?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
xen-blkfront: fix a deadlock while handling discard response
[linux-2.6.git] / drivers / block / xen-blkfront.c
blob7d148776028347caebc9f71df925cad631ec3443
1 /*
2 * blkfront.c
3 *
4 * XenLinux virtual block device driver.
5 *
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
12 *
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:
18 *
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:
25 *
26 * The above copyright notice and this permission notice shall be included in
27 * all copies or substantial portions of the Software.
28 *
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.
36 */
37
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
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>
53
54 #include <xen/interface/grant_table.h>
55 #include <xen/interface/io/blkif.h>
56 #include <xen/interface/io/protocols.h>
57
58 #include <asm/xen/hypervisor.h>
59
60 enum blkif_state {
61 BLKIF_STATE_DISCONNECTED,
62 BLKIF_STATE_CONNECTED,
63 BLKIF_STATE_SUSPENDED,
64 };
65
66 struct blk_shadow {
67 struct blkif_request req;
68 struct request *request;
69 unsigned long frame[BLKIF_MAX_SEGMENTS_PER_REQUEST];
70 };
71
72 static DEFINE_MUTEX(blkfront_mutex);
73 static const struct block_device_operations xlvbd_block_fops;
74
75 #define BLK_RING_SIZE __CONST_RING_SIZE(blkif, PAGE_SIZE)
76
77 /*
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 :-)
81 */
82 struct blkfront_info
83 {
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 unsigned int feature_discard;
102 unsigned int discard_granularity;
103 unsigned int discard_alignment;
104 int is_ready;
105 };
106
107 static DEFINE_SPINLOCK(blkif_io_lock);
108
109 static unsigned int nr_minors;
110 static unsigned long *minors;
111 static DEFINE_SPINLOCK(minor_lock);
112
113 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
114 (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
115 #define GRANT_INVALID_REF 0
116
117 #define PARTS_PER_DISK 16
118 #define PARTS_PER_EXT_DISK 256
119
120 #define BLKIF_MAJOR(dev) ((dev)>>8)
121 #define BLKIF_MINOR(dev) ((dev) & 0xff)
122
123 #define EXT_SHIFT 28
124 #define EXTENDED (1<<EXT_SHIFT)
125 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
126 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
127 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
128 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
129 #define EMULATED_SD_DISK_MINOR_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET + (4 * 16))
130 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_HD_DISK_NAME_OFFSET + 4)
131
132 #define DEV_NAME "xvd" /* name in /dev */
133
134 static int get_id_from_freelist(struct blkfront_info *info)
135 {
136 unsigned long free = info->shadow_free;
137 BUG_ON(free >= BLK_RING_SIZE);
138 info->shadow_free = info->shadow[free].req.id;
139 info->shadow[free].req.id = 0x0fffffee; /* debug */
140 return free;
141 }
142
143 static void add_id_to_freelist(struct blkfront_info *info,
144 unsigned long id)
145 {
146 info->shadow[id].req.id = info->shadow_free;
147 info->shadow[id].request = NULL;
148 info->shadow_free = id;
149 }
150
151 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
152 {
153 unsigned int end = minor + nr;
154 int rc;
155
156 if (end > nr_minors) {
157 unsigned long *bitmap, *old;
158
159 bitmap = kzalloc(BITS_TO_LONGS(end) * sizeof(*bitmap),
160 GFP_KERNEL);
161 if (bitmap == NULL)
162 return -ENOMEM;
163
164 spin_lock(&minor_lock);
165 if (end > nr_minors) {
166 old = minors;
167 memcpy(bitmap, minors,
168 BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
169 minors = bitmap;
170 nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
171 } else
172 old = bitmap;
173 spin_unlock(&minor_lock);
174 kfree(old);
175 }
176
177 spin_lock(&minor_lock);
178 if (find_next_bit(minors, end, minor) >= end) {
179 for (; minor < end; ++minor)
180 __set_bit(minor, minors);
181 rc = 0;
182 } else
183 rc = -EBUSY;
184 spin_unlock(&minor_lock);
185
186 return rc;
187 }
188
189 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
190 {
191 unsigned int end = minor + nr;
192
193 BUG_ON(end > nr_minors);
194 spin_lock(&minor_lock);
195 for (; minor < end; ++minor)
196 __clear_bit(minor, minors);
197 spin_unlock(&minor_lock);
198 }
199
200 static void blkif_restart_queue_callback(void *arg)
201 {
202 struct blkfront_info *info = (struct blkfront_info *)arg;
203 schedule_work(&info->work);
204 }
205
206 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
207 {
208 /* We don't have real geometry info, but let's at least return
209 values consistent with the size of the device */
210 sector_t nsect = get_capacity(bd->bd_disk);
211 sector_t cylinders = nsect;
212
213 hg->heads = 0xff;
214 hg->sectors = 0x3f;
215 sector_div(cylinders, hg->heads * hg->sectors);
216 hg->cylinders = cylinders;
217 if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
218 hg->cylinders = 0xffff;
219 return 0;
220 }
221
222 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
223 unsigned command, unsigned long argument)
224 {
225 struct blkfront_info *info = bdev->bd_disk->private_data;
226 int i;
227
228 dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
229 command, (long)argument);
230
231 switch (command) {
232 case CDROMMULTISESSION:
233 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
234 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
235 if (put_user(0, (char __user *)(argument + i)))
236 return -EFAULT;
237 return 0;
238
239 case CDROM_GET_CAPABILITY: {
240 struct gendisk *gd = info->gd;
241 if (gd->flags & GENHD_FL_CD)
242 return 0;
243 return -EINVAL;
244 }
245
246 default:
247 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
248 command);*/
249 return -EINVAL; /* same return as native Linux */
250 }
251
252 return 0;
253 }
254
255 /*
256 * Generate a Xen blkfront IO request from a blk layer request. Reads
257 * and writes are handled as expected.
