crypto: add compat cast5_set_key with nettle < 3.0.0
[qemu/ar7.git] / block / qed.c
blob5b24a9783fde1f0f1c762e9b8f1db8d7e3704760
1 /*
2 * QEMU Enhanced Disk Format
4 * Copyright IBM, Corp. 2010
6 * Authors:
7 * Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
8 * Anthony Liguori <aliguori@us.ibm.com>
10 * This work is licensed under the terms of the GNU LGPL, version 2 or later.
11 * See the COPYING.LIB file in the top-level directory.
15 #include "qemu/osdep.h"
16 #include "qemu/timer.h"
17 #include "trace.h"
18 #include "qed.h"
19 #include "qapi/qmp/qerror.h"
20 #include "migration/migration.h"
21 #include "sysemu/block-backend.h"
23 static const AIOCBInfo qed_aiocb_info = {
24 .aiocb_size = sizeof(QEDAIOCB),
27 static int bdrv_qed_probe(const uint8_t *buf, int buf_size,
28 const char *filename)
30 const QEDHeader *header = (const QEDHeader *)buf;
32 if (buf_size < sizeof(*header)) {
33 return 0;
35 if (le32_to_cpu(header->magic) != QED_MAGIC) {
36 return 0;
38 return 100;
41 /**
42 * Check whether an image format is raw
44 * @fmt: Backing file format, may be NULL
46 static bool qed_fmt_is_raw(const char *fmt)
48 return fmt && strcmp(fmt, "raw") == 0;
51 static void qed_header_le_to_cpu(const QEDHeader *le, QEDHeader *cpu)
53 cpu->magic = le32_to_cpu(le->magic);
54 cpu->cluster_size = le32_to_cpu(le->cluster_size);
55 cpu->table_size = le32_to_cpu(le->table_size);
56 cpu->header_size = le32_to_cpu(le->header_size);
57 cpu->features = le64_to_cpu(le->features);
58 cpu->compat_features = le64_to_cpu(le->compat_features);
59 cpu->autoclear_features = le64_to_cpu(le->autoclear_features);
60 cpu->l1_table_offset = le64_to_cpu(le->l1_table_offset);
61 cpu->image_size = le64_to_cpu(le->image_size);
62 cpu->backing_filename_offset = le32_to_cpu(le->backing_filename_offset);
63 cpu->backing_filename_size = le32_to_cpu(le->backing_filename_size);
66 static void qed_header_cpu_to_le(const QEDHeader *cpu, QEDHeader *le)
68 le->magic = cpu_to_le32(cpu->magic);
69 le->cluster_size = cpu_to_le32(cpu->cluster_size);
70 le->table_size = cpu_to_le32(cpu->table_size);
71 le->header_size = cpu_to_le32(cpu->header_size);
72 le->features = cpu_to_le64(cpu->features);
73 le->compat_features = cpu_to_le64(cpu->compat_features);
74 le->autoclear_features = cpu_to_le64(cpu->autoclear_features);
75 le->l1_table_offset = cpu_to_le64(cpu->l1_table_offset);
76 le->image_size = cpu_to_le64(cpu->image_size);
77 le->backing_filename_offset = cpu_to_le32(cpu->backing_filename_offset);
78 le->backing_filename_size = cpu_to_le32(cpu->backing_filename_size);
81 int qed_write_header_sync(BDRVQEDState *s)
83 QEDHeader le;
84 int ret;
86 qed_header_cpu_to_le(&s->header, &le);
87 ret = bdrv_pwrite(s->bs->file->bs, 0, &le, sizeof(le));
88 if (ret != sizeof(le)) {
89 return ret;
91 return 0;
94 typedef struct {
95 GenericCB gencb;
96 BDRVQEDState *s;
97 struct iovec iov;
98 QEMUIOVector qiov;
99 int nsectors;
100 uint8_t *buf;
101 } QEDWriteHeaderCB;
103 static void qed_write_header_cb(void *opaque, int ret)
105 QEDWriteHeaderCB *write_header_cb = opaque;
107 qemu_vfree(write_header_cb->buf);
108 gencb_complete(write_header_cb, ret);
111 static void qed_write_header_read_cb(void *opaque, int ret)
113 QEDWriteHeaderCB *write_header_cb = opaque;
114 BDRVQEDState *s = write_header_cb->s;
116 if (ret) {
117 qed_write_header_cb(write_header_cb, ret);
118 return;
121 /* Update header */
122 qed_header_cpu_to_le(&s->header, (QEDHeader *)write_header_cb->buf);
124 bdrv_aio_writev(s->bs->file->bs, 0, &write_header_cb->qiov,
125 write_header_cb->nsectors, qed_write_header_cb,
126 write_header_cb);
130 * Update header in-place (does not rewrite backing filename or other strings)
132 * This function only updates known header fields in-place and does not affect
133 * extra data after the QED header.
135 static void qed_write_header(BDRVQEDState *s, BlockCompletionFunc cb,
136 void *opaque)
138 /* We must write full sectors for O_DIRECT but cannot necessarily generate
139 * the data following the header if an unrecognized compat feature is
140 * active. Therefore, first read the sectors containing the header, update
141 * them, and write back.
144 int nsectors = (sizeof(QEDHeader) + BDRV_SECTOR_SIZE - 1) /
145 BDRV_SECTOR_SIZE;
146 size_t len = nsectors * BDRV_SECTOR_SIZE;
147 QEDWriteHeaderCB *write_header_cb = gencb_alloc(sizeof(*write_header_cb),
148 cb, opaque);
150 write_header_cb->s = s;
151 write_header_cb->nsectors = nsectors;
152 write_header_cb->buf = qemu_blockalign(s->bs, len);
153 write_header_cb->iov.iov_base = write_header_cb->buf;
154 write_header_cb->iov.iov_len = len;
155 qemu_iovec_init_external(&write_header_cb->qiov, &write_header_cb->iov, 1);
157 bdrv_aio_readv(s->bs->file->bs, 0, &write_header_cb->qiov, nsectors,
158 qed_write_header_read_cb, write_header_cb);
161 static uint64_t qed_max_image_size(uint32_t cluster_size, uint32_t table_size)
163 uint64_t table_entries;
164 uint64_t l2_size;
166 table_entries = (table_size * cluster_size) / sizeof(uint64_t);
167 l2_size = table_entries * cluster_size;
169 return l2_size * table_entries;
172 static bool qed_is_cluster_size_valid(uint32_t cluster_size)
174 if (cluster_size < QED_MIN_CLUSTER_SIZE ||
175 cluster_size > QED_MAX_CLUSTER_SIZE) {
176 return false;
178 if (cluster_size & (cluster_size - 1)) {
179 return false; /* not power of 2 */
181 return true;
184 static bool qed_is_table_size_valid(uint32_t table_size)
186 if (table_size < QED_MIN_TABLE_SIZE ||
187 table_size > QED_MAX_TABLE_SIZE) {
188 return false;
190 if (table_size & (table_size - 1)) {
191 return false; /* not power of 2 */
193 return true;
196 static bool qed_is_image_size_valid(uint64_t image_size, uint32_t cluster_size,
197 uint32_t table_size)
199 if (image_size % BDRV_SECTOR_SIZE != 0) {
200 return false; /* not multiple of sector size */
202 if (image_size > qed_max_image_size(cluster_size, table_size)) {
203 return false; /* image is too large */
205 return true;
209 * Read a string of known length from the image file
211 * @file: Image file
212 * @offset: File offset to start of string, in bytes
213 * @n: String length in bytes
214 * @buf: Destination buffer
215 * @buflen: Destination buffer length in bytes
216 * @ret: 0 on success, -errno on failure
218 * The string is NUL-terminated.
