xref: /openbmc/linux/drivers/block/aoe/aoecmd.c (revision b04b4f78) 
1 /* Copyright (c) 2007 Coraid, Inc.  See COPYING for GPL terms. */
2 /*
3  * aoecmd.c
4  * Filesystem request handling methods
5  */
6 
7 #include <linux/ata.h>
8 #include <linux/hdreg.h>
9 #include <linux/blkdev.h>
10 #include <linux/skbuff.h>
11 #include <linux/netdevice.h>
12 #include <linux/genhd.h>
13 #include <linux/moduleparam.h>
14 #include <net/net_namespace.h>
15 #include <asm/unaligned.h>
16 #include "aoe.h"
17 
18 static int aoe_deadsecs = 60 * 3;
19 module_param(aoe_deadsecs, int, 0644);
20 MODULE_PARM_DESC(aoe_deadsecs, "After aoe_deadsecs seconds, give up and fail dev.");
21 
22 static int aoe_maxout = 16;
23 module_param(aoe_maxout, int, 0644);
24 MODULE_PARM_DESC(aoe_maxout,
25 	"Only aoe_maxout outstanding packets for every MAC on eX.Y.");
26 
27 static struct sk_buff *
28 new_skb(ulong len)
29 {
30 	struct sk_buff *skb;
31 
32 	skb = alloc_skb(len, GFP_ATOMIC);
33 	if (skb) {
34 		skb_reset_mac_header(skb);
35 		skb_reset_network_header(skb);
36 		skb->protocol = __constant_htons(ETH_P_AOE);
37 		skb->priority = 0;
38 		skb->next = skb->prev = NULL;
39 
40 		/* tell the network layer not to perform IP checksums
41 		 * or to get the NIC to do it
42 		 */
43 		skb->ip_summed = CHECKSUM_NONE;
44 	}
45 	return skb;
46 }
47 
48 static struct frame *
49 getframe(struct aoetgt *t, int tag)
50 {
51 	struct frame *f, *e;
52 
53 	f = t->frames;
54 	e = f + t->nframes;
55 	for (; f<e; f++)
56 		if (f->tag == tag)
57 			return f;
58 	return NULL;
59 }
60 
61 /*
62  * Leave the top bit clear so we have tagspace for userland.
63  * The bottom 16 bits are the xmit tick for rexmit/rttavg processing.
64  * This driver reserves tag -1 to mean "unused frame."
65  */
66 static int
67 newtag(struct aoetgt *t)
68 {
69 	register ulong n;
70 
71 	n = jiffies & 0xffff;
72 	return n |= (++t->lasttag & 0x7fff) << 16;
73 }
74 
75 static int
76 aoehdr_atainit(struct aoedev *d, struct aoetgt *t, struct aoe_hdr *h)
77 {
78 	u32 host_tag = newtag(t);
79 
80 	memcpy(h->src, t->ifp->nd->dev_addr, sizeof h->src);
81 	memcpy(h->dst, t->addr, sizeof h->dst);
82 	h->type = __constant_cpu_to_be16(ETH_P_AOE);
83 	h->verfl = AOE_HVER;
84 	h->major = cpu_to_be16(d->aoemajor);
85 	h->minor = d->aoeminor;
86 	h->cmd = AOECMD_ATA;
87 	h->tag = cpu_to_be32(host_tag);
88 
89 	return host_tag;
90 }
91 
92 static inline void
93 put_lba(struct aoe_atahdr *ah, sector_t lba)
94 {
95 	ah->lba0 = lba;
96 	ah->lba1 = lba >>= 8;
97 	ah->lba2 = lba >>= 8;
98 	ah->lba3 = lba >>= 8;
99 	ah->lba4 = lba >>= 8;
100 	ah->lba5 = lba >>= 8;
101 }
102 
103 static void
104 ifrotate(struct aoetgt *t)
105 {
106 	t->ifp++;
107 	if (t->ifp >= &t->ifs[NAOEIFS] || t->ifp->nd == NULL)
108 		t->ifp = t->ifs;
109 	if (t->ifp->nd == NULL) {
110 		printk(KERN_INFO "aoe: no interface to rotate to\n");
111 		BUG();
112 	}
113 }
114 
115 static void
116 skb_pool_put(struct aoedev *d, struct sk_buff *skb)
117 {
118 	__skb_queue_tail(&d->skbpool, skb);
119 }
120 
121 static struct sk_buff *
122 skb_pool_get(struct aoedev *d)
123 {
124 	struct sk_buff *skb = skb_peek(&d->skbpool);
125 
126 	if (skb && atomic_read(&skb_shinfo(skb)->dataref) == 1) {
127 		__skb_unlink(skb, &d->skbpool);
128 		return skb;
129 	}
130 	if (skb_queue_len(&d->skbpool) < NSKBPOOLMAX &&
131 	    (skb = new_skb(ETH_ZLEN)))
132 		return skb;
133 
134 	return NULL;
135 }
136 
137 /* freeframe is where we do our load balancing so it's a little hairy. */
138 static struct frame *
139 freeframe(struct aoedev *d)
140 {
