Loading...
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
62enum blkif_state {
63 BLKIF_STATE_DISCONNECTED,
64 BLKIF_STATE_CONNECTED,
65 BLKIF_STATE_SUSPENDED,
66};
67
68struct grant {
69 grant_ref_t gref;
70 unsigned long pfn;
71 struct list_head node;
72};
73
74struct 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
82struct split_bio {
83 struct bio *bio;
84 atomic_t pending;
85 int err;
86};
87
88static DEFINE_MUTEX(blkfront_mutex);
89static 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
97static unsigned int xen_blkif_max_segments = 32;
98module_param_named(max, xen_blkif_max_segments, int, S_IRUGO);
99MODULE_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 */
108struct 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
139static unsigned int nr_minors;
140static unsigned long *minors;
141static 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))
166#define INDIRECT_GREFS(_segs) \
167 ((_segs + SEGS_PER_INDIRECT_FRAME - 1)/SEGS_PER_INDIRECT_FRAME)
168
169static int blkfront_setup_indirect(struct blkfront_info *info);
170
171static 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
180static 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
193static 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
220out_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
233static 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
264static 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}
281static 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
318static 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
328static void blkif_restart_queue_callback(void *arg)
329{
330 struct blkfront_info *info = (struct blkfront_info *)arg;
331 schedule_work(&info->work);
332}
333
334static 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
350static 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 */
389static 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 *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 = req->nr_phys_segments;
413 if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
414 /*
415 * If we are using indirect segments we need to account
416 * for the indirect grefs used in the request.
417 */
418 max_grefs += INDIRECT_GREFS(req->nr_phys_segments);
419
420 /* Check if we have enough grants to allocate a requests */
421 if (info->persistent_gnts_c < max_grefs) {
422 new_persistent_gnts = 1;
423 if (gnttab_alloc_grant_references(
424 max_grefs - info->persistent_gnts_c,
425 &gref_head) < 0) {
426 gnttab_request_free_callback(
427 &info->callback,
428 blkif_restart_queue_callback,
429 info,
430 max_grefs);
431 return 1;
432 }
433 } else
434 new_persistent_gnts = 0;
435
436 /* Fill out a communications ring structure. */
437 ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
438 id = get_id_from_freelist(info);
439 info->shadow[id].request = req;
440
441 if (unlikely(req->cmd_flags & (REQ_DISCARD | REQ_SECURE))) {
442 ring_req->operation = BLKIF_OP_DISCARD;
443 ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
444 ring_req->u.discard.id = id;
445 ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
446 if ((req->cmd_flags & REQ_SECURE) && info->feature_secdiscard)
447 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
448 else
449 ring_req->u.discard.flag = 0;
450 } else {
451 BUG_ON(info->max_indirect_segments == 0 &&
452 req->nr_phys_segments > BLKIF_MAX_SEGMENTS_PER_REQUEST);
453 BUG_ON(info->max_indirect_segments &&
454 req->nr_phys_segments > info->max_indirect_segments);
455 nseg = blk_rq_map_sg(req->q, req, info->shadow[id].sg);
456 ring_req->u.rw.id = id;
457 if (nseg > BLKIF_MAX_SEGMENTS_PER_REQUEST) {
458 /*
459 * The indirect operation can only be a BLKIF_OP_READ or
460 * BLKIF_OP_WRITE
461 */
462 BUG_ON(req->cmd_flags & (REQ_FLUSH | REQ_FUA));
463 ring_req->operation = BLKIF_OP_INDIRECT;
464 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
465 BLKIF_OP_WRITE : BLKIF_OP_READ;
466 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
467 ring_req->u.indirect.handle = info->handle;
468 ring_req->u.indirect.nr_segments = nseg;
469 } else {
470 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
471 ring_req->u.rw.handle = info->handle;
472 ring_req->operation = rq_data_dir(req) ?
473 BLKIF_OP_WRITE : BLKIF_OP_READ;
474 if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) {
475 /*
476 * Ideally we can do an unordered flush-to-disk. In case the
477 * backend onlysupports barriers, use that. A barrier request
478 * a superset of FUA, so we can implement it the same
479 * way. (It's also a FLUSH+FUA, since it is
480 * guaranteed ordered WRT previous writes.)
481 */
482 ring_req->operation = info->flush_op;
483 }
484 ring_req->u.rw.nr_segments = nseg;
485 }
486 for_each_sg(info->shadow[id].sg, sg, nseg, i) {
487 fsect = sg->offset >> 9;
488 lsect = fsect + (sg->length >> 9) - 1;
489
490 if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
491 (i % SEGS_PER_INDIRECT_FRAME == 0)) {
492 unsigned long uninitialized_var(pfn);
493
494 if (segments)
495 kunmap_atomic(segments);
496
497 n = i / SEGS_PER_INDIRECT_FRAME;
498 if (!info->feature_persistent) {
499 struct page *indirect_page;
500
501 /* Fetch a pre-allocated page to use for indirect grefs */
502 BUG_ON(list_empty(&info->indirect_pages));
503 indirect_page = list_first_entry(&info->indirect_pages,
504 struct page, lru);
505 list_del(&indirect_page->lru);
506 pfn = page_to_pfn(indirect_page);
507 }
508 gnt_list_entry = get_grant(&gref_head, pfn, info);
509 info->shadow[id].indirect_grants[n] = gnt_list_entry;
510 segments = kmap_atomic(pfn_to_page(gnt_list_entry->pfn));
511 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
512 }
513
514 gnt_list_entry = get_grant(&gref_head, page_to_pfn(sg_page(sg)), info);
515 ref = gnt_list_entry->gref;
516
517 info->shadow[id].grants_used[i] = gnt_list_entry;
518
519 if (rq_data_dir(req) && info->feature_persistent) {
520 char *bvec_data;
521 void *shared_data;
522
523 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
524
525 shared_data = kmap_atomic(pfn_to_page(gnt_list_entry->pfn));
526 bvec_data = kmap_atomic(sg_page(sg));
527
528 /*
529 * this does not wipe data stored outside the
530 * range sg->offset..sg->offset+sg->length.
531 * Therefore, blkback *could* see data from
532 * previous requests. This is OK as long as
533 * persistent grants are shared with just one
534 * domain. It may need refactoring if this
535 * changes
536 */
537 memcpy(shared_data + sg->offset,
538 bvec_data + sg->offset,
539 sg->length);
540
541 kunmap_atomic(bvec_data);
542 kunmap_atomic(shared_data);
543 }
544 if (ring_req->operation != BLKIF_OP_INDIRECT) {
545 ring_req->u.rw.seg[i] =
546 (struct blkif_request_segment) {
547 .gref = ref,
548 .first_sect = fsect,
549 .last_sect = lsect };
550 } else {
551 n = i % SEGS_PER_INDIRECT_FRAME;
552 segments[n] =
553 (struct blkif_request_segment) {
554 .gref = ref,
555 .first_sect = fsect,
556 .last_sect = lsect };
557 }
558 }
559 if (segments)
560 kunmap_atomic(segments);
561 }
562
563 info->ring.req_prod_pvt++;
564
565 /* Keep a private copy so we can reissue requests when recovering. */
566 info->shadow[id].req = *ring_req;
567
568 if (new_persistent_gnts)
569 gnttab_free_grant_references(gref_head);
570
571 return 0;
572}
573
574
575static inline void flush_requests(struct blkfront_info *info)
576{
577 int notify;
578
579 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
580
581 if (notify)
582 notify_remote_via_irq(info->irq);
583}
584
585/*
586 * do_blkif_request
587 * read a block; request is in a request queue
588 */
589static void do_blkif_request(struct request_queue *rq)
590{
591 struct blkfront_info *info = NULL;
592 struct request *req;
593 int queued;
594
595 pr_debug("Entered do_blkif_request\n");
596
597 queued = 0;
598
599 while ((req = blk_peek_request(rq)) != NULL) {
600 info = req->rq_disk->private_data;
601
602 if (RING_FULL(&info->ring))
603 goto wait;
604
605 blk_start_request(req);
606
607 if ((req->cmd_type != REQ_TYPE_FS) ||
608 ((req->cmd_flags & (REQ_FLUSH | REQ_FUA)) &&
609 !info->flush_op)) {
610 __blk_end_request_all(req, -EIO);
611 continue;
612 }
613
614 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
615 "(%u/%u) buffer:%p [%s]\n",
616 req, req->cmd, (unsigned long)blk_rq_pos(req),
617 blk_rq_cur_sectors(req), blk_rq_sectors(req),
618 req->buffer, rq_data_dir(req) ? "write" : "read");
619
620 if (blkif_queue_request(req)) {
621 blk_requeue_request(rq, req);
622wait:
623 /* Avoid pointless unplugs. */
624 blk_stop_queue(rq);
625 break;
626 }
627
628 queued++;
629 }
630
631 if (queued != 0)
632 flush_requests(info);
633}
634
635static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
636 unsigned int physical_sector_size,
637 unsigned int segments)
638{
639 struct request_queue *rq;
640 struct blkfront_info *info = gd->private_data;
641
642 rq = blk_init_queue(do_blkif_request, &info->io_lock);
643 if (rq == NULL)
644 return -1;
645
646 queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
647
648 if (info->feature_discard) {
649 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
650 blk_queue_max_discard_sectors(rq, get_capacity(gd));
651 rq->limits.discard_granularity = info->discard_granularity;
652 rq->limits.discard_alignment = info->discard_alignment;
653 if (info->feature_secdiscard)
654 queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD, rq);
655 }
656
657 /* Hard sector size and max sectors impersonate the equiv. hardware. */
658 blk_queue_logical_block_size(rq, sector_size);
659 blk_queue_physical_block_size(rq, physical_sector_size);
660 blk_queue_max_hw_sectors(rq, (segments * PAGE_SIZE) / 512);
661
662 /* Each segment in a request is up to an aligned page in size. */
663 blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
664 blk_queue_max_segment_size(rq, PAGE_SIZE);
665
666 /* Ensure a merged request will fit in a single I/O ring slot. */
667 blk_queue_max_segments(rq, segments);
668
669 /* Make sure buffer addresses are sector-aligned. */
670 blk_queue_dma_alignment(rq, 511);
671
672 /* Make sure we don't use bounce buffers. */
673 blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
674
675 gd->queue = rq;
676
677 return 0;
678}
679
680
681static void xlvbd_flush(struct blkfront_info *info)
682{
683 blk_queue_flush(info->rq, info->feature_flush);
684 printk(KERN_INFO "blkfront: %s: %s: %s %s %s %s %s\n",
685 info->gd->disk_name,
686 info->flush_op == BLKIF_OP_WRITE_BARRIER ?
687 "barrier" : (info->flush_op == BLKIF_OP_FLUSH_DISKCACHE ?
688 "flush diskcache" : "barrier or flush"),
689 info->feature_flush ? "enabled;" : "disabled;",
690 "persistent grants:",
691 info->feature_persistent ? "enabled;" : "disabled;",
692 "indirect descriptors:",
693 info->max_indirect_segments ? "enabled;" : "disabled;");
694}
695
696static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
697{
698 int major;
699 major = BLKIF_MAJOR(vdevice);
700 *minor = BLKIF_MINOR(vdevice);
701 switch (major) {
702 case XEN_IDE0_MAJOR:
703 *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
704 *minor = ((*minor / 64) * PARTS_PER_DISK) +
705 EMULATED_HD_DISK_MINOR_OFFSET;
706 break;
707 case XEN_IDE1_MAJOR:
708 *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
709 *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
710 EMULATED_HD_DISK_MINOR_OFFSET;
711 break;
712 case XEN_SCSI_DISK0_MAJOR:
713 *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
714 *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
715 break;
716 case XEN_SCSI_DISK1_MAJOR:
717 case XEN_SCSI_DISK2_MAJOR:
718 case XEN_SCSI_DISK3_MAJOR:
719 case XEN_SCSI_DISK4_MAJOR:
720 case XEN_SCSI_DISK5_MAJOR:
721 case XEN_SCSI_DISK6_MAJOR:
722 case XEN_SCSI_DISK7_MAJOR:
723 *offset = (*minor / PARTS_PER_DISK) +
724 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
725 EMULATED_SD_DISK_NAME_OFFSET;
726 *minor = *minor +
727 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
728 EMULATED_SD_DISK_MINOR_OFFSET;
729 break;
730 case XEN_SCSI_DISK8_MAJOR:
731 case XEN_SCSI_DISK9_MAJOR:
732 case XEN_SCSI_DISK10_MAJOR:
733 case XEN_SCSI_DISK11_MAJOR:
734 case XEN_SCSI_DISK12_MAJOR:
735 case XEN_SCSI_DISK13_MAJOR:
736 case XEN_SCSI_DISK14_MAJOR:
737 case XEN_SCSI_DISK15_MAJOR:
738 *offset = (*minor / PARTS_PER_DISK) +
739 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
740 EMULATED_SD_DISK_NAME_OFFSET;
741 *minor = *minor +
742 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
743 EMULATED_SD_DISK_MINOR_OFFSET;
744 break;
745 case XENVBD_MAJOR:
746 *offset = *minor / PARTS_PER_DISK;
747 break;
748 default:
749 printk(KERN_WARNING "blkfront: your disk configuration is "
750 "incorrect, please use an xvd device instead\n");
751 return -ENODEV;
752 }
753 return 0;
754}
755
756static char *encode_disk_name(char *ptr, unsigned int n)
757{
758 if (n >= 26)
759 ptr = encode_disk_name(ptr, n / 26 - 1);
760 *ptr = 'a' + n % 26;
761 return ptr + 1;
762}
763
764static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
765 struct blkfront_info *info,
766 u16 vdisk_info, u16 sector_size,
767 unsigned int physical_sector_size)
768{
769 struct gendisk *gd;
770 int nr_minors = 1;
771 int err;
772 unsigned int offset;
773 int minor;
774 int nr_parts;
775 char *ptr;
776
777 BUG_ON(info->gd != NULL);
778 BUG_ON(info->rq != NULL);
779
780 if ((info->vdevice>>EXT_SHIFT) > 1) {
781 /* this is above the extended range; something is wrong */
782 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
783 return -ENODEV;
784 }
785
786 if (!VDEV_IS_EXTENDED(info->vdevice)) {
787 err = xen_translate_vdev(info->vdevice, &minor, &offset);
788 if (err)
789 return err;
790 nr_parts = PARTS_PER_DISK;
791 } else {
792 minor = BLKIF_MINOR_EXT(info->vdevice);
793 nr_parts = PARTS_PER_EXT_DISK;
794 offset = minor / nr_parts;
795 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
796 printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
797 "emulated IDE disks,\n\t choose an xvd device name"
798 "from xvde on\n", info->vdevice);
799 }
800 if (minor >> MINORBITS) {
801 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
802 info->vdevice, minor);
803 return -ENODEV;
804 }
805
806 if ((minor % nr_parts) == 0)
807 nr_minors = nr_parts;
808
809 err = xlbd_reserve_minors(minor, nr_minors);
810 if (err)
811 goto out;
812 err = -ENODEV;
813
814 gd = alloc_disk(nr_minors);
815 if (gd == NULL)
816 goto release;
817
818 strcpy(gd->disk_name, DEV_NAME);
819 ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
820 BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
821 if (nr_minors > 1)
822 *ptr = 0;
823 else
824 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
825 "%d", minor & (nr_parts - 1));
826
827 gd->major = XENVBD_MAJOR;
828 gd->first_minor = minor;
829 gd->fops = &xlvbd_block_fops;
830 gd->private_data = info;
831 gd->driverfs_dev = &(info->xbdev->dev);
832 set_capacity(gd, capacity);
833
834 if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size,
835 info->max_indirect_segments ? :
836 BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
837 del_gendisk(gd);
838 goto release;
839 }
840
841 info->rq = gd->queue;
842 info->gd = gd;
843
844 xlvbd_flush(info);
845
846 if (vdisk_info & VDISK_READONLY)
847 set_disk_ro(gd, 1);
848
849 if (vdisk_info & VDISK_REMOVABLE)
850 gd->flags |= GENHD_FL_REMOVABLE;
851
852 if (vdisk_info & VDISK_CDROM)
853 gd->flags |= GENHD_FL_CD;
854
855 return 0;
856
857 release:
858 xlbd_release_minors(minor, nr_minors);
859 out:
860 return err;
861}
862
863static void xlvbd_release_gendisk(struct blkfront_info *info)
864{
865 unsigned int minor, nr_minors;
866 unsigned long flags;
867
868 if (info->rq == NULL)
869 return;
870
871 spin_lock_irqsave(&info->io_lock, flags);
872
873 /* No more blkif_request(). */
874 blk_stop_queue(info->rq);
875
876 /* No more gnttab callback work. */
877 gnttab_cancel_free_callback(&info->callback);
878 spin_unlock_irqrestore(&info->io_lock, flags);
879
880 /* Flush gnttab callback work. Must be done with no locks held. */
881 flush_work(&info->work);
882
883 del_gendisk(info->gd);
884
885 minor = info->gd->first_minor;
886 nr_minors = info->gd->minors;
887 xlbd_release_minors(minor, nr_minors);
888
889 blk_cleanup_queue(info->rq);
890 info->rq = NULL;
891
892 put_disk(info->gd);
893 info->gd = NULL;
894}
895
896static void kick_pending_request_queues(struct blkfront_info *info)
897{
898 if (!RING_FULL(&info->ring)) {
899 /* Re-enable calldowns. */
900 blk_start_queue(info->rq);
901 /* Kick things off immediately. */
902 do_blkif_request(info->rq);
903 }
904}
905
906static void blkif_restart_queue(struct work_struct *work)
907{
908 struct blkfront_info *info = container_of(work, struct blkfront_info, work);
909
910 spin_lock_irq(&info->io_lock);
911 if (info->connected == BLKIF_STATE_CONNECTED)
912 kick_pending_request_queues(info);
913 spin_unlock_irq(&info->io_lock);
914}
915
916static void blkif_free(struct blkfront_info *info, int suspend)
917{
918 struct grant *persistent_gnt;
919 struct grant *n;
920 int i, j, segs;
921
922 /* Prevent new requests being issued until we fix things up. */
923 spin_lock_irq(&info->io_lock);
924 info->connected = suspend ?
