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>
49 #include <linux/workqueue.h>
50 #include <linux/sched/mm.h>
53 #include <xen/xenbus.h>
54 #include <xen/grant_table.h>
55 #include <xen/events.h>
57 #include <xen/platform_pci.h>
59 #include <xen/interface/grant_table.h>
60 #include <xen/interface/io/blkif.h>
61 #include <xen/interface/io/protocols.h>
63 #include <asm/xen/hypervisor.h>
66 * The minimal size of segment supported by the block framework is PAGE_SIZE.
67 * When Linux is using a different page size than Xen, it may not be possible
68 * to put all the data in a single segment.
69 * This can happen when the backend doesn't support indirect descriptor and
70 * therefore the maximum amount of data that a request can carry is
71 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE = 44KB
73 * Note that we only support one extra request. So the Linux page size
74 * should be <= ( 2 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) =
77 #define HAS_EXTRA_REQ (BLKIF_MAX_SEGMENTS_PER_REQUEST < XEN_PFN_PER_PAGE)
80 BLKIF_STATE_DISCONNECTED,
81 BLKIF_STATE_CONNECTED,
82 BLKIF_STATE_SUSPENDED,
88 struct list_head node;
99 struct blkif_request req;
100 struct request *request;
101 struct grant **grants_used;
102 struct grant **indirect_grants;
103 struct scatterlist *sg;
105 enum blk_req_status status;
107 #define NO_ASSOCIATED_ID ~0UL
109 * Id of the sibling if we ever need 2 requests when handling a
112 unsigned long associated_id;
119 static inline struct blkif_req *blkif_req(struct request *rq)
121 return blk_mq_rq_to_pdu(rq);
124 static DEFINE_MUTEX(blkfront_mutex);
125 static const struct block_device_operations xlvbd_block_fops;
126 static struct delayed_work blkfront_work;
127 static LIST_HEAD(info_list);
130 * Maximum number of segments in indirect requests, the actual value used by
131 * the frontend driver is the minimum of this value and the value provided
132 * by the backend driver.
135 static unsigned int xen_blkif_max_segments = 32;
136 module_param_named(max_indirect_segments, xen_blkif_max_segments, uint, 0444);
137 MODULE_PARM_DESC(max_indirect_segments,
138 "Maximum amount of segments in indirect requests (default is 32)");
140 static unsigned int xen_blkif_max_queues = 4;
141 module_param_named(max_queues, xen_blkif_max_queues, uint, 0444);
142 MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
145 * Maximum order of pages to be used for the shared ring between front and
146 * backend, 4KB page granularity is used.
148 static unsigned int xen_blkif_max_ring_order;
149 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, 0444);
150 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
152 #define BLK_RING_SIZE(info) \
153 __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
156 * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
157 * characters are enough. Define to 20 to keep consistent with backend.
159 #define RINGREF_NAME_LEN (20)
161 * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
163 #define QUEUE_NAME_LEN (17)
167 * Every blkfront device can associate with one or more blkfront_ring_info,
168 * depending on how many hardware queues/rings to be used.
170 struct blkfront_ring_info {
171 /* Lock to protect data in every ring buffer. */
172 spinlock_t ring_lock;
173 struct blkif_front_ring ring;
174 unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
175 unsigned int evtchn, irq;
176 struct work_struct work;
177 struct gnttab_free_callback callback;
178 struct list_head indirect_pages;
179 struct list_head grants;
180 unsigned int persistent_gnts_c;
181 unsigned long shadow_free;
182 struct blkfront_info *dev_info;
183 struct blk_shadow shadow[];
187 * We have one of these per vbd, whether ide, scsi or 'other'. They
188 * hang in private_data off the gendisk structure. We may end up
189 * putting all kinds of interesting stuff here :-)
194 struct xenbus_device *xbdev;
197 unsigned int physical_sector_size;
200 enum blkif_state connected;
201 /* Number of pages per ring buffer. */
202 unsigned int nr_ring_pages;
203 struct request_queue *rq;
204 unsigned int feature_flush:1;
205 unsigned int feature_fua:1;
206 unsigned int feature_discard:1;
207 unsigned int feature_secdiscard:1;
208 unsigned int feature_persistent:1;
209 unsigned int discard_granularity;
210 unsigned int discard_alignment;
211 /* Number of 4KB segments handled */
212 unsigned int max_indirect_segments;
214 struct blk_mq_tag_set tag_set;
215 struct blkfront_ring_info *rinfo;
216 unsigned int nr_rings;
217 unsigned int rinfo_size;
218 /* Save uncomplete reqs and bios for migration. */
219 struct list_head requests;
220 struct bio_list bio_list;
221 struct list_head info_list;
224 static unsigned int nr_minors;
225 static unsigned long *minors;
226 static DEFINE_SPINLOCK(minor_lock);
228 #define GRANT_INVALID_REF 0
230 #define PARTS_PER_DISK 16
231 #define PARTS_PER_EXT_DISK 256
233 #define BLKIF_MAJOR(dev) ((dev)>>8)
234 #define BLKIF_MINOR(dev) ((dev) & 0xff)
237 #define EXTENDED (1<<EXT_SHIFT)
238 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
239 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
240 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
241 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
242 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
243 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
245 #define DEV_NAME "xvd" /* name in /dev */
248 * Grants are always the same size as a Xen page (i.e 4KB).
249 * A physical segment is always the same size as a Linux page.
250 * Number of grants per physical segment
252 #define GRANTS_PER_PSEG (PAGE_SIZE / XEN_PAGE_SIZE)
254 #define GRANTS_PER_INDIRECT_FRAME \
255 (XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
257 #define INDIRECT_GREFS(_grants) \
258 DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
260 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
261 static void blkfront_gather_backend_features(struct blkfront_info *info);
262 static int negotiate_mq(struct blkfront_info *info);
264 #define for_each_rinfo(info, ptr, idx) \
265 for ((ptr) = (info)->rinfo, (idx) = 0; \
266 (idx) < (info)->nr_rings; \
267 (idx)++, (ptr) = (void *)(ptr) + (info)->rinfo_size)
269 static inline struct blkfront_ring_info *
270 get_rinfo(const struct blkfront_info *info, unsigned int i)
272 BUG_ON(i >= info->nr_rings);
273 return (void *)info->rinfo + i * info->rinfo_size;
276 static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
278 unsigned long free = rinfo->shadow_free;
280 BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
281 rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
282 rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
286 static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
289 if (rinfo->shadow[id].req.u.rw.id != id)
291 if (rinfo->shadow[id].request == NULL)
293 rinfo->shadow[id].req.u.rw.id = rinfo->shadow_free;
294 rinfo->shadow[id].request = NULL;
295 rinfo->shadow_free = id;
299 static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
301 struct blkfront_info *info = rinfo->dev_info;
302 struct page *granted_page;
303 struct grant *gnt_list_entry, *n;
307 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
311 if (info->feature_persistent) {
312 granted_page = alloc_page(GFP_NOIO);
314 kfree(gnt_list_entry);
317 gnt_list_entry->page = granted_page;
320 gnt_list_entry->gref = GRANT_INVALID_REF;
321 list_add(&gnt_list_entry->node, &rinfo->grants);
328 list_for_each_entry_safe(gnt_list_entry, n,
329 &rinfo->grants, node) {
330 list_del(&gnt_list_entry->node);
331 if (info->feature_persistent)
332 __free_page(gnt_list_entry->page);
333 kfree(gnt_list_entry);
340 static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
342 struct grant *gnt_list_entry;
344 BUG_ON(list_empty(&rinfo->grants));
345 gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
347 list_del(&gnt_list_entry->node);
349 if (gnt_list_entry->gref != GRANT_INVALID_REF)
350 rinfo->persistent_gnts_c--;
352 return gnt_list_entry;
355 static inline void grant_foreign_access(const struct grant *gnt_list_entry,
356 const struct blkfront_info *info)
358 gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
359 info->xbdev->otherend_id,
360 gnt_list_entry->page,
364 static struct grant *get_grant(grant_ref_t *gref_head,
366 struct blkfront_ring_info *rinfo)
368 struct grant *gnt_list_entry = get_free_grant(rinfo);
369 struct blkfront_info *info = rinfo->dev_info;
371 if (gnt_list_entry->gref != GRANT_INVALID_REF)
372 return gnt_list_entry;
374 /* Assign a gref to this page */
375 gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
