 skb->ip_summed = CHECKSUM_NONE;
  41         }
  42         return skb;
  43 }
  44
  45 static struct frame *
  46 getframe(struct aoedev *d, int tag)
  47 {
  48         struct frame *f, *e;
  49
  50         f = d->frames;
  51         e = f + d->nframes;
  52         for (; f<e; f++)
  53                 if (f->tag == tag)
  54                         return f;
  55         return NULL;
  56 }
  57
  58 /*
  59  * Leave the top bit clear so we have tagspace for userland.
  60  * The bottom 16 bits are the xmit tick for rexmit/rttavg processing.
  61  * This driver reserves tag -1 to mean "unused frame."
  62  */
  63 static int
  64 newtag(struct aoedev *d)
  65 {
  66         register ulong n;
  67
  68         n = jiffies & 0xffff;
  69         return n |= (++d->lasttag & 0x7fff) << 16;
  70 }
  71
  72 static int
  73 aoehdr_atainit(struct aoedev *d, struct aoe_hdr *h)
  74 {
  75         u32 host_tag = newtag(d);
  76
  77         memcpy(h->src, d->ifp->dev_addr, sizeof h->src);
  78         memcpy(h->dst, d->addr, sizeof h->dst);
  79         h->type = __constant_cpu_to_be16(ETH_P_AOE);
  80         h->verfl = AOE_HVER;
  81         h->major = cpu_to_be16(d->aoemajor);
  82         h->minor = d->aoeminor;
  83         h->cmd = AOECMD_ATA;
  84         h->tag = cpu_to_be32(host_tag);
  85
  86         return host_tag;
  87 }
  88
  89 static inline void
  90 put_lba(struct aoe_atahdr *ah, sector_t lba)
  91 {
  92         ah->lba0 = lba;
  93         ah->lba1 = lba >>= 8;
  94         ah->lba2 = lba >>= 8;
  95         ah->lba3 = lba >>= 8;
  96         ah->lba4 = lba >>= 8;
  97         ah->lba5 = lba >>= 8;
  98 }
  99
 100 static void
 101 aoecmd_ata_rw(struct aoedev *d, struct frame *f)
 102 {
 103         struct aoe_hdr *h;
 104         struct aoe_atahdr *ah;
 105         struct buf *buf;
 106         struct sk_buff *skb;
 107         ulong bcnt;
 108         register sector_t sector;
 109         char writebit, extbit;
 110
 111         writebit = 0x10;
 112         extbit = 0x4;
 113
 114         buf = d->inprocess;
 115
 116         sector = buf->sector;
 117         bcnt = buf->bv_resid;
 118         if (bcnt > d->maxbcnt)
 119                 bcnt = d->maxbcnt;
 120
 121         /* initialize the headers & frame */
 122         skb = f->skb;
 123         h = (struct aoe_hdr *) skb->mac.raw;
 124         ah = (struct aoe_atahdr *) (h+1);
 125         skb->len = sizeof *h + sizeof *ah;
 126         memset(h, 0, ETH_ZLEN);
 127         f->tag = aoehdr_atainit(d, h);
 128         f->waited = 0;
 129         f->buf = buf;
 130         f->bufaddr = buf->bufaddr;
 131         f->bcnt = bcnt;
 132         f->lba = sector;
 133
 134         /* set up ata header */
 135         ah->scnt = bcnt >> 9;
 136         put_lba(ah, sector);
 137         if (d->flags & DEVFL_EXT) {
 138                 ah->aflags |= AOEAFL_EXT;
 139         } else {
 140                 extbit = 0;
 141                 ah->lba3 &= 0x0f;
 142                 ah->lba3 |= 0xe0;       /* LBA bit + obsolete 0xa0 */
 143         }
 144
 145         if (bio_data_dir(buf->bio) == WRITE) {
 146                 skb_fill_page_desc(skb, 0, virt_to_page(f->bufaddr),
 147                         offset_in_page(f->bufaddr), bcnt);
 148                 ah->aflags |= AOEAFL_WRITE;
 149                 skb->len += bcnt;
 150                 skb->data_len = bcnt;
 151         } else {
 152                 skb->len = ETH_ZLEN;
 153                 writebit = 0;
 154         }
 155
 156         ah->cmdstat = WIN_READ | writebit | extbit;
 157
 158         /* mark all tracking fields and load out */
 159         buf->nframesout += 1;
 160         buf->bufaddr += bcnt;
 161         buf->bv_resid -= bcnt;
 162 /* dprintk("bv_resid=%ld\n", buf->bv_resid); */
 163         buf->resid -= bcnt;
 164         buf->sector += bcnt >> 9;
 165         if (buf->resid == 0) {
