Merge tag 'for-linus-4.20a-rc2-tag' of git://git.kernel.org/pub/scm/linux/kernel...
[muen/linux.git] / drivers / block / xen-blkfront.c
1 /*
2  * blkfront.c
3  *
4  * XenLinux virtual block device driver.
5  *
6  * Copyright (c) 2003-2004, Keir Fraser & Steve Hand
7  * Modifications by Mark A. Williamson are (c) Intel Research Cambridge
8  * Copyright (c) 2004, Christian Limpach
9  * Copyright (c) 2004, Andrew Warfield
10  * Copyright (c) 2005, Christopher Clark
11  * Copyright (c) 2005, XenSource Ltd
12  *
13  * This program is free software; you can redistribute it and/or
14  * modify it under the terms of the GNU General Public License version 2
15  * as published by the Free Software Foundation; or, when distributed
16  * separately from the Linux kernel or incorporated into other
17  * software packages, subject to the following license:
18  *
19  * Permission is hereby granted, free of charge, to any person obtaining a copy
20  * of this source file (the "Software"), to deal in the Software without
21  * restriction, including without limitation the rights to use, copy, modify,
22  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
23  * and to permit persons to whom the Software is furnished to do so, subject to
24  * the following conditions:
25  *
26  * The above copyright notice and this permission notice shall be included in
27  * all copies or substantial portions of the Software.
28  *
29  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
30  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
31  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
32  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
33  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
34  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
35  * IN THE SOFTWARE.
36  */
37
38 #include <linux/interrupt.h>
39 #include <linux/blkdev.h>
40 #include <linux/blk-mq.h>
41 #include <linux/hdreg.h>
42 #include <linux/cdrom.h>
43 #include <linux/module.h>
44 #include <linux/slab.h>
45 #include <linux/mutex.h>
46 #include <linux/scatterlist.h>
47 #include <linux/bitmap.h>
48 #include <linux/list.h>
49 #include <linux/workqueue.h>
50
51 #include <xen/xen.h>
52 #include <xen/xenbus.h>
53 #include <xen/grant_table.h>
54 #include <xen/events.h>
55 #include <xen/page.h>
56 #include <xen/platform_pci.h>
57
58 #include <xen/interface/grant_table.h>
59 #include <xen/interface/io/blkif.h>
60 #include <xen/interface/io/protocols.h>
61
62 #include <asm/xen/hypervisor.h>
63
64 /*
65  * The minimal size of segment supported by the block framework is PAGE_SIZE.
66  * When Linux is using a different page size than Xen, it may not be possible
67  * to put all the data in a single segment.
68  * This can happen when the backend doesn't support indirect descriptor and
69  * therefore the maximum amount of data that a request can carry is
70  * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE = 44KB
71  *
72  * Note that we only support one extra request. So the Linux page size
73  * should be <= ( 2 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) =
74  * 88KB.
75  */
76 #define HAS_EXTRA_REQ (BLKIF_MAX_SEGMENTS_PER_REQUEST < XEN_PFN_PER_PAGE)
77
78 enum blkif_state {
79         BLKIF_STATE_DISCONNECTED,
80         BLKIF_STATE_CONNECTED,
81         BLKIF_STATE_SUSPENDED,
82 };
83
84 struct grant {
85         grant_ref_t gref;
86         struct page *page;
87         struct list_head node;
88 };
89
90 enum blk_req_status {
91         REQ_WAITING,
92         REQ_DONE,
93         REQ_ERROR,
94         REQ_EOPNOTSUPP,
95 };
96
97 struct blk_shadow {
98         struct blkif_request req;
99         struct request *request;
100         struct grant **grants_used;
101         struct grant **indirect_grants;
102         struct scatterlist *sg;
103         unsigned int num_sg;
104         enum blk_req_status status;
105
106         #define NO_ASSOCIATED_ID ~0UL
107         /*
108          * Id of the sibling if we ever need 2 requests when handling a
109          * block I/O request
110          */
111         unsigned long associated_id;
112 };
113
114 struct blkif_req {
115         blk_status_t    error;
116 };
117
118 static inline struct blkif_req *blkif_req(struct request *rq)
119 {
120         return blk_mq_rq_to_pdu(rq);
121 }
122
123 static DEFINE_MUTEX(blkfront_mutex);
124 static const struct block_device_operations xlvbd_block_fops;
125 static struct delayed_work blkfront_work;
126 static LIST_HEAD(info_list);
127
128 /*
129  * Maximum number of segments in indirect requests, the actual value used by
130  * the frontend driver is the minimum of this value and the value provided
131  * by the backend driver.
132  */
133
134 static unsigned int xen_blkif_max_segments = 32;
135 module_param_named(max_indirect_segments, xen_blkif_max_segments, uint, 0444);
136 MODULE_PARM_DESC(max_indirect_segments,
137                  "Maximum amount of segments in indirect requests (default is 32)");
138
139 static unsigned int xen_blkif_max_queues = 4;
140 module_param_named(max_queues, xen_blkif_max_queues, uint, 0444);
141 MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
142
143 /*
144  * Maximum order of pages to be used for the shared ring between front and
145  * backend, 4KB page granularity is used.
146  */
147 static unsigned int xen_blkif_max_ring_order;
148 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, 0444);
149 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
150
151 #define BLK_RING_SIZE(info)     \
152         __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
153
154 #define BLK_MAX_RING_SIZE       \
155         __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * XENBUS_MAX_RING_GRANTS)
156
157 /*
158  * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
159  * characters are enough. Define to 20 to keep consistent with backend.
160  */
161 #define RINGREF_NAME_LEN (20)
162 /*
163  * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
164  */
165 #define QUEUE_NAME_LEN (17)
166
167 /*
168  *  Per-ring info.
169  *  Every blkfront device can associate with one or more blkfront_ring_info,
170  *  depending on how many hardware queues/rings to be used.
171  */
172 struct blkfront_ring_info {
173         /* Lock to protect data in every ring buffer. */
174         spinlock_t ring_lock;
175         struct blkif_front_ring ring;
176         unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
177         unsigned int evtchn, irq;
178         struct work_struct work;
179         struct gnttab_free_callback callback;
180         struct blk_shadow shadow[BLK_MAX_RING_SIZE];
181         struct list_head indirect_pages;
182         struct list_head grants;
183         unsigned int persistent_gnts_c;
184         unsigned long shadow_free;
185         struct blkfront_info *dev_info;
186 };
187
188 /*
189  * We have one of these per vbd, whether ide, scsi or 'other'.  They
190  * hang in private_data off the gendisk structure. We may end up
191  * putting all kinds of interesting stuff here :-)
192  */
193 struct blkfront_info
194 {
195         struct mutex mutex;
196         struct xenbus_device *xbdev;
197         struct gendisk *gd;
198         u16 sector_size;
199         unsigned int physical_sector_size;
200         int vdevice;
201         blkif_vdev_t handle;
202         enum blkif_state connected;
203         /* Number of pages per ring buffer. */
204         unsigned int nr_ring_pages;
205         struct request_queue *rq;
206         unsigned int feature_flush:1;
207         unsigned int feature_fua:1;
208         unsigned int feature_discard:1;
209         unsigned int feature_secdiscard:1;
210         unsigned int feature_persistent:1;
211         unsigned int discard_granularity;
212         unsigned int discard_alignment;
213         /* Number of 4KB segments handled */
214         unsigned int max_indirect_segments;
215         int is_ready;
216         struct blk_mq_tag_set tag_set;
217         struct blkfront_ring_info *rinfo;
218         unsigned int nr_rings;
219         /* Save uncomplete reqs and bios for migration. */
220         struct list_head requests;
221         struct bio_list bio_list;
222         struct list_head info_list;
223 };
224
225 static unsigned int nr_minors;
226 static unsigned long *minors;
227 static DEFINE_SPINLOCK(minor_lock);
228
229 #define GRANT_INVALID_REF       0
230
231 #define PARTS_PER_DISK          16
232 #define PARTS_PER_EXT_DISK      256
233
234 #define BLKIF_MAJOR(dev) ((dev)>>8)
235 #define BLKIF_MINOR(dev) ((dev) & 0xff)
236
237 #define EXT_SHIFT 28
238 #define EXTENDED (1<<EXT_SHIFT)
239 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
240 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
241 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
242 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
243 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
244 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
245
246 #define DEV_NAME        "xvd"   /* name in /dev */
247
248 /*
249  * Grants are always the same size as a Xen page (i.e 4KB).
250  * A physical segment is always the same size as a Linux page.
