1 | The following changes since commit ba29883206d92a29ad5a466e679ccfc2ee6132ef: | 1 | The following changes since commit 028f2361d0c2d28d6f918fe618f389228ac22b60: |
---|---|---|---|
2 | 2 | ||
3 | Merge remote-tracking branch 'remotes/borntraeger/tags/s390x-20200310' into staging (2020-03-10 16:50:28 +0000) | 3 | Merge tag 'pull-target-arm-20220609' of https://git.linaro.org/people/pmaydell/qemu-arm into staging (2022-06-09 06:47:03 -0700) |
4 | 4 | ||
5 | are available in the Git repository at: | 5 | are available in the Git repository at: |
6 | 6 | ||
7 | git://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 8bb3b023f2055054ee119cb45b42d2b14be7fc8a: | 9 | for you to fetch changes up to 7f9a8b3342ff00d3398fdc08264948762d748edb: |
10 | 10 | ||
11 | qemu-iotests: adding LUKS cleanup for non-UTF8 secret error (2020-03-11 15:54:38 +0100) | 11 | nbd: Drop dead code spotted by Coverity (2022-06-09 18:07:17 +0200) |
12 | 12 | ||
13 | ---------------------------------------------------------------- | 13 | ---------------------------------------------------------------- |
14 | Block layer patches: | 14 | Block layer patches |
15 | 15 | ||
16 | - Relax restrictions for blockdev-snapshot (allows libvirt to do live | 16 | - Add vduse-blk export |
17 | storage migration with blockdev-mirror) | 17 | - Dirty bitmaps: Fix and improve bitmap merge |
18 | - luks: Delete created files when block_crypto_co_create_opts_luks fails | 18 | - gluster: correctly set max_pdiscard |
19 | - Fix memleaks in qmp_object_add | 19 | - rbd: report a better error when namespace does not exist |
20 | - aio_wait_kick: add missing memory barrier | ||
21 | - Code cleanups | ||
20 | 22 | ||
21 | ---------------------------------------------------------------- | 23 | ---------------------------------------------------------------- |
22 | Daniel Henrique Barboza (4): | 24 | Emanuele Giuseppe Esposito (1): |
23 | block: introducing 'bdrv_co_delete_file' interface | 25 | aio_wait_kick: add missing memory barrier |
24 | block.c: adding bdrv_co_delete_file | ||
25 | crypto.c: cleanup created file when block_crypto_co_create_opts_luks fails | ||
26 | qemu-iotests: adding LUKS cleanup for non-UTF8 secret error | ||
27 | 26 | ||
28 | Kevin Wolf (6): | 27 | Eric Blake (1): |
29 | block: Make bdrv_get_cumulative_perm() public | 28 | nbd: Drop dead code spotted by Coverity |
30 | block: Relax restrictions for blockdev-snapshot | ||
31 | iotests: Fix run_job() with use_log=False | ||
32 | iotests: Test mirror with temporarily disabled target backing file | ||
33 | block: Fix cross-AioContext blockdev-snapshot | ||
34 | iotests: Add iothread cases to 155 | ||
35 | 29 | ||
36 | Pan Nengyuan (1): | 30 | Fabian Ebner (1): |
37 | qom-qmp-cmds: fix two memleaks in qmp_object_add | 31 | block/gluster: correctly set max_pdiscard |
38 | 32 | ||
39 | Peter Krempa (1): | 33 | Stefan Hajnoczi (3): |
40 | qapi: Add '@allow-write-only-overlay' feature for 'blockdev-snapshot' | 34 | block: drop unused bdrv_co_drain() API |
35 | block: get rid of blk->guest_block_size | ||
36 | qsd: document vduse-blk exports | ||
41 | 37 | ||
42 | Philippe Mathieu-Daudé (1): | 38 | Stefano Garzarella (1): |
43 | tests/qemu-iotests: Fix socket_scm_helper build path | 39 | block/rbd: report a better error when namespace does not exist |
44 | 40 | ||
45 | qapi/block-core.json | 9 ++++- | 41 | Vladimir Sementsov-Ogievskiy (3): |
46 | include/block/block.h | 1 + | 42 | block: block_dirty_bitmap_merge(): fix error path |
47 | include/block/block_int.h | 7 ++++ | 43 | block: improve block_dirty_bitmap_merge(): don't allocate extra bitmap |
48 | block.c | 33 ++++++++++++++-- | 44 | block: simplify handling of try to merge different sized bitmaps |
49 | block/crypto.c | 18 +++++++++ | ||
50 | block/file-posix.c | 23 +++++++++++ | ||
51 | blockdev.c | 30 ++++----------- | ||
52 | qom/qom-qmp-cmds.c | 16 +++----- | ||
53 | tests/qemu-iotests/iotests.py | 5 ++- | ||
54 | tests/Makefile.include | 1 + | ||
55 | tests/qemu-iotests/085.out | 4 +- | ||
56 | tests/qemu-iotests/155 | 88 ++++++++++++++++++++++++++++++++++++------- | ||
57 | tests/qemu-iotests/155.out | 4 +- | ||
58 | tests/qemu-iotests/282 | 67 ++++++++++++++++++++++++++++++++ | ||
59 | tests/qemu-iotests/282.out | 11 ++++++ | ||
60 | tests/qemu-iotests/group | 1 + | ||
61 | tests/qtest/Makefile.include | 1 - | ||
62 | 17 files changed, 262 insertions(+), 57 deletions(-) | ||
63 | create mode 100755 tests/qemu-iotests/282 | ||
64 | create mode 100644 tests/qemu-iotests/282.out | ||
65 | 45 | ||
46 | Xie Yongji (8): | ||
47 | block: Support passing NULL ops to blk_set_dev_ops() | ||
48 | block/export: Fix incorrect length passed to vu_queue_push() | ||
49 | block/export: Abstract out the logic of virtio-blk I/O process | ||
50 | linux-headers: Add vduse.h | ||
51 | libvduse: Add VDUSE (vDPA Device in Userspace) library | ||
52 | vduse-blk: Implement vduse-blk export | ||
53 | vduse-blk: Add vduse-blk resize support | ||
54 | libvduse: Add support for reconnecting | ||
66 | 55 | ||
56 | qapi/block-export.json | 28 +- | ||
57 | docs/tools/qemu-storage-daemon.rst | 21 + | ||
58 | meson_options.txt | 4 + | ||
59 | block/export/vduse-blk.h | 20 + | ||
60 | block/export/virtio-blk-handler.h | 37 + | ||
61 | include/block/aio-wait.h | 2 + | ||
62 | include/block/block-io.h | 1 - | ||
63 | include/block/block_int-io.h | 2 +- | ||
64 | include/qemu/hbitmap.h | 15 +- | ||
65 | include/sysemu/block-backend-io.h | 1 - | ||
66 | linux-headers/linux/vduse.h | 306 ++++++ | ||
67 | subprojects/libvduse/include/atomic.h | 1 + | ||
68 | subprojects/libvduse/include/compiler.h | 1 + | ||
69 | subprojects/libvduse/libvduse.h | 247 +++++ | ||
70 | block/backup.c | 6 +- | ||
71 | block/block-backend.c | 12 +- | ||
72 | block/dirty-bitmap.c | 26 +- | ||
73 | block/export/export.c | 6 + | ||
74 | block/export/vduse-blk.c | 341 +++++++ | ||
75 | block/export/vhost-user-blk-server.c | 261 +---- | ||
76 | block/export/virtio-blk-handler.c | 240 +++++ | ||
77 | block/gluster.c | 2 +- | ||
78 | block/io.c | 15 - | ||
79 | block/monitor/bitmap-qmp-cmds.c | 40 +- | ||
80 | block/nbd.c | 8 +- | ||
81 | block/rbd.c | 24 + | ||
82 | hw/block/virtio-blk.c | 1 - | ||
83 | hw/block/xen-block.c | 1 - | ||
84 | hw/ide/core.c | 1 - | ||
85 | hw/scsi/scsi-disk.c | 1 - | ||
86 | hw/scsi/scsi-generic.c | 1 - | ||
87 | storage-daemon/qemu-storage-daemon.c | 9 + | ||
88 | subprojects/libvduse/libvduse.c | 1392 +++++++++++++++++++++++++++ | ||
89 | util/aio-wait.c | 16 +- | ||
90 | util/hbitmap.c | 25 +- | ||
91 | MAINTAINERS | 9 + | ||
92 | block/export/meson.build | 7 +- | ||
93 | meson.build | 34 + | ||
94 | scripts/meson-buildoptions.sh | 7 + | ||
95 | scripts/update-linux-headers.sh | 2 +- | ||
96 | subprojects/libvduse/linux-headers/linux | 1 + | ||
97 | subprojects/libvduse/meson.build | 10 + | ||
98 | subprojects/libvduse/standard-headers/linux | 1 + | ||
99 | 43 files changed, 2830 insertions(+), 355 deletions(-) | ||
100 | create mode 100644 block/export/vduse-blk.h | ||
101 | create mode 100644 block/export/virtio-blk-handler.h | ||
102 | create mode 100644 linux-headers/linux/vduse.h | ||
103 | create mode 120000 subprojects/libvduse/include/atomic.h | ||
104 | create mode 120000 subprojects/libvduse/include/compiler.h | ||
105 | create mode 100644 subprojects/libvduse/libvduse.h | ||
106 | create mode 100644 block/export/vduse-blk.c | ||
107 | create mode 100644 block/export/virtio-blk-handler.c | ||
108 | create mode 100644 subprojects/libvduse/libvduse.c | ||
109 | create mode 120000 subprojects/libvduse/linux-headers/linux | ||
110 | create mode 100644 subprojects/libvduse/meson.build | ||
111 | create mode 120000 subprojects/libvduse/standard-headers/linux | diff view generated by jsdifflib |
New patch | |||
---|---|---|---|
1 | From: Stefan Hajnoczi <stefanha@redhat.com> | ||
1 | 2 | ||
3 | bdrv_co_drain() has not been used since commit 9a0cec664eef ("mirror: | ||
4 | use bdrv_drained_begin/bdrv_drained_end") in 2016. Remove it so there | ||
5 | are fewer drain scenarios to worry about. | ||
6 | |||
7 | Use bdrv_drained_begin()/bdrv_drained_end() instead. They are "mixed" | ||
8 | functions that can be called from coroutine context. Unlike | ||
9 | bdrv_co_drain(), these functions provide control of the length of the | ||
10 | drained section, which is usually the right thing. | ||
11 | |||
12 | Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com> | ||
13 | Message-Id: <20220521122714.3837731-1-stefanha@redhat.com> | ||
14 | Reviewed-by: Emanuele Giuseppe Esposito <eesposit@redhat.com> | ||
15 | Reviewed-by: Alberto Faria <afaria@redhat.com> | ||
16 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> | ||
17 | --- | ||
18 | include/block/block-io.h | 1 - | ||
19 | block/io.c | 15 --------------- | ||
20 | 2 files changed, 16 deletions(-) | ||
21 | |||
22 | diff --git a/include/block/block-io.h b/include/block/block-io.h | ||
23 | index XXXXXXX..XXXXXXX 100644 | ||
24 | --- a/include/block/block-io.h | ||
25 | +++ b/include/block/block-io.h | ||
26 | @@ -XXX,XX +XXX,XX @@ void bdrv_drained_end_no_poll(BlockDriverState *bs, int *drained_end_counter); | ||
27 | cond); }) | ||
28 | |||
29 | void bdrv_drain(BlockDriverState *bs); | ||
30 | -void coroutine_fn bdrv_co_drain(BlockDriverState *bs); | ||
31 | |||
32 | int generated_co_wrapper | ||
33 | bdrv_truncate(BdrvChild *child, int64_t offset, bool exact, | ||
34 | diff --git a/block/io.c b/block/io.c | ||
35 | index XXXXXXX..XXXXXXX 100644 | ||
36 | --- a/block/io.c | ||
37 | +++ b/block/io.c | ||
38 | @@ -XXX,XX +XXX,XX @@ void bdrv_unapply_subtree_drain(BdrvChild *child, BlockDriverState *old_parent) | ||
39 | BDRV_POLL_WHILE(child->bs, qatomic_read(&drained_end_counter) > 0); | ||
40 | } | ||
41 | |||
42 | -/* | ||
43 | - * Wait for pending requests to complete on a single BlockDriverState subtree, | ||
44 | - * and suspend block driver's internal I/O until next request arrives. | ||
45 | - * | ||
46 | - * Note that unlike bdrv_drain_all(), the caller must hold the BlockDriverState | ||
47 | - * AioContext. | ||
48 | - */ | ||
49 | -void coroutine_fn bdrv_co_drain(BlockDriverState *bs) | ||
50 | -{ | ||
51 | - IO_OR_GS_CODE(); | ||
52 | - assert(qemu_in_coroutine()); | ||
53 | - bdrv_drained_begin(bs); | ||
54 | - bdrv_drained_end(bs); | ||
55 | -} | ||
56 | - | ||
57 | void bdrv_drain(BlockDriverState *bs) | ||
58 | { | ||
59 | IO_OR_GS_CODE(); | ||
60 | -- | ||
61 | 2.35.3 | diff view generated by jsdifflib |
New patch | |||
---|---|---|---|
1 | From: Stefan Hajnoczi <stefanha@redhat.com> | ||
1 | 2 | ||
3 | Commit 1b7fd729559c ("block: rename buffer_alignment to | ||
4 | guest_block_size") noted: | ||
5 | |||
6 | At this point, the field is set by the device emulation, but completely | ||
7 | ignored by the block layer. | ||
8 | |||
9 | The last time the value of buffer_alignment/guest_block_size was | ||
10 | actually used was before commit 339064d50639 ("block: Don't use guest | ||
11 | sector size for qemu_blockalign()"). | ||
12 | |||
13 | This value has not been used since 2013. Get rid of it. | ||
14 | |||
15 | Cc: Xie Yongji <xieyongji@bytedance.com> | ||
16 | Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com> | ||
17 | Message-Id: <20220518130945.2657905-1-stefanha@redhat.com> | ||
18 | Reviewed-by: Paul Durrant <paul@xen.org> | ||
19 | Reviewed-by: Eric Blake <eblake@redhat.com> | ||
20 | Reviewed-by: Alberto Faria <afaria@redhat.com> | ||
21 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> | ||
22 | --- | ||
23 | include/sysemu/block-backend-io.h | 1 - | ||
24 | block/block-backend.c | 10 ---------- | ||
25 | block/export/vhost-user-blk-server.c | 1 - | ||
26 | hw/block/virtio-blk.c | 1 - | ||
27 | hw/block/xen-block.c | 1 - | ||
28 | hw/ide/core.c | 1 - | ||
29 | hw/scsi/scsi-disk.c | 1 - | ||
30 | hw/scsi/scsi-generic.c | 1 - | ||
31 | 8 files changed, 17 deletions(-) | ||
32 | |||
33 | diff --git a/include/sysemu/block-backend-io.h b/include/sysemu/block-backend-io.h | ||
34 | index XXXXXXX..XXXXXXX 100644 | ||
35 | --- a/include/sysemu/block-backend-io.h | ||
36 | +++ b/include/sysemu/block-backend-io.h | ||
37 | @@ -XXX,XX +XXX,XX @@ void blk_error_action(BlockBackend *blk, BlockErrorAction action, | ||
38 | void blk_iostatus_set_err(BlockBackend *blk, int error); | ||
39 | int blk_get_max_iov(BlockBackend *blk); | ||
40 | int blk_get_max_hw_iov(BlockBackend *blk); | ||
41 | -void blk_set_guest_block_size(BlockBackend *blk, int align); | ||
42 | |||
43 | void blk_io_plug(BlockBackend *blk); | ||
44 | void blk_io_unplug(BlockBackend *blk); | ||
45 | diff --git a/block/block-backend.c b/block/block-backend.c | ||
46 | index XXXXXXX..XXXXXXX 100644 | ||
47 | --- a/block/block-backend.c | ||
48 | +++ b/block/block-backend.c | ||
49 | @@ -XXX,XX +XXX,XX @@ struct BlockBackend { | ||
50 | const BlockDevOps *dev_ops; | ||
51 | void *dev_opaque; | ||
52 | |||
53 | - /* the block size for which the guest device expects atomicity */ | ||
54 | - int guest_block_size; | ||
55 | - | ||
56 | /* If the BDS tree is removed, some of its options are stored here (which | ||
57 | * can be used to restore those options in the new BDS on insert) */ | ||
58 | BlockBackendRootState root_state; | ||
59 | @@ -XXX,XX +XXX,XX @@ void blk_detach_dev(BlockBackend *blk, DeviceState *dev) | ||
60 | blk->dev = NULL; | ||
61 | blk->dev_ops = NULL; | ||
62 | blk->dev_opaque = NULL; | ||
63 | - blk->guest_block_size = 512; | ||
64 | blk_set_perm(blk, 0, BLK_PERM_ALL, &error_abort); | ||
65 | blk_unref(blk); | ||
66 | } | ||
67 | @@ -XXX,XX +XXX,XX @@ int blk_get_max_iov(BlockBackend *blk) | ||
68 | return blk->root->bs->bl.max_iov; | ||
69 | } | ||
70 | |||
71 | -void blk_set_guest_block_size(BlockBackend *blk, int align) | ||
72 | -{ | ||
73 | - IO_CODE(); | ||
74 | - blk->guest_block_size = align; | ||
75 | -} | ||
76 | - | ||
77 | void *blk_try_blockalign(BlockBackend *blk, size_t size) | ||
78 | { | ||
79 | IO_CODE(); | ||
80 | diff --git a/block/export/vhost-user-blk-server.c b/block/export/vhost-user-blk-server.c | ||
81 | index XXXXXXX..XXXXXXX 100644 | ||
82 | --- a/block/export/vhost-user-blk-server.c | ||
83 | +++ b/block/export/vhost-user-blk-server.c | ||
84 | @@ -XXX,XX +XXX,XX @@ static int vu_blk_exp_create(BlockExport *exp, BlockExportOptions *opts, | ||
85 | return -EINVAL; | ||
86 | } | ||
87 | vexp->blk_size = logical_block_size; | ||
88 | - blk_set_guest_block_size(exp->blk, logical_block_size); | ||
89 | |||
90 | if (vu_opts->has_num_queues) { | ||
91 | num_queues = vu_opts->num_queues; | ||
92 | diff --git a/hw/block/virtio-blk.c b/hw/block/virtio-blk.c | ||
93 | index XXXXXXX..XXXXXXX 100644 | ||
94 | --- a/hw/block/virtio-blk.c | ||
95 | +++ b/hw/block/virtio-blk.c | ||
96 | @@ -XXX,XX +XXX,XX @@ static void virtio_blk_device_realize(DeviceState *dev, Error **errp) | ||
97 | |||
98 | s->change = qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s); | ||
99 | blk_set_dev_ops(s->blk, &virtio_block_ops, s); | ||
100 | - blk_set_guest_block_size(s->blk, s->conf.conf.logical_block_size); | ||
101 | |||
102 | blk_iostatus_enable(s->blk); | ||
103 | |||
104 | diff --git a/hw/block/xen-block.c b/hw/block/xen-block.c | ||
105 | index XXXXXXX..XXXXXXX 100644 | ||
106 | --- a/hw/block/xen-block.c | ||
107 | +++ b/hw/block/xen-block.c | ||
108 | @@ -XXX,XX +XXX,XX @@ static void xen_block_realize(XenDevice *xendev, Error **errp) | ||
109 | } | ||
110 | |||
111 | blk_set_dev_ops(blk, &xen_block_dev_ops, blockdev); | ||
112 | - blk_set_guest_block_size(blk, conf->logical_block_size); | ||
113 | |||
114 | if (conf->discard_granularity == -1) { | ||
115 | conf->discard_granularity = conf->physical_block_size; | ||
116 | diff --git a/hw/ide/core.c b/hw/ide/core.c | ||
117 | index XXXXXXX..XXXXXXX 100644 | ||
118 | --- a/hw/ide/core.c | ||
119 | +++ b/hw/ide/core.c | ||
120 | @@ -XXX,XX +XXX,XX @@ int ide_init_drive(IDEState *s, BlockBackend *blk, IDEDriveKind kind, | ||
121 | s->smart_selftest_count = 0; | ||
122 | if (kind == IDE_CD) { | ||
123 | blk_set_dev_ops(blk, &ide_cd_block_ops, s); | ||
124 | - blk_set_guest_block_size(blk, 2048); | ||
125 | } else { | ||
126 | if (!blk_is_inserted(s->blk)) { | ||
127 | error_setg(errp, "Device needs media, but drive is empty"); | ||
128 | diff --git a/hw/scsi/scsi-disk.c b/hw/scsi/scsi-disk.c | ||
129 | index XXXXXXX..XXXXXXX 100644 | ||
130 | --- a/hw/scsi/scsi-disk.c | ||
131 | +++ b/hw/scsi/scsi-disk.c | ||
132 | @@ -XXX,XX +XXX,XX @@ static void scsi_realize(SCSIDevice *dev, Error **errp) | ||
133 | } else { | ||
134 | blk_set_dev_ops(s->qdev.conf.blk, &scsi_disk_block_ops, s); | ||
135 | } | ||
136 | - blk_set_guest_block_size(s->qdev.conf.blk, s->qdev.blocksize); | ||
137 | |||
138 | blk_iostatus_enable(s->qdev.conf.blk); | ||
139 | |||
140 | diff --git a/hw/scsi/scsi-generic.c b/hw/scsi/scsi-generic.c | ||
141 | index XXXXXXX..XXXXXXX 100644 | ||
142 | --- a/hw/scsi/scsi-generic.c | ||
143 | +++ b/hw/scsi/scsi-generic.c | ||
144 | @@ -XXX,XX +XXX,XX @@ static void scsi_read_complete(void * opaque, int ret) | ||
145 | s->blocksize = ldl_be_p(&r->buf[8]); | ||
146 | s->max_lba = ldq_be_p(&r->buf[0]); | ||
147 | } | ||
148 | - blk_set_guest_block_size(s->conf.blk, s->blocksize); | ||
149 | |||
150 | /* | ||
151 | * Patch MODE SENSE device specific parameters if the BDS is opened | ||
152 | -- | ||
153 | 2.35.3 | diff view generated by jsdifflib |
1 | The 'job-complete' QMP command should be run with qmp() rather than | 1 | From: Vladimir Sementsov-Ogievskiy <vsementsov@openvz.org> |
---|---|---|---|
2 | qmp_log() if use_log=False is passed. | ||
3 | 2 | ||
4 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> | 3 | At the end we ignore failure of bdrv_merge_dirty_bitmap() and report |
5 | Message-Id: <20200310113831.27293-4-kwolf@redhat.com> | 4 | success. And still set errp. That's wrong. |
6 | Reviewed-by: Peter Krempa <pkrempa@redhat.com> | 5 | |
6 | Signed-off-by: Vladimir Sementsov-Ogievskiy <v.sementsov-og@mail.ru> | ||
7 | Reviewed-by: Nikita Lapshin <nikita.lapshin@virtuozzo.com> | ||
8 | Reviewed-by: Kevin Wolf <kwolf@redhat.com> | ||
9 | Message-Id: <20220517111206.23585-2-v.sementsov-og@mail.ru> | ||
7 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> | 10 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> |
8 | --- | 11 | --- |
9 | tests/qemu-iotests/iotests.py | 5 ++++- | 12 | block/monitor/bitmap-qmp-cmds.c | 5 ++++- |
10 | 1 file changed, 4 insertions(+), 1 deletion(-) | 13 | 1 file changed, 4 insertions(+), 1 deletion(-) |
11 | 14 | ||
12 | diff --git a/tests/qemu-iotests/iotests.py b/tests/qemu-iotests/iotests.py | 15 | diff --git a/block/monitor/bitmap-qmp-cmds.c b/block/monitor/bitmap-qmp-cmds.c |
13 | index XXXXXXX..XXXXXXX 100644 | 16 | index XXXXXXX..XXXXXXX 100644 |
14 | --- a/tests/qemu-iotests/iotests.py | 17 | --- a/block/monitor/bitmap-qmp-cmds.c |
15 | +++ b/tests/qemu-iotests/iotests.py | 18 | +++ b/block/monitor/bitmap-qmp-cmds.c |
16 | @@ -XXX,XX +XXX,XX @@ class VM(qtest.QEMUQtestMachine): | 19 | @@ -XXX,XX +XXX,XX @@ BdrvDirtyBitmap *block_dirty_bitmap_merge(const char *node, const char *target, |
17 | if use_log: | 20 | } |
18 | log('Job failed: %s' % (j['error'])) | 21 | |
19 | elif status == 'ready': | 22 | /* Merge into dst; dst is unchanged on failure. */ |
20 | - self.qmp_log('job-complete', id=job) | 23 | - bdrv_merge_dirty_bitmap(dst, anon, backup, errp); |
21 | + if use_log: | 24 | + if (!bdrv_merge_dirty_bitmap(dst, anon, backup, errp)) { |
22 | + self.qmp_log('job-complete', id=job) | 25 | + dst = NULL; |
23 | + else: | 26 | + goto out; |
24 | + self.qmp('job-complete', id=job) | 27 | + } |
25 | elif status == 'pending' and not auto_finalize: | 28 | |
26 | if pre_finalize: | 29 | out: |
27 | pre_finalize() | 30 | bdrv_release_dirty_bitmap(anon); |
28 | -- | 31 | -- |
29 | 2.20.1 | 32 | 2.35.3 |
30 | |||
31 | diff view generated by jsdifflib |
1 | external_snapshot_prepare() tries to move the overlay to the AioContext | 1 | From: Vladimir Sementsov-Ogievskiy <vsementsov@openvz.org> |
---|---|---|---|
2 | of the backing file (the snapshotted node). However, it's possible that | ||
3 | this doesn't work, but the backing file can instead be moved to the | ||
4 | overlay's AioContext (e.g. opening the backing chain for a mirror | ||
5 | target). | ||
6 | 2 | ||
7 | bdrv_append() already indirectly uses bdrv_attach_node(), which takes | 3 | We don't need extra bitmap. All we need is to backup the original |
8 | care to move nodes to make sure they use the same AioContext and which | 4 | bitmap when we do first merge. So, drop extra temporary bitmap and work |
9 | tries both directions. | 5 | directly with target and backup. |
10 | 6 | ||
11 | So the problem has a simple fix: Just delete the unnecessary extra | 7 | Still to keep old semantics, that on failure target is unchanged and |
12 | bdrv_try_set_aio_context() call in external_snapshot_prepare() and | 8 | user don't need to restore, we need a local_backup variable and do |
13 | instead assert in bdrv_append() that both nodes were indeed moved to the | 9 | restore ourselves on failure path. |
14 | same AioContext. | ||
15 | 10 | ||
16 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> | 11 | Signed-off-by: Vladimir Sementsov-Ogievskiy <v.sementsov-og@mail.ru> |
17 | Message-Id: <20200310113831.27293-6-kwolf@redhat.com> | 12 | Message-Id: <20220517111206.23585-3-v.sementsov-og@mail.ru> |
18 | Tested-by: Peter Krempa <pkrempa@redhat.com> | 13 | Reviewed-by: Eric Blake <eblake@redhat.com> |
19 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> | 14 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> |
20 | --- | 15 | --- |
21 | block.c | 1 + | 16 | block/monitor/bitmap-qmp-cmds.c | 41 +++++++++++++++++---------------- |
22 | blockdev.c | 16 ---------------- | 17 | 1 file changed, 21 insertions(+), 20 deletions(-) |
23 | 2 files changed, 1 insertion(+), 16 deletions(-) | ||
24 | 18 | ||
25 | diff --git a/block.c b/block.c | 19 | diff --git a/block/monitor/bitmap-qmp-cmds.c b/block/monitor/bitmap-qmp-cmds.c |
26 | index XXXXXXX..XXXXXXX 100644 | 20 | index XXXXXXX..XXXXXXX 100644 |
27 | --- a/block.c | 21 | --- a/block/monitor/bitmap-qmp-cmds.c |
28 | +++ b/block.c | 22 | +++ b/block/monitor/bitmap-qmp-cmds.c |
29 | @@ -XXX,XX +XXX,XX @@ void bdrv_replace_node(BlockDriverState *from, BlockDriverState *to, | 23 | @@ -XXX,XX +XXX,XX @@ BdrvDirtyBitmap *block_dirty_bitmap_merge(const char *node, const char *target, |
30 | bdrv_ref(from); | 24 | HBitmap **backup, Error **errp) |
31 | 25 | { | |
32 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); | 26 | BlockDriverState *bs; |
33 | + assert(bdrv_get_aio_context(from) == bdrv_get_aio_context(to)); | 27 | - BdrvDirtyBitmap *dst, *src, *anon; |
34 | bdrv_drained_begin(from); | 28 | + BdrvDirtyBitmap *dst, *src; |
35 | 29 | BlockDirtyBitmapOrStrList *lst; | |
36 | /* Put all parents into @list and calculate their cumulative permissions */ | 30 | + HBitmap *local_backup = NULL; |
37 | diff --git a/blockdev.c b/blockdev.c | 31 | |
38 | index XXXXXXX..XXXXXXX 100644 | 32 | GLOBAL_STATE_CODE(); |
39 | --- a/blockdev.c | 33 | |
40 | +++ b/blockdev.c | 34 | @@ -XXX,XX +XXX,XX @@ BdrvDirtyBitmap *block_dirty_bitmap_merge(const char *node, const char *target, |
41 | @@ -XXX,XX +XXX,XX @@ static void external_snapshot_prepare(BlkActionState *common, | 35 | return NULL; |
42 | DO_UPCAST(ExternalSnapshotState, common, common); | ||
43 | TransactionAction *action = common->action; | ||
44 | AioContext *aio_context; | ||
45 | - AioContext *old_context; | ||
46 | uint64_t perm, shared; | ||
47 | - int ret; | ||
48 | |||
49 | /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar | ||
50 | * purpose but a different set of parameters */ | ||
51 | @@ -XXX,XX +XXX,XX @@ static void external_snapshot_prepare(BlkActionState *common, | ||
52 | goto out; | ||
53 | } | 36 | } |
54 | 37 | ||
55 | - /* Honor bdrv_try_set_aio_context() context acquisition requirements. */ | 38 | - anon = bdrv_create_dirty_bitmap(bs, bdrv_dirty_bitmap_granularity(dst), |
56 | - old_context = bdrv_get_aio_context(state->new_bs); | 39 | - NULL, errp); |
57 | - aio_context_release(aio_context); | 40 | - if (!anon) { |
58 | - aio_context_acquire(old_context); | 41 | - return NULL; |
59 | - | ||
60 | - ret = bdrv_try_set_aio_context(state->new_bs, aio_context, errp); | ||
61 | - | ||
62 | - aio_context_release(old_context); | ||
63 | - aio_context_acquire(aio_context); | ||
64 | - | ||
65 | - if (ret < 0) { | ||
66 | - goto out; | ||
67 | - } | 42 | - } |
68 | - | 43 | - |
69 | /* This removes our old bs and adds the new bs. This is an operation that | 44 | for (lst = bms; lst; lst = lst->next) { |
70 | * can fail, so we need to do it in .prepare; undoing it for abort is | 45 | switch (lst->value->type) { |
71 | * always possible. */ | 46 | const char *name, *node; |
47 | @@ -XXX,XX +XXX,XX @@ BdrvDirtyBitmap *block_dirty_bitmap_merge(const char *node, const char *target, | ||
48 | src = bdrv_find_dirty_bitmap(bs, name); | ||
49 | if (!src) { | ||
50 | error_setg(errp, "Dirty bitmap '%s' not found", name); | ||
51 | - dst = NULL; | ||
52 | - goto out; | ||
53 | + goto fail; | ||
54 | } | ||
55 | break; | ||
56 | case QTYPE_QDICT: | ||
57 | @@ -XXX,XX +XXX,XX @@ BdrvDirtyBitmap *block_dirty_bitmap_merge(const char *node, const char *target, | ||
58 | name = lst->value->u.external.name; | ||
59 | src = block_dirty_bitmap_lookup(node, name, NULL, errp); | ||
60 | if (!src) { | ||
61 | - dst = NULL; | ||
62 | - goto out; | ||
63 | + goto fail; | ||
64 | } | ||
65 | break; | ||
66 | default: | ||
67 | abort(); | ||
68 | } | ||
69 | |||
70 | - if (!bdrv_merge_dirty_bitmap(anon, src, NULL, errp)) { | ||
71 | - dst = NULL; | ||
72 | - goto out; | ||
73 | + /* We do backup only for first merge operation */ | ||
74 | + if (!bdrv_merge_dirty_bitmap(dst, src, | ||
75 | + local_backup ? NULL : &local_backup, | ||
76 | + errp)) | ||
77 | + { | ||
78 | + goto fail; | ||
79 | } | ||
80 | } | ||
81 | |||
82 | - /* Merge into dst; dst is unchanged on failure. */ | ||
83 | - if (!bdrv_merge_dirty_bitmap(dst, anon, backup, errp)) { | ||
84 | - dst = NULL; | ||
85 | - goto out; | ||
86 | + if (backup) { | ||
87 | + *backup = local_backup; | ||
88 | + } else { | ||
89 | + hbitmap_free(local_backup); | ||
90 | } | ||
91 | |||
92 | - out: | ||
93 | - bdrv_release_dirty_bitmap(anon); | ||
94 | return dst; | ||
95 | + | ||
96 | +fail: | ||
97 | + if (local_backup) { | ||
98 | + bdrv_restore_dirty_bitmap(dst, local_backup); | ||
99 | + } | ||
100 | + | ||
101 | + return NULL; | ||
102 | } | ||
103 | |||
104 | void qmp_block_dirty_bitmap_merge(const char *node, const char *target, | ||
72 | -- | 105 | -- |
73 | 2.20.1 | 106 | 2.35.3 |
74 | |||
75 | diff view generated by jsdifflib |
1 | blockdev-snapshot returned an error if the overlay was already in use, | 1 | From: Vladimir Sementsov-Ogievskiy <vsementsov@openvz.org> |
---|---|---|---|
2 | which it defined as having any BlockBackend parent. This is in fact both | 2 | |
3 | too strict (some parents can tolerate the change of visible data caused | 3 | We have too much logic to simply check that bitmaps are of the same |
4 | by attaching a backing file) and too loose (some non-BlockBackend | 4 | size. Let's just define that hbitmap_merge() and |
5 | parents may not be happy with it). | 5 | bdrv_dirty_bitmap_merge_internal() require their argument bitmaps be of |
6 | 6 | same size, this simplifies things. | |
7 | One important use case that is prevented by the too strict check is live | 7 | |
8 | storage migration with blockdev-mirror. Here, the target node is | 8 | Let's look through the callers: |
9 | usually opened without a backing file so that the active layer is | 9 | |
10 | mirrored while its backing chain can be copied in the background. | 10 | For backup_init_bcs_bitmap() we already assert that merge can't fail. |
11 | 11 | ||
12 | The backing chain should be attached to the mirror target node when | 12 | In bdrv_reclaim_dirty_bitmap_locked() we gracefully handle the error |
13 | finalising the job, just before switching the users of the source node | 13 | that can't happen: successor always has same size as its parent, drop |
14 | to the new copy (at which point the mirror job still has a reference to | 14 | this logic. |
15 | the node). drive-mirror did this automatically, but with blockdev-mirror | 15 | |
16 | this is the job of the QMP client, so it needs a way to do this. | 16 | In bdrv_merge_dirty_bitmap() we already has assertion and separate |
17 | 17 | check. Make the check explicit and improve error message. | |
18 | blockdev-snapshot is the obvious way, so this patch makes it work in | 18 | |
19 | this scenario. The new condition is that no parent uses CONSISTENT_READ | 19 | Signed-off-by: Vladimir Sementsov-Ogievskiy <v.sementsov-og@mail.ru> |
20 | permissions. This will ensure that the operation will still be blocked | 20 | Reviewed-by: Nikita Lapshin <nikita.lapshin@virtuozzo.com> |
