1 /* Copyright (c) 2007 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 <linux/moduleparam.h>
13 #include <net/net_namespace.h>
14 #include <asm/unaligned.h>
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.");
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.");
26 static struct sk_buff *
31 skb = alloc_skb(len, GFP_ATOMIC);
33 skb_reset_mac_header(skb);
34 skb_reset_network_header(skb);
35 skb->protocol = __constant_htons(ETH_P_AOE);
37 skb->next = skb->prev = NULL;
39 /* tell the network layer not to perform IP checksums
40 * or to get the NIC to do it
42 skb->ip_summed = CHECKSUM_NONE;
48 getframe(struct aoetgt *t, int tag)
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."
66 newtag(struct aoetgt *t)
71 return n |= (++t->lasttag & 0x7fff) << 16;
75 aoehdr_atainit(struct aoedev *d, struct aoetgt *t, struct aoe_hdr *h)
77 u32 host_tag = newtag(t);
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);
83 h->major = cpu_to_be16(d->aoemajor);
84 h->minor = d->aoeminor;
86 h->tag = cpu_to_be32(host_tag);
92 put_lba(struct aoe_atahdr *ah, sector_t lba)
103 ifrotate(struct aoetgt *t)
106 if (t->ifp >= &t->ifs[NAOEIFS] || t->ifp->nd == NULL)
108 if (t->ifp->nd == NULL) {
109 printk(KERN_INFO "aoe: no interface to rotate to\n");
115 skb_pool_put(struct aoedev *d, struct sk_buff *skb)
117 __skb_queue_tail(&d->skbpool, skb);
120 static struct sk_buff *
121 skb_pool_get(struct aoedev *d)
123 struct sk_buff *skb = skb_peek(&d->skbpool);
125 if (skb && atomic_read(&skb_shinfo(skb)->dataref) == 1) {
126 __skb_unlink(skb, &d->skbpool);
129 if (skb_queue_len(&d->skbpool) < NSKBPOOLMAX &&
130 (skb = new_skb(ETH_ZLEN)))
136 /* freeframe is where we do our load balancing so it's a little hairy. */
137 static struct frame *
138 freeframe(struct aoedev *d)
140 struct frame *f, *e, *rf;
144 if (d->targets[0] == NULL) { /* shouldn't happen, but I'm paranoid */
145 printk(KERN_ERR "aoe: NULL TARGETS!\n");
150 if (t >= &d->targets[NTARGETS] || !*t)
153 if ((*t)->nout < (*t)->maxout
158 e = f + (*t)->nframes;
160 if (f->tag != FREETAG)
164 && !(f->skb = skb = new_skb(ETH_ZLEN)))
166 if (atomic_read(&skb_shinfo(skb)->dataref)
172 gotone: skb_shinfo(skb)->nr_frags = skb->data_len = 0;
178 /* Work can be done, but the network layer is
179 holding our precious packets. Try to grab
180 one from the pool. */
182 if (f == NULL) { /* more paranoia */
184 "aoe: freeframe: %s.\n",
185 "unexpected null rf");
186 d->flags |= DEVFL_KICKME;
189 skb = skb_pool_get(d);
191 skb_pool_put(d, f->skb);
197 d->flags |= DEVFL_KICKME;
199 if (t == d->tgt) /* we've looped and found nada */
202 if (t >= &d->targets[NTARGETS] || !*t)
209 aoecmd_ata_rw(struct aoedev *d)
213 struct aoe_atahdr *ah;
219 char writebit, extbit;
230 bcnt = t->ifp->maxbcnt;
233 if (bcnt > buf->bv_resid)
234 bcnt = buf->bv_resid;
235 /* initialize the headers & frame */
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);
245 f->bufaddr = page_address(bv->bv_page) + buf->bv_off;
247 f->lba = buf->sector;
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;
257 ah->lba3 |= 0xe0; /* LBA bit + obsolete 0xa0 */
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;
263 skb->data_len = bcnt;
270 ah->cmdstat = WIN_READ | writebit | extbit;
272 /* mark all tracking fields and load out */
273 buf->nframesout += 1;
275 buf->bv_resid -= bcnt;
277 buf->sector += bcnt >> 9;
278 if (buf->resid == 0) {
280 } else if (buf->bv_resid == 0) {
282 buf->bv_resid = bv->bv_len;
283 WARN_ON(buf->bv_resid == 0);
284 buf->bv_off = bv->bv_offset;
287 skb->dev = t->ifp->nd;
