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
->nh
.raw
= skb
->mac
.raw
= skb
->data
;
31 skb
->protocol
= __constant_htons(ETH_P_AOE
);
34 memset(skb
->head
, 0, len
);
35 skb
->next
= skb
->prev
= NULL
;
37 /* tell the network layer not to perform IP checksums
38 * or to get the NIC to do it
40 skb
->ip_summed
= CHECKSUM_NONE
;
46 getframe(struct aoedev
*d
, int tag
)
59 * Leave the top bit clear so we have tagspace for userland.
60 * The bottom 16 bits are the xmit tick for rexmit/rttavg processing.
61 * This driver reserves tag -1 to mean "unused frame."
64 newtag(struct aoedev
*d
)
69 return n
|= (++d
->lasttag
& 0x7fff) << 16;
73 aoehdr_atainit(struct aoedev
*d
, struct aoe_hdr
*h
)
75 u32 host_tag
= newtag(d
);
77 memcpy(h
->src
, d
->ifp
->dev_addr
, sizeof h
->src
);
78 memcpy(h
->dst
, d
->addr
, sizeof h
->dst
);
79 h
->type
= __constant_cpu_to_be16(ETH_P_AOE
);
81 h
->major
= cpu_to_be16(d
->aoemajor
);
82 h
->minor
= d
->aoeminor
;
84 h
->tag
= cpu_to_be32(host_tag
);
90 put_lba(struct aoe_atahdr
*ah
, sector_t lba
)
101 aoecmd_ata_rw(struct aoedev
*d
, struct frame
*f
)
104 struct aoe_atahdr
*ah
;
108 register sector_t sector
;
109 char writebit
, extbit
;
116 sector
= buf
->sector
;
117 bcnt
= buf
->bv_resid
;
118 if (bcnt
> d
->maxbcnt
)
121 /* initialize the headers & frame */
123 h
= (struct aoe_hdr
*) skb
->mac
.raw
;
124 ah
= (struct aoe_atahdr
*) (h
+1);
125 skb
->len
= sizeof *h
+ sizeof *ah
;
126 memset(h
, 0, ETH_ZLEN
);
127 f
->tag
= aoehdr_atainit(d
, h
);
130 f
->bufaddr
= buf
->bufaddr
;
134 /* set up ata header */
135 ah
->scnt
= bcnt
>> 9;
137 if (d
->flags
& DEVFL_EXT
) {
138 ah
->aflags
|= AOEAFL_EXT
;
142 ah
->lba3
|= 0xe0; /* LBA bit + obsolete 0xa0 */
145 if (bio_data_dir(buf
->bio
) == WRITE
) {
146 skb_fill_page_desc(skb
, 0, virt_to_page(f
->bufaddr
),
147 offset_in_page(f
->bufaddr
), bcnt
);
148 ah
->aflags
|= AOEAFL_WRITE
;
150 skb
->data_len
= bcnt
;
156 ah
->cmdstat
= WIN_READ
| writebit
| extbit
;
158 /* mark all tracking fields and load out */
159 buf
->nframesout
+= 1;
160 buf
->bufaddr
+= bcnt
;
161 buf
->bv_resid
-= bcnt
;
162 /* printk(KERN_DEBUG "aoe: bv_resid=%ld\n", buf->bv_resid); */
164 buf
->sector
+= bcnt
>> 9;
165 if (buf
->resid
== 0) {
167 } else if (buf
->bv_resid
== 0) {
169 WARN_ON(buf
->bv
->bv_len
== 0);
170 buf
->bv_resid
= buf
->bv
->bv_len
;
171 buf
->bufaddr
= page_address(buf
->bv
->bv_page
) + buf
->bv
->bv_offset
;
175 skb
= skb_clone(skb
, GFP_ATOMIC
);
179 d
->sendq_tl
->next
= skb
;
185 /* some callers cannot sleep, and they can call this function,
186 * transmitting the packets later, when interrupts are on
188 static struct sk_buff
*
189 aoecmd_cfg_pkts(ushort aoemajor
, unsigned char aoeminor
, struct sk_buff
**tail
)
192 struct aoe_cfghdr
*ch
;
193 struct sk_buff
*skb
, *sl
, *sl_tail
;
194 struct net_device
*ifp
;
