ivshmem: Fix unplug of device "ivshmem-plain"
[qemu/armbru.git] / block / qed.c
blob689ea9d4d5be5d5d3b0a0a45507c825572f5b06d
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 "block/qdict.h"
17 #include "qapi/error.h"
18 #include "qemu/timer.h"
19 #include "qemu/bswap.h"
20 #include "qemu/option.h"
21 #include "trace.h"
22 #include "qed.h"
23 #include "sysemu/block-backend.h"
24 #include "qapi/qmp/qdict.h"
25 #include "qapi/qobject-input-visitor.h"
26 #include "qapi/qapi-visit-block-core.h"
28 static QemuOptsList qed_create_opts;
30 static int bdrv_qed_probe(const uint8_t *buf, int buf_size,
31 const char *filename)
33 const QEDHeader *header = (const QEDHeader *)buf;
35 if (buf_size < sizeof(*header)) {
36 return 0;
38 if (le32_to_cpu(header->magic) != QED_MAGIC) {
39 return 0;
41 return 100;
44 /**
45 * Check whether an image format is raw
47 * @fmt: Backing file format, may be NULL
49 static bool qed_fmt_is_raw(const char *fmt)
51 return fmt && strcmp(fmt, "raw") == 0;
54 static void qed_header_le_to_cpu(const QEDHeader *le, QEDHeader *cpu)
56 cpu->magic = le32_to_cpu(le->magic);
57 cpu->cluster_size = le32_to_cpu(le->cluster_size);
58 cpu->table_size = le32_to_cpu(le->table_size);
59 cpu->header_size = le32_to_cpu(le->header_size);
60 cpu->features = le64_to_cpu(le->features);
61 cpu->compat_features = le64_to_cpu(le->compat_features);
62 cpu->autoclear_features = le64_to_cpu(le->autoclear_features);
63 cpu->l1_table_offset = le64_to_cpu(le->l1_table_offset);
64 cpu->image_size = le64_to_cpu(le->image_size);
65 cpu->backing_filename_offset = le32_to_cpu(le->backing_filename_offset);
66 cpu->backing_filename_size = le32_to_cpu(le->backing_filename_size);
69 static void qed_header_cpu_to_le(const QEDHeader *cpu, QEDHeader *le)
71 le->magic = cpu_to_le32(cpu->magic);
72 le->cluster_size = cpu_to_le32(cpu->cluster_size);
73 le->table_size = cpu_to_le32(cpu->table_size);
74 le->header_size = cpu_to_le32(cpu->header_size);
75 le->features = cpu_to_le64(cpu->features);
76 le->compat_features = cpu_to_le64(cpu->compat_features);
77 le->autoclear_features = cpu_to_le64(cpu->autoclear_features);
78 le->l1_table_offset = cpu_to_le64(cpu->l1_table_offset);
79 le->image_size = cpu_to_le64(cpu->image_size);
80 le->backing_filename_offset = cpu_to_le32(cpu->backing_filename_offset);
81 le->backing_filename_size = cpu_to_le32(cpu->backing_filename_size);
84 int qed_write_header_sync(BDRVQEDState *s)
86 QEDHeader le;
87 int ret;
89 qed_header_cpu_to_le(&s->header, &le);
90 ret = bdrv_pwrite(s->bs->file, 0, &le, sizeof(le));
91 if (ret != sizeof(le)) {
92 return ret;
94 return 0;
97 /**
98 * Update header in-place (does not rewrite backing filename or other strings)
100 * This function only updates known header fields in-place and does not affect
101 * extra data after the QED header.
103 * No new allocating reqs can start while this function runs.
105 static int coroutine_fn qed_write_header(BDRVQEDState *s)
107 /* We must write full sectors for O_DIRECT but cannot necessarily generate
108 * the data following the header if an unrecognized compat feature is
109 * active. Therefore, first read the sectors containing the header, update
110 * them, and write back.
113 int nsectors = DIV_ROUND_UP(sizeof(QEDHeader), BDRV_SECTOR_SIZE);
114 size_t len = nsectors * BDRV_SECTOR_SIZE;
115 uint8_t *buf;
116 struct iovec iov;
117 QEMUIOVector qiov;
118 int ret;
120 assert(s->allocating_acb || s->allocating_write_reqs_plugged);
122 buf = qemu_blockalign(s->bs, len);
123 iov = (struct iovec) {
124 .iov_base = buf,
125 .iov_len = len,
127 qemu_iovec_init_external(&qiov, &iov, 1);
129 ret = bdrv_co_preadv(s->bs->file, 0, qiov.size, &qiov, 0);
130 if (ret < 0) {
131 goto out;
134 /* Update header */
135 qed_header_cpu_to_le(&s->header, (QEDHeader *) buf);
137 ret = bdrv_co_pwritev(s->bs->file, 0, qiov.size, &qiov, 0);
138 if (ret < 0) {
139 goto out;
142 ret = 0;
143 out:
144 qemu_vfree(buf);
145 return ret;
148 static uint64_t qed_max_image_size(uint32_t cluster_size, uint32_t table_size)
150 uint64_t table_entries;
151 uint64_t l2_size;
153 table_entries = (table_size * cluster_size) / sizeof(uint64_t);
154 l2_size = table_entries * cluster_size;
156 return l2_size * table_entries;
159 static bool qed_is_cluster_size_valid(uint32_t cluster_size)
161 if (cluster_size < QED_MIN_CLUSTER_SIZE ||
162 cluster_size > QED_MAX_CLUSTER_SIZE) {
163 return false;
165 if (cluster_size & (cluster_size - 1)) {
166 return false; /* not power of 2 */
168 return true;
171 static bool qed_is_table_size_valid(uint32_t table_size)
173 if (table_size < QED_MIN_TABLE_SIZE ||
174 table_size > QED_MAX_TABLE_SIZE) {
175 return false;
177 if (table_size & (table_size - 1)) {
178 return false; /* not power of 2 */
180 return true;
183 static bool qed_is_image_size_valid(uint64_t image_size, uint32_t cluster_size,
184 uint32_t table_size)
186 if (image_size % BDRV_SECTOR_SIZE != 0) {
187 return false; /* not multiple of sector size */
189 if (image_size > qed_max_image_size(cluster_size, table_size)) {
190 return false; /* image is too large */
192 return true;
196 * Read a string of known length from the image file
198 * @file: Image file
199 * @offset: File offset to start of string, in bytes
200 * @n: String length in bytes
201 * @buf: Destination buffer
202 * @buflen: Destination buffer length in bytes
203 * @ret: 0 on success, -errno on failure
205 * The string is NUL-terminated.
207 static int qed_read_string(BdrvChild *file, uint64_t offset, size_t n,
208 char *buf, size_t buflen)
210 int ret;
211 if (n >= buflen) {
212 return -EINVAL;
214 ret = bdrv_pread(file, offset, buf, n);
215 if (ret < 0) {
216 return ret;
218 buf[n] = '\0';
219 return 0;
223 * Allocate new clusters
225 * @s: QED state
226 * @n: Number of contiguous clusters to allocate
227 * @ret: Offset of first allocated cluster
229 * This function only produces the offset where the new clusters should be
230 * written. It updates BDRVQEDState but does not make any changes to the image
231 * file.
