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/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
37static int batch_maps = 0;
38
39static int max_requests = 32;
40
41
42
43#define BLOCK_SIZE 512
44#define IOCB_COUNT (BLKIF_MAX_SEGMENTS_PER_REQUEST + 2)
45
46struct PersistentGrant {
47 void *page;
48 struct XenBlkDev *blkdev;
49};
50
51typedef struct PersistentGrant PersistentGrant;
52
53struct PersistentRegion {
54 void *addr;
55 int num;
56};
57
58typedef struct PersistentRegion PersistentRegion;
59
60struct ioreq {
61 blkif_request_t req;
62 int16_t status;
63
64
65 off_t start;
66 QEMUIOVector v;
67 int presync;
68 uint8_t mapped;
69
70
71 uint32_t domids[BLKIF_MAX_SEGMENTS_PER_REQUEST];
72 uint32_t refs[BLKIF_MAX_SEGMENTS_PER_REQUEST];
73 int prot;
74 void *page[BLKIF_MAX_SEGMENTS_PER_REQUEST];
75 void *pages;
76 int num_unmap;
77
78
79 int aio_inflight;
80 int aio_errors;
81
82 struct XenBlkDev *blkdev;
83 QLIST_ENTRY(ioreq) list;
84 BlockAcctCookie acct;
85};
86
87struct XenBlkDev {
88 struct XenDevice xendev;
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 int ring_ref;
98 void *sring;
99 int64_t file_blk;
100 int64_t file_size;
101 int protocol;
102 blkif_back_rings_t rings;
103 int more_work;
104 int cnt_map;
105
106
107 QLIST_HEAD(inflight_head, ioreq) inflight;
108 QLIST_HEAD(finished_head, ioreq) finished;
109 QLIST_HEAD(freelist_head, ioreq) freelist;
110 int requests_total;
111 int requests_inflight;
112 int requests_finished;
113
114
115 gboolean feature_discard;
116 gboolean feature_persistent;
117 GTree *persistent_gnts;
118 GSList *persistent_regions;
119 unsigned int persistent_gnt_count;
120 unsigned int max_grants;
121
122
123 DriveInfo *dinfo;
124 BlockBackend *blk;
125 QEMUBH *bh;
126};
127
128
129
130static void ioreq_reset(struct ioreq *ioreq)
131{
132 memset(&ioreq->req, 0, sizeof(ioreq->req));
133 ioreq->status = 0;
134 ioreq->start = 0;
135 ioreq->presync = 0;
136 ioreq->mapped = 0;
137
138 memset(ioreq->domids, 0, sizeof(ioreq->domids));
139 memset(ioreq->refs, 0, sizeof(ioreq->refs));
140 ioreq->prot = 0;
141 memset(ioreq->page, 0, sizeof(ioreq->page));
142 ioreq->pages = NULL;
143
144 ioreq->aio_inflight = 0;
145 ioreq->aio_errors = 0;
146
147 ioreq->blkdev = NULL;
148 memset(&ioreq->list, 0, sizeof(ioreq->list));
149 memset(&ioreq->acct, 0, sizeof(ioreq->acct));
150
151 qemu_iovec_reset(&ioreq->v);
152}
153
154static gint int_cmp(gconstpointer a, gconstpointer b, gpointer user_data)
155{
156 uint ua = GPOINTER_TO_UINT(a);
157 uint ub = GPOINTER_TO_UINT(b);
158 return (ua > ub) - (ua < ub);
159}
160
161static void destroy_grant(gpointer pgnt)
162{
163 PersistentGrant *grant = pgnt;
164 xengnttab_handle *gnt = grant->blkdev->xendev.gnttabdev;
165
166 if (xengnttab_unmap(gnt, grant->page, 1) != 0) {
167 xen_be_printf(&grant->blkdev->xendev, 0,
168 "xengnttab_unmap failed: %s\n",
169 strerror(errno));
170 }
171 grant->blkdev->persistent_gnt_count--;
172 xen_be_printf(&grant->blkdev->xendev, 3,
173 "unmapped grant %p\n", grant->page);
174 g_free(grant);
175}
176
177static void remove_persistent_region(gpointer data, gpointer dev)
178{
179 PersistentRegion *region = data;
180 struct XenBlkDev *blkdev = dev;
181 xengnttab_handle *gnt = blkdev->xendev.gnttabdev;
182
183 if (xengnttab_unmap(gnt, region->addr, region->num) != 0) {
184 xen_be_printf(&blkdev->xendev, 0,