220 static int qed_read_string(BlockDriverState *file, uint64_t offset, size_t n,
221 char *buf, size_t buflen)
223 int ret;
224 if (n >= buflen) {
225 return -EINVAL;
227 ret = bdrv_pread(file, offset, buf, n);
228 if (ret < 0) {
229 return ret;
231 buf[n] = '\0';
232 return 0;
236 * Allocate new clusters
238 * @s: QED state
239 * @n: Number of contiguous clusters to allocate
240 * @ret: Offset of first allocated cluster
242 * This function only produces the offset where the new clusters should be
243 * written. It updates BDRVQEDState but does not make any changes to the image
244 * file.
246 static uint64_t qed_alloc_clusters(BDRVQEDState *s, unsigned int n)
248 uint64_t offset = s->file_size;
249 s->file_size += n * s->header.cluster_size;
250 return offset;
253 QEDTable *qed_alloc_table(BDRVQEDState *s)
255 /* Honor O_DIRECT memory alignment requirements */
256 return qemu_blockalign(s->bs,
257 s->header.cluster_size * s->header.table_size);
261 * Allocate a new zeroed L2 table
263 static CachedL2Table *qed_new_l2_table(BDRVQEDState *s)
265 CachedL2Table *l2_table = qed_alloc_l2_cache_entry(&s->l2_cache);
267 l2_table->table = qed_alloc_table(s);
268 l2_table->offset = qed_alloc_clusters(s, s->header.table_size);
270 memset(l2_table->table->offsets, 0,
271 s->header.cluster_size * s->header.table_size);
272 return l2_table;
275 static void qed_aio_next_io(void *opaque, int ret);
277 static void qed_plug_allocating_write_reqs(BDRVQEDState *s)
279 assert(!s->allocating_write_reqs_plugged);
281 s->allocating_write_reqs_plugged = true;
284 static void qed_unplug_allocating_write_reqs(BDRVQEDState *s)
286 QEDAIOCB *acb;
288 assert(s->allocating_write_reqs_plugged);
290 s->allocating_write_reqs_plugged = false;
292 acb = QSIMPLEQ_FIRST(&s->allocating_write_reqs);
293 if (acb) {
294 qed_aio_next_io(acb, 0);
298 static void qed_finish_clear_need_check(void *opaque, int ret)
300 /* Do nothing */
303 static void qed_flush_after_clear_need_check(void *opaque, int ret)
305 BDRVQEDState *s = opaque;
307 bdrv_aio_flush(s->bs, qed_finish_clear_need_check, s);
309 /* No need to wait until flush completes */
310 qed_unplug_allocating_write_reqs(s);
313 static void qed_clear_need_check(void *opaque, int ret)
315 BDRVQEDState *s = opaque;
317 if (ret) {
318 qed_unplug_allocating_write_reqs(s);
319 return;
322 s->header.features &= ~QED_F_NEED_CHECK;
323 qed_write_header(s, qed_flush_after_clear_need_check, s);
326 static void qed_need_check_timer_cb(void *opaque)
328 BDRVQEDState *s = opaque;
330 /* The timer should only fire when allocating writes have drained */
331 assert(!QSIMPLEQ_FIRST(&s->allocating_write_reqs));
333 trace_qed_need_check_timer_cb(s);
335 qed_plug_allocating_write_reqs(s);
337 /* Ensure writes are on disk before clearing flag */
338 bdrv_aio_flush(s->bs, qed_clear_need_check, s);
341 static void qed_start_need_check_timer(BDRVQEDState *s)
343 trace_qed_start_need_check_timer(s);
345 /* Use QEMU_CLOCK_VIRTUAL so we don't alter the image file while suspended for
346 * migration.
348 timer_mod(s->need_check_timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) +
349 get_ticks_per_sec() * QED_NEED_CHECK_TIMEOUT);
352 /* It's okay to call this multiple times or when no timer is started */
353 static void qed_cancel_need_check_timer(BDRVQEDState *s)
355 trace_qed_cancel_need_check_timer(s);
356 timer_del(s->need_check_timer);
359 static void bdrv_qed_detach_aio_context(BlockDriverState *bs)
361 BDRVQEDState *s = bs->opaque;
363 qed_cancel_need_check_timer(s);
364 timer_free(s->need_check_timer);
367 static void bdrv_qed_attach_aio_context(BlockDriverState *bs,
368 AioContext *new_context)
370 BDRVQEDState *s = bs->opaque;
372 s->need_check_timer = aio_timer_new(new_context,
373 QEMU_CLOCK_VIRTUAL, SCALE_NS,
374 qed_need_check_timer_cb, s);
375 if (s->header.features & QED_F_NEED_CHECK) {
376 qed_start_need_check_timer(s);
380 static int bdrv_qed_open(BlockDriverState *bs, QDict *options, int flags,
381 Error **errp)
383 BDRVQEDState *s = bs->opaque;
384 QEDHeader le_header;
385 int64_t file_size;
386 int ret;
388 s->bs = bs;
389 QSIMPLEQ_INIT(&s->allocating_write_reqs);
391 ret = bdrv_pread(bs->file->bs, 0, &le_header, sizeof(le_header));
392 if (ret < 0) {
393 return ret;
395 qed_header_le_to_cpu(&le_header, &s->header);
397 if (s->header.magic != QED_MAGIC) {
398 error_setg(errp, "Image not in QED format");
399 return -EINVAL;
401 if (s->header.features & ~QED_FEATURE_MASK) {
402 /* image uses unsupported feature bits */
403 error_setg(errp, "Unsupported QED features: %" PRIx64,
404 s->header.features & ~QED_FEATURE_MASK);
405 return -ENOTSUP;
407 if (!qed_is_cluster_size_valid(s->header.cluster_size)) {
408 return -EINVAL;
411 /* Round down file size to the last cluster */
412 file_size = bdrv_getlength(bs->file->bs);
413 if (file_size < 0) {
414 return file_size;
416 s->file_size = qed_start_of_cluster(s, file_size);
418 if (!qed_is_table_size_valid(s->header.table_size)) {
419 return -EINVAL;
421 if (!qed_is_image_size_valid(s->header.image_size,
422 s->header.cluster_size,
423 s->header.table_size)) {
424 return -EINVAL;
426 if (!qed_check_table_offset(s, s->header.l1_table_offset)) {
427 return -EINVAL;
430 s->table_nelems = (s->header.cluster_size * s->header.table_size) /
431 sizeof(uint64_t);
432 s->l2_shift = ctz32(s->header.cluster_size);
433 s->l2_mask = s->table_nelems - 1;
434 s->l1_shift = s->l2_shift + ctz32(s->table_nelems);
436 /* Header size calculation must not overflow uint32_t */
437 if (s->header.header_size > UINT32_MAX / s->header.cluster_size) {
438 return -EINVAL;
441 if ((s->header.features & QED_F_BACKING_FILE)) {
442 if ((uint64_t)s->header.backing_filename_offset +
443 s->header.backing_filename_size >
444 s->header.cluster_size * s->header.header_size) {
445 return -EINVAL;
448 ret = qed_read_string(bs->file->bs, s->header.backing_filename_offset,
449 s->header.backing_filename_size, bs->backing_file,
450 sizeof(bs->backing_file));
451 if (ret < 0) {
452 return ret;
455 if (s->header.features & QED_F_BACKING_FORMAT_NO_PROBE) {
456 pstrcpy(bs->backing_format, sizeof(bs->backing_format), "raw");
460 /* Reset unknown autoclear feature bits. This is a backwards
461 * compatibility mechanism that allows images to be opened by older
462 * programs, which "knock out" unknown feature bits. When an image is
463 * opened by a newer program again it can detect that the autoclear
464 * feature is no longer valid.
