blob: 9908597c5209e19f118b01437ce0f362cab8f289 [file] [log] [blame]
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001/*
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>
Bob Liu907c3eb2015-07-13 17:55:24 +080040#include <linux/blk-mq.h>
Ian Campbell597592d2008-02-21 13:03:45 -080041#include <linux/hdreg.h>
Christian Limpach440a01a2008-06-17 10:47:08 +020042#include <linux/cdrom.h>
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -070043#include <linux/module.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090044#include <linux/slab.h>
Arnd Bergmann2a48fc02010-06-02 14:28:52 +020045#include <linux/mutex.h>
Jens Axboe9e973e62009-02-24 08:10:09 +010046#include <linux/scatterlist.h>
Akinobu Mita34ae2e42012-01-21 00:15:26 +090047#include <linux/bitmap.h>
Roger Pau Monne155b7ed2013-03-18 17:49:34 +010048#include <linux/list.h>
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -070049
Jeremy Fitzhardinge1ccbf532009-10-06 15:11:14 -070050#include <xen/xen.h>
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -070051#include <xen/xenbus.h>
52#include <xen/grant_table.h>
53#include <xen/events.h>
54#include <xen/page.h>
Stefano Stabellinic1c54132010-05-14 12:44:30 +010055#include <xen/platform_pci.h>
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -070056
57#include <xen/interface/grant_table.h>
58#include <xen/interface/io/blkif.h>
Markus Armbruster3e334232008-04-02 10:54:02 -070059#include <xen/interface/io/protocols.h>
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -070060
61#include <asm/xen/hypervisor.h>
62
Julien Grall6cc568332015-08-13 13:13:35 +010063/*
64 * The minimal size of segment supported by the block framework is PAGE_SIZE.
65 * When Linux is using a different page size than Xen, it may not be possible
66 * to put all the data in a single segment.
67 * This can happen when the backend doesn't support indirect descriptor and
68 * therefore the maximum amount of data that a request can carry is
69 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE = 44KB
70 *
71 * Note that we only support one extra request. So the Linux page size
72 * should be <= ( 2 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) =
73 * 88KB.
74 */
75#define HAS_EXTRA_REQ (BLKIF_MAX_SEGMENTS_PER_REQUEST < XEN_PFN_PER_PAGE)
76
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -070077enum blkif_state {
78 BLKIF_STATE_DISCONNECTED,
79 BLKIF_STATE_CONNECTED,
80 BLKIF_STATE_SUSPENDED,
81};
82
Roger Pau Monne0a8704a2012-10-24 18:58:45 +020083struct grant {
84 grant_ref_t gref;
Julien Gralla7a6df22015-06-30 11:58:51 +010085 struct page *page;
Roger Pau Monne155b7ed2013-03-18 17:49:34 +010086 struct list_head node;
Roger Pau Monne0a8704a2012-10-24 18:58:45 +020087};
88
Julien Grall6cc568332015-08-13 13:13:35 +010089enum blk_req_status {
90 REQ_WAITING,
91 REQ_DONE,
92 REQ_ERROR,
93 REQ_EOPNOTSUPP,
94};
95
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -070096struct blk_shadow {
97 struct blkif_request req;
Jeremy Fitzhardingea945b982010-11-01 17:03:14 -040098 struct request *request;
Roger Pau Monne402b27f2013-04-18 16:06:54 +020099 struct grant **grants_used;
100 struct grant **indirect_grants;
Roger Pau Monneb7649152013-05-02 10:58:50 +0200101 struct scatterlist *sg;
Julien Grallc004a6f2015-07-22 16:44:54 +0100102 unsigned int num_sg;
Julien Grall6cc568332015-08-13 13:13:35 +0100103 enum blk_req_status status;
104
105 #define NO_ASSOCIATED_ID ~0UL
106 /*
107 * Id of the sibling if we ever need 2 requests when handling a
108 * block I/O request
109 */
110 unsigned long associated_id;
Roger Pau Monne402b27f2013-04-18 16:06:54 +0200111};
112
113struct split_bio {
114 struct bio *bio;
115 atomic_t pending;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700116};
117
Arnd Bergmann2a48fc02010-06-02 14:28:52 +0200118static DEFINE_MUTEX(blkfront_mutex);
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -0700119static const struct block_device_operations xlvbd_block_fops;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700120
Roger Pau Monne402b27f2013-04-18 16:06:54 +0200121/*
122 * Maximum number of segments in indirect requests, the actual value used by
123 * the frontend driver is the minimum of this value and the value provided
124 * by the backend driver.
125 */
126
127static unsigned int xen_blkif_max_segments = 32;
Jan Beulich14e710f2016-02-10 04:21:15 -0700128module_param_named(max_indirect_segments, xen_blkif_max_segments, uint,
129 S_IRUGO);
130MODULE_PARM_DESC(max_indirect_segments,
131 "Maximum amount of segments in indirect requests (default is 32)");
Roger Pau Monne402b27f2013-04-18 16:06:54 +0200132
Bob Liu28d949b2015-11-14 11:12:14 +0800133static unsigned int xen_blkif_max_queues = 4;
134module_param_named(max_queues, xen_blkif_max_queues, uint, S_IRUGO);
135MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
136
Bob Liu86839c52015-06-03 13:40:03 +0800137/*
138 * Maximum order of pages to be used for the shared ring between front and
139 * backend, 4KB page granularity is used.
140 */
141static unsigned int xen_blkif_max_ring_order;
142module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, S_IRUGO);
143MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
144
Julien Grallc004a6f2015-07-22 16:44:54 +0100145#define BLK_RING_SIZE(info) \
146 __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
147
148#define BLK_MAX_RING_SIZE \
Julien Grall9cce2912015-10-13 17:50:11 +0100149 __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * XENBUS_MAX_RING_GRANTS)
Julien Grallc004a6f2015-07-22 16:44:54 +0100150
Bob Liu86839c52015-06-03 13:40:03 +0800151/*
Konrad Rzeszutek Wilk6f03a7f2015-11-16 15:14:41 -0500152 * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
153 * characters are enough. Define to 20 to keep consistent with backend.
Bob Liu86839c52015-06-03 13:40:03 +0800154 */
155#define RINGREF_NAME_LEN (20)
Bob Liu28d949b2015-11-14 11:12:14 +0800156/*
157 * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
158 */
159#define QUEUE_NAME_LEN (17)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700160
161/*
Bob Liu81f35162015-11-14 11:12:11 +0800162 * Per-ring info.
163 * Every blkfront device can associate with one or more blkfront_ring_info,
164 * depending on how many hardware queues/rings to be used.
165 */
166struct blkfront_ring_info {
Bob Liu11659562015-11-14 11:12:13 +0800167 /* Lock to protect data in every ring buffer. */
168 spinlock_t ring_lock;
Bob Liu81f35162015-11-14 11:12:11 +0800169 struct blkif_front_ring ring;
170 unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
171 unsigned int evtchn, irq;
172 struct work_struct work;
173 struct gnttab_free_callback callback;
174 struct blk_shadow shadow[BLK_MAX_RING_SIZE];
175 struct list_head indirect_pages;
Bob Liu73716df2015-11-16 16:51:39 -0500176 struct list_head grants;
177 unsigned int persistent_gnts_c;
Bob Liu81f35162015-11-14 11:12:11 +0800178 unsigned long shadow_free;
179 struct blkfront_info *dev_info;
180};
181
182/*
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700183 * We have one of these per vbd, whether ide, scsi or 'other'. They
184 * hang in private_data off the gendisk structure. We may end up
185 * putting all kinds of interesting stuff here :-)
186 */
187struct blkfront_info
188{
Daniel Stoddenb70f5fa2010-04-30 22:01:19 +0000189 struct mutex mutex;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700190 struct xenbus_device *xbdev;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700191 struct gendisk *gd;
Bob Liu172335a2016-07-01 17:43:39 -0400192 u16 sector_size;
193 unsigned int physical_sector_size;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700194 int vdevice;
195 blkif_vdev_t handle;
196 enum blkif_state connected;
Bob Liu3df0e502015-11-14 11:12:12 +0800197 /* Number of pages per ring buffer. */
Bob Liu86839c52015-06-03 13:40:03 +0800198 unsigned int nr_ring_pages;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700199 struct request_queue *rq;
Tejun Heo4913efe2010-09-03 11:56:16 +0200200 unsigned int feature_flush;
Mike Christiea4180902016-06-05 14:32:24 -0500201 unsigned int feature_fua;
Konrad Rzeszutek Wilk5ea42982011-10-12 16:23:30 -0400202 unsigned int feature_discard:1;
203 unsigned int feature_secdiscard:1;
Li Dongyanged30bf32011-09-01 18:39:09 +0800204 unsigned int discard_granularity;
205 unsigned int discard_alignment;
Roger Pau Monne0a8704a2012-10-24 18:58:45 +0200206 unsigned int feature_persistent:1;
Julien Grallc004a6f2015-07-22 16:44:54 +0100207 /* Number of 4KB segments handled */
Roger Pau Monne402b27f2013-04-18 16:06:54 +0200208 unsigned int max_indirect_segments;
Christian Limpach1d78d702008-04-02 10:54:04 -0700209 int is_ready;
Bob Liu907c3eb2015-07-13 17:55:24 +0800210 struct blk_mq_tag_set tag_set;
Bob Liu3df0e502015-11-14 11:12:12 +0800211 struct blkfront_ring_info *rinfo;
212 unsigned int nr_rings;
Bob Liu7b427a52016-06-27 16:33:24 +0800213 /* Save uncomplete reqs and bios for migration. */
214 struct list_head requests;
215 struct bio_list bio_list;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700216};
217
Jan Beulich0e345822010-08-07 18:28:55 +0200218static unsigned int nr_minors;
219static unsigned long *minors;
220static DEFINE_SPINLOCK(minor_lock);
221
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700222#define GRANT_INVALID_REF 0
223
224#define PARTS_PER_DISK 16
Chris Lalancette9246b5f2008-09-17 14:30:32 -0700225#define PARTS_PER_EXT_DISK 256
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700226
227#define BLKIF_MAJOR(dev) ((dev)>>8)
228#define BLKIF_MINOR(dev) ((dev) & 0xff)
229
Chris Lalancette9246b5f2008-09-17 14:30:32 -0700230#define EXT_SHIFT 28
231#define EXTENDED (1<<EXT_SHIFT)
232#define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
233#define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
Stefano Stabellinic80a4202010-12-02 17:55:00 +0000234#define EMULATED_HD_DISK_MINOR_OFFSET (0)
235#define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
Stefan Bader196cfe22011-07-14 15:30:22 +0200236#define EMULATED_SD_DISK_MINOR_OFFSET (0)
237#define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700238
Chris Lalancette9246b5f2008-09-17 14:30:32 -0700239#define DEV_NAME "xvd" /* name in /dev */
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700240
Julien Grallc004a6f2015-07-22 16:44:54 +0100241/*
242 * Grants are always the same size as a Xen page (i.e 4KB).
243 * A physical segment is always the same size as a Linux page.
244 * Number of grants per physical segment
245 */
246#define GRANTS_PER_PSEG (PAGE_SIZE / XEN_PAGE_SIZE)
247
248#define GRANTS_PER_INDIRECT_FRAME \
249 (XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
250
251#define PSEGS_PER_INDIRECT_FRAME \
252 (GRANTS_INDIRECT_FRAME / GRANTS_PSEGS)
253
254#define INDIRECT_GREFS(_grants) \
255 DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
256
257#define GREFS(_psegs) ((_psegs) * GRANTS_PER_PSEG)
Roger Pau Monne402b27f2013-04-18 16:06:54 +0200258
Bob Liu81f35162015-11-14 11:12:11 +0800259static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
Bob Liu3df0e502015-11-14 11:12:12 +0800260static void blkfront_gather_backend_features(struct blkfront_info *info);
Roger Pau Monne402b27f2013-04-18 16:06:54 +0200261
Bob Liu81f35162015-11-14 11:12:11 +0800262static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700263{
Bob Liu81f35162015-11-14 11:12:11 +0800264 unsigned long free = rinfo->shadow_free;
265
266 BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
267 rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
268 rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700269 return free;
270}
271
Bob Liu81f35162015-11-14 11:12:11 +0800272static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
Konrad Rzeszutek Wilk6f03a7f2015-11-16 15:14:41 -0500273 unsigned long id)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700274{
Bob Liu81f35162015-11-14 11:12:11 +0800275 if (rinfo->shadow[id].req.u.rw.id != id)
Konrad Rzeszutek Wilk6878c322012-05-25 17:34:51 -0400276 return -EINVAL;
Bob Liu81f35162015-11-14 11:12:11 +0800277 if (rinfo->shadow[id].request == NULL)
Konrad Rzeszutek Wilk6878c322012-05-25 17:34:51 -0400278 return -EINVAL;
Bob Liu81f35162015-11-14 11:12:11 +0800279 rinfo->shadow[id].req.u.rw.id = rinfo->shadow_free;
280 rinfo->shadow[id].request = NULL;
281 rinfo->shadow_free = id;
Konrad Rzeszutek Wilk6878c322012-05-25 17:34:51 -0400282 return 0;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700283}
284
Bob Liu81f35162015-11-14 11:12:11 +0800285static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
Roger Pau Monne9c1e0502013-03-18 17:49:35 +0100286{
Bob Liu81f35162015-11-14 11:12:11 +0800287 struct blkfront_info *info = rinfo->dev_info;
Roger Pau Monne9c1e0502013-03-18 17:49:35 +0100288 struct page *granted_page;
289 struct grant *gnt_list_entry, *n;
290 int i = 0;
291
Konrad Rzeszutek Wilk6f03a7f2015-11-16 15:14:41 -0500292 while (i < num) {
Roger Pau Monne9c1e0502013-03-18 17:49:35 +0100293 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
294 if (!gnt_list_entry)
295 goto out_of_memory;
296
Roger Pau Monnebfe11d62013-10-29 18:31:14 +0100297 if (info->feature_persistent) {
298 granted_page = alloc_page(GFP_NOIO);
299 if (!granted_page) {
300 kfree(gnt_list_entry);
301 goto out_of_memory;
302 }
Julien Gralla7a6df22015-06-30 11:58:51 +0100303 gnt_list_entry->page = granted_page;
Roger Pau Monne9c1e0502013-03-18 17:49:35 +0100304 }
305
Roger Pau Monne9c1e0502013-03-18 17:49:35 +0100306 gnt_list_entry->gref = GRANT_INVALID_REF;
Bob Liu73716df2015-11-16 16:51:39 -0500307 list_add(&gnt_list_entry->node, &rinfo->grants);
Roger Pau Monne9c1e0502013-03-18 17:49:35 +0100308 i++;
309 }
310
311 return 0;
312
313out_of_memory:
314 list_for_each_entry_safe(gnt_list_entry, n,
Bob Liu73716df2015-11-16 16:51:39 -0500315 &rinfo->grants, node) {
Roger Pau Monne9c1e0502013-03-18 17:49:35 +0100316 list_del(&gnt_list_entry->node);
Roger Pau Monnebfe11d62013-10-29 18:31:14 +0100317 if (info->feature_persistent)
Julien Gralla7a6df22015-06-30 11:58:51 +0100318 __free_page(gnt_list_entry->page);
Roger Pau Monne9c1e0502013-03-18 17:49:35 +0100319 kfree(gnt_list_entry);
320 i--;
321 }
322 BUG_ON(i != 0);
323 return -ENOMEM;
324}
325
Bob Liu73716df2015-11-16 16:51:39 -0500326static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
Roger Pau Monne9c1e0502013-03-18 17:49:35 +0100327{
328 struct grant *gnt_list_entry;
Roger Pau Monne9c1e0502013-03-18 17:49:35 +0100329
Bob Liu73716df2015-11-16 16:51:39 -0500330 BUG_ON(list_empty(&rinfo->grants));
331 gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
Julien Grall4f503fb2015-07-01 14:10:38 +0100332 node);
Roger Pau Monne9c1e0502013-03-18 17:49:35 +0100333 list_del(&gnt_list_entry->node);
334
Julien Grall4f503fb2015-07-01 14:10:38 +0100335 if (gnt_list_entry->gref != GRANT_INVALID_REF)
Bob Liu73716df2015-11-16 16:51:39 -0500336 rinfo->persistent_gnts_c--;
Julien Grall4f503fb2015-07-01 14:10:38 +0100337
338 return gnt_list_entry;
339}
340
341static inline void grant_foreign_access(const struct grant *gnt_list_entry,
342 const struct blkfront_info *info)
