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> |
9f27ee59 | 49 | |
1ccbf534 | 50 | #include <xen/xen.h> |
9f27ee59 JF |
51 | #include <xen/xenbus.h> |
52 | #include <xen/grant_table.h> | |
53 | #include <xen/events.h> | |
54 | #include <xen/page.h> | |
c1c5413a | 55 | #include <xen/platform_pci.h> |
9f27ee59 JF |
56 | |
57 | #include <xen/interface/grant_table.h> | |
58 | #include <xen/interface/io/blkif.h> | |
3e334239 | 59 | #include <xen/interface/io/protocols.h> |
9f27ee59 JF |
60 | |
61 | #include <asm/xen/hypervisor.h> | |
62 | ||
63 | enum blkif_state { | |
64 | BLKIF_STATE_DISCONNECTED, | |
65 | BLKIF_STATE_CONNECTED, | |
66 | BLKIF_STATE_SUSPENDED, | |
67 | }; | |
68 | ||
0a8704a5 RPM |
69 | struct grant { |
70 | grant_ref_t gref; | |
a7a6df22 | 71 | struct page *page; |
155b7edb | 72 | struct list_head node; |
0a8704a5 RPM |
73 | }; |
74 | ||
9f27ee59 JF |
75 | struct blk_shadow { |
76 | struct blkif_request req; | |
a945b980 | 77 | struct request *request; |
402b27f9 RPM |
78 | struct grant **grants_used; |
79 | struct grant **indirect_grants; | |
b7649158 | 80 | struct scatterlist *sg; |
c004a6fe | 81 | unsigned int num_sg; |
402b27f9 RPM |
82 | }; |
83 | ||
84 | struct split_bio { | |
85 | struct bio *bio; | |
86 | atomic_t pending; | |
9f27ee59 JF |
87 | }; |
88 | ||
2a48fc0a | 89 | static DEFINE_MUTEX(blkfront_mutex); |
83d5cde4 | 90 | static const struct block_device_operations xlvbd_block_fops; |
9f27ee59 | 91 | |
402b27f9 RPM |
92 | /* |
93 | * Maximum number of segments in indirect requests, the actual value used by | |
94 | * the frontend driver is the minimum of this value and the value provided | |
95 | * by the backend driver. | |
96 | */ | |
97 | ||
98 | static unsigned int xen_blkif_max_segments = 32; | |
2d5dc3ba KRW |
99 | module_param_named(max, xen_blkif_max_segments, int, S_IRUGO); |
100 | MODULE_PARM_DESC(max, "Maximum amount of segments in indirect requests (default is 32)"); | |
402b27f9 | 101 | |
86839c56 BL |
102 | /* |
103 | * Maximum order of pages to be used for the shared ring between front and | |
104 | * backend, 4KB page granularity is used. | |
105 | */ | |
106 | static unsigned int xen_blkif_max_ring_order; | |
107 | module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, S_IRUGO); | |
108 | MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring"); | |
109 | ||
c004a6fe JG |
110 | #define BLK_RING_SIZE(info) \ |
111 | __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages) | |
112 | ||
113 | #define BLK_MAX_RING_SIZE \ | |
9cce2914 | 114 | __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * XENBUS_MAX_RING_GRANTS) |
c004a6fe | 115 | |
86839c56 BL |
116 | /* |
117 | * ring-ref%i i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19 | |
118 | * characters are enough. Define to 20 to keep consist with backend. | |
119 | */ | |
120 | #define RINGREF_NAME_LEN (20) | |
9f27ee59 JF |
121 | |
122 | /* | |
123 | * We have one of these per vbd, whether ide, scsi or 'other'. They | |
124 | * hang in private_data off the gendisk structure. We may end up | |
125 | * putting all kinds of interesting stuff here :-) | |
126 | */ | |
127 | struct blkfront_info | |
128 | { | |
3467811e | 129 | spinlock_t io_lock; |
b70f5fa0 | 130 | struct mutex mutex; |
9f27ee59 | 131 | struct xenbus_device *xbdev; |
9f27ee59 JF |
132 | struct gendisk *gd; |
133 | int vdevice; | |
134 | blkif_vdev_t handle; | |
135 | enum blkif_state connected; | |
9cce2914 | 136 | int ring_ref[XENBUS_MAX_RING_GRANTS]; |
86839c56 | 137 | unsigned int nr_ring_pages; |
9f27ee59 JF |
138 | struct blkif_front_ring ring; |
139 | unsigned int evtchn, irq; | |
140 | struct request_queue *rq; | |
141 | struct work_struct work; | |
142 | struct gnttab_free_callback callback; | |
86839c56 | 143 | struct blk_shadow shadow[BLK_MAX_RING_SIZE]; |
bfe11d6d RPM |
144 | struct list_head grants; |
145 | struct list_head indirect_pages; | |
0a8704a5 | 146 | unsigned int persistent_gnts_c; |
9f27ee59 | 147 | unsigned long shadow_free; |
4913efe4 | 148 | unsigned int feature_flush; |
5ea42986 KRW |
149 | unsigned int feature_discard:1; |
150 | unsigned int feature_secdiscard:1; | |
ed30bf31 LD |
151 | unsigned int discard_granularity; |
152 | unsigned int discard_alignment; | |
0a8704a5 | 153 | unsigned int feature_persistent:1; |
c004a6fe | 154 | /* Number of 4KB segments handled */ |
402b27f9 | 155 | unsigned int max_indirect_segments; |
1d78d705 | 156 | int is_ready; |
907c3eb1 | 157 | struct blk_mq_tag_set tag_set; |
9f27ee59 JF |
158 | }; |
159 | ||
0e345826 JB |
160 | static unsigned int nr_minors; |
161 | static unsigned long *minors; | |
162 | static DEFINE_SPINLOCK(minor_lock); | |
163 | ||
9f27ee59 JF |
164 | #define GRANT_INVALID_REF 0 |
165 | ||
166 | #define PARTS_PER_DISK 16 | |
9246b5f0 | 167 | #define PARTS_PER_EXT_DISK 256 |
9f27ee59 JF |
168 | |
169 | #define BLKIF_MAJOR(dev) ((dev)>>8) | |
170 | #define BLKIF_MINOR(dev) ((dev) & 0xff) | |
171 | ||
9246b5f0 CL |
172 | #define EXT_SHIFT 28 |
173 | #define EXTENDED (1<<EXT_SHIFT) | |
174 | #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED)) | |
175 | #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED)) | |
c80a4209 SS |
176 | #define EMULATED_HD_DISK_MINOR_OFFSET (0) |
177 | #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256) | |
196cfe2a SB |
178 | #define EMULATED_SD_DISK_MINOR_OFFSET (0) |
179 | #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256) | |
9f27ee59 | 180 | |
9246b5f0 | 181 | #define DEV_NAME "xvd" /* name in /dev */ |
9f27ee59 | 182 | |
c004a6fe JG |
183 | /* |
184 | * Grants are always the same size as a Xen page (i.e 4KB). | |
185 | * A physical segment is always the same size as a Linux page. | |
186 | * Number of grants per physical segment | |
187 | */ | |
188 | #define GRANTS_PER_PSEG (PAGE_SIZE / XEN_PAGE_SIZE) | |
189 | ||
190 | #define GRANTS_PER_INDIRECT_FRAME \ | |
191 | (XEN_PAGE_SIZE / sizeof(struct blkif_request_segment)) | |
192 | ||
193 | #define PSEGS_PER_INDIRECT_FRAME \ | |
194 | (GRANTS_INDIRECT_FRAME / GRANTS_PSEGS) | |
195 | ||
196 | #define INDIRECT_GREFS(_grants) \ | |
197 | DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME) | |
198 | ||
199 | #define GREFS(_psegs) ((_psegs) * GRANTS_PER_PSEG) | |
402b27f9 RPM |
200 | |
201 | static int blkfront_setup_indirect(struct blkfront_info *info); | |
d50babbe | 202 | static int blkfront_gather_backend_features(struct blkfront_info *info); |
402b27f9 | 203 | |
9f27ee59 JF |
204 | static int get_id_from_freelist(struct blkfront_info *info) |
205 | { | |
206 | unsigned long free = info->shadow_free; | |
86839c56 | 207 | BUG_ON(free >= BLK_RING_SIZE(info)); |
97e36834 KRW |
208 | info->shadow_free = info->shadow[free].req.u.rw.id; |
209 | info->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */ | |
9f27ee59 JF |
210 | return free; |
211 | } | |
212 | ||
6878c32e | 213 | static int add_id_to_freelist(struct blkfront_info *info, |
9f27ee59 JF |
214 | unsigned long id) |
215 | { | |
6878c32e KRW |
216 | if (info->shadow[id].req.u.rw.id != id) |
217 | return -EINVAL; | |
218 | if (info->shadow[id].request == NULL) | |
219 | return -EINVAL; | |
97e36834 | 220 | info->shadow[id].req.u.rw.id = info->shadow_free; |
a945b980 | 221 | info->shadow[id].request = NULL; |
9f27ee59 | 222 | info->shadow_free = id; |
6878c32e | 223 | return 0; |
9f27ee59 JF |
224 | } |
225 | ||
9c1e050c RPM |
226 | static int fill_grant_buffer(struct blkfront_info *info, int num) |
227 | { | |
228 | struct page *granted_page; | |
229 | struct grant *gnt_list_entry, *n; | |
230 | int i = 0; | |
231 | ||
232 | while(i < num) { | |
233 | gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO); | |
234 | if (!gnt_list_entry) | |
235 | goto out_of_memory; | |
236 | ||
bfe11d6d RPM |
237 | if (info->feature_persistent) { |
238 | granted_page = alloc_page(GFP_NOIO); | |
239 | if (!granted_page) { | |
240 | kfree(gnt_list_entry); | |
241 | goto out_of_memory; | |
242 | } | |
a7a6df22 | 243 | gnt_list_entry->page = granted_page; |
9c1e050c RPM |
244 | } |
245 | ||
9c1e050c | 246 | gnt_list_entry->gref = GRANT_INVALID_REF; |
bfe11d6d | 247 | list_add(&gnt_list_entry->node, &info->grants); |
9c1e050c RPM |
248 | i++; |
249 | } | |
250 | ||
251 | return 0; | |
252 | ||
253 | out_of_memory: | |
254 | list_for_each_entry_safe(gnt_list_entry, n, | |
bfe11d6d | 255 | &info->grants, node) { |
9c1e050c | 256 | list_del(&gnt_list_entry->node); |
bfe11d6d | 257 | if (info->feature_persistent) |
a7a6df22 | 258 | __free_page(gnt_list_entry->page); |
9c1e050c RPM |
259 | kfree(gnt_list_entry); |
260 | i--; | |
261 | } | |
262 | BUG_ON(i != 0); | |
263 | return -ENOMEM; | |
264 | } | |
265 | ||
4f503fbd | 266 | static struct grant *get_free_grant(struct blkfront_info *info) |
9c1e050c RPM |
267 | { |
268 | struct grant *gnt_list_entry; | |
9c1e050c | 269 | |
bfe11d6d RPM |
270 | BUG_ON(list_empty(&info->grants)); |
271 | gnt_list_entry = list_first_entry(&info->grants, struct grant, | |
4f503fbd | 272 | node); |
9c1e050c RPM |
273 | list_del(&gnt_list_entry->node); |
274 | ||
4f503fbd | 275 | if (gnt_list_entry->gref != GRANT_INVALID_REF) |
9c1e050c | 276 | info->persistent_gnts_c--; |
4f503fbd JG |
277 | |
278 | return gnt_list_entry; | |
279 | } | |
280 | ||
281 | static inline void grant_foreign_access(const struct grant *gnt_list_entry, | |
282 | const struct blkfront_info *info) | |
283 | { | |
284 | gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref, | |
285 | info->xbdev->otherend_id, | |
286 | gnt_list_entry->page, | |
287 | 0); | |
288 | } | |
289 | ||
290 | static struct grant *get_grant(grant_ref_t *gref_head, | |
291 | unsigned long gfn, | |
292 | struct blkfront_info *info) | |
293 | { | |
294 | struct grant *gnt_list_entry = get_free_grant(info); | |
295 | ||
296 | if (gnt_list_entry->gref != GRANT_INVALID_REF) | |
9c1e050c | 297 | return gnt_list_entry; |
4f503fbd JG |
298 | |
299 | /* Assign a gref to this page */ | |
300 | gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head); | |
301 | BUG_ON(gnt_list_entry->gref == -ENOSPC); | |
302 | if (info->feature_persistent) | |
303 | grant_foreign_access(gnt_list_entry, info); | |
304 | else { | |
305 | /* Grant access to the GFN passed by the caller */ | |
306 | gnttab_grant_foreign_access_ref(gnt_list_entry->gref, | |
307 | info->xbdev->otherend_id, | |
308 | gfn, 0); | |
9c1e050c RPM |
309 | } |
310 | ||
4f503fbd JG |
311 | return gnt_list_entry; |
312 | } | |
313 | ||
314 | static struct grant *get_indirect_grant(grant_ref_t *gref_head, | |
315 | struct blkfront_info *info) | |
316 | { | |
317 | struct grant *gnt_list_entry = get_free_grant(info); | |
318 | ||
319 | if (gnt_list_entry->gref != GRANT_INVALID_REF) | |
320 | return gnt_list_entry; | |
321 | ||
9c1e050c RPM |
322 | /* Assign a gref to this page */ |
323 | gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head); | |
324 | BUG_ON(gnt_list_entry->gref == -ENOSPC); | |
