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>
51 #include <xen/xenbus.h>
52 #include <xen/grant_table.h>
53 #include <xen/events.h>
55 #include <xen/platform_pci.h>
57 #include <xen/interface/grant_table.h>
58 #include <xen/interface/io/blkif.h>
59 #include <xen/interface/io/protocols.h>
61 #include <asm/xen/hypervisor.h>
64 * The minimal size of segment supported by the block framework is PAGE_SIZE.
65 * When Linux is using a different page size than Xen, it may not be possible
66 * to put all the data in a single segment.
67 * This can happen when the backend doesn't support indirect descriptor and
68 * therefore the maximum amount of data that a request can carry is
69 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE = 44KB
71 * Note that we only support one extra request. So the Linux page size
72 * should be <= ( 2 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) =
75 #define HAS_EXTRA_REQ (BLKIF_MAX_SEGMENTS_PER_REQUEST < XEN_PFN_PER_PAGE)
78 BLKIF_STATE_DISCONNECTED,
79 BLKIF_STATE_CONNECTED,
80 BLKIF_STATE_SUSPENDED,
86 struct list_head node;
97 struct blkif_request req;
98 struct request *request;
99 struct grant **grants_used;
100 struct grant **indirect_grants;
101 struct scatterlist *sg;
103 enum blk_req_status status;
105 #define NO_ASSOCIATED_ID ~0UL
107 * Id of the sibling if we ever need 2 requests when handling a
110 unsigned long associated_id;
122 static inline struct blkif_req *blkif_req(struct request *rq)
124 return blk_mq_rq_to_pdu(rq);
127 static DEFINE_MUTEX(blkfront_mutex);
128 static const struct block_device_operations xlvbd_block_fops;
131 * Maximum number of segments in indirect requests, the actual value used by
132 * the frontend driver is the minimum of this value and the value provided
133 * by the backend driver.
136 static unsigned int xen_blkif_max_segments = 32;
137 module_param_named(max_indirect_segments, xen_blkif_max_segments, uint,
139 MODULE_PARM_DESC(max_indirect_segments,
140 "Maximum amount of segments in indirect requests (default is 32)");
142 static unsigned int xen_blkif_max_queues = 4;
143 module_param_named(max_queues, xen_blkif_max_queues, uint, S_IRUGO);
144 MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
147 * Maximum order of pages to be used for the shared ring between front and
148 * backend, 4KB page granularity is used.
150 static unsigned int xen_blkif_max_ring_order;
151 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, S_IRUGO);
152 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
154 #define BLK_RING_SIZE(info) \
155 __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
157 #define BLK_MAX_RING_SIZE \
158 __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * XENBUS_MAX_RING_GRANTS)
161 * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
162 * characters are enough. Define to 20 to keep consistent with backend.
164 #define RINGREF_NAME_LEN (20)
166 * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
168 #define QUEUE_NAME_LEN (17)
172 * Every blkfront device can associate with one or more blkfront_ring_info,
173 * depending on how many hardware queues/rings to be used.
175 struct blkfront_ring_info {
176 /* Lock to protect data in every ring buffer. */
177 spinlock_t ring_lock;
178 struct blkif_front_ring ring;
179 unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
180 unsigned int evtchn, irq;
181 struct work_struct work;
182 struct gnttab_free_callback callback;
183 struct blk_shadow shadow[BLK_MAX_RING_SIZE];
184 struct list_head indirect_pages;
185 struct list_head grants;
186 unsigned int persistent_gnts_c;
187 unsigned long shadow_free;
188 struct blkfront_info *dev_info;
192 * We have one of these per vbd, whether ide, scsi or 'other'. They
193 * hang in private_data off the gendisk structure. We may end up
194 * putting all kinds of interesting stuff here :-)
199 struct xenbus_device *xbdev;
202 unsigned int physical_sector_size;
205 enum blkif_state connected;
206 /* Number of pages per ring buffer. */
207 unsigned int nr_ring_pages;
208 struct request_queue *rq;
209 unsigned int feature_flush:1;
210 unsigned int feature_fua:1;
211 unsigned int feature_discard:1;
212 unsigned int feature_secdiscard:1;
213 unsigned int feature_persistent:1;
214 unsigned int discard_granularity;
215 unsigned int discard_alignment;
216 /* Number of 4KB segments handled */
217 unsigned int max_indirect_segments;
219 struct blk_mq_tag_set tag_set;
220 struct blkfront_ring_info *rinfo;
221 unsigned int nr_rings;
222 /* Save uncomplete reqs and bios for migration. */
223 struct list_head requests;
224 struct bio_list bio_list;
227 static unsigned int nr_minors;
228 static unsigned long *minors;
229 static DEFINE_SPINLOCK(minor_lock);
231 #define GRANT_INVALID_REF 0
233 #define PARTS_PER_DISK 16
234 #define PARTS_PER_EXT_DISK 256
236 #define BLKIF_MAJOR(dev) ((dev)>>8)
237 #define BLKIF_MINOR(dev) ((dev) & 0xff)
240 #define EXTENDED (1<<EXT_SHIFT)
241 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
242 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
243 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
244 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
245 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
246 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
248 #define DEV_NAME "xvd" /* name in /dev */
251 * Grants are always the same size as a Xen page (i.e 4KB).
252 * A physical segment is always the same size as a Linux page.
253 * Number of grants per physical segment
255 #define GRANTS_PER_PSEG (PAGE_SIZE / XEN_PAGE_SIZE)
257 #define GRANTS_PER_INDIRECT_FRAME \
258 (XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
260 #define PSEGS_PER_INDIRECT_FRAME \
261 (GRANTS_INDIRECT_FRAME / GRANTS_PSEGS)
263 #define INDIRECT_GREFS(_grants) \
264 DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
266 #define GREFS(_psegs) ((_psegs) * GRANTS_PER_PSEG)
268 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
269 static void blkfront_gather_backend_features(struct blkfront_info *info);
271 static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
273 unsigned long free = rinfo->shadow_free;
275 BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
276 rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
277 rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
281 static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
284 if (rinfo->shadow[id].req.u.rw.id != id)
286 if (rinfo->shadow[id].request == NULL)
288 rinfo->shadow[id].req.u.rw.id = rinfo->shadow_free;
289 rinfo->shadow[id].request = NULL;
290 rinfo->shadow_free = id;
294 static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
296 struct blkfront_info *info = rinfo->dev_info;
297 struct page *granted_page;
298 struct grant *gnt_list_entry, *n;
302 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
306 if (info->feature_persistent) {
307 granted_page = alloc_page(GFP_NOIO);
309 kfree(gnt_list_entry);
312 gnt_list_entry->page = granted_page;
315 gnt_list_entry->gref = GRANT_INVALID_REF;
316 list_add(&gnt_list_entry->node, &rinfo->grants);
323 list_for_each_entry_safe(gnt_list_entry, n,
324 &rinfo->grants, node) {
325 list_del(&gnt_list_entry->node);
326 if (info->feature_persistent)
327 __free_page(gnt_list_entry->page);
328 kfree(gnt_list_entry);
335 static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
337 struct grant *gnt_list_entry;
339 BUG_ON(list_empty(&rinfo->grants));
340 gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
342 list_del(&gnt_list_entry->node);
344 if (gnt_list_entry->gref != GRANT_INVALID_REF)
345 rinfo->persistent_gnts_c--;
347 return gnt_list_entry;
350 static inline void grant_foreign_access(const struct grant *gnt_list_entry,
351 const struct blkfront_info *info)
353 gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
354 info->xbdev->otherend_id,
355 gnt_list_entry->page,
359 static struct grant *get_grant(grant_ref_t *gref_head,
361 struct blkfront_ring_info *rinfo)
363 struct grant *gnt_list_entry = get_free_grant(rinfo);
364 struct blkfront_info *info = rinfo->dev_info;
366 if (gnt_list_entry->gref != GRANT_INVALID_REF)
367 return gnt_list_entry;
369 /* Assign a gref to this page */
370 gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
371 BUG_ON(gnt_list_entry->gref == -ENOSPC);
372 if (info->feature_persistent)
373 grant_foreign_access(gnt_list_entry, info);
375 /* Grant access to the GFN passed by the caller */
376 gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
377 info->xbdev->otherend_id,
381 return gnt_list_entry;
384 static struct grant *get_indirect_grant(grant_ref_t *gref_head,
385 struct blkfront_ring_info *rinfo)
387 struct grant *gnt_list_entry = get_free_grant(rinfo);
388 struct blkfront_info *info = rinfo->dev_info;
390 if (gnt_list_entry->gref != GRANT_INVALID_REF)
391 return gnt_list_entry;
393 /* Assign a gref to this page */
394 gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
395 BUG_ON(gnt_list_entry->gref == -ENOSPC);
396 if (!info->feature_persistent) {
397 struct page *indirect_page;
399 /* Fetch a pre-allocated page to use for indirect grefs */
400 BUG_ON(list_empty(&rinfo->indirect_pages));
401 indirect_page = list_first_entry(&rinfo->indirect_pages,
403 list_del(&indirect_page->lru);
404 gnt_list_entry->page = indirect_page;
406 grant_foreign_access(gnt_list_entry, info);
