1
The following changes since commit 17780edd81d27fcfdb7a802efc870a99788bd2fc:
1
The following changes since commit 281f327487c9c9b1599f93c589a408bbf4a651b8:
2
2
3
Merge tag 'quick-fix-pull-request' of https://gitlab.com/bsdimp/qemu into staging (2023-08-31 10:06:29 -0400)
3
Merge remote-tracking branch 'remotes/vivier/tags/m68k-for-2.12-pull-request' into staging (2017-12-22 00:11:36 +0000)
4
4
5
are available in the Git repository at:
5
are available in the git repository at:
6
6
7
https://repo.or.cz/qemu/kevin.git tags/for-upstream
7
git://repo.or.cz/qemu/kevin.git tags/for-upstream
8
8
9
for you to fetch changes up to bb86eb45297840c31dbc4df6bac02e50596f2376:
9
for you to fetch changes up to 1a63a907507fbbcfaee3f622907ec244b7eabda8:
10
10
11
block: Remove unnecessary variable in bdrv_block_device_info (2023-09-04 11:03:28 +0200)
11
block: Keep nodes drained between reopen_queue/multiple (2017-12-22 15:05:32 +0100)
12
12
13
----------------------------------------------------------------
13
----------------------------------------------------------------
14
Block layer patches
14
Block layer patches
15
15
16
- Process I/O in the current AioContext (instead of the BB AioContext)
16
----------------------------------------------------------------
17
- Optimise reqs_lock to make multiqueue actually scale
17
Doug Gale (1):
18
- iotests: Fix reference output for some tests after recent changes
18
nvme: Add tracing
19
- vpc: Avoid dynamic stack allocation
20
- Code cleanup, improved documentation
21
19
22
----------------------------------------------------------------
20
Edgar Kaziakhmedov (1):
23
Dmitry Frolov (1):
21
qcow2: get rid of qcow2_backing_read1 routine
24
vmdk: Clean up bdrv_open_child() return value check
25
22
26
Fabiano Rosas (2):
23
Fam Zheng (2):
27
block: Remove bdrv_query_block_node_info
24
block: Open backing image in force share mode for size probe
28
block: Remove unnecessary variable in bdrv_block_device_info
25
block: Remove unused bdrv_requests_pending
29
26
30
Fiona Ebner (1):
27
John Snow (1):
31
iotests: adapt test output for new qemu_cleanup() behavior
28
iotests: fix 197 for vpc
32
29
33
Hanna Czenczek (1):
30
Kevin Wolf (27):
34
block: Be more verbose in create fallback
31
block: Formats don't need CONSISTENT_READ with NO_IO
32
block: Make bdrv_drain_invoke() recursive
33
block: Call .drain_begin only once in bdrv_drain_all_begin()
34
test-bdrv-drain: Test BlockDriver callbacks for drain
35
block: bdrv_drain_recurse(): Remove unused begin parameter
36
block: Don't wait for requests in bdrv_drain*_end()
37
block: Unify order in drain functions
38
block: Don't acquire AioContext in hmp_qemu_io()
39
block: Document that x-blockdev-change breaks quorum children list
40
block: Assert drain_all is only called from main AioContext
41
block: Make bdrv_drain() driver callbacks non-recursive
42
test-bdrv-drain: Test callback for bdrv_drain
43
test-bdrv-drain: Test bs->quiesce_counter
44
blockjob: Pause job on draining any job BDS
45
test-bdrv-drain: Test drain vs. block jobs
46
block: Don't block_job_pause_all() in bdrv_drain_all()
47
block: Nested drain_end must still call callbacks
48
test-bdrv-drain: Test nested drain sections
49
block: Don't notify parents in drain call chain
50
block: Add bdrv_subtree_drained_begin/end()
51
test-bdrv-drain: Tests for bdrv_subtree_drain
52
test-bdrv-drain: Test behaviour in coroutine context
53
test-bdrv-drain: Recursive draining with multiple parents
54
block: Allow graph changes in subtree drained section
55
test-bdrv-drain: Test graph changes in drained section
56
commit: Simplify reopen of base
57
block: Keep nodes drained between reopen_queue/multiple
35
58
36
Kevin Wolf (1):
59
Thomas Huth (3):
37
qemu-img: Update documentation for compressed images
60
block: Remove the obsolete -drive boot=on|off parameter
61
block: Remove the deprecated -hdachs option
62
block: Mention -drive cyls/heads/secs/trans/serial/addr in deprecation chapter
38
63
39
Michael Tokarev (1):
64
qapi/block-core.json | 4 +
40
qemu-img: omit errno value in error message
65
block/qcow2.h | 3 -
66
include/block/block.h | 15 +-
67
include/block/block_int.h | 6 +-
68
block.c | 75 ++++-
69
block/commit.c | 8 +-
70
block/io.c | 164 +++++++---
71
block/qcow2.c | 51 +--
72
block/replication.c | 6 +
73
blockdev.c | 11 -
74
blockjob.c | 22 +-
75
hmp.c | 6 -
76
hw/block/nvme.c | 349 +++++++++++++++++----
77
qemu-io-cmds.c | 3 +
78
tests/test-bdrv-drain.c | 651 +++++++++++++++++++++++++++++++++++++++
79
vl.c | 86 +-----
80
hw/block/trace-events | 93 ++++++
81
qemu-doc.texi | 29 +-
82
qemu-options.hx | 19 +-
83
tests/Makefile.include | 2 +
84
tests/qemu-iotests/197 | 4 +
85
tests/qemu-iotests/common.filter | 3 +-
86
22 files changed, 1294 insertions(+), 316 deletions(-)
87
create mode 100644 tests/test-bdrv-drain.c
41
88
42
Peter Maydell (1):
43
block/iscsi: Document why we use raw malloc()
44
45
Philippe Mathieu-Daudé (1):
46
block/vpc: Avoid dynamic stack allocation
47
48
Stefan Hajnoczi (5):
49
block: minimize bs->reqs_lock section in tracked_request_end()
50
block: change reqs_lock to QemuMutex
51
block: remove AIOCBInfo->get_aio_context()
52
block-backend: process I/O in the current AioContext
53
block-backend: process zoned requests in the current AioContext
54
55
docs/tools/qemu-img.rst | 19 ++++++++++++--
56
include/block/aio.h | 1 -
57
include/block/block-global-state.h | 2 ++
58
include/block/block-io.h | 1 -
59
include/block/block_int-common.h | 2 +-
60
include/block/qapi.h | 3 ---
61
block.c | 10 ++++---
62
block/block-backend.c | 35 +++++++------------------
63
block/io.c | 53 +++++++++++++++++++-------------------
64
block/iscsi.c | 1 +
65
block/qapi.c | 32 ++---------------------
66
block/vmdk.c | 2 +-
67
block/vpc.c | 4 +--
68
hw/nvme/ctrl.c | 7 -----
69
qemu-img.c | 4 +--
70
softmmu/dma-helpers.c | 8 ------
71
util/thread-pool.c | 8 ------
72
tests/qemu-iotests/080.out | 6 ++---
73
tests/qemu-iotests/109.out | 24 +++++++++++++++++
74
tests/qemu-iotests/112.out | 6 ++---
75
tests/qemu-iotests/185 | 2 ++
76
tests/qemu-iotests/185.out | 4 +++
77
tests/qemu-iotests/244.out | 2 +-
78
23 files changed, 107 insertions(+), 129 deletions(-)
79
80
diff view generated by jsdifflib
New patch
1
Commit 1f4ad7d fixed 'qemu-img info' for raw images that are currently
2
in use as a mirror target. It is not enough for image formats, though,
3
as these still unconditionally request BLK_PERM_CONSISTENT_READ.
1
4
5
As this permission is geared towards whether the guest-visible data is
6
consistent, and has no impact on whether the metadata is sane, and
7
'qemu-img info' does not read guest-visible data (except for the raw
8
format), it makes sense to not require BLK_PERM_CONSISTENT_READ if there
9
is not going to be any guest I/O performed, regardless of image format.
10
11
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
12
---
13
block.c | 6 +++++-
14
1 file changed, 5 insertions(+), 1 deletion(-)
15
16
diff --git a/block.c b/block.c
17
index XXXXXXX..XXXXXXX 100644
18
--- a/block.c
19
+++ b/block.c
20
@@ -XXX,XX +XXX,XX @@ void bdrv_format_default_perms(BlockDriverState *bs, BdrvChild *c,
21
assert(role == &child_backing || role == &child_file);
22
23
if (!backing) {
24
+ int flags = bdrv_reopen_get_flags(reopen_queue, bs);
25
+
26
/* Apart from the modifications below, the same permissions are
27
* forwarded and left alone as for filters */
28
bdrv_filter_default_perms(bs, c, role, reopen_queue, perm, shared,
29
@@ -XXX,XX +XXX,XX @@ void bdrv_format_default_perms(BlockDriverState *bs, BdrvChild *c,
30
31
/* bs->file always needs to be consistent because of the metadata. We
32
* can never allow other users to resize or write to it. */
33
- perm |= BLK_PERM_CONSISTENT_READ;
34
+ if (!(flags & BDRV_O_NO_IO)) {
35
+ perm |= BLK_PERM_CONSISTENT_READ;
36
+ }
37
shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
38
} else {
39
/* We want consistent read from backing files if the parent needs it.
40
--
41
2.13.6
42
43
diff view generated by jsdifflib
New patch
1
From: John Snow <jsnow@redhat.com>
1
2
3
VPC has some difficulty creating geometries of particular size.
4
However, we can indeed force it to use a literal one, so let's
5
do that for the sake of test 197, which is testing some specific
6
offsets.
7
8
Signed-off-by: John Snow <jsnow@redhat.com>
9
Reviewed-by: Eric Blake <eblake@redhat.com>
10
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
11
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
12
Reviewed-by: Lukáš Doktor <ldoktor@redhat.com>
13
---
14
tests/qemu-iotests/197 | 4 ++++
15
tests/qemu-iotests/common.filter | 3 ++-
16
2 files changed, 6 insertions(+), 1 deletion(-)
17
18
diff --git a/tests/qemu-iotests/197 b/tests/qemu-iotests/197
19
index XXXXXXX..XXXXXXX 100755
20
--- a/tests/qemu-iotests/197
21
+++ b/tests/qemu-iotests/197
22
@@ -XXX,XX +XXX,XX @@ echo '=== Copy-on-read ==='
23
echo
24
25
# Prep the images
26
+# VPC rounds image sizes to a specific geometry, force a specific size.
27
+if [ "$IMGFMT" = "vpc" ]; then
28
+ IMGOPTS=$(_optstr_add "$IMGOPTS" "force_size")
29
+fi
30
_make_test_img 4G
31
$QEMU_IO -c "write -P 55 3G 1k" "$TEST_IMG" | _filter_qemu_io
32
IMGPROTO=file IMGFMT=qcow2 IMGOPTS= TEST_IMG_FILE="$TEST_WRAP" \
33
diff --git a/tests/qemu-iotests/common.filter b/tests/qemu-iotests/common.filter
34
index XXXXXXX..XXXXXXX 100644
35
--- a/tests/qemu-iotests/common.filter
36
+++ b/tests/qemu-iotests/common.filter
37
@@ -XXX,XX +XXX,XX @@ _filter_img_create()
38
-e "s# log_size=[0-9]\\+##g" \
39
-e "s# refcount_bits=[0-9]\\+##g" \
40
-e "s# key-secret=[a-zA-Z0-9]\\+##g" \
41
- -e "s# iter-time=[0-9]\\+##g"
42
+ -e "s# iter-time=[0-9]\\+##g" \
43
+ -e "s# force_size=\\(on\\|off\\)##g"
44
}
45
46
_filter_img_info()
47
--
48
2.13.6
49
50
diff view generated by jsdifflib
New patch
1
This change separates bdrv_drain_invoke(), which calls the BlockDriver
2
drain callbacks, from bdrv_drain_recurse(). Instead, the function
3
performs its own recursion now.
1
4
5
One reason for this is that bdrv_drain_recurse() can be called multiple
6
times by bdrv_drain_all_begin(), but the callbacks may only be called
7
once. The separation is necessary to fix this bug.
8
9
The other reason is that we intend to go to a model where we call all
10
driver callbacks first, and only then start polling. This is not fully
11
achieved yet with this patch, as bdrv_drain_invoke() contains a
12
BDRV_POLL_WHILE() loop for the block driver callbacks, which can still
13
call callbacks for any unrelated event. It's a step in this direction
14
anyway.
15
16
Cc: qemu-stable@nongnu.org
17
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
18
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
19
---
20
block/io.c | 14 +++++++++++---
21
1 file changed, 11 insertions(+), 3 deletions(-)
22
23
diff --git a/block/io.c b/block/io.c
24
index XXXXXXX..XXXXXXX 100644
25
--- a/block/io.c
26
+++ b/block/io.c
27
@@ -XXX,XX +XXX,XX @@ static void coroutine_fn bdrv_drain_invoke_entry(void *opaque)
28
bdrv_wakeup(bs);
29
}
30
31
+/* Recursively call BlockDriver.bdrv_co_drain_begin/end callbacks */
32
static void bdrv_drain_invoke(BlockDriverState *bs, bool begin)
33
{
34
+ BdrvChild *child, *tmp;
35
BdrvCoDrainData data = { .bs = bs, .done = false, .begin = begin};
36
37
if (!bs->drv || (begin && !bs->drv->bdrv_co_drain_begin) ||
38
@@ -XXX,XX +XXX,XX @@ static void bdrv_drain_invoke(BlockDriverState *bs, bool begin)
39
data.co = qemu_coroutine_create(bdrv_drain_invoke_entry, &data);
40
bdrv_coroutine_enter(bs, data.co);
41
BDRV_POLL_WHILE(bs, !data.done);
42
+
43
+ QLIST_FOREACH_SAFE(child, &bs->children, next, tmp) {
44
+ bdrv_drain_invoke(child->bs, begin);
45
+ }
46
}
47
48
static bool bdrv_drain_recurse(BlockDriverState *bs, bool begin)
49
@@ -XXX,XX +XXX,XX @@ static bool bdrv_drain_recurse(BlockDriverState *bs, bool begin)
50
BdrvChild *child, *tmp;
51
bool waited;
52
53
- /* Ensure any pending metadata writes are submitted to bs->file. */
54
- bdrv_drain_invoke(bs, begin);
55
-
56
/* Wait for drained requests to finish */
57
waited = BDRV_POLL_WHILE(bs, atomic_read(&bs->in_flight) > 0);
58
59
@@ -XXX,XX +XXX,XX @@ void bdrv_drained_begin(BlockDriverState *bs)
60
bdrv_parent_drained_begin(bs);
61
}
62
63
+ bdrv_drain_invoke(bs, true);
64
bdrv_drain_recurse(bs, true);
65
}
66
67
@@ -XXX,XX +XXX,XX @@ void bdrv_drained_end(BlockDriverState *bs)
68
}
69
70
bdrv_parent_drained_end(bs);
71
+ bdrv_drain_invoke(bs, false);
72
bdrv_drain_recurse(bs, false);
73
aio_enable_external(bdrv_get_aio_context(bs));
74
}
75
@@ -XXX,XX +XXX,XX @@ void bdrv_drain_all_begin(void)
76
aio_context_acquire(aio_context);
77
for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
78
if (aio_context == bdrv_get_aio_context(bs)) {
79
+ /* FIXME Calling this multiple times is wrong */
80
+ bdrv_drain_invoke(bs, true);
81
waited |= bdrv_drain_recurse(bs, true);
82
}
83
}
84
@@ -XXX,XX +XXX,XX @@ void bdrv_drain_all_end(void)
85
aio_context_acquire(aio_context);
86
aio_enable_external(aio_context);
87
bdrv_parent_drained_end(bs);
88
+ bdrv_drain_invoke(bs, false);
89
bdrv_drain_recurse(bs, false);
90
aio_context_release(aio_context);
91
}
92
--
93
2.13.6
94
95
diff view generated by jsdifflib
New patch
1
bdrv_drain_all_begin() used to call the .bdrv_co_drain_begin() driver
2
callback inside its polling loop. This means that how many times it got
3
called for each node depended on long it had to poll the event loop.
1
4
5
This is obviously not right and results in nodes that stay drained even
6
after bdrv_drain_all_end(), which calls .bdrv_co_drain_begin() once per
7
node.
8
9
Fix bdrv_drain_all_begin() to call the callback only once, too.
10
11
Cc: qemu-stable@nongnu.org
12
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
13
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
14
---
15
block/io.c | 3 +--
16
1 file changed, 1 insertion(+), 2 deletions(-)
17
18
diff --git a/block/io.c b/block/io.c
19
index XXXXXXX..XXXXXXX 100644
20
--- a/block/io.c
21
+++ b/block/io.c
22
@@ -XXX,XX +XXX,XX @@ void bdrv_drain_all_begin(void)
23
aio_context_acquire(aio_context);
24
bdrv_parent_drained_begin(bs);
25
aio_disable_external(aio_context);
26
+ bdrv_drain_invoke(bs, true);
27
aio_context_release(aio_context);
28
29
if (!g_slist_find(aio_ctxs, aio_context)) {
30
@@ -XXX,XX +XXX,XX @@ void bdrv_drain_all_begin(void)
31
aio_context_acquire(aio_context);
32
for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
33
if (aio_context == bdrv_get_aio_context(bs)) {
34
- /* FIXME Calling this multiple times is wrong */
35
- bdrv_drain_invoke(bs, true);
36
waited |= bdrv_drain_recurse(bs, true);
37
}
38
}
39
--
40
2.13.6
41
42
diff view generated by jsdifflib
New patch
1
This adds a test case that the BlockDriver callbacks for drain are
2
called in bdrv_drained_all_begin/end(), and that both of them are called
3
exactly once.
1
4
5
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
6
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
7
Reviewed-by: Eric Blake <eblake@redhat.com>
8
---
9
tests/test-bdrv-drain.c | 137 ++++++++++++++++++++++++++++++++++++++++++++++++
10
tests/Makefile.include | 2 +
11
2 files changed, 139 insertions(+)
12
create mode 100644 tests/test-bdrv-drain.c
13
14
diff --git a/tests/test-bdrv-drain.c b/tests/test-bdrv-drain.c
15
new file mode 100644
16
index XXXXXXX..XXXXXXX
17
--- /dev/null
18
+++ b/tests/test-bdrv-drain.c
19
@@ -XXX,XX +XXX,XX @@
20
+/*
21
+ * Block node draining tests
22
+ *
23
+ * Copyright (c) 2017 Kevin Wolf <kwolf@redhat.com>
24
+ *
25
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
26
+ * of this software and associated documentation files (the "Software"), to deal
27
+ * in the Software without restriction, including without limitation the rights
28
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
29
+ * copies of the Software, and to permit persons to whom the Software is
30
+ * furnished to do so, subject to the following conditions:
31
+ *
32
+ * The above copyright notice and this permission notice shall be included in
33
+ * all copies or substantial portions of the Software.
34
+ *
35
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
36
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
37
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
38
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
39
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
40
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
41
+ * THE SOFTWARE.
42
+ */
43
+
44
+#include "qemu/osdep.h"
45
+#include "block/block.h"
46
+#include "sysemu/block-backend.h"
47
+#include "qapi/error.h"
48
+
49
+typedef struct BDRVTestState {
50
+ int drain_count;
51
+} BDRVTestState;
52
+
53
+static void coroutine_fn bdrv_test_co_drain_begin(BlockDriverState *bs)
54
+{
55
+ BDRVTestState *s = bs->opaque;
56
+ s->drain_count++;
57
+}
58
+
59
+static void coroutine_fn bdrv_test_co_drain_end(BlockDriverState *bs)
60
+{
61
+ BDRVTestState *s = bs->opaque;
62
+ s->drain_count--;
63
+}
64
+
65
+static void bdrv_test_close(BlockDriverState *bs)
66
+{
67
+ BDRVTestState *s = bs->opaque;
68
+ g_assert_cmpint(s->drain_count, >, 0);
69
+}
70
+
71
+static int coroutine_fn bdrv_test_co_preadv(BlockDriverState *bs,
72
+ uint64_t offset, uint64_t bytes,
73
+ QEMUIOVector *qiov, int flags)
74
+{
75
+ /* We want this request to stay until the polling loop in drain waits for
76
+ * it to complete. We need to sleep a while as bdrv_drain_invoke() comes
77
+ * first and polls its result, too, but it shouldn't accidentally complete
78
+ * this request yet. */
79
+ qemu_co_sleep_ns(QEMU_CLOCK_REALTIME, 100000);
80
+
81
+ return 0;
82
+}
83
+
84
+static BlockDriver bdrv_test = {
85
+ .format_name = "test",
86
+ .instance_size = sizeof(BDRVTestState),
87
+
88
+ .bdrv_close = bdrv_test_close,
89
+ .bdrv_co_preadv = bdrv_test_co_preadv,
90
+
91
+ .bdrv_co_drain_begin = bdrv_test_co_drain_begin,
92
+ .bdrv_co_drain_end = bdrv_test_co_drain_end,
93
+};
94
+
95
+static void aio_ret_cb(void *opaque, int ret)
96
+{
97
+ int *aio_ret = opaque;
98
+ *aio_ret = ret;
99
+}
100
+
101
+static void test_drv_cb_drain_all(void)
102
+{
103
+ BlockBackend *blk;
104
+ BlockDriverState *bs;
105
+ BDRVTestState *s;
106
+ BlockAIOCB *acb;
107
+ int aio_ret;
108
+
109
+ QEMUIOVector qiov;
110
+ struct iovec iov = {
111
+ .iov_base = NULL,
112
+ .iov_len = 0,
113
+ };
114
+ qemu_iovec_init_external(&qiov, &iov, 1);
115
+
116
+ blk = blk_new(BLK_PERM_ALL, BLK_PERM_ALL);
117
+ bs = bdrv_new_open_driver(&bdrv_test, "test-node", BDRV_O_RDWR,
118
+ &error_abort);
119
+ s = bs->opaque;
120
+ blk_insert_bs(blk, bs, &error_abort);
121
+
122
+ /* Simple bdrv_drain_all_begin/end pair, check that CBs are called */
123
+ g_assert_cmpint(s->drain_count, ==, 0);
124
+ bdrv_drain_all_begin();
125
+ g_assert_cmpint(s->drain_count, ==, 1);
126
+ bdrv_drain_all_end();
127
+ g_assert_cmpint(s->drain_count, ==, 0);
128
+
129
+ /* Now do the same while a request is pending */
130
+ aio_ret = -EINPROGRESS;
131
+ acb = blk_aio_preadv(blk, 0, &qiov, 0, aio_ret_cb, &aio_ret);
132
+ g_assert(acb != NULL);
133
+ g_assert_cmpint(aio_ret, ==, -EINPROGRESS);
134
+
135
+ g_assert_cmpint(s->drain_count, ==, 0);
136
+ bdrv_drain_all_begin();
137
+ g_assert_cmpint(aio_ret, ==, 0);
138
+ g_assert_cmpint(s->drain_count, ==, 1);
139
+ bdrv_drain_all_end();
140
+ g_assert_cmpint(s->drain_count, ==, 0);
141
+
142
+ bdrv_unref(bs);
143
+ blk_unref(blk);
144
+}
145
+
146
+int main(int argc, char **argv)
147
+{
148
+ bdrv_init();
149
+ qemu_init_main_loop(&error_abort);
150
+
151
+ g_test_init(&argc, &argv, NULL);
152
+
153
+ g_test_add_func("/bdrv-drain/driver-cb/drain_all", test_drv_cb_drain_all);
154
+
155
+ return g_test_run();
156
+}
157
diff --git a/tests/Makefile.include b/tests/Makefile.include
158
index XXXXXXX..XXXXXXX 100644
159
--- a/tests/Makefile.include
160
+++ b/tests/Makefile.include
161
@@ -XXX,XX +XXX,XX @@ gcov-files-test-thread-pool-y = thread-pool.c
162
gcov-files-test-hbitmap-y = util/hbitmap.c
163
check-unit-y += tests/test-hbitmap$(EXESUF)
164
gcov-files-test-hbitmap-y = blockjob.c
165
+check-unit-y += tests/test-bdrv-drain$(EXESUF)
166
check-unit-y += tests/test-blockjob$(EXESUF)
167
check-unit-y += tests/test-blockjob-txn$(EXESUF)
168
check-unit-y += tests/test-x86-cpuid$(EXESUF)
169
@@ -XXX,XX +XXX,XX @@ tests/test-coroutine$(EXESUF): tests/test-coroutine.o $(test-block-obj-y)
170
tests/test-aio$(EXESUF): tests/test-aio.o $(test-block-obj-y)
171
tests/test-aio-multithread$(EXESUF): tests/test-aio-multithread.o $(test-block-obj-y)
172
tests/test-throttle$(EXESUF): tests/test-throttle.o $(test-block-obj-y)
173
+tests/test-bdrv-drain$(EXESUF): tests/test-bdrv-drain.o $(test-block-obj-y) $(test-util-obj-y)
174
tests/test-blockjob$(EXESUF): tests/test-blockjob.o $(test-block-obj-y) $(test-util-obj-y)
175
tests/test-blockjob-txn$(EXESUF): tests/test-blockjob-txn.o $(test-block-obj-y) $(test-util-obj-y)
176
tests/test-thread-pool$(EXESUF): tests/test-thread-pool.o $(test-block-obj-y)
177
--
178
2.13.6
179
180
diff view generated by jsdifflib
New patch
1
Now that the bdrv_drain_invoke() calls are pulled up to the callers of
2
bdrv_drain_recurse(), the 'begin' parameter isn't needed any more.
1
3
4
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
5
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
6
---
7
block/io.c | 12 ++++++------
8
1 file changed, 6 insertions(+), 6 deletions(-)
9
10
diff --git a/block/io.c b/block/io.c
11
index XXXXXXX..XXXXXXX 100644
12
--- a/block/io.c
13
+++ b/block/io.c
14
@@ -XXX,XX +XXX,XX @@ static void bdrv_drain_invoke(BlockDriverState *bs, bool begin)
15
}
16
}
17
18
-static bool bdrv_drain_recurse(BlockDriverState *bs, bool begin)
19
+static bool bdrv_drain_recurse(BlockDriverState *bs)
20
{
21
BdrvChild *child, *tmp;
22
bool waited;
23
@@ -XXX,XX +XXX,XX @@ static bool bdrv_drain_recurse(BlockDriverState *bs, bool begin)
24
*/
25
bdrv_ref(bs);
26
}
27
- waited |= bdrv_drain_recurse(bs, begin);
28
+ waited |= bdrv_drain_recurse(bs);
29
if (in_main_loop) {
30
bdrv_unref(bs);
31
}
32
@@ -XXX,XX +XXX,XX @@ void bdrv_drained_begin(BlockDriverState *bs)
33
}
34
35
bdrv_drain_invoke(bs, true);
36
- bdrv_drain_recurse(bs, true);
37
+ bdrv_drain_recurse(bs);
38
}
39
40
void bdrv_drained_end(BlockDriverState *bs)
41
@@ -XXX,XX +XXX,XX @@ void bdrv_drained_end(BlockDriverState *bs)
42
43
bdrv_parent_drained_end(bs);
44
bdrv_drain_invoke(bs, false);
45
- bdrv_drain_recurse(bs, false);
46
+ bdrv_drain_recurse(bs);
47
aio_enable_external(bdrv_get_aio_context(bs));
48
}
49
50
@@ -XXX,XX +XXX,XX @@ void bdrv_drain_all_begin(void)
51
aio_context_acquire(aio_context);
52
for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
53
if (aio_context == bdrv_get_aio_context(bs)) {
54
- waited |= bdrv_drain_recurse(bs, true);
55
+ waited |= bdrv_drain_recurse(bs);
56
}
57
}
58
aio_context_release(aio_context);
59
@@ -XXX,XX +XXX,XX @@ void bdrv_drain_all_end(void)
60
aio_enable_external(aio_context);
61
bdrv_parent_drained_end(bs);
62
bdrv_drain_invoke(bs, false);
63
- bdrv_drain_recurse(bs, false);
64
+ bdrv_drain_recurse(bs);
65
aio_context_release(aio_context);
66
}
67
68
--
69
2.13.6
70
71
diff view generated by jsdifflib
New patch
1
The device is drained, so there is no point in waiting for requests at
2
the end of the drained section. Remove the bdrv_drain_recurse() calls
3
there.
1
4
5
The bdrv_drain_recurse() calls were introduced in commit 481cad48e5e
6
in order to call the .bdrv_co_drain_end() driver callback. This is now
7
done by a separate bdrv_drain_invoke() call.
8
9
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
10
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
11
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
12
---
13
block/io.c | 2 --
14
1 file changed, 2 deletions(-)
15
16
diff --git a/block/io.c b/block/io.c
17
index XXXXXXX..XXXXXXX 100644
18
--- a/block/io.c
19
+++ b/block/io.c
20
@@ -XXX,XX +XXX,XX @@ void bdrv_drained_end(BlockDriverState *bs)
21
22
bdrv_parent_drained_end(bs);
23
bdrv_drain_invoke(bs, false);
24
- bdrv_drain_recurse(bs);
25
aio_enable_external(bdrv_get_aio_context(bs));
26
}
27
28
@@ -XXX,XX +XXX,XX @@ void bdrv_drain_all_end(void)
29
aio_enable_external(aio_context);
30
bdrv_parent_drained_end(bs);
31
bdrv_drain_invoke(bs, false);
32
- bdrv_drain_recurse(bs);
33
aio_context_release(aio_context);
34
}
35
36
--
37
2.13.6
38
39
diff view generated by jsdifflib
1
From: Stefan Hajnoczi <stefanha@redhat.com>
1
Drain requests are propagated to child nodes, parent nodes and directly
2
to the AioContext. The order in which this happened was different
3
between all combinations of drain/drain_all and begin/end.
2
4
3
Process zoned requests in the current thread's AioContext instead of in
5
The correct order is to keep children only drained when their parents
4
the BlockBackend's AioContext.
6
are also drained. This means that at the start of a drained section, the
7
AioContext needs to be drained first, the parents second and only then
8
the children. The correct order for the end of a drained section is the
9
opposite.
5
10
6
There is no need to use the BlockBackend's AioContext thanks to CoMutex
11
This patch changes the three other functions to follow the example of
7
bs->wps->colock, which protects zone metadata.
12
bdrv_drained_begin(), which is the only one that got it right.