bfe11d6d | 325 | if (!info->feature_persistent) { |
4f503fbd JG |
326 | struct page *indirect_page; |
327 | ||
328 | /* Fetch a pre-allocated page to use for indirect grefs */ | |
329 | BUG_ON(list_empty(&info->indirect_pages)); | |
330 | indirect_page = list_first_entry(&info->indirect_pages, | |
331 | struct page, lru); | |
332 | list_del(&indirect_page->lru); | |
333 | gnt_list_entry->page = indirect_page; | |
bfe11d6d | 334 | } |
4f503fbd JG |
335 | grant_foreign_access(gnt_list_entry, info); |
336 | ||
9c1e050c RPM |
337 | return gnt_list_entry; |
338 | } | |
339 | ||
6878c32e KRW |
340 | static const char *op_name(int op) |
341 | { | |
342 | static const char *const names[] = { | |
343 | [BLKIF_OP_READ] = "read", | |
344 | [BLKIF_OP_WRITE] = "write", | |
345 | [BLKIF_OP_WRITE_BARRIER] = "barrier", | |
346 | [BLKIF_OP_FLUSH_DISKCACHE] = "flush", | |
347 | [BLKIF_OP_DISCARD] = "discard" }; | |
348 | ||
349 | if (op < 0 || op >= ARRAY_SIZE(names)) | |
350 | return "unknown"; | |
351 | ||
352 | if (!names[op]) | |
353 | return "reserved"; | |
354 | ||
355 | return names[op]; | |
356 | } | |
0e345826 JB |
357 | static int xlbd_reserve_minors(unsigned int minor, unsigned int nr) |
358 | { | |
359 | unsigned int end = minor + nr; | |
360 | int rc; | |
361 | ||
362 | if (end > nr_minors) { | |
363 | unsigned long *bitmap, *old; | |
364 | ||
f094148a | 365 | bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap), |
0e345826 JB |
366 | GFP_KERNEL); |
367 | if (bitmap == NULL) | |
368 | return -ENOMEM; | |
369 | ||
370 | spin_lock(&minor_lock); | |
371 | if (end > nr_minors) { | |
372 | old = minors; | |
373 | memcpy(bitmap, minors, | |
374 | BITS_TO_LONGS(nr_minors) * sizeof(*bitmap)); | |
375 | minors = bitmap; | |
376 | nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG; | |
377 | } else | |
378 | old = bitmap; | |
379 | spin_unlock(&minor_lock); | |
380 | kfree(old); | |
381 | } | |
382 | ||
383 | spin_lock(&minor_lock); | |
384 | if (find_next_bit(minors, end, minor) >= end) { | |
34ae2e47 | 385 | bitmap_set(minors, minor, nr); |
0e345826 JB |
386 | rc = 0; |
387 | } else | |
388 | rc = -EBUSY; | |
389 | spin_unlock(&minor_lock); | |
390 | ||
391 | return rc; | |
392 | } | |
393 | ||
394 | static void xlbd_release_minors(unsigned int minor, unsigned int nr) | |
395 | { | |
396 | unsigned int end = minor + nr; | |
397 | ||
398 | BUG_ON(end > nr_minors); | |
399 | spin_lock(&minor_lock); | |
34ae2e47 | 400 | bitmap_clear(minors, minor, nr); |
0e345826 JB |
401 | spin_unlock(&minor_lock); |
402 | } | |
403 | ||
9f27ee59 JF |
404 | static void blkif_restart_queue_callback(void *arg) |
405 | { | |
406 | struct blkfront_info *info = (struct blkfront_info *)arg; | |
407 | schedule_work(&info->work); | |
408 | } | |
409 | ||
afe42d7d | 410 | static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg) |
597592d9 IC |
411 | { |
412 | /* We don't have real geometry info, but let's at least return | |
413 | values consistent with the size of the device */ | |
414 | sector_t nsect = get_capacity(bd->bd_disk); | |
415 | sector_t cylinders = nsect; | |
416 | ||
417 | hg->heads = 0xff; | |
418 | hg->sectors = 0x3f; | |
419 | sector_div(cylinders, hg->heads * hg->sectors); | |
420 | hg->cylinders = cylinders; | |
421 | if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect) | |
422 | hg->cylinders = 0xffff; | |
423 | return 0; | |
424 | } | |
425 | ||
a63c848b | 426 | static int blkif_ioctl(struct block_device *bdev, fmode_t mode, |
62aa0054 | 427 | unsigned command, unsigned long argument) |
440a01a7 | 428 | { |
a63c848b | 429 | struct blkfront_info *info = bdev->bd_disk->private_data; |
440a01a7 CL |
430 | int i; |
431 | ||
432 | dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n", | |
433 | command, (long)argument); | |
434 | ||
435 | switch (command) { | |
436 | case CDROMMULTISESSION: | |
437 | dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n"); | |
438 | for (i = 0; i < sizeof(struct cdrom_multisession); i++) | |
439 | if (put_user(0, (char __user *)(argument + i))) | |
440 | return -EFAULT; | |
441 | return 0; | |
442 | ||
443 | case CDROM_GET_CAPABILITY: { | |
444 | struct gendisk *gd = info->gd; | |
445 | if (gd->flags & GENHD_FL_CD) | |
446 | return 0; | |
447 | return -EINVAL; | |
448 | } | |
449 | ||
450 | default: | |
451 | /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n", | |
452 | command);*/ | |
453 | return -EINVAL; /* same return as native Linux */ | |
454 | } | |
455 | ||
456 | return 0; | |
457 | } | |
458 | ||
33204663 | 459 | static int blkif_queue_discard_req(struct request *req) |
9f27ee59 JF |
460 | { |
461 | struct blkfront_info *info = req->rq_disk->private_data; | |
9f27ee59 | 462 | struct blkif_request *ring_req; |
9f27ee59 | 463 | unsigned long id; |
33204663 JG |
464 | |
465 | /* Fill out a communications ring structure. */ | |
466 | ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt); | |
467 | id = get_id_from_freelist(info); | |
468 | info->shadow[id].request = req; | |
469 | ||
470 | ring_req->operation = BLKIF_OP_DISCARD; | |
471 | ring_req->u.discard.nr_sectors = blk_rq_sectors(req); | |
472 | ring_req->u.discard.id = id; | |
473 | ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req); | |
474 | if ((req->cmd_flags & REQ_SECURE) && info->feature_secdiscard) | |
475 | ring_req->u.discard.flag = BLKIF_DISCARD_SECURE; | |
476 | else | |
477 | ring_req->u.discard.flag = 0; | |
478 | ||
479 | info->ring.req_prod_pvt++; | |
480 | ||
481 | /* Keep a private copy so we can reissue requests when recovering. */ | |
482 | info->shadow[id].req = *ring_req; | |
483 | ||
484 | return 0; | |
485 | } | |
486 | ||
c004a6fe JG |
487 | struct setup_rw_req { |
488 | unsigned int grant_idx; | |
489 | struct blkif_request_segment *segments; | |
490 | struct blkfront_info *info; | |
491 | struct blkif_request *ring_req; | |
492 | grant_ref_t gref_head; | |
493 | unsigned int id; | |
494 | /* Only used when persistent grant is used and it's a read request */ | |
495 | bool need_copy; | |
496 | unsigned int bvec_off; | |
497 | char *bvec_data; | |
498 | }; | |
499 | ||
500 | static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset, | |
501 | unsigned int len, void *data) | |
502 | { | |
503 | struct setup_rw_req *setup = data; | |
504 | int n, ref; | |
505 | struct grant *gnt_list_entry; | |
9f27ee59 | 506 | unsigned int fsect, lsect; |
c004a6fe JG |
507 | /* Convenient aliases */ |
508 | unsigned int grant_idx = setup->grant_idx; | |
509 | struct blkif_request *ring_req = setup->ring_req; | |
510 | struct blkfront_info *info = setup->info; | |
511 | struct blk_shadow *shadow = &info->shadow[setup->id]; | |
512 | ||
513 | if ((ring_req->operation == BLKIF_OP_INDIRECT) && | |
514 | (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) { | |
515 | if (setup->segments) | |
516 | kunmap_atomic(setup->segments); | |
517 | ||
518 | n = grant_idx / GRANTS_PER_INDIRECT_FRAME; | |
519 | gnt_list_entry = get_indirect_grant(&setup->gref_head, info); | |
520 | shadow->indirect_grants[n] = gnt_list_entry; | |
521 | setup->segments = kmap_atomic(gnt_list_entry->page); | |
522 | ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref; | |
523 | } | |
524 | ||
525 | gnt_list_entry = get_grant(&setup->gref_head, gfn, info); | |
526 | ref = gnt_list_entry->gref; | |
527 | shadow->grants_used[grant_idx] = gnt_list_entry; | |
528 | ||
529 | if (setup->need_copy) { | |
530 | void *shared_data; | |
531 | ||
532 | shared_data = kmap_atomic(gnt_list_entry->page); | |
533 | /* | |
534 | * this does not wipe data stored outside the | |
535 | * range sg->offset..sg->offset+sg->length. | |
536 | * Therefore, blkback *could* see data from | |
537 | * previous requests. This is OK as long as | |
538 | * persistent grants are shared with just one | |
539 | * domain. It may need refactoring if this | |
540 | * changes | |
541 | */ | |
542 | memcpy(shared_data + offset, | |
543 | setup->bvec_data + setup->bvec_off, | |
544 | len); | |
545 | ||
546 | kunmap_atomic(shared_data); | |
547 | setup->bvec_off += len; | |
548 | } | |
549 | ||
550 | fsect = offset >> 9; | |
551 | lsect = fsect + (len >> 9) - 1; | |
552 | if (ring_req->operation != BLKIF_OP_INDIRECT) { | |
553 | ring_req->u.rw.seg[grant_idx] = | |
554 | (struct blkif_request_segment) { | |
555 | .gref = ref, | |
556 | .first_sect = fsect, | |
557 | .last_sect = lsect }; | |
558 | } else { | |
559 | setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] = | |
560 | (struct blkif_request_segment) { | |
561 | .gref = ref, | |
562 | .first_sect = fsect, | |
563 | .last_sect = lsect }; | |
564 | } | |
565 | ||
566 | (setup->grant_idx)++; | |
567 | } | |
568 | ||
33204663 | 569 | static int blkif_queue_rw_req(struct request *req) |
9f27ee59 JF |
570 | { |
571 | struct blkfront_info *info = req->rq_disk->private_data; | |
9f27ee59 | 572 | struct blkif_request *ring_req; |
9f27ee59 | 573 | unsigned long id; |
c004a6fe JG |
574 | int i; |
575 | struct setup_rw_req setup = { | |
576 | .grant_idx = 0, | |
577 | .segments = NULL, | |
578 | .info = info, | |
579 | .need_copy = rq_data_dir(req) && info->feature_persistent, | |
580 | }; | |
0a8704a5 RPM |
581 | |
582 | /* | |
583 | * Used to store if we are able to queue the request by just using | |
584 | * existing persistent grants, or if we have to get new grants, | |
585 | * as there are not sufficiently many free. | |
586 | */ | |
587 | bool new_persistent_gnts; | |
9e973e64 | 588 | struct scatterlist *sg; |
c004a6fe | 589 | int num_sg, max_grefs, num_grant; |
9f27ee59 | 590 | |
c004a6fe | 591 | max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG; |
c47206e2 RPM |
592 | if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST) |
593 | /* | |
594 | * If we are using indirect segments we need to account | |
595 | * for the indirect grefs used in the request. | |
596 | */ | |
c004a6fe | 597 | max_grefs += INDIRECT_GREFS(max_grefs); |
402b27f9 RPM |
598 | |
599 | /* Check if we have enough grants to allocate a requests */ | |
600 | if (info->persistent_gnts_c < max_grefs) { | |
0a8704a5 RPM |
601 | new_persistent_gnts = 1; |
602 | if (gnttab_alloc_grant_references( | |
402b27f9 | 603 | max_grefs - info->persistent_gnts_c, |
c004a6fe | 604 | &setup.gref_head) < 0) { |
0a8704a5 RPM |
605 | gnttab_request_free_callback( |
606 | &info->callback, | |
607 | blkif_restart_queue_callback, | |
608 | info, | |
402b27f9 | 609 | max_grefs); |
0a8704a5 RPM |
610 | return 1; |
611 | } | |
612 | } else | |
613 | new_persistent_gnts = 0; | |
9f27ee59 JF |
614 | |
615 | /* Fill out a communications ring structure. */ | |
616 | ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt); | |
617 | id = get_id_from_freelist(info); | |
a945b980 | 618 | info->shadow[id].request = req; |
9f27ee59 | 619 | |
33204663 | 620 | BUG_ON(info->max_indirect_segments == 0 && |
