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>
49 #include <xen/xenbus.h>
50 #include <xen/grant_table.h>
51 #include <xen/events.h>
53 #include <xen/platform_pci.h>
55 #include <xen/interface/grant_table.h>
56 #include <xen/interface/io/blkif.h>
57 #include <xen/interface/io/protocols.h>
59 #include <asm/xen/hypervisor.h>
62 BLKIF_STATE_DISCONNECTED,
63 BLKIF_STATE_CONNECTED,
64 BLKIF_STATE_SUSPENDED,
68 struct blkif_request req;
69 struct request *request;
70 unsigned long frame[BLKIF_MAX_SEGMENTS_PER_REQUEST];
73 static DEFINE_MUTEX(blkfront_mutex);
74 static const struct block_device_operations xlvbd_block_fops;
76 #define BLK_RING_SIZE __CONST_RING_SIZE(blkif, PAGE_SIZE)
79 * We have one of these per vbd, whether ide, scsi or 'other'. They
80 * hang in private_data off the gendisk structure. We may end up
81 * putting all kinds of interesting stuff here :-)
87 struct xenbus_device *xbdev;
91 enum blkif_state connected;
93 struct blkif_front_ring ring;
94 struct scatterlist sg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
95 unsigned int evtchn, irq;
96 struct request_queue *rq;
97 struct work_struct work;
98 struct gnttab_free_callback callback;
99 struct blk_shadow shadow[BLK_RING_SIZE];
100 unsigned long shadow_free;
101 unsigned int feature_flush;
102 unsigned int flush_op;
103 unsigned int feature_discard:1;
104 unsigned int feature_secdiscard:1;
105 unsigned int discard_granularity;
106 unsigned int discard_alignment;
110 static unsigned int nr_minors;
111 static unsigned long *minors;
112 static DEFINE_SPINLOCK(minor_lock);
114 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
115 (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
116 #define GRANT_INVALID_REF 0
118 #define PARTS_PER_DISK 16
119 #define PARTS_PER_EXT_DISK 256
121 #define BLKIF_MAJOR(dev) ((dev)>>8)
122 #define BLKIF_MINOR(dev) ((dev) & 0xff)
125 #define EXTENDED (1<<EXT_SHIFT)
126 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
127 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
128 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
129 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
130 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
131 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
133 #define DEV_NAME "xvd" /* name in /dev */
135 static int get_id_from_freelist(struct blkfront_info *info)
137 unsigned long free = info->shadow_free;
138 BUG_ON(free >= BLK_RING_SIZE);
139 info->shadow_free = info->shadow[free].req.u.rw.id;
140 info->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
144 static void add_id_to_freelist(struct blkfront_info *info,
147 info->shadow[id].req.u.rw.id = info->shadow_free;
148 info->shadow[id].request = NULL;
149 info->shadow_free = id;
152 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
154 unsigned int end = minor + nr;
157 if (end > nr_minors) {
158 unsigned long *bitmap, *old;
160 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
165 spin_lock(&minor_lock);
166 if (end > nr_minors) {
168 memcpy(bitmap, minors,
169 BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
171 nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
174 spin_unlock(&minor_lock);
178 spin_lock(&minor_lock);
179 if (find_next_bit(minors, end, minor) >= end) {
180 bitmap_set(minors, minor, nr);
184 spin_unlock(&minor_lock);
189 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
191 unsigned int end = minor + nr;
193 BUG_ON(end > nr_minors);
194 spin_lock(&minor_lock);
195 bitmap_clear(minors, minor, nr);
196 spin_unlock(&minor_lock);
199 static void blkif_restart_queue_callback(void *arg)
201 struct blkfront_info *info = (struct blkfront_info *)arg;
202 schedule_work(&info->work);
205 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
207 /* We don't have real geometry info, but let's at least return
208 values consistent with the size of the device */
209 sector_t nsect = get_capacity(bd->bd_disk);
210 sector_t cylinders = nsect;
214 sector_div(cylinders, hg->heads * hg->sectors);
215 hg->cylinders = cylinders;
216 if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
217 hg->cylinders = 0xffff;
221 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
222 unsigned command, unsigned long argument)
224 struct blkfront_info *info = bdev->bd_disk->private_data;
227 dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
228 command, (long)argument);
231 case CDROMMULTISESSION:
232 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
233 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
234 if (put_user(0, (char __user *)(argument + i)))
238 case CDROM_GET_CAPABILITY: {
239 struct gendisk *gd = info->gd;
240 if (gd->flags & GENHD_FL_CD)
246 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
248 return -EINVAL; /* same return as native Linux */
255 * Generate a Xen blkfront IO request from a blk layer request. Reads
256 * and writes are handled as expected.
