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
;
193 unsigned int physical_sector_size
;
196 enum blkif_state connected
;
197 /* Number of pages per ring buffer. */
198 unsigned int nr_ring_pages
;
199 struct request_queue
*rq
;
200 unsigned int feature_flush
;
201 unsigned int feature_fua
;
202 unsigned int feature_discard
:1;
203 unsigned int feature_secdiscard
:1;
204 unsigned int discard_granularity
;
205 unsigned int discard_alignment
;
206 unsigned int feature_persistent
:1;
207 /* Number of 4KB segments handled */
208 unsigned int max_indirect_segments
;
210 struct blk_mq_tag_set tag_set
;
211 struct blkfront_ring_info
*rinfo
;
212 unsigned int nr_rings
;
213 /* Save uncomplete reqs and bios for migration. */
214 struct list_head requests
;
215 struct bio_list bio_list
;
218 static unsigned int nr_minors
;
219 static unsigned long *minors
;
220 static DEFINE_SPINLOCK(minor_lock
);
222 #define GRANT_INVALID_REF 0
224 #define PARTS_PER_DISK 16
225 #define PARTS_PER_EXT_DISK 256
227 #define BLKIF_MAJOR(dev) ((dev)>>8)
228 #define BLKIF_MINOR(dev) ((dev) & 0xff)
231 #define EXTENDED (1<<EXT_SHIFT)
232 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
233 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
234 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
235 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
236 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
237 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
239 #define DEV_NAME "xvd" /* name in /dev */
242 * Grants are always the same size as a Xen page (i.e 4KB).
243 * A physical segment is always the same size as a Linux page.
244 * Number of grants per physical segment
246 #define GRANTS_PER_PSEG (PAGE_SIZE / XEN_PAGE_SIZE)
248 #define GRANTS_PER_INDIRECT_FRAME \
249 (XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
251 #define PSEGS_PER_INDIRECT_FRAME \
252 (GRANTS_INDIRECT_FRAME / GRANTS_PSEGS)
254 #define INDIRECT_GREFS(_grants) \
255 DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
257 #define GREFS(_psegs) ((_psegs) * GRANTS_PER_PSEG)
259 static int blkfront_setup_indirect(struct blkfront_ring_info
*rinfo
);
260 static void blkfront_gather_backend_features(struct blkfront_info
*info
);
262 static int get_id_from_freelist(struct blkfront_ring_info
*rinfo
)
264 unsigned long free
= rinfo
->shadow_free
;
266 BUG_ON(free
>= BLK_RING_SIZE(rinfo
->dev_info
));
267 rinfo
->shadow_free
= rinfo
->shadow
[free
].req
.u
.rw
.id
;
268 rinfo
->shadow
[free
].req
.u
.rw
.id
= 0x0fffffee; /* debug */
272 static int add_id_to_freelist(struct blkfront_ring_info
*rinfo
,
275 if (rinfo
->shadow
[id
].req
.u
.rw
.id
!= id
)
277 if (rinfo
->shadow
[id
].request
== NULL
)
279 rinfo
->shadow
[id
].req
.u
.rw
.id
= rinfo
->shadow_free
;
280 rinfo
->shadow
[id
].request
= NULL
;
281 rinfo
->shadow_free
= id
;
285 static int fill_grant_buffer(struct blkfront_ring_info
*rinfo
, int num
)
287 struct blkfront_info
*info
= rinfo
->dev_info
;
288 struct page
*granted_page
;
289 struct grant
*gnt_list_entry
, *n
;
293 gnt_list_entry
= kzalloc(sizeof(struct grant
), GFP_NOIO
);
297 if (info
->feature_persistent
) {
298 granted_page
= alloc_page(GFP_NOIO
);
300 kfree(gnt_list_entry
);
303 gnt_list_entry
->page
= granted_page
;
306 gnt_list_entry
->gref
= GRANT_INVALID_REF
;
307 list_add(&gnt_list_entry
->node
, &rinfo
->grants
);
314 list_for_each_entry_safe(gnt_list_entry
, n
,
315 &rinfo
->grants
, node
) {
316 list_del(&gnt_list_entry
->node
);
317 if (info
->feature_persistent
)
318 __free_page(gnt_list_entry
->page
);
319 kfree(gnt_list_entry
);
326 static struct grant
*get_free_grant(struct blkfront_ring_info
*rinfo
)
328 struct grant
*gnt_list_entry
;
330 BUG_ON(list_empty(&rinfo
->grants
));
331 gnt_list_entry
= list_first_entry(&rinfo
->grants
, struct grant
,
333 list_del(&gnt_list_entry
->node
);
335 if (gnt_list_entry
->gref
!= GRANT_INVALID_REF
)
336 rinfo
->persistent_gnts_c
--;
338 return gnt_list_entry
;
341 static inline void grant_foreign_access(const struct grant
*gnt_list_entry
,
342 const struct blkfront_info
*info
)
344 gnttab_page_grant_foreign_access_ref_one(gnt_list_entry
->gref
,
345 info
->xbdev
->otherend_id
,
346 gnt_list_entry
->page
,
350 static struct grant
*get_grant(grant_ref_t
*gref_head
,
352 struct blkfront_ring_info
*rinfo
)
354 struct grant
*gnt_list_entry
= get_free_grant(rinfo
);
355 struct blkfront_info
*info
= rinfo
->dev_info
;
357 if (gnt_list_entry
->gref
!= GRANT_INVALID_REF
)
358 return gnt_list_entry
;
360 /* Assign a gref to this page */
361 gnt_list_entry
->gref
= gnttab_claim_grant_reference(gref_head
);
362 BUG_ON(gnt_list_entry
->gref
== -ENOSPC
);
363 if (info
->feature_persistent
)
364 grant_foreign_access(gnt_list_entry
, info
);
366 /* Grant access to the GFN passed by the caller */
367 gnttab_grant_foreign_access_ref(gnt_list_entry
->gref
,
368 info
->xbdev
->otherend_id
,
372 return gnt_list_entry
;
375 static struct grant
*get_indirect_grant(grant_ref_t
*gref_head
,
376 struct blkfront_ring_info
*rinfo
)
378 struct grant
*gnt_list_entry
= get_free_grant(rinfo
);
379 struct blkfront_info
*info
= rinfo
->dev_info
;
381 if (gnt_list_entry
->gref
!= GRANT_INVALID_REF
)
382 return gnt_list_entry
;
384 /* Assign a gref to this page */
385 gnt_list_entry
->gref
= gnttab_claim_grant_reference(gref_head
);
386 BUG_ON(gnt_list_entry
->gref
== -ENOSPC
);
387 if (!info
->feature_persistent
) {
388 struct page
*indirect_page
;
390 /* Fetch a pre-allocated page to use for indirect grefs */
391 BUG_ON(list_empty(&rinfo
->indirect_pages
));
392 indirect_page
= list_first_entry(&rinfo
->indirect_pages
,
394 list_del(&indirect_page
->lru
);
395 gnt_list_entry
->page
= indirect_page
;
397 grant_foreign_access(gnt_list_entry
, info
);
399 return gnt_list_entry
;
402 static const char *op_name(int op
)
404 static const char *const names
[] = {
405 [BLKIF_OP_READ
] = "read",
406 [BLKIF_OP_WRITE
] = "write",
407 [BLKIF_OP_WRITE_BARRIER
] = "barrier",
408 [BLKIF_OP_FLUSH_DISKCACHE
] = "flush",
409 [BLKIF_OP_DISCARD
] = "discard" };
411 if (op
< 0 || op
>= ARRAY_SIZE(names
))
419 static int xlbd_reserve_minors(unsigned int minor
, unsigned int nr
)
421 unsigned int end
= minor
+ nr
;
424 if (end
> nr_minors
) {
425 unsigned long *bitmap
, *old
;
427 bitmap
= kcalloc(BITS_TO_LONGS(end
), sizeof(*bitmap
),
432 spin_lock(&minor_lock
);
433 if (end
> nr_minors
) {
435 memcpy(bitmap
, minors
,
436 BITS_TO_LONGS(nr_minors
) * sizeof(*bitmap
));
438 nr_minors
= BITS_TO_LONGS(end
) * BITS_PER_LONG
;
441 spin_unlock(&minor_lock
);
445 spin_lock(&minor_lock
);
446 if (find_next_bit(minors
, end
, minor
) >= end
) {
447 bitmap_set(minors
, minor
, nr
);
451 spin_unlock(&minor_lock
);
456 static void xlbd_release_minors(unsigned int minor
, unsigned int nr
)
458 unsigned int end
= minor
+ nr
;
460 BUG_ON(end
> nr_minors
);
461 spin_lock(&minor_lock
);
462 bitmap_clear(minors
, minor
, nr
);
463 spin_unlock(&minor_lock
);
466 static void blkif_restart_queue_callback(void *arg
)
468 struct blkfront_ring_info
*rinfo
= (struct blkfront_ring_info
*)arg
;
469 schedule_work(&rinfo
->work
);
472 static int blkif_getgeo(struct block_device
*bd
, struct hd_geometry
*hg
)
474 /* We don't have real geometry info, but let's at least return
475 values consistent with the size of the device */
476 sector_t nsect
= get_capacity(bd
->bd_disk
);
477 sector_t cylinders
= nsect
;
481 sector_div(cylinders
, hg
->heads
* hg
->sectors
);
482 hg
->cylinders
= cylinders
;
483 if ((sector_t
)(hg
->cylinders
+ 1) * hg
->heads
* hg
->sectors
< nsect
)
484 hg
->cylinders
= 0xffff;
488 static int blkif_ioctl(struct block_device
*bdev
, fmode_t mode
,
489 unsigned command
, unsigned long argument
)
491 struct blkfront_info
*info
= bdev
->bd_disk
->private_data
;
494 dev_dbg(&info
->xbdev
->dev
, "command: 0x%x, argument: 0x%lx\n",
495 command
, (long)argument
);
498 case CDROMMULTISESSION
:
499 dev_dbg(&info
->xbdev
->dev
, "FIXME: support multisession CDs later\n");
500 for (i
= 0; i
