bcache: don't attach backing with duplicate UUID
[muen/linux.git] / drivers / block / xen-blkfront.c
1 /*
2  * blkfront.c
3  *
4  * XenLinux virtual block device driver.
5  *
6  * Copyright (c) 2003-2004, Keir Fraser & Steve Hand
7  * Modifications by Mark A. Williamson are (c) Intel Research Cambridge
8  * Copyright (c) 2004, Christian Limpach
9  * Copyright (c) 2004, Andrew Warfield
10  * Copyright (c) 2005, Christopher Clark
11  * Copyright (c) 2005, XenSource Ltd
12  *
13  * This program is free software; you can redistribute it and/or
14  * modify it under the terms of the GNU General Public License version 2
15  * as published by the Free Software Foundation; or, when distributed
16  * separately from the Linux kernel or incorporated into other
17  * software packages, subject to the following license:
18  *
19  * Permission is hereby granted, free of charge, to any person obtaining a copy
20  * of this source file (the "Software"), to deal in the Software without
21  * restriction, including without limitation the rights to use, copy, modify,
22  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
23  * and to permit persons to whom the Software is furnished to do so, subject to
24  * the following conditions:
25  *
26  * The above copyright notice and this permission notice shall be included in
27  * all copies or substantial portions of the Software.
28  *
29  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
30  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
31  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
32  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
33  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
34  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
35  * IN THE SOFTWARE.
36  */
37
38 #include <linux/interrupt.h>
39 #include <linux/blkdev.h>
40 #include <linux/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
50 #include <xen/xen.h>
51 #include <xen/xenbus.h>
52 #include <xen/grant_table.h>
53 #include <xen/events.h>
54 #include <xen/page.h>
55 #include <xen/platform_pci.h>
56
57 #include <xen/interface/grant_table.h>
58 #include <xen/interface/io/blkif.h>
59 #include <xen/interface/io/protocols.h>
60
61 #include <asm/xen/hypervisor.h>
62
63 /*
64  * The minimal size of segment supported by the block framework is PAGE_SIZE.
65  * When Linux is using a different page size than Xen, it may not be possible
66  * to put all the data in a single segment.
67  * This can happen when the backend doesn't support indirect descriptor and
68  * therefore the maximum amount of data that a request can carry is
69  * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE = 44KB
70  *
71  * Note that we only support one extra request. So the Linux page size
72  * should be <= ( 2 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) =
73  * 88KB.
74  */
75 #define HAS_EXTRA_REQ (BLKIF_MAX_SEGMENTS_PER_REQUEST < XEN_PFN_PER_PAGE)
76
77 enum blkif_state {
78         BLKIF_STATE_DISCONNECTED,
79         BLKIF_STATE_CONNECTED,
80         BLKIF_STATE_SUSPENDED,
81 };
82
83 struct grant {
84         grant_ref_t gref;
85         struct page *page;
86         struct list_head node;
87 };
88
89 enum blk_req_status {
90         REQ_WAITING,
91         REQ_DONE,
92         REQ_ERROR,
93         REQ_EOPNOTSUPP,
94 };
95
96 struct blk_shadow {
97         struct blkif_request req;
98         struct request *request;
99         struct grant **grants_used;
100         struct grant **indirect_grants;
101         struct scatterlist *sg;
102         unsigned int num_sg;
103         enum blk_req_status status;
104
105         #define NO_ASSOCIATED_ID ~0UL
106         /*
107          * Id of the sibling if we ever need 2 requests when handling a
108          * block I/O request
109          */
110         unsigned long associated_id;
111 };
112
113 struct blkif_req {
114         blk_status_t    error;
115 };
116
117 static inline struct blkif_req *blkif_req(struct request *rq)
118 {
119         return blk_mq_rq_to_pdu(rq);
120 }
121
122 static DEFINE_MUTEX(blkfront_mutex);
123 static const struct block_device_operations xlvbd_block_fops;
124
125 /*
126  * Maximum number of segments in indirect requests, the actual value used by
127  * the frontend driver is the minimum of this value and the value provided
128  * by the backend driver.
129  */
130
131 static unsigned int xen_blkif_max_segments = 32;
132 module_param_named(max_indirect_segments, xen_blkif_max_segments, uint,
133                    S_IRUGO);
134 MODULE_PARM_DESC(max_indirect_segments,
135                  "Maximum amount of segments in indirect requests (default is 32)");
136
137 static unsigned int xen_blkif_max_queues = 4;
138 module_param_named(max_queues, xen_blkif_max_queues, uint, S_IRUGO);
139 MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
140
141 /*
142  * Maximum order of pages to be used for the shared ring between front and
143  * backend, 4KB page granularity is used.
144  */
145 static unsigned int xen_blkif_max_ring_order;
146 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, S_IRUGO);
147 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
148
149 #define BLK_RING_SIZE(info)     \
150         __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
151
152 #define BLK_MAX_RING_SIZE       \
153         __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * XENBUS_MAX_RING_GRANTS)
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  */
170 struct 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 blk_shadow shadow[BLK_MAX_RING_SIZE];
179         struct list_head indirect_pages;
180         struct list_head grants;
181         unsigned int persistent_gnts_c;
182         unsigned long shadow_free;
183         struct blkfront_info *dev_info;
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  */
191 struct 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         /* Save uncomplete reqs and bios for migration. */
218         struct list_head requests;
219         struct bio_list bio_list;
220 };
221
222 static unsigned int nr_minors;
223 static unsigned long *minors;
224 static DEFINE_SPINLOCK(minor_lock);
225
226 #define GRANT_INVALID_REF       0
227
228 #define PARTS_PER_DISK          16
229 #define PARTS_PER_EXT_DISK      256
230
231 #define BLKIF_MAJOR(dev) ((dev)>>8)
232 #define BLKIF_MINOR(dev) ((dev) & 0xff)
233
234 #define EXT_SHIFT 28
235 #define EXTENDED (1<<EXT_SHIFT)
236 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
237 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
238 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
239 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
240 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
241 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
242
243 #define DEV_NAME        "xvd"   /* name in /dev */
244
245 /*
246  * Grants are always the same size as a Xen page (i.e 4KB).
247  * A physical segment is always the same size as a Linux page.
248  * Number of grants per physical segment
249  */
250 #define GRANTS_PER_PSEG (PAGE_SIZE / XEN_PAGE_SIZE)
251
252 #define GRANTS_PER_INDIRECT_FRAME \
253         (XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
254
255 #define PSEGS_PER_INDIRECT_FRAME        \
256         (GRANTS_INDIRECT_FRAME / GRANTS_PSEGS)
257
258 #define INDIRECT_GREFS(_grants)         \
259         DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
260
261 #define GREFS(_psegs)   ((_psegs) * GRANTS_PER_PSEG)
262
263 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
264 static void blkfront_gather_backend_features(struct blkfront_info *info);
265
266 static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
267 {
268         unsigned long free = rinfo->shadow_free;
269
270         BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
271         rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
272         rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
273         return free;
274 }
275
276 static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
277                               unsigned long id)
278 {
279         if (rinfo->shadow[id].req.u.rw.id != id)
280                 return -EINVAL;
281         if (rinfo->shadow[id].request == NULL)
282                 return -EINVAL;
283         rinfo->shadow[id].req.u.rw.id  = rinfo->shadow_free;
284         rinfo->shadow[id].request = NULL;
285         rinfo->shadow_free = id;
286         return 0;
287 }
288
289 static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
290 {
291         struct blkfront_info *info = rinfo->dev_info;
292         struct page *granted_page;
293         struct grant *gnt_list_entry, *n;
294         int i = 0;
295
296         while (i < num) {
297                 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
298                 if (!gnt_list_entry)
299                         goto out_of_memory;
300
301                 if (info->feature_persistent) {
302                         granted_page = alloc_page(GFP_NOIO);
303                         if (!granted_page) {
304                                 kfree(gnt_list_entry);
305                                 goto out_of_memory;
306                         }
307                         gnt_list_entry->page = granted_page;
308                 }
309
310                 gnt_list_entry->gref = GRANT_INVALID_REF;
311                 list_add(&gnt_list_entry->node, &rinfo->grants);
312                 i++;
313         }
314
315         return 0;
316
317 out_of_memory:
318         list_for_each_entry_safe(gnt_list_entry, n,
319                                  &rinfo->grants, node) {
320                 list_del(&gnt_list_entry->node);
321                 if (info->feature_persistent)
322                         __free_page(gnt_list_entry->page);
323                 kfree(gnt_list_entry);
324                 i--;
325         }
326         BUG_ON(i != 0);
327         return -ENOMEM;
328 }
329
330 static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
331 {
332         struct grant *gnt_list_entry;
333
334         BUG_ON(list_empty(&rinfo->grants));
335         gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
336                                           node);
337         list_del(&gnt_list_entry->node);
338
339         if (gnt_list_entry->gref != GRANT_INVALID_REF)
340                 rinfo->persistent_gnts_c--;
341
342         return gnt_list_entry;
343 }
344
345 static inline void grant_foreign_access(const struct grant *gnt_list_entry,
346                                         const struct blkfront_info *info)
347 {
348         gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
349                                                  info->xbdev->otherend_id,
350                                                  gnt_list_entry->page,
351                                                  0);
352 }
353
354 static struct grant *get_grant(grant_ref_t *gref_head,
355                                unsigned long gfn,
356                                struct blkfront_ring_info *rinfo)
357 {
358         struct grant *gnt_list_entry = get_free_grant(rinfo);
359         struct blkfront_info *info = rinfo->dev_info;
360
361         if (gnt_list_entry->gref != GRANT_INVALID_REF)
362                 return gnt_list_entry;
363
364         /* Assign a gref to this page */
365         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
366         BUG_ON(gnt_list_entry->gref == -ENOSPC);
367         if (info->feature_persistent)
368                 grant_foreign_access(gnt_list_entry, info);
369         else {
370                 /* Grant access to the GFN passed by the caller */
371                 gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
372                                                 info->xbdev->otherend_id,
373                                                 gfn, 0);
374         }
375
376         return gnt_list_entry;
377 }
378
379 static struct grant *get_indirect_grant(grant_ref_t *gref_head,
380                                         struct blkfront_ring_info *rinfo)
381 {
382         struct grant *gnt_list_entry = get_free_grant(rinfo);
383         struct blkfront_info *info = rinfo->dev_info;
384
385         if (gnt_list_entry->gref != GRANT_INVALID_REF)
386                 return gnt_list_entry;
387
388         /* Assign a gref to this page */
389         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
390         BUG_ON(gnt_list_entry->gref == -ENOSPC);
391         if (!info->feature_persistent) {
392                 struct page *indirect_page;
393
394                 /* Fetch a pre-allocated page to use for indirect grefs */
395                 BUG_ON(list_empty(&rinfo->indirect_pages));
396                 indirect_page = list_first_entry(&rinfo->indirect_pages,
397                                                  struct page, lru);
398                 list_del(&indirect_page->lru);
399                 gnt_list_entry->page = indirect_page;
400         }
401         grant_foreign_access(gnt_list_entry, info);
402
403         return gnt_list_entry;
404 }
405
406 static const char *op_name(int op)
407 {
408         static const char *const names[] = {
409                 [BLKIF_OP_READ] = "read",
410                 [BLKIF_OP_WRITE] = "write",
411                 [BLKIF_OP_WRITE_BARRIER] = "barrier",
412                 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
413                 [BLKIF_OP_DISCARD] = "discard" };
414
415         if (op < 0 || op >= ARRAY_SIZE(names))
416                 return "unknown";
417
418         if (!names[op])
419                 return "reserved";
420
421         return names[op];
422 }
423 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
424 {
425         unsigned int end = minor + nr;
426         int rc;
427
428         if (end > nr_minors) {
429                 unsigned long *bitmap, *old;
430
