xen: add static initialization of steal_clock op to xen_time_ops
[sfrench/cifs-2.6.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 split_bio {
114         struct bio *bio;
115         atomic_t pending;
116 };
117
118 static DEFINE_MUTEX(blkfront_mutex);
119 static const struct block_device_operations xlvbd_block_fops;
120
121 /*
122  * Maximum number of segments in indirect requests, the actual value used by
123  * the frontend driver is the minimum of this value and the value provided
124  * by the backend driver.
125  */
126
127 static unsigned int xen_blkif_max_segments = 32;
128 module_param_named(max_indirect_segments, xen_blkif_max_segments, uint,
129                    S_IRUGO);
130 MODULE_PARM_DESC(max_indirect_segments,
131                  "Maximum amount of segments in indirect requests (default is 32)");
132
133 static unsigned int xen_blkif_max_queues = 4;
134 module_param_named(max_queues, xen_blkif_max_queues, uint, S_IRUGO);
135 MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
136
137 /*
138  * Maximum order of pages to be used for the shared ring between front and
139  * backend, 4KB page granularity is used.
140  */
141 static unsigned int xen_blkif_max_ring_order;
142 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, S_IRUGO);
143 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
144
145 #define BLK_RING_SIZE(info)     \
146         __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
147
148 #define BLK_MAX_RING_SIZE       \
149         __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * XENBUS_MAX_RING_GRANTS)
150
151 /*
152  * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
153  * characters are enough. Define to 20 to keep consistent with backend.
154  */
155 #define RINGREF_NAME_LEN (20)
156 /*
157  * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
158  */
159 #define QUEUE_NAME_LEN (17)
160
161 /*
162  *  Per-ring info.
163  *  Every blkfront device can associate with one or more blkfront_ring_info,
164  *  depending on how many hardware queues/rings to be used.
165  */
166 struct blkfront_ring_info {
167         /* Lock to protect data in every ring buffer. */
168         spinlock_t ring_lock;
169         struct blkif_front_ring ring;
170         unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
171         unsigned int evtchn, irq;
172         struct work_struct work;
173         struct gnttab_free_callback callback;
174         struct blk_shadow shadow[BLK_MAX_RING_SIZE];
175         struct list_head indirect_pages;
176         struct list_head grants;
177         unsigned int persistent_gnts_c;
178         unsigned long shadow_free;
179         struct blkfront_info *dev_info;
180 };
181
182 /*
183  * We have one of these per vbd, whether ide, scsi or 'other'.  They
184  * hang in private_data off the gendisk structure. We may end up
185  * putting all kinds of interesting stuff here :-)
186  */
187 struct blkfront_info
188 {
189         struct mutex mutex;
190         struct xenbus_device *xbdev;
191         struct gendisk *gd;
192         int vdevice;
193         blkif_vdev_t handle;
194         enum blkif_state connected;
195         /* Number of pages per ring buffer. */
196         unsigned int nr_ring_pages;
197         struct request_queue *rq;
198         unsigned int feature_flush;
199         unsigned int feature_discard:1;
200         unsigned int feature_secdiscard:1;
201         unsigned int discard_granularity;
202         unsigned int discard_alignment;
203         unsigned int feature_persistent:1;
204         /* Number of 4KB segments handled */
205         unsigned int max_indirect_segments;
206         int is_ready;
207         struct blk_mq_tag_set tag_set;
208         struct blkfront_ring_info *rinfo;
209         unsigned int nr_rings;
210 };
211
212 static unsigned int nr_minors;
213 static unsigned long *minors;
214 static DEFINE_SPINLOCK(minor_lock);
215
216 #define GRANT_INVALID_REF       0
217
218 #define PARTS_PER_DISK          16
219 #define PARTS_PER_EXT_DISK      256
220
221 #define BLKIF_MAJOR(dev) ((dev)>>8)
222 #define BLKIF_MINOR(dev) ((dev) & 0xff)
223
224 #define EXT_SHIFT 28
225 #define EXTENDED (1<<EXT_SHIFT)
226 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
227 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
228 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
229 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
230 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
231 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
232
233 #define DEV_NAME        "xvd"   /* name in /dev */
234
235 /*
236  * Grants are always the same size as a Xen page (i.e 4KB).
237  * A physical segment is always the same size as a Linux page.
238  * Number of grants per physical segment
239  */
240 #define GRANTS_PER_PSEG (PAGE_SIZE / XEN_PAGE_SIZE)
241
242 #define GRANTS_PER_INDIRECT_FRAME \
243         (XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
244
245 #define PSEGS_PER_INDIRECT_FRAME        \
246         (GRANTS_INDIRECT_FRAME / GRANTS_PSEGS)
247
248 #define INDIRECT_GREFS(_grants)         \
249         DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
250
251 #define GREFS(_psegs)   ((_psegs) * GRANTS_PER_PSEG)
252
253 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
254 static void blkfront_gather_backend_features(struct blkfront_info *info);
255
256 static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
257 {
258         unsigned long free = rinfo->shadow_free;
259
260         BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
261         rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
262         rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
263         return free;
264 }
265
266 static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
267                               unsigned long id)
268 {
269         if (rinfo->shadow[id].req.u.rw.id != id)
270                 return -EINVAL;
271         if (rinfo->shadow[id].request == NULL)
272                 return -EINVAL;
273         rinfo->shadow[id].req.u.rw.id  = rinfo->shadow_free;
274         rinfo->shadow[id].request = NULL;
275         rinfo->shadow_free = id;
276         return 0;
277 }
278
279 static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
280 {
281         struct blkfront_info *info = rinfo->dev_info;
282         struct page *granted_page;
283         struct grant *gnt_list_entry, *n;
284         int i = 0;
285
286         while (i < num) {
287                 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
288                 if (!gnt_list_entry)
289                         goto out_of_memory;
290
291                 if (info->feature_persistent) {
292                         granted_page = alloc_page(GFP_NOIO);
293                         if (!granted_page) {
294                                 kfree(gnt_list_entry);
295                                 goto out_of_memory;
296                         }
297                         gnt_list_entry->page = granted_page;
298                 }
299
300                 gnt_list_entry->gref = GRANT_INVALID_REF;
301                 list_add(&gnt_list_entry->node, &rinfo->grants);
302                 i++;
303         }
304
305         return 0;
306
307 out_of_memory:
308         list_for_each_entry_safe(gnt_list_entry, n,
309                                  &rinfo->grants, node) {
310                 list_del(&gnt_list_entry->node);
311                 if (info->feature_persistent)
312                         __free_page(gnt_list_entry->page);
313                 kfree(gnt_list_entry);
314                 i--;
315         }
316         BUG_ON(i != 0);
317         return -ENOMEM;
318 }
319
320 static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
321 {
322         struct grant *gnt_list_entry;
323
324         BUG_ON(list_empty(&rinfo->grants));
325         gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
326                                           node);
327         list_del(&gnt_list_entry->node);
328
329         if (gnt_list_entry->gref != GRANT_INVALID_REF)
330                 rinfo->persistent_gnts_c--;
331
332         return gnt_list_entry;
333 }
334
335 static inline void grant_foreign_access(const struct grant *gnt_list_entry,
336                                         const struct blkfront_info *info)
337 {
338         gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
339                                                  info->xbdev->otherend_id,
340                                                  gnt_list_entry->page,
341                                                  0);
342 }
343
344 static struct grant *get_grant(grant_ref_t *gref_head,
345                                unsigned long gfn,
346                                struct blkfront_ring_info *rinfo)
347 {
348         struct grant *gnt_list_entry = get_free_grant(rinfo);
349         struct blkfront_info *info = rinfo->dev_info;
350
351         if (gnt_list_entry->gref != GRANT_INVALID_REF)
352                 return gnt_list_entry;
353
354         /* Assign a gref to this page */
355         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
356         BUG_ON(gnt_list_entry->gref == -ENOSPC);
357         if (info->feature_persistent)
358                 grant_foreign_access(gnt_list_entry, info);
359         else {
360                 /* Grant access to the GFN passed by the caller */
361                 gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
362                                                 info->xbdev->otherend_id,
363                                                 gfn, 0);
364         }
365
366         return gnt_list_entry;
367 }
368
369 static struct grant *get_indirect_grant(grant_ref_t *gref_head,
370                                         struct blkfront_ring_info *rinfo)
371 {
372         struct grant *gnt_list_entry = get_free_grant(rinfo);
373         struct blkfront_info *info = rinfo->dev_info;
374
375         if (gnt_list_entry->gref != GRANT_INVALID_REF)
376                 return gnt_list_entry;
377
378         /* Assign a gref to this page */
379         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
380         BUG_ON(gnt_list_entry->gref == -ENOSPC);
381         if (!info->feature_persistent) {
382                 struct page *indirect_page;
383
384                 /* Fetch a pre-allocated page to use for indirect grefs */
385                 BUG_ON(list_empty(&rinfo->indirect_pages));
386                 indirect_page = list_first_entry(&rinfo->indirect_pages,
387                                                  struct page, lru);
388                 list_del(&indirect_page->lru);
389                 gnt_list_entry->page = indirect_page;
390         }
391         grant_foreign_access(gnt_list_entry, info);
392
393         return gnt_list_entry;
394 }
395
396 static const char *op_name(int op)
397 {
398         static const char *const names[] = {
399                 [BLKIF_OP_READ] = "read",
400                 [BLKIF_OP_WRITE] = "write",
401                 [BLKIF_OP_WRITE_BARRIER] = "barrier",
402                 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
403                 [BLKIF_OP_DISCARD] = "discard" };
404
405         if (op < 0 || op >= ARRAY_SIZE(names))
406                 return "unknown";
407
408         if (!names[op])
409                 return "reserved";
410
411         return names[op];
412 }
413 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
414 {
415         unsigned int end = minor + nr;
416         int rc;
417
418         if (end > nr_minors) {
419                 unsigned long *bitmap, *old;
420
421                 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
422                                  GFP_KERNEL);
423                 if (bitmap == NULL)
424                         return -ENOMEM;
425
426                 spin_lock(&minor_lock);
427                 if (end > nr_minors) {
428                         old = minors;
429                         memcpy(bitmap, minors,
430                                BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
431                         minors = bitmap;
432                         nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
433                 } else
434                         old = bitmap;
435                 spin_unlock(&minor_lock);
436                 kfree(old);
437         }
438
439         spin_lock(&minor_lock);
440         if (find_next_bit(minors, end, minor) >= end) {
441                 bitmap_set(minors, minor, nr);
442                 rc = 0;
443         } else
444                 rc = -EBUSY;
445         spin_unlock(&minor_lock);
446
447         return rc;
448 }
449
450 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
451 {
452         unsigned int end = minor + nr;
453
454         BUG_ON(end > nr_minors);
455         spin_lock(&minor_lock);
456         bitmap_clear(minors,  minor, nr);
457         spin_unlock(&minor_lock);
458 }
459
460 static void blkif_restart_queue_callback(void *arg)
461 {
462         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
463         schedule_work(&rinfo->work);
464 }
465
466 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
467 {
468         /* We don't have real geometry info, but let's at least return
469            values consistent with the size of the device */
470         sector_t nsect = get_capacity(bd->bd_disk);
471         sector_t cylinders = nsect;
472
473         hg->heads = 0xff;
474         hg->sectors = 0x3f;
