]>
Commit | Line | Data |
---|---|---|
9f27ee59 JF |
1 | /* |
2 | * blkfront.c | |
3 | * | |
4 | * XenLinux virtual block device driver. | |
5 | * | |
6 | * Copyright (c) 2003-2004, Keir Fraser & Steve Hand | |
7 | * Modifications by Mark A. Williamson are (c) Intel Research Cambridge | |
8 | * Copyright (c) 2004, Christian Limpach | |
9 | * Copyright (c) 2004, Andrew Warfield | |
10 | * Copyright (c) 2005, Christopher Clark | |
11 | * Copyright (c) 2005, XenSource Ltd | |
12 | * | |
13 | * This program is free software; you can redistribute it and/or | |
14 | * modify it under the terms of the GNU General Public License version 2 | |
15 | * as published by the Free Software Foundation; or, when distributed | |
16 | * separately from the Linux kernel or incorporated into other | |
17 | * software packages, subject to the following license: | |
18 | * | |
19 | * Permission is hereby granted, free of charge, to any person obtaining a copy | |
20 | * of this source file (the "Software"), to deal in the Software without | |
21 | * restriction, including without limitation the rights to use, copy, modify, | |
22 | * merge, publish, distribute, sublicense, and/or sell copies of the Software, | |
23 | * and to permit persons to whom the Software is furnished to do so, subject to | |
24 | * the following conditions: | |
25 | * | |
26 | * The above copyright notice and this permission notice shall be included in | |
27 | * all copies or substantial portions of the Software. | |
28 | * | |
29 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
30 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
31 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE | |
32 | * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
33 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING | |
34 | * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS | |
35 | * IN THE SOFTWARE. | |
36 | */ | |
37 | ||
38 | #include <linux/interrupt.h> | |
39 | #include <linux/blkdev.h> | |
907c3eb1 | 40 | #include <linux/blk-mq.h> |
597592d9 | 41 | #include <linux/hdreg.h> |
440a01a7 | 42 | #include <linux/cdrom.h> |
9f27ee59 | 43 | #include <linux/module.h> |
5a0e3ad6 | 44 | #include <linux/slab.h> |
2a48fc0a | 45 | #include <linux/mutex.h> |
9e973e64 | 46 | #include <linux/scatterlist.h> |
34ae2e47 | 47 | #include <linux/bitmap.h> |
155b7edb | 48 | #include <linux/list.h> |
a46b5367 | 49 | #include <linux/workqueue.h> |
3a169c0b | 50 | #include <linux/sched/mm.h> |
9f27ee59 | 51 | |
1ccbf534 | 52 | #include <xen/xen.h> |
9f27ee59 JF |
53 | #include <xen/xenbus.h> |
54 | #include <xen/grant_table.h> | |
55 | #include <xen/events.h> | |
56 | #include <xen/page.h> | |
c1c5413a | 57 | #include <xen/platform_pci.h> |
9f27ee59 JF |
58 | |
59 | #include <xen/interface/grant_table.h> | |
60 | #include <xen/interface/io/blkif.h> | |
3e334239 | 61 | #include <xen/interface/io/protocols.h> |
9f27ee59 JF |
62 | |
63 | #include <asm/xen/hypervisor.h> | |
64 | ||
6cc56833 JG |
65 | /* |
66 | * The minimal size of segment supported by the block framework is PAGE_SIZE. | |
67 | * When Linux is using a different page size than Xen, it may not be possible | |
68 | * to put all the data in a single segment. | |
69 | * This can happen when the backend doesn't support indirect descriptor and | |
70 | * therefore the maximum amount of data that a request can carry is | |
71 | * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE = 44KB | |
72 | * | |
73 | * Note that we only support one extra request. So the Linux page size | |
74 | * should be <= ( 2 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) = | |
75 | * 88KB. | |
76 | */ | |
77 | #define HAS_EXTRA_REQ (BLKIF_MAX_SEGMENTS_PER_REQUEST < XEN_PFN_PER_PAGE) | |
78 | ||
9f27ee59 JF |
79 | enum blkif_state { |
80 | BLKIF_STATE_DISCONNECTED, | |
81 | BLKIF_STATE_CONNECTED, | |
82 | BLKIF_STATE_SUSPENDED, | |
83 | }; | |
84 | ||
0a8704a5 RPM |
85 | struct grant { |
86 | grant_ref_t gref; | |
a7a6df22 | 87 | struct page *page; |
155b7edb | 88 | struct list_head node; |
0a8704a5 RPM |
89 | }; |
90 | ||
6cc56833 JG |
91 | enum blk_req_status { |
92 | REQ_WAITING, | |
93 | REQ_DONE, | |
94 | REQ_ERROR, | |
95 | REQ_EOPNOTSUPP, | |
96 | }; | |
97 | ||
9f27ee59 JF |
98 | struct blk_shadow { |
99 | struct blkif_request req; | |
a945b980 | 100 | struct request *request; |
402b27f9 RPM |
101 | struct grant **grants_used; |
102 | struct grant **indirect_grants; | |
b7649158 | 103 | struct scatterlist *sg; |
c004a6fe | 104 | unsigned int num_sg; |
6cc56833 JG |
105 | enum blk_req_status status; |
106 | ||
107 | #define NO_ASSOCIATED_ID ~0UL | |
108 | /* | |
109 | * Id of the sibling if we ever need 2 requests when handling a | |
110 | * block I/O request | |
111 | */ | |
112 | unsigned long associated_id; | |
402b27f9 RPM |
113 | }; |
114 | ||
2609587c | 115 | struct blkif_req { |
31c4ccc3 | 116 | blk_status_t error; |
2609587c CH |
117 | }; |
118 | ||
119 | static inline struct blkif_req *blkif_req(struct request *rq) | |
120 | { | |
121 | return blk_mq_rq_to_pdu(rq); | |
122 | } | |
123 | ||
2a48fc0a | 124 | static DEFINE_MUTEX(blkfront_mutex); |
83d5cde4 | 125 | static const struct block_device_operations xlvbd_block_fops; |
a46b5367 JG |
126 | static struct delayed_work blkfront_work; |
127 | static LIST_HEAD(info_list); | |
9f27ee59 | 128 | |
402b27f9 RPM |
129 | /* |
130 | * Maximum number of segments in indirect requests, the actual value used by | |
131 | * the frontend driver is the minimum of this value and the value provided | |
132 | * by the backend driver. | |
133 | */ | |
134 | ||
135 | static unsigned int xen_blkif_max_segments = 32; | |
5657a819 | 136 | module_param_named(max_indirect_segments, xen_blkif_max_segments, uint, 0444); |
14e710fe JB |
137 | MODULE_PARM_DESC(max_indirect_segments, |
138 | "Maximum amount of segments in indirect requests (default is 32)"); | |
402b27f9 | 139 | |
28d949bc | 140 | static unsigned int xen_blkif_max_queues = 4; |
5657a819 | 141 | module_param_named(max_queues, xen_blkif_max_queues, uint, 0444); |
28d949bc BL |
142 | MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk"); |
143 | ||
86839c56 BL |
144 | /* |
145 | * Maximum order of pages to be used for the shared ring between front and | |
146 | * backend, 4KB page granularity is used. | |
147 | */ | |
148 | static unsigned int xen_blkif_max_ring_order; | |
5657a819 | 149 | module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, 0444); |
86839c56 BL |
150 | MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring"); |
151 | ||
c004a6fe JG |
152 | #define BLK_RING_SIZE(info) \ |
153 | __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages) | |
154 | ||
86839c56 | 155 | /* |
6f03a7ff KRW |
156 | * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19 |
157 | * characters are enough. Define to 20 to keep consistent with backend. | |
86839c56 BL |
158 | */ |
159 | #define RINGREF_NAME_LEN (20) | |
28d949bc BL |
160 | /* |
161 | * queue-%u would take 7 + 10(UINT_MAX) = 17 characters. | |
162 | */ | |
163 | #define QUEUE_NAME_LEN (17) | |
9f27ee59 | 164 | |
81f35161 BL |
165 | /* |
166 | * Per-ring info. | |
167 | * Every blkfront device can associate with one or more blkfront_ring_info, | |
168 | * depending on how many hardware queues/rings to be used. | |
169 | */ | |
170 | struct blkfront_ring_info { | |
11659569 BL |
171 | /* Lock to protect data in every ring buffer. */ |
172 | spinlock_t ring_lock; | |
81f35161 BL |
173 | struct blkif_front_ring ring; |
174 | unsigned int ring_ref[XENBUS_MAX_RING_GRANTS]; | |
175 | unsigned int evtchn, irq; | |
176 | struct work_struct work; | |
177 | struct gnttab_free_callback callback; | |
81f35161 | 178 | struct list_head indirect_pages; |
73716df7 BL |
179 | struct list_head grants; |
180 | unsigned int persistent_gnts_c; | |
81f35161 BL |
181 | unsigned long shadow_free; |
182 | struct blkfront_info *dev_info; | |
0265d6e8 | 183 | struct blk_shadow shadow[]; |
81f35161 BL |
184 | }; |
185 | ||
9f27ee59 JF |
186 | /* |
187 | * We have one of these per vbd, whether ide, scsi or 'other'. They | |
188 | * hang in private_data off the gendisk structure. We may end up | |
189 | * putting all kinds of interesting stuff here :-) | |
190 | */ | |
191 | struct blkfront_info | |
192 | { | |
b70f5fa0 | 193 | struct mutex mutex; |
9f27ee59 | 194 | struct xenbus_device *xbdev; |
9f27ee59 | 195 | struct gendisk *gd; |
172335ad BL |
196 | u16 sector_size; |
197 | unsigned int physical_sector_size; | |
9f27ee59 JF |
198 | int vdevice; |
199 | blkif_vdev_t handle; | |
200 | enum blkif_state connected; | |
3df0e505 | 201 | /* Number of pages per ring buffer. */ |
86839c56 | 202 | unsigned int nr_ring_pages; |
9f27ee59 | 203 | struct request_queue *rq; |
b32728ff JB |
204 | unsigned int feature_flush:1; |
205 | unsigned int feature_fua:1; | |
5ea42986 KRW |
206 | unsigned int feature_discard:1; |
207 | unsigned int feature_secdiscard:1; | |
b32728ff | 208 | unsigned int feature_persistent:1; |
ed30bf31 LD |
209 | unsigned int discard_granularity; |
210 | unsigned int discard_alignment; | |
c004a6fe | 211 | /* Number of 4KB segments handled */ |
402b27f9 | 212 | unsigned int max_indirect_segments; |
1d78d705 | 213 | int is_ready; |
907c3eb1 | 214 | struct blk_mq_tag_set tag_set; |
3df0e505 BL |
215 | struct blkfront_ring_info *rinfo; |
216 | unsigned int nr_rings; | |
4ab50af6 | 217 | unsigned int rinfo_size; |
7b427a59 BL |
218 | /* Save uncomplete reqs and bios for migration. */ |
219 | struct list_head requests; | |
220 | struct bio_list bio_list; | |
a46b5367 | 221 | struct list_head info_list; |
9f27ee59 JF |
222 | }; |
223 | ||
0e345826 JB |
224 | static unsigned int nr_minors; |
225 | static unsigned long *minors; | |
226 | static DEFINE_SPINLOCK(minor_lock); | |
227 | ||
9f27ee59 JF |
228 | #define GRANT_INVALID_REF 0 |
229 | ||
230 | #define PARTS_PER_DISK 16 | |
9246b5f0 | 231 | #define PARTS_PER_EXT_DISK 256 |
9f27ee59 JF |
232 | |
233 | #define BLKIF_MAJOR(dev) ((dev)>>8) | |
234 | #define BLKIF_MINOR(dev) ((dev) & 0xff) | |
235 | ||
9246b5f0 CL |
236 | #define EXT_SHIFT 28 |
237 | #define EXTENDED (1<<EXT_SHIFT) | |
238 | #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED)) | |
239 | #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED)) | |
c80a4209 SS |
240 | #define EMULATED_HD_DISK_MINOR_OFFSET (0) |
241 | #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256) | |
196cfe2a SB |
242 | #define EMULATED_SD_DISK_MINOR_OFFSET (0) |
243 | #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256) | |
9f27ee59 | 244 | |
9246b5f0 | 245 | #define DEV_NAME "xvd" /* name in /dev */ |
9f27ee59 | 246 | |
c004a6fe JG |
247 | /* |
248 | * Grants are always the same size as a Xen page (i.e 4KB). | |
249 | * A physical segment is always the same size as a Linux page. | |
250 | * Number of grants per physical segment | |
251 | */ | |
252 | #define GRANTS_PER_PSEG (PAGE_SIZE / XEN_PAGE_SIZE) | |
253 | ||
254 | #define GRANTS_PER_INDIRECT_FRAME \ | |
255 | (XEN_PAGE_SIZE / sizeof(struct blkif_request_segment)) | |
256 | ||
c004a6fe JG |
257 | #define INDIRECT_GREFS(_grants) \ |
258 | DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME) | |
259 | ||
81f35161 | 260 | static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo); |
3df0e505 | 261 | static void blkfront_gather_backend_features(struct blkfront_info *info); |
7ed8ce1c | 262 | static int negotiate_mq(struct blkfront_info *info); |
402b27f9 | 263 | |
4ab50af6 JG |
264 | #define for_each_rinfo(info, ptr, idx) \ |
265 | for ((ptr) = (info)->rinfo, (idx) = 0; \ | |
266 | (idx) < (info)->nr_rings; \ | |
267 | (idx)++, (ptr) = (void *)(ptr) + (info)->rinfo_size) | |
268 | ||
269 | static inline struct blkfront_ring_info * | |
270 | get_rinfo(const struct blkfront_info *info, unsigned int i) | |
271 | { | |
272 | BUG_ON(i >= info->nr_rings); | |
273 | return (void *)info->rinfo + i * info->rinfo_size; | |
274 | } | |
275 | ||
81f35161 | 276 | static int get_id_from_freelist(struct blkfront_ring_info *rinfo) |
9f27ee59 | 277 | { |
81f35161 BL |
278 | unsigned long free = rinfo->shadow_free; |
279 | ||
280 | BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info)); | |
281 | rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id; | |
282 | rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */ | |
9f27ee59 JF |
283 | return free; |
284 | } | |
285 | ||
81f35161 | 286 | static int add_id_to_freelist(struct blkfront_ring_info *rinfo, |
6f03a7ff | 287 | unsigned long id) |
9f27ee59 | 288 | { |
81f35161 | 289 | if (rinfo->shadow[id].req.u.rw.id != id) |
6878c32e | 290 | return -EINVAL; |
81f35161 | 291 | if (rinfo->shadow[id].request == NULL) |
6878c32e | 292 | return -EINVAL; |
81f35161 BL |
293 | rinfo->shadow[id].req.u.rw.id = rinfo->shadow_free; |
294 | rinfo->shadow[id].request = NULL; | |
295 | rinfo->shadow_free = id; | |
6878c32e | 296 | return 0; |
9f27ee59 JF |
297 | } |
298 | ||
81f35161 | 299 | static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num) |
9c1e050c | 300 | { |
81f35161 | 301 | struct blkfront_info *info = rinfo->dev_info; |
9c1e050c RPM |
302 | struct page *granted_page; |
303 | struct grant *gnt_list_entry, *n; | |
304 | int i = 0; | |
305 | ||
6f03a7ff | 306 | while (i < num) { |
9c1e050c RPM |
307 | gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO); |
308 | if (!gnt_list_entry) | |
309 | goto out_of_memory; | |
310 | ||
bfe11d6d RPM |
311 | if (info->feature_persistent) { |
312 | granted_page = alloc_page(GFP_NOIO); | |
313 | if (!granted_page) { | |
314 | kfree(gnt_list_entry); | |
315 | goto out_of_memory; | |
316 | } | |
a7a6df22 | 317 | gnt_list_entry->page = granted_page; |
9c1e050c RPM |
318 | } |
319 | ||
9c1e050c | 320 | gnt_list_entry->gref = GRANT_INVALID_REF; |
