1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22#include "qemu/osdep.h"
23#include <sys/ioctl.h>
24#include <sys/mman.h>
25#include <sys/uio.h>
26
27#include "hw/hw.h"
28#include "hw/xen/xen_backend.h"
29#include "xen_blkif.h"
30#include "sysemu/blockdev.h"
31#include "sysemu/block-backend.h"
32#include "qapi/error.h"
33#include "qapi/qmp/qdict.h"
34#include "qapi/qmp/qstring.h"
35
36
37
38static int batch_maps = 0;
39
40static int max_requests = 32;
41
42
43
44#define BLOCK_SIZE 512
45#define IOCB_COUNT (BLKIF_MAX_SEGMENTS_PER_REQUEST + 2)
46
47struct PersistentGrant {
48 void *page;
49 struct XenBlkDev *blkdev;
50};
51
52typedef struct PersistentGrant PersistentGrant;
53
54struct PersistentRegion {
55 void *addr;
56 int num;
57};
58
59typedef struct PersistentRegion PersistentRegion;
60
61struct ioreq {
62 blkif_request_t req;
63 int16_t status;
64
65
66 off_t start;
67 QEMUIOVector v;
68 int presync;
69 uint8_t mapped;
70
71
72 uint32_t domids[BLKIF_MAX_SEGMENTS_PER_REQUEST];
73 uint32_t refs[BLKIF_MAX_SEGMENTS_PER_REQUEST];
74 int prot;
75 void *page[BLKIF_MAX_SEGMENTS_PER_REQUEST];
76 void *pages;
77 int num_unmap;
78
79
80 int aio_inflight;
81 int aio_errors;
82
83 struct XenBlkDev *blkdev;
84 QLIST_ENTRY(ioreq) list;
85 BlockAcctCookie acct;
86};
87
88struct XenBlkDev {
89 struct XenDevice xendev;
90 char *params;
91 char *mode;
92 char *type;
93 char *dev;
94 char *devtype;
95 bool directiosafe;
96 const char *fileproto;
97 const char *filename;
98 int 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
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
115
116 gboolean feature_discard;
117 gboolean feature_persistent;
118 GTree *persistent_gnts;
119 GSList *persistent_regions;
120 unsigned int persistent_gnt_count;
121 unsigned int max_grants;
122
123
124 DriveInfo *dinfo;
125 BlockBackend *blk;
126 QEMUBH *bh;
127};
128
129
130
131static void ioreq_reset(struct ioreq *ioreq)
132{
133 memset(&ioreq->req, 0, sizeof(ioreq->req));
134 ioreq->status = 0;
135 ioreq->start = 0;
136 ioreq->presync = 0;
137 ioreq->mapped = 0;
138
139 memset(ioreq->domids, 0, sizeof(ioreq->domids));
140 memset(ioreq->refs, 0, sizeof(ioreq->refs));
141 ioreq->prot = 0;
142 memset(ioreq->page, 0, sizeof(ioreq->page));
143 ioreq->pages = NULL;
144
145 ioreq->aio_inflight = 0;
146 ioreq->aio_errors = 0;
147
148 ioreq->blkdev = NULL;
149 memset(&ioreq->list, 0, sizeof(ioreq->list));
150 memset(&ioreq->acct, 0, sizeof(ioreq->acct));
151
152 qemu_iovec_reset(&ioreq->v);
153}
154
155static gint int_cmp(gconstpointer a, gconstpointer b, gpointer user_data)
156{
157 uint ua = GPOINTER_TO_UINT(a);
158 uint ub = GPOINTER_TO_UINT(b);
159 return (ua > ub) - (ua < ub);
160}
161
162static void destroy_grant(gpointer pgnt)
163{
164 PersistentGrant *grant = pgnt;
165 xengnttab_handle *gnt = grant->blkdev->xendev.gnttabdev;
166
167 if (xengnttab_unmap(gnt, grant->page, 1) != 0) {
168 xen_be_printf(&grant->blkdev->xendev, 0,
169 "xengnttab_unmap failed: %s\n",
170 strerror(errno));
171 }
172 grant->blkdev->persistent_gnt_count--;
173 xen_be_printf(&grant->blkdev->xendev, 3,
174 "unmapped grant %p\n", grant->page);
175 g_free(grant);
176}
177
178static void remove_persistent_region(gpointer data, gpointer dev)
179{
180 PersistentRegion *region = data;
181 struct XenBlkDev *blkdev = dev;
182 xengnttab_handle *gnt = blkdev->xendev.gnttabdev;
183
