4 * XenLinux virtual block device driver.
6 * Copyright (c) 2003-2004, Keir Fraser & Steve Hand
7 * Modifications by Mark A. Williamson are (c) Intel Research Cambridge
8 * Copyright (c) 2004, Christian Limpach
9 * Copyright (c) 2004, Andrew Warfield
10 * Copyright (c) 2005, Christopher Clark
11 * Copyright (c) 2005, XenSource Ltd
13 * This program is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU General Public License version 2
15 * as published by the Free Software Foundation; or, when distributed
16 * separately from the Linux kernel or incorporated into other
17 * software packages, subject to the following license:
19 * Permission is hereby granted, free of charge, to any person obtaining a copy
20 * of this source file (the "Software"), to deal in the Software without
21 * restriction, including without limitation the rights to use, copy, modify,
22 * merge, publish, distribute, sublicense, and/or sell copies of the Software,
23 * and to permit persons to whom the Software is furnished to do so, subject to
24 * the following conditions:
26 * The above copyright notice and this permission notice shall be included in
27 * all copies or substantial portions of the Software.
29 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
30 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
31 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
32 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
33 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
34 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
38 #include <linux/interrupt.h>
39 #include <linux/blkdev.h>
40 #include <linux/hdreg.h>
41 #include <linux/cdrom.h>
42 #include <linux/module.h>
43 #include <linux/slab.h>
44 #include <linux/mutex.h>
45 #include <linux/scatterlist.h>
48 #include <xen/xenbus.h>
49 #include <xen/grant_table.h>
50 #include <xen/events.h>
52 #include <xen/platform_pci.h>
54 #include <xen/interface/grant_table.h>
55 #include <xen/interface/io/blkif.h>
56 #include <xen/interface/io/protocols.h>
58 #include <asm/xen/hypervisor.h>
61 BLKIF_STATE_DISCONNECTED
,
62 BLKIF_STATE_CONNECTED
,
63 BLKIF_STATE_SUSPENDED
,
67 struct blkif_request req
;
68 struct request
*request
;
69 unsigned long frame
[BLKIF_MAX_SEGMENTS_PER_REQUEST
];
72 static DEFINE_MUTEX(blkfront_mutex
);
73 static const struct block_device_operations xlvbd_block_fops
;
75 #define BLK_RING_SIZE __CONST_RING_SIZE(blkif, PAGE_SIZE)
78 * We have one of these per vbd, whether ide, scsi or 'other'. They
79 * hang in private_data off the gendisk structure. We may end up
80 * putting all kinds of interesting stuff here :-)
85 struct xenbus_device
*xbdev
;
89 enum blkif_state connected
;
91 struct blkif_front_ring ring
;
92 struct scatterlist sg
[BLKIF_MAX_SEGMENTS_PER_REQUEST
];
93 unsigned int evtchn
, irq
;
94 struct request_queue
*rq
;
95 struct work_struct work
;
96 struct gnttab_free_callback callback
;
97 struct blk_shadow shadow
[BLK_RING_SIZE
];
98 unsigned long shadow_free
;
99 unsigned int feature_flush
;
100 unsigned int flush_op
;
101 unsigned int feature_discard
;
102 unsigned int discard_granularity
;
103 unsigned int discard_alignment
;
107 static DEFINE_SPINLOCK(blkif_io_lock
);
109 static unsigned int nr_minors
;
110 static unsigned long *minors
;
111 static DEFINE_SPINLOCK(minor_lock
);
113 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
114 (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
115 #define GRANT_INVALID_REF 0
117 #define PARTS_PER_DISK 16
118 #define PARTS_PER_EXT_DISK 256
120 #define BLKIF_MAJOR(dev) ((dev)>>8)
121 #define BLKIF_MINOR(dev) ((dev) & 0xff)
124 #define EXTENDED (1<<EXT_SHIFT)
125 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
126 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
127 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
128 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
129 #define EMULATED_SD_DISK_MINOR_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET + (4 * 16))
130 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_HD_DISK_NAME_OFFSET + 4)
132 #define DEV_NAME "xvd" /* name in /dev */
134 static int get_id_from_freelist(struct blkfront_info
*info
)
136 unsigned long free
= info
->shadow_free
;
137 BUG_ON(free
>= BLK_RING_SIZE
);
138 info
->shadow_free
= info
