Merge remote-tracking branch 'moduleh/module.h-split'
[linux-2.6/next.git] / drivers / block / xen-blkfront.c
blob67d656b1adc6894e1f57b5706c7b79d1868a0f69
1 /*
2 * blkfront.c
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
35 * IN THE SOFTWARE.
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>
47 #include <xen/xen.h>
48 #include <xen/xenbus.h>
49 #include <xen/grant_table.h>
50 #include <xen/events.h>
51 #include <xen/page.h>
52 #include <xen/platform_pci.h>
54 #include <xen/interface/grant_table.h>
55 #include <xen/interface/io/blkif.h>
56 #include <xen/interface/io/protocols.h>
58 #include <asm/xen/hypervisor.h>
60 enum blkif_state {
61 BLKIF_STATE_DISCONNECTED,
62 BLKIF_STATE_CONNECTED,
63 BLKIF_STATE_SUSPENDED,
66 struct blk_shadow {
67 struct blkif_request req;
68 struct request *request;
69 unsigned long frame[BLKIF_MAX_SEGMENTS_PER_REQUEST];
72 static DEFINE_MUTEX(blkfront_mutex);
73 static const struct block_device_operations xlvbd_block_fops;
75 #define BLK_RING_SIZE __CONST_RING_SIZE(blkif, PAGE_SIZE)
78 * We have one of these per vbd, whether ide, scsi or 'other'. They
79 * hang in private_data off the gendisk structure. We may end up
80 * putting all kinds of interesting stuff here :-)
82 struct blkfront_info
84 struct mutex mutex;
85 struct xenbus_device *xbdev;
86 struct gendisk *gd;
87 int vdevice;
88 blkif_vdev_t handle;
89 enum blkif_state connected;
90 int ring_ref;
91 struct blkif_front_ring ring;
92 struct scatterlist sg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
93 unsigned int evtchn, irq;
94 struct request_queue *rq;
95 struct work_struct work;
96 struct gnttab_free_callback callback;
97 struct blk_shadow shadow[BLK_RING_SIZE];
98 unsigned long shadow_free;
99 unsigned int feature_flush;
100 unsigned int flush_op;
101 int is_ready;
104 static DEFINE_SPINLOCK(blkif_io_lock);
106 static unsigned int nr_minors;
107 static unsigned long *minors;
108 static DEFINE_SPINLOCK(minor_lock);
110 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
111 (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
112 #define GRANT_INVALID_REF 0
114 #define PARTS_PER_DISK 16
115 #define PARTS_PER_EXT_DISK 256
117 #define BLKIF_MAJOR(dev) ((dev)>>8)
118 #define BLKIF_MINOR(dev) ((dev) & 0xff)
120 #define EXT_SHIFT 28
121 #define EXTENDED (1<<EXT_SHIFT)
122 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
123 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
124 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
125 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
126 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
127 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
129 #define DEV_NAME "xvd" /* name in /dev */
131 static int get_id_from_freelist(struct blkfront_info *info)
133 unsigned long free = info->shadow_free;
134 BUG_ON(free >= BLK_RING_SIZE);
135 info->shadow_free = info->shadow[free].req.id;
136 info->shadow[free].req.id = 0x0fffffee; /* debug */
137 return free;
140 static void add_id_to_freelist(struct blkfront_info *info,
141 unsigned long id)
143 info->shadow[id].req.id = info->shadow_free;
144 info->shadow[id].request = NULL;
145 info->shadow_free = id;
148 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
150 unsigned int end = minor + nr;
151 int rc;
153 if (end > nr_minors) {
154 unsigned long *bitmap, *old;
156 bitmap = kzalloc(BITS_TO_LONGS(end) * sizeof(*bitmap),
157 GFP_KERNEL);
158 if (bitmap == NULL)
159 return -ENOMEM;
161 spin_lock(&minor_lock);
162 if (end > nr_minors) {
163 old = minors;
164 memcpy(bitmap, minors,
165 BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
166 minors = bitmap;
167 nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
168 } else
169 old = bitmap;
170 spin_unlock(&minor_lock);
171 kfree(old);
174 spin_lock(&minor_lock);
175 if (find_next_bit(minors, end, minor) >= end) {
176 for (; minor < end; ++minor)
177 __set_bit(minor, minors);
178 rc = 0;
179 } else
180 rc = -EBUSY;
181 spin_unlock(&minor_lock);
183 return rc;
186 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
188 unsigned int end = minor + nr;
190 BUG_ON(end > nr_minors);
191 spin_lock(&minor_lock);
192 for (; minor < end; ++minor)
193 __clear_bit(minor, minors);
194 spin_unlock(&minor_lock);
197 static void blkif_restart_queue_callback(void *arg)
199 struct blkfront_info *info = (struct blkfront_info *)arg;
200 schedule_work(&info->work);
203 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
205 /* We don't have real geometry info, but let's at least return
206 values consistent with the size of the device */
207 sector_t nsect = get_capacity(bd->bd_disk);
208 sector_t cylinders = nsect;
210 hg->heads = 0xff;
211 hg->sectors = 0x3f;
212 sector_div(cylinders, hg->heads * hg->sectors);
213 hg->cylinders = cylinders;
214 if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
215 hg->cylinders = 0xffff;
216 return 0;
219 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
220 unsigned command, unsigned long argument)
222 struct blkfront_info *info = bdev->bd_disk->private_data;
223 int i;
225 dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
226 command, (long)argument);
228 switch (command) {
229 case CDROMMULTISESSION:
230 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
231 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
232 if (put_user(0, (char __user *)(argument + i)))
233 return -EFAULT;
234 return 0;
236 case CDROM_GET_CAPABILITY: {
237 struct gendisk *gd = info->gd;
238 if (gd->flags & GENHD_FL_CD)
239 return 0;
240 return -EINVAL;
243 default:
244 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
245 command);*/
246 return -EINVAL; /* same return as native Linux */
249 return 0;
253 * Generate a Xen blkfront IO request from a blk layer request. Reads
254 * and writes are handled as expected.