73716df7 | 321 | list_add(&gnt_list_entry->node, &rinfo->grants); |
9c1e050c RPM |
322 | i++; |
323 | } | |
324 | ||
325 | return 0; | |
326 | ||
327 | out_of_memory: | |
328 | list_for_each_entry_safe(gnt_list_entry, n, | |
73716df7 | 329 | &rinfo->grants, node) { |
9c1e050c | 330 | list_del(&gnt_list_entry->node); |
bfe11d6d | 331 | if (info->feature_persistent) |
a7a6df22 | 332 | __free_page(gnt_list_entry->page); |
9c1e050c RPM |
333 | kfree(gnt_list_entry); |
334 | i--; | |
335 | } | |
336 | BUG_ON(i != 0); | |
337 | return -ENOMEM; | |
338 | } | |
339 | ||
73716df7 | 340 | static struct grant *get_free_grant(struct blkfront_ring_info *rinfo) |
9c1e050c RPM |
341 | { |
342 | struct grant *gnt_list_entry; | |
9c1e050c | 343 | |
73716df7 BL |
344 | BUG_ON(list_empty(&rinfo->grants)); |
345 | gnt_list_entry = list_first_entry(&rinfo->grants, struct grant, | |
4f503fbd | 346 | node); |
9c1e050c RPM |
347 | list_del(&gnt_list_entry->node); |
348 | ||
4f503fbd | 349 | if (gnt_list_entry->gref != GRANT_INVALID_REF) |
73716df7 | 350 | rinfo->persistent_gnts_c--; |
4f503fbd JG |
351 | |
352 | return gnt_list_entry; | |
353 | } | |
354 | ||
355 | static inline void grant_foreign_access(const struct grant *gnt_list_entry, | |
356 | const struct blkfront_info *info) | |
357 | { | |
358 | gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref, | |
359 | info->xbdev->otherend_id, | |
360 | gnt_list_entry->page, | |
361 | 0); | |
362 | } | |
363 | ||
364 | static struct grant *get_grant(grant_ref_t *gref_head, | |
365 | unsigned long gfn, | |
73716df7 | 366 | struct blkfront_ring_info *rinfo) |
4f503fbd | 367 | { |
73716df7 BL |
368 | struct grant *gnt_list_entry = get_free_grant(rinfo); |
369 | struct blkfront_info *info = rinfo->dev_info; | |
4f503fbd JG |
370 | |
371 | if (gnt_list_entry->gref != GRANT_INVALID_REF) | |
9c1e050c | 372 | return gnt_list_entry; |
4f503fbd JG |
373 | |
374 | /* Assign a gref to this page */ | |
375 | gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head); | |
376 | BUG_ON(gnt_list_entry->gref == -ENOSPC); | |
377 | if (info->feature_persistent) | |
378 | grant_foreign_access(gnt_list_entry, info); | |
379 | else { | |
380 | /* Grant access to the GFN passed by the caller */ | |
381 | gnttab_grant_foreign_access_ref(gnt_list_entry->gref, | |
382 | info->xbdev->otherend_id, | |
383 | gfn, 0); | |
9c1e050c RPM |
384 | } |
385 | ||
4f503fbd JG |
386 | return gnt_list_entry; |
387 | } | |
388 | ||
389 | static struct grant *get_indirect_grant(grant_ref_t *gref_head, | |
73716df7 | 390 | struct blkfront_ring_info *rinfo) |
4f503fbd | 391 | { |
73716df7 BL |
392 | struct grant *gnt_list_entry = get_free_grant(rinfo); |
393 | struct blkfront_info *info = rinfo->dev_info; | |
4f503fbd JG |
394 | |
395 | if (gnt_list_entry->gref != GRANT_INVALID_REF) | |
396 | return gnt_list_entry; | |
397 | ||
9c1e050c RPM |
398 | /* Assign a gref to this page */ |
399 | gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head); | |
400 | BUG_ON(gnt_list_entry->gref == -ENOSPC); | |
bfe11d6d | 401 | if (!info->feature_persistent) { |
4f503fbd JG |
402 | struct page *indirect_page; |
403 | ||
404 | /* Fetch a pre-allocated page to use for indirect grefs */ | |
73716df7 BL |
405 | BUG_ON(list_empty(&rinfo->indirect_pages)); |
406 | indirect_page = list_first_entry(&rinfo->indirect_pages, | |
4f503fbd JG |
407 | struct page, lru); |
408 | list_del(&indirect_page->lru); | |
409 | gnt_list_entry->page = indirect_page; | |
bfe11d6d | 410 | } |
4f503fbd JG |
411 | grant_foreign_access(gnt_list_entry, info); |
412 | ||
9c1e050c RPM |
413 | return gnt_list_entry; |
414 | } | |
415 | ||
6878c32e KRW |
416 | static const char *op_name(int op) |
417 | { | |
418 | static const char *const names[] = { | |
419 | [BLKIF_OP_READ] = "read", | |
420 | [BLKIF_OP_WRITE] = "write", | |
421 | [BLKIF_OP_WRITE_BARRIER] = "barrier", | |
422 | [BLKIF_OP_FLUSH_DISKCACHE] = "flush", | |
423 | [BLKIF_OP_DISCARD] = "discard" }; | |
424 | ||
425 | if (op < 0 || op >= ARRAY_SIZE(names)) | |
426 | return "unknown"; | |
427 | ||
428 | if (!names[op]) | |
429 | return "reserved"; | |
430 | ||
431 | return names[op]; | |
432 | } | |
0e345826 JB |
433 | static int xlbd_reserve_minors(unsigned int minor, unsigned int nr) |
434 | { | |
435 | unsigned int end = minor + nr; | |
436 | int rc; | |
437 | ||
438 | if (end > nr_minors) { | |
439 | unsigned long *bitmap, *old; | |
440 | ||
f094148a | 441 | bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap), |
0e345826 JB |
442 | GFP_KERNEL); |
443 | if (bitmap == NULL) | |
444 | return -ENOMEM; | |
445 | ||
446 | spin_lock(&minor_lock); | |
447 | if (end > nr_minors) { | |
448 | old = minors; | |
449 | memcpy(bitmap, minors, | |
450 | BITS_TO_LONGS(nr_minors) * sizeof(*bitmap)); | |
451 | minors = bitmap; | |
452 | nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG; | |
453 | } else | |
454 | old = bitmap; | |
455 | spin_unlock(&minor_lock); | |
456 | kfree(old); | |
457 | } | |
458 | ||
459 | spin_lock(&minor_lock); | |
460 | if (find_next_bit(minors, end, minor) >= end) { | |
34ae2e47 | 461 | bitmap_set(minors, minor, nr); |
0e345826 JB |
462 | rc = 0; |
463 | } else | |
464 | rc = -EBUSY; | |
465 | spin_unlock(&minor_lock); | |
466 | ||
467 | return rc; | |
468 | } | |
469 | ||
470 | static void xlbd_release_minors(unsigned int minor, unsigned int nr) | |
471 | { | |
472 | unsigned int end = minor + nr; | |
473 | ||
474 | BUG_ON(end > nr_minors); | |
475 | spin_lock(&minor_lock); | |
34ae2e47 | 476 | bitmap_clear(minors, minor, nr); |
0e345826 JB |
477 | spin_unlock(&minor_lock); |
478 | } | |
479 | ||
9f27ee59 JF |
480 | static void blkif_restart_queue_callback(void *arg) |
481 | { | |
81f35161 BL |
482 | struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg; |
483 | schedule_work(&rinfo->work); | |
9f27ee59 JF |
484 | } |
485 | ||
afe42d7d | 486 | static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg) |
597592d9 IC |
487 | { |
488 | /* We don't have real geometry info, but let's at least return | |
489 | values consistent with the size of the device */ | |
490 | sector_t nsect = get_capacity(bd->bd_disk); | |
491 | sector_t cylinders = nsect; | |
492 | ||
493 | hg->heads = 0xff; | |
494 | hg->sectors = 0x3f; | |
495 | sector_div(cylinders, hg->heads * hg->sectors); | |
496 | hg->cylinders = cylinders; | |
497 | if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect) | |
498 | hg->cylinders = 0xffff; | |
499 | return 0; | |
500 | } | |
501 | ||
a63c848b | 502 | static int blkif_ioctl(struct block_device *bdev, fmode_t mode, |
62aa0054 | 503 | unsigned command, unsigned long argument) |
440a01a7 | 504 | { |
a63c848b | 505 | struct blkfront_info *info = bdev->bd_disk->private_data; |
440a01a7 CL |
506 | int i; |
507 | ||
508 | dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n", | |
509 | command, (long)argument); | |
510 | ||
511 | switch (command) { | |
512 | case CDROMMULTISESSION: | |
513 | dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n"); | |
514 | for (i = 0; i < sizeof(struct cdrom_multisession); i++) | |
515 | if (put_user(0, (char __user *)(argument + i))) | |
516 | return -EFAULT; | |
517 | return 0; | |
518 | ||
519 | case CDROM_GET_CAPABILITY: { | |
520 | struct gendisk *gd = info->gd; | |
521 | if (gd->flags & GENHD_FL_CD) | |
522 | return 0; | |
523 | return -EINVAL; | |
524 | } | |
525 | ||
526 | default: | |
527 | /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n", | |
528 | command);*/ | |
529 | return -EINVAL; /* same return as native Linux */ | |
530 | } | |
531 | ||
532 | return 0; | |
533 | } | |
534 | ||
2e073969 JG |
535 | static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo, |
536 | struct request *req, | |
537 | struct blkif_request **ring_req) | |
538 | { | |
539 | unsigned long id; | |
540 | ||
541 | *ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt); | |
542 | rinfo->ring.req_prod_pvt++; | |
543 | ||
544 | id = get_id_from_freelist(rinfo); | |
545 | rinfo->shadow[id].request = req; | |
6cc56833 JG |
546 | rinfo->shadow[id].status = REQ_WAITING; |
547 | rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID; | |
2e073969 JG |
548 | |
549 | (*ring_req)->u.rw.id = id; | |
550 | ||
551 | return id; | |
552 | } | |
553 | ||
81f35161 | 554 | static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo) |
9f27ee59 | 555 | { |
81f35161 | 556 | struct blkfront_info *info = rinfo->dev_info; |
9f27ee59 | 557 | struct blkif_request *ring_req; |
9f27ee59 | 558 | unsigned long id; |
33204663 JG |
559 | |
560 | /* Fill out a communications ring structure. */ | |
2e073969 | 561 | id = blkif_ring_get_request(rinfo, req, &ring_req); |
33204663 JG |
562 | |
563 | ring_req->operation = BLKIF_OP_DISCARD; | |
564 | ring_req->u.discard.nr_sectors = blk_rq_sectors(req); | |
565 | ring_req->u.discard.id = id; | |
566 | ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req); | |
288dab8a | 567 | if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard) |
33204663 JG |
568 | ring_req->u.discard.flag = BLKIF_DISCARD_SECURE; |
569 | else | |
570 | ring_req->u.discard.flag = 0; | |
571 | ||
33204663 | 572 | /* Keep a private copy so we can reissue requests when recovering. */ |
81f35161 | 573 | rinfo->shadow[id].req = *ring_req; |
33204663 JG |
574 | |
575 | return 0; | |
576 | } | |
577 | ||
c004a6fe JG |
578 | struct setup_rw_req { |
579 | unsigned int grant_idx; | |
580 | struct blkif_request_segment *segments; | |
81f35161 | 581 | struct blkfront_ring_info *rinfo; |
c004a6fe JG |
582 | struct blkif_request *ring_req; |
583 | grant_ref_t gref_head; | |
584 | unsigned int id; | |
585 | /* Only used when persistent grant is used and it's a read request */ | |
586 | bool need_copy; | |
587 | unsigned int bvec_off; | |
588 | char *bvec_data; | |
6cc56833 JG |
589 | |
590 | bool require_extra_req; | |
591 | struct blkif_request *extra_ring_req; | |
c004a6fe JG |
592 | }; |
593 | ||
594 | static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset, | |
595 | unsigned int len, void *data) | |
596 | { | |
597 | struct setup_rw_req *setup = data; | |
598 | int n, ref; | |
599 | struct grant *gnt_list_entry; | |
9f27ee59 | 600 | unsigned int fsect, lsect; |
c004a6fe JG |
601 | /* Convenient aliases */ |
602 | unsigned int grant_idx = setup->grant_idx; | |
603 | struct blkif_request *ring_req = setup->ring_req; | |
81f35161 | 604 | struct blkfront_ring_info *rinfo = setup->rinfo; |
6cc56833 JG |
605 | /* |
606 | * We always use the shadow of the first request to store the list | |
607 | * of grant associated to the block I/O request. This made the | |
608 | * completion more easy to handle even if the block I/O request is | |
609 | * split. | |
610 | */ | |
81f35161 | 611 | struct blk_shadow *shadow = &rinfo->shadow[setup->id]; |
c004a6fe | 612 | |
6cc56833 JG |
613 | if (unlikely(setup->require_extra_req && |
614 | grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) { | |
615 | /* | |
616 | * We are using the second request, setup grant_idx | |
617 | * to be the index of the segment array. | |
618 | */ | |
619 | grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST; | |
620 | ring_req = setup->extra_ring_req; | |
621 | } | |
622 | ||
c004a6fe JG |
623 | if ((ring_req->operation == BLKIF_OP_INDIRECT) && |
624 | (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) { | |
625 | if (setup->segments) | |
626 | kunmap_atomic(setup->segments); | |
627 | ||
628 | n = grant_idx / GRANTS_PER_INDIRECT_FRAME; | |
73716df7 | 629 | gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo); |
c004a6fe JG |
630 | shadow->indirect_grants[n] = gnt_list_entry; |
631 | setup->segments = kmap_atomic(gnt_list_entry->page); | |
632 | ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref; | |
633 | } | |
634 | ||
73716df7 | 635 | gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo); |
c004a6fe | 636 | ref = gnt_list_entry->gref; |
6cc56833 JG |
637 | /* |
638 | * All the grants are stored in the shadow of the first | |
639 | * request. Therefore we have to use the global index. | |
640 | */ | |
641 | shadow->grants_used[setup->grant_idx] = gnt_list_entry; | |
c004a6fe JG |
642 | |
643 | if (setup->need_copy) { | |
644 | void *shared_data; | |
645 | ||
646 | shared_data = kmap_atomic(gnt_list_entry->page); | |
647 | /* | |
648 | * this does not wipe data stored outside the | |
649 | * range sg->offset..sg->offset+sg->length. | |
650 | * Therefore, blkback *could* see data from | |
651 | * previous requests. This is OK as long as | |
652 | * persistent grants are shared with just one | |
653 | * domain. It may need refactoring if this | |
654 | * changes | |
655 | */ | |
656 | memcpy(shared_data + offset, | |
657 | setup->bvec_data + setup->bvec_off, | |
658 | len); | |
659 | ||
660 | kunmap_atomic(shared_data); | |
661 | setup->bvec_off += len; | |
662 | } | |
663 | ||
664 | fsect = offset >> 9; | |
665 | lsect = fsect + (len >> 9) - 1; | |
666 | if (ring_req->operation != BLKIF_OP_INDIRECT) { | |
667 | ring_req->u.rw.seg[grant_idx] = | |
668 | (struct blkif_request_segment) { | |
669 | .gref = ref, | |
670 | .first_sect = fsect, | |
671 | .last_sect = lsect }; | |
672 | } else { | |
673 | setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] = | |
674 | (struct blkif_request_segment) { | |
675 | .gref = ref, | |
676 | .first_sect = fsect, | |
677 | .last_sect = lsect }; | |
678 | } | |
679 | ||
680 | (setup->grant_idx)++; | |
681 | } | |
682 | ||
6cc56833 JG |
683 | static void blkif_setup_extra_req(struct blkif_request *first, |
684 | struct blkif_request *second) | |
685 | { | |
686 | uint16_t nr_segments = first->u.rw.nr_segments; | |
687 | ||
688 | /* | |
689 | * The second request is only present when the first request uses | |