->shadow
[free
].req
.id
;
139 info
->shadow
[free
].req
.id
= 0x0fffffee; /* debug */
143 static void add_id_to_freelist(struct blkfront_info
*info
,
146 info
->shadow
[id
].req
.id
= info
->shadow_free
;
147 info
->shadow
[id
].request
= NULL
;
148 info
->shadow_free
= id
;
151 static int xlbd_reserve_minors(unsigned int minor
, unsigned int nr
)
153 unsigned int end
= minor
+ nr
;
156 if (end
> nr_minors
) {
157 unsigned long *bitmap
, *old
;
159 bitmap
= kzalloc(BITS_TO_LONGS(end
) * sizeof(*bitmap
),
164 spin_lock(&minor_lock
);
165 if (end
> nr_minors
) {
167 memcpy(bitmap
, minors
,
168 BITS_TO_LONGS(nr_minors
) * sizeof(*bitmap
));
170 nr_minors
= BITS_TO_LONGS(end
) * BITS_PER_LONG
;
173 spin_unlock(&minor_lock
);
177 spin_lock(&minor_lock
);
178 if (find_next_bit(minors
, end
, minor
) >= end
) {
179 for (; minor
< end
; ++minor
)
180 __set_bit(minor
, minors
);
184 spin_unlock(&minor_lock
);
189 static void xlbd_release_minors(unsigned int minor
, unsigned int nr
)
191 unsigned int end
= minor
+ nr
;
193 BUG_ON(end
> nr_minors
);
194 spin_lock(&minor_lock
);
195 for (; minor
< end
; ++minor
)
196 __clear_bit(minor
, minors
);
197 spin_unlock(&minor_lock
);
200 static void blkif_restart_queue_callback(void *arg
)
202 struct blkfront_info
*info
= (struct blkfront_info
*)arg
;
203 schedule_work(&info
->work
);
206 static int blkif_getgeo(struct block_device
*bd
, struct hd_geometry
*hg
)
208 /* We don't have real geometry info, but let's at least return
209 values consistent with the size of the device */
210 sector_t nsect
= get_capacity(bd
->bd_disk
);
211 sector_t cylinders
= nsect
;
215 sector_div(cylinders
, hg
->heads
* hg
->sectors
);
216 hg
->cylinders
= cylinders
;
217 if ((sector_t
)(hg
->cylinders
+ 1) * hg
->heads
* hg
->sectors
< nsect
)
218 hg
->cylinders
= 0xffff;
222 static int blkif_ioctl(struct block_device
*bdev
, fmode_t mode
,
223 unsigned command
, unsigned long argument
)
225 struct blkfront_info
*info
= bdev
->bd_disk
->private_data
;
228 dev_dbg(&info
->xbdev
->dev
, "command: 0x%x, argument: 0x%lx\n",
229 command
, (long)argument
);
232 case CDROMMULTISESSION
:
233 dev_dbg(&info
->xbdev
->dev
, "FIXME: support multisession CDs later\n");
234 for (i
= 0; i
< sizeof(struct cdrom_multisession
); i
++)
235 if (put_user(0, (char __user
*)(argument
+ i
)))
239 case CDROM_GET_CAPABILITY
: {
240 struct gendisk
*gd
= info
->gd
;
241 if (gd
->flags
& GENHD_FL_CD
)
247 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
249 return -EINVAL
; /* same return as native Linux */
256 * Generate a Xen blkfront IO request from a blk layer request. Reads
257 * and writes are handled as expected.
259 * @req: a request struct
261 static int blkif_queue_request(struct request
*req
)
263 struct blkfront_info
*info
= req
->rq_disk
->private_data
;
264 unsigned long buffer_mfn
;
265 struct blkif_request
*ring_req
;
267 unsigned int fsect
, lsect
;
269 grant_ref_t gref_head
;
270 struct scatterlist
*sg
;
272 if (unlikely(info
->connected
!= BLKIF_STATE_CONNECTED
))
275 if (gnttab_alloc_grant_references(
276 BLKIF_MAX_SEGMENTS_PER_REQUEST
, &gref_head
) < 0) {
277 gnttab_request_free_callback(
279 blkif_restart_queue_callback
,
281 BLKIF_MAX_SEGMENTS_PER_REQUEST
);
285 /* Fill out a communications ring structure. */
286 ring_req
= RING_GET_REQUEST(&info
->ring
, info
->ring
.req_prod_pvt
);
287 id
= get_id_from_freelist(info
);
288 info
->shadow
[id
].request
= req
;
291 ring_req
->u
.rw
.sector_number
= (blkif_sector_t
)blk_rq_pos(req
);
292 ring_req
->handle
= info
->handle
;
294 ring_req
->operation
= rq_data_dir(req
) ?
295 BLKIF_OP_WRITE
: BLKIF_OP_READ
;
297 if (req
->cmd_flags
& (REQ_FLUSH
| REQ_FUA
)) {
299 * Ideally we can do an unordered flush-to-disk. In case the
300 * backend onlysupports barriers, use that. A barrier request
301 * a superset of FUA, so we can implement it the same
302 * way. (It's also a FLUSH+FUA, since it is
303 * guaranteed ordered WRT previous writes.)
305 ring_req
->operation
= info
->flush_op
;
308 if (unlikely(req
->cmd_flags
& REQ_DISCARD
)) {
309 /* id, sector_number and handle are set above. */
310 ring_req
->operation
= BLKIF_OP_DISCARD
;
311 ring_req
->nr_segments
= 0;
312 ring_req
->u
.discard
.nr_sectors
= blk_rq_sectors(req
);
314 ring_req
->nr_segments
