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/smp_lock.h>
45 #include <linux/scatterlist.h>
48 #include <xen/xenbus.h>
49 #include <xen/grant_table.h>
50 #include <xen/events.h>
53 #include <xen/interface/grant_table.h>
54 #include <xen/interface/io/blkif.h>
55 #include <xen/interface/io/protocols.h>
57 #include <asm/xen/hypervisor.h>
60 BLKIF_STATE_DISCONNECTED,
61 BLKIF_STATE_CONNECTED,
62 BLKIF_STATE_SUSPENDED,
66 struct blkif_request req;
67 unsigned long request;
68 unsigned long frame[BLKIF_MAX_SEGMENTS_PER_REQUEST];
71 static const struct block_device_operations xlvbd_block_fops;
73 #define BLK_RING_SIZE __RING_SIZE((struct blkif_sring *)0, PAGE_SIZE)
76 * We have one of these per vbd, whether ide, scsi or 'other'. They
77 * hang in private_data off the gendisk structure. We may end up
78 * putting all kinds of interesting stuff here :-)
82 struct xenbus_device *xbdev;
86 enum blkif_state connected;
88 struct blkif_front_ring ring;
89 struct scatterlist sg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
90 unsigned int evtchn, irq;
91 struct request_queue *rq;
92 struct work_struct work;
93 struct gnttab_free_callback callback;
94 struct blk_shadow shadow[BLK_RING_SIZE];
95 unsigned long shadow_free;
100 * The number of people holding this device open. We won't allow a
101 * hot-unplug unless this is 0.
106 static DEFINE_SPINLOCK(blkif_io_lock);
108 static unsigned int nr_minors;
109 static unsigned long *minors;
110 static DEFINE_SPINLOCK(minor_lock);
112 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
113 (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
114 #define GRANT_INVALID_REF 0
116 #define PARTS_PER_DISK 16
117 #define PARTS_PER_EXT_DISK 256
119 #define BLKIF_MAJOR(dev) ((dev)>>8)
120 #define BLKIF_MINOR(dev) ((dev) & 0xff)
123 #define EXTENDED (1<<EXT_SHIFT)
124 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
125 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
127 #define DEV_NAME "xvd" /* name in /dev */
129 static int get_id_from_freelist(struct blkfront_info *info)
131 unsigned long free = info->shadow_free;
132 BUG_ON(free >= BLK_RING_SIZE);
133 info->shadow_free = info->shadow[free].req.id;
134 info->shadow[free].req.id = 0x0fffffee; /* debug */
138 static void add_id_to_freelist(struct blkfront_info *info,
141 info->shadow[id].req.id = info->shadow_free;
142 info->shadow[id].request = 0;
143 info->shadow_free = id;
146 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
148 unsigned int end = minor + nr;
151 if (end > nr_minors) {
152 unsigned long *bitmap, *old;
154 bitmap = kzalloc(BITS_TO_LONGS(end) * sizeof(*bitmap),
159 spin_lock(&minor_lock);
160 if (end > nr_minors) {
162 memcpy(bitmap, minors,
163 BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
165 nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
168 spin_unlock(&minor_lock);
172 spin_lock(&minor_lock);
173 if (find_next_bit(minors, end, minor) >= end) {
174 for (; minor < end; ++minor)
175 __set_bit(minor, minors);
179 spin_unlock(&minor_lock);
184 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
186 unsigned int end = minor + nr;
188 BUG_ON(end > nr_minors);
189 spin_lock(&minor_lock);
190 for (; minor < end; ++minor)
191 __clear_bit(minor, minors);
192 spin_unlock(&minor_lock);
195 static void blkif_restart_queue_callback(void *arg)
197 struct blkfront_info *info = (struct blkfront_info *)arg;
198 schedule_work(&info->work);
201 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
203 /* We don't have real geometry info, but let's at least return
204 values consistent with the size of the device */
205 sector_t nsect = get_capacity(bd->bd_disk);
206 sector_t cylinders = nsect;
210 sector_div(cylinders, hg->heads * hg->sectors);
211 hg->cylinders = cylinders;
212 if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
213 hg->cylinders = 0xffff;
217 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
218 unsigned command, unsigned long argument)
220 struct blkfront_info *info = bdev->bd_disk->private_data;
223 dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
224 command, (long)argument);
227 case CDROMMULTISESSION:
228 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
229 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
230 if (put_user(0, (char __user *)(argument + i)))
234 case CDROM_GET_CAPABILITY: {
235 struct gendisk *gd = info->gd;
236 if (gd->flags & GENHD_FL_CD)
242 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
244 return -EINVAL; /* same return as native Linux */
251 * blkif_queue_request
255 * id: for guest use only.