475         sector_div(cylinders, hg->heads * hg->sectors);
476         hg->cylinders = cylinders;
477         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
478                 hg->cylinders = 0xffff;
479         return 0;
480 }
481
482 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
483                        unsigned command, unsigned long argument)
484 {
485         struct blkfront_info *info = bdev->bd_disk->private_data;
486         int i;
487
488         dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
489                 command, (long)argument);
490
491         switch (command) {
492         case CDROMMULTISESSION:
493                 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
494                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
495                         if (put_user(0, (char __user *)(argument + i)))
496                                 return -EFAULT;
497                 return 0;
498
499         case CDROM_GET_CAPABILITY: {
500                 struct gendisk *gd = info->gd;
501                 if (gd->flags & GENHD_FL_CD)
502                         return 0;
503                 return -EINVAL;
504         }
505
506         default:
507                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
508                   command);*/
509                 return -EINVAL; /* same return as native Linux */
510         }
511
512         return 0;
513 }
514
515 static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
516                                             struct request *req,
517                                             struct blkif_request **ring_req)
518 {
519         unsigned long id;
520
521         *ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
522         rinfo->ring.req_prod_pvt++;
523
524         id = get_id_from_freelist(rinfo);
525         rinfo->shadow[id].request = req;
526         rinfo->shadow[id].status = REQ_WAITING;
527         rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
528
529         (*ring_req)->u.rw.id = id;
530
531         return id;
532 }
533
534 static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
535 {
536         struct blkfront_info *info = rinfo->dev_info;
537         struct blkif_request *ring_req;
538         unsigned long id;
539
540         /* Fill out a communications ring structure. */
541         id = blkif_ring_get_request(rinfo, req, &ring_req);
542
543         ring_req->operation = BLKIF_OP_DISCARD;
544         ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
545         ring_req->u.discard.id = id;
546         ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
547         if ((req->cmd_flags & REQ_SECURE) && info->feature_secdiscard)
548                 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
549         else
550                 ring_req->u.discard.flag = 0;
551
552         /* Keep a private copy so we can reissue requests when recovering. */
553         rinfo->shadow[id].req = *ring_req;
554
555         return 0;
556 }
557
558 struct setup_rw_req {
559         unsigned int grant_idx;
560         struct blkif_request_segment *segments;
561         struct blkfront_ring_info *rinfo;
562         struct blkif_request *ring_req;
563         grant_ref_t gref_head;
564         unsigned int id;
565         /* Only used when persistent grant is used and it's a read request */
566         bool need_copy;
567         unsigned int bvec_off;
568         char *bvec_data;
569
570         bool require_extra_req;
571         struct blkif_request *extra_ring_req;
572 };
573
574 static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
575                                      unsigned int len, void *data)
576 {
577         struct setup_rw_req *setup = data;
578         int n, ref;
579         struct grant *gnt_list_entry;
580         unsigned int fsect, lsect;
581         /* Convenient aliases */
582         unsigned int grant_idx = setup->grant_idx;
583         struct blkif_request *ring_req = setup->ring_req;
584         struct blkfront_ring_info *rinfo = setup->rinfo;
585         /*
586          * We always use the shadow of the first request to store the list
587          * of grant associated to the block I/O request. This made the
588          * completion more easy to handle even if the block I/O request is
589          * split.
590          */
591         struct blk_shadow *shadow = &rinfo->shadow[setup->id];
592
593         if (unlikely(setup->require_extra_req &&
594                      grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
595                 /*
596                  * We are using the second request, setup grant_idx
597                  * to be the index of the segment array.
598                  */
599                 grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
600                 ring_req = setup->extra_ring_req;
601         }
602
603         if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
604             (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
605                 if (setup->segments)
606                         kunmap_atomic(setup->segments);
607
608                 n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
609                 gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
610                 shadow->indirect_grants[n] = gnt_list_entry;
611                 setup->segments = kmap_atomic(gnt_list_entry->page);
612                 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
613         }
614
615         gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
616         ref = gnt_list_entry->gref;
617         /*
618          * All the grants are stored in the shadow of the first
619          * request. Therefore we have to use the global index.
620          */
621         shadow->grants_used[setup->grant_idx] = gnt_list_entry;
622
623         if (setup->need_copy) {
624                 void *shared_data;
625
626                 shared_data = kmap_atomic(gnt_list_entry->page);
627                 /*
628                  * this does not wipe data stored outside the
629                  * range sg->offset..sg->offset+sg->length.
630                  * Therefore, blkback *could* see data from
631                  * previous requests. This is OK as long as
632                  * persistent grants are shared with just one
633                  * domain. It may need refactoring if this
634                  * changes
635                  */
636                 memcpy(shared_data + offset,
637                        setup->bvec_data + setup->bvec_off,
638                        len);
639
640                 kunmap_atomic(shared_data);
641                 setup->bvec_off += len;
642         }
643
644         fsect = offset >> 9;
645         lsect = fsect + (len >> 9) - 1;
646         if (ring_req->operation != BLKIF_OP_INDIRECT) {
647                 ring_req->u.rw.seg[grant_idx] =
648                         (struct blkif_request_segment) {
649                                 .gref       = ref,
650                                 .first_sect = fsect,
651                                 .last_sect  = lsect };
652         } else {
653                 setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
654                         (struct blkif_request_segment) {
655                                 .gref       = ref,
656                                 .first_sect = fsect,
657                                 .last_sect  = lsect };
658         }
659
660         (setup->grant_idx)++;
661 }
662
663 static void blkif_setup_extra_req(struct blkif_request *first,
664                                   struct blkif_request *second)
665 {
666         uint16_t nr_segments = first->u.rw.nr_segments;
667
668         /*
669          * The second request is only present when the first request uses
670          * all its segments. It's always the continuity of the first one.
671          */
672         first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
673
674         second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
675         second->u.rw.sector_number = first->u.rw.sector_number +
676                 (BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
677
678         second->u.rw.handle = first->u.rw.handle;
679         second->operation = first->operation;
680 }
681
682 static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
683 {
684         struct blkfront_info *info = rinfo->dev_info;
685         struct blkif_request *ring_req, *extra_ring_req = NULL;
686         unsigned long id, extra_id = NO_ASSOCIATED_ID;
687         bool require_extra_req = false;
688         int i;
689         struct setup_rw_req setup = {
690                 .grant_idx = 0,
691                 .segments = NULL,
692                 .rinfo = rinfo,
693                 .need_copy = rq_data_dir(req) && info->feature_persistent,
694         };
695
696         /*
697          * Used to store if we are able to queue the request by just using
698          * existing persistent grants, or if we have to get new grants,
699          * as there are not sufficiently many free.
700          */
701         struct scatterlist *sg;
702         int num_sg, max_grefs, num_grant;
703
704         max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
705         if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
706                 /*
707                  * If we are using indirect segments we need to account
708                  * for the indirect grefs used in the request.
709                  */
710                 max_grefs += INDIRECT_GREFS(max_grefs);
711
712         /*
713          * We have to reserve 'max_grefs' grants because persistent
714          * grants are shared by all rings.
715          */
716         if (max_grefs > 0)
717                 if (gnttab_alloc_grant_references(max_grefs, &setup.gref_head) < 0) {
718                         gnttab_request_free_callback(
719                                 &rinfo->callback,
720                                 blkif_restart_queue_callback,
721                                 rinfo,
722                                 max_grefs);
723                         return 1;
724                 }
725
726         /* Fill out a communications ring structure. */
727         id = blkif_ring_get_request(rinfo, req, &ring_req);
728
729         num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
730         num_grant = 0;
731         /* Calculate the number of grant used */
732         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
733                num_grant += gnttab_count_grant(sg->offset, sg->length);
734
735         require_extra_req = info->max_indirect_segments == 0 &&
736                 num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
737         BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
738
739         rinfo->shadow[id].num_sg = num_sg;
740         if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
741             likely(!require_extra_req)) {
742                 /*
743                  * The indirect operation can only be a BLKIF_OP_READ or
744                  * BLKIF_OP_WRITE
745                  */
746                 BUG_ON(req->cmd_flags & (REQ_FLUSH | REQ_FUA));
747                 ring_req->operation = BLKIF_OP_INDIRECT;
748                 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
749                         BLKIF_OP_WRITE : BLKIF_OP_READ;
750                 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
751                 ring_req->u.indirect.handle = info->handle;
752                 ring_req->u.indirect.nr_segments = num_grant;
753         } else {
754                 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
755                 ring_req->u.rw.handle = info->handle;
756                 ring_req->operation = rq_data_dir(req) ?
757                         BLKIF_OP_WRITE : BLKIF_OP_READ;
758                 if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) {
759                         /*
760                          * Ideally we can do an unordered flush-to-disk.
761                          * In case the backend onlysupports barriers, use that.
762                          * A barrier request a superset of FUA, so we can
763                          * implement it the same way.  (It's also a FLUSH+FUA,
764                          * since it is guaranteed ordered WRT previous writes.)
765                          */
766                         switch (info->feature_flush &
767                                 ((REQ_FLUSH|REQ_FUA))) {
768                         case REQ_FLUSH|REQ_FUA:
769                                 ring_req->operation =
770                                         BLKIF_OP_WRITE_BARRIER;
771                                 break;
772                         case REQ_FLUSH:
773                                 ring_req->operation =
774                                         BLKIF_OP_FLUSH_DISKCACHE;
775                                 break;
776                         default:
777                                 ring_req->operation = 0;
778                         }
779                 }
780                 ring_req->u.rw.nr_segments = num_grant;
781                 if (unlikely(require_extra_req)) {
782                         extra_id = blkif_ring_get_request(rinfo, req,
783                                                           &extra_ring_req);
784                         /*
785                          * Only the first request contains the scatter-gather
786                          * list.