= blk_rq_map_sg(req
->q
, req
, info
->sg
);
315 BUG_ON(ring_req
->nr_segments
> BLKIF_MAX_SEGMENTS_PER_REQUEST
);
317 for_each_sg(info
->sg
, sg
, ring_req
->nr_segments
, i
) {
318 buffer_mfn
= pfn_to_mfn(page_to_pfn(sg_page(sg
)));
319 fsect
= sg
->offset
>> 9;
320 lsect
= fsect
+ (sg
->length
>> 9) - 1;
321 /* install a grant reference. */
322 ref
= gnttab_claim_grant_reference(&gref_head
);
323 BUG_ON(ref
== -ENOSPC
);
325 gnttab_grant_foreign_access_ref(
327 info
->xbdev
->otherend_id
,
331 info
->shadow
[id
].frame
[i
] = mfn_to_pfn(buffer_mfn
);
332 ring_req
->u
.rw
.seg
[i
] =
333 (struct blkif_request_segment
) {
336 .last_sect
= lsect
};
340 info
->ring
.req_prod_pvt
++;
342 /* Keep a private copy so we can reissue requests when recovering. */
343 info
->shadow
[id
].req
= *ring_req
;
345 gnttab_free_grant_references(gref_head
);
351 static inline void flush_requests(struct blkfront_info
*info
)
355 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info
->ring
, notify
);
358 notify_remote_via_irq(info
->irq
);
363 * read a block; request is in a request queue
365 static void do_blkif_request(struct request_queue
*rq
)
367 struct blkfront_info
*info
= NULL
;
371 pr_debug("Entered do_blkif_request\n");
375 while ((req
= blk_peek_request(rq
)) != NULL
) {
376 info
= req
->rq_disk
->private_data
;
378 if (RING_FULL(&info
->ring
))
381 blk_start_request(req
);
383 if (req
->cmd_type
!= REQ_TYPE_FS
) {
384 __blk_end_request_all(req
, -EIO
);
388 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
389 "(%u/%u) buffer:%p [%s]\n",
390 req
, req
->cmd
, (unsigned long)blk_rq_pos(req
),
391 blk_rq_cur_sectors(req
), blk_rq_sectors(req
),
392 req
->buffer
, rq_data_dir(req
) ? "write" : "read");
394 if (blkif_queue_request(req
)) {
395 blk_requeue_request(rq
, req
);
397 /* Avoid pointless unplugs. */
406 flush_requests(info
);
409 static int xlvbd_init_blk_queue(struct gendisk
*gd
, u16 sector_size
)
411 struct request_queue
*rq
;
412 struct blkfront_info
*info
= gd
->private_data
;
414 rq
= blk_init_queue(do_blkif_request
, &blkif_io_lock
);
418 queue_flag_set_unlocked(QUEUE_FLAG_VIRT
, rq
);
420 if (info
->feature_discard
) {
421 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, rq
);
422 blk_queue_max_discard_sectors(rq
, get_capacity(gd
));
423 rq
->limits
.discard_granularity
= info
->discard_granularity
;
424 rq
->limits
.discard_alignment
= info
->discard_alignment
;
427 /* Hard sector size and max sectors impersonate the equiv. hardware. */
428 blk_queue_logical_block_size(rq
, sector_size
);
429 blk_queue_max_hw_sectors(rq
, 512);
431 /* Each segment in a request is up to an aligned page in size. */
432 blk_queue_segment_boundary(rq
, PAGE_SIZE
- 1);
433 blk_queue_max_segment_size(rq
, PAGE_SIZE
);
435 /* Ensure a merged request will fit in a single I/O ring slot. */
436 blk_queue_max_segments(rq
, BLKIF_MAX_SEGMENTS_PER_REQUEST
);
438 /* Make sure buffer addresses are sector-aligned. */
439 blk_queue_dma_alignment(rq
, 511);
441 /* Make sure we don't use bounce buffers. */
442 blk_queue_bounce_limit(rq
, BLK_BOUNCE_ANY
);
450 static void xlvbd_flush(struct blkfront_info
*info
)
452 blk_queue_flush(info
->rq
, info
->feature_flush
);
453 printk(KERN_INFO
"blkfront: %s: %s: %s\n",
455 info
->flush_op
== BLKIF_OP_WRITE_BARRIER
?
456 "barrier" : (info
->flush_op
== BLKIF_OP_FLUSH_DISKCACHE
?
457 "flush diskcache" : "barrier or flush"),
458 info
->feature_flush
? "enabled" : "disabled");
461 static int xen_translate_vdev(int vdevice
, int *minor
, unsigned int *offset
)
464 major
= BLKIF_MAJOR(vdevice
);
465 *minor
= BLKIF_MINOR(vdevice
);
468 *offset
= (*minor
/ 64) + EMULATED_HD_DISK_NAME_OFFSET
;
469 *minor
= ((*minor
/ 64) * PARTS_PER_DISK
) +
470 EMULATED_HD_DISK_MINOR_OFFSET
;
473 *offset
= (*minor
/ 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET
;
474 *minor
= (((*minor
/ 64) + 2) * PARTS_PER_DISK
) +
475 EMULATED_HD_DISK_MINOR_OFFSET
;
477 case XEN_SCSI_DISK0_MAJOR
:
478 *offset
= (*minor
/ PARTS_PER_DISK
) + EMULATED_SD_DISK_NAME_OFFSET
;
479 *minor
= *minor
+ EMULATED_SD_DISK_MINOR_OFFSET
;
481 case XEN_SCSI_DISK1_MAJOR
:
482 case XEN_SCSI_DISK2_MAJOR
:
483 case XEN_SCSI_DISK3_MAJOR
:
484 case XEN_SCSI_DISK4_MAJOR
:
485 case XEN_SCSI_DISK5_MAJOR
:
486 case XEN_SCSI_DISK6_MAJOR
:
