?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
nbd: Fix struct name for structured reads
[qemu/ar7.git] / hw / block / xen_disk.c
blobe431bd89e8cf394334b9ec4ee63811a9d58b84a3
1 /*
2 * xen paravirt block device backend
3 *
4 * (c) Gerd Hoffmann <kraxel@redhat.com>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; under version 2 of the License.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, see <http://www.gnu.org/licenses/>.
17 *
18 * Contributions after 2012-01-13 are licensed under the terms of the
19 * GNU GPL, version 2 or (at your option) any later version.
20 */
21
22 #include "qemu/osdep.h"
23 #include <sys/ioctl.h>
24 #include <sys/uio.h>
25
26 #include "hw/hw.h"
27 #include "hw/xen/xen_backend.h"
28 #include "xen_blkif.h"
29 #include "sysemu/blockdev.h"
30 #include "sysemu/block-backend.h"
31 #include "qapi/error.h"
32 #include "qapi/qmp/qdict.h"
33 #include "qapi/qmp/qstring.h"
34
35 /* ------------------------------------------------------------- */
36
37 static int batch_maps = 0;
38
39 /* ------------------------------------------------------------- */
40
41 #define BLOCK_SIZE 512
42 #define IOCB_COUNT (BLKIF_MAX_SEGMENTS_PER_REQUEST + 2)
43
44 struct PersistentGrant {
45 void *page;
46 struct XenBlkDev *blkdev;
47 };
48
49 typedef struct PersistentGrant PersistentGrant;
50
51 struct PersistentRegion {
52 void *addr;
53 int num;
54 };
55
56 typedef struct PersistentRegion PersistentRegion;
57
58 struct ioreq {
59 blkif_request_t req;
60 int16_t status;
61
62 /* parsed request */
63 off_t start;
64 QEMUIOVector v;
65 int presync;
66 uint8_t mapped;
67
68 /* grant mapping */
69 uint32_t domids[BLKIF_MAX_SEGMENTS_PER_REQUEST];
70 uint32_t refs[BLKIF_MAX_SEGMENTS_PER_REQUEST];
71 int prot;
72 void *page[BLKIF_MAX_SEGMENTS_PER_REQUEST];
73 void *pages;
74 int num_unmap;
75
76 /* aio status */
77 int aio_inflight;
78 int aio_errors;
79
80 struct XenBlkDev *blkdev;
81 QLIST_ENTRY(ioreq) list;
82 BlockAcctCookie acct;
83 };
84
85 #define MAX_RING_PAGE_ORDER 4
86
87 struct XenBlkDev {
88 struct XenDevice xendev; /* must be first */
89 char *params;
90 char *mode;
91 char *type;
92 char *dev;
93 char *devtype;
94 bool directiosafe;
95 const char *fileproto;
96 const char *filename;
97 unsigned int ring_ref[1 << MAX_RING_PAGE_ORDER];
98 unsigned int nr_ring_ref;
99 void *sring;
100 int64_t file_blk;
101 int64_t file_size;
102 int protocol;
103 blkif_back_rings_t rings;
104 int more_work;
105 int cnt_map;
106
107 /* request lists */
108 QLIST_HEAD(inflight_head, ioreq) inflight;
109 QLIST_HEAD(finished_head, ioreq) finished;
110 QLIST_HEAD(freelist_head, ioreq) freelist;
111 int requests_total;
112 int requests_inflight;
113 int requests_finished;
114 unsigned int max_requests;
115
116 /* Persistent grants extension */
117 gboolean feature_discard;
118 gboolean feature_persistent;
119 GTree *persistent_gnts;
120 GSList *persistent_regions;
121 unsigned int persistent_gnt_count;
122 unsigned int max_grants;
123
124 /* qemu block driver */
125 DriveInfo *dinfo;
126 BlockBackend *blk;
127 QEMUBH *bh;
128 };
129
130 /* ------------------------------------------------------------- */
131
132 static void ioreq_reset(struct ioreq *ioreq)
133 {
134 memset(&ioreq->req, 0, sizeof(ioreq->req));
135 ioreq->status = 0;
136 ioreq->start = 0;
137 ioreq->presync = 0;
138 ioreq->mapped = 0;
139
140 memset(ioreq->domids, 0, sizeof(ioreq->domids));
141 memset(ioreq->refs, 0, sizeof(ioreq->refs));
142 ioreq->prot = 0;
143 memset(ioreq->page, 0, sizeof(ioreq->page));
144 ioreq->pages = NULL;
145
146 ioreq->aio_inflight = 0;
147 ioreq->aio_errors = 0;
148
149 ioreq->blkdev = NULL;
150 memset(&ioreq->list, 0, sizeof(ioreq->list));
151 memset(&ioreq->acct, 0, sizeof(ioreq->acct));
152
153 qemu_iovec_reset(&ioreq->v);
154 }
155
156 static gint int_cmp(gconstpointer a, gconstpointer b, gpointer user_data)
157 {
158 uint ua = GPOINTER_TO_UINT(a);
159 uint ub = GPOINTER_TO_UINT(b);
160 return (ua > ub) - (ua < ub);
161 }
162
163 static void destroy_grant(gpointer pgnt)
164 {
165 PersistentGrant *grant = pgnt;
166 xengnttab_handle *gnt = grant->blkdev->xendev.gnttabdev;
167
168 if (xengnttab_unmap(gnt, grant->page, 1) != 0) {
169 xen_pv_printf(&grant->blkdev->xendev, 0,
170 "xengnttab_unmap failed: %s\n",
171 strerror(errno));
172 }
173 grant->blkdev->persistent_gnt_count--;
174 xen_pv_printf(&grant->blkdev->xendev, 3,
175 "unmapped grant %p\n", grant->page);
176 g_free(grant);
177 }
178
179 static void remove_persistent_region(gpointer data, gpointer dev)
180 {
181 PersistentRegion *region = data;
182 struct XenBlkDev *blkdev = dev;
183 xengnttab_handle *gnt = blkdev->xendev.gnttabdev;
184
185 if (xengnttab_unmap(gnt, region->addr, region->num) != 0) {
186 xen_pv_printf(&blkdev->xendev, 0,
187 "xengnttab_unmap region %p failed: %s\n",
188 region->addr, strerror(errno));
189 }
190 xen_pv_printf(&blkdev->xendev, 3,
191 "unmapped grant region %p with %d pages\n",
192 region->addr, region->num);
193 g_free(region);
