4 * XenLinux virtual block device driver.
6 * Copyright (c) 2003-2004, Keir Fraser & Steve Hand
7 * Modifications by Mark A. Williamson are (c) Intel Research Cambridge
8 * Copyright (c) 2004, Christian Limpach
9 * Copyright (c) 2004, Andrew Warfield
10 * Copyright (c) 2005, Christopher Clark
11 * Copyright (c) 2005, XenSource Ltd
13 * This program is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU General Public License version 2
15 * as published by the Free Software Foundation; or, when distributed
16 * separately from the Linux kernel or incorporated into other
17 * software packages, subject to the following license:
19 * Permission is hereby granted, free of charge, to any person obtaining a copy
20 * of this source file (the "Software"), to deal in the Software without
21 * restriction, including without limitation the rights to use, copy, modify,
22 * merge, publish, distribute, sublicense, and/or sell copies of the Software,
23 * and to permit persons to whom the Software is furnished to do so, subject to
24 * the following conditions:
26 * The above copyright notice and this permission notice shall be included in
27 * all copies or substantial portions of the Software.
29 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
30 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
31 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
32 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
33 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
34 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
38 #include <linux/interrupt.h>
39 #include <linux/blkdev.h>
40 #include <linux/blk-mq.h>
41 #include <linux/hdreg.h>
42 #include <linux/cdrom.h>
43 #include <linux/module.h>
44 #include <linux/slab.h>
45 #include <linux/mutex.h>
46 #include <linux/scatterlist.h>
47 #include <linux/bitmap.h>
48 #include <linux/list.h>
49 #include <linux/workqueue.h>
52 #include <xen/xenbus.h>
53 #include <xen/grant_table.h>
54 #include <xen/events.h>
56 #include <xen/platform_pci.h>
58 #include <xen/interface/grant_table.h>
59 #include <xen/interface/io/blkif.h>
60 #include <xen/interface/io/protocols.h>
62 #include <asm/xen/hypervisor.h>
65 * The minimal size of segment supported by the block framework is PAGE_SIZE.
66 * When Linux is using a different page size than Xen, it may not be possible
67 * to put all the data in a single segment.
68 * This can happen when the backend doesn't support indirect descriptor and
69 * therefore the maximum amount of data that a request can carry is
70 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE = 44KB
72 * Note that we only support one extra request. So the Linux page size
73 * should be <= ( 2 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) =
76 #define HAS_EXTRA_REQ (BLKIF_MAX_SEGMENTS_PER_REQUEST < XEN_PFN_PER_PAGE)
79 BLKIF_STATE_DISCONNECTED,
80 BLKIF_STATE_CONNECTED,
81 BLKIF_STATE_SUSPENDED,
87 struct list_head node;
98 struct blkif_request req;
99 struct request *request;
100 struct grant **grants_used;
101 struct grant **indirect_grants;
102 struct scatterlist *sg;
104 enum blk_req_status status;
106 #define NO_ASSOCIATED_ID ~0UL
108 * Id of the sibling if we ever need 2 requests when handling a
111 unsigned long associated_id;
118 static inline struct blkif_req *blkif_req(struct request *rq)
120 return blk_mq_rq_to_pdu(rq);
123 static DEFINE_MUTEX(blkfront_mutex);
124 static const struct block_device_operations xlvbd_block_fops;
125 static struct delayed_work blkfront_work;
126 static LIST_HEAD(info_list);
129 * Maximum number of segments in indirect requests, the actual value used by
130 * the frontend driver is the minimum of this value and the value provided
131 * by the backend driver.
134 static unsigned int xen_blkif_max_segments = 32;
135 module_param_named(max_indirect_segments, xen_blkif_max_segments, uint, 0444);
136 MODULE_PARM_DESC(max_indirect_segments,
137 "Maximum amount of segments in indirect requests (default is 32)");
139 static unsigned int xen_blkif_max_queues = 4;
140 module_param_named(max_queues, xen_blkif_max_queues, uint, 0444);
141 MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
144 * Maximum order of pages to be used for the shared ring between front and
145 * backend, 4KB page granularity is used.
147 static unsigned int xen_blkif_max_ring_order;
148 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, 0444);
149 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
151 #define BLK_RING_SIZE(info) \
152 __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
155 * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
156 * characters are enough. Define to 20 to keep consistent with backend.
158 #define RINGREF_NAME_LEN (20)
160 * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
162 #define QUEUE_NAME_LEN (17)
166 * Every blkfront device can associate with one or more blkfront_ring_info,
167 * depending on how many hardware queues/rings to be used.
169 struct blkfront_ring_info {
170 /* Lock to protect data in every ring buffer. */
171 spinlock_t ring_lock;
172 struct blkif_front_ring ring;
173 unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
174 unsigned int evtchn, irq;
175 struct work_struct work;
176 struct gnttab_free_callback callback;
177 struct list_head indirect_pages;
178 struct list_head grants;
179 unsigned int persistent_gnts_c;
180 unsigned long shadow_free;
181 struct blkfront_info *dev_info;
182 struct blk_shadow shadow[];
186 * We have one of these per vbd, whether ide, scsi or 'other'. They
187 * hang in private_data off the gendisk structure. We may end up
188 * putting all kinds of interesting stuff here :-)
193 struct xenbus_device *xbdev;
196 unsigned int physical_sector_size;
199 enum blkif_state connected;
200 /* Number of pages per ring buffer. */
201 unsigned int nr_ring_pages;
202 struct request_queue *rq;
203 unsigned int feature_flush:1;
204 unsigned int feature_fua:1;
205 unsigned int feature_discard:1;
206 unsigned int feature_secdiscard:1;
207 unsigned int feature_persistent:1;
208 unsigned int discard_granularity;
209 unsigned int discard_alignment;
210 /* Number of 4KB segments handled */
211 unsigned int max_indirect_segments;
213 struct blk_mq_tag_set tag_set;
214 struct blkfront_ring_info *rinfo;
215 unsigned int nr_rings;
216 unsigned int rinfo_size;
217 /* Save uncomplete reqs and bios for migration. */
218 struct list_head requests;
219 struct bio_list bio_list;
220 struct list_head info_list;
223 static unsigned int nr_minors;
224 static unsigned long *minors;
225 static DEFINE_SPINLOCK(minor_lock);
227 #define GRANT_INVALID_REF 0
229 #define PARTS_PER_DISK 16
230 #define PARTS_PER_EXT_DISK 256
232 #define BLKIF_MAJOR(dev) ((dev)>>8)
233 #define BLKIF_MINOR(dev) ((dev) & 0xff)
236 #define EXTENDED (1<<EXT_SHIFT)
237 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
238 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
239 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
240 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
241 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
242 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
244 #define DEV_NAME "xvd" /* name in /dev */
247 * Grants are always the same size as a Xen page (i.e 4KB).
248 * A physical segment is always the same size as a Linux page.