 166                 d->inprocess = NULL;
 167         } else if (buf->bv_resid == 0) {
 168                 buf->bv++;
 169                 buf->bv_resid = buf->bv->bv_len;
 170                 buf->bufaddr = page_address(buf->bv->bv_page) + buf->bv->bv_offset;
 171         }
 172
 173         skb->dev = d->ifp;
 174         skb = skb_clone(skb, GFP_ATOMIC);
 175         if (skb == NULL)
 176                 return;
 177         if (d->sendq_hd)
 178                 d->sendq_tl->next = skb;
 179         else
 180                 d->sendq_hd = skb;
 181         d->sendq_tl = skb;
 182 }
 183
 184 /* some callers cannot sleep, and they can call this function,
 185  * transmitting the packets later, when interrupts are on
 186  */
 187 static struct sk_buff *
 188 aoecmd_cfg_pkts(ushort aoemajor, unsigned char aoeminor, struct sk_buff **tail)
 189 {
 190         struct aoe_hdr *h;
 191         struct aoe_cfghdr *ch;
 192         struct sk_buff *skb, *sl, *sl_tail;
 193         struct net_device *ifp;
 194
 195         sl = sl_tail = NULL;
 196
 197         read_lock(&dev_base_lock);
 198         for (ifp = dev_base; ifp; dev_put(ifp), ifp = ifp->next) {
 199                 dev_hold(ifp);
 200                 if (!is_aoe_netif(ifp))
 201                         continue;
 202
 203                 skb = new_skb(sizeof *h + sizeof *ch);
 204                 if (skb == NULL) {
 205                         iprintk("skb alloc failure\n");
 206                         continue;
 207                 }
 208                 skb->dev = ifp;
 209                 if (sl_tail == NULL)
 210                         sl_tail = skb;
 211                 h = (struct aoe_hdr *) skb->mac.raw;
 212                 memset(h, 0, sizeof *h + sizeof *ch);
 213
 214                 memset(h->dst, 0xff, sizeof h->dst);
 215                 memcpy(h->src, ifp->dev_addr, sizeof h->src);
 216                 h->type = __constant_cpu_to_be16(ETH_P_AOE);
 217                 h->verfl = AOE_HVER;
 218                 h->major = cpu_to_be16(aoemajor);
 219                 h->minor = aoeminor;
 220                 h->cmd = AOECMD_CFG;
 221
 222                 skb->next = sl;
 223                 sl = skb;
 224         }
 225         read_unlock(&dev_base_lock);
 226
 227         if (tail != NULL)
 228                 *tail = sl_tail;
 229         return sl;
 230 }
 231
 232 static struct frame *
 233 freeframe(struct aoedev *d)
 234 {
 235         struct frame *f, *e;
 236         int n = 0;
 237
 238         f = d->frames;
 239         e = f + d->nframes;
 240         for (; f<e; f++) {
 241                 if (f->tag != FREETAG)
 242                         continue;
 243                 if (atomic_read(&skb_shinfo(f->skb)->dataref) == 1) {
 244                         skb_shinfo(f->skb)->nr_frags = f->skb->data_len = 0;
 245                         return f;
 246                 }
 247                 n++;
 248         }
 249         if (n == d->nframes)    /* wait for network layer */
 250                 d->flags |= DEVFL_KICKME;
 251
 252         return NULL;
 253 }
 254
 255 /* enters with d->lock held */
 256 void
 257 aoecmd_work(struct aoedev *d)
 258 {
 259         struct frame *f;
 260         struct buf *buf;
 261
 262         if (d->flags & DEVFL_PAUSE) {
 263                 if (!aoedev_isbusy(d))
 264                         d->sendq_hd = aoecmd_cfg_pkts(d->aoemajor,
 265                                                 d->aoeminor, &d->sendq_tl);
 266                 return;
 267         }
 268
 269 loop:
 270         f = freeframe(d);
 271         if (f == NULL)
 272                 return;
 273         if (d->inprocess == NULL) {
 274                 if (list_empty(&d->bufq))
 275                         return;
 276                 buf = container_of(d->bufq.next, struct buf, bufs);
 277                 list_del(d->bufq.next);
