4 * XenLinux virtual block device driver.
6 * Copyright (c) 2003-2004, Keir Fraser & Steve Hand
7 * Modifications by Mark A. Williamson are (c) Intel Research Cambridge
8 * Copyright (c) 2004, Christian Limpach
9 * Copyright (c) 2004, Andrew Warfield
10 * Copyright (c) 2005, Christopher Clark
11 * Copyright (c) 2005, XenSource Ltd
13 * This program is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU General Public License version 2
15 * as published by the Free Software Foundation; or, when distributed
16 * separately from the Linux kernel or incorporated into other
17 * software packages, subject to the following license:
19 * Permission is hereby granted, free of charge, to any person obtaining a copy
20 * of this source file (the "Software"), to deal in the Software without
21 * restriction, including without limitation the rights to use, copy, modify,
22 * merge, publish, distribute, sublicense, and/or sell copies of the Software,
23 * and to permit persons to whom the Software is furnished to do so, subject to
24 * the following conditions:
26 * The above copyright notice and this permission notice shall be included in
27 * all copies or substantial portions of the Software.
29 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
30 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
31 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
32 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
33 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
34 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
38 #include <linux/interrupt.h>
39 #include <linux/blkdev.h>
40 #include <linux/blk-mq.h>
41 #include <linux/hdreg.h>
42 #include <linux/cdrom.h>
43 #include <linux/module.h>
44 #include <linux/slab.h>
45 #include <linux/mutex.h>
46 #include <linux/scatterlist.h>
47 #include <linux/bitmap.h>
48 #include <linux/list.h>
49 #include <linux/workqueue.h>
50 #include <linux/sched/mm.h>
53 #include <xen/xenbus.h>
54 #include <xen/grant_table.h>
55 #include <xen/events.h>
57 #include <xen/platform_pci.h>
59 #include <xen/interface/grant_table.h>
60 #include <xen/interface/io/blkif.h>
61 #include <xen/interface/io/protocols.h>
63 #include <asm/xen/hypervisor.h>
66 * The minimal size of segment supported by the block framework is PAGE_SIZE.
67 * When Linux is using a different page size than Xen, it may not be possible
68 * to put all the data in a single segment.
69 * This can happen when the backend doesn't support indirect descriptor and
70 * therefore the maximum amount of data that a request can carry is
71 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE = 44KB
73 * Note that we only support one extra request. So the Linux page size
74 * should be <= ( 2 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) =
77 #define HAS_EXTRA_REQ (BLKIF_MAX_SEGMENTS_PER_REQUEST < XEN_PFN_PER_PAGE)
80 BLKIF_STATE_DISCONNECTED,
81 BLKIF_STATE_CONNECTED,
82 BLKIF_STATE_SUSPENDED,
88 struct list_head node;
99 struct blkif_request req;
100 struct request *request;
101 struct grant **grants_used;
102 struct grant **indirect_grants;
103 struct scatterlist *sg;
105 enum blk_req_status status;
107 #define NO_ASSOCIATED_ID ~0UL
109 * Id of the sibling if we ever need 2 requests when handling a
112 unsigned long associated_id;
119 static inline struct blkif_req *blkif_req(struct request *rq)
121 return blk_mq_rq_to_pdu(rq);
124 static DEFINE_MUTEX(blkfront_mutex);
125 static const struct block_device_operations xlvbd_block_fops;
126 static struct delayed_work blkfront_work;
127 static LIST_HEAD(info_list);
130 * Maximum number of segments in indirect requests, the actual value used by
131 * the frontend driver is the minimum of this value and the value provided
132 * by the backend driver.
135 static unsigned int xen_blkif_max_segments = 32;
136 module_param_named(max_indirect_segments, xen_blkif_max_segments, uint, 0444);
137 MODULE_PARM_DESC(max_indirect_segments,
138 "Maximum amount of segments in indirect requests (default is 32)");
140 static unsigned int xen_blkif_max_queues = 4;
141 module_param_named(max_queues, xen_blkif_max_queues, uint, 0444);
142 MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
145 * Maximum order of pages to be used for the shared ring between front and
146 * backend, 4KB page granularity is used.
148 static unsigned int xen_blkif_max_ring_order;
149 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, 0444);
150 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
152 #define BLK_RING_SIZE(info) \
153 __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
156 * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
157 * characters are enough. Define to 20 to keep consistent with backend.
159 #define RINGREF_NAME_LEN (20)
161 * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
163 #define QUEUE_NAME_LEN (17)
167 * Every blkfront device can associate with one or more blkfront_ring_info,
168 * depending on how many hardware queues/rings to be used.
170 struct blkfront_ring_info {
171 /* Lock to protect data in every ring buffer. */
172 spinlock_t ring_lock;
173 struct blkif_front_ring ring;
174 unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
175 unsigned int evtchn, irq;
176 struct work_struct work;
177 struct gnttab_free_callback callback;
178 struct list_head indirect_pages;
179 struct list_head grants;
180 unsigned int persistent_gnts_c;
181 unsigned long shadow_free;
182 struct blkfront_info *dev_info;
183 struct blk_shadow shadow[];
187 * We have one of these per vbd, whether ide, scsi or 'other'. They
188 * hang in private_data off the gendisk structure. We may end up
189 * putting all kinds of interesting stuff here :-)
194 struct xenbus_device *xbdev;
197 unsigned int physical_sector_size;
200 enum blkif_state connected;
201 /* Number of pages per ring buffer. */
202 unsigned int nr_ring_pages;
203 struct request_queue *rq;
204 unsigned int feature_flush:1;
205 unsigned int feature_fua:1;
206 unsigned int feature_discard:1;
207 unsigned int feature_secdiscard:1;
208 unsigned int feature_persistent:1;
209 unsigned int discard_granularity;
210 unsigned int discard_alignment;
211 /* Number of 4KB segments handled */
212 unsigned int max_indirect_segments;
214 struct blk_mq_tag_set tag_set;
215 struct blkfront_ring_info *rinfo;
216 unsigned int nr_rings;
217 unsigned int rinfo_size;
218 /* Save uncomplete reqs and bios for migration. */
219 struct list_head requests;
220 struct bio_list bio_list;
221 struct list_head info_list;
224 static unsigned int nr_minors;
225 static unsigned long *minors;
226 static DEFINE_SPINLOCK(minor_lock);
228 #define GRANT_INVALID_REF 0
230 #define PARTS_PER_DISK 16
231 #define PARTS_PER_EXT_DISK 256
233 #define BLKIF_MAJOR(dev) ((dev)>>8)
234 #define BLKIF_MINOR(dev) ((dev) & 0xff)
237 #define EXTENDED (1<<EXT_SHIFT)
238 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
239 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
240 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
241 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
242 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
243 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
245 #define DEV_NAME "xvd" /* name in /dev */
248 * Grants are always the same size as a Xen page (i.e 4KB).
249 * A physical segment is always the same size as a Linux page.