256 * operation: BLKIF_OP_{READ,WRITE,PROBE}
257 * buffer: buffer to read/write into. this should be a
258 * virtual address in the guest os.
260 static int blkif_queue_request(struct request *req)
262 struct blkfront_info *info = req->rq_disk->private_data;
263 unsigned long buffer_mfn;
264 struct blkif_request *ring_req;
266 unsigned int fsect, lsect;
268 grant_ref_t gref_head;
269 struct scatterlist *sg;
271 if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
274 if (gnttab_alloc_grant_references(
275 BLKIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) {
276 gnttab_request_free_callback(
278 blkif_restart_queue_callback,
280 BLKIF_MAX_SEGMENTS_PER_REQUEST);
284 /* Fill out a communications ring structure. */
285 ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
286 id = get_id_from_freelist(info);
287 info->shadow[id].request = (unsigned long)req;
290 ring_req->sector_number = (blkif_sector_t)blk_rq_pos(req);
291 ring_req->handle = info->handle;
293 ring_req->operation = rq_data_dir(req) ?
294 BLKIF_OP_WRITE : BLKIF_OP_READ;
295 if (req->cmd_flags & REQ_HARDBARRIER)
296 ring_req->operation = BLKIF_OP_WRITE_BARRIER;
298 ring_req->nr_segments = blk_rq_map_sg(req->q, req, info->sg);
299 BUG_ON(ring_req->nr_segments > BLKIF_MAX_SEGMENTS_PER_REQUEST);
301 for_each_sg(info->sg, sg, ring_req->nr_segments, i) {
302 buffer_mfn = pfn_to_mfn(page_to_pfn(sg_page(sg)));
303 fsect = sg->offset >> 9;
304 lsect = fsect + (sg->length >> 9) - 1;
305 /* install a grant reference. */
306 ref = gnttab_claim_grant_reference(&gref_head);
307 BUG_ON(ref == -ENOSPC);
309 gnttab_grant_foreign_access_ref(
311 info->xbdev->otherend_id,
315 info->shadow[id].frame[i] = mfn_to_pfn(buffer_mfn);
317 (struct blkif_request_segment) {
320 .last_sect = lsect };
323 info->ring.req_prod_pvt++;
325 /* Keep a private copy so we can reissue requests when recovering. */
326 info->shadow[id].req = *ring_req;
328 gnttab_free_grant_references(gref_head);
334 static inline void flush_requests(struct blkfront_info *info)
338 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
341 notify_remote_via_irq(info->irq);
346 * read a block; request is in a request queue
348 static void do_blkif_request(struct request_queue *rq)
350 struct blkfront_info *info = NULL;
354 pr_debug("Entered do_blkif_request\n");
358 while ((req = blk_peek_request(rq)) != NULL) {
359 info = req->rq_disk->private_data;
361 if (RING_FULL(&info->ring))
364 blk_start_request(req);
366 if (req->cmd_type != REQ_TYPE_FS) {
367 __blk_end_request_all(req, -EIO);
371 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
372 "(%u/%u) buffer:%p [%s]\n",
373 req, req->cmd, (unsigned long)blk_rq_pos(req),
374 blk_rq_cur_sectors(req), blk_rq_sectors(req),
375 req->buffer, rq_data_dir(req) ? "write" : "read");
377 if (blkif_queue_request(req)) {
378 blk_requeue_request(rq, req);
380 /* Avoid pointless unplugs. */
389 flush_requests(info);
392 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
394 struct request_queue *rq;
396 rq = blk_init_queue(do_blkif_request, &blkif_io_lock);
400 queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
402 /* Hard sector size and max sectors impersonate the equiv. hardware. */
