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>
52 #include <xen/xenbus.h>
53 #include <xen/grant_table.h>
54 #include <xen/events.h>
56 #include <xen/platform_pci.h>
58 #include <xen/interface/grant_table.h>
59 #include <xen/interface/io/blkif.h>
60 #include <xen/interface/io/protocols.h>
62 #include <asm/xen/hypervisor.h>
65 * The minimal size of segment supported by the block framework is PAGE_SIZE.
66 * When Linux is using a different page size than Xen, it may not be possible
67 * to put all the data in a single segment.
68 * This can happen when the backend doesn't support indirect descriptor and
69 * therefore the maximum amount of data that a request can carry is
70 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE = 44KB
72 * Note that we only support one extra request. So the Linux page size
73 * should be <= ( 2 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) =
76 #define HAS_EXTRA_REQ (BLKIF_MAX_SEGMENTS_PER_REQUEST < XEN_PFN_PER_PAGE)
79 BLKIF_STATE_DISCONNECTED,
80 BLKIF_STATE_CONNECTED,
81 BLKIF_STATE_SUSPENDED,
87 struct list_head node;
98 struct blkif_request req;
99 struct request *request;
100 struct grant **grants_used;
101 struct grant **indirect_grants;
102 struct scatterlist *sg;
104 enum blk_req_status status;
106 #define NO_ASSOCIATED_ID ~0UL
108 * Id of the sibling if we ever need 2 requests when handling a
111 unsigned long associated_id;
118 static inline struct blkif_req *blkif_req(struct request *rq)
120 return blk_mq_rq_to_pdu(rq);
123 static DEFINE_MUTEX(blkfront_mutex);
124 static const struct block_device_operations xlvbd_block_fops;
125 static struct delayed_work blkfront_work;
126 static LIST_HEAD(info_list);
129 * Maximum number of segments in indirect requests, the actual value used by
130 * the frontend driver is the minimum of this value and the value provided
131 * by the backend driver.
134 static unsigned int xen_blkif_max_segments = 32;
135 module_param_named(max_indirect_segments, xen_blkif_max_segments, uint, 0444);
136 MODULE_PARM_DESC(max_indirect_segments,
137 "Maximum amount of segments in indirect requests (default is 32)");
139 static unsigned int xen_blkif_max_queues = 4;
140 module_param_named(max_queues, xen_blkif_max_queues, uint, 0444);
141 MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
144 * Maximum order of pages to be used for the shared ring between front and
145 * backend, 4KB page granularity is used.
147 static unsigned int xen_blkif_max_ring_order;
148 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, 0444);
149 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
151 #define BLK_RING_SIZE(info) \
152 __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
155 * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
156 * characters are enough. Define to 20 to keep consistent with backend.
158 #define RINGREF_NAME_LEN (20)
160 * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
162 #define QUEUE_NAME_LEN (17)
166 * Every blkfront device can associate with one or more blkfront_ring_info,
167 * depending on how many hardware queues/rings to be used.
169 struct blkfront_ring_info {
170 /* Lock to protect data in every ring buffer. */
171 spinlock_t ring_lock;
172 struct blkif_front_ring ring;
173 unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
174 unsigned int evtchn, irq;
175 struct work_struct work;
176 struct gnttab_free_callback callback;
177 struct list_head indirect_pages;
178 struct list_head grants;
179 unsigned int persistent_gnts_c;
180 unsigned long shadow_free;
181 struct blkfront_info *dev_info;
182 struct blk_shadow shadow[];
186 * We have one of these per vbd, whether ide, scsi or 'other'. They
187 * hang in private_data off the gendisk structure. We may end up
188 * putting all kinds of interesting stuff here :-)
193 struct xenbus_device *xbdev;
196 unsigned int physical_sector_size;
199 enum blkif_state connected;
200 /* Number of pages per ring buffer. */
201 unsigned int nr_ring_pages;
202 struct request_queue *rq;
203 unsigned int feature_flush:1;
204 unsigned int feature_fua:1;
205 unsigned int feature_discard:1;
206 unsigned int feature_secdiscard:1;
207 unsigned int feature_persistent:1;
208 unsigned int discard_granularity;
209 unsigned int discard_alignment;
210 /* Number of 4KB segments handled */
211 unsigned int max_indirect_segments;
213 struct blk_mq_tag_set tag_set;
214 struct blkfront_ring_info *rinfo;
215 unsigned int nr_rings;
216 /* Save uncomplete reqs and bios for migration. */
217 struct list_head requests;
218 struct bio_list bio_list;
219 struct list_head info_list;
222 static unsigned int nr_minors;
223 static unsigned long *minors;
224 static DEFINE_SPINLOCK(minor_lock);
226 #define GRANT_INVALID_REF 0
228 #define PARTS_PER_DISK 16
229 #define PARTS_PER_EXT_DISK 256
231 #define BLKIF_MAJOR(dev) ((dev)>>8)
232 #define BLKIF_MINOR(dev) ((dev) & 0xff)
235 #define EXTENDED (1<<EXT_SHIFT)
236 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
237 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
238 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
239 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
240 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
241 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
243 #define DEV_NAME "xvd" /* name in /dev */
246 * Grants are always the same size as a Xen page (i.e 4KB).
247 * A physical segment is always the same size as a Linux page.
248 * Number of grants per physical segment
250 #define GRANTS_PER_PSEG (PAGE_SIZE / XEN_PAGE_SIZE)
252 #define GRANTS_PER_INDIRECT_FRAME \
253 (XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
255 #define INDIRECT_GREFS(_grants) \
256 DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
258 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
259 static void blkfront_gather_backend_features(struct blkfront_info *info);
260 static int negotiate_mq(struct blkfront_info *info);
262 static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
264 unsigned long free = rinfo->shadow_free;
266 BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
267 rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
268 rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
272 static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
275 if (rinfo->shadow[id].req.u.rw.id != id)
277 if (rinfo->shadow[id].request == NULL)
279 rinfo->shadow[id].req.u.rw.id = rinfo->shadow_free;
280 rinfo->shadow[id].request = NULL;
281 rinfo->shadow_free = id;
285 static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
287 struct blkfront_info *info = rinfo->dev_info;
288 struct page *granted_page;
289 struct grant *gnt_list_entry, *n;
293 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
297 if (info->feature_persistent) {
298 granted_page = alloc_page(GFP_NOIO);
300 kfree(gnt_list_entry);
303 gnt_list_entry->page = granted_page;
306 gnt_list_entry->gref = GRANT_INVALID_REF;
307 list_add(&gnt_list_entry->node, &rinfo->grants);
314 list_for_each_entry_safe(gnt_list_entry, n,
315 &rinfo->grants, node) {
316 list_del(&gnt_list_entry->node);
317 if (info->feature_persistent)
318 __free_page(gnt_list_entry->page);
319 kfree(gnt_list_entry);
326 static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
328 struct grant *gnt_list_entry;
330 BUG_ON(list_empty(&rinfo->grants));
331 gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
333 list_del(&gnt_list_entry->node);
335 if (gnt_list_entry->gref != GRANT_INVALID_REF)
336 rinfo->persistent_gnts_c--;
338 return gnt_list_entry;
341 static inline void grant_foreign_access(const struct grant *gnt_list_entry,
342 const struct blkfront_info *info)
344 gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
345 info->xbdev->otherend_id,
346 gnt_list_entry->page,
350 static struct grant *get_grant(grant_ref_t *gref_head,
352 struct blkfront_ring_info *rinfo)
354 struct grant *gnt_list_entry = get_free_grant(rinfo);
355 struct blkfront_info *info = rinfo->dev_info;
357 if (gnt_list_entry->gref != GRANT_INVALID_REF)
358 return gnt_list_entry;
360 /* Assign a gref to this page */
361 gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
362 BUG_ON(gnt_list_entry->gref == -ENOSPC);
363 if (info->feature_persistent)
364 grant_foreign_access(gnt_list_entry, info);
366 /* Grant access to the GFN passed by the caller */
367 gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
368 info->xbdev->otherend_id,
372 return gnt_list_entry;
375 static struct grant *get_indirect_grant(grant_ref_t *gref_head,
376 struct blkfront_ring_info *rinfo)
378 struct grant *gnt_list_entry = get_free_grant(rinfo);
379 struct blkfront_info *info = rinfo->dev_info;
381 if (gnt_list_entry->gref != GRANT_INVALID_REF)
382 return gnt_list_entry;
384 /* Assign a gref to this page */
385 gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
386 BUG_ON(gnt_list_entry->gref == -ENOSPC);
387 if (!info->feature_persistent) {
388 struct page *indirect_page;
390 /* Fetch a pre-allocated page to use for indirect grefs */
391 BUG_ON(list_empty(&rinfo->indirect_pages));
392 indirect_page = list_first_entry(&rinfo->indirect_pages,
394 list_del(&indirect_page->lru);
395 gnt_list_entry->page = indirect_page;
397 grant_foreign_access(gnt_list_entry, info);
399 return gnt_list_entry;
402 static const char *op_name(int op)
404 static const char *const names[] = {
405 [BLKIF_OP_READ] = "read",