431                 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
432                                  GFP_KERNEL);
433                 if (bitmap == NULL)
434                         return -ENOMEM;
435
436                 spin_lock(&minor_lock);
437                 if (end > nr_minors) {
438                         old = minors;
439                         memcpy(bitmap, minors,
440                                BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
441                         minors = bitmap;
442                         nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
443                 } else
444                         old = bitmap;
445                 spin_unlock(&minor_lock);
446                 kfree(old);
447         }
448
449         spin_lock(&minor_lock);
450         if (find_next_bit(minors, end, minor) >= end) {
451                 bitmap_set(minors, minor, nr);
452                 rc = 0;
453         } else
454                 rc = -EBUSY;
455         spin_unlock(&minor_lock);
456
457         return rc;
458 }
459
460 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
461 {
462         unsigned int end = minor + nr;
463
464         BUG_ON(end > nr_minors);
465         spin_lock(&minor_lock);
466         bitmap_clear(minors,  minor, nr);
467         spin_unlock(&minor_lock);
468 }
469
470 static void blkif_restart_queue_callback(void *arg)
471 {
472         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
473         schedule_work(&rinfo->work);
474 }
475
476 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
477 {
478         /* We don't have real geometry info, but let's at least return
479            values consistent with the size of the device */
480         sector_t nsect = get_capacity(bd->bd_disk);
481         sector_t cylinders = nsect;
482
483         hg->heads = 0xff;
484         hg->sectors = 0x3f;
485         sector_div(cylinders, hg->heads * hg->sectors);
486         hg->cylinders = cylinders;
487         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
488                 hg->cylinders = 0xffff;
489         return 0;
490 }
491
492 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
493                        unsigned command, unsigned long argument)
494 {
495         struct blkfront_info *info = bdev->bd_disk->private_data;
496         int i;
497
498         dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
499                 command, (long)argument);
500
501         switch (command) {
502         case CDROMMULTISESSION:
503                 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
504                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
505                         if (put_user(0, (char __user *)(argument + i)))
506                                 return -EFAULT;
507                 return 0;
508
509         case CDROM_GET_CAPABILITY: {
510                 struct gendisk *gd = info->gd;
511                 if (gd->flags & GENHD_FL_CD)
512                         return 0;
513                 return -EINVAL;
514         }
515
516         default:
517                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
518                   command);*/
519                 return -EINVAL; /* same return as native Linux */
520         }
521
522         return 0;
523 }
524
525 static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
526                                             struct request *req,
527                                             struct blkif_request **ring_req)
528 {
529         unsigned long id;
530
531         *ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
532         rinfo->ring.req_prod_pvt++;
533
534         id = get_id_from_freelist(rinfo);
535         rinfo->shadow[id].request = req;
536         rinfo->shadow[id].status = REQ_WAITING;
537         rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
538
539         (*ring_req)->u.rw.id = id;
540
541         return id;
542 }
543
544 static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
545 {
546         struct blkfront_info *info = rinfo->dev_info;
547         struct blkif_request *ring_req;
548         unsigned long id;
549
550         /* Fill out a communications ring structure. */
551         id = blkif_ring_get_request(rinfo, req, &ring_req);
552
553         ring_req->operation = BLKIF_OP_DISCARD;
554         ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
555         ring_req->u.discard.id = id;
556         ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
557         if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard)
558                 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
559         else
560                 ring_req->u.discard.flag = 0;
561
562         /* Keep a private copy so we can reissue requests when recovering. */
563         rinfo->shadow[id].req = *ring_req;
564
565         return 0;
566 }
567
568 struct setup_rw_req {
569         unsigned int grant_idx;
570         struct blkif_request_segment *segments;
571         struct blkfront_ring_info *rinfo;
572         struct blkif_request *ring_req;
573         grant_ref_t gref_head;
574         unsigned int id;
575         /* Only used when persistent grant is used and it's a read request */
576         bool need_copy;
577         unsigned int bvec_off;
578         char *bvec_data;
579
580         bool require_extra_req;
581         struct blkif_request *extra_ring_req;
582 };
583
584 static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
585                                      unsigned int len, void *data)
586 {
587         struct setup_rw_req *setup = data;
588         int n, ref;
589         struct grant *gnt_list_entry;
590         unsigned int fsect, lsect;
591         /* Convenient aliases */
592         unsigned int grant_idx = setup->grant_idx;
593         struct blkif_request *ring_req = setup->ring_req;
594         struct blkfront_ring_info *rinfo = setup->rinfo;
595         /*
596          * We always use the shadow of the first request to store the list
597          * of grant associated to the block I/O request. This made the
598          * completion more easy to handle even if the block I/O request is
599          * split.
600          */
601         struct blk_shadow *shadow = &rinfo->shadow[setup->id];
602
603         if (unlikely(setup->require_extra_req &&
604                      grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
605                 /*
606                  * We are using the second request, setup grant_idx
607                  * to be the index of the segment array.
608                  */
609                 grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
610                 ring_req = setup->extra_ring_req;
611         }
612
613         if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
614             (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
615                 if (setup->segments)
616                         kunmap_atomic(setup->segments);
617
618                 n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
619                 gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
620                 shadow->indirect_grants[n] = gnt_list_entry;
621                 setup->segments = kmap_atomic(gnt_list_entry->page);
622                 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
623         }
624
625         gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
626         ref = gnt_list_entry->gref;
627         /*
628          * All the grants are stored in the shadow of the first
629          * request. Therefore we have to use the global index.
630          */
631         shadow->grants_used[setup->grant_idx] = gnt_list_entry;
632
633         if (setup->need_copy) {
634                 void *shared_data;
635
636                 shared_data = kmap_atomic(gnt_list_entry->page);
637                 /*
638                  * this does not wipe data stored outside the
639                  * range sg->offset..sg->offset+sg->length.
640                  * Therefore, blkback *could* see data from
641                  * previous requests. This is OK as long as
642                  * persistent grants are shared with just one
643                  * domain. It may need refactoring if this
644                  * changes
645                  */
646                 memcpy(shared_data + offset,
647                        setup->bvec_data + setup->bvec_off,
648                        len);
649
650                 kunmap_atomic(shared_data);
651                 setup->bvec_off += len;
652         }
653
654         fsect = offset >> 9;
655         lsect = fsect + (len >> 9) - 1;
656         if (ring_req->operation != BLKIF_OP_INDIRECT) {
657                 ring_req->u.rw.seg[grant_idx] =
658                         (struct blkif_request_segment) {
659                                 .gref       = ref,
660                                 .first_sect = fsect,
661                                 .last_sect  = lsect };
662         } else {
663                 setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
664                         (struct blkif_request_segment) {
665                                 .gref       = ref,
666                                 .first_sect = fsect,
667                                 .last_sect  = lsect };
668         }
669
670         (setup->grant_idx)++;
671 }
672
673 static void blkif_setup_extra_req(struct blkif_request *first,
674                                   struct blkif_request *second)
675 {
676         uint16_t nr_segments = first->u.rw.nr_segments;
677
678         /*
679          * The second request is only present when the first request uses
680          * all its segments. It's always the continuity of the first one.
681          */
682         first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
683
684         second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
685         second->u.rw.sector_number = first->u.rw.sector_number +
686                 (BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
687
688         second->u.rw.handle = first->u.rw.handle;
689         second->operation = first->operation;
690 }
691
692 static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
693 {
694         struct blkfront_info *info = rinfo->dev_info;
695         struct blkif_request *ring_req, *extra_ring_req = NULL;
696         unsigned long id, extra_id = NO_ASSOCIATED_ID;
697         bool require_extra_req = false;
698         int i;
699         struct setup_rw_req setup = {
700                 .grant_idx = 0,
701                 .segments = NULL,
702                 .rinfo = rinfo,
703                 .need_copy = rq_data_dir(req) && info->feature_persistent,
704         };
705
706         /*
707          * Used to store if we are able to queue the request by just using
708          * existing persistent grants, or if we have to get new grants,
709          * as there are not sufficiently many free.
710          */
711         bool new_persistent_gnts = false;
712         struct scatterlist *sg;
713         int num_sg, max_grefs, num_grant;
714
715         max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
716         if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
717                 /*
718                  * If we are using indirect segments we need to account
719                  * for the indirect grefs used in the request.
720                  */
721                 max_grefs += INDIRECT_GREFS(max_grefs);
722
723         /* Check if we have enough persistent grants to allocate a requests */
724         if (rinfo->persistent_gnts_c < max_grefs) {
725                 new_persistent_gnts = true;
726
727                 if (gnttab_alloc_grant_references(
728                     max_grefs - rinfo->persistent_gnts_c,
729                     &setup.gref_head) < 0) {
730                         gnttab_request_free_callback(
731                                 &rinfo->callback,
732                                 blkif_restart_queue_callback,
733                                 rinfo,
734                                 max_grefs - rinfo->persistent_gnts_c);
735                         return 1;
736                 }
737         }
738
739         /* Fill out a communications ring structure. */
740         id = blkif_ring_get_request(rinfo, req, &ring_req);
741
742         num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
743         num_grant = 0;
744         /* Calculate the number of grant used */
745         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
746                num_grant += gnttab_count_grant(sg->offset, sg->length);
747
748         require_extra_req = info->max_indirect_segments == 0 &&
749                 num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
750         BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
751
752         rinfo->shadow[id].num_sg = num_sg;
753         if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
754             likely(!require_extra_req)) {
755                 /*
756                  * The indirect operation can only be a BLKIF_OP_READ or
757                  * BLKIF_OP_WRITE
758                  */
759                 BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA);
760                 ring_req->operation = BLKIF_OP_INDIRECT;
761                 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
762                         BLKIF_OP_WRITE : BLKIF_OP_READ;
763                 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
764                 ring_req->u.indirect.handle = info->handle;
765                 ring_req->u.indirect.nr_segments = num_grant;
766         } else {
767                 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
768                 ring_req->u.rw.handle = info->handle;
769                 ring_req->operation = rq_data_dir(req) ?