21 | when the node is attached to the guest device, so blockdev-snapshot | 21 | Reviewed-by: Kevin Wolf <kwolf@redhat.com> |
22 | remains safe. | 22 | Message-Id: <20220517111206.23585-4-v.sementsov-og@mail.ru> |
23 | |||
24 | (For the sake of completeness, x-blockdev-reopen can be used to achieve | ||
25 | the same, however it is a big hammer, performs the graph change | ||
26 | completely unchecked and is still experimental. So even with the option | ||
27 | of using x-blockdev-reopen, there are reasons why blockdev-snapshot | ||
28 | should be able to perform this operation.) | ||
29 | |||
30 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> | ||
31 | Message-Id: <20200310113831.27293-3-kwolf@redhat.com> | ||
32 | Reviewed-by: Peter Krempa <pkrempa@redhat.com> | ||
33 | Tested-by: Peter Krempa <pkrempa@redhat.com> | ||
34 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> | 23 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> |
35 | --- | 24 | --- |
36 | blockdev.c | 14 ++++++++------ | 25 | include/block/block_int-io.h | 2 +- |
37 | tests/qemu-iotests/085.out | 4 ++-- | 26 | include/qemu/hbitmap.h | 15 ++------------- |
38 | 2 files changed, 10 insertions(+), 8 deletions(-) | 27 | block/backup.c | 6 ++---- |
39 | 28 | block/dirty-bitmap.c | 26 +++++++++++--------------- | |
40 | diff --git a/blockdev.c b/blockdev.c | 29 | util/hbitmap.c | 25 +++++++------------------ |
41 | index XXXXXXX..XXXXXXX 100644 | 30 | 5 files changed, 23 insertions(+), 51 deletions(-) |
42 | --- a/blockdev.c | 31 | |
43 | +++ b/blockdev.c | 32 | diff --git a/include/block/block_int-io.h b/include/block/block_int-io.h |
44 | @@ -XXX,XX +XXX,XX @@ static void external_snapshot_prepare(BlkActionState *common, | 33 | index XXXXXXX..XXXXXXX 100644 |
45 | TransactionAction *action = common->action; | 34 | --- a/include/block/block_int-io.h |
46 | AioContext *aio_context; | 35 | +++ b/include/block/block_int-io.h |
47 | AioContext *old_context; | 36 | @@ -XXX,XX +XXX,XX @@ bool blk_dev_is_tray_open(BlockBackend *blk); |
48 | + uint64_t perm, shared; | 37 | void bdrv_set_dirty(BlockDriverState *bs, int64_t offset, int64_t bytes); |
49 | int ret; | 38 | |
50 | 39 | void bdrv_clear_dirty_bitmap(BdrvDirtyBitmap *bitmap, HBitmap **out); | |
51 | /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar | 40 | -bool bdrv_dirty_bitmap_merge_internal(BdrvDirtyBitmap *dest, |
52 | @@ -XXX,XX +XXX,XX @@ static void external_snapshot_prepare(BlkActionState *common, | 41 | +void bdrv_dirty_bitmap_merge_internal(BdrvDirtyBitmap *dest, |
42 | const BdrvDirtyBitmap *src, | ||
43 | HBitmap **backup, bool lock); | ||
44 | |||
45 | diff --git a/include/qemu/hbitmap.h b/include/qemu/hbitmap.h | ||
46 | index XXXXXXX..XXXXXXX 100644 | ||
47 | --- a/include/qemu/hbitmap.h | ||
48 | +++ b/include/qemu/hbitmap.h | ||
49 | @@ -XXX,XX +XXX,XX @@ void hbitmap_truncate(HBitmap *hb, uint64_t size); | ||
50 | * | ||
51 | * Store result of merging @a and @b into @result. | ||
52 | * @result is allowed to be equal to @a or @b. | ||
53 | - * | ||
54 | - * Return true if the merge was successful, | ||
55 | - * false if it was not attempted. | ||
56 | - */ | ||
57 | -bool hbitmap_merge(const HBitmap *a, const HBitmap *b, HBitmap *result); | ||
58 | - | ||
59 | -/** | ||
60 | - * hbitmap_can_merge: | ||
61 | - * | ||
62 | - * hbitmap_can_merge(a, b) && hbitmap_can_merge(a, result) is sufficient and | ||
63 | - * necessary for hbitmap_merge will not fail. | ||
64 | - * | ||
65 | + * All bitmaps must have same size. | ||
66 | */ | ||
67 | -bool hbitmap_can_merge(const HBitmap *a, const HBitmap *b); | ||
68 | +void hbitmap_merge(const HBitmap *a, const HBitmap *b, HBitmap *result); | ||
69 | |||
70 | /** | ||
71 | * hbitmap_empty: | ||
72 | diff --git a/block/backup.c b/block/backup.c | ||
73 | index XXXXXXX..XXXXXXX 100644 | ||
74 | --- a/block/backup.c | ||
75 | +++ b/block/backup.c | ||
76 | @@ -XXX,XX +XXX,XX @@ out: | ||
77 | |||
78 | static void backup_init_bcs_bitmap(BackupBlockJob *job) | ||
79 | { | ||
80 | - bool ret; | ||
81 | uint64_t estimate; | ||
82 | BdrvDirtyBitmap *bcs_bitmap = block_copy_dirty_bitmap(job->bcs); | ||
83 | |||
84 | if (job->sync_mode == MIRROR_SYNC_MODE_BITMAP) { | ||
85 | bdrv_clear_dirty_bitmap(bcs_bitmap, NULL); | ||
86 | - ret = bdrv_dirty_bitmap_merge_internal(bcs_bitmap, job->sync_bitmap, | ||
87 | - NULL, true); | ||
88 | - assert(ret); | ||
89 | + bdrv_dirty_bitmap_merge_internal(bcs_bitmap, job->sync_bitmap, NULL, | ||
90 | + true); | ||
91 | } else if (job->sync_mode == MIRROR_SYNC_MODE_TOP) { | ||
92 | /* | ||
93 | * We can't hog the coroutine to initialize this thoroughly. | ||
94 | diff --git a/block/dirty-bitmap.c b/block/dirty-bitmap.c | ||
95 | index XXXXXXX..XXXXXXX 100644 | ||
96 | --- a/block/dirty-bitmap.c | ||
97 | +++ b/block/dirty-bitmap.c | ||
98 | @@ -XXX,XX +XXX,XX @@ BdrvDirtyBitmap *bdrv_reclaim_dirty_bitmap_locked(BdrvDirtyBitmap *parent, | ||
99 | return NULL; | ||
100 | } | ||
101 | |||
102 | - if (!hbitmap_merge(parent->bitmap, successor->bitmap, parent->bitmap)) { | ||
103 | - error_setg(errp, "Merging of parent and successor bitmap failed"); | ||
104 | - return NULL; | ||
105 | - } | ||
106 | + hbitmap_merge(parent->bitmap, successor->bitmap, parent->bitmap); | ||
107 | |||
108 | parent->disabled = successor->disabled; | ||
109 | parent->busy = false; | ||
110 | @@ -XXX,XX +XXX,XX @@ bool bdrv_merge_dirty_bitmap(BdrvDirtyBitmap *dest, const BdrvDirtyBitmap *src, | ||
53 | goto out; | 111 | goto out; |
54 | } | 112 | } |
55 | 113 | ||
56 | - if (bdrv_has_blk(state->new_bs)) { | 114 | - if (!hbitmap_can_merge(dest->bitmap, src->bitmap)) { |
57 | + /* | 115 | - error_setg(errp, "Bitmaps are incompatible and can't be merged"); |
58 | + * Allow attaching a backing file to an overlay that's already in use only | 116 | + if (bdrv_dirty_bitmap_size(src) != bdrv_dirty_bitmap_size(dest)) { |
59 | + * if the parents don't assume that they are already seeing a valid image. | 117 | + error_setg(errp, "Bitmaps are of different sizes (destination size is %" |
60 | + * (Specifically, allow it as a mirror target, which is write-only access.) | 118 | + PRId64 ", source size is %" PRId64 ") and can't be merged", |
61 | + */ | 119 | + bdrv_dirty_bitmap_size(dest), bdrv_dirty_bitmap_size(src)); |
62 | + bdrv_get_cumulative_perm(state->new_bs, &perm, &shared); | ||
63 | + if (perm & BLK_PERM_CONSISTENT_READ) { | ||
64 | error_setg(errp, "The overlay is already in use"); | ||
65 | goto out; | 120 | goto out; |
66 | } | 121 | } |
67 | 122 | ||
68 | - if (bdrv_op_is_blocked(state->new_bs, BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, | 123 | - ret = bdrv_dirty_bitmap_merge_internal(dest, src, backup, false); |
69 | - errp)) { | 124 | - assert(ret); |
70 | - goto out; | 125 | + bdrv_dirty_bitmap_merge_internal(dest, src, backup, false); |
126 | + ret = true; | ||
127 | |||
128 | out: | ||
129 | bdrv_dirty_bitmaps_unlock(dest->bs); | ||
130 | @@ -XXX,XX +XXX,XX @@ out: | ||
131 | /** | ||
132 | * bdrv_dirty_bitmap_merge_internal: merge src into dest. | ||
133 | * Does NOT check bitmap permissions; not suitable for use as public API. | ||
134 | + * @dest, @src and @backup (if not NULL) must have same size. | ||
135 | * | ||
136 | * @backup: If provided, make a copy of dest here prior to merge. | ||
137 | * @lock: If true, lock and unlock bitmaps on the way in/out. | ||
138 | - * returns true if the merge succeeded; false if unattempted. | ||
139 | */ | ||
140 | -bool bdrv_dirty_bitmap_merge_internal(BdrvDirtyBitmap *dest, | ||
141 | +void bdrv_dirty_bitmap_merge_internal(BdrvDirtyBitmap *dest, | ||
142 | const BdrvDirtyBitmap *src, | ||
143 | HBitmap **backup, | ||
144 | bool lock) | ||
145 | { | ||
146 | - bool ret; | ||
147 | IO_CODE(); | ||
148 | |||
149 | assert(!bdrv_dirty_bitmap_readonly(dest)); | ||
150 | @@ -XXX,XX +XXX,XX @@ bool bdrv_dirty_bitmap_merge_internal(BdrvDirtyBitmap *dest, | ||
151 | if (backup) { | ||
152 | *backup = dest->bitmap; | ||
153 | dest->bitmap = hbitmap_alloc(dest->size, hbitmap_granularity(*backup)); | ||
154 | - ret = hbitmap_merge(*backup, src->bitmap, dest->bitmap); | ||
155 | + hbitmap_merge(*backup, src->bitmap, dest->bitmap); | ||
156 | } else { | ||
157 | - ret = hbitmap_merge(dest->bitmap, src->bitmap, dest->bitmap); | ||
158 | + hbitmap_merge(dest->bitmap, src->bitmap, dest->bitmap); | ||
159 | } | ||
160 | |||
161 | if (lock) { | ||
162 | @@ -XXX,XX +XXX,XX @@ bool bdrv_dirty_bitmap_merge_internal(BdrvDirtyBitmap *dest, | ||
163 | bdrv_dirty_bitmaps_unlock(src->bs); | ||
164 | } | ||
165 | } | ||
166 | - | ||
167 | - return ret; | ||
168 | } | ||
169 | diff --git a/util/hbitmap.c b/util/hbitmap.c | ||
170 | index XXXXXXX..XXXXXXX 100644 | ||
171 | --- a/util/hbitmap.c | ||
172 | +++ b/util/hbitmap.c | ||
173 | @@ -XXX,XX +XXX,XX @@ void hbitmap_truncate(HBitmap *hb, uint64_t size) | ||
174 | } | ||
175 | } | ||
176 | |||
177 | -bool hbitmap_can_merge(const HBitmap *a, const HBitmap *b) | ||
178 | -{ | ||
179 | - return (a->orig_size == b->orig_size); | ||
180 | -} | ||
181 | - | ||
182 | /** | ||
183 | * hbitmap_sparse_merge: performs dst = dst | src | ||
184 | * works with differing granularities. | ||
185 | @@ -XXX,XX +XXX,XX @@ static void hbitmap_sparse_merge(HBitmap *dst, const HBitmap *src) | ||
186 | * Given HBitmaps A and B, let R := A (BITOR) B. | ||
187 | * Bitmaps A and B will not be modified, | ||
188 | * except when bitmap R is an alias of A or B. | ||
189 | - * | ||
190 | - * @return true if the merge was successful, | ||
191 | - * false if it was not attempted. | ||
192 | + * Bitmaps must have same size. | ||
193 | */ | ||
194 | -bool hbitmap_merge(const HBitmap *a, const HBitmap *b, HBitmap *result) | ||
195 | +void hbitmap_merge(const HBitmap *a, const HBitmap *b, HBitmap *result) | ||
196 | { | ||
197 | int i; | ||
198 | uint64_t j; | ||
199 | |||
200 | - if (!hbitmap_can_merge(a, b) || !hbitmap_can_merge(a, result)) { | ||
201 | - return false; | ||
71 | - } | 202 | - } |
72 | - | 203 | - assert(hbitmap_can_merge(b, result)); |
73 | if (state->new_bs->backing != NULL) { | 204 | + assert(a->orig_size == result->orig_size); |
74 | error_setg(errp, "The overlay already has a backing image"); | 205 | + assert(b->orig_size == result->orig_size); |
75 | goto out; | 206 | |
76 | diff --git a/tests/qemu-iotests/085.out b/tests/qemu-iotests/085.out | 207 | if ((!hbitmap_count(a) && result == b) || |
77 | index XXXXXXX..XXXXXXX 100644 | 208 | (!hbitmap_count(b) && result == a)) { |
78 | --- a/tests/qemu-iotests/085.out | 209 | - return true; |
79 | +++ b/tests/qemu-iotests/085.out | 210 | + return; |
80 | @@ -XXX,XX +XXX,XX @@ Formatting 'TEST_DIR/12-snapshot-v0.IMGFMT', fmt=IMGFMT size=134217728 backing_f | 211 | } |
81 | === Invalid command - cannot create a snapshot using a file BDS === | 212 | |
82 | 213 | if (!hbitmap_count(a) && !hbitmap_count(b)) { | |
83 | { 'execute': 'blockdev-snapshot', 'arguments': { 'node':'virtio0', 'overlay':'file_12' } } | 214 | hbitmap_reset_all(result); |
84 | -{"error": {"class": "GenericError", "desc": "The overlay does not support backing images"}} | 215 | - return true; |
85 | +{"error": {"class": "GenericError", "desc": "The overlay is already in use"}} | 216 | + return; |
86 | 217 | } | |
87 | === Invalid command - snapshot node used as active layer === | 218 | |
88 | 219 | if (a->granularity != b->granularity) { | |
89 | @@ -XXX,XX +XXX,XX @@ Formatting 'TEST_DIR/12-snapshot-v0.IMGFMT', fmt=IMGFMT size=134217728 backing_f | 220 | @@ -XXX,XX +XXX,XX @@ bool hbitmap_merge(const HBitmap *a, const HBitmap *b, HBitmap *result) |
90 | === Invalid command - snapshot node used as backing hd === | 221 | if (b != result) { |
91 | 222 | hbitmap_sparse_merge(result, b); | |
92 | { 'execute': 'blockdev-snapshot', 'arguments': { 'node': 'virtio0', 'overlay':'snap_11' } } | 223 | } |
93 | -{"error": {"class": "GenericError", "desc": "Node 'snap_11' is busy: node is used as backing hd of 'snap_12'"}} | 224 | - return true; |
94 | +{"error": {"class": "GenericError", "desc": "The overlay is already in use"}} | 225 | + return; |
95 | 226 | } | |
96 | === Invalid command - snapshot node has a backing image === | 227 | |
97 | 228 | /* This merge is O(size), as BITS_PER_LONG and HBITMAP_LEVELS are constant. | |
229 | @@ -XXX,XX +XXX,XX @@ bool hbitmap_merge(const HBitmap *a, const HBitmap *b, HBitmap *result) | ||
230 | |||
231 | /* Recompute the dirty count */ | ||
232 | result->count = hb_count_between(result, 0, result->size - 1); | ||
233 | - | ||
234 | - return true; | ||
235 | } | ||
236 | |||
237 | char *hbitmap_sha256(const HBitmap *bitmap, Error **errp) | ||
98 | -- | 238 | -- |
99 | 2.20.1 | 239 | 2.35.3 |
100 | |||
101 | diff view generated by jsdifflib |
New patch | |||
---|---|---|---|
1 | From: Xie Yongji <xieyongji@bytedance.com> | ||
1 | 2 | ||
3 | This supports passing NULL ops to blk_set_dev_ops() | ||
4 | so that we can remove stale ops in some cases. | ||
5 | |||
6 | Signed-off-by: Xie Yongji <xieyongji@bytedance.com> | ||
7 | Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> | ||
8 | Message-Id: <20220523084611.91-2-xieyongji@bytedance.com> | ||
9 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> | ||
10 | --- | ||
11 | block/block-backend.c | 2 +- | ||
12 | 1 file changed, 1 insertion(+), 1 deletion(-) | ||
13 | |||
14 | diff --git a/block/block-backend.c b/block/block-backend.c | ||
15 | index XXXXXXX..XXXXXXX 100644 | ||
16 | --- a/block/block-backend.c | ||
17 | +++ b/block/block-backend.c | ||
18 | @@ -XXX,XX +XXX,XX @@ void blk_set_dev_ops(BlockBackend *blk, const BlockDevOps *ops, | ||
19 | blk->dev_opaque = opaque; | ||
20 | |||
21 | /* Are we currently quiesced? Should we enforce this right now? */ | ||
22 | - if (blk->quiesce_counter && ops->drained_begin) { | ||
23 | + if (blk->quiesce_counter && ops && ops->drained_begin) { | ||
24 | ops->drained_begin(opaque); | ||
25 | } | ||
26 | } | ||
27 | -- | ||
28 | 2.35.3 | diff view generated by jsdifflib |
New patch | |||
---|---|---|---|
1 | From: Xie Yongji <xieyongji@bytedance.com> | ||
1 | 2 | ||
3 | Now the req->size is set to the correct value only | ||
4 | when handling VIRTIO_BLK_T_GET_ID request. This patch | ||
5 | fixes it. | ||
6 | |||
7 | Signed-off-by: Xie Yongji <xieyongji@bytedance.com> | ||
8 | Message-Id: <20220523084611.91-3-xieyongji@bytedance.com> | ||
9 | Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> | ||
10 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> | ||
11 | --- | ||
12 | block/export/vhost-user-blk-server.c | 5 ++--- | ||
13 | 1 file changed, 2 insertions(+), 3 deletions(-) | ||
14 | |||
15 | diff --git a/block/export/vhost-user-blk-server.c b/block/export/vhost-user-blk-server.c | ||
16 | index XXXXXXX..XXXXXXX 100644 | ||
17 | --- a/block/export/vhost-user-blk-server.c | ||
18 | +++ b/block/export/vhost-user-blk-server.c | ||
19 | @@ -XXX,XX +XXX,XX @@ static void vu_blk_req_complete(VuBlkReq *req) | ||
20 | { | ||
21 | VuDev *vu_dev = &req->server->vu_dev; | ||
22 | |||
23 | - /* IO size with 1 extra status byte */ | ||
24 | - vu_queue_push(vu_dev, req->vq, &req->elem, req->size + 1); | ||
25 | + vu_queue_push(vu_dev, req->vq, &req->elem, req->size); | ||
26 | vu_queue_notify(vu_dev, req->vq); | ||
27 | |||
28 | free(req); | ||
29 | @@ -XXX,XX +XXX,XX @@ static void coroutine_fn vu_blk_virtio_process_req(void *opaque) | ||
30 | goto err; | ||
31 | } | ||
32 | |||
33 | + req->size = iov_size(in_iov, in_num); | ||
34 | /* We always touch the last byte, so just see how big in_iov is. */ | ||
35 | req->in = (void *)in_iov[in_num - 1].iov_base | ||
36 | + in_iov[in_num - 1].iov_len | ||
37 | @@ -XXX,XX +XXX,XX @@ static void coroutine_fn vu_blk_virtio_process_req(void *opaque) | ||
38 | VIRTIO_BLK_ID_BYTES); | ||
39 | snprintf(elem->in_sg[0].iov_base, size, "%s", "vhost_user_blk"); | ||
40 | req->in->status = VIRTIO_BLK_S_OK; | ||
41 | - req->size = elem->in_sg[0].iov_len; | ||
42 | break; | ||
43 | } | ||
44 | case VIRTIO_BLK_T_DISCARD: | ||
45 | -- | ||
46 | 2.35.3 | diff view generated by jsdifflib |
New patch | |||
---|---|---|---|
1 | From: Xie Yongji <xieyongji@bytedance.com> | ||
1 | 2 | ||
3 | Abstract the common logic of virtio-blk I/O process to a function | ||
4 | named virtio_blk_process_req(). It's needed for the following commit. | ||
5 | |||
6 | Signed-off-by: Xie Yongji <xieyongji@bytedance.com> | ||
7 | Message-Id: <20220523084611.91-4-xieyongji@bytedance.com> | ||
8 | Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> | ||
9 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> | ||
10 | --- | ||
11 | block/export/virtio-blk-handler.h | 37 ++++ | ||
12 | block/export/vhost-user-blk-server.c | 259 +++------------------------ | ||
13 | block/export/virtio-blk-handler.c | 240 +++++++++++++++++++++++++ | ||
14 | MAINTAINERS | 2 + | ||
15 | block/export/meson.build | 2 +- | ||
16 | 5 files changed, 301 insertions(+), 239 deletions(-) | ||
17 | create mode 100644 block/export/virtio-blk-handler.h | ||
18 | create mode 100644 block/export/virtio-blk-handler.c | ||
19 | |||
20 | diff --git a/block/export/virtio-blk-handler.h b/block/export/virtio-blk-handler.h | ||
21 | new file mode 100644 | ||
22 | index XXXXXXX..XXXXXXX | ||
23 | --- /dev/null | ||
24 | +++ b/block/export/virtio-blk-handler.h | ||
25 | @@ -XXX,XX +XXX,XX @@ | ||
26 | +/* | ||
27 | + * Handler for virtio-blk I/O | ||
28 | + * | ||
29 | + * Copyright (C) 2022 Bytedance Inc. and/or its affiliates. All rights reserved. | ||
30 | + * | ||
31 | + * Author: | ||
32 | + * Xie Yongji <xieyongji@bytedance.com> | ||
33 | + * | ||
34 | + * This work is licensed under the terms of the GNU GPL, version 2 or | ||
35 | + * later. See the COPYING file in the top-level directory. | ||
36 | + */ | ||
37 | + | ||
38 | +#ifndef VIRTIO_BLK_HANDLER_H | ||
39 | +#define VIRTIO_BLK_HANDLER_H | ||
40 | + | ||
41 | +#include "sysemu/block-backend.h" | ||
42 | + | ||
43 | +#define VIRTIO_BLK_SECTOR_BITS 9 | ||
44 | +#define VIRTIO_BLK_SECTOR_SIZE (1ULL << VIRTIO_BLK_SECTOR_BITS) | ||
45 | + | ||
46 | +#define VIRTIO_BLK_MAX_DISCARD_SECTORS 32768 | ||
47 | +#define VIRTIO_BLK_MAX_WRITE_ZEROES_SECTORS 32768 | ||
48 | + | ||
49 | +typedef struct { | ||
50 | + BlockBackend *blk; | ||
51 | + const char *serial; | ||
52 | + uint32_t logical_block_size; | ||
53 | + bool writable; | ||
54 | +} VirtioBlkHandler; | ||
55 | + | ||
56 | +int coroutine_fn virtio_blk_process_req(VirtioBlkHandler *handler, | ||
57 | + struct iovec *in_iov, | ||
58 | + struct iovec *out_iov, | ||
59 | + unsigned int in_num, | ||
60 | + unsigned int out_num); | ||
61 | + | ||
62 | +#endif /* VIRTIO_BLK_HANDLER_H */ | ||
63 | diff --git a/block/export/vhost-user-blk-server.c b/block/export/vhost-user-blk-server.c | ||
64 | index XXXXXXX..XXXXXXX 100644 | ||
65 | --- a/block/export/vhost-user-blk-server.c | ||
66 | +++ b/block/export/vhost-user-blk-server.c | ||
67 | @@ -XXX,XX +XXX,XX @@ | ||
68 | #include "vhost-user-blk-server.h" | ||
69 | #include "qapi/error.h" | ||
70 | #include "qom/object_interfaces.h" | ||
71 | -#include "sysemu/block-backend.h" | ||
72 | #include "util/block-helpers.h" | ||
73 | - | ||
74 | -/* | ||
75 | - * Sector units are 512 bytes regardless of the | ||
76 | - * virtio_blk_config->blk_size value. | ||
77 | - */ | ||
78 | -#define VIRTIO_BLK_SECTOR_BITS 9 | ||
79 | -#define VIRTIO_BLK_SECTOR_SIZE (1ull << VIRTIO_BLK_SECTOR_BITS) | ||
80 | +#include "virtio-blk-handler.h" | ||
81 | |||
82 | enum { | ||
83 | VHOST_USER_BLK_NUM_QUEUES_DEFAULT = 1, | ||
84 | - VHOST_USER_BLK_MAX_DISCARD_SECTORS = 32768, | ||
85 | - VHOST_USER_BLK_MAX_WRITE_ZEROES_SECTORS = 32768, | ||
86 | -}; | ||
87 | -struct virtio_blk_inhdr { | ||
88 | - unsigned char status; | ||
89 | }; | ||
90 | |||
91 | typedef struct VuBlkReq { | ||
92 | VuVirtqElement elem; | ||
93 | - int64_t sector_num; | ||
94 | - size_t size; | ||
95 | - struct virtio_blk_inhdr *in; | ||
96 | - struct virtio_blk_outhdr out; | ||
97 | VuServer *server; | ||
98 | struct VuVirtq *vq; | ||
99 | } VuBlkReq; | ||
100 | @@ -XXX,XX +XXX,XX @@ typedef struct VuBlkReq { | ||
101 | typedef struct { | ||
102 | BlockExport export; | ||
103 | VuServer vu_server; | ||
104 | - uint32_t blk_size; | ||
105 | + VirtioBlkHandler handler; | ||
106 | QIOChannelSocket *sioc; | ||
107 | struct virtio_blk_config blkcfg; | ||
108 | - bool writable; | ||
109 | } VuBlkExport; | ||
110 | |||
111 | -static void vu_blk_req_complete(VuBlkReq *req) | ||
112 | +static void vu_blk_req_complete(VuBlkReq *req, size_t in_len) | ||
113 | { | ||
114 | VuDev *vu_dev = &req->server->vu_dev; | ||
115 | |||
116 | - vu_queue_push(vu_dev, req->vq, &req->elem, req->size); | ||
117 | + vu_queue_push(vu_dev, req->vq, &req->elem, in_len); | ||
118 | vu_queue_notify(vu_dev, req->vq); | ||
119 | |||
120 | free(req); | ||
121 | } | ||
122 | |||
123 | -static bool vu_blk_sect_range_ok(VuBlkExport *vexp, uint64_t sector, | ||
124 | - size_t size) | ||
125 | -{ | ||
126 | - uint64_t nb_sectors; | ||
127 | - uint64_t total_sectors; | ||
128 | - | ||
129 | - if (size % VIRTIO_BLK_SECTOR_SIZE) { | ||
130 | - return false; | ||
131 | - } | ||
132 | - | ||
133 | - nb_sectors = size >> VIRTIO_BLK_SECTOR_BITS; | ||
134 | - | ||
135 | - QEMU_BUILD_BUG_ON(BDRV_SECTOR_SIZE != VIRTIO_BLK_SECTOR_SIZE); | ||
136 | - if (nb_sectors > BDRV_REQUEST_MAX_SECTORS) { | ||
137 | - return false; | ||
138 | - } | ||
139 | - if ((sector << VIRTIO_BLK_SECTOR_BITS) % vexp->blk_size) { | ||
140 | - return false; | ||
141 | - } | ||
142 | - blk_get_geometry(vexp->export.blk, &total_sectors); | ||
143 | - if (sector > total_sectors || nb_sectors > total_sectors - sector) { | ||
144 | - return false; | ||
145 | - } | ||
146 | - return true; | ||
147 | -} | ||
148 | - | ||
149 | -static int coroutine_fn | ||
150 | -vu_blk_discard_write_zeroes(VuBlkExport *vexp, struct iovec *iov, | ||
151 | - uint32_t iovcnt, uint32_t type) | ||
152 | -{ | ||
153 | - BlockBackend *blk = vexp->export.blk; | ||
154 | - struct virtio_blk_discard_write_zeroes desc; | ||
155 | - ssize_t size; | ||
156 | - uint64_t sector; | ||
157 | - uint32_t num_sectors; | ||
158 | - uint32_t max_sectors; | ||
159 | - uint32_t flags; | ||
160 | - int bytes; | ||
161 | - | ||
162 | - /* Only one desc is currently supported */ | ||
163 | - if (unlikely(iov_size(iov, iovcnt) > sizeof(desc))) { | ||
164 | - return VIRTIO_BLK_S_UNSUPP; | ||
165 | - } | ||
166 | - | ||
167 | - size = iov_to_buf(iov, iovcnt, 0, &desc, sizeof(desc)); | ||
168 | - if (unlikely(size != sizeof(desc))) { | ||
169 | - error_report("Invalid size %zd, expected %zu", size, sizeof(desc)); | ||
170 | - return VIRTIO_BLK_S_IOERR; | ||
171 | - } | ||
172 | - | ||
173 | - sector = le64_to_cpu(desc.sector); | ||
174 | - num_sectors = le32_to_cpu(desc.num_sectors); | ||
175 | - flags = le32_to_cpu(desc.flags); | ||
176 | - max_sectors = (type == VIRTIO_BLK_T_WRITE_ZEROES) ? | ||
177 | - VHOST_USER_BLK_MAX_WRITE_ZEROES_SECTORS : | ||
178 | - VHOST_USER_BLK_MAX_DISCARD_SECTORS; | ||
179 | - | ||
180 | - /* This check ensures that 'bytes' fits in an int */ | ||
181 | - if (unlikely(num_sectors > max_sectors)) { | ||
182 | - return VIRTIO_BLK_S_IOERR; | ||
183 | - } | ||
184 | - | ||
185 | - bytes = num_sectors << VIRTIO_BLK_SECTOR_BITS; | ||
186 | - | ||
187 | - if (unlikely(!vu_blk_sect_range_ok(vexp, sector, bytes))) { | ||
188 | - return VIRTIO_BLK_S_IOERR; | ||
189 | - } | ||
190 | - | ||
191 | - /* | ||
192 | - * The device MUST set the status byte to VIRTIO_BLK_S_UNSUPP for discard | ||
193 | - * and write zeroes commands if any unknown flag is set. | ||
194 | - */ | ||
195 | - if (unlikely(flags & ~VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP)) { | ||
196 | - return VIRTIO_BLK_S_UNSUPP; | ||
197 | - } | ||
198 | - | ||
199 | - if (type == VIRTIO_BLK_T_WRITE_ZEROES) { | ||
200 | - int blk_flags = 0; | ||
201 | - | ||
202 | - if (flags & VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP) { | ||
203 | - blk_flags |= BDRV_REQ_MAY_UNMAP; | ||
204 | - } | ||
205 | - | ||
206 | - if (blk_co_pwrite_zeroes(blk, sector << VIRTIO_BLK_SECTOR_BITS, | ||
207 | - bytes, blk_flags) == 0) { | ||
208 | - return VIRTIO_BLK_S_OK; | ||
209 | - } | ||
210 | - } else if (type == VIRTIO_BLK_T_DISCARD) { | ||
211 | - /* | ||
212 | - * The device MUST set the status byte to VIRTIO_BLK_S_UNSUPP for | ||
213 | - * discard commands if the unmap flag is set. | ||
214 | - */ | ||
215 | - if (unlikely(flags & VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP)) { | ||
216 | - return VIRTIO_BLK_S_UNSUPP; | ||
217 | - } | ||
218 | - | ||
219 | - if (blk_co_pdiscard(blk, sector << VIRTIO_BLK_SECTOR_BITS, | ||
220 | - bytes) == 0) { | ||
221 | - return VIRTIO_BLK_S_OK; | ||
222 | - } | ||
223 | - } | ||
224 | - | ||
225 | - return VIRTIO_BLK_S_IOERR; | ||
226 | -} | ||
227 | - | ||
228 | /* Called with server refcount increased, must decrease before returning */ | ||
229 | static void coroutine_fn vu_blk_virtio_process_req(void *opaque) | ||
230 | { | ||
231 | VuBlkReq *req = opaque; | ||
232 | VuServer *server = req->server; | ||
233 | VuVirtqElement *elem = &req->elem; | ||
234 | - uint32_t type; | ||
235 | - | ||
236 | VuBlkExport *vexp = container_of(server, VuBlkExport, vu_server); | ||
237 | - BlockBackend *blk = vexp->export.blk; | ||
238 | - | ||
239 | + VirtioBlkHandler *handler = &vexp->handler; | ||
240 | struct iovec *in_iov = elem->in_sg; | ||
241 | struct iovec *out_iov = elem->out_sg; | ||
242 | unsigned in_num = elem->in_num; | ||
243 | unsigned out_num = elem->out_num; | ||
244 | - | ||
245 | - /* refer to hw/block/virtio_blk.c */ | ||
246 | - if (elem->out_num < 1 || elem->in_num < 1) { | ||
247 | - error_report("virtio-blk request missing headers"); | ||
248 | - goto err; | ||
249 | - } | ||
250 | - | ||
251 | - if (unlikely(iov_to_buf(out_iov, out_num, 0, &req->out, | ||
252 | - sizeof(req->out)) != sizeof(req->out))) { | ||
253 | - error_report("virtio-blk request outhdr too short"); | ||
254 | - goto err; | ||
255 | - } | ||
256 | - | ||
257 | - iov_discard_front(&out_iov, &out_num, sizeof(req->out)); | ||
258 | - | ||
259 | - if (in_iov[in_num - 1].iov_len < sizeof(struct virtio_blk_inhdr)) { | ||
260 | - error_report("virtio-blk request inhdr too short"); | ||
261 | - goto err; | ||
262 | - } | ||
263 | - | ||
264 | - req->size = iov_size(in_iov, in_num); | ||
265 | - /* We always touch the last byte, so just see how big in_iov is. */ | ||
266 | - req->in = (void *)in_iov[in_num - 1].iov_base | ||
267 | - + in_iov[in_num - 1].iov_len | ||
268 | - - sizeof(struct virtio_blk_inhdr); | ||
269 | - iov_discard_back(in_iov, &in_num, sizeof(struct virtio_blk_inhdr)); | ||
270 | - | ||
271 | - type = le32_to_cpu(req->out.type); | ||
272 | - switch (type & ~VIRTIO_BLK_T_BARRIER) { | ||
273 | - case VIRTIO_BLK_T_IN: | ||
274 | - case VIRTIO_BLK_T_OUT: { | ||
275 | - QEMUIOVector qiov; | ||
276 | - int64_t offset; | ||
277 | - ssize_t ret = 0; | ||
278 | - bool is_write = type & VIRTIO_BLK_T_OUT; | ||
279 | - req->sector_num = le64_to_cpu(req->out.sector); | ||
280 | - | ||
281 | - if (is_write && !vexp->writable) { | ||
282 | - req->in->status = VIRTIO_BLK_S_IOERR; | ||
283 | - break; | ||
284 | - } | ||
285 | - | ||
286 | - if (is_write) { | ||
287 | - qemu_iovec_init_external(&qiov, out_iov, out_num); | ||
288 | - } else { | ||
289 | - qemu_iovec_init_external(&qiov, in_iov, in_num); | ||
290 | - } | ||
291 | - | ||
292 | - if (unlikely(!vu_blk_sect_range_ok(vexp, | ||
293 | - req->sector_num, | ||
294 | - qiov.size))) { | ||
295 | - req->in->status = VIRTIO_BLK_S_IOERR; | ||
296 | - break; | ||
297 | - } | ||
298 | - | ||
299 | - offset = req->sector_num << VIRTIO_BLK_SECTOR_BITS; | ||
300 | - | ||
301 | - if (is_write) { | ||
302 | - ret = blk_co_pwritev(blk, offset, qiov.size, &qiov, 0); | ||
303 | - } else { | ||
304 | - ret = blk_co_preadv(blk, offset, qiov.size, &qiov, 0); | ||
305 | - } | ||
306 | - if (ret >= 0) { | ||
307 | - req->in->status = VIRTIO_BLK_S_OK; | ||
308 | - } else { | ||
309 | - req->in->status = VIRTIO_BLK_S_IOERR; | ||
310 | - } | ||
311 | - break; | ||
312 | - } | ||
313 | - case VIRTIO_BLK_T_FLUSH: | ||
314 | - if (blk_co_flush(blk) == 0) { | ||
315 | - req->in->status = VIRTIO_BLK_S_OK; | ||
316 | - } else { | ||
317 | - req->in->status = VIRTIO_BLK_S_IOERR; | ||
318 | - } | ||
319 | - break; | ||
320 | - case VIRTIO_BLK_T_GET_ID: { | ||
321 | - size_t size = MIN(iov_size(&elem->in_sg[0], in_num), | ||
322 | - VIRTIO_BLK_ID_BYTES); | ||
323 | - snprintf(elem->in_sg[0].iov_base, size, "%s", "vhost_user_blk"); | ||
324 | - req->in->status = VIRTIO_BLK_S_OK; | ||
325 | - break; | ||
326 | + int in_len; | ||
327 | + | ||
328 | + in_len = virtio_blk_process_req(handler, in_iov, out_iov, | ||
329 | + in_num, out_num); | ||
330 | + if (in_len < 0) { | ||
331 | + free(req); | ||
332 | + vhost_user_server_unref(server); | ||
333 | + return; | ||
334 | } | ||