403 blk_queue_logical_block_size(rq, sector_size);
404 blk_queue_max_hw_sectors(rq, 512);
406 /* Each segment in a request is up to an aligned page in size. */
407 blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
408 blk_queue_max_segment_size(rq, PAGE_SIZE);
410 /* Ensure a merged request will fit in a single I/O ring slot. */
411 blk_queue_max_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
413 /* Make sure buffer addresses are sector-aligned. */
414 blk_queue_dma_alignment(rq, 511);
416 /* Make sure we don't use bounce buffers. */
417 blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
425 static int xlvbd_barrier(struct blkfront_info *info)
429 err = blk_queue_ordered(info->rq,
430 info->feature_barrier ? QUEUE_ORDERED_DRAIN : QUEUE_ORDERED_NONE);
435 printk(KERN_INFO "blkfront: %s: barriers %s\n",
437 info->feature_barrier ? "enabled" : "disabled");
442 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
443 struct blkfront_info *info,
444 u16 vdisk_info, u16 sector_size)
453 BUG_ON(info->gd != NULL);
454 BUG_ON(info->rq != NULL);
456 if ((info->vdevice>>EXT_SHIFT) > 1) {
457 /* this is above the extended range; something is wrong */
458 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
462 if (!VDEV_IS_EXTENDED(info->vdevice)) {
463 minor = BLKIF_MINOR(info->vdevice);
464 nr_parts = PARTS_PER_DISK;
466 minor = BLKIF_MINOR_EXT(info->vdevice);
467 nr_parts = PARTS_PER_EXT_DISK;
470 if ((minor % nr_parts) == 0)
471 nr_minors = nr_parts;
473 err = xlbd_reserve_minors(minor, nr_minors);
478 gd = alloc_disk(nr_minors);
482 offset = minor / nr_parts;
486 sprintf(gd->disk_name, "%s%c", DEV_NAME, 'a' + offset);
488 sprintf(gd->disk_name, "%s%c%c", DEV_NAME,
489 'a' + ((offset / 26)-1), 'a' + (offset % 26));
492 sprintf(gd->disk_name, "%s%c%d", DEV_NAME,
494 minor & (nr_parts - 1));
496 sprintf(gd->disk_name, "%s%c%c%d", DEV_NAME,
497 'a' + ((offset / 26) - 1),
499 minor & (nr_parts - 1));
502 gd->major = XENVBD_MAJOR;
503 gd->first_minor = minor;
504 gd->fops = &xlvbd_block_fops;
505 gd->private_data = info;
506 gd->driverfs_dev = &(info->xbdev->dev);
507 set_capacity(gd, capacity);
509 if (xlvbd_init_blk_queue(gd, sector_size)) {
514 info->rq = gd->queue;
517 if (info->feature_barrier)
520 if (vdisk_info & VDISK_READONLY)
523 if (vdisk_info & VDISK_REMOVABLE)
524 gd->flags |= GENHD_FL_REMOVABLE;
526 if (vdisk_info & VDISK_CDROM)
527 gd->flags |= GENHD_FL_CD;
532 xlbd_release_minors(minor, nr_minors);
537 static void kick_pending_request_queues(struct blkfront_info *info)
539 if (!RING_FULL(&info->ring)) {
540 /* Re-enable calldowns. */
541 blk_start_queue(info->rq);
542 /* Kick things off immediately. */
543 do_blkif_request(info->rq);
547 static void blkif_restart_queue(struct work_struct *work)
549 struct blkfront_info *info = container_of(work, struct blkfront_info, work);
551 spin_lock_irq(&blkif_io_lock);
552 if (info->connected == BLKIF_STATE_CONNECTED)
553 kick_pending_request_queues(info);
554 spin_unlock_irq(&blkif_io_lock);
557 static void blkif_free(struct blkfront_info *info, int suspend)
559 /* Prevent new requests being issued until we fix things up. */
560 spin_lock_irq(&blkif_io_lock);
561 info->connected = suspend ?
