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 do_gettimeofday(&f->sent); 402 f->sent_jiffs = (u32) jiffies; 403 __skb_queue_head_init(&queue); 404 __skb_queue_tail(&queue, skb); 405 aoenet_xmit(&queue); 406 } 407 return 1; 408 } 409 410 /* some callers cannot sleep, and they can call this function, 411 * transmitting the packets later, when interrupts are on 412 */ 413 static void 414 aoecmd_cfg_pkts(ushort aoemajor, unsigned char aoeminor, struct sk_buff_head *queue) 415 { 416 struct aoe_hdr *h; 417 struct aoe_cfghdr *ch; 418 struct sk_buff *skb; 419 struct net_device *ifp; 420 421 rcu_read_lock(); 422 for_each_netdev_rcu(&init_net, ifp) { 423 dev_hold(ifp); 424 if (!is_aoe_netif(ifp)) 425 goto cont; 426 427 skb = new_skb(sizeof *h + sizeof *ch); 428 if (skb == NULL) { 429 printk(KERN_INFO "aoe: skb alloc failure\n"); 430 goto cont; 431 } 432 skb_put(skb, sizeof *h + sizeof *ch); 433 skb->dev = ifp; 434 __skb_queue_tail(queue, skb); 435 h = (struct aoe_hdr *) skb_mac_header(skb); 436 memset(h, 0, sizeof *h + sizeof *ch); 437 438 memset(h->dst, 0xff, sizeof h->dst); 439 memcpy(h->src, ifp->dev_addr, sizeof h->src); 440 h->type = __constant_cpu_to_be16(ETH_P_AOE); 441 h->verfl = AOE_HVER; 442 h->major = cpu_to_be16(aoemajor); 443 h->minor = aoeminor; 444 h->cmd = AOECMD_CFG; 445 446 cont: 447 dev_put(ifp); 448 } 449 rcu_read_unlock(); 450 } 451 452 static void 453 resend(struct aoedev *d, struct frame *f) 454 { 455 struct sk_buff *skb; 456 struct sk_buff_head queue; 457 struct aoe_hdr *h; 458 struct aoetgt *t; 459 char buf[128]; 460 u32 n; 461 462 t = f->t; 463 n = newtag(d); 464 skb = f->skb; 465 if (ifrotate(t) == NULL) { 466 /* probably can't happen, but set it up to fail anyway */ 467 pr_info("aoe: resend: no interfaces to rotate to.\n"); 468 ktcomplete(f, NULL); 469 return; 470 } 471 h = (struct aoe_hdr *) skb_mac_header(skb); 472 473 if (!(f->flags & FFL_PROBE)) { 474 snprintf(buf, sizeof(buf), 475 "%15s e%ld.%d oldtag=%08x@%08lx newtag=%08x s=%pm d=%pm nout=%d\n", 476 "retransmit", d->aoemajor, d->aoeminor, 477 f->tag, jiffies, n, 478 h->src, h->dst, t->nout); 479 aoechr_error(buf); 480 } 481 482 f->tag = n; 483 fhash(f); 484 h->tag = cpu_to_be32(n); 485 memcpy(h->dst, t->addr, sizeof h->dst); 486 memcpy(h->src, t->ifp->nd->dev_addr, sizeof h->src); 487 488 skb->dev = t->ifp->nd; 489 skb = skb_clone(skb, GFP_ATOMIC); 490 if (skb == NULL) 491 return; 492 do_gettimeofday(&f->sent); 493 f->sent_jiffs = (u32) jiffies; 494 __skb_queue_head_init(&queue); 495 __skb_queue_tail(&queue, skb); 496 aoenet_xmit(&queue); 497 } 498 499 static int 500 tsince_hr(struct frame *f) 501 { 502 struct timeval now; 503 int n; 504 505 do_gettimeofday(&now); 506 n = now.tv_usec - f->sent.tv_usec; 507 n += (now.tv_sec - f->sent.tv_sec) * USEC_PER_SEC; 508 509 if (n < 0) 510 n = -n; 511 512 /* For relatively long periods, use jiffies to avoid 513 * discrepancies caused by updates to the system time. 514 * 515 * On system with HZ of 1000, 32-bits is over 49 days 516 * worth of jiffies, or over 71 minutes worth of usecs. 517 * 518 * Jiffies overflow is handled by subtraction of unsigned ints: 519 * (gdb) print (unsigned) 2 - (unsigned) 0xfffffffe 520 * $3 = 4 521 * (gdb) 522 */ 523 if (n > USEC_PER_SEC / 4) { 524 n = ((u32) jiffies) - f->sent_jiffs; 525 n *= USEC_PER_SEC / HZ; 526 } 527 528 return n; 529 } 530 531 static int 532 tsince(u32 tag) 533 { 534 int n; 535 536 n = jiffies & 0xffff; 537 n -= tag & 0xffff; 538 if (n < 0) 539 n += 1<<16; 540 return jiffies_to_usecs(n + 1); 541 } 542 543 static struct aoeif * 544 getif(struct aoetgt *t, struct net_device *nd) 545 { 546 struct aoeif *p, *e; 547 548 p = t->ifs; 549 e = p + NAOEIFS; 550 for (; p < e; p++) 551 if (p->nd == nd) 552 return p; 553 return NULL; 554 } 555 556 static void 557 ejectif(struct aoetgt *t, struct aoeif *ifp) 558 { 559 struct aoeif *e; 560 struct net_device *nd; 561 ulong n; 562 563 nd = ifp->nd; 564 e = t->ifs + NAOEIFS - 1; 565 n = (e - ifp) * sizeof *ifp; 566 memmove(ifp, ifp+1, n); 567 e->nd = NULL; 568 dev_put(nd); 569 } 570 571 static struct frame * 