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/major.h>
46 #include <linux/mutex.h>
47 #include <linux/scatterlist.h>
48 #include <linux/bitmap.h>
49 #include <linux/list.h>
50 #include <linux/workqueue.h>
51 #include <linux/sched/mm.h>
54 #include <xen/xenbus.h>
55 #include <xen/grant_table.h>
56 #include <xen/events.h>
58 #include <xen/platform_pci.h>
60 #include <xen/interface/grant_table.h>
61 #include <xen/interface/io/blkif.h>
62 #include <xen/interface/io/protocols.h>
64 #include <asm/xen/hypervisor.h>
67 * The minimal size of segment supported by the block framework is PAGE_SIZE.
68 * When Linux is using a different page size than Xen, it may not be possible
69 * to put all the data in a single segment.
70 * This can happen when the backend doesn't support indirect descriptor and
71 * therefore the maximum amount of data that a request can carry is
72 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE = 44KB
74 * Note that we only support one extra request. So the Linux page size
75 * should be <= ( 2 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) =
78 #define HAS_EXTRA_REQ (BLKIF_MAX_SEGMENTS_PER_REQUEST < XEN_PFN_PER_PAGE)
81 BLKIF_STATE_DISCONNECTED,
82 BLKIF_STATE_CONNECTED,
83 BLKIF_STATE_SUSPENDED,
90 struct list_head node;
102 struct blkif_request req;
103 struct request *request;
104 struct grant **grants_used;
105 struct grant **indirect_grants;
106 struct scatterlist *sg;
108 enum blk_req_status status;
110 #define NO_ASSOCIATED_ID ~0UL
112 * Id of the sibling if we ever need 2 requests when handling a
115 unsigned long associated_id;
122 static inline struct blkif_req *blkif_req(struct request *rq)
124 return blk_mq_rq_to_pdu(rq);
127 static DEFINE_MUTEX(blkfront_mutex);
128 static const struct block_device_operations xlvbd_block_fops;
129 static struct delayed_work blkfront_work;
130 static LIST_HEAD(info_list);
133 * Maximum number of segments in indirect requests, the actual value used by
134 * the frontend driver is the minimum of this value and the value provided
135 * by the backend driver.
138 static unsigned int xen_blkif_max_segments = 32;
139 module_param_named(max_indirect_segments, xen_blkif_max_segments, uint, 0444);
140 MODULE_PARM_DESC(max_indirect_segments,
141 "Maximum amount of segments in indirect requests (default is 32)");
143 static unsigned int xen_blkif_max_queues = 4;
144 module_param_named(max_queues, xen_blkif_max_queues, uint, 0444);
145 MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
148 * Maximum order of pages to be used for the shared ring between front and
149 * backend, 4KB page granularity is used.
151 static unsigned int xen_blkif_max_ring_order;
152 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, 0444);
153 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
155 #define BLK_RING_SIZE(info) \
156 __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
159 * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
160 * characters are enough. Define to 20 to keep consistent with backend.
162 #define RINGREF_NAME_LEN (20)
164 * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
166 #define QUEUE_NAME_LEN (17)
170 * Every blkfront device can associate with one or more blkfront_ring_info,
171 * depending on how many hardware queues/rings to be used.
173 struct blkfront_ring_info {
174 /* Lock to protect data in every ring buffer. */
175 spinlock_t ring_lock;
176 struct blkif_front_ring ring;
177 unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
178 unsigned int evtchn, irq;
179 struct work_struct work;
180 struct gnttab_free_callback callback;
181 struct list_head indirect_pages;
182 struct list_head grants;
183 unsigned int persistent_gnts_c;
184 unsigned long shadow_free;
185 struct blkfront_info *dev_info;
186 struct blk_shadow shadow[];
190 * We have one of these per vbd, whether ide, scsi or 'other'. They
191 * hang in private_data off the gendisk structure. We may end up
192 * putting all kinds of interesting stuff here :-)
197 struct xenbus_device *xbdev;
200 unsigned int physical_sector_size;
201 unsigned long vdisk_info;
204 enum blkif_state connected;
205 /* Number of pages per ring buffer. */
206 unsigned int nr_ring_pages;
207 struct request_queue *rq;
208 unsigned int feature_flush:1;
209 unsigned int feature_fua:1;
210 unsigned int feature_discard:1;
211 unsigned int feature_secdiscard:1;
212 unsigned int feature_persistent:1;
213 unsigned int discard_granularity;
214 unsigned int discard_alignment;
215 /* Number of 4KB segments handled */
216 unsigned int max_indirect_segments;
218 struct blk_mq_tag_set tag_set;
219 struct blkfront_ring_info *rinfo;
220 unsigned int nr_rings;
221 unsigned int rinfo_size;
222 /* Save uncomplete reqs and bios for migration. */
223 struct list_head requests;
224 struct bio_list bio_list;
225 struct list_head info_list;
228 static unsigned int nr_minors;
229 static unsigned long *minors;
230 static DEFINE_SPINLOCK(minor_lock);
232 #define GRANT_INVALID_REF 0
234 #define PARTS_PER_DISK 16
235 #define PARTS_PER_EXT_DISK 256
237 #define BLKIF_MAJOR(dev) ((dev)>>8)
238 #define BLKIF_MINOR(dev) ((dev) & 0xff)
241 #define EXTENDED (1<<EXT_SHIFT)
242 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
243 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
244 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
245 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
246 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
247 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
249 #define DEV_NAME "xvd" /* name in /dev */
252 * Grants are always the same size as a Xen page (i.e 4KB).
253 * A physical segment is always the same size as a Linux page.
254 * Number of grants per physical segment
256 #define GRANTS_PER_PSEG (PAGE_SIZE / XEN_PAGE_SIZE)
258 #define GRANTS_PER_INDIRECT_FRAME \
259 (XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
261 #define INDIRECT_GREFS(_grants) \
262 DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
264 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
265 static void blkfront_gather_backend_features(struct blkfront_info *info);
266 static int negotiate_mq(struct blkfront_info *info);
268 #define for_each_rinfo(info, ptr, idx) \
269 for ((ptr) = (info)->rinfo, (idx) = 0; \
270 (idx) < (info)->nr_rings; \
271 (idx)++, (ptr) = (void *)(ptr) + (info)->rinfo_size)
273 static inline struct blkfront_ring_info *
274 get_rinfo(const struct blkfront_info *info, unsigned int i)
276 BUG_ON(i >= info->nr_rings);
277 return (void *)info->rinfo + i * info->rinfo_size;
280 static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
282 unsigned long free = rinfo->shadow_free;
284 BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
285 rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
286 rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
290 static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
293 if (rinfo->shadow[id].req.u.rw.id != id)
295 if (rinfo->shadow[id].request == NULL)
297 rinfo->shadow[id].req.u.rw.id = rinfo->shadow_free;
298 rinfo->shadow[id].request = NULL;
299 rinfo->shadow_free = id;
303 static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
305 struct blkfront_info *info = rinfo->dev_info;
306 struct page *granted_page;
307 struct grant *gnt_list_entry, *n;
311 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
315 if (info->feature_persistent) {
316 granted_page = alloc_page(GFP_NOIO);
318 kfree(gnt_list_entry);
321 gnt_list_entry->page = granted_page;
324 gnt_list_entry->gref = GRANT_INVALID_REF;
325 list_add(&gnt_list_entry->node, &rinfo->grants);
332 list_for_each_entry_safe(gnt_list_entry, n,
333 &rinfo->grants, node) {
334 list_del(&gnt_list_entry->node);
335 if (info->feature_persistent)
336 __free_page(gnt_list_entry->page);
337 kfree(gnt_list_entry);
344 static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
346 struct grant *gnt_list_entry;
348 BUG_ON(list_empty(&rinfo->grants));
349 gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
351 list_del(&gnt_list_entry->node);
353 if (gnt_list_entry->gref != GRANT_INVALID_REF)
354 rinfo->persistent_gnts_c--;
356 return gnt_list_entry;
359 static inline void grant_foreign_access(const struct grant *gnt_list_entry,
360 const struct blkfront_info *info)
362 gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
363 info->xbdev->otherend_id,
364 gnt_list_entry->page,
368 static struct grant *get_grant(grant_ref_t *gref_head,
370 struct blkfront_ring_info *rinfo)
372 struct grant *gnt_list_entry = get_free_grant(rinfo);
373 struct blkfront_info *info = rinfo->dev_info;
375 if (gnt_list_entry->gref != GRANT_INVALID_REF)
376 return gnt_list_entry;
378 /* Assign a gref to this page */
379 gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
380 BUG_ON(gnt_list_entry->gref == -ENOSPC);