258 * @req: a request struct
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 = req;
289 ring_req->u.rw.id = id;
290 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
291 ring_req->u.rw.handle = info->handle;
293 ring_req->operation = rq_data_dir(req) ?
294 BLKIF_OP_WRITE : BLKIF_OP_READ;
296 if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) {
298 * Ideally we can do an unordered flush-to-disk. In case the
299 * backend onlysupports barriers, use that. A barrier request
300 * a superset of FUA, so we can implement it the same
301 * way. (It's also a FLUSH+FUA, since it is
302 * guaranteed ordered WRT previous writes.)
304 ring_req->operation = info->flush_op;
307 if (unlikely(req->cmd_flags & (REQ_DISCARD | REQ_SECURE))) {
308 /* id, sector_number and handle are set above. */
309 ring_req->operation = BLKIF_OP_DISCARD;
310 ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
311 if ((req->cmd_flags & REQ_SECURE) && info->feature_secdiscard)
312 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
314 ring_req->u.discard.flag = 0;
316 ring_req->u.rw.nr_segments = blk_rq_map_sg(req->q, req,
318 BUG_ON(ring_req->u.rw.nr_segments >
319 BLKIF_MAX_SEGMENTS_PER_REQUEST);
321 for_each_sg(info->sg, sg, ring_req->u.rw.nr_segments, i) {
322 buffer_mfn = pfn_to_mfn(page_to_pfn(sg_page(sg)));
323 fsect = sg->offset >> 9;
324 lsect = fsect + (sg->length >> 9) - 1;
325 /* install a grant reference. */
326 ref = gnttab_claim_grant_reference(&gref_head);
327 BUG_ON(ref == -ENOSPC);
329 gnttab_grant_foreign_access_ref(
331 info->xbdev->otherend_id,
335 info->shadow[id].frame[i] = mfn_to_pfn(buffer_mfn);
336 ring_req->u.rw.seg[i] =
337 (struct blkif_request_segment) {
340 .last_sect = lsect };
344 info->ring.req_prod_pvt++;
346 /* Keep a private copy so we can reissue requests when recovering. */
347 info->shadow[id].req = *ring_req;
349 gnttab_free_grant_references(gref_head);
355 static inline void flush_requests(struct blkfront_info *info)
359 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
362 notify_remote_via_irq(info->irq);
367 * read a block; request is in a request queue
369 static void do_blkif_request(struct request_queue *rq)
371 struct blkfront_info *info = NULL;
375 pr_debug("Entered do_blkif_request\n");
379 while ((req = blk_peek_request(rq)) != NULL) {
380 info = req->rq_disk->private_data;
382 if (RING_FULL(&info->ring))
385 blk_start_request(req);
387 if ((req->cmd_type != REQ_TYPE_FS) ||
388 ((req->cmd_flags & (REQ_FLUSH | REQ_FUA)) &&
390 __blk_end_request_all(req, -EIO);
394 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
395 "(%u/%u) buffer:%p [%s]\n",
396 req, req->cmd, (unsigned long)blk_rq_pos(req),
397 blk_rq_cur_sectors(req), blk_rq_sectors(req),
398 req->buffer, rq_data_dir(req) ? "write" : "read");
400 if (blkif_queue_request(req)) {
401 blk_requeue_request(rq, req);
403 /* Avoid pointless unplugs. */
412 flush_requests(info);
415 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
417 struct request_queue *rq;
418 struct blkfront_info *info = gd->private_data;
420 rq = blk_init_queue(do_blkif_request, &info->io_lock);
424 queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
426 if (info->feature_discard) {
427 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
428 blk_queue_max_discard_sectors(rq, get_capacity(gd));
429 rq->limits.discard_granularity = info->discard_granularity;
430 rq->limits.discard_alignment = info->discard_alignment;
431 if (info->feature_secdiscard)
432 queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD, rq);
435 /* Hard sector size and max sectors impersonate the equiv. hardware. */
436 blk_queue_logical_block_size(rq, sector_size);
437 blk_queue_max_hw_sectors(rq, 512);
439 /* Each segment in a request is up to an aligned page in size. */
440 blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
441 blk_queue_max_segment_size(rq, PAGE_SIZE);
443 /* Ensure a merged request will fit in a single I/O ring slot. */
444 blk_queue_max_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
446 /* Make sure buffer addresses are sector-aligned. */
447 blk_queue_dma_alignment(rq, 511);
449 /* Make sure we don't use bounce buffers. */
450 blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
458 static void xlvbd_flush(struct blkfront_info *info)
460 blk_queue_flush(info->rq, info->feature_flush);
461 printk(KERN_INFO "blkfront: %s: %s: %s\n",
463 info->flush_op == BLKIF_OP_WRITE_BARRIER ?
