?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
mfd: wm8350-i2c: Make sure the i2c regmap functions are compiled
[linux/fpc-iii.git] / drivers / block / xen-blkfront.c
blob0b6932c376fbf35a0d6943ca9fa97602c60b9d71
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 #include <linux/bitmap.h>
47 #include <linux/list.h>
48
49 #include <xen/xen.h>
50 #include <xen/xenbus.h>
51 #include <xen/grant_table.h>
52 #include <xen/events.h>
53 #include <xen/page.h>
54 #include <xen/platform_pci.h>
55
56 #include <xen/interface/grant_table.h>
57 #include <xen/interface/io/blkif.h>
58 #include <xen/interface/io/protocols.h>
59
60 #include <asm/xen/hypervisor.h>
61
62 enum blkif_state {
63 BLKIF_STATE_DISCONNECTED,
64 BLKIF_STATE_CONNECTED,
65 BLKIF_STATE_SUSPENDED,
66 };
67
68 struct grant {
69 grant_ref_t gref;
70 unsigned long pfn;
71 struct list_head node;
72 };
73
74 struct blk_shadow {
75 struct blkif_request req;
76 struct request *request;
77 struct grant **grants_used;
78 struct grant **indirect_grants;
79 struct scatterlist *sg;
80 };
81
82 struct split_bio {
83 struct bio *bio;
84 atomic_t pending;
85 int err;
86 };
87
88 static DEFINE_MUTEX(blkfront_mutex);
89 static const struct block_device_operations xlvbd_block_fops;
90
91 /*
92 * Maximum number of segments in indirect requests, the actual value used by
93 * the frontend driver is the minimum of this value and the value provided
94 * by the backend driver.
95 */
96
97 static unsigned int xen_blkif_max_segments = 32;
98 module_param_named(max, xen_blkif_max_segments, int, S_IRUGO);
99 MODULE_PARM_DESC(max, "Maximum amount of segments in indirect requests (default is 32)");
100
101 #define BLK_RING_SIZE __CONST_RING_SIZE(blkif, PAGE_SIZE)
102
103 /*
104 * We have one of these per vbd, whether ide, scsi or 'other'. They
105 * hang in private_data off the gendisk structure. We may end up
106 * putting all kinds of interesting stuff here :-)
107 */
108 struct blkfront_info
109 {
110 spinlock_t io_lock;
111 struct mutex mutex;
112 struct xenbus_device *xbdev;
113 struct gendisk *gd;
114 int vdevice;
115 blkif_vdev_t handle;
116 enum blkif_state connected;
117 int ring_ref;
118 struct blkif_front_ring ring;
119 unsigned int evtchn, irq;
120 struct request_queue *rq;
121 struct work_struct work;
122 struct gnttab_free_callback callback;
123 struct blk_shadow shadow[BLK_RING_SIZE];
124 struct list_head grants;
125 struct list_head indirect_pages;
126 unsigned int persistent_gnts_c;
127 unsigned long shadow_free;
128 unsigned int feature_flush;
129 unsigned int flush_op;
130 unsigned int feature_discard:1;
131 unsigned int feature_secdiscard:1;
132 unsigned int discard_granularity;
133 unsigned int discard_alignment;
134 unsigned int feature_persistent:1;
135 unsigned int max_indirect_segments;
136 int is_ready;
137 };
138
139 static unsigned int nr_minors;
140 static unsigned long *minors;
141 static DEFINE_SPINLOCK(minor_lock);
142
143 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
144 (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
145 #define GRANT_INVALID_REF 0
146
147 #define PARTS_PER_DISK 16
148 #define PARTS_PER_EXT_DISK 256
149
150 #define BLKIF_MAJOR(dev) ((dev)>>8)
151 #define BLKIF_MINOR(dev) ((dev) & 0xff)
152
153 #define EXT_SHIFT 28
154 #define EXTENDED (1<<EXT_SHIFT)
155 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
156 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
157 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
158 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
159 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
160 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
161
162 #define DEV_NAME "xvd" /* name in /dev */
163
164 #define SEGS_PER_INDIRECT_FRAME \
165 (PAGE_SIZE/sizeof(struct blkif_request_segment_aligned))
166 #define INDIRECT_GREFS(_segs) \
167 ((_segs + SEGS_PER_INDIRECT_FRAME - 1)/SEGS_PER_INDIRECT_FRAME)
168
169 static int blkfront_setup_indirect(struct blkfront_info *info);
170
171 static int get_id_from_freelist(struct blkfront_info *info)
172 {
173 unsigned long free = info->shadow_free;
174 BUG_ON(free >= BLK_RING_SIZE);
175 info->shadow_free = info->shadow[free].req.u.rw.id;
176 info->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
177 return free;
178 }
179
180 static int add_id_to_freelist(struct blkfront_info *info,
181 unsigned long id)
182 {
183 if (info->shadow[id].req.u.rw.id != id)
184 return -EINVAL;
185 if (info->shadow[id].request == NULL)
186 return -EINVAL;
187 info->shadow[id].req.u.rw.id = info->shadow_free;
188 info->shadow[id].request = NULL;
189 info->shadow_free = id;
190 return 0;
191 }
192
193 static int fill_grant_buffer(struct blkfront_info *info, int num)
194 {
195 struct page *granted_page;
196 struct grant *gnt_list_entry, *n;
197 int i = 0;
198
199 while(i < num) {
200 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
201 if (!gnt_list_entry)
202 goto out_of_memory;
203
204 if (info->feature_persistent) {
205 granted_page = alloc_page(GFP_NOIO);
206 if (!granted_page) {
207 kfree(gnt_list_entry);
208 goto out_of_memory;
209 }
210 gnt_list_entry->pfn = page_to_pfn(granted_page);
211 }
212
213 gnt_list_entry->gref = GRANT_INVALID_REF;
214 list_add(&gnt_list_entry->node, &info->grants);
215 i++;
216 }
217
218 return 0;
219
220 out_of_memory:
221 list_for_each_entry_safe(gnt_list_entry, n,
222 &info->grants, node) {
223 list_del(&gnt_list_entry->node);
224 if (info->feature_persistent)
225 __free_page(pfn_to_page(gnt_list_entry->pfn));
226 kfree(gnt_list_entry);
227 i--;
228 }
229 BUG_ON(i != 0);
230 return -ENOMEM;
231 }
232
233 static struct grant *get_grant(grant_ref_t *gref_head,
234 unsigned long pfn,
235 struct blkfront_info *info)
236 {
237 struct grant *gnt_list_entry;
238 unsigned long buffer_mfn;
239
240 BUG_ON(list_empty(&info->grants));
241 gnt_list_entry = list_first_entry(&info->grants, struct grant,
242 node);
243 list_del(&gnt_list_entry->node);
244
245 if (gnt_list_entry->gref != GRANT_INVALID_REF) {
246 info->persistent_gnts_c--;
247 return gnt_list_entry;
248 }
249
250 /* Assign a gref to this page */
251 gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
252 BUG_ON(gnt_list_entry->gref == -ENOSPC);
253 if (!info->feature_persistent) {
254 BUG_ON(!pfn);
255 gnt_list_entry->pfn = pfn;
256 }
257 buffer_mfn = pfn_to_mfn(gnt_list_entry->pfn);
258 gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
259 info->xbdev->otherend_id,
260 buffer_mfn, 0);
261 return gnt_list_entry;
262 }
263
264 static const char *op_name(int op)
265 {
266 static const char *const names[] = {
267 [BLKIF_OP_READ] = "read",
268 [BLKIF_OP_WRITE] = "write",
269 [BLKIF_OP_WRITE_BARRIER] = "barrier",
270 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
271 [BLKIF_OP_DISCARD] = "discard" };
272
273 if (op < 0 || op >= ARRAY_SIZE(names))
274 return "unknown";
275
276 if (!names[op])
277 return "reserved";
278
279 return names[op];
280 }
281 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
282 {
283 unsigned int end = minor + nr;
284 int rc;
285
286 if (end > nr_minors) {
287 unsigned long *bitmap, *old;
288
289 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
290 GFP_KERNEL);
291 if (bitmap == NULL)
292 return -ENOMEM;
293
294 spin_lock(&minor_lock);
295 if (end > nr_minors) {
296 old = minors;
297 memcpy(bitmap, minors,
298 BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
299 minors = bitmap;
300 nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
301 } else
302 old = bitmap;
303 spin_unlock(&minor_lock);
304 kfree(old);
305 }
306
307 spin_lock(&minor_lock);
308 if (find_next_bit(minors, end, minor) >= end) {
309 bitmap_set(minors, minor, nr);
310 rc = 0;
311 } else
312 rc = -EBUSY;
313 spin_unlock(&minor_lock);
314
315 return rc;
316 }
317
318 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
319 {
320 unsigned int end = minor + nr;
321
322 BUG_ON(end > nr_minors);
323 spin_lock(&minor_lock);
324 bitmap_clear(minors, minor, nr);
325 spin_unlock(&minor_lock);
326 }
327
328 static void blkif_restart_queue_callback(void *arg)
329 {
330 struct blkfront_info *info = (struct blkfront_info *)arg;
331 schedule_work(&info->work);
332 }
333
334 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
335 {
336 /* We don't have real geometry info, but let's at least return
337 values consistent with the size of the device */
338 sector_t nsect = get_capacity(bd->bd_disk);
339 sector_t cylinders = nsect;
340
341 hg->heads = 0xff;
342 hg->sectors = 0x3f;
343 sector_div(cylinders, hg->heads * hg->sectors);
344 hg->cylinders = cylinders;
345 if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
346 hg->cylinders = 0xffff;
347 return 0;
348 }
349
350 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
351 unsigned command, unsigned long argument)
352 {
353 struct blkfront_info *info = bdev->bd_disk->private_data;
354 int i;
355
356 dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
357 command, (long)argument);
358
359 switch (command) {
360 case CDROMMULTISESSION:
361 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
362 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
363 if (put_user(0, (char __user *)(argument + i)))
364 return -EFAULT;
365 return 0;
366
367 case CDROM_GET_CAPABILITY: {
368 struct gendisk *gd = info->gd;
369 if (gd->flags & GENHD_FL_CD)
370 return 0;
371 return -EINVAL;
372 }
373
374 default:
375 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
376 command);*/
377 return -EINVAL; /* same return as native Linux */
378 }
379
380 return 0;
381 }
382
383 /*
384 * Generate a Xen blkfront IO request from a blk layer request. Reads
385 * and writes are handled as expected.