194 }
195
196 static struct ioreq *ioreq_start(struct XenBlkDev *blkdev)
197 {
198 struct ioreq *ioreq = NULL;
199
200 if (QLIST_EMPTY(&blkdev->freelist)) {
201 if (blkdev->requests_total >= blkdev->max_requests) {
202 goto out;
203 }
204 /* allocate new struct */
205 ioreq = g_malloc0(sizeof(*ioreq));
206 ioreq->blkdev = blkdev;
207 blkdev->requests_total++;
208 qemu_iovec_init(&ioreq->v, BLKIF_MAX_SEGMENTS_PER_REQUEST);
209 } else {
210 /* get one from freelist */
211 ioreq = QLIST_FIRST(&blkdev->freelist);
212 QLIST_REMOVE(ioreq, list);
213 }
214 QLIST_INSERT_HEAD(&blkdev->inflight, ioreq, list);
215 blkdev->requests_inflight++;
216
217 out:
218 return ioreq;
219 }
220
221 static void ioreq_finish(struct ioreq *ioreq)
222 {
223 struct XenBlkDev *blkdev = ioreq->blkdev;
224
225 QLIST_REMOVE(ioreq, list);
226 QLIST_INSERT_HEAD(&blkdev->finished, ioreq, list);
227 blkdev->requests_inflight--;
228 blkdev->requests_finished++;
229 }
230
231 static void ioreq_release(struct ioreq *ioreq, bool finish)
232 {
233 struct XenBlkDev *blkdev = ioreq->blkdev;
234
235 QLIST_REMOVE(ioreq, list);
236 ioreq_reset(ioreq);
237 ioreq->blkdev = blkdev;
238 QLIST_INSERT_HEAD(&blkdev->freelist, ioreq, list);
239 if (finish) {
240 blkdev->requests_finished--;
241 } else {
242 blkdev->requests_inflight--;
243 }
244 }
245
246 /*
247 * translate request into iovec + start offset
248 * do sanity checks along the way
249 */
250 static int ioreq_parse(struct ioreq *ioreq)
251 {
252 struct XenBlkDev *blkdev = ioreq->blkdev;
253 uintptr_t mem;
254 size_t len;
255 int i;
256
257 xen_pv_printf(&blkdev->xendev, 3,
258 "op %d, nr %d, handle %d, id %" PRId64 ", sector %" PRId64 "\n",
259 ioreq->req.operation, ioreq->req.nr_segments,
260 ioreq->req.handle, ioreq->req.id, ioreq->req.sector_number);
261 switch (ioreq->req.operation) {
262 case BLKIF_OP_READ:
263 ioreq->prot = PROT_WRITE; /* to memory */
264 break;
265 case BLKIF_OP_FLUSH_DISKCACHE:
266 ioreq->presync = 1;
267 if (!ioreq->req.nr_segments) {
268 return 0;
269 }
270 /* fall through */
271 case BLKIF_OP_WRITE:
272 ioreq->prot = PROT_READ; /* from memory */
273 break;
274 case BLKIF_OP_DISCARD:
275 return 0;
276 default:
277 xen_pv_printf(&blkdev->xendev, 0, "error: unknown operation (%d)\n",
278 ioreq->req.operation);
279 goto err;
280 };
281
282 if (ioreq->req.operation != BLKIF_OP_READ && blkdev->mode[0] != 'w') {
283 xen_pv_printf(&blkdev->xendev, 0, "error: write req for ro device\n");
284 goto err;
285 }
286
287 ioreq->start = ioreq->req.sector_number * blkdev->file_blk;
288 for (i = 0; i < ioreq->req.nr_segments; i++) {
289 if (i == BLKIF_MAX_SEGMENTS_PER_REQUEST) {
290 xen_pv_printf(&blkdev->xendev, 0, "error: nr_segments too big\n");
291 goto err;
292 }
293 if (ioreq->req.seg[i].first_sect > ioreq->req.seg[i].last_sect) {
294 xen_pv_printf(&blkdev->xendev, 0, "error: first > last sector\n");
295 goto err;
296 }
297 if (ioreq->req.seg[i].last_sect * BLOCK_SIZE >= XC_PAGE_SIZE) {
298 xen_pv_printf(&blkdev->xendev, 0, "error: page crossing\n");
299 goto err;
300 }
301
302 ioreq->domids[i] = blkdev->xendev.dom;
303 ioreq->refs[i] = ioreq->req.seg[i].gref;
304
305 mem = ioreq->req.seg[i].first_sect * blkdev->file_blk;
306 len = (ioreq->req.seg[i].last_sect - ioreq->req.seg[i].first_sect + 1) * blkdev->file_blk;
307 qemu_iovec_add(&ioreq->v, (void*)mem, len);
308 }
309 if (ioreq->start + ioreq->v.size > blkdev->file_size) {
310 xen_pv_printf(&blkdev->xendev, 0, "error: access beyond end of file\n");
311 goto err;
312 }
313 return 0;
314
315 err:
316 ioreq->status = BLKIF_RSP_ERROR;
317 return -1;
318 }
319
320 static void ioreq_unmap(struct ioreq *ioreq)
321 {
322 xengnttab_handle *gnt = ioreq->blkdev->xendev.gnttabdev;
323 int i;
324
325 if (ioreq->num_unmap == 0 || ioreq->mapped == 0) {
326 return;
327 }
328 if (batch_maps) {
329 if (!ioreq->pages) {
330 return;
331 }
332 if (xengnttab_unmap(gnt, ioreq->pages, ioreq->num_unmap) != 0) {
333 xen_pv_printf(&ioreq->blkdev->xendev, 0,
334 "xengnttab_unmap failed: %s\n",
335 strerror(errno));
336 }
337 ioreq->blkdev->cnt_map -= ioreq->num_unmap;
338 ioreq->pages = NULL;
339 } else {
340 for (i = 0; i < ioreq->num_unmap; i++) {
341 if (!ioreq->page[i]) {
342 continue;
343 }
344 if (xengnttab_unmap(gnt, ioreq->page[i], 1) != 0) {
345 xen_pv_printf(&ioreq->blkdev->xendev, 0,
346 "xengnttab_unmap failed: %s\n",
347 strerror(errno));
348 }
349 ioreq->blkdev->cnt_map--;
350 ioreq->page[i] = NULL;
351 }
352 }
353 ioreq->mapped = 0;
354 }
355
356 static int ioreq_map(struct ioreq *ioreq)
357 {
358 xengnttab_handle *gnt = ioreq->blkdev->xendev.gnttabdev;
359 uint32_t domids[BLKIF_MAX_SEGMENTS_PER_REQUEST];
360 uint32_t refs[BLKIF_MAX_SEGMENTS_PER_REQUEST];
361 void *page[BLKIF_MAX_SEGMENTS_PER_REQUEST];
362 int i, j, new_maps = 0;
363 PersistentGrant *grant;
364 PersistentRegion *region;
365 /* domids and refs variables will contain the information necessary
366 * to map the grants that are needed to fulfill this request.