381 if (info->feature_persistent)
382 grant_foreign_access(gnt_list_entry, info);
384 /* Grant access to the GFN passed by the caller */
385 gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
386 info->xbdev->otherend_id,
390 return gnt_list_entry;
393 static struct grant *get_indirect_grant(grant_ref_t *gref_head,
394 struct blkfront_ring_info *rinfo)
396 struct grant *gnt_list_entry = get_free_grant(rinfo);
397 struct blkfront_info *info = rinfo->dev_info;
399 if (gnt_list_entry->gref != GRANT_INVALID_REF)
400 return gnt_list_entry;
402 /* Assign a gref to this page */
403 gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
404 BUG_ON(gnt_list_entry->gref == -ENOSPC);
405 if (!info->feature_persistent) {
406 struct page *indirect_page;
408 /* Fetch a pre-allocated page to use for indirect grefs */
409 BUG_ON(list_empty(&rinfo->indirect_pages));
410 indirect_page = list_first_entry(&rinfo->indirect_pages,
412 list_del(&indirect_page->lru);
413 gnt_list_entry->page = indirect_page;
415 grant_foreign_access(gnt_list_entry, info);
417 return gnt_list_entry;
420 static const char *op_name(int op)
422 static const char *const names[] = {
423 [BLKIF_OP_READ] = "read",
424 [BLKIF_OP_WRITE] = "write",
425 [BLKIF_OP_WRITE_BARRIER] = "barrier",
426 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
427 [BLKIF_OP_DISCARD] = "discard" };
429 if (op < 0 || op >= ARRAY_SIZE(names))
437 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
439 unsigned int end = minor + nr;
442 if (end > nr_minors) {
443 unsigned long *bitmap, *old;
445 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
450 spin_lock(&minor_lock);
451 if (end > nr_minors) {
453 memcpy(bitmap, minors,
454 BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
456 nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
459 spin_unlock(&minor_lock);
463 spin_lock(&minor_lock);
464 if (find_next_bit(minors, end, minor) >= end) {
465 bitmap_set(minors, minor, nr);
469 spin_unlock(&minor_lock);
474 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
476 unsigned int end = minor + nr;
478 BUG_ON(end > nr_minors);
479 spin_lock(&minor_lock);
480 bitmap_clear(minors, minor, nr);
481 spin_unlock(&minor_lock);
484 static void blkif_restart_queue_callback(void *arg)
486 struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
487 schedule_work(&rinfo->work);
490 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
492 /* We don't have real geometry info, but let's at least return
493 values consistent with the size of the device */
494 sector_t nsect = get_capacity(bd->bd_disk);
495 sector_t cylinders = nsect;
499 sector_div(cylinders, hg->heads * hg->sectors);
500 hg->cylinders = cylinders;
501 if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
502 hg->cylinders = 0xffff;
506 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
507 unsigned command, unsigned long argument)
509 struct blkfront_info *info = bdev->bd_disk->private_data;
513 case CDROMMULTISESSION:
514 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
515 if (put_user(0, (char __user *)(argument + i)))
518 case CDROM_GET_CAPABILITY:
519 if (!(info->vdisk_info & VDISK_CDROM))
527 static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
529 struct blkif_request **ring_req)
533 *ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
534 rinfo->ring.req_prod_pvt++;
536 id = get_id_from_freelist(rinfo);
537 rinfo->shadow[id].request = req;
538 rinfo->shadow[id].status = REQ_PROCESSING;
539 rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
541 rinfo->shadow[id].req.u.rw.id = id;
546 static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
548 struct blkfront_info *info = rinfo->dev_info;
549 struct blkif_request *ring_req, *final_ring_req;
552 /* Fill out a communications ring structure. */
553 id = blkif_ring_get_request(rinfo, req, &final_ring_req);
554 ring_req = &rinfo->shadow[id].req;
556 ring_req->operation = BLKIF_OP_DISCARD;
557 ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
558 ring_req->u.discard.id = id;
559 ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
560 if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard)
561 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
563 ring_req->u.discard.flag = 0;
565 /* Copy the request to the ring page. */
566 *final_ring_req = *ring_req;
567 rinfo->shadow[id].status = REQ_WAITING;
572 struct setup_rw_req {
573 unsigned int grant_idx;
574 struct blkif_request_segment *segments;
575 struct blkfront_ring_info *rinfo;
576 struct blkif_request *ring_req;
577 grant_ref_t gref_head;
579 /* Only used when persistent grant is used and it's a write request */
581 unsigned int bvec_off;
584 bool require_extra_req;
585 struct blkif_request *extra_ring_req;
588 static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
589 unsigned int len, void *data)
591 struct setup_rw_req *setup = data;
593 struct grant *gnt_list_entry;
594 unsigned int fsect, lsect;
595 /* Convenient aliases */
596 unsigned int grant_idx = setup->grant_idx;
597 struct blkif_request *ring_req = setup->ring_req;
598 struct blkfront_ring_info *rinfo = setup->rinfo;
600 * We always use the shadow of the first request to store the list
601 * of grant associated to the block I/O request. This made the
602 * completion more easy to handle even if the block I/O request is
605 struct blk_shadow *shadow = &rinfo->shadow[setup->id];
607 if (unlikely(setup->require_extra_req &&
608 grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
610 * We are using the second request, setup grant_idx
611 * to be the index of the segment array.
613 grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
614 ring_req = setup->extra_ring_req;
617 if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
618 (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
620 kunmap_atomic(setup->segments);
622 n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
623 gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
624 shadow->indirect_grants[n] = gnt_list_entry;
625 setup->segments = kmap_atomic(gnt_list_entry->page);
626 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
629 gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
630 ref = gnt_list_entry->gref;
632 * All the grants are stored in the shadow of the first
633 * request. Therefore we have to use the global index.
635 shadow->grants_used[setup->grant_idx] = gnt_list_entry;
637 if (setup->need_copy) {
640 shared_data = kmap_atomic(gnt_list_entry->page);
642 * this does not wipe data stored outside the
643 * range sg->offset..sg->offset+sg->length.
644 * Therefore, blkback *could* see data from
645 * previous requests. This is OK as long as
646 * persistent grants are shared with just one
647 * domain. It may need refactoring if this
650 memcpy(shared_data + offset,
651 setup->bvec_data + setup->bvec_off,
654 kunmap_atomic(shared_data);
655 setup->bvec_off += len;
659 lsect = fsect + (len >> 9) - 1;
660 if (ring_req->operation != BLKIF_OP_INDIRECT) {
661 ring_req->u.rw.seg[grant_idx] =
662 (struct blkif_request_segment) {
665 .last_sect = lsect };
667 setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
668 (struct blkif_request_segment) {
671 .last_sect = lsect };
674 (setup->grant_idx)++;
677 static void blkif_setup_extra_req(struct blkif_request *first,
678 struct blkif_request *second)
680 uint16_t nr_segments = first->u.rw.nr_segments;
683 * The second request is only present when the first request uses
684 * all its segments. It's always the continuity of the first one.
686 first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
688 second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
689 second->u.rw.sector_number = first->u.rw.sector_number +
690 (BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
692 second->u.rw.handle = first->u.rw.handle;
693 second->operation = first->operation;
696 static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
698 struct blkfront_info *info = rinfo->dev_info;
699 struct blkif_request *ring_req, *extra_ring_req = NULL;
700 struct blkif_request *final_ring_req, *final_extra_ring_req = NULL;
701 unsigned long id, extra_id = NO_ASSOCIATED_ID;
702 bool require_extra_req = false;
704 struct setup_rw_req setup = {
708 .need_copy = rq_data_dir(req) && info->feature_persistent,
712 * Used to store if we are able to queue the request by just using
713 * existing persistent grants, or if we have to get new grants,
714 * as there are not sufficiently many free.
716 bool new_persistent_gnts = false;
717 struct scatterlist *sg;
718 int num_sg, max_grefs, num_grant;
720 max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
721 if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
723 * If we are using indirect segments we need to account
724 * for the indirect grefs used in the request.