690 | * all its segments. It's always the continuity of the first one. | |
691 | */ | |
692 | first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST; | |
693 | ||
694 | second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST; | |
695 | second->u.rw.sector_number = first->u.rw.sector_number + | |
696 | (BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512; | |
697 | ||
698 | second->u.rw.handle = first->u.rw.handle; | |
699 | second->operation = first->operation; | |
700 | } | |
701 | ||
81f35161 | 702 | static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo) |
9f27ee59 | 703 | { |
81f35161 | 704 | struct blkfront_info *info = rinfo->dev_info; |
6cc56833 JG |
705 | struct blkif_request *ring_req, *extra_ring_req = NULL; |
706 | unsigned long id, extra_id = NO_ASSOCIATED_ID; | |
707 | bool require_extra_req = false; | |
c004a6fe JG |
708 | int i; |
709 | struct setup_rw_req setup = { | |
710 | .grant_idx = 0, | |
711 | .segments = NULL, | |
81f35161 | 712 | .rinfo = rinfo, |
c004a6fe JG |
713 | .need_copy = rq_data_dir(req) && info->feature_persistent, |
714 | }; | |
0a8704a5 RPM |
715 | |
716 | /* | |
717 | * Used to store if we are able to queue the request by just using | |
718 | * existing persistent grants, or if we have to get new grants, | |
719 | * as there are not sufficiently many free. | |
720 | */ | |
bd912ef3 | 721 | bool new_persistent_gnts = false; |
9e973e64 | 722 | struct scatterlist *sg; |
c004a6fe | 723 | int num_sg, max_grefs, num_grant; |
9f27ee59 | 724 | |
c004a6fe | 725 | max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG; |
c47206e2 RPM |
726 | if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST) |
727 | /* | |
728 | * If we are using indirect segments we need to account | |
729 | * for the indirect grefs used in the request. | |
730 | */ | |
c004a6fe | 731 | max_grefs += INDIRECT_GREFS(max_grefs); |
402b27f9 | 732 | |
bd912ef3 DZ |
733 | /* Check if we have enough persistent grants to allocate a requests */ |
734 | if (rinfo->persistent_gnts_c < max_grefs) { | |
735 | new_persistent_gnts = true; | |
736 | ||
737 | if (gnttab_alloc_grant_references( | |
738 | max_grefs - rinfo->persistent_gnts_c, | |
739 | &setup.gref_head) < 0) { | |
0a8704a5 | 740 | gnttab_request_free_callback( |
81f35161 | 741 | &rinfo->callback, |
0a8704a5 | 742 | blkif_restart_queue_callback, |
81f35161 | 743 | rinfo, |
bd912ef3 | 744 | max_grefs - rinfo->persistent_gnts_c); |
0a8704a5 RPM |
745 | return 1; |
746 | } | |
bd912ef3 | 747 | } |
9f27ee59 JF |
748 | |
749 | /* Fill out a communications ring structure. */ | |
2e073969 | 750 | id = blkif_ring_get_request(rinfo, req, &ring_req); |
9f27ee59 | 751 | |
81f35161 | 752 | num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg); |
c004a6fe JG |
753 | num_grant = 0; |
754 | /* Calculate the number of grant used */ | |
81f35161 | 755 | for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) |
c004a6fe JG |
756 | num_grant += gnttab_count_grant(sg->offset, sg->length); |
757 | ||
6cc56833 JG |
758 | require_extra_req = info->max_indirect_segments == 0 && |
759 | num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST; | |
760 | BUG_ON(!HAS_EXTRA_REQ && require_extra_req); | |
761 | ||
81f35161 | 762 | rinfo->shadow[id].num_sg = num_sg; |
6cc56833 JG |
763 | if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST && |
764 | likely(!require_extra_req)) { | |
33204663 JG |
765 | /* |
766 | * The indirect operation can only be a BLKIF_OP_READ or | |
767 | * BLKIF_OP_WRITE | |
768 | */ | |
3a5e02ce | 769 | BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA); |
33204663 JG |
770 | ring_req->operation = BLKIF_OP_INDIRECT; |
771 | ring_req->u.indirect.indirect_op = rq_data_dir(req) ? | |
772 | BLKIF_OP_WRITE : BLKIF_OP_READ; | |
773 | ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req); | |
774 | ring_req->u.indirect.handle = info->handle; | |
c004a6fe | 775 | ring_req->u.indirect.nr_segments = num_grant; |
ed30bf31 | 776 | } else { |
33204663 JG |
777 | ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req); |
778 | ring_req->u.rw.handle = info->handle; | |
779 | ring_req->operation = rq_data_dir(req) ? | |
780 | BLKIF_OP_WRITE : BLKIF_OP_READ; | |
3a5e02ce | 781 | if (req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA) { |
402b27f9 | 782 | /* |
33204663 JG |
783 | * Ideally we can do an unordered flush-to-disk. |
784 | * In case the backend onlysupports barriers, use that. | |
785 | * A barrier request a superset of FUA, so we can | |
786 | * implement it the same way. (It's also a FLUSH+FUA, | |
787 | * since it is guaranteed ordered WRT previous writes.) | |
402b27f9 | 788 | */ |
a418090a | 789 | if (info->feature_flush && info->feature_fua) |
33204663 JG |
790 | ring_req->operation = |
791 | BLKIF_OP_WRITE_BARRIER; | |
a418090a | 792 | else if (info->feature_flush) |
33204663 JG |
793 | ring_req->operation = |
794 | BLKIF_OP_FLUSH_DISKCACHE; | |
a418090a | 795 | else |
33204663 | 796 | ring_req->operation = 0; |
402b27f9 | 797 | } |
c004a6fe | 798 | ring_req->u.rw.nr_segments = num_grant; |
6cc56833 JG |
799 | if (unlikely(require_extra_req)) { |
800 | extra_id = blkif_ring_get_request(rinfo, req, | |
801 | &extra_ring_req); | |
802 | /* | |
803 | * Only the first request contains the scatter-gather | |
804 | * list. | |
805 | */ | |
806 | rinfo->shadow[extra_id].num_sg = 0; | |
807 | ||
808 | blkif_setup_extra_req(ring_req, extra_ring_req); | |
809 | ||
810 | /* Link the 2 requests together */ | |
811 | rinfo->shadow[extra_id].associated_id = id; | |
812 | rinfo->shadow[id].associated_id = extra_id; | |
813 | } | |
33204663 | 814 | } |
0a8704a5 | 815 | |
c004a6fe JG |
816 | setup.ring_req = ring_req; |
817 | setup.id = id; | |
6cc56833 JG |
818 | |
819 | setup.require_extra_req = require_extra_req; | |
820 | if (unlikely(require_extra_req)) | |
821 | setup.extra_ring_req = extra_ring_req; | |
822 | ||
81f35161 | 823 | for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) { |
c004a6fe | 824 | BUG_ON(sg->offset + sg->length > PAGE_SIZE); |
0a8704a5 | 825 | |
c004a6fe JG |
826 | if (setup.need_copy) { |
827 | setup.bvec_off = sg->offset; | |
828 | setup.bvec_data = kmap_atomic(sg_page(sg)); | |
829 | } | |
0a8704a5 | 830 | |
c004a6fe JG |
831 | gnttab_foreach_grant_in_range(sg_page(sg), |
832 | sg->offset, | |
833 | sg->length, | |
834 | blkif_setup_rw_req_grant, | |
835 | &setup); | |
0a8704a5 | 836 | |
c004a6fe JG |
837 | if (setup.need_copy) |
838 | kunmap_atomic(setup.bvec_data); | |
9f27ee59 | 839 | } |
c004a6fe JG |
840 | if (setup.segments) |
841 | kunmap_atomic(setup.segments); | |
9f27ee59 | 842 | |
9f27ee59 | 843 | /* Keep a private copy so we can reissue requests when recovering. */ |
81f35161 | 844 | rinfo->shadow[id].req = *ring_req; |
6cc56833 JG |
845 | if (unlikely(require_extra_req)) |
846 | rinfo->shadow[extra_id].req = *extra_ring_req; | |
9f27ee59 | 847 | |
bd912ef3 | 848 | if (new_persistent_gnts) |
c004a6fe | 849 | gnttab_free_grant_references(setup.gref_head); |
9f27ee59 JF |
850 | |
851 | return 0; | |
852 | } | |
853 | ||
33204663 JG |
854 | /* |
855 | * Generate a Xen blkfront IO request from a blk layer request. Reads | |
856 | * and writes are handled as expected. | |
857 | * | |
858 | * @req: a request struct | |
859 | */ | |
81f35161 | 860 | static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo) |
33204663 | 861 | { |
81f35161 | 862 | if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED)) |
33204663 JG |
863 | return 1; |
864 | ||
c2df40df | 865 | if (unlikely(req_op(req) == REQ_OP_DISCARD || |
288dab8a | 866 | req_op(req) == REQ_OP_SECURE_ERASE)) |
81f35161 | 867 | return blkif_queue_discard_req(req, rinfo); |
33204663 | 868 | else |
81f35161 | 869 | return blkif_queue_rw_req(req, rinfo); |
33204663 | 870 | } |
9f27ee59 | 871 | |
81f35161 | 872 | static inline void flush_requests(struct blkfront_ring_info *rinfo) |
9f27ee59 JF |
873 | { |
874 | int notify; | |
875 | ||
81f35161 | 876 | RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify); |
9f27ee59 JF |
877 | |
878 | if (notify) | |
81f35161 | 879 | notify_remote_via_irq(rinfo->irq); |
9f27ee59 JF |
880 | } |
881 | ||
ad42d391 VK |
882 | static inline bool blkif_request_flush_invalid(struct request *req, |
883 | struct blkfront_info *info) | |
0f1ca65e | 884 | { |
aebf526b | 885 | return (blk_rq_is_passthrough(req) || |
3a5e02ce | 886 | ((req_op(req) == REQ_OP_FLUSH) && |
a418090a | 887 | !info->feature_flush) || |
ad42d391 | 888 | ((req->cmd_flags & REQ_FUA) && |
a418090a | 889 | !info->feature_fua)); |
0f1ca65e AA |
890 | } |
891 | ||
fc17b653 | 892 | static blk_status_t blkif_queue_rq(struct blk_mq_hw_ctx *hctx, |
6f03a7ff | 893 | const struct blk_mq_queue_data *qd) |
9f27ee59 | 894 | { |
11659569 | 895 | unsigned long flags; |
2a6f71ad BL |
896 | int qid = hctx->queue_num; |
897 | struct blkfront_info *info = hctx->queue->queuedata; | |
898 | struct blkfront_ring_info *rinfo = NULL; | |
9f27ee59 | 899 | |
4ab50af6 | 900 | rinfo = get_rinfo(info, qid); |
907c3eb1 | 901 | blk_mq_start_request(qd->rq); |
11659569 | 902 | spin_lock_irqsave(&rinfo->ring_lock, flags); |
81f35161 | 903 | if (RING_FULL(&rinfo->ring)) |
907c3eb1 | 904 | goto out_busy; |
9f27ee59 | 905 | |
81f35161 | 906 | if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info)) |
907c3eb1 | 907 | goto out_err; |
296b2f6a | 908 | |
81f35161 | 909 | if (blkif_queue_request(qd->rq, rinfo)) |
907c3eb1 | 910 | goto out_busy; |
296b2f6a | 911 | |
81f35161 | 912 | flush_requests(rinfo); |
11659569 | 913 | spin_unlock_irqrestore(&rinfo->ring_lock, flags); |
fc17b653 | 914 | return BLK_STS_OK; |
9f27ee59 | 915 | |
907c3eb1 | 916 | out_err: |
11659569 | 917 | spin_unlock_irqrestore(&rinfo->ring_lock, flags); |
fc17b653 | 918 | return BLK_STS_IOERR; |
9f27ee59 | 919 | |
907c3eb1 | 920 | out_busy: |
907c3eb1 | 921 | blk_mq_stop_hw_queue(hctx); |
4b422cb9 | 922 | spin_unlock_irqrestore(&rinfo->ring_lock, flags); |
86ff7c2a | 923 | return BLK_STS_DEV_RESOURCE; |
9f27ee59 JF |
924 | } |
925 | ||
2609587c CH |
926 | static void blkif_complete_rq(struct request *rq) |
927 | { | |
928 | blk_mq_end_request(rq, blkif_req(rq)->error); | |
929 | } | |
930 | ||
f363b089 | 931 | static const struct blk_mq_ops blkfront_mq_ops = { |
907c3eb1 | 932 | .queue_rq = blkif_queue_rq, |
2609587c | 933 | .complete = blkif_complete_rq, |
907c3eb1 BL |
934 | }; |
935 | ||
172335ad BL |
936 | static void blkif_set_queue_limits(struct blkfront_info *info) |
937 | { | |
938 | struct request_queue *rq = info->rq; | |
939 | struct gendisk *gd = info->gd; | |
940 | unsigned int segments = info->max_indirect_segments ? : | |
941 | BLKIF_MAX_SEGMENTS_PER_REQUEST; | |
942 | ||
8b904b5b | 943 | blk_queue_flag_set(QUEUE_FLAG_VIRT, rq); |
172335ad BL |
944 | |
945 | if (info->feature_discard) { | |
8b904b5b | 946 | blk_queue_flag_set(QUEUE_FLAG_DISCARD, rq); |
172335ad BL |
947 | blk_queue_max_discard_sectors(rq, get_capacity(gd)); |
948 | rq->limits.discard_granularity = info->discard_granularity; | |
949 | rq->limits.discard_alignment = info->discard_alignment; | |
950 | if (info->feature_secdiscard) | |
8b904b5b | 951 | blk_queue_flag_set(QUEUE_FLAG_SECERASE, rq); |
172335ad BL |
952 | } |
953 | ||
954 | /* Hard sector size and max sectors impersonate the equiv. hardware. */ | |
955 | blk_queue_logical_block_size(rq, info->sector_size); | |
956 | blk_queue_physical_block_size(rq, info->physical_sector_size); | |
957 | blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512); | |
958 | ||
959 | /* Each segment in a request is up to an aligned page in size. */ | |
960 | blk_queue_segment_boundary(rq, PAGE_SIZE - 1); | |
961 | blk_queue_max_segment_size(rq, PAGE_SIZE); | |
962 | ||
963 | /* Ensure a merged request will fit in a single I/O ring slot. */ | |
964 | blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG); | |
965 | ||
966 | /* Make sure buffer addresses are sector-aligned. */ | |
967 | blk_queue_dma_alignment(rq, 511); | |
172335ad BL |
968 | } |
969 | ||
402b27f9 | 970 | static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size, |
172335ad | 971 | unsigned int physical_sector_size) |
9f27ee59 | 972 | { |
165125e1 | 973 | struct request_queue *rq; |
ed30bf31 | 974 | struct blkfront_info *info = gd->private_data; |
9f27ee59 | 975 | |
907c3eb1 BL |
976 | memset(&info->tag_set, 0, sizeof(info->tag_set)); |
977 | info->tag_set.ops = &blkfront_mq_ops; | |
28d949bc | 978 | info->tag_set.nr_hw_queues = info->nr_rings; |
6cc56833 JG |
979 | if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) { |
980 | /* | |
981 | * When indirect descriptior is not supported, the I/O request | |
982 | * will be split between multiple request in the ring. | |
983 | * To avoid problems when sending the request, divide by | |
984 | * 2 the depth of the queue. | |
985 | */ | |
986 | info->tag_set.queue_depth = BLK_RING_SIZE(info) / 2; | |
987 | } else | |
988 | info->tag_set.queue_depth = BLK_RING_SIZE(info); | |
907c3eb1 | 989 | info->tag_set.numa_node = NUMA_NO_NODE; |
56d18f62 | 990 | info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE; |
2609587c | 991 | info->tag_set.cmd_size = sizeof(struct blkif_req); |
907c3eb1 BL |
992 | info->tag_set.driver_data = info; |
993 | ||
994 | if (blk_mq_alloc_tag_set(&info->tag_set)) | |
bde21f73 | 995 | return -EINVAL; |
907c3eb1 BL |
996 | rq = blk_mq_init_queue(&info->tag_set); |
997 | if (IS_ERR(rq)) { | |
998 | blk_mq_free_tag_set(&info->tag_set); | |
bde21f73 | 999 | return PTR_ERR(rq); |
907c3eb1 | 1000 | } |
9f27ee59 | 1001 | |
2a6f71ad | 1002 | rq->queuedata = info; |
172335ad BL |
1003 | info->rq = gd->queue = rq; |
1004 | info->gd = gd; | |
1005 | info->sector_size = sector_size; | |