787                          */
788                         rinfo->shadow[extra_id].num_sg = 0;
789
790                         blkif_setup_extra_req(ring_req, extra_ring_req);
791
792                         /* Link the 2 requests together */
793                         rinfo->shadow[extra_id].associated_id = id;
794                         rinfo->shadow[id].associated_id = extra_id;
795                 }
796         }
797
798         setup.ring_req = ring_req;
799         setup.id = id;
800
801         setup.require_extra_req = require_extra_req;
802         if (unlikely(require_extra_req))
803                 setup.extra_ring_req = extra_ring_req;
804
805         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
806                 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
807
808                 if (setup.need_copy) {
809                         setup.bvec_off = sg->offset;
810                         setup.bvec_data = kmap_atomic(sg_page(sg));
811                 }
812
813                 gnttab_foreach_grant_in_range(sg_page(sg),
814                                               sg->offset,
815                                               sg->length,
816                                               blkif_setup_rw_req_grant,
817                                               &setup);
818
819                 if (setup.need_copy)
820                         kunmap_atomic(setup.bvec_data);
821         }
822         if (setup.segments)
823                 kunmap_atomic(setup.segments);
824
825         /* Keep a private copy so we can reissue requests when recovering. */
826         rinfo->shadow[id].req = *ring_req;
827         if (unlikely(require_extra_req))
828                 rinfo->shadow[extra_id].req = *extra_ring_req;
829
830         if (max_grefs > 0)
831                 gnttab_free_grant_references(setup.gref_head);
832
833         return 0;
834 }
835
836 /*
837  * Generate a Xen blkfront IO request from a blk layer request.  Reads
838  * and writes are handled as expected.
839  *
840  * @req: a request struct
841  */
842 static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
843 {
844         if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
845                 return 1;
846
847         if (unlikely(req->cmd_flags & (REQ_DISCARD | REQ_SECURE)))
848                 return blkif_queue_discard_req(req, rinfo);
849         else
850                 return blkif_queue_rw_req(req, rinfo);
851 }
852
853 static inline void flush_requests(struct blkfront_ring_info *rinfo)
854 {
855         int notify;
856
857         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
858
859         if (notify)
860                 notify_remote_via_irq(rinfo->irq);
861 }
862
863 static inline bool blkif_request_flush_invalid(struct request *req,
864                                                struct blkfront_info *info)
865 {
866         return ((req->cmd_type != REQ_TYPE_FS) ||
867                 ((req->cmd_flags & REQ_FLUSH) &&
868                  !(info->feature_flush & REQ_FLUSH)) ||
869                 ((req->cmd_flags & REQ_FUA) &&
870                  !(info->feature_flush & REQ_FUA)));
871 }
872
873 static int blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
874                           const struct blk_mq_queue_data *qd)
875 {
876         unsigned long flags;
877         int qid = hctx->queue_num;
878         struct blkfront_info *info = hctx->queue->queuedata;
879         struct blkfront_ring_info *rinfo = NULL;
880
881         BUG_ON(info->nr_rings <= qid);
882         rinfo = &info->rinfo[qid];
883         blk_mq_start_request(qd->rq);
884         spin_lock_irqsave(&rinfo->ring_lock, flags);
885         if (RING_FULL(&rinfo->ring))
886                 goto out_busy;
887
888         if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
889                 goto out_err;
890
891         if (blkif_queue_request(qd->rq, rinfo))
892                 goto out_busy;
893
894         flush_requests(rinfo);
895         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
896         return BLK_MQ_RQ_QUEUE_OK;
897
898 out_err:
899         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
900         return BLK_MQ_RQ_QUEUE_ERROR;
901
902 out_busy:
903         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
904         blk_mq_stop_hw_queue(hctx);
905         return BLK_MQ_RQ_QUEUE_BUSY;
906 }
907
908 static struct blk_mq_ops blkfront_mq_ops = {
909         .queue_rq = blkif_queue_rq,
910         .map_queue = blk_mq_map_queue,
911 };
912
913 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
914                                 unsigned int physical_sector_size,
915                                 unsigned int segments)
916 {
917         struct request_queue *rq;
918         struct blkfront_info *info = gd->private_data;
919
920         memset(&info->tag_set, 0, sizeof(info->tag_set));
921         info->tag_set.ops = &blkfront_mq_ops;
922         info->tag_set.nr_hw_queues = info->nr_rings;
923         if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
924                 /*
925                  * When indirect descriptior is not supported, the I/O request
926                  * will be split between multiple request in the ring.
927                  * To avoid problems when sending the request, divide by
928                  * 2 the depth of the queue.
929                  */
930                 info->tag_set.queue_depth =  BLK_RING_SIZE(info) / 2;
931         } else
932                 info->tag_set.queue_depth = BLK_RING_SIZE(info);
933         info->tag_set.numa_node = NUMA_NO_NODE;
934         info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
935         info->tag_set.cmd_size = 0;
936         info->tag_set.driver_data = info;
937
938         if (blk_mq_alloc_tag_set(&info->tag_set))
939                 return -EINVAL;
940         rq = blk_mq_init_queue(&info->tag_set);
941         if (IS_ERR(rq)) {
942                 blk_mq_free_tag_set(&info->tag_set);
943                 return PTR_ERR(rq);
944         }
945
946         rq->queuedata = info;
947         queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
948
949         if (info->feature_discard) {
950                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
951                 blk_queue_max_discard_sectors(rq, get_capacity(gd));
952                 rq->limits.discard_granularity = info->discard_granularity;
953                 rq->limits.discard_alignment = info->discard_alignment;
954                 if (info->feature_secdiscard)
955                         queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD, rq);
956         }
957
958         /* Hard sector size and max sectors impersonate the equiv. hardware. */
959         blk_queue_logical_block_size(rq, sector_size);
960         blk_queue_physical_block_size(rq, physical_sector_size);
961         blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
962
963         /* Each segment in a request is up to an aligned page in size. */
964         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
965         blk_queue_max_segment_size(rq, PAGE_SIZE);
966
967         /* Ensure a merged request will fit in a single I/O ring slot. */
968         blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
969
970         /* Make sure buffer addresses are sector-aligned. */
971         blk_queue_dma_alignment(rq, 511);
972
973         /* Make sure we don't use bounce buffers. */
974         blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
975
976         gd->queue = rq;
977
978         return 0;
979 }
980
981 static const char *flush_info(unsigned int feature_flush)
982 {
983         switch (feature_flush & ((REQ_FLUSH | REQ_FUA))) {
984         case REQ_FLUSH|REQ_FUA:
985                 return "barrier: enabled;";
986         case REQ_FLUSH:
987                 return "flush diskcache: enabled;";
988         default:
989                 return "barrier or flush: disabled;";
990         }
991 }
992
993 static void xlvbd_flush(struct blkfront_info *info)
994 {
995         blk_queue_write_cache(info->rq, info->feature_flush & REQ_FLUSH,
996                                 info->feature_flush & REQ_FUA);
997         pr_info("blkfront: %s: %s %s %s %s %s\n",
998                 info->gd->disk_name, flush_info(info->feature_flush),
999                 "persistent grants:", info->feature_persistent ?
1000                 "enabled;" : "disabled;", "indirect descriptors:",
1001                 info->max_indirect_segments ? "enabled;" : "disabled;");
1002 }
1003
1004 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
1005 {
1006         int major;
1007         major = BLKIF_MAJOR(vdevice);
1008         *minor = BLKIF_MINOR(vdevice);
1009         switch (major) {
1010                 case XEN_IDE0_MAJOR:
1011                         *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
1012                         *minor = ((*minor / 64) * PARTS_PER_DISK) +
1013                                 EMULATED_HD_DISK_MINOR_OFFSET;
1014                         break;
1015                 case XEN_IDE1_MAJOR:
1016                         *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1017                         *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1018                                 EMULATED_HD_DISK_MINOR_OFFSET;
1019                         break;
1020                 case XEN_SCSI_DISK0_MAJOR:
1021                         *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1022                         *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1023                         break;
1024                 case XEN_SCSI_DISK1_MAJOR:
1025                 case XEN_SCSI_DISK2_MAJOR:
1026                 case XEN_SCSI_DISK3_MAJOR:
1027                 case XEN_SCSI_DISK4_MAJOR:
1028                 case XEN_SCSI_DISK5_MAJOR:
1029                 case XEN_SCSI_DISK6_MAJOR:
1030                 case XEN_SCSI_DISK7_MAJOR:
1031                         *offset = (*minor / PARTS_PER_DISK) + 
1032                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1033                                 EMULATED_SD_DISK_NAME_OFFSET;
1034                         *minor = *minor +
1035                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1036                                 EMULATED_SD_DISK_MINOR_OFFSET;
1037                         break;
1038                 case XEN_SCSI_DISK8_MAJOR:
1039                 case XEN_SCSI_DISK9_MAJOR:
1040                 case XEN_SCSI_DISK10_MAJOR:
1041                 case XEN_SCSI_DISK11_MAJOR:
1042                 case XEN_SCSI_DISK12_MAJOR:
1043                 case XEN_SCSI_DISK13_MAJOR:
1044                 case XEN_SCSI_DISK14_MAJOR:
1045                 case XEN_SCSI_DISK15_MAJOR:
1046                         *offset = (*minor / PARTS_PER_DISK) + 
1047                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1048                                 EMULATED_SD_DISK_NAME_OFFSET;
1049                         *minor = *minor +
1050                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1051                                 EMULATED_SD_DISK_MINOR_OFFSET;
1052                         break;
1053                 case XENVBD_MAJOR:
1054                         *offset = *minor / PARTS_PER_DISK;
1055                         break;
1056                 default:
1057                         printk(KERN_WARNING "blkfront: your disk configuration is "
1058                                         "incorrect, please use an xvd device instead\n");
1059                         return -ENODEV;
1060         }
1061         return 0;
1062 }
1063
1064 static char *encode_disk_name(char *ptr, unsigned int n)
1065 {
1066         if (n >= 26)
1067                 ptr = encode_disk_name(ptr, n / 26 - 1);
1068         *ptr = 'a' + n % 26;
1069         return ptr + 1;
1070 }
1071
1072 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1073                                struct blkfront_info *info,
1074                                u16 vdisk_info, u16 sector_size,
1075                                unsigned int physical_sector_size)
1076 {
1077         struct gendisk *gd;
1078         int nr_minors = 1;
1079         int err;
1080         unsigned int offset;
1081         int minor;
1082         int nr_parts;
1083         char *ptr;
1084
1085         BUG_ON(info->gd != NULL);
1086         BUG_ON(info->rq != NULL);
1087
1088         if ((info->vdevice>>EXT_SHIFT) > 1) {
1089                 /* this is above the extended range; something is wrong */
1090                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1091                 return -ENODEV;
1092         }
1093
1094         if (!VDEV_IS_EXTENDED(info->vdevice)) {
1095                 err = xen_translate_vdev(info->vdevice, &minor, &offset);
1096                 if (err)
1097                         return err;             
1098                 nr_parts = PARTS_PER_DISK;
1099         } else {
1100                 minor = BLKIF_MINOR_EXT(info->vdevice);