386 *
387 * @req: a request struct
388 */
389 static int blkif_queue_request(struct request *req)
390 {
391 struct blkfront_info *info = req->rq_disk->private_data;
392 struct blkif_request *ring_req;
393 unsigned long id;
394 unsigned int fsect, lsect;
395 int i, ref, n;
396 struct blkif_request_segment_aligned *segments = NULL;
397
398 /*
399 * Used to store if we are able to queue the request by just using
400 * existing persistent grants, or if we have to get new grants,
401 * as there are not sufficiently many free.
402 */
403 bool new_persistent_gnts;
404 grant_ref_t gref_head;
405 struct grant *gnt_list_entry = NULL;
406 struct scatterlist *sg;
407 int nseg, max_grefs;
408
409 if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
410 return 1;
411
412 max_grefs = info->max_indirect_segments ?
413 info->max_indirect_segments +
414 INDIRECT_GREFS(info->max_indirect_segments) :
415 BLKIF_MAX_SEGMENTS_PER_REQUEST;
416
417 /* Check if we have enough grants to allocate a requests */
418 if (info->persistent_gnts_c < max_grefs) {
419 new_persistent_gnts = 1;
420 if (gnttab_alloc_grant_references(
421 max_grefs - info->persistent_gnts_c,
422 &gref_head) < 0) {
423 gnttab_request_free_callback(
424 &info->callback,
425 blkif_restart_queue_callback,
426 info,
427 max_grefs);
428 return 1;
429 }
430 } else
431 new_persistent_gnts = 0;
432
433 /* Fill out a communications ring structure. */
434 ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
435 id = get_id_from_freelist(info);
436 info->shadow[id].request = req;
437
438 if (unlikely(req->cmd_flags & (REQ_DISCARD | REQ_SECURE))) {
439 ring_req->operation = BLKIF_OP_DISCARD;
440 ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
441 ring_req->u.discard.id = id;
442 ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
443 if ((req->cmd_flags & REQ_SECURE) && info->feature_secdiscard)
444 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
445 else
446 ring_req->u.discard.flag = 0;
447 } else {
448 BUG_ON(info->max_indirect_segments == 0 &&
449 req->nr_phys_segments > BLKIF_MAX_SEGMENTS_PER_REQUEST);
450 BUG_ON(info->max_indirect_segments &&
451 req->nr_phys_segments > info->max_indirect_segments);
452 nseg = blk_rq_map_sg(req->q, req, info->shadow[id].sg);
453 ring_req->u.rw.id = id;
454 if (nseg > BLKIF_MAX_SEGMENTS_PER_REQUEST) {
455 /*
456 * The indirect operation can only be a BLKIF_OP_READ or
457 * BLKIF_OP_WRITE
458 */
459 BUG_ON(req->cmd_flags & (REQ_FLUSH | REQ_FUA));
460 ring_req->operation = BLKIF_OP_INDIRECT;
461 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
462 BLKIF_OP_WRITE : BLKIF_OP_READ;
463 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
464 ring_req->u.indirect.handle = info->handle;
465 ring_req->u.indirect.nr_segments = nseg;
466 } else {
467 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
468 ring_req->u.rw.handle = info->handle;
469 ring_req->operation = rq_data_dir(req) ?
470 BLKIF_OP_WRITE : BLKIF_OP_READ;
471 if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) {
472 /*
473 * Ideally we can do an unordered flush-to-disk. In case the
474 * backend onlysupports barriers, use that. A barrier request
475 * a superset of FUA, so we can implement it the same
476 * way. (It's also a FLUSH+FUA, since it is
477 * guaranteed ordered WRT previous writes.)
478 */
479 ring_req->operation = info->flush_op;
480 }
481 ring_req->u.rw.nr_segments = nseg;
482 }
483 for_each_sg(info->shadow[id].sg, sg, nseg, i) {
484 fsect = sg->offset >> 9;
485 lsect = fsect + (sg->length >> 9) - 1;
486
487 if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
488 (i % SEGS_PER_INDIRECT_FRAME == 0)) {
489 unsigned long pfn;
490
491 if (segments)
492 kunmap_atomic(segments);
493
494 n = i / SEGS_PER_INDIRECT_FRAME;
495 if (!info->feature_persistent) {
496 struct page *indirect_page;
497
498 /* Fetch a pre-allocated page to use for indirect grefs */
499 BUG_ON(list_empty(&info->indirect_pages));
500 indirect_page = list_first_entry(&info->indirect_pages,
501 struct page, lru);
502 list_del(&indirect_page->lru);
503 pfn = page_to_pfn(indirect_page);
504 }
505 gnt_list_entry = get_grant(&gref_head, pfn, info);
506 info->shadow[id].indirect_grants[n] = gnt_list_entry;
507 segments = kmap_atomic(pfn_to_page(gnt_list_entry->pfn));
508 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
509 }
510
511 gnt_list_entry = get_grant(&gref_head, page_to_pfn(sg_page(sg)), info);
512 ref = gnt_list_entry->gref;
513
514 info->shadow[id].grants_used[i] = gnt_list_entry;
515
516 if (rq_data_dir(req) && info->feature_persistent) {
517 char *bvec_data;
518 void *shared_data;
519
520 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
521
522 shared_data = kmap_atomic(pfn_to_page(gnt_list_entry->pfn));
523 bvec_data = kmap_atomic(sg_page(sg));
524
525 /*
526 * this does not wipe data stored outside the
527 * range sg->offset..sg->offset+sg->length.