233 * Called with table_lock held.
235 static uint64_t qed_alloc_clusters(BDRVQEDState *s, unsigned int n)
237 uint64_t offset = s->file_size;
238 s->file_size += n * s->header.cluster_size;
239 return offset;
242 QEDTable *qed_alloc_table(BDRVQEDState *s)
244 /* Honor O_DIRECT memory alignment requirements */
245 return qemu_blockalign(s->bs,
246 s->header.cluster_size * s->header.table_size);
250 * Allocate a new zeroed L2 table
252 * Called with table_lock held.
254 static CachedL2Table *qed_new_l2_table(BDRVQEDState *s)
256 CachedL2Table *l2_table = qed_alloc_l2_cache_entry(&s->l2_cache);
258 l2_table->table = qed_alloc_table(s);
259 l2_table->offset = qed_alloc_clusters(s, s->header.table_size);
261 memset(l2_table->table->offsets, 0,
262 s->header.cluster_size * s->header.table_size);
263 return l2_table;
266 static bool qed_plug_allocating_write_reqs(BDRVQEDState *s)
268 qemu_co_mutex_lock(&s->table_lock);
270 /* No reentrancy is allowed. */
271 assert(!s->allocating_write_reqs_plugged);
272 if (s->allocating_acb != NULL) {
273 /* Another allocating write came concurrently. This cannot happen
274 * from bdrv_qed_co_drain_begin, but it can happen when the timer runs.
276 qemu_co_mutex_unlock(&s->table_lock);
277 return false;
280 s->allocating_write_reqs_plugged = true;
281 qemu_co_mutex_unlock(&s->table_lock);
282 return true;
285 static void qed_unplug_allocating_write_reqs(BDRVQEDState *s)
287 qemu_co_mutex_lock(&s->table_lock);
288 assert(s->allocating_write_reqs_plugged);
289 s->allocating_write_reqs_plugged = false;
290 qemu_co_queue_next(&s->allocating_write_reqs);
291 qemu_co_mutex_unlock(&s->table_lock);
294 static void coroutine_fn qed_need_check_timer_entry(void *opaque)
296 BDRVQEDState *s = opaque;
297 int ret;
299 trace_qed_need_check_timer_cb(s);
301 if (!qed_plug_allocating_write_reqs(s)) {
302 return;
305 /* Ensure writes are on disk before clearing flag */
306 ret = bdrv_co_flush(s->bs->file->bs);
307 if (ret < 0) {
308 qed_unplug_allocating_write_reqs(s);
309 return;
312 s->header.features &= ~QED_F_NEED_CHECK;
313 ret = qed_write_header(s);
314 (void) ret;
316 qed_unplug_allocating_write_reqs(s);
318 ret = bdrv_co_flush(s->bs);
319 (void) ret;
322 static void qed_need_check_timer_cb(void *opaque)
324 Coroutine *co = qemu_coroutine_create(qed_need_check_timer_entry, opaque);
325 qemu_coroutine_enter(co);
328 static void qed_start_need_check_timer(BDRVQEDState *s)
330 trace_qed_start_need_check_timer(s);
332 /* Use QEMU_CLOCK_VIRTUAL so we don't alter the image file while suspended for
333 * migration.
335 timer_mod(s->need_check_timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) +
336 NANOSECONDS_PER_SECOND * QED_NEED_CHECK_TIMEOUT);
339 /* It's okay to call this multiple times or when no timer is started */
340 static void qed_cancel_need_check_timer(BDRVQEDState *s)
342 trace_qed_cancel_need_check_timer(s);
343 timer_del(s->need_check_timer);
346 static void bdrv_qed_detach_aio_context(BlockDriverState *bs)
348 BDRVQEDState *s = bs->opaque;
350 qed_cancel_need_check_timer(s);
351 timer_free(s->need_check_timer);
354 static void bdrv_qed_attach_aio_context(BlockDriverState *bs,
355 AioContext *new_context)
357 BDRVQEDState *s = bs->opaque;
359 s->need_check_timer = aio_timer_new(new_context,
360 QEMU_CLOCK_VIRTUAL, SCALE_NS,
361 qed_need_check_timer_cb, s);
362 if (s->header.features & QED_F_NEED_CHECK) {
363 qed_start_need_check_timer(s);
367 static void coroutine_fn bdrv_qed_co_drain_begin(BlockDriverState *bs)
369 BDRVQEDState *s = bs->opaque;
371 /* Fire the timer immediately in order to start doing I/O as soon as the
372 * header is flushed.
374 if (s->need_check_timer && timer_pending(s->need_check_timer)) {
375 qed_cancel_need_check_timer(s);
376 qed_need_check_timer_entry(s);
380 static void bdrv_qed_init_state(BlockDriverState *bs)
382 BDRVQEDState *s = bs->opaque;
384 memset(s, 0, sizeof(BDRVQEDState));
385 s->bs = bs;
386 qemu_co_mutex_init(&s->table_lock);
387 qemu_co_queue_init(&s->allocating_write_reqs);
390 /* Called with table_lock held. */
391 static int coroutine_fn bdrv_qed_do_open(BlockDriverState *bs, QDict *options,
392 int flags, Error **errp)
394 BDRVQEDState *s = bs->opaque;
395 QEDHeader le_header;
396 int64_t file_size;
397 int ret;
399 ret = bdrv_pread(bs->file, 0, &le_header, sizeof(le_header));
400 if (ret < 0) {
401 return ret;
403 qed_header_le_to_cpu(&le_header, &s->header);
405 if (s->header.magic != QED_MAGIC) {
406 error_setg(errp, "Image not in QED format");
407 return -EINVAL;
409 if (s->header.features & ~QED_FEATURE_MASK) {
410 /* image uses unsupported feature bits */
411 error_setg(errp, "Unsupported QED features: %" PRIx64,
412 s->header.features & ~QED_FEATURE_MASK);
413 return -ENOTSUP;
415 if (!qed_is_cluster_size_valid(s->header.cluster_size)) {
416 return -EINVAL;
419 /* Round down file size to the last cluster */
420 file_size = bdrv_getlength(bs->file->bs);
421 if (file_size < 0) {
422 return file_size;
424 s->file_size = qed_start_of_cluster(s, file_size);
426 if (!qed_is_table_size_valid(s->header.table_size)) {
427 return -EINVAL;
429 if (!qed_is_image_size_valid(s->header.image_size,
430 s->header.cluster_size,
431 s->header.table_size)) {
432 return -EINVAL;
434 if (!qed_check_table_offset(s, s->header.l1_table_offset)) {
435 return -EINVAL;
438 s->table_nelems = (s->header.cluster_size * s->header.table_size) /
439 sizeof(uint64_t);
440 s->l2_shift = ctz32(s->header.cluster_size);
441 s->l2_mask = s->table_nelems - 1;
442 s->l1_shift = s->l2_shift + ctz32(s->table_nelems);
444 /* Header size calculation must not overflow uint32_t */
445 if (s->header.header_size > UINT32_MAX / s->header.cluster_size) {
446 return -EINVAL;
449 if ((s->header.features & QED_F_BACKING_FILE)) {
450 if ((uint64_t)s->header.backing_filename_offset +
451 s->header.backing_filename_size >
452 s->header.cluster_size * s->header.header_size) {
453 return -EINVAL;
456 ret = qed_read_string(bs->file, s->header.backing_filename_offset,
457 s->header.backing_filename_size, bs->backing_file,
458 sizeof(bs->backing_file));
459 if (ret < 0) {
460 return ret;
463 if (s->header.features & QED_F_BACKING_FORMAT_NO_PROBE) {
464 pstrcpy(bs->backing_format, sizeof(bs->backing_format), "raw");
468 /* Reset unknown autoclear feature bits. This is a backwards
469 * compatibility mechanism that allows images to be opened by older
470 * programs, which "knock out" unknown feature bits. When an image is
471 * opened by a newer program again it can detect that the autoclear
472 * feature is no longer valid.