770                         BLKIF_OP_WRITE : BLKIF_OP_READ;
771                 if (req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA) {
772                         /*
773                          * Ideally we can do an unordered flush-to-disk.
774                          * In case the backend onlysupports barriers, use that.
775                          * A barrier request a superset of FUA, so we can
776                          * implement it the same way.  (It's also a FLUSH+FUA,
777                          * since it is guaranteed ordered WRT previous writes.)
778                          */
779                         if (info->feature_flush && info->feature_fua)
780                                 ring_req->operation =
781                                         BLKIF_OP_WRITE_BARRIER;
782                         else if (info->feature_flush)
783                                 ring_req->operation =
784                                         BLKIF_OP_FLUSH_DISKCACHE;
785                         else
786                                 ring_req->operation = 0;
787                 }
788                 ring_req->u.rw.nr_segments = num_grant;
789                 if (unlikely(require_extra_req)) {
790                         extra_id = blkif_ring_get_request(rinfo, req,
791                                                           &extra_ring_req);
792                         /*
793                          * Only the first request contains the scatter-gather
794                          * list.
795                          */
796                         rinfo->shadow[extra_id].num_sg = 0;
797
798                         blkif_setup_extra_req(ring_req, extra_ring_req);
799
800                         /* Link the 2 requests together */
801                         rinfo->shadow[extra_id].associated_id = id;
802                         rinfo->shadow[id].associated_id = extra_id;
803                 }
804         }
805
806         setup.ring_req = ring_req;
807         setup.id = id;
808
809         setup.require_extra_req = require_extra_req;
810         if (unlikely(require_extra_req))
811                 setup.extra_ring_req = extra_ring_req;
812
813         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
814                 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
815
816                 if (setup.need_copy) {
817                         setup.bvec_off = sg->offset;
818                         setup.bvec_data = kmap_atomic(sg_page(sg));
819                 }
820
821                 gnttab_foreach_grant_in_range(sg_page(sg),
822                                               sg->offset,
823                                               sg->length,
824                                               blkif_setup_rw_req_grant,
825                                               &setup);
826
827                 if (setup.need_copy)
828                         kunmap_atomic(setup.bvec_data);
829         }
830         if (setup.segments)
831                 kunmap_atomic(setup.segments);
832
833         /* Keep a private copy so we can reissue requests when recovering. */
834         rinfo->shadow[id].req = *ring_req;
835         if (unlikely(require_extra_req))
836                 rinfo->shadow[extra_id].req = *extra_ring_req;
837
838         if (new_persistent_gnts)
839                 gnttab_free_grant_references(setup.gref_head);
840
841         return 0;
842 }
843
844 /*
845  * Generate a Xen blkfront IO request from a blk layer request.  Reads
846  * and writes are handled as expected.
847  *
848  * @req: a request struct
849  */
850 static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
851 {
852         if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
853                 return 1;
854
855         if (unlikely(req_op(req) == REQ_OP_DISCARD ||
856                      req_op(req) == REQ_OP_SECURE_ERASE))
857                 return blkif_queue_discard_req(req, rinfo);
858         else
859                 return blkif_queue_rw_req(req, rinfo);
860 }
861
862 static inline void flush_requests(struct blkfront_ring_info *rinfo)
863 {
864         int notify;
865
866         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
867
868         if (notify)
869                 notify_remote_via_irq(rinfo->irq);
870 }
871
872 static inline bool blkif_request_flush_invalid(struct request *req,
873                                                struct blkfront_info *info)
874 {
875         return (blk_rq_is_passthrough(req) ||
876                 ((req_op(req) == REQ_OP_FLUSH) &&
877                  !info->feature_flush) ||
878                 ((req->cmd_flags & REQ_FUA) &&
879                  !info->feature_fua));
880 }
881
882 static blk_status_t blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
883                           const struct blk_mq_queue_data *qd)
884 {
885         unsigned long flags;
886         int qid = hctx->queue_num;
887         struct blkfront_info *info = hctx->queue->queuedata;
888         struct blkfront_ring_info *rinfo = NULL;
889
890         BUG_ON(info->nr_rings <= qid);
891         rinfo = &info->rinfo[qid];
892         blk_mq_start_request(qd->rq);
893         spin_lock_irqsave(&rinfo->ring_lock, flags);
894         if (RING_FULL(&rinfo->ring))
895                 goto out_busy;
896
897         if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
898                 goto out_err;
899
900         if (blkif_queue_request(qd->rq, rinfo))
901                 goto out_busy;
902
903         flush_requests(rinfo);
904         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
905         return BLK_STS_OK;
906
907 out_err:
908         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
909         return BLK_STS_IOERR;
910
911 out_busy:
912         blk_mq_stop_hw_queue(hctx);
913         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
914         return BLK_STS_DEV_RESOURCE;
915 }
916
917 static void blkif_complete_rq(struct request *rq)
918 {
919         blk_mq_end_request(rq, blkif_req(rq)->error);
920 }
921
922 static const struct blk_mq_ops blkfront_mq_ops = {
923         .queue_rq = blkif_queue_rq,
924         .complete = blkif_complete_rq,
925 };
926
927 static void blkif_set_queue_limits(struct blkfront_info *info)
928 {
929         struct request_queue *rq = info->rq;
930         struct gendisk *gd = info->gd;
931         unsigned int segments = info->max_indirect_segments ? :
932                                 BLKIF_MAX_SEGMENTS_PER_REQUEST;
933
934         queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
935
936         if (info->feature_discard) {
937                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
938                 blk_queue_max_discard_sectors(rq, get_capacity(gd));
939                 rq->limits.discard_granularity = info->discard_granularity;
940                 rq->limits.discard_alignment = info->discard_alignment;
941                 if (info->feature_secdiscard)
942                         queue_flag_set_unlocked(QUEUE_FLAG_SECERASE, rq);
943         }
944
945         /* Hard sector size and max sectors impersonate the equiv. hardware. */
946         blk_queue_logical_block_size(rq, info->sector_size);
947         blk_queue_physical_block_size(rq, info->physical_sector_size);
948         blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
949
950         /* Each segment in a request is up to an aligned page in size. */
951         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
952         blk_queue_max_segment_size(rq, PAGE_SIZE);
953
954         /* Ensure a merged request will fit in a single I/O ring slot. */
955         blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
956
957         /* Make sure buffer addresses are sector-aligned. */
958         blk_queue_dma_alignment(rq, 511);
959 }
960
961 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
962                                 unsigned int physical_sector_size)
963 {
964         struct request_queue *rq;
965         struct blkfront_info *info = gd->private_data;
966
967         memset(&info->tag_set, 0, sizeof(info->tag_set));
968         info->tag_set.ops = &blkfront_mq_ops;
969         info->tag_set.nr_hw_queues = info->nr_rings;
970         if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
971                 /*
972                  * When indirect descriptior is not supported, the I/O request
973                  * will be split between multiple request in the ring.
974                  * To avoid problems when sending the request, divide by
975                  * 2 the depth of the queue.
976                  */
977                 info->tag_set.queue_depth =  BLK_RING_SIZE(info) / 2;
978         } else
979                 info->tag_set.queue_depth = BLK_RING_SIZE(info);
980         info->tag_set.numa_node = NUMA_NO_NODE;
981         info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
982         info->tag_set.cmd_size = sizeof(struct blkif_req);
983         info->tag_set.driver_data = info;
984
985         if (blk_mq_alloc_tag_set(&info->tag_set))
986                 return -EINVAL;
987         rq = blk_mq_init_queue(&info->tag_set);
988         if (IS_ERR(rq)) {
989                 blk_mq_free_tag_set(&info->tag_set);
990                 return PTR_ERR(rq);
991         }
992
993         rq->queuedata = info;
994         info->rq = gd->queue = rq;
995         info->gd = gd;
996         info->sector_size = sector_size;
997         info->physical_sector_size = physical_sector_size;
998         blkif_set_queue_limits(info);
999
1000         return 0;
1001 }
1002
1003 static const char *flush_info(struct blkfront_info *info)
1004 {
1005         if (info->feature_flush && info->feature_fua)
1006                 return "barrier: enabled;";
1007         else if (info->feature_flush)
1008                 return "flush diskcache: enabled;";
1009         else
1010                 return "barrier or flush: disabled;";
1011 }
1012
1013 static void xlvbd_flush(struct blkfront_info *info)
1014 {
1015         blk_queue_write_cache(info->rq, info->feature_flush ? true : false,
1016                               info->feature_fua ? true : false);
1017         pr_info("blkfront: %s: %s %s %s %s %s\n",
1018                 info->gd->disk_name, flush_info(info),
1019                 "persistent grants:", info->feature_persistent ?