c004a6fe | 621 | GREFS(req->nr_phys_segments) > BLKIF_MAX_SEGMENTS_PER_REQUEST); |
33204663 | 622 | BUG_ON(info->max_indirect_segments && |
c004a6fe JG |
623 | GREFS(req->nr_phys_segments) > info->max_indirect_segments); |
624 | ||
625 | num_sg = blk_rq_map_sg(req->q, req, info->shadow[id].sg); | |
626 | num_grant = 0; | |
627 | /* Calculate the number of grant used */ | |
628 | for_each_sg(info->shadow[id].sg, sg, num_sg, i) | |
629 | num_grant += gnttab_count_grant(sg->offset, sg->length); | |
630 | ||
33204663 | 631 | ring_req->u.rw.id = id; |
c004a6fe JG |
632 | info->shadow[id].num_sg = num_sg; |
633 | if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST) { | |
33204663 JG |
634 | /* |
635 | * The indirect operation can only be a BLKIF_OP_READ or | |
636 | * BLKIF_OP_WRITE | |
637 | */ | |
638 | BUG_ON(req->cmd_flags & (REQ_FLUSH | REQ_FUA)); | |
639 | ring_req->operation = BLKIF_OP_INDIRECT; | |
640 | ring_req->u.indirect.indirect_op = rq_data_dir(req) ? | |
641 | BLKIF_OP_WRITE : BLKIF_OP_READ; | |
642 | ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req); | |
643 | ring_req->u.indirect.handle = info->handle; | |
c004a6fe | 644 | ring_req->u.indirect.nr_segments = num_grant; |
ed30bf31 | 645 | } else { |
33204663 JG |
646 | ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req); |
647 | ring_req->u.rw.handle = info->handle; | |
648 | ring_req->operation = rq_data_dir(req) ? | |
649 | BLKIF_OP_WRITE : BLKIF_OP_READ; | |
650 | if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) { | |
402b27f9 | 651 | /* |
33204663 JG |
652 | * Ideally we can do an unordered flush-to-disk. |
653 | * In case the backend onlysupports barriers, use that. | |
654 | * A barrier request a superset of FUA, so we can | |
655 | * implement it the same way. (It's also a FLUSH+FUA, | |
656 | * since it is guaranteed ordered WRT previous writes.) | |
402b27f9 | 657 | */ |
33204663 JG |
658 | switch (info->feature_flush & |
659 | ((REQ_FLUSH|REQ_FUA))) { | |
660 | case REQ_FLUSH|REQ_FUA: | |
661 | ring_req->operation = | |
662 | BLKIF_OP_WRITE_BARRIER; | |
663 | break; | |
664 | case REQ_FLUSH: | |
665 | ring_req->operation = | |
666 | BLKIF_OP_FLUSH_DISKCACHE; | |
667 | break; | |
668 | default: | |
669 | ring_req->operation = 0; | |
402b27f9 | 670 | } |
402b27f9 | 671 | } |
c004a6fe | 672 | ring_req->u.rw.nr_segments = num_grant; |
33204663 | 673 | } |
0a8704a5 | 674 | |
c004a6fe JG |
675 | setup.ring_req = ring_req; |
676 | setup.id = id; | |
677 | for_each_sg(info->shadow[id].sg, sg, num_sg, i) { | |
678 | BUG_ON(sg->offset + sg->length > PAGE_SIZE); | |
0a8704a5 | 679 | |
c004a6fe JG |
680 | if (setup.need_copy) { |
681 | setup.bvec_off = sg->offset; | |
682 | setup.bvec_data = kmap_atomic(sg_page(sg)); | |
683 | } | |
0a8704a5 | 684 | |
c004a6fe JG |
685 | gnttab_foreach_grant_in_range(sg_page(sg), |
686 | sg->offset, | |
687 | sg->length, | |
688 | blkif_setup_rw_req_grant, | |
689 | &setup); | |
0a8704a5 | 690 | |
c004a6fe JG |
691 | if (setup.need_copy) |
692 | kunmap_atomic(setup.bvec_data); | |
9f27ee59 | 693 | } |
c004a6fe JG |
694 | if (setup.segments) |
695 | kunmap_atomic(setup.segments); | |
9f27ee59 JF |
696 | |
697 | info->ring.req_prod_pvt++; | |
698 | ||
699 | /* Keep a private copy so we can reissue requests when recovering. */ | |
700 | info->shadow[id].req = *ring_req; | |
701 | ||
0a8704a5 | 702 | if (new_persistent_gnts) |
c004a6fe | 703 | gnttab_free_grant_references(setup.gref_head); |
9f27ee59 JF |
704 | |
705 | return 0; | |
706 | } | |
707 | ||
33204663 JG |
708 | /* |
709 | * Generate a Xen blkfront IO request from a blk layer request. Reads | |
710 | * and writes are handled as expected. | |
711 | * | |
712 | * @req: a request struct | |
713 | */ | |
714 | static int blkif_queue_request(struct request *req) | |
715 | { | |
716 | struct blkfront_info *info = req->rq_disk->private_data; | |
717 | ||
718 | if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) | |
719 | return 1; | |
720 | ||
721 | if (unlikely(req->cmd_flags & (REQ_DISCARD | REQ_SECURE))) | |
722 | return blkif_queue_discard_req(req); | |
723 | else | |
724 | return blkif_queue_rw_req(req); | |
725 | } | |
9f27ee59 JF |
726 | |
727 | static inline void flush_requests(struct blkfront_info *info) | |
728 | { | |
729 | int notify; | |
730 | ||
731 | RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify); | |
732 | ||
733 | if (notify) | |
734 | notify_remote_via_irq(info->irq); | |
735 | } | |
736 | ||
ad42d391 VK |
737 | static inline bool blkif_request_flush_invalid(struct request *req, |
738 | struct blkfront_info *info) | |
0f1ca65e AA |
739 | { |
740 | return ((req->cmd_type != REQ_TYPE_FS) || | |
ad42d391 VK |
741 | ((req->cmd_flags & REQ_FLUSH) && |
742 | !(info->feature_flush & REQ_FLUSH)) || | |
743 | ((req->cmd_flags & REQ_FUA) && | |
744 | !(info->feature_flush & REQ_FUA))); | |
0f1ca65e AA |
745 | } |
746 | ||
907c3eb1 BL |
747 | static int blkif_queue_rq(struct blk_mq_hw_ctx *hctx, |
748 | const struct blk_mq_queue_data *qd) | |
9f27ee59 | 749 | { |
907c3eb1 | 750 | struct blkfront_info *info = qd->rq->rq_disk->private_data; |
9f27ee59 | 751 | |
907c3eb1 BL |
752 | blk_mq_start_request(qd->rq); |
753 | spin_lock_irq(&info->io_lock); | |
754 | if (RING_FULL(&info->ring)) | |
755 | goto out_busy; | |
9f27ee59 | 756 | |
907c3eb1 BL |
757 | if (blkif_request_flush_invalid(qd->rq, info)) |
758 | goto out_err; | |
296b2f6a | 759 | |
907c3eb1 BL |
760 | if (blkif_queue_request(qd->rq)) |
761 | goto out_busy; | |
296b2f6a | 762 | |
907c3eb1 BL |
763 | flush_requests(info); |
764 | spin_unlock_irq(&info->io_lock); | |
765 | return BLK_MQ_RQ_QUEUE_OK; | |
9f27ee59 | 766 | |
907c3eb1 BL |
767 | out_err: |
768 | spin_unlock_irq(&info->io_lock); | |
769 | return BLK_MQ_RQ_QUEUE_ERROR; | |
9f27ee59 | 770 | |
907c3eb1 BL |
771 | out_busy: |
772 | spin_unlock_irq(&info->io_lock); | |
773 | blk_mq_stop_hw_queue(hctx); | |
774 | return BLK_MQ_RQ_QUEUE_BUSY; | |
9f27ee59 JF |
775 | } |
776 | ||
907c3eb1 BL |
777 | static struct blk_mq_ops blkfront_mq_ops = { |
778 | .queue_rq = blkif_queue_rq, | |
779 | .map_queue = blk_mq_map_queue, | |
780 | }; | |
781 | ||
402b27f9 | 782 | static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size, |
7c4d7d71 | 783 | unsigned int physical_sector_size, |
402b27f9 | 784 | unsigned int segments) |
9f27ee59 | 785 | { |
165125e1 | 786 | struct request_queue *rq; |
ed30bf31 | 787 | struct blkfront_info *info = gd->private_data; |
9f27ee59 | 788 | |
907c3eb1 BL |
789 | memset(&info->tag_set, 0, sizeof(info->tag_set)); |
790 | info->tag_set.ops = &blkfront_mq_ops; | |
791 | info->tag_set.nr_hw_queues = 1; | |
792 | info->tag_set.queue_depth = BLK_RING_SIZE(info); | |
793 | info->tag_set.numa_node = NUMA_NO_NODE; | |
794 | info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; | |
795 | info->tag_set.cmd_size = 0; | |
796 | info->tag_set.driver_data = info; | |
797 | ||
798 | if (blk_mq_alloc_tag_set(&info->tag_set)) | |
9f27ee59 | 799 | return -1; |
907c3eb1 BL |
800 | rq = blk_mq_init_queue(&info->tag_set); |
801 | if (IS_ERR(rq)) { | |
802 | blk_mq_free_tag_set(&info->tag_set); | |
803 | return -1; | |
804 | } | |
9f27ee59 | 805 | |
66d352e1 | 806 | queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq); |
9f27ee59 | 807 | |
ed30bf31 LD |
808 | if (info->feature_discard) { |
809 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq); | |
810 | blk_queue_max_discard_sectors(rq, get_capacity(gd)); | |
811 | rq->limits.discard_granularity = info->discard_granularity; | |
812 | rq->limits.discard_alignment = info->discard_alignment; | |
5ea42986 KRW |
813 | if (info->feature_secdiscard) |
814 | queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD, rq); | |
ed30bf31 LD |
815 | } |
816 | ||
9f27ee59 | 817 | /* Hard sector size and max sectors impersonate the equiv. hardware. */ |
e1defc4f | 818 | blk_queue_logical_block_size(rq, sector_size); |
7c4d7d71 | 819 | blk_queue_physical_block_size(rq, physical_sector_size); |
c004a6fe | 820 | blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512); |
9f27ee59 JF |
821 | |
822 | /* Each segment in a request is up to an aligned page in size. */ | |
823 | blk_queue_segment_boundary(rq, PAGE_SIZE - 1); | |
824 | blk_queue_max_segment_size(rq, PAGE_SIZE); | |
825 | ||
826 | /* Ensure a merged request will fit in a single I/O ring slot. */ | |
c004a6fe | 827 | blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG); |
9f27ee59 JF |
828 | |
829 | /* Make sure buffer addresses are sector-aligned. */ | |
830 | blk_queue_dma_alignment(rq, 511); | |
831 | ||
1c91fe1a IC |
832 | /* Make sure we don't use bounce buffers. */ |
833 | blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY); | |
834 | ||
9f27ee59 JF |
835 | gd->queue = rq; |
836 | ||
837 | return 0; | |
838 | } | |
839 | ||
fdf9b965 VK |
840 | static const char *flush_info(unsigned int feature_flush) |
841 | { | |
842 | switch (feature_flush & ((REQ_FLUSH | REQ_FUA))) { | |
843 | case REQ_FLUSH|REQ_FUA: | |
844 | return "barrier: enabled;"; | |
845 | case REQ_FLUSH: | |
846 | return "flush diskcache: enabled;"; | |
847 | default: | |
848 | return "barrier or flush: disabled;"; | |
849 | } | |
850 | } | |
9f27ee59 | 851 | |
4913efe4 | 852 | static void xlvbd_flush(struct blkfront_info *info) |
9f27ee59 | 853 | { |
4913efe4 | 854 | blk_queue_flush(info->rq, info->feature_flush); |
fdf9b965 VK |
855 | pr_info("blkfront: %s: %s %s %s %s %s\n", |
856 | info->gd->disk_name, flush_info(info->feature_flush), | |
857 | "persistent grants:", info->feature_persistent ? | |
858 | "enabled;" : "disabled;", "indirect descriptors:", | |
859 | info->max_indirect_segments ? "enabled;" : "disabled;"); | |
9f27ee59 JF |
860 | } |
861 | ||
c80a4209 SS |
862 | static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset) |
863 | { | |
864 | int major; | |
865 | major = BLKIF_MAJOR(vdevice); | |
866 | *minor = BLKIF_MINOR(vdevice); | |
867 | switch (major) { | |
868 | case XEN_IDE0_MAJOR: | |
869 | *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET; | |
870 | *minor = ((*minor / 64) * PARTS_PER_DISK) + | |
871 | EMULATED_HD_DISK_MINOR_OFFSET; | |
872 | break; | |
873 | case XEN_IDE1_MAJOR: | |
874 | *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET; | |
875 | *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) + | |
876 | EMULATED_HD_DISK_MINOR_OFFSET; | |
877 | break; | |
878 | case XEN_SCSI_DISK0_MAJOR: | |
879 | *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET; | |
880 | *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET; | |
881 | break; | |
882 | case XEN_SCSI_DISK1_MAJOR: | |
883 | case XEN_SCSI_DISK2_MAJOR: | |
884 | case XEN_SCSI_DISK3_MAJOR: | |
885 | case XEN_SCSI_DISK4_MAJOR: | |
886 | case XEN_SCSI_DISK5_MAJOR: | |
887 | case XEN_SCSI_DISK6_MAJOR: | |
888 | case XEN_SCSI_DISK7_MAJOR: | |
889 | *offset = (*minor / PARTS_PER_DISK) + | |
890 | ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) + | |
