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/blk-mq.h>
41 #include <linux/hdreg.h>
42 #include <linux/cdrom.h>
43 #include <linux/module.h>
44 #include <linux/slab.h>
45 #include <linux/mutex.h>
46 #include <linux/scatterlist.h>
47 #include <linux/bitmap.h>
48 #include <linux/list.h>
51 #include <xen/xenbus.h>
52 #include <xen/grant_table.h>
53 #include <xen/events.h>
55 #include <xen/platform_pci.h>
57 #include <xen/interface/grant_table.h>
58 #include <xen/interface/io/blkif.h>
59 #include <xen/interface/io/protocols.h>
61 #include <asm/xen/hypervisor.h>
64 * The minimal size of segment supported by the block framework is PAGE_SIZE.
65 * When Linux is using a different page size than Xen, it may not be possible
66 * to put all the data in a single segment.
67 * This can happen when the backend doesn't support indirect descriptor and
68 * therefore the maximum amount of data that a request can carry is
69 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE = 44KB
71 * Note that we only support one extra request. So the Linux page size
72 * should be <= ( 2 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) =
75 #define HAS_EXTRA_REQ (BLKIF_MAX_SEGMENTS_PER_REQUEST < XEN_PFN_PER_PAGE)
78 BLKIF_STATE_DISCONNECTED
,
79 BLKIF_STATE_CONNECTED
,
80 BLKIF_STATE_SUSPENDED
,
86 struct list_head node
;
97 struct blkif_request req
;
98 struct request
*request
;
99 struct grant
**grants_used
;
100 struct grant
**indirect_grants
;
101 struct scatterlist
*sg
;
103 enum blk_req_status status
;
105 #define NO_ASSOCIATED_ID ~0UL
107 * Id of the sibling if we ever need 2 requests when handling a
110 unsigned long associated_id
;
118 static DEFINE_MUTEX(blkfront_mutex
);
119 static const struct block_device_operations xlvbd_block_fops
;
122 * Maximum number of segments in indirect requests, the actual value used by
123 * the frontend driver is the minimum of this value and the value provided
124 * by the backend driver.
127 static unsigned int xen_blkif_max_segments
= 32;
128 module_param_named(max_indirect_segments
, xen_blkif_max_segments
, uint
,
130 MODULE_PARM_DESC(max_indirect_segments
,
131 "Maximum amount of segments in indirect requests (default is 32)");
133 static unsigned int xen_blkif_max_queues
= 4;
134 module_param_named(max_queues
, xen_blkif_max_queues
, uint
, S_IRUGO
);
135 MODULE_PARM_DESC(max_queues
, "Maximum number of hardware queues/rings used per virtual disk");
138 * Maximum order of pages to be used for the shared ring between front and
139 * backend, 4KB page granularity is used.
141 static unsigned int xen_blkif_max_ring_order
;
142 module_param_named(max_ring_page_order
, xen_blkif_max_ring_order
, int, S_IRUGO
);
143 MODULE_PARM_DESC(max_ring_page_order
, "Maximum order of pages to be used for the shared ring");
145 #define BLK_RING_SIZE(info) \
146 __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
148 #define BLK_MAX_RING_SIZE \
149 __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * XENBUS_MAX_RING_GRANTS)
152 * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
153 * characters are enough. Define to 20 to keep consistent with backend.
155 #define RINGREF_NAME_LEN (20)
157 * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
159 #define QUEUE_NAME_LEN (17)
163 * Every blkfront device can associate with one or more blkfront_ring_info,
164 * depending on how many hardware queues/rings to be used.
166 struct blkfront_ring_info
{
167 /* Lock to protect data in every ring buffer. */
168 spinlock_t ring_lock
;
169 struct blkif_front_ring ring
;
170 unsigned int ring_ref
[XENBUS_MAX_RING_GRANTS
];
171 unsigned int evtchn
, irq
;
172 struct work_struct work
;
173 struct gnttab_free_callback callback
;
174 struct blk_shadow shadow
[BLK_MAX_RING_SIZE
];
175 struct list_head indirect_pages
;
176 struct list_head grants
;
177 unsigned int persistent_gnts_c
;
178 unsigned long shadow_free
;
179 struct blkfront_info
*dev_info
;
183 * We have one of these per vbd, whether ide, scsi or 'other'. They
184 * hang in private_data off the gendisk structure. We may end up
185 * putting all kinds of interesting stuff here :-)
190 struct xenbus_device
*xbdev
;
194 enum blkif_state connected
;
195 /* Number of pages per ring buffer. */
196 unsigned int nr_ring_pages
;
197 struct request_queue
*rq
;
198 unsigned int feature_flush
;
199 unsigned int feature_fua
;
200 unsigned int feature_discard
:1;
201 unsigned int feature_secdiscard
:1;
202 unsigned int discard_granularity
;
203 unsigned int discard_alignment
;
204 unsigned int feature_persistent
:1;
205 /* Number of 4KB segments handled */
206 unsigned int max_indirect_segments
;
208 struct blk_mq_tag_set tag_set
;
209 struct blkfront_ring_info
*rinfo
;
210 unsigned int nr_rings
;
211 /* Save uncomplete reqs and bios for migration. */
212 struct list_head requests
;
213 struct bio_list bio_list
;
216 static unsigned int nr_minors
;
217 static unsigned long *minors
;
218 static DEFINE_SPINLOCK(minor_lock
);
220 #define GRANT_INVALID_REF 0
222 #define PARTS_PER_DISK 16
223 #define PARTS_PER_EXT_DISK 256
225 #define BLKIF_MAJOR(dev) ((dev)>>8)
226 #define BLKIF_MINOR(dev) ((dev) & 0xff)
229 #define EXTENDED (1<<EXT_SHIFT)
230 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
231 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
232 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
233 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
234 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
235 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
237 #define DEV_NAME "xvd" /* name in /dev */
240 * Grants are always the same size as a Xen page (i.e 4KB).
241 * A physical segment is always the same size as a Linux page.
242 * Number of grants per physical segment
244 #define GRANTS_PER_PSEG (PAGE_SIZE / XEN_PAGE_SIZE)
246 #define GRANTS_PER_INDIRECT_FRAME \
247 (XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
249 #define PSEGS_PER_INDIRECT_FRAME \
250 (GRANTS_INDIRECT_FRAME / GRANTS_PSEGS)
252 #define INDIRECT_GREFS(_grants) \
253 DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
255 #define GREFS(_psegs) ((_psegs) * GRANTS_PER_PSEG)
257 static int blkfront_setup_indirect(struct blkfront_ring_info
*rinfo
);
258 static void blkfront_gather_backend_features(struct blkfront_info
*info
);
260 static int get_id_from_freelist(struct blkfront_ring_info
*rinfo
)
262 unsigned long free
= rinfo
->shadow_free
;
264 BUG_ON(free
>= BLK_RING_SIZE(rinfo
->dev_info
));
265 rinfo
->shadow_free
= rinfo
->shadow
[free
].req
.u
.rw
.id
;
266 rinfo
->shadow
[free
].req
.u
.rw
.id
= 0x0fffffee; /* debug */
270 static int add_id_to_freelist(struct blkfront_ring_info
*rinfo
,
273 if (rinfo
->shadow
[id
].req
.u
.rw
.id
!= id
)
275 if (rinfo
->shadow
[id
].request
== NULL
)
277 rinfo
->shadow
[id
].req
.u
.rw
.id
= rinfo
->shadow_free
;
278 rinfo
->shadow
[id
].request
= NULL
;
279 rinfo
->shadow_free
= id
;
283 static int fill_grant_buffer(struct blkfront_ring_info
*rinfo
, int num
)
285 struct blkfront_info
*info
= rinfo
->dev_info
;
286 struct page
*granted_page
;
287 struct grant
*gnt_list_entry
, *n
;
291 gnt_list_entry
= kzalloc(sizeof(struct grant
), GFP_NOIO
);
295 if (info
->feature_persistent
) {
296 granted_page
= alloc_page(GFP_NOIO
);
298 kfree(gnt_list_entry
);
301 gnt_list_entry
->page
= granted_page
;
304 gnt_list_entry
->gref
= GRANT_INVALID_REF
;
305 list_add(&gnt_list_entry
->node
, &rinfo
->grants
);
312 list_for_each_entry_safe(gnt_list_entry
, n
,
313 &rinfo
->grants
, node
) {
314 list_del(&gnt_list_entry
->node
);
315 if (info
->feature_persistent
)
316 __free_page(gnt_list_entry
->page
);
317 kfree(gnt_list_entry
);
324 static struct grant
*get_free_grant(struct blkfront_ring_info
*rinfo
)
326 struct grant
*gnt_list_entry
;
328 BUG_ON(list_empty(&rinfo
->grants
));
329 gnt_list_entry
= list_first_entry(&rinfo
->grants
, struct grant
,
331 list_del(&gnt_list_entry
->node
);
333 if (gnt_list_entry
->gref
!= GRANT_INVALID_REF
)
334 rinfo
->persistent_gnts_c
--;
336 return gnt_list_entry
;
339 static inline void grant_foreign_access(const struct grant
*gnt_list_entry
,
340 const struct blkfront_info
*info
)
342 gnttab_page_grant_foreign_access_ref_one(gnt_list_entry
->gref
,
343 info
->xbdev
->otherend_id
,
344 gnt_list_entry
->page
,
348 static struct grant
*get_grant(grant_ref_t
*gref_head
,
350 struct blkfront_ring_info
*rinfo
)
352 struct grant
*gnt_list_entry
= get_free_grant(rinfo
);
353 struct blkfront_info
*info
= rinfo
->dev_info
;
355 if (gnt_list_entry
->gref
!= GRANT_INVALID_REF
)
356 return gnt_list_entry
;
358 /* Assign a gref to this page */
359 gnt_list_entry
->gref
= gnttab_claim_grant_reference(gref_head
);
360 BUG_ON(gnt_list_entry
->gref
== -ENOSPC
);
361 if (info
->feature_persistent
)
362 grant_foreign_access(gnt_list_entry
, info
);
364 /* Grant access to the GFN passed by the caller */
365 gnttab_grant_foreign_access_ref(gnt_list_entry
->gref
,
366 info
->xbdev
->otherend_id
,
370 return gnt_list_entry
;
373 static struct grant
*get_indirect_grant(grant_ref_t
*gref_head
,
374 struct blkfront_ring_info
*rinfo
)
376 struct grant
*gnt_list_entry