184 if (xengnttab_unmap(gnt, region->addr, region->num) != 0) {
185 xen_be_printf(&blkdev->xendev, 0,
186 "xengnttab_unmap region %p failed: %s\n",
187 region->addr, strerror(errno));
188 }
189 xen_be_printf(&blkdev->xendev, 3,
190 "unmapped grant region %p with %d pages\n",
191 region->addr, region->num);
192 g_free(region);
193}
194
195static struct ioreq *ioreq_start(struct XenBlkDev *blkdev)
196{
197 struct ioreq *ioreq = NULL;
198
199 if (QLIST_EMPTY(&blkdev->freelist)) {
200 if (blkdev->requests_total >= max_requests) {
201 goto out;
202 }
203
204 ioreq = g_malloc0(sizeof(*ioreq));
205 ioreq->blkdev = blkdev;
206 blkdev->requests_total++;
207 qemu_iovec_init(&ioreq->v, BLKIF_MAX_SEGMENTS_PER_REQUEST);
208 } else {
209
210 ioreq = QLIST_FIRST(&blkdev->freelist);
211 QLIST_REMOVE(ioreq, list);
212 }
213 QLIST_INSERT_HEAD(&blkdev->inflight, ioreq, list);
214 blkdev->requests_inflight++;
215
216out:
217 return ioreq;
218}
219
220static void ioreq_finish(struct ioreq *ioreq)
221{
222 struct XenBlkDev *blkdev = ioreq->blkdev;
223
224 QLIST_REMOVE(ioreq, list);
225 QLIST_INSERT_HEAD(&blkdev->finished, ioreq, list);
226 blkdev->requests_inflight--;
227 blkdev->requests_finished++;
228}
229
230static void ioreq_release(struct ioreq *ioreq, bool finish)
231{
232 struct XenBlkDev *blkdev = ioreq->blkdev;
233
234 QLIST_REMOVE(ioreq, list);
235 ioreq_reset(ioreq);
236 ioreq->blkdev = blkdev;
237 QLIST_INSERT_HEAD(&blkdev->freelist, ioreq, list);
238 if (finish) {
239 blkdev->requests_finished--;
240 } else {
241 blkdev->requests_inflight--;
242 }
243}
244
245
246
247
248
249static int ioreq_parse(struct ioreq *ioreq)
250{
251 struct XenBlkDev *blkdev = ioreq->blkdev;
252 uintptr_t mem;
253 size_t len;
254 int i;
255
256 xen_be_printf(&blkdev->xendev, 3,
257 "op %d, nr %d, handle %d, id %" PRId64 ", sector %" PRId64 "\n",
258 ioreq->req.operation, ioreq->req.nr_segments,
259 ioreq->req.handle, ioreq->req.id, ioreq->req.sector_number);
260 switch (ioreq->req.operation) {
261 case BLKIF_OP_READ:
262 ioreq->prot = PROT_WRITE;
263 break;
264 case BLKIF_OP_FLUSH_DISKCACHE:
265 ioreq->presync = 1;
266 if (!ioreq->req.nr_segments) {
267 return 0;
268 }
269
270 case BLKIF_OP_WRITE:
271 ioreq->prot = PROT_READ;
272 break;
273 case BLKIF_OP_DISCARD:
274 return 0;
275 default:
276 xen_be_printf(&blkdev->xendev, 0, "error: unknown operation (%d)\n",
277 ioreq->req.operation);
278 goto err;
279 };
280
281 if (ioreq->req.operation != BLKIF_OP_READ && blkdev->mode[0] != 'w') {
282 xen_be_printf(&blkdev->xendev, 0, "error: write req for ro device\n");
283 goto err;
284 }
285
286 ioreq->start = ioreq->req.sector_number * blkdev->file_blk;
287 for (i = 0; i < ioreq->req.nr_segments; i++) {
288 if (i == BLKIF_MAX_SEGMENTS_PER_REQUEST) {
289 xen_be_printf(&blkdev->xendev, 0, "error: nr_segments too big\n");
290 goto err;
291 }
292 if (ioreq->req.seg[i].first_sect > ioreq->req.seg[i].last_sect) {
293 xen_be_printf(&blkdev->xendev, 0, "error: first > last sector\n");
294 goto err;
295 }
296 if (ioreq->req.seg[i].last_sect * BLOCK_SIZE >= XC_PAGE_SIZE) {
297 xen_be_printf(&blkdev->xendev, 0, "error: page crossing\n");
298 goto err;
299 }
300
301 ioreq->domids[i] = blkdev->xendev.dom;
302 ioreq->refs[i] = ioreq->req.seg[i].gref;
303
304 mem = ioreq->req.seg[i].first_sect * blkdev->file_blk;
305 len = (ioreq->req.seg[i].last_sect - ioreq->req.seg[i].first_sect + 1) * blkdev->file_blk;