198 read_lock(&dev_base_lock
);
199 for (ifp
= dev_base
; ifp
; dev_put(ifp
), ifp
= ifp
->next
) {
201 if (!is_aoe_netif(ifp
))
204 skb
= new_skb(sizeof *h
+ sizeof *ch
);
206 printk(KERN_INFO
"aoe: skb alloc failure\n");
212 h
= (struct aoe_hdr
*) skb
->mac
.raw
;
213 memset(h
, 0, sizeof *h
+ sizeof *ch
);
215 memset(h
->dst
, 0xff, sizeof h
->dst
);
216 memcpy(h
->src
, ifp
->dev_addr
, sizeof h
->src
);
217 h
->type
= __constant_cpu_to_be16(ETH_P_AOE
);
219 h
->major
= cpu_to_be16(aoemajor
);
226 read_unlock(&dev_base_lock
);
233 static struct frame
*
234 freeframe(struct aoedev
*d
)
242 if (f
->tag
!= FREETAG
)
244 if (atomic_read(&skb_shinfo(f
->skb
)->dataref
) == 1) {
245 skb_shinfo(f
->skb
)->nr_frags
= f
->skb
->data_len
= 0;
250 if (n
== d
->nframes
) /* wait for network layer */
251 d
->flags
|= DEVFL_KICKME
;
256 /* enters with d->lock held */
258 aoecmd_work(struct aoedev
*d
)
263 if (d
->flags
& DEVFL_PAUSE
) {
264 if (!aoedev_isbusy(d
))
265 d
->sendq_hd
= aoecmd_cfg_pkts(d
->aoemajor
,
266 d
->aoeminor
, &d
->sendq_tl
);
274 if (d
->inprocess
== NULL
) {
275 if (list_empty(&d
->bufq
))
277 buf
= container_of(d
->bufq
.next
, struct buf
, bufs
);
278 list_del(d
->bufq
.next
);
279 /*printk(KERN_DEBUG "aoe: bi_size=%ld\n", buf->bio->bi_size); */
287 rexmit(struct aoedev
*d
, struct frame
*f
)
291 struct aoe_atahdr
*ah
;
297 snprintf(buf
, sizeof buf
,
298 "%15s e%ld.%ld oldtag=%08x@%08lx newtag=%08x\n",
300 d
->aoemajor
, d
->aoeminor
, f
->tag
, jiffies
, n
);
304 h
= (struct aoe_hdr
*) skb
->mac
.raw
;
305 ah
= (struct aoe_atahdr
*) (h
+1);
307 h
->tag
= cpu_to_be32(n
);
308 memcpy(h
->dst
, d
->addr
, sizeof h
->dst
);
309 memcpy(h
->src
, d
->ifp
->dev_addr
, sizeof h
->src
);
311 n
= DEFAULTBCNT
/ 512;
314 if (ah
->aflags
& AOEAFL_WRITE
) {
315 skb_fill_page_desc(skb
, 0, virt_to_page(f
->bufaddr
),
316 offset_in_page(f
->bufaddr
), DEFAULTBCNT
);
317 skb
->len
= sizeof *h
+ sizeof *ah
+ DEFAULTBCNT
;
318 skb
->data_len
= DEFAULTBCNT
;
320 if (++d
->lostjumbo
> (d
->nframes
<< 1))
321 if (d
->maxbcnt
!= DEFAULTBCNT
) {
322 printk(KERN_INFO
"aoe: e%ld.%ld: too many lost jumbo on %s - using 1KB frames.\n",
323 d
->aoemajor
, d
->aoeminor
, d
->ifp
->name
);
324 d
->maxbcnt
= DEFAULTBCNT
;
325 d
->flags
|= DEVFL_MAXBCNT
;
330 skb
= skb_clone(skb
, GFP_ATOMIC
);
334 d
->sendq_tl
->next
= skb
;
345 n
= jiffies
& 0xffff;
353 rexmit_timer(ulong vp
)
358 register long timeout
;
361 d
= (struct aoedev
*) vp
;
364 /* timeout is always ~150% of the moving average */
366 timeout
+= timeout
>> 1;
368 spin_lock_irqsave(&d
->lock
, flags
);
370 if (d
->flags
& DEVFL_TKILL
) {
371 spin_unlock_irqrestore(&d
->lock
, flags
);
377 if (f
->tag
!= FREETAG
&& tsince(f
->tag
) >= timeout
) {
378 n
= f
->waited
+= timeout
;
380 if (n
> aoe_deadsecs
) { /* waited too long for response */
387 if (d
->flags