572 reassign_frame(struct frame *f) 573 { 574 struct frame *nf; 575 struct sk_buff *skb; 576 577 nf = newframe(f->t->d); 578 if (!nf) 579 return NULL; 580 if (nf->t == f->t) { 581 aoe_freetframe(nf); 582 return NULL; 583 } 584 585 skb = nf->skb; 586 nf->skb = f->skb; 587 nf->buf = f->buf; 588 nf->iter = f->iter; 589 nf->waited = 0; 590 nf->waited_total = f->waited_total; 591 nf->sent = f->sent; 592 nf->sent_jiffs = f->sent_jiffs; 593 f->skb = skb; 594 595 return nf; 596 } 597 598 static void 599 probe(struct aoetgt *t) 600 { 601 struct aoedev *d; 602 struct frame *f; 603 struct sk_buff *skb; 604 struct sk_buff_head queue; 605 size_t n, m; 606 int frag; 607 608 d = t->d; 609 f = newtframe(d, t); 610 if (!f) { 611 pr_err("%s %pm for e%ld.%d: %s\n", 612 "aoe: cannot probe remote address", 613 t->addr, 614 (long) d->aoemajor, d->aoeminor, 615 "no frame available"); 616 return; 617 } 618 f->flags |= FFL_PROBE; 619 ifrotate(t); 620 f->iter.bi_size = t->d->maxbcnt ? t->d->maxbcnt : DEFAULTBCNT; 621 ata_rw_frameinit(f); 622 skb = f->skb; 623 for (frag = 0, n = f->iter.bi_size; n > 0; ++frag, n -= m) { 624 if (n < PAGE_SIZE) 625 m = n; 626 else 627 m = PAGE_SIZE; 628 skb_fill_page_desc(skb, frag, empty_page, 0, m); 629 } 630 skb->len += f->iter.bi_size; 631 skb->data_len = f->iter.bi_size; 632 skb->truesize += f->iter.bi_size; 633 634 skb = skb_clone(f->skb, GFP_ATOMIC); 635 if (skb) { 636 do_gettimeofday(&f->sent); 637 f->sent_jiffs = (u32) jiffies; 638 __skb_queue_head_init(&queue); 639 __skb_queue_tail(&queue, skb); 640 aoenet_xmit(&queue); 641 } 642 } 643 644 static long 645 rto(struct aoedev *d) 646 { 647 long t; 648 649 t = 2 * d->rttavg >> RTTSCALE; 650 t += 8 * d->rttdev >> RTTDSCALE; 651 if (t == 0) 652 t = 1; 653 654 return t; 655 } 656 657 static void 658 rexmit_deferred(struct aoedev *d) 659 { 660 struct aoetgt *t; 661 struct frame *f; 662 struct frame *nf; 663 struct list_head *pos, *nx, *head; 664 int since; 665 int untainted; 666 667 count_targets(d, &untainted); 668 669 head = &d->rexmitq; 670 list_for_each_safe(pos, nx, head) { 671 f = list_entry(pos, struct frame, head); 672 t = f->t; 673 if (t->taint) { 674 if (!(f->flags & FFL_PROBE)) { 675 nf = reassign_frame(f); 676 if (nf) { 677 if (t->nout_probes == 0 678 && untainted > 0) { 679 probe(t); 680 t->nout_probes++; 681 } 682 list_replace(&f->head, &nf->head); 683 pos = &nf->head; 684 aoe_freetframe(f); 685 f = nf; 686 t = f->t; 687 } 688 } else if (untainted < 1) { 689 /* don't probe w/o other untainted aoetgts */ 690 goto stop_probe; 691 } else if (tsince_hr(f) < t->taint * rto(d)) { 692 /* reprobe slowly when taint is high */ 693 continue; 694 } 695 } else if (f->flags & FFL_PROBE) { 696 stop_probe: /* don't probe untainted aoetgts */ 697 list_del(pos); 698 aoe_freetframe(f); 699 /* leaving d->kicked, because this is routine */ 700 f->t->d->flags |= DEVFL_KICKME; 701 continue; 702 } 703 if (t->nout >= t->maxout) 704 continue; 705 list_del(pos); 706 t->nout++; 707 if (f->flags & FFL_PROBE) 708 t->nout_probes++; 709 since = tsince_hr(f); 710 f->waited += since; 711 f->waited_total += since; 712 resend(d, f); 713 } 714 } 715 716 /* An aoetgt accumulates demerits quickly, and successful 717 * probing redeems the aoetgt slowly. 718 */ 719 static void 720 scorn(struct aoetgt *t) 721 { 722 int n; 723 724 n = t->taint++; 725 t->taint += t->taint * 2; 726 if (n > t->taint) 727 t->taint = n; 728 if (t->taint > MAX_TAINT) 729 t->taint = MAX_TAINT; 730 } 731 732 static int 733 count_targets(struct aoedev *d, int *untainted) 734 { 735 int i, good; 736 737 for (i = good = 0; i < d->ntargets && d->targets[i]; ++i) 738 if (d->targets[i]->taint == 0) 739 good++; 740 741 if (untainted) 742 *untainted = good; 743 return i; 744 } 745 746 static void 747 rexmit_timer(ulong vp) 748 { 749 struct aoedev *d; 750 struct aoetgt *t; 751 struct aoeif *ifp; 752 struct frame *f; 753 struct list_head *head, *pos, *nx; 754 LIST_HEAD(flist); 755 register long timeout; 756 ulong flags, n; 757 int i; 758 int utgts; /* number of aoetgt descriptors (not slots) */ 759 int since; 760 761 