projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
xen-blkfront: switch from llist to list
[deliverable/linux.git] / drivers / block / xen-blkfront.c
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 unsigned long frame[BLKIF_MAX_SEGMENTS_PER_REQUEST];
78 struct grant *grants_used[BLKIF_MAX_SEGMENTS_PER_REQUEST];
79 };
80
81 static DEFINE_MUTEX(blkfront_mutex);
82 static const struct block_device_operations xlvbd_block_fops;
83
84 #define BLK_RING_SIZE __CONST_RING_SIZE(blkif, PAGE_SIZE)
85
86 /*
87 * We have one of these per vbd, whether ide, scsi or 'other'. They
88 * hang in private_data off the gendisk structure. We may end up
89 * putting all kinds of interesting stuff here :-)
90 */
91 struct blkfront_info
92 {
93 spinlock_t io_lock;
94 struct mutex mutex;
95 struct xenbus_device *xbdev;
96 struct gendisk *gd;
97 int vdevice;
98 blkif_vdev_t handle;
99 enum blkif_state connected;
100 int ring_ref;
101 struct blkif_front_ring ring;
102 struct scatterlist sg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
103 unsigned int evtchn, irq;
104 struct request_queue *rq;
105 struct work_struct work;
106 struct gnttab_free_callback callback;
107 struct blk_shadow shadow[BLK_RING_SIZE];
108 struct list_head persistent_gnts;
109 unsigned int persistent_gnts_c;
110 unsigned long shadow_free;
111 unsigned int feature_flush;
112 unsigned int flush_op;
113 unsigned int feature_discard:1;
114 unsigned int feature_secdiscard:1;
115 unsigned int discard_granularity;
116 unsigned int discard_alignment;
117 unsigned int feature_persistent:1;
118 int is_ready;
119 };
120
121 static unsigned int nr_minors;
122 static unsigned long *minors;
123 static DEFINE_SPINLOCK(minor_lock);
124
125 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
126 (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
127 #define GRANT_INVALID_REF 0
128
129 #define PARTS_PER_DISK 16
130 #define PARTS_PER_EXT_DISK 256
131
132 #define BLKIF_MAJOR(dev) ((dev)>>8)
133 #define BLKIF_MINOR(dev) ((dev) & 0xff)
134
135 #define EXT_SHIFT 28
136 #define EXTENDED (1<<EXT_SHIFT)
137 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
138 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
139 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
140 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
141 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
142 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
143
144 #define DEV_NAME "xvd" /* name in /dev */
145
146 static int get_id_from_freelist(struct blkfront_info *info)
147 {
148 unsigned long free = info->shadow_free;
149 BUG_ON(free >= BLK_RING_SIZE);
150 info->shadow_free = info->shadow[free].req.u.rw.id;
151 info->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
152 return free;
153 }
154
155 static int add_id_to_freelist(struct blkfront_info *info,
156 unsigned long id)
157 {
158 if (info->shadow[id].req.u.rw.id != id)
159 return -EINVAL;
160 if (info->shadow[id].request == NULL)
161 return -EINVAL;
162 info->shadow[id].req.u.rw.id = info->shadow_free;
163 info->shadow[id].request = NULL;
164 info->shadow_free = id;
165 return 0;
166 }
167
168 static const char *op_name(int op)
169 {
170 static const char *const names[] = {
171 [BLKIF_OP_READ] = "read",
172 [BLKIF_OP_WRITE] = "write",
173 [BLKIF_OP_WRITE_BARRIER] = "barrier",
174 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
175 [BLKIF_OP_DISCARD] = "discard" };
176
177 if (op < 0 || op >= ARRAY_SIZE(names))
178 return "unknown";
179
180 if (!names[op])
181 return "reserved";
182
183 return names[op];
184 }
185 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
186 {
187 unsigned int end = minor + nr;
188 int rc;
189
190 if (end > nr_minors) {
191 unsigned long *bitmap, *old;
192
193 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
194 GFP_KERNEL);
195 if (bitmap == NULL)
196 return -ENOMEM;
197
198 spin_lock(&minor_lock);
199 if (end > nr_minors) {
200 old = minors;
201 memcpy(bitmap, minors,
202 BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
203 minors = bitmap;
204 nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
205 } else
206 old = bitmap;
207 spin_unlock(&minor_lock);
208 kfree(old);
209 }
210
211 spin_lock(&minor_lock);
212 if (find_next_bit(minors, end, minor) >= end) {
213 bitmap_set(minors, minor, nr);
214 rc = 0;
215 } else
216 rc = -EBUSY;
217 spin_unlock(&minor_lock);
218
219 return rc;
220 }
221
222 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
223 {
224 unsigned int end = minor + nr;
225
226 BUG_ON(end > nr_minors);
227 spin_lock(&minor_lock);
228 bitmap_clear(minors, minor, nr);
229 spin_unlock(&minor_lock);
230 }
231
232 static void blkif_restart_queue_callback(void *arg)
233 {
234 struct blkfront_info *info = (struct blkfront_info *)arg;
235 schedule_work(&info->work);
236 }
237
238 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
239 {
240 /* We don't have real geometry info, but let's at least return
241 values consistent with the size of the device */
242 sector_t nsect = get_capacity(bd->bd_disk);
243 sector_t cylinders = nsect;
244
245 hg->heads = 0xff;
246 hg->sectors = 0x3f;
247 sector_div(cylinders, hg->heads * hg->sectors);
248 hg->cylinders = cylinders;
249 if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
250 hg->cylinders = 0xffff;
251 return 0;
252 }
253
254 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
255 unsigned command, unsigned long argument)
256 {
257 struct blkfront_info *info = bdev->bd_disk->private_data;
258 int i;
259
260 dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
261 command, (long)argument);
262
263 switch (command) {
264 case CDROMMULTISESSION:
265 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
266 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
267 if (put_user(0, (char __user *)(argument + i)))
268 return -EFAULT;
269 return 0;
270
271 case CDROM_GET_CAPABILITY: {
272 struct gendisk *gd = info->gd;
273 if (gd->flags & GENHD_FL_CD)
274 return 0;
275 return -EINVAL;
276 }
277
278 default:
279 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
280 command);*/
281 return -EINVAL; /* same return as native Linux */
282 }
283
284 return 0;
285 }
286
287 /*
288 * Generate a Xen blkfront IO request from a blk layer request. Reads
289 * and writes are handled as expected.
