]> git.proxmox.com Git - mirror_qemu.git/blobdiff - block/stream.c
block: Pass BdrvChildRole in remaining cases
[mirror_qemu.git] / block / stream.c
index 6253c86faebcf150d776f2cf5df5624851ac23ca..aa2e7af98e37747526efbba6ef1120294e6abc55 100644 (file)
 
 enum {
     /*
-     * Size of data buffer for populating the image file.  This should be large
-     * enough to process multiple clusters in a single call, so that populating
-     * contiguous regions of the image is efficient.
+     * Maximum chunk size to feed to copy-on-read.  This should be
+     * large enough to process multiple clusters in a single call, so
+     * that populating contiguous regions of the image is efficient.
      */
-    STREAM_BUFFER_SIZE = 512 * 1024, /* in bytes */
+    STREAM_CHUNK = 512 * 1024, /* in bytes */
 };
 
 typedef struct StreamBlockJob {
     BlockJob common;
-    BlockDriverState *base;
+    BlockDriverState *bottom;
     BlockdevOnError on_error;
     char *backing_file_str;
     bool bs_read_only;
@@ -39,15 +39,12 @@ typedef struct StreamBlockJob {
 } StreamBlockJob;
 
 static int coroutine_fn stream_populate(BlockBackend *blk,
-                                        int64_t offset, uint64_t bytes,
-                                        void *buf)
+                                        int64_t offset, uint64_t bytes)
 {
-    QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes);
-
     assert(bytes < SIZE_MAX);
 
-    /* Copy-on-read the unallocated clusters */
-    return blk_co_preadv(blk, offset, qiov.size, &qiov, BDRV_REQ_COPY_ON_READ);
+    return blk_co_preadv(blk, offset, bytes, NULL,
+                         BDRV_REQ_COPY_ON_READ | BDRV_REQ_PREFETCH);
 }
 
 static void stream_abort(Job *job)
@@ -56,7 +53,7 @@ static void stream_abort(Job *job)
 
     if (s->chain_frozen) {
         BlockJob *bjob = &s->common;
-        bdrv_unfreeze_backing_chain(blk_bs(bjob->blk), s->base);
+        bdrv_unfreeze_backing_chain(blk_bs(bjob->blk), s->bottom);
     }
 }
 
@@ -65,11 +62,11 @@ static int stream_prepare(Job *job)
     StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
     BlockJob *bjob = &s->common;
     BlockDriverState *bs = blk_bs(bjob->blk);
-    BlockDriverState *base = s->base;
+    BlockDriverState *base = backing_bs(s->bottom);
     Error *local_err = NULL;
     int ret = 0;
 
-    bdrv_unfreeze_backing_chain(bs, base);
+    bdrv_unfreeze_backing_chain(bs, s->bottom);
     s->chain_frozen = false;
 
     if (bs->backing) {
@@ -80,8 +77,8 @@ static int stream_prepare(Job *job)
                 base_fmt = base->drv->format_name;
             }
         }
-        ret = bdrv_change_backing_file(bs, base_id, base_fmt);
         bdrv_set_backing_hd(bs, base, &local_err);
+        ret = bdrv_change_backing_file(bs, base_id, base_fmt);
         if (local_err) {
             error_report_err(local_err);
             return -EPERM;
@@ -112,39 +109,36 @@ static int coroutine_fn stream_run(Job *job, Error **errp)
     StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
     BlockBackend *blk = s->common.blk;
     BlockDriverState *bs = blk_bs(blk);
-    BlockDriverState *base = s->base;
+    bool enable_cor = !backing_bs(s->bottom);
     int64_t len;
     int64_t offset = 0;
     uint64_t delay_ns = 0;
     int error = 0;
-    int ret = 0;
     int64_t n = 0; /* bytes */
-    void *buf;
 
-    if (!bs->backing) {
-        goto out;
+    if (bs == s->bottom) {
+        /* Nothing to stream */
+        return 0;
     }
 
     len = bdrv_getlength(bs);
     if (len < 0) {
-        ret = len;
-        goto out;
+        return len;
     }
     job_progress_set_remaining(&s->common.job, len);
 
-    buf = qemu_blockalign(bs, STREAM_BUFFER_SIZE);
-
     /* Turn on copy-on-read for the whole block device so that guest read
      * requests help us make progress.  Only do this when copying the entire
      * backing chain since the copy-on-read operation does not take base into
      * account.
      */
-    if (!base) {
+    if (enable_cor) {
         bdrv_enable_copy_on_read(bs);
     }
 
     for ( ; offset < len; offset += n) {
         bool copy;
+        int ret;
 
         /* Note that even when no rate limit is applied we need to yield
          * with no pending I/O here so that bdrv_drain_all() returns.
@@ -156,15 +150,14 @@ static int coroutine_fn stream_run(Job *job, Error **errp)
 
         copy = false;
 