406 [BLKIF_OP_WRITE] = "write",
407 [BLKIF_OP_WRITE_BARRIER] = "barrier",
408 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
409 [BLKIF_OP_DISCARD] = "discard" };
411 if (op < 0 || op >= ARRAY_SIZE(names))
419 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
421 unsigned int end = minor + nr;
424 if (end > nr_minors) {
425 unsigned long *bitmap, *old;
427 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
432 spin_lock(&minor_lock);
433 if (end > nr_minors) {
435 memcpy(bitmap, minors,
436 BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
438 nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
441 spin_unlock(&minor_lock);
445 spin_lock(&minor_lock);
446 if (find_next_bit(minors, end, minor) >= end) {
447 bitmap_set(minors, minor, nr);
451 spin_unlock(&minor_lock);
456 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
458 unsigned int end = minor + nr;
460 BUG_ON(end > nr_minors);
461 spin_lock(&minor_lock);
462 bitmap_clear(minors, minor, nr);
463 spin_unlock(&minor_lock);
466 static void blkif_restart_queue_callback(void *arg)
468 struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
469 schedule_work(&rinfo->work);
472 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
474 /* We don't have real geometry info, but let's at least return
475 values consistent with the size of the device */
476 sector_t nsect = get_capacity(bd->bd_disk);
477 sector_t cylinders = nsect;
481 sector_div(cylinders, hg->heads * hg->sectors);
482 hg->cylinders = cylinders;
483 if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
484 hg->cylinders = 0xffff;
488 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
489 unsigned command, unsigned long argument)
491 struct blkfront_info *info = bdev->bd_disk->private_data;
494 dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
495 command, (long)argument);
498 case CDROMMULTISESSION:
499 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
500 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
501 if (put_user(0, (char __user *)(argument + i)))
505 case CDROM_GET_CAPABILITY: {
506 struct gendisk *gd = info->gd;
507 if (gd->flags & GENHD_FL_CD)
513 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
515 return -EINVAL; /* same return as native Linux */
521 static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
523 struct blkif_request **ring_req)
527 *ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
528 rinfo->ring.req_prod_pvt++;
530 id = get_id_from_freelist(rinfo);
531 rinfo->shadow[id].request = req;
532 rinfo->shadow[id].status = REQ_WAITING;
533 rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
535 (*ring_req)->u.rw.id = id;
540 static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
542 struct blkfront_info *info = rinfo->dev_info;
543 struct blkif_request *ring_req;
546 /* Fill out a communications ring structure. */
547 id = blkif_ring_get_request(rinfo, req, &ring_req);
549 ring_req->operation = BLKIF_OP_DISCARD;
550 ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
551 ring_req->u.discard.id = id;
552 ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
553 if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard)
554 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
556 ring_req->u.discard.flag = 0;
558 /* Keep a private copy so we can reissue requests when recovering. */
559 rinfo->shadow[id].req = *ring_req;
564 struct setup_rw_req {
565 unsigned int grant_idx;
566 struct blkif_request_segment *segments;
567 struct blkfront_ring_info *rinfo;
568 struct blkif_request *ring_req;
569 grant_ref_t gref_head;
571 /* Only used when persistent grant is used and it's a read request */
573 unsigned int bvec_off;
576 bool require_extra_req;
577 struct blkif_request *extra_ring_req;
580 static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
581 unsigned int len, void *data)
583 struct setup_rw_req *setup = data;
585 struct grant *gnt_list_entry;
586 unsigned int fsect, lsect;
587 /* Convenient aliases */
588 unsigned int grant_idx = setup->grant_idx;
589 struct blkif_request *ring_req = setup->ring_req;
590 struct blkfront_ring_info *rinfo = setup->rinfo;
592 * We always use the shadow of the first request to store the list
593 * of grant associated to the block I/O request. This made the
594 * completion more easy to handle even if the block I/O request is
597 struct blk_shadow *shadow = &rinfo->shadow[setup->id];
599 if (unlikely(setup->require_extra_req &&
600 grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
602 * We are using the second request, setup grant_idx
603 * to be the index of the segment array.
605 grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
606 ring_req = setup->extra_ring_req;
609 if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
610 (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
612 kunmap_atomic(setup->segments);
614 n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
615 gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
616 shadow->indirect_grants[n] = gnt_list_entry;
617 setup->segments = kmap_atomic(gnt_list_entry->page);
618 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
621 gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
622 ref = gnt_list_entry->gref;
624 * All the grants are stored in the shadow of the first
625 * request. Therefore we have to use the global index.
627 shadow->grants_used[setup->grant_idx] = gnt_list_entry;
629 if (setup->need_copy) {
632 shared_data = kmap_atomic(gnt_list_entry->page);
634 * this does not wipe data stored outside the
635 * range sg->offset..sg->offset+sg->length.
636 * Therefore, blkback *could* see data from
637 * previous requests. This is OK as long as
638 * persistent grants are shared with just one
639 * domain. It may need refactoring if this
642 memcpy(shared_data + offset,
643 setup->bvec_data + setup->bvec_off,
646 kunmap_atomic(shared_data);
647 setup->bvec_off += len;
651 lsect = fsect + (len >> 9) - 1;
652 if (ring_req->operation != BLKIF_OP_INDIRECT) {
653 ring_req->u.rw.seg[grant_idx] =
654 (struct blkif_request_segment) {
657 .last_sect = lsect };
659 setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
660 (struct blkif_request_segment) {
663 .last_sect = lsect };
666 (setup->grant_idx)++;
669 static void blkif_setup_extra_req(struct blkif_request *first,
670 struct blkif_request *second)
672 uint16_t nr_segments = first->u.rw.nr_segments;
675 * The second request is only present when the first request uses
676 * all its segments. It's always the continuity of the first one.
678 first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
680 second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
681 second->u.rw.sector_number = first->u.rw.sector_number +
682 (BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
684 second->u.rw.handle = first->u.rw.handle;
685 second->operation = first->operation;
688 static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
690 struct blkfront_info *info = rinfo->dev_info;
691 struct blkif_request *ring_req, *extra_ring_req = NULL;
692 unsigned long id, extra_id = NO_ASSOCIATED_ID;
693 bool require_extra_req = false;
695 struct setup_rw_req setup = {
699 .need_copy = rq_data_dir(req) && info->feature_persistent,
703 * Used to store if we are able to queue the request by just using
704 * existing persistent grants, or if we have to get new grants,
705 * as there are not sufficiently many free.
707 bool new_persistent_gnts = false;
708 struct scatterlist *sg;
709 int num_sg, max_grefs, num_grant;
711 max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
712 if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
714 * If we are using indirect segments we need to account
715 * for the indirect grefs used in the request.
717 max_grefs += INDIRECT_GREFS(max_grefs);
719 /* Check if we have enough persistent grants to allocate a requests */
720 if (rinfo->persistent_gnts_c < max_grefs) {
721 new_persistent_gnts = true;
723 if (gnttab_alloc_grant_references(
724 max_grefs - rinfo->persistent_gnts_c,
725 &setup.gref_head) < 0) {
726 gnttab_request_free_callback(
728 blkif_restart_queue_callback,
730 max_grefs - rinfo->persistent_gnts_c);
735 /* Fill out a communications ring structure. */
736 id = blkif_ring_get_request(rinfo, req, &ring_req);
738 num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
740 /* Calculate the number of grant used */
741 for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
742 num_grant += gnttab_count_grant(sg->offset, sg->length);
744 require_extra_req = info->max_indirect_segments == 0 &&
745 num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
746 BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
748 rinfo->shadow[id].num_sg = num_sg;
749 if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
750 likely(!require_extra_req)) {
752 * The indirect operation can only be a BLKIF_OP_READ or
755 BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA);
756 ring_req->operation = BLKIF_OP_INDIRECT;
757 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
758 BLKIF_OP_WRITE : BLKIF_OP_READ;
759 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
760 ring_req->u.indirect.handle = info->handle;
761 ring_req->u.indirect.nr_segments = num_grant;
763 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
764 ring_req->u.rw.handle = info->handle;
765 ring_req->operation = rq_data_dir(req) ?
