4 * XenLinux virtual block device driver.
6 * Copyright (c) 2003-2004, Keir Fraser & Steve Hand
7 * Modifications by Mark A. Williamson are (c) Intel Research Cambridge
8 * Copyright (c) 2004, Christian Limpach
9 * Copyright (c) 2004, Andrew Warfield
10 * Copyright (c) 2005, Christopher Clark
11 * Copyright (c) 2005, XenSource Ltd
13 * This program is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU General Public License version 2
15 * as published by the Free Software Foundation; or, when distributed
16 * separately from the Linux kernel or incorporated into other
17 * software packages, subject to the following license:
19 * Permission is hereby granted, free of charge, to any person obtaining a copy
20 * of this source file (the "Software"), to deal in the Software without
21 * restriction, including without limitation the rights to use, copy, modify,
22 * merge, publish, distribute, sublicense, and/or sell copies of the Software,
23 * and to permit persons to whom the Software is furnished to do so, subject to
24 * the following conditions:
26 * The above copyright notice and this permission notice shall be included in
27 * all copies or substantial portions of the Software.
29 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
30 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
31 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
32 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
33 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
34 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
38 #include <linux/interrupt.h>
39 #include <linux/blkdev.h>
40 #include <linux/hdreg.h>
41 #include <linux/cdrom.h>
42 #include <linux/module.h>
43 #include <linux/slab.h>
44 #include <linux/mutex.h>
45 #include <linux/scatterlist.h>
46 #include <linux/bitmap.h>
47 #include <linux/list.h>
50 #include <xen/xenbus.h>
51 #include <xen/grant_table.h>
52 #include <xen/events.h>
54 #include <xen/platform_pci.h>
56 #include <xen/interface/grant_table.h>
57 #include <xen/interface/io/blkif.h>
58 #include <xen/interface/io/protocols.h>
60 #include <asm/xen/hypervisor.h>
63 BLKIF_STATE_DISCONNECTED
,
64 BLKIF_STATE_CONNECTED
,
65 BLKIF_STATE_SUSPENDED
,
71 struct list_head node
;
75 struct blkif_request req
;
76 struct request
*request
;
77 struct grant
**grants_used
;
78 struct grant
**indirect_grants
;
79 struct scatterlist
*sg
;
88 static DEFINE_MUTEX(blkfront_mutex
);
89 static const struct block_device_operations xlvbd_block_fops
;
92 * Maximum number of segments in indirect requests, the actual value used by
93 * the frontend driver is the minimum of this value and the value provided
94 * by the backend driver.
97 static unsigned int xen_blkif_max_segments
= 32;
98 module_param_named(max
, xen_blkif_max_segments
, int, S_IRUGO
);
99 MODULE_PARM_DESC(max
, "Maximum amount of segments in indirect requests (default is 32)");
102 * Maximum order of pages to be used for the shared ring between front and
103 * backend, 4KB page granularity is used.
105 static unsigned int xen_blkif_max_ring_order
;
106 module_param_named(max_ring_page_order
, xen_blkif_max_ring_order
, int, S_IRUGO
);
107 MODULE_PARM_DESC(max_ring_page_order
, "Maximum order of pages to be used for the shared ring");
109 #define BLK_RING_SIZE(info) __CONST_RING_SIZE(blkif, PAGE_SIZE * (info)->nr_ring_pages)
110 #define BLK_MAX_RING_SIZE __CONST_RING_SIZE(blkif, PAGE_SIZE * XENBUS_MAX_RING_PAGES)
112 * ring-ref%i i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
113 * characters are enough. Define to 20 to keep consist with backend.
115 #define RINGREF_NAME_LEN (20)
118 * We have one of these per vbd, whether ide, scsi or 'other'. They
119 * hang in private_data off the gendisk structure. We may end up
120 * putting all kinds of interesting stuff here :-)
126 struct xenbus_device
*xbdev
;
130 enum blkif_state connected
;
131 int ring_ref
[XENBUS_MAX_RING_PAGES
];
132 unsigned int nr_ring_pages
;
133 struct blkif_front_ring ring
;
134 unsigned int evtchn
, irq
;
135 struct request_queue
*rq
;
136 struct work_struct work
;
137 struct gnttab_free_callback callback
;
138 struct blk_shadow shadow
[BLK_MAX_RING_SIZE
];
139 struct list_head grants
;
140 struct list_head indirect_pages
;
141 unsigned int persistent_gnts_c
;
142 unsigned long shadow_free
;
143 unsigned int feature_flush
;
144 unsigned int feature_discard
:1;
145 unsigned int feature_secdiscard
:1;
146 unsigned int discard_granularity
;
147 unsigned int discard_alignment
;
148 unsigned int feature_persistent
:1;
149 unsigned int max_indirect_segments
;
153 static unsigned int nr_minors
;
154 static unsigned long *minors
;
155 static DEFINE_SPINLOCK(minor_lock
);
157 #define GRANT_INVALID_REF 0
159 #define PARTS_PER_DISK 16
160 #define PARTS_PER_EXT_DISK 256
162 #define BLKIF_MAJOR(dev) ((dev)>>8)
163 #define BLKIF_MINOR(dev) ((dev) & 0xff)
166 #define EXTENDED (1<<EXT_SHIFT)
167 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
168 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
169 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
170 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
171 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
172 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
174 #define DEV_NAME "xvd" /* name in /dev */
176 #define SEGS_PER_INDIRECT_FRAME \
177 (PAGE_SIZE/sizeof(struct blkif_request_segment))
178 #define INDIRECT_GREFS(_segs) \
179 ((_segs + SEGS_PER_INDIRECT_FRAME - 1)/SEGS_PER_INDIRECT_FRAME)
181 static int blkfront_setup_indirect(struct blkfront_info
*info
);
183 static int get_id_from_freelist(struct blkfront_info
*info
)
185 unsigned long free
= info
->shadow_free
;
186 BUG_ON(free
>= BLK_RING_SIZE(info
));
187 info
->shadow_free
= info
->shadow
[free
].req
.u
.rw
.id
;
188 info
->shadow
[free
].req
.u
.rw
.id
= 0x0fffffee; /* debug */
192 static int add_id_to_freelist(struct blkfront_info
*info
,
195 if (info
->shadow
[id
].req
.u
.rw
.id
!= id
)
197 if (info
->shadow
[id
].request
== NULL
)
199 info
->shadow
[id
].req
.u
.rw
.id
= info
->shadow_free
;
200 info
->shadow
[id
].request
= NULL
;
201 info
->shadow_free
= id
;
205 static int fill_grant_buffer(struct blkfront_info
*info
, int num
)
207 struct page
*granted_page
;
208 struct grant
*gnt_list_entry
, *n
;
212 gnt_list_entry
= kzalloc(sizeof(struct grant
), GFP_NOIO
);
216 if (info
->feature_persistent
) {
217 granted_page
= alloc_page(GFP_NOIO
);
219 kfree(gnt_list_entry
);
222 gnt_list_entry
->pfn
= page_to_pfn(granted_page
);
225 gnt_list_entry
->gref
= GRANT_INVALID_REF
;
226 list_add(&gnt_list_entry
->node
, &info
->grants
);
233 list_for_each_entry_safe(gnt_list_entry
, n
,
234 &info
->grants
, node
) {
235 list_del(&gnt_list_entry
->node
);
236 if (info
->feature_persistent
)
237 __free_page(pfn_to_page(gnt_list_entry
->pfn
));
238 kfree(gnt_list_entry
);
245 static struct grant
*get_grant(grant_ref_t
*gref_head
,
247 struct blkfront_info
*info
)
249 struct grant
*gnt_list_entry
;
250 unsigned long buffer_mfn
;
252 BUG_ON(list_empty(&info
->grants
));
253 gnt_list_entry
= list_first_entry(&info
->grants
, struct grant
,
255 list_del(&gnt_list_entry
->node
);
257 if (gnt_list_entry
->gref
!= GRANT_INVALID_REF
) {
258 info
->persistent_gnts_c
--;
259 return gnt_list_entry
;
262 /* Assign a gref to this page */
263 gnt_list_entry
->gref
= gnttab_claim_grant_reference(gref_head
);
264 BUG_ON(gnt_list_entry
->gref
== -ENOSPC
);
265 if (!info
->feature_persistent
) {
267 gnt_list_entry
->pfn
= pfn
;
269 buffer_mfn
= pfn_to_mfn(gnt_list_entry
->pfn
);
270 gnttab_grant_foreign_access_ref(gnt_list_entry
->gref
,
271 info