249 * Number of grants per physical segment
251 #define GRANTS_PER_PSEG (PAGE_SIZE / XEN_PAGE_SIZE)
253 #define GRANTS_PER_INDIRECT_FRAME \
254 (XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
256 #define INDIRECT_GREFS(_grants) \
257 DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
259 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
260 static void blkfront_gather_backend_features(struct blkfront_info *info);
261 static int negotiate_mq(struct blkfront_info *info);
263 #define for_each_rinfo(info, ptr, idx) \
264 for ((ptr) = (info)->rinfo, (idx) = 0; \
265 (idx) < (info)->nr_rings; \
266 (idx)++, (ptr) = (void *)(ptr) + (info)->rinfo_size)
268 static inline struct blkfront_ring_info *
269 get_rinfo(const struct blkfront_info *info, unsigned int i)
271 BUG_ON(i >= info->nr_rings);
272 return (void *)info->rinfo + i * info->rinfo_size;
275 static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
277 unsigned long free = rinfo->shadow_free;
279 BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
280 rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
281 rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
285 static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
288 if (rinfo->shadow[id].req.u.rw.id != id)
290 if (rinfo->shadow[id].request == NULL)
292 rinfo->shadow[id].req.u.rw.id = rinfo->shadow_free;
293 rinfo->shadow[id].request = NULL;
294 rinfo->shadow_free = id;
298 static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
300 struct blkfront_info *info = rinfo->dev_info;
301 struct page *granted_page;
302 struct grant *gnt_list_entry, *n;
306 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
310 if (info->feature_persistent) {
311 granted_page = alloc_page(GFP_NOIO);
313 kfree(gnt_list_entry);
316 gnt_list_entry->page = granted_page;
319 gnt_list_entry->gref = GRANT_INVALID_REF;
320 list_add(&gnt_list_entry->node, &rinfo->grants);
327 list_for_each_entry_safe(gnt_list_entry, n,
328 &rinfo->grants, node) {
329 list_del(&gnt_list_entry->node);
330 if (info->feature_persistent)
331 __free_page(gnt_list_entry->page);
332 kfree(gnt_list_entry);
339 static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
341 struct grant *gnt_list_entry;
343 BUG_ON(list_empty(&rinfo->grants));
344 gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
346 list_del(&gnt_list_entry->node);
348 if (gnt_list_entry->gref != GRANT_INVALID_REF)
349 rinfo->persistent_gnts_c--;
351 return gnt_list_entry;
354 static inline void grant_foreign_access(const struct grant *gnt_list_entry,
355 const struct blkfront_info *info)
357 gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
358 info->xbdev->otherend_id,
359 gnt_list_entry->page,
363 static struct grant *get_grant(grant_ref_t *gref_head,
365 struct blkfront_ring_info *rinfo)
367 struct grant *gnt_list_entry = get_free_grant(rinfo);
368 struct blkfront_info *info = rinfo->dev_info;
370 if (gnt_list_entry->gref != GRANT_INVALID_REF)
371 return gnt_list_entry;
373 /* Assign a gref to this page */
374 gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
375 BUG_ON(gnt_list_entry->gref == -ENOSPC);
376 if (info->feature_persistent)
377 grant_foreign_access(gnt_list_entry, info);
379 /* Grant access to the GFN passed by the caller */
380 gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
381 info->xbdev->otherend_id,
385 return gnt_list_entry;
388 static struct grant *get_indirect_grant(grant_ref_t *gref_head,
389 struct blkfront_ring_info *rinfo)
391 struct grant *gnt_list_entry = get_free_grant(rinfo);
392 struct blkfront_info *info = rinfo->dev_info;
394 if (gnt_list_entry->gref != GRANT_INVALID_REF)
395 return gnt_list_entry;
397 /* Assign a gref to this page */
398 gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
399 BUG_ON(gnt_list_entry->gref == -ENOSPC);
400 if (!info->feature_persistent) {
401 struct page *indirect_page;
403 /* Fetch a pre-allocated page to use for indirect grefs */
404 BUG_ON(list_empty(&rinfo->indirect_pages));
405 indirect_page = list_first_entry(&rinfo->indirect_pages,
407 list_del(&indirect_page->lru);
408 gnt_list_entry->page = indirect_page;
410 grant_foreign_access(gnt_list_entry, info);
412 return gnt_list_entry;
415 static const char *op_name(int op)
417 static const char *const names[] = {
418 [BLKIF_OP_READ] = "read",
419 [BLKIF_OP_WRITE] = "write",
420 [BLKIF_OP_WRITE_BARRIER] = "barrier",
421 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
422 [BLKIF_OP_DISCARD] = "discard" };
424 if (op < 0 || op >= ARRAY_SIZE(names))
432 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
434 unsigned int end = minor + nr;
437 if (end > nr_minors) {
438 unsigned long *bitmap, *old;
440 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
445 spin_lock(&minor_lock);
446 if (end > nr_minors) {
448 memcpy(bitmap, minors,
449 BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
451 nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
454 spin_unlock(&minor_lock);
458 spin_lock(&minor_lock);
459 if (find_next_bit(minors, end, minor) >= end) {
460 bitmap_set(minors, minor, nr);
464 spin_unlock(&minor_lock);
469 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
471 unsigned int end = minor + nr;
473 BUG_ON(end > nr_minors);
474 spin_lock(&minor_lock);
475 bitmap_clear(minors, minor, nr);
476 spin_unlock(&minor_lock);
479 static void blkif_restart_queue_callback(void *arg)
481 struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
482 schedule_work(&rinfo->work);
485 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
487 /* We don't have real geometry info, but let's at least return
488 values consistent with the size of the device */
489 sector_t nsect = get_capacity(bd->bd_disk);
490 sector_t cylinders = nsect;
494 sector_div(cylinders, hg->heads * hg->sectors);
495 hg->cylinders = cylinders;
496 if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
497 hg->cylinders = 0xffff;
501 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
502 unsigned command, unsigned long argument)
504 struct blkfront_info *info = bdev->bd_disk->private_data;
507 dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
508 command, (long)argument);
511 case CDROMMULTISESSION:
512 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
513 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
514 if (put_user(0, (char __user *)(argument + i)))
518 case CDROM_GET_CAPABILITY: {
519 struct gendisk *gd = info->gd;
520 if (gd->flags & GENHD_FL_CD)
526 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
528 return -EINVAL; /* same return as native Linux */
534 static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
536 struct blkif_request **ring_req)
540 *ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
541 rinfo->ring.req_prod_pvt++;
543 id = get_id_from_freelist(rinfo);
544 rinfo->shadow[id].request = req;
545 rinfo->shadow[id].status = REQ_WAITING;
546 rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
548 (*ring_req)->u.rw.id = id;
553 static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
555 struct blkfront_info *info = rinfo->dev_info;
556 struct blkif_request *ring_req;
559 /* Fill out a communications ring structure. */
560 id = blkif_ring_get_request(rinfo, req, &ring_req);
562 ring_req->operation = BLKIF_OP_DISCARD;
563 ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
564 ring_req->u.discard.id = id;
565 ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
566 if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard)
567 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
569 ring_req->u.discard.flag = 0;
571 /* Keep a private copy so we can reissue requests when recovering. */
572 rinfo->shadow[id].req = *ring_req;
577 struct setup_rw_req {
578 unsigned int grant_idx;
579 struct blkif_request_segment *segments;
580 struct blkfront_ring_info *rinfo;
581 struct blkif_request *ring_req;
582 grant_ref_t gref_head;
584 /* Only used when persistent grant is used and it's a read request */
586 unsigned int bvec_off;
589 bool require_extra_req;
590 struct blkif_request *extra_ring_req;
593 static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
594 unsigned int len, void *data)
596 struct setup_rw_req *setup = data;
598 struct grant *gnt_list_entry;
599 unsigned int fsect, lsect;
600 /* Convenient aliases */
601 unsigned int grant_idx = setup->grant_idx;
602 struct blkif_request *ring_req = setup->ring_req;
603 struct blkfront_ring_info *rinfo = setup->rinfo;
605 * We always use the shadow of the first request to store the list
606 * of grant associated to the block I/O request. This made the
607 * completion more easy to handle even if the block I/O request is
610 struct blk_shadow *shadow = &rinfo->shadow[setup->id];
612 if (unlikely(setup->require_extra_req &&
613 grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
615 * We are using the second request, setup grant_idx
616 * to be the index of the segment array.
618 grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
619 ring_req = setup->extra_ring_req;
622 if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
623 (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
625 kunmap_atomic(setup->segments);
627 n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
628 gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
629 shadow->indirect_grants[n] = gnt_list_entry;
630 setup->segments = kmap_atomic(gnt_list_entry->page);
631 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
634 gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
635 ref = gnt_list_entry->gref;
637 * All the grants are stored in the shadow of the first
638 * request. Therefore we have to use the global index.
640 shadow->grants_used[setup->grant_idx] = gnt_list_entry;
642 if (setup->need_copy) {
645 shared_data = kmap_atomic(gnt_list_entry->page);
647 * this does not wipe data stored outside the
648 * range sg->offset..sg->offset+sg->length.
