1 /* Copyright (c) 2013 Coraid, Inc. See COPYING for GPL terms. */ 2 /* 3 * aoecmd.c 4 * Filesystem request handling methods 5 */ 6 7 #include <linux/ata.h> 8 #include <linux/slab.h> 9 #include <linux/hdreg.h> 10 #include <linux/blkdev.h> 11 #include <linux/skbuff.h> 12 #include <linux/netdevice.h> 13 #include <linux/genhd.h> 14 #include <linux/moduleparam.h> 15 #include <linux/workqueue.h> 16 #include <linux/kthread.h> 17 #include <net/net_namespace.h> 18 #include <asm/unaligned.h> 19 #include <linux/uio.h> 20 #include "aoe.h" 21 22 #define MAXIOC (8192) /* default meant to avoid most soft lockups */ 23 24 static void ktcomplete(struct frame *, struct sk_buff *); 25 static int count_targets(struct aoedev *d, int *untainted); 26 27 static struct buf *nextbuf(struct aoedev *); 28 29 static int aoe_deadsecs = 60 * 3; 30 module_param(aoe_deadsecs, int, 0644); 31 MODULE_PARM_DESC(aoe_deadsecs, "After aoe_deadsecs seconds, give up and fail dev."); 32 33 static int aoe_maxout = 64; 34 module_param(aoe_maxout, int, 0644); 35 MODULE_PARM_DESC(aoe_maxout, 36 "Only aoe_maxout outstanding packets for every MAC on eX.Y."); 37 38 /* The number of online cpus during module initialization gives us a 39 * convenient heuristic cap on the parallelism used for ktio threads 40 * doing I/O completion. It is not important that the cap equal the 41 * actual number of running CPUs at any given time, but because of CPU 42 * hotplug, we take care to use ncpus instead of using 43 * num_online_cpus() after module initialization. 44 */ 45 static int ncpus; 46 47 /* mutex lock used for synchronization while thread spawning */ 48 static DEFINE_MUTEX(ktio_spawn_lock); 49 50 static wait_queue_head_t *ktiowq; 51 static struct ktstate *kts; 52 53 /* io completion queue */ 54 struct iocq_ktio { 55 struct list_head head; 56 spinlock_t lock; 57 }; 58 static struct iocq_ktio *iocq; 59 60 static struct page *empty_page; 61 62 static struct sk_buff * 63 new_skb(ulong len) 64 { 65 struct sk_buff *skb; 66 67 skb = alloc_skb(len + MAX_HEADER, GFP_ATOMIC); 68 if (skb) { 69 skb_reserve(skb, MAX_HEADER); 70 skb_reset_mac_header(skb); 71 skb_reset_network_header(skb); 72 skb->protocol = __constant_htons(ETH_P_AOE); 73 skb_checksum_none_assert(skb); 74 } 75 return skb; 76 } 77 78 static struct frame * 79 getframe_deferred(struct aoedev *d, u32 tag) 80 { 81 struct list_head *head, *pos, *nx; 82 struct frame *f; 83 84 head = &d->rexmitq; 85 list_for_each_safe(pos, nx, head) { 86 f = list_entry(pos, struct frame, head); 87 if (f->tag == tag) { 88 list_del(pos); 89 return f; 90 } 91 } 92 return NULL; 93 } 94 95 static struct frame * 96 getframe(struct aoedev *d, u32 tag) 97 { 98 struct frame *f; 99 struct list_head *head, *pos, *nx; 100 u32 n; 101 102 n = tag % NFACTIVE; 103 head = &d->factive[n]; 104 list_for_each_safe(pos, nx, head) { 105 f = list_entry(pos, struct frame, head); 106 if (f->tag == tag) { 107 list_del(pos); 108 return f; 109 } 110 } 111 return NULL; 112 } 113 114 /* 115 * Leave the top bit clear so we have tagspace for userland. 116 * The bottom 16 bits are the xmit tick for rexmit/rttavg processing. 117 * This driver reserves tag -1 to mean "unused frame." 118 */ 119 static int 120 newtag(struct aoedev *d) 121 { 122 register ulong n; 123 124 n = jiffies & 0xffff; 125 return n |= (++d->lasttag & 0x7fff) << 16; 126 } 127 128 static u32 129 aoehdr_atainit(struct aoedev *d, struct aoetgt *t, struct aoe_hdr *h) 130 { 131 u32 host_tag = newtag(d); 132 133 memcpy(h->src, t->ifp->nd->dev_addr, sizeof h->src); 134 memcpy(h->dst, t->addr, sizeof h->dst); 135 h->type = __constant_cpu_to_be16(ETH_P_AOE); 136 h->verfl = AOE_HVER; 137 h->major = cpu_to_be16(d->aoemajor); 138 h->minor = d->aoeminor; 139 h->cmd = AOECMD_ATA; 140 h->tag = cpu_to_be32(host_tag); 141 142 return host_tag; 143 } 144 145 static inline void 146 put_lba(struct aoe_atahdr *ah, sector_t lba) 147 { 148 ah->lba0 = lba; 149 ah->lba1 = lba >>= 8; 150 ah->lba2 = lba >>= 8; 151 ah->lba3 = lba >>= 8; 152 ah->lba4 = lba >>= 8; 153 ah->lba5 = lba >>= 8; 154 } 155 156 static struct aoeif * 157 ifrotate(struct aoetgt *t) 158 { 159 struct aoeif *ifp; 160 161 ifp = t->ifp; 162 ifp++; 163 if (ifp >= &t->ifs[NAOEIFS] || ifp->nd == NULL) 164 ifp = t->ifs; 165 if (ifp->nd == NULL) 166 return NULL; 167 return t->ifp = ifp; 168 } 169 170 static void 171 skb_pool_put(struct aoedev *d, struct sk_buff *skb) 172 { 173 __skb_queue_tail(&d->skbpool, skb); 174 } 175 176 static struct sk_buff * 177 skb_pool_get(struct aoedev *d) 178 { 179 struct sk_buff *skb = skb_peek(&d->skbpool); 180 181 if (skb && atomic_read(&skb_shinfo(skb)->dataref) == 1) { 182 __skb_unlink(skb, &d->skbpool); 183 return skb; 184 } 185 if (skb_queue_len(&d->skbpool) < NSKBPOOLMAX && 186 (skb = new_skb(ETH_ZLEN))) 187 return skb; 188 189 return NULL; 190 } 191 192 void 193 aoe_freetframe(struct frame *f) 194 { 195 struct aoetgt *t; 196 197 t = f->t; 198 f->buf = NULL; 199 memset(&f->iter, 0, sizeof(f->iter)); 200 f->r_skb = NULL; 201 f->flags = 0; 202 list_add(&f->head, &t->ffree); 203 } 204 205 static struct frame * 206 newtframe(struct aoedev *d, struct aoetgt *t) 207 { 208 struct frame *f; 209 struct sk_buff *skb; 210 struct