< sizeof(struct cdrom_multisession
); i
++)
501 if (put_user(0, (char __user
*)(argument
+ i
)))
505 case CDROM_GET_CAPABILITY
: {
506 struct gendisk
*gd
= info
->gd
;
507 if (gd
->flags
& GENHD_FL_CD
)
513 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
515 return -EINVAL
; /* same return as native Linux */
521 static unsigned long blkif_ring_get_request(struct blkfront_ring_info
*rinfo
,
523 struct blkif_request
**ring_req
)
527 *ring_req
= RING_GET_REQUEST(&rinfo
->ring
, rinfo
->ring
.req_prod_pvt
);
528 rinfo
->ring
.req_prod_pvt
++;
530 id
= get_id_from_freelist(rinfo
);
531 rinfo
->shadow
[id
].request
= req
;
532 rinfo
->shadow
[id
].status
= REQ_WAITING
;
533 rinfo
->shadow
[id
].associated_id
= NO_ASSOCIATED_ID
;
535 (*ring_req
)->u
.rw
.id
= id
;
540 static int blkif_queue_discard_req(struct request
*req
, struct blkfront_ring_info
*rinfo
)
542 struct blkfront_info
*info
= rinfo
->dev_info
;
543 struct blkif_request
*ring_req
;
546 /* Fill out a communications ring structure. */
547 id
= blkif_ring_get_request(rinfo
, req
, &ring_req
);
549 ring_req
->operation
= BLKIF_OP_DISCARD
;
550 ring_req
->u
.discard
.nr_sectors
= blk_rq_sectors(req
);
551 ring_req
->u
.discard
.id
= id
;
552 ring_req
->u
.discard
.sector_number
= (blkif_sector_t
)blk_rq_pos(req
);
553 if (req_op(req
) == REQ_OP_SECURE_ERASE
&& info
->feature_secdiscard
)
554 ring_req
->u
.discard
.flag
= BLKIF_DISCARD_SECURE
;
556 ring_req
->u
.discard
.flag
= 0;
558 /* Keep a private copy so we can reissue requests when recovering. */
559 rinfo
->shadow
[id
].req
= *ring_req
;
564 struct setup_rw_req
{
565 unsigned int grant_idx
;
566 struct blkif_request_segment
*segments
;
567 struct blkfront_ring_info
*rinfo
;
568 struct blkif_request
*ring_req
;
569 grant_ref_t gref_head
;
571 /* Only used when persistent grant is used and it's a read request */
573 unsigned int bvec_off
;
576 bool require_extra_req
;
577 struct blkif_request
*extra_ring_req
;
580 static void blkif_setup_rw_req_grant(unsigned long gfn
, unsigned int offset
,
581 unsigned int len
, void *data
)
583 struct setup_rw_req
*setup
= data
;
585 struct grant
*gnt_list_entry
;
586 unsigned int fsect
, lsect
;
587 /* Convenient aliases */
588 unsigned int grant_idx
= setup
->grant_idx
;
589 struct blkif_request
*ring_req
= setup
->ring_req
;
590 struct blkfront_ring_info
*rinfo
= setup
->rinfo
;
592 * We always use the shadow of the first request to store the list
593 * of grant associated to the block I/O request. This made the
594 * completion more easy to handle even if the block I/O request is
597 struct blk_shadow
*shadow
= &rinfo
->shadow
[setup
->id
];
599 if (unlikely(setup
->require_extra_req
&&
600 grant_idx
>= BLKIF_MAX_SEGMENTS_PER_REQUEST
)) {
602 * We are using the second request, setup grant_idx
603 * to be the index of the segment array.
605 grant_idx
-= BLKIF_MAX_SEGMENTS_PER_REQUEST
;
606 ring_req
= setup
->extra_ring_req
;
609 if ((ring_req
->operation
== BLKIF_OP_INDIRECT
) &&
610 (grant_idx
% GRANTS_PER_INDIRECT_FRAME
== 0)) {
612 kunmap_atomic(setup
->segments
);
614 n
= grant_idx
/ GRANTS_PER_INDIRECT_FRAME
;
615 gnt_list_entry
= get_indirect_grant(&setup
->gref_head
, rinfo
);
616 shadow
->indirect_grants
[n
] = gnt_list_entry
;
617 setup
->segments
= kmap_atomic(gnt_list_entry
->page
);
618 ring_req
->u
.indirect
.indirect_grefs
[n
] = gnt_list_entry
->gref
;
621 gnt_list_entry
= get_grant(&setup
->gref_head
, gfn
, rinfo
);
622 ref
= gnt_list_entry
->gref
;
624 * All the grants are stored in the shadow of the first
625 * request. Therefore we have to use the global index.
627 shadow
->grants_used
[setup
->grant_idx
] = gnt_list_entry
;
629 if (setup
->need_copy
) {
632 shared_data
= kmap_atomic(gnt_list_entry
->page
);
634 * this does not wipe data stored outside the
635 * range sg->offset..sg->offset+sg->length.
636 * Therefore, blkback *could* see data from
637 * previous requests. This is OK as long as
638 * persistent grants are shared with just one
639 * domain. It may need refactoring if this
642 memcpy(shared_data
+ offset
,
643 setup
->bvec_data
+ setup
->bvec_off
,
646 kunmap_atomic(shared_data
);
647 setup
->bvec_off
+= len
;
651 lsect
= fsect
+ (len
>> 9) - 1;
652 if (ring_req
->operation
!= BLKIF_OP_INDIRECT
) {
653 ring_req
->u
.rw
.seg
[grant_idx
] =
654 (struct blkif_request_segment
) {
657 .last_sect
= lsect
};
659 setup
->segments
[grant_idx
% GRANTS_PER_INDIRECT_FRAME
] =
660 (struct blkif_request_segment
) {
663 .last_sect
= lsect
};
666 (setup
->grant_idx
)++;
669 static void blkif_setup_extra_req(struct blkif_request
*first
,
670 struct blkif_request
*second
)
672 uint16_t nr_segments
= first
->u
.rw
.nr_segments
;
675 * The second request is only present when the first request uses
676 * all its segments. It's always the continuity of the first one.
678 first
->u
.rw
.nr_segments
= BLKIF_MAX_SEGMENTS_PER_REQUEST
;
680 second
->u
.rw
.nr_segments
= nr_segments
- BLKIF_MAX_SEGMENTS_PER_REQUEST
;
681 second
->u
.rw
.sector_number
= first
->u
.rw
.sector_number
+
682 (BLKIF_MAX_SEGMENTS_PER_REQUEST
* XEN_PAGE_SIZE
) / 512;
684 second
->u
.rw
.handle
= first
->u
.rw
.handle
;
685 second
->operation
= first
->operation
;
688 static int blkif_queue_rw_req(struct request
*req
, struct blkfront_ring_info
*rinfo
)
690 struct blkfront_info
*info
= rinfo
->dev_info
;
691 struct blkif_request
*ring_req
, *extra_ring_req
= NULL
;
692 unsigned long id
, extra_id
= NO_ASSOCIATED_ID
;
693 bool require_extra_req
= false;
695 struct setup_rw_req setup
= {
699 .need_copy
= rq_data_dir(req
) && info
->feature_persistent
,
703 * Used to store if we are able to queue the request by just using
704 * existing persistent grants, or if we have to get new grants,
705 * as there are not sufficiently many free.
707 struct scatterlist
*sg
;
708 int num_sg
, max_grefs
, num_grant
;
710 max_grefs
= req
->nr_phys_segments
* GRANTS_PER_PSEG
;
711 if (max_grefs
> BLKIF_MAX_SEGMENTS_PER_REQUEST
)
713 * If we are using indirect segments we need to account
714 * for the indirect grefs used in the request.
716 max_grefs
+= INDIRECT_GREFS(max_grefs
);
719 * We have to reserve 'max_grefs' grants because persistent
720 * grants are shared by all rings.
723 if (gnttab_alloc_grant_references(max_grefs
, &setup
.gref_head
) < 0) {
724 gnttab_request_free_callback(
726 blkif_restart_queue_callback
,
732 /* Fill out a communications ring structure. */
733 id
= blkif_ring_get_request(rinfo
, req
, &ring_req
);
735 num_sg
= blk_rq_map_sg(req
->q
, req
, rinfo
->shadow
[id
].sg
);
737 /* Calculate the number of grant used */
738 for_each_sg(rinfo
->shadow
[id
].sg
, sg
, num_sg
, i
)
739 num_grant
+= gnttab_count_grant(sg
->offset
, sg
->length
);
741 require_extra_req
= info
->max_indirect_segments
== 0 &&
742 num_grant
> BLKIF_MAX_SEGMENTS_PER_REQUEST
;
743 BUG_ON(!HAS_EXTRA_REQ
&& require_extra_req
);
745 rinfo
->shadow
[id
].num_sg
= num_sg
;
746 if (num_grant
> BLKIF_MAX_SEGMENTS_PER_REQUEST
&&
747 likely(!require_extra_req
)) {
749 * The indirect operation can only be a BLKIF_OP_READ or
752 BUG_ON(req_op(req
) == REQ_OP_FLUSH
|| req
->cmd_flags
& REQ_FUA
);
753 ring_req
->operation
= BLKIF_OP_INDIRECT
;
754 ring_req
->u
.indirect
.indirect_op
= rq_data_dir(req
) ?
755 BLKIF_OP_WRITE
: BLKIF_OP_READ
;
756 ring_req
->u
.indirect
.sector_number
= (blkif_sector_t
)blk_rq_pos(req
);
757 ring_req
->u
.indirect
.handle
= info
->handle
;
758 ring_req
->u
.indirect
.nr_segments
= num_grant
;
760 ring_req
->u
.rw
.sector_number
= (blkif_sector_t
)blk_rq_pos(req
);
761 ring_req
->u
.rw
.handle
= info
->handle
;
762 ring_req
->operation
= rq_data_dir(req
) ?
763 BLKIF_OP_WRITE
: BLKIF_OP_READ
;
764 if (req_op(req
) == REQ_OP_FLUSH
|| req
->cmd_flags
& REQ_FUA
) {
766 * Ideally we can do an unordered flush-to-disk.
767 * In case the backend onlysupports barriers, use that.
768 * A barrier request a superset of FUA, so we can
769 * implement it the same way. (It's also a FLUSH+FUA,
770 * since it is guaranteed ordered WRT previous writes.)
