1 | The following changes since commit 6632f6ff96f0537fc34cdc00c760656fc62e23c5: | 1 | The following changes since commit f45cc81911adc7726e8a2801986b6998b91b816e: |
---|---|---|---|
2 | 2 | ||
3 | Merge remote-tracking branch 'remotes/famz/tags/block-and-testing-pull-request' into staging (2017-07-17 11:46:36 +0100) | 3 | Merge remote-tracking branch 'remotes/cschoenebeck/tags/pull-9p-20220307' into staging (2022-03-08 09:06:57 +0000) |
4 | 4 | ||
5 | are available in the git repository at: | 5 | are available in the git repository at: |
6 | 6 | ||
7 | https://github.com/jasowang/qemu.git tags/net-pull-request | 7 | https://github.com/jasowang/qemu.git tags/net-pull-request |
8 | 8 | ||
9 | for you to fetch changes up to 189ae6bb5ce1f5a322f8691d00fe942ba43dd601: | 9 | for you to fetch changes up to a10dd1e279fc56cebc7e738925e0db85d0cea089: |
10 | 10 | ||
11 | virtio-net: fix offload ctrl endian (2017-07-17 20:13:56 +0800) | 11 | vdpa: Expose VHOST_F_LOG_ALL on SVQ (2022-03-08 21:18:41 +0800) |
12 | 12 | ||
13 | ---------------------------------------------------------------- | 13 | ---------------------------------------------------------------- |
14 | 14 | ||
15 | - fix virtio-net ctrl offload endian | 15 | ---------------------------------------------------------------- |
16 | - vnet header support for variou COLO netfilters and compare thread | 16 | Eugenio Pérez (14): |
17 | vhost: Add VhostShadowVirtqueue | ||
18 | vhost: Add Shadow VirtQueue kick forwarding capabilities | ||
19 | vhost: Add Shadow VirtQueue call forwarding capabilities | ||
20 | vhost: Add vhost_svq_valid_features to shadow vq | ||
21 | virtio: Add vhost_svq_get_vring_addr | ||
22 | vdpa: adapt vhost_ops callbacks to svq | ||
23 | vhost: Shadow virtqueue buffers forwarding | ||
24 | util: Add iova_tree_alloc_map | ||
25 | util: add iova_tree_find_iova | ||
26 | vhost: Add VhostIOVATree | ||
27 | vdpa: Add custom IOTLB translations to SVQ | ||
28 | vdpa: Adapt vhost_vdpa_get_vring_base to SVQ | ||
29 | vdpa: Never set log_base addr if SVQ is enabled | ||
30 | vdpa: Expose VHOST_F_LOG_ALL on SVQ | ||
17 | 31 | ||
18 | ---------------------------------------------------------------- | ||
19 | Jason Wang (1): | 32 | Jason Wang (1): |
20 | virtio-net: fix offload ctrl endian | 33 | virtio-net: fix map leaking on error during receive |
21 | 34 | ||
22 | Michal Privoznik (1): | 35 | hw/net/virtio-net.c | 1 + |
23 | virtion-net: Prefer is_power_of_2() | 36 | hw/virtio/meson.build | 2 +- |
24 | 37 | hw/virtio/vhost-iova-tree.c | 110 +++++++ | |
25 | Zhang Chen (12): | 38 | hw/virtio/vhost-iova-tree.h | 27 ++ |
26 | net: Add vnet_hdr_len arguments in NetClientState | 39 | hw/virtio/vhost-shadow-virtqueue.c | 637 +++++++++++++++++++++++++++++++++++++ |
27 | net/net.c: Add vnet_hdr support in SocketReadState | 40 | hw/virtio/vhost-shadow-virtqueue.h | 87 +++++ |
28 | net/filter-mirror.c: Introduce parameter for filter_send() | 41 | hw/virtio/vhost-vdpa.c | 525 +++++++++++++++++++++++++++++- |
29 | net/filter-mirror.c: Make filter mirror support vnet support. | 42 | include/hw/virtio/vhost-vdpa.h | 8 + |
30 | net/filter-mirror.c: Add new option to enable vnet support for filter-redirector | 43 | include/qemu/iova-tree.h | 38 ++- |
31 | net/colo.c: Make vnet_hdr_len as packet property | 44 | util/iova-tree.c | 169 ++++++++++ |
32 | net/colo-compare.c: Introduce parameter for compare_chr_send() | 45 | 10 files changed, 1587 insertions(+), 17 deletions(-) |
33 | net/colo-compare.c: Make colo-compare support vnet_hdr_len | 46 | create mode 100644 hw/virtio/vhost-iova-tree.c |
34 | net/colo.c: Add vnet packet parse feature in colo-proxy | 47 | create mode 100644 hw/virtio/vhost-iova-tree.h |
35 | net/colo-compare.c: Add vnet packet's tcp/udp/icmp compare | 48 | create mode 100644 hw/virtio/vhost-shadow-virtqueue.c |
36 | net/filter-rewriter.c: Make filter-rewriter support vnet_hdr_len | 49 | create mode 100644 hw/virtio/vhost-shadow-virtqueue.h |
37 | docs/colo-proxy.txt: Update colo-proxy usage of net driver with vnet_header | ||
38 | |||
39 | docs/colo-proxy.txt | 26 ++++++++++++++++ | ||
40 | hw/net/virtio-net.c | 4 ++- | ||
41 | include/net/net.h | 10 ++++-- | ||
42 | net/colo-compare.c | 84 ++++++++++++++++++++++++++++++++++++++++++--------- | ||
43 | net/colo.c | 9 +++--- | ||
44 | net/colo.h | 4 ++- | ||
45 | net/filter-mirror.c | 75 +++++++++++++++++++++++++++++++++++++++++---- | ||
46 | net/filter-rewriter.c | 37 ++++++++++++++++++++++- | ||
47 | net/net.c | 37 ++++++++++++++++++++--- | ||
48 | net/socket.c | 8 ++--- | ||
49 | qemu-options.hx | 19 ++++++------ | ||
50 | 11 files changed, 265 insertions(+), 48 deletions(-) | ||
51 | 50 | ||
52 | 51 | diff view generated by jsdifflib |
1 | Spec said offloads should be le64, so use virtio_ldq_p() to guarantee | 1 | Commit bedd7e93d0196 ("virtio-net: fix use after unmap/free for sg") |
---|---|---|---|
2 | valid endian. | 2 | tries to fix the use after free of the sg by caching the virtqueue |
3 | elements in an array and unmap them at once after receiving the | ||
4 | packets, But it forgot to unmap the cached elements on error which | ||
5 | will lead to leaking of mapping and other unexpected results. | ||
3 | 6 | ||
4 | Fixes: 644c98587d4c ("virtio-net: dynamic network offloads configuration") | 7 | Fixing this by detaching the cached elements on error. This addresses |
8 | CVE-2022-26353. | ||
9 | |||
10 | Reported-by: Victor Tom <vv474172261@gmail.com> | ||
5 | Cc: qemu-stable@nongnu.org | 11 | Cc: qemu-stable@nongnu.org |
6 | Cc: Dmitry Fleytman <dfleytma@redhat.com> | 12 | Fixes: CVE-2022-26353 |
13 | Fixes: bedd7e93d0196 ("virtio-net: fix use after unmap/free for sg") | ||
14 | Reviewed-by: Michael S. Tsirkin <mst@redhat.com> | ||
7 | Signed-off-by: Jason Wang <jasowang@redhat.com> | 15 | Signed-off-by: Jason Wang <jasowang@redhat.com> |
8 | --- | 16 | --- |
9 | hw/net/virtio-net.c | 2 ++ | 17 | hw/net/virtio-net.c | 1 + |
10 | 1 file changed, 2 insertions(+) | 18 | 1 file changed, 1 insertion(+) |
11 | 19 | ||
12 | diff --git a/hw/net/virtio-net.c b/hw/net/virtio-net.c | 20 | diff --git a/hw/net/virtio-net.c b/hw/net/virtio-net.c |
13 | index XXXXXXX..XXXXXXX 100644 | 21 | index XXXXXXX..XXXXXXX 100644 |
14 | --- a/hw/net/virtio-net.c | 22 | --- a/hw/net/virtio-net.c |
15 | +++ b/hw/net/virtio-net.c | 23 | +++ b/hw/net/virtio-net.c |
16 | @@ -XXX,XX +XXX,XX @@ static int virtio_net_handle_offloads(VirtIONet *n, uint8_t cmd, | 24 | @@ -XXX,XX +XXX,XX @@ static ssize_t virtio_net_receive_rcu(NetClientState *nc, const uint8_t *buf, |
17 | if (cmd == VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET) { | 25 | |
18 | uint64_t supported_offloads; | 26 | err: |
19 | 27 | for (j = 0; j < i; j++) { | |
20 | + offloads = virtio_ldq_p(vdev, &offloads); | 28 | + virtqueue_detach_element(q->rx_vq, elems[j], lens[j]); |
21 | + | 29 | g_free(elems[j]); |
22 | if (!n->has_vnet_hdr) { | 30 | } |
23 | return VIRTIO_NET_ERR; | 31 | |
24 | } | ||
25 | -- | 32 | -- |
26 | 2.7.4 | 33 | 2.7.4 |
27 | |||
28 | diff view generated by jsdifflib |
New patch | |||
---|---|---|---|
1 | From: Eugenio Pérez <eperezma@redhat.com> | ||
1 | 2 | ||
3 | Vhost shadow virtqueue (SVQ) is an intermediate jump for virtqueue | ||
4 | notifications and buffers, allowing qemu to track them. While qemu is | ||
5 | forwarding the buffers and virtqueue changes, it is able to commit the | ||
6 | memory it's being dirtied, the same way regular qemu's VirtIO devices | ||
7 | do. | ||
8 | |||
9 | This commit only exposes basic SVQ allocation and free. Next patches of | ||
10 | the series add functionality like notifications and buffers forwarding. | ||
11 | |||
12 | Acked-by: Michael S. Tsirkin <mst@redhat.com> | ||
13 | Signed-off-by: Eugenio Pérez <eperezma@redhat.com> | ||
14 | Signed-off-by: Jason Wang <jasowang@redhat.com> | ||
15 | --- | ||
16 | hw/virtio/meson.build | 2 +- | ||
17 | hw/virtio/vhost-shadow-virtqueue.c | 62 ++++++++++++++++++++++++++++++++++++++ | ||
18 | hw/virtio/vhost-shadow-virtqueue.h | 28 +++++++++++++++++ | ||
19 | 3 files changed, 91 insertions(+), 1 deletion(-) | ||
20 | create mode 100644 hw/virtio/vhost-shadow-virtqueue.c | ||
21 | create mode 100644 hw/virtio/vhost-shadow-virtqueue.h | ||
22 | |||
23 | diff --git a/hw/virtio/meson.build b/hw/virtio/meson.build | ||
24 | index XXXXXXX..XXXXXXX 100644 | ||
25 | --- a/hw/virtio/meson.build | ||
26 | +++ b/hw/virtio/meson.build | ||
27 | @@ -XXX,XX +XXX,XX @@ softmmu_ss.add(when: 'CONFIG_ALL', if_true: files('vhost-stub.c')) | ||
28 | |||
29 | virtio_ss = ss.source_set() | ||
30 | virtio_ss.add(files('virtio.c')) | ||
31 | -virtio_ss.add(when: 'CONFIG_VHOST', if_true: files('vhost.c', 'vhost-backend.c')) | ||
32 | +virtio_ss.add(when: 'CONFIG_VHOST', if_true: files('vhost.c', 'vhost-backend.c', 'vhost-shadow-virtqueue.c')) | ||
33 | virtio_ss.add(when: 'CONFIG_VHOST_USER', if_true: files('vhost-user.c')) | ||
34 | virtio_ss.add(when: 'CONFIG_VHOST_VDPA', if_true: files('vhost-vdpa.c')) | ||
35 | virtio_ss.add(when: 'CONFIG_VIRTIO_BALLOON', if_true: files('virtio-balloon.c')) | ||
36 | diff --git a/hw/virtio/vhost-shadow-virtqueue.c b/hw/virtio/vhost-shadow-virtqueue.c | ||
37 | new file mode 100644 | ||
38 | index XXXXXXX..XXXXXXX | ||
39 | --- /dev/null | ||
40 | +++ b/hw/virtio/vhost-shadow-virtqueue.c | ||
41 | @@ -XXX,XX +XXX,XX @@ | ||
42 | +/* | ||
43 | + * vhost shadow virtqueue | ||
44 | + * | ||
45 | + * SPDX-FileCopyrightText: Red Hat, Inc. 2021 | ||
46 | + * SPDX-FileContributor: Author: Eugenio Pérez <eperezma@redhat.com> | ||
47 | + * | ||
48 | + * SPDX-License-Identifier: GPL-2.0-or-later | ||
49 | + */ | ||
50 | + | ||
51 | +#include "qemu/osdep.h" | ||
52 | +#include "hw/virtio/vhost-shadow-virtqueue.h" | ||
53 | + | ||
54 | +#include "qemu/error-report.h" | ||
55 | + | ||
56 | +/** | ||
57 | + * Creates vhost shadow virtqueue, and instructs the vhost device to use the | ||
58 | + * shadow methods and file descriptors. | ||
59 | + * | ||
60 | + * Returns the new virtqueue or NULL. | ||
61 | + * | ||
62 | + * In case of error, reason is reported through error_report. | ||
63 | + */ | ||
64 | +VhostShadowVirtqueue *vhost_svq_new(void) | ||
65 | +{ | ||
66 | + g_autofree VhostShadowVirtqueue *svq = g_new0(VhostShadowVirtqueue, 1); | ||
67 | + int r; | ||
68 | + | ||
69 | + r = event_notifier_init(&svq->hdev_kick, 0); | ||
70 | + if (r != 0) { | ||
71 | + error_report("Couldn't create kick event notifier: %s (%d)", | ||
72 | + g_strerror(errno), errno); | ||
73 | + goto err_init_hdev_kick; | ||
74 | + } | ||
75 | + | ||
76 | + r = event_notifier_init(&svq->hdev_call, 0); | ||
77 | + if (r != 0) { | ||
78 | + error_report("Couldn't create call event notifier: %s (%d)", | ||
79 | + g_strerror(errno), errno); | ||
80 | + goto err_init_hdev_call; | ||
81 | + } | ||
82 | + | ||
83 | + return g_steal_pointer(&svq); | ||
84 | + | ||
85 | +err_init_hdev_call: | ||
86 | + event_notifier_cleanup(&svq->hdev_kick); | ||
87 | + | ||
88 | +err_init_hdev_kick: | ||
89 | + return NULL; | ||
90 | +} | ||
91 | + | ||
92 | +/** | ||
93 | + * Free the resources of the shadow virtqueue. | ||
94 | + * | ||
95 | + * @pvq: gpointer to SVQ so it can be used by autofree functions. | ||
96 | + */ | ||
97 | +void vhost_svq_free(gpointer pvq) | ||
98 | +{ | ||
99 | + VhostShadowVirtqueue *vq = pvq; | ||
100 | + event_notifier_cleanup(&vq->hdev_kick); | ||
101 | + event_notifier_cleanup(&vq->hdev_call); | ||
102 | + g_free(vq); | ||
103 | +} | ||
104 | diff --git a/hw/virtio/vhost-shadow-virtqueue.h b/hw/virtio/vhost-shadow-virtqueue.h | ||
105 | new file mode 100644 | ||
106 | index XXXXXXX..XXXXXXX | ||
107 | --- /dev/null | ||
108 | +++ b/hw/virtio/vhost-shadow-virtqueue.h | ||
109 | @@ -XXX,XX +XXX,XX @@ | ||
110 | +/* | ||
111 | + * vhost shadow virtqueue | ||
112 | + * | ||
113 | + * SPDX-FileCopyrightText: Red Hat, Inc. 2021 | ||
114 | + * SPDX-FileContributor: Author: Eugenio Pérez <eperezma@redhat.com> | ||
115 | + * | ||
116 | + * SPDX-License-Identifier: GPL-2.0-or-later | ||
117 | + */ | ||
118 | + | ||
119 | +#ifndef VHOST_SHADOW_VIRTQUEUE_H | ||
120 | +#define VHOST_SHADOW_VIRTQUEUE_H | ||
121 | + | ||
122 | +#include "qemu/event_notifier.h" | ||
123 | + | ||
124 | +/* Shadow virtqueue to relay notifications */ | ||
125 | +typedef struct VhostShadowVirtqueue { | ||
126 | + /* Shadow kick notifier, sent to vhost */ | ||
127 | + EventNotifier hdev_kick; | ||
128 | + /* Shadow call notifier, sent to vhost */ | ||
129 | + EventNotifier hdev_call; | ||
130 | +} VhostShadowVirtqueue; | ||
131 | + | ||
132 | +VhostShadowVirtqueue *vhost_svq_new(void); | ||
133 | + | ||
134 | +void vhost_svq_free(gpointer vq); | ||
135 | +G_DEFINE_AUTOPTR_CLEANUP_FUNC(VhostShadowVirtqueue, vhost_svq_free); | ||
136 | + | ||
137 | +#endif | ||
138 | -- | ||
139 | 2.7.4 | ||
140 | |||
141 | diff view generated by jsdifflib |
1 | From: Michal Privoznik <mprivozn@redhat.com> | 1 | From: Eugenio Pérez <eperezma@redhat.com> |
---|---|---|---|
2 | 2 | ||
3 | We have a function that checks if given number is power of two. | 3 | At this mode no buffer forwarding will be performed in SVQ mode: Qemu |
4 | We should prefer it instead of expanding the check on our own. | 4 | will just forward the guest's kicks to the device. |
5 | 5 | ||
6 | Signed-off-by: Michal Privoznik <mprivozn@redhat.com> | 6 | Host memory notifiers regions are left out for simplicity, and they will |
7 | not be addressed in this series. | ||
8 | |||
9 | Acked-by: Michael S. Tsirkin <mst@redhat.com> | ||
10 | Signed-off-by: Eugenio Pérez <eperezma@redhat.com> | ||
7 | Signed-off-by: Jason Wang <jasowang@redhat.com> | 11 | Signed-off-by: Jason Wang <jasowang@redhat.com> |
8 | --- | 12 | --- |
9 | hw/net/virtio-net.c | 2 +- | 13 | hw/virtio/vhost-shadow-virtqueue.c | 56 ++++++++++++++ |
10 | 1 file changed, 1 insertion(+), 1 deletion(-) | 14 | hw/virtio/vhost-shadow-virtqueue.h | 14 ++++ |
11 | 15 | hw/virtio/vhost-vdpa.c | 145 ++++++++++++++++++++++++++++++++++++- | |
12 | diff --git a/hw/net/virtio-net.c b/hw/net/virtio-net.c | 16 | include/hw/virtio/vhost-vdpa.h | 4 + |
17 | 4 files changed, 217 insertions(+), 2 deletions(-) | ||
18 | |||
19 | diff --git a/hw/virtio/vhost-shadow-virtqueue.c b/hw/virtio/vhost-shadow-virtqueue.c | ||
13 | index XXXXXXX..XXXXXXX 100644 | 20 | index XXXXXXX..XXXXXXX 100644 |
14 | --- a/hw/net/virtio-net.c | 21 | --- a/hw/virtio/vhost-shadow-virtqueue.c |
15 | +++ b/hw/net/virtio-net.c | 22 | +++ b/hw/virtio/vhost-shadow-virtqueue.c |
16 | @@ -XXX,XX +XXX,XX @@ static void virtio_net_device_realize(DeviceState *dev, Error **errp) | 23 | @@ -XXX,XX +XXX,XX @@ |
17 | */ | 24 | #include "hw/virtio/vhost-shadow-virtqueue.h" |
18 | if (n->net_conf.rx_queue_size < VIRTIO_NET_RX_QUEUE_MIN_SIZE || | 25 | |
19 | n->net_conf.rx_queue_size > VIRTQUEUE_MAX_SIZE || | 26 | #include "qemu/error-report.h" |
20 | - (n->net_conf.rx_queue_size & (n->net_conf.rx_queue_size - 1))) { | 27 | +#include "qemu/main-loop.h" |
21 | + !is_power_of_2(n->net_conf.rx_queue_size)) { | 28 | +#include "linux-headers/linux/vhost.h" |
22 | error_setg(errp, "Invalid rx_queue_size (= %" PRIu16 "), " | 29 | + |
23 | "must be a power of 2 between %d and %d.", | 30 | +/** |
24 | n->net_conf.rx_queue_size, VIRTIO_NET_RX_QUEUE_MIN_SIZE, | 31 | + * Forward guest notifications. |
32 | + * | ||
33 | + * @n: guest kick event notifier, the one that guest set to notify svq. | ||
34 | + */ | ||
35 | +static void vhost_handle_guest_kick(EventNotifier *n) | ||
36 | +{ | ||
37 | + VhostShadowVirtqueue *svq = container_of(n, VhostShadowVirtqueue, | ||
38 | + svq_kick); | ||
39 | + event_notifier_test_and_clear(n); | ||
40 | + event_notifier_set(&svq->hdev_kick); | ||
41 | +} | ||
42 | + | ||
43 | +/** | ||
44 | + * Set a new file descriptor for the guest to kick the SVQ and notify for avail | ||
45 | + * | ||
46 | + * @svq: The svq | ||
47 | + * @svq_kick_fd: The svq kick fd | ||
48 | + * | ||
49 | + * Note that the SVQ will never close the old file descriptor. | ||
50 | + */ | ||
51 | +void vhost_svq_set_svq_kick_fd(VhostShadowVirtqueue *svq, int svq_kick_fd) | ||
52 | +{ | ||
53 | + EventNotifier *svq_kick = &svq->svq_kick; | ||
54 | + bool poll_stop = VHOST_FILE_UNBIND != event_notifier_get_fd(svq_kick); | ||
55 | + bool poll_start = svq_kick_fd != VHOST_FILE_UNBIND; | ||
56 | + | ||
57 | + if (poll_stop) { | ||
58 | + event_notifier_set_handler(svq_kick, NULL); | ||
59 | + } | ||
60 | + | ||
61 | + /* | ||
62 | + * event_notifier_set_handler already checks for guest's notifications if | ||
63 | + * they arrive at the new file descriptor in the switch, so there is no | ||
64 | + * need to explicitly check for them. | ||
65 | + */ | ||
66 | + if (poll_start) { | ||
67 | + event_notifier_init_fd(svq_kick, svq_kick_fd); | ||
68 | + event_notifier_set(svq_kick); | ||
69 | + event_notifier_set_handler(svq_kick, vhost_handle_guest_kick); | ||
70 | + } | ||
71 | +} | ||
72 | + | ||
73 | +/** | ||
74 | + * Stop the shadow virtqueue operation. | ||
75 | + * @svq: Shadow Virtqueue | ||
76 | + */ | ||
77 | +void vhost_svq_stop(VhostShadowVirtqueue *svq) | ||
78 | +{ | ||
79 | + event_notifier_set_handler(&svq->svq_kick, NULL); | ||
80 | +} | ||