290 *
291 * @req: a request struct
292 */
293 static int blkif_queue_request(struct request *req)
294 {
295 struct blkfront_info *info = req->rq_disk->private_data;
296 unsigned long buffer_mfn;
297 struct blkif_request *ring_req;
298 unsigned long id;
299 unsigned int fsect, lsect;
300 int i, ref;
301
302 /*
303 * Used to store if we are able to queue the request by just using
304 * existing persistent grants, or if we have to get new grants,
305 * as there are not sufficiently many free.
306 */
307 bool new_persistent_gnts;
308 grant_ref_t gref_head;
309 struct page *granted_page;
310 struct grant *gnt_list_entry = NULL;
311 struct scatterlist *sg;
312
313 if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
314 return 1;
315
316 /* Check if we have enought grants to allocate a requests */
317 if (info->persistent_gnts_c < BLKIF_MAX_SEGMENTS_PER_REQUEST) {
318 new_persistent_gnts = 1;
319 if (gnttab_alloc_grant_references(
320 BLKIF_MAX_SEGMENTS_PER_REQUEST - info->persistent_gnts_c,
321 &gref_head) < 0) {
322 gnttab_request_free_callback(
323 &info->callback,
324 blkif_restart_queue_callback,
325 info,
326 BLKIF_MAX_SEGMENTS_PER_REQUEST);
327 return 1;
328 }
329 } else
330 new_persistent_gnts = 0;
331
332 /* Fill out a communications ring structure. */
333 ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
334 id = get_id_from_freelist(info);
335 info->shadow[id].request = req;
336
337 ring_req->u.rw.id = id;
338 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
339 ring_req->u.rw.handle = info->handle;
340
341 ring_req->operation = rq_data_dir(req) ?
342 BLKIF_OP_WRITE : BLKIF_OP_READ;
343
344 if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) {
345 /*
346 * Ideally we can do an unordered flush-to-disk. In case the
347 * backend onlysupports barriers, use that. A barrier request
348 * a superset of FUA, so we can implement it the same
349 * way. (It's also a FLUSH+FUA, since it is
350 * guaranteed ordered WRT previous writes.)
351 */
352 ring_req->operation = info->flush_op;
353 }
354
355 if (unlikely(req->cmd_flags & (REQ_DISCARD | REQ_SECURE))) {
356 /* id, sector_number and handle are set above. */
357 ring_req->operation = BLKIF_OP_DISCARD;
358 ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
359 if ((req->cmd_flags & REQ_SECURE) && info->feature_secdiscard)
360 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
361 else
362 ring_req->u.discard.flag = 0;
363 } else {
364 ring_req->u.rw.nr_segments = blk_rq_map_sg(req->q, req,
365 info->sg);
366 BUG_ON(ring_req->u.rw.nr_segments >
367 BLKIF_MAX_SEGMENTS_PER_REQUEST);
368
369 for_each_sg(info->sg, sg, ring_req->u.rw.nr_segments, i) {
370 fsect = sg->offset >> 9;
371 lsect = fsect + (sg->length >> 9) - 1;
372
373 if (info->persistent_gnts_c) {
374 BUG_ON(list_empty(&info->persistent_gnts));
375 gnt_list_entry = list_first_entry(
376 &info->persistent_gnts,
377 struct grant, node);
378 list_del(&gnt_list_entry->node);
379
380 ref = gnt_list_entry->gref;
381 buffer_mfn = pfn_to_mfn(gnt_list_entry->pfn);
382 info->persistent_gnts_c--;
383 } else {
384 ref = gnttab_claim_grant_reference(&gref_head);
385 BUG_ON(ref == -ENOSPC);
386
387 gnt_list_entry =
388 kmalloc(sizeof(struct grant),
389 GFP_ATOMIC);
390 if (!gnt_list_entry)
391 return -ENOMEM;
392
393 granted_page = alloc_page(GFP_ATOMIC);
394 if (!granted_page) {
395 kfree(gnt_list_entry);
396 return -ENOMEM;
397 }
398
399 gnt_list_entry->pfn =
400 page_to_pfn(granted_page);
401 gnt_list_entry->gref = ref;
402
403 buffer_mfn = pfn_to_mfn(page_to_pfn(
404 granted_page));
405 gnttab_grant_foreign_access_ref(ref,
406 info->xbdev->otherend_id,
407 buffer_mfn, 0);
408 }
409
410 info->shadow[id].grants_used[i] = gnt_list_entry;
411
412 if (rq_data_dir(req)) {
413 char *bvec_data;
414 void *shared_data;
415
416 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
417
418 shared_data = kmap_atomic(
419 pfn_to_page(gnt_list_entry->pfn));
420 bvec_data = kmap_atomic(sg_page(sg));
421
422 /*
423 * this does not wipe data stored outside the
424 * range sg->offset..sg->offset+sg->length.