376 BUG_ON(gnt_list_entry->gref == -ENOSPC);
377 if (info->feature_persistent)
378 grant_foreign_access(gnt_list_entry, info);
380 /* Grant access to the GFN passed by the caller */
381 gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
382 info->xbdev->otherend_id,
386 return gnt_list_entry;
389 static struct grant *get_indirect_grant(grant_ref_t *gref_head,
390 struct blkfront_ring_info *rinfo)
392 struct grant *gnt_list_entry = get_free_grant(rinfo);
393 struct blkfront_info *info = rinfo->dev_info;
395 if (gnt_list_entry->gref != GRANT_INVALID_REF)
396 return gnt_list_entry;
398 /* Assign a gref to this page */
399 gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
400 BUG_ON(gnt_list_entry->gref == -ENOSPC);
401 if (!info->feature_persistent) {
402 struct page *indirect_page;
404 /* Fetch a pre-allocated page to use for indirect grefs */
405 BUG_ON(list_empty(&rinfo->indirect_pages));
406 indirect_page = list_first_entry(&rinfo->indirect_pages,
408 list_del(&indirect_page->lru);
409 gnt_list_entry->page = indirect_page;
411 grant_foreign_access(gnt_list_entry, info);
413 return gnt_list_entry;
416 static const char *op_name(int op)
418 static const char *const names[] = {
419 [BLKIF_OP_READ] = "read",
420 [BLKIF_OP_WRITE] = "write",
421 [BLKIF_OP_WRITE_BARRIER] = "barrier",
422 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
423 [BLKIF_OP_DISCARD] = "discard" };
425 if (op < 0 || op >= ARRAY_SIZE(names))
433 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
435 unsigned int end = minor + nr;
438 if (end > nr_minors) {
439 unsigned long *bitmap, *old;
441 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
446 spin_lock(&minor_lock);
447 if (end > nr_minors) {
449 memcpy(bitmap, minors,
450 BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
452 nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
455 spin_unlock(&minor_lock);
459 spin_lock(&minor_lock);
460 if (find_next_bit(minors, end, minor) >= end) {
461 bitmap_set(minors, minor, nr);
465 spin_unlock(&minor_lock);
470 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
472 unsigned int end = minor + nr;
474 BUG_ON(end > nr_minors);
475 spin_lock(&minor_lock);
476 bitmap_clear(minors, minor, nr);
477 spin_unlock(&minor_lock);
480 static void blkif_restart_queue_callback(void *arg)
482 struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
483 schedule_work(&rinfo->work);
486 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
488 /* We don't have real geometry info, but let's at least return
489 values consistent with the size of the device */
490 sector_t nsect = get_capacity(bd->bd_disk);
491 sector_t cylinders = nsect;
495 sector_div(cylinders, hg->heads * hg->sectors);
496 hg->cylinders = cylinders;
497 if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
498 hg->cylinders = 0xffff;
502 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
503 unsigned command, unsigned long argument)
508 case CDROMMULTISESSION:
509 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
510 if (put_user(0, (char __user *)(argument + i)))
513 case CDROM_GET_CAPABILITY:
514 if (bdev->bd_disk->flags & GENHD_FL_CD)
522 static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
524 struct blkif_request **ring_req)
528 *ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
529 rinfo->ring.req_prod_pvt++;
531 id = get_id_from_freelist(rinfo);
532 rinfo->shadow[id].request = req;
533 rinfo->shadow[id].status = REQ_WAITING;
534 rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
536 (*ring_req)->u.rw.id = id;
541 static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
543 struct blkfront_info *info = rinfo->dev_info;
544 struct blkif_request *ring_req;
547 /* Fill out a communications ring structure. */
548 id = blkif_ring_get_request(rinfo, req, &ring_req);
550 ring_req->operation = BLKIF_OP_DISCARD;
551 ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
552 ring_req->u.discard.id = id;
553 ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
554 if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard)
555 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
557 ring_req->u.discard.flag = 0;
559 /* Keep a private copy so we can reissue requests when recovering. */
560 rinfo->shadow[id].req = *ring_req;
565 struct setup_rw_req {
566 unsigned int grant_idx;
567 struct blkif_request_segment *segments;
568 struct blkfront_ring_info *rinfo;
569 struct blkif_request *ring_req;
570 grant_ref_t gref_head;
572 /* Only used when persistent grant is used and it's a read request */
574 unsigned int bvec_off;
577 bool require_extra_req;
578 struct blkif_request *extra_ring_req;
581 static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
582 unsigned int len, void *data)
584 struct setup_rw_req *setup = data;
586 struct grant *gnt_list_entry;
587 unsigned int fsect, lsect;
588 /* Convenient aliases */
589 unsigned int grant_idx = setup->grant_idx;
590 struct blkif_request *ring_req = setup->ring_req;
591 struct blkfront_ring_info *rinfo = setup->rinfo;
593 * We always use the shadow of the first request to store the list
594 * of grant associated to the block I/O request. This made the
595 * completion more easy to handle even if the block I/O request is
598 struct blk_shadow *shadow = &rinfo->shadow[setup->id];
600 if (unlikely(setup->require_extra_req &&
601 grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
603 * We are using the second request, setup grant_idx
604 * to be the index of the segment array.
606 grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
607 ring_req = setup->extra_ring_req;
610 if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
611 (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
613 kunmap_atomic(setup->segments);
615 n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
616 gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
617 shadow->indirect_grants[n] = gnt_list_entry;
618 setup->segments = kmap_atomic(gnt_list_entry->page);
619 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
622 gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
623 ref = gnt_list_entry->gref;
625 * All the grants are stored in the shadow of the first
626 * request. Therefore we have to use the global index.
628 shadow->grants_used[setup->grant_idx] = gnt_list_entry;
630 if (setup->need_copy) {
633 shared_data = kmap_atomic(gnt_list_entry->page);
635 * this does not wipe data stored outside the
636 * range sg->offset..sg->offset+sg->length.
637 * Therefore, blkback *could* see data from
638 * previous requests. This is OK as long as
639 * persistent grants are shared with just one
640 * domain. It may need refactoring if this
643 memcpy(shared_data + offset,
644 setup->bvec_data + setup->bvec_off,
647 kunmap_atomic(shared_data);
648 setup->bvec_off += len;
652 lsect = fsect + (len >> 9) - 1;
653 if (ring_req->operation != BLKIF_OP_INDIRECT) {
654 ring_req->u.rw.seg[grant_idx] =
655 (struct blkif_request_segment) {
658 .last_sect = lsect };
660 setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
661 (struct blkif_request_segment) {
664 .last_sect = lsect };
667 (setup->grant_idx)++;
670 static void blkif_setup_extra_req(struct blkif_request *first,
671 struct blkif_request *second)
673 uint16_t nr_segments = first->u.rw.nr_segments;
676 * The second request is only present when the first request uses
677 * all its segments. It's always the continuity of the first one.
679 first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
681 second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
682 second->u.rw.sector_number = first->u.rw.sector_number +
683 (BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
685 second->u.rw.handle = first->u.rw.handle;
686 second->operation = first->operation;
689 static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
691 struct blkfront_info *info = rinfo->dev_info;
692 struct blkif_request *ring_req, *extra_ring_req = NULL;
693 unsigned long id, extra_id = NO_ASSOCIATED_ID;
694 bool require_extra_req = false;
696 struct setup_rw_req setup = {
700 .need_copy = rq_data_dir(req) && info->feature_persistent,
704 * Used to store if we are able to queue the request by just using
705 * existing persistent grants, or if we have to get new grants,
706 * as there are not sufficiently many free.
708 bool new_persistent_gnts = false;
709 struct scatterlist *sg;
710 int num_sg, max_grefs, num_grant;
712 max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
713 if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
715 * If we are using indirect segments we need to account
716 * for the indirect grefs used in the request.