1020                 "enabled;" : "disabled;", "indirect descriptors:",
1021                 info->max_indirect_segments ? "enabled;" : "disabled;");
1022 }
1023
1024 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
1025 {
1026         int major;
1027         major = BLKIF_MAJOR(vdevice);
1028         *minor = BLKIF_MINOR(vdevice);
1029         switch (major) {
1030                 case XEN_IDE0_MAJOR:
1031                         *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
1032                         *minor = ((*minor / 64) * PARTS_PER_DISK) +
1033                                 EMULATED_HD_DISK_MINOR_OFFSET;
1034                         break;
1035                 case XEN_IDE1_MAJOR:
1036                         *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1037                         *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1038                                 EMULATED_HD_DISK_MINOR_OFFSET;
1039                         break;
1040                 case XEN_SCSI_DISK0_MAJOR:
1041                         *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1042                         *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1043                         break;
1044                 case XEN_SCSI_DISK1_MAJOR:
1045                 case XEN_SCSI_DISK2_MAJOR:
1046                 case XEN_SCSI_DISK3_MAJOR:
1047                 case XEN_SCSI_DISK4_MAJOR:
1048                 case XEN_SCSI_DISK5_MAJOR:
1049                 case XEN_SCSI_DISK6_MAJOR:
1050                 case XEN_SCSI_DISK7_MAJOR:
1051                         *offset = (*minor / PARTS_PER_DISK) + 
1052                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1053                                 EMULATED_SD_DISK_NAME_OFFSET;
1054                         *minor = *minor +
1055                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1056                                 EMULATED_SD_DISK_MINOR_OFFSET;
1057                         break;
1058                 case XEN_SCSI_DISK8_MAJOR:
1059                 case XEN_SCSI_DISK9_MAJOR:
1060                 case XEN_SCSI_DISK10_MAJOR:
1061                 case XEN_SCSI_DISK11_MAJOR:
1062                 case XEN_SCSI_DISK12_MAJOR:
1063                 case XEN_SCSI_DISK13_MAJOR:
1064                 case XEN_SCSI_DISK14_MAJOR:
1065                 case XEN_SCSI_DISK15_MAJOR:
1066                         *offset = (*minor / PARTS_PER_DISK) + 
1067                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1068                                 EMULATED_SD_DISK_NAME_OFFSET;
1069                         *minor = *minor +
1070                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1071                                 EMULATED_SD_DISK_MINOR_OFFSET;
1072                         break;
1073                 case XENVBD_MAJOR:
1074                         *offset = *minor / PARTS_PER_DISK;
1075                         break;
1076                 default:
1077                         printk(KERN_WARNING "blkfront: your disk configuration is "
1078                                         "incorrect, please use an xvd device instead\n");
1079                         return -ENODEV;
1080         }
1081         return 0;
1082 }
1083
1084 static char *encode_disk_name(char *ptr, unsigned int n)
1085 {
1086         if (n >= 26)
1087                 ptr = encode_disk_name(ptr, n / 26 - 1);
1088         *ptr = 'a' + n % 26;
1089         return ptr + 1;
1090 }
1091
1092 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1093                                struct blkfront_info *info,
1094                                u16 vdisk_info, u16 sector_size,
1095                                unsigned int physical_sector_size)
1096 {
1097         struct gendisk *gd;
1098         int nr_minors = 1;
1099         int err;
1100         unsigned int offset;
1101         int minor;
1102         int nr_parts;
1103         char *ptr;
1104
1105         BUG_ON(info->gd != NULL);
1106         BUG_ON(info->rq != NULL);
1107
1108         if ((info->vdevice>>EXT_SHIFT) > 1) {
1109                 /* this is above the extended range; something is wrong */
1110                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1111                 return -ENODEV;
1112         }
1113
1114         if (!VDEV_IS_EXTENDED(info->vdevice)) {
1115                 err = xen_translate_vdev(info->vdevice, &minor, &offset);
1116                 if (err)
1117                         return err;             
1118                 nr_parts = PARTS_PER_DISK;
1119         } else {
1120                 minor = BLKIF_MINOR_EXT(info->vdevice);
1121                 nr_parts = PARTS_PER_EXT_DISK;
1122                 offset = minor / nr_parts;
1123                 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1124                         printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1125                                         "emulated IDE disks,\n\t choose an xvd device name"
1126                                         "from xvde on\n", info->vdevice);
1127         }
1128         if (minor >> MINORBITS) {
1129                 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1130                         info->vdevice, minor);
1131                 return -ENODEV;
1132         }
1133
1134         if ((minor % nr_parts) == 0)
1135                 nr_minors = nr_parts;
1136
1137         err = xlbd_reserve_minors(minor, nr_minors);
1138         if (err)
1139                 goto out;
1140         err = -ENODEV;
1141
1142         gd = alloc_disk(nr_minors);
1143         if (gd == NULL)
1144                 goto release;
1145
1146         strcpy(gd->disk_name, DEV_NAME);
1147         ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1148         BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1149         if (nr_minors > 1)
1150                 *ptr = 0;
1151         else
1152                 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1153                          "%d", minor & (nr_parts - 1));
1154
1155         gd->major = XENVBD_MAJOR;
1156         gd->first_minor = minor;
1157         gd->fops = &xlvbd_block_fops;
1158         gd->private_data = info;
1159         set_capacity(gd, capacity);
1160
1161         if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size)) {
1162                 del_gendisk(gd);
1163                 goto release;
1164         }
1165
1166         xlvbd_flush(info);
1167
1168         if (vdisk_info & VDISK_READONLY)
1169                 set_disk_ro(gd, 1);
1170
1171         if (vdisk_info & VDISK_REMOVABLE)
1172                 gd->flags |= GENHD_FL_REMOVABLE;
1173
1174         if (vdisk_info & VDISK_CDROM)
1175                 gd->flags |= GENHD_FL_CD;
1176
1177         return 0;
1178
1179  release:
1180         xlbd_release_minors(minor, nr_minors);
1181  out:
1182         return err;
1183 }
1184
1185 static void xlvbd_release_gendisk(struct blkfront_info *info)
1186 {
1187         unsigned int minor, nr_minors, i;
1188
1189         if (info->rq == NULL)
1190                 return;
1191
1192         /* No more blkif_request(). */
1193         blk_mq_stop_hw_queues(info->rq);
1194
1195         for (i = 0; i < info->nr_rings; i++) {
1196                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1197
1198                 /* No more gnttab callback work. */
1199                 gnttab_cancel_free_callback(&rinfo->callback);
1200
1201                 /* Flush gnttab callback work. Must be done with no locks held. */
1202                 flush_work(&rinfo->work);
1203         }
1204
1205         del_gendisk(info->gd);
1206
1207         minor = info->gd->first_minor;
1208         nr_minors = info->gd->minors;
1209         xlbd_release_minors(minor, nr_minors);
1210
1211         blk_cleanup_queue(info->rq);
1212         blk_mq_free_tag_set(&info->tag_set);
1213         info->rq = NULL;
1214
1215         put_disk(info->gd);
1216         info->gd = NULL;
1217 }
1218
1219 /* Already hold rinfo->ring_lock. */
1220 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1221 {
1222         if (!RING_FULL(&rinfo->ring))
1223                 blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1224 }
1225
1226 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1227 {
1228         unsigned long flags;
1229
1230         spin_lock_irqsave(&rinfo->ring_lock, flags);
1231         kick_pending_request_queues_locked(rinfo);
1232         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1233 }
1234
1235 static void blkif_restart_queue(struct work_struct *work)
1236 {
1237         struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1238
1239         if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1240                 kick_pending_request_queues(rinfo);
1241 }
1242
1243 static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1244 {
1245         struct grant *persistent_gnt, *n;
1246         struct blkfront_info *info = rinfo->dev_info;
1247         int i, j, segs;
1248
1249         /*
1250          * Remove indirect pages, this only happens when using indirect
1251          * descriptors but not persistent grants
1252          */
1253         if (!list_empty(&rinfo->indirect_pages)) {
1254                 struct page *indirect_page, *n;
1255
1256                 BUG_ON(info->feature_persistent);
1257                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1258                         list_del(&indirect_page->lru);
1259                         __free_page(indirect_page);
1260                 }
1261         }
1262
1263         /* Remove all persistent grants. */
1264         if (!list_empty(&rinfo->grants)) {
1265                 list_for_each_entry_safe(persistent_gnt, n,
1266                                          &rinfo->grants, node) {
1267                         list_del(&persistent_gnt->node);
1268                         if (persistent_gnt->gref != GRANT_INVALID_REF) {
1269                                 gnttab_end_foreign_access(persistent_gnt->gref,
1270                                                           0, 0UL);
1271                                 rinfo->persistent_gnts_c--;
1272                         }
1273                         if (info->feature_persistent)
1274                                 __free_page(persistent_gnt->page);
1275                         kfree(persistent_gnt);
1276                 }
1277         }
1278         BUG_ON(rinfo->persistent_gnts_c != 0);
1279
1280         for (i = 0; i < BLK_RING_SIZE(info); i++) {
1281                 /*
1282                  * Clear persistent grants present in requests already
1283                  * on the shared ring
1284                  */
1285                 if (!rinfo->shadow[i].request)
1286                         goto free_shadow;
1287
1288                 segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1289                        rinfo->shadow[i].req.u.indirect.nr_segments :
1290                        rinfo->shadow[i].req.u.rw.nr_segments;
1291                 for (j = 0; j < segs; j++) {
1292                         persistent_gnt = rinfo->shadow[i].grants_used[j];
1293                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1294                         if (info->feature_persistent)
1295                                 __free_page(persistent_gnt->page);
1296                         kfree(persistent_gnt);
1297                 }
1298
1299                 if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1300                         /*
1301                          * If this is not an indirect operation don't try to
1302                          * free indirect segments
1303                          */
1304                         goto free_shadow;
1305
1306                 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1307                         persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1308                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1309                         __free_page(persistent_gnt->page);
1310                         kfree(persistent_gnt);
1311                 }
1312
1313 free_shadow:
1314                 kfree(rinfo->shadow[i].grants_used);
1315                 rinfo->shadow[i].grants_used = NULL;
1316                 kfree(rinfo->shadow[i].indirect_grants);
1317                 rinfo->shadow[i].indirect_grants = NULL;
1318                 kfree(rinfo->shadow[i].sg);
1319                 rinfo->shadow[i].sg = NULL;
1320         }
1321
1322         /* No more gnttab callback work. */
1323         gnttab_cancel_free_callback(&rinfo->callback);
1324
1325         /* Flush gnttab callback work. Must be done with no locks held. */
1326         flush_work(&rinfo->work);
1327
1328         /* Free resources associated with old device channel. */
1329         for (i = 0; i < info->nr_ring_pages; i++) {
1330                 if (rinfo->ring_ref[i] != GRANT_INVALID_REF) {
1331                         gnttab_end_foreign_access(rinfo->ring_ref[i], 0, 0);
1332                         rinfo->ring_ref[i] = GRANT_INVALID_REF;
1333                 }
1334         }
1335         free_pages((unsigned long)rinfo->ring.sring, get_order(info->nr_ring_pages * XEN_PAGE_SIZE));
1336         rinfo->ring.sring = NULL;
1337
1338         if (rinfo->irq)
1339                 unbind_from_irqhandler(rinfo->irq, rinfo);
1340         rinfo->evtchn = rinfo->irq = 0;
1341 }
1342
1343 static void blkif_free(struct blkfront_info *info, int suspend)
1344 {
1345         unsigned int i;
1346
1347         /* Prevent new requests being issued until we fix things up. */
1348         info->connected = suspend ?
