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 BLKIF_STATE_DISCONNECTED,
65 BLKIF_STATE_CONNECTED,
66 BLKIF_STATE_SUSPENDED,
72 struct list_head node;
76 struct blkif_request req;
77 struct request *request;
78 struct grant **grants_used;
79 struct grant **indirect_grants;
80 struct scatterlist *sg;
88 static DEFINE_MUTEX(blkfront_mutex);
89 static const struct block_device_operations xlvbd_block_fops;
92 * Maximum number of segments in indirect requests, the actual value used by
93 * the frontend driver is the minimum of this value and the value provided
94 * by the backend driver.
97 static unsigned int xen_blkif_max_segments = 32;
98 module_param_named(max, xen_blkif_max_segments, int, S_IRUGO);
99 MODULE_PARM_DESC(max, "Maximum amount of segments in indirect requests (default is 32)");
102 * Maximum order of pages to be used for the shared ring between front and
103 * backend, 4KB page granularity is used.
105 static unsigned int xen_blkif_max_ring_order;
106 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, S_IRUGO);
107 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
109 #define BLK_RING_SIZE(info) __CONST_RING_SIZE(blkif, PAGE_SIZE * (info)->nr_ring_pages)
110 #define BLK_MAX_RING_SIZE __CONST_RING_SIZE(blkif, PAGE_SIZE * XENBUS_MAX_RING_PAGES)
112 * ring-ref%i i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
113 * characters are enough. Define to 20 to keep consist with backend.
115 #define RINGREF_NAME_LEN (20)
118 * We have one of these per vbd, whether ide, scsi or 'other'. They
119 * hang in private_data off the gendisk structure. We may end up
120 * putting all kinds of interesting stuff here :-)
126 struct xenbus_device *xbdev;
130 enum blkif_state connected;
131 int ring_ref[XENBUS_MAX_RING_PAGES];
132 unsigned int nr_ring_pages;
133 struct blkif_front_ring ring;
134 unsigned int evtchn, irq;
135 struct request_queue *rq;
136 struct work_struct work;
137 struct gnttab_free_callback callback;
138 struct blk_shadow shadow[BLK_MAX_RING_SIZE];
139 struct list_head grants;
140 struct list_head indirect_pages;
141 unsigned int persistent_gnts_c;
142 unsigned long shadow_free;
143 unsigned int feature_flush;
144 unsigned int feature_discard:1;
145 unsigned int feature_secdiscard:1;
146 unsigned int discard_granularity;
147 unsigned int discard_alignment;
148 unsigned int feature_persistent:1;
149 unsigned int max_indirect_segments;
151 struct blk_mq_tag_set tag_set;
154 static unsigned int nr_minors;
155 static unsigned long *minors;
156 static DEFINE_SPINLOCK(minor_lock);
158 #define GRANT_INVALID_REF 0
160 #define PARTS_PER_DISK 16
161 #define PARTS_PER_EXT_DISK 256
163 #define BLKIF_MAJOR(dev) ((dev)>>8)
164 #define BLKIF_MINOR(dev) ((dev) & 0xff)
167 #define EXTENDED (1<<EXT_SHIFT)
168 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
169 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
170 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
171 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
172 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
173 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
175 #define DEV_NAME "xvd" /* name in /dev */
177 #define SEGS_PER_INDIRECT_FRAME \
178 (PAGE_SIZE/sizeof(struct blkif_request_segment))
179 #define INDIRECT_GREFS(_segs) \
180 ((_segs + SEGS_PER_INDIRECT_FRAME - 1)/SEGS_PER_INDIRECT_FRAME)
182 static int blkfront_setup_indirect(struct blkfront_info *info);
183 static int blkfront_gather_backend_features(struct blkfront_info *info);
185 static int get_id_from_freelist(struct blkfront_info *info)
187 unsigned long free = info->shadow_free;
188 BUG_ON(free >= BLK_RING_SIZE(info));
189 info->shadow_free = info->shadow[free].req.u.rw.id;
190 info->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
194 static int add_id_to_freelist(struct blkfront_info *info,
197 if (info->shadow[id].req.u.rw.id != id)
199 if (info->shadow[id].request == NULL)
201 info->shadow[id].req.u.rw.id = info->shadow_free;
202 info->shadow[id].request = NULL;
203 info->shadow_free = id;
207 static int fill_grant_buffer(struct blkfront_info *info, int num)
209 struct page *granted_page;
210 struct grant *gnt_list_entry, *n;
214 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
218 if (info->feature_persistent) {
219 granted_page = alloc_page(GFP_NOIO);
221 kfree(gnt_list_entry);
224 gnt_list_entry->page = granted_page;
227 gnt_list_entry->gref = GRANT_INVALID_REF;
228 list_add(&gnt_list_entry->node, &info->grants);
235 list_for_each_entry_safe(gnt_list_entry, n,
236 &info->grants, node) {
237 list_del(&gnt_list_entry->node);
238 if (info->feature_persistent)
239 __free_page(gnt_list_entry->page);
240 kfree(gnt_list_entry);
247 static struct grant *get_grant(grant_ref_t *gref_head,
249 struct blkfront_info *info)
251 struct grant *gnt_list_entry;
252 unsigned long buffer_gfn;
254 BUG_ON(list_empty(&info->grants));
255 gnt_list_entry = list_first_entry(&info->grants, struct grant,
257 list_del(&gnt_list_entry->node);
259 if (gnt_list_entry->gref != GRANT_INVALID_REF) {
260 info->persistent_gnts_c--;
261 return gnt_list_entry;
264 /* Assign a gref to this page */
265 gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
266 BUG_ON(gnt_list_entry->gref == -ENOSPC);
267 if (!info->feature_persistent) {
269 gnt_list_entry->page = page;
271 buffer_gfn = xen_page_to_gfn(gnt_list_entry->page);
272 gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
273 info->xbdev->otherend_id,
275 return gnt_list_entry;
278 static const char *op_name(int op)
280 static const char *const names[] = {
281 [BLKIF_OP_READ] = "read",
282 [BLKIF_OP_WRITE] = "write",
283 [BLKIF_OP_WRITE_BARRIER] = "barrier",
284 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
285 [BLKIF_OP_DISCARD] = "discard" };
287 if (op < 0 || op >= ARRAY_SIZE(names))
295 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
297 unsigned int end = minor + nr;
300 if (end > nr_minors) {
301 unsigned long *bitmap, *old;
303 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
308 spin_lock(&minor_lock);
309 if (end > nr_minors) {
311 memcpy(bitmap, minors,
312 BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
314 nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
317 spin_unlock(&minor_lock);
321 spin_lock(&minor_lock);
322 if (find_next_bit(minors, end, minor) >= end) {
323 bitmap_set(minors, minor, nr);
327 spin_unlock(&minor_lock);
332 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
334 unsigned int end = minor + nr;
336 BUG_ON(end > nr_minors);
337 spin_lock(&minor_lock);
338 bitmap_clear(minors, minor, nr);
339 spin_unlock(&minor_lock);
342 static void blkif_restart_queue_callback(void *arg)
344 struct blkfront_info *info = (struct blkfront_info *)arg;
345 schedule_work(&info->work);
348 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
350 /* We don't have real geometry info, but let's at least return