81 | |||
82 | /** | ||
83 | * Creates vhost shadow virtqueue, and instructs the vhost device to use the | ||
84 | @@ -XXX,XX +XXX,XX @@ VhostShadowVirtqueue *vhost_svq_new(void) | ||
85 | goto err_init_hdev_call; | ||
86 | } | ||
87 | |||
88 | + event_notifier_init_fd(&svq->svq_kick, VHOST_FILE_UNBIND); | ||
89 | return g_steal_pointer(&svq); | ||
90 | |||
91 | err_init_hdev_call: | ||
92 | @@ -XXX,XX +XXX,XX @@ err_init_hdev_kick: | ||
93 | void vhost_svq_free(gpointer pvq) | ||
94 | { | ||
95 | VhostShadowVirtqueue *vq = pvq; | ||
96 | + vhost_svq_stop(vq); | ||
97 | event_notifier_cleanup(&vq->hdev_kick); | ||
98 | event_notifier_cleanup(&vq->hdev_call); | ||
99 | g_free(vq); | ||
100 | diff --git a/hw/virtio/vhost-shadow-virtqueue.h b/hw/virtio/vhost-shadow-virtqueue.h | ||
101 | index XXXXXXX..XXXXXXX 100644 | ||
102 | --- a/hw/virtio/vhost-shadow-virtqueue.h | ||
103 | +++ b/hw/virtio/vhost-shadow-virtqueue.h | ||
104 | @@ -XXX,XX +XXX,XX @@ typedef struct VhostShadowVirtqueue { | ||
105 | EventNotifier hdev_kick; | ||
106 | /* Shadow call notifier, sent to vhost */ | ||
107 | EventNotifier hdev_call; | ||
108 | + | ||
109 | + /* | ||
110 | + * Borrowed virtqueue's guest to host notifier. To borrow it in this event | ||
111 | + * notifier allows to recover the VhostShadowVirtqueue from the event loop | ||
112 | + * easily. If we use the VirtQueue's one, we don't have an easy way to | ||
113 | + * retrieve VhostShadowVirtqueue. | ||
114 | + * | ||
115 | + * So shadow virtqueue must not clean it, or we would lose VirtQueue one. | ||
116 | + */ | ||
117 | + EventNotifier svq_kick; | ||
118 | } VhostShadowVirtqueue; | ||
119 | |||
120 | +void vhost_svq_set_svq_kick_fd(VhostShadowVirtqueue *svq, int svq_kick_fd); | ||
121 | + | ||
122 | +void vhost_svq_stop(VhostShadowVirtqueue *svq); | ||
123 | + | ||
124 | VhostShadowVirtqueue *vhost_svq_new(void); | ||
125 | |||
126 | void vhost_svq_free(gpointer vq); | ||
127 | diff --git a/hw/virtio/vhost-vdpa.c b/hw/virtio/vhost-vdpa.c | ||
128 | index XXXXXXX..XXXXXXX 100644 | ||
129 | --- a/hw/virtio/vhost-vdpa.c | ||
130 | +++ b/hw/virtio/vhost-vdpa.c | ||
131 | @@ -XXX,XX +XXX,XX @@ | ||
132 | #include "hw/virtio/vhost.h" | ||
133 | #include "hw/virtio/vhost-backend.h" | ||
134 | #include "hw/virtio/virtio-net.h" | ||
135 | +#include "hw/virtio/vhost-shadow-virtqueue.h" | ||
136 | #include "hw/virtio/vhost-vdpa.h" | ||
137 | #include "exec/address-spaces.h" | ||
138 | #include "qemu/main-loop.h" | ||
139 | #include "cpu.h" | ||
140 | #include "trace.h" | ||
141 | #include "qemu-common.h" | ||
142 | +#include "qapi/error.h" | ||
143 | |||
144 | /* | ||
145 | * Return one past the end of the end of section. Be careful with uint64_t | ||
146 | @@ -XXX,XX +XXX,XX @@ static bool vhost_vdpa_one_time_request(struct vhost_dev *dev) | ||
147 | return v->index != 0; | ||
148 | } | ||
149 | |||
150 | +static int vhost_vdpa_init_svq(struct vhost_dev *hdev, struct vhost_vdpa *v, | ||
151 | + Error **errp) | ||
152 | +{ | ||
153 | + g_autoptr(GPtrArray) shadow_vqs = NULL; | ||
154 | + | ||
155 | + if (!v->shadow_vqs_enabled) { | ||
156 | + return 0; | ||
157 | + } | ||
158 | + | ||
159 | + shadow_vqs = g_ptr_array_new_full(hdev->nvqs, vhost_svq_free); | ||
160 | + for (unsigned n = 0; n < hdev->nvqs; ++n) { | ||
161 | + g_autoptr(VhostShadowVirtqueue) svq = vhost_svq_new(); | ||
162 | + | ||
163 | + if (unlikely(!svq)) { | ||
164 | + error_setg(errp, "Cannot create svq %u", n); | ||
165 | + return -1; | ||
166 | + } | ||
167 | + g_ptr_array_add(shadow_vqs, g_steal_pointer(&svq)); | ||
168 | + } | ||
169 | + | ||
170 | + v->shadow_vqs = g_steal_pointer(&shadow_vqs); | ||
171 | + return 0; | ||
172 | +} | ||
173 | + | ||
174 | static int vhost_vdpa_init(struct vhost_dev *dev, void *opaque, Error **errp) | ||
175 | { | ||
176 | struct vhost_vdpa *v; | ||
177 | @@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_init(struct vhost_dev *dev, void *opaque, Error **errp) | ||
178 | dev->opaque = opaque ; | ||
179 | v->listener = vhost_vdpa_memory_listener; | ||
180 | v->msg_type = VHOST_IOTLB_MSG_V2; | ||
181 | + ret = vhost_vdpa_init_svq(dev, v, errp); | ||
182 | + if (ret) { | ||
183 | + goto err; | ||
184 | + } | ||
185 | |||
186 | vhost_vdpa_get_iova_range(v); | ||
187 | |||
188 | @@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_init(struct vhost_dev *dev, void *opaque, Error **errp) | ||
189 | VIRTIO_CONFIG_S_DRIVER); | ||
190 | |||
191 | return 0; | ||
192 | + | ||
193 | +err: | ||
194 | + ram_block_discard_disable(false); | ||
195 | + return ret; | ||
196 | } | ||
197 | |||
198 | static void vhost_vdpa_host_notifier_uninit(struct vhost_dev *dev, | ||
199 | @@ -XXX,XX +XXX,XX @@ err: | ||
200 | |||
201 | static void vhost_vdpa_host_notifiers_init(struct vhost_dev *dev) | ||
202 | { | ||
203 | + struct vhost_vdpa *v = dev->opaque; | ||
204 | int i; | ||
205 | |||
206 | + if (v->shadow_vqs_enabled) { | ||
207 | + /* FIXME SVQ is not compatible with host notifiers mr */ | ||
208 | + return; | ||
209 | + } | ||
210 | + | ||
211 | for (i = dev->vq_index; i < dev->vq_index + dev->nvqs; i++) { | ||
212 | if (vhost_vdpa_host_notifier_init(dev, i)) { | ||
213 | goto err; | ||
214 | @@ -XXX,XX +XXX,XX @@ err: | ||
215 | return; | ||
216 | } | ||
217 | |||
218 | +static void vhost_vdpa_svq_cleanup(struct vhost_dev *dev) | ||
219 | +{ | ||
220 | + struct vhost_vdpa *v = dev->opaque; | ||
221 | + size_t idx; | ||
222 | + | ||
223 | + if (!v->shadow_vqs) { | ||
224 | + return; | ||
225 | + } | ||
226 | + | ||
227 | + for (idx = 0; idx < v->shadow_vqs->len; ++idx) { | ||
228 | + vhost_svq_stop(g_ptr_array_index(v->shadow_vqs, idx)); | ||
229 | + } | ||
230 | + g_ptr_array_free(v->shadow_vqs, true); | ||
231 | +} | ||
232 | + | ||
233 | static int vhost_vdpa_cleanup(struct vhost_dev *dev) | ||
234 | { | ||
235 | struct vhost_vdpa *v; | ||
236 | @@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_cleanup(struct vhost_dev *dev) | ||
237 | trace_vhost_vdpa_cleanup(dev, v); | ||
238 | vhost_vdpa_host_notifiers_uninit(dev, dev->nvqs); | ||
239 | memory_listener_unregister(&v->listener); | ||
240 | + vhost_vdpa_svq_cleanup(dev); | ||
241 | |||
242 | dev->opaque = NULL; | ||
243 | ram_block_discard_disable(false); | ||
244 | @@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_get_device_id(struct vhost_dev *dev, | ||
245 | return ret; | ||
246 | } | ||
247 | |||
248 | +static void vhost_vdpa_reset_svq(struct vhost_vdpa *v) | ||
249 | +{ | ||
250 | + if (!v->shadow_vqs_enabled) { | ||
251 | + return; | ||
252 | + } | ||
253 | + | ||
254 | + for (unsigned i = 0; i < v->shadow_vqs->len; ++i) { | ||
255 | + VhostShadowVirtqueue *svq = g_ptr_array_index(v->shadow_vqs, i); | ||
256 | + vhost_svq_stop(svq); | ||
257 | + } | ||
258 | +} | ||
259 | + | ||
260 | static int vhost_vdpa_reset_device(struct vhost_dev *dev) | ||
261 | { | ||
262 | + struct vhost_vdpa *v = dev->opaque; | ||
263 | int ret; | ||
264 | uint8_t status = 0; | ||
265 | |||
266 | + vhost_vdpa_reset_svq(v); | ||
267 | + | ||
268 | ret = vhost_vdpa_call(dev, VHOST_VDPA_SET_STATUS, &status); | ||
269 | trace_vhost_vdpa_reset_device(dev, status); | ||
270 | return ret; | ||
271 | @@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_get_config(struct vhost_dev *dev, uint8_t *config, | ||
272 | return ret; | ||
273 | } | ||
274 | |||
275 | +static int vhost_vdpa_set_vring_dev_kick(struct vhost_dev *dev, | ||
276 | + struct vhost_vring_file *file) | ||
277 | +{ | ||
278 | + trace_vhost_vdpa_set_vring_kick(dev, file->index, file->fd); | ||
279 | + return vhost_vdpa_call(dev, VHOST_SET_VRING_KICK, file); | ||
280 | +} | ||
281 | + | ||
282 | +/** | ||
283 | + * Set the shadow virtqueue descriptors to the device | ||
284 | + * | ||
285 | + * @dev: The vhost device model | ||
286 | + * @svq: The shadow virtqueue | ||
287 | + * @idx: The index of the virtqueue in the vhost device | ||
288 | + * @errp: Error | ||
289 | + */ | ||
290 | +static bool vhost_vdpa_svq_setup(struct vhost_dev *dev, | ||
291 | + VhostShadowVirtqueue *svq, | ||
292 | + unsigned idx, | ||
293 | + Error **errp) | ||
294 | +{ | ||
295 | + struct vhost_vring_file file = { | ||
296 | + .index = dev->vq_index + idx, | ||
297 | + }; | ||
298 | + const EventNotifier *event_notifier = &svq->hdev_kick; | ||
299 | + int r; | ||
300 | + | ||
301 | + file.fd = event_notifier_get_fd(event_notifier); | ||
302 | + r = vhost_vdpa_set_vring_dev_kick(dev, &file); | ||
303 | + if (unlikely(r != 0)) { | ||
304 | + error_setg_errno(errp, -r, "Can't set device kick fd"); | ||
305 | + } | ||
306 | + | ||
307 | + return r == 0; | ||
308 | +} | ||
309 | + | ||
310 | +static bool vhost_vdpa_svqs_start(struct vhost_dev *dev) | ||
311 | +{ | ||
312 | + struct vhost_vdpa *v = dev->opaque; | ||
313 | + Error *err = NULL; | ||
314 | + unsigned i; | ||
315 | + | ||
316 | + if (!v->shadow_vqs) { | ||
317 | + return true; | ||
318 | + } | ||
319 | + | ||
320 | + for (i = 0; i < v->shadow_vqs->len; ++i) { | ||
321 | + VhostShadowVirtqueue *svq = g_ptr_array_index(v->shadow_vqs, i); | ||
322 | + bool ok = vhost_vdpa_svq_setup(dev, svq, i, &err); | ||
323 | + if (unlikely(!ok)) { | ||
324 | + error_reportf_err(err, "Cannot setup SVQ %u: ", i); | ||
325 | + return false; | ||
326 | + } | ||
327 | + } | ||
328 | + | ||
329 | + return true; | ||
330 | +} | ||
331 | + | ||
332 | static int vhost_vdpa_dev_start(struct vhost_dev *dev, bool started) | ||
333 | { | ||
334 | struct vhost_vdpa *v = dev->opaque; | ||
335 | + bool ok; | ||
336 | trace_vhost_vdpa_dev_start(dev, started); | ||
337 | |||
338 | if (started) { | ||
339 | vhost_vdpa_host_notifiers_init(dev); | ||
340 | + ok = vhost_vdpa_svqs_start(dev); | ||
341 | + if (unlikely(!ok)) { | ||
342 | + return -1; | ||
343 | + } | ||
344 | vhost_vdpa_set_vring_ready(dev); | ||
345 | } else { | ||
346 | vhost_vdpa_host_notifiers_uninit(dev, dev->nvqs); | ||
347 | @@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_get_vring_base(struct vhost_dev *dev, | ||
348 | static int vhost_vdpa_set_vring_kick(struct vhost_dev *dev, | ||
349 | struct vhost_vring_file *file) | ||
350 | { | ||
351 | - trace_vhost_vdpa_set_vring_kick(dev, file->index, file->fd); | ||
352 | - return vhost_vdpa_call(dev, VHOST_SET_VRING_KICK, file); | ||
353 | + struct vhost_vdpa *v = dev->opaque; | ||
354 | + int vdpa_idx = file->index - dev->vq_index; | ||
355 | + | ||
356 | + if (v->shadow_vqs_enabled) { | ||
357 | + VhostShadowVirtqueue *svq = g_ptr_array_index(v->shadow_vqs, vdpa_idx); | ||
358 | + vhost_svq_set_svq_kick_fd(svq, file->fd); | ||
359 | + return 0; | ||
360 | + } else { | ||
361 | + return vhost_vdpa_set_vring_dev_kick(dev, file); | ||
362 | + } | ||
363 | } | ||
364 | |||
365 | static int vhost_vdpa_set_vring_call(struct vhost_dev *dev, | ||
366 | diff --git a/include/hw/virtio/vhost-vdpa.h b/include/hw/virtio/vhost-vdpa.h | ||
367 | index XXXXXXX..XXXXXXX 100644 | ||
368 | --- a/include/hw/virtio/vhost-vdpa.h | ||
369 | +++ b/include/hw/virtio/vhost-vdpa.h | ||
370 | @@ -XXX,XX +XXX,XX @@ | ||
371 | #ifndef HW_VIRTIO_VHOST_VDPA_H | ||
372 | #define HW_VIRTIO_VHOST_VDPA_H | ||
373 | |||
374 | +#include <gmodule.h> | ||
375 | + | ||
376 | #include "hw/virtio/virtio.h" | ||
377 | #include "standard-headers/linux/vhost_types.h" | ||
378 | |||
379 | @@ -XXX,XX +XXX,XX @@ typedef struct vhost_vdpa { | ||
380 | bool iotlb_batch_begin_sent; | ||
381 | MemoryListener listener; | ||
382 | struct vhost_vdpa_iova_range iova_range; | ||
383 | + bool shadow_vqs_enabled; | ||
384 | + GPtrArray *shadow_vqs; | ||
385 | struct vhost_dev *dev; | ||
386 | VhostVDPAHostNotifier notifier[VIRTIO_QUEUE_MAX]; | ||
387 | } VhostVDPA; | ||
25 | -- | 388 | -- |
26 | 2.7.4 | 389 | 2.7.4 |
27 | 390 | ||
28 | 391 | diff view generated by jsdifflib |
1 | From: Zhang Chen <zhangchen.fnst@cn.fujitsu.com> | 1 | From: Eugenio Pérez <eperezma@redhat.com> |
---|---|---|---|
2 | 2 | ||
3 | We add the vnet_hdr_support option for filter-redirector, default is disabled. | 3 | This will make qemu aware of the device used buffers, allowing it to |
4 | If you use virtio-net-pci net driver or other driver needs vnet_hdr, please enable it. | 4 | write the guest memory with its contents if needed. |
5 | Because colo-compare or other modules needs the vnet_hdr_len to parse | ||
6 | packet, we add this new option send the len to others. | ||
7 | You can use it for example: | ||
8 | -object filter-redirector,id=r0,netdev=hn0,queue=tx,outdev=red0,vnet_hdr_support | ||
9 | 5 | ||
10 | Signed-off-by: Zhang Chen <zhangchen.fnst@cn.fujitsu.com> | 6 | Acked-by: Michael S. Tsirkin <mst@redhat.com> |
7 | Signed-off-by: Eugenio Pérez <eperezma@redhat.com> | ||
11 | Signed-off-by: Jason Wang <jasowang@redhat.com> | 8 | Signed-off-by: Jason Wang <jasowang@redhat.com> |
12 | --- | 9 | --- |
13 | net/filter-mirror.c | 23 +++++++++++++++++++++++ | 10 | hw/virtio/vhost-shadow-virtqueue.c | 38 ++++++++++++++++++++++++++++++++++++++ |
14 | qemu-options.hx | 6 +++--- | 11 | hw/virtio/vhost-shadow-virtqueue.h | 4 ++++ |
15 | 2 files changed, 26 insertions(+), 3 deletions(-) | 12 | hw/virtio/vhost-vdpa.c | 31 +++++++++++++++++++++++++++++-- |
13 | 3 files changed, 71 insertions(+), 2 deletions(-) | ||
16 | 14 | ||
17 | diff --git a/net/filter-mirror.c b/net/filter-mirror.c | 15 | diff --git a/hw/virtio/vhost-shadow-virtqueue.c b/hw/virtio/vhost-shadow-virtqueue.c |
18 | index XXXXXXX..XXXXXXX 100644 | 16 | index XXXXXXX..XXXXXXX 100644 |
19 | --- a/net/filter-mirror.c | 17 | --- a/hw/virtio/vhost-shadow-virtqueue.c |
20 | +++ b/net/filter-mirror.c | 18 | +++ b/hw/virtio/vhost-shadow-virtqueue.c |
21 | @@ -XXX,XX +XXX,XX @@ static void filter_redirector_set_outdev(Object *obj, | 19 | @@ -XXX,XX +XXX,XX @@ static void vhost_handle_guest_kick(EventNotifier *n) |
22 | s->outdev = g_strdup(value); | ||
23 | } | 20 | } |
24 | 21 | ||
25 | +static bool filter_redirector_get_vnet_hdr(Object *obj, Error **errp) | 22 | /** |
23 | + * Forward vhost notifications | ||
24 | + * | ||
25 | + * @n: hdev call event notifier, the one that device set to notify svq. | ||
26 | + */ | ||
27 | +static void vhost_svq_handle_call(EventNotifier *n) | ||
26 | +{ | 28 | +{ |
27 | + MirrorState *s = FILTER_REDIRECTOR(obj); | 29 | + VhostShadowVirtqueue *svq = container_of(n, VhostShadowVirtqueue, |
28 | + | 30 | + hdev_call); |
29 | + return s->vnet_hdr; | 31 | + event_notifier_test_and_clear(n); |
32 | + event_notifier_set(&svq->svq_call); | ||
30 | +} | 33 | +} |
31 | + | 34 | + |
32 | +static void filter_redirector_set_vnet_hdr(Object *obj, | 35 | +/** |
33 | + bool value, | 36 | + * Set the call notifier for the SVQ to call the guest |
34 | + Error **errp) | 37 | + * |
38 | + * @svq: Shadow virtqueue | ||
39 | + * @call_fd: call notifier | ||
40 | + * | ||
41 | + * Called on BQL context. | ||
42 | + */ | ||
43 | +void vhost_svq_set_svq_call_fd(VhostShadowVirtqueue *svq, int call_fd) | ||
35 | +{ | 44 | +{ |
36 | + MirrorState *s = FILTER_REDIRECTOR(obj); | 45 | + if (call_fd == VHOST_FILE_UNBIND) { |
37 | + | 46 | + /* |
38 | + s->vnet_hdr = value; | 47 | + * Fail event_notifier_set if called handling device call. |
48 | + * | ||
49 | + * SVQ still needs device notifications, since it needs to keep | ||
50 | + * forwarding used buffers even with the unbind. | ||
51 | + */ | ||
52 | + memset(&svq->svq_call, 0, sizeof(svq->svq_call)); | ||
53 | + } else { | ||
54 | + event_notifier_init_fd(&svq->svq_call, call_fd); | ||
55 | + } | ||
39 | +} | 56 | +} |
40 | + | 57 | + |
41 | static void filter_mirror_init(Object *obj) | 58 | +/** |
59 | * Set a new file descriptor for the guest to kick the SVQ and notify for avail | ||
60 | * | ||
61 | * @svq: The svq | ||
62 | @@ -XXX,XX +XXX,XX @@ VhostShadowVirtqueue *vhost_svq_new(void) | ||
63 | } | ||
64 | |||
65 | event_notifier_init_fd(&svq->svq_kick, VHOST_FILE_UNBIND); | ||
66 | + event_notifier_set_handler(&svq->hdev_call, vhost_svq_handle_call); | ||
67 | return g_steal_pointer(&svq); | ||
68 | |||
69 | err_init_hdev_call: | ||
70 | @@ -XXX,XX +XXX,XX @@ void vhost_svq_free(gpointer pvq) | ||
71 | VhostShadowVirtqueue *vq = pvq; | ||
72 | vhost_svq_stop(vq); | ||
73 | event_notifier_cleanup(&vq->hdev_kick); | ||
74 | + event_notifier_set_handler(&vq->hdev_call, NULL); | ||
75 | event_notifier_cleanup(&vq->hdev_call); | ||
76 | g_free(vq); | ||
77 | } | ||
78 | diff --git a/hw/virtio/vhost-shadow-virtqueue.h b/hw/virtio/vhost-shadow-virtqueue.h | ||
79 | index XXXXXXX..XXXXXXX 100644 | ||
80 | --- a/hw/virtio/vhost-shadow-virtqueue.h | ||
81 | +++ b/hw/virtio/vhost-shadow-virtqueue.h | ||
82 | @@ -XXX,XX +XXX,XX @@ typedef struct VhostShadowVirtqueue { | ||
83 | * So shadow virtqueue must not clean it, or we would lose VirtQueue one. | ||
84 | */ | ||
85 | EventNotifier svq_kick; | ||
86 | + | ||
87 | + /* Guest's call notifier, where the SVQ calls guest. */ | ||
88 | + EventNotifier svq_call; | ||
89 | } VhostShadowVirtqueue; | ||
90 | |||
91 | void vhost_svq_set_svq_kick_fd(VhostShadowVirtqueue *svq, int svq_kick_fd); | ||
92 | +void vhost_svq_set_svq_call_fd(VhostShadowVirtqueue *svq, int call_fd); | ||
93 | |||
94 | void vhost_svq_stop(VhostShadowVirtqueue *svq); | ||
95 | |||
