]>
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> | |
597592d9 | 40 | #include <linux/hdreg.h> |
9f27ee59 JF |
41 | #include <linux/module.h> |
42 | ||
43 | #include <xen/xenbus.h> | |
44 | #include <xen/grant_table.h> | |
45 | #include <xen/events.h> | |
46 | #include <xen/page.h> | |
47 | ||
48 | #include <xen/interface/grant_table.h> | |
49 | #include <xen/interface/io/blkif.h> | |
3e334239 | 50 | #include <xen/interface/io/protocols.h> |
9f27ee59 JF |
51 | |
52 | #include <asm/xen/hypervisor.h> | |
53 | ||
54 | enum blkif_state { | |
55 | BLKIF_STATE_DISCONNECTED, | |
56 | BLKIF_STATE_CONNECTED, | |
57 | BLKIF_STATE_SUSPENDED, | |
58 | }; | |
59 | ||
60 | struct blk_shadow { | |
61 | struct blkif_request req; | |
62 | unsigned long request; | |
63 | unsigned long frame[BLKIF_MAX_SEGMENTS_PER_REQUEST]; | |
64 | }; | |
65 | ||
66 | static struct block_device_operations xlvbd_block_fops; | |
67 | ||
68 | #define BLK_RING_SIZE __RING_SIZE((struct blkif_sring *)0, PAGE_SIZE) | |
69 | ||
70 | /* | |
71 | * We have one of these per vbd, whether ide, scsi or 'other'. They | |
72 | * hang in private_data off the gendisk structure. We may end up | |
73 | * putting all kinds of interesting stuff here :-) | |
74 | */ | |
75 | struct blkfront_info | |
76 | { | |
77 | struct xenbus_device *xbdev; | |
9f27ee59 JF |
78 | struct gendisk *gd; |
79 | int vdevice; | |
80 | blkif_vdev_t handle; | |
81 | enum blkif_state connected; | |
82 | int ring_ref; | |
83 | struct blkif_front_ring ring; | |
84 | unsigned int evtchn, irq; | |
85 | struct request_queue *rq; | |
86 | struct work_struct work; | |
87 | struct gnttab_free_callback callback; | |
88 | struct blk_shadow shadow[BLK_RING_SIZE]; | |
89 | unsigned long shadow_free; | |
90 | int feature_barrier; | |
91 | ||
92 | /** | |
93 | * The number of people holding this device open. We won't allow a | |
94 | * hot-unplug unless this is 0. | |
95 | */ | |
96 | int users; | |
97 | }; | |
98 | ||
99 | static DEFINE_SPINLOCK(blkif_io_lock); | |
100 | ||
101 | #define MAXIMUM_OUTSTANDING_BLOCK_REQS \ | |
102 | (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE) | |
103 | #define GRANT_INVALID_REF 0 | |
104 | ||
105 | #define PARTS_PER_DISK 16 | |
106 | ||
107 | #define BLKIF_MAJOR(dev) ((dev)>>8) | |
108 | #define BLKIF_MINOR(dev) ((dev) & 0xff) | |
109 | ||
110 | #define DEV_NAME "xvd" /* name in /dev */ | |
111 | ||
112 | /* Information about our VBDs. */ | |
113 | #define MAX_VBDS 64 | |
114 | static LIST_HEAD(vbds_list); | |
115 | ||
116 | static int get_id_from_freelist(struct blkfront_info *info) | |
117 | { | |
118 | unsigned long free = info->shadow_free; | |
119 | BUG_ON(free > BLK_RING_SIZE); | |
120 | info->shadow_free = info->shadow[free].req.id; | |
121 | info->shadow[free].req.id = 0x0fffffee; /* debug */ | |
122 | return free; | |
123 | } | |
124 | ||
125 | static void add_id_to_freelist(struct blkfront_info *info, | |
126 | unsigned long id) | |
127 | { | |
128 | info->shadow[id].req.id = info->shadow_free; | |
129 | info->shadow[id].request = 0; | |
130 | info->shadow_free = id; | |
131 | } | |
132 | ||
133 | static void blkif_restart_queue_callback(void *arg) | |
134 | { | |
135 | struct blkfront_info *info = (struct blkfront_info *)arg; | |
136 | schedule_work(&info->work); | |
137 | } | |
138 | ||
597592d9 IC |
139 | int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg) |
140 | { | |
141 | /* We don't have real geometry info, but let's at least return | |
142 | values consistent with the size of the device */ | |
143 | sector_t nsect = get_capacity(bd->bd_disk); | |
144 | sector_t cylinders = nsect; | |
145 | ||
146 | hg->heads = 0xff; | |
147 | hg->sectors = 0x3f; | |
148 | sector_div(cylinders, hg->heads * hg->sectors); | |
149 | hg->cylinders = cylinders; | |
150 | if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect) | |
151 | hg->cylinders = 0xffff; | |
152 | return 0; | |
153 | } | |
154 | ||
9f27ee59 JF |
155 | /* |