425 * Therefore, blkback *could* see data from
426 * previous requests. This is OK as long as
427 * persistent grants are shared with just one
428 * domain. It may need refactoring if this
429 * changes
430 */
431 memcpy(shared_data + sg->offset,
432 bvec_data + sg->offset,
433 sg->length);
434
435 kunmap_atomic(bvec_data);
436 kunmap_atomic(shared_data);
437 }
438
439 info->shadow[id].frame[i] = mfn_to_pfn(buffer_mfn);
440 ring_req->u.rw.seg[i] =
441 (struct blkif_request_segment) {
442 .gref = ref,
443 .first_sect = fsect,
444 .last_sect = lsect };
445 }
446 }
447
448 info->ring.req_prod_pvt++;
449
450 /* Keep a private copy so we can reissue requests when recovering. */
451 info->shadow[id].req = *ring_req;
452
453 if (new_persistent_gnts)
454 gnttab_free_grant_references(gref_head);
455
456 return 0;
457 }
458
459
460 static inline void flush_requests(struct blkfront_info *info)
461 {
462 int notify;
463
464 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
465
466 if (notify)
467 notify_remote_via_irq(info->irq);
468 }
469
470 /*
471 * do_blkif_request
472 * read a block; request is in a request queue
473 */
474 static void do_blkif_request(struct request_queue *rq)
475 {
476 struct blkfront_info *info = NULL;
477 struct request *req;
478 int queued;
479
480 pr_debug("Entered do_blkif_request\n");
481
482 queued = 0;
483
484 while ((req = blk_peek_request(rq)) != NULL) {
485 info = req->rq_disk->private_data;
486
487 if (RING_FULL(&info->ring))
488 goto wait;
489
490 blk_start_request(req);
491
492 if ((req->cmd_type != REQ_TYPE_FS) ||
493 ((req->cmd_flags & (REQ_FLUSH | REQ_FUA)) &&
494 !info->flush_op)) {
495 __blk_end_request_all(req, -EIO);
496 continue;
497 }
498
499 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
500 "(%u/%u) buffer:%p [%s]\n",
501 req, req->cmd, (unsigned long)blk_rq_pos(req),
502 blk_rq_cur_sectors(req), blk_rq_sectors(req),
503 req->buffer, rq_data_dir(req) ? "write" : "read");
504
505 if (blkif_queue_request(req)) {
506 blk_requeue_request(rq, req);
507 wait:
508 /* Avoid pointless unplugs. */
509 blk_stop_queue(rq);
510 break;
511 }
512
513 queued++;
514 }
515
516 if (queued != 0)
517 flush_requests(info);
518 }
519
520 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
521 {
522 struct request_queue *rq;
523 struct blkfront_info *info = gd->private_data;
524
525 rq = blk_init_queue(do_blkif_request, &info->io_lock);
526 if (rq == NULL)
527 return -1;
528
529 queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
530
531 if (info->feature_discard) {
532 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
533 blk_queue_max_discard_sectors(rq, get_capacity(gd));
534 rq->limits.discard_granularity = info->discard_granularity;
535 rq->limits.discard_alignment = info->discard_alignment;
536 if (info->feature_secdiscard)
537 queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD, rq);
538 }
539
540 /* Hard sector size and max sectors impersonate the equiv. hardware. */
541 blk_queue_logical_block_size(rq, sector_size);
542 blk_queue_max_hw_sectors(rq, 512);
543
544 /* Each segment in a request is up to an aligned page in size. */
545 blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
546 blk_queue_max_segment_size(rq, PAGE_SIZE);
547
548 /* Ensure a merged request will fit in a single I/O ring slot. */
549 blk_queue_max_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
550
551 /* Make sure buffer addresses are sector-aligned. */
552 blk_queue_dma_alignment(rq, 511);
553
554 /* Make sure we don't use bounce buffers. */
555 blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
556
557 gd->queue = rq;
558
559 return 0;
560 }
561
562
563 static void xlvbd_flush(struct blkfront_info *info)
564 {
565 blk_queue_flush(info->rq, info->feature_flush);
566 printk(KERN_INFO "blkfront: %s: %s: %s %s\n",
567 info->gd->disk_name,
568 info->flush_op == BLKIF_OP_WRITE_BARRIER ?
569 "barrier" : (info->flush_op == BLKIF_OP_FLUSH_DISKCACHE ?
570 "flush diskcache" : "barrier or flush"),
571 info->feature_flush ? "enabled" : "disabled",
572 info->feature_persistent ? "using persistent grants" : "");
573 }
574
575 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
576 {
577 int major;
578 major = BLKIF_MAJOR(vdevice);
579 *minor = BLKIF_MINOR(vdevice);
580 switch (major) {
581 case XEN_IDE0_MAJOR:
582 *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
583 *minor = ((*minor / 64) * PARTS_PER_DISK) +
584 EMULATED_HD_DISK_MINOR_OFFSET;
585 break;
586 case XEN_IDE1_MAJOR:
587 *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
588 *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
589 EMULATED_HD_DISK_MINOR_OFFSET;
590 break;
591 case XEN_SCSI_DISK0_MAJOR:
592 *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
593 *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
594 break;
595 case XEN_SCSI_DISK1_MAJOR:
596 case XEN_SCSI_DISK2_MAJOR:
597 case XEN_SCSI_DISK3_MAJOR:
598 case XEN_SCSI_DISK4_MAJOR:
599 case XEN_SCSI_DISK5_MAJOR:
600 case XEN_SCSI_DISK6_MAJOR:
601 case XEN_SCSI_DISK7_MAJOR:
602 *offset = (*minor / PARTS_PER_DISK) +
603 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
604 EMULATED_SD_DISK_NAME_OFFSET;
605 *minor = *minor +
606 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
607 EMULATED_SD_DISK_MINOR_OFFSET;
608 break;
609 case XEN_SCSI_DISK8_MAJOR:
610 case XEN_SCSI_DISK9_MAJOR:
611 case XEN_SCSI_DISK10_MAJOR:
612 case XEN_SCSI_DISK11_MAJOR:
613 case XEN_SCSI_DISK12_MAJOR:
614 case XEN_SCSI_DISK13_MAJOR:
615 case XEN_SCSI_DISK14_MAJOR:
616 case XEN_SCSI_DISK15_MAJOR:
617 *offset = (*minor / PARTS_PER_DISK) +
618 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
619 EMULATED_SD_DISK_NAME_OFFSET;
620 *minor = *minor +
621 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
622 EMULATED_SD_DISK_MINOR_OFFSET;
623 break;
624 case XENVBD_MAJOR:
625 *offset = *minor / PARTS_PER_DISK;
626 break;
627 default:
628 printk(KERN_WARNING "blkfront: your disk configuration is "
629 "incorrect, please use an xvd device instead\n");
630 return -ENODEV;
631 }
632 return 0;
633 }