351 values consistent with the size of the device */
352 sector_t nsect = get_capacity(bd->bd_disk);
353 sector_t cylinders = nsect;
357 sector_div(cylinders, hg->heads * hg->sectors);
358 hg->cylinders = cylinders;
359 if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
360 hg->cylinders = 0xffff;
364 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
365 unsigned command, unsigned long argument)
367 struct blkfront_info *info = bdev->bd_disk->private_data;
370 dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
371 command, (long)argument);
374 case CDROMMULTISESSION:
375 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
376 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
377 if (put_user(0, (char __user *)(argument + i)))
381 case CDROM_GET_CAPABILITY: {
382 struct gendisk *gd = info->gd;
383 if (gd->flags & GENHD_FL_CD)
389 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
391 return -EINVAL; /* same return as native Linux */
397 static int blkif_queue_discard_req(struct request *req)
399 struct blkfront_info *info = req->rq_disk->private_data;
400 struct blkif_request *ring_req;
403 /* Fill out a communications ring structure. */
404 ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
405 id = get_id_from_freelist(info);
406 info->shadow[id].request = req;
408 ring_req->operation = BLKIF_OP_DISCARD;
409 ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
410 ring_req->u.discard.id = id;
411 ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
412 if ((req->cmd_flags & REQ_SECURE) && info->feature_secdiscard)
413 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
415 ring_req->u.discard.flag = 0;
417 info->ring.req_prod_pvt++;
419 /* Keep a private copy so we can reissue requests when recovering. */
420 info->shadow[id].req = *ring_req;
425 static int blkif_queue_rw_req(struct request *req)
427 struct blkfront_info *info = req->rq_disk->private_data;
428 struct blkif_request *ring_req;
430 unsigned int fsect, lsect;
432 struct blkif_request_segment *segments = NULL;
435 * Used to store if we are able to queue the request by just using
436 * existing persistent grants, or if we have to get new grants,
437 * as there are not sufficiently many free.
439 bool new_persistent_gnts;
440 grant_ref_t gref_head;
441 struct grant *gnt_list_entry = NULL;
442 struct scatterlist *sg;
445 max_grefs = req->nr_phys_segments;
446 if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
448 * If we are using indirect segments we need to account
449 * for the indirect grefs used in the request.
451 max_grefs += INDIRECT_GREFS(req->nr_phys_segments);
453 /* Check if we have enough grants to allocate a requests */
454 if (info->persistent_gnts_c < max_grefs) {
455 new_persistent_gnts = 1;
456 if (gnttab_alloc_grant_references(
457 max_grefs - info->persistent_gnts_c,
459 gnttab_request_free_callback(
461 blkif_restart_queue_callback,
467 new_persistent_gnts = 0;
469 /* Fill out a communications ring structure. */
470 ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
471 id = get_id_from_freelist(info);
472 info->shadow[id].request = req;
474 BUG_ON(info->max_indirect_segments == 0 &&
475 req->nr_phys_segments > BLKIF_MAX_SEGMENTS_PER_REQUEST);
476 BUG_ON(info->max_indirect_segments &&
477 req->nr_phys_segments > info->max_indirect_segments);
478 nseg = blk_rq_map_sg(req->q, req, info->shadow[id].sg);
479 ring_req->u.rw.id = id;
480 if (nseg > BLKIF_MAX_SEGMENTS_PER_REQUEST) {
482 * The indirect operation can only be a BLKIF_OP_READ or
485 BUG_ON(req->cmd_flags & (REQ_FLUSH | REQ_FUA));
486 ring_req->operation = BLKIF_OP_INDIRECT;
487 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
488 BLKIF_OP_WRITE : BLKIF_OP_READ;
489 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
490 ring_req->u.indirect.handle = info->handle;
491 ring_req->u.indirect.nr_segments = nseg;
493 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
494 ring_req->u.rw.handle = info->handle;
495 ring_req->operation = rq_data_dir(req) ?
496 BLKIF_OP_WRITE : BLKIF_OP_READ;
497 if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) {
499 * Ideally we can do an unordered flush-to-disk.
500 * In case the backend onlysupports barriers, use that.
501 * A barrier request a superset of FUA, so we can
502 * implement it the same way. (It's also a FLUSH+FUA,
503 * since it is guaranteed ordered WRT previous writes.)
505 switch (info->feature_flush &
506 ((REQ_FLUSH|REQ_FUA))) {
507 case REQ_FLUSH|REQ_FUA:
508 ring_req->operation =
509 BLKIF_OP_WRITE_BARRIER;
512 ring_req->operation =
513 BLKIF_OP_FLUSH_DISKCACHE;
516 ring_req->operation = 0;
519 ring_req->u.rw.nr_segments = nseg;
521 for_each_sg(info->shadow[id].sg, sg, nseg, i) {
522 fsect = sg->offset >> 9;
523 lsect = fsect + (sg->length >> 9) - 1;
525 if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
526 (i % SEGS_PER_INDIRECT_FRAME == 0)) {
527 struct page *uninitialized_var(page);
530 kunmap_atomic(segments);
532 n = i / SEGS_PER_INDIRECT_FRAME;
533 if (!info->feature_persistent) {
534 struct page *indirect_page;
537 * Fetch a pre-allocated page to use for
540 BUG_ON(list_empty(&info->indirect_pages));
541 indirect_page = list_first_entry(&info->indirect_pages,
543 list_del(&indirect_page->lru);
544 page = indirect_page;
546 gnt_list_entry = get_grant(&gref_head, page, info);
547 info->shadow[id].indirect_grants[n] = gnt_list_entry;
548 segments = kmap_atomic(gnt_list_entry->page);
549 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
552 gnt_list_entry = get_grant(&gref_head, sg_page(sg), info);
553 ref = gnt_list_entry->gref;
555 info->shadow[id].grants_used[i] = gnt_list_entry;
557 if (rq_data_dir(req) && info->feature_persistent) {
561 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
563 shared_data = kmap_atomic(gnt_list_entry->page);
564 bvec_data = kmap_atomic(sg_page(sg));
567 * this does not wipe data stored outside the
568 * range sg->offset..sg->offset+sg->length.
569 * Therefore, blkback *could* see data from
570 * previous requests. This is OK as long as
571 * persistent grants are shared with just one
572 * domain. It may need refactoring if this
575 memcpy(shared_data + sg->offset,
576 bvec_data + sg->offset,
579 kunmap_atomic(bvec_data);
580 kunmap_atomic(shared_data);
582 if (ring_req->operation != BLKIF_OP_INDIRECT) {
583 ring_req->u.rw.seg[i] =
584 (struct blkif_request_segment) {
587 .last_sect = lsect };
589 n = i % SEGS_PER_INDIRECT_FRAME;
591 (struct blkif_request_segment) {
594 .last_sect = lsect };
598 kunmap_atomic(segments);
600 info->ring.req_prod_pvt++;
602 /* Keep a private copy so we can reissue requests when recovering. */
603 info->shadow[id].req = *ring_req;
605 if (new_persistent_gnts)
606 gnttab_free_grant_references(gref_head);
612 * Generate a Xen blkfront IO request from a blk layer request. Reads
613 * and writes are handled as expected.