408 return gnt_list_entry;
411 static const char *op_name(int op)
413 static const char *const names[] = {
414 [BLKIF_OP_READ] = "read",
415 [BLKIF_OP_WRITE] = "write",
416 [BLKIF_OP_WRITE_BARRIER] = "barrier",
417 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
418 [BLKIF_OP_DISCARD] = "discard" };
420 if (op < 0 || op >= ARRAY_SIZE(names))
428 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
430 unsigned int end = minor + nr;
433 if (end > nr_minors) {
434 unsigned long *bitmap, *old;
436 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
441 spin_lock(&minor_lock);
442 if (end > nr_minors) {
444 memcpy(bitmap, minors,
445 BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
447 nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
450 spin_unlock(&minor_lock);
454 spin_lock(&minor_lock);
455 if (find_next_bit(minors, end, minor) >= end) {
456 bitmap_set(minors, minor, nr);
460 spin_unlock(&minor_lock);
465 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
467 unsigned int end = minor + nr;
469 BUG_ON(end > nr_minors);
470 spin_lock(&minor_lock);
471 bitmap_clear(minors, minor, nr);
472 spin_unlock(&minor_lock);
475 static void blkif_restart_queue_callback(void *arg)
477 struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
478 schedule_work(&rinfo->work);
481 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
483 /* We don't have real geometry info, but let's at least return
484 values consistent with the size of the device */
485 sector_t nsect = get_capacity(bd->bd_disk);
486 sector_t cylinders = nsect;
490 sector_div(cylinders, hg->heads * hg->sectors);
491 hg->cylinders = cylinders;
492 if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
493 hg->cylinders = 0xffff;
497 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
498 unsigned command, unsigned long argument)
500 struct blkfront_info *info = bdev->bd_disk->private_data;
503 dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
504 command, (long)argument);
507 case CDROMMULTISESSION:
508 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
509 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
510 if (put_user(0, (char __user *)(argument + i)))
514 case CDROM_GET_CAPABILITY: {
515 struct gendisk *gd = info->gd;
516 if (gd->flags & GENHD_FL_CD)
522 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
524 return -EINVAL; /* same return as native Linux */
530 static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
532 struct blkif_request **ring_req)
536 *ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
537 rinfo->ring.req_prod_pvt++;
539 id = get_id_from_freelist(rinfo);
540 rinfo->shadow[id].request = req;
541 rinfo->shadow[id].status = REQ_WAITING;
542 rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
544 (*ring_req)->u.rw.id = id;
549 static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
551 struct blkfront_info *info = rinfo->dev_info;
552 struct blkif_request *ring_req;
555 /* Fill out a communications ring structure. */
556 id = blkif_ring_get_request(rinfo, req, &ring_req);
558 ring_req->operation = BLKIF_OP_DISCARD;
559 ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
560 ring_req->u.discard.id = id;
561 ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
562 if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard)
563 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
565 ring_req->u.discard.flag = 0;
567 /* Keep a private copy so we can reissue requests when recovering. */
568 rinfo->shadow[id].req = *ring_req;
573 struct setup_rw_req {
574 unsigned int grant_idx;
575 struct blkif_request_segment *segments;
576 struct blkfront_ring_info *rinfo;
577 struct blkif_request *ring_req;
578 grant_ref_t gref_head;
580 /* Only used when persistent grant is used and it's a read request */
582 unsigned int bvec_off;
585 bool require_extra_req;
586 struct blkif_request *extra_ring_req;
589 static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
590 unsigned int len, void *data)
592 struct setup_rw_req *setup = data;
594 struct grant *gnt_list_entry;
595 unsigned int fsect, lsect;
596 /* Convenient aliases */
597 unsigned int grant_idx = setup->grant_idx;
598 struct blkif_request *ring_req = setup->ring_req;
599 struct blkfront_ring_info *rinfo = setup->rinfo;
601 * We always use the shadow of the first request to store the list
602 * of grant associated to the block I/O request. This made the
603 * completion more easy to handle even if the block I/O request is
606 struct blk_shadow *shadow = &rinfo->shadow[setup->id];
608 if (unlikely(setup->require_extra_req &&
609 grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
611 * We are using the second request, setup grant_idx
612 * to be the index of the segment array.
614 grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
615 ring_req = setup->extra_ring_req;
618 if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
619 (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
621 kunmap_atomic(setup->segments);
623 n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
624 gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
625 shadow->indirect_grants[n] = gnt_list_entry;
626 setup->segments = kmap_atomic(gnt_list_entry->page);
627 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
630 gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
631 ref = gnt_list_entry->gref;
633 * All the grants are stored in the shadow of the first
634 * request. Therefore we have to use the global index.
636 shadow->grants_used[setup->grant_idx] = gnt_list_entry;
638 if (setup->need_copy) {
641 shared_data = kmap_atomic(gnt_list_entry->page);
643 * this does not wipe data stored outside the
644 * range sg->offset..sg->offset+sg->length.
645 * Therefore, blkback *could* see data from
646 * previous requests. This is OK as long as
647 * persistent grants are shared with just one
648 * domain. It may need refactoring if this
651 memcpy(shared_data + offset,
652 setup->bvec_data + setup->bvec_off,
655 kunmap_atomic(shared_data);
656 setup->bvec_off += len;
660 lsect = fsect + (len >> 9) - 1;
661 if (ring_req->operation != BLKIF_OP_INDIRECT) {
662 ring_req->u.rw.seg[grant_idx] =
663 (struct blkif_request_segment) {
666 .last_sect = lsect };
668 setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
669 (struct blkif_request_segment) {
672 .last_sect = lsect };
675 (setup->grant_idx)++;
678 static void blkif_setup_extra_req(struct blkif_request *first,
679 struct blkif_request *second)
681 uint16_t nr_segments = first->u.rw.nr_segments;
684 * The second request is only present when the first request uses
685 * all its segments. It's always the continuity of the first one.
687 first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
689 second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
690 second->u.rw.sector_number = first->u.rw.sector_number +
691 (BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
693 second->u.rw.handle = first->u.rw.handle;
694 second->operation = first->operation;
697 static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
699 struct blkfront_info *info = rinfo->dev_info;
700 struct blkif_request *ring_req, *extra_ring_req = NULL;
701 unsigned long id, extra_id = NO_ASSOCIATED_ID;
702 bool require_extra_req = false;
704 struct setup_rw_req setup = {
708 .need_copy = rq_data_dir(req) && info->feature_persistent,
712 * Used to store if we are able to queue the request by just using
713 * existing persistent grants, or if we have to get new grants,
714 * as there are not sufficiently many free.
716 struct scatterlist *sg;
717 int num_sg, max_grefs, num_grant;
719 max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
720 if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
722 * If we are using indirect segments we need to account
723 * for the indirect grefs used in the request.
725 max_grefs += INDIRECT_GREFS(max_grefs);
728 * We have to reserve 'max_grefs' grants because persistent
729 * grants are shared by all rings.
732 if (gnttab_alloc_grant_references(max_grefs, &setup.gref_head) < 0) {
733 gnttab_request_free_callback(
735 blkif_restart_queue_callback,
741 /* Fill out a communications ring structure. */
742 id = blkif_ring_get_request(rinfo, req, &ring_req);
744 num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
746 /* Calculate the number of grant used */
747 for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
748 num_grant += gnttab_count_grant(sg->offset, sg->length);
750 require_extra_req = info->max_indirect_segments == 0 &&
751 num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
752 BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
754 rinfo->shadow[id].num_sg = num_sg;
755 if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
756 likely(!require_extra_req)) {
758 * The indirect operation can only be a BLKIF_OP_READ or
761 BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA);
762 ring_req->operation = BLKIF_OP_INDIRECT;
763 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
764 BLKIF_OP_WRITE : BLKIF_OP_READ;
765 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
766 ring_req->u.indirect.handle = info->handle;
767 ring_req->u.indirect.nr_segments = num_grant;
769 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
770 ring_req->u.rw.handle = info->handle;
771 ring_req->operation = rq_data_dir(req) ?