562 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
563 /* No more blkif_request(). */
565 blk_stop_queue(info->rq);
566 /* No more gnttab callback work. */
567 gnttab_cancel_free_callback(&info->callback);
568 spin_unlock_irq(&blkif_io_lock);
570 /* Flush gnttab callback work. Must be done with no locks held. */
571 flush_scheduled_work();
573 /* Free resources associated with old device channel. */
574 if (info->ring_ref != GRANT_INVALID_REF) {
575 gnttab_end_foreign_access(info->ring_ref, 0,
576 (unsigned long)info->ring.sring);
577 info->ring_ref = GRANT_INVALID_REF;
578 info->ring.sring = NULL;
581 unbind_from_irqhandler(info->irq, info);
582 info->evtchn = info->irq = 0;
586 static void blkif_completion(struct blk_shadow *s)
589 for (i = 0; i < s->req.nr_segments; i++)
590 gnttab_end_foreign_access(s->req.seg[i].gref, 0, 0UL);
593 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
596 struct blkif_response *bret;
599 struct blkfront_info *info = (struct blkfront_info *)dev_id;
602 spin_lock_irqsave(&blkif_io_lock, flags);
604 if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
605 spin_unlock_irqrestore(&blkif_io_lock, flags);
610 rp = info->ring.sring->rsp_prod;
611 rmb(); /* Ensure we see queued responses up to 'rp'. */
613 for (i = info->ring.rsp_cons; i != rp; i++) {
616 bret = RING_GET_RESPONSE(&info->ring, i);
618 req = (struct request *)info->shadow[id].request;
620 blkif_completion(&info->shadow[id]);
622 add_id_to_freelist(info, id);
624 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
625 switch (bret->operation) {
626 case BLKIF_OP_WRITE_BARRIER:
627 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
628 printk(KERN_WARNING "blkfront: %s: write barrier op failed\n",
629 info->gd->disk_name);
631 info->feature_barrier = 0;
637 if (unlikely(bret->status != BLKIF_RSP_OKAY))
638 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
639 "request: %x\n", bret->status);
641 __blk_end_request_all(req, error);
648 info->ring.rsp_cons = i;
650 if (i != info->ring.req_prod_pvt) {
652 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
656 info->ring.sring->rsp_event = i + 1;
658 kick_pending_request_queues(info);
660 spin_unlock_irqrestore(&blkif_io_lock, flags);
666 static int setup_blkring(struct xenbus_device *dev,
667 struct blkfront_info *info)
669 struct blkif_sring *sring;
672 info->ring_ref = GRANT_INVALID_REF;
674 sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
676 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
679 SHARED_RING_INIT(sring);
680 FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
682 sg_init_table(info->sg, BLKIF_MAX_SEGMENTS_PER_REQUEST);
684 err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
686 free_page((unsigned long)sring);
687 info->ring.sring = NULL;
690 info->ring_ref = err;
692 err = xenbus_alloc_evtchn(dev, &info->evtchn);
696 err = bind_evtchn_to_irqhandler(info->evtchn,
698 IRQF_SAMPLE_RANDOM, "blkif", info);
700 xenbus_dev_fatal(dev, err,
701 "bind_evtchn_to_irqhandler failed");
713 /* Common code used when first setting up, and when resuming. */
714 static int talk_to_blkback(struct xenbus_device *dev,
715 struct blkfront_info *info)
717 const char *message = NULL;
718 struct xenbus_transaction xbt;
721 /* Create shared ring, alloc event channel. */
722 err = setup_blkring(dev, info);
727 err = xenbus_transaction_start(&xbt);
729 xenbus_dev_fatal(dev, err, "starting transaction");
730 goto destroy_blkring;
733 err = xenbus_printf(xbt, dev->nodename,
734 "ring-ref", "%u", info->ring_ref);
736 message = "writing ring-ref";
737 goto abort_transaction;
739 err = xenbus_printf(xbt, dev->nodename,
740 "event-channel", "%u", info->evtchn);
742 message = "writing event-channel";
743 goto abort_transaction;
745 err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
746 XEN_IO_PROTO_ABI_NATIVE);
748 message = "writing protocol";
749 goto abort_transaction;
752 err = xenbus_transaction_end(xbt, 0);
756 xenbus_dev_fatal(dev, err, "completing transaction");
757 goto destroy_blkring;
760 xenbus_switch_state(dev, XenbusStateInitialised);
765 xenbus_transaction_end(xbt, 1);
767 xenbus_dev_fatal(dev, err, "%s", message);
776 * Entry point to this code when a new device is created. Allocate the basic
777 * structures and the ring buffer for communication with the backend, and
778 * inform the backend of the appropriate details for those. Switch to
781 static int blkfront_probe(struct xenbus_device *dev,
782 const struct xenbus_device_id *id)
785 struct blkfront_info *info;
787 /* FIXME: Use dynamic device id if this is not set. */
788 err = xenbus_scanf(XBT_NIL, dev->nodename,
789 "virtual-device", "%i", &vdevice);
791 /* go looking in the extended area instead */
792 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
795 xenbus_dev_fatal(dev, err, "reading virtual-device");
800 info = kzalloc(sizeof(*info), GFP_KERNEL);
802 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
807 info->vdevice = vdevice;
808 info->connected = BLKIF_STATE_DISCONNECTED;
809 INIT_WORK(&info->work, blkif_restart_queue);
