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 feature_discard:1;
130 unsigned int feature_secdiscard:1;
131 unsigned int discard_granularity;
132 unsigned int discard_alignment;
133 unsigned int feature_persistent:1;
134 unsigned int max_indirect_segments;
138 static unsigned int nr_minors;
139 static unsigned long *minors;
140 static DEFINE_SPINLOCK(minor_lock);
142 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
143 (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
144 #define GRANT_INVALID_REF 0
146 #define PARTS_PER_DISK 16
147 #define PARTS_PER_EXT_DISK 256
149 #define BLKIF_MAJOR(dev) ((dev)>>8)
150 #define BLKIF_MINOR(dev) ((dev) & 0xff)
153 #define EXTENDED (1<<EXT_SHIFT)
154 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
155 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
156 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
157 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
158 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
159 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
161 #define DEV_NAME "xvd" /* name in /dev */
163 #define SEGS_PER_INDIRECT_FRAME \
164 (PAGE_SIZE/sizeof(struct blkif_request_segment))
165 #define INDIRECT_GREFS(_segs) \
166 ((_segs + SEGS_PER_INDIRECT_FRAME - 1)/SEGS_PER_INDIRECT_FRAME)
168 static int blkfront_setup_indirect(struct blkfront_info *info);
170 static int get_id_from_freelist(struct blkfront_info *info)
172 unsigned long free = info->shadow_free;
173 BUG_ON(free >= BLK_RING_SIZE);
174 info->shadow_free = info->shadow[free].req.u.rw.id;
175 info->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
179 static int add_id_to_freelist(struct blkfront_info *info,
182 if (info->shadow[id].req.u.rw.id != id)
184 if (info->shadow[id].request == NULL)
186 info->shadow[id].req.u.rw.id = info->shadow_free;
187 info->shadow[id].request = NULL;
188 info->shadow_free = id;
192 static int fill_grant_buffer(struct blkfront_info *info, int num)
194 struct page *granted_page;
195 struct grant *gnt_list_entry, *n;
199 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
203 if (info->feature_persistent) {
204 granted_page = alloc_page(GFP_NOIO);
206 kfree(gnt_list_entry);
209 gnt_list_entry->pfn = page_to_pfn(granted_page);
212 gnt_list_entry->gref = GRANT_INVALID_REF;
213 list_add(&gnt_list_entry->node, &info->grants);
220 list_for_each_entry_safe(gnt_list_entry, n,
221 &info->grants, node) {
222 list_del(&gnt_list_entry->node);
223 if (info->feature_persistent)
224 __free_page(pfn_to_page(gnt_list_entry->pfn));
225 kfree(gnt_list_entry);
232 static struct grant *get_grant(grant_ref_t *gref_head,
234 struct blkfront_info *info)
236 struct grant *gnt_list_entry;
237 unsigned long buffer_mfn;
239 BUG_ON(list_empty(&info->grants));
240 gnt_list_entry = list_first_entry(&info->grants, struct grant,
242 list_del(&gnt_list_entry->node);
244 if (gnt_list_entry->gref != GRANT_INVALID_REF) {
245 info->persistent_gnts_c--;
246 return gnt_list_entry;
249 /* Assign a gref to this page */
250 gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
251 BUG_ON(gnt_list_entry->gref == -ENOSPC);
252 if (!info->feature_persistent) {
254 gnt_list_entry->pfn = pfn;
256 buffer_mfn = pfn_to_mfn(gnt_list_entry->pfn);
257 gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
258 info->xbdev->otherend_id,
260 return gnt_list_entry;
263 static const char *op_name(int op)
265 static const char *const names[] = {
266 [BLKIF_OP_READ] = "read",
267 [BLKIF_OP_WRITE] = "write",
268 [BLKIF_OP_WRITE_BARRIER] = "barrier",
269 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
270 [BLKIF_OP_DISCARD] = "discard" };
272 if (op < 0 || op >= ARRAY_SIZE(names))
280 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
282 unsigned int end = minor + nr;
285 if (end > nr_minors) {
286 unsigned long *bitmap, *old;
288 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
293 spin_lock(&minor_lock);
294 if (end > nr_minors) {
296 memcpy(bitmap, minors,
297 BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
299 nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
302 spin_unlock(&minor_lock);
306 spin_lock(&minor_lock);
307 if (find_next_bit(minors, end, minor) >= end) {
308 bitmap_set(minors, minor, nr);
312 spin_unlock(&minor_lock);
317 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
319 unsigned int end = minor + nr;
321 BUG_ON(end > nr_minors);
322 spin_lock(&minor_lock);
323 bitmap_clear(minors, minor, nr);
324 spin_unlock(&minor_lock);
327 static void blkif_restart_queue_callback(void *arg)
329 struct blkfront_info *info = (struct blkfront_info *)arg;
330 schedule_work(&info->work);
333 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
335 /* We don't have real geometry info, but let's at least return
336 values consistent with the size of the device */
337 sector_t nsect = get_capacity(bd->bd_disk);
338 sector_t cylinders = nsect;
342 sector_div(cylinders, hg->heads * hg->sectors);
343 hg->cylinders = cylinders;
344 if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
345 hg->cylinders = 0xffff;
349 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
350 unsigned command, unsigned long argument)
352 struct blkfront_info *info = bdev->bd_disk->private_data;
355 dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
356 command, (long)argument);
359 case CDROMMULTISESSION:
360 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
361 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
362 if (put_user(0, (char __user *)(argument + i)))
366 case CDROM_GET_CAPABILITY: {
367 struct gendisk *gd = info->gd;
368 if (gd->flags & GENHD_FL_CD)
374 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
376 return -EINVAL; /* same return as native Linux */
383 * Generate a Xen blkfront IO request from a blk layer request. Reads
384 * and writes are handled as expected.
386 * @req: a request struct
388 static int blkif_queue_request(struct request *req)
390 struct blkfront_info *info = req->rq_disk->private_data;
391 struct blkif_request *ring_req;
393 unsigned int fsect, lsect;
395 struct blkif_request_segment *segments = NULL;
398 * Used to store if we are able to queue the request by just using
399 * existing persistent grants, or if we have to get new grants,
400 * as there are not sufficiently many free.
402 bool new_persistent_gnts;
403 grant_ref_t gref_head;
404 struct grant *gnt_list_entry = NULL;
405 struct scatterlist *sg;
408 if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
411 max_grefs = req->nr_phys_segments;
412 if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
414 * If we are using indirect segments we need to account
415 * for the indirect grefs used in the request.