925 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
926 /* No more blkif_request(). */
927 if (info->rq)
928 blk_stop_queue(info->rq);
929
930 /* Remove all persistent grants */
931 if (!list_empty(&info->grants)) {
932 list_for_each_entry_safe(persistent_gnt, n,
933 &info->grants, node) {
934 list_del(&persistent_gnt->node);
935 if (persistent_gnt->gref != GRANT_INVALID_REF) {
936 gnttab_end_foreign_access(persistent_gnt->gref,
937 0, 0UL);
938 info->persistent_gnts_c--;
939 }
940 if (info->feature_persistent)
941 __free_page(pfn_to_page(persistent_gnt->pfn));
942 kfree(persistent_gnt);
943 }
944 }
945 BUG_ON(info->persistent_gnts_c != 0);
946
947 /*
948 * Remove indirect pages, this only happens when using indirect
949 * descriptors but not persistent grants
950 */
951 if (!list_empty(&info->indirect_pages)) {
952 struct page *indirect_page, *n;
953
954 BUG_ON(info->feature_persistent);
955 list_for_each_entry_safe(indirect_page, n, &info->indirect_pages, lru) {
956 list_del(&indirect_page->lru);
957 __free_page(indirect_page);
958 }
959 }
960
961 for (i = 0; i < BLK_RING_SIZE; i++) {
962 /*
963 * Clear persistent grants present in requests already
964 * on the shared ring
965 */
966 if (!info->shadow[i].request)
967 goto free_shadow;
968
969 segs = info->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
970 info->shadow[i].req.u.indirect.nr_segments :
971 info->shadow[i].req.u.rw.nr_segments;
972 for (j = 0; j < segs; j++) {
973 persistent_gnt = info->shadow[i].grants_used[j];
974 gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
975 if (info->feature_persistent)
976 __free_page(pfn_to_page(persistent_gnt->pfn));
977 kfree(persistent_gnt);
978 }
979
980 if (info->shadow[i].req.operation != BLKIF_OP_INDIRECT)
981 /*
982 * If this is not an indirect operation don't try to
983 * free indirect segments
984 */
985 goto free_shadow;
986
987 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
988 persistent_gnt = info->shadow[i].indirect_grants[j];
989 gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
990 __free_page(pfn_to_page(persistent_gnt->pfn));
991 kfree(persistent_gnt);
992 }
993
994free_shadow:
995 kfree(info->shadow[i].grants_used);
996 info->shadow[i].grants_used = NULL;
997 kfree(info->shadow[i].indirect_grants);
998 info->shadow[i].indirect_grants = NULL;
999 kfree(info->shadow[i].sg);
1000 info->shadow[i].sg = NULL;
1001 }
1002
1003 /* No more gnttab callback work. */
1004 gnttab_cancel_free_callback(&info->callback);
1005 spin_unlock_irq(&info->io_lock);
1006
1007 /* Flush gnttab callback work. Must be done with no locks held. */
1008 flush_work(&info->work);
1009
1010 /* Free resources associated with old device channel. */
1011 if (info->ring_ref != GRANT_INVALID_REF) {
1012 gnttab_end_foreign_access(info->ring_ref, 0,
1013 (unsigned long)info->ring.sring);
1014 info->ring_ref = GRANT_INVALID_REF;
1015 info->ring.sring = NULL;
1016 }
1017 if (info->irq)
1018 unbind_from_irqhandler(info->irq, info);
1019 info->evtchn = info->irq = 0;
1020
1021}
1022
1023static void blkif_completion(struct blk_shadow *s, struct blkfront_info *info,
1024 struct blkif_response *bret)
1025{
1026 int i = 0;
1027 struct scatterlist *sg;
1028 char *bvec_data;
1029 void *shared_data;
1030 int nseg;
1031
1032 nseg = s->req.operation == BLKIF_OP_INDIRECT ?
1033 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1034
1035 if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1036 /*
1037 * Copy the data received from the backend into the bvec.
1038 * Since bv_offset can be different than 0, and bv_len different
1039 * than PAGE_SIZE, we have to keep track of the current offset,
1040 * to be sure we are copying the data from the right shared page.
1041 */
1042 for_each_sg(s->sg, sg, nseg, i) {
1043 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1044 shared_data = kmap_atomic(
1045 pfn_to_page(s->grants_used[i]->pfn));
1046 bvec_data = kmap_atomic(sg_page(sg));
1047 memcpy(bvec_data + sg->offset,
1048 shared_data + sg->offset,
1049 sg->length);
1050 kunmap_atomic(bvec_data);
1051 kunmap_atomic(shared_data);
1052 }
1053 }
1054 /* Add the persistent grant into the list of free grants */
1055 for (i = 0; i < nseg; i++) {
1056 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1057 /*
1058 * If the grant is still mapped by the backend (the
1059 * backend has chosen to make this grant persistent)
1060 * we add it at the head of the list, so it will be
1061 * reused first.
1062 */
1063 if (!info->feature_persistent)
1064 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1065 s->grants_used[i]->gref);
1066 list_add(&s->grants_used[i]->node, &info->grants);
1067 info->persistent_gnts_c++;
1068 } else {
1069 /*
1070 * If the grant is not mapped by the backend we end the
1071 * foreign access and add it to the tail of the list,
1072 * so it will not be picked again unless we run out of
1073 * persistent grants.
1074 */
1075 gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1076 s->grants_used[i]->gref = GRANT_INVALID_REF;
1077 list_add_tail(&s->grants_used[i]->node, &info->grants);
1078 }
1079 }
1080 if (s->req.operation == BLKIF_OP_INDIRECT) {
1081 for (i = 0; i < INDIRECT_GREFS(nseg); i++) {
1082 if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1083 if (!info->feature_persistent)
1084 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1085 s->indirect_grants[i]->gref);
1086 list_add(&s->indirect_grants[i]->node, &info->grants);
1087 info->persistent_gnts_c++;
1088 } else {
1089 struct page *indirect_page;
1090
1091 gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1092 /*
1093 * Add the used indirect page back to the list of
1094 * available pages for indirect grefs.
1095 */
1096 indirect_page = pfn_to_page(s->indirect_grants[i]->pfn);
1097 list_add(&indirect_page->lru, &info->indirect_pages);
1098 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1099 list_add_tail(&s->indirect_grants[i]->node, &info->grants);
1100 }
1101 }
1102 }
1103}
1104
1105static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1106{
1107 struct request *req;
1108 struct blkif_response *bret;
1109 RING_IDX i, rp;
1110 unsigned long flags;
1111 struct blkfront_info *info = (struct blkfront_info *)dev_id;
1112 int error;
1113
1114 spin_lock_irqsave(&info->io_lock, flags);
1115
1116 if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
1117 spin_unlock_irqrestore(&info->io_lock, flags);
1118 return IRQ_HANDLED;
1119 }
1120
1121 again:
1122 rp = info->ring.sring->rsp_prod;
1123 rmb(); /* Ensure we see queued responses up to 'rp'. */
1124
1125 for (i = info->ring.rsp_cons; i != rp; i++) {
1126 unsigned long id;
1127
1128 bret = RING_GET_RESPONSE(&info->ring, i);
1129 id = bret->id;
1130 /*
1131 * The backend has messed up and given us an id that we would
1132 * never have given to it (we stamp it up to BLK_RING_SIZE -
1133 * look in get_id_from_freelist.
1134 */
1135 if (id >= BLK_RING_SIZE) {
1136 WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1137 info->gd->disk_name, op_name(bret->operation), id);
1138 /* We can't safely get the 'struct request' as
1139 * the id is busted. */
1140 continue;
1141 }
1142 req = info->shadow[id].request;
1143
1144 if (bret->operation != BLKIF_OP_DISCARD)
1145 blkif_completion(&info->shadow[id], info, bret);
1146
1147 if (add_id_to_freelist(info, id)) {
1148 WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1149 info->gd->disk_name, op_name(bret->operation), id);
1150 continue;
1151 }
1152
1153 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
1154 switch (bret->operation) {
1155 case BLKIF_OP_DISCARD:
1156 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1157 struct request_queue *rq = info->rq;
1158 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1159 info->gd->disk_name, op_name(bret->operation));
1160 error = -EOPNOTSUPP;
1161 info->feature_discard = 0;
1162 info->feature_secdiscard = 0;
1163 queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1164 queue_flag_clear(QUEUE_FLAG_SECDISCARD, rq);
1165 }
1166 __blk_end_request_all(req, error);
1167 break;
1168 case BLKIF_OP_FLUSH_DISKCACHE:
1169 case BLKIF_OP_WRITE_BARRIER:
1170 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1171 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1172 info->gd->disk_name, op_name(bret->operation));
1173 error = -EOPNOTSUPP;
1174 }
1175 if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1176 info->shadow[id].req.u.rw.nr_segments == 0)) {
1177 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1178 info->gd->disk_name, op_name(bret->operation));
1179 error = -EOPNOTSUPP;
1180 }
1181 if (unlikely(error)) {
1182 if (error == -EOPNOTSUPP)
1183 error = 0;
1184 info->feature_flush = 0;
1185 info->flush_op = 0;
1186 xlvbd_flush(info);
1187 }
1188 /* fall through */
1189 case BLKIF_OP_READ:
1190 case BLKIF_OP_WRITE:
1191 if (unlikely(bret->status != BLKIF_RSP_OKAY))
1192 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1193 "request: %x\n", bret->status);
1194
1195 __blk_end_request_all(req, error);
1196 break;
1197 default:
1198 BUG();
1199 }
1200 }
1201
1202 info->ring.rsp_cons = i;
1203
1204 if (i != info->ring.req_prod_pvt) {
1205 int more_to_do;
1206 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
1207 if (more_to_do)
1208 goto again;
1209 } else
1210 info->ring.sring->rsp_event = i + 1;
1211
1212 kick_pending_request_queues(info);
1213
1214 spin_unlock_irqrestore(&info->io_lock, flags);
1215
1216 return IRQ_HANDLED;
1217}
1218
1219
1220static int setup_blkring(struct xenbus_device *dev,
1221 struct blkfront_info *info)
1222{
1223 struct blkif_sring *sring;
1224 int err;
1225
1226 info->ring_ref = GRANT_INVALID_REF;
1227
1228 sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
1229 if (!sring) {
1230 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1231 return -ENOMEM;
1232 }
1233 SHARED_RING_INIT(sring);
1234 FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
1235
1236 err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
1237 if (err < 0) {
1238 free_page((unsigned long)sring);
1239 info->ring.sring = NULL;
1240 goto fail;
1241 }
1242 info->ring_ref = err;
1243
1244 err = xenbus_alloc_evtchn(dev, &info->evtchn);
1245 if (err)
1246 goto fail;
1247
1248 err = bind_evtchn_to_irqhandler(info->evtchn, blkif_interrupt, 0,
1249 "blkif", info);
1250 if (err <= 0) {
1251 xenbus_dev_fatal(dev, err,
1252 "bind_evtchn_to_irqhandler failed");
1253 goto fail;
1254 }
1255 info->irq = err;
1256
1257 return 0;
1258fail:
1259 blkif_free(info, 0);
1260 return err;
1261}
1262
1263
1264/* Common code used when first setting up, and when resuming. */
1265static int talk_to_blkback(struct xenbus_device *dev,
1266 struct blkfront_info *info)
1267{
1268 const char *message = NULL;
1269 struct xenbus_transaction xbt;
1270 int err;
1271
1272 /* Create shared ring, alloc event channel. */
1273 err = setup_blkring(dev, info);
1274 if (err)
1275 goto out;
1276
1277again:
1278 err = xenbus_transaction_start(&xbt);
1279 if (err) {
1280 xenbus_dev_fatal(dev, err, "starting transaction");
1281 goto destroy_blkring;
1282 }
1283
1284 err = xenbus_printf(xbt, dev->nodename,
1285 "ring-ref", "%u", info->ring_ref);
1286 if (err) {
1287 message = "writing ring-ref";
1288 goto abort_transaction;
1289 }
1290 err = xenbus_printf(xbt, dev->nodename,
1291 "event-channel", "%u", info->evtchn);
1292 if (err) {
1293 message = "writing event-channel";
1294 goto abort_transaction;
1295 }
1296 err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1297 XEN_IO_PROTO_ABI_NATIVE);
1298 if (err) {
1299 message = "writing protocol";
1300 goto abort_transaction;
1301 }
1302 err = xenbus_printf(xbt, dev->nodename,
1303 "feature-persistent", "%u", 1);
1304 if (err)
1305 dev_warn(&dev->dev,
1306 "writing persistent grants feature to xenbus");
1307
1308 err = xenbus_transaction_end(xbt, 0);
1309 if (err) {
1310 if (err == -EAGAIN)
1311 goto again;
1312 xenbus_dev_fatal(dev, err, "completing transaction");
1313 goto destroy_blkring;
1314 }
1315
1316 xenbus_switch_state(dev, XenbusStateInitialised);
1317
1318 return 0;
1319
1320 abort_transaction:
1321 xenbus_transaction_end(xbt, 1);
1322 if (message)
1323 xenbus_dev_fatal(dev, err, "%s", message);
1324 destroy_blkring:
1325 blkif_free(info, 0);
1326 out:
1327 return err;
1328}
1329
1330/**
1331 * Entry point to this code when a new device is created. Allocate the basic
1332 * structures and the ring buffer for communication with the backend, and
1333 * inform the backend of the appropriate details for those. Switch to
1334 * Initialised state.
1335 */
1336static int blkfront_probe(struct xenbus_device *dev,
1337 const struct xenbus_device_id *id)
1338{
1339 int err, vdevice, i;
1340 struct blkfront_info *info;
1341
1342 /* FIXME: Use dynamic device id if this is not set. */
1343 err = xenbus_scanf(XBT_NIL, dev->nodename,
1344 "virtual-device", "%i", &vdevice);
1345 if (err != 1) {
1346 /* go looking in the extended area instead */
1347 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1348 "%i", &vdevice);
1349 if (err != 1) {
1350 xenbus_dev_fatal(dev, err, "reading virtual-device");
1351 return err;
1352 }
1353 }
1354
1355 if (xen_hvm_domain()) {
1356 char *type;
1357 int len;
1358 /* no unplug has been done: do not hook devices != xen vbds */
1359 if (xen_has_pv_and_legacy_disk_devices()) {
1360 int major;
1361
1362 if (!VDEV_IS_EXTENDED(vdevice))
1363 major = BLKIF_MAJOR(vdevice);
1364 else
1365 major = XENVBD_MAJOR;
1366
1367 if (major != XENVBD_MAJOR) {
1368 printk(KERN_INFO
1369 "%s: HVM does not support vbd %d as xen block device\n",
1370 __FUNCTION__, vdevice);
1371 return -ENODEV;
1372 }
1373 }
1374 /* do not create a PV cdrom device if we are an HVM guest */
1375 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1376 if (IS_ERR(type))
1377 return -ENODEV;
1378 if (strncmp(type, "cdrom", 5) == 0) {
1379 kfree(type);
1380 return -ENODEV;
1381 }
1382 kfree(type);
1383 }
1384 info = kzalloc(sizeof(*info), GFP_KERNEL);
1385 if (!info) {
1386 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1387 return -ENOMEM;
1388 }
1389
1390 mutex_init(&info->mutex);
1391 spin_lock_init(&info->io_lock);
1392 info->xbdev = dev;
1393 info->vdevice = vdevice;
1394 INIT_LIST_HEAD(&info->grants);
1395 INIT_LIST_HEAD(&info->indirect_pages);
1396 info->persistent_gnts_c = 0;
1397 info->connected = BLKIF_STATE_DISCONNECTED;
1398 INIT_WORK(&info->work, blkif_restart_queue);
1399
1400 for (i = 0; i < BLK_RING_SIZE; i++)
1401 info->shadow[i].req.u.rw.id = i+1;
1402 info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1403
1404 /* Front end dir is a number, which is used as the id. */
1405 info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1406 dev_set_drvdata(&dev->dev, info);
1407
1408 err = talk_to_blkback(dev, info);
1409 if (err) {
1410 kfree(info);
1411 dev_set_drvdata(&dev->dev, NULL);
1412 return err;
1413 }
1414
1415 return 0;
1416}
1417
1418static void split_bio_end(struct bio *bio, int error)
1419{
1420 struct split_bio *split_bio = bio->bi_private;
1421
1422 if (error)
1423 split_bio->err = error;
1424
1425 if (atomic_dec_and_test(&split_bio->pending)) {
1426 split_bio->bio->bi_phys_segments = 0;
1427 bio_endio(split_bio->bio, split_bio->err);
1428 kfree(split_bio);
1429 }
1430 bio_put(bio);
1431}
1432
1433static int blkif_recover(struct blkfront_info *info)
1434{
1435 int i;
1436 struct request *req, *n;
1437 struct blk_shadow *copy;
1438 int rc;
1439 struct bio *bio, *cloned_bio;
1440 struct bio_list bio_list, merge_bio;
1441 unsigned int segs, offset;
1442 int pending, size;
1443 struct split_bio *split_bio;
1444 struct list_head requests;
1445
1446 /* Stage 1: Make a safe copy of the shadow state. */
1447 copy = kmemdup(info->shadow, sizeof(info->shadow),
1448 GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
1449 if (!copy)
1450 return -ENOMEM;
1451
1452 /* Stage 2: Set up free list. */
1453 memset(&info->shadow, 0, sizeof(info->shadow));
1454 for (i = 0; i < BLK_RING_SIZE; i++)
1455 info->shadow[i].req.u.rw.id = i+1;
1456 info->shadow_free = info->ring.req_prod_pvt;
1457 info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1458
1459 rc = blkfront_setup_indirect(info);
1460 if (rc) {
1461 kfree(copy);
1462 return rc;
1463 }
1464
1465 segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
1466 blk_queue_max_segments(info->rq, segs);
1467 bio_list_init(&bio_list);
1468 INIT_LIST_HEAD(&requests);
1469 for (i = 0; i < BLK_RING_SIZE; i++) {
1470 /* Not in use? */
1471 if (!copy[i].request)
1472 continue;
1473
1474 /*
1475 * Get the bios in the request so we can re-queue them.