766 BLKIF_OP_WRITE : BLKIF_OP_READ;
767 if (req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA) {
769 * Ideally we can do an unordered flush-to-disk.
770 * In case the backend onlysupports barriers, use that.
771 * A barrier request a superset of FUA, so we can
772 * implement it the same way. (It's also a FLUSH+FUA,
773 * since it is guaranteed ordered WRT previous writes.)
775 if (info->feature_flush && info->feature_fua)
776 ring_req->operation =
777 BLKIF_OP_WRITE_BARRIER;
778 else if (info->feature_flush)
779 ring_req->operation =
780 BLKIF_OP_FLUSH_DISKCACHE;
782 ring_req->operation = 0;
784 ring_req->u.rw.nr_segments = num_grant;
785 if (unlikely(require_extra_req)) {
786 extra_id = blkif_ring_get_request(rinfo, req,
789 * Only the first request contains the scatter-gather
792 rinfo->shadow[extra_id].num_sg = 0;
794 blkif_setup_extra_req(ring_req, extra_ring_req);
796 /* Link the 2 requests together */
797 rinfo->shadow[extra_id].associated_id = id;
798 rinfo->shadow[id].associated_id = extra_id;
802 setup.ring_req = ring_req;
805 setup.require_extra_req = require_extra_req;
806 if (unlikely(require_extra_req))
807 setup.extra_ring_req = extra_ring_req;
809 for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
810 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
812 if (setup.need_copy) {
813 setup.bvec_off = sg->offset;
814 setup.bvec_data = kmap_atomic(sg_page(sg));
817 gnttab_foreach_grant_in_range(sg_page(sg),
820 blkif_setup_rw_req_grant,
824 kunmap_atomic(setup.bvec_data);
827 kunmap_atomic(setup.segments);
829 /* Keep a private copy so we can reissue requests when recovering. */
830 rinfo->shadow[id].req = *ring_req;
831 if (unlikely(require_extra_req))
832 rinfo->shadow[extra_id].req = *extra_ring_req;
834 if (new_persistent_gnts)
835 gnttab_free_grant_references(setup.gref_head);
841 * Generate a Xen blkfront IO request from a blk layer request. Reads
842 * and writes are handled as expected.
844 * @req: a request struct
846 static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
848 if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
851 if (unlikely(req_op(req) == REQ_OP_DISCARD ||
852 req_op(req) == REQ_OP_SECURE_ERASE))
853 return blkif_queue_discard_req(req, rinfo);
855 return blkif_queue_rw_req(req, rinfo);
858 static inline void flush_requests(struct blkfront_ring_info *rinfo)
862 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
865 notify_remote_via_irq(rinfo->irq);
868 static inline bool blkif_request_flush_invalid(struct request *req,
869 struct blkfront_info *info)
871 return (blk_rq_is_passthrough(req) ||
872 ((req_op(req) == REQ_OP_FLUSH) &&
873 !info->feature_flush) ||
874 ((req->cmd_flags & REQ_FUA) &&
875 !info->feature_fua));
878 static blk_status_t blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
879 const struct blk_mq_queue_data *qd)
882 int qid = hctx->queue_num;
883 struct blkfront_info *info = hctx->queue->queuedata;
884 struct blkfront_ring_info *rinfo = NULL;
886 BUG_ON(info->nr_rings <= qid);
887 rinfo = &info->rinfo[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 rq->limits.discard_alignment = info->discard_alignment;
937 if (info->feature_secdiscard)
938 blk_queue_flag_set(QUEUE_FLAG_SECERASE, rq);
941 /* Hard sector size and max sectors impersonate the equiv. hardware. */
942 blk_queue_logical_block_size(rq, info->sector_size);
943 blk_queue_physical_block_size(rq, info->physical_sector_size);
944 blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
946 /* Each segment in a request is up to an aligned page in size. */
947 blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
948 blk_queue_max_segment_size(rq, PAGE_SIZE);
950 /* Ensure a merged request will fit in a single I/O ring slot. */
951 blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
953 /* Make sure buffer addresses are sector-aligned. */
954 blk_queue_dma_alignment(rq, 511);
957 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
958 unsigned int physical_sector_size)
960 struct request_queue *rq;
961 struct blkfront_info *info = gd->private_data;
963 memset(&info->tag_set, 0, sizeof(info->tag_set));
964 info->tag_set.ops = &blkfront_mq_ops;
965 info->tag_set.nr_hw_queues = info->nr_rings;
966 if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
968 * When indirect descriptior is not supported, the I/O request
969 * will be split between multiple request in the ring.
970 * To avoid problems when sending the request, divide by
971 * 2 the depth of the queue.
973 info->tag_set.queue_depth = BLK_RING_SIZE(info) / 2;
975 info->tag_set.queue_depth = BLK_RING_SIZE(info);
976 info->tag_set.numa_node = NUMA_NO_NODE;
977 info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
978 info->tag_set.cmd_size = sizeof(struct blkif_req);
979 info->tag_set.driver_data = info;
981 if (blk_mq_alloc_tag_set(&info->tag_set))
983 rq = blk_mq_init_queue(&info->tag_set);
985 blk_mq_free_tag_set(&info->tag_set);
989 rq->queuedata = info;
990 info->rq = gd->queue = rq;
992 info->sector_size = sector_size;
993 info->physical_sector_size = physical_sector_size;
994 blkif_set_queue_limits(info);
999 static const char *flush_info(struct blkfront_info *info)
1001 if (info->feature_flush && info->feature_fua)
1002 return "barrier: enabled;";
1003 else if (info->feature_flush)
1004 return "flush diskcache: enabled;";
1006 return "barrier or flush: disabled;";
1009 static void xlvbd_flush(struct blkfront_info *info)
1011 blk_queue_write_cache(info->rq, info->feature_flush ? true : false,
1012 info->feature_fua ? true : false);
1013 pr_info("blkfront: %s: %s %s %s %s %s\n",
1014 info->gd->disk_name, flush_info(info),
1015 "persistent grants:", info->feature_persistent ?