464 "barrier" : (info->flush_op == BLKIF_OP_FLUSH_DISKCACHE ?
465 "flush diskcache" : "barrier or flush"),
466 info->feature_flush ? "enabled" : "disabled");
469 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
472 major = BLKIF_MAJOR(vdevice);
473 *minor = BLKIF_MINOR(vdevice);
476 *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
477 *minor = ((*minor / 64) * PARTS_PER_DISK) +
478 EMULATED_HD_DISK_MINOR_OFFSET;
481 *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
482 *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
483 EMULATED_HD_DISK_MINOR_OFFSET;
485 case XEN_SCSI_DISK0_MAJOR:
486 *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
487 *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
489 case XEN_SCSI_DISK1_MAJOR:
490 case XEN_SCSI_DISK2_MAJOR:
491 case XEN_SCSI_DISK3_MAJOR:
492 case XEN_SCSI_DISK4_MAJOR:
493 case XEN_SCSI_DISK5_MAJOR:
494 case XEN_SCSI_DISK6_MAJOR:
495 case XEN_SCSI_DISK7_MAJOR:
496 *offset = (*minor / PARTS_PER_DISK) +
497 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
498 EMULATED_SD_DISK_NAME_OFFSET;
500 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
501 EMULATED_SD_DISK_MINOR_OFFSET;
503 case XEN_SCSI_DISK8_MAJOR:
504 case XEN_SCSI_DISK9_MAJOR:
505 case XEN_SCSI_DISK10_MAJOR:
506 case XEN_SCSI_DISK11_MAJOR:
507 case XEN_SCSI_DISK12_MAJOR:
508 case XEN_SCSI_DISK13_MAJOR:
509 case XEN_SCSI_DISK14_MAJOR:
510 case XEN_SCSI_DISK15_MAJOR:
511 *offset = (*minor / PARTS_PER_DISK) +
512 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
513 EMULATED_SD_DISK_NAME_OFFSET;
515 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
516 EMULATED_SD_DISK_MINOR_OFFSET;
519 *offset = *minor / PARTS_PER_DISK;
522 printk(KERN_WARNING "blkfront: your disk configuration is "
523 "incorrect, please use an xvd device instead\n");
529 static char *encode_disk_name(char *ptr, unsigned int n)
532 ptr = encode_disk_name(ptr, n / 26 - 1);
537 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
538 struct blkfront_info *info,
539 u16 vdisk_info, u16 sector_size)
549 BUG_ON(info->gd != NULL);
550 BUG_ON(info->rq != NULL);
552 if ((info->vdevice>>EXT_SHIFT) > 1) {
553 /* this is above the extended range; something is wrong */
554 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
558 if (!VDEV_IS_EXTENDED(info->vdevice)) {
559 err = xen_translate_vdev(info->vdevice, &minor, &offset);
562 nr_parts = PARTS_PER_DISK;
564 minor = BLKIF_MINOR_EXT(info->vdevice);
565 nr_parts = PARTS_PER_EXT_DISK;
566 offset = minor / nr_parts;
567 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
568 printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
569 "emulated IDE disks,\n\t choose an xvd device name"
570 "from xvde on\n", info->vdevice);
572 if (minor >> MINORBITS) {
573 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
574 info->vdevice, minor);
578 if ((minor % nr_parts) == 0)
579 nr_minors = nr_parts;
581 err = xlbd_reserve_minors(minor, nr_minors);
586 gd = alloc_disk(nr_minors);
590 strcpy(gd->disk_name, DEV_NAME);
591 ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
592 BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
596 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
597 "%d", minor & (nr_parts - 1));
599 gd->major = XENVBD_MAJOR;
600 gd->first_minor = minor;
601 gd->fops = &xlvbd_block_fops;
602 gd->private_data = info;
603 gd->driverfs_dev = &(info->xbdev->dev);
604 set_capacity(gd, capacity);
606 if (xlvbd_init_blk_queue(gd, sector_size)) {
611 info->rq = gd->queue;
616 if (vdisk_info & VDISK_READONLY)
619 if (vdisk_info & VDISK_REMOVABLE)
620 gd->flags |= GENHD_FL_REMOVABLE;
622 if (vdisk_info & VDISK_CDROM)
623 gd->flags |= GENHD_FL_CD;
628 xlbd_release_minors(minor, nr_minors);