list_head *pos; 211 212 if (list_empty(&t->ffree)) { 213 if (t->falloc >= NSKBPOOLMAX*2) 214 return NULL; 215 f = kcalloc(1, sizeof(*f), GFP_ATOMIC); 216 if (f == NULL) 217 return NULL; 218 t->falloc++; 219 f->t = t; 220 } else { 221 pos = t->ffree.next; 222 list_del(pos); 223 f = list_entry(pos, struct frame, head); 224 } 225 226 skb = f->skb; 227 if (skb == NULL) { 228 f->skb = skb = new_skb(ETH_ZLEN); 229 if (!skb) { 230 bail: aoe_freetframe(f); 231 return NULL; 232 } 233 } 234 235 if (atomic_read(&skb_shinfo(skb)->dataref) != 1) { 236 skb = skb_pool_get(d); 237 if (skb == NULL) 238 goto bail; 239 skb_pool_put(d, f->skb); 240 f->skb = skb; 241 } 242 243 skb->truesize -= skb->data_len; 244 skb_shinfo(skb)->nr_frags = skb->data_len = 0; 245 skb_trim(skb, 0); 246 return f; 247 } 248 249 static struct frame * 250 newframe(struct aoedev *d) 251 { 252 struct frame *f; 253 struct aoetgt *t, **tt; 254 int totout = 0; 255 int use_tainted; 256 int has_untainted; 257 258 if (!d->targets || !d->targets[0]) { 259 printk(KERN_ERR "aoe: NULL TARGETS!\n"); 260 return NULL; 261 } 262 tt = d->tgt; /* last used target */ 263 for (use_tainted = 0, has_untainted = 0;;) { 264 tt++; 265 if (tt >= &d->targets[d->ntargets] || !*tt) 266 tt = d->targets; 267 t = *tt; 268 if (!t->taint) { 269 has_untainted = 1; 270 totout += t->nout; 271 } 272 if (t->nout < t->maxout 273 && (use_tainted || !t->taint) 274 && t->ifp->nd) { 275 f = newtframe(d, t); 276 if (f) { 277 ifrotate(t); 278 d->tgt = tt; 279 return f; 280 } 281 } 282 if (tt == d->tgt) { /* we've looped and found nada */ 283 if (!use_tainted && !has_untainted) 284 use_tainted = 1; 285 else 286 break; 287 } 288 } 289 if (totout == 0) { 290 d->kicked++; 291 d->flags |= DEVFL_KICKME; 292 } 293 return NULL; 294 } 295 296 static void 297 skb_fillup(struct sk_buff *skb, struct bio *bio, struct bvec_iter iter) 298 { 299 int frag = 0; 300 struct bio_vec bv; 301 302 __bio_for_each_segment(bv, bio, iter, iter) 303 skb_fill_page_desc(skb, frag++, bv.bv_page, 304 bv.bv_offset, bv.bv_len); 305 } 306 307 static void 308 fhash(struct frame *f) 309 { 310 struct aoedev *d = f->t->d; 311 u32 n; 312 313 n = f->tag % NFACTIVE; 314 list_add_tail(&f->head, &d->factive[n]); 315 } 316 317 static void 318 ata_rw_frameinit(struct frame *f) 319 { 320 struct aoetgt *t; 321 struct aoe_hdr *h; 322 struct aoe_atahdr *ah; 323 struct sk_buff *skb; 324 char writebit, extbit; 325 326 skb = f->skb; 327 h = (struct aoe_hdr *) skb_mac_header(skb); 328 ah = (struct aoe_atahdr *) (h + 1); 329 skb_put(skb, sizeof(*h) + sizeof(*ah)); 330 memset(h, 0, skb->len); 331 332 writebit = 0x10; 333 extbit = 0x4; 334 335 t = f->t; 336 f->tag = aoehdr_atainit(t->d, t, h); 337 fhash(f); 338 t->nout++; 339 f->waited = 0; 340 f->waited_total = 0; 341 342 /* set up ata header */ 343 ah->scnt = f->iter.bi_size >> 9; 344 put_lba(ah, f->iter.bi_sector); 345 if (t->d->flags & DEVFL_EXT) { 346 ah->aflags |= AOEAFL_EXT; 347 } else { 348 extbit = 0; 349 ah->lba3 &= 0x0f; 350 ah->lba3 |= 0xe0; /* LBA bit + obsolete 0xa0 */ 351 } 352 if (f->buf && bio_data_dir(f->buf->bio) == WRITE) { 353 skb_fillup(skb, f->buf->bio, f->iter); 354 ah->aflags |= AOEAFL_WRITE; 355 skb->len += f->iter.bi_size; 356 skb->data_len = f->iter.bi_size; 357 skb->truesize += f->iter.bi_size; 358 t->wpkts++; 359 } else { 360 t->rpkts++; 361 writebit = 0; 362 } 363 364 ah->cmdstat = ATA_CMD_PIO_READ | writebit | extbit; 365 skb->dev = t->ifp->nd; 366 } 367 368 static int 369 aoecmd_ata_rw(struct aoedev *d) 370 { 371 struct frame *f; 372 struct buf *buf; 373 struct sk_buff *skb; 374 struct sk_buff_head queue; 375 376 buf = nextbuf(d); 377 if (buf == NULL) 378 return 0; 379 f = newframe(d); 380 if (f == NULL) 381 return 0; 382 383 /* initialize the headers & frame */ 384 f->buf = buf; 385 f->iter = buf->iter; 386 f->iter.bi_size = min_t(unsigned long, 387 d->maxbcnt ?: DEFAULTBCNT, 388 f->iter.bi_size); 389 bio_advance_iter(buf->bio, &buf->iter, f->iter.bi_size); 390 391 if (!buf->iter.bi_size) 392 d->ip.buf = NULL; 393 394 /* mark all tracking fields and load out */ 395 buf->nframesout += 1; 396 397 ata_rw_frameinit(f); 398 399 skb = skb_clone(f->skb, GFP_ATOMIC); 400 if (skb) { 401 f->sent = ktime_get(); 402 __skb_queue_head_init(&queue); 403 __skb_queue_tail(&queue, skb); 404 aoenet_xmit(&queue); 405 } 406 return 1; 407 } 408 409 /* some callers cannot sleep, and they can call this function, 410 * transmitting the packets later, when interrupts are on 411 */ 412 static void 413 aoecmd_cfg_pkts(ushort aoemajor, unsigned char aoeminor, struct sk_buff_head *queue) 414 { 415 struct aoe_hdr *h; 416 struct aoe_cfghdr *ch; 417 struct sk_buff *skb; 418 struct net_device *ifp; 419 420 rcu_read_lock(); 421 for_each_netdev_rcu(&init_net, ifp) { 422 dev_hold(ifp); 423 if (!is_aoe_netif(ifp)) 424 goto cont; 425 426 skb = new_skb(sizeof *h + sizeof *ch); 427 if (skb == NULL) { 428 printk(KERN_INFO "aoe: skb alloc failure\n"); 429 goto cont; 430 } 431 skb_put(skb, sizeof *h + sizeof *ch); 432 skb->dev = ifp; 433 __skb_queue_tail(queue, skb); 434 h = (struct aoe_hdr *) skb_mac_header(skb); 435 memset(h, 0, sizeof *h + sizeof *ch); 436 437 memset(h->dst, 0xff, sizeof h->dst); 438 memcpy(h->src, ifp->dev_addr, sizeof h->src); 439 h->type = __constant_cpu_to_be16(ETH_P_AOE); 440 h->verfl = AOE_HVER; 441 h->major = cpu_to_be16(aoemajor); 442 h->minor = aoeminor; 443 h->cmd = AOECMD_CFG; 444 445 cont: 446 dev_put(ifp); 447 } 448 rcu_read_unlock(); 449 } 450 451 static void 452 resend(struct aoedev *d, struct frame *f) 453 { 454 struct sk_buff *skb; 455 struct sk_buff_head queue; 456 struct aoe_hdr *h; 457 struct aoetgt *t; 458 char buf[128]; 459 u32 n; 460 461 t = f->t; 462 n = newtag(d); 463 skb = f->skb; 464 if (ifrotate(t) == NULL) { 465 /* probably can't happen, but set it up to fail anyway */ 466 pr_info("aoe: resend: no interfaces to rotate to.