772 if (info
->feature_flush
&& info
->feature_fua
)
773 ring_req
->operation
=
774 BLKIF_OP_WRITE_BARRIER
;
775 else if (info
->feature_flush
)
776 ring_req
->operation
=
777 BLKIF_OP_FLUSH_DISKCACHE
;
779 ring_req
->operation
= 0;
781 ring_req
->u
.rw
.nr_segments
= num_grant
;
782 if (unlikely(require_extra_req
)) {
783 extra_id
= blkif_ring_get_request(rinfo
, req
,
786 * Only the first request contains the scatter-gather
789 rinfo
->shadow
[extra_id
].num_sg
= 0;
791 blkif_setup_extra_req(ring_req
, extra_ring_req
);
793 /* Link the 2 requests together */
794 rinfo
->shadow
[extra_id
].associated_id
= id
;
795 rinfo
->shadow
[id
].associated_id
= extra_id
;
799 setup
.ring_req
= ring_req
;
802 setup
.require_extra_req
= require_extra_req
;
803 if (unlikely(require_extra_req
))
804 setup
.extra_ring_req
= extra_ring_req
;
806 for_each_sg(rinfo
->shadow
[id
].sg
, sg
, num_sg
, i
) {
807 BUG_ON(sg
->offset
+ sg
->length
> PAGE_SIZE
);
809 if (setup
.need_copy
) {
810 setup
.bvec_off
= sg
->offset
;
811 setup
.bvec_data
= kmap_atomic(sg_page(sg
));
814 gnttab_foreach_grant_in_range(sg_page(sg
),
817 blkif_setup_rw_req_grant
,
821 kunmap_atomic(setup
.bvec_data
);
824 kunmap_atomic(setup
.segments
);
826 /* Keep a private copy so we can reissue requests when recovering. */
827 rinfo
->shadow
[id
].req
= *ring_req
;
828 if (unlikely(require_extra_req
))
829 rinfo
->shadow
[extra_id
].req
= *extra_ring_req
;
832 gnttab_free_grant_references(setup
.gref_head
);
838 * Generate a Xen blkfront IO request from a blk layer request. Reads
839 * and writes are handled as expected.
841 * @req: a request struct
843 static int blkif_queue_request(struct request
*req
, struct blkfront_ring_info
*rinfo
)
845 if (unlikely(rinfo
->dev_info
->connected
!= BLKIF_STATE_CONNECTED
))
848 if (unlikely(req_op(req
) == REQ_OP_DISCARD
||
849 req_op(req
) == REQ_OP_SECURE_ERASE
))
850 return blkif_queue_discard_req(req
, rinfo
);
852 return blkif_queue_rw_req(req
, rinfo
);
855 static inline void flush_requests(struct blkfront_ring_info
*rinfo
)
859 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo
->ring
, notify
);
862 notify_remote_via_irq(rinfo
->irq
);
865 static inline bool blkif_request_flush_invalid(struct request
*req
,
866 struct blkfront_info
*info
)
868 return (blk_rq_is_passthrough(req
) ||
869 ((req_op(req
) == REQ_OP_FLUSH
) &&
870 !info
->feature_flush
) ||
871 ((req
->cmd_flags
& REQ_FUA
) &&
872 !info
->feature_fua
));
875 static int blkif_queue_rq(struct blk_mq_hw_ctx
*hctx
,
876 const struct blk_mq_queue_data
*qd
)
879 int qid
= hctx
->queue_num
;
880 struct blkfront_info
*info
= hctx
->queue
->queuedata
;
881 struct blkfront_ring_info
*rinfo
= NULL
;
883 BUG_ON(info
->nr_rings
<= qid
);
884 rinfo
= &info
->rinfo
[qid
];
885 blk_mq_start_request(qd
->rq
);
886 spin_lock_irqsave(&rinfo
->ring_lock
, flags
);
887 if (RING_FULL(&rinfo
->ring
))
890 if (blkif_request_flush_invalid(qd
->rq
, rinfo
->dev_info
))
893 if (blkif_queue_request(qd
->rq
, rinfo
))
896 flush_requests(rinfo
);
897 spin_unlock_irqrestore(&rinfo
->ring_lock
, flags
);
898 return BLK_MQ_RQ_QUEUE_OK
;
901 spin_unlock_irqrestore(&rinfo
->ring_lock
, flags
);
902 return BLK_MQ_RQ_QUEUE_ERROR
;
905 spin_unlock_irqrestore(&rinfo
->ring_lock
, flags
);
906 blk_mq_stop_hw_queue(hctx
);
907 return BLK_MQ_RQ_QUEUE_BUSY
;
910 static struct blk_mq_ops blkfront_mq_ops
= {
911 .queue_rq
= blkif_queue_rq
,
914 static void blkif_set_queue_limits(struct blkfront_info
*info
)
916 struct request_queue
*rq
= info
->rq
;
917 struct gendisk
*gd
= info
->gd
;
918 unsigned int segments
= info
->max_indirect_segments
? :
919 BLKIF_MAX_SEGMENTS_PER_REQUEST
;
921 queue_flag_set_unlocked(QUEUE_FLAG_VIRT
, rq
);
923 if (info
->feature_discard
) {
924 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, rq
);
925 blk_queue_max_discard_sectors(rq
, get_capacity(gd
));
926 rq
->limits
.discard_granularity
= info
->discard_granularity
;
927 rq
->limits
.discard_alignment
= info
->discard_alignment
;
928 if (info
->feature_secdiscard
)
929 queue_flag_set_unlocked(QUEUE_FLAG_SECERASE
, rq
);
932 /* Hard sector size and max sectors impersonate the equiv. hardware. */
933 blk_queue_logical_block_size(rq
, info
->sector_size
);
934 blk_queue_physical_block_size(rq
, info
->physical_sector_size
);
935 blk_queue_max_hw_sectors(rq
, (segments
* XEN_PAGE_SIZE
) / 512);
937 /* Each segment in a request is up to an aligned page in size. */
938 blk_queue_segment_boundary(rq
, PAGE_SIZE
- 1);
939 blk_queue_max_segment_size(rq
, PAGE_SIZE
);
941 /* Ensure a merged request will fit in a single I/O ring slot. */
942 blk_queue_max_segments(rq
, segments
/ GRANTS_PER_PSEG
);
944 /* Make sure buffer addresses are sector-aligned. */
945 blk_queue_dma_alignment(rq
, 511);
947 /* Make sure we don't use bounce buffers. */
948 blk_queue_bounce_limit(rq
, BLK_BOUNCE_ANY
);
951 static int xlvbd_init_blk_queue(struct gendisk
*gd
, u16 sector_size
,
952 unsigned int physical_sector_size
)
954 struct request_queue
*rq
;
955 struct blkfront_info
*info
= gd
->private_data
;
957 memset(&info
->tag_set
, 0, sizeof(info
->tag_set
));
958 info
->tag_set
.ops
= &blkfront_mq_ops
;
959 info
->tag_set
.nr_hw_queues
= info
->nr_rings
;
960 if (HAS_EXTRA_REQ
&& info
->max_indirect_segments
== 0) {
962 * When indirect descriptior is not supported, the I/O request
963 * will be split between multiple request in the ring.
964 * To avoid problems when sending the request, divide by
965 * 2 the depth of the queue.
967 info
->tag_set
.queue_depth
= BLK_RING_SIZE(info
) / 2;
969 info
->tag_set
.queue_depth
= BLK_RING_SIZE(info
);
970 info
->tag_set
.numa_node
= NUMA_NO_NODE
;
971 info
->tag_set
.flags
= BLK_MQ_F_SHOULD_MERGE
| BLK_MQ_F_SG_MERGE
;
972 info
->tag_set
.cmd_size
= 0;
973 info
->tag_set
.driver_data
= info
;
975 if (blk_mq_alloc_tag_set(&info
->tag_set
))
977 rq
= blk_mq_init_queue(&info
->tag_set
);
979 blk_mq_free_tag_set(&info
->tag_set
);
983 rq
->queuedata
= info
;
984 info
->rq
= gd
->queue
= rq
;
986 info
->sector_size
= sector_size
;
987 info
->physical_sector_size
= physical_sector_size
;
988 blkif_set_queue_limits(info
);
993 static const char *flush_info(struct blkfront_info
*info
)
995 if (info
->feature_flush
&& info
->feature_fua
)
996 return "barrier: enabled;";
997 else if (info
->feature_flush
)
998 return "flush diskcache: enabled;";
1000 return "barrier or flush: disabled;";
1003 static void xlvbd_flush(struct blkfront_info
*info
)
1005 blk_queue_write_cache(info
->rq
, info
->feature_flush
? true : false,
1006 info
->feature_fua
? true : false);
1007 pr_info("blkfront: %s: %s %s %s %s %s\n",
1008 info
->gd
->disk_name
, flush_info(info
),
1009 "persistent grants:", info
->feature_persistent
?