251  * Number of grants per physical segment
252  */
253 #define GRANTS_PER_PSEG (PAGE_SIZE / XEN_PAGE_SIZE)
254
255 #define GRANTS_PER_INDIRECT_FRAME \
256         (XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
257
258 #define INDIRECT_GREFS(_grants)         \
259         DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
260
261 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
262 static void blkfront_gather_backend_features(struct blkfront_info *info);
263 static int negotiate_mq(struct blkfront_info *info);
264
265 static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
266 {
267         unsigned long free = rinfo->shadow_free;
268
269         BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
270         rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
271         rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
272         return free;
273 }
274
275 static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
276                               unsigned long id)
277 {
278         if (rinfo->shadow[id].req.u.rw.id != id)
279                 return -EINVAL;
280         if (rinfo->shadow[id].request == NULL)
281                 return -EINVAL;
282         rinfo->shadow[id].req.u.rw.id  = rinfo->shadow_free;
283         rinfo->shadow[id].request = NULL;
284         rinfo->shadow_free = id;
285         return 0;
286 }
287
288 static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
289 {
290         struct blkfront_info *info = rinfo->dev_info;
291         struct page *granted_page;
292         struct grant *gnt_list_entry, *n;
293         int i = 0;
294
295         while (i < num) {
296                 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
297                 if (!gnt_list_entry)
298                         goto out_of_memory;
299
300                 if (info->feature_persistent) {
301                         granted_page = alloc_page(GFP_NOIO);
302                         if (!granted_page) {
303                                 kfree(gnt_list_entry);
304                                 goto out_of_memory;
305                         }
306                         gnt_list_entry->page = granted_page;
307                 }
308
309                 gnt_list_entry->gref = GRANT_INVALID_REF;
310                 list_add(&gnt_list_entry->node, &rinfo->grants);
311                 i++;
312         }
313
314         return 0;
315
316 out_of_memory:
317         list_for_each_entry_safe(gnt_list_entry, n,
318                                  &rinfo->grants, node) {
319                 list_del(&gnt_list_entry->node);
320                 if (info->feature_persistent)
321                         __free_page(gnt_list_entry->page);
322                 kfree(gnt_list_entry);
323                 i--;
324         }
325         BUG_ON(i != 0);
326         return -ENOMEM;
327 }
328
329 static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
330 {
331         struct grant *gnt_list_entry;
332
333         BUG_ON(list_empty(&rinfo->grants));
334         gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
335                                           node);
336         list_del(&gnt_list_entry->node);
337
338         if (gnt_list_entry->gref != GRANT_INVALID_REF)
339                 rinfo->persistent_gnts_c--;
340
341         return gnt_list_entry;
342 }
343
344 static inline void grant_foreign_access(const struct grant *gnt_list_entry,
345                                         const struct blkfront_info *info)
346 {
347         gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
348                                                  info->xbdev->otherend_id,
349                                                  gnt_list_entry->page,
350                                                  0);
351 }
352
353 static struct grant *get_grant(grant_ref_t *gref_head,
354                                unsigned long gfn,
355                                struct blkfront_ring_info *rinfo)
356 {
357         struct grant *gnt_list_entry = get_free_grant(rinfo);
358         struct blkfront_info *info = rinfo->dev_info;
359
360         if (gnt_list_entry->gref != GRANT_INVALID_REF)
361                 return gnt_list_entry;
362
363         /* Assign a gref to this page */
364         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
365         BUG_ON(gnt_list_entry->gref == -ENOSPC);
366         if (info->feature_persistent)
367                 grant_foreign_access(gnt_list_entry, info);
368         else {
369                 /* Grant access to the GFN passed by the caller */
370                 gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
371                                                 info->xbdev->otherend_id,
372                                                 gfn, 0);
373         }
374
375         return gnt_list_entry;
376 }
377
378 static struct grant *get_indirect_grant(grant_ref_t *gref_head,
379                                         struct blkfront_ring_info *rinfo)
380 {
381         struct grant *gnt_list_entry = get_free_grant(rinfo);
382         struct blkfront_info *info = rinfo->dev_info;
383
384         if (gnt_list_entry->gref != GRANT_INVALID_REF)
385                 return gnt_list_entry;
386
387         /* Assign a gref to this page */
388         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
389         BUG_ON(gnt_list_entry->gref == -ENOSPC);
390         if (!info->feature_persistent) {
391                 struct page *indirect_page;
392
393                 /* Fetch a pre-allocated page to use for indirect grefs */
394                 BUG_ON(list_empty(&rinfo->indirect_pages));
395                 indirect_page = list_first_entry(&rinfo->indirect_pages,
396                                                  struct page, lru);
397                 list_del(&indirect_page->lru);
398                 gnt_list_entry->page = indirect_page;
399         }
400         grant_foreign_access(gnt_list_entry, info);
401
402         return gnt_list_entry;
403 }
404
405 static const char *op_name(int op)
406 {
407         static const char *const names[] = {
408                 [BLKIF_OP_READ] = "read",
409                 [BLKIF_OP_WRITE] = "write",
410                 [BLKIF_OP_WRITE_BARRIER] = "barrier",
411                 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
412                 [BLKIF_OP_DISCARD] = "discard" };
413
414         if (op < 0 || op >= ARRAY_SIZE(names))
415                 return "unknown";
416
417         if (!names[op])
418                 return "reserved";
419
420         return names[op];
421 }
422 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
423 {
424         unsigned int end = minor + nr;
425         int rc;
426
427         if (end > nr_minors) {
428                 unsigned long *bitmap, *old;
429
430                 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
431                                  GFP_KERNEL);
432                 if (bitmap == NULL)
433                         return -ENOMEM;
434
435                 spin_lock(&minor_lock);
436                 if (end > nr_minors) {
437                         old = minors;
438                         memcpy(bitmap, minors,
439                                BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
440                         minors = bitmap;
441                         nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
442                 } else
443                         old = bitmap;
444                 spin_unlock(&minor_lock);
445                 kfree(old);
446         }
447
448         spin_lock(&minor_lock);
449         if (find_next_bit(minors, end, minor) >= end) {
450                 bitmap_set(minors, minor, nr);
451                 rc = 0;
452         } else
453                 rc = -EBUSY;
454         spin_unlock(&minor_lock);
455
456         return rc;
457 }
458
459 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
460 {
461         unsigned int end = minor + nr;
462
463         BUG_ON(end > nr_minors);
464         spin_lock(&minor_lock);
465         bitmap_clear(minors,  minor, nr);
466         spin_unlock(&minor_lock);
467 }
468
469 static void blkif_restart_queue_callback(void *arg)
470 {
471         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
472         schedule_work(&rinfo->work);
473 }
474
475 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
476 {
477         /* We don't have real geometry info, but let's at least return
478            values consistent with the size of the device */
479         sector_t nsect = get_capacity(bd->bd_disk);
480         sector_t cylinders = nsect;
481
482         hg->heads = 0xff;
483         hg->sectors = 0x3f;
484         sector_div(cylinders, hg->heads * hg->sectors);
485         hg->cylinders = cylinders;
486         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
487                 hg->cylinders = 0xffff;
488         return 0;
489 }
490
491 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
492                        unsigned command, unsigned long argument)
493 {
494         struct blkfront_info *info = bdev->bd_disk->private_data;
495         int i;
496
497         dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
498                 command, (long)argument);
499
500         switch (command) {
501         case CDROMMULTISESSION:
502                 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
503                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
504                         if (put_user(0, (char __user *)(argument + i)))
505                                 return -EFAULT;
506                 return 0;
507
508         case CDROM_GET_CAPABILITY: {
509                 struct gendisk *gd = info->gd;
510                 if (gd->flags & GENHD_FL_CD)
511                         return 0;
512                 return -EINVAL;
513         }
514
515         default:
516                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
517                   command);*/
518                 return -EINVAL; /* same return as native Linux */
519         }
520
521         return 0;
522 }
523
524 static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
525                                             struct request *req,
526                                             struct blkif_request **ring_req)
527 {
528         unsigned long id;
529
530         *ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
531         rinfo->ring.req_prod_pvt++;
532
533         id = get_id_from_freelist(rinfo);
534         rinfo->shadow[id].request = req;
535         rinfo->shadow[id].status = REQ_WAITING;
536         rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
537
538         (*ring_req)->u.rw.id = id;
539
540         return id;
541 }
542
543 static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
544 {
545         struct blkfront_info *info = rinfo->dev_info;
546         struct blkif_request *ring_req;
547         unsigned long id;
548
549         /* Fill out a communications ring structure. */
550         id = blkif_ring_get_request(rinfo, req, &ring_req);
551
552         ring_req->operation = BLKIF_OP_DISCARD;
553         ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
554         ring_req->u.discard.id = id;
555         ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
556         if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard)
557                 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
558         else
559                 ring_req->u.discard.flag = 0;
560
561         /* Keep a private copy so we can reissue requests when recovering. */
562         rinfo->shadow[id].req = *ring_req;
563
564         return 0;
565 }
566
567 struct setup_rw_req {
568         unsigned int grant_idx;
569         struct blkif_request_segment *segments;
570         struct blkfront_ring_info *rinfo;
571         struct blkif_request *ring_req;
572         grant_ref_t gref_head;
573         unsigned int id;
574         /* Only used when persistent grant is used and it's a read request */
575         bool need_copy;
576         unsigned int bvec_off;
577         char *bvec_data;
578
579         bool require_extra_req;
580         struct blkif_request *extra_ring_req;
581 };
582
583 static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
584                                      unsigned int len, void *data)
585 {
586         struct setup_rw_req *setup = data;
587         int n, ref;
588         struct grant *gnt_list_entry;
589         unsigned int fsect, lsect;
590         /* Convenient aliases */
591         unsigned int grant_idx = setup->grant_idx;
592         struct blkif_request *ring_req = setup->ring_req;
593         struct blkfront_ring_info *rinfo = setup->rinfo;
594         /*
595          * We always use the shadow of the first request to store the list
596          * of grant associated to the block I/O request. This made the
597          * completion more easy to handle even if the block I/O request is
598          * split.
599          */
600         struct blk_shadow *shadow = &rinfo->shadow[setup->id];
601
602         if (unlikely(setup->require_extra_req &&
603                      grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
604                 /*
605                  * We are using the second request, setup grant_idx
606                  * to be the index of the segment array.
607                  */
608                 grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
609                 ring_req = setup->extra_ring_req;
610         }
611
612         if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
613             (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
614                 if (setup->segments)
615                         kunmap_atomic(setup->segments);
616
617                 n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
618                 gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
619                 shadow->indirect_grants[n] = gnt_list_entry;
620                 setup->segments = kmap_atomic(gnt_list_entry->page);
621                 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
622         }
623
624         gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
625         ref = gnt_list_entry->gref;
626         /*
627          * All the grants are stored in the shadow of the first
628          * request. Therefore we have to use the global index.
629          */
630         shadow->grants_used[setup->grant_idx] = gnt_list_entry;
631
632         if (setup->need_copy) {
633                 void *shared_data;
634
635                 shared_data = kmap_atomic(gnt_list_entry->page);
636                 /*
637                  * this does not wipe data stored outside the
638                  * range sg->offset..sg->offset+sg->length.
639                  * Therefore, blkback *could* see data from
640                  * previous requests. This is OK as long as
641                  * persistent grants are shared with just one
642                  * domain. It may need refactoring if this
643                  * changes
644                  */
645                 memcpy(shared_data + offset,
646                        setup->bvec_data + setup->bvec_off,
647                        len);
648
649                 kunmap_atomic(shared_data);
650                 setup->bvec_off += len;
651         }
652
653         fsect = offset >> 9;
654         lsect = fsect + (len >> 9) - 1;
655         if (ring_req->operation != BLKIF_OP_INDIRECT) {
656                 ring_req->u.rw.seg[grant_idx] =
657                         (struct blkif_request_segment) {
658                                 .gref       = ref,
659                                 .first_sect = fsect,
660                                 .last_sect  = lsect };
661         } else {
662                 setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
663                         (struct blkif_request_segment) {
664                                 .gref       = ref,
665                                 .first_sect = fsect,
666                                 .last_sect  = lsect };
667         }
668
669         (setup->grant_idx)++;
670 }
671
672 static void blkif_setup_extra_req(struct blkif_request *first,
673                                   struct blkif_request *second)
674 {
675         uint16_t nr_segments = first->u.rw.nr_segments;
676
677         /*
678          * The second request is only present when the first request uses
679          * all its segments. It's always the continuity of the first one.
680          */
681         first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
682
683         second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
684         second->u.rw.sector_number = first->u.rw.sector_number +
685                 (BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
686
687         second->u.rw.handle = first->u.rw.handle;
688         second->operation = first->operation;
689 }
690
691 static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
692 {
693         struct blkfront_info *info = rinfo->dev_info;
694         struct blkif_request *ring_req, *extra_ring_req = NULL;
695         unsigned long id, extra_id = NO_ASSOCIATED_ID;
696         bool require_extra_req = false;
697         int i;
698         struct setup_rw_req setup = {
699                 .grant_idx = 0,
700                 .segments = NULL,
701                 .rinfo = rinfo,
702                 .need_copy = rq_data_dir(req) && info->feature_persistent,
703         };
704
705         /*
706          * Used to store if we are able to queue the request by just using
707          * existing persistent grants, or if we have to get new grants,
708          * as there are not sufficiently many free.
709          */
710         bool new_persistent_gnts = false;
711         struct scatterlist *sg;
712         int num_sg, max_grefs, num_grant;
713
714         max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
715         if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
716                 /*
717                  * If we are using indirect segments we need to account
718                  * for the indirect grefs used in the request.