649 * Therefore, blkback *could* see data from
650 * previous requests. This is OK as long as
651 * persistent grants are shared with just one
652 * domain. It may need refactoring if this
655 memcpy(shared_data + offset,
656 setup->bvec_data + setup->bvec_off,
659 kunmap_atomic(shared_data);
660 setup->bvec_off += len;
664 lsect = fsect + (len >> 9) - 1;
665 if (ring_req->operation != BLKIF_OP_INDIRECT) {
666 ring_req->u.rw.seg[grant_idx] =
667 (struct blkif_request_segment) {
670 .last_sect = lsect };
672 setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
673 (struct blkif_request_segment) {
676 .last_sect = lsect };
679 (setup->grant_idx)++;
682 static void blkif_setup_extra_req(struct blkif_request *first,
683 struct blkif_request *second)
685 uint16_t nr_segments = first->u.rw.nr_segments;
688 * The second request is only present when the first request uses
689 * all its segments. It's always the continuity of the first one.
691 first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
693 second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
694 second->u.rw.sector_number = first->u.rw.sector_number +
695 (BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
697 second->u.rw.handle = first->u.rw.handle;
698 second->operation = first->operation;
701 static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
703 struct blkfront_info *info = rinfo->dev_info;
704 struct blkif_request *ring_req, *extra_ring_req = NULL;
705 unsigned long id, extra_id = NO_ASSOCIATED_ID;
706 bool require_extra_req = false;
708 struct setup_rw_req setup = {
712 .need_copy = rq_data_dir(req) && info->feature_persistent,
716 * Used to store if we are able to queue the request by just using
717 * existing persistent grants, or if we have to get new grants,
718 * as there are not sufficiently many free.
720 bool new_persistent_gnts = false;
721 struct scatterlist *sg;
722 int num_sg, max_grefs, num_grant;
724 max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
725 if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
727 * If we are using indirect segments we need to account
728 * for the indirect grefs used in the request.
730 max_grefs += INDIRECT_GREFS(max_grefs);
732 /* Check if we have enough persistent grants to allocate a requests */
733 if (rinfo->persistent_gnts_c < max_grefs) {
734 new_persistent_gnts = true;
736 if (gnttab_alloc_grant_references(
737 max_grefs - rinfo->persistent_gnts_c,
738 &setup.gref_head) < 0) {
739 gnttab_request_free_callback(
741 blkif_restart_queue_callback,
743 max_grefs - rinfo->persistent_gnts_c);
748 /* Fill out a communications ring structure. */
749 id = blkif_ring_get_request(rinfo, req, &ring_req);
751 num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
753 /* Calculate the number of grant used */
754 for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
755 num_grant += gnttab_count_grant(sg->offset, sg->length);
757 require_extra_req = info->max_indirect_segments == 0 &&
758 num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
759 BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
761 rinfo->shadow[id].num_sg = num_sg;
762 if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
763 likely(!require_extra_req)) {
765 * The indirect operation can only be a BLKIF_OP_READ or
768 BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA);
769 ring_req->operation = BLKIF_OP_INDIRECT;
770 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
771 BLKIF_OP_WRITE : BLKIF_OP_READ;
772 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
773 ring_req->u.indirect.handle = info->handle;
774 ring_req->u.indirect.nr_segments = num_grant;
776 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
777 ring_req->u.rw.handle = info->handle;
778 ring_req->operation = rq_data_dir(req) ?
779 BLKIF_OP_WRITE : BLKIF_OP_READ;
780 if (req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA) {
782 * Ideally we can do an unordered flush-to-disk.
783 * In case the backend onlysupports barriers, use that.
784 * A barrier request a superset of FUA, so we can
785 * implement it the same way. (It's also a FLUSH+FUA,
786 * since it is guaranteed ordered WRT previous writes.)
788 if (info->feature_flush && info->feature_fua)
789 ring_req->operation =
790 BLKIF_OP_WRITE_BARRIER;
791 else if (info->feature_flush)
792 ring_req->operation =
793 BLKIF_OP_FLUSH_DISKCACHE;
795 ring_req->operation = 0;
797 ring_req->u.rw.nr_segments = num_grant;
798 if (unlikely(require_extra_req)) {
799 extra_id = blkif_ring_get_request(rinfo, req,
802 * Only the first request contains the scatter-gather
805 rinfo->shadow[extra_id].num_sg = 0;
807 blkif_setup_extra_req(ring_req, extra_ring_req);
809 /* Link the 2 requests together */
810 rinfo->shadow[extra_id].associated_id = id;
811 rinfo->shadow[id].associated_id = extra_id;
815 setup.ring_req = ring_req;
818 setup.require_extra_req = require_extra_req;
819 if (unlikely(require_extra_req))
820 setup.extra_ring_req = extra_ring_req;
822 for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
823 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
825 if (setup.need_copy) {
826 setup.bvec_off = sg->offset;
827 setup.bvec_data = kmap_atomic(sg_page(sg));
830 gnttab_foreach_grant_in_range(sg_page(sg),
833 blkif_setup_rw_req_grant,
837 kunmap_atomic(setup.bvec_data);
840 kunmap_atomic(setup.segments);
842 /* Keep a private copy so we can reissue requests when recovering. */
843 rinfo->shadow[id].req = *ring_req;
844 if (unlikely(require_extra_req))
845 rinfo->shadow[extra_id].req = *extra_ring_req;
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 rq->limits.discard_alignment = info->discard_alignment;
949 if (info->feature_secdiscard)
950 blk_queue_flag_set(QUEUE_FLAG_SECERASE, rq);
953 /* Hard sector size and max sectors impersonate the equiv. hardware. */
954 blk_queue_logical_block_size(rq, info->sector_size);
955 blk_queue_physical_block_size(rq, info->physical_sector_size);
956 blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
958 /* Each segment in a request is up to an aligned page in size. */
959 blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
960 blk_queue_max_segment_size(rq, PAGE_SIZE);
962 /* Ensure a merged request will fit in a single I/O ring slot. */
963 blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
965 /* Make sure buffer addresses are sector-aligned. */
966 blk_queue_dma_alignment(rq, 511);
969 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
970 unsigned int physical_sector_size)
972 struct request_queue *rq;
973 struct blkfront_info *info = gd->private_data;
975 memset(&info->tag_set, 0, sizeof(info->tag_set));
976 info->tag_set.ops = &blkfront_mq_ops;
977 info->tag_set.nr_hw_queues = info->nr_rings;
978 if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
980 * When indirect descriptior is not supported, the I/O request
981 * will be split between multiple request in the ring.
982 * To avoid problems when sending the request, divide by
983 * 2 the depth of the queue.
985 info->tag_set.queue_depth = BLK_RING_SIZE(info) / 2;
987 info->tag_set.queue_depth = BLK_RING_SIZE(info);
988 info->tag_set.numa_node = NUMA_NO_NODE;
989 info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
990 info->tag_set.cmd_size = sizeof(struct blkif_req);
991 info->tag_set.driver_data = info;
993 if (blk_mq_alloc_tag_set(&info->tag_set))
995 rq = blk_mq_init_queue(&info->tag_set);
997 blk_mq_free_tag_set(&info->tag_set);
1001 rq->queuedata = info;
1002 info->rq = gd->queue = rq;
1004 info->sector_size = sector_size;
1005 info->physical_sector_size = physical_sector_size;
1006 blkif_set_queue_limits(info);
1011 static const char *flush_info(struct blkfront_info *info)
1013 if (info->feature_flush && info->feature_fua)
1014 return "barrier: enabled;";
1015 else if (info->feature_flush)
1016 return "flush diskcache: enabled;";
1018 return "barrier or flush: disabled;";
1021 static void xlvbd_flush(struct blkfront_info *info)
1023 blk_queue_write_cache(info->rq, info->feature_flush ? true : false,
1024 info->feature_fua ? true : false);
1025 pr_info("blkfront: %s: %s %s %s %s %s\n",
1026 info->gd->disk_name, flush_info(info),
1027 "persistent grants:", info->feature_persistent ?