->xbdev
->otherend_id
,
273 return gnt_list_entry
;
276 static const char *op_name(int op
)
278 static const char *const names
[] = {
279 [BLKIF_OP_READ
] = "read",
280 [BLKIF_OP_WRITE
] = "write",
281 [BLKIF_OP_WRITE_BARRIER
] = "barrier",
282 [BLKIF_OP_FLUSH_DISKCACHE
] = "flush",
283 [BLKIF_OP_DISCARD
] = "discard" };
285 if (op
< 0 || op
>= ARRAY_SIZE(names
))
293 static int xlbd_reserve_minors(unsigned int minor
, unsigned int nr
)
295 unsigned int end
= minor
+ nr
;
298 if (end
> nr_minors
) {
299 unsigned long *bitmap
, *old
;
301 bitmap
= kcalloc(BITS_TO_LONGS(end
), sizeof(*bitmap
),
306 spin_lock(&minor_lock
);
307 if (end
> nr_minors
) {
309 memcpy(bitmap
, minors
,
310 BITS_TO_LONGS(nr_minors
) * sizeof(*bitmap
));
312 nr_minors
= BITS_TO_LONGS(end
) * BITS_PER_LONG
;
315 spin_unlock(&minor_lock
);
319 spin_lock(&minor_lock
);
320 if (find_next_bit(minors
, end
, minor
) >= end
) {
321 bitmap_set(minors
, minor
, nr
);
325 spin_unlock(&minor_lock
);
330 static void xlbd_release_minors(unsigned int minor
, unsigned int nr
)
332 unsigned int end
= minor
+ nr
;
334 BUG_ON(end
> nr_minors
);
335 spin_lock(&minor_lock
);
336 bitmap_clear(minors
, minor
, nr
);
337 spin_unlock(&minor_lock
);
340 static void blkif_restart_queue_callback(void *arg
)
342 struct blkfront_info
*info
= (struct blkfront_info
*)arg
;
343 schedule_work(&info
->work
);
346 static int blkif_getgeo(struct block_device
*bd
, struct hd_geometry
*hg
)
348 /* We don't have real geometry info, but let's at least return
349 values consistent with the size of the device */
350 sector_t nsect
= get_capacity(bd
->bd_disk
);
351 sector_t cylinders
= nsect
;
355 sector_div(cylinders
, hg
->heads
* hg
->sectors
);
356 hg
->cylinders
= cylinders
;
357 if ((sector_t
)(hg
->cylinders
+ 1) * hg
->heads
* hg
->sectors
< nsect
)
358 hg
->cylinders
= 0xffff;
362 static int blkif_ioctl(struct block_device
*bdev
, fmode_t mode
,
363 unsigned command
, unsigned long argument
)
365 struct blkfront_info
*info
= bdev
->bd_disk
->private_data
;
368 dev_dbg(&info
->xbdev
->dev
, "command: 0x%x, argument: 0x%lx\n",
369 command
, (long)argument
);
372 case CDROMMULTISESSION
:
373 dev_dbg(&info
->xbdev
->dev
, "FIXME: support multisession CDs later\n");
374 for (i
= 0; i
< sizeof(struct cdrom_multisession
); i
++)
375 if (put_user(0, (char __user
*)(argument
+ i
)))
379 case CDROM_GET_CAPABILITY
: {
380 struct gendisk
*gd
= info
->gd
;
381 if (gd
->flags
& GENHD_FL_CD
)
387 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
389 return -EINVAL
; /* same return as native Linux */
396 * Generate a Xen blkfront IO request from a blk layer request. Reads
397 * and writes are handled as expected.
399 * @req: a request struct
401 static int blkif_queue_request(struct request
*req
)
403 struct blkfront_info
*info
= req
->rq_disk
->private_data
;
404 struct blkif_request
*ring_req
;
406 unsigned int fsect
, lsect
;
408 struct blkif_request_segment
*segments
= NULL
;
411 * Used to store if we are able to queue the request by just using
412 * existing persistent grants, or if we have to get new grants,
413 * as there are not sufficiently many free.
415 bool new_persistent_gnts
;
416 grant_ref_t gref_head
;
417 struct grant
*gnt_list_entry
= NULL
;
418 struct scatterlist
*sg
;
421 if (unlikely(info
->connected
!= BLKIF_STATE_CONNECTED
))
424 max_grefs
= req
->nr_phys_segments
;
425 if (max_grefs
> BLKIF_MAX_SEGMENTS_PER_REQUEST
)
427 * If we are using indirect segments we need to account
428 * for the indirect grefs used in the request.
430 max_grefs
+= INDIRECT_GREFS(req
->nr_phys_segments
);
432 /* Check if we have enough grants to allocate a requests */
433 if (info
->persistent_gnts_c
< max_grefs
) {
434 new_persistent_gnts
= 1;
435 if (gnttab_alloc_grant_references(
436 max_grefs
- info
->persistent_gnts_c
,
438 gnttab_request_free_callback(
440 blkif_restart_queue_callback
,
446 new_persistent_gnts
= 0;
448 /* Fill out a communications ring structure. */
449 ring_req
= RING_GET_REQUEST(&info
->ring
, info
->ring
.req_prod_pvt
);
450 id
= get_id_from_freelist(info
);
451 info
->shadow
[id
].request
= req
;
453 if (unlikely(req
->cmd_flags
& (REQ_DISCARD
| REQ_SECURE
))) {
454 ring_req
->operation
= BLKIF_OP_DISCARD
;
455 ring_req
->u
.discard
.nr_sectors
= blk_rq_sectors(req
);
456 ring_req
->u
.discard
.id
= id
;
457 ring_req
->u
.discard
.sector_number
= (blkif_sector_t
)blk_rq_pos(req
);
458 if ((req
->cmd_flags
& REQ_SECURE
) && info
->feature_secdiscard
)
459 ring_req
->u
.discard
.flag
= BLKIF_DISCARD_SECURE
;
461 ring_req
->u
.discard
.flag
= 0;
463 BUG_ON(info
->max_indirect_segments
== 0 &&
464 req
->nr_phys_segments
> BLKIF_MAX_SEGMENTS_PER_REQUEST
);
465 BUG_ON(info
->max_indirect_segments
&&
466 req
->nr_phys_segments
> info
->max_indirect_segments
);
467 nseg
= blk_rq_map_sg(req
->q
, req
, info
->shadow
[id
].sg
);
468 ring_req
->u
.rw
.id
= id
;
469 if (nseg
> BLKIF_MAX_SEGMENTS_PER_REQUEST
) {
471 * The indirect operation can only be a BLKIF_OP_READ or
474 BUG_ON(req
->cmd_flags
& (REQ_FLUSH
| REQ_FUA
));
475 ring_req
->operation
= BLKIF_OP_INDIRECT
;
476 ring_req
->u
.indirect
.indirect_op
= rq_data_dir(req
) ?
477 BLKIF_OP_WRITE
: BLKIF_OP_READ
;
478 ring_req
->u
.indirect
.sector_number
= (blkif_sector_t
)blk_rq_pos(req
);
479 ring_req
->u
.indirect
.handle
= info
->handle
;
480 ring_req
->u
.indirect
.nr_segments
= nseg
;
482 ring_req
->u
.rw
.sector_number
= (blkif_sector_t
)blk_rq_pos(req
);
483 ring_req
->u
.rw
.handle
= info
->handle
;
484 ring_req
->operation
= rq_data_dir(req
) ?
485 BLKIF_OP_WRITE
: BLKIF_OP_READ
;
486 if (req
->cmd_flags
& (REQ_FLUSH
| REQ_FUA
)) {
488 * Ideally we can do an unordered flush-to-disk. In case the
489 * backend onlysupports barriers, use that. A barrier request
490 * a superset of FUA, so we can implement it the same
491 * way. (It's also a FLUSH+FUA, since it is
492 * guaranteed ordered WRT previous writes.)
494 switch (info
->feature_flush
&
495 ((REQ_FLUSH
|REQ_FUA
))) {
496 case REQ_FLUSH
|REQ_FUA
:
497 ring_req
->operation
=
498 BLKIF_OP_WRITE_BARRIER
;
501 ring_req
->operation
=
502 BLKIF_OP_FLUSH_DISKCACHE
;
505 ring_req
->operation
= 0;
508 ring_req
->u
.rw
.nr_segments
= nseg
;
510 for_each_sg(info
->shadow
[id
].sg
, sg
, nseg
, i
) {
511 fsect
= sg
->offset
>> 9;
512 lsect
= fsect
+ (sg
->length
>> 9) - 1;
514 if ((ring_req
->operation
== BLKIF_OP_INDIRECT
) &&
515 (i
% SEGS_PER_INDIRECT_FRAME
== 0)) {
516 unsigned long uninitialized_var(pfn
);
519 kunmap_atomic(segments
);
521 n
= i
/ SEGS_PER_INDIRECT_FRAME
;
522 if (!info
->feature_persistent
) {
523 struct page
*indirect_page
;
525 /* Fetch a pre-allocated page to use for indirect grefs */
526 BUG_ON(list_empty(&info
->indirect_pages
));
527 indirect_page
= list_first_entry(&info
->indirect_pages
,
529 list_del(&indirect_page
->lru
);
530 pfn
= page_to_pfn(indirect_page
);
532 gnt_list_entry
= get_grant(&gref_head
, pfn
, info
);
533 info
->shadow
[id
].indirect_grants
[n
] = gnt_list_entry
;
534 segments
= kmap_atomic(pfn_to_page(gnt_list_entry
->pfn
));
535 ring_req
->u
.indirect
.indirect_grefs
[n
] = gnt_list_entry
->gref
;
538 gnt_list_entry
= get_grant(&gref_head
, page_to_pfn(sg_page(sg
)), info
);
539 ref
= gnt_list_entry
->gref
;
541 info
->shadow
[id
].grants_used
[i
] = gnt_list_entry
;
543 if (rq_data_dir(req
) && info
->feature_persistent
) {
547 BUG_ON(sg
->offset
+ sg
->length
> PAGE_SIZE
);
549 shared_data
= kmap_atomic(pfn_to_page(gnt_list_entry
->pfn
));
550 bvec_data
= kmap_atomic(sg_page(sg
));
553 * this does not wipe data stored outside the
554 * range sg->offset..sg->offset+sg->length.