335 | - case VIRTIO_BLK_T_DISCARD: | ||
336 | - case VIRTIO_BLK_T_WRITE_ZEROES: { | ||
337 | - if (!vexp->writable) { | ||
338 | - req->in->status = VIRTIO_BLK_S_IOERR; | ||
339 | - break; | ||
340 | - } | ||
341 | - | ||
342 | - req->in->status = vu_blk_discard_write_zeroes(vexp, out_iov, out_num, | ||
343 | - type); | ||
344 | - break; | ||
345 | - } | ||
346 | - default: | ||
347 | - req->in->status = VIRTIO_BLK_S_UNSUPP; | ||
348 | - break; | ||
349 | - } | ||
350 | - | ||
351 | - vu_blk_req_complete(req); | ||
352 | - vhost_user_server_unref(server); | ||
353 | - return; | ||
354 | |||
355 | -err: | ||
356 | - free(req); | ||
357 | + vu_blk_req_complete(req, in_len); | ||
358 | vhost_user_server_unref(server); | ||
359 | } | ||
360 | |||
361 | @@ -XXX,XX +XXX,XX @@ static uint64_t vu_blk_get_features(VuDev *dev) | ||
362 | 1ull << VIRTIO_RING_F_EVENT_IDX | | ||
363 | 1ull << VHOST_USER_F_PROTOCOL_FEATURES; | ||
364 | |||
365 | - if (!vexp->writable) { | ||
366 | + if (!vexp->handler.writable) { | ||
367 | features |= 1ull << VIRTIO_BLK_F_RO; | ||
368 | } | ||
369 | |||
370 | @@ -XXX,XX +XXX,XX @@ vu_blk_initialize_config(BlockDriverState *bs, | ||
371 | config->opt_io_size = cpu_to_le32(1); | ||
372 | config->num_queues = cpu_to_le16(num_queues); | ||
373 | config->max_discard_sectors = | ||
374 | - cpu_to_le32(VHOST_USER_BLK_MAX_DISCARD_SECTORS); | ||
375 | + cpu_to_le32(VIRTIO_BLK_MAX_DISCARD_SECTORS); | ||
376 | config->max_discard_seg = cpu_to_le32(1); | ||
377 | config->discard_sector_alignment = | ||
378 | cpu_to_le32(blk_size >> VIRTIO_BLK_SECTOR_BITS); | ||
379 | config->max_write_zeroes_sectors | ||
380 | - = cpu_to_le32(VHOST_USER_BLK_MAX_WRITE_ZEROES_SECTORS); | ||
381 | + = cpu_to_le32(VIRTIO_BLK_MAX_WRITE_ZEROES_SECTORS); | ||
382 | config->max_write_zeroes_seg = cpu_to_le32(1); | ||
383 | } | ||
384 | |||
385 | @@ -XXX,XX +XXX,XX @@ static int vu_blk_exp_create(BlockExport *exp, BlockExportOptions *opts, | ||
386 | uint64_t logical_block_size; | ||
387 | uint16_t num_queues = VHOST_USER_BLK_NUM_QUEUES_DEFAULT; | ||
388 | |||
389 | - vexp->writable = opts->writable; | ||
390 | vexp->blkcfg.wce = 0; | ||
391 | |||
392 | if (vu_opts->has_logical_block_size) { | ||
393 | @@ -XXX,XX +XXX,XX @@ static int vu_blk_exp_create(BlockExport *exp, BlockExportOptions *opts, | ||
394 | error_propagate(errp, local_err); | ||
395 | return -EINVAL; | ||
396 | } | ||
397 | - vexp->blk_size = logical_block_size; | ||
398 | |||
399 | if (vu_opts->has_num_queues) { | ||
400 | num_queues = vu_opts->num_queues; | ||
401 | @@ -XXX,XX +XXX,XX @@ static int vu_blk_exp_create(BlockExport *exp, BlockExportOptions *opts, | ||
402 | error_setg(errp, "num-queues must be greater than 0"); | ||
403 | return -EINVAL; | ||
404 | } | ||
405 | + vexp->handler.blk = exp->blk; | ||
406 | + vexp->handler.serial = "vhost_user_blk"; | ||
407 | + vexp->handler.logical_block_size = logical_block_size; | ||
408 | + vexp->handler.writable = opts->writable; | ||
409 | |||
410 | vu_blk_initialize_config(blk_bs(exp->blk), &vexp->blkcfg, | ||
411 | logical_block_size, num_queues); | ||
412 | diff --git a/block/export/virtio-blk-handler.c b/block/export/virtio-blk-handler.c | ||
413 | new file mode 100644 | ||
414 | index XXXXXXX..XXXXXXX | ||
415 | --- /dev/null | ||
416 | +++ b/block/export/virtio-blk-handler.c | ||
417 | @@ -XXX,XX +XXX,XX @@ | ||
418 | +/* | ||
419 | + * Handler for virtio-blk I/O | ||
420 | + * | ||
421 | + * Copyright (c) 2020 Red Hat, Inc. | ||
422 | + * Copyright (C) 2022 Bytedance Inc. and/or its affiliates. All rights reserved. | ||
423 | + * | ||
424 | + * Author: | ||
425 | + * Coiby Xu <coiby.xu@gmail.com> | ||
426 | + * Xie Yongji <xieyongji@bytedance.com> | ||
427 | + * | ||
428 | + * This work is licensed under the terms of the GNU GPL, version 2 or | ||
429 | + * later. See the COPYING file in the top-level directory. | ||
430 | + */ | ||
431 | + | ||
432 | +#include "qemu/osdep.h" | ||
433 | +#include "qemu/error-report.h" | ||
434 | +#include "virtio-blk-handler.h" | ||
435 | + | ||
436 | +#include "standard-headers/linux/virtio_blk.h" | ||
437 | + | ||
438 | +struct virtio_blk_inhdr { | ||
439 | + unsigned char status; | ||
440 | +}; | ||
441 | + | ||
442 | +static bool virtio_blk_sect_range_ok(BlockBackend *blk, uint32_t block_size, | ||
443 | + uint64_t sector, size_t size) | ||
444 | +{ | ||
445 | + uint64_t nb_sectors; | ||
446 | + uint64_t total_sectors; | ||
447 | + | ||
448 | + if (size % VIRTIO_BLK_SECTOR_SIZE) { | ||
449 | + return false; | ||
450 | + } | ||
451 | + | ||
452 | + nb_sectors = size >> VIRTIO_BLK_SECTOR_BITS; | ||
453 | + | ||
454 | + QEMU_BUILD_BUG_ON(BDRV_SECTOR_SIZE != VIRTIO_BLK_SECTOR_SIZE); | ||
455 | + if (nb_sectors > BDRV_REQUEST_MAX_SECTORS) { | ||
456 | + return false; | ||
457 | + } | ||
458 | + if ((sector << VIRTIO_BLK_SECTOR_BITS) % block_size) { | ||
459 | + return false; | ||
460 | + } | ||
461 | + blk_get_geometry(blk, &total_sectors); | ||
462 | + if (sector > total_sectors || nb_sectors > total_sectors - sector) { | ||
463 | + return false; | ||
464 | + } | ||
465 | + return true; | ||
466 | +} | ||
467 | + | ||
468 | +static int coroutine_fn | ||
469 | +virtio_blk_discard_write_zeroes(VirtioBlkHandler *handler, struct iovec *iov, | ||
470 | + uint32_t iovcnt, uint32_t type) | ||
471 | +{ | ||
472 | + BlockBackend *blk = handler->blk; | ||
473 | + struct virtio_blk_discard_write_zeroes desc; | ||
474 | + ssize_t size; | ||
475 | + uint64_t sector; | ||
476 | + uint32_t num_sectors; | ||
477 | + uint32_t max_sectors; | ||
478 | + uint32_t flags; | ||
479 | + int bytes; | ||
480 | + | ||
481 | + /* Only one desc is currently supported */ | ||
482 | + if (unlikely(iov_size(iov, iovcnt) > sizeof(desc))) { | ||
483 | + return VIRTIO_BLK_S_UNSUPP; | ||
484 | + } | ||
485 | + | ||
486 | + size = iov_to_buf(iov, iovcnt, 0, &desc, sizeof(desc)); | ||
487 | + if (unlikely(size != sizeof(desc))) { | ||
488 | + error_report("Invalid size %zd, expected %zu", size, sizeof(desc)); | ||
489 | + return VIRTIO_BLK_S_IOERR; | ||
490 | + } | ||
491 | + | ||
492 | + sector = le64_to_cpu(desc.sector); | ||
493 | + num_sectors = le32_to_cpu(desc.num_sectors); | ||
494 | + flags = le32_to_cpu(desc.flags); | ||
495 | + max_sectors = (type == VIRTIO_BLK_T_WRITE_ZEROES) ? | ||
496 | + VIRTIO_BLK_MAX_WRITE_ZEROES_SECTORS : | ||
497 | + VIRTIO_BLK_MAX_DISCARD_SECTORS; | ||
498 | + | ||
499 | + /* This check ensures that 'bytes' fits in an int */ | ||
500 | + if (unlikely(num_sectors > max_sectors)) { | ||
501 | + return VIRTIO_BLK_S_IOERR; | ||
502 | + } | ||
503 | + | ||
504 | + bytes = num_sectors << VIRTIO_BLK_SECTOR_BITS; | ||
505 | + | ||
506 | + if (unlikely(!virtio_blk_sect_range_ok(blk, handler->logical_block_size, | ||
507 | + sector, bytes))) { | ||
508 | + return VIRTIO_BLK_S_IOERR; | ||
509 | + } | ||
510 | + | ||
511 | + /* | ||
512 | + * The device MUST set the status byte to VIRTIO_BLK_S_UNSUPP for discard | ||
513 | + * and write zeroes commands if any unknown flag is set. | ||
514 | + */ | ||
515 | + if (unlikely(flags & ~VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP)) { | ||
516 | + return VIRTIO_BLK_S_UNSUPP; | ||
517 | + } | ||
518 | + | ||
519 | + if (type == VIRTIO_BLK_T_WRITE_ZEROES) { | ||
520 | + int blk_flags = 0; | ||
521 | + | ||
522 | + if (flags & VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP) { | ||
523 | + blk_flags |= BDRV_REQ_MAY_UNMAP; | ||
524 | + } | ||
525 | + | ||
526 | + if (blk_co_pwrite_zeroes(blk, sector << VIRTIO_BLK_SECTOR_BITS, | ||
527 | + bytes, blk_flags) == 0) { | ||
528 | + return VIRTIO_BLK_S_OK; | ||
529 | + } | ||
530 | + } else if (type == VIRTIO_BLK_T_DISCARD) { | ||
531 | + /* | ||
532 | + * The device MUST set the status byte to VIRTIO_BLK_S_UNSUPP for | ||
533 | + * discard commands if the unmap flag is set. | ||
534 | + */ | ||
535 | + if (unlikely(flags & VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP)) { | ||
536 | + return VIRTIO_BLK_S_UNSUPP; | ||
537 | + } | ||
538 | + | ||
539 | + if (blk_co_pdiscard(blk, sector << VIRTIO_BLK_SECTOR_BITS, | ||
540 | + bytes) == 0) { | ||
541 | + return VIRTIO_BLK_S_OK; | ||
542 | + } | ||
543 | + } | ||
544 | + | ||
545 | + return VIRTIO_BLK_S_IOERR; | ||
546 | +} | ||
547 | + | ||
548 | +int coroutine_fn virtio_blk_process_req(VirtioBlkHandler *handler, | ||
549 | + struct iovec *in_iov, | ||
550 | + struct iovec *out_iov, | ||
551 | + unsigned int in_num, | ||
552 | + unsigned int out_num) | ||
553 | +{ | ||
554 | + BlockBackend *blk = handler->blk; | ||
555 | + struct virtio_blk_inhdr *in; | ||
556 | + struct virtio_blk_outhdr out; | ||
557 | + uint32_t type; | ||
558 | + int in_len; | ||
559 | + | ||
560 | + if (out_num < 1 || in_num < 1) { | ||
561 | + error_report("virtio-blk request missing headers"); | ||
562 | + return -EINVAL; | ||
563 | + } | ||
564 | + | ||
565 | + if (unlikely(iov_to_buf(out_iov, out_num, 0, &out, | ||
566 | + sizeof(out)) != sizeof(out))) { | ||
567 | + error_report("virtio-blk request outhdr too short"); | ||
568 | + return -EINVAL; | ||
569 | + } | ||
570 | + | ||
571 | + iov_discard_front(&out_iov, &out_num, sizeof(out)); | ||
572 | + | ||
573 | + if (in_iov[in_num - 1].iov_len < sizeof(struct virtio_blk_inhdr)) { | ||
574 | + error_report("virtio-blk request inhdr too short"); | ||
575 | + return -EINVAL; | ||
576 | + } | ||
577 | + | ||
578 | + /* We always touch the last byte, so just see how big in_iov is. */ | ||
579 | + in_len = iov_size(in_iov, in_num); | ||
580 | + in = (void *)in_iov[in_num - 1].iov_base | ||
581 | + + in_iov[in_num - 1].iov_len | ||
582 | + - sizeof(struct virtio_blk_inhdr); | ||
583 | + iov_discard_back(in_iov, &in_num, sizeof(struct virtio_blk_inhdr)); | ||
584 | + | ||
585 | + type = le32_to_cpu(out.type); | ||
586 | + switch (type & ~VIRTIO_BLK_T_BARRIER) { | ||
587 | + case VIRTIO_BLK_T_IN: | ||
588 | + case VIRTIO_BLK_T_OUT: { | ||
589 | + QEMUIOVector qiov; | ||
590 | + int64_t offset; | ||
591 | + ssize_t ret = 0; | ||
592 | + bool is_write = type & VIRTIO_BLK_T_OUT; | ||
593 | + int64_t sector_num = le64_to_cpu(out.sector); | ||
594 | + | ||
595 | + if (is_write && !handler->writable) { | ||
596 | + in->status = VIRTIO_BLK_S_IOERR; | ||
597 | + break; | ||
598 | + } | ||
599 | + | ||
600 | + if (is_write) { | ||
601 | + qemu_iovec_init_external(&qiov, out_iov, out_num); | ||
602 | + } else { | ||
603 | + qemu_iovec_init_external(&qiov, in_iov, in_num); | ||
604 | + } | ||
605 | + | ||
606 | + if (unlikely(!virtio_blk_sect_range_ok(blk, | ||
607 | + handler->logical_block_size, | ||
608 | + sector_num, qiov.size))) { | ||
609 | + in->status = VIRTIO_BLK_S_IOERR; | ||
610 | + break; | ||
611 | + } | ||
612 | + | ||
613 | + offset = sector_num << VIRTIO_BLK_SECTOR_BITS; | ||
614 | + | ||
615 | + if (is_write) { | ||
616 | + ret = blk_co_pwritev(blk, offset, qiov.size, &qiov, 0); | ||
617 | + } else { | ||
618 | + ret = blk_co_preadv(blk, offset, qiov.size, &qiov, 0); | ||
619 | + } | ||
620 | + if (ret >= 0) { | ||
621 | + in->status = VIRTIO_BLK_S_OK; | ||
622 | + } else { | ||
623 | + in->status = VIRTIO_BLK_S_IOERR; | ||
624 | + } | ||
625 | + break; | ||
626 | + } | ||
627 | + case VIRTIO_BLK_T_FLUSH: | ||
628 | + if (blk_co_flush(blk) == 0) { | ||
629 | + in->status = VIRTIO_BLK_S_OK; | ||
630 | + } else { | ||
631 | + in->status = VIRTIO_BLK_S_IOERR; | ||
632 | + } | ||
633 | + break; | ||
634 | + case VIRTIO_BLK_T_GET_ID: { | ||
635 | + size_t size = MIN(strlen(handler->serial) + 1, | ||
636 | + MIN(iov_size(in_iov, in_num), | ||
637 | + VIRTIO_BLK_ID_BYTES)); | ||
638 | + iov_from_buf(in_iov, in_num, 0, handler->serial, size); | ||
639 | + in->status = VIRTIO_BLK_S_OK; | ||
640 | + break; | ||
641 | + } | ||
642 | + case VIRTIO_BLK_T_DISCARD: | ||
643 | + case VIRTIO_BLK_T_WRITE_ZEROES: | ||
644 | + if (!handler->writable) { | ||
645 | + in->status = VIRTIO_BLK_S_IOERR; | ||
646 | + break; | ||
647 | + } | ||
648 | + in->status = virtio_blk_discard_write_zeroes(handler, out_iov, | ||
649 | + out_num, type); | ||
650 | + break; | ||
651 | + default: | ||
652 | + in->status = VIRTIO_BLK_S_UNSUPP; | ||
653 | + break; | ||
654 | + } | ||
655 | + | ||
656 | + return in_len; | ||
657 | +} | ||
658 | diff --git a/MAINTAINERS b/MAINTAINERS | ||
659 | index XXXXXXX..XXXXXXX 100644 | ||
660 | --- a/MAINTAINERS | ||
661 | +++ b/MAINTAINERS | ||
662 | @@ -XXX,XX +XXX,XX @@ M: Coiby Xu <Coiby.Xu@gmail.com> | ||
663 | S: Maintained | ||
664 | F: block/export/vhost-user-blk-server.c | ||
665 | F: block/export/vhost-user-blk-server.h | ||
666 | +F: block/export/virtio-blk-handler.c | ||
667 | +F: block/export/virtio-blk-handler.h | ||
668 | F: include/qemu/vhost-user-server.h | ||
669 | F: tests/qtest/libqos/vhost-user-blk.c | ||
670 | F: tests/qtest/libqos/vhost-user-blk.h | ||
671 | diff --git a/block/export/meson.build b/block/export/meson.build | ||
672 | index XXXXXXX..XXXXXXX 100644 | ||
673 | --- a/block/export/meson.build | ||
674 | +++ b/block/export/meson.build | ||
675 | @@ -XXX,XX +XXX,XX @@ | ||
676 | blockdev_ss.add(files('export.c')) | ||
677 | |||
678 | if have_vhost_user_blk_server | ||
679 | - blockdev_ss.add(files('vhost-user-blk-server.c')) | ||
680 | + blockdev_ss.add(files('vhost-user-blk-server.c', 'virtio-blk-handler.c')) | ||
681 | endif | ||
682 | |||
683 | blockdev_ss.add(when: fuse, if_true: files('fuse.c')) | ||
684 | -- | ||
685 | 2.35.3 | diff view generated by jsdifflib |
1 | From: Philippe Mathieu-Daudé <philmd@redhat.com> | 1 | From: Xie Yongji <xieyongji@bytedance.com> |
---|---|---|---|
2 | 2 | ||
3 | The socket_scm_helper path got corrupted during the mechanical | 3 | This adds vduse header to linux headers so that the |
4 | refactor moving the qtests files into their own sub-directory. | 4 | relevant VDUSE API can be used in subsequent patches. |
5 | 5 | ||
6 | Fixes: 1e8a1fae7 ("test: Move qtests to a separate directory") | 6 | Signed-off-by: Xie Yongji <xieyongji@bytedance.com> |
7 | Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com> | 7 | Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> |
8 | Message-Id: <20200306165751.18986-1-philmd@redhat.com> | 8 | Message-Id: <20220523084611.91-5-xieyongji@bytedance.com> |
9 | Reviewed-by: Laurent Vivier <laurent@vivier.eu> | ||
10 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> | 9 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> |
11 | --- | 10 | --- |
12 | tests/Makefile.include | 1 + | 11 | linux-headers/linux/vduse.h | 306 ++++++++++++++++++++++++++++++++ |
13 | tests/qtest/Makefile.include | 1 - | 12 | scripts/update-linux-headers.sh | 2 +- |
14 | 2 files changed, 1 insertion(+), 1 deletion(-) | 13 | 2 files changed, 307 insertions(+), 1 deletion(-) |
14 | create mode 100644 linux-headers/linux/vduse.h | ||
15 | 15 | ||
16 | diff --git a/tests/Makefile.include b/tests/Makefile.include | 16 | diff --git a/linux-headers/linux/vduse.h b/linux-headers/linux/vduse.h |
17 | index XXXXXXX..XXXXXXX 100644 | 17 | new file mode 100644 |
18 | --- a/tests/Makefile.include | 18 | index XXXXXXX..XXXXXXX |
19 | +++ b/tests/Makefile.include | 19 | --- /dev/null |
20 | @@ -XXX,XX +XXX,XX @@ include $(SRC_PATH)/tests/qtest/Makefile.include | 20 | +++ b/linux-headers/linux/vduse.h |
21 | tests/test-qga$(EXESUF): qemu-ga$(EXESUF) | 21 | @@ -XXX,XX +XXX,XX @@ |
22 | tests/test-qga$(EXESUF): tests/test-qga.o $(qtest-obj-y) | 22 | +/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */ |
23 | tests/vhost-user-bridge$(EXESUF): tests/vhost-user-bridge.o $(test-util-obj-y) libvhost-user.a | 23 | +#ifndef _VDUSE_H_ |
24 | +tests/qemu-iotests/socket_scm_helper$(EXESUF): tests/qemu-iotests/socket_scm_helper.o | 24 | +#define _VDUSE_H_ |
25 | 25 | + | |
26 | SPEED = quick | 26 | +#include <linux/types.h> |
27 | 27 | + | |
28 | diff --git a/tests/qtest/Makefile.include b/tests/qtest/Makefile.include | 28 | +#define VDUSE_BASE 0x81 |
29 | index XXXXXXX..XXXXXXX 100644 | 29 | + |
30 | --- a/tests/qtest/Makefile.include | 30 | +/* The ioctls for control device (/dev/vduse/control) */ |
31 | +++ b/tests/qtest/Makefile.include | 31 | + |
32 | @@ -XXX,XX +XXX,XX @@ tests/qtest/usb-hcd-ehci-test$(EXESUF): tests/qtest/usb-hcd-ehci-test.o $(libqos | 32 | +#define VDUSE_API_VERSION 0 |
33 | tests/qtest/usb-hcd-xhci-test$(EXESUF): tests/qtest/usb-hcd-xhci-test.o $(libqos-usb-obj-y) | 33 | + |
34 | tests/qtest/cpu-plug-test$(EXESUF): tests/qtest/cpu-plug-test.o | 34 | +/* |
35 | tests/qtest/migration-test$(EXESUF): tests/qtest/migration-test.o tests/qtest/migration-helpers.o | 35 | + * Get the version of VDUSE API that kernel supported (VDUSE_API_VERSION). |
36 | -tests/qtest/qemu-iotests/qtest/socket_scm_helper$(EXESUF): tests/qtest/qemu-iotests/qtest/socket_scm_helper.o | 36 | + * This is used for future extension. |
37 | tests/qtest/test-netfilter$(EXESUF): tests/qtest/test-netfilter.o $(qtest-obj-y) | 37 | + */ |
38 | tests/qtest/test-filter-mirror$(EXESUF): tests/qtest/test-filter-mirror.o $(qtest-obj-y) | 38 | +#define VDUSE_GET_API_VERSION _IOR(VDUSE_BASE, 0x00, __u64) |
39 | tests/qtest/test-filter-redirector$(EXESUF): tests/qtest/test-filter-redirector.o $(qtest-obj-y) | 39 | + |
40 | +/* Set the version of VDUSE API that userspace supported. */ | ||
41 | +#define VDUSE_SET_API_VERSION _IOW(VDUSE_BASE, 0x01, __u64) | ||
42 | + | ||
43 | +/** | ||
44 | + * struct vduse_dev_config - basic configuration of a VDUSE device | ||
45 | + * @name: VDUSE device name, needs to be NUL terminated | ||
46 | + * @vendor_id: virtio vendor id | ||
47 | + * @device_id: virtio device id | ||
48 | + * @features: virtio features | ||
49 | + * @vq_num: the number of virtqueues | ||
50 | + * @vq_align: the allocation alignment of virtqueue's metadata | ||
51 | + * @reserved: for future use, needs to be initialized to zero | ||
52 | + * @config_size: the size of the configuration space | ||
53 | + * @config: the buffer of the configuration space | ||
54 | + * | ||
55 | + * Structure used by VDUSE_CREATE_DEV ioctl to create VDUSE device. | ||
56 | + */ | ||
57 | +struct vduse_dev_config { | ||
58 | +#define VDUSE_NAME_MAX 256 | ||
59 | + char name[VDUSE_NAME_MAX]; | ||
60 | + __u32 vendor_id; | ||
61 | + __u32 device_id; | ||
62 | + __u64 features; | ||
63 | + __u32 vq_num; | ||
64 | + __u32 vq_align; | ||
65 | + __u32 reserved[13]; | ||
66 | + __u32 config_size; | ||
67 | + __u8 config[]; | ||
68 | +}; | ||
69 | + | ||
70 | +/* Create a VDUSE device which is represented by a char device (/dev/vduse/$NAME) */ | ||
71 | +#define VDUSE_CREATE_DEV _IOW(VDUSE_BASE, 0x02, struct vduse_dev_config) | ||
72 | + | ||
73 | +/* | ||
74 | + * Destroy a VDUSE device. Make sure there are no more references | ||
75 | + * to the char device (/dev/vduse/$NAME). | ||
76 | + */ | ||
77 | +#define VDUSE_DESTROY_DEV _IOW(VDUSE_BASE, 0x03, char[VDUSE_NAME_MAX]) | ||
78 | + | ||
79 | +/* The ioctls for VDUSE device (/dev/vduse/$NAME) */ | ||
80 | + | ||
81 | +/** | ||
82 | + * struct vduse_iotlb_entry - entry of IOTLB to describe one IOVA region [start, last] | ||
83 | + * @offset: the mmap offset on returned file descriptor | ||
84 | + * @start: start of the IOVA region | ||
85 | + * @last: last of the IOVA region | ||
86 | + * @perm: access permission of the IOVA region | ||
87 | + * | ||
88 | + * Structure used by VDUSE_IOTLB_GET_FD ioctl to find an overlapped IOVA region. | ||
89 | + */ | ||
90 | +struct vduse_iotlb_entry { | ||
91 | + __u64 offset; | ||
92 | + __u64 start; | ||
93 | + __u64 last; | ||
94 | +#define VDUSE_ACCESS_RO 0x1 | ||
95 | +#define VDUSE_ACCESS_WO 0x2 | ||
96 | +#define VDUSE_ACCESS_RW 0x3 | ||
97 | + __u8 perm; | ||
98 | +}; | ||
99 | + | ||
100 | +/* | ||
101 | + * Find the first IOVA region that overlaps with the range [start, last] | ||
102 | + * and return the corresponding file descriptor. Return -EINVAL means the | ||
103 | + * IOVA region doesn't exist. Caller should set start and last fields. | ||
104 | + */ | ||
105 | +#define VDUSE_IOTLB_GET_FD _IOWR(VDUSE_BASE, 0x10, struct vduse_iotlb_entry) | ||
106 | + | ||
107 | +/* | ||
108 | + * Get the negotiated virtio features. It's a subset of the features in | ||
109 | + * struct vduse_dev_config which can be accepted by virtio driver. It's | ||
110 | + * only valid after FEATURES_OK status bit is set. | ||
111 | + */ | ||
112 | +#define VDUSE_DEV_GET_FEATURES _IOR(VDUSE_BASE, 0x11, __u64) | ||
113 | + | ||
114 | +/** | ||
115 | + * struct vduse_config_data - data used to update configuration space | ||
116 | + * @offset: the offset from the beginning of configuration space | ||
117 | + * @length: the length to write to configuration space | ||
118 | + * @buffer: the buffer used to write from | ||
119 | + * | ||
120 | + * Structure used by VDUSE_DEV_SET_CONFIG ioctl to update device | ||
121 | + * configuration space. | ||
122 | + */ | ||
123 | +struct vduse_config_data { | ||
124 | + __u32 offset; | ||
125 | + __u32 length; | ||
126 | + __u8 buffer[]; | ||
127 | +}; | ||
128 | + | ||
129 | +/* Set device configuration space */ | ||
130 | +#define VDUSE_DEV_SET_CONFIG _IOW(VDUSE_BASE, 0x12, struct vduse_config_data) | ||
131 | + | ||
132 | +/* | ||
133 | + * Inject a config interrupt. It's usually used to notify virtio driver | ||
134 | + * that device configuration space has changed. | ||
135 | + */ | ||
136 | +#define VDUSE_DEV_INJECT_CONFIG_IRQ _IO(VDUSE_BASE, 0x13) | ||
137 | + | ||
138 | +/** | ||
139 | + * struct vduse_vq_config - basic configuration of a virtqueue | ||
140 | + * @index: virtqueue index | ||
141 | + * @max_size: the max size of virtqueue | ||
142 | + * @reserved: for future use, needs to be initialized to zero | ||
143 | + * | ||
144 | + * Structure used by VDUSE_VQ_SETUP ioctl to setup a virtqueue. | ||
145 | + */ | ||
146 | +struct vduse_vq_config { | ||
147 | + __u32 index; | ||
148 | + __u16 max_size; | ||
149 | + __u16 reserved[13]; | ||
150 | +}; | ||
151 | + | ||
152 | +/* | ||
153 | + * Setup the specified virtqueue. Make sure all virtqueues have been | ||
154 | + * configured before the device is attached to vDPA bus. | ||
155 | + */ | ||
156 | +#define VDUSE_VQ_SETUP _IOW(VDUSE_BASE, 0x14, struct vduse_vq_config) | ||
157 | + | ||
158 | +/** | ||
159 | + * struct vduse_vq_state_split - split virtqueue state | ||
160 | + * @avail_index: available index | ||
161 | + */ | ||
162 | +struct vduse_vq_state_split { | ||
163 | + __u16 avail_index; | ||
164 | +}; | ||
165 | + | ||
166 | +/** | ||
167 | + * struct vduse_vq_state_packed - packed virtqueue state | ||
168 | + * @last_avail_counter: last driver ring wrap counter observed by device | ||
169 | + * @last_avail_idx: device available index | ||
170 | + * @last_used_counter: device ring wrap counter | ||
171 | + * @last_used_idx: used index | ||
172 | + */ | ||
173 | +struct vduse_vq_state_packed { | ||
174 | + __u16 last_avail_counter; | ||
175 | + __u16 last_avail_idx; | ||
176 | + __u16 last_used_counter; | ||
177 | + __u16 last_used_idx; | ||
178 | +}; | ||
179 | + | ||
180 | +/** | ||
181 | + * struct vduse_vq_info - information of a virtqueue | ||
182 | + * @index: virtqueue index | ||
183 | + * @num: the size of virtqueue | ||
184 | + * @desc_addr: address of desc area | ||
185 | + * @driver_addr: address of driver area | ||
186 | + * @device_addr: address of device area | ||
187 | + * @split: split virtqueue state | ||
188 | + * @packed: packed virtqueue state | ||
189 | + * @ready: ready status of virtqueue | ||
190 | + * | ||
191 | + * Structure used by VDUSE_VQ_GET_INFO ioctl to get virtqueue's information. | ||
192 | + */ | ||
193 | +struct vduse_vq_info { | ||
194 | + __u32 index; | ||
195 | + __u32 num; | ||
196 | + __u64 desc_addr; | ||
197 | + __u64 driver_addr; | ||
198 | + __u64 device_addr; | ||
199 | + union { | ||
200 | + struct vduse_vq_state_split split; | ||
201 | + struct vduse_vq_state_packed packed; | ||
202 | + }; | ||
203 | + __u8 ready; | ||
204 | +}; | ||
205 | + | ||
206 | +/* Get the specified virtqueue's information. Caller should set index field. */ | ||
207 | +#define VDUSE_VQ_GET_INFO _IOWR(VDUSE_BASE, 0x15, struct vduse_vq_info) | ||
208 | + | ||
209 | +/** | ||
210 | + * struct vduse_vq_eventfd - eventfd configuration for a virtqueue | ||
211 | + * @index: virtqueue index | ||
212 | + * @fd: eventfd, -1 means de-assigning the eventfd | ||
213 | + * | ||
214 | + * Structure used by VDUSE_VQ_SETUP_KICKFD ioctl to setup kick eventfd. | ||
215 | + */ | ||
216 | +struct vduse_vq_eventfd { | ||
217 | + __u32 index; | ||
218 | +#define VDUSE_EVENTFD_DEASSIGN -1 | ||
219 | + int fd; | ||
220 | +}; | ||
221 | + | ||
222 | +/* | ||
223 | + * Setup kick eventfd for specified virtqueue. The kick eventfd is used | ||
224 | + * by VDUSE kernel module to notify userspace to consume the avail vring. | ||
225 | + */ | ||
226 | +#define VDUSE_VQ_SETUP_KICKFD _IOW(VDUSE_BASE, 0x16, struct vduse_vq_eventfd) | ||
227 | + | ||
228 | +/* | ||
229 | + * Inject an interrupt for specific virtqueue. It's used to notify virtio driver | ||
230 | + * to consume the used vring. | ||
231 | + */ | ||
232 | +#define VDUSE_VQ_INJECT_IRQ _IOW(VDUSE_BASE, 0x17, __u32) | ||
233 | + | ||
234 | +/* The control messages definition for read(2)/write(2) on /dev/vduse/$NAME */ | ||
235 | + | ||
236 | +/** | ||
237 | + * enum vduse_req_type - request type | ||
238 | + * @VDUSE_GET_VQ_STATE: get the state for specified virtqueue from userspace | ||
239 | + * @VDUSE_SET_STATUS: set the device status | ||
240 | + * @VDUSE_UPDATE_IOTLB: Notify userspace to update the memory mapping for | ||
241 | + * specified IOVA range via VDUSE_IOTLB_GET_FD ioctl | ||
242 | + */ | ||
243 | +enum vduse_req_type { | ||
244 | + VDUSE_GET_VQ_STATE, | ||
245 | + VDUSE_SET_STATUS, | ||
246 | + VDUSE_UPDATE_IOTLB, | ||
247 | +}; | ||
248 | + | ||
249 | +/** | ||
250 | + * struct vduse_vq_state - virtqueue state | ||
251 | + * @index: virtqueue index | ||
252 | + * @split: split virtqueue state | ||
253 | + * @packed: packed virtqueue state | ||
254 | + */ | ||
255 | +struct vduse_vq_state { | ||
256 | + __u32 index; | ||
257 | + union { | ||
258 | + struct vduse_vq_state_split split; | ||
259 | + struct vduse_vq_state_packed packed; | ||
260 | + }; | ||
261 | +}; | ||
262 | + | ||
263 | +/** | ||
264 | + * struct vduse_dev_status - device status | ||
265 | + * @status: device status | ||
266 | + */ | ||
267 | +struct vduse_dev_status { | ||
268 | + __u8 status; | ||
269 | +}; | ||
270 | + | ||
271 | +/** | ||
272 | + * struct vduse_iova_range - IOVA range [start, last] | ||
273 | + * @start: start of the IOVA range | ||
274 | + * @last: last of the IOVA range | ||
275 | + */ | ||
276 | +struct vduse_iova_range { | ||
277 | + __u64 start; | ||
278 | + __u64 last; | ||
279 | +}; | ||
280 | + | ||
281 | +/** | ||
282 | + * struct vduse_dev_request - control request | ||
283 | + * @type: request type | ||
284 | + * @request_id: request id | ||
285 | + * @reserved: for future use | ||
286 | + * @vq_state: virtqueue state, only index field is available | ||
287 | + * @s: device status | ||
288 | + * @iova: IOVA range for updating | ||
289 | + * @padding: padding | ||
290 | + * | ||
291 | + * Structure used by read(2) on /dev/vduse/$NAME. | ||
292 | + */ | ||
293 | +struct vduse_dev_request { | ||
294 | + __u32 type; | ||
295 | + __u32 request_id; | ||
296 | + __u32 reserved[4]; | ||
297 | + union { | ||
298 | + struct vduse_vq_state vq_state; | ||
299 | + struct vduse_dev_status s; | ||
300 | + struct vduse_iova_range iova; | ||
301 | + __u32 padding[32]; | ||
302 | + }; | ||
303 | +}; | ||
304 | + | ||
305 | +/** | ||
306 | + * struct vduse_dev_response - response to control request | ||
307 | + * @request_id: corresponding request id | ||
308 | + * @result: the result of request | ||
309 | + * @reserved: for future use, needs to be initialized to zero | ||
310 | + * @vq_state: virtqueue state | ||
311 | + * @padding: padding | ||
312 | + * | ||
313 | + * Structure used by write(2) on /dev/vduse/$NAME. | ||
314 | + */ | ||
315 | +struct vduse_dev_response { | ||
316 | + __u32 request_id; | ||
317 | +#define VDUSE_REQ_RESULT_OK 0x00 | ||
318 | +#define VDUSE_REQ_RESULT_FAILED 0x01 | ||
319 | + __u32 result; | ||
320 | + __u32 reserved[4]; | ||
321 | + union { | ||
322 | + struct vduse_vq_state vq_state; | ||
323 | + __u32 padding[32]; | ||
324 | + }; | ||
325 | +}; | ||
326 | + | ||
327 | +#endif /* _VDUSE_H_ */ | ||
328 | diff --git a/scripts/update-linux-headers.sh b/scripts/update-linux-headers.sh | ||
329 | index XXXXXXX..XXXXXXX 100755 | ||
330 | --- a/scripts/update-linux-headers.sh | ||
331 | +++ b/scripts/update-linux-headers.sh | ||
332 | @@ -XXX,XX +XXX,XX @@ done | ||
333 | rm -rf "$output/linux-headers/linux" | ||
334 | mkdir -p "$output/linux-headers/linux" | ||
335 | for header in kvm.h vfio.h vfio_ccw.h vfio_zdev.h vhost.h \ | ||
336 | - psci.h psp-sev.h userfaultfd.h mman.h; do | ||
337 | + psci.h psp-sev.h userfaultfd.h mman.h vduse.h; do | ||
