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_deferred(struct aoedev *d, u32 tag)
64 struct list_head *head, *pos, *nx;
68 list_for_each_safe(pos, nx, head) {
69 f = list_entry(pos, struct frame, head);
79 getframe(struct aoedev *d, u32 tag)
82 struct list_head *head, *pos, *nx;
86 head = &d->factive[n];
87 list_for_each_safe(pos, nx, head) {
88 f = list_entry(pos, struct frame, head);
98 * Leave the top bit clear so we have tagspace for userland.
99 * The bottom 16 bits are the xmit tick for rexmit/rttavg processing.
100 * This driver reserves tag -1 to mean "unused frame."
103 newtag(struct aoedev *d)
107 n = jiffies & 0xffff;
108 return n |= (++d->lasttag & 0x7fff) << 16;
112 aoehdr_atainit(struct aoedev *d, struct aoetgt *t, struct aoe_hdr *h)
114 u32 host_tag = newtag(d);
116 memcpy(h->src, t->ifp->nd->dev_addr, sizeof h->src);
117 memcpy(h->dst, t->addr, sizeof h->dst);
118 h->type = __constant_cpu_to_be16(ETH_P_AOE);
120 h->major = cpu_to_be16(d->aoemajor);
121 h->minor = d->aoeminor;
123 h->tag = cpu_to_be32(host_tag);
129 put_lba(struct aoe_atahdr *ah, sector_t lba)
132 ah->lba1 = lba >>= 8;
133 ah->lba2 = lba >>= 8;
134 ah->lba3 = lba >>= 8;
135 ah->lba4 = lba >>= 8;
136 ah->lba5 = lba >>= 8;
139 static struct aoeif *
140 ifrotate(struct aoetgt *t)
146 if (ifp >= &t->ifs[NAOEIFS] || ifp->nd == NULL)
154 skb_pool_put(struct aoedev *d, struct sk_buff *skb)
156 __skb_queue_tail(&d->skbpool, skb);
159 static struct sk_buff *
160 skb_pool_get(struct aoedev *d)
162 struct sk_buff *skb = skb_peek(&d->skbpool);
164 if (skb && atomic_read(&skb_shinfo(skb)->dataref) == 1) {
165 __skb_unlink(skb, &d->skbpool);
168 if (skb_queue_len(&d->skbpool) < NSKBPOOLMAX &&
169 (skb = new_skb(ETH_ZLEN)))
176 aoe_freetframe(struct frame *f)
184 list_add(&f->head, &t->ffree);
187 static struct frame *
188 newtframe(struct aoedev *d, struct aoetgt *t)
192 struct list_head *pos;
194 if (list_empty(&t->ffree)) {
195 if (t->falloc >= NSKBPOOLMAX*2)
197 f = kcalloc(1, sizeof(*f), GFP_ATOMIC);
205 f = list_entry(pos, struct frame, head);
210 f->skb = skb = new_skb(ETH_ZLEN);
212 bail: aoe_freetframe(f);
217 if (atomic_read(&skb_shinfo(skb)->dataref) != 1) {
218 skb = skb_pool_get(d);
221 skb_pool_put(d, f->skb);
225 skb->truesize -= skb->data_len;
226 skb_shinfo(skb)->nr_frags = skb->data_len = 0;
231 static struct frame *
232 newframe(struct aoedev *d)
235 struct aoetgt *t, **tt;
238 if (d->targets[0] == NULL) { /* shouldn't happen, but I'm paranoid */
239 printk(KERN_ERR "aoe: NULL TARGETS!\n");
242 tt = d->tgt; /* last used target */
245 if (tt >= &d->targets[NTARGETS] || !*tt)
249 if (t->nout < t->maxout
259 if (tt == d->tgt) /* we've looped and found nada */