= get_free_grant(rinfo
);
377 struct blkfront_info
*info
= rinfo
->dev_info
;
379 if (gnt_list_entry
->gref
!= GRANT_INVALID_REF
)
380 return gnt_list_entry
;
382 /* Assign a gref to this page */
383 gnt_list_entry
->gref
= gnttab_claim_grant_reference(gref_head
);
384 BUG_ON(gnt_list_entry
->gref
== -ENOSPC
);
385 if (!info
->feature_persistent
) {
386 struct page
*indirect_page
;
388 /* Fetch a pre-allocated page to use for indirect grefs */
389 BUG_ON(list_empty(&rinfo
->indirect_pages
));
390 indirect_page
= list_first_entry(&rinfo
->indirect_pages
,
392 list_del(&indirect_page
->lru
);
393 gnt_list_entry
->page
= indirect_page
;
395 grant_foreign_access(gnt_list_entry
, info
);
397 return gnt_list_entry
;
400 static const char *op_name(int op
)
402 static const char *const names
[] = {
403 [BLKIF_OP_READ
] = "read",
404 [BLKIF_OP_WRITE
] = "write",
405 [BLKIF_OP_WRITE_BARRIER
] = "barrier",
406 [BLKIF_OP_FLUSH_DISKCACHE
] = "flush",
407 [BLKIF_OP_DISCARD
] = "discard" };
409 if (op
< 0 || op
>= ARRAY_SIZE(names
))
417 static int xlbd_reserve_minors(unsigned int minor
, unsigned int nr
)
419 unsigned int end
= minor
+ nr
;
422 if (end
> nr_minors
) {
423 unsigned long *bitmap
, *old
;
425 bitmap
= kcalloc(BITS_TO_LONGS(end
), sizeof(*bitmap
),
430 spin_lock(&minor_lock
);
431 if (end
> nr_minors
) {
433 memcpy(bitmap
, minors
,
434 BITS_TO_LONGS(nr_minors
) * sizeof(*bitmap
));
436 nr_minors
= BITS_TO_LONGS(end
) * BITS_PER_LONG
;
439 spin_unlock(&minor_lock
);
443 spin_lock(&minor_lock
);
444 if (find_next_bit(minors
, end
, minor
) >= end
) {
445 bitmap_set(minors
, minor
, nr
);
449 spin_unlock(&minor_lock
);
454 static void xlbd_release_minors(unsigned int minor
, unsigned int nr
)
456 unsigned int end
= minor
+ nr
;
458 BUG_ON(end
> nr_minors
);
459 spin_lock(&minor_lock
);
460 bitmap_clear(minors
, minor
, nr
);
461 spin_unlock(&minor_lock
);
464 static void blkif_restart_queue_callback(void *arg
)
466 struct blkfront_ring_info
*rinfo
= (struct blkfront_ring_info
*)arg
;
467 schedule_work(&rinfo
->work
);
470 static int blkif_getgeo(struct block_device
*bd
, struct hd_geometry
*hg
)
472 /* We don't have real geometry info, but let's at least return
473 values consistent with the size of the device */
474 sector_t nsect
= get_capacity(bd
->bd_disk
);
475 sector_t cylinders
= nsect
;
479 sector_div(cylinders
, hg
->heads
* hg
->sectors
);
480 hg
->cylinders
= cylinders
;
481 if ((sector_t
)(hg
->cylinders
+ 1) * hg
->heads
* hg
->sectors
< nsect
)
482 hg
->cylinders
= 0xffff;
486 static int blkif_ioctl(struct block_device
*bdev
, fmode_t mode
,
487 unsigned command
, unsigned long argument
)
489 struct blkfront_info
*info
= bdev
->bd_disk
->private_data
;
492 dev_dbg(&info
->xbdev
->dev
, "command: 0x%x, argument: 0x%lx\n",
493 command
, (long)argument
);
496 case CDROMMULTISESSION
:
497 dev_dbg(&info
->xbdev
->dev
, "FIXME: support multisession CDs later\n");
498 for (i
= 0; i
< sizeof(struct cdrom_multisession
); i
++)
499 if (put_user(0, (char __user
*)(argument
+ i
)))
503 case CDROM_GET_CAPABILITY
: {
504 struct gendisk
*gd
= info
->gd
;
505 if (gd
->flags
& GENHD_FL_CD
)
511 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
513 return -EINVAL
; /* same return as native Linux */
519 static unsigned long blkif_ring_get_request(struct blkfront_ring_info
*rinfo
,
521 struct blkif_request
**ring_req
)
525 *ring_req
= RING_GET_REQUEST(&rinfo
->ring
, rinfo
->ring
.req_prod_pvt
);
526 rinfo
->ring
.req_prod_pvt
++;
528 id
= get_id_from_freelist(rinfo
);
529 rinfo
->shadow
[id
].request
= req
;
530 rinfo
->shadow
[id
].status
= REQ_WAITING
;
531 rinfo
->shadow
[id
].associated_id
= NO_ASSOCIATED_ID
;
533 (*ring_req
)->u
.rw
.id
= id
;
538 static int blkif_queue_discard_req(struct request
*req
, struct blkfront_ring_info
*rinfo
)
540 struct blkfront_info
*info
= rinfo
->dev_info
;
541 struct blkif_request
*ring_req
;
544 /* Fill out a communications ring structure. */
545 id
= blkif_ring_get_request(rinfo
, req
, &ring_req
);
547 ring_req
->operation
= BLKIF_OP_DISCARD
;
548 ring_req
->u
.discard
.nr_sectors
= blk_rq_sectors(req
);
549 ring_req
->u
.discard
.id
= id
;
550 ring_req
->u
.discard
.sector_number
= (blkif_sector_t
)blk_rq_pos(req
);
551 if (req_op(req
) == REQ_OP_SECURE_ERASE
&& info
->feature_secdiscard
)
552 ring_req
->u
.discard
.flag
= BLKIF_DISCARD_SECURE
;
554 ring_req
->u
.discard
.flag
= 0;
556 /* Keep a private copy so we can reissue requests when recovering. */
557 rinfo
->shadow
[id
].req
= *ring_req
;
562 struct setup_rw_req
{
563 unsigned int grant_idx
;
564 struct blkif_request_segment
*segments
;
565 struct blkfront_ring_info
*rinfo
;
566 struct blkif_request
*ring_req
;
567 grant_ref_t gref_head
;
569 /* Only used when persistent grant is used and it's a read request */
571 unsigned int bvec_off
;
574 bool require_extra_req
;
575 struct blkif_request
*extra_ring_req
;
578 static void blkif_setup_rw_req_grant(unsigned long gfn
, unsigned int offset
,
579 unsigned int len
, void *data
)
581 struct setup_rw_req
*setup
= data
;
583 struct grant
*gnt_list_entry
;
584 unsigned int fsect
, lsect
;
585 /* Convenient aliases */
586 unsigned int grant_idx
= setup
->grant_idx
;
587 struct blkif_request
*ring_req
= setup
->ring_req
;
588 struct blkfront_ring_info
*rinfo
= setup
->rinfo
;
590 * We always use the shadow of the first request to store the list
591 * of grant associated to the block I/O request. This made the
592 * completion more easy to handle even if the block I/O request is
595 struct blk_shadow
*shadow
= &rinfo
->shadow
[setup
->id
];
597 if (unlikely(setup
->require_extra_req
&&
598 grant_idx
>= BLKIF_MAX_SEGMENTS_PER_REQUEST
)) {
600 * We are using the second request, setup grant_idx
601 * to be the index of the segment array.
603 grant_idx
-= BLKIF_MAX_SEGMENTS_PER_REQUEST
;
604 ring_req
= setup
->extra_ring_req
;
607 if ((ring_req
->operation
== BLKIF_OP_INDIRECT
) &&
608 (grant_idx
% GRANTS_PER_INDIRECT_FRAME
== 0)) {
610 kunmap_atomic(setup
->segments
);
612 n
= grant_idx
/ GRANTS_PER_INDIRECT_FRAME
;
613 gnt_list_entry
= get_indirect_grant(&setup
->gref_head
, rinfo
);
614 shadow
->indirect_grants
[n
] = gnt_list_entry
;
615 setup
->segments
= kmap_atomic(gnt_list_entry
->page
);
616 ring_req
->u
.indirect
.indirect_grefs
[n
] = gnt_list_entry
->gref
;
619 gnt_list_entry
= get_grant(&setup
->gref_head
, gfn
, rinfo
);
620 ref
= gnt_list_entry
->gref
;
622 * All the grants are stored in the shadow of the first
623 * request. Therefore we have to use the global index.
625 shadow
->grants_used
[setup
->grant_idx
] = gnt_list_entry
;
627 if (setup
->need_copy
) {
630 shared_data
= kmap_atomic(gnt_list_entry
->page
);
632 * this does not wipe data stored outside the
633 * range sg->offset..sg->offset+sg->length.
634 * Therefore, blkback *could* see data from
635 * previous requests. This is OK as long as
636 * persistent grants are shared with just one
637 * domain. It may need refactoring if this
640 memcpy(shared_data
+ offset
,
641 setup
->bvec_data
+ setup
->bvec_off
,
644 kunmap_atomic(shared_data
);
645 setup
->bvec_off
+= len
;
649 lsect
= fsect
+ (len
>> 9) - 1;
650 if (ring_req
->operation
!= BLKIF_OP_INDIRECT
) {
651 ring_req
->u
.rw
.seg
[grant_idx
] =
652 (struct blkif_request_segment
) {
655 .last_sect
= lsect
};
657 setup
->segments
[grant_idx
% GRANTS_PER_INDIRECT_FRAME
] =
658 (struct blkif_request_segment
) {
661 .last_sect
= lsect
};
664 (setup
->grant_idx
)++;
667 static void blkif_setup_extra_req(struct blkif_request
*first
,
668 struct blkif_request
*second
)
670 uint16_t nr_segments
= first
->u
.rw
.nr_segments
;
673 * The second request is only present when the first request uses
674 * all its segments. It's always the continuity of the first one.
676 first
->u
.rw
.nr_segments
= BLKIF_MAX_SEGMENTS_PER_REQUEST
;
678 second
->u
.rw
.nr_segments
= nr_segments
- BLKIF_MAX_SEGMENTS_PER_REQUEST
;
679 second
->u
.rw
.sector_number
= first
->u
.rw
.sector_number
+
680 (BLKIF_MAX_SEGMENTS_PER_REQUEST
* XEN_PAGE_SIZE
) / 512;
682 second
->u
.rw
.handle
= first
->u
.rw
.handle
;
683 second
->operation
= first
->operation
;
686 static int blkif_queue_rw_req(struct request
*req
, struct blkfront_ring_info
*rinfo
)
688 struct blkfront_info
*info
= rinfo
->dev_info
;
689 struct blkif_request
*ring_req
, *extra_ring_req
= NULL
;
690 unsigned long id
, extra_id
= NO_ASSOCIATED_ID
;
691 bool require_extra_req
= false;
693 struct setup_rw_req setup
= {
697 .need_copy
= rq_data_dir(req
) && info
->feature_persistent
,
701 * Used to store if we are able to queue the request by just using
702 * existing persistent grants, or if we have to get new grants,
703 * as there are not sufficiently many free.
705 struct scatterlist
*sg
;
706 int num_sg
, max_grefs
, num_grant
;
708 max_grefs
= req
->nr_phys_segments
* GRANTS_PER_PSEG
;
709 if (max_grefs
> BLKIF_MAX_SEGMENTS_PER_REQUEST
)
711 * If we are using indirect segments we need to account
712 * for the indirect grefs used in the request.
714 max_grefs
+= INDIRECT_GREFS(max_grefs
);
717 * We have to reserve 'max_grefs' grants because persistent
718 * grants are shared by all rings.