 278 /*dprintk("bi_size=%ld\n", buf->bio->bi_size); */
 279                 d->inprocess = buf;
 280         }
 281         aoecmd_ata_rw(d, f);
 282         goto loop;
 283 }
 284
 285 static void
 286 rexmit(struct aoedev *d, struct frame *f)
 287 {
 288         struct sk_buff *skb;
 289         struct aoe_hdr *h;
 290         struct aoe_atahdr *ah;
 291         char buf[128];
 292         u32 n;
 293
 294         n = newtag(d);
 295
 296         snprintf(buf, sizeof buf,
 297                 "%15s e%ld.%ld oldtag=%08x@%08lx newtag=%08x\n",
 298                 "retransmit",
 299                 d->aoemajor, d->aoeminor, f->tag, jiffies, n);
 300         aoechr_error(buf);
 301
 302         skb = f->skb;
 303         h = (struct aoe_hdr *) skb->mac.raw;
 304         ah = (struct aoe_atahdr *) (h+1);
 305         f->tag = n;
 306         h->tag = cpu_to_be32(n);
 307         memcpy(h->dst, d->addr, sizeof h->dst);
 308         memcpy(h->src, d->ifp->dev_addr, sizeof h->src);
 309
 310         n = DEFAULTBCNT / 512;
 311         if (ah->scnt > n) {
 312                 ah->scnt = n;
 313                 if (ah->aflags & AOEAFL_WRITE) {
 314                         skb_fill_page_desc(skb, 0, virt_to_page(f->bufaddr),
 315                                 offset_in_page(f->bufaddr), DEFAULTBCNT);
 316                         skb->len = sizeof *h + sizeof *ah + DEFAULTBCNT;
 317                         skb->data_len = DEFAULTBCNT;
 318                 }
 319                 if (++d->lostjumbo > (d->nframes << 1))
 320                 if (d->maxbcnt != DEFAULTBCNT) {
 321                         iprintk("e%ld.%ld: too many lost jumbo on %s - using 1KB frames.\n",
 322                                 d->aoemajor, d->aoeminor, d->ifp->name);
 323                         d->maxbcnt = DEFAULTBCNT;
 324                         d->flags |= DEVFL_MAXBCNT;
 325                 }
 326         }
 327
 328         skb->dev = d->ifp;
 329         skb = skb_clone(skb, GFP_ATOMIC);
 330         if (skb == NULL)
 331                 return;
 332         if (d->sendq_hd)
 333                 d->sendq_tl->next = skb;
 334         else
 335                 d->sendq_hd = skb;
 336         d->sendq_tl = skb;
 337 }
 338
 339 static int
 340 tsince(int tag)
 341 {
 342         int n;
 343
 344         n = jiffies & 0xffff;
 345         n -= tag & 0xffff;
 346         if (n < 0)
 347                 n += 1<<16;
 348         return n;
 349 }
 350
 351 static void
 352 rexmit_timer(ulong vp)
 353 {
 354         struct aoedev *d;
 355         struct frame *f, *e;
 356         struct sk_buff *sl;
 357         register long timeout;
 358         ulong flags, n;
 359
 360         d = (struct aoedev *) vp;
 361         sl = NULL;
 362
 363         /* timeout is always ~150% of the moving average */
 364         timeout = d->rttavg;
 365         timeout += timeout >> 1;
 366
 367         spin_lock_irqsave(&d->lock, flags);
 368
 369         if (d->flags & DEVFL_TKILL) {
 370                 spin_unlock_irqrestore(&d->lock, flags);
 371                 return;
 372         }
 373         f = d->frames;
 374         e = f + d->nframes;
 375         for (; f<e; f++) {
 376                 if (f->tag != FREETAG && tsince(f->tag) >= timeout) {
 377                         n = f->waited += timeout;
 378                         n /= HZ;
 379                         if (n > aoe_deadsecs) { /* waited too long for response */
 380                                 aoedev_downdev(d);
 381                                 break;
 382                         }
 383                         rexmit(d, f);
 384                 }
 385         }
 386         if (d->flags & DEVFL_KICKME) {
 387                 d->flags &= ~DEVFL_KICKME;
 388                 aoecmd_work(d);
 389         }
 390
 391         sl = d->sendq_hd;
 392         d->sendq_hd = d->sendq_tl = NULL;
 393         if (sl) {
