qcow2: async handling of fragmented io
v2 cover qcow2: async handling of fragmented io
Hi all!
Here is an asynchronous scheme for handling fragmented qcow2 reads and writes. Both qcow2 read and write functions loops through sequential portions of data. The series aim it to parallelize these loops iterations.
It improves performance for fragmented qcow2 images, I've tested it as described below.
v2: changed a lot, as
1. a lot of preparations around locks, hd_qiovs, threads for encryption are done
2. I decided to create separate file with async request handling API, to reuse it for backup, stream and copy-on-read to improve their performance too. Mirror and qemu-img convert has their own async request handling, may be we'll be able finally merge all these similar code into one feature.
Note that not all API calls used in qcow2, some will be needed on following steps for parallelizing other io loops.
Based-on: https://github.com/stefanha/qemu/commits/block
About testing:
I have four 4G qcow2 images (with default 64k block size) on my ssd disk:
t-seq.qcow2 - sequentially written qcow2 image
t-reverse.qcow2 - filled by writing 64k portions from end to the start t-rand.qcow2 - filled by writing 64k portions (aligned) in random order t-part-rand.qcow2 - filled by shuffling order of 64k writes in 1m clusters (see source code of image generation in the end for details)
and I've done several runs like the following (sequential io by 1mb chunks):
out=/tmp/block; echo > $out; cat /tmp/files | while read file; do for wr in {"","-w"}; do echo "$file" $wr; ./qemu-img bench -c 4096 -d 1 -f qcow2 -n -s 1m -t none $wr "$file" | grep 'Run completed in' | awk '{print $4}' >> $out; done; done
short info about parameters:
-w - do writes (otherwise do reads)
-c - count of blocks
-s - block size
-t none - disable cache
-n - native aio
-d 1 - don't use parallel requests provided by qemu-img bench itself
results:
+---------------------------+---------+---------+
| file | master | async |
+---------------------------+---------+---------+
| /ssd/t-part-rand.qcow2 | 14.671 | 9.193 |
+---------------------------+---------+---------+
| /ssd/t-part-rand.qcow2 -w | 11.434 | 8.621 |
+---------------------------+---------+---------+
| /ssd/t-rand.qcow2 | 20.421 | 10.05 |
+---------------------------+---------+---------+
| /ssd/t-rand.qcow2 -w | 11.097 | 8.915 |
+---------------------------+---------+---------+
| /ssd/t-reverse.qcow2 | 17.515 | 9.407 |
+---------------------------+---------+---------+
| /ssd/t-reverse.qcow2 -w | 11.255 | 8.649 |
+---------------------------+---------+---------+
| /ssd/t-seq.qcow2 | 9.081 | 9.072 |
+---------------------------+---------+---------+
| /ssd/t-seq.qcow2 -w | 8.761 | 8.747 |
+---------------------------+---------+---------+
| /tmp/t-part-rand.qcow2 | 41.179 | 41.37 |
+---------------------------+---------+---------+
| /tmp/t-part-rand.qcow2 -w | 54.097 | 55.323 |
+---------------------------+---------+---------+
| /tmp/t-rand.qcow2 | 711.899 | 514.339 |
+---------------------------+---------+---------+
| /tmp/t-rand.qcow2 -w | 546.259 | 642.114 |
+---------------------------+---------+---------+
| /tmp/t-reverse.qcow2 | 86.065 | 96.522 |
+---------------------------+---------+---------+
| /tmp/t-reverse.qcow2 -w | 46.557 | 48.499 |
+---------------------------+---------+---------+
| /tmp/t-seq.qcow2 | 33.804 | 33.862 |
+---------------------------+---------+---------+
| /tmp/t-seq.qcow2 -w | 34.299 | 34.233 |
+---------------------------+---------+---------+Performance gain is obvious, especially for read and especially for ssd. For hdd there is a degradation for reverse case, but this is the most impossible case and seems not critical.
How images are generated:
==== gen-writes ======
#!/usr/bin/env python
import random
import sys
size = 4 * 1024 * 1024 * 1024
block = 64 * 1024
block2 = 1024 * 1024
arg = sys.argv[1]
if arg in ('rand', 'reverse', 'seq'):
writes = list(range(0, size, block))
if arg == 'rand':
random.shuffle(writes)
elif arg == 'reverse':
writes.reverse()
elif arg == 'part-rand':
writes = []
for off in range(0, size, block2):
wr = list(range(off, off + block2, block))
random.shuffle(wr)
writes.extend(wr)
elif arg != 'seq':
sys.exit(1)
for w in writes:
print 'write -P 0xff {} {}'.format(w, block)
print 'q'
==========================
===== gen-test-images.sh =====
#!/bin/bash
IMG_PATH=/ssd
for name in seq reverse rand part-rand; do
IMG=$IMG_PATH/t-$name.qcow2
echo createing $IMG ...
rm -f $IMG
qemu-img create -f qcow2 $IMG 4G
gen-writes $name | qemu-io $IMG
done
==============================Vladimir Sementsov-Ogievskiy (4):
block: introduce aio task pool
block/qcow2: refactor qcow2_co_preadv_part
block/qcow2: refactor qcow2_co_pwritev_part
block/qcow2: introduce parallel subrequest handling in read and write
qapi/block-core.json | 2 +- block/aio_task.h | 52 +++++ block/aio_task.c | 119 +++++++++++ block/qcow2.c | 459 ++++++++++++++++++++++++++++--------------- block/Makefile.objs | 2 + block/trace-events | 1 + 6 files changed, 477 insertions(+), 158 deletions(-) create mode 100644 block/aio_task.h create mode 100644 block/aio_task.c
Signature · 1 line
2.18.0
PATCH v2 1/4 block: introduce aio task pool
Common interface for aio task loops. To be used for improving performance of synchronous io loops in qcow2, block-stream, copy-on-read, and may be other places.
