]> git.proxmox.com Git - qemu.git/blobdiff - block.c
qcow2: Zero write support
[qemu.git] / block.c
diff --git a/block.c b/block.c
index b88ee9081748a7f37d901ae8d67e99a261d09e6a..fe74dddb134d240f2857625626a922232303d02d 100644 (file)
--- a/block.c
+++ b/block.c
@@ -80,6 +80,8 @@ static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
                                                void *opaque,
                                                bool is_write);
 static void coroutine_fn bdrv_co_do_rw(void *opaque);
+static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
+    int64_t sector_num, int nb_sectors);
 
 static bool bdrv_exceed_bps_limits(BlockDriverState *bs, int nb_sectors,
         bool is_write, double elapsed_time, uint64_t *wait);
@@ -812,7 +814,13 @@ unlink_and_fail:
 
 void bdrv_close(BlockDriverState *bs)
 {
+    bdrv_flush(bs);
     if (bs->drv) {
+        if (bs->job) {
+            block_job_cancel_sync(bs->job);
+        }
+        bdrv_drain_all();
+
         if (bs == bs_snapshots) {
             bs_snapshots = NULL;
         }
@@ -889,14 +897,16 @@ void bdrv_make_anon(BlockDriverState *bs)
  * This will modify the BlockDriverState fields, and swap contents
  * between bs_new and bs_top. Both bs_new and bs_top are modified.
  *
+ * bs_new is required to be anonymous.
+ *
  * This function does not create any image files.
  */
 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top)
 {
     BlockDriverState tmp;
 
-    /* the new bs must not be in bdrv_states */
-    bdrv_make_anon(bs_new);
+    /* bs_new must be anonymous */
+    assert(bs_new->device_name[0] == '\0');
 
     tmp = *bs_new;
 
@@ -941,11 +951,18 @@ void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top)
      * swapping bs_new and bs_top contents. */
     tmp.backing_hd = bs_new;
     pstrcpy(tmp.backing_file, sizeof(tmp.backing_file), bs_top->filename);
+    bdrv_get_format(bs_top, tmp.backing_format, sizeof(tmp.backing_format));
 
     /* swap contents of the fixed new bs and the current top */
     *bs_new = *bs_top;
     *bs_top = tmp;
 
+    /* device_name[] was carried over from the old bs_top.  bs_new
+     * shouldn't be in bdrv_states, so we need to make device_name[]
+     * reflect the anonymity of bs_new
+     */
+    bs_new->device_name[0] = '\0';
+
     /* clear the copied fields in the new backing file */
     bdrv_detach_dev(bs_new, bs_new->dev);
 
@@ -966,6 +983,8 @@ void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top)
 void bdrv_delete(BlockDriverState *bs)
 {
     assert(!bs->dev);
+    assert(!bs->job);
+    assert(!bs->in_use);
 
     /* remove from list, if necessary */
     bdrv_make_anon(bs);
@@ -1463,6 +1482,17 @@ static int bdrv_rw_co(BlockDriverState *bs, int64_t sector_num, uint8_t *buf,
 
     qemu_iovec_init_external(&qiov, &iov, 1);
 
+    /**
+     * In sync call context, when the vcpu is blocked, this throttling timer
+     * will not fire; so the I/O throttling function has to be disabled here
+     * if it has been enabled.
+     */
+    if (bs->io_limits_enabled) {
+        fprintf(stderr, "Disabling I/O throttling on '%s' due "
+                        "to synchronous I/O.\n", bdrv_get_device_name(bs));
+        bdrv_io_limits_disable(bs);
+    }
+
     if (qemu_in_coroutine()) {
         /* Fast-path if already in coroutine context */
         bdrv_rw_co_entry(&rwco);
@@ -1680,8 +1710,8 @@ static int coroutine_fn bdrv_co_do_copy_on_readv(BlockDriverState *bs,
 
     if (drv->bdrv_co_write_zeroes &&
         buffer_is_zero(bounce_buffer, iov.iov_len)) {
-        ret = drv->bdrv_co_write_zeroes(bs, cluster_sector_num,
-                                        cluster_nb_sectors);
+        ret = bdrv_co_do_write_zeroes(bs, cluster_sector_num,
+                                      cluster_nb_sectors);
     } else {
         ret = drv->bdrv_co_writev(bs, cluster_sector_num, cluster_nb_sectors,
                                   &bounce_qiov);
@@ -1791,9 +1821,15 @@ static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
     struct iovec iov;
     int ret;
 
