2012-10-18 16:49:23 +02:00
|
|
|
/*
|
|
|
|
* Image mirroring
|
|
|
|
*
|
|
|
|
* Copyright Red Hat, Inc. 2012
|
|
|
|
*
|
|
|
|
* Authors:
|
|
|
|
* Paolo Bonzini <pbonzini@redhat.com>
|
|
|
|
*
|
|
|
|
* 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.
|
|
|
|
*
|
|
|
|
*/
|
|
|
|
|
2016-01-18 19:01:42 +01:00
|
|
|
#include "qemu/osdep.h"
|
2012-10-18 16:49:23 +02:00
|
|
|
#include "trace.h"
|
2012-12-17 18:19:44 +01:00
|
|
|
#include "block/blockjob.h"
|
|
|
|
#include "block/block_int.h"
|
2015-10-19 17:53:22 +02:00
|
|
|
#include "sysemu/block-backend.h"
|
include/qemu/osdep.h: Don't include qapi/error.h
Commit 57cb38b included qapi/error.h into qemu/osdep.h to get the
Error typedef. Since then, we've moved to include qemu/osdep.h
everywhere. Its file comment explains: "To avoid getting into
possible circular include dependencies, this file should not include
any other QEMU headers, with the exceptions of config-host.h,
compiler.h, os-posix.h and os-win32.h, all of which are doing a
similar job to this file and are under similar constraints."
qapi/error.h doesn't do a similar job, and it doesn't adhere to
similar constraints: it includes qapi-types.h. That's in excess of
100KiB of crap most .c files don't actually need.
Add the typedef to qemu/typedefs.h, and include that instead of
qapi/error.h. Include qapi/error.h in .c files that need it and don't
get it now. Include qapi-types.h in qom/object.h for uint16List.
Update scripts/clean-includes accordingly. Update it further to match
reality: replace config.h by config-target.h, add sysemu/os-posix.h,
sysemu/os-win32.h. Update the list of includes in the qemu/osdep.h
comment quoted above similarly.
This reduces the number of objects depending on qapi/error.h from "all
of them" to less than a third. Unfortunately, the number depending on
qapi-types.h shrinks only a little. More work is needed for that one.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
[Fix compilation without the spice devel packages. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2016-03-14 09:01:28 +01:00
|
|
|
#include "qapi/error.h"
|
2015-03-17 17:22:46 +01:00
|
|
|
#include "qapi/qmp/qerror.h"
|
2012-10-18 16:49:23 +02:00
|
|
|
#include "qemu/ratelimit.h"
|
2013-01-21 17:09:43 +01:00
|
|
|
#include "qemu/bitmap.h"
|
2012-10-18 16:49:23 +02:00
|
|
|
|
2013-01-22 09:03:14 +01:00
|
|
|
#define SLICE_TIME 100000000ULL /* ns */
|
|
|
|
#define MAX_IN_FLIGHT 16
|
2015-05-15 09:51:36 +02:00
|
|
|
#define DEFAULT_MIRROR_BUF_SIZE (10 << 20)
|
2013-01-22 09:03:14 +01:00
|
|
|
|
|
|
|
/* The mirroring buffer is a list of granularity-sized chunks.
|
|
|
|
* Free chunks are organized in a list.
|
|
|
|
*/
|
|
|
|
typedef struct MirrorBuffer {
|
|
|
|
QSIMPLEQ_ENTRY(MirrorBuffer) next;
|
|
|
|
} MirrorBuffer;
|
2012-10-18 16:49:23 +02:00
|
|
|
|
|
|
|
typedef struct MirrorBlockJob {
|
|
|
|
BlockJob common;
|
|
|
|
RateLimit limit;
|
2016-04-12 16:17:41 +02:00
|
|
|
BlockBackend *target;
|
2013-12-16 07:45:29 +01:00
|
|
|
BlockDriverState *base;
|
2014-06-27 18:25:25 +02:00
|
|
|
/* The name of the graph node to replace */
|
|
|
|
char *replaces;
|
|
|
|
/* The BDS to replace */
|
|
|
|
BlockDriverState *to_replace;
|
|
|
|
/* Used to block operations on the drive-mirror-replace target */
|
|
|
|
Error *replace_blocker;
|
2013-12-16 07:45:30 +01:00
|
|
|
bool is_none_mode;
|
2012-10-18 16:49:28 +02:00
|
|
|
BlockdevOnError on_source_error, on_target_error;
|
2012-10-18 16:49:25 +02:00
|
|
|
bool synced;
|
|
|
|
bool should_complete;
|
2013-01-21 17:09:46 +01:00
|
|
|
int64_t granularity;
|
2013-01-21 17:09:43 +01:00
|
|
|
size_t buf_size;
|
2014-10-24 15:57:36 +02:00
|
|
|
int64_t bdev_length;
|
2013-01-21 17:09:43 +01:00
|
|
|
unsigned long *cow_bitmap;
|
2013-11-13 11:29:43 +01:00
|
|
|
BdrvDirtyBitmap *dirty_bitmap;
|
2013-01-21 17:09:41 +01:00
|
|
|
HBitmapIter hbi;
|
2012-10-18 16:49:23 +02:00
|
|
|
uint8_t *buf;
|
2013-01-22 09:03:14 +01:00
|
|
|
QSIMPLEQ_HEAD(, MirrorBuffer) buf_free;
|
|
|
|
int buf_free_count;
|
2013-01-22 09:03:12 +01:00
|
|
|
|
2013-01-22 09:03:14 +01:00
|
|
|
unsigned long *in_flight_bitmap;
|
2013-01-22 09:03:12 +01:00
|
|
|
int in_flight;
|
2014-10-24 15:57:36 +02:00
|
|
|
int sectors_in_flight;
|
2013-01-22 09:03:12 +01:00
|
|
|
int ret;
|
2015-06-08 07:56:08 +02:00
|
|
|
bool unmap;
|
2015-08-13 10:41:50 +02:00
|
|
|
bool waiting_for_io;
|
2016-02-05 03:00:29 +01:00
|
|
|
int target_cluster_sectors;
|
|
|
|
int max_iov;
|
2012-10-18 16:49:23 +02:00
|
|
|
} MirrorBlockJob;
|
|
|
|
|
2013-01-22 09:03:12 +01:00
|
|
|
typedef struct MirrorOp {
|
|
|
|
MirrorBlockJob *s;
|
|
|
|
QEMUIOVector qiov;
|
|
|
|
int64_t sector_num;
|
|
|
|
int nb_sectors;
|
|
|
|
} MirrorOp;
|
|
|
|
|
2012-10-18 16:49:28 +02:00
|
|
|
static BlockErrorAction mirror_error_action(MirrorBlockJob *s, bool read,
|
|
|
|
int error)
|
|
|
|
{
|
|
|
|
s->synced = false;
|
|
|
|
if (read) {
|
2016-04-18 11:36:38 +02:00
|
|
|
return block_job_error_action(&s->common, s->on_source_error,
|
|
|
|
true, error);
|
2012-10-18 16:49:28 +02:00
|
|
|
} else {
|
2016-04-18 11:36:38 +02:00
|
|
|
return block_job_error_action(&s->common, s->on_target_error,
|
|
|
|
false, error);
|
2012-10-18 16:49:28 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2013-01-22 09:03:12 +01:00
|
|
|
static void mirror_iteration_done(MirrorOp *op, int ret)
|
|
|
|
{
|
|
|
|
MirrorBlockJob *s = op->s;
|
2013-01-22 09:03:14 +01:00
|
|
|
struct iovec *iov;
|
2013-01-22 09:03:12 +01:00
|
|
|
int64_t chunk_num;
|
2013-01-22 09:03:14 +01:00
|
|
|
int i, nb_chunks, sectors_per_chunk;
|
2013-01-22 09:03:12 +01:00
|
|
|
|
|
|
|
trace_mirror_iteration_done(s, op->sector_num, op->nb_sectors, ret);
|
|
|
|
|
|
|
|
s->in_flight--;
|
2014-10-24 15:57:36 +02:00
|
|
|
s->sectors_in_flight -= op->nb_sectors;
|
2013-01-22 09:03:14 +01:00
|
|
|
iov = op->qiov.iov;
|
|
|
|
for (i = 0; i < op->qiov.niov; i++) {
|
|
|
|
MirrorBuffer *buf = (MirrorBuffer *) iov[i].iov_base;
|
|
|
|
QSIMPLEQ_INSERT_TAIL(&s->buf_free, buf, next);
|
|
|
|
s->buf_free_count++;
|
|
|
|
}
|
|
|
|
|
2013-01-22 09:03:12 +01:00
|
|
|
sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS;
|
|
|
|
chunk_num = op->sector_num / sectors_per_chunk;
|
2016-04-20 04:48:34 +02:00
|
|
|
nb_chunks = DIV_ROUND_UP(op->nb_sectors, sectors_per_chunk);
|
2013-01-22 09:03:14 +01:00
|
|
|
bitmap_clear(s->in_flight_bitmap, chunk_num, nb_chunks);
|
2014-10-24 15:57:36 +02:00
|
|
|
if (ret >= 0) {
|
|
|
|
if (s->cow_bitmap) {
|
|
|
|
bitmap_set(s->cow_bitmap, chunk_num, nb_chunks);
|
|
|
|
}