141 	struct frame *f, *e, *rf;
142 	struct aoetgt **t;
143 	struct sk_buff *skb;
144 
145 	if (d->targets[0] == NULL) {	/* shouldn't happen, but I'm paranoid */
146 		printk(KERN_ERR "aoe: NULL TARGETS!\n");
147 		return NULL;
148 	}
149 	t = d->tgt;
150 	t++;
151 	if (t >= &d->targets[NTARGETS] || !*t)
152 		t = d->targets;
153 	for (;;) {
154 		if ((*t)->nout < (*t)->maxout
155 		&& t != d->htgt
156 		&& (*t)->ifp->nd) {
157 			rf = NULL;
158 			f = (*t)->frames;
159 			e = f + (*t)->nframes;
160 			for (; f < e; f++) {
161 				if (f->tag != FREETAG)
162 					continue;
163 				skb = f->skb;
164 				if (!skb
165 				&& !(f->skb = skb = new_skb(ETH_ZLEN)))
166 					continue;
167 				if (atomic_read(&skb_shinfo(skb)->dataref)
168 					!= 1) {
169 					if (!rf)
170 						rf = f;
171 					continue;
172 				}
173 gotone:				skb_shinfo(skb)->nr_frags = skb->data_len = 0;
174 				skb_trim(skb, 0);
175 				d->tgt = t;
176 				ifrotate(*t);
177 				return f;
178 			}
179 			/* Work can be done, but the network layer is
180 			   holding our precious packets.  Try to grab
181 			   one from the pool. */
182 			f = rf;
183 			if (f == NULL) {	/* more paranoia */
184 				printk(KERN_ERR
185 					"aoe: freeframe: %s.\n",
186 					"unexpected null rf");
187 				d->flags |= DEVFL_KICKME;
188 				return NULL;
189 			}
190 			skb = skb_pool_get(d);
191 			if (skb) {
192 				skb_pool_put(d, f->skb);
193 				f->skb = skb;
194 				goto gotone;
195 			}
196 			(*t)->dataref++;
197 			if ((*t)->nout == 0)
198 				d->flags |= DEVFL_KICKME;
199 		}
200 		if (t == d->tgt)	/* we've looped and found nada */
201 			break;
202 		t++;
203 		if (t >= &d->targets[NTARGETS] || !*t)
204 			t = d->targets;
205 	}
206 	return NULL;
207 }
208 
209 static int
210 aoecmd_ata_rw(struct aoedev *d)
211 {
212 	struct frame *f;
213 	struct aoe_hdr *h;
214 	struct aoe_atahdr *ah;
215 	struct buf *buf;
216 	struct bio_vec *bv;
217 	struct aoetgt *t;
218 	struct sk_buff *skb;
219 	ulong bcnt;
220 	char writebit, extbit;
221 
222 	writebit = 0x10;
223 	extbit = 0x4;
224 
225 	f = freeframe(d);
226 	if (f == NULL)
227 		return 0;
228 	t = *d->tgt;
229 	buf = d->inprocess;
230 	bv = buf->bv;
231 	bcnt = t->ifp->maxbcnt;
232 	if (bcnt == 0)
233 		bcnt = DEFAULTBCNT;
234 	if (bcnt > buf->bv_resid)
235 		bcnt = buf->bv_resid;
236 	/* initialize the headers & frame */
237 	skb = f->skb;
238 	h = (struct aoe_hdr *) skb_mac_header(skb);
239 	ah = (struct aoe_atahdr *) (h+1);
240 	skb_put(skb, sizeof *h + sizeof *ah);
241 	memset(h, 0, skb->len);
242 	f->tag = aoehdr_atainit(d, t, h);
243 	t->nout++;
244 	f->waited = 0;
245 	f->buf = buf;
246 	f->bufaddr = page_address(bv->bv_page) + buf->bv_off;
247 	f->bcnt = bcnt;
248 	f->lba = buf->sector;
249 
250 	/* set up ata header */
251 	ah->scnt = bcnt >> 9;
252 	put_lba(ah, buf->sector);
253 	if (d->flags & DEVFL_EXT) {
254 		ah->aflags |= AOEAFL_EXT;
255 	} else {
256 		extbit = 0;
257 		ah->lba3 &= 0x0f;
258 		ah->lba3 |= 0xe0;	/* LBA bit + obsolete 0xa0 */
259 	}
260 	if (bio_data_dir(buf->bio) == WRITE) {
261 		skb_fill_page_desc(skb, 0, bv->bv_page, buf->bv_off, bcnt);
262 		ah->aflags |= AOEAFL_WRITE;
263 		skb->len += bcnt;
264 		skb->data_len = bcnt;
265 		t->wpkts++;
266 	} else {
267 		t->rpkts++;
268 		writebit = 0;
269 	}
270 
271 	ah->cmdstat = ATA_CMD_PIO_READ | writebit | extbit;
272 
273 	/* mark all tracking fields and load out */
274 	buf->nframesout += 1;
275 	buf->bv_off += bcnt;
276 	buf->bv_resid -= bcnt;
277 	buf->resid -= bcnt;
278 	buf->sector += bcnt >> 9;
279 	if (buf->resid == 0) {