528 * Therefore, blkback *could* see data from
529 * previous requests. This is OK as long as
530 * persistent grants are shared with just one
531 * domain. It may need refactoring if this
532 * changes
533 */
534 memcpy(shared_data + sg->offset,
535 bvec_data + sg->offset,
536 sg->length);
537
538 kunmap_atomic(bvec_data);
539 kunmap_atomic(shared_data);
540 }
541 if (ring_req->operation != BLKIF_OP_INDIRECT) {
542 ring_req->u.rw.seg[i] =
543 (struct blkif_request_segment) {
544 .gref = ref,
545 .first_sect = fsect,
546 .last_sect = lsect };
547 } else {
548 n = i % SEGS_PER_INDIRECT_FRAME;
549 segments[n] =
550 (struct blkif_request_segment_aligned) {
551 .gref = ref,
552 .first_sect = fsect,
553 .last_sect = lsect };
554 }
555 }
556 if (segments)
557 kunmap_atomic(segments);
558 }
559
560 info->ring.req_prod_pvt++;
561
562 /* Keep a private copy so we can reissue requests when recovering. */
563 info->shadow[id].req = *ring_req;
564
565 if (new_persistent_gnts)
566 gnttab_free_grant_references(gref_head);
567
568 return 0;
569 }
570
571
572 static inline void flush_requests(struct blkfront_info *info)
573 {
574 int notify;
575
576 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
577
578 if (notify)
579 notify_remote_via_irq(info->irq);
580 }
581
582 /*
583 * do_blkif_request
584 * read a block; request is in a request queue
585 */
586 static void do_blkif_request(struct request_queue *rq)
587 {
588 struct blkfront_info *info = NULL;
589 struct request *req;
590 int queued;
591
592 pr_debug("Entered do_blkif_request\n");
593
594 queued = 0;
595
596 while ((req = blk_peek_request(rq)) != NULL) {
597 info = req->rq_disk->private_data;
598
599 if (RING_FULL(&info->ring))
600 goto wait;
601
602 blk_start_request(req);
603
604 if ((req->cmd_type != REQ_TYPE_FS) ||
605 ((req->cmd_flags & (REQ_FLUSH | REQ_FUA)) &&
606 !info->flush_op)) {
607 __blk_end_request_all(req, -EIO);
608 continue;
609 }
610
611 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
612 "(%u/%u) buffer:%p [%s]\n",
613 req, req->cmd, (unsigned long)blk_rq_pos(req),
614 blk_rq_cur_sectors(req), blk_rq_sectors(req),
615 req->buffer, rq_data_dir(req) ? "write" : "read");
616
617 if (blkif_queue_request(req)) {
618 blk_requeue_request(rq, req);
619 wait:
620 /* Avoid pointless unplugs. */
621 blk_stop_queue(rq);
622 break;
623 }
624
625 queued++;
626 }
627
628 if (queued != 0)
629 flush_requests(info);
630 }
631
632 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
633 unsigned int physical_sector_size,
634 unsigned int segments)
635 {
636 struct request_queue *rq;
637 struct blkfront_info *info = gd->private_data;
638
639 rq = blk_init_queue(do_blkif_request, &info->io_lock);
640 if (rq == NULL)
641 return -1;
642
643 queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
644
645 if (info->feature_discard) {
646 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
647 blk_queue_max_discard_sectors(rq, get_capacity(gd));
648 rq->limits.discard_granularity = info->discard_granularity;
649 rq->limits.discard_alignment = info->discard_alignment;
650 if (info->feature_secdiscard)
651 queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD, rq);
652 }
653
654 /* Hard sector size and max sectors impersonate the equiv. hardware. */
655 blk_queue_logical_block_size(rq, sector_size);
656 blk_queue_physical_block_size(rq, physical_sector_size);
657 blk_queue_max_hw_sectors(rq, (segments * PAGE_SIZE) / 512);
658
659 /* Each segment in a request is up to an aligned page in size. */
660 blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
661 blk_queue_max_segment_size(rq, PAGE_SIZE);
662
663 /* Ensure a merged request will fit in a single I/O ring slot. */
664 blk_queue_max_segments(rq, segments);
665
666 /* Make sure buffer addresses are sector-aligned. */
667 blk_queue_dma_alignment(rq, 511);
668
669 /* Make sure we don't use bounce buffers. */
670 blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
671
672 gd->queue = rq;
673
674 return 0;
675 }
676
677
678 static void xlvbd_flush(struct blkfront_info *info)
679 {
680 blk_queue_flush(info->rq, info->feature_flush);
681 printk(KERN_INFO "blkfront: %s: %s: %s %s %s %s %s\n",
682 info->gd->disk_name,
683 info->flush_op == BLKIF_OP_WRITE_BARRIER ?
684 "barrier" : (info->flush_op == BLKIF_OP_FLUSH_DISKCACHE ?
685 "flush diskcache" : "barrier or flush"),
686 info->feature_flush ? "enabled;" : "disabled;",
687 "persistent grants:",
688 info->feature_persistent ? "enabled;" : "disabled;",
689 "indirect descriptors:",
690 info->max_indirect_segments ? "enabled;" : "disabled;");
691 }
692
693 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
694 {
695 int major;
696 major = BLKIF_MAJOR(vdevice);
697 *minor = BLKIF_MINOR(vdevice);
698 switch (major) {
699 case XEN_IDE0_MAJOR:
700 *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
701 *minor = ((*minor / 64) * PARTS_PER_DISK) +
702 EMULATED_HD_DISK_MINOR_OFFSET;
703 break;
704 case XEN_IDE1_MAJOR:
705 *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
706 *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
707 EMULATED_HD_DISK_MINOR_OFFSET;
708 break;
709 case XEN_SCSI_DISK0_MAJOR:
710 *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
711 *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
712 break;
713 case XEN_SCSI_DISK1_MAJOR:
714 case XEN_SCSI_DISK2_MAJOR:
715 case XEN_SCSI_DISK3_MAJOR:
716 case XEN_SCSI_DISK4_MAJOR:
717 case XEN_SCSI_DISK5_MAJOR:
718 case XEN_SCSI_DISK6_MAJOR:
719 case XEN_SCSI_DISK7_MAJOR:
720 *offset = (*minor / PARTS_PER_DISK) +
721 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
722 EMULATED_SD_DISK_NAME_OFFSET;
723 *minor = *minor +
724 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
725 EMULATED_SD_DISK_MINOR_OFFSET;
726 break;
727 case XEN_SCSI_DISK8_MAJOR:
728 case XEN_SCSI_DISK9_MAJOR:
729 case XEN_SCSI_DISK10_MAJOR:
730 case XEN_SCSI_DISK11_MAJOR:
731 case XEN_SCSI_DISK12_MAJOR:
732 case XEN_SCSI_DISK13_MAJOR:
733 case XEN_SCSI_DISK14_MAJOR:
734 case XEN_SCSI_DISK15_MAJOR:
735 *offset = (*minor / PARTS_PER_DISK) +
736 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
737 EMULATED_SD_DISK_NAME_OFFSET;
738 *minor = *minor +
739 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
740 EMULATED_SD_DISK_MINOR_OFFSET;
741 break;
742 case XENVBD_MAJOR:
743 *offset = *minor / PARTS_PER_DISK;
744 break;
745 default:
746 printk(KERN_WARNING "blkfront: your disk configuration is "
747 "incorrect, please use an xvd device instead\n");
748 return -ENODEV;
749 }
750 return 0;
751 }
752
753 static char *encode_disk_name(char *ptr, unsigned int n)
754 {
755 if (n >= 26)
756 ptr = encode_disk_name(ptr, n / 26 - 1);
757 *ptr = 'a' + n % 26;
758 return ptr + 1;
759 }
760
761 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
762 struct blkfront_info *info,
763 u16 vdisk_info, u16 sector_size,
764 unsigned int physical_sector_size)
765 {
766 struct gendisk *gd;
767 int nr_minors = 1;
768 int err;
769 unsigned int offset;
770 int minor;
771 int nr_parts;
772 char *ptr;
773
774 BUG_ON(info->gd != NULL);
775 BUG_ON(info->rq != NULL);
776
777 if ((info->vdevice>>EXT_SHIFT) > 1) {
778 /* this is above the extended range; something is wrong */
779 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
780 return -ENODEV;
781 }
782
783 if (!VDEV_IS_EXTENDED(info->vdevice)) {
784 err = xen_translate_vdev(info->vdevice, &minor, &offset);
785 if (err)
786 return err;
787 nr_parts = PARTS_PER_DISK;
788 } else {
789 minor = BLKIF_MINOR_EXT(info->vdevice);
790 nr_parts = PARTS_PER_EXT_DISK;
791 offset = minor / nr_parts;
792 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
793 printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
794 "emulated IDE disks,\n\t choose an xvd device name"
795 "from xvde on\n", info->vdevice);
796 }
797 if (minor >> MINORBITS) {
798 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
799 info->vdevice, minor);
800 return -ENODEV;
801 }
802
803 if ((minor % nr_parts) == 0)
804 nr_minors = nr_parts;
805
806 err = xlbd_reserve_minors(minor, nr_minors);
807 if (err)
808 goto out;
809 err = -ENODEV;
810
811 gd = alloc_disk(nr_minors);
812 if (gd == NULL)
813 goto release;
814
815 strcpy(gd->disk_name, DEV_NAME);
816 ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
817 BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
818 if (nr_minors > 1)
819 *ptr = 0;
820 else
821 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
822 "%d", minor & (nr_parts - 1));
823
824 gd->major = XENVBD_MAJOR;
825 gd->first_minor = minor;
826 gd->fops = &xlvbd_block_fops;
827 gd->private_data = info;
828 gd->driverfs_dev = &(info->xbdev->dev);
829 set_capacity(gd, capacity);
830
831 if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size,
832 info->max_indirect_segments ? :
833 BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
834 del_gendisk(gd);
835 goto release;
836 }
837
838 info->rq = gd->queue;
839 info->gd = gd;
840
841 xlvbd_flush(info);
842
843 if (vdisk_info & VDISK_READONLY)
844 set_disk_ro(gd, 1);
845
846 if (vdisk_info & VDISK_REMOVABLE)
847 gd->flags |= GENHD_FL_REMOVABLE;
848
849 if (vdisk_info & VDISK_CDROM)
850 gd->flags |= GENHD_FL_CD;
851
852 return 0;
853
854 release:
855 xlbd_release_minors(minor, nr_minors);
856 out:
857 return err;
858 }
859
860 static void xlvbd_release_gendisk(struct blkfront_info *info)
861 {
862 unsigned int minor, nr_minors;
863 unsigned long flags;
864
865 if (info->rq == NULL)
866 return;
867
868 spin_lock_irqsave(&info->io_lock, flags);
869
870 /* No more blkif_request(). */
871 blk_stop_queue(info->rq);
872
873 /* No more gnttab callback work. */
874 gnttab_cancel_free_callback(&info->callback);
875 spin_unlock_irqrestore(&info->io_lock, flags);
876
877 /* Flush gnttab callback work. Must be done with no locks held. */
878 flush_work(&info->work);
879
880 del_gendisk(info->gd);
881
882 minor = info->gd->first_minor;
883 nr_minors = info->gd->minors;
884 xlbd_release_minors(minor, nr_minors);
885
886 blk_cleanup_queue(info->rq);
887 info->rq = NULL;
888
889 put_disk(info->gd);
890 info->gd = NULL;
891 }
892
893 static void kick_pending_request_queues(struct blkfront_info *info)
894 {
895 if (!RING_FULL(&info->ring)) {
896 /* Re-enable calldowns. */
897 blk_start_queue(info->rq);
898 /* Kick things off immediately. */
899 do_blkif_request(info->rq);
900 }
901 }
902
903 static void blkif_restart_queue(struct work_struct *work)
904 {
905 struct blkfront_info *info = container_of(work, struct blkfront_info, work);
906
907 spin_lock_irq(&info->io_lock);
908 if (info->connected == BLKIF_STATE_CONNECTED)
909 kick_pending_request_queues(info);
910 spin_unlock_irq(&info->io_lock);
911 }
912
913 static void blkif_free(struct blkfront_info *info, int suspend)
914 {
915 struct grant *persistent_gnt;
916 struct grant *n;
917 int i, j, segs;
918
919 /* Prevent new requests being issued until we fix things up. */
920 spin_lock_irq(&info->io_lock);
921 info->connected = suspend ?