1476 */
1477 if (copy[i].request->cmd_flags &
1478 (REQ_FLUSH | REQ_FUA | REQ_DISCARD | REQ_SECURE)) {
1479 /*
1480 * Flush operations don't contain bios, so
1481 * we need to requeue the whole request
1482 */
1483 list_add(©[i].request->queuelist, &requests);
1484 continue;
1485 }
1486 merge_bio.head = copy[i].request->bio;
1487 merge_bio.tail = copy[i].request->biotail;
1488 bio_list_merge(&bio_list, &merge_bio);
1489 copy[i].request->bio = NULL;
1490 blk_put_request(copy[i].request);
1491 }
1492
1493 kfree(copy);
1494
1495 /*
1496 * Empty the queue, this is important because we might have
1497 * requests in the queue with more segments than what we
1498 * can handle now.
1499 */
1500 spin_lock_irq(&info->io_lock);
1501 while ((req = blk_fetch_request(info->rq)) != NULL) {
1502 if (req->cmd_flags &
1503 (REQ_FLUSH | REQ_FUA | REQ_DISCARD | REQ_SECURE)) {
1504 list_add(&req->queuelist, &requests);
1505 continue;
1506 }
1507 merge_bio.head = req->bio;
1508 merge_bio.tail = req->biotail;
1509 bio_list_merge(&bio_list, &merge_bio);
1510 req->bio = NULL;
1511 if (req->cmd_flags & (REQ_FLUSH | REQ_FUA))
1512 pr_alert("diskcache flush request found!\n");
1513 __blk_put_request(info->rq, req);
1514 }
1515 spin_unlock_irq(&info->io_lock);
1516
1517 xenbus_switch_state(info->xbdev, XenbusStateConnected);
1518
1519 spin_lock_irq(&info->io_lock);
1520
1521 /* Now safe for us to use the shared ring */
1522 info->connected = BLKIF_STATE_CONNECTED;
1523
1524 /* Kick any other new requests queued since we resumed */
1525 kick_pending_request_queues(info);
1526
1527 list_for_each_entry_safe(req, n, &requests, queuelist) {
1528 /* Requeue pending requests (flush or discard) */
1529 list_del_init(&req->queuelist);
1530 BUG_ON(req->nr_phys_segments > segs);
1531 blk_requeue_request(info->rq, req);
1532 }
1533 spin_unlock_irq(&info->io_lock);
1534
1535 while ((bio = bio_list_pop(&bio_list)) != NULL) {
1536 /* Traverse the list of pending bios and re-queue them */
1537 if (bio_segments(bio) > segs) {
1538 /*
1539 * This bio has more segments than what we can
1540 * handle, we have to split it.
1541 */
1542 pending = (bio_segments(bio) + segs - 1) / segs;
1543 split_bio = kzalloc(sizeof(*split_bio), GFP_NOIO);
1544 BUG_ON(split_bio == NULL);
1545 atomic_set(&split_bio->pending, pending);
1546 split_bio->bio = bio;
1547 for (i = 0; i < pending; i++) {
1548 offset = (i * segs * PAGE_SIZE) >> 9;
1549 size = min((unsigned int)(segs * PAGE_SIZE) >> 9,
1550 (unsigned int)bio_sectors(bio) - offset);
1551 cloned_bio = bio_clone(bio, GFP_NOIO);
1552 BUG_ON(cloned_bio == NULL);
1553 bio_trim(cloned_bio, offset, size);
1554 cloned_bio->bi_private = split_bio;
1555 cloned_bio->bi_end_io = split_bio_end;
1556 submit_bio(cloned_bio->bi_rw, cloned_bio);
1557 }
1558 /*
1559 * Now we have to wait for all those smaller bios to
1560 * end, so we can also end the "parent" bio.
1561 */
1562 continue;
1563 }
1564 /* We don't need to split this bio */
1565 submit_bio(bio->bi_rw, bio);
1566 }
1567
1568 return 0;
1569}
1570
1571/**
1572 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1573 * driver restart. We tear down our blkif structure and recreate it, but
1574 * leave the device-layer structures intact so that this is transparent to the
1575 * rest of the kernel.
1576 */
1577static int blkfront_resume(struct xenbus_device *dev)
1578{
1579 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1580 int err;
1581
1582 dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
1583
1584 blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
1585
1586 err = talk_to_blkback(dev, info);
1587
1588 /*
1589 * We have to wait for the backend to switch to
1590 * connected state, since we want to read which
1591 * features it supports.
1592 */
1593
1594 return err;
1595}
1596
1597static void
1598blkfront_closing(struct blkfront_info *info)
1599{
1600 struct xenbus_device *xbdev = info->xbdev;
1601 struct block_device *bdev = NULL;
1602
1603 mutex_lock(&info->mutex);
1604
1605 if (xbdev->state == XenbusStateClosing) {
1606 mutex_unlock(&info->mutex);
1607 return;
1608 }
1609
1610 if (info->gd)
1611 bdev = bdget_disk(info->gd, 0);
1612
1613 mutex_unlock(&info->mutex);
1614
1615 if (!bdev) {
1616 xenbus_frontend_closed(xbdev);
1617 return;
1618 }
1619
1620 mutex_lock(&bdev->bd_mutex);
1621
1622 if (bdev->bd_openers) {
1623 xenbus_dev_error(xbdev, -EBUSY,
1624 "Device in use; refusing to close");
1625 xenbus_switch_state(xbdev, XenbusStateClosing);
1626 } else {
1627 xlvbd_release_gendisk(info);
1628 xenbus_frontend_closed(xbdev);
1629 }
1630
1631 mutex_unlock(&bdev->bd_mutex);
1632 bdput(bdev);
1633}
1634
1635static void blkfront_setup_discard(struct blkfront_info *info)
1636{
1637 int err;
1638 char *type;
1639 unsigned int discard_granularity;
1640 unsigned int discard_alignment;
1641 unsigned int discard_secure;
1642
1643 type = xenbus_read(XBT_NIL, info->xbdev->otherend, "type", NULL);
1644 if (IS_ERR(type))
1645 return;
1646
1647 info->feature_secdiscard = 0;
1648 if (strncmp(type, "phy", 3) == 0) {
1649 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1650 "discard-granularity", "%u", &discard_granularity,
1651 "discard-alignment", "%u", &discard_alignment,
1652 NULL);
1653 if (!err) {
1654 info->feature_discard = 1;
1655 info->discard_granularity = discard_granularity;
1656 info->discard_alignment = discard_alignment;
1657 }
1658 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1659 "discard-secure", "%d", &discard_secure,
1660 NULL);
1661 if (!err)
1662 info->feature_secdiscard = discard_secure;
1663
1664 } else if (strncmp(type, "file", 4) == 0)
1665 info->feature_discard = 1;
1666
1667 kfree(type);
1668}
1669
1670static int blkfront_setup_indirect(struct blkfront_info *info)
1671{
1672 unsigned int indirect_segments, segs;
1673 int err, i;
1674
1675 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1676 "feature-max-indirect-segments", "%u", &indirect_segments,
1677 NULL);
1678 if (err) {
1679 info->max_indirect_segments = 0;
1680 segs = BLKIF_MAX_SEGMENTS_PER_REQUEST;
1681 } else {
1682 info->max_indirect_segments = min(indirect_segments,
1683 xen_blkif_max_segments);
1684 segs = info->max_indirect_segments;
1685 }
1686
1687 err = fill_grant_buffer(info, (segs + INDIRECT_GREFS(segs)) * BLK_RING_SIZE);
1688 if (err)
1689 goto out_of_memory;
1690
1691 if (!info->feature_persistent && info->max_indirect_segments) {
1692 /*
1693 * We are using indirect descriptors but not persistent
1694 * grants, we need to allocate a set of pages that can be
1695 * used for mapping indirect grefs
1696 */
1697 int num = INDIRECT_GREFS(segs) * BLK_RING_SIZE;
1698
1699 BUG_ON(!list_empty(&info->indirect_pages));
1700 for (i = 0; i < num; i++) {
1701 struct page *indirect_page = alloc_page(GFP_NOIO);
1702 if (!indirect_page)
1703 goto out_of_memory;
1704 list_add(&indirect_page->lru, &info->indirect_pages);
1705 }
1706 }
1707
1708 for (i = 0; i < BLK_RING_SIZE; i++) {
1709 info->shadow[i].grants_used = kzalloc(
1710 sizeof(info->shadow[i].grants_used[0]) * segs,
1711 GFP_NOIO);
1712 info->shadow[i].sg = kzalloc(sizeof(info->shadow[i].sg[0]) * segs, GFP_NOIO);
1713 if (info->max_indirect_segments)
1714 info->shadow[i].indirect_grants = kzalloc(
1715 sizeof(info->shadow[i].indirect_grants[0]) *
1716 INDIRECT_GREFS(segs),
1717 GFP_NOIO);
1718 if ((info->shadow[i].grants_used == NULL) ||
1719 (info->shadow[i].sg == NULL) ||
1720 (info->max_indirect_segments &&
1721 (info->shadow[i].indirect_grants == NULL)))
1722 goto out_of_memory;
1723 sg_init_table(info->shadow[i].sg, segs);
1724 }
1725
1726
1727 return 0;
1728
1729out_of_memory:
1730 for (i = 0; i < BLK_RING_SIZE; i++) {
1731 kfree(info->shadow[i].grants_used);
1732 info->shadow[i].grants_used = NULL;
1733 kfree(info->shadow[i].sg);
1734 info->shadow[i].sg = NULL;
1735 kfree(info->shadow[i].indirect_grants);
1736 info->shadow[i].indirect_grants = NULL;
1737 }
1738 if (!list_empty(&info->indirect_pages)) {
1739 struct page *indirect_page, *n;
1740 list_for_each_entry_safe(indirect_page, n, &info->indirect_pages, lru) {
1741 list_del(&indirect_page->lru);
1742 __free_page(indirect_page);
1743 }
1744 }
1745 return -ENOMEM;
1746}
1747
1748/*
1749 * Invoked when the backend is finally 'ready' (and has told produced
1750 * the details about the physical device - #sectors, size, etc).
1751 */
1752static void blkfront_connect(struct blkfront_info *info)
1753{
1754 unsigned long long sectors;
1755 unsigned long sector_size;
1756 unsigned int physical_sector_size;
1757 unsigned int binfo;
1758 int err;
1759 int barrier, flush, discard, persistent;
1760
1761 switch (info->connected) {
1762 case BLKIF_STATE_CONNECTED:
1763 /*
1764 * Potentially, the back-end may be signalling
1765 * a capacity change; update the capacity.
1766 */
1767 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1768 "sectors", "%Lu", §ors);
1769 if (XENBUS_EXIST_ERR(err))
1770 return;
1771 printk(KERN_INFO "Setting capacity to %Lu\n",
1772 sectors);
1773 set_capacity(info->gd, sectors);
1774 revalidate_disk(info->gd);
1775
1776 return;
1777 case BLKIF_STATE_SUSPENDED:
1778 /*
1779 * If we are recovering from suspension, we need to wait
1780 * for the backend to announce it's features before
1781 * reconnecting, at least we need to know if the backend
1782 * supports indirect descriptors, and how many.
1783 */
1784 blkif_recover(info);
1785 return;
1786
1787 default:
1788 break;
1789 }
1790
1791 dev_dbg(&info->xbdev->dev, "%s:%s.\n",
1792 __func__, info->xbdev->otherend);
1793
1794 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1795 "sectors", "%llu", §ors,
1796 "info", "%u", &binfo,
1797 "sector-size", "%lu", §or_size,
1798 NULL);
1799 if (err) {
1800 xenbus_dev_fatal(info->xbdev, err,
1801 "reading backend fields at %s",
1802 info->xbdev->otherend);
1803 return;
1804 }
1805
1806 /*
1807 * physcial-sector-size is a newer field, so old backends may not
1808 * provide this. Assume physical sector size to be the same as
1809 * sector_size in that case.
1810 */
1811 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1812 "physical-sector-size", "%u", &physical_sector_size);
1813 if (err != 1)
1814 physical_sector_size = sector_size;
1815
1816 info->feature_flush = 0;
1817 info->flush_op = 0;
1818
1819 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1820 "feature-barrier", "%d", &barrier,
1821 NULL);
1822
1823 /*
1824 * If there's no "feature-barrier" defined, then it means
1825 * we're dealing with a very old backend which writes
1826 * synchronously; nothing to do.
1827 *
1828 * If there are barriers, then we use flush.
1829 */
1830 if (!err && barrier) {
1831 info->feature_flush = REQ_FLUSH | REQ_FUA;
1832 info->flush_op = BLKIF_OP_WRITE_BARRIER;
1833 }
1834 /*
1835 * And if there is "feature-flush-cache" use that above
1836 * barriers.
1837 */
1838 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1839 "feature-flush-cache", "%d", &flush,
1840 NULL);
1841
1842 if (!err && flush) {
1843 info->feature_flush = REQ_FLUSH;
1844 info->flush_op = BLKIF_OP_FLUSH_DISKCACHE;
1845 }
1846
1847 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1848 "feature-discard", "%d", &discard,
1849 NULL);
1850
1851 if (!err && discard)
1852 blkfront_setup_discard(info);
1853
1854 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1855 "feature-persistent", "%u", &persistent,
1856 NULL);
1857 if (err)
1858 info->feature_persistent = 0;
1859 else
1860 info->feature_persistent = persistent;
1861
1862 err = blkfront_setup_indirect(info);
1863 if (err) {
1864 xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
1865 info->xbdev->otherend);
1866 return;
1867 }
1868
1869 err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
1870 physical_sector_size);
1871 if (err) {
1872 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
1873 info->xbdev->otherend);
1874 return;
1875 }
1876
1877 xenbus_switch_state(info->xbdev, XenbusStateConnected);
1878
1879 /* Kick pending requests. */
1880 spin_lock_irq(&info->io_lock);
1881 info->connected = BLKIF_STATE_CONNECTED;
1882 kick_pending_request_queues(info);
1883 spin_unlock_irq(&info->io_lock);
1884
1885 add_disk(info->gd);
1886
1887 info->is_ready = 1;
1888}
1889
1890/**
1891 * Callback received when the backend's state changes.
1892 */
1893static void blkback_changed(struct xenbus_device *dev,
1894 enum xenbus_state backend_state)
1895{
1896 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1897
1898 dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1899
1900 switch (backend_state) {
1901 case XenbusStateInitialising:
1902 case XenbusStateInitWait:
1903 case XenbusStateInitialised:
1904 case XenbusStateReconfiguring:
1905 case XenbusStateReconfigured:
1906 case XenbusStateUnknown:
1907 break;
1908
1909 case XenbusStateConnected:
1910 blkfront_connect(info);
1911 break;
1912
1913 case XenbusStateClosed:
1914 if (dev->state == XenbusStateClosed)
1915 break;
1916 /* Missed the backend's Closing state -- fallthrough */
1917 case XenbusStateClosing:
1918 blkfront_closing(info);
1919 break;
1920 }
1921}
1922
1923static int blkfront_remove(struct xenbus_device *xbdev)
1924{
1925 struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
1926 struct block_device *bdev = NULL;
1927 struct gendisk *disk;
1928
1929 dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
1930
1931 blkif_free(info, 0);
1932
1933 mutex_lock(&info->mutex);
1934
1935 disk = info->gd;
1936 if (disk)
1937 bdev = bdget_disk(disk, 0);
1938
1939 info->xbdev = NULL;
1940 mutex_unlock(&info->mutex);
1941
1942 if (!bdev) {
1943 kfree(info);
1944 return 0;
1945 }
1946
1947 /*
1948 * The xbdev was removed before we reached the Closed
1949 * state. See if it's safe to remove the disk. If the bdev
1950 * isn't closed yet, we let release take care of it.