338 | cp "$tmpdir/include/linux/$header" "$output/linux-headers/linux" | ||
339 | done | ||
340 | |||
40 | -- | 341 | -- |
41 | 2.20.1 | 342 | 2.35.3 |
42 | |||
43 | diff view generated by jsdifflib |
1 | From: Daniel Henrique Barboza <danielhb413@gmail.com> | 1 | From: Xie Yongji <xieyongji@bytedance.com> |
---|---|---|---|
2 | 2 | ||
3 | This patch adds a new test file to exercise the case where | 3 | VDUSE [1] is a linux framework that makes it possible to implement |
4 | qemu-img fails to complete for the LUKS format when a non-UTF8 | 4 | software-emulated vDPA devices in userspace. This adds a library |
5 | secret is used. | 5 | as a subproject to help implementing VDUSE backends in QEMU. |
6 | 6 | ||
7 | Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com> | 7 | [1] https://www.kernel.org/doc/html/latest/userspace-api/vduse.html |
8 | Message-Id: <20200130213907.2830642-5-danielhb413@gmail.com> | 8 | |
9 | Signed-off-by: Xie Yongji <xieyongji@bytedance.com> | ||
10 | Message-Id: <20220523084611.91-6-xieyongji@bytedance.com> | ||
11 | Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> | ||
9 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> | 12 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> |
10 | --- | 13 | --- |
11 | tests/qemu-iotests/282 | 67 ++++++++++++++++++++++++++++++++++++++ | 14 | meson_options.txt | 2 + |
12 | tests/qemu-iotests/282.out | 11 +++++++ | 15 | subprojects/libvduse/include/atomic.h | 1 + |
13 | tests/qemu-iotests/group | 1 + | 16 | subprojects/libvduse/include/compiler.h | 1 + |
14 | 3 files changed, 79 insertions(+) | 17 | subprojects/libvduse/libvduse.h | 235 ++++ |
15 | create mode 100755 tests/qemu-iotests/282 | 18 | subprojects/libvduse/libvduse.c | 1167 +++++++++++++++++++ |
16 | create mode 100644 tests/qemu-iotests/282.out | 19 | MAINTAINERS | 5 + |
20 | meson.build | 15 + | ||
21 | scripts/meson-buildoptions.sh | 3 + | ||
22 | subprojects/libvduse/linux-headers/linux | 1 + | ||
23 | subprojects/libvduse/meson.build | 10 + | ||
24 | subprojects/libvduse/standard-headers/linux | 1 + | ||
25 | 11 files changed, 1441 insertions(+) | ||
26 | create mode 120000 subprojects/libvduse/include/atomic.h | ||
27 | create mode 120000 subprojects/libvduse/include/compiler.h | ||
28 | create mode 100644 subprojects/libvduse/libvduse.h | ||
29 | create mode 100644 subprojects/libvduse/libvduse.c | ||
30 | create mode 120000 subprojects/libvduse/linux-headers/linux | ||
31 | create mode 100644 subprojects/libvduse/meson.build | ||
32 | create mode 120000 subprojects/libvduse/standard-headers/linux | ||
17 | 33 | ||
18 | diff --git a/tests/qemu-iotests/282 b/tests/qemu-iotests/282 | 34 | diff --git a/meson_options.txt b/meson_options.txt |
19 | new file mode 100755 | 35 | index XXXXXXX..XXXXXXX 100644 |
36 | --- a/meson_options.txt | ||
37 | +++ b/meson_options.txt | ||
38 | @@ -XXX,XX +XXX,XX @@ option('virtfs', type: 'feature', value: 'auto', | ||
39 | description: 'virtio-9p support') | ||
40 | option('virtiofsd', type: 'feature', value: 'auto', | ||
41 | description: 'build virtiofs daemon (virtiofsd)') | ||
42 | +option('libvduse', type: 'feature', value: 'auto', | ||
43 | + description: 'build VDUSE Library') | ||
44 | |||
45 | option('capstone', type: 'feature', value: 'auto', | ||
46 | description: 'Whether and how to find the capstone library') | ||
47 | diff --git a/subprojects/libvduse/include/atomic.h b/subprojects/libvduse/include/atomic.h | ||
48 | new file mode 120000 | ||
20 | index XXXXXXX..XXXXXXX | 49 | index XXXXXXX..XXXXXXX |
21 | --- /dev/null | 50 | --- /dev/null |
22 | +++ b/tests/qemu-iotests/282 | 51 | +++ b/subprojects/libvduse/include/atomic.h |
23 | @@ -XXX,XX +XXX,XX @@ | 52 | @@ -0,0 +1 @@ |
24 | +#!/usr/bin/env bash | 53 | +../../../include/qemu/atomic.h |
25 | +# | 54 | \ No newline at end of file |
26 | +# Test qemu-img file cleanup for LUKS when using a non-UTF8 secret | 55 | diff --git a/subprojects/libvduse/include/compiler.h b/subprojects/libvduse/include/compiler.h |
27 | +# | 56 | new file mode 120000 |
28 | +# Copyright (C) 2020, IBM Corporation. | 57 | index XXXXXXX..XXXXXXX |
29 | +# | 58 | --- /dev/null |
30 | +# This program is free software; you can redistribute it and/or modify | 59 | +++ b/subprojects/libvduse/include/compiler.h |
31 | +# it under the terms of the GNU General Public License as published by | 60 | @@ -0,0 +1 @@ |
32 | +# the Free Software Foundation; either version 2 of the License, or | 61 | +../../../include/qemu/compiler.h |
33 | +# (at your option) any later version. | 62 | \ No newline at end of file |
34 | +# | 63 | diff --git a/subprojects/libvduse/libvduse.h b/subprojects/libvduse/libvduse.h |
35 | +# This program is distributed in the hope that it will be useful, | ||
36 | +# but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
37 | +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
38 | +# GNU General Public License for more details. | ||
39 | +# | ||
40 | +# You should have received a copy of the GNU General Public License | ||
41 | +# along with this program. If not, see <http://www.gnu.org/licenses/>. | ||
42 | +# | ||
43 | + | ||
44 | +seq=`basename $0` | ||
45 | +echo "QA output created by $seq" | ||
46 | + | ||
47 | +status=1 # failure is the default! | ||
48 | +TEST_IMAGE_FILE='vol.img' | ||
49 | + | ||
50 | +_cleanup() | ||
51 | +{ | ||
52 | + _cleanup_test_img | ||
53 | + rm non_utf8_secret | ||
54 | + rm -f $TEST_IMAGE_FILE | ||
55 | +} | ||
56 | +trap "_cleanup; exit \$status" 0 1 2 3 15 | ||
57 | + | ||
58 | +# get standard environment, filters and checks | ||
59 | +. ./common.rc | ||
60 | +. ./common.filter | ||
61 | + | ||
62 | +_supported_fmt luks | ||
63 | +_supported_proto generic | ||
64 | +_unsupported_proto vxhs | ||
65 | + | ||
66 | +echo "== Create non-UTF8 secret ==" | ||
67 | +echo -n -e '\x3a\x3c\x3b\xff' > non_utf8_secret | ||
68 | +SECRET="secret,id=sec0,file=non_utf8_secret" | ||
69 | + | ||
70 | +echo "== Throws an error because of invalid UTF-8 secret ==" | ||
71 | +$QEMU_IMG create -f $IMGFMT --object $SECRET -o "key-secret=sec0" $TEST_IMAGE_FILE 4M | ||
72 | + | ||
73 | +echo "== Image file should not exist after the error ==" | ||
74 | +if test -f "$TEST_IMAGE_FILE"; then | ||
75 | + exit 1 | ||
76 | +fi | ||
77 | + | ||
78 | +echo "== Create a stub image file and run qemu-img again ==" | ||
79 | +touch $TEST_IMAGE_FILE | ||
80 | +$QEMU_IMG create -f $IMGFMT --object $SECRET -o "key-secret=sec0" $TEST_IMAGE_FILE 4M | ||
81 | + | ||
82 | +echo "== Pre-existing image file should also be deleted after the error ==" | ||
83 | +if test -f "$TEST_IMAGE_FILE"; then | ||
84 | + exit 1 | ||
85 | +fi | ||
86 | + | ||
87 | +# success, all done | ||
88 | +echo "*** done" | ||
89 | +rm -f $seq.full | ||
90 | +status=0 | ||
91 | diff --git a/tests/qemu-iotests/282.out b/tests/qemu-iotests/282.out | ||
92 | new file mode 100644 | 64 | new file mode 100644 |
93 | index XXXXXXX..XXXXXXX | 65 | index XXXXXXX..XXXXXXX |
94 | --- /dev/null | 66 | --- /dev/null |
95 | +++ b/tests/qemu-iotests/282.out | 67 | +++ b/subprojects/libvduse/libvduse.h |
96 | @@ -XXX,XX +XXX,XX @@ | 68 | @@ -XXX,XX +XXX,XX @@ |
97 | +QA output created by 282 | 69 | +/* |
98 | +== Create non-UTF8 secret == | 70 | + * VDUSE (vDPA Device in Userspace) library |
99 | +== Throws an error because of invalid UTF-8 secret == | 71 | + * |
100 | +qemu-img: vol.img: Data from secret sec0 is not valid UTF-8 | 72 | + * Copyright (C) 2022 Bytedance Inc. and/or its affiliates. All rights reserved. |
101 | +Formatting 'vol.img', fmt=luks size=4194304 key-secret=sec0 | 73 | + * |
102 | +== Image file should not exist after the error == | 74 | + * Author: |
103 | +== Create a stub image file and run qemu-img again == | 75 | + * Xie Yongji <xieyongji@bytedance.com> |
104 | +qemu-img: vol.img: Data from secret sec0 is not valid UTF-8 | 76 | + * |
105 | +Formatting 'vol.img', fmt=luks size=4194304 key-secret=sec0 | 77 | + * This work is licensed under the terms of the GNU GPL, version 2 or |
106 | +== Pre-existing image file should also be deleted after the error == | 78 | + * later. See the COPYING file in the top-level directory. |
107 | + *** done | 79 | + */ |
108 | diff --git a/tests/qemu-iotests/group b/tests/qemu-iotests/group | 80 | + |
81 | +#ifndef LIBVDUSE_H | ||
82 | +#define LIBVDUSE_H | ||
83 | + | ||
84 | +#include <stdint.h> | ||
85 | +#include <sys/uio.h> | ||
86 | + | ||
87 | +#define VIRTQUEUE_MAX_SIZE 1024 | ||
88 | + | ||
89 | +/* VDUSE device structure */ | ||
90 | +typedef struct VduseDev VduseDev; | ||
91 | + | ||
92 | +/* Virtqueue structure */ | ||
93 | +typedef struct VduseVirtq VduseVirtq; | ||
94 | + | ||
95 | +/* Some operation of VDUSE backend */ | ||
96 | +typedef struct VduseOps { | ||
97 | + /* Called when virtqueue can be processed */ | ||
98 | + void (*enable_queue)(VduseDev *dev, VduseVirtq *vq); | ||
99 | + /* Called when virtqueue processing should be stopped */ | ||
100 | + void (*disable_queue)(VduseDev *dev, VduseVirtq *vq); | ||
101 | +} VduseOps; | ||
102 | + | ||
103 | +/* Describing elements of the I/O buffer */ | ||
104 | +typedef struct VduseVirtqElement { | ||
105 | + /* Descriptor table index */ | ||
106 | + unsigned int index; | ||
107 | + /* Number of physically-contiguous device-readable descriptors */ | ||
108 | + unsigned int out_num; | ||
109 | + /* Number of physically-contiguous device-writable descriptors */ | ||
110 | + unsigned int in_num; | ||
111 | + /* Array to store physically-contiguous device-writable descriptors */ | ||
112 | + struct iovec *in_sg; | ||
113 | + /* Array to store physically-contiguous device-readable descriptors */ | ||
114 | + struct iovec *out_sg; | ||
115 | +} VduseVirtqElement; | ||
116 | + | ||
117 | + | ||
118 | +/** | ||
119 | + * vduse_get_virtio_features: | ||
120 | + * | ||
121 | + * Get supported virtio features | ||
122 | + * | ||
123 | + * Returns: supported feature bits | ||
124 | + */ | ||
125 | +uint64_t vduse_get_virtio_features(void); | ||
126 | + | ||
127 | +/** | ||
128 | + * vduse_queue_get_dev: | ||
129 | + * @vq: specified virtqueue | ||
130 | + * | ||
131 | + * Get corresponding VDUSE device from the virtqueue. | ||
132 | + * | ||
133 | + * Returns: a pointer to VDUSE device on success, NULL on failure. | ||
134 | + */ | ||
135 | +VduseDev *vduse_queue_get_dev(VduseVirtq *vq); | ||
136 | + | ||
137 | +/** | ||
138 | + * vduse_queue_get_fd: | ||
139 | + * @vq: specified virtqueue | ||
140 | + * | ||
141 | + * Get the kick fd for the virtqueue. | ||
142 | + * | ||
143 | + * Returns: file descriptor on success, -1 on failure. | ||
144 | + */ | ||
145 | +int vduse_queue_get_fd(VduseVirtq *vq); | ||
146 | + | ||
147 | +/** | ||
148 | + * vduse_queue_pop: | ||
149 | + * @vq: specified virtqueue | ||
150 | + * @sz: the size of struct to return (must be >= VduseVirtqElement) | ||
151 | + * | ||
152 | + * Pop an element from virtqueue available ring. | ||
153 | + * | ||
154 | + * Returns: a pointer to a structure containing VduseVirtqElement on success, | ||
155 | + * NULL on failure. | ||
156 | + */ | ||
157 | +void *vduse_queue_pop(VduseVirtq *vq, size_t sz); | ||
158 | + | ||
159 | +/** | ||
160 | + * vduse_queue_push: | ||
161 | + * @vq: specified virtqueue | ||
162 | + * @elem: pointer to VduseVirtqElement returned by vduse_queue_pop() | ||
163 | + * @len: length in bytes to write | ||
164 | + * | ||
165 | + * Push an element to virtqueue used ring. | ||
166 | + */ | ||
167 | +void vduse_queue_push(VduseVirtq *vq, const VduseVirtqElement *elem, | ||
168 | + unsigned int len); | ||
169 | +/** | ||
170 | + * vduse_queue_notify: | ||
171 | + * @vq: specified virtqueue | ||
172 | + * | ||
173 | + * Request to notify the queue. | ||
174 | + */ | ||
175 | +void vduse_queue_notify(VduseVirtq *vq); | ||
176 | + | ||
177 | +/** | ||
178 | + * vduse_dev_get_priv: | ||
179 | + * @dev: VDUSE device | ||
180 | + * | ||
181 | + * Get the private pointer passed to vduse_dev_create(). | ||
182 | + * | ||
183 | + * Returns: private pointer on success, NULL on failure. | ||
184 | + */ | ||
185 | +void *vduse_dev_get_priv(VduseDev *dev); | ||
186 | + | ||
187 | +/** | ||
188 | + * vduse_dev_get_queue: | ||
189 | + * @dev: VDUSE device | ||
190 | + * @index: virtqueue index | ||
191 | + * | ||
192 | + * Get the specified virtqueue. | ||
193 | + * | ||
194 | + * Returns: a pointer to the virtqueue on success, NULL on failure. | ||
195 | + */ | ||
196 | +VduseVirtq *vduse_dev_get_queue(VduseDev *dev, int index); | ||
197 | + | ||
198 | +/** | ||
199 | + * vduse_dev_get_fd: | ||
200 | + * @dev: VDUSE device | ||
201 | + * | ||
202 | + * Get the control message fd for the VDUSE device. | ||
203 | + * | ||
204 | + * Returns: file descriptor on success, -1 on failure. | ||
205 | + */ | ||
206 | +int vduse_dev_get_fd(VduseDev *dev); | ||
207 | + | ||
208 | +/** | ||
209 | + * vduse_dev_handler: | ||
210 | + * @dev: VDUSE device | ||
211 | + * | ||
212 | + * Used to process the control message. | ||
213 | + * | ||
214 | + * Returns: file descriptor on success, -errno on failure. | ||
215 | + */ | ||
216 | +int vduse_dev_handler(VduseDev *dev); | ||
217 | + | ||
218 | +/** | ||
219 | + * vduse_dev_update_config: | ||
220 | + * @dev: VDUSE device | ||
221 | + * @size: the size to write to configuration space | ||
222 | + * @offset: the offset from the beginning of configuration space | ||
223 | + * @buffer: the buffer used to write from | ||
224 | + * | ||
225 | + * Update device configuration space and inject a config interrupt. | ||
226 | + * | ||
227 | + * Returns: 0 on success, -errno on failure. | ||
228 | + */ | ||
229 | +int vduse_dev_update_config(VduseDev *dev, uint32_t size, | ||
230 | + uint32_t offset, char *buffer); | ||
231 | + | ||
232 | +/** | ||
233 | + * vduse_dev_setup_queue: | ||
234 | + * @dev: VDUSE device | ||
235 | + * @index: virtqueue index | ||
236 | + * @max_size: the max size of virtqueue | ||
237 | + * | ||
238 | + * Setup the specified virtqueue. | ||
239 | + * | ||
240 | + * Returns: 0 on success, -errno on failure. | ||
241 | + */ | ||
242 | +int vduse_dev_setup_queue(VduseDev *dev, int index, int max_size); | ||
243 | + | ||
244 | +/** | ||
245 | + * vduse_dev_create_by_fd: | ||
246 | + * @fd: passed file descriptor | ||
247 | + * @num_queues: the number of virtqueues | ||
248 | + * @ops: the operation of VDUSE backend | ||
249 | + * @priv: private pointer | ||
250 | + * | ||
251 | + * Create VDUSE device from a passed file descriptor. | ||
252 | + * | ||
253 | + * Returns: pointer to VDUSE device on success, NULL on failure. | ||
254 | + */ | ||
255 | +VduseDev *vduse_dev_create_by_fd(int fd, uint16_t num_queues, | ||
256 | + const VduseOps *ops, void *priv); | ||
257 | + | ||
258 | +/** | ||
259 | + * vduse_dev_create_by_name: | ||
260 | + * @name: VDUSE device name | ||
261 | + * @num_queues: the number of virtqueues | ||
262 | + * @ops: the operation of VDUSE backend | ||
263 | + * @priv: private pointer | ||
264 | + * | ||
265 | + * Create VDUSE device on /dev/vduse/$NAME. | ||
266 | + * | ||
267 | + * Returns: pointer to VDUSE device on success, NULL on failure. | ||
268 | + */ | ||
269 | +VduseDev *vduse_dev_create_by_name(const char *name, uint16_t num_queues, | ||
270 | + const VduseOps *ops, void *priv); | ||
271 | + | ||
272 | +/** | ||
273 | + * vduse_dev_create: | ||
274 | + * @name: VDUSE device name | ||
275 | + * @device_id: virtio device id | ||
276 | + * @vendor_id: virtio vendor id | ||
277 | + * @features: virtio features | ||
278 | + * @num_queues: the number of virtqueues | ||
279 | + * @config_size: the size of the configuration space | ||
280 | + * @config: the buffer of the configuration space | ||
281 | + * @ops: the operation of VDUSE backend | ||
282 | + * @priv: private pointer | ||
283 | + * | ||
284 | + * Create VDUSE device. | ||
285 | + * | ||
286 | + * Returns: pointer to VDUSE device on success, NULL on failure. | ||
287 | + */ | ||
288 | +VduseDev *vduse_dev_create(const char *name, uint32_t device_id, | ||
289 | + uint32_t vendor_id, uint64_t features, | ||
290 | + uint16_t num_queues, uint32_t config_size, | ||
291 | + char *config, const VduseOps *ops, void *priv); | ||
292 | + | ||
293 | +/** | ||
294 | + * vduse_dev_destroy: | ||
295 | + * @dev: VDUSE device | ||
296 | + * | ||
297 | + * Destroy the VDUSE device. | ||
298 | + * | ||
299 | + * Returns: 0 on success, -errno on failure. | ||
300 | + */ | ||
301 | +int vduse_dev_destroy(VduseDev *dev); | ||
302 | + | ||
303 | +#endif | ||
304 | diff --git a/subprojects/libvduse/libvduse.c b/subprojects/libvduse/libvduse.c | ||
305 | new file mode 100644 | ||
306 | index XXXXXXX..XXXXXXX | ||
307 | --- /dev/null | ||
308 | +++ b/subprojects/libvduse/libvduse.c | ||
309 | @@ -XXX,XX +XXX,XX @@ | ||
310 | +/* | ||
311 | + * VDUSE (vDPA Device in Userspace) library | ||
312 | + * | ||
313 | + * Copyright (C) 2022 Bytedance Inc. and/or its affiliates. All rights reserved. | ||
314 | + * Portions of codes and concepts borrowed from libvhost-user.c, so: | ||
315 | + * Copyright IBM, Corp. 2007 | ||
316 | + * Copyright (c) 2016 Red Hat, Inc. | ||
317 | + * | ||
318 | + * Author: | ||
319 | + * Xie Yongji <xieyongji@bytedance.com> | ||
320 | + * Anthony Liguori <aliguori@us.ibm.com> | ||
321 | + * Marc-André Lureau <mlureau@redhat.com> | ||
322 | + * Victor Kaplansky <victork@redhat.com> | ||
323 | + * | ||
324 | + * This work is licensed under the terms of the GNU GPL, version 2 or | ||
325 | + * later. See the COPYING file in the top-level directory. | ||
326 | + */ | ||
327 | + | ||
328 | +#include <stdlib.h> | ||
329 | +#include <stdio.h> | ||
330 | +#include <stdbool.h> | ||
331 | +#include <stddef.h> | ||
332 | +#include <errno.h> | ||
333 | +#include <string.h> | ||
334 | +#include <assert.h> | ||
335 | +#include <endian.h> | ||
336 | +#include <unistd.h> | ||
337 | +#include <limits.h> | ||
338 | +#include <fcntl.h> | ||
339 | + | ||
340 | +#include <sys/ioctl.h> | ||
341 | +#include <sys/eventfd.h> | ||
342 | +#include <sys/mman.h> | ||
343 | + | ||
344 | +#include "include/atomic.h" | ||
345 | +#include "linux-headers/linux/virtio_ring.h" | ||
346 | +#include "linux-headers/linux/virtio_config.h" | ||
347 | +#include "linux-headers/linux/vduse.h" | ||
348 | +#include "libvduse.h" | ||
349 | + | ||
350 | +#define VDUSE_VQ_ALIGN 4096 | ||
351 | +#define MAX_IOVA_REGIONS 256 | ||
352 | + | ||
353 | +/* Round number down to multiple */ | ||
354 | +#define ALIGN_DOWN(n, m) ((n) / (m) * (m)) | ||
355 | + | ||
356 | +/* Round number up to multiple */ | ||
357 | +#define ALIGN_UP(n, m) ALIGN_DOWN((n) + (m) - 1, (m)) | ||
358 | + | ||
359 | +#ifndef unlikely | ||
360 | +#define unlikely(x) __builtin_expect(!!(x), 0) | ||
361 | +#endif | ||
362 | + | ||
363 | +typedef struct VduseRing { | ||
364 | + unsigned int num; | ||
365 | + uint64_t desc_addr; | ||
366 | + uint64_t avail_addr; | ||
367 | + uint64_t used_addr; | ||
368 | + struct vring_desc *desc; | ||
369 | + struct vring_avail *avail; | ||
370 | + struct vring_used *used; | ||
371 | +} VduseRing; | ||
372 | + | ||
373 | +struct VduseVirtq { | ||
374 | + VduseRing vring; | ||
375 | + uint16_t last_avail_idx; | ||
376 | + uint16_t shadow_avail_idx; | ||
377 | + uint16_t used_idx; | ||
378 | + uint16_t signalled_used; | ||
379 | + bool signalled_used_valid; | ||
380 | + int index; | ||
381 | + int inuse; | ||
382 | + bool ready; | ||
383 | + int fd; | ||
384 | + VduseDev *dev; | ||
385 | +}; | ||
386 | + | ||
387 | +typedef struct VduseIovaRegion { | ||
388 | + uint64_t iova; | ||
389 | + uint64_t size; | ||
390 | + uint64_t mmap_offset; | ||
391 | + uint64_t mmap_addr; | ||
392 | +} VduseIovaRegion; | ||
393 | + | ||
394 | +struct VduseDev { | ||
395 | + VduseVirtq *vqs; | ||
396 | + VduseIovaRegion regions[MAX_IOVA_REGIONS]; | ||
397 | + int num_regions; | ||
398 | + char *name; | ||
399 | + uint32_t device_id; | ||
400 | + uint32_t vendor_id; | ||
401 | + uint16_t num_queues; | ||
402 | + uint16_t queue_size; | ||
403 | + uint64_t features; | ||
404 | + const VduseOps *ops; | ||
405 | + int fd; | ||
406 | + int ctrl_fd; | ||
407 | + void *priv; | ||
408 | +}; | ||
409 | + | ||
410 | +static inline bool has_feature(uint64_t features, unsigned int fbit) | ||
411 | +{ | ||
412 | + assert(fbit < 64); | ||
413 | + return !!(features & (1ULL << fbit)); | ||
414 | +} | ||
415 | + | ||
416 | +static inline bool vduse_dev_has_feature(VduseDev *dev, unsigned int fbit) | ||
417 | +{ | ||
418 | + return has_feature(dev->features, fbit); | ||
419 | +} | ||
420 | + | ||
421 | +uint64_t vduse_get_virtio_features(void) | ||
422 | +{ | ||
423 | + return (1ULL << VIRTIO_F_IOMMU_PLATFORM) | | ||
424 | + (1ULL << VIRTIO_F_VERSION_1) | | ||
425 | + (1ULL << VIRTIO_F_NOTIFY_ON_EMPTY) | | ||
426 | + (1ULL << VIRTIO_RING_F_EVENT_IDX) | | ||
427 | + (1ULL << VIRTIO_RING_F_INDIRECT_DESC); | ||
428 | +} | ||
429 | + | ||
430 | +VduseDev *vduse_queue_get_dev(VduseVirtq *vq) | ||
431 | +{ | ||
432 | + return vq->dev; | ||
433 | +} | ||
434 | + | ||
435 | +int vduse_queue_get_fd(VduseVirtq *vq) | ||
436 | +{ | ||
437 | + return vq->fd; | ||
438 | +} | ||
439 | + | ||
440 | +void *vduse_dev_get_priv(VduseDev *dev) | ||
441 | +{ | ||
442 | + return dev->priv; | ||
443 | +} | ||
444 | + | ||
445 | +VduseVirtq *vduse_dev_get_queue(VduseDev *dev, int index) | ||
446 | +{ | ||
447 | + return &dev->vqs[index]; | ||
448 | +} | ||
449 | + | ||
450 | +int vduse_dev_get_fd(VduseDev *dev) | ||
451 | +{ | ||
452 | + return dev->fd; | ||
453 | +} | ||
454 | + | ||
455 | +static int vduse_inject_irq(VduseDev *dev, int index) | ||
456 | +{ | ||
457 | + return ioctl(dev->fd, VDUSE_VQ_INJECT_IRQ, &index); | ||
458 | +} | ||
459 | + | ||
460 | +static void vduse_iova_remove_region(VduseDev *dev, uint64_t start, | ||
461 | + uint64_t last) | ||
462 | +{ | ||
463 | + int i; | ||
464 | + | ||
465 | + if (last == start) { | ||
466 | + return; | ||
467 | + } | ||
468 | + | ||
469 | + for (i = 0; i < MAX_IOVA_REGIONS; i++) { | ||
470 | + if (!dev->regions[i].mmap_addr) { | ||
471 | + continue; | ||
472 | + } | ||
473 | + | ||
474 | + if (start <= dev->regions[i].iova && | ||
475 | + last >= (dev->regions[i].iova + dev->regions[i].size - 1)) { | ||
476 | + munmap((void *)dev->regions[i].mmap_addr, | ||
477 | + dev->regions[i].mmap_offset + dev->regions[i].size); | ||
478 | + dev->regions[i].mmap_addr = 0; | ||
479 | + dev->num_regions--; | ||
480 | + } | ||
481 | + } | ||
482 | +} | ||
483 | + | ||
484 | +static int vduse_iova_add_region(VduseDev *dev, int fd, | ||
485 | + uint64_t offset, uint64_t start, | ||
486 | + uint64_t last, int prot) | ||
487 | +{ | ||
488 | + int i; | ||
489 | + uint64_t size = last - start + 1; | ||
490 | + void *mmap_addr = mmap(0, size + offset, prot, MAP_SHARED, fd, 0); | ||
491 | + | ||
492 | + if (mmap_addr == MAP_FAILED) { | ||
493 | + close(fd); | ||
494 | + return -EINVAL; | ||
495 | + } | ||
496 | + | ||
497 | + for (i = 0; i < MAX_IOVA_REGIONS; i++) { | ||
498 | + if (!dev->regions[i].mmap_addr) { | ||
499 | + dev->regions[i].mmap_addr = (uint64_t)(uintptr_t)mmap_addr; | ||
500 | + dev->regions[i].mmap_offset = offset; | ||
501 | + dev->regions[i].iova = start; | ||
502 | + dev->regions[i].size = size; | ||
503 | + dev->num_regions++; | ||
504 | + break; | ||
505 | + } | ||
506 | + } | ||
507 | + assert(i < MAX_IOVA_REGIONS); | ||
508 | + close(fd); | ||
509 | + | ||
510 | + return 0; | ||
511 | +} | ||
512 | + | ||
513 | +static int perm_to_prot(uint8_t perm) | ||
514 | +{ | ||
515 | + int prot = 0; | ||
516 | + | ||
517 | + switch (perm) { | ||
518 | + case VDUSE_ACCESS_WO: | ||
519 | + prot |= PROT_WRITE; | ||
520 | + break; | ||
521 | + case VDUSE_ACCESS_RO: | ||
522 | + prot |= PROT_READ; | ||
523 | + break; | ||
524 | + case VDUSE_ACCESS_RW: | ||
525 | + prot |= PROT_READ | PROT_WRITE; | ||
526 | + break; | ||
527 | + default: | ||
528 | + break; | ||
529 | + } | ||
530 | + | ||
531 | + return prot; | ||
532 | +} | ||
533 | + | ||
534 | +static inline void *iova_to_va(VduseDev *dev, uint64_t *plen, uint64_t iova) | ||
535 | +{ | ||
536 | + int i, ret; | ||
537 | + struct vduse_iotlb_entry entry; | ||
538 | + | ||
539 | + for (i = 0; i < MAX_IOVA_REGIONS; i++) { | ||
540 | + VduseIovaRegion *r = &dev->regions[i]; | ||
541 | + | ||
542 | + if (!r->mmap_addr) { | ||
543 | + continue; | ||
544 | + } | ||
545 | + | ||
546 | + if ((iova >= r->iova) && (iova < (r->iova + r->size))) { | ||
547 | + if ((iova + *plen) > (r->iova + r->size)) { | ||
548 | + *plen = r->iova + r->size - iova; | ||
549 | + } | ||
550 | + return (void *)(uintptr_t)(iova - r->iova + | ||
551 | + r->mmap_addr + r->mmap_offset); | ||
552 | + } | ||
553 | + } | ||
554 | + | ||
555 | + entry.start = iova; | ||
556 | + entry.last = iova + 1; | ||
557 | + ret = ioctl(dev->fd, VDUSE_IOTLB_GET_FD, &entry); | ||
558 | + if (ret < 0) { | ||
559 | + return NULL; | ||
560 | + } | ||
561 | + | ||
562 | + if (!vduse_iova_add_region(dev, ret, entry.offset, entry.start, | ||
563 | + entry.last, perm_to_prot(entry.perm))) { | ||
564 | + return iova_to_va(dev, plen, iova); | ||
565 | + } | ||
566 | + | ||
567 | + return NULL; | ||
568 | +} | ||
569 | + | ||
570 | +static inline uint16_t vring_avail_flags(VduseVirtq *vq) | ||
571 | +{ | ||
572 | + return le16toh(vq->vring.avail->flags); | ||
573 | +} | ||
574 | + | ||
575 | +static inline uint16_t vring_avail_idx(VduseVirtq *vq) | ||
576 | +{ | ||
577 | + vq->shadow_avail_idx = le16toh(vq->vring.avail->idx); | ||
578 | + | ||
579 | + return vq->shadow_avail_idx; | ||
580 | +} | ||
581 | + | ||
582 | +static inline uint16_t vring_avail_ring(VduseVirtq *vq, int i) | ||
583 | +{ | ||
584 | + return le16toh(vq->vring.avail->ring[i]); | ||
585 | +} | ||
586 | + | ||
587 | +static inline uint16_t vring_get_used_event(VduseVirtq *vq) | ||
588 | +{ | ||
589 | + return vring_avail_ring(vq, vq->vring.num); | ||
590 | +} | ||
591 | + | ||
592 | +static bool vduse_queue_get_head(VduseVirtq *vq, unsigned int idx, | ||
593 | + unsigned int *head) | ||
594 | +{ | ||
595 | + /* | ||
596 | + * Grab the next descriptor number they're advertising, and increment | ||
597 | + * the index we've seen. | ||
598 | + */ | ||
599 | + *head = vring_avail_ring(vq, idx % vq->vring.num); | ||
600 | + | ||
601 | + /* If their number is silly, that's a fatal mistake. */ | ||
602 | + if (*head >= vq->vring.num) { | ||
603 | + fprintf(stderr, "Guest says index %u is available\n", *head); | ||
604 | + return false; | ||
605 | + } | ||
606 | + | ||
607 | + return true; | ||
608 | +} | ||
609 | + | ||
610 | +static int | ||
611 | +vduse_queue_read_indirect_desc(VduseDev *dev, struct vring_desc *desc, | ||
612 | + uint64_t addr, size_t len) | ||
613 | +{ | ||
614 | + struct vring_desc *ori_desc; | ||
615 | + uint64_t read_len; | ||
616 | + | ||
617 | + if (len > (VIRTQUEUE_MAX_SIZE * sizeof(struct vring_desc))) { | ||
618 | + return -1; | ||
619 | + } | ||
620 | + | ||
621 | + if (len == 0) { | ||
622 | + return -1; | ||
623 | + } | ||
624 | + | ||
625 | + while (len) { | ||
626 | + read_len = len; | ||
627 | + ori_desc = iova_to_va(dev, &read_len, addr); | ||
628 | + if (!ori_desc) { | ||
629 | + return -1; | ||
630 | + } | ||
631 | + | ||
632 | + memcpy(desc, ori_desc, read_len); | ||
633 | + len -= read_len; | ||
634 | + addr += read_len; | ||
635 | + desc += read_len; | ||
636 | + } | ||
637 | + | ||
638 | + return 0; | ||
639 | +} | ||
640 | + | ||
641 | +enum { | ||
642 | + VIRTQUEUE_READ_DESC_ERROR = -1, | ||
643 | + VIRTQUEUE_READ_DESC_DONE = 0, /* end of chain */ | ||
644 | + VIRTQUEUE_READ_DESC_MORE = 1, /* more buffers in chain */ | ||
645 | +}; | ||
646 | + | ||
647 | +static int vduse_queue_read_next_desc(struct vring_desc *desc, int i, | ||
648 | + unsigned int max, unsigned int *next) | ||
649 | +{ | ||
650 | + /* If this descriptor says it doesn't chain, we're done. */ | ||
651 | + if (!(le16toh(desc[i].flags) & VRING_DESC_F_NEXT)) { | ||
652 | + return VIRTQUEUE_READ_DESC_DONE; | ||
653 | + } | ||
654 | + | ||
655 | + /* Check they're not leading us off end of descriptors. */ | ||
656 | + *next = desc[i].next; | ||
657 | + /* Make sure compiler knows to grab that: we don't want it changing! */ | ||
658 | + smp_wmb(); | ||
659 | + | ||
660 | + if (*next >= max) { | ||
661 | + fprintf(stderr, "Desc next is %u\n", *next); | ||
662 | + return VIRTQUEUE_READ_DESC_ERROR; | ||
663 | + } | ||
664 | + | ||
665 | + return VIRTQUEUE_READ_DESC_MORE; | ||
666 | +} | ||
667 | + | ||
668 | +/* | ||
669 | + * Fetch avail_idx from VQ memory only when we really need to know if | ||
670 | + * guest has added some buffers. | ||
671 | + */ | ||
672 | +static bool vduse_queue_empty(VduseVirtq *vq) | ||
673 | +{ | ||
674 | + if (unlikely(!vq->vring.avail)) { | ||
675 | + return true; | ||
676 | + } | ||
677 | + | ||
678 | + if (vq->shadow_avail_idx != vq->last_avail_idx) { | ||
679 | + return false; | ||
680 | + } | ||
681 | + | ||
682 | + return vring_avail_idx(vq) == vq->last_avail_idx; | ||
683 | +} | ||
684 | + | ||