156 | * blkif_queue_request | |
157 | * | |
158 | * request block io | |
159 | * | |
160 | * id: for guest use only. | |
161 | * operation: BLKIF_OP_{READ,WRITE,PROBE} | |
162 | * buffer: buffer to read/write into. this should be a | |
163 | * virtual address in the guest os. | |
164 | */ | |
165 | static int blkif_queue_request(struct request *req) | |
166 | { | |
167 | struct blkfront_info *info = req->rq_disk->private_data; | |
168 | unsigned long buffer_mfn; | |
169 | struct blkif_request *ring_req; | |
5705f702 | 170 | struct req_iterator iter; |
9f27ee59 | 171 | struct bio_vec *bvec; |
9f27ee59 JF |
172 | unsigned long id; |
173 | unsigned int fsect, lsect; | |
174 | int ref; | |
175 | grant_ref_t gref_head; | |
176 | ||
177 | if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) | |
178 | return 1; | |
179 | ||
180 | if (gnttab_alloc_grant_references( | |
181 | BLKIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) { | |
182 | gnttab_request_free_callback( | |
183 | &info->callback, | |
184 | blkif_restart_queue_callback, | |
185 | info, | |
186 | BLKIF_MAX_SEGMENTS_PER_REQUEST); | |
187 | return 1; | |
188 | } | |
189 | ||
190 | /* Fill out a communications ring structure. */ | |
191 | ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt); | |
192 | id = get_id_from_freelist(info); | |
193 | info->shadow[id].request = (unsigned long)req; | |
194 | ||
195 | ring_req->id = id; | |
196 | ring_req->sector_number = (blkif_sector_t)req->sector; | |
197 | ring_req->handle = info->handle; | |
198 | ||
199 | ring_req->operation = rq_data_dir(req) ? | |
200 | BLKIF_OP_WRITE : BLKIF_OP_READ; | |
201 | if (blk_barrier_rq(req)) | |
202 | ring_req->operation = BLKIF_OP_WRITE_BARRIER; | |
203 | ||
204 | ring_req->nr_segments = 0; | |
5705f702 | 205 | rq_for_each_segment(bvec, req, iter) { |
6c92e699 JA |
206 | BUG_ON(ring_req->nr_segments == BLKIF_MAX_SEGMENTS_PER_REQUEST); |
207 | buffer_mfn = pfn_to_mfn(page_to_pfn(bvec->bv_page)); | |
208 | fsect = bvec->bv_offset >> 9; | |
209 | lsect = fsect + (bvec->bv_len >> 9) - 1; | |
210 | /* install a grant reference. */ | |
211 | ref = gnttab_claim_grant_reference(&gref_head); | |
212 | BUG_ON(ref == -ENOSPC); | |
213 | ||
214 | gnttab_grant_foreign_access_ref( | |
9f27ee59 JF |
215 | ref, |
216 | info->xbdev->otherend_id, | |
217 | buffer_mfn, | |
218 | rq_data_dir(req) ); | |
219 | ||
6c92e699 | 220 | info->shadow[id].frame[ring_req->nr_segments] = |
9f27ee59 JF |
221 | mfn_to_pfn(buffer_mfn); |
222 | ||
6c92e699 | 223 | ring_req->seg[ring_req->nr_segments] = |
9f27ee59 JF |
224 | (struct blkif_request_segment) { |
225 | .gref = ref, | |
226 | .first_sect = fsect, | |
227 | .last_sect = lsect }; | |
228 | ||
6c92e699 | 229 | ring_req->nr_segments++; |
9f27ee59 JF |
230 | } |
231 | ||
232 | info->ring.req_prod_pvt++; | |
233 | ||
234 | /* Keep a private copy so we can reissue requests when recovering. */ | |
235 | info->shadow[id].req = *ring_req; | |
236 | ||
237 | gnttab_free_grant_references(gref_head); | |
238 | ||
239 | return 0; | |
240 | } | |
241 | ||
242 | ||
243 | static inline void flush_requests(struct blkfront_info *info) | |
244 | { | |
245 | int notify; | |
246 | ||
247 | RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify); | |
248 | ||
249 | if (notify) | |
250 | notify_remote_via_irq(info->irq); | |
251 | } | |
252 | ||
253 | /* | |
254 | * do_blkif_request | |
255 | * read a block; request is in a request queue | |
256 | */ | |
165125e1 | 257 | static void do_blkif_request(struct request_queue *rq) |
9f27ee59 JF |
258 | { |
259 | struct blkfront_info *info = NULL; | |
260 | struct request *req; | |
261 | int queued; | |
262 | ||
263 | pr_debug("Entered do_blkif_request\n"); | |
264 | ||
265 | queued = 0; | |
266 | ||
267 | while ((req = elv_next_request(rq)) != NULL) { | |
268 | info = req->rq_disk->private_data; | |
269 | if (!blk_fs_request(req)) { | |
270 | end_request(req, 0); | |
271 | continue; | |
272 | } | |
273 | ||
274 | if (RING_FULL(&info->ring)) | |