258 *
259 * @req: a request struct
260 */
261 static int blkif_queue_request(struct request *req)
262 {
263 struct blkfront_info *info = req->rq_disk->private_data;
264 unsigned long buffer_mfn;
265 struct blkif_request *ring_req;
266 unsigned long id;
267 unsigned int fsect, lsect;
268 int i, ref;
269 grant_ref_t gref_head;
270 struct scatterlist *sg;
271
272 if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
273 return 1;
274
275 if (gnttab_alloc_grant_references(
276 BLKIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) {
277 gnttab_request_free_callback(
278 &info->callback,
279 blkif_restart_queue_callback,
280 info,
281 BLKIF_MAX_SEGMENTS_PER_REQUEST);
282 return 1;
283 }
284
285 /* Fill out a communications ring structure. */
286 ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
287 id = get_id_from_freelist(info);
288 info->shadow[id].request = req;
289
290 ring_req->id = id;
291 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
292 ring_req->handle = info->handle;
293
294 ring_req->operation = rq_data_dir(req) ?
295 BLKIF_OP_WRITE : BLKIF_OP_READ;
296
297 if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) {
298 /*
299 * Ideally we can do an unordered flush-to-disk. In case the
300 * backend onlysupports barriers, use that. A barrier request
301 * a superset of FUA, so we can implement it the same
302 * way. (It's also a FLUSH+FUA, since it is
303 * guaranteed ordered WRT previous writes.)
304 */
305 ring_req->operation = info->flush_op;
306 }
307
308 if (unlikely(req->cmd_flags & REQ_DISCARD)) {
309 /* id, sector_number and handle are set above. */
310 ring_req->operation = BLKIF_OP_DISCARD;
311 ring_req->nr_segments = 0;
312 ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
313 } else {
314 ring_req->nr_segments = blk_rq_map_sg(req->q, req, info->sg);
315 BUG_ON(ring_req->nr_segments > BLKIF_MAX_SEGMENTS_PER_REQUEST);
316
317 for_each_sg(info->sg, sg, ring_req->nr_segments, i) {
318 buffer_mfn = pfn_to_mfn(page_to_pfn(sg_page(sg)));
319 fsect = sg->offset >> 9;
320 lsect = fsect + (sg->length >> 9) - 1;
321 /* install a grant reference. */
322 ref = gnttab_claim_grant_reference(&gref_head);
323 BUG_ON(ref == -ENOSPC);
324
325 gnttab_grant_foreign_access_ref(
326 ref,
327 info->xbdev->otherend_id,
328 buffer_mfn,
329 rq_data_dir(req));
330
331 info->shadow[id].frame[i] = mfn_to_pfn(buffer_mfn);
332 ring_req->u.rw.seg[i] =
333 (struct blkif_request_segment) {
334 .gref = ref,
335 .first_sect = fsect,
336 .last_sect = lsect };
337 }
338 }
339
340 info->ring.req_prod_pvt++;
341
342 /* Keep a private copy so we can reissue requests when recovering. */
343 info->shadow[id].req = *ring_req;
344
345 gnttab_free_grant_references(gref_head);
346
347 return 0;
348 }
349
350
351 static inline void flush_requests(struct blkfront_info *info)
352 {
353 int notify;
354
355 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
356
357 if (notify)
358 notify_remote_via_irq(info->irq);
359 }
360
361 /*
362 * do_blkif_request
363 * read a block; request is in a request queue
364 */
365 static void do_blkif_request(struct request_queue *rq)
366 {
367 struct blkfront_info *info = NULL;
368 struct request *req;
369 int queued;
370
371 pr_debug("Entered do_blkif_request\n");
372
373 queued = 0;
374
375 while ((req = blk_peek_request(rq)) != NULL) {
376 info = req->rq_disk->private_data;
377
378 if (RING_FULL(&info->ring))
379 goto wait;
380
381 blk_start_request(req);
382
383 if (req->cmd_type != REQ_TYPE_FS) {
384 __blk_end_request_all(req, -EIO);
385 continue;
386 }
387
388 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
389 "(%u/%u) buffer:%p [%s]\n",
390 req, req->cmd, (unsigned long)blk_rq_pos(req),
391 blk_rq_cur_sectors(req), blk_rq_sectors(req),
392 req->buffer, rq_data_dir(req) ? "write" : "read");
393
394 if (blkif_queue_request(req)) {
395 blk_requeue_request(rq, req);
396 wait:
397 /* Avoid pointless unplugs. */
398 blk_stop_queue(rq);
399 break;
400 }
401
402 queued++;
403 }
404
405 if (queued != 0)
406 flush_requests(info);
407 }
408
409 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
410 {
411 struct request_queue *rq;
412 struct blkfront_info *info = gd->private_data;
413
414 rq = blk_init_queue(do_blkif_request, &blkif_io_lock);
415 if (rq == NULL)
416 return -1;
417
418 queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
419
420 if (info->feature_discard) {
421 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
422 blk_queue_max_discard_sectors(rq, get_capacity(gd));
423 rq->limits.discard_granularity = info->discard_granularity;
424 rq->limits.discard_alignment = info->discard_alignment;
425 }
426
427 /* Hard sector size and max sectors impersonate the equiv. hardware. */
428 blk_queue_logical_block_size(rq, sector_size);
429 blk_queue_max_hw_sectors(rq, 512);
430
431 /* Each segment in a request is up to an aligned page in size. */
432 blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
433 blk_queue_max_segment_size(rq, PAGE_SIZE);
434
435 /* Ensure a merged request will fit in a single I/O ring slot. */
436 blk_queue_max_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
437
438 /* Make sure buffer addresses are sector-aligned. */
439 blk_queue_dma_alignment(rq, 511);
440
441 /* Make sure we don't use bounce buffers. */
442 blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
443
444 gd->queue = rq;
445
446 return 0;
447 }
448
449
450 static void xlvbd_flush(struct blkfront_info *info)
451 {
452 blk_queue_flush(info->rq, info->feature_flush);
453 printk(KERN_INFO "blkfront: %s: %s: %s\n",
454 info->gd->disk_name,
455 info->flush_op == BLKIF_OP_WRITE_BARRIER ?