256 * @req: a request struct
258 static int blkif_queue_request(struct request *req)
260 struct blkfront_info *info = req->rq_disk->private_data;
261 unsigned long buffer_mfn;
262 struct blkif_request *ring_req;
263 unsigned long id;
264 unsigned int fsect, lsect;
265 int i, ref;
266 grant_ref_t gref_head;
267 struct scatterlist *sg;
269 if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
270 return 1;
272 if (gnttab_alloc_grant_references(
273 BLKIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) {
274 gnttab_request_free_callback(
275 &info->callback,
276 blkif_restart_queue_callback,
277 info,
278 BLKIF_MAX_SEGMENTS_PER_REQUEST);
279 return 1;
282 /* Fill out a communications ring structure. */
283 ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
284 id = get_id_from_freelist(info);
285 info->shadow[id].request = req;
287 ring_req->id = id;
288 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
289 ring_req->handle = info->handle;
291 ring_req->operation = rq_data_dir(req) ?
292 BLKIF_OP_WRITE : BLKIF_OP_READ;
294 if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) {
296 * Ideally we can do an unordered flush-to-disk. In case the
297 * backend onlysupports barriers, use that. A barrier request
298 * a superset of FUA, so we can implement it the same
299 * way. (It's also a FLUSH+FUA, since it is
300 * guaranteed ordered WRT previous writes.)
302 ring_req->operation = info->flush_op;
305 ring_req->nr_segments = blk_rq_map_sg(req->q, req, info->sg);
306 BUG_ON(ring_req->nr_segments > BLKIF_MAX_SEGMENTS_PER_REQUEST);
308 for_each_sg(info->sg, sg, ring_req->nr_segments, i) {
309 buffer_mfn = pfn_to_mfn(page_to_pfn(sg_page(sg)));
310 fsect = sg->offset >> 9;
311 lsect = fsect + (sg->length >> 9) - 1;
312 /* install a grant reference. */
313 ref = gnttab_claim_grant_reference(&gref_head);
314 BUG_ON(ref == -ENOSPC);
316 gnttab_grant_foreign_access_ref(
317 ref,
318 info->xbdev->otherend_id,
319 buffer_mfn,
320 rq_data_dir(req) );
322 info->shadow[id].frame[i] = mfn_to_pfn(buffer_mfn);
323 ring_req->u.rw.seg[i] =
324 (struct blkif_request_segment) {
325 .gref = ref,
326 .first_sect = fsect,
327 .last_sect = lsect };
330 info->ring.req_prod_pvt++;
332 /* Keep a private copy so we can reissue requests when recovering. */
333 info->shadow[id].req = *ring_req;
335 gnttab_free_grant_references(gref_head);
337 return 0;
341 static inline void flush_requests(struct blkfront_info *info)
343 int notify;
345 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
347 if (notify)
348 notify_remote_via_irq(info->irq);
352 * do_blkif_request
353 * read a block; request is in a request queue
355 static void do_blkif_request(struct request_queue *rq)
357 struct blkfront_info *info = NULL;
358 struct request *req;
359 int queued;
361 pr_debug("Entered do_blkif_request\n");
363 queued = 0;
365 while ((req = blk_peek_request(rq)) != NULL) {
366 info = req->rq_disk->private_data;
368 if (RING_FULL(&info->ring))
369 goto wait;
371 blk_start_request(req);
373 if (req->cmd_type != REQ_TYPE_FS) {
374 __blk_end_request_all(req, -EIO);
375 continue;
378 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
379 "(%u/%u) buffer:%p [%s]\n",
380 req, req->cmd, (unsigned long)blk_rq_pos(req),
381 blk_rq_cur_sectors(req), blk_rq_sectors(req),
382 req->buffer, rq_data_dir(req) ? "write" : "read");
384 if (blkif_queue_request(req)) {
385 blk_requeue_request(rq, req);
386 wait:
387 /* Avoid pointless unplugs. */
388 blk_stop_queue(rq);
389 break;
392 queued++;
395 if (queued != 0)
396 flush_requests(info);
399 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
401 struct request_queue *rq;
403 rq = blk_init_queue(do_blkif_request, &blkif_io_lock);
404 if (rq == NULL)
405 return -1;
407 queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
409 /* Hard sector size and max sectors impersonate the equiv. hardware. */
410 blk_queue_logical_block_size(rq, sector_size);
411 blk_queue_max_hw_sectors(rq, 512);
413 /* Each segment in a request is up to an aligned page in size. */
414 blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
415 blk_queue_max_segment_size(rq, PAGE_SIZE);
417 /* Ensure a merged request will fit in a single I/O ring slot. */
418 blk_queue_max_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
420 /* Make sure buffer addresses are sector-aligned. */
421 blk_queue_dma_alignment(rq, 511);
423 /* Make sure we don't use bounce buffers. */
424 blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
426 gd->queue = rq;
428 return 0;
432 static void xlvbd_flush(struct blkfront_info *info)
434 blk_queue_flush(info->rq, info->feature_flush);
435 printk(KERN_INFO "blkfront: %s: %s: %s\n",
436 info->gd->disk_name,
437 info->flush_op == BLKIF_OP_WRITE_BARRIER ?