1010 "enabled;" : "disabled;", "indirect descriptors:",
1011 info
->max_indirect_segments
? "enabled;" : "disabled;");
1014 static int xen_translate_vdev(int vdevice
, int *minor
, unsigned int *offset
)
1017 major
= BLKIF_MAJOR(vdevice
);
1018 *minor
= BLKIF_MINOR(vdevice
);
1020 case XEN_IDE0_MAJOR
:
1021 *offset
= (*minor
/ 64) + EMULATED_HD_DISK_NAME_OFFSET
;
1022 *minor
= ((*minor
/ 64) * PARTS_PER_DISK
) +
1023 EMULATED_HD_DISK_MINOR_OFFSET
;
1025 case XEN_IDE1_MAJOR
:
1026 *offset
= (*minor
/ 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET
;
1027 *minor
= (((*minor
/ 64) + 2) * PARTS_PER_DISK
) +
1028 EMULATED_HD_DISK_MINOR_OFFSET
;
1030 case XEN_SCSI_DISK0_MAJOR
:
1031 *offset
= (*minor
/ PARTS_PER_DISK
) + EMULATED_SD_DISK_NAME_OFFSET
;
1032 *minor
= *minor
+ EMULATED_SD_DISK_MINOR_OFFSET
;
1034 case XEN_SCSI_DISK1_MAJOR
:
1035 case XEN_SCSI_DISK2_MAJOR
:
1036 case XEN_SCSI_DISK3_MAJOR
:
1037 case XEN_SCSI_DISK4_MAJOR
:
1038 case XEN_SCSI_DISK5_MAJOR
:
1039 case XEN_SCSI_DISK6_MAJOR
:
1040 case XEN_SCSI_DISK7_MAJOR
:
1041 *offset
= (*minor
/ PARTS_PER_DISK
) +
1042 ((major
- XEN_SCSI_DISK1_MAJOR
+ 1) * 16) +
1043 EMULATED_SD_DISK_NAME_OFFSET
;
1045 ((major
- XEN_SCSI_DISK1_MAJOR
+ 1) * 16 * PARTS_PER_DISK
) +
1046 EMULATED_SD_DISK_MINOR_OFFSET
;
1048 case XEN_SCSI_DISK8_MAJOR
:
1049 case XEN_SCSI_DISK9_MAJOR
:
1050 case XEN_SCSI_DISK10_MAJOR
:
1051 case XEN_SCSI_DISK11_MAJOR
:
1052 case XEN_SCSI_DISK12_MAJOR
:
1053 case XEN_SCSI_DISK13_MAJOR
:
1054 case XEN_SCSI_DISK14_MAJOR
:
1055 case XEN_SCSI_DISK15_MAJOR
:
1056 *offset
= (*minor
/ PARTS_PER_DISK
) +
1057 ((major
- XEN_SCSI_DISK8_MAJOR
+ 8) * 16) +
1058 EMULATED_SD_DISK_NAME_OFFSET
;
1060 ((major
- XEN_SCSI_DISK8_MAJOR
+ 8) * 16 * PARTS_PER_DISK
) +
1061 EMULATED_SD_DISK_MINOR_OFFSET
;
1064 *offset
= *minor
/ PARTS_PER_DISK
;
1067 printk(KERN_WARNING
"blkfront: your disk configuration is "
1068 "incorrect, please use an xvd device instead\n");
1074 static char *encode_disk_name(char *ptr
, unsigned int n
)
1077 ptr
= encode_disk_name(ptr
, n
/ 26 - 1);
1078 *ptr
= 'a' + n
% 26;
1082 static int xlvbd_alloc_gendisk(blkif_sector_t capacity
,
1083 struct blkfront_info
*info
,
1084 u16 vdisk_info
, u16 sector_size
,
1085 unsigned int physical_sector_size
)
1090 unsigned int offset
;
1095 BUG_ON(info
->gd
!= NULL
);
1096 BUG_ON(info
->rq
!= NULL
);
1098 if ((info
->vdevice
>>EXT_SHIFT
) > 1) {
1099 /* this is above the extended range; something is wrong */
1100 printk(KERN_WARNING
"blkfront: vdevice 0x%x is above the extended range; ignoring\n", info
->vdevice
);
1104 if (!VDEV_IS_EXTENDED(info
->vdevice
)) {
1105 err
= xen_translate_vdev(info
->vdevice
, &minor
, &offset
);
1108 nr_parts
= PARTS_PER_DISK
;
1110 minor
= BLKIF_MINOR_EXT(info
->vdevice
);
1111 nr_parts
= PARTS_PER_EXT_DISK
;
1112 offset
= minor
/ nr_parts
;
1113 if (xen_hvm_domain() && offset
< EMULATED_HD_DISK_NAME_OFFSET
+ 4)
1114 printk(KERN_WARNING
"blkfront: vdevice 0x%x might conflict with "
1115 "emulated IDE disks,\n\t choose an xvd device name"
1116 "from xvde on\n", info
->vdevice
);
1118 if (minor
>> MINORBITS
) {
1119 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1120 info
->vdevice
, minor
);
1124 if ((minor
% nr_parts
) == 0)
1125 nr_minors
= nr_parts
;
1127 err
= xlbd_reserve_minors(minor
, nr_minors
);
1132 gd
= alloc_disk(nr_minors
);
1136 strcpy(gd
->disk_name
, DEV_NAME
);
1137 ptr
= encode_disk_name(gd
->disk_name
+ sizeof(DEV_NAME
) - 1, offset
);
1138 BUG_ON(ptr
>= gd
->disk_name
+ DISK_NAME_LEN
);
1142 snprintf(ptr
, gd
->disk_name
+ DISK_NAME_LEN
- ptr
,
1143 "%d", minor
& (nr_parts
- 1));
1145 gd
->major
= XENVBD_MAJOR
;
1146 gd
->first_minor
= minor
;
1147 gd
->fops
= &xlvbd_block_fops
;
1148 gd
->private_data
= info
;
1149 set_capacity(gd
, capacity
);
1151 if (xlvbd_init_blk_queue(gd
, sector_size
, physical_sector_size
)) {
1158 if (vdisk_info
& VDISK_READONLY
)
1161 if (vdisk_info
& VDISK_REMOVABLE
)
1162 gd
->flags
|= GENHD_FL_REMOVABLE
;
1164 if (vdisk_info
& VDISK_CDROM
)
1165 gd
->flags
|= GENHD_FL_CD
;
1170 xlbd_release_minors(minor
, nr_minors
);
1175 static void xlvbd_release_gendisk(struct blkfront_info
*info
)
1177 unsigned int minor
, nr_minors
, i
;
1179 if (info
->rq
== NULL
)
1182 /* No more blkif_request(). */
1183 blk_mq_stop_hw_queues(info
->rq
);
1185 for (i
= 0; i
< info
->nr_rings
; i
++) {
1186 struct blkfront_ring_info
*rinfo
= &info
->rinfo
[i
];
1188 /* No more gnttab callback work. */
1189 gnttab_cancel_free_callback(&rinfo
->callback
);
1191 /* Flush gnttab callback work. Must be done with no locks held. */
1192 flush_work(&rinfo
->work
);
1195 del_gendisk(info
->gd
);
1197 minor
= info
->gd
->first_minor
;
1198 nr_minors
= info
->gd
->minors
;
1199 xlbd_release_minors(minor
, nr_minors
);
1201 blk_cleanup_queue(info
->rq
);
1202 blk_mq_free_tag_set(&info
->tag_set
);
1209 /* Already hold rinfo->ring_lock. */
1210 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info
*rinfo
)
1212 if (!RING_FULL(&rinfo
->ring
))
1213 blk_mq_start_stopped_hw_queues(rinfo
->dev_info
->rq
, true);
1216 static void kick_pending_request_queues(struct blkfront_ring_info
*rinfo
)
1218 unsigned long flags
;
1220 spin_lock_irqsave(&rinfo
->ring_lock
, flags
);
1221 kick_pending_request_queues_locked(rinfo
);
1222 spin_unlock_irqrestore(&rinfo
->ring_lock
, flags
);
1225 static void blkif_restart_queue(struct work_struct
*work
)
1227 struct blkfront_ring_info
*rinfo
= container_of(work
, struct blkfront_ring_info
, work
);
1229 if (rinfo
->dev_info
->connected
== BLKIF_STATE_CONNECTED
)
1230 kick_pending_request_queues(rinfo
);
1233 static void blkif_free_ring(struct blkfront_ring_info
*rinfo
)
1235 struct grant
*persistent_gnt
, *n
;
1236 struct blkfront_info
*info
= rinfo
->dev_info
;
1240 * Remove indirect pages, this only happens when using indirect
1241 * descriptors but not persistent grants
1243 if (!list_empty(&rinfo
->indirect_pages
)) {
1244 struct page
*indirect_page
, *n
;
1246 BUG_ON(info
->feature_persistent
);
1247 list_for_each_entry_safe(indirect_page
, n
, &rinfo
->indirect_pages
, lru
) {
1248 list_del(&indirect_page
->lru
);
1249 __free_page(indirect_page
);
1253 /* Remove all persistent grants. */
1254 if (!list_empty(&rinfo
->grants
)) {
1255 list_for_each_entry_safe(persistent_gnt
, n
,
1256 &rinfo
->grants
, node
) {
1257 list_del(&persistent_gnt
->node
);
1258 if (persistent_gnt
->gref
!= GRANT_INVALID_REF
) {
1259 gnttab_end_foreign_access(persistent_gnt
->gref
,
1261 rinfo
->persistent_gnts_c
--;
1263 if (info
->feature_persistent
)
1264 __free_page(persistent_gnt
->page
);
1265 kfree(persistent_gnt
);
1268 BUG_ON(rinfo
->persistent_gnts_c
!= 0);
1270 for (i
= 0; i
< BLK_RING_SIZE(info
); i
++) {
1272 * Clear persistent grants present in requests already
1273 * on the shared ring
1275 if (!rinfo
->shadow
[i
].request
)
1278 segs
= rinfo
->shadow
[i
].req
.operation
== BLKIF_OP_INDIRECT
?
1279 rinfo
->shadow
[i
].req
.u
.indirect
.nr_segments
:
1280 rinfo
->shadow
[i
].req
.u
.rw
.nr_segments
;
1281 for (j
= 0; j
< segs
; j
++) {
1282 persistent_gnt
= rinfo
->shadow
[i
].grants_used
[j
];
1283 gnttab_end_foreign_access(persistent_gnt
->gref
, 0, 0UL);
1284 if (info
->feature_persistent
)
1285 __free_page(persistent_gnt
->page
);
1286 kfree(persistent_gnt
);
1289 if (rinfo
->shadow
[i
].req
.operation
!= BLKIF_OP_INDIRECT
)
1291 * If this is not an indirect operation don't try to
1292 * free indirect segments
1296 for (j
= 0; j
< INDIRECT_GREFS(segs
); j
++) {
1297 persistent_gnt
= rinfo
->shadow
[i
].indirect_grants
[j
];
1298 gnttab_end_foreign_access(persistent_gnt
->gref
, 0, 0UL);
1299 __free_page(persistent_gnt
->page
);
1300 kfree(persistent_gnt
);
1304 kfree(rinfo
->shadow
[i
].grants_used
);
1305 rinfo
->shadow
[i
].grants_used
= NULL
;
1306 kfree(rinfo
->shadow
[i
].indirect_grants
);
1307 rinfo
->shadow
[i
].indirect_grants
= NULL
;
1308 kfree(rinfo
->shadow
[i
].sg
);
1309 rinfo
->shadow
[i
].sg
= NULL
;
1312 /* No more gnttab callback work. */
1313 gnttab_cancel_free_callback(&rinfo
->callback
);
1315 /* Flush gnttab callback work. Must be done with no locks held. */
1316 flush_work(&rinfo
->work
);
1318 /* Free resources associated with old device channel. */
1319 for (i
= 0; i
< info
->nr_ring_pages
; i
++) {
1320 if (rinfo
->ring_ref
[i
] != GRANT_INVALID_REF
) {
1321 gnttab_end_foreign_access(rinfo
->ring_ref
[i
], 0, 0);
1322 rinfo
->ring_ref
[i
] = GRANT_INVALID_REF
;
1325 free_pages((unsigned long)rinfo
->ring
.sring
, get_order(info
->nr_ring_pages
* XEN_PAGE_SIZE
));
1326 rinfo
->ring
.sring
= NULL
;
1329 unbind_from_irqhandler(rinfo
->irq
, rinfo
);
1330 rinfo
->evtchn
= rinfo
->irq
= 0;
1333 static void blkif_free(struct blkfront_info
*info
, int suspend
)
1337 /* Prevent new requests being issued until we fix things up. */
1338 info
->connected
= suspend
?