487 case XEN_SCSI_DISK7_MAJOR
:
488 *offset
= (*minor
/ PARTS_PER_DISK
) +
489 ((major
- XEN_SCSI_DISK1_MAJOR
+ 1) * 16) +
490 EMULATED_SD_DISK_NAME_OFFSET
;
492 ((major
- XEN_SCSI_DISK1_MAJOR
+ 1) * 16 * PARTS_PER_DISK
) +
493 EMULATED_SD_DISK_MINOR_OFFSET
;
495 case XEN_SCSI_DISK8_MAJOR
:
496 case XEN_SCSI_DISK9_MAJOR
:
497 case XEN_SCSI_DISK10_MAJOR
:
498 case XEN_SCSI_DISK11_MAJOR
:
499 case XEN_SCSI_DISK12_MAJOR
:
500 case XEN_SCSI_DISK13_MAJOR
:
501 case XEN_SCSI_DISK14_MAJOR
:
502 case XEN_SCSI_DISK15_MAJOR
:
503 *offset
= (*minor
/ PARTS_PER_DISK
) +
504 ((major
- XEN_SCSI_DISK8_MAJOR
+ 8) * 16) +
505 EMULATED_SD_DISK_NAME_OFFSET
;
507 ((major
- XEN_SCSI_DISK8_MAJOR
+ 8) * 16 * PARTS_PER_DISK
) +
508 EMULATED_SD_DISK_MINOR_OFFSET
;
511 *offset
= *minor
/ PARTS_PER_DISK
;
514 printk(KERN_WARNING
"blkfront: your disk configuration is "
515 "incorrect, please use an xvd device instead\n");
521 static int xlvbd_alloc_gendisk(blkif_sector_t capacity
,
522 struct blkfront_info
*info
,
523 u16 vdisk_info
, u16 sector_size
)
532 BUG_ON(info
->gd
!= NULL
);
533 BUG_ON(info
->rq
!= NULL
);
535 if ((info
->vdevice
>>EXT_SHIFT
) > 1) {
536 /* this is above the extended range; something is wrong */
537 printk(KERN_WARNING
"blkfront: vdevice 0x%x is above the extended range; ignoring\n", info
->vdevice
);
541 if (!VDEV_IS_EXTENDED(info
->vdevice
)) {
542 err
= xen_translate_vdev(info
->vdevice
, &minor
, &offset
);
545 nr_parts
= PARTS_PER_DISK
;
547 minor
= BLKIF_MINOR_EXT(info
->vdevice
);
548 nr_parts
= PARTS_PER_EXT_DISK
;
549 offset
= minor
/ nr_parts
;
550 if (xen_hvm_domain() && offset
<= EMULATED_HD_DISK_NAME_OFFSET
+ 4)
551 printk(KERN_WARNING
"blkfront: vdevice 0x%x might conflict with "
552 "emulated IDE disks,\n\t choose an xvd device name"
553 "from xvde on\n", info
->vdevice
);
557 if ((minor
% nr_parts
) == 0)
558 nr_minors
= nr_parts
;
560 err
= xlbd_reserve_minors(minor
, nr_minors
);
565 gd
= alloc_disk(nr_minors
);
571 sprintf(gd
->disk_name
, "%s%c", DEV_NAME
, 'a' + offset
);
573 sprintf(gd
->disk_name
, "%s%c%c", DEV_NAME
,
574 'a' + ((offset
/ 26)-1), 'a' + (offset
% 26));
577 sprintf(gd
->disk_name
, "%s%c%d", DEV_NAME
,
579 minor
& (nr_parts
- 1));
581 sprintf(gd
->disk_name
, "%s%c%c%d", DEV_NAME
,
582 'a' + ((offset
/ 26) - 1),
584 minor
& (nr_parts
- 1));
587 gd
->major
= XENVBD_MAJOR
;
588 gd
->first_minor
= minor
;
589 gd
->fops
= &xlvbd_block_fops
;
590 gd
->private_data
= info
;
591 gd
->driverfs_dev
= &(info
->xbdev
->dev
);
592 set_capacity(gd
, capacity
);
594 if (xlvbd_init_blk_queue(gd
, sector_size
)) {
599 info
->rq
= gd
->queue
;
604 if (vdisk_info
& VDISK_READONLY
)
607 if (vdisk_info
& VDISK_REMOVABLE
)
608 gd
->flags
|= GENHD_FL_REMOVABLE
;
610 if (vdisk_info
& VDISK_CDROM
)
611 gd
->flags
|= GENHD_FL_CD
;
616 xlbd_release_minors(minor
, nr_minors
);
621 static void xlvbd_release_gendisk(struct blkfront_info
*info
)
623 unsigned int minor
, nr_minors
;
626 if (info
->rq
== NULL
)
629 spin_lock_irqsave(&blkif_io_lock
, flags
);
631 /* No more blkif_request(). */
632 blk_stop_queue(info
->rq
);
634 /* No more gnttab callback work. */
635 gnttab_cancel_free_callback(&info
->callback
);
636 spin_unlock_irqrestore(&blkif_io_lock
, flags
);
638 /* Flush gnttab callback work. Must be done with no locks held. */
639 flush_work_sync(&info
->work
);
641 del_gendisk(info
->gd
);
643 minor
= info
->gd
->first_minor
;
644 nr_minors
= info
->gd
->minors
;
645 xlbd_release_minors(minor
, nr_minors
);
647 blk_cleanup_queue(info
->rq
);
654 static void kick_pending_request_queues(struct blkfront_info
*info
)
656 if (!RING_FULL(&info
->ring
)) {
657 /* Re-enable calldowns. */
658 blk_start_queue(info
->rq
);
659 /* Kick things off immediately. */
660 do_blkif_request(info
->rq
);
664 static void blkif_restart_queue(struct work_struct
*work
)
666 struct blkfront_info
*info
= container_of(work
, struct blkfront_info
, work
);
668 spin_lock_irq(&blkif_io_lock
);
669 if (info
->connected
== BLKIF_STATE_CONNECTED
)
670 kick_pending_request_queues(info
);
671 spin_unlock_irq(&blkif_io_lock
);
674 static void blkif_free(struct blkfront_info
*info
, int suspend
)
676 /* Prevent new requests being issued until we fix things up. */
677 spin_lock_irq(&blkif_io_lock
);
678 info
->connected
= suspend
?