306 qemu_iovec_add(&ioreq->v, (void*)mem, len);
307 }
308 if (ioreq->start + ioreq->v.size > blkdev->file_size) {
309 xen_be_printf(&blkdev->xendev, 0, "error: access beyond end of file\n");
310 goto err;
311 }
312 return 0;
313
314err:
315 ioreq->status = BLKIF_RSP_ERROR;
316 return -1;
317}
318
319static void ioreq_unmap(struct ioreq *ioreq)
320{
321 xengnttab_handle *gnt = ioreq->blkdev->xendev.gnttabdev;
322 int i;
323
324 if (ioreq->num_unmap == 0 || ioreq->mapped == 0) {
325 return;
326 }
327 if (batch_maps) {
328 if (!ioreq->pages) {
329 return;
330 }
331 if (xengnttab_unmap(gnt, ioreq->pages, ioreq->num_unmap) != 0) {
332 xen_be_printf(&ioreq->blkdev->xendev, 0,
333 "xengnttab_unmap failed: %s\n",
334 strerror(errno));
335 }
336 ioreq->blkdev->cnt_map -= ioreq->num_unmap;
337 ioreq->pages = NULL;
338 } else {
339 for (i = 0; i < ioreq->num_unmap; i++) {
340 if (!ioreq->page[i]) {
341 continue;
342 }
343 if (xengnttab_unmap(gnt, ioreq->page[i], 1) != 0) {
344 xen_be_printf(&ioreq->blkdev->xendev, 0,
345 "xengnttab_unmap failed: %s\n",
346 strerror(errno));
347 }
348 ioreq->blkdev->cnt_map--;
349 ioreq->page[i] = NULL;
350 }
351 }
352 ioreq->mapped = 0;
353}
354
355static int ioreq_map(struct ioreq *ioreq)
356{
357 xengnttab_handle *gnt = ioreq->blkdev->xendev.gnttabdev;
358 uint32_t domids[BLKIF_MAX_SEGMENTS_PER_REQUEST];
359 uint32_t refs[BLKIF_MAX_SEGMENTS_PER_REQUEST];
360 void *page[BLKIF_MAX_SEGMENTS_PER_REQUEST];
361 int i, j, new_maps = 0;
362 PersistentGrant *grant;
363 PersistentRegion *region;
364
365
366
367
368
369
370
371
372 if (ioreq->v.niov == 0 || ioreq->mapped == 1) {
373 return 0;
374 }
375 if (ioreq->blkdev->feature_persistent) {
376 for (i = 0; i < ioreq->v.niov; i++) {
377 grant = g_tree_lookup(ioreq->blkdev->persistent_gnts,
378 GUINT_TO_POINTER(ioreq->refs[i]));
379
380 if (grant != NULL) {
381 page[i] = grant->page;
382 xen_be_printf(&ioreq->blkdev->xendev, 3,
383 "using persistent-grant %" PRIu32 "\n",
384 ioreq->refs[i]);
385 } else {
386
387
388
389 domids[new_maps] = ioreq->domids[i];
390 refs[new_maps] = ioreq->refs[i];
391 page[i] = NULL;
392 new_maps++;
393 }
394 }
395
396
397
398 ioreq->prot = PROT_WRITE | PROT_READ;
399 } else {
400
401 memcpy(refs, ioreq->refs, sizeof(refs));
402 memcpy(domids, ioreq->domids, sizeof(domids));
403 memset(page, 0, sizeof(page));
404 new_maps = ioreq->v.niov;
405 }
406
407 if (batch_maps && new_maps) {
408 ioreq->pages = xengnttab_map_grant_refs
409 (gnt, new_maps, domids, refs, ioreq->prot);
410 if (ioreq->pages == NULL) {
411 xen_be_printf(&ioreq->blkdev->xendev, 0,
412 "can't map %d grant refs (%s, %d maps)\n",
413 new_maps, strerror(errno), ioreq->blkdev->cnt_map);
414 return -1;
415 }
416 for (i = 0, j = 0; i < ioreq->v.niov; i++) {
417 if (page[i] == NULL) {
418 page[i] = ioreq->pages + (j++) * XC_PAGE_SIZE;
419 }
420 }
421 ioreq->blkdev->cnt_map += new_maps;
422 } else if (new_maps) {
423 for (i = 0; i < new_maps; i++) {
424 ioreq->page[i] = xengnttab_map_grant_ref
425 (gnt, domids[i], refs[i], ioreq->prot);
426 if (ioreq->page[i] == NULL) {
427 xen_be_printf(&ioreq->blkdev->xendev, 0,
428 "can't map grant ref %d (%s, %d maps)\n",
429 refs[i], strerror(errno), ioreq->blkdev->cnt_map);
430 ioreq->mapped = 1;
431 ioreq_unmap(ioreq);
432 return -1;
433 }
434 ioreq->blkdev->cnt_map++;
435 }
436 for (i = 0, j = 0; i < ioreq->v.niov; i++) {