264 d->flags |= DEVFL_KICKME;
270 skb_fillup(struct sk_buff *skb, struct bio_vec *bv, ulong off, ulong cnt)
275 fcnt = bv->bv_len - (off - bv->bv_offset);
278 skb_fill_page_desc(skb, frag++, bv->bv_page, off, fcnt);
288 fhash(struct frame *f)
290 struct aoedev *d = f->t->d;
293 n = f->tag % NFACTIVE;
294 list_add_tail(&f->head, &d->factive[n]);
298 aoecmd_ata_rw(struct aoedev *d)
302 struct aoe_atahdr *ah;
306 struct sk_buff_head queue;
308 char writebit, extbit;
323 if (bcnt > buf->resid)
327 f->bv_off = f->bv->bv_offset + (f->bv->bv_len - buf->bv_resid);
329 if (fbcnt < buf->bv_resid) {
330 buf->bv_resid -= fbcnt;
334 fbcnt -= buf->bv_resid;
335 buf->resid -= buf->bv_resid;
336 if (buf->resid == 0) {
341 buf->bv_resid = buf->bv->bv_len;
342 WARN_ON(buf->bv_resid == 0);
345 /* initialize the headers & frame */
347 h = (struct aoe_hdr *) skb_mac_header(skb);
348 ah = (struct aoe_atahdr *) (h+1);
349 skb_put(skb, sizeof *h + sizeof *ah);
350 memset(h, 0, skb->len);
351 f->tag = aoehdr_atainit(d, t, h);
357 f->lba = buf->sector;
359 /* set up ata header */
360 ah->scnt = bcnt >> 9;
361 put_lba(ah, buf->sector);
362 if (d->flags & DEVFL_EXT) {
363 ah->aflags |= AOEAFL_EXT;
367 ah->lba3 |= 0xe0; /* LBA bit + obsolete 0xa0 */
369 if (bio_data_dir(buf->bio) == WRITE) {
370 skb_fillup(skb, f->bv, f->bv_off, bcnt);
371 ah->aflags |= AOEAFL_WRITE;
373 skb->data_len = bcnt;
374 skb->truesize += bcnt;
381 ah->cmdstat = ATA_CMD_PIO_READ | writebit | extbit;
383 /* mark all tracking fields and load out */
384 buf->nframesout += 1;
385 buf->sector += bcnt >> 9;
387 skb->dev = t->ifp->nd;
388 skb = skb_clone(skb, GFP_ATOMIC);
390 __skb_queue_head_init(&queue);
391 __skb_queue_tail(&queue, skb);
397 /* some callers cannot sleep, and they can call this function,
398 * transmitting the packets later, when interrupts are on
401 aoecmd_cfg_pkts(ushort aoemajor, unsigned char aoeminor, struct sk_buff_head *queue)
404 struct aoe_cfghdr *ch;
406 struct net_device *ifp;
409 for_each_netdev_rcu(&init_net, ifp) {
411 if (!is_aoe_netif(ifp))
414 skb = new_skb(sizeof *h + sizeof *ch);
416 printk(KERN_INFO "aoe: skb alloc failure\n");
419 skb_put(skb, sizeof *h + sizeof *ch);
421 __skb_queue_tail(queue, skb);
422 h = (struct aoe_hdr *) skb_mac_header(skb);
423 memset(h, 0, sizeof *h + sizeof *ch);
425 memset(h->dst, 0xff, sizeof h->dst);
426 memcpy(h->src, ifp->dev_addr, sizeof h->src);
427 h->type = __constant_cpu_to_be16(ETH_P_AOE);
429 h->major = cpu_to_be16(aoemajor);
440 resend(struct aoedev *d, struct frame *f)
443 struct sk_buff_head queue;
445 struct aoe_atahdr *ah;
453 if (ifrotate(t) == NULL) {
