blob: 83eb9e6bf8b06673640ff5d5ef05db1555e4912f [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;
Konrad Rzeszutek Wilk2d5dc3b2013-05-15 10:39:34 -0400128module_param_named(max, xen_blkif_max_segments, int, S_IRUGO);
129MODULE_PARM_DESC(max, "Maximum amount of segments in indirect requests (default is 32)");
Roger Pau Monne402b27f2013-04-18 16:06:54 +0200130
Bob Liu28d949b2015-11-14 11:12:14 +0800131static unsigned int xen_blkif_max_queues = 4;
132module_param_named(max_queues, xen_blkif_max_queues, uint, S_IRUGO);
133MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
134
Bob Liu86839c52015-06-03 13:40:03 +0800135/*
136 * Maximum order of pages to be used for the shared ring between front and
137 * backend, 4KB page granularity is used.
138 */
139static unsigned int xen_blkif_max_ring_order;
140module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, S_IRUGO);
141MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
142
Julien Grallc004a6f2015-07-22 16:44:54 +0100143#define BLK_RING_SIZE(info) \
144 __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
145
146#define BLK_MAX_RING_SIZE \
Julien Grall9cce2912015-10-13 17:50:11 +0100147 __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * XENBUS_MAX_RING_GRANTS)
Julien Grallc004a6f2015-07-22 16:44:54 +0100148
Bob Liu86839c52015-06-03 13:40:03 +0800149/*
Konrad Rzeszutek Wilk6f03a7f2015-11-16 15:14:41 -0500150 * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
151 * characters are enough. Define to 20 to keep consistent with backend.
Bob Liu86839c52015-06-03 13:40:03 +0800152 */
153#define RINGREF_NAME_LEN (20)
Bob Liu28d949b2015-11-14 11:12:14 +0800154/*
155 * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
156 */
157#define QUEUE_NAME_LEN (17)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700158
159/*
Bob Liu81f35162015-11-14 11:12:11 +0800160 * Per-ring info.
161 * Every blkfront device can associate with one or more blkfront_ring_info,
162 * depending on how many hardware queues/rings to be used.
163 */
164struct blkfront_ring_info {
Bob Liu11659562015-11-14 11:12:13 +0800165 /* Lock to protect data in every ring buffer. */
166 spinlock_t ring_lock;
Bob Liu81f35162015-11-14 11:12:11 +0800167 struct blkif_front_ring ring;
168 unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
169 unsigned int evtchn, irq;
170 struct work_struct work;
171 struct gnttab_free_callback callback;
172 struct blk_shadow shadow[BLK_MAX_RING_SIZE];
173 struct list_head indirect_pages;
Bob Liu73716df2015-11-16 16:51:39 -0500174 struct list_head grants;
175 unsigned int persistent_gnts_c;
Bob Liu81f35162015-11-14 11:12:11 +0800176 unsigned long shadow_free;
177 struct blkfront_info *dev_info;
178};
179
180/*
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700181 * We have one of these per vbd, whether ide, scsi or 'other'. They
182 * hang in private_data off the gendisk structure. We may end up
183 * putting all kinds of interesting stuff here :-)
184 */
185struct blkfront_info
186{
Daniel Stoddenb70f5fa2010-04-30 22:01:19 +0000187 struct mutex mutex;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700188 struct xenbus_device *xbdev;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700189 struct gendisk *gd;
190 int vdevice;
191 blkif_vdev_t handle;
192 enum blkif_state connected;
Bob Liu3df0e502015-11-14 11:12:12 +0800193 /* Number of pages per ring buffer. */
Bob Liu86839c52015-06-03 13:40:03 +0800194 unsigned int nr_ring_pages;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700195 struct request_queue *rq;
Tejun Heo4913efe2010-09-03 11:56:16 +0200196 unsigned int feature_flush;
Konrad Rzeszutek Wilk5ea42982011-10-12 16:23:30 -0400197 unsigned int feature_discard:1;
198 unsigned int feature_secdiscard:1;
Li Dongyanged30bf32011-09-01 18:39:09 +0800199 unsigned int discard_granularity;
200 unsigned int discard_alignment;
Roger Pau Monne0a8704a2012-10-24 18:58:45 +0200201 unsigned int feature_persistent:1;
Julien Grallc004a6f2015-07-22 16:44:54 +0100202 /* Number of 4KB segments handled */
Roger Pau Monne402b27f2013-04-18 16:06:54 +0200203 unsigned int max_indirect_segments;
Christian Limpach1d78d702008-04-02 10:54:04 -0700204 int is_ready;
Bob Liu907c3eb2015-07-13 17:55:24 +0800205 struct blk_mq_tag_set tag_set;
Bob Liu3df0e502015-11-14 11:12:12 +0800206 struct blkfront_ring_info *rinfo;
207 unsigned int nr_rings;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700208};
209
Jan Beulich0e345822010-08-07 18:28:55 +0200210static unsigned int nr_minors;
211static unsigned long *minors;
212static DEFINE_SPINLOCK(minor_lock);
213
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700214#define GRANT_INVALID_REF 0
215
216#define PARTS_PER_DISK 16
Chris Lalancette9246b5f2008-09-17 14:30:32 -0700217#define PARTS_PER_EXT_DISK 256
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700218
219#define BLKIF_MAJOR(dev) ((dev)>>8)
220#define BLKIF_MINOR(dev) ((dev) & 0xff)
221
Chris Lalancette9246b5f2008-09-17 14:30:32 -0700222#define EXT_SHIFT 28
223#define EXTENDED (1<<EXT_SHIFT)
224#define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
225#define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
Stefano Stabellinic80a4202010-12-02 17:55:00 +0000226#define EMULATED_HD_DISK_MINOR_OFFSET (0)
227#define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
Stefan Bader196cfe22011-07-14 15:30:22 +0200228#define EMULATED_SD_DISK_MINOR_OFFSET (0)
229#define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700230
Chris Lalancette9246b5f2008-09-17 14:30:32 -0700231#define DEV_NAME "xvd" /* name in /dev */
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700232
Julien Grallc004a6f2015-07-22 16:44:54 +0100233/*
234 * Grants are always the same size as a Xen page (i.e 4KB).
235 * A physical segment is always the same size as a Linux page.
236 * Number of grants per physical segment
237 */
238#define GRANTS_PER_PSEG (PAGE_SIZE / XEN_PAGE_SIZE)
239
240#define GRANTS_PER_INDIRECT_FRAME \
241 (XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
242
243#define PSEGS_PER_INDIRECT_FRAME \
244 (GRANTS_INDIRECT_FRAME / GRANTS_PSEGS)
245
246#define INDIRECT_GREFS(_grants) \
247 DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
248
249#define GREFS(_psegs) ((_psegs) * GRANTS_PER_PSEG)
Roger Pau Monne402b27f2013-04-18 16:06:54 +0200250
Bob Liu81f35162015-11-14 11:12:11 +0800251static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
Bob Liu3df0e502015-11-14 11:12:12 +0800252static void blkfront_gather_backend_features(struct blkfront_info *info);
Roger Pau Monne402b27f2013-04-18 16:06:54 +0200253
Bob Liu81f35162015-11-14 11:12:11 +0800254static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700255{
Bob Liu81f35162015-11-14 11:12:11 +0800256 unsigned long free = rinfo->shadow_free;
257
258 BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
259 rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
260 rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700261 return free;
262}
263
Bob Liu81f35162015-11-14 11:12:11 +0800264static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
Konrad Rzeszutek Wilk6f03a7f2015-11-16 15:14:41 -0500265 unsigned long id)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700266{
Bob Liu81f35162015-11-14 11:12:11 +0800267 if (rinfo->shadow[id].req.u.rw.id != id)
Konrad Rzeszutek Wilk6878c322012-05-25 17:34:51 -0400268 return -EINVAL;
Bob Liu81f35162015-11-14 11:12:11 +0800269 if (rinfo->shadow[id].request == NULL)
Konrad Rzeszutek Wilk6878c322012-05-25 17:34:51 -0400270 return -EINVAL;
Bob Liu81f35162015-11-14 11:12:11 +0800271 rinfo->shadow[id].req.u.rw.id = rinfo->shadow_free;
272 rinfo->shadow[id].request = NULL;
273 rinfo->shadow_free = id;
Konrad Rzeszutek Wilk6878c322012-05-25 17:34:51 -0400274 return 0;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700275}
276
Bob Liu81f35162015-11-14 11:12:11 +0800277static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
Roger Pau Monne9c1e0502013-03-18 17:49:35 +0100278{
Bob Liu81f35162015-11-14 11:12:11 +0800279 struct blkfront_info *info = rinfo->dev_info;
Roger Pau Monne9c1e0502013-03-18 17:49:35 +0100280 struct page *granted_page;
281 struct grant *gnt_list_entry, *n;
282 int i = 0;
283
Konrad Rzeszutek Wilk6f03a7f2015-11-16 15:14:41 -0500284 while (i < num) {
Roger Pau Monne9c1e0502013-03-18 17:49:35 +0100285 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
286 if (!gnt_list_entry)
287 goto out_of_memory;
288
Roger Pau Monnebfe11d62013-10-29 18:31:14 +0100289 if (info->feature_persistent) {
290 granted_page = alloc_page(GFP_NOIO);
291 if (!granted_page) {
292 kfree(gnt_list_entry);
293 goto out_of_memory;
294 }
Julien Gralla7a6df22015-06-30 11:58:51 +0100295 gnt_list_entry->page = granted_page;
Roger Pau Monne9c1e0502013-03-18 17:49:35 +0100296 }
297
Roger Pau Monne9c1e0502013-03-18 17:49:35 +0100298 gnt_list_entry->gref = GRANT_INVALID_REF;
Bob Liu73716df2015-11-16 16:51:39 -0500299 list_add(&gnt_list_entry->node, &rinfo->grants);
Roger Pau Monne9c1e0502013-03-18 17:49:35 +0100300 i++;
301 }
302
303 return 0;
304
305out_of_memory:
306 list_for_each_entry_safe(gnt_list_entry, n,
Bob Liu73716df2015-11-16 16:51:39 -0500307 &rinfo->grants, node) {
Roger Pau Monne9c1e0502013-03-18 17:49:35 +0100308 list_del(&gnt_list_entry->node);
Roger Pau Monnebfe11d62013-10-29 18:31:14 +0100309 if (info->feature_persistent)
Julien Gralla7a6df22015-06-30 11:58:51 +0100310 __free_page(gnt_list_entry->page);
Roger Pau Monne9c1e0502013-03-18 17:49:35 +0100311 kfree(gnt_list_entry);
312 i--;
313 }
314 BUG_ON(i != 0);
315 return -ENOMEM;
316}
317
Bob Liu73716df2015-11-16 16:51:39 -0500318static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
Roger Pau Monne9c1e0502013-03-18 17:49:35 +0100319{
320 struct grant *gnt_list_entry;
Roger Pau Monne9c1e0502013-03-18 17:49:35 +0100321
Bob Liu73716df2015-11-16 16:51:39 -0500322 BUG_ON(list_empty(&rinfo->grants));
323 gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
Julien Grall4f503fb2015-07-01 14:10:38 +0100324 node);
Roger Pau Monne9c1e0502013-03-18 17:49:35 +0100325 list_del(&gnt_list_entry->node);
326
Julien Grall4f503fb2015-07-01 14:10:38 +0100327 if (gnt_list_entry->gref != GRANT_INVALID_REF)
Bob Liu73716df2015-11-16 16:51:39 -0500328 rinfo->persistent_gnts_c--;
Julien Grall4f503fb2015-07-01 14:10:38 +0100329
330 return gnt_list_entry;
331}
332
333static inline void grant_foreign_access(const struct grant *gnt_list_entry,
334 const struct blkfront_info *info)
335{
336 gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
337 info->xbdev->otherend_id,
338 gnt_list_entry->page,
339 0);
340}
341
342static struct grant *get_grant(grant_ref_t *gref_head,
343 unsigned long gfn,
Bob Liu73716df2015-11-16 16:51:39 -0500344 struct blkfront_ring_info *rinfo)
Julien Grall4f503fb2015-07-01 14:10:38 +0100345{
Bob Liu73716df2015-11-16 16:51:39 -0500346 struct grant *gnt_list_entry = get_free_grant(rinfo);
347 struct blkfront_info *info = rinfo->dev_info;
Julien Grall4f503fb2015-07-01 14:10:38 +0100348
349 if (gnt_list_entry->gref != GRANT_INVALID_REF)
Roger Pau Monne9c1e0502013-03-18 17:49:35 +0100350 return gnt_list_entry;
Julien Grall4f503fb2015-07-01 14:10:38 +0100351
352 /* Assign a gref to this page */
353 gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
354 BUG_ON(gnt_list_entry->gref == -ENOSPC);
355 if (info->feature_persistent)
356 grant_foreign_access(gnt_list_entry, info);
357 else {
358 /* Grant access to the GFN passed by the caller */
359 gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
360 info->xbdev->otherend_id,
361 gfn, 0);
Roger Pau Monne9c1e0502013-03-18 17:49:35 +0100362 }
363
Julien Grall4f503fb2015-07-01 14:10:38 +0100364 return gnt_list_entry;
365}
366
367static struct grant *get_indirect_grant(grant_ref_t *gref_head,
Bob Liu73716df2015-11-16 16:51:39 -0500368 struct blkfront_ring_info *rinfo)
Julien Grall4f503fb2015-07-01 14:10:38 +0100369{
Bob Liu73716df2015-11-16 16:51:39 -0500370 struct grant *gnt_list_entry = get_free_grant(rinfo);
371 struct blkfront_info *info = rinfo->dev_info;
Julien Grall4f503fb2015-07-01 14:10:38 +0100372
373 if (gnt_list_entry->gref != GRANT_INVALID_REF)
374 return gnt_list_entry;
375
Roger Pau Monne9c1e0502013-03-18 17:49:35 +0100376 /* Assign a gref to this page */
377 gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
378 BUG_ON(gnt_list_entry->gref == -ENOSPC);
Roger Pau Monnebfe11d62013-10-29 18:31:14 +0100379 if (!info->feature_persistent) {
Julien Grall4f503fb2015-07-01 14:10:38 +0100380 struct page *indirect_page;
381
382 /* Fetch a pre-allocated page to use for indirect grefs */
Bob Liu73716df2015-11-16 16:51:39 -0500383 BUG_ON(list_empty(&rinfo->indirect_pages));
384 indirect_page = list_first_entry(&rinfo->indirect_pages,
Julien Grall4f503fb2015-07-01 14:10:38 +0100385 struct page, lru);
386 list_del(&indirect_page->lru);
387 gnt_list_entry->page = indirect_page;
Roger Pau Monnebfe11d62013-10-29 18:31:14 +0100388 }
Julien Grall4f503fb2015-07-01 14:10:38 +0100389 grant_foreign_access(gnt_list_entry, info);
390
Roger Pau Monne9c1e0502013-03-18 17:49:35 +0100391 return gnt_list_entry;
392}
393
Konrad Rzeszutek Wilk6878c322012-05-25 17:34:51 -0400394static const char *op_name(int op)
395{
396 static const char *const names[] = {
397 [BLKIF_OP_READ] = "read",
398 [BLKIF_OP_WRITE] = "write",
399 [BLKIF_OP_WRITE_BARRIER] = "barrier",
400 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
401 [BLKIF_OP_DISCARD] = "discard" };
402
403 if (op < 0 || op >= ARRAY_SIZE(names))
404 return "unknown";
405
406 if (!names[op])
407 return "reserved";
408
409 return names[op];
410}
Jan Beulich0e345822010-08-07 18:28:55 +0200411static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
412{
413 unsigned int end = minor + nr;
414 int rc;
415
416 if (end > nr_minors) {
417 unsigned long *bitmap, *old;
418
Thomas Meyerf0941482011-11-29 22:08:00 +0100419 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
Jan Beulich0e345822010-08-07 18:28:55 +0200420 GFP_KERNEL);
421 if (bitmap == NULL)
422 return -ENOMEM;
423
424 spin_lock(&minor_lock);
425 if (end > nr_minors) {
426 old = minors;
427 memcpy(bitmap, minors,
428 BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