|
|
|
|
s->common.offset += (uint64_t)op->nb_sectors * BDRV_SECTOR_SIZE;
|
2013-01-22 09:03:12 +01:00
|
|
|
}
|
|
|
|
|
2014-01-23 08:59:16 +01:00
|
|
|
qemu_iovec_destroy(&op->qiov);
|
2015-10-01 13:04:39 +02:00
|
|
|
g_free(op);
|
2014-03-21 13:55:19 +01:00
|
|
|
|
2015-08-13 10:41:50 +02:00
|
|
|
if (s->waiting_for_io) {
|
2014-03-21 13:55:19 +01:00
|
|
|
qemu_coroutine_enter(s->common.co, NULL);
|
|
|
|
}
|
2013-01-22 09:03:12 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
static void mirror_write_complete(void *opaque, int ret)
|
|
|
|
{
|
|
|
|
MirrorOp *op = opaque;
|
|
|
|
MirrorBlockJob *s = op->s;
|
|
|
|
if (ret < 0) {
|
|
|
|
BlockErrorAction action;
|
|
|
|
|
2015-04-18 01:50:02 +02:00
|
|
|
bdrv_set_dirty_bitmap(s->dirty_bitmap, op->sector_num, op->nb_sectors);
|
2013-01-22 09:03:12 +01:00
|
|
|
action = mirror_error_action(s, false, -ret);
|
2014-06-18 08:43:30 +02:00
|
|
|
if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) {
|
2013-01-22 09:03:12 +01:00
|
|
|
s->ret = ret;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
mirror_iteration_done(op, ret);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void mirror_read_complete(void *opaque, int ret)
|
|
|
|
{
|
|
|
|
MirrorOp *op = opaque;
|
|
|
|
MirrorBlockJob *s = op->s;
|
|
|
|
if (ret < 0) {
|
|
|
|
BlockErrorAction action;
|
|
|
|
|
2015-04-18 01:50:02 +02:00
|
|
|
bdrv_set_dirty_bitmap(s->dirty_bitmap, op->sector_num, op->nb_sectors);
|
2013-01-22 09:03:12 +01:00
|
|
|
action = mirror_error_action(s, true, -ret);
|
2014-06-18 08:43:30 +02:00
|
|
|
if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) {
|
2013-01-22 09:03:12 +01:00
|
|
|
s->ret = ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
mirror_iteration_done(op, ret);
|
|
|
|
return;
|
|
|
|
}
|
2016-04-12 16:17:41 +02:00
|
|
|
blk_aio_pwritev(s->target, op->sector_num * BDRV_SECTOR_SIZE, &op->qiov,
|
|
|
|
op->nb_sectors * BDRV_SECTOR_SIZE,
|
2013-01-22 09:03:12 +01:00
|
|
|
mirror_write_complete, op);
|
|
|
|
}
|
|
|
|
|
2016-04-20 04:48:34 +02:00
|
|
|
static inline void mirror_clip_sectors(MirrorBlockJob *s,
|
|
|
|
int64_t sector_num,
|
|
|
|
int *nb_sectors)
|
|
|
|
{
|
|
|
|
*nb_sectors = MIN(*nb_sectors,
|
|
|
|
s->bdev_length / BDRV_SECTOR_SIZE - sector_num);
|
|
|
|
}
|
|
|
|
|
2016-02-05 03:00:29 +01:00
|
|
|
/* Round sector_num and/or nb_sectors to target cluster if COW is needed, and
|
|
|
|
* return the offset of the adjusted tail sector against original. */
|
|
|
|
static int mirror_cow_align(MirrorBlockJob *s,
|
|
|
|
int64_t *sector_num,
|
|
|
|
int *nb_sectors)
|
2012-10-18 16:49:23 +02:00
|
|
|
{
|
2016-02-05 03:00:29 +01:00
|
|
|
bool need_cow;
|
|
|
|
int ret = 0;
|
|
|
|
int chunk_sectors = s->granularity >> BDRV_SECTOR_BITS;
|
|
|
|
int64_t align_sector_num = *sector_num;
|
|
|
|
int align_nb_sectors = *nb_sectors;
|
|
|
|
int max_sectors = chunk_sectors * s->max_iov;
|
|
|
|
|
|
|
|
need_cow = !test_bit(*sector_num / chunk_sectors, s->cow_bitmap);
|
|
|
|
need_cow |= !test_bit((*sector_num + *nb_sectors - 1) / chunk_sectors,
|
|
|
|
s->cow_bitmap);
|
|
|
|
if (need_cow) {
|
2016-04-12 16:17:41 +02:00
|
|
|
bdrv_round_to_clusters(blk_bs(s->target), *sector_num, *nb_sectors,
|
2016-02-05 03:00:29 +01:00
|
|
|
&align_sector_num, &align_nb_sectors);
|
|
|
|
}
|
2015-07-09 11:56:47 +02:00
|
|
|
|
2016-02-05 03:00:29 +01:00
|
|
|
if (align_nb_sectors > max_sectors) {
|
|
|
|
align_nb_sectors = max_sectors;
|
|
|
|
if (need_cow) {
|
|
|
|
align_nb_sectors = QEMU_ALIGN_DOWN(align_nb_sectors,
|
|
|
|
s->target_cluster_sectors);
|
|
|
|
}
|
2013-01-21 17:09:41 +01:00
|
|
|
}
|
2016-04-20 04:48:34 +02:00
|
|
|
/* Clipping may result in align_nb_sectors unaligned to chunk boundary, but
|
|
|
|
* that doesn't matter because it's already the end of source image. */
|
|
|
|
mirror_clip_sectors(s, align_sector_num, &align_nb_sectors);
|
2013-01-21 17:09:41 +01:00
|
|
|
|
2016-02-05 03:00:29 +01:00
|
|
|
ret = align_sector_num + align_nb_sectors - (*sector_num + *nb_sectors);
|
|
|
|
*sector_num = align_sector_num;
|
|
|
|
*nb_sectors = align_nb_sectors;
|
|
|
|
assert(ret >= 0);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2016-02-05 03:00:30 +01:00
|
|
|
static inline void mirror_wait_for_io(MirrorBlockJob *s)
|
|
|
|
{
|
|
|
|
assert(!s->waiting_for_io);
|
|
|
|
s->waiting_for_io = true;
|
|
|
|
qemu_coroutine_yield();
|
|
|
|
s->waiting_for_io = false;
|
|
|
|
}
|
|
|
|
|
2016-02-05 03:00:29 +01:00
|
|
|
/* Submit async read while handling COW.
|
|
|
|
* Returns: nb_sectors if no alignment is necessary, or
|
|
|
|
* (new_end - sector_num) if tail is rounded up or down due to
|
|
|
|
* alignment or buffer limit.
|
|
|
|
*/
|
|
|
|
static int mirror_do_read(MirrorBlockJob *s, int64_t sector_num,
|
|
|
|
int nb_sectors)
|
|
|
|
{
|
2016-04-12 16:17:41 +02:00
|
|
|
BlockBackend *source = s->common.blk;
|
2016-02-05 03:00:29 +01:00
|
|
|
int sectors_per_chunk, nb_chunks;
|
|
|
|
int ret = nb_sectors;
|
|
|
|
MirrorOp *op;
|
|
|
|
|
2013-01-22 09:03:15 +01:00
|
|
|
sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS;
|
2013-01-22 09:03:14 +01:00
|
|
|
|
2016-02-05 03:00:29 +01:00
|
|
|
/* We can only handle as much as buf_size at a time. */
|
|
|
|
nb_sectors = MIN(s->buf_size >> BDRV_SECTOR_BITS, nb_sectors);
|
|
|
|
assert(nb_sectors);
|
2013-01-22 09:03:14 +01:00
|
|
|
|
2016-02-05 03:00:29 +01:00
|
|
|
if (s->cow_bitmap) {
|
|
|
|
ret += mirror_cow_align(s, §or_num, &nb_sectors);
|
|
|
|
}
|
|
|
|
assert(nb_sectors << BDRV_SECTOR_BITS <= s->buf_size);
|
|
|
|
/* The sector range must meet granularity because:
|
|
|
|
* 1) Caller passes in aligned values;
|
|
|
|
* 2) mirror_cow_align is used only when target cluster is larger. */
|
|
|
|
assert(!(sector_num % sectors_per_chunk));
|
2016-04-20 04:48:34 +02:00
|
|
|
nb_chunks = DIV_ROUND_UP(nb_sectors, sectors_per_chunk);
|
2016-02-05 03:00:29 +01:00
|
|
|
|
|
|
|
while (s->buf_free_count < nb_chunks) {
|
2013-01-22 09:03:14 +01:00
|
|
|
trace_mirror_yield_in_flight(s, sector_num, s->in_flight);
|
2016-02-05 03:00:30 +01:00
|
|
|
mirror_wait_for_io(s);
|
2013-01-21 17:09:43 +01:00
|
|
|
}
|
|
|
|
|
2013-01-22 09:03:12 +01:00
|
|
|
/* Allocate a MirrorOp that is used as an AIO callback. */
|
2015-10-01 13:04:39 +02:00
|
|
|
op = g_new(MirrorOp, 1);
|
2013-01-22 09:03:12 +01:00
|
|
|
op->s = s;
|
|
|
|
op->sector_num = sector_num;
|
|
|
|
op->nb_sectors = nb_sectors;
|
2013-01-22 09:03:14 +01:00
|
|
|
|
|
|
|
/* Now make a QEMUIOVector taking enough granularity-sized chunks
|
|
|
|
* from s->buf_free.