454 /* probably can't happen, but set it up to fail anyway */
455 pr_info("aoe: resend: no interfaces to rotate to.\n");
459 h = (struct aoe_hdr *) skb_mac_header(skb);
460 ah = (struct aoe_atahdr *) (h+1);
462 snprintf(buf, sizeof buf,
463 "%15s e%ld.%d oldtag=%08x@%08lx newtag=%08x s=%pm d=%pm nout=%d\n",
464 "retransmit", d->aoemajor, d->aoeminor, f->tag, jiffies, n,
465 h->src, h->dst, t->nout);
470 h->tag = cpu_to_be32(n);
471 memcpy(h->dst, t->addr, sizeof h->dst);
472 memcpy(h->src, t->ifp->nd->dev_addr, sizeof h->src);
474 skb->dev = t->ifp->nd;
475 skb = skb_clone(skb, GFP_ATOMIC);
478 __skb_queue_head_init(&queue);
479 __skb_queue_tail(&queue, skb);
488 n = jiffies & 0xffff;
495 static struct aoeif *
496 getif(struct aoetgt *t, struct net_device *nd)
509 ejectif(struct aoetgt *t, struct aoeif *ifp)
512 struct net_device *nd;
516 e = t->ifs + NAOEIFS - 1;
517 n = (e - ifp) * sizeof *ifp;
518 memmove(ifp, ifp+1, n);
524 sthtith(struct aoedev *d)
526 struct frame *f, *nf;
527 struct list_head *nx, *pos, *head;
529 struct aoetgt *ht = d->htgt;
532 for (i = 0; i < NFACTIVE; i++) {
533 head = &d->factive[i];
534 list_for_each_safe(pos, nx, head) {
535 f = list_entry(pos, struct frame, head);
543 /* remove frame from active list */
546 /* reassign all pertinent bits to new outbound frame */
553 nf->bv_off = f->bv_off;
562 /* We've cleaned up the outstanding so take away his
563 * interfaces so he won't be used. We should remove him from
564 * the target array here, but cleaning up a target is
567 memset(ht->ifs, 0, sizeof ht->ifs);
573 rexmit_deferred(struct aoedev *d)
577 struct list_head *pos, *nx, *head;
580 list_for_each_safe(pos, nx, head) {
581 f = list_entry(pos, struct frame, head);
583 if (t->nout >= t->maxout)
592 rexmit_timer(ulong vp)
598 struct list_head *head, *pos, *nx;
600 register long timeout;
604 d = (struct aoedev *) vp;
606 /* timeout based on observed timings and variations */
607 timeout = 2 * d->rttavg >> RTTSCALE;
608 timeout += 8 * d->rttdev >> RTTDSCALE;
612 spin_lock_irqsave(&d->lock, flags);
614 if (d->flags & DEVFL_TKILL) {
615 spin_unlock_irqrestore(&d->lock, flags);
619 /* collect all frames to rexmit into flist */
620 for (i = 0; i < NFACTIVE; i++) {
621 head = &d->factive[i];
622 list_for_each_safe(pos, nx, head) {
623 f = list_entry(pos, struct frame, head);
624 if (tsince(f->tag) < timeout)
625 break; /* end of expired frames */
626 /* move to flist for later processing */
627 list_move_tail(pos, &flist);
631 /* process expired frames */
632 while (!list_empty(&flist)) {
634 f = list_entry(pos, struct frame, head);
635 n = f->waited += tsince(f->tag);
637 if (n > aoe_deadsecs) {
638 /* Waited too long. Device failure.