1006 | info->physical_sector_size = physical_sector_size; | |
1007 | blkif_set_queue_limits(info); | |
9f27ee59 JF |
1008 | |
1009 | return 0; | |
1010 | } | |
1011 | ||
a418090a | 1012 | static const char *flush_info(struct blkfront_info *info) |
fdf9b965 | 1013 | { |
a418090a | 1014 | if (info->feature_flush && info->feature_fua) |
fdf9b965 | 1015 | return "barrier: enabled;"; |
a418090a | 1016 | else if (info->feature_flush) |
fdf9b965 | 1017 | return "flush diskcache: enabled;"; |
a418090a | 1018 | else |
fdf9b965 | 1019 | return "barrier or flush: disabled;"; |
fdf9b965 | 1020 | } |
9f27ee59 | 1021 | |
4913efe4 | 1022 | static void xlvbd_flush(struct blkfront_info *info) |
9f27ee59 | 1023 | { |
a418090a MC |
1024 | blk_queue_write_cache(info->rq, info->feature_flush ? true : false, |
1025 | info->feature_fua ? true : false); | |
fdf9b965 | 1026 | pr_info("blkfront: %s: %s %s %s %s %s\n", |
a418090a | 1027 | info->gd->disk_name, flush_info(info), |
fdf9b965 VK |
1028 | "persistent grants:", info->feature_persistent ? |
1029 | "enabled;" : "disabled;", "indirect descriptors:", | |
1030 | info->max_indirect_segments ? "enabled;" : "disabled;"); | |
9f27ee59 JF |
1031 | } |
1032 | ||
c80a4209 SS |
1033 | static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset) |
1034 | { | |
1035 | int major; | |
1036 | major = BLKIF_MAJOR(vdevice); | |
1037 | *minor = BLKIF_MINOR(vdevice); | |
1038 | switch (major) { | |
1039 | case XEN_IDE0_MAJOR: | |
1040 | *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET; | |
1041 | *minor = ((*minor / 64) * PARTS_PER_DISK) + | |
1042 | EMULATED_HD_DISK_MINOR_OFFSET; | |
1043 | break; | |
1044 | case XEN_IDE1_MAJOR: | |
1045 | *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET; | |
1046 | *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) + | |
1047 | EMULATED_HD_DISK_MINOR_OFFSET; | |
1048 | break; | |
1049 | case XEN_SCSI_DISK0_MAJOR: | |
1050 | *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET; | |
1051 | *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET; | |
1052 | break; | |
1053 | case XEN_SCSI_DISK1_MAJOR: | |
1054 | case XEN_SCSI_DISK2_MAJOR: | |
1055 | case XEN_SCSI_DISK3_MAJOR: | |
1056 | case XEN_SCSI_DISK4_MAJOR: | |
1057 | case XEN_SCSI_DISK5_MAJOR: | |
1058 | case XEN_SCSI_DISK6_MAJOR: | |
1059 | case XEN_SCSI_DISK7_MAJOR: | |
1060 | *offset = (*minor / PARTS_PER_DISK) + | |
1061 | ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) + | |
1062 | EMULATED_SD_DISK_NAME_OFFSET; | |
1063 | *minor = *minor + | |
1064 | ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) + | |
1065 | EMULATED_SD_DISK_MINOR_OFFSET; | |
1066 | break; | |
1067 | case XEN_SCSI_DISK8_MAJOR: | |
1068 | case XEN_SCSI_DISK9_MAJOR: | |
1069 | case XEN_SCSI_DISK10_MAJOR: | |
1070 | case XEN_SCSI_DISK11_MAJOR: | |
1071 | case XEN_SCSI_DISK12_MAJOR: | |
1072 | case XEN_SCSI_DISK13_MAJOR: | |
1073 | case XEN_SCSI_DISK14_MAJOR: | |
1074 | case XEN_SCSI_DISK15_MAJOR: | |
1075 | *offset = (*minor / PARTS_PER_DISK) + | |
1076 | ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) + | |
1077 | EMULATED_SD_DISK_NAME_OFFSET; | |
1078 | *minor = *minor + | |
1079 | ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) + | |
1080 | EMULATED_SD_DISK_MINOR_OFFSET; | |
1081 | break; | |
1082 | case XENVBD_MAJOR: | |
1083 | *offset = *minor / PARTS_PER_DISK; | |
1084 | break; | |
1085 | default: | |
1086 | printk(KERN_WARNING "blkfront: your disk configuration is " | |
1087 | "incorrect, please use an xvd device instead\n"); | |
1088 | return -ENODEV; | |
1089 | } | |
1090 | return 0; | |
1091 | } | |
9f27ee59 | 1092 | |
e77c78c0 JB |
1093 | static char *encode_disk_name(char *ptr, unsigned int n) |
1094 | { | |
1095 | if (n >= 26) | |
1096 | ptr = encode_disk_name(ptr, n / 26 - 1); | |
1097 | *ptr = 'a' + n % 26; | |
1098 | return ptr + 1; | |
1099 | } | |
1100 | ||
9246b5f0 CL |
1101 | static int xlvbd_alloc_gendisk(blkif_sector_t capacity, |
1102 | struct blkfront_info *info, | |
7c4d7d71 SB |
1103 | u16 vdisk_info, u16 sector_size, |
1104 | unsigned int physical_sector_size) | |
9f27ee59 JF |
1105 | { |
1106 | struct gendisk *gd; | |
1107 | int nr_minors = 1; | |
c80a4209 | 1108 | int err; |
9246b5f0 CL |
1109 | unsigned int offset; |
1110 | int minor; | |
1111 | int nr_parts; | |
e77c78c0 | 1112 | char *ptr; |
9f27ee59 JF |
1113 | |
1114 | BUG_ON(info->gd != NULL); | |
1115 | BUG_ON(info->rq != NULL); | |
1116 | ||
9246b5f0 CL |
1117 | if ((info->vdevice>>EXT_SHIFT) > 1) { |
1118 | /* this is above the extended range; something is wrong */ | |
1119 | printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice); | |
1120 | return -ENODEV; | |
1121 | } | |
1122 | ||
1123 | if (!VDEV_IS_EXTENDED(info->vdevice)) { | |
c80a4209 SS |
1124 | err = xen_translate_vdev(info->vdevice, &minor, &offset); |
1125 | if (err) | |
589b7289 NC |
1126 | return err; |
1127 | nr_parts = PARTS_PER_DISK; | |
9246b5f0 CL |
1128 | } else { |
1129 | minor = BLKIF_MINOR_EXT(info->vdevice); | |
1130 | nr_parts = PARTS_PER_EXT_DISK; | |
c80a4209 | 1131 | offset = minor / nr_parts; |
89153b5c | 1132 | if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4) |
c80a4209 SS |
1133 | printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with " |
1134 | "emulated IDE disks,\n\t choose an xvd device name" | |
1135 | "from xvde on\n", info->vdevice); | |
9246b5f0 | 1136 | } |
e77c78c0 JB |
1137 | if (minor >> MINORBITS) { |
1138 | pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n", | |
1139 | info->vdevice, minor); | |
1140 | return -ENODEV; | |
1141 | } | |
9246b5f0 CL |
1142 | |
1143 | if ((minor % nr_parts) == 0) | |
1144 | nr_minors = nr_parts; | |
9f27ee59 | 1145 | |
0e345826 JB |
1146 | err = xlbd_reserve_minors(minor, nr_minors); |
1147 | if (err) | |
1148 | goto out; | |
1149 | err = -ENODEV; | |
1150 | ||
9f27ee59 JF |
1151 | gd = alloc_disk(nr_minors); |
1152 | if (gd == NULL) | |
0e345826 | 1153 | goto release; |
9f27ee59 | 1154 | |
e77c78c0 JB |
1155 | strcpy(gd->disk_name, DEV_NAME); |
1156 | ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset); | |
1157 | BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN); | |
1158 | if (nr_minors > 1) | |
1159 | *ptr = 0; | |
1160 | else | |
1161 | snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr, | |
1162 | "%d", minor & (nr_parts - 1)); | |
9f27ee59 JF |
1163 | |
1164 | gd->major = XENVBD_MAJOR; | |
1165 | gd->first_minor = minor; | |
1166 | gd->fops = &xlvbd_block_fops; | |
1167 | gd->private_data = info; | |
9f27ee59 JF |
1168 | set_capacity(gd, capacity); |
1169 | ||
172335ad | 1170 | if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size)) { |
9f27ee59 | 1171 | del_gendisk(gd); |
0e345826 | 1172 | goto release; |
9f27ee59 JF |
1173 | } |
1174 | ||
4913efe4 | 1175 | xlvbd_flush(info); |
9f27ee59 JF |
1176 | |
1177 | if (vdisk_info & VDISK_READONLY) | |
1178 | set_disk_ro(gd, 1); | |
1179 | ||
1180 | if (vdisk_info & VDISK_REMOVABLE) | |
1181 | gd->flags |= GENHD_FL_REMOVABLE; | |
1182 | ||
1183 | if (vdisk_info & VDISK_CDROM) | |
1184 | gd->flags |= GENHD_FL_CD; | |
1185 | ||
1186 | return 0; | |
1187 | ||
0e345826 JB |
1188 | release: |
1189 | xlbd_release_minors(minor, nr_minors); | |
9f27ee59 JF |
1190 | out: |
1191 | return err; | |
1192 | } | |
1193 | ||
a66b5aeb DS |
1194 | static void xlvbd_release_gendisk(struct blkfront_info *info) |
1195 | { | |
3df0e505 | 1196 | unsigned int minor, nr_minors, i; |
4ab50af6 | 1197 | struct blkfront_ring_info *rinfo; |
a66b5aeb DS |
1198 | |
1199 | if (info->rq == NULL) | |
1200 | return; | |
1201 | ||
a66b5aeb | 1202 | /* No more blkif_request(). */ |
907c3eb1 | 1203 | blk_mq_stop_hw_queues(info->rq); |
a66b5aeb | 1204 | |
4ab50af6 | 1205 | for_each_rinfo(info, rinfo, i) { |
3df0e505 BL |
1206 | /* No more gnttab callback work. */ |
1207 | gnttab_cancel_free_callback(&rinfo->callback); | |
1208 | ||
1209 | /* Flush gnttab callback work. Must be done with no locks held. */ | |
1210 | flush_work(&rinfo->work); | |
1211 | } | |
a66b5aeb DS |
1212 | |
1213 | del_gendisk(info->gd); | |
1214 | ||
1215 | minor = info->gd->first_minor; | |
1216 | nr_minors = info->gd->minors; | |
1217 | xlbd_release_minors(minor, nr_minors); | |
1218 | ||
1219 | blk_cleanup_queue(info->rq); | |
907c3eb1 | 1220 | blk_mq_free_tag_set(&info->tag_set); |
a66b5aeb DS |
1221 | info->rq = NULL; |
1222 | ||
1223 | put_disk(info->gd); | |
1224 | info->gd = NULL; | |
1225 | } | |
1226 | ||
11659569 BL |
1227 | /* Already hold rinfo->ring_lock. */ |
1228 | static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo) | |
9f27ee59 | 1229 | { |
81f35161 BL |
1230 | if (!RING_FULL(&rinfo->ring)) |
1231 | blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true); | |
9f27ee59 JF |
1232 | } |
1233 | ||
11659569 BL |
1234 | static void kick_pending_request_queues(struct blkfront_ring_info *rinfo) |
1235 | { | |
1236 | unsigned long flags; | |
1237 | ||
1238 | spin_lock_irqsave(&rinfo->ring_lock, flags); | |
1239 | kick_pending_request_queues_locked(rinfo); | |
1240 | spin_unlock_irqrestore(&rinfo->ring_lock, flags); | |
1241 | } | |
1242 | ||
9f27ee59 JF |
1243 | static void blkif_restart_queue(struct work_struct *work) |
1244 | { | |
81f35161 | 1245 | struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work); |
9f27ee59 | 1246 | |
81f35161 BL |
1247 | if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED) |
1248 | kick_pending_request_queues(rinfo); | |
9f27ee59 JF |
1249 | } |
1250 | ||
3df0e505 | 1251 | static void blkif_free_ring(struct blkfront_ring_info *rinfo) |
9f27ee59 | 1252 | { |
73716df7 | 1253 | struct grant *persistent_gnt, *n; |
3df0e505 | 1254 | struct blkfront_info *info = rinfo->dev_info; |
402b27f9 | 1255 | int i, j, segs; |
0a8704a5 | 1256 | |
bfe11d6d RPM |
1257 | /* |
1258 | * Remove indirect pages, this only happens when using indirect | |
1259 | * descriptors but not persistent grants | |
1260 | */ | |
81f35161 | 1261 | if (!list_empty(&rinfo->indirect_pages)) { |
bfe11d6d RPM |
1262 | struct page *indirect_page, *n; |
1263 | ||
1264 | BUG_ON(info->feature_persistent); | |
81f35161 | 1265 | list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) { |
bfe11d6d RPM |
1266 | list_del(&indirect_page->lru); |
1267 | __free_page(indirect_page); | |
1268 | } | |
1269 | } | |
1270 | ||
73716df7 BL |
1271 | /* Remove all persistent grants. */ |
1272 | if (!list_empty(&rinfo->grants)) { | |
1273 | list_for_each_entry_safe(persistent_gnt, n, | |
1274 | &rinfo->grants, node) { | |
1275 | list_del(&persistent_gnt->node); | |
1276 | if (persistent_gnt->gref != GRANT_INVALID_REF) { | |
1277 | gnttab_end_foreign_access(persistent_gnt->gref, | |
1278 | 0, 0UL); | |
1279 | rinfo->persistent_gnts_c--; | |
1280 | } | |
1281 | if (info->feature_persistent) | |
1282 | __free_page(persistent_gnt->page); | |
1283 | kfree(persistent_gnt); | |
1284 | } | |
1285 | } | |
1286 | BUG_ON(rinfo->persistent_gnts_c != 0); | |
1287 | ||
86839c56 | 1288 | for (i = 0; i < BLK_RING_SIZE(info); i++) { |
402b27f9 RPM |
1289 | /* |
1290 | * Clear persistent grants present in requests already | |
1291 | * on the shared ring | |
1292 | */ | |
81f35161 | 1293 | if (!rinfo->shadow[i].request) |
402b27f9 RPM |
1294 | goto free_shadow; |
1295 | ||
81f35161 BL |
1296 | segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ? |
1297 | rinfo->shadow[i].req.u.indirect.nr_segments : | |
1298 | rinfo->shadow[i].req.u.rw.nr_segments; | |
402b27f9 | 1299 | for (j = 0; j < segs; j++) { |
81f35161 | 1300 | persistent_gnt = rinfo->shadow[i].grants_used[j]; |
402b27f9 | 1301 | gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL); |
bfe11d6d | 1302 | if (info->feature_persistent) |
a7a6df22 | 1303 | __free_page(persistent_gnt->page); |
402b27f9 RPM |
1304 | kfree(persistent_gnt); |
1305 | } | |
1306 | ||
81f35161 | 1307 | if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT) |
402b27f9 RPM |
1308 | /* |
1309 | * If this is not an indirect operation don't try to | |
1310 | * free indirect segments | |
1311 | */ | |
1312 | goto free_shadow; | |
1313 | ||
1314 | for (j = 0; j < INDIRECT_GREFS(segs); j++) { | |
81f35161 | 1315 | persistent_gnt = rinfo->shadow[i].indirect_grants[j]; |
402b27f9 | 1316 | gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL); |
a7a6df22 | 1317 | __free_page(persistent_gnt->page); |
402b27f9 RPM |
1318 | kfree(persistent_gnt); |
1319 | } | |
1320 | ||
1321 | free_shadow: | |
1d5c76e6 | 1322 | kvfree(rinfo->shadow[i].grants_used); |
81f35161 | 1323 | rinfo->shadow[i].grants_used = NULL; |
1d5c76e6 | 1324 | kvfree(rinfo->shadow[i].indirect_grants); |
81f35161 | 1325 | rinfo->shadow[i].indirect_grants = NULL; |
1d5c76e6 | 1326 | kvfree(rinfo->shadow[i].sg); |
81f35161 | 1327 | rinfo->shadow[i].sg = NULL; |
402b27f9 RPM |
1328 | } |
1329 | ||
9f27ee59 | 1330 | /* No more gnttab callback work. */ |
81f35161 | 1331 | gnttab_cancel_free_callback(&rinfo->callback); |
9f27ee59 JF |
1332 | |
1333 | /* Flush gnttab callback work. Must be done with no locks held. */ | |
81f35161 | 1334 | flush_work(&rinfo->work); |
9f27ee59 JF |
1335 | |
1336 | /* Free resources associated with old device channel. */ | |
86839c56 | 1337 | for (i = 0; i < info->nr_ring_pages; i++) { |
81f35161 BL |
1338 | if (rinfo->ring_ref[i] != GRANT_INVALID_REF) { |
1339 | gnttab_end_foreign_access(rinfo->ring_ref[i], 0, 0); | |
1340 | rinfo->ring_ref[i] = GRANT_INVALID_REF; | |
86839c56 | 1341 | } |
9f27ee59 | 1342 | } |
6c647b0e | 1343 | free_pages((unsigned long)rinfo->ring.sring, get_order(info->nr_ring_pages * XEN_PAGE_SIZE)); |
81f35161 | 1344 | rinfo->ring.sring = NULL; |
86839c56 | 1345 | |
81f35161 BL |
1346 | if (rinfo->irq) |
1347 | unbind_from_irqhandler(rinfo->irq, rinfo); | |
1348 | rinfo->evtchn = rinfo->irq = 0; | |
