Patchwork [1/5] RFC: Efficient VM backup for qemu (v1)

login
register
mail settings
Submitter Dietmar Maurer
Date Nov. 27, 2012, 10:24 a.m.
Message ID <24E144B8C0207547AD09C467A8259F7557832A36@lisa.maurer-it.com>
Download mbox | patch
Permalink /patch/202178/
State New
Headers show

Comments

Dietmar Maurer - Nov. 27, 2012, 10:24 a.m.
> >>> The only solution I came up with is to add before/after hooks in the
> >>> block job.  I agree with the criticism, but I think it's general
> >>> enough and at the same time easy enough to implement.

Ok, here is another try - do you think that is better?

Note: This code is not tested - I just want to ask if I am moving into the right direction.

I tried to move BackupInfo into the block job.

From 86c3ca6467e8d220369f5e7b235cdc0e88a79ff9 Mon Sep 17 00:00:00 2001
From: Dietmar Maurer <dietmar@proxmox.com>
Date: Tue, 27 Nov 2012 11:05:23 +0100
Subject: [PATCH] add basic backup support to block driver

Function bdrv_backup_init() creates a block job to backup a block device.

We call brdv_co_backup_cow() for each write during backup. That function
reads the original data and pass it to backup_dump_cb().

The tracked_request infrastructure is used to serialize access.

Currently backup cluster size is hardcoded to 65536 bytes.

Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>
---
 Makefile.objs |    1 +
 backup.c      |  302 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++
 backup.h      |   30 ++++++
 block.c       |   71 ++++++++++++-
 block.h       |    2 +
 blockjob.h    |   10 ++
 6 files changed, 410 insertions(+), 6 deletions(-)
 create mode 100644 backup.c
 create mode 100644 backup.h