639 * Hang all frames on first hash bucket for downdev
642 list_splice(&flist, &d->factive[0]);
648 if (n > aoe_deadsecs/2)
649 d->htgt = t; /* see if another target can help */
651 if (t->maxout != 1) {
652 t->ssthresh = t->maxout / 2;
656 ifp = getif(t, f->skb->dev);
657 if (ifp && ++ifp->lost > (t->nframes << 1)
658 && (ifp != t->ifs || t->ifs[1].nd)) {
662 list_move_tail(pos, &d->rexmitq);
668 if ((d->flags & DEVFL_KICKME || d->htgt) && d->blkq) {
669 d->flags &= ~DEVFL_KICKME;
670 d->blkq->request_fn(d->blkq);
673 d->timer.expires = jiffies + TIMERTICK;
674 add_timer(&d->timer);
676 spin_unlock_irqrestore(&d->lock, flags);
680 rqbiocnt(struct request *r)
685 __rq_for_each_bio(bio, r)
690 /* This can be removed if we are certain that no users of the block
691 * layer will ever use zero-count pages in bios. Otherwise we have to
692 * protect against the put_page sometimes done by the network layer.
694 * See http://oss.sgi.com/archives/xfs/2007-01/msg00594.html for
697 * We cannot use get_page in the workaround, because it insists on a
698 * positive page count as a precondition. So we use _count directly.
701 bio_pageinc(struct bio *bio)
707 bio_for_each_segment(bv, bio, i) {
709 /* Non-zero page count for non-head members of
710 * compound pages is no longer allowed by the kernel,
711 * but this has never been seen here.
713 if (unlikely(PageCompound(page)))
714 if (compound_trans_head(page) != page) {
715 pr_crit("page tail used for block I/O\n");
718 atomic_inc(&page->_count);
723 bio_pagedec(struct bio *bio)
728 bio_for_each_segment(bv, bio, i)
729 atomic_dec(&bv->bv_page->_count);
733 bufinit(struct buf *buf, struct request *rq, struct bio *bio)
737 memset(buf, 0, sizeof(*buf));
740 buf->resid = bio->bi_size;
741 buf->sector = bio->bi_sector;
743 buf->bv = bv = &bio->bi_io_vec[bio->bi_idx];
744 buf->bv_resid = bv->bv_len;
745 WARN_ON(buf->bv_resid == 0);
749 nextbuf(struct aoedev *d)
752 struct request_queue *q;
758 return NULL; /* initializing */
763 rq = blk_peek_request(q);
766 blk_start_request(rq);
768 d->ip.nxbio = rq->bio;
769 rq->special = (void *) rqbiocnt(rq);
771 buf = mempool_alloc(d->bufpool, GFP_ATOMIC);
773 pr_err("aoe: nextbuf: unable to mempool_alloc!\n");
777 bufinit(buf, rq, bio);
782 return d->ip.buf = buf;
785 /* enters with d->lock held */
787 aoecmd_work(struct aoedev *d)
789 if (d->htgt && !sthtith(d))
792 while (aoecmd_ata_rw(d))
796 /* this function performs work that has been deferred until sleeping is OK
799 aoecmd_sleepwork(struct work_struct *work)
801 struct aoedev *d = container_of(work, struct aoedev, work);
802 struct block_device *bd;
805 if (d->flags & DEVFL_GDALLOC)
808 if (d->flags & DEVFL_NEWSIZE) {
809 ssize = get_capacity(d->gd);
810 bd = bdget_disk(d->gd, 0);
812 mutex_lock(&bd->bd_inode->i_mutex);
813 i_size_write(bd->bd_inode, (loff_t)ssize<<9);
814 mutex_unlock(&bd->bd_inode->i_mutex);
817 spin_lock_irq(&d->lock);
818 d->flags |= DEVFL_UP;
819 d->flags &= ~DEVFL_NEWSIZE;