1951 */
1952
1953 mutex_lock(&bdev->bd_mutex);
1954 info = disk->private_data;
1955
1956 dev_warn(disk_to_dev(disk),
1957 "%s was hot-unplugged, %d stale handles\n",
1958 xbdev->nodename, bdev->bd_openers);
1959
1960 if (info && !bdev->bd_openers) {
1961 xlvbd_release_gendisk(info);
1962 disk->private_data = NULL;
1963 kfree(info);
1964 }
1965
1966 mutex_unlock(&bdev->bd_mutex);
1967 bdput(bdev);
1968
1969 return 0;
1970}
1971
1972static int blkfront_is_ready(struct xenbus_device *dev)
1973{
1974 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1975
1976 return info->is_ready && info->xbdev;
1977}
1978
1979static int blkif_open(struct block_device *bdev, fmode_t mode)
1980{
1981 struct gendisk *disk = bdev->bd_disk;
1982 struct blkfront_info *info;
1983 int err = 0;
1984
1985 mutex_lock(&blkfront_mutex);
1986
1987 info = disk->private_data;
1988 if (!info) {
1989 /* xbdev gone */
1990 err = -ERESTARTSYS;
1991 goto out;
1992 }
1993
1994 mutex_lock(&info->mutex);
1995
1996 if (!info->gd)
1997 /* xbdev is closed */
1998 err = -ERESTARTSYS;
1999
2000 mutex_unlock(&info->mutex);
2001
2002out:
2003 mutex_unlock(&blkfront_mutex);
2004 return err;
2005}
2006
2007static void blkif_release(struct gendisk *disk, fmode_t mode)
2008{
2009 struct blkfront_info *info = disk->private_data;
2010 struct block_device *bdev;
2011 struct xenbus_device *xbdev;
2012
2013 mutex_lock(&blkfront_mutex);
2014
2015 bdev = bdget_disk(disk, 0);
2016
2017 if (!bdev) {
2018 WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2019 goto out_mutex;
2020 }
2021 if (bdev->bd_openers)
2022 goto out;
2023
2024 /*
2025 * Check if we have been instructed to close. We will have
2026 * deferred this request, because the bdev was still open.
2027 */
2028
2029 mutex_lock(&info->mutex);
2030 xbdev = info->xbdev;
2031
2032 if (xbdev && xbdev->state == XenbusStateClosing) {
2033 /* pending switch to state closed */
2034 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2035 xlvbd_release_gendisk(info);
2036 xenbus_frontend_closed(info->xbdev);
2037 }
2038
2039 mutex_unlock(&info->mutex);
2040
2041 if (!xbdev) {
2042 /* sudden device removal */
2043 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2044 xlvbd_release_gendisk(info);
2045 disk->private_data = NULL;
2046 kfree(info);
2047 }
2048
2049out:
2050 bdput(bdev);
2051out_mutex:
2052 mutex_unlock(&blkfront_mutex);
2053}
2054
2055static const struct block_device_operations xlvbd_block_fops =
2056{
2057 .owner = THIS_MODULE,
2058 .open = blkif_open,
2059 .release = blkif_release,
2060 .getgeo = blkif_getgeo,
2061 .ioctl = blkif_ioctl,
2062};
2063
2064
2065static const struct xenbus_device_id blkfront_ids[] = {
2066 { "vbd" },
2067 { "" }
2068};
2069
2070static DEFINE_XENBUS_DRIVER(blkfront, ,
2071 .probe = blkfront_probe,
2072 .remove = blkfront_remove,
2073 .resume = blkfront_resume,
2074 .otherend_changed = blkback_changed,
2075 .is_ready = blkfront_is_ready,
2076);
2077
2078static int __init xlblk_init(void)
2079{
2080 int ret;
2081
2082 if (!xen_domain())
2083 return -ENODEV;
2084
2085 if (!xen_has_pv_disk_devices())
2086 return -ENODEV;
2087
2088 if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2089 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
2090 XENVBD_MAJOR, DEV_NAME);
2091 return -ENODEV;
2092 }
2093
2094 ret = xenbus_register_frontend(&blkfront_driver);
2095 if (ret) {
2096 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2097 return ret;
2098 }
2099
2100 return 0;
2101}
2102module_init(xlblk_init);
2103
2104
2105static void __exit xlblk_exit(void)
2106{
2107 xenbus_unregister_driver(&blkfront_driver);
2108 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2109 kfree(minors);
2110}
2111module_exit(xlblk_exit);
2112
2113MODULE_DESCRIPTION("Xen virtual block device frontend");
2114MODULE_LICENSE("GPL");
2115MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2116MODULE_ALIAS("xen:vbd");
2117MODULE_ALIAS("xenblk");
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/blk-mq.h>
41#include <linux/hdreg.h>
42#include <linux/cdrom.h>
43#include <linux/module.h>
44#include <linux/slab.h>
45#include <linux/major.h>
46#include <linux/mutex.h>
47#include <linux/scatterlist.h>
48#include <linux/bitmap.h>
49#include <linux/list.h>
50#include <linux/workqueue.h>
51#include <linux/sched/mm.h>
52
53#include <xen/xen.h>
54#include <xen/xenbus.h>
55#include <xen/grant_table.h>
56#include <xen/events.h>
57#include <xen/page.h>
58#include <xen/platform_pci.h>
59
60#include <xen/interface/grant_table.h>
61#include <xen/interface/io/blkif.h>
62#include <xen/interface/io/protocols.h>
63
64#include <asm/xen/hypervisor.h>
65
66/*
67 * The minimal size of segment supported by the block framework is PAGE_SIZE.
68 * When Linux is using a different page size than Xen, it may not be possible
69 * to put all the data in a single segment.
70 * This can happen when the backend doesn't support indirect descriptor and
71 * therefore the maximum amount of data that a request can carry is
72 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE = 44KB
73 *
74 * Note that we only support one extra request. So the Linux page size
75 * should be <= ( 2 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) =
76 * 88KB.
77 */
78#define HAS_EXTRA_REQ (BLKIF_MAX_SEGMENTS_PER_REQUEST < XEN_PFN_PER_PAGE)
79
80enum blkif_state {
81 BLKIF_STATE_DISCONNECTED,
82 BLKIF_STATE_CONNECTED,
83 BLKIF_STATE_SUSPENDED,
84 BLKIF_STATE_ERROR,
85};
86
87struct grant {
88 grant_ref_t gref;
89 struct page *page;
90 struct list_head node;
91};
92
93enum blk_req_status {
94 REQ_PROCESSING,
95 REQ_WAITING,
96 REQ_DONE,
97 REQ_ERROR,
98 REQ_EOPNOTSUPP,
99};
100
101struct blk_shadow {
102 struct blkif_request req;
103 struct request *request;
104 struct grant **grants_used;
105 struct grant **indirect_grants;
106 struct scatterlist *sg;
107 unsigned int num_sg;
108 enum blk_req_status status;
109
110 #define NO_ASSOCIATED_ID ~0UL
111 /*
112 * Id of the sibling if we ever need 2 requests when handling a
113 * block I/O request
114 */
115 unsigned long associated_id;
116};
117
118struct blkif_req {
119 blk_status_t error;
120};
121
122static inline struct blkif_req *blkif_req(struct request *rq)
123{
124 return blk_mq_rq_to_pdu(rq);
125}
126
127static DEFINE_MUTEX(blkfront_mutex);
128static const struct block_device_operations xlvbd_block_fops;
129static struct delayed_work blkfront_work;
130static LIST_HEAD(info_list);
131
132/*
133 * Maximum number of segments in indirect requests, the actual value used by
134 * the frontend driver is the minimum of this value and the value provided
135 * by the backend driver.
136 */
137
138static unsigned int xen_blkif_max_segments = 32;
139module_param_named(max_indirect_segments, xen_blkif_max_segments, uint, 0444);
140MODULE_PARM_DESC(max_indirect_segments,
141 "Maximum amount of segments in indirect requests (default is 32)");
142
143static unsigned int xen_blkif_max_queues = 4;
144module_param_named(max_queues, xen_blkif_max_queues, uint, 0444);
145MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
146
147/*
148 * Maximum order of pages to be used for the shared ring between front and
149 * backend, 4KB page granularity is used.
150 */
151static unsigned int xen_blkif_max_ring_order;
152module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, 0444);
153MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
154
155static bool __read_mostly xen_blkif_trusted = true;
156module_param_named(trusted, xen_blkif_trusted, bool, 0644);
157MODULE_PARM_DESC(trusted, "Is the backend trusted");
158
159#define BLK_RING_SIZE(info) \
160 __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
161
162/*
163 * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
164 * characters are enough. Define to 20 to keep consistent with backend.
165 */
166#define RINGREF_NAME_LEN (20)
167/*
168 * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
169 */
170#define QUEUE_NAME_LEN (17)
171
172/*
173 * Per-ring info.
174 * Every blkfront device can associate with one or more blkfront_ring_info,
175 * depending on how many hardware queues/rings to be used.
176 */
177struct blkfront_ring_info {
178 /* Lock to protect data in every ring buffer. */
179 spinlock_t ring_lock;
180 struct blkif_front_ring ring;
181 unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
182 unsigned int evtchn, irq;
183 struct work_struct work;
184 struct gnttab_free_callback callback;
185 struct list_head indirect_pages;
186 struct list_head grants;
187 unsigned int persistent_gnts_c;
188 unsigned long shadow_free;
189 struct blkfront_info *dev_info;
190 struct blk_shadow shadow[];
191};
192
193/*
194 * We have one of these per vbd, whether ide, scsi or 'other'. They
195 * hang in private_data off the gendisk structure. We may end up
196 * putting all kinds of interesting stuff here :-)
197 */
198struct blkfront_info
199{
200 struct mutex mutex;
201 struct xenbus_device *xbdev;
202 struct gendisk *gd;
203 u16 sector_size;
204 unsigned int physical_sector_size;
205 unsigned long vdisk_info;
206 int vdevice;
207 blkif_vdev_t handle;
208 enum blkif_state connected;
209 /* Number of pages per ring buffer. */
210 unsigned int nr_ring_pages;
211 struct request_queue *rq;
212 unsigned int feature_flush:1;
213 unsigned int feature_fua:1;
214 unsigned int feature_discard:1;
215 unsigned int feature_secdiscard:1;
216 /* Connect-time cached feature_persistent parameter */
217 unsigned int feature_persistent_parm:1;
218 /* Persistent grants feature negotiation result */
219 unsigned int feature_persistent:1;
220 unsigned int bounce:1;
221 unsigned int discard_granularity;
222 unsigned int discard_alignment;
223 /* Number of 4KB segments handled */
224 unsigned int max_indirect_segments;
225 int is_ready;
226 struct blk_mq_tag_set tag_set;
227 struct blkfront_ring_info *rinfo;
228 unsigned int nr_rings;
229 unsigned int rinfo_size;
230 /* Save uncomplete reqs and bios for migration. */
231 struct list_head requests;
232 struct bio_list bio_list;
233 struct list_head info_list;
234};
235
236static unsigned int nr_minors;
237static unsigned long *minors;
238static DEFINE_SPINLOCK(minor_lock);
239
240#define PARTS_PER_DISK 16
241#define PARTS_PER_EXT_DISK 256
242
243#define BLKIF_MAJOR(dev) ((dev)>>8)
244#define BLKIF_MINOR(dev) ((dev) & 0xff)
245
246#define EXT_SHIFT 28
247#define EXTENDED (1<<EXT_SHIFT)
248#define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
249#define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
250#define EMULATED_HD_DISK_MINOR_OFFSET (0)
251#define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
252#define EMULATED_SD_DISK_MINOR_OFFSET (0)
253#define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
254
255#define DEV_NAME "xvd" /* name in /dev */
256
257/*
258 * Grants are always the same size as a Xen page (i.e 4KB).
259 * A physical segment is always the same size as a Linux page.
260 * Number of grants per physical segment
261 */
262#define GRANTS_PER_PSEG (PAGE_SIZE / XEN_PAGE_SIZE)
263
264#define GRANTS_PER_INDIRECT_FRAME \
265 (XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
266
267#define INDIRECT_GREFS(_grants) \
268 DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
269
270static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
271static void blkfront_gather_backend_features(struct blkfront_info *info);
272static int negotiate_mq(struct blkfront_info *info);
273
274#define for_each_rinfo(info, ptr, idx) \
275 for ((ptr) = (info)->rinfo, (idx) = 0; \
276 (idx) < (info)->nr_rings; \
277 (idx)++, (ptr) = (void *)(ptr) + (info)->rinfo_size)
278
279static inline struct blkfront_ring_info *
280get_rinfo(const struct blkfront_info *info, unsigned int i)
281{
282 BUG_ON(i >= info->nr_rings);
283 return (void *)info->rinfo + i * info->rinfo_size;
284}
285
286static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
287{
288 unsigned long free = rinfo->shadow_free;
289
290 BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
291 rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
292 rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
293 return free;
294}
295
296static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
297 unsigned long id)
298{
299 if (rinfo->shadow[id].req.u.rw.id != id)
300 return -EINVAL;
301 if (rinfo->shadow[id].request == NULL)
302 return -EINVAL;
303 rinfo->shadow[id].req.u.rw.id = rinfo->shadow_free;
304 rinfo->shadow[id].request = NULL;
305 rinfo->shadow_free = id;
306 return 0;
307}
308
309static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
310{
311 struct blkfront_info *info = rinfo->dev_info;
312 struct page *granted_page;
313 struct grant *gnt_list_entry, *n;
314 int i = 0;
315
316 while (i < num) {
317 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
318 if (!gnt_list_entry)
319 goto out_of_memory;
320
321 if (info->bounce) {
322 granted_page = alloc_page(GFP_NOIO | __GFP_ZERO);
323 if (!granted_page) {
324 kfree(gnt_list_entry);
325 goto out_of_memory;
326 }
327 gnt_list_entry->page = granted_page;
328 }
329
330 gnt_list_entry->gref = INVALID_GRANT_REF;
331 list_add(&gnt_list_entry->node, &rinfo->grants);
332 i++;
333 }
334
335 return 0;
336
337out_of_memory:
338 list_for_each_entry_safe(gnt_list_entry, n,
339 &rinfo->grants, node) {
340 list_del(&gnt_list_entry->node);
341 if (info->bounce)
342 __free_page(gnt_list_entry->page);
343 kfree(gnt_list_entry);
344 i--;
345 }
346 BUG_ON(i != 0);
347 return -ENOMEM;
348}
349
350static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
351{
352 struct grant *gnt_list_entry;
353
354 BUG_ON(list_empty(&rinfo->grants));
355 gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
356 node);
357 list_del(&gnt_list_entry->node);
358
359 if (gnt_list_entry->gref != INVALID_GRANT_REF)
360 rinfo->persistent_gnts_c--;
361
362 return gnt_list_entry;
363}
364
365static inline void grant_foreign_access(const struct grant *gnt_list_entry,
366 const struct blkfront_info *info)
367{
368 gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
369 info->xbdev->otherend_id,
370 gnt_list_entry->page,
371 0);
372}
373
374static struct grant *get_grant(grant_ref_t *gref_head,
375 unsigned long gfn,
376 struct blkfront_ring_info *rinfo)
377{
378 struct grant *gnt_list_entry = get_free_grant(rinfo);
379 struct blkfront_info *info = rinfo->dev_info;
380
381 if (gnt_list_entry->gref != INVALID_GRANT_REF)
382 return gnt_list_entry;
383
384 /* Assign a gref to this page */
385 gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
386 BUG_ON(gnt_list_entry->gref == -ENOSPC);
387 if (info->bounce)
388 grant_foreign_access(gnt_list_entry, info);
389 else {
390 /* Grant access to the GFN passed by the caller */
391 gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
392 info->xbdev->otherend_id,
393 gfn, 0);
394 }
395
396 return gnt_list_entry;
397}
398
399static struct grant *get_indirect_grant(grant_ref_t *gref_head,
400 struct blkfront_ring_info *rinfo)
401{
402 struct grant *gnt_list_entry = get_free_grant(rinfo);
403 struct blkfront_info *info = rinfo->dev_info;
404
405 if (gnt_list_entry->gref != INVALID_GRANT_REF)
406 return gnt_list_entry;
407
408 /* Assign a gref to this page */
409 gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
410 BUG_ON(gnt_list_entry->gref == -ENOSPC);
411 if (!info->bounce) {
412 struct page *indirect_page;
413
414 /* Fetch a pre-allocated page to use for indirect grefs */
415 BUG_ON(list_empty(&rinfo->indirect_pages));
416 indirect_page = list_first_entry(&rinfo->indirect_pages,
417 struct page, lru);
418 list_del(&indirect_page->lru);
419 gnt_list_entry->page = indirect_page;
420 }
421 grant_foreign_access(gnt_list_entry, info);
422
423 return gnt_list_entry;
424}
425
426static const char *op_name(int op)
427{
428 static const char *const names[] = {
429 [BLKIF_OP_READ] = "read",
430 [BLKIF_OP_WRITE] = "write",
431 [BLKIF_OP_WRITE_BARRIER] = "barrier",
432 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
433 [BLKIF_OP_DISCARD] = "discard" };
434
435 if (op < 0 || op >= ARRAY_SIZE(names))
436 return "unknown";
437
438 if (!names[op])
439 return "reserved";
440
441 return names[op];
442}
443static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
444{
445 unsigned int end = minor + nr;
446 int rc;
447
448 if (end > nr_minors) {
449 unsigned long *bitmap, *old;
450
451 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
452 GFP_KERNEL);
453 if (bitmap == NULL)
454 return -ENOMEM;
455
456 spin_lock(&minor_lock);
457 if (end > nr_minors) {
458 old = minors;
459 memcpy(bitmap, minors,
460 BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
461 minors = bitmap;
462 nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
463 } else
464 old = bitmap;
465 spin_unlock(&minor_lock);
466 kfree(old);
467 }
468
469 spin_lock(&minor_lock);
470 if (find_next_bit(minors, end, minor) >= end) {
471 bitmap_set(minors, minor, nr);
472 rc = 0;
473 } else
474 rc = -EBUSY;
475 spin_unlock(&minor_lock);
476
477 return rc;
478}
479
480static void xlbd_release_minors(unsigned int minor, unsigned int nr)
481{
482 unsigned int end = minor + nr;
483
484 BUG_ON(end > nr_minors);
485 spin_lock(&minor_lock);
486 bitmap_clear(minors, minor, nr);
487 spin_unlock(&minor_lock);
488}
489
490static void blkif_restart_queue_callback(void *arg)
491{
492 struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
493 schedule_work(&rinfo->work);
494}
495
496static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
497{
498 /* We don't have real geometry info, but let's at least return
499 values consistent with the size of the device */
500 sector_t nsect = get_capacity(bd->bd_disk);
501 sector_t cylinders = nsect;
502
503 hg->heads = 0xff;
504 hg->sectors = 0x3f;
505 sector_div(cylinders, hg->heads * hg->sectors);
506 hg->cylinders = cylinders;
507 if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
508 hg->cylinders = 0xffff;
509 return 0;
510}
511
512static int blkif_ioctl(struct block_device *bdev, blk_mode_t mode,
513 unsigned command, unsigned long argument)
514{
515 struct blkfront_info *info = bdev->bd_disk->private_data;
516 int i;
517
518 switch (command) {
519 case CDROMMULTISESSION:
520 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
521 if (put_user(0, (char __user *)(argument + i)))
522 return -EFAULT;
523 return 0;
524 case CDROM_GET_CAPABILITY:
525 if (!(info->vdisk_info & VDISK_CDROM))
526 return -EINVAL;
527 return 0;
528 default:
529 return -EINVAL;
530 }
531}
532
533static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
534 struct request *req,
535 struct blkif_request **ring_req)
536{
537 unsigned long id;
538
539 *ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
540 rinfo->ring.req_prod_pvt++;
541
542 id = get_id_from_freelist(rinfo);
543 rinfo->shadow[id].request = req;
544 rinfo->shadow[id].status = REQ_PROCESSING;
545 rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
546
547 rinfo->shadow[id].req.u.rw.id = id;
548
549 return id;
550}
551
552static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
553{
554 struct blkfront_info *info = rinfo->dev_info;
555 struct blkif_request *ring_req, *final_ring_req;
556 unsigned long id;
557
558 /* Fill out a communications ring structure. */
559 id = blkif_ring_get_request(rinfo, req, &final_ring_req);
560 ring_req = &rinfo->shadow[id].req;
561
562 ring_req->operation = BLKIF_OP_DISCARD;
563 ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
564 ring_req->u.discard.id = id;
565 ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
566 if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard)
567 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
568 else
569 ring_req->u.discard.flag = 0;
570
571 /* Copy the request to the ring page. */
572 *final_ring_req = *ring_req;
573 rinfo->shadow[id].status = REQ_WAITING;
574
575 return 0;
576}
577
578struct setup_rw_req {
579 unsigned int grant_idx;
580 struct blkif_request_segment *segments;
581 struct blkfront_ring_info *rinfo;
582 struct blkif_request *ring_req;
583 grant_ref_t gref_head;
584 unsigned int id;
585 /* Only used when persistent grant is used and it's a write request */
586 bool need_copy;
587 unsigned int bvec_off;
588 char *bvec_data;
589
590 bool require_extra_req;
591 struct blkif_request *extra_ring_req;
592};
593
594static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
595 unsigned int len, void *data)
596{
597 struct setup_rw_req *setup = data;
598 int n, ref;
599 struct grant *gnt_list_entry;
600 unsigned int fsect, lsect;
601 /* Convenient aliases */
602 unsigned int grant_idx = setup->grant_idx;
603 struct blkif_request *ring_req = setup->ring_req;
604 struct blkfront_ring_info *rinfo = setup->rinfo;
605 /*
606 * We always use the shadow of the first request to store the list
607 * of grant associated to the block I/O request. This made the
608 * completion more easy to handle even if the block I/O request is
609 * split.