417 max_grefs += INDIRECT_GREFS(req->nr_phys_segments);
419 /* Check if we have enough grants to allocate a requests */
420 if (info->persistent_gnts_c < max_grefs) {
421 new_persistent_gnts = 1;
422 if (gnttab_alloc_grant_references(
423 max_grefs - info->persistent_gnts_c,
425 gnttab_request_free_callback(
427 blkif_restart_queue_callback,
433 new_persistent_gnts = 0;
435 /* Fill out a communications ring structure. */
436 ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
437 id = get_id_from_freelist(info);
438 info->shadow[id].request = req;
440 if (unlikely(req->cmd_flags & (REQ_DISCARD | REQ_SECURE))) {
441 ring_req->operation = BLKIF_OP_DISCARD;
442 ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
443 ring_req->u.discard.id = id;
444 ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
445 if ((req->cmd_flags & REQ_SECURE) && info->feature_secdiscard)
446 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
448 ring_req->u.discard.flag = 0;
450 BUG_ON(info->max_indirect_segments == 0 &&
451 req->nr_phys_segments > BLKIF_MAX_SEGMENTS_PER_REQUEST);
452 BUG_ON(info->max_indirect_segments &&
453 req->nr_phys_segments > info->max_indirect_segments);
454 nseg = blk_rq_map_sg(req->q, req, info->shadow[id].sg);
455 ring_req->u.rw.id = id;
456 if (nseg > BLKIF_MAX_SEGMENTS_PER_REQUEST) {
458 * The indirect operation can only be a BLKIF_OP_READ or
461 BUG_ON(req->cmd_flags & (REQ_FLUSH | REQ_FUA));
462 ring_req->operation = BLKIF_OP_INDIRECT;
463 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
464 BLKIF_OP_WRITE : BLKIF_OP_READ;
465 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
466 ring_req->u.indirect.handle = info->handle;
467 ring_req->u.indirect.nr_segments = nseg;
469 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
470 ring_req->u.rw.handle = info->handle;
471 ring_req->operation = rq_data_dir(req) ?
472 BLKIF_OP_WRITE : BLKIF_OP_READ;
473 if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) {
475 * Ideally we can do an unordered flush-to-disk. In case the
476 * backend onlysupports barriers, use that. A barrier request
477 * a superset of FUA, so we can implement it the same
478 * way. (It's also a FLUSH+FUA, since it is
479 * guaranteed ordered WRT previous writes.)
481 switch (info->feature_flush &
482 ((REQ_FLUSH|REQ_FUA))) {
483 case REQ_FLUSH|REQ_FUA:
484 ring_req->operation =
485 BLKIF_OP_WRITE_BARRIER;
488 ring_req->operation =
489 BLKIF_OP_FLUSH_DISKCACHE;
492 ring_req->operation = 0;
495 ring_req->u.rw.nr_segments = nseg;
497 for_each_sg(info->shadow[id].sg, sg, nseg, i) {
498 fsect = sg->offset >> 9;
499 lsect = fsect + (sg->length >> 9) - 1;
501 if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
502 (i % SEGS_PER_INDIRECT_FRAME == 0)) {
503 unsigned long uninitialized_var(pfn);
506 kunmap_atomic(segments);
508 n = i / SEGS_PER_INDIRECT_FRAME;
509 if (!info->feature_persistent) {
510 struct page *indirect_page;
512 /* Fetch a pre-allocated page to use for indirect grefs */
513 BUG_ON(list_empty(&info->indirect_pages));
514 indirect_page = list_first_entry(&info->indirect_pages,
516 list_del(&indirect_page->lru);
517 pfn = page_to_pfn(indirect_page);
519 gnt_list_entry = get_grant(&gref_head, pfn, info);
520 info->shadow[id].indirect_grants[n] = gnt_list_entry;
521 segments = kmap_atomic(pfn_to_page(gnt_list_entry->pfn));
522 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
525 gnt_list_entry = get_grant(&gref_head, page_to_pfn(sg_page(sg)), info);
526 ref = gnt_list_entry->gref;
528 info->shadow[id].grants_used[i] = gnt_list_entry;
530 if (rq_data_dir(req) && info->feature_persistent) {
534 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
536 shared_data = kmap_atomic(pfn_to_page(gnt_list_entry->pfn));
537 bvec_data = kmap_atomic(sg_page(sg));
540 * this does not wipe data stored outside the
541 * range sg->offset..sg->offset+sg->length.
542 * Therefore, blkback *could* see data from
543 * previous requests. This is OK as long as
544 * persistent grants are shared with just one
545 * domain. It may need refactoring if this
548 memcpy(shared_data + sg->offset,
549 bvec_data + sg->offset,
552 kunmap_atomic(bvec_data);
553 kunmap_atomic(shared_data);
555 if (ring_req->operation != BLKIF_OP_INDIRECT) {
556 ring_req->u.rw.seg[i] =
557 (struct blkif_request_segment) {
560 .last_sect = lsect };
562 n = i % SEGS_PER_INDIRECT_FRAME;
564 (struct blkif_request_segment) {
567 .last_sect = lsect };
571 kunmap_atomic(segments);
574 info->ring.req_prod_pvt++;
576 /* Keep a private copy so we can reissue requests when recovering. */
577 info->shadow[id].req = *ring_req;
579 if (new_persistent_gnts)
580 gnttab_free_grant_references(gref_head);
586 static inline void flush_requests(struct blkfront_info *info)
590 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
593 notify_remote_via_irq(info->irq);
596 static inline bool blkif_request_flush_invalid(struct request *req,
597 struct blkfront_info *info)
599 return ((req->cmd_type != REQ_TYPE_FS) ||
600 ((req->cmd_flags & REQ_FLUSH) &&
601 !(info->feature_flush & REQ_FLUSH)) ||
602 ((req->cmd_flags & REQ_FUA) &&
603 !(info->feature_flush & REQ_FUA)));
608 * read a block; request is in a request queue
610 static void do_blkif_request(struct request_queue *rq)
612 struct blkfront_info *info = NULL;
616 pr_debug("Entered do_blkif_request\n");
620 while ((req = blk_peek_request(rq)) != NULL) {
621 info = req->rq_disk->private_data;
623 if (RING_FULL(&info->ring))
626 blk_start_request(req);
628 if (blkif_request_flush_invalid(req, info)) {
629 __blk_end_request_all(req, -EOPNOTSUPP);
633 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
635 req, req->cmd, (unsigned long)blk_rq_pos(req),
636 blk_rq_cur_sectors(req), blk_rq_sectors(req),
637 rq_data_dir(req) ? "write" : "read");
639 if (blkif_queue_request(req)) {
640 blk_requeue_request(rq, req);
642 /* Avoid pointless unplugs. */
651 flush_requests(info);
654 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
655 unsigned int physical_sector_size,
656 unsigned int segments)
658 struct request_queue *rq;
659 struct blkfront_info *info = gd->private_data;
661 rq = blk_init_queue(do_blkif_request, &info->io_lock);
665 queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
667 if (info->feature_discard) {
668 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
669 blk_queue_max_discard_sectors(rq, get_capacity(gd));
670 rq->limits.discard_granularity = info->discard_granularity;
671 rq->limits.discard_alignment = info->discard_alignment;
672 if (info->feature_secdiscard)
