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->next = skb->prev = NULL;
  34
  35                 /* tell the network layer not to perform IP checksums
  36                  * or to get the NIC to do it
  37                  */
  38                 skb->ip_summed = CHECKSUM_NONE;
  39         }
  40         return skb;
  41 }
  42
  43 static struct frame *
  44 getframe(struct aoedev *d, int tag)
  45 {
  46         struct frame *f, *e;
  47
  48         f = d->frames;
  49         e = f + d->nframes;
  50         for (; f<e; f++)
  51                 if (f->tag == tag)
  52                         return f;
  53         return NULL;
  54 }
  55
  56 /*
  57  * Leave the top bit clear so we have tagspace for userland.
  58  * The bottom 16 bits are the xmit tick for rexmit/rttavg processing.
  59  * This driver reserves tag -1 to mean "unused frame."
  60  */
  61 static int
  62 newtag(struct aoedev *d)
  63 {
  64         register ulong n;
  65
  66         n = jiffies & 0xffff;
  67         return n |= (++d->lasttag & 0x7fff) << 16;
  68 }
  69
  70 static int
  71 aoehdr_atainit(struct aoedev *d, struct aoe_hdr *h)
  72 {
  73         u32 host_tag = newtag(d);
  74
  75         memcpy(h->src, d->ifp->dev_addr, sizeof h->src);
  76         memcpy(h->dst, d->addr, sizeof h->dst);
  77         h->type = __constant_cpu_to_be16(ETH_P_AOE);
  78         h->verfl = AOE_HVER;
  79         h->major = cpu_to_be16(d->aoemajor);
  80         h->minor = d->aoeminor;
  81         h->cmd = AOECMD_ATA;
  82         h->tag = cpu_to_be32(host_tag);
  83
  84         return host_tag;
  85 }
  86
  87 static inline void
  88 put_lba(struct aoe_atahdr *ah, sector_t lba)
  89 {
  90         ah->lba0 = lba;
  91         ah->lba1 = lba >>= 8;
  92         ah->lba2 = lba >>= 8;
  93         ah->lba3 = lba >>= 8;
  94         ah->lba4 = lba >>= 8;
  95         ah->lba5 = lba >>= 8;
  96 }
  97
  98 static void
  99 aoecmd_ata_rw(struct aoedev *d, struct frame *f)
 100 {
 101         struct aoe_hdr *h;
 102         struct aoe_atahdr *ah;
 103         struct buf *buf;
 104         struct sk_buff *skb;
 105         ulong bcnt;
 106         register sector_t sector;
 107         char writebit, extbit;
 108
 109         writebit = 0x10;
 110         extbit = 0x4;
 111
 112         buf = d->inprocess;
 113
 114         sector = buf->sector;
 115         bcnt = buf->bv_resid;
 116         if (bcnt > d->maxbcnt)
 117                 bcnt = d->maxbcnt;
 118
 119         /* initialize the headers & frame */
 120         skb = f->skb;
 121         h = (struct aoe_hdr *) skb->mac.raw;
 122         ah = (struct aoe_atahdr *) (h+1);
 123         skb_put(skb, sizeof *h + sizeof *ah);
 124         memset(h, 0, skb->len);
 125         f->tag = aoehdr_atainit(d, h);
 126         f->waited = 0;
 127         f->buf = buf;
 128         f->bufaddr = buf->bufaddr;
 129         f->bcnt = bcnt;
 130         f->lba = sector;
 131
 132         /* set up ata header */
 133         ah->scnt = bcnt >> 9;
 134         put_lba(ah, sector);
 135         if (d->flags & DEVFL_EXT) {
 136                 ah->aflags |= AOEAFL_EXT;
 137         } else {
 138                 extbit = 0;
 139                 ah->lba3 &= 0x0f;
 140                 ah->lba3 |= 0xe0;       /* LBA bit + obsolete 0xa0 */
 141         }
 142
 143         if (bio_data_dir(buf->bio) == WRITE) {
 144                 skb_fill_page_desc(skb, 0, virt_to_page(f->bufaddr),
 145                         offset_in_page(f->bufaddr), bcnt);
 146                 ah->aflags |= AOEAFL_WRITE;
 147                 skb->len += bcnt;
 148                 skb->data_len = bcnt;
 149         } else {
 150                 writebit = 0;
 151         }
 152
 153         ah->cmdstat = WIN_READ | writebit | extbit;
 154
 155         /* mark all tracking fields and load out */
 156         buf->nframesout += 1;
 157         buf->bufaddr += bcnt;
 158         buf->bv_resid -= bcnt;
 159 /* printk(KERN_DEBUG "aoe: bv_resid=%ld\n", buf->bv_resid); */
 160         buf->resid -= bcnt;
 161         buf->sector += bcnt >> 9;
 162         if (buf->resid == 0) {
 163                 d->inprocess = NULL;
 164         } else if (buf->bv_resid == 0) {
 165                 buf->bv++;
 166                 WARN_ON(buf->bv->bv_len == 0);
 167                 buf->bv_resid = buf->bv->bv_len;
 168                 buf->bufaddr = page_address(buf->bv->bv_page) + buf->bv->bv_offset;
 169         }
 170