\n"); 467 ktcomplete(f, NULL); 468 return; 469 } 470 h = (struct aoe_hdr *) skb_mac_header(skb); 471 472 if (!(f->flags & FFL_PROBE)) { 473 snprintf(buf, sizeof(buf), 474 "%15s e%ld.%d oldtag=%08x@%08lx newtag=%08x s=%pm d=%pm nout=%d\n", 475 "retransmit", d->aoemajor, d->aoeminor, 476 f->tag, jiffies, n, 477 h->src, h->dst, t->nout); 478 aoechr_error(buf); 479 } 480 481 f->tag = n; 482 fhash(f); 483 h->tag = cpu_to_be32(n); 484 memcpy(h->dst, t->addr, sizeof h->dst); 485 memcpy(h->src, t->ifp->nd->dev_addr, sizeof h->src); 486 487 skb->dev = t->ifp->nd; 488 skb = skb_clone(skb, GFP_ATOMIC); 489 if (skb == NULL) 490 return; 491 f->sent = ktime_get(); 492 __skb_queue_head_init(&queue); 493 __skb_queue_tail(&queue, skb); 494 aoenet_xmit(&queue); 495 } 496 497 static int 498 tsince_hr(struct frame *f) 499 { 500 u64 delta = ktime_to_ns(ktime_sub(ktime_get(), f->sent)); 501 502 /* delta is normally under 4.2 seconds, avoid 64-bit division */ 503 if (likely(delta <= UINT_MAX)) 504 return (u32)delta / NSEC_PER_USEC; 505 506 /* avoid overflow after 71 minutes */ 507 if (delta > ((u64)INT_MAX * NSEC_PER_USEC)) 508 return INT_MAX; 509 510 return div_u64(delta, NSEC_PER_USEC); 511 } 512 513 static int 514 tsince(u32 tag) 515 { 516 int n; 517 518 n = jiffies & 0xffff; 519 n -= tag & 0xffff; 520 if (n < 0) 521 n += 1<<16; 522 return jiffies_to_usecs(n + 1); 523 } 524 525 static struct aoeif * 526 getif(struct aoetgt *t, struct net_device *nd) 527 { 528 struct aoeif *p, *e; 529 530 p = t->ifs; 531 e = p + NAOEIFS; 532 for (; p < e; p++) 533 if (p->nd == nd) 534 return p; 535 return NULL; 536 } 537 538 static void 539 ejectif(struct aoetgt *t, struct aoeif *ifp) 540 { 541 struct aoeif *e; 542 struct net_device *nd; 543 ulong n; 544 545 nd = ifp->nd; 546 e = t->ifs + NAOEIFS - 1; 547 n = (e - ifp) * sizeof *ifp; 548 memmove(ifp, ifp+1, n); 549 e->nd = NULL; 550 dev_put(nd); 551 } 552 553 static struct frame * 554 reassign_frame(struct frame *f) 555 { 556 struct frame *nf; 557 struct sk_buff *skb; 558 559 nf = newframe(f->t->d); 560 if (!nf) 561 return NULL; 562 if (nf->t == f->t) { 563 aoe_freetframe(nf); 564 return NULL; 565 } 566 567 skb = nf->skb; 568 nf->skb = f->skb; 569 nf->buf = f->buf; 570 nf->iter = f->iter; 571 nf->waited = 0; 572 nf->waited_total = f->waited_total; 573 nf->sent = f->sent; 574 f->skb = skb; 575 576 return nf; 577 } 578 579 static void 580 probe(struct aoetgt *t) 581 { 582 struct aoedev *d; 583 struct frame *f; 584 struct sk_buff *skb; 585 struct sk_buff_head queue; 586 size_t n, m; 587 int frag; 588 589 d = t->d; 590 f = newtframe(d, t); 591 if (!f) { 592 pr_err("%s %pm for e%ld.%d: %s\n", 593 "aoe: cannot probe remote address", 594 t->addr, 595 (long) d->aoemajor, d->aoeminor, 596 "no frame available"); 597 return; 598 } 599 f->flags |= FFL_PROBE; 600 ifrotate(t); 601 f->iter.bi_size = t->d->maxbcnt ? t->d->maxbcnt : DEFAULTBCNT; 602 ata_rw_frameinit(f); 603 skb = f->skb; 604 for (frag = 0, n = f->iter.bi_size; n > 0; ++frag, n -= m) { 605 if (n < PAGE_SIZE) 606 m = n; 607 else 608 m = PAGE_SIZE; 609 skb_fill_page_desc(skb, frag, empty_page, 0, m); 610 } 611 skb->len += f->iter.bi_size; 612 skb->data_len = f->iter.bi_size; 613 skb->truesize += f->iter.bi_size; 614 615 skb = skb_clone(f->skb, GFP_ATOMIC); 616 if (skb) { 617 f->sent = ktime_get(); 618 __skb_queue_head_init(&queue); 619 __skb_queue_tail(&queue, skb); 620 aoenet_xmit(&queue); 621 } 622 } 623 624 static long 625 rto(struct aoedev *d) 626 { 627 long t; 628 629 t = 2 * d->rttavg >> RTTSCALE; 630 t += 8 * d->rttdev >> RTTDSCALE; 631 if (t == 0) 632 t = 1; 633 634 return t; 635 } 636 637 static void 638 rexmit_deferred(struct aoedev *d) 639 { 640 struct aoetgt *t; 641 struct frame *f; 642 struct frame *nf; 643 struct list_head *pos, *nx, *head; 644 int since; 645 int untainted; 646 647 count_targets(d, &untainted); 648 649 head = &d->rexmitq; 650 list_for_each_safe(pos, nx, head) { 651 f = list_entry(pos, struct frame, head); 652 t = f->t; 653 if (t->taint) { 654 if (!