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/hdreg.h>
41 #include <linux/cdrom.h>
42 #include <linux/module.h>
43 #include <linux/slab.h>
44 #include <linux/mutex.h>
45 #include <linux/scatterlist.h>
46 #include <linux/bitmap.h>
47 #include <linux/list.h>
50 #include <xen/xenbus.h>
51 #include <xen/grant_table.h>
52 #include <xen/events.h>
54 #include <xen/platform_pci.h>
56 #include <xen/interface/grant_table.h>
57 #include <xen/interface/io/blkif.h>
58 #include <xen/interface/io/protocols.h>
60 #include <asm/xen/hypervisor.h>
63 BLKIF_STATE_DISCONNECTED,
64 BLKIF_STATE_CONNECTED,
65 BLKIF_STATE_SUSPENDED,
71 struct list_head node;
75 struct blkif_request req;
76 struct request *request;
77 struct grant **grants_used;
78 struct grant **indirect_grants;
79 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)");
101 #define BLK_RING_SIZE __CONST_RING_SIZE(blkif, PAGE_SIZE)
104 * We have one of these per vbd, whether ide, scsi or 'other'. They
105 * hang in private_data off the gendisk structure. We may end up
106 * putting all kinds of interesting stuff here :-)
112 struct xenbus_device *xbdev;
116 enum blkif_state connected;
118 struct blkif_front_ring ring;
119 unsigned int evtchn, irq;
120 struct request_queue *rq;
121 struct work_struct work;
122 struct gnttab_free_callback callback;
123 struct blk_shadow shadow[BLK_RING_SIZE];
124 struct list_head grants;
125 struct list_head indirect_pages;
126 unsigned int persistent_gnts_c;
127 unsigned long shadow_free;
128 unsigned int feature_flush;
129 unsigned int flush_op;
130 unsigned int feature_discard:1;
131 unsigned int feature_secdiscard:1;
132 unsigned int discard_granularity;
133 unsigned int discard_alignment;
134 unsigned int feature_persistent:1;
135 unsigned int max_indirect_segments;
139 static unsigned int nr_minors;
140 static unsigned long *minors;
141 static DEFINE_SPINLOCK(minor_lock);
143 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
144 (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
145 #define GRANT_INVALID_REF 0
147 #define PARTS_PER_DISK 16
148 #define PARTS_PER_EXT_DISK 256
150 #define BLKIF_MAJOR(dev) ((dev)>>8)
151 #define BLKIF_MINOR(dev) ((dev) & 0xff)
154 #define EXTENDED (1<<EXT_SHIFT)
155 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
156 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
157 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
158 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
159 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
160 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
162 #define DEV_NAME "xvd" /* name in /dev */
164 #define SEGS_PER_INDIRECT_FRAME \
165 (PAGE_SIZE/sizeof(struct blkif_request_segment))
166 #define INDIRECT_GREFS(_segs) \
167 ((_segs + SEGS_PER_INDIRECT_FRAME - 1)/SEGS_PER_INDIRECT_FRAME)
169 static int blkfront_setup_indirect(struct blkfront_info *info);
171 static int get_id_from_freelist(struct blkfront_info *info)
173 unsigned long free = info->shadow_free;
174 BUG_ON(free >= BLK_RING_SIZE);
175 info->shadow_free = info->shadow[free].req.u.rw.id;
176 info->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
180 static int add_id_to_freelist(struct blkfront_info *info,
183 if (info->shadow[id].req.u.rw.id != id)
185 if (info->shadow[id].request == NULL)
187 info->shadow[id].req.u.rw.id = info->shadow_free;
188 info->shadow[id].request = NULL;
189 info->shadow_free = id;
193 static int fill_grant_buffer(struct blkfront_info *info, int num)
195 struct page *granted_page;
196 struct grant *gnt_list_entry, *n;
200 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
204 if (info->feature_persistent) {
205 granted_page = alloc_page(GFP_NOIO);
207 kfree(gnt_list_entry);
210 gnt_list_entry->pfn = page_to_pfn(granted_page);
213 gnt_list_entry->gref = GRANT_INVALID_REF;
214 list_add(&gnt_list_entry->node, &info->grants);
221 list_for_each_entry_safe(gnt_list_entry, n,
222 &info->grants, node) {
223 list_del(&gnt_list_entry->node);
224 if (info->feature_persistent)
225 __free_page(pfn_to_page(gnt_list_entry->pfn));
226 kfree(gnt_list_entry);
233 static struct grant *get_grant(grant_ref_t *gref_head,
235 struct blkfront_info *info)
237 struct grant *gnt_list_entry;
238 unsigned long buffer_mfn;
240 BUG_ON(list_empty(&info->grants));
241 gnt_list_entry = list_first_entry(&info->grants, struct grant,
243 list_del(&gnt_list_entry->node);
245 if (gnt_list_entry->gref != GRANT_INVALID_REF) {
246 info->persistent_gnts_c--;
247 return gnt_list_entry;
250 /* Assign a gref to this page */
251 gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
252 BUG_ON(gnt_list_entry->gref == -ENOSPC);
253 if (!info->feature_persistent) {
255 gnt_list_entry->pfn = pfn;
257 buffer_mfn = pfn_to_mfn(gnt_list_entry->pfn);
258 gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
259 info->xbdev->otherend_id,
261 return gnt_list_entry;
264 static const char *op_name(int op)
266 static const char *const names[] = {
267 [BLKIF_OP_READ] = "read",
268 [BLKIF_OP_WRITE] = "write",
269 [BLKIF_OP_WRITE_BARRIER] = "barrier",
270 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
271 [BLKIF_OP_DISCARD] = "discard" };
273 if (op < 0 || op >= ARRAY_SIZE(names))
281 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
283 unsigned int end = minor + nr;
286 if (end > nr_minors) {
287 unsigned long *bitmap, *old;
289 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
294 spin_lock(&minor_lock);
295 if (end > nr_minors) {
297 memcpy(bitmap, minors,
298 BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
300 nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
303 spin_unlock(&minor_lock);
307 spin_lock(&minor_lock);
308 if (find_next_bit(minors, end, minor) >= end) {
309 bitmap_set(minors, minor, nr);
313 spin_unlock(&minor_lock);
318 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
320 unsigned int end = minor + nr;
322 BUG_ON(end > nr_minors);
323 spin_lock(&minor_lock);
324 bitmap_clear(minors, minor, nr);
325 spin_unlock(&minor_lock);
328 static void blkif_restart_queue_callback(void *arg)
330 struct blkfront_info *info = (struct blkfront_info *)arg;
331 schedule_work(&info->work);
334 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
336 /* We don't have real geometry info, but let's at least return
337 values consistent with the size of the device */
338 sector_t nsect = get_capacity(bd->bd_disk);
339 sector_t cylinders = nsect;
343 sector_div(cylinders, hg->heads * hg->sectors);
344 hg->cylinders = cylinders;
345 if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
346 hg->cylinders = 0xffff;
350 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
351 unsigned command, unsigned long argument)
353 struct blkfront_info *info = bdev->bd_disk->private_data;
356 dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
357 command, (long)argument);
360 case CDROMMULTISESSION:
361 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
362 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
363 if (put_user(0, (char __user *)(argument + i)))
367 case CDROM_GET_CAPABILITY: {
368 struct gendisk *gd = info->gd;
369 if (gd->flags & GENHD_FL_CD)
375 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
377 return -EINVAL; /* same return as native Linux */
384 * Generate a Xen blkfront IO request from a blk layer request. Reads
385 * and writes are handled as expected.