726 max_grefs += INDIRECT_GREFS(max_grefs);
728 /* Check if we have enough persistent grants to allocate a requests */
729 if (rinfo->persistent_gnts_c < max_grefs) {
730 new_persistent_gnts = true;
732 if (gnttab_alloc_grant_references(
733 max_grefs - rinfo->persistent_gnts_c,
734 &setup.gref_head) < 0) {
735 gnttab_request_free_callback(
737 blkif_restart_queue_callback,
739 max_grefs - rinfo->persistent_gnts_c);
744 /* Fill out a communications ring structure. */
745 id = blkif_ring_get_request(rinfo, req, &final_ring_req);
746 ring_req = &rinfo->shadow[id].req;
748 num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
750 /* Calculate the number of grant used */
751 for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
752 num_grant += gnttab_count_grant(sg->offset, sg->length);
754 require_extra_req = info->max_indirect_segments == 0 &&
755 num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
756 BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
758 rinfo->shadow[id].num_sg = num_sg;
759 if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
760 likely(!require_extra_req)) {
762 * The indirect operation can only be a BLKIF_OP_READ or
765 BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA);
766 ring_req->operation = BLKIF_OP_INDIRECT;
767 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
768 BLKIF_OP_WRITE : BLKIF_OP_READ;
769 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
770 ring_req->u.indirect.handle = info->handle;
771 ring_req->u.indirect.nr_segments = num_grant;
773 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
774 ring_req->u.rw.handle = info->handle;
775 ring_req->operation = rq_data_dir(req) ?
776 BLKIF_OP_WRITE : BLKIF_OP_READ;
777 if (req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA) {
779 * Ideally we can do an unordered flush-to-disk.
780 * In case the backend onlysupports barriers, use that.
781 * A barrier request a superset of FUA, so we can
782 * implement it the same way. (It's also a FLUSH+FUA,
783 * since it is guaranteed ordered WRT previous writes.)
785 if (info->feature_flush && info->feature_fua)
786 ring_req->operation =
787 BLKIF_OP_WRITE_BARRIER;
788 else if (info->feature_flush)
789 ring_req->operation =
790 BLKIF_OP_FLUSH_DISKCACHE;
792 ring_req->operation = 0;
794 ring_req->u.rw.nr_segments = num_grant;
795 if (unlikely(require_extra_req)) {
796 extra_id = blkif_ring_get_request(rinfo, req,
797 &final_extra_ring_req);
798 extra_ring_req = &rinfo->shadow[extra_id].req;
801 * Only the first request contains the scatter-gather
804 rinfo->shadow[extra_id].num_sg = 0;
806 blkif_setup_extra_req(ring_req, extra_ring_req);
808 /* Link the 2 requests together */
809 rinfo->shadow[extra_id].associated_id = id;
810 rinfo->shadow[id].associated_id = extra_id;
814 setup.ring_req = ring_req;
817 setup.require_extra_req = require_extra_req;
818 if (unlikely(require_extra_req))
819 setup.extra_ring_req = extra_ring_req;
821 for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
822 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
824 if (setup.need_copy) {
825 setup.bvec_off = sg->offset;
826 setup.bvec_data = kmap_atomic(sg_page(sg));
829 gnttab_foreach_grant_in_range(sg_page(sg),
832 blkif_setup_rw_req_grant,
836 kunmap_atomic(setup.bvec_data);
839 kunmap_atomic(setup.segments);
841 /* Copy request(s) to the ring page. */
842 *final_ring_req = *ring_req;
843 rinfo->shadow[id].status = REQ_WAITING;
844 if (unlikely(require_extra_req)) {
845 *final_extra_ring_req = *extra_ring_req;
846 rinfo->shadow[extra_id].status = REQ_WAITING;
849 if (new_persistent_gnts)
850 gnttab_free_grant_references(setup.gref_head);
856 * Generate a Xen blkfront IO request from a blk layer request. Reads
857 * and writes are handled as expected.
859 * @req: a request struct
861 static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
863 if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
866 if (unlikely(req_op(req) == REQ_OP_DISCARD ||
867 req_op(req) == REQ_OP_SECURE_ERASE))
868 return blkif_queue_discard_req(req, rinfo);
870 return blkif_queue_rw_req(req, rinfo);
873 static inline void flush_requests(struct blkfront_ring_info *rinfo)
877 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
880 notify_remote_via_irq(rinfo->irq);
883 static inline bool blkif_request_flush_invalid(struct request *req,
884 struct blkfront_info *info)
886 return (blk_rq_is_passthrough(req) ||
887 ((req_op(req) == REQ_OP_FLUSH) &&
888 !info->feature_flush) ||
889 ((req->cmd_flags & REQ_FUA) &&
890 !info->feature_fua));
893 static blk_status_t blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
894 const struct blk_mq_queue_data *qd)
897 int qid = hctx->queue_num;
898 struct blkfront_info *info = hctx->queue->queuedata;
899 struct blkfront_ring_info *rinfo = NULL;
901 rinfo = get_rinfo(info, qid);
902 blk_mq_start_request(qd->rq);
903 spin_lock_irqsave(&rinfo->ring_lock, flags);
904 if (RING_FULL(&rinfo->ring))
907 if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
910 if (blkif_queue_request(qd->rq, rinfo))
913 flush_requests(rinfo);
914 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
918 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
919 return BLK_STS_IOERR;
922 blk_mq_stop_hw_queue(hctx);
923 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
924 return BLK_STS_DEV_RESOURCE;
927 static void blkif_complete_rq(struct request *rq)
929 blk_mq_end_request(rq, blkif_req(rq)->error);
932 static const struct blk_mq_ops blkfront_mq_ops = {
933 .queue_rq = blkif_queue_rq,
934 .complete = blkif_complete_rq,
937 static void blkif_set_queue_limits(struct blkfront_info *info)
939 struct request_queue *rq = info->rq;
940 struct gendisk *gd = info->gd;
941 unsigned int segments = info->max_indirect_segments ? :
942 BLKIF_MAX_SEGMENTS_PER_REQUEST;
944 blk_queue_flag_set(QUEUE_FLAG_VIRT, rq);
946 if (info->feature_discard) {
947 blk_queue_max_discard_sectors(rq, get_capacity(gd));
948 rq->limits.discard_granularity = info->discard_granularity ?:
949 info->physical_sector_size;
950 rq->limits.discard_alignment = info->discard_alignment;
951 if (info->feature_secdiscard)
952 blk_queue_flag_set(QUEUE_FLAG_SECERASE, rq);
955 /* Hard sector size and max sectors impersonate the equiv. hardware. */
956 blk_queue_logical_block_size(rq, info->sector_size);
957 blk_queue_physical_block_size(rq, info->physical_sector_size);
958 blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
960 /* Each segment in a request is up to an aligned page in size. */
961 blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
962 blk_queue_max_segment_size(rq, PAGE_SIZE);
964 /* Ensure a merged request will fit in a single I/O ring slot. */
965 blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
967 /* Make sure buffer addresses are sector-aligned. */
968 blk_queue_dma_alignment(rq, 511);
971 static const char *flush_info(struct blkfront_info *info)
973 if (info->feature_flush && info->feature_fua)
974 return "barrier: enabled;";
975 else if (info->feature_flush)
976 return "flush diskcache: enabled;";
978 return "barrier or flush: disabled;";
981 static void xlvbd_flush(struct blkfront_info *info)
983 blk_queue_write_cache(info->rq, info->feature_flush ? true : false,
984 info->feature_fua ? true : false);
985 pr_info("blkfront: %s: %s %s %s %s %s\n",
986 info->gd->disk_name, flush_info(info),
987 "persistent grants:", info->feature_persistent ?
