return 0;
}
+/**
+ * The copy-on-read flag is actually a reference count so multiple users may
+ * use the feature without worrying about clobbering its previous state.
+ * Copy-on-read stays enabled until all users have called to disable it.
+ */
+void bdrv_enable_copy_on_read(BlockDriverState *bs)
+{
+ bs->copy_on_read++;
+}
+
+void bdrv_disable_copy_on_read(BlockDriverState *bs)
+{
+ assert(bs->copy_on_read > 0);
+ bs->copy_on_read--;
+}
+
/*
* Common part for opening disk images and files
*/
bs->growable = 0;
bs->buffer_alignment = 512;
+ assert(bs->copy_on_read == 0); /* bdrv_new() and bdrv_close() make it so */
+ if ((flags & BDRV_O_RDWR) && (flags & BDRV_O_COPY_ON_READ)) {
+ bdrv_enable_copy_on_read(bs);
+ }
+
pstrcpy(bs->filename, sizeof(bs->filename), filename);
bs->backing_file[0] = '\0';
#endif
bs->opaque = NULL;
bs->drv = NULL;
+ bs->copy_on_read = 0;
if (bs->file != NULL) {
bdrv_close(bs->file);
}
}
+/*
+ * Wait for pending requests to complete across all BlockDriverStates
+ *
+ * This function does not flush data to disk, use bdrv_flush_all() for that
+ * after calling this function.
+ */
+void bdrv_drain_all(void)
+{
+ BlockDriverState *bs;
+
+ qemu_aio_flush();
+
+ /* If requests are still pending there is a bug somewhere */
+ QTAILQ_FOREACH(bs, &bdrv_states, list) {
+ assert(QLIST_EMPTY(&bs->tracked_requests));
+ assert(qemu_co_queue_empty(&bs->throttled_reqs));
+ }
+}
+
/* make a BlockDriverState anonymous by removing from bdrv_state list.
Also, NULL terminate the device_name to prevent double remove */
void bdrv_make_anon(BlockDriverState *bs)
return -EACCES;
}
+ if (bdrv_in_use(bs) || bdrv_in_use(bs->backing_hd)) {
+ return -EBUSY;
+ }
+
backing_drv = bs->backing_hd->drv;
ro = bs->backing_hd->read_only;
strncpy(filename, bs->backing_hd->filename, sizeof(filename));
buf = g_malloc(COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE);
for (sector = 0; sector < total_sectors; sector += n) {
- if (drv->bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n)) {
+ if (bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n)) {
if (bdrv_read(bs, sector, buf, n) != 0) {
ret = -EIO;
}
}
+struct BdrvTrackedRequest {
+ BlockDriverState *bs;
+ int64_t sector_num;
+ int nb_sectors;
+ bool is_write;
+ QLIST_ENTRY(BdrvTrackedRequest) list;
+ Coroutine *co; /* owner, used for deadlock detection */
+ CoQueue wait_queue; /* coroutines blocked on this request */
+};
+
+/**
+ * Remove an active request from the tracked requests list
+ *
+ * This function should be called when a tracked request is completing.