679 BLKIF_STATE_SUSPENDED
: BLKIF_STATE_DISCONNECTED
;
680 /* No more blkif_request(). */
682 blk_stop_queue(info
->rq
);
683 /* No more gnttab callback work. */
684 gnttab_cancel_free_callback(&info
->callback
);
685 spin_unlock_irq(&blkif_io_lock
);
687 /* Flush gnttab callback work. Must be done with no locks held. */
688 flush_work_sync(&info
->work
);
690 /* Free resources associated with old device channel. */
691 if (info
->ring_ref
!= GRANT_INVALID_REF
) {
692 gnttab_end_foreign_access(info
->ring_ref
, 0,
693 (unsigned long)info
->ring
.sring
);
694 info
->ring_ref
= GRANT_INVALID_REF
;
695 info
->ring
.sring
= NULL
;
698 unbind_from_irqhandler(info
->irq
, info
);
699 info
->evtchn
= info
->irq
= 0;
703 static void blkif_completion(struct blk_shadow
*s
)
706 for (i
= 0; i
< s
->req
.nr_segments
; i
++)
707 gnttab_end_foreign_access(s
->req
.u
.rw
.seg
[i
].gref
, 0, 0UL);
710 static irqreturn_t
blkif_interrupt(int irq
, void *dev_id
)
713 struct blkif_response
*bret
;
716 struct blkfront_info
*info
= (struct blkfront_info
*)dev_id
;
719 spin_lock_irqsave(&blkif_io_lock
, flags
);
721 if (unlikely(info
->connected
!= BLKIF_STATE_CONNECTED
)) {
722 spin_unlock_irqrestore(&blkif_io_lock
, flags
);
727 rp
= info
->ring
.sring
->rsp_prod
;
728 rmb(); /* Ensure we see queued responses up to 'rp'. */
730 for (i
= info
->ring
.rsp_cons
; i
!= rp
; i
++) {
733 bret
= RING_GET_RESPONSE(&info
->ring
, i
);
735 req
= info
->shadow
[id
].request
;
737 blkif_completion(&info
->shadow
[id
]);
739 add_id_to_freelist(info
, id
);
741 error
= (bret
->status
== BLKIF_RSP_OKAY
) ? 0 : -EIO
;
742 switch (bret
->operation
) {
743 case BLKIF_OP_DISCARD
:
744 if (unlikely(bret
->status
== BLKIF_RSP_EOPNOTSUPP
)) {
745 struct request_queue
*rq
= info
->rq
;
746 printk(KERN_WARNING
"blkfront: %s: discard op failed\n",
747 info
->gd
->disk_name
);
749 info
->feature_discard
= 0;
750 queue_flag_clear(QUEUE_FLAG_DISCARD
, rq
);
752 __blk_end_request_all(req
, error
);
754 case BLKIF_OP_FLUSH_DISKCACHE
:
755 case BLKIF_OP_WRITE_BARRIER
:
756 if (unlikely(bret
->status
== BLKIF_RSP_EOPNOTSUPP
)) {
757 printk(KERN_WARNING
"blkfront: %s: write %s op failed\n",
758 info
->flush_op
== BLKIF_OP_WRITE_BARRIER
?
759 "barrier" : "flush disk cache",
760 info
->gd
->disk_name
);
763 if (unlikely(bret
->status
== BLKIF_RSP_ERROR
&&
764 info
->shadow
[id
].req
.nr_segments
== 0)) {
765 printk(KERN_WARNING
"blkfront: %s: empty write %s op failed\n",
766 info
->flush_op
== BLKIF_OP_WRITE_BARRIER
?
767 "barrier" : "flush disk cache",
768 info
->gd
->disk_name
);
771 if (unlikely(error
)) {
772 if (error
== -EOPNOTSUPP
)
774 info
->feature_flush
= 0;
781 if (unlikely(bret
->status
!= BLKIF_RSP_OKAY
))
782 dev_dbg(&info
->xbdev
->dev
, "Bad return from blkdev data "
783 "request: %x\n", bret
->status
);
785 __blk_end_request_all(req
, error
);
792 info
->ring
.rsp_cons
= i
;
794 if (i
!= info
->ring
.req_prod_pvt
) {
796 RING_FINAL_CHECK_FOR_RESPONSES(&info
->ring
, more_to_do
);
800 info
->ring
.sring
->rsp_event
= i
+ 1;
802 kick_pending_request_queues(info
);
804 spin_unlock_irqrestore(&blkif_io_lock
, flags
);
810 static int setup_blkring(struct xenbus_device
*dev
,
811 struct blkfront_info
*info
)
813 struct blkif_sring
*sring
;
816 info
->ring_ref
= GRANT_INVALID_REF
;
818 sring
= (struct blkif_sring
*)__get_free_page(GFP_NOIO
| __GFP_HIGH
);
820 xenbus_dev_fatal(dev
, -ENOMEM
, "allocating shared ring");
823 SHARED_RING_INIT(sring
);
824 FRONT_RING_INIT(&info
->ring
, sring
, PAGE_SIZE
);
826 sg_init_table(info
->sg
, BLKIF_MAX_SEGMENTS_PER_REQUEST
);
828 err
= xenbus_grant_ring(dev
, virt_to_mfn(info
->ring
.sring
));
830 free_page((unsigned long)sring
);
831 info
->ring
.sring
= NULL
;
834 info
->ring_ref
= err
;
836 err
= xenbus_alloc_evtchn(dev
, &info
->evtchn
);
840 err
= bind_evtchn_to_irqhandler(info
->evtchn
,
842 IRQF_SAMPLE_RANDOM
, "blkif", info
);
844 xenbus_dev_fatal(dev
, err
,
845 "bind_evtchn_to_irqhandler failed");
857 /* Common code used when first setting up, and when resuming. */
858 static int talk_to_blkback(struct xenbus_device
*dev
,
859 struct blkfront_info
*info
)
861 const char *message
= NULL
;
862 struct xenbus_transaction xbt
;
865 /* Create shared ring, alloc event channel. */
866 err
= setup_blkring(dev
, info
);
871 err
= xenbus_transaction_start(&xbt
);
873 xenbus_dev_fatal(dev
, err
, "starting transaction");
874 goto destroy_blkring
;
877 err
= xenbus_printf(xbt
, dev
->nodename
,
878 "ring-ref", "%u", info
->ring_ref
);
880 message
= "writing ring-ref";
881 goto abort_transaction
;
883 err
= xenbus_printf(xbt
, dev
->nodename