275 | goto wait; | |
276 | ||
277 | pr_debug("do_blk_req %p: cmd %p, sec %lx, " | |
278 | "(%u/%li) buffer:%p [%s]\n", | |
279 | req, req->cmd, (unsigned long)req->sector, | |
280 | req->current_nr_sectors, | |
281 | req->nr_sectors, req->buffer, | |
282 | rq_data_dir(req) ? "write" : "read"); | |
283 | ||
284 | ||
285 | blkdev_dequeue_request(req); | |
286 | if (blkif_queue_request(req)) { | |
287 | blk_requeue_request(rq, req); | |
288 | wait: | |
289 | /* Avoid pointless unplugs. */ | |
290 | blk_stop_queue(rq); | |
291 | break; | |
292 | } | |
293 | ||
294 | queued++; | |
295 | } | |
296 | ||
297 | if (queued != 0) | |
298 | flush_requests(info); | |
299 | } | |
300 | ||
301 | static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size) | |
302 | { | |
165125e1 | 303 | struct request_queue *rq; |
9f27ee59 JF |
304 | |
305 | rq = blk_init_queue(do_blkif_request, &blkif_io_lock); | |
306 | if (rq == NULL) | |
307 | return -1; | |
308 | ||
309 | elevator_init(rq, "noop"); | |
310 | ||
311 | /* Hard sector size and max sectors impersonate the equiv. hardware. */ | |
312 | blk_queue_hardsect_size(rq, sector_size); | |
313 | blk_queue_max_sectors(rq, 512); | |
314 | ||
315 | /* Each segment in a request is up to an aligned page in size. */ | |
316 | blk_queue_segment_boundary(rq, PAGE_SIZE - 1); | |
317 | blk_queue_max_segment_size(rq, PAGE_SIZE); | |
318 | ||
319 | /* Ensure a merged request will fit in a single I/O ring slot. */ | |
320 | blk_queue_max_phys_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST); | |
321 | blk_queue_max_hw_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST); | |
322 | ||
323 | /* Make sure buffer addresses are sector-aligned. */ | |
324 | blk_queue_dma_alignment(rq, 511); | |
325 | ||
326 | gd->queue = rq; | |
327 | ||
328 | return 0; | |
329 | } | |
330 | ||
331 | ||
332 | static int xlvbd_barrier(struct blkfront_info *info) | |
333 | { | |
334 | int err; | |
335 | ||
336 | err = blk_queue_ordered(info->rq, | |
337 | info->feature_barrier ? QUEUE_ORDERED_DRAIN : QUEUE_ORDERED_NONE, | |
338 | NULL); | |
339 | ||
340 | if (err) | |
341 | return err; | |
342 | ||
343 | printk(KERN_INFO "blkfront: %s: barriers %s\n", | |
344 | info->gd->disk_name, | |
345 | info->feature_barrier ? "enabled" : "disabled"); | |
346 | return 0; | |
347 | } | |
348 | ||
349 | ||
350 | static int xlvbd_alloc_gendisk(int minor, blkif_sector_t capacity, | |
351 | int vdevice, u16 vdisk_info, u16 sector_size, | |
352 | struct blkfront_info *info) | |
353 | { | |
354 | struct gendisk *gd; | |
355 | int nr_minors = 1; | |
356 | int err = -ENODEV; | |
357 | ||
358 | BUG_ON(info->gd != NULL); | |
359 | BUG_ON(info->rq != NULL); | |
360 | ||
361 | if ((minor % PARTS_PER_DISK) == 0) | |
362 | nr_minors = PARTS_PER_DISK; | |
363 | ||
364 | gd = alloc_disk(nr_minors); | |
365 | if (gd == NULL) | |
366 | goto out; | |
367 | ||
368 | if (nr_minors > 1) | |
369 | sprintf(gd->disk_name, "%s%c", DEV_NAME, | |
370 | 'a' + minor / PARTS_PER_DISK); | |
371 | else | |
372 | sprintf(gd->disk_name, "%s%c%d", DEV_NAME, | |
373 | 'a' + minor / PARTS_PER_DISK, | |
374 | minor % PARTS_PER_DISK); | |
375 | ||
376 | gd->major = XENVBD_MAJOR; | |
377 | gd->first_minor = minor; | |
378 | gd->fops = &xlvbd_block_fops; | |
379 | gd->private_data = info; | |
380 | gd->driverfs_dev = &(info->xbdev->dev); | |
381 | set_capacity(gd, capacity); | |
382 | ||
383 | if (xlvbd_init_blk_queue(gd, sector_size)) { | |
384 | del_gendisk(gd); | |
385 | goto out; | |
386 | } | |
387 | ||
388 | info->rq = gd->queue; | |
389 | info->gd = gd; | |
390 | ||
391 | if (info->feature_barrier) | |
392 | xlvbd_barrier(info); | |
393 | ||
394 | if (vdisk_info & VDISK_READONLY) | |
395 | set_disk_ro(gd, 1); | |
396 | ||
397 | if (vdisk_info & VDISK_REMOVABLE) | |
398 | gd->flags |= GENHD_FL_REMOVABLE; | |
399 | ||
400 | if (vdisk_info & VDISK_CDROM) | |
401 | gd->flags |= GENHD_FL_CD; | |
402 | ||
403 | return 0; | |
404 | ||
405 | out: | |