718 max_grefs += INDIRECT_GREFS(max_grefs);
720 /* Check if we have enough persistent grants to allocate a requests */
721 if (rinfo->persistent_gnts_c < max_grefs) {
722 new_persistent_gnts = true;
724 if (gnttab_alloc_grant_references(
725 max_grefs - rinfo->persistent_gnts_c,
726 &setup.gref_head) < 0) {
727 gnttab_request_free_callback(
729 blkif_restart_queue_callback,
731 max_grefs - rinfo->persistent_gnts_c);
736 /* Fill out a communications ring structure. */
737 id = blkif_ring_get_request(rinfo, req, &ring_req);
739 num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
741 /* Calculate the number of grant used */
742 for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
743 num_grant += gnttab_count_grant(sg->offset, sg->length);
745 require_extra_req = info->max_indirect_segments == 0 &&
746 num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
747 BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
749 rinfo->shadow[id].num_sg = num_sg;
750 if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
751 likely(!require_extra_req)) {
753 * The indirect operation can only be a BLKIF_OP_READ or
756 BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA);
757 ring_req->operation = BLKIF_OP_INDIRECT;
758 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
759 BLKIF_OP_WRITE : BLKIF_OP_READ;
760 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
761 ring_req->u.indirect.handle = info->handle;
762 ring_req->u.indirect.nr_segments = num_grant;
764 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
765 ring_req->u.rw.handle = info->handle;
766 ring_req->operation = rq_data_dir(req) ?
767 BLKIF_OP_WRITE : BLKIF_OP_READ;
768 if (req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA) {
770 * Ideally we can do an unordered flush-to-disk.
771 * In case the backend onlysupports barriers, use that.
772 * A barrier request a superset of FUA, so we can
773 * implement it the same way. (It's also a FLUSH+FUA,
774 * since it is guaranteed ordered WRT previous writes.)
776 if (info->feature_flush && info->feature_fua)
777 ring_req->operation =
778 BLKIF_OP_WRITE_BARRIER;
779 else if (info->feature_flush)
780 ring_req->operation =
781 BLKIF_OP_FLUSH_DISKCACHE;
783 ring_req->operation = 0;
785 ring_req->u.rw.nr_segments = num_grant;
786 if (unlikely(require_extra_req)) {
787 extra_id = blkif_ring_get_request(rinfo, req,
790 * Only the first request contains the scatter-gather
793 rinfo->shadow[extra_id].num_sg = 0;
795 blkif_setup_extra_req(ring_req, extra_ring_req);
797 /* Link the 2 requests together */
798 rinfo->shadow[extra_id].associated_id = id;
799 rinfo->shadow[id].associated_id = extra_id;
803 setup.ring_req = ring_req;
806 setup.require_extra_req = require_extra_req;
807 if (unlikely(require_extra_req))
808 setup.extra_ring_req = extra_ring_req;
810 for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
811 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
813 if (setup.need_copy) {
814 setup.bvec_off = sg->offset;
815 setup.bvec_data = kmap_atomic(sg_page(sg));
818 gnttab_foreach_grant_in_range(sg_page(sg),
821 blkif_setup_rw_req_grant,
825 kunmap_atomic(setup.bvec_data);
828 kunmap_atomic(setup.segments);
830 /* Keep a private copy so we can reissue requests when recovering. */
831 rinfo->shadow[id].req = *ring_req;
832 if (unlikely(require_extra_req))
833 rinfo->shadow[extra_id].req = *extra_ring_req;
835 if (new_persistent_gnts)
836 gnttab_free_grant_references(setup.gref_head);
842 * Generate a Xen blkfront IO request from a blk layer request. Reads
843 * and writes are handled as expected.
845 * @req: a request struct
847 static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
849 if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
852 if (unlikely(req_op(req) == REQ_OP_DISCARD ||
853 req_op(req) == REQ_OP_SECURE_ERASE))
854 return blkif_queue_discard_req(req, rinfo);
856 return blkif_queue_rw_req(req, rinfo);
859 static inline void flush_requests(struct blkfront_ring_info *rinfo)
863 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
866 notify_remote_via_irq(rinfo->irq);
869 static inline bool blkif_request_flush_invalid(struct request *req,
870 struct blkfront_info *info)
872 return (blk_rq_is_passthrough(req) ||
873 ((req_op(req) == REQ_OP_FLUSH) &&
874 !info->feature_flush) ||
875 ((req->cmd_flags & REQ_FUA) &&
876 !info->feature_fua));
879 static blk_status_t blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
880 const struct blk_mq_queue_data *qd)
883 int qid = hctx->queue_num;
884 struct blkfront_info *info = hctx->queue->queuedata;
885 struct blkfront_ring_info *rinfo = NULL;
887 rinfo = get_rinfo(info, qid);
888 blk_mq_start_request(qd->rq);
889 spin_lock_irqsave(&rinfo->ring_lock, flags);
890 if (RING_FULL(&rinfo->ring))
893 if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
896 if (blkif_queue_request(qd->rq, rinfo))
899 flush_requests(rinfo);
900 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
904 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
905 return BLK_STS_IOERR;
908 blk_mq_stop_hw_queue(hctx);
909 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
910 return BLK_STS_DEV_RESOURCE;
913 static void blkif_complete_rq(struct request *rq)
915 blk_mq_end_request(rq, blkif_req(rq)->error);
918 static const struct blk_mq_ops blkfront_mq_ops = {
919 .queue_rq = blkif_queue_rq,
920 .complete = blkif_complete_rq,
923 static void blkif_set_queue_limits(struct blkfront_info *info)
925 struct request_queue *rq = info->rq;
926 struct gendisk *gd = info->gd;
927 unsigned int segments = info->max_indirect_segments ? :
928 BLKIF_MAX_SEGMENTS_PER_REQUEST;
930 blk_queue_flag_set(QUEUE_FLAG_VIRT, rq);
932 if (info->feature_discard) {
933 blk_queue_flag_set(QUEUE_FLAG_DISCARD, rq);
934 blk_queue_max_discard_sectors(rq, get_capacity(gd));
935 rq->limits.discard_granularity = info->discard_granularity ?:
936 info->physical_sector_size;
937 rq->limits.discard_alignment = info->discard_alignment;
938 if (info->feature_secdiscard)
939 blk_queue_flag_set(QUEUE_FLAG_SECERASE, rq);
942 /* Hard sector size and max sectors impersonate the equiv. hardware. */
943 blk_queue_logical_block_size(rq, info->sector_size);
944 blk_queue_physical_block_size(rq, info->physical_sector_size);
945 blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
947 /* Each segment in a request is up to an aligned page in size. */
948 blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
949 blk_queue_max_segment_size(rq, PAGE_SIZE);
951 /* Ensure a merged request will fit in a single I/O ring slot. */
952 blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
954 /* Make sure buffer addresses are sector-aligned. */
955 blk_queue_dma_alignment(rq, 511);
958 static const char *flush_info(struct blkfront_info *info)
960 if (info->feature_flush && info->feature_fua)
961 return "barrier: enabled;";
962 else if (info->feature_flush)
963 return "flush diskcache: enabled;";
965 return "barrier or flush: disabled;";
968 static void xlvbd_flush(struct blkfront_info *info)
970 blk_queue_write_cache(info->rq, info->feature_flush ? true : false,
971 info->feature_fua ? true : false);
972 pr_info("blkfront: %s: %s %s %s %s %s\n",
973 info->gd->disk_name, flush_info(info),
974 "persistent grants:", info->feature_persistent ?