772 BLKIF_OP_WRITE : BLKIF_OP_READ;
773 if (req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA) {
775 * Ideally we can do an unordered flush-to-disk.
776 * In case the backend onlysupports barriers, use that.
777 * A barrier request a superset of FUA, so we can
778 * implement it the same way. (It's also a FLUSH+FUA,
779 * since it is guaranteed ordered WRT previous writes.)
781 if (info->feature_flush && info->feature_fua)
782 ring_req->operation =
783 BLKIF_OP_WRITE_BARRIER;
784 else if (info->feature_flush)
785 ring_req->operation =
786 BLKIF_OP_FLUSH_DISKCACHE;
788 ring_req->operation = 0;
790 ring_req->u.rw.nr_segments = num_grant;
791 if (unlikely(require_extra_req)) {
792 extra_id = blkif_ring_get_request(rinfo, req,
795 * Only the first request contains the scatter-gather
798 rinfo->shadow[extra_id].num_sg = 0;
800 blkif_setup_extra_req(ring_req, extra_ring_req);
802 /* Link the 2 requests together */
803 rinfo->shadow[extra_id].associated_id = id;
804 rinfo->shadow[id].associated_id = extra_id;
808 setup.ring_req = ring_req;
811 setup.require_extra_req = require_extra_req;
812 if (unlikely(require_extra_req))
813 setup.extra_ring_req = extra_ring_req;
815 for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
816 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
818 if (setup.need_copy) {
819 setup.bvec_off = sg->offset;
820 setup.bvec_data = kmap_atomic(sg_page(sg));
823 gnttab_foreach_grant_in_range(sg_page(sg),
826 blkif_setup_rw_req_grant,
830 kunmap_atomic(setup.bvec_data);
833 kunmap_atomic(setup.segments);
835 /* Keep a private copy so we can reissue requests when recovering. */
836 rinfo->shadow[id].req = *ring_req;
837 if (unlikely(require_extra_req))
838 rinfo->shadow[extra_id].req = *extra_ring_req;
841 gnttab_free_grant_references(setup.gref_head);
847 * Generate a Xen blkfront IO request from a blk layer request. Reads
848 * and writes are handled as expected.
850 * @req: a request struct
852 static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
854 if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
857 if (unlikely(req_op(req) == REQ_OP_DISCARD ||
858 req_op(req) == REQ_OP_SECURE_ERASE))
859 return blkif_queue_discard_req(req, rinfo);
861 return blkif_queue_rw_req(req, rinfo);
864 static inline void flush_requests(struct blkfront_ring_info *rinfo)
868 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
871 notify_remote_via_irq(rinfo->irq);
874 static inline bool blkif_request_flush_invalid(struct request *req,
875 struct blkfront_info *info)
877 return (blk_rq_is_passthrough(req) ||
878 ((req_op(req) == REQ_OP_FLUSH) &&
879 !info->feature_flush) ||
880 ((req->cmd_flags & REQ_FUA) &&
881 !info->feature_fua));
884 static int blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
885 const struct blk_mq_queue_data *qd)
888 int qid = hctx->queue_num;
889 struct blkfront_info *info = hctx->queue->queuedata;
890 struct blkfront_ring_info *rinfo = NULL;
892 BUG_ON(info->nr_rings <= qid);
893 rinfo = &info->rinfo[qid];
894 blk_mq_start_request(qd->rq);
895 spin_lock_irqsave(&rinfo->ring_lock, flags);
896 if (RING_FULL(&rinfo->ring))
899 if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
902 if (blkif_queue_request(qd->rq, rinfo))
905 flush_requests(rinfo);
906 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
907 return BLK_MQ_RQ_QUEUE_OK;
910 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
911 return BLK_MQ_RQ_QUEUE_ERROR;
914 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
915 blk_mq_stop_hw_queue(hctx);
916 return BLK_MQ_RQ_QUEUE_BUSY;
919 static void blkif_complete_rq(struct request *rq)
921 blk_mq_end_request(rq, blkif_req(rq)->error);
924 static const struct blk_mq_ops blkfront_mq_ops = {
925 .queue_rq = blkif_queue_rq,
926 .complete = blkif_complete_rq,
929 static void blkif_set_queue_limits(struct blkfront_info *info)
931 struct request_queue *rq = info->rq;
932 struct gendisk *gd = info->gd;
933 unsigned int segments = info->max_indirect_segments ? :
934 BLKIF_MAX_SEGMENTS_PER_REQUEST;
936 queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
938 if (info->feature_discard) {
939 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
940 blk_queue_max_discard_sectors(rq, get_capacity(gd));
941 rq->limits.discard_granularity = info->discard_granularity;
942 rq->limits.discard_alignment = info->discard_alignment;
943 if (info->feature_secdiscard)
944 queue_flag_set_unlocked(QUEUE_FLAG_SECERASE, rq);
947 /* Hard sector size and max sectors impersonate the equiv. hardware. */
948 blk_queue_logical_block_size(rq, info->sector_size);
949 blk_queue_physical_block_size(rq, info->physical_sector_size);
950 blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
952 /* Each segment in a request is up to an aligned page in size. */
953 blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
954 blk_queue_max_segment_size(rq, PAGE_SIZE);
956 /* Ensure a merged request will fit in a single I/O ring slot. */
957 blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
959 /* Make sure buffer addresses are sector-aligned. */
960 blk_queue_dma_alignment(rq, 511);
962 /* Make sure we don't use bounce buffers. */
963 blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
966 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
967 unsigned int physical_sector_size)
969 struct request_queue *rq;
970 struct blkfront_info *info = gd->private_data;
972 memset(&info->tag_set, 0, sizeof(info->tag_set));
973 info->tag_set.ops = &blkfront_mq_ops;
974 info->tag_set.nr_hw_queues = info->nr_rings;
975 if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
977 * When indirect descriptior is not supported, the I/O request
978 * will be split between multiple request in the ring.
979 * To avoid problems when sending the request, divide by
980 * 2 the depth of the queue.
982 info->tag_set.queue_depth = BLK_RING_SIZE(info) / 2;
984 info->tag_set.queue_depth = BLK_RING_SIZE(info);
985 info->tag_set.numa_node = NUMA_NO_NODE;
986 info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
987 info->tag_set.cmd_size = sizeof(struct blkif_req);
988 info->tag_set.driver_data = info;
990 if (blk_mq_alloc_tag_set(&info->tag_set))
992 rq = blk_mq_init_queue(&info->tag_set);
994 blk_mq_free_tag_set(&info->tag_set);
998 rq->queuedata = info;
999 info->rq = gd->queue = rq;
1001 info->sector_size = sector_size;
1002 info->physical_sector_size = physical_sector_size;
1003 blkif_set_queue_limits(info);
1008 static const char *flush_info(struct blkfront_info *info)
1010 if (info->feature_flush && info->feature_fua)
1011 return "barrier: enabled;";
1012 else if (info->feature_flush)
1013 return "flush diskcache: enabled;";
1015 return "barrier or flush: disabled;";
1018 static void xlvbd_flush(struct blkfront_info *info)
1020 blk_queue_write_cache(info->rq, info->feature_flush ? true : false,
1021 info->feature_fua ? true : false);
1022 pr_info("blkfront: %s: %s %s %s %s %s\n",
1023 info->gd->disk_name, flush_info(info),
1024 "persistent grants:", info->feature_persistent ?