185 "xengnttab_unmap region %p failed: %s\n",
186 region->addr, strerror(errno));
187 }
188 xen_be_printf(&blkdev->xendev, 3,
189 "unmapped grant region %p with %d pages\n",
190 region->addr, region->num);
191 g_free(region);
192}
193
194static struct ioreq *ioreq_start(struct XenBlkDev *blkdev)
195{
196 struct ioreq *ioreq = NULL;
197
198 if (QLIST_EMPTY(&blkdev->freelist)) {
199 if (blkdev->requests_total >= max_requests) {
200 goto out;
201 }
202
203 ioreq = g_malloc0(sizeof(*ioreq));
204 ioreq->blkdev = blkdev;
205 blkdev->requests_total++;
206 qemu_iovec_init(&ioreq->v, BLKIF_MAX_SEGMENTS_PER_REQUEST);
207 } else {
208
209 ioreq = QLIST_FIRST(&blkdev->freelist);
210 QLIST_REMOVE(ioreq, list);
211 }
212 QLIST_INSERT_HEAD(&blkdev->inflight, ioreq, list);
213 blkdev->requests_inflight++;
214
215out:
216 return ioreq;
217}
218
219static void ioreq_finish(struct ioreq *ioreq)
220{
221 struct XenBlkDev *blkdev = ioreq->blkdev;
222
223 QLIST_REMOVE(ioreq, list);
224 QLIST_INSERT_HEAD(&blkdev->finished, ioreq, list);
225 blkdev->requests_inflight--;
226 blkdev->requests_finished++;
227}
228
229static void ioreq_release(struct ioreq *ioreq, bool finish)
230{
231 struct XenBlkDev *blkdev = ioreq->blkdev;
232
233 QLIST_REMOVE(ioreq, list);
234 ioreq_reset(ioreq);
235 ioreq->blkdev = blkdev;
236 QLIST_INSERT_HEAD(&blkdev->freelist, ioreq, list);
237 if (finish) {
238 blkdev->requests_finished--;
239 } else {
240 blkdev->requests_inflight--;
241 }
242}
243
244
245
246
247
248static int ioreq_parse(struct ioreq *ioreq)
249{
250 struct XenBlkDev *blkdev = ioreq->blkdev;
251 uintptr_t mem;
252 size_t len;
253 int i;
254
255 xen_be_printf(&blkdev->xendev, 3,
256 "op %d, nr %d, handle %d, id %" PRId64 ", sector %" PRId64 "\n",
257 ioreq->req.operation, ioreq->req.nr_segments,
258 ioreq->req.handle, ioreq->req.id, ioreq->req.sector_number);
259 switch (ioreq->req.operation) {
260 case BLKIF_OP_READ:
261 ioreq->prot = PROT_WRITE;
262 break;
263 case BLKIF_OP_FLUSH_DISKCACHE:
264 ioreq->presync = 1;
265 if (!ioreq->req.nr_segments) {
266 return 0;
267 }
268
269 case BLKIF_OP_WRITE:
270 ioreq->prot = PROT_READ;
271 break;
272 case BLKIF_OP_DISCARD:
273 return 0;
274 default:
275 xen_be_printf(&blkdev->xendev, 0, "error: unknown operation (%d)\n",
276 ioreq->req.operation);
277 goto err;
278 };
279
280 if (ioreq->req.operation != BLKIF_OP_READ && blkdev->mode[0] != 'w') {
281 xen_be_printf(&blkdev->xendev, 0, "error: write req for ro device\n");
282 goto err;
283 }
284
285 ioreq->start = ioreq->req.sector_number * blkdev->file_blk;
286 for (i = 0; i < ioreq->req.nr_segments; i++) {
287 if (i == BLKIF_MAX_SEGMENTS_PER_REQUEST) {
288 xen_be_printf(&blkdev->xendev, 0, "error: nr_segments too big\n");
289 goto err;
290 }
291 if (ioreq->req.seg[i].first_sect > ioreq->req.seg[i].last_sect) {
292 xen_be_printf(&blkdev->xendev, 0, "error: first > last sector\n");
293 goto err;
294 }
295 if (ioreq->req.seg[i].last_sect * BLOCK_SIZE >= XC_PAGE_SIZE) {
296 xen_be_printf(&blkdev->xendev, 0, "error: page crossing\n");
297 goto err;
298 }
299
300 ioreq->domids[i] = blkdev->xendev.dom;
301 ioreq->refs[i] = ioreq->req.seg[i].gref;
302
303 mem = ioreq->req.seg[i].first_sect * blkdev->file_blk;
304 len = (ioreq->req.seg[i].last_sect - ioreq->req.seg[i].first_sect + 1) * blkdev->file_blk;
305 qemu_iovec_add(&ioreq->v, (void*)mem, len);