1028 "enabled;" : "disabled;", "indirect descriptors:",
1029 info->max_indirect_segments ? "enabled;" : "disabled;");
1032 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
1035 major = BLKIF_MAJOR(vdevice);
1036 *minor = BLKIF_MINOR(vdevice);
1038 case XEN_IDE0_MAJOR:
1039 *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
1040 *minor = ((*minor / 64) * PARTS_PER_DISK) +
1041 EMULATED_HD_DISK_MINOR_OFFSET;
1043 case XEN_IDE1_MAJOR:
1044 *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1045 *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1046 EMULATED_HD_DISK_MINOR_OFFSET;
1048 case XEN_SCSI_DISK0_MAJOR:
1049 *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1050 *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1052 case XEN_SCSI_DISK1_MAJOR:
1053 case XEN_SCSI_DISK2_MAJOR:
1054 case XEN_SCSI_DISK3_MAJOR:
1055 case XEN_SCSI_DISK4_MAJOR:
1056 case XEN_SCSI_DISK5_MAJOR:
1057 case XEN_SCSI_DISK6_MAJOR:
1058 case XEN_SCSI_DISK7_MAJOR:
1059 *offset = (*minor / PARTS_PER_DISK) +
1060 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1061 EMULATED_SD_DISK_NAME_OFFSET;
1063 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1064 EMULATED_SD_DISK_MINOR_OFFSET;
1066 case XEN_SCSI_DISK8_MAJOR:
1067 case XEN_SCSI_DISK9_MAJOR:
1068 case XEN_SCSI_DISK10_MAJOR:
1069 case XEN_SCSI_DISK11_MAJOR:
1070 case XEN_SCSI_DISK12_MAJOR:
1071 case XEN_SCSI_DISK13_MAJOR:
1072 case XEN_SCSI_DISK14_MAJOR:
1073 case XEN_SCSI_DISK15_MAJOR:
1074 *offset = (*minor / PARTS_PER_DISK) +
1075 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1076 EMULATED_SD_DISK_NAME_OFFSET;
1078 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1079 EMULATED_SD_DISK_MINOR_OFFSET;
1082 *offset = *minor / PARTS_PER_DISK;
1085 printk(KERN_WARNING "blkfront: your disk configuration is "
1086 "incorrect, please use an xvd device instead\n");
1092 static char *encode_disk_name(char *ptr, unsigned int n)
1095 ptr = encode_disk_name(ptr, n / 26 - 1);
1096 *ptr = 'a' + n % 26;
1100 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1101 struct blkfront_info *info,
1102 u16 vdisk_info, u16 sector_size,
1103 unsigned int physical_sector_size)
1108 unsigned int offset;
1113 BUG_ON(info->gd != NULL);
1114 BUG_ON(info->rq != NULL);
1116 if ((info->vdevice>>EXT_SHIFT) > 1) {
1117 /* this is above the extended range; something is wrong */
1118 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1122 if (!VDEV_IS_EXTENDED(info->vdevice)) {
1123 err = xen_translate_vdev(info->vdevice, &minor, &offset);
1126 nr_parts = PARTS_PER_DISK;
1128 minor = BLKIF_MINOR_EXT(info->vdevice);
1129 nr_parts = PARTS_PER_EXT_DISK;
1130 offset = minor / nr_parts;
1131 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1132 printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1133 "emulated IDE disks,\n\t choose an xvd device name"
1134 "from xvde on\n", info->vdevice);
1136 if (minor >> MINORBITS) {
1137 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1138 info->vdevice, minor);
1142 if ((minor % nr_parts) == 0)
1143 nr_minors = nr_parts;
1145 err = xlbd_reserve_minors(minor, nr_minors);
1150 gd = alloc_disk(nr_minors);
1154 strcpy(gd->disk_name, DEV_NAME);
1155 ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1156 BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1160 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1161 "%d", minor & (nr_parts - 1));
1163 gd->major = XENVBD_MAJOR;
1164 gd->first_minor = minor;
1165 gd->fops = &xlvbd_block_fops;
1166 gd->private_data = info;
1167 set_capacity(gd, capacity);
1169 if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size)) {
1176 if (vdisk_info & VDISK_READONLY)
1179 if (vdisk_info & VDISK_REMOVABLE)
1180 gd->flags |= GENHD_FL_REMOVABLE;
1182 if (vdisk_info & VDISK_CDROM)
1183 gd->flags |= GENHD_FL_CD;
1188 xlbd_release_minors(minor, nr_minors);
1193 static void xlvbd_release_gendisk(struct blkfront_info *info)
1195 unsigned int minor, nr_minors, i;
1196 struct blkfront_ring_info *rinfo;
1198 if (info->rq == NULL)
1201 /* No more blkif_request(). */
1202 blk_mq_stop_hw_queues(info->rq);
1204 for_each_rinfo(info, rinfo, i) {
1205 /* No more gnttab callback work. */
1206 gnttab_cancel_free_callback(&rinfo->callback);
1208 /* Flush gnttab callback work. Must be done with no locks held. */
1209 flush_work(&rinfo->work);
1212 del_gendisk(info->gd);
1214 minor = info->gd->first_minor;
1215 nr_minors = info->gd->minors;
1216 xlbd_release_minors(minor, nr_minors);
1218 blk_cleanup_queue(info->rq);
1219 blk_mq_free_tag_set(&info->tag_set);
1226 /* Already hold rinfo->ring_lock. */
1227 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1229 if (!RING_FULL(&rinfo->ring))
1230 blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1233 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1235 unsigned long flags;
1237 spin_lock_irqsave(&rinfo->ring_lock, flags);
1238 kick_pending_request_queues_locked(rinfo);
1239 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1242 static void blkif_restart_queue(struct work_struct *work)
1244 struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1246 if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1247 kick_pending_request_queues(rinfo);
1250 static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1252 struct grant *persistent_gnt, *n;
1253 struct blkfront_info *info = rinfo->dev_info;
1257 * Remove indirect pages, this only happens when using indirect
1258 * descriptors but not persistent grants
1260 if (!list_empty(&rinfo->indirect_pages)) {
1261 struct page *indirect_page, *n;
1263 BUG_ON(info->feature_persistent);
1264 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1265 list_del(&indirect_page->lru);
1266 __free_page(indirect_page);
1270 /* Remove all persistent grants. */
1271 if (!list_empty(&rinfo->grants)) {
1272 list_for_each_entry_safe(persistent_gnt, n,
1273 &rinfo->grants, node) {
1274 list_del(&persistent_gnt->node);
1275 if (persistent_gnt->gref != GRANT_INVALID_REF) {
1276 gnttab_end_foreign_access(persistent_gnt->gref,
1278 rinfo->persistent_gnts_c--;
1280 if (info->feature_persistent)
1281 __free_page(persistent_gnt->page);
1282 kfree(persistent_gnt);
1285 BUG_ON(rinfo->persistent_gnts_c != 0);
1287 for (i = 0; i < BLK_RING_SIZE(info); i++) {
1289 * Clear persistent grants present in requests already
1290 * on the shared ring
1292 if (!rinfo->shadow[i].request)
1295 segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1296 rinfo->shadow[i].req.u.indirect.nr_segments :
1297 rinfo->shadow[i].req.u.rw.nr_segments;
1298 for (j = 0; j < segs; j++) {
1299 persistent_gnt = rinfo->shadow[i].grants_used[j];
1300 gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1301 if (info->feature_persistent)
1302 __free_page(persistent_gnt->page);
1303 kfree(persistent_gnt);
1306 if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1308 * If this is not an indirect operation don't try to
1309 * free indirect segments
1313 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1314 persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1315 gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1316 __free_page(persistent_gnt->page);
1317 kfree(persistent_gnt);
1321 kvfree(rinfo->shadow[i].grants_used);
1322 rinfo->shadow[i].grants_used = NULL;
1323 kvfree(rinfo->shadow[i].indirect_grants);
1324 rinfo->shadow[i].indirect_grants = NULL;
1325 kvfree(rinfo->shadow[i].sg);
1326 rinfo->shadow[i].sg = NULL;
1329 /* No more gnttab callback work. */
1330 gnttab_cancel_free_callback(&rinfo->callback);
1332 /* Flush gnttab callback work. Must be done with no locks held. */
1333 flush_work(&rinfo->work);
1335 /* Free resources associated with old device channel. */
1336 for (i = 0; i < info->nr_ring_pages; i++) {
1337 if (rinfo->ring_ref[i] != GRANT_INVALID_REF) {
1338 gnttab_end_foreign_access(rinfo->ring_ref[i], 0, 0);
1339 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1342 free_pages((unsigned long)rinfo->ring.sring, get_order(info->nr_ring_pages * XEN_PAGE_SIZE));
1343 rinfo->ring.sring = NULL;
1346 unbind_from_irqhandler(rinfo->irq, rinfo);
1347 rinfo->evtchn = rinfo->irq = 0;
1350 static void blkif_free(struct blkfront_info *info, int suspend)
1353 struct blkfront_ring_info *rinfo;
1355 /* Prevent new requests being issued until we fix things up. */
1356 info->connected = suspend ?