1016 "enabled;" : "disabled;", "indirect descriptors:",
1017 info->max_indirect_segments ? "enabled;" : "disabled;");
1020 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
1023 major = BLKIF_MAJOR(vdevice);
1024 *minor = BLKIF_MINOR(vdevice);
1026 case XEN_IDE0_MAJOR:
1027 *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
1028 *minor = ((*minor / 64) * PARTS_PER_DISK) +
1029 EMULATED_HD_DISK_MINOR_OFFSET;
1031 case XEN_IDE1_MAJOR:
1032 *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1033 *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1034 EMULATED_HD_DISK_MINOR_OFFSET;
1036 case XEN_SCSI_DISK0_MAJOR:
1037 *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1038 *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1040 case XEN_SCSI_DISK1_MAJOR:
1041 case XEN_SCSI_DISK2_MAJOR:
1042 case XEN_SCSI_DISK3_MAJOR:
1043 case XEN_SCSI_DISK4_MAJOR:
1044 case XEN_SCSI_DISK5_MAJOR:
1045 case XEN_SCSI_DISK6_MAJOR:
1046 case XEN_SCSI_DISK7_MAJOR:
1047 *offset = (*minor / PARTS_PER_DISK) +
1048 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1049 EMULATED_SD_DISK_NAME_OFFSET;
1051 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1052 EMULATED_SD_DISK_MINOR_OFFSET;
1054 case XEN_SCSI_DISK8_MAJOR:
1055 case XEN_SCSI_DISK9_MAJOR:
1056 case XEN_SCSI_DISK10_MAJOR:
1057 case XEN_SCSI_DISK11_MAJOR:
1058 case XEN_SCSI_DISK12_MAJOR:
1059 case XEN_SCSI_DISK13_MAJOR:
1060 case XEN_SCSI_DISK14_MAJOR:
1061 case XEN_SCSI_DISK15_MAJOR:
1062 *offset = (*minor / PARTS_PER_DISK) +
1063 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1064 EMULATED_SD_DISK_NAME_OFFSET;
1066 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1067 EMULATED_SD_DISK_MINOR_OFFSET;
1070 *offset = *minor / PARTS_PER_DISK;
1073 printk(KERN_WARNING "blkfront: your disk configuration is "
1074 "incorrect, please use an xvd device instead\n");
1080 static char *encode_disk_name(char *ptr, unsigned int n)
1083 ptr = encode_disk_name(ptr, n / 26 - 1);
1084 *ptr = 'a' + n % 26;
1088 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1089 struct blkfront_info *info,
1090 u16 vdisk_info, u16 sector_size,
1091 unsigned int physical_sector_size)
1096 unsigned int offset;
1101 BUG_ON(info->gd != NULL);
1102 BUG_ON(info->rq != NULL);
1104 if ((info->vdevice>>EXT_SHIFT) > 1) {
1105 /* this is above the extended range; something is wrong */
1106 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1110 if (!VDEV_IS_EXTENDED(info->vdevice)) {
1111 err = xen_translate_vdev(info->vdevice, &minor, &offset);
1114 nr_parts = PARTS_PER_DISK;
1116 minor = BLKIF_MINOR_EXT(info->vdevice);
1117 nr_parts = PARTS_PER_EXT_DISK;
1118 offset = minor / nr_parts;
1119 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1120 printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1121 "emulated IDE disks,\n\t choose an xvd device name"
1122 "from xvde on\n", info->vdevice);
1124 if (minor >> MINORBITS) {
1125 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1126 info->vdevice, minor);
1130 if ((minor % nr_parts) == 0)
1131 nr_minors = nr_parts;
1133 err = xlbd_reserve_minors(minor, nr_minors);
1138 gd = alloc_disk(nr_minors);
1142 strcpy(gd->disk_name, DEV_NAME);
1143 ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1144 BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1148 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1149 "%d", minor & (nr_parts - 1));
1151 gd->major = XENVBD_MAJOR;
1152 gd->first_minor = minor;
1153 gd->fops = &xlvbd_block_fops;
1154 gd->private_data = info;
1155 set_capacity(gd, capacity);
1157 if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size)) {
1164 if (vdisk_info & VDISK_READONLY)
1167 if (vdisk_info & VDISK_REMOVABLE)
1168 gd->flags |= GENHD_FL_REMOVABLE;
1170 if (vdisk_info & VDISK_CDROM)
1171 gd->flags |= GENHD_FL_CD;
1176 xlbd_release_minors(minor, nr_minors);
1181 static void xlvbd_release_gendisk(struct blkfront_info *info)
1183 unsigned int minor, nr_minors, i;
1185 if (info->rq == NULL)
1188 /* No more blkif_request(). */
1189 blk_mq_stop_hw_queues(info->rq);
1191 for (i = 0; i < info->nr_rings; i++) {
1192 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1194 /* No more gnttab callback work. */
1195 gnttab_cancel_free_callback(&rinfo->callback);
1197 /* Flush gnttab callback work. Must be done with no locks held. */
1198 flush_work(&rinfo->work);
1201 del_gendisk(info->gd);
1203 minor = info->gd->first_minor;
1204 nr_minors = info->gd->minors;
1205 xlbd_release_minors(minor, nr_minors);
1207 blk_cleanup_queue(info->rq);
1208 blk_mq_free_tag_set(&info->tag_set);
1215 /* Already hold rinfo->ring_lock. */
1216 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1218 if (!RING_FULL(&rinfo->ring))
1219 blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1222 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1224 unsigned long flags;
1226 spin_lock_irqsave(&rinfo->ring_lock, flags);
1227 kick_pending_request_queues_locked(rinfo);
1228 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1231 static void blkif_restart_queue(struct work_struct *work)
1233 struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1235 if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1236 kick_pending_request_queues(rinfo);
1239 static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1241 struct grant *persistent_gnt, *n;
1242 struct blkfront_info *info = rinfo->dev_info;
1246 * Remove indirect pages, this only happens when using indirect
1247 * descriptors but not persistent grants
1249 if (!list_empty(&rinfo->indirect_pages)) {
1250 struct page *indirect_page, *n;
1252 BUG_ON(info->feature_persistent);
1253 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1254 list_del(&indirect_page->lru);
1255 __free_page(indirect_page);
1259 /* Remove all persistent grants. */
1260 if (!list_empty(&rinfo->grants)) {
1261 list_for_each_entry_safe(persistent_gnt, n,
1262 &rinfo->grants, node) {
1263 list_del(&persistent_gnt->node);
1264 if (persistent_gnt->gref != GRANT_INVALID_REF) {
1265 gnttab_end_foreign_access(persistent_gnt->gref,
1267 rinfo->persistent_gnts_c--;
1269 if (info->feature_persistent)
1270 __free_page(persistent_gnt->page);
1271 kfree(persistent_gnt);
1274 BUG_ON(rinfo->persistent_gnts_c != 0);
1276 for (i = 0; i < BLK_RING_SIZE(info); i++) {
1278 * Clear persistent grants present in requests already
1279 * on the shared ring
1281 if (!rinfo->shadow[i].request)
1284 segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1285 rinfo->shadow[i].req.u.indirect.nr_segments :
1286 rinfo->shadow[i].req.u.rw.nr_segments;
1287 for (j = 0; j < segs; j++) {
1288 persistent_gnt = rinfo->shadow[i].grants_used[j];
1289 gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1290 if (info->feature_persistent)
1291 __free_page(persistent_gnt->page);
1292 kfree(persistent_gnt);
1295 if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1297 * If this is not an indirect operation don't try to
1298 * free indirect segments
1302 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1303 persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1304 gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1305 __free_page(persistent_gnt->page);
1306 kfree(persistent_gnt);
1310 kvfree(rinfo->shadow[i].grants_used);
1311 rinfo->shadow[i].grants_used = NULL;
1312 kvfree(rinfo->shadow[i].indirect_grants);
1313 rinfo->shadow[i].indirect_grants = NULL;
1314 kvfree(rinfo->shadow[i].sg);
1315 rinfo->shadow[i].sg = NULL;
1318 /* No more gnttab callback work. */
1319 gnttab_cancel_free_callback(&rinfo->callback);
1321 /* Flush gnttab callback work. Must be done with no locks held. */
1322 flush_work(&rinfo->work);
1324 /* Free resources associated with old device channel. */
1325 for (i = 0; i < info->nr_ring_pages; i++) {
1326 if (rinfo->ring_ref[i] != GRANT_INVALID_REF) {
1327 gnttab_end_foreign_access(rinfo->ring_ref[i], 0, 0);
1328 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1331 free_pages((unsigned long)rinfo->ring.sring, get_order(info->nr_ring_pages * XEN_PAGE_SIZE));
1332 rinfo->ring.sring = NULL;
1335 unbind_from_irqhandler(rinfo->irq, rinfo);
1336 rinfo->evtchn = rinfo->irq = 0;
1339 static void blkif_free(struct blkfront_info *info, int suspend)
1343 /* Prevent new requests being issued until we fix things up. */
1344 info->connected = suspend ?
1345 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1346 /* No more blkif_request(). */
1348 blk_mq_stop_hw_queues(info->rq);
1350 for (i = 0; i < info->nr_rings; i++)
1351 blkif_free_ring(&info->rinfo[i]);
1353 kvfree(info->rinfo);
1358 struct copy_from_grant {
1359 const struct blk_shadow *s;
1360 unsigned int grant_idx;
1361 unsigned int bvec_offset;
1365 static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1366 unsigned int len, void *data)
1368 struct copy_from_grant *info = data;
1370 /* Convenient aliases */
1371 const struct blk_shadow *s = info->s;
1373 shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1375 memcpy(info->bvec_data + info->bvec_offset,
1376 shared_data + offset, len);
1378 info->bvec_offset += len;
1381 kunmap_atomic(shared_data);
1384 static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1388 case BLKIF_RSP_OKAY:
1390 case BLKIF_RSP_EOPNOTSUPP:
1391 return REQ_EOPNOTSUPP;
1392 case BLKIF_RSP_ERROR:
1400 * Get the final status of the block request based on two ring response
1402 static int blkif_get_final_status(enum blk_req_status s1,
1403 enum blk_req_status s2)
1405 BUG_ON(s1 == REQ_WAITING);
1406 BUG_ON(s2 == REQ_WAITING);
1408 if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1409 return BLKIF_RSP_ERROR;
1410 else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1411 return BLKIF_RSP_EOPNOTSUPP;
1412 return BLKIF_RSP_OKAY;
1415 static bool blkif_completion(unsigned long *id,
1416 struct blkfront_ring_info *rinfo,
1417 struct blkif_response *bret)
1420 struct scatterlist *sg;
1421 int num_sg, num_grant;
1422 struct blkfront_info *info = rinfo->dev_info;
1423 struct blk_shadow *s = &rinfo->shadow[*id];
1424 struct copy_from_grant data = {
1428 num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1429 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1431 /* The I/O request may be split in two. */
1432 if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1433 struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1435 /* Keep the status of the current response in shadow. */
1436 s->status = blkif_rsp_to_req_status(bret->status);
1438 /* Wait the second response if not yet here. */
1439 if (s2->status == REQ_WAITING)
1442 bret->status = blkif_get_final_status(s->status,
1446 * All the grants is stored in the first shadow in order
1447 * to make the completion code simpler.