250 * Number of grants per physical segment
252 #define GRANTS_PER_PSEG (PAGE_SIZE / XEN_PAGE_SIZE)
254 #define GRANTS_PER_INDIRECT_FRAME \
255 (XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
257 #define INDIRECT_GREFS(_grants) \
258 DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
260 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
261 static void blkfront_gather_backend_features(struct blkfront_info *info);
262 static int negotiate_mq(struct blkfront_info *info);
264 #define for_each_rinfo(info, ptr, idx) \
265 for ((ptr) = (info)->rinfo, (idx) = 0; \
266 (idx) < (info)->nr_rings; \
267 (idx)++, (ptr) = (void *)(ptr) + (info)->rinfo_size)
269 static inline struct blkfront_ring_info *
270 get_rinfo(const struct blkfront_info *info, unsigned int i)
272 BUG_ON(i >= info->nr_rings);
273 return (void *)info->rinfo + i * info->rinfo_size;
276 static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
278 unsigned long free = rinfo->shadow_free;
280 BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
281 rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
282 rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
286 static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
289 if (rinfo->shadow[id].req.u.rw.id != id)
291 if (rinfo->shadow[id].request == NULL)
293 rinfo->shadow[id].req.u.rw.id = rinfo->shadow_free;
294 rinfo->shadow[id].request = NULL;
295 rinfo->shadow_free = id;
299 static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
301 struct blkfront_info *info = rinfo->dev_info;
302 struct page *granted_page;
303 struct grant *gnt_list_entry, *n;
307 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
311 if (info->feature_persistent) {
312 granted_page = alloc_page(GFP_NOIO);
314 kfree(gnt_list_entry);
317 gnt_list_entry->page = granted_page;
320 gnt_list_entry->gref = GRANT_INVALID_REF;
321 list_add(&gnt_list_entry->node, &rinfo->grants);
328 list_for_each_entry_safe(gnt_list_entry, n,
329 &rinfo->grants, node) {
330 list_del(&gnt_list_entry->node);
331 if (info->feature_persistent)
332 __free_page(gnt_list_entry->page);
333 kfree(gnt_list_entry);
340 static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
342 struct grant *gnt_list_entry;
344 BUG_ON(list_empty(&rinfo->grants));
345 gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
347 list_del(&gnt_list_entry->node);
349 if (gnt_list_entry->gref != GRANT_INVALID_REF)
350 rinfo->persistent_gnts_c--;
352 return gnt_list_entry;
355 static inline void grant_foreign_access(const struct grant *gnt_list_entry,
356 const struct blkfront_info *info)
358 gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
359 info->xbdev->otherend_id,
360 gnt_list_entry->page,
364 static struct grant *get_grant(grant_ref_t *gref_head,
366 struct blkfront_ring_info *rinfo)
368 struct grant *gnt_list_entry = get_free_grant(rinfo);
369 struct blkfront_info *info = rinfo->dev_info;
371 if (gnt_list_entry->gref != GRANT_INVALID_REF)
372 return gnt_list_entry;
374 /* Assign a gref to this page */
375 gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
376 BUG_ON(gnt_list_entry->gref == -ENOSPC);
377 if (info->feature_persistent)
378 grant_foreign_access(gnt_list_entry, info);
380 /* Grant access to the GFN passed by the caller */
381 gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
382 info->xbdev->otherend_id,
386 return gnt_list_entry;
389 static struct grant *get_indirect_grant(grant_ref_t *gref_head,
390 struct blkfront_ring_info *rinfo)
392 struct grant *gnt_list_entry = get_free_grant(rinfo);
393 struct blkfront_info *info = rinfo->dev_info;
395 if (gnt_list_entry->gref != GRANT_INVALID_REF)
396 return gnt_list_entry;
398 /* Assign a gref to this page */
399 gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
400 BUG_ON(gnt_list_entry->gref == -ENOSPC);
401 if (!info->feature_persistent) {
402 struct page *indirect_page;
404 /* Fetch a pre-allocated page to use for indirect grefs */
405 BUG_ON(list_empty(&rinfo->indirect_pages));
406 indirect_page = list_first_entry(&rinfo->indirect_pages,
408 list_del(&indirect_page->lru);
409 gnt_list_entry->page = indirect_page;
411 grant_foreign_access(gnt_list_entry, info);
413 return gnt_list_entry;
416 static const char *op_name(int op)
418 static const char *const names[] = {
419 [BLKIF_OP_READ] = "read",
420 [BLKIF_OP_WRITE] = "write",
421 [BLKIF_OP_WRITE_BARRIER] = "barrier",
422 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
423 [BLKIF_OP_DISCARD] = "discard" };
425 if (op < 0 || op >= ARRAY_SIZE(names))
433 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
435 unsigned int end = minor + nr;
438 if (end > nr_minors) {
439 unsigned long *bitmap, *old;
441 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
446 spin_lock(&minor_lock);
447 if (end > nr_minors) {
449 memcpy(bitmap, minors,
450 BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
452 nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
455 spin_unlock(&minor_lock);
459 spin_lock(&minor_lock);
460 if (find_next_bit(minors, end, minor) >= end) {
461 bitmap_set(minors, minor, nr);
465 spin_unlock(&minor_lock);
470 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
472 unsigned int end = minor + nr;
474 BUG_ON(end > nr_minors);
475 spin_lock(&minor_lock);
476 bitmap_clear(minors, minor, nr);
477 spin_unlock(&minor_lock);
480 static void blkif_restart_queue_callback(void *arg)
482 struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
483 schedule_work(&rinfo->work);
486 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
488 /* We don't have real geometry info, but let's at least return
489 values consistent with the size of the device */
490 sector_t nsect = get_capacity(bd->bd_disk);
491 sector_t cylinders = nsect;
495 sector_div(cylinders, hg->heads * hg->sectors);
496 hg->cylinders = cylinders;
497 if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
498 hg->cylinders = 0xffff;
502 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
503 unsigned command, unsigned long argument)
505 struct blkfront_info *info = bdev->bd_disk->private_data;
508 dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
509 command, (long)argument);
512 case CDROMMULTISESSION:
513 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
514 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
515 if (put_user(0, (char __user *)(argument + i)))
519 case CDROM_GET_CAPABILITY: {
520 struct gendisk *gd = info->gd;
521 if (gd->flags & GENHD_FL_CD)
527 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
529 return -EINVAL; /* same return as native Linux */
535 static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
537 struct blkif_request **ring_req)
541 *ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
542 rinfo->ring.req_prod_pvt++;
544 id = get_id_from_freelist(rinfo);
545 rinfo->shadow[id].request = req;
546 rinfo->shadow[id].status = REQ_WAITING;
547 rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
549 (*ring_req)->u.rw.id = id;
554 static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
556 struct blkfront_info *info = rinfo->dev_info;
557 struct blkif_request *ring_req;
560 /* Fill out a communications ring structure. */
561 id = blkif_ring_get_request(rinfo, req, &ring_req);
563 ring_req->operation = BLKIF_OP_DISCARD;
564 ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
565 ring_req->u.discard.id = id;
566 ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
567 if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard)
568 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
570 ring_req->u.discard.flag = 0;
572 /* Keep a private copy so we can reissue requests when recovering. */
573 rinfo->shadow[id].req = *ring_req;
578 struct setup_rw_req {
579 unsigned int grant_idx;
580 struct blkif_request_segment *segments;
581 struct blkfront_ring_info *rinfo;
582 struct blkif_request *ring_req;
583 grant_ref_t gref_head;
585 /* Only used when persistent grant is used and it's a read request */
587 unsigned int bvec_off;
590 bool require_extra_req;
591 struct blkif_request *extra_ring_req;
594 static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
595 unsigned int len, void *data)
597 struct setup_rw_req *setup = data;
599 struct grant *gnt_list_entry;
600 unsigned int fsect, lsect;
601 /* Convenient aliases */
602 unsigned int grant_idx = setup->grant_idx;
603 struct blkif_request *ring_req = setup->ring_req;
604 struct blkfront_ring_info *rinfo = setup->rinfo;
606 * We always use the shadow of the first request to store the list
607 * of grant associated to the block I/O request. This made the
608 * completion more easy to handle even if the block I/O request is
611 struct blk_shadow *shadow = &rinfo->shadow[setup->id];
613 if (unlikely(setup->require_extra_req &&
614 grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
616 * We are using the second request, setup grant_idx
617 * to be the index of the segment array.
619 grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
620 ring_req = setup->extra_ring_req;
623 if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
624 (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
626 kunmap_atomic(setup->segments);
628 n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
629 gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
630 shadow->indirect_grants[n] = gnt_list_entry;
631 setup->segments = kmap_atomic(gnt_list_entry->page);
632 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
635 gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
636 ref = gnt_list_entry->gref;
638 * All the grants are stored in the shadow of the first
639 * request. Therefore we have to use the global index.
641 shadow->grants_used[setup->grant_idx] = gnt_list_entry;
643 if (setup->need_copy) {
646 shared_data = kmap_atomic(gnt_list_entry->page);
648 * this does not wipe data stored outside the
649 * range sg->offset..sg->offset+sg->length.
650 * Therefore, blkback *could* see data from
651 * previous requests. This is OK as long as
652 * persistent grants are shared with just one
653 * domain. It may need refactoring if this
656 memcpy(shared_data + offset,
657 setup->bvec_data + setup->bvec_off,
660 kunmap_atomic(shared_data);
661 setup->bvec_off += len;
665 lsect = fsect + (len >> 9) - 1;
666 if (ring_req->operation != BLKIF_OP_INDIRECT) {
667 ring_req->u.rw.seg[grant_idx] =
668 (struct blkif_request_segment) {
671 .last_sect = lsect };
673 setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
674 (struct blkif_request_segment) {
677 .last_sect = lsect };
680 (setup->grant_idx)++;
683 static void blkif_setup_extra_req(struct blkif_request *first,
684 struct blkif_request *second)
686 uint16_t nr_segments = first->u.rw.nr_segments;
689 * The second request is only present when the first request uses
690 * all its segments. It's always the continuity of the first one.
692 first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
694 second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
695 second->u.rw.sector_number = first->u.rw.sector_number +
696 (BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
698 second->u.rw.handle = first->u.rw.handle;
699 second->operation = first->operation;
702 static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
704 struct blkfront_info *info = rinfo->dev_info;
705 struct blkif_request *ring_req, *extra_ring_req = NULL;
706 unsigned long id, extra_id = NO_ASSOCIATED_ID;
707 bool require_extra_req = false;
709 struct setup_rw_req setup = {
713 .need_copy = rq_data_dir(req) && info->feature_persistent,
717 * Used to store if we are able to queue the request by just using
718 * existing persistent grants, or if we have to get new grants,
719 * as there are not sufficiently many free.