367 *
368 * After mapping the needed grants, the page array will contain the
369 * memory address of each granted page in the order specified in ioreq
370 * (disregarding if it's a persistent grant or not).
371 */
372
373 if (ioreq->v.niov == 0 || ioreq->mapped == 1) {
374 return 0;
375 }
376 if (ioreq->blkdev->feature_persistent) {
377 for (i = 0; i < ioreq->v.niov; i++) {
378 grant = g_tree_lookup(ioreq->blkdev->persistent_gnts,
379 GUINT_TO_POINTER(ioreq->refs[i]));
380
381 if (grant != NULL) {
382 page[i] = grant->page;
383 xen_pv_printf(&ioreq->blkdev->xendev, 3,
384 "using persistent-grant %" PRIu32 "\n",
385 ioreq->refs[i]);
386 } else {
387 /* Add the grant to the list of grants that
388 * should be mapped
389 */
390 domids[new_maps] = ioreq->domids[i];
391 refs[new_maps] = ioreq->refs[i];
392 page[i] = NULL;
393 new_maps++;
394 }
395 }
396 /* Set the protection to RW, since grants may be reused later
397 * with a different protection than the one needed for this request
398 */
399 ioreq->prot = PROT_WRITE | PROT_READ;
400 } else {
401 /* All grants in the request should be mapped */
402 memcpy(refs, ioreq->refs, sizeof(refs));
403 memcpy(domids, ioreq->domids, sizeof(domids));
404 memset(page, 0, sizeof(page));
405 new_maps = ioreq->v.niov;
406 }
407
408 if (batch_maps && new_maps) {
409 ioreq->pages = xengnttab_map_grant_refs
410 (gnt, new_maps, domids, refs, ioreq->prot);
411 if (ioreq->pages == NULL) {
412 xen_pv_printf(&ioreq->blkdev->xendev, 0,
413 "can't map %d grant refs (%s, %d maps)\n",
414 new_maps, strerror(errno), ioreq->blkdev->cnt_map);
415 return -1;
416 }
417 for (i = 0, j = 0; i < ioreq->v.niov; i++) {
418 if (page[i] == NULL) {
419 page[i] = ioreq->pages + (j++) * XC_PAGE_SIZE;
420 }
421 }
422 ioreq->blkdev->cnt_map += new_maps;
423 } else if (new_maps) {
424 for (i = 0; i < new_maps; i++) {
425 ioreq->page[i] = xengnttab_map_grant_ref
426 (gnt, domids[i], refs[i], ioreq->prot);
427 if (ioreq->page[i] == NULL) {
428 xen_pv_printf(&ioreq->blkdev->xendev, 0,
429 "can't map grant ref %d (%s, %d maps)\n",
430 refs[i], strerror(errno), ioreq->blkdev->cnt_map);
431 ioreq->mapped = 1;
432 ioreq_unmap(ioreq);
433 return -1;
434 }
435 ioreq->blkdev->cnt_map++;
436 }
437 for (i = 0, j = 0; i < ioreq->v.niov; i++) {
438 if (page[i] == NULL) {
439 page[i] = ioreq->page[j++];
440 }
441 }
442 }
443 if (ioreq->blkdev->feature_persistent && new_maps != 0 &&
444 (!batch_maps || (ioreq->blkdev->persistent_gnt_count + new_maps <=
445 ioreq->blkdev->max_grants))) {
446 /*
447 * If we are using persistent grants and batch mappings only
448 * add the new maps to the list of persistent grants if the whole
449 * area can be persistently mapped.
450 */
451 if (batch_maps) {
452 region = g_malloc0(sizeof(*region));
453 region->addr = ioreq->pages;
454 region->num = new_maps;
455 ioreq->blkdev->persistent_regions = g_slist_append(
456 ioreq->blkdev->persistent_regions,
457 region);
458 }
459 while ((ioreq->blkdev->persistent_gnt_count < ioreq->blkdev->max_grants)
460 && new_maps) {
461 /* Go through the list of newly mapped grants and add as many
462 * as possible to the list of persistently mapped grants.
463 *
464 * Since we start at the end of ioreq->page(s), we only need
465 * to decrease new_maps to prevent this granted pages from
466 * being unmapped in ioreq_unmap.
467 */
468 grant = g_malloc0(sizeof(*grant));
469 new_maps--;
470 if (batch_maps) {
471 grant->page = ioreq->pages + (new_maps) * XC_PAGE_SIZE;
472 } else {
473 grant->page = ioreq->page[new_maps];
474 }
475 grant->blkdev = ioreq->blkdev;
476 xen_pv_printf(&ioreq->blkdev->xendev, 3,
477 "adding grant %" PRIu32 " page: %p\n",
478 refs[new_maps], grant->page);
479 g_tree_insert(ioreq->blkdev->persistent_gnts,
480 GUINT_TO_POINTER(refs[new_maps]),
481 grant);
482 ioreq->blkdev->persistent_gnt_count++;
483 }
484 assert(!batch_maps || new_maps == 0);
485 }
486 for (i = 0; i < ioreq->v.niov; i++) {
487 ioreq->v.iov[i].iov_base += (uintptr_t)page[i];
488 }
489 ioreq->mapped = 1;
490 ioreq->num_unmap = new_maps;
491 return 0;
492 }
493
494 #if CONFIG_XEN_CTRL_INTERFACE_VERSION >= 40800
495
496 static void ioreq_free_copy_buffers(struct ioreq *ioreq)
497 {
498 int i;
499
500 for (i = 0; i < ioreq->v.niov; i++) {
501 ioreq->page[i] = NULL;
502 }
503
504 qemu_vfree(ioreq->pages);
505 }
506
507 static int ioreq_init_copy_buffers(struct ioreq *ioreq)
508 {
509 int i;
510
511 if (ioreq->v.niov == 0) {
512 return 0;
513 }
514
515 ioreq->pages = qemu_memalign(XC_PAGE_SIZE, ioreq->v.niov * XC_PAGE_SIZE);
516
517 for (i = 0; i < ioreq->v.niov; i++) {
518 ioreq->page[i] = ioreq->pages + i * XC_PAGE_SIZE;
519 ioreq->v.iov[i].iov_base = ioreq->page[i];
520 }
521
522 return 0;
523 }
524
525 static int ioreq_grant_copy(struct ioreq *ioreq)
526 {
527 xengnttab_handle *gnt = ioreq->blkdev->xendev.gnttabdev;
528 xengnttab_grant_copy_segment_t segs[BLKIF_MAX_SEGMENTS_PER_REQUEST];
529 int i, count, rc;
530 int64_t file_blk = ioreq->blkdev->file_blk;
531
532 if (ioreq->v.niov == 0) {
533 return 0;
534 }
535
536 count = ioreq->v.niov;
537
538 for (i = 0; i < count; i++) {
539 if (ioreq->req.operation == BLKIF_OP_READ) {
540 segs[i].flags = GNTCOPY_dest_gref;