(f->flags & FFL_PROBE)) { 655 nf = reassign_frame(f); 656 if (nf) { 657 if (t->nout_probes == 0 658 && untainted > 0) { 659 probe(t); 660 t->nout_probes++; 661 } 662 list_replace(&f->head, &nf->head); 663 pos = &nf->head; 664 aoe_freetframe(f); 665 f = nf; 666 t = f->t; 667 } 668 } else if (untainted < 1) { 669 /* don't probe w/o other untainted aoetgts */ 670 goto stop_probe; 671 } else if (tsince_hr(f) < t->taint * rto(d)) { 672 /* reprobe slowly when taint is high */ 673 continue; 674 } 675 } else if (f->flags & FFL_PROBE) { 676 stop_probe: /* don't probe untainted aoetgts */ 677 list_del(pos); 678 aoe_freetframe(f); 679 /* leaving d->kicked, because this is routine */ 680 f->t->d->flags |= DEVFL_KICKME; 681 continue; 682 } 683 if (t->nout >= t->maxout) 684 continue; 685 list_del(pos); 686 t->nout++; 687 if (f->flags & FFL_PROBE) 688 t->nout_probes++; 689 since = tsince_hr(f); 690 f->waited += since; 691 f->waited_total += since; 692 resend(d, f); 693 } 694 } 695 696 /* An aoetgt accumulates demerits quickly, and successful 697 * probing redeems the aoetgt slowly. 698 */ 699 static void 700 scorn(struct aoetgt *t) 701 { 702 int n; 703 704 n = t->taint++; 705 t->taint += t->taint * 2; 706 if (n > t->taint) 707 t->taint = n; 708 if (t->taint > MAX_TAINT) 709 t->taint = MAX_TAINT; 710 } 711 712 static int 713 count_targets(struct aoedev *d, int *untainted) 714 { 715 int i, good; 716 717 for (i = good = 0; i < d->ntargets && d->targets[i]; ++i) 718 if (d->targets[i]->taint == 0) 719 good++; 720 721 if (untainted) 722 *untainted = good; 723 return i; 724 } 725 726 static void 727 rexmit_timer(struct timer_list *timer) 728 { 729 struct aoedev *d; 730 struct aoetgt *t; 731 struct aoeif *ifp; 732 struct frame *f; 733 struct list_head *head, *pos, *nx; 734 LIST_HEAD(flist); 735 register long timeout; 736 ulong flags, n; 737 int i; 738 int utgts; /* number of aoetgt descriptors (not slots) */ 739 int since; 740 741 d = from_timer(d, timer, timer); 742 743 spin_lock_irqsave(&d->lock, flags); 744 745 /* timeout based on observed timings and variations */ 746 timeout = rto(d); 747 748 utgts = count_targets(d, NULL); 749 750 if (d->flags & DEVFL_TKILL) { 751 spin_unlock_irqrestore(&d->lock, flags); 752 return; 753 } 754 755 /* collect all frames to rexmit into flist */ 756 for (i = 0; i < NFACTIVE; i++) { 757 head = &d->factive[i]; 758 list_for_each_safe(pos, nx, head) { 759 f = list_entry(pos, struct frame, head); 760 if (tsince_hr(f) < timeout) 761 break; /* end of expired frames */ 762 /* move to flist for later processing */ 763 list_move_tail(pos, &flist); 764 } 765 } 766 767 /* process expired frames */ 768 while (!list_empty(&flist)) { 769 pos = flist.next; 770 f = list_entry(pos, struct frame, head); 771 since = tsince_hr(f); 772 n = f->waited_total + since; 773 n /= USEC_PER_SEC; 774 if (aoe_deadsecs 775 && n > aoe_deadsecs 776 && !(f->flags & FFL_PROBE)) { 777 /* Waited too long. Device failure. 778 * Hang all frames on first hash bucket for downdev 779 * to clean up. 780 */ 781 list_splice(&flist, &d->factive[0]); 782 aoedev_downdev(d); 783 goto out; 784 } 785 786 t = f->t; 787 n = f->waited + since; 788 n /= USEC_PER_SEC; 789 if (aoe_deadsecs && utgts > 0 790 && (n > aoe_deadsecs / utgts || n > HARD_SCORN_SECS)) 791 scorn(t); /* avoid this target */ 792 793 if (t->maxout != 1) { 794 t->ssthresh = t->maxout / 2; 795 t->maxout = 1; 796 } 797 798 if (f->flags & FFL_PROBE) { 799 t->nout_probes--; 800 } else { 801 ifp = getif(t, f->skb->dev); 802 if (ifp && ++ifp->lost > (t->nframes << 1) 803 && (ifp != t->ifs || t->ifs[1].nd)) { 804 ejectif(t, ifp); 805 ifp = NULL; 806 } 807 } 808 list_move_tail(pos, &d->rexmitq); 809 t->nout--; 810 } 811 rexmit_deferred(d); 812 813 out: 814 if ((d->flags & DEVFL_KICKME) && d->blkq) { 815 d->flags &= ~DEVFL_KICKME; 816 d->blkq->request_fn(d->blkq); 817 } 818 819 d->timer.expires = jiffies + TIMERTICK; 820 add_timer(&d->timer); 821 822 spin_unlock_irqrestore(&d->lock, flags); 823 } 824 825 static unsigned long 826 rqbiocnt(struct request *r) 827 { 828 struct bio *bio; 829 unsigned long n = 0; 830 831 __rq_for_each_bio(bio, r) 832 n++; 833 return n; 834 } 835 836 static void 837 bufinit(struct buf *buf, struct request *rq, struct bio *bio) 838 { 839 memset(buf, 0, sizeof(*buf)); 840 buf->rq = rq; 841 buf->bio = bio; 842 buf->iter = bio->bi_iter; 843 } 844 845 static struct buf * 846 nextbuf(struct aoedev *d) 847 { 848 struct request *rq; 849 struct request_queue *q; 850 struct buf *buf; 851 struct bio *bio; 852 853 q = d->blkq; 854 if (q == NULL) 855 return NULL; /* initializing */ 856 if (d->ip.buf) 857 return d->ip.buf; 858 rq = d->ip.rq; 859 if (rq == NULL) { 860 rq = blk_peek_request(q); 861 if (rq == NULL) 862 return NULL; 863 blk_start_request(rq); 864 d->ip.rq = rq; 865 d->ip.nxbio = rq->bio; 866 rq->special = (void *) rqbiocnt(rq); 867 } 868 buf = mempool_alloc(d->bufpool, GFP_ATOMIC); 869 if (buf == NULL) { 870 pr_err("aoe: nextbuf: unable to mempool_alloc!