1 /* Copyright (c) 2012 Coraid, Inc. See COPYING for GPL terms. */
4 * Filesystem request handling methods
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 <linux/workqueue.h>
16 #include <linux/kthread.h>
17 #include <net/net_namespace.h>
18 #include <asm/unaligned.h>
19 #include <linux/uio.h>
22 #define MAXIOC (8192) /* default meant to avoid most soft lockups */
24 static void ktcomplete(struct frame *, struct sk_buff *);
26 static struct buf *nextbuf(struct aoedev *);
28 static int aoe_deadsecs = 60 * 3;
29 module_param(aoe_deadsecs, int, 0644);
30 MODULE_PARM_DESC(aoe_deadsecs, "After aoe_deadsecs seconds, give up and fail dev.");
32 static int aoe_maxout = 16;
33 module_param(aoe_maxout, int, 0644);
34 MODULE_PARM_DESC(aoe_maxout,
35 "Only aoe_maxout outstanding packets for every MAC on eX.Y.");
37 static wait_queue_head_t ktiowq;
38 static struct ktstate kts;
40 /* io completion queue */
42 struct list_head head;
46 static struct sk_buff *
51 skb = alloc_skb(len, GFP_ATOMIC);
53 skb_reset_mac_header(skb);
54 skb_reset_network_header(skb);
55 skb->protocol = __constant_htons(ETH_P_AOE);
56 skb_checksum_none_assert(skb);
62 getframe(struct aoedev *d, u32 tag)
65 struct list_head *head, *pos, *nx;
69 head = &d->factive[n];
70 list_for_each_safe(pos, nx, head) {
71 f = list_entry(pos, struct frame, head);
81 * Leave the top bit clear so we have tagspace for userland.
82 * The bottom 16 bits are the xmit tick for rexmit/rttavg processing.
83 * This driver reserves tag -1 to mean "unused frame."
86 newtag(struct aoedev *d)
91 return n |= (++d->lasttag & 0x7fff) << 16;
95 aoehdr_atainit(struct aoedev *d, struct aoetgt *t, struct aoe_hdr *h)
97 u32 host_tag = newtag(d);
99 memcpy(h->src, t->ifp->nd->dev_addr, sizeof h->src);
100 memcpy(h->dst, t->addr, sizeof h->dst);
101 h->type = __constant_cpu_to_be16(ETH_P_AOE);
103 h->major = cpu_to_be16(d->aoemajor);
104 h->minor = d->aoeminor;
106 h->tag = cpu_to_be32(host_tag);
112 put_lba(struct aoe_atahdr *ah, sector_t lba)
115 ah->lba1 = lba >>= 8;
116 ah->lba2 = lba >>= 8;
117 ah->lba3 = lba >>= 8;
118 ah->lba4 = lba >>= 8;
119 ah->lba5 = lba >>= 8;
122 static struct aoeif *
123 ifrotate(struct aoetgt *t)
129 if (ifp >= &t->ifs[NAOEIFS] || ifp->nd == NULL)
137 skb_pool_put(struct aoedev *d, struct sk_buff *skb)
139 __skb_queue_tail(&d->skbpool, skb);
142 static struct sk_buff *
143 skb_pool_get(struct aoedev *d)
145 struct sk_buff *skb = skb_peek(&d->skbpool);
147 if (skb && atomic_read(&skb_shinfo(skb)->dataref) == 1) {
148 __skb_unlink(skb, &d->skbpool);
151 if (skb_queue_len(&d->skbpool) < NSKBPOOLMAX &&
152 (skb = new_skb(ETH_ZLEN)))
159 aoe_freetframe(struct frame *f)
167 list_add(&f->head, &t->ffree);
170 static struct frame *
171 newtframe(struct aoedev *d, struct aoetgt *t)
175 struct list_head *pos;
177 if (list_empty(&t->ffree)) {
178 if (t->falloc >= NSKBPOOLMAX*2)
180 f = kcalloc(1, sizeof(*f), GFP_ATOMIC);
188 f = list_entry(pos, struct frame, head);
193 f->skb = skb = new_skb(ETH_ZLEN);