721 bool new_persistent_gnts = false;
722 struct scatterlist *sg;
723 int num_sg, max_grefs, num_grant;
725 max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
726 if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
728 * If we are using indirect segments we need to account
729 * for the indirect grefs used in the request.
731 max_grefs += INDIRECT_GREFS(max_grefs);
733 /* Check if we have enough persistent grants to allocate a requests */
734 if (rinfo->persistent_gnts_c < max_grefs) {
735 new_persistent_gnts = true;
737 if (gnttab_alloc_grant_references(
738 max_grefs - rinfo->persistent_gnts_c,
739 &setup.gref_head) < 0) {
740 gnttab_request_free_callback(
742 blkif_restart_queue_callback,
744 max_grefs - rinfo->persistent_gnts_c);
749 /* Fill out a communications ring structure. */
750 id = blkif_ring_get_request(rinfo, req, &ring_req);
752 num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
754 /* Calculate the number of grant used */
755 for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
756 num_grant += gnttab_count_grant(sg->offset, sg->length);
758 require_extra_req = info->max_indirect_segments == 0 &&
759 num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
760 BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
762 rinfo->shadow[id].num_sg = num_sg;
763 if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
764 likely(!require_extra_req)) {
766 * The indirect operation can only be a BLKIF_OP_READ or
769 BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA);
770 ring_req->operation = BLKIF_OP_INDIRECT;
771 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
772 BLKIF_OP_WRITE : BLKIF_OP_READ;
773 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
774 ring_req->u.indirect.handle = info->handle;
775 ring_req->u.indirect.nr_segments = num_grant;
777 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
778 ring_req->u.rw.handle = info->handle;
779 ring_req->operation = rq_data_dir(req) ?
780 BLKIF_OP_WRITE : BLKIF_OP_READ;
781 if (req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA) {
783 * Ideally we can do an unordered flush-to-disk.
784 * In case the backend onlysupports barriers, use that.
785 * A barrier request a superset of FUA, so we can
786 * implement it the same way. (It's also a FLUSH+FUA,
787 * since it is guaranteed ordered WRT previous writes.)
789 if (info->feature_flush && info->feature_fua)
790 ring_req->operation =
791 BLKIF_OP_WRITE_BARRIER;
792 else if (info->feature_flush)
793 ring_req->operation =
794 BLKIF_OP_FLUSH_DISKCACHE;
796 ring_req->operation = 0;
798 ring_req->u.rw.nr_segments = num_grant;
799 if (unlikely(require_extra_req)) {
800 extra_id = blkif_ring_get_request(rinfo, req,
803 * Only the first request contains the scatter-gather
806 rinfo->shadow[extra_id].num_sg = 0;
808 blkif_setup_extra_req(ring_req, extra_ring_req);
810 /* Link the 2 requests together */
811 rinfo->shadow[extra_id].associated_id = id;
812 rinfo->shadow[id].associated_id = extra_id;
816 setup.ring_req = ring_req;
819 setup.require_extra_req = require_extra_req;
820 if (unlikely(require_extra_req))
821 setup.extra_ring_req = extra_ring_req;
823 for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
824 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
826 if (setup.need_copy) {
827 setup.bvec_off = sg->offset;
828 setup.bvec_data = kmap_atomic(sg_page(sg));
831 gnttab_foreach_grant_in_range(sg_page(sg),
834 blkif_setup_rw_req_grant,
838 kunmap_atomic(setup.bvec_data);
841 kunmap_atomic(setup.segments);
843 /* Keep a private copy so we can reissue requests when recovering. */
844 rinfo->shadow[id].req = *ring_req;
845 if (unlikely(require_extra_req))
846 rinfo->shadow[extra_id].req = *extra_ring_req;
848 if (new_persistent_gnts)
849 gnttab_free_grant_references(setup.gref_head);
855 * Generate a Xen blkfront IO request from a blk layer request. Reads
856 * and writes are handled as expected.
858 * @req: a request struct
860 static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
862 if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
865 if (unlikely(req_op(req) == REQ_OP_DISCARD ||
866 req_op(req) == REQ_OP_SECURE_ERASE))
867 return blkif_queue_discard_req(req, rinfo);
869 return blkif_queue_rw_req(req, rinfo);
872 static inline void flush_requests(struct blkfront_ring_info *rinfo)
876 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
879 notify_remote_via_irq(rinfo->irq);
882 static inline bool blkif_request_flush_invalid(struct request *req,
883 struct blkfront_info *info)
885 return (blk_rq_is_passthrough(req) ||
886 ((req_op(req) == REQ_OP_FLUSH) &&
887 !info->feature_flush) ||
888 ((req->cmd_flags & REQ_FUA) &&
889 !info->feature_fua));
892 static blk_status_t blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
893 const struct blk_mq_queue_data *qd)
896 int qid = hctx->queue_num;
897 struct blkfront_info *info = hctx->queue->queuedata;
898 struct blkfront_ring_info *rinfo = NULL;
900 rinfo = get_rinfo(info, qid);
901 blk_mq_start_request(qd->rq);
902 spin_lock_irqsave(&rinfo->ring_lock, flags);
903 if (RING_FULL(&rinfo->ring))
906 if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
909 if (blkif_queue_request(qd->rq, rinfo))
912 flush_requests(rinfo);
913 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
917 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
918 return BLK_STS_IOERR;
921 blk_mq_stop_hw_queue(hctx);
922 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
923 return BLK_STS_DEV_RESOURCE;
926 static void blkif_complete_rq(struct request *rq)
928 blk_mq_end_request(rq, blkif_req(rq)->error);
931 static const struct blk_mq_ops blkfront_mq_ops = {
932 .queue_rq = blkif_queue_rq,
933 .complete = blkif_complete_rq,
936 static void blkif_set_queue_limits(struct blkfront_info *info)
938 struct request_queue *rq = info->rq;
939 struct gendisk *gd = info->gd;
940 unsigned int segments = info->max_indirect_segments ? :
941 BLKIF_MAX_SEGMENTS_PER_REQUEST;
943 blk_queue_flag_set(QUEUE_FLAG_VIRT, rq);
945 if (info->feature_discard) {
946 blk_queue_flag_set(QUEUE_FLAG_DISCARD, rq);
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 int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
972 unsigned int physical_sector_size)
974 struct request_queue *rq;
975 struct blkfront_info *info = gd->private_data;
977 memset(&info->tag_set, 0, sizeof(info->tag_set));
978 info->tag_set.ops = &blkfront_mq_ops;
979 info->tag_set.nr_hw_queues = info->nr_rings;
980 if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
982 * When indirect descriptior is not supported, the I/O request
983 * will be split between multiple request in the ring.
984 * To avoid problems when sending the request, divide by
985 * 2 the depth of the queue.
987 info->tag_set.queue_depth = BLK_RING_SIZE(info) / 2;
989 info->tag_set.queue_depth = BLK_RING_SIZE(info);
990 info->tag_set.numa_node = NUMA_NO_NODE;
991 info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
992 info->tag_set.cmd_size = sizeof(struct blkif_req);
993 info->tag_set.driver_data = info;
995 if (blk_mq_alloc_tag_set(&info->tag_set))
997 rq = blk_mq_init_queue(&info->tag_set);
999 blk_mq_free_tag_set(&info->tag_set);
1003 rq->queuedata = info;
1004 info->rq = gd->queue = rq;
1006 info->sector_size = sector_size;
1007 info->physical_sector_size = physical_sector_size;
1008 blkif_set_queue_limits(info);
1013 static const char *flush_info(struct blkfront_info *info)
1015 if (info->feature_flush && info->feature_fua)
1016 return "barrier: enabled;";
1017 else if (info->feature_flush)
1018 return "flush diskcache: enabled;";
1020 return "barrier or flush: disabled;";
1023 static void xlvbd_flush(struct blkfront_info *info)
1025 blk_queue_write_cache(info->rq, info->feature_flush ? true : false,
1026 info->feature_fua ? true : false);
1027 pr_info("blkfront: %s: %s %s %s %s %s\n",
1028 info->gd->disk_name, flush_info(info),
1029 "persistent grants:", info->feature_persistent ?