306 }
307 if (ioreq->start + ioreq->v.size > blkdev->file_size) {
308 xen_be_printf(&blkdev->xendev, 0, "error: access beyond end of file\n");
309 goto err;
310 }
311 return 0;
312
313err:
314 ioreq->status = BLKIF_RSP_ERROR;
315 return -1;
316}
317
318static void ioreq_unmap(struct ioreq *ioreq)
319{
320 xengnttab_handle *gnt = ioreq->blkdev->xendev.gnttabdev;
321 int i;
322
323 if (ioreq->num_unmap == 0 || ioreq->mapped == 0) {
324 return;
325 }
326 if (batch_maps) {
327 if (!ioreq->pages) {
328 return;
329 }
330 if (xengnttab_unmap(gnt, ioreq->pages, ioreq->num_unmap) != 0) {
331 xen_be_printf(&ioreq->blkdev->xendev, 0,
332 "xengnttab_unmap failed: %s\n",
333 strerror(errno));
334 }
335 ioreq->blkdev->cnt_map -= ioreq->num_unmap;
336 ioreq->pages = NULL;
337 } else {
338 for (i = 0; i < ioreq->num_unmap; i++) {
339 if (!ioreq->page[i]) {
340 continue;
341 }
342 if (xengnttab_unmap(gnt, ioreq->page[i], 1) != 0) {
343 xen_be_printf(&ioreq->blkdev->xendev, 0,
344 "xengnttab_unmap failed: %s\n",
345 strerror(errno));
346 }
347 ioreq->blkdev->cnt_map--;
348 ioreq->page[i] = NULL;
349 }
350 }
351 ioreq->mapped = 0;
352}
353
354static int ioreq_map(struct ioreq *ioreq)
355{
356 xengnttab_handle *gnt = ioreq->blkdev->xendev.gnttabdev;
357 uint32_t domids[BLKIF_MAX_SEGMENTS_PER_REQUEST];
358 uint32_t refs[BLKIF_MAX_SEGMENTS_PER_REQUEST];
359 void *page[BLKIF_MAX_SEGMENTS_PER_REQUEST];
360 int i, j, new_maps = 0;
361 PersistentGrant *grant;
362 PersistentRegion *region;
363
364
365
366
367
368
369
370
371 if (ioreq->v.niov == 0 || ioreq->mapped == 1) {
372 return 0;
373 }
374 if (ioreq->blkdev->feature_persistent) {
375 for (i = 0; i < ioreq->v.niov; i++) {
376 grant = g_tree_lookup(ioreq->blkdev->persistent_gnts,
377 GUINT_TO_POINTER(ioreq->refs[i]));
378
379 if (grant != NULL) {
380 page[i] = grant->page;
381 xen_be_printf(&ioreq->blkdev->xendev, 3,
382 "using persistent-grant %" PRIu32 "\n",
383 ioreq->refs[i]);
384 } else {
385
386
387
388 domids[new_maps] = ioreq->domids[i];
389 refs[new_maps] = ioreq->refs[i];
390 page[i] = NULL;
391 new_maps++;
392 }
393 }
394
395
396
397 ioreq->prot = PROT_WRITE | PROT_READ;
398 } else {
399
400 memcpy(refs, ioreq->refs, sizeof(refs));
401 memcpy(domids, ioreq->domids, sizeof(domids));
402 memset(page, 0, sizeof(page));
403 new_maps = ioreq->v.niov;
404 }
405
406 if (batch_maps && new_maps) {
407 ioreq->pages = xengnttab_map_grant_refs
408 (gnt, new_maps, domids, refs, ioreq->prot);
409 if (ioreq->pages == NULL) {
410 xen_be_printf(&ioreq->blkdev->xendev, 0,
411 "can't map %d grant refs (%s, %d maps)\n",
412 new_maps, strerror(errno), ioreq->blkdev->cnt_map);
413 return -1;
414 }
415 for (i = 0, j = 0; i < ioreq->v.niov; i++) {
416 if (page[i] == NULL) {
417 page[i] = ioreq->pages + (j++) * XC_PAGE_SIZE;
418 }
419 }
420 ioreq->blkdev->cnt_map += new_maps;
421 } else if (new_maps) {
422 for (i = 0; i < new_maps; i++) {
423 ioreq->page[i] = xengnttab_map_grant_ref
424 (gnt, domids[i], refs[i], ioreq->prot);
425 if (ioreq->page[i] == NULL) {
426 xen_be_printf(&ioreq->blkdev->xendev, 0,
427 "can't map grant ref %d (%s, %d maps)\n",
428 refs[i], strerror(errno), ioreq->blkdev->cnt_map);
429 ioreq->mapped = 1;
430 ioreq_unmap(ioreq);
431 return -1;
432 }
433 ioreq->blkdev->cnt_map++;
434 }
435 for (i = 0, j = 0; i < ioreq->v.niov; i++) {
436 if (page[i] == NULL) {