 394                 n = d->rttavg <<= 1;
 395                 if (n > MAXTIMER)
 396                         d->rttavg = MAXTIMER;
 397         }
 398
 399         d->timer.expires = jiffies + TIMERTICK;
 400         add_timer(&d->timer);
 401
 402         spin_unlock_irqrestore(&d->lock, flags);
 403
 404         aoenet_xmit(sl);
 405 }
 406
 407 /* this function performs work that has been deferred until sleeping is OK
 408  */
 409 void
 410 aoecmd_sleepwork(void *vp)
 411 {
 412         struct aoedev *d = (struct aoedev *) vp;
 413
 414         if (d->flags & DEVFL_GDALLOC)
 415                 aoeblk_gdalloc(d);
 416
 417         if (d->flags & DEVFL_NEWSIZE) {
 418                 struct block_device *bd;
 419                 unsigned long flags;
 420                 u64 ssize;
 421
 422                 ssize = d->gd->capacity;
 423                 bd = bdget_disk(d->gd, 0);
 424
 425                 if (bd) {
 426                         mutex_lock(&bd->bd_inode->i_mutex);
 427                         i_size_write(bd->bd_inode, (loff_t)ssize<<9);
 428                         mutex_unlock(&bd->bd_inode->i_mutex);
 429                         bdput(bd);
 430                 }
 431                 spin_lock_irqsave(&d->lock, flags);
 432                 d->flags |= DEVFL_UP;
 433                 d->flags &= ~DEVFL_NEWSIZE;
 434                 spin_unlock_irqrestore(&d->lock, flags);
 435         }
 436 }
 437
 438 static void
 439 ataid_complete(struct aoedev *d, unsigned char *id)
 440 {
 441         u64 ssize;
 442         u16 n;
 443
 444         /* word 83: command set supported */
 445         n = le16_to_cpu(get_unaligned((__le16 *) &id[83<<1]));
 446
 447         /* word 86: command set/feature enabled */
 448         n |= le16_to_cpu(get_unaligned((__le16 *) &id[86<<1]));
 449
 450         if (n & (1<<10)) {      /* bit 10: LBA 48 */
 451                 d->flags |= DEVFL_EXT;
 452
 453                 /* word 100: number lba48 sectors */
 454                 ssize = le64_to_cpu(get_unaligned((__le64 *) &id[100<<1]));
 455
 456                 /* set as in ide-disk.c:init_idedisk_capacity */
 457                 d->geo.cylinders = ssize;
 458                 d->geo.cylinders /= (255 * 63);
 459                 d->geo.heads = 255;
 460                 d->geo.sectors = 63;
 461         } else {
 462                 d->flags &= ~DEVFL_EXT;
 463
 464                 /* number lba28 sectors */
 465                 ssize = le32_to_cpu(get_unaligned((__le32 *) &id[60<<1]));
 466
 467                 /* NOTE: obsolete in ATA 6 */
 468                 d->geo.cylinders = le16_to_cpu(get_unaligned((__le16 *) &id[54<<1]));
 469                 d->geo.heads = le16_to_cpu(get_unaligned((__le16 *) &id[55<<1]));
 470                 d->geo.sectors = le16_to_cpu(get_unaligned((__le16 *) &id[56<<1]));
 471         }
 472
 473         if (d->ssize != ssize)
 474                 iprintk("%012llx e%lu.%lu v%04x has %llu sectors\n",
 475                         (unsigned long long)mac_addr(d->addr),
 476                         d->aoemajor, d->aoeminor,
 477                         d->fw_ver, (long long)ssize);
 478         d->ssize = ssize;
 479         d->geo.start = 0;
 480         if (d->gd != NULL) {
 481                 d->gd->capacity = ssize;
 482                 d->flags |= DEVFL_NEWSIZE;
 483         } else {
 484                 if (d->flags & DEVFL_GDALLOC) {
 485                         eprintk("can't schedule work for e%lu.%lu, %s\n",
 486                                d->aoemajor, d->aoeminor,
 487                                "it's already on!  This shouldn't happen.\n");
 488                         return;
 489                 }
 490                 d->flags |= DEVFL_GDALLOC;
 491         }
 492         schedule_work(&d->work);
 493 }
 494
 495 static void
 496 calc_rttavg(struct aoedev *d, int rtt)
