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;
203 enum blkif_state connected;
204 /* Number of pages per ring buffer. */
205 unsigned int nr_ring_pages;
206 struct request_queue *rq;
207 unsigned int feature_flush:1;
208 unsigned int feature_fua:1;
209 unsigned int feature_discard:1;
210 unsigned int feature_secdiscard:1;
211 unsigned int feature_persistent:1;
212 unsigned int discard_granularity;
213 unsigned int discard_alignment;
214 /* Number of 4KB segments handled */
215 unsigned int max_indirect_segments;
217 struct blk_mq_tag_set tag_set;
218 struct blkfront_ring_info *rinfo;
219 unsigned int nr_rings;
220 unsigned int rinfo_size;
221 /* Save uncomplete reqs and bios for migration. */
222 struct list_head requests;
223 struct bio_list bio_list;
224 struct list_head info_list;
227 static unsigned int nr_minors;
228 static unsigned long *minors;
229 static DEFINE_SPINLOCK(minor_lock);
231 #define GRANT_INVALID_REF 0
233 #define PARTS_PER_DISK 16
234 #define PARTS_PER_EXT_DISK 256
236 #define BLKIF_MAJOR(dev) ((dev)>>8)
237 #define BLKIF_MINOR(dev) ((dev) & 0xff)
240 #define EXTENDED (1<<EXT_SHIFT)
241 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
242 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
243 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
244 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
245 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
246 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
248 #define DEV_NAME "xvd" /* name in /dev */
251 * Grants are always the same size as a Xen page (i.e 4KB).
252 * A physical segment is always the same size as a Linux page.
253 * Number of grants per physical segment
255 #define GRANTS_PER_PSEG (PAGE_SIZE / XEN_PAGE_SIZE)
257 #define GRANTS_PER_INDIRECT_FRAME \
258 (XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
260 #define INDIRECT_GREFS(_grants) \
261 DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
263 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
264 static void blkfront_gather_backend_features(struct blkfront_info *info);
265 static int negotiate_mq(struct blkfront_info *info);
267 #define for_each_rinfo(info, ptr, idx) \
268 for ((ptr) = (info)->rinfo, (idx) = 0; \
269 (idx) < (info)->nr_rings; \
270 (idx)++, (ptr) = (void *)(ptr) + (info)->rinfo_size)
272 static inline struct blkfront_ring_info *
273 get_rinfo(const struct blkfront_info *info, unsigned int i)
275 BUG_ON(i >= info->nr_rings);
276 return (void *)info->rinfo + i * info->rinfo_size;
279 static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
281 unsigned long free = rinfo->shadow_free;
283 BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
284 rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
285 rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
289 static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
292 if (rinfo->shadow[id].req.u.rw.id != id)
294 if (rinfo->shadow[id].request == NULL)
296 rinfo->shadow[id].req.u.rw.id = rinfo->shadow_free;
297 rinfo->shadow[id].request = NULL;
298 rinfo->shadow_free = id;
302 static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
304 struct blkfront_info *info = rinfo->dev_info;
305 struct page *granted_page;
306 struct grant *gnt_list_entry, *n;
310 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
314 if (info->feature_persistent) {
315 granted_page = alloc_page(GFP_NOIO);
317 kfree(gnt_list_entry);
320 gnt_list_entry->page = granted_page;
323 gnt_list_entry->gref = GRANT_INVALID_REF;
324 list_add(&gnt_list_entry->node, &rinfo->grants);
331 list_for_each_entry_safe(gnt_list_entry, n,
332 &rinfo->grants, node) {
333 list_del(&gnt_list_entry->node);
334 if (info->feature_persistent)
335 __free_page(gnt_list_entry->page);
336 kfree(gnt_list_entry);
343 static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
345 struct grant *gnt_list_entry;
347 BUG_ON(list_empty(&rinfo->grants));
348 gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
350 list_del(&gnt_list_entry->node);
352 if (gnt_list_entry->gref != GRANT_INVALID_REF)
353 rinfo->persistent_gnts_c--;
355 return gnt_list_entry;
358 static inline void grant_foreign_access(const struct grant *gnt_list_entry,
359 const struct blkfront_info *info)
361 gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
362 info->xbdev->otherend_id,
363 gnt_list_entry->page,
367 static struct grant *get_grant(grant_ref_t *gref_head,
369 struct blkfront_ring_info *rinfo)
371 struct grant *gnt_list_entry = get_free_grant(rinfo);
372 struct blkfront_info *info = rinfo->dev_info;
374 if (gnt_list_entry->gref != GRANT_INVALID_REF)
375 return gnt_list_entry;
377 /* Assign a gref to this page */
378 gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
379 BUG_ON(gnt_list_entry->gref == -ENOSPC);
380 if (info->feature_persistent)
381 grant_foreign_access(gnt_list_entry, info);
383 /* Grant access to the GFN passed by the caller */
384 gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
385 info->xbdev->otherend_id,
389 return gnt_list_entry;
392 static struct grant *get_indirect_grant(grant_ref_t *gref_head,
393 struct blkfront_ring_info *rinfo)
395 struct grant *gnt_list_entry = get_free_grant(rinfo);
396 struct blkfront_info *info = rinfo->dev_info;
398 if (gnt_list_entry->gref != GRANT_INVALID_REF)
399 return gnt_list_entry;
401 /* Assign a gref to this page */
402 gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
403 BUG_ON(gnt_list_entry->gref == -ENOSPC);
404 if (!info->feature_persistent) {
405 struct page *indirect_page;
407 /* Fetch a pre-allocated page to use for indirect grefs */
408 BUG_ON(list_empty(&rinfo->indirect_pages));
409 indirect_page = list_first_entry(&rinfo->indirect_pages,
411 list_del(&indirect_page->lru);
412 gnt_list_entry->page = indirect_page;
414 grant_foreign_access(gnt_list_entry, info);
416 return gnt_list_entry;
419 static const char *op_name(int op)
421 static const char *const names[] = {
422 [BLKIF_OP_READ] = "read",
423 [BLKIF_OP_WRITE] = "write",
424 [BLKIF_OP_WRITE_BARRIER] = "barrier",
425 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
426 [BLKIF_OP_DISCARD] = "discard" };
428 if (op < 0 || op >= ARRAY_SIZE(names))
436 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
438 unsigned int end = minor + nr;
441 if (end > nr_minors) {
442 unsigned long *bitmap, *old;
444 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
449 spin_lock(&minor_lock);
450 if (end > nr_minors) {
452 memcpy(bitmap, minors,
453 BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
455 nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
458 spin_unlock(&minor_lock);
462 spin_lock(&minor_lock);
463 if (find_next_bit(minors, end, minor) >= end) {
464 bitmap_set(minors, minor, nr);
468 spin_unlock(&minor_lock);
473 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
475 unsigned int end = minor + nr;
477 BUG_ON(end > nr_minors);
478 spin_lock(&minor_lock);
479 bitmap_clear(minors, minor, nr);
480 spin_unlock(&minor_lock);
483 static void blkif_restart_queue_callback(void *arg)
485 struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
486 schedule_work(&rinfo->work);
489 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
491 /* We don't have real geometry info, but let's at least return
492 values consistent with the size of the device */
493 sector_t nsect = get_capacity(bd->bd_disk);
494 sector_t cylinders = nsect;
498 sector_div(cylinders, hg->heads * hg->sectors);
499 hg->cylinders = cylinders;
500 if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
501 hg->cylinders = 0xffff;
505 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
506 unsigned command, unsigned long argument)
511 case CDROMMULTISESSION:
512 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
513 if (put_user(0, (char __user *)(argument + i)))
516 case CDROM_GET_CAPABILITY:
517 if (bdev->bd_disk->flags & GENHD_FL_CD)
525 static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
527 struct blkif_request **ring_req)
531 *ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
532 rinfo->ring.req_prod_pvt++;
534 id = get_id_from_freelist(rinfo);
535 rinfo->shadow[id].request = req;
536 rinfo->shadow[id].status = REQ_PROCESSING;
537 rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
539 rinfo->shadow[id].req.u.rw.id = id;
544 static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
546 struct blkfront_info *info = rinfo->dev_info;
547 struct blkif_request *ring_req, *final_ring_req;
550 /* Fill out a communications ring structure. */
551 id = blkif_ring_get_request(rinfo, req, &final_ring_req);
552 ring_req = &rinfo->shadow[id].req;
554 ring_req->operation = BLKIF_OP_DISCARD;
555 ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
556 ring_req->u.discard.id = id;
557 ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
558 if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard)
559 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
561 ring_req->u.discard.flag = 0;
563 /* Copy the request to the ring page. */
564 *final_ring_req = *ring_req;
565 rinfo->shadow[id].status = REQ_WAITING;
570 struct setup_rw_req {
571 unsigned int grant_idx;
572 struct blkif_request_segment *segments;
573 struct blkfront_ring_info *rinfo;
574 struct blkif_request *ring_req;
575 grant_ref_t gref_head;
577 /* Only used when persistent grant is used and it's a read request */
579 unsigned int bvec_off;
582 bool require_extra_req;
583 struct blkif_request *extra_ring_req;
586 static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
587 unsigned int len, void *data)
589 struct setup_rw_req *setup = data;
591 struct grant *gnt_list_entry;
592 unsigned int fsect, lsect;
593 /* Convenient aliases */
594 unsigned int grant_idx = setup->grant_idx;
595 struct blkif_request *ring_req = setup->ring_req;
596 struct blkfront_ring_info *rinfo = setup->rinfo;
598 * We always use the shadow of the first request to store the list
599 * of grant associated to the block I/O request. This made the
600 * completion more easy to handle even if the block I/O request is
603 struct blk_shadow *shadow = &rinfo->shadow[setup->id];
605 if (unlikely(setup->require_extra_req &&
606 grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
608 * We are using the second request, setup grant_idx
609 * to be the index of the segment array.