438 "barrier" : (info->flush_op == BLKIF_OP_FLUSH_DISKCACHE ?
439 "flush diskcache" : "barrier or flush"),
440 info->feature_flush ? "enabled" : "disabled");
443 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
445 int major;
446 major = BLKIF_MAJOR(vdevice);
447 *minor = BLKIF_MINOR(vdevice);
448 switch (major) {
449 case XEN_IDE0_MAJOR:
450 *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
451 *minor = ((*minor / 64) * PARTS_PER_DISK) +
452 EMULATED_HD_DISK_MINOR_OFFSET;
453 break;
454 case XEN_IDE1_MAJOR:
455 *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
456 *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
457 EMULATED_HD_DISK_MINOR_OFFSET;
458 break;
459 case XEN_SCSI_DISK0_MAJOR:
460 *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
461 *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
462 break;
463 case XEN_SCSI_DISK1_MAJOR:
464 case XEN_SCSI_DISK2_MAJOR:
465 case XEN_SCSI_DISK3_MAJOR:
466 case XEN_SCSI_DISK4_MAJOR:
467 case XEN_SCSI_DISK5_MAJOR:
468 case XEN_SCSI_DISK6_MAJOR:
469 case XEN_SCSI_DISK7_MAJOR:
470 *offset = (*minor / PARTS_PER_DISK) +
471 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
472 EMULATED_SD_DISK_NAME_OFFSET;
473 *minor = *minor +
474 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
475 EMULATED_SD_DISK_MINOR_OFFSET;
476 break;
477 case XEN_SCSI_DISK8_MAJOR:
478 case XEN_SCSI_DISK9_MAJOR:
479 case XEN_SCSI_DISK10_MAJOR:
480 case XEN_SCSI_DISK11_MAJOR:
481 case XEN_SCSI_DISK12_MAJOR:
482 case XEN_SCSI_DISK13_MAJOR:
483 case XEN_SCSI_DISK14_MAJOR:
484 case XEN_SCSI_DISK15_MAJOR:
485 *offset = (*minor / PARTS_PER_DISK) +
486 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
487 EMULATED_SD_DISK_NAME_OFFSET;
488 *minor = *minor +
489 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
490 EMULATED_SD_DISK_MINOR_OFFSET;
491 break;
492 case XENVBD_MAJOR:
493 *offset = *minor / PARTS_PER_DISK;
494 break;
495 default:
496 printk(KERN_WARNING "blkfront: your disk configuration is "
497 "incorrect, please use an xvd device instead\n");
498 return -ENODEV;
500 return 0;
503 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
504 struct blkfront_info *info,
505 u16 vdisk_info, u16 sector_size)
507 struct gendisk *gd;
508 int nr_minors = 1;
509 int err;
510 unsigned int offset;
511 int minor;
512 int nr_parts;
514 BUG_ON(info->gd != NULL);
515 BUG_ON(info->rq != NULL);
517 if ((info->vdevice>>EXT_SHIFT) > 1) {
518 /* this is above the extended range; something is wrong */
519 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
520 return -ENODEV;
523 if (!VDEV_IS_EXTENDED(info->vdevice)) {
524 err = xen_translate_vdev(info->vdevice, &minor, &offset);
525 if (err)
526 return err;
527 nr_parts = PARTS_PER_DISK;
528 } else {
529 minor = BLKIF_MINOR_EXT(info->vdevice);
530 nr_parts = PARTS_PER_EXT_DISK;
531 offset = minor / nr_parts;
532 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
533 printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
534 "emulated IDE disks,\n\t choose an xvd device name"
535 "from xvde on\n", info->vdevice);
537 err = -ENODEV;
539 if ((minor % nr_parts) == 0)
540 nr_minors = nr_parts;
542 err = xlbd_reserve_minors(minor, nr_minors);
543 if (err)
544 goto out;
545 err = -ENODEV;
547 gd = alloc_disk(nr_minors);
548 if (gd == NULL)
549 goto release;
551 if (nr_minors > 1) {
552 if (offset < 26)
553 sprintf(gd->disk_name, "%s%c", DEV_NAME, 'a' + offset);
554 else
555 sprintf(gd->disk_name, "%s%c%c", DEV_NAME,
556 'a' + ((offset / 26)-1), 'a' + (offset % 26));
557 } else {
558 if (offset < 26)
559 sprintf(gd->disk_name, "%s%c%d", DEV_NAME,
560 'a' + offset,
561 minor & (nr_parts - 1));
562 else
563 sprintf(gd->disk_name, "%s%c%c%d", DEV_NAME,
564 'a' + ((offset / 26) - 1),
565 'a' + (offset % 26),
566 minor & (nr_parts - 1));
569 gd->major = XENVBD_MAJOR;
570 gd->first_minor = minor;
571 gd->fops = &xlvbd_block_fops;
572 gd->private_data = info;
573 gd->driverfs_dev = &(info->xbdev->dev);
574 set_capacity(gd, capacity);