685 | +static bool vduse_queue_should_notify(VduseVirtq *vq) | ||
686 | +{ | ||
687 | + VduseDev *dev = vq->dev; | ||
688 | + uint16_t old, new; | ||
689 | + bool v; | ||
690 | + | ||
691 | + /* We need to expose used array entries before checking used event. */ | ||
692 | + smp_mb(); | ||
693 | + | ||
694 | + /* Always notify when queue is empty (when feature acknowledge) */ | ||
695 | + if (vduse_dev_has_feature(dev, VIRTIO_F_NOTIFY_ON_EMPTY) && | ||
696 | + !vq->inuse && vduse_queue_empty(vq)) { | ||
697 | + return true; | ||
698 | + } | ||
699 | + | ||
700 | + if (!vduse_dev_has_feature(dev, VIRTIO_RING_F_EVENT_IDX)) { | ||
701 | + return !(vring_avail_flags(vq) & VRING_AVAIL_F_NO_INTERRUPT); | ||
702 | + } | ||
703 | + | ||
704 | + v = vq->signalled_used_valid; | ||
705 | + vq->signalled_used_valid = true; | ||
706 | + old = vq->signalled_used; | ||
707 | + new = vq->signalled_used = vq->used_idx; | ||
708 | + return !v || vring_need_event(vring_get_used_event(vq), new, old); | ||
709 | +} | ||
710 | + | ||
711 | +void vduse_queue_notify(VduseVirtq *vq) | ||
712 | +{ | ||
713 | + VduseDev *dev = vq->dev; | ||
714 | + | ||
715 | + if (unlikely(!vq->vring.avail)) { | ||
716 | + return; | ||
717 | + } | ||
718 | + | ||
719 | + if (!vduse_queue_should_notify(vq)) { | ||
720 | + return; | ||
721 | + } | ||
722 | + | ||
723 | + if (vduse_inject_irq(dev, vq->index) < 0) { | ||
724 | + fprintf(stderr, "Error inject irq for vq %d: %s\n", | ||
725 | + vq->index, strerror(errno)); | ||
726 | + } | ||
727 | +} | ||
728 | + | ||
729 | +static inline void vring_used_flags_set_bit(VduseVirtq *vq, int mask) | ||
730 | +{ | ||
731 | + uint16_t *flags; | ||
732 | + | ||
733 | + flags = (uint16_t *)((char*)vq->vring.used + | ||
734 | + offsetof(struct vring_used, flags)); | ||
735 | + *flags = htole16(le16toh(*flags) | mask); | ||
736 | +} | ||
737 | + | ||
738 | +static inline void vring_used_flags_unset_bit(VduseVirtq *vq, int mask) | ||
739 | +{ | ||
740 | + uint16_t *flags; | ||
741 | + | ||
742 | + flags = (uint16_t *)((char*)vq->vring.used + | ||
743 | + offsetof(struct vring_used, flags)); | ||
744 | + *flags = htole16(le16toh(*flags) & ~mask); | ||
745 | +} | ||
746 | + | ||
747 | +static inline void vring_set_avail_event(VduseVirtq *vq, uint16_t val) | ||
748 | +{ | ||
749 | + *((uint16_t *)&vq->vring.used->ring[vq->vring.num]) = htole16(val); | ||
750 | +} | ||
751 | + | ||
752 | +static bool vduse_queue_map_single_desc(VduseVirtq *vq, unsigned int *p_num_sg, | ||
753 | + struct iovec *iov, unsigned int max_num_sg, | ||
754 | + bool is_write, uint64_t pa, size_t sz) | ||
755 | +{ | ||
756 | + unsigned num_sg = *p_num_sg; | ||
757 | + VduseDev *dev = vq->dev; | ||
758 | + | ||
759 | + assert(num_sg <= max_num_sg); | ||
760 | + | ||
761 | + if (!sz) { | ||
762 | + fprintf(stderr, "virtio: zero sized buffers are not allowed\n"); | ||
763 | + return false; | ||
764 | + } | ||
765 | + | ||
766 | + while (sz) { | ||
767 | + uint64_t len = sz; | ||
768 | + | ||
769 | + if (num_sg == max_num_sg) { | ||
770 | + fprintf(stderr, | ||
771 | + "virtio: too many descriptors in indirect table\n"); | ||
772 | + return false; | ||
773 | + } | ||
774 | + | ||
775 | + iov[num_sg].iov_base = iova_to_va(dev, &len, pa); | ||
776 | + if (iov[num_sg].iov_base == NULL) { | ||
777 | + fprintf(stderr, "virtio: invalid address for buffers\n"); | ||
778 | + return false; | ||
779 | + } | ||
780 | + iov[num_sg++].iov_len = len; | ||
781 | + sz -= len; | ||
782 | + pa += len; | ||
783 | + } | ||
784 | + | ||
785 | + *p_num_sg = num_sg; | ||
786 | + return true; | ||
787 | +} | ||
788 | + | ||
789 | +static void *vduse_queue_alloc_element(size_t sz, unsigned out_num, | ||
790 | + unsigned in_num) | ||
791 | +{ | ||
792 | + VduseVirtqElement *elem; | ||
793 | + size_t in_sg_ofs = ALIGN_UP(sz, __alignof__(elem->in_sg[0])); | ||
794 | + size_t out_sg_ofs = in_sg_ofs + in_num * sizeof(elem->in_sg[0]); | ||
795 | + size_t out_sg_end = out_sg_ofs + out_num * sizeof(elem->out_sg[0]); | ||
796 | + | ||
797 | + assert(sz >= sizeof(VduseVirtqElement)); | ||
798 | + elem = malloc(out_sg_end); | ||
799 | + if (!elem) { | ||
800 | + return NULL; | ||
801 | + } | ||
802 | + elem->out_num = out_num; | ||
803 | + elem->in_num = in_num; | ||
804 | + elem->in_sg = (void *)elem + in_sg_ofs; | ||
805 | + elem->out_sg = (void *)elem + out_sg_ofs; | ||
806 | + return elem; | ||
807 | +} | ||
808 | + | ||
809 | +static void *vduse_queue_map_desc(VduseVirtq *vq, unsigned int idx, size_t sz) | ||
810 | +{ | ||
811 | + struct vring_desc *desc = vq->vring.desc; | ||
812 | + VduseDev *dev = vq->dev; | ||
813 | + uint64_t desc_addr, read_len; | ||
814 | + unsigned int desc_len; | ||
815 | + unsigned int max = vq->vring.num; | ||
816 | + unsigned int i = idx; | ||
817 | + VduseVirtqElement *elem; | ||
818 | + struct iovec iov[VIRTQUEUE_MAX_SIZE]; | ||
819 | + struct vring_desc desc_buf[VIRTQUEUE_MAX_SIZE]; | ||
820 | + unsigned int out_num = 0, in_num = 0; | ||
821 | + int rc; | ||
822 | + | ||
823 | + if (le16toh(desc[i].flags) & VRING_DESC_F_INDIRECT) { | ||
824 | + if (le32toh(desc[i].len) % sizeof(struct vring_desc)) { | ||
825 | + fprintf(stderr, "Invalid size for indirect buffer table\n"); | ||
826 | + return NULL; | ||
827 | + } | ||
828 | + | ||
829 | + /* loop over the indirect descriptor table */ | ||
830 | + desc_addr = le64toh(desc[i].addr); | ||
831 | + desc_len = le32toh(desc[i].len); | ||
832 | + max = desc_len / sizeof(struct vring_desc); | ||
833 | + read_len = desc_len; | ||
834 | + desc = iova_to_va(dev, &read_len, desc_addr); | ||
835 | + if (unlikely(desc && read_len != desc_len)) { | ||
836 | + /* Failed to use zero copy */ | ||
837 | + desc = NULL; | ||
838 | + if (!vduse_queue_read_indirect_desc(dev, desc_buf, | ||
839 | + desc_addr, | ||
840 | + desc_len)) { | ||
841 | + desc = desc_buf; | ||
842 | + } | ||
843 | + } | ||
844 | + if (!desc) { | ||
845 | + fprintf(stderr, "Invalid indirect buffer table\n"); | ||
846 | + return NULL; | ||
847 | + } | ||
848 | + i = 0; | ||
849 | + } | ||
850 | + | ||
851 | + /* Collect all the descriptors */ | ||
852 | + do { | ||
853 | + if (le16toh(desc[i].flags) & VRING_DESC_F_WRITE) { | ||
854 | + if (!vduse_queue_map_single_desc(vq, &in_num, iov + out_num, | ||
855 | + VIRTQUEUE_MAX_SIZE - out_num, | ||
856 | + true, le64toh(desc[i].addr), | ||
857 | + le32toh(desc[i].len))) { | ||
858 | + return NULL; | ||
859 | + } | ||
860 | + } else { | ||
861 | + if (in_num) { | ||
862 | + fprintf(stderr, "Incorrect order for descriptors\n"); | ||
863 | + return NULL; | ||
864 | + } | ||
865 | + if (!vduse_queue_map_single_desc(vq, &out_num, iov, | ||
866 | + VIRTQUEUE_MAX_SIZE, false, | ||
867 | + le64toh(desc[i].addr), | ||
868 | + le32toh(desc[i].len))) { | ||
869 | + return NULL; | ||
870 | + } | ||
871 | + } | ||
872 | + | ||
873 | + /* If we've got too many, that implies a descriptor loop. */ | ||
874 | + if ((in_num + out_num) > max) { | ||
875 | + fprintf(stderr, "Looped descriptor\n"); | ||
876 | + return NULL; | ||
877 | + } | ||
878 | + rc = vduse_queue_read_next_desc(desc, i, max, &i); | ||
879 | + } while (rc == VIRTQUEUE_READ_DESC_MORE); | ||
880 | + | ||
881 | + if (rc == VIRTQUEUE_READ_DESC_ERROR) { | ||
882 | + fprintf(stderr, "read descriptor error\n"); | ||
883 | + return NULL; | ||
884 | + } | ||
885 | + | ||
886 | + /* Now copy what we have collected and mapped */ | ||
887 | + elem = vduse_queue_alloc_element(sz, out_num, in_num); | ||
888 | + if (!elem) { | ||
889 | + fprintf(stderr, "read descriptor error\n"); | ||
890 | + return NULL; | ||
891 | + } | ||
892 | + elem->index = idx; | ||
893 | + for (i = 0; i < out_num; i++) { | ||
894 | + elem->out_sg[i] = iov[i]; | ||
895 | + } | ||
896 | + for (i = 0; i < in_num; i++) { | ||
897 | + elem->in_sg[i] = iov[out_num + i]; | ||
898 | + } | ||
899 | + | ||
900 | + return elem; | ||
901 | +} | ||
902 | + | ||
903 | +void *vduse_queue_pop(VduseVirtq *vq, size_t sz) | ||
904 | +{ | ||
905 | + unsigned int head; | ||
906 | + VduseVirtqElement *elem; | ||
907 | + VduseDev *dev = vq->dev; | ||
908 | + | ||
909 | + if (unlikely(!vq->vring.avail)) { | ||
910 | + return NULL; | ||
911 | + } | ||
912 | + | ||
913 | + if (vduse_queue_empty(vq)) { | ||
914 | + return NULL; | ||
915 | + } | ||
916 | + /* Needed after virtio_queue_empty() */ | ||
917 | + smp_rmb(); | ||
918 | + | ||
919 | + if (vq->inuse >= vq->vring.num) { | ||
920 | + fprintf(stderr, "Virtqueue size exceeded: %d\n", vq->inuse); | ||
921 | + return NULL; | ||
922 | + } | ||
923 | + | ||
924 | + if (!vduse_queue_get_head(vq, vq->last_avail_idx++, &head)) { | ||
925 | + return NULL; | ||
926 | + } | ||
927 | + | ||
928 | + if (vduse_dev_has_feature(dev, VIRTIO_RING_F_EVENT_IDX)) { | ||
929 | + vring_set_avail_event(vq, vq->last_avail_idx); | ||
930 | + } | ||
931 | + | ||
932 | + elem = vduse_queue_map_desc(vq, head, sz); | ||
933 | + | ||
934 | + if (!elem) { | ||
935 | + return NULL; | ||
936 | + } | ||
937 | + | ||
938 | + vq->inuse++; | ||
939 | + | ||
940 | + return elem; | ||
941 | +} | ||
942 | + | ||
943 | +static inline void vring_used_write(VduseVirtq *vq, | ||
944 | + struct vring_used_elem *uelem, int i) | ||
945 | +{ | ||
946 | + struct vring_used *used = vq->vring.used; | ||
947 | + | ||
948 | + used->ring[i] = *uelem; | ||
949 | +} | ||
950 | + | ||
951 | +static void vduse_queue_fill(VduseVirtq *vq, const VduseVirtqElement *elem, | ||
952 | + unsigned int len, unsigned int idx) | ||
953 | +{ | ||
954 | + struct vring_used_elem uelem; | ||
955 | + | ||
956 | + if (unlikely(!vq->vring.used)) { | ||
957 | + return; | ||
958 | + } | ||
959 | + | ||
960 | + idx = (idx + vq->used_idx) % vq->vring.num; | ||
961 | + | ||
962 | + uelem.id = htole32(elem->index); | ||
963 | + uelem.len = htole32(len); | ||
964 | + vring_used_write(vq, &uelem, idx); | ||
965 | +} | ||
966 | + | ||
967 | +static inline void vring_used_idx_set(VduseVirtq *vq, uint16_t val) | ||
968 | +{ | ||
969 | + vq->vring.used->idx = htole16(val); | ||
970 | + vq->used_idx = val; | ||
971 | +} | ||
972 | + | ||
973 | +static void vduse_queue_flush(VduseVirtq *vq, unsigned int count) | ||
974 | +{ | ||
975 | + uint16_t old, new; | ||
976 | + | ||
977 | + if (unlikely(!vq->vring.used)) { | ||
978 | + return; | ||
979 | + } | ||
980 | + | ||
981 | + /* Make sure buffer is written before we update index. */ | ||
982 | + smp_wmb(); | ||
983 | + | ||
984 | + old = vq->used_idx; | ||
985 | + new = old + count; | ||
986 | + vring_used_idx_set(vq, new); | ||
987 | + vq->inuse -= count; | ||
988 | + if (unlikely((int16_t)(new - vq->signalled_used) < (uint16_t)(new - old))) { | ||
989 | + vq->signalled_used_valid = false; | ||
990 | + } | ||
991 | +} | ||
992 | + | ||
993 | +void vduse_queue_push(VduseVirtq *vq, const VduseVirtqElement *elem, | ||
994 | + unsigned int len) | ||
995 | +{ | ||
996 | + vduse_queue_fill(vq, elem, len, 0); | ||
997 | + vduse_queue_flush(vq, 1); | ||
998 | +} | ||
999 | + | ||
1000 | +static int vduse_queue_update_vring(VduseVirtq *vq, uint64_t desc_addr, | ||
1001 | + uint64_t avail_addr, uint64_t used_addr) | ||
1002 | +{ | ||
1003 | + struct VduseDev *dev = vq->dev; | ||
1004 | + uint64_t len; | ||
1005 | + | ||
1006 | + len = sizeof(struct vring_desc); | ||
1007 | + vq->vring.desc = iova_to_va(dev, &len, desc_addr); | ||
1008 | + if (len != sizeof(struct vring_desc)) { | ||
1009 | + return -EINVAL; | ||
1010 | + } | ||
1011 | + | ||
1012 | + len = sizeof(struct vring_avail); | ||
1013 | + vq->vring.avail = iova_to_va(dev, &len, avail_addr); | ||
1014 | + if (len != sizeof(struct vring_avail)) { | ||
1015 | + return -EINVAL; | ||
1016 | + } | ||
1017 | + | ||
1018 | + len = sizeof(struct vring_used); | ||
1019 | + vq->vring.used = iova_to_va(dev, &len, used_addr); | ||
1020 | + if (len != sizeof(struct vring_used)) { | ||
1021 | + return -EINVAL; | ||
1022 | + } | ||
1023 | + | ||
1024 | + if (!vq->vring.desc || !vq->vring.avail || !vq->vring.used) { | ||
1025 | + fprintf(stderr, "Failed to get vq[%d] iova mapping\n", vq->index); | ||
1026 | + return -EINVAL; | ||
1027 | + } | ||
1028 | + | ||
1029 | + return 0; | ||
1030 | +} | ||
1031 | + | ||
1032 | +static void vduse_queue_enable(VduseVirtq *vq) | ||
1033 | +{ | ||
1034 | + struct VduseDev *dev = vq->dev; | ||
1035 | + struct vduse_vq_info vq_info; | ||
1036 | + struct vduse_vq_eventfd vq_eventfd; | ||
1037 | + int fd; | ||
1038 | + | ||
1039 | + vq_info.index = vq->index; | ||
1040 | + if (ioctl(dev->fd, VDUSE_VQ_GET_INFO, &vq_info)) { | ||
1041 | + fprintf(stderr, "Failed to get vq[%d] info: %s\n", | ||
1042 | + vq->index, strerror(errno)); | ||
1043 | + return; | ||
1044 | + } | ||
1045 | + | ||
1046 | + if (!vq_info.ready) { | ||
1047 | + return; | ||
1048 | + } | ||
1049 | + | ||
1050 | + vq->vring.num = vq_info.num; | ||
1051 | + vq->vring.desc_addr = vq_info.desc_addr; | ||
1052 | + vq->vring.avail_addr = vq_info.driver_addr; | ||
1053 | + vq->vring.used_addr = vq_info.device_addr; | ||
1054 | + | ||
1055 | + if (vduse_queue_update_vring(vq, vq_info.desc_addr, | ||
1056 | + vq_info.driver_addr, vq_info.device_addr)) { | ||
1057 | + fprintf(stderr, "Failed to update vring for vq[%d]\n", vq->index); | ||
1058 | + return; | ||
1059 | + } | ||
1060 | + | ||
1061 | + fd = eventfd(0, EFD_NONBLOCK | EFD_CLOEXEC); | ||
1062 | + if (fd < 0) { | ||
1063 | + fprintf(stderr, "Failed to init eventfd for vq[%d]\n", vq->index); | ||
1064 | + return; | ||
1065 | + } | ||
1066 | + | ||
1067 | + vq_eventfd.index = vq->index; | ||
1068 | + vq_eventfd.fd = fd; | ||
1069 | + if (ioctl(dev->fd, VDUSE_VQ_SETUP_KICKFD, &vq_eventfd)) { | ||
1070 | + fprintf(stderr, "Failed to setup kick fd for vq[%d]\n", vq->index); | ||
1071 | + close(fd); | ||
1072 | + return; | ||
1073 | + } | ||
1074 | + | ||
1075 | + vq->fd = fd; | ||
1076 | + vq->shadow_avail_idx = vq->last_avail_idx = vq_info.split.avail_index; | ||
1077 | + vq->inuse = 0; | ||
1078 | + vq->used_idx = 0; | ||
1079 | + vq->signalled_used_valid = false; | ||
1080 | + vq->ready = true; | ||
1081 | + | ||
1082 | + dev->ops->enable_queue(dev, vq); | ||
1083 | +} | ||
1084 | + | ||
1085 | +static void vduse_queue_disable(VduseVirtq *vq) | ||
1086 | +{ | ||
1087 | + struct VduseDev *dev = vq->dev; | ||
1088 | + struct vduse_vq_eventfd eventfd; | ||
1089 | + | ||
1090 | + if (!vq->ready) { | ||
1091 | + return; | ||
1092 | + } | ||
1093 | + | ||
1094 | + dev->ops->disable_queue(dev, vq); | ||
1095 | + | ||
1096 | + eventfd.index = vq->index; | ||
1097 | + eventfd.fd = VDUSE_EVENTFD_DEASSIGN; | ||
1098 | + ioctl(dev->fd, VDUSE_VQ_SETUP_KICKFD, &eventfd); | ||
1099 | + close(vq->fd); | ||
1100 | + | ||
1101 | + assert(vq->inuse == 0); | ||
1102 | + | ||
1103 | + vq->vring.num = 0; | ||
1104 | + vq->vring.desc_addr = 0; | ||
1105 | + vq->vring.avail_addr = 0; | ||
1106 | + vq->vring.used_addr = 0; | ||
1107 | + vq->vring.desc = 0; | ||
1108 | + vq->vring.avail = 0; | ||
1109 | + vq->vring.used = 0; | ||
1110 | + vq->ready = false; | ||
1111 | + vq->fd = -1; | ||
1112 | +} | ||
1113 | + | ||
1114 | +static void vduse_dev_start_dataplane(VduseDev *dev) | ||
1115 | +{ | ||
1116 | + int i; | ||
1117 | + | ||
1118 | + if (ioctl(dev->fd, VDUSE_DEV_GET_FEATURES, &dev->features)) { | ||
1119 | + fprintf(stderr, "Failed to get features: %s\n", strerror(errno)); | ||
1120 | + return; | ||
1121 | + } | ||
1122 | + assert(vduse_dev_has_feature(dev, VIRTIO_F_VERSION_1)); | ||
1123 | + | ||
1124 | + for (i = 0; i < dev->num_queues; i++) { | ||
1125 | + vduse_queue_enable(&dev->vqs[i]); | ||
1126 | + } | ||
1127 | +} | ||
1128 | + | ||
1129 | +static void vduse_dev_stop_dataplane(VduseDev *dev) | ||
1130 | +{ | ||
1131 | + int i; | ||
1132 | + | ||
1133 | + for (i = 0; i < dev->num_queues; i++) { | ||
1134 | + vduse_queue_disable(&dev->vqs[i]); | ||
1135 | + } | ||
1136 | + dev->features = 0; | ||
1137 | + vduse_iova_remove_region(dev, 0, ULONG_MAX); | ||
1138 | +} | ||
1139 | + | ||
1140 | +int vduse_dev_handler(VduseDev *dev) | ||
1141 | +{ | ||
1142 | + struct vduse_dev_request req; | ||
1143 | + struct vduse_dev_response resp = { 0 }; | ||
1144 | + VduseVirtq *vq; | ||
1145 | + int i, ret; | ||
1146 | + | ||
1147 | + ret = read(dev->fd, &req, sizeof(req)); | ||
1148 | + if (ret != sizeof(req)) { | ||
1149 | + fprintf(stderr, "Read request error [%d]: %s\n", | ||
1150 | + ret, strerror(errno)); | ||
1151 | + return -errno; | ||
1152 | + } | ||
1153 | + resp.request_id = req.request_id; | ||
1154 | + | ||
1155 | + switch (req.type) { | ||
1156 | + case VDUSE_GET_VQ_STATE: | ||
1157 | + vq = &dev->vqs[req.vq_state.index]; | ||
1158 | + resp.vq_state.split.avail_index = vq->last_avail_idx; | ||
1159 | + resp.result = VDUSE_REQ_RESULT_OK; | ||
1160 | + break; | ||
1161 | + case VDUSE_SET_STATUS: | ||
1162 | + if (req.s.status & VIRTIO_CONFIG_S_DRIVER_OK) { | ||
1163 | + vduse_dev_start_dataplane(dev); | ||
1164 | + } else if (req.s.status == 0) { | ||
1165 | + vduse_dev_stop_dataplane(dev); | ||
1166 | + } | ||
1167 | + resp.result = VDUSE_REQ_RESULT_OK; | ||
1168 | + break; | ||
1169 | + case VDUSE_UPDATE_IOTLB: | ||
1170 | + /* The iova will be updated by iova_to_va() later, so just remove it */ | ||
1171 | + vduse_iova_remove_region(dev, req.iova.start, req.iova.last); | ||
1172 | + for (i = 0; i < dev->num_queues; i++) { | ||
1173 | + VduseVirtq *vq = &dev->vqs[i]; | ||
1174 | + if (vq->ready) { | ||
1175 | + if (vduse_queue_update_vring(vq, vq->vring.desc_addr, | ||
1176 | + vq->vring.avail_addr, | ||
1177 | + vq->vring.used_addr)) { | ||
1178 | + fprintf(stderr, "Failed to update vring for vq[%d]\n", | ||
1179 | + vq->index); | ||
1180 | + } | ||
1181 | + } | ||
1182 | + } | ||
1183 | + resp.result = VDUSE_REQ_RESULT_OK; | ||
1184 | + break; | ||
1185 | + default: | ||
1186 | + resp.result = VDUSE_REQ_RESULT_FAILED; | ||
1187 | + break; | ||
1188 | + } | ||
1189 | + | ||
1190 | + ret = write(dev->fd, &resp, sizeof(resp)); | ||
1191 | + if (ret != sizeof(resp)) { | ||
1192 | + fprintf(stderr, "Write request %d error [%d]: %s\n", | ||
1193 | + req.type, ret, strerror(errno)); | ||
1194 | + return -errno; | ||
1195 | + } | ||
1196 | + return 0; | ||
1197 | +} | ||
1198 | + | ||
1199 | +int vduse_dev_update_config(VduseDev *dev, uint32_t size, | ||
1200 | + uint32_t offset, char *buffer) | ||
1201 | +{ | ||
1202 | + int ret; | ||
1203 | + struct vduse_config_data *data; | ||
1204 | + | ||
1205 | + data = malloc(offsetof(struct vduse_config_data, buffer) + size); | ||
1206 | + if (!data) { | ||
1207 | + return -ENOMEM; | ||
1208 | + } | ||
1209 | + | ||
1210 | + data->offset = offset; | ||
1211 | + data->length = size; | ||
1212 | + memcpy(data->buffer, buffer, size); | ||
1213 | + | ||
1214 | + ret = ioctl(dev->fd, VDUSE_DEV_SET_CONFIG, data); | ||
1215 | + free(data); | ||
1216 | + | ||
1217 | + if (ret) { | ||
1218 | + return -errno; | ||
1219 | + } | ||
1220 | + | ||
1221 | + if (ioctl(dev->fd, VDUSE_DEV_INJECT_CONFIG_IRQ)) { | ||
1222 | + return -errno; | ||
1223 | + } | ||
1224 | + | ||
1225 | + return 0; | ||
1226 | +} | ||
1227 | + | ||
1228 | +int vduse_dev_setup_queue(VduseDev *dev, int index, int max_size) | ||
1229 | +{ | ||
1230 | + VduseVirtq *vq = &dev->vqs[index]; | ||
1231 | + struct vduse_vq_config vq_config = { 0 }; | ||
1232 | + | ||
1233 | + if (max_size > VIRTQUEUE_MAX_SIZE) { | ||
1234 | + return -EINVAL; | ||
1235 | + } | ||
1236 | + | ||
1237 | + vq_config.index = vq->index; | ||
1238 | + vq_config.max_size = max_size; | ||
1239 | + | ||
1240 | + if (ioctl(dev->fd, VDUSE_VQ_SETUP, &vq_config)) { | ||
1241 | + return -errno; | ||
1242 | + } | ||
1243 | + | ||
1244 | + return 0; | ||
1245 | +} | ||
1246 | + | ||
1247 | +static int vduse_dev_init_vqs(VduseDev *dev, uint16_t num_queues) | ||
1248 | +{ | ||
1249 | + VduseVirtq *vqs; | ||
1250 | + int i; | ||
1251 | + | ||
1252 | + vqs = calloc(sizeof(VduseVirtq), num_queues); | ||
1253 | + if (!vqs) { | ||
1254 | + return -ENOMEM; | ||
1255 | + } | ||
1256 | + | ||
1257 | + for (i = 0; i < num_queues; i++) { | ||
1258 | + vqs[i].index = i; | ||
1259 | + vqs[i].dev = dev; | ||
1260 | + vqs[i].fd = -1; | ||
1261 | + } | ||
1262 | + dev->vqs = vqs; | ||
1263 | + | ||
1264 | + return 0; | ||
1265 | +} | ||
1266 | + | ||
1267 | +static int vduse_dev_init(VduseDev *dev, const char *name, | ||
1268 | + uint16_t num_queues, const VduseOps *ops, | ||
1269 | + void *priv) | ||
1270 | +{ | ||
1271 | + char *dev_path, *dev_name; | ||
1272 | + int ret, fd; | ||
1273 | + | ||
1274 | + dev_path = malloc(strlen(name) + strlen("/dev/vduse/") + 1); | ||
1275 | + if (!dev_path) { | ||
1276 | + return -ENOMEM; | ||
1277 | + } | ||
1278 | + sprintf(dev_path, "/dev/vduse/%s", name); | ||
1279 | + | ||
1280 | + fd = open(dev_path, O_RDWR); | ||
1281 | + free(dev_path); | ||
1282 | + if (fd < 0) { | ||
1283 | + fprintf(stderr, "Failed to open vduse dev %s: %s\n", | ||
1284 | + name, strerror(errno)); | ||
1285 | + return -errno; | ||
1286 | + } | ||
1287 | + | ||
1288 | + dev_name = strdup(name); | ||
1289 | + if (!dev_name) { | ||
1290 | + close(fd); | ||
1291 | + return -ENOMEM; | ||
1292 | + } | ||
1293 | + | ||
1294 | + ret = vduse_dev_init_vqs(dev, num_queues); | ||
1295 | + if (ret) { | ||
1296 | + free(dev_name); | ||
1297 | + close(fd); | ||
1298 | + return ret; | ||
1299 | + } | ||
1300 | + | ||
1301 | + dev->name = dev_name; | ||
1302 | + dev->num_queues = num_queues; | ||
1303 | + dev->fd = fd; | ||
1304 | + dev->ops = ops; | ||
1305 | + dev->priv = priv; | ||
1306 | + | ||
1307 | + return 0; | ||
1308 | +} | ||
1309 | + | ||
1310 | +static inline bool vduse_name_is_valid(const char *name) | ||
1311 | +{ | ||
1312 | + return strlen(name) >= VDUSE_NAME_MAX || strstr(name, ".."); | ||
1313 | +} | ||
1314 | + | ||
1315 | +VduseDev *vduse_dev_create_by_fd(int fd, uint16_t num_queues, | ||
1316 | + const VduseOps *ops, void *priv) | ||
1317 | +{ | ||
1318 | + VduseDev *dev; | ||
1319 | + int ret; | ||
1320 | + | ||
1321 | + if (!ops || !ops->enable_queue || !ops->disable_queue) { | ||
1322 | + fprintf(stderr, "Invalid parameter for vduse\n"); | ||
1323 | + return NULL; | ||
1324 | + } | ||
1325 | + | ||
1326 | + dev = calloc(sizeof(VduseDev), 1); | ||
1327 | + if (!dev) { | ||
1328 | + fprintf(stderr, "Failed to allocate vduse device\n"); | ||
1329 | + return NULL; | ||
1330 | + } | ||
1331 | + | ||
1332 | + ret = vduse_dev_init_vqs(dev, num_queues); | ||
1333 | + if (ret) { | ||
1334 | + fprintf(stderr, "Failed to init vqs\n"); | ||
1335 | + free(dev); | ||
1336 | + return NULL; | ||
1337 | + } | ||
1338 | + | ||
1339 | + dev->num_queues = num_queues; | ||
1340 | + dev->fd = fd; | ||
1341 | + dev->ops = ops; | ||
1342 | + dev->priv = priv; | ||
1343 | + | ||
1344 | + return dev; | ||
1345 | +} | ||
1346 | + | ||
1347 | +VduseDev *vduse_dev_create_by_name(const char *name, uint16_t num_queues, | ||
1348 | + const VduseOps *ops, void *priv) | ||
1349 | +{ | ||
1350 | + VduseDev *dev; | ||
1351 | + int ret; | ||
1352 | + | ||
1353 | + if (!name || vduse_name_is_valid(name) || !ops || | ||
1354 | + !ops->enable_queue || !ops->disable_queue) { | ||
1355 | + fprintf(stderr, "Invalid parameter for vduse\n"); | ||
1356 | + return NULL; | ||
1357 | + } | ||
1358 | + | ||
1359 | + dev = calloc(sizeof(VduseDev), 1); | ||
1360 | + if (!dev) { | ||
1361 | + fprintf(stderr, "Failed to allocate vduse device\n"); | ||
1362 | + return NULL; | ||
1363 | + } | ||
1364 | + | ||
1365 | + ret = vduse_dev_init(dev, name, num_queues, ops, priv); | ||
1366 | + if (ret < 0) { | ||
1367 | + fprintf(stderr, "Failed to init vduse device %s: %s\n", | ||
1368 | + name, strerror(ret)); | ||
1369 | + free(dev); | ||
1370 | + return NULL; | ||
1371 | + } | ||
1372 | + | ||
1373 | + return dev; | ||
1374 | +} | ||
1375 | + | ||
1376 | +VduseDev *vduse_dev_create(const char *name, uint32_t device_id, | ||
1377 | + uint32_t vendor_id, uint64_t features, | ||
1378 | + uint16_t num_queues, uint32_t config_size, | ||
1379 | + char *config, const VduseOps *ops, void *priv) | ||
1380 | +{ | ||
1381 | + VduseDev *dev; | ||
1382 | + int ret, ctrl_fd; | ||
1383 | + uint64_t version; | ||
1384 | + struct vduse_dev_config *dev_config; | ||
1385 | + size_t size = offsetof(struct vduse_dev_config, config); | ||
1386 | + | ||
1387 | + if (!name || vduse_name_is_valid(name) || | ||
1388 | + !has_feature(features, VIRTIO_F_VERSION_1) || !config || | ||
1389 | + !config_size || !ops || !ops->enable_queue || !ops->disable_queue) { | ||
1390 | + fprintf(stderr, "Invalid parameter for vduse\n"); | ||
1391 | + return NULL; | ||
1392 | + } | ||
1393 | + | ||
1394 | + dev = calloc(sizeof(VduseDev), 1); | ||
1395 | + if (!dev) { | ||
1396 | + fprintf(stderr, "Failed to allocate vduse device\n"); | ||
1397 | + return NULL; | ||
1398 | + } | ||
1399 | + | ||
1400 | + ctrl_fd = open("/dev/vduse/control", O_RDWR); | ||
1401 | + if (ctrl_fd < 0) { | ||
1402 | + fprintf(stderr, "Failed to open /dev/vduse/control: %s\n", | ||
1403 | + strerror(errno)); | ||
1404 | + goto err_ctrl; | ||
1405 | + } | ||
1406 | + | ||
1407 | + version = VDUSE_API_VERSION; | ||
1408 | + if (ioctl(ctrl_fd, VDUSE_SET_API_VERSION, &version)) { | ||
1409 | + fprintf(stderr, "Failed to set api version %lu: %s\n", | ||
1410 | + version, strerror(errno)); | ||
1411 | + goto err_dev; | ||
1412 | + } | ||
1413 | + | ||
1414 | + dev_config = calloc(size + config_size, 1); | ||
1415 | + if (!dev_config) { | ||
1416 | + fprintf(stderr, "Failed to allocate config space\n"); | ||
1417 | + goto err_dev; | ||
1418 | + } | ||
1419 | + | ||
1420 | + strcpy(dev_config->name, name); | ||
1421 | + dev_config->device_id = device_id; | ||
1422 | + dev_config->vendor_id = vendor_id; | ||
1423 | + dev_config->features = features; | ||
1424 | + dev_config->vq_num = num_queues; | ||
1425 | + dev_config->vq_align = VDUSE_VQ_ALIGN; | ||
1426 | + dev_config->config_size = config_size; | ||
1427 | + memcpy(dev_config->config, config, config_size); | ||
1428 | + | ||
1429 | + ret = ioctl(ctrl_fd, VDUSE_CREATE_DEV, dev_config); | ||
1430 | + free(dev_config); | ||
1431 | + if (ret < 0) { | ||
1432 | + fprintf(stderr, "Failed to create vduse device %s: %s\n", | ||
1433 | + name, strerror(errno)); | ||
1434 | + goto err_dev; | ||
1435 | + } | ||
1436 | + dev->ctrl_fd = ctrl_fd; | ||
1437 | + | ||
1438 | + ret = vduse_dev_init(dev, name, num_queues, ops, priv); | ||
1439 | + if (ret < 0) { | ||
1440 | + fprintf(stderr, "Failed to init vduse device %s: %s\n", | ||
1441 | + name, strerror(ret)); | ||
1442 | + goto err; | ||
1443 | + } | ||
1444 | + | ||
1445 | + return dev; | ||
1446 | +err: | ||
1447 | + ioctl(ctrl_fd, VDUSE_DESTROY_DEV, name); | ||
1448 | +err_dev: | ||
1449 | + close(ctrl_fd); | ||
1450 | +err_ctrl: | ||
1451 | + free(dev); | ||
1452 | + | ||
1453 | + return NULL; | ||
1454 | +} | ||
1455 | + | ||
1456 | +int vduse_dev_destroy(VduseDev *dev) | ||
1457 | +{ | ||
1458 | + int ret = 0; | ||
1459 | + | ||
1460 | + free(dev->vqs); | ||
1461 | + if (dev->fd > 0) { | ||
1462 | + close(dev->fd); | ||
1463 | + dev->fd = -1; | ||
1464 | + } | ||
1465 | + if (dev->ctrl_fd > 0) { | ||
1466 | + if (ioctl(dev->ctrl_fd, VDUSE_DESTROY_DEV, dev->name)) { | ||
1467 | + ret = -errno; | ||
1468 | + } | ||
1469 | + close(dev->ctrl_fd); | ||
1470 | + dev->ctrl_fd = -1; | ||
1471 | + } | ||
1472 | + free(dev->name); | ||
1473 | + free(dev); | ||
1474 | + | ||
1475 | + return ret; | ||
1476 | +} | ||
1477 | diff --git a/MAINTAINERS b/MAINTAINERS | ||
109 | index XXXXXXX..XXXXXXX 100644 | 1478 | index XXXXXXX..XXXXXXX 100644 |
110 | --- a/tests/qemu-iotests/group | 1479 | --- a/MAINTAINERS |
111 | +++ b/tests/qemu-iotests/group | 1480 | +++ b/MAINTAINERS |
1481 | @@ -XXX,XX +XXX,XX @@ L: qemu-block@nongnu.org | ||
1482 | S: Supported | ||
1483 | F: block/export/fuse.c | ||
1484 | |||
1485 | +VDUSE library | ||
1486 | +M: Xie Yongji <xieyongji@bytedance.com> | ||
1487 | +S: Maintained | ||
1488 | +F: subprojects/libvduse/ | ||
1489 | + | ||
1490 | Replication | ||
1491 | M: Wen Congyang <wencongyang2@huawei.com> | ||
1492 | M: Xie Changlong <xiechanglong.d@gmail.com> | ||
1493 | diff --git a/meson.build b/meson.build | ||
1494 | index XXXXXXX..XXXXXXX 100644 | ||
1495 | --- a/meson.build | ||
1496 | +++ b/meson.build | ||
1497 | @@ -XXX,XX +XXX,XX @@ if get_option('fuse_lseek').allowed() | ||
1498 | endif | ||
1499 | endif | ||
1500 | |||
1501 | +have_libvduse = (targetos == 'linux') | ||
1502 | +if get_option('libvduse').enabled() | ||
1503 | + if targetos != 'linux' | ||
1504 | + error('libvduse requires linux') | ||
1505 | + endif | ||
1506 | +elif get_option('libvduse').disabled() | ||
1507 | + have_libvduse = false | ||
1508 | +endif | ||
1509 | + | ||
1510 | +libvduse = not_found | ||
1511 | +if have_libvduse | ||
1512 | + libvduse_proj = subproject('libvduse') | ||
1513 | + libvduse = libvduse_proj.get_variable('libvduse_dep') | ||
1514 | +endif | ||
1515 | + | ||
1516 | # libbpf | ||
1517 | libbpf = dependency('libbpf', required: get_option('bpf'), method: 'pkg-config') | ||
