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,
87 struct list_head node;
99 struct blkif_request req;
100 struct request *request;
101 struct grant **grants_used;
102 struct grant **indirect_grants;
103 struct scatterlist *sg;
105 enum blk_req_status status;
107 #define NO_ASSOCIATED_ID ~0UL
109 * Id of the sibling if we ever need 2 requests when handling a
112 unsigned long associated_id;
119 static inline struct blkif_req *blkif_req(struct request *rq)
121 return blk_mq_rq_to_pdu(rq);
124 static DEFINE_MUTEX(blkfront_mutex);
125 static const struct block_device_operations xlvbd_block_fops;
128 * Maximum number of segments in indirect requests, the actual value used by
129 * the frontend driver is the minimum of this value and the value provided
130 * by the backend driver.
133 static unsigned int xen_blkif_max_segments = 32;
134 module_param_named(max_indirect_segments, xen_blkif_max_segments, uint,
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, S_IRUGO);
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, S_IRUGO);
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)
154 #define BLK_MAX_RING_SIZE \
155 __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * XENBUS_MAX_RING_GRANTS)
158 * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
159 * characters are enough. Define to 20 to keep consistent with backend.
161 #define RINGREF_NAME_LEN (20)
163 * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
165 #define QUEUE_NAME_LEN (17)
169 * Every blkfront device can associate with one or more blkfront_ring_info,
170 * depending on how many hardware queues/rings to be used.
172 struct blkfront_ring_info {
173 /* Lock to protect data in every ring buffer. */
174 spinlock_t ring_lock;
175 struct blkif_front_ring ring;
176 unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
177 unsigned int evtchn, irq;
178 struct work_struct work;
179 struct gnttab_free_callback callback;
180 struct blk_shadow shadow[BLK_MAX_RING_SIZE];
181 struct list_head indirect_pages;
182 struct list_head grants;
183 unsigned int persistent_gnts_c;
184 unsigned long shadow_free;
185 struct blkfront_info *dev_info;
189 * We have one of these per vbd, whether ide, scsi or 'other'. They
190 * hang in private_data off the gendisk structure. We may end up
191 * putting all kinds of interesting stuff here :-)
196 struct xenbus_device *xbdev;
199 unsigned int physical_sector_size;
202 enum blkif_state connected;
203 /* Number of pages per ring buffer. */
204 unsigned int nr_ring_pages;
205 struct request_queue *rq;
206 unsigned int feature_flush:1;
207 unsigned int feature_fua:1;
208 unsigned int feature_discard:1;
209 unsigned int feature_secdiscard:1;
210 unsigned int feature_persistent:1;
211 unsigned int discard_granularity;
212 unsigned int discard_alignment;
213 /* Number of 4KB segments handled */
214 unsigned int max_indirect_segments;
216 struct blk_mq_tag_set tag_set;
217 struct blkfront_ring_info *rinfo;
218 unsigned int nr_rings;
219 /* Save uncomplete reqs and bios for migration. */
220 struct list_head requests;
221 struct bio_list bio_list;
224 static unsigned int nr_minors;
225 static unsigned long *minors;
226 static DEFINE_SPINLOCK(minor_lock);
228 #define GRANT_INVALID_REF 0
230 #define PARTS_PER_DISK 16
231 #define PARTS_PER_EXT_DISK 256
233 #define BLKIF_MAJOR(dev) ((dev)>>8)
234 #define BLKIF_MINOR(dev) ((dev) & 0xff)
237 #define EXTENDED (1<<EXT_SHIFT)
238 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
239 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
240 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
241 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
242 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
243 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
245 #define DEV_NAME "xvd" /* name in /dev */
248 * Grants are always the same size as a Xen page (i.e 4KB).
249 * A physical segment is always the same size as a Linux page.
250 * Number of grants per physical segment
252 #define GRANTS_PER_PSEG (PAGE_SIZE / XEN_PAGE_SIZE)
254 #define GRANTS_PER_INDIRECT_FRAME \
255 (XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
257 #define PSEGS_PER_INDIRECT_FRAME \
258 (GRANTS_INDIRECT_FRAME / GRANTS_PSEGS)
260 #define INDIRECT_GREFS(_grants) \
261 DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
263 #define GREFS(_psegs) ((_psegs) * GRANTS_PER_PSEG)
265 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
266 static void blkfront_gather_backend_features(struct blkfront_info *info);
267 static int negotiate_mq(struct blkfront_info *info);
269 static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
271 unsigned long free = rinfo->shadow_free;
273 BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
274 rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
275 rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
279 static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
282 if (rinfo->shadow[id].req.u.rw.id != id)
284 if (rinfo->shadow[id].request == NULL)
286 rinfo->shadow[id].req.u.rw.id = rinfo->shadow_free;
287 rinfo->shadow[id].request = NULL;
288 rinfo->shadow_free = id;
292 static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
294 struct blkfront_info *info = rinfo->dev_info;
295 struct page *granted_page;
296 struct grant *gnt_list_entry, *n;
300 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
304 if (info->feature_persistent) {
305 granted_page = alloc_page(GFP_NOIO);
307 kfree(gnt_list_entry);
310 gnt_list_entry->page = granted_page;
313 gnt_list_entry->gref = GRANT_INVALID_REF;
314 list_add(&gnt_list_entry->node, &rinfo->grants);
321 list_for_each_entry_safe(gnt_list_entry, n,
322 &rinfo->grants, node) {
323 list_del(&gnt_list_entry->node);
324 if (info->feature_persistent)
325 __free_page(gnt_list_entry->page);
326 kfree(gnt_list_entry);
333 static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
335 struct grant *gnt_list_entry;
337 BUG_ON(list_empty(&rinfo->grants));
338 gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
340 list_del(&gnt_list_entry->node);
342 if (gnt_list_entry->gref != GRANT_INVALID_REF)
343 rinfo->persistent_gnts_c--;
345 return gnt_list_entry;
348 static inline void grant_foreign_access(const struct grant *gnt_list_entry,
349 const struct blkfront_info *info)
351 gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
352 info->xbdev->otherend_id,
353 gnt_list_entry->page,
357 static struct grant *get_grant(grant_ref_t *gref_head,
359 struct blkfront_ring_info *rinfo)
361 struct grant *gnt_list_entry = get_free_grant(rinfo);
362 struct blkfront_info *info = rinfo->dev_info;
364 if (gnt_list_entry->gref != GRANT_INVALID_REF)
365 return gnt_list_entry;
367 /* Assign a gref to this page */
368 gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
369 BUG_ON(gnt_list_entry->gref == -ENOSPC);
370 if (info->feature_persistent)
371 grant_foreign_access(gnt_list_entry, info);
373 /* Grant access to the GFN passed by the caller */
374 gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
375 info->xbdev->otherend_id,
379 return gnt_list_entry;
382 static struct grant *get_indirect_grant(grant_ref_t *gref_head,
383 struct blkfront_ring_info *rinfo)
385 struct grant *gnt_list_entry = get_free_grant(rinfo);
386 struct blkfront_info *info = rinfo->dev_info;
388 if (gnt_list_entry->gref != GRANT_INVALID_REF)
389 return gnt_list_entry;
391 /* Assign a gref to this page */
392 gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
393 BUG_ON(gnt_list_entry->gref == -ENOSPC);
394 if (!info->feature_persistent) {
395 struct page *indirect_page;
397 /* Fetch a pre-allocated page to use for indirect grefs */