922 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
923 /* No more blkif_request(). */
924 if (info->rq)
925 blk_stop_queue(info->rq);
926
927 /* Remove all persistent grants */
928 if (!list_empty(&info->grants)) {
929 list_for_each_entry_safe(persistent_gnt, n,
930 &info->grants, node) {
931 list_del(&persistent_gnt->node);
932 if (persistent_gnt->gref != GRANT_INVALID_REF) {
933 gnttab_end_foreign_access(persistent_gnt->gref,
934 0, 0UL);
935 info->persistent_gnts_c--;
936 }
937 if (info->feature_persistent)
938 __free_page(pfn_to_page(persistent_gnt->pfn));
939 kfree(persistent_gnt);
940 }
941 }
942 BUG_ON(info->persistent_gnts_c != 0);
943
944 /*
945 * Remove indirect pages, this only happens when using indirect
946 * descriptors but not persistent grants
947 */
948 if (!list_empty(&info->indirect_pages)) {
949 struct page *indirect_page, *n;
950
951 BUG_ON(info->feature_persistent);
952 list_for_each_entry_safe(indirect_page, n, &info->indirect_pages, lru) {
953 list_del(&indirect_page->lru);
954 __free_page(indirect_page);
955 }
956 }
957
958 for (i = 0; i < BLK_RING_SIZE; i++) {
959 /*
960 * Clear persistent grants present in requests already
961 * on the shared ring
962 */
963 if (!info->shadow[i].request)
964 goto free_shadow;
965
966 segs = info->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
967 info->shadow[i].req.u.indirect.nr_segments :
968 info->shadow[i].req.u.rw.nr_segments;
969 for (j = 0; j < segs; j++) {
970 persistent_gnt = info->shadow[i].grants_used[j];
971 gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
972 if (info->feature_persistent)
973 __free_page(pfn_to_page(persistent_gnt->pfn));
974 kfree(persistent_gnt);
975 }
976
977 if (info->shadow[i].req.operation != BLKIF_OP_INDIRECT)
978 /*
979 * If this is not an indirect operation don't try to
980 * free indirect segments
981 */
982 goto free_shadow;
983
984 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
985 persistent_gnt = info->shadow[i].indirect_grants[j];
986 gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
987 __free_page(pfn_to_page(persistent_gnt->pfn));
988 kfree(persistent_gnt);
989 }
990
991 free_shadow:
992 kfree(info->shadow[i].grants_used);
993 info->shadow[i].grants_used = NULL;
994 kfree(info->shadow[i].indirect_grants);
995 info->shadow[i].indirect_grants = NULL;
996 kfree(info->shadow[i].sg);
997 info->shadow[i].sg = NULL;
998 }
999
1000 /* No more gnttab callback work. */
1001 gnttab_cancel_free_callback(&info->callback);
1002 spin_unlock_irq(&info->io_lock);
1003
1004 /* Flush gnttab callback work. Must be done with no locks held. */
1005 flush_work(&info->work);
1006
1007 /* Free resources associated with old device channel. */
1008 if (info->ring_ref != GRANT_INVALID_REF) {
1009 gnttab_end_foreign_access(info->ring_ref, 0,
1010 (unsigned long)info->ring.sring);
1011 info->ring_ref = GRANT_INVALID_REF;
1012 info->ring.sring = NULL;
1013 }
1014 if (info->irq)
1015 unbind_from_irqhandler(info->irq, info);
1016 info->evtchn = info->irq = 0;
1017
1018 }
1019
1020 static void blkif_completion(struct blk_shadow *s, struct blkfront_info *info,
1021 struct blkif_response *bret)
1022 {
1023 int i = 0;
1024 struct scatterlist *sg;
1025 char *bvec_data;
1026 void *shared_data;
1027 int nseg;
1028
1029 nseg = s->req.operation == BLKIF_OP_INDIRECT ?
1030 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1031
1032 if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1033 /*
1034 * Copy the data received from the backend into the bvec.
1035 * Since bv_offset can be different than 0, and bv_len different
1036 * than PAGE_SIZE, we have to keep track of the current offset,
1037 * to be sure we are copying the data from the right shared page.
1038 */
1039 for_each_sg(s->sg, sg, nseg, i) {
1040 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1041 shared_data = kmap_atomic(
1042 pfn_to_page(s->grants_used[i]->pfn));
1043 bvec_data = kmap_atomic(sg_page(sg));
1044 memcpy(bvec_data + sg->offset,
1045 shared_data + sg->offset,
1046 sg->length);
1047 kunmap_atomic(bvec_data);
1048 kunmap_atomic(shared_data);
1049 }
1050 }
1051 /* Add the persistent grant into the list of free grants */
1052 for (i = 0; i < nseg; i++) {
1053 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1054 /*
1055 * If the grant is still mapped by the backend (the
1056 * backend has chosen to make this grant persistent)
1057 * we add it at the head of the list, so it will be
1058 * reused first.
1059 */
1060 if (!info->feature_persistent)
1061 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1062 s->grants_used[i]->gref);
1063 list_add(&s->grants_used[i]->node, &info->grants);
1064 info->persistent_gnts_c++;
1065 } else {
1066 /*
1067 * If the grant is not mapped by the backend we end the
1068 * foreign access and add it to the tail of the list,
1069 * so it will not be picked again unless we run out of
1070 * persistent grants.
1071 */
1072 gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1073 s->grants_used[i]->gref = GRANT_INVALID_REF;
1074 list_add_tail(&s->grants_used[i]->node, &info->grants);
1075 }
1076 }
1077 if (s->req.operation == BLKIF_OP_INDIRECT) {
1078 for (i = 0; i < INDIRECT_GREFS(nseg); i++) {
1079 if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1080 if (!info->feature_persistent)
1081 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1082 s->indirect_grants[i]->gref);
1083 list_add(&s->indirect_grants[i]->node, &info->grants);
1084 info->persistent_gnts_c++;
1085 } else {
1086 struct page *indirect_page;
1087
1088 gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1089 /*
1090 * Add the used indirect page back to the list of
1091 * available pages for indirect grefs.