975 "enabled;" : "disabled;", "indirect descriptors:",
976 info->max_indirect_segments ? "enabled;" : "disabled;");
979 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
982 major = BLKIF_MAJOR(vdevice);
983 *minor = BLKIF_MINOR(vdevice);
986 *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
987 *minor = ((*minor / 64) * PARTS_PER_DISK) +
988 EMULATED_HD_DISK_MINOR_OFFSET;
991 *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
992 *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
993 EMULATED_HD_DISK_MINOR_OFFSET;
995 case XEN_SCSI_DISK0_MAJOR:
996 *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
997 *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
999 case XEN_SCSI_DISK1_MAJOR:
1000 case XEN_SCSI_DISK2_MAJOR:
1001 case XEN_SCSI_DISK3_MAJOR:
1002 case XEN_SCSI_DISK4_MAJOR:
1003 case XEN_SCSI_DISK5_MAJOR:
1004 case XEN_SCSI_DISK6_MAJOR:
1005 case XEN_SCSI_DISK7_MAJOR:
1006 *offset = (*minor / PARTS_PER_DISK) +
1007 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1008 EMULATED_SD_DISK_NAME_OFFSET;
1010 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1011 EMULATED_SD_DISK_MINOR_OFFSET;
1013 case XEN_SCSI_DISK8_MAJOR:
1014 case XEN_SCSI_DISK9_MAJOR:
1015 case XEN_SCSI_DISK10_MAJOR:
1016 case XEN_SCSI_DISK11_MAJOR:
1017 case XEN_SCSI_DISK12_MAJOR:
1018 case XEN_SCSI_DISK13_MAJOR:
1019 case XEN_SCSI_DISK14_MAJOR:
1020 case XEN_SCSI_DISK15_MAJOR:
1021 *offset = (*minor / PARTS_PER_DISK) +
1022 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1023 EMULATED_SD_DISK_NAME_OFFSET;
1025 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1026 EMULATED_SD_DISK_MINOR_OFFSET;
1029 *offset = *minor / PARTS_PER_DISK;
1032 printk(KERN_WARNING "blkfront: your disk configuration is "
1033 "incorrect, please use an xvd device instead\n");
1039 static char *encode_disk_name(char *ptr, unsigned int n)
1042 ptr = encode_disk_name(ptr, n / 26 - 1);
1043 *ptr = 'a' + n % 26;
1047 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1048 struct blkfront_info *info,
1049 u16 vdisk_info, u16 sector_size,
1050 unsigned int physical_sector_size)
1055 unsigned int offset;
1060 BUG_ON(info->gd != NULL);
1061 BUG_ON(info->rq != NULL);
1063 if ((info->vdevice>>EXT_SHIFT) > 1) {
1064 /* this is above the extended range; something is wrong */
1065 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1069 if (!VDEV_IS_EXTENDED(info->vdevice)) {
1070 err = xen_translate_vdev(info->vdevice, &minor, &offset);
1073 nr_parts = PARTS_PER_DISK;
1075 minor = BLKIF_MINOR_EXT(info->vdevice);
1076 nr_parts = PARTS_PER_EXT_DISK;
1077 offset = minor / nr_parts;
1078 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1079 printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1080 "emulated IDE disks,\n\t choose an xvd device name"
1081 "from xvde on\n", info->vdevice);
1083 if (minor >> MINORBITS) {
1084 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1085 info->vdevice, minor);
1089 if ((minor % nr_parts) == 0)
1090 nr_minors = nr_parts;
1092 err = xlbd_reserve_minors(minor, nr_minors);
1096 memset(&info->tag_set, 0, sizeof(info->tag_set));
1097 info->tag_set.ops = &blkfront_mq_ops;
1098 info->tag_set.nr_hw_queues = info->nr_rings;
1099 if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
1101 * When indirect descriptior is not supported, the I/O request
1102 * will be split between multiple request in the ring.
1103 * To avoid problems when sending the request, divide by
1104 * 2 the depth of the queue.
1106 info->tag_set.queue_depth = BLK_RING_SIZE(info) / 2;
1108 info->tag_set.queue_depth = BLK_RING_SIZE(info);
1109 info->tag_set.numa_node = NUMA_NO_NODE;
1110 info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
1111 info->tag_set.cmd_size = sizeof(struct blkif_req);
1112 info->tag_set.driver_data = info;
1114 err = blk_mq_alloc_tag_set(&info->tag_set);
1116 goto out_release_minors;
1118 gd = blk_mq_alloc_disk(&info->tag_set, info);
1121 goto out_free_tag_set;
1124 strcpy(gd->disk_name, DEV_NAME);
1125 ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1126 BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1130 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1131 "%d", minor & (nr_parts - 1));
1133 gd->major = XENVBD_MAJOR;
1134 gd->first_minor = minor;
1135 gd->minors = nr_minors;
1136 gd->fops = &xlvbd_block_fops;
1137 gd->private_data = info;
1138 set_capacity(gd, capacity);
1140 info->rq = gd->queue;
1142 info->sector_size = sector_size;
1143 info->physical_sector_size = physical_sector_size;
1144 blkif_set_queue_limits(info);
1148 if (vdisk_info & VDISK_READONLY)
1151 if (vdisk_info & VDISK_REMOVABLE)
1152 gd->flags |= GENHD_FL_REMOVABLE;
1154 if (vdisk_info & VDISK_CDROM)
1155 gd->flags |= GENHD_FL_CD;
1160 blk_mq_free_tag_set(&info->tag_set);
1162 xlbd_release_minors(minor, nr_minors);
1166 /* Already hold rinfo->ring_lock. */
1167 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1169 if (!RING_FULL(&rinfo->ring))
1170 blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1173 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1175 unsigned long flags;
1177 spin_lock_irqsave(&rinfo->ring_lock, flags);
1178 kick_pending_request_queues_locked(rinfo);
1179 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1182 static void blkif_restart_queue(struct work_struct *work)
1184 struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1186 if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1187 kick_pending_request_queues(rinfo);
1190 static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1192 struct grant *persistent_gnt, *n;
1193 struct blkfront_info *info = rinfo->dev_info;
1197 * Remove indirect pages, this only happens when using indirect
1198 * descriptors but not persistent grants
1200 if (!list_empty(&rinfo->indirect_pages)) {
1201 struct page *indirect_page, *n;
1203 BUG_ON(info->feature_persistent);
1204 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1205 list_del(&indirect_page->lru);
1206 __free_page(indirect_page);
1210 /* Remove all persistent grants. */
1211 if (!list_empty(&rinfo->grants)) {
1212 list_for_each_entry_safe(persistent_gnt, n,
1213 &rinfo->grants, node) {
1214 list_del(&persistent_gnt->node);
1215 if (persistent_gnt->gref != GRANT_INVALID_REF) {
1216 gnttab_end_foreign_access(persistent_gnt->gref,
1218 rinfo->persistent_gnts_c--;
1220 if (info->feature_persistent)
1221 __free_page(persistent_gnt->page);
1222 kfree(persistent_gnt);
1225 BUG_ON(rinfo->persistent_gnts_c != 0);
1227 for (i = 0; i < BLK_RING_SIZE(info); i++) {
1229 * Clear persistent grants present in requests already
1230 * on the shared ring
1232 if (!rinfo->shadow[i].request)
1235 segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1236 rinfo->shadow[i].req.u.indirect.nr_segments :
1237 rinfo->shadow[i].req.u.rw.nr_segments;
1238 for (j = 0; j < segs; j++) {
1239 persistent_gnt = rinfo->shadow[i].grants_used[j];
1240 gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1241 if (info->feature_persistent)
1242 __free_page(persistent_gnt->page);
1243 kfree(persistent_gnt);
1246 if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1248 * If this is not an indirect operation don't try to
1249 * free indirect segments
1253 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1254 persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1255 gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1256 __free_page(persistent_gnt->page);
1257 kfree(persistent_gnt);
1261 kvfree(rinfo->shadow[i].grants_used);
1262 rinfo->shadow[i].grants_used = NULL;
1263 kvfree(rinfo->shadow[i].indirect_grants);
1264 rinfo->shadow[i].indirect_grants = NULL;
1265 kvfree(rinfo->shadow[i].sg);
1266 rinfo->shadow[i].sg = NULL;
1269 /* No more gnttab callback work. */
1270 gnttab_cancel_free_callback(&rinfo->callback);
1272 /* Flush gnttab callback work. Must be done with no locks held. */
1273 flush_work(&rinfo->work);
1275 /* Free resources associated with old device channel. */
1276 for (i = 0; i < info->nr_ring_pages; i++) {
1277 if (rinfo->ring_ref[i] != GRANT_INVALID_REF) {
1278 gnttab_end_foreign_access(rinfo->ring_ref[i], 0, 0);
1279 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1282 free_pages((unsigned long)rinfo->ring.sring, get_order(info->nr_ring_pages * XEN_PAGE_SIZE));
1283 rinfo->ring.sring = NULL;
1286 unbind_from_irqhandler(rinfo->irq, rinfo);
1287 rinfo->evtchn = rinfo->irq = 0;
1290 static void blkif_free(struct blkfront_info *info, int suspend)
1293 struct blkfront_ring_info *rinfo;
1295 /* Prevent new requests being issued until we fix things up. */
1296 info->connected = suspend ?
