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