633 static void xlvbd_release_gendisk(struct blkfront_info *info)
635 unsigned int minor, nr_minors;
638 if (info->rq == NULL)
641 spin_lock_irqsave(&info->io_lock, flags);
643 /* No more blkif_request(). */
644 blk_stop_queue(info->rq);
646 /* No more gnttab callback work. */
647 gnttab_cancel_free_callback(&info->callback);
648 spin_unlock_irqrestore(&info->io_lock, flags);
650 /* Flush gnttab callback work. Must be done with no locks held. */
651 flush_work_sync(&info->work);
653 del_gendisk(info->gd);
655 minor = info->gd->first_minor;
656 nr_minors = info->gd->minors;
657 xlbd_release_minors(minor, nr_minors);
659 blk_cleanup_queue(info->rq);
666 static void kick_pending_request_queues(struct blkfront_info *info)
668 if (!RING_FULL(&info->ring)) {
669 /* Re-enable calldowns. */
670 blk_start_queue(info->rq);
671 /* Kick things off immediately. */
672 do_blkif_request(info->rq);
676 static void blkif_restart_queue(struct work_struct *work)
678 struct blkfront_info *info = container_of(work, struct blkfront_info, work);
680 spin_lock_irq(&info->io_lock);
681 if (info->connected == BLKIF_STATE_CONNECTED)
682 kick_pending_request_queues(info);
683 spin_unlock_irq(&info->io_lock);
686 static void blkif_free(struct blkfront_info *info, int suspend)
688 /* Prevent new requests being issued until we fix things up. */
689 spin_lock_irq(&info->io_lock);
690 info->connected = suspend ?
691 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
692 /* No more blkif_request(). */
694 blk_stop_queue(info->rq);
695 /* No more gnttab callback work. */
696 gnttab_cancel_free_callback(&info->callback);
697 spin_unlock_irq(&info->io_lock);
699 /* Flush gnttab callback work. Must be done with no locks held. */
700 flush_work_sync(&info->work);
702 /* Free resources associated with old device channel. */
703 if (info->ring_ref != GRANT_INVALID_REF) {
704 gnttab_end_foreign_access(info->ring_ref, 0,
705 (unsigned long)info->ring.sring);
706 info->ring_ref = GRANT_INVALID_REF;
707 info->ring.sring = NULL;
710 unbind_from_irqhandler(info->irq, info);
711 info->evtchn = info->irq = 0;
715 static void blkif_completion(struct blk_shadow *s)
718 /* Do not let BLKIF_OP_DISCARD as nr_segment is in the same place
720 for (i = 0; i < s->req.u.rw.nr_segments; i++)
721 gnttab_end_foreign_access(s->req.u.rw.seg[i].gref, 0, 0UL);
724 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
727 struct blkif_response *bret;
730 struct blkfront_info *info = (struct blkfront_info *)dev_id;
733 spin_lock_irqsave(&info->io_lock, flags);
735 if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
736 spin_unlock_irqrestore(&info->io_lock, flags);
741 rp = info->ring.sring->rsp_prod;
742 rmb(); /* Ensure we see queued responses up to 'rp'. */
744 for (i = info->ring.rsp_cons; i != rp; i++) {
747 bret = RING_GET_RESPONSE(&info->ring, i);
749 req = info->shadow[id].request;
751 if (bret->operation != BLKIF_OP_DISCARD)
752 blkif_completion(&info->shadow[id]);
754 add_id_to_freelist(info, id);
756 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
757 switch (bret->operation) {
758 case BLKIF_OP_DISCARD:
759 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
760 struct request_queue *rq = info->rq;
761 printk(KERN_WARNING "blkfront: %s: discard op failed\n",
762 info->gd->disk_name);
764 info->feature_discard = 0;
765 info->feature_secdiscard = 0;
766 queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
767 queue_flag_clear(QUEUE_FLAG_SECDISCARD, rq);
769 __blk_end_request_all(req, error);
771 case BLKIF_OP_FLUSH_DISKCACHE:
772 case BLKIF_OP_WRITE_BARRIER:
773 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
774 printk(KERN_WARNING "blkfront: %s: write %s op failed\n",
775 info->flush_op == BLKIF_OP_WRITE_BARRIER ?