343{
344 gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
345 info->xbdev->otherend_id,
346 gnt_list_entry->page,
347 0);
348}
349
350static struct grant *get_grant(grant_ref_t *gref_head,
351 unsigned long gfn,
Bob Liu73716df2015-11-16 16:51:39 -0500352 struct blkfront_ring_info *rinfo)
Julien Grall4f503fb2015-07-01 14:10:38 +0100353{
Bob Liu73716df2015-11-16 16:51:39 -0500354 struct grant *gnt_list_entry = get_free_grant(rinfo);
355 struct blkfront_info *info = rinfo->dev_info;
Julien Grall4f503fb2015-07-01 14:10:38 +0100356
357 if (gnt_list_entry->gref != GRANT_INVALID_REF)
Roger Pau Monne9c1e0502013-03-18 17:49:35 +0100358 return gnt_list_entry;
Julien Grall4f503fb2015-07-01 14:10:38 +0100359
360 /* Assign a gref to this page */
361 gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
362 BUG_ON(gnt_list_entry->gref == -ENOSPC);
363 if (info->feature_persistent)
364 grant_foreign_access(gnt_list_entry, info);
365 else {
366 /* Grant access to the GFN passed by the caller */
367 gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
368 info->xbdev->otherend_id,
369 gfn, 0);
Roger Pau Monne9c1e0502013-03-18 17:49:35 +0100370 }
371
Julien Grall4f503fb2015-07-01 14:10:38 +0100372 return gnt_list_entry;
373}
374
375static struct grant *get_indirect_grant(grant_ref_t *gref_head,
Bob Liu73716df2015-11-16 16:51:39 -0500376 struct blkfront_ring_info *rinfo)
Julien Grall4f503fb2015-07-01 14:10:38 +0100377{
Bob Liu73716df2015-11-16 16:51:39 -0500378 struct grant *gnt_list_entry = get_free_grant(rinfo);
379 struct blkfront_info *info = rinfo->dev_info;
Julien Grall4f503fb2015-07-01 14:10:38 +0100380
381 if (gnt_list_entry->gref != GRANT_INVALID_REF)
382 return gnt_list_entry;
383
Roger Pau Monne9c1e0502013-03-18 17:49:35 +0100384 /* Assign a gref to this page */
385 gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
386 BUG_ON(gnt_list_entry->gref == -ENOSPC);
Roger Pau Monnebfe11d62013-10-29 18:31:14 +0100387 if (!info->feature_persistent) {
Julien Grall4f503fb2015-07-01 14:10:38 +0100388 struct page *indirect_page;
389
390 /* Fetch a pre-allocated page to use for indirect grefs */
Bob Liu73716df2015-11-16 16:51:39 -0500391 BUG_ON(list_empty(&rinfo->indirect_pages));
392 indirect_page = list_first_entry(&rinfo->indirect_pages,
Julien Grall4f503fb2015-07-01 14:10:38 +0100393 struct page, lru);
394 list_del(&indirect_page->lru);
395 gnt_list_entry->page = indirect_page;
Roger Pau Monnebfe11d62013-10-29 18:31:14 +0100396 }
Julien Grall4f503fb2015-07-01 14:10:38 +0100397 grant_foreign_access(gnt_list_entry, info);
398
Roger Pau Monne9c1e0502013-03-18 17:49:35 +0100399 return gnt_list_entry;
400}
401
Konrad Rzeszutek Wilk6878c322012-05-25 17:34:51 -0400402static const char *op_name(int op)
403{
404 static const char *const names[] = {
405 [BLKIF_OP_READ] = "read",
406 [BLKIF_OP_WRITE] = "write",
407 [BLKIF_OP_WRITE_BARRIER] = "barrier",
408 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
409 [BLKIF_OP_DISCARD] = "discard" };
410
411 if (op < 0 || op >= ARRAY_SIZE(names))
412 return "unknown";
413
414 if (!names[op])
415 return "reserved";
416
417 return names[op];
418}
Jan Beulich0e345822010-08-07 18:28:55 +0200419static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
420{
421 unsigned int end = minor + nr;
422 int rc;
423
424 if (end > nr_minors) {
425 unsigned long *bitmap, *old;
426
Thomas Meyerf0941482011-11-29 22:08:00 +0100427 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
Jan Beulich0e345822010-08-07 18:28:55 +0200428 GFP_KERNEL);
429 if (bitmap == NULL)
430 return -ENOMEM;
431
432 spin_lock(&minor_lock);
433 if (end > nr_minors) {
434 old = minors;
435 memcpy(bitmap, minors,
436 BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
437 minors = bitmap;
438 nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
439 } else
440 old = bitmap;
441 spin_unlock(&minor_lock);
442 kfree(old);
443 }
444
445 spin_lock(&minor_lock);
446 if (find_next_bit(minors, end, minor) >= end) {
Akinobu Mita34ae2e42012-01-21 00:15:26 +0900447 bitmap_set(minors, minor, nr);
Jan Beulich0e345822010-08-07 18:28:55 +0200448 rc = 0;
449 } else
450 rc = -EBUSY;
451 spin_unlock(&minor_lock);
452
453 return rc;
454}
455
456static void xlbd_release_minors(unsigned int minor, unsigned int nr)
457{
458 unsigned int end = minor + nr;
459
460 BUG_ON(end > nr_minors);
461 spin_lock(&minor_lock);
Akinobu Mita34ae2e42012-01-21 00:15:26 +0900462 bitmap_clear(minors, minor, nr);
Jan Beulich0e345822010-08-07 18:28:55 +0200463 spin_unlock(&minor_lock);
464}
465
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700466static void blkif_restart_queue_callback(void *arg)
467{
Bob Liu81f35162015-11-14 11:12:11 +0800468 struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
469 schedule_work(&rinfo->work);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700470}
471
Harvey Harrisonafe42d72008-04-29 00:59:47 -0700472static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
Ian Campbell597592d2008-02-21 13:03:45 -0800473{
474 /* We don't have real geometry info, but let's at least return
475 values consistent with the size of the device */
476 sector_t nsect = get_capacity(bd->bd_disk);
477 sector_t cylinders = nsect;
478
479 hg->heads = 0xff;
480 hg->sectors = 0x3f;
481 sector_div(cylinders, hg->heads * hg->sectors);
482 hg->cylinders = cylinders;
483 if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
484 hg->cylinders = 0xffff;
485 return 0;
486}
487
Al Viroa63c8482008-03-02 10:23:47 -0500488static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
Adrian Bunk62aa0052008-08-04 11:59:05 +0200489 unsigned command, unsigned long argument)
Christian Limpach440a01a2008-06-17 10:47:08 +0200490{
Al Viroa63c8482008-03-02 10:23:47 -0500491 struct blkfront_info *info = bdev->bd_disk->private_data;
Christian Limpach440a01a2008-06-17 10:47:08 +0200492 int i;
493
494 dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
495 command, (long)argument);
496
497 switch (command) {
498 case CDROMMULTISESSION:
499 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
500 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
501 if (put_user(0, (char __user *)(argument + i)))
502 return -EFAULT;
503 return 0;
504
505 case CDROM_GET_CAPABILITY: {
506 struct gendisk *gd = info->gd;
507 if (gd->flags & GENHD_FL_CD)
508 return 0;
509 return -EINVAL;
510 }
511
512 default:
513 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
514 command);*/
515 return -EINVAL; /* same return as native Linux */
516 }
517
518 return 0;
519}
520
Julien Grall2e073962015-08-13 19:23:10 +0100521static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
522 struct request *req,
523 struct blkif_request **ring_req)
524{
525 unsigned long id;
526
527 *ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
528 rinfo->ring.req_prod_pvt++;
529
530 id = get_id_from_freelist(rinfo);
531 rinfo->shadow[id].request = req;
Julien Grall6cc568332015-08-13 13:13:35 +0100532 rinfo->shadow[id].status = REQ_WAITING;
533 rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
Julien Grall2e073962015-08-13 19:23:10 +0100534
535 (*ring_req)->u.rw.id = id;
536
537 return id;
538}
539
Bob Liu81f35162015-11-14 11:12:11 +0800540static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700541{
Bob Liu81f35162015-11-14 11:12:11 +0800542 struct blkfront_info *info = rinfo->dev_info;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700543 struct blkif_request *ring_req;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700544 unsigned long id;
Julien Grall33204662015-06-29 17:35:24 +0100545
546 /* Fill out a communications ring structure. */
Julien Grall2e073962015-08-13 19:23:10 +0100547 id = blkif_ring_get_request(rinfo, req, &ring_req);
Julien Grall33204662015-06-29 17:35:24 +0100548
549 ring_req->operation = BLKIF_OP_DISCARD;
550 ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
551 ring_req->u.discard.id = id;
552 ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
Christoph Hellwig288dab82016-06-09 16:00:36 +0200553 if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard)
Julien Grall33204662015-06-29 17:35:24 +0100554 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
555 else
556 ring_req->u.discard.flag = 0;
557
Julien Grall33204662015-06-29 17:35:24 +0100558 /* Keep a private copy so we can reissue requests when recovering. */
Bob Liu81f35162015-11-14 11:12:11 +0800559 rinfo->shadow[id].req = *ring_req;
Julien Grall33204662015-06-29 17:35:24 +0100560
561 return 0;
562}
563
Julien Grallc004a6f2015-07-22 16:44:54 +0100564struct setup_rw_req {
565 unsigned int grant_idx;
566 struct blkif_request_segment *segments;
Bob Liu81f35162015-11-14 11:12:11 +0800567 struct blkfront_ring_info *rinfo;
Julien Grallc004a6f2015-07-22 16:44:54 +0100568 struct blkif_request *ring_req;
569 grant_ref_t gref_head;
570 unsigned int id;
571 /* Only used when persistent grant is used and it's a read request */
572 bool need_copy;
573 unsigned int bvec_off;
574 char *bvec_data;
Julien Grall6cc568332015-08-13 13:13:35 +0100575
576 bool require_extra_req;
577 struct blkif_request *extra_ring_req;
Julien Grallc004a6f2015-07-22 16:44:54 +0100578};
579
580static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
581 unsigned int len, void *data)
582{
583 struct setup_rw_req *setup = data;
584 int n, ref;
585 struct grant *gnt_list_entry;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700586 unsigned int fsect, lsect;
Julien Grallc004a6f2015-07-22 16:44:54 +0100587 /* Convenient aliases */
588 unsigned int grant_idx = setup->grant_idx;
589 struct blkif_request *ring_req = setup->ring_req;
Bob Liu81f35162015-11-14 11:12:11 +0800590 struct blkfront_ring_info *rinfo = setup->rinfo;
Julien Grall6cc568332015-08-13 13:13:35 +0100591 /*
592 * We always use the shadow of the first request to store the list
593 * of grant associated to the block I/O request. This made the
594 * completion more easy to handle even if the block I/O request is
595 * split.
596 */
Bob Liu81f35162015-11-14 11:12:11 +0800597 struct blk_shadow *shadow = &rinfo->shadow[setup->id];
Julien Grallc004a6f2015-07-22 16:44:54 +0100598
Julien Grall6cc568332015-08-13 13:13:35 +0100599 if (unlikely(setup->require_extra_req &&
600 grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
601 /*
602 * We are using the second request, setup grant_idx
603 * to be the index of the segment array.
604 */
605 grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
606 ring_req = setup->extra_ring_req;
607 }
608
Julien Grallc004a6f2015-07-22 16:44:54 +0100609 if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
610 (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
611 if (setup->segments)
612 kunmap_atomic(setup->segments);
613
614 n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
Bob Liu73716df2015-11-16 16:51:39 -0500615 gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
Julien Grallc004a6f2015-07-22 16:44:54 +0100616 shadow->indirect_grants[n] = gnt_list_entry;
617 setup->segments = kmap_atomic(gnt_list_entry->page);
618 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
619 }
620
Bob Liu73716df2015-11-16 16:51:39 -0500621 gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
Julien Grallc004a6f2015-07-22 16:44:54 +0100622 ref = gnt_list_entry->gref;
Julien Grall6cc568332015-08-13 13:13:35 +0100623 /*
624 * All the grants are stored in the shadow of the first
625 * request. Therefore we have to use the global index.
626 */
627 shadow->grants_used[setup->grant_idx] = gnt_list_entry;
Julien Grallc004a6f2015-07-22 16:44:54 +0100628
629 if (setup->need_copy) {
630 void *shared_data;
631
632 shared_data = kmap_atomic(gnt_list_entry->page);
633 /*
634 * this does not wipe data stored outside the
635 * range sg->offset..sg->offset+sg->length.
636 * Therefore, blkback *could* see data from
637 * previous requests. This is OK as long as
638 * persistent grants are shared with just one
639 * domain. It may need refactoring if this
640 * changes
641 */
642 memcpy(shared_data + offset,
643 setup->bvec_data + setup->bvec_off,
644 len);
645
646 kunmap_atomic(shared_data);
647 setup->bvec_off += len;
648 }
649
650 fsect = offset >> 9;
651 lsect = fsect + (len >> 9) - 1;
652 if (ring_req->operation != BLKIF_OP_INDIRECT) {
653 ring_req->u.rw.seg[grant_idx] =
654 (struct blkif_request_segment) {
655 .gref = ref,
656 .first_sect = fsect,
657 .last_sect = lsect };
658 } else {
659 setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
660 (struct blkif_request_segment) {
661 .gref = ref,
662 .first_sect = fsect,
663 .last_sect = lsect };
664 }
665
666 (setup->grant_idx)++;
667}
668
Julien Grall6cc568332015-08-13 13:13:35 +0100669static void blkif_setup_extra_req(struct blkif_request *first,
670 struct blkif_request *second)
671{
672 uint16_t nr_segments = first->u.rw.nr_segments;
673
674 /*
675 * The second request is only present when the first request uses
676 * all its segments. It's always the continuity of the first one.
677 */
678 first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
679
680 second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
681 second->u.rw.sector_number = first->u.rw.sector_number +
682 (BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
683
684 second->u.rw.handle = first->u.rw.handle;
685 second->operation = first->operation;
686}
687
Bob Liu81f35162015-11-14 11:12:11 +0800688static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700689{
Bob Liu81f35162015-11-14 11:12:11 +0800690 struct blkfront_info *info = rinfo->dev_info;
Julien Grall6cc568332015-08-13 13:13:35 +0100691 struct blkif_request *ring_req, *extra_ring_req = NULL;
692 unsigned long id, extra_id = NO_ASSOCIATED_ID;
693 bool require_extra_req = false;
Julien Grallc004a6f2015-07-22 16:44:54 +0100694 int i;
695 struct setup_rw_req setup = {
696 .grant_idx = 0,
697 .segments = NULL,
Bob Liu81f35162015-11-14 11:12:11 +0800698 .rinfo = rinfo,
Julien Grallc004a6f2015-07-22 16:44:54 +0100699 .need_copy = rq_data_dir(req) && info->feature_persistent,
700 };
Roger Pau Monne0a8704a2012-10-24 18:58:45 +0200701
702 /*
703 * Used to store if we are able to queue the request by just using
704 * existing persistent grants, or if we have to get new grants,
705 * as there are not sufficiently many free.
706 */
Jens Axboe9e973e62009-02-24 08:10:09 +0100707 struct scatterlist *sg;
Julien Grallc004a6f2015-07-22 16:44:54 +0100708 int num_sg, max_grefs, num_grant;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700709
Julien Grallc004a6f2015-07-22 16:44:54 +0100710 max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
Roger Pau Monnec47206e2013-08-12 12:53:43 +0200711 if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
712 /*
713 * If we are using indirect segments we need to account
714 * for the indirect grefs used in the request.
715 */
Julien Grallc004a6f2015-07-22 16:44:54 +0100716 max_grefs += INDIRECT_GREFS(max_grefs);
Roger Pau Monne402b27f2013-04-18 16:06:54 +0200717
Bob Liu3df0e502015-11-14 11:12:12 +0800718 /*
719 * We have to reserve 'max_grefs' grants because persistent
720 * grants are shared by all rings.
