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
47#include <xen/xen.h>
48#include <xen/xenbus.h>
49#include <xen/grant_table.h>
50#include <xen/events.h>
51#include <xen/page.h>
52#include <xen/platform_pci.h>
53
54#include <xen/interface/grant_table.h>
55#include <xen/interface/io/blkif.h>
56#include <xen/interface/io/protocols.h>
57
58#include <asm/xen/hypervisor.h>
59
60enum blkif_state {
61 BLKIF_STATE_DISCONNECTED,
62 BLKIF_STATE_CONNECTED,
63 BLKIF_STATE_SUSPENDED,
64};
65
66struct blk_shadow {
67 struct blkif_request req;
68 struct request *request;
69 unsigned long frame[BLKIF_MAX_SEGMENTS_PER_REQUEST];
70};
71
72static DEFINE_MUTEX(blkfront_mutex);
73static const struct block_device_operations xlvbd_block_fops;
74
75#define BLK_RING_SIZE __CONST_RING_SIZE(blkif, PAGE_SIZE)
76
77/*
78 * We have one of these per vbd, whether ide, scsi or 'other'. They
79 * hang in private_data off the gendisk structure. We may end up
80 * putting all kinds of interesting stuff here :-)
81 */
82struct blkfront_info
83{
84 struct mutex mutex;
85 struct xenbus_device *xbdev;
86 struct gendisk *gd;
87 int vdevice;
88 blkif_vdev_t handle;
89 enum blkif_state connected;
90 int ring_ref;
91 struct blkif_front_ring ring;
92 struct scatterlist sg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
93 unsigned int evtchn, irq;
94 struct request_queue *rq;
95 struct work_struct work;
96 struct gnttab_free_callback callback;
97 struct blk_shadow shadow[BLK_RING_SIZE];
98 unsigned long shadow_free;
99 unsigned int feature_flush;
100 unsigned int flush_op;
101 int is_ready;
102};
103
104static DEFINE_SPINLOCK(blkif_io_lock);
105
106static unsigned int nr_minors;
107static unsigned long *minors;
108static DEFINE_SPINLOCK(minor_lock);
109
110#define MAXIMUM_OUTSTANDING_BLOCK_REQS \
111 (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
112#define GRANT_INVALID_REF 0
113
114#define PARTS_PER_DISK 16
115#define PARTS_PER_EXT_DISK 256
116
117#define BLKIF_MAJOR(dev) ((dev)>>8)
118#define BLKIF_MINOR(dev) ((dev) & 0xff)
119
120#define EXT_SHIFT 28
121#define EXTENDED (1<<EXT_SHIFT)
122#define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
123#define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
124#define EMULATED_HD_DISK_MINOR_OFFSET (0)
125#define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
126#define EMULATED_SD_DISK_MINOR_OFFSET (0)
127#define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
128
129#define DEV_NAME "xvd" /* name in /dev */
130
131static int get_id_from_freelist(struct blkfront_info *info)
132{
133 unsigned long free = info->shadow_free;
134 BUG_ON(free >= BLK_RING_SIZE);
135 info->shadow_free = info->shadow[free].req.id;
136 info->shadow[free].req.id = 0x0fffffee; /* debug */
137 return free;
138}
139
140static void add_id_to_freelist(struct blkfront_info *info,
141 unsigned long id)
142{
143 info->shadow[id].req.id = info->shadow_free;
144 info->shadow[id].request = NULL;
145 info->shadow_free = id;
146}
147
148static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
149{
150 unsigned int end = minor + nr;
151 int rc;
152
153 if (end > nr_minors) {
154 unsigned long *bitmap, *old;
155
156 bitmap = kzalloc(BITS_TO_LONGS(end) * sizeof(*bitmap),
157 GFP_KERNEL);
158 if (bitmap == NULL)
159 return -ENOMEM;
160
161 spin_lock(&minor_lock);
162 if (end > nr_minors) {
163 old = minors;
164 memcpy(bitmap, minors,
165 BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
166 minors = bitmap;
167 nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
168 } else
169 old = bitmap;
170 spin_unlock(&minor_lock);
171 kfree(old);
172 }
173
174 spin_lock(&minor_lock);
175 if (find_next_bit(minors, end, minor) >= end) {
176 for (; minor < end; ++minor)
177 __set_bit(minor, minors);
178 rc = 0;
179 } else
180 rc = -EBUSY;
181 spin_unlock(&minor_lock);
182
183 return rc;
184}
185
186static void xlbd_release_minors(unsigned int minor, unsigned int nr)
187{
188 unsigned int end = minor + nr;
189
190 BUG_ON(end > nr_minors);
191 spin_lock(&minor_lock);
192 for (; minor < end; ++minor)
193 __clear_bit(minor, minors);
194 spin_unlock(&minor_lock);
195}
196
197static void blkif_restart_queue_callback(void *arg)
198{
199 struct blkfront_info *info = (struct blkfront_info *)arg;
200 schedule_work(&info->work);
201}
202
203static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
204{
205 /* We don't have real geometry info, but let's at least return
206 values consistent with the size of the device */
207 sector_t nsect = get_capacity(bd->bd_disk);
208 sector_t cylinders = nsect;
209
210 hg->heads = 0xff;
211 hg->sectors = 0x3f;
212 sector_div(cylinders, hg->heads * hg->sectors);
213 hg->cylinders = cylinders;
214 if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
215 hg->cylinders = 0xffff;
216 return 0;
217}
218
219static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
220 unsigned command, unsigned long argument)
221{
222 struct blkfront_info *info = bdev->bd_disk->private_data;
223 int i;
224
225 dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
226 command, (long)argument);
227
228 switch (command) {
229 case CDROMMULTISESSION:
230 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
231 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
232 if (put_user(0, (char __user *)(argument + i)))
233 return -EFAULT;
234 return 0;
235
236 case CDROM_GET_CAPABILITY: {
237 struct gendisk *gd = info->gd;
238 if (gd->flags & GENHD_FL_CD)
239 return 0;
240 return -EINVAL;
241 }
242
243 default:
244 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
245 command);*/
246 return -EINVAL; /* same return as native Linux */
247 }
248
249 return 0;
250}
251
252/*
253 * Generate a Xen blkfront IO request from a blk layer request. Reads
254 * and writes are handled as expected.
255 *
256 * @req: a request struct
257 */
258static int blkif_queue_request(struct request *req)
259{
260 struct blkfront_info *info = req->rq_disk->private_data;
261 unsigned long buffer_mfn;
262 struct blkif_request *ring_req;
263 unsigned long id;
264 unsigned int fsect, lsect;
265 int i, ref;
266 grant_ref_t gref_head;
267 struct scatterlist *sg;
268
269 if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
270 return 1;
271
272 if (gnttab_alloc_grant_references(
273 BLKIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) {
274 gnttab_request_free_callback(
275 &info->callback,
276 blkif_restart_queue_callback,
277 info,
278 BLKIF_MAX_SEGMENTS_PER_REQUEST);
279 return 1;
280 }
281
282 /* Fill out a communications ring structure. */
283 ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
284 id = get_id_from_freelist(info);
285 info->shadow[id].request = req;
286
287 ring_req->id = id;
288 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
289 ring_req->handle = info->handle;
290
291 ring_req->operation = rq_data_dir(req) ?
292 BLKIF_OP_WRITE : BLKIF_OP_READ;
293
294 if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) {
295 /*
296 * Ideally we can do an unordered flush-to-disk. In case the
297 * backend onlysupports barriers, use that. A barrier request
298 * a superset of FUA, so we can implement it the same
299 * way. (It's also a FLUSH+FUA, since it is
300 * guaranteed ordered WRT previous writes.)
301 */
302 ring_req->operation = info->flush_op;
303 }
304
305 ring_req->nr_segments = blk_rq_map_sg(req->q, req, info->sg);
306 BUG_ON(ring_req->nr_segments > BLKIF_MAX_SEGMENTS_PER_REQUEST);
307
308 for_each_sg(info->sg, sg, ring_req->nr_segments, i) {
309 buffer_mfn = pfn_to_mfn(page_to_pfn(sg_page(sg)));
310 fsect = sg->offset >> 9;
311 lsect = fsect + (sg->length >> 9) - 1;
312 /* install a grant reference. */
313 ref = gnttab_claim_grant_reference(&gref_head);
314 BUG_ON(ref == -ENOSPC);
315
316 gnttab_grant_foreign_access_ref(
317 ref,
318 info->xbdev->otherend_id,
319 buffer_mfn,
320 rq_data_dir(req) );
321
322 info->shadow[id].frame[i] = mfn_to_pfn(buffer_mfn);
323 ring_req->u.rw.seg[i] =
324 (struct blkif_request_segment) {
325 .gref = ref,
326 .first_sect = fsect,
327 .last_sect = lsect };
328 }
329
330 info->ring.req_prod_pvt++;
331
332 /* Keep a private copy so we can reissue requests when recovering. */
333 info->shadow[id].req = *ring_req;
334
335 gnttab_free_grant_references(gref_head);
336
337 return 0;
338}
339
340
341static inline void flush_requests(struct blkfront_info *info)
342{
343 int notify;
344
345 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
346
347 if (notify)
348 notify_remote_via_irq(info->irq);
349}
350
351/*
352 * do_blkif_request
353 * read a block; request is in a request queue
354 */
355static void do_blkif_request(struct request_queue *rq)
356{
357 struct blkfront_info *info = NULL;
358 struct request *req;
359 int queued;
360
361 pr_debug("Entered do_blkif_request\n");
362
363 queued = 0;
364
365 while ((req = blk_peek_request(rq)) != NULL) {
366 info = req->rq_disk->private_data;
367
368 if (RING_FULL(&info->ring))
369 goto wait;
370
371 blk_start_request(req);
372
373 if (req->cmd_type != REQ_TYPE_FS) {
374 __blk_end_request_all(req, -EIO);
375 continue;
376 }
377
378 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
379 "(%u/%u) buffer:%p [%s]\n",
380 req, req->cmd, (unsigned long)blk_rq_pos(req),
381 blk_rq_cur_sectors(req), blk_rq_sectors(req),
382 req->buffer, rq_data_dir(req) ? "write" : "read");
383
384 if (blkif_queue_request(req)) {
385 blk_requeue_request(rq, req);
386wait:
387 /* Avoid pointless unplugs. */
388 blk_stop_queue(rq);
389 break;
390 }
391
392 queued++;
393 }
394
395 if (queued != 0)
396 flush_requests(info);
397}
398
399static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
400{
401 struct request_queue *rq;
402
403 rq = blk_init_queue(do_blkif_request, &blkif_io_lock);
404 if (rq == NULL)
405 return -1;
406
407 queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
408
409 /* Hard sector size and max sectors impersonate the equiv. hardware. */
410 blk_queue_logical_block_size(rq, sector_size);
411 blk_queue_max_hw_sectors(rq, 512);
412
413 /* Each segment in a request is up to an aligned page in size. */
414 blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
415 blk_queue_max_segment_size(rq, PAGE_SIZE);
416
417 /* Ensure a merged request will fit in a single I/O ring slot. */
418 blk_queue_max_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
419
420 /* Make sure buffer addresses are sector-aligned. */
421 blk_queue_dma_alignment(rq, 511);
422
423 /* Make sure we don't use bounce buffers. */
424 blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
425
426 gd->queue = rq;
427
428 return 0;
429}
430
431
432static void xlvbd_flush(struct blkfront_info *info)
433{
434 blk_queue_flush(info->rq, info->feature_flush);
435 printk(KERN_INFO "blkfront: %s: %s: %s\n",
436 info->gd->disk_name,
437 info->flush_op == BLKIF_OP_WRITE_BARRIER ?
438 "barrier" : (info->flush_op == BLKIF_OP_FLUSH_DISKCACHE ?
439 "flush diskcache" : "barrier or flush"),
440 info->feature_flush ? "enabled" : "disabled");
441}
442
443static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
444{
445 int major;
446 major = BLKIF_MAJOR(vdevice);
447 *minor = BLKIF_MINOR(vdevice);
448 switch (major) {
449 case XEN_IDE0_MAJOR:
450 *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
451 *minor = ((*minor / 64) * PARTS_PER_DISK) +
452 EMULATED_HD_DISK_MINOR_OFFSET;
453 break;
454 case XEN_IDE1_MAJOR:
455 *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
456 *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
457 EMULATED_HD_DISK_MINOR_OFFSET;
458 break;
459 case XEN_SCSI_DISK0_MAJOR:
460 *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
461 *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
462 break;
463 case XEN_SCSI_DISK1_MAJOR:
464 case XEN_SCSI_DISK2_MAJOR:
465 case XEN_SCSI_DISK3_MAJOR:
466 case XEN_SCSI_DISK4_MAJOR:
467 case XEN_SCSI_DISK5_MAJOR:
468 case XEN_SCSI_DISK6_MAJOR:
469 case XEN_SCSI_DISK7_MAJOR:
470 *offset = (*minor / PARTS_PER_DISK) +
471 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
472 EMULATED_SD_DISK_NAME_OFFSET;
473 *minor = *minor +
474 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
475 EMULATED_SD_DISK_MINOR_OFFSET;
476 break;
477 case XEN_SCSI_DISK8_MAJOR:
478 case XEN_SCSI_DISK9_MAJOR:
479 case XEN_SCSI_DISK10_MAJOR:
480 case XEN_SCSI_DISK11_MAJOR:
481 case XEN_SCSI_DISK12_MAJOR:
482 case XEN_SCSI_DISK13_MAJOR:
483 case XEN_SCSI_DISK14_MAJOR:
484 case XEN_SCSI_DISK15_MAJOR:
485 *offset = (*minor / PARTS_PER_DISK) +
486 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
487 EMULATED_SD_DISK_NAME_OFFSET;
488 *minor = *minor +
489 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
490 EMULATED_SD_DISK_MINOR_OFFSET;
491 break;
492 case XENVBD_MAJOR:
493 *offset = *minor / PARTS_PER_DISK;
494 break;
495 default:
496 printk(KERN_WARNING "blkfront: your disk configuration is "
497 "incorrect, please use an xvd device instead\n");
498 return -ENODEV;
499 }
500 return 0;
501}
502
503static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
504 struct blkfront_info *info,
505 u16 vdisk_info, u16 sector_size)
506{
507 struct gendisk *gd;
508 int nr_minors = 1;
509 int err;
510 unsigned int offset;
511 int minor;
512 int nr_parts;
513
514 BUG_ON(info->gd != NULL);
515 BUG_ON(info->rq != NULL);
516
517 if ((info->vdevice>>EXT_SHIFT) > 1) {
518 /* this is above the extended range; something is wrong */
519 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
520 return -ENODEV;
521 }
522
523 if (!VDEV_IS_EXTENDED(info->vdevice)) {
524 err = xen_translate_vdev(info->vdevice, &minor, &offset);
525 if (err)
526 return err;
527 nr_parts = PARTS_PER_DISK;
528 } else {
529 minor = BLKIF_MINOR_EXT(info->vdevice);
530 nr_parts = PARTS_PER_EXT_DISK;
531 offset = minor / nr_parts;
532 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
533 printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
534 "emulated IDE disks,\n\t choose an xvd device name"
535 "from xvde on\n", info->vdevice);
536 }
537 err = -ENODEV;
538
539 if ((minor % nr_parts) == 0)
540 nr_minors = nr_parts;
541
542 err = xlbd_reserve_minors(minor, nr_minors);
543 if (err)
544 goto out;
545 err = -ENODEV;
546
547 gd = alloc_disk(nr_minors);
548 if (gd == NULL)
549 goto release;
550
551 if (nr_minors > 1) {
552 if (offset < 26)
553 sprintf(gd->disk_name, "%s%c", DEV_NAME, 'a' + offset);
554 else
555 sprintf(gd->disk_name, "%s%c%c", DEV_NAME,
556 'a' + ((offset / 26)-1), 'a' + (offset % 26));
557 } else {
558 if (offset < 26)
559 sprintf(gd->disk_name, "%s%c%d", DEV_NAME,
560 'a' + offset,
561 minor & (nr_parts - 1));
562 else
563 sprintf(gd->disk_name, "%s%c%c%d", DEV_NAME,
564 'a' + ((offset / 26) - 1),
565 'a' + (offset % 26),
566 minor & (nr_parts - 1));
567 }
568
569 gd->major = XENVBD_MAJOR;
570 gd->first_minor = minor;
571 gd->fops = &xlvbd_block_fops;
572 gd->private_data = info;
573 gd->driverfs_dev = &(info->xbdev->dev);
574 set_capacity(gd, capacity);
575
576 if (xlvbd_init_blk_queue(gd, sector_size)) {
577 del_gendisk(gd);
578 goto release;
579 }
580
581 info->rq = gd->queue;
582 info->gd = gd;
583
584 xlvbd_flush(info);
585
586 if (vdisk_info & VDISK_READONLY)
587 set_disk_ro(gd, 1);
588
589 if (vdisk_info & VDISK_REMOVABLE)
590 gd->flags |= GENHD_FL_REMOVABLE;
591
592 if (vdisk_info & VDISK_CDROM)
593 gd->flags |= GENHD_FL_CD;
594
595 return 0;
596
597 release:
598 xlbd_release_minors(minor, nr_minors);
599 out:
600 return err;
601}
602
603static void xlvbd_release_gendisk(struct blkfront_info *info)
604{
605 unsigned int minor, nr_minors;
606 unsigned long flags;
607
608 if (info->rq == NULL)
609 return;
610
611 spin_lock_irqsave(&blkif_io_lock, flags);
612
613 /* No more blkif_request(). */
614 blk_stop_queue(info->rq);
615
616 /* No more gnttab callback work. */
617 gnttab_cancel_free_callback(&info->callback);
618 spin_unlock_irqrestore(&blkif_io_lock, flags);
619
620 /* Flush gnttab callback work. Must be done with no locks held. */
621 flush_work_sync(&info->work);
622
623 del_gendisk(info->gd);
624
625 minor = info->gd->first_minor;
626 nr_minors = info->gd->minors;
627 xlbd_release_minors(minor, nr_minors);
628
629 blk_cleanup_queue(info->rq);
630 info->rq = NULL;
631
632 put_disk(info->gd);
633 info->gd = NULL;
634}
635
636static void kick_pending_request_queues(struct blkfront_info *info)
637{
638 if (!RING_FULL(&info->ring)) {
639 /* Re-enable calldowns. */
640 blk_start_queue(info->rq);
641 /* Kick things off immediately. */
642 do_blkif_request(info->rq);
643 }
644}
645
646static void blkif_restart_queue(struct work_struct *work)
647{
648 struct blkfront_info *info = container_of(work, struct blkfront_info, work);
649
650 spin_lock_irq(&blkif_io_lock);
651 if (info->connected == BLKIF_STATE_CONNECTED)
652 kick_pending_request_queues(info);
653 spin_unlock_irq(&blkif_io_lock);
654}
655
656static void blkif_free(struct blkfront_info *info, int suspend)
657{
658 /* Prevent new requests being issued until we fix things up. */
659 spin_lock_irq(&blkif_io_lock);
660 info->connected = suspend ?