555 * Therefore, blkback *could* see data from
556 * previous requests. This is OK as long as
557 * persistent grants are shared with just one
558 * domain. It may need refactoring if this
561 memcpy(shared_data
+ sg
->offset
,
562 bvec_data
+ sg
->offset
,
565 kunmap_atomic(bvec_data
);
566 kunmap_atomic(shared_data
);
568 if (ring_req
->operation
!= BLKIF_OP_INDIRECT
) {
569 ring_req
->u
.rw
.seg
[i
] =
570 (struct blkif_request_segment
) {
573 .last_sect
= lsect
};
575 n
= i
% SEGS_PER_INDIRECT_FRAME
;
577 (struct blkif_request_segment
) {
580 .last_sect
= lsect
};
584 kunmap_atomic(segments
);
587 info
->ring
.req_prod_pvt
++;
589 /* Keep a private copy so we can reissue requests when recovering. */
590 info
->shadow
[id
].req
= *ring_req
;
592 if (new_persistent_gnts
)
593 gnttab_free_grant_references(gref_head
);
599 static inline void flush_requests(struct blkfront_info
*info
)
603 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info
->ring
, notify
);
606 notify_remote_via_irq(info
->irq
);
609 static inline bool blkif_request_flush_invalid(struct request
*req
,
610 struct blkfront_info
*info
)
612 return ((req
->cmd_type
!= REQ_TYPE_FS
) ||
613 ((req
->cmd_flags
& REQ_FLUSH
) &&
614 !(info
->feature_flush
& REQ_FLUSH
)) ||
615 ((req
->cmd_flags
& REQ_FUA
) &&
616 !(info
->feature_flush
& REQ_FUA
)));
621 * read a block; request is in a request queue
623 static void do_blkif_request(struct request_queue
*rq
)
625 struct blkfront_info
*info
= NULL
;
629 pr_debug("Entered do_blkif_request\n");
633 while ((req
= blk_peek_request(rq
)) != NULL
) {
634 info
= req
->rq_disk
->private_data
;
636 if (RING_FULL(&info
->ring
))
639 blk_start_request(req
);
641 if (blkif_request_flush_invalid(req
, info
)) {
642 __blk_end_request_all(req
, -EOPNOTSUPP
);
646 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
648 req
, req
->cmd
, (unsigned long)blk_rq_pos(req
),
649 blk_rq_cur_sectors(req
), blk_rq_sectors(req
),
650 rq_data_dir(req
) ? "write" : "read");
652 if (blkif_queue_request(req
)) {
653 blk_requeue_request(rq
, req
);
655 /* Avoid pointless unplugs. */
664 flush_requests(info
);
667 static int xlvbd_init_blk_queue(struct gendisk
*gd
, u16 sector_size
,
668 unsigned int physical_sector_size
,
669 unsigned int segments
)
671 struct request_queue
*rq
;
672 struct blkfront_info
*info
= gd
->private_data
;
674 rq
= blk_init_queue(do_blkif_request
, &info
->io_lock
);
678 queue_flag_set_unlocked(QUEUE_FLAG_VIRT
, rq
);
680 if (info
->feature_discard
) {
681 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, rq
);
682 blk_queue_max_discard_sectors(rq
, get_capacity(gd
));
683 rq
->limits
.discard_granularity
= info
->discard_granularity
;
684 rq
->limits
.discard_alignment
= info
->discard_alignment
;
685 if (info
->feature_secdiscard
)
686 queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD
, rq
);
689 /* Hard sector size and max sectors impersonate the equiv. hardware. */
690 blk_queue_logical_block_size(rq
, sector_size
);
691 blk_queue_physical_block_size(rq
, physical_sector_size
);
692 blk_queue_max_hw_sectors(rq
, (segments
* PAGE_SIZE
) / 512);
694 /* Each segment in a request is up to an aligned page in size. */
695 blk_queue_segment_boundary(rq
, PAGE_SIZE
- 1);
696 blk_queue_max_segment_size(rq
, PAGE_SIZE
);
698 /* Ensure a merged request will fit in a single I/O ring slot. */
699 blk_queue_max_segments(rq
, segments
);
701 /* Make sure buffer addresses are sector-aligned. */
702 blk_queue_dma_alignment(rq
, 511);
704 /* Make sure we don't use bounce buffers. */
705 blk_queue_bounce_limit(rq
, BLK_BOUNCE_ANY
);
712 static const char *flush_info(unsigned int feature_flush
)
714 switch (feature_flush
& ((REQ_FLUSH
| REQ_FUA
))) {
715 case REQ_FLUSH
|REQ_FUA
:
716 return "barrier: enabled;";
718 return "flush diskcache: enabled;";
720 return "barrier or flush: disabled;";
724 static void xlvbd_flush(struct blkfront_info
*info
)
726 blk_queue_flush(info
->rq
, info
->feature_flush
);
727 pr_info("blkfront: %s: %s %s %s %s %s\n",
728 info
->gd
->disk_name
, flush_info(info
->feature_flush
),
729 "persistent grants:", info
->feature_persistent
?
730 "enabled;" : "disabled;", "indirect descriptors:",
731 info
->max_indirect_segments
? "enabled;" : "disabled;");
734 static int xen_translate_vdev(int vdevice
, int *minor
, unsigned int *offset
)
737 major
= BLKIF_MAJOR(vdevice
);
738 *minor
= BLKIF_MINOR(vdevice
);
741 *offset
= (*minor
/ 64) + EMULATED_HD_DISK_NAME_OFFSET
;
742 *minor
= ((*minor
/ 64) * PARTS_PER_DISK
) +
743 EMULATED_HD_DISK_MINOR_OFFSET
;
746 *offset
= (*minor
/ 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET
;
747 *minor
= (((*minor
/ 64) + 2) * PARTS_PER_DISK
) +
748 EMULATED_HD_DISK_MINOR_OFFSET
;
750 case XEN_SCSI_DISK0_MAJOR
:
751 *offset
= (*minor
/ PARTS_PER_DISK
) + EMULATED_SD_DISK_NAME_OFFSET
;
752 *minor
= *minor
+ EMULATED_SD_DISK_MINOR_OFFSET
;
754 case XEN_SCSI_DISK1_MAJOR
:
755 case XEN_SCSI_DISK2_MAJOR
:
756 case XEN_SCSI_DISK3_MAJOR
:
757 case XEN_SCSI_DISK4_MAJOR
:
758 case XEN_SCSI_DISK5_MAJOR
:
759 case XEN_SCSI_DISK6_MAJOR
:
760 case XEN_SCSI_DISK7_MAJOR
:
761 *offset
= (*minor
/ PARTS_PER_DISK
) +
762 ((major
- XEN_SCSI_DISK1_MAJOR
+ 1) * 16) +
763 EMULATED_SD_DISK_NAME_OFFSET
;
765 ((major
- XEN_SCSI_DISK1_MAJOR
+ 1) * 16 * PARTS_PER_DISK
) +
766 EMULATED_SD_DISK_MINOR_OFFSET
;
768 case XEN_SCSI_DISK8_MAJOR
:
769 case XEN_SCSI_DISK9_MAJOR
:
770 case XEN_SCSI_DISK10_MAJOR
:
771 case XEN_SCSI_DISK11_MAJOR
:
772 case XEN_SCSI_DISK12_MAJOR
:
773 case XEN_SCSI_DISK13_MAJOR
:
774 case XEN_SCSI_DISK14_MAJOR
:
775 case XEN_SCSI_DISK15_MAJOR
:
776 *offset
= (*minor
/ PARTS_PER_DISK
) +
777 ((major
- XEN_SCSI_DISK8_MAJOR
+ 8) * 16) +
778 EMULATED_SD_DISK_NAME_OFFSET
;
780 ((major
- XEN_SCSI_DISK8_MAJOR
+ 8) * 16 * PARTS_PER_DISK
) +
781 EMULATED_SD_DISK_MINOR_OFFSET
;
784 *offset
= *minor
/ PARTS_PER_DISK
;
787 printk(KERN_WARNING
"blkfront: your disk configuration is "
788 "incorrect, please use an xvd device instead\n");
794 static char *encode_disk_name(char *ptr
, unsigned int n
)
797 ptr
= encode_disk_name(ptr
, n
/ 26 - 1);
802 static int xlvbd_alloc_gendisk(blkif_sector_t capacity
,
803 struct blkfront_info
*info
,
804 u16 vdisk_info
, u16 sector_size
,
805 unsigned int physical_sector_size
)
815 BUG_ON(info
->gd
!= NULL
);
816 BUG_ON(info
->rq
!= NULL
);
818 if ((info
->vdevice
>>EXT_SHIFT
) > 1) {
819 /* this is above the extended range; something is wrong */
820 printk(KERN_WARNING
"blkfront: vdevice 0x%x is above the extended range; ignoring\n", info
->vdevice
);