 497 {
 498         register long n;
 499
 500         n = rtt;
 501         if (n < 0) {
 502                 n = -rtt;
 503                 if (n < MINTIMER)
 504                         n = MINTIMER;
 505                 else if (n > MAXTIMER)
 506                         n = MAXTIMER;
 507                 d->mintimer += (n - d->mintimer) >> 1;
 508         } else if (n < d->mintimer)
 509                 n = d->mintimer;
 510         else if (n > MAXTIMER)
 511                 n = MAXTIMER;
 512
 513         /* g == .25; cf. Congestion Avoidance and Control, Jacobson & Karels; 1988 */
 514         n -= d->rttavg;
 515         d->rttavg += n >> 2;
 516 }
 517
 518 void
 519 aoecmd_ata_rsp(struct sk_buff *skb)
 520 {
 521         struct aoedev *d;
 522         struct aoe_hdr *hin, *hout;
 523         struct aoe_atahdr *ahin, *ahout;
 524         struct frame *f;
 525         struct buf *buf;
 526         struct sk_buff *sl;
 527         register long n;
 528         ulong flags;
 529         char ebuf[128];
 530         u16 aoemajor;
 531
 532         hin = (struct aoe_hdr *) skb->mac.raw;
 533         aoemajor = be16_to_cpu(hin->major);
 534         d = aoedev_by_aoeaddr(aoemajor, hin->minor);
 535         if (d == NULL) {
 536                 snprintf(ebuf, sizeof ebuf, "aoecmd_ata_rsp: ata response "
 537                         "for unknown device %d.%d\n",
 538                          aoemajor, hin->minor);
 539                 aoechr_error(ebuf);
 540                 return;
 541         }
 542
 543         spin_lock_irqsave(&d->lock, flags);
 544
 545         n = be32_to_cpu(hin->tag);
 546         f = getframe(d, n);
 547         if (f == NULL) {
 548                 calc_rttavg(d, -tsince(n));
 549                 spin_unlock_irqrestore(&d->lock, flags);
 550                 snprintf(ebuf, sizeof ebuf,
 551                         "%15s e%d.%d    tag=%08x@%08lx\n",
 552                         "unexpected rsp",
 553                         be16_to_cpu(hin->major),
 554                         hin->minor,
 555                         be32_to_cpu(hin->tag),
 556                         jiffies);
 557                 aoechr_error(ebuf);
 558                 return;
 559         }
 560
 561         calc_rttavg(d, tsince(f->tag));
 562
 563         ahin = (struct aoe_atahdr *) (hin+1);
 564         hout = (struct aoe_hdr *) f->skb->mac.raw;
 565         ahout = (struct aoe_atahdr *) (hout+1);
 566         buf = f->buf;
 567
 568         if (ahout->cmdstat == WIN_IDENTIFY)
 569                 d->flags &= ~DEVFL_PAUSE;
 570         if (ahin->cmdstat & 0xa9) {     /* these bits cleared on success */
 571                 eprintk("ata error cmd=%2.2Xh stat=%2.2Xh from e%ld.%ld\n",
 572                         ahout->cmdstat, ahin->cmdstat,
 573                         d->aoemajor, d->aoeminor);
 574                 if (buf)
 575                         buf->flags |= BUFFL_FAIL;
 576         } else {
 577                 n = ahout->scnt << 9;
 578                 switch (ahout->cmdstat) {
 579                 case WIN_READ:
 580                 case WIN_READ_EXT:
 581                         if (skb->len - sizeof *hin - sizeof *ahin < n) {
 582                                 eprintk("runt data size in read.  skb->len=%d\n",
 583                                         skb->len);
 584                                 /* fail frame f?  just returning will rexmit. */
 585                                 spin_unlock_irqrestore(&d->lock, flags);
 586                                 return;
 587                         }
 588                         memcpy(f->bufaddr, ahin+1, n);
 589                 case WIN_WRITE:
 590                 case WIN_WRITE_EXT:
 591                         if (f->bcnt -= n) {
 592                                 skb = f->skb;
 593                                 f->bufaddr += n;