--- block/aio_task.h | 52 +++++++++++++++++++ block/aio_task.c | 119 ++++++++++++++++++++++++++++++++++++++++++++ block/Makefile.objs | 2 + 3 files changed, 173 insertions(+) create mode 100644 block/aio_task.h create mode 100644 block/aio_task.c
block/aio_task.h new file +52 -0
@@ -0,0 +1,52 @@ +/* +* Aio tasks loops +* +* Copyright (c) 2019 Virtuozzo International GmbH. +* +* Permission is hereby granted, free of charge, to any person obtaining a copy +* of this software and associated documentation files (the "Software"), to deal +* in the Software without restriction, including without limitation the rights +* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +* copies of the Software, and to permit persons to whom the Software is +* furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice shall be included in +* all copies or substantial portions of the Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +* THE SOFTWARE. +*/ + +#ifndef BLOCK_AIO_TASK_H +#define BLOCK_AIO_TASK_H + +#include "qemu/coroutine.h" + +typedef struct AioTaskPool AioTaskPool ; +typedef struct AioTask AioTask ; +typedef int (*AioTaskFunc )(AioTask *task); +struct AioTask { +AioTaskPool *pool; +AioTaskFunc func; +int ret; +}; + +/* +* aio_task_pool_new +* +* The caller is responsible to g_free AioTaskPool pointer after use. +*/ +AioTaskPool *aio_task_pool_new (int max_busy_tasks); +int aio_task_pool_status (AioTaskPool *pool); +bool aio_task_pool_empty (AioTaskPool *pool); +void aio_task_pool_start_task (AioTaskPool *pool,AioTask *task); +void aio_task_pool_wait_slot (AioTaskPool *pool); +void aio_task_pool_wait_one (AioTaskPool *pool); +void aio_task_pool_wait_all (AioTaskPool *pool); + +#endif /* BLOCK_AIO_TASK_H */
block/aio_task.c new file +119 -0
119 lines not on this page · raw message
block/Makefile.objs +2 -0
@@ -40,6 +40,8 @@ block-obj-y += throttle.o copy-on-read.o block-obj-y += crypto.o +block-obj-y += aio_task.o + common-obj-y += stream.o nfs.o-libs := $(LIBNFS_LIBS)
Signature · 1 line
2.18.0
Re: [Qemu-devel] [PATCH v2 1/4] block: introduce aio task pool
Quoted 6 lines · Vladimir Sementsov-Ogievskiy
Common interface for aio task loops. To be used for improving performance of synchronous io loops in qcow2, block-stream, copy-on-read, and may be other places.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>---
Looks good to me overall.
Quoted 1 line
block/aio_task.h | 52 +++++++++++++++++++
I’ve move this to include/block/.
Quoted 11 lines
block/aio_task.c | 119 ++++++++++++++++++++++++++++++++++++++++++++ block/Makefile.objs | 2 + 3 files changed, 173 insertions(+) create mode 100644 block/aio_task.h create mode 100644 block/aio_task.cblock/aio_task.h new file +0
-0
[...]
block/aio_task.h line 32
Quoted 3 lines
block/aio_task.h new file +3
-0+typedef struct AioTaskPool AioTaskPool ; +typedef struct AioTask AioTask ; +typedef int (*AioTaskFunc )(AioTask *task);
+coroutine_fn
block/aio_task.h line 42
Quoted 10 lines
block/aio_task.h new file +10
-0+struct AioTask { +AioTaskPool *pool; +AioTaskFunc func; +int ret; +}; + +/* +* aio_task_pool_new +* +* The caller is responsible to g_free AioTaskPool pointer after use.
s/to g_free/for g_freeing/ or something similar.
Or you’d just add aio_task_pool_free().
block/aio_task.h line 45
Quoted 3 lines
+ */
block/aio_task.h new file +2
-0+AioTaskPool *aio_task_pool_new (int max_busy_tasks); +int aio_task_pool_status (AioTaskPool *pool);
A comment wouldn’t hurt. It wasn’t immediately clear to me that status refers to the error code of a failing task (or 0), although it wasn’t too much of a surprise either.
block/aio_task.h line 47
Quoted 2 lines
block/aio_task.h new file +2
-0+bool aio_task_pool_empty (AioTaskPool *pool); +void aio_task_pool_start_task (AioTaskPool *pool,AioTask *task);
Maybe make a note that task->pool will be set automatically?
block/aio_task.h line 50
Quoted 3 lines
block/aio_task.h new file +3
-0+void aio_task_pool_wait_slot (AioTaskPool *pool); +void aio_task_pool_wait_one (AioTaskPool *pool); +void aio_task_pool_wait_all (AioTaskPool *pool);
Shouldn’t all of these but aio_task_pool_empty() and aio_task_pool_status() be coroutine_fns?
Quoted 6 lines
block/aio_task.h new file +1
-0+#endif /* BLOCK_AIO_TASK_H */ block/aio_task.c new file +0
-0
[...]
Quoted 1 line
+static void
aio_task_co(void *opaque)
+coroutine_fn
[...]
block/aio_task.c line 66
Quoted 6 lines
block/aio_task.c new file +6
-0+void aio_task_pool_wait_one (AioTaskPool *pool) +{ +assert (pool->busy_tasks >0 ); +assert (qemu_coroutine_self () == pool->main_co); + + pool->wait_done = true;
Hmmm, but the wait actually isn’t done yet. :-)
Maybe s/wait_done/waiting/?
Max
Quoted 5 lines
block/aio_task.c new file +5
-0+qemu_coroutine_yield (); + +assert (!pool->wait_done); +assert (pool->busy_tasks < pool->max_busy_tasks); +}
Re: [Qemu-devel] [PATCH v2 1/4] block: introduce aio task pool
Quoted 93 lines · Max Reitz
Quoted 6 lines · Vladimir Sementsov-Ogievskiy
Common interface for aio task loops. To be used for improving performance of synchronous io loops in qcow2, block-stream, copy-on-read, and may be other places.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>---
Looks good to me overall.
Quoted 1 line
block/aio_task.h | 52 +++++++++++++++++++I’ve move this to
include/block/.Quoted 11 lines
block/aio_task.c | 119 ++++++++++++++++++++++++++++++++++++++++++++ block/Makefile.objs | 2 + 3 files changed, 173 insertions(+)create mode 100644
block/aio_task.h
create mode 100644block/aio_task.cblock/aio_task.h new file +0
-0[...]
block/aio_task.h line 32
Quoted 3 lines
block/aio_task.h new file +3
-0+typedef struct AioTaskPool AioTaskPool ; +typedef struct AioTask AioTask ; +typedef int (*AioTaskFunc )(AioTask *task);+coroutine_fn
block/aio_task.h line 42
Quoted 10 lines
block/aio_task.h new file +10
-0+struct AioTask { +AioTaskPool *pool; +AioTaskFunc func; +int ret; +}; + +/* +* aio_task_pool_new +* +* The caller is responsible to g_free AioTaskPool pointer after use. s/to g_free/for g_freeing/ or something similar.