474 if ((s->header.autoclear_features & ~QED_AUTOCLEAR_FEATURE_MASK) != 0 &&
475 !bdrv_is_read_only(bs->file->bs) && !(flags & BDRV_O_INACTIVE)) {
476 s->header.autoclear_features &= QED_AUTOCLEAR_FEATURE_MASK;
478 ret = qed_write_header_sync(s);
479 if (ret) {
480 return ret;
483 /* From here on only known autoclear feature bits are valid */
484 bdrv_flush(bs->file->bs);
487 s->l1_table = qed_alloc_table(s);
488 qed_init_l2_cache(&s->l2_cache);
490 ret = qed_read_l1_table_sync(s);
491 if (ret) {
492 goto out;
495 /* If image was not closed cleanly, check consistency */
496 if (!(flags & BDRV_O_CHECK) && (s->header.features & QED_F_NEED_CHECK)) {
497 /* Read-only images cannot be fixed. There is no risk of corruption
498 * since write operations are not possible. Therefore, allow
499 * potentially inconsistent images to be opened read-only. This can
500 * aid data recovery from an otherwise inconsistent image.
502 if (!bdrv_is_read_only(bs->file->bs) &&
503 !(flags & BDRV_O_INACTIVE)) {
504 BdrvCheckResult result = {0};
506 ret = qed_check(s, &result, true);
507 if (ret) {
508 goto out;
513 bdrv_qed_attach_aio_context(bs, bdrv_get_aio_context(bs));
515 out:
516 if (ret) {
517 qed_free_l2_cache(&s->l2_cache);
518 qemu_vfree(s->l1_table);
520 return ret;
523 typedef struct QEDOpenCo {
524 BlockDriverState *bs;
525 QDict *options;
526 int flags;
527 Error **errp;
528 int ret;
529 } QEDOpenCo;
531 static void coroutine_fn bdrv_qed_open_entry(void *opaque)
533 QEDOpenCo *qoc = opaque;
534 BDRVQEDState *s = qoc->bs->opaque;
536 qemu_co_mutex_lock(&s->table_lock);
537 qoc->ret = bdrv_qed_do_open(qoc->bs, qoc->options, qoc->flags, qoc->errp);
538 qemu_co_mutex_unlock(&s->table_lock);
541 static int bdrv_qed_open(BlockDriverState *bs, QDict *options, int flags,
542 Error **errp)
544 QEDOpenCo qoc = {
545 .bs = bs,
546 .options = options,
547 .flags = flags,
548 .errp = errp,
549 .ret = -EINPROGRESS
552 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
553 false, errp);
554 if (!bs->file) {
555 return -EINVAL;
558 bdrv_qed_init_state(bs);
559 if (qemu_in_coroutine()) {
560 bdrv_qed_open_entry(&qoc);
561 } else {
562 qemu_coroutine_enter(qemu_coroutine_create(bdrv_qed_open_entry, &qoc));
563 BDRV_POLL_WHILE(bs, qoc.ret == -EINPROGRESS);
565 BDRV_POLL_WHILE(bs, qoc.ret == -EINPROGRESS);
566 return qoc.ret;
569 static void bdrv_qed_refresh_limits(BlockDriverState *bs, Error **errp)
571 BDRVQEDState *s = bs->opaque;
573 bs->bl.pwrite_zeroes_alignment = s->header.cluster_size;
576 /* We have nothing to do for QED reopen, stubs just return
577 * success */
578 static int bdrv_qed_reopen_prepare(BDRVReopenState *state,
579 BlockReopenQueue *queue, Error **errp)
581 return 0;
584 static void bdrv_qed_close(BlockDriverState *bs)
586 BDRVQEDState *s = bs->opaque;
588 bdrv_qed_detach_aio_context(bs);
590 /* Ensure writes reach stable storage */
591 bdrv_flush(bs->file->bs);
593 /* Clean shutdown, no check required on next open */
594 if (s->header.features & QED_F_NEED_CHECK) {
595 s->header.features &= ~QED_F_NEED_CHECK;
596 qed_write_header_sync(s);
599 qed_free_l2_cache(&s->l2_cache);
600 qemu_vfree(s->l1_table);
603 static int coroutine_fn bdrv_qed_co_create(BlockdevCreateOptions *opts,
604 Error **errp)
606 BlockdevCreateOptionsQed *qed_opts;
607 BlockBackend *blk = NULL;
608 BlockDriverState *bs = NULL;
610 QEDHeader header;
611 QEDHeader le_header;
612 uint8_t *l1_table = NULL;
613 size_t l1_size;
614 int ret = 0;
616 assert(opts->driver == BLOCKDEV_DRIVER_QED);
617 qed_opts = &opts->u.qed;
619 /* Validate options and set default values */
620 if (!qed_opts->has_cluster_size) {
621 qed_opts->cluster_size = QED_DEFAULT_CLUSTER_SIZE;
623 if (!qed_opts->has_table_size) {
624 qed_opts->table_size = QED_DEFAULT_TABLE_SIZE;
627 if (!qed_is_cluster_size_valid(qed_opts->cluster_size)) {
628 error_setg(errp, "QED cluster size must be within range [%u, %u] "
629 "and power of 2",
630 QED_MIN_CLUSTER_SIZE, QED_MAX_CLUSTER_SIZE);
631 return -EINVAL;
633 if (!qed_is_table_size_valid(qed_opts->table_size)) {
634 error_setg(errp, "QED table size must be within range [%u, %u] "
635 "and power of 2",
636 QED_MIN_TABLE_SIZE, QED_MAX_TABLE_SIZE);
637 return -EINVAL;
639 if (!qed_is_image_size_valid(qed_opts->size, qed_opts->cluster_size,
640 qed_opts->table_size))
642 error_setg(errp, "QED image size must be a non-zero multiple of "
643 "cluster size and less than %" PRIu64 " bytes",
644 qed_max_image_size(qed_opts->cluster_size,
645 qed_opts->table_size));
646 return -EINVAL;
649 /* Create BlockBackend to write to the image */
650 bs = bdrv_open_blockdev_ref(qed_opts->file, errp);
651 if (bs == NULL) {
652 return -EIO;
655 blk = blk_new(BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL);
656 ret = blk_insert_bs(blk, bs, errp);
657 if (ret < 0) {
658 goto out;
660 blk_set_allow_write_beyond_eof(blk, true);
662 /* Prepare image format */
663 header = (QEDHeader) {
664 .magic = QED_MAGIC,
665 .cluster_size = qed_opts->cluster_size,
666 .table_size = qed_opts->table_size,
667 .header_size = 1,