610 */
611 struct blk_shadow *shadow = &rinfo->shadow[setup->id];
612
613 if (unlikely(setup->require_extra_req &&
614 grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
615 /*
616 * We are using the second request, setup grant_idx
617 * to be the index of the segment array.
618 */
619 grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
620 ring_req = setup->extra_ring_req;
621 }
622
623 if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
624 (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
625 if (setup->segments)
626 kunmap_atomic(setup->segments);
627
628 n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
629 gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
630 shadow->indirect_grants[n] = gnt_list_entry;
631 setup->segments = kmap_atomic(gnt_list_entry->page);
632 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
633 }
634
635 gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
636 ref = gnt_list_entry->gref;
637 /*
638 * All the grants are stored in the shadow of the first
639 * request. Therefore we have to use the global index.
640 */
641 shadow->grants_used[setup->grant_idx] = gnt_list_entry;
642
643 if (setup->need_copy) {
644 void *shared_data;
645
646 shared_data = kmap_atomic(gnt_list_entry->page);
647 /*
648 * this does not wipe data stored outside the
649 * range sg->offset..sg->offset+sg->length.
650 * Therefore, blkback *could* see data from
651 * previous requests. This is OK as long as
652 * persistent grants are shared with just one
653 * domain. It may need refactoring if this
654 * changes
655 */
656 memcpy(shared_data + offset,
657 setup->bvec_data + setup->bvec_off,
658 len);
659
660 kunmap_atomic(shared_data);
661 setup->bvec_off += len;
662 }
663
664 fsect = offset >> 9;
665 lsect = fsect + (len >> 9) - 1;
666 if (ring_req->operation != BLKIF_OP_INDIRECT) {
667 ring_req->u.rw.seg[grant_idx] =
668 (struct blkif_request_segment) {
669 .gref = ref,
670 .first_sect = fsect,
671 .last_sect = lsect };
672 } else {
673 setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
674 (struct blkif_request_segment) {
675 .gref = ref,
676 .first_sect = fsect,
677 .last_sect = lsect };
678 }
679
680 (setup->grant_idx)++;
681}
682
683static void blkif_setup_extra_req(struct blkif_request *first,
684 struct blkif_request *second)
685{
686 uint16_t nr_segments = first->u.rw.nr_segments;
687
688 /*
689 * The second request is only present when the first request uses
690 * all its segments. It's always the continuity of the first one.
691 */
692 first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
693
694 second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
695 second->u.rw.sector_number = first->u.rw.sector_number +
696 (BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
697
698 second->u.rw.handle = first->u.rw.handle;
699 second->operation = first->operation;
700}
701
702static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
703{
704 struct blkfront_info *info = rinfo->dev_info;
705 struct blkif_request *ring_req, *extra_ring_req = NULL;
706 struct blkif_request *final_ring_req, *final_extra_ring_req = NULL;
707 unsigned long id, extra_id = NO_ASSOCIATED_ID;
708 bool require_extra_req = false;
709 int i;
710 struct setup_rw_req setup = {
711 .grant_idx = 0,
712 .segments = NULL,
713 .rinfo = rinfo,
714 .need_copy = rq_data_dir(req) && info->bounce,
715 };
716
717 /*
718 * Used to store if we are able to queue the request by just using
719 * existing persistent grants, or if we have to get new grants,
720 * as there are not sufficiently many free.
721 */
722 bool new_persistent_gnts = false;
723 struct scatterlist *sg;
724 int num_sg, max_grefs, num_grant;
725
726 max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
727 if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
728 /*
729 * If we are using indirect segments we need to account
730 * for the indirect grefs used in the request.
731 */
732 max_grefs += INDIRECT_GREFS(max_grefs);
733
734 /* Check if we have enough persistent grants to allocate a requests */
735 if (rinfo->persistent_gnts_c < max_grefs) {
736 new_persistent_gnts = true;
737
738 if (gnttab_alloc_grant_references(
739 max_grefs - rinfo->persistent_gnts_c,
740 &setup.gref_head) < 0) {
741 gnttab_request_free_callback(
742 &rinfo->callback,
743 blkif_restart_queue_callback,
744 rinfo,
745 max_grefs - rinfo->persistent_gnts_c);
746 return 1;
747 }
748 }
749
750 /* Fill out a communications ring structure. */
751 id = blkif_ring_get_request(rinfo, req, &final_ring_req);
752 ring_req = &rinfo->shadow[id].req;
753
754 num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
755 num_grant = 0;
756 /* Calculate the number of grant used */
757 for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
758 num_grant += gnttab_count_grant(sg->offset, sg->length);
759
760 require_extra_req = info->max_indirect_segments == 0 &&
761 num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
762 BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
763
764 rinfo->shadow[id].num_sg = num_sg;
765 if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
766 likely(!require_extra_req)) {
767 /*
768 * The indirect operation can only be a BLKIF_OP_READ or
769 * BLKIF_OP_WRITE
770 */
771 BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA);
772 ring_req->operation = BLKIF_OP_INDIRECT;
773 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
774 BLKIF_OP_WRITE : BLKIF_OP_READ;
775 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
776 ring_req->u.indirect.handle = info->handle;
777 ring_req->u.indirect.nr_segments = num_grant;
778 } else {
779 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
780 ring_req->u.rw.handle = info->handle;
781 ring_req->operation = rq_data_dir(req) ?
782 BLKIF_OP_WRITE : BLKIF_OP_READ;
783 if (req_op(req) == REQ_OP_FLUSH ||
784 (req_op(req) == REQ_OP_WRITE && (req->cmd_flags & REQ_FUA))) {
785 /*
786 * Ideally we can do an unordered flush-to-disk.
787 * In case the backend onlysupports barriers, use that.
788 * A barrier request a superset of FUA, so we can
789 * implement it the same way. (It's also a FLUSH+FUA,
790 * since it is guaranteed ordered WRT previous writes.)
791 *
792 * Note that can end up here with a FUA write and the
793 * flags cleared. This happens when the flag was
794 * run-time disabled after a failing I/O, and we'll
795 * simplify submit it as a normal write.
796 */
797 if (info->feature_flush && info->feature_fua)
798 ring_req->operation =
799 BLKIF_OP_WRITE_BARRIER;
800 else if (info->feature_flush)
801 ring_req->operation =
802 BLKIF_OP_FLUSH_DISKCACHE;
803 }
804 ring_req->u.rw.nr_segments = num_grant;
805 if (unlikely(require_extra_req)) {
806 extra_id = blkif_ring_get_request(rinfo, req,
807 &final_extra_ring_req);
808 extra_ring_req = &rinfo->shadow[extra_id].req;
809
810 /*
811 * Only the first request contains the scatter-gather
812 * list.
813 */
814 rinfo->shadow[extra_id].num_sg = 0;
815
816 blkif_setup_extra_req(ring_req, extra_ring_req);
817
818 /* Link the 2 requests together */
819 rinfo->shadow[extra_id].associated_id = id;
820 rinfo->shadow[id].associated_id = extra_id;
821 }
822 }
823
824 setup.ring_req = ring_req;
825 setup.id = id;
826
827 setup.require_extra_req = require_extra_req;
828 if (unlikely(require_extra_req))
829 setup.extra_ring_req = extra_ring_req;
830
831 for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
832 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
833
834 if (setup.need_copy) {
835 setup.bvec_off = sg->offset;
836 setup.bvec_data = kmap_atomic(sg_page(sg));
837 }
838
839 gnttab_foreach_grant_in_range(sg_page(sg),
840 sg->offset,
841 sg->length,
842 blkif_setup_rw_req_grant,
843 &setup);
844
845 if (setup.need_copy)
846 kunmap_atomic(setup.bvec_data);
847 }
848 if (setup.segments)
849 kunmap_atomic(setup.segments);
850
851 /* Copy request(s) to the ring page. */
852 *final_ring_req = *ring_req;
853 rinfo->shadow[id].status = REQ_WAITING;
854 if (unlikely(require_extra_req)) {
855 *final_extra_ring_req = *extra_ring_req;
856 rinfo->shadow[extra_id].status = REQ_WAITING;
857 }
858
859 if (new_persistent_gnts)
860 gnttab_free_grant_references(setup.gref_head);
861
862 return 0;
863}
864
865/*
866 * Generate a Xen blkfront IO request from a blk layer request. Reads
867 * and writes are handled as expected.
868 *
869 * @req: a request struct
870 */
871static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
872{
873 if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
874 return 1;
875
876 if (unlikely(req_op(req) == REQ_OP_DISCARD ||
877 req_op(req) == REQ_OP_SECURE_ERASE))
878 return blkif_queue_discard_req(req, rinfo);
879 else
880 return blkif_queue_rw_req(req, rinfo);
881}
882
883static inline void flush_requests(struct blkfront_ring_info *rinfo)
884{
885 int notify;
886
887 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
888
889 if (notify)
890 notify_remote_via_irq(rinfo->irq);
891}
892
893static blk_status_t blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
894 const struct blk_mq_queue_data *qd)
895{
896 unsigned long flags;
897 int qid = hctx->queue_num;
898 struct blkfront_info *info = hctx->queue->queuedata;
899 struct blkfront_ring_info *rinfo = NULL;
900
901 rinfo = get_rinfo(info, qid);
902 blk_mq_start_request(qd->rq);
903 spin_lock_irqsave(&rinfo->ring_lock, flags);
904
905 /*
906 * Check if the backend actually supports flushes.
907 *
908 * While the block layer won't send us flushes if we don't claim to
909 * support them, the Xen protocol allows the backend to revoke support
910 * at any time. That is of course a really bad idea and dangerous, but
911 * has been allowed for 10+ years. In that case we simply clear the
912 * flags, and directly return here for an empty flush and ignore the
913 * FUA flag later on.
914 */
915 if (unlikely(req_op(qd->rq) == REQ_OP_FLUSH && !info->feature_flush))
916 goto complete;
917
918 if (RING_FULL(&rinfo->ring))
919 goto out_busy;
920 if (blkif_queue_request(qd->rq, rinfo))
921 goto out_busy;
922
923 flush_requests(rinfo);
924 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
925 return BLK_STS_OK;
926
927out_busy:
928 blk_mq_stop_hw_queue(hctx);
929 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
930 return BLK_STS_DEV_RESOURCE;
931complete:
932 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
933 blk_mq_end_request(qd->rq, BLK_STS_OK);
934 return BLK_STS_OK;
935}
936
937static void blkif_complete_rq(struct request *rq)
938{
939 blk_mq_end_request(rq, blkif_req(rq)->error);
940}
941
942static const struct blk_mq_ops blkfront_mq_ops = {
943 .queue_rq = blkif_queue_rq,
944 .complete = blkif_complete_rq,
945};
946
947static void blkif_set_queue_limits(const struct blkfront_info *info,
948 struct queue_limits *lim)
949{
950 unsigned int segments = info->max_indirect_segments ? :
951 BLKIF_MAX_SEGMENTS_PER_REQUEST;
952
953 if (info->feature_discard) {
954 lim->max_hw_discard_sectors = UINT_MAX;
955 if (info->discard_granularity)
956 lim->discard_granularity = info->discard_granularity;
957 lim->discard_alignment = info->discard_alignment;
958 if (info->feature_secdiscard)
959 lim->max_secure_erase_sectors = UINT_MAX;
960 }
961
962 if (info->feature_flush) {
963 lim->features |= BLK_FEAT_WRITE_CACHE;
964 if (info->feature_fua)
965 lim->features |= BLK_FEAT_FUA;
966 }
967
968 /* Hard sector size and max sectors impersonate the equiv. hardware. */
969 lim->logical_block_size = info->sector_size;
970 lim->physical_block_size = info->physical_sector_size;
971 lim->max_hw_sectors = (segments * XEN_PAGE_SIZE) / 512;
972
973 /* Each segment in a request is up to an aligned page in size. */
974 lim->seg_boundary_mask = PAGE_SIZE - 1;
975 lim->max_segment_size = PAGE_SIZE;
976
977 /* Ensure a merged request will fit in a single I/O ring slot. */
978 lim->max_segments = segments / GRANTS_PER_PSEG;
979
980 /* Make sure buffer addresses are sector-aligned. */
981 lim->dma_alignment = 511;
982}
983
984static const char *flush_info(struct blkfront_info *info)
985{
986 if (info->feature_flush && info->feature_fua)
987 return "barrier: enabled;";
988 else if (info->feature_flush)
989 return "flush diskcache: enabled;";
990 else
991 return "barrier or flush: disabled;";
992}
993
994static void xlvbd_flush(struct blkfront_info *info)
995{
996 pr_info("blkfront: %s: %s %s %s %s %s %s %s\n",
997 info->gd->disk_name, flush_info(info),
998 "persistent grants:", info->feature_persistent ?