576 if (xlvbd_init_blk_queue(gd, sector_size)) {
577 del_gendisk(gd);
578 goto release;
581 info->rq = gd->queue;
582 info->gd = gd;
584 xlvbd_flush(info);
586 if (vdisk_info & VDISK_READONLY)
587 set_disk_ro(gd, 1);
589 if (vdisk_info & VDISK_REMOVABLE)
590 gd->flags |= GENHD_FL_REMOVABLE;
592 if (vdisk_info & VDISK_CDROM)
593 gd->flags |= GENHD_FL_CD;
595 return 0;
597 release:
598 xlbd_release_minors(minor, nr_minors);
599 out:
600 return err;
603 static void xlvbd_release_gendisk(struct blkfront_info *info)
605 unsigned int minor, nr_minors;
606 unsigned long flags;
608 if (info->rq == NULL)
609 return;
611 spin_lock_irqsave(&blkif_io_lock, flags);
613 /* No more blkif_request(). */
614 blk_stop_queue(info->rq);
616 /* No more gnttab callback work. */
617 gnttab_cancel_free_callback(&info->callback);
618 spin_unlock_irqrestore(&blkif_io_lock, flags);
620 /* Flush gnttab callback work. Must be done with no locks held. */
621 flush_work_sync(&info->work);
623 del_gendisk(info->gd);
625 minor = info->gd->first_minor;
626 nr_minors = info->gd->minors;
627 xlbd_release_minors(minor, nr_minors);
629 blk_cleanup_queue(info->rq);
630 info->rq = NULL;
632 put_disk(info->gd);
633 info->gd = NULL;
636 static void kick_pending_request_queues(struct blkfront_info *info)
638 if (!RING_FULL(&info->ring)) {
639 /* Re-enable calldowns. */
640 blk_start_queue(info->rq);
641 /* Kick things off immediately. */
642 do_blkif_request(info->rq);
646 static void blkif_restart_queue(struct work_struct *work)
648 struct blkfront_info *info = container_of(work, struct blkfront_info, work);
650 spin_lock_irq(&blkif_io_lock);
651 if (info->connected == BLKIF_STATE_CONNECTED)
652 kick_pending_request_queues(info);
653 spin_unlock_irq(&blkif_io_lock);
656 static void blkif_free(struct blkfront_info *info, int suspend)
658 /* Prevent new requests being issued until we fix things up. */
659 spin_lock_irq(&blkif_io_lock);
660 info->connected = suspend ?
661 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
662 /* No more blkif_request(). */
663 if (info->rq)
664 blk_stop_queue(info->rq);
665 /* No more gnttab callback work. */
666 gnttab_cancel_free_callback(&info->callback);
667 spin_unlock_irq(&blkif_io_lock);
669 /* Flush gnttab callback work. Must be done with no locks held. */
670 flush_work_sync(&info->work);
672 /* Free resources associated with old device channel. */
673 if (info->ring_ref != GRANT_INVALID_REF) {
674 gnttab_end_foreign_access(info->ring_ref, 0,
675 (unsigned long)info->ring.sring);
676 info->ring_ref = GRANT_INVALID_REF;
677 info->ring.sring = NULL;
679 if (info->irq)
680 unbind_from_irqhandler(info->irq, info);
681 info->evtchn = info->irq = 0;
685 static void blkif_completion(struct blk_shadow *s)
687 int i;
688 for (i = 0; i < s->req.nr_segments; i++)
689 gnttab_end_foreign_access(s->req.u.rw.seg[i].gref, 0, 0UL);
692 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
694 struct request *req;
695 struct blkif_response *bret;
696 RING_IDX i, rp;
697 unsigned long flags;
698 struct blkfront_info *info = (struct blkfront_info *)dev_id;
699 int error;
701 spin_lock_irqsave(&blkif_io_lock, flags);
703 if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
704 spin_unlock_irqrestore(&blkif_io_lock, flags);
705 return IRQ_HANDLED;
708 again:
709 rp = info->ring.sring->rsp_prod;
710 rmb(); /* Ensure we see queued responses up to 'rp'. */
712 for (i = info->ring.rsp_cons; i != rp; i++) {
713 unsigned long id;
715 bret = RING_GET_RESPONSE(&info->ring, i);
716 id = bret->id;
717 req = info->shadow[id].request;
719 blkif_completion(&info->shadow[id]);
721 add_id_to_freelist(info, id);
723 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
724 switch (bret->operation) {
725 case BLKIF_OP_FLUSH_DISKCACHE:
726 case BLKIF_OP_WRITE_BARRIER:
727 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
728 printk(KERN_WARNING "blkfront: %s: write %s op failed\n",
729 info->flush_op == BLKIF_OP_WRITE_BARRIER ?