891 | EMULATED_SD_DISK_NAME_OFFSET; | |
892 | *minor = *minor + | |
893 | ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) + | |
894 | EMULATED_SD_DISK_MINOR_OFFSET; | |
895 | break; | |
896 | case XEN_SCSI_DISK8_MAJOR: | |
897 | case XEN_SCSI_DISK9_MAJOR: | |
898 | case XEN_SCSI_DISK10_MAJOR: | |
899 | case XEN_SCSI_DISK11_MAJOR: | |
900 | case XEN_SCSI_DISK12_MAJOR: | |
901 | case XEN_SCSI_DISK13_MAJOR: | |
902 | case XEN_SCSI_DISK14_MAJOR: | |
903 | case XEN_SCSI_DISK15_MAJOR: | |
904 | *offset = (*minor / PARTS_PER_DISK) + | |
905 | ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) + | |
906 | EMULATED_SD_DISK_NAME_OFFSET; | |
907 | *minor = *minor + | |
908 | ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) + | |
909 | EMULATED_SD_DISK_MINOR_OFFSET; | |
910 | break; | |
911 | case XENVBD_MAJOR: | |
912 | *offset = *minor / PARTS_PER_DISK; | |
913 | break; | |
914 | default: | |
915 | printk(KERN_WARNING "blkfront: your disk configuration is " | |
916 | "incorrect, please use an xvd device instead\n"); | |
917 | return -ENODEV; | |
918 | } | |
919 | return 0; | |
920 | } | |
9f27ee59 | 921 | |
e77c78c0 JB |
922 | static char *encode_disk_name(char *ptr, unsigned int n) |
923 | { | |
924 | if (n >= 26) | |
925 | ptr = encode_disk_name(ptr, n / 26 - 1); | |
926 | *ptr = 'a' + n % 26; | |
927 | return ptr + 1; | |
928 | } | |
929 | ||
9246b5f0 CL |
930 | static int xlvbd_alloc_gendisk(blkif_sector_t capacity, |
931 | struct blkfront_info *info, | |
7c4d7d71 SB |
932 | u16 vdisk_info, u16 sector_size, |
933 | unsigned int physical_sector_size) | |
9f27ee59 JF |
934 | { |
935 | struct gendisk *gd; | |
936 | int nr_minors = 1; | |
c80a4209 | 937 | int err; |
9246b5f0 CL |
938 | unsigned int offset; |
939 | int minor; | |
940 | int nr_parts; | |
e77c78c0 | 941 | char *ptr; |
9f27ee59 JF |
942 | |
943 | BUG_ON(info->gd != NULL); | |
944 | BUG_ON(info->rq != NULL); | |
945 | ||
9246b5f0 CL |
946 | if ((info->vdevice>>EXT_SHIFT) > 1) { |
947 | /* this is above the extended range; something is wrong */ | |
948 | printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice); | |
949 | return -ENODEV; | |
950 | } | |
951 | ||
952 | if (!VDEV_IS_EXTENDED(info->vdevice)) { | |
c80a4209 SS |
953 | err = xen_translate_vdev(info->vdevice, &minor, &offset); |
954 | if (err) | |
955 | return err; | |
956 | nr_parts = PARTS_PER_DISK; | |
9246b5f0 CL |
957 | } else { |
958 | minor = BLKIF_MINOR_EXT(info->vdevice); | |
959 | nr_parts = PARTS_PER_EXT_DISK; | |
c80a4209 | 960 | offset = minor / nr_parts; |
89153b5c | 961 | if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4) |
c80a4209 SS |
962 | printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with " |
963 | "emulated IDE disks,\n\t choose an xvd device name" | |
964 | "from xvde on\n", info->vdevice); | |
9246b5f0 | 965 | } |
e77c78c0 JB |
966 | if (minor >> MINORBITS) { |
967 | pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n", | |
968 | info->vdevice, minor); | |
969 | return -ENODEV; | |
970 | } | |
9246b5f0 CL |
971 | |
972 | if ((minor % nr_parts) == 0) | |
973 | nr_minors = nr_parts; | |
9f27ee59 | 974 | |
0e345826 JB |
975 | err = xlbd_reserve_minors(minor, nr_minors); |
976 | if (err) | |
977 | goto out; | |
978 | err = -ENODEV; | |
979 | ||
9f27ee59 JF |
980 | gd = alloc_disk(nr_minors); |
981 | if (gd == NULL) | |
0e345826 | 982 | goto release; |
9f27ee59 | 983 | |
e77c78c0 JB |
984 | strcpy(gd->disk_name, DEV_NAME); |
985 | ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset); | |
986 | BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN); | |
987 | if (nr_minors > 1) | |
988 | *ptr = 0; | |
989 | else | |
990 | snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr, | |
991 | "%d", minor & (nr_parts - 1)); | |
9f27ee59 JF |
992 | |
993 | gd->major = XENVBD_MAJOR; | |
994 | gd->first_minor = minor; | |
995 | gd->fops = &xlvbd_block_fops; | |
996 | gd->private_data = info; | |
997 | gd->driverfs_dev = &(info->xbdev->dev); | |
998 | set_capacity(gd, capacity); | |
999 | ||
7c4d7d71 | 1000 | if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size, |
402b27f9 RPM |
1001 | info->max_indirect_segments ? : |
1002 | BLKIF_MAX_SEGMENTS_PER_REQUEST)) { | |
9f27ee59 | 1003 | del_gendisk(gd); |
0e345826 | 1004 | goto release; |
9f27ee59 JF |
1005 | } |
1006 | ||
1007 | info->rq = gd->queue; | |
1008 | info->gd = gd; | |
1009 | ||
4913efe4 | 1010 | xlvbd_flush(info); |
9f27ee59 JF |
1011 | |
1012 | if (vdisk_info & VDISK_READONLY) | |
1013 | set_disk_ro(gd, 1); | |
1014 | ||
1015 | if (vdisk_info & VDISK_REMOVABLE) | |
1016 | gd->flags |= GENHD_FL_REMOVABLE; | |
1017 | ||
1018 | if (vdisk_info & VDISK_CDROM) | |
1019 | gd->flags |= GENHD_FL_CD; | |
1020 | ||
1021 | return 0; | |
1022 | ||
0e345826 JB |
1023 | release: |
1024 | xlbd_release_minors(minor, nr_minors); | |
9f27ee59 JF |
1025 | out: |
1026 | return err; | |
1027 | } | |
1028 | ||
a66b5aeb DS |
1029 | static void xlvbd_release_gendisk(struct blkfront_info *info) |
1030 | { | |
1031 | unsigned int minor, nr_minors; | |
a66b5aeb DS |
1032 | |
1033 | if (info->rq == NULL) | |
1034 | return; | |
1035 | ||
a66b5aeb | 1036 | /* No more blkif_request(). */ |
907c3eb1 | 1037 | blk_mq_stop_hw_queues(info->rq); |
a66b5aeb DS |
1038 | |
1039 | /* No more gnttab callback work. */ | |
1040 | gnttab_cancel_free_callback(&info->callback); | |
a66b5aeb DS |
1041 | |
1042 | /* Flush gnttab callback work. Must be done with no locks held. */ | |
43829731 | 1043 | flush_work(&info->work); |
a66b5aeb DS |
1044 | |
1045 | del_gendisk(info->gd); | |
1046 | ||
1047 | minor = info->gd->first_minor; | |
1048 | nr_minors = info->gd->minors; | |
1049 | xlbd_release_minors(minor, nr_minors); | |
1050 | ||
1051 | blk_cleanup_queue(info->rq); | |
907c3eb1 | 1052 | blk_mq_free_tag_set(&info->tag_set); |
a66b5aeb DS |
1053 | info->rq = NULL; |
1054 | ||
1055 | put_disk(info->gd); | |
1056 | info->gd = NULL; | |
1057 | } | |
1058 | ||
907c3eb1 | 1059 | /* Must be called with io_lock holded */ |
9f27ee59 JF |
1060 | static void kick_pending_request_queues(struct blkfront_info *info) |
1061 | { | |
907c3eb1 BL |
1062 | if (!RING_FULL(&info->ring)) |
1063 | blk_mq_start_stopped_hw_queues(info->rq, true); | |
9f27ee59 JF |
1064 | } |
1065 | ||
1066 | static void blkif_restart_queue(struct work_struct *work) | |
1067 | { | |
1068 | struct blkfront_info *info = container_of(work, struct blkfront_info, work); | |
1069 | ||
3467811e | 1070 | spin_lock_irq(&info->io_lock); |
9f27ee59 JF |
1071 | if (info->connected == BLKIF_STATE_CONNECTED) |
1072 | kick_pending_request_queues(info); | |
3467811e | 1073 | spin_unlock_irq(&info->io_lock); |
9f27ee59 JF |
1074 | } |
1075 | ||
1076 | static void blkif_free(struct blkfront_info *info, int suspend) | |
1077 | { | |
155b7edb RPM |
1078 | struct grant *persistent_gnt; |
1079 | struct grant *n; | |
402b27f9 | 1080 | int i, j, segs; |
0a8704a5 | 1081 | |
9f27ee59 | 1082 | /* Prevent new requests being issued until we fix things up. */ |
3467811e | 1083 | spin_lock_irq(&info->io_lock); |
9f27ee59 JF |
1084 | info->connected = suspend ? |
1085 | BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED; | |
1086 | /* No more blkif_request(). */ | |
1087 | if (info->rq) | |
907c3eb1 | 1088 | blk_mq_stop_hw_queues(info->rq); |
0a8704a5 RPM |
1089 | |
1090 | /* Remove all persistent grants */ | |
bfe11d6d | 1091 | if (!list_empty(&info->grants)) { |
155b7edb | 1092 | list_for_each_entry_safe(persistent_gnt, n, |
bfe11d6d | 1093 | &info->grants, node) { |
155b7edb | 1094 | list_del(&persistent_gnt->node); |
9c1e050c RPM |
1095 | if (persistent_gnt->gref != GRANT_INVALID_REF) { |
1096 | gnttab_end_foreign_access(persistent_gnt->gref, | |
1097 | 0, 0UL); | |
1098 | info->persistent_gnts_c--; | |
1099 | } | |
bfe11d6d | 1100 | if (info->feature_persistent) |
a7a6df22 | 1101 | __free_page(persistent_gnt->page); |
155b7edb | 1102 | kfree(persistent_gnt); |
0a8704a5 | 1103 | } |
0a8704a5 | 1104 | } |
9c1e050c | 1105 | BUG_ON(info->persistent_gnts_c != 0); |
0a8704a5 | 1106 | |
bfe11d6d RPM |
1107 | /* |
1108 | * Remove indirect pages, this only happens when using indirect | |
1109 | * descriptors but not persistent grants | |
1110 | */ | |
1111 | if (!list_empty(&info->indirect_pages)) { | |
1112 | struct page *indirect_page, *n; | |
1113 | ||
1114 | BUG_ON(info->feature_persistent); | |
1115 | list_for_each_entry_safe(indirect_page, n, &info->indirect_pages, lru) { | |
1116 | list_del(&indirect_page->lru); | |
1117 | __free_page(indirect_page); | |
1118 | } | |
1119 | } | |
1120 | ||
86839c56 | 1121 | for (i = 0; i < BLK_RING_SIZE(info); i++) { |
402b27f9 RPM |
1122 | /* |
1123 | * Clear persistent grants present in requests already | |
1124 | * on the shared ring | |
1125 | */ | |
1126 | if (!info->shadow[i].request) | |
1127 | goto free_shadow; | |
1128 | ||
1129 | segs = info->shadow[i].req.operation == BLKIF_OP_INDIRECT ? | |
1130 | info->shadow[i].req.u.indirect.nr_segments : | |
1131 | info->shadow[i].req.u.rw.nr_segments; | |
1132 | for (j = 0; j < segs; j++) { | |
1133 | persistent_gnt = info->shadow[i].grants_used[j]; | |
1134 | gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL); | |
bfe11d6d | 1135 | if (info->feature_persistent) |
a7a6df22 | 1136 | __free_page(persistent_gnt->page); |
402b27f9 RPM |
1137 | kfree(persistent_gnt); |
1138 | } | |
1139 | ||
1140 | if (info->shadow[i].req.operation != BLKIF_OP_INDIRECT) | |
1141 | /* | |
1142 | * If this is not an indirect operation don't try to | |
1143 | * free indirect segments | |
1144 | */ | |
1145 | goto free_shadow; | |
1146 | ||
1147 | for (j = 0; j < INDIRECT_GREFS(segs); j++) { | |
1148 | persistent_gnt = info->shadow[i].indirect_grants[j]; | |
1149 | gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL); | |
a7a6df22 | 1150 | __free_page(persistent_gnt->page); |
402b27f9 RPM |
1151 | kfree(persistent_gnt); |
1152 | } | |
1153 | ||
1154 | free_shadow: | |
1155 | kfree(info->shadow[i].grants_used); | |
1156 | info->shadow[i].grants_used = NULL; | |
1157 | kfree(info->shadow[i].indirect_grants); | |
1158 | info->shadow[i].indirect_grants = NULL; | |
b7649158 RPM |
1159 | kfree(info->shadow[i].sg); |
1160 | info->shadow[i].sg = NULL; | |
402b27f9 RPM |
1161 | } |
1162 | ||
9f27ee59 JF |
1163 | /* No more gnttab callback work. */ |
1164 | gnttab_cancel_free_callback(&info->callback); | |
3467811e | 1165 | spin_unlock_irq(&info->io_lock); |
9f27ee59 JF |
1166 | |
1167 | /* Flush gnttab callback work. Must be done with no locks held. */ | |
43829731 | 1168 | flush_work(&info->work); |
9f27ee59 JF |
1169 | |
1170 | /* Free resources associated with old device channel. */ | |
86839c56 BL |
1171 | for (i = 0; i < info->nr_ring_pages; i++) { |
1172 | if (info->ring_ref[i] != GRANT_INVALID_REF) { | |
1173 | gnttab_end_foreign_access(info->ring_ref[i], 0, 0); | |
1174 | info->ring_ref[i] = GRANT_INVALID_REF; | |