1101                 nr_parts = PARTS_PER_EXT_DISK;
1102                 offset = minor / nr_parts;
1103                 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1104                         printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1105                                         "emulated IDE disks,\n\t choose an xvd device name"
1106                                         "from xvde on\n", info->vdevice);
1107         }
1108         if (minor >> MINORBITS) {
1109                 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1110                         info->vdevice, minor);
1111                 return -ENODEV;
1112         }
1113
1114         if ((minor % nr_parts) == 0)
1115                 nr_minors = nr_parts;
1116
1117         err = xlbd_reserve_minors(minor, nr_minors);
1118         if (err)
1119                 goto out;
1120         err = -ENODEV;
1121
1122         gd = alloc_disk(nr_minors);
1123         if (gd == NULL)
1124                 goto release;
1125
1126         strcpy(gd->disk_name, DEV_NAME);
1127         ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1128         BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1129         if (nr_minors > 1)
1130                 *ptr = 0;
1131         else
1132                 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1133                          "%d", minor & (nr_parts - 1));
1134
1135         gd->major = XENVBD_MAJOR;
1136         gd->first_minor = minor;
1137         gd->fops = &xlvbd_block_fops;
1138         gd->private_data = info;
1139         gd->driverfs_dev = &(info->xbdev->dev);
1140         set_capacity(gd, capacity);
1141
1142         if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size,
1143                                  info->max_indirect_segments ? :
1144                                  BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
1145                 del_gendisk(gd);
1146                 goto release;
1147         }
1148
1149         info->rq = gd->queue;
1150         info->gd = gd;
1151
1152         xlvbd_flush(info);
1153
1154         if (vdisk_info & VDISK_READONLY)
1155                 set_disk_ro(gd, 1);
1156
1157         if (vdisk_info & VDISK_REMOVABLE)
1158                 gd->flags |= GENHD_FL_REMOVABLE;
1159
1160         if (vdisk_info & VDISK_CDROM)
1161                 gd->flags |= GENHD_FL_CD;
1162
1163         return 0;
1164
1165  release:
1166         xlbd_release_minors(minor, nr_minors);
1167  out:
1168         return err;
1169 }
1170
1171 static void xlvbd_release_gendisk(struct blkfront_info *info)
1172 {
1173         unsigned int minor, nr_minors, i;
1174
1175         if (info->rq == NULL)
1176                 return;
1177
1178         /* No more blkif_request(). */
1179         blk_mq_stop_hw_queues(info->rq);
1180
1181         for (i = 0; i < info->nr_rings; i++) {
1182                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1183
1184                 /* No more gnttab callback work. */
1185                 gnttab_cancel_free_callback(&rinfo->callback);
1186
1187                 /* Flush gnttab callback work. Must be done with no locks held. */
1188                 flush_work(&rinfo->work);
1189         }
1190
1191         del_gendisk(info->gd);
1192
1193         minor = info->gd->first_minor;
1194         nr_minors = info->gd->minors;
1195         xlbd_release_minors(minor, nr_minors);
1196
1197         blk_cleanup_queue(info->rq);
1198         blk_mq_free_tag_set(&info->tag_set);
1199         info->rq = NULL;
1200
1201         put_disk(info->gd);
1202         info->gd = NULL;
1203 }
1204
1205 /* Already hold rinfo->ring_lock. */
1206 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1207 {
1208         if (!RING_FULL(&rinfo->ring))
1209                 blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1210 }
1211
1212 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1213 {
1214         unsigned long flags;
1215
1216         spin_lock_irqsave(&rinfo->ring_lock, flags);
1217         kick_pending_request_queues_locked(rinfo);
1218         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1219 }
1220
1221 static void blkif_restart_queue(struct work_struct *work)
1222 {
1223         struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1224
1225         if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1226                 kick_pending_request_queues(rinfo);
1227 }
1228
1229 static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1230 {
1231         struct grant *persistent_gnt, *n;
1232         struct blkfront_info *info = rinfo->dev_info;
1233         int i, j, segs;
1234
1235         /*
1236          * Remove indirect pages, this only happens when using indirect
1237          * descriptors but not persistent grants
1238          */
1239         if (!list_empty(&rinfo->indirect_pages)) {
1240                 struct page *indirect_page, *n;
1241
1242                 BUG_ON(info->feature_persistent);
1243                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1244                         list_del(&indirect_page->lru);
1245                         __free_page(indirect_page);
1246                 }
1247         }
1248
1249         /* Remove all persistent grants. */
1250         if (!list_empty(&rinfo->grants)) {
1251                 list_for_each_entry_safe(persistent_gnt, n,
1252                                          &rinfo->grants, node) {
1253                         list_del(&persistent_gnt->node);
1254                         if (persistent_gnt->gref != GRANT_INVALID_REF) {
1255                                 gnttab_end_foreign_access(persistent_gnt->gref,
1256                                                           0, 0UL);
1257                                 rinfo->persistent_gnts_c--;
1258                         }
1259                         if (info->feature_persistent)
1260                                 __free_page(persistent_gnt->page);
1261                         kfree(persistent_gnt);
1262                 }
1263         }
1264         BUG_ON(rinfo->persistent_gnts_c != 0);
1265
1266         for (i = 0; i < BLK_RING_SIZE(info); i++) {
1267                 /*
1268                  * Clear persistent grants present in requests already
1269                  * on the shared ring
1270                  */
1271                 if (!rinfo->shadow[i].request)
1272                         goto free_shadow;
1273
1274                 segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1275                        rinfo->shadow[i].req.u.indirect.nr_segments :
1276                        rinfo->shadow[i].req.u.rw.nr_segments;
1277                 for (j = 0; j < segs; j++) {
1278                         persistent_gnt = rinfo->shadow[i].grants_used[j];
1279                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1280                         if (info->feature_persistent)
1281                                 __free_page(persistent_gnt->page);
1282                         kfree(persistent_gnt);
1283                 }
1284
1285                 if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1286                         /*
1287                          * If this is not an indirect operation don't try to
1288                          * free indirect segments
1289                          */
1290                         goto free_shadow;
1291
1292                 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1293                         persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1294                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1295                         __free_page(persistent_gnt->page);
1296                         kfree(persistent_gnt);
1297                 }
1298
1299 free_shadow:
1300                 kfree(rinfo->shadow[i].grants_used);
1301                 rinfo->shadow[i].grants_used = NULL;
1302                 kfree(rinfo->shadow[i].indirect_grants);
1303                 rinfo->shadow[i].indirect_grants = NULL;
1304                 kfree(rinfo->shadow[i].sg);
1305                 rinfo->shadow[i].sg = NULL;
1306         }
1307
1308         /* No more gnttab callback work. */
1309         gnttab_cancel_free_callback(&rinfo->callback);
1310
1311         /* Flush gnttab callback work. Must be done with no locks held. */
1312         flush_work(&rinfo->work);
1313
1314         /* Free resources associated with old device channel. */
1315         for (i = 0; i < info->nr_ring_pages; i++) {
1316                 if (rinfo->ring_ref[i] != GRANT_INVALID_REF) {
1317                         gnttab_end_foreign_access(rinfo->ring_ref[i], 0, 0);
1318                         rinfo->ring_ref[i] = GRANT_INVALID_REF;
1319                 }
1320         }
1321         free_pages((unsigned long)rinfo->ring.sring, get_order(info->nr_ring_pages * PAGE_SIZE));
1322         rinfo->ring.sring = NULL;
1323
1324         if (rinfo->irq)
1325                 unbind_from_irqhandler(rinfo->irq, rinfo);
1326         rinfo->evtchn = rinfo->irq = 0;
1327 }
1328
1329 static void blkif_free(struct blkfront_info *info, int suspend)
1330 {
1331         unsigned int i;
1332
1333         /* Prevent new requests being issued until we fix things up. */
1334         info->connected = suspend ?
1335                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1336         /* No more blkif_request(). */
1337         if (info->rq)
1338                 blk_mq_stop_hw_queues(info->rq);
1339
1340         for (i = 0; i < info->nr_rings; i++)
1341                 blkif_free_ring(&info->rinfo[i]);
1342
1343         kfree(info->rinfo);
1344         info->rinfo = NULL;
1345         info->nr_rings = 0;
1346 }
1347
1348 struct copy_from_grant {
1349         const struct blk_shadow *s;
1350         unsigned int grant_idx;
1351         unsigned int bvec_offset;
1352         char *bvec_data;
1353 };
1354
1355 static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1356                                   unsigned int len, void *data)
1357 {
1358         struct copy_from_grant *info = data;
1359         char *shared_data;
1360         /* Convenient aliases */
1361         const struct blk_shadow *s = info->s;
1362
1363         shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1364
1365         memcpy(info->bvec_data + info->bvec_offset,
1366                shared_data + offset, len);
1367
1368         info->bvec_offset += len;
1369         info->grant_idx++;
1370
1371         kunmap_atomic(shared_data);
1372 }
1373
1374 static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1375 {
1376         switch (rsp)
1377         {
1378         case BLKIF_RSP_OKAY:
1379                 return REQ_DONE;
1380         case BLKIF_RSP_EOPNOTSUPP:
1381                 return REQ_EOPNOTSUPP;
1382         case BLKIF_RSP_ERROR:
1383                 /* Fallthrough. */
1384         default:
1385                 return REQ_ERROR;
1386         }
1387 }
1388
1389 /*
1390  * Get the final status of the block request based on two ring response
1391  */
1392 static int blkif_get_final_status(enum blk_req_status s1,
1393                                   enum blk_req_status s2)
1394 {
1395         BUG_ON(s1 == REQ_WAITING);
1396         BUG_ON(s2 == REQ_WAITING);
1397
1398         if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1399                 return BLKIF_RSP_ERROR;
1400         else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1401                 return BLKIF_RSP_EOPNOTSUPP;
1402         return BLKIF_RSP_OKAY;
1403 }
1404
1405 static bool blkif_completion(unsigned long *id,
1406                              struct blkfront_ring_info *rinfo,
1407                              struct blkif_response *bret)
1408 {
1409         int i = 0;
1410         struct scatterlist *sg;
1411         int num_sg, num_grant;
1412         struct blkfront_info *info = rinfo->dev_info;
1413         struct blk_shadow *s = &rinfo->shadow[*id];
1414         struct copy_from_grant data = {
1415                 .grant_idx = 0,
1416         };
1417
1418         num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1419                 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1420
1421         /* The I/O request may be split in two. */
1422         if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1423                 struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1424
1425                 /* Keep the status of the current response in shadow. */
1426                 s->status = blkif_rsp_to_req_status(bret->status);
1427
1428                 /* Wait the second response if not yet here. */
1429                 if (s2->status == REQ_WAITING)
1430                         return 0;
1431
1432                 bret->status = blkif_get_final_status(s->status,
1433                                                       s2->status);
1434
1435                 /*
1436                  * All the grants is stored in the first shadow in order
1437                  * to make the completion code simpler.