456 "barrier" : (info->flush_op == BLKIF_OP_FLUSH_DISKCACHE ?
457 "flush diskcache" : "barrier or flush"),
458 info->feature_flush ? "enabled" : "disabled");
459 }
460
461 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
462 {
463 int major;
464 major = BLKIF_MAJOR(vdevice);
465 *minor = BLKIF_MINOR(vdevice);
466 switch (major) {
467 case XEN_IDE0_MAJOR:
468 *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
469 *minor = ((*minor / 64) * PARTS_PER_DISK) +
470 EMULATED_HD_DISK_MINOR_OFFSET;
471 break;
472 case XEN_IDE1_MAJOR:
473 *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
474 *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
475 EMULATED_HD_DISK_MINOR_OFFSET;
476 break;
477 case XEN_SCSI_DISK0_MAJOR:
478 *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
479 *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
480 break;
481 case XEN_SCSI_DISK1_MAJOR:
482 case XEN_SCSI_DISK2_MAJOR:
483 case XEN_SCSI_DISK3_MAJOR:
484 case XEN_SCSI_DISK4_MAJOR:
485 case XEN_SCSI_DISK5_MAJOR:
486 case XEN_SCSI_DISK6_MAJOR:
487 case XEN_SCSI_DISK7_MAJOR:
488 *offset = (*minor / PARTS_PER_DISK) +
489 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
490 EMULATED_SD_DISK_NAME_OFFSET;
491 *minor = *minor +
492 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
493 EMULATED_SD_DISK_MINOR_OFFSET;
494 break;
495 case XEN_SCSI_DISK8_MAJOR:
496 case XEN_SCSI_DISK9_MAJOR:
497 case XEN_SCSI_DISK10_MAJOR:
498 case XEN_SCSI_DISK11_MAJOR:
499 case XEN_SCSI_DISK12_MAJOR:
500 case XEN_SCSI_DISK13_MAJOR:
501 case XEN_SCSI_DISK14_MAJOR:
502 case XEN_SCSI_DISK15_MAJOR:
503 *offset = (*minor / PARTS_PER_DISK) +
504 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
505 EMULATED_SD_DISK_NAME_OFFSET;
506 *minor = *minor +
507 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
508 EMULATED_SD_DISK_MINOR_OFFSET;
509 break;
510 case XENVBD_MAJOR:
511 *offset = *minor / PARTS_PER_DISK;
512 break;
513 default:
514 printk(KERN_WARNING "blkfront: your disk configuration is "
515 "incorrect, please use an xvd device instead\n");
516 return -ENODEV;
517 }
518 return 0;
519 }
520
521 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
522 struct blkfront_info *info,
523 u16 vdisk_info, u16 sector_size)
524 {
525 struct gendisk *gd;
526 int nr_minors = 1;
527 int err;
528 unsigned int offset;
529 int minor;
530 int nr_parts;
531
532 BUG_ON(info->gd != NULL);
533 BUG_ON(info->rq != NULL);
534
535 if ((info->vdevice>>EXT_SHIFT) > 1) {
536 /* this is above the extended range; something is wrong */
537 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
538 return -ENODEV;
539 }
540
541 if (!VDEV_IS_EXTENDED(info->vdevice)) {
542 err = xen_translate_vdev(info->vdevice, &minor, &offset);
543 if (err)
544 return err;
545 nr_parts = PARTS_PER_DISK;
546 } else {
547 minor = BLKIF_MINOR_EXT(info->vdevice);
548 nr_parts = PARTS_PER_EXT_DISK;
549 offset = minor / nr_parts;
550 if (xen_hvm_domain() && offset <= EMULATED_HD_DISK_NAME_OFFSET + 4)
551 printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
552 "emulated IDE disks,\n\t choose an xvd device name"
553 "from xvde on\n", info->vdevice);
554 }
555 err = -ENODEV;
556
557 if ((minor % nr_parts) == 0)
558 nr_minors = nr_parts;
559
560 err = xlbd_reserve_minors(minor, nr_minors);
561 if (err)
562 goto out;
563 err = -ENODEV;
564
565 gd = alloc_disk(nr_minors);
566 if (gd == NULL)
567 goto release;
568
569 if (nr_minors > 1) {
570 if (offset < 26)
571 sprintf(gd->disk_name, "%s%c", DEV_NAME, 'a' + offset);
572 else
573 sprintf(gd->disk_name, "%s%c%c", DEV_NAME,
574 'a' + ((offset / 26)-1), 'a' + (offset % 26));
575 } else {
576 if (offset < 26)
577 sprintf(gd->disk_name, "%s%c%d", DEV_NAME,
578 'a' + offset,
579 minor & (nr_parts - 1));
580 else
581 sprintf(gd->disk_name, "%s%c%c%d", DEV_NAME,
582 'a' + ((offset / 26) - 1),
583 'a' + (offset % 26),
584 minor & (nr_parts - 1));
585 }
586
587 gd->major = XENVBD_MAJOR;
588 gd->first_minor = minor;
589 gd->fops = &xlvbd_block_fops;
590 gd->private_data = info;
591 gd->driverfs_dev = &(info->xbdev->dev);
592 set_capacity(gd, capacity);
593
594 if (xlvbd_init_blk_queue(gd, sector_size)) {
595 del_gendisk(gd);
596 goto release;
597 }
598
599 info->rq = gd->queue;
600 info->gd = gd;
601
602 xlvbd_flush(info);
603
604 if (vdisk_info & VDISK_READONLY)
605 set_disk_ro(gd, 1);
606