195 bail: aoe_freetframe(f);
200 if (atomic_read(&skb_shinfo(skb)->dataref) != 1) {
201 skb = skb_pool_get(d);
204 skb_pool_put(d, f->skb);
208 skb->truesize -= skb->data_len;
209 skb_shinfo(skb)->nr_frags = skb->data_len = 0;
214 static struct frame *
215 newframe(struct aoedev *d)
218 struct aoetgt *t, **tt;
221 if (d->targets[0] == NULL) { /* shouldn't happen, but I'm paranoid */
222 printk(KERN_ERR "aoe: NULL TARGETS!\n");
225 tt = d->tgt; /* last used target */
228 if (tt >= &d->targets[NTARGETS] || !*tt)
232 if (t->nout < t->maxout
242 if (tt == d->tgt) /* we've looped and found nada */
247 d->flags |= DEVFL_KICKME;
253 skb_fillup(struct sk_buff *skb, struct bio_vec *bv, ulong off, ulong cnt)
258 fcnt = bv->bv_len - (off - bv->bv_offset);
261 skb_fill_page_desc(skb, frag++, bv->bv_page, off, fcnt);
271 fhash(struct frame *f)
273 struct aoedev *d = f->t->d;
276 n = f->tag % NFACTIVE;
277 list_add_tail(&f->head, &d->factive[n]);
281 aoecmd_ata_rw(struct aoedev *d)
285 struct aoe_atahdr *ah;
289 struct sk_buff_head queue;
291 char writebit, extbit;
306 if (bcnt > buf->resid)
310 f->bv_off = f->bv->bv_offset + (f->bv->bv_len - buf->bv_resid);
312 if (fbcnt < buf->bv_resid) {
313 buf->bv_resid -= fbcnt;
317 fbcnt -= buf->bv_resid;
318 buf->resid -= buf->bv_resid;
319 if (buf->resid == 0) {
324 buf->bv_resid = buf->bv->bv_len;
325 WARN_ON(buf->bv_resid == 0);
328 /* initialize the headers & frame */
330 h = (struct aoe_hdr *) skb_mac_header(skb);
331 ah = (struct aoe_atahdr *) (h+1);
332 skb_put(skb, sizeof *h + sizeof *ah);
333 memset(h, 0, skb->len);
334 f->tag = aoehdr_atainit(d, t, h);
340 f->lba = buf->sector;
342 /* set up ata header */
343 ah->scnt = bcnt >> 9;
344 put_lba(ah, buf->sector);
345 if (d->flags & DEVFL_EXT) {
346 ah->aflags |= AOEAFL_EXT;
350 ah->lba3 |= 0xe0; /* LBA bit + obsolete 0xa0 */
352 if (bio_data_dir(buf->bio) == WRITE) {
353 skb_fillup(skb, f->bv, f->bv_off, bcnt);
354 ah->aflags |= AOEAFL_WRITE;
356 skb->data_len = bcnt;
357 skb->truesize += bcnt;
364 ah->cmdstat = ATA_CMD_PIO_READ | writebit | extbit;
366 /* mark all tracking fields and load out */
367 buf->nframesout += 1;
368 buf->sector += bcnt >> 9;
370 skb->dev = t->ifp->nd;
371 skb = skb_clone(skb, GFP_ATOMIC);
373 __skb_queue_head_init(&queue);
374 __skb_queue_tail(&queue, skb);
380 /* some callers cannot sleep, and they can call this function,
381 * transmitting the packets later, when interrupts are on
384 aoecmd_cfg_pkts(ushort aoemajor, unsigned char aoeminor, struct sk_buff_head *queue)
387 struct aoe_cfghdr *ch;
389 struct net_device *ifp;
392 for_each_netdev_rcu(&init_net, ifp) {
394 if (!is_aoe_netif(ifp))
397 skb = new_skb(sizeof *h + sizeof *ch);
399 printk(KERN_INFO "aoe: skb alloc failure\n");
402 skb_put(skb, sizeof *h + sizeof *ch);
404 __skb_queue_tail(queue, skb);
405 h = (struct aoe_hdr *) skb_mac_header(skb);
406 memset(h, 0, sizeof *h + sizeof *ch);
408 memset(h->dst, 0xff, sizeof h->dst);
409 memcpy(h->src, ifp->dev_addr, sizeof h->src);