Patch

diff --git a/Makefile.objs b/Makefile.objs
index 3c7abca..cb46be5 100644
--- a/Makefile.objs
+++ b/Makefile.objs
@@ -48,6 +48,7 @@  coroutine-obj-$(CONFIG_WIN32) += coroutine-win32.o
 block-obj-y = iov.o cache-utils.o qemu-option.o module.o async.o
 block-obj-y += nbd.o block.o blockjob.o aes.o qemu-config.o
 block-obj-y += thread-pool.o qemu-progress.o qemu-sockets.o uri.o notify.o
+block-obj-y += backup.o
 block-obj-y += $(coroutine-obj-y) $(qobject-obj-y) $(version-obj-y)
 block-obj-$(CONFIG_POSIX) += event_notifier-posix.o aio-posix.o
 block-obj-$(CONFIG_WIN32) += event_notifier-win32.o aio-win32.o
diff --git a/backup.c b/backup.c
new file mode 100644
index 0000000..e86b76e
--- /dev/null
+++ b/backup.c
@@ -0,0 +1,302 @@ 
+/*
+ * QEMU backup
+ *
+ * Copyright (C) 2012 Proxmox Server Solutions
+ *
+ * Authors:
+ *  Dietmar Maurer (dietmar@proxmox.com)
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
+ * See the COPYING file in the top-level directory.
+ *
+ */
+
+#include <stdio.h>
+#include <errno.h>
+#include <unistd.h>
+
+#include "block.h"
+#include "block_int.h"
+#include "blockjob.h"
+#include "backup.h"
+
+#define DEBUG_BACKUP 0
+
+#define DPRINTF(fmt, ...) \
+    do { if (DEBUG_BACKUP) { printf("backup: " fmt, ## __VA_ARGS__); } } \
+    while (0)
+
+
+#define BITS_PER_LONG  (sizeof(unsigned long) * 8)
+
+typedef struct BackupBlockJob {
+    BlockJob common;
+    unsigned long *bitmap;
+    int bitmap_size;
+    BackupDumpFunc *backup_dump_cb;
+    BlockDriverCompletionFunc *backup_complete_cb;
+    void *opaque;
+} BackupBlockJob;
+
+static int backup_get_bitmap(BlockDriverState *bs, int64_t cluster_num)
+{
+    assert(bs);
+    BackupBlockJob *job = (BackupBlockJob *)bs->job;
+    assert(job);
+    assert(job->bitmap);
+
+    unsigned long val, idx, bit;
+
+    idx = cluster_num / BITS_PER_LONG;
+
+    assert(job->bitmap_size > idx);
+
+    bit = cluster_num % BITS_PER_LONG;
+    val = job->bitmap[idx];
+
+    return !!(val & (1UL << bit));
+}
+
+static void backup_set_bitmap(BlockDriverState *bs, int64_t cluster_num,
+                              int dirty)
+{
+    assert(bs);
+    BackupBlockJob *job = (BackupBlockJob *)bs->job;
+    assert(job);
+    assert(job->bitmap);
+
+    unsigned long val, idx, bit;
+
+    idx = cluster_num / BITS_PER_LONG;
+
+    assert(job->bitmap_size > idx);
+
+    bit = cluster_num % BITS_PER_LONG;
+    val = job->bitmap[idx];
+    if (dirty) {
+        if (!(val & (1UL << bit))) {
+            val |= 1UL << bit;
+        }
+    } else {
+        if (val & (1UL << bit)) {
+            val &= ~(1UL << bit);
+        }
+    }
+    job->bitmap[idx] = val;
+}
+
+static int backup_in_progress_count;
+
+static int coroutine_fn bdrv_co_backup_cow(BlockDriverState *bs,
+                                           int64_t sector_num, int nb_sectors)
+{
+    assert(bs);
+    BackupBlockJob *job = (BackupBlockJob *)bs->job;
+    assert(job);
+
+    BlockDriver *drv = bs->drv;
+    struct iovec iov;
+    QEMUIOVector bounce_qiov;
+    void *bounce_buffer = NULL;
+    int ret = 0;
+
+    backup_in_progress_count++;
+
+    int64_t start, end;
+
+    start = sector_num / BACKUP_BLOCKS_PER_CLUSTER;
+    end = (sector_num + nb_sectors + BACKUP_BLOCKS_PER_CLUSTER - 1) /
+        BACKUP_BLOCKS_PER_CLUSTER;
+
+    DPRINTF("brdv_co_backup_cow enter %s C%zd %zd %d\n",
+            bdrv_get_device_name(bs), start, sector_num, nb_sectors);
+
+    for (; start < end; start++) {
+        if (backup_get_bitmap(bs, start)) {
+            DPRINTF("brdv_co_backup_cow skip C%zd\n", start);
+            continue; /* already copied */
+        }
+
+        /* immediately set bitmap (avoid coroutine race) */
+        backup_set_bitmap(bs, start, 1);
+
+        DPRINTF("brdv_co_backup_cow C%zd\n", start);
+
+        if (!bounce_buffer) {
+            iov.iov_len = BACKUP_CLUSTER_SIZE;
+            iov.iov_base = bounce_buffer = qemu_blockalign(bs, iov.iov_len);
+            qemu_iovec_init_external(&bounce_qiov, &iov, 1);
+        }
+
+        ret = drv->bdrv_co_readv(bs, start * BACKUP_BLOCKS_PER_CLUSTER,
+                                 BACKUP_BLOCKS_PER_CLUSTER,
+                                 &bounce_qiov);
+        if (ret < 0) {
+            DPRINTF("brdv_co_backup_cow bdrv_read C%zd failed\n", start);
+            goto out;
+        }
+
+        ret = job->backup_dump_cb(job->opaque, bs, start, bounce_buffer);
+        if (ret < 0) {
+            DPRINTF("brdv_co_backup_cow dump_cluster_cb C%zd failed\n", start);
+            goto out;
+        }
+
+        DPRINTF("brdv_co_backup_cow done C%zd\n", start);
+    }
+
+out:
+    if (bounce_buffer) {
+        qemu_vfree(bounce_buffer);
+    }
+
+    backup_in_progress_count--;
+
+    return ret;
+}
+    
+static int coroutine_fn backup_before_read(BlockDriverState *bs, 
+                                           int64_t sector_num, 
+                                           int nb_sectors, QEMUIOVector *qiov)
+{
+    return bdrv_co_backup_cow(bs, sector_num, nb_sectors);
+}
+
+static int coroutine_fn backup_before_write(BlockDriverState *bs,