387 * @req: a request struct
389 static int blkif_queue_request(struct request *req)
391 struct blkfront_info *info = req->rq_disk->private_data;
392 struct blkif_request *ring_req;
394 unsigned int fsect, lsect;
396 struct blkif_request_segment *segments = NULL;
399 * Used to store if we are able to queue the request by just using
400 * existing persistent grants, or if we have to get new grants,
401 * as there are not sufficiently many free.
403 bool new_persistent_gnts;
404 grant_ref_t gref_head;
405 struct grant *gnt_list_entry = NULL;
406 struct scatterlist *sg;
409 if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
412 max_grefs = req->nr_phys_segments;
413 if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
415 * If we are using indirect segments we need to account
416 * for the indirect grefs used in the request.
418 max_grefs += INDIRECT_GREFS(req->nr_phys_segments);
420 /* Check if we have enough grants to allocate a requests */
421 if (info->persistent_gnts_c < max_grefs) {
422 new_persistent_gnts = 1;
423 if (gnttab_alloc_grant_references(
424 max_grefs - info->persistent_gnts_c,
426 gnttab_request_free_callback(
428 blkif_restart_queue_callback,
434 new_persistent_gnts = 0;
436 /* Fill out a communications ring structure. */
437 ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
438 id = get_id_from_freelist(info);
439 info->shadow[id].request = req;
441 if (unlikely(req->cmd_flags & (REQ_DISCARD | REQ_SECURE))) {
442 ring_req->operation = BLKIF_OP_DISCARD;
443 ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
444 ring_req->u.discard.id = id;
445 ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
446 if ((req->cmd_flags & REQ_SECURE) && info->feature_secdiscard)
447 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
449 ring_req->u.discard.flag = 0;
451 BUG_ON(info->max_indirect_segments == 0 &&
452 req->nr_phys_segments > BLKIF_MAX_SEGMENTS_PER_REQUEST);
453 BUG_ON(info->max_indirect_segments &&
454 req->nr_phys_segments > info->max_indirect_segments);
455 nseg = blk_rq_map_sg(req->q, req, info->shadow[id].sg);
456 ring_req->u.rw.id = id;
457 if (nseg > BLKIF_MAX_SEGMENTS_PER_REQUEST) {
459 * The indirect operation can only be a BLKIF_OP_READ or
462 BUG_ON(req->cmd_flags & (REQ_FLUSH | REQ_FUA));
463 ring_req->operation = BLKIF_OP_INDIRECT;
464 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
465 BLKIF_OP_WRITE : BLKIF_OP_READ;
466 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
467 ring_req->u.indirect.handle = info->handle;
468 ring_req->u.indirect.nr_segments = nseg;
470 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
471 ring_req->u.rw.handle = info->handle;
472 ring_req->operation = rq_data_dir(req) ?
473 BLKIF_OP_WRITE : BLKIF_OP_READ;
474 if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) {
476 * Ideally we can do an unordered flush-to-disk. In case the
477 * backend onlysupports barriers, use that. A barrier request
478 * a superset of FUA, so we can implement it the same
479 * way. (It's also a FLUSH+FUA, since it is
480 * guaranteed ordered WRT previous writes.)
482 ring_req->operation = info->flush_op;
484 ring_req->u.rw.nr_segments = nseg;
486 for_each_sg(info->shadow[id].sg, sg, nseg, i) {
487 fsect = sg->offset >> 9;
488 lsect = fsect + (sg->length >> 9) - 1;
490 if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
491 (i % SEGS_PER_INDIRECT_FRAME == 0)) {
492 unsigned long uninitialized_var(pfn);
495 kunmap_atomic(segments);
497 n = i / SEGS_PER_INDIRECT_FRAME;
498 if (!info->feature_persistent) {
499 struct page *indirect_page;
501 /* Fetch a pre-allocated page to use for indirect grefs */
502 BUG_ON(list_empty(&info->indirect_pages));
503 indirect_page = list_first_entry(&info->indirect_pages,
505 list_del(&indirect_page->lru);
506 pfn = page_to_pfn(indirect_page);
508 gnt_list_entry = get_grant(&gref_head, pfn, info);
509 info->shadow[id].indirect_grants[n] = gnt_list_entry;
510 segments = kmap_atomic(pfn_to_page(gnt_list_entry->pfn));
511 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
514 gnt_list_entry = get_grant(&gref_head, page_to_pfn(sg_page(sg)), info);
515 ref = gnt_list_entry->gref;
517 info->shadow[id].grants_used[i] = gnt_list_entry;
519 if (rq_data_dir(req) && info->feature_persistent) {
523 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
525 shared_data = kmap_atomic(pfn_to_page(gnt_list_entry->pfn));
526 bvec_data = kmap_atomic(sg_page(sg));
529 * this does not wipe data stored outside the
530 * range sg->offset..sg->offset+sg->length.