634
635 static char *encode_disk_name(char *ptr, unsigned int n)
636 {
637 if (n >= 26)
638 ptr = encode_disk_name(ptr, n / 26 - 1);
639 *ptr = 'a' + n % 26;
640 return ptr + 1;
641 }
642
643 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
644 struct blkfront_info *info,
645 u16 vdisk_info, u16 sector_size)
646 {
647 struct gendisk *gd;
648 int nr_minors = 1;
649 int err;
650 unsigned int offset;
651 int minor;
652 int nr_parts;
653 char *ptr;
654
655 BUG_ON(info->gd != NULL);
656 BUG_ON(info->rq != NULL);
657
658 if ((info->vdevice>>EXT_SHIFT) > 1) {
659 /* this is above the extended range; something is wrong */
660 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
661 return -ENODEV;
662 }
663
664 if (!VDEV_IS_EXTENDED(info->vdevice)) {
665 err = xen_translate_vdev(info->vdevice, &minor, &offset);
666 if (err)
667 return err;
668 nr_parts = PARTS_PER_DISK;
669 } else {
670 minor = BLKIF_MINOR_EXT(info->vdevice);
671 nr_parts = PARTS_PER_EXT_DISK;
672 offset = minor / nr_parts;
673 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
674 printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
675 "emulated IDE disks,\n\t choose an xvd device name"
676 "from xvde on\n", info->vdevice);
677 }
678 if (minor >> MINORBITS) {
679 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
680 info->vdevice, minor);
681 return -ENODEV;
682 }
683
684 if ((minor % nr_parts) == 0)
685 nr_minors = nr_parts;
686
687 err = xlbd_reserve_minors(minor, nr_minors);
688 if (err)
689 goto out;
690 err = -ENODEV;
691
692 gd = alloc_disk(nr_minors);
693 if (gd == NULL)
694 goto release;
695
696 strcpy(gd->disk_name, DEV_NAME);
697 ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
698 BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
699 if (nr_minors > 1)
700 *ptr = 0;
701 else
702 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
703 "%d", minor & (nr_parts - 1));
704
705 gd->major = XENVBD_MAJOR;
706 gd->first_minor = minor;
707 gd->fops = &xlvbd_block_fops;
708 gd->private_data = info;
709 gd->driverfs_dev = &(info->xbdev->dev);
710 set_capacity(gd, capacity);
711
712 if (xlvbd_init_blk_queue(gd, sector_size)) {
713 del_gendisk(gd);
714 goto release;
715 }
716
717 info->rq = gd->queue;
718 info->gd = gd;
719
720 xlvbd_flush(info);
721
722 if (vdisk_info & VDISK_READONLY)
723 set_disk_ro(gd, 1);
724
725 if (vdisk_info & VDISK_REMOVABLE)
726 gd->flags |= GENHD_FL_REMOVABLE;
727
728 if (vdisk_info & VDISK_CDROM)
729 gd->flags |= GENHD_FL_CD;
730
731 return 0;
732
733 release:
734 xlbd_release_minors(minor, nr_minors);
735 out:
736 return err;
737 }
738
739 static void xlvbd_release_gendisk(struct blkfront_info *info)
740 {
741 unsigned int minor, nr_minors;
742 unsigned long flags;
743
744 if (info->rq == NULL)
745 return;
746
747 spin_lock_irqsave(&info->io_lock, flags);
748
749 /* No more blkif_request(). */
750 blk_stop_queue(info->rq);
751
752 /* No more gnttab callback work. */
753 gnttab_cancel_free_callback(&info->callback);
754 spin_unlock_irqrestore(&info->io_lock, flags);
755
756 /* Flush gnttab callback work. Must be done with no locks held. */
757 flush_work(&info->work);
758
759 del_gendisk(info->gd);
760
761 minor = info->gd->first_minor;
762 nr_minors = info->gd->minors;
763 xlbd_release_minors(minor, nr_minors);
764
765 blk_cleanup_queue(info->rq);
766 info->rq = NULL;
767
768 put_disk(info->gd);
769 info->gd = NULL;
770 }
771
772 static void kick_pending_request_queues(struct blkfront_info *info)
773 {
774 if (!RING_FULL(&info->ring)) {
775 /* Re-enable calldowns. */
776 blk_start_queue(info->rq);
777 /* Kick things off immediately. */
778 do_blkif_request(info->rq);
779 }
780 }
781
782 static void blkif_restart_queue(struct work_struct *work)
783 {
784 struct blkfront_info *info = container_of(work, struct blkfront_info, work);
785
786 spin_lock_irq(&info->io_lock);
787 if (info->connected == BLKIF_STATE_CONNECTED)
788 kick_pending_request_queues(info);
789 spin_unlock_irq(&info->io_lock);
790 }
791
792 static void blkif_free(struct blkfront_info *info, int suspend)
793 {
794 struct grant *persistent_gnt;
795 struct grant *n;
796
797 /* Prevent new requests being issued until we fix things up. */
798 spin_lock_irq(&info->io_lock);
799 info->connected = suspend ?
800 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
801 /* No more blkif_request(). */
802 if (info->rq)
803 blk_stop_queue(info->rq);
804
805 /* Remove all persistent grants */
806 if (info->persistent_gnts_c) {
807 list_for_each_entry_safe(persistent_gnt, n,
808 &info->persistent_gnts, node) {
809 list_del(&persistent_gnt->node);
810 gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
811 __free_page(pfn_to_page(persistent_gnt->pfn));
812 kfree(persistent_gnt);
813 info->persistent_gnts_c--;
814 }
815 BUG_ON(info->persistent_gnts_c != 0);
816 }
817
818 /* No more gnttab callback work. */
819 gnttab_cancel_free_callback(&info->callback);
820 spin_unlock_irq(&info->io_lock);
821
822 /* Flush gnttab callback work. Must be done with no locks held. */
823 flush_work(&info->work);
824
825 /* Free resources associated with old device channel. */
826 if (info->ring_ref != GRANT_INVALID_REF) {
827 gnttab_end_foreign_access(info->ring_ref, 0,
828 (unsigned long)info->ring.sring);
829 info->ring_ref = GRANT_INVALID_REF;
830 info->ring.sring = NULL;
831 }
832 if (info->irq)
833 unbind_from_irqhandler(info->irq, info);
834 info->evtchn = info->irq = 0;
835
836 }
837
838 static void blkif_completion(struct blk_shadow *s, struct blkfront_info *info,
839 struct blkif_response *bret)
840 {
841 int i = 0;
842 struct bio_vec *bvec;
843 struct req_iterator iter;
844 unsigned long flags;
845 char *bvec_data;
846 void *shared_data;
847 unsigned int offset = 0;
848
849 if (bret->operation == BLKIF_OP_READ) {
850 /*
851 * Copy the data received from the backend into the bvec.
852 * Since bv_offset can be different than 0, and bv_len different
853 * than PAGE_SIZE, we have to keep track of the current offset,
854 * to be sure we are copying the data from the right shared page.