1438                  */
1439                 num_grant += s2->req.u.rw.nr_segments;
1440
1441                 /*
1442                  * The two responses may not come in order. Only the
1443                  * first request will store the scatter-gather list.
1444                  */
1445                 if (s2->num_sg != 0) {
1446                         /* Update "id" with the ID of the first response. */
1447                         *id = s->associated_id;
1448                         s = s2;
1449                 }
1450
1451                 /*
1452                  * We don't need anymore the second request, so recycling
1453                  * it now.
1454                  */
1455                 if (add_id_to_freelist(rinfo, s->associated_id))
1456                         WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1457                              info->gd->disk_name, s->associated_id);
1458         }
1459
1460         data.s = s;
1461         num_sg = s->num_sg;
1462
1463         if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1464                 for_each_sg(s->sg, sg, num_sg, i) {
1465                         BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1466
1467                         data.bvec_offset = sg->offset;
1468                         data.bvec_data = kmap_atomic(sg_page(sg));
1469
1470                         gnttab_foreach_grant_in_range(sg_page(sg),
1471                                                       sg->offset,
1472                                                       sg->length,
1473                                                       blkif_copy_from_grant,
1474                                                       &data);
1475
1476                         kunmap_atomic(data.bvec_data);
1477                 }
1478         }
1479         /* Add the persistent grant into the list of free grants */
1480         for (i = 0; i < num_grant; i++) {
1481                 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1482                         /*
1483                          * If the grant is still mapped by the backend (the
1484                          * backend has chosen to make this grant persistent)
1485                          * we add it at the head of the list, so it will be
1486                          * reused first.
1487                          */
1488                         if (!info->feature_persistent)
1489                                 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1490                                                      s->grants_used[i]->gref);
1491                         list_add(&s->grants_used[i]->node, &rinfo->grants);
1492                         rinfo->persistent_gnts_c++;
1493                 } else {
1494                         /*
1495                          * If the grant is not mapped by the backend we end the
1496                          * foreign access and add it to the tail of the list,
1497                          * so it will not be picked again unless we run out of
1498                          * persistent grants.
1499                          */
1500                         gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1501                         s->grants_used[i]->gref = GRANT_INVALID_REF;
1502                         list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1503                 }
1504         }
1505         if (s->req.operation == BLKIF_OP_INDIRECT) {
1506                 for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1507                         if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1508                                 if (!info->feature_persistent)
1509                                         pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1510                                                              s->indirect_grants[i]->gref);
1511                                 list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1512                                 rinfo->persistent_gnts_c++;
1513                         } else {
1514                                 struct page *indirect_page;
1515
1516                                 gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1517                                 /*
1518                                  * Add the used indirect page back to the list of
1519                                  * available pages for indirect grefs.
1520                                  */
1521                                 if (!info->feature_persistent) {
1522                                         indirect_page = s->indirect_grants[i]->page;
1523                                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
1524                                 }
1525                                 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1526                                 list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1527                         }
1528                 }
1529         }
1530
1531         return 1;
1532 }
1533
1534 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1535 {
1536         struct request *req;
1537         struct blkif_response *bret;
1538         RING_IDX i, rp;
1539         unsigned long flags;
1540         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1541         struct blkfront_info *info = rinfo->dev_info;
1542         int error;
1543
1544         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
1545                 return IRQ_HANDLED;
1546
1547         spin_lock_irqsave(&rinfo->ring_lock, flags);
1548  again:
1549         rp = rinfo->ring.sring->rsp_prod;
1550         rmb(); /* Ensure we see queued responses up to 'rp'. */
1551
1552         for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1553                 unsigned long id;
1554
1555                 bret = RING_GET_RESPONSE(&rinfo->ring, i);
1556                 id   = bret->id;
1557                 /*
1558                  * The backend has messed up and given us an id that we would
1559                  * never have given to it (we stamp it up to BLK_RING_SIZE -
1560                  * look in get_id_from_freelist.
1561                  */
1562                 if (id >= BLK_RING_SIZE(info)) {
1563                         WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1564                              info->gd->disk_name, op_name(bret->operation), id);
1565                         /* We can't safely get the 'struct request' as
1566                          * the id is busted. */
1567                         continue;
1568                 }
1569                 req  = rinfo->shadow[id].request;
1570
1571                 if (bret->operation != BLKIF_OP_DISCARD) {
1572                         /*
1573                          * We may need to wait for an extra response if the
1574                          * I/O request is split in 2
1575                          */
1576                         if (!blkif_completion(&id, rinfo, bret))
1577                                 continue;
1578                 }
1579
1580                 if (add_id_to_freelist(rinfo, id)) {
1581                         WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1582                              info->gd->disk_name, op_name(bret->operation), id);
1583                         continue;
1584                 }
1585
1586                 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
1587                 switch (bret->operation) {
1588                 case BLKIF_OP_DISCARD:
1589                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1590                                 struct request_queue *rq = info->rq;
1591                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1592                                            info->gd->disk_name, op_name(bret->operation));
1593                                 error = -EOPNOTSUPP;
1594                                 info->feature_discard = 0;
1595                                 info->feature_secdiscard = 0;
1596                                 queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1597                                 queue_flag_clear(QUEUE_FLAG_SECDISCARD, rq);
1598                         }
1599                         blk_mq_complete_request(req, error);
1600                         break;
1601                 case BLKIF_OP_FLUSH_DISKCACHE:
1602                 case BLKIF_OP_WRITE_BARRIER:
1603                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1604                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1605                                        info->gd->disk_name, op_name(bret->operation));
1606                                 error = -EOPNOTSUPP;
1607                         }
1608                         if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1609                                      rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1610                                 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1611                                        info->gd->disk_name, op_name(bret->operation));
1612                                 error = -EOPNOTSUPP;
1613                         }
1614                         if (unlikely(error)) {
1615                                 if (error == -EOPNOTSUPP)
1616                                         error = 0;
1617                                 info->feature_flush = 0;
1618                                 xlvbd_flush(info);
1619                         }
1620                         /* fall through */
1621                 case BLKIF_OP_READ:
1622                 case BLKIF_OP_WRITE:
1623                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
1624                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1625                                         "request: %x\n", bret->status);
1626
1627                         blk_mq_complete_request(req, error);
1628                         break;
1629                 default:
1630                         BUG();
1631                 }
1632         }
1633
1634         rinfo->ring.rsp_cons = i;
1635
1636         if (i != rinfo->ring.req_prod_pvt) {
1637                 int more_to_do;
1638                 RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1639                 if (more_to_do)
1640                         goto again;
1641         } else
1642                 rinfo->ring.sring->rsp_event = i + 1;
1643
1644         kick_pending_request_queues_locked(rinfo);
1645
1646         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1647
1648         return IRQ_HANDLED;
1649 }
1650
1651
1652 static int setup_blkring(struct xenbus_device *dev,
1653                          struct blkfront_ring_info *rinfo)
1654 {
1655         struct blkif_sring *sring;
1656         int err, i;
1657         struct blkfront_info *info = rinfo->dev_info;
1658         unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1659         grant_ref_t gref[XENBUS_MAX_RING_GRANTS];
1660
1661         for (i = 0; i < info->nr_ring_pages; i++)
1662                 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1663
1664         sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
1665                                                        get_order(ring_size));
1666         if (!sring) {
1667                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1668                 return -ENOMEM;
1669         }
1670         SHARED_RING_INIT(sring);
1671         FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1672
1673         err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref);
1674         if (err < 0) {
1675                 free_pages((unsigned long)sring, get_order(ring_size));
1676                 rinfo->ring.sring = NULL;
1677                 goto fail;
1678         }
1679         for (i = 0; i < info->nr_ring_pages; i++)
1680                 rinfo->ring_ref[i] = gref[i];
1681
1682         err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1683         if (err)
1684                 goto fail;
1685
1686         err = bind_evtchn_to_irqhandler(rinfo->evtchn, blkif_interrupt, 0,
1687                                         "blkif", rinfo);
1688         if (err <= 0) {
1689                 xenbus_dev_fatal(dev, err,
1690                                  "bind_evtchn_to_irqhandler failed");
1691                 goto fail;
1692         }
1693         rinfo->irq = err;
1694
1695         return 0;
1696 fail:
1697         blkif_free(info, 0);
1698         return err;
1699 }
1700
1701 /*
1702  * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1703  * ring buffer may have multi pages depending on ->nr_ring_pages.