1518 | if libbpf.found() and not cc.links(''' | ||
1519 | diff --git a/scripts/meson-buildoptions.sh b/scripts/meson-buildoptions.sh | ||
1520 | index XXXXXXX..XXXXXXX 100644 | ||
1521 | --- a/scripts/meson-buildoptions.sh | ||
1522 | +++ b/scripts/meson-buildoptions.sh | ||
1523 | @@ -XXX,XX +XXX,XX @@ meson_options_help() { | ||
1524 | printf "%s\n" ' libssh ssh block device support' | ||
1525 | printf "%s\n" ' libudev Use libudev to enumerate host devices' | ||
1526 | printf "%s\n" ' libusb libusb support for USB passthrough' | ||
1527 | + printf "%s\n" ' libvduse build VDUSE Library' | ||
1528 | printf "%s\n" ' linux-aio Linux AIO support' | ||
1529 | printf "%s\n" ' linux-io-uring Linux io_uring support' | ||
1530 | printf "%s\n" ' live-block-migration' | ||
1531 | @@ -XXX,XX +XXX,XX @@ _meson_option_parse() { | ||
1532 | --disable-libudev) printf "%s" -Dlibudev=disabled ;; | ||
1533 | --enable-libusb) printf "%s" -Dlibusb=enabled ;; | ||
1534 | --disable-libusb) printf "%s" -Dlibusb=disabled ;; | ||
1535 | + --enable-libvduse) printf "%s" -Dlibvduse=enabled ;; | ||
1536 | + --disable-libvduse) printf "%s" -Dlibvduse=disabled ;; | ||
1537 | --enable-linux-aio) printf "%s" -Dlinux_aio=enabled ;; | ||
1538 | --disable-linux-aio) printf "%s" -Dlinux_aio=disabled ;; | ||
1539 | --enable-linux-io-uring) printf "%s" -Dlinux_io_uring=enabled ;; | ||
1540 | diff --git a/subprojects/libvduse/linux-headers/linux b/subprojects/libvduse/linux-headers/linux | ||
1541 | new file mode 120000 | ||
1542 | index XXXXXXX..XXXXXXX | ||
1543 | --- /dev/null | ||
1544 | +++ b/subprojects/libvduse/linux-headers/linux | ||
1545 | @@ -0,0 +1 @@ | ||
1546 | +../../../linux-headers/linux/ | ||
1547 | \ No newline at end of file | ||
1548 | diff --git a/subprojects/libvduse/meson.build b/subprojects/libvduse/meson.build | ||
1549 | new file mode 100644 | ||
1550 | index XXXXXXX..XXXXXXX | ||
1551 | --- /dev/null | ||
1552 | +++ b/subprojects/libvduse/meson.build | ||
112 | @@ -XXX,XX +XXX,XX @@ | 1553 | @@ -XXX,XX +XXX,XX @@ |
113 | 279 rw backing quick | 1554 | +project('libvduse', 'c', |
114 | 280 rw migration quick | 1555 | + license: 'GPL-2.0-or-later', |
115 | 281 rw quick | 1556 | + default_options: ['c_std=gnu99']) |
116 | +282 rw img quick | 1557 | + |
117 | 283 auto quick | 1558 | +libvduse = static_library('vduse', |
118 | 284 rw | 1559 | + files('libvduse.c'), |
119 | 286 rw quick | 1560 | + c_args: '-D_GNU_SOURCE') |
1561 | + | ||
1562 | +libvduse_dep = declare_dependency(link_with: libvduse, | ||
1563 | + include_directories: include_directories('.')) | ||
1564 | diff --git a/subprojects/libvduse/standard-headers/linux b/subprojects/libvduse/standard-headers/linux | ||
1565 | new file mode 120000 | ||
1566 | index XXXXXXX..XXXXXXX | ||
1567 | --- /dev/null | ||
1568 | +++ b/subprojects/libvduse/standard-headers/linux | ||
1569 | @@ -0,0 +1 @@ | ||
1570 | +../../../include/standard-headers/linux/ | ||
1571 | \ No newline at end of file | ||
120 | -- | 1572 | -- |
121 | 2.20.1 | 1573 | 2.35.3 |
122 | 1574 | ||
123 | 1575 | diff view generated by jsdifflib |
1 | From: Peter Krempa <pkrempa@redhat.com> | 1 | From: Xie Yongji <xieyongji@bytedance.com> |
---|---|---|---|
2 | 2 | ||
3 | Anounce that 'blockdev-snapshot' command's permissions allow changing | 3 | This implements a VDUSE block backends based on |
4 | of the backing file if the 'consistent_read' permission is not required. | 4 | the libvduse library. We can use it to export the BDSs |
5 | 5 | for both VM and container (host) usage. | |
6 | This is useful for libvirt to allow late opening of the backing chain | 6 | |
7 | during a blockdev-mirror. | 7 | The new command-line syntax is: |
8 | 8 | ||
9 | Signed-off-by: Peter Krempa <pkrempa@redhat.com> | 9 | $ qemu-storage-daemon \ |
10 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> | 10 | --blockdev file,node-name=drive0,filename=test.img \ |
11 | Message-Id: <20200310113831.27293-8-kwolf@redhat.com> | 11 | --export vduse-blk,node-name=drive0,id=vduse-export0,writable=on |
12 | |||
13 | After the qemu-storage-daemon started, we need to use | ||
14 | the "vdpa" command to attach the device to vDPA bus: | ||
15 | |||
16 | $ vdpa dev add name vduse-export0 mgmtdev vduse | ||
17 | |||
18 | Also the device must be removed via the "vdpa" command | ||
19 | before we stop the qemu-storage-daemon. | ||
20 | |||
21 | Signed-off-by: Xie Yongji <xieyongji@bytedance.com> | ||
22 | Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> | ||
23 | Message-Id: <20220523084611.91-7-xieyongji@bytedance.com> | ||
12 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> | 24 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> |
13 | --- | 25 | --- |
14 | qapi/block-core.json | 9 ++++++++- | 26 | qapi/block-export.json | 28 +++- |
15 | 1 file changed, 8 insertions(+), 1 deletion(-) | 27 | meson_options.txt | 2 + |
16 | 28 | block/export/vduse-blk.h | 20 +++ | |
17 | diff --git a/qapi/block-core.json b/qapi/block-core.json | 29 | block/export/export.c | 6 + |
18 | index XXXXXXX..XXXXXXX 100644 | 30 | block/export/vduse-blk.c | 307 ++++++++++++++++++++++++++++++++++ |
19 | --- a/qapi/block-core.json | 31 | MAINTAINERS | 4 +- |
20 | +++ b/qapi/block-core.json | 32 | block/export/meson.build | 5 + |
21 | @@ -XXX,XX +XXX,XX @@ | 33 | meson.build | 13 ++ |
34 | scripts/meson-buildoptions.sh | 4 + | ||
35 | 9 files changed, 385 insertions(+), 4 deletions(-) | ||
36 | create mode 100644 block/export/vduse-blk.h | ||
37 | create mode 100644 block/export/vduse-blk.c | ||
38 | |||
39 | diff --git a/qapi/block-export.json b/qapi/block-export.json | ||
40 | index XXXXXXX..XXXXXXX 100644 | ||
41 | --- a/qapi/block-export.json | ||
42 | +++ b/qapi/block-export.json | ||
43 | @@ -XXX,XX +XXX,XX @@ | ||
44 | '*allow-other': 'FuseExportAllowOther' }, | ||
45 | 'if': 'CONFIG_FUSE' } | ||
46 | |||
47 | +## | ||
48 | +# @BlockExportOptionsVduseBlk: | ||
49 | +# | ||
50 | +# A vduse-blk block export. | ||
51 | +# | ||
52 | +# @num-queues: the number of virtqueues. Defaults to 1. | ||
53 | +# @queue-size: the size of virtqueue. Defaults to 256. | ||
54 | +# @logical-block-size: Logical block size in bytes. Range [512, PAGE_SIZE] | ||
55 | +# and must be power of 2. Defaults to 512 bytes. | ||
56 | +# | ||
57 | +# Since: 7.1 | ||
58 | +## | ||
59 | +{ 'struct': 'BlockExportOptionsVduseBlk', | ||
60 | + 'data': { '*num-queues': 'uint16', | ||
61 | + '*queue-size': 'uint16', | ||
62 | + '*logical-block-size': 'size'} } | ||
63 | + | ||
64 | ## | ||
65 | # @NbdServerAddOptions: | ||
22 | # | 66 | # |
23 | # For the arguments, see the documentation of BlockdevSnapshot. | 67 | @@ -XXX,XX +XXX,XX @@ |
68 | # @nbd: NBD export | ||
69 | # @vhost-user-blk: vhost-user-blk export (since 5.2) | ||
70 | # @fuse: FUSE export (since: 6.0) | ||
71 | +# @vduse-blk: vduse-blk export (since 7.1) | ||
24 | # | 72 | # |
25 | +# Features: | 73 | # Since: 4.2 |
26 | +# @allow-write-only-overlay: If present, the check whether this operation is safe | 74 | ## |
27 | +# was relaxed so that it can be used to change | 75 | @@ -XXX,XX +XXX,XX @@ |
28 | +# backing file of a destination of a blockdev-mirror. | 76 | 'data': [ 'nbd', |
29 | +# (since 5.0) | 77 | { 'name': 'vhost-user-blk', |
30 | +# | 78 | 'if': 'CONFIG_VHOST_USER_BLK_SERVER' }, |
31 | # Since: 2.5 | 79 | - { 'name': 'fuse', 'if': 'CONFIG_FUSE' } ] } |
80 | + { 'name': 'fuse', 'if': 'CONFIG_FUSE' }, | ||
81 | + { 'name': 'vduse-blk', 'if': 'CONFIG_VDUSE_BLK_EXPORT' } ] } | ||
82 | |||
83 | ## | ||
84 | # @BlockExportOptions: | ||
85 | @@ -XXX,XX +XXX,XX @@ | ||
86 | # Describes a block export, i.e. how single node should be exported on an | ||
87 | # external interface. | ||
32 | # | 88 | # |
33 | # Example: | 89 | -# @id: A unique identifier for the block export (across all export types) |
34 | @@ -XXX,XX +XXX,XX @@ | 90 | +# @id: A unique identifier for the block export (across the host for vduse-blk |
91 | +# export type or across all export types for other types) | ||
35 | # | 92 | # |
93 | # @node-name: The node name of the block node to be exported (since: 5.2) | ||
94 | # | ||
95 | @@ -XXX,XX +XXX,XX @@ | ||
96 | 'vhost-user-blk': { 'type': 'BlockExportOptionsVhostUserBlk', | ||
97 | 'if': 'CONFIG_VHOST_USER_BLK_SERVER' }, | ||
98 | 'fuse': { 'type': 'BlockExportOptionsFuse', | ||
99 | - 'if': 'CONFIG_FUSE' } | ||
100 | + 'if': 'CONFIG_FUSE' }, | ||
101 | + 'vduse-blk': { 'type': 'BlockExportOptionsVduseBlk', | ||
102 | + 'if': 'CONFIG_VDUSE_BLK_EXPORT' } | ||
103 | } } | ||
104 | |||
36 | ## | 105 | ## |
37 | { 'command': 'blockdev-snapshot', | 106 | diff --git a/meson_options.txt b/meson_options.txt |
38 | - 'data': 'BlockdevSnapshot' } | 107 | index XXXXXXX..XXXXXXX 100644 |
39 | + 'data': 'BlockdevSnapshot', | 108 | --- a/meson_options.txt |
40 | + 'features': [ 'allow-write-only-overlay' ] } | 109 | +++ b/meson_options.txt |
41 | 110 | @@ -XXX,XX +XXX,XX @@ option('virtiofsd', type: 'feature', value: 'auto', | |
42 | ## | 111 | description: 'build virtiofs daemon (virtiofsd)') |
43 | # @change-backing-file: | 112 | option('libvduse', type: 'feature', value: 'auto', |
113 | description: 'build VDUSE Library') | ||
114 | +option('vduse_blk_export', type: 'feature', value: 'auto', | ||
115 | + description: 'VDUSE block export support') | ||
116 | |||
117 | option('capstone', type: 'feature', value: 'auto', | ||
118 | description: 'Whether and how to find the capstone library') | ||
119 | diff --git a/block/export/vduse-blk.h b/block/export/vduse-blk.h | ||
120 | new file mode 100644 | ||
121 | index XXXXXXX..XXXXXXX | ||
122 | --- /dev/null | ||
123 | +++ b/block/export/vduse-blk.h | ||
124 | @@ -XXX,XX +XXX,XX @@ | ||
125 | +/* | ||
126 | + * Export QEMU block device via VDUSE | ||
127 | + * | ||
128 | + * Copyright (C) 2022 Bytedance Inc. and/or its affiliates. All rights reserved. | ||
129 | + * | ||
130 | + * Author: | ||
131 | + * Xie Yongji <xieyongji@bytedance.com> | ||
132 | + * | ||
133 | + * This work is licensed under the terms of the GNU GPL, version 2 or | ||
134 | + * later. See the COPYING file in the top-level directory. | ||
135 | + */ | ||
136 | + | ||
137 | +#ifndef VDUSE_BLK_H | ||
138 | +#define VDUSE_BLK_H | ||
139 | + | ||
140 | +#include "block/export.h" | ||
141 | + | ||
142 | +extern const BlockExportDriver blk_exp_vduse_blk; | ||
143 | + | ||
144 | +#endif /* VDUSE_BLK_H */ | ||
145 | diff --git a/block/export/export.c b/block/export/export.c | ||
146 | index XXXXXXX..XXXXXXX 100644 | ||
147 | --- a/block/export/export.c | ||
148 | +++ b/block/export/export.c | ||
149 | @@ -XXX,XX +XXX,XX @@ | ||
150 | #ifdef CONFIG_VHOST_USER_BLK_SERVER | ||
151 | #include "vhost-user-blk-server.h" | ||
152 | #endif | ||
153 | +#ifdef CONFIG_VDUSE_BLK_EXPORT | ||
154 | +#include "vduse-blk.h" | ||
155 | +#endif | ||
156 | |||
157 | static const BlockExportDriver *blk_exp_drivers[] = { | ||
158 | &blk_exp_nbd, | ||
159 | @@ -XXX,XX +XXX,XX @@ static const BlockExportDriver *blk_exp_drivers[] = { | ||
160 | #ifdef CONFIG_FUSE | ||
161 | &blk_exp_fuse, | ||
162 | #endif | ||
163 | +#ifdef CONFIG_VDUSE_BLK_EXPORT | ||
164 | + &blk_exp_vduse_blk, | ||
165 | +#endif | ||
166 | }; | ||
167 | |||
168 | /* Only accessed from the main thread */ | ||
169 | diff --git a/block/export/vduse-blk.c b/block/export/vduse-blk.c | ||
170 | new file mode 100644 | ||
171 | index XXXXXXX..XXXXXXX | ||
172 | --- /dev/null | ||
173 | +++ b/block/export/vduse-blk.c | ||
174 | @@ -XXX,XX +XXX,XX @@ | ||
175 | +/* | ||
176 | + * Export QEMU block device via VDUSE | ||
177 | + * | ||
178 | + * Copyright (C) 2022 Bytedance Inc. and/or its affiliates. All rights reserved. | ||
179 | + * | ||
180 | + * Author: | ||
181 | + * Xie Yongji <xieyongji@bytedance.com> | ||
182 | + * | ||
183 | + * This work is licensed under the terms of the GNU GPL, version 2 or | ||
184 | + * later. See the COPYING file in the top-level directory. | ||
185 | + */ | ||
186 | + | ||
187 | +#include <sys/eventfd.h> | ||
188 | + | ||
189 | +#include "qemu/osdep.h" | ||
190 | +#include "qapi/error.h" | ||
191 | +#include "block/export.h" | ||
192 | +#include "qemu/error-report.h" | ||
193 | +#include "util/block-helpers.h" | ||
194 | +#include "subprojects/libvduse/libvduse.h" | ||
195 | +#include "virtio-blk-handler.h" | ||
196 | + | ||
197 | +#include "standard-headers/linux/virtio_blk.h" | ||
198 | + | ||
199 | +#define VDUSE_DEFAULT_NUM_QUEUE 1 | ||
200 | +#define VDUSE_DEFAULT_QUEUE_SIZE 256 | ||
201 | + | ||
202 | +typedef struct VduseBlkExport { | ||
203 | + BlockExport export; | ||
204 | + VirtioBlkHandler handler; | ||
205 | + VduseDev *dev; | ||
206 | + uint16_t num_queues; | ||
207 | +} VduseBlkExport; | ||
208 | + | ||
209 | +typedef struct VduseBlkReq { | ||
210 | + VduseVirtqElement elem; | ||
211 | + VduseVirtq *vq; | ||
212 | +} VduseBlkReq; | ||
213 | + | ||
214 | +static void vduse_blk_req_complete(VduseBlkReq *req, size_t in_len) | ||
215 | +{ | ||
216 | + vduse_queue_push(req->vq, &req->elem, in_len); | ||
217 | + vduse_queue_notify(req->vq); | ||
218 | + | ||
219 | + free(req); | ||
220 | +} | ||
221 | + | ||
222 | +static void coroutine_fn vduse_blk_virtio_process_req(void *opaque) | ||
223 | +{ | ||
224 | + VduseBlkReq *req = opaque; | ||
225 | + VduseVirtq *vq = req->vq; | ||
226 | + VduseDev *dev = vduse_queue_get_dev(vq); | ||
227 | + VduseBlkExport *vblk_exp = vduse_dev_get_priv(dev); | ||
228 | + VirtioBlkHandler *handler = &vblk_exp->handler; | ||
229 | + VduseVirtqElement *elem = &req->elem; | ||
230 | + struct iovec *in_iov = elem->in_sg; | ||
231 | + struct iovec *out_iov = elem->out_sg; | ||
232 | + unsigned in_num = elem->in_num; | ||
233 | + unsigned out_num = elem->out_num; | ||
234 | + int in_len; | ||
235 | + | ||
236 | + in_len = virtio_blk_process_req(handler, in_iov, | ||
237 | + out_iov, in_num, out_num); | ||
238 | + if (in_len < 0) { | ||
239 | + free(req); | ||
240 | + return; | ||
241 | + } | ||
242 | + | ||
243 | + vduse_blk_req_complete(req, in_len); | ||
244 | +} | ||
245 | + | ||
246 | +static void vduse_blk_vq_handler(VduseDev *dev, VduseVirtq *vq) | ||
247 | +{ | ||
248 | + while (1) { | ||
249 | + VduseBlkReq *req; | ||
250 | + | ||
251 | + req = vduse_queue_pop(vq, sizeof(VduseBlkReq)); | ||
252 | + if (!req) { | ||
253 | + break; | ||
254 | + } | ||
255 | + req->vq = vq; | ||
256 | + | ||
257 | + Coroutine *co = | ||
258 | + qemu_coroutine_create(vduse_blk_virtio_process_req, req); | ||
259 | + qemu_coroutine_enter(co); | ||
260 | + } | ||
261 | +} | ||
262 | + | ||
263 | +static void on_vduse_vq_kick(void *opaque) | ||
264 | +{ | ||
265 | + VduseVirtq *vq = opaque; | ||
266 | + VduseDev *dev = vduse_queue_get_dev(vq); | ||
267 | + int fd = vduse_queue_get_fd(vq); | ||
268 | + eventfd_t kick_data; | ||
269 | + | ||
270 | + if (eventfd_read(fd, &kick_data) == -1) { | ||
271 | + error_report("failed to read data from eventfd"); | ||
272 | + return; | ||
273 | + } | ||
274 | + | ||
275 | + vduse_blk_vq_handler(dev, vq); | ||
276 | +} | ||
277 | + | ||
278 | +static void vduse_blk_enable_queue(VduseDev *dev, VduseVirtq *vq) | ||
279 | +{ | ||
280 | + VduseBlkExport *vblk_exp = vduse_dev_get_priv(dev); | ||
281 | + | ||
282 | + aio_set_fd_handler(vblk_exp->export.ctx, vduse_queue_get_fd(vq), | ||
283 | + true, on_vduse_vq_kick, NULL, NULL, NULL, vq); | ||
284 | +} | ||
285 | + | ||
286 | +static void vduse_blk_disable_queue(VduseDev *dev, VduseVirtq *vq) | ||
287 | +{ | ||
288 | + VduseBlkExport *vblk_exp = vduse_dev_get_priv(dev); | ||
289 | + | ||
290 | + aio_set_fd_handler(vblk_exp->export.ctx, vduse_queue_get_fd(vq), | ||
291 | + true, NULL, NULL, NULL, NULL, NULL); | ||
292 | +} | ||
293 | + | ||
294 | +static const VduseOps vduse_blk_ops = { | ||
295 | + .enable_queue = vduse_blk_enable_queue, | ||
296 | + .disable_queue = vduse_blk_disable_queue, | ||
297 | +}; | ||
298 | + | ||
299 | +static void on_vduse_dev_kick(void *opaque) | ||
300 | +{ | ||
301 | + VduseDev *dev = opaque; | ||
302 | + | ||
303 | + vduse_dev_handler(dev); | ||
304 | +} | ||
305 | + | ||
306 | +static void vduse_blk_attach_ctx(VduseBlkExport *vblk_exp, AioContext *ctx) | ||
307 | +{ | ||
308 | + int i; | ||
309 | + | ||
310 | + aio_set_fd_handler(vblk_exp->export.ctx, vduse_dev_get_fd(vblk_exp->dev), | ||
311 | + true, on_vduse_dev_kick, NULL, NULL, NULL, | ||
312 | + vblk_exp->dev); | ||
313 | + | ||
314 | + for (i = 0; i < vblk_exp->num_queues; i++) { | ||
315 | + VduseVirtq *vq = vduse_dev_get_queue(vblk_exp->dev, i); | ||
316 | + int fd = vduse_queue_get_fd(vq); | ||
317 | + | ||
318 | + if (fd < 0) { | ||
319 | + continue; | ||
320 | + } | ||
321 | + aio_set_fd_handler(vblk_exp->export.ctx, fd, true, | ||
322 | + on_vduse_vq_kick, NULL, NULL, NULL, vq); | ||
323 | + } | ||
324 | +} | ||
325 | + | ||
326 | +static void vduse_blk_detach_ctx(VduseBlkExport *vblk_exp) | ||
327 | +{ | ||
328 | + int i; | ||
329 | + | ||
330 | + for (i = 0; i < vblk_exp->num_queues; i++) { | ||
331 | + VduseVirtq *vq = vduse_dev_get_queue(vblk_exp->dev, i); | ||
332 | + int fd = vduse_queue_get_fd(vq); | ||
333 | + | ||
334 | + if (fd < 0) { | ||
335 | + continue; | ||
336 | + } | ||
337 | + aio_set_fd_handler(vblk_exp->export.ctx, fd, | ||
338 | + true, NULL, NULL, NULL, NULL, NULL); | ||
339 | + } | ||
340 | + aio_set_fd_handler(vblk_exp->export.ctx, vduse_dev_get_fd(vblk_exp->dev), | ||
341 | + true, NULL, NULL, NULL, NULL, NULL); | ||
342 | +} | ||
343 | + | ||
344 | + | ||
345 | +static void blk_aio_attached(AioContext *ctx, void *opaque) | ||
346 | +{ | ||
347 | + VduseBlkExport *vblk_exp = opaque; | ||
348 | + | ||
349 | + vblk_exp->export.ctx = ctx; | ||
350 | + vduse_blk_attach_ctx(vblk_exp, ctx); | ||
351 | +} | ||
352 | + | ||
353 | +static void blk_aio_detach(void *opaque) | ||
354 | +{ | ||
355 | + VduseBlkExport *vblk_exp = opaque; | ||
356 | + | ||
357 | + vduse_blk_detach_ctx(vblk_exp); | ||
358 | + vblk_exp->export.ctx = NULL; | ||
359 | +} | ||
360 | + | ||
361 | +static int vduse_blk_exp_create(BlockExport *exp, BlockExportOptions *opts, | ||
362 | + Error **errp) | ||
363 | +{ | ||
364 | + VduseBlkExport *vblk_exp = container_of(exp, VduseBlkExport, export); | ||
365 | + BlockExportOptionsVduseBlk *vblk_opts = &opts->u.vduse_blk; | ||
366 | + uint64_t logical_block_size = VIRTIO_BLK_SECTOR_SIZE; | ||
367 | + uint16_t num_queues = VDUSE_DEFAULT_NUM_QUEUE; | ||
368 | + uint16_t queue_size = VDUSE_DEFAULT_QUEUE_SIZE; | ||
369 | + Error *local_err = NULL; | ||
370 | + struct virtio_blk_config config = { 0 }; | ||
371 | + uint64_t features; | ||
372 | + int i; | ||
373 | + | ||
374 | + if (vblk_opts->has_num_queues) { | ||
375 | + num_queues = vblk_opts->num_queues; | ||
376 | + if (num_queues == 0) { | ||
377 | + error_setg(errp, "num-queues must be greater than 0"); | ||
378 | + return -EINVAL; | ||
379 | + } | ||
380 | + } | ||
381 | + | ||
382 | + if (vblk_opts->has_queue_size) { | ||
383 | + queue_size = vblk_opts->queue_size; | ||
384 | + if (queue_size <= 2 || !is_power_of_2(queue_size) || | ||
385 | + queue_size > VIRTQUEUE_MAX_SIZE) { | ||
386 | + error_setg(errp, "queue-size is invalid"); | ||
387 | + return -EINVAL; | ||
388 | + } | ||
389 | + } | ||
390 | + | ||
391 | + if (vblk_opts->has_logical_block_size) { | ||
392 | + logical_block_size = vblk_opts->logical_block_size; | ||
393 | + check_block_size(exp->id, "logical-block-size", logical_block_size, | ||
394 | + &local_err); | ||
395 | + if (local_err) { | ||
396 | + error_propagate(errp, local_err); | ||
397 | + return -EINVAL; | ||
398 | + } | ||
399 | + } | ||
400 | + vblk_exp->num_queues = num_queues; | ||
401 | + vblk_exp->handler.blk = exp->blk; | ||
402 | + vblk_exp->handler.serial = exp->id; | ||
403 | + vblk_exp->handler.logical_block_size = logical_block_size; | ||
404 | + vblk_exp->handler.writable = opts->writable; | ||
405 | + | ||
406 | + config.capacity = | ||
407 | + cpu_to_le64(blk_getlength(exp->blk) >> VIRTIO_BLK_SECTOR_BITS); | ||
408 | + config.seg_max = cpu_to_le32(queue_size - 2); | ||
409 | + config.min_io_size = cpu_to_le16(1); | ||
410 | + config.opt_io_size = cpu_to_le32(1); | ||
411 | + config.num_queues = cpu_to_le16(num_queues); | ||
412 | + config.blk_size = cpu_to_le32(logical_block_size); | ||
413 | + config.max_discard_sectors = cpu_to_le32(VIRTIO_BLK_MAX_DISCARD_SECTORS); | ||
414 | + config.max_discard_seg = cpu_to_le32(1); | ||
415 | + config.discard_sector_alignment = | ||
416 | + cpu_to_le32(logical_block_size >> VIRTIO_BLK_SECTOR_BITS); | ||
417 | + config.max_write_zeroes_sectors = | ||
418 | + cpu_to_le32(VIRTIO_BLK_MAX_WRITE_ZEROES_SECTORS); | ||
419 | + config.max_write_zeroes_seg = cpu_to_le32(1); | ||
420 | + | ||
421 | + features = vduse_get_virtio_features() | | ||
422 | + (1ULL << VIRTIO_BLK_F_SEG_MAX) | | ||
423 | + (1ULL << VIRTIO_BLK_F_TOPOLOGY) | | ||
424 | + (1ULL << VIRTIO_BLK_F_BLK_SIZE) | | ||
425 | + (1ULL << VIRTIO_BLK_F_FLUSH) | | ||
426 | + (1ULL << VIRTIO_BLK_F_DISCARD) | | ||
427 | + (1ULL << VIRTIO_BLK_F_WRITE_ZEROES); | ||
428 | + | ||
429 | + if (num_queues > 1) { | ||
430 | + features |= 1ULL << VIRTIO_BLK_F_MQ; | ||
431 | + } | ||
432 | + if (!opts->writable) { | ||
433 | + features |= 1ULL << VIRTIO_BLK_F_RO; | ||
434 | + } | ||
435 | + | ||
436 | + vblk_exp->dev = vduse_dev_create(exp->id, VIRTIO_ID_BLOCK, 0, | ||
437 | + features, num_queues, | ||
438 | + sizeof(struct virtio_blk_config), | ||
439 | + (char *)&config, &vduse_blk_ops, | ||
440 | + vblk_exp); | ||
441 | + if (!vblk_exp->dev) { | ||
442 | + error_setg(errp, "failed to create vduse device"); | ||
443 | + return -ENOMEM; | ||
444 | + } | ||
445 | + | ||
446 | + for (i = 0; i < num_queues; i++) { | ||
447 | + vduse_dev_setup_queue(vblk_exp->dev, i, queue_size); | ||
448 | + } | ||
449 | + | ||
450 | + aio_set_fd_handler(exp->ctx, vduse_dev_get_fd(vblk_exp->dev), true, | ||
451 | + on_vduse_dev_kick, NULL, NULL, NULL, vblk_exp->dev); | ||
452 | + | ||
453 | + blk_add_aio_context_notifier(exp->blk, blk_aio_attached, blk_aio_detach, | ||
454 | + vblk_exp); | ||
455 | + | ||
456 | + return 0; | ||
457 | +} | ||
458 | + | ||
459 | +static void vduse_blk_exp_delete(BlockExport *exp) | ||
460 | +{ | ||
461 | + VduseBlkExport *vblk_exp = container_of(exp, VduseBlkExport, export); | ||
462 | + | ||
463 | + blk_remove_aio_context_notifier(exp->blk, blk_aio_attached, blk_aio_detach, | ||
464 | + vblk_exp); | ||
465 | + vduse_dev_destroy(vblk_exp->dev); | ||
466 | +} | ||
467 | + | ||
468 | +static void vduse_blk_exp_request_shutdown(BlockExport *exp) | ||
469 | +{ | ||
470 | + VduseBlkExport *vblk_exp = container_of(exp, VduseBlkExport, export); | ||
471 | + | ||
472 | + vduse_blk_detach_ctx(vblk_exp); | ||
473 | +} | ||
474 | + | ||
475 | +const BlockExportDriver blk_exp_vduse_blk = { | ||
476 | + .type = BLOCK_EXPORT_TYPE_VDUSE_BLK, | ||
477 | + .instance_size = sizeof(VduseBlkExport), | ||
478 | + .create = vduse_blk_exp_create, | ||
479 | + .delete = vduse_blk_exp_delete, | ||
480 | + .request_shutdown = vduse_blk_exp_request_shutdown, | ||
481 | +}; | ||
482 | diff --git a/MAINTAINERS b/MAINTAINERS | ||
483 | index XXXXXXX..XXXXXXX 100644 | ||
484 | --- a/MAINTAINERS | ||
485 | +++ b/MAINTAINERS | ||
486 | @@ -XXX,XX +XXX,XX @@ L: qemu-block@nongnu.org | ||
487 | S: Supported | ||
488 | F: block/export/fuse.c | ||
489 | |||
490 | -VDUSE library | ||
491 | +VDUSE library and block device exports | ||
492 | M: Xie Yongji <xieyongji@bytedance.com> | ||
493 | S: Maintained | ||
494 | F: subprojects/libvduse/ | ||
495 | +F: block/export/vduse-blk.c | ||
496 | +F: block/export/vduse-blk.h | ||
497 | |||
498 | Replication | ||
499 | M: Wen Congyang <wencongyang2@huawei.com> | ||
500 | diff --git a/block/export/meson.build b/block/export/meson.build | ||
501 | index XXXXXXX..XXXXXXX 100644 | ||
502 | --- a/block/export/meson.build | ||
503 | +++ b/block/export/meson.build | ||
504 | @@ -XXX,XX +XXX,XX @@ if have_vhost_user_blk_server | ||
505 | endif | ||
506 | |||
507 | blockdev_ss.add(when: fuse, if_true: files('fuse.c')) | ||
508 | + | ||
509 | +if have_vduse_blk_export | ||
510 | + blockdev_ss.add(files('vduse-blk.c', 'virtio-blk-handler.c')) | ||
511 | + blockdev_ss.add(libvduse) | ||
512 | +endif | ||
513 | diff --git a/meson.build b/meson.build | ||
514 | index XXXXXXX..XXXXXXX 100644 | ||
515 | --- a/meson.build | ||
516 | +++ b/meson.build | ||
517 | @@ -XXX,XX +XXX,XX @@ if have_libvduse | ||
518 | libvduse = libvduse_proj.get_variable('libvduse_dep') | ||
519 | endif | ||
520 | |||
521 | +have_vduse_blk_export = (have_libvduse and targetos == 'linux') | ||
522 | +if get_option('vduse_blk_export').enabled() | ||
523 | + if targetos != 'linux' | ||
524 | + error('vduse_blk_export requires linux') | ||
525 | + elif not have_libvduse | ||
526 | + error('vduse_blk_export requires libvduse support') | ||
527 | + endif | ||
528 | +elif get_option('vduse_blk_export').disabled() | ||
529 | + have_vduse_blk_export = false | ||
530 | +endif | ||
531 | + | ||
532 | # libbpf | ||
533 | libbpf = dependency('libbpf', required: get_option('bpf'), method: 'pkg-config') | ||
534 | if libbpf.found() and not cc.links(''' | ||
535 | @@ -XXX,XX +XXX,XX @@ config_host_data.set('CONFIG_VHOST_CRYPTO', have_vhost_user_crypto) | ||
536 | config_host_data.set('CONFIG_VHOST_VDPA', have_vhost_vdpa) | ||
537 | config_host_data.set('CONFIG_VMNET', vmnet.found()) | ||
538 | config_host_data.set('CONFIG_VHOST_USER_BLK_SERVER', have_vhost_user_blk_server) | ||
539 | +config_host_data.set('CONFIG_VDUSE_BLK_EXPORT', have_vduse_blk_export) | ||
540 | config_host_data.set('CONFIG_PNG', png.found()) | ||
541 | config_host_data.set('CONFIG_VNC', vnc.found()) | ||
542 | config_host_data.set('CONFIG_VNC_JPEG', jpeg.found()) | ||
543 | @@ -XXX,XX +XXX,XX @@ if have_block | ||
544 | summary_info += {'qed support': get_option('qed').allowed()} | ||
545 | summary_info += {'parallels support': get_option('parallels').allowed()} | ||
546 | summary_info += {'FUSE exports': fuse} | ||
547 | + summary_info += {'VDUSE block exports': have_vduse_blk_export} | ||
548 | endif | ||
549 | summary(summary_info, bool_yn: true, section: 'Block layer support') | ||
550 | |||
551 | diff --git a/scripts/meson-buildoptions.sh b/scripts/meson-buildoptions.sh | ||
552 | index XXXXXXX..XXXXXXX 100644 | ||
553 | --- a/scripts/meson-buildoptions.sh | ||
554 | +++ b/scripts/meson-buildoptions.sh | ||
555 | @@ -XXX,XX +XXX,XX @@ meson_options_help() { | ||
556 | printf "%s\n" ' vhost-user vhost-user backend support' | ||
557 | printf "%s\n" ' vhost-user-blk-server' | ||
558 | printf "%s\n" ' build vhost-user-blk server' | ||
559 | + printf "%s\n" ' vduse-blk-export' | ||
560 | + printf "%s\n" ' VDUSE block export support' | ||
561 | printf "%s\n" ' vhost-vdpa vhost-vdpa kernel backend support' | ||
562 | printf "%s\n" ' virglrenderer virgl rendering support' | ||
563 | printf "%s\n" ' virtfs virtio-9p support' | ||
564 | @@ -XXX,XX +XXX,XX @@ _meson_option_parse() { | ||
565 | --disable-vhost-user) printf "%s" -Dvhost_user=disabled ;; | ||
566 | --enable-vhost-user-blk-server) printf "%s" -Dvhost_user_blk_server=enabled ;; | ||
567 | --disable-vhost-user-blk-server) printf "%s" -Dvhost_user_blk_server=disabled ;; | ||
568 | + --enable-vduse-blk-export) printf "%s" -Dvduse_blk_export=enabled ;; | ||
569 | + --disable-vduse-blk-export) printf "%s" -Dvduse_blk_export=disabled ;; | ||
570 | --enable-vhost-vdpa) printf "%s" -Dvhost_vdpa=enabled ;; | ||
571 | --disable-vhost-vdpa) printf "%s" -Dvhost_vdpa=disabled ;; | ||
572 | --enable-virglrenderer) printf "%s" -Dvirglrenderer=enabled ;; | ||
44 | -- | 573 | -- |
45 | 2.20.1 | 574 | 2.35.3 |
46 | |||
47 | diff view generated by jsdifflib |
1 | From: Daniel Henrique Barboza <danielhb413@gmail.com> | 1 | From: Xie Yongji <xieyongji@bytedance.com> |
---|---|---|---|
2 | 2 | ||
3 | Using the new 'bdrv_co_delete_file' interface, a pure co_routine function | 3 | To support block resize, this uses vduse_dev_update_config() |
4 | 'bdrv_co_delete_file' inside block.c can can be used in a way similar of | 4 | to update the capacity field in configuration space and inject |
5 | the existing bdrv_create_file to to clean up a created file. | 5 | config interrupt on the block resize callback. |
6 | 6 | ||
7 | We're creating a pure co_routine because the only caller of | 7 | Signed-off-by: Xie Yongji <xieyongji@bytedance.com> |
8 | 'bdrv_co_delete_file' will be already in co_routine context, thus there | 8 | Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> |
9 | is no need to add all the machinery to check for qemu_in_coroutine() and | 9 | Message-Id: <20220523084611.91-8-xieyongji@bytedance.com> |
10 | create a separated co_routine to do the job. | ||
11 | |||
12 | Suggested-by: Daniel P. Berrangé <berrange@redhat.com> | ||
13 | Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com> | ||
14 | Message-Id: <20200130213907.2830642-3-danielhb413@gmail.com> | ||