429 minors = bitmap;
430 nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
431 } else
432 old = bitmap;
433 spin_unlock(&minor_lock);
434 kfree(old);
435 }
436
437 spin_lock(&minor_lock);
438 if (find_next_bit(minors, end, minor) >= end) {
Akinobu Mita34ae2e42012-01-21 00:15:26 +0900439 bitmap_set(minors, minor, nr);
Jan Beulich0e345822010-08-07 18:28:55 +0200440 rc = 0;
441 } else
442 rc = -EBUSY;
443 spin_unlock(&minor_lock);
444
445 return rc;
446}
447
448static void xlbd_release_minors(unsigned int minor, unsigned int nr)
449{
450 unsigned int end = minor + nr;
451
452 BUG_ON(end > nr_minors);
453 spin_lock(&minor_lock);
Akinobu Mita34ae2e42012-01-21 00:15:26 +0900454 bitmap_clear(minors, minor, nr);
Jan Beulich0e345822010-08-07 18:28:55 +0200455 spin_unlock(&minor_lock);
456}
457
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700458static void blkif_restart_queue_callback(void *arg)
459{
Bob Liu81f35162015-11-14 11:12:11 +0800460 struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
461 schedule_work(&rinfo->work);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700462}
463
Harvey Harrisonafe42d72008-04-29 00:59:47 -0700464static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
Ian Campbell597592d2008-02-21 13:03:45 -0800465{
466 /* We don't have real geometry info, but let's at least return
467 values consistent with the size of the device */
468 sector_t nsect = get_capacity(bd->bd_disk);
469 sector_t cylinders = nsect;
470
471 hg->heads = 0xff;
472 hg->sectors = 0x3f;
473 sector_div(cylinders, hg->heads * hg->sectors);
474 hg->cylinders = cylinders;
475 if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
476 hg->cylinders = 0xffff;
477 return 0;
478}
479
Al Viroa63c8482008-03-02 10:23:47 -0500480static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
Adrian Bunk62aa0052008-08-04 11:59:05 +0200481 unsigned command, unsigned long argument)
Christian Limpach440a01a2008-06-17 10:47:08 +0200482{
Al Viroa63c8482008-03-02 10:23:47 -0500483 struct blkfront_info *info = bdev->bd_disk->private_data;
Christian Limpach440a01a2008-06-17 10:47:08 +0200484 int i;
485
486 dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
487 command, (long)argument);
488
489 switch (command) {
490 case CDROMMULTISESSION:
491 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
492 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
493 if (put_user(0, (char __user *)(argument + i)))
494 return -EFAULT;
495 return 0;
496
497 case CDROM_GET_CAPABILITY: {
498 struct gendisk *gd = info->gd;
499 if (gd->flags & GENHD_FL_CD)
500 return 0;
501 return -EINVAL;
502 }
503
504 default:
505 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
506 command);*/
507 return -EINVAL; /* same return as native Linux */
508 }
509
510 return 0;
511}
512
Julien Grall2e073962015-08-13 19:23:10 +0100513static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
514 struct request *req,
515 struct blkif_request **ring_req)
516{
517 unsigned long id;
518
519 *ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
520 rinfo->ring.req_prod_pvt++;
521
522 id = get_id_from_freelist(rinfo);
523 rinfo->shadow[id].request = req;
Julien Grall6cc568332015-08-13 13:13:35 +0100524 rinfo->shadow[id].status = REQ_WAITING;
525 rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
Julien Grall2e073962015-08-13 19:23:10 +0100526
527 (*ring_req)->u.rw.id = id;
528
529 return id;
530}
531
Bob Liu81f35162015-11-14 11:12:11 +0800532static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700533{
Bob Liu81f35162015-11-14 11:12:11 +0800534 struct blkfront_info *info = rinfo->dev_info;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700535 struct blkif_request *ring_req;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700536 unsigned long id;
Julien Grall33204662015-06-29 17:35:24 +0100537
538 /* Fill out a communications ring structure. */
Julien Grall2e073962015-08-13 19:23:10 +0100539 id = blkif_ring_get_request(rinfo, req, &ring_req);
Julien Grall33204662015-06-29 17:35:24 +0100540
541 ring_req->operation = BLKIF_OP_DISCARD;
542 ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
543 ring_req->u.discard.id = id;
544 ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
545 if ((req->cmd_flags & REQ_SECURE) && info->feature_secdiscard)
546 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
547 else
548 ring_req->u.discard.flag = 0;
549
Julien Grall33204662015-06-29 17:35:24 +0100550 /* Keep a private copy so we can reissue requests when recovering. */
Bob Liu81f35162015-11-14 11:12:11 +0800551 rinfo->shadow[id].req = *ring_req;
Julien Grall33204662015-06-29 17:35:24 +0100552
553 return 0;
554}
555
Julien Grallc004a6f2015-07-22 16:44:54 +0100556struct setup_rw_req {
557 unsigned int grant_idx;
558 struct blkif_request_segment *segments;
Bob Liu81f35162015-11-14 11:12:11 +0800559 struct blkfront_ring_info *rinfo;
Julien Grallc004a6f2015-07-22 16:44:54 +0100560 struct blkif_request *ring_req;
561 grant_ref_t gref_head;
562 unsigned int id;
563 /* Only used when persistent grant is used and it's a read request */
564 bool need_copy;
565 unsigned int bvec_off;
566 char *bvec_data;
Julien Grall6cc568332015-08-13 13:13:35 +0100567
568 bool require_extra_req;
569 struct blkif_request *extra_ring_req;
Julien Grallc004a6f2015-07-22 16:44:54 +0100570};
571
572static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
573 unsigned int len, void *data)
574{
575 struct setup_rw_req *setup = data;
576 int n, ref;
577 struct grant *gnt_list_entry;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700578 unsigned int fsect, lsect;
Julien Grallc004a6f2015-07-22 16:44:54 +0100579 /* Convenient aliases */
580 unsigned int grant_idx = setup->grant_idx;
581 struct blkif_request *ring_req = setup->ring_req;
Bob Liu81f35162015-11-14 11:12:11 +0800582 struct blkfront_ring_info *rinfo = setup->rinfo;
Julien Grall6cc568332015-08-13 13:13:35 +0100583 /*
584 * We always use the shadow of the first request to store the list
585 * of grant associated to the block I/O request. This made the
586 * completion more easy to handle even if the block I/O request is
587 * split.
588 */
Bob Liu81f35162015-11-14 11:12:11 +0800589 struct blk_shadow *shadow = &rinfo->shadow[setup->id];
Julien Grallc004a6f2015-07-22 16:44:54 +0100590
Julien Grall6cc568332015-08-13 13:13:35 +0100591 if (unlikely(setup->require_extra_req &&
592 grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
593 /*
594 * We are using the second request, setup grant_idx
595 * to be the index of the segment array.
596 */
597 grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
598 ring_req = setup->extra_ring_req;
599 }
600
Julien Grallc004a6f2015-07-22 16:44:54 +0100601 if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
602 (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
603 if (setup->segments)
604 kunmap_atomic(setup->segments);
605
606 n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
Bob Liu73716df2015-11-16 16:51:39 -0500607 gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
Julien Grallc004a6f2015-07-22 16:44:54 +0100608 shadow->indirect_grants[n] = gnt_list_entry;
609 setup->segments = kmap_atomic(gnt_list_entry->page);
610 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
611 }
612
Bob Liu73716df2015-11-16 16:51:39 -0500613 gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
Julien Grallc004a6f2015-07-22 16:44:54 +0100614 ref = gnt_list_entry->gref;
Julien Grall6cc568332015-08-13 13:13:35 +0100615 /*
616 * All the grants are stored in the shadow of the first
617 * request. Therefore we have to use the global index.
618 */
619 shadow->grants_used[setup->grant_idx] = gnt_list_entry;
Julien Grallc004a6f2015-07-22 16:44:54 +0100620
621 if (setup->need_copy) {
622 void *shared_data;
623
624 shared_data = kmap_atomic(gnt_list_entry->page);
625 /*
626 * this does not wipe data stored outside the
627 * range sg->offset..sg->offset+sg->length.
628 * Therefore, blkback *could* see data from
629 * previous requests. This is OK as long as
630 * persistent grants are shared with just one
631 * domain. It may need refactoring if this
632 * changes
633 */
634 memcpy(shared_data + offset,
635 setup->bvec_data + setup->bvec_off,
636 len);
637
638 kunmap_atomic(shared_data);
639 setup->bvec_off += len;
640 }
641
642 fsect = offset >> 9;
643 lsect = fsect + (len >> 9) - 1;
644 if (ring_req->operation != BLKIF_OP_INDIRECT) {
645 ring_req->u.rw.seg[grant_idx] =
646 (struct blkif_request_segment) {
647 .gref = ref,
648 .first_sect = fsect,
649 .last_sect = lsect };
650 } else {
651 setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
652 (struct blkif_request_segment) {
653 .gref = ref,
654 .first_sect = fsect,
655 .last_sect = lsect };
656 }
657
658 (setup->grant_idx)++;
659}
660
Julien Grall6cc568332015-08-13 13:13:35 +0100661static void blkif_setup_extra_req(struct blkif_request *first,
662 struct blkif_request *second)
663{
664 uint16_t nr_segments = first->u.rw.nr_segments;
665
666 /*
667 * The second request is only present when the first request uses
668 * all its segments. It's always the continuity of the first one.
669 */
670 first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
671
672 second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
673 second->u.rw.sector_number = first->u.rw.sector_number +
674 (BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
675
676 second->u.rw.handle = first->u.rw.handle;
677 second->operation = first->operation;
678}
679
Bob Liu81f35162015-11-14 11:12:11 +0800680static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700681{
Bob Liu81f35162015-11-14 11:12:11 +0800682 struct blkfront_info *info = rinfo->dev_info;
Julien Grall6cc568332015-08-13 13:13:35 +0100683 struct blkif_request *ring_req, *extra_ring_req = NULL;
684 unsigned long id, extra_id = NO_ASSOCIATED_ID;
685 bool require_extra_req = false;
Julien Grallc004a6f2015-07-22 16:44:54 +0100686 int i;
687 struct setup_rw_req setup = {
688 .grant_idx = 0,
689 .segments = NULL,
Bob Liu81f35162015-11-14 11:12:11 +0800690 .rinfo = rinfo,
Julien Grallc004a6f2015-07-22 16:44:54 +0100691 .need_copy = rq_data_dir(req) && info->feature_persistent,
692 };
Roger Pau Monne0a8704a2012-10-24 18:58:45 +0200693
694 /*
695 * Used to store if we are able to queue the request by just using
696 * existing persistent grants, or if we have to get new grants,
697 * as there are not sufficiently many free.
698 */
Jens Axboe9e973e62009-02-24 08:10:09 +0100699 struct scatterlist *sg;
Julien Grallc004a6f2015-07-22 16:44:54 +0100700 int num_sg, max_grefs, num_grant;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700701
Julien Grallc004a6f2015-07-22 16:44:54 +0100702 max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
Roger Pau Monnec47206e2013-08-12 12:53:43 +0200703 if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
704 /*
705 * If we are using indirect segments we need to account
706 * for the indirect grefs used in the request.
707 */
Julien Grallc004a6f2015-07-22 16:44:54 +0100708 max_grefs += INDIRECT_GREFS(max_grefs);
Roger Pau Monne402b27f2013-04-18 16:06:54 +0200709
Bob Liu3df0e502015-11-14 11:12:12 +0800710 /*
711 * We have to reserve 'max_grefs' grants because persistent
712 * grants are shared by all rings.
713 */
714 if (max_grefs > 0)
715 if (gnttab_alloc_grant_references(max_grefs, &setup.gref_head) < 0) {
Roger Pau Monne0a8704a2012-10-24 18:58:45 +0200716 gnttab_request_free_callback(
Bob Liu81f35162015-11-14 11:12:11 +0800717 &rinfo->callback,
Roger Pau Monne0a8704a2012-10-24 18:58:45 +0200718 blkif_restart_queue_callback,
Bob Liu81f35162015-11-14 11:12:11 +0800719 rinfo,
Roger Pau Monne402b27f2013-04-18 16:06:54 +0200720 max_grefs);
Roger Pau Monne0a8704a2012-10-24 18:58:45 +0200721 return 1;
722 }
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700723
724 /* Fill out a communications ring structure. */
Julien Grall2e073962015-08-13 19:23:10 +0100725 id = blkif_ring_get_request(rinfo, req, &ring_req);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700726
Bob Liu81f35162015-11-14 11:12:11 +0800727 num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
Julien Grallc004a6f2015-07-22 16:44:54 +0100728 num_grant = 0;
729 /* Calculate the number of grant used */
Bob Liu81f35162015-11-14 11:12:11 +0800730 for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
Julien Grallc004a6f2015-07-22 16:44:54 +0100731 num_grant += gnttab_count_grant(sg->offset, sg->length);
732
Julien Grall6cc568332015-08-13 13:13:35 +0100733 require_extra_req = info->max_indirect_segments == 0 &&
734 num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
735 BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
736
Bob Liu81f35162015-11-14 11:12:11 +0800737 rinfo->shadow[id].num_sg = num_sg;
Julien Grall6cc568332015-08-13 13:13:35 +0100738 if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
739 likely(!require_extra_req)) {
Julien Grall33204662015-06-29 17:35:24 +0100740 /*
741 * The indirect operation can only be a BLKIF_OP_READ or
742 * BLKIF_OP_WRITE
743 */
744 BUG_ON(req->cmd_flags & (REQ_FLUSH | REQ_FUA));
745 ring_req->operation = BLKIF_OP_INDIRECT;
746 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
747 BLKIF_OP_WRITE : BLKIF_OP_READ;
748 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
749 ring_req->u.indirect.handle = info->handle;
Julien Grallc004a6f2015-07-22 16:44:54 +0100750 ring_req->u.indirect.nr_segments = num_grant;
Li Dongyanged30bf32011-09-01 18:39:09 +0800751 } else {
Julien Grall33204662015-06-29 17:35:24 +0100752 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
753 ring_req->u.rw.handle = info->handle;
754 ring_req->operation = rq_data_dir(req) ?
755 BLKIF_OP_WRITE : BLKIF_OP_READ;
756 if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) {
Roger Pau Monne402b27f2013-04-18 16:06:54 +0200757 /*
Julien Grall33204662015-06-29 17:35:24 +0100758 * Ideally we can do an unordered flush-to-disk.
759 * In case the backend onlysupports barriers, use that.
760 * A barrier request a superset of FUA, so we can
761 * implement it the same way. (It's also a FLUSH+FUA,
762 * since it is guaranteed ordered WRT previous writes.)