466 if ((s->header.autoclear_features & ~QED_AUTOCLEAR_FEATURE_MASK) != 0 &&
467 !bdrv_is_read_only(bs->file->bs) && !(flags & BDRV_O_INACTIVE)) {
468 s->header.autoclear_features &= QED_AUTOCLEAR_FEATURE_MASK;
470 ret = qed_write_header_sync(s);
471 if (ret) {
472 return ret;
475 /* From here on only known autoclear feature bits are valid */
476 bdrv_flush(bs->file->bs);
479 s->l1_table = qed_alloc_table(s);
480 qed_init_l2_cache(&s->l2_cache);
482 ret = qed_read_l1_table_sync(s);
483 if (ret) {
484 goto out;
487 /* If image was not closed cleanly, check consistency */
488 if (!(flags & BDRV_O_CHECK) && (s->header.features & QED_F_NEED_CHECK)) {
489 /* Read-only images cannot be fixed. There is no risk of corruption
490 * since write operations are not possible. Therefore, allow
491 * potentially inconsistent images to be opened read-only. This can
492 * aid data recovery from an otherwise inconsistent image.
494 if (!bdrv_is_read_only(bs->file->bs) &&
495 !(flags & BDRV_O_INACTIVE)) {
496 BdrvCheckResult result = {0};
498 ret = qed_check(s, &result, true);
499 if (ret) {
500 goto out;
505 bdrv_qed_attach_aio_context(bs, bdrv_get_aio_context(bs));
507 out:
508 if (ret) {
509 qed_free_l2_cache(&s->l2_cache);
510 qemu_vfree(s->l1_table);
512 return ret;
515 static void bdrv_qed_refresh_limits(BlockDriverState *bs, Error **errp)
517 BDRVQEDState *s = bs->opaque;
519 bs->bl.write_zeroes_alignment = s->header.cluster_size >> BDRV_SECTOR_BITS;
522 /* We have nothing to do for QED reopen, stubs just return
523 * success */
524 static int bdrv_qed_reopen_prepare(BDRVReopenState *state,
525 BlockReopenQueue *queue, Error **errp)
527 return 0;
530 static void bdrv_qed_close(BlockDriverState *bs)
532 BDRVQEDState *s = bs->opaque;
534 bdrv_qed_detach_aio_context(bs);
536 /* Ensure writes reach stable storage */
537 bdrv_flush(bs->file->bs);
539 /* Clean shutdown, no check required on next open */
540 if (s->header.features & QED_F_NEED_CHECK) {
541 s->header.features &= ~QED_F_NEED_CHECK;
542 qed_write_header_sync(s);
545 qed_free_l2_cache(&s->l2_cache);
546 qemu_vfree(s->l1_table);
549 static int qed_create(const char *filename, uint32_t cluster_size,
550 uint64_t image_size, uint32_t table_size,
551 const char *backing_file, const char *backing_fmt,
552 QemuOpts *opts, Error **errp)
554 QEDHeader header = {
555 .magic = QED_MAGIC,
556 .cluster_size = cluster_size,
557 .table_size = table_size,
558 .header_size = 1,
559 .features = 0,
560 .compat_features = 0,
561 .l1_table_offset = cluster_size,
562 .image_size = image_size,
564 QEDHeader le_header;
565 uint8_t *l1_table = NULL;
566 size_t l1_size = header.cluster_size * header.table_size;
567 Error *local_err = NULL;
568 int ret = 0;
569 BlockBackend *blk;
571 ret = bdrv_create_file(filename, opts, &local_err);
572 if (ret < 0) {
573 error_propagate(errp, local_err);
574 return ret;
577 blk = blk_new_open(filename, NULL, NULL,
578 BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_PROTOCOL,
579 &local_err);
580 if (blk == NULL) {
581 error_propagate(errp, local_err);
582 return -EIO;
585 blk_set_allow_write_beyond_eof(blk, true);
587 /* File must start empty and grow, check truncate is supported */
588 ret = blk_truncate(blk, 0);
589 if (ret < 0) {
590 goto out;
593 if (backing_file) {
594 header.features |= QED_F_BACKING_FILE;
595 header.backing_filename_offset = sizeof(le_header);
596 header.backing_filename_size = strlen(backing_file);
598 if (qed_fmt_is_raw(backing_fmt)) {
599 header.features |= QED_F_BACKING_FORMAT_NO_PROBE;
603 qed_header_cpu_to_le(&header, &le_header);
604 ret = blk_pwrite(blk, 0, &le_header, sizeof(le_header));
605 if (ret < 0) {
606 goto out;
608 ret = blk_pwrite(blk, sizeof(le_header), backing_file,
609 header.backing_filename_size);
610 if (ret < 0) {
611 goto out;
614 l1_table = g_malloc0(l1_size);
615 ret = blk_pwrite(blk, header.l1_table_offset, l1_table, l1_size);
616 if (ret < 0) {
617 goto out;
620 ret = 0; /* success */
621 out:
622 g_free(l1_table);
623 blk_unref(blk);
624 return ret;
627 static int bdrv_qed_create(const char *filename, QemuOpts *opts, Error **errp)
629 uint64_t image_size = 0;
630 uint32_t cluster_size = QED_DEFAULT_CLUSTER_SIZE;
631 uint32_t table_size = QED_DEFAULT_TABLE_SIZE;
632 char *backing_file = NULL;
633 char *backing_fmt = NULL;
634 int ret;
636 image_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
637 BDRV_SECTOR_SIZE);
638 backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
639 backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT);
640 cluster_size = qemu_opt_get_size_del(opts,
641 BLOCK_OPT_CLUSTER_SIZE,
642 QED_DEFAULT_CLUSTER_SIZE);
643 table_size = qemu_opt_get_size_del(opts, BLOCK_OPT_TABLE_SIZE,
644 QED_DEFAULT_TABLE_SIZE);
646 if (!qed_is_cluster_size_valid(cluster_size)) {
647 error_setg(errp, "QED cluster size must be within range [%u, %u] "
648 "and power of 2",
649 QED_MIN_CLUSTER_SIZE, QED_MAX_CLUSTER_SIZE);
650 ret = -EINVAL;
651 goto finish;
653 if (!qed_is_table_size_valid(table_size)) {
654 error_setg(errp, "QED table size must be within range [%u, %u] "
655 "and power of 2",
656 QED_MIN_TABLE_SIZE, QED_MAX_TABLE_SIZE);
657 ret = -EINVAL;
658 goto finish;
660 if (!qed_is_image_size_valid(image_size, cluster_size, table_size)) {
661 error_setg(errp, "QED image size must be a non-zero multiple of "
662 "cluster size and less than %" PRIu64 " bytes",
663 qed_max_image_size(cluster_size, table_size));
664 ret = -EINVAL;
665 goto finish;
668 ret = qed_create(filename, cluster_size, image_size, table_size,
669 backing_file, backing_fmt, opts, errp);
671 finish:
672 g_free(backing_file);
673 g_free(backing_fmt);
674 return ret;
677 typedef struct {
678 BlockDriverState *bs;
679 Coroutine *co;