661 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
662 /* No more blkif_request(). */
663 if (info->rq)
664 blk_stop_queue(info->rq);
665 /* No more gnttab callback work. */
666 gnttab_cancel_free_callback(&info->callback);
667 spin_unlock_irq(&blkif_io_lock);
668
669 /* Flush gnttab callback work. Must be done with no locks held. */
670 flush_work_sync(&info->work);
671
672 /* Free resources associated with old device channel. */
673 if (info->ring_ref != GRANT_INVALID_REF) {
674 gnttab_end_foreign_access(info->ring_ref, 0,
675 (unsigned long)info->ring.sring);
676 info->ring_ref = GRANT_INVALID_REF;
677 info->ring.sring = NULL;
678 }
679 if (info->irq)
680 unbind_from_irqhandler(info->irq, info);
681 info->evtchn = info->irq = 0;
682
683}
684
685static void blkif_completion(struct blk_shadow *s)
686{
687 int i;
688 for (i = 0; i < s->req.nr_segments; i++)
689 gnttab_end_foreign_access(s->req.u.rw.seg[i].gref, 0, 0UL);
690}
691
692static irqreturn_t blkif_interrupt(int irq, void *dev_id)
693{
694 struct request *req;
695 struct blkif_response *bret;
696 RING_IDX i, rp;
697 unsigned long flags;
698 struct blkfront_info *info = (struct blkfront_info *)dev_id;
699 int error;
700
701 spin_lock_irqsave(&blkif_io_lock, flags);
702
703 if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
704 spin_unlock_irqrestore(&blkif_io_lock, flags);
705 return IRQ_HANDLED;
706 }
707
708 again:
709 rp = info->ring.sring->rsp_prod;
710 rmb(); /* Ensure we see queued responses up to 'rp'. */
711
712 for (i = info->ring.rsp_cons; i != rp; i++) {
713 unsigned long id;
714
715 bret = RING_GET_RESPONSE(&info->ring, i);
716 id = bret->id;
717 req = info->shadow[id].request;
718
719 blkif_completion(&info->shadow[id]);
720
721 add_id_to_freelist(info, id);
722
723 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
724 switch (bret->operation) {
725 case BLKIF_OP_FLUSH_DISKCACHE:
726 case BLKIF_OP_WRITE_BARRIER:
727 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
728 printk(KERN_WARNING "blkfront: %s: write %s op failed\n",
729 info->flush_op == BLKIF_OP_WRITE_BARRIER ?
730 "barrier" : "flush disk cache",
731 info->gd->disk_name);
732 error = -EOPNOTSUPP;
733 }
734 if (unlikely(bret->status == BLKIF_RSP_ERROR &&
735 info->shadow[id].req.nr_segments == 0)) {
736 printk(KERN_WARNING "blkfront: %s: empty write %s op failed\n",
737 info->flush_op == BLKIF_OP_WRITE_BARRIER ?
738 "barrier" : "flush disk cache",
739 info->gd->disk_name);
740 error = -EOPNOTSUPP;
741 }
742 if (unlikely(error)) {
743 if (error == -EOPNOTSUPP)
744 error = 0;
745 info->feature_flush = 0;
746 info->flush_op = 0;
747 xlvbd_flush(info);
748 }
749 /* fall through */
750 case BLKIF_OP_READ:
751 case BLKIF_OP_WRITE:
752 if (unlikely(bret->status != BLKIF_RSP_OKAY))
753 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
754 "request: %x\n", bret->status);
755
756 __blk_end_request_all(req, error);
757 break;
758 default:
759 BUG();
760 }
761 }
762
763 info->ring.rsp_cons = i;
764
765 if (i != info->ring.req_prod_pvt) {
766 int more_to_do;
767 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
768 if (more_to_do)
769 goto again;
770 } else
771 info->ring.sring->rsp_event = i + 1;
772
773 kick_pending_request_queues(info);
774
775 spin_unlock_irqrestore(&blkif_io_lock, flags);
776
777 return IRQ_HANDLED;
778}
779
780
781static int setup_blkring(struct xenbus_device *dev,
782 struct blkfront_info *info)
783{
784 struct blkif_sring *sring;
785 int err;
786
787 info->ring_ref = GRANT_INVALID_REF;
788
789 sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
790 if (!sring) {
791 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
792 return -ENOMEM;
793 }
794 SHARED_RING_INIT(sring);
795 FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
796
797 sg_init_table(info->sg, BLKIF_MAX_SEGMENTS_PER_REQUEST);
798
799 err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
800 if (err < 0) {
801 free_page((unsigned long)sring);
802 info->ring.sring = NULL;
803 goto fail;
804 }
805 info->ring_ref = err;
806
807 err = xenbus_alloc_evtchn(dev, &info->evtchn);
808 if (err)
809 goto fail;
810
811 err = bind_evtchn_to_irqhandler(info->evtchn,
812 blkif_interrupt,
813 IRQF_SAMPLE_RANDOM, "blkif", info);
814 if (err <= 0) {
815 xenbus_dev_fatal(dev, err,
816 "bind_evtchn_to_irqhandler failed");
817 goto fail;
818 }
819 info->irq = err;
820
821 return 0;
822fail:
823 blkif_free(info, 0);
824 return err;
825}
826
827
828/* Common code used when first setting up, and when resuming. */
829static int talk_to_blkback(struct xenbus_device *dev,
830 struct blkfront_info *info)
831{
832 const char *message = NULL;
833 struct xenbus_transaction xbt;
834 int err;
835
836 /* Create shared ring, alloc event channel. */
837 err = setup_blkring(dev, info);
838 if (err)
839 goto out;
840
841again:
842 err = xenbus_transaction_start(&xbt);
843 if (err) {
844 xenbus_dev_fatal(dev, err, "starting transaction");
845 goto destroy_blkring;
846 }
847
848 err = xenbus_printf(xbt, dev->nodename,
849 "ring-ref", "%u", info->ring_ref);
850 if (err) {
851 message = "writing ring-ref";
852 goto abort_transaction;
853 }
854 err = xenbus_printf(xbt, dev->nodename,
855 "event-channel", "%u", info->evtchn);
856 if (err) {
857 message = "writing event-channel";
858 goto abort_transaction;
859 }
860 err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
861 XEN_IO_PROTO_ABI_NATIVE);
862 if (err) {
863 message = "writing protocol";
864 goto abort_transaction;
865 }
866
867 err = xenbus_transaction_end(xbt, 0);
868 if (err) {
869 if (err == -EAGAIN)
870 goto again;
871 xenbus_dev_fatal(dev, err, "completing transaction");
872 goto destroy_blkring;
873 }
874
875 xenbus_switch_state(dev, XenbusStateInitialised);
876
877 return 0;
878
879 abort_transaction:
880 xenbus_transaction_end(xbt, 1);
881 if (message)
882 xenbus_dev_fatal(dev, err, "%s", message);
883 destroy_blkring:
884 blkif_free(info, 0);
885 out:
886 return err;
887}
888
889/**
890 * Entry point to this code when a new device is created. Allocate the basic
891 * structures and the ring buffer for communication with the backend, and
892 * inform the backend of the appropriate details for those. Switch to
893 * Initialised state.
894 */
895static int blkfront_probe(struct xenbus_device *dev,
896 const struct xenbus_device_id *id)
897{
898 int err, vdevice, i;
899 struct blkfront_info *info;
900
901 /* FIXME: Use dynamic device id if this is not set. */
902 err = xenbus_scanf(XBT_NIL, dev->nodename,
903 "virtual-device", "%i", &vdevice);
904 if (err != 1) {
905 /* go looking in the extended area instead */
906 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
907 "%i", &vdevice);
908 if (err != 1) {
909 xenbus_dev_fatal(dev, err, "reading virtual-device");
910 return err;
911 }
912 }
913
914 if (xen_hvm_domain()) {
915 char *type;
916 int len;
917 /* no unplug has been done: do not hook devices != xen vbds */
918 if (xen_platform_pci_unplug & XEN_UNPLUG_UNNECESSARY) {
919 int major;
920
921 if (!VDEV_IS_EXTENDED(vdevice))
922 major = BLKIF_MAJOR(vdevice);
923 else
924 major = XENVBD_MAJOR;
925
926 if (major != XENVBD_MAJOR) {
927 printk(KERN_INFO
928 "%s: HVM does not support vbd %d as xen block device\n",
929 __FUNCTION__, vdevice);
930 return -ENODEV;
931 }
932 }
933 /* do not create a PV cdrom device if we are an HVM guest */
934 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
935 if (IS_ERR(type))
936 return -ENODEV;
937 if (strncmp(type, "cdrom", 5) == 0) {
938 kfree(type);
939 return -ENODEV;
940 }
941 kfree(type);
942 }
943 info = kzalloc(sizeof(*info), GFP_KERNEL);
944 if (!info) {
945 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
946 return -ENOMEM;
947 }
948
949 mutex_init(&info->mutex);
950 info->xbdev = dev;
951 info->vdevice = vdevice;
952 info->connected = BLKIF_STATE_DISCONNECTED;
953 INIT_WORK(&info->work, blkif_restart_queue);
954
955 for (i = 0; i < BLK_RING_SIZE; i++)
956 info->shadow[i].req.id = i+1;
957 info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
958
959 /* Front end dir is a number, which is used as the id. */
960 info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
961 dev_set_drvdata(&dev->dev, info);
962
963 err = talk_to_blkback(dev, info);
964 if (err) {
965 kfree(info);
966 dev_set_drvdata(&dev->dev, NULL);
967 return err;
968 }
969
970 return 0;
971}
972
973
974static int blkif_recover(struct blkfront_info *info)
975{
976 int i;
977 struct blkif_request *req;
978 struct blk_shadow *copy;
979 int j;
980
981 /* Stage 1: Make a safe copy of the shadow state. */
982 copy = kmalloc(sizeof(info->shadow),
983 GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
984 if (!copy)
985 return -ENOMEM;
986 memcpy(copy, info->shadow, sizeof(info->shadow));
987
988 /* Stage 2: Set up free list. */
989 memset(&info->shadow, 0, sizeof(info->shadow));
990 for (i = 0; i < BLK_RING_SIZE; i++)
991 info->shadow[i].req.id = i+1;
992 info->shadow_free = info->ring.req_prod_pvt;
993 info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
994
995 /* Stage 3: Find pending requests and requeue them. */
996 for (i = 0; i < BLK_RING_SIZE; i++) {
997 /* Not in use? */
998 if (!copy[i].request)
999 continue;
1000
1001 /* Grab a request slot and copy shadow state into it. */
1002 req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
1003 *req = copy[i].req;
1004
1005 /* We get a new request id, and must reset the shadow state. */
1006 req->id = get_id_from_freelist(info);
1007 memcpy(&info->shadow[req->id], ©[i], sizeof(copy[i]));
1008
1009 /* Rewrite any grant references invalidated by susp/resume. */
1010 for (j = 0; j < req->nr_segments; j++)
1011 gnttab_grant_foreign_access_ref(
1012 req->u.rw.seg[j].gref,
1013 info->xbdev->otherend_id,
1014 pfn_to_mfn(info->shadow[req->id].frame[j]),
1015 rq_data_dir(info->shadow[req->id].request));
1016 info->shadow[req->id].req = *req;
1017
1018 info->ring.req_prod_pvt++;
1019 }
1020
1021 kfree(copy);
1022
1023 xenbus_switch_state(info->xbdev, XenbusStateConnected);
1024
1025 spin_lock_irq(&blkif_io_lock);
1026
1027 /* Now safe for us to use the shared ring */
1028 info->connected = BLKIF_STATE_CONNECTED;
1029
1030 /* Send off requeued requests */
1031 flush_requests(info);
1032
1033 /* Kick any other new requests queued since we resumed */
1034 kick_pending_request_queues(info);
1035
1036 spin_unlock_irq(&blkif_io_lock);
1037
1038 return 0;
1039}
1040
1041/**
1042 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1043 * driver restart. We tear down our blkif structure and recreate it, but
1044 * leave the device-layer structures intact so that this is transparent to the
1045 * rest of the kernel.