1297 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1298 /* No more blkif_request(). */
1300 blk_mq_stop_hw_queues(info->rq);
1302 for_each_rinfo(info, rinfo, i)
1303 blkif_free_ring(rinfo);
1305 kvfree(info->rinfo);
1310 struct copy_from_grant {
1311 const struct blk_shadow *s;
1312 unsigned int grant_idx;
1313 unsigned int bvec_offset;
1317 static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1318 unsigned int len, void *data)
1320 struct copy_from_grant *info = data;
1322 /* Convenient aliases */
1323 const struct blk_shadow *s = info->s;
1325 shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1327 memcpy(info->bvec_data + info->bvec_offset,
1328 shared_data + offset, len);
1330 info->bvec_offset += len;
1333 kunmap_atomic(shared_data);
1336 static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1340 case BLKIF_RSP_OKAY:
1342 case BLKIF_RSP_EOPNOTSUPP:
1343 return REQ_EOPNOTSUPP;
1344 case BLKIF_RSP_ERROR:
1351 * Get the final status of the block request based on two ring response
1353 static int blkif_get_final_status(enum blk_req_status s1,
1354 enum blk_req_status s2)
1356 BUG_ON(s1 == REQ_WAITING);
1357 BUG_ON(s2 == REQ_WAITING);
1359 if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1360 return BLKIF_RSP_ERROR;
1361 else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1362 return BLKIF_RSP_EOPNOTSUPP;
1363 return BLKIF_RSP_OKAY;
1366 static bool blkif_completion(unsigned long *id,
1367 struct blkfront_ring_info *rinfo,
1368 struct blkif_response *bret)
1371 struct scatterlist *sg;
1372 int num_sg, num_grant;
1373 struct blkfront_info *info = rinfo->dev_info;
1374 struct blk_shadow *s = &rinfo->shadow[*id];
1375 struct copy_from_grant data = {
1379 num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1380 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1382 /* The I/O request may be split in two. */
1383 if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1384 struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1386 /* Keep the status of the current response in shadow. */
1387 s->status = blkif_rsp_to_req_status(bret->status);
1389 /* Wait the second response if not yet here. */
1390 if (s2->status == REQ_WAITING)
1393 bret->status = blkif_get_final_status(s->status,
1397 * All the grants is stored in the first shadow in order
1398 * to make the completion code simpler.
1400 num_grant += s2->req.u.rw.nr_segments;
1403 * The two responses may not come in order. Only the
1404 * first request will store the scatter-gather list.
1406 if (s2->num_sg != 0) {
1407 /* Update "id" with the ID of the first response. */
1408 *id = s->associated_id;
1413 * We don't need anymore the second request, so recycling
1416 if (add_id_to_freelist(rinfo, s->associated_id))
1417 WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1418 info->gd->disk_name, s->associated_id);
1424 if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1425 for_each_sg(s->sg, sg, num_sg, i) {
1426 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1428 data.bvec_offset = sg->offset;
1429 data.bvec_data = kmap_atomic(sg_page(sg));
1431 gnttab_foreach_grant_in_range(sg_page(sg),
1434 blkif_copy_from_grant,
1437 kunmap_atomic(data.bvec_data);
1440 /* Add the persistent grant into the list of free grants */
1441 for (i = 0; i < num_grant; i++) {
1442 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1444 * If the grant is still mapped by the backend (the
1445 * backend has chosen to make this grant persistent)
1446 * we add it at the head of the list, so it will be
1449 if (!info->feature_persistent)
1450 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1451 s->grants_used[i]->gref);
1452 list_add(&s->grants_used[i]->node, &rinfo->grants);
1453 rinfo->persistent_gnts_c++;
1456 * If the grant is not mapped by the backend we end the
1457 * foreign access and add it to the tail of the list,
1458 * so it will not be picked again unless we run out of
1459 * persistent grants.
1461 gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1462 s->grants_used[i]->gref = GRANT_INVALID_REF;
1463 list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1466 if (s->req.operation == BLKIF_OP_INDIRECT) {
1467 for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1468 if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1469 if (!info->feature_persistent)
1470 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1471 s->indirect_grants[i]->gref);
1472 list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1473 rinfo->persistent_gnts_c++;
1475 struct page *indirect_page;
1477 gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1479 * Add the used indirect page back to the list of
1480 * available pages for indirect grefs.
1482 if (!info->feature_persistent) {
1483 indirect_page = s->indirect_grants[i]->page;
1484 list_add(&indirect_page->lru, &rinfo->indirect_pages);
1486 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1487 list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1495 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1497 struct request *req;
1498 struct blkif_response *bret;
1500 unsigned long flags;
1501 struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1502 struct blkfront_info *info = rinfo->dev_info;
1504 if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
1507 spin_lock_irqsave(&rinfo->ring_lock, flags);
1509 rp = rinfo->ring.sring->rsp_prod;
1510 rmb(); /* Ensure we see queued responses up to 'rp'. */
1512 for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1515 bret = RING_GET_RESPONSE(&rinfo->ring, i);
1518 * The backend has messed up and given us an id that we would
1519 * never have given to it (we stamp it up to BLK_RING_SIZE -
1520 * look in get_id_from_freelist.
1522 if (id >= BLK_RING_SIZE(info)) {
1523 WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1524 info->gd->disk_name, op_name(bret->operation), id);
1525 /* We can't safely get the 'struct request' as
1526 * the id is busted. */
1529 req = rinfo->shadow[id].request;
1531 if (bret->operation != BLKIF_OP_DISCARD) {
1533 * We may need to wait for an extra response if the
1534 * I/O request is split in 2
1536 if (!blkif_completion(&id, rinfo, bret))
1540 if (add_id_to_freelist(rinfo, id)) {
1541 WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1542 info->gd->disk_name, op_name(bret->operation), id);
1546 if (bret->status == BLKIF_RSP_OKAY)
1547 blkif_req(req)->error = BLK_STS_OK;
1549 blkif_req(req)->error = BLK_STS_IOERR;
1551 switch (bret->operation) {
1552 case BLKIF_OP_DISCARD:
1553 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1554 struct request_queue *rq = info->rq;
1555 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1556 info->gd->disk_name, op_name(bret->operation));
1557 blkif_req(req)->error = BLK_STS_NOTSUPP;
1558 info->feature_discard = 0;
1559 info->feature_secdiscard = 0;
1560 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1561 blk_queue_flag_clear(QUEUE_FLAG_SECERASE, rq);
1564 case BLKIF_OP_FLUSH_DISKCACHE:
1565 case BLKIF_OP_WRITE_BARRIER:
1566 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1567 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1568 info->gd->disk_name, op_name(bret->operation));
1569 blkif_req(req)->error = BLK_STS_NOTSUPP;
1571 if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1572 rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1573 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1574 info->gd->disk_name, op_name(bret->operation));
1575 blkif_req(req)->error = BLK_STS_NOTSUPP;
1577 if (unlikely(blkif_req(req)->error)) {
1578 if (blkif_req(req)->error == BLK_STS_NOTSUPP)
1579 blkif_req(req)->error = BLK_STS_OK;
1580 info->feature_fua = 0;
1581 info->feature_flush = 0;
1586 case BLKIF_OP_WRITE:
1587 if (unlikely(bret->status != BLKIF_RSP_OKAY))
1588 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1589 "request: %x\n", bret->status);
1596 if (likely(!blk_should_fake_timeout(req->q)))
1597 blk_mq_complete_request(req);
1600 rinfo->ring.rsp_cons = i;
1602 if (i != rinfo->ring.req_prod_pvt) {
1604 RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1608 rinfo->ring.sring->rsp_event = i + 1;
1610 kick_pending_request_queues_locked(rinfo);
1612 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1618 static int setup_blkring(struct xenbus_device *dev,
1619 struct blkfront_ring_info *rinfo)
1621 struct blkif_sring *sring;
1623 struct blkfront_info *info = rinfo->dev_info;
1624 unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1625 grant_ref_t gref[XENBUS_MAX_RING_GRANTS];
1627 for (i = 0; i < info->nr_ring_pages; i++)
1628 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1630 sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
1631 get_order(ring_size));
1633 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1636 SHARED_RING_INIT(sring);
1637 FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1639 err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref);
1641 free_pages((unsigned long)sring, get_order(ring_size));
1642 rinfo->ring.sring = NULL;
1645 for (i = 0; i < info->nr_ring_pages; i++)
1646 rinfo->ring_ref[i] = gref[i];
1648 err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1652 err = bind_evtchn_to_irqhandler(rinfo->evtchn, blkif_interrupt, 0,
1655 xenbus_dev_fatal(dev, err,
1656 "bind_evtchn_to_irqhandler failed");
1663 blkif_free(info, 0);
1668 * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1669 * ring buffer may have multi pages depending on ->nr_ring_pages.