673 queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD, rq);
676 /* Hard sector size and max sectors impersonate the equiv. hardware. */
677 blk_queue_logical_block_size(rq, sector_size);
678 blk_queue_physical_block_size(rq, physical_sector_size);
679 blk_queue_max_hw_sectors(rq, (segments * PAGE_SIZE) / 512);
681 /* Each segment in a request is up to an aligned page in size. */
682 blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
683 blk_queue_max_segment_size(rq, PAGE_SIZE);
685 /* Ensure a merged request will fit in a single I/O ring slot. */
686 blk_queue_max_segments(rq, segments);
688 /* Make sure buffer addresses are sector-aligned. */
689 blk_queue_dma_alignment(rq, 511);
691 /* Make sure we don't use bounce buffers. */
692 blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
699 static const char *flush_info(unsigned int feature_flush)
701 switch (feature_flush & ((REQ_FLUSH | REQ_FUA))) {
702 case REQ_FLUSH|REQ_FUA:
703 return "barrier: enabled;";
705 return "flush diskcache: enabled;";
707 return "barrier or flush: disabled;";
711 static void xlvbd_flush(struct blkfront_info *info)
713 blk_queue_flush(info->rq, info->feature_flush);
714 pr_info("blkfront: %s: %s %s %s %s %s\n",
715 info->gd->disk_name, flush_info(info->feature_flush),
716 "persistent grants:", info->feature_persistent ?
717 "enabled;" : "disabled;", "indirect descriptors:",
718 info->max_indirect_segments ? "enabled;" : "disabled;");
721 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
724 major = BLKIF_MAJOR(vdevice);
725 *minor = BLKIF_MINOR(vdevice);
728 *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
729 *minor = ((*minor / 64) * PARTS_PER_DISK) +
730 EMULATED_HD_DISK_MINOR_OFFSET;
733 *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
734 *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
735 EMULATED_HD_DISK_MINOR_OFFSET;
737 case XEN_SCSI_DISK0_MAJOR:
738 *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
739 *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
741 case XEN_SCSI_DISK1_MAJOR:
742 case XEN_SCSI_DISK2_MAJOR:
743 case XEN_SCSI_DISK3_MAJOR:
744 case XEN_SCSI_DISK4_MAJOR:
745 case XEN_SCSI_DISK5_MAJOR:
746 case XEN_SCSI_DISK6_MAJOR:
747 case XEN_SCSI_DISK7_MAJOR:
748 *offset = (*minor / PARTS_PER_DISK) +
749 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
750 EMULATED_SD_DISK_NAME_OFFSET;
752 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
753 EMULATED_SD_DISK_MINOR_OFFSET;
755 case XEN_SCSI_DISK8_MAJOR:
756 case XEN_SCSI_DISK9_MAJOR:
757 case XEN_SCSI_DISK10_MAJOR:
758 case XEN_SCSI_DISK11_MAJOR:
759 case XEN_SCSI_DISK12_MAJOR:
760 case XEN_SCSI_DISK13_MAJOR:
761 case XEN_SCSI_DISK14_MAJOR:
762 case XEN_SCSI_DISK15_MAJOR:
763 *offset = (*minor / PARTS_PER_DISK) +
764 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
765 EMULATED_SD_DISK_NAME_OFFSET;
767 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
768 EMULATED_SD_DISK_MINOR_OFFSET;
771 *offset = *minor / PARTS_PER_DISK;
774 printk(KERN_WARNING "blkfront: your disk configuration is "
775 "incorrect, please use an xvd device instead\n");
781 static char *encode_disk_name(char *ptr, unsigned int n)
784 ptr = encode_disk_name(ptr, n / 26 - 1);
789 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
790 struct blkfront_info *info,
791 u16 vdisk_info, u16 sector_size,
792 unsigned int physical_sector_size)
802 BUG_ON(info->gd != NULL);
803 BUG_ON(info->rq != NULL);
805 if ((info->vdevice>>EXT_SHIFT) > 1) {
806 /* this is above the extended range; something is wrong */
807 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
811 if (!VDEV_IS_EXTENDED(info->vdevice)) {
812 err = xen_translate_vdev(info->vdevice, &minor, &offset);
815 nr_parts = PARTS_PER_DISK;
817 minor = BLKIF_MINOR_EXT(info->vdevice);
818 nr_parts = PARTS_PER_EXT_DISK;
819 offset = minor / nr_parts;
820 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
821 printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
822 "emulated IDE disks,\n\t choose an xvd device name"
823 "from xvde on\n", info->vdevice);
825 if (minor >> MINORBITS) {
826 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
827 info->vdevice, minor);
831 if ((minor % nr_parts) == 0)
832 nr_minors = nr_parts;
834 err = xlbd_reserve_minors(minor, nr_minors);
839 gd = alloc_disk(nr_minors);
843 strcpy(gd->disk_name, DEV_NAME);
844 ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
845 BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
849 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
850 "%d", minor & (nr_parts - 1));
852 gd->major = XENVBD_MAJOR;
853 gd->first_minor = minor;
854 gd->fops = &xlvbd_block_fops;
855 gd->private_data = info;
856 gd->driverfs_dev = &(info->xbdev->dev);
857 set_capacity(gd, capacity);
859 if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size,
860 info->max_indirect_segments ? :
861 BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
866 info->rq = gd->queue;
871 if (vdisk_info & VDISK_READONLY)
874 if (vdisk_info & VDISK_REMOVABLE)
875 gd->flags |= GENHD_FL_REMOVABLE;
877 if (vdisk_info & VDISK_CDROM)
878 gd->flags |= GENHD_FL_CD;
883 xlbd_release_minors(minor, nr_minors);
888 static void xlvbd_release_gendisk(struct blkfront_info *info)
890 unsigned int minor, nr_minors;
893 if (info->rq == NULL)
896 spin_lock_irqsave(&info->io_lock, flags);
898 /* No more blkif_request(). */
899 blk_stop_queue(info->rq);
901 /* No more gnttab callback work. */
902 gnttab_cancel_free_callback(&info->callback);
903 spin_unlock_irqrestore(&info->io_lock, flags);
905 /* Flush gnttab callback work. Must be done with no locks held. */
906 flush_work(&info->work);
908 del_gendisk(info->gd);
910 minor = info->gd->first_minor;
911 nr_minors = info->gd->minors;
912 xlbd_release_minors(minor, nr_minors);
914 blk_cleanup_queue(info->rq);
921 static void kick_pending_request_queues(struct blkfront_info *info)
923 if (!RING_FULL(&info->ring)) {
924 /* Re-enable calldowns. */
925 blk_start_queue(info->rq);
926 /* Kick things off immediately. */
927 do_blkif_request(info->rq);
931 static void blkif_restart_queue(struct work_struct *work)
933 struct blkfront_info *info = container_of(work, struct blkfront_info, work);
935 spin_lock_irq(&info->io_lock);
936 if (info->connected == BLKIF_STATE_CONNECTED)
937 kick_pending_request_queues(info);
938 spin_unlock_irq(&info->io_lock);
941 static void blkif_free(struct blkfront_info *info, int suspend)
943 struct grant *persistent_gnt;
947 /* Prevent new requests being issued until we fix things up. */
948 spin_lock_irq(&info->io_lock);
949 info->connected = suspend ?