721 if (gnttab_alloc_grant_references(max_grefs
, &setup
.gref_head
) < 0) {
722 gnttab_request_free_callback(
724 blkif_restart_queue_callback
,
730 /* Fill out a communications ring structure. */
731 id
= blkif_ring_get_request(rinfo
, req
, &ring_req
);
733 num_sg
= blk_rq_map_sg(req
->q
, req
, rinfo
->shadow
[id
].sg
);
735 /* Calculate the number of grant used */
736 for_each_sg(rinfo
->shadow
[id
].sg
, sg
, num_sg
, i
)
737 num_grant
+= gnttab_count_grant(sg
->offset
, sg
->length
);
739 require_extra_req
= info
->max_indirect_segments
== 0 &&
740 num_grant
> BLKIF_MAX_SEGMENTS_PER_REQUEST
;
741 BUG_ON(!HAS_EXTRA_REQ
&& require_extra_req
);
743 rinfo
->shadow
[id
].num_sg
= num_sg
;
744 if (num_grant
> BLKIF_MAX_SEGMENTS_PER_REQUEST
&&
745 likely(!require_extra_req
)) {
747 * The indirect operation can only be a BLKIF_OP_READ or
750 BUG_ON(req_op(req
) == REQ_OP_FLUSH
|| req
->cmd_flags
& REQ_FUA
);
751 ring_req
->operation
= BLKIF_OP_INDIRECT
;
752 ring_req
->u
.indirect
.indirect_op
= rq_data_dir(req
) ?
753 BLKIF_OP_WRITE
: BLKIF_OP_READ
;
754 ring_req
->u
.indirect
.sector_number
= (blkif_sector_t
)blk_rq_pos(req
);
755 ring_req
->u
.indirect
.handle
= info
->handle
;
756 ring_req
->u
.indirect
.nr_segments
= num_grant
;
758 ring_req
->u
.rw
.sector_number
= (blkif_sector_t
)blk_rq_pos(req
);
759 ring_req
->u
.rw
.handle
= info
->handle
;
760 ring_req
->operation
= rq_data_dir(req
) ?
761 BLKIF_OP_WRITE
: BLKIF_OP_READ
;
762 if (req_op(req
) == REQ_OP_FLUSH
|| req
->cmd_flags
& REQ_FUA
) {
764 * Ideally we can do an unordered flush-to-disk.
765 * In case the backend onlysupports barriers, use that.
766 * A barrier request a superset of FUA, so we can
767 * implement it the same way. (It's also a FLUSH+FUA,
768 * since it is guaranteed ordered WRT previous writes.)
770 if (info
->feature_flush
&& info
->feature_fua
)
771 ring_req
->operation
=
772 BLKIF_OP_WRITE_BARRIER
;
773 else if (info
->feature_flush
)
774 ring_req
->operation
=
775 BLKIF_OP_FLUSH_DISKCACHE
;
777 ring_req
->operation
= 0;
779 ring_req
->u
.rw
.nr_segments
= num_grant
;
780 if (unlikely(require_extra_req
)) {
781 extra_id
= blkif_ring_get_request(rinfo
, req
,
784 * Only the first request contains the scatter-gather
787 rinfo
->shadow
[extra_id
].num_sg
= 0;
789 blkif_setup_extra_req(ring_req
, extra_ring_req
);
791 /* Link the 2 requests together */
792 rinfo
->shadow
[extra_id
].associated_id
= id
;
793 rinfo
->shadow
[id
].associated_id
= extra_id
;
797 setup
.ring_req
= ring_req
;
800 setup
.require_extra_req
= require_extra_req
;
801 if (unlikely(require_extra_req
))
802 setup
.extra_ring_req
= extra_ring_req
;
804 for_each_sg(rinfo
->shadow
[id
].sg
, sg
, num_sg
, i
) {
805 BUG_ON(sg
->offset
+ sg
->length
> PAGE_SIZE
);
807 if (setup
.need_copy
) {
808 setup
.bvec_off
= sg
->offset
;
809 setup
.bvec_data
= kmap_atomic(sg_page(sg
));
812 gnttab_foreach_grant_in_range(sg_page(sg
),
815 blkif_setup_rw_req_grant
,
819 kunmap_atomic(setup
.bvec_data
);
822 kunmap_atomic(setup
.segments
);
824 /* Keep a private copy so we can reissue requests when recovering. */
825 rinfo
->shadow
[id
].req
= *ring_req
;
826 if (unlikely(require_extra_req
))
827 rinfo
->shadow
[extra_id
].req
= *extra_ring_req
;
830 gnttab_free_grant_references(setup
.gref_head
);
836 * Generate a Xen blkfront IO request from a blk layer request. Reads
837 * and writes are handled as expected.
839 * @req: a request struct
841 static int blkif_queue_request(struct request
*req
, struct blkfront_ring_info
*rinfo
)
843 if (unlikely(rinfo
->dev_info
->connected
!= BLKIF_STATE_CONNECTED
))
846 if (unlikely(req_op(req
) == REQ_OP_DISCARD
||
847 req_op(req
) == REQ_OP_SECURE_ERASE
))
848 return blkif_queue_discard_req(req
, rinfo
);
850 return blkif_queue_rw_req(req
, rinfo
);
853 static inline void flush_requests(struct blkfront_ring_info
*rinfo
)
857 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo
->ring
, notify
);
860 notify_remote_via_irq(rinfo
->irq
);
863 static inline bool blkif_request_flush_invalid(struct request
*req
,
864 struct blkfront_info
*info
)
866 return ((req
->cmd_type
!= REQ_TYPE_FS
) ||
867 ((req_op(req
) == REQ_OP_FLUSH
) &&
868 !info
->feature_flush
) ||
869 ((req
->cmd_flags
& REQ_FUA
) &&
870 !info
->feature_fua
));
873 static int blkif_queue_rq(struct blk_mq_hw_ctx
*hctx
,
874 const struct blk_mq_queue_data
*qd
)
877 int qid
= hctx
->queue_num
;
878 struct blkfront_info
*info
= hctx
->queue
->queuedata
;
879 struct blkfront_ring_info
*rinfo
= NULL
;
881 BUG_ON(info
->nr_rings
<= qid
);
882 rinfo
= &info
->rinfo
[qid
];
883 blk_mq_start_request(qd
->rq
);
884 spin_lock_irqsave(&rinfo
->ring_lock
, flags
);
885 if (RING_FULL(&rinfo
->ring
))
888 if (blkif_request_flush_invalid(qd
->rq
, rinfo
->dev_info
))
891 if (blkif_queue_request(qd
->rq
, rinfo
))
894 flush_requests(rinfo
);
895 spin_unlock_irqrestore(&rinfo
->ring_lock
, flags
);
896 return BLK_MQ_RQ_QUEUE_OK
;
899 spin_unlock_irqrestore(&rinfo
->ring_lock
, flags
);
900 return BLK_MQ_RQ_QUEUE_ERROR
;
903 spin_unlock_irqrestore(&rinfo
->ring_lock
, flags
);
904 blk_mq_stop_hw_queue(hctx
);
905 return BLK_MQ_RQ_QUEUE_BUSY
;
908 static struct blk_mq_ops blkfront_mq_ops
= {
909 .queue_rq
= blkif_queue_rq
,
910 .map_queue
= blk_mq_map_queue
,
913 static int xlvbd_init_blk_queue(struct gendisk
*gd
, u16 sector_size
,
914 unsigned int physical_sector_size
,
915 unsigned int segments
)
917 struct request_queue
*rq
;
918 struct blkfront_info
*info
= gd
->private_data
;
920 memset(&info
->tag_set
, 0, sizeof(info
->tag_set
));
921 info
->tag_set
.ops
= &blkfront_mq_ops
;
922 info
->tag_set
.nr_hw_queues
= info
->nr_rings
;
923 if (HAS_EXTRA_REQ
&& info
->max_indirect_segments
== 0) {
925 * When indirect descriptior is not supported, the I/O request
926 * will be split between multiple request in the ring.
927 * To avoid problems when sending the request, divide by
928 * 2 the depth of the queue.
930 info
->tag_set
.queue_depth
= BLK_RING_SIZE(info
) / 2;
932 info
->tag_set
.queue_depth
= BLK_RING_SIZE(info
);
933 info
->tag_set
.numa_node
= NUMA_NO_NODE
;
934 info
->tag_set
.flags
= BLK_MQ_F_SHOULD_MERGE
| BLK_MQ_F_SG_MERGE
;
935 info
->tag_set
.cmd_size
= 0;
936 info
->tag_set
.driver_data
= info
;
938 if (blk_mq_alloc_tag_set(&info
->tag_set
))
940 rq
= blk_mq_init_queue(&info
->tag_set
);
942 blk_mq_free_tag_set(&info
->tag_set
);
946 rq
->queuedata
= info
;
947 queue_flag_set_unlocked(QUEUE_FLAG_VIRT
, rq
);
949 if (info
->feature_discard
) {
950 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, rq
);
951 blk_queue_max_discard_sectors(rq
, get_capacity(gd
));
952 rq
->limits
.discard_granularity
= info
->discard_granularity
;
953 rq
->limits
.discard_alignment
= info
->discard_alignment
;
954 if (info
->feature_secdiscard
)
955 queue_flag_set_unlocked(QUEUE_FLAG_SECERASE
, rq
);
958 /* Hard sector size and max sectors impersonate the equiv. hardware. */
959 blk_queue_logical_block_size(rq
, sector_size
);
960 blk_queue_physical_block_size(rq
, physical_sector_size
);
961 blk_queue_max_hw_sectors(rq
, (segments
* XEN_PAGE_SIZE
) / 512);
963 /* Each segment in a request is up to an aligned page in size. */
964 blk_queue_segment_boundary(rq
, PAGE_SIZE
- 1);
965 blk_queue_max_segment_size(rq
, PAGE_SIZE
);
967 /* Ensure a merged request will fit in a single I/O ring slot. */
968 blk_queue_max_segments(rq
, segments
/ GRANTS_PER_PSEG
);
970 /* Make sure buffer addresses are sector-aligned. */
971 blk_queue_dma_alignment(rq
, 511);
973 /* Make sure we don't use bounce buffers. */
974 blk_queue_bounce_limit(rq
, BLK_BOUNCE_ANY
);
981 static const char *flush_info(struct blkfront_info
*info
)
983 if (info
->feature_flush
&& info
->feature_fua
)
984 return "barrier: enabled;";
985 else if (info
->feature_flush
)
986 return "flush diskcache: enabled;";
988 return "barrier or flush: disabled;";
991 static void xlvbd_flush(struct blkfront_info
*info
)
993 blk_queue_write_cache(info
->rq
, info
->feature_flush
? true : false,
994 info
->feature_fua
? true : false);
995 pr_info("blkfront: %s: %s %s %s %s %s\n",
996 info
->gd
->disk_name
, flush_info(info
),
997 "persistent grants:", info
->feature_persistent
?
