1 /* Copyright (c) 2006 Coraid, Inc. See COPYING for GPL terms. */
4 * Filesystem request handling methods
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 <asm/unaligned.h>
15 #define TIMERTICK (HZ / 10)
16 #define MINTIMER (2 * TIMERTICK)
17 #define MAXTIMER (HZ << 1)
18 #define MAXWAIT (60 * 3) /* After MAXWAIT seconds, give up and fail dev */
25 skb = alloc_skb(len, GFP_ATOMIC);
27 skb->nh.raw = skb->mac.raw = skb->data;
28 skb->protocol = __constant_htons(ETH_P_AOE);
31 memset(skb->head, 0, len);
32 skb->next = skb->prev = NULL;
34 /* tell the network layer not to perform IP checksums
35 * or to get the NIC to do it
37 skb->ip_summed = CHECKSUM_NONE;
43 getframe(struct aoedev *d, int tag)
56 * Leave the top bit clear so we have tagspace for userland.
57 * The bottom 16 bits are the xmit tick for rexmit/rttavg processing.
58 * This driver reserves tag -1 to mean "unused frame."
61 newtag(struct aoedev *d)
66 return n |= (++d->lasttag & 0x7fff) << 16;
70 aoehdr_atainit(struct aoedev *d, struct aoe_hdr *h)
72 u32 host_tag = newtag(d);
74 memcpy(h->src, d->ifp->dev_addr, sizeof h->src);
75 memcpy(h->dst, d->addr, sizeof h->dst);
76 h->type = __constant_cpu_to_be16(ETH_P_AOE);
78 h->major = cpu_to_be16(d->aoemajor);
79 h->minor = d->aoeminor;
81 h->tag = cpu_to_be32(host_tag);
87 put_lba(struct aoe_atahdr *ah, sector_t lba)
98 aoecmd_ata_rw(struct aoedev *d, struct frame *f)
101 struct aoe_atahdr *ah;
105 register sector_t sector;
106 char writebit, extbit;
113 sector = buf->sector;
114 bcnt = buf->bv_resid;
115 if (bcnt > d->maxbcnt)
118 /* initialize the headers & frame */
120 h = (struct aoe_hdr *) skb->mac.raw;
121 ah = (struct aoe_atahdr *) (h+1);
122 skb->len = sizeof *h + sizeof *ah;
123 memset(h, 0, skb->len);
124 f->tag = aoehdr_atainit(d, h);
127 f->bufaddr = buf->bufaddr;
131 /* set up ata header */
132 ah->scnt = bcnt >> 9;
134 if (d->flags & DEVFL_EXT) {
135 ah->aflags |= AOEAFL_EXT;
139 ah->lba3 |= 0xe0; /* LBA bit + obsolete 0xa0 */
142 if (bio_data_dir(buf->bio) == WRITE) {
143 skb_fill_page_desc(skb, 0, virt_to_page(f->bufaddr),
144 offset_in_page(f->bufaddr), bcnt);
145 ah->aflags |= AOEAFL_WRITE;
147 skb_shinfo(skb)->nr_frags = 0;
152 ah->cmdstat = WIN_READ | writebit | extbit;
154 /* mark all tracking fields and load out */
155 buf->nframesout += 1;
156 buf->bufaddr += bcnt;
157 buf->bv_resid -= bcnt;
158 /* dprintk("bv_resid=%ld\n", buf->bv_resid); */
160 buf->sector += bcnt >> 9;
161 if (buf->resid == 0) {
163 } else if (buf->bv_resid == 0) {
165 buf->bv_resid = buf->bv->bv_len;
166 buf->bufaddr = page_address(buf->bv->bv_page) + buf->bv->bv_offset;
173 d->sendq_tl->next = skb;
179 /* some callers cannot sleep, and they can call this function,
180 * transmitting the packets later, when interrupts are on
182 static struct sk_buff *
183 aoecmd_cfg_pkts(ushort aoemajor, unsigned char aoeminor, struct sk_buff **tail)
186 struct aoe_cfghdr *ch;
187 struct sk_buff *skb, *sl, *sl_tail;
188 struct net_device *ifp;
192 read_lock(&dev_base_lock);
193 for (ifp = dev_base; ifp; dev_put(ifp), ifp = ifp->next) {
195 if (!is_aoe_netif(ifp))
198 skb = new_skb(sizeof *h + sizeof *ch);