1349                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1350         /* No more blkif_request(). */
1351         if (info->rq)
1352                 blk_mq_stop_hw_queues(info->rq);
1353
1354         for (i = 0; i < info->nr_rings; i++)
1355                 blkif_free_ring(&info->rinfo[i]);
1356
1357         kfree(info->rinfo);
1358         info->rinfo = NULL;
1359         info->nr_rings = 0;
1360 }
1361
1362 struct copy_from_grant {
1363         const struct blk_shadow *s;
1364         unsigned int grant_idx;
1365         unsigned int bvec_offset;
1366         char *bvec_data;
1367 };
1368
1369 static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1370                                   unsigned int len, void *data)
1371 {
1372         struct copy_from_grant *info = data;
1373         char *shared_data;
1374         /* Convenient aliases */
1375         const struct blk_shadow *s = info->s;
1376
1377         shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1378
1379         memcpy(info->bvec_data + info->bvec_offset,
1380                shared_data + offset, len);
1381
1382         info->bvec_offset += len;
1383         info->grant_idx++;
1384
1385         kunmap_atomic(shared_data);
1386 }
1387
1388 static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1389 {
1390         switch (rsp)
1391         {
1392         case BLKIF_RSP_OKAY:
1393                 return REQ_DONE;
1394         case BLKIF_RSP_EOPNOTSUPP:
1395                 return REQ_EOPNOTSUPP;
1396         case BLKIF_RSP_ERROR:
1397                 /* Fallthrough. */
1398         default:
1399                 return REQ_ERROR;
1400         }
1401 }
1402
1403 /*
1404  * Get the final status of the block request based on two ring response
1405  */
1406 static int blkif_get_final_status(enum blk_req_status s1,
1407                                   enum blk_req_status s2)
1408 {
1409         BUG_ON(s1 == REQ_WAITING);
1410         BUG_ON(s2 == REQ_WAITING);
1411
1412         if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1413                 return BLKIF_RSP_ERROR;
1414         else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1415                 return BLKIF_RSP_EOPNOTSUPP;
1416         return BLKIF_RSP_OKAY;
1417 }
1418
1419 static bool blkif_completion(unsigned long *id,
1420                              struct blkfront_ring_info *rinfo,
1421                              struct blkif_response *bret)
1422 {
1423         int i = 0;
1424         struct scatterlist *sg;
1425         int num_sg, num_grant;
1426         struct blkfront_info *info = rinfo->dev_info;
1427         struct blk_shadow *s = &rinfo->shadow[*id];
1428         struct copy_from_grant data = {
1429                 .grant_idx = 0,
1430         };
1431
1432         num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1433                 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1434
1435         /* The I/O request may be split in two. */
1436         if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1437                 struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1438
1439                 /* Keep the status of the current response in shadow. */
1440                 s->status = blkif_rsp_to_req_status(bret->status);
1441
1442                 /* Wait the second response if not yet here. */
1443                 if (s2->status == REQ_WAITING)
1444                         return 0;
1445
1446                 bret->status = blkif_get_final_status(s->status,
1447                                                       s2->status);
1448
1449                 /*
1450                  * All the grants is stored in the first shadow in order
1451                  * to make the completion code simpler.
1452                  */
1453                 num_grant += s2->req.u.rw.nr_segments;
1454
1455                 /*
1456                  * The two responses may not come in order. Only the
1457                  * first request will store the scatter-gather list.
1458                  */
1459                 if (s2->num_sg != 0) {
1460                         /* Update "id" with the ID of the first response. */
1461                         *id = s->associated_id;
1462                         s = s2;
1463                 }
1464
1465                 /*
1466                  * We don't need anymore the second request, so recycling
1467                  * it now.
1468                  */
1469                 if (add_id_to_freelist(rinfo, s->associated_id))
1470                         WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1471                              info->gd->disk_name, s->associated_id);
1472         }
1473
1474         data.s = s;
1475         num_sg = s->num_sg;
1476
1477         if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1478                 for_each_sg(s->sg, sg, num_sg, i) {
1479                         BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1480
1481                         data.bvec_offset = sg->offset;
1482                         data.bvec_data = kmap_atomic(sg_page(sg));
1483
1484                         gnttab_foreach_grant_in_range(sg_page(sg),
1485                                                       sg->offset,
1486                                                       sg->length,
1487                                                       blkif_copy_from_grant,
1488                                                       &data);
1489
1490                         kunmap_atomic(data.bvec_data);
1491                 }
1492         }
1493         /* Add the persistent grant into the list of free grants */
1494         for (i = 0; i < num_grant; i++) {
1495                 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1496                         /*
1497                          * If the grant is still mapped by the backend (the
1498                          * backend has chosen to make this grant persistent)
1499                          * we add it at the head of the list, so it will be
1500                          * reused first.
1501                          */
1502                         if (!info->feature_persistent)
1503                                 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1504                                                      s->grants_used[i]->gref);
1505                         list_add(&s->grants_used[i]->node, &rinfo->grants);
1506                         rinfo->persistent_gnts_c++;
1507                 } else {
1508                         /*
1509                          * If the grant is not mapped by the backend we end the
1510                          * foreign access and add it to the tail of the list,
1511                          * so it will not be picked again unless we run out of
1512                          * persistent grants.
1513                          */
1514                         gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1515                         s->grants_used[i]->gref = GRANT_INVALID_REF;
1516                         list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1517                 }
1518         }
1519         if (s->req.operation == BLKIF_OP_INDIRECT) {
1520                 for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1521                         if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1522                                 if (!info->feature_persistent)
1523                                         pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1524                                                              s->indirect_grants[i]->gref);
1525                                 list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1526                                 rinfo->persistent_gnts_c++;
1527                         } else {
1528                                 struct page *indirect_page;
1529
1530                                 gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1531                                 /*
1532                                  * Add the used indirect page back to the list of
1533                                  * available pages for indirect grefs.
1534                                  */
1535                                 if (!info->feature_persistent) {
1536                                         indirect_page = s->indirect_grants[i]->page;
1537                                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
1538                                 }
1539                                 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1540                                 list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1541                         }
1542                 }
1543         }
1544
1545         return 1;
1546 }
1547
1548 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1549 {
1550         struct request *req;
1551         struct blkif_response *bret;
1552         RING_IDX i, rp;
1553         unsigned long flags;
1554         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1555         struct blkfront_info *info = rinfo->dev_info;
1556
1557         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
1558                 return IRQ_HANDLED;
1559
1560         spin_lock_irqsave(&rinfo->ring_lock, flags);
1561  again:
1562         rp = rinfo->ring.sring->rsp_prod;
1563         rmb(); /* Ensure we see queued responses up to 'rp'. */
1564
1565         for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1566                 unsigned long id;
1567
1568                 bret = RING_GET_RESPONSE(&rinfo->ring, i);
1569                 id   = bret->id;
1570                 /*
1571                  * The backend has messed up and given us an id that we would
1572                  * never have given to it (we stamp it up to BLK_RING_SIZE -
1573                  * look in get_id_from_freelist.
1574                  */
1575                 if (id >= BLK_RING_SIZE(info)) {
1576                         WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1577                              info->gd->disk_name, op_name(bret->operation), id);
1578                         /* We can't safely get the 'struct request' as
1579                          * the id is busted. */
1580                         continue;
1581                 }
1582                 req  = rinfo->shadow[id].request;
1583
1584                 if (bret->operation != BLKIF_OP_DISCARD) {
1585                         /*
1586                          * We may need to wait for an extra response if the
1587                          * I/O request is split in 2
1588                          */
1589                         if (!blkif_completion(&id, rinfo, bret))
1590                                 continue;
1591                 }
1592
1593                 if (add_id_to_freelist(rinfo, id)) {
1594                         WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1595                              info->gd->disk_name, op_name(bret->operation), id);
1596                         continue;
1597                 }
1598
1599                 if (bret->status == BLKIF_RSP_OKAY)
1600                         blkif_req(req)->error = BLK_STS_OK;
1601                 else
1602                         blkif_req(req)->error = BLK_STS_IOERR;
1603
1604                 switch (bret->operation) {
1605                 case BLKIF_OP_DISCARD:
1606                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1607                                 struct request_queue *rq = info->rq;
1608                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1609                                            info->gd->disk_name, op_name(bret->operation));
1610                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1611                                 info->feature_discard = 0;
1612                                 info->feature_secdiscard = 0;
1613                                 queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1614                                 queue_flag_clear(QUEUE_FLAG_SECERASE, rq);
1615                         }
1616                         break;
1617                 case BLKIF_OP_FLUSH_DISKCACHE:
1618                 case BLKIF_OP_WRITE_BARRIER:
1619                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1620                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1621                                        info->gd->disk_name, op_name(bret->operation));
1622                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1623                         }
1624                         if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1625                                      rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1626                                 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1627                                        info->gd->disk_name, op_name(bret->operation));
1628                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1629                         }
1630                         if (unlikely(blkif_req(req)->error)) {
1631                                 if (blkif_req(req)->error == BLK_STS_NOTSUPP)
1632                                         blkif_req(req)->error = BLK_STS_OK;
1633                                 info->feature_fua = 0;
1634                                 info->feature_flush = 0;
1635                                 xlvbd_flush(info);
1636                         }
1637                         /* fall through */
1638                 case BLKIF_OP_READ:
1639                 case BLKIF_OP_WRITE:
1640                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
1641                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1642                                         "request: %x\n", bret->status);
1643
1644                         break;
1645                 default:
1646                         BUG();
1647                 }
1648
1649                 blk_mq_complete_request(req);
1650         }
1651
1652         rinfo->ring.rsp_cons = i;
1653
1654         if (i != rinfo->ring.req_prod_pvt) {
1655                 int more_to_do;
1656                 RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1657                 if (more_to_do)
1658                         goto again;
1659         } else
1660                 rinfo->ring.sring->rsp_event = i + 1;
1661
1662         kick_pending_request_queues_locked(rinfo);
1663
1664         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1665
1666         return IRQ_HANDLED;
1667 }
1668
1669
1670 static int setup_blkring(struct xenbus_device *dev,
1671                          struct blkfront_ring_info *rinfo)
1672 {
1673         struct blkif_sring *sring;
1674         int err, i;
1675         struct blkfront_info *info = rinfo->dev_info;
1676         unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1677         grant_ref_t gref[XENBUS_MAX_RING_GRANTS];
1678
1679         for (i = 0; i < info->nr_ring_pages; i++)
1680                 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1681
1682         sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
1683                                                        get_order(ring_size));
1684         if (!sring) {
1685                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1686                 return -ENOMEM;
1687         }
1688         SHARED_RING_INIT(sring);
1689         FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1690
1691         err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref);
1692         if (err < 0) {
1693                 free_pages((unsigned long)sring, get_order(ring_size));
1694                 rinfo->ring.sring = NULL;
1695                 goto fail;
1696         }
1697         for (i = 0; i < info->nr_ring_pages; i++)
1698                 rinfo->ring_ref[i] = gref[i];
1699
1700         err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1701         if (err)
1702                 goto fail;
1703
1704         err = bind_evtchn_to_irqhandler(rinfo->evtchn, blkif_interrupt, 0,
1705                                         "blkif", rinfo);
1706         if (err <= 0) {
1707                 xenbus_dev_fatal(dev, err,
1708                                  "bind_evtchn_to_irqhandler failed");
1709                 goto fail;
1710         }
1711         rinfo->irq = err;
1712
1713         return 0;
1714 fail:
1715         blkif_free(info, 0);
1716         return err;
1717 }
1718
1719 /*
1720  * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1721  * ring buffer may have multi pages depending on ->nr_ring_pages.