998 "enabled;" : "disabled;", "indirect descriptors:",
999 info
->max_indirect_segments
? "enabled;" : "disabled;");
1002 static int xen_translate_vdev(int vdevice
, int *minor
, unsigned int *offset
)
1005 major
= BLKIF_MAJOR(vdevice
);
1006 *minor
= BLKIF_MINOR(vdevice
);
1008 case XEN_IDE0_MAJOR
:
1009 *offset
= (*minor
/ 64) + EMULATED_HD_DISK_NAME_OFFSET
;
1010 *minor
= ((*minor
/ 64) * PARTS_PER_DISK
) +
1011 EMULATED_HD_DISK_MINOR_OFFSET
;
1013 case XEN_IDE1_MAJOR
:
1014 *offset
= (*minor
/ 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET
;
1015 *minor
= (((*minor
/ 64) + 2) * PARTS_PER_DISK
) +
1016 EMULATED_HD_DISK_MINOR_OFFSET
;
1018 case XEN_SCSI_DISK0_MAJOR
:
1019 *offset
= (*minor
/ PARTS_PER_DISK
) + EMULATED_SD_DISK_NAME_OFFSET
;
1020 *minor
= *minor
+ EMULATED_SD_DISK_MINOR_OFFSET
;
1022 case XEN_SCSI_DISK1_MAJOR
:
1023 case XEN_SCSI_DISK2_MAJOR
:
1024 case XEN_SCSI_DISK3_MAJOR
:
1025 case XEN_SCSI_DISK4_MAJOR
:
1026 case XEN_SCSI_DISK5_MAJOR
:
1027 case XEN_SCSI_DISK6_MAJOR
:
1028 case XEN_SCSI_DISK7_MAJOR
:
1029 *offset
= (*minor
/ PARTS_PER_DISK
) +
1030 ((major
- XEN_SCSI_DISK1_MAJOR
+ 1) * 16) +
1031 EMULATED_SD_DISK_NAME_OFFSET
;
1033 ((major
- XEN_SCSI_DISK1_MAJOR
+ 1) * 16 * PARTS_PER_DISK
) +
1034 EMULATED_SD_DISK_MINOR_OFFSET
;
1036 case XEN_SCSI_DISK8_MAJOR
:
1037 case XEN_SCSI_DISK9_MAJOR
:
1038 case XEN_SCSI_DISK10_MAJOR
:
1039 case XEN_SCSI_DISK11_MAJOR
:
1040 case XEN_SCSI_DISK12_MAJOR
:
1041 case XEN_SCSI_DISK13_MAJOR
:
1042 case XEN_SCSI_DISK14_MAJOR
:
1043 case XEN_SCSI_DISK15_MAJOR
:
1044 *offset
= (*minor
/ PARTS_PER_DISK
) +
1045 ((major
- XEN_SCSI_DISK8_MAJOR
+ 8) * 16) +
1046 EMULATED_SD_DISK_NAME_OFFSET
;
1048 ((major
- XEN_SCSI_DISK8_MAJOR
+ 8) * 16 * PARTS_PER_DISK
) +
1049 EMULATED_SD_DISK_MINOR_OFFSET
;
1052 *offset
= *minor
/ PARTS_PER_DISK
;
1055 printk(KERN_WARNING
"blkfront: your disk configuration is "
1056 "incorrect, please use an xvd device instead\n");
1062 static char *encode_disk_name(char *ptr
, unsigned int n
)
1065 ptr
= encode_disk_name(ptr
, n
/ 26 - 1);
1066 *ptr
= 'a' + n
% 26;
1070 static int xlvbd_alloc_gendisk(blkif_sector_t capacity
,
1071 struct blkfront_info
*info
,
1072 u16 vdisk_info
, u16 sector_size
,
1073 unsigned int physical_sector_size
)
1078 unsigned int offset
;
1083 BUG_ON(info
->gd
!= NULL
);
1084 BUG_ON(info
->rq
!= NULL
);
1086 if ((info
->vdevice
>>EXT_SHIFT
) > 1) {
1087 /* this is above the extended range; something is wrong */
1088 printk(KERN_WARNING
"blkfront: vdevice 0x%x is above the extended range; ignoring\n", info
->vdevice
);
1092 if (!VDEV_IS_EXTENDED(info
->vdevice
)) {
1093 err
= xen_translate_vdev(info
->vdevice
, &minor
, &offset
);
1096 nr_parts
= PARTS_PER_DISK
;
1098 minor
= BLKIF_MINOR_EXT(info
->vdevice
);
1099 nr_parts
= PARTS_PER_EXT_DISK
;
1100 offset
= minor
/ nr_parts
;
1101 if (xen_hvm_domain() && offset
< EMULATED_HD_DISK_NAME_OFFSET
+ 4)
1102 printk(KERN_WARNING
"blkfront: vdevice 0x%x might conflict with "
1103 "emulated IDE disks,\n\t choose an xvd device name"
1104 "from xvde on\n", info
->vdevice
);
1106 if (minor
>> MINORBITS
) {
1107 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1108 info
->vdevice
, minor
);
1112 if ((minor
% nr_parts
) == 0)
1113 nr_minors
= nr_parts
;
1115 err
= xlbd_reserve_minors(minor
, nr_minors
);
1120 gd
= alloc_disk(nr_minors
);
1124 strcpy(gd
->disk_name
, DEV_NAME
);
1125 ptr
= encode_disk_name(gd
->disk_name
+ sizeof(DEV_NAME
) - 1, offset
);
1126 BUG_ON(ptr
>= gd
->disk_name
+ DISK_NAME_LEN
);
1130 snprintf(ptr
, gd
->disk_name
+ DISK_NAME_LEN
- ptr
,
1131 "%d", minor
& (nr_parts
- 1));
1133 gd
->major
= XENVBD_MAJOR
;
1134 gd
->first_minor
= minor
;
1135 gd
->fops
= &xlvbd_block_fops
;
1136 gd
->private_data
= info
;
1137 set_capacity(gd
, capacity
);
1139 if (xlvbd_init_blk_queue(gd
, sector_size
, physical_sector_size
,
1140 info
->max_indirect_segments
? :
1141 BLKIF_MAX_SEGMENTS_PER_REQUEST
)) {
1146 info
->rq
= gd
->queue
;
1151 if (vdisk_info
& VDISK_READONLY
)
1154 if (vdisk_info
& VDISK_REMOVABLE
)
1155 gd
->flags
|= GENHD_FL_REMOVABLE
;
1157 if (vdisk_info
& VDISK_CDROM
)
1158 gd
->flags
|= GENHD_FL_CD
;
1163 xlbd_release_minors(minor
, nr_minors
);
1168 static void xlvbd_release_gendisk(struct blkfront_info
*info
)
1170 unsigned int minor
, nr_minors
, i
;
1172 if (info
->rq
== NULL
)
1175 /* No more blkif_request(). */
1176 blk_mq_stop_hw_queues(info
->rq
);
1178 for (i
= 0; i
< info
->nr_rings
; i
++) {
1179 struct blkfront_ring_info
*rinfo
= &info
->rinfo
[i
];
1181 /* No more gnttab callback work. */
1182 gnttab_cancel_free_callback(&rinfo
->callback
);
1184 /* Flush gnttab callback work. Must be done with no locks held. */
1185 flush_work(&rinfo
->work
);
1188 del_gendisk(info
->gd
);
1190 minor
= info
->gd
->first_minor
;
1191 nr_minors
= info
->gd
->minors
;
1192 xlbd_release_minors(minor
, nr_minors
);
1194 blk_cleanup_queue(info
->rq
);
1195 blk_mq_free_tag_set(&info
->tag_set
);
1202 /* Already hold rinfo->ring_lock. */
1203 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info
*rinfo
)
1205 if (!RING_FULL(&rinfo
->ring
))
1206 blk_mq_start_stopped_hw_queues(rinfo
->dev_info
->rq
, true);
1209 static void kick_pending_request_queues(struct blkfront_ring_info
*rinfo
)
1211 unsigned long flags
;
1213 spin_lock_irqsave(&rinfo
->ring_lock
, flags
);
1214 kick_pending_request_queues_locked(rinfo
);
1215 spin_unlock_irqrestore(&rinfo
->ring_lock
, flags
);
1218 static void blkif_restart_queue(struct work_struct
*work
)
1220 struct blkfront_ring_info
*rinfo
= container_of(work
, struct blkfront_ring_info
, work
);
1222 if (rinfo
->dev_info
->connected
== BLKIF_STATE_CONNECTED
)
1223 kick_pending_request_queues(rinfo
);
1226 static void blkif_free_ring(struct blkfront_ring_info
*rinfo
)
1228 struct grant
*persistent_gnt
, *n
;
1229 struct blkfront_info
*info
= rinfo
->dev_info
;
1233 * Remove indirect pages, this only happens when using indirect
1234 * descriptors but not persistent grants
1236 if (!list_empty(&rinfo
->indirect_pages
)) {
1237 struct page
*indirect_page
, *n
;
1239 BUG_ON(info
->feature_persistent
);
1240 list_for_each_entry_safe(indirect_page
, n
, &rinfo
->indirect_pages
, lru
) {
1241 list_del(&indirect_page
->lru
);
1242 __free_page(indirect_page
);
1246 /* Remove all persistent grants. */
1247 if (!list_empty(&rinfo
->grants
)) {
1248 list_for_each_entry_safe(persistent_gnt
, n
,
1249 &rinfo
->grants
, node
) {
1250 list_del(&persistent_gnt
->node
);
1251 if (persistent_gnt
->gref
!= GRANT_INVALID_REF
) {
1252 gnttab_end_foreign_access(persistent_gnt
->gref
,
1254 rinfo
->persistent_gnts_c
--;
1256 if (info
->feature_persistent
)
1257 __free_page(persistent_gnt
->page
);
1258 kfree(persistent_gnt
);
1261 BUG_ON(rinfo
->persistent_gnts_c
!= 0);
1263 for (i
= 0; i
< BLK_RING_SIZE(info
); i
++) {
1265 * Clear persistent grants present in requests already
1266 * on the shared ring
1268 if (!rinfo
->shadow
[i
].request
)
1271 segs
= rinfo
->shadow
[i
].req
.operation
== BLKIF_OP_INDIRECT
?
1272 rinfo
->shadow
[i
].req
.u
.indirect
.nr_segments
:
1273 rinfo
->shadow
[i
].req
.u
.rw
.nr_segments
;
1274 for (j
= 0; j
< segs
; j
++) {
1275 persistent_gnt
= rinfo
->shadow
[i
].grants_used
[j
];
1276 gnttab_end_foreign_access(persistent_gnt
->gref
, 0, 0UL);
1277 if (info
->feature_persistent
)
1278 __free_page(persistent_gnt
->page
);
1279 kfree(persistent_gnt
);
1282 if (rinfo
->shadow
[i
].req
.operation
!= BLKIF_OP_INDIRECT
)
1284 * If this is not an indirect operation don't try to
1285 * free indirect segments
1289 for (j
= 0; j
< INDIRECT_GREFS(segs
); j
++) {
1290 persistent_gnt
= rinfo
->shadow
[i
].indirect_grants
[j
];
1291 gnttab_end_foreign_access(persistent_gnt
->gref
, 0, 0UL);
1292 __free_page(persistent_gnt
->page
);
1293 kfree(persistent_gnt
);
1297 kfree(rinfo
->shadow
[i
].grants_used
);
1298 rinfo
->shadow
[i
].grants_used
= NULL
;
1299 kfree(rinfo
->shadow
[i
].indirect_grants
);
1300 rinfo
->shadow
[i
].indirect_grants
= NULL
;
1301 kfree(rinfo
->shadow
[i
].sg
);
1302 rinfo
->shadow
[i
].sg
= NULL
;
1305 /* No more gnttab callback work. */
1306 gnttab_cancel_free_callback(&rinfo
->callback
);
1308 /* Flush gnttab callback work. Must be done with no locks held. */
1309 flush_work(&rinfo
->work
);
1311 /* Free resources associated with old device channel. */
1312 for (i
= 0; i
< info
->nr_ring_pages
; i
++) {
1313 if (rinfo
->ring_ref
[i
] != GRANT_INVALID_REF
) {
1314 gnttab_end_foreign_access(rinfo
->ring_ref
[i
], 0, 0);
1315 rinfo
->ring_ref
[i
] = GRANT_INVALID_REF
;
1318 free_pages((unsigned long)rinfo
->ring
.sring
, get_order(info
->nr_ring_pages
* PAGE_SIZE
));
1319 rinfo
->ring
.sring
= NULL
;
1322 unbind_from_irqhandler(rinfo
->irq
, rinfo
);
1323 rinfo
->evtchn
= rinfo
->irq
= 0;
1326 static void blkif_free(struct blkfront_info
*info
, int suspend
)
1330 /* Prevent new requests being issued until we fix things up. */
1331 info
->connected
= suspend
?