950 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
951 /* No more blkif_request(). */
953 blk_stop_queue(info->rq);
955 /* Remove all persistent grants */
956 if (!list_empty(&info->grants)) {
957 list_for_each_entry_safe(persistent_gnt, n,
958 &info->grants, node) {
959 list_del(&persistent_gnt->node);
960 if (persistent_gnt->gref != GRANT_INVALID_REF) {
961 gnttab_end_foreign_access(persistent_gnt->gref,
963 info->persistent_gnts_c--;
965 if (info->feature_persistent)
966 __free_page(pfn_to_page(persistent_gnt->pfn));
967 kfree(persistent_gnt);
970 BUG_ON(info->persistent_gnts_c != 0);
973 * Remove indirect pages, this only happens when using indirect
974 * descriptors but not persistent grants
976 if (!list_empty(&info->indirect_pages)) {
977 struct page *indirect_page, *n;
979 BUG_ON(info->feature_persistent);
980 list_for_each_entry_safe(indirect_page, n, &info->indirect_pages, lru) {
981 list_del(&indirect_page->lru);
982 __free_page(indirect_page);
986 for (i = 0; i < BLK_RING_SIZE; i++) {
988 * Clear persistent grants present in requests already
991 if (!info->shadow[i].request)
994 segs = info->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
995 info->shadow[i].req.u.indirect.nr_segments :
996 info->shadow[i].req.u.rw.nr_segments;
997 for (j = 0; j < segs; j++) {
998 persistent_gnt = info->shadow[i].grants_used[j];
999 gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1000 if (info->feature_persistent)
1001 __free_page(pfn_to_page(persistent_gnt->pfn));
1002 kfree(persistent_gnt);
1005 if (info->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1007 * If this is not an indirect operation don't try to
1008 * free indirect segments
1012 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1013 persistent_gnt = info->shadow[i].indirect_grants[j];
1014 gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1015 __free_page(pfn_to_page(persistent_gnt->pfn));
1016 kfree(persistent_gnt);
1020 kfree(info->shadow[i].grants_used);
1021 info->shadow[i].grants_used = NULL;
1022 kfree(info->shadow[i].indirect_grants);
1023 info->shadow[i].indirect_grants = NULL;
1024 kfree(info->shadow[i].sg);
1025 info->shadow[i].sg = NULL;
1028 /* No more gnttab callback work. */
1029 gnttab_cancel_free_callback(&info->callback);
1030 spin_unlock_irq(&info->io_lock);
1032 /* Flush gnttab callback work. Must be done with no locks held. */
1033 flush_work(&info->work);
1035 /* Free resources associated with old device channel. */
1036 if (info->ring_ref != GRANT_INVALID_REF) {
1037 gnttab_end_foreign_access(info->ring_ref, 0,
1038 (unsigned long)info->ring.sring);
1039 info->ring_ref = GRANT_INVALID_REF;
1040 info->ring.sring = NULL;
1043 unbind_from_irqhandler(info->irq, info);
1044 info->evtchn = info->irq = 0;
1048 static void blkif_completion(struct blk_shadow *s, struct blkfront_info *info,
1049 struct blkif_response *bret)
1052 struct scatterlist *sg;
1057 nseg = s->req.operation == BLKIF_OP_INDIRECT ?
1058 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1060 if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1062 * Copy the data received from the backend into the bvec.
1063 * Since bv_offset can be different than 0, and bv_len different
1064 * than PAGE_SIZE, we have to keep track of the current offset,
1065 * to be sure we are copying the data from the right shared page.
1067 for_each_sg(s->sg, sg, nseg, i) {
1068 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1069 shared_data = kmap_atomic(
1070 pfn_to_page(s->grants_used[i]->pfn));
1071 bvec_data = kmap_atomic(sg_page(sg));
1072 memcpy(bvec_data + sg->offset,
1073 shared_data + sg->offset,
1075 kunmap_atomic(bvec_data);
1076 kunmap_atomic(shared_data);
1079 /* Add the persistent grant into the list of free grants */
1080 for (i = 0; i < nseg; i++) {
1081 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1083 * If the grant is still mapped by the backend (the
1084 * backend has chosen to make this grant persistent)
1085 * we add it at the head of the list, so it will be
1088 if (!info->feature_persistent)
1089 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1090 s->grants_used[i]->gref);
1091 list_add(&s->grants_used[i]->node, &info->grants);
1092 info->persistent_gnts_c++;
1095 * If the grant is not mapped by the backend we end the
1096 * foreign access and add it to the tail of the list,
1097 * so it will not be picked again unless we run out of
1098 * persistent grants.
