Merge branch 'stable/for-jens-4.15' 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
50 #include <xen/xen.h>
51 #include <xen/xenbus.h>
52 #include <xen/grant_table.h>
53 #include <xen/events.h>
54 #include <xen/page.h>
55 #include <xen/platform_pci.h>
56
57 #include <xen/interface/grant_table.h>
58 #include <xen/interface/io/blkif.h>
59 #include <xen/interface/io/protocols.h>
60
61 #include <asm/xen/hypervisor.h>
62
63 /*
64  * The minimal size of segment supported by the block framework is PAGE_SIZE.
65  * When Linux is using a different page size than Xen, it may not be possible
66  * to put all the data in a single segment.
67  * This can happen when the backend doesn't support indirect descriptor and
68  * therefore the maximum amount of data that a request can carry is
69  * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE = 44KB
70  *
71  * Note that we only support one extra request. So the Linux page size
72  * should be <= ( 2 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) =
73  * 88KB.
74  */
75 #define HAS_EXTRA_REQ (BLKIF_MAX_SEGMENTS_PER_REQUEST < XEN_PFN_PER_PAGE)
76
77 enum blkif_state {
78         BLKIF_STATE_DISCONNECTED,
79         BLKIF_STATE_CONNECTED,
80         BLKIF_STATE_SUSPENDED,
81 };
82
83 struct grant {
84         grant_ref_t gref;
85         struct page *page;
86         struct list_head node;
87 };
88
89 enum blk_req_status {
90         REQ_WAITING,
91         REQ_DONE,
92         REQ_ERROR,
93         REQ_EOPNOTSUPP,
94 };
95
96 struct blk_shadow {
97         struct blkif_request req;
98         struct request *request;
99         struct grant **grants_used;
100         struct grant **indirect_grants;
101         struct scatterlist *sg;
102         unsigned int num_sg;
103         enum blk_req_status status;
104
105         #define NO_ASSOCIATED_ID ~0UL
106         /*
107          * Id of the sibling if we ever need 2 requests when handling a
108          * block I/O request
109          */
110         unsigned long associated_id;
111 };
112
113 struct blkif_req {
114         blk_status_t    error;
115 };
116
117 static inline struct blkif_req *blkif_req(struct request *rq)
118 {
119         return blk_mq_rq_to_pdu(rq);
120 }
121
122 static DEFINE_MUTEX(blkfront_mutex);
123 static const struct block_device_operations xlvbd_block_fops;
124
125 /*
126  * Maximum number of segments in indirect requests, the actual value used by
127  * the frontend driver is the minimum of this value and the value provided
128  * by the backend driver.
129  */
130
131 static unsigned int xen_blkif_max_segments = 32;
132 module_param_named(max_indirect_segments, xen_blkif_max_segments, uint,
133                    S_IRUGO);
134 MODULE_PARM_DESC(max_indirect_segments,
135                  "Maximum amount of segments in indirect requests (default is 32)");
136
137 static unsigned int xen_blkif_max_queues = 4;
138 module_param_named(max_queues, xen_blkif_max_queues, uint, S_IRUGO);
139 MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
140
141 /*
142  * Maximum order of pages to be used for the shared ring between front and
143  * backend, 4KB page granularity is used.
144  */
145 static unsigned int xen_blkif_max_ring_order;
146 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, S_IRUGO);
147 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
148
149 #define BLK_RING_SIZE(info)     \
150         __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
151
152 #define BLK_MAX_RING_SIZE       \
153         __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * XENBUS_MAX_RING_GRANTS)
154
155 /*
156  * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
157  * characters are enough. Define to 20 to keep consistent with backend.
158  */
159 #define RINGREF_NAME_LEN (20)
160 /*
161  * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
162  */
163 #define QUEUE_NAME_LEN (17)
164
165 /*
166  *  Per-ring info.
167  *  Every blkfront device can associate with one or more blkfront_ring_info,
168  *  depending on how many hardware queues/rings to be used.
169  */
170 struct blkfront_ring_info {
171         /* Lock to protect data in every ring buffer. */
172         spinlock_t ring_lock;
173         struct blkif_front_ring ring;
174         unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
175         unsigned int evtchn, irq;
176         struct work_struct work;
177         struct gnttab_free_callback callback;
178         struct blk_shadow shadow[BLK_MAX_RING_SIZE];
179         struct list_head indirect_pages;
180         struct list_head grants;
181         unsigned int persistent_gnts_c;
182         unsigned long shadow_free;
183         struct blkfront_info *dev_info;
184 };
185
186 /*
187  * We have one of these per vbd, whether ide, scsi or 'other'.  They
188  * hang in private_data off the gendisk structure. We may end up
189  * putting all kinds of interesting stuff here :-)
190  */
191 struct blkfront_info
192 {
193         struct mutex mutex;
194         struct xenbus_device *xbdev;
195         struct gendisk *gd;
196         u16 sector_size;
197         unsigned int physical_sector_size;
198         int vdevice;
199         blkif_vdev_t handle;
200         enum blkif_state connected;
201         /* Number of pages per ring buffer. */
202         unsigned int nr_ring_pages;
203         struct request_queue *rq;
204         unsigned int feature_flush:1;
205         unsigned int feature_fua:1;
206         unsigned int feature_discard:1;
207         unsigned int feature_secdiscard:1;
208         unsigned int feature_persistent:1;
209         unsigned int discard_granularity;
210         unsigned int discard_alignment;
211         /* Number of 4KB segments handled */
212         unsigned int max_indirect_segments;
213         int is_ready;
214         struct blk_mq_tag_set tag_set;
215         struct blkfront_ring_info *rinfo;
216         unsigned int nr_rings;
217         /* Save uncomplete reqs and bios for migration. */
218         struct list_head requests;
219         struct bio_list bio_list;
220 };
221
222 static unsigned int nr_minors;
223 static unsigned long *minors;
224 static DEFINE_SPINLOCK(minor_lock);
225
226 #define GRANT_INVALID_REF       0
227
228 #define PARTS_PER_DISK          16
229 #define PARTS_PER_EXT_DISK      256
230
231 #define BLKIF_MAJOR(dev) ((dev)>>8)
232 #define BLKIF_MINOR(dev) ((dev) & 0xff)
233
234 #define EXT_SHIFT 28
235 #define EXTENDED (1<<EXT_SHIFT)
236 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
237 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
238 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
239 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
240 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
241 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
242
243 #define DEV_NAME        "xvd"   /* name in /dev */
244
245 /*
246  * Grants are always the same size as a Xen page (i.e 4KB).
247  * A physical segment is always the same size as a Linux page.
248  * Number of grants per physical segment
249  */
250 #define GRANTS_PER_PSEG (PAGE_SIZE / XEN_PAGE_SIZE)
251
252 #define GRANTS_PER_INDIRECT_FRAME \
253         (XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
254
255 #define PSEGS_PER_INDIRECT_FRAME        \
256         (GRANTS_INDIRECT_FRAME / GRANTS_PSEGS)
257
258 #define INDIRECT_GREFS(_grants)         \
259         DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
260
261 #define GREFS(_psegs)   ((_psegs) * GRANTS_PER_PSEG)
262
263 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
264 static void blkfront_gather_backend_features(struct blkfront_info *info);
265 static int negotiate_mq(struct blkfront_info *info);
266
267 static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
268 {
269         unsigned long free = rinfo->shadow_free;
270
271         BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
272         rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
273         rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
274         return free;
275 }
276
277 static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
278                               unsigned long id)
279 {
280         if (rinfo->shadow[id].req.u.rw.id != id)
281                 return -EINVAL;
282         if (rinfo->shadow[id].request == NULL)
283                 return -EINVAL;
284         rinfo->shadow[id].req.u.rw.id  = rinfo->shadow_free;
285         rinfo->shadow[id].request = NULL;
286         rinfo->shadow_free = id;
287         return 0;
288 }
289
290 static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
291 {
292         struct blkfront_info *info = rinfo->dev_info;
293         struct page *granted_page;
294         struct grant *gnt_list_entry, *n;
295         int i = 0;
296
297         while (i < num) {
298                 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
299                 if (!gnt_list_entry)
300                         goto out_of_memory;
301
302                 if (info->feature_persistent) {
303                         granted_page = alloc_page(GFP_NOIO);
304                         if (!granted_page) {
305                                 kfree(gnt_list_entry);
306                                 goto out_of_memory;
307                         }
308                         gnt_list_entry->page = granted_page;
309                 }
310
311                 gnt_list_entry->gref = GRANT_INVALID_REF;
312                 list_add(&gnt_list_entry->node, &rinfo->grants);
313                 i++;
314         }
315
316         return 0;
317
318 out_of_memory:
319         list_for_each_entry_safe(gnt_list_entry, n,
320                                  &rinfo->grants, node) {
321                 list_del(&gnt_list_entry->node);
322                 if (info->feature_persistent)
323                         __free_page(gnt_list_entry->page);
324                 kfree(gnt_list_entry);
325                 i--;
326         }
327         BUG_ON(i != 0);
328         return -ENOMEM;
329 }
330
331 static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
332 {
333         struct grant *gnt_list_entry;
334
335         BUG_ON(list_empty(&rinfo->grants));
336         gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
337                                           node);
338         list_del(&gnt_list_entry->node);
339
340         if (gnt_list_entry->gref != GRANT_INVALID_REF)
341                 rinfo->persistent_gnts_c--;
342
343         return gnt_list_entry;
344 }
345
346 static inline void grant_foreign_access(const struct grant *gnt_list_entry,
347                                         const struct blkfront_info *info)
348 {
349         gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
350                                                  info->xbdev->otherend_id,
351                                                  gnt_list_entry->page,
352                                                  0);
353 }
354
355 static struct grant *get_grant(grant_ref_t *gref_head,
356                                unsigned long gfn,
357                                struct blkfront_ring_info *rinfo)
358 {
359         struct grant *gnt_list_entry = get_free_grant(rinfo);
360         struct blkfront_info *info = rinfo->dev_info;
361
362         if (gnt_list_entry->gref != GRANT_INVALID_REF)
363                 return gnt_list_entry;
364
365         /* Assign a gref to this page */
366         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
367         BUG_ON(gnt_list_entry->gref == -ENOSPC);
368         if (info->feature_persistent)
369                 grant_foreign_access(gnt_list_entry, info);
370         else {
371                 /* Grant access to the GFN passed by the caller */
372                 gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
373                                                 info->xbdev->otherend_id,
374                                                 gfn, 0);
375         }
376
377         return gnt_list_entry;
378 }
379
380 static struct grant *get_indirect_grant(grant_ref_t *gref_head,
381                                         struct blkfront_ring_info *rinfo)
382 {
383         struct grant *gnt_list_entry = get_free_grant(rinfo);
384         struct blkfront_info *info = rinfo->dev_info;
385
386         if (gnt_list_entry->gref != GRANT_INVALID_REF)
387                 return gnt_list_entry;
388
389         /* Assign a gref to this page */
390         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
391         BUG_ON(gnt_list_entry->gref == -ENOSPC);
392         if (!info->feature_persistent) {
393                 struct page *indirect_page;
394
395                 /* Fetch a pre-allocated page to use for indirect grefs */