280 		d->inprocess = NULL;
281 	} else if (buf->bv_resid == 0) {
282 		buf->bv = ++bv;
283 		buf->bv_resid = bv->bv_len;
284 		WARN_ON(buf->bv_resid == 0);
285 		buf->bv_off = bv->bv_offset;
286 	}
287 
288 	skb->dev = t->ifp->nd;
289 	skb = skb_clone(skb, GFP_ATOMIC);
290 	if (skb)
291 		__skb_queue_tail(&d->sendq, skb);
292 	return 1;
293 }
294 
295 /* some callers cannot sleep, and they can call this function,
296  * transmitting the packets later, when interrupts are on
297  */
298 static void
299 aoecmd_cfg_pkts(ushort aoemajor, unsigned char aoeminor, struct sk_buff_head *queue)
300 {
301 	struct aoe_hdr *h;
302 	struct aoe_cfghdr *ch;
303 	struct sk_buff *skb;
304 	struct net_device *ifp;
305 
306 	read_lock(&dev_base_lock);
307 	for_each_netdev(&init_net, ifp) {
308 		dev_hold(ifp);
309 		if (!is_aoe_netif(ifp))
310 			goto cont;
311 
312 		skb = new_skb(sizeof *h + sizeof *ch);
313 		if (skb == NULL) {
314 			printk(KERN_INFO "aoe: skb alloc failure\n");
315 			goto cont;
316 		}
317 		skb_put(skb, sizeof *h + sizeof *ch);
318 		skb->dev = ifp;
319 		__skb_queue_tail(queue, skb);
320 		h = (struct aoe_hdr *) skb_mac_header(skb);
321 		memset(h, 0, sizeof *h + sizeof *ch);
322 
323 		memset(h->dst, 0xff, sizeof h->dst);
324 		memcpy(h->src, ifp->dev_addr, sizeof h->src);
325 		h->type = __constant_cpu_to_be16(ETH_P_AOE);
326 		h->verfl = AOE_HVER;
327 		h->major = cpu_to_be16(aoemajor);
328 		h->minor = aoeminor;
329 		h->cmd = AOECMD_CFG;
330 
331 cont:
332 		dev_put(ifp);
333 	}
334 	read_unlock(&dev_base_lock);
335 }
336 
337 static void
338 resend(struct aoedev *d, struct aoetgt *t, struct frame *f)
339 {
340 	struct sk_buff *skb;
341 	struct aoe_hdr *h;
342 	struct aoe_atahdr *ah;
343 	char buf[128];
344 	u32 n;
345 
346 	ifrotate(t);
347 	n = newtag(t);
348 	skb = f->skb;
349 	h = (struct aoe_hdr *) skb_mac_header(skb);
350 	ah = (struct aoe_atahdr *) (h+1);
351 
352 	snprintf(buf, sizeof buf,
353 		"%15s e%ld.%d oldtag=%08x@%08lx newtag=%08x s=%pm d=%pm nout=%d\n",
354 		"retransmit", d->aoemajor, d->aoeminor, f->tag, jiffies, n,
355 		h->src, h->dst, t->nout);
356 	aoechr_error(buf);
357 
358 	f->tag = n;
359 	h->tag = cpu_to_be32(n);
360 	memcpy(h->dst, t->addr, sizeof h->dst);
361 	memcpy(h->src, t->ifp->nd->dev_addr, sizeof h->src);
362 
363 	switch (ah->cmdstat) {
364 	default:
365 		break;
366 	case ATA_CMD_PIO_READ:
367 	case ATA_CMD_PIO_READ_EXT:
368 	case ATA_CMD_PIO_WRITE:
369 	case ATA_CMD_PIO_WRITE_EXT:
370 		put_lba(ah, f->lba);
371 
372 		n = f->bcnt;
373 		if (n > DEFAULTBCNT)
374 			n = DEFAULTBCNT;
375 		ah->scnt = n >> 9;
376 		if (ah->aflags & AOEAFL_WRITE) {
377 			skb_fill_page_desc(skb, 0, virt_to_page(f->bufaddr),
378 				offset_in_page(f->bufaddr), n);
379 			skb->len = sizeof *h + sizeof *ah + n;
380 			skb->data_len = n;
381 		}
382 	}
383 	skb->dev = t->ifp->nd;
384 	skb = skb_clone(skb, GFP_ATOMIC);
385 	if (skb == NULL)
386 		return;
387 	__skb_queue_tail(&d->sendq, skb);
388 }
389 
390 static int
391 tsince(int tag)
392 {
393 	int n;
394 
395 	n = jiffies & 0xffff;
396 	n -= tag & 0xffff;
397 	if (n < 0)
398 		n += 1<<16;
399 	return n;
400 }
401 
402 static struct aoeif *
403 getif(struct aoetgt *t, struct net_device *nd)
404 {
405 	struct aoeif *p, *e;
406 
407 	p = t->ifs;
408 	e = p + NAOEIFS;
409 	for (; p < e; p++)
410 		if (p->nd == nd)
411 			return p;
412 	return NULL;
413 }
414 
415 static struct aoeif *
416 addif(struct aoetgt *t, struct net_device *nd)
417 {
418 	struct aoeif *p;
419 
420 	p = getif(t, NULL);
421 	if (!p)
422 		return NULL;
423 	p->nd = nd;