855 */
856 rq_for_each_segment(bvec, s->request, iter) {
857 BUG_ON((bvec->bv_offset + bvec->bv_len) > PAGE_SIZE);
858 if (bvec->bv_offset < offset)
859 i++;
860 BUG_ON(i >= s->req.u.rw.nr_segments);
861 shared_data = kmap_atomic(
862 pfn_to_page(s->grants_used[i]->pfn));
863 bvec_data = bvec_kmap_irq(bvec, &flags);
864 memcpy(bvec_data, shared_data + bvec->bv_offset,
865 bvec->bv_len);
866 bvec_kunmap_irq(bvec_data, &flags);
867 kunmap_atomic(shared_data);
868 offset = bvec->bv_offset + bvec->bv_len;
869 }
870 }
871 /* Add the persistent grant into the list of free grants */
872 for (i = 0; i < s->req.u.rw.nr_segments; i++) {
873 list_add(&s->grants_used[i]->node, &info->persistent_gnts);
874 info->persistent_gnts_c++;
875 }
876 }
877
878 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
879 {
880 struct request *req;
881 struct blkif_response *bret;
882 RING_IDX i, rp;
883 unsigned long flags;
884 struct blkfront_info *info = (struct blkfront_info *)dev_id;
885 int error;
886
887 spin_lock_irqsave(&info->io_lock, flags);
888
889 if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
890 spin_unlock_irqrestore(&info->io_lock, flags);
891 return IRQ_HANDLED;
892 }
893
894 again:
895 rp = info->ring.sring->rsp_prod;
896 rmb(); /* Ensure we see queued responses up to 'rp'. */
897
898 for (i = info->ring.rsp_cons; i != rp; i++) {
899 unsigned long id;
900
901 bret = RING_GET_RESPONSE(&info->ring, i);
902 id = bret->id;
903 /*
904 * The backend has messed up and given us an id that we would
905 * never have given to it (we stamp it up to BLK_RING_SIZE -
906 * look in get_id_from_freelist.
907 */
908 if (id >= BLK_RING_SIZE) {
909 WARN(1, "%s: response to %s has incorrect id (%ld)\n",
910 info->gd->disk_name, op_name(bret->operation), id);
911 /* We can't safely get the 'struct request' as
912 * the id is busted. */
913 continue;
914 }
915 req = info->shadow[id].request;
916
917 if (bret->operation != BLKIF_OP_DISCARD)
918 blkif_completion(&info->shadow[id], info, bret);
919
920 if (add_id_to_freelist(info, id)) {
921 WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
922 info->gd->disk_name, op_name(bret->operation), id);
923 continue;
924 }
925
926 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
927 switch (bret->operation) {
928 case BLKIF_OP_DISCARD:
929 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
930 struct request_queue *rq = info->rq;
931 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
932 info->gd->disk_name, op_name(bret->operation));
933 error = -EOPNOTSUPP;
934 info->feature_discard = 0;
935 info->feature_secdiscard = 0;
936 queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
937 queue_flag_clear(QUEUE_FLAG_SECDISCARD, rq);
938 }
939 __blk_end_request_all(req, error);
940 break;
941 case BLKIF_OP_FLUSH_DISKCACHE:
942 case BLKIF_OP_WRITE_BARRIER:
943 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
944 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
945 info->gd->disk_name, op_name(bret->operation));
946 error = -EOPNOTSUPP;
947 }
948 if (unlikely(bret->status == BLKIF_RSP_ERROR &&
949 info->shadow[id].req.u.rw.nr_segments == 0)) {
950 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
951 info->gd->disk_name, op_name(bret->operation));
952 error = -EOPNOTSUPP;
953 }
954 if (unlikely(error)) {
955 if (error == -EOPNOTSUPP)
956 error = 0;
957 info->feature_flush = 0;
958 info->flush_op = 0;
959 xlvbd_flush(info);
960 }
961 /* fall through */
962 case BLKIF_OP_READ:
963 case BLKIF_OP_WRITE:
964 if (unlikely(bret->status != BLKIF_RSP_OKAY))
965 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
966 "request: %x\n", bret->status);
967
968 __blk_end_request_all(req, error);
969 break;
970 default:
971 BUG();
972 }
973 }
974
975 info->ring.rsp_cons = i;
976
977 if (i != info->ring.req_prod_pvt) {
978 int more_to_do;
979 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
980 if (more_to_do)
981 goto again;
982 } else
983 info->ring.sring->rsp_event = i + 1;
984
985 kick_pending_request_queues(info);
986
987 spin_unlock_irqrestore(&info->io_lock, flags);
988
989 return IRQ_HANDLED;
990 }
991
992
993 static int setup_blkring(struct xenbus_device *dev,
994 struct blkfront_info *info)
995 {
996 struct blkif_sring *sring;
997 int err;
998
999 info->ring_ref = GRANT_INVALID_REF;
1000
1001 sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
1002 if (!sring) {
1003 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1004 return -ENOMEM;
1005 }
1006 SHARED_RING_INIT(sring);
1007 FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
1008
1009 sg_init_table(info->sg, BLKIF_MAX_SEGMENTS_PER_REQUEST);
1010
1011 err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
1012 if (err < 0) {
1013 free_page((unsigned long)sring);
1014 info->ring.sring = NULL;
1015 goto fail;
1016 }
1017 info->ring_ref = err;
1018
1019 err = xenbus_alloc_evtchn(dev, &info->evtchn);
1020 if (err)
1021 goto fail;
1022
1023 err = bind_evtchn_to_irqhandler(info->evtchn, blkif_interrupt, 0,
1024 "blkif", info);
1025 if (err <= 0) {
1026 xenbus_dev_fatal(dev, err,
1027 "bind_evtchn_to_irqhandler failed");
1028 goto fail;
1029 }
1030 info->irq = err;
1031
1032 return 0;
1033 fail:
1034 blkif_free(info, 0);
1035 return err;
1036 }
1037
1038
1039 /* Common code used when first setting up, and when resuming. */
1040 static int talk_to_blkback(struct xenbus_device *dev,
1041 struct blkfront_info *info)
1042 {
1043 const char *message = NULL;
1044 struct xenbus_transaction xbt;
1045 int err;
1046
1047 /* Create shared ring, alloc event channel. */
1048 err = setup_blkring(dev, info);
1049 if (err)
1050 goto out;
1051
1052 again:
1053 err = xenbus_transaction_start(&xbt);
1054 if (err) {
1055 xenbus_dev_fatal(dev, err, "starting transaction");
1056 goto destroy_blkring;
1057 }
1058
1059 err = xenbus_printf(xbt, dev->nodename,
1060 "ring-ref", "%u", info->ring_ref);
1061 if (err) {
1062 message = "writing ring-ref";
1063 goto abort_transaction;
1064 }
1065 err = xenbus_printf(xbt, dev->nodename,
1066 "event-channel", "%u", info->evtchn);
1067 if (err) {
1068 message = "writing event-channel";
1069 goto abort_transaction;
1070 }
1071 err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1072 XEN_IO_PROTO_ABI_NATIVE);
1073 if (err) {
1074 message = "writing protocol";
1075 goto abort_transaction;
1076 }
1077 err = xenbus_printf(xbt, dev->nodename,
1078 "feature-persistent", "%u", 1);
1079 if (err)
1080 dev_warn(&dev->dev,
1081 "writing persistent grants feature to xenbus");
1082
1083 err = xenbus_transaction_end(xbt, 0);
1084 if (err) {
1085 if (err == -EAGAIN)
1086 goto again;
1087 xenbus_dev_fatal(dev, err, "completing transaction");
1088 goto destroy_blkring;
1089 }
1090
1091 xenbus_switch_state(dev, XenbusStateInitialised);
1092
1093 return 0;
1094
1095 abort_transaction:
1096 xenbus_transaction_end(xbt, 1);
1097 if (message)
1098 xenbus_dev_fatal(dev, err, "%s", message);
1099 destroy_blkring:
1100 blkif_free(info, 0);
1101 out:
1102 return err;
1103 }
1104
1105 /**
1106 * Entry point to this code when a new device is created. Allocate the basic
1107 * structures and the ring buffer for communication with the backend, and
1108 * inform the backend of the appropriate details for those. Switch to
1109 * Initialised state.