437 if (page[i] == NULL) {
438 page[i] = ioreq->page[j++];
439 }
440 }
441 }
442 if (ioreq->blkdev->feature_persistent && new_maps != 0 &&
443 (!batch_maps || (ioreq->blkdev->persistent_gnt_count + new_maps <=
444 ioreq->blkdev->max_grants))) {
445
446
447
448
449
450 if (batch_maps) {
451 region = g_malloc0(sizeof(*region));
452 region->addr = ioreq->pages;
453 region->num = new_maps;
454 ioreq->blkdev->persistent_regions = g_slist_append(
455 ioreq->blkdev->persistent_regions,
456 region);
457 }
458 while ((ioreq->blkdev->persistent_gnt_count < ioreq->blkdev->max_grants)
459 && new_maps) {
460
461
462
463
464
465
466
467 grant = g_malloc0(sizeof(*grant));
468 new_maps--;
469 if (batch_maps) {
470 grant->page = ioreq->pages + (new_maps) * XC_PAGE_SIZE;
471 } else {
472 grant->page = ioreq->page[new_maps];
473 }
474 grant->blkdev = ioreq->blkdev;
475 xen_be_printf(&ioreq->blkdev->xendev, 3,
476 "adding grant %" PRIu32 " page: %p\n",
477 refs[new_maps], grant->page);
478 g_tree_insert(ioreq->blkdev->persistent_gnts,
479 GUINT_TO_POINTER(refs[new_maps]),
480 grant);
481 ioreq->blkdev->persistent_gnt_count++;
482 }
483 assert(!batch_maps || new_maps == 0);
484 }
485 for (i = 0; i < ioreq->v.niov; i++) {
486 ioreq->v.iov[i].iov_base += (uintptr_t)page[i];
487 }
488 ioreq->mapped = 1;
489 ioreq->num_unmap = new_maps;
490 return 0;
491}
492
493static int ioreq_runio_qemu_aio(struct ioreq *ioreq);
494
495static void qemu_aio_complete(void *opaque, int ret)
496{
497 struct ioreq *ioreq = opaque;
498
499 if (ret != 0) {
500 xen_be_printf(&ioreq->blkdev->xendev, 0, "%s I/O error\n",
501 ioreq->req.operation == BLKIF_OP_READ ? "read" : "write");
502 ioreq->aio_errors++;
503 }
504
505 ioreq->aio_inflight--;
506 if (ioreq->presync) {
507 ioreq->presync = 0;
508 ioreq_runio_qemu_aio(ioreq);
509 return;
510 }
511 if (ioreq->aio_inflight > 0) {
512 return;
513 }
514
515 ioreq->status = ioreq->aio_errors ? BLKIF_RSP_ERROR : BLKIF_RSP_OKAY;
516 ioreq_unmap(ioreq);
517 ioreq_finish(ioreq);
518 switch (ioreq->req.operation) {
519 case BLKIF_OP_WRITE:
520 case BLKIF_OP_FLUSH_DISKCACHE:
521 if (!ioreq->req.nr_segments) {
522 break;
523 }
524 case BLKIF_OP_READ:
525 if (ioreq->status == BLKIF_RSP_OKAY) {
526 block_acct_done(blk_get_stats(ioreq->blkdev->blk), &ioreq->acct);
527 } else {
528 block_acct_failed(blk_get_stats(ioreq->blkdev->blk), &ioreq->acct);
529 }
530 break;
531 case BLKIF_OP_DISCARD:
532 default:
533 break;
534 }
535 qemu_bh_schedule(ioreq->blkdev->bh);
536}
537
538static int ioreq_runio_qemu_aio(struct ioreq *ioreq)
539{
540 struct XenBlkDev *blkdev = ioreq->blkdev;
541
542 if (ioreq->req.nr_segments && ioreq_map(ioreq) == -1) {
543 goto err_no_map;
544 }
545
546 ioreq->aio_inflight++;
547 if (ioreq->presync) {
548 blk_aio_flush(ioreq->blkdev->blk, qemu_aio_complete, ioreq);
549 return 0;
550 }
551
552 switch (ioreq->req.operation) {
553 case BLKIF_OP_READ:
554 block_acct_start(blk_get_stats(blkdev->blk), &ioreq->acct,
555 ioreq->v.size, BLOCK_ACCT_READ);
556 ioreq->aio_inflight++;
557 blk_aio_preadv(blkdev->blk, ioreq->start, &ioreq->v, 0,
558 qemu_aio_complete, ioreq);
559 break;
560 case BLKIF_OP_WRITE:
561 case BLKIF_OP_FLUSH_DISKCACHE:
562 if (!ioreq->req.nr_segments) {
563 break;
564 }
565
566 block_acct_start(blk_get_stats(blkdev->blk), &ioreq->acct,
567 ioreq->v.size,
568 ioreq->req.operation == BLKIF_OP_WRITE ?