680 uint64_t pos;
681 int64_t status;
682 int *pnum;
683 BlockDriverState **file;
684 } QEDIsAllocatedCB;
686 static void qed_is_allocated_cb(void *opaque, int ret, uint64_t offset, size_t len)
688 QEDIsAllocatedCB *cb = opaque;
689 BDRVQEDState *s = cb->bs->opaque;
690 *cb->pnum = len / BDRV_SECTOR_SIZE;
691 switch (ret) {
692 case QED_CLUSTER_FOUND:
693 offset |= qed_offset_into_cluster(s, cb->pos);
694 cb->status = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | offset;
695 *cb->file = cb->bs->file->bs;
696 break;
697 case QED_CLUSTER_ZERO:
698 cb->status = BDRV_BLOCK_ZERO;
699 break;
700 case QED_CLUSTER_L2:
701 case QED_CLUSTER_L1:
702 cb->status = 0;
703 break;
704 default:
705 assert(ret < 0);
706 cb->status = ret;
707 break;
710 if (cb->co) {
711 qemu_coroutine_enter(cb->co, NULL);
715 static int64_t coroutine_fn bdrv_qed_co_get_block_status(BlockDriverState *bs,
716 int64_t sector_num,
717 int nb_sectors, int *pnum,
718 BlockDriverState **file)
720 BDRVQEDState *s = bs->opaque;
721 size_t len = (size_t)nb_sectors * BDRV_SECTOR_SIZE;
722 QEDIsAllocatedCB cb = {
723 .bs = bs,
724 .pos = (uint64_t)sector_num * BDRV_SECTOR_SIZE,
725 .status = BDRV_BLOCK_OFFSET_MASK,
726 .pnum = pnum,
727 .file = file,
729 QEDRequest request = { .l2_table = NULL };
731 qed_find_cluster(s, &request, cb.pos, len, qed_is_allocated_cb, &cb);
733 /* Now sleep if the callback wasn't invoked immediately */
734 while (cb.status == BDRV_BLOCK_OFFSET_MASK) {
735 cb.co = qemu_coroutine_self();
736 qemu_coroutine_yield();
739 qed_unref_l2_cache_entry(request.l2_table);
741 return cb.status;
744 static BDRVQEDState *acb_to_s(QEDAIOCB *acb)
746 return acb->common.bs->opaque;
750 * Read from the backing file or zero-fill if no backing file
752 * @s: QED state
753 * @pos: Byte position in device
754 * @qiov: Destination I/O vector
755 * @backing_qiov: Possibly shortened copy of qiov, to be allocated here
756 * @cb: Completion function
757 * @opaque: User data for completion function
759 * This function reads qiov->size bytes starting at pos from the backing file.
760 * If there is no backing file then zeroes are read.
762 static void qed_read_backing_file(BDRVQEDState *s, uint64_t pos,
763 QEMUIOVector *qiov,
764 QEMUIOVector **backing_qiov,
765 BlockCompletionFunc *cb, void *opaque)
767 uint64_t backing_length = 0;
768 size_t size;
770 /* If there is a backing file, get its length. Treat the absence of a
771 * backing file like a zero length backing file.
773 if (s->bs->backing) {
774 int64_t l = bdrv_getlength(s->bs->backing->bs);
775 if (l < 0) {
776 cb(opaque, l);
777 return;
779 backing_length = l;
782 /* Zero all sectors if reading beyond the end of the backing file */
783 if (pos >= backing_length ||
784 pos + qiov->size > backing_length) {
785 qemu_iovec_memset(qiov, 0, 0, qiov->size);
788 /* Complete now if there are no backing file sectors to read */
789 if (pos >= backing_length) {
790 cb(opaque, 0);
791 return;
794 /* If the read straddles the end of the backing file, shorten it */
795 size = MIN((uint64_t)backing_length - pos, qiov->size);
797 assert(*backing_qiov == NULL);
798 *backing_qiov = g_new(QEMUIOVector, 1);
799 qemu_iovec_init(*backing_qiov, qiov->niov);
800 qemu_iovec_concat(*backing_qiov, qiov, 0, size);
802 BLKDBG_EVENT(s->bs->file, BLKDBG_READ_BACKING_AIO);
803 bdrv_aio_readv(s->bs->backing->bs, pos / BDRV_SECTOR_SIZE,
804 *backing_qiov, size / BDRV_SECTOR_SIZE, cb, opaque);
807 typedef struct {
808 GenericCB gencb;
809 BDRVQEDState *s;
810 QEMUIOVector qiov;
811 QEMUIOVector *backing_qiov;
812 struct iovec iov;
813 uint64_t offset;
814 } CopyFromBackingFileCB;
816 static void qed_copy_from_backing_file_cb(void *opaque, int ret)
818 CopyFromBackingFileCB *copy_cb = opaque;
819 qemu_vfree(copy_cb->iov.iov_base);
820 gencb_complete(&copy_cb->gencb, ret);
823 static void qed_copy_from_backing_file_write(void *opaque, int ret)
825 CopyFromBackingFileCB *copy_cb = opaque;
826 BDRVQEDState *s = copy_cb->s;
828 if (copy_cb->backing_qiov) {
829 qemu_iovec_destroy(copy_cb->backing_qiov);
830 g_free(copy_cb->backing_qiov);
831 copy_cb->backing_qiov = NULL;
834 if (ret) {
835 qed_copy_from_backing_file_cb(copy_cb, ret);
836 return;
839 BLKDBG_EVENT(s->bs->file, BLKDBG_COW_WRITE);
840 bdrv_aio_writev(s->bs->file->bs, copy_cb->offset / BDRV_SECTOR_SIZE,
841 &copy_cb->qiov, copy_cb->qiov.size / BDRV_SECTOR_SIZE,
842 qed_copy_from_backing_file_cb, copy_cb);
846 * Copy data from backing file into the image
848 * @s: QED state
849 * @pos: Byte position in device
850 * @len: Number of bytes
851 * @offset: Byte offset in image file
852 * @cb: Completion function
853 * @opaque: User data for completion function
855 static void qed_copy_from_backing_file(BDRVQEDState *s, uint64_t pos,
856 uint64_t len, uint64_t offset,
857 BlockCompletionFunc *cb,
858 void *opaque)
860 CopyFromBackingFileCB *copy_cb;
862 /* Skip copy entirely if there is no work to do */
863 if (len == 0) {
864 cb(opaque, 0);
865 return;
868 copy_cb = gencb_alloc(sizeof(*copy_cb), cb, opaque);
869 copy_cb->s = s;
870 copy_cb->offset = offset;
871 copy_cb->backing_qiov = NULL;
872 copy_cb->iov.iov_base = qemu_blockalign(s->bs, len);
873 copy_cb->iov.iov_len = len;
874 qemu_iovec_init_external(&copy_cb->qiov, &copy_cb->iov, 1);
876 qed_read_backing_file(s, pos, &copy_cb->qiov, &copy_cb->backing_qiov,
877 qed_copy_from_backing_file_write, copy_cb);
881 * Link one or more contiguous clusters into a table
883 * @s: QED state
884 * @table: L2 table
885 * @index: First cluster index
886 * @n: Number of contiguous clusters
887 * @cluster: First cluster offset
889 * The cluster offset may be an allocated byte offset in the image file, the
890 * zero cluster marker, or the unallocated cluster marker.