288 skb = skb_clone(skb, GFP_ATOMIC);
290 __skb_queue_tail(&d->sendq, skb);
294 /* some callers cannot sleep, and they can call this function,
295 * transmitting the packets later, when interrupts are on
298 aoecmd_cfg_pkts(ushort aoemajor, unsigned char aoeminor, struct sk_buff_head *queue)
301 struct aoe_cfghdr *ch;
303 struct net_device *ifp;
305 read_lock(&dev_base_lock);
306 for_each_netdev(&init_net, ifp) {
308 if (!is_aoe_netif(ifp))
311 skb = new_skb(sizeof *h + sizeof *ch);
313 printk(KERN_INFO "aoe: skb alloc failure\n");
316 skb_put(skb, sizeof *h + sizeof *ch);
318 __skb_queue_tail(queue, skb);
319 h = (struct aoe_hdr *) skb_mac_header(skb);
320 memset(h, 0, sizeof *h + sizeof *ch);
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);
326 h->major = cpu_to_be16(aoemajor);
333 read_unlock(&dev_base_lock);
337 resend(struct aoedev *d, struct aoetgt *t, struct frame *f)
341 struct aoe_atahdr *ah;
348 h = (struct aoe_hdr *) skb_mac_header(skb);
349 ah = (struct aoe_atahdr *) (h+1);
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);
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);
362 switch (ah->cmdstat) {
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;
382 skb->dev = t->ifp->nd;
383 skb = skb_clone(skb, GFP_ATOMIC);
386 __skb_queue_tail(&d->sendq, skb);
394 n = jiffies & 0xffff;
401 static struct aoeif *
402 getif(struct aoetgt *t, struct net_device *nd)
414 static struct aoeif *
415 addif(struct aoetgt *t, struct net_device *nd)
423 p->maxbcnt = DEFAULTBCNT;
430 ejectif(struct aoetgt *t, struct aoeif *ifp)
435 e = t->ifs + NAOEIFS - 1;
436 n = (e - ifp) * sizeof *ifp;
437 memmove(ifp, ifp+1, n);
442 sthtith(struct aoedev *d)
444 struct frame *f, *e, *nf;
446 struct aoetgt *ht = *d->htgt;
451 if (f->tag == FREETAG)
463 resend(d, *d->tgt, nf);
465 /* he's clean, he's useless. take away his interfaces */
466 memset(ht->ifs, 0, sizeof ht->ifs);
471 static inline unsigned char
472 ata_scnt(unsigned char *packet) {
474 struct aoe_atahdr *ah;
476 h = (struct aoe_hdr *) packet;
477 ah = (struct aoe_atahdr *) (h+1);
482 rexmit_timer(ulong vp)
484 struct sk_buff_head queue;
486 struct aoetgt *t, **tt, **te;
489 register long timeout;
492 d = (struct aoedev *) vp;
494 /* timeout is always ~150% of the moving average */
496 timeout += timeout >> 1;
498 spin_lock_irqsave(&d->lock, flags);
500 if (d->flags & DEVFL_TKILL) {
501 spin_unlock_irqrestore(&d->lock, flags);
506 for (; tt < te && *tt; tt++) {
511 if (f->tag == FREETAG
512 || tsince(f->tag) < timeout)
514 n = f->waited += timeout;
516 if (n > aoe_deadsecs) {
517 /* waited too long. device failure. */
522 if (n > HELPWAIT /* see if another target can help */
523 && (tt != d->targets || d->targets[1]))
526 if (t->nout == t->maxout) {
529 t->lastwadj = jiffies;
532 ifp = getif(t, f->skb->dev);
533 if (ifp && ++ifp->lost > (t->nframes << 1)
534 && (ifp != t->ifs || t->ifs[1].nd)) {
539 if (ata_scnt(skb_mac_header(f->skb)) > DEFAULTBCNT / 512
540 && ifp && ++ifp->lostjumbo > (t->nframes << 1)
541 && ifp->maxbcnt != DEFAULTBCNT) {
544 "too many lost jumbo on "
546 "falling back to %d frames.\n",
547 d->aoemajor, d->aoeminor,
548 ifp->nd->name, t->addr,
556 if (t->nout == t->maxout
557 && t->maxout < t->nframes
558 && (jiffies - t->lastwadj)/HZ > 10) {
560 t->lastwadj = jiffies;
564 if (!skb_queue_empty(&d->sendq)) {
567 d->rttavg = MAXTIMER;
570 if (d->flags & DEVFL_KICKME || d->htgt) {
571 d->flags &= ~DEVFL_KICKME;
575 __skb_queue_head_init(&queue);
576 skb_queue_splice_init(&d->sendq, &queue);
578 d->timer.expires = jiffies + TIMERTICK;