824 if (!VDEV_IS_EXTENDED(info
->vdevice
)) {
825 err
= xen_translate_vdev(info
->vdevice
, &minor
, &offset
);
828 nr_parts
= PARTS_PER_DISK
;
830 minor
= BLKIF_MINOR_EXT(info
->vdevice
);
831 nr_parts
= PARTS_PER_EXT_DISK
;
832 offset
= minor
/ nr_parts
;
833 if (xen_hvm_domain() && offset
< EMULATED_HD_DISK_NAME_OFFSET
+ 4)
834 printk(KERN_WARNING
"blkfront: vdevice 0x%x might conflict with "
835 "emulated IDE disks,\n\t choose an xvd device name"
836 "from xvde on\n", info
->vdevice
);
838 if (minor
>> MINORBITS
) {
839 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
840 info
->vdevice
, minor
);
844 if ((minor
% nr_parts
) == 0)
845 nr_minors
= nr_parts
;
847 err
= xlbd_reserve_minors(minor
, nr_minors
);
852 gd
= alloc_disk(nr_minors
);
856 strcpy(gd
->disk_name
, DEV_NAME
);
857 ptr
= encode_disk_name(gd
->disk_name
+ sizeof(DEV_NAME
) - 1, offset
);
858 BUG_ON(ptr
>= gd
->disk_name
+ DISK_NAME_LEN
);
862 snprintf(ptr
, gd
->disk_name
+ DISK_NAME_LEN
- ptr
,
863 "%d", minor
& (nr_parts
- 1));
865 gd
->major
= XENVBD_MAJOR
;
866 gd
->first_minor
= minor
;
867 gd
->fops
= &xlvbd_block_fops
;
868 gd
->private_data
= info
;
869 gd
->driverfs_dev
= &(info
->xbdev
->dev
);
870 set_capacity(gd
, capacity
);
872 if (xlvbd_init_blk_queue(gd
, sector_size
, physical_sector_size
,
873 info
->max_indirect_segments
? :
874 BLKIF_MAX_SEGMENTS_PER_REQUEST
)) {
879 info
->rq
= gd
->queue
;
884 if (vdisk_info
& VDISK_READONLY
)
887 if (vdisk_info
& VDISK_REMOVABLE
)
888 gd
->flags
|= GENHD_FL_REMOVABLE
;
890 if (vdisk_info
& VDISK_CDROM
)
891 gd
->flags
|= GENHD_FL_CD
;
896 xlbd_release_minors(minor
, nr_minors
);
901 static void xlvbd_release_gendisk(struct blkfront_info
*info
)
903 unsigned int minor
, nr_minors
;
906 if (info
->rq
== NULL
)
909 spin_lock_irqsave(&info
->io_lock
, flags
);
911 /* No more blkif_request(). */
912 blk_stop_queue(info
->rq
);
914 /* No more gnttab callback work. */
915 gnttab_cancel_free_callback(&info
->callback
);
916 spin_unlock_irqrestore(&info
->io_lock
, flags
);
918 /* Flush gnttab callback work. Must be done with no locks held. */
919 flush_work(&info
->work
);
921 del_gendisk(info
->gd
);
923 minor
= info
->gd
->first_minor
;
924 nr_minors
= info
->gd
->minors
;
925 xlbd_release_minors(minor
, nr_minors
);
927 blk_cleanup_queue(info
->rq
);
934 static void kick_pending_request_queues(struct blkfront_info
*info
)
936 if (!RING_FULL(&info
->ring
)) {
937 /* Re-enable calldowns. */
938 blk_start_queue(info
->rq
);
939 /* Kick things off immediately. */
940 do_blkif_request(info
->rq
);
944 static void blkif_restart_queue(struct work_struct
*work
)
946 struct blkfront_info
*info
= container_of(work
, struct blkfront_info
, work
);
948 spin_lock_irq(&info
->io_lock
);
949 if (info
->connected
== BLKIF_STATE_CONNECTED
)
950 kick_pending_request_queues(info
);
951 spin_unlock_irq(&info
->io_lock
);
954 static void blkif_free(struct blkfront_info
*info
, int suspend
)
956 struct grant
*persistent_gnt
;
960 /* Prevent new requests being issued until we fix things up. */
961 spin_lock_irq(&info
->io_lock
);
962 info
->connected
= suspend
?
963 BLKIF_STATE_SUSPENDED
: BLKIF_STATE_DISCONNECTED
;
964 /* No more blkif_request(). */
966 blk_stop_queue(info
->rq
);
968 /* Remove all persistent grants */
969 if (!list_empty(&info
->grants
)) {
970 list_for_each_entry_safe(persistent_gnt
, n
,
971 &info
->grants
, node
) {
972 list_del(&persistent_gnt
->node
);
973 if (persistent_gnt
->gref
!= GRANT_INVALID_REF
) {
974 gnttab_end_foreign_access(persistent_gnt
->gref
,
976 info
->persistent_gnts_c
--;
978 if (info
->feature_persistent
)
979 __free_page(pfn_to_page(persistent_gnt
->pfn
));
980 kfree(persistent_gnt
);
983 BUG_ON(info
->persistent_gnts_c
!= 0);
986 * Remove indirect pages, this only happens when using indirect
987 * descriptors but not persistent grants
989 if (!list_empty(&info
->indirect_pages
)) {
990 struct page
*indirect_page
, *n
;
992 BUG_ON(info
->feature_persistent
);
993 list_for_each_entry_safe(indirect_page
, n
, &info
->indirect_pages
, lru
) {
994 list_del(&indirect_page
->lru
);
995 __free_page(indirect_page
);
999 for (i
= 0; i
< BLK_RING_SIZE(info
); i
++) {
1001 * Clear persistent grants present in requests already
1002 * on the shared ring
1004 if (!info
->shadow
[i
].request
)
1007 segs
= info
->shadow
[i
].req
.operation
== BLKIF_OP_INDIRECT
?
1008 info
->shadow
[i
].req
.u
.indirect
.nr_segments
:
1009 info
->shadow
[i
].req
.u
.rw
.nr_segments
;
1010 for (j
= 0; j
< segs
; j
++) {
1011 persistent_gnt
= info
->shadow
[i
].grants_used
[j
];
1012 gnttab_end_foreign_access(persistent_gnt
->gref
, 0, 0UL);
1013 if (info
->feature_persistent
)
1014 __free_page(pfn_to_page(persistent_gnt
->pfn
));
1015 kfree(persistent_gnt
);
1018 if (info
->shadow
[i
].req
.operation
!= BLKIF_OP_INDIRECT
)
1020 * If this is not an indirect operation don't try to
1021 * free indirect segments
1025 for (j
= 0; j
< INDIRECT_GREFS(segs
); j
++) {
1026 persistent_gnt
= info
->shadow
[i
].indirect_grants
[j
];
1027 gnttab_end_foreign_access(persistent_gnt
->gref
, 0, 0UL);
1028 __free_page(pfn_to_page(persistent_gnt
->pfn
));
1029 kfree(persistent_gnt
);
1033 kfree(info
->shadow
[i
].grants_used
);
1034 info
->shadow
[i
].grants_used
= NULL
;
1035 kfree(info
->shadow
[i
].indirect_grants
);
1036 info
->shadow
[i
].indirect_grants
= NULL
;
1037 kfree(info
->shadow
[i
].sg
);
1038 info
->shadow
[i
].sg
= NULL
;
1041 /* No more gnttab callback work. */
1042 gnttab_cancel_free_callback(&info
->callback
);
1043 spin_unlock_irq(&info
->io_lock
);
1045 /* Flush gnttab callback work. Must be done with no locks held. */
1046 flush_work(&info
->work
);
1048 /* Free resources associated with old device channel. */
1049 for (i
= 0; i
< info
->nr_ring_pages
; i
++) {
1050 if (info
->ring_ref
[i
] != GRANT_INVALID_REF
) {
1051 gnttab_end_foreign_access(info
->ring_ref
[i
], 0, 0);
1052 info
->ring_ref
[i
] = GRANT_INVALID_REF
;
1055 free_pages((unsigned long)info
->ring
.sring
, get_order(info
->nr_ring_pages
* PAGE_SIZE
));
1056 info
->ring
.sring
= NULL
;
1059 unbind_from_irqhandler(info
->irq
, info
);
1060 info
->evtchn
= info
->irq
= 0;
1064 static void blkif_completion(struct blk_shadow
*s
, struct blkfront_info
*info
,
1065 struct blkif_response
*bret
)
1068 struct scatterlist
*sg
;
1073 nseg
= s
->req
.operation
== BLKIF_OP_INDIRECT
?
1074 s
->req
.u
.indirect
.nr_segments
: s
->req
.u
.rw
.nr_segments
;
1076 if (bret
->operation
== BLKIF_OP_READ
&& info
->feature_persistent
) {
1078 * Copy the data received from the backend into the bvec.