1030 "enabled;" : "disabled;", "indirect descriptors:",
1031 info->max_indirect_segments ? "enabled;" : "disabled;");
1034 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
1037 major = BLKIF_MAJOR(vdevice);
1038 *minor = BLKIF_MINOR(vdevice);
1040 case XEN_IDE0_MAJOR:
1041 *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
1042 *minor = ((*minor / 64) * PARTS_PER_DISK) +
1043 EMULATED_HD_DISK_MINOR_OFFSET;
1045 case XEN_IDE1_MAJOR:
1046 *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1047 *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1048 EMULATED_HD_DISK_MINOR_OFFSET;
1050 case XEN_SCSI_DISK0_MAJOR:
1051 *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1052 *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1054 case XEN_SCSI_DISK1_MAJOR:
1055 case XEN_SCSI_DISK2_MAJOR:
1056 case XEN_SCSI_DISK3_MAJOR:
1057 case XEN_SCSI_DISK4_MAJOR:
1058 case XEN_SCSI_DISK5_MAJOR:
1059 case XEN_SCSI_DISK6_MAJOR:
1060 case XEN_SCSI_DISK7_MAJOR:
1061 *offset = (*minor / PARTS_PER_DISK) +
1062 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1063 EMULATED_SD_DISK_NAME_OFFSET;
1065 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1066 EMULATED_SD_DISK_MINOR_OFFSET;
1068 case XEN_SCSI_DISK8_MAJOR:
1069 case XEN_SCSI_DISK9_MAJOR:
1070 case XEN_SCSI_DISK10_MAJOR:
1071 case XEN_SCSI_DISK11_MAJOR:
1072 case XEN_SCSI_DISK12_MAJOR:
1073 case XEN_SCSI_DISK13_MAJOR:
1074 case XEN_SCSI_DISK14_MAJOR:
1075 case XEN_SCSI_DISK15_MAJOR:
1076 *offset = (*minor / PARTS_PER_DISK) +
1077 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1078 EMULATED_SD_DISK_NAME_OFFSET;
1080 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1081 EMULATED_SD_DISK_MINOR_OFFSET;
1084 *offset = *minor / PARTS_PER_DISK;
1087 printk(KERN_WARNING "blkfront: your disk configuration is "
1088 "incorrect, please use an xvd device instead\n");
1094 static char *encode_disk_name(char *ptr, unsigned int n)
1097 ptr = encode_disk_name(ptr, n / 26 - 1);
1098 *ptr = 'a' + n % 26;
1102 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1103 struct blkfront_info *info,
1104 u16 vdisk_info, u16 sector_size,
1105 unsigned int physical_sector_size)
1110 unsigned int offset;
1115 BUG_ON(info->gd != NULL);
1116 BUG_ON(info->rq != NULL);
1118 if ((info->vdevice>>EXT_SHIFT) > 1) {
1119 /* this is above the extended range; something is wrong */
1120 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1124 if (!VDEV_IS_EXTENDED(info->vdevice)) {
1125 err = xen_translate_vdev(info->vdevice, &minor, &offset);
1128 nr_parts = PARTS_PER_DISK;
1130 minor = BLKIF_MINOR_EXT(info->vdevice);
1131 nr_parts = PARTS_PER_EXT_DISK;
1132 offset = minor / nr_parts;
1133 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1134 printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1135 "emulated IDE disks,\n\t choose an xvd device name"
1136 "from xvde on\n", info->vdevice);
1138 if (minor >> MINORBITS) {
1139 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1140 info->vdevice, minor);
1144 if ((minor % nr_parts) == 0)
1145 nr_minors = nr_parts;
1147 err = xlbd_reserve_minors(minor, nr_minors);
1152 gd = alloc_disk(nr_minors);
1156 strcpy(gd->disk_name, DEV_NAME);
1157 ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1158 BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1162 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1163 "%d", minor & (nr_parts - 1));
1165 gd->major = XENVBD_MAJOR;
1166 gd->first_minor = minor;
1167 gd->fops = &xlvbd_block_fops;
1168 gd->private_data = info;
1169 set_capacity(gd, capacity);
1171 if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size)) {
1178 if (vdisk_info & VDISK_READONLY)
1181 if (vdisk_info & VDISK_REMOVABLE)
1182 gd->flags |= GENHD_FL_REMOVABLE;
1184 if (vdisk_info & VDISK_CDROM)
1185 gd->flags |= GENHD_FL_CD;
1190 xlbd_release_minors(minor, nr_minors);
1195 static void xlvbd_release_gendisk(struct blkfront_info *info)
1197 unsigned int minor, nr_minors, i;
1198 struct blkfront_ring_info *rinfo;
1200 if (info->rq == NULL)
1203 /* No more blkif_request(). */
1204 blk_mq_stop_hw_queues(info->rq);
1206 for_each_rinfo(info, rinfo, i) {
1207 /* No more gnttab callback work. */
1208 gnttab_cancel_free_callback(&rinfo->callback);
1210 /* Flush gnttab callback work. Must be done with no locks held. */
1211 flush_work(&rinfo->work);
1214 del_gendisk(info->gd);
1216 minor = info->gd->first_minor;
1217 nr_minors = info->gd->minors;
1218 xlbd_release_minors(minor, nr_minors);
1220 blk_cleanup_queue(info->rq);
1221 blk_mq_free_tag_set(&info->tag_set);
1228 /* Already hold rinfo->ring_lock. */
1229 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1231 if (!RING_FULL(&rinfo->ring))
1232 blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1235 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1237 unsigned long flags;
1239 spin_lock_irqsave(&rinfo->ring_lock, flags);
1240 kick_pending_request_queues_locked(rinfo);
1241 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1244 static void blkif_restart_queue(struct work_struct *work)
1246 struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1248 if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1249 kick_pending_request_queues(rinfo);
1252 static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1254 struct grant *persistent_gnt, *n;
1255 struct blkfront_info *info = rinfo->dev_info;
1259 * Remove indirect pages, this only happens when using indirect
1260 * descriptors but not persistent grants
1262 if (!list_empty(&rinfo->indirect_pages)) {
1263 struct page *indirect_page, *n;
1265 BUG_ON(info->feature_persistent);
1266 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1267 list_del(&indirect_page->lru);
1268 __free_page(indirect_page);
1272 /* Remove all persistent grants. */
1273 if (!list_empty(&rinfo->grants)) {
1274 list_for_each_entry_safe(persistent_gnt, n,
1275 &rinfo->grants, node) {
1276 list_del(&persistent_gnt->node);
1277 if (persistent_gnt->gref != GRANT_INVALID_REF) {
1278 gnttab_end_foreign_access(persistent_gnt->gref,
1280 rinfo->persistent_gnts_c--;
1282 if (info->feature_persistent)
1283 __free_page(persistent_gnt->page);
1284 kfree(persistent_gnt);
1287 BUG_ON(rinfo->persistent_gnts_c != 0);
1289 for (i = 0; i < BLK_RING_SIZE(info); i++) {
1291 * Clear persistent grants present in requests already
1292 * on the shared ring
1294 if (!rinfo->shadow[i].request)
1297 segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1298 rinfo->shadow[i].req.u.indirect.nr_segments :
1299 rinfo->shadow[i].req.u.rw.nr_segments;
1300 for (j = 0; j < segs; j++) {
1301 persistent_gnt = rinfo->shadow[i].grants_used[j];
1302 gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1303 if (info->feature_persistent)
1304 __free_page(persistent_gnt->page);
1305 kfree(persistent_gnt);
1308 if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1310 * If this is not an indirect operation don't try to
1311 * free indirect segments
1315 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1316 persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1317 gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1318 __free_page(persistent_gnt->page);
1319 kfree(persistent_gnt);
1323 kvfree(rinfo->shadow[i].grants_used);
1324 rinfo->shadow[i].grants_used = NULL;
1325 kvfree(rinfo->shadow[i].indirect_grants);
1326 rinfo->shadow[i].indirect_grants = NULL;
1327 kvfree(rinfo->shadow[i].sg);
1328 rinfo->shadow[i].sg = NULL;
1331 /* No more gnttab callback work. */
1332 gnttab_cancel_free_callback(&rinfo->callback);
1334 /* Flush gnttab callback work. Must be done with no locks held. */
1335 flush_work(&rinfo->work);
1337 /* Free resources associated with old device channel. */
1338 for (i = 0; i < info->nr_ring_pages; i++) {
1339 if (rinfo->ring_ref[i] != GRANT_INVALID_REF) {
1340 gnttab_end_foreign_access(rinfo->ring_ref[i], 0, 0);
1341 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1344 free_pages((unsigned long)rinfo->ring.sring, get_order(info->nr_ring_pages * XEN_PAGE_SIZE));
1345 rinfo->ring.sring = NULL;
1348 unbind_from_irqhandler(rinfo->irq, rinfo);
1349 rinfo->evtchn = rinfo->irq = 0;
1352 static void blkif_free(struct blkfront_info *info, int suspend)
1355 struct blkfront_ring_info *rinfo;
1357 /* Prevent new requests being issued until we fix things up. */
1358 info->connected = suspend ?
