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)
19 static int aoe_deadsecs
= 60 * 3;
20 module_param(aoe_deadsecs
, int, 0644);
21 MODULE_PARM_DESC(aoe_deadsecs
, "After aoe_deadsecs seconds, give up and fail dev.");
28 skb
= alloc_skb(len
, GFP_ATOMIC
);
30 skb_reset_mac_header(skb
);
31 skb_reset_network_header(skb
);
32 skb
->protocol
= __constant_htons(ETH_P_AOE
);
34 skb
->next
= skb
->prev
= NULL
;
36 /* tell the network layer not to perform IP checksums
37 * or to get the NIC to do it
39 skb
->ip_summed
= CHECKSUM_NONE
;
45 getframe(struct aoedev
*d
, int tag
)
58 * Leave the top bit clear so we have tagspace for userland.
59 * The bottom 16 bits are the xmit tick for rexmit/rttavg processing.
60 * This driver reserves tag -1 to mean "unused frame."
63 newtag(struct aoedev
*d
)
68 return n
|= (++d
->lasttag
& 0x7fff) << 16;
72 aoehdr_atainit(struct aoedev
*d
, struct aoe_hdr
*h
)
74 u32 host_tag
= newtag(d
);
76 memcpy(h
->src
, d
->ifp
->dev_addr
, sizeof h
->src
);
77 memcpy(h
->dst
, d
->addr
, sizeof h
->dst
);
78 h
->type
= __constant_cpu_to_be16(ETH_P_AOE
);
80 h
->major
= cpu_to_be16(d
->aoemajor
);
81 h
->minor
= d
->aoeminor
;
83 h
->tag
= cpu_to_be32(host_tag
);
89 put_lba(struct aoe_atahdr
*ah
, sector_t lba
)
100 aoecmd_ata_rw(struct aoedev
*d
, struct frame
*f
)
103 struct aoe_atahdr
*ah
;
107 register sector_t sector
;
108 char writebit
, extbit
;
115 sector
= buf
->sector
;
116 bcnt
= buf
->bv_resid
;
117 if (bcnt
> d
->maxbcnt
)
120 /* initialize the headers & frame */
123 ah
= (struct aoe_atahdr
*) (h
+1);
124 skb_put(skb
, sizeof *h
+ sizeof *ah
);
125 memset(h
, 0, skb
->len
);
126 f
->tag
= aoehdr_atainit(d
, h
);
129 f
->bufaddr
= buf
->bufaddr
;
133 /* set up ata header */
134 ah
->scnt
= bcnt
>> 9;
136 if (d
->flags
& DEVFL_EXT
) {
137 ah
->aflags
|= AOEAFL_EXT
;
141 ah
->lba3
|= 0xe0; /* LBA bit + obsolete 0xa0 */
144 if (bio_data_dir(buf
->bio
) == WRITE
) {
145 skb_fill_page_desc(skb
, 0, virt_to_page(f
->bufaddr
),
146 offset_in_page(f
->bufaddr
), bcnt
);
147 ah
->aflags
|= AOEAFL_WRITE
;
149 skb
->data_len
= bcnt
;
154 ah
->cmdstat
= WIN_READ
| writebit
| extbit
;
156 /* mark all tracking fields and load out */
157 buf
->nframesout
+= 1;
158 buf
->bufaddr
+= bcnt
;
159 buf
->bv_resid
-= bcnt
;
160 /* printk(KERN_DEBUG "aoe: bv_resid=%ld\n", buf->bv_resid); */
162 buf
->sector
+= bcnt
>> 9;
163 if (buf
->resid
== 0) {
165 } else if (buf
->bv_resid
== 0) {
167 WARN_ON(buf
->bv
->bv_len
== 0);
168 buf
->bv_resid
= buf
->bv
->bv_len
;
169 buf
->bufaddr
= page_address(buf
->bv
->bv_page
) + buf
->bv
->bv_offset
;
173 skb
= skb_clone(skb
, GFP_ATOMIC
);
177 d
->sendq_tl
->next
= skb
;
183 /* some callers cannot sleep, and they can call this function,
184 * transmitting the packets later, when interrupts are on
186 static struct sk_buff
*
187 aoecmd_cfg_pkts(ushort aoemajor
, unsigned char aoeminor