668 .features = 0,
669 .compat_features = 0,
670 .l1_table_offset = qed_opts->cluster_size,
671 .image_size = qed_opts->size,
674 l1_size = header.cluster_size * header.table_size;
676 /* File must start empty and grow, check truncate is supported */
677 ret = blk_truncate(blk, 0, PREALLOC_MODE_OFF, errp);
678 if (ret < 0) {
679 goto out;
682 if (qed_opts->has_backing_file) {
683 header.features |= QED_F_BACKING_FILE;
684 header.backing_filename_offset = sizeof(le_header);
685 header.backing_filename_size = strlen(qed_opts->backing_file);
687 if (qed_opts->has_backing_fmt) {
688 const char *backing_fmt = BlockdevDriver_str(qed_opts->backing_fmt);
689 if (qed_fmt_is_raw(backing_fmt)) {
690 header.features |= QED_F_BACKING_FORMAT_NO_PROBE;
695 qed_header_cpu_to_le(&header, &le_header);
696 ret = blk_pwrite(blk, 0, &le_header, sizeof(le_header), 0);
697 if (ret < 0) {
698 goto out;
700 ret = blk_pwrite(blk, sizeof(le_header), qed_opts->backing_file,
701 header.backing_filename_size, 0);
702 if (ret < 0) {
703 goto out;
706 l1_table = g_malloc0(l1_size);
707 ret = blk_pwrite(blk, header.l1_table_offset, l1_table, l1_size, 0);
708 if (ret < 0) {
709 goto out;
712 ret = 0; /* success */
713 out:
714 g_free(l1_table);
715 blk_unref(blk);
716 bdrv_unref(bs);
717 return ret;
720 static int coroutine_fn bdrv_qed_co_create_opts(const char *filename,
721 QemuOpts *opts,
722 Error **errp)
724 BlockdevCreateOptions *create_options = NULL;
725 QDict *qdict;
726 Visitor *v;
727 BlockDriverState *bs = NULL;
728 Error *local_err = NULL;
729 int ret;
731 static const QDictRenames opt_renames[] = {
732 { BLOCK_OPT_BACKING_FILE, "backing-file" },
733 { BLOCK_OPT_BACKING_FMT, "backing-fmt" },
734 { BLOCK_OPT_CLUSTER_SIZE, "cluster-size" },
735 { BLOCK_OPT_TABLE_SIZE, "table-size" },
736 { NULL, NULL },
739 /* Parse options and convert legacy syntax */
740 qdict = qemu_opts_to_qdict_filtered(opts, NULL, &qed_create_opts, true);
742 if (!qdict_rename_keys(qdict, opt_renames, errp)) {
743 ret = -EINVAL;
744 goto fail;
747 /* Create and open the file (protocol layer) */
748 ret = bdrv_create_file(filename, opts, &local_err);
749 if (ret < 0) {
750 error_propagate(errp, local_err);
751 goto fail;
754 bs = bdrv_open(filename, NULL, NULL,
755 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
756 if (bs == NULL) {
757 ret = -EIO;
758 goto fail;
761 /* Now get the QAPI type BlockdevCreateOptions */
762 qdict_put_str(qdict, "driver", "qed");
763 qdict_put_str(qdict, "file", bs->node_name);
765 v = qobject_input_visitor_new_flat_confused(qdict, errp);
766 if (!v) {
767 ret = -EINVAL;
768 goto fail;
771 visit_type_BlockdevCreateOptions(v, NULL, &create_options, &local_err);
772 visit_free(v);
774 if (local_err) {
775 error_propagate(errp, local_err);
776 ret = -EINVAL;
777 goto fail;
780 /* Silently round up size */
781 assert(create_options->driver == BLOCKDEV_DRIVER_QED);
782 create_options->u.qed.size =
783 ROUND_UP(create_options->u.qed.size, BDRV_SECTOR_SIZE);
785 /* Create the qed image (format layer) */
786 ret = bdrv_qed_co_create(create_options, errp);
788 fail:
789 qobject_unref(qdict);
790 bdrv_unref(bs);
791 qapi_free_BlockdevCreateOptions(create_options);
792 return ret;
795 static int coroutine_fn bdrv_qed_co_block_status(BlockDriverState *bs,
796 bool want_zero,
797 int64_t pos, int64_t bytes,
798 int64_t *pnum, int64_t *map,
799 BlockDriverState **file)
801 BDRVQEDState *s = bs->opaque;
802 size_t len = MIN(bytes, SIZE_MAX);
803 int status;
804 QEDRequest request = { .l2_table = NULL };
805 uint64_t offset;
806 int ret;
808 qemu_co_mutex_lock(&s->table_lock);
809 ret = qed_find_cluster(s, &request, pos, &len, &offset);
811 *pnum = len;
812 switch (ret) {
813 case QED_CLUSTER_FOUND:
814 *map = offset | qed_offset_into_cluster(s, pos);
815 status = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
816 *file = bs->file->bs;
817 break;
818 case QED_CLUSTER_ZERO:
819 status = BDRV_BLOCK_ZERO;
820 break;
821 case QED_CLUSTER_L2:
822 case QED_CLUSTER_L1:
823 status = 0;
824 break;
825 default:
826 assert(ret < 0);
827 status = ret;
828 break;
831 qed_unref_l2_cache_entry(request.l2_table);
832 qemu_co_mutex_unlock(&s->table_lock);
834 return status;
837 static BDRVQEDState *acb_to_s(QEDAIOCB *acb)
839 return acb->bs->opaque;
843 * Read from the backing file or zero-fill if no backing file
845 * @s: QED state
846 * @pos: Byte position in device
847 * @qiov: Destination I/O vector
848 * @backing_qiov: Possibly shortened copy of qiov, to be allocated here
849 * @cb: Completion function
850 * @opaque: User data for completion function
852 * This function reads qiov->size bytes starting at pos from the backing file.
853 * If there is no backing file then zeroes are read.
855 static int coroutine_fn qed_read_backing_file(BDRVQEDState *s, uint64_t pos,
856 QEMUIOVector *qiov,
857 QEMUIOVector **backing_qiov)
859 uint64_t backing_length = 0;
860 size_t size;
861 int ret;
863 /* If there is a backing file, get its length. Treat the absence of a
864 * backing file like a zero length backing file.