1704  */
1705 static int write_per_ring_nodes(struct xenbus_transaction xbt,
1706                                 struct blkfront_ring_info *rinfo, const char *dir)
1707 {
1708         int err;
1709         unsigned int i;
1710         const char *message = NULL;
1711         struct blkfront_info *info = rinfo->dev_info;
1712
1713         if (info->nr_ring_pages == 1) {
1714                 err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1715                 if (err) {
1716                         message = "writing ring-ref";
1717                         goto abort_transaction;
1718                 }
1719         } else {
1720                 for (i = 0; i < info->nr_ring_pages; i++) {
1721                         char ring_ref_name[RINGREF_NAME_LEN];
1722
1723                         snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1724                         err = xenbus_printf(xbt, dir, ring_ref_name,
1725                                             "%u", rinfo->ring_ref[i]);
1726                         if (err) {
1727                                 message = "writing ring-ref";
1728                                 goto abort_transaction;
1729                         }
1730                 }
1731         }
1732
1733         err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1734         if (err) {
1735                 message = "writing event-channel";
1736                 goto abort_transaction;
1737         }
1738
1739         return 0;
1740
1741 abort_transaction:
1742         xenbus_transaction_end(xbt, 1);
1743         if (message)
1744                 xenbus_dev_fatal(info->xbdev, err, "%s", message);
1745
1746         return err;
1747 }
1748
1749 /* Common code used when first setting up, and when resuming. */
1750 static int talk_to_blkback(struct xenbus_device *dev,
1751                            struct blkfront_info *info)
1752 {
1753         const char *message = NULL;
1754         struct xenbus_transaction xbt;
1755         int err;
1756         unsigned int i, max_page_order = 0;
1757         unsigned int ring_page_order = 0;
1758
1759         err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1760                            "max-ring-page-order", "%u", &max_page_order);
1761         if (err != 1)
1762                 info->nr_ring_pages = 1;
1763         else {
1764                 ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1765                 info->nr_ring_pages = 1 << ring_page_order;
1766         }
1767
1768         for (i = 0; i < info->nr_rings; i++) {
1769                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1770
1771                 /* Create shared ring, alloc event channel. */
1772                 err = setup_blkring(dev, rinfo);
1773                 if (err)
1774                         goto destroy_blkring;
1775         }
1776
1777 again:
1778         err = xenbus_transaction_start(&xbt);
1779         if (err) {
1780                 xenbus_dev_fatal(dev, err, "starting transaction");
1781                 goto destroy_blkring;
1782         }
1783
1784         if (info->nr_ring_pages > 1) {
1785                 err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1786                                     ring_page_order);
1787                 if (err) {
1788                         message = "writing ring-page-order";
1789                         goto abort_transaction;
1790                 }
1791         }
1792
1793         /* We already got the number of queues/rings in _probe */
1794         if (info->nr_rings == 1) {
1795                 err = write_per_ring_nodes(xbt, &info->rinfo[0], dev->nodename);
1796                 if (err)
1797                         goto destroy_blkring;
1798         } else {
1799                 char *path;
1800                 size_t pathsize;
1801
1802                 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1803                                     info->nr_rings);
1804                 if (err) {
1805                         message = "writing multi-queue-num-queues";
1806                         goto abort_transaction;
1807                 }
1808
1809                 pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1810                 path = kmalloc(pathsize, GFP_KERNEL);
1811                 if (!path) {
1812                         err = -ENOMEM;
1813                         message = "ENOMEM while writing ring references";
1814                         goto abort_transaction;
1815                 }
1816
1817                 for (i = 0; i < info->nr_rings; i++) {
1818                         memset(path, 0, pathsize);
1819                         snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1820                         err = write_per_ring_nodes(xbt, &info->rinfo[i], path);
1821                         if (err) {
1822                                 kfree(path);
1823                                 goto destroy_blkring;
1824                         }
1825                 }
1826                 kfree(path);
1827         }
1828         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1829                             XEN_IO_PROTO_ABI_NATIVE);
1830         if (err) {
1831                 message = "writing protocol";
1832                 goto abort_transaction;
1833         }
1834         err = xenbus_printf(xbt, dev->nodename,
1835                             "feature-persistent", "%u", 1);
1836         if (err)
1837                 dev_warn(&dev->dev,
1838                          "writing persistent grants feature to xenbus");
1839
1840         err = xenbus_transaction_end(xbt, 0);
1841         if (err) {
1842                 if (err == -EAGAIN)
1843                         goto again;
1844                 xenbus_dev_fatal(dev, err, "completing transaction");
1845                 goto destroy_blkring;
1846         }
1847
1848         for (i = 0; i < info->nr_rings; i++) {
1849                 unsigned int j;
1850                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1851
1852                 for (j = 0; j < BLK_RING_SIZE(info); j++)
1853                         rinfo->shadow[j].req.u.rw.id = j + 1;
1854                 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1855         }
1856         xenbus_switch_state(dev, XenbusStateInitialised);
1857
1858         return 0;
1859
1860  abort_transaction:
1861         xenbus_transaction_end(xbt, 1);
1862         if (message)
1863                 xenbus_dev_fatal(dev, err, "%s", message);
1864  destroy_blkring:
1865         blkif_free(info, 0);
1866
1867         kfree(info);
1868         dev_set_drvdata(&dev->dev, NULL);
1869
1870         return err;
1871 }
1872
1873 static int negotiate_mq(struct blkfront_info *info)
1874 {
1875         unsigned int backend_max_queues = 0;
1876         int err;
1877         unsigned int i;
1878
1879         BUG_ON(info->nr_rings);
1880
1881         /* Check if backend supports multiple queues. */
1882         err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1883                            "multi-queue-max-queues", "%u", &backend_max_queues);
1884         if (err < 0)
1885                 backend_max_queues = 1;
1886
1887         info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1888         /* We need at least one ring. */
1889         if (!info->nr_rings)
1890                 info->nr_rings = 1;
1891
1892         info->rinfo = kzalloc(sizeof(struct blkfront_ring_info) * info->nr_rings, GFP_KERNEL);
1893         if (!info->rinfo) {
1894                 xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1895                 return -ENOMEM;
1896         }
1897
1898         for (i = 0; i < info->nr_rings; i++) {
1899                 struct blkfront_ring_info *rinfo;
1900
1901                 rinfo = &info->rinfo[i];
1902                 INIT_LIST_HEAD(&rinfo->indirect_pages);
1903                 INIT_LIST_HEAD(&rinfo->grants);
1904                 rinfo->dev_info = info;
1905                 INIT_WORK(&rinfo->work, blkif_restart_queue);
1906                 spin_lock_init(&rinfo->ring_lock);
1907         }
1908         return 0;
1909 }
1910 /**
1911  * Entry point to this code when a new device is created.  Allocate the basic
1912  * structures and the ring buffer for communication with the backend, and
1913  * inform the backend of the appropriate details for those.  Switch to
1914  * Initialised state.
1915  */
1916 static int blkfront_probe(struct xenbus_device *dev,
1917                           const struct xenbus_device_id *id)
1918 {
1919         int err, vdevice;
1920         struct blkfront_info *info;
1921
1922         /* FIXME: Use dynamic device id if this is not set. */
1923         err = xenbus_scanf(XBT_NIL, dev->nodename,
1924                            "virtual-device", "%i", &vdevice);
1925         if (err != 1) {
1926                 /* go looking in the extended area instead */
1927                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1928                                    "%i", &vdevice);
1929                 if (err != 1) {
1930                         xenbus_dev_fatal(dev, err, "reading virtual-device");
1931                         return err;
1932                 }
1933         }
1934
1935         if (xen_hvm_domain()) {
1936                 char *type;
1937                 int len;
1938                 /* no unplug has been done: do not hook devices != xen vbds */
1939                 if (xen_has_pv_and_legacy_disk_devices()) {
1940                         int major;
1941
1942                         if (!VDEV_IS_EXTENDED(vdevice))
1943                                 major = BLKIF_MAJOR(vdevice);
1944                         else
1945                                 major = XENVBD_MAJOR;
1946
1947                         if (major != XENVBD_MAJOR) {
1948                                 printk(KERN_INFO
1949                                                 "%s: HVM does not support vbd %d as xen block device\n",
1950                                                 __func__, vdevice);
1951                                 return -ENODEV;
1952                         }
1953                 }
1954                 /* do not create a PV cdrom device if we are an HVM guest */
1955                 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1956                 if (IS_ERR(type))
1957                         return -ENODEV;
1958                 if (strncmp(type, "cdrom", 5) == 0) {
1959                         kfree(type);
1960                         return -ENODEV;
1961                 }
1962                 kfree(type);
1963         }
1964         info = kzalloc(sizeof(*info), GFP_KERNEL);
1965         if (!info) {
1966                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1967                 return -ENOMEM;
1968         }
1969
1970         info->xbdev = dev;
1971         err = negotiate_mq(info);
1972         if (err) {
1973                 kfree(info);
1974                 return err;
1975         }
1976
1977         mutex_init(&info->mutex);
1978         info->vdevice = vdevice;
1979         info->connected = BLKIF_STATE_DISCONNECTED;
1980
1981         /* Front end dir is a number, which is used as the id. */
1982         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1983         dev_set_drvdata(&dev->dev, info);
1984
1985         return 0;
1986 }
1987
1988 static void split_bio_end(struct bio *bio)
1989 {
1990         struct split_bio *split_bio = bio->bi_private;
1991
1992         if (atomic_dec_and_test(&split_bio->pending)) {
1993                 split_bio->bio->bi_phys_segments = 0;
1994                 split_bio->bio->bi_error = bio->bi_error;
1995                 bio_endio(split_bio->bio);
1996                 kfree(split_bio);
1997         }
1998         bio_put(bio);
1999 }
2000
2001 static int blkif_recover(struct blkfront_info *info)
2002 {
2003         unsigned int i, r_index;
2004         struct request *req, *n;
2005         struct blk_shadow *copy;
2006         int rc;
2007         struct bio *bio, *cloned_bio;
2008         struct bio_list bio_list, merge_bio;
2009         unsigned int segs, offset;
2010         int pending, size;
2011         struct split_bio *split_bio;
2012         struct list_head requests;
2013
2014         blkfront_gather_backend_features(info);
2015         segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
2016         blk_queue_max_segments(info->rq, segs);
2017         bio_list_init(&bio_list);
2018         INIT_LIST_HEAD(&requests);
2019
2020         for (r_index = 0; r_index < info->nr_rings; r_index++) {
2021                 struct blkfront_ring_info *rinfo;
2022
2023                 rinfo = &info->rinfo[r_index];
2024                 /* Stage 1: Make a safe copy of the shadow state. */
2025                 copy = kmemdup(rinfo->shadow, sizeof(rinfo->shadow),
2026                                GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
2027                 if (!copy)
2028                         return -ENOMEM;
2029
2030                 /* Stage 2: Set up free list. */
2031                 memset(&rinfo->shadow, 0, sizeof(rinfo->shadow));
2032                 for (i = 0; i < BLK_RING_SIZE(info); i++)
2033                         rinfo->shadow[i].req.u.rw.id = i+1;
2034                 rinfo->shadow_free = rinfo->ring.req_prod_pvt;
2035                 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
2036
2037                 rc = blkfront_setup_indirect(rinfo);
2038                 if (rc) {
2039                         kfree(copy);
2040                         return rc;
2041                 }
2042
2043                 for (i = 0; i < BLK_RING_SIZE(info); i++) {
2044                         /* Not in use? */
2045                         if (!copy[i].request)
2046                                 continue;
2047
2048                         /*
2049                          * Get the bios in the request so we can re-queue them.