1100 gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1101 s->grants_used[i]->gref = GRANT_INVALID_REF;
1102 list_add_tail(&s->grants_used[i]->node, &info->grants);
1105 if (s->req.operation == BLKIF_OP_INDIRECT) {
1106 for (i = 0; i < INDIRECT_GREFS(nseg); i++) {
1107 if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1108 if (!info->feature_persistent)
1109 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1110 s->indirect_grants[i]->gref);
1111 list_add(&s->indirect_grants[i]->node, &info->grants);
1112 info->persistent_gnts_c++;
1114 struct page *indirect_page;
1116 gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1118 * Add the used indirect page back to the list of
1119 * available pages for indirect grefs.
1121 indirect_page = pfn_to_page(s->indirect_grants[i]->pfn);
1122 list_add(&indirect_page->lru, &info->indirect_pages);
1123 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1124 list_add_tail(&s->indirect_grants[i]->node, &info->grants);
1130 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1132 struct request *req;
1133 struct blkif_response *bret;
1135 unsigned long flags;
1136 struct blkfront_info *info = (struct blkfront_info *)dev_id;
1139 spin_lock_irqsave(&info->io_lock, flags);
1141 if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
1142 spin_unlock_irqrestore(&info->io_lock, flags);
1147 rp = info->ring.sring->rsp_prod;
1148 rmb(); /* Ensure we see queued responses up to 'rp'. */
1150 for (i = info->ring.rsp_cons; i != rp; i++) {
1153 bret = RING_GET_RESPONSE(&info->ring, i);
1156 * The backend has messed up and given us an id that we would
1157 * never have given to it (we stamp it up to BLK_RING_SIZE -
1158 * look in get_id_from_freelist.
1160 if (id >= BLK_RING_SIZE) {
1161 WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1162 info->gd->disk_name, op_name(bret->operation), id);
1163 /* We can't safely get the 'struct request' as
1164 * the id is busted. */
1167 req = info->shadow[id].request;
1169 if (bret->operation != BLKIF_OP_DISCARD)
1170 blkif_completion(&info->shadow[id], info, bret);
1172 if (add_id_to_freelist(info, id)) {
1173 WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1174 info->gd->disk_name, op_name(bret->operation), id);
1178 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
1179 switch (bret->operation) {
1180 case BLKIF_OP_DISCARD:
1181 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1182 struct request_queue *rq = info->rq;
1183 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1184 info->gd->disk_name, op_name(bret->operation));
1185 error = -EOPNOTSUPP;
1186 info->feature_discard = 0;
1187 info->feature_secdiscard = 0;
1188 queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1189 queue_flag_clear(QUEUE_FLAG_SECDISCARD, rq);
1191 __blk_end_request_all(req, error);
1193 case BLKIF_OP_FLUSH_DISKCACHE:
1194 case BLKIF_OP_WRITE_BARRIER:
1195 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1196 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1197 info->gd->disk_name, op_name(bret->operation));
1198 error = -EOPNOTSUPP;
1200 if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1201 info->shadow[id].req.u.rw.nr_segments == 0)) {
1202 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1203 info->gd->disk_name, op_name(bret->operation));
1204 error = -EOPNOTSUPP;
1206 if (unlikely(error)) {
1207 if (error == -EOPNOTSUPP)
1209 info->feature_flush = 0;
1214 case BLKIF_OP_WRITE:
1215 if (unlikely(bret->status != BLKIF_RSP_OKAY))
1216 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1217 "request: %x\n", bret->status);
1219 __blk_end_request_all(req, error);
1226 info->ring.rsp_cons = i;
1228 if (i != info->ring.req_prod_pvt) {
1230 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
1234 info->ring.sring->rsp_event = i + 1;
1236 kick_pending_request_queues(info);
1238 spin_unlock_irqrestore(&info->io_lock, flags);
1244 static int setup_blkring(struct xenbus_device *dev,
1245 struct blkfront_info *info)
1247 struct blkif_sring *sring;
1251 info->ring_ref = GRANT_INVALID_REF;
1253 sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
1255 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1258 SHARED_RING_INIT(sring);
1259 FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
1261 err = xenbus_grant_ring(dev, info->ring.sring, 1, &gref);
1263 free_page((unsigned long)sring);
1264 info->ring.sring = NULL;
1267 info->ring_ref = gref;
1269 err = xenbus_alloc_evtchn(dev, &info->evtchn);
1273 err = bind_evtchn_to_irqhandler(info->evtchn, blkif_interrupt, 0,
1276 xenbus_dev_fatal(dev, err,
1277 "bind_evtchn_to_irqhandler failed");
1284 blkif_free(info, 0);
1289 /* Common code used when first setting up, and when resuming. */
1290 static int talk_to_blkback(struct xenbus_device *dev,
1291 struct blkfront_info *info)
1293 const char *message = NULL;
1294 struct xenbus_transaction xbt;
1297 /* Create shared ring, alloc event channel. */
1298 err = setup_blkring(dev, info);
1303 err = xenbus_transaction_start(&xbt);
1305 xenbus_dev_fatal(dev, err, "starting transaction");
1306 goto destroy_blkring;
1309 err = xenbus_printf(xbt, dev->nodename,
1310 "ring-ref", "%u", info->ring_ref);
1312 message = "writing ring-ref";
1313 goto abort_transaction;
1315 err = xenbus_printf(xbt, dev->nodename,
1316 "event-channel", "%u", info->evtchn);
1318 message = "writing event-channel";
1319 goto abort_transaction;
1321 err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1322 XEN_IO_PROTO_ABI_NATIVE);
1324 message = "writing protocol";
1325 goto abort_transaction;
1327 err = xenbus_printf(xbt, dev->nodename,
1328 "feature-persistent", "%u", 1);
1331 "writing persistent grants feature to xenbus");
1333 err = xenbus_transaction_end(xbt, 0);
1337 xenbus_dev_fatal(dev, err, "completing transaction");
1338 goto destroy_blkring;
1341 xenbus_switch_state(dev, XenbusStateInitialised);
1346 xenbus_transaction_end(xbt, 1);
1348 xenbus_dev_fatal(dev, err, "%s", message);
1350 blkif_free(info, 0);
1356 * Entry point to this code when a new device is created. Allocate the basic
1357 * structures and the ring buffer for communication with the backend, and
1358 * inform the backend of the appropriate details for those. Switch to
1359 * Initialised state.