437 page[i] = ioreq->page[j++];
438 }
439 }
440 }
441 if (ioreq->blkdev->feature_persistent && new_maps != 0 &&
442 (!batch_maps || (ioreq->blkdev->persistent_gnt_count + new_maps <=
443 ioreq->blkdev->max_grants))) {
444
445
446
447
448
449 if (batch_maps) {
450 region = g_malloc0(sizeof(*region));
451 region->addr = ioreq->pages;
452 region->num = new_maps;
453 ioreq->blkdev->persistent_regions = g_slist_append(
454 ioreq->blkdev->persistent_regions,
455 region);
456 }
457 while ((ioreq->blkdev->persistent_gnt_count < ioreq->blkdev->max_grants)
458 && new_maps) {
459
460
461
462
463
464
465
466 grant = g_malloc0(sizeof(*grant));
467 new_maps--;
468 if (batch_maps) {
469 grant->page = ioreq->pages + (new_maps) * XC_PAGE_SIZE;
470 } else {
471 grant->page = ioreq->page[new_maps];
472 }
473 grant->blkdev = ioreq->blkdev;
474 xen_be_printf(&ioreq->blkdev->xendev, 3,
475 "adding grant %" PRIu32 " page: %p\n",
476 refs[new_maps], grant->page);
477 g_tree_insert(ioreq->blkdev->persistent_gnts,
478 GUINT_TO_POINTER(refs[new_maps]),
479 grant);
480 ioreq->blkdev->persistent_gnt_count++;
481 }
482 assert(!batch_maps || new_maps == 0);
483 }
484 for (i = 0; i < ioreq->v.niov; i++) {
485 ioreq->v.iov[i].iov_base += (uintptr_t)page[i];
486 }
487 ioreq->mapped = 1;
488 ioreq->num_unmap = new_maps;
489 return 0;
490}
491
492static int ioreq_runio_qemu_aio(struct ioreq *ioreq);
493
494static void qemu_aio_complete(void *opaque, int ret)
495{
496 struct ioreq *ioreq = opaque;
497
498 if (ret != 0) {
499 xen_be_printf(&ioreq->blkdev->xendev, 0, "%s I/O error\n",
500 ioreq->req.operation == BLKIF_OP_READ ? "read" : "write");
501 ioreq->aio_errors++;
502 }
503
504 ioreq->aio_inflight--;
505 if (ioreq->presync) {
506 ioreq->presync = 0;
507 ioreq_runio_qemu_aio(ioreq);
508 return;
509 }
510 if (ioreq->aio_inflight > 0) {
511 return;
512 }
513
514 ioreq->status = ioreq->aio_errors ? BLKIF_RSP_ERROR : BLKIF_RSP_OKAY;
515 ioreq_unmap(ioreq);
516 ioreq_finish(ioreq);
517 switch (ioreq->req.operation) {
518 case BLKIF_OP_WRITE:
519 case BLKIF_OP_FLUSH_DISKCACHE:
520 if (!ioreq->req.nr_segments) {
521 break;
522 }
523 case BLKIF_OP_READ:
524 if (ioreq->status == BLKIF_RSP_OKAY) {
525 block_acct_done(blk_get_stats(ioreq->blkdev->blk), &ioreq->acct);
526 } else {
527 block_acct_failed(blk_get_stats(ioreq->blkdev->blk), &ioreq->acct);
528 }
529 break;
530 case BLKIF_OP_DISCARD:
531 default:
532 break;
533 }
534 qemu_bh_schedule(ioreq->blkdev->bh);
535}
536
537static int ioreq_runio_qemu_aio(struct ioreq *ioreq)
538{
539 struct XenBlkDev *blkdev = ioreq->blkdev;
540
541 if (ioreq->req.nr_segments && ioreq_map(ioreq) == -1) {
542 goto err_no_map;
543 }
544
545 ioreq->aio_inflight++;
546 if (ioreq->presync) {
547 blk_aio_flush(ioreq->blkdev->blk, qemu_aio_complete, ioreq);
548 return 0;
549 }
550
551 switch (ioreq->req.operation) {
552 case BLKIF_OP_READ:
553 block_acct_start(blk_get_stats(blkdev->blk), &ioreq->acct,
554 ioreq->v.size, BLOCK_ACCT_READ);
555 ioreq->aio_inflight++;
556 blk_aio_preadv(blkdev->blk, ioreq->start, &ioreq->v, 0,
557 qemu_aio_complete, ioreq);
558 break;
559 case BLKIF_OP_WRITE:
560 case BLKIF_OP_FLUSH_DISKCACHE:
561 if (!ioreq->req.nr_segments) {
562 break;
563 }
564
565 block_acct_start(blk_get_stats(blkdev->blk), &ioreq->acct,
566 ioreq->v.size,
567 ioreq->req.operation == BLKIF_OP_WRITE ?