+ */
+static void tracked_request_end(BdrvTrackedRequest *req)
+{
+ QLIST_REMOVE(req, list);
+ qemu_co_queue_restart_all(&req->wait_queue);
+}
+
+/**
+ * Add an active request to the tracked requests list
+ */
+static void tracked_request_begin(BdrvTrackedRequest *req,
+ BlockDriverState *bs,
+ int64_t sector_num,
+ int nb_sectors, bool is_write)
+{
+ *req = (BdrvTrackedRequest){
+ .bs = bs,
+ .sector_num = sector_num,
+ .nb_sectors = nb_sectors,
+ .is_write = is_write,
+ .co = qemu_coroutine_self(),
+ };
+
+ qemu_co_queue_init(&req->wait_queue);
+
+ QLIST_INSERT_HEAD(&bs->tracked_requests, req, list);
+}
+
+/**
+ * Round a region to cluster boundaries
+ */
+static void round_to_clusters(BlockDriverState *bs,
+ int64_t sector_num, int nb_sectors,
+ int64_t *cluster_sector_num,
+ int *cluster_nb_sectors)
+{
+ BlockDriverInfo bdi;
+
+ if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) {
+ *cluster_sector_num = sector_num;
+ *cluster_nb_sectors = nb_sectors;
+ } else {
+ int64_t c = bdi.cluster_size / BDRV_SECTOR_SIZE;
+ *cluster_sector_num = QEMU_ALIGN_DOWN(sector_num, c);
+ *cluster_nb_sectors = QEMU_ALIGN_UP(sector_num - *cluster_sector_num +
+ nb_sectors, c);
+ }
+}
+
+static bool tracked_request_overlaps(BdrvTrackedRequest *req,
+ int64_t sector_num, int nb_sectors) {
+ /* aaaa bbbb */
+ if (sector_num >= req->sector_num + req->nb_sectors) {
+ return false;
+ }
+ /* bbbb aaaa */
+ if (req->sector_num >= sector_num + nb_sectors) {
+ return false;
+ }
+ return true;
+}
+
+static void coroutine_fn wait_for_overlapping_requests(BlockDriverState *bs,
+ int64_t sector_num, int nb_sectors)
+{
+ BdrvTrackedRequest *req;
+ int64_t cluster_sector_num;
+ int cluster_nb_sectors;
+ bool retry;
+
+ /* If we touch the same cluster it counts as an overlap. This guarantees
+ * that allocating writes will be serialized and not race with each other
+ * for the same cluster. For example, in copy-on-read it ensures that the
+ * CoR read and write operations are atomic and guest writes cannot
+ * interleave between them.
+ */
+ round_to_clusters(bs, sector_num, nb_sectors,
+ &cluster_sector_num, &cluster_nb_sectors);
+
+ do {
+ retry = false;
+ QLIST_FOREACH(req, &bs->tracked_requests, list) {
+ if (tracked_request_overlaps(req, cluster_sector_num,
+ cluster_nb_sectors)) {
+ /* Hitting this means there was a reentrant request, for
+ * example, a block driver issuing nested requests. This must
+ * never happen since it means deadlock.
+ */
+ assert(qemu_coroutine_self() != req->co);
+
+ qemu_co_queue_wait(&req->wait_queue);
+ retry = true;
+ break;
+ }
+ }
+ } while (retry);
+}
+
/*
* Return values:
* 0 - success
return 0;
}
+static int coroutine_fn bdrv_co_copy_on_readv(BlockDriverState *bs,
+ int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
+{
+ /* Perform I/O through a temporary buffer so that users who scribble over
+ * their read buffer while the operation is in progress do not end up
+ * modifying the image file. This is critical for zero-copy guest I/O
+ * where anything might happen inside guest memory.
+ */
+ void *bounce_buffer;
+
+ struct iovec iov;
+ QEMUIOVector bounce_qiov;
+ int64_t cluster_sector_num;
+ int cluster_nb_sectors;
+ size_t skip_bytes;
+ int ret;
+
+ /* Cover entire cluster so no additional backing file I/O is required when
+ * allocating cluster in the image file.
+ */
+ round_to_clusters(bs, sector_num, nb_sectors,
+ &cluster_sector_num, &cluster_nb_sectors);
+
+ trace_bdrv_co_copy_on_readv(bs, sector_num, nb_sectors,
+ cluster_sector_num, cluster_nb_sectors);
+
+ iov.iov_len = cluster_nb_sectors * BDRV_SECTOR_SIZE;
+ iov.iov_base = bounce_buffer = qemu_blockalign(bs, iov.iov_len);
+ qemu_iovec_init_external(&bounce_qiov, &iov, 1);
+
+ ret = bs->drv->bdrv_co_readv(bs, cluster_sector_num, cluster_nb_sectors,
+ &bounce_qiov);
+ if (ret < 0) {
+ goto err;
+ }
+
+ ret = bs->drv->bdrv_co_writev(bs, cluster_sector_num, cluster_nb_sectors,
+ &bounce_qiov);
+ if (ret < 0) {
+ /* It might be okay to ignore write errors for guest requests. If this
+ * is a deliberate copy-on-read then we don't want to ignore the error.