988 "enabled;" : "disabled;", "indirect descriptors:",
989 info->max_indirect_segments ? "enabled;" : "disabled;");
992 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
995 major = BLKIF_MAJOR(vdevice);
996 *minor = BLKIF_MINOR(vdevice);
999 *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
1000 *minor = ((*minor / 64) * PARTS_PER_DISK) +
1001 EMULATED_HD_DISK_MINOR_OFFSET;
1003 case XEN_IDE1_MAJOR:
1004 *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1005 *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1006 EMULATED_HD_DISK_MINOR_OFFSET;
1008 case XEN_SCSI_DISK0_MAJOR:
1009 *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1010 *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1012 case XEN_SCSI_DISK1_MAJOR:
1013 case XEN_SCSI_DISK2_MAJOR:
1014 case XEN_SCSI_DISK3_MAJOR:
1015 case XEN_SCSI_DISK4_MAJOR:
1016 case XEN_SCSI_DISK5_MAJOR:
1017 case XEN_SCSI_DISK6_MAJOR:
1018 case XEN_SCSI_DISK7_MAJOR:
1019 *offset = (*minor / PARTS_PER_DISK) +
1020 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1021 EMULATED_SD_DISK_NAME_OFFSET;
1023 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1024 EMULATED_SD_DISK_MINOR_OFFSET;
1026 case XEN_SCSI_DISK8_MAJOR:
1027 case XEN_SCSI_DISK9_MAJOR:
1028 case XEN_SCSI_DISK10_MAJOR:
1029 case XEN_SCSI_DISK11_MAJOR:
1030 case XEN_SCSI_DISK12_MAJOR:
1031 case XEN_SCSI_DISK13_MAJOR:
1032 case XEN_SCSI_DISK14_MAJOR:
1033 case XEN_SCSI_DISK15_MAJOR:
1034 *offset = (*minor / PARTS_PER_DISK) +
1035 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1036 EMULATED_SD_DISK_NAME_OFFSET;
1038 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1039 EMULATED_SD_DISK_MINOR_OFFSET;
1042 *offset = *minor / PARTS_PER_DISK;
1045 printk(KERN_WARNING "blkfront: your disk configuration is "
1046 "incorrect, please use an xvd device instead\n");
1052 static char *encode_disk_name(char *ptr, unsigned int n)
1055 ptr = encode_disk_name(ptr, n / 26 - 1);
1056 *ptr = 'a' + n % 26;
1060 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1061 struct blkfront_info *info, u16 sector_size,
1062 unsigned int physical_sector_size)
1067 unsigned int offset;
1072 BUG_ON(info->gd != NULL);
1073 BUG_ON(info->rq != NULL);
1075 if ((info->vdevice>>EXT_SHIFT) > 1) {
1076 /* this is above the extended range; something is wrong */
1077 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1081 if (!VDEV_IS_EXTENDED(info->vdevice)) {
1082 err = xen_translate_vdev(info->vdevice, &minor, &offset);
1085 nr_parts = PARTS_PER_DISK;
1087 minor = BLKIF_MINOR_EXT(info->vdevice);
1088 nr_parts = PARTS_PER_EXT_DISK;
1089 offset = minor / nr_parts;
1090 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1091 printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1092 "emulated IDE disks,\n\t choose an xvd device name"
1093 "from xvde on\n", info->vdevice);
1095 if (minor >> MINORBITS) {
1096 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1097 info->vdevice, minor);
1101 if ((minor % nr_parts) == 0)
1102 nr_minors = nr_parts;
1104 err = xlbd_reserve_minors(minor, nr_minors);
1108 memset(&info->tag_set, 0, sizeof(info->tag_set));
1109 info->tag_set.ops = &blkfront_mq_ops;
1110 info->tag_set.nr_hw_queues = info->nr_rings;
1111 if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
1113 * When indirect descriptior is not supported, the I/O request
1114 * will be split between multiple request in the ring.
1115 * To avoid problems when sending the request, divide by
1116 * 2 the depth of the queue.
1118 info->tag_set.queue_depth = BLK_RING_SIZE(info) / 2;
1120 info->tag_set.queue_depth = BLK_RING_SIZE(info);
1121 info->tag_set.numa_node = NUMA_NO_NODE;
1122 info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
1123 info->tag_set.cmd_size = sizeof(struct blkif_req);
1124 info->tag_set.driver_data = info;
1126 err = blk_mq_alloc_tag_set(&info->tag_set);
1128 goto out_release_minors;
1130 gd = blk_mq_alloc_disk(&info->tag_set, info);
1133 goto out_free_tag_set;
1136 strcpy(gd->disk_name, DEV_NAME);
1137 ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1138 BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1142 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1143 "%d", minor & (nr_parts - 1));
1145 gd->major = XENVBD_MAJOR;
1146 gd->first_minor = minor;
1147 gd->minors = nr_minors;
1148 gd->fops = &xlvbd_block_fops;
1149 gd->private_data = info;
1150 set_capacity(gd, capacity);
1152 info->rq = gd->queue;
1154 info->sector_size = sector_size;
1155 info->physical_sector_size = physical_sector_size;
1156 blkif_set_queue_limits(info);
1160 if (info->vdisk_info & VDISK_READONLY)
1162 if (info->vdisk_info & VDISK_REMOVABLE)
1163 gd->flags |= GENHD_FL_REMOVABLE;
1168 blk_mq_free_tag_set(&info->tag_set);
1170 xlbd_release_minors(minor, nr_minors);
1174 /* Already hold rinfo->ring_lock. */
1175 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1177 if (!RING_FULL(&rinfo->ring))
1178 blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1181 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1183 unsigned long flags;
1185 spin_lock_irqsave(&rinfo->ring_lock, flags);
1186 kick_pending_request_queues_locked(rinfo);
1187 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1190 static void blkif_restart_queue(struct work_struct *work)
1192 struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1194 if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1195 kick_pending_request_queues(rinfo);
1198 static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1200 struct grant *persistent_gnt, *n;
1201 struct blkfront_info *info = rinfo->dev_info;
1205 * Remove indirect pages, this only happens when using indirect
1206 * descriptors but not persistent grants
1208 if (!list_empty(&rinfo->indirect_pages)) {
1209 struct page *indirect_page, *n;
1211 BUG_ON(info->feature_persistent);
1212 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1213 list_del(&indirect_page->lru);
1214 __free_page(indirect_page);
1218 /* Remove all persistent grants. */
1219 if (!list_empty(&rinfo->grants)) {
1220 list_for_each_entry_safe(persistent_gnt, n,
1221 &rinfo->grants, node) {
1222 list_del(&persistent_gnt->node);
1223 if (persistent_gnt->gref != GRANT_INVALID_REF) {
1224 gnttab_end_foreign_access(persistent_gnt->gref,
1226 rinfo->persistent_gnts_c--;
1228 if (info->feature_persistent)
1229 __free_page(persistent_gnt->page);
1230 kfree(persistent_gnt);
1233 BUG_ON(rinfo->persistent_gnts_c != 0);
1235 for (i = 0; i < BLK_RING_SIZE(info); i++) {
1237 * Clear persistent grants present in requests already
1238 * on the shared ring
1240 if (!rinfo->shadow[i].request)
1243 segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1244 rinfo->shadow[i].req.u.indirect.nr_segments :
1245 rinfo->shadow[i].req.u.rw.nr_segments;
1246 for (j = 0; j < segs; j++) {
1247 persistent_gnt = rinfo->shadow[i].grants_used[j];
1248 gnttab_end_foreign_access(persistent_gnt->gref, 0UL);
1249 if (info->feature_persistent)
1250 __free_page(persistent_gnt->page);
1251 kfree(persistent_gnt);
1254 if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1256 * If this is not an indirect operation don't try to
1257 * free indirect segments
1261 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1262 persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1263 gnttab_end_foreign_access(persistent_gnt->gref, 0UL);
1264 __free_page(persistent_gnt->page);
1265 kfree(persistent_gnt);
1269 kvfree(rinfo->shadow[i].grants_used);
1270 rinfo->shadow[i].grants_used = NULL;
1271 kvfree(rinfo->shadow[i].indirect_grants);
1272 rinfo->shadow[i].indirect_grants = NULL;
1273 kvfree(rinfo->shadow[i].sg);
1274 rinfo->shadow[i].sg = NULL;
1277 /* No more gnttab callback work. */
1278 gnttab_cancel_free_callback(&rinfo->callback);
1280 /* Flush gnttab callback work. Must be done with no locks held. */
1281 flush_work(&rinfo->work);
1283 /* Free resources associated with old device channel. */
1284 for (i = 0; i < info->nr_ring_pages; i++) {
1285 if (rinfo->ring_ref[i] != GRANT_INVALID_REF) {
1286 gnttab_end_foreign_access(rinfo->ring_ref[i], 0);
1287 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1290 free_pages_exact(rinfo->ring.sring,
1291 info->nr_ring_pages * XEN_PAGE_SIZE);
1292 rinfo->ring.sring = NULL;
1295 unbind_from_irqhandler(rinfo->irq, rinfo);
1296 rinfo->evtchn = rinfo->irq = 0;
1299 static void blkif_free(struct blkfront_info *info, int suspend)
1302 struct blkfront_ring_info *rinfo;
1304 /* Prevent new requests being issued until we fix things up. */
1305 info->connected = suspend ?