607 if (vdisk_info & VDISK_REMOVABLE)
608 gd->flags |= GENHD_FL_REMOVABLE;
609
610 if (vdisk_info & VDISK_CDROM)
611 gd->flags |= GENHD_FL_CD;
612
613 return 0;
614
615 release:
616 xlbd_release_minors(minor, nr_minors);
617 out:
618 return err;
619 }
620
621 static void xlvbd_release_gendisk(struct blkfront_info *info)
622 {
623 unsigned int minor, nr_minors;
624 unsigned long flags;
625
626 if (info->rq == NULL)
627 return;
628
629 spin_lock_irqsave(&blkif_io_lock, flags);
630
631 /* No more blkif_request(). */
632 blk_stop_queue(info->rq);
633
634 /* No more gnttab callback work. */
635 gnttab_cancel_free_callback(&info->callback);
636 spin_unlock_irqrestore(&blkif_io_lock, flags);
637
638 /* Flush gnttab callback work. Must be done with no locks held. */
639 flush_work_sync(&info->work);
640
641 del_gendisk(info->gd);
642
643 minor = info->gd->first_minor;
644 nr_minors = info->gd->minors;
645 xlbd_release_minors(minor, nr_minors);
646
647 blk_cleanup_queue(info->rq);
648 info->rq = NULL;
649
650 put_disk(info->gd);
651 info->gd = NULL;
652 }
653
654 static void kick_pending_request_queues(struct blkfront_info *info)
655 {
656 if (!RING_FULL(&info->ring)) {
657 /* Re-enable calldowns. */
658 blk_start_queue(info->rq);
659 /* Kick things off immediately. */
660 do_blkif_request(info->rq);
661 }
662 }
663
664 static void blkif_restart_queue(struct work_struct *work)
665 {
666 struct blkfront_info *info = container_of(work, struct blkfront_info, work);
667
668 spin_lock_irq(&blkif_io_lock);
669 if (info->connected == BLKIF_STATE_CONNECTED)
670 kick_pending_request_queues(info);
671 spin_unlock_irq(&blkif_io_lock);
672 }
673
674 static void blkif_free(struct blkfront_info *info, int suspend)
675 {
676 /* Prevent new requests being issued until we fix things up. */
677 spin_lock_irq(&blkif_io_lock);
678 info->connected = suspend ?
679 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
680 /* No more blkif_request(). */
681 if (info->rq)
682 blk_stop_queue(info->rq);
683 /* No more gnttab callback work. */
684 gnttab_cancel_free_callback(&info->callback);
685 spin_unlock_irq(&blkif_io_lock);
686
687 /* Flush gnttab callback work. Must be done with no locks held. */
688 flush_work_sync(&info->work);
689
690 /* Free resources associated with old device channel. */
691 if (info->ring_ref != GRANT_INVALID_REF) {
692 gnttab_end_foreign_access(info->ring_ref, 0,
693 (unsigned long)info->ring.sring);
694 info->ring_ref = GRANT_INVALID_REF;
695 info->ring.sring = NULL;
696 }
697 if (info->irq)
698 unbind_from_irqhandler(info->irq, info);
699 info->evtchn = info->irq = 0;
700
701 }
702
703 static void blkif_completion(struct blk_shadow *s)
704 {
705 int i;
706 for (i = 0; i < s->req.nr_segments; i++)
707 gnttab_end_foreign_access(s->req.u.rw.seg[i].gref, 0, 0UL);
708 }
709
710 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
711 {
712 struct request *req;
713 struct blkif_response *bret;
714 RING_IDX i, rp;
715 unsigned long flags;
716 struct blkfront_info *info = (struct blkfront_info *)dev_id;
717 int error;
718
719 spin_lock_irqsave(&blkif_io_lock, flags);
720
721 if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
722 spin_unlock_irqrestore(&blkif_io_lock, flags);
723 return IRQ_HANDLED;
724 }
725
726 again:
727 rp = info->ring.sring->rsp_prod;
728 rmb(); /* Ensure we see queued responses up to 'rp'. */
729
730 for (i = info->ring.rsp_cons; i != rp; i++) {
731 unsigned long id;
732
733 bret = RING_GET_RESPONSE(&info->ring, i);
734 id = bret->id;
735 req = info->shadow[id].request;
736
737 blkif_completion(&info->shadow[id]);
738
739 add_id_to_freelist(info, id);
740
741 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
742 switch (bret->operation) {
743 case BLKIF_OP_DISCARD:
744 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
745 struct request_queue *rq = info->rq;
746 printk(KERN_WARNING "blkfront: %s: discard op failed\n",
747 info->gd->disk_name);
748 error = -EOPNOTSUPP;
749 info->feature_discard = 0;
750 queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
751 }
752 __blk_end_request_all(req, error);
753 break;
754 case BLKIF_OP_FLUSH_DISKCACHE:
755 case BLKIF_OP_WRITE_BARRIER:
756 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
757 printk(KERN_WARNING "blkfront: %s: write %s op failed\n",
758 info->flush_op == BLKIF_OP_WRITE_BARRIER ?