406 | return err; | |
407 | } | |
408 | ||
409 | static void kick_pending_request_queues(struct blkfront_info *info) | |
410 | { | |
411 | if (!RING_FULL(&info->ring)) { | |
412 | /* Re-enable calldowns. */ | |
413 | blk_start_queue(info->rq); | |
414 | /* Kick things off immediately. */ | |
415 | do_blkif_request(info->rq); | |
416 | } | |
417 | } | |
418 | ||
419 | static void blkif_restart_queue(struct work_struct *work) | |
420 | { | |
421 | struct blkfront_info *info = container_of(work, struct blkfront_info, work); | |
422 | ||
423 | spin_lock_irq(&blkif_io_lock); | |
424 | if (info->connected == BLKIF_STATE_CONNECTED) | |
425 | kick_pending_request_queues(info); | |
426 | spin_unlock_irq(&blkif_io_lock); | |
427 | } | |
428 | ||
429 | static void blkif_free(struct blkfront_info *info, int suspend) | |
430 | { | |
431 | /* Prevent new requests being issued until we fix things up. */ | |
432 | spin_lock_irq(&blkif_io_lock); | |
433 | info->connected = suspend ? | |
434 | BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED; | |
435 | /* No more blkif_request(). */ | |
436 | if (info->rq) | |
437 | blk_stop_queue(info->rq); | |
438 | /* No more gnttab callback work. */ | |
439 | gnttab_cancel_free_callback(&info->callback); | |
440 | spin_unlock_irq(&blkif_io_lock); | |
441 | ||
442 | /* Flush gnttab callback work. Must be done with no locks held. */ | |
443 | flush_scheduled_work(); | |
444 | ||
445 | /* Free resources associated with old device channel. */ | |
446 | if (info->ring_ref != GRANT_INVALID_REF) { | |
447 | gnttab_end_foreign_access(info->ring_ref, 0, | |
448 | (unsigned long)info->ring.sring); | |
449 | info->ring_ref = GRANT_INVALID_REF; | |
450 | info->ring.sring = NULL; | |
451 | } | |
452 | if (info->irq) | |
453 | unbind_from_irqhandler(info->irq, info); | |
454 | info->evtchn = info->irq = 0; | |
455 | ||
456 | } | |
457 | ||
458 | static void blkif_completion(struct blk_shadow *s) | |
459 | { | |
460 | int i; | |
461 | for (i = 0; i < s->req.nr_segments; i++) | |
462 | gnttab_end_foreign_access(s->req.seg[i].gref, 0, 0UL); | |
463 | } | |
464 | ||
465 | static irqreturn_t blkif_interrupt(int irq, void *dev_id) | |
466 | { | |
467 | struct request *req; | |
468 | struct blkif_response *bret; | |
469 | RING_IDX i, rp; | |
470 | unsigned long flags; | |
471 | struct blkfront_info *info = (struct blkfront_info *)dev_id; | |
f530f036 | 472 | int error; |
9f27ee59 JF |
473 | |
474 | spin_lock_irqsave(&blkif_io_lock, flags); | |
475 | ||
476 | if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) { | |
477 | spin_unlock_irqrestore(&blkif_io_lock, flags); | |
478 | return IRQ_HANDLED; | |
479 | } | |
480 | ||
481 | again: | |
482 | rp = info->ring.sring->rsp_prod; | |
483 | rmb(); /* Ensure we see queued responses up to 'rp'. */ | |
484 | ||
485 | for (i = info->ring.rsp_cons; i != rp; i++) { | |
486 | unsigned long id; | |
487 | int ret; | |
488 | ||
489 | bret = RING_GET_RESPONSE(&info->ring, i); | |
490 | id = bret->id; | |
491 | req = (struct request *)info->shadow[id].request; | |
492 | ||
493 | blkif_completion(&info->shadow[id]); | |
494 | ||
495 | add_id_to_freelist(info, id); | |
496 | ||
f530f036 | 497 | error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO; |
9f27ee59 JF |
498 | switch (bret->operation) { |
499 | case BLKIF_OP_WRITE_BARRIER: | |
500 | if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) { | |
501 | printk(KERN_WARNING "blkfront: %s: write barrier op failed\n", | |
502 | info->gd->disk_name); | |
f530f036 | 503 | error = -EOPNOTSUPP; |
9f27ee59 JF |
504 | info->feature_barrier = 0; |
505 | xlvbd_barrier(info); | |
506 | } | |
507 | /* fall through */ | |
508 | case BLKIF_OP_READ: | |
509 | case BLKIF_OP_WRITE: | |
510 | if (unlikely(bret->status != BLKIF_RSP_OKAY)) | |
511 | dev_dbg(&info->xbdev->dev, "Bad return from blkdev data " | |
512 | "request: %x\n", bret->status); | |
513 | ||
f530f036 | 514 | ret = __blk_end_request(req, error, blk_rq_bytes(req)); |