3df0e505 | 1349 | } |
9f27ee59 | 1350 | |
3df0e505 BL |
1351 | static void blkif_free(struct blkfront_info *info, int suspend) |
1352 | { | |
3df0e505 | 1353 | unsigned int i; |
4ab50af6 | 1354 | struct blkfront_ring_info *rinfo; |
3df0e505 BL |
1355 | |
1356 | /* Prevent new requests being issued until we fix things up. */ | |
3df0e505 BL |
1357 | info->connected = suspend ? |
1358 | BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED; | |
1359 | /* No more blkif_request(). */ | |
1360 | if (info->rq) | |
1361 | blk_mq_stop_hw_queues(info->rq); | |
1362 | ||
4ab50af6 JG |
1363 | for_each_rinfo(info, rinfo, i) |
1364 | blkif_free_ring(rinfo); | |
3df0e505 | 1365 | |
1d5c76e6 | 1366 | kvfree(info->rinfo); |
3df0e505 BL |
1367 | info->rinfo = NULL; |
1368 | info->nr_rings = 0; | |
9f27ee59 JF |
1369 | } |
1370 | ||
c004a6fe JG |
1371 | struct copy_from_grant { |
1372 | const struct blk_shadow *s; | |
1373 | unsigned int grant_idx; | |
1374 | unsigned int bvec_offset; | |
1375 | char *bvec_data; | |
1376 | }; | |
1377 | ||
1378 | static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset, | |
1379 | unsigned int len, void *data) | |
1380 | { | |
1381 | struct copy_from_grant *info = data; | |
1382 | char *shared_data; | |
1383 | /* Convenient aliases */ | |
1384 | const struct blk_shadow *s = info->s; | |
1385 | ||
1386 | shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page); | |
1387 | ||
1388 | memcpy(info->bvec_data + info->bvec_offset, | |
1389 | shared_data + offset, len); | |
1390 | ||
1391 | info->bvec_offset += len; | |
1392 | info->grant_idx++; | |
1393 | ||
1394 | kunmap_atomic(shared_data); | |
1395 | } | |
1396 | ||
6cc56833 JG |
1397 | static enum blk_req_status blkif_rsp_to_req_status(int rsp) |
1398 | { | |
1399 | switch (rsp) | |
1400 | { | |
1401 | case BLKIF_RSP_OKAY: | |
1402 | return REQ_DONE; | |
1403 | case BLKIF_RSP_EOPNOTSUPP: | |
1404 | return REQ_EOPNOTSUPP; | |
1405 | case BLKIF_RSP_ERROR: | |
6cc56833 JG |
1406 | default: |
1407 | return REQ_ERROR; | |
1408 | } | |
1409 | } | |
1410 | ||
1411 | /* | |
1412 | * Get the final status of the block request based on two ring response | |
1413 | */ | |
1414 | static int blkif_get_final_status(enum blk_req_status s1, | |
1415 | enum blk_req_status s2) | |
1416 | { | |
1417 | BUG_ON(s1 == REQ_WAITING); | |
1418 | BUG_ON(s2 == REQ_WAITING); | |
1419 | ||
1420 | if (s1 == REQ_ERROR || s2 == REQ_ERROR) | |
1421 | return BLKIF_RSP_ERROR; | |
1422 | else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP) | |
1423 | return BLKIF_RSP_EOPNOTSUPP; | |
1424 | return BLKIF_RSP_OKAY; | |
1425 | } | |
1426 | ||
1427 | static bool blkif_completion(unsigned long *id, | |
1428 | struct blkfront_ring_info *rinfo, | |
0a8704a5 | 1429 | struct blkif_response *bret) |
9f27ee59 | 1430 | { |
d62f6918 | 1431 | int i = 0; |
b7649158 | 1432 | struct scatterlist *sg; |
c004a6fe | 1433 | int num_sg, num_grant; |
81f35161 | 1434 | struct blkfront_info *info = rinfo->dev_info; |
6cc56833 | 1435 | struct blk_shadow *s = &rinfo->shadow[*id]; |
c004a6fe | 1436 | struct copy_from_grant data = { |
c004a6fe JG |
1437 | .grant_idx = 0, |
1438 | }; | |
402b27f9 | 1439 | |
c004a6fe | 1440 | num_grant = s->req.operation == BLKIF_OP_INDIRECT ? |
402b27f9 | 1441 | s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments; |
6cc56833 JG |
1442 | |
1443 | /* The I/O request may be split in two. */ | |
1444 | if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) { | |
1445 | struct blk_shadow *s2 = &rinfo->shadow[s->associated_id]; | |
1446 | ||
1447 | /* Keep the status of the current response in shadow. */ | |
1448 | s->status = blkif_rsp_to_req_status(bret->status); | |
1449 | ||
1450 | /* Wait the second response if not yet here. */ | |
1451 | if (s2->status == REQ_WAITING) | |
f87c30c9 | 1452 | return false; |
6cc56833 JG |
1453 | |
1454 | bret->status = blkif_get_final_status(s->status, | |
1455 | s2->status); | |
1456 | ||
1457 | /* | |
1458 | * All the grants is stored in the first shadow in order | |
1459 | * to make the completion code simpler. | |
1460 | */ | |
1461 | num_grant += s2->req.u.rw.nr_segments; | |
1462 | ||
1463 | /* | |
1464 | * The two responses may not come in order. Only the | |
1465 | * first request will store the scatter-gather list. | |
1466 | */ | |
1467 | if (s2->num_sg != 0) { | |
1468 | /* Update "id" with the ID of the first response. */ | |
1469 | *id = s->associated_id; | |
1470 | s = s2; | |
1471 | } | |
1472 | ||
1473 | /* | |
1474 | * We don't need anymore the second request, so recycling | |
1475 | * it now. | |
1476 | */ | |
1477 | if (add_id_to_freelist(rinfo, s->associated_id)) | |
1478 | WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n", | |
1479 | info->gd->disk_name, s->associated_id); | |
1480 | } | |
1481 | ||
1482 | data.s = s; | |
c004a6fe | 1483 | num_sg = s->num_sg; |
0a8704a5 | 1484 | |
bfe11d6d | 1485 | if (bret->operation == BLKIF_OP_READ && info->feature_persistent) { |
c004a6fe | 1486 | for_each_sg(s->sg, sg, num_sg, i) { |
b7649158 | 1487 | BUG_ON(sg->offset + sg->length > PAGE_SIZE); |
c004a6fe JG |
1488 | |
1489 | data.bvec_offset = sg->offset; | |
1490 | data.bvec_data = kmap_atomic(sg_page(sg)); | |
1491 | ||
1492 | gnttab_foreach_grant_in_range(sg_page(sg), | |
1493 | sg->offset, | |
1494 | sg->length, | |
1495 | blkif_copy_from_grant, | |
1496 | &data); | |
1497 | ||
1498 | kunmap_atomic(data.bvec_data); | |
0a8704a5 RPM |
1499 | } |
1500 | } | |
1501 | /* Add the persistent grant into the list of free grants */ | |
c004a6fe | 1502 | for (i = 0; i < num_grant; i++) { |
fbe363c4 RPM |
1503 | if (gnttab_query_foreign_access(s->grants_used[i]->gref)) { |
1504 | /* | |
1505 | * If the grant is still mapped by the backend (the | |
1506 | * backend has chosen to make this grant persistent) | |
1507 | * we add it at the head of the list, so it will be | |
1508 | * reused first. | |
1509 | */ | |
bfe11d6d RPM |
1510 | if (!info->feature_persistent) |
1511 | pr_alert_ratelimited("backed has not unmapped grant: %u\n", | |
1512 | s->grants_used[i]->gref); | |
73716df7 BL |
1513 | list_add(&s->grants_used[i]->node, &rinfo->grants); |
1514 | rinfo->persistent_gnts_c++; | |
fbe363c4 RPM |
1515 | } else { |
1516 | /* | |
1517 | * If the grant is not mapped by the backend we end the | |
1518 | * foreign access and add it to the tail of the list, | |
1519 | * so it will not be picked again unless we run out of | |
1520 | * persistent grants. | |
1521 | */ | |
1522 | gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL); | |
1523 | s->grants_used[i]->gref = GRANT_INVALID_REF; | |
73716df7 | 1524 | list_add_tail(&s->grants_used[i]->node, &rinfo->grants); |
fbe363c4 | 1525 | } |
0a8704a5 | 1526 | } |
402b27f9 | 1527 | if (s->req.operation == BLKIF_OP_INDIRECT) { |
c004a6fe | 1528 | for (i = 0; i < INDIRECT_GREFS(num_grant); i++) { |
fbe363c4 | 1529 | if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) { |
bfe11d6d RPM |
1530 | if (!info->feature_persistent) |
1531 | pr_alert_ratelimited("backed has not unmapped grant: %u\n", | |
1532 | s->indirect_grants[i]->gref); | |
73716df7 BL |
1533 | list_add(&s->indirect_grants[i]->node, &rinfo->grants); |
1534 | rinfo->persistent_gnts_c++; | |
fbe363c4 | 1535 | } else { |
bfe11d6d RPM |
1536 | struct page *indirect_page; |
1537 | ||
fbe363c4 | 1538 | gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL); |
bfe11d6d RPM |
1539 | /* |
1540 | * Add the used indirect page back to the list of | |
1541 | * available pages for indirect grefs. | |
1542 | */ | |
7b076750 | 1543 | if (!info->feature_persistent) { |
a7a6df22 | 1544 | indirect_page = s->indirect_grants[i]->page; |
81f35161 | 1545 | list_add(&indirect_page->lru, &rinfo->indirect_pages); |
7b076750 | 1546 | } |
fbe363c4 | 1547 | s->indirect_grants[i]->gref = GRANT_INVALID_REF; |
73716df7 | 1548 | list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants); |
fbe363c4 | 1549 | } |
402b27f9 RPM |
1550 | } |
1551 | } | |
6cc56833 | 1552 | |
f87c30c9 | 1553 | return true; |
9f27ee59 JF |
1554 | } |
1555 | ||
1556 | static irqreturn_t blkif_interrupt(int irq, void *dev_id) | |
1557 | { | |
1558 | struct request *req; | |
1559 | struct blkif_response *bret; | |
1560 | RING_IDX i, rp; | |
1561 | unsigned long flags; | |
81f35161 BL |
1562 | struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id; |
1563 | struct blkfront_info *info = rinfo->dev_info; | |
9f27ee59 | 1564 | |
11659569 | 1565 | if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) |
9f27ee59 | 1566 | return IRQ_HANDLED; |
9f27ee59 | 1567 | |
11659569 | 1568 | spin_lock_irqsave(&rinfo->ring_lock, flags); |
9f27ee59 | 1569 | again: |
81f35161 | 1570 | rp = rinfo->ring.sring->rsp_prod; |
9f27ee59 JF |
1571 | rmb(); /* Ensure we see queued responses up to 'rp'. */ |
1572 | ||
81f35161 | 1573 | for (i = rinfo->ring.rsp_cons; i != rp; i++) { |
9f27ee59 | 1574 | unsigned long id; |
9f27ee59 | 1575 | |
81f35161 | 1576 | bret = RING_GET_RESPONSE(&rinfo->ring, i); |
9f27ee59 | 1577 | id = bret->id; |
6878c32e KRW |
1578 | /* |
1579 | * The backend has messed up and given us an id that we would | |
1580 | * never have given to it (we stamp it up to BLK_RING_SIZE - | |
1581 | * look in get_id_from_freelist. | |
1582 | */ | |
86839c56 | 1583 | if (id >= BLK_RING_SIZE(info)) { |
6878c32e KRW |
1584 | WARN(1, "%s: response to %s has incorrect id (%ld)\n", |
1585 | info->gd->disk_name, op_name(bret->operation), id); | |
1586 | /* We can't safely get the 'struct request' as | |
1587 | * the id is busted. */ | |
1588 | continue; | |
1589 | } | |
81f35161 | 1590 | req = rinfo->shadow[id].request; |
9f27ee59 | 1591 | |
6cc56833 JG |
1592 | if (bret->operation != BLKIF_OP_DISCARD) { |
1593 | /* | |
1594 | * We may need to wait for an extra response if the | |
1595 | * I/O request is split in 2 | |
1596 | */ | |
1597 | if (!blkif_completion(&id, rinfo, bret)) | |
1598 | continue; | |
1599 | } | |
9f27ee59 | 1600 | |
81f35161 | 1601 | if (add_id_to_freelist(rinfo, id)) { |
6878c32e KRW |
1602 | WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n", |
1603 | info->gd->disk_name, op_name(bret->operation), id); | |
1604 | continue; | |
1605 | } | |
9f27ee59 | 1606 | |
2a842aca CH |
1607 | if (bret->status == BLKIF_RSP_OKAY) |
1608 | blkif_req(req)->error = BLK_STS_OK; | |
1609 | else | |
1610 | blkif_req(req)->error = BLK_STS_IOERR; | |
1611 | ||
9f27ee59 | 1612 | switch (bret->operation) { |
ed30bf31 LD |
1613 | case BLKIF_OP_DISCARD: |
1614 | if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) { | |
1615 | struct request_queue *rq = info->rq; | |
6878c32e KRW |
1616 | printk(KERN_WARNING "blkfront: %s: %s op failed\n", |
1617 | info->gd->disk_name, op_name(bret->operation)); | |
2a842aca | 1618 | blkif_req(req)->error = BLK_STS_NOTSUPP; |
ed30bf31 | 1619 | info->feature_discard = 0; |
5ea42986 | 1620 | info->feature_secdiscard = 0; |
8b904b5b BVA |
1621 | blk_queue_flag_clear(QUEUE_FLAG_DISCARD, rq); |
1622 | blk_queue_flag_clear(QUEUE_FLAG_SECERASE, rq); | |
ed30bf31 | 1623 | } |
ed30bf31 | 1624 | break; |
edf6ef59 | 1625 | case BLKIF_OP_FLUSH_DISKCACHE: |
9f27ee59 JF |
1626 | case BLKIF_OP_WRITE_BARRIER: |
1627 | if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) { | |
6878c32e KRW |
1628 | printk(KERN_WARNING "blkfront: %s: %s op failed\n", |
1629 | info->gd->disk_name, op_name(bret->operation)); | |
31c4ccc3 | 1630 | blkif_req(req)->error = BLK_STS_NOTSUPP; |
dcb8baec JF |
1631 | } |
1632 | if (unlikely(bret->status == BLKIF_RSP_ERROR && | |
81f35161 | 1633 | rinfo->shadow[id].req.u.rw.nr_segments == 0)) { |
6878c32e KRW |
1634 | printk(KERN_WARNING "blkfront: %s: empty %s op failed\n", |
1635 | info->gd->disk_name, op_name(bret->operation)); | |
2a842aca | 1636 | blkif_req(req)->error = BLK_STS_NOTSUPP; |
dcb8baec | 1637 | } |
2609587c | 1638 | if (unlikely(blkif_req(req)->error)) { |
2a842aca CH |
1639 | if (blkif_req(req)->error == BLK_STS_NOTSUPP) |
1640 | blkif_req(req)->error = BLK_STS_OK; | |
a418090a | 1641 | info->feature_fua = 0; |
4913efe4 TH |
1642 | info->feature_flush = 0; |
1643 | xlvbd_flush(info); | |
9f27ee59 | 1644 | } |
df561f66 | 1645 | fallthrough; |
9f27ee59 JF |
1646 | case BLKIF_OP_READ: |
1647 | case BLKIF_OP_WRITE: | |
1648 | if (unlikely(bret->status != BLKIF_RSP_OKAY)) | |
1649 | dev_dbg(&info->xbdev->dev, "Bad return from blkdev data " | |
1650 | "request: %x\n", bret->status); | |
1651 | ||
9f27ee59 JF |
1652 | break; |
1653 | default: | |
1654 | BUG(); | |
1655 | } | |
2609587c | 1656 | |
15f73f5b CH |
1657 | if (likely(!blk_should_fake_timeout(req->q))) |
1658 | blk_mq_complete_request(req); | |
9f27ee59 JF |
1659 | } |
1660 | ||
81f35161 | 1661 | rinfo->ring.rsp_cons = i; |
9f27ee59 | 1662 | |
81f35161 | 1663 | if (i != rinfo->ring.req_prod_pvt) { |
9f27ee59 | 1664 | int more_to_do; |
81f35161 | 1665 | RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do); |
9f27ee59 JF |
1666 | if (more_to_do) |
1667 | goto again; | |
1668 | } else | |
81f35161 | 1669 | rinfo->ring.sring->rsp_event = i + 1; |
9f27ee59 | 1670 | |
11659569 | 1671 | kick_pending_request_queues_locked(rinfo); |
9f27ee59 | 1672 | |
11659569 | 1673 | spin_unlock_irqrestore(&rinfo->ring_lock, flags); |
9f27ee59 JF |
1674 | |
1675 | return IRQ_HANDLED; | |
1676 | } | |
1677 | ||
1678 | ||
1679 | static int setup_blkring(struct xenbus_device *dev, | |
81f35161 | 1680 | struct blkfront_ring_info *rinfo) |
9f27ee59 JF |
1681 | { |
1682 | struct blkif_sring *sring; | |
86839c56 | 1683 | int err, i; |
81f35161 | 1684 | struct blkfront_info *info = rinfo->dev_info; |
c004a6fe | 1685 | unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE; |
9cce2914 | 1686 | grant_ref_t gref[XENBUS_MAX_RING_GRANTS]; |
9f27ee59 | 1687 | |