611 grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
612 ring_req = setup->extra_ring_req;
615 if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
616 (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
618 kunmap_atomic(setup->segments);
620 n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
621 gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
622 shadow->indirect_grants[n] = gnt_list_entry;
623 setup->segments = kmap_atomic(gnt_list_entry->page);
624 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
627 gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
628 ref = gnt_list_entry->gref;
630 * All the grants are stored in the shadow of the first
631 * request. Therefore we have to use the global index.
633 shadow->grants_used[setup->grant_idx] = gnt_list_entry;
635 if (setup->need_copy) {
638 shared_data = kmap_atomic(gnt_list_entry->page);
640 * this does not wipe data stored outside the
641 * range sg->offset..sg->offset+sg->length.
642 * Therefore, blkback *could* see data from
643 * previous requests. This is OK as long as
644 * persistent grants are shared with just one
645 * domain. It may need refactoring if this
648 memcpy(shared_data + offset,
649 setup->bvec_data + setup->bvec_off,
652 kunmap_atomic(shared_data);
653 setup->bvec_off += len;
657 lsect = fsect + (len >> 9) - 1;
658 if (ring_req->operation != BLKIF_OP_INDIRECT) {
659 ring_req->u.rw.seg[grant_idx] =
660 (struct blkif_request_segment) {
663 .last_sect = lsect };
665 setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
666 (struct blkif_request_segment) {
669 .last_sect = lsect };
672 (setup->grant_idx)++;
675 static void blkif_setup_extra_req(struct blkif_request *first,
676 struct blkif_request *second)
678 uint16_t nr_segments = first->u.rw.nr_segments;
681 * The second request is only present when the first request uses
682 * all its segments. It's always the continuity of the first one.
684 first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
686 second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
687 second->u.rw.sector_number = first->u.rw.sector_number +
688 (BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
690 second->u.rw.handle = first->u.rw.handle;
691 second->operation = first->operation;
694 static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
696 struct blkfront_info *info = rinfo->dev_info;
697 struct blkif_request *ring_req, *extra_ring_req = NULL;
698 struct blkif_request *final_ring_req, *final_extra_ring_req = NULL;
699 unsigned long id, extra_id = NO_ASSOCIATED_ID;
700 bool require_extra_req = false;
702 struct setup_rw_req setup = {
706 .need_copy = rq_data_dir(req) && info->feature_persistent,
710 * Used to store if we are able to queue the request by just using
711 * existing persistent grants, or if we have to get new grants,
712 * as there are not sufficiently many free.
714 bool new_persistent_gnts = false;
715 struct scatterlist *sg;
716 int num_sg, max_grefs, num_grant;
718 max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
719 if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
721 * If we are using indirect segments we need to account
722 * for the indirect grefs used in the request.
724 max_grefs += INDIRECT_GREFS(max_grefs);
726 /* Check if we have enough persistent grants to allocate a requests */
727 if (rinfo->persistent_gnts_c < max_grefs) {
728 new_persistent_gnts = true;
730 if (gnttab_alloc_grant_references(
731 max_grefs - rinfo->persistent_gnts_c,
732 &setup.gref_head) < 0) {
733 gnttab_request_free_callback(
735 blkif_restart_queue_callback,
737 max_grefs - rinfo->persistent_gnts_c);
742 /* Fill out a communications ring structure. */
743 id = blkif_ring_get_request(rinfo, req, &final_ring_req);
744 ring_req = &rinfo->shadow[id].req;
746 num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
748 /* Calculate the number of grant used */
749 for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
750 num_grant += gnttab_count_grant(sg->offset, sg->length);
752 require_extra_req = info->max_indirect_segments == 0 &&
753 num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
754 BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
756 rinfo->shadow[id].num_sg = num_sg;
757 if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
758 likely(!require_extra_req)) {
760 * The indirect operation can only be a BLKIF_OP_READ or
763 BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA);
764 ring_req->operation = BLKIF_OP_INDIRECT;
765 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
766 BLKIF_OP_WRITE : BLKIF_OP_READ;
767 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
768 ring_req->u.indirect.handle = info->handle;
769 ring_req->u.indirect.nr_segments = num_grant;
771 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
772 ring_req->u.rw.handle = info->handle;
773 ring_req->operation = rq_data_dir(req) ?
774 BLKIF_OP_WRITE : BLKIF_OP_READ;
775 if (req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA) {
777 * Ideally we can do an unordered flush-to-disk.
778 * In case the backend onlysupports barriers, use that.
779 * A barrier request a superset of FUA, so we can
780 * implement it the same way. (It's also a FLUSH+FUA,
781 * since it is guaranteed ordered WRT previous writes.)