Or you’d just add
aio_task_pool_free().block/aio_task.h line 45
Quoted 3 lines
+ */
block/aio_task.h new file +2
-0+AioTaskPool *aio_task_pool_new (int max_busy_tasks); +int aio_task_pool_status (AioTaskPool *pool);A comment wouldn’t hurt. It wasn’t immediately clear to me that status refers to the error code of a failing task (or 0), although it wasn’t too much of a surprise either.
block/aio_task.h line 47
Quoted 2 lines
block/aio_task.h new file +2
-0+bool aio_task_pool_empty (AioTaskPool *pool); +void aio_task_pool_start_task (AioTaskPool *pool,AioTask *task);Maybe make a note that
task->poolwill be set automatically?block/aio_task.h line 50
Quoted 3 lines
block/aio_task.h new file +3
-0+void aio_task_pool_wait_slot (AioTaskPool *pool); +void aio_task_pool_wait_one (AioTaskPool *pool); +void aio_task_pool_wait_all (AioTaskPool *pool);Shouldn’t all of these but
aio_task_pool_empty()andaio_task_pool_status()becoroutine_fns?Quoted 6 lines
block/aio_task.h new file +1
-0+#endif /* BLOCK_AIO_TASK_H */ block/aio_task.c new file +0
-0[...]
Quoted 1 line
+static void
aio_task_co(void *opaque)+coroutine_fn
[...]
block/aio_task.c line 66
Quoted 6 lines
block/aio_task.c new file +6
-0+void aio_task_pool_wait_one (AioTaskPool *pool) +{ +assert (pool->busy_tasks >0 ); +assert (qemu_coroutine_self () == pool->main_co); + + pool->wait_done = true;Hmmm, but the wait actually isn’t done yet. :-)
Quoted 2 lines
Maybe s/wait_done/waiting/?
Aha, really bad variable name. I meant "wait for one task done". Just "waiting" would be appropriate.
Thanks for reviewing!
Signature · 2 lines
Best regards, Vladimir
PATCH v2 2/4 block/qcow2: refactor qcow2_co_preadv_part
In reply to Vladimir Sementsov-OgievskiyFurther patch will run partial requests of iterations of qcow2_co_preadv in parallel for performance reasons. To prepare for this, separate part which may be parallelized into separate function (qcow2_co_preadv_task).
While being here, also separate encrypted clusters reading to own function, like it is done for compressed reading.
--- qapi/block-core.json | 2 +- block/qcow2.c | 206 +++++++++++++++++++++++-------------------- 2 files changed, 112 insertions(+), 96 deletions(-)
qapi/block-core.json +1 -1
@@ -3266,7 +3266,7 @@ 'pwritev_rmw_tail', 'pwritev_rmw_after_tail', 'pwritev', 'pwritev_zero', 'pwritev_done', 'empty_image_prepare', 'l1_shrink_write_table', 'l1_shrink_free_l2_clusters', - 'cor_write', 'cluster_alloc_space', 'none'] } + 'cor_write', 'cluster_alloc_space', 'none', 'read_encrypted'] } ## # @BlkdebugIOType:
block/qcow2.c +111 -95
239 lines not on this page · raw message
Signature · 1 line
2.18.0
Re: [Qemu-devel] [PATCH v2 2/4] block/qcow2: refactor qcow2_co_preadv_part
Quoted 5 lines · Vladimir Sementsov-Ogievskiy
Further patch will run partial requests of iterations of
qcow2_co_preadvin parallel for performance reasons. To prepare for this, separate part which may be parallelized into separate function (qcow2_co_preadv_task).
Quoted 8 lines
While being here, also separate encrypted clusters reading to own function, like it is done for compressed reading.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- qapi/block-core.json | 2 +- block/qcow2.c | 206 +++++++++++++++++++++++-------------------- 2 files changed, 112 insertions(+), 96 deletions(-)
Looks good to me overall, just wondering about some details, as always.
Quoted 17 lines
block/qcow2.c +9
-0@@ -1967,17 +1967,115 @@ out:return ret; } +static coroutine_fn int +qcow2_co_preadv_encrypted (BlockDriverState *bs, +uint64_t file_cluster_offset, +uint64_t offset, +uint64_t bytes, +QEMUIOVector *qiov, +uint64_t qiov_offset) +{ +int ret;
block/qcow2.c out: · line 1988
Quoted 10 lines
block/qcow2.c +10
-0+BDRVQcow2State *s = bs->opaque; +uint8_t *buf; + +assert (bs->encrypted && s->crypto); +assert (bytes <=QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); + +/* +* For encrypted images, read everything into a temporary +* contiguous buffer on which the AES functions can work. +* Note, that we can implement enctyption, working on qiov,
-, and s/enctyption/encryption/
block/qcow2.c out: · line 1990
Quoted 2 lines
block/qcow2.c +2
-0+* but we must not do decryption in guest buffers for security +* reasons.
"for security reasons" is a bit handwave-y, no?
[...]
Quoted 11 lines
block/qcow2.c +11
-0+static coroutine_fn int qcow2_co_preadv_task(BlockDriverState *bs, +QCow2ClusterType cluster_type, +uint64_t file_cluster_offset, +uint64_t offset,uint64_t bytes, +QEMUIOVector *qiov, +size_t qiov_offset) +{ +BDRVQcow2State *s = bs->opaque; +int offset_in_cluster =offset_into_cluster (s, offset); + +switch (cluster_type) {
[...]
block/qcow2.c out: · line 2061
Quoted 5 lines
block/qcow2.c +5
-0+default : +g_assert_not_reached (); +/* +* QCOW2_CLUSTER_ZERO_PLAIN and QCOW2_CLUSTER_ZERO_ALLOC handled +* in qcow2_co_preadv_part
Hmm, I’d still add them explicitly as cases and put their own g_assert_not_reach() there.
block/qcow2.c out: · line 2067
Quoted 6 lines
block/qcow2.c +6
-0+*/ + } + +g_assert_not_reached (); + +return -EIO ;
Maybe abort()ing instead of g_assert_not_reach() would save you from having to return here?
Max
Re: [Qemu-devel] [PATCH v2 2/4] block/qcow2: refactor qcow2_co_preadv_part
Quoted 47 lines · Max Reitz
Quoted 5 lines · Vladimir Sementsov-Ogievskiy
Further patch will run partial requests of iterations of
qcow2_co_preadvin parallel for performance reasons. To prepare for this, separate part which may be parallelized into separate function (qcow2_co_preadv_task).Quoted 8 lines
While being here, also separate encrypted clusters reading to own function, like it is done for compressed reading.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- qapi/block-core.json | 2 +- block/qcow2.c | 206 +++++++++++++++++++++++-------------------- 2 files changed, 112 insertions(+), 96 deletions(-)Looks good to me overall, just wondering about some details, as always.