1306 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1307 /* No more blkif_request(). */
1309 blk_mq_stop_hw_queues(info->rq);
1311 for_each_rinfo(info, rinfo, i)
1312 blkif_free_ring(rinfo);
1314 kvfree(info->rinfo);
1319 struct copy_from_grant {
1320 const struct blk_shadow *s;
1321 unsigned int grant_idx;
1322 unsigned int bvec_offset;
1326 static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1327 unsigned int len, void *data)
1329 struct copy_from_grant *info = data;
1331 /* Convenient aliases */
1332 const struct blk_shadow *s = info->s;
1334 shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1336 memcpy(info->bvec_data + info->bvec_offset,
1337 shared_data + offset, len);
1339 info->bvec_offset += len;
1342 kunmap_atomic(shared_data);
1345 static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1349 case BLKIF_RSP_OKAY:
1351 case BLKIF_RSP_EOPNOTSUPP:
1352 return REQ_EOPNOTSUPP;
1353 case BLKIF_RSP_ERROR:
1360 * Get the final status of the block request based on two ring response
1362 static int blkif_get_final_status(enum blk_req_status s1,
1363 enum blk_req_status s2)
1365 BUG_ON(s1 < REQ_DONE);
1366 BUG_ON(s2 < REQ_DONE);
1368 if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1369 return BLKIF_RSP_ERROR;
1370 else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1371 return BLKIF_RSP_EOPNOTSUPP;
1372 return BLKIF_RSP_OKAY;
1377 * 1 response processed.
1378 * 0 missing further responses.
1379 * -1 error while processing.
1381 static int blkif_completion(unsigned long *id,
1382 struct blkfront_ring_info *rinfo,
1383 struct blkif_response *bret)
1386 struct scatterlist *sg;
1387 int num_sg, num_grant;
1388 struct blkfront_info *info = rinfo->dev_info;
1389 struct blk_shadow *s = &rinfo->shadow[*id];
1390 struct copy_from_grant data = {
1394 num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1395 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1397 /* The I/O request may be split in two. */
1398 if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1399 struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1401 /* Keep the status of the current response in shadow. */
1402 s->status = blkif_rsp_to_req_status(bret->status);
1404 /* Wait the second response if not yet here. */
1405 if (s2->status < REQ_DONE)
1408 bret->status = blkif_get_final_status(s->status,
1412 * All the grants is stored in the first shadow in order
1413 * to make the completion code simpler.
1415 num_grant += s2->req.u.rw.nr_segments;
1418 * The two responses may not come in order. Only the
1419 * first request will store the scatter-gather list.
1421 if (s2->num_sg != 0) {
1422 /* Update "id" with the ID of the first response. */
1423 *id = s->associated_id;
1428 * We don't need anymore the second request, so recycling
1431 if (add_id_to_freelist(rinfo, s->associated_id))
1432 WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1433 info->gd->disk_name, s->associated_id);
1439 if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1440 for_each_sg(s->sg, sg, num_sg, i) {
1441 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1443 data.bvec_offset = sg->offset;
1444 data.bvec_data = kmap_atomic(sg_page(sg));
1446 gnttab_foreach_grant_in_range(sg_page(sg),
1449 blkif_copy_from_grant,
1452 kunmap_atomic(data.bvec_data);
1455 /* Add the persistent grant into the list of free grants */
1456 for (i = 0; i < num_grant; i++) {
1457 if (!gnttab_try_end_foreign_access(s->grants_used[i]->gref)) {
1459 * If the grant is still mapped by the backend (the
1460 * backend has chosen to make this grant persistent)
1461 * we add it at the head of the list, so it will be
1464 if (!info->feature_persistent) {
1465 pr_alert("backed has not unmapped grant: %u\n",
1466 s->grants_used[i]->gref);
1469 list_add(&s->grants_used[i]->node, &rinfo->grants);
1470 rinfo->persistent_gnts_c++;
1473 * If the grant is not mapped by the backend we add it
1474 * to the tail of the list, so it will not be picked
1475 * again unless we run out of persistent grants.
1477 s->grants_used[i]->gref = GRANT_INVALID_REF;
1478 list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1481 if (s->req.operation == BLKIF_OP_INDIRECT) {
1482 for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1483 if (!gnttab_try_end_foreign_access(s->indirect_grants[i]->gref)) {
1484 if (!info->feature_persistent) {
1485 pr_alert("backed has not unmapped grant: %u\n",
1486 s->indirect_grants[i]->gref);
1489 list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1490 rinfo->persistent_gnts_c++;
1492 struct page *indirect_page;
1495 * Add the used indirect page back to the list of
1496 * available pages for indirect grefs.
1498 if (!info->feature_persistent) {
1499 indirect_page = s->indirect_grants[i]->page;
1500 list_add(&indirect_page->lru, &rinfo->indirect_pages);
1502 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1503 list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1511 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1513 struct request *req;
1514 struct blkif_response bret;
1516 unsigned long flags;
1517 struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1518 struct blkfront_info *info = rinfo->dev_info;
1519 unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS;
1521 if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
1522 xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
1526 spin_lock_irqsave(&rinfo->ring_lock, flags);
1528 rp = READ_ONCE(rinfo->ring.sring->rsp_prod);
1529 virt_rmb(); /* Ensure we see queued responses up to 'rp'. */
1530 if (RING_RESPONSE_PROD_OVERFLOW(&rinfo->ring, rp)) {
1531 pr_alert("%s: illegal number of responses %u\n",
1532 info->gd->disk_name, rp - rinfo->ring.rsp_cons);
1536 for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1542 RING_COPY_RESPONSE(&rinfo->ring, i, &bret);
1546 * The backend has messed up and given us an id that we would
1547 * never have given to it (we stamp it up to BLK_RING_SIZE -
1548 * look in get_id_from_freelist.
1550 if (id >= BLK_RING_SIZE(info)) {
1551 pr_alert("%s: response has incorrect id (%ld)\n",
1552 info->gd->disk_name, id);
1555 if (rinfo->shadow[id].status != REQ_WAITING) {
1556 pr_alert("%s: response references no pending request\n",
1557 info->gd->disk_name);
1561 rinfo->shadow[id].status = REQ_PROCESSING;
1562 req = rinfo->shadow[id].request;
1564 op = rinfo->shadow[id].req.operation;
1565 if (op == BLKIF_OP_INDIRECT)
1566 op = rinfo->shadow[id].req.u.indirect.indirect_op;
1567 if (bret.operation != op) {
1568 pr_alert("%s: response has wrong operation (%u instead of %u)\n",
1569 info->gd->disk_name, bret.operation, op);
1573 if (bret.operation != BLKIF_OP_DISCARD) {
1577 * We may need to wait for an extra response if the
1578 * I/O request is split in 2
1580 ret = blkif_completion(&id, rinfo, &bret);
1583 if (unlikely(ret < 0))
1587 if (add_id_to_freelist(rinfo, id)) {
1588 WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1589 info->gd->disk_name, op_name(bret.operation), id);
1593 if (bret.status == BLKIF_RSP_OKAY)
1594 blkif_req(req)->error = BLK_STS_OK;
1596 blkif_req(req)->error = BLK_STS_IOERR;
1598 switch (bret.operation) {
1599 case BLKIF_OP_DISCARD:
1600 if (unlikely(bret.status == BLKIF_RSP_EOPNOTSUPP)) {
1601 struct request_queue *rq = info->rq;
1603 pr_warn_ratelimited("blkfront: %s: %s op failed\n",
1604 info->gd->disk_name, op_name(bret.operation));
1605 blkif_req(req)->error = BLK_STS_NOTSUPP;
1606 info->feature_discard = 0;
1607 info->feature_secdiscard = 0;
1608 blk_queue_max_discard_sectors(rq, 0);
1609 blk_queue_flag_clear(QUEUE_FLAG_SECERASE, rq);
1612 case BLKIF_OP_FLUSH_DISKCACHE:
1613 case BLKIF_OP_WRITE_BARRIER:
1614 if (unlikely(bret.status == BLKIF_RSP_EOPNOTSUPP)) {
1615 pr_warn_ratelimited("blkfront: %s: %s op failed\n",
1616 info->gd->disk_name, op_name(bret.operation));
1617 blkif_req(req)->error = BLK_STS_NOTSUPP;
1619 if (unlikely(bret.status == BLKIF_RSP_ERROR &&
1620 rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1621 pr_warn_ratelimited("blkfront: %s: empty %s op failed\n",
1622 info->gd->disk_name, op_name(bret.operation));
1623 blkif_req(req)->error = BLK_STS_NOTSUPP;
1625 if (unlikely(blkif_req(req)->error)) {
1626 if (blkif_req(req)->error == BLK_STS_NOTSUPP)
1627 blkif_req(req)->error = BLK_STS_OK;
1628 info->feature_fua = 0;
1629 info->feature_flush = 0;
1634 case BLKIF_OP_WRITE:
1635 if (unlikely(bret.status != BLKIF_RSP_OKAY))
1636 dev_dbg_ratelimited(&info->xbdev->dev,
1637 "Bad return from blkdev data request: %#x\n",
1645 if (likely(!blk_should_fake_timeout(req->q)))
1646 blk_mq_complete_request(req);
1649 rinfo->ring.rsp_cons = i;
1651 if (i != rinfo->ring.req_prod_pvt) {
1653 RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1657 rinfo->ring.sring->rsp_event = i + 1;
1659 kick_pending_request_queues_locked(rinfo);
1661 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1663 xen_irq_lateeoi(irq, eoiflag);
1668 info->connected = BLKIF_STATE_ERROR;
1670 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1672 /* No EOI in order to avoid further interrupts. */
1674 pr_alert("%s disabled for further use\n", info->gd->disk_name);
1679 static int setup_blkring(struct xenbus_device *dev,
1680 struct blkfront_ring_info *rinfo)
1682 struct blkif_sring *sring;
1684 struct blkfront_info *info = rinfo->dev_info;
1685 unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1686 grant_ref_t gref[XENBUS_MAX_RING_GRANTS];
1688 for (i = 0; i < info->nr_ring_pages; i++)
1689 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1691 sring = alloc_pages_exact(ring_size, GFP_NOIO);
1693 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1696 SHARED_RING_INIT(sring);
1697 FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1699 err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref);
1701 free_pages_exact(sring, ring_size);
1702 rinfo->ring.sring = NULL;
1705 for (i = 0; i < info->nr_ring_pages; i++)
1706 rinfo->ring_ref[i] = gref[i];
1708 err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1712 err = bind_evtchn_to_irqhandler_lateeoi(rinfo->evtchn, blkif_interrupt,
1715 xenbus_dev_fatal(dev, err,
1716 "bind_evtchn_to_irqhandler failed");
1723 blkif_free(info, 0);
1728 * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1729 * ring buffer may have multi pages depending on ->nr_ring_pages.