8
13
9
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
10
Message-ID: <20230823235938.1398382-4-stefanha@redhat.com>
11
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
12
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
14
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
15
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
13
---
16
---
14
block/block-backend.c | 12 ++++++------
17
block/io.c | 12 ++++++++----
15
1 file changed, 6 insertions(+), 6 deletions(-)
18
1 file changed, 8 insertions(+), 4 deletions(-)
16
19
17
diff --git a/block/block-backend.c b/block/block-backend.c
20
diff --git a/block/io.c b/block/io.c
18
index XXXXXXX..XXXXXXX 100644
21
index XXXXXXX..XXXXXXX 100644
19
--- a/block/block-backend.c
22
--- a/block/io.c
20
+++ b/block/block-backend.c
23
+++ b/block/io.c
21
@@ -XXX,XX +XXX,XX @@ BlockAIOCB *blk_aio_zone_report(BlockBackend *blk, int64_t offset,
24
@@ -XXX,XX +XXX,XX @@ void bdrv_drained_begin(BlockDriverState *bs)
22
acb->has_returned = false;
25
return;
23
24
co = qemu_coroutine_create(blk_aio_zone_report_entry, acb);
25
- aio_co_enter(blk_get_aio_context(blk), co);
26
+ aio_co_enter(qemu_get_current_aio_context(), co);
27
28
acb->has_returned = true;
29
if (acb->rwco.ret != NOT_DONE) {
30
- replay_bh_schedule_oneshot_event(blk_get_aio_context(blk),
31
+ replay_bh_schedule_oneshot_event(qemu_get_current_aio_context(),
32
blk_aio_complete_bh, acb);
33
}
26
}
34
27
35
@@ -XXX,XX +XXX,XX @@ BlockAIOCB *blk_aio_zone_mgmt(BlockBackend *blk, BlockZoneOp op,
28
+ /* Stop things in parent-to-child order */
36
acb->has_returned = false;
29
if (atomic_fetch_inc(&bs->quiesce_counter) == 0) {
37
30
aio_disable_external(bdrv_get_aio_context(bs));
38
co = qemu_coroutine_create(blk_aio_zone_mgmt_entry, acb);
31
bdrv_parent_drained_begin(bs);
39
- aio_co_enter(blk_get_aio_context(blk), co);
32
@@ -XXX,XX +XXX,XX @@ void bdrv_drained_end(BlockDriverState *bs)
40
+ aio_co_enter(qemu_get_current_aio_context(), co);
33
return;
41
42
acb->has_returned = true;
43
if (acb->rwco.ret != NOT_DONE) {
44
- replay_bh_schedule_oneshot_event(blk_get_aio_context(blk),
45
+ replay_bh_schedule_oneshot_event(qemu_get_current_aio_context(),
46
blk_aio_complete_bh, acb);
47
}
34
}
48
35
49
@@ -XXX,XX +XXX,XX @@ BlockAIOCB *blk_aio_zone_append(BlockBackend *blk, int64_t *offset,
36
- bdrv_parent_drained_end(bs);
50
acb->has_returned = false;
37
+ /* Re-enable things in child-to-parent order */
51
38
bdrv_drain_invoke(bs, false);
52
co = qemu_coroutine_create(blk_aio_zone_append_entry, acb);
39
+ bdrv_parent_drained_end(bs);
53
- aio_co_enter(blk_get_aio_context(blk), co);
40
aio_enable_external(bdrv_get_aio_context(bs));
54
+ aio_co_enter(qemu_get_current_aio_context(), co);
41
}
55
acb->has_returned = true;
42
56
if (acb->rwco.ret != NOT_DONE) {
43
@@ -XXX,XX +XXX,XX @@ void bdrv_drain_all_begin(void)
57
- replay_bh_schedule_oneshot_event(blk_get_aio_context(blk),
44
for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
58
+ replay_bh_schedule_oneshot_event(qemu_get_current_aio_context(),
45
AioContext *aio_context = bdrv_get_aio_context(bs);
59
blk_aio_complete_bh, acb);
46
47
+ /* Stop things in parent-to-child order */
48
aio_context_acquire(aio_context);
49
- bdrv_parent_drained_begin(bs);
50
aio_disable_external(aio_context);
51
+ bdrv_parent_drained_begin(bs);
52
bdrv_drain_invoke(bs, true);
53
aio_context_release(aio_context);
54
55
@@ -XXX,XX +XXX,XX @@ void bdrv_drain_all_end(void)
56
for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
57
AioContext *aio_context = bdrv_get_aio_context(bs);
58
59
+ /* Re-enable things in child-to-parent order */
60
aio_context_acquire(aio_context);
61
- aio_enable_external(aio_context);
62
- bdrv_parent_drained_end(bs);
63
bdrv_drain_invoke(bs, false);
64
+ bdrv_parent_drained_end(bs);
65
+ aio_enable_external(aio_context);
66
aio_context_release(aio_context);
60
}
67
}
61
68
62
--
69
--
63
2.41.0
70
2.13.6
71
72
diff view generated by jsdifflib
New patch
1
Commit 15afd94a047 added code to acquire and release the AioContext in
2
qemuio_command(). This means that the lock is taken twice now in the
3
call path from hmp_qemu_io(). This causes BDRV_POLL_WHILE() to hang for
4
any requests issued to nodes in a non-mainloop AioContext.
1
5
6
Dropping the first locking from hmp_qemu_io() fixes the problem.
7
8
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
9
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
10
---
11
hmp.c | 6 ------
12
1 file changed, 6 deletions(-)
13
14
diff --git a/hmp.c b/hmp.c
15
index XXXXXXX..XXXXXXX 100644
16
--- a/hmp.c
17
+++ b/hmp.c
18
@@ -XXX,XX +XXX,XX @@ void hmp_qemu_io(Monitor *mon, const QDict *qdict)
19
{
20
BlockBackend *blk;
21
BlockBackend *local_blk = NULL;
22
- AioContext *aio_context;
23
const char* device = qdict_get_str(qdict, "device");
24
const char* command = qdict_get_str(qdict, "command");
25
Error *err = NULL;
26
@@ -XXX,XX +XXX,XX @@ void hmp_qemu_io(Monitor *mon, const QDict *qdict)
27
}
28
}
29
30
- aio_context = blk_get_aio_context(blk);
31
- aio_context_acquire(aio_context);
32
-
33
/*
34
* Notably absent: Proper permission management. This is sad, but it seems
35
* almost impossible to achieve without changing the semantics and thereby
36
@@ -XXX,XX +XXX,XX @@ void hmp_qemu_io(Monitor *mon, const QDict *qdict)
37
*/
38
qemuio_command(blk, command);
39
40
- aio_context_release(aio_context);
41
-
42
fail:
43
blk_unref(local_blk);
44
hmp_handle_error(mon, &err);
45
--
46
2.13.6
47
48
diff view generated by jsdifflib
New patch
1
From: Edgar Kaziakhmedov <edgar.kaziakhmedov@virtuozzo.com>
1
2
3
Since bdrv_co_preadv does all neccessary checks including
4
reading after the end of the backing file, avoid duplication
5
of verification before bdrv_co_preadv call.
6
7
Signed-off-by: Edgar Kaziakhmedov <edgar.kaziakhmedov@virtuozzo.com>
8
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
9
Reviewed-by: Eric Blake <eblake@redhat.com>
10
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
11
---
12
block/qcow2.h | 3 ---
13
block/qcow2.c | 51 ++++++++-------------------------------------------
14
2 files changed, 8 insertions(+), 46 deletions(-)
15
16
diff --git a/block/qcow2.h b/block/qcow2.h
17
index XXXXXXX..XXXXXXX 100644
18
--- a/block/qcow2.h
19
+++ b/block/qcow2.h
20
@@ -XXX,XX +XXX,XX @@ uint32_t offset_to_reftable_index(BDRVQcow2State *s, uint64_t offset)
21
}
22
23
/* qcow2.c functions */
24
-int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov,
25
- int64_t sector_num, int nb_sectors);
26
-
27
int64_t qcow2_refcount_metadata_size(int64_t clusters, size_t cluster_size,
28
int refcount_order, bool generous_increase,
29
uint64_t *refblock_count);
30
diff --git a/block/qcow2.c b/block/qcow2.c
31
index XXXXXXX..XXXXXXX 100644
32
--- a/block/qcow2.c
33
+++ b/block/qcow2.c
34
@@ -XXX,XX +XXX,XX @@ static int64_t coroutine_fn qcow2_co_get_block_status(BlockDriverState *bs,
35
return status;
36
}
37
38
-/* handle reading after the end of the backing file */
39
-int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov,
40
- int64_t offset, int bytes)
41
-{
42
- uint64_t bs_size = bs->total_sectors * BDRV_SECTOR_SIZE;
43
- int n1;
44
-
45
- if ((offset + bytes) <= bs_size) {
46
- return bytes;
47
- }
48
-
49
- if (offset >= bs_size) {
50
- n1 = 0;
51
- } else {
52
- n1 = bs_size - offset;
53
- }
54
-
55
- qemu_iovec_memset(qiov, n1, 0, bytes - n1);
56
-
57
- return n1;
58
-}
59
-
60
static coroutine_fn int qcow2_co_preadv(BlockDriverState *bs, uint64_t offset,
61
uint64_t bytes, QEMUIOVector *qiov,
62
int flags)
63
{
64
BDRVQcow2State *s = bs->opaque;
65
- int offset_in_cluster, n1;
66
+ int offset_in_cluster;
67
int ret;
68
unsigned int cur_bytes; /* number of bytes in current iteration */
69
uint64_t cluster_offset = 0;
70
@@ -XXX,XX +XXX,XX @@ static coroutine_fn int qcow2_co_preadv(BlockDriverState *bs, uint64_t offset,
71
case QCOW2_CLUSTER_UNALLOCATED:
72
73
if (bs->backing) {
74
- /* read from the base image */
75
- n1 = qcow2_backing_read1(bs->backing->bs, &hd_qiov,
76
- offset, cur_bytes);
77
- if (n1 > 0) {
78
- QEMUIOVector local_qiov;
79
-
80
- qemu_iovec_init(&local_qiov, hd_qiov.niov);
81
- qemu_iovec_concat(&local_qiov, &hd_qiov, 0, n1);
82
-
83
- BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
84
- qemu_co_mutex_unlock(&s->lock);
85
- ret = bdrv_co_preadv(bs->backing, offset, n1,
86
- &local_qiov, 0);
87
- qemu_co_mutex_lock(&s->lock);
88
-
89
- qemu_iovec_destroy(&local_qiov);
90
-
91
- if (ret < 0) {
92
- goto fail;
93
- }
94
+ BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
95
+ qemu_co_mutex_unlock(&s->lock);
96
+ ret = bdrv_co_preadv(bs->backing, offset, cur_bytes,
97
+ &hd_qiov, 0);
98
+ qemu_co_mutex_lock(&s->lock);
99
+ if (ret < 0) {
100
+ goto fail;
101
}
102
} else {
103
/* Note: in this case, no need to wait */
104
--
105
2.13.6
106
107
diff view generated by jsdifflib
New patch
1
Removing a quorum child node with x-blockdev-change results in a quorum
2
driver state that cannot be recreated with create options because it
3
would require a list with gaps. This causes trouble in at least
4
.bdrv_refresh_filename().
1
5
6
Document this problem so that we won't accidentally mark the command
7
stable without having addressed it.
8
9
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
10
Reviewed-by: Alberto Garcia <berto@igalia.com>
11
---
12
qapi/block-core.json | 4 ++++
13
1 file changed, 4 insertions(+)
14
15
diff --git a/qapi/block-core.json b/qapi/block-core.json
16
index XXXXXXX..XXXXXXX 100644
17
--- a/qapi/block-core.json
18
+++ b/qapi/block-core.json
19
@@ -XXX,XX +XXX,XX @@
20
# does not support all kinds of operations, all kinds of children, nor
21
# all block drivers.
22
#
23
+# FIXME Removing children from a quorum node means introducing gaps in the
24
+# child indices. This cannot be represented in the 'children' list of
25
+# BlockdevOptionsQuorum, as returned by .bdrv_refresh_filename().
26
+#
27
# Warning: The data in a new quorum child MUST be consistent with that of
28
# the rest of the array.
29
#
30
--
31
2.13.6
32
33
diff view generated by jsdifflib
New patch
1
From: Doug Gale <doug16k@gmail.com>
1
2
3
Add trace output for commands, errors, and undefined behavior.
4
Add guest error log output for undefined behavior.
5
Report invalid undefined accesses to MMIO.
6
Annotate unlikely error checks with unlikely.
7
8
Signed-off-by: Doug Gale <doug16k@gmail.com>
9
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
10
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
11
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
12
---
13
hw/block/nvme.c | 349 ++++++++++++++++++++++++++++++++++++++++++--------
14
hw/block/trace-events | 93 ++++++++++++++
15
2 files changed, 390 insertions(+), 52 deletions(-)
16
17
diff --git a/hw/block/nvme.c b/hw/block/nvme.c
18
index XXXXXXX..XXXXXXX 100644
19
--- a/hw/block/nvme.c
20
+++ b/hw/block/nvme.c
21
@@ -XXX,XX +XXX,XX @@
22
#include "qapi/visitor.h"
23
#include "sysemu/block-backend.h"
24
25
+#include "qemu/log.h"
26
+#include "trace.h"
27
#include "nvme.h"
28
29
+#define NVME_GUEST_ERR(trace, fmt, ...) \
30
+ do { \
31
+ (trace_##trace)(__VA_ARGS__); \
32
+ qemu_log_mask(LOG_GUEST_ERROR, #trace \
33
+ " in %s: " fmt "\n", __func__, ## __VA_ARGS__); \
34
+ } while (0)
35
+
36
static void nvme_process_sq(void *opaque);
37
38
static void nvme_addr_read(NvmeCtrl *n, hwaddr addr, void *buf, int size)
39
@@ -XXX,XX +XXX,XX @@ static void nvme_isr_notify(NvmeCtrl *n, NvmeCQueue *cq)
40
{
41
if (cq->irq_enabled) {
42
if (msix_enabled(&(n->parent_obj))) {
43
+ trace_nvme_irq_msix(cq->vector);
44
msix_notify(&(n->parent_obj), cq->vector);
45
} else {
46
+ trace_nvme_irq_pin();
47
pci_irq_pulse(&n->parent_obj);
48
}
49
+ } else {
50
+ trace_nvme_irq_masked();
51
}
52
}
53
54
@@ -XXX,XX +XXX,XX @@ static uint16_t nvme_map_prp(QEMUSGList *qsg, QEMUIOVector *iov, uint64_t prp1,
55
trans_len = MIN(len, trans_len);
56
int num_prps = (len >> n->page_bits) + 1;
57
58
- if (!prp1) {
59
+ if (unlikely(!prp1)) {
60
+ trace_nvme_err_invalid_prp();
61
return NVME_INVALID_FIELD | NVME_DNR;
62
} else if (n->cmbsz && prp1 >= n->ctrl_mem.addr &&
63
prp1 < n->ctrl_mem.addr + int128_get64(n->ctrl_mem.size)) {
64
@@ -XXX,XX +XXX,XX @@ static uint16_t nvme_map_prp(QEMUSGList *qsg, QEMUIOVector *iov, uint64_t prp1,
65
}
66
len -= trans_len;
67
if (len) {
68
- if (!prp2) {
69
+ if (unlikely(!prp2)) {
70
+ trace_nvme_err_invalid_prp2_missing();
71
goto unmap;
72
}
73
if (len > n->page_size) {
74
@@ -XXX,XX +XXX,XX @@ static uint16_t nvme_map_prp(QEMUSGList *qsg, QEMUIOVector *iov, uint64_t prp1,
75
uint64_t prp_ent = le64_to_cpu(prp_list[i]);
76
77
if (i == n->max_prp_ents - 1 && len > n->page_size) {
78
- if (!prp_ent || prp_ent & (n->page_size - 1)) {
79
+ if (unlikely(!prp_ent || prp_ent & (n->page_size - 1))) {
80
+ trace_nvme_err_invalid_prplist_ent(prp_ent);
81
goto unmap;
82
}
83
84
@@ -XXX,XX +XXX,XX @@ static uint16_t nvme_map_prp(QEMUSGList *qsg, QEMUIOVector *iov, uint64_t prp1,
85
prp_ent = le64_to_cpu(prp_list[i]);
86
}
87
88
- if (!prp_ent || prp_ent & (n->page_size - 1)) {
89
+ if (unlikely(!prp_ent || prp_ent & (n->page_size - 1))) {
90
+ trace_nvme_err_invalid_prplist_ent(prp_ent);
91
goto unmap;
92
}
93
94
@@ -XXX,XX +XXX,XX @@ static uint16_t nvme_map_prp(QEMUSGList *qsg, QEMUIOVector *iov, uint64_t prp1,
95
i++;
96
}
97
} else {
98
- if (prp2 & (n->page_size - 1)) {
99
+ if (unlikely(prp2 & (n->page_size - 1))) {
100
+ trace_nvme_err_invalid_prp2_align(prp2);
101
goto unmap;
102
}
103
if (qsg->nsg) {
104
@@ -XXX,XX +XXX,XX @@ static uint16_t nvme_dma_read_prp(NvmeCtrl *n, uint8_t *ptr, uint32_t len,
105
QEMUIOVector iov;
106
uint16_t status = NVME_SUCCESS;
107
108
+ trace_nvme_dma_read(prp1, prp2);
109
+
110
if (nvme_map_prp(&qsg, &iov, prp1, prp2, len, n)) {
111
return NVME_INVALID_FIELD | NVME_DNR;
112
}
113
if (qsg.nsg > 0) {
114
- if (dma_buf_read(ptr, len, &qsg)) {
115
+ if (unlikely(dma_buf_read(ptr, len, &qsg))) {
116
+ trace_nvme_err_invalid_dma();
117
status = NVME_INVALID_FIELD | NVME_DNR;
118
}
119
qemu_sglist_destroy(&qsg);
120
} else {
121
- if (qemu_iovec_to_buf(&iov, 0, ptr, len) != len) {
122
+ if (unlikely(qemu_iovec_to_buf(&iov, 0, ptr, len) != len)) {
123
+ trace_nvme_err_invalid_dma();
124
status = NVME_INVALID_FIELD | NVME_DNR;
125
}
126
qemu_iovec_destroy(&iov);
127
@@ -XXX,XX +XXX,XX @@ static uint16_t nvme_write_zeros(NvmeCtrl *n, NvmeNamespace *ns, NvmeCmd *cmd,
128
uint64_t aio_slba = slba << (data_shift - BDRV_SECTOR_BITS);
129
uint32_t aio_nlb = nlb << (data_shift - BDRV_SECTOR_BITS);
130
131
- if (slba + nlb > ns->id_ns.nsze) {
132
+ if (unlikely(slba + nlb > ns->id_ns.nsze)) {
133
+ trace_nvme_err_invalid_lba_range(slba, nlb, ns->id_ns.nsze);
134
return NVME_LBA_RANGE | NVME_DNR;
135
}
136
137
@@ -XXX,XX +XXX,XX @@ static uint16_t nvme_rw(NvmeCtrl *n, NvmeNamespace *ns, NvmeCmd *cmd,
138
int is_write = rw->opcode == NVME_CMD_WRITE ? 1 : 0;
139
enum BlockAcctType acct = is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ;
140
141
- if ((slba + nlb) > ns->id_ns.nsze) {
142
+ trace_nvme_rw(is_write ? "write" : "read", nlb, data_size, slba);
143
+
144
+ if (unlikely((slba + nlb) > ns->id_ns.nsze)) {
145
block_acct_invalid(blk_get_stats(n->conf.blk), acct);
146
+ trace_nvme_err_invalid_lba_range(slba, nlb, ns->id_ns.nsze);
147
return NVME_LBA_RANGE | NVME_DNR;
148
}
149
150
@@ -XXX,XX +XXX,XX @@ static uint16_t nvme_io_cmd(NvmeCtrl *n, NvmeCmd *cmd, NvmeRequest *req)
151
NvmeNamespace *ns;
152
uint32_t nsid = le32_to_cpu(cmd->nsid);
153
154
- if (nsid == 0 || nsid > n->num_namespaces) {
155
+ if (unlikely(nsid == 0 || nsid > n->num_namespaces)) {
156
+ trace_nvme_err_invalid_ns(nsid, n->num_namespaces);
157
return NVME_INVALID_NSID | NVME_DNR;
158
}
159
160
@@ -XXX,XX +XXX,XX @@ static uint16_t nvme_io_cmd(NvmeCtrl *n, NvmeCmd *cmd, NvmeRequest *req)
161
case NVME_CMD_READ:
162
return nvme_rw(n, ns, cmd, req);
163
default:
164
+ trace_nvme_err_invalid_opc(cmd->opcode);
165
return NVME_INVALID_OPCODE | NVME_DNR;
166
}
167
}
168
@@ -XXX,XX +XXX,XX @@ static uint16_t nvme_del_sq(NvmeCtrl *n, NvmeCmd *cmd)
169
NvmeCQueue *cq;
170
uint16_t qid = le16_to_cpu(c->qid);
171
172
- if (!qid || nvme_check_sqid(n, qid)) {
173
+ if (unlikely(!qid || nvme_check_sqid(n, qid))) {
174
+ trace_nvme_err_invalid_del_sq(qid);
175
return NVME_INVALID_QID | NVME_DNR;
176
}
177
178
+ trace_nvme_del_sq(qid);
179
+
180
sq = n->sq[qid];
181
while (!QTAILQ_EMPTY(&sq->out_req_list)) {
182
req = QTAILQ_FIRST(&sq->out_req_list);
183
@@ -XXX,XX +XXX,XX @@ static uint16_t nvme_create_sq(NvmeCtrl *n, NvmeCmd *cmd)
184
uint16_t qflags = le16_to_cpu(c->sq_flags);
185
uint64_t prp1 = le64_to_cpu(c->prp1);
186
187
- if (!cqid || nvme_check_cqid(n, cqid)) {
188
+ trace_nvme_create_sq(prp1, sqid, cqid, qsize, qflags);
189
+
190
+ if (unlikely(!cqid || nvme_check_cqid(n, cqid))) {
191
+ trace_nvme_err_invalid_create_sq_cqid(cqid);
192
return NVME_INVALID_CQID | NVME_DNR;
193
}
194
- if (!sqid || !nvme_check_sqid(n, sqid)) {
195
+ if (unlikely(!sqid || !nvme_check_sqid(n, sqid))) {
196
+ trace_nvme_err_invalid_create_sq_sqid(sqid);
197
return NVME_INVALID_QID | NVME_DNR;
198
}
199
- if (!qsize || qsize > NVME_CAP_MQES(n->bar.cap)) {
200
+ if (unlikely(!qsize || qsize > NVME_CAP_MQES(n->bar.cap))) {
201
+ trace_nvme_err_invalid_create_sq_size(qsize);
202
return NVME_MAX_QSIZE_EXCEEDED | NVME_DNR;
203
}
204
- if (!prp1 || prp1 & (n->page_size - 1)) {
205
+ if (unlikely(!prp1 || prp1 & (n->page_size - 1))) {
206
+ trace_nvme_err_invalid_create_sq_addr(prp1);
207
return NVME_INVALID_FIELD | NVME_DNR;
208
}
209
- if (!(NVME_SQ_FLAGS_PC(qflags))) {
210
+ if (unlikely(!(NVME_SQ_FLAGS_PC(qflags)))) {
211
+ trace_nvme_err_invalid_create_sq_qflags(NVME_SQ_FLAGS_PC(qflags));
212
return NVME_INVALID_FIELD | NVME_DNR;
213
}
214
sq = g_malloc0(sizeof(*sq));
215
@@ -XXX,XX +XXX,XX @@ static uint16_t nvme_del_cq(NvmeCtrl *n, NvmeCmd *cmd)
216
NvmeCQueue *cq;
217
uint16_t qid = le16_to_cpu(c->qid);
218
219
- if (!qid || nvme_check_cqid(n, qid)) {
220
+ if (unlikely(!qid || nvme_check_cqid(n, qid))) {
221
+ trace_nvme_err_invalid_del_cq_cqid(qid);
222
return NVME_INVALID_CQID | NVME_DNR;
223
}
224
225
cq = n->cq[qid];
226
- if (!QTAILQ_EMPTY(&cq->sq_list)) {
227
+ if (unlikely(!QTAILQ_EMPTY(&cq->sq_list))) {
228
+ trace_nvme_err_invalid_del_cq_notempty(qid);
229
return NVME_INVALID_QUEUE_DEL;
230
}
231
+ trace_nvme_del_cq(qid);
232
nvme_free_cq(cq, n);
233
return NVME_SUCCESS;
234
}
235
@@ -XXX,XX +XXX,XX @@ static uint16_t nvme_create_cq(NvmeCtrl *n, NvmeCmd *cmd)
236
uint16_t qflags = le16_to_cpu(c->cq_flags);
237
uint64_t prp1 = le64_to_cpu(c->prp1);
238
239
- if (!cqid || !nvme_check_cqid(n, cqid)) {
240
+ trace_nvme_create_cq(prp1, cqid, vector, qsize, qflags,
241
+ NVME_CQ_FLAGS_IEN(qflags) != 0);
242
+
243
+ if (unlikely(!cqid || !nvme_check_cqid(n, cqid))) {
244
+ trace_nvme_err_invalid_create_cq_cqid(cqid);
245
return NVME_INVALID_CQID | NVME_DNR;
246
}
247
- if (!qsize || qsize > NVME_CAP_MQES(n->bar.cap)) {
248
+ if (unlikely(!qsize || qsize > NVME_CAP_MQES(n->bar.cap))) {
249
+ trace_nvme_err_invalid_create_cq_size(qsize);
250
return NVME_MAX_QSIZE_EXCEEDED | NVME_DNR;
251
}
252
- if (!prp1) {
253
+ if (unlikely(!prp1)) {
254
+ trace_nvme_err_invalid_create_cq_addr(prp1);
255
return NVME_INVALID_FIELD | NVME_DNR;
256
}
257
- if (vector > n->num_queues) {
258
+ if (unlikely(vector > n->num_queues)) {
259
+ trace_nvme_err_invalid_create_cq_vector(vector);
260
return NVME_INVALID_IRQ_VECTOR | NVME_DNR;
261
}
262
- if (!(NVME_CQ_FLAGS_PC(qflags))) {
263
+ if (unlikely(!(NVME_CQ_FLAGS_PC(qflags)))) {
264
+ trace_nvme_err_invalid_create_cq_qflags(NVME_CQ_FLAGS_PC(qflags));
265
return NVME_INVALID_FIELD | NVME_DNR;
266
}
267
268
@@ -XXX,XX +XXX,XX @@ static uint16_t nvme_identify_ctrl(NvmeCtrl *n, NvmeIdentify *c)
269
uint64_t prp1 = le64_to_cpu(c->prp1);
270
uint64_t prp2 = le64_to_cpu(c->prp2);
271
272
+ trace_nvme_identify_ctrl();
273
+
274
return nvme_dma_read_prp(n, (uint8_t *)&n->id_ctrl, sizeof(n->id_ctrl),
275
prp1, prp2);
276
}
277
@@ -XXX,XX +XXX,XX @@ static uint16_t nvme_identify_ns(NvmeCtrl *n, NvmeIdentify *c)
278
uint64_t prp1 = le64_to_cpu(c->prp1);
279
uint64_t prp2 = le64_to_cpu(c->prp2);
280
281
- if (nsid == 0 || nsid > n->num_namespaces) {
282
+ trace_nvme_identify_ns(nsid);
283
+
284
+ if (unlikely(nsid == 0 || nsid > n->num_namespaces)) {
285
+ trace_nvme_err_invalid_ns(nsid, n->num_namespaces);
286
return NVME_INVALID_NSID | NVME_DNR;
287
}
288
289
ns = &n->namespaces[nsid - 1];
290
+
291
return nvme_dma_read_prp(n, (uint8_t *)&ns->id_ns, sizeof(ns->id_ns),
292
prp1, prp2);
293
}
294
@@ -XXX,XX +XXX,XX @@ static uint16_t nvme_identify_nslist(NvmeCtrl *n, NvmeIdentify *c)
295
uint16_t ret;
296
int i, j = 0;
297
298
+ trace_nvme_identify_nslist(min_nsid);
299
+
300
list = g_malloc0(data_len);
301
for (i = 0; i < n->num_namespaces; i++) {
302
if (i < min_nsid) {
303
@@ -XXX,XX +XXX,XX @@ static uint16_t nvme_identify(NvmeCtrl *n, NvmeCmd *cmd)
304
case 0x02:
305
return nvme_identify_nslist(n, c);
306
default:
307
+ trace_nvme_err_invalid_identify_cns(le32_to_cpu(c->cns));
308
return NVME_INVALID_FIELD | NVME_DNR;
309
}
310
}
311
@@ -XXX,XX +XXX,XX @@ static uint16_t nvme_get_feature(NvmeCtrl *n, NvmeCmd *cmd, NvmeRequest *req)
312
switch (dw10) {
313
case NVME_VOLATILE_WRITE_CACHE:
314
result = blk_enable_write_cache(n->conf.blk);
315
+ trace_nvme_getfeat_vwcache(result ? "enabled" : "disabled");
316
break;
317
case NVME_NUMBER_OF_QUEUES:
318
result = cpu_to_le32((n->num_queues - 2) | ((n->num_queues - 2) << 16));
319
+ trace_nvme_getfeat_numq(result);
320
break;
321
default:
322
+ trace_nvme_err_invalid_getfeat(dw10);
323
return NVME_INVALID_FIELD | NVME_DNR;
324
}
325
326
@@ -XXX,XX +XXX,XX @@ static uint16_t nvme_set_feature(NvmeCtrl *n, NvmeCmd *cmd, NvmeRequest *req)
327
blk_set_enable_write_cache(n->conf.blk, dw11 & 1);
328
break;
329
case NVME_NUMBER_OF_QUEUES:
330
+ trace_nvme_setfeat_numq((dw11 & 0xFFFF) + 1,
331
+ ((dw11 >> 16) & 0xFFFF) + 1,
332
+ n->num_queues - 1, n->num_queues - 1);
333
req->cqe.result =
334
cpu_to_le32((n->num_queues - 2) | ((n->num_queues - 2) << 16));
335
break;
336
default:
337
+ trace_nvme_err_invalid_setfeat(dw10);
338
return NVME_INVALID_FIELD | NVME_DNR;
339
}
340
return NVME_SUCCESS;
341
@@ -XXX,XX +XXX,XX @@ static uint16_t nvme_admin_cmd(NvmeCtrl *n, NvmeCmd *cmd, NvmeRequest *req)
342
case NVME_ADM_CMD_GET_FEATURES:
343
return nvme_get_feature(n, cmd, req);
344
default:
345
+ trace_nvme_err_invalid_admin_opc(cmd->opcode);
346
return NVME_INVALID_OPCODE | NVME_DNR;
347
}
348
}
349
@@ -XXX,XX +XXX,XX @@ static int nvme_start_ctrl(NvmeCtrl *n)
350
uint32_t page_bits = NVME_CC_MPS(n->bar.cc) + 12;
351
uint32_t page_size = 1 << page_bits;
352
353
- if (n->cq[0] || n->sq[0] || !n->bar.asq || !n->bar.acq ||
354
- n->bar.asq & (page_size - 1) || n->bar.acq & (page_size - 1) ||
355
- NVME_CC_MPS(n->bar.cc) < NVME_CAP_MPSMIN(n->bar.cap) ||
356
- NVME_CC_MPS(n->bar.cc) > NVME_CAP_MPSMAX(n->bar.cap) ||
357
- NVME_CC_IOCQES(n->bar.cc) < NVME_CTRL_CQES_MIN(n->id_ctrl.cqes) ||
358
- NVME_CC_IOCQES(n->bar.cc) > NVME_CTRL_CQES_MAX(n->id_ctrl.cqes) ||
359
- NVME_CC_IOSQES(n->bar.cc) < NVME_CTRL_SQES_MIN(n->id_ctrl.sqes) ||
360
- NVME_CC_IOSQES(n->bar.cc) > NVME_CTRL_SQES_MAX(n->id_ctrl.sqes) ||
361
- !NVME_AQA_ASQS(n->bar.aqa) || !NVME_AQA_ACQS(n->bar.aqa)) {
362
+ if (unlikely(n->cq[0])) {
363
+ trace_nvme_err_startfail_cq();
364
+ return -1;
365
+ }
366
+ if (unlikely(n->sq[0])) {
367
+ trace_nvme_err_startfail_sq();
368
+ return -1;
369
+ }
370
+ if (unlikely(!n->bar.asq)) {
371
+ trace_nvme_err_startfail_nbarasq();
372
+ return -1;
373
+ }
374
+ if (unlikely(!n->bar.acq)) {
375
+ trace_nvme_err_startfail_nbaracq();
376
+ return -1;
377
+ }
378
+ if (unlikely(n->bar.asq & (page_size - 1))) {
379
+ trace_nvme_err_startfail_asq_misaligned(n->bar.asq);
380
+ return -1;
381
+ }
382
+ if (unlikely(n->bar.acq & (page_size - 1))) {
383
+ trace_nvme_err_startfail_acq_misaligned(n->bar.acq);
384
+ return -1;
385
+ }
386
+ if (unlikely(NVME_CC_MPS(n->bar.cc) <
387
+ NVME_CAP_MPSMIN(n->bar.cap))) {
388
+ trace_nvme_err_startfail_page_too_small(
389
+ NVME_CC_MPS(n->bar.cc),
390
+ NVME_CAP_MPSMIN(n->bar.cap));
391
+ return -1;
392
+ }
393
+ if (unlikely(NVME_CC_MPS(n->bar.cc) >
394
+ NVME_CAP_MPSMAX(n->bar.cap))) {
395
+ trace_nvme_err_startfail_page_too_large(
396
+ NVME_CC_MPS(n->bar.cc),
397
+ NVME_CAP_MPSMAX(n->bar.cap));
398
+ return -1;
399
+ }
400
+ if (unlikely(NVME_CC_IOCQES(n->bar.cc) <
401
+ NVME_CTRL_CQES_MIN(n->id_ctrl.cqes))) {
402
+ trace_nvme_err_startfail_cqent_too_small(
403
+ NVME_CC_IOCQES(n->bar.cc),
404
+ NVME_CTRL_CQES_MIN(n->bar.cap));
405
+ return -1;
406
+ }
407
+ if (unlikely(NVME_CC_IOCQES(n->bar.cc) >
408
+ NVME_CTRL_CQES_MAX(n->id_ctrl.cqes))) {
409
+ trace_nvme_err_startfail_cqent_too_large(
410
+ NVME_CC_IOCQES(n->bar.cc),
411
+ NVME_CTRL_CQES_MAX(n->bar.cap));
412
+ return -1;
413
+ }
414
+ if (unlikely(NVME_CC_IOSQES(n->bar.cc) <
415
+ NVME_CTRL_SQES_MIN(n->id_ctrl.sqes))) {
416
+ trace_nvme_err_startfail_sqent_too_small(
417
+ NVME_CC_IOSQES(n->bar.cc),
418
+ NVME_CTRL_SQES_MIN(n->bar.cap));
419
+ return -1;
420
+ }
421
+ if (unlikely(NVME_CC_IOSQES(n->bar.cc) >
422
+ NVME_CTRL_SQES_MAX(n->id_ctrl.sqes))) {
423
+ trace_nvme_err_startfail_sqent_too_large(
424
+ NVME_CC_IOSQES(n->bar.cc),
425
+ NVME_CTRL_SQES_MAX(n->bar.cap));
426
+ return -1;
427
+ }
428
+ if (unlikely(!NVME_AQA_ASQS(n->bar.aqa))) {
429
+ trace_nvme_err_startfail_asqent_sz_zero();
430
+ return -1;
431
+ }
432
+ if (unlikely(!NVME_AQA_ACQS(n->bar.aqa))) {
433
+ trace_nvme_err_startfail_acqent_sz_zero();
434
return -1;
435
}
436
437
@@ -XXX,XX +XXX,XX @@ static int nvme_start_ctrl(NvmeCtrl *n)
438
static void nvme_write_bar(NvmeCtrl *n, hwaddr offset, uint64_t data,
439
unsigned size)
440
{
441
+ if (unlikely(offset & (sizeof(uint32_t) - 1))) {
442
+ NVME_GUEST_ERR(nvme_ub_mmiowr_misaligned32,
443
+ "MMIO write not 32-bit aligned,"
444
+ " offset=0x%"PRIx64"", offset);
445
+ /* should be ignored, fall through for now */
446
+ }
447
+
448
+ if (unlikely(size < sizeof(uint32_t))) {
449
+ NVME_GUEST_ERR(nvme_ub_mmiowr_toosmall,
450
+ "MMIO write smaller than 32-bits,"
451
+ " offset=0x%"PRIx64", size=%u",
452
+ offset, size);
453
+ /* should be ignored, fall through for now */
454
+ }
455
+
456
switch (offset) {
457
- case 0xc:
458
+ case 0xc: /* INTMS */
459
+ if (unlikely(msix_enabled(&(n->parent_obj)))) {
460
+ NVME_GUEST_ERR(nvme_ub_mmiowr_intmask_with_msix,
461
+ "undefined access to interrupt mask set"
462
+ " when MSI-X is enabled");
463
+ /* should be ignored, fall through for now */
464
+ }
465
n->bar.intms |= data & 0xffffffff;
466
n->bar.intmc = n->bar.intms;
467
+ trace_nvme_mmio_intm_set(data & 0xffffffff,
468
+ n->bar.intmc);
469
break;
470
- case 0x10:
471
+ case 0x10: /* INTMC */
472
+ if (unlikely(msix_enabled(&(n->parent_obj)))) {
473
+ NVME_GUEST_ERR(nvme_ub_mmiowr_intmask_with_msix,
474
+ "undefined access to interrupt mask clr"
475
+ " when MSI-X is enabled");
476
+ /* should be ignored, fall through for now */
477
+ }
478
n->bar.intms &= ~(data & 0xffffffff);
479
n->bar.intmc = n->bar.intms;
480
+ trace_nvme_mmio_intm_clr(data & 0xffffffff,
481
+ n->bar.intmc);
482
break;
483
- case 0x14:
484
+ case 0x14: /* CC */
485
+ trace_nvme_mmio_cfg(data & 0xffffffff);
486
/* Windows first sends data, then sends enable bit */
487
if (!NVME_CC_EN(data) && !NVME_CC_EN(n->bar.cc) &&
488
!NVME_CC_SHN(data) && !NVME_CC_SHN(n->bar.cc))
489
@@ -XXX,XX +XXX,XX @@ static void nvme_write_bar(NvmeCtrl *n, hwaddr offset, uint64_t data,
490
491
if (NVME_CC_EN(data) && !NVME_CC_EN(n->bar.cc)) {
492
n->bar.cc = data;
493
- if (nvme_start_ctrl(n)) {
494
+ if (unlikely(nvme_start_ctrl(n))) {
495
+ trace_nvme_err_startfail();
496
n->bar.csts = NVME_CSTS_FAILED;
497
} else {
498
+ trace_nvme_mmio_start_success();
499
n->bar.csts = NVME_CSTS_READY;
500
}
501
} else if (!NVME_CC_EN(data) && NVME_CC_EN(n->bar.cc)) {
502
+ trace_nvme_mmio_stopped();
503
nvme_clear_ctrl(n);
504
n->bar.csts &= ~NVME_CSTS_READY;
505
}
506
if (NVME_CC_SHN(data) && !(NVME_CC_SHN(n->bar.cc))) {
507
- nvme_clear_ctrl(n);
508
- n->bar.cc = data;
509
- n->bar.csts |= NVME_CSTS_SHST_COMPLETE;
510
+ trace_nvme_mmio_shutdown_set();
511
+ nvme_clear_ctrl(n);
512
+ n->bar.cc = data;
513
+ n->bar.csts |= NVME_CSTS_SHST_COMPLETE;
514
} else if (!NVME_CC_SHN(data) && NVME_CC_SHN(n->bar.cc)) {
515
- n->bar.csts &= ~NVME_CSTS_SHST_COMPLETE;
516
- n->bar.cc = data;
517
+ trace_nvme_mmio_shutdown_cleared();
518
+ n->bar.csts &= ~NVME_CSTS_SHST_COMPLETE;
519
+ n->bar.cc = data;
520
+ }
521
+ break;
522
+ case 0x1C: /* CSTS */
523
+ if (data & (1 << 4)) {
524
+ NVME_GUEST_ERR(nvme_ub_mmiowr_ssreset_w1c_unsupported,
525
+ "attempted to W1C CSTS.NSSRO"
526
+ " but CAP.NSSRS is zero (not supported)");
527
+ } else if (data != 0) {
528
+ NVME_GUEST_ERR(nvme_ub_mmiowr_ro_csts,
529
+ "attempted to set a read only bit"
530
+ " of controller status");
531
+ }
532
+ break;
533
+ case 0x20: /* NSSR */
534
+ if (data == 0x4E564D65) {
535
+ trace_nvme_ub_mmiowr_ssreset_unsupported();
536
+ } else {
537
+ /* The spec says that writes of other values have no effect */
538
+ return;
539
}
540
break;
541
- case 0x24:
542
+ case 0x24: /* AQA */
543
n->bar.aqa = data & 0xffffffff;
544
+ trace_nvme_mmio_aqattr(data & 0xffffffff);
545
break;
546
- case 0x28:
547
+ case 0x28: /* ASQ */
548
n->bar.asq = data;
549
+ trace_nvme_mmio_asqaddr(data);
550
break;
551
- case 0x2c:
552
+ case 0x2c: /* ASQ hi */
553
n->bar.asq |= data << 32;
554
+ trace_nvme_mmio_asqaddr_hi(data, n->bar.asq);
555
break;
556
- case 0x30:
557
+ case 0x30: /* ACQ */
558
+ trace_nvme_mmio_acqaddr(data);
559
n->bar.acq = data;
560
break;
561
- case 0x34:
562
+ case 0x34: /* ACQ hi */
563
n->bar.acq |= data << 32;
564
+ trace_nvme_mmio_acqaddr_hi(data, n->bar.acq);
565
break;
566
+ case 0x38: /* CMBLOC */
567
+ NVME_GUEST_ERR(nvme_ub_mmiowr_cmbloc_reserved,
568
+ "invalid write to reserved CMBLOC"
569
+ " when CMBSZ is zero, ignored");
570
+ return;
571
+ case 0x3C: /* CMBSZ */
572
+ NVME_GUEST_ERR(nvme_ub_mmiowr_cmbsz_readonly,
573
+ "invalid write to read only CMBSZ, ignored");
574
+ return;
575
default:
576
+ NVME_GUEST_ERR(nvme_ub_mmiowr_invalid,
577
+ "invalid MMIO write,"
578
+ " offset=0x%"PRIx64", data=%"PRIx64"",
579
+ offset, data);
580
break;
581
}
582
}
583
@@ -XXX,XX +XXX,XX @@ static uint64_t nvme_mmio_read(void *opaque, hwaddr addr, unsigned size)
584
uint8_t *ptr = (uint8_t *)&n->bar;
585
uint64_t val = 0;
586
587
+ if (unlikely(addr & (sizeof(uint32_t) - 1))) {
588
+ NVME_GUEST_ERR(nvme_ub_mmiord_misaligned32,
589
+ "MMIO read not 32-bit aligned,"
590
+ " offset=0x%"PRIx64"", addr);
591
+ /* should RAZ, fall through for now */
592
+ } else if (unlikely(size < sizeof(uint32_t))) {
593
+ NVME_GUEST_ERR(nvme_ub_mmiord_toosmall,
594
+ "MMIO read smaller than 32-bits,"
595
+ " offset=0x%"PRIx64"", addr);
596
+ /* should RAZ, fall through for now */
597
+ }
598
+
599
if (addr < sizeof(n->bar)) {
600
memcpy(&val, ptr + addr, size);
601
+ } else {
602
+ NVME_GUEST_ERR(nvme_ub_mmiord_invalid_ofs,
603
+ "MMIO read beyond last register,"
604
+ " offset=0x%"PRIx64", returning 0", addr);
605
}
606
+
607
return val;
608
}
609
610
@@ -XXX,XX +XXX,XX @@ static void nvme_process_db(NvmeCtrl *n, hwaddr addr, int val)
611
{
612
uint32_t qid;
613
614
- if (addr & ((1 << 2) - 1)) {
615
+ if (unlikely(addr & ((1 << 2) - 1))) {
616
+ NVME_GUEST_ERR(nvme_ub_db_wr_misaligned,
617
+ "doorbell write not 32-bit aligned,"
618
+ " offset=0x%"PRIx64", ignoring", addr);
619
return;
620
}
621
622
if (((addr - 0x1000) >> 2) & 1) {
623
+ /* Completion queue doorbell write */
624
+
625
uint16_t new_head = val & 0xffff;
626
int start_sqs;
627
NvmeCQueue *cq;
628
629
qid = (addr - (0x1000 + (1 << 2))) >> 3;
630
- if (nvme_check_cqid(n, qid)) {
631
+ if (unlikely(nvme_check_cqid(n, qid))) {
632
+ NVME_GUEST_ERR(nvme_ub_db_wr_invalid_cq,
633
+ "completion queue doorbell write"
634
+ " for nonexistent queue,"
635
+ " sqid=%"PRIu32", ignoring", qid);
636
return;
637
}
638
639
cq = n->cq[qid];
640
- if (new_head >= cq->size) {
641
+ if (unlikely(new_head >= cq->size)) {
642
+ NVME_GUEST_ERR(nvme_ub_db_wr_invalid_cqhead,
643
+ "completion queue doorbell write value"
644
+ " beyond queue size, sqid=%"PRIu32","
645
+ " new_head=%"PRIu16", ignoring",
646
+ qid, new_head);
647
return;
648
}
649
650
@@ -XXX,XX +XXX,XX @@ static void nvme_process_db(NvmeCtrl *n, hwaddr addr, int val)
651
nvme_isr_notify(n, cq);
652
}
653
} else {
654
+ /* Submission queue doorbell write */
655
+
656
uint16_t new_tail = val & 0xffff;
657
NvmeSQueue *sq;
658
659
qid = (addr - 0x1000) >> 3;
660
- if (nvme_check_sqid(n, qid)) {
661
+ if (unlikely(nvme_check_sqid(n, qid))) {
662
+ NVME_GUEST_ERR(nvme_ub_db_wr_invalid_sq,
663
+ "submission queue doorbell write"
664
+ " for nonexistent queue,"
665
+ " sqid=%"PRIu32", ignoring", qid);
666
return;
667
}
668
669
sq = n->sq[qid];
670
- if (new_tail >= sq->size) {
671
+ if (unlikely(new_tail >= sq->size)) {
672
+ NVME_GUEST_ERR(nvme_ub_db_wr_invalid_sqtail,
673
+ "submission queue doorbell write value"
674
+ " beyond queue size, sqid=%"PRIu32","
675
+ " new_tail=%"PRIu16", ignoring",
676
+ qid, new_tail);
677
return;
678
}
679
680
diff --git a/hw/block/trace-events b/hw/block/trace-events
681
index XXXXXXX..XXXXXXX 100644
682
--- a/hw/block/trace-events
683
+++ b/hw/block/trace-events
684
@@ -XXX,XX +XXX,XX @@ virtio_blk_submit_multireq(void *vdev, void *mrb, int start, int num_reqs, uint6
685
hd_geometry_lchs_guess(void *blk, int cyls, int heads, int secs) "blk %p LCHS %d %d %d"
686
hd_geometry_guess(void *blk, uint32_t cyls, uint32_t heads, uint32_t secs, int trans) "blk %p CHS %u %u %u trans %d"
687
688
+# hw/block/nvme.c
689
+# nvme traces for successful events
690
+nvme_irq_msix(uint32_t vector) "raising MSI-X IRQ vector %u"
691
+nvme_irq_pin(void) "pulsing IRQ pin"
692
+nvme_irq_masked(void) "IRQ is masked"
693
+nvme_dma_read(uint64_t prp1, uint64_t prp2) "DMA read, prp1=0x%"PRIx64" prp2=0x%"PRIx64""
694
+nvme_rw(char const *verb, uint32_t blk_count, uint64_t byte_count, uint64_t lba) "%s %"PRIu32" blocks (%"PRIu64" bytes) from LBA %"PRIu64""
695
+nvme_create_sq(uint64_t addr, uint16_t sqid, uint16_t cqid, uint16_t qsize, uint16_t qflags) "create submission queue, addr=0x%"PRIx64", sqid=%"PRIu16", cqid=%"PRIu16", qsize=%"PRIu16", qflags=%"PRIu16""
696
+nvme_create_cq(uint64_t addr, uint16_t cqid, uint16_t vector, uint16_t size, uint16_t qflags, int ien) "create completion queue, addr=0x%"PRIx64", cqid=%"PRIu16", vector=%"PRIu16", qsize=%"PRIu16", qflags=%"PRIu16", ien=%d"
697
+nvme_del_sq(uint16_t qid) "deleting submission queue sqid=%"PRIu16""
698
+nvme_del_cq(uint16_t cqid) "deleted completion queue, sqid=%"PRIu16""
699
+nvme_identify_ctrl(void) "identify controller"
700
+nvme_identify_ns(uint16_t ns) "identify namespace, nsid=%"PRIu16""
701
+nvme_identify_nslist(uint16_t ns) "identify namespace list, nsid=%"PRIu16""
702
+nvme_getfeat_vwcache(char const* result) "get feature volatile write cache, result=%s"
703
+nvme_getfeat_numq(int result) "get feature number of queues, result=%d"
704
+nvme_setfeat_numq(int reqcq, int reqsq, int gotcq, int gotsq) "requested cq_count=%d sq_count=%d, responding with cq_count=%d sq_count=%d"
705
+nvme_mmio_intm_set(uint64_t data, uint64_t new_mask) "wrote MMIO, interrupt mask set, data=0x%"PRIx64", new_mask=0x%"PRIx64""
706
+nvme_mmio_intm_clr(uint64_t data, uint64_t new_mask) "wrote MMIO, interrupt mask clr, data=0x%"PRIx64", new_mask=0x%"PRIx64""
707
+nvme_mmio_cfg(uint64_t data) "wrote MMIO, config controller config=0x%"PRIx64""
708
+nvme_mmio_aqattr(uint64_t data) "wrote MMIO, admin queue attributes=0x%"PRIx64""
709
+nvme_mmio_asqaddr(uint64_t data) "wrote MMIO, admin submission queue address=0x%"PRIx64""
710
+nvme_mmio_acqaddr(uint64_t data) "wrote MMIO, admin completion queue address=0x%"PRIx64""
711
+nvme_mmio_asqaddr_hi(uint64_t data, uint64_t new_addr) "wrote MMIO, admin submission queue high half=0x%"PRIx64", new_address=0x%"PRIx64""
712
+nvme_mmio_acqaddr_hi(uint64_t data, uint64_t new_addr) "wrote MMIO, admin completion queue high half=0x%"PRIx64", new_address=0x%"PRIx64""
713
+nvme_mmio_start_success(void) "setting controller enable bit succeeded"
714
+nvme_mmio_stopped(void) "cleared controller enable bit"
715
+nvme_mmio_shutdown_set(void) "shutdown bit set"
716
+nvme_mmio_shutdown_cleared(void) "shutdown bit cleared"
717
+
718
+# nvme traces for error conditions
719
+nvme_err_invalid_dma(void) "PRP/SGL is too small for transfer size"
720
+nvme_err_invalid_prplist_ent(uint64_t prplist) "PRP list entry is null or not page aligned: 0x%"PRIx64""
721
+nvme_err_invalid_prp2_align(uint64_t prp2) "PRP2 is not page aligned: 0x%"PRIx64""
722
+nvme_err_invalid_prp2_missing(void) "PRP2 is null and more data to be transferred"
723
+nvme_err_invalid_field(void) "invalid field"
724
+nvme_err_invalid_prp(void) "invalid PRP"
725
+nvme_err_invalid_sgl(void) "invalid SGL"
726
+nvme_err_invalid_ns(uint32_t ns, uint32_t limit) "invalid namespace %u not within 1-%u"
727
+nvme_err_invalid_opc(uint8_t opc) "invalid opcode 0x%"PRIx8""
728
+nvme_err_invalid_admin_opc(uint8_t opc) "invalid admin opcode 0x%"PRIx8""
729
+nvme_err_invalid_lba_range(uint64_t start, uint64_t len, uint64_t limit) "Invalid LBA start=%"PRIu64" len=%"PRIu64" limit=%"PRIu64""
730
+nvme_err_invalid_del_sq(uint16_t qid) "invalid submission queue deletion, sid=%"PRIu16""
731
+nvme_err_invalid_create_sq_cqid(uint16_t cqid) "failed creating submission queue, invalid cqid=%"PRIu16""
732
+nvme_err_invalid_create_sq_sqid(uint16_t sqid) "failed creating submission queue, invalid sqid=%"PRIu16""
733
+nvme_err_invalid_create_sq_size(uint16_t qsize) "failed creating submission queue, invalid qsize=%"PRIu16""
734
+nvme_err_invalid_create_sq_addr(uint64_t addr) "failed creating submission queue, addr=0x%"PRIx64""
735
+nvme_err_invalid_create_sq_qflags(uint16_t qflags) "failed creating submission queue, qflags=%"PRIu16""
736
+nvme_err_invalid_del_cq_cqid(uint16_t cqid) "failed deleting completion queue, cqid=%"PRIu16""
737
+nvme_err_invalid_del_cq_notempty(uint16_t cqid) "failed deleting completion queue, it is not empty, cqid=%"PRIu16""
738
+nvme_err_invalid_create_cq_cqid(uint16_t cqid) "failed creating completion queue, cqid=%"PRIu16""
739
+nvme_err_invalid_create_cq_size(uint16_t size) "failed creating completion queue, size=%"PRIu16""
740
+nvme_err_invalid_create_cq_addr(uint64_t addr) "failed creating completion queue, addr=0x%"PRIx64""
741
+nvme_err_invalid_create_cq_vector(uint16_t vector) "failed creating completion queue, vector=%"PRIu16""
742
+nvme_err_invalid_create_cq_qflags(uint16_t qflags) "failed creating completion queue, qflags=%"PRIu16""
743
+nvme_err_invalid_identify_cns(uint16_t cns) "identify, invalid cns=0x%"PRIx16""
744
+nvme_err_invalid_getfeat(int dw10) "invalid get features, dw10=0x%"PRIx32""
745
+nvme_err_invalid_setfeat(uint32_t dw10) "invalid set features, dw10=0x%"PRIx32""
746
+nvme_err_startfail_cq(void) "nvme_start_ctrl failed because there are non-admin completion queues"
747
+nvme_err_startfail_sq(void) "nvme_start_ctrl failed because there are non-admin submission queues"
748
+nvme_err_startfail_nbarasq(void) "nvme_start_ctrl failed because the admin submission queue address is null"
749
+nvme_err_startfail_nbaracq(void) "nvme_start_ctrl failed because the admin completion queue address is null"
750
+nvme_err_startfail_asq_misaligned(uint64_t addr) "nvme_start_ctrl failed because the admin submission queue address is misaligned: 0x%"PRIx64""
751
+nvme_err_startfail_acq_misaligned(uint64_t addr) "nvme_start_ctrl failed because the admin completion queue address is misaligned: 0x%"PRIx64""
752
+nvme_err_startfail_page_too_small(uint8_t log2ps, uint8_t maxlog2ps) "nvme_start_ctrl failed because the page size is too small: log2size=%u, min=%u"
753
+nvme_err_startfail_page_too_large(uint8_t log2ps, uint8_t maxlog2ps) "nvme_start_ctrl failed because the page size is too large: log2size=%u, max=%u"
754
+nvme_err_startfail_cqent_too_small(uint8_t log2ps, uint8_t maxlog2ps) "nvme_start_ctrl failed because the completion queue entry size is too small: log2size=%u, min=%u"
755
+nvme_err_startfail_cqent_too_large(uint8_t log2ps, uint8_t maxlog2ps) "nvme_start_ctrl failed because the completion queue entry size is too large: log2size=%u, max=%u"
756
+nvme_err_startfail_sqent_too_small(uint8_t log2ps, uint8_t maxlog2ps) "nvme_start_ctrl failed because the submission queue entry size is too small: log2size=%u, min=%u"
757
+nvme_err_startfail_sqent_too_large(uint8_t log2ps, uint8_t maxlog2ps) "nvme_start_ctrl failed because the submission queue entry size is too large: log2size=%u, max=%u"
758
+nvme_err_startfail_asqent_sz_zero(void) "nvme_start_ctrl failed because the admin submission queue size is zero"
759
+nvme_err_startfail_acqent_sz_zero(void) "nvme_start_ctrl failed because the admin completion queue size is zero"
760
+nvme_err_startfail(void) "setting controller enable bit failed"
761
+
762
+# Traces for undefined behavior
763
+nvme_ub_mmiowr_misaligned32(uint64_t offset) "MMIO write not 32-bit aligned, offset=0x%"PRIx64""
764
+nvme_ub_mmiowr_toosmall(uint64_t offset, unsigned size) "MMIO write smaller than 32 bits, offset=0x%"PRIx64", size=%u"
765
+nvme_ub_mmiowr_intmask_with_msix(void) "undefined access to interrupt mask set when MSI-X is enabled"
766
+nvme_ub_mmiowr_ro_csts(void) "attempted to set a read only bit of controller status"
767
+nvme_ub_mmiowr_ssreset_w1c_unsupported(void) "attempted to W1C CSTS.NSSRO but CAP.NSSRS is zero (not supported)"
768
+nvme_ub_mmiowr_ssreset_unsupported(void) "attempted NVM subsystem reset but CAP.NSSRS is zero (not supported)"
769
+nvme_ub_mmiowr_cmbloc_reserved(void) "invalid write to reserved CMBLOC when CMBSZ is zero, ignored"
770
+nvme_ub_mmiowr_cmbsz_readonly(void) "invalid write to read only CMBSZ, ignored"
771
+nvme_ub_mmiowr_invalid(uint64_t offset, uint64_t data) "invalid MMIO write, offset=0x%"PRIx64", data=0x%"PRIx64""
772
+nvme_ub_mmiord_misaligned32(uint64_t offset) "MMIO read not 32-bit aligned, offset=0x%"PRIx64""
773
+nvme_ub_mmiord_toosmall(uint64_t offset) "MMIO read smaller than 32-bits, offset=0x%"PRIx64""
774
+nvme_ub_mmiord_invalid_ofs(uint64_t offset) "MMIO read beyond last register, offset=0x%"PRIx64", returning 0"
775
+nvme_ub_db_wr_misaligned(uint64_t offset) "doorbell write not 32-bit aligned, offset=0x%"PRIx64", ignoring"
776
+nvme_ub_db_wr_invalid_cq(uint32_t qid) "completion queue doorbell write for nonexistent queue, cqid=%"PRIu32", ignoring"
777
+nvme_ub_db_wr_invalid_cqhead(uint32_t qid, uint16_t new_head) "completion queue doorbell write value beyond queue size, cqid=%"PRIu32", new_head=%"PRIu16", ignoring"
778
+nvme_ub_db_wr_invalid_sq(uint32_t qid) "submission queue doorbell write for nonexistent queue, sqid=%"PRIu32", ignoring"
779
+nvme_ub_db_wr_invalid_sqtail(uint32_t qid, uint16_t new_tail) "submission queue doorbell write value beyond queue size, sqid=%"PRIu32", new_head=%"PRIu16", ignoring"
780
+
781
# hw/block/xen_disk.c
782
xen_disk_alloc(char *name) "%s"
783
xen_disk_init(char *name) "%s"
784
--
785
2.13.6
786
787
diff view generated by jsdifflib
New patch
1
From: Fam Zheng <famz@redhat.com>
1
2
3
Management tools create overlays of running guests with qemu-img:
4
5
$ qemu-img create -b /image/in/use.qcow2 -f qcow2 /overlay/image.qcow2
6
7
but this doesn't work anymore due to image locking:
8
9
qemu-img: /overlay/image.qcow2: Failed to get shared "write" lock
10
Is another process using the image?
11
Could not open backing image to determine size.
12
Use the force share option to allow this use case again.
13
14
Cc: qemu-stable@nongnu.org
15
Signed-off-by: Fam Zheng <famz@redhat.com>
16
Reviewed-by: Eric Blake <eblake@redhat.com>
17
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
18
---
19
block.c | 3 ++-
20
1 file changed, 2 insertions(+), 1 deletion(-)
21
22
diff --git a/block.c b/block.c
23
index XXXXXXX..XXXXXXX 100644
24
--- a/block.c
25
+++ b/block.c
26
@@ -XXX,XX +XXX,XX @@ void bdrv_img_create(const char *filename, const char *fmt,
27
back_flags = flags;
28
back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
29
30
+ backing_options = qdict_new();
31
if (backing_fmt) {
32
- backing_options = qdict_new();
33
qdict_put_str(backing_options, "driver", backing_fmt);
34
}
35
+ qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true);
36
37
bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
38
&local_err);
39
--
40
2.13.6
41
42
diff view generated by jsdifflib
1
From: Peter Maydell <peter.maydell@linaro.org>
1
From: Thomas Huth <thuth@redhat.com>
2
2
3
In block/iscsi.c we use a raw malloc() call, which is unusual
3
It's not working anymore since QEMU v1.3.0 - time to remove it now.