+                                            int64_t sector_num, 
+                                            int nb_sectors, QEMUIOVector *qiov)
+{
+    return bdrv_co_backup_cow(bs, sector_num, nb_sectors);
+}
+
+
+static BlockJobType backup_job_type = {
+    .instance_size = sizeof(BackupBlockJob),
+    .before_read = backup_before_read,
+    .before_write = backup_before_write,
+    .job_type      = "backup",
+};
+
+static void coroutine_fn backup_run(void *opaque)
+{
+    BackupBlockJob *job = opaque;
+    BlockDriverState *bs = job->common.bs;
+    assert(bs);
+
+    int64_t start, end;
+
+    start = 0;
+    end = (bs->total_sectors + BACKUP_BLOCKS_PER_CLUSTER - 1) /
+        BACKUP_BLOCKS_PER_CLUSTER;
+
+    DPRINTF("backup_run start %s %zd %zd\n", bdrv_get_device_name(bs),
+            start, end);
+
+    int ret = 0;
+
+    for (; start < end; start++) {
+        if (block_job_is_cancelled(&job->common)) {
+            ret = -1;
+            break;
+        }
+
+        if (backup_get_bitmap(bs, start)) {
+            continue; /* already copied */
+        }
+
+        /* we need to yield so that qemu_aio_flush() returns.
+         * (without, VM does not reboot)
+         * todo: can we avoid that?
+         */
+        co_sleep_ns(rt_clock, 0);
+        if (block_job_is_cancelled(&job->common)) {
+            ret = -1;
+            break;
+        }
+        DPRINTF("backup_run loop C%zd\n", start);
+
+        /**
+         * This triggers a cluster copy
+         * Note: avoid direct call to brdv_co_backup_cow, because
+         * this does not call tracked_request_begin()
+         */
+        ret = bdrv_co_backup(bs, start*BACKUP_BLOCKS_PER_CLUSTER, 1);
+        if (ret < 0) {
+            break;
+        }
+        /* Publish progress */
+        job->common.offset += BACKUP_CLUSTER_SIZE;
+    }
+
+    while (backup_in_progress_count > 0) {
+        DPRINTF("backup_run backup_in_progress_count != 0 (%d)",
+                backup_in_progress_count);
+        co_sleep_ns(rt_clock, 10000);
+    }
+
+    DPRINTF("backup_run complete %d\n", ret);
+    block_job_completed(&job->common, ret);
+}
+
+static void backup_job_cleanup_cb(void *opaque, int ret)
+{
+    BlockDriverState *bs = opaque;
+    assert(bs);
+    BackupBlockJob *job = (BackupBlockJob *)bs->job;
+    assert(job);
+
+    DPRINTF("backup_job_cleanup_cb start %d\n", ret);
+
+    job->backup_complete_cb(job->opaque, ret);
+
+    DPRINTF("backup_job_cleanup_cb end\n");
+
+    g_free(job->bitmap);
+}
+
+int
+bdrv_backup_init(BlockDriverState *bs, BackupDumpFunc *backup_dump_cb,
+                 BlockDriverCompletionFunc *backup_complete_cb,
+                 void *opaque)
+{
+    assert(bs);
+    assert(backup_dump_cb);
+    assert(backup_complete_cb);
+
+    if (bs->job) {
+        DPRINTF("bdrv_backup_init failed - running job on %s\n",
+                bdrv_get_device_name(bs));
+        return -1;
+    }
+
+    int64_t bitmap_size;
+    const char *devname = bdrv_get_device_name(bs);
+
+    if (!devname || !devname[0]) {
+        return -1;
+    }
+
+    DPRINTF("bdrv_backup_init %s\n", bdrv_get_device_name(bs));
+
+    Error *errp;
+    BackupBlockJob *job = block_job_create(&backup_job_type, bs, 0,
+                                           backup_job_cleanup_cb, bs, &errp);
+
+    job->common.cluster_size = BACKUP_CLUSTER_SIZE;
+
+    bitmap_size = bs->total_sectors +
+        BACKUP_BLOCKS_PER_CLUSTER * BITS_PER_LONG - 1;
+    bitmap_size /= BACKUP_BLOCKS_PER_CLUSTER * BITS_PER_LONG;
+
+    job->backup_dump_cb = backup_dump_cb;
+    job->backup_complete_cb = backup_complete_cb;
+    job->opaque = opaque;
+    job->bitmap_size = bitmap_size;
+    job->bitmap = g_new0(unsigned long, bitmap_size);
+
+    job->common.len = bs->total_sectors*BDRV_SECTOR_SIZE;
+    job->common.co = qemu_coroutine_create(backup_run);
+    qemu_coroutine_enter(job->common.co, job);
+
+    return 0;
+}
diff --git a/backup.h b/backup.h
new file mode 100644
index 0000000..577ac19
--- /dev/null
+++ b/backup.h
@@ -0,0 +1,30 @@ 
+/*
+ * QEMU backup related definitions
+ *
+ * Copyright (C) Proxmox Server Solutions
+ *
+ * Authors:
+ *  Dietmar Maurer (dietmar@proxmox.com)
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
+ * See the COPYING file in the top-level directory.
+ *
+ */
+
+#ifndef QEMU_BACKUP_H
+#define QEMU_BACKUP_H
+
+#include "block.h"
+
+#define BACKUP_CLUSTER_BITS 16
+#define BACKUP_CLUSTER_SIZE (1<<BACKUP_CLUSTER_BITS)
+#define BACKUP_BLOCKS_PER_CLUSTER (BACKUP_CLUSTER_SIZE/BDRV_SECTOR_SIZE)
+
+typedef int BackupDumpFunc(void *opaque, BlockDriverState *bs,
+                           int64_t cluster_num, unsigned char *buf);
+
+int bdrv_backup_init(BlockDriverState *bs, BackupDumpFunc *backup_dump_cb,
+		     BlockDriverCompletionFunc *backup_complete_cb,
+		     void *opaque);
+
+#endif /* QEMU_BACKUP_H */
diff --git a/block.c b/block.c
index c05875f..2f7c2eb 100644
--- a/block.c
+++ b/block.c
@@ -54,6 +54,7 @@ 
 typedef enum {
     BDRV_REQ_COPY_ON_READ = 0x1,
     BDRV_REQ_ZERO_WRITE   = 0x2,
+    BDRV_REQ_BACKUP_ONLY  = 0x4,
 } BdrvRequestFlags;
 