1110 */
1111 static int blkfront_probe(struct xenbus_device *dev,
1112 const struct xenbus_device_id *id)
1113 {
1114 int err, vdevice, i;
1115 struct blkfront_info *info;
1116
1117 /* FIXME: Use dynamic device id if this is not set. */
1118 err = xenbus_scanf(XBT_NIL, dev->nodename,
1119 "virtual-device", "%i", &vdevice);
1120 if (err != 1) {
1121 /* go looking in the extended area instead */
1122 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1123 "%i", &vdevice);
1124 if (err != 1) {
1125 xenbus_dev_fatal(dev, err, "reading virtual-device");
1126 return err;
1127 }
1128 }
1129
1130 if (xen_hvm_domain()) {
1131 char *type;
1132 int len;
1133 /* no unplug has been done: do not hook devices != xen vbds */
1134 if (xen_platform_pci_unplug & XEN_UNPLUG_UNNECESSARY) {
1135 int major;
1136
1137 if (!VDEV_IS_EXTENDED(vdevice))
1138 major = BLKIF_MAJOR(vdevice);
1139 else
1140 major = XENVBD_MAJOR;
1141
1142 if (major != XENVBD_MAJOR) {
1143 printk(KERN_INFO
1144 "%s: HVM does not support vbd %d as xen block device\n",
1145 __FUNCTION__, vdevice);
1146 return -ENODEV;
1147 }
1148 }
1149 /* do not create a PV cdrom device if we are an HVM guest */
1150 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1151 if (IS_ERR(type))
1152 return -ENODEV;
1153 if (strncmp(type, "cdrom", 5) == 0) {
1154 kfree(type);
1155 return -ENODEV;
1156 }
1157 kfree(type);
1158 }
1159 info = kzalloc(sizeof(*info), GFP_KERNEL);
1160 if (!info) {
1161 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1162 return -ENOMEM;
1163 }
1164
1165 mutex_init(&info->mutex);
1166 spin_lock_init(&info->io_lock);
1167 info->xbdev = dev;
1168 info->vdevice = vdevice;
1169 INIT_LIST_HEAD(&info->persistent_gnts);
1170 info->persistent_gnts_c = 0;
1171 info->connected = BLKIF_STATE_DISCONNECTED;
1172 INIT_WORK(&info->work, blkif_restart_queue);
1173
1174 for (i = 0; i < BLK_RING_SIZE; i++)
1175 info->shadow[i].req.u.rw.id = i+1;
1176 info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1177
1178 /* Front end dir is a number, which is used as the id. */
1179 info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1180 dev_set_drvdata(&dev->dev, info);
1181
1182 err = talk_to_blkback(dev, info);
1183 if (err) {
1184 kfree(info);
1185 dev_set_drvdata(&dev->dev, NULL);
1186 return err;
1187 }
1188
1189 return 0;
1190 }
1191
1192
1193 static int blkif_recover(struct blkfront_info *info)
1194 {
1195 int i;
1196 struct blkif_request *req;
1197 struct blk_shadow *copy;
1198 int j;
1199
1200 /* Stage 1: Make a safe copy of the shadow state. */
1201 copy = kmemdup(info->shadow, sizeof(info->shadow),
1202 GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
1203 if (!copy)
1204 return -ENOMEM;
1205
1206 /* Stage 2: Set up free list. */
1207 memset(&info->shadow, 0, sizeof(info->shadow));
1208 for (i = 0; i < BLK_RING_SIZE; i++)
1209 info->shadow[i].req.u.rw.id = i+1;
1210 info->shadow_free = info->ring.req_prod_pvt;
1211 info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1212
1213 /* Stage 3: Find pending requests and requeue them. */
1214 for (i = 0; i < BLK_RING_SIZE; i++) {
1215 /* Not in use? */
1216 if (!copy[i].request)
1217 continue;
1218
1219 /* Grab a request slot and copy shadow state into it. */
1220 req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
1221 *req = copy[i].req;
1222
1223 /* We get a new request id, and must reset the shadow state. */
1224 req->u.rw.id = get_id_from_freelist(info);
1225 memcpy(&info->shadow[req->u.rw.id], &copy[i], sizeof(copy[i]));
1226
1227 if (req->operation != BLKIF_OP_DISCARD) {
1228 /* Rewrite any grant references invalidated by susp/resume. */
1229 for (j = 0; j < req->u.rw.nr_segments; j++)
1230 gnttab_grant_foreign_access_ref(
1231 req->u.rw.seg[j].gref,
1232 info->xbdev->otherend_id,
1233 pfn_to_mfn(info->shadow[req->u.rw.id].frame[j]),
1234 0);
1235 }
1236 info->shadow[req->u.rw.id].req = *req;
1237
1238 info->ring.req_prod_pvt++;
1239 }
1240
1241 kfree(copy);
1242
1243 xenbus_switch_state(info->xbdev, XenbusStateConnected);
1244
1245 spin_lock_irq(&info->io_lock);
1246
1247 /* Now safe for us to use the shared ring */
1248 info->connected = BLKIF_STATE_CONNECTED;
1249
1250 /* Send off requeued requests */
1251 flush_requests(info);
1252
1253 /* Kick any other new requests queued since we resumed */
1254 kick_pending_request_queues(info);
1255
1256 spin_unlock_irq(&info->io_lock);
1257
1258 return 0;
1259 }
1260
1261 /**
1262 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1263 * driver restart. We tear down our blkif structure and recreate it, but
1264 * leave the device-layer structures intact so that this is transparent to the
1265 * rest of the kernel.