15 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> | 10 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> |
16 | --- | 11 | --- |
17 | include/block/block.h | 1 + | 12 | block/export/vduse-blk.c | 20 ++++++++++++++++++++ |
18 | block.c | 26 ++++++++++++++++++++++++++ | 13 | 1 file changed, 20 insertions(+) |
19 | 2 files changed, 27 insertions(+) | ||
20 | 14 | ||
21 | diff --git a/include/block/block.h b/include/block/block.h | 15 | diff --git a/block/export/vduse-blk.c b/block/export/vduse-blk.c |
22 | index XXXXXXX..XXXXXXX 100644 | 16 | index XXXXXXX..XXXXXXX 100644 |
23 | --- a/include/block/block.h | 17 | --- a/block/export/vduse-blk.c |
24 | +++ b/include/block/block.h | 18 | +++ b/block/export/vduse-blk.c |
25 | @@ -XXX,XX +XXX,XX @@ bool bdrv_is_backing_chain_frozen(BlockDriverState *bs, BlockDriverState *base, | 19 | @@ -XXX,XX +XXX,XX @@ static void blk_aio_detach(void *opaque) |
26 | int bdrv_freeze_backing_chain(BlockDriverState *bs, BlockDriverState *base, | 20 | vblk_exp->export.ctx = NULL; |
27 | Error **errp); | ||
28 | void bdrv_unfreeze_backing_chain(BlockDriverState *bs, BlockDriverState *base); | ||
29 | +int coroutine_fn bdrv_co_delete_file(BlockDriverState *bs, Error **errp); | ||
30 | |||
31 | |||
32 | typedef struct BdrvCheckResult { | ||
33 | diff --git a/block.c b/block.c | ||
34 | index XXXXXXX..XXXXXXX 100644 | ||
35 | --- a/block.c | ||
36 | +++ b/block.c | ||
37 | @@ -XXX,XX +XXX,XX @@ int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp) | ||
38 | } | ||
39 | } | 21 | } |
40 | 22 | ||
41 | +int coroutine_fn bdrv_co_delete_file(BlockDriverState *bs, Error **errp) | 23 | +static void vduse_blk_resize(void *opaque) |
42 | +{ | 24 | +{ |
43 | + Error *local_err = NULL; | 25 | + BlockExport *exp = opaque; |
44 | + int ret; | 26 | + VduseBlkExport *vblk_exp = container_of(exp, VduseBlkExport, export); |
27 | + struct virtio_blk_config config; | ||
45 | + | 28 | + |
46 | + assert(bs != NULL); | 29 | + config.capacity = |
47 | + | 30 | + cpu_to_le64(blk_getlength(exp->blk) >> VIRTIO_BLK_SECTOR_BITS); |
48 | + if (!bs->drv) { | 31 | + vduse_dev_update_config(vblk_exp->dev, sizeof(config.capacity), |
49 | + error_setg(errp, "Block node '%s' is not opened", bs->filename); | 32 | + offsetof(struct virtio_blk_config, capacity), |
50 | + return -ENOMEDIUM; | 33 | + (char *)&config.capacity); |
51 | + } | ||
52 | + | ||
53 | + if (!bs->drv->bdrv_co_delete_file) { | ||
54 | + error_setg(errp, "Driver '%s' does not support image deletion", | ||
55 | + bs->drv->format_name); | ||
56 | + return -ENOTSUP; | ||
57 | + } | ||
58 | + | ||
59 | + ret = bs->drv->bdrv_co_delete_file(bs, &local_err); | ||
60 | + if (ret < 0) { | ||
61 | + error_propagate(errp, local_err); | ||
62 | + } | ||
63 | + | ||
64 | + return ret; | ||
65 | +} | 34 | +} |
66 | + | 35 | + |
67 | /** | 36 | +static const BlockDevOps vduse_block_ops = { |
68 | * Try to get @bs's logical and physical block size. | 37 | + .resize_cb = vduse_blk_resize, |
69 | * On success, store them in @bsz struct and return 0. | 38 | +}; |
39 | + | ||
40 | static int vduse_blk_exp_create(BlockExport *exp, BlockExportOptions *opts, | ||
41 | Error **errp) | ||
42 | { | ||
43 | @@ -XXX,XX +XXX,XX @@ static int vduse_blk_exp_create(BlockExport *exp, BlockExportOptions *opts, | ||
44 | blk_add_aio_context_notifier(exp->blk, blk_aio_attached, blk_aio_detach, | ||
45 | vblk_exp); | ||
46 | |||
47 | + blk_set_dev_ops(exp->blk, &vduse_block_ops, exp); | ||
48 | + | ||
49 | return 0; | ||
50 | } | ||
51 | |||
52 | @@ -XXX,XX +XXX,XX @@ static void vduse_blk_exp_delete(BlockExport *exp) | ||
53 | |||
54 | blk_remove_aio_context_notifier(exp->blk, blk_aio_attached, blk_aio_detach, | ||
55 | vblk_exp); | ||
56 | + blk_set_dev_ops(exp->blk, NULL, NULL); | ||
57 | vduse_dev_destroy(vblk_exp->dev); | ||
58 | } | ||
59 | |||
70 | -- | 60 | -- |
71 | 2.20.1 | 61 | 2.35.3 |
72 | |||
73 | diff view generated by jsdifflib |
1 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> | 1 | From: Xie Yongji <xieyongji@bytedance.com> |
---|---|---|---|
2 | Message-Id: <20200310113831.27293-2-kwolf@redhat.com> | 2 | |
3 | Reviewed-by: Peter Krempa <pkrempa@redhat.com> | 3 | To support reconnecting after restart or crash, VDUSE backend |
4 | might need to resubmit inflight I/Os. This stores the metadata | ||
5 | such as the index of inflight I/O's descriptors to a shm file so | ||
6 | that VDUSE backend can restore them during reconnecting. | ||
7 | |||
8 | Signed-off-by: Xie Yongji <xieyongji@bytedance.com> | ||
9 | Message-Id: <20220523084611.91-9-xieyongji@bytedance.com> | ||
10 | Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> | ||
4 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> | 11 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> |
5 | --- | 12 | --- |
6 | include/block/block_int.h | 3 +++ | 13 | subprojects/libvduse/libvduse.h | 12 ++ |
7 | block.c | 6 ++---- | 14 | block/export/vduse-blk.c | 14 ++ |
8 | 2 files changed, 5 insertions(+), 4 deletions(-) | 15 | subprojects/libvduse/libvduse.c | 235 +++++++++++++++++++++++++++++++- |
16 | 3 files changed, 256 insertions(+), 5 deletions(-) | ||
9 | 17 | ||
10 | diff --git a/include/block/block_int.h b/include/block/block_int.h | 18 | diff --git a/subprojects/libvduse/libvduse.h b/subprojects/libvduse/libvduse.h |
11 | index XXXXXXX..XXXXXXX 100644 | 19 | index XXXXXXX..XXXXXXX 100644 |
12 | --- a/include/block/block_int.h | 20 | --- a/subprojects/libvduse/libvduse.h |
13 | +++ b/include/block/block_int.h | 21 | +++ b/subprojects/libvduse/libvduse.h |
14 | @@ -XXX,XX +XXX,XX @@ BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs, | 22 | @@ -XXX,XX +XXX,XX @@ int vduse_dev_update_config(VduseDev *dev, uint32_t size, |
15 | void *opaque, Error **errp); | 23 | */ |
16 | void bdrv_root_unref_child(BdrvChild *child); | 24 | int vduse_dev_setup_queue(VduseDev *dev, int index, int max_size); |
17 | 25 | ||
18 | +void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm, | 26 | +/** |
19 | + uint64_t *shared_perm); | 27 | + * vduse_set_reconnect_log_file: |
28 | + * @dev: VDUSE device | ||
29 | + * @file: filename of reconnect log | ||
30 | + * | ||
31 | + * Specify the file to store log for reconnecting. It should | ||
32 | + * be called before vduse_dev_setup_queue(). | ||
33 | + * | ||
34 | + * Returns: 0 on success, -errno on failure. | ||
35 | + */ | ||
36 | +int vduse_set_reconnect_log_file(VduseDev *dev, const char *filename); | ||
20 | + | 37 | + |
21 | /** | 38 | /** |
22 | * Sets a BdrvChild's permissions. Avoid if the parent is a BDS; use | 39 | * vduse_dev_create_by_fd: |
23 | * bdrv_child_refresh_perms() instead and make the parent's | 40 | * @fd: passed file descriptor |
24 | diff --git a/block.c b/block.c | 41 | diff --git a/block/export/vduse-blk.c b/block/export/vduse-blk.c |
25 | index XXXXXXX..XXXXXXX 100644 | 42 | index XXXXXXX..XXXXXXX 100644 |
26 | --- a/block.c | 43 | --- a/block/export/vduse-blk.c |
27 | +++ b/block.c | 44 | +++ b/block/export/vduse-blk.c |
28 | @@ -XXX,XX +XXX,XX @@ static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q, | 45 | @@ -XXX,XX +XXX,XX @@ typedef struct VduseBlkExport { |
29 | bool *tighten_restrictions, Error **errp); | 46 | VirtioBlkHandler handler; |
30 | static void bdrv_child_abort_perm_update(BdrvChild *c); | 47 | VduseDev *dev; |
31 | static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared); | 48 | uint16_t num_queues; |
32 | -static void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm, | 49 | + char *recon_file; |
33 | - uint64_t *shared_perm); | 50 | } VduseBlkExport; |
34 | 51 | ||
35 | typedef struct BlockReopenQueueEntry { | 52 | typedef struct VduseBlkReq { |
36 | bool prepared; | 53 | @@ -XXX,XX +XXX,XX @@ static void vduse_blk_enable_queue(VduseDev *dev, VduseVirtq *vq) |
37 | @@ -XXX,XX +XXX,XX @@ static void bdrv_set_perm(BlockDriverState *bs, uint64_t cumulative_perms, | 54 | |
38 | } | 55 | aio_set_fd_handler(vblk_exp->export.ctx, vduse_queue_get_fd(vq), |
39 | } | 56 | true, on_vduse_vq_kick, NULL, NULL, NULL, vq); |
40 | 57 | + /* Make sure we don't miss any kick afer reconnecting */ | |
41 | -static void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm, | 58 | + eventfd_write(vduse_queue_get_fd(vq), 1); |
42 | - uint64_t *shared_perm) | 59 | } |
43 | +void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm, | 60 | |
44 | + uint64_t *shared_perm) | 61 | static void vduse_blk_disable_queue(VduseDev *dev, VduseVirtq *vq) |
62 | @@ -XXX,XX +XXX,XX @@ static int vduse_blk_exp_create(BlockExport *exp, BlockExportOptions *opts, | ||
63 | return -ENOMEM; | ||
64 | } | ||
65 | |||
66 | + vblk_exp->recon_file = g_strdup_printf("%s/vduse-blk-%s", | ||
67 | + g_get_tmp_dir(), exp->id); | ||
68 | + if (vduse_set_reconnect_log_file(vblk_exp->dev, vblk_exp->recon_file)) { | ||
69 | + error_setg(errp, "failed to set reconnect log file"); | ||
70 | + vduse_dev_destroy(vblk_exp->dev); | ||
71 | + g_free(vblk_exp->recon_file); | ||
72 | + return -EINVAL; | ||
73 | + } | ||
74 | + | ||
75 | for (i = 0; i < num_queues; i++) { | ||
76 | vduse_dev_setup_queue(vblk_exp->dev, i, queue_size); | ||
77 | } | ||
78 | @@ -XXX,XX +XXX,XX @@ static void vduse_blk_exp_delete(BlockExport *exp) | ||
79 | vblk_exp); | ||
80 | blk_set_dev_ops(exp->blk, NULL, NULL); | ||
81 | vduse_dev_destroy(vblk_exp->dev); | ||
82 | + unlink(vblk_exp->recon_file); | ||
83 | + g_free(vblk_exp->recon_file); | ||
84 | } | ||
85 | |||
86 | static void vduse_blk_exp_request_shutdown(BlockExport *exp) | ||
87 | diff --git a/subprojects/libvduse/libvduse.c b/subprojects/libvduse/libvduse.c | ||
88 | index XXXXXXX..XXXXXXX 100644 | ||
89 | --- a/subprojects/libvduse/libvduse.c | ||
90 | +++ b/subprojects/libvduse/libvduse.c | ||
91 | @@ -XXX,XX +XXX,XX @@ | ||
92 | #define VDUSE_VQ_ALIGN 4096 | ||
93 | #define MAX_IOVA_REGIONS 256 | ||
94 | |||
95 | +#define LOG_ALIGNMENT 64 | ||
96 | + | ||
97 | /* Round number down to multiple */ | ||
98 | #define ALIGN_DOWN(n, m) ((n) / (m) * (m)) | ||
99 | |||
100 | @@ -XXX,XX +XXX,XX @@ | ||
101 | #define unlikely(x) __builtin_expect(!!(x), 0) | ||
102 | #endif | ||
103 | |||
104 | +typedef struct VduseDescStateSplit { | ||
105 | + uint8_t inflight; | ||
106 | + uint8_t padding[5]; | ||
107 | + uint16_t next; | ||
108 | + uint64_t counter; | ||
109 | +} VduseDescStateSplit; | ||
110 | + | ||
111 | +typedef struct VduseVirtqLogInflight { | ||
112 | + uint64_t features; | ||
113 | + uint16_t version; | ||
114 | + uint16_t desc_num; | ||
115 | + uint16_t last_batch_head; | ||
116 | + uint16_t used_idx; | ||
117 | + VduseDescStateSplit desc[]; | ||
118 | +} VduseVirtqLogInflight; | ||
119 | + | ||
120 | +typedef struct VduseVirtqLog { | ||
121 | + VduseVirtqLogInflight inflight; | ||
122 | +} VduseVirtqLog; | ||
123 | + | ||
124 | +typedef struct VduseVirtqInflightDesc { | ||
125 | + uint16_t index; | ||
126 | + uint64_t counter; | ||
127 | +} VduseVirtqInflightDesc; | ||
128 | + | ||
129 | typedef struct VduseRing { | ||
130 | unsigned int num; | ||
131 | uint64_t desc_addr; | ||
132 | @@ -XXX,XX +XXX,XX @@ struct VduseVirtq { | ||
133 | bool ready; | ||
134 | int fd; | ||
135 | VduseDev *dev; | ||
136 | + VduseVirtqInflightDesc *resubmit_list; | ||
137 | + uint16_t resubmit_num; | ||
138 | + uint64_t counter; | ||
139 | + VduseVirtqLog *log; | ||
140 | }; | ||
141 | |||
142 | typedef struct VduseIovaRegion { | ||
143 | @@ -XXX,XX +XXX,XX @@ struct VduseDev { | ||
144 | int fd; | ||
145 | int ctrl_fd; | ||
146 | void *priv; | ||
147 | + void *log; | ||
148 | }; | ||
149 | |||
150 | +static inline size_t vduse_vq_log_size(uint16_t queue_size) | ||
151 | +{ | ||
152 | + return ALIGN_UP(sizeof(VduseDescStateSplit) * queue_size + | ||
153 | + sizeof(VduseVirtqLogInflight), LOG_ALIGNMENT); | ||
154 | +} | ||
155 | + | ||
156 | +static void *vduse_log_get(const char *filename, size_t size) | ||
157 | +{ | ||
158 | + void *ptr = MAP_FAILED; | ||
159 | + int fd; | ||
160 | + | ||
161 | + fd = open(filename, O_RDWR | O_CREAT, 0600); | ||
162 | + if (fd == -1) { | ||
163 | + return MAP_FAILED; | ||
164 | + } | ||
165 | + | ||
166 | + if (ftruncate(fd, size) == -1) { | ||
167 | + goto out; | ||
168 | + } | ||
169 | + | ||
170 | + ptr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0); | ||
171 | + | ||
172 | +out: | ||
173 | + close(fd); | ||
174 | + return ptr; | ||
175 | +} | ||
176 | + | ||
177 | static inline bool has_feature(uint64_t features, unsigned int fbit) | ||
45 | { | 178 | { |
46 | BdrvChild *c; | 179 | assert(fbit < 64); |
47 | uint64_t cumulative_perms = 0; | 180 | @@ -XXX,XX +XXX,XX @@ static int vduse_inject_irq(VduseDev *dev, int index) |
181 | return ioctl(dev->fd, VDUSE_VQ_INJECT_IRQ, &index); | ||
182 | } | ||
183 | |||
184 | +static int inflight_desc_compare(const void *a, const void *b) | ||
185 | +{ | ||
186 | + VduseVirtqInflightDesc *desc0 = (VduseVirtqInflightDesc *)a, | ||
187 | + *desc1 = (VduseVirtqInflightDesc *)b; | ||
188 | + | ||
189 | + if (desc1->counter > desc0->counter && | ||
190 | + (desc1->counter - desc0->counter) < VIRTQUEUE_MAX_SIZE * 2) { | ||
191 | + return 1; | ||
192 | + } | ||
193 | + | ||
194 | + return -1; | ||
195 | +} | ||
196 | + | ||
197 | +static int vduse_queue_check_inflights(VduseVirtq *vq) | ||
198 | +{ | ||
199 | + int i = 0; | ||
200 | + VduseDev *dev = vq->dev; | ||
201 | + | ||
202 | + vq->used_idx = le16toh(vq->vring.used->idx); | ||
203 | + vq->resubmit_num = 0; | ||
204 | + vq->resubmit_list = NULL; | ||
205 | + vq->counter = 0; | ||
206 | + | ||
207 | + if (unlikely(vq->log->inflight.used_idx != vq->used_idx)) { | ||
208 | + if (vq->log->inflight.last_batch_head > VIRTQUEUE_MAX_SIZE) { | ||
209 | + return -1; | ||
210 | + } | ||
211 | + | ||
212 | + vq->log->inflight.desc[vq->log->inflight.last_batch_head].inflight = 0; | ||
213 | + | ||
214 | + barrier(); | ||
215 | + | ||
216 | + vq->log->inflight.used_idx = vq->used_idx; | ||
217 | + } | ||
218 | + | ||
219 | + for (i = 0; i < vq->log->inflight.desc_num; i++) { | ||
220 | + if (vq->log->inflight.desc[i].inflight == 1) { | ||
221 | + vq->inuse++; | ||
222 | + } | ||
223 | + } | ||
224 | + | ||
225 | + vq->shadow_avail_idx = vq->last_avail_idx = vq->inuse + vq->used_idx; | ||
226 | + | ||
227 | + if (vq->inuse) { | ||
228 | + vq->resubmit_list = calloc(vq->inuse, sizeof(VduseVirtqInflightDesc)); | ||
229 | + if (!vq->resubmit_list) { | ||
230 | + return -1; | ||
231 | + } | ||
232 | + | ||
233 | + for (i = 0; i < vq->log->inflight.desc_num; i++) { | ||
234 | + if (vq->log->inflight.desc[i].inflight) { | ||
235 | + vq->resubmit_list[vq->resubmit_num].index = i; | ||
236 | + vq->resubmit_list[vq->resubmit_num].counter = | ||
237 | + vq->log->inflight.desc[i].counter; | ||
238 | + vq->resubmit_num++; | ||
239 | + } | ||
240 | + } | ||
241 | + | ||
242 | + if (vq->resubmit_num > 1) { | ||
243 | + qsort(vq->resubmit_list, vq->resubmit_num, | ||
244 | + sizeof(VduseVirtqInflightDesc), inflight_desc_compare); | ||
245 | + } | ||
246 | + vq->counter = vq->resubmit_list[0].counter + 1; | ||
247 | + } | ||
248 | + | ||
249 | + vduse_inject_irq(dev, vq->index); | ||
250 | + | ||
251 | + return 0; | ||
252 | +} | ||
253 | + | ||
254 | +static int vduse_queue_inflight_get(VduseVirtq *vq, int desc_idx) | ||
255 | +{ | ||
256 | + vq->log->inflight.desc[desc_idx].counter = vq->counter++; | ||
257 | + | ||
258 | + barrier(); | ||
259 | + | ||
260 | + vq->log->inflight.desc[desc_idx].inflight = 1; | ||
261 | + | ||
262 | + return 0; | ||
263 | +} | ||
264 | + | ||
265 | +static int vduse_queue_inflight_pre_put(VduseVirtq *vq, int desc_idx) | ||
266 | +{ | ||
267 | + vq->log->inflight.last_batch_head = desc_idx; | ||
268 | + | ||
269 | + return 0; | ||
270 | +} | ||
271 | + | ||
272 | +static int vduse_queue_inflight_post_put(VduseVirtq *vq, int desc_idx) | ||
273 | +{ | ||
274 | + vq->log->inflight.desc[desc_idx].inflight = 0; | ||
275 | + | ||
276 | + barrier(); | ||
277 | + | ||
278 | + vq->log->inflight.used_idx = vq->used_idx; | ||
279 | + | ||
280 | + return 0; | ||
281 | +} | ||
282 | + | ||
283 | static void vduse_iova_remove_region(VduseDev *dev, uint64_t start, | ||
284 | uint64_t last) | ||
285 | { | ||
286 | @@ -XXX,XX +XXX,XX @@ void *vduse_queue_pop(VduseVirtq *vq, size_t sz) | ||
287 | unsigned int head; | ||
288 | VduseVirtqElement *elem; | ||
289 | VduseDev *dev = vq->dev; | ||
290 | + int i; | ||
291 | |||
292 | if (unlikely(!vq->vring.avail)) { | ||
293 | return NULL; | ||
294 | } | ||
295 | |||
296 | + if (unlikely(vq->resubmit_list && vq->resubmit_num > 0)) { | ||
297 | + i = (--vq->resubmit_num); | ||
298 | + elem = vduse_queue_map_desc(vq, vq->resubmit_list[i].index, sz); | ||
299 | + | ||
300 | + if (!vq->resubmit_num) { | ||
301 | + free(vq->resubmit_list); | ||
302 | + vq->resubmit_list = NULL; | ||
303 | + } | ||
304 | + | ||
305 | + return elem; | ||
306 | + } | ||
307 | + | ||
308 | if (vduse_queue_empty(vq)) { | ||
309 | return NULL; | ||
310 | } | ||
311 | @@ -XXX,XX +XXX,XX @@ void *vduse_queue_pop(VduseVirtq *vq, size_t sz) | ||
312 | |||
313 | vq->inuse++; | ||
314 | |||
315 | + vduse_queue_inflight_get(vq, head); | ||
316 | + | ||
317 | return elem; | ||
318 | } | ||
319 | |||
320 | @@ -XXX,XX +XXX,XX @@ void vduse_queue_push(VduseVirtq *vq, const VduseVirtqElement *elem, | ||
321 | unsigned int len) | ||
322 | { | ||
323 | vduse_queue_fill(vq, elem, len, 0); | ||
324 | + vduse_queue_inflight_pre_put(vq, elem->index); | ||
325 | vduse_queue_flush(vq, 1); | ||
326 | + vduse_queue_inflight_post_put(vq, elem->index); | ||
327 | } | ||
328 | |||
329 | static int vduse_queue_update_vring(VduseVirtq *vq, uint64_t desc_addr, | ||
330 | @@ -XXX,XX +XXX,XX @@ static void vduse_queue_enable(VduseVirtq *vq) | ||
331 | } | ||
332 | |||
333 | vq->fd = fd; | ||
334 | - vq->shadow_avail_idx = vq->last_avail_idx = vq_info.split.avail_index; | ||
335 | - vq->inuse = 0; | ||
336 | - vq->used_idx = 0; | ||
337 | vq->signalled_used_valid = false; | ||
338 | vq->ready = true; | ||
339 | |||
340 | + if (vduse_queue_check_inflights(vq)) { | ||
341 | + fprintf(stderr, "Failed to check inflights for vq[%d]\n", vq->index); | ||
342 | + close(fd); | ||
343 | + return; | ||
344 | + } | ||
345 | + | ||
346 | dev->ops->enable_queue(dev, vq); | ||
347 | } | ||
348 | |||
349 | @@ -XXX,XX +XXX,XX @@ static void vduse_dev_start_dataplane(VduseDev *dev) | ||
350 | |||
351 | static void vduse_dev_stop_dataplane(VduseDev *dev) | ||
352 | { | ||
353 | + size_t log_size = dev->num_queues * vduse_vq_log_size(VIRTQUEUE_MAX_SIZE); | ||
354 | int i; | ||
355 | |||
356 | for (i = 0; i < dev->num_queues; i++) { | ||
357 | vduse_queue_disable(&dev->vqs[i]); | ||
358 | } | ||
359 | + if (dev->log) { | ||
360 | + memset(dev->log, 0, log_size); | ||
361 | + } | ||
362 | dev->features = 0; | ||
363 | vduse_iova_remove_region(dev, 0, ULONG_MAX); | ||
364 | } | ||
365 | @@ -XXX,XX +XXX,XX @@ int vduse_dev_setup_queue(VduseDev *dev, int index, int max_size) | ||
366 | return -errno; | ||
367 | } | ||
368 | |||
369 | + vduse_queue_enable(vq); | ||
370 | + | ||
371 | + return 0; | ||
372 | +} | ||
373 | + | ||
374 | +int vduse_set_reconnect_log_file(VduseDev *dev, const char *filename) | ||
375 | +{ | ||
376 | + | ||
377 | + size_t log_size = dev->num_queues * vduse_vq_log_size(VIRTQUEUE_MAX_SIZE); | ||
378 | + void *log; | ||
379 | + int i; | ||
380 | + | ||
381 | + dev->log = log = vduse_log_get(filename, log_size); | ||
382 | + if (log == MAP_FAILED) { | ||
383 | + fprintf(stderr, "Failed to get vduse log\n"); | ||
384 | + return -EINVAL; | ||
385 | + } | ||
386 | + | ||
387 | + for (i = 0; i < dev->num_queues; i++) { | ||
388 | + dev->vqs[i].log = log; | ||
389 | + dev->vqs[i].log->inflight.desc_num = VIRTQUEUE_MAX_SIZE; | ||
390 | + log = (void *)((char *)log + vduse_vq_log_size(VIRTQUEUE_MAX_SIZE)); | ||
391 | + } | ||
392 | + | ||
393 | return 0; | ||
394 | } | ||
395 | |||
396 | @@ -XXX,XX +XXX,XX @@ static int vduse_dev_init(VduseDev *dev, const char *name, | ||
397 | return -errno; | ||
398 | } | ||
399 | |||
400 | + if (ioctl(fd, VDUSE_DEV_GET_FEATURES, &dev->features)) { | ||
401 | + fprintf(stderr, "Failed to get features: %s\n", strerror(errno)); | ||
402 | + close(fd); | ||
403 | + return -errno; | ||
404 | + } | ||
405 | + | ||
406 | dev_name = strdup(name); | ||
407 | if (!dev_name) { | ||
408 | close(fd); | ||
409 | @@ -XXX,XX +XXX,XX @@ VduseDev *vduse_dev_create_by_fd(int fd, uint16_t num_queues, | ||
410 | return NULL; | ||
411 | } | ||
412 | |||
413 | + if (ioctl(fd, VDUSE_DEV_GET_FEATURES, &dev->features)) { | ||
414 | + fprintf(stderr, "Failed to get features: %s\n", strerror(errno)); | ||
415 | + free(dev); | ||
416 | + return NULL; | ||
417 | + } | ||
418 | + | ||
419 | ret = vduse_dev_init_vqs(dev, num_queues); | ||
420 | if (ret) { | ||
421 | fprintf(stderr, "Failed to init vqs\n"); | ||
422 | @@ -XXX,XX +XXX,XX @@ VduseDev *vduse_dev_create(const char *name, uint32_t device_id, | ||
423 | |||
424 | ret = ioctl(ctrl_fd, VDUSE_CREATE_DEV, dev_config); | ||
425 | free(dev_config); | ||
426 | - if (ret < 0) { | ||
427 | + if (ret && errno != EEXIST) { | ||
428 | fprintf(stderr, "Failed to create vduse device %s: %s\n", | ||
429 | name, strerror(errno)); | ||
430 | goto err_dev; | ||
431 | @@ -XXX,XX +XXX,XX @@ err_ctrl: | ||
432 | |||
433 | int vduse_dev_destroy(VduseDev *dev) | ||
434 | { | ||
435 | - int ret = 0; | ||
436 | + size_t log_size = dev->num_queues * vduse_vq_log_size(VIRTQUEUE_MAX_SIZE); | ||
437 | + int i, ret = 0; | ||
438 | |||
439 | + if (dev->log) { | ||
440 | + munmap(dev->log, log_size); | ||
441 | + } | ||
442 | + for (i = 0; i < dev->num_queues; i++) { | ||
443 | + free(dev->vqs[i].resubmit_list); | ||
444 | + } | ||
445 | free(dev->vqs); | ||
446 | if (dev->fd > 0) { | ||
447 | close(dev->fd); | ||
48 | -- | 448 | -- |
49 | 2.20.1 | 449 | 2.35.3 |
50 | |||
51 | diff view generated by jsdifflib |
1 | This patch adds test cases for attaching the backing chain to a mirror | 1 | From: Stefan Hajnoczi <stefanha@redhat.com> |
---|---|---|---|
2 | job target right before finalising the job, where the image is in a | ||
3 | non-mainloop AioContext (i.e. the backing chain needs to be moved to the | ||
4 | AioContext of the mirror target). | ||
5 | 2 | ||
6 | This requires switching the test case from virtio-blk to virtio-scsi | 3 | Document vduse-blk exports in qemu-storage-daemon --help and the |
7 | because virtio-blk only actually starts using the iothreads when the | 4 | qemu-storage-daemon(1) man page. |
8 | guest driver initialises the device (which never happens in a test case | ||
9 | without a guest OS). virtio-scsi always keeps its block nodes in the | ||
10 | AioContext of the the requested iothread without guest interaction. | ||
11 | 5 | ||
12 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> | 6 | Based-on: <20220523084611.91-1-xieyongji@bytedance.com> |
13 | Message-Id: <20200310113831.27293-7-kwolf@redhat.com> | 7 | Cc: Xie Yongji <xieyongji@bytedance.com> |
14 | Reviewed-by: Peter Krempa <pkrempa@redhat.com> | 8 | Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com> |
9 | Message-Id: <20220525121947.859820-1-stefanha@redhat.com> | ||
15 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> | 10 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> |
16 | --- | 11 | --- |
17 | tests/qemu-iotests/155 | 32 +++++++++++++++++++++++--------- | 12 | docs/tools/qemu-storage-daemon.rst | 21 +++++++++++++++++++++ |
18 | tests/qemu-iotests/155.out | 4 ++-- | 13 | storage-daemon/qemu-storage-daemon.c | 9 +++++++++ |
19 | 2 files changed, 25 insertions(+), 11 deletions(-) | 14 | 2 files changed, 30 insertions(+) |
20 | 15 | ||
21 | diff --git a/tests/qemu-iotests/155 b/tests/qemu-iotests/155 | 16 | diff --git a/docs/tools/qemu-storage-daemon.rst b/docs/tools/qemu-storage-daemon.rst |
22 | index XXXXXXX..XXXXXXX 100755 | 17 | index XXXXXXX..XXXXXXX 100644 |
23 | --- a/tests/qemu-iotests/155 | 18 | --- a/docs/tools/qemu-storage-daemon.rst |
24 | +++ b/tests/qemu-iotests/155 | 19 | +++ b/docs/tools/qemu-storage-daemon.rst |
25 | @@ -XXX,XX +XXX,XX @@ target_img = os.path.join(iotests.test_dir, 'target.' + iotests.imgfmt) | 20 | @@ -XXX,XX +XXX,XX @@ Standard options: |
26 | # chain opened right away. If False, blockdev-add | 21 | --export [type=]vhost-user-blk,id=<id>,node-name=<node-name>,addr.type=unix,addr.path=<socket-path>[,writable=on|off][,logical-block-size=<block-size>][,num-queues=<num-queues>] |
27 | # opens it without a backing file and job completion | 22 | --export [type=]vhost-user-blk,id=<id>,node-name=<node-name>,addr.type=fd,addr.str=<fd>[,writable=on|off][,logical-block-size=<block-size>][,num-queues=<num-queues>] |
28 | # is supposed to open the backing chain. | 23 | --export [type=]fuse,id=<id>,node-name=<node-name>,mountpoint=<file>[,growable=on|off][,writable=on|off][,allow-other=on|off|auto] |
29 | +# use_iothread: If True, an iothread is configured for the virtio-blk device | 24 | + --export [type=]vduse-blk,id=<id>,node-name=<node-name>[,writable=on|off][,num-queues=<num-queues>][,queue-size=<queue-size>][,logical-block-size=<block-size>] |
30 | +# that uses the image being mirrored | 25 | |
31 | 26 | is a block export definition. ``node-name`` is the block node that should be | |
32 | class BaseClass(iotests.QMPTestCase): | 27 | exported. ``writable`` determines whether or not the export allows write |
33 | target_blockdev_backing = None | 28 | @@ -XXX,XX +XXX,XX @@ Standard options: |
34 | target_real_backing = None | 29 | ``allow-other`` to auto (the default) will try enabling this option, and on |
35 | target_open_with_backing = True | 30 | error fall back to disabling it. |
36 | + use_iothread = False | 31 | |
37 | 32 | + The ``vduse-blk`` export type uses the ``id`` as the VDUSE device name. | |
38 | def setUp(self): | 33 | + ``num-queues`` sets the number of virtqueues (the default is 1). |
39 | qemu_img('create', '-f', iotests.imgfmt, back0_img, '1440K') | 34 | + ``queue-size`` sets the virtqueue descriptor table size (the default is 256). |
40 | @@ -XXX,XX +XXX,XX @@ class BaseClass(iotests.QMPTestCase): | ||
41 | 'file': {'driver': 'file', | ||
42 | 'filename': source_img}} | ||
43 | self.vm.add_blockdev(self.vm.qmp_to_opts(blockdev)) | ||
44 | - self.vm.add_device('virtio-blk,id=qdev0,drive=source') | ||
45 | + | 35 | + |
46 | + if self.use_iothread: | 36 | + The instantiated VDUSE device must then be added to the vDPA bus using the |
47 | + self.vm.add_object('iothread,id=iothread0') | 37 | + vdpa(8) command from the iproute2 project:: |
48 | + iothread = ",iothread=iothread0" | ||
49 | + else: | ||
50 | + iothread = "" | ||
51 | + | 38 | + |
52 | + self.vm.add_device('virtio-scsi%s' % iothread) | 39 | + # vdpa dev add name <id> mgmtdev vduse |
53 | + self.vm.add_device('scsi-hd,id=qdev0,drive=source') | ||
54 | + | 40 | + |
55 | self.vm.launch() | 41 | + The device can be removed from the vDPA bus later as follows:: |
56 | |||
57 | self.assertIntactSourceBackingChain() | ||
58 | @@ -XXX,XX +XXX,XX @@ class MirrorBaseClass(BaseClass): | ||
59 | def testFull(self): | ||
60 | self.runMirror('full') | ||
61 | |||
62 | - node = self.findBlockNode('target', | ||
63 | - '/machine/peripheral/qdev0/virtio-backend') | ||
64 | + node = self.findBlockNode('target', 'qdev0') | ||
65 | self.assertCorrectBackingImage(node, None) | ||
66 | self.assertIntactSourceBackingChain() | ||
67 | |||
68 | def testTop(self): | ||
69 | self.runMirror('top') | ||
70 | |||
71 | - node = self.findBlockNode('target', | ||
72 | - '/machine/peripheral/qdev0/virtio-backend') | ||
73 | + node = self.findBlockNode('target', 'qdev0') | ||
74 | self.assertCorrectBackingImage(node, back2_img) | ||
75 | self.assertIntactSourceBackingChain() | ||
76 | |||
77 | def testNone(self): | ||
78 | self.runMirror('none') | ||
79 | |||
80 | - node = self.findBlockNode('target', | ||
81 | - '/machine/peripheral/qdev0/virtio-backend') | ||
82 | + node = self.findBlockNode('target', 'qdev0') | ||
83 | self.assertCorrectBackingImage(node, source_img) | ||
84 | self.assertIntactSourceBackingChain() | ||
85 | |||
86 | @@ -XXX,XX +XXX,XX @@ class TestBlockdevMirrorReopen(MirrorBaseClass): | ||
87 | backing="backing") | ||
88 | self.assert_qmp(result, 'return', {}) | ||
89 | |||
90 | +class TestBlockdevMirrorReopenIothread(TestBlockdevMirrorReopen): | ||
91 | + use_iothread = True | ||
92 | + | 42 | + |
93 | # Attach the backing chain only during completion, with blockdev-snapshot | 43 | + # vdpa dev del <id> |
94 | class TestBlockdevMirrorSnapshot(MirrorBaseClass): | ||
95 | cmd = 'blockdev-mirror' | ||
96 | @@ -XXX,XX +XXX,XX @@ class TestBlockdevMirrorSnapshot(MirrorBaseClass): | ||
97 | overlay="target") | ||
98 | self.assert_qmp(result, 'return', {}) | ||
99 | |||
100 | +class TestBlockdevMirrorSnapshotIothread(TestBlockdevMirrorSnapshot): | ||
101 | + use_iothread = True | ||
102 | + | 44 | + |
103 | class TestCommit(BaseClass): | 45 | + For more information about attaching vDPA devices to the host with |