1046 */
1047static int blkfront_resume(struct xenbus_device *dev)
1048{
1049 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1050 int err;
1051
1052 dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
1053
1054 blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
1055
1056 err = talk_to_blkback(dev, info);
1057 if (info->connected == BLKIF_STATE_SUSPENDED && !err)
1058 err = blkif_recover(info);
1059
1060 return err;
1061}
1062
1063static void
1064blkfront_closing(struct blkfront_info *info)
1065{
1066 struct xenbus_device *xbdev = info->xbdev;
1067 struct block_device *bdev = NULL;
1068
1069 mutex_lock(&info->mutex);
1070
1071 if (xbdev->state == XenbusStateClosing) {
1072 mutex_unlock(&info->mutex);
1073 return;
1074 }
1075
1076 if (info->gd)
1077 bdev = bdget_disk(info->gd, 0);
1078
1079 mutex_unlock(&info->mutex);
1080
1081 if (!bdev) {
1082 xenbus_frontend_closed(xbdev);
1083 return;
1084 }
1085
1086 mutex_lock(&bdev->bd_mutex);
1087
1088 if (bdev->bd_openers) {
1089 xenbus_dev_error(xbdev, -EBUSY,
1090 "Device in use; refusing to close");
1091 xenbus_switch_state(xbdev, XenbusStateClosing);
1092 } else {
1093 xlvbd_release_gendisk(info);
1094 xenbus_frontend_closed(xbdev);
1095 }
1096
1097 mutex_unlock(&bdev->bd_mutex);
1098 bdput(bdev);
1099}
1100
1101/*
1102 * Invoked when the backend is finally 'ready' (and has told produced
1103 * the details about the physical device - #sectors, size, etc).
1104 */
1105static void blkfront_connect(struct blkfront_info *info)
1106{
1107 unsigned long long sectors;
1108 unsigned long sector_size;
1109 unsigned int binfo;
1110 int err;
1111 int barrier, flush;
1112
1113 switch (info->connected) {
1114 case BLKIF_STATE_CONNECTED:
1115 /*
1116 * Potentially, the back-end may be signalling
1117 * a capacity change; update the capacity.
1118 */
1119 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1120 "sectors", "%Lu", §ors);
1121 if (XENBUS_EXIST_ERR(err))
1122 return;
1123 printk(KERN_INFO "Setting capacity to %Lu\n",
1124 sectors);
1125 set_capacity(info->gd, sectors);
1126 revalidate_disk(info->gd);
1127
1128 /* fall through */
1129 case BLKIF_STATE_SUSPENDED:
1130 return;
1131
1132 default:
1133 break;
1134 }
1135
1136 dev_dbg(&info->xbdev->dev, "%s:%s.\n",
1137 __func__, info->xbdev->otherend);
1138
1139 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1140 "sectors", "%llu", §ors,
1141 "info", "%u", &binfo,
1142 "sector-size", "%lu", §or_size,
1143 NULL);
1144 if (err) {
1145 xenbus_dev_fatal(info->xbdev, err,
1146 "reading backend fields at %s",
1147 info->xbdev->otherend);
1148 return;
1149 }
1150
1151 info->feature_flush = 0;
1152 info->flush_op = 0;
1153
1154 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1155 "feature-barrier", "%d", &barrier,
1156 NULL);
1157
1158 /*
1159 * If there's no "feature-barrier" defined, then it means
1160 * we're dealing with a very old backend which writes
1161 * synchronously; nothing to do.
1162 *
1163 * If there are barriers, then we use flush.
1164 */
1165 if (!err && barrier) {
1166 info->feature_flush = REQ_FLUSH | REQ_FUA;
1167 info->flush_op = BLKIF_OP_WRITE_BARRIER;
1168 }
1169 /*
1170 * And if there is "feature-flush-cache" use that above
1171 * barriers.
1172 */
1173 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1174 "feature-flush-cache", "%d", &flush,
1175 NULL);
1176
1177 if (!err && flush) {
1178 info->feature_flush = REQ_FLUSH;
1179 info->flush_op = BLKIF_OP_FLUSH_DISKCACHE;
1180 }
1181
1182 err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size);
1183 if (err) {
1184 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
1185 info->xbdev->otherend);
1186 return;
1187 }
1188
1189 xenbus_switch_state(info->xbdev, XenbusStateConnected);
1190
1191 /* Kick pending requests. */
1192 spin_lock_irq(&blkif_io_lock);
1193 info->connected = BLKIF_STATE_CONNECTED;
1194 kick_pending_request_queues(info);
1195 spin_unlock_irq(&blkif_io_lock);
1196
1197 add_disk(info->gd);
1198
1199 info->is_ready = 1;
1200}
1201
1202/**
1203 * Callback received when the backend's state changes.
1204 */
1205static void blkback_changed(struct xenbus_device *dev,
1206 enum xenbus_state backend_state)
1207{
1208 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1209
1210 dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1211
1212 switch (backend_state) {
1213 case XenbusStateInitialising:
1214 case XenbusStateInitWait:
1215 case XenbusStateInitialised:
1216 case XenbusStateReconfiguring:
1217 case XenbusStateReconfigured:
1218 case XenbusStateUnknown:
1219 case XenbusStateClosed:
1220 break;
1221
1222 case XenbusStateConnected:
1223 blkfront_connect(info);
1224 break;
1225
1226 case XenbusStateClosing:
1227 blkfront_closing(info);
1228 break;
1229 }
1230}
1231
1232static int blkfront_remove(struct xenbus_device *xbdev)
1233{
1234 struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
1235 struct block_device *bdev = NULL;
1236 struct gendisk *disk;
1237
1238 dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
1239
1240 blkif_free(info, 0);
1241
1242 mutex_lock(&info->mutex);
1243
1244 disk = info->gd;
1245 if (disk)
1246 bdev = bdget_disk(disk, 0);
1247
1248 info->xbdev = NULL;
1249 mutex_unlock(&info->mutex);
1250
1251 if (!bdev) {
1252 kfree(info);
1253 return 0;
1254 }
1255
1256 /*
1257 * The xbdev was removed before we reached the Closed
1258 * state. See if it's safe to remove the disk. If the bdev
1259 * isn't closed yet, we let release take care of it.
1260 */
1261
1262 mutex_lock(&bdev->bd_mutex);
1263 info = disk->private_data;
1264
1265 dev_warn(disk_to_dev(disk),
1266 "%s was hot-unplugged, %d stale handles\n",
1267 xbdev->nodename, bdev->bd_openers);
1268
1269 if (info && !bdev->bd_openers) {
1270 xlvbd_release_gendisk(info);
1271 disk->private_data = NULL;
1272 kfree(info);
1273 }
1274
1275 mutex_unlock(&bdev->bd_mutex);
1276 bdput(bdev);
1277
1278 return 0;
1279}
1280
1281static int blkfront_is_ready(struct xenbus_device *dev)
1282{
1283 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1284
1285 return info->is_ready && info->xbdev;
1286}
1287
1288static int blkif_open(struct block_device *bdev, fmode_t mode)
1289{
1290 struct gendisk *disk = bdev->bd_disk;
1291 struct blkfront_info *info;
1292 int err = 0;
1293
1294 mutex_lock(&blkfront_mutex);
1295
1296 info = disk->private_data;
1297 if (!info) {
1298 /* xbdev gone */
1299 err = -ERESTARTSYS;
1300 goto out;
1301 }
1302
1303 mutex_lock(&info->mutex);
1304
1305 if (!info->gd)
1306 /* xbdev is closed */
1307 err = -ERESTARTSYS;
1308
1309 mutex_unlock(&info->mutex);
1310
1311out:
1312 mutex_unlock(&blkfront_mutex);
1313 return err;
1314}
1315
1316static int blkif_release(struct gendisk *disk, fmode_t mode)
1317{
1318 struct blkfront_info *info = disk->private_data;
1319 struct block_device *bdev;
1320 struct xenbus_device *xbdev;
1321
1322 mutex_lock(&blkfront_mutex);
1323
1324 bdev = bdget_disk(disk, 0);
1325 bdput(bdev);
1326
1327 if (bdev->bd_openers)
1328 goto out;
1329
1330 /*
1331 * Check if we have been instructed to close. We will have
1332 * deferred this request, because the bdev was still open.
1333 */
1334
1335 mutex_lock(&info->mutex);
1336 xbdev = info->xbdev;
1337
1338 if (xbdev && xbdev->state == XenbusStateClosing) {
1339 /* pending switch to state closed */
1340 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1341 xlvbd_release_gendisk(info);
1342 xenbus_frontend_closed(info->xbdev);
1343 }
1344
1345 mutex_unlock(&info->mutex);
1346
1347 if (!xbdev) {
1348 /* sudden device removal */
1349 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1350 xlvbd_release_gendisk(info);
1351 disk->private_data = NULL;
1352 kfree(info);
1353 }
1354
1355out:
1356 mutex_unlock(&blkfront_mutex);
1357 return 0;
1358}
1359
1360static const struct block_device_operations xlvbd_block_fops =
1361{
1362 .owner = THIS_MODULE,
1363 .open = blkif_open,
1364 .release = blkif_release,
1365 .getgeo = blkif_getgeo,
1366 .ioctl = blkif_ioctl,
1367};
1368
1369
1370static const struct xenbus_device_id blkfront_ids[] = {
1371 { "vbd" },
1372 { "" }
1373};
1374
1375static struct xenbus_driver blkfront = {
1376 .name = "vbd",
1377 .owner = THIS_MODULE,
1378 .ids = blkfront_ids,
1379 .probe = blkfront_probe,
1380 .remove = blkfront_remove,
1381 .resume = blkfront_resume,
1382 .otherend_changed = blkback_changed,
1383 .is_ready = blkfront_is_ready,
1384};
1385
1386static int __init xlblk_init(void)
1387{
1388 if (!xen_domain())
1389 return -ENODEV;
1390
1391 if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
1392 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
1393 XENVBD_MAJOR, DEV_NAME);
1394 return -ENODEV;
1395 }
1396
1397 return xenbus_register_frontend(&blkfront);
1398}
1399module_init(xlblk_init);
1400
1401
1402static void __exit xlblk_exit(void)
1403{
1404 return xenbus_unregister_driver(&blkfront);
1405}
1406module_exit(xlblk_exit);
1407
1408MODULE_DESCRIPTION("Xen virtual block device frontend");
1409MODULE_LICENSE("GPL");
1410MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
1411MODULE_ALIAS("xen:vbd");
1412MODULE_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/mutex.h>
46#include <linux/scatterlist.h>
47#include <linux/bitmap.h>
48#include <linux/list.h>
49#include <linux/workqueue.h>
50#include <linux/sched/mm.h>
51
52#include <xen/xen.h>
53#include <xen/xenbus.h>
54#include <xen/grant_table.h>
55#include <xen/events.h>
56#include <xen/page.h>
57#include <xen/platform_pci.h>
58
59#include <xen/interface/grant_table.h>
60#include <xen/interface/io/blkif.h>
61#include <xen/interface/io/protocols.h>
62
63#include <asm/xen/hypervisor.h>
64
65/*
66 * The minimal size of segment supported by the block framework is PAGE_SIZE.
67 * When Linux is using a different page size than Xen, it may not be possible
68 * to put all the data in a single segment.
69 * This can happen when the backend doesn't support indirect descriptor and
70 * therefore the maximum amount of data that a request can carry is
71 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE = 44KB
72 *
73 * Note that we only support one extra request. So the Linux page size
74 * should be <= ( 2 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) =
75 * 88KB.
76 */
77#define HAS_EXTRA_REQ (BLKIF_MAX_SEGMENTS_PER_REQUEST < XEN_PFN_PER_PAGE)
78
79enum blkif_state {
80 BLKIF_STATE_DISCONNECTED,
81 BLKIF_STATE_CONNECTED,
82 BLKIF_STATE_SUSPENDED,
83};
84
85struct grant {
86 grant_ref_t gref;
87 struct page *page;
88 struct list_head node;
89};
90
91enum blk_req_status {
92 REQ_WAITING,
93 REQ_DONE,
94 REQ_ERROR,
95 REQ_EOPNOTSUPP,
96};
97
98struct blk_shadow {
99 struct blkif_request req;
100 struct request *request;
101 struct grant **grants_used;
102 struct grant **indirect_grants;
103 struct scatterlist *sg;
104 unsigned int num_sg;
105 enum blk_req_status status;
106
107 #define NO_ASSOCIATED_ID ~0UL
108 /*
109 * Id of the sibling if we ever need 2 requests when handling a
110 * block I/O request
111 */
112 unsigned long associated_id;
113};
114
115struct blkif_req {
116 blk_status_t error;
117};
118
119static inline struct blkif_req *blkif_req(struct request *rq)
120{
121 return blk_mq_rq_to_pdu(rq);
122}
123
124static DEFINE_MUTEX(blkfront_mutex);
125static const struct block_device_operations xlvbd_block_fops;
126static struct delayed_work blkfront_work;
127static LIST_HEAD(info_list);
128
129/*
130 * Maximum number of segments in indirect requests, the actual value used by
131 * the frontend driver is the minimum of this value and the value provided
132 * by the backend driver.
133 */
134
135static unsigned int xen_blkif_max_segments = 32;
136module_param_named(max_indirect_segments, xen_blkif_max_segments, uint, 0444);
137MODULE_PARM_DESC(max_indirect_segments,
138 "Maximum amount of segments in indirect requests (default is 32)");
139
140static unsigned int xen_blkif_max_queues = 4;
141module_param_named(max_queues, xen_blkif_max_queues, uint, 0444);
142MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
143
144/*
145 * Maximum order of pages to be used for the shared ring between front and
146 * backend, 4KB page granularity is used.
147 */
148static unsigned int xen_blkif_max_ring_order;
149module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, 0444);
150MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
151
152#define BLK_RING_SIZE(info) \
153 __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
154
155/*
156 * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
157 * characters are enough. Define to 20 to keep consistent with backend.
158 */
159#define RINGREF_NAME_LEN (20)
160/*
161 * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
162 */
163#define QUEUE_NAME_LEN (17)
164
165/*
166 * Per-ring info.
167 * Every blkfront device can associate with one or more blkfront_ring_info,
168 * depending on how many hardware queues/rings to be used.