568 BLOCK_ACCT_WRITE : BLOCK_ACCT_FLUSH);
569 ioreq->aio_inflight++;
570 blk_aio_pwritev(blkdev->blk, ioreq->start, &ioreq->v, 0,
571 qemu_aio_complete, ioreq);
572 break;
573 case BLKIF_OP_DISCARD:
574 {
575 struct blkif_request_discard *discard_req = (void *)&ioreq->req;
576 ioreq->aio_inflight++;
577 blk_aio_pdiscard(blkdev->blk,
578 discard_req->sector_number << BDRV_SECTOR_BITS,
579 discard_req->nr_sectors << BDRV_SECTOR_BITS,
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
683 barrier();
684 return 0;
685}
686
687static void blk_handle_requests(struct XenBlkDev *blkdev)
688{
689 RING_IDX rc, rp;
690 struct ioreq *ioreq;
691
692 blkdev->more_work = 0;
693
694 rc = blkdev->rings.common.req_cons;
695 rp = blkdev->rings.common.sring->req_prod;
696 xen_rmb();
697
698 blk_send_response_all(blkdev);
699 while (rc != rp) {
700
701 if (RING_REQUEST_CONS_OVERFLOW(&blkdev->rings.common, rc)) {
702 break;
703 }
704 ioreq = ioreq_start(blkdev);
705 if (ioreq == NULL) {
706 blkdev->more_work++;
707 break;
708 }
709 blk_get_request(blkdev, ioreq, rc);
710 blkdev->rings.common.req_cons = ++rc;
711
712
713 if (ioreq_parse(ioreq) != 0) {
714
715 switch (ioreq->req.operation) {
716 case BLKIF_OP_READ:
717 block_acct_invalid(blk_get_stats(blkdev->blk),
718 BLOCK_ACCT_READ);
719 break;
720 case BLKIF_OP_WRITE:
721 block_acct_invalid(blk_get_stats(blkdev->blk),
722 BLOCK_ACCT_WRITE);
723 break;
724 case BLKIF_OP_FLUSH_DISKCACHE:
725 block_acct_invalid(blk_get_stats(blkdev->blk),
726 BLOCK_ACCT_FLUSH);
727 default:
728 break;
729 };
730
731 if (blk_send_response_one(ioreq)) {
732 xen_be_send_notify(&blkdev->xendev);
733 }
734 ioreq_release(ioreq, false);
735 continue;
736 }
737
738 ioreq_runio_qemu_aio(ioreq);
739 }
740
741 if (blkdev->more_work && blkdev->requests_inflight < max_requests) {
742 qemu_bh_schedule(blkdev->bh);
743 }
744}
745
746
747
748static void blk_bh(void *opaque)
749{
750 struct XenBlkDev *blkdev = opaque;
751 blk_handle_requests(blkdev);
752}
753
754
755
756
757
758
759
760
761#define MAX_GRANTS(max_req, max_seg) (2 * (max_req) * (max_seg))
762
763static void blk_alloc(struct XenDevice *xendev)
764{
765 struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
766
767 QLIST_INIT(&blkdev->inflight);
768 QLIST_INIT(&blkdev->finished);
769 QLIST_INIT(&blkdev->freelist);
770 blkdev->bh = qemu_bh_new(blk_bh, blkdev);
771 if (xen_mode != XEN_EMULATE) {
772 batch_maps = 1;
773 }
774 if (xengnttab_set_max_grants(xendev->gnttabdev,
775 MAX_GRANTS(max_requests, BLKIF_MAX_SEGMENTS_PER_REQUEST)) < 0) {
776 xen_be_printf(xendev, 0, "xengnttab_set_max_grants failed: %s\n",
777 strerror(errno));
778 }
779}
780
781static void blk_parse_discard(struct XenBlkDev *blkdev)
782{
783 int enable;
784
785 blkdev->feature_discard = true;
786
787 if (xenstore_read_be_int(&blkdev->xendev, "discard-enable", &enable) == 0) {
788 blkdev->feature_discard = !!enable;
789 }
790
791 if (blkdev->feature_discard) {
792 xenstore_write_be_int(&blkdev->xendev, "feature-discard", 1);
793 }
794}
795
796static int blk_init(struct XenDevice *xendev)
797{
798 struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
799 int info = 0;
800 char *directiosafe = NULL;
801
802
803 if (blkdev->params == NULL) {