424 	p->maxbcnt = DEFAULTBCNT;
425 	p->lost = 0;
426 	p->lostjumbo = 0;
427 	return p;
428 }
429 
430 static void
431 ejectif(struct aoetgt *t, struct aoeif *ifp)
432 {
433 	struct aoeif *e;
434 	ulong n;
435 
436 	e = t->ifs + NAOEIFS - 1;
437 	n = (e - ifp) * sizeof *ifp;
438 	memmove(ifp, ifp+1, n);
439 	e->nd = NULL;
440 }
441 
442 static int
443 sthtith(struct aoedev *d)
444 {
445 	struct frame *f, *e, *nf;
446 	struct sk_buff *skb;
447 	struct aoetgt *ht = *d->htgt;
448 
449 	f = ht->frames;
450 	e = f + ht->nframes;
451 	for (; f < e; f++) {
452 		if (f->tag == FREETAG)
453 			continue;
454 		nf = freeframe(d);
455 		if (!nf)
456 			return 0;
457 		skb = nf->skb;
458 		*nf = *f;
459 		f->skb = skb;
460 		f->tag = FREETAG;
461 		nf->waited = 0;
462 		ht->nout--;
463 		(*d->tgt)->nout++;
464 		resend(d, *d->tgt, nf);
465 	}
466 	/* he's clean, he's useless.  take away his interfaces */
467 	memset(ht->ifs, 0, sizeof ht->ifs);
468 	d->htgt = NULL;
469 	return 1;
470 }
471 
472 static inline unsigned char
473 ata_scnt(unsigned char *packet) {
474 	struct aoe_hdr *h;
475 	struct aoe_atahdr *ah;
476 
477 	h = (struct aoe_hdr *) packet;
478 	ah = (struct aoe_atahdr *) (h+1);
479 	return ah->scnt;
480 }
481 
482 static void
483 rexmit_timer(ulong vp)
484 {
485 	struct sk_buff_head queue;
486 	struct aoedev *d;
487 	struct aoetgt *t, **tt, **te;
488 	struct aoeif *ifp;
489 	struct frame *f, *e;
490 	register long timeout;
491 	ulong flags, n;
492 
493 	d = (struct aoedev *) vp;
494 
495 	/* timeout is always ~150% of the moving average */
496 	timeout = d->rttavg;
497 	timeout += timeout >> 1;
498 
499 	spin_lock_irqsave(&d->lock, flags);
500 
501 	if (d->flags & DEVFL_TKILL) {
502 		spin_unlock_irqrestore(&d->lock, flags);
503 		return;
504 	}
505 	tt = d->targets;
506 	te = tt + NTARGETS;
507 	for (; tt < te && *tt; tt++) {
508 		t = *tt;
509 		f = t->frames;
510 		e = f + t->nframes;
511 		for (; f < e; f++) {
512 			if (f->tag == FREETAG
513 			|| tsince(f->tag) < timeout)
514 				continue;
515 			n = f->waited += timeout;
516 			n /= HZ;
517 			if (n > aoe_deadsecs) {
518 				/* waited too long.  device failure. */
519 				aoedev_downdev(d);
520 				break;
521 			}
522 
523 			if (n > HELPWAIT /* see if another target can help */
524 			&& (tt != d->targets || d->targets[1]))
525 				d->htgt = tt;
526 
527 			if (t->nout == t->maxout) {
528 				if (t->maxout > 1)
529 					t->maxout--;
530 				t->lastwadj = jiffies;
531 			}
532 
533 			ifp = getif(t, f->skb->dev);
534 			if (ifp && ++ifp->lost > (t->nframes << 1)
535 			&& (ifp != t->ifs || t->ifs[1].nd)) {
536 				ejectif(t, ifp);
537 				ifp = NULL;
538 			}
539 
540 			if (ata_scnt(skb_mac_header(f->skb)) > DEFAULTBCNT / 512
541 			&& ifp && ++ifp->lostjumbo > (t->nframes << 1)
542 			&& ifp->maxbcnt != DEFAULTBCNT) {
543 				printk(KERN_INFO
544 					"aoe: e%ld.%d: "
545 					"too many lost jumbo on "
546 					"%s:%pm - "
547 					"falling back to %d frames.\n",
548 					d->aoemajor, d->aoeminor,
549 					ifp->nd->name, t->addr,
550 					DEFAULTBCNT);
551 				ifp->maxbcnt = 0;
552 			}
553 			resend(d, t, f);
554 		}
555 
556 		/* window check */
557 		if (t->nout == t->maxout
558 		&& t->maxout < t->nframes
559 		&& (jiffies - t->lastwadj)/HZ > 10) {
560 			t->maxout++;
561 			t->lastwadj = jiffies;
562 		}
563 	}
564 
565 	if (!skb_queue_empty(&d->sendq)) {
566 		n = d->rttavg <<= 1;
567 		if (n > MAXTIMER)
568 			d->rttavg = MAXTIMER;
569 	}
570 
571 	if (d->flags & DEVFL_KICKME || d->htgt) {
572 		d->flags &= ~DEVFL_KICKME;
573 		aoecmd_work(d);