+ * Simply report it in all cases.
+ */
+ goto err;
+ }
+
+ skip_bytes = (sector_num - cluster_sector_num) * BDRV_SECTOR_SIZE;
+ qemu_iovec_from_buffer(qiov, bounce_buffer + skip_bytes,
+ nb_sectors * BDRV_SECTOR_SIZE);
+
+err:
+ qemu_vfree(bounce_buffer);
+ return ret;
+}
+
/*
* Handle a read request in coroutine context
*/
int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
{
BlockDriver *drv = bs->drv;
+ BdrvTrackedRequest req;
+ int ret;
if (!drv) {
return -ENOMEDIUM;
bdrv_io_limits_intercept(bs, false, nb_sectors);
}
- return drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov);
+ if (bs->copy_on_read) {
+ wait_for_overlapping_requests(bs, sector_num, nb_sectors);
+ }
+
+ tracked_request_begin(&req, bs, sector_num, nb_sectors, false);
+
+ if (bs->copy_on_read) {
+ int pnum;
+
+ ret = bdrv_co_is_allocated(bs, sector_num, nb_sectors, &pnum);
+ if (ret < 0) {
+ goto out;
+ }
+
+ if (!ret || pnum != nb_sectors) {
+ ret = bdrv_co_copy_on_readv(bs, sector_num, nb_sectors, qiov);
+ goto out;
+ }
+ }
+
+ ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov);
+
+out:
+ tracked_request_end(&req);
+ return ret;
}
int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num,
int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
{
BlockDriver *drv = bs->drv;
+ BdrvTrackedRequest req;
int ret;
if (!bs->drv) {
bdrv_io_limits_intercept(bs, true, nb_sectors);
}
+ if (bs->copy_on_read) {
+ wait_for_overlapping_requests(bs, sector_num, nb_sectors);
+ }
+
+ tracked_request_begin(&req, bs, sector_num, nb_sectors, true);
+
ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov);
if (bs->dirty_bitmap) {
bs->wr_highest_sector = sector_num + nb_sectors - 1;
}
+ tracked_request_end(&req);
+
return ret;
}
return 1;
}
+typedef struct BdrvCoIsAllocatedData {
+ BlockDriverState *bs;
+ int64_t sector_num;
+ int nb_sectors;
+ int *pnum;
+ int ret;
+ bool done;
+} BdrvCoIsAllocatedData;
+
/*
* Returns true iff the specified sector is present in the disk image. Drivers
* not implementing the functionality are assumed to not support backing files,
* hence all their sectors are reported as allocated.
*
+ * If 'sector_num' is beyond the end of the disk image the return value is 0
+ * and 'pnum' is set to 0.
+ *
* 'pnum' is set to the number of sectors (including and immediately following
* the specified sector) that are known to be in the same
* allocated/unallocated state.
*
- * 'nb_sectors' is the max value 'pnum' should be set to.
+ * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
+ * beyond the end of the disk image it will be clamped.