999 "enabled;" : "disabled;", "indirect descriptors:",
1000 info->max_indirect_segments ? "enabled;" : "disabled;",
1001 "bounce buffer:", info->bounce ? "enabled" : "disabled;");
1002}
1003
1004static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
1005{
1006 int major;
1007 major = BLKIF_MAJOR(vdevice);
1008 *minor = BLKIF_MINOR(vdevice);
1009 switch (major) {
1010 case XEN_IDE0_MAJOR:
1011 *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
1012 *minor = ((*minor / 64) * PARTS_PER_DISK) +
1013 EMULATED_HD_DISK_MINOR_OFFSET;
1014 break;
1015 case XEN_IDE1_MAJOR:
1016 *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1017 *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1018 EMULATED_HD_DISK_MINOR_OFFSET;
1019 break;
1020 case XEN_SCSI_DISK0_MAJOR:
1021 *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1022 *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1023 break;
1024 case XEN_SCSI_DISK1_MAJOR:
1025 case XEN_SCSI_DISK2_MAJOR:
1026 case XEN_SCSI_DISK3_MAJOR:
1027 case XEN_SCSI_DISK4_MAJOR:
1028 case XEN_SCSI_DISK5_MAJOR:
1029 case XEN_SCSI_DISK6_MAJOR:
1030 case XEN_SCSI_DISK7_MAJOR:
1031 *offset = (*minor / PARTS_PER_DISK) +
1032 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1033 EMULATED_SD_DISK_NAME_OFFSET;
1034 *minor = *minor +
1035 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1036 EMULATED_SD_DISK_MINOR_OFFSET;
1037 break;
1038 case XEN_SCSI_DISK8_MAJOR:
1039 case XEN_SCSI_DISK9_MAJOR:
1040 case XEN_SCSI_DISK10_MAJOR:
1041 case XEN_SCSI_DISK11_MAJOR:
1042 case XEN_SCSI_DISK12_MAJOR:
1043 case XEN_SCSI_DISK13_MAJOR:
1044 case XEN_SCSI_DISK14_MAJOR:
1045 case XEN_SCSI_DISK15_MAJOR:
1046 *offset = (*minor / PARTS_PER_DISK) +
1047 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1048 EMULATED_SD_DISK_NAME_OFFSET;
1049 *minor = *minor +
1050 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1051 EMULATED_SD_DISK_MINOR_OFFSET;
1052 break;
1053 case XENVBD_MAJOR:
1054 *offset = *minor / PARTS_PER_DISK;
1055 break;
1056 default:
1057 printk(KERN_WARNING "blkfront: your disk configuration is "
1058 "incorrect, please use an xvd device instead\n");
1059 return -ENODEV;
1060 }
1061 return 0;
1062}
1063
1064static char *encode_disk_name(char *ptr, unsigned int n)
1065{
1066 if (n >= 26)
1067 ptr = encode_disk_name(ptr, n / 26 - 1);
1068 *ptr = 'a' + n % 26;
1069 return ptr + 1;
1070}
1071
1072static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1073 struct blkfront_info *info)
1074{
1075 struct queue_limits lim = {};
1076 struct gendisk *gd;
1077 int nr_minors = 1;
1078 int err;
1079 unsigned int offset;
1080 int minor;
1081 int nr_parts;
1082 char *ptr;
1083
1084 BUG_ON(info->gd != NULL);
1085 BUG_ON(info->rq != NULL);
1086
1087 if ((info->vdevice>>EXT_SHIFT) > 1) {
1088 /* this is above the extended range; something is wrong */
1089 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1090 return -ENODEV;
1091 }
1092
1093 if (!VDEV_IS_EXTENDED(info->vdevice)) {
1094 err = xen_translate_vdev(info->vdevice, &minor, &offset);
1095 if (err)
1096 return err;
1097 nr_parts = PARTS_PER_DISK;
1098 } else {
1099 minor = BLKIF_MINOR_EXT(info->vdevice);
1100 nr_parts = PARTS_PER_EXT_DISK;
1101 offset = minor / nr_parts;
1102 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1103 printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1104 "emulated IDE disks,\n\t choose an xvd device name"
1105 "from xvde on\n", info->vdevice);
1106 }
1107 if (minor >> MINORBITS) {
1108 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1109 info->vdevice, minor);
1110 return -ENODEV;
1111 }
1112
1113 if ((minor % nr_parts) == 0)
1114 nr_minors = nr_parts;
1115
1116 err = xlbd_reserve_minors(minor, nr_minors);
1117 if (err)
1118 return err;
1119
1120 memset(&info->tag_set, 0, sizeof(info->tag_set));
1121 info->tag_set.ops = &blkfront_mq_ops;
1122 info->tag_set.nr_hw_queues = info->nr_rings;
1123 if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
1124 /*
1125 * When indirect descriptior is not supported, the I/O request
1126 * will be split between multiple request in the ring.
1127 * To avoid problems when sending the request, divide by
1128 * 2 the depth of the queue.
1129 */
1130 info->tag_set.queue_depth = BLK_RING_SIZE(info) / 2;
1131 } else
1132 info->tag_set.queue_depth = BLK_RING_SIZE(info);
1133 info->tag_set.numa_node = NUMA_NO_NODE;
1134 info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
1135 info->tag_set.cmd_size = sizeof(struct blkif_req);
1136 info->tag_set.driver_data = info;
1137
1138 err = blk_mq_alloc_tag_set(&info->tag_set);
1139 if (err)
1140 goto out_release_minors;
1141
1142 blkif_set_queue_limits(info, &lim);
1143 gd = blk_mq_alloc_disk(&info->tag_set, &lim, info);
1144 if (IS_ERR(gd)) {
1145 err = PTR_ERR(gd);
1146 goto out_free_tag_set;
1147 }
1148
1149 strcpy(gd->disk_name, DEV_NAME);
1150 ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1151 BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1152 if (nr_minors > 1)
1153 *ptr = 0;
1154 else
1155 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1156 "%d", minor & (nr_parts - 1));
1157
1158 gd->major = XENVBD_MAJOR;
1159 gd->first_minor = minor;
1160 gd->minors = nr_minors;
1161 gd->fops = &xlvbd_block_fops;
1162 gd->private_data = info;
1163 set_capacity(gd, capacity);
1164
1165 info->rq = gd->queue;
1166 info->gd = gd;
1167
1168 xlvbd_flush(info);
1169
1170 if (info->vdisk_info & VDISK_READONLY)
1171 set_disk_ro(gd, 1);
1172 if (info->vdisk_info & VDISK_REMOVABLE)
1173 gd->flags |= GENHD_FL_REMOVABLE;
1174
1175 return 0;
1176
1177out_free_tag_set:
1178 blk_mq_free_tag_set(&info->tag_set);
1179out_release_minors:
1180 xlbd_release_minors(minor, nr_minors);
1181 return err;
1182}
1183
1184/* Already hold rinfo->ring_lock. */
1185static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1186{
1187 if (!RING_FULL(&rinfo->ring))
1188 blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1189}
1190
1191static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1192{
1193 unsigned long flags;
1194
1195 spin_lock_irqsave(&rinfo->ring_lock, flags);
1196 kick_pending_request_queues_locked(rinfo);
1197 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1198}
1199
1200static void blkif_restart_queue(struct work_struct *work)
1201{
1202 struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1203
1204 if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1205 kick_pending_request_queues(rinfo);
1206}
1207
1208static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1209{
1210 struct grant *persistent_gnt, *n;
1211 struct blkfront_info *info = rinfo->dev_info;
1212 int i, j, segs;
1213
1214 /*
1215 * Remove indirect pages, this only happens when using indirect
1216 * descriptors but not persistent grants
1217 */
1218 if (!list_empty(&rinfo->indirect_pages)) {
1219 struct page *indirect_page, *n;
1220
1221 BUG_ON(info->bounce);
1222 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1223 list_del(&indirect_page->lru);
1224 __free_page(indirect_page);
1225 }
1226 }
1227
1228 /* Remove all persistent grants. */
1229 if (!list_empty(&rinfo->grants)) {
1230 list_for_each_entry_safe(persistent_gnt, n,
1231 &rinfo->grants, node) {
1232 list_del(&persistent_gnt->node);
1233 if (persistent_gnt->gref != INVALID_GRANT_REF) {
1234 gnttab_end_foreign_access(persistent_gnt->gref,
1235 NULL);
1236 rinfo->persistent_gnts_c--;
1237 }
1238 if (info->bounce)
1239 __free_page(persistent_gnt->page);
1240 kfree(persistent_gnt);
1241 }
1242 }
1243 BUG_ON(rinfo->persistent_gnts_c != 0);
1244
1245 for (i = 0; i < BLK_RING_SIZE(info); i++) {
1246 /*
1247 * Clear persistent grants present in requests already
1248 * on the shared ring
1249 */
1250 if (!rinfo->shadow[i].request)
1251 goto free_shadow;
1252
1253 segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1254 rinfo->shadow[i].req.u.indirect.nr_segments :
1255 rinfo->shadow[i].req.u.rw.nr_segments;
1256 for (j = 0; j < segs; j++) {
1257 persistent_gnt = rinfo->shadow[i].grants_used[j];
1258 gnttab_end_foreign_access(persistent_gnt->gref, NULL);
1259 if (info->bounce)
1260 __free_page(persistent_gnt->page);
1261 kfree(persistent_gnt);
1262 }
1263
1264 if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1265 /*
1266 * If this is not an indirect operation don't try to
1267 * free indirect segments
1268 */
1269 goto free_shadow;
1270
1271 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1272 persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1273 gnttab_end_foreign_access(persistent_gnt->gref, NULL);
1274 __free_page(persistent_gnt->page);
1275 kfree(persistent_gnt);
1276 }
1277
1278free_shadow:
1279 kvfree(rinfo->shadow[i].grants_used);
1280 rinfo->shadow[i].grants_used = NULL;
1281 kvfree(rinfo->shadow[i].indirect_grants);
1282 rinfo->shadow[i].indirect_grants = NULL;
1283 kvfree(rinfo->shadow[i].sg);
1284 rinfo->shadow[i].sg = NULL;
1285 }
1286
1287 /* No more gnttab callback work. */
1288 gnttab_cancel_free_callback(&rinfo->callback);
1289
1290 /* Flush gnttab callback work. Must be done with no locks held. */
1291 flush_work(&rinfo->work);
1292
1293 /* Free resources associated with old device channel. */
1294 xenbus_teardown_ring((void **)&rinfo->ring.sring, info->nr_ring_pages,
1295 rinfo->ring_ref);
1296
1297 if (rinfo->irq)
1298 unbind_from_irqhandler(rinfo->irq, rinfo);
1299 rinfo->evtchn = rinfo->irq = 0;
1300}
1301
1302static void blkif_free(struct blkfront_info *info, int suspend)
1303{
1304 unsigned int i;
1305 struct blkfront_ring_info *rinfo;
1306
1307 /* Prevent new requests being issued until we fix things up. */
1308 info->connected = suspend ?
1309 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1310 /* No more blkif_request(). */
1311 if (info->rq)
1312 blk_mq_stop_hw_queues(info->rq);
1313
1314 for_each_rinfo(info, rinfo, i)
1315 blkif_free_ring(rinfo);
1316
1317 kvfree(info->rinfo);
1318 info->rinfo = NULL;
1319 info->nr_rings = 0;
1320}
1321
1322struct copy_from_grant {
1323 const struct blk_shadow *s;
1324 unsigned int grant_idx;
1325 unsigned int bvec_offset;
1326 char *bvec_data;
1327};
1328
1329static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1330 unsigned int len, void *data)
1331{
1332 struct copy_from_grant *info = data;
1333 char *shared_data;
1334 /* Convenient aliases */
1335 const struct blk_shadow *s = info->s;
1336
1337 shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1338
1339 memcpy(info->bvec_data + info->bvec_offset,
1340 shared_data + offset, len);
1341
1342 info->bvec_offset += len;
1343 info->grant_idx++;
1344
1345 kunmap_atomic(shared_data);
1346}
1347
1348static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1349{
1350 switch (rsp)
1351 {
1352 case BLKIF_RSP_OKAY:
1353 return REQ_DONE;
1354 case BLKIF_RSP_EOPNOTSUPP:
1355 return REQ_EOPNOTSUPP;
1356 case BLKIF_RSP_ERROR:
1357 default:
1358 return REQ_ERROR;
1359 }
1360}
1361
1362/*
1363 * Get the final status of the block request based on two ring response
1364 */
1365static int blkif_get_final_status(enum blk_req_status s1,
1366 enum blk_req_status s2)
1367{
1368 BUG_ON(s1 < REQ_DONE);
1369 BUG_ON(s2 < REQ_DONE);
1370
1371 if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1372 return BLKIF_RSP_ERROR;
1373 else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1374 return BLKIF_RSP_EOPNOTSUPP;
1375 return BLKIF_RSP_OKAY;
1376}
1377
1378/*
1379 * Return values:
1380 * 1 response processed.
1381 * 0 missing further responses.
1382 * -1 error while processing.
1383 */
1384static int blkif_completion(unsigned long *id,
1385 struct blkfront_ring_info *rinfo,
1386 struct blkif_response *bret)
1387{
1388 int i = 0;
1389 struct scatterlist *sg;
1390 int num_sg, num_grant;
1391 struct blkfront_info *info = rinfo->dev_info;
1392 struct blk_shadow *s = &rinfo->shadow[*id];
1393 struct copy_from_grant data = {
1394 .grant_idx = 0,
1395 };
1396
1397 num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1398 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1399
1400 /* The I/O request may be split in two. */
1401 if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1402 struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1403
1404 /* Keep the status of the current response in shadow. */
1405 s->status = blkif_rsp_to_req_status(bret->status);
1406
1407 /* Wait the second response if not yet here. */
1408 if (s2->status < REQ_DONE)
1409 return 0;
1410
1411 bret->status = blkif_get_final_status(s->status,
1412 s2->status);
1413
1414 /*
1415 * All the grants is stored in the first shadow in order
1416 * to make the completion code simpler.
1417 */
1418 num_grant += s2->req.u.rw.nr_segments;
1419
1420 /*
1421 * The two responses may not come in order. Only the
1422 * first request will store the scatter-gather list.
1423 */
1424 if (s2->num_sg != 0) {
1425 /* Update "id" with the ID of the first response. */
1426 *id = s->associated_id;
1427 s = s2;
1428 }
1429
1430 /*
1431 * We don't need anymore the second request, so recycling
1432 * it now.
1433 */
1434 if (add_id_to_freelist(rinfo, s->associated_id))
1435 WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1436 info->gd->disk_name, s->associated_id);
1437 }
1438
1439 data.s = s;
1440 num_sg = s->num_sg;
1441
1442 if (bret->operation == BLKIF_OP_READ && info->bounce) {
1443 for_each_sg(s->sg, sg, num_sg, i) {
1444 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1445
1446 data.bvec_offset = sg->offset;
1447 data.bvec_data = kmap_atomic(sg_page(sg));
1448
1449 gnttab_foreach_grant_in_range(sg_page(sg),
1450 sg->offset,
1451 sg->length,
1452 blkif_copy_from_grant,
1453 &data);
1454
1455 kunmap_atomic(data.bvec_data);
1456 }
1457 }
1458 /* Add the persistent grant into the list of free grants */
1459 for (i = 0; i < num_grant; i++) {
1460 if (!gnttab_try_end_foreign_access(s->grants_used[i]->gref)) {
1461 /*
1462 * If the grant is still mapped by the backend (the
1463 * backend has chosen to make this grant persistent)
1464 * we add it at the head of the list, so it will be
1465 * reused first.
1466 */
1467 if (!info->feature_persistent) {
1468 pr_alert("backed has not unmapped grant: %u\n",
1469 s->grants_used[i]->gref);
1470 return -1;
1471 }
1472 list_add(&s->grants_used[i]->node, &rinfo->grants);
1473 rinfo->persistent_gnts_c++;
1474 } else {
1475 /*
1476 * If the grant is not mapped by the backend we add it
1477 * to the tail of the list, so it will not be picked
1478 * again unless we run out of persistent grants.
1479 */
1480 s->grants_used[i]->gref = INVALID_GRANT_REF;
1481 list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1482 }
1483 }
1484 if (s->req.operation == BLKIF_OP_INDIRECT) {
1485 for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1486 if (!gnttab_try_end_foreign_access(s->indirect_grants[i]->gref)) {
1487 if (!info->feature_persistent) {
1488 pr_alert("backed has not unmapped grant: %u\n",
1489 s->indirect_grants[i]->gref);
1490 return -1;
1491 }
1492 list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1493 rinfo->persistent_gnts_c++;
1494 } else {
1495 struct page *indirect_page;
1496
1497 /*
1498 * Add the used indirect page back to the list of
1499 * available pages for indirect grefs.
1500 */
1501 if (!info->bounce) {
1502 indirect_page = s->indirect_grants[i]->page;
1503 list_add(&indirect_page->lru, &rinfo->indirect_pages);
1504 }
1505 s->indirect_grants[i]->gref = INVALID_GRANT_REF;
1506 list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1507 }
1508 }
1509 }
1510
1511 return 1;
1512}
1513
1514static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1515{
1516 struct request *req;
1517 struct blkif_response bret;
1518 RING_IDX i, rp;
1519 unsigned long flags;
1520 struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1521 struct blkfront_info *info = rinfo->dev_info;
1522 unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS;
1523
1524 if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
1525 xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
1526 return IRQ_HANDLED;
1527 }
1528
1529 spin_lock_irqsave(&rinfo->ring_lock, flags);
1530 again:
1531 rp = READ_ONCE(rinfo->ring.sring->rsp_prod);
1532 virt_rmb(); /* Ensure we see queued responses up to 'rp'. */
1533 if (RING_RESPONSE_PROD_OVERFLOW(&rinfo->ring, rp)) {
1534 pr_alert("%s: illegal number of responses %u\n",
1535 info->gd->disk_name, rp - rinfo->ring.rsp_cons);
1536 goto err;
1537 }
1538
1539 for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1540 unsigned long id;
1541 unsigned int op;
1542
1543 eoiflag = 0;
1544
1545 RING_COPY_RESPONSE(&rinfo->ring, i, &bret);
1546 id = bret.id;
1547
1548 /*
1549 * The backend has messed up and given us an id that we would
1550 * never have given to it (we stamp it up to BLK_RING_SIZE -
1551 * look in get_id_from_freelist.