1175 | } | |
9f27ee59 | 1176 | } |
86839c56 BL |
1177 | free_pages((unsigned long)info->ring.sring, get_order(info->nr_ring_pages * PAGE_SIZE)); |
1178 | info->ring.sring = NULL; | |
1179 | ||
9f27ee59 JF |
1180 | if (info->irq) |
1181 | unbind_from_irqhandler(info->irq, info); | |
1182 | info->evtchn = info->irq = 0; | |
1183 | ||
1184 | } | |
1185 | ||
c004a6fe JG |
1186 | struct copy_from_grant { |
1187 | const struct blk_shadow *s; | |
1188 | unsigned int grant_idx; | |
1189 | unsigned int bvec_offset; | |
1190 | char *bvec_data; | |
1191 | }; | |
1192 | ||
1193 | static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset, | |
1194 | unsigned int len, void *data) | |
1195 | { | |
1196 | struct copy_from_grant *info = data; | |
1197 | char *shared_data; | |
1198 | /* Convenient aliases */ | |
1199 | const struct blk_shadow *s = info->s; | |
1200 | ||
1201 | shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page); | |
1202 | ||
1203 | memcpy(info->bvec_data + info->bvec_offset, | |
1204 | shared_data + offset, len); | |
1205 | ||
1206 | info->bvec_offset += len; | |
1207 | info->grant_idx++; | |
1208 | ||
1209 | kunmap_atomic(shared_data); | |
1210 | } | |
1211 | ||
0a8704a5 RPM |
1212 | static void blkif_completion(struct blk_shadow *s, struct blkfront_info *info, |
1213 | struct blkif_response *bret) | |
9f27ee59 | 1214 | { |
d62f6918 | 1215 | int i = 0; |
b7649158 | 1216 | struct scatterlist *sg; |
c004a6fe JG |
1217 | int num_sg, num_grant; |
1218 | struct copy_from_grant data = { | |
1219 | .s = s, | |
1220 | .grant_idx = 0, | |
1221 | }; | |
402b27f9 | 1222 | |
c004a6fe | 1223 | num_grant = s->req.operation == BLKIF_OP_INDIRECT ? |
402b27f9 | 1224 | s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments; |
c004a6fe | 1225 | num_sg = s->num_sg; |
0a8704a5 | 1226 | |
bfe11d6d | 1227 | if (bret->operation == BLKIF_OP_READ && info->feature_persistent) { |
c004a6fe | 1228 | for_each_sg(s->sg, sg, num_sg, i) { |
b7649158 | 1229 | BUG_ON(sg->offset + sg->length > PAGE_SIZE); |
c004a6fe JG |
1230 | |
1231 | data.bvec_offset = sg->offset; | |
1232 | data.bvec_data = kmap_atomic(sg_page(sg)); | |
1233 | ||
1234 | gnttab_foreach_grant_in_range(sg_page(sg), | |
1235 | sg->offset, | |
1236 | sg->length, | |
1237 | blkif_copy_from_grant, | |
1238 | &data); | |
1239 | ||
1240 | kunmap_atomic(data.bvec_data); | |
0a8704a5 RPM |
1241 | } |
1242 | } | |
1243 | /* Add the persistent grant into the list of free grants */ | |
c004a6fe | 1244 | for (i = 0; i < num_grant; i++) { |
fbe363c4 RPM |
1245 | if (gnttab_query_foreign_access(s->grants_used[i]->gref)) { |
1246 | /* | |
1247 | * If the grant is still mapped by the backend (the | |
1248 | * backend has chosen to make this grant persistent) | |
1249 | * we add it at the head of the list, so it will be | |
1250 | * reused first. | |
1251 | */ | |
bfe11d6d RPM |
1252 | if (!info->feature_persistent) |
1253 | pr_alert_ratelimited("backed has not unmapped grant: %u\n", | |
1254 | s->grants_used[i]->gref); | |
1255 | list_add(&s->grants_used[i]->node, &info->grants); | |
fbe363c4 RPM |
1256 | info->persistent_gnts_c++; |
1257 | } else { | |
1258 | /* | |
1259 | * If the grant is not mapped by the backend we end the | |
1260 | * foreign access and add it to the tail of the list, | |
1261 | * so it will not be picked again unless we run out of | |
1262 | * persistent grants. | |
1263 | */ | |
1264 | gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL); | |
1265 | s->grants_used[i]->gref = GRANT_INVALID_REF; | |
bfe11d6d | 1266 | list_add_tail(&s->grants_used[i]->node, &info->grants); |
fbe363c4 | 1267 | } |
0a8704a5 | 1268 | } |
402b27f9 | 1269 | if (s->req.operation == BLKIF_OP_INDIRECT) { |
c004a6fe | 1270 | for (i = 0; i < INDIRECT_GREFS(num_grant); i++) { |
fbe363c4 | 1271 | if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) { |
bfe11d6d RPM |
1272 | if (!info->feature_persistent) |
1273 | pr_alert_ratelimited("backed has not unmapped grant: %u\n", | |
1274 | s->indirect_grants[i]->gref); | |
1275 | list_add(&s->indirect_grants[i]->node, &info->grants); | |
fbe363c4 RPM |
1276 | info->persistent_gnts_c++; |
1277 | } else { | |
bfe11d6d RPM |
1278 | struct page *indirect_page; |
1279 | ||
fbe363c4 | 1280 | gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL); |
bfe11d6d RPM |
1281 | /* |
1282 | * Add the used indirect page back to the list of | |
1283 | * available pages for indirect grefs. | |
1284 | */ | |
7b076750 | 1285 | if (!info->feature_persistent) { |
a7a6df22 | 1286 | indirect_page = s->indirect_grants[i]->page; |
7b076750 BL |
1287 | list_add(&indirect_page->lru, &info->indirect_pages); |
1288 | } | |
fbe363c4 | 1289 | s->indirect_grants[i]->gref = GRANT_INVALID_REF; |
bfe11d6d | 1290 | list_add_tail(&s->indirect_grants[i]->node, &info->grants); |
fbe363c4 | 1291 | } |
402b27f9 RPM |
1292 | } |
1293 | } | |
9f27ee59 JF |
1294 | } |
1295 | ||
1296 | static irqreturn_t blkif_interrupt(int irq, void *dev_id) | |
1297 | { | |
1298 | struct request *req; | |
1299 | struct blkif_response *bret; | |
1300 | RING_IDX i, rp; | |
1301 | unsigned long flags; | |
1302 | struct blkfront_info *info = (struct blkfront_info *)dev_id; | |
f4829a9b | 1303 | int error; |
9f27ee59 | 1304 | |
3467811e | 1305 | spin_lock_irqsave(&info->io_lock, flags); |
9f27ee59 JF |
1306 | |
1307 | if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) { | |
3467811e | 1308 | spin_unlock_irqrestore(&info->io_lock, flags); |
9f27ee59 JF |
1309 | return IRQ_HANDLED; |
1310 | } | |
1311 | ||
1312 | again: | |
1313 | rp = info->ring.sring->rsp_prod; | |
1314 | rmb(); /* Ensure we see queued responses up to 'rp'. */ | |
1315 | ||
1316 | for (i = info->ring.rsp_cons; i != rp; i++) { | |
1317 | unsigned long id; | |
9f27ee59 JF |
1318 | |
1319 | bret = RING_GET_RESPONSE(&info->ring, i); | |
1320 | id = bret->id; | |
6878c32e KRW |
1321 | /* |
1322 | * The backend has messed up and given us an id that we would | |
1323 | * never have given to it (we stamp it up to BLK_RING_SIZE - | |
1324 | * look in get_id_from_freelist. | |
1325 | */ | |
86839c56 | 1326 | if (id >= BLK_RING_SIZE(info)) { |
6878c32e KRW |
1327 | WARN(1, "%s: response to %s has incorrect id (%ld)\n", |
1328 | info->gd->disk_name, op_name(bret->operation), id); | |
1329 | /* We can't safely get the 'struct request' as | |
1330 | * the id is busted. */ | |
1331 | continue; | |
1332 | } | |
a945b980 | 1333 | req = info->shadow[id].request; |
9f27ee59 | 1334 | |
5ea42986 | 1335 | if (bret->operation != BLKIF_OP_DISCARD) |
0a8704a5 | 1336 | blkif_completion(&info->shadow[id], info, bret); |
9f27ee59 | 1337 | |
6878c32e KRW |
1338 | if (add_id_to_freelist(info, id)) { |
1339 | WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n", | |
1340 | info->gd->disk_name, op_name(bret->operation), id); | |
1341 | continue; | |
1342 | } | |
9f27ee59 | 1343 | |
f4829a9b | 1344 | error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO; |
9f27ee59 | 1345 | switch (bret->operation) { |
ed30bf31 LD |
1346 | case BLKIF_OP_DISCARD: |
1347 | if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) { | |
1348 | struct request_queue *rq = info->rq; | |
6878c32e KRW |
1349 | printk(KERN_WARNING "blkfront: %s: %s op failed\n", |
1350 | info->gd->disk_name, op_name(bret->operation)); | |
f4829a9b | 1351 | error = -EOPNOTSUPP; |
ed30bf31 | 1352 | info->feature_discard = 0; |
5ea42986 | 1353 | info->feature_secdiscard = 0; |
ed30bf31 | 1354 | queue_flag_clear(QUEUE_FLAG_DISCARD, rq); |
5ea42986 | 1355 | queue_flag_clear(QUEUE_FLAG_SECDISCARD, rq); |
ed30bf31 | 1356 | } |
f4829a9b | 1357 | blk_mq_complete_request(req, error); |
ed30bf31 | 1358 | break; |
edf6ef59 | 1359 | case BLKIF_OP_FLUSH_DISKCACHE: |
9f27ee59 JF |
1360 | case BLKIF_OP_WRITE_BARRIER: |
1361 | if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) { | |
6878c32e KRW |
1362 | printk(KERN_WARNING "blkfront: %s: %s op failed\n", |
1363 | info->gd->disk_name, op_name(bret->operation)); | |
f4829a9b | 1364 | error = -EOPNOTSUPP; |
dcb8baec JF |
1365 | } |
1366 | if (unlikely(bret->status == BLKIF_RSP_ERROR && | |
97e36834 | 1367 | info->shadow[id].req.u.rw.nr_segments == 0)) { |
6878c32e KRW |
1368 | printk(KERN_WARNING "blkfront: %s: empty %s op failed\n", |
1369 | info->gd->disk_name, op_name(bret->operation)); | |
f4829a9b | 1370 | error = -EOPNOTSUPP; |
dcb8baec | 1371 | } |
f4829a9b CH |
1372 | if (unlikely(error)) { |
1373 | if (error == -EOPNOTSUPP) | |
1374 | error = 0; | |
4913efe4 TH |
1375 | info->feature_flush = 0; |
1376 | xlvbd_flush(info); | |
9f27ee59 JF |
1377 | } |
1378 | /* fall through */ | |
1379 | case BLKIF_OP_READ: | |
1380 | case BLKIF_OP_WRITE: | |
1381 | if (unlikely(bret->status != BLKIF_RSP_OKAY)) | |
1382 | dev_dbg(&info->xbdev->dev, "Bad return from blkdev data " | |
1383 | "request: %x\n", bret->status); | |
1384 | ||
f4829a9b | 1385 | blk_mq_complete_request(req, error); |
9f27ee59 JF |
1386 | break; |
1387 | default: | |
1388 | BUG(); | |
1389 | } | |
1390 | } | |
1391 | ||
1392 | info->ring.rsp_cons = i; | |
1393 | ||
1394 | if (i != info->ring.req_prod_pvt) { | |
1395 | int more_to_do; | |
1396 | RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do); | |
1397 | if (more_to_do) | |
1398 | goto again; | |
1399 | } else | |
1400 | info->ring.sring->rsp_event = i + 1; | |
1401 | ||
1402 | kick_pending_request_queues(info); | |
1403 | ||
3467811e | 1404 | spin_unlock_irqrestore(&info->io_lock, flags); |
9f27ee59 JF |
1405 | |
1406 | return IRQ_HANDLED; | |
1407 | } | |
1408 | ||
1409 | ||
1410 | static int setup_blkring(struct xenbus_device *dev, | |
1411 | struct blkfront_info *info) | |
1412 | { | |
1413 | struct blkif_sring *sring; | |
86839c56 | 1414 | int err, i; |
c004a6fe | 1415 | unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE; |
9cce2914 | 1416 | grant_ref_t gref[XENBUS_MAX_RING_GRANTS]; |
9f27ee59 | 1417 | |
86839c56 BL |
1418 | for (i = 0; i < info->nr_ring_pages; i++) |
1419 | info->ring_ref[i] = GRANT_INVALID_REF; | |
9f27ee59 | 1420 | |
86839c56 BL |
1421 | sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH, |
1422 | get_order(ring_size)); | |
9f27ee59 JF |
1423 | if (!sring) { |
1424 | xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring"); | |
1425 | return -ENOMEM; | |
1426 | } | |
1427 | SHARED_RING_INIT(sring); | |
86839c56 | 1428 | FRONT_RING_INIT(&info->ring, sring, ring_size); |
9e973e64 | 1429 | |
86839c56 | 1430 | err = xenbus_grant_ring(dev, info->ring.sring, info->nr_ring_pages, gref); |
9f27ee59 | 1431 | if (err < 0) { |
86839c56 | 1432 | free_pages((unsigned long)sring, get_order(ring_size)); |
9f27ee59 JF |
1433 | info->ring.sring = NULL; |
1434 | goto fail; | |
1435 | } | |
86839c56 BL |
1436 | for (i = 0; i < info->nr_ring_pages; i++) |
1437 | info->ring_ref[i] = gref[i]; | |
9f27ee59 JF |
1438 | |
1439 | err = xenbus_alloc_evtchn(dev, &info->evtchn); | |
1440 | if (err) | |
1441 | goto fail; | |