4
given the project standard is to use the glib memory allocation
5
functions. Document why we do so, to avoid it being converted
6
to g_malloc() by mistake.
7
4
8
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
5
Signed-off-by: Thomas Huth <thuth@redhat.com>
9
Message-ID: <20230727150705.2664464-1-peter.maydell@linaro.org>
6
Reviewed-by: John Snow <jsnow@redhat.com>
10
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
7
Reviewed-by: Markus Armbruster <armbru@redhat.com>
11
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
8
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
12
---
9
---
13
block/iscsi.c | 1 +
10
blockdev.c | 11 -----------
14
1 file changed, 1 insertion(+)
11
qemu-doc.texi | 6 ------
12
2 files changed, 17 deletions(-)
15
13
16
diff --git a/block/iscsi.c b/block/iscsi.c
14
diff --git a/blockdev.c b/blockdev.c
17
index XXXXXXX..XXXXXXX 100644
15
index XXXXXXX..XXXXXXX 100644
18
--- a/block/iscsi.c
16
--- a/blockdev.c
19
+++ b/block/iscsi.c
17
+++ b/blockdev.c
20
@@ -XXX,XX +XXX,XX @@ static BlockAIOCB *iscsi_aio_ioctl(BlockDriverState *bs,
18
@@ -XXX,XX +XXX,XX @@ QemuOptsList qemu_legacy_drive_opts = {
21
return NULL;
19
.type = QEMU_OPT_STRING,
20
.help = "chs translation (auto, lba, none)",
21
},{
22
- .name = "boot",
23
- .type = QEMU_OPT_BOOL,
24
- .help = "(deprecated, ignored)",
25
- },{
26
.name = "addr",
27
.type = QEMU_OPT_STRING,
28
.help = "pci address (virtio only)",
29
@@ -XXX,XX +XXX,XX @@ DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type)
30
goto fail;
22
}
31
}
23
32
24
+ /* Must use malloc(): this is freed via scsi_free_scsi_task() */
33
- /* Deprecated option boot=[on|off] */
25
acb->task = malloc(sizeof(struct scsi_task));
34
- if (qemu_opt_get(legacy_opts, "boot") != NULL) {
26
if (acb->task == NULL) {
35
- fprintf(stderr, "qemu-kvm: boot=on|off is deprecated and will be "
27
error_report("iSCSI: Failed to allocate task for scsi command. %s",
36
- "ignored. Future versions will reject this parameter. Please "
37
- "update your scripts.\n");
38
- }
39
-
40
/* Other deprecated options */
41
if (!qtest_enabled()) {
42
for (i = 0; i < ARRAY_SIZE(deprecated); i++) {
43
diff --git a/qemu-doc.texi b/qemu-doc.texi
44
index XXXXXXX..XXXXXXX 100644
45
--- a/qemu-doc.texi
46
+++ b/qemu-doc.texi
47
@@ -XXX,XX +XXX,XX @@ deprecated.
48
49
@section System emulator command line arguments
50
51
-@subsection -drive boot=on|off (since 1.3.0)
52
-
53
-The ``boot=on|off'' option to the ``-drive'' argument is
54
-ignored. Applications should use the ``bootindex=N'' parameter
55
-to set an absolute ordering between devices instead.
56
-
57
@subsection -tdf (since 1.3.0)
58
59
The ``-tdf'' argument is ignored. The behaviour implemented
28
--
60
--
29
2.41.0
61
2.13.6
62
63
diff view generated by jsdifflib
New patch
1
1
From: Thomas Huth <thuth@redhat.com>
2
3
It's been marked as deprecated since QEMU v2.10.0, and so far nobody
4
complained that we should keep it, so let's remove this legacy option
5
now to simplify the code quite a bit.
6
7
Signed-off-by: Thomas Huth <thuth@redhat.com>
8
Reviewed-by: John Snow <jsnow@redhat.com>
9
Reviewed-by: Markus Armbruster <armbru@redhat.com>
10
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
11
---
12
vl.c | 86 ++-------------------------------------------------------
13
qemu-doc.texi | 8 ------
14
qemu-options.hx | 19 ++-----------
15
3 files changed, 4 insertions(+), 109 deletions(-)
16
17
diff --git a/vl.c b/vl.c
18
index XXXXXXX..XXXXXXX 100644
19
--- a/vl.c
20
+++ b/vl.c
21
@@ -XXX,XX +XXX,XX @@ int main(int argc, char **argv, char **envp)
22
const char *boot_order = NULL;
23
const char *boot_once = NULL;
24
DisplayState *ds;
25
- int cyls, heads, secs, translation;
26
QemuOpts *opts, *machine_opts;
27
- QemuOpts *hda_opts = NULL, *icount_opts = NULL, *accel_opts = NULL;
28
+ QemuOpts *icount_opts = NULL, *accel_opts = NULL;
29
QemuOptsList *olist;
30
int optind;
31
const char *optarg;
32
@@ -XXX,XX +XXX,XX @@ int main(int argc, char **argv, char **envp)
33
34
cpu_model = NULL;
35
snapshot = 0;
36
- cyls = heads = secs = 0;
37
- translation = BIOS_ATA_TRANSLATION_AUTO;
38
39
nb_nics = 0;
40
41
@@ -XXX,XX +XXX,XX @@ int main(int argc, char **argv, char **envp)
42
if (optind >= argc)
43
break;
44
if (argv[optind][0] != '-') {
45
- hda_opts = drive_add(IF_DEFAULT, 0, argv[optind++], HD_OPTS);
46
+ drive_add(IF_DEFAULT, 0, argv[optind++], HD_OPTS);
47
} else {
48
const QEMUOption *popt;
49
50
@@ -XXX,XX +XXX,XX @@ int main(int argc, char **argv, char **envp)
51
cpu_model = optarg;
52
break;
53
case QEMU_OPTION_hda:
54
- {
55
- char buf[256];
56
- if (cyls == 0)
57
- snprintf(buf, sizeof(buf), "%s", HD_OPTS);
58
- else
59
- snprintf(buf, sizeof(buf),
60
- "%s,cyls=%d,heads=%d,secs=%d%s",
61
- HD_OPTS , cyls, heads, secs,
62
- translation == BIOS_ATA_TRANSLATION_LBA ?
63
- ",trans=lba" :
64
- translation == BIOS_ATA_TRANSLATION_NONE ?
65
- ",trans=none" : "");
66
- drive_add(IF_DEFAULT, 0, optarg, buf);
67
- break;
68
- }
69
case QEMU_OPTION_hdb:
70
case QEMU_OPTION_hdc:
71
case QEMU_OPTION_hdd:
72
@@ -XXX,XX +XXX,XX @@ int main(int argc, char **argv, char **envp)
73
case QEMU_OPTION_snapshot:
74
snapshot = 1;
75
break;
76
- case QEMU_OPTION_hdachs:
77
- {
78
- const char *p;
79
- p = optarg;
80
- cyls = strtol(p, (char **)&p, 0);
81
- if (cyls < 1 || cyls > 16383)
82
- goto chs_fail;
83
- if (*p != ',')
84
- goto chs_fail;
85
- p++;
86
- heads = strtol(p, (char **)&p, 0);
87
- if (heads < 1 || heads > 16)
88
- goto chs_fail;
89
- if (*p != ',')
90
- goto chs_fail;
91
- p++;
92
- secs = strtol(p, (char **)&p, 0);
93
- if (secs < 1 || secs > 63)
94
- goto chs_fail;
95
- if (*p == ',') {
96
- p++;
97
- if (!strcmp(p, "large")) {
98
- translation = BIOS_ATA_TRANSLATION_LARGE;
99
- } else if (!strcmp(p, "rechs")) {
100
- translation = BIOS_ATA_TRANSLATION_RECHS;
101
- } else if (!strcmp(p, "none")) {
102
- translation = BIOS_ATA_TRANSLATION_NONE;
103
- } else if (!strcmp(p, "lba")) {
104
- translation = BIOS_ATA_TRANSLATION_LBA;
105
- } else if (!strcmp(p, "auto")) {
106
- translation = BIOS_ATA_TRANSLATION_AUTO;
107
- } else {
108
- goto chs_fail;
109
- }
110
- } else if (*p != '\0') {
111
- chs_fail:
112
- error_report("invalid physical CHS format");
113
- exit(1);
114
- }
115
- if (hda_opts != NULL) {
116
- qemu_opt_set_number(hda_opts, "cyls", cyls,
117
- &error_abort);
118
- qemu_opt_set_number(hda_opts, "heads", heads,
119
- &error_abort);
120
- qemu_opt_set_number(hda_opts, "secs", secs,
121
- &error_abort);
122
- if (translation == BIOS_ATA_TRANSLATION_LARGE) {
123
- qemu_opt_set(hda_opts, "trans", "large",
124
- &error_abort);
125
- } else if (translation == BIOS_ATA_TRANSLATION_RECHS) {
126
- qemu_opt_set(hda_opts, "trans", "rechs",
127
- &error_abort);
128
- } else if (translation == BIOS_ATA_TRANSLATION_LBA) {
129
- qemu_opt_set(hda_opts, "trans", "lba",
130
- &error_abort);
131
- } else if (translation == BIOS_ATA_TRANSLATION_NONE) {
132
- qemu_opt_set(hda_opts, "trans", "none",
133
- &error_abort);
134
- }
135
- }
136
- }
137
- error_report("'-hdachs' is deprecated, please use '-device"
138
- " ide-hd,cyls=c,heads=h,secs=s,...' instead");
139
- break;
140
case QEMU_OPTION_numa:
141
opts = qemu_opts_parse_noisily(qemu_find_opts("numa"),
142
optarg, true);
143
diff --git a/qemu-doc.texi b/qemu-doc.texi
144
index XXXXXXX..XXXXXXX 100644
145
--- a/qemu-doc.texi
146
+++ b/qemu-doc.texi
147
@@ -XXX,XX +XXX,XX @@ The ``--net dump'' argument is now replaced with the
148
``-object filter-dump'' argument which works in combination
149
with the modern ``-netdev`` backends instead.
150
151
-@subsection -hdachs (since 2.10.0)
152
-
153
-The ``-hdachs'' argument is now a synonym for setting
154
-the ``cyls'', ``heads'', ``secs'', and ``trans'' properties
155
-on the ``ide-hd'' device using the ``-device'' argument.
156
-The new syntax allows different settings to be provided
157
-per disk.
158
-
159
@subsection -usbdevice (since 2.10.0)
160
161
The ``-usbdevice DEV'' argument is now a synonym for setting
162
diff --git a/qemu-options.hx b/qemu-options.hx
163
index XXXXXXX..XXXXXXX 100644
164
--- a/qemu-options.hx
165
+++ b/qemu-options.hx
166
@@ -XXX,XX +XXX,XX @@ of available connectors of a given interface type.
167
@item media=@var{media}
168
This option defines the type of the media: disk or cdrom.
169
@item cyls=@var{c},heads=@var{h},secs=@var{s}[,trans=@var{t}]
170
-These options have the same definition as they have in @option{-hdachs}.
171
-These parameters are deprecated, use the corresponding parameters
172
+Force disk physical geometry and the optional BIOS translation (trans=none or
173
+lba). These parameters are deprecated, use the corresponding parameters
174
of @code{-device} instead.
175
@item snapshot=@var{snapshot}
176
@var{snapshot} is "on" or "off" and controls snapshot mode for the given drive
177
@@ -XXX,XX +XXX,XX @@ the raw disk image you use is not written back. You can however force
178
the write back by pressing @key{C-a s} (@pxref{disk_images}).
179
ETEXI
180
181
-DEF("hdachs", HAS_ARG, QEMU_OPTION_hdachs, \
182
- "-hdachs c,h,s[,t]\n" \
183
- " force hard disk 0 physical geometry and the optional BIOS\n" \
184
- " translation (t=none or lba) (usually QEMU can guess them)\n",
185
- QEMU_ARCH_ALL)
186
-STEXI
187
-@item -hdachs @var{c},@var{h},@var{s},[,@var{t}]
188
-@findex -hdachs
189
-Force hard disk 0 physical geometry (1 <= @var{c} <= 16383, 1 <=
190
-@var{h} <= 16, 1 <= @var{s} <= 63) and optionally force the BIOS
191
-translation mode (@var{t}=none, lba or auto). Usually QEMU can guess
192
-all those parameters. This option is deprecated, please use
193
-@code{-device ide-hd,cyls=c,heads=h,secs=s,...} instead.
194
-ETEXI
195
-
196
DEF("fsdev", HAS_ARG, QEMU_OPTION_fsdev,
197
"-fsdev fsdriver,id=id[,path=path,][security_model={mapped-xattr|mapped-file|passthrough|none}]\n"
198
" [,writeout=immediate][,readonly][,socket=socket|sock_fd=sock_fd][,fmode=fmode][,dmode=dmode]\n"
199
--
200
2.13.6
201
202
diff view generated by jsdifflib
New patch
1
From: Thomas Huth <thuth@redhat.com>
1
2
3
Looks like we forgot to announce the deprecation of these options in
4
the corresponding chapter of the qemu-doc text, so let's do that now.
5
6
Signed-off-by: Thomas Huth <thuth@redhat.com>
7
Reviewed-by: John Snow <jsnow@redhat.com>
8
Reviewed-by: Markus Armbruster <armbru@redhat.com>
9
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
10
---
11
qemu-doc.texi | 15 +++++++++++++++
12
1 file changed, 15 insertions(+)
13
14
diff --git a/qemu-doc.texi b/qemu-doc.texi
15
index XXXXXXX..XXXXXXX 100644
16
--- a/qemu-doc.texi
17
+++ b/qemu-doc.texi
18
@@ -XXX,XX +XXX,XX @@ longer be directly supported in QEMU.
19
The ``-drive if=scsi'' argument is replaced by the the
20
``-device BUS-TYPE'' argument combined with ``-drive if=none''.
21
22
+@subsection -drive cyls=...,heads=...,secs=...,trans=... (since 2.10.0)
23
+
24
+The drive geometry arguments are replaced by the the geometry arguments
25
+that can be specified with the ``-device'' parameter.
26
+
27
+@subsection -drive serial=... (since 2.10.0)
28
+
29
+The drive serial argument is replaced by the the serial argument
30
+that can be specified with the ``-device'' parameter.
31
+
32
+@subsection -drive addr=... (since 2.10.0)
33
+
34
+The drive addr argument is replaced by the the addr argument
35
+that can be specified with the ``-device'' parameter.
36
+
37
@subsection -net dump (since 2.10.0)
38
39
The ``--net dump'' argument is now replaced with the
40
--
41
2.13.6
42
43
diff view generated by jsdifflib
1
From: Fabiano Rosas <farosas@suse.de>
1
From: Fam Zheng <famz@redhat.com>
2
2
3
The last call site of this function has been removed by commit
3
Signed-off-by: Fam Zheng <famz@redhat.com>
4
c04d0ab026 ("qemu-img: Let info print block graph").
5
6
Reviewed-by: Claudio Fontana <cfontana@suse.de>
7
Signed-off-by: Fabiano Rosas <farosas@suse.de>
8
Message-ID: <20230901184605.32260-2-farosas@suse.de>
9
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
10
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
4
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
11
---
5
---
12
include/block/qapi.h | 3 ---
6
include/block/block_int.h | 1 -
13
block/qapi.c | 27 ---------------------------
7
block/io.c | 18 ------------------
14
2 files changed, 30 deletions(-)
8
2 files changed, 19 deletions(-)
15
9
16
diff --git a/include/block/qapi.h b/include/block/qapi.h
10
diff --git a/include/block/block_int.h b/include/block/block_int.h
17
index XXXXXXX..XXXXXXX 100644
11
index XXXXXXX..XXXXXXX 100644
18
--- a/include/block/qapi.h
12
--- a/include/block/block_int.h
19
+++ b/include/block/qapi.h
13
+++ b/include/block/block_int.h
20
@@ -XXX,XX +XXX,XX @@ BlockDeviceInfo *bdrv_block_device_info(BlockBackend *blk,
14
@@ -XXX,XX +XXX,XX @@ bool blk_dev_is_tray_open(BlockBackend *blk);
21
int bdrv_query_snapshot_info_list(BlockDriverState *bs,
15
bool blk_dev_is_medium_locked(BlockBackend *blk);
22
SnapshotInfoList **p_list,
16
23
Error **errp);
17
void bdrv_set_dirty(BlockDriverState *bs, int64_t offset, int64_t bytes);
24
-void bdrv_query_block_node_info(BlockDriverState *bs,
18
-bool bdrv_requests_pending(BlockDriverState *bs);
25
- BlockNodeInfo **p_info,
19
26
- Error **errp);
20
void bdrv_clear_dirty_bitmap(BdrvDirtyBitmap *bitmap, HBitmap **out);
27
void bdrv_query_image_info(BlockDriverState *bs,
21
void bdrv_undo_clear_dirty_bitmap(BdrvDirtyBitmap *bitmap, HBitmap *in);
28
ImageInfo **p_info,
22
diff --git a/block/io.c b/block/io.c
29
bool flat,
30
diff --git a/block/qapi.c b/block/qapi.c
31
index XXXXXXX..XXXXXXX 100644
23
index XXXXXXX..XXXXXXX 100644
32
--- a/block/qapi.c
24
--- a/block/io.c
33
+++ b/block/qapi.c
25
+++ b/block/io.c
34
@@ -XXX,XX +XXX,XX @@ out:
26
@@ -XXX,XX +XXX,XX @@ void bdrv_disable_copy_on_read(BlockDriverState *bs)
35
aio_context_release(bdrv_get_aio_context(bs));
27
assert(old >= 1);
36
}
28
}
37
29
38
-/**
30
-/* Check if any requests are in-flight (including throttled requests) */
39
- * bdrv_query_block_node_info:
31
-bool bdrv_requests_pending(BlockDriverState *bs)
40
- * @bs: block node to examine
41
- * @p_info: location to store node information
42
- * @errp: location to store error information
43
- *
44
- * Store image information about @bs in @p_info.
45
- *
46
- * @p_info will be set only on success. On error, store error in @errp.
47
- */
48
-void bdrv_query_block_node_info(BlockDriverState *bs,
49
- BlockNodeInfo **p_info,
50
- Error **errp)
51
-{
32
-{
52
- BlockNodeInfo *info;
33
- BdrvChild *child;
53
- ERRP_GUARD();
54
-
34
-
55
- info = g_new0(BlockNodeInfo, 1);
35
- if (atomic_read(&bs->in_flight)) {
56
- bdrv_do_query_node_info(bs, info, errp);
36
- return true;
57
- if (*errp) {
58
- qapi_free_BlockNodeInfo(info);
59
- return;
60
- }
37
- }
61
-
38
-
62
- *p_info = info;
39
- QLIST_FOREACH(child, &bs->children, next) {
40
- if (bdrv_requests_pending(child->bs)) {
41
- return true;
42
- }
43
- }
44
-
45
- return false;
63
-}
46
-}
64
-
47
-
65
/**
48
typedef struct {
66
* bdrv_query_image_info:
49
Coroutine *co;
67
* @bs: block node to examine
50
BlockDriverState *bs;
68
--
51
--
69
2.41.0
52
2.13.6
53
54
diff view generated by jsdifflib
New patch
1
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2
Reviewed-by: Fam Zheng <famz@redhat.com>
3
---
4
block/io.c | 6 ++++++
5
1 file changed, 6 insertions(+)
1
6
7
diff --git a/block/io.c b/block/io.c
8
index XXXXXXX..XXXXXXX 100644
9
--- a/block/io.c
10
+++ b/block/io.c
11
@@ -XXX,XX +XXX,XX @@ void bdrv_drain_all_begin(void)
12
BdrvNextIterator it;
13
GSList *aio_ctxs = NULL, *ctx;
14
15
+ /* BDRV_POLL_WHILE() for a node can only be called from its own I/O thread
16
+ * or the main loop AioContext. We potentially use BDRV_POLL_WHILE() on
17
+ * nodes in several different AioContexts, so make sure we're in the main
18
+ * context. */
19
+ assert(qemu_get_current_aio_context() == qemu_get_aio_context());
20
+
21
block_job_pause_all();
22
23
for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
24
--
25
2.13.6
26
27
diff view generated by jsdifflib
New patch
1
bdrv_drained_begin() doesn't increase bs->quiesce_counter recursively
2
and also doesn't notify other parent nodes of children, which both means
3
that the child nodes are not actually drained, and bdrv_drained_begin()
4
is providing useful functionality only on a single node.
1
5
6
To keep things consistent, we also shouldn't call the block driver
7
callbacks recursively.
8
9
A proper recursive drain version that provides an actually working
10
drained section for child nodes will be introduced later.
11
12
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
13
Reviewed-by: Fam Zheng <famz@redhat.com>
14
---
15
block/io.c | 16 +++++++++-------
16
1 file changed, 9 insertions(+), 7 deletions(-)
17
18
diff --git a/block/io.c b/block/io.c
19
index XXXXXXX..XXXXXXX 100644
20
--- a/block/io.c
21
+++ b/block/io.c
22
@@ -XXX,XX +XXX,XX @@ static void coroutine_fn bdrv_drain_invoke_entry(void *opaque)
23
}
24
25
/* Recursively call BlockDriver.bdrv_co_drain_begin/end callbacks */
26
-static void bdrv_drain_invoke(BlockDriverState *bs, bool begin)
27
+static void bdrv_drain_invoke(BlockDriverState *bs, bool begin, bool recursive)
28
{
29
BdrvChild *child, *tmp;
30
BdrvCoDrainData data = { .bs = bs, .done = false, .begin = begin};
31
@@ -XXX,XX +XXX,XX @@ static void bdrv_drain_invoke(BlockDriverState *bs, bool begin)
32
bdrv_coroutine_enter(bs, data.co);
33
BDRV_POLL_WHILE(bs, !data.done);
34
35
- QLIST_FOREACH_SAFE(child, &bs->children, next, tmp) {
36
- bdrv_drain_invoke(child->bs, begin);
37
+ if (recursive) {
38
+ QLIST_FOREACH_SAFE(child, &bs->children, next, tmp) {
39
+ bdrv_drain_invoke(child->bs, begin, true);
40
+ }
41
}
42
}
43
44
@@ -XXX,XX +XXX,XX @@ void bdrv_drained_begin(BlockDriverState *bs)
45
bdrv_parent_drained_begin(bs);
46
}
47
48
- bdrv_drain_invoke(bs, true);
49
+ bdrv_drain_invoke(bs, true, false);
50
bdrv_drain_recurse(bs);
51
}
52
53
@@ -XXX,XX +XXX,XX @@ void bdrv_drained_end(BlockDriverState *bs)
54
}
55
56
/* Re-enable things in child-to-parent order */
57
- bdrv_drain_invoke(bs, false);
58
+ bdrv_drain_invoke(bs, false, false);
59
bdrv_parent_drained_end(bs);
60
aio_enable_external(bdrv_get_aio_context(bs));
61
}
62
@@ -XXX,XX +XXX,XX @@ void bdrv_drain_all_begin(void)
63
aio_context_acquire(aio_context);
64
aio_disable_external(aio_context);
65
bdrv_parent_drained_begin(bs);
66
- bdrv_drain_invoke(bs, true);
67
+ bdrv_drain_invoke(bs, true, true);
68
aio_context_release(aio_context);
69
70
if (!g_slist_find(aio_ctxs, aio_context)) {
71
@@ -XXX,XX +XXX,XX @@ void bdrv_drain_all_end(void)
72
73
/* Re-enable things in child-to-parent order */
74
aio_context_acquire(aio_context);
75
- bdrv_drain_invoke(bs, false);
76
+ bdrv_drain_invoke(bs, false, true);
77
bdrv_parent_drained_end(bs);
78
aio_enable_external(aio_context);
79
aio_context_release(aio_context);
80
--
81
2.13.6
82
83
diff view generated by jsdifflib
New patch
1
The existing test is for bdrv_drain_all_begin/end() only. Generalise the
2
test case so that it can be run for the other variants as well. At the
3
moment this is only bdrv_drain_begin/end(), but in a while, we'll add
4
another one.
1
5
6
Also, add a backing file to the test node to test whether the operations
7
work recursively.
8
9
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
10
---
11
tests/test-bdrv-drain.c | 69 ++++++++++++++++++++++++++++++++++++++++++++-----
12
1 file changed, 62 insertions(+), 7 deletions(-)
13
14
diff --git a/tests/test-bdrv-drain.c b/tests/test-bdrv-drain.c
15
index XXXXXXX..XXXXXXX 100644
16
--- a/tests/test-bdrv-drain.c
17
+++ b/tests/test-bdrv-drain.c
18
@@ -XXX,XX +XXX,XX @@ static BlockDriver bdrv_test = {
19
20
.bdrv_co_drain_begin = bdrv_test_co_drain_begin,
21
.bdrv_co_drain_end = bdrv_test_co_drain_end,
22
+
23
+ .bdrv_child_perm = bdrv_format_default_perms,
24
};
25
26
static void aio_ret_cb(void *opaque, int ret)
27
@@ -XXX,XX +XXX,XX @@ static void aio_ret_cb(void *opaque, int ret)
28
*aio_ret = ret;
29
}
30
31
-static void test_drv_cb_drain_all(void)
32
+enum drain_type {
33
+ BDRV_DRAIN_ALL,
34
+ BDRV_DRAIN,
35
+};
36
+
37
+static void do_drain_begin(enum drain_type drain_type, BlockDriverState *bs)
38
+{
39
+ switch (drain_type) {
40
+ case BDRV_DRAIN_ALL: bdrv_drain_all_begin(); break;
41
+ case BDRV_DRAIN: bdrv_drained_begin(bs); break;
42
+ default: g_assert_not_reached();
43
+ }
44
+}
45
+
46
+static void do_drain_end(enum drain_type drain_type, BlockDriverState *bs)
47
+{
48
+ switch (drain_type) {
49
+ case BDRV_DRAIN_ALL: bdrv_drain_all_end(); break;
50
+ case BDRV_DRAIN: bdrv_drained_end(bs); break;
51
+ default: g_assert_not_reached();
52
+ }
53
+}
54
+
55
+static void test_drv_cb_common(enum drain_type drain_type, bool recursive)
56
{
57
BlockBackend *blk;
58
- BlockDriverState *bs;
59
- BDRVTestState *s;
60
+ BlockDriverState *bs, *backing;
61
+ BDRVTestState *s, *backing_s;
62
BlockAIOCB *acb;
63
int aio_ret;
64
65
@@ -XXX,XX +XXX,XX @@ static void test_drv_cb_drain_all(void)
66
s = bs->opaque;
67
blk_insert_bs(blk, bs, &error_abort);
68
69
+ backing = bdrv_new_open_driver(&bdrv_test, "backing", 0, &error_abort);
70
+ backing_s = backing->opaque;
71
+ bdrv_set_backing_hd(bs, backing, &error_abort);
72
+
73
/* Simple bdrv_drain_all_begin/end pair, check that CBs are called */
74
g_assert_cmpint(s->drain_count, ==, 0);
75
- bdrv_drain_all_begin();
76
+ g_assert_cmpint(backing_s->drain_count, ==, 0);
77
+
78
+ do_drain_begin(drain_type, bs);
79
+
80
g_assert_cmpint(s->drain_count, ==, 1);
81
- bdrv_drain_all_end();
82
+ g_assert_cmpint(backing_s->drain_count, ==, !!recursive);
83
+
84
+ do_drain_end(drain_type, bs);
85
+
86
g_assert_cmpint(s->drain_count, ==, 0);
87
+ g_assert_cmpint(backing_s->drain_count, ==, 0);
88
89
/* Now do the same while a request is pending */
90
aio_ret = -EINPROGRESS;
91
@@ -XXX,XX +XXX,XX @@ static void test_drv_cb_drain_all(void)
92
g_assert_cmpint(aio_ret, ==, -EINPROGRESS);
93
94
g_assert_cmpint(s->drain_count, ==, 0);
95
- bdrv_drain_all_begin();
96
+ g_assert_cmpint(backing_s->drain_count, ==, 0);
97
+
98
+ do_drain_begin(drain_type, bs);
99
+
100
g_assert_cmpint(aio_ret, ==, 0);
101
g_assert_cmpint(s->drain_count, ==, 1);
102
- bdrv_drain_all_end();
103
+ g_assert_cmpint(backing_s->drain_count, ==, !!recursive);
104
+
105
+ do_drain_end(drain_type, bs);
106
+
107
g_assert_cmpint(s->drain_count, ==, 0);
108
+ g_assert_cmpint(backing_s->drain_count, ==, 0);
109
110
+ bdrv_unref(backing);
111
bdrv_unref(bs);
112
blk_unref(blk);
113
}
114
115
+static void test_drv_cb_drain_all(void)
116
+{
117
+ test_drv_cb_common(BDRV_DRAIN_ALL, true);
118
+}
119
+
120
+static void test_drv_cb_drain(void)
121
+{
122
+ test_drv_cb_common(BDRV_DRAIN, false);
123
+}
124
+
125
int main(int argc, char **argv)
126
{
127
bdrv_init();
128
@@ -XXX,XX +XXX,XX @@ int main(int argc, char **argv)
129
g_test_init(&argc, &argv, NULL);
130
131
g_test_add_func("/bdrv-drain/driver-cb/drain_all", test_drv_cb_drain_all);
132
+ g_test_add_func("/bdrv-drain/driver-cb/drain", test_drv_cb_drain);
133
134
return g_test_run();
135
}
136
--
137
2.13.6
138
139
diff view generated by jsdifflib
New patch
1
This is currently only working correctly for bdrv_drain(), not for
2
bdrv_drain_all(). Leave a comment for the drain_all case, we'll address
3
it later.