 static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load);
@@ -1542,7 +1543,7 @@  int bdrv_commit(BlockDriverState *bs)
 
     if (!drv)
         return -ENOMEDIUM;
-    
+
     if (!bs->backing_hd) {
         return -ENOTSUP;
     }
@@ -1679,6 +1680,22 @@  static void round_to_clusters(BlockDriverState *bs,
     }
 }
 
+/**
+ * Round a region to job cluster boundaries
+ */
+static void round_to_job_clusters(BlockDriverState *bs,
+                                  int64_t sector_num, int nb_sectors,
+                                  int job_cluster_size,
+                                  int64_t *cluster_sector_num,
+                                  int *cluster_nb_sectors)
+{
+    int64_t c = job_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 */
@@ -1693,7 +1710,9 @@  static bool tracked_request_overlaps(BdrvTrackedRequest *req,
 }
 
 static void coroutine_fn wait_for_overlapping_requests(BlockDriverState *bs,
-        int64_t sector_num, int nb_sectors)
+                                                       int64_t sector_num,
+                                                       int nb_sectors,
+                                                       int job_cluster_size)
 {
     BdrvTrackedRequest *req;
     int64_t cluster_sector_num;
@@ -1709,6 +1728,11 @@  static void coroutine_fn wait_for_overlapping_requests(BlockDriverState *bs,
     round_to_clusters(bs, sector_num, nb_sectors,
                       &cluster_sector_num, &cluster_nb_sectors);
 
+    if (job_cluster_size) {
+        round_to_job_clusters(bs, sector_num, nb_sectors, job_cluster_size,
+                              &cluster_sector_num, &cluster_nb_sectors);
+    }
+
     do {
         retry = false;
         QLIST_FOREACH(req, &bs->tracked_requests, list) {
@@ -2278,12 +2302,24 @@  static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs,
         bs->copy_on_read_in_flight++;
     }
 
