]> git.proxmox.com Git - qemu.git/blobdiff - block/commit.c
block: add live block commit functionality
[qemu.git] / block / commit.c
diff --git a/block/commit.c b/block/commit.c
new file mode 100644 (file)
index 0000000..624ec5f
--- /dev/null
@@ -0,0 +1,267 @@
+/*
+ * Live block commit
+ *
+ * Copyright Red Hat, Inc. 2012
+ *
+ * Authors:
+ *  Jeff Cody   <jcody@redhat.com>
+ *  Based on stream.c by Stefan Hajnoczi
+ *
+ * This work is licensed under the terms of the GNU LGPL, version 2 or later.
+ * See the COPYING.LIB file in the top-level directory.
+ *
+ */
+
+#include "trace.h"
+#include "block_int.h"
+#include "qemu/ratelimit.h"
+
+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.
+     */
+    COMMIT_BUFFER_SIZE = 512 * 1024, /* in bytes */
+};
+
+#define SLICE_TIME 100000000ULL /* ns */
+
+typedef struct CommitBlockJob {
+    BlockJob common;
+    RateLimit limit;
+    BlockDriverState *active;
+    BlockDriverState *top;
+    BlockDriverState *base;
+    BlockErrorAction on_error;
+    int base_flags;
+    int orig_overlay_flags;
+} CommitBlockJob;
+
+static int coroutine_fn commit_populate(BlockDriverState *bs,
+                                        BlockDriverState *base,
+                                        int64_t sector_num, int nb_sectors,
+                                        void *buf)
+{
+    int ret = 0;
+
+    ret = bdrv_read(bs, sector_num, buf, nb_sectors);
+    if (ret) {
+        return ret;
+    }
+
+    ret = bdrv_write(base, sector_num, buf, nb_sectors);
+    if (ret) {
+        return ret;
+    }
+
+    return 0;
+}
+
+static void coroutine_fn commit_run(void *opaque)
+{
+    CommitBlockJob *s = opaque;
+    BlockDriverState *active = s->active;
+    BlockDriverState *top = s->top;
+    BlockDriverState *base = s->base;
+    BlockDriverState *overlay_bs = NULL;
+    int64_t sector_num, end;
+    int ret = 0;
+    int n = 0;
+    void *buf;
+    int bytes_written = 0;
+    int64_t base_len;
+
+    ret = s->common.len = bdrv_getlength(top);
+
+
+    if (s->common.len < 0) {
+        goto exit_restore_reopen;
+    }
+
+    ret = base_len = bdrv_getlength(base);
+    if (base_len < 0) {
+        goto exit_restore_reopen;
+    }
+
+    if (base_len < s->common.len) {
+        ret = bdrv_truncate(base, s->common.len);
+        if (ret) {
+            goto exit_restore_reopen;
+        }
+    }
+
+    overlay_bs = bdrv_find_overlay(active, top);
+
+    end = s->common.len >> BDRV_SECTOR_BITS;
+    buf = qemu_blockalign(top, COMMIT_BUFFER_SIZE);
+
+    for (sector_num = 0; sector_num < end; sector_num += n) {
+        uint64_t delay_ns = 0;
+        bool copy;
+
+wait:
+        /* Note that even when no rate limit is applied we need to yield
+         * with no pending I/O here so that qemu_aio_flush() returns.