574 	}
575 
576 	__skb_queue_head_init(&queue);
577 	skb_queue_splice_init(&d->sendq, &queue);
578 
579 	d->timer.expires = jiffies + TIMERTICK;
580 	add_timer(&d->timer);
581 
582 	spin_unlock_irqrestore(&d->lock, flags);
583 
584 	aoenet_xmit(&queue);
585 }
586 
587 /* enters with d->lock held */
588 void
589 aoecmd_work(struct aoedev *d)
590 {
591 	struct buf *buf;
592 loop:
593 	if (d->htgt && !sthtith(d))
594 		return;
595 	if (d->inprocess == NULL) {
596 		if (list_empty(&d->bufq))
597 			return;
598 		buf = container_of(d->bufq.next, struct buf, bufs);
599 		list_del(d->bufq.next);
600 		d->inprocess = buf;
601 	}
602 	if (aoecmd_ata_rw(d))
603 		goto loop;
604 }
605 
606 /* this function performs work that has been deferred until sleeping is OK
607  */
608 void
609 aoecmd_sleepwork(struct work_struct *work)
610 {
611 	struct aoedev *d = container_of(work, struct aoedev, work);
612 
613 	if (d->flags & DEVFL_GDALLOC)
614 		aoeblk_gdalloc(d);
615 
616 	if (d->flags & DEVFL_NEWSIZE) {
617 		struct block_device *bd;
618 		unsigned long flags;
619 		u64 ssize;
620 
621 		ssize = get_capacity(d->gd);
622 		bd = bdget_disk(d->gd, 0);
623 
624 		if (bd) {
625 			mutex_lock(&bd->bd_inode->i_mutex);
626 			i_size_write(bd->bd_inode, (loff_t)ssize<<9);
627 			mutex_unlock(&bd->bd_inode->i_mutex);
628 			bdput(bd);
629 		}
630 		spin_lock_irqsave(&d->lock, flags);
631 		d->flags |= DEVFL_UP;
632 		d->flags &= ~DEVFL_NEWSIZE;
633 		spin_unlock_irqrestore(&d->lock, flags);
634 	}
635 }
636 
637 static void
638 ataid_complete(struct aoedev *d, struct aoetgt *t, unsigned char *id)
639 {
640 	u64 ssize;
641 	u16 n;
642 
643 	/* word 83: command set supported */
644 	n = get_unaligned_le16(&id[83 << 1]);
645 
646 	/* word 86: command set/feature enabled */
647 	n |= get_unaligned_le16(&id[86 << 1]);
648 
649 	if (n & (1<<10)) {	/* bit 10: LBA 48 */
650 		d->flags |= DEVFL_EXT;
651 
652 		/* word 100: number lba48 sectors */
653 		ssize = get_unaligned_le64(&id[100 << 1]);
654 
655 		/* set as in ide-disk.c:init_idedisk_capacity */
656 		d->geo.cylinders = ssize;
657 		d->geo.cylinders /= (255 * 63);
658 		d->geo.heads = 255;
659 		d->geo.sectors = 63;
660 	} else {
661 		d->flags &= ~DEVFL_EXT;
662 
663 		/* number lba28 sectors */
664 		ssize = get_unaligned_le32(&id[60 << 1]);
665 
666 		/* NOTE: obsolete in ATA 6 */
667 		d->geo.cylinders = get_unaligned_le16(&id[54 << 1]);
668 		d->geo.heads = get_unaligned_le16(&id[55 << 1]);
669 		d->geo.sectors = get_unaligned_le16(&id[56 << 1]);
670 	}
671 
672 	if (d->ssize != ssize)
673 		printk(KERN_INFO
674 			"aoe: %pm e%ld.%d v%04x has %llu sectors\n",
675 			t->addr,
676 			d->aoemajor, d->aoeminor,
677 			d->fw_ver, (long long)ssize);
678 	d->ssize = ssize;
679 	d->geo.start = 0;
680 	if (d->flags & (DEVFL_GDALLOC|DEVFL_NEWSIZE))
681 		return;
682 	if (d->gd != NULL) {
683 		set_capacity(d->gd, ssize);
684 		d->flags |= DEVFL_NEWSIZE;
685 	} else
686 		d->flags |= DEVFL_GDALLOC;
687 	schedule_work(&d->work);
688 }
689 
690 static void
691 calc_rttavg(struct aoedev *d, int rtt)
692 {
693 	register long n;
694 
695 	n = rtt;
696 	if (n < 0) {
697 		n = -rtt;
698 		if (n < MINTIMER)
699 			n = MINTIMER;
700 		else if (n > MAXTIMER)
701 			n = MAXTIMER;
702 		d->mintimer += (n - d->mintimer) >> 1;
703 	} else if (n < d->mintimer)
704 		n = d->mintimer;
705 	else if (n > MAXTIMER)
706 		n = MAXTIMER;
707 
708 	/* g == .25; cf. Congestion Avoidance and Control, Jacobson & Karels; 1988 */
709 	n -= d->rttavg;
710 	d->rttavg += n >> 2;
711 }
712 
713 static struct aoetgt *