1079 * Since bv_offset can be different than 0, and bv_len different
1080 * than PAGE_SIZE, we have to keep track of the current offset,
1081 * to be sure we are copying the data from the right shared page.
1083 for_each_sg(s
->sg
, sg
, nseg
, i
) {
1084 BUG_ON(sg
->offset
+ sg
->length
> PAGE_SIZE
);
1085 shared_data
= kmap_atomic(
1086 pfn_to_page(s
->grants_used
[i
]->pfn
));
1087 bvec_data
= kmap_atomic(sg_page(sg
));
1088 memcpy(bvec_data
+ sg
->offset
,
1089 shared_data
+ sg
->offset
,
1091 kunmap_atomic(bvec_data
);
1092 kunmap_atomic(shared_data
);
1095 /* Add the persistent grant into the list of free grants */
1096 for (i
= 0; i
< nseg
; i
++) {
1097 if (gnttab_query_foreign_access(s
->grants_used
[i
]->gref
)) {
1099 * If the grant is still mapped by the backend (the
1100 * backend has chosen to make this grant persistent)
1101 * we add it at the head of the list, so it will be
1104 if (!info
->feature_persistent
)
1105 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1106 s
->grants_used
[i
]->gref
);
1107 list_add(&s
->grants_used
[i
]->node
, &info
->grants
);
1108 info
->persistent_gnts_c
++;
1111 * If the grant is not mapped by the backend we end the
1112 * foreign access and add it to the tail of the list,
1113 * so it will not be picked again unless we run out of
1114 * persistent grants.
1116 gnttab_end_foreign_access(s
->grants_used
[i
]->gref
, 0, 0UL);
1117 s
->grants_used
[i
]->gref
= GRANT_INVALID_REF
;
1118 list_add_tail(&s
->grants_used
[i
]->node
, &info
->grants
);
1121 if (s
->req
.operation
== BLKIF_OP_INDIRECT
) {
1122 for (i
= 0; i
< INDIRECT_GREFS(nseg
); i
++) {
1123 if (gnttab_query_foreign_access(s
->indirect_grants
[i
]->gref
)) {
1124 if (!info
->feature_persistent
)
1125 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1126 s
->indirect_grants
[i
]->gref
);
1127 list_add(&s
->indirect_grants
[i
]->node
, &info
->grants
);
1128 info
->persistent_gnts_c
++;
1130 struct page
*indirect_page
;
1132 gnttab_end_foreign_access(s
->indirect_grants
[i
]->gref
, 0, 0UL);
1134 * Add the used indirect page back to the list of
1135 * available pages for indirect grefs.
1137 indirect_page
= pfn_to_page(s
->indirect_grants
[i
]->pfn
);
1138 list_add(&indirect_page
->lru
, &info
->indirect_pages
);
1139 s
->indirect_grants
[i
]->gref
= GRANT_INVALID_REF
;
1140 list_add_tail(&s
->indirect_grants
[i
]->node
, &info
->grants
);
1146 static irqreturn_t
blkif_interrupt(int irq
, void *dev_id
)
1148 struct request
*req
;
1149 struct blkif_response
*bret
;
1151 unsigned long flags
;
1152 struct blkfront_info
*info
= (struct blkfront_info
*)dev_id
;
1155 spin_lock_irqsave(&info
->io_lock
, flags
);
1157 if (unlikely(info
->connected
!= BLKIF_STATE_CONNECTED
)) {
1158 spin_unlock_irqrestore(&info
->io_lock
, flags
);
1163 rp
= info
->ring
.sring
->rsp_prod
;
1164 rmb(); /* Ensure we see queued responses up to 'rp'. */
1166 for (i
= info
->ring
.rsp_cons
; i
!= rp
; i
++) {
1169 bret
= RING_GET_RESPONSE(&info
->ring
, i
);
1172 * The backend has messed up and given us an id that we would
1173 * never have given to it (we stamp it up to BLK_RING_SIZE -
1174 * look in get_id_from_freelist.
1176 if (id
>= BLK_RING_SIZE(info
)) {
1177 WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1178 info
->gd
->disk_name
, op_name(bret
->operation
), id
);
1179 /* We can't safely get the 'struct request' as
1180 * the id is busted. */
1183 req
= info
->shadow
[id
].request
;
1185 if (bret
->operation
!= BLKIF_OP_DISCARD
)
1186 blkif_completion(&info
->shadow
[id
], info
, bret
);
1188 if (add_id_to_freelist(info
, id
)) {
1189 WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1190 info
->gd
->disk_name
, op_name(bret
->operation
), id
);
1194 error
= (bret
->status
== BLKIF_RSP_OKAY
) ? 0 : -EIO
;
1195 switch (bret
->operation
) {
1196 case BLKIF_OP_DISCARD
:
1197 if (unlikely(bret
->status
== BLKIF_RSP_EOPNOTSUPP
)) {
1198 struct request_queue
*rq
= info
->rq
;
1199 printk(KERN_WARNING
"blkfront: %s: %s op failed\n",
1200 info
->gd
->disk_name
, op_name(bret
->operation
));
1201 error
= -EOPNOTSUPP
;
1202 info
->feature_discard
= 0;
1203 info
->feature_secdiscard
= 0;
1204 queue_flag_clear(QUEUE_FLAG_DISCARD
, rq
);
1205 queue_flag_clear(QUEUE_FLAG_SECDISCARD
, rq
);
1207 __blk_end_request_all(req
, error
);
1209 case BLKIF_OP_FLUSH_DISKCACHE
:
1210 case BLKIF_OP_WRITE_BARRIER
:
1211 if (unlikely(bret
->status
== BLKIF_RSP_EOPNOTSUPP
)) {
1212 printk(KERN_WARNING
"blkfront: %s: %s op failed\n",
1213 info
->gd
->disk_name
, op_name(bret
->operation
));
1214 error
= -EOPNOTSUPP
;
1216 if (unlikely(bret
->status
== BLKIF_RSP_ERROR
&&
1217 info
->shadow
[id
].req
.u
.rw
.nr_segments
== 0)) {
1218 printk(KERN_WARNING
"blkfront: %s: empty %s op failed\n",
1219 info
->gd
->disk_name
, op_name(bret
->operation
));
1220 error
= -EOPNOTSUPP
;
1222 if (unlikely(error
)) {
1223 if (error
== -EOPNOTSUPP
)
1225 info
->feature_flush
= 0;
1230 case BLKIF_OP_WRITE
:
1231 if (unlikely(bret
->status
!= BLKIF_RSP_OKAY
))
1232 dev_dbg(&info
->xbdev
->dev
, "Bad return from blkdev data "
1233 "request: %x\n", bret
->status
);
1235 __blk_end_request_all(req
, error
);
1242 info
->ring
.rsp_cons
= i
;
1244 if (i
!= info
->ring
.req_prod_pvt
) {
1246 RING_FINAL_CHECK_FOR_RESPONSES(&info
->ring
, more_to_do
);
1250 info
->ring
.sring
->rsp_event
= i
+ 1;
1252 kick_pending_request_queues(info
);
1254 spin_unlock_irqrestore(&info
->io_lock
, flags
);
1260 static int setup_blkring(struct xenbus_device
*dev
,
1261 struct blkfront_info
*info
)
1263 struct blkif_sring
*sring
;
1265 unsigned long ring_size
= info
->nr_ring_pages
* PAGE_SIZE
;
1266 grant_ref_t gref
[XENBUS_MAX_RING_PAGES
];
1268 for (i
= 0; i
< info
->nr_ring_pages
; i
++)
1269 info
->ring_ref
[i
] = GRANT_INVALID_REF
;
1271 sring
= (struct blkif_sring
*)__get_free_pages(GFP_NOIO
| __GFP_HIGH
,
1272 get_order(ring_size
));
1274 xenbus_dev_fatal(dev
, -ENOMEM
, "allocating shared ring");
1277 SHARED_RING_INIT(sring
);
1278 FRONT_RING_INIT(&info
->ring
, sring
, ring_size
);
1280 err
= xenbus_grant_ring(dev
, info
->ring
.sring
, info
->nr_ring_pages
, gref
);
1282 free_pages((unsigned long)sring
, get_order(ring_size
));
1283 info
->ring
.sring
= NULL
;
1286 for (i
= 0; i
< info
->nr_ring_pages
; i
++)
1287 info
->ring_ref
[i
] = gref
[i
];
1289 err
= xenbus_alloc_evtchn(dev
, &info
->evtchn
);
1293 err
= bind_evtchn_to_irqhandler(info
->evtchn
, blkif_interrupt
, 0,
1296 xenbus_dev_fatal(dev
, err
,
1297 "bind_evtchn_to_irqhandler failed");
1304 blkif_free(info
, 0);
1309 /* Common code used when first setting up, and when resuming. */
1310 static int talk_to_blkback(struct xenbus_device
*dev
,
1311 struct blkfront_info
*info
)
1313 const char *message
= NULL
;
1314 struct xenbus_transaction xbt
;
1316 unsigned int max_page_order
= 0;
1317 unsigned int ring_page_order
= 0;
1319 err
= xenbus_scanf(XBT_NIL
, info
->xbdev
->otherend
,
1320 "max-ring-page-order", "%u", &max_page_order
);
1322 info
->nr_ring_pages
= 1;
1324 ring_page_order
= min(xen_blkif_max_ring_order
, max_page_order
);
1325 info
->nr_ring_pages
= 1 << ring_page_order
;
1328 /* Create shared ring, alloc event channel. */
