Re: [Qemu-devel] [PATCH v4 2/8] block: add basic backup support to block driver

[Kevin Wolf]

Am 16.05.2013 um 10:36 hat Stefan Hajnoczi geschrieben:
> From: Dietmar Maurer <address@hidden>
> 
> backup_start() creates a block job that copies a point-in-time snapshot
> of a block device to a target block device.
> 
> We call backup_do_cow() for each write during backup. That function
> reads the original data from the block device before it gets
> overwritten.  The data is then written to the target device.
> 
> Currently backup cluster size is hardcoded to 65536 bytes.
> 
> [I made a number of changes to Dietmar's original patch and folded them
> in to make code review easy.  Here is the full list:
> 
>  * Drop BackupDumpFunc interface in favor of a target block device
>  * Detect zero clusters with buffer_is_zero()
>  * Don't write zero clusters to the target
>  * Use 0 delay instead of 1us, like other block jobs
>  * Unify creation/start functions into backup_start()
>  * Simplify cleanup, free bitmap in backup_run() instead of cb
>  * function
>  * Use HBitmap to avoid duplicating bitmap code
>  * Use bdrv_getlength() instead of accessing ->total_sectors
>  * directly
>  * Delete the backup.h header file, it is no longer necessary
>  * Move ./backup.c to block/backup.c
>  * Remove #ifdefed out code
>  * Coding style and whitespace cleanups
>  * Use bdrv_add_before_write_notifier() instead of blockjob-specific hooks
>  * Keep our own in-flight CowRequest list instead of using block.c
>    tracked requests.  This means a little code duplication but is much
>    simpler than trying to share the tracked requests list and use the
>    backup block size.
> 
> -- stefanha]
> 
> Signed-off-by: Dietmar Maurer <address@hidden>
> Signed-off-by: Stefan Hajnoczi <address@hidden>
> ---
>  block/Makefile.objs       |   1 +
>  block/backup.c            | 283 
> ++++++++++++++++++++++++++++++++++++++++++++++
>  include/block/block_int.h |  16 +++
>  3 files changed, 300 insertions(+)
>  create mode 100644 block/backup.c
> 
> diff --git a/block/Makefile.objs b/block/Makefile.objs
> index 5f0358a..88bd101 100644
> --- a/block/Makefile.objs
> +++ b/block/Makefile.objs
> @@ -20,5 +20,6 @@ endif
>  common-obj-y += stream.o
>  common-obj-y += commit.o
>  common-obj-y += mirror.o
> +common-obj-y += backup.o
>  
>  $(obj)/curl.o: QEMU_CFLAGS+=$(CURL_CFLAGS)
> diff --git a/block/backup.c b/block/backup.c
> new file mode 100644
> index 0000000..5438e26
> --- /dev/null
> +++ b/block/backup.c
> @@ -0,0 +1,283 @@
> +/*
> + * QEMU backup
> + *
> + * Copyright (C) 2013 Proxmox Server Solutions
> + *
> + * Authors:
> + *  Dietmar Maurer (address@hidden)
> + *
> + * 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/block.h"
> +#include "block/block_int.h"
> +#include "block/blockjob.h"
> +#include "qemu/ratelimit.h"
> +
> +#define DEBUG_BACKUP 0
> +
> +#define DPRINTF(fmt, ...) \
> +    do { \
> +        if (DEBUG_BACKUP) { \
> +            fprintf(stderr, "backup: " fmt, ## __VA_ARGS__); \
> +        } \
> +    } while (0)
> +
> +#define BACKUP_CLUSTER_BITS 16
> +#define BACKUP_CLUSTER_SIZE (1 << BACKUP_CLUSTER_BITS)
> +#define BACKUP_BLOCKS_PER_CLUSTER (BACKUP_CLUSTER_SIZE / BDRV_SECTOR_SIZE)

BACKUP_SECTORS_PER_CLUSTER for more consistent naming?