86839c56 | 1688 | for (i = 0; i < info->nr_ring_pages; i++) |
81f35161 | 1689 | rinfo->ring_ref[i] = GRANT_INVALID_REF; |
9f27ee59 | 1690 | |
86839c56 BL |
1691 | sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH, |
1692 | get_order(ring_size)); | |
9f27ee59 JF |
1693 | if (!sring) { |
1694 | xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring"); | |
1695 | return -ENOMEM; | |
1696 | } | |
1697 | SHARED_RING_INIT(sring); | |
81f35161 | 1698 | FRONT_RING_INIT(&rinfo->ring, sring, ring_size); |
9e973e64 | 1699 | |
81f35161 | 1700 | err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref); |
9f27ee59 | 1701 | if (err < 0) { |
86839c56 | 1702 | free_pages((unsigned long)sring, get_order(ring_size)); |
81f35161 | 1703 | rinfo->ring.sring = NULL; |
9f27ee59 JF |
1704 | goto fail; |
1705 | } | |
86839c56 | 1706 | for (i = 0; i < info->nr_ring_pages; i++) |
81f35161 | 1707 | rinfo->ring_ref[i] = gref[i]; |
9f27ee59 | 1708 | |
81f35161 | 1709 | err = xenbus_alloc_evtchn(dev, &rinfo->evtchn); |
9f27ee59 JF |
1710 | if (err) |
1711 | goto fail; | |
1712 | ||
81f35161 BL |
1713 | err = bind_evtchn_to_irqhandler(rinfo->evtchn, blkif_interrupt, 0, |
1714 | "blkif", rinfo); | |
9f27ee59 JF |
1715 | if (err <= 0) { |
1716 | xenbus_dev_fatal(dev, err, | |
1717 | "bind_evtchn_to_irqhandler failed"); | |
1718 | goto fail; | |
1719 | } | |
81f35161 | 1720 | rinfo->irq = err; |
9f27ee59 JF |
1721 | |
1722 | return 0; | |
1723 | fail: | |
1724 | blkif_free(info, 0); | |
1725 | return err; | |
1726 | } | |
1727 | ||
28d949bc BL |
1728 | /* |
1729 | * Write out per-ring/queue nodes including ring-ref and event-channel, and each | |
1730 | * ring buffer may have multi pages depending on ->nr_ring_pages. | |
1731 | */ | |
1732 | static int write_per_ring_nodes(struct xenbus_transaction xbt, | |
1733 | struct blkfront_ring_info *rinfo, const char *dir) | |
1734 | { | |
1735 | int err; | |
1736 | unsigned int i; | |
1737 | const char *message = NULL; | |
1738 | struct blkfront_info *info = rinfo->dev_info; | |
1739 | ||
1740 | if (info->nr_ring_pages == 1) { | |
1741 | err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]); | |
1742 | if (err) { | |
1743 | message = "writing ring-ref"; | |
1744 | goto abort_transaction; | |
1745 | } | |
1746 | } else { | |
1747 | for (i = 0; i < info->nr_ring_pages; i++) { | |
1748 | char ring_ref_name[RINGREF_NAME_LEN]; | |
1749 | ||
1750 | snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i); | |
1751 | err = xenbus_printf(xbt, dir, ring_ref_name, | |
1752 | "%u", rinfo->ring_ref[i]); | |
1753 | if (err) { | |
1754 | message = "writing ring-ref"; | |
1755 | goto abort_transaction; | |
1756 | } | |
1757 | } | |
1758 | } | |
1759 | ||
1760 | err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn); | |
1761 | if (err) { | |
1762 | message = "writing event-channel"; | |
1763 | goto abort_transaction; | |
1764 | } | |
1765 | ||
1766 | return 0; | |
1767 | ||
1768 | abort_transaction: | |
1769 | xenbus_transaction_end(xbt, 1); | |
1770 | if (message) | |
1771 | xenbus_dev_fatal(info->xbdev, err, "%s", message); | |
1772 | ||
1773 | return err; | |
1774 | } | |
9f27ee59 | 1775 | |
a46b5367 JG |
1776 | static void free_info(struct blkfront_info *info) |
1777 | { | |
1778 | list_del(&info->info_list); | |
1779 | kfree(info); | |
1780 | } | |
1781 | ||
9f27ee59 | 1782 | /* Common code used when first setting up, and when resuming. */ |
203fd61f | 1783 | static int talk_to_blkback(struct xenbus_device *dev, |
9f27ee59 JF |
1784 | struct blkfront_info *info) |
1785 | { | |
1786 | const char *message = NULL; | |
1787 | struct xenbus_transaction xbt; | |
28d949bc | 1788 | int err; |
f27dc1ac JG |
1789 | unsigned int i, max_page_order; |
1790 | unsigned int ring_page_order; | |
4ab50af6 | 1791 | struct blkfront_ring_info *rinfo; |
86839c56 | 1792 | |
7ed8ce1c BD |
1793 | if (!info) |
1794 | return -ENODEV; | |
1795 | ||
f27dc1ac JG |
1796 | max_page_order = xenbus_read_unsigned(info->xbdev->otherend, |
1797 | "max-ring-page-order", 0); | |
1798 | ring_page_order = min(xen_blkif_max_ring_order, max_page_order); | |
1799 | info->nr_ring_pages = 1 << ring_page_order; | |
9f27ee59 | 1800 | |
7ed8ce1c BD |
1801 | err = negotiate_mq(info); |
1802 | if (err) | |
1803 | goto destroy_blkring; | |
1804 | ||
4ab50af6 | 1805 | for_each_rinfo(info, rinfo, i) { |
3df0e505 BL |
1806 | /* Create shared ring, alloc event channel. */ |
1807 | err = setup_blkring(dev, rinfo); | |
1808 | if (err) | |
1809 | goto destroy_blkring; | |
1810 | } | |
9f27ee59 JF |
1811 | |
1812 | again: | |
1813 | err = xenbus_transaction_start(&xbt); | |
1814 | if (err) { | |
1815 | xenbus_dev_fatal(dev, err, "starting transaction"); | |
1816 | goto destroy_blkring; | |
1817 | } | |
1818 | ||
28d949bc BL |
1819 | if (info->nr_ring_pages > 1) { |
1820 | err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u", | |
1821 | ring_page_order); | |
1822 | if (err) { | |
1823 | message = "writing ring-page-order"; | |
1824 | goto abort_transaction; | |
1825 | } | |
1826 | } | |
3df0e505 | 1827 | |
28d949bc BL |
1828 | /* We already got the number of queues/rings in _probe */ |
1829 | if (info->nr_rings == 1) { | |
4ab50af6 | 1830 | err = write_per_ring_nodes(xbt, info->rinfo, dev->nodename); |
28d949bc BL |
1831 | if (err) |
1832 | goto destroy_blkring; | |
1833 | } else { | |
1834 | char *path; | |
1835 | size_t pathsize; | |
3df0e505 | 1836 | |
28d949bc BL |
1837 | err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u", |
1838 | info->nr_rings); | |
3df0e505 | 1839 | if (err) { |
28d949bc | 1840 | message = "writing multi-queue-num-queues"; |
3df0e505 BL |
1841 | goto abort_transaction; |
1842 | } | |
28d949bc BL |
1843 | |
1844 | pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN; | |
1845 | path = kmalloc(pathsize, GFP_KERNEL); | |
1846 | if (!path) { | |
1847 | err = -ENOMEM; | |
1848 | message = "ENOMEM while writing ring references"; | |
1849 | goto abort_transaction; | |
1850 | } | |
1851 | ||
4ab50af6 | 1852 | for_each_rinfo(info, rinfo, i) { |
28d949bc BL |
1853 | memset(path, 0, pathsize); |
1854 | snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i); | |
4ab50af6 | 1855 | err = write_per_ring_nodes(xbt, rinfo, path); |
28d949bc BL |
1856 | if (err) { |
1857 | kfree(path); | |
1858 | goto destroy_blkring; | |
1859 | } | |
1860 | } | |
1861 | kfree(path); | |
9f27ee59 | 1862 | } |
3e334239 MA |
1863 | err = xenbus_printf(xbt, dev->nodename, "protocol", "%s", |
1864 | XEN_IO_PROTO_ABI_NATIVE); | |
1865 | if (err) { | |
1866 | message = "writing protocol"; | |
1867 | goto abort_transaction; | |
1868 | } | |
0a8704a5 | 1869 | err = xenbus_printf(xbt, dev->nodename, |
cb5bd4d1 | 1870 | "feature-persistent", "%u", 1); |
0a8704a5 RPM |
1871 | if (err) |
1872 | dev_warn(&dev->dev, | |
1873 | "writing persistent grants feature to xenbus"); | |
9f27ee59 JF |
1874 | |
1875 | err = xenbus_transaction_end(xbt, 0); | |
1876 | if (err) { | |
1877 | if (err == -EAGAIN) | |
1878 | goto again; | |
1879 | xenbus_dev_fatal(dev, err, "completing transaction"); | |
1880 | goto destroy_blkring; | |
1881 | } | |
1882 | ||
4ab50af6 | 1883 | for_each_rinfo(info, rinfo, i) { |
3df0e505 | 1884 | unsigned int j; |
3df0e505 BL |
1885 | |
1886 | for (j = 0; j < BLK_RING_SIZE(info); j++) | |
1887 | rinfo->shadow[j].req.u.rw.id = j + 1; | |
1888 | rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff; | |
1889 | } | |
9f27ee59 JF |
1890 | xenbus_switch_state(dev, XenbusStateInitialised); |
1891 | ||
1892 | return 0; | |
1893 | ||
1894 | abort_transaction: | |
1895 | xenbus_transaction_end(xbt, 1); | |
1896 | if (message) | |
1897 | xenbus_dev_fatal(dev, err, "%s", message); | |
1898 | destroy_blkring: | |
1899 | blkif_free(info, 0); | |
3df0e505 | 1900 | |
a46b5367 JG |
1901 | mutex_lock(&blkfront_mutex); |
1902 | free_info(info); | |
1903 | mutex_unlock(&blkfront_mutex); | |
1904 | ||
c31ecf6c KRW |
1905 | dev_set_drvdata(&dev->dev, NULL); |
1906 | ||
9f27ee59 JF |
1907 | return err; |
1908 | } | |
1909 | ||
3db70a85 BL |
1910 | static int negotiate_mq(struct blkfront_info *info) |
1911 | { | |
f27dc1ac | 1912 | unsigned int backend_max_queues; |
3db70a85 | 1913 | unsigned int i; |
4ab50af6 | 1914 | struct blkfront_ring_info *rinfo; |
3db70a85 BL |
1915 | |
1916 | BUG_ON(info->nr_rings); | |
1917 | ||
1918 | /* Check if backend supports multiple queues. */ | |
f27dc1ac JG |
1919 | backend_max_queues = xenbus_read_unsigned(info->xbdev->otherend, |
1920 | "multi-queue-max-queues", 1); | |
3db70a85 BL |
1921 | info->nr_rings = min(backend_max_queues, xen_blkif_max_queues); |
1922 | /* We need at least one ring. */ | |
1923 | if (!info->nr_rings) | |
1924 | info->nr_rings = 1; | |
1925 | ||
4ab50af6 JG |
1926 | info->rinfo_size = struct_size(info->rinfo, shadow, |
1927 | BLK_RING_SIZE(info)); | |
1928 | info->rinfo = kvcalloc(info->nr_rings, info->rinfo_size, GFP_KERNEL); | |
3db70a85 BL |
1929 | if (!info->rinfo) { |
1930 | xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure"); | |
6cc4a086 | 1931 | info->nr_rings = 0; |
3db70a85 BL |
1932 | return -ENOMEM; |
1933 | } | |
1934 | ||
4ab50af6 | 1935 | for_each_rinfo(info, rinfo, i) { |
3db70a85 BL |
1936 | INIT_LIST_HEAD(&rinfo->indirect_pages); |
1937 | INIT_LIST_HEAD(&rinfo->grants); | |
1938 | rinfo->dev_info = info; | |
1939 | INIT_WORK(&rinfo->work, blkif_restart_queue); | |
1940 | spin_lock_init(&rinfo->ring_lock); | |
1941 | } | |
1942 | return 0; | |
1943 | } | |
9f27ee59 JF |
1944 | /** |
1945 | * Entry point to this code when a new device is created. Allocate the basic | |
1946 | * structures and the ring buffer for communication with the backend, and | |
1947 | * inform the backend of the appropriate details for those. Switch to | |
1948 | * Initialised state. | |
1949 | */ | |
1950 | static int blkfront_probe(struct xenbus_device *dev, | |
1951 | const struct xenbus_device_id *id) | |
1952 | { | |
86839c56 | 1953 | int err, vdevice; |
9f27ee59 JF |
1954 | struct blkfront_info *info; |
1955 | ||
1956 | /* FIXME: Use dynamic device id if this is not set. */ | |
1957 | err = xenbus_scanf(XBT_NIL, dev->nodename, | |
1958 | "virtual-device", "%i", &vdevice); | |
1959 | if (err != 1) { | |
9246b5f0 CL |
1960 | /* go looking in the extended area instead */ |
1961 | err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext", | |
1962 | "%i", &vdevice); | |
1963 | if (err != 1) { | |
1964 | xenbus_dev_fatal(dev, err, "reading virtual-device"); | |
1965 | return err; | |
1966 | } | |
9f27ee59 JF |
1967 | } |
1968 | ||
b98a409b SS |
1969 | if (xen_hvm_domain()) { |
1970 | char *type; | |
1971 | int len; | |
1972 | /* no unplug has been done: do not hook devices != xen vbds */ | |
51c71a3b | 1973 | if (xen_has_pv_and_legacy_disk_devices()) { |
b98a409b SS |
1974 | int major; |
1975 | ||
1976 | if (!VDEV_IS_EXTENDED(vdevice)) | |
1977 | major = BLKIF_MAJOR(vdevice); | |
1978 | else | |
1979 | major = XENVBD_MAJOR; | |
1980 | ||
1981 | if (major != XENVBD_MAJOR) { | |
1982 | printk(KERN_INFO | |
1983 | "%s: HVM does not support vbd %d as xen block device\n", | |
02f1f217 | 1984 | __func__, vdevice); |
b98a409b SS |
1985 | return -ENODEV; |
1986 | } | |
1987 | } | |
1988 | /* do not create a PV cdrom device if we are an HVM guest */ | |
1989 | type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len); | |
1990 | if (IS_ERR(type)) | |
1991 | return -ENODEV; | |
1992 | if (strncmp(type, "cdrom", 5) == 0) { | |
1993 | kfree(type); | |
c1c5413a SS |
1994 | return -ENODEV; |
1995 | } | |
b98a409b | 1996 | kfree(type); |
c1c5413a | 1997 | } |
9f27ee59 JF |
1998 | info = kzalloc(sizeof(*info), GFP_KERNEL); |
1999 | if (!info) { | |
2000 | xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure"); | |
2001 | return -ENOMEM; | |
2002 | } | |
2003 | ||
28d949bc | 2004 | info->xbdev = dev; |
81f35161 | 2005 | |
b70f5fa0 | 2006 | mutex_init(&info->mutex); |
9f27ee59 JF |
2007 | info->vdevice = vdevice; |
2008 | info->connected = BLKIF_STATE_DISCONNECTED; | |
9f27ee59 | 2009 | |
9f27ee59 JF |
2010 | /* Front end dir is a number, which is used as the id. */ |
2011 | info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0); | |
a1b4b12b | 2012 | dev_set_drvdata(&dev->dev, info); |
9f27ee59 | 2013 | |
a46b5367 JG |
2014 | mutex_lock(&blkfront_mutex); |
2015 | list_add(&info->info_list, &info_list); | |
2016 | mutex_unlock(&blkfront_mutex); | |
2017 | ||
9f27ee59 JF |
2018 | return 0; |
2019 | } | |
2020 | ||
9f27ee59 JF |
2021 | static int blkif_recover(struct blkfront_info *info) |
2022 | { | |
4559fa55 | 2023 | unsigned int r_index; |
402b27f9 | 2024 | struct request *req, *n; |
402b27f9 | 2025 | int rc; |
4559fa55 N |
2026 | struct bio *bio; |
2027 | unsigned int segs; | |
4ab50af6 | 2028 | struct blkfront_ring_info *rinfo; |
402b27f9 | 2029 | |
3df0e505 | 2030 | blkfront_gather_backend_features(info); |
172335ad BL |
2031 | /* Reset limits changed by blk_mq_update_nr_hw_queues(). */ |
2032 | blkif_set_queue_limits(info); | |
402b27f9 | 2033 | segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST; |
6c647b0e | 2034 | blk_queue_max_segments(info->rq, segs / GRANTS_PER_PSEG); |
9f27ee59 | 2035 | |
4ab50af6 | 2036 | for_each_rinfo(info, rinfo, r_index) { |
3df0e505 | 2037 | rc = blkfront_setup_indirect(rinfo); |
7b427a59 | 2038 | if (rc) |
3df0e505 | 2039 | return rc; |
3df0e505 | 2040 | } |
9f27ee59 JF |
2041 | xenbus_switch_state(info->xbdev, XenbusStateConnected); |
2042 | ||
9f27ee59 JF |
2043 | /* Now safe for us to use the shared ring */ |
2044 | info->connected = BLKIF_STATE_CONNECTED; | |
2045 | ||
4ab50af6 | 2046 | for_each_rinfo(info, rinfo, r_index) { |
3df0e505 BL |
2047 | /* Kick any other new requests queued since we resumed */ |
2048 | kick_pending_request_queues(rinfo); | |