615 * @req: a request struct
617 static int blkif_queue_request(struct request *req)
619 struct blkfront_info *info = req->rq_disk->private_data;
621 if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
624 if (unlikely(req->cmd_flags & (REQ_DISCARD | REQ_SECURE)))
625 return blkif_queue_discard_req(req);
627 return blkif_queue_rw_req(req);
630 static inline void flush_requests(struct blkfront_info *info)
634 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
637 notify_remote_via_irq(info->irq);
640 static inline bool blkif_request_flush_invalid(struct request *req,
641 struct blkfront_info *info)
643 return ((req->cmd_type != REQ_TYPE_FS) ||
644 ((req->cmd_flags & REQ_FLUSH) &&
645 !(info->feature_flush & REQ_FLUSH)) ||
646 ((req->cmd_flags & REQ_FUA) &&
647 !(info->feature_flush & REQ_FUA)));
650 static int blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
651 const struct blk_mq_queue_data *qd)
653 struct blkfront_info *info = qd->rq->rq_disk->private_data;
655 blk_mq_start_request(qd->rq);
656 spin_lock_irq(&info->io_lock);
657 if (RING_FULL(&info->ring))
660 if (blkif_request_flush_invalid(qd->rq, info))
663 if (blkif_queue_request(qd->rq))
666 flush_requests(info);
667 spin_unlock_irq(&info->io_lock);
668 return BLK_MQ_RQ_QUEUE_OK;
671 spin_unlock_irq(&info->io_lock);
672 return BLK_MQ_RQ_QUEUE_ERROR;
675 spin_unlock_irq(&info->io_lock);
676 blk_mq_stop_hw_queue(hctx);
677 return BLK_MQ_RQ_QUEUE_BUSY;
680 static struct blk_mq_ops blkfront_mq_ops = {
681 .queue_rq = blkif_queue_rq,
682 .map_queue = blk_mq_map_queue,
685 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
686 unsigned int physical_sector_size,
687 unsigned int segments)
689 struct request_queue *rq;
690 struct blkfront_info *info = gd->private_data;
692 memset(&info->tag_set, 0, sizeof(info->tag_set));
693 info->tag_set.ops = &blkfront_mq_ops;
694 info->tag_set.nr_hw_queues = 1;
695 info->tag_set.queue_depth = BLK_RING_SIZE(info);
696 info->tag_set.numa_node = NUMA_NO_NODE;
697 info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
698 info->tag_set.cmd_size = 0;
699 info->tag_set.driver_data = info;
701 if (blk_mq_alloc_tag_set(&info->tag_set))
703 rq = blk_mq_init_queue(&info->tag_set);
705 blk_mq_free_tag_set(&info->tag_set);
709 queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
711 if (info->feature_discard) {
712 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
713 blk_queue_max_discard_sectors(rq, get_capacity(gd));
714 rq->limits.discard_granularity = info->discard_granularity;
715 rq->limits.discard_alignment = info->discard_alignment;
716 if (info->feature_secdiscard)
717 queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD, rq);
720 /* Hard sector size and max sectors impersonate the equiv. hardware. */
721 blk_queue_logical_block_size(rq, sector_size);
722 blk_queue_physical_block_size(rq, physical_sector_size);
723 blk_queue_max_hw_sectors(rq, (segments * PAGE_SIZE) / 512);
725 /* Each segment in a request is up to an aligned page in size. */
726 blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
727 blk_queue_max_segment_size(rq, PAGE_SIZE);
729 /* Ensure a merged request will fit in a single I/O ring slot. */
730 blk_queue_max_segments(rq, segments);
732 /* Make sure buffer addresses are sector-aligned. */
733 blk_queue_dma_alignment(rq, 511);
735 /* Make sure we don't use bounce buffers. */
736 blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
743 static const char *flush_info(unsigned int feature_flush)
745 switch (feature_flush & ((REQ_FLUSH | REQ_FUA))) {
746 case REQ_FLUSH|REQ_FUA:
747 return "barrier: enabled;";
749 return "flush diskcache: enabled;";
751 return "barrier or flush: disabled;";
755 static void xlvbd_flush(struct blkfront_info *info)
757 blk_queue_flush(info->rq, info->feature_flush);
758 pr_info("blkfront: %s: %s %s %s %s %s\n",
759 info->gd->disk_name, flush_info(info->feature_flush),
760 "persistent grants:", info->feature_persistent ?
761 "enabled;" : "disabled;", "indirect descriptors:",
762 info->max_indirect_segments ? "enabled;" : "disabled;");
765 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
768 major = BLKIF_MAJOR(vdevice);
769 *minor = BLKIF_MINOR(vdevice);
772 *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
773 *minor = ((*minor / 64) * PARTS_PER_DISK) +
774 EMULATED_HD_DISK_MINOR_OFFSET;
777 *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
778 *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
779 EMULATED_HD_DISK_MINOR_OFFSET;
781 case XEN_SCSI_DISK0_MAJOR:
782 *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
783 *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
785 case XEN_SCSI_DISK1_MAJOR:
786 case XEN_SCSI_DISK2_MAJOR:
787 case XEN_SCSI_DISK3_MAJOR:
788 case XEN_SCSI_DISK4_MAJOR:
789 case XEN_SCSI_DISK5_MAJOR:
790 case XEN_SCSI_DISK6_MAJOR:
791 case XEN_SCSI_DISK7_MAJOR:
792 *offset = (*minor / PARTS_PER_DISK) +
793 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
794 EMULATED_SD_DISK_NAME_OFFSET;
796 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
797 EMULATED_SD_DISK_MINOR_OFFSET;
799 case XEN_SCSI_DISK8_MAJOR:
800 case XEN_SCSI_DISK9_MAJOR:
801 case XEN_SCSI_DISK10_MAJOR:
802 case XEN_SCSI_DISK11_MAJOR:
803 case XEN_SCSI_DISK12_MAJOR:
804 case XEN_SCSI_DISK13_MAJOR:
805 case XEN_SCSI_DISK14_MAJOR:
806 case XEN_SCSI_DISK15_MAJOR:
807 *offset = (*minor / PARTS_PER_DISK) +
808 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
809 EMULATED_SD_DISK_NAME_OFFSET;
811 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
812 EMULATED_SD_DISK_MINOR_OFFSET;
815 *offset = *minor / PARTS_PER_DISK;
818 printk(KERN_WARNING "blkfront: your disk configuration is "
819 "incorrect, please use an xvd device instead\n");
825 static char *encode_disk_name(char *ptr, unsigned int n)
828 ptr = encode_disk_name(ptr, n / 26 - 1);
833 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
834 struct blkfront_info *info,
835 u16 vdisk_info, u16 sector_size,
836 unsigned int physical_sector_size)
846 BUG_ON(info->gd != NULL);
847 BUG_ON(info->rq != NULL);
849 if ((info->vdevice>>EXT_SHIFT) > 1) {