> +
> +#define SLICE_TIME 100000000ULL /* ns */
> +
> +typedef struct CowRequest {
> +    int64_t start;
> +    int64_t end;
> +    QLIST_ENTRY(CowRequest) list;
> +    CoQueue wait_queue; /* coroutines blocked on this request */
> +} CowRequest;
> +
> +typedef struct BackupBlockJob {
> +    BlockJob common;
> +    BlockDriverState *target;
> +    RateLimit limit;
> +    CoRwlock flush_rwlock;
> +    uint64_t sectors_read;
> +    HBitmap *bitmap;
> +    QLIST_HEAD(, CowRequest) inflight_reqs;
> +} BackupBlockJob;
> +
> +/* See if in-flight requests overlap and wait for them to complete */
> +static void coroutine_fn wait_for_overlapping_requests(BackupBlockJob *job,
> +                                                       int64_t start,
> +                                                       int64_t end)
> +{
> +    CowRequest *req;
> +    bool retry;
> +
> +    do {
> +        retry = false;
> +        QLIST_FOREACH(req, &job->inflight_reqs, list) {
> +            if (end > req->start && start < req->end) {
> +                qemu_co_queue_wait(&req->wait_queue);
> +                retry = true;
> +                break;
> +            }
> +        }
> +    } while (retry);
> +}
> +
> +/* Keep track of an in-flight request */
> +static void cow_request_begin(CowRequest *req, BackupBlockJob *job,
> +                                     int64_t start, int64_t end)
> +{
> +    req->start = start;
> +    req->end = end;
> +    qemu_co_queue_init(&req->wait_queue);
> +    QLIST_INSERT_HEAD(&job->inflight_reqs, req, list);
> +}
> +
> +/* Forget about a completed request */
> +static void cow_request_end(CowRequest *req)
> +{
> +    QLIST_REMOVE(req, list);
> +    qemu_co_queue_restart_all(&req->wait_queue);
> +}
> +
> +static int coroutine_fn backup_do_cow(BlockDriverState *bs,
> +                                      int64_t sector_num, int nb_sectors)
> +{
> +    BackupBlockJob *job = (BackupBlockJob *)bs->job;
> +    CowRequest cow_request;
> +    struct iovec iov;
> +    QEMUIOVector bounce_qiov;
> +    void *bounce_buffer = NULL;
> +    int ret = 0;
> +    int64_t start, end;
> +
> +    qemu_co_rwlock_rdlock(&job->flush_rwlock);
> +
> +    start = sector_num / BACKUP_BLOCKS_PER_CLUSTER;
> +    end = DIV_ROUND_UP(sector_num + nb_sectors, BACKUP_BLOCKS_PER_CLUSTER);
> +
> +    DPRINTF("brdv_co_backup_cow enter %s C%" PRId64 " %" PRId64 " %d\n",
> +            bdrv_get_device_name(bs), start, sector_num, nb_sectors);
> +
> +    wait_for_overlapping_requests(job, start, end);
> +    cow_request_begin(&cow_request, job, start, end);
> +
> +    for (; start < end; start++) {
> +        if (hbitmap_get(job->bitmap, start)) {
> +            DPRINTF("brdv_co_backup_cow skip C%" PRId64 "\n", start);
> +            continue; /* already copied */
> +        }
> +
> +        /* immediately set bitmap (avoid coroutine race) */
> +        hbitmap_set(job->bitmap, start, 1);

Hm, what kind of race? Doesn't wait_for_overlapping_requests() already
serialise everything so that it doesn't matter where exactly the bit is
set, as long as it's between cow_request_begin/end?

> +
> +        DPRINTF("brdv_co_backup_cow C%" PRId64 "\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 = 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%" PRId64 " failed\n",
> +                    start);
> +            goto out;
> +        }
> +
> +        job->sectors_read += BACKUP_BLOCKS_PER_CLUSTER;
> +
> +        if (!buffer_is_zero(bounce_buffer, BACKUP_CLUSTER_SIZE)) {
> +            ret = bdrv_co_writev(job->target, start * 
> BACKUP_BLOCKS_PER_CLUSTER,
> +                                 BACKUP_BLOCKS_PER_CLUSTER,
> +                                 &bounce_qiov);
> +            if (ret < 0) {
> +                DPRINTF("brdv_co_backup_cow dump_cluster_cb C%" PRId64
> +                        " failed\n", start);
> +                goto out;
> +            }
> +        }

This series seems to only allow standalone target images without a
backing file, so this is okay. But we've been talking about use cases
like using the original image as a backing file and exposing the backup
via NBD in order to provide consistent inspection. Then we can't simply
ignore all-zero clusters.

Maybe we should use bdrv_co_write_zeroes() from the beginning?

> +
> +        DPRINTF("brdv_co_backup_cow done C%" PRId64 "\n", start);
> +    }
> +
> +out:
> +    if (bounce_buffer) {
> +        qemu_vfree(bounce_buffer);
> +    }
> +
> +    cow_request_end(&cow_request);
> +
> +    qemu_co_rwlock_unlock(&job->flush_rwlock);
> +
> +    return ret;
> +}
> +
> +static void coroutine_fn backup_before_write_notify(Notifier *notifier,
> +                                                    void *opaque)
> +{
> +    BdrvTrackedRequest *req = opaque;
> +    backup_do_cow(req->bs, req->sector_num, req->nb_sectors);
> +}

I don't think you can ignore errors here. Not sure if we can stop the VM
and resume later or something like that, but if we can't, the backup
will be invalid and we must fail the job.

Kevin