721 */
722 if (max_grefs > 0)
723 if (gnttab_alloc_grant_references(max_grefs, &setup.gref_head) < 0) {
Roger Pau Monne0a8704a2012-10-24 18:58:45 +0200724 gnttab_request_free_callback(
Bob Liu81f35162015-11-14 11:12:11 +0800725 &rinfo->callback,
Roger Pau Monne0a8704a2012-10-24 18:58:45 +0200726 blkif_restart_queue_callback,
Bob Liu81f35162015-11-14 11:12:11 +0800727 rinfo,
Roger Pau Monne402b27f2013-04-18 16:06:54 +0200728 max_grefs);
Roger Pau Monne0a8704a2012-10-24 18:58:45 +0200729 return 1;
730 }
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700731
732 /* Fill out a communications ring structure. */
Julien Grall2e073962015-08-13 19:23:10 +0100733 id = blkif_ring_get_request(rinfo, req, &ring_req);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700734
Bob Liu81f35162015-11-14 11:12:11 +0800735 num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
Julien Grallc004a6f2015-07-22 16:44:54 +0100736 num_grant = 0;
737 /* Calculate the number of grant used */
Bob Liu81f35162015-11-14 11:12:11 +0800738 for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
Julien Grallc004a6f2015-07-22 16:44:54 +0100739 num_grant += gnttab_count_grant(sg->offset, sg->length);
740
Julien Grall6cc568332015-08-13 13:13:35 +0100741 require_extra_req = info->max_indirect_segments == 0 &&
742 num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
743 BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
744
Bob Liu81f35162015-11-14 11:12:11 +0800745 rinfo->shadow[id].num_sg = num_sg;
Julien Grall6cc568332015-08-13 13:13:35 +0100746 if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
747 likely(!require_extra_req)) {
Julien Grall33204662015-06-29 17:35:24 +0100748 /*
749 * The indirect operation can only be a BLKIF_OP_READ or
750 * BLKIF_OP_WRITE
751 */
Mike Christie3a5e02c2016-06-05 14:32:23 -0500752 BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA);
Julien Grall33204662015-06-29 17:35:24 +0100753 ring_req->operation = BLKIF_OP_INDIRECT;
754 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
755 BLKIF_OP_WRITE : BLKIF_OP_READ;
756 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
757 ring_req->u.indirect.handle = info->handle;
Julien Grallc004a6f2015-07-22 16:44:54 +0100758 ring_req->u.indirect.nr_segments = num_grant;
Li Dongyanged30bf32011-09-01 18:39:09 +0800759 } else {
Julien Grall33204662015-06-29 17:35:24 +0100760 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
761 ring_req->u.rw.handle = info->handle;
762 ring_req->operation = rq_data_dir(req) ?
763 BLKIF_OP_WRITE : BLKIF_OP_READ;
Mike Christie3a5e02c2016-06-05 14:32:23 -0500764 if (req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA) {
Roger Pau Monne402b27f2013-04-18 16:06:54 +0200765 /*
Julien Grall33204662015-06-29 17:35:24 +0100766 * Ideally we can do an unordered flush-to-disk.
767 * In case the backend onlysupports barriers, use that.
768 * A barrier request a superset of FUA, so we can
769 * implement it the same way. (It's also a FLUSH+FUA,
770 * since it is guaranteed ordered WRT previous writes.)
Roger Pau Monne402b27f2013-04-18 16:06:54 +0200771 */
Mike Christiea4180902016-06-05 14:32:24 -0500772 if (info->feature_flush && info->feature_fua)
Julien Grall33204662015-06-29 17:35:24 +0100773 ring_req->operation =
774 BLKIF_OP_WRITE_BARRIER;
Mike Christiea4180902016-06-05 14:32:24 -0500775 else if (info->feature_flush)
Julien Grall33204662015-06-29 17:35:24 +0100776 ring_req->operation =
777 BLKIF_OP_FLUSH_DISKCACHE;
Mike Christiea4180902016-06-05 14:32:24 -0500778 else
Julien Grall33204662015-06-29 17:35:24 +0100779 ring_req->operation = 0;
Li Dongyanged30bf32011-09-01 18:39:09 +0800780 }
Julien Grallc004a6f2015-07-22 16:44:54 +0100781 ring_req->u.rw.nr_segments = num_grant;
Julien Grall6cc568332015-08-13 13:13:35 +0100782 if (unlikely(require_extra_req)) {
783 extra_id = blkif_ring_get_request(rinfo, req,
784 &extra_ring_req);
785 /*
786 * Only the first request contains the scatter-gather
787 * list.
788 */
789 rinfo->shadow[extra_id].num_sg = 0;
790
791 blkif_setup_extra_req(ring_req, extra_ring_req);
792
793 /* Link the 2 requests together */
794 rinfo->shadow[extra_id].associated_id = id;
795 rinfo->shadow[id].associated_id = extra_id;
796 }
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700797 }
798
Julien Grallc004a6f2015-07-22 16:44:54 +0100799 setup.ring_req = ring_req;
800 setup.id = id;
Julien Grall6cc568332015-08-13 13:13:35 +0100801
802 setup.require_extra_req = require_extra_req;
803 if (unlikely(require_extra_req))
804 setup.extra_ring_req = extra_ring_req;
805
Bob Liu81f35162015-11-14 11:12:11 +0800806 for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
Julien Grallc004a6f2015-07-22 16:44:54 +0100807 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
Julien Grall33204662015-06-29 17:35:24 +0100808
Julien Grallc004a6f2015-07-22 16:44:54 +0100809 if (setup.need_copy) {
810 setup.bvec_off = sg->offset;
811 setup.bvec_data = kmap_atomic(sg_page(sg));
Julien Grall33204662015-06-29 17:35:24 +0100812 }
813
Julien Grallc004a6f2015-07-22 16:44:54 +0100814 gnttab_foreach_grant_in_range(sg_page(sg),
815 sg->offset,
816 sg->length,
817 blkif_setup_rw_req_grant,
818 &setup);
Julien Grall33204662015-06-29 17:35:24 +0100819
Julien Grallc004a6f2015-07-22 16:44:54 +0100820 if (setup.need_copy)
821 kunmap_atomic(setup.bvec_data);
Julien Grall33204662015-06-29 17:35:24 +0100822 }
Julien Grallc004a6f2015-07-22 16:44:54 +0100823 if (setup.segments)
824 kunmap_atomic(setup.segments);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700825
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700826 /* Keep a private copy so we can reissue requests when recovering. */
Bob Liu81f35162015-11-14 11:12:11 +0800827 rinfo->shadow[id].req = *ring_req;
Julien Grall6cc568332015-08-13 13:13:35 +0100828 if (unlikely(require_extra_req))
829 rinfo->shadow[extra_id].req = *extra_ring_req;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700830
Bob Liu3df0e502015-11-14 11:12:12 +0800831 if (max_grefs > 0)
Julien Grallc004a6f2015-07-22 16:44:54 +0100832 gnttab_free_grant_references(setup.gref_head);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700833
834 return 0;
835}
836
Julien Grall33204662015-06-29 17:35:24 +0100837/*
838 * Generate a Xen blkfront IO request from a blk layer request. Reads
839 * and writes are handled as expected.
840 *
841 * @req: a request struct
842 */
Bob Liu81f35162015-11-14 11:12:11 +0800843static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
Julien Grall33204662015-06-29 17:35:24 +0100844{
Bob Liu81f35162015-11-14 11:12:11 +0800845 if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
Julien Grall33204662015-06-29 17:35:24 +0100846 return 1;
847
Mike Christiec2df40d2016-06-05 14:32:17 -0500848 if (unlikely(req_op(req) == REQ_OP_DISCARD ||
Christoph Hellwig288dab82016-06-09 16:00:36 +0200849 req_op(req) == REQ_OP_SECURE_ERASE))
Bob Liu81f35162015-11-14 11:12:11 +0800850 return blkif_queue_discard_req(req, rinfo);
Julien Grall33204662015-06-29 17:35:24 +0100851 else
Bob Liu81f35162015-11-14 11:12:11 +0800852 return blkif_queue_rw_req(req, rinfo);
Julien Grall33204662015-06-29 17:35:24 +0100853}
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700854
Bob Liu81f35162015-11-14 11:12:11 +0800855static inline void flush_requests(struct blkfront_ring_info *rinfo)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700856{
857 int notify;
858
Bob Liu81f35162015-11-14 11:12:11 +0800859 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700860
861 if (notify)
Bob Liu81f35162015-11-14 11:12:11 +0800862 notify_remote_via_irq(rinfo->irq);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700863}
864
Vitaly Kuznetsovad42d392014-12-01 14:01:13 +0100865static inline bool blkif_request_flush_invalid(struct request *req,
866 struct blkfront_info *info)
Arianna Avanzini0f1ca652014-08-22 13:20:02 +0200867{
868 return ((req->cmd_type != REQ_TYPE_FS) ||
Mike Christie3a5e02c2016-06-05 14:32:23 -0500869 ((req_op(req) == REQ_OP_FLUSH) &&
Mike Christiea4180902016-06-05 14:32:24 -0500870 !info->feature_flush) ||
Vitaly Kuznetsovad42d392014-12-01 14:01:13 +0100871 ((req->cmd_flags & REQ_FUA) &&
Mike Christiea4180902016-06-05 14:32:24 -0500872 !info->feature_fua));
Arianna Avanzini0f1ca652014-08-22 13:20:02 +0200873}
874
Bob Liu907c3eb2015-07-13 17:55:24 +0800875static int blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
Konrad Rzeszutek Wilk6f03a7f2015-11-16 15:14:41 -0500876 const struct blk_mq_queue_data *qd)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700877{
Bob Liu11659562015-11-14 11:12:13 +0800878 unsigned long flags;
Bob Liu2a6f71a2016-05-31 16:59:17 +0800879 int qid = hctx->queue_num;
880 struct blkfront_info *info = hctx->queue->queuedata;
881 struct blkfront_ring_info *rinfo = NULL;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700882
Bob Liu2a6f71a2016-05-31 16:59:17 +0800883 BUG_ON(info->nr_rings <= qid);
884 rinfo = &info->rinfo[qid];
Bob Liu907c3eb2015-07-13 17:55:24 +0800885 blk_mq_start_request(qd->rq);
Bob Liu11659562015-11-14 11:12:13 +0800886 spin_lock_irqsave(&rinfo->ring_lock, flags);
Bob Liu81f35162015-11-14 11:12:11 +0800887 if (RING_FULL(&rinfo->ring))
Bob Liu907c3eb2015-07-13 17:55:24 +0800888 goto out_busy;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700889
Bob Liu81f35162015-11-14 11:12:11 +0800890 if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
Bob Liu907c3eb2015-07-13 17:55:24 +0800891 goto out_err;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700892
Bob Liu81f35162015-11-14 11:12:11 +0800893 if (blkif_queue_request(qd->rq, rinfo))
Bob Liu907c3eb2015-07-13 17:55:24 +0800894 goto out_busy;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700895
Bob Liu81f35162015-11-14 11:12:11 +0800896 flush_requests(rinfo);
Bob Liu11659562015-11-14 11:12:13 +0800897 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
Bob Liu907c3eb2015-07-13 17:55:24 +0800898 return BLK_MQ_RQ_QUEUE_OK;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700899
Bob Liu907c3eb2015-07-13 17:55:24 +0800900out_err:
Bob Liu11659562015-11-14 11:12:13 +0800901 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
Bob Liu907c3eb2015-07-13 17:55:24 +0800902 return BLK_MQ_RQ_QUEUE_ERROR;
Tejun Heo296b2f62009-05-08 11:54:15 +0900903
Bob Liu907c3eb2015-07-13 17:55:24 +0800904out_busy:
Bob Liu11659562015-11-14 11:12:13 +0800905 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
Bob Liu907c3eb2015-07-13 17:55:24 +0800906 blk_mq_stop_hw_queue(hctx);
907 return BLK_MQ_RQ_QUEUE_BUSY;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700908}
909
Bob Liu907c3eb2015-07-13 17:55:24 +0800910static struct blk_mq_ops blkfront_mq_ops = {
911 .queue_rq = blkif_queue_rq,
Bob Liu907c3eb2015-07-13 17:55:24 +0800912};
913
Bob Liu172335a2016-07-01 17:43:39 -0400914static void blkif_set_queue_limits(struct blkfront_info *info)
915{
916 struct request_queue *rq = info->rq;
917 struct gendisk *gd = info->gd;
918 unsigned int segments = info->max_indirect_segments ? :
919 BLKIF_MAX_SEGMENTS_PER_REQUEST;
920
921 queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
922
923 if (info->feature_discard) {
924 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
925 blk_queue_max_discard_sectors(rq, get_capacity(gd));
926 rq->limits.discard_granularity = info->discard_granularity;
927 rq->limits.discard_alignment = info->discard_alignment;
928 if (info->feature_secdiscard)
929 queue_flag_set_unlocked(QUEUE_FLAG_SECERASE, rq);
930 }
931
932 /* Hard sector size and max sectors impersonate the equiv. hardware. */
933 blk_queue_logical_block_size(rq, info->sector_size);
934 blk_queue_physical_block_size(rq, info->physical_sector_size);
935 blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
936
937 /* Each segment in a request is up to an aligned page in size. */
938 blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
939 blk_queue_max_segment_size(rq, PAGE_SIZE);
940
941 /* Ensure a merged request will fit in a single I/O ring slot. */
942 blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
943
944 /* Make sure buffer addresses are sector-aligned. */
945 blk_queue_dma_alignment(rq, 511);
946
947 /* Make sure we don't use bounce buffers. */
948 blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
949}
950
Roger Pau Monne402b27f2013-04-18 16:06:54 +0200951static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
Bob Liu172335a2016-07-01 17:43:39 -0400952 unsigned int physical_sector_size)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700953{
Jens Axboe165125e2007-07-24 09:28:11 +0200954 struct request_queue *rq;
Li Dongyanged30bf32011-09-01 18:39:09 +0800955 struct blkfront_info *info = gd->private_data;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700956
Bob Liu907c3eb2015-07-13 17:55:24 +0800957 memset(&info->tag_set, 0, sizeof(info->tag_set));
958 info->tag_set.ops = &blkfront_mq_ops;
Bob Liu28d949b2015-11-14 11:12:14 +0800959 info->tag_set.nr_hw_queues = info->nr_rings;
Julien Grall6cc568332015-08-13 13:13:35 +0100960 if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
961 /*
962 * When indirect descriptior is not supported, the I/O request
963 * will be split between multiple request in the ring.
964 * To avoid problems when sending the request, divide by
965 * 2 the depth of the queue.
966 */
967 info->tag_set.queue_depth = BLK_RING_SIZE(info) / 2;
968 } else
969 info->tag_set.queue_depth = BLK_RING_SIZE(info);
Bob Liu907c3eb2015-07-13 17:55:24 +0800970 info->tag_set.numa_node = NUMA_NO_NODE;
971 info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
972 info->tag_set.cmd_size = 0;
973 info->tag_set.driver_data = info;
974
975 if (blk_mq_alloc_tag_set(&info->tag_set))
Konrad Rzeszutek Wilkbde21f72015-11-25 13:07:39 -0500976 return -EINVAL;
Bob Liu907c3eb2015-07-13 17:55:24 +0800977 rq = blk_mq_init_queue(&info->tag_set);
978 if (IS_ERR(rq)) {
979 blk_mq_free_tag_set(&info->tag_set);
Konrad Rzeszutek Wilkbde21f72015-11-25 13:07:39 -0500980 return PTR_ERR(rq);
Bob Liu907c3eb2015-07-13 17:55:24 +0800981 }
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700982
Bob Liu2a6f71a2016-05-31 16:59:17 +0800983 rq->queuedata = info;
Bob Liu172335a2016-07-01 17:43:39 -0400984 info->rq = gd->queue = rq;
985 info->gd = gd;
986 info->sector_size = sector_size;
987 info->physical_sector_size = physical_sector_size;
988 blkif_set_queue_limits(info);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700989
990 return 0;
991}
992
Mike Christiea4180902016-06-05 14:32:24 -0500993static const char *flush_info(struct blkfront_info *info)
Vitaly Kuznetsovfdf9b962014-12-09 15:25:10 +0100994{
Mike Christiea4180902016-06-05 14:32:24 -0500995 if (info->feature_flush && info->feature_fua)
Vitaly Kuznetsovfdf9b962014-12-09 15:25:10 +0100996 return "barrier: enabled;";
Mike Christiea4180902016-06-05 14:32:24 -0500997 else if (info->feature_flush)
Vitaly Kuznetsovfdf9b962014-12-09 15:25:10 +0100998 return "flush diskcache: enabled;";
Mike Christiea4180902016-06-05 14:32:24 -0500999 else
Vitaly Kuznetsovfdf9b962014-12-09 15:25:10 +01001000 return "barrier or flush: disabled;";
Vitaly Kuznetsovfdf9b962014-12-09 15:25:10 +01001001}
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001002
Tejun Heo4913efe2010-09-03 11:56:16 +02001003static void xlvbd_flush(struct blkfront_info *info)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001004{
Mike Christiea4180902016-06-05 14:32:24 -05001005 blk_queue_write_cache(info->rq, info->feature_flush ? true : false,
1006 info->feature_fua ? true : false);
Vitaly Kuznetsovfdf9b962014-12-09 15:25:10 +01001007 pr_info("blkfront: %s: %s %s %s %s %s\n",
Mike Christiea4180902016-06-05 14:32:24 -05001008 info->gd->disk_name, flush_info(info),
Vitaly Kuznetsovfdf9b962014-12-09 15:25:10 +01001009 "persistent grants:", info->feature_persistent ?