531 * Therefore, blkback *could* see data from
532 * previous requests. This is OK as long as
533 * persistent grants are shared with just one
534 * domain. It may need refactoring if this
537 memcpy(shared_data + sg->offset,
538 bvec_data + sg->offset,
541 kunmap_atomic(bvec_data);
542 kunmap_atomic(shared_data);
544 if (ring_req->operation != BLKIF_OP_INDIRECT) {
545 ring_req->u.rw.seg[i] =
546 (struct blkif_request_segment) {
549 .last_sect = lsect };
551 n = i % SEGS_PER_INDIRECT_FRAME;
553 (struct blkif_request_segment) {
556 .last_sect = lsect };
560 kunmap_atomic(segments);
563 info->ring.req_prod_pvt++;
565 /* Keep a private copy so we can reissue requests when recovering. */
566 info->shadow[id].req = *ring_req;
568 if (new_persistent_gnts)
569 gnttab_free_grant_references(gref_head);
575 static inline void flush_requests(struct blkfront_info *info)
579 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
582 notify_remote_via_irq(info->irq);
585 static inline bool blkif_request_flush_valid(struct request *req,
586 struct blkfront_info *info)
588 return ((req->cmd_type != REQ_TYPE_FS) ||
589 ((req->cmd_flags & (REQ_FLUSH | REQ_FUA)) &&
595 * read a block; request is in a request queue
597 static void do_blkif_request(struct request_queue *rq)
599 struct blkfront_info *info = NULL;
603 pr_debug("Entered do_blkif_request\n");
607 while ((req = blk_peek_request(rq)) != NULL) {
608 info = req->rq_disk->private_data;
610 if (RING_FULL(&info->ring))
613 blk_start_request(req);
615 if (blkif_request_flush_valid(req, info)) {
616 __blk_end_request_all(req, -EIO);
620 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
622 req, req->cmd, (unsigned long)blk_rq_pos(req),
623 blk_rq_cur_sectors(req), blk_rq_sectors(req),
624 rq_data_dir(req) ? "write" : "read");
626 if (blkif_queue_request(req)) {
627 blk_requeue_request(rq, req);
629 /* Avoid pointless unplugs. */
638 flush_requests(info);
641 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
642 unsigned int physical_sector_size,
643 unsigned int segments)
645 struct request_queue *rq;
646 struct blkfront_info *info = gd->private_data;
648 rq = blk_init_queue(do_blkif_request, &info->io_lock);
652 queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
654 if (info->feature_discard) {
655 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
656 blk_queue_max_discard_sectors(rq, get_capacity(gd));
657 rq->limits.discard_granularity = info->discard_granularity;
658 rq->limits.discard_alignment = info->discard_alignment;
659 if (info->feature_secdiscard)
660 queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD, rq);
663 /* Hard sector size and max sectors impersonate the equiv. hardware. */
664 blk_queue_logical_block_size(rq, sector_size);
665 blk_queue_physical_block_size(rq, physical_sector_size);
666 blk_queue_max_hw_sectors(rq, (segments * PAGE_SIZE) / 512);
668 /* Each segment in a request is up to an aligned page in size. */
669 blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
670 blk_queue_max_segment_size(rq, PAGE_SIZE);
672 /* Ensure a merged request will fit in a single I/O ring slot. */
673 blk_queue_max_segments(rq, segments);
675 /* Make sure buffer addresses are sector-aligned. */
676 blk_queue_dma_alignment(rq, 511);
678 /* Make sure we don't use bounce buffers. */
679 blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
687 static void xlvbd_flush(struct blkfront_info *info)
689 blk_queue_flush(info->rq, info->feature_flush);
690 printk(KERN_INFO "blkfront: %s: %s: %s %s %s %s %s\n",
692 info->flush_op == BLKIF_OP_WRITE_BARRIER ?
693 "barrier" : (info->flush_op == BLKIF_OP_FLUSH_DISKCACHE ?
694 "flush diskcache" : "barrier or flush"),
695 info->feature_flush ? "enabled;" : "disabled;",
696 "persistent grants:",
697 info->feature_persistent ? "enabled;" : "disabled;",
698 "indirect descriptors:",
699 info->max_indirect_segments ? "enabled;" : "disabled;");
702 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
705 major = BLKIF_MAJOR(vdevice);
706 *minor = BLKIF_MINOR(vdevice);
709 *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
710 *minor = ((*minor / 64) * PARTS_PER_DISK) +
711 EMULATED_HD_DISK_MINOR_OFFSET;
714 *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
715 *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
716 EMULATED_HD_DISK_MINOR_OFFSET;
718 case XEN_SCSI_DISK0_MAJOR:
719 *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
720 *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
722 case XEN_SCSI_DISK1_MAJOR:
723 case XEN_SCSI_DISK2_MAJOR:
724 case XEN_SCSI_DISK3_MAJOR:
725 case XEN_SCSI_DISK4_MAJOR:
726 case XEN_SCSI_DISK5_MAJOR:
727 case XEN_SCSI_DISK6_MAJOR:
728 case XEN_SCSI_DISK7_MAJOR:
729 *offset = (*minor / PARTS_PER_DISK) +
730 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
731 EMULATED_SD_DISK_NAME_OFFSET;
733 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
734 EMULATED_SD_DISK_MINOR_OFFSET;
736 case XEN_SCSI_DISK8_MAJOR:
737 case XEN_SCSI_DISK9_MAJOR:
738 case XEN_SCSI_DISK10_MAJOR:
739 case XEN_SCSI_DISK11_MAJOR:
740 case XEN_SCSI_DISK12_MAJOR:
741 case XEN_SCSI_DISK13_MAJOR:
742 case XEN_SCSI_DISK14_MAJOR:
743 case XEN_SCSI_DISK15_MAJOR:
744 *offset = (*minor / PARTS_PER_DISK) +
745 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
746 EMULATED_SD_DISK_NAME_OFFSET;
748 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
749 EMULATED_SD_DISK_MINOR_OFFSET;
752 *offset = *minor / PARTS_PER_DISK;
755 printk(KERN_WARNING "blkfront: your disk configuration is "
756 "incorrect, please use an xvd device instead\n");
762 static char *encode_disk_name(char *ptr, unsigned int n)
765 ptr = encode_disk_name(ptr, n / 26 - 1);
770 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
771 struct blkfront_info *info,
772 u16 vdisk_info, u16 sector_size,
773 unsigned int physical_sector_size)
783 BUG_ON(info->gd != NULL);
784 BUG_ON(info->rq != NULL);
786 if ((info->vdevice>>EXT_SHIFT) > 1) {
787 /* this is above the extended range; something is wrong */
788 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
792 if (!VDEV_IS_EXTENDED(info->vdevice)) {
793 err = xen_translate_vdev(info->vdevice, &minor, &offset);
796 nr_parts = PARTS_PER_DISK;
798 minor = BLKIF_MINOR_EXT(info->vdevice);
799 nr_parts = PARTS_PER_EXT_DISK;
800 offset = minor / nr_parts;
801 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
802 printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
803 "emulated IDE disks,\n\t choose an xvd device name"
804 "from xvde on\n", info->vdevice);
806 if (minor >> MINORBITS) {
807 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
808 info->vdevice, minor);
812 if ((minor % nr_parts) == 0)
813 nr_minors = nr_parts;
815 err = xlbd_reserve_minors(minor, nr_minors);