1
4
5
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
6
---
7
tests/test-bdrv-drain.c | 45 +++++++++++++++++++++++++++++++++++++++++++++
8
1 file changed, 45 insertions(+)
9
10
diff --git a/tests/test-bdrv-drain.c b/tests/test-bdrv-drain.c
11
index XXXXXXX..XXXXXXX 100644
12
--- a/tests/test-bdrv-drain.c
13
+++ b/tests/test-bdrv-drain.c
14
@@ -XXX,XX +XXX,XX @@ static void test_drv_cb_drain(void)
15
test_drv_cb_common(BDRV_DRAIN, false);
16
}
17
18
+static void test_quiesce_common(enum drain_type drain_type, bool recursive)
19
+{
20
+ BlockBackend *blk;
21
+ BlockDriverState *bs, *backing;
22
+
23
+ blk = blk_new(BLK_PERM_ALL, BLK_PERM_ALL);
24
+ bs = bdrv_new_open_driver(&bdrv_test, "test-node", BDRV_O_RDWR,
25
+ &error_abort);
26
+ blk_insert_bs(blk, bs, &error_abort);
27
+
28
+ backing = bdrv_new_open_driver(&bdrv_test, "backing", 0, &error_abort);
29
+ bdrv_set_backing_hd(bs, backing, &error_abort);
30
+
31
+ g_assert_cmpint(bs->quiesce_counter, ==, 0);
32
+ g_assert_cmpint(backing->quiesce_counter, ==, 0);
33
+
34
+ do_drain_begin(drain_type, bs);
35
+
36
+ g_assert_cmpint(bs->quiesce_counter, ==, 1);
37
+ g_assert_cmpint(backing->quiesce_counter, ==, !!recursive);
38
+
39
+ do_drain_end(drain_type, bs);
40
+
41
+ g_assert_cmpint(bs->quiesce_counter, ==, 0);
42
+ g_assert_cmpint(backing->quiesce_counter, ==, 0);
43
+
44
+ bdrv_unref(backing);
45
+ bdrv_unref(bs);
46
+ blk_unref(blk);
47
+}
48
+
49
+static void test_quiesce_drain_all(void)
50
+{
51
+ // XXX drain_all doesn't quiesce
52
+ //test_quiesce_common(BDRV_DRAIN_ALL, true);
53
+}
54
+
55
+static void test_quiesce_drain(void)
56
+{
57
+ test_quiesce_common(BDRV_DRAIN, false);
58
+}
59
+
60
int main(int argc, char **argv)
61
{
62
bdrv_init();
63
@@ -XXX,XX +XXX,XX @@ int main(int argc, char **argv)
64
g_test_add_func("/bdrv-drain/driver-cb/drain_all", test_drv_cb_drain_all);
65
g_test_add_func("/bdrv-drain/driver-cb/drain", test_drv_cb_drain);
66
67
+ g_test_add_func("/bdrv-drain/quiesce/drain_all", test_quiesce_drain_all);
68
+ g_test_add_func("/bdrv-drain/quiesce/drain", test_quiesce_drain);
69
+
70
return g_test_run();
71
}
72
--
73
2.13.6
74
75
diff view generated by jsdifflib
New patch
1
Block jobs already paused themselves when their main BlockBackend
2
entered a drained section. This is not good enough: We also want to
3
pause a block job and may not submit new requests if, for example, the
4
mirror target node should be drained.
1
5
6
This implements .drained_begin/end callbacks in child_job in order to
7
consider all block nodes related to the job, and removes the
8
BlockBackend callbacks which are unnecessary now because the root of the
9
job main BlockBackend is always referenced with a child_job, too.
10
11
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
12
---
13
blockjob.c | 22 +++++++++-------------
14
1 file changed, 9 insertions(+), 13 deletions(-)
15
16
diff --git a/blockjob.c b/blockjob.c
17
index XXXXXXX..XXXXXXX 100644
18
--- a/blockjob.c
19
+++ b/blockjob.c
20
@@ -XXX,XX +XXX,XX @@ static char *child_job_get_parent_desc(BdrvChild *c)
21
job->id);
22
}
23
24
-static const BdrvChildRole child_job = {
25
- .get_parent_desc = child_job_get_parent_desc,
26
- .stay_at_node = true,
27
-};
28
-
29
-static void block_job_drained_begin(void *opaque)
30
+static void child_job_drained_begin(BdrvChild *c)
31
{
32
- BlockJob *job = opaque;
33
+ BlockJob *job = c->opaque;
34
block_job_pause(job);
35
}
36
37
-static void block_job_drained_end(void *opaque)
38
+static void child_job_drained_end(BdrvChild *c)
39
{
40
- BlockJob *job = opaque;
41
+ BlockJob *job = c->opaque;
42
block_job_resume(job);
43
}
44
45
-static const BlockDevOps block_job_dev_ops = {
46
- .drained_begin = block_job_drained_begin,
47
- .drained_end = block_job_drained_end,
48
+static const BdrvChildRole child_job = {
49
+ .get_parent_desc = child_job_get_parent_desc,
50
+ .drained_begin = child_job_drained_begin,
51
+ .drained_end = child_job_drained_end,
52
+ .stay_at_node = true,
53
};
54
55
void block_job_remove_all_bdrv(BlockJob *job)
56
@@ -XXX,XX +XXX,XX @@ void *block_job_create(const char *job_id, const BlockJobDriver *driver,
57
block_job_add_bdrv(job, "main node", bs, 0, BLK_PERM_ALL, &error_abort);
58
bs->job = job;
59
60
- blk_set_dev_ops(blk, &block_job_dev_ops, job);
61
bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker);
62
63
QLIST_INSERT_HEAD(&block_jobs, job, job_list);
64
--
65
2.13.6
66
67
diff view generated by jsdifflib
1
From: Fabiano Rosas <farosas@suse.de>
1
Block jobs must be paused if any of the involved nodes are drained.
2
2
3
The commit 5d8813593f ("block/qapi: Let bdrv_query_image_info()
4
recurse") removed the loop where we set the 'bs0' variable, so now it
5
is just the same as 'bs'.
6
7
Signed-off-by: Fabiano Rosas <farosas@suse.de>
8
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
9
Message-ID: <20230901184605.32260-3-farosas@suse.de>
10
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
11
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
3
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
12
---
4
---
13
block/qapi.c | 5 ++---
5
tests/test-bdrv-drain.c | 121 ++++++++++++++++++++++++++++++++++++++++++++++++
14
1 file changed, 2 insertions(+), 3 deletions(-)
6
1 file changed, 121 insertions(+)
15
7
16
diff --git a/block/qapi.c b/block/qapi.c
8
diff --git a/tests/test-bdrv-drain.c b/tests/test-bdrv-drain.c
17
index XXXXXXX..XXXXXXX 100644
9
index XXXXXXX..XXXXXXX 100644
18
--- a/block/qapi.c
10
--- a/tests/test-bdrv-drain.c
19
+++ b/block/qapi.c
11
+++ b/tests/test-bdrv-drain.c
20
@@ -XXX,XX +XXX,XX @@ BlockDeviceInfo *bdrv_block_device_info(BlockBackend *blk,
12
@@ -XXX,XX +XXX,XX @@
13
14
#include "qemu/osdep.h"
15
#include "block/block.h"
16
+#include "block/blockjob_int.h"
17
#include "sysemu/block-backend.h"
18
#include "qapi/error.h"
19
20
@@ -XXX,XX +XXX,XX @@ static void test_quiesce_drain(void)
21
test_quiesce_common(BDRV_DRAIN, false);
22
}
23
24
+
25
+typedef struct TestBlockJob {
26
+ BlockJob common;
27
+ bool should_complete;
28
+} TestBlockJob;
29
+
30
+static void test_job_completed(BlockJob *job, void *opaque)
31
+{
32
+ block_job_completed(job, 0);
33
+}
34
+
35
+static void coroutine_fn test_job_start(void *opaque)
36
+{
37
+ TestBlockJob *s = opaque;
38
+
39
+ while (!s->should_complete) {
40
+ block_job_sleep_ns(&s->common, 100000);
41
+ }
42
+
43
+ block_job_defer_to_main_loop(&s->common, test_job_completed, NULL);
44
+}
45
+
46
+static void test_job_complete(BlockJob *job, Error **errp)
47
+{
48
+ TestBlockJob *s = container_of(job, TestBlockJob, common);
49
+ s->should_complete = true;
50
+}
51
+
52
+BlockJobDriver test_job_driver = {
53
+ .instance_size = sizeof(TestBlockJob),
54
+ .start = test_job_start,
55
+ .complete = test_job_complete,
56
+};
57
+
58
+static void test_blockjob_common(enum drain_type drain_type)
59
+{
60
+ BlockBackend *blk_src, *blk_target;
61
+ BlockDriverState *src, *target;
62
+ BlockJob *job;
63
+ int ret;
64
+
65
+ src = bdrv_new_open_driver(&bdrv_test, "source", BDRV_O_RDWR,
66
+ &error_abort);
67
+ blk_src = blk_new(BLK_PERM_ALL, BLK_PERM_ALL);
68
+ blk_insert_bs(blk_src, src, &error_abort);
69
+
70
+ target = bdrv_new_open_driver(&bdrv_test, "target", BDRV_O_RDWR,
71
+ &error_abort);
72
+ blk_target = blk_new(BLK_PERM_ALL, BLK_PERM_ALL);
73
+ blk_insert_bs(blk_target, target, &error_abort);
74
+
75
+ job = block_job_create("job0", &test_job_driver, src, 0, BLK_PERM_ALL, 0,
76
+ 0, NULL, NULL, &error_abort);
77
+ block_job_add_bdrv(job, "target", target, 0, BLK_PERM_ALL, &error_abort);
78
+ block_job_start(job);
79
+
80
+ g_assert_cmpint(job->pause_count, ==, 0);
81
+ g_assert_false(job->paused);
82
+ g_assert_false(job->busy); /* We're in block_job_sleep_ns() */
83
+
84
+ do_drain_begin(drain_type, src);
85
+
86
+ if (drain_type == BDRV_DRAIN_ALL) {
87
+ /* bdrv_drain_all() drains both src and target, and involves an
88
+ * additional block_job_pause_all() */
89
+ g_assert_cmpint(job->pause_count, ==, 3);
90
+ } else {
91
+ g_assert_cmpint(job->pause_count, ==, 1);
92
+ }
93
+ /* XXX We don't wait until the job is actually paused. Is this okay? */
94
+ /* g_assert_true(job->paused); */
95
+ g_assert_false(job->busy); /* The job is paused */
96
+
97
+ do_drain_end(drain_type, src);
98
+
99
+ g_assert_cmpint(job->pause_count, ==, 0);
100
+ g_assert_false(job->paused);
101
+ g_assert_false(job->busy); /* We're in block_job_sleep_ns() */
102
+
103
+ do_drain_begin(drain_type, target);
104
+
105
+ if (drain_type == BDRV_DRAIN_ALL) {
106
+ /* bdrv_drain_all() drains both src and target, and involves an
107
+ * additional block_job_pause_all() */
108
+ g_assert_cmpint(job->pause_count, ==, 3);
109
+ } else {
110
+ g_assert_cmpint(job->pause_count, ==, 1);
111
+ }
112
+ /* XXX We don't wait until the job is actually paused. Is this okay? */
113
+ /* g_assert_true(job->paused); */
114
+ g_assert_false(job->busy); /* The job is paused */
115
+
116
+ do_drain_end(drain_type, target);
117
+
118
+ g_assert_cmpint(job->pause_count, ==, 0);
119
+ g_assert_false(job->paused);
120
+ g_assert_false(job->busy); /* We're in block_job_sleep_ns() */
121
+
122
+ ret = block_job_complete_sync(job, &error_abort);
123
+ g_assert_cmpint(ret, ==, 0);
124
+
125
+ blk_unref(blk_src);
126
+ blk_unref(blk_target);
127
+ bdrv_unref(src);
128
+ bdrv_unref(target);
129
+}
130
+
131
+static void test_blockjob_drain_all(void)
132
+{
133
+ test_blockjob_common(BDRV_DRAIN_ALL);
134
+}
135
+
136
+static void test_blockjob_drain(void)
137
+{
138
+ test_blockjob_common(BDRV_DRAIN);
139
+}
140
+
141
int main(int argc, char **argv)
21
{
142
{
22
ImageInfo **p_image_info;
143
bdrv_init();
23
ImageInfo *backing_info;
144
@@ -XXX,XX +XXX,XX @@ int main(int argc, char **argv)
24
- BlockDriverState *bs0, *backing;
145
g_test_add_func("/bdrv-drain/quiesce/drain_all", test_quiesce_drain_all);
25
+ BlockDriverState *backing;
146
g_test_add_func("/bdrv-drain/quiesce/drain", test_quiesce_drain);
26
BlockDeviceInfo *info;
147
27
ERRP_GUARD();
148
+ g_test_add_func("/bdrv-drain/blockjob/drain_all", test_blockjob_drain_all);
28
149
+ g_test_add_func("/bdrv-drain/blockjob/drain", test_blockjob_drain);
29
@@ -XXX,XX +XXX,XX @@ BlockDeviceInfo *bdrv_block_device_info(BlockBackend *blk,
150
+
30
151
return g_test_run();
31
info->write_threshold = bdrv_write_threshold_get(bs);
152
}
32
33
- bs0 = bs;
34
p_image_info = &info->image;
35
info->backing_file_depth = 0;
36
37
@@ -XXX,XX +XXX,XX @@ BlockDeviceInfo *bdrv_block_device_info(BlockBackend *blk,
38
* Skip automatically inserted nodes that the user isn't aware of for
39
* query-block (blk != NULL), but not for query-named-block-nodes
40
*/
41
- bdrv_query_image_info(bs0, p_image_info, flat, blk != NULL, errp);
42
+ bdrv_query_image_info(bs, p_image_info, flat, blk != NULL, errp);
43
if (*errp) {
44
qapi_free_BlockDeviceInfo(info);
45
return NULL;
46
--
153
--
47
2.41.0
154
2.13.6
48
155
49
156
diff view generated by jsdifflib
1
From: Dmitry Frolov <frolov@swemel.ru>
1
Block jobs are already paused using the BdrvChildRole drain callbacks,
2
so we don't need an additional block_job_pause_all() call.
2
3
3
bdrv_open_child() may return NULL.
4
Usually return value is checked for this function.
5
Check for return value is more reliable.
6
7
Fixes: 24bc15d1f6 ("vmdk: Use BdrvChild instead of BDS for references to extents")
8
9
Signed-off-by: Dmitry Frolov <frolov@swemel.ru>
10
Message-ID: <20230831125926.796205-1-frolov@swemel.ru>
11
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
12
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
4
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
13
---
5
---
14
block/vmdk.c | 2 +-
6
block/io.c | 4 ----
15
1 file changed, 1 insertion(+), 1 deletion(-)
7
tests/test-bdrv-drain.c | 10 ++++------
8
2 files changed, 4 insertions(+), 10 deletions(-)
16
9
17
diff --git a/block/vmdk.c b/block/vmdk.c
10
diff --git a/block/io.c b/block/io.c
18
index XXXXXXX..XXXXXXX 100644
11
index XXXXXXX..XXXXXXX 100644
19
--- a/block/vmdk.c
12
--- a/block/io.c
20
+++ b/block/vmdk.c
13
+++ b/block/io.c
21
@@ -XXX,XX +XXX,XX @@ static int vmdk_parse_extents(const char *desc, BlockDriverState *bs,
14
@@ -XXX,XX +XXX,XX @@ void bdrv_drain_all_begin(void)
22
bs, &child_of_bds, extent_role, false,
15
* context. */
23
&local_err);
16
assert(qemu_get_current_aio_context() == qemu_get_aio_context());
24
g_free(extent_path);
17
25
- if (local_err) {
18
- block_job_pause_all();
26
+ if (!extent_file) {
19
-
27
error_propagate(errp, local_err);
20
for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
28
ret = -EINVAL;
21
AioContext *aio_context = bdrv_get_aio_context(bs);
29
goto out;
22
23
@@ -XXX,XX +XXX,XX @@ void bdrv_drain_all_end(void)
24
aio_enable_external(aio_context);
25
aio_context_release(aio_context);
26
}
27
-
28
- block_job_resume_all();
29
}
30
31
void bdrv_drain_all(void)
32
diff --git a/tests/test-bdrv-drain.c b/tests/test-bdrv-drain.c
33
index XXXXXXX..XXXXXXX 100644
34
--- a/tests/test-bdrv-drain.c
35
+++ b/tests/test-bdrv-drain.c
36
@@ -XXX,XX +XXX,XX @@ static void test_blockjob_common(enum drain_type drain_type)
37
do_drain_begin(drain_type, src);
38
39
if (drain_type == BDRV_DRAIN_ALL) {
40
- /* bdrv_drain_all() drains both src and target, and involves an
41
- * additional block_job_pause_all() */
42
- g_assert_cmpint(job->pause_count, ==, 3);
43
+ /* bdrv_drain_all() drains both src and target */
44
+ g_assert_cmpint(job->pause_count, ==, 2);
45
} else {
46
g_assert_cmpint(job->pause_count, ==, 1);
47
}
48
@@ -XXX,XX +XXX,XX @@ static void test_blockjob_common(enum drain_type drain_type)
49
do_drain_begin(drain_type, target);
50
51
if (drain_type == BDRV_DRAIN_ALL) {
52
- /* bdrv_drain_all() drains both src and target, and involves an
53
- * additional block_job_pause_all() */
54
- g_assert_cmpint(job->pause_count, ==, 3);
55
+ /* bdrv_drain_all() drains both src and target */
56
+ g_assert_cmpint(job->pause_count, ==, 2);
57
} else {
58
g_assert_cmpint(job->pause_count, ==, 1);
59
}
30
--
60
--
31
2.41.0
61
2.13.6
62
63
diff view generated by jsdifflib
1
From: Stefan Hajnoczi <stefanha@redhat.com>
1
bdrv_do_drained_begin() restricts the call of parent callbacks and
2
aio_disable_external() to the outermost drain section, but the block
3
driver callbacks are always called. bdrv_do_drained_end() must match
4
this behaviour, otherwise nodes stay drained even if begin/end calls
5
were balanced.
2
6
3
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
4
Message-ID: <20230808155852.2745350-2-stefanha@redhat.com>
5
Reviewed-by: Eric Blake <eblake@redhat.com>
6
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
7
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
7
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
8
---
8
---
9
block/io.c | 8 +++++++-
9
block/io.c | 12 +++++++-----
10
1 file changed, 7 insertions(+), 1 deletion(-)
10
1 file changed, 7 insertions(+), 5 deletions(-)
11
11
12
diff --git a/block/io.c b/block/io.c
12
diff --git a/block/io.c b/block/io.c
13
index XXXXXXX..XXXXXXX 100644
13
index XXXXXXX..XXXXXXX 100644
14
--- a/block/io.c
14
--- a/block/io.c
15
+++ b/block/io.c
15
+++ b/block/io.c
16
@@ -XXX,XX +XXX,XX @@ static void coroutine_fn tracked_request_end(BdrvTrackedRequest *req)
16
@@ -XXX,XX +XXX,XX @@ void bdrv_drained_begin(BlockDriverState *bs)
17
17
18
qemu_co_mutex_lock(&req->bs->reqs_lock);
18
void bdrv_drained_end(BlockDriverState *bs)
19
QLIST_REMOVE(req, list);
19
{
20
- qemu_co_queue_restart_all(&req->wait_queue);
20
+ int old_quiesce_counter;
21
qemu_co_mutex_unlock(&req->bs->reqs_lock);
22
+
21
+
23
+ /*
22
if (qemu_in_coroutine()) {
24
+ * At this point qemu_co_queue_wait(&req->wait_queue, ...) won't be called
23
bdrv_co_yield_to_drain(bs, false);
25
+ * anymore because the request has been removed from the list, so it's safe
24
return;
26
+ * to restart the queue outside reqs_lock to minimize the critical section.
25
}
27
+ */
26
assert(bs->quiesce_counter > 0);
28
+ qemu_co_queue_restart_all(&req->wait_queue);
27
- if (atomic_fetch_dec(&bs->quiesce_counter) > 1) {
28
- return;
29
- }
30
+ old_quiesce_counter = atomic_fetch_dec(&bs->quiesce_counter);
31
32
/* Re-enable things in child-to-parent order */
33
bdrv_drain_invoke(bs, false, false);
34
- bdrv_parent_drained_end(bs);
35
- aio_enable_external(bdrv_get_aio_context(bs));
36
+ if (old_quiesce_counter == 1) {
37
+ bdrv_parent_drained_end(bs);
38
+ aio_enable_external(bdrv_get_aio_context(bs));
39
+ }
29
}
40
}
30
41
31
/**
42
/*
32
--
43
--
33
2.41.0
44
2.13.6
45
46
diff view generated by jsdifflib
1
From: Stefan Hajnoczi <stefanha@redhat.com>
2
3
Switch blk_aio_*() APIs over to multi-queue by using
4
qemu_get_current_aio_context() instead of blk_get_aio_context(). This
5
change will allow devices to process I/O in multiple IOThreads in the
6
future.
7
8
I audited existing blk_aio_*() callers:
9
- migration/block.c: blk_mig_lock() protects the data accessed by the
10
completion callback.
11
- The remaining emulated devices and exports run with
12
qemu_get_aio_context() == blk_get_aio_context().
13
14
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
15
Message-ID: <20230823235938.1398382-3-stefanha@redhat.com>
16
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
17
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
1
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
18
---
2
---
19
block/block-backend.c | 6 +++---
3
tests/test-bdrv-drain.c | 57 +++++++++++++++++++++++++++++++++++++++++++++++++
20
1 file changed, 3 insertions(+), 3 deletions(-)
4
1 file changed, 57 insertions(+)
21
5
22
diff --git a/block/block-backend.c b/block/block-backend.c
6
diff --git a/tests/test-bdrv-drain.c b/tests/test-bdrv-drain.c
23
index XXXXXXX..XXXXXXX 100644
7
index XXXXXXX..XXXXXXX 100644
24
--- a/block/block-backend.c
8
--- a/tests/test-bdrv-drain.c
25
+++ b/block/block-backend.c
9
+++ b/tests/test-bdrv-drain.c
26
@@ -XXX,XX +XXX,XX @@ BlockAIOCB *blk_abort_aio_request(BlockBackend *blk,
10
@@ -XXX,XX +XXX,XX @@ static void aio_ret_cb(void *opaque, int ret)
27
acb->blk = blk;
11
enum drain_type {
28
acb->ret = ret;
12
BDRV_DRAIN_ALL,
29
13
BDRV_DRAIN,
30
- replay_bh_schedule_oneshot_event(blk_get_aio_context(blk),
14
+ DRAIN_TYPE_MAX,
31
+ replay_bh_schedule_oneshot_event(qemu_get_current_aio_context(),
15
};
32
error_callback_bh, acb);
16
33
return &acb->common;
17
static void do_drain_begin(enum drain_type drain_type, BlockDriverState *bs)
18
@@ -XXX,XX +XXX,XX @@ static void test_quiesce_drain(void)
19
test_quiesce_common(BDRV_DRAIN, false);
34
}
20
}
35
@@ -XXX,XX +XXX,XX @@ static BlockAIOCB *blk_aio_prwv(BlockBackend *blk, int64_t offset,
21
36
acb->has_returned = false;
22
+static void test_nested(void)
37
23
+{
38
co = qemu_coroutine_create(co_entry, acb);
24
+ BlockBackend *blk;
39
- aio_co_enter(blk_get_aio_context(blk), co);
25
+ BlockDriverState *bs, *backing;
40
+ aio_co_enter(qemu_get_current_aio_context(), co);
26
+ BDRVTestState *s, *backing_s;
41
27
+ enum drain_type outer, inner;
42
acb->has_returned = true;
28
+
43
if (acb->rwco.ret != NOT_DONE) {
29
+ blk = blk_new(BLK_PERM_ALL, BLK_PERM_ALL);
44
- replay_bh_schedule_oneshot_event(blk_get_aio_context(blk),
30
+ bs = bdrv_new_open_driver(&bdrv_test, "test-node", BDRV_O_RDWR,
45
+ replay_bh_schedule_oneshot_event(qemu_get_current_aio_context(),
31
+ &error_abort);
46
blk_aio_complete_bh, acb);
32
+ s = bs->opaque;
47
}
33
+ blk_insert_bs(blk, bs, &error_abort);
34
+
35
+ backing = bdrv_new_open_driver(&bdrv_test, "backing", 0, &error_abort);
36
+ backing_s = backing->opaque;
37
+ bdrv_set_backing_hd(bs, backing, &error_abort);
38
+
39
+ for (outer = 0; outer < DRAIN_TYPE_MAX; outer++) {
40
+ for (inner = 0; inner < DRAIN_TYPE_MAX; inner++) {
41
+ /* XXX bdrv_drain_all() doesn't increase the quiesce_counter */
42
+ int bs_quiesce = (outer != BDRV_DRAIN_ALL) +
43
+ (inner != BDRV_DRAIN_ALL);
44
+ int backing_quiesce = 0;
45
+ int backing_cb_cnt = (outer != BDRV_DRAIN) +
46
+ (inner != BDRV_DRAIN);
47
+
48
+ g_assert_cmpint(bs->quiesce_counter, ==, 0);
49
+ g_assert_cmpint(backing->quiesce_counter, ==, 0);
50
+ g_assert_cmpint(s->drain_count, ==, 0);
51
+ g_assert_cmpint(backing_s->drain_count, ==, 0);
52
+
53
+ do_drain_begin(outer, bs);
54
+ do_drain_begin(inner, bs);
55
+
56
+ g_assert_cmpint(bs->quiesce_counter, ==, bs_quiesce);
57
+ g_assert_cmpint(backing->quiesce_counter, ==, backing_quiesce);
58
+ g_assert_cmpint(s->drain_count, ==, 2);
59
+ g_assert_cmpint(backing_s->drain_count, ==, backing_cb_cnt);
60
+
61
+ do_drain_end(inner, bs);
62
+ do_drain_end(outer, bs);
63
+
64
+ g_assert_cmpint(bs->quiesce_counter, ==, 0);
65
+ g_assert_cmpint(backing->quiesce_counter, ==, 0);
66
+ g_assert_cmpint(s->drain_count, ==, 0);
67
+ g_assert_cmpint(backing_s->drain_count, ==, 0);
68
+ }
69
+ }
70
+
71
+ bdrv_unref(backing);
72
+ bdrv_unref(bs);
73
+ blk_unref(blk);
74
+}
75
+
76
77
typedef struct TestBlockJob {
78
BlockJob common;
79
@@ -XXX,XX +XXX,XX @@ int main(int argc, char **argv)
80
g_test_add_func("/bdrv-drain/quiesce/drain_all", test_quiesce_drain_all);
81
g_test_add_func("/bdrv-drain/quiesce/drain", test_quiesce_drain);
82
83
+ g_test_add_func("/bdrv-drain/nested", test_nested);
84
+
85
g_test_add_func("/bdrv-drain/blockjob/drain_all", test_blockjob_drain_all);
86
g_test_add_func("/bdrv-drain/blockjob/drain", test_blockjob_drain);
48
87
49
--
88
--
50
2.41.0
89
2.13.6
90
91
diff view generated by jsdifflib
1
From: Hanna Czenczek <hreitz@redhat.com>
1
This is in preparation for subtree drains, i.e. drained sections that
2
2
affect not only a single node, but recursively all child nodes, too.
3
For image creation code, we have central fallback code for protocols
3
4
that do not support creating new images (like NBD or iscsi). So for
4
Calling the parent callbacks for drain is pointless when we just came
5
them, you can only specify existing paths/exports that are overwritten
5
from that parent node recursively and leads to multiple increases of
6
to make clean new images. In such a case, if the given path cannot be
6
bs->quiesce_counter in a single drain call. Don't do it.
7
opened (assuming a pre-existing image there), we print an error message
7
8
that tries to describe what is going on: That with this protocol, you
8
In order for this to work correctly, the parent callback must be called
9
cannot create new images, but only overwrite existing ones; and the
9
for every bdrv_drain_begin/end() call, not only for the outermost one:
10
given path could not be opened as a pre-existing image.
10
11
11
If we have a node N with two parents A and B, recursive draining of A
12
However, the current message is confusing, because it does not say that
12
should cause the quiesce_counter of B to increase because its child N is
13
the protocol in question does not support creating new images, but
13
drained independently of B. If now B is recursively drained, too, A must
14
instead that "image creation" is unsupported. This can be interpreted
14
increase its quiesce_counter because N is drained independently of A
15
to mean that `qemu-img create` will not work in principle, which is not
15
only now, even if N is going from quiesce_counter 1 to 2.
16
true. Be more verbose for clarity.
16
17
18
Buglink: https://bugzilla.redhat.com/show_bug.cgi?id=2217204
19
Signed-off-by: Hanna Czenczek <hreitz@redhat.com>
20
Message-ID: <20230720140024.46836-1-hreitz@redhat.com>
21
Reviewed-by: Eric Blake <eblake@redhat.com>
22
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
23
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
17
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
24
---
18
---
25
block.c | 6 ++++--
19
include/block/block.h | 4 ++--
26
1 file changed, 4 insertions(+), 2 deletions(-)
20
block.c | 13 +++++++++----
27
21
block/io.c | 47 ++++++++++++++++++++++++++++++++++-------------
22
3 files changed, 45 insertions(+), 19 deletions(-)
23
24
diff --git a/include/block/block.h b/include/block/block.h
25
index XXXXXXX..XXXXXXX 100644
26
--- a/include/block/block.h
27
+++ b/include/block/block.h
28
@@ -XXX,XX +XXX,XX @@ void bdrv_io_unplug(BlockDriverState *bs);
29
* Begin a quiesced section of all users of @bs. This is part of
30
* bdrv_drained_begin.
31
*/
32
-void bdrv_parent_drained_begin(BlockDriverState *bs);
33
+void bdrv_parent_drained_begin(BlockDriverState *bs, BdrvChild *ignore);
34
35
/**
36
* bdrv_parent_drained_end:
37
@@ -XXX,XX +XXX,XX @@ void bdrv_parent_drained_begin(BlockDriverState *bs);
38
* End a quiesced section of all users of @bs. This is part of
39
* bdrv_drained_end.