2049 | } | |
9f27ee59 | 2050 | |
7b427a59 | 2051 | list_for_each_entry_safe(req, n, &info->requests, queuelist) { |
402b27f9 RPM |
2052 | /* Requeue pending requests (flush or discard) */ |
2053 | list_del_init(&req->queuelist); | |
2054 | BUG_ON(req->nr_phys_segments > segs); | |
2b053aca | 2055 | blk_mq_requeue_request(req, false); |
402b27f9 | 2056 | } |
52d7f1b5 | 2057 | blk_mq_start_stopped_hw_queues(info->rq, true); |
907c3eb1 | 2058 | blk_mq_kick_requeue_list(info->rq); |
9f27ee59 | 2059 | |
7b427a59 | 2060 | while ((bio = bio_list_pop(&info->bio_list)) != NULL) { |
402b27f9 | 2061 | /* Traverse the list of pending bios and re-queue them */ |
4e49ea4a | 2062 | submit_bio(bio); |
402b27f9 RPM |
2063 | } |
2064 | ||
9f27ee59 JF |
2065 | return 0; |
2066 | } | |
2067 | ||
2068 | /** | |
2069 | * We are reconnecting to the backend, due to a suspend/resume, or a backend | |
2070 | * driver restart. We tear down our blkif structure and recreate it, but | |
2071 | * leave the device-layer structures intact so that this is transparent to the | |
2072 | * rest of the kernel. | |
2073 | */ | |
2074 | static int blkfront_resume(struct xenbus_device *dev) | |
2075 | { | |
a1b4b12b | 2076 | struct blkfront_info *info = dev_get_drvdata(&dev->dev); |
3db70a85 | 2077 | int err = 0; |
7b427a59 | 2078 | unsigned int i, j; |
4ab50af6 | 2079 | struct blkfront_ring_info *rinfo; |
9f27ee59 JF |
2080 | |
2081 | dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename); | |
2082 | ||
7b427a59 BL |
2083 | bio_list_init(&info->bio_list); |
2084 | INIT_LIST_HEAD(&info->requests); | |
4ab50af6 | 2085 | for_each_rinfo(info, rinfo, i) { |
7b427a59 BL |
2086 | struct bio_list merge_bio; |
2087 | struct blk_shadow *shadow = rinfo->shadow; | |
2088 | ||
2089 | for (j = 0; j < BLK_RING_SIZE(info); j++) { | |
2090 | /* Not in use? */ | |
2091 | if (!shadow[j].request) | |
2092 | continue; | |
2093 | ||
2094 | /* | |
2095 | * Get the bios in the request so we can re-queue them. | |
2096 | */ | |
b15bd8cb MK |
2097 | if (req_op(shadow[j].request) == REQ_OP_FLUSH || |
2098 | req_op(shadow[j].request) == REQ_OP_DISCARD || | |
2099 | req_op(shadow[j].request) == REQ_OP_SECURE_ERASE || | |
3fc9d690 | 2100 | shadow[j].request->cmd_flags & REQ_FUA) { |
7b427a59 BL |
2101 | /* |
2102 | * Flush operations don't contain bios, so | |
2103 | * we need to requeue the whole request | |
3fc9d690 LT |
2104 | * |
2105 | * XXX: but this doesn't make any sense for a | |
2106 | * write with the FUA flag set.. | |
7b427a59 BL |
2107 | */ |
2108 | list_add(&shadow[j].request->queuelist, &info->requests); | |
2109 | continue; | |
2110 | } | |
2111 | merge_bio.head = shadow[j].request->bio; | |
2112 | merge_bio.tail = shadow[j].request->biotail; | |
2113 | bio_list_merge(&info->bio_list, &merge_bio); | |
2114 | shadow[j].request->bio = NULL; | |
2a842aca | 2115 | blk_mq_end_request(shadow[j].request, BLK_STS_OK); |
7b427a59 BL |
2116 | } |
2117 | } | |
2118 | ||
9f27ee59 JF |
2119 | blkif_free(info, info->connected == BLKIF_STATE_CONNECTED); |
2120 | ||
203fd61f | 2121 | err = talk_to_blkback(dev, info); |
2a6f71ad BL |
2122 | if (!err) |
2123 | blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings); | |
402b27f9 RPM |
2124 | |
2125 | /* | |
2126 | * We have to wait for the backend to switch to | |
2127 | * connected state, since we want to read which | |
2128 | * features it supports. | |
2129 | */ | |
9f27ee59 JF |
2130 | |
2131 | return err; | |
2132 | } | |
2133 | ||
6f03a7ff | 2134 | static void blkfront_closing(struct blkfront_info *info) |
b70f5fa0 DS |
2135 | { |
2136 | struct xenbus_device *xbdev = info->xbdev; | |
2137 | struct block_device *bdev = NULL; | |
2138 | ||
2139 | mutex_lock(&info->mutex); | |
2140 | ||
2141 | if (xbdev->state == XenbusStateClosing) { | |
2142 | mutex_unlock(&info->mutex); | |
2143 | return; | |
2144 | } | |
2145 | ||
2146 | if (info->gd) | |
2147 | bdev = bdget_disk(info->gd, 0); | |
2148 | ||
2149 | mutex_unlock(&info->mutex); | |
2150 | ||
2151 | if (!bdev) { | |
2152 | xenbus_frontend_closed(xbdev); | |
2153 | return; | |
2154 | } | |
2155 | ||
2156 | mutex_lock(&bdev->bd_mutex); | |
2157 | ||
7b32d104 | 2158 | if (bdev->bd_openers) { |
b70f5fa0 DS |
2159 | xenbus_dev_error(xbdev, -EBUSY, |
2160 | "Device in use; refusing to close"); | |
2161 | xenbus_switch_state(xbdev, XenbusStateClosing); | |
2162 | } else { | |
2163 | xlvbd_release_gendisk(info); | |
2164 | xenbus_frontend_closed(xbdev); | |
2165 | } | |
2166 | ||
2167 | mutex_unlock(&bdev->bd_mutex); | |
2168 | bdput(bdev); | |
2169 | } | |
9f27ee59 | 2170 | |
ed30bf31 LD |
2171 | static void blkfront_setup_discard(struct blkfront_info *info) |
2172 | { | |
2173 | int err; | |
ed30bf31 LD |
2174 | unsigned int discard_granularity; |
2175 | unsigned int discard_alignment; | |
2176 | ||
1c8cad6c OH |
2177 | info->feature_discard = 1; |
2178 | err = xenbus_gather(XBT_NIL, info->xbdev->otherend, | |
2179 | "discard-granularity", "%u", &discard_granularity, | |
2180 | "discard-alignment", "%u", &discard_alignment, | |
2181 | NULL); | |
2182 | if (!err) { | |
2183 | info->discard_granularity = discard_granularity; | |
2184 | info->discard_alignment = discard_alignment; | |
2185 | } | |
f27dc1ac JG |
2186 | info->feature_secdiscard = |
2187 | !!xenbus_read_unsigned(info->xbdev->otherend, "discard-secure", | |
2188 | 0); | |
ed30bf31 LD |
2189 | } |
2190 | ||
81f35161 | 2191 | static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo) |
402b27f9 | 2192 | { |
3a169c0b | 2193 | unsigned int psegs, grants, memflags; |
402b27f9 | 2194 | int err, i; |
81f35161 | 2195 | struct blkfront_info *info = rinfo->dev_info; |
402b27f9 | 2196 | |
3a169c0b JG |
2197 | memflags = memalloc_noio_save(); |
2198 | ||
6cc56833 JG |
2199 | if (info->max_indirect_segments == 0) { |
2200 | if (!HAS_EXTRA_REQ) | |
2201 | grants = BLKIF_MAX_SEGMENTS_PER_REQUEST; | |
2202 | else { | |
2203 | /* | |
2204 | * When an extra req is required, the maximum | |
2205 | * grants supported is related to the size of the | |
2206 | * Linux block segment. | |
2207 | */ | |
2208 | grants = GRANTS_PER_PSEG; | |
2209 | } | |
2210 | } | |
d50babbe | 2211 | else |
c004a6fe | 2212 | grants = info->max_indirect_segments; |
3b4f1884 | 2213 | psegs = DIV_ROUND_UP(grants, GRANTS_PER_PSEG); |
402b27f9 | 2214 | |
81f35161 | 2215 | err = fill_grant_buffer(rinfo, |
c004a6fe | 2216 | (grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info)); |
402b27f9 RPM |
2217 | if (err) |
2218 | goto out_of_memory; | |
2219 | ||
bfe11d6d RPM |
2220 | if (!info->feature_persistent && info->max_indirect_segments) { |
2221 | /* | |
2222 | * We are using indirect descriptors but not persistent | |
2223 | * grants, we need to allocate a set of pages that can be | |
2224 | * used for mapping indirect grefs | |
2225 | */ | |
c004a6fe | 2226 | int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info); |
bfe11d6d | 2227 | |
81f35161 | 2228 | BUG_ON(!list_empty(&rinfo->indirect_pages)); |
bfe11d6d | 2229 | for (i = 0; i < num; i++) { |
3a169c0b | 2230 | struct page *indirect_page = alloc_page(GFP_KERNEL); |
bfe11d6d RPM |
2231 | if (!indirect_page) |
2232 | goto out_of_memory; | |
81f35161 | 2233 | list_add(&indirect_page->lru, &rinfo->indirect_pages); |
bfe11d6d RPM |
2234 | } |
2235 | } | |
2236 | ||
86839c56 | 2237 | for (i = 0; i < BLK_RING_SIZE(info); i++) { |
6396bb22 | 2238 | rinfo->shadow[i].grants_used = |
1d5c76e6 RPM |
2239 | kvcalloc(grants, |
2240 | sizeof(rinfo->shadow[i].grants_used[0]), | |
3a169c0b | 2241 | GFP_KERNEL); |
1d5c76e6 RPM |
2242 | rinfo->shadow[i].sg = kvcalloc(psegs, |
2243 | sizeof(rinfo->shadow[i].sg[0]), | |
3a169c0b | 2244 | GFP_KERNEL); |
6396bb22 KC |
2245 | if (info->max_indirect_segments) |
2246 | rinfo->shadow[i].indirect_grants = | |
1d5c76e6 RPM |
2247 | kvcalloc(INDIRECT_GREFS(grants), |
2248 | sizeof(rinfo->shadow[i].indirect_grants[0]), | |
3a169c0b | 2249 | GFP_KERNEL); |
81f35161 BL |
2250 | if ((rinfo->shadow[i].grants_used == NULL) || |
2251 | (rinfo->shadow[i].sg == NULL) || | |
402b27f9 | 2252 | (info->max_indirect_segments && |
81f35161 | 2253 | (rinfo->shadow[i].indirect_grants == NULL))) |
402b27f9 | 2254 | goto out_of_memory; |
81f35161 | 2255 | sg_init_table(rinfo->shadow[i].sg, psegs); |
402b27f9 RPM |
2256 | } |
2257 | ||
3a169c0b | 2258 | memalloc_noio_restore(memflags); |
402b27f9 RPM |
2259 | |
2260 | return 0; | |
2261 | ||
2262 | out_of_memory: | |
86839c56 | 2263 | for (i = 0; i < BLK_RING_SIZE(info); i++) { |
1d5c76e6 | 2264 | kvfree(rinfo->shadow[i].grants_used); |
81f35161 | 2265 | rinfo->shadow[i].grants_used = NULL; |
1d5c76e6 | 2266 | kvfree(rinfo->shadow[i].sg); |
81f35161 | 2267 | rinfo->shadow[i].sg = NULL; |
1d5c76e6 | 2268 | kvfree(rinfo->shadow[i].indirect_grants); |
81f35161 | 2269 | rinfo->shadow[i].indirect_grants = NULL; |
402b27f9 | 2270 | } |
81f35161 | 2271 | if (!list_empty(&rinfo->indirect_pages)) { |
bfe11d6d | 2272 | struct page *indirect_page, *n; |
81f35161 | 2273 | list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) { |
bfe11d6d RPM |
2274 | list_del(&indirect_page->lru); |
2275 | __free_page(indirect_page); | |
2276 | } | |
2277 | } | |
3a169c0b JG |
2278 | |
2279 | memalloc_noio_restore(memflags); | |
2280 | ||
402b27f9 RPM |
2281 | return -ENOMEM; |
2282 | } | |
2283 | ||
d50babbe BL |
2284 | /* |
2285 | * Gather all backend feature-* | |
2286 | */ | |
3df0e505 | 2287 | static void blkfront_gather_backend_features(struct blkfront_info *info) |
d50babbe | 2288 | { |
d50babbe BL |
2289 | unsigned int indirect_segments; |
2290 | ||
2291 | info->feature_flush = 0; | |
a418090a | 2292 | info->feature_fua = 0; |
d50babbe | 2293 | |
d50babbe BL |
2294 | /* |
2295 | * If there's no "feature-barrier" defined, then it means | |
2296 | * we're dealing with a very old backend which writes | |
2297 | * synchronously; nothing to do. | |
2298 | * | |
2299 | * If there are barriers, then we use flush. | |
2300 | */ | |
f27dc1ac | 2301 | if (xenbus_read_unsigned(info->xbdev->otherend, "feature-barrier", 0)) { |
a418090a MC |
2302 | info->feature_flush = 1; |
2303 | info->feature_fua = 1; | |
2304 | } | |
2305 | ||
d50babbe BL |
2306 | /* |
2307 | * And if there is "feature-flush-cache" use that above | |
2308 | * barriers. | |
2309 | */ | |
f27dc1ac JG |
2310 | if (xenbus_read_unsigned(info->xbdev->otherend, "feature-flush-cache", |
2311 | 0)) { | |
a418090a MC |
2312 | info->feature_flush = 1; |
2313 | info->feature_fua = 0; | |
2314 | } | |
d50babbe | 2315 | |
f27dc1ac | 2316 | if (xenbus_read_unsigned(info->xbdev->otherend, "feature-discard", 0)) |
d50babbe BL |
2317 | blkfront_setup_discard(info); |
2318 | ||
f27dc1ac | 2319 | info->feature_persistent = |
b32728ff JB |
2320 | !!xenbus_read_unsigned(info->xbdev->otherend, |
2321 | "feature-persistent", 0); | |
d50babbe | 2322 | |
f27dc1ac JG |
2323 | indirect_segments = xenbus_read_unsigned(info->xbdev->otherend, |
2324 | "feature-max-indirect-segments", 0); | |
3b4f1884 JB |
2325 | if (indirect_segments > xen_blkif_max_segments) |
2326 | indirect_segments = xen_blkif_max_segments; | |
2327 | if (indirect_segments <= BLKIF_MAX_SEGMENTS_PER_REQUEST) | |
2328 | indirect_segments = 0; | |
2329 | info->max_indirect_segments = indirect_segments; | |
a46b5367 JG |
2330 | |
2331 | if (info->feature_persistent) { | |
2332 | mutex_lock(&blkfront_mutex); | |
2333 | schedule_delayed_work(&blkfront_work, HZ * 10); | |
2334 | mutex_unlock(&blkfront_mutex); | |
2335 | } | |
d50babbe BL |
2336 | } |
2337 | ||
9f27ee59 JF |
2338 | /* |
2339 | * Invoked when the backend is finally 'ready' (and has told produced | |
2340 | * the details about the physical device - #sectors, size, etc). | |
2341 | */ | |
2342 | static void blkfront_connect(struct blkfront_info *info) | |
2343 | { | |
2344 | unsigned long long sectors; | |
2345 | unsigned long sector_size; | |
7c4d7d71 | 2346 | unsigned int physical_sector_size; |
9f27ee59 | 2347 | unsigned int binfo; |
3df0e505 | 2348 | int err, i; |
4ab50af6 | 2349 | struct blkfront_ring_info *rinfo; |
9f27ee59 | 2350 | |
1fa73be6 S |
2351 | switch (info->connected) { |
2352 | case BLKIF_STATE_CONNECTED: | |
2353 | /* | |
2354 | * Potentially, the back-end may be signalling | |
2355 | * a capacity change; update the capacity. | |
2356 | */ | |
2357 | err = xenbus_scanf(XBT_NIL, info->xbdev->otherend, | |
2358 | "sectors", "%Lu", §ors); | |
2359 | if (XENBUS_EXIST_ERR(err)) | |
2360 | return; | |
2361 | printk(KERN_INFO "Setting capacity to %Lu\n", | |
2362 | sectors); | |
3cbc28bb | 2363 | set_capacity_revalidate_and_notify(info->gd, sectors, true); |
1fa73be6 | 2364 | |
402b27f9 | 2365 | return; |
1fa73be6 | 2366 | case BLKIF_STATE_SUSPENDED: |
402b27f9 RPM |
2367 | /* |
2368 | * If we are recovering from suspension, we need to wait | |
2369 | * for the backend to announce it's features before | |
2370 | * reconnecting, at least we need to know if the backend | |
2371 | * supports indirect descriptors, and how many. | |
2372 | */ | |
2373 | blkif_recover(info); | |
9f27ee59 JF |
2374 | return; |
2375 | ||
b4dddb49 JF |
2376 | default: |
2377 | break; | |
1fa73be6 | 2378 | } |
9f27ee59 JF |
2379 | |
2380 | dev_dbg(&info->xbdev->dev, "%s:%s.\n", | |
2381 | __func__, info->xbdev->otherend); | |
2382 | ||
2383 | err = xenbus_gather(XBT_NIL, info->xbdev->otherend, | |
2384 | "sectors", "%llu", §ors, | |
2385 | "info", "%u", &binfo, | |
2386 | "sector-size", "%lu", §or_size, | |
2387 | NULL); | |
2388 | if (err) { | |
2389 | xenbus_dev_fatal(info->xbdev, err, | |
2390 | "reading backend fields at %s", | |
2391 | info->xbdev->otherend); | |
2392 | return; | |
2393 | } | |
2394 | ||
7c4d7d71 SB |
2395 | /* |
2396 | * physcial-sector-size is a newer field, so old backends may not | |