820 gd = alloc_disk(nr_minors);
824 strcpy(gd->disk_name, DEV_NAME);
825 ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
826 BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
830 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
831 "%d", minor & (nr_parts - 1));
833 gd->major = XENVBD_MAJOR;
834 gd->first_minor = minor;
835 gd->fops = &xlvbd_block_fops;
836 gd->private_data = info;
837 gd->driverfs_dev = &(info->xbdev->dev);
838 set_capacity(gd, capacity);
840 if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size,
841 info->max_indirect_segments ? :
842 BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
847 info->rq = gd->queue;
852 if (vdisk_info & VDISK_READONLY)
855 if (vdisk_info & VDISK_REMOVABLE)
856 gd->flags |= GENHD_FL_REMOVABLE;
858 if (vdisk_info & VDISK_CDROM)
859 gd->flags |= GENHD_FL_CD;
864 xlbd_release_minors(minor, nr_minors);
869 static void xlvbd_release_gendisk(struct blkfront_info *info)
871 unsigned int minor, nr_minors;
874 if (info->rq == NULL)
877 spin_lock_irqsave(&info->io_lock, flags);
879 /* No more blkif_request(). */
880 blk_stop_queue(info->rq);
882 /* No more gnttab callback work. */
883 gnttab_cancel_free_callback(&info->callback);
884 spin_unlock_irqrestore(&info->io_lock, flags);
886 /* Flush gnttab callback work. Must be done with no locks held. */
887 flush_work(&info->work);
889 del_gendisk(info->gd);
891 minor = info->gd->first_minor;
892 nr_minors = info->gd->minors;
893 xlbd_release_minors(minor, nr_minors);
895 blk_cleanup_queue(info->rq);
902 static void kick_pending_request_queues(struct blkfront_info *info)
904 if (!RING_FULL(&info->ring)) {
905 /* Re-enable calldowns. */
906 blk_start_queue(info->rq);
907 /* Kick things off immediately. */
908 do_blkif_request(info->rq);
912 static void blkif_restart_queue(struct work_struct *work)
914 struct blkfront_info *info = container_of(work, struct blkfront_info, work);
916 spin_lock_irq(&info->io_lock);
917 if (info->connected == BLKIF_STATE_CONNECTED)
918 kick_pending_request_queues(info);
919 spin_unlock_irq(&info->io_lock);
922 static void blkif_free(struct blkfront_info *info, int suspend)
924 struct grant *persistent_gnt;
928 /* Prevent new requests being issued until we fix things up. */
929 spin_lock_irq(&info->io_lock);
930 info->connected = suspend ?
931 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
932 /* No more blkif_request(). */
934 blk_stop_queue(info->rq);
936 /* Remove all persistent grants */
937 if (!list_empty(&info->grants)) {
938 list_for_each_entry_safe(persistent_gnt, n,
939 &info->grants, node) {
940 list_del(&persistent_gnt->node);
941 if (persistent_gnt->gref != GRANT_INVALID_REF) {
942 gnttab_end_foreign_access(persistent_gnt->gref,
944 info->persistent_gnts_c--;
946 if (info->feature_persistent)
947 __free_page(pfn_to_page(persistent_gnt->pfn));
948 kfree(persistent_gnt);
951 BUG_ON(info->persistent_gnts_c != 0);
954 * Remove indirect pages, this only happens when using indirect
955 * descriptors but not persistent grants
957 if (!list_empty(&info->indirect_pages)) {
958 struct page *indirect_page, *n;
960 BUG_ON(info->feature_persistent);
961 list_for_each_entry_safe(indirect_page, n, &info->indirect_pages, lru) {
962 list_del(&indirect_page->lru);
963 __free_page(indirect_page);
967 for (i = 0; i < BLK_RING_SIZE; i++) {
969 * Clear persistent grants present in requests already
972 if (!info->shadow[i].request)
975 segs = info->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
976 info->shadow[i].req.u.indirect.nr_segments :
977 info->shadow[i].req.u.rw.nr_segments;
978 for (j = 0; j < segs; j++) {
979 persistent_gnt = info->shadow[i].grants_used[j];
980 gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
981 if (info->feature_persistent)
982 __free_page(pfn_to_page(persistent_gnt->pfn));
983 kfree(persistent_gnt);
986 if (info->shadow[i].req.operation != BLKIF_OP_INDIRECT)
988 * If this is not an indirect operation don't try to
989 * free indirect segments
993 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
994 persistent_gnt = info->shadow[i].indirect_grants[j];
995 gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
996 __free_page(pfn_to_page(persistent_gnt->pfn));
997 kfree(persistent_gnt);
1001 kfree(info->shadow[i].grants_used);
1002 info->shadow[i].grants_used = NULL;
1003 kfree(info->shadow[i].indirect_grants);
1004 info->shadow[i].indirect_grants = NULL;
1005 kfree(info->shadow[i].sg);
1006 info->shadow[i].sg = NULL;
1009 /* No more gnttab callback work. */
1010 gnttab_cancel_free_callback(&info->callback);
1011 spin_unlock_irq(&info->io_lock);
1013 /* Flush gnttab callback work. Must be done with no locks held. */
1014 flush_work(&info->work);
1016 /* Free resources associated with old device channel. */
1017 if (info->ring_ref != GRANT_INVALID_REF) {
1018 gnttab_end_foreign_access(info->ring_ref, 0,
1019 (unsigned long)info->ring.sring);
1020 info->ring_ref = GRANT_INVALID_REF;
1021 info->ring.sring = NULL;
1024 unbind_from_irqhandler(info->irq, info);
1025 info->evtchn = info->irq = 0;
1029 static void blkif_completion(struct blk_shadow *s, struct blkfront_info *info,
1030 struct blkif_response *bret)
1033 struct scatterlist *sg;
1038 nseg = s->req.operation == BLKIF_OP_INDIRECT ?
1039 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1041 if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1043 * Copy the data received from the backend into the bvec.
1044 * Since bv_offset can be different than 0, and bv_len different
1045 * than PAGE_SIZE, we have to keep track of the current offset,
1046 * to be sure we are copying the data from the right shared page.
1048 for_each_sg(s->sg, sg, nseg, i) {
1049 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1050 shared_data = kmap_atomic(
1051 pfn_to_page(s->grants_used[i]->pfn));
1052 bvec_data = kmap_atomic(sg_page(sg));
1053 memcpy(bvec_data + sg->offset,
1054 shared_data + sg->offset,
1056 kunmap_atomic(bvec_data);
1057 kunmap_atomic(shared_data);
1060 /* Add the persistent grant into the list of free grants */
1061 for (i = 0; i < nseg; i++) {
1062 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1064 * If the grant is still mapped by the backend (the
1065 * backend has chosen to make this grant persistent)
1066 * we add it at the head of the list, so it will be
1069 if (!info->feature_persistent)
1070 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1071 s->grants_used[i]->gref);
1072 list_add(&s->grants_used[i]->node, &info->grants);
1073 info->persistent_gnts_c++;
1076 * If the grant is not mapped by the backend we end the
1077 * foreign access and add it to the tail of the list,
1078 * so it will not be picked again unless we run out of
1079 * persistent grants.