1010 "enabled;" : "disabled;", "indirect descriptors:",
1011 info->max_indirect_segments ? "enabled;" : "disabled;");
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001012}
1013
Stefano Stabellinic80a4202010-12-02 17:55:00 +00001014static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
1015{
1016 int major;
1017 major = BLKIF_MAJOR(vdevice);
1018 *minor = BLKIF_MINOR(vdevice);
1019 switch (major) {
1020 case XEN_IDE0_MAJOR:
1021 *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
1022 *minor = ((*minor / 64) * PARTS_PER_DISK) +
1023 EMULATED_HD_DISK_MINOR_OFFSET;
1024 break;
1025 case XEN_IDE1_MAJOR:
1026 *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1027 *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1028 EMULATED_HD_DISK_MINOR_OFFSET;
1029 break;
1030 case XEN_SCSI_DISK0_MAJOR:
1031 *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1032 *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1033 break;
1034 case XEN_SCSI_DISK1_MAJOR:
1035 case XEN_SCSI_DISK2_MAJOR:
1036 case XEN_SCSI_DISK3_MAJOR:
1037 case XEN_SCSI_DISK4_MAJOR:
1038 case XEN_SCSI_DISK5_MAJOR:
1039 case XEN_SCSI_DISK6_MAJOR:
1040 case XEN_SCSI_DISK7_MAJOR:
1041 *offset = (*minor / PARTS_PER_DISK) +
1042 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1043 EMULATED_SD_DISK_NAME_OFFSET;
1044 *minor = *minor +
1045 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1046 EMULATED_SD_DISK_MINOR_OFFSET;
1047 break;
1048 case XEN_SCSI_DISK8_MAJOR:
1049 case XEN_SCSI_DISK9_MAJOR:
1050 case XEN_SCSI_DISK10_MAJOR:
1051 case XEN_SCSI_DISK11_MAJOR:
1052 case XEN_SCSI_DISK12_MAJOR:
1053 case XEN_SCSI_DISK13_MAJOR:
1054 case XEN_SCSI_DISK14_MAJOR:
1055 case XEN_SCSI_DISK15_MAJOR:
1056 *offset = (*minor / PARTS_PER_DISK) +
1057 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1058 EMULATED_SD_DISK_NAME_OFFSET;
1059 *minor = *minor +
1060 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1061 EMULATED_SD_DISK_MINOR_OFFSET;
1062 break;
1063 case XENVBD_MAJOR:
1064 *offset = *minor / PARTS_PER_DISK;
1065 break;
1066 default:
1067 printk(KERN_WARNING "blkfront: your disk configuration is "
1068 "incorrect, please use an xvd device instead\n");
1069 return -ENODEV;
1070 }
1071 return 0;
1072}
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001073
Jan Beuliche77c78c2012-04-05 16:37:22 +01001074static char *encode_disk_name(char *ptr, unsigned int n)
1075{
1076 if (n >= 26)
1077 ptr = encode_disk_name(ptr, n / 26 - 1);
1078 *ptr = 'a' + n % 26;
1079 return ptr + 1;
1080}
1081
Chris Lalancette9246b5f2008-09-17 14:30:32 -07001082static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1083 struct blkfront_info *info,
Stefan Bader7c4d7d72013-05-13 16:28:15 +02001084 u16 vdisk_info, u16 sector_size,
1085 unsigned int physical_sector_size)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001086{
1087 struct gendisk *gd;
1088 int nr_minors = 1;
Stefano Stabellinic80a4202010-12-02 17:55:00 +00001089 int err;
Chris Lalancette9246b5f2008-09-17 14:30:32 -07001090 unsigned int offset;
1091 int minor;
1092 int nr_parts;
Jan Beuliche77c78c2012-04-05 16:37:22 +01001093 char *ptr;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001094
1095 BUG_ON(info->gd != NULL);
1096 BUG_ON(info->rq != NULL);
1097
Chris Lalancette9246b5f2008-09-17 14:30:32 -07001098 if ((info->vdevice>>EXT_SHIFT) > 1) {
1099 /* this is above the extended range; something is wrong */
1100 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1101 return -ENODEV;
1102 }
1103
1104 if (!VDEV_IS_EXTENDED(info->vdevice)) {
Stefano Stabellinic80a4202010-12-02 17:55:00 +00001105 err = xen_translate_vdev(info->vdevice, &minor, &offset);
1106 if (err)
1107 return err;
1108 nr_parts = PARTS_PER_DISK;
Chris Lalancette9246b5f2008-09-17 14:30:32 -07001109 } else {
1110 minor = BLKIF_MINOR_EXT(info->vdevice);
1111 nr_parts = PARTS_PER_EXT_DISK;
Stefano Stabellinic80a4202010-12-02 17:55:00 +00001112 offset = minor / nr_parts;
Stefan Bader89153b52011-07-14 15:30:37 +02001113 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
Stefano Stabellinic80a4202010-12-02 17:55:00 +00001114 printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1115 "emulated IDE disks,\n\t choose an xvd device name"
1116 "from xvde on\n", info->vdevice);
Chris Lalancette9246b5f2008-09-17 14:30:32 -07001117 }
Jan Beuliche77c78c2012-04-05 16:37:22 +01001118 if (minor >> MINORBITS) {
1119 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1120 info->vdevice, minor);
1121 return -ENODEV;
1122 }
Chris Lalancette9246b5f2008-09-17 14:30:32 -07001123
1124 if ((minor % nr_parts) == 0)
1125 nr_minors = nr_parts;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001126
Jan Beulich0e345822010-08-07 18:28:55 +02001127 err = xlbd_reserve_minors(minor, nr_minors);
1128 if (err)
1129 goto out;
1130 err = -ENODEV;
1131
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001132 gd = alloc_disk(nr_minors);
1133 if (gd == NULL)
Jan Beulich0e345822010-08-07 18:28:55 +02001134 goto release;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001135
Jan Beuliche77c78c2012-04-05 16:37:22 +01001136 strcpy(gd->disk_name, DEV_NAME);
1137 ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1138 BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1139 if (nr_minors > 1)
1140 *ptr = 0;
1141 else
1142 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1143 "%d", minor & (nr_parts - 1));
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001144
1145 gd->major = XENVBD_MAJOR;
1146 gd->first_minor = minor;
1147 gd->fops = &xlvbd_block_fops;
1148 gd->private_data = info;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001149 set_capacity(gd, capacity);
1150
Bob Liu172335a2016-07-01 17:43:39 -04001151 if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size)) {
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001152 del_gendisk(gd);
Jan Beulich0e345822010-08-07 18:28:55 +02001153 goto release;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001154 }
1155
Tejun Heo4913efe2010-09-03 11:56:16 +02001156 xlvbd_flush(info);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001157
1158 if (vdisk_info & VDISK_READONLY)
1159 set_disk_ro(gd, 1);
1160
1161 if (vdisk_info & VDISK_REMOVABLE)
1162 gd->flags |= GENHD_FL_REMOVABLE;
1163
1164 if (vdisk_info & VDISK_CDROM)
1165 gd->flags |= GENHD_FL_CD;
1166
1167 return 0;
1168
Jan Beulich0e345822010-08-07 18:28:55 +02001169 release:
1170 xlbd_release_minors(minor, nr_minors);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001171 out:
1172 return err;
1173}
1174
Daniel Stoddena66b5ae2010-08-07 18:33:17 +02001175static void xlvbd_release_gendisk(struct blkfront_info *info)
1176{
Bob Liu3df0e502015-11-14 11:12:12 +08001177 unsigned int minor, nr_minors, i;
Daniel Stoddena66b5ae2010-08-07 18:33:17 +02001178
1179 if (info->rq == NULL)
1180 return;
1181
Daniel Stoddena66b5ae2010-08-07 18:33:17 +02001182 /* No more blkif_request(). */
Bob Liu907c3eb2015-07-13 17:55:24 +08001183 blk_mq_stop_hw_queues(info->rq);
Daniel Stoddena66b5ae2010-08-07 18:33:17 +02001184
Bob Liu3df0e502015-11-14 11:12:12 +08001185 for (i = 0; i < info->nr_rings; i++) {
1186 struct blkfront_ring_info *rinfo = &info->rinfo[i];
Daniel Stoddena66b5ae2010-08-07 18:33:17 +02001187
Bob Liu3df0e502015-11-14 11:12:12 +08001188 /* No more gnttab callback work. */
1189 gnttab_cancel_free_callback(&rinfo->callback);
1190
1191 /* Flush gnttab callback work. Must be done with no locks held. */
1192 flush_work(&rinfo->work);
1193 }
Daniel Stoddena66b5ae2010-08-07 18:33:17 +02001194
1195 del_gendisk(info->gd);
1196
1197 minor = info->gd->first_minor;
1198 nr_minors = info->gd->minors;
1199 xlbd_release_minors(minor, nr_minors);
1200
1201 blk_cleanup_queue(info->rq);
Bob Liu907c3eb2015-07-13 17:55:24 +08001202 blk_mq_free_tag_set(&info->tag_set);
Daniel Stoddena66b5ae2010-08-07 18:33:17 +02001203 info->rq = NULL;
1204
1205 put_disk(info->gd);
1206 info->gd = NULL;
1207}
1208
Bob Liu11659562015-11-14 11:12:13 +08001209/* Already hold rinfo->ring_lock. */
1210static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001211{
Bob Liu81f35162015-11-14 11:12:11 +08001212 if (!RING_FULL(&rinfo->ring))
1213 blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001214}
1215
Bob Liu11659562015-11-14 11:12:13 +08001216static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1217{
1218 unsigned long flags;
1219
1220 spin_lock_irqsave(&rinfo->ring_lock, flags);
1221 kick_pending_request_queues_locked(rinfo);
1222 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1223}
1224
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001225static void blkif_restart_queue(struct work_struct *work)
1226{
Bob Liu81f35162015-11-14 11:12:11 +08001227 struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001228
Bob Liu81f35162015-11-14 11:12:11 +08001229 if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1230 kick_pending_request_queues(rinfo);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001231}
1232
Bob Liu3df0e502015-11-14 11:12:12 +08001233static void blkif_free_ring(struct blkfront_ring_info *rinfo)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001234{
Bob Liu73716df2015-11-16 16:51:39 -05001235 struct grant *persistent_gnt, *n;
Bob Liu3df0e502015-11-14 11:12:12 +08001236 struct blkfront_info *info = rinfo->dev_info;
Roger Pau Monne402b27f2013-04-18 16:06:54 +02001237 int i, j, segs;
Roger Pau Monne0a8704a2012-10-24 18:58:45 +02001238
Roger Pau Monnebfe11d62013-10-29 18:31:14 +01001239 /*
1240 * Remove indirect pages, this only happens when using indirect
1241 * descriptors but not persistent grants
1242 */
Bob Liu81f35162015-11-14 11:12:11 +08001243 if (!list_empty(&rinfo->indirect_pages)) {
Roger Pau Monnebfe11d62013-10-29 18:31:14 +01001244 struct page *indirect_page, *n;
1245
1246 BUG_ON(info->feature_persistent);
Bob Liu81f35162015-11-14 11:12:11 +08001247 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
Roger Pau Monnebfe11d62013-10-29 18:31:14 +01001248 list_del(&indirect_page->lru);
1249 __free_page(indirect_page);
1250 }
1251 }
1252
Bob Liu73716df2015-11-16 16:51:39 -05001253 /* Remove all persistent grants. */
1254 if (!list_empty(&rinfo->grants)) {
1255 list_for_each_entry_safe(persistent_gnt, n,
1256 &rinfo->grants, node) {
1257 list_del(&persistent_gnt->node);
1258 if (persistent_gnt->gref != GRANT_INVALID_REF) {
1259 gnttab_end_foreign_access(persistent_gnt->gref,
1260 0, 0UL);
1261 rinfo->persistent_gnts_c--;
1262 }
1263 if (info->feature_persistent)
1264 __free_page(persistent_gnt->page);
1265 kfree(persistent_gnt);
1266 }
1267 }
1268 BUG_ON(rinfo->persistent_gnts_c != 0);
1269
Bob Liu86839c52015-06-03 13:40:03 +08001270 for (i = 0; i < BLK_RING_SIZE(info); i++) {
Roger Pau Monne402b27f2013-04-18 16:06:54 +02001271 /*
1272 * Clear persistent grants present in requests already
1273 * on the shared ring
1274 */
Bob Liu81f35162015-11-14 11:12:11 +08001275 if (!rinfo->shadow[i].request)
Roger Pau Monne402b27f2013-04-18 16:06:54 +02001276 goto free_shadow;
1277
Bob Liu81f35162015-11-14 11:12:11 +08001278 segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1279 rinfo->shadow[i].req.u.indirect.nr_segments :
1280 rinfo->shadow[i].req.u.rw.nr_segments;
Roger Pau Monne402b27f2013-04-18 16:06:54 +02001281 for (j = 0; j < segs; j++) {
Bob Liu81f35162015-11-14 11:12:11 +08001282 persistent_gnt = rinfo->shadow[i].grants_used[j];
Roger Pau Monne402b27f2013-04-18 16:06:54 +02001283 gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
Roger Pau Monnebfe11d62013-10-29 18:31:14 +01001284 if (info->feature_persistent)
Julien Gralla7a6df22015-06-30 11:58:51 +01001285 __free_page(persistent_gnt->page);
Roger Pau Monne402b27f2013-04-18 16:06:54 +02001286 kfree(persistent_gnt);
1287 }
1288
Bob Liu81f35162015-11-14 11:12:11 +08001289 if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
Roger Pau Monne402b27f2013-04-18 16:06:54 +02001290 /*
1291 * If this is not an indirect operation don't try to
1292 * free indirect segments
1293 */
1294 goto free_shadow;
1295
1296 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
Bob Liu81f35162015-11-14 11:12:11 +08001297 persistent_gnt = rinfo->shadow[i].indirect_grants[j];
Roger Pau Monne402b27f2013-04-18 16:06:54 +02001298 gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
Julien Gralla7a6df22015-06-30 11:58:51 +01001299 __free_page(persistent_gnt->page);
Roger Pau Monne402b27f2013-04-18 16:06:54 +02001300 kfree(persistent_gnt);
1301 }
1302
1303free_shadow:
Bob Liu81f35162015-11-14 11:12:11 +08001304 kfree(rinfo->shadow[i].grants_used);
1305 rinfo->shadow[i].grants_used = NULL;
1306 kfree(rinfo->shadow[i].indirect_grants);
1307 rinfo->shadow[i].indirect_grants = NULL;
1308 kfree(rinfo->shadow[i].sg);
1309 rinfo->shadow[i].sg = NULL;
Roger Pau Monne402b27f2013-04-18 16:06:54 +02001310 }
1311
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001312 /* No more gnttab callback work. */
Bob Liu81f35162015-11-14 11:12:11 +08001313 gnttab_cancel_free_callback(&rinfo->callback);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001314
1315 /* Flush gnttab callback work. Must be done with no locks held. */
Bob Liu81f35162015-11-14 11:12:11 +08001316 flush_work(&rinfo->work);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001317
1318 /* Free resources associated with old device channel. */
Bob Liu86839c52015-06-03 13:40:03 +08001319 for (i = 0; i < info->nr_ring_pages; i++) {
Bob Liu81f35162015-11-14 11:12:11 +08001320 if (rinfo->ring_ref[i] != GRANT_INVALID_REF) {
1321 gnttab_end_foreign_access(rinfo->ring_ref[i], 0, 0);
1322 rinfo->ring_ref[i] = GRANT_INVALID_REF;
Bob Liu86839c52015-06-03 13:40:03 +08001323 }
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001324 }
Bob Liu6c647b02016-07-01 15:45:57 -04001325 free_pages((unsigned long)rinfo->ring.sring, get_order(info->nr_ring_pages * XEN_PAGE_SIZE));
Bob Liu81f35162015-11-14 11:12:11 +08001326 rinfo->ring.sring = NULL;
Bob Liu86839c52015-06-03 13:40:03 +08001327
Bob Liu81f35162015-11-14 11:12:11 +08001328 if (rinfo->irq)
1329 unbind_from_irqhandler(rinfo->irq, rinfo);
1330 rinfo->evtchn = rinfo->irq = 0;
Bob Liu3df0e502015-11-14 11:12:12 +08001331}
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001332
Bob Liu3df0e502015-11-14 11:12:12 +08001333static void blkif_free(struct blkfront_info *info, int suspend)
1334{
Bob Liu3df0e502015-11-14 11:12:12 +08001335 unsigned int i;
1336
1337 /* Prevent new requests being issued until we fix things up. */
Bob Liu3df0e502015-11-14 11:12:12 +08001338 info->connected = suspend ?