1722  */
1723 static int write_per_ring_nodes(struct xenbus_transaction xbt,
1724                                 struct blkfront_ring_info *rinfo, const char *dir)
1725 {
1726         int err;
1727         unsigned int i;
1728         const char *message = NULL;
1729         struct blkfront_info *info = rinfo->dev_info;
1730
1731         if (info->nr_ring_pages == 1) {
1732                 err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1733                 if (err) {
1734                         message = "writing ring-ref";
1735                         goto abort_transaction;
1736                 }
1737         } else {
1738                 for (i = 0; i < info->nr_ring_pages; i++) {
1739                         char ring_ref_name[RINGREF_NAME_LEN];
1740
1741                         snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1742                         err = xenbus_printf(xbt, dir, ring_ref_name,
1743                                             "%u", rinfo->ring_ref[i]);
1744                         if (err) {
1745                                 message = "writing ring-ref";
1746                                 goto abort_transaction;
1747                         }
1748                 }
1749         }
1750
1751         err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1752         if (err) {
1753                 message = "writing event-channel";
1754                 goto abort_transaction;
1755         }
1756
1757         return 0;
1758
1759 abort_transaction:
1760         xenbus_transaction_end(xbt, 1);
1761         if (message)
1762                 xenbus_dev_fatal(info->xbdev, err, "%s", message);
1763
1764         return err;
1765 }
1766
1767 /* Common code used when first setting up, and when resuming. */
1768 static int talk_to_blkback(struct xenbus_device *dev,
1769                            struct blkfront_info *info)
1770 {
1771         const char *message = NULL;
1772         struct xenbus_transaction xbt;
1773         int err;
1774         unsigned int i, max_page_order;
1775         unsigned int ring_page_order;
1776
1777         max_page_order = xenbus_read_unsigned(info->xbdev->otherend,
1778                                               "max-ring-page-order", 0);
1779         ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1780         info->nr_ring_pages = 1 << ring_page_order;
1781
1782         for (i = 0; i < info->nr_rings; i++) {
1783                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1784
1785                 /* Create shared ring, alloc event channel. */
1786                 err = setup_blkring(dev, rinfo);
1787                 if (err)
1788                         goto destroy_blkring;
1789         }
1790
1791 again:
1792         err = xenbus_transaction_start(&xbt);
1793         if (err) {
1794                 xenbus_dev_fatal(dev, err, "starting transaction");
1795                 goto destroy_blkring;
1796         }
1797
1798         if (info->nr_ring_pages > 1) {
1799                 err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1800                                     ring_page_order);
1801                 if (err) {
1802                         message = "writing ring-page-order";
1803                         goto abort_transaction;
1804                 }
1805         }
1806
1807         /* We already got the number of queues/rings in _probe */
1808         if (info->nr_rings == 1) {
1809                 err = write_per_ring_nodes(xbt, &info->rinfo[0], dev->nodename);
1810                 if (err)
1811                         goto destroy_blkring;
1812         } else {
1813                 char *path;
1814                 size_t pathsize;
1815
1816                 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1817                                     info->nr_rings);
1818                 if (err) {
1819                         message = "writing multi-queue-num-queues";
1820                         goto abort_transaction;
1821                 }
1822
1823                 pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1824                 path = kmalloc(pathsize, GFP_KERNEL);
1825                 if (!path) {
1826                         err = -ENOMEM;
1827                         message = "ENOMEM while writing ring references";
1828                         goto abort_transaction;
1829                 }
1830
1831                 for (i = 0; i < info->nr_rings; i++) {
1832                         memset(path, 0, pathsize);
1833                         snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1834                         err = write_per_ring_nodes(xbt, &info->rinfo[i], path);
1835                         if (err) {
1836                                 kfree(path);
1837                                 goto destroy_blkring;
1838                         }
1839                 }
1840                 kfree(path);
1841         }
1842         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1843                             XEN_IO_PROTO_ABI_NATIVE);
1844         if (err) {
1845                 message = "writing protocol";
1846                 goto abort_transaction;
1847         }
1848         err = xenbus_printf(xbt, dev->nodename,
1849                             "feature-persistent", "%u", 1);
1850         if (err)
1851                 dev_warn(&dev->dev,
1852                          "writing persistent grants feature to xenbus");
1853
1854         err = xenbus_transaction_end(xbt, 0);
1855         if (err) {
1856                 if (err == -EAGAIN)
1857                         goto again;
1858                 xenbus_dev_fatal(dev, err, "completing transaction");
1859                 goto destroy_blkring;
1860         }
1861
1862         for (i = 0; i < info->nr_rings; i++) {
1863                 unsigned int j;
1864                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1865
1866                 for (j = 0; j < BLK_RING_SIZE(info); j++)
1867                         rinfo->shadow[j].req.u.rw.id = j + 1;
1868                 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1869         }
1870         xenbus_switch_state(dev, XenbusStateInitialised);
1871
1872         return 0;
1873
1874  abort_transaction:
1875         xenbus_transaction_end(xbt, 1);
1876         if (message)
1877                 xenbus_dev_fatal(dev, err, "%s", message);
1878  destroy_blkring:
1879         blkif_free(info, 0);
1880
1881         kfree(info);
1882         dev_set_drvdata(&dev->dev, NULL);
1883
1884         return err;
1885 }
1886
1887 static int negotiate_mq(struct blkfront_info *info)
1888 {
1889         unsigned int backend_max_queues;
1890         unsigned int i;
1891
1892         BUG_ON(info->nr_rings);
1893
1894         /* Check if backend supports multiple queues. */
1895         backend_max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1896                                                   "multi-queue-max-queues", 1);
1897         info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1898         /* We need at least one ring. */
1899         if (!info->nr_rings)
1900                 info->nr_rings = 1;
1901
1902         info->rinfo = kzalloc(sizeof(struct blkfront_ring_info) * info->nr_rings, GFP_KERNEL);
1903         if (!info->rinfo) {
1904                 xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1905                 return -ENOMEM;
1906         }
1907
1908         for (i = 0; i < info->nr_rings; i++) {
1909                 struct blkfront_ring_info *rinfo;
1910
1911                 rinfo = &info->rinfo[i];
1912                 INIT_LIST_HEAD(&rinfo->indirect_pages);
1913                 INIT_LIST_HEAD(&rinfo->grants);
1914                 rinfo->dev_info = info;
1915                 INIT_WORK(&rinfo->work, blkif_restart_queue);
1916                 spin_lock_init(&rinfo->ring_lock);
1917         }
1918         return 0;
1919 }
1920 /**
1921  * Entry point to this code when a new device is created.  Allocate the basic
1922  * structures and the ring buffer for communication with the backend, and
1923  * inform the backend of the appropriate details for those.  Switch to
1924  * Initialised state.
1925  */
1926 static int blkfront_probe(struct xenbus_device *dev,
1927                           const struct xenbus_device_id *id)
1928 {
1929         int err, vdevice;
1930         struct blkfront_info *info;
1931
1932         /* FIXME: Use dynamic device id if this is not set. */
1933         err = xenbus_scanf(XBT_NIL, dev->nodename,
1934                            "virtual-device", "%i", &vdevice);
1935         if (err != 1) {
1936                 /* go looking in the extended area instead */
1937                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1938                                    "%i", &vdevice);
1939                 if (err != 1) {
1940                         xenbus_dev_fatal(dev, err, "reading virtual-device");
1941                         return err;
1942                 }
1943         }
1944
1945         if (xen_hvm_domain()) {
1946                 char *type;
1947                 int len;
1948                 /* no unplug has been done: do not hook devices != xen vbds */
1949                 if (xen_has_pv_and_legacy_disk_devices()) {
1950                         int major;
1951
1952                         if (!VDEV_IS_EXTENDED(vdevice))
1953                                 major = BLKIF_MAJOR(vdevice);
1954                         else
1955                                 major = XENVBD_MAJOR;
1956
1957                         if (major != XENVBD_MAJOR) {
1958                                 printk(KERN_INFO
1959                                                 "%s: HVM does not support vbd %d as xen block device\n",
1960                                                 __func__, vdevice);
1961                                 return -ENODEV;
1962                         }
1963                 }
1964                 /* do not create a PV cdrom device if we are an HVM guest */
1965                 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1966                 if (IS_ERR(type))
1967                         return -ENODEV;
1968                 if (strncmp(type, "cdrom", 5) == 0) {
1969                         kfree(type);
1970                         return -ENODEV;
1971                 }
1972                 kfree(type);
1973         }
1974         info = kzalloc(sizeof(*info), GFP_KERNEL);
1975         if (!info) {
1976                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1977                 return -ENOMEM;
1978         }
1979
1980         info->xbdev = dev;
1981         err = negotiate_mq(info);
1982         if (err) {
1983                 kfree(info);
1984                 return err;
1985         }
1986
1987         mutex_init(&info->mutex);
1988         info->vdevice = vdevice;
1989         info->connected = BLKIF_STATE_DISCONNECTED;
1990
1991         /* Front end dir is a number, which is used as the id. */
1992         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1993         dev_set_drvdata(&dev->dev, info);
1994
1995         return 0;
1996 }
1997
1998 static int blkif_recover(struct blkfront_info *info)
1999 {
2000         unsigned int r_index;
2001         struct request *req, *n;
2002         int rc;
2003         struct bio *bio;
2004         unsigned int segs;
2005
2006         blkfront_gather_backend_features(info);
2007         /* Reset limits changed by blk_mq_update_nr_hw_queues(). */
2008         blkif_set_queue_limits(info);
2009         segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
2010         blk_queue_max_segments(info->rq, segs / GRANTS_PER_PSEG);
2011
2012         for (r_index = 0; r_index < info->nr_rings; r_index++) {
2013                 struct blkfront_ring_info *rinfo = &info->rinfo[r_index];
2014
2015                 rc = blkfront_setup_indirect(rinfo);
2016                 if (rc)
2017                         return rc;
2018         }
2019         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2020
2021         /* Now safe for us to use the shared ring */
2022         info->connected = BLKIF_STATE_CONNECTED;
2023
2024         for (r_index = 0; r_index < info->nr_rings; r_index++) {
2025                 struct blkfront_ring_info *rinfo;
2026
2027                 rinfo = &info->rinfo[r_index];
2028                 /* Kick any other new requests queued since we resumed */
2029                 kick_pending_request_queues(rinfo);
2030         }
2031
2032         list_for_each_entry_safe(req, n, &info->requests, queuelist) {
2033                 /* Requeue pending requests (flush or discard) */
2034                 list_del_init(&req->queuelist);
2035                 BUG_ON(req->nr_phys_segments > segs);
2036                 blk_mq_requeue_request(req, false);
2037         }
2038         blk_mq_start_stopped_hw_queues(info->rq, true);
2039         blk_mq_kick_requeue_list(info->rq);
2040
2041         while ((bio = bio_list_pop(&info->bio_list)) != NULL) {
2042                 /* Traverse the list of pending bios and re-queue them */
2043                 submit_bio(bio);
2044         }
2045
2046         return 0;
2047 }
2048
2049 /**
2050  * We are reconnecting to the backend, due to a suspend/resume, or a backend
2051  * driver restart.  We tear down our blkif structure and recreate it, but
2052  * leave the device-layer structures intact so that this is transparent to the
2053  * rest of the kernel.