719                  */
720                 max_grefs += INDIRECT_GREFS(max_grefs);
721
722         /* Check if we have enough persistent grants to allocate a requests */
723         if (rinfo->persistent_gnts_c < max_grefs) {
724                 new_persistent_gnts = true;
725
726                 if (gnttab_alloc_grant_references(
727                     max_grefs - rinfo->persistent_gnts_c,
728                     &setup.gref_head) < 0) {
729                         gnttab_request_free_callback(
730                                 &rinfo->callback,
731                                 blkif_restart_queue_callback,
732                                 rinfo,
733                                 max_grefs - rinfo->persistent_gnts_c);
734                         return 1;
735                 }
736         }
737
738         /* Fill out a communications ring structure. */
739         id = blkif_ring_get_request(rinfo, req, &ring_req);
740
741         num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
742         num_grant = 0;
743         /* Calculate the number of grant used */
744         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
745                num_grant += gnttab_count_grant(sg->offset, sg->length);
746
747         require_extra_req = info->max_indirect_segments == 0 &&
748                 num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
749         BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
750
751         rinfo->shadow[id].num_sg = num_sg;
752         if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
753             likely(!require_extra_req)) {
754                 /*
755                  * The indirect operation can only be a BLKIF_OP_READ or
756                  * BLKIF_OP_WRITE
757                  */
758                 BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA);
759                 ring_req->operation = BLKIF_OP_INDIRECT;
760                 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
761                         BLKIF_OP_WRITE : BLKIF_OP_READ;
762                 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
763                 ring_req->u.indirect.handle = info->handle;
764                 ring_req->u.indirect.nr_segments = num_grant;
765         } else {
766                 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
767                 ring_req->u.rw.handle = info->handle;
768                 ring_req->operation = rq_data_dir(req) ?
769                         BLKIF_OP_WRITE : BLKIF_OP_READ;
770                 if (req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA) {
771                         /*
772                          * Ideally we can do an unordered flush-to-disk.
773                          * In case the backend onlysupports barriers, use that.
774                          * A barrier request a superset of FUA, so we can
775                          * implement it the same way.  (It's also a FLUSH+FUA,
776                          * since it is guaranteed ordered WRT previous writes.)
777                          */
778                         if (info->feature_flush && info->feature_fua)
779                                 ring_req->operation =
780                                         BLKIF_OP_WRITE_BARRIER;
781                         else if (info->feature_flush)
782                                 ring_req->operation =
783                                         BLKIF_OP_FLUSH_DISKCACHE;
784                         else
785                                 ring_req->operation = 0;
786                 }
787                 ring_req->u.rw.nr_segments = num_grant;
788                 if (unlikely(require_extra_req)) {
789                         extra_id = blkif_ring_get_request(rinfo, req,
790                                                           &extra_ring_req);
791                         /*
792                          * Only the first request contains the scatter-gather
793                          * list.
794                          */
795                         rinfo->shadow[extra_id].num_sg = 0;
796
797                         blkif_setup_extra_req(ring_req, extra_ring_req);
798
799                         /* Link the 2 requests together */
800                         rinfo->shadow[extra_id].associated_id = id;
801                         rinfo->shadow[id].associated_id = extra_id;
802                 }
803         }
804
805         setup.ring_req = ring_req;
806         setup.id = id;
807
808         setup.require_extra_req = require_extra_req;
809         if (unlikely(require_extra_req))
810                 setup.extra_ring_req = extra_ring_req;
811
812         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
813                 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
814
815                 if (setup.need_copy) {
816                         setup.bvec_off = sg->offset;
817                         setup.bvec_data = kmap_atomic(sg_page(sg));
818                 }
819
820                 gnttab_foreach_grant_in_range(sg_page(sg),
821                                               sg->offset,
822                                               sg->length,
823                                               blkif_setup_rw_req_grant,
824                                               &setup);
825
826                 if (setup.need_copy)
827                         kunmap_atomic(setup.bvec_data);
828         }
829         if (setup.segments)
830                 kunmap_atomic(setup.segments);
831
832         /* Keep a private copy so we can reissue requests when recovering. */
833         rinfo->shadow[id].req = *ring_req;
834         if (unlikely(require_extra_req))
835                 rinfo->shadow[extra_id].req = *extra_ring_req;
836
837         if (new_persistent_gnts)
838                 gnttab_free_grant_references(setup.gref_head);
839
840         return 0;
841 }
842
843 /*
844  * Generate a Xen blkfront IO request from a blk layer request.  Reads
845  * and writes are handled as expected.
846  *
847  * @req: a request struct
848  */
849 static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
850 {
851         if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
852                 return 1;
853
854         if (unlikely(req_op(req) == REQ_OP_DISCARD ||
855                      req_op(req) == REQ_OP_SECURE_ERASE))
856                 return blkif_queue_discard_req(req, rinfo);
857         else
858                 return blkif_queue_rw_req(req, rinfo);
859 }
860
861 static inline void flush_requests(struct blkfront_ring_info *rinfo)
862 {
863         int notify;
864
865         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
866
867         if (notify)
868                 notify_remote_via_irq(rinfo->irq);
869 }
870
871 static inline bool blkif_request_flush_invalid(struct request *req,
872                                                struct blkfront_info *info)
873 {
874         return (blk_rq_is_passthrough(req) ||
875                 ((req_op(req) == REQ_OP_FLUSH) &&
876                  !info->feature_flush) ||
877                 ((req->cmd_flags & REQ_FUA) &&
878                  !info->feature_fua));
879 }
880
881 static blk_status_t blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
882                           const struct blk_mq_queue_data *qd)
883 {
884         unsigned long flags;
885         int qid = hctx->queue_num;
886         struct blkfront_info *info = hctx->queue->queuedata;
887         struct blkfront_ring_info *rinfo = NULL;
888
889         BUG_ON(info->nr_rings <= qid);
890         rinfo = &info->rinfo[qid];
891         blk_mq_start_request(qd->rq);
892         spin_lock_irqsave(&rinfo->ring_lock, flags);
893         if (RING_FULL(&rinfo->ring))
894                 goto out_busy;
895
896         if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
897                 goto out_err;
898
899         if (blkif_queue_request(qd->rq, rinfo))
900                 goto out_busy;
901
902         flush_requests(rinfo);
903         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
904         return BLK_STS_OK;
905
906 out_err:
907         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
908         return BLK_STS_IOERR;
909
910 out_busy:
911         blk_mq_stop_hw_queue(hctx);
912         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
913         return BLK_STS_DEV_RESOURCE;
914 }
915
916 static void blkif_complete_rq(struct request *rq)
917 {
918         blk_mq_end_request(rq, blkif_req(rq)->error);
919 }
920
921 static const struct blk_mq_ops blkfront_mq_ops = {
922         .queue_rq = blkif_queue_rq,
923         .complete = blkif_complete_rq,
924 };
925
926 static void blkif_set_queue_limits(struct blkfront_info *info)
927 {
928         struct request_queue *rq = info->rq;
929         struct gendisk *gd = info->gd;
930         unsigned int segments = info->max_indirect_segments ? :
931                                 BLKIF_MAX_SEGMENTS_PER_REQUEST;
932
933         blk_queue_flag_set(QUEUE_FLAG_VIRT, rq);
934
935         if (info->feature_discard) {
936                 blk_queue_flag_set(QUEUE_FLAG_DISCARD, rq);
937                 blk_queue_max_discard_sectors(rq, get_capacity(gd));
938                 rq->limits.discard_granularity = info->discard_granularity;
939                 rq->limits.discard_alignment = info->discard_alignment;
940                 if (info->feature_secdiscard)
941                         blk_queue_flag_set(QUEUE_FLAG_SECERASE, rq);
942         }
943
944         /* Hard sector size and max sectors impersonate the equiv. hardware. */
945         blk_queue_logical_block_size(rq, info->sector_size);
946         blk_queue_physical_block_size(rq, info->physical_sector_size);
947         blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
948
949         /* Each segment in a request is up to an aligned page in size. */
950         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
951         blk_queue_max_segment_size(rq, PAGE_SIZE);
952
953         /* Ensure a merged request will fit in a single I/O ring slot. */
954         blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
955
956         /* Make sure buffer addresses are sector-aligned. */
957         blk_queue_dma_alignment(rq, 511);
958 }
959
960 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
961                                 unsigned int physical_sector_size)
962 {
963         struct request_queue *rq;
964         struct blkfront_info *info = gd->private_data;
965
966         memset(&info->tag_set, 0, sizeof(info->tag_set));
967         info->tag_set.ops = &blkfront_mq_ops;
968         info->tag_set.nr_hw_queues = info->nr_rings;
969         if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
970                 /*
971                  * When indirect descriptior is not supported, the I/O request
972                  * will be split between multiple request in the ring.
973                  * To avoid problems when sending the request, divide by
974                  * 2 the depth of the queue.
975                  */
976                 info->tag_set.queue_depth =  BLK_RING_SIZE(info) / 2;
977         } else
978                 info->tag_set.queue_depth = BLK_RING_SIZE(info);
979         info->tag_set.numa_node = NUMA_NO_NODE;
980         info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
981         info->tag_set.cmd_size = sizeof(struct blkif_req);
982         info->tag_set.driver_data = info;
983
984         if (blk_mq_alloc_tag_set(&info->tag_set))
985                 return -EINVAL;
986         rq = blk_mq_init_queue(&info->tag_set);
987         if (IS_ERR(rq)) {
988                 blk_mq_free_tag_set(&info->tag_set);
989                 return PTR_ERR(rq);
990         }
991
992         rq->queuedata = info;
993         info->rq = gd->queue = rq;
994         info->gd = gd;
995         info->sector_size = sector_size;
996         info->physical_sector_size = physical_sector_size;
997         blkif_set_queue_limits(info);
998
999         return 0;
1000 }
1001
1002 static const char *flush_info(struct blkfront_info *info)
1003 {
1004         if (info->feature_flush && info->feature_fua)
1005                 return "barrier: enabled;";
1006         else if (info->feature_flush)
1007                 return "flush diskcache: enabled;";
1008         else
1009                 return "barrier or flush: disabled;";
1010 }
1011
1012 static void xlvbd_flush(struct blkfront_info *info)
1013 {
1014         blk_queue_write_cache(info->rq, info->feature_flush ? true : false,
1015                               info->feature_fua ? true : false);
1016         pr_info("blkfront: %s: %s %s %s %s %s\n",
1017                 info->gd->disk_name, flush_info(info),
1018                 "persistent grants:", info->feature_persistent ?