1339 BLKIF_STATE_SUSPENDED
: BLKIF_STATE_DISCONNECTED
;
1340 /* No more blkif_request(). */
1342 blk_mq_stop_hw_queues(info
->rq
);
1344 for (i
= 0; i
< info
->nr_rings
; i
++)
1345 blkif_free_ring(&info
->rinfo
[i
]);
1352 struct copy_from_grant
{
1353 const struct blk_shadow
*s
;
1354 unsigned int grant_idx
;
1355 unsigned int bvec_offset
;
1359 static void blkif_copy_from_grant(unsigned long gfn
, unsigned int offset
,
1360 unsigned int len
, void *data
)
1362 struct copy_from_grant
*info
= data
;
1364 /* Convenient aliases */
1365 const struct blk_shadow
*s
= info
->s
;
1367 shared_data
= kmap_atomic(s
->grants_used
[info
->grant_idx
]->page
);
1369 memcpy(info
->bvec_data
+ info
->bvec_offset
,
1370 shared_data
+ offset
, len
);
1372 info
->bvec_offset
+= len
;
1375 kunmap_atomic(shared_data
);
1378 static enum blk_req_status
blkif_rsp_to_req_status(int rsp
)
1382 case BLKIF_RSP_OKAY
:
1384 case BLKIF_RSP_EOPNOTSUPP
:
1385 return REQ_EOPNOTSUPP
;
1386 case BLKIF_RSP_ERROR
:
1394 * Get the final status of the block request based on two ring response
1396 static int blkif_get_final_status(enum blk_req_status s1
,
1397 enum blk_req_status s2
)
1399 BUG_ON(s1
== REQ_WAITING
);
1400 BUG_ON(s2
== REQ_WAITING
);
1402 if (s1
== REQ_ERROR
|| s2
== REQ_ERROR
)
1403 return BLKIF_RSP_ERROR
;
1404 else if (s1
== REQ_EOPNOTSUPP
|| s2
== REQ_EOPNOTSUPP
)
1405 return BLKIF_RSP_EOPNOTSUPP
;
1406 return BLKIF_RSP_OKAY
;
1409 static bool blkif_completion(unsigned long *id
,
1410 struct blkfront_ring_info
*rinfo
,
1411 struct blkif_response
*bret
)
1414 struct scatterlist
*sg
;
1415 int num_sg
, num_grant
;
1416 struct blkfront_info
*info
= rinfo
->dev_info
;
1417 struct blk_shadow
*s
= &rinfo
->shadow
[*id
];
1418 struct copy_from_grant data
= {
1422 num_grant
= s
->req
.operation
== BLKIF_OP_INDIRECT
?
1423 s
->req
.u
.indirect
.nr_segments
: s
->req
.u
.rw
.nr_segments
;
1425 /* The I/O request may be split in two. */
1426 if (unlikely(s
->associated_id
!= NO_ASSOCIATED_ID
)) {
1427 struct blk_shadow
*s2
= &rinfo
->shadow
[s
->associated_id
];
1429 /* Keep the status of the current response in shadow. */
1430 s
->status
= blkif_rsp_to_req_status(bret
->status
);
1432 /* Wait the second response if not yet here. */
1433 if (s2
->status
== REQ_WAITING
)
1436 bret
->status
= blkif_get_final_status(s
->status
,
1440 * All the grants is stored in the first shadow in order
1441 * to make the completion code simpler.
1443 num_grant
+= s2
->req
.u
.rw
.nr_segments
;
1446 * The two responses may not come in order. Only the
1447 * first request will store the scatter-gather list.
1449 if (s2
->num_sg
!= 0) {
1450 /* Update "id" with the ID of the first response. */
1451 *id
= s
->associated_id
;
1456 * We don't need anymore the second request, so recycling
1459 if (add_id_to_freelist(rinfo
, s
->associated_id
))
1460 WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1461 info
->gd
->disk_name
, s
->associated_id
);
1467 if (bret
->operation
== BLKIF_OP_READ
&& info
->feature_persistent
) {
1468 for_each_sg(s
->sg
, sg
, num_sg
, i
) {
1469 BUG_ON(sg
->offset
+ sg
->length
> PAGE_SIZE
);
1471 data
.bvec_offset
= sg
->offset
;
1472 data
.bvec_data
= kmap_atomic(sg_page(sg
));
1474 gnttab_foreach_grant_in_range(sg_page(sg
),
1477 blkif_copy_from_grant
,
1480 kunmap_atomic(data
.bvec_data
);
1483 /* Add the persistent grant into the list of free grants */
1484 for (i
= 0; i
< num_grant
; i
++) {
1485 if (gnttab_query_foreign_access(s
->grants_used
[i
]->gref
)) {
1487 * If the grant is still mapped by the backend (the
1488 * backend has chosen to make this grant persistent)
1489 * we add it at the head of the list, so it will be
1492 if (!info
->feature_persistent
)
1493 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1494 s
->grants_used
[i
]->gref
);
1495 list_add(&s
->grants_used
[i
]->node
, &rinfo
->grants
);
1496 rinfo
->persistent_gnts_c
++;
1499 * If the grant is not mapped by the backend we end the
1500 * foreign access and add it to the tail of the list,
1501 * so it will not be picked again unless we run out of
1502 * persistent grants.
1504 gnttab_end_foreign_access(s
->grants_used
[i
]->gref
, 0, 0UL);
1505 s
->grants_used
[i
]->gref
= GRANT_INVALID_REF
;
1506 list_add_tail(&s
->grants_used
[i
]->node
, &rinfo
->grants
);
1509 if (s
->req
.operation
== BLKIF_OP_INDIRECT
) {
1510 for (i
= 0; i
< INDIRECT_GREFS(num_grant
); i
++) {
1511 if (gnttab_query_foreign_access(s
->indirect_grants
[i
]->gref
)) {
1512 if (!info
->feature_persistent
)
1513 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1514 s
->indirect_grants
[i
]->gref
);
1515 list_add(&s
->indirect_grants
[i
]->node
, &rinfo
->grants
);
1516 rinfo
->persistent_gnts_c
++;
1518 struct page
*indirect_page
;
1520 gnttab_end_foreign_access(s
->indirect_grants
[i
]->gref
, 0, 0UL);
1522 * Add the used indirect page back to the list of
1523 * available pages for indirect grefs.
1525 if (!info
->feature_persistent
) {
1526 indirect_page
= s
->indirect_grants
[i
]->page
;
1527 list_add(&indirect_page
->lru
, &rinfo
->indirect_pages
);
1529 s
->indirect_grants
[i
]->gref
= GRANT_INVALID_REF
;
1530 list_add_tail(&s
->indirect_grants
[i
]->node
, &rinfo
->grants
);
1538 static irqreturn_t
blkif_interrupt(int irq
, void *dev_id
)
1540 struct request
*req
;
1541 struct blkif_response
*bret
;
1543 unsigned long flags
;
1544 struct blkfront_ring_info
*rinfo
= (struct blkfront_ring_info
*)dev_id
;
1545 struct blkfront_info
*info
= rinfo
->dev_info
;
1548 if (unlikely(info
->connected
!= BLKIF_STATE_CONNECTED
))
1551 spin_lock_irqsave(&rinfo
->ring_lock
, flags
);
1553 rp
= rinfo
->ring
.sring
->rsp_prod
;
1554 rmb(); /* Ensure we see queued responses up to 'rp'. */
1556 for (i
= rinfo
->ring
.rsp_cons
; i
!= rp
; i
++) {
1559 bret
= RING_GET_RESPONSE(&rinfo
->ring
, i
);
1562 * The backend has messed up and given us an id that we would
1563 * never have given to it (we stamp it up to BLK_RING_SIZE -
1564 * look in get_id_from_freelist.