1081 gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1082 s->grants_used[i]->gref = GRANT_INVALID_REF;
1083 list_add_tail(&s->grants_used[i]->node, &info->grants);
1086 if (s->req.operation == BLKIF_OP_INDIRECT) {
1087 for (i = 0; i < INDIRECT_GREFS(nseg); i++) {
1088 if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1089 if (!info->feature_persistent)
1090 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1091 s->indirect_grants[i]->gref);
1092 list_add(&s->indirect_grants[i]->node, &info->grants);
1093 info->persistent_gnts_c++;
1095 struct page *indirect_page;
1097 gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1099 * Add the used indirect page back to the list of
1100 * available pages for indirect grefs.
1102 indirect_page = pfn_to_page(s->indirect_grants[i]->pfn);
1103 list_add(&indirect_page->lru, &info->indirect_pages);
1104 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1105 list_add_tail(&s->indirect_grants[i]->node, &info->grants);
1111 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1113 struct request *req;
1114 struct blkif_response *bret;
1116 unsigned long flags;
1117 struct blkfront_info *info = (struct blkfront_info *)dev_id;
1120 spin_lock_irqsave(&info->io_lock, flags);
1122 if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
1123 spin_unlock_irqrestore(&info->io_lock, flags);
1128 rp = info->ring.sring->rsp_prod;
1129 rmb(); /* Ensure we see queued responses up to 'rp'. */
1131 for (i = info->ring.rsp_cons; i != rp; i++) {
1134 bret = RING_GET_RESPONSE(&info->ring, i);
1137 * The backend has messed up and given us an id that we would
1138 * never have given to it (we stamp it up to BLK_RING_SIZE -
1139 * look in get_id_from_freelist.
1141 if (id >= BLK_RING_SIZE) {
1142 WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1143 info->gd->disk_name, op_name(bret->operation), id);
1144 /* We can't safely get the 'struct request' as
1145 * the id is busted. */
1148 req = info->shadow[id].request;
1150 if (bret->operation != BLKIF_OP_DISCARD)
1151 blkif_completion(&info->shadow[id], info, bret);
1153 if (add_id_to_freelist(info, id)) {
1154 WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1155 info->gd->disk_name, op_name(bret->operation), id);
1159 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
1160 switch (bret->operation) {
1161 case BLKIF_OP_DISCARD:
1162 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1163 struct request_queue *rq = info->rq;
1164 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1165 info->gd->disk_name, op_name(bret->operation));
1166 error = -EOPNOTSUPP;
1167 info->feature_discard = 0;
1168 info->feature_secdiscard = 0;
1169 queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1170 queue_flag_clear(QUEUE_FLAG_SECDISCARD, rq);
1172 __blk_end_request_all(req, error);
1174 case BLKIF_OP_FLUSH_DISKCACHE:
1175 case BLKIF_OP_WRITE_BARRIER:
1176 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1177 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1178 info->gd->disk_name, op_name(bret->operation));
1179 error = -EOPNOTSUPP;
1181 if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1182 info->shadow[id].req.u.rw.nr_segments == 0)) {
1183 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1184 info->gd->disk_name, op_name(bret->operation));
1185 error = -EOPNOTSUPP;
1187 if (unlikely(error)) {
1188 if (error == -EOPNOTSUPP)
1190 info->feature_flush = 0;
1196 case BLKIF_OP_WRITE:
1197 if (unlikely(bret->status != BLKIF_RSP_OKAY))
1198 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1199 "request: %x\n", bret->status);
1201 __blk_end_request_all(req, error);
1208 info->ring.rsp_cons = i;
1210 if (i != info->ring.req_prod_pvt) {
1212 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
1216 info->ring.sring->rsp_event = i + 1;
1218 kick_pending_request_queues(info);
1220 spin_unlock_irqrestore(&info->io_lock, flags);
1226 static int setup_blkring(struct xenbus_device *dev,
1227 struct blkfront_info *info)
1229 struct blkif_sring *sring;
1232 info->ring_ref = GRANT_INVALID_REF;
1234 sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
1236 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1239 SHARED_RING_INIT(sring);
1240 FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
1242 err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
1244 free_page((unsigned long)sring);
1245 info->ring.sring = NULL;
1248 info->ring_ref = err;
1250 err = xenbus_alloc_evtchn(dev, &info->evtchn);
1254 err = bind_evtchn_to_irqhandler(info->evtchn, blkif_interrupt, 0,
1257 xenbus_dev_fatal(dev, err,
1258 "bind_evtchn_to_irqhandler failed");
1265 blkif_free(info, 0);
1270 /* Common code used when first setting up, and when resuming. */
1271 static int talk_to_blkback(struct xenbus_device *dev,
1272 struct blkfront_info *info)
1274 const char *message = NULL;
1275 struct xenbus_transaction xbt;
1278 /* Create shared ring, alloc event channel. */
1279 err = setup_blkring(dev, info);
1284 err = xenbus_transaction_start(&xbt);
1286 xenbus_dev_fatal(dev, err, "starting transaction");
1287 goto destroy_blkring;
1290 err = xenbus_printf(xbt, dev->nodename,
1291 "ring-ref", "%u", info->ring_ref);
1293 message = "writing ring-ref";
1294 goto abort_transaction;
1296 err = xenbus_printf(xbt, dev->nodename,
1297 "event-channel", "%u", info->evtchn);
1299 message = "writing event-channel";
1300 goto abort_transaction;
1302 err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1303 XEN_IO_PROTO_ABI_NATIVE);
1305 message = "writing protocol";
1306 goto abort_transaction;
1308 err = xenbus_printf(xbt, dev->nodename,
1309 "feature-persistent", "%u", 1);
1312 "writing persistent grants feature to xenbus");
1314 err = xenbus_transaction_end(xbt, 0);
1318 xenbus_dev_fatal(dev, err, "completing transaction");
1319 goto destroy_blkring;
1322 xenbus_switch_state(dev, XenbusStateInitialised);
1327 xenbus_transaction_end(xbt, 1);
1329 xenbus_dev_fatal(dev, err, "%s", message);
1331 blkif_free(info, 0);
1337 * Entry point to this code when a new device is created. Allocate the basic
1338 * structures and the ring buffer for communication with the backend, and
1339 * inform the backend of the appropriate details for those. Switch to
1340 * Initialised state.