1361 static int blkfront_probe(struct xenbus_device *dev,
1362 const struct xenbus_device_id *id)
1364 int err, vdevice, i;
1365 struct blkfront_info *info;
1367 /* FIXME: Use dynamic device id if this is not set. */
1368 err = xenbus_scanf(XBT_NIL, dev->nodename,
1369 "virtual-device", "%i", &vdevice);
1371 /* go looking in the extended area instead */
1372 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1375 xenbus_dev_fatal(dev, err, "reading virtual-device");
1380 if (xen_hvm_domain()) {
1383 /* no unplug has been done: do not hook devices != xen vbds */
1384 if (xen_has_pv_and_legacy_disk_devices()) {
1387 if (!VDEV_IS_EXTENDED(vdevice))
1388 major = BLKIF_MAJOR(vdevice);
1390 major = XENVBD_MAJOR;
1392 if (major != XENVBD_MAJOR) {
1394 "%s: HVM does not support vbd %d as xen block device\n",
1399 /* do not create a PV cdrom device if we are an HVM guest */
1400 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1403 if (strncmp(type, "cdrom", 5) == 0) {
1409 info = kzalloc(sizeof(*info), GFP_KERNEL);
1411 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1415 mutex_init(&info->mutex);
1416 spin_lock_init(&info->io_lock);
1418 info->vdevice = vdevice;
1419 INIT_LIST_HEAD(&info->grants);
1420 INIT_LIST_HEAD(&info->indirect_pages);
1421 info->persistent_gnts_c = 0;
1422 info->connected = BLKIF_STATE_DISCONNECTED;
1423 INIT_WORK(&info->work, blkif_restart_queue);
1425 for (i = 0; i < BLK_RING_SIZE; i++)
1426 info->shadow[i].req.u.rw.id = i+1;
1427 info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1429 /* Front end dir is a number, which is used as the id. */
1430 info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1431 dev_set_drvdata(&dev->dev, info);
1433 err = talk_to_blkback(dev, info);
1436 dev_set_drvdata(&dev->dev, NULL);
1443 static void split_bio_end(struct bio *bio, int error)
1445 struct split_bio *split_bio = bio->bi_private;
1448 split_bio->err = error;
1450 if (atomic_dec_and_test(&split_bio->pending)) {
1451 split_bio->bio->bi_phys_segments = 0;
1452 bio_endio(split_bio->bio, split_bio->err);
1458 static int blkif_recover(struct blkfront_info *info)
1461 struct request *req, *n;
1462 struct blk_shadow *copy;
1464 struct bio *bio, *cloned_bio;
1465 struct bio_list bio_list, merge_bio;
1466 unsigned int segs, offset;
1468 struct split_bio *split_bio;
1469 struct list_head requests;
1471 /* Stage 1: Make a safe copy of the shadow state. */
1472 copy = kmemdup(info->shadow, sizeof(info->shadow),
1473 GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
1477 /* Stage 2: Set up free list. */
1478 memset(&info->shadow, 0, sizeof(info->shadow));
1479 for (i = 0; i < BLK_RING_SIZE; i++)
1480 info->shadow[i].req.u.rw.id = i+1;
1481 info->shadow_free = info->ring.req_prod_pvt;
1482 info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1484 rc = blkfront_setup_indirect(info);
1490 segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
1491 blk_queue_max_segments(info->rq, segs);
1492 bio_list_init(&bio_list);
1493 INIT_LIST_HEAD(&requests);
1494 for (i = 0; i < BLK_RING_SIZE; i++) {
1496 if (!copy[i].request)
1500 * Get the bios in the request so we can re-queue them.
1502 if (copy[i].request->cmd_flags &
1503 (REQ_FLUSH | REQ_FUA | REQ_DISCARD | REQ_SECURE)) {
1505 * Flush operations don't contain bios, so
1506 * we need to requeue the whole request
1508 list_add(©[i].request->queuelist, &requests);
1511 merge_bio.head = copy[i].request->bio;
1512 merge_bio.tail = copy[i].request->biotail;
1513 bio_list_merge(&bio_list, &merge_bio);
1514 copy[i].request->bio = NULL;
1515 blk_end_request_all(copy[i].request, 0);
1521 * Empty the queue, this is important because we might have
1522 * requests in the queue with more segments than what we
1525 spin_lock_irq(&info->io_lock);
1526 while ((req = blk_fetch_request(info->rq)) != NULL) {
1527 if (req->cmd_flags &
1528 (REQ_FLUSH | REQ_FUA | REQ_DISCARD | REQ_SECURE)) {
1529 list_add(&req->queuelist, &requests);
1532 merge_bio.head = req->bio;
1533 merge_bio.tail = req->biotail;
1534 bio_list_merge(&bio_list, &merge_bio);
1536 if (req->cmd_flags & (REQ_FLUSH | REQ_FUA))
1537 pr_alert("diskcache flush request found!\n");
1538 __blk_end_request_all(req, 0);
1540 spin_unlock_irq(&info->io_lock);
1542 xenbus_switch_state(info->xbdev, XenbusStateConnected);
1544 spin_lock_irq(&info->io_lock);
1546 /* Now safe for us to use the shared ring */
1547 info->connected = BLKIF_STATE_CONNECTED;
1549 /* Kick any other new requests queued since we resumed */
1550 kick_pending_request_queues(info);
1552 list_for_each_entry_safe(req, n, &requests, queuelist) {
1553 /* Requeue pending requests (flush or discard) */
1554 list_del_init(&req->queuelist);
1555 BUG_ON(req->nr_phys_segments > segs);
1556 blk_requeue_request(info->rq, req);
1558 spin_unlock_irq(&info->io_lock);
1560 while ((bio = bio_list_pop(&bio_list)) != NULL) {
1561 /* Traverse the list of pending bios and re-queue them */
1562 if (bio_segments(bio) > segs) {
1564 * This bio has more segments than what we can
1565 * handle, we have to split it.
1567 pending = (bio_segments(bio) + segs - 1) / segs;
1568 split_bio = kzalloc(sizeof(*split_bio), GFP_NOIO);
1569 BUG_ON(split_bio == NULL);
1570 atomic_set(&split_bio->pending, pending);
1571 split_bio->bio = bio;
1572 for (i = 0; i < pending; i++) {
1573 offset = (i * segs * PAGE_SIZE) >> 9;
1574 size = min((unsigned int)(segs * PAGE_SIZE) >> 9,
1575 (unsigned int)bio_sectors(bio) - offset);
1576 cloned_bio = bio_clone(bio, GFP_NOIO);
1577 BUG_ON(cloned_bio == NULL);
1578 bio_trim(cloned_bio, offset, size);
1579 cloned_bio->bi_private = split_bio;
1580 cloned_bio->bi_end_io = split_bio_end;
1581 submit_bio(cloned_bio->bi_rw, cloned_bio);
1584 * Now we have to wait for all those smaller bios to
1585 * end, so we can also end the "parent" bio.