104 | existing = False | 46 | + virtio_vdpa.ko or attaching them to guests with vhost_vdpa.ko, see |
105 | 47 | + https://vdpa-dev.gitlab.io/. | |
106 | @@ -XXX,XX +XXX,XX @@ class TestCommit(BaseClass): | 48 | + |
107 | 49 | + For more information about VDUSE, see | |
108 | self.vm.event_wait('BLOCK_JOB_COMPLETED') | 50 | + https://docs.kernel.org/userspace-api/vduse.html. |
109 | 51 | + | |
110 | - node = self.findBlockNode(None, | 52 | .. option:: --monitor MONITORDEF |
111 | - '/machine/peripheral/qdev0/virtio-backend') | 53 | |
112 | + node = self.findBlockNode(None, 'qdev0') | 54 | is a QMP monitor definition. See the :manpage:`qemu(1)` manual page for |
113 | self.assert_qmp(node, 'image' + '/backing-image' * 0 + '/filename', | 55 | diff --git a/storage-daemon/qemu-storage-daemon.c b/storage-daemon/qemu-storage-daemon.c |
114 | back1_img) | ||
115 | self.assert_qmp(node, 'image' + '/backing-image' * 1 + '/filename', | ||
116 | diff --git a/tests/qemu-iotests/155.out b/tests/qemu-iotests/155.out | ||
117 | index XXXXXXX..XXXXXXX 100644 | 56 | index XXXXXXX..XXXXXXX 100644 |
118 | --- a/tests/qemu-iotests/155.out | 57 | --- a/storage-daemon/qemu-storage-daemon.c |
119 | +++ b/tests/qemu-iotests/155.out | 58 | +++ b/storage-daemon/qemu-storage-daemon.c |
120 | @@ -XXX,XX +XXX,XX @@ | 59 | @@ -XXX,XX +XXX,XX @@ static void help(void) |
121 | -......................... | 60 | " vhost-user-blk device over file descriptor\n" |
122 | +............................... | 61 | "\n" |
123 | ---------------------------------------------------------------------- | 62 | #endif /* CONFIG_VHOST_USER_BLK_SERVER */ |
124 | -Ran 25 tests | 63 | +#ifdef CONFIG_VDUSE_BLK_EXPORT |
125 | +Ran 31 tests | 64 | +" --export [type=]vduse-blk,id=<id>,node-name=<node-name>\n" |
126 | 65 | +" [,writable=on|off][,num-queues=<num-queues>]\n" | |
127 | OK | 66 | +" [,queue-size=<queue-size>]\n" |
67 | +" [,logical-block-size=<logical-block-size>]\n" | ||
68 | +" export the specified block node as a vduse-blk\n" | ||
69 | +" device using the id as the VDUSE device name\n" | ||
70 | +"\n" | ||
71 | +#endif /* CONFIG_VDUSE_BLK_EXPORT */ | ||
72 | " --monitor [chardev=]name[,mode=control][,pretty[=on|off]]\n" | ||
73 | " configure a QMP monitor\n" | ||
74 | "\n" | ||
128 | -- | 75 | -- |
129 | 2.20.1 | 76 | 2.35.3 |
130 | |||
131 | diff view generated by jsdifflib |
1 | From: Daniel Henrique Barboza <danielhb413@gmail.com> | 1 | From: Stefano Garzarella <sgarzare@redhat.com> |
---|---|---|---|
2 | 2 | ||
3 | When using a non-UTF8 secret to create a volume using qemu-img, the | 3 | If the namespace does not exist, rbd_create() fails with -ENOENT and |
4 | following error happens: | 4 | QEMU reports a generic "error rbd create: No such file or directory": |
5 | 5 | ||
6 | $ qemu-img create -f luks --object secret,id=vol_1_encrypt0,file=vol_resize_pool.vol_1.secret.qzVQrI -o key-secret=vol_1_encrypt0 /var/tmp/pool_target/vol_1 10240K | 6 | $ qemu-img create rbd:rbd/namespace/image 1M |
7 | Formatting 'rbd:rbd/namespace/image', fmt=raw size=1048576 | ||
8 | qemu-img: rbd:rbd/namespace/image: error rbd create: No such file or directory | ||
7 | 9 | ||
8 | Formatting '/var/tmp/pool_target/vol_1', fmt=luks size=10485760 key-secret=vol_1_encrypt0 | 10 | Unfortunately rados_ioctx_set_namespace() does not fail if the namespace |
9 | qemu-img: /var/tmp/pool_target/vol_1: Data from secret vol_1_encrypt0 is not valid UTF-8 | 11 | does not exist, so let's use rbd_namespace_exists() in qemu_rbd_connect() |
12 | to check if the namespace exists, reporting a more understandable error: | ||
10 | 13 | ||
11 | However, the created file '/var/tmp/pool_target/vol_1' is left behind in the | 14 | $ qemu-img create rbd:rbd/namespace/image 1M |
12 | file system after the failure. This behavior can be observed when creating | 15 | Formatting 'rbd:rbd/namespace/image', fmt=raw size=1048576 |
13 | the volume using Libvirt, via 'virsh vol-create', and then getting "volume | 16 | qemu-img: rbd:rbd/namespace/image: namespace 'namespace' does not exist |
14 | target path already exist" errors when trying to re-create the volume. | ||
15 | 17 | ||
16 | The volume file is created inside block_crypto_co_create_opts_luks(), in | 18 | Reported-by: Tingting Mao <timao@redhat.com> |
17 | block/crypto.c. If the bdrv_create_file() call is successful but any | 19 | Reviewed-by: Ilya Dryomov <idryomov@gmail.com> |
18 | succeeding step fails*, the existing 'fail' label does not take into | 20 | Signed-off-by: Stefano Garzarella <sgarzare@redhat.com> |
19 | account the created file, leaving it behind. | 21 | Message-Id: <20220517071012.6120-1-sgarzare@redhat.com> |
20 | |||
21 | This patch changes block_crypto_co_create_opts_luks() to delete | ||
22 | 'filename' in case of failure. A failure in this point means that | ||
23 | the volume is now truncated/corrupted, so even if 'filename' was an | ||
24 | existing volume before calling qemu-img, it is now unusable. Deleting | ||
25 | the file it is not much worse than leaving it in the filesystem in | ||
26 | this scenario, and we don't have to deal with checking the file | ||
27 | pre-existence in the code. | ||
28 | |||
29 | * in our case, block_crypto_co_create_generic calls qcrypto_block_create, | ||
30 | which calls qcrypto_block_luks_create, and this function fails when | ||
31 | calling qcrypto_secret_lookup_as_utf8. | ||
32 | |||
33 | Reported-by: Srikanth Aithal <bssrikanth@in.ibm.com> | ||
34 | Suggested-by: Kevin Wolf <kwolf@redhat.com> | ||
35 | Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com> | ||
36 | Message-Id: <20200130213907.2830642-4-danielhb413@gmail.com> | ||
37 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> | 22 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> |
38 | --- | 23 | --- |
39 | block/crypto.c | 18 ++++++++++++++++++ | 24 | block/rbd.c | 24 ++++++++++++++++++++++++ |
40 | 1 file changed, 18 insertions(+) | 25 | meson.build | 6 ++++++ |
26 | 2 files changed, 30 insertions(+) | ||
41 | 27 | ||
42 | diff --git a/block/crypto.c b/block/crypto.c | 28 | diff --git a/block/rbd.c b/block/rbd.c |
43 | index XXXXXXX..XXXXXXX 100644 | 29 | index XXXXXXX..XXXXXXX 100644 |
44 | --- a/block/crypto.c | 30 | --- a/block/rbd.c |
45 | +++ b/block/crypto.c | 31 | +++ b/block/rbd.c |
46 | @@ -XXX,XX +XXX,XX @@ | 32 | @@ -XXX,XX +XXX,XX @@ static int qemu_rbd_connect(rados_t *cluster, rados_ioctx_t *io_ctx, |
47 | #include "qapi/error.h" | 33 | error_setg_errno(errp, -r, "error opening pool %s", opts->pool); |
48 | #include "qemu/module.h" | 34 | goto failed_shutdown; |
49 | #include "qemu/option.h" | 35 | } |
50 | +#include "qemu/cutils.h" | 36 | + |
51 | #include "crypto.h" | 37 | +#ifdef HAVE_RBD_NAMESPACE_EXISTS |
52 | 38 | + if (opts->has_q_namespace && strlen(opts->q_namespace) > 0) { | |
53 | typedef struct BlockCrypto BlockCrypto; | 39 | + bool exists; |
54 | @@ -XXX,XX +XXX,XX @@ static int coroutine_fn block_crypto_co_create_opts_luks(const char *filename, | 40 | + |
55 | 41 | + r = rbd_namespace_exists(*io_ctx, opts->q_namespace, &exists); | |
56 | ret = 0; | 42 | + if (r < 0) { |
57 | fail: | 43 | + error_setg_errno(errp, -r, "error checking namespace"); |
58 | + /* | 44 | + goto failed_ioctx_destroy; |
59 | + * If an error occurred, delete 'filename'. Even if the file existed | 45 | + } |
60 | + * beforehand, it has been truncated and corrupted in the process. | 46 | + |
61 | + */ | 47 | + if (!exists) { |
62 | + if (ret && bs) { | 48 | + error_setg(errp, "namespace '%s' does not exist", |
63 | + Error *local_delete_err = NULL; | 49 | + opts->q_namespace); |
64 | + int r_del = bdrv_co_delete_file(bs, &local_delete_err); | 50 | + r = -ENOENT; |
65 | + /* | 51 | + goto failed_ioctx_destroy; |
66 | + * ENOTSUP will happen if the block driver doesn't support | ||
67 | + * the 'bdrv_co_delete_file' interface. This is a predictable | ||
68 | + * scenario and shouldn't be reported back to the user. | ||
69 | + */ | ||
70 | + if ((r_del < 0) && (r_del != -ENOTSUP)) { | ||
71 | + error_report_err(local_delete_err); | ||
72 | + } | 52 | + } |
73 | + } | 53 | + } |
54 | +#endif | ||
74 | + | 55 | + |
75 | bdrv_unref(bs); | 56 | /* |
76 | qapi_free_QCryptoBlockCreateOptions(create_opts); | 57 | * Set the namespace after opening the io context on the pool, |
77 | qobject_unref(cryptoopts); | 58 | * if nspace == NULL or if nspace == "", it is just as we did nothing |
59 | @@ -XXX,XX +XXX,XX @@ static int qemu_rbd_connect(rados_t *cluster, rados_ioctx_t *io_ctx, | ||
60 | r = 0; | ||
61 | goto out; | ||
62 | |||
63 | +#ifdef HAVE_RBD_NAMESPACE_EXISTS | ||
64 | +failed_ioctx_destroy: | ||
65 | + rados_ioctx_destroy(*io_ctx); | ||
66 | +#endif | ||
67 | failed_shutdown: | ||
68 | rados_shutdown(*cluster); | ||
69 | out: | ||
70 | diff --git a/meson.build b/meson.build | ||
71 | index XXXXXXX..XXXXXXX 100644 | ||
72 | --- a/meson.build | ||
73 | +++ b/meson.build | ||
74 | @@ -XXX,XX +XXX,XX @@ config_host_data.set('HAVE_GETIFADDRS', cc.has_function('getifaddrs')) | ||
75 | config_host_data.set('HAVE_OPENPTY', cc.has_function('openpty', dependencies: util)) | ||
76 | config_host_data.set('HAVE_STRCHRNUL', cc.has_function('strchrnul')) | ||
77 | config_host_data.set('HAVE_SYSTEM_FUNCTION', cc.has_function('system', prefix: '#include <stdlib.h>')) | ||
78 | +if rbd.found() | ||
79 | + config_host_data.set('HAVE_RBD_NAMESPACE_EXISTS', | ||
80 | + cc.has_function('rbd_namespace_exists', | ||
81 | + dependencies: rbd, | ||
82 | + prefix: '#include <rbd/librbd.h>')) | ||
83 | +endif | ||
84 | if rdma.found() | ||
85 | config_host_data.set('HAVE_IBV_ADVISE_MR', | ||
86 | cc.has_function('ibv_advise_mr', | ||
78 | -- | 87 | -- |
79 | 2.20.1 | 88 | 2.35.3 |
80 | |||
81 | diff view generated by jsdifflib |
1 | From: Daniel Henrique Barboza <danielhb413@gmail.com> | 1 | From: Fabian Ebner <f.ebner@proxmox.com> |
---|---|---|---|
2 | 2 | ||
3 | Adding to Block Drivers the capability of being able to clean up | 3 | On 64-bit platforms, assigning SIZE_MAX to the int64_t max_pdiscard |
4 | its created files can be useful in certain situations. For the | 4 | results in a negative value, and the following assertion would trigger |
5 | LUKS driver, for instance, a failure in one of its authentication | 5 | down the line (it's not the same max_pdiscard, but computed from the |
6 | steps can leave files in the host that weren't there before. | 6 | other one): |
7 | qemu-system-x86_64: ../block/io.c:3166: bdrv_co_pdiscard: Assertion | ||
8 | `max_pdiscard >= bs->bl.request_alignment' failed. | ||
7 | 9 | ||
8 | This patch adds the 'bdrv_co_delete_file' interface to block | 10 | On 32-bit platforms, it's fine to keep using SIZE_MAX. |
9 | drivers and add it to the 'file' driver in file-posix.c. The | ||
10 | implementation is given by 'raw_co_delete_file'. | ||
11 | 11 | ||
12 | Suggested-by: Daniel P. Berrangé <berrange@redhat.com> | 12 | The assertion in qemu_gluster_co_pdiscard() is checking that the value |
13 | Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com> | 13 | of 'bytes' can safely be passed to glfs_discard_async(), which takes a |
14 | Message-Id: <20200130213907.2830642-2-danielhb413@gmail.com> | 14 | size_t for the argument in question, so it is kept as is. And since |
15 | max_pdiscard is still <= SIZE_MAX, relying on max_pdiscard is still | ||
16 | fine. | ||
17 | |||
18 | Fixes: 0c8022876f ("block: use int64_t instead of int in driver discard handlers") | ||
19 | Cc: qemu-stable@nongnu.org | ||
20 | Signed-off-by: Fabian Ebner <f.ebner@proxmox.com> | ||
21 | Message-Id: <20220520075922.43972-1-f.ebner@proxmox.com> | ||
22 | Reviewed-by: Eric Blake <eblake@redhat.com> | ||
23 | Reviewed-by: Stefano Garzarella <sgarzare@redhat.com> | ||
24 | Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru> | ||
15 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> | 25 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> |
16 | --- | 26 | --- |
17 | include/block/block_int.h | 4 ++++ | 27 | block/gluster.c | 2 +- |
18 | block/file-posix.c | 23 +++++++++++++++++++++++ | 28 | 1 file changed, 1 insertion(+), 1 deletion(-) |
19 | 2 files changed, 27 insertions(+) | ||
20 | 29 | ||
21 | diff --git a/include/block/block_int.h b/include/block/block_int.h | 30 | diff --git a/block/gluster.c b/block/gluster.c |
22 | index XXXXXXX..XXXXXXX 100644 | 31 | index XXXXXXX..XXXXXXX 100644 |
23 | --- a/include/block/block_int.h | 32 | --- a/block/gluster.c |
24 | +++ b/include/block/block_int.h | 33 | +++ b/block/gluster.c |
25 | @@ -XXX,XX +XXX,XX @@ struct BlockDriver { | 34 | @@ -XXX,XX +XXX,XX @@ out: |
26 | */ | 35 | static void qemu_gluster_refresh_limits(BlockDriverState *bs, Error **errp) |
27 | int coroutine_fn (*bdrv_co_flush)(BlockDriverState *bs); | 36 | { |
28 | 37 | bs->bl.max_transfer = GLUSTER_MAX_TRANSFER; | |
29 | + /* Delete a created file. */ | 38 | - bs->bl.max_pdiscard = SIZE_MAX; |
30 | + int coroutine_fn (*bdrv_co_delete_file)(BlockDriverState *bs, | 39 | + bs->bl.max_pdiscard = MIN(SIZE_MAX, INT64_MAX); |
31 | + Error **errp); | ||
32 | + | ||
33 | /* | ||
34 | * Flushes all data that was already written to the OS all the way down to | ||
35 | * the disk (for example file-posix.c calls fsync()). | ||
36 | diff --git a/block/file-posix.c b/block/file-posix.c | ||
37 | index XXXXXXX..XXXXXXX 100644 | ||
38 | --- a/block/file-posix.c | ||
39 | +++ b/block/file-posix.c | ||
40 | @@ -XXX,XX +XXX,XX @@ static int coroutine_fn raw_co_create_opts(const char *filename, QemuOpts *opts, | ||
41 | return raw_co_create(&options, errp); | ||
42 | } | 40 | } |
43 | 41 | ||
44 | +static int coroutine_fn raw_co_delete_file(BlockDriverState *bs, | 42 | static int qemu_gluster_reopen_prepare(BDRVReopenState *state, |
45 | + Error **errp) | ||
46 | +{ | ||
47 | + struct stat st; | ||
48 | + int ret; | ||
49 | + | ||
50 | + if (!(stat(bs->filename, &st) == 0) || !S_ISREG(st.st_mode)) { | ||
51 | + error_setg_errno(errp, ENOENT, "%s is not a regular file", | ||
52 | + bs->filename); | ||
53 | + return -ENOENT; | ||
54 | + } | ||
55 | + | ||
56 | + ret = unlink(bs->filename); | ||
57 | + if (ret < 0) { | ||
58 | + ret = -errno; | ||
59 | + error_setg_errno(errp, -ret, "Error when deleting file %s", | ||
60 | + bs->filename); | ||
61 | + } | ||
62 | + | ||
63 | + return ret; | ||
64 | +} | ||
65 | + | ||
66 | /* | ||
67 | * Find allocation range in @bs around offset @start. | ||
68 | * May change underlying file descriptor's file offset. | ||
69 | @@ -XXX,XX +XXX,XX @@ BlockDriver bdrv_file = { | ||
70 | .bdrv_co_block_status = raw_co_block_status, | ||
71 | .bdrv_co_invalidate_cache = raw_co_invalidate_cache, | ||
72 | .bdrv_co_pwrite_zeroes = raw_co_pwrite_zeroes, | ||
73 | + .bdrv_co_delete_file = raw_co_delete_file, | ||
74 | |||
75 | .bdrv_co_preadv = raw_co_preadv, | ||
76 | .bdrv_co_pwritev = raw_co_pwritev, | ||
77 | -- | 43 | -- |
78 | 2.20.1 | 44 | 2.35.3 |
79 | |||
80 | diff view generated by jsdifflib |
1 | The newly tested scenario is a common live storage migration scenario: | 1 | From: Emanuele Giuseppe Esposito <eesposit@redhat.com> |
---|---|---|---|
2 | The target node is opened without a backing file so that the active | ||
3 | layer is mirrored while its backing chain can be copied in the | ||
4 | background. | ||
5 | 2 | ||
6 | The backing chain should be attached to the mirror target node when | 3 | It seems that aio_wait_kick always required a memory barrier |
7 | finalising the job, just before switching the users of the source node | 4 | or atomic operation in the caller, but nobody actually |
8 | to the new copy (at which point the mirror job still has a reference to | 5 | took care of doing it. |
9 | the node). drive-mirror did this automatically, but with blockdev-mirror | ||
10 | this is the job of the QMP client. | ||
11 | 6 | ||
12 | This patch adds test cases for two ways to achieve the desired result, | 7 | Let's put the barrier in the function instead, and pair it |
13 | using either x-blockdev-reopen or blockdev-snapshot. | 8 | with another one in AIO_WAIT_WHILE. Read aio_wait_kick() |
9 | comment for further explanation. | ||
14 | 10 | ||
15 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> | 11 | Suggested-by: Paolo Bonzini <pbonzini@redhat.com> |
16 | Message-Id: <20200310113831.27293-5-kwolf@redhat.com> | 12 | Signed-off-by: Emanuele Giuseppe Esposito <eesposit@redhat.com> |
17 | Reviewed-by: Peter Krempa <pkrempa@redhat.com> | 13 | Message-Id: <20220524173054.12651-1-eesposit@redhat.com> |
14 | Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru> | ||
15 | Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> | ||
18 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> | 16 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> |
19 | --- | 17 | --- |
20 | tests/qemu-iotests/155 | 56 ++++++++++++++++++++++++++++++++++---- | 18 | include/block/aio-wait.h | 2 ++ |
21 | tests/qemu-iotests/155.out | 4 +-- | 19 | util/aio-wait.c | 16 +++++++++++++++- |
22 | 2 files changed, 53 insertions(+), 7 deletions(-) | 20 | 2 files changed, 17 insertions(+), 1 deletion(-) |
23 | 21 | ||
24 | diff --git a/tests/qemu-iotests/155 b/tests/qemu-iotests/155 | 22 | diff --git a/include/block/aio-wait.h b/include/block/aio-wait.h |
25 | index XXXXXXX..XXXXXXX 100755 | 23 | index XXXXXXX..XXXXXXX 100644 |
26 | --- a/tests/qemu-iotests/155 | 24 | --- a/include/block/aio-wait.h |
27 | +++ b/tests/qemu-iotests/155 | 25 | +++ b/include/block/aio-wait.h |
28 | @@ -XXX,XX +XXX,XX @@ target_img = os.path.join(iotests.test_dir, 'target.' + iotests.imgfmt) | 26 | @@ -XXX,XX +XXX,XX @@ extern AioWait global_aio_wait; |
29 | # image during runtime, only makes sense if | 27 | AioContext *ctx_ = (ctx); \ |
30 | # target_blockdev_backing is not None | 28 | /* Increment wait_->num_waiters before evaluating cond. */ \ |
31 | # (None: same as target_backing) | 29 | qatomic_inc(&wait_->num_waiters); \ |
32 | +# target_open_with_backing: If True, the target image is added with its backing | 30 | + /* Paired with smp_mb in aio_wait_kick(). */ \ |
33 | +# chain opened right away. If False, blockdev-add | 31 | + smp_mb(); \ |
34 | +# opens it without a backing file and job completion | 32 | if (ctx_ && in_aio_context_home_thread(ctx_)) { \ |
35 | +# is supposed to open the backing chain. | 33 | while ((cond)) { \ |
36 | 34 | aio_poll(ctx_, true); \ | |
37 | class BaseClass(iotests.QMPTestCase): | 35 | diff --git a/util/aio-wait.c b/util/aio-wait.c |
38 | target_blockdev_backing = None | 36 | index XXXXXXX..XXXXXXX 100644 |
39 | target_real_backing = None | 37 | --- a/util/aio-wait.c |
40 | + target_open_with_backing = True | 38 | +++ b/util/aio-wait.c |
41 | 39 | @@ -XXX,XX +XXX,XX @@ static void dummy_bh_cb(void *opaque) | |
42 | def setUp(self): | 40 | |
43 | qemu_img('create', '-f', iotests.imgfmt, back0_img, '1440K') | 41 | void aio_wait_kick(void) |
44 | @@ -XXX,XX +XXX,XX @@ class BaseClass(iotests.QMPTestCase): | 42 | { |
45 | options = { 'node-name': 'target', | 43 | - /* The barrier (or an atomic op) is in the caller. */ |
46 | 'driver': iotests.imgfmt, | 44 | + /* |
47 | 'file': { 'driver': 'file', | 45 | + * Paired with smp_mb in AIO_WAIT_WHILE. Here we have: |
48 | + 'node-name': 'target-file', | 46 | + * write(condition); |
49 | 'filename': target_img } } | 47 | + * aio_wait_kick() { |
50 | - if self.target_blockdev_backing: | 48 | + * smp_mb(); |
51 | - options['backing'] = self.target_blockdev_backing | 49 | + * read(num_waiters); |
50 | + * } | ||
51 | + * | ||
52 | + * And in AIO_WAIT_WHILE: | ||
53 | + * write(num_waiters); | ||
54 | + * smp_mb(); | ||
55 | + * read(condition); | ||
56 | + */ | ||
57 | + smp_mb(); | ||
52 | + | 58 | + |
53 | + if not self.target_open_with_backing: | 59 | if (qatomic_read(&global_aio_wait.num_waiters)) { |
54 | + options['backing'] = None | 60 | aio_bh_schedule_oneshot(qemu_get_aio_context(), dummy_bh_cb, NULL); |
55 | + elif self.target_blockdev_backing: | 61 | } |
56 | + options['backing'] = self.target_blockdev_backing | ||
57 | |||
58 | result = self.vm.qmp('blockdev-add', **options) | ||
59 | self.assert_qmp(result, 'return', {}) | ||
60 | @@ -XXX,XX +XXX,XX @@ class BaseClass(iotests.QMPTestCase): | ||
61 | # cmd: Mirroring command to execute, either drive-mirror or blockdev-mirror | ||
62 | |||
63 | class MirrorBaseClass(BaseClass): | ||
64 | + def openBacking(self): | ||
65 | + pass | ||
66 | + | ||
67 | def runMirror(self, sync): | ||
68 | if self.cmd == 'blockdev-mirror': | ||
69 | result = self.vm.qmp(self.cmd, job_id='mirror-job', device='source', | ||
70 | - sync=sync, target='target') | ||
71 | + sync=sync, target='target', | ||
72 | + auto_finalize=False) | ||
73 | else: | ||
74 | if self.existing: | ||
75 | mode = 'existing' | ||
76 | @@ -XXX,XX +XXX,XX @@ class MirrorBaseClass(BaseClass): | ||
77 | result = self.vm.qmp(self.cmd, job_id='mirror-job', device='source', | ||
78 | sync=sync, target=target_img, | ||
79 | format=iotests.imgfmt, mode=mode, | ||
80 | - node_name='target') | ||
81 | + node_name='target', auto_finalize=False) | ||
82 | |||
83 | self.assert_qmp(result, 'return', {}) | ||
84 | |||
85 | - self.complete_and_wait('mirror-job') | ||
86 | + self.vm.run_job('mirror-job', use_log=False, auto_finalize=False, | ||
87 | + pre_finalize=self.openBacking, auto_dismiss=True) | ||
88 | |||
89 | def testFull(self): | ||
90 | self.runMirror('full') | ||
91 | @@ -XXX,XX +XXX,XX @@ class TestBlockdevMirrorForcedBacking(MirrorBaseClass): | ||
92 | target_blockdev_backing = { 'driver': 'null-co' } | ||
93 | target_real_backing = 'null-co://' | ||
94 | |||
95 | +# Attach the backing chain only during completion, with blockdev-reopen | ||
96 | +class TestBlockdevMirrorReopen(MirrorBaseClass): | ||
97 | + cmd = 'blockdev-mirror' | ||
98 | + existing = True | ||
99 | + target_backing = 'null-co://' | ||
100 | + target_open_with_backing = False | ||
101 | + | ||
102 | + def openBacking(self): | ||
103 | + if not self.target_open_with_backing: | ||
104 | + result = self.vm.qmp('blockdev-add', node_name="backing", | ||
105 | + driver="null-co") | ||
106 | + self.assert_qmp(result, 'return', {}) | ||
107 | + result = self.vm.qmp('x-blockdev-reopen', node_name="target", | ||
108 | + driver=iotests.imgfmt, file="target-file", | ||
109 | + backing="backing") | ||
110 | + self.assert_qmp(result, 'return', {}) | ||
111 | + | ||
112 | +# Attach the backing chain only during completion, with blockdev-snapshot | ||
113 | +class TestBlockdevMirrorSnapshot(MirrorBaseClass): | ||
114 | + cmd = 'blockdev-mirror' | ||
115 | + existing = True | ||
116 | + target_backing = 'null-co://' | ||
117 | + target_open_with_backing = False | ||
118 | + | ||
119 | + def openBacking(self): | ||
120 | + if not self.target_open_with_backing: | ||
121 | + result = self.vm.qmp('blockdev-add', node_name="backing", | ||
122 | + driver="null-co") | ||
123 | + self.assert_qmp(result, 'return', {}) | ||
124 | + result = self.vm.qmp('blockdev-snapshot', node="backing", | ||
125 | + overlay="target") | ||
126 | + self.assert_qmp(result, 'return', {}) | ||
127 | |||
128 | class TestCommit(BaseClass): | ||
129 | existing = False | ||
130 | diff --git a/tests/qemu-iotests/155.out b/tests/qemu-iotests/155.out | ||
131 | index XXXXXXX..XXXXXXX 100644 | ||
132 | --- a/tests/qemu-iotests/155.out | ||
133 | +++ b/tests/qemu-iotests/155.out | ||
134 | @@ -XXX,XX +XXX,XX @@ | ||
135 | -................... | ||
136 | +......................... | ||
137 | ---------------------------------------------------------------------- | ||
138 | -Ran 19 tests | ||
139 | +Ran 25 tests | ||
140 | |||
141 | OK | ||
142 | -- | 62 | -- |
143 | 2.20.1 | 63 | 2.35.3 |
144 | |||
145 | diff view generated by jsdifflib |
1 | From: Pan Nengyuan <pannengyuan@huawei.com> | 1 | From: Eric Blake <eblake@redhat.com> |
---|---|---|---|
2 | 2 | ||
3 | 'type/id' forgot to free in qmp_object_add, this patch fix that. | 3 | CID 1488362 points out that the second 'rc >= 0' check is now dead |
4 | code. | ||
4 | 5 | ||
5 | The leak stack: | 6 | Reported-by: Peter Maydell <peter.maydell@linaro.org> |
6 | Direct leak of 84 byte(s) in 6 object(s) allocated from: | 7 | Fixes: 172f5f1a40(nbd: remove peppering of nbd_client_connected) |
7 | #0 0x7fe2a5ebf768 in __interceptor_malloc (/lib64/libasan.so.5+0xef768) | 8 | Signed-off-by: Eric Blake <eblake@redhat.com> |
8 | #1 0x7fe2a5044445 in g_malloc (/lib64/libglib-2.0.so.0+0x52445) | 9 | Message-Id: <20220516210519.76135-1-eblake@redhat.com> |
9 | #2 0x7fe2a505dd92 in g_strdup (/lib64/libglib-2.0.so.0+0x6bd92) | 10 | Reviewed-by: Peter Maydell <peter.maydell@linaro.org> |
10 | #3 0x56344954e692 in qmp_object_add /mnt/sdb/qemu-new/qemu_test/qemu/qom/qom-qmp-cmds.c:258 | 11 | Reviewed-by: Vladimir Sementsov-Ogievskiy <v.sementsov-og@mail.ru> |
11 | #4 0x563449960f5a in do_qmp_dispatch /mnt/sdb/qemu-new/qemu_test/qemu/qapi/qmp-dispatch.c:132 | ||
12 | #5 0x563449960f5a in qmp_dispatch /mnt/sdb/qemu-new/qemu_test/qemu/qapi/qmp-dispatch.c:175 | ||
13 | #6 0x563449498a30 in monitor_qmp_dispatch /mnt/sdb/qemu-new/qemu_test/qemu/monitor/qmp.c:145 | ||
14 | #7 0x56344949a64f in monitor_qmp_bh_dispatcher /mnt/sdb/qemu-new/qemu_test/qemu/monitor/qmp.c:234 | ||
15 | #8 0x563449a92a3a in aio_bh_call /mnt/sdb/qemu-new/qemu_test/qemu/util/async.c:136 | ||
16 | |||
17 | Direct leak of 54 byte(s) in 6 object(s) allocated from: | ||
18 | #0 0x7fe2a5ebf768 in __interceptor_malloc (/lib64/libasan.so.5+0xef768) | ||
19 | #1 0x7fe2a5044445 in g_malloc (/lib64/libglib-2.0.so.0+0x52445) | ||
20 | #2 0x7fe2a505dd92 in g_strdup (/lib64/libglib-2.0.so.0+0x6bd92) | ||
21 | #3 0x56344954e6c4 in qmp_object_add /mnt/sdb/qemu-new/qemu_test/qemu/qom/qom-qmp-cmds.c:267 | ||
22 | #4 0x563449960f5a in do_qmp_dispatch /mnt/sdb/qemu-new/qemu_test/qemu/qapi/qmp-dispatch.c:132 | ||
23 | #5 0x563449960f5a in qmp_dispatch /mnt/sdb/qemu-new/qemu_test/qemu/qapi/qmp-dispatch.c:175 | ||
24 | #6 0x563449498a30 in monitor_qmp_dispatch /mnt/sdb/qemu-new/qemu_test/qemu/monitor/qmp.c:145 | ||
25 | #7 0x56344949a64f in monitor_qmp_bh_dispatcher /mnt/sdb/qemu-new/qemu_test/qemu/monitor/qmp.c:234 | ||
26 | #8 0x563449a92a3a in aio_bh_call /mnt/sdb/qemu-new/qemu_test/qemu/util/async.c:136 | ||
27 | |||
28 | Fixes: 5f07c4d60d091320186e7b0edaf9ed2cc16b2d1e | ||
29 | Reported-by: Euler Robot <euler.robot@huawei.com> | ||
30 | Signed-off-by: Pan Nengyuan <pannengyuan@huawei.com> | ||
31 | Message-Id: <20200310064640.5059-1-pannengyuan@huawei.com> | ||
32 | Reviewed-by: Daniel P. Berrangé <berrange@redhat.com> | ||
33 | Acked-by: Igor Mammedov <imammedo@redhat.com> | ||
34 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> | 12 | Signed-off-by: Kevin Wolf <kwolf@redhat.com> |
35 | --- | 13 | --- |
36 | qom/qom-qmp-cmds.c | 16 ++++++---------- | 14 | block/nbd.c | 8 ++------ |
37 | 1 file changed, 6 insertions(+), 10 deletions(-) | 15 | 1 file changed, 2 insertions(+), 6 deletions(-) |
38 | 16 | ||
39 | diff --git a/qom/qom-qmp-cmds.c b/qom/qom-qmp-cmds.c | 17 | diff --git a/block/nbd.c b/block/nbd.c |
40 | index XXXXXXX..XXXXXXX 100644 | 18 | index XXXXXXX..XXXXXXX 100644 |
41 | --- a/qom/qom-qmp-cmds.c | 19 | --- a/block/nbd.c |
42 | +++ b/qom/qom-qmp-cmds.c | 20 | +++ b/block/nbd.c |
43 | @@ -XXX,XX +XXX,XX @@ void qmp_object_add(QDict *qdict, QObject **ret_data, Error **errp) | 21 | @@ -XXX,XX +XXX,XX @@ static int coroutine_fn nbd_co_send_request(BlockDriverState *bs, |
44 | QDict *pdict; | 22 | if (qiov) { |
45 | Visitor *v; | 23 | qio_channel_set_cork(s->ioc, true); |
46 | Object *obj; | 24 | rc = nbd_send_request(s->ioc, request); |
47 | - const char *type; | 25 | - if (rc >= 0) { |
48 | - const char *id; | 26 | - if (qio_channel_writev_all(s->ioc, qiov->iov, qiov->niov, |
49 | + g_autofree char *type = NULL; | 27 | - NULL) < 0) { |
50 | + g_autofree char *id = NULL; | 28 | - rc = -EIO; |
51 | 29 | - } | |
52 | - type = qdict_get_try_str(qdict, "qom-type"); | 30 | - } else if (rc >= 0) { |
53 | + type = g_strdup(qdict_get_try_str(qdict, "qom-type")); | 31 | + if (rc >= 0 && qio_channel_writev_all(s->ioc, qiov->iov, qiov->niov, |
54 | if (!type) { | 32 | + NULL) < 0) { |
55 | error_setg(errp, QERR_MISSING_PARAMETER, "qom-type"); | 33 | rc = -EIO; |
56 | return; | 34 | } |
57 | - } else { | 35 | qio_channel_set_cork(s->ioc, false); |
58 | - type = g_strdup(type); | ||
59 | - qdict_del(qdict, "qom-type"); | ||
60 | } | ||
61 | + qdict_del(qdict, "qom-type"); | ||
62 | |||
63 | - id = qdict_get_try_str(qdict, "id"); | ||
64 | + id = g_strdup(qdict_get_try_str(qdict, "id")); | ||
65 | if (!id) { | ||
66 | error_setg(errp, QERR_MISSING_PARAMETER, "id"); | ||
67 | return; | ||
68 | - } else { | ||
69 | - id = g_strdup(id); | ||
70 | - qdict_del(qdict, "id"); | ||
71 | } | ||
72 | + qdict_del(qdict, "id"); | ||
73 | |||
74 | props = qdict_get(qdict, "props"); | ||
75 | if (props) { | ||
76 | -- | 36 | -- |
77 | 2.20.1 | 37 | 2.35.3 |
78 | |||
79 | diff view generated by jsdifflib |