1019                 "enabled;" : "disabled;", "indirect descriptors:",
1020                 info->max_indirect_segments ? "enabled;" : "disabled;");
1021 }
1022
1023 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
1024 {
1025         int major;
1026         major = BLKIF_MAJOR(vdevice);
1027         *minor = BLKIF_MINOR(vdevice);
1028         switch (major) {
1029                 case XEN_IDE0_MAJOR:
1030                         *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
1031                         *minor = ((*minor / 64) * PARTS_PER_DISK) +
1032                                 EMULATED_HD_DISK_MINOR_OFFSET;
1033                         break;
1034                 case XEN_IDE1_MAJOR:
1035                         *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1036                         *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1037                                 EMULATED_HD_DISK_MINOR_OFFSET;
1038                         break;
1039                 case XEN_SCSI_DISK0_MAJOR:
1040                         *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1041                         *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1042                         break;
1043                 case XEN_SCSI_DISK1_MAJOR:
1044                 case XEN_SCSI_DISK2_MAJOR:
1045                 case XEN_SCSI_DISK3_MAJOR:
1046                 case XEN_SCSI_DISK4_MAJOR:
1047                 case XEN_SCSI_DISK5_MAJOR:
1048                 case XEN_SCSI_DISK6_MAJOR:
1049                 case XEN_SCSI_DISK7_MAJOR:
1050                         *offset = (*minor / PARTS_PER_DISK) + 
1051                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1052                                 EMULATED_SD_DISK_NAME_OFFSET;
1053                         *minor = *minor +
1054                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1055                                 EMULATED_SD_DISK_MINOR_OFFSET;
1056                         break;
1057                 case XEN_SCSI_DISK8_MAJOR:
1058                 case XEN_SCSI_DISK9_MAJOR:
1059                 case XEN_SCSI_DISK10_MAJOR:
1060                 case XEN_SCSI_DISK11_MAJOR:
1061                 case XEN_SCSI_DISK12_MAJOR:
1062                 case XEN_SCSI_DISK13_MAJOR:
1063                 case XEN_SCSI_DISK14_MAJOR:
1064                 case XEN_SCSI_DISK15_MAJOR:
1065                         *offset = (*minor / PARTS_PER_DISK) + 
1066                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1067                                 EMULATED_SD_DISK_NAME_OFFSET;
1068                         *minor = *minor +
1069                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1070                                 EMULATED_SD_DISK_MINOR_OFFSET;
1071                         break;
1072                 case XENVBD_MAJOR:
1073                         *offset = *minor / PARTS_PER_DISK;
1074                         break;
1075                 default:
1076                         printk(KERN_WARNING "blkfront: your disk configuration is "
1077                                         "incorrect, please use an xvd device instead\n");
1078                         return -ENODEV;
1079         }
1080         return 0;
1081 }
1082
1083 static char *encode_disk_name(char *ptr, unsigned int n)
1084 {
1085         if (n >= 26)
1086                 ptr = encode_disk_name(ptr, n / 26 - 1);
1087         *ptr = 'a' + n % 26;
1088         return ptr + 1;
1089 }
1090
1091 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1092                                struct blkfront_info *info,
1093                                u16 vdisk_info, u16 sector_size,
1094                                unsigned int physical_sector_size)
1095 {
1096         struct gendisk *gd;
1097         int nr_minors = 1;
1098         int err;
1099         unsigned int offset;
1100         int minor;
1101         int nr_parts;
1102         char *ptr;
1103
1104         BUG_ON(info->gd != NULL);
1105         BUG_ON(info->rq != NULL);
1106
1107         if ((info->vdevice>>EXT_SHIFT) > 1) {
1108                 /* this is above the extended range; something is wrong */
1109                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1110                 return -ENODEV;
1111         }
1112
1113         if (!VDEV_IS_EXTENDED(info->vdevice)) {
1114                 err = xen_translate_vdev(info->vdevice, &minor, &offset);
1115                 if (err)
1116                         return err;             
1117                 nr_parts = PARTS_PER_DISK;
1118         } else {
1119                 minor = BLKIF_MINOR_EXT(info->vdevice);
1120                 nr_parts = PARTS_PER_EXT_DISK;
1121                 offset = minor / nr_parts;
1122                 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1123                         printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1124                                         "emulated IDE disks,\n\t choose an xvd device name"
1125                                         "from xvde on\n", info->vdevice);
1126         }
1127         if (minor >> MINORBITS) {
1128                 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1129                         info->vdevice, minor);
1130                 return -ENODEV;
1131         }
1132
1133         if ((minor % nr_parts) == 0)
1134                 nr_minors = nr_parts;
1135
1136         err = xlbd_reserve_minors(minor, nr_minors);
1137         if (err)
1138                 goto out;
1139         err = -ENODEV;
1140
1141         gd = alloc_disk(nr_minors);
1142         if (gd == NULL)
1143                 goto release;
1144
1145         strcpy(gd->disk_name, DEV_NAME);
1146         ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1147         BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1148         if (nr_minors > 1)
1149                 *ptr = 0;
1150         else
1151                 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1152                          "%d", minor & (nr_parts - 1));
1153
1154         gd->major = XENVBD_MAJOR;
1155         gd->first_minor = minor;
1156         gd->fops = &xlvbd_block_fops;
1157         gd->private_data = info;
1158         set_capacity(gd, capacity);
1159
1160         if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size)) {
1161                 del_gendisk(gd);
1162                 goto release;
1163         }
1164
1165         xlvbd_flush(info);
1166
1167         if (vdisk_info & VDISK_READONLY)
1168                 set_disk_ro(gd, 1);
1169
1170         if (vdisk_info & VDISK_REMOVABLE)
1171                 gd->flags |= GENHD_FL_REMOVABLE;
1172
1173         if (vdisk_info & VDISK_CDROM)
1174                 gd->flags |= GENHD_FL_CD;
1175
1176         return 0;
1177
1178  release:
1179         xlbd_release_minors(minor, nr_minors);
1180  out:
1181         return err;
1182 }
1183
1184 static void xlvbd_release_gendisk(struct blkfront_info *info)
1185 {
1186         unsigned int minor, nr_minors, i;
1187
1188         if (info->rq == NULL)
1189                 return;
1190
1191         /* No more blkif_request(). */
1192         blk_mq_stop_hw_queues(info->rq);
1193
1194         for (i = 0; i < info->nr_rings; i++) {
1195                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1196
1197                 /* No more gnttab callback work. */
1198                 gnttab_cancel_free_callback(&rinfo->callback);
1199
1200                 /* Flush gnttab callback work. Must be done with no locks held. */
1201                 flush_work(&rinfo->work);
1202         }
1203
1204         del_gendisk(info->gd);
1205
1206         minor = info->gd->first_minor;
1207         nr_minors = info->gd->minors;
1208         xlbd_release_minors(minor, nr_minors);
1209
1210         blk_cleanup_queue(info->rq);
1211         blk_mq_free_tag_set(&info->tag_set);
1212         info->rq = NULL;
1213
1214         put_disk(info->gd);
1215         info->gd = NULL;
1216 }
1217
1218 /* Already hold rinfo->ring_lock. */
1219 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1220 {
1221         if (!RING_FULL(&rinfo->ring))
1222                 blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1223 }
1224
1225 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1226 {
1227         unsigned long flags;
1228
1229         spin_lock_irqsave(&rinfo->ring_lock, flags);
1230         kick_pending_request_queues_locked(rinfo);
1231         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1232 }
1233
1234 static void blkif_restart_queue(struct work_struct *work)
1235 {
1236         struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1237
1238         if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1239                 kick_pending_request_queues(rinfo);
1240 }
1241
1242 static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1243 {
1244         struct grant *persistent_gnt, *n;
1245         struct blkfront_info *info = rinfo->dev_info;
1246         int i, j, segs;
1247
1248         /*
1249          * Remove indirect pages, this only happens when using indirect
1250          * descriptors but not persistent grants
1251          */
1252         if (!list_empty(&rinfo->indirect_pages)) {
1253                 struct page *indirect_page, *n;
1254
1255                 BUG_ON(info->feature_persistent);
1256                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1257                         list_del(&indirect_page->lru);
1258                         __free_page(indirect_page);
1259                 }
1260         }
1261
1262         /* Remove all persistent grants. */
1263         if (!list_empty(&rinfo->grants)) {
1264                 list_for_each_entry_safe(persistent_gnt, n,
1265                                          &rinfo->grants, node) {
1266                         list_del(&persistent_gnt->node);
1267                         if (persistent_gnt->gref != GRANT_INVALID_REF) {
1268                                 gnttab_end_foreign_access(persistent_gnt->gref,
1269                                                           0, 0UL);
1270                                 rinfo->persistent_gnts_c--;
1271                         }
1272                         if (info->feature_persistent)
1273                                 __free_page(persistent_gnt->page);
1274                         kfree(persistent_gnt);
1275                 }
1276         }
1277         BUG_ON(rinfo->persistent_gnts_c != 0);
1278
1279         for (i = 0; i < BLK_RING_SIZE(info); i++) {
1280                 /*
1281                  * Clear persistent grants present in requests already
1282                  * on the shared ring
1283                  */
1284                 if (!rinfo->shadow[i].request)
1285                         goto free_shadow;
1286
1287                 segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1288                        rinfo->shadow[i].req.u.indirect.nr_segments :
1289                        rinfo->shadow[i].req.u.rw.nr_segments;
1290                 for (j = 0; j < segs; j++) {
1291                         persistent_gnt = rinfo->shadow[i].grants_used[j];
1292                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1293                         if (info->feature_persistent)
1294                                 __free_page(persistent_gnt->page);
1295                         kfree(persistent_gnt);
1296                 }
1297
1298                 if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1299                         /*
1300                          * If this is not an indirect operation don't try to
1301                          * free indirect segments
1302                          */
1303                         goto free_shadow;
1304
1305                 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1306                         persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1307                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1308                         __free_page(persistent_gnt->page);
1309                         kfree(persistent_gnt);
1310                 }
1311
1312 free_shadow:
1313                 kfree(rinfo->shadow[i].grants_used);
1314                 rinfo->shadow[i].grants_used = NULL;
1315                 kfree(rinfo->shadow[i].indirect_grants);
1316                 rinfo->shadow[i].indirect_grants = NULL;
1317                 kfree(rinfo->shadow[i].sg);
1318                 rinfo->shadow[i].sg = NULL;
1319         }
1320
1321         /* No more gnttab callback work. */
1322         gnttab_cancel_free_callback(&rinfo->callback);
1323
1324         /* Flush gnttab callback work. Must be done with no locks held. */
1325         flush_work(&rinfo->work);
1326
1327         /* Free resources associated with old device channel. */
1328         for (i = 0; i < info->nr_ring_pages; i++) {
1329                 if (rinfo->ring_ref[i] != GRANT_INVALID_REF) {
1330                         gnttab_end_foreign_access(rinfo->ring_ref[i], 0, 0);
1331                         rinfo->ring_ref[i] = GRANT_INVALID_REF;
1332                 }
1333         }
1334         free_pages((unsigned long)rinfo->ring.sring, get_order(info->nr_ring_pages * XEN_PAGE_SIZE));
1335         rinfo->ring.sring = NULL;
1336
1337         if (rinfo->irq)
1338                 unbind_from_irqhandler(rinfo->irq, rinfo);
1339         rinfo->evtchn = rinfo->irq = 0;
1340 }
1341
1342 static void blkif_free(struct blkfront_info *info, int suspend)
1343 {
1344         unsigned int i;
1345
1346         /* Prevent new requests being issued until we fix things up. */
1347         info->connected = suspend ?