d = (struct aoedev *) vp; 762 763 spin_lock_irqsave(&d->lock, flags); 764 765 /* timeout based on observed timings and variations */ 766 timeout = rto(d); 767 768 utgts = count_targets(d, NULL); 769 770 if (d->flags & DEVFL_TKILL) { 771 spin_unlock_irqrestore(&d->lock, flags); 772 return; 773 } 774 775 /* collect all frames to rexmit into flist */ 776 for (i = 0; i < NFACTIVE; i++) { 777 head = &d->factive[i]; 778 list_for_each_safe(pos, nx, head) { 779 f = list_entry(pos, struct frame, head); 780 if (tsince_hr(f) < timeout) 781 break; /* end of expired frames */ 782 /* move to flist for later processing */ 783 list_move_tail(pos, &flist); 784 } 785 } 786 787 /* process expired frames */ 788 while (!list_empty(&flist)) { 789 pos = flist.next; 790 f = list_entry(pos, struct frame, head); 791 since = tsince_hr(f); 792 n = f->waited_total + since; 793 n /= USEC_PER_SEC; 794 if (aoe_deadsecs 795 && n > aoe_deadsecs 796 && !(f->flags & FFL_PROBE)) { 797 /* Waited too long. Device failure. 798 * Hang all frames on first hash bucket for downdev 799 * to clean up. 800 */ 801 list_splice(&flist, &d->factive[0]); 802 aoedev_downdev(d); 803 goto out; 804 } 805 806 t = f->t; 807 n = f->waited + since; 808 n /= USEC_PER_SEC; 809 if (aoe_deadsecs && utgts > 0 810 && (n > aoe_deadsecs / utgts || n > HARD_SCORN_SECS)) 811 scorn(t); /* avoid this target */ 812 813 if (t->maxout != 1) { 814 t->ssthresh = t->maxout / 2; 815 t->maxout = 1; 816 } 817 818 if (f->flags & FFL_PROBE) { 819 t->nout_probes--; 820 } else { 821 ifp = getif(t, f->skb->dev); 822 if (ifp && ++ifp->lost > (t->nframes << 1) 823 && (ifp != t->ifs || t->ifs[1].nd)) { 824 ejectif(t, ifp); 825 ifp = NULL; 826 } 827 } 828 list_move_tail(pos, &d->rexmitq); 829 t->nout--; 830 } 831 rexmit_deferred(d); 832 833 out: 834 if ((d->flags & DEVFL_KICKME) && d->blkq) { 835 d->flags &= ~DEVFL_KICKME; 836 d->blkq->request_fn(d->blkq); 837 } 838 839 d->timer.expires = jiffies + TIMERTICK; 840 add_timer(&d->timer); 841 842 spin_unlock_irqrestore(&d->lock, flags); 843 } 844 845 static unsigned long 846 rqbiocnt(struct request *r) 847 { 848 struct bio *bio; 849 unsigned long n = 0; 850 851 __rq_for_each_bio(bio, r) 852 n++; 853 return n; 854 } 855 856 static void 857 bufinit(struct buf *buf, struct request *rq, struct bio *bio) 858 { 859 memset(buf, 0, sizeof(*buf)); 860 buf->rq = rq; 861 buf->bio = bio; 862 buf->iter = bio->bi_iter; 863 } 864 865 static struct buf * 866 nextbuf(struct aoedev *d) 867 { 868 struct request *rq; 869 struct request_queue *q; 870 struct buf *buf; 871 struct bio *bio; 872 873 q = d->blkq; 874 if (q == NULL) 875 return NULL; /* initializing */ 876 if (d->ip.buf) 877 return d->ip.buf; 878 rq = d->ip.rq; 879 if (rq == NULL) { 880 rq = blk_peek_request(q); 881 if (rq == NULL) 882 return NULL; 883 blk_start_request(rq); 884 d->ip.rq = rq; 885 d->ip.nxbio = rq->bio; 886 rq->special = (void *) rqbiocnt(rq); 887 } 888 buf = mempool_alloc(d->bufpool, GFP_ATOMIC); 889 if (buf == NULL) { 890 pr_err("aoe: nextbuf: unable to mempool_alloc!\n"); 891 return NULL; 892 } 893 bio = d->ip.nxbio; 894 bufinit(buf, rq, bio); 895 bio = bio->bi_next; 896 d->ip.nxbio = bio; 897 if (bio == NULL) 898 d->ip.rq = NULL; 899 return d->ip.buf = buf; 900 } 901 902 /* enters with d->lock held */ 903 void 904 aoecmd_work(struct aoedev *d) 905 { 906 rexmit_deferred(d); 907 while (aoecmd_ata_rw(d)) 908 ; 909 } 910 911 /* this function performs work that has been deferred until sleeping is OK 912 */ 913 void 914 aoecmd_sleepwork(struct work_struct *work) 915 { 916 struct aoedev *d = container_of(work, struct aoedev, work); 917 struct block_device *bd; 918 u64 ssize; 919 920 if (d->flags & DEVFL_GDALLOC) 921 aoeblk_gdalloc(d); 922 923 if (d->flags & DEVFL_NEWSIZE) { 924 ssize = get_capacity(d->gd); 925 bd = bdget_disk(d->gd, 0); 926 if (bd) { 927 inode_lock(bd->bd_inode); 928 i_size_write(bd->bd_inode, (loff_t)ssize<<9); 929 inode_unlock(bd->bd_inode); 930 bdput(bd); 931 } 932 spin_lock_irq(&d->lock); 933 d->flags |= DEVFL_UP; 934 d->flags &= ~DEVFL_NEWSIZE; 935 spin_unlock_irq(&d->lock); 936 } 937 } 938 939 static void 940 ata_ident_fixstring(u16 *id, int ns) 941 { 942 u16 s; 943 944 while (ns-- > 0) { 945 s = *id; 946 *id++ = s >> 8 | s << 8; 947 } 948 } 949 950 static void 951 ataid_complete(struct aoedev *d, struct aoetgt *t, unsigned char *id) 952 { 953 u64 ssize; 954 u16 n; 955 956 /* word 83: command set supported */ 957 n = get_unaligned_le16(&id[83 << 1]); 958 959 /* word 86: command set/feature enabled */ 960 n |= get_unaligned_le16(&id[86 << 1]); 961 962 if (n & (1<<10)) { /* bit 10: LBA 48 */ 963 d->flags |= DEVFL_EXT; 964 965 /* word 100: number lba48 sectors */ 966 ssize = get_unaligned_le64(&id[100 << 1]); 967 968 /* set as in ide-disk.c:init_idedisk_capacity */ 969 d->geo.cylinders = ssize; 970 d->geo.cylinders /= (255 * 63); 971 d->geo.heads = 255; 972 d->geo.sectors = 63; 973 } else { 974 d->flags &= ~DEVFL_EXT; 975 976 /* number lba28 sectors */ 977 ssize = get_unaligned_le32(&id[60 << 1]); 978 979 /* NOTE: obsolete in ATA 6 */ 980 d->geo.cylinders = get_unaligned_le16(&id[54 << 1]); 981 d->geo.heads = get_unaligned_le16(&id[55 << 1]); 982 d->geo.sectors = get_unaligned_le16(&id[56 << 1]); 983 } 984 985 ata_ident_fixstring((u16 *) &id[10<<1], 10); /* serial */ 986 ata_ident_fixstring((u16 *) &id[23<<1], 4); /* firmware */ 987 ata_ident_fixstring((u16 *) &id[27<<1], 20); /* model */ 988 memcpy(d->ident, id, sizeof(d->ident)); 989 990 if (d->ssize != ssize) 991 printk(KERN_INFO 992 "aoe: %pm e%ld.%d v%04x has %llu sectors\n", 993 t->addr, 994 d->aoemajor, d->aoeminor, 995 d->fw_ver, (long long)ssize); 996 d->ssize = ssize; 997 d->geo.start = 0; 998 if (d->flags & (DEVFL_GDALLOC|DEVFL_NEWSIZE)) 999 return; 1000 if (d->gd != NULL) { 1001 set_capacity(d->gd, ssize); 1002 d->flags |= DEVFL_NEWSIZE; 1003 } else 1004 d->flags |= DEVFL_GDALLOC; 1005 schedule_work(&d->work); 1006 } 1007 1008 static void 1009 calc_rttavg(struct aoedev *d, struct aoetgt *t, int rtt) 1010 { 1011 register long n; 1012 1013 n = rtt; 1014 1015 /* cf. Congestion Avoidance and Control, Jacobson & Karels, 1988 */ 1016 n -= d->rttavg >> RTTSCALE; 1017 d->rttavg += n; 1018 if (n < 0) 1019 n = -n; 1020 n -= d->rttdev >> RTTDSCALE; 1021 d->rttdev += n; 1022 1023 if (!t || t->maxout >= t->nframes) 1024 return; 1025 if (t->maxout < t->ssthresh) 1026 t->maxout += 1; 1027 else if (t->nout == t->maxout && t->next_cwnd-- == 0) { 1028 t->maxout += 1; 1029 t->next_cwnd = t->maxout; 1030 } 1031 } 1032 1033 static struct aoetgt * 1034 gettgt(struct aoedev *d, char *addr) 1035 { 1036 struct aoetgt **t, **e; 1037 1038 t = d->targets; 1039 e = t + d->ntargets; 1040 for (; t < e && *t; t++) 1041 if (memcmp((*t)->addr, addr, sizeof((*t)->addr)) == 0) 1042 return *t; 1043 return NULL; 1044 } 1045 1046 static void 1047 bvcpy(struct sk_buff *skb, struct bio *bio, struct bvec_iter iter, long cnt) 1048 { 1049 int soff = 0; 1050 struct bio_vec bv; 1051 1052 iter.bi_size = cnt; 1053 1054 __bio_for_each_segment(bv, bio, iter, iter) { 1055 char *p = page_address(bv.bv_page) + bv.bv_offset; 1056 skb_copy_bits(skb, soff, p, bv.bv_len); 1057 soff += bv.bv_len; 1058 } 1059 } 1060 1061 void 1062 aoe_end_request(struct aoedev *d, struct request *rq, int fastfail) 1063 { 1064 struct bio *bio; 1065 int bok; 1066 struct request_queue *q; 1067 1068 q = d->blkq; 1069 if (rq == d->ip.rq) 1070 d->ip.rq = NULL; 1071 do { 1072 bio = rq->bio; 1073 bok = !fastfail && !bio->bi_status; 1074 } while (__blk_end_request(rq, bok ? BLK_STS_OK : BLK_STS_IOERR, bio->bi_iter.bi_size)); 1075 1076 /* cf. http://lkml.org/lkml/2006/10/31/28 */ 1077 if (!fastfail) 1078 __blk_run_queue(q); 1079 } 1080 1081 static void 1082 aoe_end_buf(struct aoedev *d, struct buf *buf) 1083 { 1084 struct request *rq; 1085 unsigned long n; 1086 1087 if (buf == d->ip.buf) 1088 d->ip.buf = NULL; 1089 rq = buf->rq; 1090 mempool_free(buf, d->bufpool); 1091 n = (unsigned long) rq->special; 1092 rq->special = (void *) --n; 1093 if (n == 0) 1094 aoe_end_request(d, rq, 0); 1095 } 1096 1097 static void 1098 ktiocomplete(struct frame *f) 1099 { 1100 struct aoe_hdr *hin, *hout; 1101 struct aoe_atahdr *ahin, *ahout; 1102 struct buf *buf; 1103 struct sk_buff *skb; 1104 struct aoetgt *t; 1105 struct aoeif *ifp; 1106 struct aoedev *d; 1107 long n; 1108 int untainted; 1109 1110 if (f == NULL) 1111 return; 1112 1113 t = f->t; 1114 d = t->d; 1115 skb = f->r_skb; 1116 buf = f->buf; 1117 if (f->flags & FFL_PROBE) 1118 goto out; 1119 if (!skb) /* just fail the buf. */ 1120 goto noskb; 1121 1122 hout = (struct aoe_hdr *) skb_mac_header(f->skb); 1123 ahout = (struct aoe_atahdr *) (hout+1); 1124 1125 hin = (struct aoe_hdr *) skb->data; 1126 skb_pull(skb, sizeof(*hin)); 1127 ahin = (struct aoe_atahdr *) skb->data; 1128 skb_pull(skb, sizeof(*ahin)); 1129 if (ahin->cmdstat & 0xa9) { /* these bits cleared on success */ 1130 pr_err("aoe: ata error cmd=%2.2Xh stat=%2.2Xh from e%ld.%d\n", 1131 ahout->cmdstat, ahin->cmdstat, 1132 d->aoemajor, d->aoeminor); 1133 noskb: if (buf) 1134 buf->bio->bi_status = BLK_STS_IOERR; 1135 goto out; 1136 } 1137 1138 n = ahout->scnt << 9; 1139 switch (ahout->cmdstat) { 1140 case ATA_CMD_PIO_READ: 1141 case ATA_CMD_PIO_READ_EXT: 1142 if (skb->len < n) { 1143 pr_err("%s e%ld.%d. skb->len=%d need=%ld\n", 1144 "aoe: runt data size in read from", 1145 (long) d->aoemajor, d->aoeminor, 1146 skb->len, n); 1147 buf->bio->bi_status = BLK_STS_IOERR; 1148 break; 1149 } 1150 if (n > f->iter.bi_size) { 1151 pr_err_ratelimited("%s e%ld.%d. bytes=%ld need=%u\n", 1152 "aoe: too-large data size in read from", 1153 (long) d->aoemajor, d->aoeminor, 1154 n, f->iter.bi_size); 1155 buf->bio->bi_status = BLK_STS_IOERR; 1156 break; 1157 } 1158 bvcpy(skb, f->buf->bio, f->iter, n); 1159 case ATA_CMD_PIO_WRITE: 1160 case ATA_CMD_PIO_WRITE_EXT: 1161 spin_lock_irq(&d->lock); 1162 ifp = getif(t, skb->dev); 1163 if (ifp) 1164 ifp->lost = 0; 1165 spin_unlock_irq(&d->lock); 1166 break; 1167 case ATA_CMD_ID_ATA: 1168 if (skb->len < 512) { 1169 pr_info("%s e%ld.%d. skb->len=%d need=512\n", 1170 "aoe: runt data size in ataid from", 1171 (long) d->aoemajor, d->aoeminor, 1172 skb->len); 1173 break; 1174 } 1175 if (skb_linearize(skb)) 1176 break; 1177 spin_lock_irq(&d->lock); 1178 ataid_complete(d, t, skb->data); 1179 spin_unlock_irq(&d->lock); 1180 break; 1181 default: 1182 pr_info("aoe: unrecognized ata command %2.2Xh for %d.%d\n", 1183 ahout->cmdstat, 1184 be16_to_cpu(get_unaligned(&hin->major)), 1185 hin->minor); 1186 } 1187 out: 1188 spin_lock_irq(&d->lock); 1189 if (t->taint > 0 1190 && --t->taint > 0 1191 && t->nout_probes == 0) { 1192 count_targets(d, &untainted); 1193 if (untainted > 0) { 1194 probe(t); 1195 t->nout_probes++; 1196 } 1197 } 1198 1199 aoe_freetframe(f); 1200 1201 if (buf && --buf->nframesout == 0 && buf->iter.bi_size == 0) 1202 aoe_end_buf(d, buf); 1203 1204 spin_unlock_irq(&d->lock); 1205 aoedev_put(d); 1206 dev_kfree_skb(skb); 1207 } 1208 1209 /* Enters with iocq.lock held. 1210 * Returns true iff responses needing processing remain. 1211 */ 1212 static int 1213 ktio(int id) 1214 { 1215 struct frame *f; 1216 struct list_head *pos; 1217 int i; 1218 int actual_id; 1219 1220 for (i = 0; ; ++i) { 1221 if (i == MAXIOC) 1222 return 1; 1223 if (list_empty(&iocq[id].head)) 1224 return 0; 1225 pos = iocq[id].head.next; 1226 list_del(pos); 1227 f = list_entry(pos, struct frame, head); 1228 spin_unlock_irq(&iocq[id].lock); 1229 ktiocomplete(f); 1230 1231 /* Figure out if extra threads are required. */ 1232 actual_id = f->t->d->aoeminor % ncpus; 1233 1234 if (!kts[actual_id].active) { 1235 BUG_ON(id != 0); 1236 mutex_lock(&ktio_spawn_lock); 1237 if (!kts[actual_id].active 1238 && aoe_ktstart(&kts[actual_id]) == 0) 1239 kts[actual_id].active = 1; 1240 mutex_unlock(&ktio_spawn_lock); 1241 } 1242 spin_lock_irq(&iocq[id].lock); 1243 } 1244 } 1245 1246 static int 1247 kthread(void *vp) 1248 { 1249 struct ktstate *k; 1250 DECLARE_WAITQUEUE(wait, current); 1251 int more; 1252 1253 k = vp; 1254 current->flags |= PF_NOFREEZE; 1255 set_user_nice(current, -10); 1256 complete(&k->rendez); /* tell spawner we're running */ 1257 do { 1258 spin_lock_irq(k->lock); 1259 more = k->fn(k->id); 1260 if (!more) { 1261 add_wait_queue(k->waitq, &wait); 