 171         skb->dev = d->ifp;
 172         skb = skb_clone(skb, GFP_ATOMIC);
 173         if (skb == NULL)
 174                 return;
 175         if (d->sendq_hd)
 176                 d->sendq_tl->next = skb;
 177         else
 178                 d->sendq_hd = skb;
 179         d->sendq_tl = skb;
 180 }
 181
 182 /* some callers cannot sleep, and they can call this function,
 183  * transmitting the packets later, when interrupts are on
 184  */
 185 static struct sk_buff *
 186 aoecmd_cfg_pkts(ushort aoemajor, unsigned char aoeminor, struct sk_buff **tail)
 187 {
 188         struct aoe_hdr *h;
 189         struct aoe_cfghdr *ch;
 190         struct sk_buff *skb, *sl, *sl_tail;
 191         struct net_device *ifp;
 192
 193         sl = sl_tail = NULL;
 194
 195         read_lock(&dev_base_lock);
 196         for (ifp = dev_base; ifp; dev_put(ifp), ifp = ifp->next) {
 197                 dev_hold(ifp);
 198                 if (!is_aoe_netif(ifp))
 199                         continue;
 200
 201                 skb = new_skb(sizeof *h + sizeof *ch);
 202                 if (skb == NULL) {
 203                         printk(KERN_INFO "aoe: skb alloc failure\n");
 204                         continue;
 205                 }
 206                 skb_put(skb, sizeof *h + sizeof *ch);
 207                 skb->dev = ifp;
 208                 if (sl_tail == NULL)
 209                         sl_tail = skb;
 210                 h = (struct aoe_hdr *) skb->mac.raw;
 211                 memset(h, 0, sizeof *h + sizeof *ch);
 212
 213                 memset(h->dst, 0xff, sizeof h->dst);
 214                 memcpy(h->src, ifp->dev_addr, sizeof h->src);
 215                 h->type = __constant_cpu_to_be16(ETH_P_AOE);
 216                 h->verfl = AOE_HVER;
 217                 h->major = cpu_to_be16(aoemajor);
 218                 h->minor = aoeminor;
 219                 h->cmd = AOECMD_CFG;
 220
 221                 skb->next = sl;
 222                 sl = skb;
 223         }
 224         read_unlock(&dev_base_lock);
 225
 226         if (tail != NULL)
 227                 *tail = sl_tail;
 228         return sl;
 229 }
 230
 231 static struct frame *
 232 freeframe(struct aoedev *d)
 233 {
 234         struct frame *f, *e;
 235         int n = 0;
 236
 237         f = d->frames;
 238         e = f + d->nframes;
 239         for (; f<e; f++) {
 240                 if (f->tag != FREETAG)
 241                         continue;
 242                 if (atomic_read(&skb_shinfo(f->skb)->dataref) == 1) {
 243                         skb_shinfo(f->skb)->nr_frags = f->skb->data_len = 0;
 244                         skb_trim(f->skb, 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 /*printk(KERN_DEBUG "aoe: 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                         printk(KERN_INFO "aoe: 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(struct work_struct *work)
 411 {
 412         struct aoedev *d = container_of(work, struct aoedev, work);
 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                 printk(KERN_INFO "aoe: %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                         printk(KERN_ERR "aoe: 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(get_unaligned(&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(get_unaligned(&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(get_unaligned(&hin->major)),
 554                         hin->minor,
 555                         be32_to_cpu(get_unaligned(&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                 printk(KERN_ERR
 572                         "aoe: ata error cmd=%2.2Xh stat=%2.2Xh from e%ld.%ld\n",
 573                         ahout->cmdstat, ahin->cmdstat,
 574                         d->aoemajor, d->aoeminor);
 575                 if (buf)
 576                         buf->flags |= BUFFL_FAIL;
 577         } else {
 578                 n = ahout->scnt << 9;
 579                 switch (ahout->cmdstat) {
 580                 case WIN_READ:
 581                 case WIN_READ_EXT:
 582                         if (skb->len - sizeof *hin - sizeof *ahin < n) {
 583                                 printk(KERN_ERR
 584                                         "aoe: runt data size in read.  skb->len=%d\n",
 585                                         skb->len);
 586                                 /* fail frame f?  just returning will rexmit. */