892 static void qed_update_l2_table(BDRVQEDState *s, QEDTable *table, int index,
893 unsigned int n, uint64_t cluster)
895 int i;
896 for (i = index; i < index + n; i++) {
897 table->offsets[i] = cluster;
898 if (!qed_offset_is_unalloc_cluster(cluster) &&
899 !qed_offset_is_zero_cluster(cluster)) {
900 cluster += s->header.cluster_size;
905 static void qed_aio_complete_bh(void *opaque)
907 QEDAIOCB *acb = opaque;
908 BlockCompletionFunc *cb = acb->common.cb;
909 void *user_opaque = acb->common.opaque;
910 int ret = acb->bh_ret;
912 qemu_bh_delete(acb->bh);
913 qemu_aio_unref(acb);
915 /* Invoke callback */
916 cb(user_opaque, ret);
919 static void qed_aio_complete(QEDAIOCB *acb, int ret)
921 BDRVQEDState *s = acb_to_s(acb);
923 trace_qed_aio_complete(s, acb, ret);
925 /* Free resources */
926 qemu_iovec_destroy(&acb->cur_qiov);
927 qed_unref_l2_cache_entry(acb->request.l2_table);
929 /* Free the buffer we may have allocated for zero writes */
930 if (acb->flags & QED_AIOCB_ZERO) {
931 qemu_vfree(acb->qiov->iov[0].iov_base);
932 acb->qiov->iov[0].iov_base = NULL;
935 /* Arrange for a bh to invoke the completion function */
936 acb->bh_ret = ret;
937 acb->bh = aio_bh_new(bdrv_get_aio_context(acb->common.bs),
938 qed_aio_complete_bh, acb);
939 qemu_bh_schedule(acb->bh);
941 /* Start next allocating write request waiting behind this one. Note that
942 * requests enqueue themselves when they first hit an unallocated cluster
943 * but they wait until the entire request is finished before waking up the
944 * next request in the queue. This ensures that we don't cycle through
945 * requests multiple times but rather finish one at a time completely.
947 if (acb == QSIMPLEQ_FIRST(&s->allocating_write_reqs)) {
948 QSIMPLEQ_REMOVE_HEAD(&s->allocating_write_reqs, next);
949 acb = QSIMPLEQ_FIRST(&s->allocating_write_reqs);
950 if (acb) {
951 qed_aio_next_io(acb, 0);
952 } else if (s->header.features & QED_F_NEED_CHECK) {
953 qed_start_need_check_timer(s);
959 * Commit the current L2 table to the cache
961 static void qed_commit_l2_update(void *opaque, int ret)
963 QEDAIOCB *acb = opaque;
964 BDRVQEDState *s = acb_to_s(acb);
965 CachedL2Table *l2_table = acb->request.l2_table;
966 uint64_t l2_offset = l2_table->offset;
968 qed_commit_l2_cache_entry(&s->l2_cache, l2_table);
970 /* This is guaranteed to succeed because we just committed the entry to the
971 * cache.
973 acb->request.l2_table = qed_find_l2_cache_entry(&s->l2_cache, l2_offset);
974 assert(acb->request.l2_table != NULL);
976 qed_aio_next_io(opaque, ret);
980 * Update L1 table with new L2 table offset and write it out
982 static void qed_aio_write_l1_update(void *opaque, int ret)
984 QEDAIOCB *acb = opaque;
985 BDRVQEDState *s = acb_to_s(acb);
986 int index;
988 if (ret) {
989 qed_aio_complete(acb, ret);
990 return;
993 index = qed_l1_index(s, acb->cur_pos);
994 s->l1_table->offsets[index] = acb->request.l2_table->offset;
996 qed_write_l1_table(s, index, 1, qed_commit_l2_update, acb);
1000 * Update L2 table with new cluster offsets and write them out
1002 static void qed_aio_write_l2_update(QEDAIOCB *acb, int ret, uint64_t offset)
1004 BDRVQEDState *s = acb_to_s(acb);
1005 bool need_alloc = acb->find_cluster_ret == QED_CLUSTER_L1;
1006 int index;
1008 if (ret) {
1009 goto err;
1012 if (need_alloc) {
1013 qed_unref_l2_cache_entry(acb->request.l2_table);
1014 acb->request.l2_table = qed_new_l2_table(s);
1017 index = qed_l2_index(s, acb->cur_pos);
1018 qed_update_l2_table(s, acb->request.l2_table->table, index, acb->cur_nclusters,
1019 offset);
1021 if (need_alloc) {
1022 /* Write out the whole new L2 table */
1023 qed_write_l2_table(s, &acb->request, 0, s->table_nelems, true,
1024 qed_aio_write_l1_update, acb);
1025 } else {
1026 /* Write out only the updated part of the L2 table */
1027 qed_write_l2_table(s, &acb->request, index, acb->cur_nclusters, false,
1028 qed_aio_next_io, acb);
1030 return;
1032 err:
1033 qed_aio_complete(acb, ret);
1036 static void qed_aio_write_l2_update_cb(void *opaque, int ret)
1038 QEDAIOCB *acb = opaque;
1039 qed_aio_write_l2_update(acb, ret, acb->cur_cluster);
1043 * Flush new data clusters before updating the L2 table
1045 * This flush is necessary when a backing file is in use. A crash during an
1046 * allocating write could result in empty clusters in the image. If the write
1047 * only touched a subregion of the cluster, then backing image sectors have
1048 * been lost in the untouched region. The solution is to flush after writing a
1049 * new data cluster and before updating the L2 table.