776 "barrier" : "flush disk cache",
777 info->gd->disk_name);
780 if (unlikely(bret->status == BLKIF_RSP_ERROR &&
781 info->shadow[id].req.u.rw.nr_segments == 0)) {
782 printk(KERN_WARNING "blkfront: %s: empty write %s op failed\n",
783 info->flush_op == BLKIF_OP_WRITE_BARRIER ?
784 "barrier" : "flush disk cache",
785 info->gd->disk_name);
788 if (unlikely(error)) {
789 if (error == -EOPNOTSUPP)
791 info->feature_flush = 0;
798 if (unlikely(bret->status != BLKIF_RSP_OKAY))
799 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
800 "request: %x\n", bret->status);
802 __blk_end_request_all(req, error);
809 info->ring.rsp_cons = i;
811 if (i != info->ring.req_prod_pvt) {
813 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
817 info->ring.sring->rsp_event = i + 1;
819 kick_pending_request_queues(info);
821 spin_unlock_irqrestore(&info->io_lock, flags);
827 static int setup_blkring(struct xenbus_device *dev,
828 struct blkfront_info *info)
830 struct blkif_sring *sring;
833 info->ring_ref = GRANT_INVALID_REF;
835 sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
837 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
840 SHARED_RING_INIT(sring);
841 FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
843 sg_init_table(info->sg, BLKIF_MAX_SEGMENTS_PER_REQUEST);
845 err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
847 free_page((unsigned long)sring);
848 info->ring.sring = NULL;
851 info->ring_ref = err;
853 err = xenbus_alloc_evtchn(dev, &info->evtchn);
857 err = bind_evtchn_to_irqhandler(info->evtchn,
859 IRQF_SAMPLE_RANDOM, "blkif", info);
861 xenbus_dev_fatal(dev, err,
862 "bind_evtchn_to_irqhandler failed");
874 /* Common code used when first setting up, and when resuming. */
875 static int talk_to_blkback(struct xenbus_device *dev,
876 struct blkfront_info *info)
878 const char *message = NULL;
879 struct xenbus_transaction xbt;
882 /* Create shared ring, alloc event channel. */
883 err = setup_blkring(dev, info);
888 err = xenbus_transaction_start(&xbt);
890 xenbus_dev_fatal(dev, err, "starting transaction");
891 goto destroy_blkring;
894 err = xenbus_printf(xbt, dev->nodename,
895 "ring-ref", "%u", info->ring_ref);
897 message = "writing ring-ref";
898 goto abort_transaction;
900 err = xenbus_printf(xbt, dev->nodename,
901 "event-channel", "%u", info->evtchn);
903 message = "writing event-channel";
904 goto abort_transaction;
906 err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
907 XEN_IO_PROTO_ABI_NATIVE);
909 message = "writing protocol";
910 goto abort_transaction;
913 err = xenbus_transaction_end(xbt, 0);
917 xenbus_dev_fatal(dev, err, "completing transaction");
918 goto destroy_blkring;
921 xenbus_switch_state(dev, XenbusStateInitialised);
926 xenbus_transaction_end(xbt, 1);
928 xenbus_dev_fatal(dev, err, "%s", message);
936 * Entry point to this code when a new device is created. Allocate the basic
937 * structures and the ring buffer for communication with the backend, and
938 * inform the backend of the appropriate details for those. Switch to
941 static int blkfront_probe(struct xenbus_device *dev,
942 const struct xenbus_device_id *id)
945 struct blkfront_info *info;
947 /* FIXME: Use dynamic device id if this is not set. */
948 err = xenbus_scanf(XBT_NIL, dev->nodename,
949 "virtual-device", "%i", &vdevice);
951 /* go looking in the extended area instead */
952 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