1332 BLKIF_STATE_SUSPENDED
: BLKIF_STATE_DISCONNECTED
;
1333 /* No more blkif_request(). */
1335 blk_mq_stop_hw_queues(info
->rq
);
1337 for (i
= 0; i
< info
->nr_rings
; i
++)
1338 blkif_free_ring(&info
->rinfo
[i
]);
1345 struct copy_from_grant
{
1346 const struct blk_shadow
*s
;
1347 unsigned int grant_idx
;
1348 unsigned int bvec_offset
;
1352 static void blkif_copy_from_grant(unsigned long gfn
, unsigned int offset
,
1353 unsigned int len
, void *data
)
1355 struct copy_from_grant
*info
= data
;
1357 /* Convenient aliases */
1358 const struct blk_shadow
*s
= info
->s
;
1360 shared_data
= kmap_atomic(s
->grants_used
[info
->grant_idx
]->page
);
1362 memcpy(info
->bvec_data
+ info
->bvec_offset
,
1363 shared_data
+ offset
, len
);
1365 info
->bvec_offset
+= len
;
1368 kunmap_atomic(shared_data
);
1371 static enum blk_req_status
blkif_rsp_to_req_status(int rsp
)
1375 case BLKIF_RSP_OKAY
:
1377 case BLKIF_RSP_EOPNOTSUPP
:
1378 return REQ_EOPNOTSUPP
;
1379 case BLKIF_RSP_ERROR
:
1387 * Get the final status of the block request based on two ring response
1389 static int blkif_get_final_status(enum blk_req_status s1
,
1390 enum blk_req_status s2
)
1392 BUG_ON(s1
== REQ_WAITING
);
1393 BUG_ON(s2
== REQ_WAITING
);
1395 if (s1
== REQ_ERROR
|| s2
== REQ_ERROR
)
1396 return BLKIF_RSP_ERROR
;
1397 else if (s1
== REQ_EOPNOTSUPP
|| s2
== REQ_EOPNOTSUPP
)
1398 return BLKIF_RSP_EOPNOTSUPP
;
1399 return BLKIF_RSP_OKAY
;
1402 static bool blkif_completion(unsigned long *id
,
1403 struct blkfront_ring_info
*rinfo
,
1404 struct blkif_response
*bret
)
1407 struct scatterlist
*sg
;
1408 int num_sg
, num_grant
;
1409 struct blkfront_info
*info
= rinfo
->dev_info
;
1410 struct blk_shadow
*s
= &rinfo
->shadow
[*id
];
1411 struct copy_from_grant data
= {
1415 num_grant
= s
->req
.operation
== BLKIF_OP_INDIRECT
?
1416 s
->req
.u
.indirect
.nr_segments
: s
->req
.u
.rw
.nr_segments
;
1418 /* The I/O request may be split in two. */
1419 if (unlikely(s
->associated_id
!= NO_ASSOCIATED_ID
)) {
1420 struct blk_shadow
*s2
= &rinfo
->shadow
[s
->associated_id
];
1422 /* Keep the status of the current response in shadow. */
1423 s
->status
= blkif_rsp_to_req_status(bret
->status
);
1425 /* Wait the second response if not yet here. */
1426 if (s2
->status
== REQ_WAITING
)
1429 bret
->status
= blkif_get_final_status(s
->status
,
1433 * All the grants is stored in the first shadow in order
1434 * to make the completion code simpler.
1436 num_grant
+= s2
->req
.u
.rw
.nr_segments
;
1439 * The two responses may not come in order. Only the
1440 * first request will store the scatter-gather list.
1442 if (s2
->num_sg
!= 0) {
1443 /* Update "id" with the ID of the first response. */
1444 *id
= s
->associated_id
;
1449 * We don't need anymore the second request, so recycling
1452 if (add_id_to_freelist(rinfo
, s
->associated_id
))
1453 WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1454 info
->gd
->disk_name
, s
->associated_id
);
1460 if (bret
->operation
== BLKIF_OP_READ
&& info
->feature_persistent
) {
1461 for_each_sg(s
->sg
, sg
, num_sg
, i
) {
1462 BUG_ON(sg
->offset
+ sg
->length
> PAGE_SIZE
);
1464 data
.bvec_offset
= sg
->offset
;
1465 data
.bvec_data
= kmap_atomic(sg_page(sg
));
1467 gnttab_foreach_grant_in_range(sg_page(sg
),
1470 blkif_copy_from_grant
,
1473 kunmap_atomic(data
.bvec_data
);
1476 /* Add the persistent grant into the list of free grants */
1477 for (i
= 0; i
< num_grant
; i
++) {
1478 if (gnttab_query_foreign_access(s
->grants_used
[i
]->gref
)) {
1480 * If the grant is still mapped by the backend (the
1481 * backend has chosen to make this grant persistent)
1482 * we add it at the head of the list, so it will be
1485 if (!info
->feature_persistent
)
1486 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1487 s
->grants_used
[i
]->gref
);
1488 list_add(&s
->grants_used
[i
]->node
, &rinfo
->grants
);
1489 rinfo
->persistent_gnts_c
++;
1492 * If the grant is not mapped by the backend we end the
1493 * foreign access and add it to the tail of the list,
1494 * so it will not be picked again unless we run out of
1495 * persistent grants.
1497 gnttab_end_foreign_access(s
->grants_used
[i
]->gref
, 0, 0UL);
1498 s
->grants_used
[i
]->gref
= GRANT_INVALID_REF
;
1499 list_add_tail(&s
->grants_used
[i
]->node
, &rinfo
->grants
);
1502 if (s
->req
.operation
== BLKIF_OP_INDIRECT
) {
1503 for (i
= 0; i
< INDIRECT_GREFS(num_grant
); i
++) {
1504 if (gnttab_query_foreign_access(s
->indirect_grants
[i
]->gref
)) {
1505 if (!info
->feature_persistent
)
1506 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1507 s
->indirect_grants
[i
]->gref
);
1508 list_add(&s
->indirect_grants
[i
]->node
, &rinfo
->grants
);
1509 rinfo
->persistent_gnts_c
++;
1511 struct page
*indirect_page
;
1513 gnttab_end_foreign_access(s
->indirect_grants
[i
]->gref
, 0, 0UL);
1515 * Add the used indirect page back to the list of
1516 * available pages for indirect grefs.
1518 if (!info
->feature_persistent
) {
1519 indirect_page
= s
->indirect_grants
[i
]->page
;
1520 list_add(&indirect_page
->lru
, &rinfo
->indirect_pages
);
1522 s
->indirect_grants
[i
]->gref
= GRANT_INVALID_REF
;
1523 list_add_tail(&s
->indirect_grants
[i
]->node
, &rinfo
->grants
);
1531 static irqreturn_t
blkif_interrupt(int irq
, void *dev_id
)
1533 struct request
*req
;
1534 struct blkif_response
*bret
;
1536 unsigned long flags
;
1537 struct blkfront_ring_info
*rinfo
= (struct blkfront_ring_info
*)dev_id
;
1538 struct blkfront_info
*info
= rinfo
->dev_info
;
1541 if (unlikely(info
->connected
!= BLKIF_STATE_CONNECTED
))
1544 spin_lock_irqsave(&rinfo
->ring_lock
, flags
);
1546 rp
= rinfo
->ring
.sring
->rsp_prod
;
1547 rmb(); /* Ensure we see queued responses up to 'rp'. */
1549 for (i
= rinfo
->ring
.rsp_cons
; i
!= rp
; i
++) {
1552 bret
= RING_GET_RESPONSE(&rinfo
->ring
, i
);
1555 * The backend has messed up and given us an id that we would
1556 * never have given to it (we stamp it up to BLK_RING_SIZE -
1557 * look in get_id_from_freelist.
1559 if (id
>= BLK_RING_SIZE(info
)) {
1560 WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1561 info
->gd
->disk_name
, op_name(bret
->operation
), id
);
1562 /* We can't safely get the 'struct request' as
1563 * the id is busted. */
1566 req
= rinfo
->shadow
[id
].request
;
1568 if (bret
->operation
!= BLKIF_OP_DISCARD
) {
1570 * We may need to wait for an extra response if the
1571 * I/O request is split in 2
1573 if (!blkif_completion(&id
, rinfo
, bret
))
1577 if (add_id_to_freelist(rinfo
, id
)) {
1578 WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1579 info
->gd
->disk_name
, op_name(bret
->operation
), id
);
1583 error
= (bret
->status
== BLKIF_RSP_OKAY
) ? 0 : -EIO
;
1584 switch (bret
->operation
) {
1585 case BLKIF_OP_DISCARD
:
1586 if (unlikely(bret
->status
== BLKIF_RSP_EOPNOTSUPP
)) {
1587 struct request_queue
*rq
= info
->rq
;
1588 printk(KERN_WARNING
"blkfront: %s: %s op failed\n",
1589 info
->gd
->disk_name
, op_name(bret
->operation
));
1590 error
= -EOPNOTSUPP
;
1591 info
->feature_discard
= 0;
1592 info
->feature_secdiscard
= 0;
1593 queue_flag_clear(QUEUE_FLAG_DISCARD
, rq
);
1594 queue_flag_clear(QUEUE_FLAG_SECERASE
, rq
);
1596 blk_mq_complete_request(req
, error
);
1598 case BLKIF_OP_FLUSH_DISKCACHE
:
1599 case BLKIF_OP_WRITE_BARRIER
:
1600 if (unlikely(bret
->status
== BLKIF_RSP_EOPNOTSUPP
)) {
1601 printk(KERN_WARNING
"blkfront: %s: %s op failed\n",
1602 info
->gd
->disk_name
, op_name(bret
->operation
));
1603 error
= -EOPNOTSUPP
;
1605 if (unlikely(bret
->status
== BLKIF_RSP_ERROR
&&
1606 rinfo
->shadow
[id
].req
.u
.rw
.nr_segments
== 0)) {
1607 printk(KERN_WARNING
"blkfront: %s: empty %s op failed\n",
1608 info
->gd
->disk_name
, op_name(bret
->operation
));
1609 error
= -EOPNOTSUPP
;
1611 if (unlikely(error
)) {
1612 if (error
== -EOPNOTSUPP
)
1614 info
->feature_fua
= 0;
1615 info
->feature_flush
= 0;
1620 case BLKIF_OP_WRITE
:
1621 if (unlikely(bret
->status
!= BLKIF_RSP_OKAY
))
1622 dev_dbg(&info
->xbdev
->dev
, "Bad return from blkdev data "
1623 "request: %x\n", bret
->status
);
1625 blk_mq_complete_request(req
, error
);
1632 rinfo
->ring
.rsp_cons
= i
;
1634 if (i
!= rinfo
->ring
.req_prod_pvt
) {
1636 RING_FINAL_CHECK_FOR_RESPONSES(&rinfo
->ring
, more_to_do
);
1640 rinfo
->ring
.sring
->rsp_event
= i
+ 1;
1642 kick_pending_request_queues_locked(rinfo
);
1644 spin_unlock_irqrestore(&rinfo
->ring_lock
, flags
);
1650 static int setup_blkring(struct xenbus_device
*dev
,
1651 struct blkfront_ring_info
*rinfo
)
1653 struct blkif_sring
*sring
;
1655 struct blkfront_info
*info
= rinfo
->dev_info
;
1656 unsigned long ring_size
= info
->nr_ring_pages
* XEN_PAGE_SIZE
;
1657 grant_ref_t gref
[XENBUS_MAX_RING_GRANTS
];
1659 for (i
= 0; i
< info
->nr_ring_pages
; i
++)
1660 rinfo
->ring_ref
[i
] = GRANT_INVALID_REF
;
1662 sring
= (struct blkif_sring
*)__get_free_pages(GFP_NOIO
| __GFP_HIGH
,
1663 get_order(ring_size
));
1665 xenbus_dev_fatal(dev
, -ENOMEM
, "allocating shared ring");
1668 SHARED_RING_INIT(sring
);
1669 FRONT_RING_INIT(&rinfo
->ring
, sring
, ring_size
);
1671 err
= xenbus_grant_ring(dev
, rinfo
->ring
.sring
, info
->nr_ring_pages
, gref
);
1673 free_pages((unsigned long)sring
, get_order(ring_size
));
1674 rinfo
->ring
.sring
= NULL
;
1677 for (i
= 0; i
< info
->nr_ring_pages
; i
++)
1678 rinfo
->ring_ref
[i
] = gref
[i
];
1680 err
= xenbus_alloc_evtchn(dev
, &rinfo
->evtchn
);
1684 err
= bind_evtchn_to_irqhandler(rinfo
->evtchn
, blkif_interrupt
, 0,
1687 xenbus_dev_fatal(dev
, err
,
1688 "bind_evtchn_to_irqhandler failed");
1695 blkif_free(info
, 0);
1700 * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1701 * ring buffer may have multi pages depending on ->nr_ring_pages.