1442 | ||
89c30f16 TT |
1443 | err = bind_evtchn_to_irqhandler(info->evtchn, blkif_interrupt, 0, |
1444 | "blkif", info); | |
9f27ee59 JF |
1445 | if (err <= 0) { |
1446 | xenbus_dev_fatal(dev, err, | |
1447 | "bind_evtchn_to_irqhandler failed"); | |
1448 | goto fail; | |
1449 | } | |
1450 | info->irq = err; | |
1451 | ||
1452 | return 0; | |
1453 | fail: | |
1454 | blkif_free(info, 0); | |
1455 | return err; | |
1456 | } | |
1457 | ||
1458 | ||
1459 | /* Common code used when first setting up, and when resuming. */ | |
203fd61f | 1460 | static int talk_to_blkback(struct xenbus_device *dev, |
9f27ee59 JF |
1461 | struct blkfront_info *info) |
1462 | { | |
1463 | const char *message = NULL; | |
1464 | struct xenbus_transaction xbt; | |
86839c56 BL |
1465 | int err, i; |
1466 | unsigned int max_page_order = 0; | |
1467 | unsigned int ring_page_order = 0; | |
1468 | ||
1469 | err = xenbus_scanf(XBT_NIL, info->xbdev->otherend, | |
1470 | "max-ring-page-order", "%u", &max_page_order); | |
1471 | if (err != 1) | |
1472 | info->nr_ring_pages = 1; | |
1473 | else { | |
1474 | ring_page_order = min(xen_blkif_max_ring_order, max_page_order); | |
1475 | info->nr_ring_pages = 1 << ring_page_order; | |
1476 | } | |
9f27ee59 JF |
1477 | |
1478 | /* Create shared ring, alloc event channel. */ | |
1479 | err = setup_blkring(dev, info); | |
1480 | if (err) | |
1481 | goto out; | |
1482 | ||
1483 | again: | |
1484 | err = xenbus_transaction_start(&xbt); | |
1485 | if (err) { | |
1486 | xenbus_dev_fatal(dev, err, "starting transaction"); | |
1487 | goto destroy_blkring; | |
1488 | } | |
1489 | ||
86839c56 BL |
1490 | if (info->nr_ring_pages == 1) { |
1491 | err = xenbus_printf(xbt, dev->nodename, | |
1492 | "ring-ref", "%u", info->ring_ref[0]); | |
1493 | if (err) { | |
1494 | message = "writing ring-ref"; | |
1495 | goto abort_transaction; | |
1496 | } | |
1497 | } else { | |
1498 | err = xenbus_printf(xbt, dev->nodename, | |
1499 | "ring-page-order", "%u", ring_page_order); | |
1500 | if (err) { | |
1501 | message = "writing ring-page-order"; | |
1502 | goto abort_transaction; | |
1503 | } | |
1504 | ||
1505 | for (i = 0; i < info->nr_ring_pages; i++) { | |
1506 | char ring_ref_name[RINGREF_NAME_LEN]; | |
1507 | ||
1508 | snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i); | |
1509 | err = xenbus_printf(xbt, dev->nodename, ring_ref_name, | |
1510 | "%u", info->ring_ref[i]); | |
1511 | if (err) { | |
1512 | message = "writing ring-ref"; | |
1513 | goto abort_transaction; | |
1514 | } | |
1515 | } | |
9f27ee59 JF |
1516 | } |
1517 | err = xenbus_printf(xbt, dev->nodename, | |
1518 | "event-channel", "%u", info->evtchn); | |
1519 | if (err) { | |
1520 | message = "writing event-channel"; | |
1521 | goto abort_transaction; | |
1522 | } | |
3e334239 MA |
1523 | err = xenbus_printf(xbt, dev->nodename, "protocol", "%s", |
1524 | XEN_IO_PROTO_ABI_NATIVE); | |
1525 | if (err) { | |
1526 | message = "writing protocol"; | |
1527 | goto abort_transaction; | |
1528 | } | |
0a8704a5 | 1529 | err = xenbus_printf(xbt, dev->nodename, |
cb5bd4d1 | 1530 | "feature-persistent", "%u", 1); |
0a8704a5 RPM |
1531 | if (err) |
1532 | dev_warn(&dev->dev, | |
1533 | "writing persistent grants feature to xenbus"); | |
9f27ee59 JF |
1534 | |
1535 | err = xenbus_transaction_end(xbt, 0); | |
1536 | if (err) { | |
1537 | if (err == -EAGAIN) | |
1538 | goto again; | |
1539 | xenbus_dev_fatal(dev, err, "completing transaction"); | |
1540 | goto destroy_blkring; | |
1541 | } | |
1542 | ||
86839c56 BL |
1543 | for (i = 0; i < BLK_RING_SIZE(info); i++) |
1544 | info->shadow[i].req.u.rw.id = i+1; | |
1545 | info->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff; | |
9f27ee59 JF |
1546 | xenbus_switch_state(dev, XenbusStateInitialised); |
1547 | ||
1548 | return 0; | |
1549 | ||
1550 | abort_transaction: | |
1551 | xenbus_transaction_end(xbt, 1); | |
1552 | if (message) | |
1553 | xenbus_dev_fatal(dev, err, "%s", message); | |
1554 | destroy_blkring: | |
1555 | blkif_free(info, 0); | |
1556 | out: | |
1557 | return err; | |
1558 | } | |
1559 | ||
9f27ee59 JF |
1560 | /** |
1561 | * Entry point to this code when a new device is created. Allocate the basic | |
1562 | * structures and the ring buffer for communication with the backend, and | |
1563 | * inform the backend of the appropriate details for those. Switch to | |
1564 | * Initialised state. | |
1565 | */ | |
1566 | static int blkfront_probe(struct xenbus_device *dev, | |
1567 | const struct xenbus_device_id *id) | |
1568 | { | |
86839c56 | 1569 | int err, vdevice; |
9f27ee59 JF |
1570 | struct blkfront_info *info; |
1571 | ||
1572 | /* FIXME: Use dynamic device id if this is not set. */ | |
1573 | err = xenbus_scanf(XBT_NIL, dev->nodename, | |
1574 | "virtual-device", "%i", &vdevice); | |
1575 | if (err != 1) { | |
9246b5f0 CL |
1576 | /* go looking in the extended area instead */ |
1577 | err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext", | |
1578 | "%i", &vdevice); | |
1579 | if (err != 1) { | |
1580 | xenbus_dev_fatal(dev, err, "reading virtual-device"); | |
1581 | return err; | |
1582 | } | |
9f27ee59 JF |
1583 | } |
1584 | ||
b98a409b SS |
1585 | if (xen_hvm_domain()) { |
1586 | char *type; | |
1587 | int len; | |
1588 | /* no unplug has been done: do not hook devices != xen vbds */ | |
51c71a3b | 1589 | if (xen_has_pv_and_legacy_disk_devices()) { |
b98a409b SS |
1590 | int major; |
1591 | ||
1592 | if (!VDEV_IS_EXTENDED(vdevice)) | |
1593 | major = BLKIF_MAJOR(vdevice); | |
1594 | else | |
1595 | major = XENVBD_MAJOR; | |
1596 | ||
1597 | if (major != XENVBD_MAJOR) { | |
1598 | printk(KERN_INFO | |
1599 | "%s: HVM does not support vbd %d as xen block device\n", | |
02f1f217 | 1600 | __func__, vdevice); |
b98a409b SS |
1601 | return -ENODEV; |
1602 | } | |
1603 | } | |
1604 | /* do not create a PV cdrom device if we are an HVM guest */ | |
1605 | type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len); | |
1606 | if (IS_ERR(type)) | |
1607 | return -ENODEV; | |
1608 | if (strncmp(type, "cdrom", 5) == 0) { | |
1609 | kfree(type); | |
c1c5413a SS |
1610 | return -ENODEV; |
1611 | } | |
b98a409b | 1612 | kfree(type); |
c1c5413a | 1613 | } |
9f27ee59 JF |
1614 | info = kzalloc(sizeof(*info), GFP_KERNEL); |
1615 | if (!info) { | |
1616 | xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure"); | |
1617 | return -ENOMEM; | |
1618 | } | |
1619 | ||
b70f5fa0 | 1620 | mutex_init(&info->mutex); |
3467811e | 1621 | spin_lock_init(&info->io_lock); |
9f27ee59 JF |
1622 | info->xbdev = dev; |
1623 | info->vdevice = vdevice; | |
bfe11d6d RPM |
1624 | INIT_LIST_HEAD(&info->grants); |
1625 | INIT_LIST_HEAD(&info->indirect_pages); | |
0a8704a5 | 1626 | info->persistent_gnts_c = 0; |
9f27ee59 JF |
1627 | info->connected = BLKIF_STATE_DISCONNECTED; |
1628 | INIT_WORK(&info->work, blkif_restart_queue); | |
1629 | ||
9f27ee59 JF |
1630 | /* Front end dir is a number, which is used as the id. */ |
1631 | info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0); | |
a1b4b12b | 1632 | dev_set_drvdata(&dev->dev, info); |
9f27ee59 | 1633 | |
9f27ee59 JF |
1634 | return 0; |
1635 | } | |
1636 | ||
4246a0b6 | 1637 | static void split_bio_end(struct bio *bio) |
402b27f9 RPM |
1638 | { |
1639 | struct split_bio *split_bio = bio->bi_private; | |
1640 | ||
402b27f9 RPM |
1641 | if (atomic_dec_and_test(&split_bio->pending)) { |
1642 | split_bio->bio->bi_phys_segments = 0; | |
4246a0b6 CH |
1643 | split_bio->bio->bi_error = bio->bi_error; |
1644 | bio_endio(split_bio->bio); | |
402b27f9 RPM |
1645 | kfree(split_bio); |
1646 | } | |
1647 | bio_put(bio); | |
1648 | } | |
9f27ee59 JF |
1649 | |
1650 | static int blkif_recover(struct blkfront_info *info) | |
1651 | { | |
1652 | int i; | |
402b27f9 | 1653 | struct request *req, *n; |
9f27ee59 | 1654 | struct blk_shadow *copy; |
402b27f9 RPM |
1655 | int rc; |
1656 | struct bio *bio, *cloned_bio; | |
1657 | struct bio_list bio_list, merge_bio; | |
1658 | unsigned int segs, offset; | |
1659 | int pending, size; | |
1660 | struct split_bio *split_bio; | |
1661 | struct list_head requests; | |
9f27ee59 JF |
1662 | |
1663 | /* Stage 1: Make a safe copy of the shadow state. */ | |
29d0b218 | 1664 | copy = kmemdup(info->shadow, sizeof(info->shadow), |
a144ff09 | 1665 | GFP_NOIO | __GFP_REPEAT | __GFP_HIGH); |
9f27ee59 JF |
1666 | if (!copy) |
1667 | return -ENOMEM; | |
9f27ee59 JF |
1668 | |
1669 | /* Stage 2: Set up free list. */ | |
1670 | memset(&info->shadow, 0, sizeof(info->shadow)); | |
86839c56 | 1671 | for (i = 0; i < BLK_RING_SIZE(info); i++) |
97e36834 | 1672 | info->shadow[i].req.u.rw.id = i+1; |
9f27ee59 | 1673 | info->shadow_free = info->ring.req_prod_pvt; |
86839c56 | 1674 | info->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff; |
9f27ee59 | 1675 | |
d50babbe | 1676 | rc = blkfront_gather_backend_features(info); |
402b27f9 RPM |
1677 | if (rc) { |
1678 | kfree(copy); | |
1679 | return rc; | |
1680 | } | |
1681 | ||
1682 | segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST; | |
1683 | blk_queue_max_segments(info->rq, segs); | |
1684 | bio_list_init(&bio_list); | |
1685 | INIT_LIST_HEAD(&requests); | |
86839c56 | 1686 | for (i = 0; i < BLK_RING_SIZE(info); i++) { |
9f27ee59 | 1687 | /* Not in use? */ |
a945b980 | 1688 | if (!copy[i].request) |
9f27ee59 JF |
1689 | continue; |
1690 | ||
402b27f9 RPM |
1691 | /* |
1692 | * Get the bios in the request so we can re-queue them. | |
1693 | */ | |
1694 | if (copy[i].request->cmd_flags & | |
1695 | (REQ_FLUSH | REQ_FUA | REQ_DISCARD | REQ_SECURE)) { | |
1696 | /* | |
1697 | * Flush operations don't contain bios, so | |
1698 | * we need to requeue the whole request | |
1699 | */ | |
1700 | list_add(©[i].request->queuelist, &requests); | |
1701 | continue; | |
5ea42986 | 1702 | } |
402b27f9 RPM |
1703 | merge_bio.head = copy[i].request->bio; |
1704 | merge_bio.tail = copy[i].request->biotail; | |
1705 | bio_list_merge(&bio_list, &merge_bio); | |
1706 | copy[i].request->bio = NULL; | |
3bb8c98e | 1707 | blk_end_request_all(copy[i].request, 0); |
9f27ee59 JF |
1708 | } |
1709 | ||
1710 | kfree(copy); | |
1711 | ||
1712 | xenbus_switch_state(info->xbdev, XenbusStateConnected); | |
1713 | ||
3467811e | 1714 | spin_lock_irq(&info->io_lock); |
9f27ee59 JF |
1715 | |
1716 | /* Now safe for us to use the shared ring */ | |
1717 | info->connected = BLKIF_STATE_CONNECTED; | |
1718 | ||
9f27ee59 JF |
1719 | /* Kick any other new requests queued since we resumed */ |
1720 | kick_pending_request_queues(info); | |
1721 | ||
402b27f9 RPM |
1722 | list_for_each_entry_safe(req, n, &requests, queuelist) { |
1723 | /* Requeue pending requests (flush or discard) */ | |
1724 | list_del_init(&req->queuelist); | |
1725 | BUG_ON(req->nr_phys_segments > segs); | |
907c3eb1 | 1726 | blk_mq_requeue_request(req); |
402b27f9 | 1727 | } |
3467811e | 1728 | spin_unlock_irq(&info->io_lock); |
907c3eb1 | 1729 | blk_mq_kick_requeue_list(info->rq); |