820 spin_unlock_irq(&d->lock);
825 ata_ident_fixstring(u16 *id, int ns)
831 *id++ = s >> 8 | s << 8;
836 ataid_complete(struct aoedev *d, struct aoetgt *t, unsigned char *id)
841 /* word 83: command set supported */
842 n = get_unaligned_le16(&id[83 << 1]);
844 /* word 86: command set/feature enabled */
845 n |= get_unaligned_le16(&id[86 << 1]);
847 if (n & (1<<10)) { /* bit 10: LBA 48 */
848 d->flags |= DEVFL_EXT;
850 /* word 100: number lba48 sectors */
851 ssize = get_unaligned_le64(&id[100 << 1]);
853 /* set as in ide-disk.c:init_idedisk_capacity */
854 d->geo.cylinders = ssize;
855 d->geo.cylinders /= (255 * 63);
859 d->flags &= ~DEVFL_EXT;
861 /* number lba28 sectors */
862 ssize = get_unaligned_le32(&id[60 << 1]);
864 /* NOTE: obsolete in ATA 6 */
865 d->geo.cylinders = get_unaligned_le16(&id[54 << 1]);
866 d->geo.heads = get_unaligned_le16(&id[55 << 1]);
867 d->geo.sectors = get_unaligned_le16(&id[56 << 1]);
870 ata_ident_fixstring((u16 *) &id[10<<1], 10); /* serial */
871 ata_ident_fixstring((u16 *) &id[23<<1], 4); /* firmware */
872 ata_ident_fixstring((u16 *) &id[27<<1], 20); /* model */
873 memcpy(d->ident, id, sizeof(d->ident));
875 if (d->ssize != ssize)
877 "aoe: %pm e%ld.%d v%04x has %llu sectors\n",
879 d->aoemajor, d->aoeminor,
880 d->fw_ver, (long long)ssize);
883 if (d->flags & (DEVFL_GDALLOC|DEVFL_NEWSIZE))
886 set_capacity(d->gd, ssize);
887 d->flags |= DEVFL_NEWSIZE;
889 d->flags |= DEVFL_GDALLOC;
890 schedule_work(&d->work);
894 calc_rttavg(struct aoedev *d, struct aoetgt *t, int rtt)
900 /* cf. Congestion Avoidance and Control, Jacobson & Karels, 1988 */
901 n -= d->rttavg >> RTTSCALE;
905 n -= d->rttdev >> RTTDSCALE;
908 if (!t || t->maxout >= t->nframes)
910 if (t->maxout < t->ssthresh)
912 else if (t->nout == t->maxout && t->next_cwnd-- == 0) {
914 t->next_cwnd = t->maxout;
918 static struct aoetgt *
919 gettgt(struct aoedev *d, char *addr)
921 struct aoetgt **t, **e;
925 for (; t < e && *t; t++)
926 if (memcmp((*t)->addr, addr, sizeof((*t)->addr)) == 0)
932 bvcpy(struct bio_vec *bv, ulong off, struct sk_buff *skb, long cnt)
938 fcnt = bv->bv_len - (off - bv->bv_offset);
941 p = page_address(bv->bv_page) + off;
942 skb_copy_bits(skb, soff, p, fcnt);
953 aoe_end_request(struct aoedev *d, struct request *rq, int fastfail)
957 struct request_queue *q;
964 bok = !fastfail && test_bit(BIO_UPTODATE, &bio->bi_flags);
965 } while (__blk_end_request(rq, bok ? 0 : -EIO, bio->bi_size));
967 /* cf. http://lkml.org/lkml/2006/10/31/28 */
973 aoe_end_buf(struct aoedev *d, struct buf *buf)
978 if (buf == d->ip.buf)
981 bio_pagedec(buf->bio);
982 mempool_free(buf, d->bufpool);
983 n = (unsigned long) rq->special;
984 rq->special = (void *) --n;
986 aoe_end_request(d, rq, 0);
990 ktiocomplete(struct frame *f)
992 struct aoe_hdr *hin, *hout;
993 struct aoe_atahdr *ahin, *ahout;
1007 hout = (struct aoe_hdr *) skb_mac_header(f->skb);
1008 ahout = (struct aoe_atahdr *) (hout+1);
1012 goto noskb; /* just fail the buf. */