1339 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1340 /* No more blkif_request(). */
1341 if (info->rq)
1342 blk_mq_stop_hw_queues(info->rq);
1343
Bob Liu3df0e502015-11-14 11:12:12 +08001344 for (i = 0; i < info->nr_rings; i++)
1345 blkif_free_ring(&info->rinfo[i]);
1346
1347 kfree(info->rinfo);
1348 info->rinfo = NULL;
1349 info->nr_rings = 0;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001350}
1351
Julien Grallc004a6f2015-07-22 16:44:54 +01001352struct copy_from_grant {
1353 const struct blk_shadow *s;
1354 unsigned int grant_idx;
1355 unsigned int bvec_offset;
1356 char *bvec_data;
1357};
1358
1359static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1360 unsigned int len, void *data)
1361{
1362 struct copy_from_grant *info = data;
1363 char *shared_data;
1364 /* Convenient aliases */
1365 const struct blk_shadow *s = info->s;
1366
1367 shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1368
1369 memcpy(info->bvec_data + info->bvec_offset,
1370 shared_data + offset, len);
1371
1372 info->bvec_offset += len;
1373 info->grant_idx++;
1374
1375 kunmap_atomic(shared_data);
1376}
1377
Julien Grall6cc568332015-08-13 13:13:35 +01001378static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1379{
1380 switch (rsp)
1381 {
1382 case BLKIF_RSP_OKAY:
1383 return REQ_DONE;
1384 case BLKIF_RSP_EOPNOTSUPP:
1385 return REQ_EOPNOTSUPP;
1386 case BLKIF_RSP_ERROR:
1387 /* Fallthrough. */
1388 default:
1389 return REQ_ERROR;
1390 }
1391}
1392
1393/*
1394 * Get the final status of the block request based on two ring response
1395 */
1396static int blkif_get_final_status(enum blk_req_status s1,
1397 enum blk_req_status s2)
1398{
1399 BUG_ON(s1 == REQ_WAITING);
1400 BUG_ON(s2 == REQ_WAITING);
1401
1402 if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1403 return BLKIF_RSP_ERROR;
1404 else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1405 return BLKIF_RSP_EOPNOTSUPP;
1406 return BLKIF_RSP_OKAY;
1407}
1408
1409static bool blkif_completion(unsigned long *id,
1410 struct blkfront_ring_info *rinfo,
Roger Pau Monne0a8704a2012-10-24 18:58:45 +02001411 struct blkif_response *bret)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001412{
Roger Pau Monned62f6912012-12-07 19:00:31 +01001413 int i = 0;
Roger Pau Monneb7649152013-05-02 10:58:50 +02001414 struct scatterlist *sg;
Julien Grallc004a6f2015-07-22 16:44:54 +01001415 int num_sg, num_grant;
Bob Liu81f35162015-11-14 11:12:11 +08001416 struct blkfront_info *info = rinfo->dev_info;
Julien Grall6cc568332015-08-13 13:13:35 +01001417 struct blk_shadow *s = &rinfo->shadow[*id];
Julien Grallc004a6f2015-07-22 16:44:54 +01001418 struct copy_from_grant data = {
Julien Grallc004a6f2015-07-22 16:44:54 +01001419 .grant_idx = 0,
1420 };
Roger Pau Monne402b27f2013-04-18 16:06:54 +02001421
Julien Grallc004a6f2015-07-22 16:44:54 +01001422 num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
Roger Pau Monne402b27f2013-04-18 16:06:54 +02001423 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
Julien Grall6cc568332015-08-13 13:13:35 +01001424
1425 /* The I/O request may be split in two. */
1426 if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1427 struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1428
1429 /* Keep the status of the current response in shadow. */
1430 s->status = blkif_rsp_to_req_status(bret->status);
1431
1432 /* Wait the second response if not yet here. */
1433 if (s2->status == REQ_WAITING)
1434 return 0;
1435
1436 bret->status = blkif_get_final_status(s->status,
1437 s2->status);
1438
1439 /*
1440 * All the grants is stored in the first shadow in order
1441 * to make the completion code simpler.
1442 */
1443 num_grant += s2->req.u.rw.nr_segments;
1444
1445 /*
1446 * The two responses may not come in order. Only the
1447 * first request will store the scatter-gather list.
1448 */
1449 if (s2->num_sg != 0) {
1450 /* Update "id" with the ID of the first response. */
1451 *id = s->associated_id;
1452 s = s2;
1453 }
1454
1455 /*
1456 * We don't need anymore the second request, so recycling
1457 * it now.
1458 */
1459 if (add_id_to_freelist(rinfo, s->associated_id))
1460 WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1461 info->gd->disk_name, s->associated_id);
1462 }
1463
1464 data.s = s;
Julien Grallc004a6f2015-07-22 16:44:54 +01001465 num_sg = s->num_sg;
Roger Pau Monne0a8704a2012-10-24 18:58:45 +02001466
Roger Pau Monnebfe11d62013-10-29 18:31:14 +01001467 if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
Julien Grallc004a6f2015-07-22 16:44:54 +01001468 for_each_sg(s->sg, sg, num_sg, i) {
Roger Pau Monneb7649152013-05-02 10:58:50 +02001469 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
Julien Grallc004a6f2015-07-22 16:44:54 +01001470
1471 data.bvec_offset = sg->offset;
1472 data.bvec_data = kmap_atomic(sg_page(sg));
1473
1474 gnttab_foreach_grant_in_range(sg_page(sg),
1475 sg->offset,
1476 sg->length,
1477 blkif_copy_from_grant,
1478 &data);
1479
1480 kunmap_atomic(data.bvec_data);
Roger Pau Monne0a8704a2012-10-24 18:58:45 +02001481 }
1482 }
1483 /* Add the persistent grant into the list of free grants */
Julien Grallc004a6f2015-07-22 16:44:54 +01001484 for (i = 0; i < num_grant; i++) {
Roger Pau Monnefbe363c2013-08-12 12:53:44 +02001485 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1486 /*
1487 * If the grant is still mapped by the backend (the
1488 * backend has chosen to make this grant persistent)
1489 * we add it at the head of the list, so it will be
1490 * reused first.
1491 */
Roger Pau Monnebfe11d62013-10-29 18:31:14 +01001492 if (!info->feature_persistent)
1493 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1494 s->grants_used[i]->gref);
Bob Liu73716df2015-11-16 16:51:39 -05001495 list_add(&s->grants_used[i]->node, &rinfo->grants);
1496 rinfo->persistent_gnts_c++;
Roger Pau Monnefbe363c2013-08-12 12:53:44 +02001497 } else {
1498 /*
1499 * If the grant is not mapped by the backend we end the
1500 * foreign access and add it to the tail of the list,
1501 * so it will not be picked again unless we run out of
1502 * persistent grants.
1503 */
1504 gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1505 s->grants_used[i]->gref = GRANT_INVALID_REF;
Bob Liu73716df2015-11-16 16:51:39 -05001506 list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
Roger Pau Monnefbe363c2013-08-12 12:53:44 +02001507 }
Roger Pau Monne0a8704a2012-10-24 18:58:45 +02001508 }
Roger Pau Monne402b27f2013-04-18 16:06:54 +02001509 if (s->req.operation == BLKIF_OP_INDIRECT) {
Julien Grallc004a6f2015-07-22 16:44:54 +01001510 for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
Roger Pau Monnefbe363c2013-08-12 12:53:44 +02001511 if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
Roger Pau Monnebfe11d62013-10-29 18:31:14 +01001512 if (!info->feature_persistent)
1513 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1514 s->indirect_grants[i]->gref);
Bob Liu73716df2015-11-16 16:51:39 -05001515 list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1516 rinfo->persistent_gnts_c++;
Roger Pau Monnefbe363c2013-08-12 12:53:44 +02001517 } else {
Roger Pau Monnebfe11d62013-10-29 18:31:14 +01001518 struct page *indirect_page;
1519
Roger Pau Monnefbe363c2013-08-12 12:53:44 +02001520 gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
Roger Pau Monnebfe11d62013-10-29 18:31:14 +01001521 /*
1522 * Add the used indirect page back to the list of
1523 * available pages for indirect grefs.
1524 */
Bob Liu7b076752015-07-22 14:40:09 +08001525 if (!info->feature_persistent) {
Julien Gralla7a6df22015-06-30 11:58:51 +01001526 indirect_page = s->indirect_grants[i]->page;
Bob Liu81f35162015-11-14 11:12:11 +08001527 list_add(&indirect_page->lru, &rinfo->indirect_pages);
Bob Liu7b076752015-07-22 14:40:09 +08001528 }
Roger Pau Monnefbe363c2013-08-12 12:53:44 +02001529 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
Bob Liu73716df2015-11-16 16:51:39 -05001530 list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
Roger Pau Monnefbe363c2013-08-12 12:53:44 +02001531 }
Roger Pau Monne402b27f2013-04-18 16:06:54 +02001532 }
1533 }
Julien Grall6cc568332015-08-13 13:13:35 +01001534
1535 return 1;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001536}
1537
1538static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1539{
1540 struct request *req;
1541 struct blkif_response *bret;
1542 RING_IDX i, rp;
1543 unsigned long flags;
Bob Liu81f35162015-11-14 11:12:11 +08001544 struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1545 struct blkfront_info *info = rinfo->dev_info;
Christoph Hellwigf4829a92015-09-27 21:01:50 +02001546 int error;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001547
Bob Liu11659562015-11-14 11:12:13 +08001548 if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001549 return IRQ_HANDLED;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001550
Bob Liu11659562015-11-14 11:12:13 +08001551 spin_lock_irqsave(&rinfo->ring_lock, flags);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001552 again:
Bob Liu81f35162015-11-14 11:12:11 +08001553 rp = rinfo->ring.sring->rsp_prod;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001554 rmb(); /* Ensure we see queued responses up to 'rp'. */
1555
Bob Liu81f35162015-11-14 11:12:11 +08001556 for (i = rinfo->ring.rsp_cons; i != rp; i++) {
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001557 unsigned long id;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001558
Bob Liu81f35162015-11-14 11:12:11 +08001559 bret = RING_GET_RESPONSE(&rinfo->ring, i);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001560 id = bret->id;
Konrad Rzeszutek Wilk6878c322012-05-25 17:34:51 -04001561 /*
1562 * The backend has messed up and given us an id that we would
1563 * never have given to it (we stamp it up to BLK_RING_SIZE -
1564 * look in get_id_from_freelist.
1565 */
Bob Liu86839c52015-06-03 13:40:03 +08001566 if (id >= BLK_RING_SIZE(info)) {
Konrad Rzeszutek Wilk6878c322012-05-25 17:34:51 -04001567 WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1568 info->gd->disk_name, op_name(bret->operation), id);
1569 /* We can't safely get the 'struct request' as
1570 * the id is busted. */
1571 continue;
1572 }
Bob Liu81f35162015-11-14 11:12:11 +08001573 req = rinfo->shadow[id].request;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001574
Julien Grall6cc568332015-08-13 13:13:35 +01001575 if (bret->operation != BLKIF_OP_DISCARD) {
1576 /*
1577 * We may need to wait for an extra response if the
1578 * I/O request is split in 2
1579 */
1580 if (!blkif_completion(&id, rinfo, bret))
1581 continue;
1582 }
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001583
Bob Liu81f35162015-11-14 11:12:11 +08001584 if (add_id_to_freelist(rinfo, id)) {
Konrad Rzeszutek Wilk6878c322012-05-25 17:34:51 -04001585 WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1586 info->gd->disk_name, op_name(bret->operation), id);
1587 continue;
1588 }
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001589
Christoph Hellwigf4829a92015-09-27 21:01:50 +02001590 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001591 switch (bret->operation) {
Li Dongyanged30bf32011-09-01 18:39:09 +08001592 case BLKIF_OP_DISCARD:
1593 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1594 struct request_queue *rq = info->rq;
Konrad Rzeszutek Wilk6878c322012-05-25 17:34:51 -04001595 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1596 info->gd->disk_name, op_name(bret->operation));
Christoph Hellwigf4829a92015-09-27 21:01:50 +02001597 error = -EOPNOTSUPP;
Li Dongyanged30bf32011-09-01 18:39:09 +08001598 info->feature_discard = 0;
Konrad Rzeszutek Wilk5ea42982011-10-12 16:23:30 -04001599 info->feature_secdiscard = 0;
Li Dongyanged30bf32011-09-01 18:39:09 +08001600 queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
Christoph Hellwig288dab82016-06-09 16:00:36 +02001601 queue_flag_clear(QUEUE_FLAG_SECERASE, rq);
Li Dongyanged30bf32011-09-01 18:39:09 +08001602 }
Christoph Hellwigf4829a92015-09-27 21:01:50 +02001603 blk_mq_complete_request(req, error);
Li Dongyanged30bf32011-09-01 18:39:09 +08001604 break;
Konrad Rzeszutek Wilkedf6ef52011-05-03 12:01:11 -04001605 case BLKIF_OP_FLUSH_DISKCACHE:
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001606 case BLKIF_OP_WRITE_BARRIER:
1607 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
Konrad Rzeszutek Wilk6878c322012-05-25 17:34:51 -04001608 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1609 info->gd->disk_name, op_name(bret->operation));
Christoph Hellwigf4829a92015-09-27 21:01:50 +02001610 error = -EOPNOTSUPP;
Jeremy Fitzhardingedcb8bae2010-11-02 11:55:58 -04001611 }
1612 if (unlikely(bret->status == BLKIF_RSP_ERROR &&
Bob Liu81f35162015-11-14 11:12:11 +08001613 rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
Konrad Rzeszutek Wilk6878c322012-05-25 17:34:51 -04001614 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1615 info->gd->disk_name, op_name(bret->operation));
Christoph Hellwigf4829a92015-09-27 21:01:50 +02001616 error = -EOPNOTSUPP;
Jeremy Fitzhardingedcb8bae2010-11-02 11:55:58 -04001617 }
Christoph Hellwigf4829a92015-09-27 21:01:50 +02001618 if (unlikely(error)) {
1619 if (error == -EOPNOTSUPP)
1620 error = 0;
Mike Christiea4180902016-06-05 14:32:24 -05001621 info->feature_fua = 0;
Tejun Heo4913efe2010-09-03 11:56:16 +02001622 info->feature_flush = 0;
1623 xlvbd_flush(info);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001624 }
1625 /* fall through */
1626 case BLKIF_OP_READ:
1627 case BLKIF_OP_WRITE:
1628 if (unlikely(bret->status != BLKIF_RSP_OKAY))
1629 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1630 "request: %x\n", bret->status);
1631
Christoph Hellwigf4829a92015-09-27 21:01:50 +02001632 blk_mq_complete_request(req, error);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001633 break;
1634 default:
1635 BUG();
1636 }
1637 }
1638
Bob Liu81f35162015-11-14 11:12:11 +08001639 rinfo->ring.rsp_cons = i;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001640
Bob Liu81f35162015-11-14 11:12:11 +08001641 if (i != rinfo->ring.req_prod_pvt) {
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001642 int more_to_do;
Bob Liu81f35162015-11-14 11:12:11 +08001643 RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001644 if (more_to_do)
1645 goto again;
1646 } else
Bob Liu81f35162015-11-14 11:12:11 +08001647 rinfo->ring.sring->rsp_event = i + 1;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001648
Bob Liu11659562015-11-14 11:12:13 +08001649 kick_pending_request_queues_locked(rinfo);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001650
Bob Liu11659562015-11-14 11:12:13 +08001651 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001652
1653 return IRQ_HANDLED;
1654}
1655
1656
1657static int setup_blkring(struct xenbus_device *dev,
Bob Liu81f35162015-11-14 11:12:11 +08001658 struct blkfront_ring_info *rinfo)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001659{
1660 struct blkif_sring *sring;
Bob Liu86839c52015-06-03 13:40:03 +08001661 int err, i;
Bob Liu81f35162015-11-14 11:12:11 +08001662 struct blkfront_info *info = rinfo->dev_info;
Julien Grallc004a6f2015-07-22 16:44:54 +01001663 unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
Julien Grall9cce2912015-10-13 17:50:11 +01001664 grant_ref_t gref[XENBUS_MAX_RING_GRANTS];
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001665
Bob Liu86839c52015-06-03 13:40:03 +08001666 for (i = 0; i < info->nr_ring_pages; i++)
Bob Liu81f35162015-11-14 11:12:11 +08001667 rinfo->ring_ref[i] = GRANT_INVALID_REF;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001668
Bob Liu86839c52015-06-03 13:40:03 +08001669 sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
1670 get_order(ring_size));
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001671 if (!sring) {
1672 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1673 return -ENOMEM;
1674 }
1675 SHARED_RING_INIT(sring);
Bob Liu81f35162015-11-14 11:12:11 +08001676 FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001677
Bob Liu81f35162015-11-14 11:12:11 +08001678 err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001679 if (err < 0) {
Bob Liu86839c52015-06-03 13:40:03 +08001680 free_pages((unsigned long)sring, get_order(ring_size));
Bob Liu81f35162015-11-14 11:12:11 +08001681 rinfo->ring.sring = NULL;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001682 goto fail;
1683 }
Bob Liu86839c52015-06-03 13:40:03 +08001684 for (i = 0; i < info->nr_ring_pages; i++)
Bob Liu81f35162015-11-14 11:12:11 +08001685 rinfo->ring_ref[i] = gref[i];
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001686
Bob Liu81f35162015-11-14 11:12:11 +08001687 err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001688 if (err)
1689 goto fail;
1690
Bob Liu81f35162015-11-14 11:12:11 +08001691 err = bind_evtchn_to_irqhandler(rinfo->evtchn, blkif_interrupt, 0,
1692 "blkif", rinfo);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001693 if (err <= 0) {
1694 xenbus_dev_fatal(dev, err,
1695 "bind_evtchn_to_irqhandler failed");
1696 goto fail;
1697 }
Bob Liu81f35162015-11-14 11:12:11 +08001698 rinfo->irq = err;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001699
1700 return 0;
1701fail:
1702 blkif_free(info, 0);
1703 return err;
1704}
1705
Bob Liu28d949b2015-11-14 11:12:14 +08001706/*
1707 * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1708 * ring buffer may have multi pages depending on ->nr_ring_pages.