Roger Pau Monne402b27f2013-04-18 16:06:54 +0200763 */
Julien Grall33204662015-06-29 17:35:24 +0100764 switch (info->feature_flush &
765 ((REQ_FLUSH|REQ_FUA))) {
766 case REQ_FLUSH|REQ_FUA:
767 ring_req->operation =
768 BLKIF_OP_WRITE_BARRIER;
769 break;
770 case REQ_FLUSH:
771 ring_req->operation =
772 BLKIF_OP_FLUSH_DISKCACHE;
773 break;
774 default:
775 ring_req->operation = 0;
Roger Pau Monne402b27f2013-04-18 16:06:54 +0200776 }
Li Dongyanged30bf32011-09-01 18:39:09 +0800777 }
Julien Grallc004a6f2015-07-22 16:44:54 +0100778 ring_req->u.rw.nr_segments = num_grant;
Julien Grall6cc568332015-08-13 13:13:35 +0100779 if (unlikely(require_extra_req)) {
780 extra_id = blkif_ring_get_request(rinfo, req,
781 &extra_ring_req);
782 /*
783 * Only the first request contains the scatter-gather
784 * list.
785 */
786 rinfo->shadow[extra_id].num_sg = 0;
787
788 blkif_setup_extra_req(ring_req, extra_ring_req);
789
790 /* Link the 2 requests together */
791 rinfo->shadow[extra_id].associated_id = id;
792 rinfo->shadow[id].associated_id = extra_id;
793 }
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700794 }
795
Julien Grallc004a6f2015-07-22 16:44:54 +0100796 setup.ring_req = ring_req;
797 setup.id = id;
Julien Grall6cc568332015-08-13 13:13:35 +0100798
799 setup.require_extra_req = require_extra_req;
800 if (unlikely(require_extra_req))
801 setup.extra_ring_req = extra_ring_req;
802
Bob Liu81f35162015-11-14 11:12:11 +0800803 for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
Julien Grallc004a6f2015-07-22 16:44:54 +0100804 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
Julien Grall33204662015-06-29 17:35:24 +0100805
Julien Grallc004a6f2015-07-22 16:44:54 +0100806 if (setup.need_copy) {
807 setup.bvec_off = sg->offset;
808 setup.bvec_data = kmap_atomic(sg_page(sg));
Julien Grall33204662015-06-29 17:35:24 +0100809 }
810
Julien Grallc004a6f2015-07-22 16:44:54 +0100811 gnttab_foreach_grant_in_range(sg_page(sg),
812 sg->offset,
813 sg->length,
814 blkif_setup_rw_req_grant,
815 &setup);
Julien Grall33204662015-06-29 17:35:24 +0100816
Julien Grallc004a6f2015-07-22 16:44:54 +0100817 if (setup.need_copy)
818 kunmap_atomic(setup.bvec_data);
Julien Grall33204662015-06-29 17:35:24 +0100819 }
Julien Grallc004a6f2015-07-22 16:44:54 +0100820 if (setup.segments)
821 kunmap_atomic(setup.segments);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700822
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700823 /* Keep a private copy so we can reissue requests when recovering. */
Bob Liu81f35162015-11-14 11:12:11 +0800824 rinfo->shadow[id].req = *ring_req;
Julien Grall6cc568332015-08-13 13:13:35 +0100825 if (unlikely(require_extra_req))
826 rinfo->shadow[extra_id].req = *extra_ring_req;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700827
Bob Liu3df0e502015-11-14 11:12:12 +0800828 if (max_grefs > 0)
Julien Grallc004a6f2015-07-22 16:44:54 +0100829 gnttab_free_grant_references(setup.gref_head);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700830
831 return 0;
832}
833
Julien Grall33204662015-06-29 17:35:24 +0100834/*
835 * Generate a Xen blkfront IO request from a blk layer request. Reads
836 * and writes are handled as expected.
837 *
838 * @req: a request struct
839 */
Bob Liu81f35162015-11-14 11:12:11 +0800840static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
Julien Grall33204662015-06-29 17:35:24 +0100841{
Bob Liu81f35162015-11-14 11:12:11 +0800842 if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
Julien Grall33204662015-06-29 17:35:24 +0100843 return 1;
844
845 if (unlikely(req->cmd_flags & (REQ_DISCARD | REQ_SECURE)))
Bob Liu81f35162015-11-14 11:12:11 +0800846 return blkif_queue_discard_req(req, rinfo);
Julien Grall33204662015-06-29 17:35:24 +0100847 else
Bob Liu81f35162015-11-14 11:12:11 +0800848 return blkif_queue_rw_req(req, rinfo);
Julien Grall33204662015-06-29 17:35:24 +0100849}
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700850
Bob Liu81f35162015-11-14 11:12:11 +0800851static inline void flush_requests(struct blkfront_ring_info *rinfo)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700852{
853 int notify;
854
Bob Liu81f35162015-11-14 11:12:11 +0800855 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700856
857 if (notify)
Bob Liu81f35162015-11-14 11:12:11 +0800858 notify_remote_via_irq(rinfo->irq);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700859}
860
Vitaly Kuznetsovad42d392014-12-01 14:01:13 +0100861static inline bool blkif_request_flush_invalid(struct request *req,
862 struct blkfront_info *info)
Arianna Avanzini0f1ca652014-08-22 13:20:02 +0200863{
864 return ((req->cmd_type != REQ_TYPE_FS) ||
Vitaly Kuznetsovad42d392014-12-01 14:01:13 +0100865 ((req->cmd_flags & REQ_FLUSH) &&
866 !(info->feature_flush & REQ_FLUSH)) ||
867 ((req->cmd_flags & REQ_FUA) &&
868 !(info->feature_flush & REQ_FUA)));
Arianna Avanzini0f1ca652014-08-22 13:20:02 +0200869}
870
Bob Liu907c3eb2015-07-13 17:55:24 +0800871static int blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
Konrad Rzeszutek Wilk6f03a7f2015-11-16 15:14:41 -0500872 const struct blk_mq_queue_data *qd)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700873{
Bob Liu11659562015-11-14 11:12:13 +0800874 unsigned long flags;
Bob Liu81f35162015-11-14 11:12:11 +0800875 struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)hctx->driver_data;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700876
Bob Liu907c3eb2015-07-13 17:55:24 +0800877 blk_mq_start_request(qd->rq);
Bob Liu11659562015-11-14 11:12:13 +0800878 spin_lock_irqsave(&rinfo->ring_lock, flags);
Bob Liu81f35162015-11-14 11:12:11 +0800879 if (RING_FULL(&rinfo->ring))
Bob Liu907c3eb2015-07-13 17:55:24 +0800880 goto out_busy;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700881
Bob Liu81f35162015-11-14 11:12:11 +0800882 if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
Bob Liu907c3eb2015-07-13 17:55:24 +0800883 goto out_err;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700884
Bob Liu81f35162015-11-14 11:12:11 +0800885 if (blkif_queue_request(qd->rq, rinfo))
Bob Liu907c3eb2015-07-13 17:55:24 +0800886 goto out_busy;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700887
Bob Liu81f35162015-11-14 11:12:11 +0800888 flush_requests(rinfo);
Bob Liu11659562015-11-14 11:12:13 +0800889 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
Bob Liu907c3eb2015-07-13 17:55:24 +0800890 return BLK_MQ_RQ_QUEUE_OK;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700891
Bob Liu907c3eb2015-07-13 17:55:24 +0800892out_err:
Bob Liu11659562015-11-14 11:12:13 +0800893 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
Bob Liu907c3eb2015-07-13 17:55:24 +0800894 return BLK_MQ_RQ_QUEUE_ERROR;
Tejun Heo296b2f62009-05-08 11:54:15 +0900895
Bob Liu907c3eb2015-07-13 17:55:24 +0800896out_busy:
Bob Liu11659562015-11-14 11:12:13 +0800897 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
Bob Liu907c3eb2015-07-13 17:55:24 +0800898 blk_mq_stop_hw_queue(hctx);
899 return BLK_MQ_RQ_QUEUE_BUSY;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700900}
901
Bob Liu81f35162015-11-14 11:12:11 +0800902static int blk_mq_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
903 unsigned int index)
904{
905 struct blkfront_info *info = (struct blkfront_info *)data;
906
Bob Liu3df0e502015-11-14 11:12:12 +0800907 BUG_ON(info->nr_rings <= index);
908 hctx->driver_data = &info->rinfo[index];
Bob Liu81f35162015-11-14 11:12:11 +0800909 return 0;
910}
911
Bob Liu907c3eb2015-07-13 17:55:24 +0800912static struct blk_mq_ops blkfront_mq_ops = {
913 .queue_rq = blkif_queue_rq,
914 .map_queue = blk_mq_map_queue,
Bob Liu81f35162015-11-14 11:12:11 +0800915 .init_hctx = blk_mq_init_hctx,
Bob Liu907c3eb2015-07-13 17:55:24 +0800916};
917
Roger Pau Monne402b27f2013-04-18 16:06:54 +0200918static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
Stefan Bader7c4d7d72013-05-13 16:28:15 +0200919 unsigned int physical_sector_size,
Roger Pau Monne402b27f2013-04-18 16:06:54 +0200920 unsigned int segments)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700921{
Jens Axboe165125e2007-07-24 09:28:11 +0200922 struct request_queue *rq;
Li Dongyanged30bf32011-09-01 18:39:09 +0800923 struct blkfront_info *info = gd->private_data;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700924
Bob Liu907c3eb2015-07-13 17:55:24 +0800925 memset(&info->tag_set, 0, sizeof(info->tag_set));
926 info->tag_set.ops = &blkfront_mq_ops;
Bob Liu28d949b2015-11-14 11:12:14 +0800927 info->tag_set.nr_hw_queues = info->nr_rings;
Julien Grall6cc568332015-08-13 13:13:35 +0100928 if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
929 /*
930 * When indirect descriptior is not supported, the I/O request
931 * will be split between multiple request in the ring.
932 * To avoid problems when sending the request, divide by
933 * 2 the depth of the queue.
934 */
935 info->tag_set.queue_depth = BLK_RING_SIZE(info) / 2;
936 } else
937 info->tag_set.queue_depth = BLK_RING_SIZE(info);
Bob Liu907c3eb2015-07-13 17:55:24 +0800938 info->tag_set.numa_node = NUMA_NO_NODE;
939 info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
940 info->tag_set.cmd_size = 0;
941 info->tag_set.driver_data = info;
942
943 if (blk_mq_alloc_tag_set(&info->tag_set))
Konrad Rzeszutek Wilkbde21f72015-11-25 13:07:39 -0500944 return -EINVAL;
Bob Liu907c3eb2015-07-13 17:55:24 +0800945 rq = blk_mq_init_queue(&info->tag_set);
946 if (IS_ERR(rq)) {
947 blk_mq_free_tag_set(&info->tag_set);
Konrad Rzeszutek Wilkbde21f72015-11-25 13:07:39 -0500948 return PTR_ERR(rq);
Bob Liu907c3eb2015-07-13 17:55:24 +0800949 }
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700950
Fernando Luis Vázquez Cao66d352e2008-10-27 18:45:54 +0900951 queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700952
Li Dongyanged30bf32011-09-01 18:39:09 +0800953 if (info->feature_discard) {
954 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
955 blk_queue_max_discard_sectors(rq, get_capacity(gd));
956 rq->limits.discard_granularity = info->discard_granularity;
957 rq->limits.discard_alignment = info->discard_alignment;
Konrad Rzeszutek Wilk5ea42982011-10-12 16:23:30 -0400958 if (info->feature_secdiscard)
959 queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD, rq);
Li Dongyanged30bf32011-09-01 18:39:09 +0800960 }
961
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700962 /* Hard sector size and max sectors impersonate the equiv. hardware. */
Martin K. Petersene1defc42009-05-22 17:17:49 -0400963 blk_queue_logical_block_size(rq, sector_size);
Stefan Bader7c4d7d72013-05-13 16:28:15 +0200964 blk_queue_physical_block_size(rq, physical_sector_size);
Julien Grallc004a6f2015-07-22 16:44:54 +0100965 blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700966
967 /* Each segment in a request is up to an aligned page in size. */
968 blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
969 blk_queue_max_segment_size(rq, PAGE_SIZE);
970
971 /* Ensure a merged request will fit in a single I/O ring slot. */
Julien Grallc004a6f2015-07-22 16:44:54 +0100972 blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700973
974 /* Make sure buffer addresses are sector-aligned. */
975 blk_queue_dma_alignment(rq, 511);
976
Ian Campbell1c91fe12008-06-17 10:47:08 +0200977 /* Make sure we don't use bounce buffers. */
978 blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
979
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700980 gd->queue = rq;
981
982 return 0;
983}
984
Vitaly Kuznetsovfdf9b962014-12-09 15:25:10 +0100985static const char *flush_info(unsigned int feature_flush)
986{
987 switch (feature_flush & ((REQ_FLUSH | REQ_FUA))) {
988 case REQ_FLUSH|REQ_FUA:
989 return "barrier: enabled;";
990 case REQ_FLUSH:
991 return "flush diskcache: enabled;";
992 default:
993 return "barrier or flush: disabled;";
994 }
995}
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700996
Tejun Heo4913efe2010-09-03 11:56:16 +0200997static void xlvbd_flush(struct blkfront_info *info)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -0700998{
Tejun Heo4913efe2010-09-03 11:56:16 +0200999 blk_queue_flush(info->rq, info->feature_flush);
Vitaly Kuznetsovfdf9b962014-12-09 15:25:10 +01001000 pr_info("blkfront: %s: %s %s %s %s %s\n",
1001 info->gd->disk_name, flush_info(info->feature_flush),
1002 "persistent grants:", info->feature_persistent ?