396                 BUG_ON(list_empty(&rinfo->indirect_pages));
397                 indirect_page = list_first_entry(&rinfo->indirect_pages,
398                                                  struct page, lru);
399                 list_del(&indirect_page->lru);
400                 gnt_list_entry->page = indirect_page;
401         }
402         grant_foreign_access(gnt_list_entry, info);
403
404         return gnt_list_entry;
405 }
406
407 static const char *op_name(int op)
408 {
409         static const char *const names[] = {
410                 [BLKIF_OP_READ] = "read",
411                 [BLKIF_OP_WRITE] = "write",
412                 [BLKIF_OP_WRITE_BARRIER] = "barrier",
413                 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
414                 [BLKIF_OP_DISCARD] = "discard" };
415
416         if (op < 0 || op >= ARRAY_SIZE(names))
417                 return "unknown";
418
419         if (!names[op])
420                 return "reserved";
421
422         return names[op];
423 }
424 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
425 {
426         unsigned int end = minor + nr;
427         int rc;
428
429         if (end > nr_minors) {
430                 unsigned long *bitmap, *old;
431
432                 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
433                                  GFP_KERNEL);
434                 if (bitmap == NULL)
435                         return -ENOMEM;
436
437                 spin_lock(&minor_lock);
438                 if (end > nr_minors) {
439                         old = minors;
440                         memcpy(bitmap, minors,
441                                BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
442                         minors = bitmap;
443                         nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
444                 } else
445                         old = bitmap;
446                 spin_unlock(&minor_lock);
447                 kfree(old);
448         }
449
450         spin_lock(&minor_lock);
451         if (find_next_bit(minors, end, minor) >= end) {
452                 bitmap_set(minors, minor, nr);
453                 rc = 0;
454         } else
455                 rc = -EBUSY;
456         spin_unlock(&minor_lock);
457
458         return rc;
459 }
460
461 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
462 {
463         unsigned int end = minor + nr;
464
465         BUG_ON(end > nr_minors);
466         spin_lock(&minor_lock);
467         bitmap_clear(minors,  minor, nr);
468         spin_unlock(&minor_lock);
469 }
470
471 static void blkif_restart_queue_callback(void *arg)
472 {
473         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
474         schedule_work(&rinfo->work);
475 }
476
477 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
478 {
479         /* We don't have real geometry info, but let's at least return
480            values consistent with the size of the device */
481         sector_t nsect = get_capacity(bd->bd_disk);
482         sector_t cylinders = nsect;
483
484         hg->heads = 0xff;
485         hg->sectors = 0x3f;
486         sector_div(cylinders, hg->heads * hg->sectors);
487         hg->cylinders = cylinders;
488         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
489                 hg->cylinders = 0xffff;
490         return 0;
491 }
492
493 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
494                        unsigned command, unsigned long argument)
495 {
496         struct blkfront_info *info = bdev->bd_disk->private_data;
497         int i;
498
499         dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
500                 command, (long)argument);
501
502         switch (command) {
503         case CDROMMULTISESSION:
504                 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
505                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
506                         if (put_user(0, (char __user *)(argument + i)))
507                                 return -EFAULT;
508                 return 0;
509
510         case CDROM_GET_CAPABILITY: {
511                 struct gendisk *gd = info->gd;
512                 if (gd->flags & GENHD_FL_CD)
513                         return 0;
514                 return -EINVAL;
515         }
516
517         default:
518                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
519                   command);*/
520                 return -EINVAL; /* same return as native Linux */
521         }
522
523         return 0;
524 }
525
526 static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
527                                             struct request *req,
528                                             struct blkif_request **ring_req)
529 {
530         unsigned long id;
531
532         *ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
533         rinfo->ring.req_prod_pvt++;
534
535         id = get_id_from_freelist(rinfo);
536         rinfo->shadow[id].request = req;
537         rinfo->shadow[id].status = REQ_WAITING;
538         rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
539
540         (*ring_req)->u.rw.id = id;
541
542         return id;
543 }
544
545 static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
546 {
547         struct blkfront_info *info = rinfo->dev_info;
548         struct blkif_request *ring_req;
549         unsigned long id;
550
551         /* Fill out a communications ring structure. */
552         id = blkif_ring_get_request(rinfo, req, &ring_req);
553
554         ring_req->operation = BLKIF_OP_DISCARD;
555         ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
556         ring_req->u.discard.id = id;
557         ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
558         if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard)
559                 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
560         else
561                 ring_req->u.discard.flag = 0;
562
563         /* Keep a private copy so we can reissue requests when recovering. */
564         rinfo->shadow[id].req = *ring_req;
565
566         return 0;
567 }
568
569 struct setup_rw_req {
570         unsigned int grant_idx;
571         struct blkif_request_segment *segments;
572         struct blkfront_ring_info *rinfo;
573         struct blkif_request *ring_req;
574         grant_ref_t gref_head;
575         unsigned int id;
576         /* Only used when persistent grant is used and it's a read request */
577         bool need_copy;
578         unsigned int bvec_off;
579         char *bvec_data;
580
581         bool require_extra_req;
582         struct blkif_request *extra_ring_req;
583 };
584
585 static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
586                                      unsigned int len, void *data)
587 {
588         struct setup_rw_req *setup = data;
589         int n, ref;
590         struct grant *gnt_list_entry;
591         unsigned int fsect, lsect;
592         /* Convenient aliases */
593         unsigned int grant_idx = setup->grant_idx;
594         struct blkif_request *ring_req = setup->ring_req;
595         struct blkfront_ring_info *rinfo = setup->rinfo;
596         /*
597          * We always use the shadow of the first request to store the list
598          * of grant associated to the block I/O request. This made the
599          * completion more easy to handle even if the block I/O request is
600          * split.
601          */
602         struct blk_shadow *shadow = &rinfo->shadow[setup->id];
603
604         if (unlikely(setup->require_extra_req &&
605                      grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
606                 /*
607                  * We are using the second request, setup grant_idx
608                  * to be the index of the segment array.
609                  */
610                 grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
611                 ring_req = setup->extra_ring_req;
612         }
613
614         if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
615             (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
616                 if (setup->segments)
617                         kunmap_atomic(setup->segments);
618
619                 n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
620                 gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
621                 shadow->indirect_grants[n] = gnt_list_entry;
622                 setup->segments = kmap_atomic(gnt_list_entry->page);
623                 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
624         }
625
626         gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
627         ref = gnt_list_entry->gref;
628         /*
629          * All the grants are stored in the shadow of the first
630          * request. Therefore we have to use the global index.
631          */
632         shadow->grants_used[setup->grant_idx] = gnt_list_entry;
633
634         if (setup->need_copy) {
635                 void *shared_data;
636
637                 shared_data = kmap_atomic(gnt_list_entry->page);
638                 /*
639                  * this does not wipe data stored outside the
640                  * range sg->offset..sg->offset+sg->length.
641                  * Therefore, blkback *could* see data from
642                  * previous requests. This is OK as long as
643                  * persistent grants are shared with just one
644                  * domain. It may need refactoring if this
645                  * changes
646                  */
647                 memcpy(shared_data + offset,
648                        setup->bvec_data + setup->bvec_off,
649                        len);
650
651                 kunmap_atomic(shared_data);
652                 setup->bvec_off += len;
653         }
654
655         fsect = offset >> 9;
656         lsect = fsect + (len >> 9) - 1;
657         if (ring_req->operation != BLKIF_OP_INDIRECT) {
658                 ring_req->u.rw.seg[grant_idx] =
659                         (struct blkif_request_segment) {
660                                 .gref       = ref,
661                                 .first_sect = fsect,
662                                 .last_sect  = lsect };
663         } else {
664                 setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
665                         (struct blkif_request_segment) {
666                                 .gref       = ref,
667                                 .first_sect = fsect,
668                                 .last_sect  = lsect };
669         }
670
671         (setup->grant_idx)++;
672 }
673
674 static void blkif_setup_extra_req(struct blkif_request *first,
675                                   struct blkif_request *second)
676 {
677         uint16_t nr_segments = first->u.rw.nr_segments;
678
679         /*
680          * The second request is only present when the first request uses
681          * all its segments. It's always the continuity of the first one.
682          */
683         first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
684
685         second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
686         second->u.rw.sector_number = first->u.rw.sector_number +
687                 (BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
688
689         second->u.rw.handle = first->u.rw.handle;
690         second->operation = first->operation;
691 }
692
693 static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
694 {
695         struct blkfront_info *info = rinfo->dev_info;
696         struct blkif_request *ring_req, *extra_ring_req = NULL;
697         unsigned long id, extra_id = NO_ASSOCIATED_ID;
698         bool require_extra_req = false;
699         int i;
700         struct setup_rw_req setup = {
701                 .grant_idx = 0,
702                 .segments = NULL,
703                 .rinfo = rinfo,
704                 .need_copy = rq_data_dir(req) && info->feature_persistent,
705         };
706
707         /*
708          * Used to store if we are able to queue the request by just using
709          * existing persistent grants, or if we have to get new grants,
710          * as there are not sufficiently many free.
711          */
712         bool new_persistent_gnts = false;
713         struct scatterlist *sg;
714         int num_sg, max_grefs, num_grant;
715
716         max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
717         if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
718                 /*
719                  * If we are using indirect segments we need to account
720                  * for the indirect grefs used in the request.