804 char *h = NULL;
805 blkdev->params = xenstore_read_be_str(&blkdev->xendev, "params");
806 if (blkdev->params != NULL) {
807 h = strchr(blkdev->params, ':');
808 }
809 if (h != NULL) {
810 blkdev->fileproto = blkdev->params;
811 blkdev->filename = h+1;
812 *h = 0;
813 } else {
814 blkdev->fileproto = "<unset>";
815 blkdev->filename = blkdev->params;
816 }
817 }
818 if (!strcmp("aio", blkdev->fileproto)) {
819 blkdev->fileproto = "raw";
820 }
821 if (!strcmp("vhd", blkdev->fileproto)) {
822 blkdev->fileproto = "vpc";
823 }
824 if (blkdev->mode == NULL) {
825 blkdev->mode = xenstore_read_be_str(&blkdev->xendev, "mode");
826 }
827 if (blkdev->type == NULL) {
828 blkdev->type = xenstore_read_be_str(&blkdev->xendev, "type");
829 }
830 if (blkdev->dev == NULL) {
831 blkdev->dev = xenstore_read_be_str(&blkdev->xendev, "dev");
832 }
833 if (blkdev->devtype == NULL) {
834 blkdev->devtype = xenstore_read_be_str(&blkdev->xendev, "device-type");
835 }
836 directiosafe = xenstore_read_be_str(&blkdev->xendev, "direct-io-safe");
837 blkdev->directiosafe = (directiosafe && atoi(directiosafe));
838
839
840 if (blkdev->params == NULL ||
841 blkdev->mode == NULL ||
842 blkdev->type == NULL ||
843 blkdev->dev == NULL) {
844 goto out_error;
845 }
846
847
848 if (strcmp(blkdev->mode, "w")) {
849 info |= VDISK_READONLY;
850 }
851
852
853 if (blkdev->devtype && !strcmp(blkdev->devtype, "cdrom")) {
854 info |= VDISK_CDROM;
855 }
856
857 blkdev->file_blk = BLOCK_SIZE;
858
859
860
861
862 xenstore_write_be_int(&blkdev->xendev, "feature-flush-cache", 1);
863 xenstore_write_be_int(&blkdev->xendev, "feature-persistent", 1);
864 xenstore_write_be_int(&blkdev->xendev, "info", info);
865
866 blk_parse_discard(blkdev);
867
868 g_free(directiosafe);
869 return 0;
870
871out_error:
872 g_free(blkdev->params);
873 blkdev->params = NULL;
874 g_free(blkdev->mode);
875 blkdev->mode = NULL;
876 g_free(blkdev->type);
877 blkdev->type = NULL;
878 g_free(blkdev->dev);
879 blkdev->dev = NULL;
880 g_free(blkdev->devtype);
881 blkdev->devtype = NULL;
882 g_free(directiosafe);
883 blkdev->directiosafe = false;
884 return -1;
885}
886
887static int blk_connect(struct XenDevice *xendev)
888{
889 struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
890 int pers, index, qflags;
891 bool readonly = true;
892 bool writethrough = true;
893
894
895 if (blkdev->directiosafe) {
896 qflags = BDRV_O_NOCACHE | BDRV_O_NATIVE_AIO;
897 } else {
898 qflags = 0;
899 writethrough = false;
900 }
901 if (strcmp(blkdev->mode, "w") == 0) {
902 qflags |= BDRV_O_RDWR;
903 readonly = false;
904 }
905 if (blkdev->feature_discard) {
906 qflags |= BDRV_O_UNMAP;
907 }
908
909
910 index = (blkdev->xendev.dev - 202 * 256) / 16;
911 blkdev->dinfo = drive_get(IF_XEN, 0, index);
912 if (!blkdev->dinfo) {
913 Error *local_err = NULL;
914 QDict *options = NULL;
915
916 if (strcmp(blkdev->fileproto, "<unset>")) {
917 options = qdict_new();
918 qdict_put(options, "driver", qstring_from_str(blkdev->fileproto));
919 }
920
921
922 xen_be_printf(&blkdev->xendev, 2, "create new bdrv (xenbus setup)\n");
923 blkdev->blk = blk_new_open(blkdev->filename, NULL, options,
924 qflags, &local_err);
925 if (!blkdev->blk) {