2050                          */
2051                         if (copy[i].request->cmd_flags &
2052                             (REQ_FLUSH | REQ_FUA | REQ_DISCARD | REQ_SECURE)) {
2053                                 /*
2054                                  * Flush operations don't contain bios, so
2055                                  * we need to requeue the whole request
2056                                  */
2057                                 list_add(&copy[i].request->queuelist, &requests);
2058                                 continue;
2059                         }
2060                         merge_bio.head = copy[i].request->bio;
2061                         merge_bio.tail = copy[i].request->biotail;
2062                         bio_list_merge(&bio_list, &merge_bio);
2063                         copy[i].request->bio = NULL;
2064                         blk_end_request_all(copy[i].request, 0);
2065                 }
2066
2067                 kfree(copy);
2068         }
2069         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2070
2071         /* Now safe for us to use the shared ring */
2072         info->connected = BLKIF_STATE_CONNECTED;
2073
2074         for (r_index = 0; r_index < info->nr_rings; r_index++) {
2075                 struct blkfront_ring_info *rinfo;
2076
2077                 rinfo = &info->rinfo[r_index];
2078                 /* Kick any other new requests queued since we resumed */
2079                 kick_pending_request_queues(rinfo);
2080         }
2081
2082         list_for_each_entry_safe(req, n, &requests, queuelist) {
2083                 /* Requeue pending requests (flush or discard) */
2084                 list_del_init(&req->queuelist);
2085                 BUG_ON(req->nr_phys_segments > segs);
2086                 blk_mq_requeue_request(req);
2087         }
2088         blk_mq_kick_requeue_list(info->rq);
2089
2090         while ((bio = bio_list_pop(&bio_list)) != NULL) {
2091                 /* Traverse the list of pending bios and re-queue them */
2092                 if (bio_segments(bio) > segs) {
2093                         /*
2094                          * This bio has more segments than what we can
2095                          * handle, we have to split it.
2096                          */
2097                         pending = (bio_segments(bio) + segs - 1) / segs;
2098                         split_bio = kzalloc(sizeof(*split_bio), GFP_NOIO);
2099                         BUG_ON(split_bio == NULL);
2100                         atomic_set(&split_bio->pending, pending);
2101                         split_bio->bio = bio;
2102                         for (i = 0; i < pending; i++) {
2103                                 offset = (i * segs * XEN_PAGE_SIZE) >> 9;
2104                                 size = min((unsigned int)(segs * XEN_PAGE_SIZE) >> 9,
2105                                            (unsigned int)bio_sectors(bio) - offset);
2106                                 cloned_bio = bio_clone(bio, GFP_NOIO);
2107                                 BUG_ON(cloned_bio == NULL);
2108                                 bio_trim(cloned_bio, offset, size);
2109                                 cloned_bio->bi_private = split_bio;
2110                                 cloned_bio->bi_end_io = split_bio_end;
2111                                 submit_bio(cloned_bio->bi_rw, cloned_bio);
2112                         }
2113                         /*
2114                          * Now we have to wait for all those smaller bios to
2115                          * end, so we can also end the "parent" bio.
2116                          */
2117                         continue;
2118                 }
2119                 /* We don't need to split this bio */
2120                 submit_bio(bio->bi_rw, bio);
2121         }
2122
2123         return 0;
2124 }
2125
2126 /**
2127  * We are reconnecting to the backend, due to a suspend/resume, or a backend
2128  * driver restart.  We tear down our blkif structure and recreate it, but
2129  * leave the device-layer structures intact so that this is transparent to the
2130  * rest of the kernel.
2131  */
2132 static int blkfront_resume(struct xenbus_device *dev)
2133 {
2134         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2135         int err = 0;
2136
2137         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2138
2139         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2140
2141         err = negotiate_mq(info);
2142         if (err)
2143                 return err;
2144
2145         err = talk_to_blkback(dev, info);
2146         if (!err)
2147                 blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings);
2148
2149         /*
2150          * We have to wait for the backend to switch to
2151          * connected state, since we want to read which
2152          * features it supports.
2153          */
2154
2155         return err;
2156 }
2157
2158 static void blkfront_closing(struct blkfront_info *info)
2159 {
2160         struct xenbus_device *xbdev = info->xbdev;
2161         struct block_device *bdev = NULL;
2162
2163         mutex_lock(&info->mutex);
2164
2165         if (xbdev->state == XenbusStateClosing) {
2166                 mutex_unlock(&info->mutex);
2167                 return;
2168         }
2169
2170         if (info->gd)
2171                 bdev = bdget_disk(info->gd, 0);
2172
2173         mutex_unlock(&info->mutex);
2174
2175         if (!bdev) {
2176                 xenbus_frontend_closed(xbdev);
2177                 return;
2178         }
2179
2180         mutex_lock(&bdev->bd_mutex);
2181
2182         if (bdev->bd_openers) {
2183                 xenbus_dev_error(xbdev, -EBUSY,
2184                                  "Device in use; refusing to close");
2185                 xenbus_switch_state(xbdev, XenbusStateClosing);
2186         } else {
2187                 xlvbd_release_gendisk(info);
2188                 xenbus_frontend_closed(xbdev);
2189         }
2190
2191         mutex_unlock(&bdev->bd_mutex);
2192         bdput(bdev);
2193 }
2194
2195 static void blkfront_setup_discard(struct blkfront_info *info)
2196 {
2197         int err;
2198         unsigned int discard_granularity;
2199         unsigned int discard_alignment;
2200         unsigned int discard_secure;
2201
2202         info->feature_discard = 1;
2203         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2204                 "discard-granularity", "%u", &discard_granularity,
2205                 "discard-alignment", "%u", &discard_alignment,
2206                 NULL);
2207         if (!err) {
2208                 info->discard_granularity = discard_granularity;
2209                 info->discard_alignment = discard_alignment;
2210         }
2211         err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2212                            "discard-secure", "%u", &discard_secure);
2213         if (err > 0)
2214                 info->feature_secdiscard = !!discard_secure;
2215 }
2216
2217 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2218 {
2219         unsigned int psegs, grants;
2220         int err, i;
2221         struct blkfront_info *info = rinfo->dev_info;
2222
2223         if (info->max_indirect_segments == 0) {
2224                 if (!HAS_EXTRA_REQ)
2225                         grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2226                 else {
2227                         /*
2228                          * When an extra req is required, the maximum
2229                          * grants supported is related to the size of the
2230                          * Linux block segment.
2231                          */
2232                         grants = GRANTS_PER_PSEG;
2233                 }
2234         }
2235         else
2236                 grants = info->max_indirect_segments;
2237         psegs = grants / GRANTS_PER_PSEG;
2238
2239         err = fill_grant_buffer(rinfo,
2240                                 (grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2241         if (err)
2242                 goto out_of_memory;
2243
2244         if (!info->feature_persistent && info->max_indirect_segments) {
2245                 /*
2246                  * We are using indirect descriptors but not persistent
2247                  * grants, we need to allocate a set of pages that can be
2248                  * used for mapping indirect grefs
2249                  */
2250                 int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2251
2252                 BUG_ON(!list_empty(&rinfo->indirect_pages));
2253                 for (i = 0; i < num; i++) {
2254                         struct page *indirect_page = alloc_page(GFP_NOIO);
2255                         if (!indirect_page)
2256                                 goto out_of_memory;
2257                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
2258                 }
2259         }
2260
2261         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2262                 rinfo->shadow[i].grants_used = kzalloc(
2263                         sizeof(rinfo->shadow[i].grants_used[0]) * grants,
2264                         GFP_NOIO);
2265                 rinfo->shadow[i].sg = kzalloc(sizeof(rinfo->shadow[i].sg[0]) * psegs, GFP_NOIO);
2266                 if (info->max_indirect_segments)
2267                         rinfo->shadow[i].indirect_grants = kzalloc(
2268                                 sizeof(rinfo->shadow[i].indirect_grants[0]) *
2269                                 INDIRECT_GREFS(grants),
2270                                 GFP_NOIO);
2271                 if ((rinfo->shadow[i].grants_used == NULL) ||
2272                         (rinfo->shadow[i].sg == NULL) ||
2273                      (info->max_indirect_segments &&
2274                      (rinfo->shadow[i].indirect_grants == NULL)))
2275                         goto out_of_memory;
2276                 sg_init_table(rinfo->shadow[i].sg, psegs);
2277         }
2278
2279
2280         return 0;
2281
2282 out_of_memory:
2283         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2284                 kfree(rinfo->shadow[i].grants_used);
2285                 rinfo->shadow[i].grants_used = NULL;
2286                 kfree(rinfo->shadow[i].sg);
2287                 rinfo->shadow[i].sg = NULL;
2288                 kfree(rinfo->shadow[i].indirect_grants);
2289                 rinfo->shadow[i].indirect_grants = NULL;
2290         }
2291         if (!list_empty(&rinfo->indirect_pages)) {
2292                 struct page *indirect_page, *n;
2293                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2294                         list_del(&indirect_page->lru);
2295                         __free_page(indirect_page);
2296                 }
2297         }
2298         return -ENOMEM;
2299 }
2300
2301 /*
2302  * Gather all backend feature-*
2303  */
2304 static void blkfront_gather_backend_features(struct blkfront_info *info)
2305 {
2306         int err;
2307         int barrier, flush, discard, persistent;
2308         unsigned int indirect_segments;
2309
2310         info->feature_flush = 0;
2311
2312         err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2313                            "feature-barrier", "%d", &barrier);
2314
2315         /*
2316          * If there's no "feature-barrier" defined, then it means
2317          * we're dealing with a very old backend which writes
2318          * synchronously; nothing to do.
2319          *
2320          * If there are barriers, then we use flush.
2321          */
2322         if (err > 0 && barrier)
2323                 info->feature_flush = REQ_FLUSH | REQ_FUA;
2324         /*
2325          * And if there is "feature-flush-cache" use that above
2326          * barriers.
2327          */
2328         err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2329                            "feature-flush-cache", "%d", &flush);
2330
2331         if (err > 0 && flush)
2332                 info->feature_flush = REQ_FLUSH;
2333
2334         err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2335                            "feature-discard", "%d", &discard);
2336
2337         if (err > 0 && discard)
2338                 blkfront_setup_discard(info);
2339
2340         err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2341                            "feature-persistent", "%d", &persistent);
2342         if (err <= 0)
2343                 info->feature_persistent = 0;
2344         else
2345                 info->feature_persistent = persistent;
2346
2347         err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2348                            "feature-max-indirect-segments", "%u",
2349                            &indirect_segments);
2350         if (err <= 0)
2351                 info->max_indirect_segments = 0;
2352         else
2353                 info->max_indirect_segments = min(indirect_segments,
2354                                                   xen_blkif_max_segments);
2355 }
2356
2357 /*
2358  * Invoked when the backend is finally 'ready' (and has told produced
2359  * the details about the physical device - #sectors, size, etc).