759 "barrier" : "flush disk cache",
760 info->gd->disk_name);
761 error = -EOPNOTSUPP;
762 }
763 if (unlikely(bret->status == BLKIF_RSP_ERROR &&
764 info->shadow[id].req.nr_segments == 0)) {
765 printk(KERN_WARNING "blkfront: %s: empty write %s op failed\n",
766 info->flush_op == BLKIF_OP_WRITE_BARRIER ?
767 "barrier" : "flush disk cache",
768 info->gd->disk_name);
769 error = -EOPNOTSUPP;
770 }
771 if (unlikely(error)) {
772 if (error == -EOPNOTSUPP)
773 error = 0;
774 info->feature_flush = 0;
775 info->flush_op = 0;
776 xlvbd_flush(info);
777 }
778 /* fall through */
779 case BLKIF_OP_READ:
780 case BLKIF_OP_WRITE:
781 if (unlikely(bret->status != BLKIF_RSP_OKAY))
782 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
783 "request: %x\n", bret->status);
784
785 __blk_end_request_all(req, error);
786 break;
787 default:
788 BUG();
789 }
790 }
791
792 info->ring.rsp_cons = i;
793
794 if (i != info->ring.req_prod_pvt) {
795 int more_to_do;
796 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
797 if (more_to_do)
798 goto again;
799 } else
800 info->ring.sring->rsp_event = i + 1;
801
802 kick_pending_request_queues(info);
803
804 spin_unlock_irqrestore(&blkif_io_lock, flags);
805
806 return IRQ_HANDLED;
807 }
808
809
810 static int setup_blkring(struct xenbus_device *dev,
811 struct blkfront_info *info)
812 {
813 struct blkif_sring *sring;
814 int err;
815
816 info->ring_ref = GRANT_INVALID_REF;
817
818 sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
819 if (!sring) {
820 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
821 return -ENOMEM;
822 }
823 SHARED_RING_INIT(sring);
824 FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
825
826 sg_init_table(info->sg, BLKIF_MAX_SEGMENTS_PER_REQUEST);
827
828 err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
829 if (err < 0) {
830 free_page((unsigned long)sring);
831 info->ring.sring = NULL;
832 goto fail;
833 }
834 info->ring_ref = err;
835
836 err = xenbus_alloc_evtchn(dev, &info->evtchn);
837 if (err)
838 goto fail;
839
840 err = bind_evtchn_to_irqhandler(info->evtchn,
841 blkif_interrupt,
842 IRQF_SAMPLE_RANDOM, "blkif", info);
843 if (err <= 0) {
844 xenbus_dev_fatal(dev, err,
845 "bind_evtchn_to_irqhandler failed");
846 goto fail;
847 }
848 info->irq = err;
849
850 return 0;
851 fail:
852 blkif_free(info, 0);
853 return err;
854 }
855
856
857 /* Common code used when first setting up, and when resuming. */
858 static int talk_to_blkback(struct xenbus_device *dev,
859 struct blkfront_info *info)
860 {
861 const char *message = NULL;
862 struct xenbus_transaction xbt;
863 int err;
864
865 /* Create shared ring, alloc event channel. */
866 err = setup_blkring(dev, info);
867 if (err)
868 goto out;
869
870 again:
871 err = xenbus_transaction_start(&xbt);
872 if (err) {
873 xenbus_dev_fatal(dev, err, "starting transaction");
874 goto destroy_blkring;
875 }
876
877 err = xenbus_printf(xbt, dev->nodename,
878 "ring-ref", "%u", info->ring_ref);
879 if (err) {
880 message = "writing ring-ref";
881 goto abort_transaction;
882 }
883 err = xenbus_printf(xbt, dev->nodename,
884 "event-channel", "%u", info->evtchn);
885 if (err) {
886 message = "writing event-channel";
887 goto abort_transaction;
888 }
889 err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
890 XEN_IO_PROTO_ABI_NATIVE);
891 if (err) {
892 message = "writing protocol";
893 goto abort_transaction;
894 }
895
896 err = xenbus_transaction_end(xbt, 0);
897 if (err) {
898 if (err == -EAGAIN)
899 goto again;
900 xenbus_dev_fatal(dev, err, "completing transaction");
901 goto destroy_blkring;
902 }
903
904 xenbus_switch_state(dev, XenbusStateInitialised);
905
906 return 0;
907
908 abort_transaction:
909 xenbus_transaction_end(xbt, 1);
910 if (message)
911 xenbus_dev_fatal(dev, err, "%s", message);
912 destroy_blkring:
913 blkif_free(info, 0);
914 out:
915 return err;
916 }
917
918 /**
919 * Entry point to this code when a new device is created. Allocate the basic
920 * structures and the ring buffer for communication with the backend, and
921 * inform the backend of the appropriate details for those. Switch to
922 * Initialised state.