1003 "enabled;" : "disabled;", "indirect descriptors:",
1004 info->max_indirect_segments ? "enabled;" : "disabled;");
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001005}
1006
Stefano Stabellinic80a4202010-12-02 17:55:00 +00001007static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
1008{
1009 int major;
1010 major = BLKIF_MAJOR(vdevice);
1011 *minor = BLKIF_MINOR(vdevice);
1012 switch (major) {
1013 case XEN_IDE0_MAJOR:
1014 *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
1015 *minor = ((*minor / 64) * PARTS_PER_DISK) +
1016 EMULATED_HD_DISK_MINOR_OFFSET;
1017 break;
1018 case XEN_IDE1_MAJOR:
1019 *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1020 *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1021 EMULATED_HD_DISK_MINOR_OFFSET;
1022 break;
1023 case XEN_SCSI_DISK0_MAJOR:
1024 *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1025 *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1026 break;
1027 case XEN_SCSI_DISK1_MAJOR:
1028 case XEN_SCSI_DISK2_MAJOR:
1029 case XEN_SCSI_DISK3_MAJOR:
1030 case XEN_SCSI_DISK4_MAJOR:
1031 case XEN_SCSI_DISK5_MAJOR:
1032 case XEN_SCSI_DISK6_MAJOR:
1033 case XEN_SCSI_DISK7_MAJOR:
1034 *offset = (*minor / PARTS_PER_DISK) +
1035 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1036 EMULATED_SD_DISK_NAME_OFFSET;
1037 *minor = *minor +
1038 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1039 EMULATED_SD_DISK_MINOR_OFFSET;
1040 break;
1041 case XEN_SCSI_DISK8_MAJOR:
1042 case XEN_SCSI_DISK9_MAJOR:
1043 case XEN_SCSI_DISK10_MAJOR:
1044 case XEN_SCSI_DISK11_MAJOR:
1045 case XEN_SCSI_DISK12_MAJOR:
1046 case XEN_SCSI_DISK13_MAJOR:
1047 case XEN_SCSI_DISK14_MAJOR:
1048 case XEN_SCSI_DISK15_MAJOR:
1049 *offset = (*minor / PARTS_PER_DISK) +
1050 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1051 EMULATED_SD_DISK_NAME_OFFSET;
1052 *minor = *minor +
1053 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1054 EMULATED_SD_DISK_MINOR_OFFSET;
1055 break;
1056 case XENVBD_MAJOR:
1057 *offset = *minor / PARTS_PER_DISK;
1058 break;
1059 default:
1060 printk(KERN_WARNING "blkfront: your disk configuration is "
1061 "incorrect, please use an xvd device instead\n");
1062 return -ENODEV;
1063 }
1064 return 0;
1065}
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001066
Jan Beuliche77c78c2012-04-05 16:37:22 +01001067static char *encode_disk_name(char *ptr, unsigned int n)
1068{
1069 if (n >= 26)
1070 ptr = encode_disk_name(ptr, n / 26 - 1);
1071 *ptr = 'a' + n % 26;
1072 return ptr + 1;
1073}
1074
Chris Lalancette9246b5f2008-09-17 14:30:32 -07001075static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1076 struct blkfront_info *info,
Stefan Bader7c4d7d72013-05-13 16:28:15 +02001077 u16 vdisk_info, u16 sector_size,
1078 unsigned int physical_sector_size)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001079{
1080 struct gendisk *gd;
1081 int nr_minors = 1;
Stefano Stabellinic80a4202010-12-02 17:55:00 +00001082 int err;
Chris Lalancette9246b5f2008-09-17 14:30:32 -07001083 unsigned int offset;
1084 int minor;
1085 int nr_parts;
Jan Beuliche77c78c2012-04-05 16:37:22 +01001086 char *ptr;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001087
1088 BUG_ON(info->gd != NULL);
1089 BUG_ON(info->rq != NULL);
1090
Chris Lalancette9246b5f2008-09-17 14:30:32 -07001091 if ((info->vdevice>>EXT_SHIFT) > 1) {
1092 /* this is above the extended range; something is wrong */
1093 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1094 return -ENODEV;
1095 }
1096
1097 if (!VDEV_IS_EXTENDED(info->vdevice)) {
Stefano Stabellinic80a4202010-12-02 17:55:00 +00001098 err = xen_translate_vdev(info->vdevice, &minor, &offset);
1099 if (err)
1100 return err;
1101 nr_parts = PARTS_PER_DISK;
Chris Lalancette9246b5f2008-09-17 14:30:32 -07001102 } else {
1103 minor = BLKIF_MINOR_EXT(info->vdevice);
1104 nr_parts = PARTS_PER_EXT_DISK;
Stefano Stabellinic80a4202010-12-02 17:55:00 +00001105 offset = minor / nr_parts;
Stefan Bader89153b52011-07-14 15:30:37 +02001106 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
Stefano Stabellinic80a4202010-12-02 17:55:00 +00001107 printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1108 "emulated IDE disks,\n\t choose an xvd device name"
1109 "from xvde on\n", info->vdevice);
Chris Lalancette9246b5f2008-09-17 14:30:32 -07001110 }
Jan Beuliche77c78c2012-04-05 16:37:22 +01001111 if (minor >> MINORBITS) {
1112 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1113 info->vdevice, minor);
1114 return -ENODEV;
1115 }
Chris Lalancette9246b5f2008-09-17 14:30:32 -07001116
1117 if ((minor % nr_parts) == 0)
1118 nr_minors = nr_parts;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001119
Jan Beulich0e345822010-08-07 18:28:55 +02001120 err = xlbd_reserve_minors(minor, nr_minors);
1121 if (err)
1122 goto out;
1123 err = -ENODEV;
1124
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001125 gd = alloc_disk(nr_minors);
1126 if (gd == NULL)
Jan Beulich0e345822010-08-07 18:28:55 +02001127 goto release;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001128
Jan Beuliche77c78c2012-04-05 16:37:22 +01001129 strcpy(gd->disk_name, DEV_NAME);
1130 ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1131 BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1132 if (nr_minors > 1)
1133 *ptr = 0;
1134 else
1135 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1136 "%d", minor & (nr_parts - 1));
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001137
1138 gd->major = XENVBD_MAJOR;
1139 gd->first_minor = minor;
1140 gd->fops = &xlvbd_block_fops;
1141 gd->private_data = info;
1142 gd->driverfs_dev = &(info->xbdev->dev);
1143 set_capacity(gd, capacity);
1144
Stefan Bader7c4d7d72013-05-13 16:28:15 +02001145 if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size,
Roger Pau Monne402b27f2013-04-18 16:06:54 +02001146 info->max_indirect_segments ? :
1147 BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001148 del_gendisk(gd);
Jan Beulich0e345822010-08-07 18:28:55 +02001149 goto release;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001150 }
1151
1152 info->rq = gd->queue;
1153 info->gd = gd;
1154
Tejun Heo4913efe2010-09-03 11:56:16 +02001155 xlvbd_flush(info);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001156
1157 if (vdisk_info & VDISK_READONLY)
1158 set_disk_ro(gd, 1);
1159
1160 if (vdisk_info & VDISK_REMOVABLE)
1161 gd->flags |= GENHD_FL_REMOVABLE;
1162
1163 if (vdisk_info & VDISK_CDROM)
1164 gd->flags |= GENHD_FL_CD;
1165
1166 return 0;
1167
Jan Beulich0e345822010-08-07 18:28:55 +02001168 release:
1169 xlbd_release_minors(minor, nr_minors);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001170 out:
1171 return err;
1172}
1173
Daniel Stoddena66b5ae2010-08-07 18:33:17 +02001174static void xlvbd_release_gendisk(struct blkfront_info *info)
1175{
Bob Liu3df0e502015-11-14 11:12:12 +08001176 unsigned int minor, nr_minors, i;
Daniel Stoddena66b5ae2010-08-07 18:33:17 +02001177
1178 if (info->rq == NULL)
1179 return;
1180
Daniel Stoddena66b5ae2010-08-07 18:33:17 +02001181 /* No more blkif_request(). */
Bob Liu907c3eb2015-07-13 17:55:24 +08001182 blk_mq_stop_hw_queues(info->rq);
Daniel Stoddena66b5ae2010-08-07 18:33:17 +02001183
Bob Liu3df0e502015-11-14 11:12:12 +08001184 for (i = 0; i < info->nr_rings; i++) {
1185 struct blkfront_ring_info *rinfo = &info->rinfo[i];
Daniel Stoddena66b5ae2010-08-07 18:33:17 +02001186
Bob Liu3df0e502015-11-14 11:12:12 +08001187 /* No more gnttab callback work. */
1188 gnttab_cancel_free_callback(&rinfo->callback);
1189
1190 /* Flush gnttab callback work. Must be done with no locks held. */
1191 flush_work(&rinfo->work);
1192 }
Daniel Stoddena66b5ae2010-08-07 18:33:17 +02001193
1194 del_gendisk(info->gd);
1195
1196 minor = info->gd->first_minor;
1197 nr_minors = info->gd->minors;
1198 xlbd_release_minors(minor, nr_minors);
1199
1200 blk_cleanup_queue(info->rq);
Bob Liu907c3eb2015-07-13 17:55:24 +08001201 blk_mq_free_tag_set(&info->tag_set);
Daniel Stoddena66b5ae2010-08-07 18:33:17 +02001202 info->rq = NULL;
1203
1204 put_disk(info->gd);
1205 info->gd = NULL;
1206}
1207
Bob Liu11659562015-11-14 11:12:13 +08001208/* Already hold rinfo->ring_lock. */
1209static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001210{
Bob Liu81f35162015-11-14 11:12:11 +08001211 if (!RING_FULL(&rinfo->ring))
1212 blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001213}
1214
Bob Liu11659562015-11-14 11:12:13 +08001215static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1216{
1217 unsigned long flags;
1218
1219 spin_lock_irqsave(&rinfo->ring_lock, flags);
1220 kick_pending_request_queues_locked(rinfo);
1221 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1222}
1223
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001224static void blkif_restart_queue(struct work_struct *work)
1225{
Bob Liu81f35162015-11-14 11:12:11 +08001226 struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001227
Bob Liu81f35162015-11-14 11:12:11 +08001228 if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1229 kick_pending_request_queues(rinfo);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001230}
1231
Bob Liu3df0e502015-11-14 11:12:12 +08001232static void blkif_free_ring(struct blkfront_ring_info *rinfo)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001233{
Bob Liu73716df2015-11-16 16:51:39 -05001234 struct grant *persistent_gnt, *n;
Bob Liu3df0e502015-11-14 11:12:12 +08001235 struct blkfront_info *info = rinfo->dev_info;
Roger Pau Monne402b27f2013-04-18 16:06:54 +02001236 int i, j, segs;
Roger Pau Monne0a8704a2012-10-24 18:58:45 +02001237
Roger Pau Monnebfe11d62013-10-29 18:31:14 +01001238 /*
1239 * Remove indirect pages, this only happens when using indirect
1240 * descriptors but not persistent grants
1241 */
Bob Liu81f35162015-11-14 11:12:11 +08001242 if (!list_empty(&rinfo->indirect_pages)) {
Roger Pau Monnebfe11d62013-10-29 18:31:14 +01001243 struct page *indirect_page, *n;
1244
1245 BUG_ON(info->feature_persistent);
Bob Liu81f35162015-11-14 11:12:11 +08001246 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
Roger Pau Monnebfe11d62013-10-29 18:31:14 +01001247 list_del(&indirect_page->lru);
1248 __free_page(indirect_page);
1249 }
1250 }
1251
Bob Liu73716df2015-11-16 16:51:39 -05001252 /* Remove all persistent grants. */
1253 if (!list_empty(&rinfo->grants)) {
1254 list_for_each_entry_safe(persistent_gnt, n,
1255 &rinfo->grants, node) {
1256 list_del(&persistent_gnt->node);
1257 if (persistent_gnt->gref != GRANT_INVALID_REF) {
1258 gnttab_end_foreign_access(persistent_gnt->gref,
1259 0, 0UL);
1260 rinfo->persistent_gnts_c--;
1261 }
1262 if (info->feature_persistent)
1263 __free_page(persistent_gnt->page);
1264 kfree(persistent_gnt);
1265 }
1266 }
1267 BUG_ON(rinfo->persistent_gnts_c != 0);
1268
Bob Liu86839c52015-06-03 13:40:03 +08001269 for (i = 0; i < BLK_RING_SIZE(info); i++) {
Roger Pau Monne402b27f2013-04-18 16:06:54 +02001270 /*
1271 * Clear persistent grants present in requests already
1272 * on the shared ring
1273 */
Bob Liu81f35162015-11-14 11:12:11 +08001274 if (!rinfo->shadow[i].request)
Roger Pau Monne402b27f2013-04-18 16:06:54 +02001275 goto free_shadow;
1276
Bob Liu81f35162015-11-14 11:12:11 +08001277 segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1278 rinfo->shadow[i].req.u.indirect.nr_segments :
1279 rinfo->shadow[i].req.u.rw.nr_segments;
Roger Pau Monne402b27f2013-04-18 16:06:54 +02001280 for (j = 0; j < segs; j++) {
Bob Liu81f35162015-11-14 11:12:11 +08001281 persistent_gnt = rinfo->shadow[i].grants_used[j];
Roger Pau Monne402b27f2013-04-18 16:06:54 +02001282 gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
Roger Pau Monnebfe11d62013-10-29 18:31:14 +01001283 if (info->feature_persistent)
Julien Gralla7a6df22015-06-30 11:58:51 +01001284 __free_page(persistent_gnt->page);
Roger Pau Monne402b27f2013-04-18 16:06:54 +02001285 kfree(persistent_gnt);
1286 }
1287
Bob Liu81f35162015-11-14 11:12:11 +08001288 if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
Roger Pau Monne402b27f2013-04-18 16:06:54 +02001289 /*
1290 * If this is not an indirect operation don't try to
1291 * free indirect segments
1292 */
1293 goto free_shadow;
1294
1295 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
Bob Liu81f35162015-11-14 11:12:11 +08001296 persistent_gnt = rinfo->shadow[i].indirect_grants[j];
Roger Pau Monne402b27f2013-04-18 16:06:54 +02001297 gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
Julien Gralla7a6df22015-06-30 11:58:51 +01001298 __free_page(persistent_gnt->page);
Roger Pau Monne402b27f2013-04-18 16:06:54 +02001299 kfree(persistent_gnt);
1300 }
1301
1302free_shadow:
Bob Liu81f35162015-11-14 11:12:11 +08001303 kfree(rinfo->shadow[i].grants_used);
1304 rinfo->shadow[i].grants_used = NULL;
1305 kfree(rinfo->shadow[i].indirect_grants);
1306 rinfo->shadow[i].indirect_grants = NULL;
1307 kfree(rinfo->shadow[i].sg);
1308 rinfo->shadow[i].sg = NULL;
Roger Pau Monne402b27f2013-04-18 16:06:54 +02001309 }
1310
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001311 /* No more gnttab callback work. */
Bob Liu81f35162015-11-14 11:12:11 +08001312 gnttab_cancel_free_callback(&rinfo->callback);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001313
1314 /* Flush gnttab callback work. Must be done with no locks held. */
Bob Liu81f35162015-11-14 11:12:11 +08001315 flush_work(&rinfo->work);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001316
1317 /* Free resources associated with old device channel. */
Bob Liu86839c52015-06-03 13:40:03 +08001318 for (i = 0; i < info->nr_ring_pages; i++) {
Bob Liu81f35162015-11-14 11:12:11 +08001319 if (rinfo->ring_ref[i] != GRANT_INVALID_REF) {
1320 gnttab_end_foreign_access(rinfo->ring_ref[i], 0, 0);
1321 rinfo->ring_ref[i] = GRANT_INVALID_REF;
Bob Liu86839c52015-06-03 13:40:03 +08001322 }
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001323 }
Bob Liu81f35162015-11-14 11:12:11 +08001324 free_pages((unsigned long)rinfo->ring.sring, get_order(info->nr_ring_pages * PAGE_SIZE));
1325 rinfo->ring.sring = NULL;
Bob Liu86839c52015-06-03 13:40:03 +08001326
Bob Liu81f35162015-11-14 11:12:11 +08001327 if (rinfo->irq)
1328 unbind_from_irqhandler(rinfo->irq, rinfo);
1329 rinfo->evtchn = rinfo->irq = 0;
Bob Liu3df0e502015-11-14 11:12:12 +08001330}
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001331
Bob Liu3df0e502015-11-14 11:12:12 +08001332static void blkif_free(struct blkfront_info *info, int suspend)
1333{
Bob Liu3df0e502015-11-14 11:12:12 +08001334 unsigned int i;
1335
1336 /* Prevent new requests being issued until we fix things up. */
Bob Liu3df0e502015-11-14 11:12:12 +08001337 info->connected = suspend ?