1449 num_grant += s2->req.u.rw.nr_segments;
1452 * The two responses may not come in order. Only the
1453 * first request will store the scatter-gather list.
1455 if (s2->num_sg != 0) {
1456 /* Update "id" with the ID of the first response. */
1457 *id = s->associated_id;
1462 * We don't need anymore the second request, so recycling
1465 if (add_id_to_freelist(rinfo, s->associated_id))
1466 WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1467 info->gd->disk_name, s->associated_id);
1473 if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1474 for_each_sg(s->sg, sg, num_sg, i) {
1475 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1477 data.bvec_offset = sg->offset;
1478 data.bvec_data = kmap_atomic(sg_page(sg));
1480 gnttab_foreach_grant_in_range(sg_page(sg),
1483 blkif_copy_from_grant,
1486 kunmap_atomic(data.bvec_data);
1489 /* Add the persistent grant into the list of free grants */
1490 for (i = 0; i < num_grant; i++) {
1491 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1493 * If the grant is still mapped by the backend (the
1494 * backend has chosen to make this grant persistent)
1495 * we add it at the head of the list, so it will be
1498 if (!info->feature_persistent)
1499 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1500 s->grants_used[i]->gref);
1501 list_add(&s->grants_used[i]->node, &rinfo->grants);
1502 rinfo->persistent_gnts_c++;
1505 * If the grant is not mapped by the backend we end the
1506 * foreign access and add it to the tail of the list,
1507 * so it will not be picked again unless we run out of
1508 * persistent grants.
1510 gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1511 s->grants_used[i]->gref = GRANT_INVALID_REF;
1512 list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1515 if (s->req.operation == BLKIF_OP_INDIRECT) {
1516 for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1517 if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1518 if (!info->feature_persistent)
1519 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1520 s->indirect_grants[i]->gref);
1521 list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1522 rinfo->persistent_gnts_c++;
1524 struct page *indirect_page;
1526 gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1528 * Add the used indirect page back to the list of
1529 * available pages for indirect grefs.
1531 if (!info->feature_persistent) {
1532 indirect_page = s->indirect_grants[i]->page;
1533 list_add(&indirect_page->lru, &rinfo->indirect_pages);
1535 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1536 list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1544 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1546 struct request *req;
1547 struct blkif_response *bret;
1549 unsigned long flags;
1550 struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1551 struct blkfront_info *info = rinfo->dev_info;
1553 if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
1556 spin_lock_irqsave(&rinfo->ring_lock, flags);
1558 rp = rinfo->ring.sring->rsp_prod;
1559 rmb(); /* Ensure we see queued responses up to 'rp'. */
1561 for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1564 bret = RING_GET_RESPONSE(&rinfo->ring, i);
1567 * The backend has messed up and given us an id that we would
1568 * never have given to it (we stamp it up to BLK_RING_SIZE -
1569 * look in get_id_from_freelist.
1571 if (id >= BLK_RING_SIZE(info)) {
1572 WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1573 info->gd->disk_name, op_name(bret->operation), id);
1574 /* We can't safely get the 'struct request' as
1575 * the id is busted. */
1578 req = rinfo->shadow[id].request;
1580 if (bret->operation != BLKIF_OP_DISCARD) {
1582 * We may need to wait for an extra response if the
1583 * I/O request is split in 2
1585 if (!blkif_completion(&id, rinfo, bret))
1589 if (add_id_to_freelist(rinfo, id)) {
1590 WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1591 info->gd->disk_name, op_name(bret->operation), id);
1595 if (bret->status == BLKIF_RSP_OKAY)
1596 blkif_req(req)->error = BLK_STS_OK;
1598 blkif_req(req)->error = BLK_STS_IOERR;
1600 switch (bret->operation) {
1601 case BLKIF_OP_DISCARD:
1602 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1603 struct request_queue *rq = info->rq;
1604 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1605 info->gd->disk_name, op_name(bret->operation));
1606 blkif_req(req)->error = BLK_STS_NOTSUPP;
1607 info->feature_discard = 0;
1608 info->feature_secdiscard = 0;
1609 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1610 blk_queue_flag_clear(QUEUE_FLAG_SECERASE, rq);
1613 case BLKIF_OP_FLUSH_DISKCACHE:
1614 case BLKIF_OP_WRITE_BARRIER:
1615 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1616 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1617 info->gd->disk_name, op_name(bret->operation));
1618 blkif_req(req)->error = BLK_STS_NOTSUPP;
1620 if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1621 rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1622 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1623 info->gd->disk_name, op_name(bret->operation));
1624 blkif_req(req)->error = BLK_STS_NOTSUPP;
1626 if (unlikely(blkif_req(req)->error)) {
1627 if (blkif_req(req)->error == BLK_STS_NOTSUPP)
1628 blkif_req(req)->error = BLK_STS_OK;
1629 info->feature_fua = 0;
1630 info->feature_flush = 0;
1635 case BLKIF_OP_WRITE:
1636 if (unlikely(bret->status != BLKIF_RSP_OKAY))
1637 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1638 "request: %x\n", bret->status);
1645 blk_mq_complete_request(req);
1648 rinfo->ring.rsp_cons = i;
1650 if (i != rinfo->ring.req_prod_pvt) {
1652 RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1656 rinfo->ring.sring->rsp_event = i + 1;
1658 kick_pending_request_queues_locked(rinfo);
1660 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1666 static int setup_blkring(struct xenbus_device *dev,
1667 struct blkfront_ring_info *rinfo)
1669 struct blkif_sring *sring;
1671 struct blkfront_info *info = rinfo->dev_info;
1672 unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1673 grant_ref_t gref[XENBUS_MAX_RING_GRANTS];
1675 for (i = 0; i < info->nr_ring_pages; i++)
1676 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1678 sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
1679 get_order(ring_size));
1681 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1684 SHARED_RING_INIT(sring);
1685 FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1687 err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref);
1689 free_pages((unsigned long)sring, get_order(ring_size));
1690 rinfo->ring.sring = NULL;
1693 for (i = 0; i < info->nr_ring_pages; i++)
1694 rinfo->ring_ref[i] = gref[i];
1696 err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1700 err = bind_evtchn_to_irqhandler(rinfo->evtchn, blkif_interrupt, 0,
1703 xenbus_dev_fatal(dev, err,
1704 "bind_evtchn_to_irqhandler failed");
1711 blkif_free(info, 0);
1716 * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1717 * ring buffer may have multi pages depending on ->nr_ring_pages.