866 if (s->bs->backing) {
867 int64_t l = bdrv_getlength(s->bs->backing->bs);
868 if (l < 0) {
869 return l;
871 backing_length = l;
874 /* Zero all sectors if reading beyond the end of the backing file */
875 if (pos >= backing_length ||
876 pos + qiov->size > backing_length) {
877 qemu_iovec_memset(qiov, 0, 0, qiov->size);
880 /* Complete now if there are no backing file sectors to read */
881 if (pos >= backing_length) {
882 return 0;
885 /* If the read straddles the end of the backing file, shorten it */
886 size = MIN((uint64_t)backing_length - pos, qiov->size);
888 assert(*backing_qiov == NULL);
889 *backing_qiov = g_new(QEMUIOVector, 1);
890 qemu_iovec_init(*backing_qiov, qiov->niov);
891 qemu_iovec_concat(*backing_qiov, qiov, 0, size);
893 BLKDBG_EVENT(s->bs->file, BLKDBG_READ_BACKING_AIO);
894 ret = bdrv_co_preadv(s->bs->backing, pos, size, *backing_qiov, 0);
895 if (ret < 0) {
896 return ret;
898 return 0;
902 * Copy data from backing file into the image
904 * @s: QED state
905 * @pos: Byte position in device
906 * @len: Number of bytes
907 * @offset: Byte offset in image file
909 static int coroutine_fn qed_copy_from_backing_file(BDRVQEDState *s,
910 uint64_t pos, uint64_t len,
911 uint64_t offset)
913 QEMUIOVector qiov;
914 QEMUIOVector *backing_qiov = NULL;
915 struct iovec iov;
916 int ret;
918 /* Skip copy entirely if there is no work to do */
919 if (len == 0) {
920 return 0;
923 iov = (struct iovec) {
924 .iov_base = qemu_blockalign(s->bs, len),
925 .iov_len = len,
927 qemu_iovec_init_external(&qiov, &iov, 1);
929 ret = qed_read_backing_file(s, pos, &qiov, &backing_qiov);
931 if (backing_qiov) {
932 qemu_iovec_destroy(backing_qiov);
933 g_free(backing_qiov);
934 backing_qiov = NULL;
937 if (ret) {
938 goto out;
941 BLKDBG_EVENT(s->bs->file, BLKDBG_COW_WRITE);
942 ret = bdrv_co_pwritev(s->bs->file, offset, qiov.size, &qiov, 0);
943 if (ret < 0) {
944 goto out;
946 ret = 0;
947 out:
948 qemu_vfree(iov.iov_base);
949 return ret;
953 * Link one or more contiguous clusters into a table
955 * @s: QED state
956 * @table: L2 table
957 * @index: First cluster index
958 * @n: Number of contiguous clusters
959 * @cluster: First cluster offset
961 * The cluster offset may be an allocated byte offset in the image file, the
962 * zero cluster marker, or the unallocated cluster marker.
964 * Called with table_lock held.
966 static void coroutine_fn qed_update_l2_table(BDRVQEDState *s, QEDTable *table,
967 int index, unsigned int n,
968 uint64_t cluster)
970 int i;
971 for (i = index; i < index + n; i++) {
972 table->offsets[i] = cluster;
973 if (!qed_offset_is_unalloc_cluster(cluster) &&
974 !qed_offset_is_zero_cluster(cluster)) {
975 cluster += s->header.cluster_size;
980 /* Called with table_lock held. */
981 static void coroutine_fn qed_aio_complete(QEDAIOCB *acb)
983 BDRVQEDState *s = acb_to_s(acb);
985 /* Free resources */
986 qemu_iovec_destroy(&acb->cur_qiov);
987 qed_unref_l2_cache_entry(acb->request.l2_table);
989 /* Free the buffer we may have allocated for zero writes */
990 if (acb->flags & QED_AIOCB_ZERO) {
991 qemu_vfree(acb->qiov->iov[0].iov_base);
992 acb->qiov->iov[0].iov_base = NULL;
995 /* Start next allocating write request waiting behind this one. Note that
996 * requests enqueue themselves when they first hit an unallocated cluster
997 * but they wait until the entire request is finished before waking up the
998 * next request in the queue. This ensures that we don't cycle through
999 * requests multiple times but rather finish one at a time completely.
1001 if (acb == s->allocating_acb) {
1002 s->allocating_acb = NULL;
1003 if (!qemu_co_queue_empty(&s->allocating_write_reqs)) {
1004 qemu_co_queue_next(&s->allocating_write_reqs);
1005 } else if (s->header.features & QED_F_NEED_CHECK) {
1006 qed_start_need_check_timer(s);
1012 * Update L1 table with new L2 table offset and write it out
1014 * Called with table_lock held.
1016 static int coroutine_fn qed_aio_write_l1_update(QEDAIOCB *acb)
1018 BDRVQEDState *s = acb_to_s(acb);
1019 CachedL2Table *l2_table = acb->request.l2_table;
1020 uint64_t l2_offset = l2_table->offset;
1021 int index, ret;
1023 index = qed_l1_index(s, acb->cur_pos);
1024 s->l1_table->offsets[index] = l2_table->offset;
1026 ret = qed_write_l1_table(s, index, 1);
1028 /* Commit the current L2 table to the cache */
1029 qed_commit_l2_cache_entry(&s->l2_cache, l2_table);
1031 /* This is guaranteed to succeed because we just committed the entry to the
1032 * cache.
1034 acb->request.l2_table = qed_find_l2_cache_entry(&s->l2_cache, l2_offset);
1035 assert(acb->request.l2_table != NULL);
1037 return ret;
1042 * Update L2 table with new cluster offsets and write them out
1044 * Called with table_lock held.
1046 static int coroutine_fn qed_aio_write_l2_update(QEDAIOCB *acb, uint64_t offset)
1048 BDRVQEDState *s = acb_to_s(acb);
1049 bool need_alloc = acb->find_cluster_ret == QED_CLUSTER_L1;
1050 int index, ret;
1052 if (need_alloc) {
1053 qed_unref_l2_cache_entry(acb->request.l2_table);
1054 acb->request.l2_table = qed_new_l2_table(s);
1057 index = qed_l2_index(s, acb->cur_pos);
1058 qed_update_l2_table(s, acb->request.l2_table->table, index, acb->cur_nclusters,
1059 offset);
1061 if (need_alloc) {
1062 /* Write out the whole new L2 table */
1063 ret = qed_write_l2_table(s, &acb->request, 0, s->table_nelems, true);
1064 if (ret) {
1065 return ret;
1067 return qed_aio_write_l1_update(acb);
1068 } else {
1069 /* Write out only the updated part of the L2 table */
1070 ret = qed_write_l2_table(s, &acb->request, index, acb->cur_nclusters,
1071 false);
1072 if (ret) {
1073 return ret;
1076 return 0;
1080 * Write data to the image file
1082 * Called with table_lock *not* held.
1084 static int coroutine_fn qed_aio_write_main(QEDAIOCB *acb)
1086 BDRVQEDState *s = acb_to_s(acb);
1087 uint64_t offset = acb->cur_cluster +
1088 qed_offset_into_cluster(s, acb->cur_pos);
1090 trace_qed_aio_write_main(s, acb, 0, offset, acb->cur_qiov.size);
1092 BLKDBG_EVENT(s->bs->file, BLKDBG_WRITE_AIO);
1093 return bdrv_co_pwritev(s->bs->file, offset, acb->cur_qiov.size,
1094 &acb->cur_qiov, 0);
1098 * Populate untouched regions of new data cluster
1100 * Called with table_lock held.