1329 err
= setup_blkring(dev
, info
);
1334 err
= xenbus_transaction_start(&xbt
);
1336 xenbus_dev_fatal(dev
, err
, "starting transaction");
1337 goto destroy_blkring
;
1340 if (info
->nr_ring_pages
== 1) {
1341 err
= xenbus_printf(xbt
, dev
->nodename
,
1342 "ring-ref", "%u", info
->ring_ref
[0]);
1344 message
= "writing ring-ref";
1345 goto abort_transaction
;
1348 err
= xenbus_printf(xbt
, dev
->nodename
,
1349 "ring-page-order", "%u", ring_page_order
);
1351 message
= "writing ring-page-order";
1352 goto abort_transaction
;
1355 for (i
= 0; i
< info
->nr_ring_pages
; i
++) {
1356 char ring_ref_name
[RINGREF_NAME_LEN
];
1358 snprintf(ring_ref_name
, RINGREF_NAME_LEN
, "ring-ref%u", i
);
1359 err
= xenbus_printf(xbt
, dev
->nodename
, ring_ref_name
,
1360 "%u", info
->ring_ref
[i
]);
1362 message
= "writing ring-ref";
1363 goto abort_transaction
;
1367 err
= xenbus_printf(xbt
, dev
->nodename
,
1368 "event-channel", "%u", info
->evtchn
);
1370 message
= "writing event-channel";
1371 goto abort_transaction
;
1373 err
= xenbus_printf(xbt
, dev
->nodename
, "protocol", "%s",
1374 XEN_IO_PROTO_ABI_NATIVE
);
1376 message
= "writing protocol";
1377 goto abort_transaction
;
1379 err
= xenbus_printf(xbt
, dev
->nodename
,
1380 "feature-persistent", "%u", 1);
1383 "writing persistent grants feature to xenbus");
1385 err
= xenbus_transaction_end(xbt
, 0);
1389 xenbus_dev_fatal(dev
, err
, "completing transaction");
1390 goto destroy_blkring
;
1393 for (i
= 0; i
< BLK_RING_SIZE(info
); i
++)
1394 info
->shadow
[i
].req
.u
.rw
.id
= i
+1;
1395 info
->shadow
[BLK_RING_SIZE(info
)-1].req
.u
.rw
.id
= 0x0fffffff;
1396 xenbus_switch_state(dev
, XenbusStateInitialised
);
1401 xenbus_transaction_end(xbt
, 1);
1403 xenbus_dev_fatal(dev
, err
, "%s", message
);
1405 blkif_free(info
, 0);
1411 * Entry point to this code when a new device is created. Allocate the basic
1412 * structures and the ring buffer for communication with the backend, and
1413 * inform the backend of the appropriate details for those. Switch to
1414 * Initialised state.
1416 static int blkfront_probe(struct xenbus_device
*dev
,
1417 const struct xenbus_device_id
*id
)
1420 struct blkfront_info
*info
;
1422 /* FIXME: Use dynamic device id if this is not set. */
1423 err
= xenbus_scanf(XBT_NIL
, dev
->nodename
,
1424 "virtual-device", "%i", &vdevice
);
1426 /* go looking in the extended area instead */
1427 err
= xenbus_scanf(XBT_NIL
, dev
->nodename
, "virtual-device-ext",
1430 xenbus_dev_fatal(dev
, err
, "reading virtual-device");
1435 if (xen_hvm_domain()) {
1438 /* no unplug has been done: do not hook devices != xen vbds */
1439 if (xen_has_pv_and_legacy_disk_devices()) {
1442 if (!VDEV_IS_EXTENDED(vdevice
))
1443 major
= BLKIF_MAJOR(vdevice
);
1445 major
= XENVBD_MAJOR
;
1447 if (major
!= XENVBD_MAJOR
) {
1449 "%s: HVM does not support vbd %d as xen block device\n",
1454 /* do not create a PV cdrom device if we are an HVM guest */
1455 type
= xenbus_read(XBT_NIL
, dev
->nodename
, "device-type", &len
);
1458 if (strncmp(type
, "cdrom", 5) == 0) {
1464 info
= kzalloc(sizeof(*info
), GFP_KERNEL
);
1466 xenbus_dev_fatal(dev
, -ENOMEM
, "allocating info structure");
1470 mutex_init(&info
->mutex
);
1471 spin_lock_init(&info
->io_lock
);
1473 info
->vdevice
= vdevice
;
1474 INIT_LIST_HEAD(&info
->grants
);
1475 INIT_LIST_HEAD(&info
->indirect_pages
);
1476 info
->persistent_gnts_c
= 0;
1477 info
->connected
= BLKIF_STATE_DISCONNECTED
;
1478 INIT_WORK(&info
->work
, blkif_restart_queue
);
1480 /* Front end dir is a number, which is used as the id. */
1481 info
->handle
= simple_strtoul(strrchr(dev
->nodename
, '/')+1, NULL
, 0);
1482 dev_set_drvdata(&dev
->dev
, info
);
1487 static void split_bio_end(struct bio
*bio
, int error
)
1489 struct split_bio
*split_bio
= bio
->bi_private
;
1492 split_bio
->err
= error
;
1494 if (atomic_dec_and_test(&split_bio
->pending
)) {
1495 split_bio
->bio
->bi_phys_segments
= 0;
1496 bio_endio(split_bio
->bio
, split_bio
->err
);
1502 static int blkif_recover(struct blkfront_info
*info
)
1505 struct request
*req
, *n
;
1506 struct blk_shadow
*copy
;
1508 struct bio
*bio
, *cloned_bio
;
1509 struct bio_list bio_list
, merge_bio
;
1510 unsigned int segs
, offset
;
1512 struct split_bio
*split_bio
;
1513 struct list_head requests
;
1515 /* Stage 1: Make a safe copy of the shadow state. */
1516 copy
= kmemdup(info
->shadow
, sizeof(info
->shadow
),
1517 GFP_NOIO
| __GFP_REPEAT
| __GFP_HIGH
);
1521 /* Stage 2: Set up free list. */
1522 memset(&info
->shadow
, 0, sizeof(info
->shadow
));
1523 for (i
= 0; i
< BLK_RING_SIZE(info
); i
++)
1524 info
->shadow
[i
].req
.u
.rw
.id
= i
+1;
1525 info
->shadow_free
= info
->ring
.req_prod_pvt
;
1526 info
->shadow
[BLK_RING_SIZE(info
)-1].req
.u
.rw
.id
= 0x0fffffff;
1528 rc
= blkfront_setup_indirect(info
);
1534 segs
= info
->max_indirect_segments
? : BLKIF_MAX_SEGMENTS_PER_REQUEST
;
1535 blk_queue_max_segments(info
->rq
, segs
);
1536 bio_list_init(&bio_list
);
1537 INIT_LIST_HEAD(&requests
);
1538 for (i
= 0; i
< BLK_RING_SIZE(info
); i
++) {
1540 if (!copy
[i
].request
)
1544 * Get the bios in the request so we can re-queue them.
1546 if (copy
[i
].request
->cmd_flags
&
1547 (REQ_FLUSH
| REQ_FUA
| REQ_DISCARD
| REQ_SECURE
)) {
1549 * Flush operations don't contain bios, so
1550 * we need to requeue the whole request
1552 list_add(©
[i
].request
->queuelist
, &requests
);
1555 merge_bio
.head
= copy
[i
].request
->bio
;
1556 merge_bio
.tail
= copy
[i
].request
->biotail
;
1557 bio_list_merge(&bio_list
, &merge_bio
);
1558 copy
[i
].request
->bio
= NULL
;
1559 blk_end_request_all(copy
[i
].request
, 0);
1565 * Empty the queue, this is important because we might have
1566 * requests in the queue with more segments than what we
1569 spin_lock_irq(&info
->io_lock
);
1570 while ((req
= blk_fetch_request(info
->rq
)) != NULL
) {
1571 if (req
->cmd_flags
&
1572 (REQ_FLUSH
| REQ_FUA
| REQ_DISCARD
| REQ_SECURE
)) {
1573 list_add(&req
->queuelist
, &requests
);
1576 merge_bio
.head
= req
->bio
;
1577 merge_bio
.tail
= req
->biotail
;
1578 bio_list_merge(&bio_list
, &merge_bio
);
1580 if (req
->cmd_flags
& (REQ_FLUSH
| REQ_FUA
))
1581 pr_alert("diskcache flush request found!\n");
1582 __blk_end_request_all(req
, 0);
1584 spin_unlock_irq(&info
->io_lock
);
1586 xenbus_switch_state(info
->xbdev
, XenbusStateConnected
);
1588 spin_lock_irq(&info
->io_lock
);
1590 /* Now safe for us to use the shared ring */
1591 info
->connected
= BLKIF_STATE_CONNECTED
;
1593 /* Kick any other new requests queued since we resumed */
1594 kick_pending_request_queues(info
);
1596 list_for_each_entry_safe(req
, n
, &requests
, queuelist
) {
1597 /* Requeue pending requests (flush or discard) */
1598 list_del_init(&req
->queuelist
);
1599 BUG_ON(req
->nr_phys_segments
> segs
);
1600 blk_requeue_request(info
->rq
, req
);
1602 spin_unlock_irq(&info
->io_lock
);
1604 while ((bio
= bio_list_pop(&bio_list
)) != NULL
) {
1605 /* Traverse the list of pending bios and re-queue them */
1606 if (bio_segments(bio
) > segs
) {
1608 * This bio has more segments than what we can
1609 * handle, we have to split it.