541 segs[i].dest.foreign.ref = ioreq->refs[i];
542 segs[i].dest.foreign.domid = ioreq->domids[i];
543 segs[i].dest.foreign.offset = ioreq->req.seg[i].first_sect * file_blk;
544 segs[i].source.virt = ioreq->v.iov[i].iov_base;
545 } else {
546 segs[i].flags = GNTCOPY_source_gref;
547 segs[i].source.foreign.ref = ioreq->refs[i];
548 segs[i].source.foreign.domid = ioreq->domids[i];
549 segs[i].source.foreign.offset = ioreq->req.seg[i].first_sect * file_blk;
550 segs[i].dest.virt = ioreq->v.iov[i].iov_base;
551 }
552 segs[i].len = (ioreq->req.seg[i].last_sect
553 - ioreq->req.seg[i].first_sect + 1) * file_blk;
554 }
555
556 rc = xengnttab_grant_copy(gnt, count, segs);
557
558 if (rc) {
559 xen_pv_printf(&ioreq->blkdev->xendev, 0,
560 "failed to copy data %d\n", rc);
561 ioreq->aio_errors++;
562 return -1;
563 }
564
565 for (i = 0; i < count; i++) {
566 if (segs[i].status != GNTST_okay) {
567 xen_pv_printf(&ioreq->blkdev->xendev, 3,
568 "failed to copy data %d for gref %d, domid %d\n",
569 segs[i].status, ioreq->refs[i], ioreq->domids[i]);
570 ioreq->aio_errors++;
571 rc = -1;
572 }
573 }
574
575 return rc;
576 }
577 #else
578 static void ioreq_free_copy_buffers(struct ioreq *ioreq)
579 {
580 abort();
581 }
582
583 static int ioreq_init_copy_buffers(struct ioreq *ioreq)
584 {
585 abort();
586 }
587
588 static int ioreq_grant_copy(struct ioreq *ioreq)
589 {
590 abort();
591 }
592 #endif
593
594 static int ioreq_runio_qemu_aio(struct ioreq *ioreq);
595
596 static void qemu_aio_complete(void *opaque, int ret)
597 {
598 struct ioreq *ioreq = opaque;
599
600 if (ret != 0) {
601 xen_pv_printf(&ioreq->blkdev->xendev, 0, "%s I/O error\n",
602 ioreq->req.operation == BLKIF_OP_READ ? "read" : "write");
603 ioreq->aio_errors++;
604 }
605
606 ioreq->aio_inflight--;
607 if (ioreq->presync) {
608 ioreq->presync = 0;
609 ioreq_runio_qemu_aio(ioreq);
610 return;
611 }
612 if (ioreq->aio_inflight > 0) {
613 return;
614 }
615
616 if (xen_feature_grant_copy) {
617 switch (ioreq->req.operation) {
618 case BLKIF_OP_READ:
619 /* in case of failure ioreq->aio_errors is increased */
620 if (ret == 0) {
621 ioreq_grant_copy(ioreq);
622 }
623 ioreq_free_copy_buffers(ioreq);
624 break;
625 case BLKIF_OP_WRITE:
626 case BLKIF_OP_FLUSH_DISKCACHE:
627 if (!ioreq->req.nr_segments) {
628 break;
629 }
630 ioreq_free_copy_buffers(ioreq);
631 break;
632 default:
633 break;
634 }
635 }
636
637 ioreq->status = ioreq->aio_errors ? BLKIF_RSP_ERROR : BLKIF_RSP_OKAY;
638 if (!xen_feature_grant_copy) {
639 ioreq_unmap(ioreq);
640 }
641 ioreq_finish(ioreq);
642 switch (ioreq->req.operation) {
643 case BLKIF_OP_WRITE:
644 case BLKIF_OP_FLUSH_DISKCACHE:
645 if (!ioreq->req.nr_segments) {
646 break;
647 }
648 case BLKIF_OP_READ:
649 if (ioreq->status == BLKIF_RSP_OKAY) {
650 block_acct_done(blk_get_stats(ioreq->blkdev->blk), &ioreq->acct);
651 } else {
652 block_acct_failed(blk_get_stats(ioreq->blkdev->blk), &ioreq->acct);
653 }
654 break;
655 case BLKIF_OP_DISCARD:
656 default:
657 break;
658 }
659 qemu_bh_schedule(ioreq->blkdev->bh);
660 }
661
662 static bool blk_split_discard(struct ioreq *ioreq, blkif_sector_t sector_number,
663 uint64_t nr_sectors)
664 {
665 struct XenBlkDev *blkdev = ioreq->blkdev;
666 int64_t byte_offset;
667 int byte_chunk;
668 uint64_t byte_remaining, limit;
669 uint64_t sec_start = sector_number;
670 uint64_t sec_count = nr_sectors;
671
672 /* Wrap around, or overflowing byte limit? */
673 if (sec_start + sec_count < sec_count ||
674 sec_start + sec_count > INT64_MAX >> BDRV_SECTOR_BITS) {
675 return false;
676 }
677
678 limit = BDRV_REQUEST_MAX_SECTORS << BDRV_SECTOR_BITS;
679 byte_offset = sec_start << BDRV_SECTOR_BITS;
680 byte_remaining = sec_count << BDRV_SECTOR_BITS;
681
682 do {
683 byte_chunk = byte_remaining > limit ? limit : byte_remaining;
684 ioreq->aio_inflight++;
685 blk_aio_pdiscard(blkdev->blk, byte_offset, byte_chunk,
686 qemu_aio_complete, ioreq);
687 byte_remaining -= byte_chunk;
688 byte_offset += byte_chunk;
689 } while (byte_remaining > 0);
690
691 return true;
692 }
693
694 static int ioreq_runio_qemu_aio(struct ioreq *ioreq)
695 {
696 struct XenBlkDev *blkdev = ioreq->blkdev;
697
698 if (xen_feature_grant_copy) {
699 ioreq_init_copy_buffers(ioreq);
700 if (ioreq->req.nr_segments && (ioreq->req.operation == BLKIF_OP_WRITE ||
701 ioreq->req.operation == BLKIF_OP_FLUSH_DISKCACHE) &&
702 ioreq_grant_copy(ioreq)) {
703 ioreq_free_copy_buffers(ioreq);
704 goto err;
705 }
706 } else {
707 if (ioreq->req.nr_segments && ioreq_map(ioreq)) {
708 goto err;
709 }
710 }
711
712 ioreq->aio_inflight++;
713 if (ioreq->presync) {
714 blk_aio_flush(ioreq->blkdev->blk, qemu_aio_complete, ioreq);
715 return 0;
716 }
717
718 switch (ioreq->req.operation) {
719 case BLKIF_OP_READ:
720 block_acct_start(blk_get_stats(blkdev->blk), &ioreq->acct,
721 ioreq->v.size, BLOCK_ACCT_READ);
722 ioreq->aio_inflight++;
723 blk_aio_preadv(blkdev->blk, ioreq->start, &ioreq->v, 0,
724 qemu_aio_complete, ioreq);
725 break;
726 case BLKIF_OP_WRITE:
727 case BLKIF_OP_FLUSH_DISKCACHE:
728 if (!ioreq->req.nr_segments) {
729 break;
730 }
731
732 block_acct_start(blk_get_stats(blkdev->blk), &ioreq->acct,
733 ioreq->v.size,
734 ioreq->req.operation == BLKIF_OP_WRITE ?