9f27ee59 | 515 | BUG_ON(ret); |
9f27ee59 JF |
516 | break; |
517 | default: | |
518 | BUG(); | |
519 | } | |
520 | } | |
521 | ||
522 | info->ring.rsp_cons = i; | |
523 | ||
524 | if (i != info->ring.req_prod_pvt) { | |
525 | int more_to_do; | |
526 | RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do); | |
527 | if (more_to_do) | |
528 | goto again; | |
529 | } else | |
530 | info->ring.sring->rsp_event = i + 1; | |
531 | ||
532 | kick_pending_request_queues(info); | |
533 | ||
534 | spin_unlock_irqrestore(&blkif_io_lock, flags); | |
535 | ||
536 | return IRQ_HANDLED; | |
537 | } | |
538 | ||
539 | ||
540 | static int setup_blkring(struct xenbus_device *dev, | |
541 | struct blkfront_info *info) | |
542 | { | |
543 | struct blkif_sring *sring; | |
544 | int err; | |
545 | ||
546 | info->ring_ref = GRANT_INVALID_REF; | |
547 | ||
548 | sring = (struct blkif_sring *)__get_free_page(GFP_KERNEL); | |
549 | if (!sring) { | |
550 | xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring"); | |
551 | return -ENOMEM; | |
552 | } | |
553 | SHARED_RING_INIT(sring); | |
554 | FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE); | |
555 | ||
556 | err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring)); | |
557 | if (err < 0) { | |
558 | free_page((unsigned long)sring); | |
559 | info->ring.sring = NULL; | |
560 | goto fail; | |
561 | } | |
562 | info->ring_ref = err; | |
563 | ||
564 | err = xenbus_alloc_evtchn(dev, &info->evtchn); | |
565 | if (err) | |
566 | goto fail; | |
567 | ||
568 | err = bind_evtchn_to_irqhandler(info->evtchn, | |
569 | blkif_interrupt, | |
570 | IRQF_SAMPLE_RANDOM, "blkif", info); | |
571 | if (err <= 0) { | |
572 | xenbus_dev_fatal(dev, err, | |
573 | "bind_evtchn_to_irqhandler failed"); | |
574 | goto fail; | |
575 | } | |
576 | info->irq = err; | |
577 | ||
578 | return 0; | |
579 | fail: | |
580 | blkif_free(info, 0); | |
581 | return err; | |
582 | } | |
583 | ||
584 | ||
585 | /* Common code used when first setting up, and when resuming. */ | |
586 | static int talk_to_backend(struct xenbus_device *dev, | |
587 | struct blkfront_info *info) | |
588 | { | |
589 | const char *message = NULL; | |
590 | struct xenbus_transaction xbt; | |
591 | int err; | |
592 | ||
593 | /* Create shared ring, alloc event channel. */ | |
594 | err = setup_blkring(dev, info); | |
595 | if (err) | |
596 | goto out; | |
597 | ||
598 | again: | |
599 | err = xenbus_transaction_start(&xbt); | |
600 | if (err) { | |
601 | xenbus_dev_fatal(dev, err, "starting transaction"); | |
602 | goto destroy_blkring; | |
603 | } | |
604 | ||
605 | err = xenbus_printf(xbt, dev->nodename, | |
606 | "ring-ref", "%u", info->ring_ref); | |
607 | if (err) { | |
608 | message = "writing ring-ref"; | |
609 | goto abort_transaction; | |
610 | } | |
611 | err = xenbus_printf(xbt, dev->nodename, | |
612 | "event-channel", "%u", info->evtchn); | |
613 | if (err) { | |
614 | message = "writing event-channel"; | |
615 | goto abort_transaction; | |
616 | } | |
3e334239 MA |
617 | err = xenbus_printf(xbt, dev->nodename, "protocol", "%s", |
618 | XEN_IO_PROTO_ABI_NATIVE); | |
619 | if (err) { | |
620 | message = "writing protocol"; | |
621 | goto abort_transaction; | |
622 | } | |
9f27ee59 JF |
623 | |
624 | err = xenbus_transaction_end(xbt, 0); | |
625 | if (err) { | |
626 | if (err == -EAGAIN) | |
627 | goto again; | |
628 | xenbus_dev_fatal(dev, err, "completing transaction"); | |
629 | goto destroy_blkring; | |
630 | } | |
631 | ||
632 | xenbus_switch_state(dev, XenbusStateInitialised); | |
633 | ||
634 | return 0; | |
635 | ||
636 | abort_transaction: | |
637 | xenbus_transaction_end(xbt, 1); | |
638 | if (message) | |
639 | xenbus_dev_fatal(dev, err, "%s", message); | |
640 | destroy_blkring: | |
641 | blkif_free(info, 0); | |
642 | out: | |
643 | return err; | |
644 | } | |
645 | ||
646 | ||
647 | /** | |
648 | * Entry point to this code when a new device is created. Allocate the basic | |