, struct sk_buff
**tail
)
190 struct aoe_cfghdr
*ch
;
191 struct sk_buff
*skb
, *sl
, *sl_tail
;
192 struct net_device
*ifp
;
196 read_lock(&dev_base_lock
);
197 for_each_netdev(ifp
) {
199 if (!is_aoe_netif(ifp
))
202 skb
= new_skb(sizeof *h
+ sizeof *ch
);
204 printk(KERN_INFO
"aoe: skb alloc failure\n");
207 skb_put(skb
, sizeof *h
+ sizeof *ch
);
212 memset(h
, 0, sizeof *h
+ sizeof *ch
);
214 memset(h
->dst
, 0xff, sizeof h
->dst
);
215 memcpy(h
->src
, ifp
->dev_addr
, sizeof h
->src
);
216 h
->type
= __constant_cpu_to_be16(ETH_P_AOE
);
218 h
->major
= cpu_to_be16(aoemajor
);
227 read_unlock(&dev_base_lock
);
234 static struct frame
*
235 freeframe(struct aoedev
*d
)
243 if (f
->tag
!= FREETAG
)
245 if (atomic_read(&skb_shinfo(f
->skb
)->dataref
) == 1) {
246 skb_shinfo(f
->skb
)->nr_frags
= f
->skb
->data_len
= 0;
252 if (n
== d
->nframes
) /* wait for network layer */
253 d
->flags
|= DEVFL_KICKME
;
258 /* enters with d->lock held */
260 aoecmd_work(struct aoedev
*d
)
265 if (d
->flags
& DEVFL_PAUSE
) {
266 if (!aoedev_isbusy(d
))
267 d
->sendq_hd
= aoecmd_cfg_pkts(d
->aoemajor
,
268 d
->aoeminor
, &d
->sendq_tl
);
276 if (d
->inprocess
== NULL
) {
277 if (list_empty(&d
->bufq
))
279 buf
= container_of(d
->bufq
.next
, struct buf
, bufs
);
280 list_del(d
->bufq
.next
);
281 /*printk(KERN_DEBUG "aoe: bi_size=%ld\n", buf->bio->bi_size); */
289 rexmit(struct aoedev
*d
, struct frame
*f
)
293 struct aoe_atahdr
*ah
;
299 snprintf(buf
, sizeof buf
,
300 "%15s e%ld.%ld oldtag=%08x@%08lx newtag=%08x\n",
302 d
->aoemajor
, d
->aoeminor
, f
->tag
, jiffies
, n
);
307 ah
= (struct aoe_atahdr
*) (h
+1);
309 h
->tag
= cpu_to_be32(n
);
310 memcpy(h
->dst
, d
->addr
, sizeof h
->dst
);
311 memcpy(h
->src
, d
->ifp
->dev_addr
, sizeof h
->src
);
313 n
= DEFAULTBCNT
/ 512;
316 if (ah
->aflags
& AOEAFL_WRITE
) {
317 skb_fill_page_desc(skb
, 0, virt_to_page(f
->bufaddr
),
318 offset_in_page(f
->bufaddr
), DEFAULTBCNT
);
319 skb
->len
= sizeof *h
+ sizeof *ah
+ DEFAULTBCNT
;
320 skb
->data_len
= DEFAULTBCNT
;
322 if (++d
->lostjumbo
> (d
->nframes
<< 1))
323 if (d
->maxbcnt
!= DEFAULTBCNT
) {
324 printk(KERN_INFO
"aoe: e%ld.%ld: too many lost jumbo on %s - using 1KB frames.\n",
325 d
->aoemajor
, d
->aoeminor
, d
->ifp
->name
);
326 d
->maxbcnt
= DEFAULTBCNT
;
327 d
->flags
|= DEVFL_MAXBCNT
;
332 skb
= skb_clone(skb
, GFP_ATOMIC
);
336 d
->sendq_tl
->next
= skb
;
347 n
= jiffies
& 0xffff;
355 rexmit_timer(ulong vp
)
360 register long timeout
;
363 d
= (struct aoedev
*) vp
;
366 /* timeout is always ~150% of the moving average */
368 timeout
+= timeout
>> 1;
370 spin_lock_irqsave(&d
->lock
, flags
);
372 if (d
->flags
& DEVFL_TKILL
) {
373 spin_unlock_irqrestore(&d
->lock
, flags
);
379 if (f
->tag
!= FREETAG
&& tsince(f
->tag
) >= timeout
) {
380 n
= f
->waited
+= timeout
;
382 if (n
> aoe_deadsecs
) { /* waited too long for response */
389 if (d