398 BUG_ON(list_empty(&rinfo->indirect_pages));
399 indirect_page = list_first_entry(&rinfo->indirect_pages,
401 list_del(&indirect_page->lru);
402 gnt_list_entry->page = indirect_page;
404 grant_foreign_access(gnt_list_entry, info);
406 return gnt_list_entry;
409 static const char *op_name(int op)
411 static const char *const names[] = {
412 [BLKIF_OP_READ] = "read",
413 [BLKIF_OP_WRITE] = "write",
414 [BLKIF_OP_WRITE_BARRIER] = "barrier",
415 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
416 [BLKIF_OP_DISCARD] = "discard" };
418 if (op < 0 || op >= ARRAY_SIZE(names))
426 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
428 unsigned int end = minor + nr;
431 if (end > nr_minors) {
432 unsigned long *bitmap, *old;
434 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
439 spin_lock(&minor_lock);
440 if (end > nr_minors) {
442 memcpy(bitmap, minors,
443 BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
445 nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
448 spin_unlock(&minor_lock);
452 spin_lock(&minor_lock);
453 if (find_next_bit(minors, end, minor) >= end) {
454 bitmap_set(minors, minor, nr);
458 spin_unlock(&minor_lock);
463 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
465 unsigned int end = minor + nr;
467 BUG_ON(end > nr_minors);
468 spin_lock(&minor_lock);
469 bitmap_clear(minors, minor, nr);
470 spin_unlock(&minor_lock);
473 static void blkif_restart_queue_callback(void *arg)
475 struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
476 schedule_work(&rinfo->work);
479 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
481 /* We don't have real geometry info, but let's at least return
482 values consistent with the size of the device */
483 sector_t nsect = get_capacity(bd->bd_disk);
484 sector_t cylinders = nsect;
488 sector_div(cylinders, hg->heads * hg->sectors);
489 hg->cylinders = cylinders;
490 if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
491 hg->cylinders = 0xffff;
495 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
496 unsigned command, unsigned long argument)
498 struct blkfront_info *info = bdev->bd_disk->private_data;
501 dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
502 command, (long)argument);
505 case CDROMMULTISESSION:
506 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
507 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
508 if (put_user(0, (char __user *)(argument + i)))
512 case CDROM_GET_CAPABILITY: {
513 struct gendisk *gd = info->gd;
514 if (gd->flags & GENHD_FL_CD)
520 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
522 return -EINVAL; /* same return as native Linux */
528 static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
530 struct blkif_request **ring_req)
534 *ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
535 rinfo->ring.req_prod_pvt++;
537 id = get_id_from_freelist(rinfo);
538 rinfo->shadow[id].request = req;
539 rinfo->shadow[id].status = REQ_PROCESSING;
540 rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
542 rinfo->shadow[id].req.u.rw.id = id;
547 static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
549 struct blkfront_info *info = rinfo->dev_info;
550 struct blkif_request *ring_req, *final_ring_req;
553 /* Fill out a communications ring structure. */
554 id = blkif_ring_get_request(rinfo, req, &final_ring_req);
555 ring_req = &rinfo->shadow[id].req;
557 ring_req->operation = BLKIF_OP_DISCARD;
558 ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
559 ring_req->u.discard.id = id;
560 ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
561 if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard)
562 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
564 ring_req->u.discard.flag = 0;
566 /* Copy the request to the ring page. */
567 *final_ring_req = *ring_req;
568 rinfo->shadow[id].status = REQ_WAITING;
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 struct blkif_request *final_ring_req, *final_extra_ring_req = NULL;
702 unsigned long id, extra_id = NO_ASSOCIATED_ID;
703 bool require_extra_req = false;
705 struct setup_rw_req setup = {
709 .need_copy = rq_data_dir(req) && info->feature_persistent,
713 * Used to store if we are able to queue the request by just using
714 * existing persistent grants, or if we have to get new grants,
715 * as there are not sufficiently many free.
717 bool new_persistent_gnts = false;
718 struct scatterlist *sg;
719 int num_sg, max_grefs, num_grant;
721 max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
722 if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
724 * If we are using indirect segments we need to account
725 * for the indirect grefs used in the request.
727 max_grefs += INDIRECT_GREFS(max_grefs);
729 /* Check if we have enough persistent grants to allocate a requests */
730 if (rinfo->persistent_gnts_c < max_grefs) {
731 new_persistent_gnts = true;
733 if (gnttab_alloc_grant_references(
734 max_grefs - rinfo->persistent_gnts_c,
735 &setup.gref_head) < 0) {
736 gnttab_request_free_callback(
738 blkif_restart_queue_callback,
740 max_grefs - rinfo->persistent_gnts_c);
745 /* Fill out a communications ring structure. */
746 id = blkif_ring_get_request(rinfo, req, &final_ring_req);
747 ring_req = &rinfo->shadow[id].req;
749 num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
751 /* Calculate the number of grant used */
752 for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
753 num_grant += gnttab_count_grant(sg->offset, sg->length);
755 require_extra_req = info->max_indirect_segments == 0 &&
756 num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
757 BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
759 rinfo->shadow[id].num_sg = num_sg;
760 if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
761 likely(!require_extra_req)) {
763 * The indirect operation can only be a BLKIF_OP_READ or
766 BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA);
767 ring_req->operation = BLKIF_OP_INDIRECT;
768 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
769 BLKIF_OP_WRITE : BLKIF_OP_READ;
770 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
771 ring_req->u.indirect.handle = info->handle;
772 ring_req->u.indirect.nr_segments = num_grant;
774 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
775 ring_req->u.rw.handle = info->handle;
776 ring_req->operation = rq_data_dir(req) ?
777 BLKIF_OP_WRITE : BLKIF_OP_READ;
778 if (req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA) {
780 * Ideally we can do an unordered flush-to-disk.
781 * In case the backend onlysupports barriers, use that.
782 * A barrier request a superset of FUA, so we can
783 * implement it the same way. (It's also a FLUSH+FUA,
784 * since it is guaranteed ordered WRT previous writes.)
786 if (info->feature_flush && info->feature_fua)
787 ring_req->operation =
788 BLKIF_OP_WRITE_BARRIER;
789 else if (info->feature_flush)
790 ring_req->operation =
791 BLKIF_OP_FLUSH_DISKCACHE;
793 ring_req->operation = 0;
795 ring_req->u.rw.nr_segments = num_grant;
796 if (unlikely(require_extra_req)) {
797 extra_id = blkif_ring_get_request(rinfo, req,
798 &final_extra_ring_req);
799 extra_ring_req = &rinfo->shadow[extra_id].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 /* Copy request(s) to the ring page. */
843 *final_ring_req = *ring_req;
844 rinfo->shadow[id].status = REQ_WAITING;
845 if (unlikely(require_extra_req)) {
846 *final_extra_ring_req = *extra_ring_req;
847 rinfo->shadow[extra_id].status = REQ_WAITING;
850 if (new_persistent_gnts)
851 gnttab_free_grant_references(setup.gref_head);
857 * Generate a Xen blkfront IO request from a blk layer request. Reads
858 * and writes are handled as expected.