,
884 "event-channel", "%u", info
->evtchn
);
886 message
= "writing event-channel";
887 goto abort_transaction
;
889 err
= xenbus_printf(xbt
, dev
->nodename
, "protocol", "%s",
890 XEN_IO_PROTO_ABI_NATIVE
);
892 message
= "writing protocol";
893 goto abort_transaction
;
896 err
= xenbus_transaction_end(xbt
, 0);
900 xenbus_dev_fatal(dev
, err
, "completing transaction");
901 goto destroy_blkring
;
904 xenbus_switch_state(dev
, XenbusStateInitialised
);
909 xenbus_transaction_end(xbt
, 1);
911 xenbus_dev_fatal(dev
, err
, "%s", message
);
919 * Entry point to this code when a new device is created. Allocate the basic
920 * structures and the ring buffer for communication with the backend, and
921 * inform the backend of the appropriate details for those. Switch to
924 static int blkfront_probe(struct xenbus_device
*dev
,
925 const struct xenbus_device_id
*id
)
928 struct blkfront_info
*info
;
930 /* FIXME: Use dynamic device id if this is not set. */
931 err
= xenbus_scanf(XBT_NIL
, dev
->nodename
,
932 "virtual-device", "%i", &vdevice
);
934 /* go looking in the extended area instead */
935 err
= xenbus_scanf(XBT_NIL
, dev
->nodename
, "virtual-device-ext",
938 xenbus_dev_fatal(dev
, err
, "reading virtual-device");
943 if (xen_hvm_domain()) {
946 /* no unplug has been done: do not hook devices != xen vbds */
947 if (xen_platform_pci_unplug
& XEN_UNPLUG_UNNECESSARY
) {
950 if (!VDEV_IS_EXTENDED(vdevice
))
951 major
= BLKIF_MAJOR(vdevice
);
953 major
= XENVBD_MAJOR
;
955 if (major
!= XENVBD_MAJOR
) {
957 "%s: HVM does not support vbd %d as xen block device\n",
958 __FUNCTION__
, vdevice
);
962 /* do not create a PV cdrom device if we are an HVM guest */
963 type
= xenbus_read(XBT_NIL
, dev
->nodename
, "device-type", &len
);
966 if (strncmp(type
, "cdrom", 5) == 0) {
972 info
= kzalloc(sizeof(*info
), GFP_KERNEL
);
974 xenbus_dev_fatal(dev
, -ENOMEM
, "allocating info structure");
978 mutex_init(&info
->mutex
);
980 info
->vdevice
= vdevice
;
981 info
->connected
= BLKIF_STATE_DISCONNECTED
;
982 INIT_WORK(&info
->work
, blkif_restart_queue
);
984 for (i
= 0; i
< BLK_RING_SIZE
; i
++)
985 info
->shadow
[i
].req
.id
= i
+1;
986 info
->shadow
[BLK_RING_SIZE
-1].req
.id
= 0x0fffffff;
988 /* Front end dir is a number, which is used as the id. */
989 info
->handle
= simple_strtoul(strrchr(dev
->nodename
, '/')+1, NULL
, 0);
990 dev_set_drvdata(&dev
->dev
, info
);
992 err
= talk_to_blkback(dev
, info
);
995 dev_set_drvdata(&dev
->dev
, NULL
);
1003 static int blkif_recover(struct blkfront_info
*info
)
1006 struct blkif_request
*req
;
1007 struct blk_shadow
*copy
;
1010 /* Stage 1: Make a safe copy of the shadow state. */
1011 copy
= kmalloc(sizeof(info
->shadow
),
1012 GFP_NOIO
| __GFP_REPEAT
| __GFP_HIGH
);
1015 memcpy(copy
, info
->shadow
, sizeof(info
->shadow
));
1017 /* Stage 2: Set up free list. */
1018 memset(&info
->shadow
, 0, sizeof(info
->shadow
));
1019 for (i
= 0; i
< BLK_RING_SIZE
; i
++)
1020 info
->shadow
[i
].req
.id
= i
+1;
1021 info
->shadow_free
= info
->ring
.req_prod_pvt
;
1022 info
->shadow
[BLK_RING_SIZE
-1].req
.id
= 0x0fffffff;
1024 /* Stage 3: Find pending requests and requeue them. */
1025 for (i
= 0; i
< BLK_RING_SIZE
; i
++) {
1027 if (!copy
[i
].request
)
1030 /* Grab a request slot and copy shadow state into it. */
1031 req
= RING_GET_REQUEST(&info
->ring
, info
->ring
.req_prod_pvt
);
1034 /* We get a new request id, and must reset the shadow state. */
1035 req
->id
= get_id_from_freelist(info
);
1036 memcpy(&info
->shadow
[req
->id
], ©
[i
], sizeof(copy
[i
]));
1038 /* Rewrite any grant references invalidated by susp/resume. */
1039 for (j
= 0; j
< req
->nr_segments
; j
++)
1040 gnttab_grant_foreign_access_ref(
1041 req
->u
.rw
.seg
[j
].gref
,
1042 info
->xbdev
->otherend_id
,
1043 pfn_to_mfn(info
->shadow
[req
->id
].frame
[j
]),
1044 rq_data_dir(info
->shadow
[req
->id
].request
));
1045 info
->shadow
[req
->id
].req
= *req
;
1047 info
->ring
.req_prod_pvt
++;
1052 xenbus_switch_state(info
->xbdev
, XenbusStateConnected
);
1054 spin_lock_irq(&blkif_io_lock
);
1056 /* Now safe for us to use the shared ring */
1057 info
->connected
= BLKIF_STATE_CONNECTED
;
1059 /* Send off requeued requests */
1060 flush_requests(info
);
1062 /* Kick any other new requests queued since we resumed */
1063 kick_pending_request_queues(info
);
1065 spin_unlock_irq(&blkif_io_lock
);
1071 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1072 * driver restart. We tear down our blkif structure and recreate it, but
1073 * leave the device-layer structures intact so that this is transparent to the
1074 * rest of the kernel.