1338 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1339 /* No more blkif_request(). */
1340 if (info->rq)
1341 blk_mq_stop_hw_queues(info->rq);
1342
Bob Liu3df0e502015-11-14 11:12:12 +08001343 for (i = 0; i < info->nr_rings; i++)
1344 blkif_free_ring(&info->rinfo[i]);
1345
1346 kfree(info->rinfo);
1347 info->rinfo = NULL;
1348 info->nr_rings = 0;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001349}
1350
Julien Grallc004a6f2015-07-22 16:44:54 +01001351struct copy_from_grant {
1352 const struct blk_shadow *s;
1353 unsigned int grant_idx;
1354 unsigned int bvec_offset;
1355 char *bvec_data;
1356};
1357
1358static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1359 unsigned int len, void *data)
1360{
1361 struct copy_from_grant *info = data;
1362 char *shared_data;
1363 /* Convenient aliases */
1364 const struct blk_shadow *s = info->s;
1365
1366 shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1367
1368 memcpy(info->bvec_data + info->bvec_offset,
1369 shared_data + offset, len);
1370
1371 info->bvec_offset += len;
1372 info->grant_idx++;
1373
1374 kunmap_atomic(shared_data);
1375}
1376
Julien Grall6cc568332015-08-13 13:13:35 +01001377static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1378{
1379 switch (rsp)
1380 {
1381 case BLKIF_RSP_OKAY:
1382 return REQ_DONE;
1383 case BLKIF_RSP_EOPNOTSUPP:
1384 return REQ_EOPNOTSUPP;
1385 case BLKIF_RSP_ERROR:
1386 /* Fallthrough. */
1387 default:
1388 return REQ_ERROR;
1389 }
1390}
1391
1392/*
1393 * Get the final status of the block request based on two ring response
1394 */
1395static int blkif_get_final_status(enum blk_req_status s1,
1396 enum blk_req_status s2)
1397{
1398 BUG_ON(s1 == REQ_WAITING);
1399 BUG_ON(s2 == REQ_WAITING);
1400
1401 if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1402 return BLKIF_RSP_ERROR;
1403 else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1404 return BLKIF_RSP_EOPNOTSUPP;
1405 return BLKIF_RSP_OKAY;
1406}
1407
1408static bool blkif_completion(unsigned long *id,
1409 struct blkfront_ring_info *rinfo,
Roger Pau Monne0a8704a2012-10-24 18:58:45 +02001410 struct blkif_response *bret)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001411{
Roger Pau Monned62f6912012-12-07 19:00:31 +01001412 int i = 0;
Roger Pau Monneb7649152013-05-02 10:58:50 +02001413 struct scatterlist *sg;
Julien Grallc004a6f2015-07-22 16:44:54 +01001414 int num_sg, num_grant;
Bob Liu81f35162015-11-14 11:12:11 +08001415 struct blkfront_info *info = rinfo->dev_info;
Julien Grall6cc568332015-08-13 13:13:35 +01001416 struct blk_shadow *s = &rinfo->shadow[*id];
Julien Grallc004a6f2015-07-22 16:44:54 +01001417 struct copy_from_grant data = {
Julien Grallc004a6f2015-07-22 16:44:54 +01001418 .grant_idx = 0,
1419 };
Roger Pau Monne402b27f2013-04-18 16:06:54 +02001420
Julien Grallc004a6f2015-07-22 16:44:54 +01001421 num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
Roger Pau Monne402b27f2013-04-18 16:06:54 +02001422 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
Julien Grall6cc568332015-08-13 13:13:35 +01001423
1424 /* The I/O request may be split in two. */
1425 if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1426 struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1427
1428 /* Keep the status of the current response in shadow. */
1429 s->status = blkif_rsp_to_req_status(bret->status);
1430
1431 /* Wait the second response if not yet here. */
1432 if (s2->status == REQ_WAITING)
1433 return 0;
1434
1435 bret->status = blkif_get_final_status(s->status,
1436 s2->status);
1437
1438 /*
1439 * All the grants is stored in the first shadow in order
1440 * to make the completion code simpler.
1441 */
1442 num_grant += s2->req.u.rw.nr_segments;
1443
1444 /*
1445 * The two responses may not come in order. Only the
1446 * first request will store the scatter-gather list.
1447 */
1448 if (s2->num_sg != 0) {
1449 /* Update "id" with the ID of the first response. */
1450 *id = s->associated_id;
1451 s = s2;
1452 }
1453
1454 /*
1455 * We don't need anymore the second request, so recycling
1456 * it now.
1457 */
1458 if (add_id_to_freelist(rinfo, s->associated_id))
1459 WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1460 info->gd->disk_name, s->associated_id);
1461 }
1462
1463 data.s = s;
Julien Grallc004a6f2015-07-22 16:44:54 +01001464 num_sg = s->num_sg;
Roger Pau Monne0a8704a2012-10-24 18:58:45 +02001465
Roger Pau Monnebfe11d62013-10-29 18:31:14 +01001466 if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
Julien Grallc004a6f2015-07-22 16:44:54 +01001467 for_each_sg(s->sg, sg, num_sg, i) {
Roger Pau Monneb7649152013-05-02 10:58:50 +02001468 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
Julien Grallc004a6f2015-07-22 16:44:54 +01001469
1470 data.bvec_offset = sg->offset;
1471 data.bvec_data = kmap_atomic(sg_page(sg));
1472
1473 gnttab_foreach_grant_in_range(sg_page(sg),
1474 sg->offset,
1475 sg->length,
1476 blkif_copy_from_grant,
1477 &data);
1478
1479 kunmap_atomic(data.bvec_data);
Roger Pau Monne0a8704a2012-10-24 18:58:45 +02001480 }
1481 }
1482 /* Add the persistent grant into the list of free grants */
Julien Grallc004a6f2015-07-22 16:44:54 +01001483 for (i = 0; i < num_grant; i++) {
Roger Pau Monnefbe363c2013-08-12 12:53:44 +02001484 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1485 /*
1486 * If the grant is still mapped by the backend (the
1487 * backend has chosen to make this grant persistent)
1488 * we add it at the head of the list, so it will be
1489 * reused first.
1490 */
Roger Pau Monnebfe11d62013-10-29 18:31:14 +01001491 if (!info->feature_persistent)
1492 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1493 s->grants_used[i]->gref);
Bob Liu73716df2015-11-16 16:51:39 -05001494 list_add(&s->grants_used[i]->node, &rinfo->grants);
1495 rinfo->persistent_gnts_c++;
Roger Pau Monnefbe363c2013-08-12 12:53:44 +02001496 } else {
1497 /*
1498 * If the grant is not mapped by the backend we end the
1499 * foreign access and add it to the tail of the list,
1500 * so it will not be picked again unless we run out of
1501 * persistent grants.
1502 */
1503 gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1504 s->grants_used[i]->gref = GRANT_INVALID_REF;
Bob Liu73716df2015-11-16 16:51:39 -05001505 list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
Roger Pau Monnefbe363c2013-08-12 12:53:44 +02001506 }
Roger Pau Monne0a8704a2012-10-24 18:58:45 +02001507 }
Roger Pau Monne402b27f2013-04-18 16:06:54 +02001508 if (s->req.operation == BLKIF_OP_INDIRECT) {
Julien Grallc004a6f2015-07-22 16:44:54 +01001509 for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
Roger Pau Monnefbe363c2013-08-12 12:53:44 +02001510 if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
Roger Pau Monnebfe11d62013-10-29 18:31:14 +01001511 if (!info->feature_persistent)
1512 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1513 s->indirect_grants[i]->gref);
Bob Liu73716df2015-11-16 16:51:39 -05001514 list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1515 rinfo->persistent_gnts_c++;
Roger Pau Monnefbe363c2013-08-12 12:53:44 +02001516 } else {
Roger Pau Monnebfe11d62013-10-29 18:31:14 +01001517 struct page *indirect_page;
1518
Roger Pau Monnefbe363c2013-08-12 12:53:44 +02001519 gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
Roger Pau Monnebfe11d62013-10-29 18:31:14 +01001520 /*
1521 * Add the used indirect page back to the list of
1522 * available pages for indirect grefs.
1523 */
Bob Liu7b076752015-07-22 14:40:09 +08001524 if (!info->feature_persistent) {
Julien Gralla7a6df22015-06-30 11:58:51 +01001525 indirect_page = s->indirect_grants[i]->page;
Bob Liu81f35162015-11-14 11:12:11 +08001526 list_add(&indirect_page->lru, &rinfo->indirect_pages);
Bob Liu7b076752015-07-22 14:40:09 +08001527 }
Roger Pau Monnefbe363c2013-08-12 12:53:44 +02001528 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
Bob Liu73716df2015-11-16 16:51:39 -05001529 list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
Roger Pau Monnefbe363c2013-08-12 12:53:44 +02001530 }
Roger Pau Monne402b27f2013-04-18 16:06:54 +02001531 }
1532 }
Julien Grall6cc568332015-08-13 13:13:35 +01001533
1534 return 1;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001535}
1536
1537static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1538{
1539 struct request *req;
1540 struct blkif_response *bret;
1541 RING_IDX i, rp;
1542 unsigned long flags;
Bob Liu81f35162015-11-14 11:12:11 +08001543 struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1544 struct blkfront_info *info = rinfo->dev_info;
Christoph Hellwigf4829a92015-09-27 21:01:50 +02001545 int error;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001546
Bob Liu11659562015-11-14 11:12:13 +08001547 if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001548 return IRQ_HANDLED;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001549
Bob Liu11659562015-11-14 11:12:13 +08001550 spin_lock_irqsave(&rinfo->ring_lock, flags);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001551 again:
Bob Liu81f35162015-11-14 11:12:11 +08001552 rp = rinfo->ring.sring->rsp_prod;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001553 rmb(); /* Ensure we see queued responses up to 'rp'. */
1554
Bob Liu81f35162015-11-14 11:12:11 +08001555 for (i = rinfo->ring.rsp_cons; i != rp; i++) {
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001556 unsigned long id;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001557
Bob Liu81f35162015-11-14 11:12:11 +08001558 bret = RING_GET_RESPONSE(&rinfo->ring, i);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001559 id = bret->id;
Konrad Rzeszutek Wilk6878c322012-05-25 17:34:51 -04001560 /*
1561 * The backend has messed up and given us an id that we would
1562 * never have given to it (we stamp it up to BLK_RING_SIZE -
1563 * look in get_id_from_freelist.
1564 */
Bob Liu86839c52015-06-03 13:40:03 +08001565 if (id >= BLK_RING_SIZE(info)) {
Konrad Rzeszutek Wilk6878c322012-05-25 17:34:51 -04001566 WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1567 info->gd->disk_name, op_name(bret->operation), id);
1568 /* We can't safely get the 'struct request' as
1569 * the id is busted. */
1570 continue;
1571 }
Bob Liu81f35162015-11-14 11:12:11 +08001572 req = rinfo->shadow[id].request;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001573
Julien Grall6cc568332015-08-13 13:13:35 +01001574 if (bret->operation != BLKIF_OP_DISCARD) {
1575 /*
1576 * We may need to wait for an extra response if the
1577 * I/O request is split in 2
1578 */
1579 if (!blkif_completion(&id, rinfo, bret))
1580 continue;
1581 }
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001582
Bob Liu81f35162015-11-14 11:12:11 +08001583 if (add_id_to_freelist(rinfo, id)) {
Konrad Rzeszutek Wilk6878c322012-05-25 17:34:51 -04001584 WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1585 info->gd->disk_name, op_name(bret->operation), id);
1586 continue;
1587 }
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001588
Christoph Hellwigf4829a92015-09-27 21:01:50 +02001589 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001590 switch (bret->operation) {
Li Dongyanged30bf32011-09-01 18:39:09 +08001591 case BLKIF_OP_DISCARD:
1592 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1593 struct request_queue *rq = info->rq;
Konrad Rzeszutek Wilk6878c322012-05-25 17:34:51 -04001594 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1595 info->gd->disk_name, op_name(bret->operation));
Christoph Hellwigf4829a92015-09-27 21:01:50 +02001596 error = -EOPNOTSUPP;
Li Dongyanged30bf32011-09-01 18:39:09 +08001597 info->feature_discard = 0;
Konrad Rzeszutek Wilk5ea42982011-10-12 16:23:30 -04001598 info->feature_secdiscard = 0;
Li Dongyanged30bf32011-09-01 18:39:09 +08001599 queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
Konrad Rzeszutek Wilk5ea42982011-10-12 16:23:30 -04001600 queue_flag_clear(QUEUE_FLAG_SECDISCARD, rq);
Li Dongyanged30bf32011-09-01 18:39:09 +08001601 }
Christoph Hellwigf4829a92015-09-27 21:01:50 +02001602 blk_mq_complete_request(req, error);
Li Dongyanged30bf32011-09-01 18:39:09 +08001603 break;
Konrad Rzeszutek Wilkedf6ef52011-05-03 12:01:11 -04001604 case BLKIF_OP_FLUSH_DISKCACHE:
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001605 case BLKIF_OP_WRITE_BARRIER:
1606 if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
Konrad Rzeszutek Wilk6878c322012-05-25 17:34:51 -04001607 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1608 info->gd->disk_name, op_name(bret->operation));
Christoph Hellwigf4829a92015-09-27 21:01:50 +02001609 error = -EOPNOTSUPP;
Jeremy Fitzhardingedcb8bae2010-11-02 11:55:58 -04001610 }
1611 if (unlikely(bret->status == BLKIF_RSP_ERROR &&
Bob Liu81f35162015-11-14 11:12:11 +08001612 rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
Konrad Rzeszutek Wilk6878c322012-05-25 17:34:51 -04001613 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1614 info->gd->disk_name, op_name(bret->operation));
Christoph Hellwigf4829a92015-09-27 21:01:50 +02001615 error = -EOPNOTSUPP;
Jeremy Fitzhardingedcb8bae2010-11-02 11:55:58 -04001616 }
Christoph Hellwigf4829a92015-09-27 21:01:50 +02001617 if (unlikely(error)) {
1618 if (error == -EOPNOTSUPP)
1619 error = 0;
Tejun Heo4913efe2010-09-03 11:56:16 +02001620 info->feature_flush = 0;
1621 xlvbd_flush(info);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001622 }
1623 /* fall through */
1624 case BLKIF_OP_READ:
1625 case BLKIF_OP_WRITE:
1626 if (unlikely(bret->status != BLKIF_RSP_OKAY))
1627 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1628 "request: %x\n", bret->status);
1629
Christoph Hellwigf4829a92015-09-27 21:01:50 +02001630 blk_mq_complete_request(req, error);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001631 break;
1632 default:
1633 BUG();
1634 }
1635 }
1636
Bob Liu81f35162015-11-14 11:12:11 +08001637 rinfo->ring.rsp_cons = i;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001638
Bob Liu81f35162015-11-14 11:12:11 +08001639 if (i != rinfo->ring.req_prod_pvt) {
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001640 int more_to_do;
Bob Liu81f35162015-11-14 11:12:11 +08001641 RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001642 if (more_to_do)
1643 goto again;
1644 } else
Bob Liu81f35162015-11-14 11:12:11 +08001645 rinfo->ring.sring->rsp_event = i + 1;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001646
Bob Liu11659562015-11-14 11:12:13 +08001647 kick_pending_request_queues_locked(rinfo);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001648
Bob Liu11659562015-11-14 11:12:13 +08001649 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001650
1651 return IRQ_HANDLED;
1652}
1653
1654
1655static int setup_blkring(struct xenbus_device *dev,
Bob Liu81f35162015-11-14 11:12:11 +08001656 struct blkfront_ring_info *rinfo)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001657{
1658 struct blkif_sring *sring;
Bob Liu86839c52015-06-03 13:40:03 +08001659 int err, i;
Bob Liu81f35162015-11-14 11:12:11 +08001660 struct blkfront_info *info = rinfo->dev_info;
Julien Grallc004a6f2015-07-22 16:44:54 +01001661 unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
Julien Grall9cce2912015-10-13 17:50:11 +01001662 grant_ref_t gref[XENBUS_MAX_RING_GRANTS];
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001663
Bob Liu86839c52015-06-03 13:40:03 +08001664 for (i = 0; i < info->nr_ring_pages; i++)
Bob Liu81f35162015-11-14 11:12:11 +08001665 rinfo->ring_ref[i] = GRANT_INVALID_REF;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001666
Bob Liu86839c52015-06-03 13:40:03 +08001667 sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
1668 get_order(ring_size));
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001669 if (!sring) {
1670 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1671 return -ENOMEM;
1672 }
1673 SHARED_RING_INIT(sring);
Bob Liu81f35162015-11-14 11:12:11 +08001674 FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001675
Bob Liu81f35162015-11-14 11:12:11 +08001676 err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001677 if (err < 0) {
Bob Liu86839c52015-06-03 13:40:03 +08001678 free_pages((unsigned long)sring, get_order(ring_size));
Bob Liu81f35162015-11-14 11:12:11 +08001679 rinfo->ring.sring = NULL;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001680 goto fail;
1681 }
Bob Liu86839c52015-06-03 13:40:03 +08001682 for (i = 0; i < info->nr_ring_pages; i++)
Bob Liu81f35162015-11-14 11:12:11 +08001683 rinfo->ring_ref[i] = gref[i];
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001684
Bob Liu81f35162015-11-14 11:12:11 +08001685 err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001686 if (err)
1687 goto fail;
1688
Bob Liu81f35162015-11-14 11:12:11 +08001689 err = bind_evtchn_to_irqhandler(rinfo->evtchn, blkif_interrupt, 0,
1690 "blkif", rinfo);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001691 if (err <= 0) {
1692 xenbus_dev_fatal(dev, err,
1693 "bind_evtchn_to_irqhandler failed");
1694 goto fail;
1695 }
Bob Liu81f35162015-11-14 11:12:11 +08001696 rinfo->irq = err;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001697
1698 return 0;
1699fail:
1700 blkif_free(info, 0);
1701 return err;
1702}
1703
Bob Liu28d949b2015-11-14 11:12:14 +08001704/*
1705 * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1706 * ring buffer may have multi pages depending on ->nr_ring_pages.