410 h->type = __constant_cpu_to_be16(ETH_P_AOE);
412 h->major = cpu_to_be16(aoemajor);
423 resend(struct aoedev *d, struct frame *f)
426 struct sk_buff_head queue;
428 struct aoe_atahdr *ah;
436 if (ifrotate(t) == NULL) {
437 /* probably can't happen, but set it up to fail anyway */
438 pr_info("aoe: resend: no interfaces to rotate to.\n");
442 h = (struct aoe_hdr *) skb_mac_header(skb);
443 ah = (struct aoe_atahdr *) (h+1);
445 snprintf(buf, sizeof buf,
446 "%15s e%ld.%d oldtag=%08x@%08lx newtag=%08x s=%pm d=%pm nout=%d\n",
447 "retransmit", d->aoemajor, d->aoeminor, f->tag, jiffies, n,
448 h->src, h->dst, t->nout);
453 h->tag = cpu_to_be32(n);
454 memcpy(h->dst, t->addr, sizeof h->dst);
455 memcpy(h->src, t->ifp->nd->dev_addr, sizeof h->src);
457 skb->dev = t->ifp->nd;
458 skb = skb_clone(skb, GFP_ATOMIC);
461 __skb_queue_head_init(&queue);
462 __skb_queue_tail(&queue, skb);
471 n = jiffies & 0xffff;
478 static struct aoeif *
479 getif(struct aoetgt *t, struct net_device *nd)
492 ejectif(struct aoetgt *t, struct aoeif *ifp)
495 struct net_device *nd;
499 e = t->ifs + NAOEIFS - 1;
500 n = (e - ifp) * sizeof *ifp;
501 memmove(ifp, ifp+1, n);
507 sthtith(struct aoedev *d)
509 struct frame *f, *nf;
510 struct list_head *nx, *pos, *head;
512 struct aoetgt *ht = d->htgt;
515 for (i = 0; i < NFACTIVE; i++) {
516 head = &d->factive[i];
517 list_for_each_safe(pos, nx, head) {
518 f = list_entry(pos, struct frame, head);
526 /* remove frame from active list */
529 /* reassign all pertinent bits to new outbound frame */
536 nf->bv_off = f->bv_off;
545 /* We've cleaned up the outstanding so take away his
546 * interfaces so he won't be used. We should remove him from
547 * the target array here, but cleaning up a target is
550 memset(ht->ifs, 0, sizeof ht->ifs);
556 rexmit_timer(ulong vp)
559 struct aoetgt *t, **tt, **te;
562 struct list_head *head, *pos, *nx;
564 register long timeout;
568 d = (struct aoedev *) vp;
570 /* timeout is always ~150% of the moving average */
572 timeout += timeout >> 1;
574 spin_lock_irqsave(&d->lock, flags);
576 if (d->flags & DEVFL_TKILL) {
577 spin_unlock_irqrestore(&d->lock, flags);
581 /* collect all frames to rexmit into flist */
582 for (i = 0; i < NFACTIVE; i++) {
583 head = &d->factive[i];
584 list_for_each_safe(pos, nx, head) {
585 f = list_entry(pos, struct frame, head);
586 if (tsince(f->tag) < timeout)
587 break; /* end of expired frames */
588 /* move to flist for later processing */
589 list_move_tail(pos, &flist);
594 te = tt + d->ntargets;
595 for (; tt < te && (t = *tt); tt++) {
596 if (t->nout == t->maxout
597 && t->maxout < t->nframes
598 && (jiffies - t->lastwadj)/HZ > 10) {
600 t->lastwadj = jiffies;
604 if (!list_empty(&flist)) { /* retransmissions necessary */
607 d->rttavg = MAXTIMER;
610 /* process expired frames */
611 while (!list_empty(&flist)) {
613 f = list_entry(pos, struct frame, head);
614 n = f->waited += timeout;
616 if (n > aoe_deadsecs) {
617 /* Waited too long. Device failure.