1552 */
1553 if (id >= BLK_RING_SIZE(info)) {
1554 pr_alert("%s: response has incorrect id (%ld)\n",
1555 info->gd->disk_name, id);
1556 goto err;
1557 }
1558 if (rinfo->shadow[id].status != REQ_WAITING) {
1559 pr_alert("%s: response references no pending request\n",
1560 info->gd->disk_name);
1561 goto err;
1562 }
1563
1564 rinfo->shadow[id].status = REQ_PROCESSING;
1565 req = rinfo->shadow[id].request;
1566
1567 op = rinfo->shadow[id].req.operation;
1568 if (op == BLKIF_OP_INDIRECT)
1569 op = rinfo->shadow[id].req.u.indirect.indirect_op;
1570 if (bret.operation != op) {
1571 pr_alert("%s: response has wrong operation (%u instead of %u)\n",
1572 info->gd->disk_name, bret.operation, op);
1573 goto err;
1574 }
1575
1576 if (bret.operation != BLKIF_OP_DISCARD) {
1577 int ret;
1578
1579 /*
1580 * We may need to wait for an extra response if the
1581 * I/O request is split in 2
1582 */
1583 ret = blkif_completion(&id, rinfo, &bret);
1584 if (!ret)
1585 continue;
1586 if (unlikely(ret < 0))
1587 goto err;
1588 }
1589
1590 if (add_id_to_freelist(rinfo, id)) {
1591 WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1592 info->gd->disk_name, op_name(bret.operation), id);
1593 continue;
1594 }
1595
1596 if (bret.status == BLKIF_RSP_OKAY)
1597 blkif_req(req)->error = BLK_STS_OK;
1598 else
1599 blkif_req(req)->error = BLK_STS_IOERR;
1600
1601 switch (bret.operation) {
1602 case BLKIF_OP_DISCARD:
1603 if (unlikely(bret.status == BLKIF_RSP_EOPNOTSUPP)) {
1604 struct request_queue *rq = info->rq;
1605
1606 pr_warn_ratelimited("blkfront: %s: %s op failed\n",
1607 info->gd->disk_name, op_name(bret.operation));
1608 blkif_req(req)->error = BLK_STS_NOTSUPP;
1609 info->feature_discard = 0;
1610 info->feature_secdiscard = 0;
1611 blk_queue_disable_discard(rq);
1612 blk_queue_disable_secure_erase(rq);
1613 }
1614 break;
1615 case BLKIF_OP_FLUSH_DISKCACHE:
1616 case BLKIF_OP_WRITE_BARRIER:
1617 if (unlikely(bret.status == BLKIF_RSP_EOPNOTSUPP)) {
1618 pr_warn_ratelimited("blkfront: %s: %s op failed\n",
1619 info->gd->disk_name, op_name(bret.operation));
1620 blkif_req(req)->error = BLK_STS_NOTSUPP;
1621 }
1622 if (unlikely(bret.status == BLKIF_RSP_ERROR &&
1623 rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1624 pr_warn_ratelimited("blkfront: %s: empty %s op failed\n",
1625 info->gd->disk_name, op_name(bret.operation));
1626 blkif_req(req)->error = BLK_STS_NOTSUPP;
1627 }
1628 if (unlikely(blkif_req(req)->error)) {
1629 if (blkif_req(req)->error == BLK_STS_NOTSUPP)
1630 blkif_req(req)->error = BLK_STS_OK;
1631 info->feature_fua = 0;
1632 info->feature_flush = 0;
1633 }
1634 fallthrough;
1635 case BLKIF_OP_READ:
1636 case BLKIF_OP_WRITE:
1637 if (unlikely(bret.status != BLKIF_RSP_OKAY))
1638 dev_dbg_ratelimited(&info->xbdev->dev,
1639 "Bad return from blkdev data request: %#x\n",
1640 bret.status);
1641
1642 break;
1643 default:
1644 BUG();
1645 }
1646
1647 if (likely(!blk_should_fake_timeout(req->q)))
1648 blk_mq_complete_request(req);
1649 }
1650
1651 rinfo->ring.rsp_cons = i;
1652
1653 if (i != rinfo->ring.req_prod_pvt) {
1654 int more_to_do;
1655 RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1656 if (more_to_do)
1657 goto again;
1658 } else
1659 rinfo->ring.sring->rsp_event = i + 1;
1660
1661 kick_pending_request_queues_locked(rinfo);
1662
1663 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1664
1665 xen_irq_lateeoi(irq, eoiflag);
1666
1667 return IRQ_HANDLED;
1668
1669 err:
1670 info->connected = BLKIF_STATE_ERROR;
1671
1672 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1673
1674 /* No EOI in order to avoid further interrupts. */
1675
1676 pr_alert("%s disabled for further use\n", info->gd->disk_name);
1677 return IRQ_HANDLED;
1678}
1679
1680
1681static int setup_blkring(struct xenbus_device *dev,
1682 struct blkfront_ring_info *rinfo)
1683{
1684 struct blkif_sring *sring;
1685 int err;
1686 struct blkfront_info *info = rinfo->dev_info;
1687 unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1688
1689 err = xenbus_setup_ring(dev, GFP_NOIO, (void **)&sring,
1690 info->nr_ring_pages, rinfo->ring_ref);
1691 if (err)
1692 goto fail;
1693
1694 XEN_FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1695
1696 err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1697 if (err)
1698 goto fail;
1699
1700 err = bind_evtchn_to_irqhandler_lateeoi(rinfo->evtchn, blkif_interrupt,
1701 0, "blkif", rinfo);
1702 if (err <= 0) {
1703 xenbus_dev_fatal(dev, err,
1704 "bind_evtchn_to_irqhandler failed");
1705 goto fail;
1706 }
1707 rinfo->irq = err;
1708
1709 return 0;
1710fail:
1711 blkif_free(info, 0);
1712 return err;
1713}
1714
1715/*
1716 * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1717 * ring buffer may have multi pages depending on ->nr_ring_pages.
1718 */
1719static int write_per_ring_nodes(struct xenbus_transaction xbt,
1720 struct blkfront_ring_info *rinfo, const char *dir)
1721{
1722 int err;
1723 unsigned int i;
1724 const char *message = NULL;
1725 struct blkfront_info *info = rinfo->dev_info;
1726
1727 if (info->nr_ring_pages == 1) {
1728 err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1729 if (err) {
1730 message = "writing ring-ref";
1731 goto abort_transaction;
1732 }
1733 } else {
1734 for (i = 0; i < info->nr_ring_pages; i++) {
1735 char ring_ref_name[RINGREF_NAME_LEN];
1736
1737 snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1738 err = xenbus_printf(xbt, dir, ring_ref_name,
1739 "%u", rinfo->ring_ref[i]);
1740 if (err) {
1741 message = "writing ring-ref";
1742 goto abort_transaction;
1743 }
1744 }
1745 }
1746
1747 err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1748 if (err) {
1749 message = "writing event-channel";
1750 goto abort_transaction;
1751 }
1752
1753 return 0;
1754
1755abort_transaction:
1756 xenbus_transaction_end(xbt, 1);
1757 if (message)
1758 xenbus_dev_fatal(info->xbdev, err, "%s", message);
1759
1760 return err;
1761}
1762
1763/* Enable the persistent grants feature. */
1764static bool feature_persistent = true;
1765module_param(feature_persistent, bool, 0644);
1766MODULE_PARM_DESC(feature_persistent,
1767 "Enables the persistent grants feature");
1768
1769/* Common code used when first setting up, and when resuming. */
1770static int talk_to_blkback(struct xenbus_device *dev,
1771 struct blkfront_info *info)
1772{
1773 const char *message = NULL;
1774 struct xenbus_transaction xbt;
1775 int err;
1776 unsigned int i, max_page_order;
1777 unsigned int ring_page_order;
1778 struct blkfront_ring_info *rinfo;
1779
1780 if (!info)
1781 return -ENODEV;
1782
1783 /* Check if backend is trusted. */
1784 info->bounce = !xen_blkif_trusted ||
1785 !xenbus_read_unsigned(dev->nodename, "trusted", 1);
1786
1787 max_page_order = xenbus_read_unsigned(info->xbdev->otherend,
1788 "max-ring-page-order", 0);
1789 ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1790 info->nr_ring_pages = 1 << ring_page_order;
1791
1792 err = negotiate_mq(info);
1793 if (err)
1794 goto destroy_blkring;
1795
1796 for_each_rinfo(info, rinfo, i) {
1797 /* Create shared ring, alloc event channel. */
1798 err = setup_blkring(dev, rinfo);
1799 if (err)
1800 goto destroy_blkring;
1801 }
1802
1803again:
1804 err = xenbus_transaction_start(&xbt);
1805 if (err) {
1806 xenbus_dev_fatal(dev, err, "starting transaction");
1807 goto destroy_blkring;
1808 }
1809
1810 if (info->nr_ring_pages > 1) {
1811 err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1812 ring_page_order);
1813 if (err) {
1814 message = "writing ring-page-order";
1815 goto abort_transaction;
1816 }
1817 }
1818
1819 /* We already got the number of queues/rings in _probe */
1820 if (info->nr_rings == 1) {
1821 err = write_per_ring_nodes(xbt, info->rinfo, dev->nodename);
1822 if (err)
1823 goto destroy_blkring;
1824 } else {
1825 char *path;
1826 size_t pathsize;
1827
1828 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1829 info->nr_rings);
1830 if (err) {
1831 message = "writing multi-queue-num-queues";
1832 goto abort_transaction;
1833 }
1834
1835 pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1836 path = kmalloc(pathsize, GFP_KERNEL);
1837 if (!path) {
1838 err = -ENOMEM;
1839 message = "ENOMEM while writing ring references";
1840 goto abort_transaction;
1841 }
1842
1843 for_each_rinfo(info, rinfo, i) {
1844 memset(path, 0, pathsize);
1845 snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1846 err = write_per_ring_nodes(xbt, rinfo, path);
1847 if (err) {
1848 kfree(path);
1849 goto destroy_blkring;
1850 }
1851 }
1852 kfree(path);
1853 }
1854 err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1855 XEN_IO_PROTO_ABI_NATIVE);
1856 if (err) {
1857 message = "writing protocol";
1858 goto abort_transaction;
1859 }
1860 info->feature_persistent_parm = feature_persistent;
1861 err = xenbus_printf(xbt, dev->nodename, "feature-persistent", "%u",
1862 info->feature_persistent_parm);
1863 if (err)
1864 dev_warn(&dev->dev,
1865 "writing persistent grants feature to xenbus");
1866
1867 err = xenbus_transaction_end(xbt, 0);
1868 if (err) {
1869 if (err == -EAGAIN)
1870 goto again;
1871 xenbus_dev_fatal(dev, err, "completing transaction");
1872 goto destroy_blkring;
1873 }
1874
1875 for_each_rinfo(info, rinfo, i) {
1876 unsigned int j;
1877
1878 for (j = 0; j < BLK_RING_SIZE(info); j++)
1879 rinfo->shadow[j].req.u.rw.id = j + 1;
1880 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1881 }
1882 xenbus_switch_state(dev, XenbusStateInitialised);
1883
1884 return 0;
1885
1886 abort_transaction:
1887 xenbus_transaction_end(xbt, 1);
1888 if (message)
1889 xenbus_dev_fatal(dev, err, "%s", message);
1890 destroy_blkring:
1891 blkif_free(info, 0);
1892 return err;
1893}
1894
1895static int negotiate_mq(struct blkfront_info *info)
1896{
1897 unsigned int backend_max_queues;
1898 unsigned int i;
1899 struct blkfront_ring_info *rinfo;
1900
1901 BUG_ON(info->nr_rings);
1902
1903 /* Check if backend supports multiple queues. */
1904 backend_max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1905 "multi-queue-max-queues", 1);
1906 info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1907 /* We need at least one ring. */
1908 if (!info->nr_rings)
1909 info->nr_rings = 1;
1910
1911 info->rinfo_size = struct_size(info->rinfo, shadow,
1912 BLK_RING_SIZE(info));
1913 info->rinfo = kvcalloc(info->nr_rings, info->rinfo_size, GFP_KERNEL);
1914 if (!info->rinfo) {
1915 xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1916 info->nr_rings = 0;
1917 return -ENOMEM;
1918 }
1919
1920 for_each_rinfo(info, rinfo, i) {
1921 INIT_LIST_HEAD(&rinfo->indirect_pages);
1922 INIT_LIST_HEAD(&rinfo->grants);
1923 rinfo->dev_info = info;
1924 INIT_WORK(&rinfo->work, blkif_restart_queue);
1925 spin_lock_init(&rinfo->ring_lock);
1926 }
1927 return 0;
1928}
1929
1930/*
1931 * Entry point to this code when a new device is created. Allocate the basic
1932 * structures and the ring buffer for communication with the backend, and
1933 * inform the backend of the appropriate details for those. Switch to
1934 * Initialised state.
1935 */
1936static int blkfront_probe(struct xenbus_device *dev,
1937 const struct xenbus_device_id *id)
1938{
1939 int err, vdevice;
1940 struct blkfront_info *info;
1941
1942 /* FIXME: Use dynamic device id if this is not set. */
1943 err = xenbus_scanf(XBT_NIL, dev->nodename,
1944 "virtual-device", "%i", &vdevice);
1945 if (err != 1) {
1946 /* go looking in the extended area instead */
1947 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1948 "%i", &vdevice);
1949 if (err != 1) {
1950 xenbus_dev_fatal(dev, err, "reading virtual-device");
1951 return err;
1952 }
1953 }
1954
1955 if (xen_hvm_domain()) {
1956 char *type;
1957 int len;
1958 /* no unplug has been done: do not hook devices != xen vbds */
1959 if (xen_has_pv_and_legacy_disk_devices()) {
1960 int major;
1961
1962 if (!VDEV_IS_EXTENDED(vdevice))
1963 major = BLKIF_MAJOR(vdevice);
1964 else
1965 major = XENVBD_MAJOR;
1966
1967 if (major != XENVBD_MAJOR) {
1968 printk(KERN_INFO
1969 "%s: HVM does not support vbd %d as xen block device\n",
1970 __func__, vdevice);
1971 return -ENODEV;
1972 }
1973 }
1974 /* do not create a PV cdrom device if we are an HVM guest */
1975 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1976 if (IS_ERR(type))
1977 return -ENODEV;
1978 if (strncmp(type, "cdrom", 5) == 0) {
1979 kfree(type);
1980 return -ENODEV;
1981 }
1982 kfree(type);
1983 }
1984 info = kzalloc(sizeof(*info), GFP_KERNEL);
1985 if (!info) {
1986 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1987 return -ENOMEM;
1988 }
1989
1990 info->xbdev = dev;
1991
1992 mutex_init(&info->mutex);
1993 info->vdevice = vdevice;
1994 info->connected = BLKIF_STATE_DISCONNECTED;
1995
1996 /* Front end dir is a number, which is used as the id. */
1997 info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1998 dev_set_drvdata(&dev->dev, info);
1999
2000 mutex_lock(&blkfront_mutex);
2001 list_add(&info->info_list, &info_list);
2002 mutex_unlock(&blkfront_mutex);
2003
2004 return 0;
2005}
2006
2007static int blkif_recover(struct blkfront_info *info)
2008{
2009 struct queue_limits lim;
2010 unsigned int r_index;
2011 struct request *req, *n;
2012 int rc;
2013 struct bio *bio;
2014 struct blkfront_ring_info *rinfo;
2015
2016 lim = queue_limits_start_update(info->rq);
2017 blkfront_gather_backend_features(info);
2018 blkif_set_queue_limits(info, &lim);
2019 rc = queue_limits_commit_update(info->rq, &lim);
2020 if (rc)
2021 return rc;
2022
2023 for_each_rinfo(info, rinfo, r_index) {
2024 rc = blkfront_setup_indirect(rinfo);
2025 if (rc)
2026 return rc;
2027 }
2028 xenbus_switch_state(info->xbdev, XenbusStateConnected);
2029
2030 /* Now safe for us to use the shared ring */
2031 info->connected = BLKIF_STATE_CONNECTED;
2032
2033 for_each_rinfo(info, rinfo, r_index) {
2034 /* Kick any other new requests queued since we resumed */
2035 kick_pending_request_queues(rinfo);
2036 }
2037
2038 list_for_each_entry_safe(req, n, &info->requests, queuelist) {
2039 /* Requeue pending requests (flush or discard) */
2040 list_del_init(&req->queuelist);
2041 BUG_ON(req->nr_phys_segments >
2042 (info->max_indirect_segments ? :
2043 BLKIF_MAX_SEGMENTS_PER_REQUEST));
2044 blk_mq_requeue_request(req, false);
2045 }
2046 blk_mq_start_stopped_hw_queues(info->rq, true);
2047 blk_mq_kick_requeue_list(info->rq);
2048
2049 while ((bio = bio_list_pop(&info->bio_list)) != NULL) {
2050 /* Traverse the list of pending bios and re-queue them */
2051 submit_bio(bio);
2052 }
2053
2054 return 0;
2055}
2056
2057/*
2058 * We are reconnecting to the backend, due to a suspend/resume, or a backend
2059 * driver restart. We tear down our blkif structure and recreate it, but
2060 * leave the device-layer structures intact so that this is transparent to the
2061 * rest of the kernel.
2062 */
2063static int blkfront_resume(struct xenbus_device *dev)
2064{
2065 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2066 int err = 0;
2067 unsigned int i, j;
2068 struct blkfront_ring_info *rinfo;
2069
2070 dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2071
2072 bio_list_init(&info->bio_list);
2073 INIT_LIST_HEAD(&info->requests);
2074 for_each_rinfo(info, rinfo, i) {
2075 struct bio_list merge_bio;
2076 struct blk_shadow *shadow = rinfo->shadow;
2077
2078 for (j = 0; j < BLK_RING_SIZE(info); j++) {
2079 /* Not in use? */
2080 if (!shadow[j].request)
2081 continue;
2082
2083 /*
2084 * Get the bios in the request so we can re-queue them.
2085 */
2086 if (req_op(shadow[j].request) == REQ_OP_FLUSH ||
2087 req_op(shadow[j].request) == REQ_OP_DISCARD ||
2088 req_op(shadow[j].request) == REQ_OP_SECURE_ERASE ||
2089 shadow[j].request->cmd_flags & REQ_FUA) {
2090 /*
2091 * Flush operations don't contain bios, so
2092 * we need to requeue the whole request
2093 *
2094 * XXX: but this doesn't make any sense for a
2095 * write with the FUA flag set..
