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