721                  */
722                 max_grefs += INDIRECT_GREFS(max_grefs);
723
724         /* Check if we have enough persistent grants to allocate a requests */
725         if (rinfo->persistent_gnts_c < max_grefs) {
726                 new_persistent_gnts = true;
727
728                 if (gnttab_alloc_grant_references(
729                     max_grefs - rinfo->persistent_gnts_c,
730                     &setup.gref_head) < 0) {
731                         gnttab_request_free_callback(
732                                 &rinfo->callback,
733                                 blkif_restart_queue_callback,
734                                 rinfo,
735                                 max_grefs - rinfo->persistent_gnts_c);
736                         return 1;
737                 }
738         }
739
740         /* Fill out a communications ring structure. */
741         id = blkif_ring_get_request(rinfo, req, &ring_req);
742
743         num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
744         num_grant = 0;
745         /* Calculate the number of grant used */
746         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
747                num_grant += gnttab_count_grant(sg->offset, sg->length);
748
749         require_extra_req = info->max_indirect_segments == 0 &&
750                 num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
751         BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
752
753         rinfo->shadow[id].num_sg = num_sg;
754         if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
755             likely(!require_extra_req)) {
756                 /*
757                  * The indirect operation can only be a BLKIF_OP_READ or
758                  * BLKIF_OP_WRITE
759                  */
760                 BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA);
761                 ring_req->operation = BLKIF_OP_INDIRECT;
762                 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
763                         BLKIF_OP_WRITE : BLKIF_OP_READ;
764                 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
765                 ring_req->u.indirect.handle = info->handle;
766                 ring_req->u.indirect.nr_segments = num_grant;
767         } else {
768                 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
769                 ring_req->u.rw.handle = info->handle;
770                 ring_req->operation = rq_data_dir(req) ?
771                         BLKIF_OP_WRITE : BLKIF_OP_READ;
772                 if (req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA) {
773                         /*
774                          * Ideally we can do an unordered flush-to-disk.
775                          * In case the backend onlysupports barriers, use that.
776                          * A barrier request a superset of FUA, so we can
777                          * implement it the same way.  (It's also a FLUSH+FUA,
778                          * since it is guaranteed ordered WRT previous writes.)
779                          */
780                         if (info->feature_flush && info->feature_fua)
781                                 ring_req->operation =
782                                         BLKIF_OP_WRITE_BARRIER;
783                         else if (info->feature_flush)
784                                 ring_req->operation =
785                                         BLKIF_OP_FLUSH_DISKCACHE;
786                         else
787                                 ring_req->operation = 0;
788                 }
789                 ring_req->u.rw.nr_segments = num_grant;
790                 if (unlikely(require_extra_req)) {
791                         extra_id = blkif_ring_get_request(rinfo, req,
792                                                           &extra_ring_req);
793                         /*
794                          * Only the first request contains the scatter-gather
795                          * list.
796                          */
797                         rinfo->shadow[extra_id].num_sg = 0;
798
799                         blkif_setup_extra_req(ring_req, extra_ring_req);
800
801                         /* Link the 2 requests together */
802                         rinfo->shadow[extra_id].associated_id = id;
803                         rinfo->shadow[id].associated_id = extra_id;
804                 }
805         }
806
807         setup.ring_req = ring_req;
808         setup.id = id;
809
810         setup.require_extra_req = require_extra_req;
811         if (unlikely(require_extra_req))
812                 setup.extra_ring_req = extra_ring_req;
813
814         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
815                 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
816
817                 if (setup.need_copy) {
818                         setup.bvec_off = sg->offset;
819                         setup.bvec_data = kmap_atomic(sg_page(sg));
820                 }
821
822                 gnttab_foreach_grant_in_range(sg_page(sg),
823                                               sg->offset,
824                                               sg->length,
825                                               blkif_setup_rw_req_grant,
826                                               &setup);
827
828                 if (setup.need_copy)
829                         kunmap_atomic(setup.bvec_data);
830         }
831         if (setup.segments)
832                 kunmap_atomic(setup.segments);
833
834         /* Keep a private copy so we can reissue requests when recovering. */
835         rinfo->shadow[id].req = *ring_req;
836         if (unlikely(require_extra_req))
837                 rinfo->shadow[extra_id].req = *extra_ring_req;
838
839         if (new_persistent_gnts)
840                 gnttab_free_grant_references(setup.gref_head);
841
842         return 0;
843 }
844
845 /*
846  * Generate a Xen blkfront IO request from a blk layer request.  Reads
847  * and writes are handled as expected.
848  *
849  * @req: a request struct
850  */
851 static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
852 {
853         if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
854                 return 1;
855
856         if (unlikely(req_op(req) == REQ_OP_DISCARD ||
857                      req_op(req) == REQ_OP_SECURE_ERASE))
858                 return blkif_queue_discard_req(req, rinfo);
859         else
860                 return blkif_queue_rw_req(req, rinfo);
861 }
862
863 static inline void flush_requests(struct blkfront_ring_info *rinfo)
864 {
865         int notify;
866
867         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
868
869         if (notify)
870                 notify_remote_via_irq(rinfo->irq);
871 }
872
873 static inline bool blkif_request_flush_invalid(struct request *req,
874                                                struct blkfront_info *info)
875 {
876         return (blk_rq_is_passthrough(req) ||
877                 ((req_op(req) == REQ_OP_FLUSH) &&
878                  !info->feature_flush) ||
879                 ((req->cmd_flags & REQ_FUA) &&
880                  !info->feature_fua));
881 }
882
883 static blk_status_t blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
884                           const struct blk_mq_queue_data *qd)
885 {
886         unsigned long flags;
887         int qid = hctx->queue_num;
888         struct blkfront_info *info = hctx->queue->queuedata;
889         struct blkfront_ring_info *rinfo = NULL;
890
891         BUG_ON(info->nr_rings <= qid);
892         rinfo = &info->rinfo[qid];
893         blk_mq_start_request(qd->rq);
894         spin_lock_irqsave(&rinfo->ring_lock, flags);
895         if (RING_FULL(&rinfo->ring))
896                 goto out_busy;
897
898         if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
899                 goto out_err;
900
901         if (blkif_queue_request(qd->rq, rinfo))
902                 goto out_busy;
903
904         flush_requests(rinfo);
905         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
906         return BLK_STS_OK;
907
908 out_err:
909         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
910         return BLK_STS_IOERR;
911
912 out_busy:
913         blk_mq_stop_hw_queue(hctx);
914         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
915         return BLK_STS_DEV_RESOURCE;
916 }
917
918 static void blkif_complete_rq(struct request *rq)
919 {
920         blk_mq_end_request(rq, blkif_req(rq)->error);
921 }
922
923 static const struct blk_mq_ops blkfront_mq_ops = {
924         .queue_rq = blkif_queue_rq,
925         .complete = blkif_complete_rq,
926 };
927
928 static void blkif_set_queue_limits(struct blkfront_info *info)
929 {
930         struct request_queue *rq = info->rq;
931         struct gendisk *gd = info->gd;
932         unsigned int segments = info->max_indirect_segments ? :
933                                 BLKIF_MAX_SEGMENTS_PER_REQUEST;
934
935         queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
936
937         if (info->feature_discard) {
938                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
939                 blk_queue_max_discard_sectors(rq, get_capacity(gd));
940                 rq->limits.discard_granularity = info->discard_granularity;
941                 rq->limits.discard_alignment = info->discard_alignment;
942                 if (info->feature_secdiscard)
943                         queue_flag_set_unlocked(QUEUE_FLAG_SECERASE, rq);
944         }
945
946         /* Hard sector size and max sectors impersonate the equiv. hardware. */
947         blk_queue_logical_block_size(rq, info->sector_size);
948         blk_queue_physical_block_size(rq, info->physical_sector_size);
949         blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
950
951         /* Each segment in a request is up to an aligned page in size. */
952         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
953         blk_queue_max_segment_size(rq, PAGE_SIZE);
954
955         /* Ensure a merged request will fit in a single I/O ring slot. */
956         blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
957
958         /* Make sure buffer addresses are sector-aligned. */
959         blk_queue_dma_alignment(rq, 511);
960 }
961
962 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
963                                 unsigned int physical_sector_size)
964 {
965         struct request_queue *rq;
966         struct blkfront_info *info = gd->private_data;
967
968         memset(&info->tag_set, 0, sizeof(info->tag_set));
969         info->tag_set.ops = &blkfront_mq_ops;
970         info->tag_set.nr_hw_queues = info->nr_rings;
971         if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
972                 /*
973                  * When indirect descriptior is not supported, the I/O request
974                  * will be split between multiple request in the ring.
975                  * To avoid problems when sending the request, divide by
976                  * 2 the depth of the queue.
977                  */
978                 info->tag_set.queue_depth =  BLK_RING_SIZE(info) / 2;
979         } else
980                 info->tag_set.queue_depth = BLK_RING_SIZE(info);
981         info->tag_set.numa_node = NUMA_NO_NODE;
982         info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
983         info->tag_set.cmd_size = sizeof(struct blkif_req);
984         info->tag_set.driver_data = info;
985
986         if (blk_mq_alloc_tag_set(&info->tag_set))
987                 return -EINVAL;
988         rq = blk_mq_init_queue(&info->tag_set);
989         if (IS_ERR(rq)) {
990                 blk_mq_free_tag_set(&info->tag_set);
991                 return PTR_ERR(rq);
992         }
993
994         rq->queuedata = info;
995         info->rq = gd->queue = rq;
996         info->gd = gd;
997         info->sector_size = sector_size;
998         info->physical_sector_size = physical_sector_size;
999         blkif_set_queue_limits(info);
1000
1001         return 0;
1002 }
1003
1004 static const char *flush_info(struct blkfront_info *info)
1005 {
1006         if (info->feature_flush && info->feature_fua)
1007                 return "barrier: enabled;";
1008         else if (info->feature_flush)
1009                 return "flush diskcache: enabled;";
1010         else
1011                 return "barrier or flush: disabled;";
1012 }
1013
1014 static void xlvbd_flush(struct blkfront_info *info)
1015 {
1016         blk_queue_write_cache(info->rq, info->feature_flush ? true : false,
1017                               info->feature_fua ? true : false);
1018         pr_info("blkfront: %s: %s %s %s %s %s\n",
1019                 info->gd->disk_name, flush_info(info),
1020                 "persistent grants:", info->feature_persistent ?