783 if (info->feature_flush && info->feature_fua)
784 ring_req->operation =
785 BLKIF_OP_WRITE_BARRIER;
786 else if (info->feature_flush)
787 ring_req->operation =
788 BLKIF_OP_FLUSH_DISKCACHE;
790 ring_req->operation = 0;
792 ring_req->u.rw.nr_segments = num_grant;
793 if (unlikely(require_extra_req)) {
794 extra_id = blkif_ring_get_request(rinfo, req,
795 &final_extra_ring_req);
796 extra_ring_req = &rinfo->shadow[extra_id].req;
799 * Only the first request contains the scatter-gather
802 rinfo->shadow[extra_id].num_sg = 0;
804 blkif_setup_extra_req(ring_req, extra_ring_req);
806 /* Link the 2 requests together */
807 rinfo->shadow[extra_id].associated_id = id;
808 rinfo->shadow[id].associated_id = extra_id;
812 setup.ring_req = ring_req;
815 setup.require_extra_req = require_extra_req;
816 if (unlikely(require_extra_req))
817 setup.extra_ring_req = extra_ring_req;
819 for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
820 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
822 if (setup.need_copy) {
823 setup.bvec_off = sg->offset;
824 setup.bvec_data = kmap_atomic(sg_page(sg));
827 gnttab_foreach_grant_in_range(sg_page(sg),
830 blkif_setup_rw_req_grant,
834 kunmap_atomic(setup.bvec_data);
837 kunmap_atomic(setup.segments);
839 /* Copy request(s) to the ring page. */
840 *final_ring_req = *ring_req;
841 rinfo->shadow[id].status = REQ_WAITING;
842 if (unlikely(require_extra_req)) {
843 *final_extra_ring_req = *extra_ring_req;
844 rinfo->shadow[extra_id].status = REQ_WAITING;
847 if (new_persistent_gnts)
848 gnttab_free_grant_references(setup.gref_head);
854 * Generate a Xen blkfront IO request from a blk layer request. Reads
855 * and writes are handled as expected.
857 * @req: a request struct
859 static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
861 if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
864 if (unlikely(req_op(req) == REQ_OP_DISCARD ||
865 req_op(req) == REQ_OP_SECURE_ERASE))
866 return blkif_queue_discard_req(req, rinfo);
868 return blkif_queue_rw_req(req, rinfo);
871 static inline void flush_requests(struct blkfront_ring_info *rinfo)
875 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
878 notify_remote_via_irq(rinfo->irq);
881 static inline bool blkif_request_flush_invalid(struct request *req,
882 struct blkfront_info *info)
884 return (blk_rq_is_passthrough(req) ||
885 ((req_op(req) == REQ_OP_FLUSH) &&
886 !info->feature_flush) ||
887 ((req->cmd_flags & REQ_FUA) &&
888 !info->feature_fua));
891 static blk_status_t blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
892 const struct blk_mq_queue_data *qd)
895 int qid = hctx->queue_num;
896 struct blkfront_info *info = hctx->queue->queuedata;
897 struct blkfront_ring_info *rinfo = NULL;
899 rinfo = get_rinfo(info, qid);
900 blk_mq_start_request(qd->rq);
901 spin_lock_irqsave(&rinfo->ring_lock, flags);
902 if (RING_FULL(&rinfo->ring))
905 if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
908 if (blkif_queue_request(qd->rq, rinfo))
911 flush_requests(rinfo);
912 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
916 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
917 return BLK_STS_IOERR;
920 blk_mq_stop_hw_queue(hctx);
921 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
922 return BLK_STS_DEV_RESOURCE;
925 static void blkif_complete_rq(struct request *rq)
927 blk_mq_end_request(rq, blkif_req(rq)->error);
930 static const struct blk_mq_ops blkfront_mq_ops = {
931 .queue_rq = blkif_queue_rq,
932 .complete = blkif_complete_rq,
935 static void blkif_set_queue_limits(struct blkfront_info *info)
937 struct request_queue *rq = info->rq;
938 struct gendisk *gd = info->gd;
939 unsigned int segments = info->max_indirect_segments ? :
940 BLKIF_MAX_SEGMENTS_PER_REQUEST;
942 blk_queue_flag_set(QUEUE_FLAG_VIRT, rq);
944 if (info->feature_discard) {
945 blk_queue_flag_set(QUEUE_FLAG_DISCARD, rq);
946 blk_queue_max_discard_sectors(rq, get_capacity(gd));
947 rq->limits.discard_granularity = info->discard_granularity ?:
948 info->physical_sector_size;
949 rq->limits.discard_alignment = info->discard_alignment;
950 if (info->feature_secdiscard)
951 blk_queue_flag_set(QUEUE_FLAG_SECERASE, rq);
954 /* Hard sector size and max sectors impersonate the equiv. hardware. */
955 blk_queue_logical_block_size(rq, info->sector_size);
956 blk_queue_physical_block_size(rq, info->physical_sector_size);
957 blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
959 /* Each segment in a request is up to an aligned page in size. */
960 blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
961 blk_queue_max_segment_size(rq, PAGE_SIZE);
963 /* Ensure a merged request will fit in a single I/O ring slot. */
964 blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
966 /* Make sure buffer addresses are sector-aligned. */
967 blk_queue_dma_alignment(rq, 511);
970 static const char *flush_info(struct blkfront_info *info)
972 if (info->feature_flush && info->feature_fua)
973 return "barrier: enabled;";
974 else if (info->feature_flush)
975 return "flush diskcache: enabled;";
977 return "barrier or flush: disabled;";
980 static void xlvbd_flush(struct blkfront_info *info)
982 blk_queue_write_cache(info->rq, info->feature_flush ? true : false,
983 info->feature_fua ? true : false);
984 pr_info("blkfront: %s: %s %s %s %s %s\n",
985 info->gd->disk_name, flush_info(info),
986 "persistent grants:", info->feature_persistent ?
987 "enabled;" : "disabled;", "indirect descriptors:",
988 info->max_indirect_segments ? "enabled;" : "disabled;");
991 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
994 major = BLKIF_MAJOR(vdevice);
995 *minor = BLKIF_MINOR(vdevice);
998 *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
999 *minor = ((*minor / 64) * PARTS_PER_DISK) +
1000 EMULATED_HD_DISK_MINOR_OFFSET;
1002 case XEN_IDE1_MAJOR:
1003 *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1004 *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1005 EMULATED_HD_DISK_MINOR_OFFSET;
1007 case XEN_SCSI_DISK0_MAJOR:
1008 *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1009 *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1011 case XEN_SCSI_DISK1_MAJOR:
1012 case XEN_SCSI_DISK2_MAJOR:
1013 case XEN_SCSI_DISK3_MAJOR:
1014 case XEN_SCSI_DISK4_MAJOR:
1015 case XEN_SCSI_DISK5_MAJOR:
1016 case XEN_SCSI_DISK6_MAJOR:
1017 case XEN_SCSI_DISK7_MAJOR:
1018 *offset = (*minor / PARTS_PER_DISK) +
1019 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1020 EMULATED_SD_DISK_NAME_OFFSET;
1022 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1023 EMULATED_SD_DISK_MINOR_OFFSET;
1025 case XEN_SCSI_DISK8_MAJOR:
1026 case XEN_SCSI_DISK9_MAJOR:
1027 case XEN_SCSI_DISK10_MAJOR:
1028 case XEN_SCSI_DISK11_MAJOR:
1029 case XEN_SCSI_DISK12_MAJOR:
1030 case XEN_SCSI_DISK13_MAJOR:
1031 case XEN_SCSI_DISK14_MAJOR:
1032 case XEN_SCSI_DISK15_MAJOR:
1033 *offset = (*minor / PARTS_PER_DISK) +
1034 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1035 EMULATED_SD_DISK_NAME_OFFSET;
1037 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1038 EMULATED_SD_DISK_MINOR_OFFSET;
1041 *offset = *minor / PARTS_PER_DISK;
1044 printk(KERN_WARNING "blkfront: your disk configuration is "
1045 "incorrect, please use an xvd device instead\n");
1051 static char *encode_disk_name(char *ptr, unsigned int n)
1054 ptr = encode_disk_name(ptr, n / 26 - 1);
1055 *ptr = 'a' + n % 26;
1059 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1060 struct blkfront_info *info,
1061 u16 vdisk_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 (vdisk_info & VDISK_READONLY)
1163 if (vdisk_info & VDISK_REMOVABLE)
1164 gd->flags |= GENHD_FL_REMOVABLE;
1166 if (vdisk_info & VDISK_CDROM)
1167 gd->flags |= GENHD_FL_CD;
1172 blk_mq_free_tag_set(&info->tag_set);
1174 xlbd_release_minors(minor, nr_minors);