618 * Hang all frames on first hash bucket for downdev
621 list_splice(&flist, &d->factive[0]);
628 if (n > aoe_deadsecs/2)
629 d->htgt = t; /* see if another target can help */
631 if (t->nout == t->maxout) {
634 t->lastwadj = jiffies;
637 ifp = getif(t, f->skb->dev);
638 if (ifp && ++ifp->lost > (t->nframes << 1)
639 && (ifp != t->ifs || t->ifs[1].nd)) {
646 if ((d->flags & DEVFL_KICKME || d->htgt) && d->blkq) {
647 d->flags &= ~DEVFL_KICKME;
648 d->blkq->request_fn(d->blkq);
651 d->timer.expires = jiffies + TIMERTICK;
652 add_timer(&d->timer);
654 spin_unlock_irqrestore(&d->lock, flags);
658 rqbiocnt(struct request *r)
663 __rq_for_each_bio(bio, r)
668 /* This can be removed if we are certain that no users of the block
669 * layer will ever use zero-count pages in bios. Otherwise we have to
670 * protect against the put_page sometimes done by the network layer.
672 * See http://oss.sgi.com/archives/xfs/2007-01/msg00594.html for
675 * We cannot use get_page in the workaround, because it insists on a
676 * positive page count as a precondition. So we use _count directly.
679 bio_pageinc(struct bio *bio)
685 bio_for_each_segment(bv, bio, i) {
687 /* Non-zero page count for non-head members of
688 * compound pages is no longer allowed by the kernel,
689 * but this has never been seen here.
691 if (unlikely(PageCompound(page)))
692 if (compound_trans_head(page) != page) {
693 pr_crit("page tail used for block I/O\n");
696 atomic_inc(&page->_count);
701 bio_pagedec(struct bio *bio)
706 bio_for_each_segment(bv, bio, i)
707 atomic_dec(&bv->bv_page->_count);
711 bufinit(struct buf *buf, struct request *rq, struct bio *bio)
715 memset(buf, 0, sizeof(*buf));
718 buf->resid = bio->bi_size;
719 buf->sector = bio->bi_sector;
721 buf->bv = bv = &bio->bi_io_vec[bio->bi_idx];
722 buf->bv_resid = bv->bv_len;
723 WARN_ON(buf->bv_resid == 0);
727 nextbuf(struct aoedev *d)
730 struct request_queue *q;
736 return NULL; /* initializing */
741 rq = blk_peek_request(q);
744 blk_start_request(rq);
746 d->ip.nxbio = rq->bio;
747 rq->special = (void *) rqbiocnt(rq);
749 buf = mempool_alloc(d->bufpool, GFP_ATOMIC);
751 pr_err("aoe: nextbuf: unable to mempool_alloc!\n");
755 bufinit(buf, rq, bio);
760 return d->ip.buf = buf;
763 /* enters with d->lock held */
765 aoecmd_work(struct aoedev *d)
767 if (d->htgt && !sthtith(d))
769 while (aoecmd_ata_rw(d))
773 /* this function performs work that has been deferred until sleeping is OK
776 aoecmd_sleepwork(struct work_struct *work)
778 struct aoedev *d = container_of(work, struct aoedev, work);
779 struct block_device *bd;
782 if (d->flags & DEVFL_GDALLOC)
785 if (d->flags & DEVFL_NEWSIZE) {
786 ssize = get_capacity(d->gd);
787 bd = bdget_disk(d->gd, 0);
789 mutex_lock(&bd->bd_inode->i_mutex);
790 i_size_write(bd->bd_inode, (loff_t)ssize<<9);
791 mutex_unlock(&bd->bd_inode->i_mutex);