1589 /* We don't need to split this bio */
1590 submit_bio(bio->bi_rw, bio);
1597 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1598 * driver restart. We tear down our blkif structure and recreate it, but
1599 * leave the device-layer structures intact so that this is transparent to the
1600 * rest of the kernel.
1602 static int blkfront_resume(struct xenbus_device *dev)
1604 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1607 dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
1609 blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
1611 err = talk_to_blkback(dev, info);
1614 * We have to wait for the backend to switch to
1615 * connected state, since we want to read which
1616 * features it supports.
1623 blkfront_closing(struct blkfront_info *info)
1625 struct xenbus_device *xbdev = info->xbdev;
1626 struct block_device *bdev = NULL;
1628 mutex_lock(&info->mutex);
1630 if (xbdev->state == XenbusStateClosing) {
1631 mutex_unlock(&info->mutex);
1636 bdev = bdget_disk(info->gd, 0);
1638 mutex_unlock(&info->mutex);
1641 xenbus_frontend_closed(xbdev);
1645 mutex_lock(&bdev->bd_mutex);
1647 if (bdev->bd_openers) {
1648 xenbus_dev_error(xbdev, -EBUSY,
1649 "Device in use; refusing to close");
1650 xenbus_switch_state(xbdev, XenbusStateClosing);
1652 xlvbd_release_gendisk(info);
1653 xenbus_frontend_closed(xbdev);
1656 mutex_unlock(&bdev->bd_mutex);
1660 static void blkfront_setup_discard(struct blkfront_info *info)
1663 unsigned int discard_granularity;
1664 unsigned int discard_alignment;
1665 unsigned int discard_secure;
1667 info->feature_discard = 1;
1668 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1669 "discard-granularity", "%u", &discard_granularity,
1670 "discard-alignment", "%u", &discard_alignment,
1673 info->discard_granularity = discard_granularity;
1674 info->discard_alignment = discard_alignment;
1676 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1677 "discard-secure", "%d", &discard_secure,
1680 info->feature_secdiscard = !!discard_secure;
1683 static int blkfront_setup_indirect(struct blkfront_info *info)
1685 unsigned int indirect_segments, segs;
1688 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1689 "feature-max-indirect-segments", "%u", &indirect_segments,
1692 info->max_indirect_segments = 0;
1693 segs = BLKIF_MAX_SEGMENTS_PER_REQUEST;
1695 info->max_indirect_segments = min(indirect_segments,
1696 xen_blkif_max_segments);
1697 segs = info->max_indirect_segments;
1700 err = fill_grant_buffer(info, (segs + INDIRECT_GREFS(segs)) * BLK_RING_SIZE);
1704 if (!info->feature_persistent && info->max_indirect_segments) {
1706 * We are using indirect descriptors but not persistent
1707 * grants, we need to allocate a set of pages that can be
1708 * used for mapping indirect grefs
1710 int num = INDIRECT_GREFS(segs) * BLK_RING_SIZE;
1712 BUG_ON(!list_empty(&info->indirect_pages));
1713 for (i = 0; i < num; i++) {
1714 struct page *indirect_page = alloc_page(GFP_NOIO);
1717 list_add(&indirect_page->lru, &info->indirect_pages);
1721 for (i = 0; i < BLK_RING_SIZE; i++) {
1722 info->shadow[i].grants_used = kzalloc(
1723 sizeof(info->shadow[i].grants_used[0]) * segs,
1725 info->shadow[i].sg = kzalloc(sizeof(info->shadow[i].sg[0]) * segs, GFP_NOIO);
1726 if (info->max_indirect_segments)
1727 info->shadow[i].indirect_grants = kzalloc(
1728 sizeof(info->shadow[i].indirect_grants[0]) *
1729 INDIRECT_GREFS(segs),
1731 if ((info->shadow[i].grants_used == NULL) ||
1732 (info->shadow[i].sg == NULL) ||
1733 (info->max_indirect_segments &&
1734 (info->shadow[i].indirect_grants == NULL)))
1736 sg_init_table(info->shadow[i].sg, segs);
1743 for (i = 0; i < BLK_RING_SIZE; i++) {
1744 kfree(info->shadow[i].grants_used);
1745 info->shadow[i].grants_used = NULL;
1746 kfree(info->shadow[i].sg);
1747 info->shadow[i].sg = NULL;
1748 kfree(info->shadow[i].indirect_grants);
1749 info->shadow[i].indirect_grants = NULL;
1751 if (!list_empty(&info->indirect_pages)) {
1752 struct page *indirect_page, *n;
1753 list_for_each_entry_safe(indirect_page, n, &info->indirect_pages, lru) {
1754 list_del(&indirect_page->lru);
1755 __free_page(indirect_page);
1762 * Invoked when the backend is finally 'ready' (and has told produced
1763 * the details about the physical device - #sectors, size, etc).
1765 static void blkfront_connect(struct blkfront_info *info)
1767 unsigned long long sectors;
1768 unsigned long sector_size;
1769 unsigned int physical_sector_size;
1772 int barrier, flush, discard, persistent;
1774 switch (info->connected) {
1775 case BLKIF_STATE_CONNECTED:
1777 * Potentially, the back-end may be signalling
1778 * a capacity change; update the capacity.
1780 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1781 "sectors", "%Lu", §ors);
1782 if (XENBUS_EXIST_ERR(err))
1784 printk(KERN_INFO "Setting capacity to %Lu\n",
1786 set_capacity(info->gd, sectors);
1787 revalidate_disk(info->gd);
1790 case BLKIF_STATE_SUSPENDED:
1792 * If we are recovering from suspension, we need to wait
1793 * for the backend to announce it's features before
1794 * reconnecting, at least we need to know if the backend
1795 * supports indirect descriptors, and how many.