1021                 "enabled;" : "disabled;", "indirect descriptors:",
1022                 info->max_indirect_segments ? "enabled;" : "disabled;");
1023 }
1024
1025 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
1026 {
1027         int major;
1028         major = BLKIF_MAJOR(vdevice);
1029         *minor = BLKIF_MINOR(vdevice);
1030         switch (major) {
1031                 case XEN_IDE0_MAJOR:
1032                         *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
1033                         *minor = ((*minor / 64) * PARTS_PER_DISK) +
1034                                 EMULATED_HD_DISK_MINOR_OFFSET;
1035                         break;
1036                 case XEN_IDE1_MAJOR:
1037                         *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1038                         *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1039                                 EMULATED_HD_DISK_MINOR_OFFSET;
1040                         break;
1041                 case XEN_SCSI_DISK0_MAJOR:
1042                         *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1043                         *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1044                         break;
1045                 case XEN_SCSI_DISK1_MAJOR:
1046                 case XEN_SCSI_DISK2_MAJOR:
1047                 case XEN_SCSI_DISK3_MAJOR:
1048                 case XEN_SCSI_DISK4_MAJOR:
1049                 case XEN_SCSI_DISK5_MAJOR:
1050                 case XEN_SCSI_DISK6_MAJOR:
1051                 case XEN_SCSI_DISK7_MAJOR:
1052                         *offset = (*minor / PARTS_PER_DISK) + 
1053                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1054                                 EMULATED_SD_DISK_NAME_OFFSET;
1055                         *minor = *minor +
1056                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1057                                 EMULATED_SD_DISK_MINOR_OFFSET;
1058                         break;
1059                 case XEN_SCSI_DISK8_MAJOR:
1060                 case XEN_SCSI_DISK9_MAJOR:
1061                 case XEN_SCSI_DISK10_MAJOR:
1062                 case XEN_SCSI_DISK11_MAJOR:
1063                 case XEN_SCSI_DISK12_MAJOR:
1064                 case XEN_SCSI_DISK13_MAJOR:
1065                 case XEN_SCSI_DISK14_MAJOR:
1066                 case XEN_SCSI_DISK15_MAJOR:
1067                         *offset = (*minor / PARTS_PER_DISK) + 
1068                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1069                                 EMULATED_SD_DISK_NAME_OFFSET;
1070                         *minor = *minor +
1071                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1072                                 EMULATED_SD_DISK_MINOR_OFFSET;
1073                         break;
1074                 case XENVBD_MAJOR:
1075                         *offset = *minor / PARTS_PER_DISK;
1076                         break;
1077                 default:
1078                         printk(KERN_WARNING "blkfront: your disk configuration is "
1079                                         "incorrect, please use an xvd device instead\n");
1080                         return -ENODEV;
1081         }
1082         return 0;
1083 }
1084
1085 static char *encode_disk_name(char *ptr, unsigned int n)
1086 {
1087         if (n >= 26)
1088                 ptr = encode_disk_name(ptr, n / 26 - 1);
1089         *ptr = 'a' + n % 26;
1090         return ptr + 1;
1091 }
1092
1093 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1094                                struct blkfront_info *info,
1095                                u16 vdisk_info, u16 sector_size,
1096                                unsigned int physical_sector_size)
1097 {
1098         struct gendisk *gd;
1099         int nr_minors = 1;
1100         int err;
1101         unsigned int offset;
1102         int minor;
1103         int nr_parts;
1104         char *ptr;
1105
1106         BUG_ON(info->gd != NULL);
1107         BUG_ON(info->rq != NULL);
1108
1109         if ((info->vdevice>>EXT_SHIFT) > 1) {
1110                 /* this is above the extended range; something is wrong */
1111                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1112                 return -ENODEV;
1113         }
1114
1115         if (!VDEV_IS_EXTENDED(info->vdevice)) {
1116                 err = xen_translate_vdev(info->vdevice, &minor, &offset);
1117                 if (err)
1118                         return err;             
1119                 nr_parts = PARTS_PER_DISK;
1120         } else {
1121                 minor = BLKIF_MINOR_EXT(info->vdevice);
1122                 nr_parts = PARTS_PER_EXT_DISK;
1123                 offset = minor / nr_parts;
1124                 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1125                         printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1126                                         "emulated IDE disks,\n\t choose an xvd device name"
1127                                         "from xvde on\n", info->vdevice);
1128         }
1129         if (minor >> MINORBITS) {
1130                 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1131                         info->vdevice, minor);
1132                 return -ENODEV;
1133         }
1134
1135         if ((minor % nr_parts) == 0)
1136                 nr_minors = nr_parts;
1137
1138         err = xlbd_reserve_minors(minor, nr_minors);
1139         if (err)
1140                 goto out;
1141         err = -ENODEV;
1142
1143         gd = alloc_disk(nr_minors);
1144         if (gd == NULL)
1145                 goto release;
1146
1147         strcpy(gd->disk_name, DEV_NAME);
1148         ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1149         BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1150         if (nr_minors > 1)
1151                 *ptr = 0;
1152         else
1153                 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1154                          "%d", minor & (nr_parts - 1));
1155
1156         gd->major = XENVBD_MAJOR;
1157         gd->first_minor = minor;
1158         gd->fops = &xlvbd_block_fops;
1159         gd->private_data = info;
1160         set_capacity(gd, capacity);
1161
1162         if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size)) {
1163                 del_gendisk(gd);
1164                 goto release;
1165         }
1166
1167         xlvbd_flush(info);
1168
1169         if (vdisk_info & VDISK_READONLY)
1170                 set_disk_ro(gd, 1);
1171
1172         if (vdisk_info & VDISK_REMOVABLE)
1173                 gd->flags |= GENHD_FL_REMOVABLE;
1174
1175         if (vdisk_info & VDISK_CDROM)
1176                 gd->flags |= GENHD_FL_CD;
1177
1178         return 0;
1179
1180  release:
1181         xlbd_release_minors(minor, nr_minors);
1182  out:
1183         return err;
1184 }
1185
1186 static void xlvbd_release_gendisk(struct blkfront_info *info)
1187 {
1188         unsigned int minor, nr_minors, i;
1189
1190         if (info->rq == NULL)
1191                 return;
1192
1193         /* No more blkif_request(). */
1194         blk_mq_stop_hw_queues(info->rq);
1195
1196         for (i = 0; i < info->nr_rings; i++) {
1197                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1198
1199                 /* No more gnttab callback work. */
1200                 gnttab_cancel_free_callback(&rinfo->callback);
1201
1202                 /* Flush gnttab callback work. Must be done with no locks held. */
1203                 flush_work(&rinfo->work);
1204         }
1205
1206         del_gendisk(info->gd);
1207
1208         minor = info->gd->first_minor;
1209         nr_minors = info->gd->minors;
1210         xlbd_release_minors(minor, nr_minors);
1211
1212         blk_cleanup_queue(info->rq);
1213         blk_mq_free_tag_set(&info->tag_set);
1214         info->rq = NULL;
1215
1216         put_disk(info->gd);
1217         info->gd = NULL;
1218 }
1219
1220 /* Already hold rinfo->ring_lock. */
1221 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1222 {
1223         if (!RING_FULL(&rinfo->ring))
1224                 blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1225 }
1226
1227 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1228 {
1229         unsigned long flags;
1230
1231         spin_lock_irqsave(&rinfo->ring_lock, flags);
1232         kick_pending_request_queues_locked(rinfo);
1233         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1234 }
1235
1236 static void blkif_restart_queue(struct work_struct *work)
1237 {
1238         struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1239
1240         if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1241                 kick_pending_request_queues(rinfo);
1242 }
1243
1244 static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1245 {
1246         struct grant *persistent_gnt, *n;
1247         struct blkfront_info *info = rinfo->dev_info;
1248         int i, j, segs;
1249
1250         /*
1251          * Remove indirect pages, this only happens when using indirect
1252          * descriptors but not persistent grants
1253          */
1254         if (!list_empty(&rinfo->indirect_pages)) {
1255                 struct page *indirect_page, *n;
1256
1257                 BUG_ON(info->feature_persistent);
1258                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1259                         list_del(&indirect_page->lru);
1260                         __free_page(indirect_page);
1261                 }
1262         }
1263
1264         /* Remove all persistent grants. */
1265         if (!list_empty(&rinfo->grants)) {
1266                 list_for_each_entry_safe(persistent_gnt, n,
1267                                          &rinfo->grants, node) {
1268                         list_del(&persistent_gnt->node);
1269                         if (persistent_gnt->gref != GRANT_INVALID_REF) {
1270                                 gnttab_end_foreign_access(persistent_gnt->gref,
1271                                                           0, 0UL);
1272                                 rinfo->persistent_gnts_c--;
1273                         }
1274                         if (info->feature_persistent)
1275                                 __free_page(persistent_gnt->page);
1276                         kfree(persistent_gnt);
1277                 }
1278         }
1279         BUG_ON(rinfo->persistent_gnts_c != 0);
1280
1281         for (i = 0; i < BLK_RING_SIZE(info); i++) {
1282                 /*
1283                  * Clear persistent grants present in requests already
1284                  * on the shared ring
1285                  */
1286                 if (!rinfo->shadow[i].request)
1287                         goto free_shadow;
1288
1289                 segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1290                        rinfo->shadow[i].req.u.indirect.nr_segments :
1291                        rinfo->shadow[i].req.u.rw.nr_segments;
1292                 for (j = 0; j < segs; j++) {
1293                         persistent_gnt = rinfo->shadow[i].grants_used[j];
1294                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1295                         if (info->feature_persistent)
1296                                 __free_page(persistent_gnt->page);
1297                         kfree(persistent_gnt);
1298                 }
1299
1300                 if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1301                         /*
1302                          * If this is not an indirect operation don't try to
1303                          * free indirect segments
1304                          */
1305                         goto free_shadow;
1306
1307                 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1308                         persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1309                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1310                         __free_page(persistent_gnt->page);
1311                         kfree(persistent_gnt);
1312                 }
1313
1314 free_shadow:
1315                 kfree(rinfo->shadow[i].grants_used);
1316                 rinfo->shadow[i].grants_used = NULL;
1317                 kfree(rinfo->shadow[i].indirect_grants);
1318                 rinfo->shadow[i].indirect_grants = NULL;
1319                 kfree(rinfo->shadow[i].sg);
1320                 rinfo->shadow[i].sg = NULL;
1321         }
1322
1323         /* No more gnttab callback work. */
1324         gnttab_cancel_free_callback(&rinfo->callback);
1325
1326         /* Flush gnttab callback work. Must be done with no locks held. */
1327         flush_work(&rinfo->work);
1328
1329         /* Free resources associated with old device channel. */
1330         for (i = 0; i < info->nr_ring_pages; i++) {
1331                 if (rinfo->ring_ref[i] != GRANT_INVALID_REF) {
1332                         gnttab_end_foreign_access(rinfo->ring_ref[i], 0, 0);
1333                         rinfo->ring_ref[i] = GRANT_INVALID_REF;
1334                 }
1335         }
1336         free_pages((unsigned long)rinfo->ring.sring, get_order(info->nr_ring_pages * XEN_PAGE_SIZE));
1337         rinfo->ring.sring = NULL;
1338
1339         if (rinfo->irq)
1340                 unbind_from_irqhandler(rinfo->irq, rinfo);
1341         rinfo->evtchn = rinfo->irq = 0;
1342 }
1343
1344 static void blkif_free(struct blkfront_info *info, int suspend)
1345 {
1346         unsigned int i;
1347
1348         /* Prevent new requests being issued until we fix things up. */
1349         info->connected = suspend ?