\n"); 871 return NULL; 872 } 873 bio = d->ip.nxbio; 874 bufinit(buf, rq, bio); 875 bio = bio->bi_next; 876 d->ip.nxbio = bio; 877 if (bio == NULL) 878 d->ip.rq = NULL; 879 return d->ip.buf = buf; 880 } 881 882 /* enters with d->lock held */ 883 void 884 aoecmd_work(struct aoedev *d) 885 { 886 rexmit_deferred(d); 887 while (aoecmd_ata_rw(d)) 888 ; 889 } 890 891 /* this function performs work that has been deferred until sleeping is OK 892 */ 893 void 894 aoecmd_sleepwork(struct work_struct *work) 895 { 896 struct aoedev *d = container_of(work, struct aoedev, work); 897 struct block_device *bd; 898 u64 ssize; 899 900 if (d->flags & DEVFL_GDALLOC) 901 aoeblk_gdalloc(d); 902 903 if (d->flags & DEVFL_NEWSIZE) { 904 ssize = get_capacity(d->gd); 905 bd = bdget_disk(d->gd, 0); 906 if (bd) { 907 inode_lock(bd->bd_inode); 908 i_size_write(bd->bd_inode, (loff_t)ssize<<9); 909 inode_unlock(bd->bd_inode); 910 bdput(bd); 911 } 912 spin_lock_irq(&d->lock); 913 d->flags |= DEVFL_UP; 914 d->flags &= ~DEVFL_NEWSIZE; 915 spin_unlock_irq(&d->lock); 916 } 917 } 918 919 static void 920 ata_ident_fixstring(u16 *id, int ns) 921 { 922 u16 s; 923 924 while (ns-- > 0) { 925 s = *id; 926 *id++ = s >> 8 | s << 8; 927 } 928 } 929 930 static void 931 ataid_complete(struct aoedev *d, struct aoetgt *t, unsigned char *id) 932 { 933 u64 ssize; 934 u16 n; 935 936 /* word 83: command set supported */ 937 n = get_unaligned_le16(&id[83 << 1]); 938 939 /* word 86: command set/feature enabled */ 940 n |= get_unaligned_le16(&id[86 << 1]); 941 942 if (n & (1<<10)) { /* bit 10: LBA 48 */ 943 d->flags |= DEVFL_EXT; 944 945 /* word 100: number lba48 sectors */ 946 ssize = get_unaligned_le64(&id[100 << 1]); 947 948 /* set as in ide-disk.c:init_idedisk_capacity */ 949 d->geo.cylinders = ssize; 950 d->geo.cylinders /= (255 * 63); 951 d->geo.heads = 255; 952 d->geo.sectors = 63; 953 } else { 954 d->flags &= ~DEVFL_EXT; 955 956 /* number lba28 sectors */ 957 ssize = get_unaligned_le32(&id[60 << 1]); 958 959 /* NOTE: obsolete in ATA 6 */ 960 d->geo.cylinders = get_unaligned_le16(&id[54 << 1]); 961 d->geo.heads = get_unaligned_le16(&id[55 << 1]); 962 d->geo.sectors = get_unaligned_le16(&id[56 << 1]); 963 } 964 965 ata_ident_fixstring((u16 *) &id[10<<1], 10); /* serial */ 966 ata_ident_fixstring((u16 *) &id[23<<1], 4); /* firmware */ 967 ata_ident_fixstring((u16 *) &id[27<<1], 20); /* model */ 968 memcpy(d->ident, id, sizeof(d->ident)); 969 970 if (d->ssize != ssize) 971 printk(KERN_INFO 972 "aoe: %pm e%ld.%d v%04x has %llu sectors\n", 973 t->addr, 974 d->aoemajor, d->aoeminor, 975 d->fw_ver, (long long)ssize); 976 d->ssize = ssize; 977 d->geo.start = 0; 978 if (d->flags & (DEVFL_GDALLOC|DEVFL_NEWSIZE)) 979 return; 980 if (d->gd != NULL) { 981 set_capacity(d->gd, ssize); 982 d->flags |= DEVFL_NEWSIZE; 983 } else 984 d->flags |= DEVFL_GDALLOC; 985 schedule_work(&d->work); 986 } 987 988 static void 989 calc_rttavg(struct aoedev *d, struct aoetgt *t, int rtt) 990 { 991 register long n; 992 993 n = rtt; 994 995 /* cf. Congestion Avoidance and Control, Jacobson & Karels, 1988 */ 996 n -= d->rttavg >> RTTSCALE; 997 d->rttavg += n; 998 if (n < 0) 999 n = -n; 1000 n -= d->rttdev >> RTTDSCALE; 1001 d->rttdev += n; 1002 1003 if (!t || t->maxout >= t->nframes) 1004 return; 1005 if (t->maxout < t->ssthresh) 1006 t->maxout += 1; 1007 else if (t->nout == t->maxout && t->next_cwnd-- == 0) { 1008 t->maxout += 1; 1009 t->next_cwnd = t->maxout; 1010 } 1011 } 1012 1013 static struct aoetgt * 1014 gettgt(struct aoedev *d, char *addr) 1015 { 1016 struct aoetgt **t, **e; 1017 1018 t = d->targets; 1019 e = t + d->ntargets; 1020 for (; t < e && *t; t++) 1021 if (memcmp((*t)->addr, addr, sizeof((*t)->addr)) == 0) 1022 return *t; 1023 return NULL; 1024 } 1025 1026 static void 1027 bvcpy(struct sk_buff *skb, struct bio *bio, struct bvec_iter iter, long cnt) 1028 { 1029 int soff = 0; 1030 struct bio_vec bv; 1031 1032 iter.bi_size = cnt; 1033 1034 __bio_for_each_segment(bv, bio, iter, iter) { 1035 char *p = kmap_atomic(bv.bv_page) + bv.bv_offset; 1036 skb_copy_bits(skb, soff, p, bv.bv_len); 1037 kunmap_atomic(p); 1038 soff += bv.bv_len; 1039 } 1040 } 1041 1042 void 1043 aoe_end_request(struct aoedev *d, struct request *rq, int fastfail) 1044 { 1045 struct bio *bio; 1046 int bok; 1047 struct request_queue *q; 1048 1049 q = d->blkq; 1050 if (rq == d->ip.rq) 1051 d->ip.rq = NULL; 1052 do { 1053 bio = rq->bio; 1054 bok = !fastfail && !bio->bi_status; 1055 } while (__blk_end_request(rq, bok ? BLK_STS_OK : BLK_STS_IOERR, bio->bi_iter.bi_size)); 1056 1057 /* cf. http://lkml.org/lkml/2006/10/31/28 */ 1058 if (!fastfail) 1059 __blk_run_queue(q); 1060 } 1061 1062 static void 1063 aoe_end_buf(struct aoedev *d, struct buf *buf) 1064 { 1065 struct request *rq; 1066 unsigned long n; 1067 1068 if (buf == d->ip.buf) 1069 d->ip.buf = NULL; 1070 rq = buf->rq; 1071 mempool_free(buf, d->bufpool); 1072 n = (unsigned long) rq->special; 1073 rq->special = (void *) --n; 1074 if (n == 0) 1075 aoe_end_request(d, rq, 0); 1076 } 1077 1078 static void 1079 ktiocomplete(struct frame *f) 1080 { 1081 struct aoe_hdr *hin, *hout; 1082 struct aoe_atahdr *ahin, *ahout; 1083 struct buf *buf; 1084 struct sk_buff *skb; 1085 struct aoetgt *t; 1086 struct aoeif *ifp; 1087 struct aoedev *d; 1088 long n; 1089 int untainted; 1090 1091 if (f == NULL) 1092 return; 1093 1094 t = f->t; 1095 d = t->d; 1096 skb = f->r_skb; 1097 buf = f->buf; 1098 if (f->flags & FFL_PROBE) 1099 goto out; 1100 if (!skb) /* just fail the buf. */ 1101 goto noskb; 1102 1103 hout = (struct aoe_hdr *) skb_mac_header(f->skb); 1104 ahout = (struct aoe_atahdr *) (hout+1); 1105 1106 hin = (struct aoe_hdr *) skb->data; 1107 skb_pull(skb, sizeof(*hin)); 1108 ahin = (struct aoe_atahdr *) skb->data; 1109 skb_pull(skb, sizeof(*ahin)); 1110 if (ahin->cmdstat & 0xa9) { /* these bits cleared on success */ 1111 pr_err("aoe: ata error cmd=%2.2Xh stat=%2.2Xh from e%ld.