|
|
|
|
*/
|
|
|
|
qemu_iovec_init(&op->qiov, nb_chunks);
|
|
|
|
while (nb_chunks-- > 0) {
|
|
|
|
MirrorBuffer *buf = QSIMPLEQ_FIRST(&s->buf_free);
|
2016-02-05 03:00:29 +01:00
|
|
|
size_t remaining = nb_sectors * BDRV_SECTOR_SIZE - op->qiov.size;
|
2014-07-01 16:52:21 +02:00
|
|
|
|
2013-01-22 09:03:14 +01:00
|
|
|
QSIMPLEQ_REMOVE_HEAD(&s->buf_free, next);
|
|
|
|
s->buf_free_count--;
|
2014-07-01 16:52:21 +02:00
|
|
|
qemu_iovec_add(&op->qiov, buf, MIN(s->granularity, remaining));
|
2013-01-22 09:03:14 +01:00
|
|
|
}
|
2013-01-22 09:03:12 +01:00
|
|
|
|
2012-10-18 16:49:23 +02:00
|
|
|
/* Copy the dirty cluster. */
|
2013-01-22 09:03:12 +01:00
|
|
|
s->in_flight++;
|
2014-10-24 15:57:36 +02:00
|
|
|
s->sectors_in_flight += nb_sectors;
|
2013-01-21 17:09:43 +01:00
|
|
|
trace_mirror_one_iteration(s, sector_num, nb_sectors);
|
2015-06-08 07:56:09 +02:00
|
|
|
|
2016-04-12 16:17:41 +02:00
|
|
|
blk_aio_preadv(source, sector_num * BDRV_SECTOR_SIZE, &op->qiov,
|
|
|
|
nb_sectors * BDRV_SECTOR_SIZE,
|
2016-02-05 03:00:29 +01:00
|
|
|
mirror_read_complete, op);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void mirror_do_zero_or_discard(MirrorBlockJob *s,
|
|
|
|
int64_t sector_num,
|
|
|
|
int nb_sectors,
|
|
|
|
bool is_discard)
|
|
|
|
{
|
|
|
|
MirrorOp *op;
|
|
|
|
|
|
|
|
/* Allocate a MirrorOp that is used as an AIO callback. The qiov is zeroed
|
|
|
|
* so the freeing in mirror_iteration_done is nop. */
|
|
|
|
op = g_new0(MirrorOp, 1);
|
|
|
|
op->s = s;
|
|
|
|
op->sector_num = sector_num;
|
|
|
|
op->nb_sectors = nb_sectors;
|
|
|
|
|
|
|
|
s->in_flight++;
|
|
|
|
s->sectors_in_flight += nb_sectors;
|
|
|
|
if (is_discard) {
|
2016-04-12 16:17:41 +02:00
|
|
|
blk_aio_discard(s->target, sector_num, op->nb_sectors,
|
|
|
|
mirror_write_complete, op);
|
2016-02-05 03:00:29 +01:00
|
|
|
} else {
|
2016-04-12 16:17:41 +02:00
|
|
|
blk_aio_pwrite_zeroes(s->target, sector_num * BDRV_SECTOR_SIZE,
|
|
|
|
op->nb_sectors * BDRV_SECTOR_SIZE,
|
2015-06-08 07:56:09 +02:00
|
|
|
s->unmap ? BDRV_REQ_MAY_UNMAP : 0,
|
|
|
|
mirror_write_complete, op);
|
2016-02-05 03:00:29 +01:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s)
|
|
|
|
{
|
2016-04-12 16:17:41 +02:00
|
|
|
BlockDriverState *source = blk_bs(s->common.blk);
|
2016-04-20 00:59:47 +02:00
|
|
|
int64_t sector_num, first_chunk;
|
2016-02-05 03:00:29 +01:00
|
|
|
uint64_t delay_ns = 0;
|
|
|
|
/* At least the first dirty chunk is mirrored in one iteration. */
|
|
|
|
int nb_chunks = 1;
|
|
|
|
int64_t end = s->bdev_length / BDRV_SECTOR_SIZE;
|
|
|
|
int sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS;
|
|
|
|
|
|
|
|
sector_num = hbitmap_iter_next(&s->hbi);
|
|
|
|
if (sector_num < 0) {
|
|
|
|
bdrv_dirty_iter_init(s->dirty_bitmap, &s->hbi);
|
|
|
|
sector_num = hbitmap_iter_next(&s->hbi);
|
|
|
|
trace_mirror_restart_iter(s, bdrv_get_dirty_count(s->dirty_bitmap));
|
|
|
|
assert(sector_num >= 0);
|
|
|
|
}
|
|
|
|
|
2016-04-20 00:59:47 +02:00
|
|
|
first_chunk = sector_num / sectors_per_chunk;
|
|
|
|
while (test_bit(first_chunk, s->in_flight_bitmap)) {
|
|
|
|
trace_mirror_yield_in_flight(s, first_chunk, s->in_flight);
|
|
|
|
mirror_wait_for_io(s);
|
|
|
|
}
|
|
|
|
|
2016-02-05 03:00:29 +01:00
|
|
|
/* Find the number of consective dirty chunks following the first dirty
|
|
|
|
* one, and wait for in flight requests in them. */
|
|
|
|
while (nb_chunks * sectors_per_chunk < (s->buf_size >> BDRV_SECTOR_BITS)) {
|
|
|
|
int64_t hbitmap_next;
|
|
|
|
int64_t next_sector = sector_num + nb_chunks * sectors_per_chunk;
|
|
|
|
int64_t next_chunk = next_sector / sectors_per_chunk;
|
|
|
|
if (next_sector >= end ||
|
|
|
|
!bdrv_get_dirty(source, s->dirty_bitmap, next_sector)) {
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
if (test_bit(next_chunk, s->in_flight_bitmap)) {
|
2016-04-20 00:59:47 +02:00
|
|
|
break;
|
2016-02-05 03:00:29 +01:00
|
|
|
}
|
2016-04-20 00:59:47 +02:00
|
|
|
|
|
|
|
hbitmap_next = hbitmap_iter_next(&s->hbi);
|
2016-04-20 00:59:48 +02:00
|
|
|
if (hbitmap_next > next_sector || hbitmap_next < 0) {
|
|
|
|
/* The bitmap iterator's cache is stale, refresh it */
|
|
|
|
bdrv_set_dirty_iter(&s->hbi, next_sector);
|
|
|
|
hbitmap_next = hbitmap_iter_next(&s->hbi);
|
|
|
|
}
|
2016-04-20 00:59:47 +02:00
|
|
|
assert(hbitmap_next == next_sector);
|
|
|
|
nb_chunks++;
|
2016-02-05 03:00:29 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Clear dirty bits before querying the block status, because
|
|
|
|
* calling bdrv_get_block_status_above could yield - if some blocks are
|
|
|
|
* marked dirty in this window, we need to know.