40
*/
41
-void bdrv_parent_drained_end(BlockDriverState *bs);
42
+void bdrv_parent_drained_end(BlockDriverState *bs, BdrvChild *ignore);
43
44
/**
45
* bdrv_drained_begin:
28
diff --git a/block.c b/block.c
46
diff --git a/block.c b/block.c
29
index XXXXXXX..XXXXXXX 100644
47
index XXXXXXX..XXXXXXX 100644
30
--- a/block.c
48
--- a/block.c
31
+++ b/block.c
49
+++ b/block.c
32
@@ -XXX,XX +XXX,XX @@ int coroutine_fn bdrv_co_create_opts_simple(BlockDriver *drv,
50
@@ -XXX,XX +XXX,XX @@ static void bdrv_replace_child_noperm(BdrvChild *child,
33
blk = blk_co_new_open(filename, NULL, options,
51
BlockDriverState *new_bs)
34
BDRV_O_RDWR | BDRV_O_RESIZE, errp);
52
{
35
if (!blk) {
53
BlockDriverState *old_bs = child->bs;
36
- error_prepend(errp, "Protocol driver '%s' does not support image "
54
+ int i;
37
- "creation, and opening the image failed: ",
55
38
+ error_prepend(errp, "Protocol driver '%s' does not support creating "
56
if (old_bs && new_bs) {
39
+ "new images, so an existing image must be selected as "
57
assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
40
+ "the target; however, opening the given target as an "
58
}
41
+ "existing image failed: ",
59
if (old_bs) {
42
drv->format_name);
60
if (old_bs->quiesce_counter && child->role->drained_end) {
43
return -EINVAL;
61
- child->role->drained_end(child);
62
+ for (i = 0; i < old_bs->quiesce_counter; i++) {
63
+ child->role->drained_end(child);
64
+ }
65
}
66
if (child->role->detach) {
67
child->role->detach(child);
68
@@ -XXX,XX +XXX,XX @@ static void bdrv_replace_child_noperm(BdrvChild *child,
69
if (new_bs) {
70
QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
71
if (new_bs->quiesce_counter && child->role->drained_begin) {
72
- child->role->drained_begin(child);
73
+ for (i = 0; i < new_bs->quiesce_counter; i++) {
74
+ child->role->drained_begin(child);
75
+ }
76
}
77
78
if (child->role->attach) {
79
@@ -XXX,XX +XXX,XX @@ void bdrv_set_aio_context(BlockDriverState *bs, AioContext *new_context)
80
AioContext *ctx = bdrv_get_aio_context(bs);
81
82
aio_disable_external(ctx);
83
- bdrv_parent_drained_begin(bs);
84
+ bdrv_parent_drained_begin(bs, NULL);
85
bdrv_drain(bs); /* ensure there are no in-flight requests */
86
87
while (aio_poll(ctx, false)) {
88
@@ -XXX,XX +XXX,XX @@ void bdrv_set_aio_context(BlockDriverState *bs, AioContext *new_context)
89
*/
90
aio_context_acquire(new_context);
91
bdrv_attach_aio_context(bs, new_context);
92
- bdrv_parent_drained_end(bs);
93
+ bdrv_parent_drained_end(bs, NULL);
94
aio_enable_external(ctx);
95
aio_context_release(new_context);
96
}
97
diff --git a/block/io.c b/block/io.c
98
index XXXXXXX..XXXXXXX 100644
99
--- a/block/io.c
100
+++ b/block/io.c
101
@@ -XXX,XX +XXX,XX @@
102
static int coroutine_fn bdrv_co_do_pwrite_zeroes(BlockDriverState *bs,
103
int64_t offset, int bytes, BdrvRequestFlags flags);
104
105
-void bdrv_parent_drained_begin(BlockDriverState *bs)
106
+void bdrv_parent_drained_begin(BlockDriverState *bs, BdrvChild *ignore)
107
{
108
BdrvChild *c, *next;
109
110
QLIST_FOREACH_SAFE(c, &bs->parents, next_parent, next) {
111
+ if (c == ignore) {
112
+ continue;
113
+ }
114
if (c->role->drained_begin) {
115
c->role->drained_begin(c);
116
}
117
}
118
}
119
120
-void bdrv_parent_drained_end(BlockDriverState *bs)
121
+void bdrv_parent_drained_end(BlockDriverState *bs, BdrvChild *ignore)
122
{
123
BdrvChild *c, *next;
124
125
QLIST_FOREACH_SAFE(c, &bs->parents, next_parent, next) {
126
+ if (c == ignore) {
127
+ continue;
128
+ }
129
if (c->role->drained_end) {
130
c->role->drained_end(c);
131
}
132
@@ -XXX,XX +XXX,XX @@ typedef struct {
133
BlockDriverState *bs;
134
bool done;
135
bool begin;
136
+ BdrvChild *parent;
137
} BdrvCoDrainData;
138
139
static void coroutine_fn bdrv_drain_invoke_entry(void *opaque)
140
@@ -XXX,XX +XXX,XX @@ static bool bdrv_drain_recurse(BlockDriverState *bs)
141
return waited;
142
}
143
144
+static void bdrv_do_drained_begin(BlockDriverState *bs, BdrvChild *parent);
145
+static void bdrv_do_drained_end(BlockDriverState *bs, BdrvChild *parent);
146
+
147
static void bdrv_co_drain_bh_cb(void *opaque)
148
{
149
BdrvCoDrainData *data = opaque;
150
@@ -XXX,XX +XXX,XX @@ static void bdrv_co_drain_bh_cb(void *opaque)
151
152
bdrv_dec_in_flight(bs);
153
if (data->begin) {
154
- bdrv_drained_begin(bs);
155
+ bdrv_do_drained_begin(bs, data->parent);
156
} else {
157
- bdrv_drained_end(bs);
158
+ bdrv_do_drained_end(bs, data->parent);
159
}
160
161
data->done = true;
162
@@ -XXX,XX +XXX,XX @@ static void bdrv_co_drain_bh_cb(void *opaque)
163
}
164
165
static void coroutine_fn bdrv_co_yield_to_drain(BlockDriverState *bs,
166
- bool begin)
167
+ bool begin, BdrvChild *parent)
168
{
169
BdrvCoDrainData data;
170
171
@@ -XXX,XX +XXX,XX @@ static void coroutine_fn bdrv_co_yield_to_drain(BlockDriverState *bs,
172
.bs = bs,
173
.done = false,
174
.begin = begin,
175
+ .parent = parent,
176
};
177
bdrv_inc_in_flight(bs);
178
aio_bh_schedule_oneshot(bdrv_get_aio_context(bs),
179
@@ -XXX,XX +XXX,XX @@ static void coroutine_fn bdrv_co_yield_to_drain(BlockDriverState *bs,
180
assert(data.done);
181
}
182
183
-void bdrv_drained_begin(BlockDriverState *bs)
184
+static void bdrv_do_drained_begin(BlockDriverState *bs, BdrvChild *parent)
185
{
186
if (qemu_in_coroutine()) {
187
- bdrv_co_yield_to_drain(bs, true);
188
+ bdrv_co_yield_to_drain(bs, true, parent);
189
return;
190
}
191
192
/* Stop things in parent-to-child order */
193
if (atomic_fetch_inc(&bs->quiesce_counter) == 0) {
194
aio_disable_external(bdrv_get_aio_context(bs));
195
- bdrv_parent_drained_begin(bs);
196
}
197
198
+ bdrv_parent_drained_begin(bs, parent);
199
bdrv_drain_invoke(bs, true, false);
200
bdrv_drain_recurse(bs);
201
}
202
203
-void bdrv_drained_end(BlockDriverState *bs)
204
+void bdrv_drained_begin(BlockDriverState *bs)
205
+{
206
+ bdrv_do_drained_begin(bs, NULL);
207
+}
208
+
209
+static void bdrv_do_drained_end(BlockDriverState *bs, BdrvChild *parent)
210
{
211
int old_quiesce_counter;
212
213
if (qemu_in_coroutine()) {
214
- bdrv_co_yield_to_drain(bs, false);
215
+ bdrv_co_yield_to_drain(bs, false, parent);
216
return;
217
}
218
assert(bs->quiesce_counter > 0);
219
@@ -XXX,XX +XXX,XX @@ void bdrv_drained_end(BlockDriverState *bs)
220
221
/* Re-enable things in child-to-parent order */
222
bdrv_drain_invoke(bs, false, false);
223
+ bdrv_parent_drained_end(bs, parent);
224
if (old_quiesce_counter == 1) {
225
- bdrv_parent_drained_end(bs);
226
aio_enable_external(bdrv_get_aio_context(bs));
227
}
228
}
229
230
+void bdrv_drained_end(BlockDriverState *bs)
231
+{
232
+ bdrv_do_drained_end(bs, NULL);
233
+}
234
+
235
/*
236
* Wait for pending requests to complete on a single BlockDriverState subtree,
237
* and suspend block driver's internal I/O until next request arrives.
238
@@ -XXX,XX +XXX,XX @@ void bdrv_drain_all_begin(void)
239
/* Stop things in parent-to-child order */
240
aio_context_acquire(aio_context);
241
aio_disable_external(aio_context);
242
- bdrv_parent_drained_begin(bs);
243
+ bdrv_parent_drained_begin(bs, NULL);
244
bdrv_drain_invoke(bs, true, true);
245
aio_context_release(aio_context);
246
247
@@ -XXX,XX +XXX,XX @@ void bdrv_drain_all_end(void)
248
/* Re-enable things in child-to-parent order */
249
aio_context_acquire(aio_context);
250
bdrv_drain_invoke(bs, false, true);
251
- bdrv_parent_drained_end(bs);
252
+ bdrv_parent_drained_end(bs, NULL);
253
aio_enable_external(aio_context);
254
aio_context_release(aio_context);
44
}
255
}
45
--
256
--
46
2.41.0
257
2.13.6
258
259
diff view generated by jsdifflib
1
From: Stefan Hajnoczi <stefanha@redhat.com>
1
bdrv_drained_begin() waits for the completion of requests in the whole
2
subtree, but it only actually keeps its immediate bs parameter quiesced
3
until bdrv_drained_end().
2
4
3
The synchronous bdrv_aio_cancel() function needs the acb's AioContext so
5
Add a version that keeps the whole subtree drained. As of this commit,
4
it can call aio_poll() to wait for cancellation.
6
graph changes cannot be allowed during a subtree drained section, but
7
this will be fixed soon.
5
8
6
It turns out that all users run under the BQL in the main AioContext, so
7
this callback is not needed.
8
9
Remove the callback, mark bdrv_aio_cancel() GLOBAL_STATE_CODE just like
10
its blk_aio_cancel() caller, and poll the main loop AioContext.
11
12
The purpose of this cleanup is to identify bdrv_aio_cancel() as an API
13
that does not work with the multi-queue block layer.
14
15
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
16
Message-ID: <20230823235938.1398382-2-stefanha@redhat.com>
17
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
18
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
9
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
19
---
10
---
20
include/block/aio.h | 1 -
11
include/block/block.h | 13 +++++++++++++
21
include/block/block-global-state.h | 2 ++
12
block/io.c | 54 ++++++++++++++++++++++++++++++++++++++++-----------
22
include/block/block-io.h | 1 -
13
2 files changed, 56 insertions(+), 11 deletions(-)
23
block/block-backend.c | 17 -----------------
24
block/io.c | 23 ++++++++---------------
25
hw/nvme/ctrl.c | 7 -------
26
softmmu/dma-helpers.c | 8 --------
27
util/thread-pool.c | 8 --------
28
8 files changed, 10 insertions(+), 57 deletions(-)
29
14
30
diff --git a/include/block/aio.h b/include/block/aio.h
15
diff --git a/include/block/block.h b/include/block/block.h
31
index XXXXXXX..XXXXXXX 100644
16
index XXXXXXX..XXXXXXX 100644
32
--- a/include/block/aio.h
17
--- a/include/block/block.h
33
+++ b/include/block/aio.h
18
+++ b/include/block/block.h
34
@@ -XXX,XX +XXX,XX @@ typedef void BlockCompletionFunc(void *opaque, int ret);
19
@@ -XXX,XX +XXX,XX @@ void bdrv_parent_drained_end(BlockDriverState *bs, BdrvChild *ignore);
35
20
void bdrv_drained_begin(BlockDriverState *bs);
36
typedef struct AIOCBInfo {
21
37
void (*cancel_async)(BlockAIOCB *acb);
22
/**
38
- AioContext *(*get_aio_context)(BlockAIOCB *acb);
23
+ * Like bdrv_drained_begin, but recursively begins a quiesced section for
39
size_t aiocb_size;
24
+ * exclusive access to all child nodes as well.
40
} AIOCBInfo;
25
+ *
41
26
+ * Graph changes are not allowed during a subtree drain section.
42
diff --git a/include/block/block-global-state.h b/include/block/block-global-state.h
27
+ */
43
index XXXXXXX..XXXXXXX 100644
28
+void bdrv_subtree_drained_begin(BlockDriverState *bs);
44
--- a/include/block/block-global-state.h
45
+++ b/include/block/block-global-state.h
46
@@ -XXX,XX +XXX,XX @@ void bdrv_drain_all_begin_nopoll(void);
47
void bdrv_drain_all_end(void);
48
void bdrv_drain_all(void);
49
50
+void bdrv_aio_cancel(BlockAIOCB *acb);
51
+
29
+
52
int bdrv_has_zero_init_1(BlockDriverState *bs);
30
+/**
53
int bdrv_has_zero_init(BlockDriverState *bs);
31
* bdrv_drained_end:
54
BlockDriverState *bdrv_find_node(const char *node_name);
32
*
55
diff --git a/include/block/block-io.h b/include/block/block-io.h
33
* End a quiescent section started by bdrv_drained_begin().
56
index XXXXXXX..XXXXXXX 100644
34
*/
57
--- a/include/block/block-io.h
35
void bdrv_drained_end(BlockDriverState *bs);
58
+++ b/include/block/block-io.h
36
59
@@ -XXX,XX +XXX,XX @@ bdrv_co_delete_file_noerr(BlockDriverState *bs);
37
+/**
60
38
+ * End a quiescent section started by bdrv_subtree_drained_begin().
61
39
+ */
62
/* async block I/O */
40
+void bdrv_subtree_drained_end(BlockDriverState *bs);
63
-void bdrv_aio_cancel(BlockAIOCB *acb);
41
+
64
void bdrv_aio_cancel_async(BlockAIOCB *acb);
42
void bdrv_add_child(BlockDriverState *parent, BlockDriverState *child,
65
43
Error **errp);
66
/* sg packet commands */
44
void bdrv_del_child(BlockDriverState *parent, BdrvChild *child, Error **errp);
67
diff --git a/block/block-backend.c b/block/block-backend.c
68
index XXXXXXX..XXXXXXX 100644
69
--- a/block/block-backend.c
70
+++ b/block/block-backend.c
71
@@ -XXX,XX +XXX,XX @@
72
73
#define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
74
75
-static AioContext *blk_aiocb_get_aio_context(BlockAIOCB *acb);
76
-
77
typedef struct BlockBackendAioNotifier {
78
void (*attached_aio_context)(AioContext *new_context, void *opaque);
79
void (*detach_aio_context)(void *opaque);
80
@@ -XXX,XX +XXX,XX @@ typedef struct BlockBackendAIOCB {
81
} BlockBackendAIOCB;
82
83
static const AIOCBInfo block_backend_aiocb_info = {
84
- .get_aio_context = blk_aiocb_get_aio_context,
85
.aiocb_size = sizeof(BlockBackendAIOCB),
86
};
87
88
@@ -XXX,XX +XXX,XX @@ typedef struct BlkAioEmAIOCB {
89
bool has_returned;
90
} BlkAioEmAIOCB;
91
92
-static AioContext *blk_aio_em_aiocb_get_aio_context(BlockAIOCB *acb_)
93
-{
94
- BlkAioEmAIOCB *acb = container_of(acb_, BlkAioEmAIOCB, common);
95
-
96
- return blk_get_aio_context(acb->rwco.blk);
97
-}
98
-
99
static const AIOCBInfo blk_aio_em_aiocb_info = {
100
.aiocb_size = sizeof(BlkAioEmAIOCB),
101
- .get_aio_context = blk_aio_em_aiocb_get_aio_context,
102
};
103
104
static void blk_aio_complete(BlkAioEmAIOCB *acb)
105
@@ -XXX,XX +XXX,XX @@ AioContext *blk_get_aio_context(BlockBackend *blk)
106
return blk->ctx;
107
}
108
109
-static AioContext *blk_aiocb_get_aio_context(BlockAIOCB *acb)
110
-{
111
- BlockBackendAIOCB *blk_acb = DO_UPCAST(BlockBackendAIOCB, common, acb);
112
- return blk_get_aio_context(blk_acb->blk);
113
-}
114
-
115
int blk_set_aio_context(BlockBackend *blk, AioContext *new_context,
116
Error **errp)
117
{
118
diff --git a/block/io.c b/block/io.c
45
diff --git a/block/io.c b/block/io.c
119
index XXXXXXX..XXXXXXX 100644
46
index XXXXXXX..XXXXXXX 100644
120
--- a/block/io.c
47
--- a/block/io.c
121
+++ b/block/io.c
48
+++ b/block/io.c
122
@@ -XXX,XX +XXX,XX @@ int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
49
@@ -XXX,XX +XXX,XX @@ typedef struct {
123
/**************************************************************/
50
BlockDriverState *bs;
124
/* async I/Os */
51
bool done;
125
52
bool begin;
126
+/**
53
+ bool recursive;
127
+ * Synchronously cancels an acb. Must be called with the BQL held and the acb
54
BdrvChild *parent;
128
+ * must be processed with the BQL held too (IOThreads are not allowed).
55
} BdrvCoDrainData;
129
+ *
56
130
+ * Use bdrv_aio_cancel_async() instead when possible.
57
@@ -XXX,XX +XXX,XX @@ static bool bdrv_drain_recurse(BlockDriverState *bs)
131
+ */
58
return waited;
132
void bdrv_aio_cancel(BlockAIOCB *acb)
59
}
60
61
-static void bdrv_do_drained_begin(BlockDriverState *bs, BdrvChild *parent);
62
-static void bdrv_do_drained_end(BlockDriverState *bs, BdrvChild *parent);
63
+static void bdrv_do_drained_begin(BlockDriverState *bs, bool recursive,
64
+ BdrvChild *parent);
65
+static void bdrv_do_drained_end(BlockDriverState *bs, bool recursive,
66
+ BdrvChild *parent);
67
68
static void bdrv_co_drain_bh_cb(void *opaque)
133
{
69
{
134
- IO_CODE();
70
@@ -XXX,XX +XXX,XX @@ static void bdrv_co_drain_bh_cb(void *opaque)
135
+ GLOBAL_STATE_CODE();
71
136
qemu_aio_ref(acb);
72
bdrv_dec_in_flight(bs);
137
bdrv_aio_cancel_async(acb);
73
if (data->begin) {
138
- while (acb->refcnt > 1) {
74
- bdrv_do_drained_begin(bs, data->parent);
139
- if (acb->aiocb_info->get_aio_context) {
75
+ bdrv_do_drained_begin(bs, data->recursive, data->parent);
140
- aio_poll(acb->aiocb_info->get_aio_context(acb), true);
76
} else {
141
- } else if (acb->bs) {
77
- bdrv_do_drained_end(bs, data->parent);
142
- /* qemu_aio_ref and qemu_aio_unref are not thread-safe, so
78
+ bdrv_do_drained_end(bs, data->recursive, data->parent);
143
- * assert that we're not using an I/O thread. Thread-safe
79
}
144
- * code should use bdrv_aio_cancel_async exclusively.
80
145
- */
81
data->done = true;
146
- assert(bdrv_get_aio_context(acb->bs) == qemu_get_aio_context());
82
@@ -XXX,XX +XXX,XX @@ static void bdrv_co_drain_bh_cb(void *opaque)
147
- aio_poll(bdrv_get_aio_context(acb->bs), true);
148
- } else {
149
- abort();
150
- }
151
- }
152
+ AIO_WAIT_WHILE_UNLOCKED(NULL, acb->refcnt > 1);
153
qemu_aio_unref(acb);
154
}
83
}
155
84
156
diff --git a/hw/nvme/ctrl.c b/hw/nvme/ctrl.c
85
static void coroutine_fn bdrv_co_yield_to_drain(BlockDriverState *bs,
157
index XXXXXXX..XXXXXXX 100644
86
- bool begin, BdrvChild *parent)
158
--- a/hw/nvme/ctrl.c
87
+ bool begin, bool recursive,
159
+++ b/hw/nvme/ctrl.c
88
+ BdrvChild *parent)
160
@@ -XXX,XX +XXX,XX @@ static inline bool nvme_is_write(NvmeRequest *req)
89
{
161
rw->opcode == NVME_CMD_WRITE_ZEROES;
90
BdrvCoDrainData data;
91
92
@@ -XXX,XX +XXX,XX @@ static void coroutine_fn bdrv_co_yield_to_drain(BlockDriverState *bs,
93
.bs = bs,
94
.done = false,
95
.begin = begin,
96
+ .recursive = recursive,
97
.parent = parent,
98
};
99
bdrv_inc_in_flight(bs);
100
@@ -XXX,XX +XXX,XX @@ static void coroutine_fn bdrv_co_yield_to_drain(BlockDriverState *bs,
101
assert(data.done);
162
}
102
}
163
103
164
-static AioContext *nvme_get_aio_context(BlockAIOCB *acb)
104
-static void bdrv_do_drained_begin(BlockDriverState *bs, BdrvChild *parent)
165
-{
105
+static void bdrv_do_drained_begin(BlockDriverState *bs, bool recursive,
166
- return qemu_get_aio_context();
106
+ BdrvChild *parent)
167
-}
168
-
169
static void nvme_misc_cb(void *opaque, int ret)
170
{
107
{
171
NvmeRequest *req = opaque;
108
+ BdrvChild *child, *next;
172
@@ -XXX,XX +XXX,XX @@ static void nvme_flush_cancel(BlockAIOCB *acb)
109
+
173
static const AIOCBInfo nvme_flush_aiocb_info = {
110
if (qemu_in_coroutine()) {
174
.aiocb_size = sizeof(NvmeFlushAIOCB),
111
- bdrv_co_yield_to_drain(bs, true, parent);
175
.cancel_async = nvme_flush_cancel,
112
+ bdrv_co_yield_to_drain(bs, true, recursive, parent);
176
- .get_aio_context = nvme_get_aio_context,
113
return;
177
};
178
179
static void nvme_do_flush(NvmeFlushAIOCB *iocb);
180
@@ -XXX,XX +XXX,XX @@ static void nvme_format_cancel(BlockAIOCB *aiocb)
181
static const AIOCBInfo nvme_format_aiocb_info = {
182
.aiocb_size = sizeof(NvmeFormatAIOCB),
183
.cancel_async = nvme_format_cancel,
184
- .get_aio_context = nvme_get_aio_context,
185
};
186
187
static void nvme_format_set(NvmeNamespace *ns, uint8_t lbaf, uint8_t mset,
188
diff --git a/softmmu/dma-helpers.c b/softmmu/dma-helpers.c
189
index XXXXXXX..XXXXXXX 100644
190
--- a/softmmu/dma-helpers.c
191
+++ b/softmmu/dma-helpers.c
192
@@ -XXX,XX +XXX,XX @@ static void dma_aio_cancel(BlockAIOCB *acb)
193
}
114
}
115
116
@@ -XXX,XX +XXX,XX @@ static void bdrv_do_drained_begin(BlockDriverState *bs, BdrvChild *parent)
117
bdrv_parent_drained_begin(bs, parent);
118
bdrv_drain_invoke(bs, true, false);
119
bdrv_drain_recurse(bs);
120
+
121
+ if (recursive) {
122
+ QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
123
+ bdrv_do_drained_begin(child->bs, true, child);
124
+ }
125
+ }
194
}
126
}
195
127
196
-static AioContext *dma_get_aio_context(BlockAIOCB *acb)
128
void bdrv_drained_begin(BlockDriverState *bs)
197
-{
129
{
198
- DMAAIOCB *dbs = container_of(acb, DMAAIOCB, common);
130
- bdrv_do_drained_begin(bs, NULL);
199
-
131
+ bdrv_do_drained_begin(bs, false, NULL);
200
- return dbs->ctx;
132
+}
201
-}
133
+
202
-
134
+void bdrv_subtree_drained_begin(BlockDriverState *bs)
203
static const AIOCBInfo dma_aiocb_info = {
135
+{
204
.aiocb_size = sizeof(DMAAIOCB),
136
+ bdrv_do_drained_begin(bs, true, NULL);
205
.cancel_async = dma_aio_cancel,
206
- .get_aio_context = dma_get_aio_context,
207
};
208
209
BlockAIOCB *dma_blk_io(AioContext *ctx,
210
diff --git a/util/thread-pool.c b/util/thread-pool.c
211
index XXXXXXX..XXXXXXX 100644
212
--- a/util/thread-pool.c
213
+++ b/util/thread-pool.c
214
@@ -XXX,XX +XXX,XX @@ static void thread_pool_cancel(BlockAIOCB *acb)
215
216
}
137
}
217
138
218
-static AioContext *thread_pool_get_aio_context(BlockAIOCB *acb)
139
-static void bdrv_do_drained_end(BlockDriverState *bs, BdrvChild *parent)
219
-{
140
+static void bdrv_do_drained_end(BlockDriverState *bs, bool recursive,
220
- ThreadPoolElement *elem = (ThreadPoolElement *)acb;
141
+ BdrvChild *parent)
221
- ThreadPool *pool = elem->pool;
142
{
222
- return pool->ctx;
143
+ BdrvChild *child, *next;
223
-}
144
int old_quiesce_counter;
224
-
145
225
static const AIOCBInfo thread_pool_aiocb_info = {
146
if (qemu_in_coroutine()) {
226
.aiocb_size = sizeof(ThreadPoolElement),
147
- bdrv_co_yield_to_drain(bs, false, parent);
227
.cancel_async = thread_pool_cancel,
148
+ bdrv_co_yield_to_drain(bs, false, recursive, parent);
228
- .get_aio_context = thread_pool_get_aio_context,
149
return;
229
};
150
}
230
151
assert(bs->quiesce_counter > 0);
231
BlockAIOCB *thread_pool_submit_aio(ThreadPoolFunc *func, void *arg,
152
@@ -XXX,XX +XXX,XX @@ static void bdrv_do_drained_end(BlockDriverState *bs, BdrvChild *parent)
153
if (old_quiesce_counter == 1) {
154
aio_enable_external(bdrv_get_aio_context(bs));
155
}
156
+
157
+ if (recursive) {
158
+ QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
159
+ bdrv_do_drained_end(child->bs, true, child);
160
+ }
161
+ }
162
}
163
164
void bdrv_drained_end(BlockDriverState *bs)
165
{
166
- bdrv_do_drained_end(bs, NULL);
167
+ bdrv_do_drained_end(bs, false, NULL);
168
+}
169
+
170
+void bdrv_subtree_drained_end(BlockDriverState *bs)
171
+{
172
+ bdrv_do_drained_end(bs, true, NULL);
173
}
174
175
/*
232
--
176
--
233
2.41.0
177
2.13.6
178
179
diff view generated by jsdifflib
1
Document the 'compression_type' option for qcow2, and mention that
1
Add a subtree drain version to the existing test cases.
2
streamOptimized vmdk supports compression, too.
3
2
4
Reported-by: Richard W.M. Jones <rjones@redhat.com>
5
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
6
Message-ID: <20230901102430.23856-1-kwolf@redhat.com>
7
Reviewed-by: Richard W.M. Jones <rjones@redhat.com>
8
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
3
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
9
---
4
---
10
docs/tools/qemu-img.rst | 19 +++++++++++++++++--
5
tests/test-bdrv-drain.c | 27 ++++++++++++++++++++++++++-
11
1 file changed, 17 insertions(+), 2 deletions(-)
6
1 file changed, 26 insertions(+), 1 deletion(-)
12
7
13
diff --git a/docs/tools/qemu-img.rst b/docs/tools/qemu-img.rst
8
diff --git a/tests/test-bdrv-drain.c b/tests/test-bdrv-drain.c
14
index XXXXXXX..XXXXXXX 100644
9
index XXXXXXX..XXXXXXX 100644
15
--- a/docs/tools/qemu-img.rst
10
--- a/tests/test-bdrv-drain.c
16
+++ b/docs/tools/qemu-img.rst
11
+++ b/tests/test-bdrv-drain.c
17
@@ -XXX,XX +XXX,XX @@ by the used format or see the format descriptions below for details.
12
@@ -XXX,XX +XXX,XX @@ static void aio_ret_cb(void *opaque, int ret)
18
13
enum drain_type {
19
.. option:: -c
14
BDRV_DRAIN_ALL,
20
15
BDRV_DRAIN,
21
- Indicates that target image must be compressed (qcow format only).
16
+ BDRV_SUBTREE_DRAIN,
22
+ Indicates that target image must be compressed (qcow/qcow2 and vmdk with
17
DRAIN_TYPE_MAX,
23
+ streamOptimized subformat only).
18
};
19
20
@@ -XXX,XX +XXX,XX @@ static void do_drain_begin(enum drain_type drain_type, BlockDriverState *bs)
21
switch (drain_type) {
22
case BDRV_DRAIN_ALL: bdrv_drain_all_begin(); break;
23
case BDRV_DRAIN: bdrv_drained_begin(bs); break;
24
+ case BDRV_SUBTREE_DRAIN: bdrv_subtree_drained_begin(bs); break;
25
default: g_assert_not_reached();
26
}
27
}
28
@@ -XXX,XX +XXX,XX @@ static void do_drain_end(enum drain_type drain_type, BlockDriverState *bs)
29
switch (drain_type) {
30
case BDRV_DRAIN_ALL: bdrv_drain_all_end(); break;
31
case BDRV_DRAIN: bdrv_drained_end(bs); break;
32
+ case BDRV_SUBTREE_DRAIN: bdrv_subtree_drained_end(bs); break;
33
default: g_assert_not_reached();
34
}
35
}
36
@@ -XXX,XX +XXX,XX @@ static void test_drv_cb_drain(void)
37
test_drv_cb_common(BDRV_DRAIN, false);
38
}
39
40
+static void test_drv_cb_drain_subtree(void)
41
+{
42
+ test_drv_cb_common(BDRV_SUBTREE_DRAIN, true);
43
+}
24
+
44
+
25
+ For qcow2, the compression algorithm can be specified with the ``-o
45
static void test_quiesce_common(enum drain_type drain_type, bool recursive)
26
+ compression_type=...`` option (see below).
46
{
27
47
BlockBackend *blk;
28
.. option:: -h
48
@@ -XXX,XX +XXX,XX @@ static void test_quiesce_drain(void)
29
49
test_quiesce_common(BDRV_DRAIN, false);
30
@@ -XXX,XX +XXX,XX @@ Supported image file formats:
50
}
31
51
32
QEMU image format, the most versatile format. Use it to have smaller
52
+static void test_quiesce_drain_subtree(void)
33
images (useful if your filesystem does not supports holes, for example
53
+{
34
- on Windows), optional AES encryption, zlib based compression and
54
+ test_quiesce_common(BDRV_SUBTREE_DRAIN, true);
35
+ on Windows), optional AES encryption, zlib or zstd based compression and
55
+}
36
support of multiple VM snapshots.