811 for (i = 0; i < BLK_RING_SIZE; i++)
812 info->shadow[i].req.id = i+1;
813 info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
815 /* Front end dir is a number, which is used as the id. */
816 info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
817 dev_set_drvdata(&dev->dev, info);
819 err = talk_to_blkback(dev, info);
822 dev_set_drvdata(&dev->dev, NULL);
830 static int blkif_recover(struct blkfront_info *info)
833 struct blkif_request *req;
834 struct blk_shadow *copy;
837 /* Stage 1: Make a safe copy of the shadow state. */
838 copy = kmalloc(sizeof(info->shadow),
839 GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
842 memcpy(copy, info->shadow, sizeof(info->shadow));
844 /* Stage 2: Set up free list. */
845 memset(&info->shadow, 0, sizeof(info->shadow));
846 for (i = 0; i < BLK_RING_SIZE; i++)
847 info->shadow[i].req.id = i+1;
848 info->shadow_free = info->ring.req_prod_pvt;
849 info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
851 /* Stage 3: Find pending requests and requeue them. */
852 for (i = 0; i < BLK_RING_SIZE; i++) {
854 if (copy[i].request == 0)
857 /* Grab a request slot and copy shadow state into it. */
858 req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
861 /* We get a new request id, and must reset the shadow state. */
862 req->id = get_id_from_freelist(info);
863 memcpy(&info->shadow[req->id], ©[i], sizeof(copy[i]));
865 /* Rewrite any grant references invalidated by susp/resume. */
866 for (j = 0; j < req->nr_segments; j++)
867 gnttab_grant_foreign_access_ref(
869 info->xbdev->otherend_id,
870 pfn_to_mfn(info->shadow[req->id].frame[j]),
873 info->shadow[req->id].request));
874 info->shadow[req->id].req = *req;
876 info->ring.req_prod_pvt++;
881 xenbus_switch_state(info->xbdev, XenbusStateConnected);
883 spin_lock_irq(&blkif_io_lock);
885 /* Now safe for us to use the shared ring */
886 info->connected = BLKIF_STATE_CONNECTED;
888 /* Send off requeued requests */
889 flush_requests(info);
891 /* Kick any other new requests queued since we resumed */
892 kick_pending_request_queues(info);
894 spin_unlock_irq(&blkif_io_lock);
900 * We are reconnecting to the backend, due to a suspend/resume, or a backend
901 * driver restart. We tear down our blkif structure and recreate it, but
902 * leave the device-layer structures intact so that this is transparent to the
903 * rest of the kernel.
905 static int blkfront_resume(struct xenbus_device *dev)
907 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
910 dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
912 blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
914 err = talk_to_blkback(dev, info);
915 if (info->connected == BLKIF_STATE_SUSPENDED && !err)
916 err = blkif_recover(info);
923 * Invoked when the backend is finally 'ready' (and has told produced
924 * the details about the physical device - #sectors, size, etc).
926 static void blkfront_connect(struct blkfront_info *info)
928 unsigned long long sectors;
929 unsigned long sector_size;
933 if ((info->connected == BLKIF_STATE_CONNECTED) ||
934 (info->connected == BLKIF_STATE_SUSPENDED) )
937 dev_dbg(&info->xbdev->dev, "%s:%s.\n",
938 __func__, info->xbdev->otherend);
940 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
941 "sectors", "%llu", §ors,
942 "info", "%u", &binfo,
943 "sector-size", "%lu", §or_size,
946 xenbus_dev_fatal(info->xbdev, err,
947 "reading backend fields at %s",
948 info->xbdev->otherend);
952 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
953 "feature-barrier", "%lu", &info->feature_barrier,
956 info->feature_barrier = 0;
958 err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size);
960 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
961 info->xbdev->otherend);
965 xenbus_switch_state(info->xbdev, XenbusStateConnected);
967 /* Kick pending requests. */
968 spin_lock_irq(&blkif_io_lock);
969 info->connected = BLKIF_STATE_CONNECTED;
970 kick_pending_request_queues(info);
971 spin_unlock_irq(&blkif_io_lock);
979 * Handle the change of state of the backend to Closing. We must delete our
980 * device-layer structures now, to ensure that writes are flushed through to
981 * the backend. Once is this done, we can switch to Closed in
984 static void blkfront_closing(struct xenbus_device *dev)
986 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
987 unsigned int minor, nr_minors;
990 dev_dbg(&dev->dev, "blkfront_closing: %s removed\n", dev->nodename);
992 if (info->rq == NULL)
995 spin_lock_irqsave(&blkif_io_lock, flags);
997 /* No more blkif_request(). */
998 blk_stop_queue(info->rq);
1000 /* No more gnttab callback work. */
1001 gnttab_cancel_free_callback(&info->callback);
1002 spin_unlock_irqrestore(&blkif_io_lock, flags);
1004 /* Flush gnttab callback work. Must be done with no locks held. */
1005 flush_scheduled_work();
1007 blk_cleanup_queue(info->rq);
1010 minor = info->gd->first_minor;
1011 nr_minors = info->gd->minors;
1012 del_gendisk(info->gd);
1013 xlbd_release_minors(minor, nr_minors);
1016 xenbus_frontend_closed(dev);
1020 * Callback received when the backend's state changes.