|
|
|
|
*/
|
|
|
|
bdrv_reset_dirty_bitmap(s->dirty_bitmap, sector_num,
|
|
|
|
nb_chunks * sectors_per_chunk);
|
|
|
|
bitmap_set(s->in_flight_bitmap, sector_num / sectors_per_chunk, nb_chunks);
|
|
|
|
while (nb_chunks > 0 && sector_num < end) {
|
|
|
|
int ret;
|
|
|
|
int io_sectors;
|
|
|
|
BlockDriverState *file;
|
|
|
|
enum MirrorMethod {
|
|
|
|
MIRROR_METHOD_COPY,
|
|
|
|
MIRROR_METHOD_ZERO,
|
|
|
|
MIRROR_METHOD_DISCARD
|
|
|
|
} mirror_method = MIRROR_METHOD_COPY;
|
|
|
|
|
|
|
|
assert(!(sector_num % sectors_per_chunk));
|
|
|
|
ret = bdrv_get_block_status_above(source, NULL, sector_num,
|
|
|
|
nb_chunks * sectors_per_chunk,
|
|
|
|
&io_sectors, &file);
|
|
|
|
if (ret < 0) {
|
|
|
|
io_sectors = nb_chunks * sectors_per_chunk;
|
|
|
|
}
|
|
|
|
|
|
|
|
io_sectors -= io_sectors % sectors_per_chunk;
|
|
|
|
if (io_sectors < sectors_per_chunk) {
|
|
|
|
io_sectors = sectors_per_chunk;
|
|
|
|
} else if (ret >= 0 && !(ret & BDRV_BLOCK_DATA)) {
|
|
|
|
int64_t target_sector_num;
|
|
|
|
int target_nb_sectors;
|
2016-04-12 16:17:41 +02:00
|
|
|
bdrv_round_to_clusters(blk_bs(s->target), sector_num, io_sectors,
|
2016-02-05 03:00:29 +01:00
|
|
|
&target_sector_num, &target_nb_sectors);
|
|
|
|
if (target_sector_num == sector_num &&
|
|
|
|
target_nb_sectors == io_sectors) {
|
|
|
|
mirror_method = ret & BDRV_BLOCK_ZERO ?
|
|
|
|
MIRROR_METHOD_ZERO :
|
|
|
|
MIRROR_METHOD_DISCARD;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2016-04-20 04:48:34 +02:00
|
|
|
mirror_clip_sectors(s, sector_num, &io_sectors);
|
2016-02-05 03:00:29 +01:00
|
|
|
switch (mirror_method) {
|
|
|
|
case MIRROR_METHOD_COPY:
|
|
|
|
io_sectors = mirror_do_read(s, sector_num, io_sectors);
|
|
|
|
break;
|
|
|
|
case MIRROR_METHOD_ZERO:
|
|
|
|
mirror_do_zero_or_discard(s, sector_num, io_sectors, false);
|
|
|
|
break;
|
|
|
|
case MIRROR_METHOD_DISCARD:
|
|
|
|
mirror_do_zero_or_discard(s, sector_num, io_sectors, true);
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
abort();
|
|
|
|
}
|
|
|
|
assert(io_sectors);
|
|
|
|
sector_num += io_sectors;
|
2016-04-20 04:48:34 +02:00
|
|
|
nb_chunks -= DIV_ROUND_UP(io_sectors, sectors_per_chunk);
|
2016-02-05 03:00:29 +01:00
|
|
|
delay_ns += ratelimit_calculate_delay(&s->limit, io_sectors);
|
2015-06-08 07:56:09 +02:00
|
|
|
}
|
2014-03-21 13:55:18 +01:00
|
|
|
return delay_ns;
|
2013-01-22 09:03:12 +01:00
|
|
|
}
|
2012-10-18 16:49:28 +02:00
|
|
|
|
2013-01-22 09:03:14 +01:00
|
|
|
static void mirror_free_init(MirrorBlockJob *s)
|
|
|
|
{
|
|
|
|
int granularity = s->granularity;
|
|
|
|
size_t buf_size = s->buf_size;
|
|
|
|
uint8_t *buf = s->buf;
|
|
|
|
|
|
|
|
assert(s->buf_free_count == 0);
|
|
|
|
QSIMPLEQ_INIT(&s->buf_free);
|
|
|
|
while (buf_size != 0) {
|
|
|
|
MirrorBuffer *cur = (MirrorBuffer *)buf;
|
|
|
|
QSIMPLEQ_INSERT_TAIL(&s->buf_free, cur, next);
|
|
|
|
s->buf_free_count++;
|
|
|
|
buf_size -= granularity;
|
|
|
|
buf += granularity;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2013-01-22 09:03:12 +01:00
|
|
|
static void mirror_drain(MirrorBlockJob *s)
|
|
|
|
{
|
|
|
|
while (s->in_flight > 0) {
|
2016-02-05 03:00:30 +01:00
|
|
|
mirror_wait_for_io(s);
|
2013-01-22 09:03:12 +01:00
|
|
|
}
|
2012-10-18 16:49:23 +02:00
|
|
|
}
|
|
|
|
|
2014-10-21 13:03:58 +02:00
|
|
|
typedef struct {
|
|
|
|
int ret;
|
|
|
|
} MirrorExitData;
|
|
|
|
|
|
|
|
static void mirror_exit(BlockJob *job, void *opaque)
|
|
|
|
{
|
|
|
|
MirrorBlockJob *s = container_of(job, MirrorBlockJob, common);
|
|
|
|
MirrorExitData *data = opaque;
|
|
|
|
AioContext *replace_aio_context = NULL;
|
2016-04-12 16:17:41 +02:00
|
|
|
BlockDriverState *src = blk_bs(s->common.blk);
|
|
|
|
BlockDriverState *target_bs = blk_bs(s->target);
|
2015-09-15 11:58:23 +02:00
|
|
|
|
|
|
|
/* Make sure that the source BDS doesn't go away before we called
|
|
|
|
* block_job_completed(). */
|
|
|
|
bdrv_ref(src);
|
2014-10-21 13:03:58 +02:00
|
|
|
|
|
|
|
if (s->to_replace) {
|
|
|
|
replace_aio_context = bdrv_get_aio_context(s->to_replace);
|
|
|
|
aio_context_acquire(replace_aio_context);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (s->should_complete && data->ret == 0) {
|
2016-04-12 16:17:41 +02:00
|
|
|
BlockDriverState *to_replace = src;
|
2014-10-21 13:03:58 +02:00
|
|
|
if (s->to_replace) {
|
|
|
|
to_replace = s->to_replace;
|
|
|
|
}
|
2015-10-28 13:24:26 +01:00
|
|
|
|
2016-04-12 16:17:41 +02:00
|
|
|
if (bdrv_get_flags(target_bs) != bdrv_get_flags(to_replace)) {
|
|
|
|
bdrv_reopen(target_bs, bdrv_get_flags(to_replace), NULL);
|
2014-10-21 13:03:58 +02:00
|
|
|
}
|
2016-04-12 16:20:59 +02:00
|
|
|
|
|
|
|
/* The mirror job has no requests in flight any more, but we need to
|
|
|
|
* drain potential other users of the BDS before changing the graph. */
|
2016-04-12 16:17:41 +02:00
|
|
|
bdrv_drained_begin(target_bs);
|
|
|
|
bdrv_replace_in_backing_chain(to_replace, target_bs);
|
|
|
|
bdrv_drained_end(target_bs);
|
2016-04-12 16:20:59 +02:00
|
|
|
|
2016-04-08 14:51:09 +02:00
|
|
|
/* We just changed the BDS the job BB refers to */
|
|
|
|
blk_remove_bs(job->blk);
|
|
|
|
blk_insert_bs(job->blk, src);
|
2014-10-21 13:03:58 +02:00
|
|
|
}
|
|
|
|
if (s->to_replace) {
|
|
|
|
bdrv_op_unblock_all(s->to_replace, s->replace_blocker);
|
|
|
|
error_free(s->replace_blocker);
|
|
|
|
bdrv_unref(s->to_replace);
|
|
|
|
}
|
|
|
|
if (replace_aio_context) {
|
|
|
|
aio_context_release(replace_aio_context);
|
|
|
|
}
|
|
|
|
g_free(s->replaces);
|
2016-04-12 16:17:41 +02:00
|
|
|
bdrv_op_unblock_all(target_bs, s->common.blocker);
|
|
|
|
blk_unref(s->target);
|
2014-10-21 13:03:58 +02:00
|
|
|
block_job_completed(&s->common, data->ret);
|
|
|
|
g_free(data);
|
2015-11-23 03:28:04 +01:00
|
|
|
bdrv_drained_end(src);
|
mirror: Workaround for unexpected iohandler events during completion
Commit 5a7e7a0ba moved mirror_exit to a BH handler but didn't add any
protection against new requests that could sneak in just before the
BH is dispatched. For example (assuming a code base at that commit):
main_loop_wait # 1
os_host_main_loop_wait
g_main_context_dispatch
aio_ctx_dispatch
aio_dispatch
...
mirror_run
bdrv_drain
(a) block_job_defer_to_main_loop
qemu_iohandler_poll
virtio_queue_host_notifier_read
...
virtio_submit_multiwrite
(b) blk_aio_multiwrite
main_loop_wait # 2
<snip>
aio_dispatch
aio_bh_poll
(c) mirror_exit
At (a) we know the BDS has no pending request. However, the same
main_loop_wait call is going to dispatch iohandlers (EventNotifier
events), which may lead to a new I/O from guest. So the invariant is
already broken at (c). Data loss.