730 "barrier" : "flush disk cache",
731 info->gd->disk_name);
732 error = -EOPNOTSUPP;
734 if (unlikely(bret->status == BLKIF_RSP_ERROR &&
735 info->shadow[id].req.nr_segments == 0)) {
736 printk(KERN_WARNING "blkfront: %s: empty write %s op failed\n",
737 info->flush_op == BLKIF_OP_WRITE_BARRIER ?
738 "barrier" : "flush disk cache",
739 info->gd->disk_name);
740 error = -EOPNOTSUPP;
742 if (unlikely(error)) {
743 if (error == -EOPNOTSUPP)
744 error = 0;
745 info->feature_flush = 0;
746 info->flush_op = 0;
747 xlvbd_flush(info);
749 /* fall through */
750 case BLKIF_OP_READ:
751 case BLKIF_OP_WRITE:
752 if (unlikely(bret->status != BLKIF_RSP_OKAY))
753 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
754 "request: %x\n", bret->status);
756 __blk_end_request_all(req, error);
757 break;
758 default:
759 BUG();
763 info->ring.rsp_cons = i;
765 if (i != info->ring.req_prod_pvt) {
766 int more_to_do;
767 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
768 if (more_to_do)
769 goto again;
770 } else
771 info->ring.sring->rsp_event = i + 1;
773 kick_pending_request_queues(info);
775 spin_unlock_irqrestore(&blkif_io_lock, flags);
777 return IRQ_HANDLED;
781 static int setup_blkring(struct xenbus_device *dev,
782 struct blkfront_info *info)
784 struct blkif_sring *sring;
785 int err;
787 info->ring_ref = GRANT_INVALID_REF;
789 sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
790 if (!sring) {
791 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
792 return -ENOMEM;
794 SHARED_RING_INIT(sring);
795 FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
797 sg_init_table(info->sg, BLKIF_MAX_SEGMENTS_PER_REQUEST);
799 err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
800 if (err < 0) {
801 free_page((unsigned long)sring);
802 info->ring.sring = NULL;
803 goto fail;
805 info->ring_ref = err;
807 err = xenbus_alloc_evtchn(dev, &info->evtchn);
808 if (err)
809 goto fail;
811 err = bind_evtchn_to_irqhandler(info->evtchn,
812 blkif_interrupt,
813 IRQF_SAMPLE_RANDOM, "blkif", info);
814 if (err <= 0) {
815 xenbus_dev_fatal(dev, err,
816 "bind_evtchn_to_irqhandler failed");
817 goto fail;
819 info->irq = err;
821 return 0;
822 fail:
823 blkif_free(info, 0);
824 return err;
828 /* Common code used when first setting up, and when resuming. */
829 static int talk_to_blkback(struct xenbus_device *dev,
830 struct blkfront_info *info)
832 const char *message = NULL;
833 struct xenbus_transaction xbt;
834 int err;
836 /* Create shared ring, alloc event channel. */
837 err = setup_blkring(dev, info);
838 if (err)
839 goto out;
841 again:
842 err = xenbus_transaction_start(&xbt);
843 if (err) {
844 xenbus_dev_fatal(dev, err, "starting transaction");
845 goto destroy_blkring;
848 err = xenbus_printf(xbt, dev->nodename,
849 "ring-ref", "%u", info->ring_ref);
850 if (err) {
851 message = "writing ring-ref";
852 goto abort_transaction;
854 err = xenbus_printf(xbt, dev->nodename,
855 "event-channel", "%u", info->evtchn);
856 if (err) {
857 message = "writing event-channel";
858 goto abort_transaction;
860 err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
861 XEN_IO_PROTO_ABI_NATIVE);
862 if (err) {
863 message = "writing protocol";
864 goto abort_transaction;
867 err = xenbus_transaction_end(xbt, 0);
868 if (err) {
869 if (err == -EAGAIN)
870 goto again;
871 xenbus_dev_fatal(dev, err, "completing transaction");
872 goto destroy_blkring;
875 xenbus_switch_state(dev, XenbusStateInitialised);
877 return 0;
879 abort_transaction:
880 xenbus_transaction_end(xbt, 1);
881 if (message)
882 xenbus_dev_fatal(dev, err, "%s", message);
883 destroy_blkring:
884 blkif_free(info, 0);
885 out:
886 return err;
890 * Entry point to this code when a new device is created. Allocate the basic
891 * structures and the ring buffer for communication with the backend, and
892 * inform the backend of the appropriate details for those. Switch to
893 * Initialised state.