1357 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1358 /* No more blkif_request(). */
1360 blk_mq_stop_hw_queues(info->rq);
1362 for_each_rinfo(info, rinfo, i)
1363 blkif_free_ring(rinfo);
1365 kvfree(info->rinfo);
1370 struct copy_from_grant {
1371 const struct blk_shadow *s;
1372 unsigned int grant_idx;
1373 unsigned int bvec_offset;
1377 static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1378 unsigned int len, void *data)
1380 struct copy_from_grant *info = data;
1382 /* Convenient aliases */
1383 const struct blk_shadow *s = info->s;
1385 shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1387 memcpy(info->bvec_data + info->bvec_offset,
1388 shared_data + offset, len);
1390 info->bvec_offset += len;
1393 kunmap_atomic(shared_data);
1396 static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1400 case BLKIF_RSP_OKAY:
1402 case BLKIF_RSP_EOPNOTSUPP:
1403 return REQ_EOPNOTSUPP;
1404 case BLKIF_RSP_ERROR:
1412 * Get the final status of the block request based on two ring response
1414 static int blkif_get_final_status(enum blk_req_status s1,
1415 enum blk_req_status s2)
1417 BUG_ON(s1 == REQ_WAITING);
1418 BUG_ON(s2 == REQ_WAITING);
1420 if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1421 return BLKIF_RSP_ERROR;
1422 else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1423 return BLKIF_RSP_EOPNOTSUPP;
1424 return BLKIF_RSP_OKAY;
1427 static bool blkif_completion(unsigned long *id,
1428 struct blkfront_ring_info *rinfo,
1429 struct blkif_response *bret)
1432 struct scatterlist *sg;
1433 int num_sg, num_grant;
1434 struct blkfront_info *info = rinfo->dev_info;
1435 struct blk_shadow *s = &rinfo->shadow[*id];
1436 struct copy_from_grant data = {
1440 num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1441 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1443 /* The I/O request may be split in two. */
1444 if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1445 struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1447 /* Keep the status of the current response in shadow. */
1448 s->status = blkif_rsp_to_req_status(bret->status);
1450 /* Wait the second response if not yet here. */
1451 if (s2->status == REQ_WAITING)
1454 bret->status = blkif_get_final_status(s->status,
1458 * All the grants is stored in the first shadow in order
1459 * to make the completion code simpler.
1461 num_grant += s2->req.u.rw.nr_segments;
1464 * The two responses may not come in order. Only the
1465 * first request will store the scatter-gather list.
1467 if (s2->num_sg != 0) {
1468 /* Update "id" with the ID of the first response. */
1469 *id = s->associated_id;
1474 * We don't need anymore the second request, so recycling
1477 if (add_id_to_freelist(rinfo, s->associated_id))
1478 WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1479 info->gd->disk_name, s->associated_id);
1485 if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1486 for_each_sg(s->sg, sg, num_sg, i) {
1487 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1489 data.bvec_offset = sg->offset;
1490 data.bvec_data = kmap_atomic(sg_page(sg));
1492 gnttab_foreach_grant_in_range(sg_page(sg),
1495 blkif_copy_from_grant,
1498 kunmap_atomic(data.bvec_data);
1501 /* Add the persistent grant into the list of free grants */
1502 for (i = 0; i < num_grant; i++) {
1503 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1505 * If the grant is still mapped by the backend (the
1506 * backend has chosen to make this grant persistent)
1507 * we add it at the head of the list, so it will be
1510 if (!info->feature_persistent)
1511 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1512 s->grants_used[i]->gref);
1513 list_add(&s->grants_used[i]->node, &rinfo->grants);
1514 rinfo->persistent_gnts_c++;
1517 * If the grant is not mapped by the backend we end the
1518 * foreign access and add it to the tail of the list,
1519 * so it will not be picked again unless we run out of
1520 * persistent grants.
1522 gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1523 s->grants_used[i]->gref = GRANT_INVALID_REF;
1524 list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1527 if (s->req.operation == BLKIF_OP_INDIRECT) {
1528 for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1529 if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1530 if (!info->feature_persistent)
1531 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1532 s->indirect_grants[i]->gref);
1533 list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1534 rinfo->persistent_gnts_c++;
1536 struct page *indirect_page;
1538 gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1540 * Add the used indirect page back to the list of
1541 * available pages for indirect grefs.
1543 if (!info->feature_persistent) {
1544 indirect_page = s->indirect_grants[i]->page;
1545 list_add(&indirect_page->lru, &rinfo->indirect_pages);
1547 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1548 list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1556 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1558 struct request *req;
1559 struct blkif_response *bret;
1561 unsigned long flags;
1562 struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1563 struct blkfront_info *info = rinfo->dev_info;
1565 if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
1568 spin_lock_irqsave(&rinfo->ring_lock, flags);
1570 rp = rinfo->ring.sring->rsp_prod;
1571 rmb(); /* Ensure we see queued responses up to 'rp'. */
1573 for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1576 bret = RING_GET_RESPONSE(&rinfo->ring, i);
1579 * The backend has messed up and given us an id that we would
1580 * never have given to it (we stamp it up to BLK_RING_SIZE -
1581 * look in get_id_from_freelist.