2054  */
2055 static int blkfront_resume(struct xenbus_device *dev)
2056 {
2057         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2058         int err = 0;
2059         unsigned int i, j;
2060
2061         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2062
2063         bio_list_init(&info->bio_list);
2064         INIT_LIST_HEAD(&info->requests);
2065         for (i = 0; i < info->nr_rings; i++) {
2066                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
2067                 struct bio_list merge_bio;
2068                 struct blk_shadow *shadow = rinfo->shadow;
2069
2070                 for (j = 0; j < BLK_RING_SIZE(info); j++) {
2071                         /* Not in use? */
2072                         if (!shadow[j].request)
2073                                 continue;
2074
2075                         /*
2076                          * Get the bios in the request so we can re-queue them.
2077                          */
2078                         if (req_op(shadow[j].request) == REQ_OP_FLUSH ||
2079                             req_op(shadow[j].request) == REQ_OP_DISCARD ||
2080                             req_op(shadow[j].request) == REQ_OP_SECURE_ERASE ||
2081                             shadow[j].request->cmd_flags & REQ_FUA) {
2082                                 /*
2083                                  * Flush operations don't contain bios, so
2084                                  * we need to requeue the whole request
2085                                  *
2086                                  * XXX: but this doesn't make any sense for a
2087                                  * write with the FUA flag set..
2088                                  */
2089                                 list_add(&shadow[j].request->queuelist, &info->requests);
2090                                 continue;
2091                         }
2092                         merge_bio.head = shadow[j].request->bio;
2093                         merge_bio.tail = shadow[j].request->biotail;
2094                         bio_list_merge(&info->bio_list, &merge_bio);
2095                         shadow[j].request->bio = NULL;
2096                         blk_mq_end_request(shadow[j].request, BLK_STS_OK);
2097                 }
2098         }
2099
2100         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2101
2102         err = negotiate_mq(info);
2103         if (err)
2104                 return err;
2105
2106         err = talk_to_blkback(dev, info);
2107         if (!err)
2108                 blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings);
2109
2110         /*
2111          * We have to wait for the backend to switch to
2112          * connected state, since we want to read which
2113          * features it supports.
2114          */
2115
2116         return err;
2117 }
2118
2119 static void blkfront_closing(struct blkfront_info *info)
2120 {
2121         struct xenbus_device *xbdev = info->xbdev;
2122         struct block_device *bdev = NULL;
2123
2124         mutex_lock(&info->mutex);
2125
2126         if (xbdev->state == XenbusStateClosing) {
2127                 mutex_unlock(&info->mutex);
2128                 return;
2129         }
2130
2131         if (info->gd)
2132                 bdev = bdget_disk(info->gd, 0);
2133
2134         mutex_unlock(&info->mutex);
2135
2136         if (!bdev) {
2137                 xenbus_frontend_closed(xbdev);
2138                 return;
2139         }
2140
2141         mutex_lock(&bdev->bd_mutex);
2142
2143         if (bdev->bd_openers) {
2144                 xenbus_dev_error(xbdev, -EBUSY,
2145                                  "Device in use; refusing to close");
2146                 xenbus_switch_state(xbdev, XenbusStateClosing);
2147         } else {
2148                 xlvbd_release_gendisk(info);
2149                 xenbus_frontend_closed(xbdev);
2150         }
2151
2152         mutex_unlock(&bdev->bd_mutex);
2153         bdput(bdev);
2154 }
2155
2156 static void blkfront_setup_discard(struct blkfront_info *info)
2157 {
2158         int err;
2159         unsigned int discard_granularity;
2160         unsigned int discard_alignment;
2161
2162         info->feature_discard = 1;
2163         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2164                 "discard-granularity", "%u", &discard_granularity,
2165                 "discard-alignment", "%u", &discard_alignment,
2166                 NULL);
2167         if (!err) {
2168                 info->discard_granularity = discard_granularity;
2169                 info->discard_alignment = discard_alignment;
2170         }
2171         info->feature_secdiscard =
2172                 !!xenbus_read_unsigned(info->xbdev->otherend, "discard-secure",
2173                                        0);
2174 }
2175
2176 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2177 {
2178         unsigned int psegs, grants;
2179         int err, i;
2180         struct blkfront_info *info = rinfo->dev_info;
2181
2182         if (info->max_indirect_segments == 0) {
2183                 if (!HAS_EXTRA_REQ)
2184                         grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2185                 else {
2186                         /*
2187                          * When an extra req is required, the maximum
2188                          * grants supported is related to the size of the
2189                          * Linux block segment.
2190                          */
2191                         grants = GRANTS_PER_PSEG;
2192                 }
2193         }
2194         else
2195                 grants = info->max_indirect_segments;
2196         psegs = DIV_ROUND_UP(grants, GRANTS_PER_PSEG);
2197
2198         err = fill_grant_buffer(rinfo,
2199                                 (grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2200         if (err)
2201                 goto out_of_memory;
2202
2203         if (!info->feature_persistent && info->max_indirect_segments) {
2204                 /*
2205                  * We are using indirect descriptors but not persistent
2206                  * grants, we need to allocate a set of pages that can be
2207                  * used for mapping indirect grefs
2208                  */
2209                 int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2210
2211                 BUG_ON(!list_empty(&rinfo->indirect_pages));
2212                 for (i = 0; i < num; i++) {
2213                         struct page *indirect_page = alloc_page(GFP_NOIO);
2214                         if (!indirect_page)
2215                                 goto out_of_memory;
2216                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
2217                 }
2218         }
2219
2220         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2221                 rinfo->shadow[i].grants_used = kzalloc(
2222                         sizeof(rinfo->shadow[i].grants_used[0]) * grants,
2223                         GFP_NOIO);
2224                 rinfo->shadow[i].sg = kzalloc(sizeof(rinfo->shadow[i].sg[0]) * psegs, GFP_NOIO);
2225                 if (info->max_indirect_segments)
2226                         rinfo->shadow[i].indirect_grants = kzalloc(
2227                                 sizeof(rinfo->shadow[i].indirect_grants[0]) *
2228                                 INDIRECT_GREFS(grants),
2229                                 GFP_NOIO);
2230                 if ((rinfo->shadow[i].grants_used == NULL) ||
2231                         (rinfo->shadow[i].sg == NULL) ||
2232                      (info->max_indirect_segments &&
2233                      (rinfo->shadow[i].indirect_grants == NULL)))
2234                         goto out_of_memory;
2235                 sg_init_table(rinfo->shadow[i].sg, psegs);
2236         }
2237
2238
2239         return 0;
2240
2241 out_of_memory:
2242         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2243                 kfree(rinfo->shadow[i].grants_used);
2244                 rinfo->shadow[i].grants_used = NULL;
2245                 kfree(rinfo->shadow[i].sg);
2246                 rinfo->shadow[i].sg = NULL;
2247                 kfree(rinfo->shadow[i].indirect_grants);
2248                 rinfo->shadow[i].indirect_grants = NULL;
2249         }
2250         if (!list_empty(&rinfo->indirect_pages)) {
2251                 struct page *indirect_page, *n;
2252                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2253                         list_del(&indirect_page->lru);
2254                         __free_page(indirect_page);
2255                 }
2256         }
2257         return -ENOMEM;
2258 }
2259
2260 /*
2261  * Gather all backend feature-*
2262  */
2263 static void blkfront_gather_backend_features(struct blkfront_info *info)
2264 {
2265         unsigned int indirect_segments;
2266
2267         info->feature_flush = 0;
2268         info->feature_fua = 0;
2269
2270         /*
2271          * If there's no "feature-barrier" defined, then it means
2272          * we're dealing with a very old backend which writes
2273          * synchronously; nothing to do.
2274          *
2275          * If there are barriers, then we use flush.
2276          */
2277         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-barrier", 0)) {
2278                 info->feature_flush = 1;
2279                 info->feature_fua = 1;
2280         }
2281
2282         /*
2283          * And if there is "feature-flush-cache" use that above
2284          * barriers.
2285          */
2286         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-flush-cache",
2287                                  0)) {
2288                 info->feature_flush = 1;
2289                 info->feature_fua = 0;
2290         }
2291
2292         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-discard", 0))
2293                 blkfront_setup_discard(info);
2294
2295         info->feature_persistent =
2296                 !!xenbus_read_unsigned(info->xbdev->otherend,
2297                                        "feature-persistent", 0);
2298
2299         indirect_segments = xenbus_read_unsigned(info->xbdev->otherend,
2300                                         "feature-max-indirect-segments", 0);
2301         if (indirect_segments > xen_blkif_max_segments)
2302                 indirect_segments = xen_blkif_max_segments;
2303         if (indirect_segments <= BLKIF_MAX_SEGMENTS_PER_REQUEST)
2304                 indirect_segments = 0;
2305         info->max_indirect_segments = indirect_segments;
2306 }
2307
2308 /*
2309  * Invoked when the backend is finally 'ready' (and has told produced
2310  * the details about the physical device - #sectors, size, etc).
2311  */
2312 static void blkfront_connect(struct blkfront_info *info)
2313 {
2314         unsigned long long sectors;
2315         unsigned long sector_size;
2316         unsigned int physical_sector_size;
2317         unsigned int binfo;
2318         char *envp[] = { "RESIZE=1", NULL };
2319         int err, i;
2320
2321         switch (info->connected) {
2322         case BLKIF_STATE_CONNECTED:
2323                 /*
2324                  * Potentially, the back-end may be signalling
2325                  * a capacity change; update the capacity.