1719 static int write_per_ring_nodes(struct xenbus_transaction xbt,
1720 struct blkfront_ring_info *rinfo, const char *dir)
1724 const char *message = NULL;
1725 struct blkfront_info *info = rinfo->dev_info;
1727 if (info->nr_ring_pages == 1) {
1728 err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1730 message = "writing ring-ref";
1731 goto abort_transaction;
1734 for (i = 0; i < info->nr_ring_pages; i++) {
1735 char ring_ref_name[RINGREF_NAME_LEN];
1737 snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1738 err = xenbus_printf(xbt, dir, ring_ref_name,
1739 "%u", rinfo->ring_ref[i]);
1741 message = "writing ring-ref";
1742 goto abort_transaction;
1747 err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1749 message = "writing event-channel";
1750 goto abort_transaction;
1756 xenbus_transaction_end(xbt, 1);
1758 xenbus_dev_fatal(info->xbdev, err, "%s", message);
1763 static void free_info(struct blkfront_info *info)
1765 list_del(&info->info_list);
1769 /* Common code used when first setting up, and when resuming. */
1770 static int talk_to_blkback(struct xenbus_device *dev,
1771 struct blkfront_info *info)
1773 const char *message = NULL;
1774 struct xenbus_transaction xbt;
1776 unsigned int i, max_page_order;
1777 unsigned int ring_page_order;
1782 max_page_order = xenbus_read_unsigned(info->xbdev->otherend,
1783 "max-ring-page-order", 0);
1784 ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1785 info->nr_ring_pages = 1 << ring_page_order;
1787 err = negotiate_mq(info);
1789 goto destroy_blkring;
1791 for (i = 0; i < info->nr_rings; i++) {
1792 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1794 /* Create shared ring, alloc event channel. */
1795 err = setup_blkring(dev, rinfo);
1797 goto destroy_blkring;
1801 err = xenbus_transaction_start(&xbt);
1803 xenbus_dev_fatal(dev, err, "starting transaction");
1804 goto destroy_blkring;
1807 if (info->nr_ring_pages > 1) {
1808 err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1811 message = "writing ring-page-order";
1812 goto abort_transaction;
1816 /* We already got the number of queues/rings in _probe */
1817 if (info->nr_rings == 1) {
1818 err = write_per_ring_nodes(xbt, &info->rinfo[0], dev->nodename);
1820 goto destroy_blkring;
1825 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1828 message = "writing multi-queue-num-queues";
1829 goto abort_transaction;
1832 pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1833 path = kmalloc(pathsize, GFP_KERNEL);
1836 message = "ENOMEM while writing ring references";
1837 goto abort_transaction;
1840 for (i = 0; i < info->nr_rings; i++) {
1841 memset(path, 0, pathsize);
1842 snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1843 err = write_per_ring_nodes(xbt, &info->rinfo[i], path);
1846 goto destroy_blkring;
1851 err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1852 XEN_IO_PROTO_ABI_NATIVE);
1854 message = "writing protocol";
1855 goto abort_transaction;
1857 err = xenbus_printf(xbt, dev->nodename,
1858 "feature-persistent", "%u", 1);
1861 "writing persistent grants feature to xenbus");
1863 err = xenbus_transaction_end(xbt, 0);
1867 xenbus_dev_fatal(dev, err, "completing transaction");
1868 goto destroy_blkring;
1871 for (i = 0; i < info->nr_rings; i++) {
1873 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1875 for (j = 0; j < BLK_RING_SIZE(info); j++)
1876 rinfo->shadow[j].req.u.rw.id = j + 1;
1877 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1879 xenbus_switch_state(dev, XenbusStateInitialised);
1884 xenbus_transaction_end(xbt, 1);
1886 xenbus_dev_fatal(dev, err, "%s", message);
1888 blkif_free(info, 0);
1890 mutex_lock(&blkfront_mutex);
1892 mutex_unlock(&blkfront_mutex);
1894 dev_set_drvdata(&dev->dev, NULL);
1899 static int negotiate_mq(struct blkfront_info *info)
1901 unsigned int backend_max_queues;
1904 BUG_ON(info->nr_rings);
1906 /* Check if backend supports multiple queues. */
1907 backend_max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1908 "multi-queue-max-queues", 1);
1909 info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1910 /* We need at least one ring. */
1911 if (!info->nr_rings)
1914 info->rinfo = kvcalloc(info->nr_rings,
1915 struct_size(info->rinfo, shadow,
1916 BLK_RING_SIZE(info)),
1919 xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1924 for (i = 0; i < info->nr_rings; i++) {
1925 struct blkfront_ring_info *rinfo;
1927 rinfo = &info->rinfo[i];
1928 INIT_LIST_HEAD(&rinfo->indirect_pages);
1929 INIT_LIST_HEAD(&rinfo->grants);
1930 rinfo->dev_info = info;
1931 INIT_WORK(&rinfo->work, blkif_restart_queue);
1932 spin_lock_init(&rinfo->ring_lock);
1937 * Entry point to this code when a new device is created. Allocate the basic
1938 * structures and the ring buffer for communication with the backend, and
1939 * inform the backend of the appropriate details for those. Switch to
1940 * Initialised state.
1942 static int blkfront_probe(struct xenbus_device *dev,
1943 const struct xenbus_device_id *id)
1946 struct blkfront_info *info;
1948 /* FIXME: Use dynamic device id if this is not set. */
1949 err = xenbus_scanf(XBT_NIL, dev->nodename,
1950 "virtual-device", "%i", &vdevice);
1952 /* go looking in the extended area instead */
1953 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1956 xenbus_dev_fatal(dev, err, "reading virtual-device");
1961 if (xen_hvm_domain()) {
1964 /* no unplug has been done: do not hook devices != xen vbds */
1965 if (xen_has_pv_and_legacy_disk_devices()) {
1968 if (!VDEV_IS_EXTENDED(vdevice))
1969 major = BLKIF_MAJOR(vdevice);
1971 major = XENVBD_MAJOR;
1973 if (major != XENVBD_MAJOR) {
1975 "%s: HVM does not support vbd %d as xen block device\n",
1980 /* do not create a PV cdrom device if we are an HVM guest */
1981 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1984 if (strncmp(type, "cdrom", 5) == 0) {
1990 info = kzalloc(sizeof(*info), GFP_KERNEL);
1992 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1998 mutex_init(&info->mutex);
1999 info->vdevice = vdevice;
2000 info->connected = BLKIF_STATE_DISCONNECTED;
2002 /* Front end dir is a number, which is used as the id. */
2003 info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
2004 dev_set_drvdata(&dev->dev, info);
2006 mutex_lock(&blkfront_mutex);
2007 list_add(&info->info_list, &info_list);
2008 mutex_unlock(&blkfront_mutex);
2013 static int blkif_recover(struct blkfront_info *info)
2015 unsigned int r_index;
2016 struct request *req, *n;
2021 blkfront_gather_backend_features(info);
2022 /* Reset limits changed by blk_mq_update_nr_hw_queues(). */
2023 blkif_set_queue_limits(info);
2024 segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
2025 blk_queue_max_segments(info->rq, segs / GRANTS_PER_PSEG);
2027 for (r_index = 0; r_index < info->nr_rings; r_index++) {
2028 struct blkfront_ring_info *rinfo = &info->rinfo[r_index];
2030 rc = blkfront_setup_indirect(rinfo);
2034 xenbus_switch_state(info->xbdev, XenbusStateConnected);
2036 /* Now safe for us to use the shared ring */
2037 info->connected = BLKIF_STATE_CONNECTED;
2039 for (r_index = 0; r_index < info->nr_rings; r_index++) {
2040 struct blkfront_ring_info *rinfo;
2042 rinfo = &info->rinfo[r_index];
2043 /* Kick any other new requests queued since we resumed */
2044 kick_pending_request_queues(rinfo);
2047 list_for_each_entry_safe(req, n, &info->requests, queuelist) {
2048 /* Requeue pending requests (flush or discard) */
2049 list_del_init(&req->queuelist);
2050 BUG_ON(req->nr_phys_segments > segs);
2051 blk_mq_requeue_request(req, false);
2053 blk_mq_start_stopped_hw_queues(info->rq, true);
2054 blk_mq_kick_requeue_list(info->rq);
2056 while ((bio = bio_list_pop(&info->bio_list)) != NULL) {
2057 /* Traverse the list of pending bios and re-queue them */
2065 * We are reconnecting to the backend, due to a suspend/resume, or a backend
2066 * driver restart. We tear down our blkif structure and recreate it, but
2067 * leave the device-layer structures intact so that this is transparent to the
2068 * rest of the kernel.
2070 static int blkfront_resume(struct xenbus_device *dev)
2072 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2076 dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2078 bio_list_init(&info->bio_list);
2079 INIT_LIST_HEAD(&info->requests);
2080 for (i = 0; i < info->nr_rings; i++) {
2081 struct blkfront_ring_info *rinfo = &info->rinfo[i];
2082 struct bio_list merge_bio;
2083 struct blk_shadow *shadow = rinfo->shadow;
2085 for (j = 0; j < BLK_RING_SIZE(info); j++) {
2087 if (!shadow[j].request)
2091 * Get the bios in the request so we can re-queue them.