->flags
& DEVFL_KICKME
) {
390 d
->flags
&= ~DEVFL_KICKME
;
395 d
->sendq_hd
= d
->sendq_tl
= NULL
;
399 d
->rttavg
= MAXTIMER
;
402 d
->timer
.expires
= jiffies
+ TIMERTICK
;
403 add_timer(&d
->timer
);
405 spin_unlock_irqrestore(&d
->lock
, flags
);
410 /* this function performs work that has been deferred until sleeping is OK
413 aoecmd_sleepwork(struct work_struct
*work
)
415 struct aoedev
*d
= container_of(work
, struct aoedev
, work
);
417 if (d
->flags
& DEVFL_GDALLOC
)
420 if (d
->flags
& DEVFL_NEWSIZE
) {
421 struct block_device
*bd
;
425 ssize
= d
->gd
->capacity
;
426 bd
= bdget_disk(d
->gd
, 0);
429 mutex_lock(&bd
->bd_inode
->i_mutex
);
430 i_size_write(bd
->bd_inode
, (loff_t
)ssize
<<9);
431 mutex_unlock(&bd
->bd_inode
->i_mutex
);
434 spin_lock_irqsave(&d
->lock
, flags
);
435 d
->flags
|= DEVFL_UP
;
436 d
->flags
&= ~DEVFL_NEWSIZE
;
437 spin_unlock_irqrestore(&d
->lock
, flags
);
442 ataid_complete(struct aoedev
*d
, unsigned char *id
)
447 /* word 83: command set supported */
448 n
= le16_to_cpu(get_unaligned((__le16
*) &id
[83<<1]));
450 /* word 86: command set/feature enabled */
451 n
|= le16_to_cpu(get_unaligned((__le16
*) &id
[86<<1]));
453 if (n
& (1<<10)) { /* bit 10: LBA 48 */
454 d
->flags
|= DEVFL_EXT
;
456 /* word 100: number lba48 sectors */
457 ssize
= le64_to_cpu(get_unaligned((__le64
*) &id
[100<<1]));
459 /* set as in ide-disk.c:init_idedisk_capacity */
460 d
->geo
.cylinders
= ssize
;
461 d
->geo
.cylinders
/= (255 * 63);
465 d
->flags
&= ~DEVFL_EXT
;
467 /* number lba28 sectors */
468 ssize
= le32_to_cpu(get_unaligned((__le32
*) &id
[60<<1]));
470 /* NOTE: obsolete in ATA 6 */
471 d
->geo
.cylinders
= le16_to_cpu(get_unaligned((__le16
*) &id
[54<<1]));
472 d
->geo
.heads
= le16_to_cpu(get_unaligned((__le16
*) &id
[55<<1]));
473 d
->geo
.sectors
= le16_to_cpu(get_unaligned((__le16
*) &id
[56<<1]));
476 if (d
->ssize
!= ssize
)
477 printk(KERN_INFO
"aoe: %012llx e%lu.%lu v%04x has %llu sectors\n",
478 (unsigned long long)mac_addr(d
->addr
),
479 d
->aoemajor
, d
->aoeminor
,
480 d
->fw_ver
, (long long)ssize
);
484 d
->gd
->capacity
= ssize
;
485 d
->flags
|= DEVFL_NEWSIZE
;
487 if (d
->flags
& DEVFL_GDALLOC
) {
488 printk(KERN_ERR
"aoe: can't schedule work for e%lu.%lu, %s\n",
489 d
->aoemajor
, d
->aoeminor
,
490 "it's already on! This shouldn't happen.\n");
493 d
->flags
|= DEVFL_GDALLOC
;
495 schedule_work(&d
->work
);
499 calc_rttavg(struct aoedev
*d
, int rtt
)
508 else if (n
> MAXTIMER
)
510 d
->mintimer
+= (n
- d
->mintimer
) >> 1;
511 } else if (n
< d
->mintimer
)
513 else if (n
> MAXTIMER
)
516 /* g == .25; cf. Congestion Avoidance and Control, Jacobson & Karels; 1988 */
522 aoecmd_ata_rsp(struct sk_buff
*skb
)
525 struct aoe_hdr
*hin
, *hout
;
526 struct aoe_atahdr
*ahin
, *ahout
;
536 aoemajor
= be16_to_cpu(get_unaligned(&hin
->major
));
537 d
= aoedev_by_aoeaddr(aoemajor
, hin
->minor
);