1671 static int write_per_ring_nodes(struct xenbus_transaction xbt,
1672 struct blkfront_ring_info *rinfo, const char *dir)
1676 const char *message = NULL;
1677 struct blkfront_info *info = rinfo->dev_info;
1679 if (info->nr_ring_pages == 1) {
1680 err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1682 message = "writing ring-ref";
1683 goto abort_transaction;
1686 for (i = 0; i < info->nr_ring_pages; i++) {
1687 char ring_ref_name[RINGREF_NAME_LEN];
1689 snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1690 err = xenbus_printf(xbt, dir, ring_ref_name,
1691 "%u", rinfo->ring_ref[i]);
1693 message = "writing ring-ref";
1694 goto abort_transaction;
1699 err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1701 message = "writing event-channel";
1702 goto abort_transaction;
1708 xenbus_transaction_end(xbt, 1);
1710 xenbus_dev_fatal(info->xbdev, err, "%s", message);
1715 /* Common code used when first setting up, and when resuming. */
1716 static int talk_to_blkback(struct xenbus_device *dev,
1717 struct blkfront_info *info)
1719 const char *message = NULL;
1720 struct xenbus_transaction xbt;
1722 unsigned int i, max_page_order;
1723 unsigned int ring_page_order;
1724 struct blkfront_ring_info *rinfo;
1729 max_page_order = xenbus_read_unsigned(info->xbdev->otherend,
1730 "max-ring-page-order", 0);
1731 ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1732 info->nr_ring_pages = 1 << ring_page_order;
1734 err = negotiate_mq(info);
1736 goto destroy_blkring;
1738 for_each_rinfo(info, rinfo, i) {
1739 /* Create shared ring, alloc event channel. */
1740 err = setup_blkring(dev, rinfo);
1742 goto destroy_blkring;
1746 err = xenbus_transaction_start(&xbt);
1748 xenbus_dev_fatal(dev, err, "starting transaction");
1749 goto destroy_blkring;
1752 if (info->nr_ring_pages > 1) {
1753 err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1756 message = "writing ring-page-order";
1757 goto abort_transaction;
1761 /* We already got the number of queues/rings in _probe */
1762 if (info->nr_rings == 1) {
1763 err = write_per_ring_nodes(xbt, info->rinfo, dev->nodename);
1765 goto destroy_blkring;
1770 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1773 message = "writing multi-queue-num-queues";
1774 goto abort_transaction;
1777 pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1778 path = kmalloc(pathsize, GFP_KERNEL);
1781 message = "ENOMEM while writing ring references";
1782 goto abort_transaction;
1785 for_each_rinfo(info, rinfo, i) {
1786 memset(path, 0, pathsize);
1787 snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1788 err = write_per_ring_nodes(xbt, rinfo, path);
1791 goto destroy_blkring;
1796 err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1797 XEN_IO_PROTO_ABI_NATIVE);
1799 message = "writing protocol";
1800 goto abort_transaction;
1802 err = xenbus_printf(xbt, dev->nodename, "feature-persistent", "%u",
1803 info->feature_persistent);
1806 "writing persistent grants feature to xenbus");
1808 err = xenbus_transaction_end(xbt, 0);
1812 xenbus_dev_fatal(dev, err, "completing transaction");
1813 goto destroy_blkring;
1816 for_each_rinfo(info, rinfo, i) {
1819 for (j = 0; j < BLK_RING_SIZE(info); j++)
1820 rinfo->shadow[j].req.u.rw.id = j + 1;
1821 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1823 xenbus_switch_state(dev, XenbusStateInitialised);
1828 xenbus_transaction_end(xbt, 1);
1830 xenbus_dev_fatal(dev, err, "%s", message);
1832 blkif_free(info, 0);
1836 static int negotiate_mq(struct blkfront_info *info)
1838 unsigned int backend_max_queues;
1840 struct blkfront_ring_info *rinfo;
1842 BUG_ON(info->nr_rings);
1844 /* Check if backend supports multiple queues. */
1845 backend_max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1846 "multi-queue-max-queues", 1);
1847 info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1848 /* We need at least one ring. */
1849 if (!info->nr_rings)
1852 info->rinfo_size = struct_size(info->rinfo, shadow,
1853 BLK_RING_SIZE(info));
1854 info->rinfo = kvcalloc(info->nr_rings, info->rinfo_size, GFP_KERNEL);
1856 xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1861 for_each_rinfo(info, rinfo, i) {
1862 INIT_LIST_HEAD(&rinfo->indirect_pages);
1863 INIT_LIST_HEAD(&rinfo->grants);
1864 rinfo->dev_info = info;
1865 INIT_WORK(&rinfo->work, blkif_restart_queue);
1866 spin_lock_init(&rinfo->ring_lock);
1871 /* Enable the persistent grants feature. */
1872 static bool feature_persistent = true;
1873 module_param(feature_persistent, bool, 0644);
1874 MODULE_PARM_DESC(feature_persistent,
1875 "Enables the persistent grants feature");
1878 * Entry point to this code when a new device is created. Allocate the basic
1879 * structures and the ring buffer for communication with the backend, and
1880 * inform the backend of the appropriate details for those. Switch to
1881 * Initialised state.
1883 static int blkfront_probe(struct xenbus_device *dev,
1884 const struct xenbus_device_id *id)
1887 struct blkfront_info *info;
1889 /* FIXME: Use dynamic device id if this is not set. */
1890 err = xenbus_scanf(XBT_NIL, dev->nodename,
1891 "virtual-device", "%i", &vdevice);
1893 /* go looking in the extended area instead */
1894 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1897 xenbus_dev_fatal(dev, err, "reading virtual-device");
1902 if (xen_hvm_domain()) {
1905 /* no unplug has been done: do not hook devices != xen vbds */
1906 if (xen_has_pv_and_legacy_disk_devices()) {
1909 if (!VDEV_IS_EXTENDED(vdevice))
1910 major = BLKIF_MAJOR(vdevice);
1912 major = XENVBD_MAJOR;
1914 if (major != XENVBD_MAJOR) {
1916 "%s: HVM does not support vbd %d as xen block device\n",
1921 /* do not create a PV cdrom device if we are an HVM guest */
1922 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1925 if (strncmp(type, "cdrom", 5) == 0) {
1931 info = kzalloc(sizeof(*info), GFP_KERNEL);
1933 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1939 mutex_init(&info->mutex);
1940 info->vdevice = vdevice;
1941 info->connected = BLKIF_STATE_DISCONNECTED;
1943 info->feature_persistent = feature_persistent;
1945 /* Front end dir is a number, which is used as the id. */
1946 info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1947 dev_set_drvdata(&dev->dev, info);
1949 mutex_lock(&blkfront_mutex);
1950 list_add(&info->info_list, &info_list);
1951 mutex_unlock(&blkfront_mutex);
1956 static int blkif_recover(struct blkfront_info *info)
1958 unsigned int r_index;
1959 struct request *req, *n;
1963 struct blkfront_ring_info *rinfo;
1965 blkfront_gather_backend_features(info);
1966 /* Reset limits changed by blk_mq_update_nr_hw_queues(). */
1967 blkif_set_queue_limits(info);
1968 segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
1969 blk_queue_max_segments(info->rq, segs / GRANTS_PER_PSEG);
1971 for_each_rinfo(info, rinfo, r_index) {
1972 rc = blkfront_setup_indirect(rinfo);
1976 xenbus_switch_state(info->xbdev, XenbusStateConnected);
1978 /* Now safe for us to use the shared ring */
1979 info->connected = BLKIF_STATE_CONNECTED;
1981 for_each_rinfo(info, rinfo, r_index) {
1982 /* Kick any other new requests queued since we resumed */
1983 kick_pending_request_queues(rinfo);
1986 list_for_each_entry_safe(req, n, &info->requests, queuelist) {
1987 /* Requeue pending requests (flush or discard) */
1988 list_del_init(&req->queuelist);
1989 BUG_ON(req->nr_phys_segments > segs);
1990 blk_mq_requeue_request(req, false);
1992 blk_mq_start_stopped_hw_queues(info->rq, true);
1993 blk_mq_kick_requeue_list(info->rq);
1995 while ((bio = bio_list_pop(&info->bio_list)) != NULL) {
1996 /* Traverse the list of pending bios and re-queue them */
2004 * We are reconnecting to the backend, due to a suspend/resume, or a backend
2005 * driver restart. We tear down our blkif structure and recreate it, but
2006 * leave the device-layer structures intact so that this is transparent to the
2007 * rest of the kernel.