895 static int blkfront_probe(struct xenbus_device *dev,
896 const struct xenbus_device_id *id)
898 int err, vdevice, i;
899 struct blkfront_info *info;
901 /* FIXME: Use dynamic device id if this is not set. */
902 err = xenbus_scanf(XBT_NIL, dev->nodename,
903 "virtual-device", "%i", &vdevice);
904 if (err != 1) {
905 /* go looking in the extended area instead */
906 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
907 "%i", &vdevice);
908 if (err != 1) {
909 xenbus_dev_fatal(dev, err, "reading virtual-device");
910 return err;
914 if (xen_hvm_domain()) {
915 char *type;
916 int len;
917 /* no unplug has been done: do not hook devices != xen vbds */
918 if (xen_platform_pci_unplug & XEN_UNPLUG_UNNECESSARY) {
919 int major;
921 if (!VDEV_IS_EXTENDED(vdevice))
922 major = BLKIF_MAJOR(vdevice);
923 else
924 major = XENVBD_MAJOR;
926 if (major != XENVBD_MAJOR) {
927 printk(KERN_INFO
928 "%s: HVM does not support vbd %d as xen block device\n",
929 __FUNCTION__, vdevice);
930 return -ENODEV;
933 /* do not create a PV cdrom device if we are an HVM guest */
934 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
935 if (IS_ERR(type))
936 return -ENODEV;
937 if (strncmp(type, "cdrom", 5) == 0) {
938 kfree(type);
939 return -ENODEV;
941 kfree(type);
943 info = kzalloc(sizeof(*info), GFP_KERNEL);
944 if (!info) {
945 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
946 return -ENOMEM;
949 mutex_init(&info->mutex);
950 info->xbdev = dev;
951 info->vdevice = vdevice;
952 info->connected = BLKIF_STATE_DISCONNECTED;
953 INIT_WORK(&info->work, blkif_restart_queue);
955 for (i = 0; i < BLK_RING_SIZE; i++)
956 info->shadow[i].req.id = i+1;
957 info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
959 /* Front end dir is a number, which is used as the id. */
960 info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
961 dev_set_drvdata(&dev->dev, info);
963 err = talk_to_blkback(dev, info);
964 if (err) {
965 kfree(info);
966 dev_set_drvdata(&dev->dev, NULL);
967 return err;
970 return 0;
974 static int blkif_recover(struct blkfront_info *info)
976 int i;
977 struct blkif_request *req;
978 struct blk_shadow *copy;
979 int j;
981 /* Stage 1: Make a safe copy of the shadow state. */
982 copy = kmalloc(sizeof(info->shadow),
983 GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
984 if (!copy)
985 return -ENOMEM;
986 memcpy(copy, info->shadow, sizeof(info->shadow));
988 /* Stage 2: Set up free list. */
989 memset(&info->shadow, 0, sizeof(info->shadow));
990 for (i = 0; i < BLK_RING_SIZE; i++)
991 info->shadow[i].req.id = i+1;
992 info->shadow_free = info->ring.req_prod_pvt;
993 info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
995 /* Stage 3: Find pending requests and requeue them. */
996 for (i = 0; i < BLK_RING_SIZE; i++) {
997 /* Not in use? */
998 if (!copy[i].request)
999 continue;
1001 /* Grab a request slot and copy shadow state into it. */
1002 req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
1003 *req = copy[i].req;
1005 /* We get a new request id, and must reset the shadow state. */
1006 req->id = get_id_from_freelist(info);
1007 memcpy(&info->shadow[req->id], &copy[i], sizeof(copy[i]));
1009 /* Rewrite any grant references invalidated by susp/resume. */
1010 for (j = 0; j < req->nr_segments; j++)
1011 gnttab_grant_foreign_access_ref(
1012 req->u.rw.seg[j].gref,
1013 info->xbdev->otherend_id,
1014 pfn_to_mfn(info->shadow[req->id].frame[j]),
1015 rq_data_dir(info->shadow[req->id].request));
1016 info->shadow[req->id].req = *req;
1018 info->ring.req_prod_pvt++;
1021 kfree(copy);
1023 xenbus_switch_state(info->xbdev, XenbusStateConnected);
1025 spin_lock_irq(&blkif_io_lock);
1027 /* Now safe for us to use the shared ring */
1028 info->connected = BLKIF_STATE_CONNECTED;
1030 /* Send off requeued requests */
1031 flush_requests(info);
1033 /* Kick any other new requests queued since we resumed */
1034 kick_pending_request_queues(info);
1036 spin_unlock_irq(&blkif_io_lock);
1038 return 0;
1042 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1043 * driver restart. We tear down our blkif structure and recreate it, but
1044 * leave the device-layer structures intact so that this is transparent to the
1045 * rest of the kernel.