860 * @req: a request struct
862 static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
864 if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
867 if (unlikely(req_op(req) == REQ_OP_DISCARD ||
868 req_op(req) == REQ_OP_SECURE_ERASE))
869 return blkif_queue_discard_req(req, rinfo);
871 return blkif_queue_rw_req(req, rinfo);
874 static inline void flush_requests(struct blkfront_ring_info *rinfo)
878 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
881 notify_remote_via_irq(rinfo->irq);
884 static inline bool blkif_request_flush_invalid(struct request *req,
885 struct blkfront_info *info)
887 return (blk_rq_is_passthrough(req) ||
888 ((req_op(req) == REQ_OP_FLUSH) &&
889 !info->feature_flush) ||
890 ((req->cmd_flags & REQ_FUA) &&
891 !info->feature_fua));
894 static blk_status_t blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
895 const struct blk_mq_queue_data *qd)
898 int qid = hctx->queue_num;
899 struct blkfront_info *info = hctx->queue->queuedata;
900 struct blkfront_ring_info *rinfo = NULL;
902 BUG_ON(info->nr_rings <= qid);
903 rinfo = &info->rinfo[qid];
904 blk_mq_start_request(qd->rq);
905 spin_lock_irqsave(&rinfo->ring_lock, flags);
906 if (RING_FULL(&rinfo->ring))
909 if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
912 if (blkif_queue_request(qd->rq, rinfo))
915 flush_requests(rinfo);
916 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
920 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
921 return BLK_STS_IOERR;
924 blk_mq_stop_hw_queue(hctx);
925 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
926 return BLK_STS_RESOURCE;
929 static void blkif_complete_rq(struct request *rq)
931 blk_mq_end_request(rq, blkif_req(rq)->error);
934 static const struct blk_mq_ops blkfront_mq_ops = {
935 .queue_rq = blkif_queue_rq,
936 .complete = blkif_complete_rq,
939 static void blkif_set_queue_limits(struct blkfront_info *info)
941 struct request_queue *rq = info->rq;
942 struct gendisk *gd = info->gd;
943 unsigned int segments = info->max_indirect_segments ? :
944 BLKIF_MAX_SEGMENTS_PER_REQUEST;
946 queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
948 if (info->feature_discard) {
949 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
950 blk_queue_max_discard_sectors(rq, get_capacity(gd));
951 rq->limits.discard_granularity = info->discard_granularity ?:
952 info->physical_sector_size;
953 rq->limits.discard_alignment = info->discard_alignment;
954 if (info->feature_secdiscard)
955 queue_flag_set_unlocked(QUEUE_FLAG_SECERASE, rq);
958 /* Hard sector size and max sectors impersonate the equiv. hardware. */
959 blk_queue_logical_block_size(rq, info->sector_size);
960 blk_queue_physical_block_size(rq, info->physical_sector_size);
961 blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
963 /* Each segment in a request is up to an aligned page in size. */
964 blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
965 blk_queue_max_segment_size(rq, PAGE_SIZE);
967 /* Ensure a merged request will fit in a single I/O ring slot. */
968 blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
970 /* Make sure buffer addresses are sector-aligned. */
971 blk_queue_dma_alignment(rq, 511);
974 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
975 unsigned int physical_sector_size)
977 struct request_queue *rq;
978 struct blkfront_info *info = gd->private_data;
980 memset(&info->tag_set, 0, sizeof(info->tag_set));
981 info->tag_set.ops = &blkfront_mq_ops;
982 info->tag_set.nr_hw_queues = info->nr_rings;
983 if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
985 * When indirect descriptior is not supported, the I/O request
986 * will be split between multiple request in the ring.
987 * To avoid problems when sending the request, divide by
988 * 2 the depth of the queue.
990 info->tag_set.queue_depth = BLK_RING_SIZE(info) / 2;
992 info->tag_set.queue_depth = BLK_RING_SIZE(info);
993 info->tag_set.numa_node = NUMA_NO_NODE;
994 info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
995 info->tag_set.cmd_size = sizeof(struct blkif_req);
996 info->tag_set.driver_data = info;
998 if (blk_mq_alloc_tag_set(&info->tag_set))
1000 rq = blk_mq_init_queue(&info->tag_set);
1002 blk_mq_free_tag_set(&info->tag_set);
1006 rq->queuedata = info;
1007 info->rq = gd->queue = rq;
1009 info->sector_size = sector_size;
1010 info->physical_sector_size = physical_sector_size;
1011 blkif_set_queue_limits(info);
1016 static const char *flush_info(struct blkfront_info *info)
1018 if (info->feature_flush && info->feature_fua)
1019 return "barrier: enabled;";
1020 else if (info->feature_flush)
1021 return "flush diskcache: enabled;";
1023 return "barrier or flush: disabled;";
1026 static void xlvbd_flush(struct blkfront_info *info)
1028 blk_queue_write_cache(info->rq, info->feature_flush ? true : false,
1029 info->feature_fua ? true : false);
1030 pr_info("blkfront: %s: %s %s %s %s %s\n",
1031 info->gd->disk_name, flush_info(info),
1032 "persistent grants:", info->feature_persistent ?
1033 "enabled;" : "disabled;", "indirect descriptors:",
1034 info->max_indirect_segments ? "enabled;" : "disabled;");
1037 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
1040 major = BLKIF_MAJOR(vdevice);
1041 *minor = BLKIF_MINOR(vdevice);
1043 case XEN_IDE0_MAJOR:
1044 *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
1045 *minor = ((*minor / 64) * PARTS_PER_DISK) +
1046 EMULATED_HD_DISK_MINOR_OFFSET;
1048 case XEN_IDE1_MAJOR:
1049 *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1050 *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1051 EMULATED_HD_DISK_MINOR_OFFSET;
1053 case XEN_SCSI_DISK0_MAJOR:
1054 *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1055 *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1057 case XEN_SCSI_DISK1_MAJOR:
1058 case XEN_SCSI_DISK2_MAJOR:
1059 case XEN_SCSI_DISK3_MAJOR:
1060 case XEN_SCSI_DISK4_MAJOR:
1061 case XEN_SCSI_DISK5_MAJOR:
1062 case XEN_SCSI_DISK6_MAJOR:
1063 case XEN_SCSI_DISK7_MAJOR:
1064 *offset = (*minor / PARTS_PER_DISK) +
1065 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1066 EMULATED_SD_DISK_NAME_OFFSET;
1068 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1069 EMULATED_SD_DISK_MINOR_OFFSET;
1071 case XEN_SCSI_DISK8_MAJOR:
1072 case XEN_SCSI_DISK9_MAJOR:
1073 case XEN_SCSI_DISK10_MAJOR:
1074 case XEN_SCSI_DISK11_MAJOR:
1075 case XEN_SCSI_DISK12_MAJOR:
1076 case XEN_SCSI_DISK13_MAJOR:
1077 case XEN_SCSI_DISK14_MAJOR:
1078 case XEN_SCSI_DISK15_MAJOR:
1079 *offset = (*minor / PARTS_PER_DISK) +
1080 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1081 EMULATED_SD_DISK_NAME_OFFSET;
1083 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1084 EMULATED_SD_DISK_MINOR_OFFSET;
