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