1025 "enabled;" : "disabled;", "indirect descriptors:",
1026 info->max_indirect_segments ? "enabled;" : "disabled;");
1029 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
1032 major = BLKIF_MAJOR(vdevice);
1033 *minor = BLKIF_MINOR(vdevice);
1035 case XEN_IDE0_MAJOR:
1036 *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
1037 *minor = ((*minor / 64) * PARTS_PER_DISK) +
1038 EMULATED_HD_DISK_MINOR_OFFSET;
1040 case XEN_IDE1_MAJOR:
1041 *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1042 *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1043 EMULATED_HD_DISK_MINOR_OFFSET;
1045 case XEN_SCSI_DISK0_MAJOR:
1046 *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1047 *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1049 case XEN_SCSI_DISK1_MAJOR:
1050 case XEN_SCSI_DISK2_MAJOR:
1051 case XEN_SCSI_DISK3_MAJOR:
1052 case XEN_SCSI_DISK4_MAJOR:
1053 case XEN_SCSI_DISK5_MAJOR:
1054 case XEN_SCSI_DISK6_MAJOR:
1055 case XEN_SCSI_DISK7_MAJOR:
1056 *offset = (*minor / PARTS_PER_DISK) +
1057 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1058 EMULATED_SD_DISK_NAME_OFFSET;
1060 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1061 EMULATED_SD_DISK_MINOR_OFFSET;
1063 case XEN_SCSI_DISK8_MAJOR:
1064 case XEN_SCSI_DISK9_MAJOR:
1065 case XEN_SCSI_DISK10_MAJOR:
1066 case XEN_SCSI_DISK11_MAJOR:
1067 case XEN_SCSI_DISK12_MAJOR:
1068 case XEN_SCSI_DISK13_MAJOR:
1069 case XEN_SCSI_DISK14_MAJOR:
1070 case XEN_SCSI_DISK15_MAJOR:
1071 *offset = (*minor / PARTS_PER_DISK) +
1072 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1073 EMULATED_SD_DISK_NAME_OFFSET;
1075 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1076 EMULATED_SD_DISK_MINOR_OFFSET;
1079 *offset = *minor / PARTS_PER_DISK;
1082 printk(KERN_WARNING "blkfront: your disk configuration is "
1083 "incorrect, please use an xvd device instead\n");
1089 static char *encode_disk_name(char *ptr, unsigned int n)
1092 ptr = encode_disk_name(ptr, n / 26 - 1);
1093 *ptr = 'a' + n % 26;
1097 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1098 struct blkfront_info *info,
1099 u16 vdisk_info, u16 sector_size,
1100 unsigned int physical_sector_size)
1105 unsigned int offset;
1110 BUG_ON(info->gd != NULL);
1111 BUG_ON(info->rq != NULL);
1113 if ((info->vdevice>>EXT_SHIFT) > 1) {
1114 /* this is above the extended range; something is wrong */
1115 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1119 if (!VDEV_IS_EXTENDED(info->vdevice)) {
1120 err = xen_translate_vdev(info->vdevice, &minor, &offset);
1123 nr_parts = PARTS_PER_DISK;
1125 minor = BLKIF_MINOR_EXT(info->vdevice);
1126 nr_parts = PARTS_PER_EXT_DISK;
1127 offset = minor / nr_parts;
1128 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1129 printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1130 "emulated IDE disks,\n\t choose an xvd device name"
1131 "from xvde on\n", info->vdevice);
1133 if (minor >> MINORBITS) {
1134 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1135 info->vdevice, minor);
1139 if ((minor % nr_parts) == 0)
1140 nr_minors = nr_parts;
1142 err = xlbd_reserve_minors(minor, nr_minors);
1147 gd = alloc_disk(nr_minors);
1151 strcpy(gd->disk_name, DEV_NAME);
1152 ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1153 BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1157 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1158 "%d", minor & (nr_parts - 1));
1160 gd->major = XENVBD_MAJOR;
1161 gd->first_minor = minor;
1162 gd->fops = &xlvbd_block_fops;
1163 gd->private_data = info;
1164 set_capacity(gd, capacity);
1166 if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size)) {
1173 if (vdisk_info & VDISK_READONLY)
1176 if (vdisk_info & VDISK_REMOVABLE)
1177 gd->flags |= GENHD_FL_REMOVABLE;
1179 if (vdisk_info & VDISK_CDROM)
1180 gd->flags |= GENHD_FL_CD;
1185 xlbd_release_minors(minor, nr_minors);
1190 static void xlvbd_release_gendisk(struct blkfront_info *info)
1192 unsigned int minor, nr_minors, i;
1194 if (info->rq == NULL)
1197 /* No more blkif_request(). */
1198 blk_mq_stop_hw_queues(info->rq);
1200 for (i = 0; i < info->nr_rings; i++) {
1201 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1203 /* No more gnttab callback work. */
1204 gnttab_cancel_free_callback(&rinfo->callback);
1206 /* Flush gnttab callback work. Must be done with no locks held. */
1207 flush_work(&rinfo->work);
1210 del_gendisk(info->gd);
1212 minor = info->gd->first_minor;
1213 nr_minors = info->gd->minors;
1214 xlbd_release_minors(minor, nr_minors);
1216 blk_cleanup_queue(info->rq);
1217 blk_mq_free_tag_set(&info->tag_set);
1224 /* Already hold rinfo->ring_lock. */
1225 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1227 if (!RING_FULL(&rinfo->ring))
1228 blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1231 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1233 unsigned long flags;
1235 spin_lock_irqsave(&rinfo->ring_lock, flags);
1236 kick_pending_request_queues_locked(rinfo);
1237 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1240 static void blkif_restart_queue(struct work_struct *work)
1242 struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1244 if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1245 kick_pending_request_queues(rinfo);
1248 static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1250 struct grant *persistent_gnt, *n;
1251 struct blkfront_info *info = rinfo->dev_info;
1255 * Remove indirect pages, this only happens when using indirect
1256 * descriptors but not persistent grants
1258 if (!list_empty(&rinfo->indirect_pages)) {
1259 struct page *indirect_page, *n;
1261 BUG_ON(info->feature_persistent);
1262 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1263 list_del(&indirect_page->lru);
1264 __free_page(indirect_page);
1268 /* Remove all persistent grants. */
1269 if (!list_empty(&rinfo->grants)) {
1270 list_for_each_entry_safe(persistent_gnt, n,
1271 &rinfo->grants, node) {
1272 list_del(&persistent_gnt->node);
1273 if (persistent_gnt->gref != GRANT_INVALID_REF) {
1274 gnttab_end_foreign_access(persistent_gnt->gref,
1276 rinfo->persistent_gnts_c--;
1278 if (info->feature_persistent)
1279 __free_page(persistent_gnt->page);
1280 kfree(persistent_gnt);
1283 BUG_ON(rinfo->persistent_gnts_c != 0);
1285 for (i = 0; i < BLK_RING_SIZE(info); i++) {
1287 * Clear persistent grants present in requests already
1288 * on the shared ring
1290 if (!rinfo->shadow[i].request)
1293 segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1294 rinfo->shadow[i].req.u.indirect.nr_segments :
1295 rinfo->shadow[i].req.u.rw.nr_segments;
1296 for (j = 0; j < segs; j++) {
1297 persistent_gnt = rinfo->shadow[i].grants_used[j];
1298 gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1299 if (info->feature_persistent)
1300 __free_page(persistent_gnt->page);
1301 kfree(persistent_gnt);
1304 if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1306 * If this is not an indirect operation don't try to
1307 * free indirect segments
1311 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1312 persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1313 gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1314 __free_page(persistent_gnt->page);
1315 kfree(persistent_gnt);
1319 kfree(rinfo->shadow[i].grants_used);
1320 rinfo->shadow[i].grants_used = NULL;
1321 kfree(rinfo->shadow[i].indirect_grants);
1322 rinfo->shadow[i].indirect_grants = NULL;
1323 kfree(rinfo->shadow[i].sg);
1324 rinfo->shadow[i].sg = NULL;
1327 /* No more gnttab callback work. */
1328 gnttab_cancel_free_callback(&rinfo->callback);
1330 /* Flush gnttab callback work. Must be done with no locks held. */
1331 flush_work(&rinfo->work);
1333 /* Free resources associated with old device channel. */
1334 for (i = 0; i < info->nr_ring_pages; i++) {
1335 if (rinfo->ring_ref[i] != GRANT_INVALID_REF) {
1336 gnttab_end_foreign_access(rinfo->ring_ref[i], 0, 0);
1337 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1340 free_pages((unsigned long)rinfo->ring.sring, get_order(info->nr_ring_pages * XEN_PAGE_SIZE));
1341 rinfo->ring.sring = NULL;
1344 unbind_from_irqhandler(rinfo->irq, rinfo);
1345 rinfo->evtchn = rinfo->irq = 0;
1348 static void blkif_free(struct blkfront_info *info, int suspend)
1352 /* Prevent new requests being issued until we fix things up. */
1353 info->connected = suspend ?