169 */
170struct blkfront_ring_info {
171 /* Lock to protect data in every ring buffer. */
172 spinlock_t ring_lock;
173 struct blkif_front_ring ring;
174 unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
175 unsigned int evtchn, irq;
176 struct work_struct work;
177 struct gnttab_free_callback callback;
178 struct list_head indirect_pages;
179 struct list_head grants;
180 unsigned int persistent_gnts_c;
181 unsigned long shadow_free;
182 struct blkfront_info *dev_info;
183 struct blk_shadow shadow[];
184};
185
186/*
187 * We have one of these per vbd, whether ide, scsi or 'other'. They
188 * hang in private_data off the gendisk structure. We may end up
189 * putting all kinds of interesting stuff here :-)
190 */
191struct blkfront_info
192{
193 struct mutex mutex;
194 struct xenbus_device *xbdev;
195 struct gendisk *gd;
196 u16 sector_size;
197 unsigned int physical_sector_size;
198 int vdevice;
199 blkif_vdev_t handle;
200 enum blkif_state connected;
201 /* Number of pages per ring buffer. */
202 unsigned int nr_ring_pages;
203 struct request_queue *rq;
204 unsigned int feature_flush:1;
205 unsigned int feature_fua:1;
206 unsigned int feature_discard:1;
207 unsigned int feature_secdiscard:1;
208 unsigned int feature_persistent:1;
209 unsigned int discard_granularity;
210 unsigned int discard_alignment;
211 /* Number of 4KB segments handled */
212 unsigned int max_indirect_segments;
213 int is_ready;
214 struct blk_mq_tag_set tag_set;
215 struct blkfront_ring_info *rinfo;
216 unsigned int nr_rings;
217 unsigned int rinfo_size;
218 /* Save uncomplete reqs and bios for migration. */
219 struct list_head requests;
220 struct bio_list bio_list;
221 struct list_head info_list;
222};
223
224static unsigned int nr_minors;
225static unsigned long *minors;
226static DEFINE_SPINLOCK(minor_lock);
227
228#define GRANT_INVALID_REF 0
229
230#define PARTS_PER_DISK 16
231#define PARTS_PER_EXT_DISK 256
232
233#define BLKIF_MAJOR(dev) ((dev)>>8)
234#define BLKIF_MINOR(dev) ((dev) & 0xff)
235
236#define EXT_SHIFT 28
237#define EXTENDED (1<<EXT_SHIFT)
238#define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
239#define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
240#define EMULATED_HD_DISK_MINOR_OFFSET (0)
241#define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
242#define EMULATED_SD_DISK_MINOR_OFFSET (0)
243#define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
244
245#define DEV_NAME "xvd" /* name in /dev */
246
247/*
248 * Grants are always the same size as a Xen page (i.e 4KB).
249 * A physical segment is always the same size as a Linux page.
250 * Number of grants per physical segment
251 */
252#define GRANTS_PER_PSEG (PAGE_SIZE / XEN_PAGE_SIZE)
253
254#define GRANTS_PER_INDIRECT_FRAME \
255 (XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
256
257#define INDIRECT_GREFS(_grants) \
258 DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
259
260static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
261static void blkfront_gather_backend_features(struct blkfront_info *info);
262static int negotiate_mq(struct blkfront_info *info);
263
264#define for_each_rinfo(info, ptr, idx) \
265 for ((ptr) = (info)->rinfo, (idx) = 0; \
266 (idx) < (info)->nr_rings; \
267 (idx)++, (ptr) = (void *)(ptr) + (info)->rinfo_size)
268
269static inline struct blkfront_ring_info *
270get_rinfo(const struct blkfront_info *info, unsigned int i)
271{
272 BUG_ON(i >= info->nr_rings);
273 return (void *)info->rinfo + i * info->rinfo_size;
274}
275
276static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
277{
278 unsigned long free = rinfo->shadow_free;
279
280 BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
281 rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
282 rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
283 return free;
284}
285
286static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
287 unsigned long id)
288{
289 if (rinfo->shadow[id].req.u.rw.id != id)
290 return -EINVAL;
291 if (rinfo->shadow[id].request == NULL)
292 return -EINVAL;
293 rinfo->shadow[id].req.u.rw.id = rinfo->shadow_free;
294 rinfo->shadow[id].request = NULL;
295 rinfo->shadow_free = id;
296 return 0;
297}
298
299static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
300{
301 struct blkfront_info *info = rinfo->dev_info;
302 struct page *granted_page;
303 struct grant *gnt_list_entry, *n;
304 int i = 0;
305
306 while (i < num) {
307 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
308 if (!gnt_list_entry)
309 goto out_of_memory;
310
311 if (info->feature_persistent) {
312 granted_page = alloc_page(GFP_NOIO);
313 if (!granted_page) {
314 kfree(gnt_list_entry);
315 goto out_of_memory;
316 }
317 gnt_list_entry->page = granted_page;
318 }
319
320 gnt_list_entry->gref = GRANT_INVALID_REF;
321 list_add(&gnt_list_entry->node, &rinfo->grants);
322 i++;
323 }
324
325 return 0;
326
327out_of_memory:
328 list_for_each_entry_safe(gnt_list_entry, n,
329 &rinfo->grants, node) {
330 list_del(&gnt_list_entry->node);
331 if (info->feature_persistent)
332 __free_page(gnt_list_entry->page);
333 kfree(gnt_list_entry);
334 i--;
335 }
336 BUG_ON(i != 0);
337 return -ENOMEM;
338}
339
340static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
341{
342 struct grant *gnt_list_entry;
343
344 BUG_ON(list_empty(&rinfo->grants));
345 gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
346 node);
347 list_del(&gnt_list_entry->node);
348
349 if (gnt_list_entry->gref != GRANT_INVALID_REF)
350 rinfo->persistent_gnts_c--;
351
352 return gnt_list_entry;
353}
354
355static inline void grant_foreign_access(const struct grant *gnt_list_entry,
356 const struct blkfront_info *info)
357{
358 gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
359 info->xbdev->otherend_id,
360 gnt_list_entry->page,
361 0);
362}
363
364static struct grant *get_grant(grant_ref_t *gref_head,
365 unsigned long gfn,
366 struct blkfront_ring_info *rinfo)
367{
368 struct grant *gnt_list_entry = get_free_grant(rinfo);
369 struct blkfront_info *info = rinfo->dev_info;
370
371 if (gnt_list_entry->gref != GRANT_INVALID_REF)
372 return gnt_list_entry;
373
374 /* Assign a gref to this page */
375 gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
376 BUG_ON(gnt_list_entry->gref == -ENOSPC);
377 if (info->feature_persistent)
378 grant_foreign_access(gnt_list_entry, info);
379 else {
380 /* Grant access to the GFN passed by the caller */
381 gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
382 info->xbdev->otherend_id,
383 gfn, 0);
384 }
385
386 return gnt_list_entry;
387}
388
389static struct grant *get_indirect_grant(grant_ref_t *gref_head,
390 struct blkfront_ring_info *rinfo)
391{
392 struct grant *gnt_list_entry = get_free_grant(rinfo);
393 struct blkfront_info *info = rinfo->dev_info;
394
395 if (gnt_list_entry->gref != GRANT_INVALID_REF)
396 return gnt_list_entry;
397
398 /* Assign a gref to this page */
399 gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
400 BUG_ON(gnt_list_entry->gref == -ENOSPC);
401 if (!info->feature_persistent) {
402 struct page *indirect_page;
403
404 /* Fetch a pre-allocated page to use for indirect grefs */
405 BUG_ON(list_empty(&rinfo->indirect_pages));
406 indirect_page = list_first_entry(&rinfo->indirect_pages,
407 struct page, lru);
408 list_del(&indirect_page->lru);
409 gnt_list_entry->page = indirect_page;
410 }
411 grant_foreign_access(gnt_list_entry, info);
412
413 return gnt_list_entry;
414}
415
416static const char *op_name(int op)
417{
418 static const char *const names[] = {
419 [BLKIF_OP_READ] = "read",
420 [BLKIF_OP_WRITE] = "write",
421 [BLKIF_OP_WRITE_BARRIER] = "barrier",
422 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
423 [BLKIF_OP_DISCARD] = "discard" };
424
425 if (op < 0 || op >= ARRAY_SIZE(names))
426 return "unknown";
427
428 if (!names[op])
429 return "reserved";
430
431 return names[op];
432}
433static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
434{
435 unsigned int end = minor + nr;
436 int rc;
437
438 if (end > nr_minors) {
439 unsigned long *bitmap, *old;
440
441 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
442 GFP_KERNEL);
443 if (bitmap == NULL)
444 return -ENOMEM;
445
446 spin_lock(&minor_lock);
447 if (end > nr_minors) {
448 old = minors;
449 memcpy(bitmap, minors,
450 BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
451 minors = bitmap;
452 nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
453 } else
454 old = bitmap;
455 spin_unlock(&minor_lock);
456 kfree(old);
457 }
458
459 spin_lock(&minor_lock);
460 if (find_next_bit(minors, end, minor) >= end) {
461 bitmap_set(minors, minor, nr);
462 rc = 0;
463 } else
464 rc = -EBUSY;
465 spin_unlock(&minor_lock);
466
467 return rc;
468}
469
470static void xlbd_release_minors(unsigned int minor, unsigned int nr)
471{
472 unsigned int end = minor + nr;
473
474 BUG_ON(end > nr_minors);
475 spin_lock(&minor_lock);
476 bitmap_clear(minors, minor, nr);
477 spin_unlock(&minor_lock);
478}
479
480static void blkif_restart_queue_callback(void *arg)
481{
482 struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
483 schedule_work(&rinfo->work);
484}
485
486static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
487{
488 /* We don't have real geometry info, but let's at least return
489 values consistent with the size of the device */
490 sector_t nsect = get_capacity(bd->bd_disk);
491 sector_t cylinders = nsect;
492
493 hg->heads = 0xff;
494 hg->sectors = 0x3f;
495 sector_div(cylinders, hg->heads * hg->sectors);
496 hg->cylinders = cylinders;
497 if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
498 hg->cylinders = 0xffff;
499 return 0;
500}
501
502static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
503 unsigned command, unsigned long argument)
504{
505 struct blkfront_info *info = bdev->bd_disk->private_data;
506 int i;
507
508 dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
509 command, (long)argument);
510
511 switch (command) {
512 case CDROMMULTISESSION:
513 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
514 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
515 if (put_user(0, (char __user *)(argument + i)))
516 return -EFAULT;
517 return 0;
518
519 case CDROM_GET_CAPABILITY: {
520 struct gendisk *gd = info->gd;
521 if (gd->flags & GENHD_FL_CD)
522 return 0;
523 return -EINVAL;
524 }
525
526 default:
527 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
528 command);*/
529 return -EINVAL; /* same return as native Linux */
530 }
531
532 return 0;
533}
534
535static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
536 struct request *req,
537 struct blkif_request **ring_req)
538{
539 unsigned long id;
540
541 *ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
542 rinfo->ring.req_prod_pvt++;
543
544 id = get_id_from_freelist(rinfo);
545 rinfo->shadow[id].request = req;
546 rinfo->shadow[id].status = REQ_WAITING;
547 rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
548
549 (*ring_req)->u.rw.id = id;
550
551 return id;
552}
553
554static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
555{
556 struct blkfront_info *info = rinfo->dev_info;
557 struct blkif_request *ring_req;
558 unsigned long id;
559
560 /* Fill out a communications ring structure. */
561 id = blkif_ring_get_request(rinfo, req, &ring_req);
562
563 ring_req->operation = BLKIF_OP_DISCARD;
564 ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
565 ring_req->u.discard.id = id;
566 ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
567 if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard)
568 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
569 else
570 ring_req->u.discard.flag = 0;
571
572 /* Keep a private copy so we can reissue requests when recovering. */
573 rinfo->shadow[id].req = *ring_req;
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 read 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 unsigned long id, extra_id = NO_ASSOCIATED_ID;
707 bool require_extra_req = false;
708 int i;
709 struct setup_rw_req setup = {
710 .grant_idx = 0,
711 .segments = NULL,
712 .rinfo = rinfo,
713 .need_copy = rq_data_dir(req) && info->feature_persistent,
714 };
715
716 /*
717 * Used to store if we are able to queue the request by just using
718 * existing persistent grants, or if we have to get new grants,
719 * as there are not sufficiently many free.
720 */
721 bool new_persistent_gnts = false;
722 struct scatterlist *sg;
723 int num_sg, max_grefs, num_grant;
724
725 max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
726 if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
727 /*
728 * If we are using indirect segments we need to account
729 * for the indirect grefs used in the request.
730 */
731 max_grefs += INDIRECT_GREFS(max_grefs);
732
733 /* Check if we have enough persistent grants to allocate a requests */
734 if (rinfo->persistent_gnts_c < max_grefs) {
735 new_persistent_gnts = true;
736
737 if (gnttab_alloc_grant_references(
738 max_grefs - rinfo->persistent_gnts_c,
739 &setup.gref_head) < 0) {
740 gnttab_request_free_callback(
741 &rinfo->callback,
742 blkif_restart_queue_callback,
743 rinfo,
744 max_grefs - rinfo->persistent_gnts_c);
745 return 1;
746 }
747 }
748
749 /* Fill out a communications ring structure. */
750 id = blkif_ring_get_request(rinfo, req, &ring_req);
751
752 num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
753 num_grant = 0;
754 /* Calculate the number of grant used */
755 for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
756 num_grant += gnttab_count_grant(sg->offset, sg->length);
757
758 require_extra_req = info->max_indirect_segments == 0 &&
759 num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
760 BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
761
762 rinfo->shadow[id].num_sg = num_sg;
763 if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
764 likely(!require_extra_req)) {
765 /*
766 * The indirect operation can only be a BLKIF_OP_READ or
767 * BLKIF_OP_WRITE
768 */
769 BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA);
770 ring_req->operation = BLKIF_OP_INDIRECT;
771 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
772 BLKIF_OP_WRITE : BLKIF_OP_READ;
773 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
774 ring_req->u.indirect.handle = info->handle;
775 ring_req->u.indirect.nr_segments = num_grant;
776 } else {
777 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
778 ring_req->u.rw.handle = info->handle;
779 ring_req->operation = rq_data_dir(req) ?
780 BLKIF_OP_WRITE : BLKIF_OP_READ;
781 if (req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA) {
782 /*
783 * Ideally we can do an unordered flush-to-disk.
784 * In case the backend onlysupports barriers, use that.
785 * A barrier request a superset of FUA, so we can
786 * implement it the same way. (It's also a FLUSH+FUA,
787 * since it is guaranteed ordered WRT previous writes.)
788 */
789 if (info->feature_flush && info->feature_fua)
790 ring_req->operation =
791 BLKIF_OP_WRITE_BARRIER;
792 else if (info->feature_flush)
793 ring_req->operation =
794 BLKIF_OP_FLUSH_DISKCACHE;
795 else
796 ring_req->operation = 0;
797 }
798 ring_req->u.rw.nr_segments = num_grant;
799 if (unlikely(require_extra_req)) {
800 extra_id = blkif_ring_get_request(rinfo, req,
801 &extra_ring_req);
802 /*
803 * Only the first request contains the scatter-gather
804 * list.
805 */
806 rinfo->shadow[extra_id].num_sg = 0;
807
808 blkif_setup_extra_req(ring_req, extra_ring_req);
809
810 /* Link the 2 requests together */
811 rinfo->shadow[extra_id].associated_id = id;
812 rinfo->shadow[id].associated_id = extra_id;
813 }
814 }
815
816 setup.ring_req = ring_req;
817 setup.id = id;
818
819 setup.require_extra_req = require_extra_req;
820 if (unlikely(require_extra_req))
821 setup.extra_ring_req = extra_ring_req;
822
823 for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
824 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
825
826 if (setup.need_copy) {
827 setup.bvec_off = sg->offset;
828 setup.bvec_data = kmap_atomic(sg_page(sg));
829 }
830
831 gnttab_foreach_grant_in_range(sg_page(sg),
832 sg->offset,
833 sg->length,
834 blkif_setup_rw_req_grant,
835 &setup);
836
837 if (setup.need_copy)
838 kunmap_atomic(setup.bvec_data);
839 }
840 if (setup.segments)
841 kunmap_atomic(setup.segments);
842
843 /* Keep a private copy so we can reissue requests when recovering. */
844 rinfo->shadow[id].req = *ring_req;
845 if (unlikely(require_extra_req))
846 rinfo->shadow[extra_id].req = *extra_ring_req;
847
848 if (new_persistent_gnts)
849 gnttab_free_grant_references(setup.gref_head);
850
851 return 0;
852}
853
854/*
855 * Generate a Xen blkfront IO request from a blk layer request. Reads
856 * and writes are handled as expected.