1178 /* Already hold rinfo->ring_lock. */
1179 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1181 if (!RING_FULL(&rinfo->ring))
1182 blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1185 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1187 unsigned long flags;
1189 spin_lock_irqsave(&rinfo->ring_lock, flags);
1190 kick_pending_request_queues_locked(rinfo);
1191 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1194 static void blkif_restart_queue(struct work_struct *work)
1196 struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1198 if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1199 kick_pending_request_queues(rinfo);
1202 static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1204 struct grant *persistent_gnt, *n;
1205 struct blkfront_info *info = rinfo->dev_info;
1209 * Remove indirect pages, this only happens when using indirect
1210 * descriptors but not persistent grants
1212 if (!list_empty(&rinfo->indirect_pages)) {
1213 struct page *indirect_page, *n;
1215 BUG_ON(info->feature_persistent);
1216 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1217 list_del(&indirect_page->lru);
1218 __free_page(indirect_page);
1222 /* Remove all persistent grants. */
1223 if (!list_empty(&rinfo->grants)) {
1224 list_for_each_entry_safe(persistent_gnt, n,
1225 &rinfo->grants, node) {
1226 list_del(&persistent_gnt->node);
1227 if (persistent_gnt->gref != GRANT_INVALID_REF) {
1228 gnttab_end_foreign_access(persistent_gnt->gref,
1230 rinfo->persistent_gnts_c--;
1232 if (info->feature_persistent)
1233 __free_page(persistent_gnt->page);
1234 kfree(persistent_gnt);
1237 BUG_ON(rinfo->persistent_gnts_c != 0);
1239 for (i = 0; i < BLK_RING_SIZE(info); i++) {
1241 * Clear persistent grants present in requests already
1242 * on the shared ring
1244 if (!rinfo->shadow[i].request)
1247 segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1248 rinfo->shadow[i].req.u.indirect.nr_segments :
1249 rinfo->shadow[i].req.u.rw.nr_segments;
1250 for (j = 0; j < segs; j++) {
1251 persistent_gnt = rinfo->shadow[i].grants_used[j];
1252 gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1253 if (info->feature_persistent)
1254 __free_page(persistent_gnt->page);
1255 kfree(persistent_gnt);
1258 if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1260 * If this is not an indirect operation don't try to
1261 * free indirect segments
1265 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1266 persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1267 gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1268 __free_page(persistent_gnt->page);
1269 kfree(persistent_gnt);
1273 kvfree(rinfo->shadow[i].grants_used);
1274 rinfo->shadow[i].grants_used = NULL;
1275 kvfree(rinfo->shadow[i].indirect_grants);
1276 rinfo->shadow[i].indirect_grants = NULL;
1277 kvfree(rinfo->shadow[i].sg);
1278 rinfo->shadow[i].sg = NULL;
1281 /* No more gnttab callback work. */
1282 gnttab_cancel_free_callback(&rinfo->callback);
1284 /* Flush gnttab callback work. Must be done with no locks held. */
1285 flush_work(&rinfo->work);
1287 /* Free resources associated with old device channel. */
1288 for (i = 0; i < info->nr_ring_pages; i++) {
1289 if (rinfo->ring_ref[i] != GRANT_INVALID_REF) {
1290 gnttab_end_foreign_access(rinfo->ring_ref[i], 0, 0);
1291 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1294 free_pages_exact(rinfo->ring.sring,
1295 info->nr_ring_pages * XEN_PAGE_SIZE);
1296 rinfo->ring.sring = NULL;
1299 unbind_from_irqhandler(rinfo->irq, rinfo);
1300 rinfo->evtchn = rinfo->irq = 0;
1303 static void blkif_free(struct blkfront_info *info, int suspend)
1306 struct blkfront_ring_info *rinfo;
1308 /* Prevent new requests being issued until we fix things up. */
1309 info->connected = suspend ?
1310 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1311 /* No more blkif_request(). */
1313 blk_mq_stop_hw_queues(info->rq);
1315 for_each_rinfo(info, rinfo, i)
1316 blkif_free_ring(rinfo);
1318 kvfree(info->rinfo);
1323 struct copy_from_grant {
1324 const struct blk_shadow *s;
1325 unsigned int grant_idx;
1326 unsigned int bvec_offset;
1330 static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1331 unsigned int len, void *data)
1333 struct copy_from_grant *info = data;
1335 /* Convenient aliases */
1336 const struct blk_shadow *s = info->s;
1338 shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1340 memcpy(info->bvec_data + info->bvec_offset,
1341 shared_data + offset, len);
1343 info->bvec_offset += len;
1346 kunmap_atomic(shared_data);
1349 static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1353 case BLKIF_RSP_OKAY:
1355 case BLKIF_RSP_EOPNOTSUPP:
1356 return REQ_EOPNOTSUPP;
1357 case BLKIF_RSP_ERROR:
1364 * Get the final status of the block request based on two ring response
1366 static int blkif_get_final_status(enum blk_req_status s1,
1367 enum blk_req_status s2)
1369 BUG_ON(s1 < REQ_DONE);
1370 BUG_ON(s2 < REQ_DONE);
1372 if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1373 return BLKIF_RSP_ERROR;
1374 else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1375 return BLKIF_RSP_EOPNOTSUPP;
1376 return BLKIF_RSP_OKAY;
1381 * 1 response processed.
1382 * 0 missing further responses.
1383 * -1 error while processing.
1385 static int blkif_completion(unsigned long *id,
1386 struct blkfront_ring_info *rinfo,
1387 struct blkif_response *bret)
1390 struct scatterlist *sg;
1391 int num_sg, num_grant;
1392 struct blkfront_info *info = rinfo->dev_info;
1393 struct blk_shadow *s = &rinfo->shadow[*id];
1394 struct copy_from_grant data = {
1398 num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1399 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1401 /* The I/O request may be split in two. */
1402 if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1403 struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1405 /* Keep the status of the current response in shadow. */
1406 s->status = blkif_rsp_to_req_status(bret->status);
1408 /* Wait the second response if not yet here. */
1409 if (s2->status < REQ_DONE)
1412 bret->status = blkif_get_final_status(s->status,
1416 * All the grants is stored in the first shadow in order
1417 * to make the completion code simpler.
1419 num_grant += s2->req.u.rw.nr_segments;
1422 * The two responses may not come in order. Only the
1423 * first request will store the scatter-gather list.
1425 if (s2->num_sg != 0) {
1426 /* Update "id" with the ID of the first response. */
1427 *id = s->associated_id;
1432 * We don't need anymore the second request, so recycling
1435 if (add_id_to_freelist(rinfo, s->associated_id))
1436 WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1437 info->gd->disk_name, s->associated_id);
1443 if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1444 for_each_sg(s->sg, sg, num_sg, i) {
1445 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1447 data.bvec_offset = sg->offset;
1448 data.bvec_data = kmap_atomic(sg_page(sg));
1450 gnttab_foreach_grant_in_range(sg_page(sg),
1453 blkif_copy_from_grant,
1456 kunmap_atomic(data.bvec_data);
1459 /* Add the persistent grant into the list of free grants */
1460 for (i = 0; i < num_grant; i++) {
1461 if (!gnttab_try_end_foreign_access(s->grants_used[i]->gref)) {
1463 * If the grant is still mapped by the backend (the
1464 * backend has chosen to make this grant persistent)
1465 * we add it at the head of the list, so it will be
1468 if (!info->feature_persistent) {
1469 pr_alert("backed has not unmapped grant: %u\n",
1470 s->grants_used[i]->gref);
1473 list_add(&s->grants_used[i]->node, &rinfo->grants);
1474 rinfo->persistent_gnts_c++;
1477 * If the grant is not mapped by the backend we add it
1478 * to the tail of the list, so it will not be picked
1479 * again unless we run out of persistent grants.