2397 | * provide this. Assume physical sector size to be the same as | |
2398 | * sector_size in that case. | |
2399 | */ | |
f27dc1ac JG |
2400 | physical_sector_size = xenbus_read_unsigned(info->xbdev->otherend, |
2401 | "physical-sector-size", | |
2402 | sector_size); | |
3df0e505 | 2403 | blkfront_gather_backend_features(info); |
4ab50af6 JG |
2404 | for_each_rinfo(info, rinfo, i) { |
2405 | err = blkfront_setup_indirect(rinfo); | |
3df0e505 BL |
2406 | if (err) { |
2407 | xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s", | |
2408 | info->xbdev->otherend); | |
2409 | blkif_free(info, 0); | |
2410 | break; | |
2411 | } | |
402b27f9 RPM |
2412 | } |
2413 | ||
7c4d7d71 SB |
2414 | err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size, |
2415 | physical_sector_size); | |
9f27ee59 JF |
2416 | if (err) { |
2417 | xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s", | |
2418 | info->xbdev->otherend); | |
4e876c2b | 2419 | goto fail; |
9f27ee59 JF |
2420 | } |
2421 | ||
2422 | xenbus_switch_state(info->xbdev, XenbusStateConnected); | |
2423 | ||
2424 | /* Kick pending requests. */ | |
9f27ee59 | 2425 | info->connected = BLKIF_STATE_CONNECTED; |
4ab50af6 JG |
2426 | for_each_rinfo(info, rinfo, i) |
2427 | kick_pending_request_queues(rinfo); | |
9f27ee59 | 2428 | |
fef912bf | 2429 | device_add_disk(&info->xbdev->dev, info->gd, NULL); |
1d78d705 CL |
2430 | |
2431 | info->is_ready = 1; | |
4e876c2b BL |
2432 | return; |
2433 | ||
2434 | fail: | |
2435 | blkif_free(info, 0); | |
2436 | return; | |
9f27ee59 JF |
2437 | } |
2438 | ||
9f27ee59 JF |
2439 | /** |
2440 | * Callback received when the backend's state changes. | |
2441 | */ | |
203fd61f | 2442 | static void blkback_changed(struct xenbus_device *dev, |
9f27ee59 JF |
2443 | enum xenbus_state backend_state) |
2444 | { | |
a1b4b12b | 2445 | struct blkfront_info *info = dev_get_drvdata(&dev->dev); |
9f27ee59 | 2446 | |
203fd61f | 2447 | dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state); |
9f27ee59 JF |
2448 | |
2449 | switch (backend_state) { | |
9f27ee59 | 2450 | case XenbusStateInitWait: |
a9b54bb9 BL |
2451 | if (dev->state != XenbusStateInitialising) |
2452 | break; | |
c31ecf6c | 2453 | if (talk_to_blkback(dev, info)) |
8ab0144a | 2454 | break; |
8ab0144a | 2455 | case XenbusStateInitialising: |
9f27ee59 | 2456 | case XenbusStateInitialised: |
b78c9512 NI |
2457 | case XenbusStateReconfiguring: |
2458 | case XenbusStateReconfigured: | |
9f27ee59 | 2459 | case XenbusStateUnknown: |
9f27ee59 JF |
2460 | break; |
2461 | ||
2462 | case XenbusStateConnected: | |
efd15352 BL |
2463 | /* |
2464 | * talk_to_blkback sets state to XenbusStateInitialised | |
2465 | * and blkfront_connect sets it to XenbusStateConnected | |
2466 | * (if connection went OK). | |
2467 | * | |
2468 | * If the backend (or toolstack) decides to poke at backend | |
2469 | * state (and re-trigger the watch by setting the state repeatedly | |
2470 | * to XenbusStateConnected (4)) we need to deal with this. | |
2471 | * This is allowed as this is used to communicate to the guest | |
2472 | * that the size of disk has changed! | |
2473 | */ | |
2474 | if ((dev->state != XenbusStateInitialised) && | |
2475 | (dev->state != XenbusStateConnected)) { | |
c31ecf6c KRW |
2476 | if (talk_to_blkback(dev, info)) |
2477 | break; | |
2478 | } | |
efd15352 | 2479 | |
9f27ee59 JF |
2480 | blkfront_connect(info); |
2481 | break; | |
2482 | ||
36613717 DV |
2483 | case XenbusStateClosed: |
2484 | if (dev->state == XenbusStateClosed) | |
2485 | break; | |
df561f66 | 2486 | fallthrough; |
9f27ee59 | 2487 | case XenbusStateClosing: |
a54c8f0f CA |
2488 | if (info) |
2489 | blkfront_closing(info); | |
9f27ee59 JF |
2490 | break; |
2491 | } | |
2492 | } | |
2493 | ||
fa1bd359 | 2494 | static int blkfront_remove(struct xenbus_device *xbdev) |
9f27ee59 | 2495 | { |
fa1bd359 DS |
2496 | struct blkfront_info *info = dev_get_drvdata(&xbdev->dev); |
2497 | struct block_device *bdev = NULL; | |
2498 | struct gendisk *disk; | |
9f27ee59 | 2499 | |
fa1bd359 | 2500 | dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename); |
9f27ee59 | 2501 | |
f92898e7 VL |
2502 | if (!info) |
2503 | return 0; | |
2504 | ||
9f27ee59 JF |
2505 | blkif_free(info, 0); |
2506 | ||
fa1bd359 DS |
2507 | mutex_lock(&info->mutex); |
2508 | ||
2509 | disk = info->gd; | |
2510 | if (disk) | |
2511 | bdev = bdget_disk(disk, 0); | |
2512 | ||
2513 | info->xbdev = NULL; | |
2514 | mutex_unlock(&info->mutex); | |
2515 | ||
2516 | if (!bdev) { | |
a46b5367 JG |
2517 | mutex_lock(&blkfront_mutex); |
2518 | free_info(info); | |
2519 | mutex_unlock(&blkfront_mutex); | |
fa1bd359 DS |
2520 | return 0; |
2521 | } | |
2522 | ||
2523 | /* | |
2524 | * The xbdev was removed before we reached the Closed | |
2525 | * state. See if it's safe to remove the disk. If the bdev | |
2526 | * isn't closed yet, we let release take care of it. | |
2527 | */ | |
2528 | ||
2529 | mutex_lock(&bdev->bd_mutex); | |
2530 | info = disk->private_data; | |
2531 | ||
d54142c7 DS |
2532 | dev_warn(disk_to_dev(disk), |
2533 | "%s was hot-unplugged, %d stale handles\n", | |
2534 | xbdev->nodename, bdev->bd_openers); | |
2535 | ||
7b32d104 | 2536 | if (info && !bdev->bd_openers) { |
fa1bd359 DS |
2537 | xlvbd_release_gendisk(info); |
2538 | disk->private_data = NULL; | |
a46b5367 JG |
2539 | mutex_lock(&blkfront_mutex); |
2540 | free_info(info); | |
2541 | mutex_unlock(&blkfront_mutex); | |
fa1bd359 DS |
2542 | } |
2543 | ||
2544 | mutex_unlock(&bdev->bd_mutex); | |
2545 | bdput(bdev); | |
9f27ee59 JF |
2546 | |
2547 | return 0; | |
2548 | } | |
2549 | ||
1d78d705 CL |
2550 | static int blkfront_is_ready(struct xenbus_device *dev) |
2551 | { | |
a1b4b12b | 2552 | struct blkfront_info *info = dev_get_drvdata(&dev->dev); |
1d78d705 | 2553 | |
5d7ed20e | 2554 | return info->is_ready && info->xbdev; |
1d78d705 CL |
2555 | } |
2556 | ||
a63c848b | 2557 | static int blkif_open(struct block_device *bdev, fmode_t mode) |
9f27ee59 | 2558 | { |
13961743 DS |
2559 | struct gendisk *disk = bdev->bd_disk; |
2560 | struct blkfront_info *info; | |
2561 | int err = 0; | |
6e9624b8 | 2562 | |
2a48fc0a | 2563 | mutex_lock(&blkfront_mutex); |
6e9624b8 | 2564 | |
13961743 DS |
2565 | info = disk->private_data; |
2566 | if (!info) { | |
2567 | /* xbdev gone */ | |
2568 | err = -ERESTARTSYS; | |
2569 | goto out; | |
2570 | } | |
2571 | ||
2572 | mutex_lock(&info->mutex); | |
2573 | ||
2574 | if (!info->gd) | |
2575 | /* xbdev is closed */ | |
2576 | err = -ERESTARTSYS; | |
2577 | ||
2578 | mutex_unlock(&info->mutex); | |
2579 | ||
13961743 | 2580 | out: |
2a48fc0a | 2581 | mutex_unlock(&blkfront_mutex); |
13961743 | 2582 | return err; |
9f27ee59 JF |
2583 | } |
2584 | ||
db2a144b | 2585 | static void blkif_release(struct gendisk *disk, fmode_t mode) |
9f27ee59 | 2586 | { |
a63c848b | 2587 | struct blkfront_info *info = disk->private_data; |
7fd152f4 DS |
2588 | struct block_device *bdev; |
2589 | struct xenbus_device *xbdev; | |
2590 | ||
2a48fc0a | 2591 | mutex_lock(&blkfront_mutex); |
7fd152f4 DS |
2592 | |
2593 | bdev = bdget_disk(disk, 0); | |
7fd152f4 | 2594 | |
2f089cb8 FP |
2595 | if (!bdev) { |
2596 | WARN(1, "Block device %s yanked out from us!\n", disk->disk_name); | |
2597 | goto out_mutex; | |
2598 | } | |
acfca3c6 DS |
2599 | if (bdev->bd_openers) |
2600 | goto out; | |
2601 | ||
7fd152f4 DS |
2602 | /* |
2603 | * Check if we have been instructed to close. We will have | |
2604 | * deferred this request, because the bdev was still open. | |
2605 | */ | |
2606 | ||
2607 | mutex_lock(&info->mutex); | |
2608 | xbdev = info->xbdev; | |
2609 | ||
2610 | if (xbdev && xbdev->state == XenbusStateClosing) { | |
2611 | /* pending switch to state closed */ | |
d54142c7 | 2612 | dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n"); |
7fd152f4 DS |
2613 | xlvbd_release_gendisk(info); |
2614 | xenbus_frontend_closed(info->xbdev); | |
2615 | } | |
2616 | ||
2617 | mutex_unlock(&info->mutex); | |
2618 | ||
2619 | if (!xbdev) { | |
2620 | /* sudden device removal */ | |
d54142c7 | 2621 | dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n"); |
7fd152f4 DS |
2622 | xlvbd_release_gendisk(info); |
2623 | disk->private_data = NULL; | |
a46b5367 | 2624 | free_info(info); |
9f27ee59 | 2625 | } |
7fd152f4 | 2626 | |
a4cc14ec | 2627 | out: |
dad5cf65 | 2628 | bdput(bdev); |
2f089cb8 | 2629 | out_mutex: |
2a48fc0a | 2630 | mutex_unlock(&blkfront_mutex); |
9f27ee59 JF |
2631 | } |
2632 | ||
83d5cde4 | 2633 | static const struct block_device_operations xlvbd_block_fops = |
9f27ee59 JF |
2634 | { |
2635 | .owner = THIS_MODULE, | |
a63c848b AV |
2636 | .open = blkif_open, |
2637 | .release = blkif_release, | |
597592d9 | 2638 | .getgeo = blkif_getgeo, |
8a6cfeb6 | 2639 | .ioctl = blkif_ioctl, |
9452b1a3 | 2640 | .compat_ioctl = blkdev_compat_ptr_ioctl, |
9f27ee59 JF |
2641 | }; |
2642 | ||
2643 | ||
ec9c42ec | 2644 | static const struct xenbus_device_id blkfront_ids[] = { |
9f27ee59 JF |
2645 | { "vbd" }, |
2646 | { "" } | |
2647 | }; | |
2648 | ||
95afae48 DV |
2649 | static struct xenbus_driver blkfront_driver = { |
2650 | .ids = blkfront_ids, | |
9f27ee59 JF |
2651 | .probe = blkfront_probe, |
2652 | .remove = blkfront_remove, | |
2653 | .resume = blkfront_resume, | |
203fd61f | 2654 | .otherend_changed = blkback_changed, |
1d78d705 | 2655 | .is_ready = blkfront_is_ready, |
95afae48 | 2656 | }; |
9f27ee59 | 2657 | |
a46b5367 JG |
2658 | static void purge_persistent_grants(struct blkfront_info *info) |
2659 | { | |
2660 | unsigned int i; | |
2661 | unsigned long flags; | |
4ab50af6 | 2662 | struct blkfront_ring_info *rinfo; |
a46b5367 | 2663 | |
4ab50af6 | 2664 | for_each_rinfo(info, rinfo, i) { |
a46b5367 JG |
2665 | struct grant *gnt_list_entry, *tmp; |
2666 | ||
2667 | spin_lock_irqsave(&rinfo->ring_lock, flags); | |
2668 | ||
2669 | if (rinfo->persistent_gnts_c == 0) { | |
2670 | spin_unlock_irqrestore(&rinfo->ring_lock, flags); | |
2671 | continue; | |
2672 | } | |
2673 | ||
2674 | list_for_each_entry_safe(gnt_list_entry, tmp, &rinfo->grants, | |
2675 | node) { | |
2676 | if (gnt_list_entry->gref == GRANT_INVALID_REF || | |
2677 | gnttab_query_foreign_access(gnt_list_entry->gref)) | |
2678 | continue; | |
2679 | ||
2680 | list_del(&gnt_list_entry->node); | |
2681 | gnttab_end_foreign_access(gnt_list_entry->gref, 0, 0UL); | |
2682 | rinfo->persistent_gnts_c--; | |
6c767867 JG |
2683 | gnt_list_entry->gref = GRANT_INVALID_REF; |
2684 | list_add_tail(&gnt_list_entry->node, &rinfo->grants); | |
a46b5367 JG |
2685 | } |
2686 | ||
2687 | spin_unlock_irqrestore(&rinfo->ring_lock, flags); | |
2688 | } | |
2689 | } | |
2690 | ||
2691 | static void blkfront_delay_work(struct work_struct *work) | |
2692 | { | |
2693 | struct blkfront_info *info; | |
2694 | bool need_schedule_work = false; | |
2695 | ||
2696 | mutex_lock(&blkfront_mutex); | |
2697 | ||
2698 | list_for_each_entry(info, &info_list, info_list) { | |
2699 | if (info->feature_persistent) { | |
2700 | need_schedule_work = true; | |
2701 | mutex_lock(&info->mutex); | |
2702 | purge_persistent_grants(info); | |
2703 | mutex_unlock(&info->mutex); | |
2704 | } | |
2705 | } | |
2706 | ||
2707 | if (need_schedule_work) | |
2708 | schedule_delayed_work(&blkfront_work, HZ * 10); | |
2709 | ||
2710 | mutex_unlock(&blkfront_mutex); | |
2711 | } | |
2712 | ||
9f27ee59 JF |
2713 | static int __init xlblk_init(void) |
2714 | { | |
469738e6 | 2715 | int ret; |
28d949bc | 2716 | int nr_cpus = num_online_cpus(); |
469738e6 | 2717 | |
6e833587 | 2718 | if (!xen_domain()) |
9f27ee59 JF |
2719 | return -ENODEV; |
2720 | ||
4bcddbae JG |
2721 | if (!xen_has_pv_disk_devices()) |
2722 | return -ENODEV; | |
2723 | ||
2724 | if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) { | |
2725 | pr_warn("xen_blk: can't get major %d with name %s\n", | |
2726 | XENVBD_MAJOR, DEV_NAME); | |
2727 | return -ENODEV; | |
2728 | } | |
2729 | ||
3b4f1884 JB |
2730 | if (xen_blkif_max_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST) |
2731 | xen_blkif_max_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST; | |
2732 | ||
9cce2914 | 2733 | if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) { |
86839c56 | 2734 | pr_info("Invalid max_ring_order (%d), will use default max: %d.\n", |
9cce2914 | 2735 | xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER); |
45fc8264 | 2736 | xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER; |
86839c56 BL |
2737 | } |
2738 | ||
28d949bc BL |
2739 | if (xen_blkif_max_queues > nr_cpus) { |
2740 | pr_info("Invalid max_queues (%d), will use default max: %d.\n", | |
2741 | xen_blkif_max_queues, nr_cpus); | |
2742 | xen_blkif_max_queues = nr_cpus; | |
2743 | } | |
2744 | ||
a46b5367 JG |
2745 | INIT_DELAYED_WORK(&blkfront_work, blkfront_delay_work); |
2746 | ||
73db144b | 2747 | ret = xenbus_register_frontend(&blkfront_driver); |
469738e6 LE |
2748 | if (ret) { |
2749 | unregister_blkdev(XENVBD_MAJOR, DEV_NAME); | |
2750 | return ret; | |
2751 | } | |
2752 | ||
2753 | return 0; | |
9f27ee59 JF |
2754 | } |
2755 | module_init(xlblk_init); | |
2756 | ||
2757 | ||
5a60d0cd | 2758 | static void __exit xlblk_exit(void) |
9f27ee59 | 2759 | { |
a46b5367 JG |
2760 | cancel_delayed_work_sync(&blkfront_work); |
2761 | ||
8605067f JB |
2762 | xenbus_unregister_driver(&blkfront_driver); |
2763 | unregister_blkdev(XENVBD_MAJOR, DEV_NAME); | |
2764 | kfree(minors); | |
9f27ee59 JF |
2765 | } |
2766 | module_exit(xlblk_exit); | |
2767 | ||
2768 | MODULE_DESCRIPTION("Xen virtual block device frontend"); | |
2769 | MODULE_LICENSE("GPL"); | |
2770 | MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR); | |
d2f0c52b | 2771 | MODULE_ALIAS("xen:vbd"); |
4f93f09b | 2772 | MODULE_ALIAS("xenblk"); |