569 BLOCK_ACCT_WRITE : BLOCK_ACCT_FLUSH);
570 ioreq->aio_inflight++;
571 blk_aio_pwritev(blkdev->blk, ioreq->start, &ioreq->v, 0,
572 qemu_aio_complete, ioreq);
573 break;
574 case BLKIF_OP_DISCARD:
575 {
576 struct blkif_request_discard *discard_req = (void *)&ioreq->req;
577 ioreq->aio_inflight++;
578 blk_aio_discard(blkdev->blk,
579 discard_req->sector_number, discard_req->nr_sectors,
580 qemu_aio_complete, ioreq);
581 break;
582 }
583 default:
584
585 goto err;
586 }
587
588 qemu_aio_complete(ioreq, 0);
589
590 return 0;
591
592err:
593 ioreq_unmap(ioreq);
594err_no_map:
595 ioreq_finish(ioreq);
596 ioreq->status = BLKIF_RSP_ERROR;
597 return -1;
598}
599
600static int blk_send_response_one(struct ioreq *ioreq)
601{
602 struct XenBlkDev *blkdev = ioreq->blkdev;
603 int send_notify = 0;
604 int have_requests = 0;
605 blkif_response_t resp;
606 void *dst;
607
608 resp.id = ioreq->req.id;
609 resp.operation = ioreq->req.operation;
610 resp.status = ioreq->status;
611
612
613 switch (blkdev->protocol) {
614 case BLKIF_PROTOCOL_NATIVE:
615 dst = RING_GET_RESPONSE(&blkdev->rings.native, blkdev->rings.native.rsp_prod_pvt);
616 break;
617 case BLKIF_PROTOCOL_X86_32:
618 dst = RING_GET_RESPONSE(&blkdev->rings.x86_32_part,
619 blkdev->rings.x86_32_part.rsp_prod_pvt);
620 break;
621 case BLKIF_PROTOCOL_X86_64:
622 dst = RING_GET_RESPONSE(&blkdev->rings.x86_64_part,
623 blkdev->rings.x86_64_part.rsp_prod_pvt);
624 break;
625 default:
626 dst = NULL;
627 return 0;
628 }
629 memcpy(dst, &resp, sizeof(resp));
630 blkdev->rings.common.rsp_prod_pvt++;
631
632 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&blkdev->rings.common, send_notify);
633 if (blkdev->rings.common.rsp_prod_pvt == blkdev->rings.common.req_cons) {
634
635
636
637
638
639 RING_FINAL_CHECK_FOR_REQUESTS(&blkdev->rings.common, have_requests);
640 } else if (RING_HAS_UNCONSUMED_REQUESTS(&blkdev->rings.common)) {
641 have_requests = 1;
642 }
643
644 if (have_requests) {
645 blkdev->more_work++;
646 }
647 return send_notify;
648}
649
650
651static void blk_send_response_all(struct XenBlkDev *blkdev)
652{
653 struct ioreq *ioreq;
654 int send_notify = 0;
655
656 while (!QLIST_EMPTY(&blkdev->finished)) {
657 ioreq = QLIST_FIRST(&blkdev->finished);
658 send_notify += blk_send_response_one(ioreq);
659 ioreq_release(ioreq, true);
660 }
661 if (send_notify) {
662 xen_be_send_notify(&blkdev->xendev);
663 }
664}
665
666static int blk_get_request(struct XenBlkDev *blkdev, struct ioreq *ioreq, RING_IDX rc)
667{
668 switch (blkdev->protocol) {
669 case BLKIF_PROTOCOL_NATIVE:
670 memcpy(&ioreq->req, RING_GET_REQUEST(&blkdev->rings.native, rc),
671 sizeof(ioreq->req));
672 break;
673 case BLKIF_PROTOCOL_X86_32:
674 blkif_get_x86_32_req(&ioreq->req,
675 RING_GET_REQUEST(&blkdev->rings.x86_32_part, rc));
676 break;
677 case BLKIF_PROTOCOL_X86_64:
678 blkif_get_x86_64_req(&ioreq->req,
679 RING_GET_REQUEST(&blkdev->rings.x86_64_part, rc));
680 break;
681 }
682 return 0;
683}
684
685static void blk_handle_requests(struct XenBlkDev *blkdev)
686{
687 RING_IDX rc, rp;
688 struct ioreq *ioreq;
689
690 blkdev->more_work = 0;
691
692 rc = blkdev->rings.common.req_cons;
693 rp = blkdev->rings.common.sring->req_prod;
694 xen_rmb();
695
696 blk_send_response_all(blkdev);
697 while (rc != rp) {
698
699 if (RING_REQUEST_CONS_OVERFLOW(&blkdev->rings.common, rc)) {
700 break;
701 }
702 ioreq = ioreq_start(blkdev);
703 if (ioreq == NULL) {
704 blkdev->more_work++;
705 break;
706 }
707 blk_get_request(blkdev, ioreq, rc);
708 blkdev->rings.common.req_cons = ++rc;
709
710
711 if (ioreq_parse(ioreq) != 0) {
712
713 switch (ioreq->req.operation) {