649 | * structures and the ring buffer for communication with the backend, and | |
650 | * inform the backend of the appropriate details for those. Switch to | |
651 | * Initialised state. | |
652 | */ | |
653 | static int blkfront_probe(struct xenbus_device *dev, | |
654 | const struct xenbus_device_id *id) | |
655 | { | |
656 | int err, vdevice, i; | |
657 | struct blkfront_info *info; | |
658 | ||
659 | /* FIXME: Use dynamic device id if this is not set. */ | |
660 | err = xenbus_scanf(XBT_NIL, dev->nodename, | |
661 | "virtual-device", "%i", &vdevice); | |
662 | if (err != 1) { | |
663 | xenbus_dev_fatal(dev, err, "reading virtual-device"); | |
664 | return err; | |
665 | } | |
666 | ||
667 | info = kzalloc(sizeof(*info), GFP_KERNEL); | |
668 | if (!info) { | |
669 | xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure"); | |
670 | return -ENOMEM; | |
671 | } | |
672 | ||
673 | info->xbdev = dev; | |
674 | info->vdevice = vdevice; | |
675 | info->connected = BLKIF_STATE_DISCONNECTED; | |
676 | INIT_WORK(&info->work, blkif_restart_queue); | |
677 | ||
678 | for (i = 0; i < BLK_RING_SIZE; i++) | |
679 | info->shadow[i].req.id = i+1; | |
680 | info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff; | |
681 | ||
682 | /* Front end dir is a number, which is used as the id. */ | |
683 | info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0); | |
684 | dev->dev.driver_data = info; | |
685 | ||
686 | err = talk_to_backend(dev, info); | |
687 | if (err) { | |
688 | kfree(info); | |
689 | dev->dev.driver_data = NULL; | |
690 | return err; | |
691 | } | |
692 | ||
693 | return 0; | |
694 | } | |
695 | ||
696 | ||
697 | static int blkif_recover(struct blkfront_info *info) | |
698 | { | |
699 | int i; | |
700 | struct blkif_request *req; | |
701 | struct blk_shadow *copy; | |
702 | int j; | |
703 | ||
704 | /* Stage 1: Make a safe copy of the shadow state. */ | |
705 | copy = kmalloc(sizeof(info->shadow), GFP_KERNEL); | |
706 | if (!copy) | |
707 | return -ENOMEM; | |
708 | memcpy(copy, info->shadow, sizeof(info->shadow)); | |
709 | ||
710 | /* Stage 2: Set up free list. */ | |
711 | memset(&info->shadow, 0, sizeof(info->shadow)); | |
712 | for (i = 0; i < BLK_RING_SIZE; i++) | |
713 | info->shadow[i].req.id = i+1; | |
714 | info->shadow_free = info->ring.req_prod_pvt; | |
715 | info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff; | |
716 | ||
717 | /* Stage 3: Find pending requests and requeue them. */ | |
718 | for (i = 0; i < BLK_RING_SIZE; i++) { | |
719 | /* Not in use? */ | |
720 | if (copy[i].request == 0) | |
721 | continue; | |
722 | ||
723 | /* Grab a request slot and copy shadow state into it. */ | |
724 | req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt); | |
725 | *req = copy[i].req; | |
726 | ||
727 | /* We get a new request id, and must reset the shadow state. */ | |
728 | req->id = get_id_from_freelist(info); | |
729 | memcpy(&info->shadow[req->id], ©[i], sizeof(copy[i])); | |
730 | ||
731 | /* Rewrite any grant references invalidated by susp/resume. */ | |
732 | for (j = 0; j < req->nr_segments; j++) | |
733 | gnttab_grant_foreign_access_ref( | |
734 | req->seg[j].gref, | |
735 | info->xbdev->otherend_id, | |
736 | pfn_to_mfn(info->shadow[req->id].frame[j]), | |
737 | rq_data_dir( | |
738 | (struct request *) | |
739 | info->shadow[req->id].request)); | |
740 | info->shadow[req->id].req = *req; | |
741 | ||
742 | info->ring.req_prod_pvt++; | |
743 | } | |
744 | ||
745 | kfree(copy); | |
746 | ||
747 | xenbus_switch_state(info->xbdev, XenbusStateConnected); | |
748 | ||
749 | spin_lock_irq(&blkif_io_lock); | |
750 | ||
751 | /* Now safe for us to use the shared ring */ | |
752 | info->connected = BLKIF_STATE_CONNECTED; | |
753 | ||
754 | /* Send off requeued requests */ | |
755 | flush_requests(info); | |
756 | ||
757 | /* Kick any other new requests queued since we resumed */ | |
758 | kick_pending_request_queues(info); | |
759 | ||
760 | spin_unlock_irq(&blkif_io_lock); | |