1359 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1360 /* No more blkif_request(). */
1362 blk_mq_stop_hw_queues(info->rq);
1364 for_each_rinfo(info, rinfo, i)
1365 blkif_free_ring(rinfo);
1367 kvfree(info->rinfo);
1372 struct copy_from_grant {
1373 const struct blk_shadow *s;
1374 unsigned int grant_idx;
1375 unsigned int bvec_offset;
1379 static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1380 unsigned int len, void *data)
1382 struct copy_from_grant *info = data;
1384 /* Convenient aliases */
1385 const struct blk_shadow *s = info->s;
1387 shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1389 memcpy(info->bvec_data + info->bvec_offset,
1390 shared_data + offset, len);
1392 info->bvec_offset += len;
1395 kunmap_atomic(shared_data);
1398 static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1402 case BLKIF_RSP_OKAY:
1404 case BLKIF_RSP_EOPNOTSUPP:
1405 return REQ_EOPNOTSUPP;
1406 case BLKIF_RSP_ERROR:
1413 * Get the final status of the block request based on two ring response
1415 static int blkif_get_final_status(enum blk_req_status s1,
1416 enum blk_req_status s2)
1418 BUG_ON(s1 == REQ_WAITING);
1419 BUG_ON(s2 == REQ_WAITING);
1421 if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1422 return BLKIF_RSP_ERROR;
1423 else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1424 return BLKIF_RSP_EOPNOTSUPP;
1425 return BLKIF_RSP_OKAY;
1428 static bool blkif_completion(unsigned long *id,
1429 struct blkfront_ring_info *rinfo,
1430 struct blkif_response *bret)
1433 struct scatterlist *sg;
1434 int num_sg, num_grant;
1435 struct blkfront_info *info = rinfo->dev_info;
1436 struct blk_shadow *s = &rinfo->shadow[*id];
1437 struct copy_from_grant data = {
1441 num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1442 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1444 /* The I/O request may be split in two. */
1445 if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1446 struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1448 /* Keep the status of the current response in shadow. */
1449 s->status = blkif_rsp_to_req_status(bret->status);
1451 /* Wait the second response if not yet here. */
1452 if (s2->status == REQ_WAITING)
1455 bret->status = blkif_get_final_status(s->status,
1459 * All the grants is stored in the first shadow in order
1460 * to make the completion code simpler.
1462 num_grant += s2->req.u.rw.nr_segments;
1465 * The two responses may not come in order. Only the
1466 * first request will store the scatter-gather list.
1468 if (s2->num_sg != 0) {
1469 /* Update "id" with the ID of the first response. */
1470 *id = s->associated_id;
1475 * We don't need anymore the second request, so recycling
1478 if (add_id_to_freelist(rinfo, s->associated_id))
1479 WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1480 info->gd->disk_name, s->associated_id);
1486 if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1487 for_each_sg(s->sg, sg, num_sg, i) {
1488 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1490 data.bvec_offset = sg->offset;
1491 data.bvec_data = kmap_atomic(sg_page(sg));
1493 gnttab_foreach_grant_in_range(sg_page(sg),
1496 blkif_copy_from_grant,
1499 kunmap_atomic(data.bvec_data);
1502 /* Add the persistent grant into the list of free grants */
1503 for (i = 0; i < num_grant; i++) {
1504 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1506 * If the grant is still mapped by the backend (the
1507 * backend has chosen to make this grant persistent)
1508 * we add it at the head of the list, so it will be
1511 if (!info->feature_persistent)
1512 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1513 s->grants_used[i]->gref);
1514 list_add(&s->grants_used[i]->node, &rinfo->grants);
1515 rinfo->persistent_gnts_c++;
1518 * If the grant is not mapped by the backend we end the
1519 * foreign access and add it to the tail of the list,
1520 * so it will not be picked again unless we run out of
1521 * persistent grants.
1523 gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1524 s->grants_used[i]->gref = GRANT_INVALID_REF;
1525 list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1528 if (s->req.operation == BLKIF_OP_INDIRECT) {
1529 for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1530 if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1531 if (!info->feature_persistent)
1532 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1533 s->indirect_grants[i]->gref);
1534 list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1535 rinfo->persistent_gnts_c++;
1537 struct page *indirect_page;
1539 gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1541 * Add the used indirect page back to the list of
1542 * available pages for indirect grefs.
1544 if (!info->feature_persistent) {
1545 indirect_page = s->indirect_grants[i]->page;
1546 list_add(&indirect_page->lru, &rinfo->indirect_pages);
1548 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1549 list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1557 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1559 struct request *req;
1560 struct blkif_response *bret;
1562 unsigned long flags;
1563 struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1564 struct blkfront_info *info = rinfo->dev_info;
1566 if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
1569 spin_lock_irqsave(&rinfo->ring_lock, flags);
1571 rp = rinfo->ring.sring->rsp_prod;
1572 rmb(); /* Ensure we see queued responses up to 'rp'. */
1574 for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1577 bret = RING_GET_RESPONSE(&rinfo->ring, i);
1580 * The backend has messed up and given us an id that we would
1581 * never have given to it (we stamp it up to BLK_RING_SIZE -
1582 * look in get_id_from_freelist.
1584 if (id >= BLK_RING_SIZE(info)) {
1585 WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1586 info->gd->disk_name, op_name(bret->operation), id);
1587 /* We can't safely get the 'struct request' as
1588 * the id is busted. */
1591 req = rinfo->shadow[id].request;
1593 if (bret->operation != BLKIF_OP_DISCARD) {
1595 * We may need to wait for an extra response if the
1596 * I/O request is split in 2
1598 if (!blkif_completion(&id, rinfo, bret))
1602 if (add_id_to_freelist(rinfo, id)) {
1603 WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1604 info->gd->disk_name, op_name(bret->operation), id);
1608 if (bret->status == BLKIF_RSP_OKAY)
1609 blkif_req(req)->error = BLK_STS_OK;
1611 blkif_req(req)->error = BLK_STS_IOERR;
1613 switch (bret->operation) {
1614 case BLKIF_OP_DISCARD:
1615 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1616 struct request_queue *rq = info->rq;
1617 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1618 info->gd->disk_name, op_name(bret->operation));
1619 blkif_req(req)->error = BLK_STS_NOTSUPP;
1620 info->feature_discard = 0;
1621 info->feature_secdiscard = 0;
1622 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1623 blk_queue_flag_clear(QUEUE_FLAG_SECERASE, rq);
1626 case BLKIF_OP_FLUSH_DISKCACHE:
1627 case BLKIF_OP_WRITE_BARRIER:
1628 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1629 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1630 info->gd->disk_name, op_name(bret->operation));
1631 blkif_req(req)->error = BLK_STS_NOTSUPP;
1633 if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1634 rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1635 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1636 info->gd->disk_name, op_name(bret->operation));
1637 blkif_req(req)->error = BLK_STS_NOTSUPP;
1639 if (unlikely(blkif_req(req)->error)) {
1640 if (blkif_req(req)->error == BLK_STS_NOTSUPP)
1641 blkif_req(req)->error = BLK_STS_OK;
1642 info->feature_fua = 0;
1643 info->feature_flush = 0;
1648 case BLKIF_OP_WRITE:
1649 if (unlikely(bret->status != BLKIF_RSP_OKAY))
1650 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1651 "request: %x\n", bret->status);
1658 if (likely(!blk_should_fake_timeout(req->q)))
1659 blk_mq_complete_request(req);
1662 rinfo->ring.rsp_cons = i;
1664 if (i != rinfo->ring.req_prod_pvt) {
1666 RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1670 rinfo->ring.sring->rsp_event = i + 1;
1672 kick_pending_request_queues_locked(rinfo);
1674 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1680 static int setup_blkring(struct xenbus_device *dev,
1681 struct blkfront_ring_info *rinfo)
1683 struct blkif_sring *sring;
1685 struct blkfront_info *info = rinfo->dev_info;
1686 unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1687 grant_ref_t gref[XENBUS_MAX_RING_GRANTS];
1689 for (i = 0; i < info->nr_ring_pages; i++)
1690 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1692 sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
1693 get_order(ring_size));
1695 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1698 SHARED_RING_INIT(sring);
1699 FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1701 err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref);
1703 free_pages((unsigned long)sring, get_order(ring_size));
1704 rinfo->ring.sring = NULL;
1707 for (i = 0; i < info->nr_ring_pages; i++)
1708 rinfo->ring_ref[i] = gref[i];
1710 err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1714 err = bind_evtchn_to_irqhandler(rinfo->evtchn, blkif_interrupt, 0,
1717 xenbus_dev_fatal(dev, err,
1718 "bind_evtchn_to_irqhandler failed");
1725 blkif_free(info, 0);
1730 * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1731 * ring buffer may have multi pages depending on ->nr_ring_pages.