1350                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1351         /* No more blkif_request(). */
1352         if (info->rq)
1353                 blk_mq_stop_hw_queues(info->rq);
1354
1355         for (i = 0; i < info->nr_rings; i++)
1356                 blkif_free_ring(&info->rinfo[i]);
1357
1358         kfree(info->rinfo);
1359         info->rinfo = NULL;
1360         info->nr_rings = 0;
1361 }
1362
1363 struct copy_from_grant {
1364         const struct blk_shadow *s;
1365         unsigned int grant_idx;
1366         unsigned int bvec_offset;
1367         char *bvec_data;
1368 };
1369
1370 static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1371                                   unsigned int len, void *data)
1372 {
1373         struct copy_from_grant *info = data;
1374         char *shared_data;
1375         /* Convenient aliases */
1376         const struct blk_shadow *s = info->s;
1377
1378         shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1379
1380         memcpy(info->bvec_data + info->bvec_offset,
1381                shared_data + offset, len);
1382
1383         info->bvec_offset += len;
1384         info->grant_idx++;
1385
1386         kunmap_atomic(shared_data);
1387 }
1388
1389 static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1390 {
1391         switch (rsp)
1392         {
1393         case BLKIF_RSP_OKAY:
1394                 return REQ_DONE;
1395         case BLKIF_RSP_EOPNOTSUPP:
1396                 return REQ_EOPNOTSUPP;
1397         case BLKIF_RSP_ERROR:
1398                 /* Fallthrough. */
1399         default:
1400                 return REQ_ERROR;
1401         }
1402 }
1403
1404 /*
1405  * Get the final status of the block request based on two ring response
1406  */
1407 static int blkif_get_final_status(enum blk_req_status s1,
1408                                   enum blk_req_status s2)
1409 {
1410         BUG_ON(s1 == REQ_WAITING);
1411         BUG_ON(s2 == REQ_WAITING);
1412
1413         if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1414                 return BLKIF_RSP_ERROR;
1415         else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1416                 return BLKIF_RSP_EOPNOTSUPP;
1417         return BLKIF_RSP_OKAY;
1418 }
1419
1420 static bool blkif_completion(unsigned long *id,
1421                              struct blkfront_ring_info *rinfo,
1422                              struct blkif_response *bret)
1423 {
1424         int i = 0;
1425         struct scatterlist *sg;
1426         int num_sg, num_grant;
1427         struct blkfront_info *info = rinfo->dev_info;
1428         struct blk_shadow *s = &rinfo->shadow[*id];
1429         struct copy_from_grant data = {
1430                 .grant_idx = 0,
1431         };
1432
1433         num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1434                 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1435
1436         /* The I/O request may be split in two. */
1437         if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1438                 struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1439
1440                 /* Keep the status of the current response in shadow. */
1441                 s->status = blkif_rsp_to_req_status(bret->status);
1442
1443                 /* Wait the second response if not yet here. */
1444                 if (s2->status == REQ_WAITING)
1445                         return 0;
1446
1447                 bret->status = blkif_get_final_status(s->status,
1448                                                       s2->status);
1449
1450                 /*
1451                  * All the grants is stored in the first shadow in order
1452                  * to make the completion code simpler.
1453                  */
1454                 num_grant += s2->req.u.rw.nr_segments;
1455
1456                 /*
1457                  * The two responses may not come in order. Only the
1458                  * first request will store the scatter-gather list.
1459                  */
1460                 if (s2->num_sg != 0) {
1461                         /* Update "id" with the ID of the first response. */
1462                         *id = s->associated_id;
1463                         s = s2;
1464                 }
1465
1466                 /*
1467                  * We don't need anymore the second request, so recycling
1468                  * it now.
1469                  */
1470                 if (add_id_to_freelist(rinfo, s->associated_id))
1471                         WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1472                              info->gd->disk_name, s->associated_id);
1473         }
1474
1475         data.s = s;
1476         num_sg = s->num_sg;
1477
1478         if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1479                 for_each_sg(s->sg, sg, num_sg, i) {
1480                         BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1481
1482                         data.bvec_offset = sg->offset;
1483                         data.bvec_data = kmap_atomic(sg_page(sg));
1484
1485                         gnttab_foreach_grant_in_range(sg_page(sg),
1486                                                       sg->offset,
1487                                                       sg->length,
1488                                                       blkif_copy_from_grant,
1489                                                       &data);
1490
1491                         kunmap_atomic(data.bvec_data);
1492                 }
1493         }
1494         /* Add the persistent grant into the list of free grants */
1495         for (i = 0; i < num_grant; i++) {
1496                 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1497                         /*
1498                          * If the grant is still mapped by the backend (the
1499                          * backend has chosen to make this grant persistent)
1500                          * we add it at the head of the list, so it will be
1501                          * reused first.
1502                          */
1503                         if (!info->feature_persistent)
1504                                 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1505                                                      s->grants_used[i]->gref);
1506                         list_add(&s->grants_used[i]->node, &rinfo->grants);
1507                         rinfo->persistent_gnts_c++;
1508                 } else {
1509                         /*
1510                          * If the grant is not mapped by the backend we end the
1511                          * foreign access and add it to the tail of the list,
1512                          * so it will not be picked again unless we run out of
1513                          * persistent grants.
1514                          */
1515                         gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1516                         s->grants_used[i]->gref = GRANT_INVALID_REF;
1517                         list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1518                 }
1519         }
1520         if (s->req.operation == BLKIF_OP_INDIRECT) {
1521                 for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1522                         if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1523                                 if (!info->feature_persistent)
1524                                         pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1525                                                              s->indirect_grants[i]->gref);
1526                                 list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1527                                 rinfo->persistent_gnts_c++;
1528                         } else {
1529                                 struct page *indirect_page;
1530
1531                                 gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1532                                 /*
1533                                  * Add the used indirect page back to the list of
1534                                  * available pages for indirect grefs.
1535                                  */
1536                                 if (!info->feature_persistent) {
1537                                         indirect_page = s->indirect_grants[i]->page;
1538                                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
1539                                 }
1540                                 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1541                                 list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1542                         }
1543                 }
1544         }
1545
1546         return 1;
1547 }
1548
1549 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1550 {
1551         struct request *req;
1552         struct blkif_response *bret;
1553         RING_IDX i, rp;
1554         unsigned long flags;
1555         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1556         struct blkfront_info *info = rinfo->dev_info;
1557
1558         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
1559                 return IRQ_HANDLED;
1560
1561         spin_lock_irqsave(&rinfo->ring_lock, flags);
1562  again:
1563         rp = rinfo->ring.sring->rsp_prod;
1564         rmb(); /* Ensure we see queued responses up to 'rp'. */
1565
1566         for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1567                 unsigned long id;
1568
1569                 bret = RING_GET_RESPONSE(&rinfo->ring, i);
1570                 id   = bret->id;
1571                 /*
1572                  * The backend has messed up and given us an id that we would
1573                  * never have given to it (we stamp it up to BLK_RING_SIZE -
1574                  * look in get_id_from_freelist.
1575                  */
1576                 if (id >= BLK_RING_SIZE(info)) {
1577                         WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1578                              info->gd->disk_name, op_name(bret->operation), id);
1579                         /* We can't safely get the 'struct request' as
1580                          * the id is busted. */
1581                         continue;
1582                 }
1583                 req  = rinfo->shadow[id].request;
1584
1585                 if (bret->operation != BLKIF_OP_DISCARD) {
1586                         /*
1587                          * We may need to wait for an extra response if the
1588                          * I/O request is split in 2
1589                          */
1590                         if (!blkif_completion(&id, rinfo, bret))
1591                                 continue;
1592                 }
1593
1594                 if (add_id_to_freelist(rinfo, id)) {
1595                         WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1596                              info->gd->disk_name, op_name(bret->operation), id);
1597                         continue;
1598                 }
1599
1600                 if (bret->status == BLKIF_RSP_OKAY)
1601                         blkif_req(req)->error = BLK_STS_OK;
1602                 else
1603                         blkif_req(req)->error = BLK_STS_IOERR;
1604
1605                 switch (bret->operation) {
1606                 case BLKIF_OP_DISCARD:
1607                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1608                                 struct request_queue *rq = info->rq;
1609                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1610                                            info->gd->disk_name, op_name(bret->operation));
1611                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1612                                 info->feature_discard = 0;
1613                                 info->feature_secdiscard = 0;
1614                                 queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1615                                 queue_flag_clear(QUEUE_FLAG_SECERASE, rq);
1616                         }
1617                         break;
1618                 case BLKIF_OP_FLUSH_DISKCACHE:
1619                 case BLKIF_OP_WRITE_BARRIER:
1620                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1621                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1622                                        info->gd->disk_name, op_name(bret->operation));
1623                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1624                         }
1625                         if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1626                                      rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1627                                 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1628                                        info->gd->disk_name, op_name(bret->operation));
1629                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1630                         }
1631                         if (unlikely(blkif_req(req)->error)) {
1632                                 if (blkif_req(req)->error == BLK_STS_NOTSUPP)
1633                                         blkif_req(req)->error = BLK_STS_OK;
1634                                 info->feature_fua = 0;
1635                                 info->feature_flush = 0;
1636                                 xlvbd_flush(info);
1637                         }
1638                         /* fall through */
1639                 case BLKIF_OP_READ:
1640                 case BLKIF_OP_WRITE:
1641                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
1642                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1643                                         "request: %x\n", bret->status);
1644
1645                         break;
1646                 default:
1647                         BUG();
1648                 }
1649
1650                 blk_mq_complete_request(req);
1651         }
1652
1653         rinfo->ring.rsp_cons = i;
1654
1655         if (i != rinfo->ring.req_prod_pvt) {
1656                 int more_to_do;
1657                 RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1658                 if (more_to_do)
1659                         goto again;
1660         } else
1661                 rinfo->ring.sring->rsp_event = i + 1;
1662
1663         kick_pending_request_queues_locked(rinfo);
1664
1665         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1666
1667         return IRQ_HANDLED;
1668 }
1669
1670
1671 static int setup_blkring(struct xenbus_device *dev,
1672                          struct blkfront_ring_info *rinfo)
1673 {
1674         struct blkif_sring *sring;
1675         int err, i;
1676         struct blkfront_info *info = rinfo->dev_info;
1677         unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1678         grant_ref_t gref[XENBUS_MAX_RING_GRANTS];
1679
1680         for (i = 0; i < info->nr_ring_pages; i++)
1681                 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1682
1683         sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
1684                                                        get_order(ring_size));
1685         if (!sring) {
1686                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1687                 return -ENOMEM;
1688         }
1689         SHARED_RING_INIT(sring);
1690         FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1691
1692         err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref);
1693         if (err < 0) {
1694                 free_pages((unsigned long)sring, get_order(ring_size));
1695                 rinfo->ring.sring = NULL;
1696                 goto fail;
1697         }
1698         for (i = 0; i < info->nr_ring_pages; i++)
1699                 rinfo->ring_ref[i] = gref[i];
1700
1701         err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1702         if (err)
1703                 goto fail;
1704
1705         err = bind_evtchn_to_irqhandler(rinfo->evtchn, blkif_interrupt, 0,
1706                                         "blkif", rinfo);
1707         if (err <= 0) {
1708                 xenbus_dev_fatal(dev, err,
1709                                  "bind_evtchn_to_irqhandler failed");
1710                 goto fail;
1711         }
1712         rinfo->irq = err;
1713
1714         return 0;
1715 fail:
1716         blkif_free(info, 0);
1717         return err;
1718 }
1719
1720 /*
1721  * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1722  * ring buffer may have multi pages depending on ->nr_ring_pages.