1348                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1349         /* No more blkif_request(). */
1350         if (info->rq)
1351                 blk_mq_stop_hw_queues(info->rq);
1352
1353         for (i = 0; i < info->nr_rings; i++)
1354                 blkif_free_ring(&info->rinfo[i]);
1355
1356         kfree(info->rinfo);
1357         info->rinfo = NULL;
1358         info->nr_rings = 0;
1359 }
1360
1361 struct copy_from_grant {
1362         const struct blk_shadow *s;
1363         unsigned int grant_idx;
1364         unsigned int bvec_offset;
1365         char *bvec_data;
1366 };
1367
1368 static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1369                                   unsigned int len, void *data)
1370 {
1371         struct copy_from_grant *info = data;
1372         char *shared_data;
1373         /* Convenient aliases */
1374         const struct blk_shadow *s = info->s;
1375
1376         shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1377
1378         memcpy(info->bvec_data + info->bvec_offset,
1379                shared_data + offset, len);
1380
1381         info->bvec_offset += len;
1382         info->grant_idx++;
1383
1384         kunmap_atomic(shared_data);
1385 }
1386
1387 static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1388 {
1389         switch (rsp)
1390         {
1391         case BLKIF_RSP_OKAY:
1392                 return REQ_DONE;
1393         case BLKIF_RSP_EOPNOTSUPP:
1394                 return REQ_EOPNOTSUPP;
1395         case BLKIF_RSP_ERROR:
1396                 /* Fallthrough. */
1397         default:
1398                 return REQ_ERROR;
1399         }
1400 }
1401
1402 /*
1403  * Get the final status of the block request based on two ring response
1404  */
1405 static int blkif_get_final_status(enum blk_req_status s1,
1406                                   enum blk_req_status s2)
1407 {
1408         BUG_ON(s1 == REQ_WAITING);
1409         BUG_ON(s2 == REQ_WAITING);
1410
1411         if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1412                 return BLKIF_RSP_ERROR;
1413         else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1414                 return BLKIF_RSP_EOPNOTSUPP;
1415         return BLKIF_RSP_OKAY;
1416 }
1417
1418 static bool blkif_completion(unsigned long *id,
1419                              struct blkfront_ring_info *rinfo,
1420                              struct blkif_response *bret)
1421 {
1422         int i = 0;
1423         struct scatterlist *sg;
1424         int num_sg, num_grant;
1425         struct blkfront_info *info = rinfo->dev_info;
1426         struct blk_shadow *s = &rinfo->shadow[*id];
1427         struct copy_from_grant data = {
1428                 .grant_idx = 0,
1429         };
1430
1431         num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1432                 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1433
1434         /* The I/O request may be split in two. */
1435         if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1436                 struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1437
1438                 /* Keep the status of the current response in shadow. */
1439                 s->status = blkif_rsp_to_req_status(bret->status);
1440
1441                 /* Wait the second response if not yet here. */
1442                 if (s2->status == REQ_WAITING)
1443                         return false;
1444
1445                 bret->status = blkif_get_final_status(s->status,
1446                                                       s2->status);
1447
1448                 /*
1449                  * All the grants is stored in the first shadow in order
1450                  * to make the completion code simpler.
1451                  */
1452                 num_grant += s2->req.u.rw.nr_segments;
1453
1454                 /*
1455                  * The two responses may not come in order. Only the
1456                  * first request will store the scatter-gather list.
1457                  */
1458                 if (s2->num_sg != 0) {
1459                         /* Update "id" with the ID of the first response. */
1460                         *id = s->associated_id;
1461                         s = s2;
1462                 }
1463
1464                 /*
1465                  * We don't need anymore the second request, so recycling
1466                  * it now.
1467                  */
1468                 if (add_id_to_freelist(rinfo, s->associated_id))
1469                         WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1470                              info->gd->disk_name, s->associated_id);
1471         }
1472
1473         data.s = s;
1474         num_sg = s->num_sg;
1475
1476         if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1477                 for_each_sg(s->sg, sg, num_sg, i) {
1478                         BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1479
1480                         data.bvec_offset = sg->offset;
1481                         data.bvec_data = kmap_atomic(sg_page(sg));
1482
1483                         gnttab_foreach_grant_in_range(sg_page(sg),
1484                                                       sg->offset,
1485                                                       sg->length,
1486                                                       blkif_copy_from_grant,
1487                                                       &data);
1488
1489                         kunmap_atomic(data.bvec_data);
1490                 }
1491         }
1492         /* Add the persistent grant into the list of free grants */
1493         for (i = 0; i < num_grant; i++) {
1494                 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1495                         /*
1496                          * If the grant is still mapped by the backend (the
1497                          * backend has chosen to make this grant persistent)
1498                          * we add it at the head of the list, so it will be
1499                          * reused first.
1500                          */
1501                         if (!info->feature_persistent)
1502                                 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1503                                                      s->grants_used[i]->gref);
1504                         list_add(&s->grants_used[i]->node, &rinfo->grants);
1505                         rinfo->persistent_gnts_c++;
1506                 } else {
1507                         /*
1508                          * If the grant is not mapped by the backend we end the
1509                          * foreign access and add it to the tail of the list,
1510                          * so it will not be picked again unless we run out of
1511                          * persistent grants.
1512                          */
1513                         gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1514                         s->grants_used[i]->gref = GRANT_INVALID_REF;
1515                         list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1516                 }
1517         }
1518         if (s->req.operation == BLKIF_OP_INDIRECT) {
1519                 for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1520                         if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1521                                 if (!info->feature_persistent)
1522                                         pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1523                                                              s->indirect_grants[i]->gref);
1524                                 list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1525                                 rinfo->persistent_gnts_c++;
1526                         } else {
1527                                 struct page *indirect_page;
1528
1529                                 gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1530                                 /*
1531                                  * Add the used indirect page back to the list of
1532                                  * available pages for indirect grefs.
1533                                  */
1534                                 if (!info->feature_persistent) {
1535                                         indirect_page = s->indirect_grants[i]->page;
1536                                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
1537                                 }
1538                                 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1539                                 list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1540                         }
1541                 }
1542         }
1543
1544         return true;
1545 }
1546
1547 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1548 {
1549         struct request *req;
1550         struct blkif_response *bret;
1551         RING_IDX i, rp;
1552         unsigned long flags;
1553         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1554         struct blkfront_info *info = rinfo->dev_info;
1555
1556         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
1557                 return IRQ_HANDLED;
1558
1559         spin_lock_irqsave(&rinfo->ring_lock, flags);
1560  again:
1561         rp = rinfo->ring.sring->rsp_prod;
1562         rmb(); /* Ensure we see queued responses up to 'rp'. */
1563
1564         for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1565                 unsigned long id;
1566
1567                 bret = RING_GET_RESPONSE(&rinfo->ring, i);
1568                 id   = bret->id;
1569                 /*
1570                  * The backend has messed up and given us an id that we would
1571                  * never have given to it (we stamp it up to BLK_RING_SIZE -
1572                  * look in get_id_from_freelist.
1573                  */
1574                 if (id >= BLK_RING_SIZE(info)) {
1575                         WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1576                              info->gd->disk_name, op_name(bret->operation), id);
1577                         /* We can't safely get the 'struct request' as
1578                          * the id is busted. */
1579                         continue;
1580                 }
1581                 req  = rinfo->shadow[id].request;
1582
1583                 if (bret->operation != BLKIF_OP_DISCARD) {
1584                         /*
1585                          * We may need to wait for an extra response if the
1586                          * I/O request is split in 2
1587                          */
1588                         if (!blkif_completion(&id, rinfo, bret))
1589                                 continue;
1590                 }
1591
1592                 if (add_id_to_freelist(rinfo, id)) {
1593                         WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1594                              info->gd->disk_name, op_name(bret->operation), id);
1595                         continue;
1596                 }
1597
1598                 if (bret->status == BLKIF_RSP_OKAY)
1599                         blkif_req(req)->error = BLK_STS_OK;
1600                 else
1601                         blkif_req(req)->error = BLK_STS_IOERR;
1602
1603                 switch (bret->operation) {
1604                 case BLKIF_OP_DISCARD:
1605                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1606                                 struct request_queue *rq = info->rq;
1607                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1608                                            info->gd->disk_name, op_name(bret->operation));
1609                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1610                                 info->feature_discard = 0;
1611                                 info->feature_secdiscard = 0;
1612                                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1613                                 blk_queue_flag_clear(QUEUE_FLAG_SECERASE, rq);
1614                         }
1615                         break;
1616                 case BLKIF_OP_FLUSH_DISKCACHE:
1617                 case BLKIF_OP_WRITE_BARRIER:
1618                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1619                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1620                                        info->gd->disk_name, op_name(bret->operation));
1621                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1622                         }
1623                         if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1624                                      rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1625                                 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1626                                        info->gd->disk_name, op_name(bret->operation));
1627                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1628                         }
1629                         if (unlikely(blkif_req(req)->error)) {
1630                                 if (blkif_req(req)->error == BLK_STS_NOTSUPP)
1631                                         blkif_req(req)->error = BLK_STS_OK;
1632                                 info->feature_fua = 0;
1633                                 info->feature_flush = 0;
1634                                 xlvbd_flush(info);
1635                         }
1636                         /* fall through */
1637                 case BLKIF_OP_READ:
1638                 case BLKIF_OP_WRITE:
1639                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
1640                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1641                                         "request: %x\n", bret->status);
1642
1643                         break;
1644                 default:
1645                         BUG();
1646                 }
1647
1648                 blk_mq_complete_request(req);
1649         }
1650
1651         rinfo->ring.rsp_cons = i;
1652
1653         if (i != rinfo->ring.req_prod_pvt) {
1654                 int more_to_do;
1655                 RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1656                 if (more_to_do)
1657                         goto again;
1658         } else
1659                 rinfo->ring.sring->rsp_event = i + 1;
1660
1661         kick_pending_request_queues_locked(rinfo);
1662
1663         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1664
1665         return IRQ_HANDLED;
1666 }
1667
1668
1669 static int setup_blkring(struct xenbus_device *dev,
1670                          struct blkfront_ring_info *rinfo)
1671 {
1672         struct blkif_sring *sring;
1673         int err, i;
1674         struct blkfront_info *info = rinfo->dev_info;
1675         unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1676         grant_ref_t gref[XENBUS_MAX_RING_GRANTS];
1677
1678         for (i = 0; i < info->nr_ring_pages; i++)
1679                 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1680
1681         sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
1682                                                        get_order(ring_size));
1683         if (!sring) {
1684                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1685                 return -ENOMEM;
1686         }
1687         SHARED_RING_INIT(sring);
1688         FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1689
1690         err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref);
1691         if (err < 0) {
1692                 free_pages((unsigned long)sring, get_order(ring_size));
1693                 rinfo->ring.sring = NULL;
1694                 goto fail;
1695         }
1696         for (i = 0; i < info->nr_ring_pages; i++)
1697                 rinfo->ring_ref[i] = gref[i];
1698
1699         err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1700         if (err)
1701                 goto fail;
1702
1703         err = bind_evtchn_to_irqhandler(rinfo->evtchn, blkif_interrupt, 0,
1704                                         "blkif", rinfo);
1705         if (err <= 0) {
1706                 xenbus_dev_fatal(dev, err,
1707                                  "bind_evtchn_to_irqhandler failed");
1708                 goto fail;
1709         }
1710         rinfo->irq = err;
1711
1712         return 0;
1713 fail:
1714         blkif_free(info, 0);
1715         return err;
1716 }
1717
1718 /*
1719  * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1720  * ring buffer may have multi pages depending on ->nr_ring_pages.