1087 *offset = *minor / PARTS_PER_DISK;
1090 printk(KERN_WARNING "blkfront: your disk configuration is "
1091 "incorrect, please use an xvd device instead\n");
1097 static char *encode_disk_name(char *ptr, unsigned int n)
1100 ptr = encode_disk_name(ptr, n / 26 - 1);
1101 *ptr = 'a' + n % 26;
1105 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1106 struct blkfront_info *info,
1107 u16 vdisk_info, u16 sector_size,
1108 unsigned int physical_sector_size)
1113 unsigned int offset;
1118 BUG_ON(info->gd != NULL);
1119 BUG_ON(info->rq != NULL);
1121 if ((info->vdevice>>EXT_SHIFT) > 1) {
1122 /* this is above the extended range; something is wrong */
1123 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1127 if (!VDEV_IS_EXTENDED(info->vdevice)) {
1128 err = xen_translate_vdev(info->vdevice, &minor, &offset);
1131 nr_parts = PARTS_PER_DISK;
1133 minor = BLKIF_MINOR_EXT(info->vdevice);
1134 nr_parts = PARTS_PER_EXT_DISK;
1135 offset = minor / nr_parts;
1136 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1137 printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1138 "emulated IDE disks,\n\t choose an xvd device name"
1139 "from xvde on\n", info->vdevice);
1141 if (minor >> MINORBITS) {
1142 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1143 info->vdevice, minor);
1147 if ((minor % nr_parts) == 0)
1148 nr_minors = nr_parts;
1150 err = xlbd_reserve_minors(minor, nr_minors);
1155 gd = alloc_disk(nr_minors);
1159 strcpy(gd->disk_name, DEV_NAME);
1160 ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1161 BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1165 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1166 "%d", minor & (nr_parts - 1));
1168 gd->major = XENVBD_MAJOR;
1169 gd->first_minor = minor;
1170 gd->fops = &xlvbd_block_fops;
1171 gd->private_data = info;
1172 set_capacity(gd, capacity);
1174 if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size)) {
1181 if (vdisk_info & VDISK_READONLY)
1184 if (vdisk_info & VDISK_REMOVABLE)
1185 gd->flags |= GENHD_FL_REMOVABLE;
1187 if (vdisk_info & VDISK_CDROM)
1188 gd->flags |= GENHD_FL_CD;
1193 xlbd_release_minors(minor, nr_minors);
1198 static void xlvbd_release_gendisk(struct blkfront_info *info)
1200 unsigned int minor, nr_minors, i;
1202 if (info->rq == NULL)
1205 /* No more blkif_request(). */
1206 blk_mq_stop_hw_queues(info->rq);
1208 for (i = 0; i < info->nr_rings; i++) {
1209 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1211 /* No more gnttab callback work. */
1212 gnttab_cancel_free_callback(&rinfo->callback);
1214 /* Flush gnttab callback work. Must be done with no locks held. */
1215 flush_work(&rinfo->work);
1218 del_gendisk(info->gd);
1220 minor = info->gd->first_minor;
1221 nr_minors = info->gd->minors;
1222 xlbd_release_minors(minor, nr_minors);
1224 blk_cleanup_queue(info->rq);
1225 blk_mq_free_tag_set(&info->tag_set);
1232 /* Already hold rinfo->ring_lock. */
1233 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1235 if (!RING_FULL(&rinfo->ring))
1236 blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1239 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1241 unsigned long flags;
1243 spin_lock_irqsave(&rinfo->ring_lock, flags);
1244 kick_pending_request_queues_locked(rinfo);
1245 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1248 static void blkif_restart_queue(struct work_struct *work)
1250 struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1252 if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1253 kick_pending_request_queues(rinfo);
1256 static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1258 struct grant *persistent_gnt, *n;
1259 struct blkfront_info *info = rinfo->dev_info;
1263 * Remove indirect pages, this only happens when using indirect
1264 * descriptors but not persistent grants
1266 if (!list_empty(&rinfo->indirect_pages)) {
1267 struct page *indirect_page, *n;
1269 BUG_ON(info->feature_persistent);
1270 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1271 list_del(&indirect_page->lru);
1272 __free_page(indirect_page);
1276 /* Remove all persistent grants. */
1277 if (!list_empty(&rinfo->grants)) {
1278 list_for_each_entry_safe(persistent_gnt, n,
1279 &rinfo->grants, node) {
1280 list_del(&persistent_gnt->node);
1281 if (persistent_gnt->gref != GRANT_INVALID_REF) {
1282 gnttab_end_foreign_access(persistent_gnt->gref,
1284 rinfo->persistent_gnts_c--;
1286 if (info->feature_persistent)
1287 __free_page(persistent_gnt->page);
1288 kfree(persistent_gnt);
1291 BUG_ON(rinfo->persistent_gnts_c != 0);
1293 for (i = 0; i < BLK_RING_SIZE(info); i++) {
1295 * Clear persistent grants present in requests already
1296 * on the shared ring
1298 if (!rinfo->shadow[i].request)
1301 segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1302 rinfo->shadow[i].req.u.indirect.nr_segments :
1303 rinfo->shadow[i].req.u.rw.nr_segments;
1304 for (j = 0; j < segs; j++) {
1305 persistent_gnt = rinfo->shadow[i].grants_used[j];
1306 gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1307 if (info->feature_persistent)
1308 __free_page(persistent_gnt->page);
1309 kfree(persistent_gnt);
1312 if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1314 * If this is not an indirect operation don't try to
1315 * free indirect segments
1319 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1320 persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1321 gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1322 __free_page(persistent_gnt->page);
1323 kfree(persistent_gnt);
1327 kfree(rinfo->shadow[i].grants_used);
1328 rinfo->shadow[i].grants_used = NULL;
1329 kfree(rinfo->shadow[i].indirect_grants);
1330 rinfo->shadow[i].indirect_grants = NULL;
1331 kfree(rinfo->shadow[i].sg);
1332 rinfo->shadow[i].sg = NULL;
1335 /* No more gnttab callback work. */
1336 gnttab_cancel_free_callback(&rinfo->callback);
1338 /* Flush gnttab callback work. Must be done with no locks held. */
1339 flush_work(&rinfo->work);
1341 /* Free resources associated with old device channel. */
1342 for (i = 0; i < info->nr_ring_pages; i++) {
1343 if (rinfo->ring_ref[i] != GRANT_INVALID_REF) {
1344 gnttab_end_foreign_access(rinfo->ring_ref[i], 0, 0);
1345 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1348 free_pages((unsigned long)rinfo->ring.sring, get_order(info->nr_ring_pages * XEN_PAGE_SIZE));
1349 rinfo->ring.sring = NULL;
1352 unbind_from_irqhandler(rinfo->irq, rinfo);
1353 rinfo->evtchn = rinfo->irq = 0;
1356 static void blkif_free(struct blkfront_info *info, int suspend)
1360 /* Prevent new requests being issued until we fix things up. */
1361 info->connected = suspend ?