1611 pending
= (bio_segments(bio
) + segs
- 1) / segs
;
1612 split_bio
= kzalloc(sizeof(*split_bio
), GFP_NOIO
);
1613 BUG_ON(split_bio
== NULL
);
1614 atomic_set(&split_bio
->pending
, pending
);
1615 split_bio
->bio
= bio
;
1616 for (i
= 0; i
< pending
; i
++) {
1617 offset
= (i
* segs
* PAGE_SIZE
) >> 9;
1618 size
= min((unsigned int)(segs
* PAGE_SIZE
) >> 9,
1619 (unsigned int)bio_sectors(bio
) - offset
);
1620 cloned_bio
= bio_clone(bio
, GFP_NOIO
);
1621 BUG_ON(cloned_bio
== NULL
);
1622 bio_trim(cloned_bio
, offset
, size
);
1623 cloned_bio
->bi_private
= split_bio
;
1624 cloned_bio
->bi_end_io
= split_bio_end
;
1625 submit_bio(cloned_bio
->bi_rw
, cloned_bio
);
1628 * Now we have to wait for all those smaller bios to
1629 * end, so we can also end the "parent" bio.
1633 /* We don't need to split this bio */
1634 submit_bio(bio
->bi_rw
, bio
);
1641 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1642 * driver restart. We tear down our blkif structure and recreate it, but
1643 * leave the device-layer structures intact so that this is transparent to the
1644 * rest of the kernel.
1646 static int blkfront_resume(struct xenbus_device
*dev
)
1648 struct blkfront_info
*info
= dev_get_drvdata(&dev
->dev
);
1651 dev_dbg(&dev
->dev
, "blkfront_resume: %s\n", dev
->nodename
);
1653 blkif_free(info
, info
->connected
== BLKIF_STATE_CONNECTED
);
1655 err
= talk_to_blkback(dev
, info
);
1658 * We have to wait for the backend to switch to
1659 * connected state, since we want to read which
1660 * features it supports.
1667 blkfront_closing(struct blkfront_info
*info
)
1669 struct xenbus_device
*xbdev
= info
->xbdev
;
1670 struct block_device
*bdev
= NULL
;
1672 mutex_lock(&info
->mutex
);
1674 if (xbdev
->state
== XenbusStateClosing
) {
1675 mutex_unlock(&info
->mutex
);
1680 bdev
= bdget_disk(info
->gd
, 0);
1682 mutex_unlock(&info
->mutex
);
1685 xenbus_frontend_closed(xbdev
);
1689 mutex_lock(&bdev
->bd_mutex
);
1691 if (bdev
->bd_openers
) {
1692 xenbus_dev_error(xbdev
, -EBUSY
,
1693 "Device in use; refusing to close");
1694 xenbus_switch_state(xbdev
, XenbusStateClosing
);
1696 xlvbd_release_gendisk(info
);
1697 xenbus_frontend_closed(xbdev
);
1700 mutex_unlock(&bdev
->bd_mutex
);
1704 static void blkfront_setup_discard(struct blkfront_info
*info
)
1707 unsigned int discard_granularity
;
1708 unsigned int discard_alignment
;
1709 unsigned int discard_secure
;
1711 info
->feature_discard
= 1;
1712 err
= xenbus_gather(XBT_NIL
, info
->xbdev
->otherend
,
1713 "discard-granularity", "%u", &discard_granularity
,
1714 "discard-alignment", "%u", &discard_alignment
,
1717 info
->discard_granularity
= discard_granularity
;
1718 info
->discard_alignment
= discard_alignment
;
1720 err
= xenbus_gather(XBT_NIL
, info
->xbdev
->otherend
,
1721 "discard-secure", "%d", &discard_secure
,
1724 info
->feature_secdiscard
= !!discard_secure
;
1727 static int blkfront_setup_indirect(struct blkfront_info
*info
)
1729 unsigned int indirect_segments
, segs
;
1732 err
= xenbus_gather(XBT_NIL
, info
->xbdev
->otherend
,
1733 "feature-max-indirect-segments", "%u", &indirect_segments
,
1736 info
->max_indirect_segments
= 0;
1737 segs
= BLKIF_MAX_SEGMENTS_PER_REQUEST
;
1739 info
->max_indirect_segments
= min(indirect_segments
,
1740 xen_blkif_max_segments
);
1741 segs
= info
->max_indirect_segments
;
1744 err
= fill_grant_buffer(info
, (segs
+ INDIRECT_GREFS(segs
)) * BLK_RING_SIZE(info
));
1748 if (!info
->feature_persistent
&& info
->max_indirect_segments
) {
1750 * We are using indirect descriptors but not persistent
1751 * grants, we need to allocate a set of pages that can be
1752 * used for mapping indirect grefs
1754 int num
= INDIRECT_GREFS(segs
) * BLK_RING_SIZE(info
);
1756 BUG_ON(!list_empty(&info
->indirect_pages
));
1757 for (i
= 0; i
< num
; i
++) {
1758 struct page
*indirect_page
= alloc_page(GFP_NOIO
);
1761 list_add(&indirect_page
->lru
, &info
->indirect_pages
);
1765 for (i
= 0; i
< BLK_RING_SIZE(info
); i
++) {
1766 info
->shadow
[i
].grants_used
= kzalloc(
1767 sizeof(info
->shadow
[i
].grants_used
[0]) * segs
,
1769 info
->shadow
[i
].sg
= kzalloc(sizeof(info
->shadow
[i
].sg
[0]) * segs
, GFP_NOIO
);
1770 if (info
->max_indirect_segments
)
1771 info
->shadow
[i
].indirect_grants
= kzalloc(
1772 sizeof(info
->shadow
[i
].indirect_grants
[0]) *
1773 INDIRECT_GREFS(segs
),
1775 if ((info
->shadow
[i
].grants_used
== NULL
) ||
1776 (info
->shadow
[i
].sg
== NULL
) ||
1777 (info
->max_indirect_segments
&&
1778 (info
->shadow
[i
].indirect_grants
== NULL
)))
1780 sg_init_table(info
->shadow
[i
].sg
, segs
);
1787 for (i
= 0; i
< BLK_RING_SIZE(info
); i
++) {
1788 kfree(info
->shadow
[i
].grants_used
);
1789 info
->shadow
[i
].grants_used
= NULL
;
1790 kfree(info
->shadow
[i
].sg
);
1791 info
->shadow
[i
].sg
= NULL
;
1792 kfree(info
->shadow
[i
].indirect_grants
);
1793 info
->shadow
[i
].indirect_grants
= NULL
;
1795 if (!list_empty(&info
->indirect_pages
)) {
1796 struct page
*indirect_page
, *n
;
1797 list_for_each_entry_safe(indirect_page
, n
, &info
->indirect_pages
, lru
) {
1798 list_del(&indirect_page
->lru
);
1799 __free_page(indirect_page
);
1806 * Invoked when the backend is finally 'ready' (and has told produced
1807 * the details about the physical device - #sectors, size, etc).
1809 static void blkfront_connect(struct blkfront_info
*info
)
1811 unsigned long long sectors
;
1812 unsigned long sector_size
;
1813 unsigned int physical_sector_size
;
1816 int barrier
, flush
, discard
, persistent
;
1818 switch (info
->connected
) {
1819 case BLKIF_STATE_CONNECTED
:
1821 * Potentially, the back-end may be signalling
1822 * a capacity change; update the capacity.
1824 err
= xenbus_scanf(XBT_NIL
, info
->xbdev
->otherend
,
1825 "sectors", "%Lu", §ors
);
1826 if (XENBUS_EXIST_ERR(err
))
1828 printk(KERN_INFO
"Setting capacity to %Lu\n",
1830 set_capacity(info
->gd
, sectors
);
1831 revalidate_disk(info
->gd
);
1834 case BLKIF_STATE_SUSPENDED
:
1836 * If we are recovering from suspension, we need to wait
1837 * for the backend to announce it's features before
1838 * reconnecting, at least we need to know if the backend
1839 * supports indirect descriptors, and how many.