1707 */
1708static int write_per_ring_nodes(struct xenbus_transaction xbt,
1709 struct blkfront_ring_info *rinfo, const char *dir)
1710{
1711 int err;
1712 unsigned int i;
1713 const char *message = NULL;
1714 struct blkfront_info *info = rinfo->dev_info;
1715
1716 if (info->nr_ring_pages == 1) {
1717 err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1718 if (err) {
1719 message = "writing ring-ref";
1720 goto abort_transaction;
1721 }
1722 } else {
1723 for (i = 0; i < info->nr_ring_pages; i++) {
1724 char ring_ref_name[RINGREF_NAME_LEN];
1725
1726 snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1727 err = xenbus_printf(xbt, dir, ring_ref_name,
1728 "%u", rinfo->ring_ref[i]);
1729 if (err) {
1730 message = "writing ring-ref";
1731 goto abort_transaction;
1732 }
1733 }
1734 }
1735
1736 err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1737 if (err) {
1738 message = "writing event-channel";
1739 goto abort_transaction;
1740 }
1741
1742 return 0;
1743
1744abort_transaction:
1745 xenbus_transaction_end(xbt, 1);
1746 if (message)
1747 xenbus_dev_fatal(info->xbdev, err, "%s", message);
1748
1749 return err;
1750}
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001751
1752/* Common code used when first setting up, and when resuming. */
Ian Campbell203fd612009-12-04 15:33:54 +00001753static int talk_to_blkback(struct xenbus_device *dev,
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001754 struct blkfront_info *info)
1755{
1756 const char *message = NULL;
1757 struct xenbus_transaction xbt;
Bob Liu28d949b2015-11-14 11:12:14 +08001758 int err;
1759 unsigned int i, max_page_order = 0;
Bob Liu86839c52015-06-03 13:40:03 +08001760 unsigned int ring_page_order = 0;
1761
1762 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1763 "max-ring-page-order", "%u", &max_page_order);
1764 if (err != 1)
1765 info->nr_ring_pages = 1;
1766 else {
1767 ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1768 info->nr_ring_pages = 1 << ring_page_order;
1769 }
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001770
Bob Liu3df0e502015-11-14 11:12:12 +08001771 for (i = 0; i < info->nr_rings; i++) {
Bob Liu28d949b2015-11-14 11:12:14 +08001772 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1773
Bob Liu3df0e502015-11-14 11:12:12 +08001774 /* Create shared ring, alloc event channel. */
1775 err = setup_blkring(dev, rinfo);
1776 if (err)
1777 goto destroy_blkring;
1778 }
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001779
1780again:
1781 err = xenbus_transaction_start(&xbt);
1782 if (err) {
1783 xenbus_dev_fatal(dev, err, "starting transaction");
1784 goto destroy_blkring;
1785 }
1786
Bob Liu28d949b2015-11-14 11:12:14 +08001787 if (info->nr_ring_pages > 1) {
1788 err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1789 ring_page_order);
Bob Liu3df0e502015-11-14 11:12:12 +08001790 if (err) {
Bob Liu28d949b2015-11-14 11:12:14 +08001791 message = "writing ring-page-order";
Bob Liu3df0e502015-11-14 11:12:12 +08001792 goto abort_transaction;
1793 }
Bob Liu28d949b2015-11-14 11:12:14 +08001794 }
1795
1796 /* We already got the number of queues/rings in _probe */
1797 if (info->nr_rings == 1) {
1798 err = write_per_ring_nodes(xbt, &info->rinfo[0], dev->nodename);
1799 if (err)
1800 goto destroy_blkring;
Bob Liu3df0e502015-11-14 11:12:12 +08001801 } else {
Bob Liu28d949b2015-11-14 11:12:14 +08001802 char *path;
1803 size_t pathsize;
1804
1805 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1806 info->nr_rings);
1807 if (err) {
1808 message = "writing multi-queue-num-queues";
1809 goto abort_transaction;
1810 }
1811
1812 pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1813 path = kmalloc(pathsize, GFP_KERNEL);
1814 if (!path) {
1815 err = -ENOMEM;
1816 message = "ENOMEM while writing ring references";
1817 goto abort_transaction;
1818 }
1819
1820 for (i = 0; i < info->nr_rings; i++) {
1821 memset(path, 0, pathsize);
1822 snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1823 err = write_per_ring_nodes(xbt, &info->rinfo[i], path);
1824 if (err) {
1825 kfree(path);
1826 goto destroy_blkring;
1827 }
1828 }
1829 kfree(path);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001830 }
Markus Armbruster3e334232008-04-02 10:54:02 -07001831 err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1832 XEN_IO_PROTO_ABI_NATIVE);
1833 if (err) {
1834 message = "writing protocol";
1835 goto abort_transaction;
1836 }
Roger Pau Monne0a8704a2012-10-24 18:58:45 +02001837 err = xenbus_printf(xbt, dev->nodename,
Roger Pau Monnecb5bd4d2012-11-02 16:43:04 +01001838 "feature-persistent", "%u", 1);
Roger Pau Monne0a8704a2012-10-24 18:58:45 +02001839 if (err)
1840 dev_warn(&dev->dev,
1841 "writing persistent grants feature to xenbus");
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001842
1843 err = xenbus_transaction_end(xbt, 0);
1844 if (err) {
1845 if (err == -EAGAIN)
1846 goto again;
1847 xenbus_dev_fatal(dev, err, "completing transaction");
1848 goto destroy_blkring;
1849 }
1850
Bob Liu3df0e502015-11-14 11:12:12 +08001851 for (i = 0; i < info->nr_rings; i++) {
1852 unsigned int j;
Bob Liu28d949b2015-11-14 11:12:14 +08001853 struct blkfront_ring_info *rinfo = &info->rinfo[i];
Bob Liu3df0e502015-11-14 11:12:12 +08001854
1855 for (j = 0; j < BLK_RING_SIZE(info); j++)
1856 rinfo->shadow[j].req.u.rw.id = j + 1;
1857 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1858 }
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001859 xenbus_switch_state(dev, XenbusStateInitialised);
1860
1861 return 0;
1862
1863 abort_transaction:
1864 xenbus_transaction_end(xbt, 1);
1865 if (message)
1866 xenbus_dev_fatal(dev, err, "%s", message);
1867 destroy_blkring:
1868 blkif_free(info, 0);
Bob Liu3df0e502015-11-14 11:12:12 +08001869
Konrad Rzeszutek Wilkc31ecf62015-12-18 16:28:53 -05001870 kfree(info);
1871 dev_set_drvdata(&dev->dev, NULL);
1872
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001873 return err;
1874}
1875
Bob Liu3db70a82015-11-25 17:52:55 -05001876static int negotiate_mq(struct blkfront_info *info)
1877{
1878 unsigned int backend_max_queues = 0;
1879 int err;
1880 unsigned int i;
1881
1882 BUG_ON(info->nr_rings);
1883
1884 /* Check if backend supports multiple queues. */
1885 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1886 "multi-queue-max-queues", "%u", &backend_max_queues);
1887 if (err < 0)
1888 backend_max_queues = 1;
1889
1890 info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1891 /* We need at least one ring. */
1892 if (!info->nr_rings)
1893 info->nr_rings = 1;
1894
1895 info->rinfo = kzalloc(sizeof(struct blkfront_ring_info) * info->nr_rings, GFP_KERNEL);
1896 if (!info->rinfo) {
1897 xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1898 return -ENOMEM;
1899 }
1900
1901 for (i = 0; i < info->nr_rings; i++) {
1902 struct blkfront_ring_info *rinfo;
1903
1904 rinfo = &info->rinfo[i];
1905 INIT_LIST_HEAD(&rinfo->indirect_pages);
1906 INIT_LIST_HEAD(&rinfo->grants);
1907 rinfo->dev_info = info;
1908 INIT_WORK(&rinfo->work, blkif_restart_queue);
1909 spin_lock_init(&rinfo->ring_lock);
1910 }
1911 return 0;
1912}
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001913/**
1914 * Entry point to this code when a new device is created. Allocate the basic
1915 * structures and the ring buffer for communication with the backend, and
1916 * inform the backend of the appropriate details for those. Switch to
1917 * Initialised state.
1918 */
1919static int blkfront_probe(struct xenbus_device *dev,
1920 const struct xenbus_device_id *id)
1921{
Bob Liu86839c52015-06-03 13:40:03 +08001922 int err, vdevice;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001923 struct blkfront_info *info;
1924
1925 /* FIXME: Use dynamic device id if this is not set. */
1926 err = xenbus_scanf(XBT_NIL, dev->nodename,
1927 "virtual-device", "%i", &vdevice);
1928 if (err != 1) {
Chris Lalancette9246b5f2008-09-17 14:30:32 -07001929 /* go looking in the extended area instead */
1930 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1931 "%i", &vdevice);
1932 if (err != 1) {
1933 xenbus_dev_fatal(dev, err, "reading virtual-device");
1934 return err;
1935 }
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001936 }
1937
Stefano Stabellinib98a4092010-07-29 14:53:16 +01001938 if (xen_hvm_domain()) {
1939 char *type;
1940 int len;
1941 /* no unplug has been done: do not hook devices != xen vbds */
Konrad Rzeszutek Wilk51c71a32013-11-26 15:05:40 -05001942 if (xen_has_pv_and_legacy_disk_devices()) {
Stefano Stabellinib98a4092010-07-29 14:53:16 +01001943 int major;
Stefano Stabellinic1c54132010-05-14 12:44:30 +01001944
Stefano Stabellinib98a4092010-07-29 14:53:16 +01001945 if (!VDEV_IS_EXTENDED(vdevice))
1946 major = BLKIF_MAJOR(vdevice);
1947 else
1948 major = XENVBD_MAJOR;
Stefano Stabellinic1c54132010-05-14 12:44:30 +01001949
Stefano Stabellinib98a4092010-07-29 14:53:16 +01001950 if (major != XENVBD_MAJOR) {
1951 printk(KERN_INFO
1952 "%s: HVM does not support vbd %d as xen block device\n",
Rasmus Villemoes02f1f212015-02-12 15:01:31 -08001953 __func__, vdevice);
Stefano Stabellinib98a4092010-07-29 14:53:16 +01001954 return -ENODEV;
1955 }
1956 }
1957 /* do not create a PV cdrom device if we are an HVM guest */
1958 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1959 if (IS_ERR(type))
1960 return -ENODEV;
1961 if (strncmp(type, "cdrom", 5) == 0) {
1962 kfree(type);
Stefano Stabellinic1c54132010-05-14 12:44:30 +01001963 return -ENODEV;
1964 }
Stefano Stabellinib98a4092010-07-29 14:53:16 +01001965 kfree(type);
Stefano Stabellinic1c54132010-05-14 12:44:30 +01001966 }
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001967 info = kzalloc(sizeof(*info), GFP_KERNEL);
1968 if (!info) {
1969 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1970 return -ENOMEM;
1971 }
1972
Bob Liu28d949b2015-11-14 11:12:14 +08001973 info->xbdev = dev;
Bob Liu3db70a82015-11-25 17:52:55 -05001974 err = negotiate_mq(info);
1975 if (err) {
Bob Liu3df0e502015-11-14 11:12:12 +08001976 kfree(info);
Bob Liu3db70a82015-11-25 17:52:55 -05001977 return err;
Bob Liu3df0e502015-11-14 11:12:12 +08001978 }
Bob Liu81f35162015-11-14 11:12:11 +08001979
Daniel Stoddenb70f5fa2010-04-30 22:01:19 +00001980 mutex_init(&info->mutex);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001981 info->vdevice = vdevice;
1982 info->connected = BLKIF_STATE_DISCONNECTED;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001983
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001984 /* Front end dir is a number, which is used as the id. */
1985 info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
Greg Kroah-Hartmana1b4b122009-04-30 14:43:31 -07001986 dev_set_drvdata(&dev->dev, info);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001987
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07001988 return 0;
1989}
1990
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001991static void split_bio_end(struct bio *bio)
Roger Pau Monne402b27f2013-04-18 16:06:54 +02001992{
1993 struct split_bio *split_bio = bio->bi_private;
1994
Roger Pau Monne402b27f2013-04-18 16:06:54 +02001995 if (atomic_dec_and_test(&split_bio->pending)) {
1996 split_bio->bio->bi_phys_segments = 0;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001997 split_bio->bio->bi_error = bio->bi_error;
1998 bio_endio(split_bio->bio);
Roger Pau Monne402b27f2013-04-18 16:06:54 +02001999 kfree(split_bio);
2000 }
2001 bio_put(bio);
2002}
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002003
2004static int blkif_recover(struct blkfront_info *info)
2005{
Bob Liu3df0e502015-11-14 11:12:12 +08002006 unsigned int i, r_index;
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002007 struct request *req, *n;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002008 struct blk_shadow *copy;
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002009 int rc;
2010 struct bio *bio, *cloned_bio;
2011 struct bio_list bio_list, merge_bio;
2012 unsigned int segs, offset;
2013 int pending, size;
2014 struct split_bio *split_bio;
2015 struct list_head requests;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002016
Bob Liu3df0e502015-11-14 11:12:12 +08002017 blkfront_gather_backend_features(info);
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002018 segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
2019 blk_queue_max_segments(info->rq, segs);
2020 bio_list_init(&bio_list);
2021 INIT_LIST_HEAD(&requests);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002022
Bob Liu3df0e502015-11-14 11:12:12 +08002023 for (r_index = 0; r_index < info->nr_rings; r_index++) {
2024 struct blkfront_ring_info *rinfo;
2025
2026 rinfo = &info->rinfo[r_index];
2027 /* Stage 1: Make a safe copy of the shadow state. */
2028 copy = kmemdup(rinfo->shadow, sizeof(rinfo->shadow),
2029 GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
2030 if (!copy)
2031 return -ENOMEM;
2032
2033 /* Stage 2: Set up free list. */
2034 memset(&rinfo->shadow, 0, sizeof(rinfo->shadow));
2035 for (i = 0; i < BLK_RING_SIZE(info); i++)
2036 rinfo->shadow[i].req.u.rw.id = i+1;
2037 rinfo->shadow_free = rinfo->ring.req_prod_pvt;
2038 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
2039
2040 rc = blkfront_setup_indirect(rinfo);
2041 if (rc) {
2042 kfree(copy);
2043 return rc;
Konrad Rzeszutek Wilk5ea42982011-10-12 16:23:30 -04002044 }
Bob Liu3df0e502015-11-14 11:12:12 +08002045
2046 for (i = 0; i < BLK_RING_SIZE(info); i++) {
2047 /* Not in use? */
2048 if (!copy[i].request)
2049 continue;
2050
2051 /*
2052 * Get the bios in the request so we can re-queue them.