1342 static int blkfront_probe(struct xenbus_device *dev,
1343 const struct xenbus_device_id *id)
1345 int err, vdevice, i;
1346 struct blkfront_info *info;
1348 /* FIXME: Use dynamic device id if this is not set. */
1349 err = xenbus_scanf(XBT_NIL, dev->nodename,
1350 "virtual-device", "%i", &vdevice);
1352 /* go looking in the extended area instead */
1353 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1356 xenbus_dev_fatal(dev, err, "reading virtual-device");
1361 if (xen_hvm_domain()) {
1364 /* no unplug has been done: do not hook devices != xen vbds */
1365 if (xen_has_pv_and_legacy_disk_devices()) {
1368 if (!VDEV_IS_EXTENDED(vdevice))
1369 major = BLKIF_MAJOR(vdevice);
1371 major = XENVBD_MAJOR;
1373 if (major != XENVBD_MAJOR) {
1375 "%s: HVM does not support vbd %d as xen block device\n",
1376 __FUNCTION__, vdevice);
1380 /* do not create a PV cdrom device if we are an HVM guest */
1381 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1384 if (strncmp(type, "cdrom", 5) == 0) {
1390 info = kzalloc(sizeof(*info), GFP_KERNEL);
1392 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1396 mutex_init(&info->mutex);
1397 spin_lock_init(&info->io_lock);
1399 info->vdevice = vdevice;
1400 INIT_LIST_HEAD(&info->grants);
1401 INIT_LIST_HEAD(&info->indirect_pages);
1402 info->persistent_gnts_c = 0;
1403 info->connected = BLKIF_STATE_DISCONNECTED;
1404 INIT_WORK(&info->work, blkif_restart_queue);
1406 for (i = 0; i < BLK_RING_SIZE; i++)
1407 info->shadow[i].req.u.rw.id = i+1;
1408 info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1410 /* Front end dir is a number, which is used as the id. */
1411 info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1412 dev_set_drvdata(&dev->dev, info);
1414 err = talk_to_blkback(dev, info);
1417 dev_set_drvdata(&dev->dev, NULL);
1424 static void split_bio_end(struct bio *bio, int error)
1426 struct split_bio *split_bio = bio->bi_private;
1429 split_bio->err = error;
1431 if (atomic_dec_and_test(&split_bio->pending)) {
1432 split_bio->bio->bi_phys_segments = 0;
1433 bio_endio(split_bio->bio, split_bio->err);
1439 static int blkif_recover(struct blkfront_info *info)
1442 struct request *req, *n;
1443 struct blk_shadow *copy;
1445 struct bio *bio, *cloned_bio;
1446 struct bio_list bio_list, merge_bio;
1447 unsigned int segs, offset;
1449 struct split_bio *split_bio;
1450 struct list_head requests;
1452 /* Stage 1: Make a safe copy of the shadow state. */
1453 copy = kmemdup(info->shadow, sizeof(info->shadow),
1454 GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
1458 /* Stage 2: Set up free list. */
1459 memset(&info->shadow, 0, sizeof(info->shadow));
1460 for (i = 0; i < BLK_RING_SIZE; i++)
1461 info->shadow[i].req.u.rw.id = i+1;
1462 info->shadow_free = info->ring.req_prod_pvt;
1463 info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1465 rc = blkfront_setup_indirect(info);
1471 segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
1472 blk_queue_max_segments(info->rq, segs);
1473 bio_list_init(&bio_list);
1474 INIT_LIST_HEAD(&requests);
1475 for (i = 0; i < BLK_RING_SIZE; i++) {
1477 if (!copy[i].request)
1481 * Get the bios in the request so we can re-queue them.
1483 if (copy[i].request->cmd_flags &
1484 (REQ_FLUSH | REQ_FUA | REQ_DISCARD | REQ_SECURE)) {
1486 * Flush operations don't contain bios, so
1487 * we need to requeue the whole request
1489 list_add(©[i].request->queuelist, &requests);
1492 merge_bio.head = copy[i].request->bio;
1493 merge_bio.tail = copy[i].request->biotail;
1494 bio_list_merge(&bio_list, &merge_bio);
1495 copy[i].request->bio = NULL;
1496 blk_put_request(copy[i].request);
1502 * Empty the queue, this is important because we might have
1503 * requests in the queue with more segments than what we
1506 spin_lock_irq(&info->io_lock);
1507 while ((req = blk_fetch_request(info->rq)) != NULL) {
1508 if (req->cmd_flags &
1509 (REQ_FLUSH | REQ_FUA | REQ_DISCARD | REQ_SECURE)) {
1510 list_add(&req->queuelist, &requests);
1513 merge_bio.head = req->bio;
1514 merge_bio.tail = req->biotail;
1515 bio_list_merge(&bio_list, &merge_bio);
1517 if (req->cmd_flags & (REQ_FLUSH | REQ_FUA))
1518 pr_alert("diskcache flush request found!\n");
1519 __blk_put_request(info->rq, req);
1521 spin_unlock_irq(&info->io_lock);
1523 xenbus_switch_state(info->xbdev, XenbusStateConnected);
1525 spin_lock_irq(&info->io_lock);
1527 /* Now safe for us to use the shared ring */
1528 info->connected = BLKIF_STATE_CONNECTED;
1530 /* Kick any other new requests queued since we resumed */
1531 kick_pending_request_queues(info);
1533 list_for_each_entry_safe(req, n, &requests, queuelist) {
1534 /* Requeue pending requests (flush or discard) */
1535 list_del_init(&req->queuelist);
1536 BUG_ON(req->nr_phys_segments > segs);
1537 blk_requeue_request(info->rq, req);
1539 spin_unlock_irq(&info->io_lock);
1541 while ((bio = bio_list_pop(&bio_list)) != NULL) {
1542 /* Traverse the list of pending bios and re-queue them */
1543 if (bio_segments(bio) > segs) {
1545 * This bio has more segments than what we can
1546 * handle, we have to split it.