9f27ee59 | 1730 | |
402b27f9 RPM |
1731 | while ((bio = bio_list_pop(&bio_list)) != NULL) { |
1732 | /* Traverse the list of pending bios and re-queue them */ | |
1733 | if (bio_segments(bio) > segs) { | |
1734 | /* | |
1735 | * This bio has more segments than what we can | |
1736 | * handle, we have to split it. | |
1737 | */ | |
1738 | pending = (bio_segments(bio) + segs - 1) / segs; | |
1739 | split_bio = kzalloc(sizeof(*split_bio), GFP_NOIO); | |
1740 | BUG_ON(split_bio == NULL); | |
1741 | atomic_set(&split_bio->pending, pending); | |
1742 | split_bio->bio = bio; | |
1743 | for (i = 0; i < pending; i++) { | |
c004a6fe JG |
1744 | offset = (i * segs * XEN_PAGE_SIZE) >> 9; |
1745 | size = min((unsigned int)(segs * XEN_PAGE_SIZE) >> 9, | |
4f024f37 | 1746 | (unsigned int)bio_sectors(bio) - offset); |
402b27f9 RPM |
1747 | cloned_bio = bio_clone(bio, GFP_NOIO); |
1748 | BUG_ON(cloned_bio == NULL); | |
6678d83f | 1749 | bio_trim(cloned_bio, offset, size); |
402b27f9 RPM |
1750 | cloned_bio->bi_private = split_bio; |
1751 | cloned_bio->bi_end_io = split_bio_end; | |
1752 | submit_bio(cloned_bio->bi_rw, cloned_bio); | |
1753 | } | |
1754 | /* | |
1755 | * Now we have to wait for all those smaller bios to | |
1756 | * end, so we can also end the "parent" bio. | |
1757 | */ | |
1758 | continue; | |
1759 | } | |
1760 | /* We don't need to split this bio */ | |
1761 | submit_bio(bio->bi_rw, bio); | |
1762 | } | |
1763 | ||
9f27ee59 JF |
1764 | return 0; |
1765 | } | |
1766 | ||
1767 | /** | |
1768 | * We are reconnecting to the backend, due to a suspend/resume, or a backend | |
1769 | * driver restart. We tear down our blkif structure and recreate it, but | |
1770 | * leave the device-layer structures intact so that this is transparent to the | |
1771 | * rest of the kernel. | |
1772 | */ | |
1773 | static int blkfront_resume(struct xenbus_device *dev) | |
1774 | { | |
a1b4b12b | 1775 | struct blkfront_info *info = dev_get_drvdata(&dev->dev); |
9f27ee59 JF |
1776 | int err; |
1777 | ||
1778 | dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename); | |
1779 | ||
1780 | blkif_free(info, info->connected == BLKIF_STATE_CONNECTED); | |
1781 | ||
203fd61f | 1782 | err = talk_to_blkback(dev, info); |
402b27f9 RPM |
1783 | |
1784 | /* | |
1785 | * We have to wait for the backend to switch to | |
1786 | * connected state, since we want to read which | |
1787 | * features it supports. | |
1788 | */ | |
9f27ee59 JF |
1789 | |
1790 | return err; | |
1791 | } | |
1792 | ||
b70f5fa0 DS |
1793 | static void |
1794 | blkfront_closing(struct blkfront_info *info) | |
1795 | { | |
1796 | struct xenbus_device *xbdev = info->xbdev; | |
1797 | struct block_device *bdev = NULL; | |
1798 | ||
1799 | mutex_lock(&info->mutex); | |
1800 | ||
1801 | if (xbdev->state == XenbusStateClosing) { | |
1802 | mutex_unlock(&info->mutex); | |
1803 | return; | |
1804 | } | |
1805 | ||
1806 | if (info->gd) | |
1807 | bdev = bdget_disk(info->gd, 0); | |
1808 | ||
1809 | mutex_unlock(&info->mutex); | |
1810 | ||
1811 | if (!bdev) { | |
1812 | xenbus_frontend_closed(xbdev); | |
1813 | return; | |
1814 | } | |
1815 | ||
1816 | mutex_lock(&bdev->bd_mutex); | |
1817 | ||
7b32d104 | 1818 | if (bdev->bd_openers) { |
b70f5fa0 DS |
1819 | xenbus_dev_error(xbdev, -EBUSY, |
1820 | "Device in use; refusing to close"); | |
1821 | xenbus_switch_state(xbdev, XenbusStateClosing); | |
1822 | } else { | |
1823 | xlvbd_release_gendisk(info); | |
1824 | xenbus_frontend_closed(xbdev); | |
1825 | } | |
1826 | ||
1827 | mutex_unlock(&bdev->bd_mutex); | |
1828 | bdput(bdev); | |
1829 | } | |
9f27ee59 | 1830 | |
ed30bf31 LD |
1831 | static void blkfront_setup_discard(struct blkfront_info *info) |
1832 | { | |
1833 | int err; | |
ed30bf31 LD |
1834 | unsigned int discard_granularity; |
1835 | unsigned int discard_alignment; | |
5ea42986 | 1836 | unsigned int discard_secure; |
ed30bf31 | 1837 | |
1c8cad6c OH |
1838 | info->feature_discard = 1; |
1839 | err = xenbus_gather(XBT_NIL, info->xbdev->otherend, | |
1840 | "discard-granularity", "%u", &discard_granularity, | |
1841 | "discard-alignment", "%u", &discard_alignment, | |
1842 | NULL); | |
1843 | if (!err) { | |
1844 | info->discard_granularity = discard_granularity; | |
1845 | info->discard_alignment = discard_alignment; | |
1846 | } | |
1847 | err = xenbus_gather(XBT_NIL, info->xbdev->otherend, | |
1848 | "discard-secure", "%d", &discard_secure, | |
1849 | NULL); | |
1850 | if (!err) | |
1851 | info->feature_secdiscard = !!discard_secure; | |
ed30bf31 LD |
1852 | } |
1853 | ||
402b27f9 RPM |
1854 | static int blkfront_setup_indirect(struct blkfront_info *info) |
1855 | { | |
c004a6fe | 1856 | unsigned int psegs, grants; |
402b27f9 RPM |
1857 | int err, i; |
1858 | ||
d50babbe | 1859 | if (info->max_indirect_segments == 0) |
c004a6fe | 1860 | grants = BLKIF_MAX_SEGMENTS_PER_REQUEST; |
d50babbe | 1861 | else |
c004a6fe JG |
1862 | grants = info->max_indirect_segments; |
1863 | psegs = grants / GRANTS_PER_PSEG; | |
402b27f9 | 1864 | |
c004a6fe JG |
1865 | err = fill_grant_buffer(info, |
1866 | (grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info)); | |
402b27f9 RPM |
1867 | if (err) |
1868 | goto out_of_memory; | |
1869 | ||
bfe11d6d RPM |
1870 | if (!info->feature_persistent && info->max_indirect_segments) { |
1871 | /* | |
1872 | * We are using indirect descriptors but not persistent | |
1873 | * grants, we need to allocate a set of pages that can be | |
1874 | * used for mapping indirect grefs | |
1875 | */ | |
c004a6fe | 1876 | int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info); |
bfe11d6d RPM |
1877 | |
1878 | BUG_ON(!list_empty(&info->indirect_pages)); | |
1879 | for (i = 0; i < num; i++) { | |
1880 | struct page *indirect_page = alloc_page(GFP_NOIO); | |
1881 | if (!indirect_page) | |
1882 | goto out_of_memory; | |
1883 | list_add(&indirect_page->lru, &info->indirect_pages); | |
1884 | } | |
1885 | } | |
1886 | ||
86839c56 | 1887 | for (i = 0; i < BLK_RING_SIZE(info); i++) { |
402b27f9 | 1888 | info->shadow[i].grants_used = kzalloc( |
c004a6fe | 1889 | sizeof(info->shadow[i].grants_used[0]) * grants, |
402b27f9 | 1890 | GFP_NOIO); |
c004a6fe | 1891 | info->shadow[i].sg = kzalloc(sizeof(info->shadow[i].sg[0]) * psegs, GFP_NOIO); |
402b27f9 RPM |
1892 | if (info->max_indirect_segments) |
1893 | info->shadow[i].indirect_grants = kzalloc( | |
1894 | sizeof(info->shadow[i].indirect_grants[0]) * | |
c004a6fe | 1895 | INDIRECT_GREFS(grants), |
402b27f9 RPM |
1896 | GFP_NOIO); |
1897 | if ((info->shadow[i].grants_used == NULL) || | |
b7649158 | 1898 | (info->shadow[i].sg == NULL) || |
402b27f9 RPM |
1899 | (info->max_indirect_segments && |
1900 | (info->shadow[i].indirect_grants == NULL))) | |
1901 | goto out_of_memory; | |
c004a6fe | 1902 | sg_init_table(info->shadow[i].sg, psegs); |
402b27f9 RPM |
1903 | } |
1904 | ||
1905 | ||
1906 | return 0; | |
1907 | ||
1908 | out_of_memory: | |
86839c56 | 1909 | for (i = 0; i < BLK_RING_SIZE(info); i++) { |
402b27f9 RPM |
1910 | kfree(info->shadow[i].grants_used); |
1911 | info->shadow[i].grants_used = NULL; | |
b7649158 RPM |
1912 | kfree(info->shadow[i].sg); |
1913 | info->shadow[i].sg = NULL; | |
402b27f9 RPM |
1914 | kfree(info->shadow[i].indirect_grants); |
1915 | info->shadow[i].indirect_grants = NULL; | |
1916 | } | |
bfe11d6d RPM |
1917 | if (!list_empty(&info->indirect_pages)) { |
1918 | struct page *indirect_page, *n; | |
1919 | list_for_each_entry_safe(indirect_page, n, &info->indirect_pages, lru) { | |
1920 | list_del(&indirect_page->lru); | |
1921 | __free_page(indirect_page); | |
1922 | } | |
1923 | } | |
402b27f9 RPM |
1924 | return -ENOMEM; |
1925 | } | |
1926 | ||
d50babbe BL |
1927 | /* |
1928 | * Gather all backend feature-* | |
1929 | */ | |
1930 | static int blkfront_gather_backend_features(struct blkfront_info *info) | |
1931 | { | |
1932 | int err; | |
1933 | int barrier, flush, discard, persistent; | |
1934 | unsigned int indirect_segments; | |
1935 | ||
1936 | info->feature_flush = 0; | |
1937 | ||
1938 | err = xenbus_gather(XBT_NIL, info->xbdev->otherend, | |
1939 | "feature-barrier", "%d", &barrier, | |
1940 | NULL); | |
1941 | ||
1942 | /* | |
1943 | * If there's no "feature-barrier" defined, then it means | |
1944 | * we're dealing with a very old backend which writes | |
1945 | * synchronously; nothing to do. | |
1946 | * | |
1947 | * If there are barriers, then we use flush. | |
1948 | */ | |
1949 | if (!err && barrier) | |
1950 | info->feature_flush = REQ_FLUSH | REQ_FUA; | |
1951 | /* | |
1952 | * And if there is "feature-flush-cache" use that above | |
1953 | * barriers. | |
1954 | */ | |
1955 | err = xenbus_gather(XBT_NIL, info->xbdev->otherend, | |
1956 | "feature-flush-cache", "%d", &flush, | |
1957 | NULL); | |
1958 | ||
1959 | if (!err && flush) | |
1960 | info->feature_flush = REQ_FLUSH; | |
1961 | ||
1962 | err = xenbus_gather(XBT_NIL, info->xbdev->otherend, | |
1963 | "feature-discard", "%d", &discard, | |
1964 | NULL); | |
1965 | ||
1966 | if (!err && discard) | |
1967 | blkfront_setup_discard(info); | |
1968 | ||
1969 | err = xenbus_gather(XBT_NIL, info->xbdev->otherend, | |
1970 | "feature-persistent", "%u", &persistent, | |
1971 | NULL); | |
1972 | if (err) | |
1973 | info->feature_persistent = 0; | |
1974 | else | |
1975 | info->feature_persistent = persistent; | |
1976 | ||
1977 | err = xenbus_gather(XBT_NIL, info->xbdev->otherend, | |
1978 | "feature-max-indirect-segments", "%u", &indirect_segments, | |
1979 | NULL); | |
1980 | if (err) | |
1981 | info->max_indirect_segments = 0; | |
1982 | else | |
1983 | info->max_indirect_segments = min(indirect_segments, | |
1984 | xen_blkif_max_segments); | |
1985 | ||
1986 | return blkfront_setup_indirect(info); | |
1987 | } | |
1988 | ||
9f27ee59 JF |
1989 | /* |
1990 | * Invoked when the backend is finally 'ready' (and has told produced | |
1991 | * the details about the physical device - #sectors, size, etc). | |
1992 | */ | |
1993 | static void blkfront_connect(struct blkfront_info *info) | |
1994 | { | |
1995 | unsigned long long sectors; | |
1996 | unsigned long sector_size; | |
7c4d7d71 | 1997 | unsigned int physical_sector_size; |
9f27ee59 JF |
1998 | unsigned int binfo; |
1999 | int err; | |
2000 | ||
1fa73be6 S |
2001 | switch (info->connected) { |
2002 | case BLKIF_STATE_CONNECTED: | |
2003 | /* | |
2004 | * Potentially, the back-end may be signalling | |
2005 | * a capacity change; update the capacity. | |
2006 | */ | |
2007 | err = xenbus_scanf(XBT_NIL, info->xbdev->otherend, | |
2008 | "sectors", "%Lu", §ors); | |
2009 | if (XENBUS_EXIST_ERR(err)) | |
2010 | return; | |
2011 | printk(KERN_INFO "Setting capacity to %Lu\n", | |
2012 | sectors); | |
2013 | set_capacity(info->gd, sectors); | |
2def141e | 2014 | revalidate_disk(info->gd); |
1fa73be6 | 2015 | |
402b27f9 | 2016 | return; |
1fa73be6 | 2017 | case BLKIF_STATE_SUSPENDED: |
402b27f9 RPM |
2018 | /* |