735 BLOCK_ACCT_WRITE : BLOCK_ACCT_FLUSH);
736 ioreq->aio_inflight++;
737 blk_aio_pwritev(blkdev->blk, ioreq->start, &ioreq->v, 0,
738 qemu_aio_complete, ioreq);
739 break;
740 case BLKIF_OP_DISCARD:
741 {
742 struct blkif_request_discard *req = (void *)&ioreq->req;
743 if (!blk_split_discard(ioreq, req->sector_number, req->nr_sectors)) {
744 goto err;
745 }
746 break;
747 }
748 default:
749 /* unknown operation (shouldn't happen -- parse catches this) */
750 if (!xen_feature_grant_copy) {
751 ioreq_unmap(ioreq);
752 }
753 goto err;
754 }
755
756 qemu_aio_complete(ioreq, 0);
757
758 return 0;
759
760 err:
761 ioreq_finish(ioreq);
762 ioreq->status = BLKIF_RSP_ERROR;
763 return -1;
764 }
765
766 static int blk_send_response_one(struct ioreq *ioreq)
767 {
768 struct XenBlkDev *blkdev = ioreq->blkdev;
769 int send_notify = 0;
770 int have_requests = 0;
771 blkif_response_t *resp;
772
773 /* Place on the response ring for the relevant domain. */
774 switch (blkdev->protocol) {
775 case BLKIF_PROTOCOL_NATIVE:
776 resp = (blkif_response_t *) RING_GET_RESPONSE(&blkdev->rings.native,
777 blkdev->rings.native.rsp_prod_pvt);
778 break;
779 case BLKIF_PROTOCOL_X86_32:
780 resp = (blkif_response_t *) RING_GET_RESPONSE(&blkdev->rings.x86_32_part,
781 blkdev->rings.x86_32_part.rsp_prod_pvt);
782 break;
783 case BLKIF_PROTOCOL_X86_64:
784 resp = (blkif_response_t *) RING_GET_RESPONSE(&blkdev->rings.x86_64_part,
785 blkdev->rings.x86_64_part.rsp_prod_pvt);
786 break;
787 default:
788 return 0;
789 }
790
791 resp->id = ioreq->req.id;
792 resp->operation = ioreq->req.operation;
793 resp->status = ioreq->status;
794
795 blkdev->rings.common.rsp_prod_pvt++;
796
797 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&blkdev->rings.common, send_notify);
798 if (blkdev->rings.common.rsp_prod_pvt == blkdev->rings.common.req_cons) {
799 /*
800 * Tail check for pending requests. Allows frontend to avoid
801 * notifications if requests are already in flight (lower
802 * overheads and promotes batching).
803 */
804 RING_FINAL_CHECK_FOR_REQUESTS(&blkdev->rings.common, have_requests);
805 } else if (RING_HAS_UNCONSUMED_REQUESTS(&blkdev->rings.common)) {
806 have_requests = 1;
807 }
808
809 if (have_requests) {
810 blkdev->more_work++;
811 }
812 return send_notify;
813 }
814
815 /* walk finished list, send outstanding responses, free requests */
816 static void blk_send_response_all(struct XenBlkDev *blkdev)
817 {
818 struct ioreq *ioreq;
819 int send_notify = 0;
820
821 while (!QLIST_EMPTY(&blkdev->finished)) {
822 ioreq = QLIST_FIRST(&blkdev->finished);
823 send_notify += blk_send_response_one(ioreq);
824 ioreq_release(ioreq, true);
825 }
826 if (send_notify) {
827 xen_pv_send_notify(&blkdev->xendev);
828 }
829 }
830
831 static int blk_get_request(struct XenBlkDev *blkdev, struct ioreq *ioreq, RING_IDX rc)
832 {
833 switch (blkdev->protocol) {
834 case BLKIF_PROTOCOL_NATIVE:
835 memcpy(&ioreq->req, RING_GET_REQUEST(&blkdev->rings.native, rc),
836 sizeof(ioreq->req));
837 break;
838 case BLKIF_PROTOCOL_X86_32:
839 blkif_get_x86_32_req(&ioreq->req,
840 RING_GET_REQUEST(&blkdev->rings.x86_32_part, rc));
841 break;
842 case BLKIF_PROTOCOL_X86_64:
843 blkif_get_x86_64_req(&ioreq->req,
844 RING_GET_REQUEST(&blkdev->rings.x86_64_part, rc));
845 break;
846 }
847 /* Prevent the compiler from accessing the on-ring fields instead. */
848 barrier();
849 return 0;
850 }
851
852 static void blk_handle_requests(struct XenBlkDev *blkdev)
853 {
854 RING_IDX rc, rp;
855 struct ioreq *ioreq;
856
857 blkdev->more_work = 0;
858
859 rc = blkdev->rings.common.req_cons;
860 rp = blkdev->rings.common.sring->req_prod;
861 xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
862
863 blk_send_response_all(blkdev);
864 while (rc != rp) {
865 /* pull request from ring */
866 if (RING_REQUEST_CONS_OVERFLOW(&blkdev->rings.common, rc)) {
867 break;
868 }
869 ioreq = ioreq_start(blkdev);
870 if (ioreq == NULL) {
871 blkdev->more_work++;
872 break;
873 }
874 blk_get_request(blkdev, ioreq, rc);
875 blkdev->rings.common.req_cons = ++rc;
876
877 /* parse them */
878 if (ioreq_parse(ioreq) != 0) {
879
880 switch (ioreq->req.operation) {
881 case BLKIF_OP_READ:
882 block_acct_invalid(blk_get_stats(blkdev->blk),
883 BLOCK_ACCT_READ);
884 break;
885 case BLKIF_OP_WRITE:
886 block_acct_invalid(blk_get_stats(blkdev->blk),
887 BLOCK_ACCT_WRITE);
888 break;
889 case BLKIF_OP_FLUSH_DISKCACHE:
890 block_acct_invalid(blk_get_stats(blkdev->blk),
891 BLOCK_ACCT_FLUSH);
892 default:
893 break;
894 };
895
896 if (blk_send_response_one(ioreq)) {
897 xen_pv_send_notify(&blkdev->xendev);
898 }
899 ioreq_release(ioreq, false);
900 continue;
901 }
902
903 ioreq_runio_qemu_aio(ioreq);
904 }
905
906 if (blkdev->more_work && blkdev->requests_inflight < blkdev->max_requests) {
907 qemu_bh_schedule(blkdev->bh);
908 }
909 }
910
911 /* ------------------------------------------------------------- */
912
913 static void blk_bh(void *opaque)
914 {
915 struct XenBlkDev *blkdev = opaque;
916 blk_handle_requests(blkdev);
917 }
918
919 static void blk_alloc(struct XenDevice *xendev)
920 {
921 struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
922
923 QLIST_INIT(&blkdev->inflight);
924 QLIST_INIT(&blkdev->finished);
925 QLIST_INIT(&blkdev->freelist);
926 blkdev->bh = qemu_bh_new(blk_bh, blkdev);
927 if (xen_mode != XEN_EMULATE) {
928 batch_maps = 1;
929 }
930 }
931
932 static void blk_parse_discard(struct XenBlkDev *blkdev)
933 {
934 int enable;
935
936 blkdev->feature_discard = true;
937
938 if (xenstore_read_be_int(&blkdev->xendev, "discard-enable", &enable) == 0) {
939 blkdev->feature_discard = !!enable;
940 }
941
942 if (blkdev->feature_discard) {
943 xenstore_write_be_int(&blkdev->xendev, "feature-discard", 1);
944 }
945 }
946
947 static int blk_init(struct XenDevice *xendev)
948 {
949 struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
950 int info = 0;
951 char *directiosafe = NULL;
952
953 /* read xenstore entries */
954 if (blkdev->params == NULL) {
955 char *h = NULL;
956 blkdev->params = xenstore_read_be_str(&blkdev->xendev, "params");
957 if (blkdev->params != NULL) {
958 h = strchr(blkdev->params, ':');
959 }
960 if (h != NULL) {
961 blkdev->fileproto = blkdev->params;
962 blkdev->filename = h+1;
963 *h = 0;
964 } else {
965 blkdev->fileproto = "<unset>";
966 blkdev->filename = blkdev->params;
967 }
968 }
969 if (!strcmp("aio", blkdev->fileproto)) {
970 blkdev->fileproto = "raw";
971 }
972 if (!strcmp("vhd", blkdev->fileproto)) {
973 blkdev->fileproto = "vpc";
974 }
975 if (blkdev->mode == NULL) {
976 blkdev->mode = xenstore_read_be_str(&blkdev->xendev, "mode");
977 }
978 if (blkdev->type == NULL) {
979 blkdev->type = xenstore_read_be_str(&blkdev->xendev, "type");
980 }
981 if (blkdev->dev == NULL) {
982 blkdev->dev = xenstore_read_be_str(&blkdev->xendev, "dev");
983 }
984 if (blkdev->devtype == NULL) {
985 blkdev->devtype = xenstore_read_be_str(&blkdev->xendev, "device-type");
986 }
987 directiosafe = xenstore_read_be_str(&blkdev->xendev, "direct-io-safe");
988 blkdev->directiosafe = (directiosafe && atoi(directiosafe));
989
990 /* do we have all we need? */
991 if (blkdev->params == NULL ||
992 blkdev->mode == NULL ||
993 blkdev->type == NULL ||
994 blkdev->dev == NULL) {
995 goto out_error;
996 }
997
998 /* read-only ? */
999 if (strcmp(blkdev->mode, "w")) {
1000 info |= VDISK_READONLY;
1001 }
1002
1003 /* cdrom ? */
1004 if (blkdev->devtype && !strcmp(blkdev->devtype, "cdrom")) {
1005 info |= VDISK_CDROM;
1006 }
1007
1008 blkdev->file_blk = BLOCK_SIZE;
1009
1010 xen_pv_printf(&blkdev->xendev, 3, "grant copy operation %s\n",
1011 xen_feature_grant_copy ? "enabled" : "disabled");
1012
1013 /* fill info
1014 * blk_connect supplies sector-size and sectors
1015 */
1016 xenstore_write_be_int(&blkdev->xendev, "feature-flush-cache", 1);
1017 xenstore_write_be_int(&blkdev->xendev, "feature-persistent",
1018 !xen_feature_grant_copy);
1019 xenstore_write_be_int(&blkdev->xendev, "info", info);
1020
1021 xenstore_write_be_int(&blkdev->xendev, "max-ring-page-order",
1022 MAX_RING_PAGE_ORDER);
1023
1024 blk_parse_discard(blkdev);
1025
1026 g_free(directiosafe);
1027 return 0;
1028
1029 out_error:
1030 g_free(blkdev->params);
1031 blkdev->params = NULL;
1032 g_free(blkdev->mode);
1033 blkdev->mode = NULL;
1034 g_free(blkdev->type);
1035 blkdev->type = NULL;
1036 g_free(blkdev->dev);
1037 blkdev->dev = NULL;
1038 g_free(blkdev->devtype);
1039 blkdev->devtype = NULL;
1040 g_free(directiosafe);
1041 blkdev->directiosafe = false;
1042 return -1;
1043 }
1044
1045 /*
1046 * We need to account for the grant allocations requiring contiguous
1047 * chunks; the worst case number would be
1048 * max_req * max_seg + (max_req - 1) * (max_seg - 1) + 1,
1049 * but in order to keep things simple just use
1050 * 2 * max_req * max_seg.