Commit f3926945c8 made iohandler to use aio API. The order of
virtio_queue_host_notifier_read and block_job_defer_to_main_loop within
a main_loop_wait becomes unpredictable, and even worse, if the host
notifier event arrives at the next main_loop_wait call, the
unpredictable order between mirror_exit and
virtio_queue_host_notifier_read is also a trouble. As shown below, this
commit made the bug easier to trigger:
- Bug case 1:
main_loop_wait # 1
os_host_main_loop_wait
g_main_context_dispatch
aio_ctx_dispatch (qemu_aio_context)
...
mirror_run
bdrv_drain
(a) block_job_defer_to_main_loop
aio_ctx_dispatch (iohandler_ctx)
virtio_queue_host_notifier_read
...
virtio_submit_multiwrite
(b) blk_aio_multiwrite
main_loop_wait # 2
...
aio_dispatch
aio_bh_poll
(c) mirror_exit
- Bug case 2:
main_loop_wait # 1
os_host_main_loop_wait
g_main_context_dispatch
aio_ctx_dispatch (qemu_aio_context)
...
mirror_run
bdrv_drain
(a) block_job_defer_to_main_loop
main_loop_wait # 2
...
aio_ctx_dispatch (iohandler_ctx)
virtio_queue_host_notifier_read
...
virtio_submit_multiwrite
(b) blk_aio_multiwrite
aio_dispatch
aio_bh_poll
(c) mirror_exit
In both cases, (b) breaks the invariant wanted by (a) and (c).
Until then, the request loss has been silent. Later, 3f09bfbc7be added
asserts at (c) to check the invariant (in
bdrv_replace_in_backing_chain), and Max reported an assertion failure
first visible there, by doing active committing while the guest is
running bonnie++.
2.5 added bdrv_drained_begin at (a) to protect the dataplane case from
similar problems, but we never realize the main loop bug until now.
As a bandage, this patch disables iohandler's external events
temporarily together with bs->ctx.
Launchpad Bug: 1570134
Cc: qemu-stable@nongnu.org
Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Jeff Cody <jcody@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2016-04-22 15:53:56 +02:00
|
|
|
if (qemu_get_aio_context() == bdrv_get_aio_context(src)) {
|
|
|
|
aio_enable_external(iohandler_get_aio_context());
|
|
|
|
}
|
2015-09-15 11:58:23 +02:00
|
|
|
bdrv_unref(src);
|
2014-10-21 13:03:58 +02:00
|
|
|
}
|
|
|
|
|
2012-10-18 16:49:23 +02:00
|
|
|
static void coroutine_fn mirror_run(void *opaque)
|
|
|
|
{
|
|
|
|
MirrorBlockJob *s = opaque;
|
2014-10-21 13:03:58 +02:00
|
|
|
MirrorExitData *data;
|
2016-04-12 16:17:41 +02:00
|
|
|
BlockDriverState *bs = blk_bs(s->common.blk);
|
|
|
|
BlockDriverState *target_bs = blk_bs(s->target);
|
2015-07-09 05:47:58 +02:00
|
|
|
int64_t sector_num, end, length;
|
2013-01-22 09:03:12 +01:00
|
|
|
uint64_t last_pause_ns;
|
2013-01-21 17:09:43 +01:00
|
|
|
BlockDriverInfo bdi;
|
2015-01-22 14:03:29 +01:00
|
|
|
char backing_filename[2]; /* we only need 2 characters because we are only
|
|
|
|
checking for a NULL string */
|
2012-10-18 16:49:23 +02:00
|
|
|
int ret = 0;
|
|
|
|
int n;
|
2016-02-05 03:00:29 +01:00
|
|
|
int target_cluster_size = BDRV_SECTOR_SIZE;
|
2012-10-18 16:49:23 +02:00
|
|
|
|
|
|
|
if (block_job_is_cancelled(&s->common)) {
|
|
|
|
goto immediate_exit;
|
|
|
|
}
|
|
|
|
|
2014-10-24 15:57:36 +02:00
|
|
|
s->bdev_length = bdrv_getlength(bs);
|
|
|
|
if (s->bdev_length < 0) {
|
|
|
|
ret = s->bdev_length;
|
2014-04-29 12:09:09 +02:00
|
|
|
goto immediate_exit;
|
2014-10-24 15:57:36 +02:00
|
|
|
} else if (s->bdev_length == 0) {
|
2014-06-24 14:26:36 +02:00
|
|
|
/* Report BLOCK_JOB_READY and wait for complete. */
|
|
|
|
block_job_event_ready(&s->common);
|
|
|
|
s->synced = true;
|
|
|
|
while (!block_job_is_cancelled(&s->common) && !s->should_complete) {
|
|
|
|
block_job_yield(&s->common);
|
|
|
|
}
|
|
|
|
s->common.cancelled = false;
|
|
|
|
goto immediate_exit;
|
2012-10-18 16:49:23 +02:00
|
|
|
}
|
|
|
|
|
2014-10-24 15:57:36 +02:00
|
|
|
length = DIV_ROUND_UP(s->bdev_length, s->granularity);
|
2013-01-22 09:03:14 +01:00
|
|
|
s->in_flight_bitmap = bitmap_new(length);
|
|
|
|
|
2013-01-21 17:09:43 +01:00
|
|
|
/* If we have no backing file yet in the destination, we cannot let
|
|
|
|
* the destination do COW. Instead, we copy sectors around the
|
|
|
|
* dirty data if needed. We need a bitmap to do that.
|
|
|
|
*/
|
2016-04-12 16:17:41 +02:00
|
|
|
bdrv_get_backing_filename(target_bs, backing_filename,
|
2013-01-21 17:09:43 +01:00
|
|
|
sizeof(backing_filename));
|
2016-04-12 16:17:41 +02:00
|
|
|
if (!bdrv_get_info(target_bs, &bdi) && bdi.cluster_size) {
|
2016-02-05 03:00:29 +01:00
|
|
|
target_cluster_size = bdi.cluster_size;
|
|
|
|
}
|
2016-04-12 16:17:41 +02:00
|
|
|
if (backing_filename[0] && !target_bs->backing
|
2016-02-05 03:00:29 +01:00
|
|
|
&& s->granularity < target_cluster_size) {
|
|
|
|
s->buf_size = MAX(s->buf_size, target_cluster_size);
|
|
|
|
s->cow_bitmap = bitmap_new(length);
|
2013-01-21 17:09:43 +01:00
|
|
|
}
|
2016-02-05 03:00:29 +01:00
|
|
|
s->target_cluster_sectors = target_cluster_size >> BDRV_SECTOR_BITS;
|
2016-04-12 16:17:41 +02:00
|
|
|
s->max_iov = MIN(bs->bl.max_iov, target_bs->bl.max_iov);
|
2013-01-21 17:09:43 +01:00
|
|
|
|
2014-10-24 15:57:36 +02:00
|
|
|
end = s->bdev_length / BDRV_SECTOR_SIZE;
|
2014-05-21 18:16:21 +02:00
|
|
|
s->buf = qemu_try_blockalign(bs, s->buf_size);
|
|
|
|
if (s->buf == NULL) {
|
|
|
|
ret = -ENOMEM;
|
|
|
|
goto immediate_exit;
|
|
|
|
}
|
|
|
|
|
2013-01-22 09:03:14 +01:00
|
|
|
mirror_free_init(s);
|
2012-10-18 16:49:23 +02:00
|
|
|
|
2015-05-13 05:11:13 +02:00
|
|
|
last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
|
2013-12-16 07:45:30 +01:00
|
|
|
if (!s->is_none_mode) {
|
2012-10-18 16:49:23 +02:00
|
|
|
/* First part, loop on the sectors and initialize the dirty bitmap. */
|
2013-12-16 07:45:29 +01:00
|
|
|
BlockDriverState *base = s->base;
|
2016-04-12 16:17:41 +02:00
|
|
|
bool mark_all_dirty = s->base == NULL && !bdrv_has_zero_init(target_bs);
|
2015-10-01 06:06:37 +02:00
|
|
|
|
2012-10-18 16:49:23 +02:00
|
|
|
for (sector_num = 0; sector_num < end; ) {
|
2015-07-09 05:47:58 +02:00
|
|
|
/* Just to make sure we are not exceeding int limit. */
|
|
|
|
int nb_sectors = MIN(INT_MAX >> BDRV_SECTOR_BITS,
|
|
|
|
end - sector_num);
|
2015-05-13 05:11:13 +02:00
|
|
|
int64_t now = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
|
|
|
|
|
|
|
|
if (now - last_pause_ns > SLICE_TIME) {
|
|
|
|
last_pause_ns = now;
|
|
|
|
block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, 0);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (block_job_is_cancelled(&s->common)) {
|
|
|
|
goto immediate_exit;
|
|
|
|
}
|
|
|
|
|
2015-07-09 05:47:58 +02:00
|
|
|
ret = bdrv_is_allocated_above(bs, base, sector_num, nb_sectors, &n);
|
2012-10-18 16:49:23 +02:00
|
|
|
|
|
|
|
if (ret < 0) {
|
|
|
|
goto immediate_exit;
|
|
|
|
}
|
|
|
|
|
|
|
|
assert(n > 0);
|
2015-10-01 06:06:37 +02:00
|
|
|
if (ret == 1 || mark_all_dirty) {
|
2015-04-18 01:50:02 +02:00
|
|
|
bdrv_set_dirty_bitmap(s->dirty_bitmap, sector_num, n);
|
2012-10-18 16:49:23 +02:00
|
|
|
}
|
2015-07-09 05:47:58 +02:00
|
|
|
sector_num += n;
|
2012-10-18 16:49:23 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2015-04-18 01:50:02 +02:00
|
|
|
bdrv_dirty_iter_init(s->dirty_bitmap, &s->hbi);
|
2012-10-18 16:49:23 +02:00
|
|
|
for (;;) {
|
2014-03-21 13:55:18 +01:00
|
|
|
uint64_t delay_ns = 0;
|
2012-10-18 16:49:23 +02:00
|
|
|
int64_t cnt;
|
|
|
|
bool should_complete;
|
|
|
|
|
2013-01-22 09:03:12 +01:00
|
|
|
if (s->ret < 0) {
|
|
|
|
ret = s->ret;
|
|
|
|
goto immediate_exit;
|
|
|
|
}
|
|
|
|
|
2015-04-18 01:50:02 +02:00
|
|
|
cnt = bdrv_get_dirty_count(s->dirty_bitmap);
|
2014-10-24 15:57:36 +02:00
|
|
|
/* s->common.offset contains the number of bytes already processed so
|
|
|
|
* far, cnt is the number of dirty sectors remaining and
|
|
|
|
* s->sectors_in_flight is the number of sectors currently being
|
|
|
|
* processed; together those are the current total operation length */
|
|
|
|
s->common.len = s->common.offset +
|
|
|
|
(cnt + s->sectors_in_flight) * BDRV_SECTOR_SIZE;
|
2013-01-22 09:03:12 +01:00
|
|
|
|
|
|
|
/* Note that even when no rate limit is applied we need to yield
|
2015-04-03 16:05:21 +02:00
|
|
|
* periodically with no pending I/O so that bdrv_drain_all() returns.