714 case BLKIF_OP_READ:
715 block_acct_invalid(blk_get_stats(blkdev->blk),
716 BLOCK_ACCT_READ);
717 break;
718 case BLKIF_OP_WRITE:
719 block_acct_invalid(blk_get_stats(blkdev->blk),
720 BLOCK_ACCT_WRITE);
721 break;
722 case BLKIF_OP_FLUSH_DISKCACHE:
723 block_acct_invalid(blk_get_stats(blkdev->blk),
724 BLOCK_ACCT_FLUSH);
725 default:
726 break;
727 };
728
729 if (blk_send_response_one(ioreq)) {
730 xen_be_send_notify(&blkdev->xendev);
731 }
732 ioreq_release(ioreq, false);
733 continue;
734 }
735
736 ioreq_runio_qemu_aio(ioreq);
737 }
738
739 if (blkdev->more_work && blkdev->requests_inflight < max_requests) {
740 qemu_bh_schedule(blkdev->bh);
741 }
742}
743
744
745
746static void blk_bh(void *opaque)
747{
748 struct XenBlkDev *blkdev = opaque;
749 blk_handle_requests(blkdev);
750}
751
752
753
754
755
756
757
758
759#define MAX_GRANTS(max_req, max_seg) (2 * (max_req) * (max_seg))
760
761static void blk_alloc(struct XenDevice *xendev)
762{
763 struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
764
765 QLIST_INIT(&blkdev->inflight);
766 QLIST_INIT(&blkdev->finished);
767 QLIST_INIT(&blkdev->freelist);
768 blkdev->bh = qemu_bh_new(blk_bh, blkdev);
769 if (xen_mode != XEN_EMULATE) {
770 batch_maps = 1;
771 }
772 if (xengnttab_set_max_grants(xendev->gnttabdev,
773 MAX_GRANTS(max_requests, BLKIF_MAX_SEGMENTS_PER_REQUEST)) < 0) {
774 xen_be_printf(xendev, 0, "xengnttab_set_max_grants failed: %s\n",
775 strerror(errno));
776 }
777}
778
779static void blk_parse_discard(struct XenBlkDev *blkdev)
780{
781 int enable;
782
783 blkdev->feature_discard = true;
784
785 if (xenstore_read_be_int(&blkdev->xendev, "discard-enable", &enable) == 0) {
786 blkdev->feature_discard = !!enable;
787 }
788
789 if (blkdev->feature_discard) {
790 xenstore_write_be_int(&blkdev->xendev, "feature-discard", 1);
791 }
792}
793
794static int blk_init(struct XenDevice *xendev)
795{
796 struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
797 int info = 0;
798 char *directiosafe = NULL;
799
800
801 if (blkdev->params == NULL) {
802 char *h = NULL;
803 blkdev->params = xenstore_read_be_str(&blkdev->xendev, "params");
804 if (blkdev->params != NULL) {
805 h = strchr(blkdev->params, ':');
806 }
807 if (h != NULL) {
808 blkdev->fileproto = blkdev->params;
809 blkdev->filename = h+1;
810 *h = 0;
811 } else {
812 blkdev->fileproto = "<unset>";
813 blkdev->filename = blkdev->params;
814 }
815 }
816 if (!strcmp("aio", blkdev->fileproto)) {
817 blkdev->fileproto = "raw";
818 }
819 if (!strcmp("vhd", blkdev->fileproto)) {
820 blkdev->fileproto = "vpc";
821 }
822 if (blkdev->mode == NULL) {
823 blkdev->mode = xenstore_read_be_str(&blkdev->xendev, "mode");
824 }
825 if (blkdev->type == NULL) {
826 blkdev->type = xenstore_read_be_str(&blkdev->xendev, "type");
827 }
828 if (blkdev->dev == NULL) {
829 blkdev->dev = xenstore_read_be_str(&blkdev->xendev, "dev");
830 }
831 if (blkdev->devtype == NULL) {
832 blkdev->devtype = xenstore_read_be_str(&blkdev->xendev, "device-type");
833 }
834 directiosafe = xenstore_read_be_str(&blkdev->xendev, "direct-io-safe");
835 blkdev->directiosafe = (directiosafe && atoi(directiosafe));
836
837
838 if (blkdev->params == NULL ||
839 blkdev->mode == NULL ||
840 blkdev->type == NULL ||
841 blkdev->dev == NULL) {
842 goto out_error;
843 }
844
845
846 if (strcmp(blkdev->mode, "w")) {
847 info |= VDISK_READONLY;
848 }
849
850
851 if (blkdev->devtype && !strcmp(blkdev->devtype, "cdrom")) {
852 info |= VDISK_CDROM;
853 }
854
855 blkdev->file_blk = BLOCK_SIZE;
856
857
858
859
860 xenstore_write_be_int(&blkdev->xendev, "feature-flush-cache", 1);
861 xenstore_write_be_int(&blkdev->xendev, "feature-persistent", 1);