1051 */
1052 #define MAX_GRANTS(max_req, max_seg) (2 * (max_req) * (max_seg))
1053
1054 static int blk_connect(struct XenDevice *xendev)
1055 {
1056 struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
1057 int pers, index, qflags;
1058 bool readonly = true;
1059 bool writethrough = true;
1060 int order, ring_ref;
1061 unsigned int ring_size, max_grants;
1062 unsigned int i;
1063 uint32_t *domids;
1064
1065 /* read-only ? */
1066 if (blkdev->directiosafe) {
1067 qflags = BDRV_O_NOCACHE | BDRV_O_NATIVE_AIO;
1068 } else {
1069 qflags = 0;
1070 writethrough = false;
1071 }
1072 if (strcmp(blkdev->mode, "w") == 0) {
1073 qflags |= BDRV_O_RDWR;
1074 readonly = false;
1075 }
1076 if (blkdev->feature_discard) {
1077 qflags |= BDRV_O_UNMAP;
1078 }
1079
1080 /* init qemu block driver */
1081 index = (blkdev->xendev.dev - 202 * 256) / 16;
1082 blkdev->dinfo = drive_get(IF_XEN, 0, index);
1083 if (!blkdev->dinfo) {
1084 Error *local_err = NULL;
1085 QDict *options = NULL;
1086
1087 if (strcmp(blkdev->fileproto, "<unset>")) {
1088 options = qdict_new();
1089 qdict_put_str(options, "driver", blkdev->fileproto);
1090 }
1091
1092 /* setup via xenbus -> create new block driver instance */
1093 xen_pv_printf(&blkdev->xendev, 2, "create new bdrv (xenbus setup)\n");
1094 blkdev->blk = blk_new_open(blkdev->filename, NULL, options,
1095 qflags, &local_err);
1096 if (!blkdev->blk) {
1097 xen_pv_printf(&blkdev->xendev, 0, "error: %s\n",
1098 error_get_pretty(local_err));
1099 error_free(local_err);
1100 return -1;
1101 }
1102 blk_set_enable_write_cache(blkdev->blk, !writethrough);
1103 } else {
1104 /* setup via qemu cmdline -> already setup for us */
1105 xen_pv_printf(&blkdev->xendev, 2,
1106 "get configured bdrv (cmdline setup)\n");
1107 blkdev->blk = blk_by_legacy_dinfo(blkdev->dinfo);
1108 if (blk_is_read_only(blkdev->blk) && !readonly) {
1109 xen_pv_printf(&blkdev->xendev, 0, "Unexpected read-only drive");
1110 blkdev->blk = NULL;
1111 return -1;
1112 }
1113 /* blkdev->blk is not create by us, we get a reference
1114 * so we can blk_unref() unconditionally */
1115 blk_ref(blkdev->blk);
1116 }
1117 blk_attach_dev_legacy(blkdev->blk, blkdev);
1118 blkdev->file_size = blk_getlength(blkdev->blk);
1119 if (blkdev->file_size < 0) {
1120 BlockDriverState *bs = blk_bs(blkdev->blk);
1121 const char *drv_name = bs ? bdrv_get_format_name(bs) : NULL;
1122 xen_pv_printf(&blkdev->xendev, 1, "blk_getlength: %d (%s) | drv %s\n",
1123 (int)blkdev->file_size, strerror(-blkdev->file_size),
1124 drv_name ?: "-");
1125 blkdev->file_size = 0;
1126 }
1127
1128 xen_pv_printf(xendev, 1, "type \"%s\", fileproto \"%s\", filename \"%s\","
1129 " size %" PRId64 " (%" PRId64 " MB)\n",
1130 blkdev->type, blkdev->fileproto, blkdev->filename,
1131 blkdev->file_size, blkdev->file_size >> 20);
1132
1133 /* Fill in number of sector size and number of sectors */
1134 xenstore_write_be_int(&blkdev->xendev, "sector-size", blkdev->file_blk);
1135 xenstore_write_be_int64(&blkdev->xendev, "sectors",
1136 blkdev->file_size / blkdev->file_blk);
1137
1138 if (xenstore_read_fe_int(&blkdev->xendev, "ring-page-order",
1139 &order) == -1) {
1140 blkdev->nr_ring_ref = 1;
1141
1142 if (xenstore_read_fe_int(&blkdev->xendev, "ring-ref",
1143 &ring_ref) == -1) {
1144 return -1;
1145 }
1146 blkdev->ring_ref[0] = ring_ref;
1147
1148 } else if (order >= 0 && order <= MAX_RING_PAGE_ORDER) {
1149 blkdev->nr_ring_ref = 1 << order;
1150
1151 for (i = 0; i < blkdev->nr_ring_ref; i++) {
1152 char *key;
1153
1154 key = g_strdup_printf("ring-ref%u", i);
1155 if (!key) {
1156 return -1;
1157 }
1158
1159 if (xenstore_read_fe_int(&blkdev->xendev, key,
1160 &ring_ref) == -1) {
1161 g_free(key);
1162 return -1;
1163 }
1164 blkdev->ring_ref[i] = ring_ref;
1165
1166 g_free(key);
1167 }
1168 } else {
1169 xen_pv_printf(xendev, 0, "invalid ring-page-order: %d\n",
1170 order);
1171 return -1;
1172 }
1173
1174 if (xenstore_read_fe_int(&blkdev->xendev, "event-channel",
1175 &blkdev->xendev.remote_port) == -1) {
1176 return -1;
1177 }
1178 if (xenstore_read_fe_int(&blkdev->xendev, "feature-persistent", &pers)) {
1179 blkdev->feature_persistent = FALSE;
1180 } else {
1181 blkdev->feature_persistent = !!pers;
1182 }
1183
1184 if (!blkdev->xendev.protocol) {
1185 blkdev->protocol = BLKIF_PROTOCOL_NATIVE;
1186 } else if (strcmp(blkdev->xendev.protocol, XEN_IO_PROTO_ABI_NATIVE) == 0) {
1187 blkdev->protocol = BLKIF_PROTOCOL_NATIVE;
1188 } else if (strcmp(blkdev->xendev.protocol, XEN_IO_PROTO_ABI_X86_32) == 0) {
1189 blkdev->protocol = BLKIF_PROTOCOL_X86_32;
1190 } else if (strcmp(blkdev->xendev.protocol, XEN_IO_PROTO_ABI_X86_64) == 0) {
1191 blkdev->protocol = BLKIF_PROTOCOL_X86_64;
1192 } else {
1193 blkdev->protocol = BLKIF_PROTOCOL_NATIVE;
1194 }
1195
1196 ring_size = XC_PAGE_SIZE * blkdev->nr_ring_ref;
1197 switch (blkdev->protocol) {
1198 case BLKIF_PROTOCOL_NATIVE:
1199 {
1200 blkdev->max_requests = __CONST_RING_SIZE(blkif, ring_size);
1201 break;
1202 }
1203 case BLKIF_PROTOCOL_X86_32:
1204 {
1205 blkdev->max_requests = __CONST_RING_SIZE(blkif_x86_32, ring_size);