%d\n", 1112 ahout->cmdstat, ahin->cmdstat, 1113 d->aoemajor, d->aoeminor); 1114 noskb: if (buf) 1115 buf->bio->bi_status = BLK_STS_IOERR; 1116 goto out; 1117 } 1118 1119 n = ahout->scnt << 9; 1120 switch (ahout->cmdstat) { 1121 case ATA_CMD_PIO_READ: 1122 case ATA_CMD_PIO_READ_EXT: 1123 if (skb->len < n) { 1124 pr_err("%s e%ld.%d. skb->len=%d need=%ld\n", 1125 "aoe: runt data size in read from", 1126 (long) d->aoemajor, d->aoeminor, 1127 skb->len, n); 1128 buf->bio->bi_status = BLK_STS_IOERR; 1129 break; 1130 } 1131 if (n > f->iter.bi_size) { 1132 pr_err_ratelimited("%s e%ld.%d. bytes=%ld need=%u\n", 1133 "aoe: too-large data size in read from", 1134 (long) d->aoemajor, d->aoeminor, 1135 n, f->iter.bi_size); 1136 buf->bio->bi_status = BLK_STS_IOERR; 1137 break; 1138 } 1139 bvcpy(skb, f->buf->bio, f->iter, n); 1140 /* fall through */ 1141 case ATA_CMD_PIO_WRITE: 1142 case ATA_CMD_PIO_WRITE_EXT: 1143 spin_lock_irq(&d->lock); 1144 ifp = getif(t, skb->dev); 1145 if (ifp) 1146 ifp->lost = 0; 1147 spin_unlock_irq(&d->lock); 1148 break; 1149 case ATA_CMD_ID_ATA: 1150 if (skb->len < 512) { 1151 pr_info("%s e%ld.%d. skb->len=%d need=512\n", 1152 "aoe: runt data size in ataid from", 1153 (long) d->aoemajor, d->aoeminor, 1154 skb->len); 1155 break; 1156 } 1157 if (skb_linearize(skb)) 1158 break; 1159 spin_lock_irq(&d->lock); 1160 ataid_complete(d, t, skb->data); 1161 spin_unlock_irq(&d->lock); 1162 break; 1163 default: 1164 pr_info("aoe: unrecognized ata command %2.2Xh for %d.%d\n", 1165 ahout->cmdstat, 1166 be16_to_cpu(get_unaligned(&hin->major)), 1167 hin->minor); 1168 } 1169 out: 1170 spin_lock_irq(&d->lock); 1171 if (t->taint > 0 1172 && --t->taint > 0 1173 && t->nout_probes == 0) { 1174 count_targets(d, &untainted); 1175 if (untainted > 0) { 1176 probe(t); 1177 t->nout_probes++; 1178 } 1179 } 1180 1181 aoe_freetframe(f); 1182 1183 if (buf && --buf->nframesout == 0 && buf->iter.bi_size == 0) 1184 aoe_end_buf(d, buf); 1185 1186 spin_unlock_irq(&d->lock); 1187 aoedev_put(d); 1188 dev_kfree_skb(skb); 1189 } 1190 1191 /* Enters with iocq.lock held. 1192 * Returns true iff responses needing processing remain. 1193 */ 1194 static int 1195 ktio(int id) 1196 { 1197 struct frame *f; 1198 struct list_head *pos; 1199 int i; 1200 int actual_id; 1201 1202 for (i = 0; ; ++i) { 1203 if (i == MAXIOC) 1204 return 1; 1205 if (list_empty(&iocq[id].head)) 1206 return 0; 1207 pos = iocq[id].head.next; 1208 list_del(pos); 1209 f = list_entry(pos, struct frame, head); 1210 spin_unlock_irq(&iocq[id].lock); 1211 ktiocomplete(f); 1212 1213 /* Figure out if extra threads are required. */ 1214 actual_id = f->t->d->aoeminor % ncpus; 1215 1216 if (!kts[actual_id].active) { 1217 BUG_ON(id != 0); 1218 mutex_lock(&ktio_spawn_lock); 1219 if (!kts[actual_id].active 1220 && aoe_ktstart(&kts[actual_id]) == 0) 1221 kts[actual_id].active = 1; 1222 mutex_unlock(&ktio_spawn_lock); 1223 } 1224 spin_lock_irq(&iocq[id].lock); 1225 } 1226 } 1227 1228 static int 1229 kthread(void *vp) 1230 { 1231 struct ktstate *k; 1232 DECLARE_WAITQUEUE(wait, current); 1233 int more; 1234 1235 k = vp; 1236 current->flags |= PF_NOFREEZE; 1237 set_user_nice(current, -10); 1238 complete(&k->rendez); /* tell spawner we're running */ 1239 do { 1240 spin_lock_irq(k->lock); 1241 more = k->fn(k->id); 1242 if (!more) { 1243 add_wait_queue(k->waitq, &wait); 1244 __set_current_state(TASK_INTERRUPTIBLE); 1245 } 1246 spin_unlock_irq(k->lock); 1247 if (!more) { 1248 schedule(); 1249 remove_wait_queue(k->waitq, &wait); 1250 } else 1251 cond_resched(); 1252 } while (!kthread_should_stop()); 1253 complete(&k->rendez); /* tell spawner we're stopping */ 1254 return 0; 1255 } 1256 1257 void 1258 aoe_ktstop(struct ktstate *k) 1259 { 1260 kthread_stop(k->task); 1261 wait_for_completion(&k->rendez); 1262 } 1263 1264 int 1265 aoe_ktstart(struct ktstate *k) 1266 { 1267 struct task_struct *task; 1268 1269 init_completion(&k->rendez); 1270 task = kthread_run(kthread, k, "%s", k->name); 1271 if (task == NULL || IS_ERR(task)) 1272 return -ENOMEM; 1273 k->task = task; 1274 wait_for_completion(&k->rendez); /* allow kthread to start */ 1275 init_completion(&k->rendez); /* for waiting for exit later */ 1276 return 0; 1277 } 1278 1279 /* pass it off to kthreads for processing */ 1280 static void 1281 ktcomplete(struct frame *f, struct sk_buff *skb) 1282 { 1283 int id; 1284 ulong flags; 1285 1286 f->r_skb = skb; 1287 id = f->t->d->aoeminor % ncpus; 1288 spin_lock_irqsave(&iocq[id].lock, flags); 1289 if (!kts[id].active) { 1290 spin_unlock_irqrestore(&iocq[id].lock, flags); 1291 /* The thread with id has not been spawned yet, 1292 * so delegate the work to the main thread and 1293 * try spawning a new thread. 1294 */ 1295 id = 0; 1296 spin_lock_irqsave(&iocq[id].lock, flags); 1297 } 1298 list_add_tail(&f->head, &iocq[id].head); 1299 spin_unlock_irqrestore(&iocq[id].lock, flags); 1300 wake_up(&ktiowq[id]); 1301 } 1302 1303 struct sk_buff * 1304 aoecmd_ata_rsp(struct sk_buff *skb) 1305 { 1306 struct aoedev *d; 1307 struct aoe_hdr *h; 1308 struct frame *f; 1309 u32 n; 1310 ulong flags; 1311 char ebuf[128]; 1312 u16 aoemajor; 1313 1314 h = (struct aoe_hdr *) skb->data; 1315 aoemajor = be16_to_cpu(get_unaligned(&h->major)); 1316 d = aoedev_by_aoeaddr(aoemajor, h->minor, 0); 1317 if (d == NULL) { 1318 snprintf(ebuf, sizeof ebuf, "aoecmd_ata_rsp: ata response " 1319 "for unknown device %d.