2093 if (req_op(shadow[j].request) == REQ_OP_FLUSH ||
2094 req_op(shadow[j].request) == REQ_OP_DISCARD ||
2095 req_op(shadow[j].request) == REQ_OP_SECURE_ERASE ||
2096 shadow[j].request->cmd_flags & REQ_FUA) {
2098 * Flush operations don't contain bios, so
2099 * we need to requeue the whole request
2101 * XXX: but this doesn't make any sense for a
2102 * write with the FUA flag set..
2104 list_add(&shadow[j].request->queuelist, &info->requests);
2107 merge_bio.head = shadow[j].request->bio;
2108 merge_bio.tail = shadow[j].request->biotail;
2109 bio_list_merge(&info->bio_list, &merge_bio);
2110 shadow[j].request->bio = NULL;
2111 blk_mq_end_request(shadow[j].request, BLK_STS_OK);
2115 blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2117 err = talk_to_blkback(dev, info);
2119 blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings);
2122 * We have to wait for the backend to switch to
2123 * connected state, since we want to read which
2124 * features it supports.
2130 static void blkfront_closing(struct blkfront_info *info)
2132 struct xenbus_device *xbdev = info->xbdev;
2133 struct block_device *bdev = NULL;
2135 mutex_lock(&info->mutex);
2137 if (xbdev->state == XenbusStateClosing) {
2138 mutex_unlock(&info->mutex);
2143 bdev = bdget_disk(info->gd, 0);
2145 mutex_unlock(&info->mutex);
2148 xenbus_frontend_closed(xbdev);
2152 mutex_lock(&bdev->bd_mutex);
2154 if (bdev->bd_openers) {
2155 xenbus_dev_error(xbdev, -EBUSY,
2156 "Device in use; refusing to close");
2157 xenbus_switch_state(xbdev, XenbusStateClosing);
2159 xlvbd_release_gendisk(info);
2160 xenbus_frontend_closed(xbdev);
2163 mutex_unlock(&bdev->bd_mutex);
2167 static void blkfront_setup_discard(struct blkfront_info *info)
2170 unsigned int discard_granularity;
2171 unsigned int discard_alignment;
2173 info->feature_discard = 1;
2174 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2175 "discard-granularity", "%u", &discard_granularity,
2176 "discard-alignment", "%u", &discard_alignment,
2179 info->discard_granularity = discard_granularity;
2180 info->discard_alignment = discard_alignment;
2182 info->feature_secdiscard =
2183 !!xenbus_read_unsigned(info->xbdev->otherend, "discard-secure",
2187 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2189 unsigned int psegs, grants;
2191 struct blkfront_info *info = rinfo->dev_info;
2193 if (info->max_indirect_segments == 0) {
2195 grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2198 * When an extra req is required, the maximum
2199 * grants supported is related to the size of the
2200 * Linux block segment.
2202 grants = GRANTS_PER_PSEG;
2206 grants = info->max_indirect_segments;
2207 psegs = DIV_ROUND_UP(grants, GRANTS_PER_PSEG);
2209 err = fill_grant_buffer(rinfo,
2210 (grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2214 if (!info->feature_persistent && info->max_indirect_segments) {
2216 * We are using indirect descriptors but not persistent
2217 * grants, we need to allocate a set of pages that can be
2218 * used for mapping indirect grefs
2220 int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2222 BUG_ON(!list_empty(&rinfo->indirect_pages));
2223 for (i = 0; i < num; i++) {
2224 struct page *indirect_page = alloc_page(GFP_NOIO);
2227 list_add(&indirect_page->lru, &rinfo->indirect_pages);
2231 for (i = 0; i < BLK_RING_SIZE(info); i++) {
2232 rinfo->shadow[i].grants_used =
2234 sizeof(rinfo->shadow[i].grants_used[0]),
2236 rinfo->shadow[i].sg = kvcalloc(psegs,
2237 sizeof(rinfo->shadow[i].sg[0]),
2239 if (info->max_indirect_segments)
2240 rinfo->shadow[i].indirect_grants =
2241 kvcalloc(INDIRECT_GREFS(grants),
2242 sizeof(rinfo->shadow[i].indirect_grants[0]),
2244 if ((rinfo->shadow[i].grants_used == NULL) ||
2245 (rinfo->shadow[i].sg == NULL) ||
2246 (info->max_indirect_segments &&
2247 (rinfo->shadow[i].indirect_grants == NULL)))
2249 sg_init_table(rinfo->shadow[i].sg, psegs);
2256 for (i = 0; i < BLK_RING_SIZE(info); i++) {
2257 kvfree(rinfo->shadow[i].grants_used);
2258 rinfo->shadow[i].grants_used = NULL;
2259 kvfree(rinfo->shadow[i].sg);
2260 rinfo->shadow[i].sg = NULL;
2261 kvfree(rinfo->shadow[i].indirect_grants);
2262 rinfo->shadow[i].indirect_grants = NULL;
2264 if (!list_empty(&rinfo->indirect_pages)) {
2265 struct page *indirect_page, *n;
2266 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2267 list_del(&indirect_page->lru);
2268 __free_page(indirect_page);
2275 * Gather all backend feature-*
2277 static void blkfront_gather_backend_features(struct blkfront_info *info)
2279 unsigned int indirect_segments;
2281 info->feature_flush = 0;
2282 info->feature_fua = 0;
2285 * If there's no "feature-barrier" defined, then it means
2286 * we're dealing with a very old backend which writes
2287 * synchronously; nothing to do.
2289 * If there are barriers, then we use flush.
2291 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-barrier", 0)) {
2292 info->feature_flush = 1;
2293 info->feature_fua = 1;
2297 * And if there is "feature-flush-cache" use that above
2300 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-flush-cache",
2302 info->feature_flush = 1;
2303 info->feature_fua = 0;
2306 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-discard", 0))
2307 blkfront_setup_discard(info);
2309 info->feature_persistent =
2310 !!xenbus_read_unsigned(info->xbdev->otherend,
2311 "feature-persistent", 0);
2313 indirect_segments = xenbus_read_unsigned(info->xbdev->otherend,
2314 "feature-max-indirect-segments", 0);
2315 if (indirect_segments > xen_blkif_max_segments)
2316 indirect_segments = xen_blkif_max_segments;
2317 if (indirect_segments <= BLKIF_MAX_SEGMENTS_PER_REQUEST)
2318 indirect_segments = 0;
2319 info->max_indirect_segments = indirect_segments;
2321 if (info->feature_persistent) {
2322 mutex_lock(&blkfront_mutex);
2323 schedule_delayed_work(&blkfront_work, HZ * 10);
2324 mutex_unlock(&blkfront_mutex);
2329 * Invoked when the backend is finally 'ready' (and has told produced
2330 * the details about the physical device - #sectors, size, etc).
2332 static void blkfront_connect(struct blkfront_info *info)
2334 unsigned long long sectors;
2335 unsigned long sector_size;
2336 unsigned int physical_sector_size;
2338 char *envp[] = { "RESIZE=1", NULL };
2341 switch (info->connected) {
2342 case BLKIF_STATE_CONNECTED:
2344 * Potentially, the back-end may be signalling
2345 * a capacity change; update the capacity.
2347 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2348 "sectors", "%Lu", §ors);
2349 if (XENBUS_EXIST_ERR(err))
2351 printk(KERN_INFO "Setting capacity to %Lu\n",
2353 set_capacity(info->gd, sectors);
2354 revalidate_disk(info->gd);
2355 kobject_uevent_env(&disk_to_dev(info->gd)->kobj,
2359 case BLKIF_STATE_SUSPENDED:
2361 * If we are recovering from suspension, we need to wait
2362 * for the backend to announce it's features before
2363 * reconnecting, at least we need to know if the backend
2364 * supports indirect descriptors, and how many.
2366 blkif_recover(info);
2373 dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2374 __func__, info->xbdev->otherend);
2376 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2377 "sectors", "%llu", §ors,
2378 "info", "%u", &binfo,
2379 "sector-size", "%lu", §or_size,
2382 xenbus_dev_fatal(info->xbdev, err,
2383 "reading backend fields at %s",
2384 info->xbdev->otherend);
2389 * physcial-sector-size is a newer field, so old backends may not
2390 * provide this. Assume physical sector size to be the same as
2391 * sector_size in that case.