1797 blkif_recover(info);
1804 dev_dbg(&info->xbdev->dev, "%s:%s.\n",
1805 __func__, info->xbdev->otherend);
1807 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1808 "sectors", "%llu", §ors,
1809 "info", "%u", &binfo,
1810 "sector-size", "%lu", §or_size,
1813 xenbus_dev_fatal(info->xbdev, err,
1814 "reading backend fields at %s",
1815 info->xbdev->otherend);
1820 * physcial-sector-size is a newer field, so old backends may not
1821 * provide this. Assume physical sector size to be the same as
1822 * sector_size in that case.
1824 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1825 "physical-sector-size", "%u", &physical_sector_size);
1827 physical_sector_size = sector_size;
1829 info->feature_flush = 0;
1831 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1832 "feature-barrier", "%d", &barrier,
1836 * If there's no "feature-barrier" defined, then it means
1837 * we're dealing with a very old backend which writes
1838 * synchronously; nothing to do.
1840 * If there are barriers, then we use flush.
1842 if (!err && barrier)
1843 info->feature_flush = REQ_FLUSH | REQ_FUA;
1845 * And if there is "feature-flush-cache" use that above
1848 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1849 "feature-flush-cache", "%d", &flush,
1853 info->feature_flush = REQ_FLUSH;
1855 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1856 "feature-discard", "%d", &discard,
1859 if (!err && discard)
1860 blkfront_setup_discard(info);
1862 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1863 "feature-persistent", "%u", &persistent,
1866 info->feature_persistent = 0;
1868 info->feature_persistent = persistent;
1870 err = blkfront_setup_indirect(info);
1872 xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
1873 info->xbdev->otherend);
1877 err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
1878 physical_sector_size);
1880 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
1881 info->xbdev->otherend);
1885 xenbus_switch_state(info->xbdev, XenbusStateConnected);
1887 /* Kick pending requests. */
1888 spin_lock_irq(&info->io_lock);
1889 info->connected = BLKIF_STATE_CONNECTED;
1890 kick_pending_request_queues(info);
1891 spin_unlock_irq(&info->io_lock);
1899 * Callback received when the backend's state changes.
1901 static void blkback_changed(struct xenbus_device *dev,
1902 enum xenbus_state backend_state)
1904 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1906 dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1908 switch (backend_state) {
1909 case XenbusStateInitialising:
1910 case XenbusStateInitWait:
1911 case XenbusStateInitialised:
1912 case XenbusStateReconfiguring:
1913 case XenbusStateReconfigured:
1914 case XenbusStateUnknown:
1917 case XenbusStateConnected:
1918 blkfront_connect(info);
1921 case XenbusStateClosed:
1922 if (dev->state == XenbusStateClosed)
1924 /* Missed the backend's Closing state -- fallthrough */
1925 case XenbusStateClosing:
1926 blkfront_closing(info);
1931 static int blkfront_remove(struct xenbus_device *xbdev)
1933 struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
1934 struct block_device *bdev = NULL;
1935 struct gendisk *disk;
1937 dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
1939 blkif_free(info, 0);
1941 mutex_lock(&info->mutex);
1945 bdev = bdget_disk(disk, 0);
1948 mutex_unlock(&info->mutex);
1956 * The xbdev was removed before we reached the Closed
1957 * state. See if it's safe to remove the disk. If the bdev
1958 * isn't closed yet, we let release take care of it.
1961 mutex_lock(&bdev->bd_mutex);
1962 info = disk->private_data;
1964 dev_warn(disk_to_dev(disk),
1965 "%s was hot-unplugged, %d stale handles\n",
1966 xbdev->nodename, bdev->bd_openers);
1968 if (info && !bdev->bd_openers) {
1969 xlvbd_release_gendisk(info);
1970 disk->private_data = NULL;
1974 mutex_unlock(&bdev->bd_mutex);
1980 static int blkfront_is_ready(struct xenbus_device *dev)
1982 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1984 return info->is_ready && info->xbdev;
1987 static int blkif_open(struct block_device *bdev, fmode_t mode)
1989 struct gendisk *disk = bdev->bd_disk;
1990 struct blkfront_info *info;
1993 mutex_lock(&blkfront_mutex);
1995 info = disk->private_data;
2002 mutex_lock(&info->mutex);
2005 /* xbdev is closed */
2008 mutex_unlock(&info->mutex);
2011 mutex_unlock(&blkfront_mutex);
2015 static void blkif_release(struct gendisk *disk, fmode_t mode)
2017 struct blkfront_info *info = disk->private_data;
2018 struct block_device *bdev;
2019 struct xenbus_device *xbdev;
2021 mutex_lock(&blkfront_mutex);
2023 bdev = bdget_disk(disk, 0);
2026 WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2029 if (bdev->bd_openers)
2033 * Check if we have been instructed to close. We will have
2034 * deferred this request, because the bdev was still open.
2037 mutex_lock(&info->mutex);
2038 xbdev = info->xbdev;
2040 if (xbdev && xbdev->state == XenbusStateClosing) {
2041 /* pending switch to state closed */
2042 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2043 xlvbd_release_gendisk(info);
2044 xenbus_frontend_closed(info->xbdev);
2047 mutex_unlock(&info->mutex);
2050 /* sudden device removal */
2051 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2052 xlvbd_release_gendisk(info);
2053 disk->private_data = NULL;
2060 mutex_unlock(&blkfront_mutex);
2063 static const struct block_device_operations xlvbd_block_fops =
2065 .owner = THIS_MODULE,
2067 .release = blkif_release,
2068 .getgeo = blkif_getgeo,
2069 .ioctl = blkif_ioctl,
2073 static const struct xenbus_device_id blkfront_ids[] = {
2078 static struct xenbus_driver blkfront_driver = {
2079 .ids = blkfront_ids,
2080 .probe = blkfront_probe,
2081 .remove = blkfront_remove,
2082 .resume = blkfront_resume,
2083 .otherend_changed = blkback_changed,
2084 .is_ready = blkfront_is_ready,
2087 static int __init xlblk_init(void)
2094 if (!xen_has_pv_disk_devices())
2097 if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2098 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
2099 XENVBD_MAJOR, DEV_NAME);
2103 ret = xenbus_register_frontend(&blkfront_driver);
2105 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2111 module_init(xlblk_init);
2114 static void __exit xlblk_exit(void)
2116 xenbus_unregister_driver(&blkfront_driver);
2117 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2120 module_exit(xlblk_exit);
2122 MODULE_DESCRIPTION("Xen virtual block device frontend");
2123 MODULE_LICENSE("GPL");
2124 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2125 MODULE_ALIAS("xen:vbd");
2126 MODULE_ALIAS("xenblk");