1731 static int write_per_ring_nodes(struct xenbus_transaction xbt,
1732 struct blkfront_ring_info *rinfo, const char *dir)
1736 const char *message = NULL;
1737 struct blkfront_info *info = rinfo->dev_info;
1739 if (info->nr_ring_pages == 1) {
1740 err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1742 message = "writing ring-ref";
1743 goto abort_transaction;
1746 for (i = 0; i < info->nr_ring_pages; i++) {
1747 char ring_ref_name[RINGREF_NAME_LEN];
1749 snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1750 err = xenbus_printf(xbt, dir, ring_ref_name,
1751 "%u", rinfo->ring_ref[i]);
1753 message = "writing ring-ref";
1754 goto abort_transaction;
1759 err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1761 message = "writing event-channel";
1762 goto abort_transaction;
1768 xenbus_transaction_end(xbt, 1);
1770 xenbus_dev_fatal(info->xbdev, err, "%s", message);
1775 /* Common code used when first setting up, and when resuming. */
1776 static int talk_to_blkback(struct xenbus_device *dev,
1777 struct blkfront_info *info)
1779 const char *message = NULL;
1780 struct xenbus_transaction xbt;
1782 unsigned int i, max_page_order;
1783 unsigned int ring_page_order;
1784 struct blkfront_ring_info *rinfo;
1789 max_page_order = xenbus_read_unsigned(info->xbdev->otherend,
1790 "max-ring-page-order", 0);
1791 ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1792 info->nr_ring_pages = 1 << ring_page_order;
1794 err = negotiate_mq(info);
1796 goto destroy_blkring;
1798 for_each_rinfo(info, rinfo, i) {
1799 /* Create shared ring, alloc event channel. */
1800 err = setup_blkring(dev, rinfo);
1802 goto destroy_blkring;
1806 err = xenbus_transaction_start(&xbt);
1808 xenbus_dev_fatal(dev, err, "starting transaction");
1809 goto destroy_blkring;
1812 if (info->nr_ring_pages > 1) {
1813 err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1816 message = "writing ring-page-order";
1817 goto abort_transaction;
1821 /* We already got the number of queues/rings in _probe */
1822 if (info->nr_rings == 1) {
1823 err = write_per_ring_nodes(xbt, info->rinfo, dev->nodename);
1825 goto destroy_blkring;
1830 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1833 message = "writing multi-queue-num-queues";
1834 goto abort_transaction;
1837 pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1838 path = kmalloc(pathsize, GFP_KERNEL);
1841 message = "ENOMEM while writing ring references";
1842 goto abort_transaction;
1845 for_each_rinfo(info, rinfo, i) {
1846 memset(path, 0, pathsize);
1847 snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1848 err = write_per_ring_nodes(xbt, rinfo, path);
1851 goto destroy_blkring;
1856 err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1857 XEN_IO_PROTO_ABI_NATIVE);
1859 message = "writing protocol";
1860 goto abort_transaction;
1862 err = xenbus_printf(xbt, dev->nodename, "feature-persistent", "%u",
1863 info->feature_persistent);
1866 "writing persistent grants feature to xenbus");
1868 err = xenbus_transaction_end(xbt, 0);
1872 xenbus_dev_fatal(dev, err, "completing transaction");
1873 goto destroy_blkring;
1876 for_each_rinfo(info, rinfo, i) {
1879 for (j = 0; j < BLK_RING_SIZE(info); j++)
1880 rinfo->shadow[j].req.u.rw.id = j + 1;
1881 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1883 xenbus_switch_state(dev, XenbusStateInitialised);
1888 xenbus_transaction_end(xbt, 1);
1890 xenbus_dev_fatal(dev, err, "%s", message);
1892 blkif_free(info, 0);
1896 static int negotiate_mq(struct blkfront_info *info)
1898 unsigned int backend_max_queues;
1900 struct blkfront_ring_info *rinfo;
1902 BUG_ON(info->nr_rings);
1904 /* Check if backend supports multiple queues. */
1905 backend_max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1906 "multi-queue-max-queues", 1);
1907 info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1908 /* We need at least one ring. */
1909 if (!info->nr_rings)
1912 info->rinfo_size = struct_size(info->rinfo, shadow,
1913 BLK_RING_SIZE(info));
1914 info->rinfo = kvcalloc(info->nr_rings, info->rinfo_size, GFP_KERNEL);
1916 xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1921 for_each_rinfo(info, rinfo, i) {
1922 INIT_LIST_HEAD(&rinfo->indirect_pages);
1923 INIT_LIST_HEAD(&rinfo->grants);
1924 rinfo->dev_info = info;
1925 INIT_WORK(&rinfo->work, blkif_restart_queue);
1926 spin_lock_init(&rinfo->ring_lock);
1931 /* Enable the persistent grants feature. */
1932 static bool feature_persistent = true;
1933 module_param(feature_persistent, bool, 0644);
1934 MODULE_PARM_DESC(feature_persistent,
1935 "Enables the persistent grants feature");
1938 * Entry point to this code when a new device is created. Allocate the basic
1939 * structures and the ring buffer for communication with the backend, and
1940 * inform the backend of the appropriate details for those. Switch to
1941 * Initialised state.