923 */
924 static int blkfront_probe(struct xenbus_device *dev,
925 const struct xenbus_device_id *id)
926 {
927 int err, vdevice, i;
928 struct blkfront_info *info;
929
930 /* FIXME: Use dynamic device id if this is not set. */
931 err = xenbus_scanf(XBT_NIL, dev->nodename,
932 "virtual-device", "%i", &vdevice);
933 if (err != 1) {
934 /* go looking in the extended area instead */
935 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
936 "%i", &vdevice);
937 if (err != 1) {
938 xenbus_dev_fatal(dev, err, "reading virtual-device");
939 return err;
940 }
941 }
942
943 if (xen_hvm_domain()) {
944 char *type;
945 int len;
946 /* no unplug has been done: do not hook devices != xen vbds */
947 if (xen_platform_pci_unplug & XEN_UNPLUG_UNNECESSARY) {
948 int major;
949
950 if (!VDEV_IS_EXTENDED(vdevice))
951 major = BLKIF_MAJOR(vdevice);
952 else
953 major = XENVBD_MAJOR;
954
955 if (major != XENVBD_MAJOR) {
956 printk(KERN_INFO
957 "%s: HVM does not support vbd %d as xen block device\n",
958 __FUNCTION__, vdevice);
959 return -ENODEV;
960 }
961 }
962 /* do not create a PV cdrom device if we are an HVM guest */
963 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
964 if (IS_ERR(type))
965 return -ENODEV;
966 if (strncmp(type, "cdrom", 5) == 0) {
967 kfree(type);
968 return -ENODEV;
969 }
970 kfree(type);
971 }
972 info = kzalloc(sizeof(*info), GFP_KERNEL);
973 if (!info) {
974 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
975 return -ENOMEM;
976 }
977
978 mutex_init(&info->mutex);
979 info->xbdev = dev;
980 info->vdevice = vdevice;
981 info->connected = BLKIF_STATE_DISCONNECTED;
982 INIT_WORK(&info->work, blkif_restart_queue);
983
984 for (i = 0; i < BLK_RING_SIZE; i++)
985 info->shadow[i].req.id = i+1;
986 info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
987
988 /* Front end dir is a number, which is used as the id. */
989 info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
990 dev_set_drvdata(&dev->dev, info);
991
992 err = talk_to_blkback(dev, info);
993 if (err) {
994 kfree(info);
995 dev_set_drvdata(&dev->dev, NULL);
996 return err;
997 }
998
999 return 0;
1000 }
1001
1002
1003 static int blkif_recover(struct blkfront_info *info)
1004 {
1005 int i;
1006 struct blkif_request *req;
1007 struct blk_shadow *copy;
1008 int j;
1009
1010 /* Stage 1: Make a safe copy of the shadow state. */
1011 copy = kmalloc(sizeof(info->shadow),
1012 GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
1013 if (!copy)
1014 return -ENOMEM;
1015 memcpy(copy, info->shadow, sizeof(info->shadow));
1016
1017 /* Stage 2: Set up free list. */
1018 memset(&info->shadow, 0, sizeof(info->shadow));
1019 for (i = 0; i < BLK_RING_SIZE; i++)
1020 info->shadow[i].req.id = i+1;
1021 info->shadow_free = info->ring.req_prod_pvt;
1022 info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
1023
1024 /* Stage 3: Find pending requests and requeue them. */
1025 for (i = 0; i < BLK_RING_SIZE; i++) {
1026 /* Not in use? */
1027 if (!copy[i].request)
1028 continue;
1029
1030 /* Grab a request slot and copy shadow state into it. */
1031 req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
1032 *req = copy[i].req;
1033
1034 /* We get a new request id, and must reset the shadow state. */
1035 req->id = get_id_from_freelist(info);
1036 memcpy(&info->shadow[req->id], &copy[i], sizeof(copy[i]));
1037
1038 /* Rewrite any grant references invalidated by susp/resume. */
1039 for (j = 0; j < req->nr_segments; j++)
1040 gnttab_grant_foreign_access_ref(
1041 req->u.rw.seg[j].gref,
1042 info->xbdev->otherend_id,
1043 pfn_to_mfn(info->shadow[req->id].frame[j]),
1044 rq_data_dir(info->shadow[req->id].request));
1045 info->shadow[req->id].req = *req;
1046
1047 info->ring.req_prod_pvt++;
1048 }
1049
1050 kfree(copy);
1051
1052 xenbus_switch_state(info->xbdev, XenbusStateConnected);
1053
1054 spin_lock_irq(&blkif_io_lock);
1055
1056 /* Now safe for us to use the shared ring */
1057 info->connected = BLKIF_STATE_CONNECTED;
1058
1059 /* Send off requeued requests */
1060 flush_requests(info);
1061
1062 /* Kick any other new requests queued since we resumed */
1063 kick_pending_request_queues(info);
1064
1065 spin_unlock_irq(&blkif_io_lock);
1066
1067 return 0;
1068 }
1069
1070 /**
1071 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1072 * driver restart. We tear down our blkif structure and recreate it, but
1073 * leave the device-layer structures intact so that this is transparent to the
1074 * rest of the kernel.