1354 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1355 /* No more blkif_request(). */
1357 blk_mq_stop_hw_queues(info->rq);
1359 for (i = 0; i < info->nr_rings; i++)
1360 blkif_free_ring(&info->rinfo[i]);
1367 struct copy_from_grant {
1368 const struct blk_shadow *s;
1369 unsigned int grant_idx;
1370 unsigned int bvec_offset;
1374 static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1375 unsigned int len, void *data)
1377 struct copy_from_grant *info = data;
1379 /* Convenient aliases */
1380 const struct blk_shadow *s = info->s;
1382 shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1384 memcpy(info->bvec_data + info->bvec_offset,
1385 shared_data + offset, len);
1387 info->bvec_offset += len;
1390 kunmap_atomic(shared_data);
1393 static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1397 case BLKIF_RSP_OKAY:
1399 case BLKIF_RSP_EOPNOTSUPP:
1400 return REQ_EOPNOTSUPP;
1401 case BLKIF_RSP_ERROR:
1409 * Get the final status of the block request based on two ring response
1411 static int blkif_get_final_status(enum blk_req_status s1,
1412 enum blk_req_status s2)
1414 BUG_ON(s1 == REQ_WAITING);
1415 BUG_ON(s2 == REQ_WAITING);
1417 if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1418 return BLKIF_RSP_ERROR;
1419 else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1420 return BLKIF_RSP_EOPNOTSUPP;
1421 return BLKIF_RSP_OKAY;
1424 static bool blkif_completion(unsigned long *id,
1425 struct blkfront_ring_info *rinfo,
1426 struct blkif_response *bret)
1429 struct scatterlist *sg;
1430 int num_sg, num_grant;
1431 struct blkfront_info *info = rinfo->dev_info;
1432 struct blk_shadow *s = &rinfo->shadow[*id];
1433 struct copy_from_grant data = {
1437 num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1438 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1440 /* The I/O request may be split in two. */
1441 if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1442 struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1444 /* Keep the status of the current response in shadow. */
1445 s->status = blkif_rsp_to_req_status(bret->status);
1447 /* Wait the second response if not yet here. */
1448 if (s2->status == REQ_WAITING)
1451 bret->status = blkif_get_final_status(s->status,
1455 * All the grants is stored in the first shadow in order
1456 * to make the completion code simpler.
1458 num_grant += s2->req.u.rw.nr_segments;
1461 * The two responses may not come in order. Only the
1462 * first request will store the scatter-gather list.
1464 if (s2->num_sg != 0) {
1465 /* Update "id" with the ID of the first response. */
1466 *id = s->associated_id;
1471 * We don't need anymore the second request, so recycling
1474 if (add_id_to_freelist(rinfo, s->associated_id))
1475 WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1476 info->gd->disk_name, s->associated_id);
1482 if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1483 for_each_sg(s->sg, sg, num_sg, i) {
1484 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1486 data.bvec_offset = sg->offset;
1487 data.bvec_data = kmap_atomic(sg_page(sg));
1489 gnttab_foreach_grant_in_range(sg_page(sg),
1492 blkif_copy_from_grant,
1495 kunmap_atomic(data.bvec_data);
1498 /* Add the persistent grant into the list of free grants */
1499 for (i = 0; i < num_grant; i++) {
1500 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1502 * If the grant is still mapped by the backend (the
1503 * backend has chosen to make this grant persistent)
1504 * we add it at the head of the list, so it will be
1507 if (!info->feature_persistent)
1508 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1509 s->grants_used[i]->gref);
1510 list_add(&s->grants_used[i]->node, &rinfo->grants);
1511 rinfo->persistent_gnts_c++;
1514 * If the grant is not mapped by the backend we end the
1515 * foreign access and add it to the tail of the list,
1516 * so it will not be picked again unless we run out of
1517 * persistent grants.
1519 gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1520 s->grants_used[i]->gref = GRANT_INVALID_REF;
1521 list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1524 if (s->req.operation == BLKIF_OP_INDIRECT) {
1525 for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1526 if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1527 if (!info->feature_persistent)
1528 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1529 s->indirect_grants[i]->gref);
1530 list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1531 rinfo->persistent_gnts_c++;
1533 struct page *indirect_page;
1535 gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1537 * Add the used indirect page back to the list of
1538 * available pages for indirect grefs.
1540 if (!info->feature_persistent) {
1541 indirect_page = s->indirect_grants[i]->page;
1542 list_add(&indirect_page->lru, &rinfo->indirect_pages);
1544 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1545 list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1553 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1555 struct request *req;
1556 struct blkif_response *bret;
1558 unsigned long flags;
1559 struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1560 struct blkfront_info *info = rinfo->dev_info;
1562 if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
1565 spin_lock_irqsave(&rinfo->ring_lock, flags);
1567 rp = rinfo->ring.sring->rsp_prod;
1568 rmb(); /* Ensure we see queued responses up to 'rp'. */
1570 for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1573 bret = RING_GET_RESPONSE(&rinfo->ring, i);
1576 * The backend has messed up and given us an id that we would
1577 * never have given to it (we stamp it up to BLK_RING_SIZE -
1578 * look in get_id_from_freelist.
1580 if (id >= BLK_RING_SIZE(info)) {
1581 WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1582 info->gd->disk_name, op_name(bret->operation), id);
1583 /* We can't safely get the 'struct request' as
1584 * the id is busted. */
1587 req = rinfo->shadow[id].request;
1589 if (bret->operation != BLKIF_OP_DISCARD) {
1591 * We may need to wait for an extra response if the
1592 * I/O request is split in 2
1594 if (!blkif_completion(&id, rinfo, bret))
1598 if (add_id_to_freelist(rinfo, id)) {
1599 WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1600 info->gd->disk_name, op_name(bret->operation), id);
1604 if (bret->status == BLKIF_RSP_OKAY)
1605 blkif_req(req)->error = BLK_STS_OK;
1607 blkif_req(req)->error = BLK_STS_IOERR;
1609 switch (bret->operation) {
1610 case BLKIF_OP_DISCARD:
1611 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1612 struct request_queue *rq = info->rq;
1613 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1614 info->gd->disk_name, op_name(bret->operation));
1615 blkif_req(req)->error = BLK_STS_NOTSUPP;
1616 info->feature_discard = 0;
1617 info->feature_secdiscard = 0;
1618 queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1619 queue_flag_clear(QUEUE_FLAG_SECERASE, rq);
1622 case BLKIF_OP_FLUSH_DISKCACHE:
1623 case BLKIF_OP_WRITE_BARRIER:
1624 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1625 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1626 info->gd->disk_name, op_name(bret->operation));
1627 blkif_req(req)->error = -EOPNOTSUPP;
1629 if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1630 rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1631 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1632 info->gd->disk_name, op_name(bret->operation));
1633 blkif_req(req)->error = BLK_STS_NOTSUPP;
1635 if (unlikely(blkif_req(req)->error)) {
1636 if (blkif_req(req)->error == BLK_STS_NOTSUPP)
1637 blkif_req(req)->error = BLK_STS_OK;
1638 info->feature_fua = 0;
1639 info->feature_flush = 0;
1644 case BLKIF_OP_WRITE:
1645 if (unlikely(bret->status != BLKIF_RSP_OKAY))
1646 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1647 "request: %x\n", bret->status);
1654 blk_mq_complete_request(req);
1657 rinfo->ring.rsp_cons = i;
1659 if (i != rinfo->ring.req_prod_pvt) {
1661 RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1665 rinfo->ring.sring->rsp_event = i + 1;
1667 kick_pending_request_queues_locked(rinfo);
1669 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1675 static int setup_blkring(struct xenbus_device *dev,
1676 struct blkfront_ring_info *rinfo)
1678 struct blkif_sring *sring;
1680 struct blkfront_info *info = rinfo->dev_info;
1681 unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1682 grant_ref_t gref[XENBUS_MAX_RING_GRANTS];
1684 for (i = 0; i < info->nr_ring_pages; i++)
1685 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1687 sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
1688 get_order(ring_size));
1690 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1693 SHARED_RING_INIT(sring);
1694 FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1696 err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref);
1698 free_pages((unsigned long)sring, get_order(ring_size));
1699 rinfo->ring.sring = NULL;
1702 for (i = 0; i < info->nr_ring_pages; i++)
1703 rinfo->ring_ref[i] = gref[i];
1705 err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1709 err = bind_evtchn_to_irqhandler(rinfo->evtchn, blkif_interrupt, 0,
1712 xenbus_dev_fatal(dev, err,
1713 "bind_evtchn_to_irqhandler failed");
1720 blkif_free(info, 0);
1725 * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1726 * ring buffer may have multi pages depending on ->nr_ring_pages.