96 | diff --git a/hw/virtio/vhost-vdpa.c b/hw/virtio/vhost-vdpa.c | ||
97 | index XXXXXXX..XXXXXXX 100644 | ||
98 | --- a/hw/virtio/vhost-vdpa.c | ||
99 | +++ b/hw/virtio/vhost-vdpa.c | ||
100 | @@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_set_vring_dev_kick(struct vhost_dev *dev, | ||
101 | return vhost_vdpa_call(dev, VHOST_SET_VRING_KICK, file); | ||
102 | } | ||
103 | |||
104 | +static int vhost_vdpa_set_vring_dev_call(struct vhost_dev *dev, | ||
105 | + struct vhost_vring_file *file) | ||
106 | +{ | ||
107 | + trace_vhost_vdpa_set_vring_call(dev, file->index, file->fd); | ||
108 | + return vhost_vdpa_call(dev, VHOST_SET_VRING_CALL, file); | ||
109 | +} | ||
110 | + | ||
111 | /** | ||
112 | * Set the shadow virtqueue descriptors to the device | ||
113 | * | ||
114 | @@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_set_vring_dev_kick(struct vhost_dev *dev, | ||
115 | * @svq: The shadow virtqueue | ||
116 | * @idx: The index of the virtqueue in the vhost device | ||
117 | * @errp: Error | ||
118 | + * | ||
119 | + * Note that this function does not rewind kick file descriptor if cannot set | ||
120 | + * call one. | ||
121 | */ | ||
122 | static bool vhost_vdpa_svq_setup(struct vhost_dev *dev, | ||
123 | VhostShadowVirtqueue *svq, | ||
124 | @@ -XXX,XX +XXX,XX @@ static bool vhost_vdpa_svq_setup(struct vhost_dev *dev, | ||
125 | r = vhost_vdpa_set_vring_dev_kick(dev, &file); | ||
126 | if (unlikely(r != 0)) { | ||
127 | error_setg_errno(errp, -r, "Can't set device kick fd"); | ||
128 | + return false; | ||
129 | + } | ||
130 | + | ||
131 | + event_notifier = &svq->hdev_call; | ||
132 | + file.fd = event_notifier_get_fd(event_notifier); | ||
133 | + r = vhost_vdpa_set_vring_dev_call(dev, &file); | ||
134 | + if (unlikely(r != 0)) { | ||
135 | + error_setg_errno(errp, -r, "Can't set device call fd"); | ||
136 | } | ||
137 | |||
138 | return r == 0; | ||
139 | @@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_set_vring_kick(struct vhost_dev *dev, | ||
140 | static int vhost_vdpa_set_vring_call(struct vhost_dev *dev, | ||
141 | struct vhost_vring_file *file) | ||
42 | { | 142 | { |
43 | MirrorState *s = FILTER_MIRROR(obj); | 143 | - trace_vhost_vdpa_set_vring_call(dev, file->index, file->fd); |
44 | @@ -XXX,XX +XXX,XX @@ static void filter_mirror_init(Object *obj) | 144 | - return vhost_vdpa_call(dev, VHOST_SET_VRING_CALL, file); |
45 | 145 | + struct vhost_vdpa *v = dev->opaque; | |
46 | static void filter_redirector_init(Object *obj) | ||
47 | { | ||
48 | + MirrorState *s = FILTER_REDIRECTOR(obj); | ||
49 | + | 146 | + |
50 | object_property_add_str(obj, "indev", filter_redirector_get_indev, | 147 | + if (v->shadow_vqs_enabled) { |
51 | filter_redirector_set_indev, NULL); | 148 | + int vdpa_idx = file->index - dev->vq_index; |
52 | object_property_add_str(obj, "outdev", filter_redirector_get_outdev, | 149 | + VhostShadowVirtqueue *svq = g_ptr_array_index(v->shadow_vqs, vdpa_idx); |
53 | filter_redirector_set_outdev, NULL); | ||
54 | + | 150 | + |
55 | + s->vnet_hdr = false; | 151 | + vhost_svq_set_svq_call_fd(svq, file->fd); |
56 | + object_property_add_bool(obj, "vnet_hdr_support", | 152 | + return 0; |
57 | + filter_redirector_get_vnet_hdr, | 153 | + } else { |
58 | + filter_redirector_set_vnet_hdr, NULL); | 154 | + return vhost_vdpa_set_vring_dev_call(dev, file); |
155 | + } | ||
59 | } | 156 | } |
60 | 157 | ||
61 | static void filter_mirror_fini(Object *obj) | 158 | static int vhost_vdpa_get_features(struct vhost_dev *dev, |
62 | diff --git a/qemu-options.hx b/qemu-options.hx | ||
63 | index XXXXXXX..XXXXXXX 100644 | ||
64 | --- a/qemu-options.hx | ||
65 | +++ b/qemu-options.hx | ||
66 | @@ -XXX,XX +XXX,XX @@ queue @var{all|rx|tx} is an option that can be applied to any netfilter. | ||
67 | |||
68 | filter-mirror on netdev @var{netdevid},mirror net packet to chardev@var{chardevid}, if it has the vnet_hdr_support flag, filter-mirror will mirror packet with vnet_hdr_len. | ||
69 | |||
70 | -@item -object filter-redirector,id=@var{id},netdev=@var{netdevid},indev=@var{chardevid}, | ||
71 | -outdev=@var{chardevid}[,queue=@var{all|rx|tx}] | ||
72 | +@item -object filter-redirector,id=@var{id},netdev=@var{netdevid},indev=@var{chardevid},outdev=@var{chardevid},queue=@var{all|rx|tx}[,vnet_hdr_support] | ||
73 | |||
74 | filter-redirector on netdev @var{netdevid},redirect filter's net packet to chardev | ||
75 | -@var{chardevid},and redirect indev's packet to filter. | ||
76 | +@var{chardevid},and redirect indev's packet to filter.if it has the vnet_hdr_support flag, | ||
77 | +filter-redirector will redirect packet with vnet_hdr_len. | ||
78 | Create a filter-redirector we need to differ outdev id from indev id, id can not | ||
79 | be the same. we can just use indev or outdev, but at least one of indev or outdev | ||
80 | need to be specified. | ||
81 | -- | 159 | -- |
82 | 2.7.4 | 160 | 2.7.4 |
83 | 161 | ||
84 | 162 | diff view generated by jsdifflib |
1 | From: Zhang Chen <zhangchen.fnst@cn.fujitsu.com> | 1 | From: Eugenio Pérez <eperezma@redhat.com> |
---|---|---|---|
2 | 2 | ||
3 | Signed-off-by: Zhang Chen <zhangchen.fnst@cn.fujitsu.com> | 3 | This allows SVQ to negotiate features with the guest and the device. For |
4 | the device, SVQ is a driver. While this function bypasses all | ||
5 | non-transport features, it needs to disable the features that SVQ does | ||
6 | not support when forwarding buffers. This includes packed vq layout, | ||
7 | indirect descriptors or event idx. | ||
8 | |||
9 | Future changes can add support to offer more features to the guest, | ||
10 | since the use of VirtQueue gives this for free. This is left out at the | ||
11 | moment for simplicity. | ||
12 | |||
13 | Acked-by: Michael S. Tsirkin <mst@redhat.com> | ||
14 | Signed-off-by: Eugenio Pérez <eperezma@redhat.com> | ||
4 | Signed-off-by: Jason Wang <jasowang@redhat.com> | 15 | Signed-off-by: Jason Wang <jasowang@redhat.com> |
5 | --- | 16 | --- |
6 | docs/colo-proxy.txt | 26 ++++++++++++++++++++++++++ | 17 | hw/virtio/vhost-shadow-virtqueue.c | 44 ++++++++++++++++++++++++++++++++++++++ |
7 | 1 file changed, 26 insertions(+) | 18 | hw/virtio/vhost-shadow-virtqueue.h | 2 ++ |
19 | hw/virtio/vhost-vdpa.c | 15 +++++++++++++ | ||
20 | 3 files changed, 61 insertions(+) | ||
8 | 21 | ||
9 | diff --git a/docs/colo-proxy.txt b/docs/colo-proxy.txt | 22 | diff --git a/hw/virtio/vhost-shadow-virtqueue.c b/hw/virtio/vhost-shadow-virtqueue.c |
10 | index XXXXXXX..XXXXXXX 100644 | 23 | index XXXXXXX..XXXXXXX 100644 |
11 | --- a/docs/colo-proxy.txt | 24 | --- a/hw/virtio/vhost-shadow-virtqueue.c |
12 | +++ b/docs/colo-proxy.txt | 25 | +++ b/hw/virtio/vhost-shadow-virtqueue.c |
13 | @@ -XXX,XX +XXX,XX @@ Secondary(ip:3.3.3.8): | 26 | @@ -XXX,XX +XXX,XX @@ |
14 | -chardev socket,id=red1,host=3.3.3.3,port=9004 | 27 | #include "hw/virtio/vhost-shadow-virtqueue.h" |
15 | -object filter-redirector,id=f1,netdev=hn0,queue=tx,indev=red0 | 28 | |
16 | -object filter-redirector,id=f2,netdev=hn0,queue=rx,outdev=red1 | 29 | #include "qemu/error-report.h" |
17 | +-object filter-rewriter,id=f3,netdev=hn0,queue=all | 30 | +#include "qapi/error.h" |
31 | #include "qemu/main-loop.h" | ||
32 | #include "linux-headers/linux/vhost.h" | ||
33 | |||
34 | /** | ||
35 | + * Validate the transport device features that both guests can use with the SVQ | ||
36 | + * and SVQs can use with the device. | ||
37 | + * | ||
38 | + * @dev_features: The features | ||
39 | + * @errp: Error pointer | ||
40 | + */ | ||
41 | +bool vhost_svq_valid_features(uint64_t features, Error **errp) | ||
42 | +{ | ||
43 | + bool ok = true; | ||
44 | + uint64_t svq_features = features; | ||
18 | + | 45 | + |
19 | +If you want to use virtio-net-pci or other driver with vnet_header: | 46 | + for (uint64_t b = VIRTIO_TRANSPORT_F_START; b <= VIRTIO_TRANSPORT_F_END; |
47 | + ++b) { | ||
48 | + switch (b) { | ||
49 | + case VIRTIO_F_ANY_LAYOUT: | ||
50 | + continue; | ||
20 | + | 51 | + |
21 | +Primary(ip:3.3.3.3): | 52 | + case VIRTIO_F_ACCESS_PLATFORM: |
22 | +-netdev tap,id=hn0,vhost=off,script=/etc/qemu-ifup,downscript=/etc/qemu-ifdown | 53 | + /* SVQ trust in the host's IOMMU to translate addresses */ |
23 | +-device e1000,id=e0,netdev=hn0,mac=52:a4:00:12:78:66 | 54 | + case VIRTIO_F_VERSION_1: |
24 | +-chardev socket,id=mirror0,host=3.3.3.3,port=9003,server,nowait | 55 | + /* SVQ trust that the guest vring is little endian */ |
25 | +-chardev socket,id=compare1,host=3.3.3.3,port=9004,server,nowait | 56 | + if (!(svq_features & BIT_ULL(b))) { |
26 | +-chardev socket,id=compare0,host=3.3.3.3,port=9001,server,nowait | 57 | + set_bit(b, &svq_features); |
27 | +-chardev socket,id=compare0-0,host=3.3.3.3,port=9001 | 58 | + ok = false; |
28 | +-chardev socket,id=compare_out,host=3.3.3.3,port=9005,server,nowait | 59 | + } |
29 | +-chardev socket,id=compare_out0,host=3.3.3.3,port=9005 | 60 | + continue; |
30 | +-object filter-mirror,id=m0,netdev=hn0,queue=tx,outdev=mirror0,vnet_hdr_support | ||
31 | +-object filter-redirector,netdev=hn0,id=redire0,queue=rx,indev=compare_out,vnet_hdr_support | ||
32 | +-object filter-redirector,netdev=hn0,id=redire1,queue=rx,outdev=compare0,vnet_hdr_support | ||
33 | +-object colo-compare,id=comp0,primary_in=compare0-0,secondary_in=compare1,outdev=compare_out0,vnet_hdr_support | ||
34 | + | 61 | + |
35 | +Secondary(ip:3.3.3.8): | 62 | + default: |
36 | +-netdev tap,id=hn0,vhost=off,script=/etc/qemu-ifup,down script=/etc/qemu-ifdown | 63 | + if (svq_features & BIT_ULL(b)) { |
37 | +-device e1000,netdev=hn0,mac=52:a4:00:12:78:66 | 64 | + clear_bit(b, &svq_features); |
38 | +-chardev socket,id=red0,host=3.3.3.3,port=9003 | 65 | + ok = false; |
39 | +-chardev socket,id=red1,host=3.3.3.3,port=9004 | 66 | + } |
40 | +-object filter-redirector,id=f1,netdev=hn0,queue=tx,indev=red0,vnet_hdr_support | 67 | + } |
41 | +-object filter-redirector,id=f2,netdev=hn0,queue=rx,outdev=red1,vnet_hdr_support | 68 | + } |
42 | +-object filter-rewriter,id=f3,netdev=hn0,queue=all,vnet_hdr_support | 69 | + |
43 | 70 | + if (!ok) { | |
44 | Note: | 71 | + error_setg(errp, "SVQ Invalid device feature flags, offer: 0x%"PRIx64 |
45 | a.COLO-proxy must work with COLO-frame and Block-replication. | 72 | + ", ok: 0x%"PRIx64, features, svq_features); |
73 | + } | ||
74 | + return ok; | ||
75 | +} | ||
76 | + | ||
77 | +/** | ||
78 | * Forward guest notifications. | ||
79 | * | ||
80 | * @n: guest kick event notifier, the one that guest set to notify svq. | ||
81 | diff --git a/hw/virtio/vhost-shadow-virtqueue.h b/hw/virtio/vhost-shadow-virtqueue.h | ||
82 | index XXXXXXX..XXXXXXX 100644 | ||
83 | --- a/hw/virtio/vhost-shadow-virtqueue.h | ||
84 | +++ b/hw/virtio/vhost-shadow-virtqueue.h | ||
85 | @@ -XXX,XX +XXX,XX @@ typedef struct VhostShadowVirtqueue { | ||
86 | EventNotifier svq_call; | ||
87 | } VhostShadowVirtqueue; | ||
88 | |||
89 | +bool vhost_svq_valid_features(uint64_t features, Error **errp); | ||
90 | + | ||
91 | void vhost_svq_set_svq_kick_fd(VhostShadowVirtqueue *svq, int svq_kick_fd); | ||
92 | void vhost_svq_set_svq_call_fd(VhostShadowVirtqueue *svq, int call_fd); | ||
93 | |||
94 | diff --git a/hw/virtio/vhost-vdpa.c b/hw/virtio/vhost-vdpa.c | ||
95 | index XXXXXXX..XXXXXXX 100644 | ||
96 | --- a/hw/virtio/vhost-vdpa.c | ||
97 | +++ b/hw/virtio/vhost-vdpa.c | ||
98 | @@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_init_svq(struct vhost_dev *hdev, struct vhost_vdpa *v, | ||
99 | Error **errp) | ||
100 | { | ||
101 | g_autoptr(GPtrArray) shadow_vqs = NULL; | ||
102 | + uint64_t dev_features, svq_features; | ||
103 | + int r; | ||
104 | + bool ok; | ||
105 | |||
106 | if (!v->shadow_vqs_enabled) { | ||
107 | return 0; | ||
108 | } | ||
109 | |||
110 | + r = hdev->vhost_ops->vhost_get_features(hdev, &dev_features); | ||
111 | + if (r != 0) { | ||
112 | + error_setg_errno(errp, -r, "Can't get vdpa device features"); | ||
113 | + return r; | ||
114 | + } | ||
115 | + | ||
116 | + svq_features = dev_features; | ||
117 | + ok = vhost_svq_valid_features(svq_features, errp); | ||
118 | + if (unlikely(!ok)) { | ||
119 | + return -1; | ||
120 | + } | ||
121 | + | ||
122 | shadow_vqs = g_ptr_array_new_full(hdev->nvqs, vhost_svq_free); | ||
123 | for (unsigned n = 0; n < hdev->nvqs; ++n) { | ||
124 | g_autoptr(VhostShadowVirtqueue) svq = vhost_svq_new(); | ||
46 | -- | 125 | -- |
47 | 2.7.4 | 126 | 2.7.4 |
48 | 127 | ||
49 | 128 | diff view generated by jsdifflib |
1 | From: Zhang Chen <zhangchen.fnst@cn.fujitsu.com> | 1 | From: Eugenio Pérez <eperezma@redhat.com> |
---|---|---|---|
2 | 2 | ||
3 | We add the vnet_hdr_support option for filter-rewriter, default is disabled. | 3 | It reports the shadow virtqueue address from qemu virtual address space. |
4 | If you use virtio-net-pci or other driver needs vnet_hdr, please enable it. | ||
5 | You can use it for example: | ||
6 | -object filter-rewriter,id=rew0,netdev=hn0,queue=all,vnet_hdr_support | ||
7 | 4 | ||
8 | We get the vnet_hdr_len from NetClientState that make us | 5 | Since this will be different from the guest's vaddr, but the device can |
9 | parse net packet correctly. | 6 | access it, SVQ takes special care about its alignment & lack of garbage |
7 | data. It assumes that IOMMU will work in host_page_size ranges for that. | ||
10 | 8 | ||
11 | Signed-off-by: Zhang Chen <zhangchen.fnst@cn.fujitsu.com> | 9 | Acked-by: Michael S. Tsirkin <mst@redhat.com> |
10 | Signed-off-by: Eugenio Pérez <eperezma@redhat.com> | ||
12 | Signed-off-by: Jason Wang <jasowang@redhat.com> | 11 | Signed-off-by: Jason Wang <jasowang@redhat.com> |
13 | --- | 12 | --- |
14 | net/filter-rewriter.c | 37 ++++++++++++++++++++++++++++++++++++- | 13 | hw/virtio/vhost-shadow-virtqueue.c | 29 +++++++++++++++++++++++++++++ |
15 | qemu-options.hx | 4 ++-- | 14 | hw/virtio/vhost-shadow-virtqueue.h | 9 +++++++++ |
16 | 2 files changed, 38 insertions(+), 3 deletions(-) | 15 | 2 files changed, 38 insertions(+) |
17 | 16 | ||
18 | diff --git a/net/filter-rewriter.c b/net/filter-rewriter.c | 17 | diff --git a/hw/virtio/vhost-shadow-virtqueue.c b/hw/virtio/vhost-shadow-virtqueue.c |
19 | index XXXXXXX..XXXXXXX 100644 | 18 | index XXXXXXX..XXXXXXX 100644 |
20 | --- a/net/filter-rewriter.c | 19 | --- a/hw/virtio/vhost-shadow-virtqueue.c |
21 | +++ b/net/filter-rewriter.c | 20 | +++ b/hw/virtio/vhost-shadow-virtqueue.c |
22 | @@ -XXX,XX +XXX,XX @@ | 21 | @@ -XXX,XX +XXX,XX @@ void vhost_svq_set_svq_call_fd(VhostShadowVirtqueue *svq, int call_fd) |
23 | #include "qemu-common.h" | ||
24 | #include "qapi/error.h" | ||
25 | #include "qapi/qmp/qerror.h" | ||
26 | +#include "qemu/error-report.h" | ||
27 | #include "qapi-visit.h" | ||
28 | #include "qom/object.h" | ||
29 | #include "qemu/main-loop.h" | ||
30 | @@ -XXX,XX +XXX,XX @@ typedef struct RewriterState { | ||
31 | NetQueue *incoming_queue; | ||
32 | /* hashtable to save connection */ | ||
33 | GHashTable *connection_track_table; | ||
34 | + bool vnet_hdr; | ||
35 | } RewriterState; | ||
36 | |||
37 | static void filter_rewriter_flush(NetFilterState *nf) | ||
38 | @@ -XXX,XX +XXX,XX @@ static ssize_t colo_rewriter_receive_iov(NetFilterState *nf, | ||
39 | ConnectionKey key; | ||
40 | Packet *pkt; | ||
41 | ssize_t size = iov_size(iov, iovcnt); | ||
42 | + ssize_t vnet_hdr_len = 0; | ||
43 | char *buf = g_malloc0(size); | ||
44 | |||
45 | iov_to_buf(iov, iovcnt, 0, buf, size); | ||
46 | - pkt = packet_new(buf, size, 0); | ||
47 | + | ||
48 | + if (s->vnet_hdr) { | ||
49 | + vnet_hdr_len = nf->netdev->vnet_hdr_len; | ||
50 | + } | ||
51 | + | ||
52 | + pkt = packet_new(buf, size, vnet_hdr_len); | ||
53 | g_free(buf); | ||
54 | |||
55 | /* | ||
56 | @@ -XXX,XX +XXX,XX @@ static void colo_rewriter_setup(NetFilterState *nf, Error **errp) | ||
57 | s->incoming_queue = qemu_new_net_queue(qemu_netfilter_pass_to_next, nf); | ||
58 | } | 22 | } |
59 | 23 | ||
60 | +static bool filter_rewriter_get_vnet_hdr(Object *obj, Error **errp) | 24 | /** |
25 | + * Get the shadow vq vring address. | ||
26 | + * @svq: Shadow virtqueue | ||
27 | + * @addr: Destination to store address | ||
28 | + */ | ||
29 | +void vhost_svq_get_vring_addr(const VhostShadowVirtqueue *svq, | ||
30 | + struct vhost_vring_addr *addr) | ||
61 | +{ | 31 | +{ |
62 | + RewriterState *s = FILTER_COLO_REWRITER(obj); | 32 | + addr->desc_user_addr = (uint64_t)svq->vring.desc; |
63 | + | 33 | + addr->avail_user_addr = (uint64_t)svq->vring.avail; |
64 | + return s->vnet_hdr; | 34 | + addr->used_user_addr = (uint64_t)svq->vring.used; |
65 | +} | 35 | +} |
66 | + | 36 | + |
67 | +static void filter_rewriter_set_vnet_hdr(Object *obj, | 37 | +size_t vhost_svq_driver_area_size(const VhostShadowVirtqueue *svq) |
68 | + bool value, | ||
69 | + Error **errp) | ||
70 | +{ | 38 | +{ |
71 | + RewriterState *s = FILTER_COLO_REWRITER(obj); | 39 | + size_t desc_size = sizeof(vring_desc_t) * svq->vring.num; |
40 | + size_t avail_size = offsetof(vring_avail_t, ring) + | ||
41 | + sizeof(uint16_t) * svq->vring.num; | ||
72 | + | 42 | + |
73 | + s->vnet_hdr = value; | 43 | + return ROUND_UP(desc_size + avail_size, qemu_real_host_page_size); |
74 | +} | 44 | +} |
75 | + | 45 | + |
76 | +static void filter_rewriter_init(Object *obj) | 46 | +size_t vhost_svq_device_area_size(const VhostShadowVirtqueue *svq) |