1723  */
1724 static int write_per_ring_nodes(struct xenbus_transaction xbt,
1725                                 struct blkfront_ring_info *rinfo, const char *dir)
1726 {
1727         int err;
1728         unsigned int i;
1729         const char *message = NULL;
1730         struct blkfront_info *info = rinfo->dev_info;
1731
1732         if (info->nr_ring_pages == 1) {
1733                 err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1734                 if (err) {
1735                         message = "writing ring-ref";
1736                         goto abort_transaction;
1737                 }
1738         } else {
1739                 for (i = 0; i < info->nr_ring_pages; i++) {
1740                         char ring_ref_name[RINGREF_NAME_LEN];
1741
1742                         snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1743                         err = xenbus_printf(xbt, dir, ring_ref_name,
1744                                             "%u", rinfo->ring_ref[i]);
1745                         if (err) {
1746                                 message = "writing ring-ref";
1747                                 goto abort_transaction;
1748                         }
1749                 }
1750         }
1751
1752         err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1753         if (err) {
1754                 message = "writing event-channel";
1755                 goto abort_transaction;
1756         }
1757
1758         return 0;
1759
1760 abort_transaction:
1761         xenbus_transaction_end(xbt, 1);
1762         if (message)
1763                 xenbus_dev_fatal(info->xbdev, err, "%s", message);
1764
1765         return err;
1766 }
1767
1768 /* Common code used when first setting up, and when resuming. */
1769 static int talk_to_blkback(struct xenbus_device *dev,
1770                            struct blkfront_info *info)
1771 {
1772         const char *message = NULL;
1773         struct xenbus_transaction xbt;
1774         int err;
1775         unsigned int i, max_page_order;
1776         unsigned int ring_page_order;
1777
1778         if (!info)
1779                 return -ENODEV;
1780
1781         max_page_order = xenbus_read_unsigned(info->xbdev->otherend,
1782                                               "max-ring-page-order", 0);
1783         ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1784         info->nr_ring_pages = 1 << ring_page_order;
1785
1786         err = negotiate_mq(info);
1787         if (err)
1788                 goto destroy_blkring;
1789
1790         for (i = 0; i < info->nr_rings; i++) {
1791                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1792
1793                 /* Create shared ring, alloc event channel. */
1794                 err = setup_blkring(dev, rinfo);
1795                 if (err)
1796                         goto destroy_blkring;
1797         }
1798
1799 again:
1800         err = xenbus_transaction_start(&xbt);
1801         if (err) {
1802                 xenbus_dev_fatal(dev, err, "starting transaction");
1803                 goto destroy_blkring;
1804         }
1805
1806         if (info->nr_ring_pages > 1) {
1807                 err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1808                                     ring_page_order);
1809                 if (err) {
1810                         message = "writing ring-page-order";
1811                         goto abort_transaction;
1812                 }
1813         }
1814
1815         /* We already got the number of queues/rings in _probe */
1816         if (info->nr_rings == 1) {
1817                 err = write_per_ring_nodes(xbt, &info->rinfo[0], dev->nodename);
1818                 if (err)
1819                         goto destroy_blkring;
1820         } else {
1821                 char *path;
1822                 size_t pathsize;
1823
1824                 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1825                                     info->nr_rings);
1826                 if (err) {
1827                         message = "writing multi-queue-num-queues";
1828                         goto abort_transaction;
1829                 }
1830
1831                 pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1832                 path = kmalloc(pathsize, GFP_KERNEL);
1833                 if (!path) {
1834                         err = -ENOMEM;
1835                         message = "ENOMEM while writing ring references";
1836                         goto abort_transaction;
1837                 }
1838
1839                 for (i = 0; i < info->nr_rings; i++) {
1840                         memset(path, 0, pathsize);
1841                         snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1842                         err = write_per_ring_nodes(xbt, &info->rinfo[i], path);
1843                         if (err) {
1844                                 kfree(path);
1845                                 goto destroy_blkring;
1846                         }
1847                 }
1848                 kfree(path);
1849         }
1850         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1851                             XEN_IO_PROTO_ABI_NATIVE);
1852         if (err) {
1853                 message = "writing protocol";
1854                 goto abort_transaction;
1855         }
1856         err = xenbus_printf(xbt, dev->nodename,
1857                             "feature-persistent", "%u", 1);
1858         if (err)
1859                 dev_warn(&dev->dev,
1860                          "writing persistent grants feature to xenbus");
1861
1862         err = xenbus_transaction_end(xbt, 0);
1863         if (err) {
1864                 if (err == -EAGAIN)
1865                         goto again;
1866                 xenbus_dev_fatal(dev, err, "completing transaction");
1867                 goto destroy_blkring;
1868         }
1869
1870         for (i = 0; i < info->nr_rings; i++) {
1871                 unsigned int j;
1872                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1873
1874                 for (j = 0; j < BLK_RING_SIZE(info); j++)
1875                         rinfo->shadow[j].req.u.rw.id = j + 1;
1876                 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1877         }
1878         xenbus_switch_state(dev, XenbusStateInitialised);
1879
1880         return 0;
1881
1882  abort_transaction:
1883         xenbus_transaction_end(xbt, 1);
1884         if (message)
1885                 xenbus_dev_fatal(dev, err, "%s", message);
1886  destroy_blkring:
1887         blkif_free(info, 0);
1888
1889         kfree(info);
1890         dev_set_drvdata(&dev->dev, NULL);
1891
1892         return err;
1893 }
1894
1895 static int negotiate_mq(struct blkfront_info *info)
1896 {
1897         unsigned int backend_max_queues;
1898         unsigned int i;
1899
1900         BUG_ON(info->nr_rings);
1901
1902         /* Check if backend supports multiple queues. */
1903         backend_max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1904                                                   "multi-queue-max-queues", 1);
1905         info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1906         /* We need at least one ring. */
1907         if (!info->nr_rings)
1908                 info->nr_rings = 1;
1909
1910         info->rinfo = kzalloc(sizeof(struct blkfront_ring_info) * info->nr_rings, GFP_KERNEL);
1911         if (!info->rinfo) {
1912                 xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1913                 return -ENOMEM;
1914         }
1915
1916         for (i = 0; i < info->nr_rings; i++) {
1917                 struct blkfront_ring_info *rinfo;
1918
1919                 rinfo = &info->rinfo[i];
1920                 INIT_LIST_HEAD(&rinfo->indirect_pages);
1921                 INIT_LIST_HEAD(&rinfo->grants);
1922                 rinfo->dev_info = info;
1923                 INIT_WORK(&rinfo->work, blkif_restart_queue);
1924                 spin_lock_init(&rinfo->ring_lock);
1925         }
1926         return 0;
1927 }
1928 /**
1929  * Entry point to this code when a new device is created.  Allocate the basic
1930  * structures and the ring buffer for communication with the backend, and
1931  * inform the backend of the appropriate details for those.  Switch to
1932  * Initialised state.
1933  */
1934 static int blkfront_probe(struct xenbus_device *dev,
1935                           const struct xenbus_device_id *id)
1936 {
1937         int err, vdevice;
1938         struct blkfront_info *info;
1939
1940         /* FIXME: Use dynamic device id if this is not set. */
1941         err = xenbus_scanf(XBT_NIL, dev->nodename,
1942                            "virtual-device", "%i", &vdevice);
1943         if (err != 1) {
1944                 /* go looking in the extended area instead */
1945                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1946                                    "%i", &vdevice);
1947                 if (err != 1) {
1948                         xenbus_dev_fatal(dev, err, "reading virtual-device");
1949                         return err;
1950                 }
1951         }
1952
1953         if (xen_hvm_domain()) {
1954                 char *type;
1955                 int len;
1956                 /* no unplug has been done: do not hook devices != xen vbds */
1957                 if (xen_has_pv_and_legacy_disk_devices()) {
1958                         int major;
1959
1960                         if (!VDEV_IS_EXTENDED(vdevice))
1961                                 major = BLKIF_MAJOR(vdevice);
1962                         else
1963                                 major = XENVBD_MAJOR;
1964
1965                         if (major != XENVBD_MAJOR) {
1966                                 printk(KERN_INFO
1967                                                 "%s: HVM does not support vbd %d as xen block device\n",
1968                                                 __func__, vdevice);
1969                                 return -ENODEV;
1970                         }
1971                 }
1972                 /* do not create a PV cdrom device if we are an HVM guest */
1973                 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1974                 if (IS_ERR(type))
1975                         return -ENODEV;
1976                 if (strncmp(type, "cdrom", 5) == 0) {
1977                         kfree(type);
1978                         return -ENODEV;
1979                 }
1980                 kfree(type);
1981         }
1982         info = kzalloc(sizeof(*info), GFP_KERNEL);
1983         if (!info) {
1984                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1985                 return -ENOMEM;
1986         }
1987
1988         info->xbdev = dev;
1989
1990         mutex_init(&info->mutex);
1991         info->vdevice = vdevice;
1992         info->connected = BLKIF_STATE_DISCONNECTED;
1993
1994         /* Front end dir is a number, which is used as the id. */
1995         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1996         dev_set_drvdata(&dev->dev, info);
1997
1998         return 0;
1999 }
2000
2001 static int blkif_recover(struct blkfront_info *info)
2002 {
2003         unsigned int r_index;
2004         struct request *req, *n;
2005         int rc;
2006         struct bio *bio;
2007         unsigned int segs;
2008
2009         blkfront_gather_backend_features(info);
2010         /* Reset limits changed by blk_mq_update_nr_hw_queues(). */
2011         blkif_set_queue_limits(info);
2012         segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
2013         blk_queue_max_segments(info->rq, segs / GRANTS_PER_PSEG);
2014
2015         for (r_index = 0; r_index < info->nr_rings; r_index++) {
2016                 struct blkfront_ring_info *rinfo = &info->rinfo[r_index];
2017
2018                 rc = blkfront_setup_indirect(rinfo);
2019                 if (rc)
2020                         return rc;
2021         }
2022         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2023
2024         /* Now safe for us to use the shared ring */
2025         info->connected = BLKIF_STATE_CONNECTED;
2026
2027         for (r_index = 0; r_index < info->nr_rings; r_index++) {
2028                 struct blkfront_ring_info *rinfo;
2029
2030                 rinfo = &info->rinfo[r_index];
2031                 /* Kick any other new requests queued since we resumed */
2032                 kick_pending_request_queues(rinfo);
2033         }
2034
2035         list_for_each_entry_safe(req, n, &info->requests, queuelist) {
2036                 /* Requeue pending requests (flush or discard) */
2037                 list_del_init(&req->queuelist);
2038                 BUG_ON(req->nr_phys_segments > segs);
2039                 blk_mq_requeue_request(req, false);
2040         }
2041         blk_mq_start_stopped_hw_queues(info->rq, true);
2042         blk_mq_kick_requeue_list(info->rq);
2043
2044         while ((bio = bio_list_pop(&info->bio_list)) != NULL) {
2045                 /* Traverse the list of pending bios and re-queue them */
2046                 submit_bio(bio);
2047         }
2048
2049         return 0;
2050 }
2051
2052 /**
2053  * We are reconnecting to the backend, due to a suspend/resume, or a backend
2054  * driver restart.  We tear down our blkif structure and recreate it, but
2055  * leave the device-layer structures intact so that this is transparent to the
2056  * rest of the kernel.