1102 static int coroutine_fn qed_aio_write_cow(QEDAIOCB *acb)
1104 BDRVQEDState *s = acb_to_s(acb);
1105 uint64_t start, len, offset;
1106 int ret;
1108 qemu_co_mutex_unlock(&s->table_lock);
1110 /* Populate front untouched region of new data cluster */
1111 start = qed_start_of_cluster(s, acb->cur_pos);
1112 len = qed_offset_into_cluster(s, acb->cur_pos);
1114 trace_qed_aio_write_prefill(s, acb, start, len, acb->cur_cluster);
1115 ret = qed_copy_from_backing_file(s, start, len, acb->cur_cluster);
1116 if (ret < 0) {
1117 goto out;
1120 /* Populate back untouched region of new data cluster */
1121 start = acb->cur_pos + acb->cur_qiov.size;
1122 len = qed_start_of_cluster(s, start + s->header.cluster_size - 1) - start;
1123 offset = acb->cur_cluster +
1124 qed_offset_into_cluster(s, acb->cur_pos) +
1125 acb->cur_qiov.size;
1127 trace_qed_aio_write_postfill(s, acb, start, len, offset);
1128 ret = qed_copy_from_backing_file(s, start, len, offset);
1129 if (ret < 0) {
1130 goto out;
1133 ret = qed_aio_write_main(acb);
1134 if (ret < 0) {
1135 goto out;
1138 if (s->bs->backing) {
1140 * Flush new data clusters before updating the L2 table
1142 * This flush is necessary when a backing file is in use. A crash
1143 * during an allocating write could result in empty clusters in the
1144 * image. If the write only touched a subregion of the cluster,
1145 * then backing image sectors have been lost in the untouched
1146 * region. The solution is to flush after writing a new data
1147 * cluster and before updating the L2 table.
1149 ret = bdrv_co_flush(s->bs->file->bs);
1152 out:
1153 qemu_co_mutex_lock(&s->table_lock);
1154 return ret;
1158 * Check if the QED_F_NEED_CHECK bit should be set during allocating write
1160 static bool qed_should_set_need_check(BDRVQEDState *s)
1162 /* The flush before L2 update path ensures consistency */
1163 if (s->bs->backing) {
1164 return false;
1167 return !(s->header.features & QED_F_NEED_CHECK);
1171 * Write new data cluster
1173 * @acb: Write request
1174 * @len: Length in bytes
1176 * This path is taken when writing to previously unallocated clusters.
1178 * Called with table_lock held.
1180 static int coroutine_fn qed_aio_write_alloc(QEDAIOCB *acb, size_t len)
1182 BDRVQEDState *s = acb_to_s(acb);
1183 int ret;
1185 /* Cancel timer when the first allocating request comes in */
1186 if (s->allocating_acb == NULL) {
1187 qed_cancel_need_check_timer(s);
1190 /* Freeze this request if another allocating write is in progress */
1191 if (s->allocating_acb != acb || s->allocating_write_reqs_plugged) {
1192 if (s->allocating_acb != NULL) {
1193 qemu_co_queue_wait(&s->allocating_write_reqs, &s->table_lock);
1194 assert(s->allocating_acb == NULL);
1196 s->allocating_acb = acb;
1197 return -EAGAIN; /* start over with looking up table entries */
1200 acb->cur_nclusters = qed_bytes_to_clusters(s,
1201 qed_offset_into_cluster(s, acb->cur_pos) + len);
1202 qemu_iovec_concat(&acb->cur_qiov, acb->qiov, acb->qiov_offset, len);
1204 if (acb->flags & QED_AIOCB_ZERO) {
1205 /* Skip ahead if the clusters are already zero */
1206 if (acb->find_cluster_ret == QED_CLUSTER_ZERO) {
1207 return 0;
1209 acb->cur_cluster = 1;
1210 } else {
1211 acb->cur_cluster = qed_alloc_clusters(s, acb->cur_nclusters);
1214 if (qed_should_set_need_check(s)) {
1215 s->header.features |= QED_F_NEED_CHECK;
1216 ret = qed_write_header(s);
1217 if (ret < 0) {
1218 return ret;
1222 if (!(acb->flags & QED_AIOCB_ZERO)) {
1223 ret = qed_aio_write_cow(acb);
1224 if (ret < 0) {
1225 return ret;
1229 return qed_aio_write_l2_update(acb, acb->cur_cluster);
1233 * Write data cluster in place
1235 * @acb: Write request
1236 * @offset: Cluster offset in bytes
1237 * @len: Length in bytes
1239 * This path is taken when writing to already allocated clusters.
1241 * Called with table_lock held.
1243 static int coroutine_fn qed_aio_write_inplace(QEDAIOCB *acb, uint64_t offset,
1244 size_t len)
1246 BDRVQEDState *s = acb_to_s(acb);
1247 int r;
1249 qemu_co_mutex_unlock(&s->table_lock);
1251 /* Allocate buffer for zero writes */
1252 if (acb->flags & QED_AIOCB_ZERO) {
1253 struct iovec *iov = acb->qiov->iov;
1255 if (!iov->iov_base) {
1256 iov->iov_base = qemu_try_blockalign(acb->bs, iov->iov_len);
1257 if (iov->iov_base == NULL) {
1258 r = -ENOMEM;
1259 goto out;
1261 memset(iov->iov_base, 0, iov->iov_len);
1265 /* Calculate the I/O vector */
1266 acb->cur_cluster = offset;
1267 qemu_iovec_concat(&acb->cur_qiov, acb->qiov, acb->qiov_offset, len);
1269 /* Do the actual write. */
1270 r = qed_aio_write_main(acb);
1271 out:
1272 qemu_co_mutex_lock(&s->table_lock);
1273 return r;
1277 * Write data cluster
1279 * @opaque: Write request
1280 * @ret: QED_CLUSTER_FOUND, QED_CLUSTER_L2 or QED_CLUSTER_L1
1281 * @offset: Cluster offset in bytes
1282 * @len: Length in bytes
1284 * Called with table_lock held.
1286 static int coroutine_fn qed_aio_write_data(void *opaque, int ret,
1287 uint64_t offset, size_t len)
1289 QEDAIOCB *acb = opaque;
1291 trace_qed_aio_write_data(acb_to_s(acb), acb, ret, offset, len);
1293 acb->find_cluster_ret = ret;
1295 switch (ret) {
1296 case QED_CLUSTER_FOUND:
1297 return qed_aio_write_inplace(acb, offset, len);
1299 case QED_CLUSTER_L2:
1300 case QED_CLUSTER_L1:
1301 case QED_CLUSTER_ZERO:
1302 return qed_aio_write_alloc(acb, len);
1304 default:
1305 g_assert_not_reached();
1310 * Read data cluster
1312 * @opaque: Read request
1313 * @ret: QED_CLUSTER_FOUND, QED_CLUSTER_L2 or QED_CLUSTER_L1
1314 * @offset: Cluster offset in bytes
1315 * @len: Length in bytes
1317 * Called with table_lock held.