1075 */
1076 static int blkfront_resume(struct xenbus_device *dev)
1077 {
1078 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1079 int err;
1080
1081 dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
1082
1083 blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
1084
1085 err = talk_to_blkback(dev, info);
1086 if (info->connected == BLKIF_STATE_SUSPENDED && !err)
1087 err = blkif_recover(info);
1088
1089 return err;
1090 }
1091
1092 static void
1093 blkfront_closing(struct blkfront_info *info)
1094 {
1095 struct xenbus_device *xbdev = info->xbdev;
1096 struct block_device *bdev = NULL;
1097
1098 mutex_lock(&info->mutex);
1099
1100 if (xbdev->state == XenbusStateClosing) {
1101 mutex_unlock(&info->mutex);
1102 return;
1103 }
1104
1105 if (info->gd)
1106 bdev = bdget_disk(info->gd, 0);
1107
1108 mutex_unlock(&info->mutex);
1109
1110 if (!bdev) {
1111 xenbus_frontend_closed(xbdev);
1112 return;
1113 }
1114
1115 mutex_lock(&bdev->bd_mutex);
1116
1117 if (bdev->bd_openers) {
1118 xenbus_dev_error(xbdev, -EBUSY,
1119 "Device in use; refusing to close");
1120 xenbus_switch_state(xbdev, XenbusStateClosing);
1121 } else {
1122 xlvbd_release_gendisk(info);
1123 xenbus_frontend_closed(xbdev);
1124 }
1125
1126 mutex_unlock(&bdev->bd_mutex);
1127 bdput(bdev);
1128 }
1129
1130 static void blkfront_setup_discard(struct blkfront_info *info)
1131 {
1132 int err;
1133 char *type;
1134 unsigned int discard_granularity;
1135 unsigned int discard_alignment;
1136
1137 type = xenbus_read(XBT_NIL, info->xbdev->otherend, "type", NULL);
1138 if (IS_ERR(type))
1139 return;
1140
1141 if (strncmp(type, "phy", 3) == 0) {
1142 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1143 "discard-granularity", "%u", &discard_granularity,
1144 "discard-alignment", "%u", &discard_alignment,
1145 NULL);
1146 if (!err) {
1147 info->feature_discard = 1;
1148 info->discard_granularity = discard_granularity;
1149 info->discard_alignment = discard_alignment;
1150 }
1151 } else if (strncmp(type, "file", 4) == 0)
1152 info->feature_discard = 1;
1153
1154 kfree(type);
1155 }
1156
1157 /*
1158 * Invoked when the backend is finally 'ready' (and has told produced
1159 * the details about the physical device - #sectors, size, etc).
1160 */
1161 static void blkfront_connect(struct blkfront_info *info)
1162 {
1163 unsigned long long sectors;
1164 unsigned long sector_size;
1165 unsigned int binfo;
1166 int err;
1167 int barrier, flush, discard;
1168
1169 switch (info->connected) {
1170 case BLKIF_STATE_CONNECTED:
1171 /*
1172 * Potentially, the back-end may be signalling
1173 * a capacity change; update the capacity.
1174 */
1175 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1176 "sectors", "%Lu", &sectors);
1177 if (XENBUS_EXIST_ERR(err))
1178 return;
1179 printk(KERN_INFO "Setting capacity to %Lu\n",
1180 sectors);
1181 set_capacity(info->gd, sectors);
1182 revalidate_disk(info->gd);
1183
1184 /* fall through */
1185 case BLKIF_STATE_SUSPENDED:
1186 return;
1187
1188 default:
1189 break;
1190 }
1191
1192 dev_dbg(&info->xbdev->dev, "%s:%s.\n",
1193 __func__, info->xbdev->otherend);
1194
1195 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1196 "sectors", "%llu", &sectors,
1197 "info", "%u", &binfo,
1198 "sector-size", "%lu", &sector_size,
1199 NULL);
1200 if (err) {
1201 xenbus_dev_fatal(info->xbdev, err,
1202 "reading backend fields at %s",
1203 info->xbdev->otherend);
1204 return;
1205 }
1206
1207 info->feature_flush = 0;
1208 info->flush_op = 0;
1209
1210 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1211 "feature-barrier", "%d", &barrier,
1212 NULL);
1213
1214 /*
1215 * If there's no "feature-barrier" defined, then it means
1216 * we're dealing with a very old backend which writes
1217 * synchronously; nothing to do.
1218 *
1219 * If there are barriers, then we use flush.
1220 */
1221 if (!err && barrier) {
1222 info->feature_flush = REQ_FLUSH | REQ_FUA;
1223 info->flush_op = BLKIF_OP_WRITE_BARRIER;
1224 }
1225 /*
1226 * And if there is "feature-flush-cache" use that above
1227 * barriers.
1228 */
1229 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1230 "feature-flush-cache", "%d", &flush,
1231 NULL);
1232
1233 if (!err && flush) {
1234 info->feature_flush = REQ_FLUSH;
1235 info->flush_op = BLKIF_OP_FLUSH_DISKCACHE;
1236 }
1237
1238 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1239 "feature-discard", "%d", &discard,
1240 NULL);
1241
1242 if (!err && discard)
1243 blkfront_setup_discard(info);
1244
1245 err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size);
1246 if (err) {
1247 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
1248 info->xbdev->otherend);
1249 return;
1250 }
1251
1252 xenbus_switch_state(info->xbdev, XenbusStateConnected);
1253
1254 /* Kick pending requests. */
1255 spin_lock_irq(&blkif_io_lock);
1256 info->connected = BLKIF_STATE_CONNECTED;
1257 kick_pending_request_queues(info);
1258 spin_unlock_irq(&blkif_io_lock);
1259
1260 add_disk(info->gd);
1261
1262 info->is_ready = 1;
1263 }
1264
1265 /**
1266 * Callback received when the backend's state changes.