*/
-int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
- int *pnum)
+int coroutine_fn bdrv_co_is_allocated(BlockDriverState *bs, int64_t sector_num,
+ int nb_sectors, int *pnum)
{
int64_t n;
- if (!bs->drv->bdrv_is_allocated) {
- if (sector_num >= bs->total_sectors) {
- *pnum = 0;
- return 0;
- }
- n = bs->total_sectors - sector_num;
- *pnum = (n < nb_sectors) ? (n) : (nb_sectors);
+
+ if (sector_num >= bs->total_sectors) {
+ *pnum = 0;
+ return 0;
+ }
+
+ n = bs->total_sectors - sector_num;
+ if (n < nb_sectors) {
+ nb_sectors = n;
+ }
+
+ if (!bs->drv->bdrv_co_is_allocated) {
+ *pnum = nb_sectors;
return 1;
}
- return bs->drv->bdrv_is_allocated(bs, sector_num, nb_sectors, pnum);
+
+ return bs->drv->bdrv_co_is_allocated(bs, sector_num, nb_sectors, pnum);
+}
+
+/* Coroutine wrapper for bdrv_is_allocated() */
+static void coroutine_fn bdrv_is_allocated_co_entry(void *opaque)
+{
+ BdrvCoIsAllocatedData *data = opaque;
+ BlockDriverState *bs = data->bs;
+
+ data->ret = bdrv_co_is_allocated(bs, data->sector_num, data->nb_sectors,
+ data->pnum);
+ data->done = true;
+}
+
+/*
+ * Synchronous wrapper around bdrv_co_is_allocated().
+ *
+ * See bdrv_co_is_allocated() for details.
+ */
+int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
+ int *pnum)
+{
+ Coroutine *co;
+ BdrvCoIsAllocatedData data = {
+ .bs = bs,
+ .sector_num = sector_num,
+ .nb_sectors = nb_sectors,
+ .pnum = pnum,
+ .done = false,
+ };
+
+ co = qemu_coroutine_create(bdrv_is_allocated_co_entry);
+ qemu_coroutine_enter(co, &data);
+ while (!data.done) {
+ qemu_aio_wait();
+ }
+ return data.ret;
}
void bdrv_mon_event(const BlockDriverState *bdrv,
*/
int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
{
- BlockDriverAIOCB *acb;
MultiwriteCB *mcb;
int i;
trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs);
- /*
- * Run the aio requests. As soon as one request can't be submitted
- * successfully, fail all requests that are not yet submitted (we must
- * return failure for all requests anyway)
- *
- * num_requests cannot be set to the right value immediately: If
- * bdrv_aio_writev fails for some request, num_requests would be too high
- * and therefore multiwrite_cb() would never recognize the multiwrite
- * request as completed. We also cannot use the loop variable i to set it
- * when the first request fails because the callback may already have been
- * called for previously submitted requests. Thus, num_requests must be
- * incremented for each request that is submitted.
- *
- * The problem that callbacks may be called early also means that we need
- * to take care that num_requests doesn't become 0 before all requests are
- * submitted - multiwrite_cb() would consider the multiwrite request
- * completed. A dummy request that is "completed" by a manual call to
- * multiwrite_cb() takes care of this.
- */
- mcb->num_requests = 1;
-
- // Run the aio requests
+ /* Run the aio requests. */
+ mcb->num_requests = num_reqs;
for (i = 0; i < num_reqs; i++) {
- mcb->num_requests++;
- acb = bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov,
+ bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov,
reqs[i].nb_sectors, multiwrite_cb, mcb);
-
- if (acb == NULL) {
- // We can only fail the whole thing if no request has been
- // submitted yet. Otherwise we'll wait for the submitted AIOs to
- // complete and report the error in the callback.
- if (i == 0) {
- trace_bdrv_aio_multiwrite_earlyfail(mcb);
- goto fail;
- } else {
- trace_bdrv_aio_multiwrite_latefail(mcb, i);
- multiwrite_cb(mcb, -EIO);
- break;
- }
- }
}
- /* Complete the dummy request */
- multiwrite_cb(mcb, 0);
-
return 0;
-
-fail:
- for (i = 0; i < mcb->num_callbacks; i++) {
- reqs[i].error = -EIO;
- }
- g_free(mcb);
- return -1;
}
void bdrv_aio_cancel(BlockDriverAIOCB *acb)
acb->is_write = is_write;
acb->qiov = qiov;
acb->bounce = qemu_blockalign(bs, qiov->size);
-
- if (!acb->bh)
- acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
+ acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
if (is_write) {
qemu_iovec_to_buffer(acb->qiov, acb->bounce);