1566 if (id
>= BLK_RING_SIZE(info
)) {
1567 WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1568 info
->gd
->disk_name
, op_name(bret
->operation
), id
);
1569 /* We can't safely get the 'struct request' as
1570 * the id is busted. */
1573 req
= rinfo
->shadow
[id
].request
;
1575 if (bret
->operation
!= BLKIF_OP_DISCARD
) {
1577 * We may need to wait for an extra response if the
1578 * I/O request is split in 2
1580 if (!blkif_completion(&id
, rinfo
, bret
))
1584 if (add_id_to_freelist(rinfo
, id
)) {
1585 WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1586 info
->gd
->disk_name
, op_name(bret
->operation
), id
);
1590 error
= (bret
->status
== BLKIF_RSP_OKAY
) ? 0 : -EIO
;
1591 switch (bret
->operation
) {
1592 case BLKIF_OP_DISCARD
:
1593 if (unlikely(bret
->status
== BLKIF_RSP_EOPNOTSUPP
)) {
1594 struct request_queue
*rq
= info
->rq
;
1595 printk(KERN_WARNING
"blkfront: %s: %s op failed\n",
1596 info
->gd
->disk_name
, op_name(bret
->operation
));
1597 error
= -EOPNOTSUPP
;
1598 info
->feature_discard
= 0;
1599 info
->feature_secdiscard
= 0;
1600 queue_flag_clear(QUEUE_FLAG_DISCARD
, rq
);
1601 queue_flag_clear(QUEUE_FLAG_SECERASE
, rq
);
1603 blk_mq_complete_request(req
, error
);
1605 case BLKIF_OP_FLUSH_DISKCACHE
:
1606 case BLKIF_OP_WRITE_BARRIER
:
1607 if (unlikely(bret
->status
== BLKIF_RSP_EOPNOTSUPP
)) {
1608 printk(KERN_WARNING
"blkfront: %s: %s op failed\n",
1609 info
->gd
->disk_name
, op_name(bret
->operation
));
1610 error
= -EOPNOTSUPP
;
1612 if (unlikely(bret
->status
== BLKIF_RSP_ERROR
&&
1613 rinfo
->shadow
[id
].req
.u
.rw
.nr_segments
== 0)) {
1614 printk(KERN_WARNING
"blkfront: %s: empty %s op failed\n",
1615 info
->gd
->disk_name
, op_name(bret
->operation
));
1616 error
= -EOPNOTSUPP
;
1618 if (unlikely(error
)) {
1619 if (error
== -EOPNOTSUPP
)
1621 info
->feature_fua
= 0;
1622 info
->feature_flush
= 0;
1627 case BLKIF_OP_WRITE
:
1628 if (unlikely(bret
->status
!= BLKIF_RSP_OKAY
))
1629 dev_dbg(&info
->xbdev
->dev
, "Bad return from blkdev data "
1630 "request: %x\n", bret
->status
);
1632 blk_mq_complete_request(req
, error
);
1639 rinfo
->ring
.rsp_cons
= i
;
1641 if (i
!= rinfo
->ring
.req_prod_pvt
) {
1643 RING_FINAL_CHECK_FOR_RESPONSES(&rinfo
->ring
, more_to_do
);
1647 rinfo
->ring
.sring
->rsp_event
= i
+ 1;
1649 kick_pending_request_queues_locked(rinfo
);
1651 spin_unlock_irqrestore(&rinfo
->ring_lock
, flags
);
1657 static int setup_blkring(struct xenbus_device
*dev
,
1658 struct blkfront_ring_info
*rinfo
)
1660 struct blkif_sring
*sring
;
1662 struct blkfront_info
*info
= rinfo
->dev_info
;
1663 unsigned long ring_size
= info
->nr_ring_pages
* XEN_PAGE_SIZE
;
1664 grant_ref_t gref
[XENBUS_MAX_RING_GRANTS
];
1666 for (i
= 0; i
< info
->nr_ring_pages
; i
++)
1667 rinfo
->ring_ref
[i
] = GRANT_INVALID_REF
;
1669 sring
= (struct blkif_sring
*)__get_free_pages(GFP_NOIO
| __GFP_HIGH
,
1670 get_order(ring_size
));
1672 xenbus_dev_fatal(dev
, -ENOMEM
, "allocating shared ring");
1675 SHARED_RING_INIT(sring
);
1676 FRONT_RING_INIT(&rinfo
->ring
, sring
, ring_size
);
1678 err
= xenbus_grant_ring(dev
, rinfo
->ring
.sring
, info
->nr_ring_pages
, gref
);
1680 free_pages((unsigned long)sring
, get_order(ring_size
));
1681 rinfo
->ring
.sring
= NULL
;
1684 for (i
= 0; i
< info
->nr_ring_pages
; i
++)
1685 rinfo
->ring_ref
[i
] = gref
[i
];
1687 err
= xenbus_alloc_evtchn(dev
, &rinfo
->evtchn
);
1691 err
= bind_evtchn_to_irqhandler(rinfo
->evtchn
, blkif_interrupt
, 0,
1694 xenbus_dev_fatal(dev
, err
,
1695 "bind_evtchn_to_irqhandler failed");
1702 blkif_free(info
, 0);
1707 * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1708 * ring buffer may have multi pages depending on ->nr_ring_pages.
1710 static int write_per_ring_nodes(struct xenbus_transaction xbt
,
1711 struct blkfront_ring_info
*rinfo
, const char *dir
)
1715 const char *message
= NULL
;
1716 struct blkfront_info
*info
= rinfo
->dev_info
;
1718 if (info
->nr_ring_pages
== 1) {
1719 err
= xenbus_printf(xbt
, dir
, "ring-ref", "%u", rinfo
->ring_ref
[0]);
1721 message
= "writing ring-ref";
1722 goto abort_transaction
;
1725 for (i
= 0; i
< info
->nr_ring_pages
; i
++) {
1726 char ring_ref_name
[RINGREF_NAME_LEN
];
1728 snprintf(ring_ref_name
, RINGREF_NAME_LEN
, "ring-ref%u", i
);
1729 err
= xenbus_printf(xbt
, dir
, ring_ref_name
,
1730 "%u", rinfo
->ring_ref
[i
]);
1732 message
= "writing ring-ref";
1733 goto abort_transaction
;
1738 err
= xenbus_printf(xbt
, dir
, "event-channel", "%u", rinfo
->evtchn
);
1740 message
= "writing event-channel";
1741 goto abort_transaction
;
1747 xenbus_transaction_end(xbt
, 1);
1749 xenbus_dev_fatal(info
->xbdev
, err
, "%s", message
);
1754 /* Common code used when first setting up, and when resuming. */
1755 static int talk_to_blkback(struct xenbus_device
*dev
,
1756 struct blkfront_info
*info
)
1758 const char *message
= NULL
;
1759 struct xenbus_transaction xbt
;
1761 unsigned int i
, max_page_order
;
1762 unsigned int ring_page_order
;
1764 max_page_order
= xenbus_read_unsigned(info
->xbdev
->otherend
,
1765 "max-ring-page-order", 0);
1766 ring_page_order
= min(xen_blkif_max_ring_order
, max_page_order
);
1767 info
->nr_ring_pages
= 1 << ring_page_order
;
1769 for (i
= 0; i
< info
->nr_rings
; i
++) {
1770 struct blkfront_ring_info
*rinfo
= &info
->rinfo
[i
];
1772 /* Create shared ring, alloc event channel. */
1773 err
= setup_blkring(dev
, rinfo
);
1775 goto destroy_blkring
;
1779 err
= xenbus_transaction_start(&xbt
);
1781 xenbus_dev_fatal(dev
, err
, "starting transaction");
1782 goto destroy_blkring
;
1785 if (info
->nr_ring_pages
> 1) {
1786 err
= xenbus_printf(xbt
, dev
->nodename
, "ring-page-order", "%u",
1789 message
= "writing ring-page-order";
1790 goto abort_transaction
;
1794 /* We already got the number of queues/rings in _probe */
1795 if (info
->nr_rings
== 1) {
1796 err
= write_per_ring_nodes(xbt
, &info
->rinfo
[0], dev
->nodename
);
1798 goto destroy_blkring
;
1803 err
= xenbus_printf(xbt
, dev
->nodename
, "multi-queue-num-queues", "%u",
1806 message
= "writing multi-queue-num-queues";
1807 goto abort_transaction
;
1810 pathsize
= strlen(dev
->nodename
) + QUEUE_NAME_LEN
;
1811 path
= kmalloc(pathsize
, GFP_KERNEL
);
1814 message
= "ENOMEM while writing ring references";
1815 goto abort_transaction
;
1818 for (i
= 0; i
< info
->nr_rings
; i
++) {
1819 memset(path
, 0, pathsize
);
1820 snprintf(path
, pathsize
, "%s/queue-%u", dev
->nodename
, i
);
1821 err
= write_per_ring_nodes(xbt
, &info
->rinfo
[i
], path
);
1824 goto destroy_blkring
;
1829 err
= xenbus_printf(xbt
, dev
->nodename
, "protocol", "%s",
1830 XEN_IO_PROTO_ABI_NATIVE
);
1832 message
= "writing protocol";
1833 goto abort_transaction
;
1835 err
= xenbus_printf(xbt
, dev
->nodename
,
1836 "feature-persistent", "%u", 1);
1839 "writing persistent grants feature to xenbus");
1841 err
= xenbus_transaction_end(xbt
, 0);
1845 xenbus_dev_fatal(dev
, err
, "completing transaction");
1846 goto destroy_blkring
;
1849 for (i
= 0; i
< info
->nr_rings
; i
++) {
1851 struct blkfront_ring_info
*rinfo
= &info
->rinfo
[i
];
1853 for (j
= 0; j
< BLK_RING_SIZE(info
); j
++)
1854 rinfo
->shadow
[j
].req
.u
.rw
.id
= j
+ 1;
1855 rinfo
->shadow
[BLK_RING_SIZE(info
)-1].req
.u
.rw
.id
= 0x0fffffff;
1857 xenbus_switch_state(dev
, XenbusStateInitialised
);
1862 xenbus_transaction_end(xbt
, 1);
1864 xenbus_dev_fatal(dev
, err
, "%s", message
);
1866 blkif_free(info
, 0);
1869 dev_set_drvdata(&dev
->dev
, NULL
);
1874 static int negotiate_mq(struct blkfront_info
*info
)
1876 unsigned int backend_max_queues
;
1879 BUG_ON(info
->nr_rings
);
1881 /* Check if backend supports multiple queues. */
1882 backend_max_queues
= xenbus_read_unsigned(info
->xbdev
->otherend
,
1883 "multi-queue-max-queues", 1);
1884 info
->nr_rings
= min(backend_max_queues
, xen_blkif_max_queues
);
1885 /* We need at least one ring. */
1886 if (!info
->nr_rings
)
1889 info
->rinfo
= kzalloc(sizeof(struct blkfront_ring_info
) * info
->nr_rings
, GFP_KERNEL
);
1891 xenbus_dev_fatal(info
->xbdev
, -ENOMEM
, "allocating ring_info structure");
1895 for (i
= 0; i
< info
->nr_rings
; i
++) {
1896 struct blkfront_ring_info
*rinfo
;
1898 rinfo
= &info
->rinfo
[i
];
1899 INIT_LIST_HEAD(&rinfo
->indirect_pages
);
1900 INIT_LIST_HEAD(&rinfo
->grants
);
1901 rinfo
->dev_info
= info
;
1902 INIT_WORK(&rinfo
->work
, blkif_restart_queue
);
1903 spin_lock_init(&rinfo
->ring_lock
);
1908 * Entry point to this code when a new device is created. Allocate the basic
1909 * structures and the ring buffer for communication with the backend, and
1910 * inform the backend of the appropriate details for those. Switch to
1911 * Initialised state.