-    if (bs->copy_on_read_in_flight) {
-        wait_for_overlapping_requests(bs, sector_num, nb_sectors);
+    int job_cluster_size = bs->job && bs->job->cluster_size ?
+        bs->job->cluster_size : 0;
+
+    if (bs->copy_on_read_in_flight || job_cluster_size) {
+        wait_for_overlapping_requests(bs, sector_num, nb_sectors,
+                                      job_cluster_size);
     }
 
     tracked_request_begin(&req, bs, sector_num, nb_sectors, false);
 
+    if (bs->job && bs->job->job_type->before_read) {
+        ret = bs->job->job_type->before_read(bs, sector_num, nb_sectors, qiov);
+        if (flags & BDRV_REQ_BACKUP_ONLY) {
+            /* Note: We do not return any data to the caller */
+            goto out;
+        }
+    }
+
     if (flags & BDRV_REQ_COPY_ON_READ) {
         int pnum;
 
@@ -2327,6 +2363,17 @@  int coroutine_fn bdrv_co_copy_on_readv(BlockDriverState *bs,
                             BDRV_REQ_COPY_ON_READ);
 }
 
+int coroutine_fn bdrv_co_backup(BlockDriverState *bs,
+    int64_t sector_num, int nb_sectors)
+{
+    if (!bs->job) {
+        return -ENOTSUP;
+    }
+
+    return bdrv_co_do_readv(bs, sector_num, nb_sectors, NULL,
+                            BDRV_REQ_BACKUP_ONLY);
+}
+
 static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
     int64_t sector_num, int nb_sectors)
 {
@@ -2384,12 +2431,23 @@  static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
         bdrv_io_limits_intercept(bs, true, nb_sectors);
     }
 
-    if (bs->copy_on_read_in_flight) {
-        wait_for_overlapping_requests(bs, sector_num, nb_sectors);
+    int job_cluster_size = bs->job && bs->job->cluster_size ?
+        bs->job->cluster_size : 0;
+
+    if (bs->copy_on_read_in_flight || job_cluster_size) {
+        wait_for_overlapping_requests(bs, sector_num, nb_sectors,
+                                      job_cluster_size);
     }
 
     tracked_request_begin(&req, bs, sector_num, nb_sectors, true);
 
+    if (bs->job && bs->job->job_type->before_write) {
+        ret = bs->job->job_type->before_write(bs, sector_num, nb_sectors, qiov);
+        if (ret < 0) {
+            goto out;
+        }
+    }
+
     if (flags & BDRV_REQ_ZERO_WRITE) {
         ret = bdrv_co_do_write_zeroes(bs, sector_num, nb_sectors);
     } else {
@@ -2408,6 +2466,7 @@  static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
         bs->wr_highest_sector = sector_num + nb_sectors - 1;
     }
 
+out:
     tracked_request_end(&req);
 
     return ret;
diff --git a/block.h b/block.h
index 722c620..94e5903 100644
--- a/block.h
+++ b/block.h
@@ -172,6 +172,8 @@  int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num,
     int nb_sectors, QEMUIOVector *qiov);
 int coroutine_fn bdrv_co_copy_on_readv(BlockDriverState *bs,
     int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);
+int coroutine_fn bdrv_co_backup(BlockDriverState *bs,
+    int64_t sector_num, int nb_sectors);
 int coroutine_fn bdrv_co_writev(BlockDriverState *bs, int64_t sector_num,
     int nb_sectors, QEMUIOVector *qiov);
 /*
diff --git a/blockjob.h b/blockjob.h
index 3792b73..1d3f4fd 100644
--- a/blockjob.h
+++ b/blockjob.h
@@ -50,6 +50,13 @@  typedef struct BlockJobType {
      * manually.
      */
     void (*complete)(BlockJob *job, Error **errp);
+
+    /** tracked requests */
+    int coroutine_fn (*before_read)(BlockDriverState *bs, int64_t sector_num, 
+                                    int nb_sectors, QEMUIOVector *qiov);
+    int coroutine_fn (*before_write)(BlockDriverState *bs, int64_t sector_num, 
+                                     int nb_sectors, QEMUIOVector *qiov);
+
 } BlockJobType;
 
 /**
@@ -103,6 +110,9 @@  struct BlockJob {
     /** Speed that was set with @block_job_set_speed.  */
     int64_t speed;
 
+    /** tracked requests */
+    int cluster_size;
+
     /** The completion function that will be called when the job completes.  */
     BlockDriverCompletionFunc *cb;