200 iprintk("skb alloc failure\n");
206 h = (struct aoe_hdr *) skb->mac.raw;
207 memset(h, 0, sizeof *h + sizeof *ch);
209 memset(h->dst, 0xff, sizeof h->dst);
210 memcpy(h->src, ifp->dev_addr, sizeof h->src);
211 h->type = __constant_cpu_to_be16(ETH_P_AOE);
213 h->major = cpu_to_be16(aoemajor);
220 read_unlock(&dev_base_lock);
227 /* enters with d->lock held */
229 aoecmd_work(struct aoedev *d)
234 if (d->flags & DEVFL_PAUSE) {
235 if (!aoedev_isbusy(d))
236 d->sendq_hd = aoecmd_cfg_pkts(d->aoemajor,
237 d->aoeminor, &d->sendq_tl);
242 f = getframe(d, FREETAG);
245 if (d->inprocess == NULL) {
246 if (list_empty(&d->bufq))
248 buf = container_of(d->bufq.next, struct buf, bufs);
249 list_del(d->bufq.next);
250 /*dprintk("bi_size=%ld\n", buf->bio->bi_size); */
258 rexmit(struct aoedev *d, struct frame *f)
262 struct aoe_atahdr *ah;
268 snprintf(buf, sizeof buf,
269 "%15s e%ld.%ld oldtag=%08x@%08lx newtag=%08x\n",
271 d->aoemajor, d->aoeminor, f->tag, jiffies, n);
275 h = (struct aoe_hdr *) skb->mac.raw;
276 ah = (struct aoe_atahdr *) (h+1);
278 h->tag = cpu_to_be32(n);
279 memcpy(h->dst, d->addr, sizeof h->dst);
280 memcpy(h->src, d->ifp->dev_addr, sizeof h->src);
282 n = DEFAULTBCNT / 512;
285 if (ah->aflags & AOEAFL_WRITE)
286 skb_fill_page_desc(skb, 0, virt_to_page(f->bufaddr),
287 offset_in_page(f->bufaddr), DEFAULTBCNT);
288 if (++d->lostjumbo > (d->nframes << 1))
289 if (d->maxbcnt != DEFAULTBCNT) {
290 iprintk("too many lost jumbo - using 1KB frames.\n");
291 d->maxbcnt = DEFAULTBCNT;
292 d->flags |= DEVFL_MAXBCNT;
300 d->sendq_tl->next = skb;
311 n = jiffies & 0xffff;
319 rexmit_timer(ulong vp)
324 register long timeout;
327 d = (struct aoedev *) vp;
330 /* timeout is always ~150% of the moving average */
332 timeout += timeout >> 1;
334 spin_lock_irqsave(&d->lock, flags);
336 if (d->flags & DEVFL_TKILL) {
337 spin_unlock_irqrestore(&d->lock, flags);
343 if (f->tag != FREETAG && tsince(f->tag) >= timeout) {
344 n = f->waited += timeout;
346 if (n > MAXWAIT) { /* waited too long. device failure. */
355 d->sendq_hd = d->sendq_tl = NULL;
359 d->rttavg = MAXTIMER;
362 d->timer.expires = jiffies + TIMERTICK;
363 add_timer(&d->timer);
365 spin_unlock_irqrestore(&d->lock, flags);
370 /* this function performs work that has been deferred until sleeping is OK
373 aoecmd_sleepwork(void *vp)
375 struct aoedev *d = (struct aoedev *) vp;
377 if (d->flags & DEVFL_GDALLOC)
380 if (d->flags & DEVFL_NEWSIZE) {
381 struct block_device *bd;
385 ssize = d->gd->capacity;
386 bd = bdget_disk(d->gd, 0);
389 mutex_lock(&bd->bd_inode->i_mutex);
390 i_size_write(bd->bd_inode, (loff_t)ssize<<9);
391 mutex_unlock(&bd->bd_inode->i_mutex);
394 spin_lock_irqsave(&d->lock, flags);
395 d->flags |= DEVFL_UP;
396 d->flags &= ~DEVFL_NEWSIZE;
397 spin_unlock_irqrestore(&d->lock, flags);
402 ataid_complete(struct aoedev *d, unsigned char *id)
407 /* word 83: command set supported */
408 n = le16_to_cpu(get_unaligned((__le16 *) &id[83<<1]));
410 /* word 86: command set/feature enabled */
411 n |= le16_to_cpu(get_unaligned((__le16 *) &id[86<<1]));
413 if (n & (1<<10)) { /* bit 10: LBA 48 */