Quoted 17 lines
block/qcow2.c +9
-0@@ -1967,17 +1967,115 @@ out:return ret; } +static coroutine_fn int +qcow2_co_preadv_encrypted (BlockDriverState *bs, +uint64_t file_cluster_offset, +uint64_t offset, +uint64_t bytes, +QEMUIOVector *qiov, +uint64_t qiov_offset) +{ +int ret;block/qcow2.c out: · line 1988
Quoted 10 lines
block/qcow2.c +10
-0+BDRVQcow2State *s = bs->opaque; +uint8_t *buf; + +assert (bs->encrypted && s->crypto); +assert (bytes <=QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); + +/* +* For encrypted images, read everything into a temporary +* contiguous buffer on which the AES functions can work. +* Note, that we can implement enctyption, working on qiov, -, and s/enctyption/encryption/
Quoted 4 lines
block/qcow2.c out: · line 1990
Quoted 2 lines
block/qcow2.c +2
-0+* but we must not do decryption in guest buffers for security +* reasons. "for security reasons" is a bit handwave-y, no?
Hmm, let's think of it a bit.
WRITE
1. We can't do any operations on write buffers, as guest may use them for something else and not prepared for their change. [thx to Den, pointed to this fact]
READ
Hmm, here otherwise, guest should not expect something meaningful in buffers until the end of read operation, so theoretically we may decrypt directly in guest buffer.. What is bad with it?
1. Making read-part different from write and implementing support of qiov for decryptin for little outcome (hmm, don't double allocation for reads, is it little or not? [*]).
2. Guest can read its buffers.
So, it may see encrypted data and guess something about it. Ideally guest should know nothing about encryption, but on the other hand, is there any real damage? I don't sure..
3. Guest can modify its buffers.
3.1 I think there is no guarantee that guest will not modify its data before we finished copying to separate buffer, so what guest finally reads is not predictable anyway. 3.2 But, modifying during decryption may possibly lead to guest visible error (which will never be if we operate on separated cluster)
So if we don't afraid of [2] and [3.2], and in a specific case [*] is significant, we may want implement decryption on guest buffers at least as an option..
But all it looks for me like we'll never do it.
===
So, I'd rewrite my "Note" like this:
Also, decryption in separate buffer is better as it hides from the guest information it doesn't own (about encrypted nature of virtual disk).
Quoted 26 lines
[...]
Quoted 11 lines
block/qcow2.c +11
-0+static coroutine_fn int qcow2_co_preadv_task(BlockDriverState *bs, +QCow2ClusterType cluster_type, +uint64_t file_cluster_offset, +uint64_t offset,uint64_t bytes, +QEMUIOVector *qiov, +size_t qiov_offset) +{ +BDRVQcow2State *s = bs->opaque; +int offset_in_cluster =offset_into_cluster (s, offset); + +switch (cluster_type) {[...]
block/qcow2.c out: · line 2061
Quoted 5 lines
block/qcow2.c +5
-0+default : +g_assert_not_reached (); +/* +* QCOW2_CLUSTER_ZERO_PLAIN and QCOW2_CLUSTER_ZERO_ALLOC handled +* in qcow2_co_preadv_part Hmm, I’d still add them explicitly as cases and put their own
g_assert_not_reach()there.
Quoted 10 lines
block/qcow2.c out: · line 2067
Quoted 6 lines
block/qcow2.c +6
-0+*/ + } + +g_assert_not_reached (); + +return -EIO ;Maybe abort()ing instead of
g_assert_not_reach()would save you from having to return here?
Hmm, will check. Any reason to use g_assert_not_reached() instead of abort() in "default"? I just kept it like it was. But it seems to be more often practice to use just abort() in Qemu code.
Signature · 2 lines
Best regards, Vladimir
Re: [Qemu-devel] [PATCH v2 2/4] block/qcow2: refactor qcow2_co_preadv_part
Quoted 51 lines · Vladimir Sementsov-Ogievskiy
Quoted 50 lines · Max Reitz
Quoted 5 lines · Vladimir Sementsov-Ogievskiy
Further patch will run partial requests of iterations of
qcow2_co_preadvin parallel for performance reasons. To prepare for this, separate part which may be parallelized into separate function (qcow2_co_preadv_task).Quoted 8 lines
While being here, also separate encrypted clusters reading to own function, like it is done for compressed reading.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- qapi/block-core.json | 2 +- block/qcow2.c | 206 +++++++++++++++++++++++-------------------- 2 files changed, 112 insertions(+), 96 deletions(-)Looks good to me overall, just wondering about some details, as always.
Quoted 17 lines
block/qcow2.c +9
-0@@ -1967,17 +1967,115 @@ out:return ret; } +static coroutine_fn int +qcow2_co_preadv_encrypted (BlockDriverState *bs, +uint64_t file_cluster_offset, +uint64_t offset, +uint64_t bytes, +QEMUIOVector *qiov, +uint64_t qiov_offset) +{ +int ret;block/qcow2.c out: · line 1988
Quoted 10 lines
block/qcow2.c +10
-0+BDRVQcow2State *s = bs->opaque; +uint8_t *buf; + +assert (bs->encrypted && s->crypto); +assert (bytes <=QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); + +/* +* For encrypted images, read everything into a temporary +* contiguous buffer on which the AES functions can work. +* Note, that we can implement enctyption, working on qiov, -, and s/enctyption/encryption/
Quoted 2 lines
block/qcow2.c +2
-0+* but we must not do decryption in guest buffers for security +* reasons.
Quoted 8 lines
Quoted 1 line
"for security reasons" is a bit handwave-y, no?
Hmm, let's think of it a bit.
WRITE
1. We can't do any operations on write buffers, as guest may use them for something else and not prepared for their change. [thx to Den, pointed to this fact]
Yep. So we actually cannot implement encryption in the guest buffer. :-)
Quoted 20 lines
READ
Hmm, here otherwise, guest should not expect something meaningful in buffers until the end of read operation, so theoretically we may decrypt directly in guest buffer.. What is bad with it?
1. Making read-part different from write and implementing support of qiov for decryptin for little outcome (hmm, don't double allocation for reads, is it little or not? [*]).
2. Guest can read its buffers.
So, it may see encrypted data and guess something about it. Ideally guest should know nothing about encryption, but on the other hand, is there any real damage? I don't sure..3. Guest can modify its buffers.
3.1 I think there is no guarantee that guest will not modify its data before we finished copying to separate buffer, so what guest finally reads is not predictable anyway. 3.2 But, modifying during decryption may possibly lead to guest visible error (which will never be if we operate on separated cluster)
Quoted 3 lines
So if we don't afraid of [2] and [3.2], and in a specific case [*] is significant, we may want implement decryption on guest buffers at least as an option..
But all it looks for me like we'll never do it.
Well, I do think it would be weird from a guest perspective to see data changing twice. I just couldn’t figure out what the security problem might be.