926 xen_be_printf(&blkdev->xendev, 0, "error: %s\n",
927 error_get_pretty(local_err));
928 error_free(local_err);
929 return -1;
930 }
931 blk_set_enable_write_cache(blkdev->blk, !writethrough);
932 } else {
933
934 xen_be_printf(&blkdev->xendev, 2, "get configured bdrv (cmdline setup)\n");
935 blkdev->blk = blk_by_legacy_dinfo(blkdev->dinfo);
936 if (blk_is_read_only(blkdev->blk) && !readonly) {
937 xen_be_printf(&blkdev->xendev, 0, "Unexpected read-only drive");
938 blkdev->blk = NULL;
939 return -1;
940 }
941
942
943 blk_ref(blkdev->blk);
944 }
945 blk_attach_dev_nofail(blkdev->blk, blkdev);
946 blkdev->file_size = blk_getlength(blkdev->blk);
947 if (blkdev->file_size < 0) {
948 BlockDriverState *bs = blk_bs(blkdev->blk);
949 const char *drv_name = bs ? bdrv_get_format_name(bs) : NULL;
950 xen_be_printf(&blkdev->xendev, 1, "blk_getlength: %d (%s) | drv %s\n",
951 (int)blkdev->file_size, strerror(-blkdev->file_size),
952 drv_name ?: "-");
953 blkdev->file_size = 0;
954 }
955
956 xen_be_printf(xendev, 1, "type \"%s\", fileproto \"%s\", filename \"%s\","
957 " size %" PRId64 " (%" PRId64 " MB)\n",
958 blkdev->type, blkdev->fileproto, blkdev->filename,
959 blkdev->file_size, blkdev->file_size >> 20);
960
961
962 xenstore_write_be_int(&blkdev->xendev, "sector-size", blkdev->file_blk);
963 xenstore_write_be_int64(&blkdev->xendev, "sectors",
964 blkdev->file_size / blkdev->file_blk);
965
966 if (xenstore_read_fe_int(&blkdev->xendev, "ring-ref", &blkdev->ring_ref) == -1) {
967 return -1;
968 }
969 if (xenstore_read_fe_int(&blkdev->xendev, "event-channel",
970 &blkdev->xendev.remote_port) == -1) {
971 return -1;
972 }
973 if (xenstore_read_fe_int(&blkdev->xendev, "feature-persistent", &pers)) {
974 blkdev->feature_persistent = FALSE;
975 } else {
976 blkdev->feature_persistent = !!pers;
977 }
978
979 blkdev->protocol = BLKIF_PROTOCOL_NATIVE;
980 if (blkdev->xendev.protocol) {
981 if (strcmp(blkdev->xendev.protocol, XEN_IO_PROTO_ABI_X86_32) == 0) {
982 blkdev->protocol = BLKIF_PROTOCOL_X86_32;
983 }
984 if (strcmp(blkdev->xendev.protocol, XEN_IO_PROTO_ABI_X86_64) == 0) {
985 blkdev->protocol = BLKIF_PROTOCOL_X86_64;
986 }
987 }
988
989 blkdev->sring = xengnttab_map_grant_ref(blkdev->xendev.gnttabdev,
990 blkdev->xendev.dom,
991 blkdev->ring_ref,
992 PROT_READ | PROT_WRITE);
993 if (!blkdev->sring) {
994 return -1;
995 }
996 blkdev->cnt_map++;
997
998 switch (blkdev->protocol) {
999 case BLKIF_PROTOCOL_NATIVE:
1000 {
1001 blkif_sring_t *sring_native = blkdev->sring;
1002 BACK_RING_INIT(&blkdev->rings.native, sring_native, XC_PAGE_SIZE);
1003 break;
1004 }
1005 case BLKIF_PROTOCOL_X86_32:
1006 {
1007 blkif_x86_32_sring_t *sring_x86_32 = blkdev->sring;
1008
1009 BACK_RING_INIT(&blkdev->rings.x86_32_part, sring_x86_32, XC_PAGE_SIZE);
1010 break;
1011 }
1012 case BLKIF_PROTOCOL_X86_64:
1013 {
1014 blkif_x86_64_sring_t *sring_x86_64 = blkdev->sring;
1015
1016 BACK_RING_INIT(&blkdev->rings.x86_64_part, sring_x86_64, XC_PAGE_SIZE);
1017 break;
1018 }
1019 }
1020
1021 if (blkdev->feature_persistent) {
1022
1023 blkdev->max_grants = max_requests * BLKIF_MAX_SEGMENTS_PER_REQUEST;
1024 blkdev->persistent_gnts = g_tree_new_full((GCompareDataFunc)int_cmp,
1025 NULL, NULL,
1026 batch_maps ?