1913 static int blkfront_probe(struct xenbus_device
*dev
,
1914 const struct xenbus_device_id
*id
)
1917 struct blkfront_info
*info
;
1919 /* FIXME: Use dynamic device id if this is not set. */
1920 err
= xenbus_scanf(XBT_NIL
, dev
->nodename
,
1921 "virtual-device", "%i", &vdevice
);
1923 /* go looking in the extended area instead */
1924 err
= xenbus_scanf(XBT_NIL
, dev
->nodename
, "virtual-device-ext",
1927 xenbus_dev_fatal(dev
, err
, "reading virtual-device");
1932 if (xen_hvm_domain()) {
1935 /* no unplug has been done: do not hook devices != xen vbds */
1936 if (xen_has_pv_and_legacy_disk_devices()) {
1939 if (!VDEV_IS_EXTENDED(vdevice
))
1940 major
= BLKIF_MAJOR(vdevice
);
1942 major
= XENVBD_MAJOR
;
1944 if (major
!= XENVBD_MAJOR
) {
1946 "%s: HVM does not support vbd %d as xen block device\n",
1951 /* do not create a PV cdrom device if we are an HVM guest */
1952 type
= xenbus_read(XBT_NIL
, dev
->nodename
, "device-type", &len
);
1955 if (strncmp(type
, "cdrom", 5) == 0) {
1961 info
= kzalloc(sizeof(*info
), GFP_KERNEL
);
1963 xenbus_dev_fatal(dev
, -ENOMEM
, "allocating info structure");
1968 err
= negotiate_mq(info
);
1974 mutex_init(&info
->mutex
);
1975 info
->vdevice
= vdevice
;
1976 info
->connected
= BLKIF_STATE_DISCONNECTED
;
1978 /* Front end dir is a number, which is used as the id. */
1979 info
->handle
= simple_strtoul(strrchr(dev
->nodename
, '/')+1, NULL
, 0);
1980 dev_set_drvdata(&dev
->dev
, info
);
1985 static void split_bio_end(struct bio
*bio
)
1987 struct split_bio
*split_bio
= bio
->bi_private
;
1989 if (atomic_dec_and_test(&split_bio
->pending
)) {
1990 split_bio
->bio
->bi_phys_segments
= 0;
1991 split_bio
->bio
->bi_error
= bio
->bi_error
;
1992 bio_endio(split_bio
->bio
);
1998 static int blkif_recover(struct blkfront_info
*info
)
2000 unsigned int i
, r_index
;
2001 struct request
*req
, *n
;
2003 struct bio
*bio
, *cloned_bio
;
2004 unsigned int segs
, offset
;
2006 struct split_bio
*split_bio
;
2008 blkfront_gather_backend_features(info
);
2009 /* Reset limits changed by blk_mq_update_nr_hw_queues(). */
2010 blkif_set_queue_limits(info
);
2011 segs
= info
->max_indirect_segments
? : BLKIF_MAX_SEGMENTS_PER_REQUEST
;
2012 blk_queue_max_segments(info
->rq
, segs
/ GRANTS_PER_PSEG
);
2014 for (r_index
= 0; r_index
< info
->nr_rings
; r_index
++) {
2015 struct blkfront_ring_info
*rinfo
= &info
->rinfo
[r_index
];
2017 rc
= blkfront_setup_indirect(rinfo
);
2021 xenbus_switch_state(info
->xbdev
, XenbusStateConnected
);
2023 /* Now safe for us to use the shared ring */
2024 info
->connected
= BLKIF_STATE_CONNECTED
;
2026 for (r_index
= 0; r_index
< info
->nr_rings
; r_index
++) {
2027 struct blkfront_ring_info
*rinfo
;
2029 rinfo
= &info
->rinfo
[r_index
];
2030 /* Kick any other new requests queued since we resumed */
2031 kick_pending_request_queues(rinfo
);
2034 list_for_each_entry_safe(req
, n
, &info
->requests
, queuelist
) {
2035 /* Requeue pending requests (flush or discard) */
2036 list_del_init(&req
->queuelist
);
2037 BUG_ON(req
->nr_phys_segments
> segs
);
2038 blk_mq_requeue_request(req
, false);
2040 blk_mq_start_stopped_hw_queues(info
->rq
, true);
2041 blk_mq_kick_requeue_list(info
->rq
);
2043 while ((bio
= bio_list_pop(&info
->bio_list
)) != NULL
) {
2044 /* Traverse the list of pending bios and re-queue them */
2045 if (bio_segments(bio
) > segs
) {
2047 * This bio has more segments than what we can
2048 * handle, we have to split it.
2050 pending
= (bio_segments(bio
) + segs
- 1) / segs
;
2051 split_bio
= kzalloc(sizeof(*split_bio
), GFP_NOIO
);
2052 BUG_ON(split_bio
== NULL
);
2053 atomic_set(&split_bio
->pending
, pending
);
2054 split_bio
->bio
= bio
;
2055 for (i
= 0; i
< pending
; i
++) {
2056 offset
= (i
* segs
* XEN_PAGE_SIZE
) >> 9;
2057 size
= min((unsigned int)(segs
* XEN_PAGE_SIZE
) >> 9,
2058 (unsigned int)bio_sectors(bio
) - offset
);
2059 cloned_bio
= bio_clone(bio
, GFP_NOIO
);
2060 BUG_ON(cloned_bio
== NULL
);
2061 bio_trim(cloned_bio
, offset
, size
);
2062 cloned_bio
->bi_private
= split_bio
;
2063 cloned_bio
->bi_end_io
= split_bio_end
;
2064 submit_bio(cloned_bio
);
2067 * Now we have to wait for all those smaller bios to
2068 * end, so we can also end the "parent" bio.
2072 /* We don't need to split this bio */
2080 * We are reconnecting to the backend, due to a suspend/resume, or a backend
2081 * driver restart. We tear down our blkif structure and recreate it, but
2082 * leave the device-layer structures intact so that this is transparent to the
2083 * rest of the kernel.
2085 static int blkfront_resume(struct xenbus_device
*dev
)
2087 struct blkfront_info
*info
= dev_get_drvdata(&dev
->dev
);
2091 dev_dbg(&dev
->dev
, "blkfront_resume: %s\n", dev
->nodename
);
2093 bio_list_init(&info
->bio_list
);
2094 INIT_LIST_HEAD(&info
->requests
);
2095 for (i
= 0; i
< info
->nr_rings
; i
++) {
2096 struct blkfront_ring_info
*rinfo
= &info
->rinfo
[i
];
2097 struct bio_list merge_bio
;
2098 struct blk_shadow
*shadow
= rinfo
->shadow
;
2100 for (j
= 0; j
< BLK_RING_SIZE(info
); j
++) {
2102 if (!shadow
[j
].request
)
2106 * Get the bios in the request so we can re-queue them.
2108 if (req_op(shadow
[i
].request
) == REQ_OP_FLUSH
||
2109 req_op(shadow
[i
].request
) == REQ_OP_DISCARD
||
2110 req_op(shadow
[i
].request
) == REQ_OP_SECURE_ERASE
||
2111 shadow
[j
].request
->cmd_flags
& REQ_FUA
) {
2113 * Flush operations don't contain bios, so
2114 * we need to requeue the whole request
2116 * XXX: but this doesn't make any sense for a
2117 * write with the FUA flag set..
2119 list_add(&shadow
[j
].request
->queuelist
, &info
->requests
);
2122 merge_bio
.head
= shadow
[j
].request
->bio
;
2123 merge_bio
.tail
= shadow
[j
].request
->biotail
;
2124 bio_list_merge(&info
->bio_list
, &merge_bio
);
2125 shadow
[j
].request
->bio
= NULL
;
2126 blk_mq_end_request(shadow
[j
].request
, 0);
2130 blkif_free(info
, info
->connected
== BLKIF_STATE_CONNECTED
);
2132 err
= negotiate_mq(info
);
2136 err
= talk_to_blkback(dev
, info
);
2138 blk_mq_update_nr_hw_queues(&info
->tag_set
, info
->nr_rings
);
2141 * We have to wait for the backend to switch to
2142 * connected state, since we want to read which
2143 * features it supports.
2149 static void blkfront_closing(struct blkfront_info
*info
)
2151 struct xenbus_device
*xbdev
= info
->xbdev
;
2152 struct block_device
*bdev
= NULL
;
2154 mutex_lock(&info
->mutex
);
2156 if (xbdev
->state
== XenbusStateClosing
) {
2157 mutex_unlock(&info
->mutex
);
2162 bdev
= bdget_disk(info
->gd
, 0);
2164 mutex_unlock(&info
->mutex
);
2167 xenbus_frontend_closed(xbdev
);
2171 mutex_lock(&bdev
->bd_mutex
);
2173 if (bdev
->bd_openers
) {
2174 xenbus_dev_error(xbdev
, -EBUSY
,
2175 "Device in use; refusing to close");
2176 xenbus_switch_state(xbdev
, XenbusStateClosing
);
2178 xlvbd_release_gendisk(info
);
2179 xenbus_frontend_closed(xbdev
);
2182 mutex_unlock(&bdev
->bd_mutex
);
2186 static void blkfront_setup_discard(struct blkfront_info
*info
)
2189 unsigned int discard_granularity
;
2190 unsigned int discard_alignment
;
2192 info
->feature_discard
= 1;
2193 err
= xenbus_gather(XBT_NIL
, info
->xbdev
->otherend
,
2194 "discard-granularity", "%u", &discard_granularity
,
2195 "discard-alignment", "%u", &discard_alignment
,
2198 info
->discard_granularity
= discard_granularity
;
2199 info
->discard_alignment
= discard_alignment
;
2201 info
->feature_secdiscard
=
2202 !!xenbus_read_unsigned(info
->xbdev
->otherend
, "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
)
2295 unsigned int indirect_segments
;
2297 info
->feature_flush
= 0;
2298 info
->feature_fua
= 0;
2301 * If there's no "feature-barrier" defined, then it means
2302 * we're dealing with a very old backend which writes
2303 * synchronously; nothing to do.
2305 * If there are barriers, then we use flush.
2307 if (xenbus_read_unsigned(info
->xbdev
->otherend
, "feature-barrier", 0)) {
2308 info
->feature_flush
= 1;
2309 info
->feature_fua
= 1;
2313 * And if there is "feature-flush-cache" use that above
2316 if (xenbus_read_unsigned(info
->xbdev
->otherend
, "feature-flush-cache",
2318 info
->feature_flush
= 1;
2319 info
->feature_fua
= 0;
2322 if (xenbus_read_unsigned(info
->xbdev
->otherend
, "feature-discard", 0))
2323 blkfront_setup_discard(info
);
2325 info
->feature_persistent
=
2326 xenbus_read_unsigned(info
->xbdev
->otherend
,
2327 "feature-persistent", 0);
2329 indirect_segments
= xenbus_read_unsigned(info
->xbdev
->otherend
,
2330 "feature-max-indirect-segments", 0);
2331 info
->max_indirect_segments
= min(indirect_segments
,
2332 xen_blkif_max_segments
);
2336 * Invoked when the backend is finally 'ready' (and has told produced
2337 * the details about the physical device - #sectors, size, etc).