1721  */
1722 static int write_per_ring_nodes(struct xenbus_transaction xbt,
1723                                 struct blkfront_ring_info *rinfo, const char *dir)
1724 {
1725         int err;
1726         unsigned int i;
1727         const char *message = NULL;
1728         struct blkfront_info *info = rinfo->dev_info;
1729
1730         if (info->nr_ring_pages == 1) {
1731                 err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1732                 if (err) {
1733                         message = "writing ring-ref";
1734                         goto abort_transaction;
1735                 }
1736         } else {
1737                 for (i = 0; i < info->nr_ring_pages; i++) {
1738                         char ring_ref_name[RINGREF_NAME_LEN];
1739
1740                         snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1741                         err = xenbus_printf(xbt, dir, ring_ref_name,
1742                                             "%u", rinfo->ring_ref[i]);
1743                         if (err) {
1744                                 message = "writing ring-ref";
1745                                 goto abort_transaction;
1746                         }
1747                 }
1748         }
1749
1750         err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1751         if (err) {
1752                 message = "writing event-channel";
1753                 goto abort_transaction;
1754         }
1755
1756         return 0;
1757
1758 abort_transaction:
1759         xenbus_transaction_end(xbt, 1);
1760         if (message)
1761                 xenbus_dev_fatal(info->xbdev, err, "%s", message);
1762
1763         return err;
1764 }
1765
1766 static void free_info(struct blkfront_info *info)
1767 {
1768         list_del(&info->info_list);
1769         kfree(info);
1770 }
1771
1772 /* Common code used when first setting up, and when resuming. */
1773 static int talk_to_blkback(struct xenbus_device *dev,
1774                            struct blkfront_info *info)
1775 {
1776         const char *message = NULL;
1777         struct xenbus_transaction xbt;
1778         int err;
1779         unsigned int i, max_page_order;
1780         unsigned int ring_page_order;
1781
1782         if (!info)
1783                 return -ENODEV;
1784
1785         max_page_order = xenbus_read_unsigned(info->xbdev->otherend,
1786                                               "max-ring-page-order", 0);
1787         ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1788         info->nr_ring_pages = 1 << ring_page_order;
1789
1790         err = negotiate_mq(info);
1791         if (err)
1792                 goto destroy_blkring;
1793
1794         for (i = 0; i < info->nr_rings; i++) {
1795                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1796
1797                 /* Create shared ring, alloc event channel. */
1798                 err = setup_blkring(dev, rinfo);
1799                 if (err)
1800                         goto destroy_blkring;
1801         }
1802
1803 again:
1804         err = xenbus_transaction_start(&xbt);
1805         if (err) {
1806                 xenbus_dev_fatal(dev, err, "starting transaction");
1807                 goto destroy_blkring;
1808         }
1809
1810         if (info->nr_ring_pages > 1) {
1811                 err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1812                                     ring_page_order);
1813                 if (err) {
1814                         message = "writing ring-page-order";
1815                         goto abort_transaction;
1816                 }
1817         }
1818
1819         /* We already got the number of queues/rings in _probe */
1820         if (info->nr_rings == 1) {
1821                 err = write_per_ring_nodes(xbt, &info->rinfo[0], dev->nodename);
1822                 if (err)
1823                         goto destroy_blkring;
1824         } else {
1825                 char *path;
1826                 size_t pathsize;
1827
1828                 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1829                                     info->nr_rings);
1830                 if (err) {
1831                         message = "writing multi-queue-num-queues";
1832                         goto abort_transaction;
1833                 }
1834
1835                 pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1836                 path = kmalloc(pathsize, GFP_KERNEL);
1837                 if (!path) {
1838                         err = -ENOMEM;
1839                         message = "ENOMEM while writing ring references";
1840                         goto abort_transaction;
1841                 }
1842
1843                 for (i = 0; i < info->nr_rings; i++) {
1844                         memset(path, 0, pathsize);
1845                         snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1846                         err = write_per_ring_nodes(xbt, &info->rinfo[i], path);
1847                         if (err) {
1848                                 kfree(path);
1849                                 goto destroy_blkring;
1850                         }
1851                 }
1852                 kfree(path);
1853         }
1854         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1855                             XEN_IO_PROTO_ABI_NATIVE);
1856         if (err) {
1857                 message = "writing protocol";
1858                 goto abort_transaction;
1859         }
1860         err = xenbus_printf(xbt, dev->nodename,
1861                             "feature-persistent", "%u", 1);
1862         if (err)
1863                 dev_warn(&dev->dev,
1864                          "writing persistent grants feature to xenbus");
1865
1866         err = xenbus_transaction_end(xbt, 0);
1867         if (err) {
1868                 if (err == -EAGAIN)
1869                         goto again;
1870                 xenbus_dev_fatal(dev, err, "completing transaction");
1871                 goto destroy_blkring;
1872         }
1873
1874         for (i = 0; i < info->nr_rings; i++) {
1875                 unsigned int j;
1876                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1877
1878                 for (j = 0; j < BLK_RING_SIZE(info); j++)
1879                         rinfo->shadow[j].req.u.rw.id = j + 1;
1880                 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1881         }
1882         xenbus_switch_state(dev, XenbusStateInitialised);
1883
1884         return 0;
1885
1886  abort_transaction:
1887         xenbus_transaction_end(xbt, 1);
1888         if (message)
1889                 xenbus_dev_fatal(dev, err, "%s", message);
1890  destroy_blkring:
1891         blkif_free(info, 0);
1892
1893         mutex_lock(&blkfront_mutex);
1894         free_info(info);
1895         mutex_unlock(&blkfront_mutex);
1896
1897         dev_set_drvdata(&dev->dev, NULL);
1898
1899         return err;
1900 }
1901
1902 static int negotiate_mq(struct blkfront_info *info)
1903 {
1904         unsigned int backend_max_queues;
1905         unsigned int i;
1906
1907         BUG_ON(info->nr_rings);
1908
1909         /* Check if backend supports multiple queues. */
1910         backend_max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1911                                                   "multi-queue-max-queues", 1);
1912         info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1913         /* We need at least one ring. */
1914         if (!info->nr_rings)
1915                 info->nr_rings = 1;
1916
1917         info->rinfo = kcalloc(info->nr_rings,
1918                               sizeof(struct blkfront_ring_info),
1919                               GFP_KERNEL);
1920         if (!info->rinfo) {
1921                 xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1922                 info->nr_rings = 0;
1923                 return -ENOMEM;
1924         }
1925
1926         for (i = 0; i < info->nr_rings; i++) {
1927                 struct blkfront_ring_info *rinfo;
1928
1929                 rinfo = &info->rinfo[i];
1930                 INIT_LIST_HEAD(&rinfo->indirect_pages);
1931                 INIT_LIST_HEAD(&rinfo->grants);
1932                 rinfo->dev_info = info;
1933                 INIT_WORK(&rinfo->work, blkif_restart_queue);
1934                 spin_lock_init(&rinfo->ring_lock);
1935         }
1936         return 0;
1937 }
1938 /**
1939  * Entry point to this code when a new device is created.  Allocate the basic
1940  * structures and the ring buffer for communication with the backend, and
1941  * inform the backend of the appropriate details for those.  Switch to
1942  * Initialised state.
1943  */
1944 static int blkfront_probe(struct xenbus_device *dev,
1945                           const struct xenbus_device_id *id)
1946 {
1947         int err, vdevice;
1948         struct blkfront_info *info;
1949
1950         /* FIXME: Use dynamic device id if this is not set. */
1951         err = xenbus_scanf(XBT_NIL, dev->nodename,
1952                            "virtual-device", "%i", &vdevice);
1953         if (err != 1) {
1954                 /* go looking in the extended area instead */
1955                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1956                                    "%i", &vdevice);
1957                 if (err != 1) {
1958                         xenbus_dev_fatal(dev, err, "reading virtual-device");
1959                         return err;
1960                 }
1961         }
1962
1963         if (xen_hvm_domain()) {
1964                 char *type;
1965                 int len;
1966                 /* no unplug has been done: do not hook devices != xen vbds */
1967                 if (xen_has_pv_and_legacy_disk_devices()) {
1968                         int major;
1969
1970                         if (!VDEV_IS_EXTENDED(vdevice))
1971                                 major = BLKIF_MAJOR(vdevice);
1972                         else
1973                                 major = XENVBD_MAJOR;
1974
1975                         if (major != XENVBD_MAJOR) {
1976                                 printk(KERN_INFO
1977                                                 "%s: HVM does not support vbd %d as xen block device\n",
1978                                                 __func__, vdevice);
1979                                 return -ENODEV;
1980                         }
1981                 }
1982                 /* do not create a PV cdrom device if we are an HVM guest */
1983                 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1984                 if (IS_ERR(type))
1985                         return -ENODEV;
1986                 if (strncmp(type, "cdrom", 5) == 0) {
1987                         kfree(type);
1988                         return -ENODEV;
1989                 }
1990                 kfree(type);
1991         }
1992         info = kzalloc(sizeof(*info), GFP_KERNEL);
1993         if (!info) {
1994                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1995                 return -ENOMEM;
1996         }
1997
1998         info->xbdev = dev;
1999
2000         mutex_init(&info->mutex);
2001         info->vdevice = vdevice;
2002         info->connected = BLKIF_STATE_DISCONNECTED;
2003
2004         /* Front end dir is a number, which is used as the id. */
2005         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
2006         dev_set_drvdata(&dev->dev, info);
2007
2008         mutex_lock(&blkfront_mutex);
2009         list_add(&info->info_list, &info_list);
2010         mutex_unlock(&blkfront_mutex);
2011
2012         return 0;
2013 }
2014
2015 static int blkif_recover(struct blkfront_info *info)
2016 {
2017         unsigned int r_index;
2018         struct request *req, *n;
2019         int rc;
2020         struct bio *bio;
2021         unsigned int segs;
2022
2023         blkfront_gather_backend_features(info);
2024         /* Reset limits changed by blk_mq_update_nr_hw_queues(). */
2025         blkif_set_queue_limits(info);
2026         segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
2027         blk_queue_max_segments(info->rq, segs / GRANTS_PER_PSEG);
2028
2029         for (r_index = 0; r_index < info->nr_rings; r_index++) {
2030                 struct blkfront_ring_info *rinfo = &info->rinfo[r_index];
2031
2032                 rc = blkfront_setup_indirect(rinfo);
2033                 if (rc)
2034                         return rc;
2035         }
2036         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2037
2038         /* Now safe for us to use the shared ring */
2039         info->connected = BLKIF_STATE_CONNECTED;
2040
2041         for (r_index = 0; r_index < info->nr_rings; r_index++) {
2042                 struct blkfront_ring_info *rinfo;
2043
2044                 rinfo = &info->rinfo[r_index];
2045                 /* Kick any other new requests queued since we resumed */
2046                 kick_pending_request_queues(rinfo);
2047         }
2048
2049         list_for_each_entry_safe(req, n, &info->requests, queuelist) {
2050                 /* Requeue pending requests (flush or discard) */
2051                 list_del_init(&req->queuelist);
2052                 BUG_ON(req->nr_phys_segments > segs);
2053                 blk_mq_requeue_request(req, false);
2054         }
2055         blk_mq_start_stopped_hw_queues(info->rq, true);
2056         blk_mq_kick_requeue_list(info->rq);
2057
2058         while ((bio = bio_list_pop(&info->bio_list)) != NULL) {
2059                 /* Traverse the list of pending bios and re-queue them */
2060                 submit_bio(bio);
2061         }
2062
2063         return 0;
2064 }
2065
2066 /**
2067  * We are reconnecting to the backend, due to a suspend/resume, or a backend
2068  * driver restart.  We tear down our blkif structure and recreate it, but
2069  * leave the device-layer structures intact so that this is transparent to the
2070  * rest of the kernel.