1092 */
1093 if (!info->feature_persistent) {
1094 indirect_page = pfn_to_page(s->indirect_grants[i]->pfn);
1095 list_add(&indirect_page->lru, &info->indirect_pages);
1096 }
1097 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1098 list_add_tail(&s->indirect_grants[i]->node, &info->grants);
1099 }
1100 }
1101 }
1102 }
1103
1104 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1105 {
1106 struct request *req;
1107 struct blkif_response *bret;
1108 RING_IDX i, rp;
1109 unsigned long flags;
1110 struct blkfront_info *info = (struct blkfront_info *)dev_id;
1111 int error;
1112
1113 spin_lock_irqsave(&info->io_lock, flags);
1114
1115 if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
1116 spin_unlock_irqrestore(&info->io_lock, flags);
1117 return IRQ_HANDLED;
1118 }
1119
1120 again:
1121 rp = info->ring.sring->rsp_prod;
1122 rmb(); /* Ensure we see queued responses up to 'rp'. */
1123
1124 for (i = info->ring.rsp_cons; i != rp; i++) {
1125 unsigned long id;
1126
1127 bret = RING_GET_RESPONSE(&info->ring, i);
1128 id = bret->id;
1129 /*
1130 * The backend has messed up and given us an id that we would
1131 * never have given to it (we stamp it up to BLK_RING_SIZE -
1132 * look in get_id_from_freelist.
1133 */
1134 if (id >= BLK_RING_SIZE) {
1135 WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1136 info->gd->disk_name, op_name(bret->operation), id);
1137 /* We can't safely get the 'struct request' as
1138 * the id is busted. */
1139 continue;
1140 }
1141 req = info->shadow[id].request;
1142
1143 if (bret->operation != BLKIF_OP_DISCARD)
1144 blkif_completion(&info->shadow[id], info, bret);
1145
1146 if (add_id_to_freelist(info, id)) {
1147 WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1148 info->gd->disk_name, op_name(bret->operation), id);
1149 continue;
1150 }
1151
1152 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
1153 switch (bret->operation) {
1154 case BLKIF_OP_DISCARD:
1155 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1156 struct request_queue *rq = info->rq;
1157 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1158 info->gd->disk_name, op_name(bret->operation));
1159 error = -EOPNOTSUPP;
1160 info->feature_discard = 0;
1161 info->feature_secdiscard = 0;
1162 queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1163 queue_flag_clear(QUEUE_FLAG_SECDISCARD, rq);
1164 }
1165 __blk_end_request_all(req, error);
1166 break;
1167 case BLKIF_OP_FLUSH_DISKCACHE:
1168 case BLKIF_OP_WRITE_BARRIER:
1169 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1170 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1171 info->gd->disk_name, op_name(bret->operation));
1172 error = -EOPNOTSUPP;
1173 }
1174 if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1175 info->shadow[id].req.u.rw.nr_segments == 0)) {
1176 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1177 info->gd->disk_name, op_name(bret->operation));
1178 error = -EOPNOTSUPP;
1179 }
1180 if (unlikely(error)) {
1181 if (error == -EOPNOTSUPP)
1182 error = 0;
1183 info->feature_flush = 0;
1184 info->flush_op = 0;
1185 xlvbd_flush(info);
1186 }
1187 /* fall through */
1188 case BLKIF_OP_READ:
1189 case BLKIF_OP_WRITE:
1190 if (unlikely(bret->status != BLKIF_RSP_OKAY))
1191 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1192 "request: %x\n", bret->status);
1193
1194 __blk_end_request_all(req, error);
1195 break;
1196 default:
1197 BUG();
1198 }
1199 }
1200
1201 info->ring.rsp_cons = i;
1202
1203 if (i != info->ring.req_prod_pvt) {
1204 int more_to_do;
1205 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
1206 if (more_to_do)
1207 goto again;
1208 } else
1209 info->ring.sring->rsp_event = i + 1;
1210
1211 kick_pending_request_queues(info);
1212
1213 spin_unlock_irqrestore(&info->io_lock, flags);
1214
1215 return IRQ_HANDLED;
1216 }
1217
1218
1219 static int setup_blkring(struct xenbus_device *dev,
1220 struct blkfront_info *info)
1221 {
1222 struct blkif_sring *sring;
1223 int err;
1224
1225 info->ring_ref = GRANT_INVALID_REF;
1226
1227 sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
1228 if (!sring) {
1229 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1230 return -ENOMEM;
1231 }
1232 SHARED_RING_INIT(sring);
1233 FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
1234
1235 err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
1236 if (err < 0) {
1237 free_page((unsigned long)sring);
1238 info->ring.sring = NULL;
1239 goto fail;
1240 }
1241 info->ring_ref = err;
1242
1243 err = xenbus_alloc_evtchn(dev, &info->evtchn);
1244 if (err)
1245 goto fail;
1246
1247 err = bind_evtchn_to_irqhandler(info->evtchn, blkif_interrupt, 0,
1248 "blkif", info);
1249 if (err <= 0) {
1250 xenbus_dev_fatal(dev, err,
1251 "bind_evtchn_to_irqhandler failed");
1252 goto fail;
1253 }
1254 info->irq = err;
1255
1256 return 0;
1257 fail:
1258 blkif_free(info, 0);
1259 return err;
1260 }
1261
1262
1263 /* Common code used when first setting up, and when resuming. */
1264 static int talk_to_blkback(struct xenbus_device *dev,
1265 struct blkfront_info *info)
1266 {
1267 const char *message = NULL;
1268 struct xenbus_transaction xbt;
1269 int err;
1270
1271 /* Create shared ring, alloc event channel. */
1272 err = setup_blkring(dev, info);
1273 if (err)
1274 goto out;
1275
1276 again:
1277 err = xenbus_transaction_start(&xbt);
1278 if (err) {
1279 xenbus_dev_fatal(dev, err, "starting transaction");
1280 goto destroy_blkring;
1281 }
1282
1283 err = xenbus_printf(xbt, dev->nodename,
1284 "ring-ref", "%u", info->ring_ref);
1285 if (err) {
1286 message = "writing ring-ref";
1287 goto abort_transaction;
1288 }
1289 err = xenbus_printf(xbt, dev->nodename,
1290 "event-channel", "%u", info->evtchn);
1291 if (err) {
1292 message = "writing event-channel";
1293 goto abort_transaction;
1294 }
1295 err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1296 XEN_IO_PROTO_ABI_NATIVE);
1297 if (err) {
1298 message = "writing protocol";
1299 goto abort_transaction;
1300 }
1301 err = xenbus_printf(xbt, dev->nodename,
1302 "feature-persistent", "%u", 1);
1303 if (err)
1304 dev_warn(&dev->dev,
1305 "writing persistent grants feature to xenbus");
1306
1307 err = xenbus_transaction_end(xbt, 0);
1308 if (err) {
1309 if (err == -EAGAIN)
1310 goto again;
1311 xenbus_dev_fatal(dev, err, "completing transaction");
1312 goto destroy_blkring;
1313 }
1314
1315 xenbus_switch_state(dev, XenbusStateInitialised);
1316
1317 return 0;
1318
1319 abort_transaction:
1320 xenbus_transaction_end(xbt, 1);
1321 if (message)
1322 xenbus_dev_fatal(dev, err, "%s", message);
1323 destroy_blkring:
1324 blkif_free(info, 0);
1325 out:
1326 return err;
1327 }
1328
1329 /**
1330 * Entry point to this code when a new device is created. Allocate the basic
1331 * structures and the ring buffer for communication with the backend, and
1332 * inform the backend of the appropriate details for those. Switch to
1333 * Initialised state.