1583 if (id >= BLK_RING_SIZE(info)) {
1584 WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1585 info->gd->disk_name, op_name(bret->operation), id);
1586 /* We can't safely get the 'struct request' as
1587 * the id is busted. */
1590 req = rinfo->shadow[id].request;
1592 if (bret->operation != BLKIF_OP_DISCARD) {
1594 * We may need to wait for an extra response if the
1595 * I/O request is split in 2
1597 if (!blkif_completion(&id, rinfo, bret))
1601 if (add_id_to_freelist(rinfo, id)) {
1602 WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1603 info->gd->disk_name, op_name(bret->operation), id);
1607 if (bret->status == BLKIF_RSP_OKAY)
1608 blkif_req(req)->error = BLK_STS_OK;
1610 blkif_req(req)->error = BLK_STS_IOERR;
1612 switch (bret->operation) {
1613 case BLKIF_OP_DISCARD:
1614 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1615 struct request_queue *rq = info->rq;
1616 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1617 info->gd->disk_name, op_name(bret->operation));
1618 blkif_req(req)->error = BLK_STS_NOTSUPP;
1619 info->feature_discard = 0;
1620 info->feature_secdiscard = 0;
1621 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1622 blk_queue_flag_clear(QUEUE_FLAG_SECERASE, rq);
1625 case BLKIF_OP_FLUSH_DISKCACHE:
1626 case BLKIF_OP_WRITE_BARRIER:
1627 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1628 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1629 info->gd->disk_name, op_name(bret->operation));
1630 blkif_req(req)->error = BLK_STS_NOTSUPP;
1632 if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1633 rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1634 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1635 info->gd->disk_name, op_name(bret->operation));
1636 blkif_req(req)->error = BLK_STS_NOTSUPP;
1638 if (unlikely(blkif_req(req)->error)) {
1639 if (blkif_req(req)->error == BLK_STS_NOTSUPP)
1640 blkif_req(req)->error = BLK_STS_OK;
1641 info->feature_fua = 0;
1642 info->feature_flush = 0;
1647 case BLKIF_OP_WRITE:
1648 if (unlikely(bret->status != BLKIF_RSP_OKAY))
1649 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1650 "request: %x\n", bret->status);
1657 blk_mq_complete_request(req);
1660 rinfo->ring.rsp_cons = i;
1662 if (i != rinfo->ring.req_prod_pvt) {
1664 RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1668 rinfo->ring.sring->rsp_event = i + 1;
1670 kick_pending_request_queues_locked(rinfo);
1672 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1678 static int setup_blkring(struct xenbus_device *dev,
1679 struct blkfront_ring_info *rinfo)
1681 struct blkif_sring *sring;
1683 struct blkfront_info *info = rinfo->dev_info;
1684 unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1685 grant_ref_t gref[XENBUS_MAX_RING_GRANTS];
1687 for (i = 0; i < info->nr_ring_pages; i++)
1688 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1690 sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
1691 get_order(ring_size));
1693 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1696 SHARED_RING_INIT(sring);
1697 FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1699 err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref);
1701 free_pages((unsigned long)sring, get_order(ring_size));
1702 rinfo->ring.sring = NULL;
1705 for (i = 0; i < info->nr_ring_pages; i++)
1706 rinfo->ring_ref[i] = gref[i];
1708 err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1712 err = bind_evtchn_to_irqhandler(rinfo->evtchn, blkif_interrupt, 0,
1715 xenbus_dev_fatal(dev, err,
1716 "bind_evtchn_to_irqhandler failed");
1723 blkif_free(info, 0);
1728 * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1729 * ring buffer may have multi pages depending on ->nr_ring_pages.
1731 static int write_per_ring_nodes(struct xenbus_transaction xbt,
1732 struct blkfront_ring_info *rinfo, const char *dir)
1736 const char *message = NULL;
1737 struct blkfront_info *info = rinfo->dev_info;
1739 if (info->nr_ring_pages == 1) {
1740 err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1742 message = "writing ring-ref";
1743 goto abort_transaction;
1746 for (i = 0; i < info->nr_ring_pages; i++) {
1747 char ring_ref_name[RINGREF_NAME_LEN];
1749 snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1750 err = xenbus_printf(xbt, dir, ring_ref_name,
1751 "%u", rinfo->ring_ref[i]);
1753 message = "writing ring-ref";
1754 goto abort_transaction;
1759 err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1761 message = "writing event-channel";
1762 goto abort_transaction;
1768 xenbus_transaction_end(xbt, 1);
1770 xenbus_dev_fatal(info->xbdev, err, "%s", message);
1775 static void free_info(struct blkfront_info *info)
1777 list_del(&info->info_list);
1781 /* Common code used when first setting up, and when resuming. */
1782 static int talk_to_blkback(struct xenbus_device *dev,
1783 struct blkfront_info *info)
1785 const char *message = NULL;
1786 struct xenbus_transaction xbt;
1788 unsigned int i, max_page_order;
1789 unsigned int ring_page_order;
1790 struct blkfront_ring_info *rinfo;
1795 max_page_order = xenbus_read_unsigned(info->xbdev->otherend,
1796 "max-ring-page-order", 0);
1797 ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1798 info->nr_ring_pages = 1 << ring_page_order;
1800 err = negotiate_mq(info);
1802 goto destroy_blkring;
1804 for_each_rinfo(info, rinfo, i) {
1805 /* Create shared ring, alloc event channel. */
1806 err = setup_blkring(dev, rinfo);
1808 goto destroy_blkring;
1812 err = xenbus_transaction_start(&xbt);
1814 xenbus_dev_fatal(dev, err, "starting transaction");
1815 goto destroy_blkring;
1818 if (info->nr_ring_pages > 1) {
1819 err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1822 message = "writing ring-page-order";
1823 goto abort_transaction;
1827 /* We already got the number of queues/rings in _probe */
1828 if (info->nr_rings == 1) {
1829 err = write_per_ring_nodes(xbt, info->rinfo, dev->nodename);
1831 goto destroy_blkring;
1836 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1839 message = "writing multi-queue-num-queues";
1840 goto abort_transaction;
1843 pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1844 path = kmalloc(pathsize, GFP_KERNEL);
1847 message = "ENOMEM while writing ring references";
1848 goto abort_transaction;
1851 for_each_rinfo(info, rinfo, i) {
1852 memset(path, 0, pathsize);
1853 snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1854 err = write_per_ring_nodes(xbt, rinfo, path);
1857 goto destroy_blkring;
1862 err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1863 XEN_IO_PROTO_ABI_NATIVE);
1865 message = "writing protocol";
1866 goto abort_transaction;
1868 err = xenbus_printf(xbt, dev->nodename,
1869 "feature-persistent", "%u", 1);
1872 "writing persistent grants feature to xenbus");
1874 err = xenbus_transaction_end(xbt, 0);
1878 xenbus_dev_fatal(dev, err, "completing transaction");
1879 goto destroy_blkring;
1882 for_each_rinfo(info, rinfo, i) {
1885 for (j = 0; j < BLK_RING_SIZE(info); j++)
1886 rinfo->shadow[j].req.u.rw.id = j + 1;
1887 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1889 xenbus_switch_state(dev, XenbusStateInitialised);
1894 xenbus_transaction_end(xbt, 1);
1896 xenbus_dev_fatal(dev, err, "%s", message);
1898 blkif_free(info, 0);
1900 mutex_lock(&blkfront_mutex);
1902 mutex_unlock(&blkfront_mutex);
1904 dev_set_drvdata(&dev->dev, NULL);
1909 static int negotiate_mq(struct blkfront_info *info)
1911 unsigned int backend_max_queues;
1913 struct blkfront_ring_info *rinfo;
1915 BUG_ON(info->nr_rings);
1917 /* Check if backend supports multiple queues. */
1918 backend_max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1919 "multi-queue-max-queues", 1);
1920 info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1921 /* We need at least one ring. */
1922 if (!info->nr_rings)
1925 info->rinfo_size = struct_size(info->rinfo, shadow,
1926 BLK_RING_SIZE(info));
1927 info->rinfo = kvcalloc(info->nr_rings, info->rinfo_size, GFP_KERNEL);
1929 xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1934 for_each_rinfo(info, rinfo, i) {
1935 INIT_LIST_HEAD(&rinfo->indirect_pages);
1936 INIT_LIST_HEAD(&rinfo->grants);
1937 rinfo->dev_info = info;
1938 INIT_WORK(&rinfo->work, blkif_restart_queue);
1939 spin_lock_init(&rinfo->ring_lock);
1944 * Entry point to this code when a new device is created. Allocate the basic
1945 * structures and the ring buffer for communication with the backend, and
1946 * inform the backend of the appropriate details for those. Switch to
1947 * Initialised state.