2053 */
2054 if (copy[i].request->cmd_flags &
2055 (REQ_FLUSH | REQ_FUA | REQ_DISCARD | REQ_SECURE)) {
2056 /*
2057 * Flush operations don't contain bios, so
2058 * we need to requeue the whole request
2059 */
2060 list_add(&copy[i].request->queuelist, &requests);
2061 continue;
2062 }
2063 merge_bio.head = copy[i].request->bio;
2064 merge_bio.tail = copy[i].request->biotail;
2065 bio_list_merge(&bio_list, &merge_bio);
2066 copy[i].request->bio = NULL;
2067 blk_end_request_all(copy[i].request, 0);
2068 }
2069
2070 kfree(copy);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002071 }
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002072 xenbus_switch_state(info->xbdev, XenbusStateConnected);
2073
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002074 /* Now safe for us to use the shared ring */
2075 info->connected = BLKIF_STATE_CONNECTED;
2076
Bob Liu3df0e502015-11-14 11:12:12 +08002077 for (r_index = 0; r_index < info->nr_rings; r_index++) {
2078 struct blkfront_ring_info *rinfo;
2079
2080 rinfo = &info->rinfo[r_index];
2081 /* Kick any other new requests queued since we resumed */
2082 kick_pending_request_queues(rinfo);
2083 }
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002084
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002085 list_for_each_entry_safe(req, n, &requests, queuelist) {
2086 /* Requeue pending requests (flush or discard) */
2087 list_del_init(&req->queuelist);
2088 BUG_ON(req->nr_phys_segments > segs);
Bob Liu907c3eb2015-07-13 17:55:24 +08002089 blk_mq_requeue_request(req);
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002090 }
Bob Liu907c3eb2015-07-13 17:55:24 +08002091 blk_mq_kick_requeue_list(info->rq);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002092
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002093 while ((bio = bio_list_pop(&bio_list)) != NULL) {
2094 /* Traverse the list of pending bios and re-queue them */
2095 if (bio_segments(bio) > segs) {
2096 /*
2097 * This bio has more segments than what we can
2098 * handle, we have to split it.
2099 */
2100 pending = (bio_segments(bio) + segs - 1) / segs;
2101 split_bio = kzalloc(sizeof(*split_bio), GFP_NOIO);
2102 BUG_ON(split_bio == NULL);
2103 atomic_set(&split_bio->pending, pending);
2104 split_bio->bio = bio;
2105 for (i = 0; i < pending; i++) {
Julien Grallc004a6f2015-07-22 16:44:54 +01002106 offset = (i * segs * XEN_PAGE_SIZE) >> 9;
2107 size = min((unsigned int)(segs * XEN_PAGE_SIZE) >> 9,
Kent Overstreet4f024f32013-10-11 15:44:27 -07002108 (unsigned int)bio_sectors(bio) - offset);
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002109 cloned_bio = bio_clone(bio, GFP_NOIO);
2110 BUG_ON(cloned_bio == NULL);
Kent Overstreet6678d832013-08-07 11:14:32 -07002111 bio_trim(cloned_bio, offset, size);
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002112 cloned_bio->bi_private = split_bio;
2113 cloned_bio->bi_end_io = split_bio_end;
2114 submit_bio(cloned_bio->bi_rw, cloned_bio);
2115 }
2116 /*
2117 * Now we have to wait for all those smaller bios to
2118 * end, so we can also end the "parent" bio.
2119 */
2120 continue;
2121 }
2122 /* We don't need to split this bio */
2123 submit_bio(bio->bi_rw, bio);
2124 }
2125
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002126 return 0;
2127}
2128
2129/**
2130 * We are reconnecting to the backend, due to a suspend/resume, or a backend
2131 * driver restart. We tear down our blkif structure and recreate it, but
2132 * leave the device-layer structures intact so that this is transparent to the
2133 * rest of the kernel.
2134 */
2135static int blkfront_resume(struct xenbus_device *dev)
2136{
Greg Kroah-Hartmana1b4b122009-04-30 14:43:31 -07002137 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
Bob Liu3db70a82015-11-25 17:52:55 -05002138 int err = 0;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002139
2140 dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2141
2142 blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2143
Bob Liu3db70a82015-11-25 17:52:55 -05002144 err = negotiate_mq(info);
2145 if (err)
2146 return err;
2147
Ian Campbell203fd612009-12-04 15:33:54 +00002148 err = talk_to_blkback(dev, info);
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002149
2150 /*
2151 * We have to wait for the backend to switch to
2152 * connected state, since we want to read which
2153 * features it supports.
2154 */
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002155
2156 return err;
2157}
2158
Konrad Rzeszutek Wilk6f03a7f2015-11-16 15:14:41 -05002159static void blkfront_closing(struct blkfront_info *info)
Daniel Stoddenb70f5fa2010-04-30 22:01:19 +00002160{
2161 struct xenbus_device *xbdev = info->xbdev;
2162 struct block_device *bdev = NULL;
2163
2164 mutex_lock(&info->mutex);
2165
2166 if (xbdev->state == XenbusStateClosing) {
2167 mutex_unlock(&info->mutex);
2168 return;
2169 }
2170
2171 if (info->gd)
2172 bdev = bdget_disk(info->gd, 0);
2173
2174 mutex_unlock(&info->mutex);
2175
2176 if (!bdev) {
2177 xenbus_frontend_closed(xbdev);
2178 return;
2179 }
2180
2181 mutex_lock(&bdev->bd_mutex);
2182
Daniel Stodden7b32d102010-04-30 22:01:23 +00002183 if (bdev->bd_openers) {
Daniel Stoddenb70f5fa2010-04-30 22:01:19 +00002184 xenbus_dev_error(xbdev, -EBUSY,
2185 "Device in use; refusing to close");
2186 xenbus_switch_state(xbdev, XenbusStateClosing);
2187 } else {
2188 xlvbd_release_gendisk(info);
2189 xenbus_frontend_closed(xbdev);
2190 }
2191
2192 mutex_unlock(&bdev->bd_mutex);
2193 bdput(bdev);
2194}
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002195
Li Dongyanged30bf32011-09-01 18:39:09 +08002196static void blkfront_setup_discard(struct blkfront_info *info)
2197{
2198 int err;
Li Dongyanged30bf32011-09-01 18:39:09 +08002199 unsigned int discard_granularity;
2200 unsigned int discard_alignment;
Konrad Rzeszutek Wilk5ea42982011-10-12 16:23:30 -04002201 unsigned int discard_secure;
Li Dongyanged30bf32011-09-01 18:39:09 +08002202
Olaf Hering1c8cad62014-05-21 16:32:40 +02002203 info->feature_discard = 1;
2204 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2205 "discard-granularity", "%u", &discard_granularity,
2206 "discard-alignment", "%u", &discard_alignment,
2207 NULL);
2208 if (!err) {
2209 info->discard_granularity = discard_granularity;
2210 info->discard_alignment = discard_alignment;
2211 }
2212 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2213 "discard-secure", "%d", &discard_secure,
2214 NULL);
2215 if (!err)
2216 info->feature_secdiscard = !!discard_secure;
Li Dongyanged30bf32011-09-01 18:39:09 +08002217}
2218
Bob Liu81f35162015-11-14 11:12:11 +08002219static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002220{
Julien Grallc004a6f2015-07-22 16:44:54 +01002221 unsigned int psegs, grants;
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002222 int err, i;
Bob Liu81f35162015-11-14 11:12:11 +08002223 struct blkfront_info *info = rinfo->dev_info;
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002224
Julien Grall6cc568332015-08-13 13:13:35 +01002225 if (info->max_indirect_segments == 0) {
2226 if (!HAS_EXTRA_REQ)
2227 grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2228 else {
2229 /*
2230 * When an extra req is required, the maximum
2231 * grants supported is related to the size of the
2232 * Linux block segment.
2233 */
2234 grants = GRANTS_PER_PSEG;
2235 }
2236 }
Bob Liud50babb2015-07-22 14:40:08 +08002237 else
Julien Grallc004a6f2015-07-22 16:44:54 +01002238 grants = info->max_indirect_segments;
2239 psegs = grants / GRANTS_PER_PSEG;
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002240
Bob Liu81f35162015-11-14 11:12:11 +08002241 err = fill_grant_buffer(rinfo,
Julien Grallc004a6f2015-07-22 16:44:54 +01002242 (grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002243 if (err)
2244 goto out_of_memory;
2245
Roger Pau Monnebfe11d62013-10-29 18:31:14 +01002246 if (!info->feature_persistent && info->max_indirect_segments) {
2247 /*
2248 * We are using indirect descriptors but not persistent
2249 * grants, we need to allocate a set of pages that can be
2250 * used for mapping indirect grefs
2251 */
Julien Grallc004a6f2015-07-22 16:44:54 +01002252 int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
Roger Pau Monnebfe11d62013-10-29 18:31:14 +01002253
Bob Liu81f35162015-11-14 11:12:11 +08002254 BUG_ON(!list_empty(&rinfo->indirect_pages));
Roger Pau Monnebfe11d62013-10-29 18:31:14 +01002255 for (i = 0; i < num; i++) {
2256 struct page *indirect_page = alloc_page(GFP_NOIO);
2257 if (!indirect_page)
2258 goto out_of_memory;
Bob Liu81f35162015-11-14 11:12:11 +08002259 list_add(&indirect_page->lru, &rinfo->indirect_pages);
Roger Pau Monnebfe11d62013-10-29 18:31:14 +01002260 }
2261 }
2262
Bob Liu86839c52015-06-03 13:40:03 +08002263 for (i = 0; i < BLK_RING_SIZE(info); i++) {
Bob Liu81f35162015-11-14 11:12:11 +08002264 rinfo->shadow[i].grants_used = kzalloc(
2265 sizeof(rinfo->shadow[i].grants_used[0]) * grants,
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002266 GFP_NOIO);
Bob Liu81f35162015-11-14 11:12:11 +08002267 rinfo->shadow[i].sg = kzalloc(sizeof(rinfo->shadow[i].sg[0]) * psegs, GFP_NOIO);
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002268 if (info->max_indirect_segments)
Bob Liu81f35162015-11-14 11:12:11 +08002269 rinfo->shadow[i].indirect_grants = kzalloc(
2270 sizeof(rinfo->shadow[i].indirect_grants[0]) *
Julien Grallc004a6f2015-07-22 16:44:54 +01002271 INDIRECT_GREFS(grants),
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002272 GFP_NOIO);
Bob Liu81f35162015-11-14 11:12:11 +08002273 if ((rinfo->shadow[i].grants_used == NULL) ||
2274 (rinfo->shadow[i].sg == NULL) ||
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002275 (info->max_indirect_segments &&
Bob Liu81f35162015-11-14 11:12:11 +08002276 (rinfo->shadow[i].indirect_grants == NULL)))
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002277 goto out_of_memory;
Bob Liu81f35162015-11-14 11:12:11 +08002278 sg_init_table(rinfo->shadow[i].sg, psegs);
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002279 }
2280
2281
2282 return 0;
2283
2284out_of_memory:
Bob Liu86839c52015-06-03 13:40:03 +08002285 for (i = 0; i < BLK_RING_SIZE(info); i++) {
Bob Liu81f35162015-11-14 11:12:11 +08002286 kfree(rinfo->shadow[i].grants_used);
2287 rinfo->shadow[i].grants_used = NULL;
2288 kfree(rinfo->shadow[i].sg);
2289 rinfo->shadow[i].sg = NULL;
2290 kfree(rinfo->shadow[i].indirect_grants);
2291 rinfo->shadow[i].indirect_grants = NULL;
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002292 }
Bob Liu81f35162015-11-14 11:12:11 +08002293 if (!list_empty(&rinfo->indirect_pages)) {
Roger Pau Monnebfe11d62013-10-29 18:31:14 +01002294 struct page *indirect_page, *n;
Bob Liu81f35162015-11-14 11:12:11 +08002295 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
Roger Pau Monnebfe11d62013-10-29 18:31:14 +01002296 list_del(&indirect_page->lru);
2297 __free_page(indirect_page);
2298 }
2299 }
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002300 return -ENOMEM;
2301}
2302
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002303/*
Bob Liud50babb2015-07-22 14:40:08 +08002304 * Gather all backend feature-*
2305 */
Bob Liu3df0e502015-11-14 11:12:12 +08002306static void blkfront_gather_backend_features(struct blkfront_info *info)
Bob Liud50babb2015-07-22 14:40:08 +08002307{
2308 int err;
2309 int barrier, flush, discard, persistent;
2310 unsigned int indirect_segments;
2311
2312 info->feature_flush = 0;
2313
2314 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2315 "feature-barrier", "%d", &barrier,
2316 NULL);
2317
2318 /*
2319 * If there's no "feature-barrier" defined, then it means
2320 * we're dealing with a very old backend which writes
2321 * synchronously; nothing to do.
2322 *
2323 * If there are barriers, then we use flush.
2324 */
2325 if (!err && barrier)
2326 info->feature_flush = REQ_FLUSH | REQ_FUA;
2327 /*
2328 * And if there is "feature-flush-cache" use that above
2329 * barriers.
2330 */
2331 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2332 "feature-flush-cache", "%d", &flush,
2333 NULL);
2334
2335 if (!err && flush)
2336 info->feature_flush = REQ_FLUSH;
2337
2338 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2339 "feature-discard", "%d", &discard,
2340 NULL);
2341
2342 if (!err && discard)
2343 blkfront_setup_discard(info);
2344
2345 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2346 "feature-persistent", "%u", &persistent,
2347 NULL);
2348 if (err)
2349 info->feature_persistent = 0;
2350 else
2351 info->feature_persistent = persistent;
2352
2353 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2354 "feature-max-indirect-segments", "%u", &indirect_segments,
2355 NULL);
2356 if (err)
2357 info->max_indirect_segments = 0;
2358 else
2359 info->max_indirect_segments = min(indirect_segments,
2360 xen_blkif_max_segments);
Bob Liud50babb2015-07-22 14:40:08 +08002361}
2362
2363/*
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002364 * Invoked when the backend is finally 'ready' (and has told produced
2365 * the details about the physical device - #sectors, size, etc).
2366 */
2367static void blkfront_connect(struct blkfront_info *info)
2368{
2369 unsigned long long sectors;
2370 unsigned long sector_size;
Stefan Bader7c4d7d72013-05-13 16:28:15 +02002371 unsigned int physical_sector_size;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002372 unsigned int binfo;
Bob Liu3df0e502015-11-14 11:12:12 +08002373 int err, i;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002374
K. Y. Srinivasan1fa73be2010-03-11 13:42:26 -08002375 switch (info->connected) {
2376 case BLKIF_STATE_CONNECTED:
2377 /*
2378 * Potentially, the back-end may be signalling
2379 * a capacity change; update the capacity.