857 *
858 * @req: a request struct
859 */
860static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
861{
862 if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
863 return 1;
864
865 if (unlikely(req_op(req) == REQ_OP_DISCARD ||
866 req_op(req) == REQ_OP_SECURE_ERASE))
867 return blkif_queue_discard_req(req, rinfo);
868 else
869 return blkif_queue_rw_req(req, rinfo);
870}
871
872static inline void flush_requests(struct blkfront_ring_info *rinfo)
873{
874 int notify;
875
876 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
877
878 if (notify)
879 notify_remote_via_irq(rinfo->irq);
880}
881
882static inline bool blkif_request_flush_invalid(struct request *req,
883 struct blkfront_info *info)
884{
885 return (blk_rq_is_passthrough(req) ||
886 ((req_op(req) == REQ_OP_FLUSH) &&
887 !info->feature_flush) ||
888 ((req->cmd_flags & REQ_FUA) &&
889 !info->feature_fua));
890}
891
892static blk_status_t blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
893 const struct blk_mq_queue_data *qd)
894{
895 unsigned long flags;
896 int qid = hctx->queue_num;
897 struct blkfront_info *info = hctx->queue->queuedata;
898 struct blkfront_ring_info *rinfo = NULL;
899
900 rinfo = get_rinfo(info, qid);
901 blk_mq_start_request(qd->rq);
902 spin_lock_irqsave(&rinfo->ring_lock, flags);
903 if (RING_FULL(&rinfo->ring))
904 goto out_busy;
905
906 if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
907 goto out_err;
908
909 if (blkif_queue_request(qd->rq, rinfo))
910 goto out_busy;
911
912 flush_requests(rinfo);
913 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
914 return BLK_STS_OK;
915
916out_err:
917 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
918 return BLK_STS_IOERR;
919
920out_busy:
921 blk_mq_stop_hw_queue(hctx);
922 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
923 return BLK_STS_DEV_RESOURCE;
924}
925
926static void blkif_complete_rq(struct request *rq)
927{
928 blk_mq_end_request(rq, blkif_req(rq)->error);
929}
930
931static const struct blk_mq_ops blkfront_mq_ops = {
932 .queue_rq = blkif_queue_rq,
933 .complete = blkif_complete_rq,
934};
935
936static void blkif_set_queue_limits(struct blkfront_info *info)
937{
938 struct request_queue *rq = info->rq;
939 struct gendisk *gd = info->gd;
940 unsigned int segments = info->max_indirect_segments ? :
941 BLKIF_MAX_SEGMENTS_PER_REQUEST;
942
943 blk_queue_flag_set(QUEUE_FLAG_VIRT, rq);
944
945 if (info->feature_discard) {
946 blk_queue_flag_set(QUEUE_FLAG_DISCARD, rq);
947 blk_queue_max_discard_sectors(rq, get_capacity(gd));
948 rq->limits.discard_granularity = info->discard_granularity;
949 rq->limits.discard_alignment = info->discard_alignment;
950 if (info->feature_secdiscard)
951 blk_queue_flag_set(QUEUE_FLAG_SECERASE, rq);
952 }
953
954 /* Hard sector size and max sectors impersonate the equiv. hardware. */
955 blk_queue_logical_block_size(rq, info->sector_size);
956 blk_queue_physical_block_size(rq, info->physical_sector_size);
957 blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
958
959 /* Each segment in a request is up to an aligned page in size. */
960 blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
961 blk_queue_max_segment_size(rq, PAGE_SIZE);
962
963 /* Ensure a merged request will fit in a single I/O ring slot. */
964 blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
965
966 /* Make sure buffer addresses are sector-aligned. */
967 blk_queue_dma_alignment(rq, 511);
968}
969
970static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
971 unsigned int physical_sector_size)
972{
973 struct request_queue *rq;
974 struct blkfront_info *info = gd->private_data;
975
976 memset(&info->tag_set, 0, sizeof(info->tag_set));
977 info->tag_set.ops = &blkfront_mq_ops;
978 info->tag_set.nr_hw_queues = info->nr_rings;
979 if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
980 /*
981 * When indirect descriptior is not supported, the I/O request
982 * will be split between multiple request in the ring.
983 * To avoid problems when sending the request, divide by
984 * 2 the depth of the queue.
985 */
986 info->tag_set.queue_depth = BLK_RING_SIZE(info) / 2;
987 } else
988 info->tag_set.queue_depth = BLK_RING_SIZE(info);
989 info->tag_set.numa_node = NUMA_NO_NODE;
990 info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
991 info->tag_set.cmd_size = sizeof(struct blkif_req);
992 info->tag_set.driver_data = info;
993
994 if (blk_mq_alloc_tag_set(&info->tag_set))
995 return -EINVAL;
996 rq = blk_mq_init_queue(&info->tag_set);
997 if (IS_ERR(rq)) {
998 blk_mq_free_tag_set(&info->tag_set);
999 return PTR_ERR(rq);
1000 }
1001
1002 rq->queuedata = info;
1003 info->rq = gd->queue = rq;
1004 info->gd = gd;
1005 info->sector_size = sector_size;
1006 info->physical_sector_size = physical_sector_size;
1007 blkif_set_queue_limits(info);
1008
1009 return 0;
1010}
1011
1012static const char *flush_info(struct blkfront_info *info)
1013{
1014 if (info->feature_flush && info->feature_fua)
1015 return "barrier: enabled;";
1016 else if (info->feature_flush)
1017 return "flush diskcache: enabled;";
1018 else
1019 return "barrier or flush: disabled;";
1020}
1021
1022static void xlvbd_flush(struct blkfront_info *info)
1023{
1024 blk_queue_write_cache(info->rq, info->feature_flush ? true : false,
1025 info->feature_fua ? true : false);
1026 pr_info("blkfront: %s: %s %s %s %s %s\n",
1027 info->gd->disk_name, flush_info(info),
1028 "persistent grants:", info->feature_persistent ?
1029 "enabled;" : "disabled;", "indirect descriptors:",
1030 info->max_indirect_segments ? "enabled;" : "disabled;");
1031}
1032
1033static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
1034{
1035 int major;
1036 major = BLKIF_MAJOR(vdevice);
1037 *minor = BLKIF_MINOR(vdevice);
1038 switch (major) {
1039 case XEN_IDE0_MAJOR:
1040 *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
1041 *minor = ((*minor / 64) * PARTS_PER_DISK) +
1042 EMULATED_HD_DISK_MINOR_OFFSET;
1043 break;
1044 case XEN_IDE1_MAJOR:
1045 *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1046 *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1047 EMULATED_HD_DISK_MINOR_OFFSET;
1048 break;
1049 case XEN_SCSI_DISK0_MAJOR:
1050 *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1051 *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1052 break;
1053 case XEN_SCSI_DISK1_MAJOR:
1054 case XEN_SCSI_DISK2_MAJOR:
1055 case XEN_SCSI_DISK3_MAJOR:
1056 case XEN_SCSI_DISK4_MAJOR:
1057 case XEN_SCSI_DISK5_MAJOR:
1058 case XEN_SCSI_DISK6_MAJOR:
1059 case XEN_SCSI_DISK7_MAJOR:
1060 *offset = (*minor / PARTS_PER_DISK) +
1061 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1062 EMULATED_SD_DISK_NAME_OFFSET;
1063 *minor = *minor +
1064 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1065 EMULATED_SD_DISK_MINOR_OFFSET;
1066 break;
1067 case XEN_SCSI_DISK8_MAJOR:
1068 case XEN_SCSI_DISK9_MAJOR:
1069 case XEN_SCSI_DISK10_MAJOR:
1070 case XEN_SCSI_DISK11_MAJOR:
1071 case XEN_SCSI_DISK12_MAJOR:
1072 case XEN_SCSI_DISK13_MAJOR:
1073 case XEN_SCSI_DISK14_MAJOR:
1074 case XEN_SCSI_DISK15_MAJOR:
1075 *offset = (*minor / PARTS_PER_DISK) +
1076 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1077 EMULATED_SD_DISK_NAME_OFFSET;
1078 *minor = *minor +
1079 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1080 EMULATED_SD_DISK_MINOR_OFFSET;
1081 break;
1082 case XENVBD_MAJOR:
1083 *offset = *minor / PARTS_PER_DISK;
1084 break;
1085 default:
1086 printk(KERN_WARNING "blkfront: your disk configuration is "
1087 "incorrect, please use an xvd device instead\n");
1088 return -ENODEV;
1089 }
1090 return 0;
1091}
1092
1093static char *encode_disk_name(char *ptr, unsigned int n)
1094{
1095 if (n >= 26)
1096 ptr = encode_disk_name(ptr, n / 26 - 1);
1097 *ptr = 'a' + n % 26;
1098 return ptr + 1;
1099}
1100
1101static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1102 struct blkfront_info *info,
1103 u16 vdisk_info, u16 sector_size,
1104 unsigned int physical_sector_size)
1105{
1106 struct gendisk *gd;
1107 int nr_minors = 1;
1108 int err;
1109 unsigned int offset;
1110 int minor;
1111 int nr_parts;
1112 char *ptr;
1113
1114 BUG_ON(info->gd != NULL);
1115 BUG_ON(info->rq != NULL);
1116
1117 if ((info->vdevice>>EXT_SHIFT) > 1) {
1118 /* this is above the extended range; something is wrong */
1119 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1120 return -ENODEV;
1121 }
1122
1123 if (!VDEV_IS_EXTENDED(info->vdevice)) {
1124 err = xen_translate_vdev(info->vdevice, &minor, &offset);
1125 if (err)
1126 return err;
1127 nr_parts = PARTS_PER_DISK;
1128 } else {
1129 minor = BLKIF_MINOR_EXT(info->vdevice);
1130 nr_parts = PARTS_PER_EXT_DISK;
1131 offset = minor / nr_parts;
1132 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1133 printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1134 "emulated IDE disks,\n\t choose an xvd device name"
1135 "from xvde on\n", info->vdevice);
1136 }
1137 if (minor >> MINORBITS) {
1138 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1139 info->vdevice, minor);
1140 return -ENODEV;
1141 }
1142
1143 if ((minor % nr_parts) == 0)
1144 nr_minors = nr_parts;
1145
1146 err = xlbd_reserve_minors(minor, nr_minors);
1147 if (err)
1148 goto out;
1149 err = -ENODEV;
1150
1151 gd = alloc_disk(nr_minors);
1152 if (gd == NULL)
1153 goto release;
1154
1155 strcpy(gd->disk_name, DEV_NAME);
1156 ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1157 BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1158 if (nr_minors > 1)
1159 *ptr = 0;
1160 else
1161 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1162 "%d", minor & (nr_parts - 1));
1163
1164 gd->major = XENVBD_MAJOR;
1165 gd->first_minor = minor;
1166 gd->fops = &xlvbd_block_fops;
1167 gd->private_data = info;
1168 set_capacity(gd, capacity);
1169
1170 if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size)) {
1171 del_gendisk(gd);
1172 goto release;
1173 }
1174
1175 xlvbd_flush(info);
1176
1177 if (vdisk_info & VDISK_READONLY)
1178 set_disk_ro(gd, 1);
1179
1180 if (vdisk_info & VDISK_REMOVABLE)
1181 gd->flags |= GENHD_FL_REMOVABLE;
1182
1183 if (vdisk_info & VDISK_CDROM)
1184 gd->flags |= GENHD_FL_CD;
1185
1186 return 0;
1187
1188 release:
1189 xlbd_release_minors(minor, nr_minors);
1190 out:
1191 return err;
1192}
1193
1194static void xlvbd_release_gendisk(struct blkfront_info *info)
1195{
1196 unsigned int minor, nr_minors, i;
1197 struct blkfront_ring_info *rinfo;
1198
1199 if (info->rq == NULL)
1200 return;
1201
1202 /* No more blkif_request(). */
1203 blk_mq_stop_hw_queues(info->rq);
1204
1205 for_each_rinfo(info, rinfo, i) {
1206 /* No more gnttab callback work. */
1207 gnttab_cancel_free_callback(&rinfo->callback);
1208
1209 /* Flush gnttab callback work. Must be done with no locks held. */
1210 flush_work(&rinfo->work);
1211 }
1212
1213 del_gendisk(info->gd);
1214
1215 minor = info->gd->first_minor;
1216 nr_minors = info->gd->minors;
1217 xlbd_release_minors(minor, nr_minors);
1218
1219 blk_cleanup_queue(info->rq);
1220 blk_mq_free_tag_set(&info->tag_set);
1221 info->rq = NULL;
1222
1223 put_disk(info->gd);
1224 info->gd = NULL;
1225}
1226
1227/* Already hold rinfo->ring_lock. */
1228static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1229{
1230 if (!RING_FULL(&rinfo->ring))
1231 blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1232}
1233
1234static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1235{
1236 unsigned long flags;
1237
1238 spin_lock_irqsave(&rinfo->ring_lock, flags);
1239 kick_pending_request_queues_locked(rinfo);
1240 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1241}
1242
1243static void blkif_restart_queue(struct work_struct *work)
1244{
1245 struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1246
1247 if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1248 kick_pending_request_queues(rinfo);
1249}
1250
1251static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1252{
1253 struct grant *persistent_gnt, *n;
1254 struct blkfront_info *info = rinfo->dev_info;
1255 int i, j, segs;
1256
1257 /*
1258 * Remove indirect pages, this only happens when using indirect
1259 * descriptors but not persistent grants
1260 */
1261 if (!list_empty(&rinfo->indirect_pages)) {
1262 struct page *indirect_page, *n;
1263
1264 BUG_ON(info->feature_persistent);
1265 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1266 list_del(&indirect_page->lru);
1267 __free_page(indirect_page);
1268 }
1269 }
1270
1271 /* Remove all persistent grants. */
1272 if (!list_empty(&rinfo->grants)) {
1273 list_for_each_entry_safe(persistent_gnt, n,
1274 &rinfo->grants, node) {
1275 list_del(&persistent_gnt->node);
1276 if (persistent_gnt->gref != GRANT_INVALID_REF) {
1277 gnttab_end_foreign_access(persistent_gnt->gref,
1278 0, 0UL);
1279 rinfo->persistent_gnts_c--;
1280 }
1281 if (info->feature_persistent)
1282 __free_page(persistent_gnt->page);
1283 kfree(persistent_gnt);
1284 }
1285 }
1286 BUG_ON(rinfo->persistent_gnts_c != 0);
1287
1288 for (i = 0; i < BLK_RING_SIZE(info); i++) {
1289 /*
1290 * Clear persistent grants present in requests already
1291 * on the shared ring
1292 */
1293 if (!rinfo->shadow[i].request)
1294 goto free_shadow;
1295
1296 segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1297 rinfo->shadow[i].req.u.indirect.nr_segments :
1298 rinfo->shadow[i].req.u.rw.nr_segments;
1299 for (j = 0; j < segs; j++) {
1300 persistent_gnt = rinfo->shadow[i].grants_used[j];
1301 gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1302 if (info->feature_persistent)
1303 __free_page(persistent_gnt->page);
1304 kfree(persistent_gnt);
1305 }
1306
1307 if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1308 /*
1309 * If this is not an indirect operation don't try to
1310 * free indirect segments
1311 */
1312 goto free_shadow;
1313
1314 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1315 persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1316 gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1317 __free_page(persistent_gnt->page);
1318 kfree(persistent_gnt);
1319 }
1320
1321free_shadow:
1322 kvfree(rinfo->shadow[i].grants_used);
1323 rinfo->shadow[i].grants_used = NULL;
1324 kvfree(rinfo->shadow[i].indirect_grants);
1325 rinfo->shadow[i].indirect_grants = NULL;
1326 kvfree(rinfo->shadow[i].sg);
1327 rinfo->shadow[i].sg = NULL;
1328 }
1329
1330 /* No more gnttab callback work. */
1331 gnttab_cancel_free_callback(&rinfo->callback);
1332
1333 /* Flush gnttab callback work. Must be done with no locks held. */
1334 flush_work(&rinfo->work);
1335
1336 /* Free resources associated with old device channel. */
1337 for (i = 0; i < info->nr_ring_pages; i++) {
1338 if (rinfo->ring_ref[i] != GRANT_INVALID_REF) {
1339 gnttab_end_foreign_access(rinfo->ring_ref[i], 0, 0);
1340 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1341 }
1342 }
1343 free_pages((unsigned long)rinfo->ring.sring, get_order(info->nr_ring_pages * XEN_PAGE_SIZE));
1344 rinfo->ring.sring = NULL;
1345
1346 if (rinfo->irq)
1347 unbind_from_irqhandler(rinfo->irq, rinfo);
1348 rinfo->evtchn = rinfo->irq = 0;
1349}
1350
1351static void blkif_free(struct blkfront_info *info, int suspend)
1352{
1353 unsigned int i;
1354 struct blkfront_ring_info *rinfo;
1355
1356 /* Prevent new requests being issued until we fix things up. */
1357 info->connected = suspend ?