1267 */
1268 static void blkback_changed(struct xenbus_device *dev,
1269 enum xenbus_state backend_state)
1270 {
1271 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1272
1273 dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1274
1275 switch (backend_state) {
1276 case XenbusStateInitialising:
1277 case XenbusStateInitWait:
1278 case XenbusStateInitialised:
1279 case XenbusStateReconfiguring:
1280 case XenbusStateReconfigured:
1281 case XenbusStateUnknown:
1282 case XenbusStateClosed:
1283 break;
1284
1285 case XenbusStateConnected:
1286 blkfront_connect(info);
1287 break;
1288
1289 case XenbusStateClosing:
1290 blkfront_closing(info);
1291 break;
1292 }
1293 }
1294
1295 static int blkfront_remove(struct xenbus_device *xbdev)
1296 {
1297 struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
1298 struct block_device *bdev = NULL;
1299 struct gendisk *disk;
1300
1301 dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
1302
1303 blkif_free(info, 0);
1304
1305 mutex_lock(&info->mutex);
1306
1307 disk = info->gd;
1308 if (disk)
1309 bdev = bdget_disk(disk, 0);
1310
1311 info->xbdev = NULL;
1312 mutex_unlock(&info->mutex);
1313
1314 if (!bdev) {
1315 kfree(info);
1316 return 0;
1317 }
1318
1319 /*
1320 * The xbdev was removed before we reached the Closed
1321 * state. See if it's safe to remove the disk. If the bdev
1322 * isn't closed yet, we let release take care of it.
1323 */
1324
1325 mutex_lock(&bdev->bd_mutex);
1326 info = disk->private_data;
1327
1328 dev_warn(disk_to_dev(disk),
1329 "%s was hot-unplugged, %d stale handles\n",
1330 xbdev->nodename, bdev->bd_openers);
1331
1332 if (info && !bdev->bd_openers) {
1333 xlvbd_release_gendisk(info);
1334 disk->private_data = NULL;
1335 kfree(info);
1336 }
1337
1338 mutex_unlock(&bdev->bd_mutex);
1339 bdput(bdev);
1340
1341 return 0;
1342 }
1343
1344 static int blkfront_is_ready(struct xenbus_device *dev)
1345 {
1346 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1347
1348 return info->is_ready && info->xbdev;
1349 }
1350
1351 static int blkif_open(struct block_device *bdev, fmode_t mode)
1352 {
1353 struct gendisk *disk = bdev->bd_disk;
1354 struct blkfront_info *info;
1355 int err = 0;
1356
1357 mutex_lock(&blkfront_mutex);
1358
1359 info = disk->private_data;
1360 if (!info) {
1361 /* xbdev gone */
1362 err = -ERESTARTSYS;
1363 goto out;
1364 }
1365
1366 mutex_lock(&info->mutex);
1367
1368 if (!info->gd)
1369 /* xbdev is closed */
1370 err = -ERESTARTSYS;
1371
1372 mutex_unlock(&info->mutex);
1373
1374 out:
1375 mutex_unlock(&blkfront_mutex);
1376 return err;
1377 }
1378
1379 static int blkif_release(struct gendisk *disk, fmode_t mode)
1380 {
1381 struct blkfront_info *info = disk->private_data;
1382 struct block_device *bdev;
1383 struct xenbus_device *xbdev;
1384
1385 mutex_lock(&blkfront_mutex);
1386
1387 bdev = bdget_disk(disk, 0);
1388 bdput(bdev);
1389
1390 if (bdev->bd_openers)
1391 goto out;
1392
1393 /*
1394 * Check if we have been instructed to close. We will have
1395 * deferred this request, because the bdev was still open.
1396 */
1397
1398 mutex_lock(&info->mutex);
1399 xbdev = info->xbdev;
1400
1401 if (xbdev && xbdev->state == XenbusStateClosing) {
1402 /* pending switch to state closed */
1403 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1404 xlvbd_release_gendisk(info);
1405 xenbus_frontend_closed(info->xbdev);
1406 }
1407
1408 mutex_unlock(&info->mutex);
1409
1410 if (!xbdev) {
1411 /* sudden device removal */
1412 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1413 xlvbd_release_gendisk(info);
1414 disk->private_data = NULL;
1415 kfree(info);
1416 }
1417
1418 out:
1419 mutex_unlock(&blkfront_mutex);
1420 return 0;
1421 }
1422
1423 static const struct block_device_operations xlvbd_block_fops =
1424 {
1425 .owner = THIS_MODULE,
1426 .open = blkif_open,
1427 .release = blkif_release,
1428 .getgeo = blkif_getgeo,
1429 .ioctl = blkif_ioctl,
1430 };
1431
1432
1433 static const struct xenbus_device_id blkfront_ids[] = {
1434 { "vbd" },
1435 { "" }
1436 };
1437
1438 static struct xenbus_driver blkfront = {
1439 .name = "vbd",
1440 .owner = THIS_MODULE,
1441 .ids = blkfront_ids,
1442 .probe = blkfront_probe,
1443 .remove = blkfront_remove,
1444 .resume = blkfront_resume,
1445 .otherend_changed = blkback_changed,
1446 .is_ready = blkfront_is_ready,
1447 };
1448
1449 static int __init xlblk_init(void)
1450 {
1451 if (!xen_domain())
1452 return -ENODEV;
1453
1454 if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
1455 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
1456 XENVBD_MAJOR, DEV_NAME);
1457 return -ENODEV;
1458 }
1459
1460 return xenbus_register_frontend(&blkfront);
1461 }
1462 module_init(xlblk_init);
1463
1464
1465 static void __exit xlblk_exit(void)
1466 {
1467 return xenbus_unregister_driver(&blkfront);
1468 }
1469 module_exit(xlblk_exit);
1470
1471 MODULE_DESCRIPTION("Xen virtual block device frontend");
1472 MODULE_LICENSE("GPL");
1473 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
1474 MODULE_ALIAS("xen:vbd");
1475 MODULE_ALIAS("xenblk");
Linus' kernel tree