 594                                 put_lba(ahout, f->lba += ahout->scnt);
 595                                 n = f->bcnt;
 596                                 if (n > DEFAULTBCNT)
 597                                         n = DEFAULTBCNT;
 598                                 ahout->scnt = n >> 9;
 599                                 if (ahout->aflags & AOEAFL_WRITE) {
 600                                         skb_fill_page_desc(skb, 0,
 601                                                 virt_to_page(f->bufaddr),
 602                                                 offset_in_page(f->bufaddr), n);
 603                                         skb->len = sizeof *hout + sizeof *ahout + n;
 604                                         skb->data_len = n;
 605                                 }
 606                                 f->tag = newtag(d);
 607                                 hout->tag = cpu_to_be32(f->tag);
 608                                 skb->dev = d->ifp;
 609                                 skb = skb_clone(skb, GFP_ATOMIC);
 610                                 spin_unlock_irqrestore(&d->lock, flags);
 611                                 if (skb)
 612                                         aoenet_xmit(skb);
 613                                 return;
 614                         }
 615                         if (n > DEFAULTBCNT)
 616                                 d->lostjumbo = 0;
 617                         break;
 618                 case WIN_IDENTIFY:
 619                         if (skb->len - sizeof *hin - sizeof *ahin < 512) {
 620                                 iprintk("runt data size in ataid.  skb->len=%d\n",
 621                                         skb->len);
 622                                 spin_unlock_irqrestore(&d->lock, flags);
 623                                 return;
 624                         }
 625                         ataid_complete(d, (char *) (ahin+1));
 626                         break;
 627                 default:
 628                         iprintk("unrecognized ata command %2.2Xh for %d.%d\n",
 629                                 ahout->cmdstat,
 630                                 be16_to_cpu(hin->major),
 631                                 hin->minor);
 632                 }
 633         }
 634
 635         if (buf) {
 636                 buf->nframesout -= 1;
 637                 if (buf->nframesout == 0 && buf->resid == 0) {
 638                         unsigned long duration = jiffies - buf->start_time;
 639                         unsigned long n_sect = buf->bio->bi_size >> 9;
 640                         struct gendisk *disk = d->gd;
 641                         const int rw = bio_data_dir(buf->bio);
 642
 643                         disk_stat_inc(disk, ios[rw]);
 644                         disk_stat_add(disk, ticks[rw], duration);
 645                         disk_stat_add(disk, sectors[rw], n_sect);
 646                         disk_stat_add(disk, io_ticks, duration);
 647                         n = (buf->flags & BUFFL_FAIL) ? -EIO : 0;
 648                         bio_endio(buf->bio, buf->bio->bi_size, n);
 649                         mempool_free(buf, d->bufpool);
 650                 }
 651         }
 652
 653         f->buf = NULL;
 654         f->tag = FREETAG;
 655
 656         aoecmd_work(d);
 657         sl = d->sendq_hd;
 658         d->sendq_hd = d->sendq_tl = NULL;
 659
 660         spin_unlock_irqrestore(&d->lock, flags);
 661         aoenet_xmit(sl);
 662 }
 663
 664 void
 665 aoecmd_cfg(ushort aoemajor, unsigned char aoeminor)
 666 {
 667         struct sk_buff *sl;
 668
 669         sl = aoecmd_cfg_pkts(aoemajor, aoeminor, NULL);
 670
 671         aoenet_xmit(sl);
 672 }
 673
 674 /*
 675  * Since we only call this in one place (and it only prepares one frame)
 676  * we just return the skb.  Usually we'd chain it up to the aoedev sendq.