2009 static int blkfront_resume(struct xenbus_device *dev)
2011 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2014 struct blkfront_ring_info *rinfo;
2016 dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2018 bio_list_init(&info->bio_list);
2019 INIT_LIST_HEAD(&info->requests);
2020 for_each_rinfo(info, rinfo, i) {
2021 struct bio_list merge_bio;
2022 struct blk_shadow *shadow = rinfo->shadow;
2024 for (j = 0; j < BLK_RING_SIZE(info); j++) {
2026 if (!shadow[j].request)
2030 * Get the bios in the request so we can re-queue them.
2032 if (req_op(shadow[j].request) == REQ_OP_FLUSH ||
2033 req_op(shadow[j].request) == REQ_OP_DISCARD ||
2034 req_op(shadow[j].request) == REQ_OP_SECURE_ERASE ||
2035 shadow[j].request->cmd_flags & REQ_FUA) {
2037 * Flush operations don't contain bios, so
2038 * we need to requeue the whole request
2040 * XXX: but this doesn't make any sense for a
2041 * write with the FUA flag set..
2043 list_add(&shadow[j].request->queuelist, &info->requests);
2046 merge_bio.head = shadow[j].request->bio;
2047 merge_bio.tail = shadow[j].request->biotail;
2048 bio_list_merge(&info->bio_list, &merge_bio);
2049 shadow[j].request->bio = NULL;
2050 blk_mq_end_request(shadow[j].request, BLK_STS_OK);
2054 blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2056 err = talk_to_blkback(dev, info);
2058 blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings);
2061 * We have to wait for the backend to switch to
2062 * connected state, since we want to read which
2063 * features it supports.
2069 static void blkfront_closing(struct blkfront_info *info)
2071 struct xenbus_device *xbdev = info->xbdev;
2072 struct blkfront_ring_info *rinfo;
2075 if (xbdev->state == XenbusStateClosing)
2078 /* No more blkif_request(). */
2079 blk_mq_stop_hw_queues(info->rq);
2080 blk_set_queue_dying(info->rq);
2081 set_capacity(info->gd, 0);
2083 for_each_rinfo(info, rinfo, i) {
2084 /* No more gnttab callback work. */
2085 gnttab_cancel_free_callback(&rinfo->callback);
2087 /* Flush gnttab callback work. Must be done with no locks held. */
2088 flush_work(&rinfo->work);
2091 xenbus_frontend_closed(xbdev);
2094 static void blkfront_setup_discard(struct blkfront_info *info)
2096 info->feature_discard = 1;
2097 info->discard_granularity = xenbus_read_unsigned(info->xbdev->otherend,
2098 "discard-granularity",
2100 info->discard_alignment = xenbus_read_unsigned(info->xbdev->otherend,
2101 "discard-alignment", 0);
2102 info->feature_secdiscard =
2103 !!xenbus_read_unsigned(info->xbdev->otherend, "discard-secure",
2107 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2109 unsigned int psegs, grants, memflags;
2111 struct blkfront_info *info = rinfo->dev_info;
2113 memflags = memalloc_noio_save();
2115 if (info->max_indirect_segments == 0) {
2117 grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2120 * When an extra req is required, the maximum
2121 * grants supported is related to the size of the
2122 * Linux block segment.
2124 grants = GRANTS_PER_PSEG;
2128 grants = info->max_indirect_segments;
2129 psegs = DIV_ROUND_UP(grants, GRANTS_PER_PSEG);
2131 err = fill_grant_buffer(rinfo,
2132 (grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2136 if (!info->feature_persistent && info->max_indirect_segments) {
2138 * We are using indirect descriptors but not persistent
2139 * grants, we need to allocate a set of pages that can be
2140 * used for mapping indirect grefs
2142 int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2144 BUG_ON(!list_empty(&rinfo->indirect_pages));
2145 for (i = 0; i < num; i++) {
2146 struct page *indirect_page = alloc_page(GFP_KERNEL);
2149 list_add(&indirect_page->lru, &rinfo->indirect_pages);
2153 for (i = 0; i < BLK_RING_SIZE(info); i++) {
2154 rinfo->shadow[i].grants_used =
2156 sizeof(rinfo->shadow[i].grants_used[0]),
2158 rinfo->shadow[i].sg = kvcalloc(psegs,
2159 sizeof(rinfo->shadow[i].sg[0]),
2161 if (info->max_indirect_segments)
2162 rinfo->shadow[i].indirect_grants =
2163 kvcalloc(INDIRECT_GREFS(grants),
2164 sizeof(rinfo->shadow[i].indirect_grants[0]),
2166 if ((rinfo->shadow[i].grants_used == NULL) ||
2167 (rinfo->shadow[i].sg == NULL) ||
2168 (info->max_indirect_segments &&
2169 (rinfo->shadow[i].indirect_grants == NULL)))
2171 sg_init_table(rinfo->shadow[i].sg, psegs);
2174 memalloc_noio_restore(memflags);
2179 for (i = 0; i < BLK_RING_SIZE(info); i++) {
2180 kvfree(rinfo->shadow[i].grants_used);
2181 rinfo->shadow[i].grants_used = NULL;
2182 kvfree(rinfo->shadow[i].sg);
2183 rinfo->shadow[i].sg = NULL;
2184 kvfree(rinfo->shadow[i].indirect_grants);
2185 rinfo->shadow[i].indirect_grants = NULL;
2187 if (!list_empty(&rinfo->indirect_pages)) {
2188 struct page *indirect_page, *n;
2189 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2190 list_del(&indirect_page->lru);
2191 __free_page(indirect_page);
2195 memalloc_noio_restore(memflags);
2201 * Gather all backend feature-*
2203 static void blkfront_gather_backend_features(struct blkfront_info *info)
2205 unsigned int indirect_segments;
2207 info->feature_flush = 0;
2208 info->feature_fua = 0;
2211 * If there's no "feature-barrier" defined, then it means
2212 * we're dealing with a very old backend which writes
2213 * synchronously; nothing to do.
2215 * If there are barriers, then we use flush.
2217 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-barrier", 0)) {
2218 info->feature_flush = 1;
2219 info->feature_fua = 1;
2223 * And if there is "feature-flush-cache" use that above
2226 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-flush-cache",
2228 info->feature_flush = 1;
2229 info->feature_fua = 0;
2232 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-discard", 0))
2233 blkfront_setup_discard(info);
2235 if (info->feature_persistent)
2236 info->feature_persistent =
2237 !!xenbus_read_unsigned(info->xbdev->otherend,
2238 "feature-persistent", 0);
2240 indirect_segments = xenbus_read_unsigned(info->xbdev->otherend,
2241 "feature-max-indirect-segments", 0);
2242 if (indirect_segments > xen_blkif_max_segments)
2243 indirect_segments = xen_blkif_max_segments;
2244 if (indirect_segments <= BLKIF_MAX_SEGMENTS_PER_REQUEST)
2245 indirect_segments = 0;
2246 info->max_indirect_segments = indirect_segments;
2248 if (info->feature_persistent) {
2249 mutex_lock(&blkfront_mutex);
2250 schedule_delayed_work(&blkfront_work, HZ * 10);
2251 mutex_unlock(&blkfront_mutex);
2256 * Invoked when the backend is finally 'ready' (and has told produced
2257 * the details about the physical device - #sectors, size, etc).