1014 hin = (struct aoe_hdr *) skb->data;
1015 skb_pull(skb, sizeof(*hin));
1016 ahin = (struct aoe_atahdr *) skb->data;
1017 skb_pull(skb, sizeof(*ahin));
1018 if (ahin->cmdstat & 0xa9) { /* these bits cleared on success */
1019 pr_err("aoe: ata error cmd=%2.2Xh stat=%2.2Xh from e%ld.%d\n",
1020 ahout->cmdstat, ahin->cmdstat,
1021 d->aoemajor, d->aoeminor);
1023 clear_bit(BIO_UPTODATE, &buf->bio->bi_flags);
1027 n = ahout->scnt << 9;
1028 switch (ahout->cmdstat) {
1029 case ATA_CMD_PIO_READ:
1030 case ATA_CMD_PIO_READ_EXT:
1032 pr_err("aoe: runt data size in read. skb->len=%d need=%ld\n",
1034 clear_bit(BIO_UPTODATE, &buf->bio->bi_flags);
1037 bvcpy(f->bv, f->bv_off, skb, n);
1038 case ATA_CMD_PIO_WRITE:
1039 case ATA_CMD_PIO_WRITE_EXT:
1040 spin_lock_irq(&d->lock);
1041 ifp = getif(t, skb->dev);
1044 if (d->htgt == t) /* I'll help myself, thank you. */
1046 spin_unlock_irq(&d->lock);
1048 case ATA_CMD_ID_ATA:
1049 if (skb->len < 512) {
1050 pr_info("aoe: runt data size in ataid. skb->len=%d\n",
1054 if (skb_linearize(skb))
1056 spin_lock_irq(&d->lock);
1057 ataid_complete(d, t, skb->data);
1058 spin_unlock_irq(&d->lock);
1061 pr_info("aoe: unrecognized ata command %2.2Xh for %d.%d\n",
1063 be16_to_cpu(get_unaligned(&hin->major)),
1067 spin_lock_irq(&d->lock);
1071 if (buf && --buf->nframesout == 0 && buf->resid == 0)
1072 aoe_end_buf(d, buf);
1076 spin_unlock_irq(&d->lock);
1081 /* Enters with iocq.lock held.
1082 * Returns true iff responses needing processing remain.
1088 struct list_head *pos;
1091 for (i = 0; ; ++i) {
1094 if (list_empty(&iocq.head))
1096 pos = iocq.head.next;
1098 spin_unlock_irq(&iocq.lock);
1099 f = list_entry(pos, struct frame, head);
1101 spin_lock_irq(&iocq.lock);
1109 DECLARE_WAITQUEUE(wait, current);
1113 current->flags |= PF_NOFREEZE;
1114 set_user_nice(current, -10);
1115 complete(&k->rendez); /* tell spawner we're running */
1117 spin_lock_irq(k->lock);
1120 add_wait_queue(k->waitq, &wait);
1121 __set_current_state(TASK_INTERRUPTIBLE);
1123 spin_unlock_irq(k->lock);
1126 remove_wait_queue(k->waitq, &wait);
1129 } while (!kthread_should_stop());
1130 complete(&k->rendez); /* tell spawner we're stopping */
1135 aoe_ktstop(struct ktstate *k)
1137 kthread_stop(k->task);
1138 wait_for_completion(&k->rendez);
1142 aoe_ktstart(struct ktstate *k)
1144 struct task_struct *task;
1146 init_completion(&k->rendez);
1147 task = kthread_run(kthread, k, k->name);
1148 if (task == NULL || IS_ERR(task))
1151 wait_for_completion(&k->rendez); /* allow kthread to start */
1152 init_completion(&k->rendez); /* for waiting for exit later */
1156 /* pass it off to kthreads for processing */
1158 ktcomplete(struct frame *f, struct sk_buff *skb)
1163 spin_lock_irqsave(&iocq.lock, flags);
1164 list_add_tail(&f->head, &iocq.head);
1165 spin_unlock_irqrestore(&iocq.lock, flags);
1170 aoecmd_ata_rsp(struct sk_buff *skb)
1180 h = (struct aoe_hdr *) skb->data;
1181 aoemajor = be16_to_cpu(get_unaligned(&h->major));