Quoted 6 lines
===
So, I'd rewrite my "Note" like this:
Also, decryption in separate buffer is better as it hides from the guest information it doesn't own (about encrypted nature of virtual disk).
Sounds good.
Quoted 36 lines
Quoted 25 lines
[...]
Quoted 11 lines
block/qcow2.c +11
-0+static coroutine_fn int qcow2_co_preadv_task(BlockDriverState *bs, +QCow2ClusterType cluster_type, +uint64_t file_cluster_offset, +uint64_t offset,uint64_t bytes, +QEMUIOVector *qiov, +size_t qiov_offset) +{ +BDRVQcow2State *s = bs->opaque; +int offset_in_cluster =offset_into_cluster (s, offset); + +switch (cluster_type) {[...]
block/qcow2.c out: · line 2061
Quoted 5 lines
block/qcow2.c +5
-0+default : +g_assert_not_reached (); +/* +* QCOW2_CLUSTER_ZERO_PLAIN and QCOW2_CLUSTER_ZERO_ALLOC handled +* in qcow2_co_preadv_part Hmm, I’d still add them explicitly as cases and put their own
g_assert_not_reach()there.Quoted 10 lines
block/qcow2.c out: · line 2067
Quoted 6 lines
block/qcow2.c +6
-0+*/ + } + +g_assert_not_reached (); + +return -EIO ;Maybe abort()ing instead of
g_assert_not_reach()would save you from having to return here?
Quoted 1 line
Hmm, will check. Any reason to use
g_assert_not_reached()instead ofabort()in "default"?
g_assert_not_reached() makes it more explicit what it’s about.
Quoted 2 lines
I just kept it like it was. But it seems to be more often practice to use just
abort()in Qemu code.
Yes, we often use abort() instead because it always has _Noreturn (and thus saves us from such useless return statements).
Max
Re: [Qemu-devel] [PATCH v2 2/4] block/qcow2: refactor qcow2_co_preadv_part
In reply to Vladimir Sementsov-OgievskiyQuoted 22 lines · Vladimir Sementsov-Ogievskiy
Quoted 9 lines · Max Reitz
Quoted 8 lines · Vladimir Sementsov-Ogievskiy
Further patch will run partial requests of iterations of
qcow2_co_preadvin parallel for performance reasons. To prepare for this, separate part which may be parallelized into separate function (qcow2_co_preadv_task).While being here, also separate encrypted clusters reading to own function, like it is done for compressed reading.
Quoted 7 lines
block/qcow2.c out: · line 1990
Quoted 5 lines
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>---
block/qcow2.c +2
-0+* but we must not do decryption in guest buffers for security +* reasons. "for security reasons" is a bit handwave-y, no?
Hmm, let's think of it a bit.
WRITE
Quoted 8 lines
1. We can't do any operations on write buffers, as guest may use them for something else and not prepared for their change. [thx to Den, pointed to this fact]
READ
Hmm, here otherwise, guest should not expect something meaningful in buffers until the end of read operation, so theoretically we may decrypt directly in guest buffer.. What is bad with it?
The badness is that the guest can theoretically reverse-engineer the encryption keys if they are savvy enough to grab the contents of the buffer before and after. The guest must NEVER be able to see the encrypted bits, which means decryption requires a bounce buffer.
Quoted 8 lines
1. Making read-part different from write and implementing support of qiov for decryptin for little outcome (hmm, don't double allocation for reads, is it little or not? [*]).
2. Guest can read its buffers.
So, it may see encrypted data and guess something about it. Ideally guest should know nothing about encryption, but on the other hand, is there any real damage? I don't sure..
Yes, this is the security risk.
Quoted 11 lines
3. Guest can modify its buffers.
3.1 I think there is no guarantee that guest will not modify its data before we finished copying to separate buffer, so what guest finally reads is not predictable anyway. 3.2 But, modifying during decryption may possibly lead to guest visible error (which will never be if we operate on separated cluster)So if we don't afraid of [2] and [3.2], and in a specific case [*] is significant, we may want implement decryption on guest buffers at least as an option..
But all it looks for me like we'll never do it.
Quoted 6 lines
===
So, I'd rewrite my "Note" like this:
Also, decryption in separate buffer is better as it hides from the guest information it doesn't own (about encrypted nature of virtual disk).
Possible wording tweak:
Also, decryption in a separate buffer is better as it prevents the guest from learning information about the encrypted nature of the virtual disk.
Quoted 10 lines
Quoted 9 lines
block/qcow2.c out: · line 2067
Quoted 5 lines
block/qcow2.c +5
-0+ } + +g_assert_not_reached (); + +return -EIO ;Maybe abort()ing instead of
g_assert_not_reach()would save you from having to return here?
Quoted 3 lines
Hmm, will check. Any reason to use
g_assert_not_reached()instead ofabort()in "default"? I just kept it like it was. But it seems to be more often practice to use justabort()in Qemu code.
Both are used. abort() is shorter to type, but g_assert_not_reach() is slightly friendlier to developers (which are the only people that would ever see the failure). As both are marked noreturn, the real fix is to drop the dead return -EIO line, the compiler is smart enough to not need a return statement after a noreturn function.
Signature · 3 lines
Eric Blake, Principal Software Engineer Red Hat, Inc. +1-919-301-3226 Virtualization: qemu.org | libvirt.org
Re: [Qemu-devel] [PATCH v2 2/4] block/qcow2: refactor qcow2_co_preadv_part
Quoted 28 lines · Eric Blake
Quoted 27 lines · Vladimir Sementsov-Ogievskiy
Quoted 9 lines · Max Reitz
Quoted 8 lines · Vladimir Sementsov-Ogievskiy
Further patch will run partial requests of iterations of
qcow2_co_preadvin parallel for performance reasons. To prepare for this, separate part which may be parallelized into separate function (qcow2_co_preadv_task).While being here, also separate encrypted clusters reading to own function, like it is done for compressed reading.
Quoted 7 lines
block/qcow2.c out: · line 1990
Quoted 5 lines
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>---
block/qcow2.c +2
-0+* but we must not do decryption in guest buffers for security +* reasons. "for security reasons" is a bit handwave-y, no?
Hmm, let's think of it a bit.
WRITE
1. We can't do any operations on write buffers, as guest may use them for something else and not prepared for their change. [thx to Den, pointed to this fact]
READ
Quoted 8 lines
Quoted 3 lines
Hmm, here otherwise, guest should not expect something meaningful in buffers until the end of read operation, so theoretically we may decrypt directly in guest buffer.. What is bad with it?
The badness is that the guest can theoretically reverse-engineer the encryption keys if they are savvy enough to grab the contents of the buffer before and after. The guest must NEVER be able to see the encrypted bits, which means decryption requires a bounce buffer.