2380 */
2381 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2382 "sectors", "%Lu", &sectors);
2383 if (XENBUS_EXIST_ERR(err))
2384 return;
2385 printk(KERN_INFO "Setting capacity to %Lu\n",
2386 sectors);
2387 set_capacity(info->gd, sectors);
K. Y. Srinivasan2def1412010-03-18 15:00:54 -07002388 revalidate_disk(info->gd);
K. Y. Srinivasan1fa73be2010-03-11 13:42:26 -08002389
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002390 return;
K. Y. Srinivasan1fa73be2010-03-11 13:42:26 -08002391 case BLKIF_STATE_SUSPENDED:
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002392 /*
2393 * If we are recovering from suspension, we need to wait
2394 * for the backend to announce it's features before
2395 * reconnecting, at least we need to know if the backend
2396 * supports indirect descriptors, and how many.
2397 */
2398 blkif_recover(info);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002399 return;
2400
Jeremy Fitzhardingeb4dddb42010-03-11 15:10:40 -08002401 default:
2402 break;
K. Y. Srinivasan1fa73be2010-03-11 13:42:26 -08002403 }
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002404
2405 dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2406 __func__, info->xbdev->otherend);
2407
2408 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2409 "sectors", "%llu", &sectors,
2410 "info", "%u", &binfo,
2411 "sector-size", "%lu", &sector_size,
2412 NULL);
2413 if (err) {
2414 xenbus_dev_fatal(info->xbdev, err,
2415 "reading backend fields at %s",
2416 info->xbdev->otherend);
2417 return;
2418 }
2419
Stefan Bader7c4d7d72013-05-13 16:28:15 +02002420 /*
2421 * physcial-sector-size is a newer field, so old backends may not
2422 * provide this. Assume physical sector size to be the same as
2423 * sector_size in that case.
2424 */
2425 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2426 "physical-sector-size", "%u", &physical_sector_size);
2427 if (err != 1)
2428 physical_sector_size = sector_size;
2429
Bob Liu3df0e502015-11-14 11:12:12 +08002430 blkfront_gather_backend_features(info);
2431 for (i = 0; i < info->nr_rings; i++) {
2432 err = blkfront_setup_indirect(&info->rinfo[i]);
2433 if (err) {
2434 xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2435 info->xbdev->otherend);
2436 blkif_free(info, 0);
2437 break;
2438 }
Roger Pau Monne402b27f2013-04-18 16:06:54 +02002439 }
2440
Stefan Bader7c4d7d72013-05-13 16:28:15 +02002441 err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
2442 physical_sector_size);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002443 if (err) {
2444 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2445 info->xbdev->otherend);
2446 return;
2447 }
2448
2449 xenbus_switch_state(info->xbdev, XenbusStateConnected);
2450
2451 /* Kick pending requests. */
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002452 info->connected = BLKIF_STATE_CONNECTED;
Bob Liu3df0e502015-11-14 11:12:12 +08002453 for (i = 0; i < info->nr_rings; i++)
2454 kick_pending_request_queues(&info->rinfo[i]);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002455
2456 add_disk(info->gd);
Christian Limpach1d78d702008-04-02 10:54:04 -07002457
2458 info->is_ready = 1;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002459}
2460
2461/**
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002462 * Callback received when the backend's state changes.
2463 */
Ian Campbell203fd612009-12-04 15:33:54 +00002464static void blkback_changed(struct xenbus_device *dev,
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002465 enum xenbus_state backend_state)
2466{
Greg Kroah-Hartmana1b4b122009-04-30 14:43:31 -07002467 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002468
Ian Campbell203fd612009-12-04 15:33:54 +00002469 dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002470
2471 switch (backend_state) {
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002472 case XenbusStateInitWait:
Bob Liua9b54bb2015-06-19 00:23:00 -04002473 if (dev->state != XenbusStateInitialising)
2474 break;
Konrad Rzeszutek Wilkc31ecf62015-12-18 16:28:53 -05002475 if (talk_to_blkback(dev, info))
Bob Liu8ab01442015-06-03 13:40:02 +08002476 break;
Bob Liu8ab01442015-06-03 13:40:02 +08002477 case XenbusStateInitialising:
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002478 case XenbusStateInitialised:
Noboru Iwamatsub78c9512009-10-13 17:22:29 -04002479 case XenbusStateReconfiguring:
2480 case XenbusStateReconfigured:
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002481 case XenbusStateUnknown:
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002482 break;
2483
2484 case XenbusStateConnected:
Konrad Rzeszutek Wilkc31ecf62015-12-18 16:28:53 -05002485 if (dev->state != XenbusStateInitialised) {
2486 if (talk_to_blkback(dev, info))
2487 break;
2488 }
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002489 blkfront_connect(info);
2490 break;
2491
David Vrabel36613712014-02-04 18:53:56 +00002492 case XenbusStateClosed:
2493 if (dev->state == XenbusStateClosed)
2494 break;
2495 /* Missed the backend's Closing state -- fallthrough */
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002496 case XenbusStateClosing:
Cathy Averya54c8f02015-10-02 09:35:01 -04002497 if (info)
2498 blkfront_closing(info);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002499 break;
2500 }
2501}
2502
Daniel Stoddenfa1bd352010-04-30 22:01:22 +00002503static int blkfront_remove(struct xenbus_device *xbdev)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002504{
Daniel Stoddenfa1bd352010-04-30 22:01:22 +00002505 struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2506 struct block_device *bdev = NULL;
2507 struct gendisk *disk;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002508
Daniel Stoddenfa1bd352010-04-30 22:01:22 +00002509 dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002510
2511 blkif_free(info, 0);
2512
Daniel Stoddenfa1bd352010-04-30 22:01:22 +00002513 mutex_lock(&info->mutex);
2514
2515 disk = info->gd;
2516 if (disk)
2517 bdev = bdget_disk(disk, 0);
2518
2519 info->xbdev = NULL;
2520 mutex_unlock(&info->mutex);
2521
2522 if (!bdev) {
Jan Beulich0e345822010-08-07 18:28:55 +02002523 kfree(info);
Daniel Stoddenfa1bd352010-04-30 22:01:22 +00002524 return 0;
2525 }
2526
2527 /*
2528 * The xbdev was removed before we reached the Closed
2529 * state. See if it's safe to remove the disk. If the bdev
2530 * isn't closed yet, we let release take care of it.
2531 */
2532
2533 mutex_lock(&bdev->bd_mutex);
2534 info = disk->private_data;
2535
Daniel Stoddend54142c2010-08-07 18:51:21 +02002536 dev_warn(disk_to_dev(disk),
2537 "%s was hot-unplugged, %d stale handles\n",
2538 xbdev->nodename, bdev->bd_openers);
2539
Daniel Stodden7b32d102010-04-30 22:01:23 +00002540 if (info && !bdev->bd_openers) {
Daniel Stoddenfa1bd352010-04-30 22:01:22 +00002541 xlvbd_release_gendisk(info);
2542 disk->private_data = NULL;
2543 kfree(info);
2544 }
2545
2546 mutex_unlock(&bdev->bd_mutex);
2547 bdput(bdev);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002548
2549 return 0;
2550}
2551
Christian Limpach1d78d702008-04-02 10:54:04 -07002552static int blkfront_is_ready(struct xenbus_device *dev)
2553{
Greg Kroah-Hartmana1b4b122009-04-30 14:43:31 -07002554 struct blkfront_info *info = dev_get_drvdata(&dev->dev);
Christian Limpach1d78d702008-04-02 10:54:04 -07002555
Jan Beulich5d7ed202010-08-07 18:31:12 +02002556 return info->is_ready && info->xbdev;
Christian Limpach1d78d702008-04-02 10:54:04 -07002557}
2558
Al Viroa63c8482008-03-02 10:23:47 -05002559static int blkif_open(struct block_device *bdev, fmode_t mode)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002560{
Daniel Stodden13961742010-08-07 18:36:53 +02002561 struct gendisk *disk = bdev->bd_disk;
2562 struct blkfront_info *info;
2563 int err = 0;
Arnd Bergmann6e9624b2010-08-07 18:25:34 +02002564
Arnd Bergmann2a48fc02010-06-02 14:28:52 +02002565 mutex_lock(&blkfront_mutex);
Arnd Bergmann6e9624b2010-08-07 18:25:34 +02002566
Daniel Stodden13961742010-08-07 18:36:53 +02002567 info = disk->private_data;
2568 if (!info) {
2569 /* xbdev gone */
2570 err = -ERESTARTSYS;
2571 goto out;
2572 }
2573
2574 mutex_lock(&info->mutex);
2575
2576 if (!info->gd)
2577 /* xbdev is closed */
2578 err = -ERESTARTSYS;
2579
2580 mutex_unlock(&info->mutex);
2581
Daniel Stodden13961742010-08-07 18:36:53 +02002582out:
Arnd Bergmann2a48fc02010-06-02 14:28:52 +02002583 mutex_unlock(&blkfront_mutex);
Daniel Stodden13961742010-08-07 18:36:53 +02002584 return err;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002585}
2586
Al Virodb2a1442013-05-05 21:52:57 -04002587static void blkif_release(struct gendisk *disk, fmode_t mode)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002588{
Al Viroa63c8482008-03-02 10:23:47 -05002589 struct blkfront_info *info = disk->private_data;
Daniel Stodden7fd152f2010-08-07 18:45:12 +02002590 struct block_device *bdev;
2591 struct xenbus_device *xbdev;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002592
Arnd Bergmann2a48fc02010-06-02 14:28:52 +02002593 mutex_lock(&blkfront_mutex);
Daniel Stodden7fd152f2010-08-07 18:45:12 +02002594
2595 bdev = bdget_disk(disk, 0);
Daniel Stodden7fd152f2010-08-07 18:45:12 +02002596
Felipe Pena2f089cb2013-11-09 13:36:09 -02002597 if (!bdev) {
2598 WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2599 goto out_mutex;
2600 }
Daniel Stoddenacfca3c2010-08-07 18:47:26 +02002601 if (bdev->bd_openers)
2602 goto out;
2603
Daniel Stodden7fd152f2010-08-07 18:45:12 +02002604 /*
2605 * Check if we have been instructed to close. We will have
2606 * deferred this request, because the bdev was still open.
2607 */
2608
2609 mutex_lock(&info->mutex);
2610 xbdev = info->xbdev;
2611
2612 if (xbdev && xbdev->state == XenbusStateClosing) {
2613 /* pending switch to state closed */
Daniel Stoddend54142c2010-08-07 18:51:21 +02002614 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
Daniel Stodden7fd152f2010-08-07 18:45:12 +02002615 xlvbd_release_gendisk(info);
2616 xenbus_frontend_closed(info->xbdev);
2617 }
2618
2619 mutex_unlock(&info->mutex);
2620
2621 if (!xbdev) {
2622 /* sudden device removal */
Daniel Stoddend54142c2010-08-07 18:51:21 +02002623 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
Daniel Stodden7fd152f2010-08-07 18:45:12 +02002624 xlvbd_release_gendisk(info);
2625 disk->private_data = NULL;
2626 kfree(info);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002627 }
Daniel Stodden7fd152f2010-08-07 18:45:12 +02002628
Jens Axboea4cc14e2010-08-08 21:50:05 -04002629out:
Andrew Jonesdad5cf62012-02-16 13:16:25 +01002630 bdput(bdev);
Felipe Pena2f089cb2013-11-09 13:36:09 -02002631out_mutex:
Arnd Bergmann2a48fc02010-06-02 14:28:52 +02002632 mutex_unlock(&blkfront_mutex);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002633}
2634
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07002635static const struct block_device_operations xlvbd_block_fops =
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002636{
2637 .owner = THIS_MODULE,
Al Viroa63c8482008-03-02 10:23:47 -05002638 .open = blkif_open,
2639 .release = blkif_release,
Ian Campbell597592d2008-02-21 13:03:45 -08002640 .getgeo = blkif_getgeo,
Arnd Bergmann8a6cfeb2010-07-08 10:18:46 +02002641 .ioctl = blkif_ioctl,
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002642};
2643
2644
Márton Némethec9c42e2010-01-10 13:39:52 +01002645static const struct xenbus_device_id blkfront_ids[] = {
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002646 { "vbd" },
2647 { "" }
2648};
2649
David Vrabel95afae42014-09-08 17:30:41 +01002650static struct xenbus_driver blkfront_driver = {
2651 .ids = blkfront_ids,
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002652 .probe = blkfront_probe,
2653 .remove = blkfront_remove,
2654 .resume = blkfront_resume,
Ian Campbell203fd612009-12-04 15:33:54 +00002655 .otherend_changed = blkback_changed,
Christian Limpach1d78d702008-04-02 10:54:04 -07002656 .is_ready = blkfront_is_ready,
David Vrabel95afae42014-09-08 17:30:41 +01002657};
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002658
2659static int __init xlblk_init(void)
2660{
Laszlo Ersek469738e2011-10-07 21:34:38 +02002661 int ret;
Bob Liu28d949b2015-11-14 11:12:14 +08002662 int nr_cpus = num_online_cpus();
Laszlo Ersek469738e2011-10-07 21:34:38 +02002663
Jeremy Fitzhardinge6e833582008-08-19 13:16:17 -07002664 if (!xen_domain())
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002665 return -ENODEV;
2666
Julien Grall9cce2912015-10-13 17:50:11 +01002667 if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
Bob Liu86839c52015-06-03 13:40:03 +08002668 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
Julien Grall9cce2912015-10-13 17:50:11 +01002669 xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
Peng Fan45fc8262015-11-25 18:26:01 +08002670 xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
Bob Liu86839c52015-06-03 13:40:03 +08002671 }
2672
Bob Liu28d949b2015-11-14 11:12:14 +08002673 if (xen_blkif_max_queues > nr_cpus) {
2674 pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2675 xen_blkif_max_queues, nr_cpus);
2676 xen_blkif_max_queues = nr_cpus;
2677 }
2678
Konrad Rzeszutek Wilk51c71a32013-11-26 15:05:40 -05002679 if (!xen_has_pv_disk_devices())
Igor Mammedovb9136d22012-03-21 15:08:38 +01002680 return -ENODEV;
2681
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002682 if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2683 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
2684 XENVBD_MAJOR, DEV_NAME);
2685 return -ENODEV;
2686 }
2687
Jan Beulich73db1442011-12-22 09:08:13 +00002688 ret = xenbus_register_frontend(&blkfront_driver);
Laszlo Ersek469738e2011-10-07 21:34:38 +02002689 if (ret) {
2690 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2691 return ret;
2692 }
2693
2694 return 0;
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002695}
2696module_init(xlblk_init);
2697
2698
Jan Beulich5a60d0c2008-06-17 10:47:08 +02002699static void __exit xlblk_exit(void)
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002700{
Jan Beulich86050672012-04-05 16:04:52 +01002701 xenbus_unregister_driver(&blkfront_driver);
2702 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2703 kfree(minors);
Jeremy Fitzhardinge9f27ee52007-07-17 18:37:06 -07002704}
2705module_exit(xlblk_exit);
2706
2707MODULE_DESCRIPTION("Xen virtual block device frontend");
2708MODULE_LICENSE("GPL");
2709MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
Mark McLoughlind2f0c522008-04-02 10:54:05 -07002710MODULE_ALIAS("xen:vbd");
Mark McLoughlin4f93f09b2008-04-02 10:54:06 -07002711MODULE_ALIAS("xenblk");