1051 static void qed_aio_write_flush_before_l2_update(void *opaque, int ret)
1053 QEDAIOCB *acb = opaque;
1054 BDRVQEDState *s = acb_to_s(acb);
1056 if (!bdrv_aio_flush(s->bs->file->bs, qed_aio_write_l2_update_cb, opaque)) {
1057 qed_aio_complete(acb, -EIO);
1062 * Write data to the image file
1064 static void qed_aio_write_main(void *opaque, int ret)
1066 QEDAIOCB *acb = opaque;
1067 BDRVQEDState *s = acb_to_s(acb);
1068 uint64_t offset = acb->cur_cluster +
1069 qed_offset_into_cluster(s, acb->cur_pos);
1070 BlockCompletionFunc *next_fn;
1072 trace_qed_aio_write_main(s, acb, ret, offset, acb->cur_qiov.size);
1074 if (ret) {
1075 qed_aio_complete(acb, ret);
1076 return;
1079 if (acb->find_cluster_ret == QED_CLUSTER_FOUND) {
1080 next_fn = qed_aio_next_io;
1081 } else {
1082 if (s->bs->backing) {
1083 next_fn = qed_aio_write_flush_before_l2_update;
1084 } else {
1085 next_fn = qed_aio_write_l2_update_cb;
1089 BLKDBG_EVENT(s->bs->file, BLKDBG_WRITE_AIO);
1090 bdrv_aio_writev(s->bs->file->bs, offset / BDRV_SECTOR_SIZE,
1091 &acb->cur_qiov, acb->cur_qiov.size / BDRV_SECTOR_SIZE,
1092 next_fn, acb);
1096 * Populate back untouched region of new data cluster
1098 static void qed_aio_write_postfill(void *opaque, int ret)
1100 QEDAIOCB *acb = opaque;
1101 BDRVQEDState *s = acb_to_s(acb);
1102 uint64_t start = acb->cur_pos + acb->cur_qiov.size;
1103 uint64_t len =
1104 qed_start_of_cluster(s, start + s->header.cluster_size - 1) - start;
1105 uint64_t offset = acb->cur_cluster +
1106 qed_offset_into_cluster(s, acb->cur_pos) +
1107 acb->cur_qiov.size;
1109 if (ret) {
1110 qed_aio_complete(acb, ret);
1111 return;
1114 trace_qed_aio_write_postfill(s, acb, start, len, offset);
1115 qed_copy_from_backing_file(s, start, len, offset,
1116 qed_aio_write_main, acb);
1120 * Populate front untouched region of new data cluster
1122 static void qed_aio_write_prefill(void *opaque, int ret)
1124 QEDAIOCB *acb = opaque;
1125 BDRVQEDState *s = acb_to_s(acb);
1126 uint64_t start = qed_start_of_cluster(s, acb->cur_pos);
1127 uint64_t len = qed_offset_into_cluster(s, acb->cur_pos);
1129 trace_qed_aio_write_prefill(s, acb, start, len, acb->cur_cluster);
1130 qed_copy_from_backing_file(s, start, len, acb->cur_cluster,
1131 qed_aio_write_postfill, acb);
1135 * Check if the QED_F_NEED_CHECK bit should be set during allocating write
1137 static bool qed_should_set_need_check(BDRVQEDState *s)
1139 /* The flush before L2 update path ensures consistency */
1140 if (s->bs->backing) {
1141 return false;
1144 return !(s->header.features & QED_F_NEED_CHECK);
1147 static void qed_aio_write_zero_cluster(void *opaque, int ret)
1149 QEDAIOCB *acb = opaque;
1151 if (ret) {
1152 qed_aio_complete(acb, ret);
1153 return;
1156 qed_aio_write_l2_update(acb, 0, 1);
1160 * Write new data cluster
1162 * @acb: Write request
1163 * @len: Length in bytes
1165 * This path is taken when writing to previously unallocated clusters.
1167 static void qed_aio_write_alloc(QEDAIOCB *acb, size_t len)
1169 BDRVQEDState *s = acb_to_s(acb);
1170 BlockCompletionFunc *cb;
1172 /* Cancel timer when the first allocating request comes in */
1173 if (QSIMPLEQ_EMPTY(&s->allocating_write_reqs)) {
1174 qed_cancel_need_check_timer(s);
1177 /* Freeze this request if another allocating write is in progress */
1178 if (acb != QSIMPLEQ_FIRST(&s->allocating_write_reqs)) {
1179 QSIMPLEQ_INSERT_TAIL(&s->allocating_write_reqs, acb, next);
1181 if (acb != QSIMPLEQ_FIRST(&s->allocating_write_reqs) ||
1182 s->allocating_write_reqs_plugged) {
1183 return; /* wait for existing request to finish */
1186 acb->cur_nclusters = qed_bytes_to_clusters(s,
1187 qed_offset_into_cluster(s, acb->cur_pos) + len);
1188 qemu_iovec_concat(&acb->cur_qiov, acb->qiov, acb->qiov_offset, len);
1190 if (acb->flags & QED_AIOCB_ZERO) {
1191 /* Skip ahead if the clusters are already zero */
1192 if (acb->find_cluster_ret == QED_CLUSTER_ZERO) {
1193 qed_aio_next_io(acb, 0);
1194 return;
1197 cb = qed_aio_write_zero_cluster;
1198 } else {
1199 cb = qed_aio_write_prefill;
1200 acb->cur_cluster = qed_alloc_clusters(s, acb->cur_nclusters);
1203 if (qed_should_set_need_check(s)) {
1204 s->header.features |= QED_F_NEED_CHECK;
1205 qed_write_header(s, cb, acb);
1206 } else {
1207 cb(acb, 0);
1212 * Write data cluster in place
1214 * @acb: Write request
1215 * @offset: Cluster offset in bytes
1216 * @len: Length in bytes
1218 * This path is taken when writing to already allocated clusters.
1220 static void qed_aio_write_inplace(QEDAIOCB *acb, uint64_t offset, size_t len)
1222 /* Allocate buffer for zero writes */
1223 if (acb->flags & QED_AIOCB_ZERO) {
1224 struct iovec *iov = acb->qiov->iov;
1226 if (!iov->iov_base) {
1227 iov->iov_base = qemu_try_blockalign(acb->common.bs, iov->iov_len);
1228 if (iov->iov_base == NULL) {
1229 qed_aio_complete(acb, -ENOMEM);
1230 return;
1232 memset(iov->iov_base, 0, iov->iov_len);
1236 /* Calculate the I/O vector */
1237 acb->cur_cluster = offset;
1238 qemu_iovec_concat(&acb->cur_qiov, acb->qiov, acb->qiov_offset, len);
1240 /* Do the actual write */
1241 qed_aio_write_main(acb, 0);
1245 * Write data cluster
1247 * @opaque: Write request
1248 * @ret: QED_CLUSTER_FOUND, QED_CLUSTER_L2, QED_CLUSTER_L1,
1249 * or -errno
1250 * @offset: Cluster offset in bytes
1251 * @len: Length in bytes
1253 * Callback from qed_find_cluster().
1255 static void qed_aio_write_data(void *opaque, int ret,
1256 uint64_t offset, size_t len)
1258 QEDAIOCB *acb = opaque;
1260 trace_qed_aio_write_data(acb_to_s(acb), acb, ret, offset, len);
1262 acb->find_cluster_ret = ret;
1264 switch (ret) {
1265 case QED_CLUSTER_FOUND:
1266 qed_aio_write_inplace(acb, offset, len);
1267 break;
1269 case QED_CLUSTER_L2:
1270 case QED_CLUSTER_L1:
1271 case QED_CLUSTER_ZERO:
1272 qed_aio_write_alloc(acb, len);
1273 break;
1275 default:
1276 qed_aio_complete(acb, ret);
1277 break;
1282 * Read data cluster
1284 * @opaque: Read request
1285 * @ret: QED_CLUSTER_FOUND, QED_CLUSTER_L2, QED_CLUSTER_L1,
1286 * or -errno
1287 * @offset: Cluster offset in bytes
1288 * @len: Length in bytes
1290 * Callback from qed_find_cluster().