1548 pending = (bio_segments(bio) + segs - 1) / segs;
1549 split_bio = kzalloc(sizeof(*split_bio), GFP_NOIO);
1550 BUG_ON(split_bio == NULL);
1551 atomic_set(&split_bio->pending, pending);
1552 split_bio->bio = bio;
1553 for (i = 0; i < pending; i++) {
1554 offset = (i * segs * PAGE_SIZE) >> 9;
1555 size = min((unsigned int)(segs * PAGE_SIZE) >> 9,
1556 (unsigned int)bio_sectors(bio) - offset);
1557 cloned_bio = bio_clone(bio, GFP_NOIO);
1558 BUG_ON(cloned_bio == NULL);
1559 bio_trim(cloned_bio, offset, size);
1560 cloned_bio->bi_private = split_bio;
1561 cloned_bio->bi_end_io = split_bio_end;
1562 submit_bio(cloned_bio->bi_rw, cloned_bio);
1565 * Now we have to wait for all those smaller bios to
1566 * end, so we can also end the "parent" bio.
1570 /* We don't need to split this bio */
1571 submit_bio(bio->bi_rw, bio);
1578 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1579 * driver restart. We tear down our blkif structure and recreate it, but
1580 * leave the device-layer structures intact so that this is transparent to the
1581 * rest of the kernel.
1583 static int blkfront_resume(struct xenbus_device *dev)
1585 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1588 dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
1590 blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
1592 err = talk_to_blkback(dev, info);
1595 * We have to wait for the backend to switch to
1596 * connected state, since we want to read which
1597 * features it supports.
1604 blkfront_closing(struct blkfront_info *info)
1606 struct xenbus_device *xbdev = info->xbdev;
1607 struct block_device *bdev = NULL;
1609 mutex_lock(&info->mutex);
1611 if (xbdev->state == XenbusStateClosing) {
1612 mutex_unlock(&info->mutex);
1617 bdev = bdget_disk(info->gd, 0);
1619 mutex_unlock(&info->mutex);
1622 xenbus_frontend_closed(xbdev);
1626 mutex_lock(&bdev->bd_mutex);
1628 if (bdev->bd_openers) {
1629 xenbus_dev_error(xbdev, -EBUSY,
1630 "Device in use; refusing to close");
1631 xenbus_switch_state(xbdev, XenbusStateClosing);
1633 xlvbd_release_gendisk(info);
1634 xenbus_frontend_closed(xbdev);
1637 mutex_unlock(&bdev->bd_mutex);
1641 static void blkfront_setup_discard(struct blkfront_info *info)
1644 unsigned int discard_granularity;
1645 unsigned int discard_alignment;
1646 unsigned int discard_secure;
1648 info->feature_discard = 1;
1649 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1650 "discard-granularity", "%u", &discard_granularity,
1651 "discard-alignment", "%u", &discard_alignment,
1654 info->discard_granularity = discard_granularity;
1655 info->discard_alignment = discard_alignment;
1657 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1658 "discard-secure", "%d", &discard_secure,
1661 info->feature_secdiscard = !!discard_secure;
1664 static int blkfront_setup_indirect(struct blkfront_info *info)
1666 unsigned int indirect_segments, segs;
1669 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1670 "feature-max-indirect-segments", "%u", &indirect_segments,
1673 info->max_indirect_segments = 0;
1674 segs = BLKIF_MAX_SEGMENTS_PER_REQUEST;
1676 info->max_indirect_segments = min(indirect_segments,
1677 xen_blkif_max_segments);
1678 segs = info->max_indirect_segments;
1681 err = fill_grant_buffer(info, (segs + INDIRECT_GREFS(segs)) * BLK_RING_SIZE);
1685 if (!info->feature_persistent && info->max_indirect_segments) {
1687 * We are using indirect descriptors but not persistent
1688 * grants, we need to allocate a set of pages that can be
1689 * used for mapping indirect grefs
1691 int num = INDIRECT_GREFS(segs) * BLK_RING_SIZE;
1693 BUG_ON(!list_empty(&info->indirect_pages));
1694 for (i = 0; i < num; i++) {
1695 struct page *indirect_page = alloc_page(GFP_NOIO);
1698 list_add(&indirect_page->lru, &info->indirect_pages);
1702 for (i = 0; i < BLK_RING_SIZE; i++) {
1703 info->shadow[i].grants_used = kzalloc(
1704 sizeof(info->shadow[i].grants_used[0]) * segs,
1706 info->shadow[i].sg = kzalloc(sizeof(info->shadow[i].sg[0]) * segs, GFP_NOIO);
1707 if (info->max_indirect_segments)
1708 info->shadow[i].indirect_grants = kzalloc(
1709 sizeof(info->shadow[i].indirect_grants[0]) *
1710 INDIRECT_GREFS(segs),
1712 if ((info->shadow[i].grants_used == NULL) ||
1713 (info->shadow[i].sg == NULL) ||
1714 (info->max_indirect_segments &&
1715 (info->shadow[i].indirect_grants == NULL)))
1717 sg_init_table(info->shadow[i].sg, segs);
1724 for (i = 0; i < BLK_RING_SIZE; i++) {
1725 kfree(info->shadow[i].grants_used);
1726 info->shadow[i].grants_used = NULL;
1727 kfree(info->shadow[i].sg);
1728 info->shadow[i].sg = NULL;
1729 kfree(info->shadow[i].indirect_grants);
1730 info->shadow[i].indirect_grants = NULL;
1732 if (!list_empty(&info->indirect_pages)) {
1733 struct page *indirect_page, *n;
1734 list_for_each_entry_safe(indirect_page, n, &info->indirect_pages, lru) {
1735 list_del(&indirect_page->lru);
1736 __free_page(indirect_page);
1743 * Invoked when the backend is finally 'ready' (and has told produced
1744 * the details about the physical device - #sectors, size, etc).
1746 static void blkfront_connect(struct blkfront_info *info)
1748 unsigned long long sectors;
1749 unsigned long sector_size;
1750 unsigned int physical_sector_size;
1753 int barrier, flush, discard, persistent;
1755 switch (info->connected) {
1756 case BLKIF_STATE_CONNECTED:
1758 * Potentially, the back-end may be signalling
1759 * a capacity change; update the capacity.
1761 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1762 "sectors", "%Lu", §ors);
1763 if (XENBUS_EXIST_ERR(err))
1765 printk(KERN_INFO "Setting capacity to %Lu\n",
1767 set_capacity(info->gd, sectors);
1768 revalidate_disk(info->gd);
1771 case BLKIF_STATE_SUSPENDED:
1773 * If we are recovering from suspension, we need to wait
1774 * for the backend to announce it's features before
1775 * reconnecting, at least we need to know if the backend
1776 * supports indirect descriptors, and how many.