|
2013-01-22 09:03:12 +01:00
|
|
|
* We do so every SLICE_TIME nanoseconds, or when there is an error,
|
|
|
|
* or when the source is clean, whichever comes first.
|
|
|
|
*/
|
2013-08-21 17:03:08 +02:00
|
|
|
if (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - last_pause_ns < SLICE_TIME &&
|
2013-01-22 09:03:12 +01:00
|
|
|
s->common.iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
|
2013-01-22 09:03:14 +01:00
|
|
|
if (s->in_flight == MAX_IN_FLIGHT || s->buf_free_count == 0 ||
|
|
|
|
(cnt == 0 && s->in_flight > 0)) {
|
|
|
|
trace_mirror_yield(s, s->in_flight, s->buf_free_count, cnt);
|
2016-02-05 03:00:30 +01:00
|
|
|
mirror_wait_for_io(s);
|
2013-01-22 09:03:12 +01:00
|
|
|
continue;
|
|
|
|
} else if (cnt != 0) {
|
2014-03-21 13:55:18 +01:00
|
|
|
delay_ns = mirror_iteration(s);
|
2012-10-18 16:49:23 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
should_complete = false;
|
2013-01-22 09:03:12 +01:00
|
|
|
if (s->in_flight == 0 && cnt == 0) {
|
2012-10-18 16:49:23 +02:00
|
|
|
trace_mirror_before_flush(s);
|
2016-04-12 16:17:41 +02:00
|
|
|
ret = blk_flush(s->target);
|
2012-10-18 16:49:23 +02:00
|
|
|
if (ret < 0) {
|
2014-06-18 08:43:30 +02:00
|
|
|
if (mirror_error_action(s, false, -ret) ==
|
|
|
|
BLOCK_ERROR_ACTION_REPORT) {
|
2012-10-18 16:49:28 +02:00
|
|
|
goto immediate_exit;
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
/* We're out of the streaming phase. From now on, if the job
|
|
|
|
* is cancelled we will actually complete all pending I/O and
|
|
|
|
* report completion. This way, block-job-cancel will leave
|
|
|
|
* the target in a consistent state.
|
|
|
|
*/
|
|
|
|
if (!s->synced) {
|
2014-06-18 08:43:47 +02:00
|
|
|
block_job_event_ready(&s->common);
|
2012-10-18 16:49:28 +02:00
|
|
|
s->synced = true;
|
|
|
|
}
|
|
|
|
|
|
|
|
should_complete = s->should_complete ||
|
|
|
|
block_job_is_cancelled(&s->common);
|
2015-04-18 01:50:02 +02:00
|
|
|
cnt = bdrv_get_dirty_count(s->dirty_bitmap);
|
2012-10-18 16:49:25 +02:00
|
|
|
}
|
2012-10-18 16:49:23 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
if (cnt == 0 && should_complete) {
|
|
|
|
/* The dirty bitmap is not updated while operations are pending.
|
|
|
|
* If we're about to exit, wait for pending operations before
|
|
|
|
* calling bdrv_get_dirty_count(bs), or we may exit while the
|
|
|
|
* source has dirty data to copy!
|
|
|
|
*
|
|
|
|
* Note that I/O can be submitted by the guest while
|
|
|
|
* mirror_populate runs.
|
|
|
|
*/
|
|
|
|
trace_mirror_before_drain(s, cnt);
|
2016-04-05 13:20:53 +02:00
|
|
|
bdrv_co_drain(bs);
|
2015-04-18 01:50:02 +02:00
|
|
|
cnt = bdrv_get_dirty_count(s->dirty_bitmap);
|
2012-10-18 16:49:23 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
ret = 0;
|
2014-03-21 13:55:18 +01:00
|
|
|
trace_mirror_before_sleep(s, cnt, s->synced, delay_ns);
|
2012-10-18 16:49:25 +02:00
|
|
|
if (!s->synced) {
|
2013-08-21 17:03:05 +02:00
|
|
|
block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns);
|
2012-10-18 16:49:23 +02:00
|
|
|
if (block_job_is_cancelled(&s->common)) {
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
} else if (!should_complete) {
|
2013-01-22 09:03:12 +01:00
|
|
|
delay_ns = (s->in_flight == 0 && cnt == 0 ? SLICE_TIME : 0);
|
2013-08-21 17:03:05 +02:00
|
|
|
block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns);
|
2012-10-18 16:49:23 +02:00
|
|
|
} else if (cnt == 0) {
|
|
|
|
/* The two disks are in sync. Exit and report successful
|
|
|
|
* completion.
|
|
|
|
*/
|
|
|
|
assert(QLIST_EMPTY(&bs->tracked_requests));
|
|
|
|
s->common.cancelled = false;
|
|
|
|
break;
|
|
|
|
}
|
2013-08-21 17:03:08 +02:00
|
|
|
last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
|
2012-10-18 16:49:23 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
immediate_exit:
|
2013-01-22 09:03:12 +01:00
|
|
|
if (s->in_flight > 0) {
|
|
|
|
/* We get here only if something went wrong. Either the job failed,
|
|
|
|
* or it was cancelled prematurely so that we do not guarantee that
|
|
|
|
* the target is a copy of the source.
|
|
|
|
*/
|
|
|
|
assert(ret < 0 || (!s->synced && block_job_is_cancelled(&s->common)));
|
|
|
|
mirror_drain(s);
|
|
|
|
}
|
|
|
|
|
|
|
|
assert(s->in_flight == 0);
|
2013-01-15 15:29:10 +01:00
|
|
|
qemu_vfree(s->buf);
|
2013-01-21 17:09:43 +01:00
|
|
|
g_free(s->cow_bitmap);
|
2013-01-22 09:03:14 +01:00
|
|
|
g_free(s->in_flight_bitmap);
|
2013-11-13 11:29:43 +01:00
|
|
|
bdrv_release_dirty_bitmap(bs, s->dirty_bitmap);
|
2014-10-21 13:03:58 +02:00
|
|
|
|
|
|
|
data = g_malloc(sizeof(*data));
|
|
|
|
data->ret = ret;
|
2015-11-23 03:28:04 +01:00
|
|
|
/* Before we switch to target in mirror_exit, make sure data doesn't
|
|
|
|
* change. */
|
2016-04-12 16:17:41 +02:00
|
|
|
bdrv_drained_begin(bs);
|
mirror: Workaround for unexpected iohandler events during completion
Commit 5a7e7a0ba moved mirror_exit to a BH handler but didn't add any
protection against new requests that could sneak in just before the
BH is dispatched. For example (assuming a code base at that commit):
main_loop_wait # 1
os_host_main_loop_wait
g_main_context_dispatch
aio_ctx_dispatch
aio_dispatch
...
mirror_run
bdrv_drain
(a) block_job_defer_to_main_loop
qemu_iohandler_poll
virtio_queue_host_notifier_read
...
virtio_submit_multiwrite
(b) blk_aio_multiwrite
main_loop_wait # 2
<snip>
aio_dispatch
aio_bh_poll
(c) mirror_exit
At (a) we know the BDS has no pending request. However, the same
main_loop_wait call is going to dispatch iohandlers (EventNotifier
events), which may lead to a new I/O from guest. So the invariant is
already broken at (c). Data loss.