2071  */
2072 static int blkfront_resume(struct xenbus_device *dev)
2073 {
2074         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2075         int err = 0;
2076         unsigned int i, j;
2077
2078         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2079
2080         bio_list_init(&info->bio_list);
2081         INIT_LIST_HEAD(&info->requests);
2082         for (i = 0; i < info->nr_rings; i++) {
2083                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
2084                 struct bio_list merge_bio;
2085                 struct blk_shadow *shadow = rinfo->shadow;
2086
2087                 for (j = 0; j < BLK_RING_SIZE(info); j++) {
2088                         /* Not in use? */
2089                         if (!shadow[j].request)
2090                                 continue;
2091
2092                         /*
2093                          * Get the bios in the request so we can re-queue them.
2094                          */
2095                         if (req_op(shadow[j].request) == REQ_OP_FLUSH ||
2096                             req_op(shadow[j].request) == REQ_OP_DISCARD ||
2097                             req_op(shadow[j].request) == REQ_OP_SECURE_ERASE ||
2098                             shadow[j].request->cmd_flags & REQ_FUA) {
2099                                 /*
2100                                  * Flush operations don't contain bios, so
2101                                  * we need to requeue the whole request
2102                                  *
2103                                  * XXX: but this doesn't make any sense for a
2104                                  * write with the FUA flag set..
2105                                  */
2106                                 list_add(&shadow[j].request->queuelist, &info->requests);
2107                                 continue;
2108                         }
2109                         merge_bio.head = shadow[j].request->bio;
2110                         merge_bio.tail = shadow[j].request->biotail;
2111                         bio_list_merge(&info->bio_list, &merge_bio);
2112                         shadow[j].request->bio = NULL;
2113                         blk_mq_end_request(shadow[j].request, BLK_STS_OK);
2114                 }
2115         }
2116
2117         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2118
2119         err = talk_to_blkback(dev, info);
2120         if (!err)
2121                 blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings);
2122
2123         /*
2124          * We have to wait for the backend to switch to
2125          * connected state, since we want to read which
2126          * features it supports.
2127          */
2128
2129         return err;
2130 }
2131
2132 static void blkfront_closing(struct blkfront_info *info)
2133 {
2134         struct xenbus_device *xbdev = info->xbdev;
2135         struct block_device *bdev = NULL;
2136
2137         mutex_lock(&info->mutex);
2138
2139         if (xbdev->state == XenbusStateClosing) {
2140                 mutex_unlock(&info->mutex);
2141                 return;
2142         }
2143
2144         if (info->gd)
2145                 bdev = bdget_disk(info->gd, 0);
2146
2147         mutex_unlock(&info->mutex);
2148
2149         if (!bdev) {
2150                 xenbus_frontend_closed(xbdev);
2151                 return;
2152         }
2153
2154         mutex_lock(&bdev->bd_mutex);
2155
2156         if (bdev->bd_openers) {
2157                 xenbus_dev_error(xbdev, -EBUSY,
2158                                  "Device in use; refusing to close");
2159                 xenbus_switch_state(xbdev, XenbusStateClosing);
2160         } else {
2161                 xlvbd_release_gendisk(info);
2162                 xenbus_frontend_closed(xbdev);
2163         }
2164
2165         mutex_unlock(&bdev->bd_mutex);
2166         bdput(bdev);
2167 }
2168
2169 static void blkfront_setup_discard(struct blkfront_info *info)
2170 {
2171         int err;
2172         unsigned int discard_granularity;
2173         unsigned int discard_alignment;
2174
2175         info->feature_discard = 1;
2176         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2177                 "discard-granularity", "%u", &discard_granularity,
2178                 "discard-alignment", "%u", &discard_alignment,
2179                 NULL);
2180         if (!err) {
2181                 info->discard_granularity = discard_granularity;
2182                 info->discard_alignment = discard_alignment;
2183         }
2184         info->feature_secdiscard =
2185                 !!xenbus_read_unsigned(info->xbdev->otherend, "discard-secure",
2186                                        0);
2187 }
2188
2189 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2190 {
2191         unsigned int psegs, grants;
2192         int err, i;
2193         struct blkfront_info *info = rinfo->dev_info;
2194
2195         if (info->max_indirect_segments == 0) {
2196                 if (!HAS_EXTRA_REQ)
2197                         grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2198                 else {
2199                         /*
2200                          * When an extra req is required, the maximum
2201                          * grants supported is related to the size of the
2202                          * Linux block segment.
2203                          */
2204                         grants = GRANTS_PER_PSEG;
2205                 }
2206         }
2207         else
2208                 grants = info->max_indirect_segments;
2209         psegs = DIV_ROUND_UP(grants, GRANTS_PER_PSEG);
2210
2211         err = fill_grant_buffer(rinfo,
2212                                 (grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2213         if (err)
2214                 goto out_of_memory;
2215
2216         if (!info->feature_persistent && info->max_indirect_segments) {
2217                 /*
2218                  * We are using indirect descriptors but not persistent
2219                  * grants, we need to allocate a set of pages that can be
2220                  * used for mapping indirect grefs
2221                  */
2222                 int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2223
2224                 BUG_ON(!list_empty(&rinfo->indirect_pages));
2225                 for (i = 0; i < num; i++) {
2226                         struct page *indirect_page = alloc_page(GFP_NOIO);
2227                         if (!indirect_page)
2228                                 goto out_of_memory;
2229                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
2230                 }
2231         }
2232
2233         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2234                 rinfo->shadow[i].grants_used =
2235                         kcalloc(grants,
2236                                 sizeof(rinfo->shadow[i].grants_used[0]),
2237                                 GFP_NOIO);
2238                 rinfo->shadow[i].sg = kcalloc(psegs,
2239                                               sizeof(rinfo->shadow[i].sg[0]),
2240                                               GFP_NOIO);
2241                 if (info->max_indirect_segments)
2242                         rinfo->shadow[i].indirect_grants =
2243                                 kcalloc(INDIRECT_GREFS(grants),
2244                                         sizeof(rinfo->shadow[i].indirect_grants[0]),
2245                                         GFP_NOIO);
2246                 if ((rinfo->shadow[i].grants_used == NULL) ||
2247                         (rinfo->shadow[i].sg == NULL) ||
2248                      (info->max_indirect_segments &&
2249                      (rinfo->shadow[i].indirect_grants == NULL)))
2250                         goto out_of_memory;
2251                 sg_init_table(rinfo->shadow[i].sg, psegs);
2252         }
2253
2254
2255         return 0;
2256
2257 out_of_memory:
2258         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2259                 kfree(rinfo->shadow[i].grants_used);
2260                 rinfo->shadow[i].grants_used = NULL;
2261                 kfree(rinfo->shadow[i].sg);
2262                 rinfo->shadow[i].sg = NULL;
2263                 kfree(rinfo->shadow[i].indirect_grants);
2264                 rinfo->shadow[i].indirect_grants = NULL;
2265         }
2266         if (!list_empty(&rinfo->indirect_pages)) {
2267                 struct page *indirect_page, *n;
2268                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2269                         list_del(&indirect_page->lru);
2270                         __free_page(indirect_page);
2271                 }
2272         }
2273         return -ENOMEM;
2274 }
2275
2276 /*
2277  * Gather all backend feature-*
2278  */
2279 static void blkfront_gather_backend_features(struct blkfront_info *info)
2280 {
2281         unsigned int indirect_segments;
2282
2283         info->feature_flush = 0;
2284         info->feature_fua = 0;
2285
2286         /*
2287          * If there's no "feature-barrier" defined, then it means
2288          * we're dealing with a very old backend which writes
2289          * synchronously; nothing to do.
2290          *
2291          * If there are barriers, then we use flush.
2292          */
2293         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-barrier", 0)) {
2294                 info->feature_flush = 1;
2295                 info->feature_fua = 1;
2296         }
2297
2298         /*
2299          * And if there is "feature-flush-cache" use that above
2300          * barriers.
2301          */
2302         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-flush-cache",
2303                                  0)) {
2304                 info->feature_flush = 1;
2305                 info->feature_fua = 0;
2306         }
2307
2308         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-discard", 0))
2309                 blkfront_setup_discard(info);
2310
2311         info->feature_persistent =
2312                 !!xenbus_read_unsigned(info->xbdev->otherend,
2313                                        "feature-persistent", 0);
2314
2315         indirect_segments = xenbus_read_unsigned(info->xbdev->otherend,
2316                                         "feature-max-indirect-segments", 0);
2317         if (indirect_segments > xen_blkif_max_segments)
2318                 indirect_segments = xen_blkif_max_segments;
2319         if (indirect_segments <= BLKIF_MAX_SEGMENTS_PER_REQUEST)
2320                 indirect_segments = 0;
2321         info->max_indirect_segments = indirect_segments;
2322
2323         if (info->feature_persistent) {
2324                 mutex_lock(&blkfront_mutex);
2325                 schedule_delayed_work(&blkfront_work, HZ * 10);
2326                 mutex_unlock(&blkfront_mutex);
2327         }
2328 }
2329
2330 /*
2331  * Invoked when the backend is finally 'ready' (and has told produced
2332  * the details about the physical device - #sectors, size, etc).
2333  */
2334 static void blkfront_connect(struct blkfront_info *info)
2335 {
2336         unsigned long long sectors;
2337         unsigned long sector_size;
2338         unsigned int physical_sector_size;
2339         unsigned int binfo;
2340         char *envp[] = { "RESIZE=1", NULL };
2341         int err, i;
2342
2343         switch (info->connected) {
2344         case BLKIF_STATE_CONNECTED:
2345                 /*
2346                  * Potentially, the back-end may be signalling
2347                  * a capacity change; update the capacity.
2348                  */
2349                 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2350                                    "sectors", "%Lu", &sectors);
2351                 if (XENBUS_EXIST_ERR(err))
2352                         return;
2353                 printk(KERN_INFO "Setting capacity to %Lu\n",
2354                        sectors);
2355                 set_capacity(info->gd, sectors);
2356                 revalidate_disk(info->gd);
2357                 kobject_uevent_env(&disk_to_dev(info->gd)->kobj,
2358                                    KOBJ_CHANGE, envp);
2359
2360                 return;
2361         case BLKIF_STATE_SUSPENDED:
2362                 /*
2363                  * If we are recovering from suspension, we need to wait
2364                  * for the backend to announce it's features before
2365                  * reconnecting, at least we need to know if the backend
2366                  * supports indirect descriptors, and how many.
2367                  */
2368                 blkif_recover(info);
2369                 return;
2370
2371         default:
2372                 break;
2373         }
2374
2375         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2376                 __func__, info->xbdev->otherend);
2377
2378         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2379                             "sectors", "%llu", &sectors,
2380                             "info", "%u", &binfo,
2381                             "sector-size", "%lu", &sector_size,
2382                             NULL);
2383         if (err) {
2384                 xenbus_dev_fatal(info->xbdev, err,
2385                                  "reading backend fields at %s",
2386                                  info->xbdev->otherend);
2387                 return;
2388         }
2389
2390         /*
2391          * physcial-sector-size is a newer field, so old backends may not
2392          * provide this. Assume physical sector size to be the same as
2393          * sector_size in that case.