1778 blkif_recover(info);
1785 dev_dbg(&info->xbdev->dev, "%s:%s.\n",
1786 __func__, info->xbdev->otherend);
1788 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1789 "sectors", "%llu", §ors,
1790 "info", "%u", &binfo,
1791 "sector-size", "%lu", §or_size,
1794 xenbus_dev_fatal(info->xbdev, err,
1795 "reading backend fields at %s",
1796 info->xbdev->otherend);
1801 * physcial-sector-size is a newer field, so old backends may not
1802 * provide this. Assume physical sector size to be the same as
1803 * sector_size in that case.
1805 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1806 "physical-sector-size", "%u", &physical_sector_size);
1808 physical_sector_size = sector_size;
1810 info->feature_flush = 0;
1813 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1814 "feature-barrier", "%d", &barrier,
1818 * If there's no "feature-barrier" defined, then it means
1819 * we're dealing with a very old backend which writes
1820 * synchronously; nothing to do.
1822 * If there are barriers, then we use flush.
1824 if (!err && barrier) {
1825 info->feature_flush = REQ_FLUSH | REQ_FUA;
1826 info->flush_op = BLKIF_OP_WRITE_BARRIER;
1829 * And if there is "feature-flush-cache" use that above
1832 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1833 "feature-flush-cache", "%d", &flush,
1836 if (!err && flush) {
1837 info->feature_flush = REQ_FLUSH;
1838 info->flush_op = BLKIF_OP_FLUSH_DISKCACHE;
1841 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1842 "feature-discard", "%d", &discard,
1845 if (!err && discard)
1846 blkfront_setup_discard(info);
1848 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1849 "feature-persistent", "%u", &persistent,
1852 info->feature_persistent = 0;
1854 info->feature_persistent = persistent;
1856 err = blkfront_setup_indirect(info);
1858 xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
1859 info->xbdev->otherend);
1863 err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
1864 physical_sector_size);
1866 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
1867 info->xbdev->otherend);
1871 xenbus_switch_state(info->xbdev, XenbusStateConnected);
1873 /* Kick pending requests. */
1874 spin_lock_irq(&info->io_lock);
1875 info->connected = BLKIF_STATE_CONNECTED;
1876 kick_pending_request_queues(info);
1877 spin_unlock_irq(&info->io_lock);
1885 * Callback received when the backend's state changes.
1887 static void blkback_changed(struct xenbus_device *dev,
1888 enum xenbus_state backend_state)
1890 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1892 dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1894 switch (backend_state) {
1895 case XenbusStateInitialising:
1896 case XenbusStateInitWait:
1897 case XenbusStateInitialised:
1898 case XenbusStateReconfiguring:
1899 case XenbusStateReconfigured:
1900 case XenbusStateUnknown:
1903 case XenbusStateConnected:
1904 blkfront_connect(info);
1907 case XenbusStateClosed:
1908 if (dev->state == XenbusStateClosed)
1910 /* Missed the backend's Closing state -- fallthrough */
1911 case XenbusStateClosing:
1912 blkfront_closing(info);
1917 static int blkfront_remove(struct xenbus_device *xbdev)
1919 struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
1920 struct block_device *bdev = NULL;
1921 struct gendisk *disk;
1923 dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
1925 blkif_free(info, 0);
1927 mutex_lock(&info->mutex);
1931 bdev = bdget_disk(disk, 0);
1934 mutex_unlock(&info->mutex);
1942 * The xbdev was removed before we reached the Closed
1943 * state. See if it's safe to remove the disk. If the bdev
1944 * isn't closed yet, we let release take care of it.
1947 mutex_lock(&bdev->bd_mutex);
1948 info = disk->private_data;
1950 dev_warn(disk_to_dev(disk),
1951 "%s was hot-unplugged, %d stale handles\n",
1952 xbdev->nodename, bdev->bd_openers);
1954 if (info && !bdev->bd_openers) {
1955 xlvbd_release_gendisk(info);
1956 disk->private_data = NULL;
1960 mutex_unlock(&bdev->bd_mutex);
1966 static int blkfront_is_ready(struct xenbus_device *dev)
1968 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1970 return info->is_ready && info->xbdev;
1973 static int blkif_open(struct block_device *bdev, fmode_t mode)
1975 struct gendisk *disk = bdev->bd_disk;
1976 struct blkfront_info *info;
1979 mutex_lock(&blkfront_mutex);
1981 info = disk->private_data;
1988 mutex_lock(&info->mutex);
1991 /* xbdev is closed */
1994 mutex_unlock(&info->mutex);
1997 mutex_unlock(&blkfront_mutex);
2001 static void blkif_release(struct gendisk *disk, fmode_t mode)
2003 struct blkfront_info *info = disk->private_data;
2004 struct block_device *bdev;
2005 struct xenbus_device *xbdev;
2007 mutex_lock(&blkfront_mutex);
2009 bdev = bdget_disk(disk, 0);
2012 WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2015 if (bdev->bd_openers)
2019 * Check if we have been instructed to close. We will have
2020 * deferred this request, because the bdev was still open.
2023 mutex_lock(&info->mutex);
2024 xbdev = info->xbdev;
2026 if (xbdev && xbdev->state == XenbusStateClosing) {
2027 /* pending switch to state closed */
2028 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2029 xlvbd_release_gendisk(info);
2030 xenbus_frontend_closed(info->xbdev);
2033 mutex_unlock(&info->mutex);
2036 /* sudden device removal */
2037 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2038 xlvbd_release_gendisk(info);
2039 disk->private_data = NULL;
2046 mutex_unlock(&blkfront_mutex);
2049 static const struct block_device_operations xlvbd_block_fops =
2051 .owner = THIS_MODULE,
2053 .release = blkif_release,
2054 .getgeo = blkif_getgeo,
2055 .ioctl = blkif_ioctl,
2059 static const struct xenbus_device_id blkfront_ids[] = {
2064 static struct xenbus_driver blkfront_driver = {
2065 .ids = blkfront_ids,
2066 .probe = blkfront_probe,
2067 .remove = blkfront_remove,
2068 .resume = blkfront_resume,
2069 .otherend_changed = blkback_changed,
2070 .is_ready = blkfront_is_ready,
2073 static int __init xlblk_init(void)
2080 if (!xen_has_pv_disk_devices())
2083 if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2084 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
2085 XENVBD_MAJOR, DEV_NAME);
2089 ret = xenbus_register_frontend(&blkfront_driver);
2091 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2097 module_init(xlblk_init);
2100 static void __exit xlblk_exit(void)
2102 xenbus_unregister_driver(&blkfront_driver);
2103 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2106 module_exit(xlblk_exit);
2108 MODULE_DESCRIPTION("Xen virtual block device frontend");
2109 MODULE_LICENSE("GPL");
2110 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2111 MODULE_ALIAS("xen:vbd");
2112 MODULE_ALIAS("xenblk");