77 | +{ | 47 | +{ |
78 | + RewriterState *s = FILTER_COLO_REWRITER(obj); | 48 | + size_t used_size = offsetof(vring_used_t, ring) + |
79 | + | 49 | + sizeof(vring_used_elem_t) * svq->vring.num; |
80 | + s->vnet_hdr = false; | 50 | + return ROUND_UP(used_size, qemu_real_host_page_size); |
81 | + object_property_add_bool(obj, "vnet_hdr_support", | ||
82 | + filter_rewriter_get_vnet_hdr, | ||
83 | + filter_rewriter_set_vnet_hdr, NULL); | ||
84 | +} | 51 | +} |
85 | + | 52 | + |
86 | static void colo_rewriter_class_init(ObjectClass *oc, void *data) | 53 | +/** |
87 | { | 54 | * Set a new file descriptor for the guest to kick the SVQ and notify for avail |
88 | NetFilterClass *nfc = NETFILTER_CLASS(oc); | 55 | * |
89 | @@ -XXX,XX +XXX,XX @@ static const TypeInfo colo_rewriter_info = { | 56 | * @svq: The svq |
90 | .name = TYPE_FILTER_REWRITER, | 57 | diff --git a/hw/virtio/vhost-shadow-virtqueue.h b/hw/virtio/vhost-shadow-virtqueue.h |
91 | .parent = TYPE_NETFILTER, | ||
92 | .class_init = colo_rewriter_class_init, | ||
93 | + .instance_init = filter_rewriter_init, | ||
94 | .instance_size = sizeof(RewriterState), | ||
95 | }; | ||
96 | |||
97 | diff --git a/qemu-options.hx b/qemu-options.hx | ||
98 | index XXXXXXX..XXXXXXX 100644 | 58 | index XXXXXXX..XXXXXXX 100644 |
99 | --- a/qemu-options.hx | 59 | --- a/hw/virtio/vhost-shadow-virtqueue.h |
100 | +++ b/qemu-options.hx | 60 | +++ b/hw/virtio/vhost-shadow-virtqueue.h |
101 | @@ -XXX,XX +XXX,XX @@ Create a filter-redirector we need to differ outdev id from indev id, id can not | 61 | @@ -XXX,XX +XXX,XX @@ |
102 | be the same. we can just use indev or outdev, but at least one of indev or outdev | 62 | #define VHOST_SHADOW_VIRTQUEUE_H |
103 | need to be specified. | 63 | |
104 | 64 | #include "qemu/event_notifier.h" | |
105 | -@item -object filter-rewriter,id=@var{id},netdev=@var{netdevid}[,queue=@var{all|rx|tx}] | 65 | +#include "hw/virtio/virtio.h" |
106 | +@item -object filter-rewriter,id=@var{id},netdev=@var{netdevid},queue=@var{all|rx|tx},[vnet_hdr_support] | 66 | +#include "standard-headers/linux/vhost_types.h" |
107 | 67 | ||
108 | Filter-rewriter is a part of COLO project.It will rewrite tcp packet to | 68 | /* Shadow virtqueue to relay notifications */ |
109 | secondary from primary to keep secondary tcp connection,and rewrite | 69 | typedef struct VhostShadowVirtqueue { |
110 | tcp packet to primary from secondary make tcp packet can be handled by | 70 | + /* Shadow vring */ |
111 | -client. | 71 | + struct vring vring; |
112 | +client.if it has the vnet_hdr_support flag, we can parse packet with vnet header. | 72 | + |
113 | 73 | /* Shadow kick notifier, sent to vhost */ | |
114 | usage: | 74 | EventNotifier hdev_kick; |
115 | colo secondary: | 75 | /* Shadow call notifier, sent to vhost */ |
76 | @@ -XXX,XX +XXX,XX @@ bool vhost_svq_valid_features(uint64_t features, Error **errp); | ||
77 | |||
78 | void vhost_svq_set_svq_kick_fd(VhostShadowVirtqueue *svq, int svq_kick_fd); | ||
79 | void vhost_svq_set_svq_call_fd(VhostShadowVirtqueue *svq, int call_fd); | ||
80 | +void vhost_svq_get_vring_addr(const VhostShadowVirtqueue *svq, | ||
81 | + struct vhost_vring_addr *addr); | ||
82 | +size_t vhost_svq_driver_area_size(const VhostShadowVirtqueue *svq); | ||
83 | +size_t vhost_svq_device_area_size(const VhostShadowVirtqueue *svq); | ||
84 | |||
85 | void vhost_svq_stop(VhostShadowVirtqueue *svq); | ||
86 | |||
116 | -- | 87 | -- |
117 | 2.7.4 | 88 | 2.7.4 |
118 | 89 | ||
119 | 90 | diff view generated by jsdifflib |
1 | From: Zhang Chen <zhangchen.fnst@cn.fujitsu.com> | 1 | From: Eugenio Pérez <eperezma@redhat.com> |
---|---|---|---|
2 | 2 | ||
3 | We add the vnet_hdr_support option for colo-compare, default is disabled. | 3 | First half of the buffers forwarding part, preparing vhost-vdpa |
4 | If you use virtio-net-pci or other driver needs vnet_hdr, please enable it. | 4 | callbacks to SVQ to offer it. QEMU cannot enable it at this moment, so |
5 | You can use it for example: | 5 | this is effectively dead code at the moment, but it helps to reduce |
6 | -object colo-compare,id=comp0,primary_in=compare0-0,secondary_in=compare1,outdev=compare_out0,vnet_hdr_support | 6 | patch size. |
7 | 7 | ||
8 | COLO-compare can get vnet header length from filter, | 8 | Acked-by: Michael S. Tsirkin <mst@redhat.com> |
9 | Add vnet_hdr_len to struct packet and output packet with | 9 | Signed-off-by: Eugenio Pérez <eperezma@redhat.com> |
10 | the vnet_hdr_len. | ||
11 | |||
12 | Signed-off-by: Zhang Chen <zhangchen.fnst@cn.fujitsu.com> | ||
13 | Signed-off-by: Jason Wang <jasowang@redhat.com> | 10 | Signed-off-by: Jason Wang <jasowang@redhat.com> |
14 | --- | 11 | --- |
15 | net/colo-compare.c | 60 +++++++++++++++++++++++++++++++++++++++++++++++------- | 12 | hw/virtio/vhost-vdpa.c | 48 +++++++++++++++++++++++++++++++++++++++++------- |
16 | qemu-options.hx | 4 ++-- | 13 | 1 file changed, 41 insertions(+), 7 deletions(-) |
17 | 2 files changed, 55 insertions(+), 9 deletions(-) | ||
18 | 14 | ||
19 | diff --git a/net/colo-compare.c b/net/colo-compare.c | 15 | diff --git a/hw/virtio/vhost-vdpa.c b/hw/virtio/vhost-vdpa.c |
20 | index XXXXXXX..XXXXXXX 100644 | 16 | index XXXXXXX..XXXXXXX 100644 |
21 | --- a/net/colo-compare.c | 17 | --- a/hw/virtio/vhost-vdpa.c |
22 | +++ b/net/colo-compare.c | 18 | +++ b/hw/virtio/vhost-vdpa.c |
23 | @@ -XXX,XX +XXX,XX @@ typedef struct CompareState { | 19 | @@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_get_config(struct vhost_dev *dev, uint8_t *config, |
24 | CharBackend chr_out; | 20 | return ret; |
25 | SocketReadState pri_rs; | 21 | } |
26 | SocketReadState sec_rs; | 22 | |
27 | + bool vnet_hdr; | 23 | +static int vhost_vdpa_set_dev_vring_base(struct vhost_dev *dev, |
28 | 24 | + struct vhost_vring_state *ring) | |
29 | /* connection list: the connections belonged to this NIC could be found | 25 | +{ |
30 | * in this list. | 26 | + trace_vhost_vdpa_set_vring_base(dev, ring->index, ring->num); |
31 | @@ -XXX,XX +XXX,XX @@ enum { | 27 | + return vhost_vdpa_call(dev, VHOST_SET_VRING_BASE, ring); |
32 | 28 | +} | |
33 | static int compare_chr_send(CompareState *s, | 29 | + |
34 | const uint8_t *buf, | 30 | static int vhost_vdpa_set_vring_dev_kick(struct vhost_dev *dev, |
35 | - uint32_t size); | 31 | struct vhost_vring_file *file) |
36 | + uint32_t size, | ||
37 | + uint32_t vnet_hdr_len); | ||
38 | |||
39 | static gint seq_sorter(Packet *a, Packet *b, gpointer data) | ||
40 | { | 32 | { |
41 | @@ -XXX,XX +XXX,XX @@ static void colo_compare_connection(void *opaque, void *user_data) | 33 | @@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_set_vring_dev_call(struct vhost_dev *dev, |
42 | } | 34 | return vhost_vdpa_call(dev, VHOST_SET_VRING_CALL, file); |
43 | 35 | } | |
44 | if (result) { | 36 | |
45 | - ret = compare_chr_send(s, pkt->data, pkt->size); | 37 | +static int vhost_vdpa_set_vring_dev_addr(struct vhost_dev *dev, |
46 | + ret = compare_chr_send(s, | 38 | + struct vhost_vring_addr *addr) |
47 | + pkt->data, | 39 | +{ |
48 | + pkt->size, | 40 | + trace_vhost_vdpa_set_vring_addr(dev, addr->index, addr->flags, |
49 | + pkt->vnet_hdr_len); | 41 | + addr->desc_user_addr, addr->used_user_addr, |
50 | if (ret < 0) { | 42 | + addr->avail_user_addr, |
51 | error_report("colo_send_primary_packet failed"); | 43 | + addr->log_guest_addr); |
52 | } | 44 | + |
53 | @@ -XXX,XX +XXX,XX @@ static void colo_compare_connection(void *opaque, void *user_data) | 45 | + return vhost_vdpa_call(dev, VHOST_SET_VRING_ADDR, addr); |
54 | 46 | + | |
55 | static int compare_chr_send(CompareState *s, | 47 | +} |
56 | const uint8_t *buf, | 48 | + |
57 | - uint32_t size) | 49 | /** |
58 | + uint32_t size, | 50 | * Set the shadow virtqueue descriptors to the device |
59 | + uint32_t vnet_hdr_len) | 51 | * |
52 | @@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_set_log_base(struct vhost_dev *dev, uint64_t base, | ||
53 | static int vhost_vdpa_set_vring_addr(struct vhost_dev *dev, | ||
54 | struct vhost_vring_addr *addr) | ||
60 | { | 55 | { |
61 | int ret = 0; | 56 | - trace_vhost_vdpa_set_vring_addr(dev, addr->index, addr->flags, |
62 | uint32_t len = htonl(size); | 57 | - addr->desc_user_addr, addr->used_user_addr, |
63 | @@ -XXX,XX +XXX,XX @@ static int compare_chr_send(CompareState *s, | 58 | - addr->avail_user_addr, |
64 | goto err; | 59 | - addr->log_guest_addr); |
65 | } | 60 | - return vhost_vdpa_call(dev, VHOST_SET_VRING_ADDR, addr); |
66 | 61 | + struct vhost_vdpa *v = dev->opaque; | |
67 | + if (s->vnet_hdr) { | 62 | + |
63 | + if (v->shadow_vqs_enabled) { | ||
68 | + /* | 64 | + /* |
69 | + * We send vnet header len make other module(like filter-redirector) | 65 | + * Device vring addr was set at device start. SVQ base is handled by |
70 | + * know how to parse net packet correctly. | 66 | + * VirtQueue code. |
71 | + */ | 67 | + */ |
72 | + len = htonl(vnet_hdr_len); | 68 | + return 0; |
73 | + ret = qemu_chr_fe_write_all(&s->chr_out, (uint8_t *)&len, sizeof(len)); | ||
74 | + if (ret != sizeof(len)) { | ||
75 | + goto err; | ||
76 | + } | ||
77 | + } | 69 | + } |
78 | + | 70 | + |
79 | ret = qemu_chr_fe_write_all(&s->chr_out, (uint8_t *)buf, size); | 71 | + return vhost_vdpa_set_vring_dev_addr(dev, addr); |
80 | if (ret != size) { | ||
81 | goto err; | ||
82 | @@ -XXX,XX +XXX,XX @@ static void compare_set_outdev(Object *obj, const char *value, Error **errp) | ||
83 | s->outdev = g_strdup(value); | ||
84 | } | 72 | } |
85 | 73 | ||
86 | +static bool compare_get_vnet_hdr(Object *obj, Error **errp) | 74 | static int vhost_vdpa_set_vring_num(struct vhost_dev *dev, |
87 | +{ | 75 | @@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_set_vring_num(struct vhost_dev *dev, |
88 | + CompareState *s = COLO_COMPARE(obj); | 76 | static int vhost_vdpa_set_vring_base(struct vhost_dev *dev, |
77 | struct vhost_vring_state *ring) | ||
78 | { | ||
79 | - trace_vhost_vdpa_set_vring_base(dev, ring->index, ring->num); | ||
80 | - return vhost_vdpa_call(dev, VHOST_SET_VRING_BASE, ring); | ||
81 | + struct vhost_vdpa *v = dev->opaque; | ||
89 | + | 82 | + |
90 | + return s->vnet_hdr; | 83 | + if (v->shadow_vqs_enabled) { |
91 | +} | 84 | + /* |
85 | + * Device vring base was set at device start. SVQ base is handled by | ||
86 | + * VirtQueue code. | ||
87 | + */ | ||
88 | + return 0; | ||
89 | + } | ||
92 | + | 90 | + |
93 | +static void compare_set_vnet_hdr(Object *obj, | 91 | + return vhost_vdpa_set_dev_vring_base(dev, ring); |
94 | + bool value, | ||
95 | + Error **errp) | ||
96 | +{ | ||
97 | + CompareState *s = COLO_COMPARE(obj); | ||
98 | + | ||
99 | + s->vnet_hdr = value; | ||
100 | +} | ||
101 | + | ||
102 | static void compare_pri_rs_finalize(SocketReadState *pri_rs) | ||
103 | { | ||
104 | CompareState *s = container_of(pri_rs, CompareState, pri_rs); | ||
105 | |||
106 | if (packet_enqueue(s, PRIMARY_IN)) { | ||
107 | trace_colo_compare_main("primary: unsupported packet in"); | ||
108 | - compare_chr_send(s, pri_rs->buf, pri_rs->packet_len); | ||
109 | + compare_chr_send(s, | ||
110 | + pri_rs->buf, | ||
111 | + pri_rs->packet_len, | ||
112 | + pri_rs->vnet_hdr_len); | ||
113 | } else { | ||
114 | /* compare connection */ | ||
115 | g_queue_foreach(&s->conn_list, colo_compare_connection, s); | ||
116 | @@ -XXX,XX +XXX,XX @@ static void colo_compare_complete(UserCreatable *uc, Error **errp) | ||
117 | return; | ||
118 | } | ||
119 | |||
120 | - net_socket_rs_init(&s->pri_rs, compare_pri_rs_finalize, false); | ||
121 | - net_socket_rs_init(&s->sec_rs, compare_sec_rs_finalize, false); | ||
122 | + net_socket_rs_init(&s->pri_rs, compare_pri_rs_finalize, s->vnet_hdr); | ||
123 | + net_socket_rs_init(&s->sec_rs, compare_sec_rs_finalize, s->vnet_hdr); | ||
124 | |||
125 | g_queue_init(&s->conn_list); | ||
126 | |||
127 | @@ -XXX,XX +XXX,XX @@ static void colo_flush_packets(void *opaque, void *user_data) | ||
128 | |||
129 | while (!g_queue_is_empty(&conn->primary_list)) { | ||
130 | pkt = g_queue_pop_head(&conn->primary_list); | ||
131 | - compare_chr_send(s, pkt->data, pkt->size); | ||
132 | + compare_chr_send(s, | ||
133 | + pkt->data, | ||
134 | + pkt->size, | ||
135 | + pkt->vnet_hdr_len); | ||
136 | packet_destroy(pkt, NULL); | ||
137 | } | ||
138 | while (!g_queue_is_empty(&conn->secondary_list)) { | ||
139 | @@ -XXX,XX +XXX,XX @@ static void colo_compare_class_init(ObjectClass *oc, void *data) | ||
140 | |||
141 | static void colo_compare_init(Object *obj) | ||
142 | { | ||
143 | + CompareState *s = COLO_COMPARE(obj); | ||
144 | + | ||
145 | object_property_add_str(obj, "primary_in", | ||
146 | compare_get_pri_indev, compare_set_pri_indev, | ||
147 | NULL); | ||
148 | @@ -XXX,XX +XXX,XX @@ static void colo_compare_init(Object *obj) | ||
149 | object_property_add_str(obj, "outdev", | ||
150 | compare_get_outdev, compare_set_outdev, | ||
151 | NULL); | ||
152 | + | ||
153 | + s->vnet_hdr = false; | ||
154 | + object_property_add_bool(obj, "vnet_hdr_support", compare_get_vnet_hdr, | ||
155 | + compare_set_vnet_hdr, NULL); | ||
156 | } | 92 | } |
157 | 93 | ||
158 | static void colo_compare_finalize(Object *obj) | 94 | static int vhost_vdpa_get_vring_base(struct vhost_dev *dev, |
159 | diff --git a/qemu-options.hx b/qemu-options.hx | ||
160 | index XXXXXXX..XXXXXXX 100644 | ||
161 | --- a/qemu-options.hx | ||
162 | +++ b/qemu-options.hx | ||
163 | @@ -XXX,XX +XXX,XX @@ Dump the network traffic on netdev @var{dev} to the file specified by | ||
164 | The file format is libpcap, so it can be analyzed with tools such as tcpdump | ||
165 | or Wireshark. | ||
166 | |||
167 | -@item -object colo-compare,id=@var{id},primary_in=@var{chardevid},secondary_in=@var{chardevid}, | ||
168 | -outdev=@var{chardevid} | ||
169 | +@item -object colo-compare,id=@var{id},primary_in=@var{chardevid},secondary_in=@var{chardevid},outdev=@var{chardevid}[,vnet_hdr_support] | ||
170 | |||
171 | Colo-compare gets packet from primary_in@var{chardevid} and secondary_in@var{chardevid}, than compare primary packet with | ||
172 | secondary packet. If the packets are same, we will output primary | ||
173 | packet to outdev@var{chardevid}, else we will notify colo-frame | ||
174 | do checkpoint and send primary packet to outdev@var{chardevid}. | ||
175 | +if it has the vnet_hdr_support flag, colo compare will send/recv packet with vnet_hdr_len. | ||
176 | |||
177 | we must use it with the help of filter-mirror and filter-redirector. | ||
178 | |||
179 | -- | 95 | -- |
180 | 2.7.4 | 96 | 2.7.4 |
181 | 97 | ||
182 | 98 | diff view generated by jsdifflib |
1 | From: Zhang Chen <zhangchen.fnst@cn.fujitsu.com> | 1 | From: Eugenio Pérez <eperezma@redhat.com> |
---|---|---|---|
2 | 2 | ||
3 | We add a flag to decide whether net_fill_rstate() need read | 3 | Initial version of shadow virtqueue that actually forward buffers. There |
4 | the vnet_hdr_len or not. | 4 | is no iommu support at the moment, and that will be addressed in future |
5 | 5 | patches of this series. Since all vhost-vdpa devices use forced IOMMU, | |
6 | Signed-off-by: Zhang Chen <zhangchen.fnst@cn.fujitsu.com> | 6 | this means that SVQ is not usable at this point of the series on any |
7 | Suggested-by: Jason Wang <jasowang@redhat.com> | 7 | device. |
8 | |||
9 | For simplicity it only supports modern devices, that expects vring | ||
10 | in little endian, with split ring and no event idx or indirect | ||
11 | descriptors. Support for them will not be added in this series. | ||
12 | |||
13 | It reuses the VirtQueue code for the device part. The driver part is | ||
14 | based on Linux's virtio_ring driver, but with stripped functionality | ||
15 | and optimizations so it's easier to review. | ||
16 | |||
17 | However, forwarding buffers have some particular pieces: One of the most | ||
18 | unexpected ones is that a guest's buffer can expand through more than | ||
19 | one descriptor in SVQ. While this is handled gracefully by qemu's | ||
20 | emulated virtio devices, it may cause unexpected SVQ queue full. This | ||
21 | patch also solves it by checking for this condition at both guest's | ||
22 | kicks and device's calls. The code may be more elegant in the future if | ||
23 | SVQ code runs in its own iocontext. | ||
24 | |||
25 | Acked-by: Michael S. Tsirkin <mst@redhat.com> | ||
26 | Signed-off-by: Eugenio Pérez <eperezma@redhat.com> | ||
8 | Signed-off-by: Jason Wang <jasowang@redhat.com> | 27 | Signed-off-by: Jason Wang <jasowang@redhat.com> |
9 | --- | 28 | --- |
10 | include/net/net.h | 9 +++++++-- | 29 | hw/virtio/vhost-shadow-virtqueue.c | 353 ++++++++++++++++++++++++++++++++++++- |
11 | net/colo-compare.c | 4 ++-- | 30 | hw/virtio/vhost-shadow-virtqueue.h | 26 +++ |
12 | net/filter-mirror.c | 2 +- | 31 | hw/virtio/vhost-vdpa.c | 159 ++++++++++++++++- |
13 | net/net.c | 36 ++++++++++++++++++++++++++++++++---- | 32 | 3 files changed, 526 insertions(+), 12 deletions(-) |
14 | net/socket.c | 8 ++++---- | 33 | |
15 | 5 files changed, 46 insertions(+), 13 deletions(-) | 34 | diff --git a/hw/virtio/vhost-shadow-virtqueue.c b/hw/virtio/vhost-shadow-virtqueue.c |
16 | |||
17 | diff --git a/include/net/net.h b/include/net/net.h | ||
18 | index XXXXXXX..XXXXXXX 100644 | 35 | index XXXXXXX..XXXXXXX 100644 |