37
38
Supported options:
39
@@ -XXX,XX +XXX,XX @@ Supported image file formats:
40
``backing_fmt``
41
Image format of the base image
42
43
+ ``compression_type``
44
+ This option configures which compression algorithm will be used for
45
+ compressed clusters on the image. Note that setting this option doesn't yet
46
+ cause the image to actually receive compressed writes. It is most commonly
47
+ used with the ``-c`` option of ``qemu-img convert``, but can also be used
48
+ with the ``compress`` filter driver or backup block jobs with compression
49
+ enabled.
50
+
56
+
51
+ Valid values are ``zlib`` and ``zstd``. For images that use
57
static void test_nested(void)
52
+ ``compat=0.10``, only ``zlib`` compression is available.
58
{
59
BlockBackend *blk;
60
@@ -XXX,XX +XXX,XX @@ static void test_nested(void)
61
/* XXX bdrv_drain_all() doesn't increase the quiesce_counter */
62
int bs_quiesce = (outer != BDRV_DRAIN_ALL) +
63
(inner != BDRV_DRAIN_ALL);
64
- int backing_quiesce = 0;
65
+ int backing_quiesce = (outer == BDRV_SUBTREE_DRAIN) +
66
+ (inner == BDRV_SUBTREE_DRAIN);
67
int backing_cb_cnt = (outer != BDRV_DRAIN) +
68
(inner != BDRV_DRAIN);
69
70
@@ -XXX,XX +XXX,XX @@ static void test_blockjob_drain(void)
71
test_blockjob_common(BDRV_DRAIN);
72
}
73
74
+static void test_blockjob_drain_subtree(void)
75
+{
76
+ test_blockjob_common(BDRV_SUBTREE_DRAIN);
77
+}
53
+
78
+
54
``encryption``
79
int main(int argc, char **argv)
55
If this option is set to ``on``, the image is encrypted with
80
{
56
128-bit AES-CBC.
81
bdrv_init();
82
@@ -XXX,XX +XXX,XX @@ int main(int argc, char **argv)
83
84
g_test_add_func("/bdrv-drain/driver-cb/drain_all", test_drv_cb_drain_all);
85
g_test_add_func("/bdrv-drain/driver-cb/drain", test_drv_cb_drain);
86
+ g_test_add_func("/bdrv-drain/driver-cb/drain_subtree",
87
+ test_drv_cb_drain_subtree);
88
89
g_test_add_func("/bdrv-drain/quiesce/drain_all", test_quiesce_drain_all);
90
g_test_add_func("/bdrv-drain/quiesce/drain", test_quiesce_drain);
91
+ g_test_add_func("/bdrv-drain/quiesce/drain_subtree",
92
+ test_quiesce_drain_subtree);
93
94
g_test_add_func("/bdrv-drain/nested", test_nested);
95
96
g_test_add_func("/bdrv-drain/blockjob/drain_all", test_blockjob_drain_all);
97
g_test_add_func("/bdrv-drain/blockjob/drain", test_blockjob_drain);
98
+ g_test_add_func("/bdrv-drain/blockjob/drain_subtree",
99
+ test_blockjob_drain_subtree);
100
101
return g_test_run();
102
}
57
--
103
--
58
2.41.0
104
2.13.6
105
106
diff view generated by jsdifflib
1
From: Michael Tokarev <mjt@tls.msk.ru>
1
If bdrv_do_drained_begin/end() are called in coroutine context, they
2
first use a BH to get out of the coroutine context. Call some existing
3
tests again from a coroutine to cover this code path.
2
4
3
I'm getting io-qcow2-244 test failure on mips*
4
due to output mismatch:
5
6
Take an internal snapshot:
7
-qemu-img: Could not create snapshot 'test': -95 (Operation not supported)
8
+qemu-img: Could not create snapshot 'test': -122 (Operation not supported)
9
No errors were found on the image.
10
11
This is because errno values might be different across
12
different architectures.
13
14
This error message in qemu-img.c is the only one which
15
prints errno directly, all the rest print strerror(errno)
16
only. Fix this error message and the expected output
17
of the 3 test cases too.
18
19
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
20
Message-ID: <20230811110946.2435067-1-mjt@tls.msk.ru>
21
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
22
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
5
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
23
---
6
---
24
qemu-img.c | 4 ++--
7
tests/test-bdrv-drain.c | 59 +++++++++++++++++++++++++++++++++++++++++++++++++
25
tests/qemu-iotests/080.out | 6 +++---
8
1 file changed, 59 insertions(+)
26
tests/qemu-iotests/112.out | 6 +++---
27
tests/qemu-iotests/244.out | 2 +-
28
4 files changed, 9 insertions(+), 9 deletions(-)
29
9
30
diff --git a/qemu-img.c b/qemu-img.c
10
diff --git a/tests/test-bdrv-drain.c b/tests/test-bdrv-drain.c
31
index XXXXXXX..XXXXXXX 100644
11
index XXXXXXX..XXXXXXX 100644
32
--- a/qemu-img.c
12
--- a/tests/test-bdrv-drain.c
33
+++ b/qemu-img.c
13
+++ b/tests/test-bdrv-drain.c
34
@@ -XXX,XX +XXX,XX @@ static int img_snapshot(int argc, char **argv)
14
@@ -XXX,XX +XXX,XX @@ static void aio_ret_cb(void *opaque, int ret)
35
15
*aio_ret = ret;
36
ret = bdrv_snapshot_create(bs, &sn);
16
}
37
if (ret) {
17
38
- error_report("Could not create snapshot '%s': %d (%s)",
18
+typedef struct CallInCoroutineData {
39
- snapshot_name, ret, strerror(-ret));
19
+ void (*entry)(void);
40
+ error_report("Could not create snapshot '%s': %s",
20
+ bool done;
41
+ snapshot_name, strerror(-ret));
21
+} CallInCoroutineData;
42
}
22
+
43
break;
23
+static coroutine_fn void call_in_coroutine_entry(void *opaque)
44
24
+{
45
diff --git a/tests/qemu-iotests/080.out b/tests/qemu-iotests/080.out
25
+ CallInCoroutineData *data = opaque;
46
index XXXXXXX..XXXXXXX 100644
26
+
47
--- a/tests/qemu-iotests/080.out
27
+ data->entry();
48
+++ b/tests/qemu-iotests/080.out
28
+ data->done = true;
49
@@ -XXX,XX +XXX,XX @@ qemu-io: can't open device TEST_DIR/t.qcow2: Snapshot table offset invalid
29
+}
50
30
+
51
== Hitting snapshot table size limit ==
31
+static void call_in_coroutine(void (*entry)(void))
52
Formatting 'TEST_DIR/t.IMGFMT', fmt=IMGFMT size=67108864
32
+{
53
-qemu-img: Could not create snapshot 'test': -27 (File too large)
33
+ Coroutine *co;
54
+qemu-img: Could not create snapshot 'test': File too large
34
+ CallInCoroutineData data = {
55
read 512/512 bytes at offset 0
35
+ .entry = entry,
56
512 bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
36
+ .done = false,
57
37
+ };
58
@@ -XXX,XX +XXX,XX @@ qemu-io: can't open device TEST_DIR/t.qcow2: Backing file name too long
38
+
59
Formatting 'TEST_DIR/t.IMGFMT', fmt=IMGFMT size=67108864
39
+ co = qemu_coroutine_create(call_in_coroutine_entry, &data);
60
wrote 512/512 bytes at offset 0
40
+ qemu_coroutine_enter(co);
61
512 bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
41
+ while (!data.done) {
62
-qemu-img: Could not create snapshot 'test': -27 (File too large)
42
+ aio_poll(qemu_get_aio_context(), true);
63
-qemu-img: Could not create snapshot 'test': -11 (Resource temporarily unavailable)
43
+ }
64
+qemu-img: Could not create snapshot 'test': File too large
44
+}
65
+qemu-img: Could not create snapshot 'test': Resource temporarily unavailable
45
+
66
46
enum drain_type {
67
== Invalid snapshot L1 table offset ==
47
BDRV_DRAIN_ALL,
68
Formatting 'TEST_DIR/t.IMGFMT', fmt=IMGFMT size=67108864
48
BDRV_DRAIN,
69
diff --git a/tests/qemu-iotests/112.out b/tests/qemu-iotests/112.out
49
@@ -XXX,XX +XXX,XX @@ static void test_drv_cb_drain_subtree(void)
70
index XXXXXXX..XXXXXXX 100644
50
test_drv_cb_common(BDRV_SUBTREE_DRAIN, true);
71
--- a/tests/qemu-iotests/112.out
51
}
72
+++ b/tests/qemu-iotests/112.out
52
73
@@ -XXX,XX +XXX,XX @@ Formatting 'TEST_DIR/t.IMGFMT', fmt=IMGFMT size=67108864
53
+static void test_drv_cb_co_drain(void)
74
refcount bits: 1
54
+{
75
wrote 512/512 bytes at offset 0
55
+ call_in_coroutine(test_drv_cb_drain);
76
512 bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
56
+}
77
-qemu-img: Could not create snapshot 'foo': -22 (Invalid argument)
57
+
78
+qemu-img: Could not create snapshot 'foo': Invalid argument
58
+static void test_drv_cb_co_drain_subtree(void)
79
Leaked cluster 6 refcount=1 reference=0
59
+{
80
60
+ call_in_coroutine(test_drv_cb_drain_subtree);
81
1 leaked clusters were found on the image.
61
+}
82
@@ -XXX,XX +XXX,XX @@ Formatting 'TEST_DIR/t.IMGFMT', fmt=IMGFMT size=67108864
62
+
83
refcount bits: 2
63
static void test_quiesce_common(enum drain_type drain_type, bool recursive)
84
wrote 512/512 bytes at offset 0
64
{
85
512 bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
65
BlockBackend *blk;
86
-qemu-img: Could not create snapshot 'baz': -22 (Invalid argument)
66
@@ -XXX,XX +XXX,XX @@ static void test_quiesce_drain_subtree(void)
87
+qemu-img: Could not create snapshot 'baz': Invalid argument
67
test_quiesce_common(BDRV_SUBTREE_DRAIN, true);
88
Leaked cluster 7 refcount=1 reference=0
68
}
89
69
90
1 leaked clusters were found on the image.
70
+static void test_quiesce_co_drain(void)
91
@@ -XXX,XX +XXX,XX @@ Formatting 'TEST_DIR/t.IMGFMT', fmt=IMGFMT size=67108864
71
+{
92
refcount bits: 64
72
+ call_in_coroutine(test_quiesce_drain);
93
wrote 512/512 bytes at offset 0
73
+}
94
512 bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
74
+
95
-qemu-img: Could not create snapshot 'foo': -22 (Invalid argument)
75
+static void test_quiesce_co_drain_subtree(void)
96
+qemu-img: Could not create snapshot 'foo': Invalid argument
76
+{
97
Leaked cluster 5 refcount=18446744073709551615 reference=1
77
+ call_in_coroutine(test_quiesce_drain_subtree);
98
Leaked cluster 6 refcount=1 reference=0
78
+}
99
79
+
100
diff --git a/tests/qemu-iotests/244.out b/tests/qemu-iotests/244.out
80
static void test_nested(void)
101
index XXXXXXX..XXXXXXX 100644
81
{
102
--- a/tests/qemu-iotests/244.out
82
BlockBackend *blk;
103
+++ b/tests/qemu-iotests/244.out
83
@@ -XXX,XX +XXX,XX @@ int main(int argc, char **argv)
104
@@ -XXX,XX +XXX,XX @@ write failed: Operation not supported
84
g_test_add_func("/bdrv-drain/driver-cb/drain_subtree",
105
No errors were found on the image.
85
test_drv_cb_drain_subtree);
106
86
107
Take an internal snapshot:
87
+ // XXX bdrv_drain_all() doesn't work in coroutine context
108
-qemu-img: Could not create snapshot 'test': -95 (Operation not supported)
88
+ g_test_add_func("/bdrv-drain/driver-cb/co/drain", test_drv_cb_co_drain);
109
+qemu-img: Could not create snapshot 'test': Operation not supported
89
+ g_test_add_func("/bdrv-drain/driver-cb/co/drain_subtree",
110
No errors were found on the image.
90
+ test_drv_cb_co_drain_subtree);
111
91
+
112
=== Standalone image with external data file (efficient) ===
92
+
93
g_test_add_func("/bdrv-drain/quiesce/drain_all", test_quiesce_drain_all);
94
g_test_add_func("/bdrv-drain/quiesce/drain", test_quiesce_drain);
95
g_test_add_func("/bdrv-drain/quiesce/drain_subtree",
96
test_quiesce_drain_subtree);
97
98
+ // XXX bdrv_drain_all() doesn't work in coroutine context
99
+ g_test_add_func("/bdrv-drain/quiesce/co/drain", test_quiesce_co_drain);
100
+ g_test_add_func("/bdrv-drain/quiesce/co/drain_subtree",
101
+ test_quiesce_co_drain_subtree);
102
+
103
g_test_add_func("/bdrv-drain/nested", test_nested);
104
105
g_test_add_func("/bdrv-drain/blockjob/drain_all", test_blockjob_drain_all);
113
--
106
--
114
2.41.0
107
2.13.6
108
109
diff view generated by jsdifflib
1
From: Fiona Ebner <f.ebner@proxmox.com>
1
Test that drain sections are correctly propagated through the graph.
2
2
3
Since commit ca2a5e630d ("qemu_cleanup: begin drained section after
4
vm_shutdown()"), there will be an additional pause for jobs during
5
qemu_cleanup(). The reason is that the bdrv_drain_all() call in
6
do_vm_stop() is not inside the drained section used by qemu_cleanup()
7
anymore. I.e., there is a second drained section now that ends before
8
the final one in qemu_cleanup() starts. Thus, job_pause() is called
9
twice during cleanup (via child_job_drained_begin()).
10
11
Test 185 needs to be adapted directly too, because it waits for a
12
specific number of JOB_STATUS_CHANGE events before the
13
BLOCK_JOB_CANCELLED event.
14
15
Reported-by: Kevin Wolf <kwolf@redhat.com>
16
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
17
Message-ID: <20230817112538.255111-1-f.ebner@proxmox.com>
18
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
19
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
3
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
20
---
4
---
21
tests/qemu-iotests/109.out | 24 ++++++++++++++++++++++++
5
tests/test-bdrv-drain.c | 74 +++++++++++++++++++++++++++++++++++++++++++++++++
22
tests/qemu-iotests/185 | 2 ++
6
1 file changed, 74 insertions(+)
23
tests/qemu-iotests/185.out | 4 ++++
24
3 files changed, 30 insertions(+)
25
7
26
diff --git a/tests/qemu-iotests/109.out b/tests/qemu-iotests/109.out
8
diff --git a/tests/test-bdrv-drain.c b/tests/test-bdrv-drain.c
27
index XXXXXXX..XXXXXXX 100644
9
index XXXXXXX..XXXXXXX 100644
28
--- a/tests/qemu-iotests/109.out
10
--- a/tests/test-bdrv-drain.c
29
+++ b/tests/qemu-iotests/109.out
11
+++ b/tests/test-bdrv-drain.c
30
@@ -XXX,XX +XXX,XX @@ read 512/512 bytes at offset 0
12
@@ -XXX,XX +XXX,XX @@ static void test_nested(void)
31
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "SHUTDOWN", "data": {"guest": false, "reason": "host-qmp-quit"}}
13
blk_unref(blk);
32
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "standby", "id": "src"}}
14
}
33
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "ready", "id": "src"}}
15
34
+{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "standby", "id": "src"}}
16
+static void test_multiparent(void)
35
+{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "ready", "id": "src"}}
17
+{
36
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "aborting", "id": "src"}}
18
+ BlockBackend *blk_a, *blk_b;
37
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "BLOCK_JOB_CANCELLED", "data": {"device": "src", "len": 1024, "offset": 1024, "speed": 0, "type": "mirror"}}
19
+ BlockDriverState *bs_a, *bs_b, *backing;
38
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "concluded", "id": "src"}}
20
+ BDRVTestState *a_s, *b_s, *backing_s;
39
@@ -XXX,XX +XXX,XX @@ read 512/512 bytes at offset 0
21
+
40
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "SHUTDOWN", "data": {"guest": false, "reason": "host-qmp-quit"}}
22
+ blk_a = blk_new(BLK_PERM_ALL, BLK_PERM_ALL);
41
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "standby", "id": "src"}}
23
+ bs_a = bdrv_new_open_driver(&bdrv_test, "test-node-a", BDRV_O_RDWR,
42
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "ready", "id": "src"}}
24
+ &error_abort);
43
+{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "standby", "id": "src"}}
25
+ a_s = bs_a->opaque;
44
+{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "ready", "id": "src"}}
26
+ blk_insert_bs(blk_a, bs_a, &error_abort);
45
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "aborting", "id": "src"}}
27
+
46
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "BLOCK_JOB_CANCELLED", "data": {"device": "src", "len": 197120, "offset": 197120, "speed": 0, "type": "mirror"}}
28
+ blk_b = blk_new(BLK_PERM_ALL, BLK_PERM_ALL);
47
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "concluded", "id": "src"}}
29
+ bs_b = bdrv_new_open_driver(&bdrv_test, "test-node-b", BDRV_O_RDWR,
48
@@ -XXX,XX +XXX,XX @@ read 512/512 bytes at offset 0
30
+ &error_abort);
49
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "SHUTDOWN", "data": {"guest": false, "reason": "host-qmp-quit"}}
31
+ b_s = bs_b->opaque;
50
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "standby", "id": "src"}}
32
+ blk_insert_bs(blk_b, bs_b, &error_abort);
51
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "ready", "id": "src"}}
33
+
52
+{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "standby", "id": "src"}}
34
+ backing = bdrv_new_open_driver(&bdrv_test, "backing", 0, &error_abort);
53
+{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "ready", "id": "src"}}
35
+ backing_s = backing->opaque;
54
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "aborting", "id": "src"}}
36
+ bdrv_set_backing_hd(bs_a, backing, &error_abort);
55
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "BLOCK_JOB_CANCELLED", "data": {"device": "src", "len": 327680, "offset": 327680, "speed": 0, "type": "mirror"}}
37
+ bdrv_set_backing_hd(bs_b, backing, &error_abort);
56
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "concluded", "id": "src"}}
38
+
57
@@ -XXX,XX +XXX,XX @@ read 512/512 bytes at offset 0
39
+ g_assert_cmpint(bs_a->quiesce_counter, ==, 0);
58
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "SHUTDOWN", "data": {"guest": false, "reason": "host-qmp-quit"}}
40
+ g_assert_cmpint(bs_b->quiesce_counter, ==, 0);
59
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "standby", "id": "src"}}
41
+ g_assert_cmpint(backing->quiesce_counter, ==, 0);
60
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "ready", "id": "src"}}
42
+ g_assert_cmpint(a_s->drain_count, ==, 0);
61
+{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "standby", "id": "src"}}
43
+ g_assert_cmpint(b_s->drain_count, ==, 0);
62
+{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "ready", "id": "src"}}
44
+ g_assert_cmpint(backing_s->drain_count, ==, 0);
63
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "aborting", "id": "src"}}
45
+
64
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "BLOCK_JOB_CANCELLED", "data": {"device": "src", "len": 1024, "offset": 1024, "speed": 0, "type": "mirror"}}
46
+ do_drain_begin(BDRV_SUBTREE_DRAIN, bs_a);
65
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "concluded", "id": "src"}}
47
+
66
@@ -XXX,XX +XXX,XX @@ read 512/512 bytes at offset 0
48
+ g_assert_cmpint(bs_a->quiesce_counter, ==, 1);
67
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "SHUTDOWN", "data": {"guest": false, "reason": "host-qmp-quit"}}
49
+ g_assert_cmpint(bs_b->quiesce_counter, ==, 1);
68
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "standby", "id": "src"}}
50
+ g_assert_cmpint(backing->quiesce_counter, ==, 1);
69
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "ready", "id": "src"}}
51
+ g_assert_cmpint(a_s->drain_count, ==, 1);
70
+{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "standby", "id": "src"}}
52
+ g_assert_cmpint(b_s->drain_count, ==, 1);
71
+{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "ready", "id": "src"}}
53
+ g_assert_cmpint(backing_s->drain_count, ==, 1);
72
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "aborting", "id": "src"}}
54
+
73
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "BLOCK_JOB_CANCELLED", "data": {"device": "src", "len": 65536, "offset": 65536, "speed": 0, "type": "mirror"}}
55
+ do_drain_begin(BDRV_SUBTREE_DRAIN, bs_b);
74
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "concluded", "id": "src"}}
56
+
75
@@ -XXX,XX +XXX,XX @@ read 512/512 bytes at offset 0
57
+ g_assert_cmpint(bs_a->quiesce_counter, ==, 2);
76
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "SHUTDOWN", "data": {"guest": false, "reason": "host-qmp-quit"}}
58
+ g_assert_cmpint(bs_b->quiesce_counter, ==, 2);
77
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "standby", "id": "src"}}
59
+ g_assert_cmpint(backing->quiesce_counter, ==, 2);
78
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "ready", "id": "src"}}
60
+ g_assert_cmpint(a_s->drain_count, ==, 2);
79
+{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "standby", "id": "src"}}
61
+ g_assert_cmpint(b_s->drain_count, ==, 2);
80
+{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "ready", "id": "src"}}
62
+ g_assert_cmpint(backing_s->drain_count, ==, 2);
81
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "aborting", "id": "src"}}
63
+
82
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "BLOCK_JOB_CANCELLED", "data": {"device": "src", "len": 2560, "offset": 2560, "speed": 0, "type": "mirror"}}
64
+ do_drain_end(BDRV_SUBTREE_DRAIN, bs_b);
83
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "concluded", "id": "src"}}
65
+
84
@@ -XXX,XX +XXX,XX @@ read 512/512 bytes at offset 0
66
+ g_assert_cmpint(bs_a->quiesce_counter, ==, 1);
85
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "SHUTDOWN", "data": {"guest": false, "reason": "host-qmp-quit"}}
67
+ g_assert_cmpint(bs_b->quiesce_counter, ==, 1);
86
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "standby", "id": "src"}}
68
+ g_assert_cmpint(backing->quiesce_counter, ==, 1);
87
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "ready", "id": "src"}}
69
+ g_assert_cmpint(a_s->drain_count, ==, 1);
88
+{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "standby", "id": "src"}}
70
+ g_assert_cmpint(b_s->drain_count, ==, 1);
89
+{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "ready", "id": "src"}}
71
+ g_assert_cmpint(backing_s->drain_count, ==, 1);
90
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "aborting", "id": "src"}}
72
+
91
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "BLOCK_JOB_CANCELLED", "data": {"device": "src", "len": 2560, "offset": 2560, "speed": 0, "type": "mirror"}}
73
+ do_drain_end(BDRV_SUBTREE_DRAIN, bs_a);
92
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "concluded", "id": "src"}}
74
+
93
@@ -XXX,XX +XXX,XX @@ read 512/512 bytes at offset 0
75
+ g_assert_cmpint(bs_a->quiesce_counter, ==, 0);
94
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "SHUTDOWN", "data": {"guest": false, "reason": "host-qmp-quit"}}
76
+ g_assert_cmpint(bs_b->quiesce_counter, ==, 0);
95
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "standby", "id": "src"}}
77
+ g_assert_cmpint(backing->quiesce_counter, ==, 0);
96
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "ready", "id": "src"}}
78
+ g_assert_cmpint(a_s->drain_count, ==, 0);
97
+{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "standby", "id": "src"}}
79
+ g_assert_cmpint(b_s->drain_count, ==, 0);
98
+{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "ready", "id": "src"}}
80
+ g_assert_cmpint(backing_s->drain_count, ==, 0);
99
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "aborting", "id": "src"}}
81
+
100
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "BLOCK_JOB_CANCELLED", "data": {"device": "src", "len": 31457280, "offset": 31457280, "speed": 0, "type": "mirror"}}
82
+ bdrv_unref(backing);
101
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "concluded", "id": "src"}}
83
+ bdrv_unref(bs_a);
102
@@ -XXX,XX +XXX,XX @@ read 512/512 bytes at offset 0
84
+ bdrv_unref(bs_b);
103
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "SHUTDOWN", "data": {"guest": false, "reason": "host-qmp-quit"}}
85
+ blk_unref(blk_a);
104
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "standby", "id": "src"}}
86
+ blk_unref(blk_b);
105
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "ready", "id": "src"}}
87
+}
106
+{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "standby", "id": "src"}}
88
+
107
+{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "ready", "id": "src"}}
89
108
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "aborting", "id": "src"}}
90
typedef struct TestBlockJob {
109
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "BLOCK_JOB_CANCELLED", "data": {"device": "src", "len": 327680, "offset": 327680, "speed": 0, "type": "mirror"}}
91
BlockJob common;
110
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "concluded", "id": "src"}}
92
@@ -XXX,XX +XXX,XX @@ int main(int argc, char **argv)
111
@@ -XXX,XX +XXX,XX @@ read 512/512 bytes at offset 0
93
test_quiesce_co_drain_subtree);
112
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "SHUTDOWN", "data": {"guest": false, "reason": "host-qmp-quit"}}
94
113
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "standby", "id": "src"}}
95
g_test_add_func("/bdrv-drain/nested", test_nested);
114
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "ready", "id": "src"}}
96
+ g_test_add_func("/bdrv-drain/multiparent", test_multiparent);
115
+{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "standby", "id": "src"}}
97
116
+{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "ready", "id": "src"}}
98
g_test_add_func("/bdrv-drain/blockjob/drain_all", test_blockjob_drain_all);
117
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "aborting", "id": "src"}}
99
g_test_add_func("/bdrv-drain/blockjob/drain", test_blockjob_drain);
118
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "BLOCK_JOB_CANCELLED", "data": {"device": "src", "len": 2048, "offset": 2048, "speed": 0, "type": "mirror"}}
119
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "concluded", "id": "src"}}
120
@@ -XXX,XX +XXX,XX @@ WARNING: Image format was not specified for 'TEST_DIR/t.raw' and probing guessed
121
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "SHUTDOWN", "data": {"guest": false, "reason": "host-qmp-quit"}}
122
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "standby", "id": "src"}}
123
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "ready", "id": "src"}}
124
+{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "standby", "id": "src"}}
125
+{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "ready", "id": "src"}}
126
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "aborting", "id": "src"}}
127
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "BLOCK_JOB_CANCELLED", "data": {"device": "src", "len": 512, "offset": 512, "speed": 0, "type": "mirror"}}
128
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "concluded", "id": "src"}}
129
@@ -XXX,XX +XXX,XX @@ Images are identical.
130
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "SHUTDOWN", "data": {"guest": false, "reason": "host-qmp-quit"}}
131
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "standby", "id": "src"}}
132
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "ready", "id": "src"}}
133
+{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "standby", "id": "src"}}
134
+{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "ready", "id": "src"}}
135
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "aborting", "id": "src"}}
136
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "BLOCK_JOB_CANCELLED", "data": {"device": "src", "len": 512, "offset": 512, "speed": 0, "type": "mirror"}}
137
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "concluded", "id": "src"}}
138
diff --git a/tests/qemu-iotests/185 b/tests/qemu-iotests/185
139
index XXXXXXX..XXXXXXX 100755
140
--- a/tests/qemu-iotests/185
141
+++ b/tests/qemu-iotests/185
142
@@ -XXX,XX +XXX,XX @@ wait_for_job_and_quit() {
143
QEMU_EVENTS= # Ignore all JOB_STATUS_CHANGE events that came before SHUTDOWN
144
_wait_event $h 'JOB_STATUS_CHANGE' # standby
145
_wait_event $h 'JOB_STATUS_CHANGE' # ready
146
+ _wait_event $h 'JOB_STATUS_CHANGE' # standby
147
+ _wait_event $h 'JOB_STATUS_CHANGE' # ready
148
_wait_event $h 'JOB_STATUS_CHANGE' # aborting
149
# Filter the offset (depends on when exactly `quit` was issued)
150
_wait_event $h 'BLOCK_JOB_CANCELLED' \
151
diff --git a/tests/qemu-iotests/185.out b/tests/qemu-iotests/185.out
152
index XXXXXXX..XXXXXXX 100644
153
--- a/tests/qemu-iotests/185.out
154
+++ b/tests/qemu-iotests/185.out
155
@@ -XXX,XX +XXX,XX @@ Formatting 'TEST_DIR/t.IMGFMT', fmt=IMGFMT size=67108864
156
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "SHUTDOWN", "data": {"guest": false, "reason": "host-qmp-quit"}}
157
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "standby", "id": "mirror"}}
158
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "ready", "id": "mirror"}}
159
+{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "standby", "id": "mirror"}}
160
+{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "ready", "id": "mirror"}}
161
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "aborting", "id": "mirror"}}
162
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "BLOCK_JOB_CANCELLED", "data": {"device": "mirror", "len": 33554432, "offset": (filtered), "speed": 0, "type": "mirror"}}
163
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "concluded", "id": "mirror"}}
164
@@ -XXX,XX +XXX,XX @@ Formatting 'TEST_DIR/t.IMGFMT', fmt=IMGFMT size=67108864
165
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "SHUTDOWN", "data": {"guest": false, "reason": "host-qmp-quit"}}
166
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "standby", "id": "commit"}}
167
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "ready", "id": "commit"}}
168
+{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "standby", "id": "commit"}}
169
+{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "ready", "id": "commit"}}
170
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "aborting", "id": "commit"}}
171
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "BLOCK_JOB_CANCELLED", "data": {"device": "commit", "len": 33554432, "offset": (filtered), "speed": 0, "type": "commit"}}
172
{"timestamp": {"seconds": TIMESTAMP, "microseconds": TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "concluded", "id": "commit"}}
173
--
100
--
174
2.41.0
101
2.13.6
102
103
diff view generated by jsdifflib
1
From: Stefan Hajnoczi <stefanha@redhat.com>
1
We need to remember how many of the drain sections in which a node is
2
2
were recursive (i.e. subtree drain rather than node drain), so that they
3
CoMutex has poor performance when lock contention is high. The tracked
3
can be correctly applied when children are added or removed during the
4
requests list is accessed frequently and performance suffers in QEMU
4
drained section.
5
multi-queue block layer scenarios.
5
6
6
With this change, it is safe to modify the graph even inside a
7
It is not necessary to use CoMutex for the requests lock. The lock is
7
bdrv_subtree_drained_begin/end() section.
8
always released across coroutine yield operations. It is held for
8
9
relatively short periods of time and it is not beneficial to yield when
10
the lock is held by another coroutine.