794 spin_lock_irq(&d->lock);
795 d->flags |= DEVFL_UP;
796 d->flags &= ~DEVFL_NEWSIZE;
797 spin_unlock_irq(&d->lock);
802 ataid_complete(struct aoedev *d, struct aoetgt *t, unsigned char *id)
807 /* word 83: command set supported */
808 n = get_unaligned_le16(&id[83 << 1]);
810 /* word 86: command set/feature enabled */
811 n |= get_unaligned_le16(&id[86 << 1]);
813 if (n & (1<<10)) { /* bit 10: LBA 48 */
814 d->flags |= DEVFL_EXT;
816 /* word 100: number lba48 sectors */
817 ssize = get_unaligned_le64(&id[100 << 1]);
819 /* set as in ide-disk.c:init_idedisk_capacity */
820 d->geo.cylinders = ssize;
821 d->geo.cylinders /= (255 * 63);
825 d->flags &= ~DEVFL_EXT;
827 /* number lba28 sectors */
828 ssize = get_unaligned_le32(&id[60 << 1]);
830 /* NOTE: obsolete in ATA 6 */
831 d->geo.cylinders = get_unaligned_le16(&id[54 << 1]);
832 d->geo.heads = get_unaligned_le16(&id[55 << 1]);
833 d->geo.sectors = get_unaligned_le16(&id[56 << 1]);
836 if (d->ssize != ssize)
838 "aoe: %pm e%ld.%d v%04x has %llu sectors\n",
840 d->aoemajor, d->aoeminor,
841 d->fw_ver, (long long)ssize);
844 if (d->flags & (DEVFL_GDALLOC|DEVFL_NEWSIZE))
847 set_capacity(d->gd, ssize);
848 d->flags |= DEVFL_NEWSIZE;
850 d->flags |= DEVFL_GDALLOC;
851 schedule_work(&d->work);
855 calc_rttavg(struct aoedev *d, int rtt)
864 else if (n > MAXTIMER)
866 d->mintimer += (n - d->mintimer) >> 1;
867 } else if (n < d->mintimer)
869 else if (n > MAXTIMER)
872 /* g == .25; cf. Congestion Avoidance and Control, Jacobson & Karels; 1988 */
877 static struct aoetgt *
878 gettgt(struct aoedev *d, char *addr)
880 struct aoetgt **t, **e;
884 for (; t < e && *t; t++)
885 if (memcmp((*t)->addr, addr, sizeof((*t)->addr)) == 0)
891 bvcpy(struct bio_vec *bv, ulong off, struct sk_buff *skb, long cnt)
897 fcnt = bv->bv_len - (off - bv->bv_offset);
900 p = page_address(bv->bv_page) + off;
901 skb_copy_bits(skb, soff, p, fcnt);
912 aoe_end_request(struct aoedev *d, struct request *rq, int fastfail)
916 struct request_queue *q;
923 bok = !fastfail && test_bit(BIO_UPTODATE, &bio->bi_flags);
924 } while (__blk_end_request(rq, bok ? 0 : -EIO, bio->bi_size));
926 /* cf. http://lkml.org/lkml/2006/10/31/28 */
932 aoe_end_buf(struct aoedev *d, struct buf *buf)
937 if (buf == d->ip.buf)
940 bio_pagedec(buf->bio);
941 mempool_free(buf, d->bufpool);
942 n = (unsigned long) rq->special;
943 rq->special = (void *) --n;
945 aoe_end_request(d, rq, 0);
949 ktiocomplete(struct frame *f)
951 struct aoe_hdr *hin, *hout;
952 struct aoe_atahdr *ahin, *ahout;
966 hout = (struct aoe_hdr *) skb_mac_header(f->skb);
967 ahout = (struct aoe_atahdr *) (hout+1);
971 goto noskb; /* just fail the buf. */
973 hin = (struct aoe_hdr *) skb->data;
974 skb_pull(skb, sizeof(*hin));
975 ahin = (struct aoe_atahdr *) skb->data;
976 skb_pull(skb, sizeof(*ahin));