1709 */
1710static int write_per_ring_nodes(struct xenbus_transaction xbt,
1711 struct blkfront_ring_info *rinfo, const char *dir)
1712{
1713 int err;
1714 unsigned int i;
1715 const char *message = NULL;
1716 struct blkfront_info *info = rinfo->dev_info;
1717
1718 if (info->nr_ring_pages == 1) {
1719 err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1720 if (err) {
1721 message = "writing ring-ref";
1722 goto abort_transaction;
1723 }
1724 } else {
1725 for (i = 0; i < info->nr_ring_pages; i++) {
1726 char ring_ref_name[RINGREF_NAME_LEN];
1727
1728 snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1729 err = xenbus_printf(xbt, dir, ring_ref_name,
1730 "%u", rinfo->ring_ref[i]);
1731 if (err) {
1732 message = "writing ring-ref";
1733 goto abort_transaction;
1734 }
1735 }
1736 }
1737
1738 err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1739 if (err) {
1740 message = "writing event-channel";
1741 goto abort_transaction;
1742 }
1743
1744 return 0;
1745
1746abort_transaction:
1747 xenbus_transaction_end(xbt, 1);
1748 if (message)
1749 xenbus_dev_fatal(info->xbdev, err, "%s", message);
1750
1751 return err;
1752}
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001753
1754/* Common code used when first setting up, and when resuming. */
Ian Campbell203fd612009-12-04 15:33:54 +00001755static int talk_to_blkback(struct xenbus_device *dev,
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001756 struct blkfront_info *info)
1757{
1758 const char *message = NULL;
1759 struct xenbus_transaction xbt;
Bob Liu28d949b2015-11-14 11:12:14 +08001760 int err;
1761 unsigned int i, max_page_order = 0;
Bob Liu86839c52015-06-03 13:40:03 +08001762 unsigned int ring_page_order = 0;
1763
1764 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1765 "max-ring-page-order", "%u", &max_page_order);
1766 if (err != 1)
1767 info->nr_ring_pages = 1;
1768 else {
1769 ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1770 info->nr_ring_pages = 1 << ring_page_order;
1771 }
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001772
Bob Liu3df0e502015-11-14 11:12:12 +08001773 for (i = 0; i < info->nr_rings; i++) {
Bob Liu28d949b2015-11-14 11:12:14 +08001774 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1775
Bob Liu3df0e502015-11-14 11:12:12 +08001776 /* Create shared ring, alloc event channel. */
1777 err = setup_blkring(dev, rinfo);
1778 if (err)
1779 goto destroy_blkring;
1780 }
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001781
1782again:
1783 err = xenbus_transaction_start(&xbt);
1784 if (err) {
1785 xenbus_dev_fatal(dev, err, "starting transaction");
1786 goto destroy_blkring;
1787 }
1788
Bob Liu28d949b2015-11-14 11:12:14 +08001789 if (info->nr_ring_pages > 1) {
1790 err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1791 ring_page_order);
Bob Liu3df0e502015-11-14 11:12:12 +08001792 if (err) {
Bob Liu28d949b2015-11-14 11:12:14 +08001793 message = "writing ring-page-order";
Bob Liu3df0e502015-11-14 11:12:12 +08001794 goto abort_transaction;
1795 }
Bob Liu28d949b2015-11-14 11:12:14 +08001796 }
1797
1798 /* We already got the number of queues/rings in _probe */
1799 if (info->nr_rings == 1) {
1800 err = write_per_ring_nodes(xbt, &info->rinfo[0], dev->nodename);
1801 if (err)
1802 goto destroy_blkring;
Bob Liu3df0e502015-11-14 11:12:12 +08001803 } else {
Bob Liu28d949b2015-11-14 11:12:14 +08001804 char *path;
1805 size_t pathsize;
1806
1807 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1808 info->nr_rings);
1809 if (err) {
1810 message = "writing multi-queue-num-queues";
1811 goto abort_transaction;
1812 }
1813
1814 pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1815 path = kmalloc(pathsize, GFP_KERNEL);
1816 if (!path) {
1817 err = -ENOMEM;
1818 message = "ENOMEM while writing ring references";
1819 goto abort_transaction;
1820 }
1821
1822 for (i = 0; i < info->nr_rings; i++) {
1823 memset(path, 0, pathsize);
1824 snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1825 err = write_per_ring_nodes(xbt, &info->rinfo[i], path);
1826 if (err) {
1827 kfree(path);
1828 goto destroy_blkring;
1829 }
1830 }
1831 kfree(path);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001832 }
Markus Armbruster3e334232008-04-02 10:54:02 -07001833 err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1834 XEN_IO_PROTO_ABI_NATIVE);
1835 if (err) {
1836 message = "writing protocol";
1837 goto abort_transaction;
1838 }
Roger Pau Monne0a8704a2012-10-24 18:58:45 +02001839 err = xenbus_printf(xbt, dev->nodename,
Roger Pau Monnecb5bd4d2012-11-02 16:43:04 +01001840 "feature-persistent", "%u", 1);
Roger Pau Monne0a8704a2012-10-24 18:58:45 +02001841 if (err)
1842 dev_warn(&dev->dev,
1843 "writing persistent grants feature to xenbus");
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001844
1845 err = xenbus_transaction_end(xbt, 0);
1846 if (err) {
1847 if (err == -EAGAIN)
1848 goto again;
1849 xenbus_dev_fatal(dev, err, "completing transaction");
1850 goto destroy_blkring;
1851 }
1852
Bob Liu3df0e502015-11-14 11:12:12 +08001853 for (i = 0; i < info->nr_rings; i++) {
1854 unsigned int j;
Bob Liu28d949b2015-11-14 11:12:14 +08001855 struct blkfront_ring_info *rinfo = &info->rinfo[i];
Bob Liu3df0e502015-11-14 11:12:12 +08001856
1857 for (j = 0; j < BLK_RING_SIZE(info); j++)
1858 rinfo->shadow[j].req.u.rw.id = j + 1;
1859 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1860 }
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001861 xenbus_switch_state(dev, XenbusStateInitialised);
1862
1863 return 0;
1864
1865 abort_transaction:
1866 xenbus_transaction_end(xbt, 1);
1867 if (message)
1868 xenbus_dev_fatal(dev, err, "%s", message);
1869 destroy_blkring:
1870 blkif_free(info, 0);
Bob Liu3df0e502015-11-14 11:12:12 +08001871
Konrad Rzeszutek Wilkc31ecf62015-12-18 16:28:53 -05001872 kfree(info);
1873 dev_set_drvdata(&dev->dev, NULL);
1874
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001875 return err;
1876}
1877
Bob Liu3db70a82015-11-25 17:52:55 -05001878static int negotiate_mq(struct blkfront_info *info)
1879{
1880 unsigned int backend_max_queues = 0;
1881 int err;
1882 unsigned int i;
1883
1884 BUG_ON(info->nr_rings);
1885
1886 /* Check if backend supports multiple queues. */
1887 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1888 "multi-queue-max-queues", "%u", &backend_max_queues);
1889 if (err < 0)
1890 backend_max_queues = 1;
1891
1892 info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1893 /* We need at least one ring. */
1894 if (!info->nr_rings)
1895 info->nr_rings = 1;
1896
1897 info->rinfo = kzalloc(sizeof(struct blkfront_ring_info) * info->nr_rings, GFP_KERNEL);
1898 if (!info->rinfo) {
1899 xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1900 return -ENOMEM;
1901 }
1902
1903 for (i = 0; i < info->nr_rings; i++) {
1904 struct blkfront_ring_info *rinfo;
1905
1906 rinfo = &info->rinfo[i];
1907 INIT_LIST_HEAD(&rinfo->indirect_pages);
1908 INIT_LIST_HEAD(&rinfo->grants);
1909 rinfo->dev_info = info;
1910 INIT_WORK(&rinfo->work, blkif_restart_queue);
1911 spin_lock_init(&rinfo->ring_lock);
1912 }
1913 return 0;
1914}
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001915/**
1916 * Entry point to this code when a new device is created. Allocate the basic
1917 * structures and the ring buffer for communication with the backend, and
1918 * inform the backend of the appropriate details for those. Switch to
1919 * Initialised state.
1920 */
1921static int blkfront_probe(struct xenbus_device *dev,
1922 const struct xenbus_device_id *id)
1923{
Bob Liu86839c52015-06-03 13:40:03 +08001924 int err, vdevice;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001925 struct blkfront_info *info;
1926
1927 /* FIXME: Use dynamic device id if this is not set. */
1928 err = xenbus_scanf(XBT_NIL, dev->nodename,
1929 "virtual-device", "%i", &vdevice);
1930 if (err != 1) {
Chris Lalancette9246b5f2008-09-17 14:30:32 -07001931 /* go looking in the extended area instead */
1932 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1933 "%i", &vdevice);
1934 if (err != 1) {
1935 xenbus_dev_fatal(dev, err, "reading virtual-device");
1936 return err;
1937 }
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001938 }
1939
Stefano Stabellinib98a4092010-07-29 14:53:16 +01001940 if (xen_hvm_domain()) {
1941 char *type;
1942 int len;
1943 /* no unplug has been done: do not hook devices != xen vbds */
Konrad Rzeszutek Wilk51c71a32013-11-26 15:05:40 -05001944 if (xen_has_pv_and_legacy_disk_devices()) {
Stefano Stabellinib98a4092010-07-29 14:53:16 +01001945 int major;
Stefano Stabellinic1c54132010-05-14 12:44:30 +01001946
Stefano Stabellinib98a4092010-07-29 14:53:16 +01001947 if (!VDEV_IS_EXTENDED(vdevice))
1948 major = BLKIF_MAJOR(vdevice);
1949 else
1950 major = XENVBD_MAJOR;
Stefano Stabellinic1c54132010-05-14 12:44:30 +01001951
Stefano Stabellinib98a4092010-07-29 14:53:16 +01001952 if (major != XENVBD_MAJOR) {
1953 printk(KERN_INFO
1954 "%s: HVM does not support vbd %d as xen block device\n",
Rasmus Villemoes02f1f212015-02-12 15:01:31 -08001955 __func__, vdevice);
Stefano Stabellinib98a4092010-07-29 14:53:16 +01001956 return -ENODEV;
1957 }
1958 }
1959 /* do not create a PV cdrom device if we are an HVM guest */
1960 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1961 if (IS_ERR(type))
1962 return -ENODEV;
1963 if (strncmp(type, "cdrom", 5) == 0) {
1964 kfree(type);
Stefano Stabellinic1c54132010-05-14 12:44:30 +01001965 return -ENODEV;
1966 }
Stefano Stabellinib98a4092010-07-29 14:53:16 +01001967 kfree(type);
Stefano Stabellinic1c54132010-05-14 12:44:30 +01001968 }
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001969 info = kzalloc(sizeof(*info), GFP_KERNEL);
1970 if (!info) {
1971 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1972 return -ENOMEM;
1973 }
1974
Bob Liu28d949b2015-11-14 11:12:14 +08001975 info->xbdev = dev;
Bob Liu3db70a82015-11-25 17:52:55 -05001976 err = negotiate_mq(info);
1977 if (err) {
Bob Liu3df0e502015-11-14 11:12:12 +08001978 kfree(info);
Bob Liu3db70a82015-11-25 17:52:55 -05001979 return err;
Bob Liu3df0e502015-11-14 11:12:12 +08001980 }
Bob Liu81f35162015-11-14 11:12:11 +08001981
Daniel Stoddenb70f5fa2010-04-30 22:01:19 +00001982 mutex_init(&info->mutex);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001983 info->vdevice = vdevice;
1984 info->connected = BLKIF_STATE_DISCONNECTED;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001985
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001986 /* Front end dir is a number, which is used as the id. */
1987 info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
Greg Kroah-Hartmana1b4b122009-04-30 14:43:31 -07001988 dev_set_drvdata(&dev->dev, info);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001989
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001990 return 0;
1991}
1992
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001993static void split_bio_end(struct bio *bio)
Roger Pau Monne402b27f2013-04-18 16:06:54 +02001994{
1995 struct split_bio *split_bio = bio->bi_private;
1996
Roger Pau Monne402b27f2013-04-18 16:06:54 +02001997 if (atomic_dec_and_test(&split_bio->pending)) {
1998 split_bio->bio->bi_phys_segments = 0;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001999 split_bio->bio->bi_error = bio->bi_error;
2000 bio_endio(split_bio->bio);
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002001 kfree(split_bio);
2002 }
2003 bio_put(bio);
2004}
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002005
2006static int blkif_recover(struct blkfront_info *info)
2007{
Bob Liu3df0e502015-11-14 11:12:12 +08002008 unsigned int i, r_index;
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002009 struct request *req, *n;
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002010 int rc;
2011 struct bio *bio, *cloned_bio;
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002012 unsigned int segs, offset;
2013 int pending, size;
2014 struct split_bio *split_bio;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002015
Bob Liu3df0e502015-11-14 11:12:12 +08002016 blkfront_gather_backend_features(info);
Bob Liu172335a2016-07-01 17:43:39 -04002017 /* Reset limits changed by blk_mq_update_nr_hw_queues(). */
2018 blkif_set_queue_limits(info);
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002019 segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
Bob Liu6c647b02016-07-01 15:45:57 -04002020 blk_queue_max_segments(info->rq, segs / GRANTS_PER_PSEG);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002021
Bob Liu3df0e502015-11-14 11:12:12 +08002022 for (r_index = 0; r_index < info->nr_rings; r_index++) {
Bob Liu7b427a52016-06-27 16:33:24 +08002023 struct blkfront_ring_info *rinfo = &info->rinfo[r_index];
Bob Liu3df0e502015-11-14 11:12:12 +08002024
2025 rc = blkfront_setup_indirect(rinfo);
Bob Liu7b427a52016-06-27 16:33:24 +08002026 if (rc)
Bob Liu3df0e502015-11-14 11:12:12 +08002027 return rc;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002028 }
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002029 xenbus_switch_state(info->xbdev, XenbusStateConnected);
2030
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002031 /* Now safe for us to use the shared ring */
2032 info->connected = BLKIF_STATE_CONNECTED;
2033
Bob Liu3df0e502015-11-14 11:12:12 +08002034 for (r_index = 0; r_index < info->nr_rings; r_index++) {
2035 struct blkfront_ring_info *rinfo;
2036
2037 rinfo = &info->rinfo[r_index];
2038 /* Kick any other new requests queued since we resumed */
2039 kick_pending_request_queues(rinfo);
2040 }
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002041
Bob Liu7b427a52016-06-27 16:33:24 +08002042 list_for_each_entry_safe(req, n, &info->requests, queuelist) {
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002043 /* Requeue pending requests (flush or discard) */
2044 list_del_init(&req->queuelist);
2045 BUG_ON(req->nr_phys_segments > segs);
Bob Liu907c3eb2015-07-13 17:55:24 +08002046 blk_mq_requeue_request(req);
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002047 }
Bob Liu907c3eb2015-07-13 17:55:24 +08002048 blk_mq_kick_requeue_list(info->rq);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002049
Bob Liu7b427a52016-06-27 16:33:24 +08002050 while ((bio = bio_list_pop(&info->bio_list)) != NULL) {
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002051 /* Traverse the list of pending bios and re-queue them */
2052 if (bio_segments(bio) > segs) {
2053 /*
2054 * This bio has more segments than what we can
2055 * handle, we have to split it.