1949 static int blkfront_probe(struct xenbus_device *dev,
1950 const struct xenbus_device_id *id)
1953 struct blkfront_info *info;
1955 /* FIXME: Use dynamic device id if this is not set. */
1956 err = xenbus_scanf(XBT_NIL, dev->nodename,
1957 "virtual-device", "%i", &vdevice);
1959 /* go looking in the extended area instead */
1960 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1963 xenbus_dev_fatal(dev, err, "reading virtual-device");
1968 if (xen_hvm_domain()) {
1971 /* no unplug has been done: do not hook devices != xen vbds */
1972 if (xen_has_pv_and_legacy_disk_devices()) {
1975 if (!VDEV_IS_EXTENDED(vdevice))
1976 major = BLKIF_MAJOR(vdevice);
1978 major = XENVBD_MAJOR;
1980 if (major != XENVBD_MAJOR) {
1982 "%s: HVM does not support vbd %d as xen block device\n",
1987 /* do not create a PV cdrom device if we are an HVM guest */
1988 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1991 if (strncmp(type, "cdrom", 5) == 0) {
1997 info = kzalloc(sizeof(*info), GFP_KERNEL);
1999 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
2005 mutex_init(&info->mutex);
2006 info->vdevice = vdevice;
2007 info->connected = BLKIF_STATE_DISCONNECTED;
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 block_device *bdev = NULL;
2138 mutex_lock(&info->mutex);
2140 if (xbdev->state == XenbusStateClosing) {
2141 mutex_unlock(&info->mutex);
2146 bdev = bdget_disk(info->gd, 0);
2148 mutex_unlock(&info->mutex);
2151 xenbus_frontend_closed(xbdev);
2155 mutex_lock(&bdev->bd_mutex);
2157 if (bdev->bd_openers) {
2158 xenbus_dev_error(xbdev, -EBUSY,
2159 "Device in use; refusing to close");
2160 xenbus_switch_state(xbdev, XenbusStateClosing);
2162 xlvbd_release_gendisk(info);
2163 xenbus_frontend_closed(xbdev);
2166 mutex_unlock(&bdev->bd_mutex);
2170 static void blkfront_setup_discard(struct blkfront_info *info)
2173 unsigned int discard_granularity;
2174 unsigned int discard_alignment;
2176 info->feature_discard = 1;
2177 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2178 "discard-granularity", "%u", &discard_granularity,
2179 "discard-alignment", "%u", &discard_alignment,
2182 info->discard_granularity = discard_granularity;
2183 info->discard_alignment = discard_alignment;
2185 info->feature_secdiscard =
2186 !!xenbus_read_unsigned(info->xbdev->otherend, "discard-secure",
2190 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2192 unsigned int psegs, grants;
2194 struct blkfront_info *info = rinfo->dev_info;
2196 if (info->max_indirect_segments == 0) {
2198 grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2201 * When an extra req is required, the maximum
2202 * grants supported is related to the size of the
2203 * Linux block segment.
2205 grants = GRANTS_PER_PSEG;
2209 grants = info->max_indirect_segments;
2210 psegs = DIV_ROUND_UP(grants, GRANTS_PER_PSEG);
2212 err = fill_grant_buffer(rinfo,
2213 (grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2217 if (!info->feature_persistent && info->max_indirect_segments) {
2219 * We are using indirect descriptors but not persistent
2220 * grants, we need to allocate a set of pages that can be
2221 * used for mapping indirect grefs
2223 int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2225 BUG_ON(!list_empty(&rinfo->indirect_pages));
2226 for (i = 0; i < num; i++) {
2227 struct page *indirect_page = alloc_page(GFP_NOIO);
2230 list_add(&indirect_page->lru, &rinfo->indirect_pages);
2234 for (i = 0; i < BLK_RING_SIZE(info); i++) {
2235 rinfo->shadow[i].grants_used =
2237 sizeof(rinfo->shadow[i].grants_used[0]),
2239 rinfo->shadow[i].sg = kvcalloc(psegs,
2240 sizeof(rinfo->shadow[i].sg[0]),
2242 if (info->max_indirect_segments)
2243 rinfo->shadow[i].indirect_grants =
2244 kvcalloc(INDIRECT_GREFS(grants),
2245 sizeof(rinfo->shadow[i].indirect_grants[0]),
2247 if ((rinfo->shadow[i].grants_used == NULL) ||
2248 (rinfo->shadow[i].sg == NULL) ||
2249 (info->max_indirect_segments &&
2250 (rinfo->shadow[i].indirect_grants == NULL)))
2252 sg_init_table(rinfo->shadow[i].sg, psegs);
2259 for (i = 0; i < BLK_RING_SIZE(info); i++) {
2260 kvfree(rinfo->shadow[i].grants_used);
2261 rinfo->shadow[i].grants_used = NULL;
2262 kvfree(rinfo->shadow[i].sg);
2263 rinfo->shadow[i].sg = NULL;
2264 kvfree(rinfo->shadow[i].indirect_grants);
2265 rinfo->shadow[i].indirect_grants = NULL;
2267 if (!list_empty(&rinfo->indirect_pages)) {
2268 struct page *indirect_page, *n;
2269 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2270 list_del(&indirect_page->lru);
2271 __free_page(indirect_page);
2278 * Gather all backend feature-*
2280 static void blkfront_gather_backend_features(struct blkfront_info *info)
2282 unsigned int indirect_segments;
2284 info->feature_flush = 0;
2285 info->feature_fua = 0;
2288 * If there's no "feature-barrier" defined, then it means
2289 * we're dealing with a very old backend which writes
2290 * synchronously; nothing to do.
2292 * If there are barriers, then we use flush.
2294 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-barrier", 0)) {
2295 info->feature_flush = 1;
2296 info->feature_fua = 1;
2300 * And if there is "feature-flush-cache" use that above
2303 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-flush-cache",
2305 info->feature_flush = 1;
2306 info->feature_fua = 0;
2309 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-discard", 0))
2310 blkfront_setup_discard(info);
2312 info->feature_persistent =
2313 !!xenbus_read_unsigned(info->xbdev->otherend,
2314 "feature-persistent", 0);
2316 indirect_segments = xenbus_read_unsigned(info->xbdev->otherend,
2317 "feature-max-indirect-segments", 0);
2318 if (indirect_segments > xen_blkif_max_segments)
2319 indirect_segments = xen_blkif_max_segments;
2320 if (indirect_segments <= BLKIF_MAX_SEGMENTS_PER_REQUEST)
2321 indirect_segments = 0;
2322 info->max_indirect_segments = indirect_segments;
2324 if (info->feature_persistent) {
2325 mutex_lock(&blkfront_mutex);
2326 schedule_delayed_work(&blkfront_work, HZ * 10);
2327 mutex_unlock(&blkfront_mutex);
2332 * Invoked when the backend is finally 'ready' (and has told produced
2333 * the details about the physical device - #sectors, size, etc).
2335 static void blkfront_connect(struct blkfront_info *info)
2337 unsigned long long sectors;
2338 unsigned long sector_size;
2339 unsigned int physical_sector_size;
2342 struct blkfront_ring_info *rinfo;
2344 switch (info->connected) {
2345 case BLKIF_STATE_CONNECTED:
2347 * Potentially, the back-end may be signalling
2348 * a capacity change; update the capacity.
2350 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2351 "sectors", "%Lu", §ors);
2352 if (XENBUS_EXIST_ERR(err))
2354 printk(KERN_INFO "Setting capacity to %Lu\n",
2356 set_capacity_revalidate_and_notify(info->gd, sectors, true);
2359 case BLKIF_STATE_SUSPENDED:
2361 * If we are recovering from suspension, we need to wait
2362 * for the backend to announce it's features before
2363 * reconnecting, at least we need to know if the backend
2364 * supports indirect descriptors, and how many.
2366 blkif_recover(info);
2373 dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2374 __func__, info->xbdev->otherend);
2376 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2377 "sectors", "%llu", §ors,
2378 "info", "%u", &binfo,
2379 "sector-size", "%lu", §or_size,
2382 xenbus_dev_fatal(info->xbdev, err,
2383 "reading backend fields at %s",
2384 info->xbdev->otherend);
2389 * physcial-sector-size is a newer field, so old backends may not
2390 * provide this. Assume physical sector size to be the same as
2391 * sector_size in that case.
2393 physical_sector_size = xenbus_read_unsigned(info->xbdev->otherend,
2394 "physical-sector-size",
2396 blkfront_gather_backend_features(info);
2397 for_each_rinfo(info, rinfo, i) {
2398 err = blkfront_setup_indirect(rinfo);
2400 xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2401 info->xbdev->otherend);
2402 blkif_free(info, 0);
2407 err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
2408 physical_sector_size);
2410 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2411 info->xbdev->otherend);
2415 xenbus_switch_state(info->xbdev, XenbusStateConnected);
2417 /* Kick pending requests. */
2418 info->connected = BLKIF_STATE_CONNECTED;
2419 for_each_rinfo(info, rinfo, i)
2420 kick_pending_request_queues(rinfo);
2422 device_add_disk(&info->xbdev->dev, info->gd, NULL);
2428 blkif_free(info, 0);
2433 * Callback received when the backend's state changes.