2339 static void blkfront_connect(struct blkfront_info
*info
)
2341 unsigned long long sectors
;
2342 unsigned long sector_size
;
2343 unsigned int physical_sector_size
;
2347 switch (info
->connected
) {
2348 case BLKIF_STATE_CONNECTED
:
2350 * Potentially, the back-end may be signalling
2351 * a capacity change; update the capacity.
2353 err
= xenbus_scanf(XBT_NIL
, info
->xbdev
->otherend
,
2354 "sectors", "%Lu", §ors
);
2355 if (XENBUS_EXIST_ERR(err
))
2357 printk(KERN_INFO
"Setting capacity to %Lu\n",
2359 set_capacity(info
->gd
, sectors
);
2360 revalidate_disk(info
->gd
);
2363 case BLKIF_STATE_SUSPENDED
:
2365 * If we are recovering from suspension, we need to wait
2366 * for the backend to announce it's features before
2367 * reconnecting, at least we need to know if the backend
2368 * supports indirect descriptors, and how many.
2370 blkif_recover(info
);
2377 dev_dbg(&info
->xbdev
->dev
, "%s:%s.\n",
2378 __func__
, info
->xbdev
->otherend
);
2380 err
= xenbus_gather(XBT_NIL
, info
->xbdev
->otherend
,
2381 "sectors", "%llu", §ors
,
2382 "info", "%u", &binfo
,
2383 "sector-size", "%lu", §or_size
,
2386 xenbus_dev_fatal(info
->xbdev
, err
,
2387 "reading backend fields at %s",
2388 info
->xbdev
->otherend
);
2393 * physcial-sector-size is a newer field, so old backends may not
2394 * provide this. Assume physical sector size to be the same as
2395 * sector_size in that case.
2397 physical_sector_size
= xenbus_read_unsigned(info
->xbdev
->otherend
,
2398 "physical-sector-size",
2400 blkfront_gather_backend_features(info
);
2401 for (i
= 0; i
< info
->nr_rings
; i
++) {
2402 err
= blkfront_setup_indirect(&info
->rinfo
[i
]);
2404 xenbus_dev_fatal(info
->xbdev
, err
, "setup_indirect at %s",
2405 info
->xbdev
->otherend
);
2406 blkif_free(info
, 0);
2411 err
= xlvbd_alloc_gendisk(sectors
, info
, binfo
, sector_size
,
2412 physical_sector_size
);
2414 xenbus_dev_fatal(info
->xbdev
, err
, "xlvbd_add at %s",
2415 info
->xbdev
->otherend
);
2419 xenbus_switch_state(info
->xbdev
, XenbusStateConnected
);
2421 /* Kick pending requests. */
2422 info
->connected
= BLKIF_STATE_CONNECTED
;
2423 for (i
= 0; i
< info
->nr_rings
; i
++)
2424 kick_pending_request_queues(&info
->rinfo
[i
]);
2426 device_add_disk(&info
->xbdev
->dev
, info
->gd
);
2432 blkif_free(info
, 0);
2437 * Callback received when the backend's state changes.
2439 static void blkback_changed(struct xenbus_device
*dev
,
2440 enum xenbus_state backend_state
)
2442 struct blkfront_info
*info
= dev_get_drvdata(&dev
->dev
);
2444 dev_dbg(&dev
->dev
, "blkfront:blkback_changed to state %d.\n", backend_state
);
2446 switch (backend_state
) {
2447 case XenbusStateInitWait
:
2448 if (dev
->state
!= XenbusStateInitialising
)
2450 if (talk_to_blkback(dev
, info
))
2452 case XenbusStateInitialising
:
2453 case XenbusStateInitialised
:
2454 case XenbusStateReconfiguring
:
2455 case XenbusStateReconfigured
:
2456 case XenbusStateUnknown
:
2459 case XenbusStateConnected
:
2461 * talk_to_blkback sets state to XenbusStateInitialised
2462 * and blkfront_connect sets it to XenbusStateConnected
2463 * (if connection went OK).
2465 * If the backend (or toolstack) decides to poke at backend
2466 * state (and re-trigger the watch by setting the state repeatedly
2467 * to XenbusStateConnected (4)) we need to deal with this.
2468 * This is allowed as this is used to communicate to the guest
2469 * that the size of disk has changed!
2471 if ((dev
->state
!= XenbusStateInitialised
) &&
2472 (dev
->state
!= XenbusStateConnected
)) {
2473 if (talk_to_blkback(dev
, info
))
2477 blkfront_connect(info
);
2480 case XenbusStateClosed
:
2481 if (dev
->state
== XenbusStateClosed
)
2483 /* Missed the backend's Closing state -- fallthrough */
2484 case XenbusStateClosing
:
2486 blkfront_closing(info
);
2491 static int blkfront_remove(struct xenbus_device
*xbdev
)
2493 struct blkfront_info
*info
= dev_get_drvdata(&xbdev
->dev
);
2494 struct block_device
*bdev
= NULL
;
2495 struct gendisk
*disk
;
2497 dev_dbg(&xbdev
->dev
, "%s removed", xbdev
->nodename
);
2499 blkif_free(info
, 0);
2501 mutex_lock(&info
->mutex
);
2505 bdev
= bdget_disk(disk
, 0);
2508 mutex_unlock(&info
->mutex
);
2516 * The xbdev was removed before we reached the Closed
2517 * state. See if it's safe to remove the disk. If the bdev
2518 * isn't closed yet, we let release take care of it.
2521 mutex_lock(&bdev
->bd_mutex
);
2522 info
= disk
->private_data
;
2524 dev_warn(disk_to_dev(disk
),
2525 "%s was hot-unplugged, %d stale handles\n",
2526 xbdev
->nodename
, bdev
->bd_openers
);
2528 if (info
&& !bdev
->bd_openers
) {
2529 xlvbd_release_gendisk(info
);
2530 disk
->private_data
= NULL
;
2534 mutex_unlock(&bdev
->bd_mutex
);
2540 static int blkfront_is_ready(struct xenbus_device
*dev
)
2542 struct blkfront_info
*info
= dev_get_drvdata(&dev
->dev
);
2544 return info
->is_ready
&& info
->xbdev
;
2547 static int blkif_open(struct block_device
*bdev
, fmode_t mode
)
2549 struct gendisk
*disk
= bdev
->bd_disk
;
2550 struct blkfront_info
*info
;
2553 mutex_lock(&blkfront_mutex
);
2555 info
= disk
->private_data
;
2562 mutex_lock(&info
->mutex
);
2565 /* xbdev is closed */
2568 mutex_unlock(&info
->mutex
);
2571 mutex_unlock(&blkfront_mutex
);
2575 static void blkif_release(struct gendisk
*disk
, fmode_t mode
)
2577 struct blkfront_info
*info
= disk
->private_data
;
2578 struct block_device
*bdev
;
2579 struct xenbus_device
*xbdev
;
2581 mutex_lock(&blkfront_mutex
);
2583 bdev
= bdget_disk(disk
, 0);
2586 WARN(1, "Block device %s yanked out from us!\n", disk
->disk_name
);
2589 if (bdev
->bd_openers
)
2593 * Check if we have been instructed to close. We will have
2594 * deferred this request, because the bdev was still open.
2597 mutex_lock(&info
->mutex
);
2598 xbdev
= info
->xbdev
;
2600 if (xbdev
&& xbdev
->state
== XenbusStateClosing
) {
2601 /* pending switch to state closed */
2602 dev_info(disk_to_dev(bdev
->bd_disk
), "releasing disk\n");
2603 xlvbd_release_gendisk(info
);
2604 xenbus_frontend_closed(info
->xbdev
);
2607 mutex_unlock(&info
->mutex
);
2610 /* sudden device removal */
2611 dev_info(disk_to_dev(bdev
->bd_disk
), "releasing disk\n");
2612 xlvbd_release_gendisk(info
);
2613 disk
->private_data
= NULL
;
2620 mutex_unlock(&blkfront_mutex
);
2623 static const struct block_device_operations xlvbd_block_fops
=
2625 .owner
= THIS_MODULE
,
2627 .release
= blkif_release
,
2628 .getgeo
= blkif_getgeo
,
2629 .ioctl
= blkif_ioctl
,
2633 static const struct xenbus_device_id blkfront_ids
[] = {
2638 static struct xenbus_driver blkfront_driver
= {
2639 .ids
= blkfront_ids
,
2640 .probe
= blkfront_probe
,
2641 .remove
= blkfront_remove
,
2642 .resume
= blkfront_resume
,
2643 .otherend_changed
= blkback_changed
,
2644 .is_ready
= blkfront_is_ready
,
2647 static int __init
xlblk_init(void)
2650 int nr_cpus
= num_online_cpus();
2655 if (xen_blkif_max_ring_order
> XENBUS_MAX_RING_GRANT_ORDER
) {
2656 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2657 xen_blkif_max_ring_order
, XENBUS_MAX_RING_GRANT_ORDER
);
2658 xen_blkif_max_ring_order
= XENBUS_MAX_RING_GRANT_ORDER
;
2661 if (xen_blkif_max_queues
> nr_cpus
) {
2662 pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2663 xen_blkif_max_queues
, nr_cpus
);
2664 xen_blkif_max_queues
= nr_cpus
;
2667 if (!xen_has_pv_disk_devices())
2670 if (register_blkdev(XENVBD_MAJOR
, DEV_NAME
)) {
2671 printk(KERN_WARNING
"xen_blk: can't get major %d with name %s\n",
2672 XENVBD_MAJOR
, DEV_NAME
);
2676 ret
= xenbus_register_frontend(&blkfront_driver
);
2678 unregister_blkdev(XENVBD_MAJOR
, DEV_NAME
);
2684 module_init(xlblk_init
);
2687 static void __exit
xlblk_exit(void)
2689 xenbus_unregister_driver(&blkfront_driver
);
2690 unregister_blkdev(XENVBD_MAJOR
, DEV_NAME
);
2693 module_exit(xlblk_exit
);
2695 MODULE_DESCRIPTION("Xen virtual block device frontend");
2696 MODULE_LICENSE("GPL");
2697 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR
);
2698 MODULE_ALIAS("xen:vbd");
2699 MODULE_ALIAS("xenblk");