Our encryption does not protect against a known-plaintext attack?
Max
PATCH v2 3/4 block/qcow2: refactor qcow2_co_pwritev_part
In reply to Vladimir Sementsov-OgievskiySimilarly to previous commit, prepare for parallelizing write-loop iterations.
--- block/qcow2.c | 150 +++++++++++++++++++++++++++++--------------------- 1 file changed, 88 insertions(+), 62 deletions(-)
block/qcow2.c +88 -62
@@ -2235,6 +2235,87 @@ static int handle_alloc_space(BlockDriverState *bs, QCowL2Meta *l2meta)return 0 ; } +/* +* qcow2_co_pwritev_task +* Called with s->lock unlocked +* l2meta - if not NULL, qcow2_co_do_pwritev() will consume it. Caller must not +* use it somehow after qcow2_co_pwritev_task() call +*/ +static coroutine_fn int qcow2_co_pwritev_task(BlockDriverState *bs, +uint64_t file_cluster_offset, +uint64_t offset,uint64_t bytes, +QEMUIOVector *qiov, +uint64_t qiov_offset, +QCowL2Meta *l2meta) +{ +int ret; +BDRVQcow2State *s = bs->opaque; +void *crypt_buf =NULL ; +int offset_in_cluster =offset_into_cluster (s, offset); +QEMUIOVector encrypted_qiov; + +if (bs->encrypted) { +assert (s->crypto); +assert (bytes <=QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); + crypt_buf =qemu_try_blockalign (bs->file->bs, bytes); +if (crypt_buf ==NULL ) { + ret = -ENOMEM ; + goto out_unlocked; + } +qemu_iovec_to_buf (qiov, qiov_offset, crypt_buf, bytes); + +if (qcow2_co_encrypt (bs, file_cluster_offset, offset, + crypt_buf, bytes) <0 ) { + ret = -EIO ; + goto out_unlocked; + } + +qemu_iovec_init_buf (&encrypted_qiov, crypt_buf, bytes); + qiov = &encrypted_qiov; + qiov_offset =0 ; + } + +/* Try to efficiently initialize the physical space with zeroes */ + ret =handle_alloc_space (bs, l2meta); +if (ret <0 ) { + goto out_unlocked; + } + +/* +* If we need to do COW, check if it's possible to merge the +* writing of the guest data together with that of the COW regions. +* If it's not possible (or not necessary) then write the +* guest data now. +*/ +if (!merge_cow (offset, bytes, qiov, qiov_offset, l2meta)) { +BLKDBG_EVENT (bs->file,BLKDBG_WRITE_AIO ); +trace_qcow2_writev_data (qemu_coroutine_self (), + file_cluster_offset + offset_in_cluster); + ret =bdrv_co_pwritev_part (s->data_file, + file_cluster_offset + offset_in_cluster, + bytes, qiov, qiov_offset,0 ); +if (ret <0 ) { + goto out_unlocked; + } + } + +qemu_co_mutex_lock (&s->lock); + + ret =qcow2_handle_l2meta (bs, &l2meta, true); + goto out_locked; + +out_unlocked: +qemu_co_mutex_lock (&s->lock); + +out_locked: +qcow2_handle_l2meta (bs, &l2meta, false); +qemu_co_mutex_unlock (&s->lock); + +qemu_vfree (crypt_buf); + +return ret; +} + static coroutine_fn int qcow2_co_pwritev_part(BlockDriverState *bs,uint64_t offset,uint64_t bytes,QEMUIOVector *qiov,size_t qiov_offset,int flags)
@@ -2244,15 +2325,11 @@ static coroutine_fn int qcow2_co_pwritev_part(int ret;unsigned int cur_bytes;/* number of sectors in current iteration */ uint64_t cluster_offset; -QEMUIOVector encrypted_qiov;uint64_t bytes_done =0 ; -uint8_t *cluster_data =NULL ;QCowL2Meta *l2meta =NULL ;trace_qcow2_writev_start_req (qemu_coroutine_self (), offset, bytes); -qemu_co_mutex_lock (&s->lock); -while (bytes !=0 ) { l2meta =NULL ;
@@ -2266,6 +2343,8 @@ static coroutine_fn int qcow2_co_pwritev_part( - offset_in_cluster); } +qemu_co_mutex_lock (&s->lock); + ret =qcow2_alloc_cluster_offset (bs, offset, &cur_bytes, &cluster_offset, &l2meta);if (ret <0 ) {
@@ -2283,62 +2362,11 @@ static coroutine_fn int qcow2_co_pwritev_part(qemu_co_mutex_unlock (&s->lock); -if (bs->encrypted) { -assert (s->crypto); -if (!cluster_data) { - cluster_data =qemu_try_blockalign (bs->file->bs, -QCOW_MAX_CRYPT_CLUSTERS - * s->cluster_size); -if (cluster_data ==NULL ) { - ret = -ENOMEM ; - goto out_unlocked; - } - } - -assert (cur_bytes <=QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); -qemu_iovec_to_buf (qiov, qiov_offset + bytes_done, - cluster_data, cur_bytes); - -if (qcow2_co_encrypt (bs, cluster_offset, offset, - cluster_data, cur_bytes) <0 ) { - ret = -EIO ; - goto out_unlocked; - } - -qemu_iovec_init_buf (&encrypted_qiov, cluster_data, cur_bytes); - } - -/* Try to efficiently initialize the physical space with zeroes */ - ret =handle_alloc_space (bs, l2meta); + ret =qcow2_co_pwritev_task (bs, cluster_offset, offset, cur_bytes, + qiov, bytes_done, l2meta); + l2meta =NULL ;/* l2meta is consumed by qcow2_co_do_pwritev() */ if (ret <0 ) { - goto out_unlocked; - } - -/* If we need to do COW, check if it's possible to merge the -* writing of the guest data together with that of the COW regions. -* If it's not possible (or not necessary) then write the -* guest data now. */ -if (!merge_cow (offset, cur_bytes, - bs->encrypted ? &encrypted_qiov : qiov, - bs->encrypted ?0 : qiov_offset + bytes_done, l2meta)) - { -BLKDBG_EVENT (bs->file,BLKDBG_WRITE_AIO ); -trace_qcow2_writev_data (qemu_coroutine_self (), - cluster_offset + offset_in_cluster); - ret =bdrv_co_pwritev_part ( - s->data_file, cluster_offset + offset_in_cluster, cur_bytes, - bs->encrypted ? &encrypted_qiov : qiov, - bs->encrypted ?0 : qiov_offset + bytes_done,0 ); -if (ret <0 ) { - goto out_unlocked; - } - } - -qemu_co_mutex_lock (&s->lock); - - ret =qcow2_handle_l2meta (bs, &l2meta, true); -if (ret) { - goto out_locked; + goto fail_nometa; } bytes -= cur_bytes;
@@ -2347,9 +2375,7 @@ static coroutine_fn int qcow2_co_pwritev_part(trace_qcow2_writev_done_part (qemu_coroutine_self (), cur_bytes); } ret =0 ; - goto out_locked; -out_unlocked:qemu_co_mutex_lock (&s->lock); out_locked:
@@ -2357,7 +2383,7 @@ out_locked:qemu_co_mutex_unlock (&s->lock); -qemu_vfree (cluster_data); +fail_nometa:trace_qcow2_writev_done_req (qemu_coroutine_self (), ret);return ret;
Signature · 1 line
2.18.0
Re: [Qemu-devel] [PATCH v2 3/4] block/qcow2: refactor qcow2_co_pwritev_part
Quoted 12 lines · Vladimir Sementsov-Ogievskiy
Similarly to previous commit, prepare for parallelizing write-loop iterations.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- block/qcow2.c | 150 +++++++++++++++++++++++++++++--------------------- 1 file changed, 88 insertions(+), 62 deletions(-)block/qcow2.c +0
-0
[...]