1182 d = aoedev_by_aoeaddr(aoemajor, h->minor, 0);
1184 snprintf(ebuf, sizeof ebuf, "aoecmd_ata_rsp: ata response "
1185 "for unknown device %d.%d\n",
1186 aoemajor, h->minor);
1191 spin_lock_irqsave(&d->lock, flags);
1193 n = be32_to_cpu(get_unaligned(&h->tag));
1196 calc_rttavg(d, f->t, tsince(n));
1199 f = getframe_deferred(d, n);
1201 calc_rttavg(d, NULL, tsince(n));
1203 calc_rttavg(d, NULL, tsince(n));
1204 spin_unlock_irqrestore(&d->lock, flags);
1206 snprintf(ebuf, sizeof(ebuf),
1207 "%15s e%d.%d tag=%08x@%08lx s=%pm d=%pm\n",
1209 get_unaligned_be16(&h->major),
1211 get_unaligned_be32(&h->tag),
1221 spin_unlock_irqrestore(&d->lock, flags);
1226 * Note here that we do not perform an aoedev_put, as we are
1227 * leaving this reference for the ktio to release.
1233 aoecmd_cfg(ushort aoemajor, unsigned char aoeminor)
1235 struct sk_buff_head queue;
1237 __skb_queue_head_init(&queue);
1238 aoecmd_cfg_pkts(aoemajor, aoeminor, &queue);
1239 aoenet_xmit(&queue);
1243 aoecmd_ata_id(struct aoedev *d)
1246 struct aoe_atahdr *ah;
1248 struct sk_buff *skb;
1257 /* initialize the headers & frame */
1259 h = (struct aoe_hdr *) skb_mac_header(skb);
1260 ah = (struct aoe_atahdr *) (h+1);
1261 skb_put(skb, sizeof *h + sizeof *ah);
1262 memset(h, 0, skb->len);
1263 f->tag = aoehdr_atainit(d, t, h);
1268 /* set up ata header */
1270 ah->cmdstat = ATA_CMD_ID_ATA;
1273 skb->dev = t->ifp->nd;
1275 d->rttavg = RTTAVG_INIT;
1276 d->rttdev = RTTDEV_INIT;
1277 d->timer.function = rexmit_timer;
1279 return skb_clone(skb, GFP_ATOMIC);
1282 static struct aoetgt *
1283 addtgt(struct aoedev *d, char *addr, ulong nframes)
1285 struct aoetgt *t, **tt, **te;
1289 for (; tt < te && *tt; tt++)
1294 "aoe: device addtgt failure; too many targets\n");
1297 t = kzalloc(sizeof(*t), GFP_ATOMIC);
1299 printk(KERN_INFO "aoe: cannot allocate memory to add target\n");
1304 t->nframes = nframes;
1306 memcpy(t->addr, addr, sizeof t->addr);
1309 INIT_LIST_HEAD(&t->ffree);
1314 setdbcnt(struct aoedev *d)
1316 struct aoetgt **t, **e;
1321 for (; t < e && *t; t++)
1322 if (bcnt == 0 || bcnt > (*t)->minbcnt)
1323 bcnt = (*t)->minbcnt;
1324 if (bcnt != d->maxbcnt) {
1326 pr_info("aoe: e%ld.%d: setting %d byte data frames\n",
1327 d->aoemajor, d->aoeminor, bcnt);
1332 setifbcnt(struct aoetgt *t, struct net_device *nd, int bcnt)
1335 struct aoeif *p, *e;
1342 for (; p < e; p++) {
1344 break; /* end of the valid interfaces */
1346 p->bcnt = bcnt; /* we're updating */
1348 } else if (minbcnt > p->bcnt)
1349 minbcnt = p->bcnt; /* find the min interface */
1353 pr_err("aoe: device setifbcnt failure; too many interfaces.\n");
1360 t->minbcnt = minbcnt;
1365 aoecmd_cfg_rsp(struct sk_buff *skb)
1369 struct aoe_cfghdr *ch;
1371 ulong flags, aoemajor;
1373 struct sk_buff_head queue;
1377 h = (struct aoe_hdr *) skb_mac_header(skb);
1378 ch = (struct aoe_cfghdr *) (h+1);
1381 * Enough people have their dip switches set backwards to
1382 * warrant a loud message for this special case.