579 add_timer(&d->timer);
581 spin_unlock_irqrestore(&d->lock, flags);
586 /* enters with d->lock held */
588 aoecmd_work(struct aoedev *d)
592 if (d->htgt && !sthtith(d))
594 if (d->inprocess == NULL) {
595 if (list_empty(&d->bufq))
597 buf = container_of(d->bufq.next, struct buf, bufs);
598 list_del(d->bufq.next);
601 if (aoecmd_ata_rw(d))
605 /* this function performs work that has been deferred until sleeping is OK
608 aoecmd_sleepwork(struct work_struct *work)
610 struct aoedev *d = container_of(work, struct aoedev, work);
612 if (d->flags & DEVFL_GDALLOC)
615 if (d->flags & DEVFL_NEWSIZE) {
616 struct block_device *bd;
620 ssize = get_capacity(d->gd);
621 bd = bdget_disk(d->gd, 0);
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);
629 spin_lock_irqsave(&d->lock, flags);
630 d->flags |= DEVFL_UP;
631 d->flags &= ~DEVFL_NEWSIZE;
632 spin_unlock_irqrestore(&d->lock, flags);
637 ataid_complete(struct aoedev *d, struct aoetgt *t, unsigned char *id)
642 /* word 83: command set supported */
643 n = get_unaligned_le16(&id[83 << 1]);
645 /* word 86: command set/feature enabled */
646 n |= get_unaligned_le16(&id[86 << 1]);
648 if (n & (1<<10)) { /* bit 10: LBA 48 */
649 d->flags |= DEVFL_EXT;
651 /* word 100: number lba48 sectors */
652 ssize = get_unaligned_le64(&id[100 << 1]);
654 /* set as in ide-disk.c:init_idedisk_capacity */
655 d->geo.cylinders = ssize;
656 d->geo.cylinders /= (255 * 63);
660 d->flags &= ~DEVFL_EXT;
662 /* number lba28 sectors */
663 ssize = get_unaligned_le32(&id[60 << 1]);
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]);
671 if (d->ssize != ssize)
673 "aoe: %pm e%ld.%d v%04x has %llu sectors\n",
675 d->aoemajor, d->aoeminor,
676 d->fw_ver, (long long)ssize);
679 if (d->flags & (DEVFL_GDALLOC|DEVFL_NEWSIZE))
682 set_capacity(d->gd, ssize);
683 d->flags |= DEVFL_NEWSIZE;
685 d->flags |= DEVFL_GDALLOC;
686 schedule_work(&d->work);
690 calc_rttavg(struct aoedev *d, int rtt)
699 else if (n > MAXTIMER)
701 d->mintimer += (n - d->mintimer) >> 1;
702 } else if (n < d->mintimer)
704 else if (n > MAXTIMER)
707 /* g == .25; cf. Congestion Avoidance and Control, Jacobson & Karels; 1988 */
712 static struct aoetgt *
713 gettgt(struct aoedev *d, char *addr)
715 struct aoetgt **t, **e;
719 for (; t < e && *t; t++)
720 if (memcmp((*t)->addr, addr, sizeof((*t)->addr)) == 0)
726 diskstats(struct gendisk *disk, struct bio *bio, ulong duration, sector_t sector)
728 unsigned long n_sect = bio->bi_size >> 9;
729 const int rw = bio_data_dir(bio);
730 struct hd_struct *part;
733 cpu = part_stat_lock();
734 part = disk_map_sector_rcu(disk, sector);
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);
745 aoecmd_ata_rsp(struct sk_buff *skb)
747 struct sk_buff_head queue;
749 struct aoe_hdr *hin, *hout;
750 struct aoe_atahdr *ahin, *ahout;
760 hin = (struct aoe_hdr *) skb_mac_header(skb);
761 aoemajor = get_unaligned_be16(&hin->major);
762 d = aoedev_by_aoeaddr(aoemajor, hin->minor);
764 snprintf(ebuf, sizeof ebuf, "aoecmd_ata_rsp: ata response "
765 "for unknown device %d.%d\n",
766 aoemajor, hin->minor);
771 spin_lock_irqsave(&d->lock, flags);
773 n = get_unaligned_be32(&hin->tag);
774 t = gettgt(d, hin->src);
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);
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",
788 get_unaligned_be16(&hin->major),
790 get_unaligned_be32(&hin->tag),
796 calc_rttavg(d, tsince(f->tag));
798 ahin = (struct aoe_atahdr *) (hin+1);
799 hout = (struct aoe_hdr *) skb_mac_header(f->skb);
800 ahout = (struct aoe_atahdr *) (hout+1);