539 snprintf(ebuf
, sizeof ebuf
, "aoecmd_ata_rsp: ata response "
540 "for unknown device %d.%d\n",
541 aoemajor
, hin
->minor
);
546 spin_lock_irqsave(&d
->lock
, flags
);
548 n
= be32_to_cpu(get_unaligned(&hin
->tag
));
551 calc_rttavg(d
, -tsince(n
));
552 spin_unlock_irqrestore(&d
->lock
, flags
);
553 snprintf(ebuf
, sizeof ebuf
,
554 "%15s e%d.%d tag=%08x@%08lx\n",
556 be16_to_cpu(get_unaligned(&hin
->major
)),
558 be32_to_cpu(get_unaligned(&hin
->tag
)),
564 calc_rttavg(d
, tsince(f
->tag
));
566 ahin
= (struct aoe_atahdr
*) (hin
+1);
567 hout
= aoe_hdr(f
->skb
);
568 ahout
= (struct aoe_atahdr
*) (hout
+1);
571 if (ahout
->cmdstat
== WIN_IDENTIFY
)
572 d
->flags
&= ~DEVFL_PAUSE
;
573 if (ahin
->cmdstat
& 0xa9) { /* these bits cleared on success */
575 "aoe: ata error cmd=%2.2Xh stat=%2.2Xh from e%ld.%ld\n",
576 ahout
->cmdstat
, ahin
->cmdstat
,
577 d
->aoemajor
, d
->aoeminor
);
579 buf
->flags
|= BUFFL_FAIL
;
581 n
= ahout
->scnt
<< 9;
582 switch (ahout
->cmdstat
) {
585 if (skb
->len
- sizeof *hin
- sizeof *ahin
< n
) {
587 "aoe: runt data size in read. skb->len=%d\n",
589 /* fail frame f? just returning will rexmit. */
590 spin_unlock_irqrestore(&d
->lock
, flags
);
593 memcpy(f
->bufaddr
, ahin
+1, n
);
599 put_lba(ahout
, f
->lba
+= ahout
->scnt
);
603 ahout
->scnt
= n
>> 9;
604 if (ahout
->aflags
& AOEAFL_WRITE
) {
605 skb_fill_page_desc(skb
, 0,
606 virt_to_page(f
->bufaddr
),
607 offset_in_page(f
->bufaddr
), n
);
608 skb
->len
= sizeof *hout
+ sizeof *ahout
+ n
;
612 hout
->tag
= cpu_to_be32(f
->tag
);
614 skb
= skb_clone(skb
, GFP_ATOMIC
);
615 spin_unlock_irqrestore(&d
->lock
, flags
);
624 if (skb
->len
- sizeof *hin
- sizeof *ahin
< 512) {
626 "aoe: runt data size in ataid. skb->len=%d\n",
628 spin_unlock_irqrestore(&d
->lock
, flags
);
631 ataid_complete(d
, (char *) (ahin
+1));
635 "aoe: unrecognized ata command %2.2Xh for %d.%d\n",
637 be16_to_cpu(get_unaligned(&hin
->major
)),
643 buf
->nframesout
-= 1;
644 if (buf
->nframesout
== 0 && buf
->resid
== 0) {
645 unsigned long duration
= jiffies
- buf
->start_time
;
646 unsigned long n_sect
= buf
->bio
->bi_size
>> 9;
647 struct gendisk
*disk
= d
->gd
;
648 const int rw
= bio_data_dir(buf
->bio
);
650 disk_stat_inc(disk
, ios
[rw
]);
651 disk_stat_add(disk
, ticks
[rw
], duration
);
652 disk_stat_add(disk
, sectors
[rw
], n_sect
);
653 disk_stat_add(disk
, io_ticks
, duration
);
654 n
= (buf
->flags
& BUFFL_FAIL
) ? -EIO
: 0;
655 bio_endio(buf
->bio
, buf
->bio
->bi_size
, n
);
656 mempool_free(buf
, d
->bufpool
);
665 d
->sendq_hd
= d
->sendq_tl
= NULL
;
667 spin_unlock_irqrestore(&d
->lock
, flags
);
672 aoecmd_cfg(ushort aoemajor
, unsigned char aoeminor
)
676 sl
= aoecmd_cfg_pkts(aoemajor
, aoeminor
, NULL
);
682 * Since we only call this in one place (and it only prepares one frame)
683 * we just return the skb. Usually we'd chain it up to the aoedev sendq.