1362 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1363 /* No more blkif_request(). */
1365 blk_mq_stop_hw_queues(info->rq);
1367 for (i = 0; i < info->nr_rings; i++)
1368 blkif_free_ring(&info->rinfo[i]);
1375 struct copy_from_grant {
1376 const struct blk_shadow *s;
1377 unsigned int grant_idx;
1378 unsigned int bvec_offset;
1382 static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1383 unsigned int len, void *data)
1385 struct copy_from_grant *info = data;
1387 /* Convenient aliases */
1388 const struct blk_shadow *s = info->s;
1390 shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1392 memcpy(info->bvec_data + info->bvec_offset,
1393 shared_data + offset, len);
1395 info->bvec_offset += len;
1398 kunmap_atomic(shared_data);
1401 static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1405 case BLKIF_RSP_OKAY:
1407 case BLKIF_RSP_EOPNOTSUPP:
1408 return REQ_EOPNOTSUPP;
1409 case BLKIF_RSP_ERROR:
1417 * Get the final status of the block request based on two ring response
1419 static int blkif_get_final_status(enum blk_req_status s1,
1420 enum blk_req_status s2)
1422 BUG_ON(s1 < REQ_DONE);
1423 BUG_ON(s2 < REQ_DONE);
1425 if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1426 return BLKIF_RSP_ERROR;
1427 else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1428 return BLKIF_RSP_EOPNOTSUPP;
1429 return BLKIF_RSP_OKAY;
1432 static bool blkif_completion(unsigned long *id,
1433 struct blkfront_ring_info *rinfo,
1434 struct blkif_response *bret)
1437 struct scatterlist *sg;
1438 int num_sg, num_grant;
1439 struct blkfront_info *info = rinfo->dev_info;
1440 struct blk_shadow *s = &rinfo->shadow[*id];
1441 struct copy_from_grant data = {
1445 num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1446 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1448 /* The I/O request may be split in two. */
1449 if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1450 struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1452 /* Keep the status of the current response in shadow. */
1453 s->status = blkif_rsp_to_req_status(bret->status);
1455 /* Wait the second response if not yet here. */
1456 if (s2->status < REQ_DONE)
1459 bret->status = blkif_get_final_status(s->status,
1463 * All the grants is stored in the first shadow in order
1464 * to make the completion code simpler.
1466 num_grant += s2->req.u.rw.nr_segments;
1469 * The two responses may not come in order. Only the
1470 * first request will store the scatter-gather list.
1472 if (s2->num_sg != 0) {
1473 /* Update "id" with the ID of the first response. */
1474 *id = s->associated_id;
1479 * We don't need anymore the second request, so recycling
1482 if (add_id_to_freelist(rinfo, s->associated_id))
1483 WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1484 info->gd->disk_name, s->associated_id);
1490 if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1491 for_each_sg(s->sg, sg, num_sg, i) {
1492 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1494 data.bvec_offset = sg->offset;
1495 data.bvec_data = kmap_atomic(sg_page(sg));
1497 gnttab_foreach_grant_in_range(sg_page(sg),
1500 blkif_copy_from_grant,
1503 kunmap_atomic(data.bvec_data);
1506 /* Add the persistent grant into the list of free grants */
1507 for (i = 0; i < num_grant; i++) {
1508 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1510 * If the grant is still mapped by the backend (the
1511 * backend has chosen to make this grant persistent)
1512 * we add it at the head of the list, so it will be
1515 if (!info->feature_persistent)
1516 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1517 s->grants_used[i]->gref);
1518 list_add(&s->grants_used[i]->node, &rinfo->grants);
1519 rinfo->persistent_gnts_c++;
1522 * If the grant is not mapped by the backend we end the
1523 * foreign access and add it to the tail of the list,
1524 * so it will not be picked again unless we run out of
1525 * persistent grants.
1527 gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1528 s->grants_used[i]->gref = GRANT_INVALID_REF;
1529 list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1532 if (s->req.operation == BLKIF_OP_INDIRECT) {
1533 for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1534 if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1535 if (!info->feature_persistent)
1536 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1537 s->indirect_grants[i]->gref);
1538 list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1539 rinfo->persistent_gnts_c++;
1541 struct page *indirect_page;
1543 gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1545 * Add the used indirect page back to the list of
1546 * available pages for indirect grefs.
1548 if (!info->feature_persistent) {
1549 indirect_page = s->indirect_grants[i]->page;
1550 list_add(&indirect_page->lru, &rinfo->indirect_pages);
1552 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1553 list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1561 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1563 struct request *req;
1564 struct blkif_response bret;
1566 unsigned long flags;
1567 struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1568 struct blkfront_info *info = rinfo->dev_info;
1569 unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS;
1571 if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
1572 xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
1576 spin_lock_irqsave(&rinfo->ring_lock, flags);
1578 rp = READ_ONCE(rinfo->ring.sring->rsp_prod);
1579 virt_rmb(); /* Ensure we see queued responses up to 'rp'. */
1580 if (RING_RESPONSE_PROD_OVERFLOW(&rinfo->ring, rp)) {
1581 pr_alert("%s: illegal number of responses %u\n",
1582 info->gd->disk_name, rp - rinfo->ring.rsp_cons);
1586 for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1592 RING_COPY_RESPONSE(&rinfo->ring, i, &bret);
1596 * The backend has messed up and given us an id that we would
1597 * never have given to it (we stamp it up to BLK_RING_SIZE -
1598 * look in get_id_from_freelist.
1600 if (id >= BLK_RING_SIZE(info)) {
1601 pr_alert("%s: response has incorrect id (%ld)\n",
1602 info->gd->disk_name, id);
1605 if (rinfo->shadow[id].status != REQ_WAITING) {
1606 pr_alert("%s: response references no pending request\n",
1607 info->gd->disk_name);
1611 rinfo->shadow[id].status = REQ_PROCESSING;
1612 req = rinfo->shadow[id].request;
1614 op = rinfo->shadow[id].req.operation;
1615 if (op == BLKIF_OP_INDIRECT)
1616 op = rinfo->shadow[id].req.u.indirect.indirect_op;
1617 if (bret.operation != op) {
1618 pr_alert("%s: response has wrong operation (%u instead of %u)\n",
1619 info->gd->disk_name, bret.operation, op);
1623 if (bret.operation != BLKIF_OP_DISCARD) {
1625 * We may need to wait for an extra response if the
1626 * I/O request is split in 2
1628 if (!blkif_completion(&id, rinfo, &bret))
1632 if (add_id_to_freelist(rinfo, id)) {
1633 WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1634 info->gd->disk_name, op_name(bret.operation), id);
1638 if (bret.status == BLKIF_RSP_OKAY)
1639 blkif_req(req)->error = BLK_STS_OK;
1641 blkif_req(req)->error = BLK_STS_IOERR;
1643 switch (bret.operation) {
1644 case BLKIF_OP_DISCARD:
1645 if (unlikely(bret.status == BLKIF_RSP_EOPNOTSUPP)) {
1646 struct request_queue *rq = info->rq;
1648 pr_warn_ratelimited("blkfront: %s: %s op failed\n",
1649 info->gd->disk_name, op_name(bret.operation));
1650 blkif_req(req)->error = BLK_STS_NOTSUPP;
1651 info->feature_discard = 0;
1652 info->feature_secdiscard = 0;