761 | ||
762 | return 0; | |
763 | } | |
764 | ||
765 | /** | |
766 | * We are reconnecting to the backend, due to a suspend/resume, or a backend | |
767 | * driver restart. We tear down our blkif structure and recreate it, but | |
768 | * leave the device-layer structures intact so that this is transparent to the | |
769 | * rest of the kernel. | |
770 | */ | |
771 | static int blkfront_resume(struct xenbus_device *dev) | |
772 | { | |
773 | struct blkfront_info *info = dev->dev.driver_data; | |
774 | int err; | |
775 | ||
776 | dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename); | |
777 | ||
778 | blkif_free(info, info->connected == BLKIF_STATE_CONNECTED); | |
779 | ||
780 | err = talk_to_backend(dev, info); | |
781 | if (info->connected == BLKIF_STATE_SUSPENDED && !err) | |
782 | err = blkif_recover(info); | |
783 | ||
784 | return err; | |
785 | } | |
786 | ||
787 | ||
788 | /* | |
789 | * Invoked when the backend is finally 'ready' (and has told produced | |
790 | * the details about the physical device - #sectors, size, etc). | |
791 | */ | |
792 | static void blkfront_connect(struct blkfront_info *info) | |
793 | { | |
794 | unsigned long long sectors; | |
795 | unsigned long sector_size; | |
796 | unsigned int binfo; | |
797 | int err; | |
798 | ||
799 | if ((info->connected == BLKIF_STATE_CONNECTED) || | |
800 | (info->connected == BLKIF_STATE_SUSPENDED) ) | |
801 | return; | |
802 | ||
803 | dev_dbg(&info->xbdev->dev, "%s:%s.\n", | |
804 | __func__, info->xbdev->otherend); | |
805 | ||
806 | err = xenbus_gather(XBT_NIL, info->xbdev->otherend, | |
807 | "sectors", "%llu", §ors, | |
808 | "info", "%u", &binfo, | |
809 | "sector-size", "%lu", §or_size, | |
810 | NULL); | |
811 | if (err) { | |
812 | xenbus_dev_fatal(info->xbdev, err, | |
813 | "reading backend fields at %s", | |
814 | info->xbdev->otherend); | |
815 | return; | |
816 | } | |
817 | ||
818 | err = xenbus_gather(XBT_NIL, info->xbdev->otherend, | |
819 | "feature-barrier", "%lu", &info->feature_barrier, | |
820 | NULL); | |
821 | if (err) | |
822 | info->feature_barrier = 0; | |
823 | ||
824 | err = xlvbd_alloc_gendisk(BLKIF_MINOR(info->vdevice), | |
825 | sectors, info->vdevice, | |
826 | binfo, sector_size, info); | |
827 | if (err) { | |
828 | xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s", | |
829 | info->xbdev->otherend); | |
830 | return; | |
831 | } | |
832 | ||
833 | xenbus_switch_state(info->xbdev, XenbusStateConnected); | |
834 | ||
835 | /* Kick pending requests. */ | |
836 | spin_lock_irq(&blkif_io_lock); | |
837 | info->connected = BLKIF_STATE_CONNECTED; | |
838 | kick_pending_request_queues(info); | |
839 | spin_unlock_irq(&blkif_io_lock); | |
840 | ||
841 | add_disk(info->gd); | |
842 | } | |
843 | ||
844 | /** | |
845 | * Handle the change of state of the backend to Closing. We must delete our | |
846 | * device-layer structures now, to ensure that writes are flushed through to | |
847 | * the backend. Once is this done, we can switch to Closed in | |
848 | * acknowledgement. | |
849 | */ | |
850 | static void blkfront_closing(struct xenbus_device *dev) | |
851 | { | |
852 | struct blkfront_info *info = dev->dev.driver_data; | |
853 | unsigned long flags; | |
854 | ||
855 | dev_dbg(&dev->dev, "blkfront_closing: %s removed\n", dev->nodename); | |
856 | ||
857 | if (info->rq == NULL) | |
858 | goto out; | |
859 | ||
860 | spin_lock_irqsave(&blkif_io_lock, flags); | |
861 | ||
862 | del_gendisk(info->gd); | |
863 | ||
864 | /* No more blkif_request(). */ | |
865 | blk_stop_queue(info->rq); | |
866 | ||
867 | /* No more gnttab callback work. */ | |
868 | gnttab_cancel_free_callback(&info->callback); | |
869 | spin_unlock_irqrestore(&blkif_io_lock, flags); | |
870 | ||
871 | /* Flush gnttab callback work. Must be done with no locks held. */ | |
872 | flush_scheduled_work(); | |
873 | ||
874 | blk_cleanup_queue(info->rq); | |
875 | info->rq = NULL; | |
876 | ||
877 | out: | |
878 | xenbus_frontend_closed(dev); | |