1334 */
1335 static int blkfront_probe(struct xenbus_device *dev,
1336 const struct xenbus_device_id *id)
1337 {
1338 int err, vdevice, i;
1339 struct blkfront_info *info;
1340
1341 /* FIXME: Use dynamic device id if this is not set. */
1342 err = xenbus_scanf(XBT_NIL, dev->nodename,
1343 "virtual-device", "%i", &vdevice);
1344 if (err != 1) {
1345 /* go looking in the extended area instead */
1346 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1347 "%i", &vdevice);
1348 if (err != 1) {
1349 xenbus_dev_fatal(dev, err, "reading virtual-device");
1350 return err;
1351 }
1352 }
1353
1354 if (xen_hvm_domain()) {
1355 char *type;
1356 int len;
1357 /* no unplug has been done: do not hook devices != xen vbds */
1358 if (xen_has_pv_and_legacy_disk_devices()) {
1359 int major;
1360
1361 if (!VDEV_IS_EXTENDED(vdevice))
1362 major = BLKIF_MAJOR(vdevice);
1363 else
1364 major = XENVBD_MAJOR;
1365
1366 if (major != XENVBD_MAJOR) {
1367 printk(KERN_INFO
1368 "%s: HVM does not support vbd %d as xen block device\n",
1369 __FUNCTION__, vdevice);
1370 return -ENODEV;
1371 }
1372 }
1373 /* do not create a PV cdrom device if we are an HVM guest */
1374 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1375 if (IS_ERR(type))
1376 return -ENODEV;
1377 if (strncmp(type, "cdrom", 5) == 0) {
1378 kfree(type);
1379 return -ENODEV;
1380 }
1381 kfree(type);
1382 }
1383 info = kzalloc(sizeof(*info), GFP_KERNEL);
1384 if (!info) {
1385 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1386 return -ENOMEM;
1387 }
1388
1389 mutex_init(&info->mutex);
1390 spin_lock_init(&info->io_lock);
1391 info->xbdev = dev;
1392 info->vdevice = vdevice;
1393 INIT_LIST_HEAD(&info->grants);
1394 INIT_LIST_HEAD(&info->indirect_pages);
1395 info->persistent_gnts_c = 0;
1396 info->connected = BLKIF_STATE_DISCONNECTED;
1397 INIT_WORK(&info->work, blkif_restart_queue);
1398
1399 for (i = 0; i < BLK_RING_SIZE; i++)
1400 info->shadow[i].req.u.rw.id = i+1;
1401 info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1402
1403 /* Front end dir is a number, which is used as the id. */
1404 info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1405 dev_set_drvdata(&dev->dev, info);
1406
1407 err = talk_to_blkback(dev, info);
1408 if (err) {
1409 kfree(info);
1410 dev_set_drvdata(&dev->dev, NULL);
1411 return err;
1412 }
1413
1414 return 0;
1415 }
1416
1417 /*
1418 * This is a clone of md_trim_bio, used to split a bio into smaller ones
1419 */
1420 static void trim_bio(struct bio *bio, int offset, int size)
1421 {
1422 /* 'bio' is a cloned bio which we need to trim to match
1423 * the given offset and size.
1424 * This requires adjusting bi_sector, bi_size, and bi_io_vec
1425 */
1426 int i;
1427 struct bio_vec *bvec;
1428 int sofar = 0;
1429
1430 size <<= 9;
1431 if (offset == 0 && size == bio->bi_size)
1432 return;
1433
1434 bio->bi_sector += offset;
1435 bio->bi_size = size;
1436 offset <<= 9;
1437 clear_bit(BIO_SEG_VALID, &bio->bi_flags);
1438
1439 while (bio->bi_idx < bio->bi_vcnt &&
1440 bio->bi_io_vec[bio->bi_idx].bv_len <= offset) {
1441 /* remove this whole bio_vec */
1442 offset -= bio->bi_io_vec[bio->bi_idx].bv_len;
1443 bio->bi_idx++;
1444 }
1445 if (bio->bi_idx < bio->bi_vcnt) {
1446 bio->bi_io_vec[bio->bi_idx].bv_offset += offset;
1447 bio->bi_io_vec[bio->bi_idx].bv_len -= offset;
1448 }
1449 /* avoid any complications with bi_idx being non-zero*/
1450 if (bio->bi_idx) {
1451 memmove(bio->bi_io_vec, bio->bi_io_vec+bio->bi_idx,
1452 (bio->bi_vcnt - bio->bi_idx) * sizeof(struct bio_vec));
1453 bio->bi_vcnt -= bio->bi_idx;
1454 bio->bi_idx = 0;
1455 }
1456 /* Make sure vcnt and last bv are not too big */
1457 bio_for_each_segment(bvec, bio, i) {
1458 if (sofar + bvec->bv_len > size)
1459 bvec->bv_len = size - sofar;
1460 if (bvec->bv_len == 0) {
1461 bio->bi_vcnt = i;
1462 break;
1463 }
1464 sofar += bvec->bv_len;
1465 }
1466 }
1467
1468 static void split_bio_end(struct bio *bio, int error)
1469 {
1470 struct split_bio *split_bio = bio->bi_private;
1471
1472 if (error)
1473 split_bio->err = error;
1474
1475 if (atomic_dec_and_test(&split_bio->pending)) {
1476 split_bio->bio->bi_phys_segments = 0;
1477 bio_endio(split_bio->bio, split_bio->err);
1478 kfree(split_bio);
1479 }
1480 bio_put(bio);
1481 }
1482
1483 static int blkif_recover(struct blkfront_info *info)
1484 {
1485 int i;
1486 struct request *req, *n;
1487 struct blk_shadow *copy;
1488 int rc;
1489 struct bio *bio, *cloned_bio;
1490 struct bio_list bio_list, merge_bio;
1491 unsigned int segs, offset;
1492 int pending, size;
1493 struct split_bio *split_bio;
1494 struct list_head requests;
1495
1496 /* Stage 1: Make a safe copy of the shadow state. */
1497 copy = kmemdup(info->shadow, sizeof(info->shadow),
1498 GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
1499 if (!copy)
1500 return -ENOMEM;
1501
1502 /* Stage 2: Set up free list. */
1503 memset(&info->shadow, 0, sizeof(info->shadow));
1504 for (i = 0; i < BLK_RING_SIZE; i++)
1505 info->shadow[i].req.u.rw.id = i+1;
1506 info->shadow_free = info->ring.req_prod_pvt;
1507 info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1508
1509 rc = blkfront_setup_indirect(info);
1510 if (rc) {
1511 kfree(copy);
1512 return rc;
1513 }
1514
1515 segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
1516 blk_queue_max_segments(info->rq, segs);
1517 bio_list_init(&bio_list);
1518 INIT_LIST_HEAD(&requests);
1519 for (i = 0; i < BLK_RING_SIZE; i++) {
1520 /* Not in use? */
1521 if (!copy[i].request)
1522 continue;
1523
1524 /*
1525 * Get the bios in the request so we can re-queue them.
1526 */
1527 if (copy[i].request->cmd_flags &
1528 (REQ_FLUSH | REQ_FUA | REQ_DISCARD | REQ_SECURE)) {
1529 /*
1530 * Flush operations don't contain bios, so
1531 * we need to requeue the whole request
1532 */
1533 list_add(&copy[i].request->queuelist, &requests);
1534 continue;
1535 }
1536 merge_bio.head = copy[i].request->bio;
1537 merge_bio.tail = copy[i].request->biotail;
1538 bio_list_merge(&bio_list, &merge_bio);
1539 copy[i].request->bio = NULL;
1540 blk_put_request(copy[i].request);
1541 }
1542
1543 kfree(copy);
1544
1545 /*
1546 * Empty the queue, this is important because we might have
1547 * requests in the queue with more segments than what we
1548 * can handle now.
1549 */
1550 spin_lock_irq(&info->io_lock);
1551 while ((req = blk_fetch_request(info->rq)) != NULL) {
1552 if (req->cmd_flags &
1553 (REQ_FLUSH | REQ_FUA | REQ_DISCARD | REQ_SECURE)) {
1554 list_add(&req->queuelist, &requests);
1555 continue;
1556 }
1557 merge_bio.head = req->bio;
1558 merge_bio.tail = req->biotail;
1559 bio_list_merge(&bio_list, &merge_bio);
1560 req->bio = NULL;
1561 if (req->cmd_flags & (REQ_FLUSH | REQ_FUA))
1562 pr_alert("diskcache flush request found!\n");
1563 __blk_put_request(info->rq, req);
1564 }
1565 spin_unlock_irq(&info->io_lock);
1566
1567 xenbus_switch_state(info->xbdev, XenbusStateConnected);
1568
1569 spin_lock_irq(&info->io_lock);
1570
1571 /* Now safe for us to use the shared ring */
1572 info->connected = BLKIF_STATE_CONNECTED;
1573
1574 /* Kick any other new requests queued since we resumed */
1575 kick_pending_request_queues(info);
1576
1577 list_for_each_entry_safe(req, n, &requests, queuelist) {
1578 /* Requeue pending requests (flush or discard) */
1579 list_del_init(&req->queuelist);
1580 BUG_ON(req->nr_phys_segments > segs);
1581 blk_requeue_request(info->rq, req);
1582 }
1583 spin_unlock_irq(&info->io_lock);
1584
1585 while ((bio = bio_list_pop(&bio_list)) != NULL) {
1586 /* Traverse the list of pending bios and re-queue them */
1587 if (bio_segments(bio) > segs) {
1588 /*
1589 * This bio has more segments than what we can
1590 * handle, we have to split it.
1591 */
1592 pending = (bio_segments(bio) + segs - 1) / segs;
1593 split_bio = kzalloc(sizeof(*split_bio), GFP_NOIO);
1594 BUG_ON(split_bio == NULL);
1595 atomic_set(&split_bio->pending, pending);
1596 split_bio->bio = bio;
1597 for (i = 0; i < pending; i++) {
1598 offset = (i * segs * PAGE_SIZE) >> 9;
1599 size = min((unsigned int)(segs * PAGE_SIZE) >> 9,
1600 (unsigned int)(bio->bi_size >> 9) - offset);
1601 cloned_bio = bio_clone(bio, GFP_NOIO);
1602 BUG_ON(cloned_bio == NULL);
1603 trim_bio(cloned_bio, offset, size);
1604 cloned_bio->bi_private = split_bio;
1605 cloned_bio->bi_end_io = split_bio_end;
1606 submit_bio(cloned_bio->bi_rw, cloned_bio);
1607 }
1608 /*
1609 * Now we have to wait for all those smaller bios to
1610 * end, so we can also end the "parent" bio.