977 if (ahin->cmdstat & 0xa9) { /* these bits cleared on success */
978 pr_err("aoe: ata error cmd=%2.2Xh stat=%2.2Xh from e%ld.%d\n",
979 ahout->cmdstat, ahin->cmdstat,
980 d->aoemajor, d->aoeminor);
982 clear_bit(BIO_UPTODATE, &buf->bio->bi_flags);
986 n = ahout->scnt << 9;
987 switch (ahout->cmdstat) {
988 case ATA_CMD_PIO_READ:
989 case ATA_CMD_PIO_READ_EXT:
991 pr_err("aoe: runt data size in read. skb->len=%d need=%ld\n",
993 clear_bit(BIO_UPTODATE, &buf->bio->bi_flags);
996 bvcpy(f->bv, f->bv_off, skb, n);
997 case ATA_CMD_PIO_WRITE:
998 case ATA_CMD_PIO_WRITE_EXT:
999 spin_lock_irq(&d->lock);
1000 ifp = getif(t, skb->dev);
1003 if (d->htgt == t) /* I'll help myself, thank you. */
1005 spin_unlock_irq(&d->lock);
1007 case ATA_CMD_ID_ATA:
1008 if (skb->len < 512) {
1009 pr_info("aoe: runt data size in ataid. skb->len=%d\n",
1013 if (skb_linearize(skb))
1015 spin_lock_irq(&d->lock);
1016 ataid_complete(d, t, skb->data);
1017 spin_unlock_irq(&d->lock);
1020 pr_info("aoe: unrecognized ata command %2.2Xh for %d.%d\n",
1022 be16_to_cpu(get_unaligned(&hin->major)),
1026 spin_lock_irq(&d->lock);
1030 if (buf && --buf->nframesout == 0 && buf->resid == 0)
1031 aoe_end_buf(d, buf);
1035 spin_unlock_irq(&d->lock);
1040 /* Enters with iocq.lock held.
1041 * Returns true iff responses needing processing remain.
1047 struct list_head *pos;
1050 for (i = 0; ; ++i) {
1053 if (list_empty(&iocq.head))
1055 pos = iocq.head.next;
1057 spin_unlock_irq(&iocq.lock);
1058 f = list_entry(pos, struct frame, head);
1060 spin_lock_irq(&iocq.lock);
1068 DECLARE_WAITQUEUE(wait, current);
1072 current->flags |= PF_NOFREEZE;
1073 set_user_nice(current, -10);
1074 complete(&k->rendez); /* tell spawner we're running */
1076 spin_lock_irq(k->lock);
1079 add_wait_queue(k->waitq, &wait);
1080 __set_current_state(TASK_INTERRUPTIBLE);
1082 spin_unlock_irq(k->lock);
1085 remove_wait_queue(k->waitq, &wait);
1088 } while (!kthread_should_stop());
1089 complete(&k->rendez); /* tell spawner we're stopping */
1094 aoe_ktstop(struct ktstate *k)
1096 kthread_stop(k->task);
1097 wait_for_completion(&k->rendez);
1101 aoe_ktstart(struct ktstate *k)
1103 struct task_struct *task;
1105 init_completion(&k->rendez);
1106 task = kthread_run(kthread, k, k->name);
1107 if (task == NULL || IS_ERR(task))
1110 wait_for_completion(&k->rendez); /* allow kthread to start */
1111 init_completion(&k->rendez); /* for waiting for exit later */
1115 /* pass it off to kthreads for processing */
1117 ktcomplete(struct frame *f, struct sk_buff *skb)
1122 spin_lock_irqsave(&iocq.lock, flags);
1123 list_add_tail(&f->head, &iocq.head);
1124 spin_unlock_irqrestore(&iocq.lock, flags);
1129 aoecmd_ata_rsp(struct sk_buff *skb)
1140 h = (struct aoe_hdr *) skb->data;
1141 aoemajor = be16_to_cpu(get_unaligned(&h->major));
1142 d = aoedev_by_aoeaddr(aoemajor, h->minor, 0);