803 if (ahin->cmdstat & 0xa9) { /* these bits cleared on success */
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);
809 buf->flags |= BUFFL_FAIL;
811 if (d->htgt && t == *d->htgt) /* I'll help myself, thank you. */
813 n = ahout->scnt << 9;
814 switch (ahout->cmdstat) {
817 if (skb->len - sizeof *hin - sizeof *ahin < n) {
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);
825 memcpy(f->bufaddr, ahin+1, n);
828 ifp = getif(t, skb->dev);
842 if (skb->len - sizeof *hin - sizeof *ahin < 512) {
844 "aoe: runt data size in ataid. skb->len=%d\n",
846 spin_unlock_irqrestore(&d->lock, flags);
849 ataid_complete(d, t, (char *) (ahin+1));
853 "aoe: unrecognized ata command %2.2Xh for %d.%d\n",
855 get_unaligned_be16(&hin->major),
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);
873 __skb_queue_head_init(&queue);
874 skb_queue_splice_init(&d->sendq, &queue);
876 spin_unlock_irqrestore(&d->lock, flags);
881 aoecmd_cfg(ushort aoemajor, unsigned char aoeminor)
883 struct sk_buff_head queue;
885 __skb_queue_head_init(&queue);
886 aoecmd_cfg_pkts(aoemajor, aoeminor, &queue);
891 aoecmd_ata_id(struct aoedev *d)
894 struct aoe_atahdr *ah;
905 /* initialize the headers & frame */
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);
915 /* set up ata header */
917 ah->cmdstat = WIN_IDENTIFY;
920 skb->dev = t->ifp->nd;
922 d->rttavg = MAXTIMER;
923 d->timer.function = rexmit_timer;
925 return skb_clone(skb, GFP_ATOMIC);
928 static struct aoetgt *
929 addtgt(struct aoedev *d, char *addr, ulong nframes)
931 struct aoetgt *t, **tt, **te;
936 for (; tt < te && *tt; tt++)
941 "aoe: device addtgt failure; too many targets\n");
944 t = kcalloc(1, sizeof *t, GFP_ATOMIC);
945 f = kcalloc(nframes, sizeof *f, GFP_ATOMIC);
949 printk(KERN_INFO "aoe: cannot allocate memory to add target\n");
953 t->nframes = nframes;
958 memcpy(t->addr, addr, sizeof t->addr);
960 t->maxout = t->nframes;
965 aoecmd_cfg_rsp(struct sk_buff *skb)
969 struct aoe_cfghdr *ch;
972 ulong flags, sysminor, aoemajor;
976 h = (struct aoe_hdr *) skb_mac_header(skb);
977 ch = (struct aoe_cfghdr *) (h+1);
980 * Enough people have their dip switches set backwards to
981 * warrant a loud message for this special case.
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");
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);
997 n = be16_to_cpu(ch->bufcnt);
998 if (n > aoe_maxout) /* keep it reasonable */
1001 d = aoedev_by_sysminor_m(sysminor);
1003 printk(KERN_INFO "aoe: device sysminor_m failure\n");
1007 spin_lock_irqsave(&d->lock, flags);
1009 t = gettgt(d, h->src);
1011 t = addtgt(d, h->src, n);
1013 spin_unlock_irqrestore(&d->lock, flags);
1017 ifp = getif(t, skb->dev);
1019 ifp = addif(t, skb->dev);
1022 "aoe: device addif failure; "
1023 "too many interfaces?\n");
1024 spin_unlock_irqrestore(&d->lock, flags);
1030 n -= sizeof (struct aoe_hdr) + sizeof (struct aoe_atahdr);
1034 n = n ? n * 512 : DEFAULTBCNT;
1035 if (n != ifp->maxbcnt) {
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,
1045 /* don't change users' perspective */
1047 spin_unlock_irqrestore(&d->lock, flags);
1050 d->fw_ver = be16_to_cpu(ch->fwver);
1052 sl = aoecmd_ata_id(d);
1054 spin_unlock_irqrestore(&d->lock, flags);
1057 struct sk_buff_head queue;
1058 __skb_queue_head_init(&queue);
1059 __skb_queue_tail(&queue, sl);
1060 aoenet_xmit(&queue);
1065 aoecmd_cleanslate(struct aoedev *d)
1067 struct aoetgt **t, **te;
1068 struct aoeif *p, *e;
1070 d->mintimer = MINTIMER;
1074 for (; t < te && *t; t++) {
1075 (*t)->maxout = (*t)->nframes;
1078 for (; p < e; p++) {
1081 p->maxbcnt = DEFAULTBCNT;