1022 static void blkback_changed(struct xenbus_device *dev,
1023 enum xenbus_state backend_state)
1025 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1026 struct block_device *bd;
1028 dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1030 switch (backend_state) {
1031 case XenbusStateInitialising:
1032 case XenbusStateInitWait:
1033 case XenbusStateInitialised:
1034 case XenbusStateUnknown:
1035 case XenbusStateClosed:
1038 case XenbusStateConnected:
1039 blkfront_connect(info);
1042 case XenbusStateClosing:
1043 if (info->gd == NULL) {
1044 xenbus_frontend_closed(dev);
1047 bd = bdget_disk(info->gd, 0);
1049 xenbus_dev_fatal(dev, -ENODEV, "bdget failed");
1051 mutex_lock(&bd->bd_mutex);
1052 if (info->users > 0)
1053 xenbus_dev_error(dev, -EBUSY,
1054 "Device in use; refusing to close");
1056 blkfront_closing(dev);
1057 mutex_unlock(&bd->bd_mutex);
1063 static int blkfront_remove(struct xenbus_device *dev)
1065 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1067 dev_dbg(&dev->dev, "blkfront_remove: %s removed\n", dev->nodename);
1069 blkif_free(info, 0);
1071 if(info->users == 0)
1074 info->is_ready = -1;
1079 static int blkfront_is_ready(struct xenbus_device *dev)
1081 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1083 return info->is_ready > 0;
1086 static int blkif_open(struct block_device *bdev, fmode_t mode)
1088 struct blkfront_info *info = bdev->bd_disk->private_data;
1092 if (info->is_ready < 0)
1101 static int blkif_release(struct gendisk *disk, fmode_t mode)
1103 struct blkfront_info *info = disk->private_data;
1106 if (info->users == 0) {
1107 /* Check whether we have been instructed to close. We will
1108 have ignored this request initially, as the device was
1110 struct xenbus_device *dev = info->xbdev;
1111 enum xenbus_state state = xenbus_read_driver_state(dev->otherend);
1113 if(info->is_ready < 0) {
1114 blkfront_closing(dev);
1116 } else if (state == XenbusStateClosing && info->is_ready)
1117 blkfront_closing(dev);
1123 static const struct block_device_operations xlvbd_block_fops =
1125 .owner = THIS_MODULE,
1127 .release = blkif_release,
1128 .getgeo = blkif_getgeo,
1129 .ioctl = blkif_ioctl,
1133 static const struct xenbus_device_id blkfront_ids[] = {
1138 static struct xenbus_driver blkfront = {
1140 .owner = THIS_MODULE,
1141 .ids = blkfront_ids,
1142 .probe = blkfront_probe,
1143 .remove = blkfront_remove,
1144 .resume = blkfront_resume,
1145 .otherend_changed = blkback_changed,
1146 .is_ready = blkfront_is_ready,
1149 static int __init xlblk_init(void)
1154 if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
1155 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
1156 XENVBD_MAJOR, DEV_NAME);
1160 return xenbus_register_frontend(&blkfront);
1162 module_init(xlblk_init);
1165 static void __exit xlblk_exit(void)
1167 return xenbus_unregister_driver(&blkfront);
1169 module_exit(xlblk_exit);
1171 MODULE_DESCRIPTION("Xen virtual block device frontend");
1172 MODULE_LICENSE("GPL");
1173 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
1174 MODULE_ALIAS("xen:vbd");
1175 MODULE_ALIAS("xenblk");