1076 static int blkfront_resume(struct xenbus_device
*dev
)
1078 struct blkfront_info
*info
= dev_get_drvdata(&dev
->dev
);
1081 dev_dbg(&dev
->dev
, "blkfront_resume: %s\n", dev
->nodename
);
1083 blkif_free(info
, info
->connected
== BLKIF_STATE_CONNECTED
);
1085 err
= talk_to_blkback(dev
, info
);
1086 if (info
->connected
== BLKIF_STATE_SUSPENDED
&& !err
)
1087 err
= blkif_recover(info
);
1093 blkfront_closing(struct blkfront_info
*info
)
1095 struct xenbus_device
*xbdev
= info
->xbdev
;
1096 struct block_device
*bdev
= NULL
;
1098 mutex_lock(&info
->mutex
);
1100 if (xbdev
->state
== XenbusStateClosing
) {
1101 mutex_unlock(&info
->mutex
);
1106 bdev
= bdget_disk(info
->gd
, 0);
1108 mutex_unlock(&info
->mutex
);
1111 xenbus_frontend_closed(xbdev
);
1115 mutex_lock(&bdev
->bd_mutex
);
1117 if (bdev
->bd_openers
) {
1118 xenbus_dev_error(xbdev
, -EBUSY
,
1119 "Device in use; refusing to close");
1120 xenbus_switch_state(xbdev
, XenbusStateClosing
);
1122 xlvbd_release_gendisk(info
);
1123 xenbus_frontend_closed(xbdev
);
1126 mutex_unlock(&bdev
->bd_mutex
);
1130 static void blkfront_setup_discard(struct blkfront_info
*info
)
1134 unsigned int discard_granularity
;
1135 unsigned int discard_alignment
;
1137 type
= xenbus_read(XBT_NIL
, info
->xbdev
->otherend
, "type", NULL
);
1141 if (strncmp(type
, "phy", 3) == 0) {
1142 err
= xenbus_gather(XBT_NIL
, info
->xbdev
->otherend
,
1143 "discard-granularity", "%u", &discard_granularity
,
1144 "discard-alignment", "%u", &discard_alignment
,
1147 info
->feature_discard
= 1;
1148 info
->discard_granularity
= discard_granularity
;
1149 info
->discard_alignment
= discard_alignment
;
1151 } else if (strncmp(type
, "file", 4) == 0)
1152 info
->feature_discard
= 1;
1158 * Invoked when the backend is finally 'ready' (and has told produced
1159 * the details about the physical device - #sectors, size, etc).
1161 static void blkfront_connect(struct blkfront_info
*info
)
1163 unsigned long long sectors
;
1164 unsigned long sector_size
;
1167 int barrier
, flush
, discard
;
1169 switch (info
->connected
) {
1170 case BLKIF_STATE_CONNECTED
:
1172 * Potentially, the back-end may be signalling
1173 * a capacity change; update the capacity.
1175 err
= xenbus_scanf(XBT_NIL
, info
->xbdev
->otherend
,
1176 "sectors", "%Lu", §ors
);
1177 if (XENBUS_EXIST_ERR(err
))
1179 printk(KERN_INFO
"Setting capacity to %Lu\n",
1181 set_capacity(info
->gd
, sectors
);
1182 revalidate_disk(info
->gd
);
1185 case BLKIF_STATE_SUSPENDED
:
1192 dev_dbg(&info
->xbdev
->dev
, "%s:%s.\n",
1193 __func__
, info
->xbdev
->otherend
);
1195 err
= xenbus_gather(XBT_NIL
, info
->xbdev
->otherend
,
1196 "sectors", "%llu", §ors
,
1197 "info", "%u", &binfo
,
1198 "sector-size", "%lu", §or_size
,
1201 xenbus_dev_fatal(info
->xbdev
, err
,
1202 "reading backend fields at %s",
1203 info
->xbdev
->otherend
);
1207 info
->feature_flush
= 0;
1210 err
= xenbus_gather(XBT_NIL
, info
->xbdev
->otherend
,
1211 "feature-barrier", "%d", &barrier
,
1215 * If there's no "feature-barrier" defined, then it means
1216 * we're dealing with a very old backend which writes
1217 * synchronously; nothing to do.
1219 * If there are barriers, then we use flush.
1221 if (!err
&& barrier
) {
1222 info
->feature_flush
= REQ_FLUSH
| REQ_FUA
;
1223 info
->flush_op
= BLKIF_OP_WRITE_BARRIER
;
1226 * And if there is "feature-flush-cache" use that above
1229 err
= xenbus_gather(XBT_NIL
, info
->xbdev
->otherend
,
1230 "feature-flush-cache", "%d", &flush
,
1233 if (!err
&& flush
) {
1234 info
->feature_flush
= REQ_FLUSH
;
1235 info
->flush_op
= BLKIF_OP_FLUSH_DISKCACHE
;
1238 err
= xenbus_gather(XBT_NIL
, info
->xbdev
->otherend
,
1239 "feature-discard", "%d", &discard
,
1242 if (!err
&& discard
)
1243 blkfront_setup_discard(info
);
1245 err
= xlvbd_alloc_gendisk(sectors
, info
, binfo
, sector_size
);
1247 xenbus_dev_fatal(info
->xbdev
, err
, "xlvbd_add at %s",
1248 info
->xbdev
->otherend
);
1252 xenbus_switch_state(info
->xbdev
, XenbusStateConnected
);
1254 /* Kick pending requests. */
1255 spin_lock_irq(&blkif_io_lock
);
1256 info
->connected
= BLKIF_STATE_CONNECTED
;
1257 kick_pending_request_queues(info
);
1258 spin_unlock_irq(&blkif_io_lock
);
1266 * Callback received when the backend's state changes.