1047 static int blkfront_resume(struct xenbus_device *dev)
1049 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1050 int err;
1052 dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
1054 blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
1056 err = talk_to_blkback(dev, info);
1057 if (info->connected == BLKIF_STATE_SUSPENDED && !err)
1058 err = blkif_recover(info);
1060 return err;
1063 static void
1064 blkfront_closing(struct blkfront_info *info)
1066 struct xenbus_device *xbdev = info->xbdev;
1067 struct block_device *bdev = NULL;
1069 mutex_lock(&info->mutex);
1071 if (xbdev->state == XenbusStateClosing) {
1072 mutex_unlock(&info->mutex);
1073 return;
1076 if (info->gd)
1077 bdev = bdget_disk(info->gd, 0);
1079 mutex_unlock(&info->mutex);
1081 if (!bdev) {
1082 xenbus_frontend_closed(xbdev);
1083 return;
1086 mutex_lock(&bdev->bd_mutex);
1088 if (bdev->bd_openers) {
1089 xenbus_dev_error(xbdev, -EBUSY,
1090 "Device in use; refusing to close");
1091 xenbus_switch_state(xbdev, XenbusStateClosing);
1092 } else {
1093 xlvbd_release_gendisk(info);
1094 xenbus_frontend_closed(xbdev);
1097 mutex_unlock(&bdev->bd_mutex);
1098 bdput(bdev);
1102 * Invoked when the backend is finally 'ready' (and has told produced
1103 * the details about the physical device - #sectors, size, etc).
1105 static void blkfront_connect(struct blkfront_info *info)
1107 unsigned long long sectors;
1108 unsigned long sector_size;
1109 unsigned int binfo;
1110 int err;
1111 int barrier, flush;
1113 switch (info->connected) {
1114 case BLKIF_STATE_CONNECTED:
1116 * Potentially, the back-end may be signalling
1117 * a capacity change; update the capacity.
1119 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1120 "sectors", "%Lu", &sectors);
1121 if (XENBUS_EXIST_ERR(err))
1122 return;
1123 printk(KERN_INFO "Setting capacity to %Lu\n",
1124 sectors);
1125 set_capacity(info->gd, sectors);
1126 revalidate_disk(info->gd);
1128 /* fall through */
1129 case BLKIF_STATE_SUSPENDED:
1130 return;
1132 default:
1133 break;
1136 dev_dbg(&info->xbdev->dev, "%s:%s.\n",
1137 __func__, info->xbdev->otherend);
1139 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1140 "sectors", "%llu", &sectors,
1141 "info", "%u", &binfo,
1142 "sector-size", "%lu", &sector_size,
1143 NULL);
1144 if (err) {
1145 xenbus_dev_fatal(info->xbdev, err,
1146 "reading backend fields at %s",
1147 info->xbdev->otherend);
1148 return;
1151 info->feature_flush = REQ_FLUSH | REQ_FUA;
1152 info->flush_op = 0;
1154 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1155 "feature-barrier", "%d", &barrier,
1156 NULL);
1159 * If there's no "feature-barrier" defined, then it means
1160 * we're dealing with a very old backend which writes
1161 * synchronously; nothing to do.
1163 * If there are barriers, then we use flush.
1165 if (!err && barrier) {
1166 info->flush_op = BLKIF_OP_WRITE_BARRIER;
1169 * And if there is "feature-flush-cache" use that above
1170 * barriers.
1172 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1173 "feature-flush-cache", "%d", &flush,
1174 NULL);
1176 if (!err && flush) {
1177 info->flush_op = BLKIF_OP_FLUSH_DISKCACHE;
1180 if (!flush && !barrier)
1181 info->feature_flush = 0;
1183 err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size);
1184 if (err) {
1185 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
1186 info->xbdev->otherend);
1187 return;
1190 xenbus_switch_state(info->xbdev, XenbusStateConnected);
1192 /* Kick pending requests. */
1193 spin_lock_irq(&blkif_io_lock);
1194 info->connected = BLKIF_STATE_CONNECTED;
1195 kick_pending_request_queues(info);
1196 spin_unlock_irq(&blkif_io_lock);
1198 add_disk(info->gd);
1200 info->is_ready = 1;
1204 * Callback received when the backend's state changes.