1319 static int coroutine_fn qed_aio_read_data(void *opaque, int ret,
1320 uint64_t offset, size_t len)
1322 QEDAIOCB *acb = opaque;
1323 BDRVQEDState *s = acb_to_s(acb);
1324 BlockDriverState *bs = acb->bs;
1325 int r;
1327 qemu_co_mutex_unlock(&s->table_lock);
1329 /* Adjust offset into cluster */
1330 offset += qed_offset_into_cluster(s, acb->cur_pos);
1332 trace_qed_aio_read_data(s, acb, ret, offset, len);
1334 qemu_iovec_concat(&acb->cur_qiov, acb->qiov, acb->qiov_offset, len);
1336 /* Handle zero cluster and backing file reads, otherwise read
1337 * data cluster directly.
1339 if (ret == QED_CLUSTER_ZERO) {
1340 qemu_iovec_memset(&acb->cur_qiov, 0, 0, acb->cur_qiov.size);
1341 r = 0;
1342 } else if (ret != QED_CLUSTER_FOUND) {
1343 r = qed_read_backing_file(s, acb->cur_pos, &acb->cur_qiov,
1344 &acb->backing_qiov);
1345 } else {
1346 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
1347 r = bdrv_co_preadv(bs->file, offset, acb->cur_qiov.size,
1348 &acb->cur_qiov, 0);
1351 qemu_co_mutex_lock(&s->table_lock);
1352 return r;
1356 * Begin next I/O or complete the request
1358 static int coroutine_fn qed_aio_next_io(QEDAIOCB *acb)
1360 BDRVQEDState *s = acb_to_s(acb);
1361 uint64_t offset;
1362 size_t len;
1363 int ret;
1365 qemu_co_mutex_lock(&s->table_lock);
1366 while (1) {
1367 trace_qed_aio_next_io(s, acb, 0, acb->cur_pos + acb->cur_qiov.size);
1369 if (acb->backing_qiov) {
1370 qemu_iovec_destroy(acb->backing_qiov);
1371 g_free(acb->backing_qiov);
1372 acb->backing_qiov = NULL;
1375 acb->qiov_offset += acb->cur_qiov.size;
1376 acb->cur_pos += acb->cur_qiov.size;
1377 qemu_iovec_reset(&acb->cur_qiov);
1379 /* Complete request */
1380 if (acb->cur_pos >= acb->end_pos) {
1381 ret = 0;
1382 break;
1385 /* Find next cluster and start I/O */
1386 len = acb->end_pos - acb->cur_pos;
1387 ret = qed_find_cluster(s, &acb->request, acb->cur_pos, &len, &offset);
1388 if (ret < 0) {
1389 break;
1392 if (acb->flags & QED_AIOCB_WRITE) {
1393 ret = qed_aio_write_data(acb, ret, offset, len);
1394 } else {
1395 ret = qed_aio_read_data(acb, ret, offset, len);
1398 if (ret < 0 && ret != -EAGAIN) {
1399 break;
1403 trace_qed_aio_complete(s, acb, ret);
1404 qed_aio_complete(acb);
1405 qemu_co_mutex_unlock(&s->table_lock);
1406 return ret;
1409 static int coroutine_fn qed_co_request(BlockDriverState *bs, int64_t sector_num,
1410 QEMUIOVector *qiov, int nb_sectors,
1411 int flags)
1413 QEDAIOCB acb = {
1414 .bs = bs,
1415 .cur_pos = (uint64_t) sector_num * BDRV_SECTOR_SIZE,
1416 .end_pos = (sector_num + nb_sectors) * BDRV_SECTOR_SIZE,
1417 .qiov = qiov,
1418 .flags = flags,
1420 qemu_iovec_init(&acb.cur_qiov, qiov->niov);
1422 trace_qed_aio_setup(bs->opaque, &acb, sector_num, nb_sectors, NULL, flags);
1424 /* Start request */
1425 return qed_aio_next_io(&acb);
1428 static int coroutine_fn bdrv_qed_co_readv(BlockDriverState *bs,
1429 int64_t sector_num, int nb_sectors,
1430 QEMUIOVector *qiov)
1432 return qed_co_request(bs, sector_num, qiov, nb_sectors, 0);
1435 static int coroutine_fn bdrv_qed_co_writev(BlockDriverState *bs,
1436 int64_t sector_num, int nb_sectors,
1437 QEMUIOVector *qiov, int flags)
1439 assert(!flags);
1440 return qed_co_request(bs, sector_num, qiov, nb_sectors, QED_AIOCB_WRITE);
1443 static int coroutine_fn bdrv_qed_co_pwrite_zeroes(BlockDriverState *bs,
1444 int64_t offset,
1445 int bytes,
1446 BdrvRequestFlags flags)
1448 BDRVQEDState *s = bs->opaque;
1449 QEMUIOVector qiov;
1450 struct iovec iov;
1452 /* Fall back if the request is not aligned */
1453 if (qed_offset_into_cluster(s, offset) ||
1454 qed_offset_into_cluster(s, bytes)) {
1455 return -ENOTSUP;
1458 /* Zero writes start without an I/O buffer. If a buffer becomes necessary
1459 * then it will be allocated during request processing.
1461 iov.iov_base = NULL;
1462 iov.iov_len = bytes;
1464 qemu_iovec_init_external(&qiov, &iov, 1);
1465 return qed_co_request(bs, offset >> BDRV_SECTOR_BITS, &qiov,
1466 bytes >> BDRV_SECTOR_BITS,
1467 QED_AIOCB_WRITE | QED_AIOCB_ZERO);
1470 static int coroutine_fn bdrv_qed_co_truncate(BlockDriverState *bs,
1471 int64_t offset,
1472 PreallocMode prealloc,
1473 Error **errp)
1475 BDRVQEDState *s = bs->opaque;
1476 uint64_t old_image_size;
1477 int ret;
1479 if (prealloc != PREALLOC_MODE_OFF) {
1480 error_setg(errp, "Unsupported preallocation mode '%s'",
1481 PreallocMode_str(prealloc));
1482 return -ENOTSUP;
1485 if (!qed_is_image_size_valid(offset, s->header.cluster_size,
1486 s->header.table_size)) {
1487 error_setg(errp, "Invalid image size specified");
1488 return -EINVAL;
1491 if ((uint64_t)offset < s->header.image_size) {
1492 error_setg(errp, "Shrinking images is currently not supported");
1493 return -ENOTSUP;
1496 old_image_size = s->header.image_size;
1497 s->header.image_size = offset;
1498 ret = qed_write_header_sync(s);
1499 if (ret < 0) {
1500 s->header.image_size = old_image_size;
1501 error_setg_errno(errp, -ret, "Failed to update the image size");
1503 return ret;
1506 static int64_t bdrv_qed_getlength(BlockDriverState *bs)
1508 BDRVQEDState *s = bs->opaque;
1509 return s->header.image_size;
1512 static int bdrv_qed_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1514 BDRVQEDState *s = bs->opaque;
1516 memset(bdi, 0, sizeof(*bdi));
1517 bdi->cluster_size = s->header.cluster_size;
1518 bdi->is_dirty = s->header.features & QED_F_NEED_CHECK;
1519 bdi->unallocated_blocks_are_zero = true;
1520 return 0;
1523 static int bdrv_qed_change_backing_file(BlockDriverState *bs,
1524 const char *backing_file,
1525 const char *backing_fmt)
1527 BDRVQEDState *s = bs->opaque;
1528 QEDHeader new_header, le_header;
1529 void *buffer;
1530 size_t buffer_len, backing_file_len;
1531 int ret;
1533 /* Refuse to set backing filename if unknown compat feature bits are
1534 * active. If the image uses an unknown compat feature then we may not
1535 * know the layout of data following the header structure and cannot safely
1536 * add a new string.