714 gettgt(struct aoedev *d, char *addr)
715 {
716 	struct aoetgt **t, **e;
717 
718 	t = d->targets;
719 	e = t + NTARGETS;
720 	for (; t < e && *t; t++)
721 		if (memcmp((*t)->addr, addr, sizeof((*t)->addr)) == 0)
722 			return *t;
723 	return NULL;
724 }
725 
726 static inline void
727 diskstats(struct gendisk *disk, struct bio *bio, ulong duration, sector_t sector)
728 {
729 	unsigned long n_sect = bio->bi_size >> 9;
730 	const int rw = bio_data_dir(bio);
731 	struct hd_struct *part;
732 	int cpu;
733 
734 	cpu = part_stat_lock();
735 	part = disk_map_sector_rcu(disk, sector);
736 
737 	part_stat_inc(cpu, part, ios[rw]);
738 	part_stat_add(cpu, part, ticks[rw], duration);
739 	part_stat_add(cpu, part, sectors[rw], n_sect);
740 	part_stat_add(cpu, part, io_ticks, duration);
741 
742 	part_stat_unlock();
743 }
744 
745 void
746 aoecmd_ata_rsp(struct sk_buff *skb)
747 {
748 	struct sk_buff_head queue;
749 	struct aoedev *d;
750 	struct aoe_hdr *hin, *hout;
751 	struct aoe_atahdr *ahin, *ahout;
752 	struct frame *f;
753 	struct buf *buf;
754 	struct aoetgt *t;
755 	struct aoeif *ifp;
756 	register long n;
757 	ulong flags;
758 	char ebuf[128];
759 	u16 aoemajor;
760 
761 	hin = (struct aoe_hdr *) skb_mac_header(skb);
762 	aoemajor = get_unaligned_be16(&hin->major);
763 	d = aoedev_by_aoeaddr(aoemajor, hin->minor);
764 	if (d == NULL) {
765 		snprintf(ebuf, sizeof ebuf, "aoecmd_ata_rsp: ata response "
766 			"for unknown device %d.%d\n",
767 			 aoemajor, hin->minor);
768 		aoechr_error(ebuf);
769 		return;
770 	}
771 
772 	spin_lock_irqsave(&d->lock, flags);
773 
774 	n = get_unaligned_be32(&hin->tag);
775 	t = gettgt(d, hin->src);
776 	if (t == NULL) {
777 		printk(KERN_INFO "aoe: can't find target e%ld.%d:%pm\n",
778 			d->aoemajor, d->aoeminor, hin->src);
779 		spin_unlock_irqrestore(&d->lock, flags);
780 		return;
781 	}
782 	f = getframe(t, n);
783 	if (f == NULL) {
784 		calc_rttavg(d, -tsince(n));
785 		spin_unlock_irqrestore(&d->lock, flags);
786 		snprintf(ebuf, sizeof ebuf,
787 			"%15s e%d.%d    tag=%08x@%08lx\n",
788 			"unexpected rsp",
789 			get_unaligned_be16(&hin->major),
790 			hin->minor,
791 			get_unaligned_be32(&hin->tag),
792 			jiffies);
793 		aoechr_error(ebuf);
794 		return;
795 	}
796 
797 	calc_rttavg(d, tsince(f->tag));
798 
799 	ahin = (struct aoe_atahdr *) (hin+1);
800 	hout = (struct aoe_hdr *) skb_mac_header(f->skb);
801 	ahout = (struct aoe_atahdr *) (hout+1);
802 	buf = f->buf;
803 
804 	if (ahin->cmdstat & 0xa9) {	/* these bits cleared on success */
805 		printk(KERN_ERR
806 			"aoe: ata error cmd=%2.2Xh stat=%2.2Xh from e%ld.%d\n",
807 			ahout->cmdstat, ahin->cmdstat,
808 			d->aoemajor, d->aoeminor);
809 		if (buf)
810 			buf->flags |= BUFFL_FAIL;
811 	} else {
812 		if (d->htgt && t == *d->htgt) /* I'll help myself, thank you. */
813 			d->htgt = NULL;
814 		n = ahout->scnt << 9;
815 		switch (ahout->cmdstat) {
816 		case ATA_CMD_PIO_READ:
817 		case ATA_CMD_PIO_READ_EXT:
818 			if (skb->len - sizeof *hin - sizeof *ahin < n) {
819 				printk(KERN_ERR
820 					"aoe: %s.  skb->len=%d need=%ld\n",
821 					"runt data size in read", skb->len, n);
822 				/* fail frame f?  just returning will rexmit. */
823 				spin_unlock_irqrestore(&d->lock, flags);
824 				return;
825 			}
826 			memcpy(f->bufaddr, ahin+1, n);
827 		case ATA_CMD_PIO_WRITE:
828 		case ATA_CMD_PIO_WRITE_EXT:
829 			ifp = getif(t, skb->dev);
830 			if (ifp) {
831 				ifp->lost = 0;
832 				if (n > DEFAULTBCNT)
833 					ifp->lostjumbo = 0;
834 			}
835 			if (f->bcnt -= n) {
836 				f->lba += n >> 9;