2393 physical_sector_size = xenbus_read_unsigned(info->xbdev->otherend,
2394 "physical-sector-size",
2396 blkfront_gather_backend_features(info);
2397 for (i = 0; i < info->nr_rings; i++) {
2398 err = blkfront_setup_indirect(&info->rinfo[i]);
2400 xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2401 info->xbdev->otherend);
2402 blkif_free(info, 0);
2407 err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
2408 physical_sector_size);
2410 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2411 info->xbdev->otherend);
2415 xenbus_switch_state(info->xbdev, XenbusStateConnected);
2417 /* Kick pending requests. */
2418 info->connected = BLKIF_STATE_CONNECTED;
2419 for (i = 0; i < info->nr_rings; i++)
2420 kick_pending_request_queues(&info->rinfo[i]);
2422 device_add_disk(&info->xbdev->dev, info->gd, NULL);
2428 blkif_free(info, 0);
2433 * Callback received when the backend's state changes.
2435 static void blkback_changed(struct xenbus_device *dev,
2436 enum xenbus_state backend_state)
2438 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2440 dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2442 switch (backend_state) {
2443 case XenbusStateInitWait:
2444 if (dev->state != XenbusStateInitialising)
2446 if (talk_to_blkback(dev, info))
2448 case XenbusStateInitialising:
2449 case XenbusStateInitialised:
2450 case XenbusStateReconfiguring:
2451 case XenbusStateReconfigured:
2452 case XenbusStateUnknown:
2455 case XenbusStateConnected:
2457 * talk_to_blkback sets state to XenbusStateInitialised
2458 * and blkfront_connect sets it to XenbusStateConnected
2459 * (if connection went OK).
2461 * If the backend (or toolstack) decides to poke at backend
2462 * state (and re-trigger the watch by setting the state repeatedly
2463 * to XenbusStateConnected (4)) we need to deal with this.
2464 * This is allowed as this is used to communicate to the guest
2465 * that the size of disk has changed!
2467 if ((dev->state != XenbusStateInitialised) &&
2468 (dev->state != XenbusStateConnected)) {
2469 if (talk_to_blkback(dev, info))
2473 blkfront_connect(info);
2476 case XenbusStateClosed:
2477 if (dev->state == XenbusStateClosed)
2480 case XenbusStateClosing:
2482 blkfront_closing(info);
2487 static int blkfront_remove(struct xenbus_device *xbdev)
2489 struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2490 struct block_device *bdev = NULL;
2491 struct gendisk *disk;
2493 dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2498 blkif_free(info, 0);
2500 mutex_lock(&info->mutex);
2504 bdev = bdget_disk(disk, 0);
2507 mutex_unlock(&info->mutex);
2510 mutex_lock(&blkfront_mutex);
2512 mutex_unlock(&blkfront_mutex);
2517 * The xbdev was removed before we reached the Closed
2518 * state. See if it's safe to remove the disk. If the bdev
2519 * isn't closed yet, we let release take care of it.
2522 mutex_lock(&bdev->bd_mutex);
2523 info = disk->private_data;
2525 dev_warn(disk_to_dev(disk),
2526 "%s was hot-unplugged, %d stale handles\n",
2527 xbdev->nodename, bdev->bd_openers);
2529 if (info && !bdev->bd_openers) {
2530 xlvbd_release_gendisk(info);
2531 disk->private_data = NULL;
2532 mutex_lock(&blkfront_mutex);
2534 mutex_unlock(&blkfront_mutex);
2537 mutex_unlock(&bdev->bd_mutex);
2543 static int blkfront_is_ready(struct xenbus_device *dev)
2545 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2547 return info->is_ready && info->xbdev;
2550 static int blkif_open(struct block_device *bdev, fmode_t mode)
2552 struct gendisk *disk = bdev->bd_disk;
2553 struct blkfront_info *info;
2556 mutex_lock(&blkfront_mutex);
2558 info = disk->private_data;
2565 mutex_lock(&info->mutex);
2568 /* xbdev is closed */
2571 mutex_unlock(&info->mutex);
2574 mutex_unlock(&blkfront_mutex);
2578 static void blkif_release(struct gendisk *disk, fmode_t mode)
2580 struct blkfront_info *info = disk->private_data;
2581 struct block_device *bdev;
2582 struct xenbus_device *xbdev;
2584 mutex_lock(&blkfront_mutex);
2586 bdev = bdget_disk(disk, 0);
2589 WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2592 if (bdev->bd_openers)
2596 * Check if we have been instructed to close. We will have
2597 * deferred this request, because the bdev was still open.
2600 mutex_lock(&info->mutex);
2601 xbdev = info->xbdev;
2603 if (xbdev && xbdev->state == XenbusStateClosing) {
2604 /* pending switch to state closed */
2605 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2606 xlvbd_release_gendisk(info);
2607 xenbus_frontend_closed(info->xbdev);
2610 mutex_unlock(&info->mutex);
2613 /* sudden device removal */
2614 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2615 xlvbd_release_gendisk(info);
2616 disk->private_data = NULL;
2623 mutex_unlock(&blkfront_mutex);
2626 static const struct block_device_operations xlvbd_block_fops =
2628 .owner = THIS_MODULE,
2630 .release = blkif_release,
2631 .getgeo = blkif_getgeo,
2632 .ioctl = blkif_ioctl,
2633 .compat_ioctl = blkdev_compat_ptr_ioctl,
2637 static const struct xenbus_device_id blkfront_ids[] = {
2642 static struct xenbus_driver blkfront_driver = {
2643 .ids = blkfront_ids,
2644 .probe = blkfront_probe,
2645 .remove = blkfront_remove,
2646 .resume = blkfront_resume,
2647 .otherend_changed = blkback_changed,
2648 .is_ready = blkfront_is_ready,
2651 static void purge_persistent_grants(struct blkfront_info *info)
2654 unsigned long flags;
2656 for (i = 0; i < info->nr_rings; i++) {
2657 struct blkfront_ring_info *rinfo = &info->rinfo[i];
2658 struct grant *gnt_list_entry, *tmp;
2660 spin_lock_irqsave(&rinfo->ring_lock, flags);
2662 if (rinfo->persistent_gnts_c == 0) {
2663 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2667 list_for_each_entry_safe(gnt_list_entry, tmp, &rinfo->grants,
2669 if (gnt_list_entry->gref == GRANT_INVALID_REF ||
2670 gnttab_query_foreign_access(gnt_list_entry->gref))
2673 list_del(&gnt_list_entry->node);
2674 gnttab_end_foreign_access(gnt_list_entry->gref, 0, 0UL);
2675 rinfo->persistent_gnts_c--;
2676 gnt_list_entry->gref = GRANT_INVALID_REF;
2677 list_add_tail(&gnt_list_entry->node, &rinfo->grants);
2680 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2684 static void blkfront_delay_work(struct work_struct *work)
2686 struct blkfront_info *info;
2687 bool need_schedule_work = false;
2689 mutex_lock(&blkfront_mutex);
2691 list_for_each_entry(info, &info_list, info_list) {
2692 if (info->feature_persistent) {
2693 need_schedule_work = true;
2694 mutex_lock(&info->mutex);
2695 purge_persistent_grants(info);
2696 mutex_unlock(&info->mutex);
2700 if (need_schedule_work)
2701 schedule_delayed_work(&blkfront_work, HZ * 10);
2703 mutex_unlock(&blkfront_mutex);
2706 static int __init xlblk_init(void)
2709 int nr_cpus = num_online_cpus();
2714 if (!xen_has_pv_disk_devices())
2717 if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2718 pr_warn("xen_blk: can't get major %d with name %s\n",
2719 XENVBD_MAJOR, DEV_NAME);
2723 if (xen_blkif_max_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST)
2724 xen_blkif_max_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2726 if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2727 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2728 xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2729 xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2732 if (xen_blkif_max_queues > nr_cpus) {
2733 pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2734 xen_blkif_max_queues, nr_cpus);
2735 xen_blkif_max_queues = nr_cpus;
2738 INIT_DELAYED_WORK(&blkfront_work, blkfront_delay_work);
2740 ret = xenbus_register_frontend(&blkfront_driver);
2742 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2748 module_init(xlblk_init);
2751 static void __exit xlblk_exit(void)
2753 cancel_delayed_work_sync(&blkfront_work);
2755 xenbus_unregister_driver(&blkfront_driver);
2756 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2759 module_exit(xlblk_exit);
2761 MODULE_DESCRIPTION("Xen virtual block device frontend");
2762 MODULE_LICENSE("GPL");
2763 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2764 MODULE_ALIAS("xen:vbd");
2765 MODULE_ALIAS("xenblk");