1262 __set_current_state(TASK_INTERRUPTIBLE); 1263 } 1264 spin_unlock_irq(k->lock); 1265 if (!more) { 1266 schedule(); 1267 remove_wait_queue(k->waitq, &wait); 1268 } else 1269 cond_resched(); 1270 } while (!kthread_should_stop()); 1271 complete(&k->rendez); /* tell spawner we're stopping */ 1272 return 0; 1273 } 1274 1275 void 1276 aoe_ktstop(struct ktstate *k) 1277 { 1278 kthread_stop(k->task); 1279 wait_for_completion(&k->rendez); 1280 } 1281 1282 int 1283 aoe_ktstart(struct ktstate *k) 1284 { 1285 struct task_struct *task; 1286 1287 init_completion(&k->rendez); 1288 task = kthread_run(kthread, k, "%s", k->name); 1289 if (task == NULL || IS_ERR(task)) 1290 return -ENOMEM; 1291 k->task = task; 1292 wait_for_completion(&k->rendez); /* allow kthread to start */ 1293 init_completion(&k->rendez); /* for waiting for exit later */ 1294 return 0; 1295 } 1296 1297 /* pass it off to kthreads for processing */ 1298 static void 1299 ktcomplete(struct frame *f, struct sk_buff *skb) 1300 { 1301 int id; 1302 ulong flags; 1303 1304 f->r_skb = skb; 1305 id = f->t->d->aoeminor % ncpus; 1306 spin_lock_irqsave(&iocq[id].lock, flags); 1307 if (!kts[id].active) { 1308 spin_unlock_irqrestore(&iocq[id].lock, flags); 1309 /* The thread with id has not been spawned yet, 1310 * so delegate the work to the main thread and 1311 * try spawning a new thread. 1312 */ 1313 id = 0; 1314 spin_lock_irqsave(&iocq[id].lock, flags); 1315 } 1316 list_add_tail(&f->head, &iocq[id].head); 1317 spin_unlock_irqrestore(&iocq[id].lock, flags); 1318 wake_up(&ktiowq[id]); 1319 } 1320 1321 struct sk_buff * 1322 aoecmd_ata_rsp(struct sk_buff *skb) 1323 { 1324 struct aoedev *d; 1325 struct aoe_hdr *h; 1326 struct frame *f; 1327 u32 n; 1328 ulong flags; 1329 char ebuf[128]; 1330 u16 aoemajor; 1331 1332 h = (struct aoe_hdr *) skb->data; 1333 aoemajor = be16_to_cpu(get_unaligned(&h->major)); 1334 d = aoedev_by_aoeaddr(aoemajor, h->minor, 0); 1335 if (d == NULL) { 1336 snprintf(ebuf, sizeof ebuf, "aoecmd_ata_rsp: ata response " 1337 "for unknown device %d.%d\n", 1338 aoemajor, h->minor); 1339 aoechr_error(ebuf); 1340 return skb; 1341 } 1342 1343 spin_lock_irqsave(&d->lock, flags); 1344 1345 n = be32_to_cpu(get_unaligned(&h->tag)); 1346 f = getframe(d, n); 1347 if (f) { 1348 calc_rttavg(d, f->t, tsince_hr(f)); 1349 f->t->nout--; 1350 if (f->flags & FFL_PROBE) 1351 f->t->nout_probes--; 1352 } else { 1353 f = getframe_deferred(d, n); 1354 if (f) { 1355 calc_rttavg(d, NULL, tsince_hr(f)); 1356 } else { 1357 calc_rttavg(d, NULL, tsince(n)); 1358 spin_unlock_irqrestore(&d->lock, flags); 1359 aoedev_put(d); 1360 snprintf(ebuf, sizeof(ebuf), 1361 "%15s e%d.%d tag=%08x@%08lx s=%pm d=%pm\n", 1362 "unexpected rsp", 1363 get_unaligned_be16(&h->major), 1364 h->minor, 1365 get_unaligned_be32(&h->tag), 1366 jiffies, 1367 h->src, 1368 h->dst); 1369 aoechr_error(ebuf); 1370 return skb; 1371 } 1372 } 1373 aoecmd_work(d); 1374 1375 spin_unlock_irqrestore(&d->lock, flags); 1376 1377 ktcomplete(f, skb); 1378 1379 /* 1380 * Note here that we do not perform an aoedev_put, as we are 1381 * leaving this reference for the ktio to release. 1382 */ 1383 return NULL; 1384 } 1385 1386 void 1387 aoecmd_cfg(ushort aoemajor, unsigned char aoeminor) 1388 { 1389 struct sk_buff_head queue; 1390 1391 __skb_queue_head_init(&queue); 1392 aoecmd_cfg_pkts(aoemajor, aoeminor, &queue); 1393 aoenet_xmit(&queue); 1394 } 1395 1396 struct sk_buff * 1397 aoecmd_ata_id(struct aoedev *d) 1398 { 1399 struct aoe_hdr *h; 1400 struct aoe_atahdr *ah; 1401 struct frame *f; 1402 struct sk_buff *skb; 1403 struct aoetgt *t; 1404 1405 f = newframe(d); 1406 if (f == NULL) 1407 return NULL; 1408 1409 t = *d->tgt; 1410 1411 /* initialize the headers & frame */ 1412 skb = f->skb; 1413 h = (struct aoe_hdr *) skb_mac_header(skb); 1414 ah = (struct aoe_atahdr *) (h+1); 1415 skb_put(skb, sizeof *h + sizeof *ah); 1416 memset(h, 0, skb->len); 1417 f->tag = aoehdr_atainit(d, t, h); 1418 fhash(f); 1419 t->nout++; 1420 f->waited = 0; 1421 f->waited_total = 0; 1422 1423 /* set up ata header */ 1424 ah->scnt = 1; 1425 ah->cmdstat = ATA_CMD_ID_ATA; 1426 ah->lba3 = 0xa0; 1427 1428 skb->dev = t->ifp->nd; 1429 1430 d->rttavg = RTTAVG_INIT; 1431 d->rttdev = RTTDEV_INIT; 1432 d->timer.function = rexmit_timer; 1433 1434 skb = skb_clone(skb, GFP_ATOMIC); 1435 if (skb) { 1436 do_gettimeofday(&f->sent); 1437 f->sent_jiffs = (u32) jiffies; 1438 } 1439 1440 return skb; 1441 } 1442 1443 static struct aoetgt ** 1444 grow_targets(struct aoedev *d) 1445 { 1446 ulong oldn, newn; 1447 struct aoetgt **tt; 1448 1449 oldn = d->ntargets; 1450 newn = oldn * 2; 1451 tt = kcalloc(newn, sizeof(*d->targets), GFP_ATOMIC); 1452 if (!tt) 1453 return NULL; 1454 memmove(tt, d->targets, sizeof(*d->targets) * oldn); 1455 d->tgt = tt + (d->tgt - d->targets); 1456 kfree(d->targets); 1457 d->targets = tt; 1458 d->ntargets = newn; 1459 1460 return &d->targets[oldn]; 1461 } 1462 1463 static struct aoetgt * 1464 addtgt(struct aoedev *d, char *addr, ulong nframes) 1465 { 1466 struct aoetgt *t, **tt, **te; 1467 1468 tt = d->targets; 1469 te = tt + d->ntargets; 1470 for (; tt < te && *tt; tt++) 1471 ; 1472 1473 if (tt == te) { 1474 tt = grow_targets(d); 1475 if (!tt) 1476 goto nomem; 1477 } 1478 t = kzalloc(sizeof(*t), GFP_ATOMIC); 1479 if (!t) 1480 goto nomem; 1481 t->nframes = nframes; 1482 t->d = d; 1483 memcpy(t->addr, addr, sizeof t->addr); 1484 t->ifp = t->ifs; 1485 aoecmd_wreset(t); 1486 t->maxout = t->nframes / 2; 1487 INIT_LIST_HEAD(&t->ffree); 1488 return *tt = t; 1489 1490 nomem: 1491 pr_info("aoe: cannot allocate memory to add target\n"); 1492 return NULL; 1493 } 1494 1495 static void 1496 setdbcnt(struct aoedev *d) 1497 { 1498 struct aoetgt **t, **e; 1499 int bcnt = 0; 1500 1501 t = d->targets; 1502 e = t + d->ntargets; 1503 for (; t < e && *t; t++) 1504 if (bcnt == 0 || bcnt > (*t)->minbcnt) 1505 bcnt = (*t)->minbcnt; 1506 if (bcnt != d->maxbcnt) { 1507 d->maxbcnt = bcnt; 1508 pr_info("aoe: e%ld.%d: setting %d byte data frames\n", 1509 d->aoemajor, d->aoeminor, bcnt); 1510 } 1511 } 1512 1513 static void 1514 setifbcnt(struct aoetgt *t, struct net_device *nd, int bcnt) 1515 { 1516 struct aoedev *d; 1517 struct aoeif *p, *e; 1518 int minbcnt; 1519 1520 d = t->d; 1521 minbcnt = bcnt; 1522 p = t->ifs; 1523 e = p + NAOEIFS; 1524 for (; p < e; p++) { 1525 if (p->nd == NULL) 1526 break; /* end of the valid interfaces */ 1527 if (p->nd == nd) { 1528 p->bcnt = bcnt; /* we're updating */ 1529 nd = NULL; 1530 } else if (minbcnt > p->bcnt) 1531 minbcnt = p->bcnt; /* find the min interface */ 1532 } 1533 if (nd) { 1534 if (p == e) { 1535 pr_err("aoe: device setifbcnt failure; too many interfaces.\n"); 1536 return; 1537 } 1538 dev_hold(nd); 1539 p->nd = nd; 1540 p->bcnt = bcnt; 1541 } 1542 t->minbcnt = minbcnt; 1543 setdbcnt(d); 1544 } 1545 1546 void 1547 aoecmd_cfg_rsp(struct sk_buff *skb) 1548 { 1549 struct aoedev *d; 1550 struct aoe_hdr *h; 1551 struct aoe_cfghdr *ch; 1552 struct aoetgt *t; 1553 ulong flags, aoemajor; 1554 struct sk_buff *sl; 1555 struct sk_buff_head queue; 1556 u16 n; 1557 1558 sl = NULL; 1559 h = (struct aoe_hdr *) skb_mac_header(skb); 1560 ch = (struct aoe_cfghdr *) (h+1); 1561 1562 /* 1563 * Enough people have their dip switches set backwards to 1564 * warrant a loud message for this special case. 1565 */ 1566 aoemajor = get_unaligned_be16(&h->major); 1567 if (aoemajor == 0xfff) { 1568 printk(KERN_ERR "aoe: Warning: shelf address is all ones. " 1569 "Check shelf dip switches.\n"); 1570 return; 1571 } 1572 if (aoemajor == 0xffff) { 1573 pr_info("aoe: e%ld.%d: broadcast shelf number invalid\n", 1574 aoemajor, (int) h->minor); 1575 return; 1576 } 1577 if (h->minor == 0xff) { 1578 pr_info("aoe: e%ld.