879 | } | |
880 | ||
881 | /** | |
882 | * Callback received when the backend's state changes. | |
883 | */ | |
884 | static void backend_changed(struct xenbus_device *dev, | |
885 | enum xenbus_state backend_state) | |
886 | { | |
887 | struct blkfront_info *info = dev->dev.driver_data; | |
888 | struct block_device *bd; | |
889 | ||
890 | dev_dbg(&dev->dev, "blkfront:backend_changed.\n"); | |
891 | ||
892 | switch (backend_state) { | |
893 | case XenbusStateInitialising: | |
894 | case XenbusStateInitWait: | |
895 | case XenbusStateInitialised: | |
896 | case XenbusStateUnknown: | |
897 | case XenbusStateClosed: | |
898 | break; | |
899 | ||
900 | case XenbusStateConnected: | |
901 | blkfront_connect(info); | |
902 | break; | |
903 | ||
904 | case XenbusStateClosing: | |
53f0e8af | 905 | bd = bdget_disk(info->gd, 0); |
9f27ee59 JF |
906 | if (bd == NULL) |
907 | xenbus_dev_fatal(dev, -ENODEV, "bdget failed"); | |
908 | ||
909 | mutex_lock(&bd->bd_mutex); | |
910 | if (info->users > 0) | |
911 | xenbus_dev_error(dev, -EBUSY, | |
912 | "Device in use; refusing to close"); | |
913 | else | |
914 | blkfront_closing(dev); | |
915 | mutex_unlock(&bd->bd_mutex); | |
916 | bdput(bd); | |
917 | break; | |
918 | } | |
919 | } | |
920 | ||
921 | static int blkfront_remove(struct xenbus_device *dev) | |
922 | { | |
923 | struct blkfront_info *info = dev->dev.driver_data; | |
924 | ||
925 | dev_dbg(&dev->dev, "blkfront_remove: %s removed\n", dev->nodename); | |
926 | ||
927 | blkif_free(info, 0); | |
928 | ||
929 | kfree(info); | |
930 | ||
931 | return 0; | |
932 | } | |
933 | ||
934 | static int blkif_open(struct inode *inode, struct file *filep) | |
935 | { | |
936 | struct blkfront_info *info = inode->i_bdev->bd_disk->private_data; | |
937 | info->users++; | |
938 | return 0; | |
939 | } | |
940 | ||
941 | static int blkif_release(struct inode *inode, struct file *filep) | |
942 | { | |
943 | struct blkfront_info *info = inode->i_bdev->bd_disk->private_data; | |
944 | info->users--; | |
945 | if (info->users == 0) { | |
946 | /* Check whether we have been instructed to close. We will | |
947 | have ignored this request initially, as the device was | |
948 | still mounted. */ | |
949 | struct xenbus_device *dev = info->xbdev; | |
950 | enum xenbus_state state = xenbus_read_driver_state(dev->otherend); | |
951 | ||
952 | if (state == XenbusStateClosing) | |
953 | blkfront_closing(dev); | |
954 | } | |
955 | return 0; | |
956 | } | |
957 | ||
958 | static struct block_device_operations xlvbd_block_fops = | |
959 | { | |
960 | .owner = THIS_MODULE, | |
961 | .open = blkif_open, | |
962 | .release = blkif_release, | |
597592d9 | 963 | .getgeo = blkif_getgeo, |
9f27ee59 JF |
964 | }; |
965 | ||
966 | ||
967 | static struct xenbus_device_id blkfront_ids[] = { | |
968 | { "vbd" }, | |
969 | { "" } | |
970 | }; | |
971 | ||
972 | static struct xenbus_driver blkfront = { | |
973 | .name = "vbd", | |
974 | .owner = THIS_MODULE, | |
975 | .ids = blkfront_ids, | |
976 | .probe = blkfront_probe, | |
977 | .remove = blkfront_remove, | |
978 | .resume = blkfront_resume, | |
979 | .otherend_changed = backend_changed, | |
980 | }; | |
981 | ||
982 | static int __init xlblk_init(void) | |
983 | { | |
984 | if (!is_running_on_xen()) | |
985 | return -ENODEV; | |
986 | ||
987 | if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) { | |
988 | printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n", | |
989 | XENVBD_MAJOR, DEV_NAME); | |
990 | return -ENODEV; | |
991 | } | |
992 | ||
993 | return xenbus_register_frontend(&blkfront); | |
994 | } | |
995 | module_init(xlblk_init); | |
996 | ||
997 | ||
998 | static void xlblk_exit(void) | |
999 | { | |
1000 | return xenbus_unregister_driver(&blkfront); | |
1001 | } | |
1002 | module_exit(xlblk_exit); | |
1003 | ||
1004 | MODULE_DESCRIPTION("Xen virtual block device frontend"); | |
1005 | MODULE_LICENSE("GPL"); | |
1006 | MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR); |