1358 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1359 /* No more blkif_request(). */
1360 if (info->rq)
1361 blk_mq_stop_hw_queues(info->rq);
1362
1363 for_each_rinfo(info, rinfo, i)
1364 blkif_free_ring(rinfo);
1365
1366 kvfree(info->rinfo);
1367 info->rinfo = NULL;
1368 info->nr_rings = 0;
1369}
1370
1371struct copy_from_grant {
1372 const struct blk_shadow *s;
1373 unsigned int grant_idx;
1374 unsigned int bvec_offset;
1375 char *bvec_data;
1376};
1377
1378static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1379 unsigned int len, void *data)
1380{
1381 struct copy_from_grant *info = data;
1382 char *shared_data;
1383 /* Convenient aliases */
1384 const struct blk_shadow *s = info->s;
1385
1386 shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1387
1388 memcpy(info->bvec_data + info->bvec_offset,
1389 shared_data + offset, len);
1390
1391 info->bvec_offset += len;
1392 info->grant_idx++;
1393
1394 kunmap_atomic(shared_data);
1395}
1396
1397static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1398{
1399 switch (rsp)
1400 {
1401 case BLKIF_RSP_OKAY:
1402 return REQ_DONE;
1403 case BLKIF_RSP_EOPNOTSUPP:
1404 return REQ_EOPNOTSUPP;
1405 case BLKIF_RSP_ERROR:
1406 default:
1407 return REQ_ERROR;
1408 }
1409}
1410
1411/*
1412 * Get the final status of the block request based on two ring response
1413 */
1414static int blkif_get_final_status(enum blk_req_status s1,
1415 enum blk_req_status s2)
1416{
1417 BUG_ON(s1 == REQ_WAITING);
1418 BUG_ON(s2 == REQ_WAITING);
1419
1420 if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1421 return BLKIF_RSP_ERROR;
1422 else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1423 return BLKIF_RSP_EOPNOTSUPP;
1424 return BLKIF_RSP_OKAY;
1425}
1426
1427static bool blkif_completion(unsigned long *id,
1428 struct blkfront_ring_info *rinfo,
1429 struct blkif_response *bret)
1430{
1431 int i = 0;
1432 struct scatterlist *sg;
1433 int num_sg, num_grant;
1434 struct blkfront_info *info = rinfo->dev_info;
1435 struct blk_shadow *s = &rinfo->shadow[*id];
1436 struct copy_from_grant data = {
1437 .grant_idx = 0,
1438 };
1439
1440 num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1441 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1442
1443 /* The I/O request may be split in two. */
1444 if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1445 struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1446
1447 /* Keep the status of the current response in shadow. */
1448 s->status = blkif_rsp_to_req_status(bret->status);
1449
1450 /* Wait the second response if not yet here. */
1451 if (s2->status == REQ_WAITING)
1452 return false;
1453
1454 bret->status = blkif_get_final_status(s->status,
1455 s2->status);
1456
1457 /*
1458 * All the grants is stored in the first shadow in order
1459 * to make the completion code simpler.
1460 */
1461 num_grant += s2->req.u.rw.nr_segments;
1462
1463 /*
1464 * The two responses may not come in order. Only the
1465 * first request will store the scatter-gather list.
1466 */
1467 if (s2->num_sg != 0) {
1468 /* Update "id" with the ID of the first response. */
1469 *id = s->associated_id;
1470 s = s2;
1471 }
1472
1473 /*
1474 * We don't need anymore the second request, so recycling
1475 * it now.
1476 */
1477 if (add_id_to_freelist(rinfo, s->associated_id))
1478 WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1479 info->gd->disk_name, s->associated_id);
1480 }
1481
1482 data.s = s;
1483 num_sg = s->num_sg;
1484
1485 if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1486 for_each_sg(s->sg, sg, num_sg, i) {
1487 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1488
1489 data.bvec_offset = sg->offset;
1490 data.bvec_data = kmap_atomic(sg_page(sg));
1491
1492 gnttab_foreach_grant_in_range(sg_page(sg),
1493 sg->offset,
1494 sg->length,
1495 blkif_copy_from_grant,
1496 &data);
1497
1498 kunmap_atomic(data.bvec_data);
1499 }
1500 }
1501 /* Add the persistent grant into the list of free grants */
1502 for (i = 0; i < num_grant; i++) {
1503 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1504 /*
1505 * If the grant is still mapped by the backend (the
1506 * backend has chosen to make this grant persistent)
1507 * we add it at the head of the list, so it will be
1508 * reused first.
1509 */
1510 if (!info->feature_persistent)
1511 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1512 s->grants_used[i]->gref);
1513 list_add(&s->grants_used[i]->node, &rinfo->grants);
1514 rinfo->persistent_gnts_c++;
1515 } else {
1516 /*
1517 * If the grant is not mapped by the backend we end the
1518 * foreign access and add it to the tail of the list,
1519 * so it will not be picked again unless we run out of
1520 * persistent grants.
1521 */
1522 gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1523 s->grants_used[i]->gref = GRANT_INVALID_REF;
1524 list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1525 }
1526 }
1527 if (s->req.operation == BLKIF_OP_INDIRECT) {
1528 for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1529 if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1530 if (!info->feature_persistent)
1531 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1532 s->indirect_grants[i]->gref);
1533 list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1534 rinfo->persistent_gnts_c++;
1535 } else {
1536 struct page *indirect_page;
1537
1538 gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1539 /*
1540 * Add the used indirect page back to the list of
1541 * available pages for indirect grefs.
1542 */
1543 if (!info->feature_persistent) {
1544 indirect_page = s->indirect_grants[i]->page;
1545 list_add(&indirect_page->lru, &rinfo->indirect_pages);
1546 }
1547 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1548 list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1549 }
1550 }
1551 }
1552
1553 return true;
1554}
1555
1556static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1557{
1558 struct request *req;
1559 struct blkif_response *bret;
1560 RING_IDX i, rp;
1561 unsigned long flags;
1562 struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1563 struct blkfront_info *info = rinfo->dev_info;
1564
1565 if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
1566 return IRQ_HANDLED;
1567
1568 spin_lock_irqsave(&rinfo->ring_lock, flags);
1569 again:
1570 rp = rinfo->ring.sring->rsp_prod;
1571 rmb(); /* Ensure we see queued responses up to 'rp'. */
1572
1573 for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1574 unsigned long id;
1575
1576 bret = RING_GET_RESPONSE(&rinfo->ring, i);
1577 id = bret->id;
1578 /*
1579 * The backend has messed up and given us an id that we would
1580 * never have given to it (we stamp it up to BLK_RING_SIZE -
1581 * look in get_id_from_freelist.
1582 */
1583 if (id >= BLK_RING_SIZE(info)) {
1584 WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1585 info->gd->disk_name, op_name(bret->operation), id);
1586 /* We can't safely get the 'struct request' as
1587 * the id is busted. */
1588 continue;
1589 }
1590 req = rinfo->shadow[id].request;
1591
1592 if (bret->operation != BLKIF_OP_DISCARD) {
1593 /*
1594 * We may need to wait for an extra response if the
1595 * I/O request is split in 2
1596 */
1597 if (!blkif_completion(&id, rinfo, bret))
1598 continue;
1599 }
1600
1601 if (add_id_to_freelist(rinfo, id)) {
1602 WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1603 info->gd->disk_name, op_name(bret->operation), id);
1604 continue;
1605 }
1606
1607 if (bret->status == BLKIF_RSP_OKAY)
1608 blkif_req(req)->error = BLK_STS_OK;
1609 else
1610 blkif_req(req)->error = BLK_STS_IOERR;
1611
1612 switch (bret->operation) {
1613 case BLKIF_OP_DISCARD:
1614 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1615 struct request_queue *rq = info->rq;
1616 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1617 info->gd->disk_name, op_name(bret->operation));
1618 blkif_req(req)->error = BLK_STS_NOTSUPP;
1619 info->feature_discard = 0;
1620 info->feature_secdiscard = 0;
1621 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1622 blk_queue_flag_clear(QUEUE_FLAG_SECERASE, rq);
1623 }
1624 break;
1625 case BLKIF_OP_FLUSH_DISKCACHE:
1626 case BLKIF_OP_WRITE_BARRIER:
1627 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1628 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1629 info->gd->disk_name, op_name(bret->operation));
1630 blkif_req(req)->error = BLK_STS_NOTSUPP;
1631 }
1632 if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1633 rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1634 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1635 info->gd->disk_name, op_name(bret->operation));
1636 blkif_req(req)->error = BLK_STS_NOTSUPP;
1637 }
1638 if (unlikely(blkif_req(req)->error)) {
1639 if (blkif_req(req)->error == BLK_STS_NOTSUPP)
1640 blkif_req(req)->error = BLK_STS_OK;
1641 info->feature_fua = 0;
1642 info->feature_flush = 0;
1643 xlvbd_flush(info);
1644 }
1645 fallthrough;
1646 case BLKIF_OP_READ:
1647 case BLKIF_OP_WRITE:
1648 if (unlikely(bret->status != BLKIF_RSP_OKAY))
1649 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1650 "request: %x\n", bret->status);
1651
1652 break;
1653 default:
1654 BUG();
1655 }
1656
1657 if (likely(!blk_should_fake_timeout(req->q)))
1658 blk_mq_complete_request(req);
1659 }
1660
1661 rinfo->ring.rsp_cons = i;
1662
1663 if (i != rinfo->ring.req_prod_pvt) {
1664 int more_to_do;
1665 RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1666 if (more_to_do)
1667 goto again;
1668 } else
1669 rinfo->ring.sring->rsp_event = i + 1;
1670
1671 kick_pending_request_queues_locked(rinfo);
1672
1673 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1674
1675 return IRQ_HANDLED;
1676}
1677
1678
1679static int setup_blkring(struct xenbus_device *dev,
1680 struct blkfront_ring_info *rinfo)
1681{
1682 struct blkif_sring *sring;
1683 int err, i;
1684 struct blkfront_info *info = rinfo->dev_info;
1685 unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1686 grant_ref_t gref[XENBUS_MAX_RING_GRANTS];
1687
1688 for (i = 0; i < info->nr_ring_pages; i++)
1689 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1690
1691 sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
1692 get_order(ring_size));
1693 if (!sring) {
1694 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1695 return -ENOMEM;
1696 }
1697 SHARED_RING_INIT(sring);
1698 FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1699
1700 err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref);
1701 if (err < 0) {
1702 free_pages((unsigned long)sring, get_order(ring_size));
1703 rinfo->ring.sring = NULL;
1704 goto fail;
1705 }
1706 for (i = 0; i < info->nr_ring_pages; i++)
1707 rinfo->ring_ref[i] = gref[i];
1708
1709 err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1710 if (err)
1711 goto fail;
1712
1713 err = bind_evtchn_to_irqhandler(rinfo->evtchn, blkif_interrupt, 0,
1714 "blkif", rinfo);
1715 if (err <= 0) {
1716 xenbus_dev_fatal(dev, err,
1717 "bind_evtchn_to_irqhandler failed");
1718 goto fail;
1719 }
1720 rinfo->irq = err;
1721
1722 return 0;
1723fail:
1724 blkif_free(info, 0);
1725 return err;
1726}
1727
1728/*
1729 * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1730 * ring buffer may have multi pages depending on ->nr_ring_pages.