850 /* this is above the extended range; something is wrong */
851 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
855 if (!VDEV_IS_EXTENDED(info->vdevice)) {
856 err = xen_translate_vdev(info->vdevice, &minor, &offset);
859 nr_parts = PARTS_PER_DISK;
861 minor = BLKIF_MINOR_EXT(info->vdevice);
862 nr_parts = PARTS_PER_EXT_DISK;
863 offset = minor / nr_parts;
864 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
865 printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
866 "emulated IDE disks,\n\t choose an xvd device name"
867 "from xvde on\n", info->vdevice);
869 if (minor >> MINORBITS) {
870 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
871 info->vdevice, minor);
875 if ((minor % nr_parts) == 0)
876 nr_minors = nr_parts;
878 err = xlbd_reserve_minors(minor, nr_minors);
883 gd = alloc_disk(nr_minors);
887 strcpy(gd->disk_name, DEV_NAME);
888 ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
889 BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
893 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
894 "%d", minor & (nr_parts - 1));
896 gd->major = XENVBD_MAJOR;
897 gd->first_minor = minor;
898 gd->fops = &xlvbd_block_fops;
899 gd->private_data = info;
900 gd->driverfs_dev = &(info->xbdev->dev);
901 set_capacity(gd, capacity);
903 if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size,
904 info->max_indirect_segments ? :
905 BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
910 info->rq = gd->queue;
915 if (vdisk_info & VDISK_READONLY)
918 if (vdisk_info & VDISK_REMOVABLE)
919 gd->flags |= GENHD_FL_REMOVABLE;
921 if (vdisk_info & VDISK_CDROM)
922 gd->flags |= GENHD_FL_CD;
927 xlbd_release_minors(minor, nr_minors);
932 static void xlvbd_release_gendisk(struct blkfront_info *info)
934 unsigned int minor, nr_minors;
936 if (info->rq == NULL)
939 /* No more blkif_request(). */
940 blk_mq_stop_hw_queues(info->rq);
942 /* No more gnttab callback work. */
943 gnttab_cancel_free_callback(&info->callback);
945 /* Flush gnttab callback work. Must be done with no locks held. */
946 flush_work(&info->work);
948 del_gendisk(info->gd);
950 minor = info->gd->first_minor;
951 nr_minors = info->gd->minors;
952 xlbd_release_minors(minor, nr_minors);
954 blk_cleanup_queue(info->rq);
955 blk_mq_free_tag_set(&info->tag_set);
962 /* Must be called with io_lock holded */
963 static void kick_pending_request_queues(struct blkfront_info *info)
965 if (!RING_FULL(&info->ring))
966 blk_mq_start_stopped_hw_queues(info->rq, true);
969 static void blkif_restart_queue(struct work_struct *work)
971 struct blkfront_info *info = container_of(work, struct blkfront_info, work);
973 spin_lock_irq(&info->io_lock);
974 if (info->connected == BLKIF_STATE_CONNECTED)
975 kick_pending_request_queues(info);
976 spin_unlock_irq(&info->io_lock);
979 static void blkif_free(struct blkfront_info *info, int suspend)
981 struct grant *persistent_gnt;
985 /* Prevent new requests being issued until we fix things up. */
986 spin_lock_irq(&info->io_lock);
987 info->connected = suspend ?
988 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
989 /* No more blkif_request(). */
991 blk_mq_stop_hw_queues(info->rq);
993 /* Remove all persistent grants */
994 if (!list_empty(&info->grants)) {
995 list_for_each_entry_safe(persistent_gnt, n,
996 &info->grants, node) {
997 list_del(&persistent_gnt->node);
998 if (persistent_gnt->gref != GRANT_INVALID_REF) {
999 gnttab_end_foreign_access(persistent_gnt->gref,
1001 info->persistent_gnts_c--;
1003 if (info->feature_persistent)
1004 __free_page(persistent_gnt->page);
1005 kfree(persistent_gnt);
1008 BUG_ON(info->persistent_gnts_c != 0);
1011 * Remove indirect pages, this only happens when using indirect
1012 * descriptors but not persistent grants
1014 if (!list_empty(&info->indirect_pages)) {
1015 struct page *indirect_page, *n;
1017 BUG_ON(info->feature_persistent);
1018 list_for_each_entry_safe(indirect_page, n, &info->indirect_pages, lru) {
1019 list_del(&indirect_page->lru);
1020 __free_page(indirect_page);
1024 for (i = 0; i < BLK_RING_SIZE(info); i++) {
1026 * Clear persistent grants present in requests already
1027 * on the shared ring
1029 if (!info->shadow[i].request)
1032 segs = info->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1033 info->shadow[i].req.u.indirect.nr_segments :
1034 info->shadow[i].req.u.rw.nr_segments;
1035 for (j = 0; j < segs; j++) {
1036 persistent_gnt = info->shadow[i].grants_used[j];
1037 gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1038 if (info->feature_persistent)
1039 __free_page(persistent_gnt->page);
1040 kfree(persistent_gnt);
1043 if (info->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1045 * If this is not an indirect operation don't try to
1046 * free indirect segments
1050 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1051 persistent_gnt = info->shadow[i].indirect_grants[j];
1052 gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1053 __free_page(persistent_gnt->page);
1054 kfree(persistent_gnt);
1058 kfree(info->shadow[i].grants_used);
1059 info->shadow[i].grants_used = NULL;
1060 kfree(info->shadow[i].indirect_grants);
1061 info->shadow[i].indirect_grants = NULL;
1062 kfree(info->shadow[i].sg);
1063 info->shadow[i].sg = NULL;
1066 /* No more gnttab callback work. */
1067 gnttab_cancel_free_callback(&info->callback);
1068 spin_unlock_irq(&info->io_lock);
1070 /* Flush gnttab callback work. Must be done with no locks held. */
1071 flush_work(&info->work);
1073 /* Free resources associated with old device channel. */
1074 for (i = 0; i < info->nr_ring_pages; i++) {
1075 if (info->ring_ref[i] != GRANT_INVALID_REF) {
1076 gnttab_end_foreign_access(info->ring_ref[i], 0, 0);
1077 info->ring_ref[i] = GRANT_INVALID_REF;
1080 free_pages((unsigned long)info->ring.sring, get_order(info->nr_ring_pages * PAGE_SIZE));
1081 info->ring.sring = NULL;
1084 unbind_from_irqhandler(info->irq, info);
1085 info->evtchn = info->irq = 0;
1089 static void blkif_completion(struct blk_shadow *s, struct blkfront_info *info,
1090 struct blkif_response *bret)
1093 struct scatterlist *sg;
1098 nseg = s->req.operation == BLKIF_OP_INDIRECT ?