1266 */
1267 static int blkfront_resume(struct xenbus_device *dev)
1268 {
1269 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1270 int err;
1271
1272 dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
1273
1274 blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
1275
1276 err = talk_to_blkback(dev, info);
1277 if (info->connected == BLKIF_STATE_SUSPENDED && !err)
1278 err = blkif_recover(info);
1279
1280 return err;
1281 }
1282
1283 static void
1284 blkfront_closing(struct blkfront_info *info)
1285 {
1286 struct xenbus_device *xbdev = info->xbdev;
1287 struct block_device *bdev = NULL;
1288
1289 mutex_lock(&info->mutex);
1290
1291 if (xbdev->state == XenbusStateClosing) {
1292 mutex_unlock(&info->mutex);
1293 return;
1294 }
1295
1296 if (info->gd)
1297 bdev = bdget_disk(info->gd, 0);
1298
1299 mutex_unlock(&info->mutex);
1300
1301 if (!bdev) {
1302 xenbus_frontend_closed(xbdev);
1303 return;
1304 }
1305
1306 mutex_lock(&bdev->bd_mutex);
1307
1308 if (bdev->bd_openers) {
1309 xenbus_dev_error(xbdev, -EBUSY,
1310 "Device in use; refusing to close");
1311 xenbus_switch_state(xbdev, XenbusStateClosing);
1312 } else {
1313 xlvbd_release_gendisk(info);
1314 xenbus_frontend_closed(xbdev);
1315 }
1316
1317 mutex_unlock(&bdev->bd_mutex);
1318 bdput(bdev);
1319 }
1320
1321 static void blkfront_setup_discard(struct blkfront_info *info)
1322 {
1323 int err;
1324 char *type;
1325 unsigned int discard_granularity;
1326 unsigned int discard_alignment;
1327 unsigned int discard_secure;
1328
1329 type = xenbus_read(XBT_NIL, info->xbdev->otherend, "type", NULL);
1330 if (IS_ERR(type))
1331 return;
1332
1333 info->feature_secdiscard = 0;
1334 if (strncmp(type, "phy", 3) == 0) {
1335 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1336 "discard-granularity", "%u", &discard_granularity,
1337 "discard-alignment", "%u", &discard_alignment,
1338 NULL);
1339 if (!err) {
1340 info->feature_discard = 1;
1341 info->discard_granularity = discard_granularity;
1342 info->discard_alignment = discard_alignment;
1343 }
1344 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1345 "discard-secure", "%d", &discard_secure,
1346 NULL);
1347 if (!err)
1348 info->feature_secdiscard = discard_secure;
1349
1350 } else if (strncmp(type, "file", 4) == 0)
1351 info->feature_discard = 1;
1352
1353 kfree(type);
1354 }
1355
1356 /*
1357 * Invoked when the backend is finally 'ready' (and has told produced
1358 * the details about the physical device - #sectors, size, etc).
1359 */
1360 static void blkfront_connect(struct blkfront_info *info)
1361 {
1362 unsigned long long sectors;
1363 unsigned long sector_size;
1364 unsigned int binfo;
1365 int err;
1366 int barrier, flush, discard, persistent;
1367
1368 switch (info->connected) {
1369 case BLKIF_STATE_CONNECTED:
1370 /*
1371 * Potentially, the back-end may be signalling
1372 * a capacity change; update the capacity.
1373 */
1374 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1375 "sectors", "%Lu", &sectors);
1376 if (XENBUS_EXIST_ERR(err))
1377 return;
1378 printk(KERN_INFO "Setting capacity to %Lu\n",
1379 sectors);
1380 set_capacity(info->gd, sectors);
1381 revalidate_disk(info->gd);
1382
1383 /* fall through */
1384 case BLKIF_STATE_SUSPENDED:
1385 return;
1386
1387 default:
1388 break;
1389 }
1390
1391 dev_dbg(&info->xbdev->dev, "%s:%s.\n",
1392 __func__, info->xbdev->otherend);
1393
1394 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1395 "sectors", "%llu", &sectors,
1396 "info", "%u", &binfo,
1397 "sector-size", "%lu", &sector_size,
1398 NULL);
1399 if (err) {
1400 xenbus_dev_fatal(info->xbdev, err,
1401 "reading backend fields at %s",
1402 info->xbdev->otherend);
1403 return;
1404 }
1405
1406 info->feature_flush = 0;
1407 info->flush_op = 0;
1408
1409 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1410 "feature-barrier", "%d", &barrier,
1411 NULL);
1412
1413 /*
1414 * If there's no "feature-barrier" defined, then it means
1415 * we're dealing with a very old backend which writes
1416 * synchronously; nothing to do.
1417 *
1418 * If there are barriers, then we use flush.
1419 */
1420 if (!err && barrier) {
1421 info->feature_flush = REQ_FLUSH | REQ_FUA;
1422 info->flush_op = BLKIF_OP_WRITE_BARRIER;
1423 }
1424 /*
1425 * And if there is "feature-flush-cache" use that above
1426 * barriers.
1427 */
1428 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1429 "feature-flush-cache", "%d", &flush,
1430 NULL);
1431
1432 if (!err && flush) {
1433 info->feature_flush = REQ_FLUSH;
1434 info->flush_op = BLKIF_OP_FLUSH_DISKCACHE;
1435 }
1436
1437 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1438 "feature-discard", "%d", &discard,
1439 NULL);
1440
1441 if (!err && discard)
1442 blkfront_setup_discard(info);
1443
1444 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1445 "feature-persistent", "%u", &persistent,
1446 NULL);
1447 if (err)
1448 info->feature_persistent = 0;
1449 else
1450 info->feature_persistent = persistent;
1451
1452 err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size);
1453 if (err) {
1454 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
1455 info->xbdev->otherend);
1456 return;
1457 }
1458
1459 xenbus_switch_state(info->xbdev, XenbusStateConnected);
1460
1461 /* Kick pending requests. */
1462 spin_lock_irq(&info->io_lock);
1463 info->connected = BLKIF_STATE_CONNECTED;
1464 kick_pending_request_queues(info);
1465 spin_unlock_irq(&info->io_lock);
1466
1467 add_disk(info->gd);
1468
1469 info->is_ready = 1;
1470 }
1471
1472 /**
1473 * Callback received when the backend's state changes.