2056 */
2057 pending = (bio_segments(bio) + segs - 1) / segs;
2058 split_bio = kzalloc(sizeof(*split_bio), GFP_NOIO);
2059 BUG_ON(split_bio == NULL);
2060 atomic_set(&split_bio->pending, pending);
2061 split_bio->bio = bio;
2062 for (i = 0; i < pending; i++) {
Julien Grallc004a6f2015-07-22 16:44:54 +01002063 offset = (i * segs * XEN_PAGE_SIZE) >> 9;
2064 size = min((unsigned int)(segs * XEN_PAGE_SIZE) >> 9,
Kent Overstreet4f024f32013-10-11 15:44:27 -07002065 (unsigned int)bio_sectors(bio) - offset);
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002066 cloned_bio = bio_clone(bio, GFP_NOIO);
2067 BUG_ON(cloned_bio == NULL);
Kent Overstreet6678d832013-08-07 11:14:32 -07002068 bio_trim(cloned_bio, offset, size);
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002069 cloned_bio->bi_private = split_bio;
2070 cloned_bio->bi_end_io = split_bio_end;
Mike Christie4e49ea42016-06-05 14:31:41 -05002071 submit_bio(cloned_bio);
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002072 }
2073 /*
2074 * Now we have to wait for all those smaller bios to
2075 * end, so we can also end the "parent" bio.
2076 */
2077 continue;
2078 }
2079 /* We don't need to split this bio */
Mike Christie4e49ea42016-06-05 14:31:41 -05002080 submit_bio(bio);
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002081 }
2082
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002083 return 0;
2084}
2085
2086/**
2087 * We are reconnecting to the backend, due to a suspend/resume, or a backend
2088 * driver restart. We tear down our blkif structure and recreate it, but
2089 * leave the device-layer structures intact so that this is transparent to the
2090 * rest of the kernel.
2091 */
2092static int blkfront_resume(struct xenbus_device *dev)
2093{
Greg Kroah-Hartmana1b4b122009-04-30 14:43:31 -07002094 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
Bob Liu3db70a82015-11-25 17:52:55 -05002095 int err = 0;
Bob Liu7b427a52016-06-27 16:33:24 +08002096 unsigned int i, j;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002097
2098 dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2099
Bob Liu7b427a52016-06-27 16:33:24 +08002100 bio_list_init(&info->bio_list);
2101 INIT_LIST_HEAD(&info->requests);
2102 for (i = 0; i < info->nr_rings; i++) {
2103 struct blkfront_ring_info *rinfo = &info->rinfo[i];
2104 struct bio_list merge_bio;
2105 struct blk_shadow *shadow = rinfo->shadow;
2106
2107 for (j = 0; j < BLK_RING_SIZE(info); j++) {
2108 /* Not in use? */
2109 if (!shadow[j].request)
2110 continue;
2111
2112 /*
2113 * Get the bios in the request so we can re-queue them.
2114 */
Linus Torvaldsd05d7f42016-07-26 15:03:07 -07002115 if (req_op(shadow[i].request) == REQ_OP_FLUSH ||
2116 req_op(shadow[i].request) == REQ_OP_DISCARD ||
Linus Torvalds3fc9d692016-07-26 15:37:51 -07002117 req_op(shadow[i].request) == REQ_OP_SECURE_ERASE ||
2118 shadow[j].request->cmd_flags & REQ_FUA) {
Bob Liu7b427a52016-06-27 16:33:24 +08002119 /*
2120 * Flush operations don't contain bios, so
2121 * we need to requeue the whole request
Linus Torvalds3fc9d692016-07-26 15:37:51 -07002122 *
2123 * XXX: but this doesn't make any sense for a
2124 * write with the FUA flag set..
Bob Liu7b427a52016-06-27 16:33:24 +08002125 */
2126 list_add(&shadow[j].request->queuelist, &info->requests);
2127 continue;
2128 }
2129 merge_bio.head = shadow[j].request->bio;
2130 merge_bio.tail = shadow[j].request->biotail;
2131 bio_list_merge(&info->bio_list, &merge_bio);
2132 shadow[j].request->bio = NULL;
2133 blk_mq_end_request(shadow[j].request, 0);
2134 }
2135 }
2136
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002137 blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2138
Bob Liu3db70a82015-11-25 17:52:55 -05002139 err = negotiate_mq(info);
2140 if (err)
2141 return err;
2142
Ian Campbell203fd612009-12-04 15:33:54 +00002143 err = talk_to_blkback(dev, info);
Bob Liu2a6f71a2016-05-31 16:59:17 +08002144 if (!err)
2145 blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings);
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002146
2147 /*
2148 * We have to wait for the backend to switch to
2149 * connected state, since we want to read which
2150 * features it supports.
2151 */
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002152
2153 return err;
2154}
2155
Konrad Rzeszutek Wilk6f03a7f2015-11-16 15:14:41 -05002156static void blkfront_closing(struct blkfront_info *info)
Daniel Stoddenb70f5fa2010-04-30 22:01:19 +00002157{
2158 struct xenbus_device *xbdev = info->xbdev;
2159 struct block_device *bdev = NULL;
2160
2161 mutex_lock(&info->mutex);
2162
2163 if (xbdev->state == XenbusStateClosing) {
2164 mutex_unlock(&info->mutex);
2165 return;
2166 }
2167
2168 if (info->gd)
2169 bdev = bdget_disk(info->gd, 0);
2170
2171 mutex_unlock(&info->mutex);
2172
2173 if (!bdev) {
2174 xenbus_frontend_closed(xbdev);
2175 return;
2176 }
2177
2178 mutex_lock(&bdev->bd_mutex);
2179
Daniel Stodden7b32d102010-04-30 22:01:23 +00002180 if (bdev->bd_openers) {
Daniel Stoddenb70f5fa2010-04-30 22:01:19 +00002181 xenbus_dev_error(xbdev, -EBUSY,
2182 "Device in use; refusing to close");
2183 xenbus_switch_state(xbdev, XenbusStateClosing);
2184 } else {
2185 xlvbd_release_gendisk(info);
2186 xenbus_frontend_closed(xbdev);
2187 }
2188
2189 mutex_unlock(&bdev->bd_mutex);
2190 bdput(bdev);
2191}
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002192
Li Dongyanged30bf32011-09-01 18:39:09 +08002193static void blkfront_setup_discard(struct blkfront_info *info)
2194{
2195 int err;
Li Dongyanged30bf32011-09-01 18:39:09 +08002196 unsigned int discard_granularity;
2197 unsigned int discard_alignment;
Konrad Rzeszutek Wilk5ea42982011-10-12 16:23:30 -04002198 unsigned int discard_secure;
Li Dongyanged30bf32011-09-01 18:39:09 +08002199
Olaf Hering1c8cad62014-05-21 16:32:40 +02002200 info->feature_discard = 1;
2201 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2202 "discard-granularity", "%u", &discard_granularity,
2203 "discard-alignment", "%u", &discard_alignment,
2204 NULL);
2205 if (!err) {
2206 info->discard_granularity = discard_granularity;
2207 info->discard_alignment = discard_alignment;
2208 }
Jan Beulichff595322016-07-07 02:05:46 -06002209 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2210 "discard-secure", "%u", &discard_secure);
2211 if (err > 0)
Olaf Hering1c8cad62014-05-21 16:32:40 +02002212 info->feature_secdiscard = !!discard_secure;
Li Dongyanged30bf32011-09-01 18:39:09 +08002213}
2214
Bob Liu81f35162015-11-14 11:12:11 +08002215static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002216{
Julien Grallc004a6f2015-07-22 16:44:54 +01002217 unsigned int psegs, grants;
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002218 int err, i;
Bob Liu81f35162015-11-14 11:12:11 +08002219 struct blkfront_info *info = rinfo->dev_info;
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002220
Julien Grall6cc568332015-08-13 13:13:35 +01002221 if (info->max_indirect_segments == 0) {
2222 if (!HAS_EXTRA_REQ)
2223 grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2224 else {
2225 /*
2226 * When an extra req is required, the maximum
2227 * grants supported is related to the size of the
2228 * Linux block segment.
2229 */
2230 grants = GRANTS_PER_PSEG;
2231 }
2232 }
Bob Liud50babb2015-07-22 14:40:08 +08002233 else
Julien Grallc004a6f2015-07-22 16:44:54 +01002234 grants = info->max_indirect_segments;
2235 psegs = grants / GRANTS_PER_PSEG;
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002236
Bob Liu81f35162015-11-14 11:12:11 +08002237 err = fill_grant_buffer(rinfo,
Julien Grallc004a6f2015-07-22 16:44:54 +01002238 (grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002239 if (err)
2240 goto out_of_memory;
2241
Roger Pau Monnebfe11d62013-10-29 18:31:14 +01002242 if (!info->feature_persistent && info->max_indirect_segments) {
2243 /*
2244 * We are using indirect descriptors but not persistent
2245 * grants, we need to allocate a set of pages that can be
2246 * used for mapping indirect grefs
2247 */
Julien Grallc004a6f2015-07-22 16:44:54 +01002248 int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
Roger Pau Monnebfe11d62013-10-29 18:31:14 +01002249
Bob Liu81f35162015-11-14 11:12:11 +08002250 BUG_ON(!list_empty(&rinfo->indirect_pages));
Roger Pau Monnebfe11d62013-10-29 18:31:14 +01002251 for (i = 0; i < num; i++) {
2252 struct page *indirect_page = alloc_page(GFP_NOIO);
2253 if (!indirect_page)
2254 goto out_of_memory;
Bob Liu81f35162015-11-14 11:12:11 +08002255 list_add(&indirect_page->lru, &rinfo->indirect_pages);
Roger Pau Monnebfe11d62013-10-29 18:31:14 +01002256 }
2257 }
2258
Bob Liu86839c52015-06-03 13:40:03 +08002259 for (i = 0; i < BLK_RING_SIZE(info); i++) {
Bob Liu81f35162015-11-14 11:12:11 +08002260 rinfo->shadow[i].grants_used = kzalloc(
2261 sizeof(rinfo->shadow[i].grants_used[0]) * grants,
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002262 GFP_NOIO);
Bob Liu81f35162015-11-14 11:12:11 +08002263 rinfo->shadow[i].sg = kzalloc(sizeof(rinfo->shadow[i].sg[0]) * psegs, GFP_NOIO);
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002264 if (info->max_indirect_segments)
Bob Liu81f35162015-11-14 11:12:11 +08002265 rinfo->shadow[i].indirect_grants = kzalloc(
2266 sizeof(rinfo->shadow[i].indirect_grants[0]) *
Julien Grallc004a6f2015-07-22 16:44:54 +01002267 INDIRECT_GREFS(grants),
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002268 GFP_NOIO);
Bob Liu81f35162015-11-14 11:12:11 +08002269 if ((rinfo->shadow[i].grants_used == NULL) ||
2270 (rinfo->shadow[i].sg == NULL) ||
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002271 (info->max_indirect_segments &&
Bob Liu81f35162015-11-14 11:12:11 +08002272 (rinfo->shadow[i].indirect_grants == NULL)))
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002273 goto out_of_memory;
Bob Liu81f35162015-11-14 11:12:11 +08002274 sg_init_table(rinfo->shadow[i].sg, psegs);
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002275 }
2276
2277
2278 return 0;
2279
2280out_of_memory:
Bob Liu86839c52015-06-03 13:40:03 +08002281 for (i = 0; i < BLK_RING_SIZE(info); i++) {
Bob Liu81f35162015-11-14 11:12:11 +08002282 kfree(rinfo->shadow[i].grants_used);
2283 rinfo->shadow[i].grants_used = NULL;
2284 kfree(rinfo->shadow[i].sg);
2285 rinfo->shadow[i].sg = NULL;
2286 kfree(rinfo->shadow[i].indirect_grants);
2287 rinfo->shadow[i].indirect_grants = NULL;
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002288 }
Bob Liu81f35162015-11-14 11:12:11 +08002289 if (!list_empty(&rinfo->indirect_pages)) {
Roger Pau Monnebfe11d62013-10-29 18:31:14 +01002290 struct page *indirect_page, *n;
Bob Liu81f35162015-11-14 11:12:11 +08002291 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
Roger Pau Monnebfe11d62013-10-29 18:31:14 +01002292 list_del(&indirect_page->lru);
2293 __free_page(indirect_page);
2294 }
2295 }
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002296 return -ENOMEM;
2297}
2298
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002299/*
Bob Liud50babb2015-07-22 14:40:08 +08002300 * Gather all backend feature-*
2301 */
Bob Liu3df0e502015-11-14 11:12:12 +08002302static void blkfront_gather_backend_features(struct blkfront_info *info)
Bob Liud50babb2015-07-22 14:40:08 +08002303{
2304 int err;
2305 int barrier, flush, discard, persistent;
2306 unsigned int indirect_segments;
2307
2308 info->feature_flush = 0;
Mike Christiea4180902016-06-05 14:32:24 -05002309 info->feature_fua = 0;
Bob Liud50babb2015-07-22 14:40:08 +08002310
Jan Beulichff595322016-07-07 02:05:46 -06002311 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2312 "feature-barrier", "%d", &barrier);
Bob Liud50babb2015-07-22 14:40:08 +08002313
2314 /*
2315 * If there's no "feature-barrier" defined, then it means
2316 * we're dealing with a very old backend which writes
2317 * synchronously; nothing to do.
2318 *
2319 * If there are barriers, then we use flush.
2320 */
Linus Torvalds08fd8c172016-07-27 11:35:37 -07002321 if (err > 0 && barrier) {
Mike Christiea4180902016-06-05 14:32:24 -05002322 info->feature_flush = 1;
2323 info->feature_fua = 1;
2324 }
2325
Bob Liud50babb2015-07-22 14:40:08 +08002326 /*
2327 * And if there is "feature-flush-cache" use that above
2328 * barriers.
2329 */
Jan Beulichff595322016-07-07 02:05:46 -06002330 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2331 "feature-flush-cache", "%d", &flush);
Bob Liud50babb2015-07-22 14:40:08 +08002332
Linus Torvalds08fd8c172016-07-27 11:35:37 -07002333 if (err > 0 && flush) {
Mike Christiea4180902016-06-05 14:32:24 -05002334 info->feature_flush = 1;
2335 info->feature_fua = 0;
2336 }
Bob Liud50babb2015-07-22 14:40:08 +08002337
Jan Beulichff595322016-07-07 02:05:46 -06002338 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2339 "feature-discard", "%d", &discard);
Bob Liud50babb2015-07-22 14:40:08 +08002340
Jan Beulichff595322016-07-07 02:05:46 -06002341 if (err > 0 && discard)
Bob Liud50babb2015-07-22 14:40:08 +08002342 blkfront_setup_discard(info);
2343
Jan Beulichff595322016-07-07 02:05:46 -06002344 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2345 "feature-persistent", "%d", &persistent);
2346 if (err <= 0)
Bob Liud50babb2015-07-22 14:40:08 +08002347 info->feature_persistent = 0;
2348 else
2349 info->feature_persistent = persistent;
2350
Jan Beulichff595322016-07-07 02:05:46 -06002351 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2352 "feature-max-indirect-segments", "%u",
2353 &indirect_segments);
2354 if (err <= 0)
Bob Liud50babb2015-07-22 14:40:08 +08002355 info->max_indirect_segments = 0;
2356 else
2357 info->max_indirect_segments = min(indirect_segments,
2358 xen_blkif_max_segments);
Bob Liud50babb2015-07-22 14:40:08 +08002359}
2360
2361/*
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002362 * Invoked when the backend is finally 'ready' (and has told produced
2363 * the details about the physical device - #sectors, size, etc).