1099 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1101 if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1102 for_each_sg(s->sg, sg, nseg, i) {
1103 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1104 shared_data = kmap_atomic(s->grants_used[i]->page);
1105 bvec_data = kmap_atomic(sg_page(sg));
1106 memcpy(bvec_data + sg->offset,
1107 shared_data + sg->offset,
1109 kunmap_atomic(bvec_data);
1110 kunmap_atomic(shared_data);
1113 /* Add the persistent grant into the list of free grants */
1114 for (i = 0; i < nseg; i++) {
1115 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1117 * If the grant is still mapped by the backend (the
1118 * backend has chosen to make this grant persistent)
1119 * we add it at the head of the list, so it will be
1122 if (!info->feature_persistent)
1123 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1124 s->grants_used[i]->gref);
1125 list_add(&s->grants_used[i]->node, &info->grants);
1126 info->persistent_gnts_c++;
1129 * If the grant is not mapped by the backend we end the
1130 * foreign access and add it to the tail of the list,
1131 * so it will not be picked again unless we run out of
1132 * persistent grants.
1134 gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1135 s->grants_used[i]->gref = GRANT_INVALID_REF;
1136 list_add_tail(&s->grants_used[i]->node, &info->grants);
1139 if (s->req.operation == BLKIF_OP_INDIRECT) {
1140 for (i = 0; i < INDIRECT_GREFS(nseg); i++) {
1141 if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1142 if (!info->feature_persistent)
1143 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1144 s->indirect_grants[i]->gref);
1145 list_add(&s->indirect_grants[i]->node, &info->grants);
1146 info->persistent_gnts_c++;
1148 struct page *indirect_page;
1150 gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1152 * Add the used indirect page back to the list of
1153 * available pages for indirect grefs.
1155 if (!info->feature_persistent) {
1156 indirect_page = s->indirect_grants[i]->page;
1157 list_add(&indirect_page->lru, &info->indirect_pages);
1159 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1160 list_add_tail(&s->indirect_grants[i]->node, &info->grants);
1166 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1168 struct request *req;
1169 struct blkif_response *bret;
1171 unsigned long flags;
1172 struct blkfront_info *info = (struct blkfront_info *)dev_id;
1175 spin_lock_irqsave(&info->io_lock, flags);
1177 if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
1178 spin_unlock_irqrestore(&info->io_lock, flags);
1183 rp = info->ring.sring->rsp_prod;
1184 rmb(); /* Ensure we see queued responses up to 'rp'. */
1186 for (i = info->ring.rsp_cons; i != rp; i++) {
1189 bret = RING_GET_RESPONSE(&info->ring, i);
1192 * The backend has messed up and given us an id that we would
1193 * never have given to it (we stamp it up to BLK_RING_SIZE -
1194 * look in get_id_from_freelist.
1196 if (id >= BLK_RING_SIZE(info)) {
1197 WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1198 info->gd->disk_name, op_name(bret->operation), id);
1199 /* We can't safely get the 'struct request' as
1200 * the id is busted. */
1203 req = info->shadow[id].request;
1205 if (bret->operation != BLKIF_OP_DISCARD)
1206 blkif_completion(&info->shadow[id], info, bret);
1208 if (add_id_to_freelist(info, id)) {
1209 WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1210 info->gd->disk_name, op_name(bret->operation), id);
1214 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
1215 switch (bret->operation) {
1216 case BLKIF_OP_DISCARD:
1217 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1218 struct request_queue *rq = info->rq;
1219 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1220 info->gd->disk_name, op_name(bret->operation));
1221 error = -EOPNOTSUPP;
1222 info->feature_discard = 0;
1223 info->feature_secdiscard = 0;
1224 queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1225 queue_flag_clear(QUEUE_FLAG_SECDISCARD, rq);
1227 blk_mq_complete_request(req, error);
1229 case BLKIF_OP_FLUSH_DISKCACHE:
1230 case BLKIF_OP_WRITE_BARRIER:
1231 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1232 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1233 info->gd->disk_name, op_name(bret->operation));
1234 error = -EOPNOTSUPP;
1236 if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1237 info->shadow[id].req.u.rw.nr_segments == 0)) {
1238 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1239 info->gd->disk_name, op_name(bret->operation));
1240 error = -EOPNOTSUPP;
1242 if (unlikely(error)) {
1243 if (error == -EOPNOTSUPP)
1245 info->feature_flush = 0;
1250 case BLKIF_OP_WRITE:
1251 if (unlikely(bret->status != BLKIF_RSP_OKAY))
1252 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1253 "request: %x\n", bret->status);
1255 blk_mq_complete_request(req, error);
1262 info->ring.rsp_cons = i;
1264 if (i != info->ring.req_prod_pvt) {
1266 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
1270 info->ring.sring->rsp_event = i + 1;
1272 kick_pending_request_queues(info);
1274 spin_unlock_irqrestore(&info->io_lock, flags);
1280 static int setup_blkring(struct xenbus_device *dev,
1281 struct blkfront_info *info)
1283 struct blkif_sring *sring;
1285 unsigned long ring_size = info->nr_ring_pages * PAGE_SIZE;
1286 grant_ref_t gref[XENBUS_MAX_RING_PAGES];
1288 for (i = 0; i < info->nr_ring_pages; i++)
1289 info->ring_ref[i] = GRANT_INVALID_REF;
1291 sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
1292 get_order(ring_size));
1294 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1297 SHARED_RING_INIT(sring);
1298 FRONT_RING_INIT(&info->ring, sring, ring_size);
1300 err = xenbus_grant_ring(dev, info->ring.sring, info->nr_ring_pages, gref);
1302 free_pages((unsigned long)sring, get_order(ring_size));
1303 info->ring.sring = NULL;
1306 for (i = 0; i < info->nr_ring_pages; i++)
1307 info->ring_ref[i] = gref[i];
1309 err = xenbus_alloc_evtchn(dev, &info->evtchn);
1313 err = bind_evtchn_to_irqhandler(info->evtchn, blkif_interrupt, 0,
1316 xenbus_dev_fatal(dev, err,
1317 "bind_evtchn_to_irqhandler failed");
1324 blkif_free(info, 0);
1329 /* Common code used when first setting up, and when resuming. */
1330 static int talk_to_blkback(struct xenbus_device *dev,
1331 struct blkfront_info *info)
1333 const char *message = NULL;
1334 struct xenbus_transaction xbt;
1336 unsigned int max_page_order = 0;
1337 unsigned int ring_page_order = 0;
1339 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1340 "max-ring-page-order", "%u", &max_page_order);
1342 info->nr_ring_pages = 1;
1344 ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1345 info->nr_ring_pages = 1 << ring_page_order;
1348 /* Create shared ring, alloc event channel. */
1349 err = setup_blkring(dev, info);
1354 err = xenbus_transaction_start(&xbt);
1356 xenbus_dev_fatal(dev, err, "starting transaction");
1357 goto destroy_blkring;
1360 if (info->nr_ring_pages == 1) {
1361 err = xenbus_printf(xbt, dev->nodename,
1362 "ring-ref", "%u", info->ring_ref[0]);
1364 message = "writing ring-ref";
1365 goto abort_transaction;
1368 err = xenbus_printf(xbt, dev->nodename,
1369 "ring-page-order", "%u", ring_page_order);
1371 message = "writing ring-page-order";
1372 goto abort_transaction;
1375 for (i = 0; i < info->nr_ring_pages; i++) {
1376 char ring_ref_name[RINGREF_NAME_LEN];
1378 snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1379 err = xenbus_printf(xbt, dev->nodename, ring_ref_name,
1380 "%u", info->ring_ref[i]);
1382 message = "writing ring-ref";
1383 goto abort_transaction;
1387 err = xenbus_printf(xbt, dev->nodename,
1388 "event-channel", "%u", info->evtchn);
1390 message = "writing event-channel";
1391 goto abort_transaction;
1393 err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1394 XEN_IO_PROTO_ABI_NATIVE);
1396 message = "writing protocol";
1397 goto abort_transaction;
1399 err = xenbus_printf(xbt, dev->nodename,
1400 "feature-persistent", "%u", 1);
1403 "writing persistent grants feature to xenbus");
1405 err = xenbus_transaction_end(xbt, 0);
1409 xenbus_dev_fatal(dev, err, "completing transaction");
1410 goto destroy_blkring;
1413 for (i = 0; i < BLK_RING_SIZE(info); i++)
1414 info->shadow[i].req.u.rw.id = i+1;
1415 info->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1416 xenbus_switch_state(dev, XenbusStateInitialised);
1421 xenbus_transaction_end(xbt, 1);
1423 xenbus_dev_fatal(dev, err, "%s", message);
1425 blkif_free(info, 0);
1431 * Entry point to this code when a new device is created. Allocate the basic
1432 * structures and the ring buffer for communication with the backend, and
1433 * inform the backend of the appropriate details for those. Switch to
1434 * Initialised state.