 587                                 spin_unlock_irqrestore(&d->lock, flags);
 588                                 return;
 589                         }
 590                         memcpy(f->bufaddr, ahin+1, n);
 591                 case WIN_WRITE:
 592                 case WIN_WRITE_EXT:
 593                         if (f->bcnt -= n) {
 594                                 skb = f->skb;
 595                                 f->bufaddr += n;
 596                                 put_lba(ahout, f->lba += ahout->scnt);
 597                                 n = f->bcnt;
 598                                 if (n > DEFAULTBCNT)
 599                                         n = DEFAULTBCNT;
 600                                 ahout->scnt = n >> 9;
 601                                 if (ahout->aflags & AOEAFL_WRITE) {
 602                                         skb_fill_page_desc(skb, 0,
 603                                                 virt_to_page(f->bufaddr),
 604                                                 offset_in_page(f->bufaddr), n);
 605                                         skb->len = sizeof *hout + sizeof *ahout + n;
 606                                         skb->data_len = n;
 607                                 }
 608                                 f->tag = newtag(d);
 609                                 hout->tag = cpu_to_be32(f->tag);
 610                                 skb->dev = d->ifp;
 611                                 skb = skb_clone(skb, GFP_ATOMIC);
 612                                 spin_unlock_irqrestore(&d->lock, flags);
 613                                 if (skb)
 614                                         aoenet_xmit(skb);
 615                                 return;
 616                         }
 617                         if (n > DEFAULTBCNT)
 618                                 d->lostjumbo = 0;
 619                         break;
 620                 case WIN_IDENTIFY:
 621                         if (skb->len - sizeof *hin - sizeof *ahin < 512) {
 622                                 printk(KERN_INFO
 623                                         "aoe: runt data size in ataid.  skb->len=%d\n",
 624                                         skb->len);
 625                                 spin_unlock_irqrestore(&d->lock, flags);
 626                                 return;
 627                         }
 628                         ataid_complete(d, (char *) (ahin+1));
 629                         break;
 630                 default:
 631                         printk(KERN_INFO
 632                                 "aoe: unrecognized ata command %2.2Xh for %d.%d\n",
 633                                 ahout->cmdstat,
 634                                 be16_to_cpu(get_unaligned(&hin->major)),
 635                                 hin->minor);
 636                 }
 637         }
 638
 639         if (buf) {
 640                 buf->nframesout -= 1;
 641                 if (buf->nframesout == 0 && buf->resid == 0) {
 642                         unsigned long duration = jiffies - buf->start_time;
 643                         unsigned long n_sect = buf->bio->bi_size >> 9;
 644                         struct gendisk *disk = d->gd;
 645                         const int rw = bio_data_dir(buf->bio);
 646
 647                         disk_stat_inc(disk, ios[rw]);
 648                         disk_stat_add(disk, ticks[rw], duration);
 649                         disk_stat_add(disk, sectors[rw], n_sect);
 650                         disk_stat_add(disk, io_ticks, duration);
 651                         n = (buf->flags & BUFFL_FAIL) ? -EIO : 0;
 652                         bio_endio(buf->bio, buf->bio->bi_size, n);
 653                         mempool_free(buf, d->bufpool);
 654                 }
 655         }
 656
 657         f->buf = NULL;
 658         f->tag = FREETAG;
 659
 660         aoecmd_work(d);
 661         sl = d->sendq_hd;
 662         d->sendq_hd = d->sendq_tl = NULL;
 663
 664         spin_unlock_irqrestore(&d->lock, flags);
 665         aoenet_xmit(sl);
 666 }
 667
 668 void
 669 aoecmd_cfg(ushort aoemajor, unsigned char aoeminor)
 670 {
 671         struct sk_buff *sl;
 672
 673         sl = aoecmd_cfg_pkts(aoemajor, aoeminor, NULL);
 674
 675         aoenet_xmit(sl);
 676 }
 677
 678 /*
 679  * Since we only call this in one place (and it only prepares one frame)
 680  * we just return the skb.  Usually we'd chain it up to the aoedev sendq.