1653 queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1654 queue_flag_clear(QUEUE_FLAG_SECERASE, rq);
1657 case BLKIF_OP_FLUSH_DISKCACHE:
1658 case BLKIF_OP_WRITE_BARRIER:
1659 if (unlikely(bret.status == BLKIF_RSP_EOPNOTSUPP)) {
1660 pr_warn_ratelimited("blkfront: %s: %s op failed\n",
1661 info->gd->disk_name, op_name(bret.operation));
1662 blkif_req(req)->error = BLK_STS_NOTSUPP;
1664 if (unlikely(bret.status == BLKIF_RSP_ERROR &&
1665 rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1666 pr_warn_ratelimited("blkfront: %s: empty %s op failed\n",
1667 info->gd->disk_name, op_name(bret.operation));
1668 blkif_req(req)->error = BLK_STS_NOTSUPP;
1670 if (unlikely(blkif_req(req)->error)) {
1671 if (blkif_req(req)->error == BLK_STS_NOTSUPP)
1672 blkif_req(req)->error = BLK_STS_OK;
1673 info->feature_fua = 0;
1674 info->feature_flush = 0;
1679 case BLKIF_OP_WRITE:
1680 if (unlikely(bret.status != BLKIF_RSP_OKAY))
1681 dev_dbg_ratelimited(&info->xbdev->dev,
1682 "Bad return from blkdev data request: %#x\n",
1690 blk_mq_complete_request(req);
1693 rinfo->ring.rsp_cons = i;
1695 if (i != rinfo->ring.req_prod_pvt) {
1697 RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1701 rinfo->ring.sring->rsp_event = i + 1;
1703 kick_pending_request_queues_locked(rinfo);
1705 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1707 xen_irq_lateeoi(irq, eoiflag);
1712 info->connected = BLKIF_STATE_ERROR;
1714 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1716 /* No EOI in order to avoid further interrupts. */
1718 pr_alert("%s disabled for further use\n", info->gd->disk_name);
1723 static int setup_blkring(struct xenbus_device *dev,
1724 struct blkfront_ring_info *rinfo)
1726 struct blkif_sring *sring;
1728 struct blkfront_info *info = rinfo->dev_info;
1729 unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1730 grant_ref_t gref[XENBUS_MAX_RING_GRANTS];
1732 for (i = 0; i < info->nr_ring_pages; i++)
1733 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1735 sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
1736 get_order(ring_size));
1738 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1741 SHARED_RING_INIT(sring);
1742 FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1744 err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref);
1746 free_pages((unsigned long)sring, get_order(ring_size));
1747 rinfo->ring.sring = NULL;
1750 for (i = 0; i < info->nr_ring_pages; i++)
1751 rinfo->ring_ref[i] = gref[i];
1753 err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1757 err = bind_evtchn_to_irqhandler_lateeoi(rinfo->evtchn, blkif_interrupt,
1760 xenbus_dev_fatal(dev, err,
1761 "bind_evtchn_to_irqhandler failed");
1768 blkif_free(info, 0);
1773 * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1774 * ring buffer may have multi pages depending on ->nr_ring_pages.
1776 static int write_per_ring_nodes(struct xenbus_transaction xbt,
1777 struct blkfront_ring_info *rinfo, const char *dir)
1781 const char *message = NULL;
1782 struct blkfront_info *info = rinfo->dev_info;
1784 if (info->nr_ring_pages == 1) {
1785 err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1787 message = "writing ring-ref";
1788 goto abort_transaction;
1791 for (i = 0; i < info->nr_ring_pages; i++) {
1792 char ring_ref_name[RINGREF_NAME_LEN];
1794 snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1795 err = xenbus_printf(xbt, dir, ring_ref_name,
1796 "%u", rinfo->ring_ref[i]);
1798 message = "writing ring-ref";
1799 goto abort_transaction;
1804 err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1806 message = "writing event-channel";
1807 goto abort_transaction;
1813 xenbus_transaction_end(xbt, 1);
1815 xenbus_dev_fatal(info->xbdev, err, "%s", message);
1820 /* Common code used when first setting up, and when resuming. */
1821 static int talk_to_blkback(struct xenbus_device *dev,
1822 struct blkfront_info *info)
1824 const char *message = NULL;
1825 struct xenbus_transaction xbt;
1827 unsigned int i, max_page_order;
1828 unsigned int ring_page_order;
1833 max_page_order = xenbus_read_unsigned(info->xbdev->otherend,
1834 "max-ring-page-order", 0);
1835 ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1836 info->nr_ring_pages = 1 << ring_page_order;
1838 err = negotiate_mq(info);
1840 goto destroy_blkring;
1842 for (i = 0; i < info->nr_rings; i++) {
1843 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1845 /* Create shared ring, alloc event channel. */
1846 err = setup_blkring(dev, rinfo);
1848 goto destroy_blkring;
1852 err = xenbus_transaction_start(&xbt);
1854 xenbus_dev_fatal(dev, err, "starting transaction");
1855 goto destroy_blkring;
1858 if (info->nr_ring_pages > 1) {
1859 err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1862 message = "writing ring-page-order";
1863 goto abort_transaction;
1867 /* We already got the number of queues/rings in _probe */
1868 if (info->nr_rings == 1) {
1869 err = write_per_ring_nodes(xbt, &info->rinfo[0], dev->nodename);
1871 goto destroy_blkring;
1876 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1879 message = "writing multi-queue-num-queues";
1880 goto abort_transaction;
1883 pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1884 path = kmalloc(pathsize, GFP_KERNEL);
1887 message = "ENOMEM while writing ring references";
1888 goto abort_transaction;
1891 for (i = 0; i < info->nr_rings; i++) {
1892 memset(path, 0, pathsize);
1893 snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1894 err = write_per_ring_nodes(xbt, &info->rinfo[i], path);
1897 goto destroy_blkring;
1902 err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1903 XEN_IO_PROTO_ABI_NATIVE);
1905 message = "writing protocol";
1906 goto abort_transaction;
1908 err = xenbus_printf(xbt, dev->nodename,
1909 "feature-persistent", "%u", 1);
1912 "writing persistent grants feature to xenbus");
1914 err = xenbus_transaction_end(xbt, 0);
1918 xenbus_dev_fatal(dev, err, "completing transaction");
1919 goto destroy_blkring;
1922 for (i = 0; i < info->nr_rings; i++) {
1924 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1926 for (j = 0; j < BLK_RING_SIZE(info); j++)
1927 rinfo->shadow[j].req.u.rw.id = j + 1;
1928 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1930 xenbus_switch_state(dev, XenbusStateInitialised);
1935 xenbus_transaction_end(xbt, 1);
1937 xenbus_dev_fatal(dev, err, "%s", message);
1939 blkif_free(info, 0);
1942 dev_set_drvdata(&dev->dev, NULL);
1947 static int negotiate_mq(struct blkfront_info *info)
1949 unsigned int backend_max_queues;
1952 BUG_ON(info->nr_rings);
1954 /* Check if backend supports multiple queues. */
1955 backend_max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1956 "multi-queue-max-queues", 1);
1957 info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1958 /* We need at least one ring. */
1959 if (!info->nr_rings)
1962 info->rinfo = kzalloc(sizeof(struct blkfront_ring_info) * info->nr_rings, GFP_KERNEL);
1964 xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1969 for (i = 0; i < info->nr_rings; i++) {
1970 struct blkfront_ring_info *rinfo;
1972 rinfo = &info->rinfo[i];
1973 INIT_LIST_HEAD(&rinfo->indirect_pages);
1974 INIT_LIST_HEAD(&rinfo->grants);
1975 rinfo->dev_info = info;
1976 INIT_WORK(&rinfo->work, blkif_restart_queue);
1977 spin_lock_init(&rinfo->ring_lock);
1982 * Entry point to this code when a new device is created. Allocate the basic
1983 * structures and the ring buffer for communication with the backend, and
1984 * inform the backend of the appropriate details for those. Switch to
1985 * Initialised state.