Quoted 1 line
block/qcow2.c +0
-0@@ -2283,62 +2362,11 @@ static coroutine_fn int qcow2_co_pwritev_part(
[...]
block/qcow2.c static coroutine_fn int qcow2_co_pwritev_part( · line 2366
Quoted 2 lines
block/qcow2.c +2
-0+ ret =qcow2_co_pwritev_task (bs, cluster_offset, offset, cur_bytes, + qiov, bytes_done, l2meta);
You’re passing bytes_done as qiov_offset here. That is initialized to 0, so it ignores the qiov_offset given to qcow2_co_pwritev_part().
Max
Re: [Qemu-devel] [PATCH v2 3/4] block/qcow2: refactor qcow2_co_pwritev_part
In reply to Vladimir Sementsov-OgievskiyQuoted 12 lines · Vladimir Sementsov-Ogievskiy
Similarly to previous commit, prepare for parallelizing write-loop iterations.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- block/qcow2.c | 150 +++++++++++++++++++++++++++++--------------------- 1 file changed, 88 insertions(+), 62 deletions(-)block/qcow2.c +0
-0
[...]
Quoted 2 lines
+ * l2meta - if not NULL,
qcow2_co_do_pwritev()will consume it. Caller must not
+ * use it somehow afterqcow2_co_pwritev_task()call
[...]
Quoted 1 line
+ l2meta = NULL; /* l2meta is consumed by
qcow2_co_do_pwritev()*/
By the way, qcow2_co_do_pwritev() does not exist. :-)
Max
PATCH v2 4/4 block/qcow2: introduce parallel subrequest handling in read and write
In reply to Vladimir Sementsov-OgievskiyIt improves performance for fragmented qcow2 images.
--- block/qcow2.c | 125 +++++++++++++++++++++++++++++++++++++++++---- block/trace-events | 1 + 2 files changed, 115 insertions(+), 11 deletions(-)
block/qcow2.c +114 -11
@@ -40,6 +40,7 @@#include "qapi/qobject-input-visitor.h" #include "qapi/qapi-visit-block-core.h" #include "crypto.h" +#include "aio_task.h" /* Differences with QCOW:
@@ -2017,6 +2018,62 @@ fail:return ret; } +typedef struct Qcow2AioTask { +AioTask task; + +BlockDriverState *bs; +QCow2ClusterType cluster_type;/* only for read */ +uint64_t file_cluster_offset; +uint64_t offset; +uint64_t bytes; +QEMUIOVector *qiov; +uint64_t qiov_offset; +QCowL2Meta *l2meta;/* only for write */ +}Qcow2AioTask ; + +#define QCOW2_MAX_WORKERS 8 + +static coroutine_fn int qcow2_co_preadv_task_entry(AioTask *task); +static coroutine_fn int qcow2_add_task(BlockDriverState *bs, +AioTaskPool *pool, +AioTaskFunc func, +QCow2ClusterType cluster_type, +uint64_t file_cluster_offset, +uint64_t offset, +uint64_t bytes, +QEMUIOVector *qiov, +size_t qiov_offset, +QCowL2Meta *l2meta) +{ +Qcow2AioTask local_task; +Qcow2AioTask *task = pool ?g_new (Qcow2AioTask,1 ) : &local_task; + + *task = (Qcow2AioTask ) { + .task.func = func, + .bs = bs, + .cluster_type = cluster_type, + .qiov = qiov, + .file_cluster_offset = file_cluster_offset, + .offset = offset, + .bytes = bytes, + .qiov_offset = qiov_offset, + .l2meta = l2meta, + }; + +trace_qcow2_add_task (qemu_coroutine_self (), bs, pool, + func == qcow2_co_preadv_task_entry ?"read" :"write" , + cluster_type, file_cluster_offset, offset, bytes, + qiov, qiov_offset); + +if (!pool) { +return func (&task->task); + } + +aio_task_pool_start_task (pool, &task->task); + +return 0 ; +} + static coroutine_fn int qcow2_co_preadv_task(BlockDriverState *bs, QCow2ClusterType cluster_type, uint64_t file_cluster_offset,
@@ -2067,6 +2124,16 @@ static coroutine_fn int qcow2_co_preadv_task(BlockDriverState *bs,return -EIO ; } +static coroutine_fn int qcow2_co_preadv_task_entry(AioTask *task) +{ +Qcow2AioTask *t =container_of (task, Qcow2AioTask, task); + +assert (!t->l2meta); + +return qcow2_co_preadv_task (t->bs, t->cluster_type, t->file_cluster_offset, + t->offset, t->bytes, t->qiov, t->qiov_offset); +} + static coroutine_fn int qcow2_co_preadv_part(BlockDriverState *bs, uint64_t offset, uint64_t bytes, QEMUIOVector *qiov,
@@ -2076,9 +2143,9 @@ static coroutine_fn int qcow2_co_preadv_part(BlockDriverState *bs,int ret;unsigned int cur_bytes;/* number of bytes in current iteration */ uint64_t cluster_offset =0 ; +AioTaskPool *aio =NULL ; -while (bytes !=0 ) { - +while (bytes !=0 &&aio_task_pool_status (aio) ==0 ) {/* prepare next request */ cur_bytes =MIN (bytes,INT_MAX );if (s->crypto) {
@@ -2090,7 +2157,7 @@ static coroutine_fn int qcow2_co_preadv_part(BlockDriverState *bs, ret =qcow2_get_cluster_offset (bs, offset, &cur_bytes, &cluster_offset);qemu_co_mutex_unlock (&s->lock);if (ret <0 ) { -return ret; + goto out; } if (ret == QCOW2_CLUSTER_ZERO_PLAIN ||
@@ -2099,11 +2166,14 @@ static coroutine_fn int qcow2_co_preadv_part(BlockDriverState *bs, {qemu_iovec_memset (qiov, qiov_offset,0 , cur_bytes); } else { - ret =qcow2_co_preadv_task (bs, ret, - cluster_offset, offset, cur_bytes, - qiov, qiov_offset); +if (!aio && cur_bytes != bytes) { + aio =aio_task_pool_new (QCOW2_MAX_WORKERS ); + } + ret =qcow2_add_task (bs, aio, qcow2_co_preadv_task_entry, ret, + cluster_offset, offset, cur_bytes, + qiov, qiov_offset,NULL );if (ret <0 ) { -return ret; + goto out; } }