1943 static int blkfront_probe(struct xenbus_device *dev,
1944 const struct xenbus_device_id *id)
1947 struct blkfront_info *info;
1949 /* FIXME: Use dynamic device id if this is not set. */
1950 err = xenbus_scanf(XBT_NIL, dev->nodename,
1951 "virtual-device", "%i", &vdevice);
1953 /* go looking in the extended area instead */
1954 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1957 xenbus_dev_fatal(dev, err, "reading virtual-device");
1962 if (xen_hvm_domain()) {
1965 /* no unplug has been done: do not hook devices != xen vbds */
1966 if (xen_has_pv_and_legacy_disk_devices()) {
1969 if (!VDEV_IS_EXTENDED(vdevice))
1970 major = BLKIF_MAJOR(vdevice);
1972 major = XENVBD_MAJOR;
1974 if (major != XENVBD_MAJOR) {
1976 "%s: HVM does not support vbd %d as xen block device\n",
1981 /* do not create a PV cdrom device if we are an HVM guest */
1982 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1985 if (strncmp(type, "cdrom", 5) == 0) {
1991 info = kzalloc(sizeof(*info), GFP_KERNEL);
1993 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1999 mutex_init(&info->mutex);
2000 info->vdevice = vdevice;
2001 info->connected = BLKIF_STATE_DISCONNECTED;
2003 info->feature_persistent = feature_persistent;
2005 /* Front end dir is a number, which is used as the id. */
2006 info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
2007 dev_set_drvdata(&dev->dev, info);
2009 mutex_lock(&blkfront_mutex);
2010 list_add(&info->info_list, &info_list);
2011 mutex_unlock(&blkfront_mutex);
2016 static int blkif_recover(struct blkfront_info *info)
2018 unsigned int r_index;
2019 struct request *req, *n;
2023 struct blkfront_ring_info *rinfo;
2025 blkfront_gather_backend_features(info);
2026 /* Reset limits changed by blk_mq_update_nr_hw_queues(). */
2027 blkif_set_queue_limits(info);
2028 segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
2029 blk_queue_max_segments(info->rq, segs / GRANTS_PER_PSEG);
2031 for_each_rinfo(info, rinfo, r_index) {
2032 rc = blkfront_setup_indirect(rinfo);
2036 xenbus_switch_state(info->xbdev, XenbusStateConnected);
2038 /* Now safe for us to use the shared ring */
2039 info->connected = BLKIF_STATE_CONNECTED;
2041 for_each_rinfo(info, rinfo, r_index) {
2042 /* Kick any other new requests queued since we resumed */
2043 kick_pending_request_queues(rinfo);
2046 list_for_each_entry_safe(req, n, &info->requests, queuelist) {
2047 /* Requeue pending requests (flush or discard) */
2048 list_del_init(&req->queuelist);
2049 BUG_ON(req->nr_phys_segments > segs);
2050 blk_mq_requeue_request(req, false);
2052 blk_mq_start_stopped_hw_queues(info->rq, true);
2053 blk_mq_kick_requeue_list(info->rq);
2055 while ((bio = bio_list_pop(&info->bio_list)) != NULL) {
2056 /* Traverse the list of pending bios and re-queue them */
2064 * We are reconnecting to the backend, due to a suspend/resume, or a backend
2065 * driver restart. We tear down our blkif structure and recreate it, but
2066 * leave the device-layer structures intact so that this is transparent to the
2067 * rest of the kernel.
2069 static int blkfront_resume(struct xenbus_device *dev)
2071 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2074 struct blkfront_ring_info *rinfo;
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_each_rinfo(info, rinfo, i) {
2081 struct bio_list merge_bio;
2082 struct blk_shadow *shadow = rinfo->shadow;
2084 for (j = 0; j < BLK_RING_SIZE(info); j++) {
2086 if (!shadow[j].request)
2090 * Get the bios in the request so we can re-queue them.
2092 if (req_op(shadow[j].request) == REQ_OP_FLUSH ||
2093 req_op(shadow[j].request) == REQ_OP_DISCARD ||
2094 req_op(shadow[j].request) == REQ_OP_SECURE_ERASE ||
2095 shadow[j].request->cmd_flags & REQ_FUA) {
2097 * Flush operations don't contain bios, so
2098 * we need to requeue the whole request
2100 * XXX: but this doesn't make any sense for a
2101 * write with the FUA flag set..
2103 list_add(&shadow[j].request->queuelist, &info->requests);
2106 merge_bio.head = shadow[j].request->bio;
2107 merge_bio.tail = shadow[j].request->biotail;
2108 bio_list_merge(&info->bio_list, &merge_bio);
2109 shadow[j].request->bio = NULL;
2110 blk_mq_end_request(shadow[j].request, BLK_STS_OK);
2114 blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2116 err = talk_to_blkback(dev, info);
2118 blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings);
2121 * We have to wait for the backend to switch to
2122 * connected state, since we want to read which
2123 * features it supports.
2129 static void blkfront_closing(struct blkfront_info *info)
2131 struct xenbus_device *xbdev = info->xbdev;
2132 struct blkfront_ring_info *rinfo;
2135 if (xbdev->state == XenbusStateClosing)
2138 /* No more blkif_request(). */
2139 blk_mq_stop_hw_queues(info->rq);
2140 blk_mark_disk_dead(info->gd);
2141 set_capacity(info->gd, 0);
2143 for_each_rinfo(info, rinfo, i) {
2144 /* No more gnttab callback work. */
2145 gnttab_cancel_free_callback(&rinfo->callback);
2147 /* Flush gnttab callback work. Must be done with no locks held. */
2148 flush_work(&rinfo->work);
2151 xenbus_frontend_closed(xbdev);
2154 static void blkfront_setup_discard(struct blkfront_info *info)
2156 info->feature_discard = 1;
2157 info->discard_granularity = xenbus_read_unsigned(info->xbdev->otherend,
2158 "discard-granularity",
2160 info->discard_alignment = xenbus_read_unsigned(info->xbdev->otherend,
2161 "discard-alignment", 0);
2162 info->feature_secdiscard =
2163 !!xenbus_read_unsigned(info->xbdev->otherend, "discard-secure",
2167 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2169 unsigned int psegs, grants, memflags;
2171 struct blkfront_info *info = rinfo->dev_info;
2173 memflags = memalloc_noio_save();
2175 if (info->max_indirect_segments == 0) {
2177 grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2180 * When an extra req is required, the maximum
2181 * grants supported is related to the size of the
2182 * Linux block segment.
2184 grants = GRANTS_PER_PSEG;
2188 grants = info->max_indirect_segments;
2189 psegs = DIV_ROUND_UP(grants, GRANTS_PER_PSEG);
2191 err = fill_grant_buffer(rinfo,
2192 (grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2196 if (!info->feature_persistent && info->max_indirect_segments) {
2198 * We are using indirect descriptors but not persistent
2199 * grants, we need to allocate a set of pages that can be
2200 * used for mapping indirect grefs
2202 int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2204 BUG_ON(!list_empty(&rinfo->indirect_pages));
2205 for (i = 0; i < num; i++) {
2206 struct page *indirect_page = alloc_page(GFP_KERNEL);
2209 list_add(&indirect_page->lru, &rinfo->indirect_pages);
2213 for (i = 0; i < BLK_RING_SIZE(info); i++) {
2214 rinfo->shadow[i].grants_used =
2216 sizeof(rinfo->shadow[i].grants_used[0]),
2218 rinfo->shadow[i].sg = kvcalloc(psegs,
2219 sizeof(rinfo->shadow[i].sg[0]),
2221 if (info->max_indirect_segments)
2222 rinfo->shadow[i].indirect_grants =
2223 kvcalloc(INDIRECT_GREFS(grants),
2224 sizeof(rinfo->shadow[i].indirect_grants[0]),
2226 if ((rinfo->shadow[i].grants_used == NULL) ||
2227 (rinfo->shadow[i].sg == NULL) ||
2228 (info->max_indirect_segments &&
2229 (rinfo->shadow[i].indirect_grants == NULL)))
2231 sg_init_table(rinfo->shadow[i].sg, psegs);
2234 memalloc_noio_restore(memflags);
2239 for (i = 0; i < BLK_RING_SIZE(info); i++) {
2240 kvfree(rinfo->shadow[i].grants_used);
2241 rinfo->shadow[i].grants_used = NULL;
2242 kvfree(rinfo->shadow[i].sg);
2243 rinfo->shadow[i].sg = NULL;
2244 kvfree(rinfo->shadow[i].indirect_grants);
2245 rinfo->shadow[i].indirect_grants = NULL;
2247 if (!list_empty(&rinfo->indirect_pages)) {
2248 struct page *indirect_page, *n;
2249 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2250 list_del(&indirect_page->lru);
2251 __free_page(indirect_page);
2255 memalloc_noio_restore(memflags);
2261 * Gather all backend feature-*
2263 static void blkfront_gather_backend_features(struct blkfront_info *info)
2265 unsigned int indirect_segments;
2267 info->feature_flush = 0;
2268 info->feature_fua = 0;
2271 * If there's no "feature-barrier" defined, then it means
2272 * we're dealing with a very old backend which writes
2273 * synchronously; nothing to do.
2275 * If there are barriers, then we use flush.
2277 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-barrier", 0)) {
2278 info->feature_flush = 1;
2279 info->feature_fua = 1;
2283 * And if there is "feature-flush-cache" use that above
2286 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-flush-cache",
2288 info->feature_flush = 1;
2289 info->feature_fua = 0;
2292 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-discard", 0))
2293 blkfront_setup_discard(info);
2295 if (info->feature_persistent)
2296 info->feature_persistent =
2297 !!xenbus_read_unsigned(info->xbdev->otherend,
2298 "feature-persistent", 0);
2300 indirect_segments = xenbus_read_unsigned(info->xbdev->otherend,
2301 "feature-max-indirect-segments", 0);
2302 if (indirect_segments > xen_blkif_max_segments)
2303 indirect_segments = xen_blkif_max_segments;
2304 if (indirect_segments <= BLKIF_MAX_SEGMENTS_PER_REQUEST)
2305 indirect_segments = 0;
2306 info->max_indirect_segments = indirect_segments;
2308 if (info->feature_persistent) {
2309 mutex_lock(&blkfront_mutex);
2310 schedule_delayed_work(&blkfront_work, HZ * 10);
2311 mutex_unlock(&blkfront_mutex);
2316 * Invoked when the backend is finally 'ready' (and has told produced
2317 * the details about the physical device - #sectors, size, etc).