2435 static void blkback_changed(struct xenbus_device *dev,
2436 enum xenbus_state backend_state)
2438 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2440 dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2442 switch (backend_state) {
2443 case XenbusStateInitWait:
2444 if (dev->state != XenbusStateInitialising)
2446 if (talk_to_blkback(dev, info))
2448 case XenbusStateInitialising:
2449 case XenbusStateInitialised:
2450 case XenbusStateReconfiguring:
2451 case XenbusStateReconfigured:
2452 case XenbusStateUnknown:
2455 case XenbusStateConnected:
2457 * talk_to_blkback sets state to XenbusStateInitialised
2458 * and blkfront_connect sets it to XenbusStateConnected
2459 * (if connection went OK).
2461 * If the backend (or toolstack) decides to poke at backend
2462 * state (and re-trigger the watch by setting the state repeatedly
2463 * to XenbusStateConnected (4)) we need to deal with this.
2464 * This is allowed as this is used to communicate to the guest
2465 * that the size of disk has changed!
2467 if ((dev->state != XenbusStateInitialised) &&
2468 (dev->state != XenbusStateConnected)) {
2469 if (talk_to_blkback(dev, info))
2473 blkfront_connect(info);
2476 case XenbusStateClosed:
2477 if (dev->state == XenbusStateClosed)
2480 case XenbusStateClosing:
2482 blkfront_closing(info);
2487 static int blkfront_remove(struct xenbus_device *xbdev)
2489 struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2490 struct block_device *bdev = NULL;
2491 struct gendisk *disk;
2493 dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2498 blkif_free(info, 0);
2500 mutex_lock(&info->mutex);
2504 bdev = bdget_disk(disk, 0);
2507 mutex_unlock(&info->mutex);
2510 mutex_lock(&blkfront_mutex);
2512 mutex_unlock(&blkfront_mutex);
2517 * The xbdev was removed before we reached the Closed
2518 * state. See if it's safe to remove the disk. If the bdev
2519 * isn't closed yet, we let release take care of it.
2522 mutex_lock(&bdev->bd_mutex);
2523 info = disk->private_data;
2525 dev_warn(disk_to_dev(disk),
2526 "%s was hot-unplugged, %d stale handles\n",
2527 xbdev->nodename, bdev->bd_openers);
2529 if (info && !bdev->bd_openers) {
2530 xlvbd_release_gendisk(info);
2531 disk->private_data = NULL;
2532 mutex_lock(&blkfront_mutex);
2534 mutex_unlock(&blkfront_mutex);
2537 mutex_unlock(&bdev->bd_mutex);
2543 static int blkfront_is_ready(struct xenbus_device *dev)
2545 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2547 return info->is_ready && info->xbdev;
2550 static int blkif_open(struct block_device *bdev, fmode_t mode)
2552 struct gendisk *disk = bdev->bd_disk;
2553 struct blkfront_info *info;
2556 mutex_lock(&blkfront_mutex);
2558 info = disk->private_data;
2565 mutex_lock(&info->mutex);
2568 /* xbdev is closed */
2571 mutex_unlock(&info->mutex);
2574 mutex_unlock(&blkfront_mutex);
2578 static void blkif_release(struct gendisk *disk, fmode_t mode)
2580 struct blkfront_info *info = disk->private_data;
2581 struct block_device *bdev;
2582 struct xenbus_device *xbdev;
2584 mutex_lock(&blkfront_mutex);
2586 bdev = bdget_disk(disk, 0);
2589 WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2592 if (bdev->bd_openers)
2596 * Check if we have been instructed to close. We will have
2597 * deferred this request, because the bdev was still open.
2600 mutex_lock(&info->mutex);
2601 xbdev = info->xbdev;
2603 if (xbdev && xbdev->state == XenbusStateClosing) {
2604 /* pending switch to state closed */
2605 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2606 xlvbd_release_gendisk(info);
2607 xenbus_frontend_closed(info->xbdev);
2610 mutex_unlock(&info->mutex);
2613 /* sudden device removal */
2614 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2615 xlvbd_release_gendisk(info);
2616 disk->private_data = NULL;
2623 mutex_unlock(&blkfront_mutex);
2626 static const struct block_device_operations xlvbd_block_fops =
2628 .owner = THIS_MODULE,
2630 .release = blkif_release,
2631 .getgeo = blkif_getgeo,
2632 .ioctl = blkif_ioctl,
2633 .compat_ioctl = blkdev_compat_ptr_ioctl,
2637 static const struct xenbus_device_id blkfront_ids[] = {
2642 static struct xenbus_driver blkfront_driver = {
2643 .ids = blkfront_ids,
2644 .probe = blkfront_probe,
2645 .remove = blkfront_remove,
2646 .resume = blkfront_resume,
2647 .otherend_changed = blkback_changed,
2648 .is_ready = blkfront_is_ready,
2651 static void purge_persistent_grants(struct blkfront_info *info)
2654 unsigned long flags;
2655 struct blkfront_ring_info *rinfo;
2657 for_each_rinfo(info, rinfo, i) {
2658 struct grant *gnt_list_entry, *tmp;
2660 spin_lock_irqsave(&rinfo->ring_lock, flags);
2662 if (rinfo->persistent_gnts_c == 0) {
2663 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2667 list_for_each_entry_safe(gnt_list_entry, tmp, &rinfo->grants,
2669 if (gnt_list_entry->gref == GRANT_INVALID_REF ||
2670 gnttab_query_foreign_access(gnt_list_entry->gref))
2673 list_del(&gnt_list_entry->node);
2674 gnttab_end_foreign_access(gnt_list_entry->gref, 0, 0UL);
2675 rinfo->persistent_gnts_c--;
2676 gnt_list_entry->gref = GRANT_INVALID_REF;
2677 list_add_tail(&gnt_list_entry->node, &rinfo->grants);
2680 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2684 static void blkfront_delay_work(struct work_struct *work)
2686 struct blkfront_info *info;
2687 bool need_schedule_work = false;
2689 mutex_lock(&blkfront_mutex);
2691 list_for_each_entry(info, &info_list, info_list) {
2692 if (info->feature_persistent) {
2693 need_schedule_work = true;
2694 mutex_lock(&info->mutex);
2695 purge_persistent_grants(info);
2696 mutex_unlock(&info->mutex);
2700 if (need_schedule_work)
2701 schedule_delayed_work(&blkfront_work, HZ * 10);
2703 mutex_unlock(&blkfront_mutex);
2706 static int __init xlblk_init(void)
2709 int nr_cpus = num_online_cpus();
2714 if (!xen_has_pv_disk_devices())
2717 if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2718 pr_warn("xen_blk: can't get major %d with name %s\n",
2719 XENVBD_MAJOR, DEV_NAME);
2723 if (xen_blkif_max_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST)
2724 xen_blkif_max_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2726 if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2727 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2728 xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2729 xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2732 if (xen_blkif_max_queues > nr_cpus) {
2733 pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2734 xen_blkif_max_queues, nr_cpus);
2735 xen_blkif_max_queues = nr_cpus;
2738 INIT_DELAYED_WORK(&blkfront_work, blkfront_delay_work);
2740 ret = xenbus_register_frontend(&blkfront_driver);
2742 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2748 module_init(xlblk_init);
2751 static void __exit xlblk_exit(void)
2753 cancel_delayed_work_sync(&blkfront_work);
2755 xenbus_unregister_driver(&blkfront_driver);
2756 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2759 module_exit(xlblk_exit);
2761 MODULE_DESCRIPTION("Xen virtual block device frontend");
2762 MODULE_LICENSE("GPL");
2763 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2764 MODULE_ALIAS("xen:vbd");
2765 MODULE_ALIAS("xenblk");