11
12
Change the lock type from CoMutex to QemuMutex to improve multi-queue
13
block layer performance. fio randread bs=4k iodepth=64 with 4 IOThreads
14
handling a virtio-blk device with 8 virtqueues improves from 254k to
15
517k IOPS (+203%). Full benchmark results and configuration details are
16
available here:
17
https://gitlab.com/stefanha/virt-playbooks/-/commit/980c40845d540e3669add1528739503c2e817b57
18
19
In the future we may wish to introduce thread-local tracked requests
20
lists to avoid lock contention completely. That would be much more
21
involved though.
22
23
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
24
Message-ID: <20230808155852.2745350-3-stefanha@redhat.com>
25
Reviewed-by: Eric Blake <eblake@redhat.com>
26
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
27
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
9
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
28
---
10
---
29
include/block/block_int-common.h | 2 +-
11
include/block/block.h | 2 --
30
block.c | 4 +++-
12
include/block/block_int.h | 5 +++++
31
block/io.c | 24 ++++++++++++------------
13
block.c | 32 +++++++++++++++++++++++++++++---
32
3 files changed, 16 insertions(+), 14 deletions(-)
14
block/io.c | 28 ++++++++++++++++++++++++----
33
15
4 files changed, 58 insertions(+), 9 deletions(-)
34
diff --git a/include/block/block_int-common.h b/include/block/block_int-common.h
16
35
index XXXXXXX..XXXXXXX 100644
17
diff --git a/include/block/block.h b/include/block/block.h
36
--- a/include/block/block_int-common.h
18
index XXXXXXX..XXXXXXX 100644
37
+++ b/include/block/block_int-common.h
19
--- a/include/block/block.h
20
+++ b/include/block/block.h
21
@@ -XXX,XX +XXX,XX @@ void bdrv_drained_begin(BlockDriverState *bs);
22
/**
23
* Like bdrv_drained_begin, but recursively begins a quiesced section for
24
* exclusive access to all child nodes as well.
25
- *
26
- * Graph changes are not allowed during a subtree drain section.
27
*/
28
void bdrv_subtree_drained_begin(BlockDriverState *bs);
29
30
diff --git a/include/block/block_int.h b/include/block/block_int.h
31
index XXXXXXX..XXXXXXX 100644
32
--- a/include/block/block_int.h
33
+++ b/include/block/block_int.h
38
@@ -XXX,XX +XXX,XX @@ struct BlockDriverState {
34
@@ -XXX,XX +XXX,XX @@ struct BlockDriverState {
35
36
/* Accessed with atomic ops. */
37
int quiesce_counter;
38
+ int recursive_quiesce_counter;
39
+
39
unsigned int write_gen; /* Current data generation */
40
unsigned int write_gen; /* Current data generation */
40
41
41
/* Protected by reqs_lock. */
42
/* Protected by reqs_lock. */
42
- CoMutex reqs_lock;
43
@@ -XXX,XX +XXX,XX @@ int coroutine_fn bdrv_co_pwritev(BdrvChild *child,
43
+ QemuMutex reqs_lock;
44
int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
44
QLIST_HEAD(, BdrvTrackedRequest) tracked_requests;
45
BdrvRequestFlags flags);
45
CoQueue flush_queue; /* Serializing flush queue */
46
46
bool active_flush_req; /* Flush request in flight? */
47
+void bdrv_apply_subtree_drain(BdrvChild *child, BlockDriverState *new_parent);
48
+void bdrv_unapply_subtree_drain(BdrvChild *child, BlockDriverState *old_parent);
49
+
50
int get_tmp_filename(char *filename, int size);
51
BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
52
const char *filename);
47
diff --git a/block.c b/block.c
53
diff --git a/block.c b/block.c
48
index XXXXXXX..XXXXXXX 100644
54
index XXXXXXX..XXXXXXX 100644
49
--- a/block.c
55
--- a/block.c
50
+++ b/block.c
56
+++ b/block.c
51
@@ -XXX,XX +XXX,XX @@ BlockDriverState *bdrv_new(void)
57
@@ -XXX,XX +XXX,XX @@ static void bdrv_child_cb_drained_end(BdrvChild *child)
52
for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
58
bdrv_drained_end(bs);
53
QLIST_INIT(&bs->op_blockers[i]);
59
}
60
61
+static void bdrv_child_cb_attach(BdrvChild *child)
62
+{
63
+ BlockDriverState *bs = child->opaque;
64
+ bdrv_apply_subtree_drain(child, bs);
65
+}
66
+
67
+static void bdrv_child_cb_detach(BdrvChild *child)
68
+{
69
+ BlockDriverState *bs = child->opaque;
70
+ bdrv_unapply_subtree_drain(child, bs);
71
+}
72
+
73
static int bdrv_child_cb_inactivate(BdrvChild *child)
74
{
75
BlockDriverState *bs = child->opaque;
76
@@ -XXX,XX +XXX,XX @@ const BdrvChildRole child_file = {
77
.inherit_options = bdrv_inherited_options,
78
.drained_begin = bdrv_child_cb_drained_begin,
79
.drained_end = bdrv_child_cb_drained_end,
80
+ .attach = bdrv_child_cb_attach,
81
+ .detach = bdrv_child_cb_detach,
82
.inactivate = bdrv_child_cb_inactivate,
83
};
84
85
@@ -XXX,XX +XXX,XX @@ const BdrvChildRole child_format = {
86
.inherit_options = bdrv_inherited_fmt_options,
87
.drained_begin = bdrv_child_cb_drained_begin,
88
.drained_end = bdrv_child_cb_drained_end,
89
+ .attach = bdrv_child_cb_attach,
90
+ .detach = bdrv_child_cb_detach,
91
.inactivate = bdrv_child_cb_inactivate,
92
};
93
94
@@ -XXX,XX +XXX,XX @@ static void bdrv_backing_attach(BdrvChild *c)
95
parent->backing_blocker);
96
bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
97
parent->backing_blocker);
98
+
99
+ bdrv_child_cb_attach(c);
100
}
101
102
static void bdrv_backing_detach(BdrvChild *c)
103
@@ -XXX,XX +XXX,XX @@ static void bdrv_backing_detach(BdrvChild *c)
104
bdrv_op_unblock_all(c->bs, parent->backing_blocker);
105
error_free(parent->backing_blocker);
106
parent->backing_blocker = NULL;
107
+
108
+ bdrv_child_cb_detach(c);
109
}
110
111
/*
112
@@ -XXX,XX +XXX,XX @@ static void bdrv_replace_child_noperm(BdrvChild *child,
113
assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
54
}
114
}
55
- qemu_co_mutex_init(&bs->reqs_lock);
115
if (old_bs) {
56
+ qemu_mutex_init(&bs->reqs_lock);
116
+ /* Detach first so that the recursive drain sections coming from @child
57
qemu_mutex_init(&bs->dirty_bitmap_mutex);
117
+ * are already gone and we only end the drain sections that came from
58
bs->refcnt = 1;
118
+ * elsewhere. */
59
bs->aio_context = qemu_get_aio_context();
119
+ if (child->role->detach) {
60
@@ -XXX,XX +XXX,XX @@ static void bdrv_delete(BlockDriverState *bs)
120
+ child->role->detach(child);
61
121
+ }
62
bdrv_close(bs);
122
if (old_bs->quiesce_counter && child->role->drained_end) {
63
123
for (i = 0; i < old_bs->quiesce_counter; i++) {
64
+ qemu_mutex_destroy(&bs->reqs_lock);
124
child->role->drained_end(child);
65
+
125
}
66
g_free(bs);
126
}
67
}
127
- if (child->role->detach) {
68
128
- child->role->detach(child);
129
- }
130
QLIST_REMOVE(child, next_parent);
131
}
132
133
@@ -XXX,XX +XXX,XX @@ static void bdrv_replace_child_noperm(BdrvChild *child,
134
}
135
}
136
137
+ /* Attach only after starting new drained sections, so that recursive
138
+ * drain sections coming from @child don't get an extra .drained_begin
139
+ * callback. */
140
if (child->role->attach) {
141
child->role->attach(child);
142
}
69
diff --git a/block/io.c b/block/io.c
143
diff --git a/block/io.c b/block/io.c
70
index XXXXXXX..XXXXXXX 100644
144
index XXXXXXX..XXXXXXX 100644
71
--- a/block/io.c
145
--- a/block/io.c
72
+++ b/block/io.c
146
+++ b/block/io.c
73
@@ -XXX,XX +XXX,XX @@ static void coroutine_fn tracked_request_end(BdrvTrackedRequest *req)
147
@@ -XXX,XX +XXX,XX @@ static void coroutine_fn bdrv_co_yield_to_drain(BlockDriverState *bs,
74
qatomic_dec(&req->bs->serialising_in_flight);
148
assert(data.done);
149
}
150
151
-static void bdrv_do_drained_begin(BlockDriverState *bs, bool recursive,
152
- BdrvChild *parent)
153
+void bdrv_do_drained_begin(BlockDriverState *bs, bool recursive,
154
+ BdrvChild *parent)
155
{
156
BdrvChild *child, *next;
157
158
@@ -XXX,XX +XXX,XX @@ static void bdrv_do_drained_begin(BlockDriverState *bs, bool recursive,
159
bdrv_drain_recurse(bs);
160
161
if (recursive) {
162
+ bs->recursive_quiesce_counter++;
163
QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
164
bdrv_do_drained_begin(child->bs, true, child);
165
}
166
@@ -XXX,XX +XXX,XX @@ void bdrv_subtree_drained_begin(BlockDriverState *bs)
167
bdrv_do_drained_begin(bs, true, NULL);
168
}
169
170
-static void bdrv_do_drained_end(BlockDriverState *bs, bool recursive,
171
- BdrvChild *parent)
172
+void bdrv_do_drained_end(BlockDriverState *bs, bool recursive,
173
+ BdrvChild *parent)
174
{
175
BdrvChild *child, *next;
176
int old_quiesce_counter;
177
@@ -XXX,XX +XXX,XX @@ static void bdrv_do_drained_end(BlockDriverState *bs, bool recursive,
75
}
178
}
76
179
77
- qemu_co_mutex_lock(&req->bs->reqs_lock);
180
if (recursive) {
78
+ qemu_mutex_lock(&req->bs->reqs_lock);
181
+ bs->recursive_quiesce_counter--;
79
QLIST_REMOVE(req, list);
182
QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
80
- qemu_co_mutex_unlock(&req->bs->reqs_lock);
183
bdrv_do_drained_end(child->bs, true, child);
81
+ qemu_mutex_unlock(&req->bs->reqs_lock);
184
}
82
185
@@ -XXX,XX +XXX,XX @@ void bdrv_subtree_drained_end(BlockDriverState *bs)
83
/*
186
bdrv_do_drained_end(bs, true, NULL);
84
* At this point qemu_co_queue_wait(&req->wait_queue, ...) won't be called
187
}
85
@@ -XXX,XX +XXX,XX @@ static void coroutine_fn tracked_request_begin(BdrvTrackedRequest *req,
188
86
189
+void bdrv_apply_subtree_drain(BdrvChild *child, BlockDriverState *new_parent)
87
qemu_co_queue_init(&req->wait_queue);
190
+{
88
191
+ int i;
89
- qemu_co_mutex_lock(&bs->reqs_lock);
192
+
90
+ qemu_mutex_lock(&bs->reqs_lock);
193
+ for (i = 0; i < new_parent->recursive_quiesce_counter; i++) {
91
QLIST_INSERT_HEAD(&bs->tracked_requests, req, list);
194
+ bdrv_do_drained_begin(child->bs, true, child);
92
- qemu_co_mutex_unlock(&bs->reqs_lock);
195
+ }
93
+ qemu_mutex_unlock(&bs->reqs_lock);
196
+}
94
}
197
+
95
198
+void bdrv_unapply_subtree_drain(BdrvChild *child, BlockDriverState *old_parent)
96
static bool tracked_request_overlaps(BdrvTrackedRequest *req,
199
+{
97
@@ -XXX,XX +XXX,XX @@ bdrv_wait_serialising_requests(BdrvTrackedRequest *self)
200
+ int i;
98
return;
201
+
99
}
202
+ for (i = 0; i < old_parent->recursive_quiesce_counter; i++) {
100
203
+ bdrv_do_drained_end(child->bs, true, child);
101
- qemu_co_mutex_lock(&bs->reqs_lock);
204
+ }
102
+ qemu_mutex_lock(&bs->reqs_lock);
205
+}
103
bdrv_wait_serialising_requests_locked(self);
206
+
104
- qemu_co_mutex_unlock(&bs->reqs_lock);
207
/*
105
+ qemu_mutex_unlock(&bs->reqs_lock);
208
* Wait for pending requests to complete on a single BlockDriverState subtree,
106
}
209
* and suspend block driver's internal I/O until next request arrives.
107
108
void coroutine_fn bdrv_make_request_serialising(BdrvTrackedRequest *req,
109
@@ -XXX,XX +XXX,XX @@ void coroutine_fn bdrv_make_request_serialising(BdrvTrackedRequest *req,
110
{
111
IO_CODE();
112
113
- qemu_co_mutex_lock(&req->bs->reqs_lock);
114
+ qemu_mutex_lock(&req->bs->reqs_lock);
115
116
tracked_request_set_serialising(req, align);
117
bdrv_wait_serialising_requests_locked(req);
118
119
- qemu_co_mutex_unlock(&req->bs->reqs_lock);
120
+ qemu_mutex_unlock(&req->bs->reqs_lock);
121
}
122
123
int bdrv_check_qiov_request(int64_t offset, int64_t bytes,
124
@@ -XXX,XX +XXX,XX @@ int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
125
goto early_exit;
126
}
127
128
- qemu_co_mutex_lock(&bs->reqs_lock);
129
+ qemu_mutex_lock(&bs->reqs_lock);
130
current_gen = qatomic_read(&bs->write_gen);
131
132
/* Wait until any previous flushes are completed */
133
@@ -XXX,XX +XXX,XX @@ int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
134
135
/* Flushes reach this point in nondecreasing current_gen order. */
136
bs->active_flush_req = true;
137
- qemu_co_mutex_unlock(&bs->reqs_lock);
138
+ qemu_mutex_unlock(&bs->reqs_lock);
139
140
/* Write back all layers by calling one driver function */
141
if (bs->drv->bdrv_co_flush) {
142
@@ -XXX,XX +XXX,XX @@ out:
143
bs->flushed_gen = current_gen;
144
}
145
146
- qemu_co_mutex_lock(&bs->reqs_lock);
147
+ qemu_mutex_lock(&bs->reqs_lock);
148
bs->active_flush_req = false;
149
/* Return value is ignored - it's ok if wait queue is empty */
150
qemu_co_queue_next(&bs->flush_queue);
151
- qemu_co_mutex_unlock(&bs->reqs_lock);
152
+ qemu_mutex_unlock(&bs->reqs_lock);
153
154
early_exit:
155
bdrv_dec_in_flight(bs);
156
--
210
--
157
2.41.0
211
2.13.6
212
213
diff view generated by jsdifflib
1
From: Philippe Mathieu-Daudé <philmd@redhat.com>
2
3
Use autofree heap allocation instead of variable-length array on the
4
stack. Here we don't expect the bitmap size to be enormous, and
5
since we're about to read/write it to disk the overhead of the
6
allocation should be fine.
7
8
The codebase has very few VLAs, and if we can get rid of them all we
9
can make the compiler error on new additions. This is a defensive
10
measure against security bugs where an on-stack dynamic allocation
11
isn't correctly size-checked (e.g. CVE-2021-3527).
12
13
Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
14
[PMM: expanded commit message]
15
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
16
Message-ID: <20230811175229.808139-1-peter.maydell@linaro.org>
17
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
18
Reviewed-by: Francisco Iglesias <frasse.iglesias@gmail.com>
19
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
1
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
20
---
2
---
21
block/vpc.c | 4 ++--
3
tests/test-bdrv-drain.c | 80 +++++++++++++++++++++++++++++++++++++++++++++++++
22
1 file changed, 2 insertions(+), 2 deletions(-)
4
1 file changed, 80 insertions(+)
23
5
24
diff --git a/block/vpc.c b/block/vpc.c
6
diff --git a/tests/test-bdrv-drain.c b/tests/test-bdrv-drain.c
25
index XXXXXXX..XXXXXXX 100644
7
index XXXXXXX..XXXXXXX 100644
26
--- a/block/vpc.c
8
--- a/tests/test-bdrv-drain.c
27
+++ b/block/vpc.c
9
+++ b/tests/test-bdrv-drain.c
28
@@ -XXX,XX +XXX,XX @@ get_image_offset(BlockDriverState *bs, uint64_t offset, bool write, int *err)
10
@@ -XXX,XX +XXX,XX @@ static void test_multiparent(void)
29
miss sparse read optimization, but it's not a problem in terms of
11
blk_unref(blk_b);
30
correctness. */
12
}
31
if (write && (s->last_bitmap_offset != bitmap_offset)) {
13
32
- uint8_t bitmap[s->bitmap_size];
14
+static void test_graph_change(void)
33
+ g_autofree uint8_t *bitmap = g_malloc(s->bitmap_size);
15
+{
34
int r;
16
+ BlockBackend *blk_a, *blk_b;
35
17
+ BlockDriverState *bs_a, *bs_b, *backing;
36
s->last_bitmap_offset = bitmap_offset;
18
+ BDRVTestState *a_s, *b_s, *backing_s;
37
@@ -XXX,XX +XXX,XX @@ alloc_block(BlockDriverState *bs, int64_t offset)
19
+
38
int64_t bat_offset;
20
+ blk_a = blk_new(BLK_PERM_ALL, BLK_PERM_ALL);
39
uint32_t index, bat_value;
21
+ bs_a = bdrv_new_open_driver(&bdrv_test, "test-node-a", BDRV_O_RDWR,
40
int ret;
22
+ &error_abort);
41
- uint8_t bitmap[s->bitmap_size];
23
+ a_s = bs_a->opaque;
42
+ g_autofree uint8_t *bitmap = g_malloc(s->bitmap_size);
24
+ blk_insert_bs(blk_a, bs_a, &error_abort);
43
25
+
44
/* Check if sector_num is valid */
26
+ blk_b = blk_new(BLK_PERM_ALL, BLK_PERM_ALL);
45
if ((offset < 0) || (offset > bs->total_sectors * BDRV_SECTOR_SIZE)) {
27
+ bs_b = bdrv_new_open_driver(&bdrv_test, "test-node-b", BDRV_O_RDWR,
28
+ &error_abort);
29
+ b_s = bs_b->opaque;
30
+ blk_insert_bs(blk_b, bs_b, &error_abort);
31
+
32
+ backing = bdrv_new_open_driver(&bdrv_test, "backing", 0, &error_abort);
33
+ backing_s = backing->opaque;
34
+ bdrv_set_backing_hd(bs_a, backing, &error_abort);
35
+
36
+ g_assert_cmpint(bs_a->quiesce_counter, ==, 0);
37
+ g_assert_cmpint(bs_b->quiesce_counter, ==, 0);
38
+ g_assert_cmpint(backing->quiesce_counter, ==, 0);
39
+ g_assert_cmpint(a_s->drain_count, ==, 0);
40
+ g_assert_cmpint(b_s->drain_count, ==, 0);
41
+ g_assert_cmpint(backing_s->drain_count, ==, 0);
42
+
43
+ do_drain_begin(BDRV_SUBTREE_DRAIN, bs_a);
44
+ do_drain_begin(BDRV_SUBTREE_DRAIN, bs_a);
45
+ do_drain_begin(BDRV_SUBTREE_DRAIN, bs_a);
46
+ do_drain_begin(BDRV_SUBTREE_DRAIN, bs_b);
47
+ do_drain_begin(BDRV_SUBTREE_DRAIN, bs_b);
48
+
49
+ bdrv_set_backing_hd(bs_b, backing, &error_abort);
50
+ g_assert_cmpint(bs_a->quiesce_counter, ==, 5);
51
+ g_assert_cmpint(bs_b->quiesce_counter, ==, 5);
52
+ g_assert_cmpint(backing->quiesce_counter, ==, 5);
53
+ g_assert_cmpint(a_s->drain_count, ==, 5);
54
+ g_assert_cmpint(b_s->drain_count, ==, 5);
55
+ g_assert_cmpint(backing_s->drain_count, ==, 5);
56
+
57
+ bdrv_set_backing_hd(bs_b, NULL, &error_abort);
58
+ g_assert_cmpint(bs_a->quiesce_counter, ==, 3);
59
+ g_assert_cmpint(bs_b->quiesce_counter, ==, 2);
60
+ g_assert_cmpint(backing->quiesce_counter, ==, 3);
61
+ g_assert_cmpint(a_s->drain_count, ==, 3);
62
+ g_assert_cmpint(b_s->drain_count, ==, 2);
63
+ g_assert_cmpint(backing_s->drain_count, ==, 3);
64
+
65
+ bdrv_set_backing_hd(bs_b, backing, &error_abort);
66
+ g_assert_cmpint(bs_a->quiesce_counter, ==, 5);
67
+ g_assert_cmpint(bs_b->quiesce_counter, ==, 5);
68
+ g_assert_cmpint(backing->quiesce_counter, ==, 5);
69
+ g_assert_cmpint(a_s->drain_count, ==, 5);
70
+ g_assert_cmpint(b_s->drain_count, ==, 5);
71
+ g_assert_cmpint(backing_s->drain_count, ==, 5);
72
+
73
+ do_drain_end(BDRV_SUBTREE_DRAIN, bs_b);
74
+ do_drain_end(BDRV_SUBTREE_DRAIN, bs_b);
75
+ do_drain_end(BDRV_SUBTREE_DRAIN, bs_a);
76
+ do_drain_end(BDRV_SUBTREE_DRAIN, bs_a);
77
+ do_drain_end(BDRV_SUBTREE_DRAIN, bs_a);
78
+
79
+ g_assert_cmpint(bs_a->quiesce_counter, ==, 0);
80
+ g_assert_cmpint(bs_b->quiesce_counter, ==, 0);
81
+ g_assert_cmpint(backing->quiesce_counter, ==, 0);
82
+ g_assert_cmpint(a_s->drain_count, ==, 0);
83
+ g_assert_cmpint(b_s->drain_count, ==, 0);
84
+ g_assert_cmpint(backing_s->drain_count, ==, 0);
85
+
86
+ bdrv_unref(backing);
87
+ bdrv_unref(bs_a);
88
+ bdrv_unref(bs_b);
89
+ blk_unref(blk_a);
90
+ blk_unref(blk_b);
91
+}
92
+
93
94
typedef struct TestBlockJob {
95
BlockJob common;
96
@@ -XXX,XX +XXX,XX @@ int main(int argc, char **argv)
97
98
g_test_add_func("/bdrv-drain/nested", test_nested);
99
g_test_add_func("/bdrv-drain/multiparent", test_multiparent);
100
+ g_test_add_func("/bdrv-drain/graph-change", test_graph_change);
101
102
g_test_add_func("/bdrv-drain/blockjob/drain_all", test_blockjob_drain_all);
103
g_test_add_func("/bdrv-drain/blockjob/drain", test_blockjob_drain);
46
--
104
--
47
2.41.0
105
2.13.6
48
106
49
107
diff view generated by jsdifflib
New patch
1
Since commit bde70715, base is the only node that is reopened in
2
commit_start(). This means that the code, which still involves an
3
explicit BlockReopenQueue, can now be simplified by using bdrv_reopen().
1
4
5
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
6
Reviewed-by: Fam Zheng <famz@redhat.com>
7
---
8
block/commit.c | 8 +-------
9
1 file changed, 1 insertion(+), 7 deletions(-)
10
11
diff --git a/block/commit.c b/block/commit.c
12
index XXXXXXX..XXXXXXX 100644
13
--- a/block/commit.c
14
+++ b/block/commit.c
15
@@ -XXX,XX +XXX,XX @@ void commit_start(const char *job_id, BlockDriverState *bs,
16
const char *filter_node_name, Error **errp)
17
{
18
CommitBlockJob *s;
19
- BlockReopenQueue *reopen_queue = NULL;
20
int orig_base_flags;
21
BlockDriverState *iter;
22
BlockDriverState *commit_top_bs = NULL;
23
@@ -XXX,XX +XXX,XX @@ void commit_start(const char *job_id, BlockDriverState *bs,
24
/* convert base to r/w, if necessary */
25
orig_base_flags = bdrv_get_flags(base);
26
if (!(orig_base_flags & BDRV_O_RDWR)) {
27
- reopen_queue = bdrv_reopen_queue(reopen_queue, base, NULL,
28
- orig_base_flags | BDRV_O_RDWR);
29
- }
30
-
31
- if (reopen_queue) {
32
- bdrv_reopen_multiple(bdrv_get_aio_context(bs), reopen_queue, &local_err);
33
+ bdrv_reopen(base, orig_base_flags | BDRV_O_RDWR, &local_err);
34
if (local_err != NULL) {
35
error_propagate(errp, local_err);
36
goto fail;
37
--
38
2.13.6
39
40
diff view generated by jsdifflib
New patch
1
The bdrv_reopen*() implementation doesn't like it if the graph is
2
changed between queuing nodes for reopen and actually reopening them
3
(one of the reasons is that queuing can be recursive).
1
4
5
So instead of draining the device only in bdrv_reopen_multiple(),
6
require that callers already drained all affected nodes, and assert this
7
in bdrv_reopen_queue().
8
9
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
10
Reviewed-by: Fam Zheng <famz@redhat.com>
11
---
12
block.c | 23 ++++++++++++++++-------
13
block/replication.c | 6 ++++++
14
qemu-io-cmds.c | 3 +++
15
3 files changed, 25 insertions(+), 7 deletions(-)
16
17
diff --git a/block.c b/block.c
18
index XXXXXXX..XXXXXXX 100644
19
--- a/block.c
20
+++ b/block.c
21
@@ -XXX,XX +XXX,XX @@ BlockDriverState *bdrv_open(const char *filename, const char *reference,
22
* returns a pointer to bs_queue, which is either the newly allocated
23
* bs_queue, or the existing bs_queue being used.
24
*
25
+ * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
26
*/
27
static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
28
BlockDriverState *bs,
29
@@ -XXX,XX +XXX,XX @@ static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
30
BdrvChild *child;
31
QDict *old_options, *explicit_options;
32
33
+ /* Make sure that the caller remembered to use a drained section. This is
34
+ * important to avoid graph changes between the recursive queuing here and
35
+ * bdrv_reopen_multiple(). */
36
+ assert(bs->quiesce_counter > 0);
37
+
38
if (bs_queue == NULL) {
39
bs_queue = g_new0(BlockReopenQueue, 1);
40
QSIMPLEQ_INIT(bs_queue);
41
@@ -XXX,XX +XXX,XX @@ BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
42
* If all devices prepare successfully, then the changes are committed
43
* to all devices.
44
*
45
+ * All affected nodes must be drained between bdrv_reopen_queue() and
46
+ * bdrv_reopen_multiple().
47
*/
48
int bdrv_reopen_multiple(AioContext *ctx, BlockReopenQueue *bs_queue, Error **errp)
49
{
50
@@ -XXX,XX +XXX,XX @@ int bdrv_reopen_multiple(AioContext *ctx, BlockReopenQueue *bs_queue, Error **er
51
52
assert(bs_queue != NULL);
53
54
- aio_context_release(ctx);
55
- bdrv_drain_all_begin();
56
- aio_context_acquire(ctx);
57
-
58
QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
59
+ assert(bs_entry->state.bs->quiesce_counter > 0);
60
if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, &local_err)) {
61
error_propagate(errp, local_err);
62
goto cleanup;
63
@@ -XXX,XX +XXX,XX @@ cleanup:
64
}
65
g_free(bs_queue);
66
67
- bdrv_drain_all_end();
68
-
69
return ret;
70
}
71
72
@@ -XXX,XX +XXX,XX @@ int bdrv_reopen(BlockDriverState *bs, int bdrv_flags, Error **errp)
73
{
74
int ret = -1;
75
Error *local_err = NULL;
76
- BlockReopenQueue *queue = bdrv_reopen_queue(NULL, bs, NULL, bdrv_flags);
77
+ BlockReopenQueue *queue;
78
79
+ bdrv_subtree_drained_begin(bs);
80
+
81
+ queue = bdrv_reopen_queue(NULL, bs, NULL, bdrv_flags);
82
ret = bdrv_reopen_multiple(bdrv_get_aio_context(bs), queue, &local_err);
83
if (local_err != NULL) {
84
error_propagate(errp, local_err);
85
}
86
+
87
+ bdrv_subtree_drained_end(bs);
88
+
89
return ret;
90
}
91
92
diff --git a/block/replication.c b/block/replication.c
93
index XXXXXXX..XXXXXXX 100644
94
--- a/block/replication.c
95
+++ b/block/replication.c
96
@@ -XXX,XX +XXX,XX @@ static void reopen_backing_file(BlockDriverState *bs, bool writable,
97
new_secondary_flags = s->orig_secondary_flags;
98
}
99
100
+ bdrv_subtree_drained_begin(s->hidden_disk->bs);
101
+ bdrv_subtree_drained_begin(s->secondary_disk->bs);
102
+
103
if (orig_hidden_flags != new_hidden_flags) {
104
reopen_queue = bdrv_reopen_queue(reopen_queue, s->hidden_disk->bs, NULL,
105
new_hidden_flags);
106
@@ -XXX,XX +XXX,XX @@ static void reopen_backing_file(BlockDriverState *bs, bool writable,
107
reopen_queue, &local_err);
108
error_propagate(errp, local_err);
109
}
110
+
111
+ bdrv_subtree_drained_end(s->hidden_disk->bs);
112
+ bdrv_subtree_drained_end(s->secondary_disk->bs);
113
}
114
115
static void backup_job_cleanup(BlockDriverState *bs)
116
diff --git a/qemu-io-cmds.c b/qemu-io-cmds.c
117
index XXXXXXX..XXXXXXX 100644
118
--- a/qemu-io-cmds.c
119
+++ b/qemu-io-cmds.c
120
@@ -XXX,XX +XXX,XX @@ static int reopen_f(BlockBackend *blk, int argc, char **argv)
121
opts = qopts ? qemu_opts_to_qdict(qopts, NULL) : NULL;
122
qemu_opts_reset(&reopen_opts);
123
124
+ bdrv_subtree_drained_begin(bs);
125
brq = bdrv_reopen_queue(NULL, bs, opts, flags);
126
bdrv_reopen_multiple(bdrv_get_aio_context(bs), brq, &local_err);
127
+ bdrv_subtree_drained_end(bs);
128
+
129
if (local_err) {
130
error_report_err(local_err);
131
} else {
132
--
133
2.13.6
134
135
diff view generated by jsdifflib