1728 static int write_per_ring_nodes(struct xenbus_transaction xbt,
1729 struct blkfront_ring_info *rinfo, const char *dir)
1733 const char *message = NULL;
1734 struct blkfront_info *info = rinfo->dev_info;
1736 if (info->nr_ring_pages == 1) {
1737 err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1739 message = "writing ring-ref";
1740 goto abort_transaction;
1743 for (i = 0; i < info->nr_ring_pages; i++) {
1744 char ring_ref_name[RINGREF_NAME_LEN];
1746 snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1747 err = xenbus_printf(xbt, dir, ring_ref_name,
1748 "%u", rinfo->ring_ref[i]);
1750 message = "writing ring-ref";
1751 goto abort_transaction;
1756 err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1758 message = "writing event-channel";
1759 goto abort_transaction;
1765 xenbus_transaction_end(xbt, 1);
1767 xenbus_dev_fatal(info->xbdev, err, "%s", message);
1772 /* Common code used when first setting up, and when resuming. */
1773 static int talk_to_blkback(struct xenbus_device *dev,
1774 struct blkfront_info *info)
1776 const char *message = NULL;
1777 struct xenbus_transaction xbt;
1779 unsigned int i, max_page_order;
1780 unsigned int ring_page_order;
1782 max_page_order = xenbus_read_unsigned(info->xbdev->otherend,
1783 "max-ring-page-order", 0);
1784 ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1785 info->nr_ring_pages = 1 << ring_page_order;
1787 for (i = 0; i < info->nr_rings; i++) {
1788 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1790 /* Create shared ring, alloc event channel. */
1791 err = setup_blkring(dev, rinfo);
1793 goto destroy_blkring;
1797 err = xenbus_transaction_start(&xbt);
1799 xenbus_dev_fatal(dev, err, "starting transaction");
1800 goto destroy_blkring;
1803 if (info->nr_ring_pages > 1) {
1804 err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1807 message = "writing ring-page-order";
1808 goto abort_transaction;
1812 /* We already got the number of queues/rings in _probe */
1813 if (info->nr_rings == 1) {
1814 err = write_per_ring_nodes(xbt, &info->rinfo[0], dev->nodename);
1816 goto destroy_blkring;
1821 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1824 message = "writing multi-queue-num-queues";
1825 goto abort_transaction;
1828 pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1829 path = kmalloc(pathsize, GFP_KERNEL);
1832 message = "ENOMEM while writing ring references";
1833 goto abort_transaction;
1836 for (i = 0; i < info->nr_rings; i++) {
1837 memset(path, 0, pathsize);
1838 snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1839 err = write_per_ring_nodes(xbt, &info->rinfo[i], path);
1842 goto destroy_blkring;
1847 err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1848 XEN_IO_PROTO_ABI_NATIVE);
1850 message = "writing protocol";
1851 goto abort_transaction;
1853 err = xenbus_printf(xbt, dev->nodename,
1854 "feature-persistent", "%u", 1);
1857 "writing persistent grants feature to xenbus");
1859 err = xenbus_transaction_end(xbt, 0);
1863 xenbus_dev_fatal(dev, err, "completing transaction");
1864 goto destroy_blkring;
1867 for (i = 0; i < info->nr_rings; i++) {
1869 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1871 for (j = 0; j < BLK_RING_SIZE(info); j++)
1872 rinfo->shadow[j].req.u.rw.id = j + 1;
1873 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1875 xenbus_switch_state(dev, XenbusStateInitialised);
1880 xenbus_transaction_end(xbt, 1);
1882 xenbus_dev_fatal(dev, err, "%s", message);
1884 blkif_free(info, 0);
1887 dev_set_drvdata(&dev->dev, NULL);
1892 static int negotiate_mq(struct blkfront_info *info)
1894 unsigned int backend_max_queues;
1897 BUG_ON(info->nr_rings);
1899 /* Check if backend supports multiple queues. */
1900 backend_max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1901 "multi-queue-max-queues", 1);
1902 info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1903 /* We need at least one ring. */
1904 if (!info->nr_rings)
1907 info->rinfo = kzalloc(sizeof(struct blkfront_ring_info) * info->nr_rings, GFP_KERNEL);
1909 xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1913 for (i = 0; i < info->nr_rings; i++) {
1914 struct blkfront_ring_info *rinfo;
1916 rinfo = &info->rinfo[i];
1917 INIT_LIST_HEAD(&rinfo->indirect_pages);
1918 INIT_LIST_HEAD(&rinfo->grants);
1919 rinfo->dev_info = info;
1920 INIT_WORK(&rinfo->work, blkif_restart_queue);
1921 spin_lock_init(&rinfo->ring_lock);
1926 * Entry point to this code when a new device is created. Allocate the basic
1927 * structures and the ring buffer for communication with the backend, and
1928 * inform the backend of the appropriate details for those. Switch to
1929 * Initialised state.
1931 static int blkfront_probe(struct xenbus_device *dev,
1932 const struct xenbus_device_id *id)
1935 struct blkfront_info *info;
1937 /* FIXME: Use dynamic device id if this is not set. */
1938 err = xenbus_scanf(XBT_NIL, dev->nodename,
1939 "virtual-device", "%i", &vdevice);
1941 /* go looking in the extended area instead */
1942 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1945 xenbus_dev_fatal(dev, err, "reading virtual-device");
1950 if (xen_hvm_domain()) {
1953 /* no unplug has been done: do not hook devices != xen vbds */
1954 if (xen_has_pv_and_legacy_disk_devices()) {
1957 if (!VDEV_IS_EXTENDED(vdevice))
1958 major = BLKIF_MAJOR(vdevice);
1960 major = XENVBD_MAJOR;
1962 if (major != XENVBD_MAJOR) {
1964 "%s: HVM does not support vbd %d as xen block device\n",
1969 /* do not create a PV cdrom device if we are an HVM guest */
1970 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1973 if (strncmp(type, "cdrom", 5) == 0) {
1979 info = kzalloc(sizeof(*info), GFP_KERNEL);
1981 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1986 err = negotiate_mq(info);
1992 mutex_init(&info->mutex);
1993 info->vdevice = vdevice;
1994 info->connected = BLKIF_STATE_DISCONNECTED;
1996 /* Front end dir is a number, which is used as the id. */
1997 info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1998 dev_set_drvdata(&dev->dev, info);
2003 static void split_bio_end(struct bio *bio)
2005 struct split_bio *split_bio = bio->bi_private;
2007 if (atomic_dec_and_test(&split_bio->pending)) {
2008 split_bio->bio->bi_phys_segments = 0;
2009 split_bio->bio->bi_error = bio->bi_error;
2010 bio_endio(split_bio->bio);
2016 static int blkif_recover(struct blkfront_info *info)
2018 unsigned int i, r_index;
2019 struct request *req, *n;
2021 struct bio *bio, *cloned_bio;
2022 unsigned int segs, offset;
2024 struct split_bio *split_bio;
2026 blkfront_gather_backend_features(info);
2027 /* Reset limits changed by blk_mq_update_nr_hw_queues(). */
2028 blkif_set_queue_limits(info);
2029 segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
2030 blk_queue_max_segments(info->rq, segs / GRANTS_PER_PSEG);
2032 for (r_index = 0; r_index < info->nr_rings; r_index++) {
2033 struct blkfront_ring_info *rinfo = &info->rinfo[r_index];
2035 rc = blkfront_setup_indirect(rinfo);
2039 xenbus_switch_state(info->xbdev, XenbusStateConnected);
2041 /* Now safe for us to use the shared ring */
2042 info->connected = BLKIF_STATE_CONNECTED;
2044 for (r_index = 0; r_index < info->nr_rings; r_index++) {
2045 struct blkfront_ring_info *rinfo;
2047 rinfo = &info->rinfo[r_index];
2048 /* Kick any other new requests queued since we resumed */
2049 kick_pending_request_queues(rinfo);
2052 list_for_each_entry_safe(req, n, &info->requests, queuelist) {
2053 /* Requeue pending requests (flush or discard) */
2054 list_del_init(&req->queuelist);
2055 BUG_ON(req->nr_phys_segments > segs);
2056 blk_mq_requeue_request(req, false);
2058 blk_mq_start_stopped_hw_queues(info->rq, true);
2059 blk_mq_kick_requeue_list(info->rq);
2061 while ((bio = bio_list_pop(&info->bio_list)) != NULL) {
2062 /* Traverse the list of pending bios and re-queue them */
2063 if (bio_segments(bio) > segs) {
2065 * This bio has more segments than what we can
2066 * handle, we have to split it.