19 | --- a/include/net/net.h | 36 | --- a/hw/virtio/vhost-shadow-virtqueue.c |
20 | +++ b/include/net/net.h | 37 | +++ b/hw/virtio/vhost-shadow-virtqueue.c |
21 | @@ -XXX,XX +XXX,XX @@ typedef struct NICState { | 38 | @@ -XXX,XX +XXX,XX @@ |
22 | } NICState; | 39 | #include "qemu/error-report.h" |
23 | 40 | #include "qapi/error.h" | |
24 | struct SocketReadState { | 41 | #include "qemu/main-loop.h" |
25 | - int state; /* 0 = getting length, 1 = getting data */ | 42 | +#include "qemu/log.h" |
26 | + /* 0 = getting length, 1 = getting vnet header length, 2 = getting data */ | 43 | #include "linux-headers/linux/vhost.h" |
27 | + int state; | 44 | |
28 | + /* This flag decide whether to read the vnet_hdr_len field */ | 45 | /** |
29 | + bool vnet_hdr; | 46 | @@ -XXX,XX +XXX,XX @@ bool vhost_svq_valid_features(uint64_t features, Error **errp) |
30 | uint32_t index; | 47 | } |
31 | uint32_t packet_len; | 48 | |
32 | + uint32_t vnet_hdr_len; | 49 | /** |
33 | uint8_t buf[NET_BUFSIZE]; | 50 | - * Forward guest notifications. |
34 | SocketReadStateFinalize *finalize; | 51 | + * Number of descriptors that the SVQ can make available from the guest. |
35 | }; | 52 | + * |
36 | @@ -XXX,XX +XXX,XX @@ ssize_t qemu_deliver_packet_iov(NetClientState *sender, | 53 | + * @svq: The svq |
37 | void print_net_client(Monitor *mon, NetClientState *nc); | 54 | + */ |
38 | void hmp_info_network(Monitor *mon, const QDict *qdict); | 55 | +static uint16_t vhost_svq_available_slots(const VhostShadowVirtqueue *svq) |
39 | void net_socket_rs_init(SocketReadState *rs, | 56 | +{ |
40 | - SocketReadStateFinalize *finalize); | 57 | + return svq->vring.num - (svq->shadow_avail_idx - svq->shadow_used_idx); |
41 | + SocketReadStateFinalize *finalize, | 58 | +} |
42 | + bool vnet_hdr); | 59 | + |
43 | 60 | +static void vhost_vring_write_descs(VhostShadowVirtqueue *svq, | |
44 | /* NIC info */ | 61 | + const struct iovec *iovec, |
45 | 62 | + size_t num, bool more_descs, bool write) | |
46 | diff --git a/net/colo-compare.c b/net/colo-compare.c | 63 | +{ |
64 | + uint16_t i = svq->free_head, last = svq->free_head; | ||
65 | + unsigned n; | ||
66 | + uint16_t flags = write ? cpu_to_le16(VRING_DESC_F_WRITE) : 0; | ||
67 | + vring_desc_t *descs = svq->vring.desc; | ||
68 | + | ||
69 | + if (num == 0) { | ||
70 | + return; | ||
71 | + } | ||
72 | + | ||
73 | + for (n = 0; n < num; n++) { | ||
74 | + if (more_descs || (n + 1 < num)) { | ||
75 | + descs[i].flags = flags | cpu_to_le16(VRING_DESC_F_NEXT); | ||
76 | + } else { | ||
77 | + descs[i].flags = flags; | ||
78 | + } | ||
79 | + descs[i].addr = cpu_to_le64((hwaddr)iovec[n].iov_base); | ||
80 | + descs[i].len = cpu_to_le32(iovec[n].iov_len); | ||
81 | + | ||
82 | + last = i; | ||
83 | + i = cpu_to_le16(descs[i].next); | ||
84 | + } | ||
85 | + | ||
86 | + svq->free_head = le16_to_cpu(descs[last].next); | ||
87 | +} | ||
88 | + | ||
89 | +static bool vhost_svq_add_split(VhostShadowVirtqueue *svq, | ||
90 | + VirtQueueElement *elem, | ||
91 | + unsigned *head) | ||
92 | +{ | ||
93 | + unsigned avail_idx; | ||
94 | + vring_avail_t *avail = svq->vring.avail; | ||
95 | + | ||
96 | + *head = svq->free_head; | ||
97 | + | ||
98 | + /* We need some descriptors here */ | ||
99 | + if (unlikely(!elem->out_num && !elem->in_num)) { | ||
100 | + qemu_log_mask(LOG_GUEST_ERROR, | ||
101 | + "Guest provided element with no descriptors"); | ||
102 | + return false; | ||
103 | + } | ||
104 | + | ||
105 | + vhost_vring_write_descs(svq, elem->out_sg, elem->out_num, | ||
106 | + elem->in_num > 0, false); | ||
107 | + vhost_vring_write_descs(svq, elem->in_sg, elem->in_num, false, true); | ||
108 | + | ||
109 | + /* | ||
110 | + * Put the entry in the available array (but don't update avail->idx until | ||
111 | + * they do sync). | ||
112 | + */ | ||
113 | + avail_idx = svq->shadow_avail_idx & (svq->vring.num - 1); | ||
114 | + avail->ring[avail_idx] = cpu_to_le16(*head); | ||
115 | + svq->shadow_avail_idx++; | ||
116 | + | ||
117 | + /* Update the avail index after write the descriptor */ | ||
118 | + smp_wmb(); | ||
119 | + avail->idx = cpu_to_le16(svq->shadow_avail_idx); | ||
120 | + | ||
121 | + return true; | ||
122 | +} | ||
123 | + | ||
124 | +static bool vhost_svq_add(VhostShadowVirtqueue *svq, VirtQueueElement *elem) | ||
125 | +{ | ||
126 | + unsigned qemu_head; | ||
127 | + bool ok = vhost_svq_add_split(svq, elem, &qemu_head); | ||
128 | + if (unlikely(!ok)) { | ||
129 | + return false; | ||
130 | + } | ||
131 | + | ||
132 | + svq->ring_id_maps[qemu_head] = elem; | ||
133 | + return true; | ||
134 | +} | ||
135 | + | ||
136 | +static void vhost_svq_kick(VhostShadowVirtqueue *svq) | ||
137 | +{ | ||
138 | + /* | ||
139 | + * We need to expose the available array entries before checking the used | ||
140 | + * flags | ||
141 | + */ | ||
142 | + smp_mb(); | ||
143 | + if (svq->vring.used->flags & VRING_USED_F_NO_NOTIFY) { | ||
144 | + return; | ||
145 | + } | ||
146 | + | ||
147 | + event_notifier_set(&svq->hdev_kick); | ||
148 | +} | ||
149 | + | ||
150 | +/** | ||
151 | + * Forward available buffers. | ||
152 | + * | ||
153 | + * @svq: Shadow VirtQueue | ||
154 | + * | ||
155 | + * Note that this function does not guarantee that all guest's available | ||
156 | + * buffers are available to the device in SVQ avail ring. The guest may have | ||
157 | + * exposed a GPA / GIOVA contiguous buffer, but it may not be contiguous in | ||
158 | + * qemu vaddr. | ||
159 | + * | ||
160 | + * If that happens, guest's kick notifications will be disabled until the | ||
161 | + * device uses some buffers. | ||
162 | + */ | ||
163 | +static void vhost_handle_guest_kick(VhostShadowVirtqueue *svq) | ||
164 | +{ | ||
165 | + /* Clear event notifier */ | ||
166 | + event_notifier_test_and_clear(&svq->svq_kick); | ||
167 | + | ||
168 | + /* Forward to the device as many available buffers as possible */ | ||
169 | + do { | ||
170 | + virtio_queue_set_notification(svq->vq, false); | ||
171 | + | ||
172 | + while (true) { | ||
173 | + VirtQueueElement *elem; | ||
174 | + bool ok; | ||
175 | + | ||
176 | + if (svq->next_guest_avail_elem) { | ||
177 | + elem = g_steal_pointer(&svq->next_guest_avail_elem); | ||
178 | + } else { | ||
179 | + elem = virtqueue_pop(svq->vq, sizeof(*elem)); | ||
180 | + } | ||
181 | + | ||
182 | + if (!elem) { | ||
183 | + break; | ||
184 | + } | ||
185 | + | ||
186 | + if (elem->out_num + elem->in_num > | ||
187 | + vhost_svq_available_slots(svq)) { | ||
188 | + /* | ||
189 | + * This condition is possible since a contiguous buffer in GPA | ||
190 | + * does not imply a contiguous buffer in qemu's VA | ||
191 | + * scatter-gather segments. If that happens, the buffer exposed | ||
192 | + * to the device needs to be a chain of descriptors at this | ||
193 | + * moment. | ||
194 | + * | ||
195 | + * SVQ cannot hold more available buffers if we are here: | ||
196 | + * queue the current guest descriptor and ignore further kicks | ||
197 | + * until some elements are used. | ||
198 | + */ | ||
199 | + svq->next_guest_avail_elem = elem; | ||
200 | + return; | ||
201 | + } | ||
202 | + | ||
203 | + ok = vhost_svq_add(svq, elem); | ||
204 | + if (unlikely(!ok)) { | ||
205 | + /* VQ is broken, just return and ignore any other kicks */ | ||
206 | + return; | ||
207 | + } | ||
208 | + vhost_svq_kick(svq); | ||
209 | + } | ||
210 | + | ||
211 | + virtio_queue_set_notification(svq->vq, true); | ||
212 | + } while (!virtio_queue_empty(svq->vq)); | ||
213 | +} | ||
214 | + | ||
215 | +/** | ||
216 | + * Handle guest's kick. | ||
217 | * | ||
218 | * @n: guest kick event notifier, the one that guest set to notify svq. | ||
219 | */ | ||
220 | -static void vhost_handle_guest_kick(EventNotifier *n) | ||
221 | +static void vhost_handle_guest_kick_notifier(EventNotifier *n) | ||
222 | { | ||
223 | VhostShadowVirtqueue *svq = container_of(n, VhostShadowVirtqueue, | ||
224 | svq_kick); | ||
225 | event_notifier_test_and_clear(n); | ||
226 | - event_notifier_set(&svq->hdev_kick); | ||
227 | + vhost_handle_guest_kick(svq); | ||
228 | +} | ||
229 | + | ||
230 | +static bool vhost_svq_more_used(VhostShadowVirtqueue *svq) | ||
231 | +{ | ||
232 | + if (svq->last_used_idx != svq->shadow_used_idx) { | ||
233 | + return true; | ||
234 | + } | ||
235 | + | ||
236 | + svq->shadow_used_idx = cpu_to_le16(svq->vring.used->idx); | ||
237 | + | ||
238 | + return svq->last_used_idx != svq->shadow_used_idx; | ||
239 | } | ||
240 | |||
241 | /** | ||
242 | - * Forward vhost notifications | ||
243 | + * Enable vhost device calls after disable them. | ||
244 | + * | ||
245 | + * @svq: The svq | ||
246 | + * | ||
247 | + * It returns false if there are pending used buffers from the vhost device, | ||
248 | + * avoiding the possible races between SVQ checking for more work and enabling | ||
249 | + * callbacks. True if SVQ used vring has no more pending buffers. | ||
250 | + */ | ||
251 | +static bool vhost_svq_enable_notification(VhostShadowVirtqueue *svq) | ||
252 | +{ | ||
253 | + svq->vring.avail->flags &= ~cpu_to_le16(VRING_AVAIL_F_NO_INTERRUPT); | ||
254 | + /* Make sure the flag is written before the read of used_idx */ | ||
255 | + smp_mb(); | ||
256 | + return !vhost_svq_more_used(svq); | ||
257 | +} | ||
258 | + | ||
259 | +static void vhost_svq_disable_notification(VhostShadowVirtqueue *svq) | ||
260 | +{ | ||
261 | + svq->vring.avail->flags |= cpu_to_le16(VRING_AVAIL_F_NO_INTERRUPT); | ||
262 | +} | ||
263 | + | ||
264 | +static VirtQueueElement *vhost_svq_get_buf(VhostShadowVirtqueue *svq, | ||
265 | + uint32_t *len) | ||
266 | +{ | ||
267 | + vring_desc_t *descs = svq->vring.desc; | ||
268 | + const vring_used_t *used = svq->vring.used; | ||
269 | + vring_used_elem_t used_elem; | ||
270 | + uint16_t last_used; | ||
271 | + | ||
272 | + if (!vhost_svq_more_used(svq)) { | ||
273 | + return NULL; | ||
274 | + } | ||
275 | + | ||
276 | + /* Only get used array entries after they have been exposed by dev */ | ||
277 | + smp_rmb(); | ||
278 | + last_used = svq->last_used_idx & (svq->vring.num - 1); | ||
279 | + used_elem.id = le32_to_cpu(used->ring[last_used].id); | ||
280 | + used_elem.len = le32_to_cpu(used->ring[last_used].len); | ||
281 | + | ||
282 | + svq->last_used_idx++; | ||
283 | + if (unlikely(used_elem.id >= svq->vring.num)) { | ||
284 | + qemu_log_mask(LOG_GUEST_ERROR, "Device %s says index %u is used", | ||
285 | + svq->vdev->name, used_elem.id); | ||
286 | + return NULL; | ||
287 | + } | ||
288 | + | ||
289 | + if (unlikely(!svq->ring_id_maps[used_elem.id])) { | ||
290 | + qemu_log_mask(LOG_GUEST_ERROR, | ||
291 | + "Device %s says index %u is used, but it was not available", | ||
292 | + svq->vdev->name, used_elem.id); | ||
293 | + return NULL; | ||
294 | + } | ||
295 | + | ||
296 | + descs[used_elem.id].next = svq->free_head; | ||
297 | + svq->free_head = used_elem.id; | ||
298 | + | ||
299 | + *len = used_elem.len; | ||
300 | + return g_steal_pointer(&svq->ring_id_maps[used_elem.id]); | ||
301 | +} | ||
302 | + | ||
303 | +static void vhost_svq_flush(VhostShadowVirtqueue *svq, | ||
304 | + bool check_for_avail_queue) | ||
305 | +{ | ||
306 | + VirtQueue *vq = svq->vq; | ||
307 | + | ||
308 | + /* Forward as many used buffers as possible. */ | ||
309 | + do { | ||
310 | + unsigned i = 0; | ||
311 | + | ||
312 | + vhost_svq_disable_notification(svq); | ||
313 | + while (true) { | ||
314 | + uint32_t len; | ||
315 | + g_autofree VirtQueueElement *elem = vhost_svq_get_buf(svq, &len); | ||
316 | + if (!elem) { | ||
317 | + break; | ||
318 | + } | ||
319 | + | ||
320 | + if (unlikely(i >= svq->vring.num)) { | ||
321 | + qemu_log_mask(LOG_GUEST_ERROR, | ||
322 | + "More than %u used buffers obtained in a %u size SVQ", | ||
323 | + i, svq->vring.num); | ||
324 | + virtqueue_fill(vq, elem, len, i); | ||
325 | + virtqueue_flush(vq, i); | ||
326 | + return; | ||
327 | + } | ||
328 | + virtqueue_fill(vq, elem, len, i++); | ||
329 | + } | ||
330 | + | ||
331 | + virtqueue_flush(vq, i); | ||
332 | + event_notifier_set(&svq->svq_call); | ||
333 | + | ||
334 | + if (check_for_avail_queue && svq->next_guest_avail_elem) { | ||
335 | + /* | ||
336 | + * Avail ring was full when vhost_svq_flush was called, so it's a | ||
337 | + * good moment to make more descriptors available if possible. | ||
338 | + */ | ||
339 | + vhost_handle_guest_kick(svq); | ||
340 | + } | ||
341 | + } while (!vhost_svq_enable_notification(svq)); | ||
342 | +} | ||
343 | + | ||
344 | +/** | ||
345 | + * Forward used buffers. | ||
346 | * | ||
347 | * @n: hdev call event notifier, the one that device set to notify svq. | ||
348 | + * | ||
349 | + * Note that we are not making any buffers available in the loop, there is no | ||
350 | + * way that it runs more than virtqueue size times. | ||
351 | */ | ||
352 | static void vhost_svq_handle_call(EventNotifier *n) | ||
353 | { | ||
354 | VhostShadowVirtqueue *svq = container_of(n, VhostShadowVirtqueue, | ||
355 | hdev_call); | ||
356 | event_notifier_test_and_clear(n); | ||
357 | - event_notifier_set(&svq->svq_call); | ||
358 | + vhost_svq_flush(svq, true); | ||
359 | } | ||
360 | |||
361 | /** | ||
362 | @@ -XXX,XX +XXX,XX @@ void vhost_svq_set_svq_kick_fd(VhostShadowVirtqueue *svq, int svq_kick_fd) | ||
363 | if (poll_start) { | ||
364 | event_notifier_init_fd(svq_kick, svq_kick_fd); | ||
365 | event_notifier_set(svq_kick); | ||
366 | - event_notifier_set_handler(svq_kick, vhost_handle_guest_kick); | ||
367 | + event_notifier_set_handler(svq_kick, vhost_handle_guest_kick_notifier); | ||
368 | + } | ||
369 | +} | ||
370 | + | ||
371 | +/** | ||
372 | + * Start the shadow virtqueue operation. | ||
373 | + * | ||
374 | + * @svq: Shadow Virtqueue | ||
375 | + * @vdev: VirtIO device | ||
376 | + * @vq: Virtqueue to shadow | ||
377 | + */ | ||
378 | +void vhost_svq_start(VhostShadowVirtqueue *svq, VirtIODevice *vdev, | ||
379 | + VirtQueue *vq) | ||
380 | +{ | ||
381 | + size_t desc_size, driver_size, device_size; | ||
382 | + | ||
383 | + svq->next_guest_avail_elem = NULL; | ||
384 | + svq->shadow_avail_idx = 0; | ||
385 | + svq->shadow_used_idx = 0; | ||
386 | + svq->last_used_idx = 0; | ||
387 | + svq->vdev = vdev; | ||
388 | + svq->vq = vq; | ||
389 | + | ||
390 | + svq->vring.num = virtio_queue_get_num(vdev, virtio_get_queue_index(vq)); | ||
391 | + driver_size = vhost_svq_driver_area_size(svq); | ||
392 | + device_size = vhost_svq_device_area_size(svq); | ||
393 | + svq->vring.desc = qemu_memalign(qemu_real_host_page_size, driver_size); | ||
394 | + desc_size = sizeof(vring_desc_t) * svq->vring.num; | ||
395 | + svq->vring.avail = (void *)((char *)svq->vring.desc + desc_size); | ||
396 | + memset(svq->vring.desc, 0, driver_size); | ||
397 | + svq->vring.used = qemu_memalign(qemu_real_host_page_size, device_size); | ||
398 | + memset(svq->vring.used, 0, device_size); | ||
399 | + svq->ring_id_maps = g_new0(VirtQueueElement *, svq->vring.num); | ||
400 | + for (unsigned i = 0; i < svq->vring.num - 1; i++) { | ||
401 | + svq->vring.desc[i].next = cpu_to_le16(i + 1); | ||
402 | } | ||
403 | } | ||
404 | |||
405 | @@ -XXX,XX +XXX,XX @@ void vhost_svq_set_svq_kick_fd(VhostShadowVirtqueue *svq, int svq_kick_fd) | ||
406 | void vhost_svq_stop(VhostShadowVirtqueue *svq) | ||
407 | { | ||
408 | event_notifier_set_handler(&svq->svq_kick, NULL); | ||
409 | + g_autofree VirtQueueElement *next_avail_elem = NULL; | ||
410 | + | ||
411 | + if (!svq->vq) { | ||
412 | + return; | ||
413 | + } | ||
414 | + | ||
415 | + /* Send all pending used descriptors to guest */ | ||
416 | + vhost_svq_flush(svq, false); | ||
417 | + | ||
418 | + for (unsigned i = 0; i < svq->vring.num; ++i) { | ||
419 | + g_autofree VirtQueueElement *elem = NULL; | ||
420 | + elem = g_steal_pointer(&svq->ring_id_maps[i]); | ||
421 | + if (elem) { | ||
422 | + virtqueue_detach_element(svq->vq, elem, 0); | ||
423 | + } | ||
424 | + } | ||
425 | + | ||
426 | + next_avail_elem = g_steal_pointer(&svq->next_guest_avail_elem); | ||
427 | + if (next_avail_elem) { | ||
428 | + virtqueue_detach_element(svq->vq, next_avail_elem, 0); | ||
429 | + } | ||
430 | + svq->vq = NULL; | ||
431 | + g_free(svq->ring_id_maps); | ||
432 | + qemu_vfree(svq->vring.desc); | ||
433 | + qemu_vfree(svq->vring.used); | ||
434 | } | ||
435 | |||
436 | /** | ||
437 | diff --git a/hw/virtio/vhost-shadow-virtqueue.h b/hw/virtio/vhost-shadow-virtqueue.h | ||
47 | index XXXXXXX..XXXXXXX 100644 | 438 | index XXXXXXX..XXXXXXX 100644 |
48 | --- a/net/colo-compare.c | 439 | --- a/hw/virtio/vhost-shadow-virtqueue.h |
49 | +++ b/net/colo-compare.c | 440 | +++ b/hw/virtio/vhost-shadow-virtqueue.h |
50 | @@ -XXX,XX +XXX,XX @@ static void colo_compare_complete(UserCreatable *uc, Error **errp) | 441 | @@ -XXX,XX +XXX,XX @@ typedef struct VhostShadowVirtqueue { |
51 | return; | 442 | |
443 | /* Guest's call notifier, where the SVQ calls guest. */ | ||
444 | EventNotifier svq_call; | ||
445 | + | ||
446 | + /* Virtio queue shadowing */ | ||
447 | + VirtQueue *vq; | ||
448 | + | ||
449 | + /* Virtio device */ | ||
450 | + VirtIODevice *vdev; | ||
451 | + | ||
452 | + /* Map for use the guest's descriptors */ | ||
453 | + VirtQueueElement **ring_id_maps; | ||
454 | + | ||