1481 s->grants_used[i]->gref = GRANT_INVALID_REF;
1482 list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1485 if (s->req.operation == BLKIF_OP_INDIRECT) {
1486 for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1487 if (!gnttab_try_end_foreign_access(s->indirect_grants[i]->gref)) {
1488 if (!info->feature_persistent) {
1489 pr_alert("backed has not unmapped grant: %u\n",
1490 s->indirect_grants[i]->gref);
1493 list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1494 rinfo->persistent_gnts_c++;
1496 struct page *indirect_page;
1499 * Add the used indirect page back to the list of
1500 * available pages for indirect grefs.
1502 if (!info->feature_persistent) {
1503 indirect_page = s->indirect_grants[i]->page;
1504 list_add(&indirect_page->lru, &rinfo->indirect_pages);
1506 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1507 list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1515 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1517 struct request *req;
1518 struct blkif_response bret;
1520 unsigned long flags;
1521 struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1522 struct blkfront_info *info = rinfo->dev_info;
1523 unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS;
1525 if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
1526 xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
1530 spin_lock_irqsave(&rinfo->ring_lock, flags);
1532 rp = READ_ONCE(rinfo->ring.sring->rsp_prod);
1533 virt_rmb(); /* Ensure we see queued responses up to 'rp'. */
1534 if (RING_RESPONSE_PROD_OVERFLOW(&rinfo->ring, rp)) {
1535 pr_alert("%s: illegal number of responses %u\n",
1536 info->gd->disk_name, rp - rinfo->ring.rsp_cons);
1540 for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1546 RING_COPY_RESPONSE(&rinfo->ring, i, &bret);
1550 * The backend has messed up and given us an id that we would
1551 * never have given to it (we stamp it up to BLK_RING_SIZE -
1552 * look in get_id_from_freelist.
1554 if (id >= BLK_RING_SIZE(info)) {
1555 pr_alert("%s: response has incorrect id (%ld)\n",
1556 info->gd->disk_name, id);
1559 if (rinfo->shadow[id].status != REQ_WAITING) {
1560 pr_alert("%s: response references no pending request\n",
1561 info->gd->disk_name);
1565 rinfo->shadow[id].status = REQ_PROCESSING;
1566 req = rinfo->shadow[id].request;
1568 op = rinfo->shadow[id].req.operation;
1569 if (op == BLKIF_OP_INDIRECT)
1570 op = rinfo->shadow[id].req.u.indirect.indirect_op;
1571 if (bret.operation != op) {
1572 pr_alert("%s: response has wrong operation (%u instead of %u)\n",
1573 info->gd->disk_name, bret.operation, op);
1577 if (bret.operation != BLKIF_OP_DISCARD) {
1581 * We may need to wait for an extra response if the
1582 * I/O request is split in 2
1584 ret = blkif_completion(&id, rinfo, &bret);
1587 if (unlikely(ret < 0))
1591 if (add_id_to_freelist(rinfo, id)) {
1592 WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1593 info->gd->disk_name, op_name(bret.operation), id);
1597 if (bret.status == BLKIF_RSP_OKAY)
1598 blkif_req(req)->error = BLK_STS_OK;
1600 blkif_req(req)->error = BLK_STS_IOERR;
1602 switch (bret.operation) {
1603 case BLKIF_OP_DISCARD:
1604 if (unlikely(bret.status == BLKIF_RSP_EOPNOTSUPP)) {
1605 struct request_queue *rq = info->rq;
1607 pr_warn_ratelimited("blkfront: %s: %s op failed\n",
1608 info->gd->disk_name, op_name(bret.operation));
1609 blkif_req(req)->error = BLK_STS_NOTSUPP;
1610 info->feature_discard = 0;
1611 info->feature_secdiscard = 0;
1612 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1613 blk_queue_flag_clear(QUEUE_FLAG_SECERASE, rq);
1616 case BLKIF_OP_FLUSH_DISKCACHE:
1617 case BLKIF_OP_WRITE_BARRIER:
1618 if (unlikely(bret.status == BLKIF_RSP_EOPNOTSUPP)) {
1619 pr_warn_ratelimited("blkfront: %s: %s op failed\n",
1620 info->gd->disk_name, op_name(bret.operation));
1621 blkif_req(req)->error = BLK_STS_NOTSUPP;
1623 if (unlikely(bret.status == BLKIF_RSP_ERROR &&
1624 rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1625 pr_warn_ratelimited("blkfront: %s: empty %s op failed\n",
1626 info->gd->disk_name, op_name(bret.operation));
1627 blkif_req(req)->error = BLK_STS_NOTSUPP;
1629 if (unlikely(blkif_req(req)->error)) {
1630 if (blkif_req(req)->error == BLK_STS_NOTSUPP)
1631 blkif_req(req)->error = BLK_STS_OK;
1632 info->feature_fua = 0;
1633 info->feature_flush = 0;
1638 case BLKIF_OP_WRITE:
1639 if (unlikely(bret.status != BLKIF_RSP_OKAY))
1640 dev_dbg_ratelimited(&info->xbdev->dev,
1641 "Bad return from blkdev data request: %#x\n",
1649 if (likely(!blk_should_fake_timeout(req->q)))
1650 blk_mq_complete_request(req);
1653 rinfo->ring.rsp_cons = i;
1655 if (i != rinfo->ring.req_prod_pvt) {
1657 RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1661 rinfo->ring.sring->rsp_event = i + 1;
1663 kick_pending_request_queues_locked(rinfo);
1665 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1667 xen_irq_lateeoi(irq, eoiflag);
1672 info->connected = BLKIF_STATE_ERROR;
1674 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1676 /* No EOI in order to avoid further interrupts. */
1678 pr_alert("%s disabled for further use\n", info->gd->disk_name);
1683 static int setup_blkring(struct xenbus_device *dev,
1684 struct blkfront_ring_info *rinfo)
1686 struct blkif_sring *sring;
1688 struct blkfront_info *info = rinfo->dev_info;
1689 unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1690 grant_ref_t gref[XENBUS_MAX_RING_GRANTS];
1692 for (i = 0; i < info->nr_ring_pages; i++)
1693 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1695 sring = alloc_pages_exact(ring_size, GFP_NOIO);
1697 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1700 SHARED_RING_INIT(sring);
1701 FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1703 err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref);
1705 free_pages_exact(sring, ring_size);
1706 rinfo->ring.sring = NULL;
1709 for (i = 0; i < info->nr_ring_pages; i++)
1710 rinfo->ring_ref[i] = gref[i];
1712 err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1716 err = bind_evtchn_to_irqhandler_lateeoi(rinfo->evtchn, blkif_interrupt,
1719 xenbus_dev_fatal(dev, err,
1720 "bind_evtchn_to_irqhandler failed");
1727 blkif_free(info, 0);
1732 * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1733 * ring buffer may have multi pages depending on ->nr_ring_pages.