1268 static void blkback_changed(struct xenbus_device
*dev
,
1269 enum xenbus_state backend_state
)
1271 struct blkfront_info
*info
= dev_get_drvdata(&dev
->dev
);
1273 dev_dbg(&dev
->dev
, "blkfront:blkback_changed to state %d.\n", backend_state
);
1275 switch (backend_state
) {
1276 case XenbusStateInitialising
:
1277 case XenbusStateInitWait
:
1278 case XenbusStateInitialised
:
1279 case XenbusStateReconfiguring
:
1280 case XenbusStateReconfigured
:
1281 case XenbusStateUnknown
:
1282 case XenbusStateClosed
:
1285 case XenbusStateConnected
:
1286 blkfront_connect(info
);
1289 case XenbusStateClosing
:
1290 blkfront_closing(info
);
1295 static int blkfront_remove(struct xenbus_device
*xbdev
)
1297 struct blkfront_info
*info
= dev_get_drvdata(&xbdev
->dev
);
1298 struct block_device
*bdev
= NULL
;
1299 struct gendisk
*disk
;
1301 dev_dbg(&xbdev
->dev
, "%s removed", xbdev
->nodename
);
1303 blkif_free(info
, 0);
1305 mutex_lock(&info
->mutex
);
1309 bdev
= bdget_disk(disk
, 0);
1312 mutex_unlock(&info
->mutex
);
1320 * The xbdev was removed before we reached the Closed
1321 * state. See if it's safe to remove the disk. If the bdev
1322 * isn't closed yet, we let release take care of it.
1325 mutex_lock(&bdev
->bd_mutex
);
1326 info
= disk
->private_data
;
1328 dev_warn(disk_to_dev(disk
),
1329 "%s was hot-unplugged, %d stale handles\n",
1330 xbdev
->nodename
, bdev
->bd_openers
);
1332 if (info
&& !bdev
->bd_openers
) {
1333 xlvbd_release_gendisk(info
);
1334 disk
->private_data
= NULL
;
1338 mutex_unlock(&bdev
->bd_mutex
);
1344 static int blkfront_is_ready(struct xenbus_device
*dev
)
1346 struct blkfront_info
*info
= dev_get_drvdata(&dev
->dev
);
1348 return info
->is_ready
&& info
->xbdev
;
1351 static int blkif_open(struct block_device
*bdev
, fmode_t mode
)
1353 struct gendisk
*disk
= bdev
->bd_disk
;
1354 struct blkfront_info
*info
;
1357 mutex_lock(&blkfront_mutex
);
1359 info
= disk
->private_data
;
1366 mutex_lock(&info
->mutex
);
1369 /* xbdev is closed */
1372 mutex_unlock(&info
->mutex
);
1375 mutex_unlock(&blkfront_mutex
);
1379 static int blkif_release(struct gendisk
*disk
, fmode_t mode
)
1381 struct blkfront_info
*info
= disk
->private_data
;
1382 struct block_device
*bdev
;
1383 struct xenbus_device
*xbdev
;
1385 mutex_lock(&blkfront_mutex
);
1387 bdev
= bdget_disk(disk
, 0);
1390 if (bdev
->bd_openers
)
1394 * Check if we have been instructed to close. We will have
1395 * deferred this request, because the bdev was still open.
1398 mutex_lock(&info
->mutex
);
1399 xbdev
= info
->xbdev
;
1401 if (xbdev
&& xbdev
->state
== XenbusStateClosing
) {
1402 /* pending switch to state closed */
1403 dev_info(disk_to_dev(bdev
->bd_disk
), "releasing disk\n");
1404 xlvbd_release_gendisk(info
);
1405 xenbus_frontend_closed(info
->xbdev
);
1408 mutex_unlock(&info
->mutex
);
1411 /* sudden device removal */
1412 dev_info(disk_to_dev(bdev
->bd_disk
), "releasing disk\n");
1413 xlvbd_release_gendisk(info
);
1414 disk
->private_data
= NULL
;
1419 mutex_unlock(&blkfront_mutex
);
1423 static const struct block_device_operations xlvbd_block_fops
=
1425 .owner
= THIS_MODULE
,
1427 .release
= blkif_release
,
1428 .getgeo
= blkif_getgeo
,
1429 .ioctl
= blkif_ioctl
,
1433 static const struct xenbus_device_id blkfront_ids
[] = {
1438 static struct xenbus_driver blkfront
= {
1440 .owner
= THIS_MODULE
,
1441 .ids
= blkfront_ids
,
1442 .probe
= blkfront_probe
,
1443 .remove
= blkfront_remove
,
1444 .resume
= blkfront_resume
,
1445 .otherend_changed
= blkback_changed
,
1446 .is_ready
= blkfront_is_ready
,
1449 static int __init
xlblk_init(void)
1454 if (register_blkdev(XENVBD_MAJOR
, DEV_NAME
)) {
1455 printk(KERN_WARNING
"xen_blk: can't get major %d with name %s\n",
1456 XENVBD_MAJOR
, DEV_NAME
);
1460 return xenbus_register_frontend(&blkfront
);
1462 module_init(xlblk_init
);
1465 static void __exit
xlblk_exit(void)
1467 return xenbus_unregister_driver(&blkfront
);
1469 module_exit(xlblk_exit
);
1471 MODULE_DESCRIPTION("Xen virtual block device frontend");
1472 MODULE_LICENSE("GPL");
1473 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR
);
1474 MODULE_ALIAS("xen:vbd");
1475 MODULE_ALIAS("xenblk");