862 xenstore_write_be_int(&blkdev->xendev, "info", info);
863
864 blk_parse_discard(blkdev);
865
866 g_free(directiosafe);
867 return 0;
868
869out_error:
870 g_free(blkdev->params);
871 blkdev->params = NULL;
872 g_free(blkdev->mode);
873 blkdev->mode = NULL;
874 g_free(blkdev->type);
875 blkdev->type = NULL;
876 g_free(blkdev->dev);
877 blkdev->dev = NULL;
878 g_free(blkdev->devtype);
879 blkdev->devtype = NULL;
880 g_free(directiosafe);
881 blkdev->directiosafe = false;
882 return -1;
883}
884
885static int blk_connect(struct XenDevice *xendev)
886{
887 struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
888 int pers, index, qflags;
889 bool readonly = true;
890 bool writethrough = true;
891
892
893 if (blkdev->directiosafe) {
894 qflags = BDRV_O_NOCACHE | BDRV_O_NATIVE_AIO;
895 } else {
896 qflags = 0;
897 writethrough = false;
898 }
899 if (strcmp(blkdev->mode, "w") == 0) {
900 qflags |= BDRV_O_RDWR;
901 readonly = false;
902 }
903 if (blkdev->feature_discard) {
904 qflags |= BDRV_O_UNMAP;
905 }
906
907
908 index = (blkdev->xendev.dev - 202 * 256) / 16;
909 blkdev->dinfo = drive_get(IF_XEN, 0, index);
910 if (!blkdev->dinfo) {
911 Error *local_err = NULL;
912 QDict *options = NULL;
913
914 if (strcmp(blkdev->fileproto, "<unset>")) {
915 options = qdict_new();
916 qdict_put(options, "driver", qstring_from_str(blkdev->fileproto));
917 }
918
919
920 xen_be_printf(&blkdev->xendev, 2, "create new bdrv (xenbus setup)\n");
921 blkdev->blk = blk_new_open(blkdev->filename, NULL, options,
922 qflags, &local_err);
923 if (!blkdev->blk) {
924 xen_be_printf(&blkdev->xendev, 0, "error: %s\n",
925 error_get_pretty(local_err));
926 error_free(local_err);
927 return -1;
928 }
929 blk_set_enable_write_cache(blkdev->blk, !writethrough);
930 } else {
931
932 xen_be_printf(&blkdev->xendev, 2, "get configured bdrv (cmdline setup)\n");
933 blkdev->blk = blk_by_legacy_dinfo(blkdev->dinfo);
934 if (blk_is_read_only(blkdev->blk) && !readonly) {
935 xen_be_printf(&blkdev->xendev, 0, "Unexpected read-only drive");
936 blkdev->blk = NULL;
937 return -1;
938 }
939
940
941 blk_ref(blkdev->blk);
942 }
943 blk_attach_dev_nofail(blkdev->blk, blkdev);
944 blkdev->file_size = blk_getlength(blkdev->blk);
945 if (blkdev->file_size < 0) {
946 BlockDriverState *bs = blk_bs(blkdev->blk);
947 const char *drv_name = bs ? bdrv_get_format_name(bs) : NULL;
948 xen_be_printf(&blkdev->xendev, 1, "blk_getlength: %d (%s) | drv %s\n",
949 (int)blkdev->file_size, strerror(-blkdev->file_size),
950 drv_name ?: "-");
951 blkdev->file_size = 0;
952 }
953
954 xen_be_printf(xendev, 1, "type \"%s\", fileproto \"%s\", filename \"%s\","
955 " size %" PRId64 " (%" PRId64 " MB)\n",
956 blkdev->type, blkdev->fileproto, blkdev->filename,
957 blkdev->file_size, blkdev->file_size >> 20);
958
959
960 xenstore_write_be_int(&blkdev->xendev, "sector-size", blkdev->file_blk);
961 xenstore_write_be_int64(&blkdev->xendev, "sectors",
962 blkdev->file_size / blkdev->file_blk);
963
964 if (xenstore_read_fe_int(&blkdev->xendev, "ring-ref", &blkdev->ring_ref) == -1) {
965 return -1;
966 }
967 if (xenstore_read_fe_int(&blkdev->xendev, "event-channel",
968 &blkdev->xendev.remote_port) == -1) {
969 return -1;
970 }
971 if (xenstore_read_fe_int(&blkdev->xendev, "feature-persistent", &pers)) {
972 blkdev->feature_persistent = FALSE;
973 } else {
974 blkdev->feature_persistent = !!pers;
975 }
976
977 blkdev->protocol = BLKIF_PROTOCOL_NATIVE;
978 if (blkdev->xendev.protocol) {
979 if (strcmp(blkdev->xendev.protocol, XEN_IO_PROTO_ABI_X86_32) == 0) {
980 blkdev->protocol = BLKIF_PROTOCOL_X86_32;
981 }