1292 static void qed_aio_read_data(void *opaque, int ret,
1293 uint64_t offset, size_t len)
1295 QEDAIOCB *acb = opaque;
1296 BDRVQEDState *s = acb_to_s(acb);
1297 BlockDriverState *bs = acb->common.bs;
1299 /* Adjust offset into cluster */
1300 offset += qed_offset_into_cluster(s, acb->cur_pos);
1302 trace_qed_aio_read_data(s, acb, ret, offset, len);
1304 if (ret < 0) {
1305 goto err;
1308 qemu_iovec_concat(&acb->cur_qiov, acb->qiov, acb->qiov_offset, len);
1310 /* Handle zero cluster and backing file reads */
1311 if (ret == QED_CLUSTER_ZERO) {
1312 qemu_iovec_memset(&acb->cur_qiov, 0, 0, acb->cur_qiov.size);
1313 qed_aio_next_io(acb, 0);
1314 return;
1315 } else if (ret != QED_CLUSTER_FOUND) {
1316 qed_read_backing_file(s, acb->cur_pos, &acb->cur_qiov,
1317 &acb->backing_qiov, qed_aio_next_io, acb);
1318 return;
1321 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
1322 bdrv_aio_readv(bs->file->bs, offset / BDRV_SECTOR_SIZE,
1323 &acb->cur_qiov, acb->cur_qiov.size / BDRV_SECTOR_SIZE,
1324 qed_aio_next_io, acb);
1325 return;
1327 err:
1328 qed_aio_complete(acb, ret);
1332 * Begin next I/O or complete the request
1334 static void qed_aio_next_io(void *opaque, int ret)
1336 QEDAIOCB *acb = opaque;
1337 BDRVQEDState *s = acb_to_s(acb);
1338 QEDFindClusterFunc *io_fn = (acb->flags & QED_AIOCB_WRITE) ?
1339 qed_aio_write_data : qed_aio_read_data;
1341 trace_qed_aio_next_io(s, acb, ret, acb->cur_pos + acb->cur_qiov.size);
1343 if (acb->backing_qiov) {
1344 qemu_iovec_destroy(acb->backing_qiov);
1345 g_free(acb->backing_qiov);
1346 acb->backing_qiov = NULL;
1349 /* Handle I/O error */
1350 if (ret) {
1351 qed_aio_complete(acb, ret);
1352 return;
1355 acb->qiov_offset += acb->cur_qiov.size;
1356 acb->cur_pos += acb->cur_qiov.size;
1357 qemu_iovec_reset(&acb->cur_qiov);
1359 /* Complete request */
1360 if (acb->cur_pos >= acb->end_pos) {
1361 qed_aio_complete(acb, 0);
1362 return;
1365 /* Find next cluster and start I/O */
1366 qed_find_cluster(s, &acb->request,
1367 acb->cur_pos, acb->end_pos - acb->cur_pos,
1368 io_fn, acb);
1371 static BlockAIOCB *qed_aio_setup(BlockDriverState *bs,
1372 int64_t sector_num,
1373 QEMUIOVector *qiov, int nb_sectors,
1374 BlockCompletionFunc *cb,
1375 void *opaque, int flags)
1377 QEDAIOCB *acb = qemu_aio_get(&qed_aiocb_info, bs, cb, opaque);
1379 trace_qed_aio_setup(bs->opaque, acb, sector_num, nb_sectors,
1380 opaque, flags);
1382 acb->flags = flags;
1383 acb->qiov = qiov;
1384 acb->qiov_offset = 0;
1385 acb->cur_pos = (uint64_t)sector_num * BDRV_SECTOR_SIZE;
1386 acb->end_pos = acb->cur_pos + nb_sectors * BDRV_SECTOR_SIZE;
1387 acb->backing_qiov = NULL;
1388 acb->request.l2_table = NULL;
1389 qemu_iovec_init(&acb->cur_qiov, qiov->niov);
1391 /* Start request */
1392 qed_aio_next_io(acb, 0);
1393 return &acb->common;
1396 static BlockAIOCB *bdrv_qed_aio_readv(BlockDriverState *bs,
1397 int64_t sector_num,
1398 QEMUIOVector *qiov, int nb_sectors,
1399 BlockCompletionFunc *cb,
1400 void *opaque)
1402 return qed_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
1405 static BlockAIOCB *bdrv_qed_aio_writev(BlockDriverState *bs,
1406 int64_t sector_num,
1407 QEMUIOVector *qiov, int nb_sectors,
1408 BlockCompletionFunc *cb,
1409 void *opaque)
1411 return qed_aio_setup(bs, sector_num, qiov, nb_sectors, cb,
1412 opaque, QED_AIOCB_WRITE);
1415 typedef struct {
1416 Coroutine *co;
1417 int ret;
1418 bool done;
1419 } QEDWriteZeroesCB;
1421 static void coroutine_fn qed_co_write_zeroes_cb(void *opaque, int ret)
1423 QEDWriteZeroesCB *cb = opaque;
1425 cb->done = true;
1426 cb->ret = ret;
1427 if (cb->co) {
1428 qemu_coroutine_enter(cb->co, NULL);
1432 static int coroutine_fn bdrv_qed_co_write_zeroes(BlockDriverState *bs,
1433 int64_t sector_num,
1434 int nb_sectors,
1435 BdrvRequestFlags flags)
1437 BlockAIOCB *blockacb;
1438 BDRVQEDState *s = bs->opaque;
1439 QEDWriteZeroesCB cb = { .done = false };
1440 QEMUIOVector qiov;
1441 struct iovec iov;
1443 /* Refuse if there are untouched backing file sectors */
1444 if (bs->backing) {
1445 if (qed_offset_into_cluster(s, sector_num * BDRV_SECTOR_SIZE) != 0) {
1446 return -ENOTSUP;
1448 if (qed_offset_into_cluster(s, nb_sectors * BDRV_SECTOR_SIZE) != 0) {
1449 return -ENOTSUP;
1453 /* Zero writes start without an I/O buffer. If a buffer becomes necessary
1454 * then it will be allocated during request processing.