1611 */
1612 continue;
1613 }
1614 /* We don't need to split this bio */
1615 submit_bio(bio->bi_rw, bio);
1616 }
1617
1618 return 0;
1619 }
1620
1621 /**
1622 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1623 * driver restart. We tear down our blkif structure and recreate it, but
1624 * leave the device-layer structures intact so that this is transparent to the
1625 * rest of the kernel.
1626 */
1627 static int blkfront_resume(struct xenbus_device *dev)
1628 {
1629 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1630 int err;
1631
1632 dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
1633
1634 blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
1635
1636 err = talk_to_blkback(dev, info);
1637
1638 /*
1639 * We have to wait for the backend to switch to
1640 * connected state, since we want to read which
1641 * features it supports.
1642 */
1643
1644 return err;
1645 }
1646
1647 static void
1648 blkfront_closing(struct blkfront_info *info)
1649 {
1650 struct xenbus_device *xbdev = info->xbdev;
1651 struct block_device *bdev = NULL;
1652
1653 mutex_lock(&info->mutex);
1654
1655 if (xbdev->state == XenbusStateClosing) {
1656 mutex_unlock(&info->mutex);
1657 return;
1658 }
1659
1660 if (info->gd)
1661 bdev = bdget_disk(info->gd, 0);
1662
1663 mutex_unlock(&info->mutex);
1664
1665 if (!bdev) {
1666 xenbus_frontend_closed(xbdev);
1667 return;
1668 }
1669
1670 mutex_lock(&bdev->bd_mutex);
1671
1672 if (bdev->bd_openers) {
1673 xenbus_dev_error(xbdev, -EBUSY,
1674 "Device in use; refusing to close");
1675 xenbus_switch_state(xbdev, XenbusStateClosing);
1676 } else {
1677 xlvbd_release_gendisk(info);
1678 xenbus_frontend_closed(xbdev);
1679 }
1680
1681 mutex_unlock(&bdev->bd_mutex);
1682 bdput(bdev);
1683 }
1684
1685 static void blkfront_setup_discard(struct blkfront_info *info)
1686 {
1687 int err;
1688 char *type;
1689 unsigned int discard_granularity;
1690 unsigned int discard_alignment;
1691 unsigned int discard_secure;
1692
1693 type = xenbus_read(XBT_NIL, info->xbdev->otherend, "type", NULL);
1694 if (IS_ERR(type))
1695 return;
1696
1697 info->feature_secdiscard = 0;
1698 if (strncmp(type, "phy", 3) == 0) {
1699 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1700 "discard-granularity", "%u", &discard_granularity,
1701 "discard-alignment", "%u", &discard_alignment,
1702 NULL);
1703 if (!err) {
1704 info->feature_discard = 1;
1705 info->discard_granularity = discard_granularity;
1706 info->discard_alignment = discard_alignment;
1707 }
1708 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1709 "discard-secure", "%d", &discard_secure,
1710 NULL);
1711 if (!err)
1712 info->feature_secdiscard = discard_secure;
1713
1714 } else if (strncmp(type, "file", 4) == 0)
1715 info->feature_discard = 1;
1716
1717 kfree(type);
1718 }
1719
1720 static int blkfront_setup_indirect(struct blkfront_info *info)
1721 {
1722 unsigned int indirect_segments, segs;
1723 int err, i;
1724
1725 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1726 "feature-max-indirect-segments", "%u", &indirect_segments,
1727 NULL);
1728 if (err) {
1729 info->max_indirect_segments = 0;
1730 segs = BLKIF_MAX_SEGMENTS_PER_REQUEST;
1731 } else {
1732 info->max_indirect_segments = min(indirect_segments,
1733 xen_blkif_max_segments);
1734 segs = info->max_indirect_segments;
1735 }
1736
1737 err = fill_grant_buffer(info, (segs + INDIRECT_GREFS(segs)) * BLK_RING_SIZE);
1738 if (err)
1739 goto out_of_memory;
1740
1741 if (!info->feature_persistent && info->max_indirect_segments) {
1742 /*
1743 * We are using indirect descriptors but not persistent
1744 * grants, we need to allocate a set of pages that can be
1745 * used for mapping indirect grefs
1746 */
1747 int num = INDIRECT_GREFS(segs) * BLK_RING_SIZE;
1748
1749 BUG_ON(!list_empty(&info->indirect_pages));
1750 for (i = 0; i < num; i++) {
1751 struct page *indirect_page = alloc_page(GFP_NOIO);
1752 if (!indirect_page)
1753 goto out_of_memory;
1754 list_add(&indirect_page->lru, &info->indirect_pages);
1755 }
1756 }
1757
1758 for (i = 0; i < BLK_RING_SIZE; i++) {
1759 info->shadow[i].grants_used = kzalloc(
1760 sizeof(info->shadow[i].grants_used[0]) * segs,
1761 GFP_NOIO);
1762 info->shadow[i].sg = kzalloc(sizeof(info->shadow[i].sg[0]) * segs, GFP_NOIO);
1763 if (info->max_indirect_segments)
1764 info->shadow[i].indirect_grants = kzalloc(
1765 sizeof(info->shadow[i].indirect_grants[0]) *
1766 INDIRECT_GREFS(segs),
1767 GFP_NOIO);
1768 if ((info->shadow[i].grants_used == NULL) ||
1769 (info->shadow[i].sg == NULL) ||
1770 (info->max_indirect_segments &&
1771 (info->shadow[i].indirect_grants == NULL)))
1772 goto out_of_memory;
1773 sg_init_table(info->shadow[i].sg, segs);
1774 }
1775
1776
1777 return 0;
1778
1779 out_of_memory:
1780 for (i = 0; i < BLK_RING_SIZE; i++) {
1781 kfree(info->shadow[i].grants_used);
1782 info->shadow[i].grants_used = NULL;
1783 kfree(info->shadow[i].sg);
1784 info->shadow[i].sg = NULL;
1785 kfree(info->shadow[i].indirect_grants);
1786 info->shadow[i].indirect_grants = NULL;
1787 }
1788 if (!list_empty(&info->indirect_pages)) {
1789 struct page *indirect_page, *n;
1790 list_for_each_entry_safe(indirect_page, n, &info->indirect_pages, lru) {
1791 list_del(&indirect_page->lru);
1792 __free_page(indirect_page);
1793 }
1794 }
1795 return -ENOMEM;
1796 }
1797
1798 /*
1799 * Invoked when the backend is finally 'ready' (and has told produced
1800 * the details about the physical device - #sectors, size, etc).
1801 */
1802 static void blkfront_connect(struct blkfront_info *info)
1803 {
1804 unsigned long long sectors;
1805 unsigned long sector_size;
1806 unsigned int physical_sector_size;
1807 unsigned int binfo;
1808 int err;
1809 int barrier, flush, discard, persistent;
1810
1811 switch (info->connected) {
1812 case BLKIF_STATE_CONNECTED:
1813 /*
1814 * Potentially, the back-end may be signalling
1815 * a capacity change; update the capacity.
1816 */
1817 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1818 "sectors", "%Lu", &sectors);
1819 if (XENBUS_EXIST_ERR(err))
1820 return;
1821 printk(KERN_INFO "Setting capacity to %Lu\n",
1822 sectors);
1823 set_capacity(info->gd, sectors);
1824 revalidate_disk(info->gd);
1825
1826 return;
1827 case BLKIF_STATE_SUSPENDED:
1828 /*
1829 * If we are recovering from suspension, we need to wait
1830 * for the backend to announce it's features before
1831 * reconnecting, at least we need to know if the backend
1832 * supports indirect descriptors, and how many.
1833 */
1834 blkif_recover(info);
1835 return;
1836
1837 default:
1838 break;
1839 }
1840
1841 dev_dbg(&info->xbdev->dev, "%s:%s.\n",
1842 __func__, info->xbdev->otherend);
1843
1844 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1845 "sectors", "%llu", &sectors,
1846 "info", "%u", &binfo,
1847 "sector-size", "%lu", &sector_size,
1848 NULL);
1849 if (err) {
1850 xenbus_dev_fatal(info->xbdev, err,
1851 "reading backend fields at %s",
1852 info->xbdev->otherend);
1853 return;
1854 }
1855
1856 /*
1857 * physcial-sector-size is a newer field, so old backends may not
1858 * provide this. Assume physical sector size to be the same as
1859 * sector_size in that case.