1703 static int write_per_ring_nodes(struct xenbus_transaction xbt
,
1704 struct blkfront_ring_info
*rinfo
, const char *dir
)
1708 const char *message
= NULL
;
1709 struct blkfront_info
*info
= rinfo
->dev_info
;
1711 if (info
->nr_ring_pages
== 1) {
1712 err
= xenbus_printf(xbt
, dir
, "ring-ref", "%u", rinfo
->ring_ref
[0]);
1714 message
= "writing ring-ref";
1715 goto abort_transaction
;
1718 for (i
= 0; i
< info
->nr_ring_pages
; i
++) {
1719 char ring_ref_name
[RINGREF_NAME_LEN
];
1721 snprintf(ring_ref_name
, RINGREF_NAME_LEN
, "ring-ref%u", i
);
1722 err
= xenbus_printf(xbt
, dir
, ring_ref_name
,
1723 "%u", rinfo
->ring_ref
[i
]);
1725 message
= "writing ring-ref";
1726 goto abort_transaction
;
1731 err
= xenbus_printf(xbt
, dir
, "event-channel", "%u", rinfo
->evtchn
);
1733 message
= "writing event-channel";
1734 goto abort_transaction
;
1740 xenbus_transaction_end(xbt
, 1);
1742 xenbus_dev_fatal(info
->xbdev
, err
, "%s", message
);
1747 /* Common code used when first setting up, and when resuming. */
1748 static int talk_to_blkback(struct xenbus_device
*dev
,
1749 struct blkfront_info
*info
)
1751 const char *message
= NULL
;
1752 struct xenbus_transaction xbt
;
1754 unsigned int i
, max_page_order
= 0;
1755 unsigned int ring_page_order
= 0;
1757 err
= xenbus_scanf(XBT_NIL
, info
->xbdev
->otherend
,
1758 "max-ring-page-order", "%u", &max_page_order
);
1760 info
->nr_ring_pages
= 1;
1762 ring_page_order
= min(xen_blkif_max_ring_order
, max_page_order
);
1763 info
->nr_ring_pages
= 1 << ring_page_order
;
1766 for (i
= 0; i
< info
->nr_rings
; i
++) {
1767 struct blkfront_ring_info
*rinfo
= &info
->rinfo
[i
];
1769 /* Create shared ring, alloc event channel. */
1770 err
= setup_blkring(dev
, rinfo
);
1772 goto destroy_blkring
;
1776 err
= xenbus_transaction_start(&xbt
);
1778 xenbus_dev_fatal(dev
, err
, "starting transaction");
1779 goto destroy_blkring
;
1782 if (info
->nr_ring_pages
> 1) {
1783 err
= xenbus_printf(xbt
, dev
->nodename
, "ring-page-order", "%u",
1786 message
= "writing ring-page-order";
1787 goto abort_transaction
;
1791 /* We already got the number of queues/rings in _probe */
1792 if (info
->nr_rings
== 1) {
1793 err
= write_per_ring_nodes(xbt
, &info
->rinfo
[0], dev
->nodename
);
1795 goto destroy_blkring
;
1800 err
= xenbus_printf(xbt
, dev
->nodename
, "multi-queue-num-queues", "%u",
1803 message
= "writing multi-queue-num-queues";
1804 goto abort_transaction
;
1807 pathsize
= strlen(dev
->nodename
) + QUEUE_NAME_LEN
;
1808 path
= kmalloc(pathsize
, GFP_KERNEL
);
1811 message
= "ENOMEM while writing ring references";
1812 goto abort_transaction
;
1815 for (i
= 0; i
< info
->nr_rings
; i
++) {
1816 memset(path
, 0, pathsize
);
1817 snprintf(path
, pathsize
, "%s/queue-%u", dev
->nodename
, i
);
1818 err
= write_per_ring_nodes(xbt
, &info
->rinfo
[i
], path
);
1821 goto destroy_blkring
;
1826 err
= xenbus_printf(xbt
, dev
->nodename
, "protocol", "%s",
1827 XEN_IO_PROTO_ABI_NATIVE
);
1829 message
= "writing protocol";
1830 goto abort_transaction
;
1832 err
= xenbus_printf(xbt
, dev
->nodename
,
1833 "feature-persistent", "%u", 1);
1836 "writing persistent grants feature to xenbus");
1838 err
= xenbus_transaction_end(xbt
, 0);
1842 xenbus_dev_fatal(dev
, err
, "completing transaction");
1843 goto destroy_blkring
;
1846 for (i
= 0; i
< info
->nr_rings
; i
++) {
1848 struct blkfront_ring_info
*rinfo
= &info
->rinfo
[i
];
1850 for (j
= 0; j
< BLK_RING_SIZE(info
); j
++)
1851 rinfo
->shadow
[j
].req
.u
.rw
.id
= j
+ 1;
1852 rinfo
->shadow
[BLK_RING_SIZE(info
)-1].req
.u
.rw
.id
= 0x0fffffff;
1854 xenbus_switch_state(dev
, XenbusStateInitialised
);
1859 xenbus_transaction_end(xbt
, 1);
1861 xenbus_dev_fatal(dev
, err
, "%s", message
);
1863 blkif_free(info
, 0);
1866 dev_set_drvdata(&dev
->dev
, NULL
);
1871 static int negotiate_mq(struct blkfront_info
*info
)
1873 unsigned int backend_max_queues
= 0;
1877 BUG_ON(info
->nr_rings
);
1879 /* Check if backend supports multiple queues. */
1880 err
= xenbus_scanf(XBT_NIL
, info
->xbdev
->otherend
,
1881 "multi-queue-max-queues", "%u", &backend_max_queues
);
1883 backend_max_queues
= 1;
1885 info
->nr_rings
= min(backend_max_queues
, xen_blkif_max_queues
);
1886 /* We need at least one ring. */
1887 if (!info
->nr_rings
)
1890 info
->rinfo
= kzalloc(sizeof(struct blkfront_ring_info
) * info
->nr_rings
, GFP_KERNEL
);
1892 xenbus_dev_fatal(info
->xbdev
, -ENOMEM
, "allocating ring_info structure");
1896 for (i
= 0; i
< info
->nr_rings
; i
++) {
1897 struct blkfront_ring_info
*rinfo
;
1899 rinfo
= &info
->rinfo
[i
];
1900 INIT_LIST_HEAD(&rinfo
->indirect_pages
);
1901 INIT_LIST_HEAD(&rinfo
->grants
);
1902 rinfo
->dev_info
= info
;
1903 INIT_WORK(&rinfo
->work
, blkif_restart_queue
);
1904 spin_lock_init(&rinfo
->ring_lock
);
1909 * Entry point to this code when a new device is created. Allocate the basic
1910 * structures and the ring buffer for communication with the backend, and
1911 * inform the backend of the appropriate details for those. Switch to
1912 * Initialised state.
1914 static int blkfront_probe(struct xenbus_device
*dev
,
1915 const struct xenbus_device_id
*id
)
1918 struct blkfront_info
*info
;
1920 /* FIXME: Use dynamic device id if this is not set. */
1921 err
= xenbus_scanf(XBT_NIL
, dev
->nodename
,
1922 "virtual-device", "%i", &vdevice
);
1924 /* go looking in the extended area instead */
1925 err
= xenbus_scanf(XBT_NIL
, dev
->nodename
, "virtual-device-ext",
1928 xenbus_dev_fatal(dev
, err
, "reading virtual-device");
1933 if (xen_hvm_domain()) {
1936 /* no unplug has been done: do not hook devices != xen vbds */
1937 if (xen_has_pv_and_legacy_disk_devices()) {
1940 if (!VDEV_IS_EXTENDED(vdevice
))
1941 major
= BLKIF_MAJOR(vdevice
);
1943 major
= XENVBD_MAJOR
;
1945 if (major
!= XENVBD_MAJOR
) {
1947 "%s: HVM does not support vbd %d as xen block device\n",
1952 /* do not create a PV cdrom device if we are an HVM guest */
1953 type
= xenbus_read(XBT_NIL
, dev
->nodename
, "device-type", &len
);
1956 if (strncmp(type
, "cdrom", 5) == 0) {
1962 info
= kzalloc(sizeof(*info
), GFP_KERNEL
);
1964 xenbus_dev_fatal(dev
, -ENOMEM
, "allocating info structure");
1969 err
= negotiate_mq(info
);
1975 mutex_init(&info
->mutex
);
1976 info
->vdevice
= vdevice
;
1977 info
->connected
= BLKIF_STATE_DISCONNECTED
;
1979 /* Front end dir is a number, which is used as the id. */
1980 info
->handle
= simple_strtoul(strrchr(dev
->nodename
, '/')+1, NULL
, 0);
1981 dev_set_drvdata(&dev
->dev
, info
);
1986 static void split_bio_end(struct bio
*bio
)
1988 struct split_bio
*split_bio
= bio
->bi_private
;
1990 if (atomic_dec_and_test(&split_bio
->pending
)) {
1991 split_bio
->bio
->bi_phys_segments
= 0;
1992 split_bio
->bio
->bi_error
= bio
->bi_error
;
1993 bio_endio(split_bio
->bio
);
1999 static int blkif_recover(struct blkfront_info
*info
)
2001 unsigned int i
, r_index
;
2002 struct request
*req
, *n
;
2004 struct bio
*bio
, *cloned_bio
;
2005 unsigned int segs
, offset
;
2007 struct split_bio
*split_bio
;
2009 blkfront_gather_backend_features(info
);
2010 segs
= info
->max_indirect_segments
? : BLKIF_MAX_SEGMENTS_PER_REQUEST
;
2011 blk_queue_max_segments(info
->rq
, segs
);
2013 for (r_index
= 0; r_index
< info
->nr_rings
; r_index
++) {
2014 struct blkfront_ring_info
*rinfo
= &info
->rinfo
[r_index
];
2016 rc
= blkfront_setup_indirect(rinfo
);
2020 xenbus_switch_state(info
->xbdev
, XenbusStateConnected
);
2022 /* Now safe for us to use the shared ring */
2023 info
->connected
= BLKIF_STATE_CONNECTED
;
2025 for (r_index
= 0; r_index
< info
->nr_rings
; r_index
++) {
2026 struct blkfront_ring_info
*rinfo
;
2028 rinfo
= &info
->rinfo
[r_index
];
2029 /* Kick any other new requests queued since we resumed */
2030 kick_pending_request_queues(rinfo
);
2033 list_for_each_entry_safe(req
, n
, &info
->requests
, queuelist
) {
2034 /* Requeue pending requests (flush or discard) */
2035 list_del_init(&req
->queuelist
);
2036 BUG_ON(req
->nr_phys_segments
> segs
);
2037 blk_mq_requeue_request(req
);
2039 blk_mq_kick_requeue_list(info
->rq
);
2041 while ((bio
= bio_list_pop(&info
->bio_list
)) != NULL
) {
2042 /* Traverse the list of pending bios and re-queue them */
2043 if (bio_segments(bio
) > segs
) {
2045 * This bio has more segments than what we can
2046 * handle, we have to split it.