1436 static int blkfront_probe(struct xenbus_device *dev,
1437 const struct xenbus_device_id *id)
1440 struct blkfront_info *info;
1442 /* FIXME: Use dynamic device id if this is not set. */
1443 err = xenbus_scanf(XBT_NIL, dev->nodename,
1444 "virtual-device", "%i", &vdevice);
1446 /* go looking in the extended area instead */
1447 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1450 xenbus_dev_fatal(dev, err, "reading virtual-device");
1455 if (xen_hvm_domain()) {
1458 /* no unplug has been done: do not hook devices != xen vbds */
1459 if (xen_has_pv_and_legacy_disk_devices()) {
1462 if (!VDEV_IS_EXTENDED(vdevice))
1463 major = BLKIF_MAJOR(vdevice);
1465 major = XENVBD_MAJOR;
1467 if (major != XENVBD_MAJOR) {
1469 "%s: HVM does not support vbd %d as xen block device\n",
1474 /* do not create a PV cdrom device if we are an HVM guest */
1475 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1478 if (strncmp(type, "cdrom", 5) == 0) {
1484 info = kzalloc(sizeof(*info), GFP_KERNEL);
1486 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1490 mutex_init(&info->mutex);
1491 spin_lock_init(&info->io_lock);
1493 info->vdevice = vdevice;
1494 INIT_LIST_HEAD(&info->grants);
1495 INIT_LIST_HEAD(&info->indirect_pages);
1496 info->persistent_gnts_c = 0;
1497 info->connected = BLKIF_STATE_DISCONNECTED;
1498 INIT_WORK(&info->work, blkif_restart_queue);
1500 /* Front end dir is a number, which is used as the id. */
1501 info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1502 dev_set_drvdata(&dev->dev, info);
1507 static void split_bio_end(struct bio *bio)
1509 struct split_bio *split_bio = bio->bi_private;
1511 if (atomic_dec_and_test(&split_bio->pending)) {
1512 split_bio->bio->bi_phys_segments = 0;
1513 split_bio->bio->bi_error = bio->bi_error;
1514 bio_endio(split_bio->bio);
1520 static int blkif_recover(struct blkfront_info *info)
1523 struct request *req, *n;
1524 struct blk_shadow *copy;
1526 struct bio *bio, *cloned_bio;
1527 struct bio_list bio_list, merge_bio;
1528 unsigned int segs, offset;
1530 struct split_bio *split_bio;
1531 struct list_head requests;
1533 /* Stage 1: Make a safe copy of the shadow state. */
1534 copy = kmemdup(info->shadow, sizeof(info->shadow),
1535 GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
1539 /* Stage 2: Set up free list. */
1540 memset(&info->shadow, 0, sizeof(info->shadow));
1541 for (i = 0; i < BLK_RING_SIZE(info); i++)
1542 info->shadow[i].req.u.rw.id = i+1;
1543 info->shadow_free = info->ring.req_prod_pvt;
1544 info->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1546 rc = blkfront_gather_backend_features(info);
1552 segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
1553 blk_queue_max_segments(info->rq, segs);
1554 bio_list_init(&bio_list);
1555 INIT_LIST_HEAD(&requests);
1556 for (i = 0; i < BLK_RING_SIZE(info); i++) {
1558 if (!copy[i].request)
1562 * Get the bios in the request so we can re-queue them.
1564 if (copy[i].request->cmd_flags &
1565 (REQ_FLUSH | REQ_FUA | REQ_DISCARD | REQ_SECURE)) {
1567 * Flush operations don't contain bios, so
1568 * we need to requeue the whole request
1570 list_add(©[i].request->queuelist, &requests);
1573 merge_bio.head = copy[i].request->bio;
1574 merge_bio.tail = copy[i].request->biotail;
1575 bio_list_merge(&bio_list, &merge_bio);
1576 copy[i].request->bio = NULL;
1577 blk_end_request_all(copy[i].request, 0);
1582 xenbus_switch_state(info->xbdev, XenbusStateConnected);
1584 spin_lock_irq(&info->io_lock);
1586 /* Now safe for us to use the shared ring */
1587 info->connected = BLKIF_STATE_CONNECTED;
1589 /* Kick any other new requests queued since we resumed */
1590 kick_pending_request_queues(info);
1592 list_for_each_entry_safe(req, n, &requests, queuelist) {
1593 /* Requeue pending requests (flush or discard) */
1594 list_del_init(&req->queuelist);
1595 BUG_ON(req->nr_phys_segments > segs);
1596 blk_mq_requeue_request(req);
1598 spin_unlock_irq(&info->io_lock);
1599 blk_mq_kick_requeue_list(info->rq);
1601 while ((bio = bio_list_pop(&bio_list)) != NULL) {
1602 /* Traverse the list of pending bios and re-queue them */
1603 if (bio_segments(bio) > segs) {
1605 * This bio has more segments than what we can
1606 * handle, we have to split it.