455 | + /* Next VirtQueue element that guest made available */ | ||
456 | + VirtQueueElement *next_guest_avail_elem; | ||
457 | + | ||
458 | + /* Next head to expose to the device */ | ||
459 | + uint16_t shadow_avail_idx; | ||
460 | + | ||
461 | + /* Next free descriptor */ | ||
462 | + uint16_t free_head; | ||
463 | + | ||
464 | + /* Last seen used idx */ | ||
465 | + uint16_t shadow_used_idx; | ||
466 | + | ||
467 | + /* Next head to consume from the device */ | ||
468 | + uint16_t last_used_idx; | ||
469 | } VhostShadowVirtqueue; | ||
470 | |||
471 | bool vhost_svq_valid_features(uint64_t features, Error **errp); | ||
472 | @@ -XXX,XX +XXX,XX @@ void vhost_svq_get_vring_addr(const VhostShadowVirtqueue *svq, | ||
473 | size_t vhost_svq_driver_area_size(const VhostShadowVirtqueue *svq); | ||
474 | size_t vhost_svq_device_area_size(const VhostShadowVirtqueue *svq); | ||
475 | |||
476 | +void vhost_svq_start(VhostShadowVirtqueue *svq, VirtIODevice *vdev, | ||
477 | + VirtQueue *vq); | ||
478 | void vhost_svq_stop(VhostShadowVirtqueue *svq); | ||
479 | |||
480 | VhostShadowVirtqueue *vhost_svq_new(void); | ||
481 | diff --git a/hw/virtio/vhost-vdpa.c b/hw/virtio/vhost-vdpa.c | ||
482 | index XXXXXXX..XXXXXXX 100644 | ||
483 | --- a/hw/virtio/vhost-vdpa.c | ||
484 | +++ b/hw/virtio/vhost-vdpa.c | ||
485 | @@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_set_vring_dev_addr(struct vhost_dev *dev, | ||
486 | * Note that this function does not rewind kick file descriptor if cannot set | ||
487 | * call one. | ||
488 | */ | ||
489 | -static bool vhost_vdpa_svq_setup(struct vhost_dev *dev, | ||
490 | - VhostShadowVirtqueue *svq, | ||
491 | - unsigned idx, | ||
492 | - Error **errp) | ||
493 | +static int vhost_vdpa_svq_set_fds(struct vhost_dev *dev, | ||
494 | + VhostShadowVirtqueue *svq, | ||
495 | + unsigned idx, | ||
496 | + Error **errp) | ||
497 | { | ||
498 | struct vhost_vring_file file = { | ||
499 | .index = dev->vq_index + idx, | ||
500 | @@ -XXX,XX +XXX,XX @@ static bool vhost_vdpa_svq_setup(struct vhost_dev *dev, | ||
501 | r = vhost_vdpa_set_vring_dev_kick(dev, &file); | ||
502 | if (unlikely(r != 0)) { | ||
503 | error_setg_errno(errp, -r, "Can't set device kick fd"); | ||
504 | - return false; | ||
505 | + return r; | ||
52 | } | 506 | } |
53 | 507 | ||
54 | - net_socket_rs_init(&s->pri_rs, compare_pri_rs_finalize); | 508 | event_notifier = &svq->hdev_call; |
55 | - net_socket_rs_init(&s->sec_rs, compare_sec_rs_finalize); | 509 | @@ -XXX,XX +XXX,XX @@ static bool vhost_vdpa_svq_setup(struct vhost_dev *dev, |
56 | + net_socket_rs_init(&s->pri_rs, compare_pri_rs_finalize, false); | 510 | error_setg_errno(errp, -r, "Can't set device call fd"); |
57 | + net_socket_rs_init(&s->sec_rs, compare_sec_rs_finalize, false); | 511 | } |
58 | 512 | ||
59 | g_queue_init(&s->conn_list); | 513 | + return r; |
60 | 514 | +} | |
61 | diff --git a/net/filter-mirror.c b/net/filter-mirror.c | 515 | + |
62 | index XXXXXXX..XXXXXXX 100644 | 516 | +/** |
63 | --- a/net/filter-mirror.c | 517 | + * Unmap a SVQ area in the device |
64 | +++ b/net/filter-mirror.c | 518 | + */ |
65 | @@ -XXX,XX +XXX,XX @@ static void filter_redirector_setup(NetFilterState *nf, Error **errp) | 519 | +static bool vhost_vdpa_svq_unmap_ring(struct vhost_vdpa *v, hwaddr iova, |
520 | + hwaddr size) | ||
521 | +{ | ||
522 | + int r; | ||
523 | + | ||
524 | + size = ROUND_UP(size, qemu_real_host_page_size); | ||
525 | + r = vhost_vdpa_dma_unmap(v, iova, size); | ||
526 | + return r == 0; | ||
527 | +} | ||
528 | + | ||
529 | +static bool vhost_vdpa_svq_unmap_rings(struct vhost_dev *dev, | ||
530 | + const VhostShadowVirtqueue *svq) | ||
531 | +{ | ||
532 | + struct vhost_vdpa *v = dev->opaque; | ||
533 | + struct vhost_vring_addr svq_addr; | ||
534 | + size_t device_size = vhost_svq_device_area_size(svq); | ||
535 | + size_t driver_size = vhost_svq_driver_area_size(svq); | ||
536 | + bool ok; | ||
537 | + | ||
538 | + vhost_svq_get_vring_addr(svq, &svq_addr); | ||
539 | + | ||
540 | + ok = vhost_vdpa_svq_unmap_ring(v, svq_addr.desc_user_addr, driver_size); | ||
541 | + if (unlikely(!ok)) { | ||
542 | + return false; | ||
543 | + } | ||
544 | + | ||
545 | + return vhost_vdpa_svq_unmap_ring(v, svq_addr.used_user_addr, device_size); | ||
546 | +} | ||
547 | + | ||
548 | +/** | ||
549 | + * Map the shadow virtqueue rings in the device | ||
550 | + * | ||
551 | + * @dev: The vhost device | ||
552 | + * @svq: The shadow virtqueue | ||
553 | + * @addr: Assigned IOVA addresses | ||
554 | + * @errp: Error pointer | ||
555 | + */ | ||
556 | +static bool vhost_vdpa_svq_map_rings(struct vhost_dev *dev, | ||
557 | + const VhostShadowVirtqueue *svq, | ||
558 | + struct vhost_vring_addr *addr, | ||
559 | + Error **errp) | ||
560 | +{ | ||
561 | + struct vhost_vdpa *v = dev->opaque; | ||
562 | + size_t device_size = vhost_svq_device_area_size(svq); | ||
563 | + size_t driver_size = vhost_svq_driver_area_size(svq); | ||
564 | + int r; | ||
565 | + | ||
566 | + ERRP_GUARD(); | ||
567 | + vhost_svq_get_vring_addr(svq, addr); | ||
568 | + | ||
569 | + r = vhost_vdpa_dma_map(v, addr->desc_user_addr, driver_size, | ||
570 | + (void *)addr->desc_user_addr, true); | ||
571 | + if (unlikely(r != 0)) { | ||
572 | + error_setg_errno(errp, -r, "Cannot create vq driver region: "); | ||
573 | + return false; | ||
574 | + } | ||
575 | + | ||
576 | + r = vhost_vdpa_dma_map(v, addr->used_user_addr, device_size, | ||
577 | + (void *)addr->used_user_addr, false); | ||
578 | + if (unlikely(r != 0)) { | ||
579 | + error_setg_errno(errp, -r, "Cannot create vq device region: "); | ||
580 | + } | ||
581 | + | ||
582 | + return r == 0; | ||
583 | +} | ||
584 | + | ||
585 | +static bool vhost_vdpa_svq_setup(struct vhost_dev *dev, | ||
586 | + VhostShadowVirtqueue *svq, | ||
587 | + unsigned idx, | ||
588 | + Error **errp) | ||
589 | +{ | ||
590 | + uint16_t vq_index = dev->vq_index + idx; | ||
591 | + struct vhost_vring_state s = { | ||
592 | + .index = vq_index, | ||
593 | + }; | ||
594 | + int r; | ||
595 | + | ||
596 | + r = vhost_vdpa_set_dev_vring_base(dev, &s); | ||
597 | + if (unlikely(r)) { | ||
598 | + error_setg_errno(errp, -r, "Cannot set vring base"); | ||
599 | + return false; | ||
600 | + } | ||
601 | + | ||
602 | + r = vhost_vdpa_svq_set_fds(dev, svq, idx, errp); | ||
603 | return r == 0; | ||
604 | } | ||
605 | |||
606 | @@ -XXX,XX +XXX,XX @@ static bool vhost_vdpa_svqs_start(struct vhost_dev *dev) | ||
607 | } | ||
608 | |||
609 | for (i = 0; i < v->shadow_vqs->len; ++i) { | ||
610 | + VirtQueue *vq = virtio_get_queue(dev->vdev, dev->vq_index + i); | ||
611 | VhostShadowVirtqueue *svq = g_ptr_array_index(v->shadow_vqs, i); | ||
612 | + struct vhost_vring_addr addr = { | ||
613 | + .index = i, | ||
614 | + }; | ||
615 | + int r; | ||
616 | bool ok = vhost_vdpa_svq_setup(dev, svq, i, &err); | ||
617 | if (unlikely(!ok)) { | ||
618 | - error_reportf_err(err, "Cannot setup SVQ %u: ", i); | ||
619 | + goto err; | ||
620 | + } | ||
621 | + | ||
622 | + vhost_svq_start(svq, dev->vdev, vq); | ||
623 | + ok = vhost_vdpa_svq_map_rings(dev, svq, &addr, &err); | ||
624 | + if (unlikely(!ok)) { | ||
625 | + goto err_map; | ||
626 | + } | ||
627 | + | ||
628 | + /* Override vring GPA set by vhost subsystem */ | ||
629 | + r = vhost_vdpa_set_vring_dev_addr(dev, &addr); | ||
630 | + if (unlikely(r != 0)) { | ||
631 | + error_setg_errno(&err, -r, "Cannot set device address"); | ||
632 | + goto err_set_addr; | ||
633 | + } | ||
634 | + } | ||
635 | + | ||
636 | + return true; | ||
637 | + | ||
638 | +err_set_addr: | ||
639 | + vhost_vdpa_svq_unmap_rings(dev, g_ptr_array_index(v->shadow_vqs, i)); | ||
640 | + | ||
641 | +err_map: | ||
642 | + vhost_svq_stop(g_ptr_array_index(v->shadow_vqs, i)); | ||
643 | + | ||
644 | +err: | ||
645 | + error_reportf_err(err, "Cannot setup SVQ %u: ", i); | ||
646 | + for (unsigned j = 0; j < i; ++j) { | ||
647 | + VhostShadowVirtqueue *svq = g_ptr_array_index(v->shadow_vqs, j); | ||
648 | + vhost_vdpa_svq_unmap_rings(dev, svq); | ||
649 | + vhost_svq_stop(svq); | ||
650 | + } | ||
651 | + | ||
652 | + return false; | ||
653 | +} | ||
654 | + | ||
655 | +static bool vhost_vdpa_svqs_stop(struct vhost_dev *dev) | ||
656 | +{ | ||
657 | + struct vhost_vdpa *v = dev->opaque; | ||
658 | + | ||
659 | + if (!v->shadow_vqs) { | ||
660 | + return true; | ||
661 | + } | ||
662 | + | ||
663 | + for (unsigned i = 0; i < v->shadow_vqs->len; ++i) { | ||
664 | + VhostShadowVirtqueue *svq = g_ptr_array_index(v->shadow_vqs, | ||
665 | + i); | ||
666 | + bool ok = vhost_vdpa_svq_unmap_rings(dev, svq); | ||
667 | + if (unlikely(!ok)) { | ||
668 | return false; | ||
66 | } | 669 | } |
67 | } | 670 | } |
68 | 671 | @@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_dev_start(struct vhost_dev *dev, bool started) | |
69 | - net_socket_rs_init(&s->rs, redirector_rs_finalize); | 672 | } |
70 | + net_socket_rs_init(&s->rs, redirector_rs_finalize, false); | 673 | vhost_vdpa_set_vring_ready(dev); |
71 | 674 | } else { | |
72 | if (s->indev) { | 675 | + ok = vhost_vdpa_svqs_stop(dev); |
73 | chr = qemu_chr_find(s->indev); | 676 | + if (unlikely(!ok)) { |
74 | diff --git a/net/net.c b/net/net.c | 677 | + return -1; |
75 | index XXXXXXX..XXXXXXX 100644 | 678 | + } |
76 | --- a/net/net.c | 679 | vhost_vdpa_host_notifiers_uninit(dev, dev->nvqs); |
77 | +++ b/net/net.c | 680 | } |
78 | @@ -XXX,XX +XXX,XX @@ QemuOptsList qemu_net_opts = { | 681 | |
79 | }; | ||
80 | |||
81 | void net_socket_rs_init(SocketReadState *rs, | ||
82 | - SocketReadStateFinalize *finalize) | ||
83 | + SocketReadStateFinalize *finalize, | ||
84 | + bool vnet_hdr) | ||
85 | { | ||
86 | rs->state = 0; | ||
87 | + rs->vnet_hdr = vnet_hdr; | ||
88 | rs->index = 0; | ||
89 | rs->packet_len = 0; | ||
90 | + rs->vnet_hdr_len = 0; | ||
91 | memset(rs->buf, 0, sizeof(rs->buf)); | ||
92 | rs->finalize = finalize; | ||
93 | } | ||
94 | @@ -XXX,XX +XXX,XX @@ int net_fill_rstate(SocketReadState *rs, const uint8_t *buf, int size) | ||
95 | unsigned int l; | ||
96 | |||
97 | while (size > 0) { | ||
98 | - /* reassemble a packet from the network */ | ||
99 | - switch (rs->state) { /* 0 = getting length, 1 = getting data */ | ||
100 | + /* Reassemble a packet from the network. | ||
101 | + * 0 = getting length. | ||
102 | + * 1 = getting vnet header length. | ||
103 | + * 2 = getting data. | ||
104 | + */ | ||
105 | + switch (rs->state) { | ||
106 | case 0: | ||
107 | l = 4 - rs->index; | ||
108 | if (l > size) { | ||
109 | @@ -XXX,XX +XXX,XX @@ int net_fill_rstate(SocketReadState *rs, const uint8_t *buf, int size) | ||
110 | /* got length */ | ||
111 | rs->packet_len = ntohl(*(uint32_t *)rs->buf); | ||
112 | rs->index = 0; | ||
113 | - rs->state = 1; | ||
114 | + if (rs->vnet_hdr) { | ||
115 | + rs->state = 1; | ||
116 | + } else { | ||
117 | + rs->state = 2; | ||
118 | + rs->vnet_hdr_len = 0; | ||
119 | + } | ||
120 | } | ||
121 | break; | ||
122 | case 1: | ||
123 | + l = 4 - rs->index; | ||
124 | + if (l > size) { | ||
125 | + l = size; | ||
126 | + } | ||
127 | + memcpy(rs->buf + rs->index, buf, l); | ||
128 | + buf += l; | ||
129 | + size -= l; | ||
130 | + rs->index += l; | ||
131 | + if (rs->index == 4) { | ||
132 | + /* got vnet header length */ | ||
133 | + rs->vnet_hdr_len = ntohl(*(uint32_t *)rs->buf); | ||
134 | + rs->index = 0; | ||
135 | + rs->state = 2; | ||
136 | + } | ||
137 | + break; | ||
138 | + case 2: | ||
139 | l = rs->packet_len - rs->index; | ||
140 | if (l > size) { | ||
141 | l = size; | ||
142 | diff --git a/net/socket.c b/net/socket.c | ||
143 | index XXXXXXX..XXXXXXX 100644 | ||
144 | --- a/net/socket.c | ||
145 | +++ b/net/socket.c | ||
146 | @@ -XXX,XX +XXX,XX @@ static void net_socket_send(void *opaque) | ||
147 | closesocket(s->fd); | ||
148 | |||
149 | s->fd = -1; | ||
150 | - net_socket_rs_init(&s->rs, net_socket_rs_finalize); | ||
151 | + net_socket_rs_init(&s->rs, net_socket_rs_finalize, false); | ||
152 | s->nc.link_down = true; | ||
153 | memset(s->nc.info_str, 0, sizeof(s->nc.info_str)); | ||
154 | |||
155 | @@ -XXX,XX +XXX,XX @@ static NetSocketState *net_socket_fd_init_dgram(NetClientState *peer, | ||
156 | s->fd = fd; | ||
157 | s->listen_fd = -1; | ||
158 | s->send_fn = net_socket_send_dgram; | ||
159 | - net_socket_rs_init(&s->rs, net_socket_rs_finalize); | ||
160 | + net_socket_rs_init(&s->rs, net_socket_rs_finalize, false); | ||
161 | net_socket_read_poll(s, true); | ||
162 | |||
163 | /* mcast: save bound address as dst */ | ||
164 | @@ -XXX,XX +XXX,XX @@ static NetSocketState *net_socket_fd_init_stream(NetClientState *peer, | ||
165 | |||
166 | s->fd = fd; | ||
167 | s->listen_fd = -1; | ||
168 | - net_socket_rs_init(&s->rs, net_socket_rs_finalize); | ||
169 | + net_socket_rs_init(&s->rs, net_socket_rs_finalize, false); | ||
170 | |||
171 | /* Disable Nagle algorithm on TCP sockets to reduce latency */ | ||
172 | socket_set_nodelay(fd); | ||
173 | @@ -XXX,XX +XXX,XX @@ static int net_socket_listen_init(NetClientState *peer, | ||
174 | s->fd = -1; | ||
175 | s->listen_fd = fd; | ||
176 | s->nc.link_down = true; | ||
177 | - net_socket_rs_init(&s->rs, net_socket_rs_finalize); | ||
178 | + net_socket_rs_init(&s->rs, net_socket_rs_finalize, false); | ||
179 | |||
180 | qemu_set_fd_handler(s->listen_fd, net_socket_accept, NULL, s); | ||
181 | return 0; | ||
182 | -- | 682 | -- |
183 | 2.7.4 | 683 | 2.7.4 |
184 | 684 | ||
185 | 685 | diff view generated by jsdifflib |
1 | From: Zhang Chen <zhangchen.fnst@cn.fujitsu.com> | 1 | From: Eugenio Pérez <eperezma@redhat.com> |
---|---|---|---|
2 | 2 | ||
3 | Add vnet_hdr_len arguments in NetClientState | 3 | This iova tree function allows it to look for a hole in allocated |
4 | that make other module get real vnet_hdr_len easily. | 4 | regions and return a totally new translation for a given translated |
5 | 5 | address. | |
6 | Signed-off-by: Zhang Chen <zhangchen.fnst@cn.fujitsu.com> | 6 | |
7 | It's usage is mainly to allow devices to access qemu address space, | ||
8 | remapping guest's one into a new iova space where qemu can add chunks of | ||
9 | addresses. | ||
10 | |||
11 | Acked-by: Michael S. Tsirkin <mst@redhat.com> | ||
12 | Signed-off-by: Eugenio Pérez <eperezma@redhat.com> | ||
13 | Reviewed-by: Peter Xu <peterx@redhat.com> | ||
7 | Signed-off-by: Jason Wang <jasowang@redhat.com> | 14 | Signed-off-by: Jason Wang <jasowang@redhat.com> |
8 | --- | 15 | --- |
9 | include/net/net.h | 1 + | 16 | include/qemu/iova-tree.h | 18 +++++++ |
10 | net/net.c | 1 + | 17 | util/iova-tree.c | 135 +++++++++++++++++++++++++++++++++++++++++++++++ |
11 | 2 files changed, 2 insertions(+) | 18 | 2 files changed, 153 insertions(+) |
12 | 19 | ||
13 | diff --git a/include/net/net.h b/include/net/net.h | 20 | diff --git a/include/qemu/iova-tree.h b/include/qemu/iova-tree.h |
14 | index XXXXXXX..XXXXXXX 100644 | 21 | index XXXXXXX..XXXXXXX 100644 |
15 | --- a/include/net/net.h | 22 | --- a/include/qemu/iova-tree.h |
16 | +++ b/include/net/net.h | 23 | +++ b/include/qemu/iova-tree.h |
17 | @@ -XXX,XX +XXX,XX @@ struct NetClientState { | 24 | @@ -XXX,XX +XXX,XX @@ |
18 | unsigned int queue_index; | 25 | #define IOVA_OK (0) |
19 | unsigned rxfilter_notify_enabled:1; | 26 | #define IOVA_ERR_INVALID (-1) /* Invalid parameters */ |
20 | int vring_enable; | 27 | #define IOVA_ERR_OVERLAP (-2) /* IOVA range overlapped */ |
21 | + int vnet_hdr_len; | 28 | +#define IOVA_ERR_NOMEM (-3) /* Cannot allocate */ |
22 | QTAILQ_HEAD(NetFilterHead, NetFilterState) filters; | 29 | |
30 | typedef struct IOVATree IOVATree; | ||
31 | typedef struct DMAMap { | ||
32 | @@ -XXX,XX +XXX,XX @@ const DMAMap *iova_tree_find_address(const IOVATree *tree, hwaddr iova); | ||
33 | void iova_tree_foreach(IOVATree *tree, iova_tree_iterator iterator); | ||
34 | |||
35 | /** | ||
36 | + * iova_tree_alloc_map: | ||
37 | + * | ||
38 | + * @tree: the iova tree to allocate from | ||
39 | + * @map: the new map (as translated addr & size) to allocate in the iova region | ||
40 | + * @iova_begin: the minimum address of the allocation | ||
41 | + * @iova_end: the maximum addressable direction of the allocation | ||
42 | + * | ||
43 | + * Allocates a new region of a given size, between iova_min and iova_max. | ||
44 | + * | ||
45 | + * Return: Same as iova_tree_insert, but cannot overlap and can return error if | ||
46 | + * iova tree is out of free contiguous range. The caller gets the assigned iova | ||
47 | + * in map->iova. | ||
48 | + */ | ||
49 | +int iova_tree_alloc_map(IOVATree *tree, DMAMap *map, hwaddr iova_begin, | ||
50 | + hwaddr iova_end); | ||
51 | + | ||
52 | +/** | ||
53 | * iova_tree_destroy: | ||
54 | * | ||
55 | * @tree: the iova tree to destroy | ||
56 | diff --git a/util/iova-tree.c b/util/iova-tree.c | ||
57 | index XXXXXXX..XXXXXXX 100644 | ||
58 | --- a/util/iova-tree.c | ||
59 | +++ b/util/iova-tree.c | ||
60 | @@ -XXX,XX +XXX,XX @@ struct IOVATree { | ||
61 | GTree *tree; | ||
23 | }; | 62 | }; |
24 | 63 | ||
25 | diff --git a/net/net.c b/net/net.c | 64 | +/* Args to pass to iova_tree_alloc foreach function. */ |
26 | index XXXXXXX..XXXXXXX 100644 | 65 | +struct IOVATreeAllocArgs { |
27 | --- a/net/net.c | 66 | + /* Size of the desired allocation */ |