955 xenbus_dev_fatal(dev, err, "reading virtual-device");
960 if (xen_hvm_domain()) {
963 /* no unplug has been done: do not hook devices != xen vbds */
964 if (xen_platform_pci_unplug & XEN_UNPLUG_UNNECESSARY) {
967 if (!VDEV_IS_EXTENDED(vdevice))
968 major = BLKIF_MAJOR(vdevice);
970 major = XENVBD_MAJOR;
972 if (major != XENVBD_MAJOR) {
974 "%s: HVM does not support vbd %d as xen block device\n",
975 __FUNCTION__, vdevice);
979 /* do not create a PV cdrom device if we are an HVM guest */
980 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
983 if (strncmp(type, "cdrom", 5) == 0) {
989 info = kzalloc(sizeof(*info), GFP_KERNEL);
991 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
995 mutex_init(&info->mutex);
996 spin_lock_init(&info->io_lock);
998 info->vdevice = vdevice;
999 info->connected = BLKIF_STATE_DISCONNECTED;
1000 INIT_WORK(&info->work, blkif_restart_queue);
1002 for (i = 0; i < BLK_RING_SIZE; i++)
1003 info->shadow[i].req.u.rw.id = i+1;
1004 info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1006 /* Front end dir is a number, which is used as the id. */
1007 info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1008 dev_set_drvdata(&dev->dev, info);
1010 err = talk_to_blkback(dev, info);
1013 dev_set_drvdata(&dev->dev, NULL);
1021 static int blkif_recover(struct blkfront_info *info)
1024 struct blkif_request *req;
1025 struct blk_shadow *copy;
1028 /* Stage 1: Make a safe copy of the shadow state. */
1029 copy = kmalloc(sizeof(info->shadow),
1030 GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
1033 memcpy(copy, info->shadow, sizeof(info->shadow));
1035 /* Stage 2: Set up free list. */
1036 memset(&info->shadow, 0, sizeof(info->shadow));
1037 for (i = 0; i < BLK_RING_SIZE; i++)
1038 info->shadow[i].req.u.rw.id = i+1;
1039 info->shadow_free = info->ring.req_prod_pvt;
1040 info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1042 /* Stage 3: Find pending requests and requeue them. */
1043 for (i = 0; i < BLK_RING_SIZE; i++) {
1045 if (!copy[i].request)
1048 /* Grab a request slot and copy shadow state into it. */
1049 req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
1052 /* We get a new request id, and must reset the shadow state. */
1053 req->u.rw.id = get_id_from_freelist(info);
1054 memcpy(&info->shadow[req->u.rw.id], ©[i], sizeof(copy[i]));
1056 if (req->operation != BLKIF_OP_DISCARD) {
1057 /* Rewrite any grant references invalidated by susp/resume. */
1058 for (j = 0; j < req->u.rw.nr_segments; j++)
1059 gnttab_grant_foreign_access_ref(
1060 req->u.rw.seg[j].gref,
1061 info->xbdev->otherend_id,
1062 pfn_to_mfn(info->shadow[req->u.rw.id].frame[j]),
1063 rq_data_dir(info->shadow[req->u.rw.id].request));
1065 info->shadow[req->u.rw.id].req = *req;
1067 info->ring.req_prod_pvt++;
1072 xenbus_switch_state(info->xbdev, XenbusStateConnected);
1074 spin_lock_irq(&info->io_lock);
1076 /* Now safe for us to use the shared ring */
1077 info->connected = BLKIF_STATE_CONNECTED;
1079 /* Send off requeued requests */
1080 flush_requests(info);
1082 /* Kick any other new requests queued since we resumed */
1083 kick_pending_request_queues(info);
1085 spin_unlock_irq(&info->io_lock);
1091 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1092 * driver restart. We tear down our blkif structure and recreate it, but
1093 * leave the device-layer structures intact so that this is transparent to the
1094 * rest of the kernel.