2326                  */
2327                 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2328                                    "sectors", "%Lu", &sectors);
2329                 if (XENBUS_EXIST_ERR(err))
2330                         return;
2331                 printk(KERN_INFO "Setting capacity to %Lu\n",
2332                        sectors);
2333                 set_capacity(info->gd, sectors);
2334                 revalidate_disk(info->gd);
2335                 kobject_uevent_env(&disk_to_dev(info->gd)->kobj,
2336                                    KOBJ_CHANGE, envp);
2337
2338                 return;
2339         case BLKIF_STATE_SUSPENDED:
2340                 /*
2341                  * If we are recovering from suspension, we need to wait
2342                  * for the backend to announce it's features before
2343                  * reconnecting, at least we need to know if the backend
2344                  * supports indirect descriptors, and how many.
2345                  */
2346                 blkif_recover(info);
2347                 return;
2348
2349         default:
2350                 break;
2351         }
2352
2353         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2354                 __func__, info->xbdev->otherend);
2355
2356         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2357                             "sectors", "%llu", &sectors,
2358                             "info", "%u", &binfo,
2359                             "sector-size", "%lu", &sector_size,
2360                             NULL);
2361         if (err) {
2362                 xenbus_dev_fatal(info->xbdev, err,
2363                                  "reading backend fields at %s",
2364                                  info->xbdev->otherend);
2365                 return;
2366         }
2367
2368         /*
2369          * physcial-sector-size is a newer field, so old backends may not
2370          * provide this. Assume physical sector size to be the same as
2371          * sector_size in that case.
2372          */
2373         physical_sector_size = xenbus_read_unsigned(info->xbdev->otherend,
2374                                                     "physical-sector-size",
2375                                                     sector_size);
2376         blkfront_gather_backend_features(info);
2377         for (i = 0; i < info->nr_rings; i++) {
2378                 err = blkfront_setup_indirect(&info->rinfo[i]);
2379                 if (err) {
2380                         xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2381                                          info->xbdev->otherend);
2382                         blkif_free(info, 0);
2383                         break;
2384                 }
2385         }
2386
2387         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
2388                                   physical_sector_size);
2389         if (err) {
2390                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2391                                  info->xbdev->otherend);
2392                 goto fail;
2393         }
2394
2395         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2396
2397         /* Kick pending requests. */
2398         info->connected = BLKIF_STATE_CONNECTED;
2399         for (i = 0; i < info->nr_rings; i++)
2400                 kick_pending_request_queues(&info->rinfo[i]);
2401
2402         device_add_disk(&info->xbdev->dev, info->gd);
2403
2404         info->is_ready = 1;
2405         return;
2406
2407 fail:
2408         blkif_free(info, 0);
2409         return;
2410 }
2411
2412 /**
2413  * Callback received when the backend's state changes.
2414  */
2415 static void blkback_changed(struct xenbus_device *dev,
2416                             enum xenbus_state backend_state)
2417 {
2418         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2419
2420         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2421
2422         switch (backend_state) {
2423         case XenbusStateInitWait:
2424                 if (dev->state != XenbusStateInitialising)
2425                         break;
2426                 if (talk_to_blkback(dev, info))
2427                         break;
2428         case XenbusStateInitialising:
2429         case XenbusStateInitialised:
2430         case XenbusStateReconfiguring:
2431         case XenbusStateReconfigured:
2432         case XenbusStateUnknown:
2433                 break;
2434
2435         case XenbusStateConnected:
2436                 /*
2437                  * talk_to_blkback sets state to XenbusStateInitialised
2438                  * and blkfront_connect sets it to XenbusStateConnected
2439                  * (if connection went OK).
2440                  *
2441                  * If the backend (or toolstack) decides to poke at backend
2442                  * state (and re-trigger the watch by setting the state repeatedly
2443                  * to XenbusStateConnected (4)) we need to deal with this.
2444                  * This is allowed as this is used to communicate to the guest
2445                  * that the size of disk has changed!
2446                  */
2447                 if ((dev->state != XenbusStateInitialised) &&
2448                     (dev->state != XenbusStateConnected)) {
2449                         if (talk_to_blkback(dev, info))
2450                                 break;
2451                 }
2452
2453                 blkfront_connect(info);
2454                 break;
2455
2456         case XenbusStateClosed:
2457                 if (dev->state == XenbusStateClosed)
2458                         break;
2459                 /* fall through */
2460         case XenbusStateClosing:
2461                 if (info)
2462                         blkfront_closing(info);
2463                 break;
2464         }
2465 }
2466
2467 static int blkfront_remove(struct xenbus_device *xbdev)
2468 {
2469         struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2470         struct block_device *bdev = NULL;
2471         struct gendisk *disk;
2472
2473         dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2474
2475         blkif_free(info, 0);
2476
2477         mutex_lock(&info->mutex);
2478
2479         disk = info->gd;
2480         if (disk)
2481                 bdev = bdget_disk(disk, 0);
2482
2483         info->xbdev = NULL;
2484         mutex_unlock(&info->mutex);
2485
2486         if (!bdev) {
2487                 kfree(info);
2488                 return 0;
2489         }
2490
2491         /*
2492          * The xbdev was removed before we reached the Closed
2493          * state. See if it's safe to remove the disk. If the bdev
2494          * isn't closed yet, we let release take care of it.
2495          */
2496
2497         mutex_lock(&bdev->bd_mutex);
2498         info = disk->private_data;
2499
2500         dev_warn(disk_to_dev(disk),
2501                  "%s was hot-unplugged, %d stale handles\n",
2502                  xbdev->nodename, bdev->bd_openers);
2503
2504         if (info && !bdev->bd_openers) {
2505                 xlvbd_release_gendisk(info);
2506                 disk->private_data = NULL;
2507                 kfree(info);
2508         }
2509
2510         mutex_unlock(&bdev->bd_mutex);
2511         bdput(bdev);
2512
2513         return 0;
2514 }
2515
2516 static int blkfront_is_ready(struct xenbus_device *dev)
2517 {
2518         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2519
2520         return info->is_ready && info->xbdev;
2521 }
2522
2523 static int blkif_open(struct block_device *bdev, fmode_t mode)
2524 {
2525         struct gendisk *disk = bdev->bd_disk;
2526         struct blkfront_info *info;
2527         int err = 0;
2528
2529         mutex_lock(&blkfront_mutex);
2530
2531         info = disk->private_data;
2532         if (!info) {
2533                 /* xbdev gone */
2534                 err = -ERESTARTSYS;
2535                 goto out;
2536         }
2537
2538         mutex_lock(&info->mutex);
2539
2540         if (!info->gd)
2541                 /* xbdev is closed */
2542                 err = -ERESTARTSYS;
2543
2544         mutex_unlock(&info->mutex);
2545
2546 out:
2547         mutex_unlock(&blkfront_mutex);
2548         return err;
2549 }
2550
2551 static void blkif_release(struct gendisk *disk, fmode_t mode)
2552 {
2553         struct blkfront_info *info = disk->private_data;
2554         struct block_device *bdev;
2555         struct xenbus_device *xbdev;
2556
2557         mutex_lock(&blkfront_mutex);
2558
2559         bdev = bdget_disk(disk, 0);
2560
2561         if (!bdev) {
2562                 WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2563                 goto out_mutex;
2564         }
2565         if (bdev->bd_openers)
2566                 goto out;
2567
2568         /*
2569          * Check if we have been instructed to close. We will have
2570          * deferred this request, because the bdev was still open.
2571          */
2572
2573         mutex_lock(&info->mutex);
2574         xbdev = info->xbdev;
2575
2576         if (xbdev && xbdev->state == XenbusStateClosing) {
2577                 /* pending switch to state closed */
2578                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2579                 xlvbd_release_gendisk(info);
2580                 xenbus_frontend_closed(info->xbdev);
2581         }
2582
2583         mutex_unlock(&info->mutex);
2584
2585         if (!xbdev) {
2586                 /* sudden device removal */
2587                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2588                 xlvbd_release_gendisk(info);
2589                 disk->private_data = NULL;
2590                 kfree(info);
2591         }
2592
2593 out:
2594         bdput(bdev);
2595 out_mutex:
2596         mutex_unlock(&blkfront_mutex);
2597 }
2598
2599 static const struct block_device_operations xlvbd_block_fops =
2600 {
2601         .owner = THIS_MODULE,
2602         .open = blkif_open,
2603         .release = blkif_release,
2604         .getgeo = blkif_getgeo,
2605         .ioctl = blkif_ioctl,
2606 };
2607
2608
2609 static const struct xenbus_device_id blkfront_ids[] = {
2610         { "vbd" },
2611         { "" }
2612 };
2613
2614 static struct xenbus_driver blkfront_driver = {
2615         .ids  = blkfront_ids,
2616         .probe = blkfront_probe,
2617         .remove = blkfront_remove,
2618         .resume = blkfront_resume,
2619         .otherend_changed = blkback_changed,
2620         .is_ready = blkfront_is_ready,
2621 };
2622
2623 static int __init xlblk_init(void)
2624 {
2625         int ret;
2626         int nr_cpus = num_online_cpus();
2627
2628         if (!xen_domain())
2629                 return -ENODEV;
2630
2631         if (xen_blkif_max_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST)
2632                 xen_blkif_max_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2633
2634         if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2635                 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2636                         xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2637                 xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2638         }
2639
2640         if (xen_blkif_max_queues > nr_cpus) {
2641                 pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2642                         xen_blkif_max_queues, nr_cpus);
2643                 xen_blkif_max_queues = nr_cpus;
2644         }
2645
2646         if (!xen_has_pv_disk_devices())
2647                 return -ENODEV;
2648
2649         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2650                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
2651                        XENVBD_MAJOR, DEV_NAME);
2652                 return -ENODEV;
2653         }
2654
2655         ret = xenbus_register_frontend(&blkfront_driver);
2656         if (ret) {
2657                 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2658                 return ret;
2659         }
2660
2661         return 0;
2662 }
2663 module_init(xlblk_init);
2664
2665
2666 static void __exit xlblk_exit(void)
2667 {
2668         xenbus_unregister_driver(&blkfront_driver);
2669         unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2670         kfree(minors);
2671 }
2672 module_exit(xlblk_exit);
2673
2674 MODULE_DESCRIPTION("Xen virtual block device frontend");
2675 MODULE_LICENSE("GPL");
2676 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2677 MODULE_ALIAS("xen:vbd");
2678 MODULE_ALIAS("xenblk");