28 | +++ b/net/net.c | 67 | + size_t new_size; |
29 | @@ -XXX,XX +XXX,XX @@ void qemu_set_vnet_hdr_len(NetClientState *nc, int len) | 68 | + |
30 | return; | 69 | + /* The minimum address allowed in the allocation */ |
31 | } | 70 | + hwaddr iova_begin; |
32 | 71 | + | |
33 | + nc->vnet_hdr_len = len; | 72 | + /* Map at the left of the hole, can be NULL if "this" is first one */ |
34 | nc->info->set_vnet_hdr_len(nc, len); | 73 | + const DMAMap *prev; |
74 | + | ||
75 | + /* Map at the right of the hole, can be NULL if "prev" is the last one */ | ||
76 | + const DMAMap *this; | ||
77 | + | ||
78 | + /* If found, we fill in the IOVA here */ | ||
79 | + hwaddr iova_result; | ||
80 | + | ||
81 | + /* Whether have we found a valid IOVA */ | ||
82 | + bool iova_found; | ||
83 | +}; | ||
84 | + | ||
85 | +/** | ||
86 | + * Iterate args to the next hole | ||
87 | + * | ||
88 | + * @args: The alloc arguments | ||
89 | + * @next: The next mapping in the tree. Can be NULL to signal the last one | ||
90 | + */ | ||
91 | +static void iova_tree_alloc_args_iterate(struct IOVATreeAllocArgs *args, | ||
92 | + const DMAMap *next) { | ||
93 | + args->prev = args->this; | ||
94 | + args->this = next; | ||
95 | +} | ||
96 | + | ||
97 | static int iova_tree_compare(gconstpointer a, gconstpointer b, gpointer data) | ||
98 | { | ||
99 | const DMAMap *m1 = a, *m2 = b; | ||
100 | @@ -XXX,XX +XXX,XX @@ int iova_tree_remove(IOVATree *tree, const DMAMap *map) | ||
101 | return IOVA_OK; | ||
35 | } | 102 | } |
36 | 103 | ||
104 | +/** | ||
105 | + * Try to find an unallocated IOVA range between prev and this elements. | ||
106 | + * | ||
107 | + * @args: Arguments to allocation | ||
108 | + * | ||
109 | + * Cases: | ||
110 | + * | ||
111 | + * (1) !prev, !this: No entries allocated, always succeed | ||
112 | + * | ||
113 | + * (2) !prev, this: We're iterating at the 1st element. | ||
114 | + * | ||
115 | + * (3) prev, !this: We're iterating at the last element. | ||
116 | + * | ||
117 | + * (4) prev, this: this is the most common case, we'll try to find a hole | ||
118 | + * between "prev" and "this" mapping. | ||
119 | + * | ||
120 | + * Note that this function assumes the last valid iova is HWADDR_MAX, but it | ||
121 | + * searches linearly so it's easy to discard the result if it's not the case. | ||
122 | + */ | ||
123 | +static void iova_tree_alloc_map_in_hole(struct IOVATreeAllocArgs *args) | ||
124 | +{ | ||
125 | + const DMAMap *prev = args->prev, *this = args->this; | ||
126 | + uint64_t hole_start, hole_last; | ||
127 | + | ||
128 | + if (this && this->iova + this->size < args->iova_begin) { | ||
129 | + return; | ||
130 | + } | ||
131 | + | ||
132 | + hole_start = MAX(prev ? prev->iova + prev->size + 1 : 0, args->iova_begin); | ||
133 | + hole_last = this ? this->iova : HWADDR_MAX; | ||
134 | + | ||
135 | + if (hole_last - hole_start > args->new_size) { | ||
136 | + args->iova_result = hole_start; | ||
137 | + args->iova_found = true; | ||
138 | + } | ||
139 | +} | ||
140 | + | ||
141 | +/** | ||
142 | + * Foreach dma node in the tree, compare if there is a hole with its previous | ||
143 | + * node (or minimum iova address allowed) and the node. | ||
144 | + * | ||
145 | + * @key: Node iterating | ||
146 | + * @value: Node iterating | ||
147 | + * @pargs: Struct to communicate with the outside world | ||
148 | + * | ||
149 | + * Return: false to keep iterating, true if needs break. | ||
150 | + */ | ||
151 | +static gboolean iova_tree_alloc_traverse(gpointer key, gpointer value, | ||
152 | + gpointer pargs) | ||
153 | +{ | ||
154 | + struct IOVATreeAllocArgs *args = pargs; | ||
155 | + DMAMap *node = value; | ||
156 | + | ||
157 | + assert(key == value); | ||
158 | + | ||
159 | + iova_tree_alloc_args_iterate(args, node); | ||
160 | + iova_tree_alloc_map_in_hole(args); | ||
161 | + return args->iova_found; | ||
162 | +} | ||
163 | + | ||
164 | +int iova_tree_alloc_map(IOVATree *tree, DMAMap *map, hwaddr iova_begin, | ||
165 | + hwaddr iova_last) | ||
166 | +{ | ||
167 | + struct IOVATreeAllocArgs args = { | ||
168 | + .new_size = map->size, | ||
169 | + .iova_begin = iova_begin, | ||
170 | + }; | ||
171 | + | ||
172 | + if (unlikely(iova_last < iova_begin)) { | ||
173 | + return IOVA_ERR_INVALID; | ||
174 | + } | ||
175 | + | ||
176 | + /* | ||
177 | + * Find a valid hole for the mapping | ||
178 | + * | ||
179 | + * Assuming low iova_begin, so no need to do a binary search to | ||
180 | + * locate the first node. | ||
181 | + * | ||
182 | + * TODO: Replace all this with g_tree_node_first/next/last when available | ||
183 | + * (from glib since 2.68). To do it with g_tree_foreach complicates the | ||
184 | + * code a lot. | ||
185 | + * | ||
186 | + */ | ||
187 | + g_tree_foreach(tree->tree, iova_tree_alloc_traverse, &args); | ||
188 | + if (!args.iova_found) { | ||
189 | + /* | ||
190 | + * Either tree is empty or the last hole is still not checked. | ||
191 | + * g_tree_foreach does not compare (last, iova_last] range, so we check | ||
192 | + * it here. | ||
193 | + */ | ||
194 | + iova_tree_alloc_args_iterate(&args, NULL); | ||
195 | + iova_tree_alloc_map_in_hole(&args); | ||
196 | + } | ||
197 | + | ||
198 | + if (!args.iova_found || args.iova_result + map->size > iova_last) { | ||
199 | + return IOVA_ERR_NOMEM; | ||
200 | + } | ||
201 | + | ||
202 | + map->iova = args.iova_result; | ||
203 | + return iova_tree_insert(tree, map); | ||
204 | +} | ||
205 | + | ||
206 | void iova_tree_destroy(IOVATree *tree) | ||
207 | { | ||
208 | g_tree_destroy(tree->tree); | ||
37 | -- | 209 | -- |
38 | 2.7.4 | 210 | 2.7.4 |
39 | 211 | ||
40 | 212 | diff view generated by jsdifflib |
1 | From: Zhang Chen <zhangchen.fnst@cn.fujitsu.com> | 1 | From: Eugenio Pérez <eperezma@redhat.com> |
---|---|---|---|
2 | 2 | ||
3 | We add the vnet_hdr_support option for filter-mirror, default is disabled. | 3 | This function does the reverse operation of iova_tree_find: To look for |
4 | If you use virtio-net-pci or other driver needs vnet_hdr, please enable it. | 4 | a mapping that match a translated address so we can do the reverse. |
5 | You can use it for example: | ||
6 | -object filter-mirror,id=m0,netdev=hn0,queue=tx,outdev=mirror0,vnet_hdr_support | ||
7 | 5 | ||
8 | If it has vnet_hdr_support flag, we will change the sending packet format from | 6 | This have linear complexity instead of logarithmic, but it supports |
9 | struct {int size; const uint8_t buf[];} to {int size; int vnet_hdr_len; const uint8_t buf[];}. | 7 | overlapping HVA. Future developments could reduce it. |
10 | make other module(like colo-compare) know how to parse net packet correctly. | ||
11 | 8 | ||
12 | Signed-off-by: Zhang Chen <zhangchen.fnst@cn.fujitsu.com> | 9 | Acked-by: Michael S. Tsirkin <mst@redhat.com> |
10 | Signed-off-by: Eugenio Pérez <eperezma@redhat.com> | ||
13 | Signed-off-by: Jason Wang <jasowang@redhat.com> | 11 | Signed-off-by: Jason Wang <jasowang@redhat.com> |
14 | --- | 12 | --- |
15 | net/filter-mirror.c | 42 +++++++++++++++++++++++++++++++++++++++++- | 13 | include/qemu/iova-tree.h | 20 +++++++++++++++++++- |
16 | qemu-options.hx | 5 ++--- | 14 | util/iova-tree.c | 34 ++++++++++++++++++++++++++++++++++ |
17 | 2 files changed, 43 insertions(+), 4 deletions(-) | 15 | 2 files changed, 53 insertions(+), 1 deletion(-) |
18 | 16 | ||
19 | diff --git a/net/filter-mirror.c b/net/filter-mirror.c | 17 | diff --git a/include/qemu/iova-tree.h b/include/qemu/iova-tree.h |
20 | index XXXXXXX..XXXXXXX 100644 | 18 | index XXXXXXX..XXXXXXX 100644 |
21 | --- a/net/filter-mirror.c | 19 | --- a/include/qemu/iova-tree.h |
22 | +++ b/net/filter-mirror.c | 20 | +++ b/include/qemu/iova-tree.h |
23 | @@ -XXX,XX +XXX,XX @@ typedef struct MirrorState { | 21 | @@ -XXX,XX +XXX,XX @@ int iova_tree_remove(IOVATree *tree, const DMAMap *map); |
24 | CharBackend chr_in; | 22 | * @tree: the iova tree to search from |
25 | CharBackend chr_out; | 23 | * @map: the mapping to search |
26 | SocketReadState rs; | 24 | * |
27 | + bool vnet_hdr; | 25 | - * Search for a mapping in the iova tree that overlaps with the |
28 | } MirrorState; | 26 | + * Search for a mapping in the iova tree that iova overlaps with the |
29 | 27 | * mapping range specified. Only the first found mapping will be | |
30 | static int filter_send(MirrorState *s, | 28 | * returned. |
31 | const struct iovec *iov, | 29 | * |
32 | int iovcnt) | 30 | @@ -XXX,XX +XXX,XX @@ int iova_tree_remove(IOVATree *tree, const DMAMap *map); |
33 | { | 31 | const DMAMap *iova_tree_find(const IOVATree *tree, const DMAMap *map); |
34 | + NetFilterState *nf = NETFILTER(s); | 32 | |
35 | int ret = 0; | 33 | /** |
36 | ssize_t size = 0; | 34 | + * iova_tree_find_iova: |
37 | uint32_t len = 0; | 35 | + * |
38 | @@ -XXX,XX +XXX,XX @@ static int filter_send(MirrorState *s, | 36 | + * @tree: the iova tree to search from |
39 | goto err; | 37 | + * @map: the mapping to search |
40 | } | 38 | + * |
41 | 39 | + * Search for a mapping in the iova tree that translated_addr overlaps with the | |
42 | + if (s->vnet_hdr) { | 40 | + * mapping range specified. Only the first found mapping will be |
43 | + /* | 41 | + * returned. |
44 | + * If vnet_hdr = on, we send vnet header len to make other | 42 | + * |
45 | + * module(like colo-compare) know how to parse net | 43 | + * Return: DMAMap pointer if found, or NULL if not found. Note that |
46 | + * packet correctly. | 44 | + * the returned DMAMap pointer is maintained internally. User should |
47 | + */ | 45 | + * only read the content but never modify or free the content. Also, |
48 | + ssize_t vnet_hdr_len; | 46 | + * user is responsible to make sure the pointer is valid (say, no |
47 | + * concurrent deletion in progress). | ||
48 | + */ | ||
49 | +const DMAMap *iova_tree_find_iova(const IOVATree *tree, const DMAMap *map); | ||
49 | + | 50 | + |
50 | + vnet_hdr_len = nf->netdev->vnet_hdr_len; | 51 | +/** |
52 | * iova_tree_find_address: | ||
53 | * | ||
54 | * @tree: the iova tree to search from | ||
55 | diff --git a/util/iova-tree.c b/util/iova-tree.c | ||
56 | index XXXXXXX..XXXXXXX 100644 | ||
57 | --- a/util/iova-tree.c | ||
58 | +++ b/util/iova-tree.c | ||
59 | @@ -XXX,XX +XXX,XX @@ struct IOVATreeAllocArgs { | ||
60 | bool iova_found; | ||
61 | }; | ||
62 | |||
63 | +typedef struct IOVATreeFindIOVAArgs { | ||
64 | + const DMAMap *needle; | ||
65 | + const DMAMap *result; | ||
66 | +} IOVATreeFindIOVAArgs; | ||
51 | + | 67 | + |
52 | + len = htonl(vnet_hdr_len); | 68 | /** |
53 | + ret = qemu_chr_fe_write_all(&s->chr_out, (uint8_t *)&len, sizeof(len)); | 69 | * Iterate args to the next hole |
54 | + if (ret != sizeof(len)) { | 70 | * |
55 | + goto err; | 71 | @@ -XXX,XX +XXX,XX @@ const DMAMap *iova_tree_find(const IOVATree *tree, const DMAMap *map) |
56 | + } | 72 | return g_tree_lookup(tree->tree, map); |
73 | } | ||
74 | |||
75 | +static gboolean iova_tree_find_address_iterator(gpointer key, gpointer value, | ||
76 | + gpointer data) | ||
77 | +{ | ||
78 | + const DMAMap *map = key; | ||
79 | + IOVATreeFindIOVAArgs *args = data; | ||
80 | + const DMAMap *needle; | ||
81 | + | ||
82 | + g_assert(key == value); | ||
83 | + | ||
84 | + needle = args->needle; | ||
85 | + if (map->translated_addr + map->size < needle->translated_addr || | ||
86 | + needle->translated_addr + needle->size < map->translated_addr) { | ||
87 | + return false; | ||
57 | + } | 88 | + } |
58 | + | 89 | + |
59 | buf = g_malloc(size); | 90 | + args->result = map; |
60 | iov_to_buf(iov, iovcnt, 0, buf, size); | 91 | + return true; |
61 | ret = qemu_chr_fe_write_all(&s->chr_out, (uint8_t *)buf, size); | ||
62 | @@ -XXX,XX +XXX,XX @@ static void filter_redirector_setup(NetFilterState *nf, Error **errp) | ||
63 | } | ||
64 | } | ||
65 | |||
66 | - net_socket_rs_init(&s->rs, redirector_rs_finalize, false); | ||
67 | + net_socket_rs_init(&s->rs, redirector_rs_finalize, s->vnet_hdr); | ||
68 | |||
69 | if (s->indev) { | ||
70 | chr = qemu_chr_find(s->indev); | ||
71 | @@ -XXX,XX +XXX,XX @@ static void filter_mirror_set_outdev(Object *obj, | ||
72 | } | ||
73 | } | ||
74 | |||
75 | +static bool filter_mirror_get_vnet_hdr(Object *obj, Error **errp) | ||
76 | +{ | ||
77 | + MirrorState *s = FILTER_MIRROR(obj); | ||
78 | + | ||
79 | + return s->vnet_hdr; | ||
80 | +} | 92 | +} |
81 | + | 93 | + |
82 | +static void filter_mirror_set_vnet_hdr(Object *obj, bool value, Error **errp) | 94 | +const DMAMap *iova_tree_find_iova(const IOVATree *tree, const DMAMap *map) |
83 | +{ | 95 | +{ |
84 | + MirrorState *s = FILTER_MIRROR(obj); | 96 | + IOVATreeFindIOVAArgs args = { |
97 | + .needle = map, | ||
98 | + }; | ||
85 | + | 99 | + |
86 | + s->vnet_hdr = value; | 100 | + g_tree_foreach(tree->tree, iova_tree_find_address_iterator, &args); |
101 | + return args.result; | ||
87 | +} | 102 | +} |
88 | + | 103 | + |
89 | static char *filter_redirector_get_outdev(Object *obj, Error **errp) | 104 | const DMAMap *iova_tree_find_address(const IOVATree *tree, hwaddr iova) |
90 | { | 105 | { |
91 | MirrorState *s = FILTER_REDIRECTOR(obj); | 106 | const DMAMap map = { .iova = iova, .size = 0 }; |
92 | @@ -XXX,XX +XXX,XX @@ static void filter_redirector_set_outdev(Object *obj, | ||
93 | |||
94 | static void filter_mirror_init(Object *obj) | ||
95 | { | ||
96 | + MirrorState *s = FILTER_MIRROR(obj); | ||
97 | + | ||
98 | object_property_add_str(obj, "outdev", filter_mirror_get_outdev, | ||
99 | filter_mirror_set_outdev, NULL); | ||
100 | + | ||
101 | + s->vnet_hdr = false; | ||
102 | + object_property_add_bool(obj, "vnet_hdr_support", | ||
103 | + filter_mirror_get_vnet_hdr, | ||
104 | + filter_mirror_set_vnet_hdr, NULL); | ||
105 | } | ||
106 | |||
107 | static void filter_redirector_init(Object *obj) | ||
108 | diff --git a/qemu-options.hx b/qemu-options.hx | ||
109 | index XXXXXXX..XXXXXXX 100644 | ||
110 | --- a/qemu-options.hx | ||
111 | +++ b/qemu-options.hx | ||
112 | @@ -XXX,XX +XXX,XX @@ queue @var{all|rx|tx} is an option that can be applied to any netfilter. | ||
113 | @option{tx}: the filter is attached to the transmit queue of the netdev, | ||
114 | where it will receive packets sent by the netdev. | ||
115 | |||
116 | -@item -object filter-mirror,id=@var{id},netdev=@var{netdevid},outdev=@var{chardevid}[,queue=@var{all|rx|tx}] | ||
117 | +@item -object filter-mirror,id=@var{id},netdev=@var{netdevid},outdev=@var{chardevid},queue=@var{all|rx|tx}[,vnet_hdr_support] | ||
118 | |||
119 | -filter-mirror on netdev @var{netdevid},mirror net packet to chardev | ||
120 | -@var{chardevid} | ||
121 | +filter-mirror on netdev @var{netdevid},mirror net packet to chardev@var{chardevid}, if it has the vnet_hdr_support flag, filter-mirror will mirror packet with vnet_hdr_len. | ||
122 | |||
123 | @item -object filter-redirector,id=@var{id},netdev=@var{netdevid},indev=@var{chardevid}, | ||
124 | outdev=@var{chardevid}[,queue=@var{all|rx|tx}] | ||
125 | -- | 107 | -- |
126 | 2.7.4 | 108 | 2.7.4 |
127 | 109 | ||
128 | 110 | diff view generated by jsdifflib |
1 | From: Zhang Chen <zhangchen.fnst@cn.fujitsu.com> | 1 | From: Eugenio Pérez <eperezma@redhat.com> |
---|---|---|---|
2 | 2 | ||
3 | Make colo-compare and filter-rewriter can parse vnet packet. | 3 | This tree is able to look for a translated address from an IOVA address. |
4 | 4 | ||
5 | Signed-off-by: Zhang Chen <zhangchen.fnst@cn.fujitsu.com> | 5 | At first glance it is similar to util/iova-tree. However, SVQ working on |
6 | devices with limited IOVA space need more capabilities, like allocating | ||
7 | IOVA chunks or performing reverse translations (qemu addresses to iova). | ||
8 | |||
9 | The allocation capability, as "assign a free IOVA address to this chunk | ||
10 | of memory in qemu's address space" allows shadow virtqueue to create a | ||
11 | new address space that is not restricted by guest's addressable one, so | ||
12 | we can allocate shadow vqs vrings outside of it. | ||
13 | |||
14 | It duplicates the tree so it can search efficiently in both directions, | ||
15 | and it will signal overlap if iova or the translated address is present | ||
16 | in any tree. | ||
17 | |||
18 | Acked-by: Michael S. Tsirkin <mst@redhat.com> | ||
19 | Signed-off-by: Eugenio Pérez <eperezma@redhat.com> | ||
6 | Signed-off-by: Jason Wang <jasowang@redhat.com> | 20 | Signed-off-by: Jason Wang <jasowang@redhat.com> |
7 | --- | 21 | --- |
8 | net/colo.c | 6 +++--- | 22 | hw/virtio/meson.build | 2 +- |
9 | 1 file changed, 3 insertions(+), 3 deletions(-) | 23 | hw/virtio/vhost-iova-tree.c | 110 ++++++++++++++++++++++++++++++++++++++++++++ |
24 | hw/virtio/vhost-iova-tree.h | 27 +++++++++++ | ||
25 | 3 files changed, 138 insertions(+), 1 deletion(-) | ||
26 | create mode 100644 hw/virtio/vhost-iova-tree.c | ||
27 | create mode 100644 hw/virtio/vhost-iova-tree.h | ||
10 | 28 | ||
11 | diff --git a/net/colo.c b/net/colo.c | 29 | diff --git a/hw/virtio/meson.build b/hw/virtio/meson.build |
12 | index XXXXXXX..XXXXXXX 100644 | 30 | index XXXXXXX..XXXXXXX 100644 |
13 | --- a/net/colo.c | 31 | --- a/hw/virtio/meson.build |
14 | +++ b/net/colo.c | 32 | +++ b/hw/virtio/meson.build |
15 | @@ -XXX,XX +XXX,XX @@ int parse_packet_early(Packet *pkt) | 33 | @@ -XXX,XX +XXX,XX @@ softmmu_ss.add(when: 'CONFIG_ALL', if_true: files('vhost-stub.c')) |