1860 */
1861 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1862 "physical-sector-size", "%u", &physical_sector_size);
1863 if (err != 1)
1864 physical_sector_size = sector_size;
1865
1866 info->feature_flush = 0;
1867 info->flush_op = 0;
1868
1869 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1870 "feature-barrier", "%d", &barrier,
1871 NULL);
1872
1873 /*
1874 * If there's no "feature-barrier" defined, then it means
1875 * we're dealing with a very old backend which writes
1876 * synchronously; nothing to do.
1877 *
1878 * If there are barriers, then we use flush.
1879 */
1880 if (!err && barrier) {
1881 info->feature_flush = REQ_FLUSH | REQ_FUA;
1882 info->flush_op = BLKIF_OP_WRITE_BARRIER;
1883 }
1884 /*
1885 * And if there is "feature-flush-cache" use that above
1886 * barriers.
1887 */
1888 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1889 "feature-flush-cache", "%d", &flush,
1890 NULL);
1891
1892 if (!err && flush) {
1893 info->feature_flush = REQ_FLUSH;
1894 info->flush_op = BLKIF_OP_FLUSH_DISKCACHE;
1895 }
1896
1897 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1898 "feature-discard", "%d", &discard,
1899 NULL);
1900
1901 if (!err && discard)
1902 blkfront_setup_discard(info);
1903
1904 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1905 "feature-persistent", "%u", &persistent,
1906 NULL);
1907 if (err)
1908 info->feature_persistent = 0;
1909 else
1910 info->feature_persistent = persistent;
1911
1912 err = blkfront_setup_indirect(info);
1913 if (err) {
1914 xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
1915 info->xbdev->otherend);
1916 return;
1917 }
1918
1919 err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
1920 physical_sector_size);
1921 if (err) {
1922 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
1923 info->xbdev->otherend);
1924 return;
1925 }
1926
1927 xenbus_switch_state(info->xbdev, XenbusStateConnected);
1928
1929 /* Kick pending requests. */
1930 spin_lock_irq(&info->io_lock);
1931 info->connected = BLKIF_STATE_CONNECTED;
1932 kick_pending_request_queues(info);
1933 spin_unlock_irq(&info->io_lock);
1934
1935 add_disk(info->gd);
1936
1937 info->is_ready = 1;
1938 }
1939
1940 /**
1941 * Callback received when the backend's state changes.
1942 */
1943 static void blkback_changed(struct xenbus_device *dev,
1944 enum xenbus_state backend_state)
1945 {
1946 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1947
1948 dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1949
1950 switch (backend_state) {
1951 case XenbusStateInitialising:
1952 case XenbusStateInitWait:
1953 case XenbusStateInitialised:
1954 case XenbusStateReconfiguring:
1955 case XenbusStateReconfigured:
1956 case XenbusStateUnknown:
1957 break;
1958
1959 case XenbusStateConnected:
1960 blkfront_connect(info);
1961 break;
1962
1963 case XenbusStateClosed:
1964 if (dev->state == XenbusStateClosed)
1965 break;
1966 /* Missed the backend's Closing state -- fallthrough */
1967 case XenbusStateClosing:
1968 if (info)
1969 blkfront_closing(info);
1970 break;
1971 }
1972 }
1973
1974 static int blkfront_remove(struct xenbus_device *xbdev)
1975 {
1976 struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
1977 struct block_device *bdev = NULL;
1978 struct gendisk *disk;
1979
1980 dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
1981
1982 blkif_free(info, 0);
1983
1984 mutex_lock(&info->mutex);
1985
1986 disk = info->gd;
1987 if (disk)
1988 bdev = bdget_disk(disk, 0);
1989
1990 info->xbdev = NULL;
1991 mutex_unlock(&info->mutex);
1992
1993 if (!bdev) {
1994 kfree(info);
1995 return 0;
1996 }
1997
1998 /*
1999 * The xbdev was removed before we reached the Closed
2000 * state. See if it's safe to remove the disk. If the bdev
2001 * isn't closed yet, we let release take care of it.
2002 */
2003
2004 mutex_lock(&bdev->bd_mutex);
2005 info = disk->private_data;
2006
2007 dev_warn(disk_to_dev(disk),
2008 "%s was hot-unplugged, %d stale handles\n",
2009 xbdev->nodename, bdev->bd_openers);
2010
2011 if (info && !bdev->bd_openers) {
2012 xlvbd_release_gendisk(info);
2013 disk->private_data = NULL;
2014 kfree(info);
2015 }
2016
2017 mutex_unlock(&bdev->bd_mutex);
2018 bdput(bdev);
2019
2020 return 0;
2021 }
2022
2023 static int blkfront_is_ready(struct xenbus_device *dev)
2024 {
2025 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2026
2027 return info->is_ready && info->xbdev;
2028 }
2029
2030 static int blkif_open(struct block_device *bdev, fmode_t mode)
2031 {
2032 struct gendisk *disk = bdev->bd_disk;
2033 struct blkfront_info *info;
2034 int err = 0;
2035
2036 mutex_lock(&blkfront_mutex);
2037
2038 info = disk->private_data;
2039 if (!info) {
2040 /* xbdev gone */
2041 err = -ERESTARTSYS;
2042 goto out;
2043 }
2044
2045 mutex_lock(&info->mutex);
2046
2047 if (!info->gd)
2048 /* xbdev is closed */
2049 err = -ERESTARTSYS;
2050
2051 mutex_unlock(&info->mutex);
2052
2053 out:
2054 mutex_unlock(&blkfront_mutex);
2055 return err;
2056 }
2057
2058 static void blkif_release(struct gendisk *disk, fmode_t mode)
2059 {
2060 struct blkfront_info *info = disk->private_data;
2061 struct block_device *bdev;
2062 struct xenbus_device *xbdev;
2063
2064 mutex_lock(&blkfront_mutex);
2065
2066 bdev = bdget_disk(disk, 0);
2067
2068 if (bdev->bd_openers)
2069 goto out;
2070
2071 /*
2072 * Check if we have been instructed to close. We will have
2073 * deferred this request, because the bdev was still open.
2074 */
2075
2076 mutex_lock(&info->mutex);
2077 xbdev = info->xbdev;
2078
2079 if (xbdev && xbdev->state == XenbusStateClosing) {
2080 /* pending switch to state closed */
2081 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2082 xlvbd_release_gendisk(info);
2083 xenbus_frontend_closed(info->xbdev);
2084 }
2085
2086 mutex_unlock(&info->mutex);
2087
2088 if (!xbdev) {
2089 /* sudden device removal */
2090 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2091 xlvbd_release_gendisk(info);
2092 disk->private_data = NULL;
2093 kfree(info);
2094 }
2095
2096 out:
2097 bdput(bdev);
2098 mutex_unlock(&blkfront_mutex);
2099 }
2100
2101 static const struct block_device_operations xlvbd_block_fops =
2102 {
2103 .owner = THIS_MODULE,
2104 .open = blkif_open,
2105 .release = blkif_release,
2106 .getgeo = blkif_getgeo,
2107 .ioctl = blkif_ioctl,
2108 };
2109
2110
2111 static const struct xenbus_device_id blkfront_ids[] = {
2112 { "vbd" },
2113 { "" }
2114 };
2115
2116 static DEFINE_XENBUS_DRIVER(blkfront, ,
2117 .probe = blkfront_probe,
2118 .remove = blkfront_remove,
2119 .resume = blkfront_resume,
2120 .otherend_changed = blkback_changed,
2121 .is_ready = blkfront_is_ready,
2122 );
2123
2124 static int __init xlblk_init(void)
2125 {
2126 int ret;
2127
2128 if (!xen_domain())
2129 return -ENODEV;
2130
2131 if (!xen_has_pv_disk_devices())
2132 return -ENODEV;
2133
2134 if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2135 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
2136 XENVBD_MAJOR, DEV_NAME);
2137 return -ENODEV;
2138 }
2139
2140 ret = xenbus_register_frontend(&blkfront_driver);
2141 if (ret) {
2142 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2143 return ret;
2144 }
2145
2146 return 0;
2147 }
2148 module_init(xlblk_init);
2149
2150
2151 static void __exit xlblk_exit(void)
2152 {
2153 xenbus_unregister_driver(&blkfront_driver);
2154 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2155 kfree(minors);
2156 }
2157 module_exit(xlblk_exit);
2158
2159 MODULE_DESCRIPTION("Xen virtual block device frontend");
2160 MODULE_LICENSE("GPL");
2161 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2162 MODULE_ALIAS("xen:vbd");
2163 MODULE_ALIAS("xenblk");
Linux with added FPC-III support