1384 aoemajor = get_unaligned_be16(&h->major);
1385 if (aoemajor == 0xfff) {
1386 printk(KERN_ERR "aoe: Warning: shelf address is all ones. "
1387 "Check shelf dip switches.\n");
1390 if (aoemajor == 0xffff) {
1391 pr_info("aoe: e%ld.%d: broadcast shelf number invalid\n",
1392 aoemajor, (int) h->minor);
1395 if (h->minor == 0xff) {
1396 pr_info("aoe: e%ld.%d: broadcast slot number invalid\n",
1397 aoemajor, (int) h->minor);
1401 n = be16_to_cpu(ch->bufcnt);
1402 if (n > aoe_maxout) /* keep it reasonable */
1405 d = aoedev_by_aoeaddr(aoemajor, h->minor, 1);
1407 pr_info("aoe: device allocation failure\n");
1411 spin_lock_irqsave(&d->lock, flags);
1413 t = gettgt(d, h->src);
1419 t = addtgt(d, h->src, n);
1424 n -= sizeof(struct aoe_hdr) + sizeof(struct aoe_atahdr);
1428 n = n ? n * 512 : DEFAULTBCNT;
1429 setifbcnt(t, skb->dev, n);
1431 /* don't change users' perspective */
1432 if (d->nopen == 0) {
1433 d->fw_ver = be16_to_cpu(ch->fwver);
1434 sl = aoecmd_ata_id(d);
1437 spin_unlock_irqrestore(&d->lock, flags);
1440 __skb_queue_head_init(&queue);
1441 __skb_queue_tail(&queue, sl);
1442 aoenet_xmit(&queue);
1447 aoecmd_wreset(struct aoetgt *t)
1450 t->ssthresh = t->nframes / 2;
1451 t->next_cwnd = t->nframes;
1455 aoecmd_cleanslate(struct aoedev *d)
1457 struct aoetgt **t, **te;
1459 d->rttavg = RTTAVG_INIT;
1460 d->rttdev = RTTDEV_INIT;
1465 for (; t < te && *t; t++)
1470 aoe_failbuf(struct aoedev *d, struct buf *buf)
1475 clear_bit(BIO_UPTODATE, &buf->bio->bi_flags);
1476 if (buf->nframesout == 0)
1477 aoe_end_buf(d, buf);
1481 aoe_flush_iocq(void)
1486 struct list_head *pos;
1487 struct sk_buff *skb;
1490 spin_lock_irqsave(&iocq.lock, flags);
1491 list_splice_init(&iocq.head, &flist);
1492 spin_unlock_irqrestore(&iocq.lock, flags);
1493 while (!list_empty(&flist)) {
1496 f = list_entry(pos, struct frame, head);
1499 spin_lock_irqsave(&d->lock, flags);
1501 f->buf->nframesout--;
1502 aoe_failbuf(d, f->buf);
1505 spin_unlock_irqrestore(&d->lock, flags);
1514 INIT_LIST_HEAD(&iocq.head);
1515 spin_lock_init(&iocq.lock);
1516 init_waitqueue_head(&ktiowq);
1517 kts.name = "aoe_ktio";
1519 kts.waitq = &ktiowq;
1520 kts.lock = &iocq.lock;
1521 return aoe_ktstart(&kts);