1096 static int blkfront_resume(struct xenbus_device *dev)
1098 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1101 dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
1103 blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
1105 err = talk_to_blkback(dev, info);
1106 if (info->connected == BLKIF_STATE_SUSPENDED && !err)
1107 err = blkif_recover(info);
1113 blkfront_closing(struct blkfront_info *info)
1115 struct xenbus_device *xbdev = info->xbdev;
1116 struct block_device *bdev = NULL;
1118 mutex_lock(&info->mutex);
1120 if (xbdev->state == XenbusStateClosing) {
1121 mutex_unlock(&info->mutex);
1126 bdev = bdget_disk(info->gd, 0);
1128 mutex_unlock(&info->mutex);
1131 xenbus_frontend_closed(xbdev);
1135 mutex_lock(&bdev->bd_mutex);
1137 if (bdev->bd_openers) {
1138 xenbus_dev_error(xbdev, -EBUSY,
1139 "Device in use; refusing to close");
1140 xenbus_switch_state(xbdev, XenbusStateClosing);
1142 xlvbd_release_gendisk(info);
1143 xenbus_frontend_closed(xbdev);
1146 mutex_unlock(&bdev->bd_mutex);
1150 static void blkfront_setup_discard(struct blkfront_info *info)
1154 unsigned int discard_granularity;
1155 unsigned int discard_alignment;
1156 unsigned int discard_secure;
1158 type = xenbus_read(XBT_NIL, info->xbdev->otherend, "type", NULL);
1162 info->feature_secdiscard = 0;
1163 if (strncmp(type, "phy", 3) == 0) {
1164 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1165 "discard-granularity", "%u", &discard_granularity,
1166 "discard-alignment", "%u", &discard_alignment,
1169 info->feature_discard = 1;
1170 info->discard_granularity = discard_granularity;
1171 info->discard_alignment = discard_alignment;
1173 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1174 "discard-secure", "%d", &discard_secure,
1177 info->feature_secdiscard = discard_secure;
1179 } else if (strncmp(type, "file", 4) == 0)
1180 info->feature_discard = 1;
1186 * Invoked when the backend is finally 'ready' (and has told produced
1187 * the details about the physical device - #sectors, size, etc).
1189 static void blkfront_connect(struct blkfront_info *info)
1191 unsigned long long sectors;
1192 unsigned long sector_size;
1195 int barrier, flush, discard;
1197 switch (info->connected) {
1198 case BLKIF_STATE_CONNECTED:
1200 * Potentially, the back-end may be signalling
1201 * a capacity change; update the capacity.
1203 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1204 "sectors", "%Lu", §ors);
1205 if (XENBUS_EXIST_ERR(err))
1207 printk(KERN_INFO "Setting capacity to %Lu\n",
1209 set_capacity(info->gd, sectors);
1210 revalidate_disk(info->gd);
1213 case BLKIF_STATE_SUSPENDED:
1220 dev_dbg(&info->xbdev->dev, "%s:%s.\n",
1221 __func__, info->xbdev->otherend);
1223 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1224 "sectors", "%llu", §ors,
1225 "info", "%u", &binfo,
1226 "sector-size", "%lu", §or_size,
1229 xenbus_dev_fatal(info->xbdev, err,
1230 "reading backend fields at %s",
1231 info->xbdev->otherend);
1235 info->feature_flush = 0;
1238 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1239 "feature-barrier", "%d", &barrier,
1243 * If there's no "feature-barrier" defined, then it means
1244 * we're dealing with a very old backend which writes
1245 * synchronously; nothing to do.
1247 * If there are barriers, then we use flush.
1249 if (!err && barrier) {
1250 info->feature_flush = REQ_FLUSH | REQ_FUA;
1251 info->flush_op = BLKIF_OP_WRITE_BARRIER;
1254 * And if there is "feature-flush-cache" use that above
1257 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1258 "feature-flush-cache", "%d", &flush,
1261 if (!err && flush) {
1262 info->feature_flush = REQ_FLUSH;
1263 info->flush_op = BLKIF_OP_FLUSH_DISKCACHE;
1266 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1267 "feature-discard", "%d", &discard,
1270 if (!err && discard)
1271 blkfront_setup_discard(info);
1273 err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size);
1275 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
1276 info->xbdev->otherend);
1280 xenbus_switch_state(info->xbdev, XenbusStateConnected);
1282 /* Kick pending requests. */
1283 spin_lock_irq(&info->io_lock);
1284 info->connected = BLKIF_STATE_CONNECTED;
1285 kick_pending_request_queues(info);
1286 spin_unlock_irq(&info->io_lock);
1294 * Callback received when the backend's state changes.