1608 pending = (bio_segments(bio) + segs - 1) / segs;
1609 split_bio = kzalloc(sizeof(*split_bio), GFP_NOIO);
1610 BUG_ON(split_bio == NULL);
1611 atomic_set(&split_bio->pending, pending);
1612 split_bio->bio = bio;
1613 for (i = 0; i < pending; i++) {
1614 offset = (i * segs * PAGE_SIZE) >> 9;
1615 size = min((unsigned int)(segs * PAGE_SIZE) >> 9,
1616 (unsigned int)bio_sectors(bio) - offset);
1617 cloned_bio = bio_clone(bio, GFP_NOIO);
1618 BUG_ON(cloned_bio == NULL);
1619 bio_trim(cloned_bio, offset, size);
1620 cloned_bio->bi_private = split_bio;
1621 cloned_bio->bi_end_io = split_bio_end;
1622 submit_bio(cloned_bio->bi_rw, cloned_bio);
1625 * Now we have to wait for all those smaller bios to
1626 * end, so we can also end the "parent" bio.
1630 /* We don't need to split this bio */
1631 submit_bio(bio->bi_rw, bio);
1638 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1639 * driver restart. We tear down our blkif structure and recreate it, but
1640 * leave the device-layer structures intact so that this is transparent to the
1641 * rest of the kernel.
1643 static int blkfront_resume(struct xenbus_device *dev)
1645 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1648 dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
1650 blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
1652 err = talk_to_blkback(dev, info);
1655 * We have to wait for the backend to switch to
1656 * connected state, since we want to read which
1657 * features it supports.
1664 blkfront_closing(struct blkfront_info *info)
1666 struct xenbus_device *xbdev = info->xbdev;
1667 struct block_device *bdev = NULL;
1669 mutex_lock(&info->mutex);
1671 if (xbdev->state == XenbusStateClosing) {
1672 mutex_unlock(&info->mutex);
1677 bdev = bdget_disk(info->gd, 0);
1679 mutex_unlock(&info->mutex);
1682 xenbus_frontend_closed(xbdev);
1686 mutex_lock(&bdev->bd_mutex);
1688 if (bdev->bd_openers) {
1689 xenbus_dev_error(xbdev, -EBUSY,
1690 "Device in use; refusing to close");
1691 xenbus_switch_state(xbdev, XenbusStateClosing);
1693 xlvbd_release_gendisk(info);
1694 xenbus_frontend_closed(xbdev);
1697 mutex_unlock(&bdev->bd_mutex);
1701 static void blkfront_setup_discard(struct blkfront_info *info)
1704 unsigned int discard_granularity;
1705 unsigned int discard_alignment;
1706 unsigned int discard_secure;
1708 info->feature_discard = 1;
1709 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1710 "discard-granularity", "%u", &discard_granularity,
1711 "discard-alignment", "%u", &discard_alignment,
1714 info->discard_granularity = discard_granularity;
1715 info->discard_alignment = discard_alignment;
1717 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1718 "discard-secure", "%d", &discard_secure,
1721 info->feature_secdiscard = !!discard_secure;
1724 static int blkfront_setup_indirect(struct blkfront_info *info)
1729 if (info->max_indirect_segments == 0)
1730 segs = BLKIF_MAX_SEGMENTS_PER_REQUEST;
1732 segs = info->max_indirect_segments;
1734 err = fill_grant_buffer(info, (segs + INDIRECT_GREFS(segs)) * BLK_RING_SIZE(info));
1738 if (!info->feature_persistent && info->max_indirect_segments) {
1740 * We are using indirect descriptors but not persistent
1741 * grants, we need to allocate a set of pages that can be
1742 * used for mapping indirect grefs
1744 int num = INDIRECT_GREFS(segs) * BLK_RING_SIZE(info);
1746 BUG_ON(!list_empty(&info->indirect_pages));
1747 for (i = 0; i < num; i++) {
1748 struct page *indirect_page = alloc_page(GFP_NOIO);
1751 list_add(&indirect_page->lru, &info->indirect_pages);
1755 for (i = 0; i < BLK_RING_SIZE(info); i++) {
1756 info->shadow[i].grants_used = kzalloc(
1757 sizeof(info->shadow[i].grants_used[0]) * segs,
1759 info->shadow[i].sg = kzalloc(sizeof(info->shadow[i].sg[0]) * segs, GFP_NOIO);
1760 if (info->max_indirect_segments)
1761 info->shadow[i].indirect_grants = kzalloc(
1762 sizeof(info->shadow[i].indirect_grants[0]) *
1763 INDIRECT_GREFS(segs),
1765 if ((info->shadow[i].grants_used == NULL) ||
1766 (info->shadow[i].sg == NULL) ||
1767 (info->max_indirect_segments &&
1768 (info->shadow[i].indirect_grants == NULL)))
1770 sg_init_table(info->shadow[i].sg, segs);
1777 for (i = 0; i < BLK_RING_SIZE(info); i++) {
1778 kfree(info->shadow[i].grants_used);
1779 info->shadow[i].grants_used = NULL;
1780 kfree(info->shadow[i].sg);
1781 info->shadow[i].sg = NULL;
1782 kfree(info->shadow[i].indirect_grants);
1783 info->shadow[i].indirect_grants = NULL;
1785 if (!list_empty(&info->indirect_pages)) {
1786 struct page *indirect_page, *n;
1787 list_for_each_entry_safe(indirect_page, n, &info->indirect_pages, lru) {
1788 list_del(&indirect_page->lru);
1789 __free_page(indirect_page);
1796 * Gather all backend feature-*
1798 static int blkfront_gather_backend_features(struct blkfront_info *info)
1801 int barrier, flush, discard, persistent;
1802 unsigned int indirect_segments;
1804 info->feature_flush = 0;
1806 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1807 "feature-barrier", "%d", &barrier,
1811 * If there's no "feature-barrier" defined, then it means
1812 * we're dealing with a very old backend which writes
1813 * synchronously; nothing to do.
1815 * If there are barriers, then we use flush.
1817 if (!err && barrier)
1818 info->feature_flush = REQ_FLUSH | REQ_FUA;
1820 * And if there is "feature-flush-cache" use that above
1823 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1824 "feature-flush-cache", "%d", &flush,
1828 info->feature_flush = REQ_FLUSH;
1830 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1831 "feature-discard", "%d", &discard,
1834 if (!err && discard)
1835 blkfront_setup_discard(info);
1837 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1838 "feature-persistent", "%u", &persistent,
1841 info->feature_persistent = 0;
1843 info->feature_persistent = persistent;
1845 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1846 "feature-max-indirect-segments", "%u", &indirect_segments,
1849 info->max_indirect_segments = 0;
1851 info->max_indirect_segments = min(indirect_segments,
1852 xen_blkif_max_segments);
1854 return blkfront_setup_indirect(info);
1858 * Invoked when the backend is finally 'ready' (and has told produced
1859 * the details about the physical device - #sectors, size, etc).