2057  */
2058 static int blkfront_resume(struct xenbus_device *dev)
2059 {
2060         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2061         int err = 0;
2062         unsigned int i, j;
2063
2064         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2065
2066         bio_list_init(&info->bio_list);
2067         INIT_LIST_HEAD(&info->requests);
2068         for (i = 0; i < info->nr_rings; i++) {
2069                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
2070                 struct bio_list merge_bio;
2071                 struct blk_shadow *shadow = rinfo->shadow;
2072
2073                 for (j = 0; j < BLK_RING_SIZE(info); j++) {
2074                         /* Not in use? */
2075                         if (!shadow[j].request)
2076                                 continue;
2077
2078                         /*
2079                          * Get the bios in the request so we can re-queue them.
2080                          */
2081                         if (req_op(shadow[j].request) == REQ_OP_FLUSH ||
2082                             req_op(shadow[j].request) == REQ_OP_DISCARD ||
2083                             req_op(shadow[j].request) == REQ_OP_SECURE_ERASE ||
2084                             shadow[j].request->cmd_flags & REQ_FUA) {
2085                                 /*
2086                                  * Flush operations don't contain bios, so
2087                                  * we need to requeue the whole request
2088                                  *
2089                                  * XXX: but this doesn't make any sense for a
2090                                  * write with the FUA flag set..
2091                                  */
2092                                 list_add(&shadow[j].request->queuelist, &info->requests);
2093                                 continue;
2094                         }
2095                         merge_bio.head = shadow[j].request->bio;
2096                         merge_bio.tail = shadow[j].request->biotail;
2097                         bio_list_merge(&info->bio_list, &merge_bio);
2098                         shadow[j].request->bio = NULL;
2099                         blk_mq_end_request(shadow[j].request, BLK_STS_OK);
2100                 }
2101         }
2102
2103         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2104
2105         err = talk_to_blkback(dev, info);
2106         if (!err)
2107                 blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings);
2108
2109         /*
2110          * We have to wait for the backend to switch to
2111          * connected state, since we want to read which
2112          * features it supports.
2113          */
2114
2115         return err;
2116 }
2117
2118 static void blkfront_closing(struct blkfront_info *info)
2119 {
2120         struct xenbus_device *xbdev = info->xbdev;
2121         struct block_device *bdev = NULL;
2122
2123         mutex_lock(&info->mutex);
2124
2125         if (xbdev->state == XenbusStateClosing) {
2126                 mutex_unlock(&info->mutex);
2127                 return;
2128         }
2129
2130         if (info->gd)
2131                 bdev = bdget_disk(info->gd, 0);
2132
2133         mutex_unlock(&info->mutex);
2134
2135         if (!bdev) {
2136                 xenbus_frontend_closed(xbdev);
2137                 return;
2138         }
2139
2140         mutex_lock(&bdev->bd_mutex);
2141
2142         if (bdev->bd_openers) {
2143                 xenbus_dev_error(xbdev, -EBUSY,
2144                                  "Device in use; refusing to close");
2145                 xenbus_switch_state(xbdev, XenbusStateClosing);
2146         } else {
2147                 xlvbd_release_gendisk(info);
2148                 xenbus_frontend_closed(xbdev);
2149         }
2150
2151         mutex_unlock(&bdev->bd_mutex);
2152         bdput(bdev);
2153 }
2154
2155 static void blkfront_setup_discard(struct blkfront_info *info)
2156 {
2157         int err;
2158         unsigned int discard_granularity;
2159         unsigned int discard_alignment;
2160
2161         info->feature_discard = 1;
2162         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2163                 "discard-granularity", "%u", &discard_granularity,
2164                 "discard-alignment", "%u", &discard_alignment,
2165                 NULL);
2166         if (!err) {
2167                 info->discard_granularity = discard_granularity;
2168                 info->discard_alignment = discard_alignment;
2169         }
2170         info->feature_secdiscard =
2171                 !!xenbus_read_unsigned(info->xbdev->otherend, "discard-secure",
2172                                        0);
2173 }
2174
2175 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2176 {
2177         unsigned int psegs, grants;
2178         int err, i;
2179         struct blkfront_info *info = rinfo->dev_info;
2180
2181         if (info->max_indirect_segments == 0) {
2182                 if (!HAS_EXTRA_REQ)
2183                         grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2184                 else {
2185                         /*
2186                          * When an extra req is required, the maximum
2187                          * grants supported is related to the size of the
2188                          * Linux block segment.
2189                          */
2190                         grants = GRANTS_PER_PSEG;
2191                 }
2192         }
2193         else
2194                 grants = info->max_indirect_segments;
2195         psegs = DIV_ROUND_UP(grants, GRANTS_PER_PSEG);
2196
2197         err = fill_grant_buffer(rinfo,
2198                                 (grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2199         if (err)
2200                 goto out_of_memory;
2201
2202         if (!info->feature_persistent && info->max_indirect_segments) {
2203                 /*
2204                  * We are using indirect descriptors but not persistent
2205                  * grants, we need to allocate a set of pages that can be
2206                  * used for mapping indirect grefs
2207                  */
2208                 int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2209
2210                 BUG_ON(!list_empty(&rinfo->indirect_pages));
2211                 for (i = 0; i < num; i++) {
2212                         struct page *indirect_page = alloc_page(GFP_NOIO);
2213                         if (!indirect_page)
2214                                 goto out_of_memory;
2215                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
2216                 }
2217         }
2218
2219         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2220                 rinfo->shadow[i].grants_used = kzalloc(
2221                         sizeof(rinfo->shadow[i].grants_used[0]) * grants,
2222                         GFP_NOIO);
2223                 rinfo->shadow[i].sg = kzalloc(sizeof(rinfo->shadow[i].sg[0]) * psegs, GFP_NOIO);
2224                 if (info->max_indirect_segments)
2225                         rinfo->shadow[i].indirect_grants = kzalloc(
2226                                 sizeof(rinfo->shadow[i].indirect_grants[0]) *
2227                                 INDIRECT_GREFS(grants),
2228                                 GFP_NOIO);
2229                 if ((rinfo->shadow[i].grants_used == NULL) ||
2230                         (rinfo->shadow[i].sg == NULL) ||
2231                      (info->max_indirect_segments &&
2232                      (rinfo->shadow[i].indirect_grants == NULL)))
2233                         goto out_of_memory;
2234                 sg_init_table(rinfo->shadow[i].sg, psegs);
2235         }
2236
2237
2238         return 0;
2239
2240 out_of_memory:
2241         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2242                 kfree(rinfo->shadow[i].grants_used);
2243                 rinfo->shadow[i].grants_used = NULL;
2244                 kfree(rinfo->shadow[i].sg);
2245                 rinfo->shadow[i].sg = NULL;
2246                 kfree(rinfo->shadow[i].indirect_grants);
2247                 rinfo->shadow[i].indirect_grants = NULL;
2248         }
2249         if (!list_empty(&rinfo->indirect_pages)) {
2250                 struct page *indirect_page, *n;
2251                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2252                         list_del(&indirect_page->lru);
2253                         __free_page(indirect_page);
2254                 }
2255         }
2256         return -ENOMEM;
2257 }
2258
2259 /*
2260  * Gather all backend feature-*
2261  */
2262 static void blkfront_gather_backend_features(struct blkfront_info *info)
2263 {
2264         unsigned int indirect_segments;
2265
2266         info->feature_flush = 0;
2267         info->feature_fua = 0;
2268
2269         /*
2270          * If there's no "feature-barrier" defined, then it means
2271          * we're dealing with a very old backend which writes
2272          * synchronously; nothing to do.
2273          *
2274          * If there are barriers, then we use flush.
2275          */
2276         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-barrier", 0)) {
2277                 info->feature_flush = 1;
2278                 info->feature_fua = 1;
2279         }
2280
2281         /*
2282          * And if there is "feature-flush-cache" use that above
2283          * barriers.
2284          */
2285         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-flush-cache",
2286                                  0)) {
2287                 info->feature_flush = 1;
2288                 info->feature_fua = 0;
2289         }
2290
2291         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-discard", 0))
2292                 blkfront_setup_discard(info);
2293
2294         info->feature_persistent =
2295                 !!xenbus_read_unsigned(info->xbdev->otherend,
2296                                        "feature-persistent", 0);
2297
2298         indirect_segments = xenbus_read_unsigned(info->xbdev->otherend,
2299                                         "feature-max-indirect-segments", 0);
2300         if (indirect_segments > xen_blkif_max_segments)
2301                 indirect_segments = xen_blkif_max_segments;
2302         if (indirect_segments <= BLKIF_MAX_SEGMENTS_PER_REQUEST)
2303                 indirect_segments = 0;
2304         info->max_indirect_segments = indirect_segments;
2305 }
2306
2307 /*
2308  * Invoked when the backend is finally 'ready' (and has told produced
2309  * the details about the physical device - #sectors, size, etc).
2310  */
2311 static void blkfront_connect(struct blkfront_info *info)
2312 {
2313         unsigned long long sectors;
2314         unsigned long sector_size;
2315         unsigned int physical_sector_size;
2316         unsigned int binfo;
2317         char *envp[] = { "RESIZE=1", NULL };
2318         int err, i;
2319
2320         switch (info->connected) {
2321         case BLKIF_STATE_CONNECTED:
2322                 /*
2323                  * Potentially, the back-end may be signalling
2324                  * a capacity change; update the capacity.