%d: broadcast slot number invalid\n", 1579 aoemajor, (int) h->minor); 1580 return; 1581 } 1582 1583 n = be16_to_cpu(ch->bufcnt); 1584 if (n > aoe_maxout) /* keep it reasonable */ 1585 n = aoe_maxout; 1586 1587 d = aoedev_by_aoeaddr(aoemajor, h->minor, 1); 1588 if (d == NULL) { 1589 pr_info("aoe: device allocation failure\n"); 1590 return; 1591 } 1592 1593 spin_lock_irqsave(&d->lock, flags); 1594 1595 t = gettgt(d, h->src); 1596 if (t) { 1597 t->nframes = n; 1598 if (n < t->maxout) 1599 aoecmd_wreset(t); 1600 } else { 1601 t = addtgt(d, h->src, n); 1602 if (!t) 1603 goto bail; 1604 } 1605 n = skb->dev->mtu; 1606 n -= sizeof(struct aoe_hdr) + sizeof(struct aoe_atahdr); 1607 n /= 512; 1608 if (n > ch->scnt) 1609 n = ch->scnt; 1610 n = n ? n * 512 : DEFAULTBCNT; 1611 setifbcnt(t, skb->dev, n); 1612 1613 /* don't change users' perspective */ 1614 if (d->nopen == 0) { 1615 d->fw_ver = be16_to_cpu(ch->fwver); 1616 sl = aoecmd_ata_id(d); 1617 } 1618 bail: 1619 spin_unlock_irqrestore(&d->lock, flags); 1620 aoedev_put(d); 1621 if (sl) { 1622 __skb_queue_head_init(&queue); 1623 __skb_queue_tail(&queue, sl); 1624 aoenet_xmit(&queue); 1625 } 1626 } 1627 1628 void 1629 aoecmd_wreset(struct aoetgt *t) 1630 { 1631 t->maxout = 1; 1632 t->ssthresh = t->nframes / 2; 1633 t->next_cwnd = t->nframes; 1634 } 1635 1636 void 1637 aoecmd_cleanslate(struct aoedev *d) 1638 { 1639 struct aoetgt **t, **te; 1640 1641 d->rttavg = RTTAVG_INIT; 1642 d->rttdev = RTTDEV_INIT; 1643 d->maxbcnt = 0; 1644 1645 t = d->targets; 1646 te = t + d->ntargets; 1647 for (; t < te && *t; t++) 1648 aoecmd_wreset(*t); 1649 } 1650 1651 void 1652 aoe_failbuf(struct aoedev *d, struct buf *buf) 1653 { 1654 if (buf == NULL) 1655 return; 1656 buf->iter.bi_size = 0; 1657 buf->bio->bi_status = BLK_STS_IOERR; 1658 if (buf->nframesout == 0) 1659 aoe_end_buf(d, buf); 1660 } 1661 1662 void 1663 aoe_flush_iocq(void) 1664 { 1665 int i; 1666 1667 for (i = 0; i < ncpus; i++) { 1668 if (kts[i].active) 1669 aoe_flush_iocq_by_index(i); 1670 } 1671 } 1672 1673 void 1674 aoe_flush_iocq_by_index(int id) 1675 { 1676 struct frame *f; 1677 struct aoedev *d; 1678 LIST_HEAD(flist); 1679 struct list_head *pos; 1680 struct sk_buff *skb; 1681 ulong flags; 1682 1683 spin_lock_irqsave(&iocq[id].lock, flags); 1684 list_splice_init(&iocq[id].head, &flist); 1685 spin_unlock_irqrestore(&iocq[id].lock, flags); 1686 while (!list_empty(&flist)) { 1687 pos = flist.next; 1688 list_del(pos); 1689 f = list_entry(pos, struct frame, head); 1690 d = f->t->d; 1691 skb = f->r_skb; 1692 spin_lock_irqsave(&d->lock, flags); 1693 if (f->buf) { 1694 f->buf->nframesout--; 1695 aoe_failbuf(d, f->buf); 1696 } 1697 aoe_freetframe(f); 1698 spin_unlock_irqrestore(&d->lock, flags); 1699 dev_kfree_skb(skb); 1700 aoedev_put(d); 1701 } 1702 } 1703 1704 int __init 1705 aoecmd_init(void) 1706 { 1707 void *p; 1708 int i; 1709 int ret; 1710 1711 /* get_zeroed_page returns page with ref count 1 */ 1712 p = (void *) get_zeroed_page(GFP_KERNEL); 1713 if (!p) 1714 return -ENOMEM; 1715 empty_page = virt_to_page(p); 1716 1717 ncpus = num_online_cpus(); 1718 1719 iocq = kcalloc(ncpus, sizeof(struct iocq_ktio), GFP_KERNEL); 1720 if (!iocq) 1721 return -ENOMEM; 1722 1723 kts = kcalloc(ncpus, sizeof(struct ktstate), GFP_KERNEL); 1724 if (!kts) { 1725 ret = -ENOMEM; 1726 goto kts_fail; 1727 } 1728 1729 ktiowq = kcalloc(ncpus, sizeof(wait_queue_head_t), GFP_KERNEL); 1730 if (!ktiowq) { 1731 ret = -ENOMEM; 1732 goto ktiowq_fail; 1733 } 1734 1735 mutex_init(&ktio_spawn_lock); 1736 1737 for (i = 0; i < ncpus; i++) { 1738 INIT_LIST_HEAD(&iocq[i].head); 1739 spin_lock_init(&iocq[i].lock); 1740 init_waitqueue_head(&ktiowq[i]); 1741 snprintf(kts[i].name, sizeof(kts[i].name), "aoe_ktio%d", i); 1742 kts[i].fn = ktio; 1743 kts[i].waitq = &ktiowq[i]; 1744 kts[i].lock = &iocq[i].lock; 1745 kts[i].id = i; 1746 kts[i].active = 0; 1747 } 1748 kts[0].active = 1; 1749 if (aoe_ktstart(&kts[0])) { 1750 ret = -ENOMEM; 1751 goto ktstart_fail; 1752 } 1753 return 0; 1754 1755 ktstart_fail: 1756 kfree(ktiowq); 1757 ktiowq_fail: 1758 kfree(kts); 1759 kts_fail: 1760 kfree(iocq); 1761 1762 return ret; 1763 } 1764 1765 void 1766 aoecmd_exit(void) 1767 { 1768 int i; 1769 1770 for (i = 0; i < ncpus; i++) 1771 if (kts[i].active) 1772 aoe_ktstop(&kts[i]); 1773 1774 aoe_flush_iocq(); 1775 1776 /* Free up the iocq and thread speicific configuration 1777 * allocated during startup. 1778 */ 1779 kfree(iocq); 1780 kfree(kts); 1781 kfree(ktiowq); 1782 1783 free_page((unsigned long) page_address(empty_page)); 1784 empty_page = NULL; 1785 } 1786