2360  */
2361 static void blkfront_connect(struct blkfront_info *info)
2362 {
2363         unsigned long long sectors;
2364         unsigned long sector_size;
2365         unsigned int physical_sector_size;
2366         unsigned int binfo;
2367         int err, i;
2368
2369         switch (info->connected) {
2370         case BLKIF_STATE_CONNECTED:
2371                 /*
2372                  * Potentially, the back-end may be signalling
2373                  * a capacity change; update the capacity.
2374                  */
2375                 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2376                                    "sectors", "%Lu", &sectors);
2377                 if (XENBUS_EXIST_ERR(err))
2378                         return;
2379                 printk(KERN_INFO "Setting capacity to %Lu\n",
2380                        sectors);
2381                 set_capacity(info->gd, sectors);
2382                 revalidate_disk(info->gd);
2383
2384                 return;
2385         case BLKIF_STATE_SUSPENDED:
2386                 /*
2387                  * If we are recovering from suspension, we need to wait
2388                  * for the backend to announce it's features before
2389                  * reconnecting, at least we need to know if the backend
2390                  * supports indirect descriptors, and how many.
2391                  */
2392                 blkif_recover(info);
2393                 return;
2394
2395         default:
2396                 break;
2397         }
2398
2399         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2400                 __func__, info->xbdev->otherend);
2401
2402         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2403                             "sectors", "%llu", &sectors,
2404                             "info", "%u", &binfo,
2405                             "sector-size", "%lu", &sector_size,
2406                             NULL);
2407         if (err) {
2408                 xenbus_dev_fatal(info->xbdev, err,
2409                                  "reading backend fields at %s",
2410                                  info->xbdev->otherend);
2411                 return;
2412         }
2413
2414         /*
2415          * physcial-sector-size is a newer field, so old backends may not
2416          * provide this. Assume physical sector size to be the same as
2417          * sector_size in that case.
2418          */
2419         err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2420                            "physical-sector-size", "%u", &physical_sector_size);
2421         if (err != 1)
2422                 physical_sector_size = sector_size;
2423
2424         blkfront_gather_backend_features(info);
2425         for (i = 0; i < info->nr_rings; i++) {
2426                 err = blkfront_setup_indirect(&info->rinfo[i]);
2427                 if (err) {
2428                         xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2429                                          info->xbdev->otherend);
2430                         blkif_free(info, 0);
2431                         break;
2432                 }
2433         }
2434
2435         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
2436                                   physical_sector_size);
2437         if (err) {
2438                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2439                                  info->xbdev->otherend);
2440                 return;
2441         }
2442
2443         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2444
2445         /* Kick pending requests. */
2446         info->connected = BLKIF_STATE_CONNECTED;
2447         for (i = 0; i < info->nr_rings; i++)
2448                 kick_pending_request_queues(&info->rinfo[i]);
2449
2450         add_disk(info->gd);
2451
2452         info->is_ready = 1;
2453 }
2454
2455 /**
2456  * Callback received when the backend's state changes.
2457  */
2458 static void blkback_changed(struct xenbus_device *dev,
2459                             enum xenbus_state backend_state)
2460 {
2461         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2462
2463         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2464
2465         switch (backend_state) {
2466         case XenbusStateInitWait:
2467                 if (dev->state != XenbusStateInitialising)
2468                         break;
2469                 if (talk_to_blkback(dev, info))
2470                         break;
2471         case XenbusStateInitialising:
2472         case XenbusStateInitialised:
2473         case XenbusStateReconfiguring:
2474         case XenbusStateReconfigured:
2475         case XenbusStateUnknown:
2476                 break;
2477
2478         case XenbusStateConnected:
2479                 /*
2480                  * talk_to_blkback sets state to XenbusStateInitialised
2481                  * and blkfront_connect sets it to XenbusStateConnected
2482                  * (if connection went OK).
2483                  *
2484                  * If the backend (or toolstack) decides to poke at backend
2485                  * state (and re-trigger the watch by setting the state repeatedly
2486                  * to XenbusStateConnected (4)) we need to deal with this.
2487                  * This is allowed as this is used to communicate to the guest
2488                  * that the size of disk has changed!
2489                  */
2490                 if ((dev->state != XenbusStateInitialised) &&
2491                     (dev->state != XenbusStateConnected)) {
2492                         if (talk_to_blkback(dev, info))
2493                                 break;
2494                 }
2495
2496                 blkfront_connect(info);
2497                 break;
2498
2499         case XenbusStateClosed:
2500                 if (dev->state == XenbusStateClosed)
2501                         break;
2502                 /* Missed the backend's Closing state -- fallthrough */
2503         case XenbusStateClosing:
2504                 if (info)
2505                         blkfront_closing(info);
2506                 break;
2507         }
2508 }
2509
2510 static int blkfront_remove(struct xenbus_device *xbdev)
2511 {
2512         struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2513         struct block_device *bdev = NULL;
2514         struct gendisk *disk;
2515
2516         dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2517
2518         blkif_free(info, 0);
2519
2520         mutex_lock(&info->mutex);
2521
2522         disk = info->gd;
2523         if (disk)
2524                 bdev = bdget_disk(disk, 0);
2525
2526         info->xbdev = NULL;
2527         mutex_unlock(&info->mutex);
2528
2529         if (!bdev) {
2530                 kfree(info);
2531                 return 0;
2532         }
2533
2534         /*
2535          * The xbdev was removed before we reached the Closed
2536          * state. See if it's safe to remove the disk. If the bdev
2537          * isn't closed yet, we let release take care of it.
2538          */
2539
2540         mutex_lock(&bdev->bd_mutex);
2541         info = disk->private_data;
2542
2543         dev_warn(disk_to_dev(disk),
2544                  "%s was hot-unplugged, %d stale handles\n",
2545                  xbdev->nodename, bdev->bd_openers);
2546
2547         if (info && !bdev->bd_openers) {
2548                 xlvbd_release_gendisk(info);
2549                 disk->private_data = NULL;
2550                 kfree(info);
2551         }
2552
2553         mutex_unlock(&bdev->bd_mutex);
2554         bdput(bdev);
2555
2556         return 0;
2557 }
2558
2559 static int blkfront_is_ready(struct xenbus_device *dev)
2560 {
2561         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2562
2563         return info->is_ready && info->xbdev;
2564 }
2565
2566 static int blkif_open(struct block_device *bdev, fmode_t mode)
2567 {
2568         struct gendisk *disk = bdev->bd_disk;
2569         struct blkfront_info *info;
2570         int err = 0;
2571
2572         mutex_lock(&blkfront_mutex);
2573
2574         info = disk->private_data;
2575         if (!info) {
2576                 /* xbdev gone */
2577                 err = -ERESTARTSYS;
2578                 goto out;
2579         }
2580
2581         mutex_lock(&info->mutex);
2582
2583         if (!info->gd)
2584                 /* xbdev is closed */
2585                 err = -ERESTARTSYS;
2586
2587         mutex_unlock(&info->mutex);
2588
2589 out:
2590         mutex_unlock(&blkfront_mutex);
2591         return err;
2592 }
2593
2594 static void blkif_release(struct gendisk *disk, fmode_t mode)
2595 {
2596         struct blkfront_info *info = disk->private_data;
2597         struct block_device *bdev;
2598         struct xenbus_device *xbdev;
2599
2600         mutex_lock(&blkfront_mutex);
2601
2602         bdev = bdget_disk(disk, 0);
2603
2604         if (!bdev) {
2605                 WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2606                 goto out_mutex;
2607         }
2608         if (bdev->bd_openers)
2609                 goto out;
2610
2611         /*
2612          * Check if we have been instructed to close. We will have
2613          * deferred this request, because the bdev was still open.
2614          */
2615
2616         mutex_lock(&info->mutex);
2617         xbdev = info->xbdev;
2618
2619         if (xbdev && xbdev->state == XenbusStateClosing) {
2620                 /* pending switch to state closed */
2621                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2622                 xlvbd_release_gendisk(info);
2623                 xenbus_frontend_closed(info->xbdev);
2624         }
2625
2626         mutex_unlock(&info->mutex);
2627
2628         if (!xbdev) {
2629                 /* sudden device removal */
2630                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2631                 xlvbd_release_gendisk(info);
2632                 disk->private_data = NULL;
2633                 kfree(info);
2634         }
2635
2636 out:
2637         bdput(bdev);
2638 out_mutex:
2639         mutex_unlock(&blkfront_mutex);
2640 }
2641
2642 static const struct block_device_operations xlvbd_block_fops =
2643 {
2644         .owner = THIS_MODULE,
2645         .open = blkif_open,
2646         .release = blkif_release,
2647         .getgeo = blkif_getgeo,
2648         .ioctl = blkif_ioctl,
2649 };
2650
2651
2652 static const struct xenbus_device_id blkfront_ids[] = {
2653         { "vbd" },
2654         { "" }
2655 };
2656
2657 static struct xenbus_driver blkfront_driver = {
2658         .ids  = blkfront_ids,
2659         .probe = blkfront_probe,
2660         .remove = blkfront_remove,
2661         .resume = blkfront_resume,
2662         .otherend_changed = blkback_changed,
2663         .is_ready = blkfront_is_ready,
2664 };
2665
2666 static int __init xlblk_init(void)
2667 {
2668         int ret;
2669         int nr_cpus = num_online_cpus();
2670
2671         if (!xen_domain())
2672                 return -ENODEV;
2673
2674         if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2675                 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2676                         xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2677                 xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2678         }
2679
2680         if (xen_blkif_max_queues > nr_cpus) {
2681                 pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2682                         xen_blkif_max_queues, nr_cpus);
2683                 xen_blkif_max_queues = nr_cpus;
2684         }
2685
2686         if (!xen_has_pv_disk_devices())
2687                 return -ENODEV;
2688
2689         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2690                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
2691                        XENVBD_MAJOR, DEV_NAME);
2692                 return -ENODEV;
2693         }
2694
2695         ret = xenbus_register_frontend(&blkfront_driver);
2696         if (ret) {
2697                 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2698                 return ret;
2699         }
2700
2701         return 0;
2702 }
2703 module_init(xlblk_init);
2704
2705
2706 static void __exit xlblk_exit(void)
2707 {
2708         xenbus_unregister_driver(&blkfront_driver);
2709         unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2710         kfree(minors);
2711 }
2712 module_exit(xlblk_exit);
2713
2714 MODULE_DESCRIPTION("Xen virtual block device frontend");
2715 MODULE_LICENSE("GPL");
2716 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2717 MODULE_ALIAS("xen:vbd");
2718 MODULE_ALIAS("xenblk");