& DEVFL_KICKME
) {
388 d
->flags
&= ~DEVFL_KICKME
;
393 d
->sendq_hd
= d
->sendq_tl
= NULL
;
397 d
->rttavg
= MAXTIMER
;
400 d
->timer
.expires
= jiffies
+ TIMERTICK
;
401 add_timer(&d
->timer
);
403 spin_unlock_irqrestore(&d
->lock
, flags
);
408 /* this function performs work that has been deferred until sleeping is OK
411 aoecmd_sleepwork(void *vp
)
413 struct aoedev
*d
= (struct aoedev
*) vp
;
415 if (d
->flags
& DEVFL_GDALLOC
)
418 if (d
->flags
& DEVFL_NEWSIZE
) {
419 struct block_device
*bd
;
423 ssize
= d
->gd
->capacity
;
424 bd
= bdget_disk(d
->gd
, 0);
427 mutex_lock(&bd
->bd_inode
->i_mutex
);
428 i_size_write(bd
->bd_inode
, (loff_t
)ssize
<<9);
429 mutex_unlock(&bd
->bd_inode
->i_mutex
);
432 spin_lock_irqsave(&d
->lock
, flags
);
433 d
->flags
|= DEVFL_UP
;
434 d
->flags
&= ~DEVFL_NEWSIZE
;
435 spin_unlock_irqrestore(&d
->lock
, flags
);
440 ataid_complete(struct aoedev
*d
, unsigned char *id
)
445 /* word 83: command set supported */
446 n
= le16_to_cpu(get_unaligned((__le16
*) &id
[83<<1]));
448 /* word 86: command set/feature enabled */
449 n
|= le16_to_cpu(get_unaligned((__le16
*) &id
[86<<1]));
451 if (n
& (1<<10)) { /* bit 10: LBA 48 */
452 d
->flags
|= DEVFL_EXT
;
454 /* word 100: number lba48 sectors */
455 ssize
= le64_to_cpu(get_unaligned((__le64
*) &id
[100<<1]));
457 /* set as in ide-disk.c:init_idedisk_capacity */
458 d
->geo
.cylinders
= ssize
;
459 d
->geo
.cylinders
/= (255 * 63);
463 d
->flags
&= ~DEVFL_EXT
;
465 /* number lba28 sectors */
466 ssize
= le32_to_cpu(get_unaligned((__le32
*) &id
[60<<1]));
468 /* NOTE: obsolete in ATA 6 */
469 d
->geo
.cylinders
= le16_to_cpu(get_unaligned((__le16
*) &id
[54<<1]));
470 d
->geo
.heads
= le16_to_cpu(get_unaligned((__le16
*) &id
[55<<1]));
471 d
->geo
.sectors
= le16_to_cpu(get_unaligned((__le16
*) &id
[56<<1]));
474 if (d
->ssize
!= ssize
)
475 printk(KERN_INFO
"aoe: %012llx e%lu.%lu v%04x has %llu sectors\n",
476 (unsigned long long)mac_addr(d
->addr
),
477 d
->aoemajor
, d
->aoeminor
,
478 d
->fw_ver
, (long long)ssize
);
482 d
->gd
->capacity
= ssize
;
483 d
->flags
|= DEVFL_NEWSIZE
;
485 if (d
->flags
& DEVFL_GDALLOC
) {
486 printk(KERN_ERR
"aoe: can't schedule work for e%lu.%lu, %s\n",
487 d
->aoemajor
, d
->aoeminor
,
488 "it's already on! This shouldn't happen.\n");
491 d
->flags
|= DEVFL_GDALLOC
;
493 schedule_work(&d
->work
);
497 calc_rttavg(struct aoedev
*d
, int rtt
)
506 else if (n
> MAXTIMER
)
508 d
->mintimer
+= (n
- d
->mintimer
) >> 1;
509 } else if (n
< d
->mintimer
)
511 else if (n
> MAXTIMER
)
514 /* g == .25; cf. Congestion Avoidance and Control, Jacobson & Karels; 1988 */
520 aoecmd_ata_rsp(struct sk_buff
*skb
)
523 struct aoe_hdr
*hin
, *hout
;
524 struct aoe_atahdr
*ahin
, *ahout
;
533 hin
= (struct aoe_hdr
*) skb
->mac
.raw
;
534 aoemajor
= be16_to_cpu(hin
->major
);
535 d
= aoedev_by_aoeaddr(aoemajor
, hin