1987 static int blkfront_probe(struct xenbus_device *dev,
1988 const struct xenbus_device_id *id)
1991 struct blkfront_info *info;
1993 /* FIXME: Use dynamic device id if this is not set. */
1994 err = xenbus_scanf(XBT_NIL, dev->nodename,
1995 "virtual-device", "%i", &vdevice);
1997 /* go looking in the extended area instead */
1998 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
2001 xenbus_dev_fatal(dev, err, "reading virtual-device");
2006 if (xen_hvm_domain()) {
2009 /* no unplug has been done: do not hook devices != xen vbds */
2010 if (xen_has_pv_and_legacy_disk_devices()) {
2013 if (!VDEV_IS_EXTENDED(vdevice))
2014 major = BLKIF_MAJOR(vdevice);
2016 major = XENVBD_MAJOR;
2018 if (major != XENVBD_MAJOR) {
2020 "%s: HVM does not support vbd %d as xen block device\n",
2025 /* do not create a PV cdrom device if we are an HVM guest */
2026 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
2029 if (strncmp(type, "cdrom", 5) == 0) {
2035 info = kzalloc(sizeof(*info), GFP_KERNEL);
2037 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
2043 mutex_init(&info->mutex);
2044 info->vdevice = vdevice;
2045 info->connected = BLKIF_STATE_DISCONNECTED;
2047 /* Front end dir is a number, which is used as the id. */
2048 info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
2049 dev_set_drvdata(&dev->dev, info);
2054 static int blkif_recover(struct blkfront_info *info)
2056 unsigned int r_index;
2057 struct request *req, *n;
2062 blkfront_gather_backend_features(info);
2063 /* Reset limits changed by blk_mq_update_nr_hw_queues(). */
2064 blkif_set_queue_limits(info);
2065 segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
2066 blk_queue_max_segments(info->rq, segs / GRANTS_PER_PSEG);
2068 for (r_index = 0; r_index < info->nr_rings; r_index++) {
2069 struct blkfront_ring_info *rinfo = &info->rinfo[r_index];
2071 rc = blkfront_setup_indirect(rinfo);
2075 xenbus_switch_state(info->xbdev, XenbusStateConnected);
2077 /* Now safe for us to use the shared ring */
2078 info->connected = BLKIF_STATE_CONNECTED;
2080 for (r_index = 0; r_index < info->nr_rings; r_index++) {
2081 struct blkfront_ring_info *rinfo;
2083 rinfo = &info->rinfo[r_index];
2084 /* Kick any other new requests queued since we resumed */
2085 kick_pending_request_queues(rinfo);
2088 list_for_each_entry_safe(req, n, &info->requests, queuelist) {
2089 /* Requeue pending requests (flush or discard) */
2090 list_del_init(&req->queuelist);
2091 BUG_ON(req->nr_phys_segments > segs);
2092 blk_mq_requeue_request(req, false);
2094 blk_mq_start_stopped_hw_queues(info->rq, true);
2095 blk_mq_kick_requeue_list(info->rq);
2097 while ((bio = bio_list_pop(&info->bio_list)) != NULL) {
2098 /* Traverse the list of pending bios and re-queue them */
2106 * We are reconnecting to the backend, due to a suspend/resume, or a backend
2107 * driver restart. We tear down our blkif structure and recreate it, but
2108 * leave the device-layer structures intact so that this is transparent to the
2109 * rest of the kernel.
2111 static int blkfront_resume(struct xenbus_device *dev)
2113 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2117 dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2119 bio_list_init(&info->bio_list);
2120 INIT_LIST_HEAD(&info->requests);
2121 for (i = 0; i < info->nr_rings; i++) {
2122 struct blkfront_ring_info *rinfo = &info->rinfo[i];
2123 struct bio_list merge_bio;
2124 struct blk_shadow *shadow = rinfo->shadow;
2126 for (j = 0; j < BLK_RING_SIZE(info); j++) {
2128 if (!shadow[j].request)
2132 * Get the bios in the request so we can re-queue them.
2134 if (req_op(shadow[j].request) == REQ_OP_FLUSH ||
2135 req_op(shadow[j].request) == REQ_OP_DISCARD ||
2136 req_op(shadow[j].request) == REQ_OP_SECURE_ERASE ||
2137 shadow[j].request->cmd_flags & REQ_FUA) {
2139 * Flush operations don't contain bios, so
2140 * we need to requeue the whole request
2142 * XXX: but this doesn't make any sense for a
2143 * write with the FUA flag set..
2145 list_add(&shadow[j].request->queuelist, &info->requests);
2148 merge_bio.head = shadow[j].request->bio;
2149 merge_bio.tail = shadow[j].request->biotail;
2150 bio_list_merge(&info->bio_list, &merge_bio);
2151 shadow[j].request->bio = NULL;
2152 blk_mq_end_request(shadow[j].request, BLK_STS_OK);
2156 blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2158 err = talk_to_blkback(dev, info);
2160 blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings);
2163 * We have to wait for the backend to switch to
2164 * connected state, since we want to read which
2165 * features it supports.
2171 static void blkfront_closing(struct blkfront_info *info)
2173 struct xenbus_device *xbdev = info->xbdev;
2174 struct block_device *bdev = NULL;
2176 mutex_lock(&info->mutex);
2178 if (xbdev->state == XenbusStateClosing) {
2179 mutex_unlock(&info->mutex);
2184 bdev = bdget_disk(info->gd, 0);
2186 mutex_unlock(&info->mutex);
2189 xenbus_frontend_closed(xbdev);
2193 mutex_lock(&bdev->bd_mutex);
2195 if (bdev->bd_openers) {
2196 xenbus_dev_error(xbdev, -EBUSY,
2197 "Device in use; refusing to close");
2198 xenbus_switch_state(xbdev, XenbusStateClosing);
2200 xlvbd_release_gendisk(info);
2201 xenbus_frontend_closed(xbdev);
2204 mutex_unlock(&bdev->bd_mutex);
2208 static void blkfront_setup_discard(struct blkfront_info *info)
2210 info->feature_discard = 1;
2211 info->discard_granularity = xenbus_read_unsigned(info->xbdev->otherend,
2212 "discard-granularity",
2214 info->discard_alignment = xenbus_read_unsigned(info->xbdev->otherend,
2215 "discard-alignment", 0);
2216 info->feature_secdiscard =
2217 !!xenbus_read_unsigned(info->xbdev->otherend, "discard-secure",
2221 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2223 unsigned int psegs, grants;
2225 struct blkfront_info *info = rinfo->dev_info;
2227 if (info->max_indirect_segments == 0) {
2229 grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2232 * When an extra req is required, the maximum
2233 * grants supported is related to the size of the
2234 * Linux block segment.