1206 static void blkback_changed(struct xenbus_device *dev,
1207 enum xenbus_state backend_state)
1209 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1211 dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1213 switch (backend_state) {
1214 case XenbusStateInitialising:
1215 case XenbusStateInitWait:
1216 case XenbusStateInitialised:
1217 case XenbusStateReconfiguring:
1218 case XenbusStateReconfigured:
1219 case XenbusStateUnknown:
1220 case XenbusStateClosed:
1221 break;
1223 case XenbusStateConnected:
1224 blkfront_connect(info);
1225 break;
1227 case XenbusStateClosing:
1228 blkfront_closing(info);
1229 break;
1233 static int blkfront_remove(struct xenbus_device *xbdev)
1235 struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
1236 struct block_device *bdev = NULL;
1237 struct gendisk *disk;
1239 dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
1241 blkif_free(info, 0);
1243 mutex_lock(&info->mutex);
1245 disk = info->gd;
1246 if (disk)
1247 bdev = bdget_disk(disk, 0);
1249 info->xbdev = NULL;
1250 mutex_unlock(&info->mutex);
1252 if (!bdev) {
1253 kfree(info);
1254 return 0;
1258 * The xbdev was removed before we reached the Closed
1259 * state. See if it's safe to remove the disk. If the bdev
1260 * isn't closed yet, we let release take care of it.
1263 mutex_lock(&bdev->bd_mutex);
1264 info = disk->private_data;
1266 dev_warn(disk_to_dev(disk),
1267 "%s was hot-unplugged, %d stale handles\n",
1268 xbdev->nodename, bdev->bd_openers);
1270 if (info && !bdev->bd_openers) {
1271 xlvbd_release_gendisk(info);
1272 disk->private_data = NULL;
1273 kfree(info);
1276 mutex_unlock(&bdev->bd_mutex);
1277 bdput(bdev);
1279 return 0;
1282 static int blkfront_is_ready(struct xenbus_device *dev)
1284 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1286 return info->is_ready && info->xbdev;
1289 static int blkif_open(struct block_device *bdev, fmode_t mode)
1291 struct gendisk *disk = bdev->bd_disk;
1292 struct blkfront_info *info;
1293 int err = 0;
1295 mutex_lock(&blkfront_mutex);
1297 info = disk->private_data;
1298 if (!info) {
1299 /* xbdev gone */
1300 err = -ERESTARTSYS;
1301 goto out;
1304 mutex_lock(&info->mutex);
1306 if (!info->gd)
1307 /* xbdev is closed */
1308 err = -ERESTARTSYS;
1310 mutex_unlock(&info->mutex);
1312 out:
1313 mutex_unlock(&blkfront_mutex);
1314 return err;
1317 static int blkif_release(struct gendisk *disk, fmode_t mode)
1319 struct blkfront_info *info = disk->private_data;
1320 struct block_device *bdev;
1321 struct xenbus_device *xbdev;
1323 mutex_lock(&blkfront_mutex);
1325 bdev = bdget_disk(disk, 0);
1326 bdput(bdev);
1328 if (bdev->bd_openers)
1329 goto out;
1332 * Check if we have been instructed to close. We will have
1333 * deferred this request, because the bdev was still open.
1336 mutex_lock(&info->mutex);
1337 xbdev = info->xbdev;
1339 if (xbdev && xbdev->state == XenbusStateClosing) {
1340 /* pending switch to state closed */
1341 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1342 xlvbd_release_gendisk(info);
1343 xenbus_frontend_closed(info->xbdev);
1346 mutex_unlock(&info->mutex);
1348 if (!xbdev) {
1349 /* sudden device removal */
1350 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1351 xlvbd_release_gendisk(info);
1352 disk->private_data = NULL;
1353 kfree(info);
1356 out:
1357 mutex_unlock(&blkfront_mutex);
1358 return 0;
1361 static const struct block_device_operations xlvbd_block_fops =
1363 .owner = THIS_MODULE,
1364 .open = blkif_open,
1365 .release = blkif_release,
1366 .getgeo = blkif_getgeo,
1367 .ioctl = blkif_ioctl,
1371 static const struct xenbus_device_id blkfront_ids[] = {
1372 { "vbd" },
1373 { "" }
1376 static struct xenbus_driver blkfront = {
1377 .name = "vbd",
1378 .owner = THIS_MODULE,
1379 .ids = blkfront_ids,
1380 .probe = blkfront_probe,
1381 .remove = blkfront_remove,
1382 .resume = blkfront_resume,
1383 .otherend_changed = blkback_changed,
1384 .is_ready = blkfront_is_ready,
1387 static int __init xlblk_init(void)
1389 if (!xen_domain())
1390 return -ENODEV;
1392 if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
1393 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
1394 XENVBD_MAJOR, DEV_NAME);
1395 return -ENODEV;
1398 return xenbus_register_frontend(&blkfront);
1400 module_init(xlblk_init);
1403 static void __exit xlblk_exit(void)
1405 return xenbus_unregister_driver(&blkfront);
1407 module_exit(xlblk_exit);
1409 MODULE_DESCRIPTION("Xen virtual block device frontend");
1410 MODULE_LICENSE("GPL");
1411 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
1412 MODULE_ALIAS("xen:vbd");
1413 MODULE_ALIAS("xenblk");