1735 static int write_per_ring_nodes(struct xenbus_transaction xbt,
1736 struct blkfront_ring_info *rinfo, const char *dir)
1740 const char *message = NULL;
1741 struct blkfront_info *info = rinfo->dev_info;
1743 if (info->nr_ring_pages == 1) {
1744 err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1746 message = "writing ring-ref";
1747 goto abort_transaction;
1750 for (i = 0; i < info->nr_ring_pages; i++) {
1751 char ring_ref_name[RINGREF_NAME_LEN];
1753 snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1754 err = xenbus_printf(xbt, dir, ring_ref_name,
1755 "%u", rinfo->ring_ref[i]);
1757 message = "writing ring-ref";
1758 goto abort_transaction;
1763 err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1765 message = "writing event-channel";
1766 goto abort_transaction;
1772 xenbus_transaction_end(xbt, 1);
1774 xenbus_dev_fatal(info->xbdev, err, "%s", message);
1779 /* Common code used when first setting up, and when resuming. */
1780 static int talk_to_blkback(struct xenbus_device *dev,
1781 struct blkfront_info *info)
1783 const char *message = NULL;
1784 struct xenbus_transaction xbt;
1786 unsigned int i, max_page_order;
1787 unsigned int ring_page_order;
1788 struct blkfront_ring_info *rinfo;
1793 max_page_order = xenbus_read_unsigned(info->xbdev->otherend,
1794 "max-ring-page-order", 0);
1795 ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1796 info->nr_ring_pages = 1 << ring_page_order;
1798 err = negotiate_mq(info);
1800 goto destroy_blkring;
1802 for_each_rinfo(info, rinfo, i) {
1803 /* Create shared ring, alloc event channel. */
1804 err = setup_blkring(dev, rinfo);
1806 goto destroy_blkring;
1810 err = xenbus_transaction_start(&xbt);
1812 xenbus_dev_fatal(dev, err, "starting transaction");
1813 goto destroy_blkring;
1816 if (info->nr_ring_pages > 1) {
1817 err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1820 message = "writing ring-page-order";
1821 goto abort_transaction;
1825 /* We already got the number of queues/rings in _probe */
1826 if (info->nr_rings == 1) {
1827 err = write_per_ring_nodes(xbt, info->rinfo, dev->nodename);
1829 goto destroy_blkring;
1834 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1837 message = "writing multi-queue-num-queues";
1838 goto abort_transaction;
1841 pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1842 path = kmalloc(pathsize, GFP_KERNEL);
1845 message = "ENOMEM while writing ring references";
1846 goto abort_transaction;
1849 for_each_rinfo(info, rinfo, i) {
1850 memset(path, 0, pathsize);
1851 snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1852 err = write_per_ring_nodes(xbt, rinfo, path);
1855 goto destroy_blkring;
1860 err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1861 XEN_IO_PROTO_ABI_NATIVE);
1863 message = "writing protocol";
1864 goto abort_transaction;
1866 err = xenbus_printf(xbt, dev->nodename, "feature-persistent", "%u",
1867 info->feature_persistent);
1870 "writing persistent grants feature to xenbus");
1872 err = xenbus_transaction_end(xbt, 0);
1876 xenbus_dev_fatal(dev, err, "completing transaction");
1877 goto destroy_blkring;
1880 for_each_rinfo(info, rinfo, i) {
1883 for (j = 0; j < BLK_RING_SIZE(info); j++)
1884 rinfo->shadow[j].req.u.rw.id = j + 1;
1885 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1887 xenbus_switch_state(dev, XenbusStateInitialised);
1892 xenbus_transaction_end(xbt, 1);
1894 xenbus_dev_fatal(dev, err, "%s", message);
1896 blkif_free(info, 0);
1900 static int negotiate_mq(struct blkfront_info *info)
1902 unsigned int backend_max_queues;
1904 struct blkfront_ring_info *rinfo;
1906 BUG_ON(info->nr_rings);
1908 /* Check if backend supports multiple queues. */
1909 backend_max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1910 "multi-queue-max-queues", 1);
1911 info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1912 /* We need at least one ring. */
1913 if (!info->nr_rings)
1916 info->rinfo_size = struct_size(info->rinfo, shadow,
1917 BLK_RING_SIZE(info));
1918 info->rinfo = kvcalloc(info->nr_rings, info->rinfo_size, GFP_KERNEL);
1920 xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1925 for_each_rinfo(info, rinfo, i) {
1926 INIT_LIST_HEAD(&rinfo->indirect_pages);
1927 INIT_LIST_HEAD(&rinfo->grants);
1928 rinfo->dev_info = info;
1929 INIT_WORK(&rinfo->work, blkif_restart_queue);
1930 spin_lock_init(&rinfo->ring_lock);
1935 /* Enable the persistent grants feature. */
1936 static bool feature_persistent = true;
1937 module_param(feature_persistent, bool, 0644);
1938 MODULE_PARM_DESC(feature_persistent,
1939 "Enables the persistent grants feature");
1942 * Entry point to this code when a new device is created. Allocate the basic
1943 * structures and the ring buffer for communication with the backend, and
1944 * inform the backend of the appropriate details for those. Switch to
1945 * Initialised state.
1947 static int blkfront_probe(struct xenbus_device *dev,
1948 const struct xenbus_device_id *id)
1951 struct blkfront_info *info;
1953 /* FIXME: Use dynamic device id if this is not set. */
1954 err = xenbus_scanf(XBT_NIL, dev->nodename,
1955 "virtual-device", "%i", &vdevice);
1957 /* go looking in the extended area instead */
1958 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1961 xenbus_dev_fatal(dev, err, "reading virtual-device");
1966 if (xen_hvm_domain()) {
1969 /* no unplug has been done: do not hook devices != xen vbds */
1970 if (xen_has_pv_and_legacy_disk_devices()) {
1973 if (!VDEV_IS_EXTENDED(vdevice))
1974 major = BLKIF_MAJOR(vdevice);
1976 major = XENVBD_MAJOR;
1978 if (major != XENVBD_MAJOR) {
1980 "%s: HVM does not support vbd %d as xen block device\n",
1985 /* do not create a PV cdrom device if we are an HVM guest */
1986 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1989 if (strncmp(type, "cdrom", 5) == 0) {
1995 info = kzalloc(sizeof(*info), GFP_KERNEL);
1997 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
2003 mutex_init(&info->mutex);
2004 info->vdevice = vdevice;
2005 info->connected = BLKIF_STATE_DISCONNECTED;
2007 info->feature_persistent = feature_persistent;
2009 /* Front end dir is a number, which is used as the id. */
2010 info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
2011 dev_set_drvdata(&dev->dev, info);
2013 mutex_lock(&blkfront_mutex);
2014 list_add(&info->info_list, &info_list);
2015 mutex_unlock(&blkfront_mutex);
2020 static int blkif_recover(struct blkfront_info *info)
2022 unsigned int r_index;
2023 struct request *req, *n;
2027 struct blkfront_ring_info *rinfo;
2029 blkfront_gather_backend_features(info);
2030 /* Reset limits changed by blk_mq_update_nr_hw_queues(). */
2031 blkif_set_queue_limits(info);
2032 segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
2033 blk_queue_max_segments(info->rq, segs / GRANTS_PER_PSEG);
2035 for_each_rinfo(info, rinfo, r_index) {
2036 rc = blkfront_setup_indirect(rinfo);
2040 xenbus_switch_state(info->xbdev, XenbusStateConnected);
2042 /* Now safe for us to use the shared ring */
2043 info->connected = BLKIF_STATE_CONNECTED;
2045 for_each_rinfo(info, rinfo, r_index) {
2046 /* Kick any other new requests queued since we resumed */
2047 kick_pending_request_queues(rinfo);
2050 list_for_each_entry_safe(req, n, &info->requests, queuelist) {
2051 /* Requeue pending requests (flush or discard) */
2052 list_del_init(&req->queuelist);
2053 BUG_ON(req->nr_phys_segments > segs);
2054 blk_mq_requeue_request(req, false);
2056 blk_mq_start_stopped_hw_queues(info->rq, true);
2057 blk_mq_kick_requeue_list(info->rq);
2059 while ((bio = bio_list_pop(&info->bio_list)) != NULL) {
2060 /* Traverse the list of pending bios and re-queue them */
2068 * We are reconnecting to the backend, due to a suspend/resume, or a backend
2069 * driver restart. We tear down our blkif structure and recreate it, but
2070 * leave the device-layer structures intact so that this is transparent to the
2071 * rest of the kernel.