2068 pending = (bio_segments(bio) + segs - 1) / segs;
2069 split_bio = kzalloc(sizeof(*split_bio), GFP_NOIO);
2070 BUG_ON(split_bio == NULL);
2071 atomic_set(&split_bio->pending, pending);
2072 split_bio->bio = bio;
2073 for (i = 0; i < pending; i++) {
2074 offset = (i * segs * XEN_PAGE_SIZE) >> 9;
2075 size = min((unsigned int)(segs * XEN_PAGE_SIZE) >> 9,
2076 (unsigned int)bio_sectors(bio) - offset);
2077 cloned_bio = bio_clone(bio, GFP_NOIO);
2078 BUG_ON(cloned_bio == NULL);
2079 bio_trim(cloned_bio, offset, size);
2080 cloned_bio->bi_private = split_bio;
2081 cloned_bio->bi_end_io = split_bio_end;
2082 submit_bio(cloned_bio);
2085 * Now we have to wait for all those smaller bios to
2086 * end, so we can also end the "parent" bio.
2090 /* We don't need to split this bio */
2098 * We are reconnecting to the backend, due to a suspend/resume, or a backend
2099 * driver restart. We tear down our blkif structure and recreate it, but
2100 * leave the device-layer structures intact so that this is transparent to the
2101 * rest of the kernel.
2103 static int blkfront_resume(struct xenbus_device *dev)
2105 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2109 dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2111 bio_list_init(&info->bio_list);
2112 INIT_LIST_HEAD(&info->requests);
2113 for (i = 0; i < info->nr_rings; i++) {
2114 struct blkfront_ring_info *rinfo = &info->rinfo[i];
2115 struct bio_list merge_bio;
2116 struct blk_shadow *shadow = rinfo->shadow;
2118 for (j = 0; j < BLK_RING_SIZE(info); j++) {
2120 if (!shadow[j].request)
2124 * Get the bios in the request so we can re-queue them.
2126 if (req_op(shadow[i].request) == REQ_OP_FLUSH ||
2127 req_op(shadow[i].request) == REQ_OP_DISCARD ||
2128 req_op(shadow[i].request) == REQ_OP_SECURE_ERASE ||
2129 shadow[j].request->cmd_flags & REQ_FUA) {
2131 * Flush operations don't contain bios, so
2132 * we need to requeue the whole request
2134 * XXX: but this doesn't make any sense for a
2135 * write with the FUA flag set..
2137 list_add(&shadow[j].request->queuelist, &info->requests);
2140 merge_bio.head = shadow[j].request->bio;
2141 merge_bio.tail = shadow[j].request->biotail;
2142 bio_list_merge(&info->bio_list, &merge_bio);
2143 shadow[j].request->bio = NULL;
2144 blk_mq_end_request(shadow[j].request, BLK_STS_OK);
2148 blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2150 err = negotiate_mq(info);
2154 err = talk_to_blkback(dev, info);
2156 blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings);
2159 * We have to wait for the backend to switch to
2160 * connected state, since we want to read which
2161 * features it supports.
2167 static void blkfront_closing(struct blkfront_info *info)
2169 struct xenbus_device *xbdev = info->xbdev;
2170 struct block_device *bdev = NULL;
2172 mutex_lock(&info->mutex);
2174 if (xbdev->state == XenbusStateClosing) {
2175 mutex_unlock(&info->mutex);
2180 bdev = bdget_disk(info->gd, 0);
2182 mutex_unlock(&info->mutex);
2185 xenbus_frontend_closed(xbdev);
2189 mutex_lock(&bdev->bd_mutex);
2191 if (bdev->bd_openers) {
2192 xenbus_dev_error(xbdev, -EBUSY,
2193 "Device in use; refusing to close");
2194 xenbus_switch_state(xbdev, XenbusStateClosing);
2196 xlvbd_release_gendisk(info);
2197 xenbus_frontend_closed(xbdev);
2200 mutex_unlock(&bdev->bd_mutex);
2204 static void blkfront_setup_discard(struct blkfront_info *info)
2207 unsigned int discard_granularity;
2208 unsigned int discard_alignment;
2210 info->feature_discard = 1;
2211 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2212 "discard-granularity", "%u", &discard_granularity,
2213 "discard-alignment", "%u", &discard_alignment,
2216 info->discard_granularity = discard_granularity;
2217 info->discard_alignment = discard_alignment;
2219 info->feature_secdiscard =
2220 !!xenbus_read_unsigned(info->xbdev->otherend, "discard-secure",
2224 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2226 unsigned int psegs, grants;
2228 struct blkfront_info *info = rinfo->dev_info;
2230 if (info->max_indirect_segments == 0) {
2232 grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2235 * When an extra req is required, the maximum
2236 * grants supported is related to the size of the
2237 * Linux block segment.
2239 grants = GRANTS_PER_PSEG;
2243 grants = info->max_indirect_segments;
2244 psegs = DIV_ROUND_UP(grants, GRANTS_PER_PSEG);
2246 err = fill_grant_buffer(rinfo,
2247 (grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2251 if (!info->feature_persistent && info->max_indirect_segments) {
2253 * We are using indirect descriptors but not persistent
2254 * grants, we need to allocate a set of pages that can be
2255 * used for mapping indirect grefs
2257 int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2259 BUG_ON(!list_empty(&rinfo->indirect_pages));
2260 for (i = 0; i < num; i++) {
2261 struct page *indirect_page = alloc_page(GFP_NOIO);
2264 list_add(&indirect_page->lru, &rinfo->indirect_pages);
2268 for (i = 0; i < BLK_RING_SIZE(info); i++) {
2269 rinfo->shadow[i].grants_used = kzalloc(
2270 sizeof(rinfo->shadow[i].grants_used[0]) * grants,
2272 rinfo->shadow[i].sg = kzalloc(sizeof(rinfo->shadow[i].sg[0]) * psegs, GFP_NOIO);
2273 if (info->max_indirect_segments)
2274 rinfo->shadow[i].indirect_grants = kzalloc(
2275 sizeof(rinfo->shadow[i].indirect_grants[0]) *
2276 INDIRECT_GREFS(grants),
2278 if ((rinfo->shadow[i].grants_used == NULL) ||
2279 (rinfo->shadow[i].sg == NULL) ||
2280 (info->max_indirect_segments &&
2281 (rinfo->shadow[i].indirect_grants == NULL)))
2283 sg_init_table(rinfo->shadow[i].sg, psegs);
2290 for (i = 0; i < BLK_RING_SIZE(info); i++) {
2291 kfree(rinfo->shadow[i].grants_used);
2292 rinfo->shadow[i].grants_used = NULL;
2293 kfree(rinfo->shadow[i].sg);
2294 rinfo->shadow[i].sg = NULL;
2295 kfree(rinfo->shadow[i].indirect_grants);
2296 rinfo->shadow[i].indirect_grants = NULL;
2298 if (!list_empty(&rinfo->indirect_pages)) {
2299 struct page *indirect_page, *n;
2300 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2301 list_del(&indirect_page->lru);
2302 __free_page(indirect_page);
2309 * Gather all backend feature-*
2311 static void blkfront_gather_backend_features(struct blkfront_info *info)
2313 unsigned int indirect_segments;
2315 info->feature_flush = 0;
2316 info->feature_fua = 0;
2319 * If there's no "feature-barrier" defined, then it means
2320 * we're dealing with a very old backend which writes
2321 * synchronously; nothing to do.
2323 * If there are barriers, then we use flush.
2325 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-barrier", 0)) {
2326 info->feature_flush = 1;
2327 info->feature_fua = 1;
2331 * And if there is "feature-flush-cache" use that above
2334 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-flush-cache",
2336 info->feature_flush = 1;
2337 info->feature_fua = 0;
2340 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-discard", 0))
2341 blkfront_setup_discard(info);
2343 info->feature_persistent =
2344 !!xenbus_read_unsigned(info->xbdev->otherend,
2345 "feature-persistent", 0);
2347 indirect_segments = xenbus_read_unsigned(info->xbdev->otherend,
2348 "feature-max-indirect-segments", 0);
2349 if (indirect_segments > xen_blkif_max_segments)
2350 indirect_segments = xen_blkif_max_segments;
2351 if (indirect_segments <= BLKIF_MAX_SEGMENTS_PER_REQUEST)
2352 indirect_segments = 0;
2353 info->max_indirect_segments = indirect_segments;
2357 * Invoked when the backend is finally 'ready' (and has told produced
2358 * the details about the physical device - #sectors, size, etc).