1456 iov.iov_base = NULL,
1457 iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE,
1459 qemu_iovec_init_external(&qiov, &iov, 1);
1460 blockacb = qed_aio_setup(bs, sector_num, &qiov, nb_sectors,
1461 qed_co_write_zeroes_cb, &cb,
1462 QED_AIOCB_WRITE | QED_AIOCB_ZERO);
1463 if (!blockacb) {
1464 return -EIO;
1466 if (!cb.done) {
1467 cb.co = qemu_coroutine_self();
1468 qemu_coroutine_yield();
1470 assert(cb.done);
1471 return cb.ret;
1474 static int bdrv_qed_truncate(BlockDriverState *bs, int64_t offset)
1476 BDRVQEDState *s = bs->opaque;
1477 uint64_t old_image_size;
1478 int ret;
1480 if (!qed_is_image_size_valid(offset, s->header.cluster_size,
1481 s->header.table_size)) {
1482 return -EINVAL;
1485 /* Shrinking is currently not supported */
1486 if ((uint64_t)offset < s->header.image_size) {
1487 return -ENOTSUP;
1490 old_image_size = s->header.image_size;
1491 s->header.image_size = offset;
1492 ret = qed_write_header_sync(s);
1493 if (ret < 0) {
1494 s->header.image_size = old_image_size;
1496 return ret;
1499 static int64_t bdrv_qed_getlength(BlockDriverState *bs)
1501 BDRVQEDState *s = bs->opaque;
1502 return s->header.image_size;
1505 static int bdrv_qed_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1507 BDRVQEDState *s = bs->opaque;
1509 memset(bdi, 0, sizeof(*bdi));
1510 bdi->cluster_size = s->header.cluster_size;
1511 bdi->is_dirty = s->header.features & QED_F_NEED_CHECK;
1512 bdi->unallocated_blocks_are_zero = true;
1513 bdi->can_write_zeroes_with_unmap = true;
1514 return 0;
1517 static int bdrv_qed_change_backing_file(BlockDriverState *bs,
1518 const char *backing_file,
1519 const char *backing_fmt)
1521 BDRVQEDState *s = bs->opaque;
1522 QEDHeader new_header, le_header;
1523 void *buffer;
1524 size_t buffer_len, backing_file_len;
1525 int ret;
1527 /* Refuse to set backing filename if unknown compat feature bits are
1528 * active. If the image uses an unknown compat feature then we may not
1529 * know the layout of data following the header structure and cannot safely
1530 * add a new string.
1532 if (backing_file && (s->header.compat_features &
1533 ~QED_COMPAT_FEATURE_MASK)) {
1534 return -ENOTSUP;
1537 memcpy(&new_header, &s->header, sizeof(new_header));
1539 new_header.features &= ~(QED_F_BACKING_FILE |
1540 QED_F_BACKING_FORMAT_NO_PROBE);
1542 /* Adjust feature flags */
1543 if (backing_file) {
1544 new_header.features |= QED_F_BACKING_FILE;
1546 if (qed_fmt_is_raw(backing_fmt)) {
1547 new_header.features |= QED_F_BACKING_FORMAT_NO_PROBE;
1551 /* Calculate new header size */
1552 backing_file_len = 0;
1554 if (backing_file) {
1555 backing_file_len = strlen(backing_file);
1558 buffer_len = sizeof(new_header);
1559 new_header.backing_filename_offset = buffer_len;
1560 new_header.backing_filename_size = backing_file_len;
1561 buffer_len += backing_file_len;
1563 /* Make sure we can rewrite header without failing */
1564 if (buffer_len > new_header.header_size * new_header.cluster_size) {
1565 return -ENOSPC;
1568 /* Prepare new header */
1569 buffer = g_malloc(buffer_len);
1571 qed_header_cpu_to_le(&new_header, &le_header);
1572 memcpy(buffer, &le_header, sizeof(le_header));
1573 buffer_len = sizeof(le_header);
1575 if (backing_file) {
1576 memcpy(buffer + buffer_len, backing_file, backing_file_len);
1577 buffer_len += backing_file_len;
1580 /* Write new header */
1581 ret = bdrv_pwrite_sync(bs->file->bs, 0, buffer, buffer_len);
1582 g_free(buffer);
1583 if (ret == 0) {
1584 memcpy(&s->header, &new_header, sizeof(new_header));
1586 return ret;
1589 static void bdrv_qed_invalidate_cache(BlockDriverState *bs, Error **errp)
1591 BDRVQEDState *s = bs->opaque;
1592 Error *local_err = NULL;
1593 int ret;
1595 bdrv_qed_close(bs);
1597 bdrv_invalidate_cache(bs->file->bs, &local_err);
1598 if (local_err) {
1599 error_propagate(errp, local_err);
1600 return;
1603 memset(s, 0, sizeof(BDRVQEDState));
1604 ret = bdrv_qed_open(bs, NULL, bs->open_flags, &local_err);
1605 if (local_err) {
1606 error_propagate(errp, local_err);
1607 error_prepend(errp, "Could not reopen qed layer: ");
1608 return;
1609 } else if (ret < 0) {
1610 error_setg_errno(errp, -ret, "Could not reopen qed layer");
1611 return;
1615 static int bdrv_qed_check(BlockDriverState *bs, BdrvCheckResult *result,
1616 BdrvCheckMode fix)
1618 BDRVQEDState *s = bs->opaque;
1620 return qed_check(s, result, !!fix);
1623 static QemuOptsList qed_create_opts = {
1624 .name = "qed-create-opts",
1625 .head = QTAILQ_HEAD_INITIALIZER(qed_create_opts.head),
1626 .desc = {
1628 .name = BLOCK_OPT_SIZE,
1629 .type = QEMU_OPT_SIZE,
1630 .help = "Virtual disk size"
1633 .name = BLOCK_OPT_BACKING_FILE,
1634 .type = QEMU_OPT_STRING,
1635 .help = "File name of a base image"
1638 .name = BLOCK_OPT_BACKING_FMT,
1639 .type = QEMU_OPT_STRING,
1640 .help = "Image format of the base image"
1643 .name = BLOCK_OPT_CLUSTER_SIZE,
1644 .type = QEMU_OPT_SIZE,
1645 .help = "Cluster size (in bytes)",
1646 .def_value_str = stringify(QED_DEFAULT_CLUSTER_SIZE)
1649 .name = BLOCK_OPT_TABLE_SIZE,
1650 .type = QEMU_OPT_SIZE,
1651 .help = "L1/L2 table size (in clusters)"
1653 { /* end of list */ }
1657 static BlockDriver bdrv_qed = {
1658 .format_name = "qed",
1659 .instance_size = sizeof(BDRVQEDState),
1660 .create_opts = &qed_create_opts,
1661 .supports_backing = true,
1663 .bdrv_probe = bdrv_qed_probe,
1664 .bdrv_open = bdrv_qed_open,
1665 .bdrv_close = bdrv_qed_close,
1666 .bdrv_reopen_prepare = bdrv_qed_reopen_prepare,
1667 .bdrv_create = bdrv_qed_create,
1668 .bdrv_has_zero_init = bdrv_has_zero_init_1,
1669 .bdrv_co_get_block_status = bdrv_qed_co_get_block_status,
1670 .bdrv_aio_readv = bdrv_qed_aio_readv,
1671 .bdrv_aio_writev = bdrv_qed_aio_writev,
1672 .bdrv_co_write_zeroes = bdrv_qed_co_write_zeroes,
1673 .bdrv_truncate = bdrv_qed_truncate,
1674 .bdrv_getlength = bdrv_qed_getlength,
1675 .bdrv_get_info = bdrv_qed_get_info,
1676 .bdrv_refresh_limits = bdrv_qed_refresh_limits,
1677 .bdrv_change_backing_file = bdrv_qed_change_backing_file,
1678 .bdrv_invalidate_cache = bdrv_qed_invalidate_cache,
1679 .bdrv_check = bdrv_qed_check,
1680 .bdrv_detach_aio_context = bdrv_qed_detach_aio_context,
1681 .bdrv_attach_aio_context = bdrv_qed_attach_aio_context,
1684 static void bdrv_qed_init(void)
1686 bdrv_register(&bdrv_qed);
1689 block_init(bdrv_qed_init);