@@ -2112,7 +2182,16 @@ static coroutine_fn int qcow2_co_preadv_part(BlockDriverState *bs, qiov_offset += cur_bytes; } -return 0 ; +out: +if (aio) { +aio_task_pool_wait_all (aio); +if (ret ==0 ) { + ret =aio_task_pool_status (aio); + } +g_free (aio); + } + +return ret; }/* Check if it's possible to merge a write request with the writing of
@@ -2316,6 +2395,17 @@ out_locked:return ret; } +static coroutine_fn int qcow2_co_pwritev_task_entry(AioTask *task) +{ +Qcow2AioTask *t =container_of (task, Qcow2AioTask, task); + +assert (!t->cluster_type); + +return qcow2_co_pwritev_task (t->bs, t->file_cluster_offset, + t->offset, t->bytes, t->qiov, t->qiov_offset, + t->l2meta); +} + static coroutine_fn int qcow2_co_pwritev_part(BlockDriverState *bs,uint64_t offset,uint64_t bytes,QEMUIOVector *qiov,size_t qiov_offset,int flags)
@@ -2327,10 +2417,11 @@ static coroutine_fn int qcow2_co_pwritev_part(uint64_t cluster_offset;uint64_t bytes_done =0 ;QCowL2Meta *l2meta =NULL ; +AioTaskPool *aio =NULL ;trace_qcow2_writev_start_req (qemu_coroutine_self (), offset, bytes); -while (bytes !=0 ) { +while (bytes !=0 &&aio_task_pool_status (aio) ==0 ) { l2meta =NULL ;
@@ -2362,8 +2453,12 @@ static coroutine_fn int qcow2_co_pwritev_part(qemu_co_mutex_unlock (&s->lock); - ret =qcow2_co_pwritev_task (bs, cluster_offset, offset, cur_bytes, - qiov, bytes_done, l2meta); +if (!aio && cur_bytes != bytes) { + aio =aio_task_pool_new (QCOW2_MAX_WORKERS ); + } + ret =qcow2_add_task (bs, aio, qcow2_co_pwritev_task_entry,0 , + cluster_offset, offset, cur_bytes, + qiov, bytes_done, l2meta); l2meta =NULL ;/* l2meta is consumed by qcow2_co_do_pwritev() */ if (ret <0 ) { goto fail_nometa;
@@ -2384,6 +2479,14 @@ out_locked:qemu_co_mutex_unlock (&s->lock); fail_nometa: +if (aio) { +aio_task_pool_wait_all (aio); +if (ret ==0 ) { + ret =aio_task_pool_status (aio); + } +g_free (aio); + } +trace_qcow2_writev_done_req (qemu_coroutine_self (), ret);return ret;
block/trace-events +1 -0
@@ -61,6 +61,7 @@ file_paio_submit(void *acb, void *opaque, int64_t offset, int count, int type) " file_copy_file_range(void *bs, int src, int64_t src_off, int dst, int64_t dst_off, int64_t bytes, int flags, int64_t ret) "bs %p src_fd %d offset %"PRIu64" dst_fd %d offset %"PRIu64" bytes %"PRIu64" flags %d ret %"PRId64 # qcow2.c +qcow2_add_task(void *co, void *bs, void *pool, const char *action, int cluster_type, uint64_t file_cluster_offset, uint64_t offset, uint64_t bytes, void *qiov, size_t qiov_offset) "co %p bs %p pool %p: %s: cluster_type %d file_cluster_offset %" PRIu64 " offset %" PRIu64 " bytes %" PRIu64 " qiov %p qiov_offset %zu" qcow2_writev_start_req(void *co, int64_t offset, int bytes) "co %p offset 0x%" PRIx64 " bytes %d" qcow2_writev_done_req(void *co, int ret) "co %p ret %d" qcow2_writev_start_part(void *co) "co %p"
Signature · 1 line
2.18.0
Re: [Qemu-devel] [PATCH v2 4/4] block/qcow2: introduce parallel subrequest handling in read and write
Quoted 12 lines · Vladimir Sementsov-Ogievskiy
It improves performance for fragmented qcow2 images.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- block/qcow2.c | 125 +++++++++++++++++++++++++++++++++++++++++---- block/trace-events | 1 + 2 files changed, 115 insertions(+), 11 deletions(-)block/qcow2.c +0
-0
[...]
Quoted 8 lines
block/qcow2.c +4
-0@@ -2017,6 +2018,62 @@ fail:return ret; } +typedef struct Qcow2AioTask { + AioTask task; + + BlockDriverState *bs;
block/qcow2.c fail: · line 2034
Quoted 10 lines
block/qcow2.c +10
-0+QCow2ClusterType cluster_type;/* only for read */ +uint64_t file_cluster_offset; +uint64_t offset; +uint64_t bytes; +QEMUIOVector *qiov; +uint64_t qiov_offset; +QCowL2Meta *l2meta;/* only for write */ +} Qcow2AioTask; + +#define QCOW2_MAX_WORKERS 8
Maybe move this to the top, or even qcow2.h?
[...]
Quoted 5 lines
block/qcow2.c +0
-1@@ -2112,7 +2182,16 @@ static coroutine_fn int qcow2_co_preadv_part(BlockDriverState *bs, qiov_offset += cur_bytes; } -return 0 ;
block/qcow2.c static coroutine_fn int qcow2_co_preadv_part(BlockDriverState *bs, · line 2194
Quoted 10 lines
block/qcow2.c +10
-0+out: +if (aio) { +aio_task_pool_wait_all (aio); +if (ret ==0 ) { + ret =aio_task_pool_status (aio); + } +g_free (aio); + } + +return ret;
My gcc complains here that ret may be initialized. (Which is indeed the case if @bytes is 0.)
The rest looks good to me.
Max