2360 static void blkfront_connect(struct blkfront_info *info)
2362 unsigned long long sectors;
2363 unsigned long sector_size;
2364 unsigned int physical_sector_size;
2366 char *envp[] = { "RESIZE=1", NULL };
2369 switch (info->connected) {
2370 case BLKIF_STATE_CONNECTED:
2372 * Potentially, the back-end may be signalling
2373 * a capacity change; update the capacity.
2375 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2376 "sectors", "%Lu", §ors);
2377 if (XENBUS_EXIST_ERR(err))
2379 printk(KERN_INFO "Setting capacity to %Lu\n",
2381 set_capacity(info->gd, sectors);
2382 revalidate_disk(info->gd);
2383 kobject_uevent_env(&disk_to_dev(info->gd)->kobj,
2387 case BLKIF_STATE_SUSPENDED:
2389 * If we are recovering from suspension, we need to wait
2390 * for the backend to announce it's features before
2391 * reconnecting, at least we need to know if the backend
2392 * supports indirect descriptors, and how many.
2394 blkif_recover(info);
2401 dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2402 __func__, info->xbdev->otherend);
2404 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2405 "sectors", "%llu", §ors,
2406 "info", "%u", &binfo,
2407 "sector-size", "%lu", §or_size,
2410 xenbus_dev_fatal(info->xbdev, err,
2411 "reading backend fields at %s",
2412 info->xbdev->otherend);
2417 * physcial-sector-size is a newer field, so old backends may not
2418 * provide this. Assume physical sector size to be the same as
2419 * sector_size in that case.
2421 physical_sector_size = xenbus_read_unsigned(info->xbdev->otherend,
2422 "physical-sector-size",
2424 blkfront_gather_backend_features(info);
2425 for (i = 0; i < info->nr_rings; i++) {
2426 err = blkfront_setup_indirect(&info->rinfo[i]);
2428 xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2429 info->xbdev->otherend);
2430 blkif_free(info, 0);
2435 err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
2436 physical_sector_size);
2438 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2439 info->xbdev->otherend);
2443 xenbus_switch_state(info->xbdev, XenbusStateConnected);
2445 /* Kick pending requests. */
2446 info->connected = BLKIF_STATE_CONNECTED;
2447 for (i = 0; i < info->nr_rings; i++)
2448 kick_pending_request_queues(&info->rinfo[i]);
2450 device_add_disk(&info->xbdev->dev, info->gd);
2456 blkif_free(info, 0);
2461 * Callback received when the backend's state changes.
2463 static void blkback_changed(struct xenbus_device *dev,
2464 enum xenbus_state backend_state)
2466 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2468 dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2470 switch (backend_state) {
2471 case XenbusStateInitWait:
2472 if (dev->state != XenbusStateInitialising)
2474 if (talk_to_blkback(dev, info))
2476 case XenbusStateInitialising:
2477 case XenbusStateInitialised:
2478 case XenbusStateReconfiguring:
2479 case XenbusStateReconfigured:
2480 case XenbusStateUnknown:
2483 case XenbusStateConnected:
2485 * talk_to_blkback sets state to XenbusStateInitialised
2486 * and blkfront_connect sets it to XenbusStateConnected
2487 * (if connection went OK).
2489 * If the backend (or toolstack) decides to poke at backend
2490 * state (and re-trigger the watch by setting the state repeatedly
2491 * to XenbusStateConnected (4)) we need to deal with this.
2492 * This is allowed as this is used to communicate to the guest
2493 * that the size of disk has changed!
2495 if ((dev->state != XenbusStateInitialised) &&
2496 (dev->state != XenbusStateConnected)) {
2497 if (talk_to_blkback(dev, info))
2501 blkfront_connect(info);
2504 case XenbusStateClosed:
2505 if (dev->state == XenbusStateClosed)
2507 /* Missed the backend's Closing state -- fallthrough */
2508 case XenbusStateClosing:
2510 blkfront_closing(info);
2515 static int blkfront_remove(struct xenbus_device *xbdev)
2517 struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2518 struct block_device *bdev = NULL;
2519 struct gendisk *disk;
2521 dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2523 blkif_free(info, 0);
2525 mutex_lock(&info->mutex);
2529 bdev = bdget_disk(disk, 0);
2532 mutex_unlock(&info->mutex);
2540 * The xbdev was removed before we reached the Closed
2541 * state. See if it's safe to remove the disk. If the bdev
2542 * isn't closed yet, we let release take care of it.
2545 mutex_lock(&bdev->bd_mutex);
2546 info = disk->private_data;
2548 dev_warn(disk_to_dev(disk),
2549 "%s was hot-unplugged, %d stale handles\n",
2550 xbdev->nodename, bdev->bd_openers);
2552 if (info && !bdev->bd_openers) {
2553 xlvbd_release_gendisk(info);
2554 disk->private_data = NULL;
2558 mutex_unlock(&bdev->bd_mutex);
2564 static int blkfront_is_ready(struct xenbus_device *dev)
2566 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2568 return info->is_ready && info->xbdev;
2571 static int blkif_open(struct block_device *bdev, fmode_t mode)
2573 struct gendisk *disk = bdev->bd_disk;
2574 struct blkfront_info *info;
2577 mutex_lock(&blkfront_mutex);
2579 info = disk->private_data;
2586 mutex_lock(&info->mutex);
2589 /* xbdev is closed */
2592 mutex_unlock(&info->mutex);
2595 mutex_unlock(&blkfront_mutex);
2599 static void blkif_release(struct gendisk *disk, fmode_t mode)
2601 struct blkfront_info *info = disk->private_data;
2602 struct block_device *bdev;
2603 struct xenbus_device *xbdev;
2605 mutex_lock(&blkfront_mutex);
2607 bdev = bdget_disk(disk, 0);
2610 WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2613 if (bdev->bd_openers)
2617 * Check if we have been instructed to close. We will have
2618 * deferred this request, because the bdev was still open.
2621 mutex_lock(&info->mutex);
2622 xbdev = info->xbdev;
2624 if (xbdev && xbdev->state == XenbusStateClosing) {
2625 /* pending switch to state closed */
2626 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2627 xlvbd_release_gendisk(info);
2628 xenbus_frontend_closed(info->xbdev);
2631 mutex_unlock(&info->mutex);
2634 /* sudden device removal */
2635 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2636 xlvbd_release_gendisk(info);
2637 disk->private_data = NULL;
2644 mutex_unlock(&blkfront_mutex);
2647 static const struct block_device_operations xlvbd_block_fops =
2649 .owner = THIS_MODULE,
2651 .release = blkif_release,
2652 .getgeo = blkif_getgeo,
2653 .ioctl = blkif_ioctl,
2657 static const struct xenbus_device_id blkfront_ids[] = {
2662 static struct xenbus_driver blkfront_driver = {
2663 .ids = blkfront_ids,
2664 .probe = blkfront_probe,
2665 .remove = blkfront_remove,
2666 .resume = blkfront_resume,
2667 .otherend_changed = blkback_changed,
2668 .is_ready = blkfront_is_ready,
2671 static int __init xlblk_init(void)
2674 int nr_cpus = num_online_cpus();
2679 if (xen_blkif_max_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST)
2680 xen_blkif_max_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2682 if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2683 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2684 xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2685 xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2688 if (xen_blkif_max_queues > nr_cpus) {
2689 pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2690 xen_blkif_max_queues, nr_cpus);
2691 xen_blkif_max_queues = nr_cpus;
2694 if (!xen_has_pv_disk_devices())
2697 if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2698 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
2699 XENVBD_MAJOR, DEV_NAME);
2703 ret = xenbus_register_frontend(&blkfront_driver);
2705 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2711 module_init(xlblk_init);
2714 static void __exit xlblk_exit(void)
2716 xenbus_unregister_driver(&blkfront_driver);
2717 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2720 module_exit(xlblk_exit);
2722 MODULE_DESCRIPTION("Xen virtual block device frontend");
2723 MODULE_LICENSE("GPL");
2724 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2725 MODULE_ALIAS("xen:vbd");
2726 MODULE_ALIAS("xenblk");