2319 static void blkfront_connect(struct blkfront_info *info)
2321 unsigned long long sectors;
2322 unsigned long sector_size;
2323 unsigned int physical_sector_size;
2325 struct blkfront_ring_info *rinfo;
2327 switch (info->connected) {
2328 case BLKIF_STATE_CONNECTED:
2330 * Potentially, the back-end may be signalling
2331 * a capacity change; update the capacity.
2333 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2334 "sectors", "%Lu", §ors);
2335 if (XENBUS_EXIST_ERR(err))
2337 printk(KERN_INFO "Setting capacity to %Lu\n",
2339 set_capacity_and_notify(info->gd, sectors);
2342 case BLKIF_STATE_SUSPENDED:
2344 * If we are recovering from suspension, we need to wait
2345 * for the backend to announce it's features before
2346 * reconnecting, at least we need to know if the backend
2347 * supports indirect descriptors, and how many.
2349 blkif_recover(info);
2356 dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2357 __func__, info->xbdev->otherend);
2359 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2360 "sectors", "%llu", §ors,
2361 "info", "%u", &info->vdisk_info,
2362 "sector-size", "%lu", §or_size,
2365 xenbus_dev_fatal(info->xbdev, err,
2366 "reading backend fields at %s",
2367 info->xbdev->otherend);
2372 * physical-sector-size is a newer field, so old backends may not
2373 * provide this. Assume physical sector size to be the same as
2374 * sector_size in that case.
2376 physical_sector_size = xenbus_read_unsigned(info->xbdev->otherend,
2377 "physical-sector-size",
2379 blkfront_gather_backend_features(info);
2380 for_each_rinfo(info, rinfo, i) {
2381 err = blkfront_setup_indirect(rinfo);
2383 xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2384 info->xbdev->otherend);
2385 blkif_free(info, 0);
2390 err = xlvbd_alloc_gendisk(sectors, info, sector_size,
2391 physical_sector_size);
2393 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2394 info->xbdev->otherend);
2398 xenbus_switch_state(info->xbdev, XenbusStateConnected);
2400 /* Kick pending requests. */
2401 info->connected = BLKIF_STATE_CONNECTED;
2402 for_each_rinfo(info, rinfo, i)
2403 kick_pending_request_queues(rinfo);
2405 err = device_add_disk(&info->xbdev->dev, info->gd, NULL);
2407 blk_cleanup_disk(info->gd);
2408 blk_mq_free_tag_set(&info->tag_set);
2417 blkif_free(info, 0);
2422 * Callback received when the backend's state changes.
2424 static void blkback_changed(struct xenbus_device *dev,
2425 enum xenbus_state backend_state)
2427 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2429 dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2431 switch (backend_state) {
2432 case XenbusStateInitWait:
2433 if (dev->state != XenbusStateInitialising)
2435 if (talk_to_blkback(dev, info))
2438 case XenbusStateInitialising:
2439 case XenbusStateInitialised:
2440 case XenbusStateReconfiguring:
2441 case XenbusStateReconfigured:
2442 case XenbusStateUnknown:
2445 case XenbusStateConnected:
2447 * talk_to_blkback sets state to XenbusStateInitialised
2448 * and blkfront_connect sets it to XenbusStateConnected
2449 * (if connection went OK).
2451 * If the backend (or toolstack) decides to poke at backend
2452 * state (and re-trigger the watch by setting the state repeatedly
2453 * to XenbusStateConnected (4)) we need to deal with this.
2454 * This is allowed as this is used to communicate to the guest
2455 * that the size of disk has changed!
2457 if ((dev->state != XenbusStateInitialised) &&
2458 (dev->state != XenbusStateConnected)) {
2459 if (talk_to_blkback(dev, info))
2463 blkfront_connect(info);
2466 case XenbusStateClosed:
2467 if (dev->state == XenbusStateClosed)
2470 case XenbusStateClosing:
2471 blkfront_closing(info);
2476 static int blkfront_remove(struct xenbus_device *xbdev)
2478 struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2480 dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2482 del_gendisk(info->gd);
2484 mutex_lock(&blkfront_mutex);
2485 list_del(&info->info_list);
2486 mutex_unlock(&blkfront_mutex);
2488 blkif_free(info, 0);
2489 xlbd_release_minors(info->gd->first_minor, info->gd->minors);
2490 blk_cleanup_disk(info->gd);
2491 blk_mq_free_tag_set(&info->tag_set);
2497 static int blkfront_is_ready(struct xenbus_device *dev)
2499 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2501 return info->is_ready && info->xbdev;
2504 static const struct block_device_operations xlvbd_block_fops =
2506 .owner = THIS_MODULE,
2507 .getgeo = blkif_getgeo,
2508 .ioctl = blkif_ioctl,
2509 .compat_ioctl = blkdev_compat_ptr_ioctl,
2513 static const struct xenbus_device_id blkfront_ids[] = {
2518 static struct xenbus_driver blkfront_driver = {
2519 .ids = blkfront_ids,
2520 .probe = blkfront_probe,
2521 .remove = blkfront_remove,
2522 .resume = blkfront_resume,
2523 .otherend_changed = blkback_changed,
2524 .is_ready = blkfront_is_ready,
2527 static void purge_persistent_grants(struct blkfront_info *info)
2530 unsigned long flags;
2531 struct blkfront_ring_info *rinfo;
2533 for_each_rinfo(info, rinfo, i) {
2534 struct grant *gnt_list_entry, *tmp;
2537 spin_lock_irqsave(&rinfo->ring_lock, flags);
2539 if (rinfo->persistent_gnts_c == 0) {
2540 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2544 list_for_each_entry_safe(gnt_list_entry, tmp, &rinfo->grants,
2546 if (gnt_list_entry->gref == GRANT_INVALID_REF ||
2547 !gnttab_try_end_foreign_access(gnt_list_entry->gref))
2550 list_del(&gnt_list_entry->node);
2551 rinfo->persistent_gnts_c--;
2552 gnt_list_entry->gref = GRANT_INVALID_REF;
2553 list_add_tail(&gnt_list_entry->node, &grants);
2556 list_splice_tail(&grants, &rinfo->grants);
2558 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2562 static void blkfront_delay_work(struct work_struct *work)
2564 struct blkfront_info *info;
2565 bool need_schedule_work = false;
2567 mutex_lock(&blkfront_mutex);
2569 list_for_each_entry(info, &info_list, info_list) {
2570 if (info->feature_persistent) {
2571 need_schedule_work = true;
2572 mutex_lock(&info->mutex);
2573 purge_persistent_grants(info);
2574 mutex_unlock(&info->mutex);
2578 if (need_schedule_work)
2579 schedule_delayed_work(&blkfront_work, HZ * 10);
2581 mutex_unlock(&blkfront_mutex);
2584 static int __init xlblk_init(void)
2587 int nr_cpus = num_online_cpus();
2592 if (!xen_has_pv_disk_devices())
2595 if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2596 pr_warn("xen_blk: can't get major %d with name %s\n",
2597 XENVBD_MAJOR, DEV_NAME);
2601 if (xen_blkif_max_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST)
2602 xen_blkif_max_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2604 if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2605 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2606 xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2607 xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2610 if (xen_blkif_max_queues > nr_cpus) {
2611 pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2612 xen_blkif_max_queues, nr_cpus);
2613 xen_blkif_max_queues = nr_cpus;
2616 INIT_DELAYED_WORK(&blkfront_work, blkfront_delay_work);
2618 ret = xenbus_register_frontend(&blkfront_driver);
2620 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2626 module_init(xlblk_init);
2629 static void __exit xlblk_exit(void)
2631 cancel_delayed_work_sync(&blkfront_work);
2633 xenbus_unregister_driver(&blkfront_driver);
2634 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2637 module_exit(xlblk_exit);
2639 MODULE_DESCRIPTION("Xen virtual block device frontend");
2640 MODULE_LICENSE("GPL");
2641 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2642 MODULE_ALIAS("xen:vbd");
2643 MODULE_ALIAS("xenblk");