414 d->flags |= DEVFL_EXT;
416 /* word 100: number lba48 sectors */
417 ssize = le64_to_cpu(get_unaligned((__le64 *) &id[100<<1]));
419 /* set as in ide-disk.c:init_idedisk_capacity */
420 d->geo.cylinders = ssize;
421 d->geo.cylinders /= (255 * 63);
425 d->flags &= ~DEVFL_EXT;
427 /* number lba28 sectors */
428 ssize = le32_to_cpu(get_unaligned((__le32 *) &id[60<<1]));
430 /* NOTE: obsolete in ATA 6 */
431 d->geo.cylinders = le16_to_cpu(get_unaligned((__le16 *) &id[54<<1]));
432 d->geo.heads = le16_to_cpu(get_unaligned((__le16 *) &id[55<<1]));
433 d->geo.sectors = le16_to_cpu(get_unaligned((__le16 *) &id[56<<1]));
436 if (d->ssize != ssize)
437 iprintk("%012llx e%lu.%lu v%04x has %llu sectors\n",
438 (unsigned long long)mac_addr(d->addr),
439 d->aoemajor, d->aoeminor,
440 d->fw_ver, (long long)ssize);
444 d->gd->capacity = ssize;
445 d->flags |= DEVFL_NEWSIZE;
447 if (d->flags & DEVFL_GDALLOC) {
448 eprintk("can't schedule work for e%lu.%lu, %s\n",
449 d->aoemajor, d->aoeminor,
450 "it's already on! This shouldn't happen.\n");
453 d->flags |= DEVFL_GDALLOC;
455 schedule_work(&d->work);
459 calc_rttavg(struct aoedev *d, int rtt)
468 else if (n > MAXTIMER)
470 d->mintimer += (n - d->mintimer) >> 1;
471 } else if (n < d->mintimer)
473 else if (n > MAXTIMER)
476 /* g == .25; cf. Congestion Avoidance and Control, Jacobson & Karels; 1988 */
482 aoecmd_ata_rsp(struct sk_buff *skb)
485 struct aoe_hdr *hin, *hout;
486 struct aoe_atahdr *ahin, *ahout;
495 hin = (struct aoe_hdr *) skb->mac.raw;
496 aoemajor = be16_to_cpu(hin->major);
497 d = aoedev_by_aoeaddr(aoemajor, hin->minor);
499 snprintf(ebuf, sizeof ebuf, "aoecmd_ata_rsp: ata response "
500 "for unknown device %d.%d\n",
501 aoemajor, hin->minor);
506 spin_lock_irqsave(&d->lock, flags);
508 n = be32_to_cpu(hin->tag);
511 calc_rttavg(d, -tsince(n));
512 spin_unlock_irqrestore(&d->lock, flags);
513 snprintf(ebuf, sizeof ebuf,
514 "%15s e%d.%d tag=%08x@%08lx\n",
516 be16_to_cpu(hin->major),
518 be32_to_cpu(hin->tag),
524 calc_rttavg(d, tsince(f->tag));
526 ahin = (struct aoe_atahdr *) (hin+1);
527 hout = (struct aoe_hdr *) f->skb->mac.raw;
528 ahout = (struct aoe_atahdr *) (hout+1);
531 if (ahout->cmdstat == WIN_IDENTIFY)
532 d->flags &= ~DEVFL_PAUSE;
533 if (ahin->cmdstat & 0xa9) { /* these bits cleared on success */
534 eprintk("ata error cmd=%2.2Xh stat=%2.2Xh from e%ld.%ld\n",
535 ahout->cmdstat, ahin->cmdstat,
536 d->aoemajor, d->aoeminor);
538 buf->flags |= BUFFL_FAIL;
540 n = ahout->scnt << 9;
541 switch (ahout->cmdstat) {
544 if (skb->len - sizeof *hin - sizeof *ahin < n) {
545 eprintk("runt data size in read. skb->len=%d\n",
547 /* fail frame f? just returning will rexmit. */
548 spin_unlock_irqrestore(&d->lock, flags);
551 memcpy(f->bufaddr, ahin+1, n);
556 put_lba(ahout, f->lba += ahout->scnt);
560 ahout->scnt = n >> 9;
561 if (ahout->aflags & AOEAFL_WRITE)
562 skb_fill_page_desc(f->skb, 0,
563 virt_to_page(f->bufaddr),
564 offset_in_page(f->bufaddr), n);
566 hout->tag = cpu_to_be32(f->tag);
570 spin_unlock_irqrestore(&d->lock, flags);
578 if (skb->len - sizeof *hin - sizeof *ahin < 512) {
579 iprintk("runt data size in ataid. skb->len=%d\n",