2073 static int blkfront_resume(struct xenbus_device *dev)
2075 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2078 struct blkfront_ring_info *rinfo;
2080 dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2082 bio_list_init(&info->bio_list);
2083 INIT_LIST_HEAD(&info->requests);
2084 for_each_rinfo(info, rinfo, i) {
2085 struct bio_list merge_bio;
2086 struct blk_shadow *shadow = rinfo->shadow;
2088 for (j = 0; j < BLK_RING_SIZE(info); j++) {
2090 if (!shadow[j].request)
2094 * Get the bios in the request so we can re-queue them.
2096 if (req_op(shadow[j].request) == REQ_OP_FLUSH ||
2097 req_op(shadow[j].request) == REQ_OP_DISCARD ||
2098 req_op(shadow[j].request) == REQ_OP_SECURE_ERASE ||
2099 shadow[j].request->cmd_flags & REQ_FUA) {
2101 * Flush operations don't contain bios, so
2102 * we need to requeue the whole request
2104 * XXX: but this doesn't make any sense for a
2105 * write with the FUA flag set..
2107 list_add(&shadow[j].request->queuelist, &info->requests);
2110 merge_bio.head = shadow[j].request->bio;
2111 merge_bio.tail = shadow[j].request->biotail;
2112 bio_list_merge(&info->bio_list, &merge_bio);
2113 shadow[j].request->bio = NULL;
2114 blk_mq_end_request(shadow[j].request, BLK_STS_OK);
2118 blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2120 err = talk_to_blkback(dev, info);
2122 blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings);
2125 * We have to wait for the backend to switch to
2126 * connected state, since we want to read which
2127 * features it supports.
2133 static void blkfront_closing(struct blkfront_info *info)
2135 struct xenbus_device *xbdev = info->xbdev;
2136 struct blkfront_ring_info *rinfo;
2139 if (xbdev->state == XenbusStateClosing)
2142 /* No more blkif_request(). */
2143 if (info->rq && info->gd) {
2144 blk_mq_stop_hw_queues(info->rq);
2145 blk_mark_disk_dead(info->gd);
2146 set_capacity(info->gd, 0);
2149 for_each_rinfo(info, rinfo, i) {
2150 /* No more gnttab callback work. */
2151 gnttab_cancel_free_callback(&rinfo->callback);
2153 /* Flush gnttab callback work. Must be done with no locks held. */
2154 flush_work(&rinfo->work);
2157 xenbus_frontend_closed(xbdev);
2160 static void blkfront_setup_discard(struct blkfront_info *info)
2162 info->feature_discard = 1;
2163 info->discard_granularity = xenbus_read_unsigned(info->xbdev->otherend,
2164 "discard-granularity",
2166 info->discard_alignment = xenbus_read_unsigned(info->xbdev->otherend,
2167 "discard-alignment", 0);
2168 info->feature_secdiscard =
2169 !!xenbus_read_unsigned(info->xbdev->otherend, "discard-secure",
2173 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2175 unsigned int psegs, grants, memflags;
2177 struct blkfront_info *info = rinfo->dev_info;
2179 memflags = memalloc_noio_save();
2181 if (info->max_indirect_segments == 0) {
2183 grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2186 * When an extra req is required, the maximum
2187 * grants supported is related to the size of the
2188 * Linux block segment.
2190 grants = GRANTS_PER_PSEG;
2194 grants = info->max_indirect_segments;
2195 psegs = DIV_ROUND_UP(grants, GRANTS_PER_PSEG);
2197 err = fill_grant_buffer(rinfo,
2198 (grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2202 if (!info->feature_persistent && info->max_indirect_segments) {
2204 * We are using indirect descriptors but not persistent
2205 * grants, we need to allocate a set of pages that can be
2206 * used for mapping indirect grefs
2208 int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2210 BUG_ON(!list_empty(&rinfo->indirect_pages));
2211 for (i = 0; i < num; i++) {
2212 struct page *indirect_page = alloc_page(GFP_KERNEL);
2215 list_add(&indirect_page->lru, &rinfo->indirect_pages);
2219 for (i = 0; i < BLK_RING_SIZE(info); i++) {
2220 rinfo->shadow[i].grants_used =
2222 sizeof(rinfo->shadow[i].grants_used[0]),
2224 rinfo->shadow[i].sg = kvcalloc(psegs,
2225 sizeof(rinfo->shadow[i].sg[0]),
2227 if (info->max_indirect_segments)
2228 rinfo->shadow[i].indirect_grants =
2229 kvcalloc(INDIRECT_GREFS(grants),
2230 sizeof(rinfo->shadow[i].indirect_grants[0]),
2232 if ((rinfo->shadow[i].grants_used == NULL) ||
2233 (rinfo->shadow[i].sg == NULL) ||
2234 (info->max_indirect_segments &&
2235 (rinfo->shadow[i].indirect_grants == NULL)))
2237 sg_init_table(rinfo->shadow[i].sg, psegs);
2240 memalloc_noio_restore(memflags);
2245 for (i = 0; i < BLK_RING_SIZE(info); i++) {
2246 kvfree(rinfo->shadow[i].grants_used);
2247 rinfo->shadow[i].grants_used = NULL;
2248 kvfree(rinfo->shadow[i].sg);
2249 rinfo->shadow[i].sg = NULL;
2250 kvfree(rinfo->shadow[i].indirect_grants);
2251 rinfo->shadow[i].indirect_grants = NULL;
2253 if (!list_empty(&rinfo->indirect_pages)) {
2254 struct page *indirect_page, *n;
2255 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2256 list_del(&indirect_page->lru);
2257 __free_page(indirect_page);
2261 memalloc_noio_restore(memflags);
2267 * Gather all backend feature-*
2269 static void blkfront_gather_backend_features(struct blkfront_info *info)
2271 unsigned int indirect_segments;
2273 info->feature_flush = 0;
2274 info->feature_fua = 0;
2277 * If there's no "feature-barrier" defined, then it means
2278 * we're dealing with a very old backend which writes
2279 * synchronously; nothing to do.
2281 * If there are barriers, then we use flush.
2283 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-barrier", 0)) {
2284 info->feature_flush = 1;
2285 info->feature_fua = 1;
2289 * And if there is "feature-flush-cache" use that above
2292 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-flush-cache",
2294 info->feature_flush = 1;
2295 info->feature_fua = 0;
2298 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-discard", 0))
2299 blkfront_setup_discard(info);
2301 if (info->feature_persistent)
2302 info->feature_persistent =
2303 !!xenbus_read_unsigned(info->xbdev->otherend,
2304 "feature-persistent", 0);
2306 indirect_segments = xenbus_read_unsigned(info->xbdev->otherend,
2307 "feature-max-indirect-segments", 0);
2308 if (indirect_segments > xen_blkif_max_segments)
2309 indirect_segments = xen_blkif_max_segments;
2310 if (indirect_segments <= BLKIF_MAX_SEGMENTS_PER_REQUEST)
2311 indirect_segments = 0;
2312 info->max_indirect_segments = indirect_segments;
2314 if (info->feature_persistent) {
2315 mutex_lock(&blkfront_mutex);
2316 schedule_delayed_work(&blkfront_work, HZ * 10);
2317 mutex_unlock(&blkfront_mutex);
2322 * Invoked when the backend is finally 'ready' (and has told produced
2323 * the details about the physical device - #sectors, size, etc).