1144 snprintf(ebuf, sizeof ebuf, "aoecmd_ata_rsp: ata response "
1145 "for unknown device %d.%d\n",
1146 aoemajor, h->minor);
1151 spin_lock_irqsave(&d->lock, flags);
1153 n = be32_to_cpu(get_unaligned(&h->tag));
1156 calc_rttavg(d, -tsince(n));
1157 spin_unlock_irqrestore(&d->lock, flags);
1159 snprintf(ebuf, sizeof ebuf,
1160 "%15s e%d.%d tag=%08x@%08lx\n",
1162 get_unaligned_be16(&h->major),
1164 get_unaligned_be32(&h->tag),
1170 calc_rttavg(d, tsince(f->tag));
1174 spin_unlock_irqrestore(&d->lock, flags);
1179 * Note here that we do not perform an aoedev_put, as we are
1180 * leaving this reference for the ktio to release.
1186 aoecmd_cfg(ushort aoemajor, unsigned char aoeminor)
1188 struct sk_buff_head queue;
1190 __skb_queue_head_init(&queue);
1191 aoecmd_cfg_pkts(aoemajor, aoeminor, &queue);
1192 aoenet_xmit(&queue);
1196 aoecmd_ata_id(struct aoedev *d)
1199 struct aoe_atahdr *ah;
1201 struct sk_buff *skb;
1210 /* initialize the headers & frame */
1212 h = (struct aoe_hdr *) skb_mac_header(skb);
1213 ah = (struct aoe_atahdr *) (h+1);
1214 skb_put(skb, sizeof *h + sizeof *ah);
1215 memset(h, 0, skb->len);
1216 f->tag = aoehdr_atainit(d, t, h);
1221 /* set up ata header */
1223 ah->cmdstat = ATA_CMD_ID_ATA;
1226 skb->dev = t->ifp->nd;
1228 d->rttavg = MAXTIMER;
1229 d->timer.function = rexmit_timer;
1231 return skb_clone(skb, GFP_ATOMIC);
1234 static struct aoetgt *
1235 addtgt(struct aoedev *d, char *addr, ulong nframes)
1237 struct aoetgt *t, **tt, **te;
1241 for (; tt < te && *tt; tt++)
1246 "aoe: device addtgt failure; too many targets\n");
1249 t = kzalloc(sizeof(*t), GFP_ATOMIC);
1251 printk(KERN_INFO "aoe: cannot allocate memory to add target\n");
1256 t->nframes = nframes;
1258 memcpy(t->addr, addr, sizeof t->addr);
1260 t->maxout = t->nframes;
1261 INIT_LIST_HEAD(&t->ffree);
1266 setdbcnt(struct aoedev *d)
1268 struct aoetgt **t, **e;
1273 for (; t < e && *t; t++)
1274 if (bcnt == 0 || bcnt > (*t)->minbcnt)
1275 bcnt = (*t)->minbcnt;
1276 if (bcnt != d->maxbcnt) {
1278 pr_info("aoe: e%ld.%d: setting %d byte data frames\n",
1279 d->aoemajor, d->aoeminor, bcnt);
1284 setifbcnt(struct aoetgt *t, struct net_device *nd, int bcnt)
1287 struct aoeif *p, *e;
1294 for (; p < e; p++) {
1296 break; /* end of the valid interfaces */
1298 p->bcnt = bcnt; /* we're updating */
1300 } else if (minbcnt > p->bcnt)
1301 minbcnt = p->bcnt; /* find the min interface */
1305 pr_err("aoe: device setifbcnt failure; too many interfaces.\n");
1312 t->minbcnt = minbcnt;
1317 aoecmd_cfg_rsp(struct sk_buff *skb)
1321 struct aoe_cfghdr *ch;
1323 ulong flags, aoemajor;
1325 struct sk_buff_head queue;
1329 h = (struct aoe_hdr *) skb_mac_header(skb);
1330 ch = (struct aoe_cfghdr *) (h+1);
1333 * Enough people have their dip switches set backwards to
1334 * warrant a loud message for this special case.