-        ret = bdrv_is_allocated(bs, offset, STREAM_BUFFER_SIZE, &n);
+        ret = bdrv_is_allocated(bs, offset, STREAM_CHUNK, &n);
         if (ret == 1) {
             /* Allocated in the top, no need to copy.  */
         } else if (ret >= 0) {
             /* Copy if allocated in the intermediate images.  Limit to the
              * known-unallocated area [offset, offset+n*BDRV_SECTOR_SIZE).  */
-            ret = bdrv_is_allocated_above(backing_bs(bs), base,
+            ret = bdrv_is_allocated_above(backing_bs(bs), s->bottom, true,
                                           offset, n, &n);
-
             /* Finish early if end of backing file has been reached */
             if (ret == 0 && n == 0) {
                 n = len - offset;
@@ -174,7 +167,7 @@ static int coroutine_fn stream_run(Job *job, Error **errp)
         }
         trace_stream_one_iteration(s, offset, n, ret);
         if (copy) {
-            ret = stream_populate(blk, offset, n, buf);
+            ret = stream_populate(blk, offset, n);
         }
         if (ret < 0) {
             BlockErrorAction action =
@@ -190,7 +183,6 @@ static int coroutine_fn stream_run(Job *job, Error **errp)
                 break;
             }
         }
-        ret = 0;
 
         /* Publish progress */
         job_progress_update(&s->common.job, n);
@@ -201,18 +193,12 @@ static int coroutine_fn stream_run(Job *job, Error **errp)
         }
     }
 
-    if (!base) {
+    if (enable_cor) {
         bdrv_disable_copy_on_read(bs);
     }
 
-    /* Do not remove the backing file if an error was there but ignored.  */
-    ret = error;
-
-    qemu_vfree(buf);
-
-out:
-    /* Modify backing chain and close BDSes in main loop */
-    return ret;
+    /* Do not remove the backing file if an error was there but ignored. */
+    return error;
 }
 
 static const BlockJobDriver stream_job_driver = {
@@ -225,7 +211,6 @@ static const BlockJobDriver stream_job_driver = {
         .abort         = stream_abort,
         .clean         = stream_clean,
         .user_resume   = block_job_user_resume,
-        .drain         = block_job_drain,
     },
 };
 
@@ -237,12 +222,19 @@ void stream_start(const char *job_id, BlockDriverState *bs,
     StreamBlockJob *s;
     BlockDriverState *iter;
     bool bs_read_only;
+    int basic_flags = BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED;
+    BlockDriverState *bottom = bdrv_find_overlay(bs, base);
+
+    if (bdrv_freeze_backing_chain(bs, bottom, errp) < 0) {
+        return;
+    }
 
     /* Make sure that the image is opened in read-write mode */
     bs_read_only = bdrv_is_read_only(bs);
     if (bs_read_only) {
         if (bdrv_reopen_set_read_only(bs, false, errp) != 0) {
-            return;
+            bs_read_only = false;
+            goto fail;
         }
     }
 
@@ -250,10 +242,8 @@ void stream_start(const char *job_id, BlockDriverState *bs,
      * already have our own plans. Also don't allow resize as the image size is
      * queried only at the job start and then cached. */
     s = block_job_create(job_id, &stream_job_driver, NULL, bs,
-                         BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
-                         BLK_PERM_GRAPH_MOD,
-                         BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
-                         BLK_PERM_WRITE,
+                         basic_flags | BLK_PERM_GRAPH_MOD,
+                         basic_flags | BLK_PERM_WRITE,
                          speed, creation_flags, NULL, NULL, errp);
     if (!s) {
         goto fail;
@@ -261,20 +251,18 @@ void stream_start(const char *job_id, BlockDriverState *bs,
 
     /* Block all intermediate nodes between bs and base, because they will
      * disappear from the chain after this operation. The streaming job reads
-     * every block only once, assuming that it doesn't change, so block writes
-     * and resizes. */
+     * every block only once, assuming that it doesn't change, so forbid writes
+     * and resizes. Reassign the base node pointer because the backing BS of the
+     * bottom node might change after the call to bdrv_reopen_set_read_only()
+     * due to parallel block jobs running.
+     */
+    base = backing_bs(bottom);
     for (iter = backing_bs(bs); iter && iter != base; iter = backing_bs(iter)) {
         block_job_add_bdrv(&s->common, "intermediate node", iter, 0,
-                           BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED,
-                           &error_abort);
-    }
-
-    if (bdrv_freeze_backing_chain(bs, base, errp) < 0) {
-        job_early_fail(&s->common.job);
-        goto fail;
+                           basic_flags, &error_abort);
     }
 
-    s->base = base;
+    s->bottom = bottom;
     s->backing_file_str = g_strdup(backing_file_str);
     s->bs_read_only = bs_read_only;
     s->chain_frozen = true;
@@ -288,4 +276,5 @@ fail:
     if (bs_read_only) {
         bdrv_reopen_set_read_only(bs, true, NULL);
     }
+    bdrv_unfreeze_backing_chain(bs, bottom);
 }