2096 */
2097 list_add(&shadow[j].request->queuelist, &info->requests);
2098 continue;
2099 }
2100 merge_bio.head = shadow[j].request->bio;
2101 merge_bio.tail = shadow[j].request->biotail;
2102 bio_list_merge(&info->bio_list, &merge_bio);
2103 shadow[j].request->bio = NULL;
2104 blk_mq_end_request(shadow[j].request, BLK_STS_OK);
2105 }
2106 }
2107
2108 blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2109
2110 err = talk_to_blkback(dev, info);
2111 if (!err)
2112 blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings);
2113
2114 /*
2115 * We have to wait for the backend to switch to
2116 * connected state, since we want to read which
2117 * features it supports.
2118 */
2119
2120 return err;
2121}
2122
2123static void blkfront_closing(struct blkfront_info *info)
2124{
2125 struct xenbus_device *xbdev = info->xbdev;
2126 struct blkfront_ring_info *rinfo;
2127 unsigned int i;
2128
2129 if (xbdev->state == XenbusStateClosing)
2130 return;
2131
2132 /* No more blkif_request(). */
2133 if (info->rq && info->gd) {
2134 blk_mq_stop_hw_queues(info->rq);
2135 blk_mark_disk_dead(info->gd);
2136 }
2137
2138 for_each_rinfo(info, rinfo, i) {
2139 /* No more gnttab callback work. */
2140 gnttab_cancel_free_callback(&rinfo->callback);
2141
2142 /* Flush gnttab callback work. Must be done with no locks held. */
2143 flush_work(&rinfo->work);
2144 }
2145
2146 xenbus_frontend_closed(xbdev);
2147}
2148
2149static void blkfront_setup_discard(struct blkfront_info *info)
2150{
2151 info->feature_discard = 1;
2152 info->discard_granularity = xenbus_read_unsigned(info->xbdev->otherend,
2153 "discard-granularity",
2154 0);
2155 info->discard_alignment = xenbus_read_unsigned(info->xbdev->otherend,
2156 "discard-alignment", 0);
2157 info->feature_secdiscard =
2158 !!xenbus_read_unsigned(info->xbdev->otherend, "discard-secure",
2159 0);
2160}
2161
2162static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2163{
2164 unsigned int psegs, grants, memflags;
2165 int err, i;
2166 struct blkfront_info *info = rinfo->dev_info;
2167
2168 memflags = memalloc_noio_save();
2169
2170 if (info->max_indirect_segments == 0) {
2171 if (!HAS_EXTRA_REQ)
2172 grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2173 else {
2174 /*
2175 * When an extra req is required, the maximum
2176 * grants supported is related to the size of the
2177 * Linux block segment.
2178 */
2179 grants = GRANTS_PER_PSEG;
2180 }
2181 }
2182 else
2183 grants = info->max_indirect_segments;
2184 psegs = DIV_ROUND_UP(grants, GRANTS_PER_PSEG);
2185
2186 err = fill_grant_buffer(rinfo,
2187 (grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2188 if (err)
2189 goto out_of_memory;
2190
2191 if (!info->bounce && info->max_indirect_segments) {
2192 /*
2193 * We are using indirect descriptors but don't have a bounce
2194 * buffer, we need to allocate a set of pages that can be
2195 * used for mapping indirect grefs
2196 */
2197 int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2198
2199 BUG_ON(!list_empty(&rinfo->indirect_pages));
2200 for (i = 0; i < num; i++) {
2201 struct page *indirect_page = alloc_page(GFP_KERNEL |
2202 __GFP_ZERO);
2203 if (!indirect_page)
2204 goto out_of_memory;
2205 list_add(&indirect_page->lru, &rinfo->indirect_pages);
2206 }
2207 }
2208
2209 for (i = 0; i < BLK_RING_SIZE(info); i++) {
2210 rinfo->shadow[i].grants_used =
2211 kvcalloc(grants,
2212 sizeof(rinfo->shadow[i].grants_used[0]),
2213 GFP_KERNEL);
2214 rinfo->shadow[i].sg = kvcalloc(psegs,
2215 sizeof(rinfo->shadow[i].sg[0]),
2216 GFP_KERNEL);
2217 if (info->max_indirect_segments)
2218 rinfo->shadow[i].indirect_grants =
2219 kvcalloc(INDIRECT_GREFS(grants),
2220 sizeof(rinfo->shadow[i].indirect_grants[0]),
2221 GFP_KERNEL);
2222 if ((rinfo->shadow[i].grants_used == NULL) ||
2223 (rinfo->shadow[i].sg == NULL) ||
2224 (info->max_indirect_segments &&
2225 (rinfo->shadow[i].indirect_grants == NULL)))
2226 goto out_of_memory;
2227 sg_init_table(rinfo->shadow[i].sg, psegs);
2228 }
2229
2230 memalloc_noio_restore(memflags);
2231
2232 return 0;
2233
2234out_of_memory:
2235 for (i = 0; i < BLK_RING_SIZE(info); i++) {
2236 kvfree(rinfo->shadow[i].grants_used);
2237 rinfo->shadow[i].grants_used = NULL;
2238 kvfree(rinfo->shadow[i].sg);
2239 rinfo->shadow[i].sg = NULL;
2240 kvfree(rinfo->shadow[i].indirect_grants);
2241 rinfo->shadow[i].indirect_grants = NULL;
2242 }
2243 if (!list_empty(&rinfo->indirect_pages)) {
2244 struct page *indirect_page, *n;
2245 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2246 list_del(&indirect_page->lru);
2247 __free_page(indirect_page);
2248 }
2249 }
2250
2251 memalloc_noio_restore(memflags);
2252
2253 return -ENOMEM;
2254}
2255
2256/*
2257 * Gather all backend feature-*
2258 */
2259static void blkfront_gather_backend_features(struct blkfront_info *info)
2260{
2261 unsigned int indirect_segments;
2262
2263 info->feature_flush = 0;
2264 info->feature_fua = 0;
2265
2266 /*
2267 * If there's no "feature-barrier" defined, then it means
2268 * we're dealing with a very old backend which writes
2269 * synchronously; nothing to do.
2270 *
2271 * If there are barriers, then we use flush.
2272 */
2273 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-barrier", 0)) {
2274 info->feature_flush = 1;
2275 info->feature_fua = 1;
2276 }
2277
2278 /*
2279 * And if there is "feature-flush-cache" use that above
2280 * barriers.
2281 */
2282 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-flush-cache",
2283 0)) {
2284 info->feature_flush = 1;
2285 info->feature_fua = 0;
2286 }
2287
2288 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-discard", 0))
2289 blkfront_setup_discard(info);
2290
2291 if (info->feature_persistent_parm)
2292 info->feature_persistent =
2293 !!xenbus_read_unsigned(info->xbdev->otherend,
2294 "feature-persistent", 0);
2295 if (info->feature_persistent)
2296 info->bounce = true;
2297
2298 indirect_segments = xenbus_read_unsigned(info->xbdev->otherend,
2299 "feature-max-indirect-segments", 0);
2300 if (indirect_segments > xen_blkif_max_segments)
2301 indirect_segments = xen_blkif_max_segments;
2302 if (indirect_segments <= BLKIF_MAX_SEGMENTS_PER_REQUEST)
2303 indirect_segments = 0;
2304 info->max_indirect_segments = indirect_segments;
2305
2306 if (info->feature_persistent) {
2307 mutex_lock(&blkfront_mutex);
2308 schedule_delayed_work(&blkfront_work, HZ * 10);
2309 mutex_unlock(&blkfront_mutex);
2310 }
2311}
2312
2313/*
2314 * Invoked when the backend is finally 'ready' (and has told produced
2315 * the details about the physical device - #sectors, size, etc).
2316 */
2317static void blkfront_connect(struct blkfront_info *info)
2318{
2319 unsigned long long sectors;
2320 int err, i;
2321 struct blkfront_ring_info *rinfo;
2322
2323 switch (info->connected) {
2324 case BLKIF_STATE_CONNECTED:
2325 /*
2326 * Potentially, the back-end may be signalling
2327 * a capacity change; update the capacity.
2328 */
2329 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2330 "sectors", "%Lu", §ors);
2331 if (XENBUS_EXIST_ERR(err))
2332 return;
2333 printk(KERN_INFO "Setting capacity to %Lu\n",
2334 sectors);
2335 set_capacity_and_notify(info->gd, sectors);
2336
2337 return;
2338 case BLKIF_STATE_SUSPENDED:
2339 /*
2340 * If we are recovering from suspension, we need to wait
2341 * for the backend to announce it's features before
2342 * reconnecting, at least we need to know if the backend
2343 * supports indirect descriptors, and how many.
2344 */
2345 blkif_recover(info);
2346 return;
2347
2348 default:
2349 break;
2350 }
2351
2352 dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2353 __func__, info->xbdev->otherend);
2354
2355 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2356 "sectors", "%llu", §ors,
2357 "info", "%u", &info->vdisk_info,
2358 "sector-size", "%lu", &info->sector_size,
2359 NULL);
2360 if (err) {
2361 xenbus_dev_fatal(info->xbdev, err,
2362 "reading backend fields at %s",
2363 info->xbdev->otherend);
2364 return;
2365 }
2366
2367 /*
2368 * physical-sector-size is a newer field, so old backends may not
2369 * provide this. Assume physical sector size to be the same as
2370 * sector_size in that case.
2371 */
2372 info->physical_sector_size = xenbus_read_unsigned(info->xbdev->otherend,
2373 "physical-sector-size",
2374 info->sector_size);
2375 blkfront_gather_backend_features(info);
2376 for_each_rinfo(info, rinfo, i) {
2377 err = blkfront_setup_indirect(rinfo);
2378 if (err) {
2379 xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2380 info->xbdev->otherend);
2381 blkif_free(info, 0);
2382 break;
2383 }
2384 }
2385
2386 err = xlvbd_alloc_gendisk(sectors, info);
2387 if (err) {
2388 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2389 info->xbdev->otherend);
2390 goto fail;
2391 }
2392
2393 xenbus_switch_state(info->xbdev, XenbusStateConnected);
2394
2395 /* Kick pending requests. */
2396 info->connected = BLKIF_STATE_CONNECTED;
2397 for_each_rinfo(info, rinfo, i)
2398 kick_pending_request_queues(rinfo);
2399
2400 err = device_add_disk(&info->xbdev->dev, info->gd, NULL);
2401 if (err) {
2402 put_disk(info->gd);
2403 blk_mq_free_tag_set(&info->tag_set);
2404 info->rq = NULL;
2405 goto fail;
2406 }
2407
2408 info->is_ready = 1;
2409 return;
2410
2411fail:
2412 blkif_free(info, 0);
2413 return;
2414}
2415
2416/*
2417 * Callback received when the backend's state changes.
2418 */
2419static void blkback_changed(struct xenbus_device *dev,
2420 enum xenbus_state backend_state)
2421{
2422 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2423
2424 dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2425
2426 switch (backend_state) {
2427 case XenbusStateInitWait:
2428 if (dev->state != XenbusStateInitialising)
2429 break;
2430 if (talk_to_blkback(dev, info))
2431 break;
2432 break;
2433 case XenbusStateInitialising:
2434 case XenbusStateInitialised:
2435 case XenbusStateReconfiguring:
2436 case XenbusStateReconfigured:
2437 case XenbusStateUnknown:
2438 break;
2439
2440 case XenbusStateConnected:
2441 /*
2442 * talk_to_blkback sets state to XenbusStateInitialised
2443 * and blkfront_connect sets it to XenbusStateConnected
2444 * (if connection went OK).
2445 *
2446 * If the backend (or toolstack) decides to poke at backend
2447 * state (and re-trigger the watch by setting the state repeatedly
2448 * to XenbusStateConnected (4)) we need to deal with this.
2449 * This is allowed as this is used to communicate to the guest
2450 * that the size of disk has changed!
2451 */
2452 if ((dev->state != XenbusStateInitialised) &&
2453 (dev->state != XenbusStateConnected)) {
2454 if (talk_to_blkback(dev, info))
2455 break;
2456 }
2457
2458 blkfront_connect(info);
2459 break;
2460
2461 case XenbusStateClosed:
2462 if (dev->state == XenbusStateClosed)
2463 break;
2464 fallthrough;
2465 case XenbusStateClosing:
2466 blkfront_closing(info);
2467 break;
2468 }
2469}
2470
2471static void blkfront_remove(struct xenbus_device *xbdev)
2472{
2473 struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2474
2475 dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2476
2477 if (info->gd)
2478 del_gendisk(info->gd);
2479
2480 mutex_lock(&blkfront_mutex);
2481 list_del(&info->info_list);
2482 mutex_unlock(&blkfront_mutex);
2483
2484 blkif_free(info, 0);
2485 if (info->gd) {
2486 xlbd_release_minors(info->gd->first_minor, info->gd->minors);
2487 put_disk(info->gd);
2488 blk_mq_free_tag_set(&info->tag_set);
2489 }
2490
2491 kfree(info);
2492}
2493
2494static int blkfront_is_ready(struct xenbus_device *dev)
2495{
2496 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2497
2498 return info->is_ready && info->xbdev;
2499}
2500
2501static const struct block_device_operations xlvbd_block_fops =
2502{
2503 .owner = THIS_MODULE,
2504 .getgeo = blkif_getgeo,
2505 .ioctl = blkif_ioctl,
2506 .compat_ioctl = blkdev_compat_ptr_ioctl,
2507};
2508
2509
2510static const struct xenbus_device_id blkfront_ids[] = {
2511 { "vbd" },
2512 { "" }
2513};
2514
2515static struct xenbus_driver blkfront_driver = {
2516 .ids = blkfront_ids,
2517 .probe = blkfront_probe,
2518 .remove = blkfront_remove,
2519 .resume = blkfront_resume,
2520 .otherend_changed = blkback_changed,
2521 .is_ready = blkfront_is_ready,
2522};
2523
2524static void purge_persistent_grants(struct blkfront_info *info)
2525{
2526 unsigned int i;
2527 unsigned long flags;
2528 struct blkfront_ring_info *rinfo;
2529
2530 for_each_rinfo(info, rinfo, i) {
2531 struct grant *gnt_list_entry, *tmp;
2532 LIST_HEAD(grants);
2533
2534 spin_lock_irqsave(&rinfo->ring_lock, flags);
2535
2536 if (rinfo->persistent_gnts_c == 0) {
2537 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2538 continue;
2539 }
2540
2541 list_for_each_entry_safe(gnt_list_entry, tmp, &rinfo->grants,
2542 node) {
2543 if (gnt_list_entry->gref == INVALID_GRANT_REF ||
2544 !gnttab_try_end_foreign_access(gnt_list_entry->gref))
2545 continue;
2546
2547 list_del(&gnt_list_entry->node);
2548 rinfo->persistent_gnts_c--;
2549 gnt_list_entry->gref = INVALID_GRANT_REF;
2550 list_add_tail(&gnt_list_entry->node, &grants);
2551 }
2552
2553 list_splice_tail(&grants, &rinfo->grants);
2554
2555 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2556 }
2557}
2558
2559static void blkfront_delay_work(struct work_struct *work)
2560{
2561 struct blkfront_info *info;
2562 bool need_schedule_work = false;
2563
2564 /*
2565 * Note that when using bounce buffers but not persistent grants
2566 * there's no need to run blkfront_delay_work because grants are
2567 * revoked in blkif_completion or else an error is reported and the
2568 * connection is closed.
2569 */
2570
2571 mutex_lock(&blkfront_mutex);
2572
2573 list_for_each_entry(info, &info_list, info_list) {
2574 if (info->feature_persistent) {
2575 need_schedule_work = true;
2576 mutex_lock(&info->mutex);
2577 purge_persistent_grants(info);
2578 mutex_unlock(&info->mutex);
2579 }
2580 }
2581
2582 if (need_schedule_work)
2583 schedule_delayed_work(&blkfront_work, HZ * 10);
2584
2585 mutex_unlock(&blkfront_mutex);
2586}
2587
2588static int __init xlblk_init(void)
2589{
2590 int ret;
2591 int nr_cpus = num_online_cpus();
2592
2593 if (!xen_domain())
2594 return -ENODEV;
2595
2596 if (!xen_has_pv_disk_devices())
2597 return -ENODEV;
2598
2599 if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2600 pr_warn("xen_blk: can't get major %d with name %s\n",
2601 XENVBD_MAJOR, DEV_NAME);
2602 return -ENODEV;
2603 }
2604
2605 if (xen_blkif_max_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST)
2606 xen_blkif_max_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2607
2608 if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2609 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2610 xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2611 xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2612 }
2613
2614 if (xen_blkif_max_queues > nr_cpus) {
2615 pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2616 xen_blkif_max_queues, nr_cpus);
2617 xen_blkif_max_queues = nr_cpus;
2618 }
2619
2620 INIT_DELAYED_WORK(&blkfront_work, blkfront_delay_work);
2621
2622 ret = xenbus_register_frontend(&blkfront_driver);
2623 if (ret) {
2624 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2625 return ret;
2626 }
2627
2628 return 0;
2629}
2630module_init(xlblk_init);
2631
2632
2633static void __exit xlblk_exit(void)
2634{
2635 cancel_delayed_work_sync(&blkfront_work);
2636
2637 xenbus_unregister_driver(&blkfront_driver);
2638 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2639 kfree(minors);
2640}
2641module_exit(xlblk_exit);
2642
2643MODULE_DESCRIPTION("Xen virtual block device frontend");
2644MODULE_LICENSE("GPL");
2645MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2646MODULE_ALIAS("xen:vbd");
2647MODULE_ALIAS("xenblk");