1733 static int write_per_ring_nodes(struct xenbus_transaction xbt,
1734 struct blkfront_ring_info *rinfo, const char *dir)
1738 const char *message = NULL;
1739 struct blkfront_info *info = rinfo->dev_info;
1741 if (info->nr_ring_pages == 1) {
1742 err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1744 message = "writing ring-ref";
1745 goto abort_transaction;
1748 for (i = 0; i < info->nr_ring_pages; i++) {
1749 char ring_ref_name[RINGREF_NAME_LEN];
1751 snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1752 err = xenbus_printf(xbt, dir, ring_ref_name,
1753 "%u", rinfo->ring_ref[i]);
1755 message = "writing ring-ref";
1756 goto abort_transaction;
1761 err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1763 message = "writing event-channel";
1764 goto abort_transaction;
1770 xenbus_transaction_end(xbt, 1);
1772 xenbus_dev_fatal(info->xbdev, err, "%s", message);
1777 static void free_info(struct blkfront_info *info)
1779 list_del(&info->info_list);
1783 /* Common code used when first setting up, and when resuming. */
1784 static int talk_to_blkback(struct xenbus_device *dev,
1785 struct blkfront_info *info)
1787 const char *message = NULL;
1788 struct xenbus_transaction xbt;
1790 unsigned int i, max_page_order;
1791 unsigned int ring_page_order;
1792 struct blkfront_ring_info *rinfo;
1797 max_page_order = xenbus_read_unsigned(info->xbdev->otherend,
1798 "max-ring-page-order", 0);
1799 ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1800 info->nr_ring_pages = 1 << ring_page_order;
1802 err = negotiate_mq(info);
1804 goto destroy_blkring;
1806 for_each_rinfo(info, rinfo, i) {
1807 /* Create shared ring, alloc event channel. */
1808 err = setup_blkring(dev, rinfo);
1810 goto destroy_blkring;
1814 err = xenbus_transaction_start(&xbt);
1816 xenbus_dev_fatal(dev, err, "starting transaction");
1817 goto destroy_blkring;
1820 if (info->nr_ring_pages > 1) {
1821 err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1824 message = "writing ring-page-order";
1825 goto abort_transaction;
1829 /* We already got the number of queues/rings in _probe */
1830 if (info->nr_rings == 1) {
1831 err = write_per_ring_nodes(xbt, info->rinfo, dev->nodename);
1833 goto destroy_blkring;
1838 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1841 message = "writing multi-queue-num-queues";
1842 goto abort_transaction;
1845 pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1846 path = kmalloc(pathsize, GFP_KERNEL);
1849 message = "ENOMEM while writing ring references";
1850 goto abort_transaction;
1853 for_each_rinfo(info, rinfo, i) {
1854 memset(path, 0, pathsize);
1855 snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1856 err = write_per_ring_nodes(xbt, rinfo, path);
1859 goto destroy_blkring;
1864 err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1865 XEN_IO_PROTO_ABI_NATIVE);
1867 message = "writing protocol";
1868 goto abort_transaction;
1870 err = xenbus_printf(xbt, dev->nodename, "feature-persistent", "%u",
1871 info->feature_persistent);
1874 "writing persistent grants feature to xenbus");
1876 err = xenbus_transaction_end(xbt, 0);
1880 xenbus_dev_fatal(dev, err, "completing transaction");
1881 goto destroy_blkring;
1884 for_each_rinfo(info, rinfo, i) {
1887 for (j = 0; j < BLK_RING_SIZE(info); j++)
1888 rinfo->shadow[j].req.u.rw.id = j + 1;
1889 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1891 xenbus_switch_state(dev, XenbusStateInitialised);
1896 xenbus_transaction_end(xbt, 1);
1898 xenbus_dev_fatal(dev, err, "%s", message);
1900 blkif_free(info, 0);
1902 mutex_lock(&blkfront_mutex);
1904 mutex_unlock(&blkfront_mutex);
1906 dev_set_drvdata(&dev->dev, NULL);
1911 static int negotiate_mq(struct blkfront_info *info)
1913 unsigned int backend_max_queues;
1915 struct blkfront_ring_info *rinfo;
1917 BUG_ON(info->nr_rings);
1919 /* Check if backend supports multiple queues. */
1920 backend_max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1921 "multi-queue-max-queues", 1);
1922 info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1923 /* We need at least one ring. */
1924 if (!info->nr_rings)
1927 info->rinfo_size = struct_size(info->rinfo, shadow,
1928 BLK_RING_SIZE(info));
1929 info->rinfo = kvcalloc(info->nr_rings, info->rinfo_size, GFP_KERNEL);
1931 xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1936 for_each_rinfo(info, rinfo, i) {
1937 INIT_LIST_HEAD(&rinfo->indirect_pages);
1938 INIT_LIST_HEAD(&rinfo->grants);
1939 rinfo->dev_info = info;
1940 INIT_WORK(&rinfo->work, blkif_restart_queue);
1941 spin_lock_init(&rinfo->ring_lock);
1946 /* Enable the persistent grants feature. */
1947 static bool feature_persistent = true;
1948 module_param(feature_persistent, bool, 0644);
1949 MODULE_PARM_DESC(feature_persistent,
1950 "Enables the persistent grants feature");
1953 * Entry point to this code when a new device is created. Allocate the basic
1954 * structures and the ring buffer for communication with the backend, and
1955 * inform the backend of the appropriate details for those. Switch to
1956 * Initialised state.
1958 static int blkfront_probe(struct xenbus_device *dev,
1959 const struct xenbus_device_id *id)
1962 struct blkfront_info *info;
1964 /* FIXME: Use dynamic device id if this is not set. */
1965 err = xenbus_scanf(XBT_NIL, dev->nodename,
1966 "virtual-device", "%i", &vdevice);
1968 /* go looking in the extended area instead */
1969 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1972 xenbus_dev_fatal(dev, err, "reading virtual-device");
1977 if (xen_hvm_domain()) {
1980 /* no unplug has been done: do not hook devices != xen vbds */
1981 if (xen_has_pv_and_legacy_disk_devices()) {
1984 if (!VDEV_IS_EXTENDED(vdevice))
1985 major = BLKIF_MAJOR(vdevice);
1987 major = XENVBD_MAJOR;
1989 if (major != XENVBD_MAJOR) {
1991 "%s: HVM does not support vbd %d as xen block device\n",
1996 /* do not create a PV cdrom device if we are an HVM guest */
1997 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
2000 if (strncmp(type, "cdrom", 5) == 0) {
2006 info = kzalloc(sizeof(*info), GFP_KERNEL);
2008 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
2014 mutex_init(&info->mutex);
2015 info->vdevice = vdevice;
2016 info->connected = BLKIF_STATE_DISCONNECTED;
2018 info->feature_persistent = feature_persistent;
2020 /* Front end dir is a number, which is used as the id. */
2021 info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
2022 dev_set_drvdata(&dev->dev, info);
2024 mutex_lock(&blkfront_mutex);
2025 list_add(&info->info_list, &info_list);
2026 mutex_unlock(&blkfront_mutex);
2031 static int blkif_recover(struct blkfront_info *info)
2033 unsigned int r_index;
2034 struct request *req, *n;
2038 struct blkfront_ring_info *rinfo;
2040 blkfront_gather_backend_features(info);
2041 /* Reset limits changed by blk_mq_update_nr_hw_queues(). */
2042 blkif_set_queue_limits(info);
2043 segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
2044 blk_queue_max_segments(info->rq, segs / GRANTS_PER_PSEG);
2046 for_each_rinfo(info, rinfo, r_index) {
2047 rc = blkfront_setup_indirect(rinfo);
2051 xenbus_switch_state(info->xbdev, XenbusStateConnected);
2053 /* Now safe for us to use the shared ring */
2054 info->connected = BLKIF_STATE_CONNECTED;
2056 for_each_rinfo(info, rinfo, r_index) {
2057 /* Kick any other new requests queued since we resumed */
2058 kick_pending_request_queues(rinfo);
2061 list_for_each_entry_safe(req, n, &info->requests, queuelist) {
2062 /* Requeue pending requests (flush or discard) */
2063 list_del_init(&req->queuelist);
2064 BUG_ON(req->nr_phys_segments > segs);
2065 blk_mq_requeue_request(req, false);
2067 blk_mq_start_stopped_hw_queues(info->rq, true);
2068 blk_mq_kick_requeue_list(info->rq);
2070 while ((bio = bio_list_pop(&info->bio_list)) != NULL) {
2071 /* Traverse the list of pending bios and re-queue them */
2079 * We are reconnecting to the backend, due to a suspend/resume, or a backend
2080 * driver restart. We tear down our blkif structure and recreate it, but
2081 * leave the device-layer structures intact so that this is transparent to the
2082 * rest of the kernel.