2259 static void blkfront_connect(struct blkfront_info *info)
2261 unsigned long long sectors;
2262 unsigned long sector_size;
2263 unsigned int physical_sector_size;
2266 struct blkfront_ring_info *rinfo;
2268 switch (info->connected) {
2269 case BLKIF_STATE_CONNECTED:
2271 * Potentially, the back-end may be signalling
2272 * a capacity change; update the capacity.
2274 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2275 "sectors", "%Lu", §ors);
2276 if (XENBUS_EXIST_ERR(err))
2278 printk(KERN_INFO "Setting capacity to %Lu\n",
2280 set_capacity_and_notify(info->gd, sectors);
2283 case BLKIF_STATE_SUSPENDED:
2285 * If we are recovering from suspension, we need to wait
2286 * for the backend to announce it's features before
2287 * reconnecting, at least we need to know if the backend
2288 * supports indirect descriptors, and how many.
2290 blkif_recover(info);
2297 dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2298 __func__, info->xbdev->otherend);
2300 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2301 "sectors", "%llu", §ors,
2302 "info", "%u", &binfo,
2303 "sector-size", "%lu", §or_size,
2306 xenbus_dev_fatal(info->xbdev, err,
2307 "reading backend fields at %s",
2308 info->xbdev->otherend);
2313 * physical-sector-size is a newer field, so old backends may not
2314 * provide this. Assume physical sector size to be the same as
2315 * sector_size in that case.
2317 physical_sector_size = xenbus_read_unsigned(info->xbdev->otherend,
2318 "physical-sector-size",
2320 blkfront_gather_backend_features(info);
2321 for_each_rinfo(info, rinfo, i) {
2322 err = blkfront_setup_indirect(rinfo);
2324 xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2325 info->xbdev->otherend);
2326 blkif_free(info, 0);
2331 err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
2332 physical_sector_size);
2334 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2335 info->xbdev->otherend);
2339 xenbus_switch_state(info->xbdev, XenbusStateConnected);
2341 /* Kick pending requests. */
2342 info->connected = BLKIF_STATE_CONNECTED;
2343 for_each_rinfo(info, rinfo, i)
2344 kick_pending_request_queues(rinfo);
2346 device_add_disk(&info->xbdev->dev, info->gd, NULL);
2352 blkif_free(info, 0);
2357 * Callback received when the backend's state changes.
2359 static void blkback_changed(struct xenbus_device *dev,
2360 enum xenbus_state backend_state)
2362 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2364 dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2366 switch (backend_state) {
2367 case XenbusStateInitWait:
2368 if (dev->state != XenbusStateInitialising)
2370 if (talk_to_blkback(dev, info))
2373 case XenbusStateInitialising:
2374 case XenbusStateInitialised:
2375 case XenbusStateReconfiguring:
2376 case XenbusStateReconfigured:
2377 case XenbusStateUnknown:
2380 case XenbusStateConnected:
2382 * talk_to_blkback sets state to XenbusStateInitialised
2383 * and blkfront_connect sets it to XenbusStateConnected
2384 * (if connection went OK).
2386 * If the backend (or toolstack) decides to poke at backend
2387 * state (and re-trigger the watch by setting the state repeatedly
2388 * to XenbusStateConnected (4)) we need to deal with this.
2389 * This is allowed as this is used to communicate to the guest
2390 * that the size of disk has changed!
2392 if ((dev->state != XenbusStateInitialised) &&
2393 (dev->state != XenbusStateConnected)) {
2394 if (talk_to_blkback(dev, info))
2398 blkfront_connect(info);
2401 case XenbusStateClosed:
2402 if (dev->state == XenbusStateClosed)
2405 case XenbusStateClosing:
2406 blkfront_closing(info);
2411 static int blkfront_remove(struct xenbus_device *xbdev)
2413 struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2415 dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2417 del_gendisk(info->gd);
2419 mutex_lock(&blkfront_mutex);
2420 list_del(&info->info_list);
2421 mutex_unlock(&blkfront_mutex);
2423 blkif_free(info, 0);
2424 xlbd_release_minors(info->gd->first_minor, info->gd->minors);
2425 blk_cleanup_disk(info->gd);
2426 blk_mq_free_tag_set(&info->tag_set);
2432 static int blkfront_is_ready(struct xenbus_device *dev)
2434 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2436 return info->is_ready && info->xbdev;
2439 static const struct block_device_operations xlvbd_block_fops =
2441 .owner = THIS_MODULE,
2442 .getgeo = blkif_getgeo,
2443 .ioctl = blkif_ioctl,
2444 .compat_ioctl = blkdev_compat_ptr_ioctl,
2448 static const struct xenbus_device_id blkfront_ids[] = {
2453 static struct xenbus_driver blkfront_driver = {
2454 .ids = blkfront_ids,
2455 .probe = blkfront_probe,
2456 .remove = blkfront_remove,
2457 .resume = blkfront_resume,
2458 .otherend_changed = blkback_changed,
2459 .is_ready = blkfront_is_ready,
2462 static void purge_persistent_grants(struct blkfront_info *info)
2465 unsigned long flags;
2466 struct blkfront_ring_info *rinfo;
2468 for_each_rinfo(info, rinfo, i) {
2469 struct grant *gnt_list_entry, *tmp;
2471 spin_lock_irqsave(&rinfo->ring_lock, flags);
2473 if (rinfo->persistent_gnts_c == 0) {
2474 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2478 list_for_each_entry_safe(gnt_list_entry, tmp, &rinfo->grants,
2480 if (gnt_list_entry->gref == GRANT_INVALID_REF ||
2481 gnttab_query_foreign_access(gnt_list_entry->gref))
2484 list_del(&gnt_list_entry->node);
2485 gnttab_end_foreign_access(gnt_list_entry->gref, 0, 0UL);
2486 rinfo->persistent_gnts_c--;
2487 gnt_list_entry->gref = GRANT_INVALID_REF;
2488 list_add_tail(&gnt_list_entry->node, &rinfo->grants);
2491 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2495 static void blkfront_delay_work(struct work_struct *work)
2497 struct blkfront_info *info;
2498 bool need_schedule_work = false;
2500 mutex_lock(&blkfront_mutex);
2502 list_for_each_entry(info, &info_list, info_list) {
2503 if (info->feature_persistent) {
2504 need_schedule_work = true;
2505 mutex_lock(&info->mutex);
2506 purge_persistent_grants(info);
2507 mutex_unlock(&info->mutex);
2511 if (need_schedule_work)
2512 schedule_delayed_work(&blkfront_work, HZ * 10);
2514 mutex_unlock(&blkfront_mutex);
2517 static int __init xlblk_init(void)
2520 int nr_cpus = num_online_cpus();
2525 if (!xen_has_pv_disk_devices())
2528 if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2529 pr_warn("xen_blk: can't get major %d with name %s\n",
2530 XENVBD_MAJOR, DEV_NAME);
2534 if (xen_blkif_max_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST)
2535 xen_blkif_max_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2537 if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2538 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2539 xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2540 xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2543 if (xen_blkif_max_queues > nr_cpus) {
2544 pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2545 xen_blkif_max_queues, nr_cpus);
2546 xen_blkif_max_queues = nr_cpus;
2549 INIT_DELAYED_WORK(&blkfront_work, blkfront_delay_work);
2551 ret = xenbus_register_frontend(&blkfront_driver);
2553 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2559 module_init(xlblk_init);
2562 static void __exit xlblk_exit(void)
2564 cancel_delayed_work_sync(&blkfront_work);
2566 xenbus_unregister_driver(&blkfront_driver);
2567 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2570 module_exit(xlblk_exit);
2572 MODULE_DESCRIPTION("Xen virtual block device frontend");
2573 MODULE_LICENSE("GPL");
2574 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2575 MODULE_ALIAS("xen:vbd");
2576 MODULE_ALIAS("xenblk");