1731 */
1732static int write_per_ring_nodes(struct xenbus_transaction xbt,
1733 struct blkfront_ring_info *rinfo, const char *dir)
1734{
1735 int err;
1736 unsigned int i;
1737 const char *message = NULL;
1738 struct blkfront_info *info = rinfo->dev_info;
1739
1740 if (info->nr_ring_pages == 1) {
1741 err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1742 if (err) {
1743 message = "writing ring-ref";
1744 goto abort_transaction;
1745 }
1746 } else {
1747 for (i = 0; i < info->nr_ring_pages; i++) {
1748 char ring_ref_name[RINGREF_NAME_LEN];
1749
1750 snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1751 err = xenbus_printf(xbt, dir, ring_ref_name,
1752 "%u", rinfo->ring_ref[i]);
1753 if (err) {
1754 message = "writing ring-ref";
1755 goto abort_transaction;
1756 }
1757 }
1758 }
1759
1760 err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1761 if (err) {
1762 message = "writing event-channel";
1763 goto abort_transaction;
1764 }
1765
1766 return 0;
1767
1768abort_transaction:
1769 xenbus_transaction_end(xbt, 1);
1770 if (message)
1771 xenbus_dev_fatal(info->xbdev, err, "%s", message);
1772
1773 return err;
1774}
1775
1776static void free_info(struct blkfront_info *info)
1777{
1778 list_del(&info->info_list);
1779 kfree(info);
1780}
1781
1782/* Common code used when first setting up, and when resuming. */
1783static int talk_to_blkback(struct xenbus_device *dev,
1784 struct blkfront_info *info)
1785{
1786 const char *message = NULL;
1787 struct xenbus_transaction xbt;
1788 int err;
1789 unsigned int i, max_page_order;
1790 unsigned int ring_page_order;
1791 struct blkfront_ring_info *rinfo;
1792
1793 if (!info)
1794 return -ENODEV;
1795
1796 max_page_order = xenbus_read_unsigned(info->xbdev->otherend,
1797 "max-ring-page-order", 0);
1798 ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1799 info->nr_ring_pages = 1 << ring_page_order;
1800
1801 err = negotiate_mq(info);
1802 if (err)
1803 goto destroy_blkring;
1804
1805 for_each_rinfo(info, rinfo, i) {
1806 /* Create shared ring, alloc event channel. */
1807 err = setup_blkring(dev, rinfo);
1808 if (err)
1809 goto destroy_blkring;
1810 }
1811
1812again:
1813 err = xenbus_transaction_start(&xbt);
1814 if (err) {
1815 xenbus_dev_fatal(dev, err, "starting transaction");
1816 goto destroy_blkring;
1817 }
1818
1819 if (info->nr_ring_pages > 1) {
1820 err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1821 ring_page_order);
1822 if (err) {
1823 message = "writing ring-page-order";
1824 goto abort_transaction;
1825 }
1826 }
1827
1828 /* We already got the number of queues/rings in _probe */
1829 if (info->nr_rings == 1) {
1830 err = write_per_ring_nodes(xbt, info->rinfo, dev->nodename);
1831 if (err)
1832 goto destroy_blkring;
1833 } else {
1834 char *path;
1835 size_t pathsize;
1836
1837 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1838 info->nr_rings);
1839 if (err) {
1840 message = "writing multi-queue-num-queues";
1841 goto abort_transaction;
1842 }
1843
1844 pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1845 path = kmalloc(pathsize, GFP_KERNEL);
1846 if (!path) {
1847 err = -ENOMEM;
1848 message = "ENOMEM while writing ring references";
1849 goto abort_transaction;
1850 }
1851
1852 for_each_rinfo(info, rinfo, i) {
1853 memset(path, 0, pathsize);
1854 snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1855 err = write_per_ring_nodes(xbt, rinfo, path);
1856 if (err) {
1857 kfree(path);
1858 goto destroy_blkring;
1859 }
1860 }
1861 kfree(path);
1862 }
1863 err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1864 XEN_IO_PROTO_ABI_NATIVE);
1865 if (err) {
1866 message = "writing protocol";
1867 goto abort_transaction;
1868 }
1869 err = xenbus_printf(xbt, dev->nodename,
1870 "feature-persistent", "%u", 1);
1871 if (err)
1872 dev_warn(&dev->dev,
1873 "writing persistent grants feature to xenbus");
1874
1875 err = xenbus_transaction_end(xbt, 0);
1876 if (err) {
1877 if (err == -EAGAIN)
1878 goto again;
1879 xenbus_dev_fatal(dev, err, "completing transaction");
1880 goto destroy_blkring;
1881 }
1882
1883 for_each_rinfo(info, rinfo, i) {
1884 unsigned int j;
1885
1886 for (j = 0; j < BLK_RING_SIZE(info); j++)
1887 rinfo->shadow[j].req.u.rw.id = j + 1;
1888 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1889 }
1890 xenbus_switch_state(dev, XenbusStateInitialised);
1891
1892 return 0;
1893
1894 abort_transaction:
1895 xenbus_transaction_end(xbt, 1);
1896 if (message)
1897 xenbus_dev_fatal(dev, err, "%s", message);
1898 destroy_blkring:
1899 blkif_free(info, 0);
1900
1901 mutex_lock(&blkfront_mutex);
1902 free_info(info);
1903 mutex_unlock(&blkfront_mutex);
1904
1905 dev_set_drvdata(&dev->dev, NULL);
1906
1907 return err;
1908}
1909
1910static int negotiate_mq(struct blkfront_info *info)
1911{
1912 unsigned int backend_max_queues;
1913 unsigned int i;
1914 struct blkfront_ring_info *rinfo;
1915
1916 BUG_ON(info->nr_rings);
1917
1918 /* Check if backend supports multiple queues. */
1919 backend_max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1920 "multi-queue-max-queues", 1);
1921 info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1922 /* We need at least one ring. */
1923 if (!info->nr_rings)
1924 info->nr_rings = 1;
1925
1926 info->rinfo_size = struct_size(info->rinfo, shadow,
1927 BLK_RING_SIZE(info));
1928 info->rinfo = kvcalloc(info->nr_rings, info->rinfo_size, GFP_KERNEL);
1929 if (!info->rinfo) {
1930 xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1931 info->nr_rings = 0;
1932 return -ENOMEM;
1933 }
1934
1935 for_each_rinfo(info, rinfo, i) {
1936 INIT_LIST_HEAD(&rinfo->indirect_pages);
1937 INIT_LIST_HEAD(&rinfo->grants);
1938 rinfo->dev_info = info;
1939 INIT_WORK(&rinfo->work, blkif_restart_queue);
1940 spin_lock_init(&rinfo->ring_lock);
1941 }
1942 return 0;
1943}
1944/**
1945 * Entry point to this code when a new device is created. Allocate the basic
1946 * structures and the ring buffer for communication with the backend, and
1947 * inform the backend of the appropriate details for those. Switch to
1948 * Initialised state.
1949 */
1950static int blkfront_probe(struct xenbus_device *dev,
1951 const struct xenbus_device_id *id)
1952{
1953 int err, vdevice;
1954 struct blkfront_info *info;
1955
1956 /* FIXME: Use dynamic device id if this is not set. */
1957 err = xenbus_scanf(XBT_NIL, dev->nodename,
1958 "virtual-device", "%i", &vdevice);
1959 if (err != 1) {
1960 /* go looking in the extended area instead */
1961 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1962 "%i", &vdevice);
1963 if (err != 1) {
1964 xenbus_dev_fatal(dev, err, "reading virtual-device");
1965 return err;
1966 }
1967 }
1968
1969 if (xen_hvm_domain()) {
1970 char *type;
1971 int len;
1972 /* no unplug has been done: do not hook devices != xen vbds */
1973 if (xen_has_pv_and_legacy_disk_devices()) {
1974 int major;
1975
1976 if (!VDEV_IS_EXTENDED(vdevice))
1977 major = BLKIF_MAJOR(vdevice);
1978 else
1979 major = XENVBD_MAJOR;
1980
1981 if (major != XENVBD_MAJOR) {
1982 printk(KERN_INFO
1983 "%s: HVM does not support vbd %d as xen block device\n",
1984 __func__, vdevice);
1985 return -ENODEV;
1986 }
1987 }
1988 /* do not create a PV cdrom device if we are an HVM guest */
1989 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1990 if (IS_ERR(type))
1991 return -ENODEV;
1992 if (strncmp(type, "cdrom", 5) == 0) {
1993 kfree(type);
1994 return -ENODEV;
1995 }
1996 kfree(type);
1997 }
1998 info = kzalloc(sizeof(*info), GFP_KERNEL);
1999 if (!info) {
2000 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
2001 return -ENOMEM;
2002 }
2003
2004 info->xbdev = dev;
2005
2006 mutex_init(&info->mutex);
2007 info->vdevice = vdevice;
2008 info->connected = BLKIF_STATE_DISCONNECTED;
2009
2010 /* Front end dir is a number, which is used as the id. */
2011 info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
2012 dev_set_drvdata(&dev->dev, info);
2013
2014 mutex_lock(&blkfront_mutex);
2015 list_add(&info->info_list, &info_list);
2016 mutex_unlock(&blkfront_mutex);
2017
2018 return 0;
2019}
2020
2021static int blkif_recover(struct blkfront_info *info)
2022{
2023 unsigned int r_index;
2024 struct request *req, *n;
2025 int rc;
2026 struct bio *bio;
2027 unsigned int segs;
2028 struct blkfront_ring_info *rinfo;
2029
2030 blkfront_gather_backend_features(info);
2031 /* Reset limits changed by blk_mq_update_nr_hw_queues(). */
2032 blkif_set_queue_limits(info);
2033 segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
2034 blk_queue_max_segments(info->rq, segs / GRANTS_PER_PSEG);
2035
2036 for_each_rinfo(info, rinfo, r_index) {
2037 rc = blkfront_setup_indirect(rinfo);
2038 if (rc)
2039 return rc;
2040 }
2041 xenbus_switch_state(info->xbdev, XenbusStateConnected);
2042
2043 /* Now safe for us to use the shared ring */
2044 info->connected = BLKIF_STATE_CONNECTED;
2045
2046 for_each_rinfo(info, rinfo, r_index) {
2047 /* Kick any other new requests queued since we resumed */
2048 kick_pending_request_queues(rinfo);
2049 }
2050
2051 list_for_each_entry_safe(req, n, &info->requests, queuelist) {
2052 /* Requeue pending requests (flush or discard) */
2053 list_del_init(&req->queuelist);
2054 BUG_ON(req->nr_phys_segments > segs);
2055 blk_mq_requeue_request(req, false);
2056 }
2057 blk_mq_start_stopped_hw_queues(info->rq, true);
2058 blk_mq_kick_requeue_list(info->rq);
2059
2060 while ((bio = bio_list_pop(&info->bio_list)) != NULL) {
2061 /* Traverse the list of pending bios and re-queue them */
2062 submit_bio(bio);
2063 }
2064
2065 return 0;
2066}
2067
2068/**
2069 * We are reconnecting to the backend, due to a suspend/resume, or a backend
2070 * driver restart. We tear down our blkif structure and recreate it, but
2071 * leave the device-layer structures intact so that this is transparent to the
2072 * rest of the kernel.
2073 */
2074static int blkfront_resume(struct xenbus_device *dev)
2075{
2076 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2077 int err = 0;
2078 unsigned int i, j;
2079 struct blkfront_ring_info *rinfo;
2080
2081 dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2082
2083 bio_list_init(&info->bio_list);
2084 INIT_LIST_HEAD(&info->requests);
2085 for_each_rinfo(info, rinfo, i) {
2086 struct bio_list merge_bio;
2087 struct blk_shadow *shadow = rinfo->shadow;
2088
2089 for (j = 0; j < BLK_RING_SIZE(info); j++) {
2090 /* Not in use? */
2091 if (!shadow[j].request)
2092 continue;
2093
2094 /*
2095 * Get the bios in the request so we can re-queue them.
2096 */
2097 if (req_op(shadow[j].request) == REQ_OP_FLUSH ||
2098 req_op(shadow[j].request) == REQ_OP_DISCARD ||
2099 req_op(shadow[j].request) == REQ_OP_SECURE_ERASE ||
2100 shadow[j].request->cmd_flags & REQ_FUA) {
2101 /*
2102 * Flush operations don't contain bios, so
2103 * we need to requeue the whole request
2104 *
2105 * XXX: but this doesn't make any sense for a
2106 * write with the FUA flag set..
2107 */
2108 list_add(&shadow[j].request->queuelist, &info->requests);
2109 continue;
2110 }
2111 merge_bio.head = shadow[j].request->bio;
2112 merge_bio.tail = shadow[j].request->biotail;
2113 bio_list_merge(&info->bio_list, &merge_bio);
2114 shadow[j].request->bio = NULL;
2115 blk_mq_end_request(shadow[j].request, BLK_STS_OK);
2116 }
2117 }
2118
2119 blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2120
2121 err = talk_to_blkback(dev, info);
2122 if (!err)
2123 blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings);
2124
2125 /*
2126 * We have to wait for the backend to switch to
2127 * connected state, since we want to read which
2128 * features it supports.
2129 */
2130
2131 return err;
2132}
2133
2134static void blkfront_closing(struct blkfront_info *info)
2135{
2136 struct xenbus_device *xbdev = info->xbdev;
2137 struct block_device *bdev = NULL;
2138
2139 mutex_lock(&info->mutex);
2140
2141 if (xbdev->state == XenbusStateClosing) {
2142 mutex_unlock(&info->mutex);
2143 return;
2144 }
2145
2146 if (info->gd)
2147 bdev = bdget_disk(info->gd, 0);
2148
2149 mutex_unlock(&info->mutex);
2150
2151 if (!bdev) {
2152 xenbus_frontend_closed(xbdev);
2153 return;
2154 }
2155
2156 mutex_lock(&bdev->bd_mutex);
2157
2158 if (bdev->bd_openers) {
2159 xenbus_dev_error(xbdev, -EBUSY,
2160 "Device in use; refusing to close");
2161 xenbus_switch_state(xbdev, XenbusStateClosing);
2162 } else {
2163 xlvbd_release_gendisk(info);
2164 xenbus_frontend_closed(xbdev);
2165 }
2166
2167 mutex_unlock(&bdev->bd_mutex);
2168 bdput(bdev);
2169}
2170
2171static void blkfront_setup_discard(struct blkfront_info *info)
2172{
2173 int err;
2174 unsigned int discard_granularity;
2175 unsigned int discard_alignment;
2176
2177 info->feature_discard = 1;
2178 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2179 "discard-granularity", "%u", &discard_granularity,
2180 "discard-alignment", "%u", &discard_alignment,
2181 NULL);
2182 if (!err) {
2183 info->discard_granularity = discard_granularity;
2184 info->discard_alignment = discard_alignment;
2185 }
2186 info->feature_secdiscard =
2187 !!xenbus_read_unsigned(info->xbdev->otherend, "discard-secure",
2188 0);
2189}
2190
2191static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2192{
2193 unsigned int psegs, grants, memflags;
2194 int err, i;
2195 struct blkfront_info *info = rinfo->dev_info;
2196
2197 memflags = memalloc_noio_save();
2198
2199 if (info->max_indirect_segments == 0) {
2200 if (!HAS_EXTRA_REQ)
2201 grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2202 else {
2203 /*
2204 * When an extra req is required, the maximum
2205 * grants supported is related to the size of the
2206 * Linux block segment.
2207 */
2208 grants = GRANTS_PER_PSEG;
2209 }
2210 }
2211 else
2212 grants = info->max_indirect_segments;
2213 psegs = DIV_ROUND_UP(grants, GRANTS_PER_PSEG);
2214
2215 err = fill_grant_buffer(rinfo,
2216 (grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2217 if (err)
2218 goto out_of_memory;
2219
2220 if (!info->feature_persistent && info->max_indirect_segments) {
2221 /*
2222 * We are using indirect descriptors but not persistent
2223 * grants, we need to allocate a set of pages that can be
2224 * used for mapping indirect grefs
2225 */
2226 int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2227
2228 BUG_ON(!list_empty(&rinfo->indirect_pages));
2229 for (i = 0; i < num; i++) {
2230 struct page *indirect_page = alloc_page(GFP_KERNEL);
2231 if (!indirect_page)
2232 goto out_of_memory;
2233 list_add(&indirect_page->lru, &rinfo->indirect_pages);
2234 }
2235 }
2236
2237 for (i = 0; i < BLK_RING_SIZE(info); i++) {
2238 rinfo->shadow[i].grants_used =
2239 kvcalloc(grants,
2240 sizeof(rinfo->shadow[i].grants_used[0]),
2241 GFP_KERNEL);
2242 rinfo->shadow[i].sg = kvcalloc(psegs,
2243 sizeof(rinfo->shadow[i].sg[0]),
2244 GFP_KERNEL);
2245 if (info->max_indirect_segments)
2246 rinfo->shadow[i].indirect_grants =
2247 kvcalloc(INDIRECT_GREFS(grants),
2248 sizeof(rinfo->shadow[i].indirect_grants[0]),
2249 GFP_KERNEL);
2250 if ((rinfo->shadow[i].grants_used == NULL) ||
2251 (rinfo->shadow[i].sg == NULL) ||
2252 (info->max_indirect_segments &&
2253 (rinfo->shadow[i].indirect_grants == NULL)))
2254 goto out_of_memory;
2255 sg_init_table(rinfo->shadow[i].sg, psegs);
2256 }
2257
2258 memalloc_noio_restore(memflags);
2259
2260 return 0;
2261
2262out_of_memory:
2263 for (i = 0; i < BLK_RING_SIZE(info); i++) {
2264 kvfree(rinfo->shadow[i].grants_used);
2265 rinfo->shadow[i].grants_used = NULL;
2266 kvfree(rinfo->shadow[i].sg);
2267 rinfo->shadow[i].sg = NULL;
2268 kvfree(rinfo->shadow[i].indirect_grants);
2269 rinfo->shadow[i].indirect_grants = NULL;
2270 }
2271 if (!list_empty(&rinfo->indirect_pages)) {
2272 struct page *indirect_page, *n;
2273 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2274 list_del(&indirect_page->lru);
2275 __free_page(indirect_page);
2276 }
2277 }
2278
2279 memalloc_noio_restore(memflags);
2280
2281 return -ENOMEM;
2282}
2283
2284/*
2285 * Gather all backend feature-*
2286 */
2287static void blkfront_gather_backend_features(struct blkfront_info *info)
2288{
2289 unsigned int indirect_segments;
2290
2291 info->feature_flush = 0;
2292 info->feature_fua = 0;
2293
2294 /*
2295 * If there's no "feature-barrier" defined, then it means
2296 * we're dealing with a very old backend which writes
2297 * synchronously; nothing to do.