2084 static int blkfront_resume(struct xenbus_device *dev)
2086 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2089 struct blkfront_ring_info *rinfo;
2091 dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2093 bio_list_init(&info->bio_list);
2094 INIT_LIST_HEAD(&info->requests);
2095 for_each_rinfo(info, rinfo, i) {
2096 struct bio_list merge_bio;
2097 struct blk_shadow *shadow = rinfo->shadow;
2099 for (j = 0; j < BLK_RING_SIZE(info); j++) {
2101 if (!shadow[j].request)
2105 * Get the bios in the request so we can re-queue them.
2107 if (req_op(shadow[j].request) == REQ_OP_FLUSH ||
2108 req_op(shadow[j].request) == REQ_OP_DISCARD ||
2109 req_op(shadow[j].request) == REQ_OP_SECURE_ERASE ||
2110 shadow[j].request->cmd_flags & REQ_FUA) {
2112 * Flush operations don't contain bios, so
2113 * we need to requeue the whole request
2115 * XXX: but this doesn't make any sense for a
2116 * write with the FUA flag set..
2118 list_add(&shadow[j].request->queuelist, &info->requests);
2121 merge_bio.head = shadow[j].request->bio;
2122 merge_bio.tail = shadow[j].request->biotail;
2123 bio_list_merge(&info->bio_list, &merge_bio);
2124 shadow[j].request->bio = NULL;
2125 blk_mq_end_request(shadow[j].request, BLK_STS_OK);
2129 blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2131 err = talk_to_blkback(dev, info);
2133 blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings);
2136 * We have to wait for the backend to switch to
2137 * connected state, since we want to read which
2138 * features it supports.
2144 static void blkfront_closing(struct blkfront_info *info)
2146 struct xenbus_device *xbdev = info->xbdev;
2147 struct block_device *bdev = NULL;
2149 mutex_lock(&info->mutex);
2151 if (xbdev->state == XenbusStateClosing) {
2152 mutex_unlock(&info->mutex);
2157 bdev = bdgrab(info->gd->part0);
2159 mutex_unlock(&info->mutex);
2162 xenbus_frontend_closed(xbdev);
2166 mutex_lock(&bdev->bd_mutex);
2168 if (bdev->bd_openers) {
2169 xenbus_dev_error(xbdev, -EBUSY,
2170 "Device in use; refusing to close");
2171 xenbus_switch_state(xbdev, XenbusStateClosing);
2173 xlvbd_release_gendisk(info);
2174 xenbus_frontend_closed(xbdev);
2177 mutex_unlock(&bdev->bd_mutex);
2181 static void blkfront_setup_discard(struct blkfront_info *info)
2183 info->feature_discard = 1;
2184 info->discard_granularity = xenbus_read_unsigned(info->xbdev->otherend,
2185 "discard-granularity",
2187 info->discard_alignment = xenbus_read_unsigned(info->xbdev->otherend,
2188 "discard-alignment", 0);
2189 info->feature_secdiscard =
2190 !!xenbus_read_unsigned(info->xbdev->otherend, "discard-secure",
2194 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2196 unsigned int psegs, grants, memflags;
2198 struct blkfront_info *info = rinfo->dev_info;
2200 memflags = memalloc_noio_save();
2202 if (info->max_indirect_segments == 0) {
2204 grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2207 * When an extra req is required, the maximum
2208 * grants supported is related to the size of the
2209 * Linux block segment.
2211 grants = GRANTS_PER_PSEG;
2215 grants = info->max_indirect_segments;
2216 psegs = DIV_ROUND_UP(grants, GRANTS_PER_PSEG);
2218 err = fill_grant_buffer(rinfo,
2219 (grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2223 if (!info->feature_persistent && info->max_indirect_segments) {
2225 * We are using indirect descriptors but not persistent
2226 * grants, we need to allocate a set of pages that can be
2227 * used for mapping indirect grefs
2229 int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2231 BUG_ON(!list_empty(&rinfo->indirect_pages));
2232 for (i = 0; i < num; i++) {
2233 struct page *indirect_page = alloc_page(GFP_KERNEL);
2236 list_add(&indirect_page->lru, &rinfo->indirect_pages);
2240 for (i = 0; i < BLK_RING_SIZE(info); i++) {
2241 rinfo->shadow[i].grants_used =
2243 sizeof(rinfo->shadow[i].grants_used[0]),
2245 rinfo->shadow[i].sg = kvcalloc(psegs,
2246 sizeof(rinfo->shadow[i].sg[0]),
2248 if (info->max_indirect_segments)
2249 rinfo->shadow[i].indirect_grants =
2250 kvcalloc(INDIRECT_GREFS(grants),
2251 sizeof(rinfo->shadow[i].indirect_grants[0]),
2253 if ((rinfo->shadow[i].grants_used == NULL) ||
2254 (rinfo->shadow[i].sg == NULL) ||
2255 (info->max_indirect_segments &&
2256 (rinfo->shadow[i].indirect_grants == NULL)))
2258 sg_init_table(rinfo->shadow[i].sg, psegs);
2261 memalloc_noio_restore(memflags);
2266 for (i = 0; i < BLK_RING_SIZE(info); i++) {
2267 kvfree(rinfo->shadow[i].grants_used);
2268 rinfo->shadow[i].grants_used = NULL;
2269 kvfree(rinfo->shadow[i].sg);
2270 rinfo->shadow[i].sg = NULL;
2271 kvfree(rinfo->shadow[i].indirect_grants);
2272 rinfo->shadow[i].indirect_grants = NULL;
2274 if (!list_empty(&rinfo->indirect_pages)) {
2275 struct page *indirect_page, *n;
2276 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2277 list_del(&indirect_page->lru);
2278 __free_page(indirect_page);
2282 memalloc_noio_restore(memflags);
2288 * Gather all backend feature-*
2290 static void blkfront_gather_backend_features(struct blkfront_info *info)
2292 unsigned int indirect_segments;
2294 info->feature_flush = 0;
2295 info->feature_fua = 0;
2298 * If there's no "feature-barrier" defined, then it means
2299 * we're dealing with a very old backend which writes
2300 * synchronously; nothing to do.
2302 * If there are barriers, then we use flush.
2304 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-barrier", 0)) {
2305 info->feature_flush = 1;
2306 info->feature_fua = 1;
2310 * And if there is "feature-flush-cache" use that above
2313 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-flush-cache",
2315 info->feature_flush = 1;
2316 info->feature_fua = 0;
2319 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-discard", 0))
2320 blkfront_setup_discard(info);
2322 if (info->feature_persistent)
2323 info->feature_persistent =
2324 !!xenbus_read_unsigned(info->xbdev->otherend,
2325 "feature-persistent", 0);
2327 indirect_segments = xenbus_read_unsigned(info->xbdev->otherend,
2328 "feature-max-indirect-segments", 0);
2329 if (indirect_segments > xen_blkif_max_segments)
2330 indirect_segments = xen_blkif_max_segments;
2331 if (indirect_segments <= BLKIF_MAX_SEGMENTS_PER_REQUEST)
2332 indirect_segments = 0;
2333 info->max_indirect_segments = indirect_segments;
2335 if (info->feature_persistent) {
2336 mutex_lock(&blkfront_mutex);
2337 schedule_delayed_work(&blkfront_work, HZ * 10);
2338 mutex_unlock(&blkfront_mutex);
2343 * Invoked when the backend is finally 'ready' (and has told produced
2344 * the details about the physical device - #sectors, size, etc).
2346 static void blkfront_connect(struct blkfront_info *info)
2348 unsigned long long sectors;
2349 unsigned long sector_size;
2350 unsigned int physical_sector_size;
2353 struct blkfront_ring_info *rinfo;
2355 switch (info->connected) {
2356 case BLKIF_STATE_CONNECTED:
2358 * Potentially, the back-end may be signalling
2359 * a capacity change; update the capacity.
2361 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2362 "sectors", "%Lu", §ors);
2363 if (XENBUS_EXIST_ERR(err))
2365 printk(KERN_INFO "Setting capacity to %Lu\n",
2367 set_capacity_and_notify(info->gd, sectors);
2370 case BLKIF_STATE_SUSPENDED:
2372 * If we are recovering from suspension, we need to wait
2373 * for the backend to announce it's features before
2374 * reconnecting, at least we need to know if the backend
2375 * supports indirect descriptors, and how many.
2377 blkif_recover(info);
2384 dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2385 __func__, info->xbdev->otherend);
2387 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2388 "sectors", "%llu", §ors,
2389 "info", "%u", &binfo,
2390 "sector-size", "%lu", §or_size,
2393 xenbus_dev_fatal(info->xbdev, err,
2394 "reading backend fields at %s",
2395 info->xbdev->otherend);
2400 * physcial-sector-size is a newer field, so old backends may not
2401 * provide this. Assume physical sector size to be the same as
2402 * sector_size in that case.