982 if (strcmp(blkdev->xendev.protocol, XEN_IO_PROTO_ABI_X86_64) == 0) {
983 blkdev->protocol = BLKIF_PROTOCOL_X86_64;
984 }
985 }
986
987 blkdev->sring = xengnttab_map_grant_ref(blkdev->xendev.gnttabdev,
988 blkdev->xendev.dom,
989 blkdev->ring_ref,
990 PROT_READ | PROT_WRITE);
991 if (!blkdev->sring) {
992 return -1;
993 }
994 blkdev->cnt_map++;
995
996 switch (blkdev->protocol) {
997 case BLKIF_PROTOCOL_NATIVE:
998 {
999 blkif_sring_t *sring_native = blkdev->sring;
1000 BACK_RING_INIT(&blkdev->rings.native, sring_native, XC_PAGE_SIZE);
1001 break;
1002 }
1003 case BLKIF_PROTOCOL_X86_32:
1004 {
1005 blkif_x86_32_sring_t *sring_x86_32 = blkdev->sring;
1006
1007 BACK_RING_INIT(&blkdev->rings.x86_32_part, sring_x86_32, XC_PAGE_SIZE);
1008 break;
1009 }
1010 case BLKIF_PROTOCOL_X86_64:
1011 {
1012 blkif_x86_64_sring_t *sring_x86_64 = blkdev->sring;
1013
1014 BACK_RING_INIT(&blkdev->rings.x86_64_part, sring_x86_64, XC_PAGE_SIZE);
1015 break;
1016 }
1017 }
1018
1019 if (blkdev->feature_persistent) {
1020
1021 blkdev->max_grants = max_requests * BLKIF_MAX_SEGMENTS_PER_REQUEST;
1022 blkdev->persistent_gnts = g_tree_new_full((GCompareDataFunc)int_cmp,
1023 NULL, NULL,
1024 batch_maps ?
1025 (GDestroyNotify)g_free :
1026 (GDestroyNotify)destroy_grant);
1027 blkdev->persistent_regions = NULL;
1028 blkdev->persistent_gnt_count = 0;
1029 }
1030
1031 xen_be_bind_evtchn(&blkdev->xendev);
1032
1033 xen_be_printf(&blkdev->xendev, 1, "ok: proto %s, ring-ref %d, "
1034 "remote port %d, local port %d\n",
1035 blkdev->xendev.protocol, blkdev->ring_ref,
1036 blkdev->xendev.remote_port, blkdev->xendev.local_port);
1037 return 0;
1038}
1039
1040static void blk_disconnect(struct XenDevice *xendev)
1041{
1042 struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
1043
1044 if (blkdev->blk) {
1045 blk_detach_dev(blkdev->blk, blkdev);
1046 blk_unref(blkdev->blk);
1047 blkdev->blk = NULL;
1048 }
1049 xen_be_unbind_evtchn(&blkdev->xendev);
1050
1051 if (blkdev->sring) {
1052 xengnttab_unmap(blkdev->xendev.gnttabdev, blkdev->sring, 1);
1053 blkdev->cnt_map--;
1054 blkdev->sring = NULL;
1055 }
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065 if (blkdev->feature_persistent) {
1066 g_tree_destroy(blkdev->persistent_gnts);
1067 assert(batch_maps || blkdev->persistent_gnt_count == 0);
1068 if (batch_maps) {
1069 blkdev->persistent_gnt_count = 0;
1070 g_slist_foreach(blkdev->persistent_regions,
1071 (GFunc)remove_persistent_region, blkdev);
1072 g_slist_free(blkdev->persistent_regions);
1073 }
1074 blkdev->feature_persistent = false;
1075 }
1076}
1077
1078static int blk_free(struct XenDevice *xendev)
1079{
1080 struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
1081 struct ioreq *ioreq;
1082
1083 if (blkdev->blk || blkdev->sring) {
1084 blk_disconnect(xendev);
1085 }
1086
1087 while (!QLIST_EMPTY(&blkdev->freelist)) {
1088 ioreq = QLIST_FIRST(&blkdev->freelist);
1089 QLIST_REMOVE(ioreq, list);
1090 qemu_iovec_destroy(&ioreq->v);
1091 g_free(ioreq);
1092 }
1093
1094 g_free(blkdev->params);
1095 g_free(blkdev->mode);
1096 g_free(blkdev->type);
1097 g_free(blkdev->dev);
1098 g_free(blkdev->devtype);
1099 qemu_bh_delete(blkdev->bh);
1100 return 0;
1101}
1102
1103static void blk_event(struct XenDevice *xendev)
1104{
1105 struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
1106
1107 qemu_bh_schedule(blkdev->bh);
1108}
1109
1110struct XenDevOps xen_blkdev_ops = {
1111 .size = sizeof(struct XenBlkDev),
1112 .flags = DEVOPS_FLAG_NEED_GNTDEV,
1113 .alloc = blk_alloc,
1114 .init = blk_init,
1115 .initialise = blk_connect,
1116 .disconnect = blk_disconnect,
1117 .event = blk_event,
1118 .free = blk_free,
1119};
1120