1841 blkif_recover(info
);
1848 dev_dbg(&info
->xbdev
->dev
, "%s:%s.\n",
1849 __func__
, info
->xbdev
->otherend
);
1851 err
= xenbus_gather(XBT_NIL
, info
->xbdev
->otherend
,
1852 "sectors", "%llu", §ors
,
1853 "info", "%u", &binfo
,
1854 "sector-size", "%lu", §or_size
,
1857 xenbus_dev_fatal(info
->xbdev
, err
,
1858 "reading backend fields at %s",
1859 info
->xbdev
->otherend
);
1864 * physcial-sector-size is a newer field, so old backends may not
1865 * provide this. Assume physical sector size to be the same as
1866 * sector_size in that case.
1868 err
= xenbus_scanf(XBT_NIL
, info
->xbdev
->otherend
,
1869 "physical-sector-size", "%u", &physical_sector_size
);
1871 physical_sector_size
= sector_size
;
1873 info
->feature_flush
= 0;
1875 err
= xenbus_gather(XBT_NIL
, info
->xbdev
->otherend
,
1876 "feature-barrier", "%d", &barrier
,
1880 * If there's no "feature-barrier" defined, then it means
1881 * we're dealing with a very old backend which writes
1882 * synchronously; nothing to do.
1884 * If there are barriers, then we use flush.
1886 if (!err
&& barrier
)
1887 info
->feature_flush
= REQ_FLUSH
| REQ_FUA
;
1889 * And if there is "feature-flush-cache" use that above
1892 err
= xenbus_gather(XBT_NIL
, info
->xbdev
->otherend
,
1893 "feature-flush-cache", "%d", &flush
,
1897 info
->feature_flush
= REQ_FLUSH
;
1899 err
= xenbus_gather(XBT_NIL
, info
->xbdev
->otherend
,
1900 "feature-discard", "%d", &discard
,
1903 if (!err
&& discard
)
1904 blkfront_setup_discard(info
);
1906 err
= xenbus_gather(XBT_NIL
, info
->xbdev
->otherend
,
1907 "feature-persistent", "%u", &persistent
,
1910 info
->feature_persistent
= 0;
1912 info
->feature_persistent
= persistent
;
1914 err
= blkfront_setup_indirect(info
);
1916 xenbus_dev_fatal(info
->xbdev
, err
, "setup_indirect at %s",
1917 info
->xbdev
->otherend
);
1921 err
= xlvbd_alloc_gendisk(sectors
, info
, binfo
, sector_size
,
1922 physical_sector_size
);
1924 xenbus_dev_fatal(info
->xbdev
, err
, "xlvbd_add at %s",
1925 info
->xbdev
->otherend
);
1929 xenbus_switch_state(info
->xbdev
, XenbusStateConnected
);
1931 /* Kick pending requests. */
1932 spin_lock_irq(&info
->io_lock
);
1933 info
->connected
= BLKIF_STATE_CONNECTED
;
1934 kick_pending_request_queues(info
);
1935 spin_unlock_irq(&info
->io_lock
);
1943 * Callback received when the backend's state changes.
1945 static void blkback_changed(struct xenbus_device
*dev
,
1946 enum xenbus_state backend_state
)
1948 struct blkfront_info
*info
= dev_get_drvdata(&dev
->dev
);
1950 dev_dbg(&dev
->dev
, "blkfront:blkback_changed to state %d.\n", backend_state
);
1952 switch (backend_state
) {
1953 case XenbusStateInitWait
:
1954 if (dev
->state
!= XenbusStateInitialising
)
1956 if (talk_to_blkback(dev
, info
)) {
1958 dev_set_drvdata(&dev
->dev
, NULL
);
1961 case XenbusStateInitialising
:
1962 case XenbusStateInitialised
:
1963 case XenbusStateReconfiguring
:
1964 case XenbusStateReconfigured
:
1965 case XenbusStateUnknown
:
1968 case XenbusStateConnected
:
1969 blkfront_connect(info
);
1972 case XenbusStateClosed
:
1973 if (dev
->state
== XenbusStateClosed
)
1975 /* Missed the backend's Closing state -- fallthrough */
1976 case XenbusStateClosing
:
1977 blkfront_closing(info
);
1982 static int blkfront_remove(struct xenbus_device
*xbdev
)
1984 struct blkfront_info
*info
= dev_get_drvdata(&xbdev
->dev
);
1985 struct block_device
*bdev
= NULL
;
1986 struct gendisk
*disk
;
1988 dev_dbg(&xbdev
->dev
, "%s removed", xbdev
->nodename
);
1990 blkif_free(info
, 0);
1992 mutex_lock(&info
->mutex
);
1996 bdev
= bdget_disk(disk
, 0);
1999 mutex_unlock(&info
->mutex
);
2007 * The xbdev was removed before we reached the Closed
2008 * state. See if it's safe to remove the disk. If the bdev
2009 * isn't closed yet, we let release take care of it.
2012 mutex_lock(&bdev
->bd_mutex
);
2013 info
= disk
->private_data
;
2015 dev_warn(disk_to_dev(disk
),
2016 "%s was hot-unplugged, %d stale handles\n",
2017 xbdev
->nodename
, bdev
->bd_openers
);
2019 if (info
&& !bdev
->bd_openers
) {
2020 xlvbd_release_gendisk(info
);
2021 disk
->private_data
= NULL
;
2025 mutex_unlock(&bdev
->bd_mutex
);
2031 static int blkfront_is_ready(struct xenbus_device
*dev
)
2033 struct blkfront_info
*info
= dev_get_drvdata(&dev
->dev
);
2035 return info
->is_ready
&& info
->xbdev
;
2038 static int blkif_open(struct block_device
*bdev
, fmode_t mode
)
2040 struct gendisk
*disk
= bdev
->bd_disk
;
2041 struct blkfront_info
*info
;
2044 mutex_lock(&blkfront_mutex
);
2046 info
= disk
->private_data
;
2053 mutex_lock(&info
->mutex
);
2056 /* xbdev is closed */
2059 mutex_unlock(&info
->mutex
);
2062 mutex_unlock(&blkfront_mutex
);
2066 static void blkif_release(struct gendisk
*disk
, fmode_t mode
)
2068 struct blkfront_info
*info
= disk
->private_data
;
2069 struct block_device
*bdev
;
2070 struct xenbus_device
*xbdev
;
2072 mutex_lock(&blkfront_mutex
);
2074 bdev
= bdget_disk(disk
, 0);
2077 WARN(1, "Block device %s yanked out from us!\n", disk
->disk_name
);
2080 if (bdev
->bd_openers
)
2084 * Check if we have been instructed to close. We will have
2085 * deferred this request, because the bdev was still open.
2088 mutex_lock(&info
->mutex
);
2089 xbdev
= info
->xbdev
;
2091 if (xbdev
&& xbdev
->state
== XenbusStateClosing
) {
2092 /* pending switch to state closed */
2093 dev_info(disk_to_dev(bdev
->bd_disk
), "releasing disk\n");
2094 xlvbd_release_gendisk(info
);
2095 xenbus_frontend_closed(info
->xbdev
);
2098 mutex_unlock(&info
->mutex
);
2101 /* sudden device removal */
2102 dev_info(disk_to_dev(bdev
->bd_disk
), "releasing disk\n");
2103 xlvbd_release_gendisk(info
);
2104 disk
->private_data
= NULL
;
2111 mutex_unlock(&blkfront_mutex
);
2114 static const struct block_device_operations xlvbd_block_fops
=
2116 .owner
= THIS_MODULE
,
2118 .release
= blkif_release
,
2119 .getgeo
= blkif_getgeo
,
2120 .ioctl
= blkif_ioctl
,
2124 static const struct xenbus_device_id blkfront_ids
[] = {
2129 static struct xenbus_driver blkfront_driver
= {
2130 .ids
= blkfront_ids
,
2131 .probe
= blkfront_probe
,
2132 .remove
= blkfront_remove
,
2133 .resume
= blkfront_resume
,
2134 .otherend_changed
= blkback_changed
,
2135 .is_ready
= blkfront_is_ready
,
2138 static int __init
xlblk_init(void)
2145 if (xen_blkif_max_ring_order
> XENBUS_MAX_RING_PAGE_ORDER
) {
2146 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2147 xen_blkif_max_ring_order
, XENBUS_MAX_RING_PAGE_ORDER
);
2148 xen_blkif_max_ring_order
= 0;
2151 if (!xen_has_pv_disk_devices())
2154 if (register_blkdev(XENVBD_MAJOR
, DEV_NAME
)) {
2155 printk(KERN_WARNING
"xen_blk: can't get major %d with name %s\n",
2156 XENVBD_MAJOR
, DEV_NAME
);
2160 ret
= xenbus_register_frontend(&blkfront_driver
);
2162 unregister_blkdev(XENVBD_MAJOR
, DEV_NAME
);
2168 module_init(xlblk_init
);
2171 static void __exit
xlblk_exit(void)
2173 xenbus_unregister_driver(&blkfront_driver
);
2174 unregister_blkdev(XENVBD_MAJOR
, DEV_NAME
);
2177 module_exit(xlblk_exit
);
2179 MODULE_DESCRIPTION("Xen virtual block device frontend");
2180 MODULE_LICENSE("GPL");
2181 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR
);
2182 MODULE_ALIAS("xen:vbd");
2183 MODULE_ALIAS("xenblk");