685 static struct sk_buff
*
686 aoecmd_ata_id(struct aoedev
*d
)
689 struct aoe_atahdr
*ah
;
695 printk(KERN_ERR
"aoe: can't get a frame. This shouldn't happen.\n");
699 /* initialize the headers & frame */
702 ah
= (struct aoe_atahdr
*) (h
+1);
703 skb_put(skb
, sizeof *h
+ sizeof *ah
);
704 memset(h
, 0, skb
->len
);
705 f
->tag
= aoehdr_atainit(d
, h
);
708 /* set up ata header */
710 ah
->cmdstat
= WIN_IDENTIFY
;
715 d
->rttavg
= MAXTIMER
;
716 d
->timer
.function
= rexmit_timer
;
718 return skb_clone(skb
, GFP_ATOMIC
);
722 aoecmd_cfg_rsp(struct sk_buff
*skb
)
726 struct aoe_cfghdr
*ch
;
727 ulong flags
, sysminor
, aoemajor
;
729 enum { MAXFRAMES
= 16 };
733 ch
= (struct aoe_cfghdr
*) (h
+1);
736 * Enough people have their dip switches set backwards to
737 * warrant a loud message for this special case.
739 aoemajor
= be16_to_cpu(get_unaligned(&h
->major
));
740 if (aoemajor
== 0xfff) {
741 printk(KERN_ERR
"aoe: Warning: shelf address is all ones. "
742 "Check shelf dip switches.\n");
746 sysminor
= SYSMINOR(aoemajor
, h
->minor
);
747 if (sysminor
* AOE_PARTITIONS
+ AOE_PARTITIONS
> MINORMASK
) {
748 printk(KERN_INFO
"aoe: e%ld.%d: minor number too large\n",
749 aoemajor
, (int) h
->minor
);
753 n
= be16_to_cpu(ch
->bufcnt
);
754 if (n
> MAXFRAMES
) /* keep it reasonable */
757 d
= aoedev_by_sysminor_m(sysminor
, n
);
759 printk(KERN_INFO
"aoe: device sysminor_m failure\n");
763 spin_lock_irqsave(&d
->lock
, flags
);
765 /* permit device to migrate mac and network interface */
767 memcpy(d
->addr
, h
->src
, sizeof d
->addr
);
768 if (!(d
->flags
& DEVFL_MAXBCNT
)) {
770 n
-= sizeof (struct aoe_hdr
) + sizeof (struct aoe_atahdr
);
774 n
= n
? n
* 512 : DEFAULTBCNT
;
775 if (n
!= d
->maxbcnt
) {
777 "aoe: e%ld.%ld: setting %d byte data frames on %s\n",
778 d
->aoemajor
, d
->aoeminor
, n
, d
->ifp
->name
);
783 /* don't change users' perspective */
784 if (d
->nopen
&& !(d
->flags
& DEVFL_PAUSE
)) {
785 spin_unlock_irqrestore(&d
->lock
, flags
);
788 d
->flags
|= DEVFL_PAUSE
; /* force pause */
789 d
->mintimer
= MINTIMER
;
790 d
->fw_ver
= be16_to_cpu(ch
->fwver
);
792 /* check for already outstanding ataid */
793 sl
= aoedev_isbusy(d
) == 0 ? aoecmd_ata_id(d
) : NULL
;
795 spin_unlock_irqrestore(&d
->lock
, flags
);