+         */
+        block_job_sleep_ns(&s->common, rt_clock, delay_ns);
+        if (block_job_is_cancelled(&s->common)) {
+            break;
+        }
+        /* Copy if allocated above the base */
+        ret = bdrv_co_is_allocated_above(top, base, sector_num,
+                                         COMMIT_BUFFER_SIZE / BDRV_SECTOR_SIZE,
+                                         &n);
+        copy = (ret == 1);
+        trace_commit_one_iteration(s, sector_num, n, ret);
+        if (copy) {
+            if (s->common.speed) {
+                delay_ns = ratelimit_calculate_delay(&s->limit, n);
+                if (delay_ns > 0) {
+                    goto wait;
+                }
+            }
+            ret = commit_populate(top, base, sector_num, n, buf);
+            bytes_written += n * BDRV_SECTOR_SIZE;
+        }
+        if (ret < 0) {
+            if (s->on_error == BLOCK_ERR_STOP_ANY    ||
+                s->on_error == BLOCK_ERR_REPORT      ||
+                (s->on_error == BLOCK_ERR_STOP_ENOSPC && ret == -ENOSPC)) {
+                goto exit_free_buf;
+            } else {
+                n = 0;
+                continue;
+            }
+        }
+        /* Publish progress */
+        s->common.offset += n * BDRV_SECTOR_SIZE;
+    }
+
+    ret = 0;
+
+    if (!block_job_is_cancelled(&s->common) && sector_num == end) {
+        /* success */
+        ret = bdrv_drop_intermediate(active, top, base);
+    }
+
+exit_free_buf:
+    qemu_vfree(buf);
+
+exit_restore_reopen:
+    /* restore base open flags here if appropriate (e.g., change the base back
+     * to r/o). These reopens do not need to be atomic, since we won't abort
+     * even on failure here */
+    if (s->base_flags != bdrv_get_flags(base)) {
+        bdrv_reopen(base, s->base_flags, NULL);
+    }
+    if (s->orig_overlay_flags != bdrv_get_flags(overlay_bs)) {
+        bdrv_reopen(overlay_bs, s->orig_overlay_flags, NULL);
+    }
+
+    block_job_complete(&s->common, ret);
+}
+
+static void commit_set_speed(BlockJob *job, int64_t speed, Error **errp)
+{
+    CommitBlockJob *s = container_of(job, CommitBlockJob, common);
+
+    if (speed < 0) {
+        error_set(errp, QERR_INVALID_PARAMETER, "speed");
+        return;
+    }
+    ratelimit_set_speed(&s->limit, speed / BDRV_SECTOR_SIZE, SLICE_TIME);
+}
+
+static BlockJobType commit_job_type = {
+    .instance_size = sizeof(CommitBlockJob),
+    .job_type      = "commit",
+    .set_speed     = commit_set_speed,
+};
+
+void commit_start(BlockDriverState *bs, BlockDriverState *base,
+                  BlockDriverState *top, int64_t speed,
+                  BlockErrorAction on_error, BlockDriverCompletionFunc *cb,
+                  void *opaque, Error **errp)
+{
+    CommitBlockJob *s;
+    BlockReopenQueue *reopen_queue = NULL;
+    int orig_overlay_flags;
+    int orig_base_flags;
+    BlockDriverState *overlay_bs;
+    Error *local_err = NULL;
+
+    if ((on_error == BLOCK_ERR_STOP_ANY ||
+         on_error == BLOCK_ERR_STOP_ENOSPC) &&
+        !bdrv_iostatus_is_enabled(bs)) {
+        error_set(errp, QERR_INVALID_PARAMETER_COMBINATION);
+        return;
+    }
+
+    /* Once we support top == active layer, remove this check */
+    if (top == bs) {
+        error_setg(errp,
+                   "Top image as the active layer is currently unsupported");
+        return;
+    }
+
+    if (top == base) {
+        error_setg(errp, "Invalid files for merge: top and base are the same");
+        return;
+    }
+
+    /* top and base may be valid, but let's make sure that base is reachable
+     * from top */
+    if (bdrv_find_backing_image(top, base->filename) != base) {
+        error_setg(errp,
+                   "Base (%s) is not reachable from top (%s)",
+                   base->filename, top->filename);
+        return;
+    }
+
+    overlay_bs = bdrv_find_overlay(bs, top);
+
+    if (overlay_bs == NULL) {
+        error_setg(errp, "Could not find overlay image for %s:", top->filename);
+        return;
+    }
+
+    orig_base_flags    = bdrv_get_flags(base);
+    orig_overlay_flags = bdrv_get_flags(overlay_bs);
+
+    /* convert base & overlay_bs to r/w, if necessary */
+    if (!(orig_base_flags & BDRV_O_RDWR)) {
+        reopen_queue = bdrv_reopen_queue(reopen_queue, base,
+                                         orig_base_flags | BDRV_O_RDWR);
+    }
+    if (!(orig_overlay_flags & BDRV_O_RDWR)) {
+        reopen_queue = bdrv_reopen_queue(reopen_queue, overlay_bs,
+                                         orig_overlay_flags | BDRV_O_RDWR);
+    }
+    if (reopen_queue) {
+        bdrv_reopen_multiple(reopen_queue, &local_err);
+        if (local_err != NULL) {
+            error_propagate(errp, local_err);
+            return;
+        }
+    }
+
+
+    s = block_job_create(&commit_job_type, bs, speed, cb, opaque, errp);
+    if (!s) {
+        return;
+    }
+
+    s->base   = base;
+    s->top    = top;
+    s->active = bs;
+
+    s->base_flags          = orig_base_flags;
+    s->orig_overlay_flags  = orig_overlay_flags;
+
+    s->on_error = on_error;
+    s->common.co = qemu_coroutine_create(commit_run);
+
+    trace_commit_start(bs, base, top, s, s->common.co, opaque);
+    qemu_coroutine_enter(s->common.co, s);
+}