1474 */
1475 static void blkback_changed(struct xenbus_device *dev,
1476 enum xenbus_state backend_state)
1477 {
1478 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1479
1480 dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1481
1482 switch (backend_state) {
1483 case XenbusStateInitialising:
1484 case XenbusStateInitWait:
1485 case XenbusStateInitialised:
1486 case XenbusStateReconfiguring:
1487 case XenbusStateReconfigured:
1488 case XenbusStateUnknown:
1489 case XenbusStateClosed:
1490 break;
1491
1492 case XenbusStateConnected:
1493 blkfront_connect(info);
1494 break;
1495
1496 case XenbusStateClosing:
1497 blkfront_closing(info);
1498 break;
1499 }
1500 }
1501
1502 static int blkfront_remove(struct xenbus_device *xbdev)
1503 {
1504 struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
1505 struct block_device *bdev = NULL;
1506 struct gendisk *disk;
1507
1508 dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
1509
1510 blkif_free(info, 0);
1511
1512 mutex_lock(&info->mutex);
1513
1514 disk = info->gd;
1515 if (disk)
1516 bdev = bdget_disk(disk, 0);
1517
1518 info->xbdev = NULL;
1519 mutex_unlock(&info->mutex);
1520
1521 if (!bdev) {
1522 kfree(info);
1523 return 0;
1524 }
1525
1526 /*
1527 * The xbdev was removed before we reached the Closed
1528 * state. See if it's safe to remove the disk. If the bdev
1529 * isn't closed yet, we let release take care of it.
1530 */
1531
1532 mutex_lock(&bdev->bd_mutex);
1533 info = disk->private_data;
1534
1535 dev_warn(disk_to_dev(disk),
1536 "%s was hot-unplugged, %d stale handles\n",
1537 xbdev->nodename, bdev->bd_openers);
1538
1539 if (info && !bdev->bd_openers) {
1540 xlvbd_release_gendisk(info);
1541 disk->private_data = NULL;
1542 kfree(info);
1543 }
1544
1545 mutex_unlock(&bdev->bd_mutex);
1546 bdput(bdev);
1547
1548 return 0;
1549 }
1550
1551 static int blkfront_is_ready(struct xenbus_device *dev)
1552 {
1553 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1554
1555 return info->is_ready && info->xbdev;
1556 }
1557
1558 static int blkif_open(struct block_device *bdev, fmode_t mode)
1559 {
1560 struct gendisk *disk = bdev->bd_disk;
1561 struct blkfront_info *info;
1562 int err = 0;
1563
1564 mutex_lock(&blkfront_mutex);
1565
1566 info = disk->private_data;
1567 if (!info) {
1568 /* xbdev gone */
1569 err = -ERESTARTSYS;
1570 goto out;
1571 }
1572
1573 mutex_lock(&info->mutex);
1574
1575 if (!info->gd)
1576 /* xbdev is closed */
1577 err = -ERESTARTSYS;
1578
1579 mutex_unlock(&info->mutex);
1580
1581 out:
1582 mutex_unlock(&blkfront_mutex);
1583 return err;
1584 }
1585
1586 static int blkif_release(struct gendisk *disk, fmode_t mode)
1587 {
1588 struct blkfront_info *info = disk->private_data;
1589 struct block_device *bdev;
1590 struct xenbus_device *xbdev;
1591
1592 mutex_lock(&blkfront_mutex);
1593
1594 bdev = bdget_disk(disk, 0);
1595
1596 if (bdev->bd_openers)
1597 goto out;
1598
1599 /*
1600 * Check if we have been instructed to close. We will have
1601 * deferred this request, because the bdev was still open.
1602 */
1603
1604 mutex_lock(&info->mutex);
1605 xbdev = info->xbdev;
1606
1607 if (xbdev && xbdev->state == XenbusStateClosing) {
1608 /* pending switch to state closed */
1609 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1610 xlvbd_release_gendisk(info);
1611 xenbus_frontend_closed(info->xbdev);
1612 }
1613
1614 mutex_unlock(&info->mutex);
1615
1616 if (!xbdev) {
1617 /* sudden device removal */
1618 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1619 xlvbd_release_gendisk(info);
1620 disk->private_data = NULL;
1621 kfree(info);
1622 }
1623
1624 out:
1625 bdput(bdev);
1626 mutex_unlock(&blkfront_mutex);
1627 return 0;
1628 }
1629
1630 static const struct block_device_operations xlvbd_block_fops =
1631 {
1632 .owner = THIS_MODULE,
1633 .open = blkif_open,
1634 .release = blkif_release,
1635 .getgeo = blkif_getgeo,
1636 .ioctl = blkif_ioctl,
1637 };
1638
1639
1640 static const struct xenbus_device_id blkfront_ids[] = {
1641 { "vbd" },
1642 { "" }
1643 };
1644
1645 static DEFINE_XENBUS_DRIVER(blkfront, ,
1646 .probe = blkfront_probe,
1647 .remove = blkfront_remove,
1648 .resume = blkfront_resume,
1649 .otherend_changed = blkback_changed,
1650 .is_ready = blkfront_is_ready,
1651 );
1652
1653 static int __init xlblk_init(void)
1654 {
1655 int ret;
1656
1657 if (!xen_domain())
1658 return -ENODEV;
1659
1660 if (xen_hvm_domain() && !xen_platform_pci_unplug)
1661 return -ENODEV;
1662
1663 if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
1664 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
1665 XENVBD_MAJOR, DEV_NAME);
1666 return -ENODEV;
1667 }
1668
1669 ret = xenbus_register_frontend(&blkfront_driver);
1670 if (ret) {
1671 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
1672 return ret;
1673 }
1674
1675 return 0;
1676 }
1677 module_init(xlblk_init);
1678
1679
1680 static void __exit xlblk_exit(void)
1681 {
1682 xenbus_unregister_driver(&blkfront_driver);
1683 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
1684 kfree(minors);
1685 }
1686 module_exit(xlblk_exit);
1687
1688 MODULE_DESCRIPTION("Xen virtual block device frontend");
1689 MODULE_LICENSE("GPL");
1690 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
1691 MODULE_ALIAS("xen:vbd");
1692 MODULE_ALIAS("xenblk");
Linux kernel - Deliverables
This page took 0.065194 seconds and 6 git commands to generate.