2236 grants = GRANTS_PER_PSEG;
2240 grants = info->max_indirect_segments;
2241 psegs = DIV_ROUND_UP(grants, GRANTS_PER_PSEG);
2243 err = fill_grant_buffer(rinfo,
2244 (grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2248 if (!info->feature_persistent && info->max_indirect_segments) {
2250 * We are using indirect descriptors but not persistent
2251 * grants, we need to allocate a set of pages that can be
2252 * used for mapping indirect grefs
2254 int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2256 BUG_ON(!list_empty(&rinfo->indirect_pages));
2257 for (i = 0; i < num; i++) {
2258 struct page *indirect_page = alloc_page(GFP_NOIO);
2261 list_add(&indirect_page->lru, &rinfo->indirect_pages);
2265 for (i = 0; i < BLK_RING_SIZE(info); i++) {
2266 rinfo->shadow[i].grants_used = kzalloc(
2267 sizeof(rinfo->shadow[i].grants_used[0]) * grants,
2269 rinfo->shadow[i].sg = kzalloc(sizeof(rinfo->shadow[i].sg[0]) * psegs, GFP_NOIO);
2270 if (info->max_indirect_segments)
2271 rinfo->shadow[i].indirect_grants = kzalloc(
2272 sizeof(rinfo->shadow[i].indirect_grants[0]) *
2273 INDIRECT_GREFS(grants),
2275 if ((rinfo->shadow[i].grants_used == NULL) ||
2276 (rinfo->shadow[i].sg == NULL) ||
2277 (info->max_indirect_segments &&
2278 (rinfo->shadow[i].indirect_grants == NULL)))
2280 sg_init_table(rinfo->shadow[i].sg, psegs);
2287 for (i = 0; i < BLK_RING_SIZE(info); i++) {
2288 kfree(rinfo->shadow[i].grants_used);
2289 rinfo->shadow[i].grants_used = NULL;
2290 kfree(rinfo->shadow[i].sg);
2291 rinfo->shadow[i].sg = NULL;
2292 kfree(rinfo->shadow[i].indirect_grants);
2293 rinfo->shadow[i].indirect_grants = NULL;
2295 if (!list_empty(&rinfo->indirect_pages)) {
2296 struct page *indirect_page, *n;
2297 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2298 list_del(&indirect_page->lru);
2299 __free_page(indirect_page);
2306 * Gather all backend feature-*
2308 static void blkfront_gather_backend_features(struct blkfront_info *info)
2310 unsigned int indirect_segments;
2312 info->feature_flush = 0;
2313 info->feature_fua = 0;
2316 * If there's no "feature-barrier" defined, then it means
2317 * we're dealing with a very old backend which writes
2318 * synchronously; nothing to do.
2320 * If there are barriers, then we use flush.
2322 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-barrier", 0)) {
2323 info->feature_flush = 1;
2324 info->feature_fua = 1;
2328 * And if there is "feature-flush-cache" use that above
2331 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-flush-cache",
2333 info->feature_flush = 1;
2334 info->feature_fua = 0;
2337 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-discard", 0))
2338 blkfront_setup_discard(info);
2340 info->feature_persistent =
2341 !!xenbus_read_unsigned(info->xbdev->otherend,
2342 "feature-persistent", 0);
2344 indirect_segments = xenbus_read_unsigned(info->xbdev->otherend,
2345 "feature-max-indirect-segments", 0);
2346 if (indirect_segments > xen_blkif_max_segments)
2347 indirect_segments = xen_blkif_max_segments;
2348 if (indirect_segments <= BLKIF_MAX_SEGMENTS_PER_REQUEST)
2349 indirect_segments = 0;
2350 info->max_indirect_segments = indirect_segments;
2354 * Invoked when the backend is finally 'ready' (and has told produced
2355 * the details about the physical device - #sectors, size, etc).
2357 static void blkfront_connect(struct blkfront_info *info)
2359 unsigned long long sectors;
2360 unsigned long sector_size;
2361 unsigned int physical_sector_size;
2363 char *envp[] = { "RESIZE=1", NULL };
2366 switch (info->connected) {
2367 case BLKIF_STATE_CONNECTED:
2369 * Potentially, the back-end may be signalling
2370 * a capacity change; update the capacity.
2372 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2373 "sectors", "%Lu", §ors);
2374 if (XENBUS_EXIST_ERR(err))
2376 printk(KERN_INFO "Setting capacity to %Lu\n",
2378 set_capacity(info->gd, sectors);
2379 revalidate_disk(info->gd);
2380 kobject_uevent_env(&disk_to_dev(info->gd)->kobj,
2384 case BLKIF_STATE_SUSPENDED:
2386 * If we are recovering from suspension, we need to wait
2387 * for the backend to announce it's features before
2388 * reconnecting, at least we need to know if the backend
2389 * supports indirect descriptors, and how many.
2391 blkif_recover(info);
2398 dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2399 __func__, info->xbdev->otherend);
2401 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2402 "sectors", "%llu", §ors,
2403 "info", "%u", &binfo,
2404 "sector-size", "%lu", §or_size,
2407 xenbus_dev_fatal(info->xbdev, err,
2408 "reading backend fields at %s",
2409 info->xbdev->otherend);
2414 * physcial-sector-size is a newer field, so old backends may not
2415 * provide this. Assume physical sector size to be the same as
2416 * sector_size in that case.
2418 physical_sector_size = xenbus_read_unsigned(info->xbdev->otherend,
2419 "physical-sector-size",
2421 blkfront_gather_backend_features(info);
2422 for (i = 0; i < info->nr_rings; i++) {
2423 err = blkfront_setup_indirect(&info->rinfo[i]);
2425 xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2426 info->xbdev->otherend);
2427 blkif_free(info, 0);
2432 err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
2433 physical_sector_size);
2435 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2436 info->xbdev->otherend);
2440 xenbus_switch_state(info->xbdev, XenbusStateConnected);
2442 /* Kick pending requests. */
2443 info->connected = BLKIF_STATE_CONNECTED;
2444 for (i = 0; i < info->nr_rings; i++)
2445 kick_pending_request_queues(&info->rinfo[i]);
2447 device_add_disk(&info->xbdev->dev, info->gd);
2453 blkif_free(info, 0);
2458 * Callback received when the backend's state changes.
2460 static void blkback_changed(struct xenbus_device *dev,
2461 enum xenbus_state backend_state)
2463 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2465 dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2467 switch (backend_state) {
2468 case XenbusStateInitWait:
2469 if (dev->state != XenbusStateInitialising)
2471 if (talk_to_blkback(dev, info))
2473 case XenbusStateInitialising:
2474 case XenbusStateInitialised:
2475 case XenbusStateReconfiguring:
2476 case XenbusStateReconfigured:
2477 case XenbusStateUnknown:
2480 case XenbusStateConnected:
2482 * talk_to_blkback sets state to XenbusStateInitialised
2483 * and blkfront_connect sets it to XenbusStateConnected
2484 * (if connection went OK).
2486 * If the backend (or toolstack) decides to poke at backend
2487 * state (and re-trigger the watch by setting the state repeatedly
2488 * to XenbusStateConnected (4)) we need to deal with this.
2489 * This is allowed as this is used to communicate to the guest
2490 * that the size of disk has changed!
2492 if ((dev->state != XenbusStateInitialised) &&
2493 (dev->state != XenbusStateConnected)) {
2494 if (talk_to_blkback(dev, info))
2498 blkfront_connect(info);
2501 case XenbusStateClosed:
2502 if (dev->state == XenbusStateClosed)
2505 case XenbusStateClosing:
2507 blkfront_closing(info);
2512 static int blkfront_remove(struct xenbus_device *xbdev)
2514 struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2515 struct block_device *bdev = NULL;
2516 struct gendisk *disk;
2518 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");