2325                  */
2326                 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2327                                    "sectors", "%Lu", &sectors);
2328                 if (XENBUS_EXIST_ERR(err))
2329                         return;
2330                 printk(KERN_INFO "Setting capacity to %Lu\n",
2331                        sectors);
2332                 set_capacity(info->gd, sectors);
2333                 revalidate_disk(info->gd);
2334                 kobject_uevent_env(&disk_to_dev(info->gd)->kobj,
2335                                    KOBJ_CHANGE, envp);
2336
2337                 return;
2338         case BLKIF_STATE_SUSPENDED:
2339                 /*
2340                  * If we are recovering from suspension, we need to wait
2341                  * for the backend to announce it's features before
2342                  * reconnecting, at least we need to know if the backend
2343                  * supports indirect descriptors, and how many.
2344                  */
2345                 blkif_recover(info);
2346                 return;
2347
2348         default:
2349                 break;
2350         }
2351
2352         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2353                 __func__, info->xbdev->otherend);
2354
2355         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2356                             "sectors", "%llu", &sectors,
2357                             "info", "%u", &binfo,
2358                             "sector-size", "%lu", &sector_size,
2359                             NULL);
2360         if (err) {
2361                 xenbus_dev_fatal(info->xbdev, err,
2362                                  "reading backend fields at %s",
2363                                  info->xbdev->otherend);
2364                 return;
2365         }
2366
2367         /*
2368          * physcial-sector-size is a newer field, so old backends may not
2369          * provide this. Assume physical sector size to be the same as
2370          * sector_size in that case.
2371          */
2372         physical_sector_size = xenbus_read_unsigned(info->xbdev->otherend,
2373                                                     "physical-sector-size",
2374                                                     sector_size);
2375         blkfront_gather_backend_features(info);
2376         for (i = 0; i < info->nr_rings; i++) {
2377                 err = blkfront_setup_indirect(&info->rinfo[i]);
2378                 if (err) {
2379                         xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2380                                          info->xbdev->otherend);
2381                         blkif_free(info, 0);
2382                         break;
2383                 }
2384         }
2385
2386         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
2387                                   physical_sector_size);
2388         if (err) {
2389                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2390                                  info->xbdev->otherend);
2391                 goto fail;
2392         }
2393
2394         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2395
2396         /* Kick pending requests. */
2397         info->connected = BLKIF_STATE_CONNECTED;
2398         for (i = 0; i < info->nr_rings; i++)
2399                 kick_pending_request_queues(&info->rinfo[i]);
2400
2401         device_add_disk(&info->xbdev->dev, info->gd);
2402
2403         info->is_ready = 1;
2404         return;
2405
2406 fail:
2407         blkif_free(info, 0);
2408         return;
2409 }
2410
2411 /**
2412  * Callback received when the backend's state changes.
2413  */
2414 static void blkback_changed(struct xenbus_device *dev,
2415                             enum xenbus_state backend_state)
2416 {
2417         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2418
2419         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2420
2421         switch (backend_state) {
2422         case XenbusStateInitWait:
2423                 if (dev->state != XenbusStateInitialising)
2424                         break;
2425                 if (talk_to_blkback(dev, info))
2426                         break;
2427         case XenbusStateInitialising:
2428         case XenbusStateInitialised:
2429         case XenbusStateReconfiguring:
2430         case XenbusStateReconfigured:
2431         case XenbusStateUnknown:
2432                 break;
2433
2434         case XenbusStateConnected:
2435                 /*
2436                  * talk_to_blkback sets state to XenbusStateInitialised
2437                  * and blkfront_connect sets it to XenbusStateConnected
2438                  * (if connection went OK).
2439                  *
2440                  * If the backend (or toolstack) decides to poke at backend
2441                  * state (and re-trigger the watch by setting the state repeatedly
2442                  * to XenbusStateConnected (4)) we need to deal with this.
2443                  * This is allowed as this is used to communicate to the guest
2444                  * that the size of disk has changed!
2445                  */
2446                 if ((dev->state != XenbusStateInitialised) &&
2447                     (dev->state != XenbusStateConnected)) {
2448                         if (talk_to_blkback(dev, info))
2449                                 break;
2450                 }
2451
2452                 blkfront_connect(info);
2453                 break;
2454
2455         case XenbusStateClosed:
2456                 if (dev->state == XenbusStateClosed)
2457                         break;
2458                 /* fall through */
2459         case XenbusStateClosing:
2460                 if (info)
2461                         blkfront_closing(info);
2462                 break;
2463         }
2464 }
2465
2466 static int blkfront_remove(struct xenbus_device *xbdev)
2467 {
2468         struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2469         struct block_device *bdev = NULL;
2470         struct gendisk *disk;
2471
2472         dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2473
2474         blkif_free(info, 0);
2475
2476         mutex_lock(&info->mutex);
2477
2478         disk = info->gd;
2479         if (disk)
2480                 bdev = bdget_disk(disk, 0);
2481
2482         info->xbdev = NULL;
2483         mutex_unlock(&info->mutex);
2484
2485         if (!bdev) {
2486                 kfree(info);
2487                 return 0;
2488         }
2489
2490         /*
2491          * The xbdev was removed before we reached the Closed
2492          * state. See if it's safe to remove the disk. If the bdev
2493          * isn't closed yet, we let release take care of it.
2494          */
2495
2496         mutex_lock(&bdev->bd_mutex);
2497         info = disk->private_data;
2498
2499         dev_warn(disk_to_dev(disk),
2500                  "%s was hot-unplugged, %d stale handles\n",
2501                  xbdev->nodename, bdev->bd_openers);
2502
2503         if (info && !bdev->bd_openers) {
2504                 xlvbd_release_gendisk(info);
2505                 disk->private_data = NULL;
2506                 kfree(info);
2507         }
2508
2509         mutex_unlock(&bdev->bd_mutex);
2510         bdput(bdev);
2511
2512         return 0;
2513 }
2514
2515 static int blkfront_is_ready(struct xenbus_device *dev)
2516 {
2517         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2518
2519         return info->is_ready && info->xbdev;
2520 }
2521
2522 static int blkif_open(struct block_device *bdev, fmode_t mode)
2523 {
2524         struct gendisk *disk = bdev->bd_disk;
2525         struct blkfront_info *info;
2526         int err = 0;
2527
2528         mutex_lock(&blkfront_mutex);
2529
2530         info = disk->private_data;
2531         if (!info) {
2532                 /* xbdev gone */
2533                 err = -ERESTARTSYS;
2534                 goto out;
2535         }
2536
2537         mutex_lock(&info->mutex);
2538
2539         if (!info->gd)
2540                 /* xbdev is closed */
2541                 err = -ERESTARTSYS;
2542
2543         mutex_unlock(&info->mutex);
2544
2545 out:
2546         mutex_unlock(&blkfront_mutex);
2547         return err;
2548 }
2549
2550 static void blkif_release(struct gendisk *disk, fmode_t mode)
2551 {
2552         struct blkfront_info *info = disk->private_data;
2553         struct block_device *bdev;
2554         struct xenbus_device *xbdev;
2555
2556         mutex_lock(&blkfront_mutex);
2557
2558         bdev = bdget_disk(disk, 0);
2559
2560         if (!bdev) {
2561                 WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2562                 goto out_mutex;
2563         }
2564         if (bdev->bd_openers)
2565                 goto out;
2566
2567         /*
2568          * Check if we have been instructed to close. We will have
2569          * deferred this request, because the bdev was still open.
2570          */
2571
2572         mutex_lock(&info->mutex);
2573         xbdev = info->xbdev;
2574
2575         if (xbdev && xbdev->state == XenbusStateClosing) {
2576                 /* pending switch to state closed */
2577                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2578                 xlvbd_release_gendisk(info);
2579                 xenbus_frontend_closed(info->xbdev);
2580         }
2581
2582         mutex_unlock(&info->mutex);
2583
2584         if (!xbdev) {
2585                 /* sudden device removal */
2586                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2587                 xlvbd_release_gendisk(info);
2588                 disk->private_data = NULL;
2589                 kfree(info);
2590         }
2591
2592 out:
2593         bdput(bdev);
2594 out_mutex:
2595         mutex_unlock(&blkfront_mutex);
2596 }
2597
2598 static const struct block_device_operations xlvbd_block_fops =
2599 {
2600         .owner = THIS_MODULE,
2601         .open = blkif_open,
2602         .release = blkif_release,
2603         .getgeo = blkif_getgeo,
2604         .ioctl = blkif_ioctl,
2605 };
2606
2607
2608 static const struct xenbus_device_id blkfront_ids[] = {
2609         { "vbd" },
2610         { "" }
2611 };
2612
2613 static struct xenbus_driver blkfront_driver = {
2614         .ids  = blkfront_ids,
2615         .probe = blkfront_probe,
2616         .remove = blkfront_remove,
2617         .resume = blkfront_resume,
2618         .otherend_changed = blkback_changed,
2619         .is_ready = blkfront_is_ready,
2620 };
2621
2622 static int __init xlblk_init(void)
2623 {
2624         int ret;
2625         int nr_cpus = num_online_cpus();
2626
2627         if (!xen_domain())
2628                 return -ENODEV;
2629
2630         if (xen_blkif_max_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST)
2631                 xen_blkif_max_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2632
2633         if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2634                 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2635                         xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2636                 xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2637         }
2638
2639         if (xen_blkif_max_queues > nr_cpus) {
2640                 pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2641                         xen_blkif_max_queues, nr_cpus);
2642                 xen_blkif_max_queues = nr_cpus;
2643         }
2644
2645         if (!xen_has_pv_disk_devices())
2646                 return -ENODEV;
2647
2648         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2649                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
2650                        XENVBD_MAJOR, DEV_NAME);
2651                 return -ENODEV;
2652         }
2653
2654         ret = xenbus_register_frontend(&blkfront_driver);
2655         if (ret) {
2656                 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2657                 return ret;
2658         }
2659
2660         return 0;
2661 }
2662 module_init(xlblk_init);
2663
2664
2665 static void __exit xlblk_exit(void)
2666 {
2667         xenbus_unregister_driver(&blkfront_driver);
2668         unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2669         kfree(minors);
2670 }
2671 module_exit(xlblk_exit);
2672
2673 MODULE_DESCRIPTION("Xen virtual block device frontend");
2674 MODULE_LICENSE("GPL");
2675 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2676 MODULE_ALIAS("xen:vbd");
2677 MODULE_ALIAS("xenblk");