2048 pending
= (bio_segments(bio
) + segs
- 1) / segs
;
2049 split_bio
= kzalloc(sizeof(*split_bio
), GFP_NOIO
);
2050 BUG_ON(split_bio
== NULL
);
2051 atomic_set(&split_bio
->pending
, pending
);
2052 split_bio
->bio
= bio
;
2053 for (i
= 0; i
< pending
; i
++) {
2054 offset
= (i
* segs
* XEN_PAGE_SIZE
) >> 9;
2055 size
= min((unsigned int)(segs
* XEN_PAGE_SIZE
) >> 9,
2056 (unsigned int)bio_sectors(bio
) - offset
);
2057 cloned_bio
= bio_clone(bio
, GFP_NOIO
);
2058 BUG_ON(cloned_bio
== NULL
);
2059 bio_trim(cloned_bio
, offset
, size
);
2060 cloned_bio
->bi_private
= split_bio
;
2061 cloned_bio
->bi_end_io
= split_bio_end
;
2062 submit_bio(cloned_bio
);
2065 * Now we have to wait for all those smaller bios to
2066 * end, so we can also end the "parent" bio.
2070 /* We don't need to split this bio */
2078 * We are reconnecting to the backend, due to a suspend/resume, or a backend
2079 * driver restart. We tear down our blkif structure and recreate it, but
2080 * leave the device-layer structures intact so that this is transparent to the
2081 * rest of the kernel.
2083 static int blkfront_resume(struct xenbus_device
*dev
)
2085 struct blkfront_info
*info
= dev_get_drvdata(&dev
->dev
);
2089 dev_dbg(&dev
->dev
, "blkfront_resume: %s\n", dev
->nodename
);
2091 bio_list_init(&info
->bio_list
);
2092 INIT_LIST_HEAD(&info
->requests
);
2093 for (i
= 0; i
< info
->nr_rings
; i
++) {
2094 struct blkfront_ring_info
*rinfo
= &info
->rinfo
[i
];
2095 struct bio_list merge_bio
;
2096 struct blk_shadow
*shadow
= rinfo
->shadow
;
2098 for (j
= 0; j
< BLK_RING_SIZE(info
); j
++) {
2100 if (!shadow
[j
].request
)
2104 * Get the bios in the request so we can re-queue them.
2106 if (req_op(shadow
[i
].request
) == REQ_OP_FLUSH
||
2107 req_op(shadow
[i
].request
) == REQ_OP_DISCARD
||
2108 req_op(shadow
[i
].request
) == REQ_OP_SECURE_ERASE
||
2109 shadow
[j
].request
->cmd_flags
& REQ_FUA
) {
2111 * Flush operations don't contain bios, so
2112 * we need to requeue the whole request
2114 * XXX: but this doesn't make any sense for a
2115 * write with the FUA flag set..
2117 list_add(&shadow
[j
].request
->queuelist
, &info
->requests
);
2120 merge_bio
.head
= shadow
[j
].request
->bio
;
2121 merge_bio
.tail
= shadow
[j
].request
->biotail
;
2122 bio_list_merge(&info
->bio_list
, &merge_bio
);
2123 shadow
[j
].request
->bio
= NULL
;
2124 blk_mq_end_request(shadow
[j
].request
, 0);
2128 blkif_free(info
, info
->connected
== BLKIF_STATE_CONNECTED
);
2130 err
= negotiate_mq(info
);
2134 err
= talk_to_blkback(dev
, info
);
2136 blk_mq_update_nr_hw_queues(&info
->tag_set
, info
->nr_rings
);
2139 * We have to wait for the backend to switch to
2140 * connected state, since we want to read which
2141 * features it supports.
2147 static void blkfront_closing(struct blkfront_info
*info
)
2149 struct xenbus_device
*xbdev
= info
->xbdev
;
2150 struct block_device
*bdev
= NULL
;
2152 mutex_lock(&info
->mutex
);
2154 if (xbdev
->state
== XenbusStateClosing
) {
2155 mutex_unlock(&info
->mutex
);
2160 bdev
= bdget_disk(info
->gd
, 0);
2162 mutex_unlock(&info
->mutex
);
2165 xenbus_frontend_closed(xbdev
);
2169 mutex_lock(&bdev
->bd_mutex
);
2171 if (bdev
->bd_openers
) {
2172 xenbus_dev_error(xbdev
, -EBUSY
,
2173 "Device in use; refusing to close");
2174 xenbus_switch_state(xbdev
, XenbusStateClosing
);
2176 xlvbd_release_gendisk(info
);
2177 xenbus_frontend_closed(xbdev
);
2180 mutex_unlock(&bdev
->bd_mutex
);
2184 static void blkfront_setup_discard(struct blkfront_info
*info
)
2187 unsigned int discard_granularity
;
2188 unsigned int discard_alignment
;
2189 unsigned int discard_secure
;
2191 info
->feature_discard
= 1;
2192 err
= xenbus_gather(XBT_NIL
, info
->xbdev
->otherend
,
2193 "discard-granularity", "%u", &discard_granularity
,
2194 "discard-alignment", "%u", &discard_alignment
,
2197 info
->discard_granularity
= discard_granularity
;
2198 info
->discard_alignment
= discard_alignment
;
2200 err
= xenbus_scanf(XBT_NIL
, info
->xbdev
->otherend
,
2201 "discard-secure", "%u", &discard_secure
);
2203 info
->feature_secdiscard
= !!discard_secure
;
2206 static int blkfront_setup_indirect(struct blkfront_ring_info
*rinfo
)
2208 unsigned int psegs
, grants
;
2210 struct blkfront_info
*info
= rinfo
->dev_info
;
2212 if (info
->max_indirect_segments
== 0) {
2214 grants
= BLKIF_MAX_SEGMENTS_PER_REQUEST
;
2217 * When an extra req is required, the maximum
2218 * grants supported is related to the size of the
2219 * Linux block segment.
2221 grants
= GRANTS_PER_PSEG
;
2225 grants
= info
->max_indirect_segments
;
2226 psegs
= grants
/ GRANTS_PER_PSEG
;
2228 err
= fill_grant_buffer(rinfo
,
2229 (grants
+ INDIRECT_GREFS(grants
)) * BLK_RING_SIZE(info
));
2233 if (!info
->feature_persistent
&& info
->max_indirect_segments
) {
2235 * We are using indirect descriptors but not persistent
2236 * grants, we need to allocate a set of pages that can be
2237 * used for mapping indirect grefs
2239 int num
= INDIRECT_GREFS(grants
) * BLK_RING_SIZE(info
);
2241 BUG_ON(!list_empty(&rinfo
->indirect_pages
));
2242 for (i
= 0; i
< num
; i
++) {
2243 struct page
*indirect_page
= alloc_page(GFP_NOIO
);
2246 list_add(&indirect_page
->lru
, &rinfo
->indirect_pages
);
2250 for (i
= 0; i
< BLK_RING_SIZE(info
); i
++) {
2251 rinfo
->shadow
[i
].grants_used
= kzalloc(
2252 sizeof(rinfo
->shadow
[i
].grants_used
[0]) * grants
,
2254 rinfo
->shadow
[i
].sg
= kzalloc(sizeof(rinfo
->shadow
[i
].sg
[0]) * psegs
, GFP_NOIO
);
2255 if (info
->max_indirect_segments
)
2256 rinfo
->shadow
[i
].indirect_grants
= kzalloc(
2257 sizeof(rinfo
->shadow
[i
].indirect_grants
[0]) *
2258 INDIRECT_GREFS(grants
),
2260 if ((rinfo
->shadow
[i
].grants_used
== NULL
) ||
2261 (rinfo
->shadow
[i
].sg
== NULL
) ||
2262 (info
->max_indirect_segments
&&
2263 (rinfo
->shadow
[i
].indirect_grants
== NULL
)))
2265 sg_init_table(rinfo
->shadow
[i
].sg
, psegs
);
2272 for (i
= 0; i
< BLK_RING_SIZE(info
); i
++) {
2273 kfree(rinfo
->shadow
[i
].grants_used
);
2274 rinfo
->shadow
[i
].grants_used
= NULL
;
2275 kfree(rinfo
->shadow
[i
].sg
);
2276 rinfo
->shadow
[i
].sg
= NULL
;
2277 kfree(rinfo
->shadow
[i
].indirect_grants
);
2278 rinfo
->shadow
[i
].indirect_grants
= NULL
;
2280 if (!list_empty(&rinfo
->indirect_pages
)) {
2281 struct page
*indirect_page
, *n
;
2282 list_for_each_entry_safe(indirect_page
, n
, &rinfo
->indirect_pages
, lru
) {
2283 list_del(&indirect_page
->lru
);
2284 __free_page(indirect_page
);
2291 * Gather all backend feature-*
2293 static void blkfront_gather_backend_features(struct blkfront_info
*info
)
2296 int barrier
, flush
, discard
, persistent
;
2297 unsigned int indirect_segments
;
2299 info
->feature_flush
= 0;
2300 info
->feature_fua
= 0;
2302 err
= xenbus_scanf(XBT_NIL
, info
->xbdev
->otherend
,
2303 "feature-barrier", "%d", &barrier
);
2306 * If there's no "feature-barrier" defined, then it means
2307 * we're dealing with a very old backend which writes
2308 * synchronously; nothing to do.
2310 * If there are barriers, then we use flush.