%d\n", 1320 aoemajor, h->minor); 1321 aoechr_error(ebuf); 1322 return skb; 1323 } 1324 1325 spin_lock_irqsave(&d->lock, flags); 1326 1327 n = be32_to_cpu(get_unaligned(&h->tag)); 1328 f = getframe(d, n); 1329 if (f) { 1330 calc_rttavg(d, f->t, tsince_hr(f)); 1331 f->t->nout--; 1332 if (f->flags & FFL_PROBE) 1333 f->t->nout_probes--; 1334 } else { 1335 f = getframe_deferred(d, n); 1336 if (f) { 1337 calc_rttavg(d, NULL, tsince_hr(f)); 1338 } else { 1339 calc_rttavg(d, NULL, tsince(n)); 1340 spin_unlock_irqrestore(&d->lock, flags); 1341 aoedev_put(d); 1342 snprintf(ebuf, sizeof(ebuf), 1343 "%15s e%d.%d tag=%08x@%08lx s=%pm d=%pm\n", 1344 "unexpected rsp", 1345 get_unaligned_be16(&h->major), 1346 h->minor, 1347 get_unaligned_be32(&h->tag), 1348 jiffies, 1349 h->src, 1350 h->dst); 1351 aoechr_error(ebuf); 1352 return skb; 1353 } 1354 } 1355 aoecmd_work(d); 1356 1357 spin_unlock_irqrestore(&d->lock, flags); 1358 1359 ktcomplete(f, skb); 1360 1361 /* 1362 * Note here that we do not perform an aoedev_put, as we are 1363 * leaving this reference for the ktio to release. 1364 */ 1365 return NULL; 1366 } 1367 1368 void 1369 aoecmd_cfg(ushort aoemajor, unsigned char aoeminor) 1370 { 1371 struct sk_buff_head queue; 1372 1373 __skb_queue_head_init(&queue); 1374 aoecmd_cfg_pkts(aoemajor, aoeminor, &queue); 1375 aoenet_xmit(&queue); 1376 } 1377 1378 struct sk_buff * 1379 aoecmd_ata_id(struct aoedev *d) 1380 { 1381 struct aoe_hdr *h; 1382 struct aoe_atahdr *ah; 1383 struct frame *f; 1384 struct sk_buff *skb; 1385 struct aoetgt *t; 1386 1387 f = newframe(d); 1388 if (f == NULL) 1389 return NULL; 1390 1391 t = *d->tgt; 1392 1393 /* initialize the headers & frame */ 1394 skb = f->skb; 1395 h = (struct aoe_hdr *) skb_mac_header(skb); 1396 ah = (struct aoe_atahdr *) (h+1); 1397 skb_put(skb, sizeof *h + sizeof *ah); 1398 memset(h, 0, skb->len); 1399 f->tag = aoehdr_atainit(d, t, h); 1400 fhash(f); 1401 t->nout++; 1402 f->waited = 0; 1403 f->waited_total = 0; 1404 1405 /* set up ata header */ 1406 ah->scnt = 1; 1407 ah->cmdstat = ATA_CMD_ID_ATA; 1408 ah->lba3 = 0xa0; 1409 1410 skb->dev = t->ifp->nd; 1411 1412 d->rttavg = RTTAVG_INIT; 1413 d->rttdev = RTTDEV_INIT; 1414 d->timer.function = rexmit_timer; 1415 1416 skb = skb_clone(skb, GFP_ATOMIC); 1417 if (skb) 1418 f->sent = ktime_get(); 1419 1420 return skb; 1421 } 1422 1423 static struct aoetgt ** 1424 grow_targets(struct aoedev *d) 1425 { 1426 ulong oldn, newn; 1427 struct aoetgt **tt; 1428 1429 oldn = d->ntargets; 1430 newn = oldn * 2; 1431 tt = kcalloc(newn, sizeof(*d->targets), GFP_ATOMIC); 1432 if (!tt) 1433 return NULL; 1434 memmove(tt, d->targets, sizeof(*d->targets) * oldn); 1435 d->tgt = tt + (d->tgt - d->targets); 1436 kfree(d->targets); 1437 d->targets = tt; 1438 d->ntargets = newn; 1439 1440 return &d->targets[oldn]; 1441 } 1442 1443 static struct aoetgt * 1444 addtgt(struct aoedev *d, char *addr, ulong nframes) 1445 { 1446 struct aoetgt *t, **tt, **te; 1447 1448 tt = d->targets; 1449 te = tt + d->ntargets; 1450 for (; tt < te && *tt; tt++) 1451 ; 1452 1453 if (tt == te) { 1454 tt = grow_targets(d); 1455 if (!tt) 1456 goto nomem; 1457 } 1458 t = kzalloc(sizeof(*t), GFP_ATOMIC); 1459 if (!t) 1460 goto nomem; 1461 t->nframes = nframes; 1462 t->d = d; 1463 memcpy(t->addr, addr, sizeof t->addr); 1464 t->ifp = t->ifs; 1465 aoecmd_wreset(t); 1466 t->maxout = t->nframes / 2; 1467 INIT_LIST_HEAD(&t->ffree); 1468 return *tt = t; 1469 1470 nomem: 1471 pr_info("aoe: cannot allocate memory to add target\n"); 1472 return NULL; 1473 } 1474 1475 static void 1476 setdbcnt(struct aoedev *d) 1477 { 1478 struct aoetgt **t, **e; 1479 int bcnt = 0; 1480 1481 t = d->targets; 1482 e = t + d->ntargets; 1483 for (; t < e && *t; t++) 1484 if (bcnt == 0 || bcnt > (*t)->minbcnt) 1485 bcnt = (*t)->minbcnt; 1486 if (bcnt != d->maxbcnt) { 1487 d->maxbcnt = bcnt; 1488 pr_info("aoe: e%ld.%d: setting %d byte data frames\n", 1489 d->aoemajor, d->aoeminor, bcnt); 1490 } 1491 } 1492 1493 static void 1494 setifbcnt(struct aoetgt *t, struct net_device *nd, int bcnt) 1495 { 1496 struct aoedev *d; 1497 struct aoeif *p, *e; 1498 int minbcnt; 1499 1500 d = t->d; 1501 minbcnt = bcnt; 1502 p = t->ifs; 1503 e = p + NAOEIFS; 1504 for (; p < e; p++) { 1505 if (p->nd == NULL) 1506 break; /* end of the valid interfaces */ 1507 if (p->nd == nd) { 1508 p->bcnt = bcnt; /* we're updating */ 1509 nd = NULL; 1510 } else if (minbcnt > p->bcnt) 1511 minbcnt = p->bcnt; /* find the min interface */ 1512 } 1513 if (nd) { 1514 if (p == e) { 1515 pr_err("aoe: device setifbcnt failure; too many interfaces.\n"); 1516 return; 1517 } 1518 dev_hold(nd); 1519 p->nd = nd; 1520 p->bcnt = bcnt; 1521 } 1522 t->minbcnt = minbcnt; 1523 setdbcnt(d); 1524 } 1525 1526 void 1527 aoecmd_cfg_rsp(struct sk_buff *skb) 1528 { 1529 struct aoedev *d; 1530 struct aoe_hdr *h; 1531 struct aoe_cfghdr *ch; 1532 struct aoetgt *t; 1533 ulong flags, aoemajor; 1534 struct sk_buff *sl; 1535 struct sk_buff_head queue; 1536 u16 n; 1537 1538 sl = NULL; 1539 h = (struct aoe_hdr *) skb_mac_header(skb); 1540 ch = (struct aoe_cfghdr *) (h+1); 1541 1542 /* 1543 * Enough people have their dip switches set backwards to 1544 * warrant a loud message for this special case. 1545 */ 1546 aoemajor = get_unaligned_be16(&h->major); 1547 if (aoemajor == 0xfff) { 1548 printk(KERN_ERR "aoe: Warning: shelf address is all ones. " 1549 "Check shelf dip switches.