1296 static void blkback_changed(struct xenbus_device *dev,
1297 enum xenbus_state backend_state)
1299 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1301 dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1303 switch (backend_state) {
1304 case XenbusStateInitialising:
1305 case XenbusStateInitWait:
1306 case XenbusStateInitialised:
1307 case XenbusStateReconfiguring:
1308 case XenbusStateReconfigured:
1309 case XenbusStateUnknown:
1310 case XenbusStateClosed:
1313 case XenbusStateConnected:
1314 blkfront_connect(info);
1317 case XenbusStateClosing:
1318 blkfront_closing(info);
1323 static int blkfront_remove(struct xenbus_device *xbdev)
1325 struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
1326 struct block_device *bdev = NULL;
1327 struct gendisk *disk;
1329 dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
1331 blkif_free(info, 0);
1333 mutex_lock(&info->mutex);
1337 bdev = bdget_disk(disk, 0);
1340 mutex_unlock(&info->mutex);
1348 * The xbdev was removed before we reached the Closed
1349 * state. See if it's safe to remove the disk. If the bdev
1350 * isn't closed yet, we let release take care of it.
1353 mutex_lock(&bdev->bd_mutex);
1354 info = disk->private_data;
1356 dev_warn(disk_to_dev(disk),
1357 "%s was hot-unplugged, %d stale handles\n",
1358 xbdev->nodename, bdev->bd_openers);
1360 if (info && !bdev->bd_openers) {
1361 xlvbd_release_gendisk(info);
1362 disk->private_data = NULL;
1366 mutex_unlock(&bdev->bd_mutex);
1372 static int blkfront_is_ready(struct xenbus_device *dev)
1374 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1376 return info->is_ready && info->xbdev;
1379 static int blkif_open(struct block_device *bdev, fmode_t mode)
1381 struct gendisk *disk = bdev->bd_disk;
1382 struct blkfront_info *info;
1385 mutex_lock(&blkfront_mutex);
1387 info = disk->private_data;
1394 mutex_lock(&info->mutex);
1397 /* xbdev is closed */
1400 mutex_unlock(&info->mutex);
1403 mutex_unlock(&blkfront_mutex);
1407 static int blkif_release(struct gendisk *disk, fmode_t mode)
1409 struct blkfront_info *info = disk->private_data;
1410 struct block_device *bdev;
1411 struct xenbus_device *xbdev;
1413 mutex_lock(&blkfront_mutex);
1415 bdev = bdget_disk(disk, 0);
1417 if (bdev->bd_openers)
1421 * Check if we have been instructed to close. We will have
1422 * deferred this request, because the bdev was still open.
1425 mutex_lock(&info->mutex);
1426 xbdev = info->xbdev;
1428 if (xbdev && xbdev->state == XenbusStateClosing) {
1429 /* pending switch to state closed */
1430 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1431 xlvbd_release_gendisk(info);
1432 xenbus_frontend_closed(info->xbdev);
1435 mutex_unlock(&info->mutex);
1438 /* sudden device removal */
1439 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1440 xlvbd_release_gendisk(info);
1441 disk->private_data = NULL;
1447 mutex_unlock(&blkfront_mutex);
1451 static const struct block_device_operations xlvbd_block_fops =
1453 .owner = THIS_MODULE,
1455 .release = blkif_release,
1456 .getgeo = blkif_getgeo,
1457 .ioctl = blkif_ioctl,
1461 static const struct xenbus_device_id blkfront_ids[] = {
1466 static DEFINE_XENBUS_DRIVER(blkfront, ,
1467 .probe = blkfront_probe,
1468 .remove = blkfront_remove,
1469 .resume = blkfront_resume,
1470 .otherend_changed = blkback_changed,
1471 .is_ready = blkfront_is_ready,
1474 static int __init xlblk_init(void)
1481 if (xen_hvm_domain() && !xen_platform_pci_unplug)
1484 if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
1485 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
1486 XENVBD_MAJOR, DEV_NAME);
1490 ret = xenbus_register_frontend(&blkfront_driver);
1492 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
1498 module_init(xlblk_init);
1501 static void __exit xlblk_exit(void)
1503 xenbus_unregister_driver(&blkfront_driver);
1504 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
1507 module_exit(xlblk_exit);
1509 MODULE_DESCRIPTION("Xen virtual block device frontend");
1510 MODULE_LICENSE("GPL");
1511 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
1512 MODULE_ALIAS("xen:vbd");
1513 MODULE_ALIAS("xenblk");