2394          */
2395         physical_sector_size = xenbus_read_unsigned(info->xbdev->otherend,
2396                                                     "physical-sector-size",
2397                                                     sector_size);
2398         blkfront_gather_backend_features(info);
2399         for (i = 0; i < info->nr_rings; i++) {
2400                 err = blkfront_setup_indirect(&info->rinfo[i]);
2401                 if (err) {
2402                         xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2403                                          info->xbdev->otherend);
2404                         blkif_free(info, 0);
2405                         break;
2406                 }
2407         }
2408
2409         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
2410                                   physical_sector_size);
2411         if (err) {
2412                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2413                                  info->xbdev->otherend);
2414                 goto fail;
2415         }
2416
2417         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2418
2419         /* Kick pending requests. */
2420         info->connected = BLKIF_STATE_CONNECTED;
2421         for (i = 0; i < info->nr_rings; i++)
2422                 kick_pending_request_queues(&info->rinfo[i]);
2423
2424         device_add_disk(&info->xbdev->dev, info->gd, NULL);
2425
2426         info->is_ready = 1;
2427         return;
2428
2429 fail:
2430         blkif_free(info, 0);
2431         return;
2432 }
2433
2434 /**
2435  * Callback received when the backend's state changes.
2436  */
2437 static void blkback_changed(struct xenbus_device *dev,
2438                             enum xenbus_state backend_state)
2439 {
2440         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2441
2442         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2443
2444         switch (backend_state) {
2445         case XenbusStateInitWait:
2446                 if (dev->state != XenbusStateInitialising)
2447                         break;
2448                 if (talk_to_blkback(dev, info))
2449                         break;
2450         case XenbusStateInitialising:
2451         case XenbusStateInitialised:
2452         case XenbusStateReconfiguring:
2453         case XenbusStateReconfigured:
2454         case XenbusStateUnknown:
2455                 break;
2456
2457         case XenbusStateConnected:
2458                 /*
2459                  * talk_to_blkback sets state to XenbusStateInitialised
2460                  * and blkfront_connect sets it to XenbusStateConnected
2461                  * (if connection went OK).
2462                  *
2463                  * If the backend (or toolstack) decides to poke at backend
2464                  * state (and re-trigger the watch by setting the state repeatedly
2465                  * to XenbusStateConnected (4)) we need to deal with this.
2466                  * This is allowed as this is used to communicate to the guest
2467                  * that the size of disk has changed!
2468                  */
2469                 if ((dev->state != XenbusStateInitialised) &&
2470                     (dev->state != XenbusStateConnected)) {
2471                         if (talk_to_blkback(dev, info))
2472                                 break;
2473                 }
2474
2475                 blkfront_connect(info);
2476                 break;
2477
2478         case XenbusStateClosed:
2479                 if (dev->state == XenbusStateClosed)
2480                         break;
2481                 /* fall through */
2482         case XenbusStateClosing:
2483                 if (info)
2484                         blkfront_closing(info);
2485                 break;
2486         }
2487 }
2488
2489 static int blkfront_remove(struct xenbus_device *xbdev)
2490 {
2491         struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2492         struct block_device *bdev = NULL;
2493         struct gendisk *disk;
2494
2495         dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2496
2497         if (!info)
2498                 return 0;
2499
2500         blkif_free(info, 0);
2501
2502         mutex_lock(&info->mutex);
2503
2504         disk = info->gd;
2505         if (disk)
2506                 bdev = bdget_disk(disk, 0);
2507
2508         info->xbdev = NULL;
2509         mutex_unlock(&info->mutex);
2510
2511         if (!bdev) {
2512                 mutex_lock(&blkfront_mutex);
2513                 free_info(info);
2514                 mutex_unlock(&blkfront_mutex);
2515                 return 0;
2516         }
2517
2518         /*
2519          * The xbdev was removed before we reached the Closed
2520          * state. See if it's safe to remove the disk. If the bdev
2521          * isn't closed yet, we let release take care of it.
2522          */
2523
2524         mutex_lock(&bdev->bd_mutex);
2525         info = disk->private_data;
2526
2527         dev_warn(disk_to_dev(disk),
2528                  "%s was hot-unplugged, %d stale handles\n",
2529                  xbdev->nodename, bdev->bd_openers);
2530
2531         if (info && !bdev->bd_openers) {
2532                 xlvbd_release_gendisk(info);
2533                 disk->private_data = NULL;
2534                 mutex_lock(&blkfront_mutex);
2535                 free_info(info);
2536                 mutex_unlock(&blkfront_mutex);
2537         }
2538
2539         mutex_unlock(&bdev->bd_mutex);
2540         bdput(bdev);
2541
2542         return 0;
2543 }
2544
2545 static int blkfront_is_ready(struct xenbus_device *dev)
2546 {
2547         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2548
2549         return info->is_ready && info->xbdev;
2550 }
2551
2552 static int blkif_open(struct block_device *bdev, fmode_t mode)
2553 {
2554         struct gendisk *disk = bdev->bd_disk;
2555         struct blkfront_info *info;
2556         int err = 0;
2557
2558         mutex_lock(&blkfront_mutex);
2559
2560         info = disk->private_data;
2561         if (!info) {
2562                 /* xbdev gone */
2563                 err = -ERESTARTSYS;
2564                 goto out;
2565         }
2566
2567         mutex_lock(&info->mutex);
2568
2569         if (!info->gd)
2570                 /* xbdev is closed */
2571                 err = -ERESTARTSYS;
2572
2573         mutex_unlock(&info->mutex);
2574
2575 out:
2576         mutex_unlock(&blkfront_mutex);
2577         return err;
2578 }
2579
2580 static void blkif_release(struct gendisk *disk, fmode_t mode)
2581 {
2582         struct blkfront_info *info = disk->private_data;
2583         struct block_device *bdev;
2584         struct xenbus_device *xbdev;
2585
2586         mutex_lock(&blkfront_mutex);
2587
2588         bdev = bdget_disk(disk, 0);
2589
2590         if (!bdev) {
2591                 WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2592                 goto out_mutex;
2593         }
2594         if (bdev->bd_openers)
2595                 goto out;
2596
2597         /*
2598          * Check if we have been instructed to close. We will have
2599          * deferred this request, because the bdev was still open.
2600          */
2601
2602         mutex_lock(&info->mutex);
2603         xbdev = info->xbdev;
2604
2605         if (xbdev && xbdev->state == XenbusStateClosing) {
2606                 /* pending switch to state closed */
2607                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2608                 xlvbd_release_gendisk(info);
2609                 xenbus_frontend_closed(info->xbdev);
2610         }
2611
2612         mutex_unlock(&info->mutex);
2613
2614         if (!xbdev) {
2615                 /* sudden device removal */
2616                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2617                 xlvbd_release_gendisk(info);
2618                 disk->private_data = NULL;
2619                 free_info(info);
2620         }
2621
2622 out:
2623         bdput(bdev);
2624 out_mutex:
2625         mutex_unlock(&blkfront_mutex);
2626 }
2627
2628 static const struct block_device_operations xlvbd_block_fops =
2629 {
2630         .owner = THIS_MODULE,
2631         .open = blkif_open,
2632         .release = blkif_release,
2633         .getgeo = blkif_getgeo,
2634         .ioctl = blkif_ioctl,
2635 };
2636
2637
2638 static const struct xenbus_device_id blkfront_ids[] = {
2639         { "vbd" },
2640         { "" }
2641 };
2642
2643 static struct xenbus_driver blkfront_driver = {
2644         .ids  = blkfront_ids,
2645         .probe = blkfront_probe,
2646         .remove = blkfront_remove,
2647         .resume = blkfront_resume,
2648         .otherend_changed = blkback_changed,
2649         .is_ready = blkfront_is_ready,
2650 };
2651
2652 static void purge_persistent_grants(struct blkfront_info *info)
2653 {
2654         unsigned int i;
2655         unsigned long flags;
2656
2657         for (i = 0; i < info->nr_rings; i++) {
2658                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
2659                 struct grant *gnt_list_entry, *tmp;
2660
2661                 spin_lock_irqsave(&rinfo->ring_lock, flags);
2662
2663                 if (rinfo->persistent_gnts_c == 0) {
2664                         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2665                         continue;
2666                 }
2667
2668                 list_for_each_entry_safe(gnt_list_entry, tmp, &rinfo->grants,
2669                                          node) {
2670                         if (gnt_list_entry->gref == GRANT_INVALID_REF ||
2671                             gnttab_query_foreign_access(gnt_list_entry->gref))
2672                                 continue;
2673
2674                         list_del(&gnt_list_entry->node);
2675                         gnttab_end_foreign_access(gnt_list_entry->gref, 0, 0UL);
2676                         rinfo->persistent_gnts_c--;
2677                         gnt_list_entry->gref = GRANT_INVALID_REF;
2678                         list_add_tail(&gnt_list_entry->node, &rinfo->grants);
2679                 }
2680
2681                 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2682         }
2683 }
2684
2685 static void blkfront_delay_work(struct work_struct *work)
2686 {
2687         struct blkfront_info *info;
2688         bool need_schedule_work = false;
2689
2690         mutex_lock(&blkfront_mutex);
2691
2692         list_for_each_entry(info, &info_list, info_list) {
2693                 if (info->feature_persistent) {
2694                         need_schedule_work = true;
2695                         mutex_lock(&info->mutex);
2696                         purge_persistent_grants(info);
2697                         mutex_unlock(&info->mutex);
2698                 }
2699         }
2700
2701         if (need_schedule_work)
2702                 schedule_delayed_work(&blkfront_work, HZ * 10);
2703
2704         mutex_unlock(&blkfront_mutex);
2705 }
2706
2707 static int __init xlblk_init(void)
2708 {
2709         int ret;
2710         int nr_cpus = num_online_cpus();
2711
2712         if (!xen_domain())
2713                 return -ENODEV;
2714
2715         if (!xen_has_pv_disk_devices())
2716                 return -ENODEV;
2717
2718         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2719                 pr_warn("xen_blk: can't get major %d with name %s\n",
2720                         XENVBD_MAJOR, DEV_NAME);
2721                 return -ENODEV;
2722         }
2723
2724         if (xen_blkif_max_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST)
2725                 xen_blkif_max_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2726
2727         if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2728                 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2729                         xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2730                 xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2731         }
2732
2733         if (xen_blkif_max_queues > nr_cpus) {
2734                 pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2735                         xen_blkif_max_queues, nr_cpus);
2736                 xen_blkif_max_queues = nr_cpus;
2737         }
2738
2739         INIT_DELAYED_WORK(&blkfront_work, blkfront_delay_work);
2740
2741         ret = xenbus_register_frontend(&blkfront_driver);
2742         if (ret) {
2743                 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2744                 return ret;
2745         }
2746
2747         return 0;
2748 }
2749 module_init(xlblk_init);
2750
2751
2752 static void __exit xlblk_exit(void)
2753 {
2754         cancel_delayed_work_sync(&blkfront_work);
2755
2756         xenbus_unregister_driver(&blkfront_driver);
2757         unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2758         kfree(minors);
2759 }
2760 module_exit(xlblk_exit);
2761
2762 MODULE_DESCRIPTION("Xen virtual block device frontend");
2763 MODULE_LICENSE("GPL");
2764 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2765 MODULE_ALIAS("xen:vbd");
2766 MODULE_ALIAS("xenblk");