Commit f3926945c8 made iohandler to use aio API. The order of
virtio_queue_host_notifier_read and block_job_defer_to_main_loop within
a main_loop_wait becomes unpredictable, and even worse, if the host
notifier event arrives at the next main_loop_wait call, the
unpredictable order between mirror_exit and
virtio_queue_host_notifier_read is also a trouble. As shown below, this
commit made the bug easier to trigger:
- Bug case 1:
main_loop_wait # 1
os_host_main_loop_wait
g_main_context_dispatch
aio_ctx_dispatch (qemu_aio_context)
...
mirror_run
bdrv_drain
(a) block_job_defer_to_main_loop
aio_ctx_dispatch (iohandler_ctx)
virtio_queue_host_notifier_read
...
virtio_submit_multiwrite
(b) blk_aio_multiwrite
main_loop_wait # 2
...
aio_dispatch
aio_bh_poll
(c) mirror_exit
- Bug case 2:
main_loop_wait # 1
os_host_main_loop_wait
g_main_context_dispatch
aio_ctx_dispatch (qemu_aio_context)
...
mirror_run
bdrv_drain
(a) block_job_defer_to_main_loop
main_loop_wait # 2
...
aio_ctx_dispatch (iohandler_ctx)
virtio_queue_host_notifier_read
...
virtio_submit_multiwrite
(b) blk_aio_multiwrite
aio_dispatch
aio_bh_poll
(c) mirror_exit
In both cases, (b) breaks the invariant wanted by (a) and (c).
Until then, the request loss has been silent. Later, 3f09bfbc7be added
asserts at (c) to check the invariant (in
bdrv_replace_in_backing_chain), and Max reported an assertion failure
first visible there, by doing active committing while the guest is
running bonnie++.
2.5 added bdrv_drained_begin at (a) to protect the dataplane case from
similar problems, but we never realize the main loop bug until now.
As a bandage, this patch disables iohandler's external events
temporarily together with bs->ctx.
Launchpad Bug: 1570134
Cc: qemu-stable@nongnu.org
Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Jeff Cody <jcody@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2016-04-22 15:53:56 +02:00
|
|
|
if (qemu_get_aio_context() == bdrv_get_aio_context(bs)) {
|
|
|
|
/* FIXME: virtio host notifiers run on iohandler_ctx, therefore the
|
|
|
|
* above bdrv_drained_end isn't enough to quiesce it. This is ugly, we
|
|
|
|
* need a block layer API change to achieve this. */
|
|
|
|
aio_disable_external(iohandler_get_aio_context());
|
|
|
|
}
|
2014-10-21 13:03:58 +02:00
|
|
|
block_job_defer_to_main_loop(&s->common, mirror_exit, data);
|
2012-10-18 16:49:23 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
static void mirror_set_speed(BlockJob *job, int64_t speed, Error **errp)
|
|
|
|
{
|
|
|
|
MirrorBlockJob *s = container_of(job, MirrorBlockJob, common);
|
|
|
|
|
|
|
|
if (speed < 0) {
|
2015-03-17 11:54:50 +01:00
|
|
|
error_setg(errp, QERR_INVALID_PARAMETER, "speed");
|
2012-10-18 16:49:23 +02:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
ratelimit_set_speed(&s->limit, speed / BDRV_SECTOR_SIZE, SLICE_TIME);
|
|
|
|
}
|
|
|
|
|
2012-10-18 16:49:25 +02:00
|
|
|
static void mirror_complete(BlockJob *job, Error **errp)
|
|
|
|
{
|
|
|
|
MirrorBlockJob *s = container_of(job, MirrorBlockJob, common);
|
2013-09-05 14:45:29 +02:00
|
|
|
Error *local_err = NULL;
|
2012-10-18 16:49:25 +02:00
|
|
|
int ret;
|
|
|
|
|
2016-04-12 16:17:41 +02:00
|
|
|
ret = bdrv_open_backing_file(blk_bs(s->target), NULL, "backing",
|
|
|
|
&local_err);
|
2012-10-18 16:49:25 +02:00
|
|
|
if (ret < 0) {
|
2013-09-05 14:45:29 +02:00
|
|
|
error_propagate(errp, local_err);
|
2012-10-18 16:49:25 +02:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
if (!s->synced) {
|
2015-09-16 13:34:54 +02:00
|
|
|
error_setg(errp, QERR_BLOCK_JOB_NOT_READY, job->id);
|
2012-10-18 16:49:25 +02:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2014-06-27 18:25:25 +02:00
|
|
|
/* check the target bs is not blocked and block all operations on it */
|
|
|
|
if (s->replaces) {
|
2014-10-21 13:03:58 +02:00
|
|
|
AioContext *replace_aio_context;
|
|
|
|
|
2015-07-17 04:12:22 +02:00
|
|
|
s->to_replace = bdrv_find_node(s->replaces);
|
2014-06-27 18:25:25 +02:00
|
|
|
if (!s->to_replace) {
|
2015-07-17 04:12:22 +02:00
|
|
|
error_setg(errp, "Node name '%s' not found", s->replaces);
|
2014-06-27 18:25:25 +02:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2014-10-21 13:03:58 +02:00
|
|
|
replace_aio_context = bdrv_get_aio_context(s->to_replace);
|
|
|
|
aio_context_acquire(replace_aio_context);
|
|
|
|
|
2014-06-27 18:25:25 +02:00
|
|
|
error_setg(&s->replace_blocker,
|
|
|
|
"block device is in use by block-job-complete");
|
|
|
|
bdrv_op_block_all(s->to_replace, s->replace_blocker);
|
|
|
|
bdrv_ref(s->to_replace);
|
2014-10-21 13:03:58 +02:00
|
|
|
|
|
|
|
aio_context_release(replace_aio_context);
|
2014-06-27 18:25:25 +02:00
|
|
|
}
|
|
|
|
|
2012-10-18 16:49:25 +02:00
|
|
|
s->should_complete = true;
|
2015-04-03 16:05:18 +02:00
|
|
|
block_job_enter(&s->common);
|
2012-10-18 16:49:25 +02:00
|
|
|
}
|
|
|
|
|
2013-10-08 11:29:38 +02:00
|
|
|
static const BlockJobDriver mirror_job_driver = {
|
2012-10-18 16:49:23 +02:00
|
|
|
.instance_size = sizeof(MirrorBlockJob),
|
2013-10-08 11:29:40 +02:00
|
|
|
.job_type = BLOCK_JOB_TYPE_MIRROR,
|
2012-10-18 16:49:23 +02:00
|
|
|
.set_speed = mirror_set_speed,
|
2012-10-18 16:49:25 +02:00
|
|
|
.complete = mirror_complete,
|
2012-10-18 16:49:23 +02:00
|
|
|
};
|
|
|
|
|
2013-12-16 07:45:30 +01:00
|
|
|
static const BlockJobDriver commit_active_job_driver = {
|
|
|
|
.instance_size = sizeof(MirrorBlockJob),
|
|
|
|
.job_type = BLOCK_JOB_TYPE_COMMIT,
|
|
|
|
.set_speed = mirror_set_speed,
|
|
|
|
.complete = mirror_complete,
|
|
|
|
};
|
|
|
|
|
|
|
|
static void mirror_start_job(BlockDriverState *bs, BlockDriverState *target,
|
2014-06-27 18:25:25 +02:00
|
|
|
const char *replaces,
|
2015-04-18 01:49:51 +02:00
|
|
|
int64_t speed, uint32_t granularity,
|
2014-06-27 18:25:25 +02:00
|
|
|
int64_t buf_size,
|
|
|
|
BlockdevOnError on_source_error,
|
|
|
|
BlockdevOnError on_target_error,
|
2015-06-08 07:56:08 +02:00
|
|
|
bool unmap,
|
2014-10-07 13:59:15 +02:00
|
|
|
BlockCompletionFunc *cb,
|
2014-06-27 18:25:25 +02:00
|
|
|
void *opaque, Error **errp,