1027 (GDestroyNotify)g_free :
1028 (GDestroyNotify)destroy_grant);
1029 blkdev->persistent_regions = NULL;
1030 blkdev->persistent_gnt_count = 0;
1031 }
1032
1033 xen_be_bind_evtchn(&blkdev->xendev);
1034
1035 xen_be_printf(&blkdev->xendev, 1, "ok: proto %s, ring-ref %d, "
1036 "remote port %d, local port %d\n",
1037 blkdev->xendev.protocol, blkdev->ring_ref,
1038 blkdev->xendev.remote_port, blkdev->xendev.local_port);
1039 return 0;
1040}
1041
1042static void blk_disconnect(struct XenDevice *xendev)
1043{
1044 struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
1045
1046 if (blkdev->blk) {
1047 blk_detach_dev(blkdev->blk, blkdev);
1048 blk_unref(blkdev->blk);
1049 blkdev->blk = NULL;
1050 }
1051 xen_be_unbind_evtchn(&blkdev->xendev);
1052
1053 if (blkdev->sring) {
1054 xengnttab_unmap(blkdev->xendev.gnttabdev, blkdev->sring, 1);
1055 blkdev->cnt_map--;
1056 blkdev->sring = NULL;
1057 }
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067 if (blkdev->feature_persistent) {
1068 g_tree_destroy(blkdev->persistent_gnts);
1069 assert(batch_maps || blkdev->persistent_gnt_count == 0);
1070 if (batch_maps) {
1071 blkdev->persistent_gnt_count = 0;
1072 g_slist_foreach(blkdev->persistent_regions,
1073 (GFunc)remove_persistent_region, blkdev);
1074 g_slist_free(blkdev->persistent_regions);
1075 }
1076 blkdev->feature_persistent = false;
1077 }
1078}
1079
1080static int blk_free(struct XenDevice *xendev)
1081{
1082 struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
1083 struct ioreq *ioreq;
1084
1085 if (blkdev->blk || blkdev->sring) {
1086 blk_disconnect(xendev);
1087 }
1088
1089 while (!QLIST_EMPTY(&blkdev->freelist)) {
1090 ioreq = QLIST_FIRST(&blkdev->freelist);
1091 QLIST_REMOVE(ioreq, list);
1092 qemu_iovec_destroy(&ioreq->v);
1093 g_free(ioreq);
1094 }
1095
1096 g_free(blkdev->params);
1097 g_free(blkdev->mode);
1098 g_free(blkdev->type);
1099 g_free(blkdev->dev);
1100 g_free(blkdev->devtype);
1101 qemu_bh_delete(blkdev->bh);
1102 return 0;
1103}
1104
1105static void blk_event(struct XenDevice *xendev)
1106{
1107 struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
1108
1109 qemu_bh_schedule(blkdev->bh);
1110}
1111
1112struct XenDevOps xen_blkdev_ops = {
1113 .size = sizeof(struct XenBlkDev),
1114 .flags = DEVOPS_FLAG_NEED_GNTDEV,
1115 .alloc = blk_alloc,
1116 .init = blk_init,
1117 .initialise = blk_connect,
1118 .disconnect = blk_disconnect,
1119 .event = blk_event,
1120 .free = blk_free,
1121};
1122