1336 aoemajor = get_unaligned_be16(&h->major);
1337 if (aoemajor == 0xfff) {
1338 printk(KERN_ERR "aoe: Warning: shelf address is all ones. "
1339 "Check shelf dip switches.\n");
1342 if (aoemajor == 0xffff) {
1343 pr_info("aoe: e%ld.%d: broadcast shelf number invalid\n",
1344 aoemajor, (int) h->minor);
1347 if (h->minor == 0xff) {
1348 pr_info("aoe: e%ld.%d: broadcast slot number invalid\n",
1349 aoemajor, (int) h->minor);
1353 n = be16_to_cpu(ch->bufcnt);
1354 if (n > aoe_maxout) /* keep it reasonable */
1357 d = aoedev_by_aoeaddr(aoemajor, h->minor, 1);
1359 pr_info("aoe: device allocation failure\n");
1363 spin_lock_irqsave(&d->lock, flags);
1365 t = gettgt(d, h->src);
1371 t = addtgt(d, h->src, n);
1376 n -= sizeof(struct aoe_hdr) + sizeof(struct aoe_atahdr);
1380 n = n ? n * 512 : DEFAULTBCNT;
1381 setifbcnt(t, skb->dev, n);
1383 /* don't change users' perspective */
1384 if (d->nopen == 0) {
1385 d->fw_ver = be16_to_cpu(ch->fwver);
1386 sl = aoecmd_ata_id(d);
1389 spin_unlock_irqrestore(&d->lock, flags);
1392 __skb_queue_head_init(&queue);
1393 __skb_queue_tail(&queue, sl);
1394 aoenet_xmit(&queue);
1399 aoecmd_cleanslate(struct aoedev *d)
1401 struct aoetgt **t, **te;
1403 d->mintimer = MINTIMER;
1408 for (; t < te && *t; t++)
1409 (*t)->maxout = (*t)->nframes;
1413 aoe_failbuf(struct aoedev *d, struct buf *buf)
1418 clear_bit(BIO_UPTODATE, &buf->bio->bi_flags);
1419 if (buf->nframesout == 0)
1420 aoe_end_buf(d, buf);
1424 aoe_flush_iocq(void)
1429 struct list_head *pos;
1430 struct sk_buff *skb;
1433 spin_lock_irqsave(&iocq.lock, flags);
1434 list_splice_init(&iocq.head, &flist);
1435 spin_unlock_irqrestore(&iocq.lock, flags);
1436 while (!list_empty(&flist)) {
1439 f = list_entry(pos, struct frame, head);
1442 spin_lock_irqsave(&d->lock, flags);
1444 f->buf->nframesout--;
1445 aoe_failbuf(d, f->buf);
1448 spin_unlock_irqrestore(&d->lock, flags);
1457 INIT_LIST_HEAD(&iocq.head);
1458 spin_lock_init(&iocq.lock);
1459 init_waitqueue_head(&ktiowq);
1460 kts.name = "aoe_ktio";
1462 kts.waitq = &ktiowq;
1463 kts.lock = &iocq.lock;
1464 return aoe_ktstart(&kts);