837 				f->bufaddr += n;
838 				resend(d, t, f);
839 				goto xmit;
840 			}
841 			break;
842 		case ATA_CMD_ID_ATA:
843 			if (skb->len - sizeof *hin - sizeof *ahin < 512) {
844 				printk(KERN_INFO
845 					"aoe: runt data size in ataid.  skb->len=%d\n",
846 					skb->len);
847 				spin_unlock_irqrestore(&d->lock, flags);
848 				return;
849 			}
850 			ataid_complete(d, t, (char *) (ahin+1));
851 			break;
852 		default:
853 			printk(KERN_INFO
854 				"aoe: unrecognized ata command %2.2Xh for %d.%d\n",
855 				ahout->cmdstat,
856 				get_unaligned_be16(&hin->major),
857 				hin->minor);
858 		}
859 	}
860 
861 	if (buf && --buf->nframesout == 0 && buf->resid == 0) {
862 		diskstats(d->gd, buf->bio, jiffies - buf->stime, buf->sector);
863 		n = (buf->flags & BUFFL_FAIL) ? -EIO : 0;
864 		bio_endio(buf->bio, n);
865 		mempool_free(buf, d->bufpool);
866 	}
867 
868 	f->buf = NULL;
869 	f->tag = FREETAG;
870 	t->nout--;
871 
872 	aoecmd_work(d);
873 xmit:
874 	__skb_queue_head_init(&queue);
875 	skb_queue_splice_init(&d->sendq, &queue);
876 
877 	spin_unlock_irqrestore(&d->lock, flags);
878 	aoenet_xmit(&queue);
879 }
880 
881 void
882 aoecmd_cfg(ushort aoemajor, unsigned char aoeminor)
883 {
884 	struct sk_buff_head queue;
885 
886 	__skb_queue_head_init(&queue);
887 	aoecmd_cfg_pkts(aoemajor, aoeminor, &queue);
888 	aoenet_xmit(&queue);
889 }
890 
891 struct sk_buff *
892 aoecmd_ata_id(struct aoedev *d)
893 {
894 	struct aoe_hdr *h;
895 	struct aoe_atahdr *ah;
896 	struct frame *f;
897 	struct sk_buff *skb;
898 	struct aoetgt *t;
899 
900 	f = freeframe(d);
901 	if (f == NULL)
902 		return NULL;
903 
904 	t = *d->tgt;
905 
906 	/* initialize the headers & frame */
907 	skb = f->skb;
908 	h = (struct aoe_hdr *) skb_mac_header(skb);
909 	ah = (struct aoe_atahdr *) (h+1);
910 	skb_put(skb, sizeof *h + sizeof *ah);
911 	memset(h, 0, skb->len);
912 	f->tag = aoehdr_atainit(d, t, h);
913 	t->nout++;
914 	f->waited = 0;
915 
916 	/* set up ata header */
917 	ah->scnt = 1;
918 	ah->cmdstat = ATA_CMD_ID_ATA;
919 	ah->lba3 = 0xa0;
920 
921 	skb->dev = t->ifp->nd;
922 
923 	d->rttavg = MAXTIMER;
924 	d->timer.function = rexmit_timer;
925 
926 	return skb_clone(skb, GFP_ATOMIC);
927 }
928 
929 static struct aoetgt *
930 addtgt(struct aoedev *d, char *addr, ulong nframes)
931 {
932 	struct aoetgt *t, **tt, **te;
933 	struct frame *f, *e;
934 
935 	tt = d->targets;
936 	te = tt + NTARGETS;
937 	for (; tt < te && *tt; tt++)
938 		;
939 
940 	if (tt == te) {
941 		printk(KERN_INFO
942 			"aoe: device addtgt failure; too many targets\n");
943 		return NULL;
944 	}
945 	t = kcalloc(1, sizeof *t, GFP_ATOMIC);
946 	f = kcalloc(nframes, sizeof *f, GFP_ATOMIC);
947 	if (!t || !f) {
948 		kfree(f);
949 		kfree(t);
950 		printk(KERN_INFO "aoe: cannot allocate memory to add target\n");
951 		return NULL;
952 	}
953 
954 	t->nframes = nframes;
955 	t->frames = f;
956 	e = f + nframes;
957 	for (; f < e; f++)
958 		f->tag = FREETAG;
959 	memcpy(t->addr, addr, sizeof t->addr);
960 	t->ifp = t->ifs;
961 	t->maxout = t->nframes;
962 	return *tt = t;
963 }
964 
965 void
966 aoecmd_cfg_rsp(struct sk_buff *skb)
967 {
968 	struct aoedev *d;
969 	struct aoe_hdr *h;
970 	struct aoe_cfghdr *ch;
971 	struct aoetgt *t;
972 	struct aoeif *ifp;
973 	ulong flags, sysminor, aoemajor;
974 	struct sk_buff *sl;
975 	u16 n;
976 
977 	h = (struct aoe_hdr *) skb_mac_header(skb);
978 	ch = (struct aoe_cfghdr *) (h+1);
979 
980 	/*
981 	 * Enough people have their dip switches set backwards to
982 	 * warrant a loud message for this special case.