|
|
|
|
const BlockJobDriver *driver,
|
|
|
|
bool is_none_mode, BlockDriverState *base)
|
2012-10-18 16:49:23 +02:00
|
|
|
{
|
|
|
|
MirrorBlockJob *s;
|
|
|
|
|
2013-01-21 17:09:46 +01:00
|
|
|
if (granularity == 0) {
|
2015-04-18 01:49:52 +02:00
|
|
|
granularity = bdrv_get_default_bitmap_granularity(target);
|
2013-01-21 17:09:46 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
assert ((granularity & (granularity - 1)) == 0);
|
|
|
|
|
2015-05-15 09:51:36 +02:00
|
|
|
if (buf_size < 0) {
|
|
|
|
error_setg(errp, "Invalid parameter 'buf-size'");
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (buf_size == 0) {
|
|
|
|
buf_size = DEFAULT_MIRROR_BUF_SIZE;
|
|
|
|
}
|
2013-12-16 07:45:29 +01:00
|
|
|
|
2013-12-16 07:45:30 +01:00
|
|
|
s = block_job_create(driver, bs, speed, cb, opaque, errp);
|
2012-10-18 16:49:23 +02:00
|
|
|
if (!s) {
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2016-04-12 16:17:41 +02:00
|
|
|
s->target = blk_new();
|
|
|
|
blk_insert_bs(s->target, target);
|
|
|
|
|
2014-06-27 18:25:25 +02:00
|
|
|
s->replaces = g_strdup(replaces);
|
2012-10-18 16:49:28 +02:00
|
|
|
s->on_source_error = on_source_error;
|
|
|
|
s->on_target_error = on_target_error;
|
2013-12-16 07:45:30 +01:00
|
|
|
s->is_none_mode = is_none_mode;
|
2013-12-16 07:45:29 +01:00
|
|
|
s->base = base;
|
2013-01-21 17:09:46 +01:00
|
|
|
s->granularity = granularity;
|
2015-05-15 09:51:36 +02:00
|
|
|
s->buf_size = ROUND_UP(buf_size, granularity);
|
2015-06-08 07:56:08 +02:00
|
|
|
s->unmap = unmap;
|
2013-01-21 17:09:43 +01:00
|
|
|
|
2015-04-18 01:49:50 +02:00
|
|
|
s->dirty_bitmap = bdrv_create_dirty_bitmap(bs, granularity, NULL, errp);
|
2014-04-16 03:34:30 +02:00
|
|
|
if (!s->dirty_bitmap) {
|
2015-06-26 11:37:35 +02:00
|
|
|
g_free(s->replaces);
|
2016-04-12 16:17:41 +02:00
|
|
|
blk_unref(s->target);
|
2015-11-06 00:13:11 +01:00
|
|
|
block_job_unref(&s->common);
|
2014-04-16 03:34:30 +02:00
|
|
|
return;
|
|
|
|
}
|
2015-11-02 15:51:53 +01:00
|
|
|
|
2016-04-12 16:17:41 +02:00
|
|
|
bdrv_op_block_all(target, s->common.blocker);
|
2015-11-02 15:51:53 +01:00
|
|
|
|
2012-10-18 16:49:23 +02:00
|
|
|
s->common.co = qemu_coroutine_create(mirror_run);
|
|
|
|
trace_mirror_start(bs, s, s->common.co, opaque);
|
|
|
|
qemu_coroutine_enter(s->common.co, s);
|
|
|
|
}
|
2013-12-16 07:45:30 +01:00
|
|
|
|
|
|
|
void mirror_start(BlockDriverState *bs, BlockDriverState *target,
|
2014-06-27 18:25:25 +02:00
|
|
|
const char *replaces,
|
2015-04-18 01:49:51 +02:00
|
|
|
int64_t speed, uint32_t granularity, int64_t buf_size,
|
2013-12-16 07:45:30 +01:00
|
|
|
MirrorSyncMode mode, BlockdevOnError on_source_error,
|
|
|
|
BlockdevOnError on_target_error,
|
2015-06-08 07:56:08 +02:00
|
|
|
bool unmap,
|
2014-10-07 13:59:15 +02:00
|
|
|
BlockCompletionFunc *cb,
|
2013-12-16 07:45:30 +01:00
|
|
|
void *opaque, Error **errp)
|
|
|
|
{
|
|
|
|
bool is_none_mode;
|
|
|
|
BlockDriverState *base;
|
|
|
|
|
2015-06-05 02:20:34 +02:00
|
|
|
if (mode == MIRROR_SYNC_MODE_INCREMENTAL) {
|
|
|
|
error_setg(errp, "Sync mode 'incremental' not supported");
|
2015-04-18 01:49:58 +02:00
|
|
|
return;
|
|
|
|
}
|
2013-12-16 07:45:30 +01:00
|
|
|
is_none_mode = mode == MIRROR_SYNC_MODE_NONE;
|
2015-06-17 14:55:21 +02:00
|
|
|
base = mode == MIRROR_SYNC_MODE_TOP ? backing_bs(bs) : NULL;
|
2014-06-27 18:25:25 +02:00
|
|
|
mirror_start_job(bs, target, replaces,
|
|
|
|
speed, granularity, buf_size,
|
2015-06-08 07:56:08 +02:00
|
|
|
on_source_error, on_target_error, unmap, cb, opaque, errp,
|
2013-12-16 07:45:30 +01:00
|
|
|
&mirror_job_driver, is_none_mode, base);
|
|
|
|
}
|
|
|
|
|
|
|
|
void commit_active_start(BlockDriverState *bs, BlockDriverState *base,
|
|
|
|
int64_t speed,
|
|
|
|
BlockdevOnError on_error,
|
2014-10-07 13:59:15 +02:00
|
|
|
BlockCompletionFunc *cb,
|
2013-12-16 07:45:30 +01:00
|
|
|
void *opaque, Error **errp)
|
|
|
|
{
|
2014-01-24 15:02:36 +01:00
|
|
|
int64_t length, base_length;
|
|
|
|
int orig_base_flags;
|
2014-02-12 20:46:24 +01:00
|
|
|
int ret;
|
2014-02-13 15:23:38 +01:00
|
|
|
Error *local_err = NULL;
|
2014-01-24 15:02:36 +01:00
|
|
|
|
|
|
|
orig_base_flags = bdrv_get_flags(base);
|
|
|
|
|
2013-12-16 07:45:31 +01:00
|
|
|
if (bdrv_reopen(base, bs->open_flags, errp)) {
|
|
|
|
return;
|
|
|
|
}
|
2014-01-24 15:02:36 +01:00
|
|
|
|
|
|
|
length = bdrv_getlength(bs);
|
|
|
|
if (length < 0) {
|
2014-02-12 20:46:24 +01:00
|
|
|
error_setg_errno(errp, -length,
|
|
|
|
"Unable to determine length of %s", bs->filename);
|
2014-01-24 15:02:36 +01:00
|
|
|
goto error_restore_flags;
|
|
|
|
}
|
|
|
|
|
|
|
|
base_length = bdrv_getlength(base);
|
|
|
|
if (base_length < 0) {
|
2014-02-12 20:46:24 +01:00
|
|
|
error_setg_errno(errp, -base_length,
|
|
|
|
"Unable to determine length of %s", base->filename);
|
2014-01-24 15:02:36 +01:00
|
|
|
goto error_restore_flags;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (length > base_length) {
|
2014-02-12 20:46:24 +01:00
|
|
|
ret = bdrv_truncate(base, length);
|
|
|
|
if (ret < 0) {
|
|
|
|
error_setg_errno(errp, -ret,
|
|
|
|
"Top image %s is larger than base image %s, and "
|
2014-01-24 15:02:36 +01:00
|
|
|
"resize of base image failed",
|
|
|
|
bs->filename, base->filename);
|
|
|
|
goto error_restore_flags;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2014-06-27 18:25:25 +02:00
|
|
|
mirror_start_job(bs, base, NULL, speed, 0, 0,
|
2015-06-08 07:56:08 +02:00
|
|
|
on_error, on_error, false, cb, opaque, &local_err,
|
2013-12-16 07:45:30 +01:00
|
|
|
&commit_active_job_driver, false, base);
|
2014-04-25 16:50:31 +02:00
|
|
|
if (local_err) {
|
2014-02-13 15:23:38 +01:00
|
|
|
error_propagate(errp, local_err);
|
2014-01-24 15:02:36 +01:00
|
|
|
goto error_restore_flags;
|
|
|
|
}
|
|
|
|
|
|
|
|
return;
|
|
|
|
|
|
|
|
error_restore_flags:
|
|
|
|
/* ignore error and errp for bdrv_reopen, because we want to propagate
|
|
|
|
* the original error */
|
|
|
|
bdrv_reopen(base, orig_base_flags, NULL);
|
|
|
|
return;
|
2013-12-16 07:45:30 +01:00
|
|
|
}
|