->minor
);
537 snprintf(ebuf
, sizeof ebuf
, "aoecmd_ata_rsp: ata response "
538 "for unknown device %d.%d\n",
539 aoemajor
, hin
->minor
);
544 spin_lock_irqsave(&d
->lock
, flags
);
546 n
= be32_to_cpu(hin
->tag
);
549 calc_rttavg(d
, -tsince(n
));
550 spin_unlock_irqrestore(&d
->lock
, flags
);
551 snprintf(ebuf
, sizeof ebuf
,
552 "%15s e%d.%d tag=%08x@%08lx\n",
554 be16_to_cpu(hin
->major
),
556 be32_to_cpu(hin
->tag
),
562 calc_rttavg(d
, tsince(f
->tag
));
564 ahin
= (struct aoe_atahdr
*) (hin
+1);
565 hout
= (struct aoe_hdr
*) f
->skb
->mac
.raw
;
566 ahout
= (struct aoe_atahdr
*) (hout
+1);
569 if (ahout
->cmdstat
== WIN_IDENTIFY
)
570 d
->flags
&= ~DEVFL_PAUSE
;
571 if (ahin
->cmdstat
& 0xa9) { /* these bits cleared on success */
573 "aoe: ata error cmd=%2.2Xh stat=%2.2Xh from e%ld.%ld\n",
574 ahout
->cmdstat
, ahin
->cmdstat
,
575 d
->aoemajor
, d
->aoeminor
);
577 buf
->flags
|= BUFFL_FAIL
;
579 n
= ahout
->scnt
<< 9;
580 switch (ahout
->cmdstat
) {
583 if (skb
->len
- sizeof *hin
- sizeof *ahin
< n
) {
585 "aoe: runt data size in read. skb->len=%d\n",
587 /* fail frame f? just returning will rexmit. */
588 spin_unlock_irqrestore(&d
->lock
, flags
);
591 memcpy(f
->bufaddr
, ahin
+1, n
);
597 put_lba(ahout
, f
->lba
+= ahout
->scnt
);
601 ahout
->scnt
= n
>> 9;
602 if (ahout
->aflags
& AOEAFL_WRITE
) {
603 skb_fill_page_desc(skb
, 0,
604 virt_to_page(f
->bufaddr
),
605 offset_in_page(f
->bufaddr
), n
);
606 skb
->len
= sizeof *hout
+ sizeof *ahout
+ n
;
610 hout
->tag
= cpu_to_be32(f
->tag
);
612 skb
= skb_clone(skb
, GFP_ATOMIC
);
613 spin_unlock_irqrestore(&d
->lock
, flags
);
622 if (skb
->len
- sizeof *hin
- sizeof *ahin
< 512) {
624 "aoe: runt data size in ataid. skb->len=%d\n",
626 spin_unlock_irqrestore(&d
->lock
, flags
);
629 ataid_complete(d
, (char *) (ahin
+1));
633 "aoe: unrecognized ata command %2.2Xh for %d.%d\n",
635 be16_to_cpu(hin
->major
),
641 buf
->nframesout
-= 1;
642 if (buf
->nframesout
== 0 && buf
->resid
== 0) {
643 unsigned long duration
= jiffies
- buf
->start_time
;
644 unsigned long n_sect
= buf
->bio
->bi_size
>> 9;
645 struct gendisk
*disk
= d
->gd
;
646 const int rw
= bio_data_dir(buf
->bio
);
648 disk_stat_inc(disk
, ios
[rw
]);
649 disk_stat_add(disk
, ticks
[rw
], duration
);
650 disk_stat_add(disk
, sectors
[rw
], n_sect
);
651 disk_stat_add(disk
, io_ticks
, duration
);
652 n
= (buf
->flags
& BUFFL_FAIL
) ? -EIO
: 0;
653 bio_endio(buf
->bio
, buf
->bio
->bi_size
, n
);
654 mempool_free(buf
, d
->bufpool
);
663 d
->sendq_hd
= d
->sendq_tl
= NULL
;
665 spin_unlock_irqrestore(&d
->lock
, flags
);
670 aoecmd_cfg(ushort aoemajor
, unsigned char aoeminor
)
674 sl
= aoecmd_cfg_pkts(aoemajor
, aoeminor
, NULL
);
680 * Since we only call this in one place (and it only prepares one frame)
681 * we just return the skb. Usually we'd chain it up to the aoedev sendq.