983 	 */
984 	aoemajor = get_unaligned_be16(&h->major);
985 	if (aoemajor == 0xfff) {
986 		printk(KERN_ERR "aoe: Warning: shelf address is all ones.  "
987 			"Check shelf dip switches.\n");
988 		return;
989 	}
990 
991 	sysminor = SYSMINOR(aoemajor, h->minor);
992 	if (sysminor * AOE_PARTITIONS + AOE_PARTITIONS > MINORMASK) {
993 		printk(KERN_INFO "aoe: e%ld.%d: minor number too large\n",
994 			aoemajor, (int) h->minor);
995 		return;
996 	}
997 
998 	n = be16_to_cpu(ch->bufcnt);
999 	if (n > aoe_maxout)	/* keep it reasonable */
1000 		n = aoe_maxout;
1001 
1002 	d = aoedev_by_sysminor_m(sysminor);
1003 	if (d == NULL) {
1004 		printk(KERN_INFO "aoe: device sysminor_m failure\n");
1005 		return;
1006 	}
1007 
1008 	spin_lock_irqsave(&d->lock, flags);
1009 
1010 	t = gettgt(d, h->src);
1011 	if (!t) {
1012 		t = addtgt(d, h->src, n);
1013 		if (!t) {
1014 			spin_unlock_irqrestore(&d->lock, flags);
1015 			return;
1016 		}
1017 	}
1018 	ifp = getif(t, skb->dev);
1019 	if (!ifp) {
1020 		ifp = addif(t, skb->dev);
1021 		if (!ifp) {
1022 			printk(KERN_INFO
1023 				"aoe: device addif failure; "
1024 				"too many interfaces?\n");
1025 			spin_unlock_irqrestore(&d->lock, flags);
1026 			return;
1027 		}
1028 	}
1029 	if (ifp->maxbcnt) {
1030 		n = ifp->nd->mtu;
1031 		n -= sizeof (struct aoe_hdr) + sizeof (struct aoe_atahdr);
1032 		n /= 512;
1033 		if (n > ch->scnt)
1034 			n = ch->scnt;
1035 		n = n ? n * 512 : DEFAULTBCNT;
1036 		if (n != ifp->maxbcnt) {
1037 			printk(KERN_INFO
1038 				"aoe: e%ld.%d: setting %d%s%s:%pm\n",
1039 				d->aoemajor, d->aoeminor, n,
1040 				" byte data frames on ", ifp->nd->name,
1041 				t->addr);
1042 			ifp->maxbcnt = n;
1043 		}
1044 	}
1045 
1046 	/* don't change users' perspective */
1047 	if (d->nopen) {
1048 		spin_unlock_irqrestore(&d->lock, flags);
1049 		return;
1050 	}
1051 	d->fw_ver = be16_to_cpu(ch->fwver);
1052 
1053 	sl = aoecmd_ata_id(d);
1054 
1055 	spin_unlock_irqrestore(&d->lock, flags);
1056 
1057 	if (sl) {
1058 		struct sk_buff_head queue;
1059 		__skb_queue_head_init(&queue);
1060 		__skb_queue_tail(&queue, sl);
1061 		aoenet_xmit(&queue);
1062 	}
1063 }
1064 
1065 void
1066 aoecmd_cleanslate(struct aoedev *d)
1067 {
1068 	struct aoetgt **t, **te;
1069 	struct aoeif *p, *e;
1070 
1071 	d->mintimer = MINTIMER;
1072 
1073 	t = d->targets;
1074 	te = t + NTARGETS;
1075 	for (; t < te && *t; t++) {
1076 		(*t)->maxout = (*t)->nframes;
1077 		p = (*t)->ifs;
1078 		e = p + NAOEIFS;
1079 		for (; p < e; p++) {
1080 			p->lostjumbo = 0;
1081 			p->lost = 0;
1082 			p->maxbcnt = DEFAULTBCNT;
1083 		}
1084 	}
1085 }
1086