2404 physical_sector_size = xenbus_read_unsigned(info->xbdev->otherend,
2405 "physical-sector-size",
2407 blkfront_gather_backend_features(info);
2408 for_each_rinfo(info, rinfo, i) {
2409 err = blkfront_setup_indirect(rinfo);
2411 xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2412 info->xbdev->otherend);
2413 blkif_free(info, 0);
2418 err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
2419 physical_sector_size);
2421 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2422 info->xbdev->otherend);
2426 xenbus_switch_state(info->xbdev, XenbusStateConnected);
2428 /* Kick pending requests. */
2429 info->connected = BLKIF_STATE_CONNECTED;
2430 for_each_rinfo(info, rinfo, i)
2431 kick_pending_request_queues(rinfo);
2433 device_add_disk(&info->xbdev->dev, info->gd, NULL);
2439 blkif_free(info, 0);
2444 * Callback received when the backend's state changes.
2446 static void blkback_changed(struct xenbus_device *dev,
2447 enum xenbus_state backend_state)
2449 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2451 dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2453 switch (backend_state) {
2454 case XenbusStateInitWait:
2455 if (dev->state != XenbusStateInitialising)
2457 if (talk_to_blkback(dev, info))
2460 case XenbusStateInitialising:
2461 case XenbusStateInitialised:
2462 case XenbusStateReconfiguring:
2463 case XenbusStateReconfigured:
2464 case XenbusStateUnknown:
2467 case XenbusStateConnected:
2469 * talk_to_blkback sets state to XenbusStateInitialised
2470 * and blkfront_connect sets it to XenbusStateConnected
2471 * (if connection went OK).
2473 * If the backend (or toolstack) decides to poke at backend
2474 * state (and re-trigger the watch by setting the state repeatedly
2475 * to XenbusStateConnected (4)) we need to deal with this.
2476 * This is allowed as this is used to communicate to the guest
2477 * that the size of disk has changed!
2479 if ((dev->state != XenbusStateInitialised) &&
2480 (dev->state != XenbusStateConnected)) {
2481 if (talk_to_blkback(dev, info))
2485 blkfront_connect(info);
2488 case XenbusStateClosed:
2489 if (dev->state == XenbusStateClosed)
2492 case XenbusStateClosing:
2494 blkfront_closing(info);
2499 static int blkfront_remove(struct xenbus_device *xbdev)
2501 struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2502 struct block_device *bdev = NULL;
2503 struct gendisk *disk;
2505 dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2510 blkif_free(info, 0);
2512 mutex_lock(&info->mutex);
2516 bdev = bdgrab(disk->part0);
2519 mutex_unlock(&info->mutex);
2522 mutex_lock(&blkfront_mutex);
2524 mutex_unlock(&blkfront_mutex);
2529 * The xbdev was removed before we reached the Closed
2530 * state. See if it's safe to remove the disk. If the bdev
2531 * isn't closed yet, we let release take care of it.
2534 mutex_lock(&bdev->bd_mutex);
2535 info = disk->private_data;
2537 dev_warn(disk_to_dev(disk),
2538 "%s was hot-unplugged, %d stale handles\n",
2539 xbdev->nodename, bdev->bd_openers);
2541 if (info && !bdev->bd_openers) {
2542 xlvbd_release_gendisk(info);
2543 disk->private_data = NULL;
2544 mutex_lock(&blkfront_mutex);
2546 mutex_unlock(&blkfront_mutex);
2549 mutex_unlock(&bdev->bd_mutex);
2555 static int blkfront_is_ready(struct xenbus_device *dev)
2557 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2559 return info->is_ready && info->xbdev;
2562 static int blkif_open(struct block_device *bdev, fmode_t mode)
2564 struct gendisk *disk = bdev->bd_disk;
2565 struct blkfront_info *info;
2568 mutex_lock(&blkfront_mutex);
2570 info = disk->private_data;
2577 mutex_lock(&info->mutex);
2580 /* xbdev is closed */
2583 mutex_unlock(&info->mutex);
2586 mutex_unlock(&blkfront_mutex);
2590 static void blkif_release(struct gendisk *disk, fmode_t mode)
2592 struct blkfront_info *info = disk->private_data;
2593 struct xenbus_device *xbdev;
2595 mutex_lock(&blkfront_mutex);
2596 if (disk->part0->bd_openers)
2600 * Check if we have been instructed to close. We will have
2601 * deferred this request, because the bdev was still open.
2604 mutex_lock(&info->mutex);
2605 xbdev = info->xbdev;
2607 if (xbdev && xbdev->state == XenbusStateClosing) {
2608 /* pending switch to state closed */
2609 dev_info(disk_to_dev(disk), "releasing disk\n");
2610 xlvbd_release_gendisk(info);
2611 xenbus_frontend_closed(info->xbdev);
2614 mutex_unlock(&info->mutex);
2617 /* sudden device removal */
2618 dev_info(disk_to_dev(disk), "releasing disk\n");
2619 xlvbd_release_gendisk(info);
2620 disk->private_data = NULL;
2625 mutex_unlock(&blkfront_mutex);
2628 static const struct block_device_operations xlvbd_block_fops =
2630 .owner = THIS_MODULE,
2632 .release = blkif_release,
2633 .getgeo = blkif_getgeo,
2634 .ioctl = blkif_ioctl,
2635 .compat_ioctl = blkdev_compat_ptr_ioctl,
2639 static const struct xenbus_device_id blkfront_ids[] = {
2644 static struct xenbus_driver blkfront_driver = {
2645 .ids = blkfront_ids,
2646 .probe = blkfront_probe,
2647 .remove = blkfront_remove,
2648 .resume = blkfront_resume,
2649 .otherend_changed = blkback_changed,
2650 .is_ready = blkfront_is_ready,
2653 static void purge_persistent_grants(struct blkfront_info *info)
2656 unsigned long flags;
2657 struct blkfront_ring_info *rinfo;
2659 for_each_rinfo(info, rinfo, i) {
2660 struct grant *gnt_list_entry, *tmp;
2662 spin_lock_irqsave(&rinfo->ring_lock, flags);
2664 if (rinfo->persistent_gnts_c == 0) {
2665 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2669 list_for_each_entry_safe(gnt_list_entry, tmp, &rinfo->grants,
2671 if (gnt_list_entry->gref == GRANT_INVALID_REF ||
2672 gnttab_query_foreign_access(gnt_list_entry->gref))
2675 list_del(&gnt_list_entry->node);
2676 gnttab_end_foreign_access(gnt_list_entry->gref, 0, 0UL);
2677 rinfo->persistent_gnts_c--;
2678 gnt_list_entry->gref = GRANT_INVALID_REF;
2679 list_add_tail(&gnt_list_entry->node, &rinfo->grants);
2682 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2686 static void blkfront_delay_work(struct work_struct *work)
2688 struct blkfront_info *info;
2689 bool need_schedule_work = false;
2691 mutex_lock(&blkfront_mutex);
2693 list_for_each_entry(info, &info_list, info_list) {
2694 if (info->feature_persistent) {
2695 need_schedule_work = true;
2696 mutex_lock(&info->mutex);
2697 purge_persistent_grants(info);
2698 mutex_unlock(&info->mutex);
2702 if (need_schedule_work)
2703 schedule_delayed_work(&blkfront_work, HZ * 10);
2705 mutex_unlock(&blkfront_mutex);
2708 static int __init xlblk_init(void)
2711 int nr_cpus = num_online_cpus();
2716 if (!xen_has_pv_disk_devices())
2719 if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2720 pr_warn("xen_blk: can't get major %d with name %s\n",
2721 XENVBD_MAJOR, DEV_NAME);
2725 if (xen_blkif_max_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST)
2726 xen_blkif_max_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2728 if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2729 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2730 xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2731 xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2734 if (xen_blkif_max_queues > nr_cpus) {
2735 pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2736 xen_blkif_max_queues, nr_cpus);
2737 xen_blkif_max_queues = nr_cpus;
2740 INIT_DELAYED_WORK(&blkfront_work, blkfront_delay_work);
2742 ret = xenbus_register_frontend(&blkfront_driver);
2744 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2750 module_init(xlblk_init);
2753 static void __exit xlblk_exit(void)
2755 cancel_delayed_work_sync(&blkfront_work);
2757 xenbus_unregister_driver(&blkfront_driver);
2758 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2761 module_exit(xlblk_exit);
2763 MODULE_DESCRIPTION("Xen virtual block device frontend");
2764 MODULE_LICENSE("GPL");
2765 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2766 MODULE_ALIAS("xen:vbd");
2767 MODULE_ALIAS("xenblk");