\n"); 1550 return; 1551 } 1552 if (aoemajor == 0xffff) { 1553 pr_info("aoe: e%ld.%d: broadcast shelf number invalid\n", 1554 aoemajor, (int) h->minor); 1555 return; 1556 } 1557 if (h->minor == 0xff) { 1558 pr_info("aoe: e%ld.%d: broadcast slot number invalid\n", 1559 aoemajor, (int) h->minor); 1560 return; 1561 } 1562 1563 n = be16_to_cpu(ch->bufcnt); 1564 if (n > aoe_maxout) /* keep it reasonable */ 1565 n = aoe_maxout; 1566 1567 d = aoedev_by_aoeaddr(aoemajor, h->minor, 1); 1568 if (d == NULL) { 1569 pr_info("aoe: device allocation failure\n"); 1570 return; 1571 } 1572 1573 spin_lock_irqsave(&d->lock, flags); 1574 1575 t = gettgt(d, h->src); 1576 if (t) { 1577 t->nframes = n; 1578 if (n < t->maxout) 1579 aoecmd_wreset(t); 1580 } else { 1581 t = addtgt(d, h->src, n); 1582 if (!t) 1583 goto bail; 1584 } 1585 n = skb->dev->mtu; 1586 n -= sizeof(struct aoe_hdr) + sizeof(struct aoe_atahdr); 1587 n /= 512; 1588 if (n > ch->scnt) 1589 n = ch->scnt; 1590 n = n ? n * 512 : DEFAULTBCNT; 1591 setifbcnt(t, skb->dev, n); 1592 1593 /* don't change users' perspective */ 1594 if (d->nopen == 0) { 1595 d->fw_ver = be16_to_cpu(ch->fwver); 1596 sl = aoecmd_ata_id(d); 1597 } 1598 bail: 1599 spin_unlock_irqrestore(&d->lock, flags); 1600 aoedev_put(d); 1601 if (sl) { 1602 __skb_queue_head_init(&queue); 1603 __skb_queue_tail(&queue, sl); 1604 aoenet_xmit(&queue); 1605 } 1606 } 1607 1608 void 1609 aoecmd_wreset(struct aoetgt *t) 1610 { 1611 t->maxout = 1; 1612 t->ssthresh = t->nframes / 2; 1613 t->next_cwnd = t->nframes; 1614 } 1615 1616 void 1617 aoecmd_cleanslate(struct aoedev *d) 1618 { 1619 struct aoetgt **t, **te; 1620 1621 d->rttavg = RTTAVG_INIT; 1622 d->rttdev = RTTDEV_INIT; 1623 d->maxbcnt = 0; 1624 1625 t = d->targets; 1626 te = t + d->ntargets; 1627 for (; t < te && *t; t++) 1628 aoecmd_wreset(*t); 1629 } 1630 1631 void 1632 aoe_failbuf(struct aoedev *d, struct buf *buf) 1633 { 1634 if (buf == NULL) 1635 return; 1636 buf->iter.bi_size = 0; 1637 buf->bio->bi_status = BLK_STS_IOERR; 1638 if (buf->nframesout == 0) 1639 aoe_end_buf(d, buf); 1640 } 1641 1642 void 1643 aoe_flush_iocq(void) 1644 { 1645 int i; 1646 1647 for (i = 0; i < ncpus; i++) { 1648 if (kts[i].active) 1649 aoe_flush_iocq_by_index(i); 1650 } 1651 } 1652 1653 void 1654 aoe_flush_iocq_by_index(int id) 1655 { 1656 struct frame *f; 1657 struct aoedev *d; 1658 LIST_HEAD(flist); 1659 struct list_head *pos; 1660 struct sk_buff *skb; 1661 ulong flags; 1662 1663 spin_lock_irqsave(&iocq[id].lock, flags); 1664 list_splice_init(&iocq[id].head, &flist); 1665 spin_unlock_irqrestore(&iocq[id].lock, flags); 1666 while (!list_empty(&flist)) { 1667 pos = flist.next; 1668 list_del(pos); 1669 f = list_entry(pos, struct frame, head); 1670 d = f->t->d; 1671 skb = f->r_skb; 1672 spin_lock_irqsave(&d->lock, flags); 1673 if (f->buf) { 1674 f->buf->nframesout--; 1675 aoe_failbuf(d, f->buf); 1676 } 1677 aoe_freetframe(f); 1678 spin_unlock_irqrestore(&d->lock, flags); 1679 dev_kfree_skb(skb); 1680 aoedev_put(d); 1681 } 1682 } 1683 1684 int __init 1685 aoecmd_init(void) 1686 { 1687 void *p; 1688 int i; 1689 int ret; 1690 1691 /* get_zeroed_page returns page with ref count 1 */ 1692 p = (void *) get_zeroed_page(GFP_KERNEL); 1693 if (!p) 1694 return -ENOMEM; 1695 empty_page = virt_to_page(p); 1696 1697 ncpus = num_online_cpus(); 1698 1699 iocq = kcalloc(ncpus, sizeof(struct iocq_ktio), GFP_KERNEL); 1700 if (!iocq) 1701 return -ENOMEM; 1702 1703 kts = kcalloc(ncpus, sizeof(struct ktstate), GFP_KERNEL); 1704 if (!kts) { 1705 ret = -ENOMEM; 1706 goto kts_fail; 1707 } 1708 1709 ktiowq = kcalloc(ncpus, sizeof(wait_queue_head_t), GFP_KERNEL); 1710 if (!ktiowq) { 1711 ret = -ENOMEM; 1712 goto ktiowq_fail; 1713 } 1714 1715 mutex_init(&ktio_spawn_lock); 1716 1717 for (i = 0; i < ncpus; i++) { 1718 INIT_LIST_HEAD(&iocq[i].head); 1719 spin_lock_init(&iocq[i].lock); 1720 init_waitqueue_head(&ktiowq[i]); 1721 snprintf(kts[i].name, sizeof(kts[i].name), "aoe_ktio%d", i); 1722 kts[i].fn = ktio; 1723 kts[i].waitq = &ktiowq[i]; 1724 kts[i].lock = &iocq[i].lock; 1725 kts[i].id = i; 1726 kts[i].active = 0; 1727 } 1728 kts[0].active = 1; 1729 if (aoe_ktstart(&kts[0])) { 1730 ret = -ENOMEM; 1731 goto ktstart_fail; 1732 } 1733 return 0; 1734 1735 ktstart_fail: 1736 kfree(ktiowq); 1737 ktiowq_fail: 1738 kfree(kts); 1739 kts_fail: 1740 kfree(iocq); 1741 1742 return ret; 1743 } 1744 1745 void 1746 aoecmd_exit(void) 1747 { 1748 int i; 1749 1750 for (i = 0; i < ncpus; i++) 1751 if (kts[i].active) 1752 aoe_ktstop(&kts[i]); 1753 1754 aoe_flush_iocq(); 1755 1756 /* Free up the iocq and thread speicific configuration 1757 * allocated during startup. 1758 */ 1759 kfree(iocq); 1760 kfree(kts); 1761 kfree(ktiowq); 1762 1763 free_page((unsigned long) page_address(empty_page)); 1764 empty_page = NULL; 1765 } 1766