 681  */
 682 static struct sk_buff *
 683 aoecmd_ata_id(struct aoedev *d)
 684 {
 685         struct aoe_hdr *h;
 686         struct aoe_atahdr *ah;
 687         struct frame *f;
 688         struct sk_buff *skb;
 689
 690         f = freeframe(d);
 691         if (f == NULL) {
 692                 printk(KERN_ERR "aoe: can't get a frame. This shouldn't happen.\n");
 693                 return NULL;
 694         }
 695
 696         /* initialize the headers & frame */
 697         skb = f->skb;
 698         h = (struct aoe_hdr *) skb->mac.raw;
 699         ah = (struct aoe_atahdr *) (h+1);
 700         skb_put(skb, sizeof *h + sizeof *ah);
 701         memset(h, 0, skb->len);
 702         f->tag = aoehdr_atainit(d, h);
 703         f->waited = 0;
 704
 705         /* set up ata header */
 706         ah->scnt = 1;
 707         ah->cmdstat = WIN_IDENTIFY;
 708         ah->lba3 = 0xa0;
 709
 710         skb->dev = d->ifp;
 711
 712         d->rttavg = MAXTIMER;
 713         d->timer.function = rexmit_timer;
 714
 715         return skb_clone(skb, GFP_ATOMIC);
 716 }
 717
 718 void
 719 aoecmd_cfg_rsp(struct sk_buff *skb)
 720 {
 721         struct aoedev *d;
 722         struct aoe_hdr *h;
 723         struct aoe_cfghdr *ch;
 724         ulong flags, sysminor, aoemajor;
 725         struct sk_buff *sl;
 726         enum { MAXFRAMES = 16 };
 727         u16 n;
 728
 729         h = (struct aoe_hdr *) skb->mac.raw;
 730         ch = (struct aoe_cfghdr *) (h+1);
 731
 732         /*
 733          * Enough people have their dip switches set backwards to
 734          * warrant a loud message for this special case.
 735          */
 736         aoemajor = be16_to_cpu(get_unaligned(&h->major));
 737         if (aoemajor == 0xfff) {
 738                 printk(KERN_ERR "aoe: Warning: shelf address is all ones.  "
 739                         "Check shelf dip switches.\n");
 740                 return;
 741         }
 742
 743         sysminor = SYSMINOR(aoemajor, h->minor);
 744         if (sysminor * AOE_PARTITIONS + AOE_PARTITIONS > MINORMASK) {
 745                 printk(KERN_INFO "aoe: e%ld.%d: minor number too large\n",
 746                         aoemajor, (int) h->minor);
 747                 return;
 748         }
 749
 750         n = be16_to_cpu(ch->bufcnt);
 751         if (n > MAXFRAMES)      /* keep it reasonable */
 752                 n = MAXFRAMES;
 753
 754         d = aoedev_by_sysminor_m(sysminor, n);
 755         if (d == NULL) {
 756                 printk(KERN_INFO "aoe: device sysminor_m failure\n");
 757                 return;
 758         }
 759
 760         spin_lock_irqsave(&d->lock, flags);
 761
 762         /* permit device to migrate mac and network interface */
 763         d->ifp = skb->dev;
 764         memcpy(d->addr, h->src, sizeof d->addr);
 765         if (!(d->flags & DEVFL_MAXBCNT)) {
 766                 n = d->ifp->mtu;
 767                 n -= sizeof (struct aoe_hdr) + sizeof (struct aoe_atahdr);
 768                 n /= 512;
 769                 if (n > ch->scnt)
 770                         n = ch->scnt;
 771                 n = n ? n * 512 : DEFAULTBCNT;
 772                 if (n != d->maxbcnt) {
 773                         printk(KERN_INFO
 774                                 "aoe: e%ld.%ld: setting %d byte data frames on %s\n",
 775                                 d->aoemajor, d->aoeminor, n, d->ifp->name);
 776                         d->maxbcnt = n;
 777                 }
 778         }
 779
 780         /* don't change users' perspective */
 781         if (d->nopen && !(d->flags & DEVFL_PAUSE)) {
 782                 spin_unlock_irqrestore(&d->lock, flags);
 783                 return;
 784         }
 785         d->flags |= DEVFL_PAUSE;        /* force pause */
 786         d->mintimer = MINTIMER;
 787         d->fw_ver = be16_to_cpu(ch->fwver);
 788
 789         /* check for already outstanding ataid */
 790         sl = aoedev_isbusy(d) == 0 ? aoecmd_ata_id(d) : NULL;
 791
 792         spin_unlock_irqrestore(&d->lock, flags);
 793
 794         aoenet_xmit(sl);
 795 }
 796