1206 break;
1207 }
1208 case BLKIF_PROTOCOL_X86_64:
1209 {
1210 blkdev->max_requests = __CONST_RING_SIZE(blkif_x86_64, ring_size);
1211 break;
1212 }
1213 default:
1214 return -1;
1215 }
1216
1217 /* Calculate the maximum number of grants needed by ioreqs */
1218 max_grants = MAX_GRANTS(blkdev->max_requests,
1219 BLKIF_MAX_SEGMENTS_PER_REQUEST);
1220 /* Add on the number needed for the ring pages */
1221 max_grants += blkdev->nr_ring_ref;
1222
1223 blkdev->xendev.gnttabdev = xengnttab_open(NULL, 0);
1224 if (blkdev->xendev.gnttabdev == NULL) {
1225 xen_pv_printf(xendev, 0, "xengnttab_open failed: %s\n",
1226 strerror(errno));
1227 return -1;
1228 }
1229 if (xengnttab_set_max_grants(blkdev->xendev.gnttabdev, max_grants)) {
1230 xen_pv_printf(xendev, 0, "xengnttab_set_max_grants failed: %s\n",
1231 strerror(errno));
1232 return -1;
1233 }
1234
1235 domids = g_new0(uint32_t, blkdev->nr_ring_ref);
1236 for (i = 0; i < blkdev->nr_ring_ref; i++) {
1237 domids[i] = blkdev->xendev.dom;
1238 }
1239
1240 blkdev->sring = xengnttab_map_grant_refs(blkdev->xendev.gnttabdev,
1241 blkdev->nr_ring_ref,
1242 domids,
1243 blkdev->ring_ref,
1244 PROT_READ | PROT_WRITE);
1245
1246 g_free(domids);
1247
1248 if (!blkdev->sring) {
1249 return -1;
1250 }
1251
1252 blkdev->cnt_map++;
1253
1254 switch (blkdev->protocol) {
1255 case BLKIF_PROTOCOL_NATIVE:
1256 {
1257 blkif_sring_t *sring_native = blkdev->sring;
1258 BACK_RING_INIT(&blkdev->rings.native, sring_native, ring_size);
1259 break;
1260 }
1261 case BLKIF_PROTOCOL_X86_32:
1262 {
1263 blkif_x86_32_sring_t *sring_x86_32 = blkdev->sring;
1264
1265 BACK_RING_INIT(&blkdev->rings.x86_32_part, sring_x86_32, ring_size);
1266 break;
1267 }
1268 case BLKIF_PROTOCOL_X86_64:
1269 {
1270 blkif_x86_64_sring_t *sring_x86_64 = blkdev->sring;
1271
1272 BACK_RING_INIT(&blkdev->rings.x86_64_part, sring_x86_64, ring_size);
1273 break;
1274 }
1275 }
1276
1277 if (blkdev->feature_persistent) {
1278 /* Init persistent grants */
1279 blkdev->max_grants = blkdev->max_requests *
1280 BLKIF_MAX_SEGMENTS_PER_REQUEST;
1281 blkdev->persistent_gnts = g_tree_new_full((GCompareDataFunc)int_cmp,
1282 NULL, NULL,
1283 batch_maps ?
1284 (GDestroyNotify)g_free :
1285 (GDestroyNotify)destroy_grant);
1286 blkdev->persistent_regions = NULL;
1287 blkdev->persistent_gnt_count = 0;
1288 }
1289
1290 xen_be_bind_evtchn(&blkdev->xendev);
1291
1292 xen_pv_printf(&blkdev->xendev, 1, "ok: proto %s, nr-ring-ref %u, "
1293 "remote port %d, local port %d\n",
1294 blkdev->xendev.protocol, blkdev->nr_ring_ref,
1295 blkdev->xendev.remote_port, blkdev->xendev.local_port);
1296 return 0;
1297 }
1298
1299 static void blk_disconnect(struct XenDevice *xendev)
1300 {
1301 struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
1302
1303 if (blkdev->blk) {
1304 blk_detach_dev(blkdev->blk, blkdev);
1305 blk_unref(blkdev->blk);
1306 blkdev->blk = NULL;
1307 }
1308 xen_pv_unbind_evtchn(&blkdev->xendev);
1309
1310 if (blkdev->sring) {
1311 xengnttab_unmap(blkdev->xendev.gnttabdev, blkdev->sring,
1312 blkdev->nr_ring_ref);
1313 blkdev->cnt_map--;
1314 blkdev->sring = NULL;
1315 }
1316
1317 /*
1318 * Unmap persistent grants before switching to the closed state
1319 * so the frontend can free them.
1320 *
1321 * In the !batch_maps case g_tree_destroy will take care of unmapping
1322 * the grant, but in the batch_maps case we need to iterate over every
1323 * region in persistent_regions and unmap it.
1324 */
1325 if (blkdev->feature_persistent) {
1326 g_tree_destroy(blkdev->persistent_gnts);
1327 assert(batch_maps || blkdev->persistent_gnt_count == 0);
1328 if (batch_maps) {
1329 blkdev->persistent_gnt_count = 0;
1330 g_slist_foreach(blkdev->persistent_regions,
1331 (GFunc)remove_persistent_region, blkdev);
1332 g_slist_free(blkdev->persistent_regions);
1333 }
1334 blkdev->feature_persistent = false;
1335 }
1336
1337 if (blkdev->xendev.gnttabdev) {
1338 xengnttab_close(blkdev->xendev.gnttabdev);
1339 blkdev->xendev.gnttabdev = NULL;
1340 }
1341 }
1342
1343 static int blk_free(struct XenDevice *xendev)
1344 {
1345 struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
1346 struct ioreq *ioreq;
1347
1348 blk_disconnect(xendev);
1349
1350 while (!QLIST_EMPTY(&blkdev->freelist)) {
1351 ioreq = QLIST_FIRST(&blkdev->freelist);
1352 QLIST_REMOVE(ioreq, list);
1353 qemu_iovec_destroy(&ioreq->v);
1354 g_free(ioreq);
1355 }
1356
1357 g_free(blkdev->params);
1358 g_free(blkdev->mode);
1359 g_free(blkdev->type);
1360 g_free(blkdev->dev);
1361 g_free(blkdev->devtype);
1362 qemu_bh_delete(blkdev->bh);
1363 return 0;
1364 }
1365
1366 static void blk_event(struct XenDevice *xendev)
1367 {
1368 struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
1369
1370 qemu_bh_schedule(blkdev->bh);
1371 }
1372
1373 struct XenDevOps xen_blkdev_ops = {
1374 .size = sizeof(struct XenBlkDev),
1375 .alloc = blk_alloc,
1376 .init = blk_init,
1377 .initialise = blk_connect,
1378 .disconnect = blk_disconnect,
1379 .event = blk_event,
1380 .free = blk_free,
1381 };
AR7 emulation for QEMU