1538 if (backing_file && (s->header.compat_features &
1539 ~QED_COMPAT_FEATURE_MASK)) {
1540 return -ENOTSUP;
1543 memcpy(&new_header, &s->header, sizeof(new_header));
1545 new_header.features &= ~(QED_F_BACKING_FILE |
1546 QED_F_BACKING_FORMAT_NO_PROBE);
1548 /* Adjust feature flags */
1549 if (backing_file) {
1550 new_header.features |= QED_F_BACKING_FILE;
1552 if (qed_fmt_is_raw(backing_fmt)) {
1553 new_header.features |= QED_F_BACKING_FORMAT_NO_PROBE;
1557 /* Calculate new header size */
1558 backing_file_len = 0;
1560 if (backing_file) {
1561 backing_file_len = strlen(backing_file);
1564 buffer_len = sizeof(new_header);
1565 new_header.backing_filename_offset = buffer_len;
1566 new_header.backing_filename_size = backing_file_len;
1567 buffer_len += backing_file_len;
1569 /* Make sure we can rewrite header without failing */
1570 if (buffer_len > new_header.header_size * new_header.cluster_size) {
1571 return -ENOSPC;
1574 /* Prepare new header */
1575 buffer = g_malloc(buffer_len);
1577 qed_header_cpu_to_le(&new_header, &le_header);
1578 memcpy(buffer, &le_header, sizeof(le_header));
1579 buffer_len = sizeof(le_header);
1581 if (backing_file) {
1582 memcpy(buffer + buffer_len, backing_file, backing_file_len);
1583 buffer_len += backing_file_len;
1586 /* Write new header */
1587 ret = bdrv_pwrite_sync(bs->file, 0, buffer, buffer_len);
1588 g_free(buffer);
1589 if (ret == 0) {
1590 memcpy(&s->header, &new_header, sizeof(new_header));
1592 return ret;
1595 static void coroutine_fn bdrv_qed_co_invalidate_cache(BlockDriverState *bs,
1596 Error **errp)
1598 BDRVQEDState *s = bs->opaque;
1599 Error *local_err = NULL;
1600 int ret;
1602 bdrv_qed_close(bs);
1604 bdrv_qed_init_state(bs);
1605 qemu_co_mutex_lock(&s->table_lock);
1606 ret = bdrv_qed_do_open(bs, NULL, bs->open_flags, &local_err);
1607 qemu_co_mutex_unlock(&s->table_lock);
1608 if (local_err) {
1609 error_propagate(errp, local_err);
1610 error_prepend(errp, "Could not reopen qed layer: ");
1611 return;
1612 } else if (ret < 0) {
1613 error_setg_errno(errp, -ret, "Could not reopen qed layer");
1614 return;
1618 static int bdrv_qed_co_check(BlockDriverState *bs, BdrvCheckResult *result,
1619 BdrvCheckMode fix)
1621 BDRVQEDState *s = bs->opaque;
1622 int ret;
1624 qemu_co_mutex_lock(&s->table_lock);
1625 ret = qed_check(s, result, !!fix);
1626 qemu_co_mutex_unlock(&s->table_lock);
1628 return ret;
1631 static QemuOptsList qed_create_opts = {
1632 .name = "qed-create-opts",
1633 .head = QTAILQ_HEAD_INITIALIZER(qed_create_opts.head),
1634 .desc = {
1636 .name = BLOCK_OPT_SIZE,
1637 .type = QEMU_OPT_SIZE,
1638 .help = "Virtual disk size"
1641 .name = BLOCK_OPT_BACKING_FILE,
1642 .type = QEMU_OPT_STRING,
1643 .help = "File name of a base image"
1646 .name = BLOCK_OPT_BACKING_FMT,
1647 .type = QEMU_OPT_STRING,
1648 .help = "Image format of the base image"
1651 .name = BLOCK_OPT_CLUSTER_SIZE,
1652 .type = QEMU_OPT_SIZE,
1653 .help = "Cluster size (in bytes)",
1654 .def_value_str = stringify(QED_DEFAULT_CLUSTER_SIZE)
1657 .name = BLOCK_OPT_TABLE_SIZE,
1658 .type = QEMU_OPT_SIZE,
1659 .help = "L1/L2 table size (in clusters)"
1661 { /* end of list */ }
1665 static BlockDriver bdrv_qed = {
1666 .format_name = "qed",
1667 .instance_size = sizeof(BDRVQEDState),
1668 .create_opts = &qed_create_opts,
1669 .supports_backing = true,
1671 .bdrv_probe = bdrv_qed_probe,
1672 .bdrv_open = bdrv_qed_open,
1673 .bdrv_close = bdrv_qed_close,
1674 .bdrv_reopen_prepare = bdrv_qed_reopen_prepare,
1675 .bdrv_child_perm = bdrv_format_default_perms,
1676 .bdrv_co_create = bdrv_qed_co_create,
1677 .bdrv_co_create_opts = bdrv_qed_co_create_opts,
1678 .bdrv_has_zero_init = bdrv_has_zero_init_1,
1679 .bdrv_co_block_status = bdrv_qed_co_block_status,
1680 .bdrv_co_readv = bdrv_qed_co_readv,
1681 .bdrv_co_writev = bdrv_qed_co_writev,
1682 .bdrv_co_pwrite_zeroes = bdrv_qed_co_pwrite_zeroes,
1683 .bdrv_co_truncate = bdrv_qed_co_truncate,
1684 .bdrv_getlength = bdrv_qed_getlength,
1685 .bdrv_get_info = bdrv_qed_get_info,
1686 .bdrv_refresh_limits = bdrv_qed_refresh_limits,
1687 .bdrv_change_backing_file = bdrv_qed_change_backing_file,
1688 .bdrv_co_invalidate_cache = bdrv_qed_co_invalidate_cache,
1689 .bdrv_co_check = bdrv_qed_co_check,
1690 .bdrv_detach_aio_context = bdrv_qed_detach_aio_context,
1691 .bdrv_attach_aio_context = bdrv_qed_attach_aio_context,
1692 .bdrv_co_drain_begin = bdrv_qed_co_drain_begin,
1695 static void bdrv_qed_init(void)
1697 bdrv_register(&bdrv_qed);
1700 block_init(bdrv_qed_init);