2312 if (err
> 0 && barrier
) {
2313 info
->feature_flush
= 1;
2314 info
->feature_fua
= 1;
2318 * And if there is "feature-flush-cache" use that above
2321 err
= xenbus_scanf(XBT_NIL
, info
->xbdev
->otherend
,
2322 "feature-flush-cache", "%d", &flush
);
2324 if (err
> 0 && flush
) {
2325 info
->feature_flush
= 1;
2326 info
->feature_fua
= 0;
2329 err
= xenbus_scanf(XBT_NIL
, info
->xbdev
->otherend
,
2330 "feature-discard", "%d", &discard
);
2332 if (err
> 0 && discard
)
2333 blkfront_setup_discard(info
);
2335 err
= xenbus_scanf(XBT_NIL
, info
->xbdev
->otherend
,
2336 "feature-persistent", "%d", &persistent
);
2338 info
->feature_persistent
= 0;
2340 info
->feature_persistent
= persistent
;
2342 err
= xenbus_scanf(XBT_NIL
, info
->xbdev
->otherend
,
2343 "feature-max-indirect-segments", "%u",
2344 &indirect_segments
);
2346 info
->max_indirect_segments
= 0;
2348 info
->max_indirect_segments
= min(indirect_segments
,
2349 xen_blkif_max_segments
);
2353 * Invoked when the backend is finally 'ready' (and has told produced
2354 * the details about the physical device - #sectors, size, etc).
2356 static void blkfront_connect(struct blkfront_info
*info
)
2358 unsigned long long sectors
;
2359 unsigned long sector_size
;
2360 unsigned int physical_sector_size
;
2364 switch (info
->connected
) {
2365 case BLKIF_STATE_CONNECTED
:
2367 * Potentially, the back-end may be signalling
2368 * a capacity change; update the capacity.
2370 err
= xenbus_scanf(XBT_NIL
, info
->xbdev
->otherend
,
2371 "sectors", "%Lu", §ors
);
2372 if (XENBUS_EXIST_ERR(err
))
2374 printk(KERN_INFO
"Setting capacity to %Lu\n",
2376 set_capacity(info
->gd
, sectors
);
2377 revalidate_disk(info
->gd
);
2380 case BLKIF_STATE_SUSPENDED
:
2382 * If we are recovering from suspension, we need to wait
2383 * for the backend to announce it's features before
2384 * reconnecting, at least we need to know if the backend
2385 * supports indirect descriptors, and how many.
2387 blkif_recover(info
);
2394 dev_dbg(&info
->xbdev
->dev
, "%s:%s.\n",
2395 __func__
, info
->xbdev
->otherend
);
2397 err
= xenbus_gather(XBT_NIL
, info
->xbdev
->otherend
,
2398 "sectors", "%llu", §ors
,
2399 "info", "%u", &binfo
,
2400 "sector-size", "%lu", §or_size
,
2403 xenbus_dev_fatal(info
->xbdev
, err
,
2404 "reading backend fields at %s",
2405 info
->xbdev
->otherend
);
2410 * physcial-sector-size is a newer field, so old backends may not
2411 * provide this. Assume physical sector size to be the same as
2412 * sector_size in that case.
2414 err
= xenbus_scanf(XBT_NIL
, info
->xbdev
->otherend
,
2415 "physical-sector-size", "%u", &physical_sector_size
);
2417 physical_sector_size
= sector_size
;
2419 blkfront_gather_backend_features(info
);
2420 for (i
= 0; i
< info
->nr_rings
; i
++) {
2421 err
= blkfront_setup_indirect(&info
->rinfo
[i
]);
2423 xenbus_dev_fatal(info
->xbdev
, err
, "setup_indirect at %s",
2424 info
->xbdev
->otherend
);
2425 blkif_free(info
, 0);
2430 err
= xlvbd_alloc_gendisk(sectors
, info
, binfo
, sector_size
,
2431 physical_sector_size
);
2433 xenbus_dev_fatal(info
->xbdev
, err
, "xlvbd_add at %s",
2434 info
->xbdev
->otherend
);
2438 xenbus_switch_state(info
->xbdev
, XenbusStateConnected
);
2440 /* Kick pending requests. */
2441 info
->connected
= BLKIF_STATE_CONNECTED
;
2442 for (i
= 0; i
< info
->nr_rings
; i
++)
2443 kick_pending_request_queues(&info
->rinfo
[i
]);
2445 device_add_disk(&info
->xbdev
->dev
, info
->gd
);
2451 * Callback received when the backend's state changes.
2453 static void blkback_changed(struct xenbus_device
*dev
,
2454 enum xenbus_state backend_state
)
2456 struct blkfront_info
*info
= dev_get_drvdata(&dev
->dev
);
2458 dev_dbg(&dev
->dev
, "blkfront:blkback_changed to state %d.\n", backend_state
);
2460 switch (backend_state
) {
2461 case XenbusStateInitWait
:
2462 if (dev
->state
!= XenbusStateInitialising
)
2464 if (talk_to_blkback(dev
, info
))
2466 case XenbusStateInitialising
:
2467 case XenbusStateInitialised
:
2468 case XenbusStateReconfiguring
:
2469 case XenbusStateReconfigured
:
2470 case XenbusStateUnknown
:
2473 case XenbusStateConnected
:
2475 * talk_to_blkback sets state to XenbusStateInitialised
2476 * and blkfront_connect sets it to XenbusStateConnected
2477 * (if connection went OK).
2479 * If the backend (or toolstack) decides to poke at backend
2480 * state (and re-trigger the watch by setting the state repeatedly
2481 * to XenbusStateConnected (4)) we need to deal with this.
2482 * This is allowed as this is used to communicate to the guest
2483 * that the size of disk has changed!
2485 if ((dev
->state
!= XenbusStateInitialised
) &&
2486 (dev
->state
!= XenbusStateConnected
)) {
2487 if (talk_to_blkback(dev
, info
))
2491 blkfront_connect(info
);
2494 case XenbusStateClosed
:
2495 if (dev
->state
== XenbusStateClosed
)
2497 /* Missed the backend's Closing state -- fallthrough */
2498 case XenbusStateClosing
:
2500 blkfront_closing(info
);
2505 static int blkfront_remove(struct xenbus_device
*xbdev
)
2507 struct blkfront_info
*info
= dev_get_drvdata(&xbdev
->dev
);
2508 struct block_device
*bdev
= NULL
;
2509 struct gendisk
*disk
;
2511 dev_dbg(&xbdev
->dev
, "%s removed", xbdev
->nodename
);
2513 blkif_free(info
, 0);
2515 mutex_lock(&info
->mutex
);
2519 bdev
= bdget_disk(disk
, 0);
2522 mutex_unlock(&info
->mutex
);
2530 * The xbdev was removed before we reached the Closed
2531 * state. See if it's safe to remove the disk. If the bdev
2532 * isn't closed yet, we let release take care of it.
2535 mutex_lock(&bdev
->bd_mutex
);
2536 info
= disk
->private_data
;
2538 dev_warn(disk_to_dev(disk
),
2539 "%s was hot-unplugged, %d stale handles\n",
2540 xbdev
->nodename
, bdev
->bd_openers
);
2542 if (info
&& !bdev
->bd_openers
) {
2543 xlvbd_release_gendisk(info
);
2544 disk
->private_data
= NULL
;
2548 mutex_unlock(&bdev
->bd_mutex
);
2554 static int blkfront_is_ready(struct xenbus_device
*dev
)
2556 struct blkfront_info
*info
= dev_get_drvdata(&dev
->dev
);
2558 return info
->is_ready
&& info
->xbdev
;
2561 static int blkif_open(struct block_device
*bdev
, fmode_t mode
)
2563 struct gendisk
*disk
= bdev
->bd_disk
;
2564 struct blkfront_info
*info
;
2567 mutex_lock(&blkfront_mutex
);
2569 info
= disk
->private_data
;
2576 mutex_lock(&info
->mutex
);
2579 /* xbdev is closed */
2582 mutex_unlock(&info
->mutex
);
2585 mutex_unlock(&blkfront_mutex
);
2589 static void blkif_release(struct gendisk
*disk
, fmode_t mode
)
2591 struct blkfront_info
*info
= disk
->private_data
;
2592 struct block_device
*bdev
;
2593 struct xenbus_device
*xbdev
;
2595 mutex_lock(&blkfront_mutex
);
2597 bdev
= bdget_disk(disk
, 0);
2600 WARN(1, "Block device %s yanked out from us!\n", disk
->disk_name
);
2603 if (bdev
->bd_openers
)
2607 * Check if we have been instructed to close. We will have
2608 * deferred this request, because the bdev was still open.
2611 mutex_lock(&info
->mutex
);
2612 xbdev
= info
->xbdev
;
2614 if (xbdev
&& xbdev
->state
== XenbusStateClosing
) {
2615 /* pending switch to state closed */
2616 dev_info(disk_to_dev(bdev
->bd_disk
), "releasing disk\n");
2617 xlvbd_release_gendisk(info
);
2618 xenbus_frontend_closed(info
->xbdev
);
2621 mutex_unlock(&info
->mutex
);
2624 /* sudden device removal */
2625 dev_info(disk_to_dev(bdev
->bd_disk
), "releasing disk\n");
2626 xlvbd_release_gendisk(info
);
2627 disk
->private_data
= NULL
;
2634 mutex_unlock(&blkfront_mutex
);
2637 static const struct block_device_operations xlvbd_block_fops
=
2639 .owner
= THIS_MODULE
,
2641 .release
= blkif_release
,
2642 .getgeo
= blkif_getgeo
,
2643 .ioctl
= blkif_ioctl
,
2647 static const struct xenbus_device_id blkfront_ids
[] = {
2652 static struct xenbus_driver blkfront_driver
= {
2653 .ids
= blkfront_ids
,
2654 .probe
= blkfront_probe
,
2655 .remove
= blkfront_remove
,
2656 .resume
= blkfront_resume
,
2657 .otherend_changed
= blkback_changed
,
2658 .is_ready
= blkfront_is_ready
,
2661 static int __init
xlblk_init(void)
2664 int nr_cpus
= num_online_cpus();
2669 if (xen_blkif_max_ring_order
> XENBUS_MAX_RING_GRANT_ORDER
) {
2670 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2671 xen_blkif_max_ring_order
, XENBUS_MAX_RING_GRANT_ORDER
);
2672 xen_blkif_max_ring_order
= XENBUS_MAX_RING_GRANT_ORDER
;
2675 if (xen_blkif_max_queues
> nr_cpus
) {
2676 pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2677 xen_blkif_max_queues
, nr_cpus
);
2678 xen_blkif_max_queues
= nr_cpus
;
2681 if (!xen_has_pv_disk_devices())
2684 if (register_blkdev(XENVBD_MAJOR
, DEV_NAME
)) {
2685 printk(KERN_WARNING
"xen_blk: can't get major %d with name %s\n",
2686 XENVBD_MAJOR
, DEV_NAME
);
2690 ret
= xenbus_register_frontend(&blkfront_driver
);
2692 unregister_blkdev(XENVBD_MAJOR
, DEV_NAME
);
2698 module_init(xlblk_init
);
2701 static void __exit
xlblk_exit(void)
2703 xenbus_unregister_driver(&blkfront_driver
);
2704 unregister_blkdev(XENVBD_MAJOR
, DEV_NAME
);
2707 module_exit(xlblk_exit
);
2709 MODULE_DESCRIPTION("Xen virtual block device frontend");
2710 MODULE_LICENSE("GPL");
2711 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR
);
2712 MODULE_ALIAS("xen:vbd");
2713 MODULE_ALIAS("xenblk");