2298 *
2299 * If there are barriers, then we use flush.
2300 */
2301 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-barrier", 0)) {
2302 info->feature_flush = 1;
2303 info->feature_fua = 1;
2304 }
2305
2306 /*
2307 * And if there is "feature-flush-cache" use that above
2308 * barriers.
2309 */
2310 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-flush-cache",
2311 0)) {
2312 info->feature_flush = 1;
2313 info->feature_fua = 0;
2314 }
2315
2316 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-discard", 0))
2317 blkfront_setup_discard(info);
2318
2319 info->feature_persistent =
2320 !!xenbus_read_unsigned(info->xbdev->otherend,
2321 "feature-persistent", 0);
2322
2323 indirect_segments = xenbus_read_unsigned(info->xbdev->otherend,
2324 "feature-max-indirect-segments", 0);
2325 if (indirect_segments > xen_blkif_max_segments)
2326 indirect_segments = xen_blkif_max_segments;
2327 if (indirect_segments <= BLKIF_MAX_SEGMENTS_PER_REQUEST)
2328 indirect_segments = 0;
2329 info->max_indirect_segments = indirect_segments;
2330
2331 if (info->feature_persistent) {
2332 mutex_lock(&blkfront_mutex);
2333 schedule_delayed_work(&blkfront_work, HZ * 10);
2334 mutex_unlock(&blkfront_mutex);
2335 }
2336}
2337
2338/*
2339 * Invoked when the backend is finally 'ready' (and has told produced
2340 * the details about the physical device - #sectors, size, etc).
2341 */
2342static void blkfront_connect(struct blkfront_info *info)
2343{
2344 unsigned long long sectors;
2345 unsigned long sector_size;
2346 unsigned int physical_sector_size;
2347 unsigned int binfo;
2348 int err, i;
2349 struct blkfront_ring_info *rinfo;
2350
2351 switch (info->connected) {
2352 case BLKIF_STATE_CONNECTED:
2353 /*
2354 * Potentially, the back-end may be signalling
2355 * a capacity change; update the capacity.
2356 */
2357 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2358 "sectors", "%Lu", §ors);
2359 if (XENBUS_EXIST_ERR(err))
2360 return;
2361 printk(KERN_INFO "Setting capacity to %Lu\n",
2362 sectors);
2363 set_capacity_revalidate_and_notify(info->gd, sectors, true);
2364
2365 return;
2366 case BLKIF_STATE_SUSPENDED:
2367 /*
2368 * If we are recovering from suspension, we need to wait
2369 * for the backend to announce it's features before
2370 * reconnecting, at least we need to know if the backend
2371 * supports indirect descriptors, and how many.
2372 */
2373 blkif_recover(info);
2374 return;
2375
2376 default:
2377 break;
2378 }
2379
2380 dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2381 __func__, info->xbdev->otherend);
2382
2383 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2384 "sectors", "%llu", §ors,
2385 "info", "%u", &binfo,
2386 "sector-size", "%lu", §or_size,
2387 NULL);
2388 if (err) {
2389 xenbus_dev_fatal(info->xbdev, err,
2390 "reading backend fields at %s",
2391 info->xbdev->otherend);
2392 return;
2393 }
2394
2395 /*
2396 * physcial-sector-size is a newer field, so old backends may not
2397 * provide this. Assume physical sector size to be the same as
2398 * sector_size in that case.
2399 */
2400 physical_sector_size = xenbus_read_unsigned(info->xbdev->otherend,
2401 "physical-sector-size",
2402 sector_size);
2403 blkfront_gather_backend_features(info);
2404 for_each_rinfo(info, rinfo, i) {
2405 err = blkfront_setup_indirect(rinfo);
2406 if (err) {
2407 xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2408 info->xbdev->otherend);
2409 blkif_free(info, 0);
2410 break;
2411 }
2412 }
2413
2414 err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
2415 physical_sector_size);
2416 if (err) {
2417 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2418 info->xbdev->otherend);
2419 goto fail;
2420 }
2421
2422 xenbus_switch_state(info->xbdev, XenbusStateConnected);
2423
2424 /* Kick pending requests. */
2425 info->connected = BLKIF_STATE_CONNECTED;
2426 for_each_rinfo(info, rinfo, i)
2427 kick_pending_request_queues(rinfo);
2428
2429 device_add_disk(&info->xbdev->dev, info->gd, NULL);
2430
2431 info->is_ready = 1;
2432 return;
2433
2434fail:
2435 blkif_free(info, 0);
2436 return;
2437}
2438
2439/**
2440 * Callback received when the backend's state changes.
2441 */
2442static void blkback_changed(struct xenbus_device *dev,
2443 enum xenbus_state backend_state)
2444{
2445 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2446
2447 dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2448
2449 switch (backend_state) {
2450 case XenbusStateInitWait:
2451 if (dev->state != XenbusStateInitialising)
2452 break;
2453 if (talk_to_blkback(dev, info))
2454 break;
2455 case XenbusStateInitialising:
2456 case XenbusStateInitialised:
2457 case XenbusStateReconfiguring:
2458 case XenbusStateReconfigured:
2459 case XenbusStateUnknown:
2460 break;
2461
2462 case XenbusStateConnected:
2463 /*
2464 * talk_to_blkback sets state to XenbusStateInitialised
2465 * and blkfront_connect sets it to XenbusStateConnected
2466 * (if connection went OK).
2467 *
2468 * If the backend (or toolstack) decides to poke at backend
2469 * state (and re-trigger the watch by setting the state repeatedly
2470 * to XenbusStateConnected (4)) we need to deal with this.
2471 * This is allowed as this is used to communicate to the guest
2472 * that the size of disk has changed!
2473 */
2474 if ((dev->state != XenbusStateInitialised) &&
2475 (dev->state != XenbusStateConnected)) {
2476 if (talk_to_blkback(dev, info))
2477 break;
2478 }
2479
2480 blkfront_connect(info);
2481 break;
2482
2483 case XenbusStateClosed:
2484 if (dev->state == XenbusStateClosed)
2485 break;
2486 fallthrough;
2487 case XenbusStateClosing:
2488 if (info)
2489 blkfront_closing(info);
2490 break;
2491 }
2492}
2493
2494static int blkfront_remove(struct xenbus_device *xbdev)
2495{
2496 struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2497 struct block_device *bdev = NULL;
2498 struct gendisk *disk;
2499
2500 dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2501
2502 if (!info)
2503 return 0;
2504
2505 blkif_free(info, 0);
2506
2507 mutex_lock(&info->mutex);
2508
2509 disk = info->gd;
2510 if (disk)
2511 bdev = bdget_disk(disk, 0);
2512
2513 info->xbdev = NULL;
2514 mutex_unlock(&info->mutex);
2515
2516 if (!bdev) {
2517 mutex_lock(&blkfront_mutex);
2518 free_info(info);
2519 mutex_unlock(&blkfront_mutex);
2520 return 0;
2521 }
2522
2523 /*
2524 * The xbdev was removed before we reached the Closed
2525 * state. See if it's safe to remove the disk. If the bdev
2526 * isn't closed yet, we let release take care of it.
2527 */
2528
2529 mutex_lock(&bdev->bd_mutex);
2530 info = disk->private_data;
2531
2532 dev_warn(disk_to_dev(disk),
2533 "%s was hot-unplugged, %d stale handles\n",
2534 xbdev->nodename, bdev->bd_openers);
2535
2536 if (info && !bdev->bd_openers) {
2537 xlvbd_release_gendisk(info);
2538 disk->private_data = NULL;
2539 mutex_lock(&blkfront_mutex);
2540 free_info(info);
2541 mutex_unlock(&blkfront_mutex);
2542 }
2543
2544 mutex_unlock(&bdev->bd_mutex);
2545 bdput(bdev);
2546
2547 return 0;
2548}
2549
2550static int blkfront_is_ready(struct xenbus_device *dev)
2551{
2552 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2553
2554 return info->is_ready && info->xbdev;
2555}
2556
2557static int blkif_open(struct block_device *bdev, fmode_t mode)
2558{
2559 struct gendisk *disk = bdev->bd_disk;
2560 struct blkfront_info *info;
2561 int err = 0;
2562
2563 mutex_lock(&blkfront_mutex);
2564
2565 info = disk->private_data;
2566 if (!info) {
2567 /* xbdev gone */
2568 err = -ERESTARTSYS;
2569 goto out;
2570 }
2571
2572 mutex_lock(&info->mutex);
2573
2574 if (!info->gd)
2575 /* xbdev is closed */
2576 err = -ERESTARTSYS;
2577
2578 mutex_unlock(&info->mutex);
2579
2580out:
2581 mutex_unlock(&blkfront_mutex);
2582 return err;
2583}
2584
2585static void blkif_release(struct gendisk *disk, fmode_t mode)
2586{
2587 struct blkfront_info *info = disk->private_data;
2588 struct block_device *bdev;
2589 struct xenbus_device *xbdev;
2590
2591 mutex_lock(&blkfront_mutex);
2592
2593 bdev = bdget_disk(disk, 0);
2594
2595 if (!bdev) {
2596 WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2597 goto out_mutex;
2598 }
2599 if (bdev->bd_openers)
2600 goto out;
2601
2602 /*
2603 * Check if we have been instructed to close. We will have
2604 * deferred this request, because the bdev was still open.
2605 */
2606
2607 mutex_lock(&info->mutex);
2608 xbdev = info->xbdev;
2609
2610 if (xbdev && xbdev->state == XenbusStateClosing) {
2611 /* pending switch to state closed */
2612 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2613 xlvbd_release_gendisk(info);
2614 xenbus_frontend_closed(info->xbdev);
2615 }
2616
2617 mutex_unlock(&info->mutex);
2618
2619 if (!xbdev) {
2620 /* sudden device removal */
2621 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2622 xlvbd_release_gendisk(info);
2623 disk->private_data = NULL;
2624 free_info(info);
2625 }
2626
2627out:
2628 bdput(bdev);
2629out_mutex:
2630 mutex_unlock(&blkfront_mutex);
2631}
2632
2633static const struct block_device_operations xlvbd_block_fops =
2634{
2635 .owner = THIS_MODULE,
2636 .open = blkif_open,
2637 .release = blkif_release,
2638 .getgeo = blkif_getgeo,
2639 .ioctl = blkif_ioctl,
2640 .compat_ioctl = blkdev_compat_ptr_ioctl,
2641};
2642
2643
2644static const struct xenbus_device_id blkfront_ids[] = {
2645 { "vbd" },
2646 { "" }
2647};
2648
2649static struct xenbus_driver blkfront_driver = {
2650 .ids = blkfront_ids,
2651 .probe = blkfront_probe,
2652 .remove = blkfront_remove,
2653 .resume = blkfront_resume,
2654 .otherend_changed = blkback_changed,
2655 .is_ready = blkfront_is_ready,
2656};
2657
2658static void purge_persistent_grants(struct blkfront_info *info)
2659{
2660 unsigned int i;
2661 unsigned long flags;
2662 struct blkfront_ring_info *rinfo;
2663
2664 for_each_rinfo(info, rinfo, i) {
2665 struct grant *gnt_list_entry, *tmp;
2666
2667 spin_lock_irqsave(&rinfo->ring_lock, flags);
2668
2669 if (rinfo->persistent_gnts_c == 0) {
2670 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2671 continue;
2672 }
2673
2674 list_for_each_entry_safe(gnt_list_entry, tmp, &rinfo->grants,
2675 node) {
2676 if (gnt_list_entry->gref == GRANT_INVALID_REF ||
2677 gnttab_query_foreign_access(gnt_list_entry->gref))
2678 continue;
2679
2680 list_del(&gnt_list_entry->node);
2681 gnttab_end_foreign_access(gnt_list_entry->gref, 0, 0UL);
2682 rinfo->persistent_gnts_c--;
2683 gnt_list_entry->gref = GRANT_INVALID_REF;
2684 list_add_tail(&gnt_list_entry->node, &rinfo->grants);
2685 }
2686
2687 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2688 }
2689}
2690
2691static void blkfront_delay_work(struct work_struct *work)
2692{
2693 struct blkfront_info *info;
2694 bool need_schedule_work = false;
2695
2696 mutex_lock(&blkfront_mutex);
2697
2698 list_for_each_entry(info, &info_list, info_list) {
2699 if (info->feature_persistent) {
2700 need_schedule_work = true;
2701 mutex_lock(&info->mutex);
2702 purge_persistent_grants(info);
2703 mutex_unlock(&info->mutex);
2704 }
2705 }
2706
2707 if (need_schedule_work)
2708 schedule_delayed_work(&blkfront_work, HZ * 10);
2709
2710 mutex_unlock(&blkfront_mutex);
2711}
2712
2713static int __init xlblk_init(void)
2714{
2715 int ret;
2716 int nr_cpus = num_online_cpus();
2717
2718 if (!xen_domain())
2719 return -ENODEV;
2720
2721 if (!xen_has_pv_disk_devices())
2722 return -ENODEV;
2723
2724 if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2725 pr_warn("xen_blk: can't get major %d with name %s\n",
2726 XENVBD_MAJOR, DEV_NAME);
2727 return -ENODEV;
2728 }
2729
2730 if (xen_blkif_max_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST)
2731 xen_blkif_max_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2732
2733 if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2734 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2735 xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2736 xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2737 }
2738
2739 if (xen_blkif_max_queues > nr_cpus) {
2740 pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2741 xen_blkif_max_queues, nr_cpus);
2742 xen_blkif_max_queues = nr_cpus;
2743 }
2744
2745 INIT_DELAYED_WORK(&blkfront_work, blkfront_delay_work);
2746
2747 ret = xenbus_register_frontend(&blkfront_driver);
2748 if (ret) {
2749 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2750 return ret;
2751 }
2752
2753 return 0;
2754}
2755module_init(xlblk_init);
2756
2757
2758static void __exit xlblk_exit(void)
2759{
2760 cancel_delayed_work_sync(&blkfront_work);
2761
2762 xenbus_unregister_driver(&blkfront_driver);
2763 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2764 kfree(minors);
2765}
2766module_exit(xlblk_exit);
2767
2768MODULE_DESCRIPTION("Xen virtual block device frontend");
2769MODULE_LICENSE("GPL");
2770MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2771MODULE_ALIAS("xen:vbd");
2772MODULE_ALIAS("xenblk");