+    /* TODO Emulate only part of misaligned requests instead of letting block
+     * drivers return -ENOTSUP and emulate everything */
+
     /* First try the efficient write zeroes operation */
     if (drv->bdrv_co_write_zeroes) {
-        return drv->bdrv_co_write_zeroes(bs, sector_num, nb_sectors);
+        ret = drv->bdrv_co_write_zeroes(bs, sector_num, nb_sectors);
+        if (ret != -ENOTSUP) {
+            return ret;
+        }
     }
 
     /* Fall back to bounce buffer if write zeroes is unsupported */
@@ -1969,10 +2005,19 @@ static int guess_disk_lchs(BlockDriverState *bs,
     struct partition *p;
     uint32_t nr_sects;
     uint64_t nb_sectors;
+    bool enabled;
 
     bdrv_get_geometry(bs, &nb_sectors);
 
+    /**
+     * The function will be invoked during startup not only in sync I/O mode,
+     * but also in async I/O mode. So the I/O throttling function has to
+     * be disabled temporarily here, not permanently.
+     */
+    enabled = bs->io_limits_enabled;
+    bs->io_limits_enabled = false;
     ret = bdrv_read(bs, 0, buf, 1);
+    bs->io_limits_enabled = enabled;
     if (ret < 0)
         return -1;
     /* test msdos magic */
@@ -2331,9 +2376,7 @@ void bdrv_flush_all(void)
     BlockDriverState *bs;
 
     QTAILQ_FOREACH(bs, &bdrv_states, list) {
-        if (!bdrv_is_read_only(bs) && bdrv_is_inserted(bs)) {
-            bdrv_flush(bs);
-        }
+        bdrv_flush(bs);
     }
 }
 
@@ -3520,7 +3563,7 @@ int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
 {
     int ret;
 
-    if (!bs->drv) {
+    if (!bs || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
         return 0;
     }
 
@@ -3538,7 +3581,7 @@ int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
     }
 
     if (bs->drv->bdrv_co_flush_to_disk) {
-        return bs->drv->bdrv_co_flush_to_disk(bs);
+        ret = bs->drv->bdrv_co_flush_to_disk(bs);
     } else if (bs->drv->bdrv_aio_flush) {
         BlockDriverAIOCB *acb;
         CoroutineIOCompletion co = {
@@ -3547,10 +3590,10 @@ int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
 
         acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co);
         if (acb == NULL) {
-            return -EIO;
+            ret = -EIO;
         } else {
             qemu_coroutine_yield();
-            return co.ret;
+            ret = co.ret;
         }
     } else {
         /*
@@ -3564,8 +3607,16 @@ int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
          *
          * Let's hope the user knows what he's doing.
          */
-        return 0;
+        ret = 0;
     }
+    if (ret < 0) {
+        return ret;
+    }
+
+    /* Now flush the underlying protocol.  It will also have BDRV_O_NO_FLUSH
+     * in the case of cache=unsafe, so there are no useless flushes.
+     */
+    return bdrv_co_flush(bs->file);
 }
 
 void bdrv_invalidate_cache(BlockDriverState *bs)
@@ -3584,6 +3635,15 @@ void bdrv_invalidate_cache_all(void)
     }
 }
 
+void bdrv_clear_incoming_migration_all(void)
+{
+    BlockDriverState *bs;
+
+    QTAILQ_FOREACH(bs, &bdrv_states, list) {
+        bs->open_flags = bs->open_flags & ~(BDRV_O_INCOMING);
+    }
+}
+
 int bdrv_flush(BlockDriverState *bs)
 {
     Coroutine *co;
@@ -4054,10 +4114,16 @@ void block_job_complete(BlockJob *job, int ret)
 
 int block_job_set_speed(BlockJob *job, int64_t value)
 {
+    int rc;
+
     if (!job->job_type->set_speed) {
         return -ENOTSUP;
     }
-    return job->job_type->set_speed(job, value);
+    rc = job->job_type->set_speed(job, value);
+    if (rc == 0) {
+        job->speed = value;
+    }
+    return rc;
 }
 
 void block_job_cancel(BlockJob *job)
@@ -4069,3 +4135,14 @@ bool block_job_is_cancelled(BlockJob *job)
 {
     return job->cancelled;
 }
+
+void block_job_cancel_sync(BlockJob *job)
+{
+    BlockDriverState *bs = job->bs;
+
+    assert(bs->job == job);
+    block_job_cancel(job);
+    while (bs->job != NULL && bs->job->busy) {
+        qemu_aio_wait();
+    }
+}