 677  */
 678 static struct sk_buff *
 679 aoecmd_ata_id(struct aoedev *d)
 680 {
 681         struct aoe_hdr *h;
 682         struct aoe_atahdr *ah;
 683         struct frame *f;
 684         struct sk_buff *skb;
 685
 686         f = freeframe(d);
 687         if (f == NULL) {
 688                 eprintk("can't get a frame. This shouldn't happen.\n");
 689                 return NULL;
 690         }
 691
 692         /* initialize the headers & frame */
 693         skb = f->skb;
 694         h = (struct aoe_hdr *) skb->mac.raw;
 695         ah = (struct aoe_atahdr *) (h+1);
 696         skb->len = ETH_ZLEN;
 697         memset(h, 0, ETH_ZLEN);
 698         f->tag = aoehdr_atainit(d, h);
 699         f->waited = 0;
 700
 701         /* set up ata header */
 702         ah->scnt = 1;
 703         ah->cmdstat = WIN_IDENTIFY;
 704         ah->lba3 = 0xa0;
 705
 706         skb->dev = d->ifp;
 707
 708         d->rttavg = MAXTIMER;
 709         d->timer.function = rexmit_timer;
 710
 711         return skb_clone(skb, GFP_ATOMIC);
 712 }
 713
 714 void
 715 aoecmd_cfg_rsp(struct sk_buff *skb)
 716 {
 717         struct aoedev *d;
 718         struct aoe_hdr *h;
 719         struct aoe_cfghdr *ch;
 720         ulong flags, sysminor, aoemajor;
 721         struct sk_buff *sl;
 722         enum { MAXFRAMES = 16 };
 723         u16 n;
 724
 725         h = (struct aoe_hdr *) skb->mac.raw;
 726         ch = (struct aoe_cfghdr *) (h+1);
 727
 728         /*
 729          * Enough people have their dip switches set backwards to
 730          * warrant a loud message for this special case.
 731          */
 732         aoemajor = be16_to_cpu(h->major);
 733         if (aoemajor == 0xfff) {
 734                 eprintk("Warning: shelf address is all ones.  "
 735                         "Check shelf dip switches.\n");
 736                 return;
 737         }
 738
 739         sysminor = SYSMINOR(aoemajor, h->minor);
 740         if (sysminor * AOE_PARTITIONS + AOE_PARTITIONS > MINORMASK) {
 741                 iprintk("e%ld.%d: minor number too large\n",
 742                         aoemajor, (int) h->minor);
 743                 return;
 744         }
 745
 746         n = be16_to_cpu(ch->bufcnt);
 747         if (n > MAXFRAMES)      /* keep it reasonable */
 748                 n = MAXFRAMES;
 749
 750         d = aoedev_by_sysminor_m(sysminor, n);
 751         if (d == NULL) {
 752                 iprintk("device sysminor_m failure\n");
 753                 return;
 754         }
 755
 756         spin_lock_irqsave(&d->lock, flags);
 757
 758         /* permit device to migrate mac and network interface */
 759         d->ifp = skb->dev;
 760         memcpy(d->addr, h->src, sizeof d->addr);
 761         if (!(d->flags & DEVFL_MAXBCNT)) {
 762                 n = d->ifp->mtu;
 763                 n -= sizeof (struct aoe_hdr) + sizeof (struct aoe_atahdr);
 764                 n /= 512;
 765                 if (n > ch->scnt)
 766                         n = ch->scnt;
 767                 n = n ? n * 512 : DEFAULTBCNT;
 768                 if (n != d->maxbcnt) {
 769                         iprintk("e%ld.%ld: setting %d byte data frames on %s\n",
 770                                 d->aoemajor, d->aoeminor, n, d->ifp->name);
 771                         d->maxbcnt = n;
 772                 }
 773         }
 774
 775         /* don't change users' perspective */
 776         if (d->nopen && !(d->flags & DEVFL_PAUSE)) {
 777                 spin_unlock_irqrestore(&d->lock, flags);
 778                 return;
 779         }
 780         d->flags |= DEVFL_PAUSE;        /* force pause */
 781         d->mintimer = MINTIMER;
 782         d->fw_ver = be16_to_cpu(ch->fwver);
 783
 784         /* check for already outstanding ataid */
 785         sl = aoedev_isbusy(d) == 0 ? aoecmd_ata_id(d) : NULL;
 786
 787         spin_unlock_irqrestore(&d->lock, flags);
 788
 789         aoenet_xmit(sl);
 790 }
 791