683 static struct sk_buff
*
684 aoecmd_ata_id(struct aoedev
*d
)
687 struct aoe_atahdr
*ah
;
693 printk(KERN_ERR
"aoe: can't get a frame. This shouldn't happen.\n");
697 /* initialize the headers & frame */
699 h
= (struct aoe_hdr
*) skb
->mac
.raw
;
700 ah
= (struct aoe_atahdr
*) (h
+1);
702 memset(h
, 0, ETH_ZLEN
);
703 f
->tag
= aoehdr_atainit(d
, h
);
706 /* set up ata header */
708 ah
->cmdstat
= WIN_IDENTIFY
;
713 d
->rttavg
= MAXTIMER
;
714 d
->timer
.function
= rexmit_timer
;
716 return skb_clone(skb
, GFP_ATOMIC
);
720 aoecmd_cfg_rsp(struct sk_buff
*skb
)
724 struct aoe_cfghdr
*ch
;
725 ulong flags
, sysminor
, aoemajor
;
727 enum { MAXFRAMES
= 16 };
730 h
= (struct aoe_hdr
*) skb
->mac
.raw
;
731 ch
= (struct aoe_cfghdr
*) (h
+1);
734 * Enough people have their dip switches set backwards to
735 * warrant a loud message for this special case.
737 aoemajor
= be16_to_cpu(h
->major
);
738 if (aoemajor
== 0xfff) {
739 printk(KERN_ERR
"aoe: Warning: shelf address is all ones. "
740 "Check shelf dip switches.\n");
744 sysminor
= SYSMINOR(aoemajor
, h
->minor
);
745 if (sysminor
* AOE_PARTITIONS
+ AOE_PARTITIONS
> MINORMASK
) {
746 printk(KERN_INFO
"aoe: e%ld.%d: minor number too large\n",
747 aoemajor
, (int) h
->minor
);
751 n
= be16_to_cpu(ch
->bufcnt
);
752 if (n
> MAXFRAMES
) /* keep it reasonable */
755 d
= aoedev_by_sysminor_m(sysminor
, n
);
757 printk(KERN_INFO
"aoe: device sysminor_m failure\n");
761 spin_lock_irqsave(&d
->lock
, flags
);
763 /* permit device to migrate mac and network interface */
765 memcpy(d
->addr
, h
->src
, sizeof d
->addr
);
766 if (!(d
->flags
& DEVFL_MAXBCNT
)) {
768 n
-= sizeof (struct aoe_hdr
) + sizeof (struct aoe_atahdr
);
772 n
= n
? n
* 512 : DEFAULTBCNT
;
773 if (n
!= d
->maxbcnt
) {
775 "aoe: e%ld.%ld: setting %d byte data frames on %s\n",
776 d
->aoemajor
, d
->aoeminor
, n
, d
->ifp
->name
);
781 /* don't change users' perspective */
782 if (d
->nopen
&& !(d
->flags
& DEVFL_PAUSE
)) {
783 spin_unlock_irqrestore(&d
->lock
, flags
);
786 d
->flags
|= DEVFL_PAUSE
; /* force pause */
787 d
->mintimer
= MINTIMER
;
788 d
->fw_ver
= be16_to_cpu(ch
->fwver
);
790 /* check for already outstanding ataid */
791 sl
= aoedev_isbusy(d
) == 0 ? aoecmd_ata_id(d
) : NULL
;
793 spin_unlock_irqrestore(&d
->lock
, flags
);