581 spin_unlock_irqrestore(&d->lock, flags);
584 ataid_complete(d, (char *) (ahin+1));
587 iprintk("unrecognized ata command %2.2Xh for %d.%d\n",
589 be16_to_cpu(hin->major),
595 buf->nframesout -= 1;
596 if (buf->nframesout == 0 && buf->resid == 0) {
597 unsigned long duration = jiffies - buf->start_time;
598 unsigned long n_sect = buf->bio->bi_size >> 9;
599 struct gendisk *disk = d->gd;
600 const int rw = bio_data_dir(buf->bio);
602 disk_stat_inc(disk, ios[rw]);
603 disk_stat_add(disk, ticks[rw], duration);
604 disk_stat_add(disk, sectors[rw], n_sect);
605 disk_stat_add(disk, io_ticks, duration);
606 n = (buf->flags & BUFFL_FAIL) ? -EIO : 0;
607 bio_endio(buf->bio, buf->bio->bi_size, n);
608 mempool_free(buf, d->bufpool);
617 d->sendq_hd = d->sendq_tl = NULL;
619 spin_unlock_irqrestore(&d->lock, flags);
624 aoecmd_cfg(ushort aoemajor, unsigned char aoeminor)
628 sl = aoecmd_cfg_pkts(aoemajor, aoeminor, NULL);
634 * Since we only call this in one place (and it only prepares one frame)
635 * we just return the skb. Usually we'd chain it up to the aoedev sendq.
637 static struct sk_buff *
638 aoecmd_ata_id(struct aoedev *d)
641 struct aoe_atahdr *ah;
645 f = getframe(d, FREETAG);
647 eprintk("can't get a frame. This shouldn't happen.\n");
651 /* initialize the headers & frame */
653 h = (struct aoe_hdr *) skb->mac.raw;
654 ah = (struct aoe_atahdr *) (h+1);
655 skb->len = sizeof *h + sizeof *ah;
656 memset(h, 0, skb->len);
657 f->tag = aoehdr_atainit(d, h);
660 /* set up ata header */
662 ah->cmdstat = WIN_IDENTIFY;
668 d->rttavg = MAXTIMER;
669 d->timer.function = rexmit_timer;
675 aoecmd_cfg_rsp(struct sk_buff *skb)
679 struct aoe_cfghdr *ch;
680 ulong flags, sysminor, aoemajor;
682 enum { MAXFRAMES = 16 };
685 h = (struct aoe_hdr *) skb->mac.raw;
686 ch = (struct aoe_cfghdr *) (h+1);
689 * Enough people have their dip switches set backwards to
690 * warrant a loud message for this special case.
692 aoemajor = be16_to_cpu(h->major);
693 if (aoemajor == 0xfff) {
694 eprintk("Warning: shelf address is all ones. "
695 "Check shelf dip switches.\n");
699 sysminor = SYSMINOR(aoemajor, h->minor);
700 if (sysminor * AOE_PARTITIONS + AOE_PARTITIONS > MINORMASK) {
701 iprintk("e%ld.%d: minor number too large\n",
702 aoemajor, (int) h->minor);
706 n = be16_to_cpu(ch->bufcnt);
707 if (n > MAXFRAMES) /* keep it reasonable */
710 d = aoedev_by_sysminor_m(sysminor, n);
712 iprintk("device sysminor_m failure\n");
716 spin_lock_irqsave(&d->lock, flags);
718 /* permit device to migrate mac and network interface */
720 memcpy(d->addr, h->src, sizeof d->addr);
721 if (!(d->flags & DEVFL_MAXBCNT)) {
723 n -= sizeof (struct aoe_hdr) + sizeof (struct aoe_atahdr);
727 d->maxbcnt = n ? n * 512 : DEFAULTBCNT;
730 /* don't change users' perspective */
731 if (d->nopen && !(d->flags & DEVFL_PAUSE)) {
732 spin_unlock_irqrestore(&d->lock, flags);
735 d->flags |= DEVFL_PAUSE; /* force pause */
736 d->mintimer = MINTIMER;
737 d->fw_ver = be16_to_cpu(ch->fwver);
739 /* check for already outstanding ataid */
740 sl = aoedev_isbusy(d) == 0 ? aoecmd_ata_id(d) : NULL;
742 spin_unlock_irqrestore(&d->lock, flags);