2019 | * If we are recovering from suspension, we need to wait | |
2020 | * for the backend to announce it's features before | |
2021 | * reconnecting, at least we need to know if the backend | |
2022 | * supports indirect descriptors, and how many. | |
2023 | */ | |
2024 | blkif_recover(info); | |
9f27ee59 JF |
2025 | return; |
2026 | ||
b4dddb49 JF |
2027 | default: |
2028 | break; | |
1fa73be6 | 2029 | } |
9f27ee59 JF |
2030 | |
2031 | dev_dbg(&info->xbdev->dev, "%s:%s.\n", | |
2032 | __func__, info->xbdev->otherend); | |
2033 | ||
2034 | err = xenbus_gather(XBT_NIL, info->xbdev->otherend, | |
2035 | "sectors", "%llu", §ors, | |
2036 | "info", "%u", &binfo, | |
2037 | "sector-size", "%lu", §or_size, | |
2038 | NULL); | |
2039 | if (err) { | |
2040 | xenbus_dev_fatal(info->xbdev, err, | |
2041 | "reading backend fields at %s", | |
2042 | info->xbdev->otherend); | |
2043 | return; | |
2044 | } | |
2045 | ||
7c4d7d71 SB |
2046 | /* |
2047 | * physcial-sector-size is a newer field, so old backends may not | |
2048 | * provide this. Assume physical sector size to be the same as | |
2049 | * sector_size in that case. | |
2050 | */ | |
2051 | err = xenbus_scanf(XBT_NIL, info->xbdev->otherend, | |
2052 | "physical-sector-size", "%u", &physical_sector_size); | |
2053 | if (err != 1) | |
2054 | physical_sector_size = sector_size; | |
2055 | ||
d50babbe | 2056 | err = blkfront_gather_backend_features(info); |
402b27f9 RPM |
2057 | if (err) { |
2058 | xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s", | |
2059 | info->xbdev->otherend); | |
2060 | return; | |
2061 | } | |
2062 | ||
7c4d7d71 SB |
2063 | err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size, |
2064 | physical_sector_size); | |
9f27ee59 JF |
2065 | if (err) { |
2066 | xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s", | |
2067 | info->xbdev->otherend); | |
2068 | return; | |
2069 | } | |
2070 | ||
2071 | xenbus_switch_state(info->xbdev, XenbusStateConnected); | |
2072 | ||
2073 | /* Kick pending requests. */ | |
3467811e | 2074 | spin_lock_irq(&info->io_lock); |
9f27ee59 JF |
2075 | info->connected = BLKIF_STATE_CONNECTED; |
2076 | kick_pending_request_queues(info); | |
3467811e | 2077 | spin_unlock_irq(&info->io_lock); |
9f27ee59 JF |
2078 | |
2079 | add_disk(info->gd); | |
1d78d705 CL |
2080 | |
2081 | info->is_ready = 1; | |
9f27ee59 JF |
2082 | } |
2083 | ||
9f27ee59 JF |
2084 | /** |
2085 | * Callback received when the backend's state changes. | |
2086 | */ | |
203fd61f | 2087 | static void blkback_changed(struct xenbus_device *dev, |
9f27ee59 JF |
2088 | enum xenbus_state backend_state) |
2089 | { | |
a1b4b12b | 2090 | struct blkfront_info *info = dev_get_drvdata(&dev->dev); |
9f27ee59 | 2091 | |
203fd61f | 2092 | dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state); |
9f27ee59 JF |
2093 | |
2094 | switch (backend_state) { | |
9f27ee59 | 2095 | case XenbusStateInitWait: |
a9b54bb9 BL |
2096 | if (dev->state != XenbusStateInitialising) |
2097 | break; | |
8ab0144a BL |
2098 | if (talk_to_blkback(dev, info)) { |
2099 | kfree(info); | |
2100 | dev_set_drvdata(&dev->dev, NULL); | |
2101 | break; | |
2102 | } | |
2103 | case XenbusStateInitialising: | |
9f27ee59 | 2104 | case XenbusStateInitialised: |
b78c9512 NI |
2105 | case XenbusStateReconfiguring: |
2106 | case XenbusStateReconfigured: | |
9f27ee59 | 2107 | case XenbusStateUnknown: |
9f27ee59 JF |
2108 | break; |
2109 | ||
2110 | case XenbusStateConnected: | |
2111 | blkfront_connect(info); | |
2112 | break; | |
2113 | ||
36613717 DV |
2114 | case XenbusStateClosed: |
2115 | if (dev->state == XenbusStateClosed) | |
2116 | break; | |
2117 | /* Missed the backend's Closing state -- fallthrough */ | |
9f27ee59 | 2118 | case XenbusStateClosing: |
a54c8f0f CA |
2119 | if (info) |
2120 | blkfront_closing(info); | |
9f27ee59 JF |
2121 | break; |
2122 | } | |
2123 | } | |
2124 | ||
fa1bd359 | 2125 | static int blkfront_remove(struct xenbus_device *xbdev) |
9f27ee59 | 2126 | { |
fa1bd359 DS |
2127 | struct blkfront_info *info = dev_get_drvdata(&xbdev->dev); |
2128 | struct block_device *bdev = NULL; | |
2129 | struct gendisk *disk; | |
9f27ee59 | 2130 | |
fa1bd359 | 2131 | dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename); |
9f27ee59 JF |
2132 | |
2133 | blkif_free(info, 0); | |
2134 | ||
fa1bd359 DS |
2135 | mutex_lock(&info->mutex); |
2136 | ||
2137 | disk = info->gd; | |
2138 | if (disk) | |
2139 | bdev = bdget_disk(disk, 0); | |
2140 | ||
2141 | info->xbdev = NULL; | |
2142 | mutex_unlock(&info->mutex); | |
2143 | ||
2144 | if (!bdev) { | |
2145 | kfree(info); | |
2146 | return 0; | |
2147 | } | |
2148 | ||
2149 | /* | |
2150 | * The xbdev was removed before we reached the Closed | |
2151 | * state. See if it's safe to remove the disk. If the bdev | |
2152 | * isn't closed yet, we let release take care of it. | |
2153 | */ | |
2154 | ||
2155 | mutex_lock(&bdev->bd_mutex); | |
2156 | info = disk->private_data; | |
2157 | ||
d54142c7 DS |
2158 | dev_warn(disk_to_dev(disk), |
2159 | "%s was hot-unplugged, %d stale handles\n", | |
2160 | xbdev->nodename, bdev->bd_openers); | |
2161 | ||
7b32d104 | 2162 | if (info && !bdev->bd_openers) { |
fa1bd359 DS |
2163 | xlvbd_release_gendisk(info); |
2164 | disk->private_data = NULL; | |
0e345826 | 2165 | kfree(info); |
fa1bd359 DS |
2166 | } |
2167 | ||
2168 | mutex_unlock(&bdev->bd_mutex); | |
2169 | bdput(bdev); | |
9f27ee59 JF |
2170 | |
2171 | return 0; | |
2172 | } | |
2173 | ||
1d78d705 CL |
2174 | static int blkfront_is_ready(struct xenbus_device *dev) |
2175 | { | |
a1b4b12b | 2176 | struct blkfront_info *info = dev_get_drvdata(&dev->dev); |
1d78d705 | 2177 | |
5d7ed20e | 2178 | return info->is_ready && info->xbdev; |
1d78d705 CL |
2179 | } |
2180 | ||
a63c848b | 2181 | static int blkif_open(struct block_device *bdev, fmode_t mode) |
9f27ee59 | 2182 | { |
13961743 DS |
2183 | struct gendisk *disk = bdev->bd_disk; |
2184 | struct blkfront_info *info; | |
2185 | int err = 0; | |
6e9624b8 | 2186 | |
2a48fc0a | 2187 | mutex_lock(&blkfront_mutex); |
6e9624b8 | 2188 | |
13961743 DS |
2189 | info = disk->private_data; |
2190 | if (!info) { | |
2191 | /* xbdev gone */ | |
2192 | err = -ERESTARTSYS; | |
2193 | goto out; | |
2194 | } | |
2195 | ||
2196 | mutex_lock(&info->mutex); | |
2197 | ||
2198 | if (!info->gd) | |
2199 | /* xbdev is closed */ | |
2200 | err = -ERESTARTSYS; | |
2201 | ||
2202 | mutex_unlock(&info->mutex); | |
2203 | ||
13961743 | 2204 | out: |
2a48fc0a | 2205 | mutex_unlock(&blkfront_mutex); |
13961743 | 2206 | return err; |
9f27ee59 JF |
2207 | } |
2208 | ||
db2a144b | 2209 | static void blkif_release(struct gendisk *disk, fmode_t mode) |
9f27ee59 | 2210 | { |
a63c848b | 2211 | struct blkfront_info *info = disk->private_data; |
7fd152f4 DS |
2212 | struct block_device *bdev; |
2213 | struct xenbus_device *xbdev; | |
2214 | ||
2a48fc0a | 2215 | mutex_lock(&blkfront_mutex); |
7fd152f4 DS |
2216 | |
2217 | bdev = bdget_disk(disk, 0); | |
7fd152f4 | 2218 | |
2f089cb8 FP |
2219 | if (!bdev) { |
2220 | WARN(1, "Block device %s yanked out from us!\n", disk->disk_name); | |
2221 | goto out_mutex; | |
2222 | } | |
acfca3c6 DS |
2223 | if (bdev->bd_openers) |
2224 | goto out; | |
2225 | ||
7fd152f4 DS |
2226 | /* |
2227 | * Check if we have been instructed to close. We will have | |
2228 | * deferred this request, because the bdev was still open. | |
2229 | */ | |
2230 | ||
2231 | mutex_lock(&info->mutex); | |
2232 | xbdev = info->xbdev; | |
2233 | ||
2234 | if (xbdev && xbdev->state == XenbusStateClosing) { | |
2235 | /* pending switch to state closed */ | |
d54142c7 | 2236 | dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n"); |
7fd152f4 DS |
2237 | xlvbd_release_gendisk(info); |
2238 | xenbus_frontend_closed(info->xbdev); | |
2239 | } | |
2240 | ||
2241 | mutex_unlock(&info->mutex); | |
2242 | ||
2243 | if (!xbdev) { | |
2244 | /* sudden device removal */ | |
d54142c7 | 2245 | dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n"); |
7fd152f4 DS |
2246 | xlvbd_release_gendisk(info); |
2247 | disk->private_data = NULL; | |
2248 | kfree(info); | |
9f27ee59 | 2249 | } |
7fd152f4 | 2250 | |
a4cc14ec | 2251 | out: |
dad5cf65 | 2252 | bdput(bdev); |
2f089cb8 | 2253 | out_mutex: |
2a48fc0a | 2254 | mutex_unlock(&blkfront_mutex); |
9f27ee59 JF |
2255 | } |
2256 | ||
83d5cde4 | 2257 | static const struct block_device_operations xlvbd_block_fops = |
9f27ee59 JF |
2258 | { |
2259 | .owner = THIS_MODULE, | |
a63c848b AV |
2260 | .open = blkif_open, |
2261 | .release = blkif_release, | |
597592d9 | 2262 | .getgeo = blkif_getgeo, |
8a6cfeb6 | 2263 | .ioctl = blkif_ioctl, |
9f27ee59 JF |
2264 | }; |
2265 | ||
2266 | ||
ec9c42ec | 2267 | static const struct xenbus_device_id blkfront_ids[] = { |
9f27ee59 JF |
2268 | { "vbd" }, |
2269 | { "" } | |
2270 | }; | |
2271 | ||
95afae48 DV |
2272 | static struct xenbus_driver blkfront_driver = { |
2273 | .ids = blkfront_ids, | |
9f27ee59 JF |
2274 | .probe = blkfront_probe, |
2275 | .remove = blkfront_remove, | |
2276 | .resume = blkfront_resume, | |
203fd61f | 2277 | .otherend_changed = blkback_changed, |
1d78d705 | 2278 | .is_ready = blkfront_is_ready, |
95afae48 | 2279 | }; |
9f27ee59 JF |
2280 | |
2281 | static int __init xlblk_init(void) | |
2282 | { | |
469738e6 LE |
2283 | int ret; |
2284 | ||
6e833587 | 2285 | if (!xen_domain()) |
9f27ee59 JF |
2286 | return -ENODEV; |
2287 | ||
9cce2914 | 2288 | if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) { |
86839c56 | 2289 | pr_info("Invalid max_ring_order (%d), will use default max: %d.\n", |
9cce2914 | 2290 | xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER); |
86839c56 BL |
2291 | xen_blkif_max_ring_order = 0; |
2292 | } | |
2293 | ||
51c71a3b | 2294 | if (!xen_has_pv_disk_devices()) |
b9136d20 IM |
2295 | return -ENODEV; |
2296 | ||
9f27ee59 JF |
2297 | if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) { |
2298 | printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n", | |
2299 | XENVBD_MAJOR, DEV_NAME); | |
2300 | return -ENODEV; | |
2301 | } | |
2302 | ||
73db144b | 2303 | ret = xenbus_register_frontend(&blkfront_driver); |
469738e6 LE |
2304 | if (ret) { |
2305 | unregister_blkdev(XENVBD_MAJOR, DEV_NAME); | |
2306 | return ret; | |
2307 | } | |
2308 | ||
2309 | return 0; | |
9f27ee59 JF |
2310 | } |
2311 | module_init(xlblk_init); | |
2312 | ||
2313 | ||
5a60d0cd | 2314 | static void __exit xlblk_exit(void) |
9f27ee59 | 2315 | { |
8605067f JB |
2316 | xenbus_unregister_driver(&blkfront_driver); |
2317 | unregister_blkdev(XENVBD_MAJOR, DEV_NAME); | |
2318 | kfree(minors); | |
9f27ee59 JF |
2319 | } |
2320 | module_exit(xlblk_exit); | |
2321 | ||
2322 | MODULE_DESCRIPTION("Xen virtual block device frontend"); | |
2323 | MODULE_LICENSE("GPL"); | |
2324 | MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR); | |
d2f0c52b | 2325 | MODULE_ALIAS("xen:vbd"); |
4f93f09b | 2326 | MODULE_ALIAS("xenblk"); |