1861 static void blkfront_connect(struct blkfront_info *info)
1863 unsigned long long sectors;
1864 unsigned long sector_size;
1865 unsigned int physical_sector_size;
1869 switch (info->connected) {
1870 case BLKIF_STATE_CONNECTED:
1872 * Potentially, the back-end may be signalling
1873 * a capacity change; update the capacity.
1875 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1876 "sectors", "%Lu", §ors);
1877 if (XENBUS_EXIST_ERR(err))
1879 printk(KERN_INFO "Setting capacity to %Lu\n",
1881 set_capacity(info->gd, sectors);
1882 revalidate_disk(info->gd);
1885 case BLKIF_STATE_SUSPENDED:
1887 * If we are recovering from suspension, we need to wait
1888 * for the backend to announce it's features before
1889 * reconnecting, at least we need to know if the backend
1890 * supports indirect descriptors, and how many.
1892 blkif_recover(info);
1899 dev_dbg(&info->xbdev->dev, "%s:%s.\n",
1900 __func__, info->xbdev->otherend);
1902 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1903 "sectors", "%llu", §ors,
1904 "info", "%u", &binfo,
1905 "sector-size", "%lu", §or_size,
1908 xenbus_dev_fatal(info->xbdev, err,
1909 "reading backend fields at %s",
1910 info->xbdev->otherend);
1915 * physcial-sector-size is a newer field, so old backends may not
1916 * provide this. Assume physical sector size to be the same as
1917 * sector_size in that case.
1919 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1920 "physical-sector-size", "%u", &physical_sector_size);
1922 physical_sector_size = sector_size;
1924 err = blkfront_gather_backend_features(info);
1926 xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
1927 info->xbdev->otherend);
1931 err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
1932 physical_sector_size);
1934 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
1935 info->xbdev->otherend);
1939 xenbus_switch_state(info->xbdev, XenbusStateConnected);
1941 /* Kick pending requests. */
1942 spin_lock_irq(&info->io_lock);
1943 info->connected = BLKIF_STATE_CONNECTED;
1944 kick_pending_request_queues(info);
1945 spin_unlock_irq(&info->io_lock);
1953 * Callback received when the backend's state changes.
1955 static void blkback_changed(struct xenbus_device *dev,
1956 enum xenbus_state backend_state)
1958 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1960 dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1962 switch (backend_state) {
1963 case XenbusStateInitWait:
1964 if (dev->state != XenbusStateInitialising)
1966 if (talk_to_blkback(dev, info)) {
1968 dev_set_drvdata(&dev->dev, NULL);
1971 case XenbusStateInitialising:
1972 case XenbusStateInitialised:
1973 case XenbusStateReconfiguring:
1974 case XenbusStateReconfigured:
1975 case XenbusStateUnknown:
1978 case XenbusStateConnected:
1979 blkfront_connect(info);
1982 case XenbusStateClosed:
1983 if (dev->state == XenbusStateClosed)
1985 /* Missed the backend's Closing state -- fallthrough */
1986 case XenbusStateClosing:
1987 blkfront_closing(info);
1992 static int blkfront_remove(struct xenbus_device *xbdev)
1994 struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
1995 struct block_device *bdev = NULL;
1996 struct gendisk *disk;
1998 dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2000 blkif_free(info, 0);
2002 mutex_lock(&info->mutex);
2006 bdev = bdget_disk(disk, 0);
2009 mutex_unlock(&info->mutex);
2017 * The xbdev was removed before we reached the Closed
2018 * state. See if it's safe to remove the disk. If the bdev
2019 * isn't closed yet, we let release take care of it.
2022 mutex_lock(&bdev->bd_mutex);
2023 info = disk->private_data;
2025 dev_warn(disk_to_dev(disk),
2026 "%s was hot-unplugged, %d stale handles\n",
2027 xbdev->nodename, bdev->bd_openers);
2029 if (info && !bdev->bd_openers) {
2030 xlvbd_release_gendisk(info);
2031 disk->private_data = NULL;
2035 mutex_unlock(&bdev->bd_mutex);
2041 static int blkfront_is_ready(struct xenbus_device *dev)
2043 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2045 return info->is_ready && info->xbdev;
2048 static int blkif_open(struct block_device *bdev, fmode_t mode)
2050 struct gendisk *disk = bdev->bd_disk;
2051 struct blkfront_info *info;
2054 mutex_lock(&blkfront_mutex);
2056 info = disk->private_data;
2063 mutex_lock(&info->mutex);
2066 /* xbdev is closed */
2069 mutex_unlock(&info->mutex);
2072 mutex_unlock(&blkfront_mutex);
2076 static void blkif_release(struct gendisk *disk, fmode_t mode)
2078 struct blkfront_info *info = disk->private_data;
2079 struct block_device *bdev;
2080 struct xenbus_device *xbdev;
2082 mutex_lock(&blkfront_mutex);
2084 bdev = bdget_disk(disk, 0);
2087 WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2090 if (bdev->bd_openers)
2094 * Check if we have been instructed to close. We will have
2095 * deferred this request, because the bdev was still open.
2098 mutex_lock(&info->mutex);
2099 xbdev = info->xbdev;
2101 if (xbdev && xbdev->state == XenbusStateClosing) {
2102 /* pending switch to state closed */
2103 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2104 xlvbd_release_gendisk(info);
2105 xenbus_frontend_closed(info->xbdev);
2108 mutex_unlock(&info->mutex);
2111 /* sudden device removal */
2112 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2113 xlvbd_release_gendisk(info);
2114 disk->private_data = NULL;
2121 mutex_unlock(&blkfront_mutex);
2124 static const struct block_device_operations xlvbd_block_fops =
2126 .owner = THIS_MODULE,
2128 .release = blkif_release,
2129 .getgeo = blkif_getgeo,
2130 .ioctl = blkif_ioctl,
2134 static const struct xenbus_device_id blkfront_ids[] = {
2139 static struct xenbus_driver blkfront_driver = {
2140 .ids = blkfront_ids,
2141 .probe = blkfront_probe,
2142 .remove = blkfront_remove,
2143 .resume = blkfront_resume,
2144 .otherend_changed = blkback_changed,
2145 .is_ready = blkfront_is_ready,
2148 static int __init xlblk_init(void)
2155 if (xen_blkif_max_ring_order > XENBUS_MAX_RING_PAGE_ORDER) {
2156 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2157 xen_blkif_max_ring_order, XENBUS_MAX_RING_PAGE_ORDER);
2158 xen_blkif_max_ring_order = 0;
2161 if (!xen_has_pv_disk_devices())
2164 if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2165 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
2166 XENVBD_MAJOR, DEV_NAME);
2170 ret = xenbus_register_frontend(&blkfront_driver);
2172 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2178 module_init(xlblk_init);
2181 static void __exit xlblk_exit(void)
2183 xenbus_unregister_driver(&blkfront_driver);
2184 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2187 module_exit(xlblk_exit);
2189 MODULE_DESCRIPTION("Xen virtual block device frontend");
2190 MODULE_LICENSE("GPL");
2191 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2192 MODULE_ALIAS("xen:vbd");
2193 MODULE_ALIAS("xenblk");