2325 static void blkfront_connect(struct blkfront_info *info)
2327 unsigned long long sectors;
2328 unsigned long sector_size;
2329 unsigned int physical_sector_size;
2332 struct blkfront_ring_info *rinfo;
2334 switch (info->connected) {
2335 case BLKIF_STATE_CONNECTED:
2337 * Potentially, the back-end may be signalling
2338 * a capacity change; update the capacity.
2340 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2341 "sectors", "%Lu", §ors);
2342 if (XENBUS_EXIST_ERR(err))
2344 printk(KERN_INFO "Setting capacity to %Lu\n",
2346 set_capacity_and_notify(info->gd, sectors);
2349 case BLKIF_STATE_SUSPENDED:
2351 * If we are recovering from suspension, we need to wait
2352 * for the backend to announce it's features before
2353 * reconnecting, at least we need to know if the backend
2354 * supports indirect descriptors, and how many.
2356 blkif_recover(info);
2363 dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2364 __func__, info->xbdev->otherend);
2366 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2367 "sectors", "%llu", §ors,
2368 "info", "%u", &binfo,
2369 "sector-size", "%lu", §or_size,
2372 xenbus_dev_fatal(info->xbdev, err,
2373 "reading backend fields at %s",
2374 info->xbdev->otherend);
2379 * physical-sector-size is a newer field, so old backends may not
2380 * provide this. Assume physical sector size to be the same as
2381 * sector_size in that case.
2383 physical_sector_size = xenbus_read_unsigned(info->xbdev->otherend,
2384 "physical-sector-size",
2386 blkfront_gather_backend_features(info);
2387 for_each_rinfo(info, rinfo, i) {
2388 err = blkfront_setup_indirect(rinfo);
2390 xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2391 info->xbdev->otherend);
2392 blkif_free(info, 0);
2397 err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
2398 physical_sector_size);
2400 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2401 info->xbdev->otherend);
2405 xenbus_switch_state(info->xbdev, XenbusStateConnected);
2407 /* Kick pending requests. */
2408 info->connected = BLKIF_STATE_CONNECTED;
2409 for_each_rinfo(info, rinfo, i)
2410 kick_pending_request_queues(rinfo);
2412 device_add_disk(&info->xbdev->dev, info->gd, NULL);
2418 blkif_free(info, 0);
2423 * Callback received when the backend's state changes.
2425 static void blkback_changed(struct xenbus_device *dev,
2426 enum xenbus_state backend_state)
2428 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2430 dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2432 switch (backend_state) {
2433 case XenbusStateInitWait:
2434 if (dev->state != XenbusStateInitialising)
2436 if (talk_to_blkback(dev, info))
2439 case XenbusStateInitialising:
2440 case XenbusStateInitialised:
2441 case XenbusStateReconfiguring:
2442 case XenbusStateReconfigured:
2443 case XenbusStateUnknown:
2446 case XenbusStateConnected:
2448 * talk_to_blkback sets state to XenbusStateInitialised
2449 * and blkfront_connect sets it to XenbusStateConnected
2450 * (if connection went OK).
2452 * If the backend (or toolstack) decides to poke at backend
2453 * state (and re-trigger the watch by setting the state repeatedly
2454 * to XenbusStateConnected (4)) we need to deal with this.
2455 * This is allowed as this is used to communicate to the guest
2456 * that the size of disk has changed!
2458 if ((dev->state != XenbusStateInitialised) &&
2459 (dev->state != XenbusStateConnected)) {
2460 if (talk_to_blkback(dev, info))
2464 blkfront_connect(info);
2467 case XenbusStateClosed:
2468 if (dev->state == XenbusStateClosed)
2471 case XenbusStateClosing:
2472 blkfront_closing(info);
2477 static int blkfront_remove(struct xenbus_device *xbdev)
2479 struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2481 dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2484 del_gendisk(info->gd);
2486 mutex_lock(&blkfront_mutex);
2487 list_del(&info->info_list);
2488 mutex_unlock(&blkfront_mutex);
2490 blkif_free(info, 0);
2492 xlbd_release_minors(info->gd->first_minor, info->gd->minors);
2493 blk_cleanup_disk(info->gd);
2494 blk_mq_free_tag_set(&info->tag_set);
2501 static int blkfront_is_ready(struct xenbus_device *dev)
2503 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2505 return info->is_ready && info->xbdev;
2508 static const struct block_device_operations xlvbd_block_fops =
2510 .owner = THIS_MODULE,
2511 .getgeo = blkif_getgeo,
2512 .ioctl = blkif_ioctl,
2513 .compat_ioctl = blkdev_compat_ptr_ioctl,
2517 static const struct xenbus_device_id blkfront_ids[] = {
2522 static struct xenbus_driver blkfront_driver = {
2523 .ids = blkfront_ids,
2524 .probe = blkfront_probe,
2525 .remove = blkfront_remove,
2526 .resume = blkfront_resume,
2527 .otherend_changed = blkback_changed,
2528 .is_ready = blkfront_is_ready,
2531 static void purge_persistent_grants(struct blkfront_info *info)
2534 unsigned long flags;
2535 struct blkfront_ring_info *rinfo;
2537 for_each_rinfo(info, rinfo, i) {
2538 struct grant *gnt_list_entry, *tmp;
2540 spin_lock_irqsave(&rinfo->ring_lock, flags);
2542 if (rinfo->persistent_gnts_c == 0) {
2543 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2547 list_for_each_entry_safe(gnt_list_entry, tmp, &rinfo->grants,
2549 if (gnt_list_entry->gref == GRANT_INVALID_REF ||
2550 !gnttab_try_end_foreign_access(gnt_list_entry->gref))
2553 list_del(&gnt_list_entry->node);
2554 rinfo->persistent_gnts_c--;
2555 gnt_list_entry->gref = GRANT_INVALID_REF;
2556 list_add_tail(&gnt_list_entry->node, &rinfo->grants);
2559 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2563 static void blkfront_delay_work(struct work_struct *work)
2565 struct blkfront_info *info;
2566 bool need_schedule_work = false;
2568 mutex_lock(&blkfront_mutex);
2570 list_for_each_entry(info, &info_list, info_list) {
2571 if (info->feature_persistent) {
2572 need_schedule_work = true;
2573 mutex_lock(&info->mutex);
2574 purge_persistent_grants(info);
2575 mutex_unlock(&info->mutex);
2579 if (need_schedule_work)
2580 schedule_delayed_work(&blkfront_work, HZ * 10);
2582 mutex_unlock(&blkfront_mutex);
2585 static int __init xlblk_init(void)
2588 int nr_cpus = num_online_cpus();
2593 if (!xen_has_pv_disk_devices())
2596 if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2597 pr_warn("xen_blk: can't get major %d with name %s\n",
2598 XENVBD_MAJOR, DEV_NAME);
2602 if (xen_blkif_max_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST)
2603 xen_blkif_max_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2605 if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2606 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2607 xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2608 xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2611 if (xen_blkif_max_queues > nr_cpus) {
2612 pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2613 xen_blkif_max_queues, nr_cpus);
2614 xen_blkif_max_queues = nr_cpus;
2617 INIT_DELAYED_WORK(&blkfront_work, blkfront_delay_work);
2619 ret = xenbus_register_frontend(&blkfront_driver);
2621 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2627 module_init(xlblk_init);
2630 static void __exit xlblk_exit(void)
2632 cancel_delayed_work_sync(&blkfront_work);
2634 xenbus_unregister_driver(&blkfront_driver);
2635 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2638 module_exit(xlblk_exit);
2640 MODULE_DESCRIPTION("Xen virtual block device frontend");
2641 MODULE_LICENSE("GPL");
2642 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2643 MODULE_ALIAS("xen:vbd");
2644 MODULE_ALIAS("xenblk");