2364 */
2365static void blkfront_connect(struct blkfront_info *info)
2366{
2367 unsigned long long sectors;
2368 unsigned long sector_size;
Stefan Bader7c4d7d72013-05-13 16:28:15 +02002369 unsigned int physical_sector_size;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002370 unsigned int binfo;
Bob Liu3df0e502015-11-14 11:12:12 +08002371 int err, i;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002372
K. Y. Srinivasan1fa73be2010-03-11 13:42:26 -08002373 switch (info->connected) {
2374 case BLKIF_STATE_CONNECTED:
2375 /*
2376 * Potentially, the back-end may be signalling
2377 * a capacity change; update the capacity.
2378 */
2379 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2380 "sectors", "%Lu", &sectors);
2381 if (XENBUS_EXIST_ERR(err))
2382 return;
2383 printk(KERN_INFO "Setting capacity to %Lu\n",
2384 sectors);
2385 set_capacity(info->gd, sectors);
K. Y. Srinivasan2def1412010-03-18 15:00:54 -07002386 revalidate_disk(info->gd);
K. Y. Srinivasan1fa73be2010-03-11 13:42:26 -08002387
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002388 return;
K. Y. Srinivasan1fa73be2010-03-11 13:42:26 -08002389 case BLKIF_STATE_SUSPENDED:
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002390 /*
2391 * If we are recovering from suspension, we need to wait
2392 * for the backend to announce it's features before
2393 * reconnecting, at least we need to know if the backend
2394 * supports indirect descriptors, and how many.
2395 */
2396 blkif_recover(info);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002397 return;
2398
Jeremy Fitzhardingeb4dddb42010-03-11 15:10:40 -08002399 default:
2400 break;
K. Y. Srinivasan1fa73be2010-03-11 13:42:26 -08002401 }
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002402
2403 dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2404 __func__, info->xbdev->otherend);
2405
2406 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2407 "sectors", "%llu", &sectors,
2408 "info", "%u", &binfo,
2409 "sector-size", "%lu", &sector_size,
2410 NULL);
2411 if (err) {
2412 xenbus_dev_fatal(info->xbdev, err,
2413 "reading backend fields at %s",
2414 info->xbdev->otherend);
2415 return;
2416 }
2417
Stefan Bader7c4d7d72013-05-13 16:28:15 +02002418 /*
2419 * physcial-sector-size is a newer field, so old backends may not
2420 * provide this. Assume physical sector size to be the same as
2421 * sector_size in that case.
2422 */
2423 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2424 "physical-sector-size", "%u", &physical_sector_size);
2425 if (err != 1)
2426 physical_sector_size = sector_size;
2427
Bob Liu3df0e502015-11-14 11:12:12 +08002428 blkfront_gather_backend_features(info);
2429 for (i = 0; i < info->nr_rings; i++) {
2430 err = blkfront_setup_indirect(&info->rinfo[i]);
2431 if (err) {
2432 xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2433 info->xbdev->otherend);
2434 blkif_free(info, 0);
2435 break;
2436 }
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002437 }
2438
Stefan Bader7c4d7d72013-05-13 16:28:15 +02002439 err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
2440 physical_sector_size);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002441 if (err) {
2442 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2443 info->xbdev->otherend);
Bob Liu4e876c2b2016-07-27 17:42:04 +08002444 goto fail;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002445 }
2446
2447 xenbus_switch_state(info->xbdev, XenbusStateConnected);
2448
2449 /* Kick pending requests. */
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002450 info->connected = BLKIF_STATE_CONNECTED;
Bob Liu3df0e502015-11-14 11:12:12 +08002451 for (i = 0; i < info->nr_rings; i++)
2452 kick_pending_request_queues(&info->rinfo[i]);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002453
Dan Williams0d52c7562016-06-15 19:44:20 -07002454 device_add_disk(&info->xbdev->dev, info->gd);
Christian Limpach1d78d702008-04-02 10:54:04 -07002455
2456 info->is_ready = 1;
Bob Liu4e876c2b2016-07-27 17:42:04 +08002457 return;
2458
2459fail:
2460 blkif_free(info, 0);
2461 return;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002462}
2463
2464/**
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002465 * Callback received when the backend's state changes.
2466 */
Ian Campbell203fd612009-12-04 15:33:54 +00002467static void blkback_changed(struct xenbus_device *dev,
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002468 enum xenbus_state backend_state)
2469{
Greg Kroah-Hartmana1b4b122009-04-30 14:43:31 -07002470 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002471
Ian Campbell203fd612009-12-04 15:33:54 +00002472 dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002473
2474 switch (backend_state) {
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002475 case XenbusStateInitWait:
Bob Liua9b54bb2015-06-19 00:23:00 -04002476 if (dev->state != XenbusStateInitialising)
2477 break;
Konrad Rzeszutek Wilkc31ecf62015-12-18 16:28:53 -05002478 if (talk_to_blkback(dev, info))
Bob Liu8ab01442015-06-03 13:40:02 +08002479 break;
Bob Liu8ab01442015-06-03 13:40:02 +08002480 case XenbusStateInitialising:
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002481 case XenbusStateInitialised:
Noboru Iwamatsub78c9512009-10-13 17:22:29 -04002482 case XenbusStateReconfiguring:
2483 case XenbusStateReconfigured:
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002484 case XenbusStateUnknown:
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002485 break;
2486
2487 case XenbusStateConnected:
Bob Liuefd15352016-06-07 10:43:15 -04002488 /*
2489 * talk_to_blkback sets state to XenbusStateInitialised
2490 * and blkfront_connect sets it to XenbusStateConnected
2491 * (if connection went OK).
2492 *
2493 * If the backend (or toolstack) decides to poke at backend
2494 * state (and re-trigger the watch by setting the state repeatedly
2495 * to XenbusStateConnected (4)) we need to deal with this.
2496 * This is allowed as this is used to communicate to the guest
2497 * that the size of disk has changed!
2498 */
2499 if ((dev->state != XenbusStateInitialised) &&
2500 (dev->state != XenbusStateConnected)) {
Konrad Rzeszutek Wilkc31ecf62015-12-18 16:28:53 -05002501 if (talk_to_blkback(dev, info))
2502 break;
2503 }
Bob Liuefd15352016-06-07 10:43:15 -04002504
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002505 blkfront_connect(info);
2506 break;
2507
David Vrabel36613712014-02-04 18:53:56 +00002508 case XenbusStateClosed:
2509 if (dev->state == XenbusStateClosed)
2510 break;
2511 /* Missed the backend's Closing state -- fallthrough */
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002512 case XenbusStateClosing:
Cathy Averya54c8f02015-10-02 09:35:01 -04002513 if (info)
2514 blkfront_closing(info);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002515 break;
2516 }
2517}
2518
Daniel Stoddenfa1bd352010-04-30 22:01:22 +00002519static int blkfront_remove(struct xenbus_device *xbdev)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002520{
Daniel Stoddenfa1bd352010-04-30 22:01:22 +00002521 struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2522 struct block_device *bdev = NULL;
2523 struct gendisk *disk;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002524
Daniel Stoddenfa1bd352010-04-30 22:01:22 +00002525 dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002526
2527 blkif_free(info, 0);
2528
Daniel Stoddenfa1bd352010-04-30 22:01:22 +00002529 mutex_lock(&info->mutex);
2530
2531 disk = info->gd;
2532 if (disk)
2533 bdev = bdget_disk(disk, 0);
2534
2535 info->xbdev = NULL;
2536 mutex_unlock(&info->mutex);
2537
2538 if (!bdev) {
Jan Beulich0e345822010-08-07 18:28:55 +02002539 kfree(info);
Daniel Stoddenfa1bd352010-04-30 22:01:22 +00002540 return 0;
2541 }
2542
2543 /*
2544 * The xbdev was removed before we reached the Closed
2545 * state. See if it's safe to remove the disk. If the bdev
2546 * isn't closed yet, we let release take care of it.
2547 */
2548
2549 mutex_lock(&bdev->bd_mutex);
2550 info = disk->private_data;
2551
Daniel Stoddend54142c2010-08-07 18:51:21 +02002552 dev_warn(disk_to_dev(disk),
2553 "%s was hot-unplugged, %d stale handles\n",
2554 xbdev->nodename, bdev->bd_openers);
2555
Daniel Stodden7b32d102010-04-30 22:01:23 +00002556 if (info && !bdev->bd_openers) {
Daniel Stoddenfa1bd352010-04-30 22:01:22 +00002557 xlvbd_release_gendisk(info);
2558 disk->private_data = NULL;
2559 kfree(info);
2560 }
2561
2562 mutex_unlock(&bdev->bd_mutex);
2563 bdput(bdev);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002564
2565 return 0;
2566}
2567
Christian Limpach1d78d702008-04-02 10:54:04 -07002568static int blkfront_is_ready(struct xenbus_device *dev)
2569{
Greg Kroah-Hartmana1b4b122009-04-30 14:43:31 -07002570 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
Christian Limpach1d78d702008-04-02 10:54:04 -07002571
Jan Beulich5d7ed202010-08-07 18:31:12 +02002572 return info->is_ready && info->xbdev;
Christian Limpach1d78d702008-04-02 10:54:04 -07002573}
2574
Al Viroa63c8482008-03-02 10:23:47 -05002575static int blkif_open(struct block_device *bdev, fmode_t mode)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002576{
Daniel Stodden13961742010-08-07 18:36:53 +02002577 struct gendisk *disk = bdev->bd_disk;
2578 struct blkfront_info *info;
2579 int err = 0;
Arnd Bergmann6e9624b2010-08-07 18:25:34 +02002580
Arnd Bergmann2a48fc02010-06-02 14:28:52 +02002581 mutex_lock(&blkfront_mutex);
Arnd Bergmann6e9624b2010-08-07 18:25:34 +02002582
Daniel Stodden13961742010-08-07 18:36:53 +02002583 info = disk->private_data;
2584 if (!info) {
2585 /* xbdev gone */
2586 err = -ERESTARTSYS;
2587 goto out;
2588 }
2589
2590 mutex_lock(&info->mutex);
2591
2592 if (!info->gd)
2593 /* xbdev is closed */
2594 err = -ERESTARTSYS;
2595
2596 mutex_unlock(&info->mutex);
2597
Daniel Stodden13961742010-08-07 18:36:53 +02002598out:
Arnd Bergmann2a48fc02010-06-02 14:28:52 +02002599 mutex_unlock(&blkfront_mutex);
Daniel Stodden13961742010-08-07 18:36:53 +02002600 return err;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002601}
2602
Al Virodb2a1442013-05-05 21:52:57 -04002603static void blkif_release(struct gendisk *disk, fmode_t mode)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002604{
Al Viroa63c8482008-03-02 10:23:47 -05002605 struct blkfront_info *info = disk->private_data;
Daniel Stodden7fd152f2010-08-07 18:45:12 +02002606 struct block_device *bdev;
2607 struct xenbus_device *xbdev;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002608
Arnd Bergmann2a48fc02010-06-02 14:28:52 +02002609 mutex_lock(&blkfront_mutex);
Daniel Stodden7fd152f2010-08-07 18:45:12 +02002610
2611 bdev = bdget_disk(disk, 0);
Daniel Stodden7fd152f2010-08-07 18:45:12 +02002612
Felipe Pena2f089cb2013-11-09 13:36:09 -02002613 if (!bdev) {
2614 WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2615 goto out_mutex;
2616 }
Daniel Stoddenacfca3c2010-08-07 18:47:26 +02002617 if (bdev->bd_openers)
2618 goto out;
2619
Daniel Stodden7fd152f2010-08-07 18:45:12 +02002620 /*
2621 * Check if we have been instructed to close. We will have
2622 * deferred this request, because the bdev was still open.
2623 */
2624
2625 mutex_lock(&info->mutex);
2626 xbdev = info->xbdev;
2627
2628 if (xbdev && xbdev->state == XenbusStateClosing) {
2629 /* pending switch to state closed */
Daniel Stoddend54142c2010-08-07 18:51:21 +02002630 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
Daniel Stodden7fd152f2010-08-07 18:45:12 +02002631 xlvbd_release_gendisk(info);
2632 xenbus_frontend_closed(info->xbdev);
2633 }
2634
2635 mutex_unlock(&info->mutex);
2636
2637 if (!xbdev) {
2638 /* sudden device removal */
Daniel Stoddend54142c2010-08-07 18:51:21 +02002639 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
Daniel Stodden7fd152f2010-08-07 18:45:12 +02002640 xlvbd_release_gendisk(info);
2641 disk->private_data = NULL;
2642 kfree(info);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002643 }
Daniel Stodden7fd152f2010-08-07 18:45:12 +02002644
Jens Axboea4cc14e2010-08-08 21:50:05 -04002645out:
Andrew Jonesdad5cf62012-02-16 13:16:25 +01002646 bdput(bdev);
Felipe Pena2f089cb2013-11-09 13:36:09 -02002647out_mutex:
Arnd Bergmann2a48fc02010-06-02 14:28:52 +02002648 mutex_unlock(&blkfront_mutex);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002649}
2650
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07002651static const struct block_device_operations xlvbd_block_fops =
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002652{
2653 .owner = THIS_MODULE,
Al Viroa63c8482008-03-02 10:23:47 -05002654 .open = blkif_open,
2655 .release = blkif_release,
Ian Campbell597592d2008-02-21 13:03:45 -08002656 .getgeo = blkif_getgeo,
Arnd Bergmann8a6cfeb2010-07-08 10:18:46 +02002657 .ioctl = blkif_ioctl,
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002658};
2659
2660
Márton Némethec9c42e2010-01-10 13:39:52 +01002661static const struct xenbus_device_id blkfront_ids[] = {
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002662 { "vbd" },
2663 { "" }
2664};
2665
David Vrabel95afae42014-09-08 17:30:41 +01002666static struct xenbus_driver blkfront_driver = {
2667 .ids = blkfront_ids,
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002668 .probe = blkfront_probe,
2669 .remove = blkfront_remove,
2670 .resume = blkfront_resume,
Ian Campbell203fd612009-12-04 15:33:54 +00002671 .otherend_changed = blkback_changed,
Christian Limpach1d78d702008-04-02 10:54:04 -07002672 .is_ready = blkfront_is_ready,
David Vrabel95afae42014-09-08 17:30:41 +01002673};
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002674
2675static int __init xlblk_init(void)
2676{
Laszlo Ersek469738e2011-10-07 21:34:38 +02002677 int ret;
Bob Liu28d949b2015-11-14 11:12:14 +08002678 int nr_cpus = num_online_cpus();
Laszlo Ersek469738e2011-10-07 21:34:38 +02002679
Jeremy Fitzhardinge6e833582008-08-19 13:16:17 -07002680 if (!xen_domain())
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002681 return -ENODEV;
2682
Julien Grall9cce2912015-10-13 17:50:11 +01002683 if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
Bob Liu86839c52015-06-03 13:40:03 +08002684 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
Julien Grall9cce2912015-10-13 17:50:11 +01002685 xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
Peng Fan45fc8262015-11-25 18:26:01 +08002686 xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
Bob Liu86839c52015-06-03 13:40:03 +08002687 }
2688
Bob Liu28d949b2015-11-14 11:12:14 +08002689 if (xen_blkif_max_queues > nr_cpus) {
2690 pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2691 xen_blkif_max_queues, nr_cpus);
2692 xen_blkif_max_queues = nr_cpus;
2693 }
2694
Konrad Rzeszutek Wilk51c71a32013-11-26 15:05:40 -05002695 if (!xen_has_pv_disk_devices())
Igor Mammedovb9136d22012-03-21 15:08:38 +01002696 return -ENODEV;
2697
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002698 if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2699 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
2700 XENVBD_MAJOR, DEV_NAME);
2701 return -ENODEV;
2702 }
2703
Jan Beulich73db1442011-12-22 09:08:13 +00002704 ret = xenbus_register_frontend(&blkfront_driver);
Laszlo Ersek469738e2011-10-07 21:34:38 +02002705 if (ret) {
2706 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2707 return ret;
2708 }
2709
2710 return 0;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002711}
2712module_init(xlblk_init);
2713
2714
Jan Beulich5a60d0c2008-06-17 10:47:08 +02002715static void __exit xlblk_exit(void)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002716{
Jan Beulich86050672012-04-05 16:04:52 +01002717 xenbus_unregister_driver(&blkfront_driver);
2718 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2719 kfree(minors);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002720}
2721module_exit(xlblk_exit);
2722
2723MODULE_DESCRIPTION("Xen virtual block device frontend");
2724MODULE_LICENSE("GPL");
2725MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
Mark McLoughlind2f0c522008-04-02 10:54:05 -07002726MODULE_ALIAS("xen:vbd");
Mark McLoughlin4f93f09b2008-04-02 10:54:06 -07002727MODULE_ALIAS("xenblk");