16 | { | 34 | |
17 | int network_length; | 35 | virtio_ss = ss.source_set() |
18 | static const uint8_t vlan[] = {0x81, 0x00}; | 36 | virtio_ss.add(files('virtio.c')) |
19 | - uint8_t *data = pkt->data; | 37 | -virtio_ss.add(when: 'CONFIG_VHOST', if_true: files('vhost.c', 'vhost-backend.c', 'vhost-shadow-virtqueue.c')) |
20 | + uint8_t *data = pkt->data + pkt->vnet_hdr_len; | 38 | +virtio_ss.add(when: 'CONFIG_VHOST', if_true: files('vhost.c', 'vhost-backend.c', 'vhost-shadow-virtqueue.c', 'vhost-iova-tree.c')) |
21 | uint16_t l3_proto; | 39 | virtio_ss.add(when: 'CONFIG_VHOST_USER', if_true: files('vhost-user.c')) |
22 | ssize_t l2hdr_len = eth_get_l2_hdr_length(data); | 40 | virtio_ss.add(when: 'CONFIG_VHOST_VDPA', if_true: files('vhost-vdpa.c')) |
23 | 41 | virtio_ss.add(when: 'CONFIG_VIRTIO_BALLOON', if_true: files('virtio-balloon.c')) | |
24 | - if (pkt->size < ETH_HLEN) { | 42 | diff --git a/hw/virtio/vhost-iova-tree.c b/hw/virtio/vhost-iova-tree.c |
25 | + if (pkt->size < ETH_HLEN + pkt->vnet_hdr_len) { | 43 | new file mode 100644 |
26 | trace_colo_proxy_main("pkt->size < ETH_HLEN"); | 44 | index XXXXXXX..XXXXXXX |
27 | return 1; | 45 | --- /dev/null |
28 | } | 46 | +++ b/hw/virtio/vhost-iova-tree.c |
29 | @@ -XXX,XX +XXX,XX @@ int parse_packet_early(Packet *pkt) | 47 | @@ -XXX,XX +XXX,XX @@ |
30 | } | 48 | +/* |
31 | 49 | + * vhost software live migration iova tree | |
32 | network_length = pkt->ip->ip_hl * 4; | 50 | + * |
33 | - if (pkt->size < l2hdr_len + network_length) { | 51 | + * SPDX-FileCopyrightText: Red Hat, Inc. 2021 |
34 | + if (pkt->size < l2hdr_len + network_length + pkt->vnet_hdr_len) { | 52 | + * SPDX-FileContributor: Author: Eugenio Pérez <eperezma@redhat.com> |
35 | trace_colo_proxy_main("pkt->size < network_header + network_length"); | 53 | + * |
36 | return 1; | 54 | + * SPDX-License-Identifier: GPL-2.0-or-later |
37 | } | 55 | + */ |
56 | + | ||
57 | +#include "qemu/osdep.h" | ||
58 | +#include "qemu/iova-tree.h" | ||
59 | +#include "vhost-iova-tree.h" | ||
60 | + | ||
61 | +#define iova_min_addr qemu_real_host_page_size | ||
62 | + | ||
63 | +/** | ||
64 | + * VhostIOVATree, able to: | ||
65 | + * - Translate iova address | ||
66 | + * - Reverse translate iova address (from translated to iova) | ||
67 | + * - Allocate IOVA regions for translated range (linear operation) | ||
68 | + */ | ||
69 | +struct VhostIOVATree { | ||
70 | + /* First addressable iova address in the device */ | ||
71 | + uint64_t iova_first; | ||
72 | + | ||
73 | + /* Last addressable iova address in the device */ | ||
74 | + uint64_t iova_last; | ||
75 | + | ||
76 | + /* IOVA address to qemu memory maps. */ | ||
77 | + IOVATree *iova_taddr_map; | ||
78 | +}; | ||
79 | + | ||
80 | +/** | ||
81 | + * Create a new IOVA tree | ||
82 | + * | ||
83 | + * Returns the new IOVA tree | ||
84 | + */ | ||
85 | +VhostIOVATree *vhost_iova_tree_new(hwaddr iova_first, hwaddr iova_last) | ||
86 | +{ | ||
87 | + VhostIOVATree *tree = g_new(VhostIOVATree, 1); | ||
88 | + | ||
89 | + /* Some devices do not like 0 addresses */ | ||
90 | + tree->iova_first = MAX(iova_first, iova_min_addr); | ||
91 | + tree->iova_last = iova_last; | ||
92 | + | ||
93 | + tree->iova_taddr_map = iova_tree_new(); | ||
94 | + return tree; | ||
95 | +} | ||
96 | + | ||
97 | +/** | ||
98 | + * Delete an iova tree | ||
99 | + */ | ||
100 | +void vhost_iova_tree_delete(VhostIOVATree *iova_tree) | ||
101 | +{ | ||
102 | + iova_tree_destroy(iova_tree->iova_taddr_map); | ||
103 | + g_free(iova_tree); | ||
104 | +} | ||
105 | + | ||
106 | +/** | ||
107 | + * Find the IOVA address stored from a memory address | ||
108 | + * | ||
109 | + * @tree: The iova tree | ||
110 | + * @map: The map with the memory address | ||
111 | + * | ||
112 | + * Return the stored mapping, or NULL if not found. | ||
113 | + */ | ||
114 | +const DMAMap *vhost_iova_tree_find_iova(const VhostIOVATree *tree, | ||
115 | + const DMAMap *map) | ||
116 | +{ | ||
117 | + return iova_tree_find_iova(tree->iova_taddr_map, map); | ||
118 | +} | ||
119 | + | ||
120 | +/** | ||
121 | + * Allocate a new mapping | ||
122 | + * | ||
123 | + * @tree: The iova tree | ||
124 | + * @map: The iova map | ||
125 | + * | ||
126 | + * Returns: | ||
127 | + * - IOVA_OK if the map fits in the container | ||
128 | + * - IOVA_ERR_INVALID if the map does not make sense (like size overflow) | ||
129 | + * - IOVA_ERR_NOMEM if tree cannot allocate more space. | ||
130 | + * | ||
131 | + * It returns assignated iova in map->iova if return value is VHOST_DMA_MAP_OK. | ||
132 | + */ | ||
133 | +int vhost_iova_tree_map_alloc(VhostIOVATree *tree, DMAMap *map) | ||
134 | +{ | ||
135 | + /* Some vhost devices do not like addr 0. Skip first page */ | ||
136 | + hwaddr iova_first = tree->iova_first ?: qemu_real_host_page_size; | ||
137 | + | ||
138 | + if (map->translated_addr + map->size < map->translated_addr || | ||
139 | + map->perm == IOMMU_NONE) { | ||
140 | + return IOVA_ERR_INVALID; | ||
141 | + } | ||
142 | + | ||
143 | + /* Allocate a node in IOVA address */ | ||
144 | + return iova_tree_alloc_map(tree->iova_taddr_map, map, iova_first, | ||
145 | + tree->iova_last); | ||
146 | +} | ||
147 | + | ||
148 | +/** | ||
149 | + * Remove existing mappings from iova tree | ||
150 | + * | ||
151 | + * @iova_tree: The vhost iova tree | ||
152 | + * @map: The map to remove | ||
153 | + */ | ||
154 | +void vhost_iova_tree_remove(VhostIOVATree *iova_tree, const DMAMap *map) | ||
155 | +{ | ||
156 | + iova_tree_remove(iova_tree->iova_taddr_map, map); | ||
157 | +} | ||
158 | diff --git a/hw/virtio/vhost-iova-tree.h b/hw/virtio/vhost-iova-tree.h | ||
159 | new file mode 100644 | ||
160 | index XXXXXXX..XXXXXXX | ||
161 | --- /dev/null | ||
162 | +++ b/hw/virtio/vhost-iova-tree.h | ||
163 | @@ -XXX,XX +XXX,XX @@ | ||
164 | +/* | ||
165 | + * vhost software live migration iova tree | ||
166 | + * | ||
167 | + * SPDX-FileCopyrightText: Red Hat, Inc. 2021 | ||
168 | + * SPDX-FileContributor: Author: Eugenio Pérez <eperezma@redhat.com> | ||
169 | + * | ||
170 | + * SPDX-License-Identifier: GPL-2.0-or-later | ||
171 | + */ | ||
172 | + | ||
173 | +#ifndef HW_VIRTIO_VHOST_IOVA_TREE_H | ||
174 | +#define HW_VIRTIO_VHOST_IOVA_TREE_H | ||
175 | + | ||
176 | +#include "qemu/iova-tree.h" | ||
177 | +#include "exec/memory.h" | ||
178 | + | ||
179 | +typedef struct VhostIOVATree VhostIOVATree; | ||
180 | + | ||
181 | +VhostIOVATree *vhost_iova_tree_new(uint64_t iova_first, uint64_t iova_last); | ||
182 | +void vhost_iova_tree_delete(VhostIOVATree *iova_tree); | ||
183 | +G_DEFINE_AUTOPTR_CLEANUP_FUNC(VhostIOVATree, vhost_iova_tree_delete); | ||
184 | + | ||
185 | +const DMAMap *vhost_iova_tree_find_iova(const VhostIOVATree *iova_tree, | ||
186 | + const DMAMap *map); | ||
187 | +int vhost_iova_tree_map_alloc(VhostIOVATree *iova_tree, DMAMap *map); | ||
188 | +void vhost_iova_tree_remove(VhostIOVATree *iova_tree, const DMAMap *map); | ||
189 | + | ||
190 | +#endif | ||
38 | -- | 191 | -- |
39 | 2.7.4 | 192 | 2.7.4 |
40 | 193 | ||
41 | 194 | diff view generated by jsdifflib |
1 | From: Zhang Chen <zhangchen.fnst@cn.fujitsu.com> | 1 | From: Eugenio Pérez <eperezma@redhat.com> |
---|---|---|---|
2 | 2 | ||
3 | We can use this property flush and send packet with vnet_hdr_len. | 3 | Use translations added in VhostIOVATree in SVQ. |
4 | 4 | ||
5 | Signed-off-by: Zhang Chen <zhangchen.fnst@cn.fujitsu.com> | 5 | Only introduce usage here, not allocation and deallocation. As with |
6 | previous patches, we use the dead code paths of shadow_vqs_enabled to | ||
7 | avoid commiting too many changes at once. These are impossible to take | ||
8 | at the moment. | ||
9 | |||
10 | Acked-by: Michael S. Tsirkin <mst@redhat.com> | ||
11 | Signed-off-by: Eugenio Pérez <eperezma@redhat.com> | ||
6 | Signed-off-by: Jason Wang <jasowang@redhat.com> | 12 | Signed-off-by: Jason Wang <jasowang@redhat.com> |
7 | --- | 13 | --- |
8 | net/colo-compare.c | 8 ++++++-- | 14 | hw/virtio/vhost-shadow-virtqueue.c | 75 +++++++++++++++++++++-- |
9 | net/colo.c | 3 ++- | 15 | hw/virtio/vhost-shadow-virtqueue.h | 6 +- |
10 | net/colo.h | 4 +++- | 16 | hw/virtio/vhost-vdpa.c | 122 +++++++++++++++++++++++++++++++------ |
11 | net/filter-rewriter.c | 2 +- | 17 | include/hw/virtio/vhost-vdpa.h | 3 + |
12 | 4 files changed, 12 insertions(+), 5 deletions(-) | 18 | 4 files changed, 181 insertions(+), 25 deletions(-) |
13 | 19 | ||
14 | diff --git a/net/colo-compare.c b/net/colo-compare.c | 20 | diff --git a/hw/virtio/vhost-shadow-virtqueue.c b/hw/virtio/vhost-shadow-virtqueue.c |
15 | index XXXXXXX..XXXXXXX 100644 | 21 | index XXXXXXX..XXXXXXX 100644 |
16 | --- a/net/colo-compare.c | 22 | --- a/hw/virtio/vhost-shadow-virtqueue.c |
17 | +++ b/net/colo-compare.c | 23 | +++ b/hw/virtio/vhost-shadow-virtqueue.c |
18 | @@ -XXX,XX +XXX,XX @@ static int packet_enqueue(CompareState *s, int mode) | 24 | @@ -XXX,XX +XXX,XX @@ static uint16_t vhost_svq_available_slots(const VhostShadowVirtqueue *svq) |
19 | Connection *conn; | 25 | return svq->vring.num - (svq->shadow_avail_idx - svq->shadow_used_idx); |
20 | 26 | } | |
21 | if (mode == PRIMARY_IN) { | 27 | |
22 | - pkt = packet_new(s->pri_rs.buf, s->pri_rs.packet_len); | 28 | +/** |
23 | + pkt = packet_new(s->pri_rs.buf, | 29 | + * Translate addresses between the qemu's virtual address and the SVQ IOVA |
24 | + s->pri_rs.packet_len, | 30 | + * |
25 | + s->pri_rs.vnet_hdr_len); | 31 | + * @svq: Shadow VirtQueue |
26 | } else { | 32 | + * @vaddr: Translated IOVA addresses |
27 | - pkt = packet_new(s->sec_rs.buf, s->sec_rs.packet_len); | 33 | + * @iovec: Source qemu's VA addresses |
28 | + pkt = packet_new(s->sec_rs.buf, | 34 | + * @num: Length of iovec and minimum length of vaddr |
29 | + s->sec_rs.packet_len, | 35 | + */ |
30 | + s->sec_rs.vnet_hdr_len); | 36 | +static bool vhost_svq_translate_addr(const VhostShadowVirtqueue *svq, |
37 | + void **addrs, const struct iovec *iovec, | ||
38 | + size_t num) | ||
39 | +{ | ||
40 | + if (num == 0) { | ||
41 | + return true; | ||
42 | + } | ||
43 | + | ||
44 | + for (size_t i = 0; i < num; ++i) { | ||
45 | + DMAMap needle = { | ||
46 | + .translated_addr = (hwaddr)iovec[i].iov_base, | ||
47 | + .size = iovec[i].iov_len, | ||
48 | + }; | ||
49 | + size_t off; | ||
50 | + | ||
51 | + const DMAMap *map = vhost_iova_tree_find_iova(svq->iova_tree, &needle); | ||
52 | + /* | ||
53 | + * Map cannot be NULL since iova map contains all guest space and | ||
54 | + * qemu already has a physical address mapped | ||
55 | + */ | ||
56 | + if (unlikely(!map)) { | ||
57 | + qemu_log_mask(LOG_GUEST_ERROR, | ||
58 | + "Invalid address 0x%"HWADDR_PRIx" given by guest", | ||
59 | + needle.translated_addr); | ||
60 | + return false; | ||
61 | + } | ||
62 | + | ||
63 | + off = needle.translated_addr - map->translated_addr; | ||
64 | + addrs[i] = (void *)(map->iova + off); | ||
65 | + | ||
66 | + if (unlikely(int128_gt(int128_add(needle.translated_addr, | ||
67 | + iovec[i].iov_len), | ||
68 | + map->translated_addr + map->size))) { | ||
69 | + qemu_log_mask(LOG_GUEST_ERROR, | ||
70 | + "Guest buffer expands over iova range"); | ||
71 | + return false; | ||
72 | + } | ||
73 | + } | ||
74 | + | ||
75 | + return true; | ||
76 | +} | ||
77 | + | ||
78 | static void vhost_vring_write_descs(VhostShadowVirtqueue *svq, | ||
79 | + void * const *sg, | ||
80 | const struct iovec *iovec, | ||
81 | size_t num, bool more_descs, bool write) | ||
82 | { | ||
83 | @@ -XXX,XX +XXX,XX @@ static void vhost_vring_write_descs(VhostShadowVirtqueue *svq, | ||
84 | } else { | ||
85 | descs[i].flags = flags; | ||
86 | } | ||
87 | - descs[i].addr = cpu_to_le64((hwaddr)iovec[n].iov_base); | ||
88 | + descs[i].addr = cpu_to_le64((hwaddr)sg[n]); | ||
89 | descs[i].len = cpu_to_le32(iovec[n].iov_len); | ||
90 | |||
91 | last = i; | ||
92 | @@ -XXX,XX +XXX,XX @@ static bool vhost_svq_add_split(VhostShadowVirtqueue *svq, | ||
93 | { | ||
94 | unsigned avail_idx; | ||
95 | vring_avail_t *avail = svq->vring.avail; | ||
96 | + bool ok; | ||
97 | + g_autofree void **sgs = g_new(void *, MAX(elem->out_num, elem->in_num)); | ||
98 | |||
99 | *head = svq->free_head; | ||
100 | |||
101 | @@ -XXX,XX +XXX,XX @@ static bool vhost_svq_add_split(VhostShadowVirtqueue *svq, | ||
102 | return false; | ||
31 | } | 103 | } |
32 | 104 | ||
33 | if (parse_packet_early(pkt)) { | 105 | - vhost_vring_write_descs(svq, elem->out_sg, elem->out_num, |
34 | diff --git a/net/colo.c b/net/colo.c | 106 | + ok = vhost_svq_translate_addr(svq, sgs, elem->out_sg, elem->out_num); |
107 | + if (unlikely(!ok)) { | ||
108 | + return false; | ||
109 | + } | ||
110 | + vhost_vring_write_descs(svq, sgs, elem->out_sg, elem->out_num, | ||
111 | elem->in_num > 0, false); | ||
112 | - vhost_vring_write_descs(svq, elem->in_sg, elem->in_num, false, true); | ||
113 | + | ||
114 | + | ||
115 | + ok = vhost_svq_translate_addr(svq, sgs, elem->in_sg, elem->in_num); | ||
116 | + if (unlikely(!ok)) { | ||
117 | + return false; | ||
118 | + } | ||
119 | + | ||
120 | + vhost_vring_write_descs(svq, sgs, elem->in_sg, elem->in_num, false, true); | ||
121 | |||
122 | /* | ||
123 | * Put the entry in the available array (but don't update avail->idx until | ||
124 | @@ -XXX,XX +XXX,XX @@ void vhost_svq_stop(VhostShadowVirtqueue *svq) | ||
125 | * Creates vhost shadow virtqueue, and instructs the vhost device to use the | ||
126 | * shadow methods and file descriptors. | ||
127 | * | ||
128 | + * @iova_tree: Tree to perform descriptors translations | ||
129 | + * | ||
130 | * Returns the new virtqueue or NULL. | ||
131 | * | ||
132 | * In case of error, reason is reported through error_report. | ||
133 | */ | ||
134 | -VhostShadowVirtqueue *vhost_svq_new(void) | ||
135 | +VhostShadowVirtqueue *vhost_svq_new(VhostIOVATree *iova_tree) | ||
136 | { | ||
137 | g_autofree VhostShadowVirtqueue *svq = g_new0(VhostShadowVirtqueue, 1); | ||
138 | int r; | ||
139 | @@ -XXX,XX +XXX,XX @@ VhostShadowVirtqueue *vhost_svq_new(void) | ||
140 | |||
141 | event_notifier_init_fd(&svq->svq_kick, VHOST_FILE_UNBIND); | ||
142 | event_notifier_set_handler(&svq->hdev_call, vhost_svq_handle_call); | ||
143 | + svq->iova_tree = iova_tree; | ||
144 | return g_steal_pointer(&svq); | ||
145 | |||
146 | err_init_hdev_call: | ||
147 | diff --git a/hw/virtio/vhost-shadow-virtqueue.h b/hw/virtio/vhost-shadow-virtqueue.h | ||
35 | index XXXXXXX..XXXXXXX 100644 | 148 | index XXXXXXX..XXXXXXX 100644 |
36 | --- a/net/colo.c | 149 | --- a/hw/virtio/vhost-shadow-virtqueue.h |
37 | +++ b/net/colo.c | 150 | +++ b/hw/virtio/vhost-shadow-virtqueue.h |
38 | @@ -XXX,XX +XXX,XX @@ void connection_destroy(void *opaque) | 151 | @@ -XXX,XX +XXX,XX @@ |
39 | g_slice_free(Connection, conn); | 152 | #include "qemu/event_notifier.h" |
153 | #include "hw/virtio/virtio.h" | ||
154 | #include "standard-headers/linux/vhost_types.h" | ||
155 | +#include "hw/virtio/vhost-iova-tree.h" | ||
156 | |||
157 | /* Shadow virtqueue to relay notifications */ | ||
158 | typedef struct VhostShadowVirtqueue { | ||
159 | @@ -XXX,XX +XXX,XX @@ typedef struct VhostShadowVirtqueue { | ||
160 | /* Virtio device */ | ||
161 | VirtIODevice *vdev; | ||
162 | |||
163 | + /* IOVA mapping */ | ||
164 | + VhostIOVATree *iova_tree; | ||
165 | + | ||
166 | /* Map for use the guest's descriptors */ | ||
167 | VirtQueueElement **ring_id_maps; | ||
168 | |||
169 | @@ -XXX,XX +XXX,XX @@ void vhost_svq_start(VhostShadowVirtqueue *svq, VirtIODevice *vdev, | ||
170 | VirtQueue *vq); | ||
171 | void vhost_svq_stop(VhostShadowVirtqueue *svq); | ||
172 | |||
173 | -VhostShadowVirtqueue *vhost_svq_new(void); | ||
174 | +VhostShadowVirtqueue *vhost_svq_new(VhostIOVATree *iova_tree); | ||
175 | |||
176 | void vhost_svq_free(gpointer vq); | ||
177 | G_DEFINE_AUTOPTR_CLEANUP_FUNC(VhostShadowVirtqueue, vhost_svq_free); | ||
178 | diff --git a/hw/virtio/vhost-vdpa.c b/hw/virtio/vhost-vdpa.c | ||
179 | index XXXXXXX..XXXXXXX 100644 | ||
180 | --- a/hw/virtio/vhost-vdpa.c | ||
181 | +++ b/hw/virtio/vhost-vdpa.c | ||
182 | @@ -XXX,XX +XXX,XX @@ static void vhost_vdpa_listener_region_add(MemoryListener *listener, | ||
183 | vaddr, section->readonly); | ||
184 | |||
185 | llsize = int128_sub(llend, int128_make64(iova)); | ||
186 | + if (v->shadow_vqs_enabled) { | ||
187 | + DMAMap mem_region = { | ||
188 | + .translated_addr = (hwaddr)vaddr, | ||
189 | + .size = int128_get64(llsize) - 1, | ||
190 | + .perm = IOMMU_ACCESS_FLAG(true, section->readonly), | ||
191 | + }; | ||
192 | + | ||
193 | + int r = vhost_iova_tree_map_alloc(v->iova_tree, &mem_region); | ||
194 | + if (unlikely(r != IOVA_OK)) { | ||
195 | + error_report("Can't allocate a mapping (%d)", r); | ||
196 | + goto fail; | ||
197 | + } | ||
198 | + | ||
199 | + iova = mem_region.iova; | ||
200 | + } | ||
201 | |||
202 | vhost_vdpa_iotlb_batch_begin_once(v); | ||
203 | ret = vhost_vdpa_dma_map(v, iova, int128_get64(llsize), | ||
204 | @@ -XXX,XX +XXX,XX @@ static void vhost_vdpa_listener_region_del(MemoryListener *listener, | ||
205 | |||
206 | llsize = int128_sub(llend, int128_make64(iova)); | ||
207 | |||
208 | + if (v->shadow_vqs_enabled) { | ||
209 | + const DMAMap *result; | ||
210 | + const void *vaddr = memory_region_get_ram_ptr(section->mr) + | ||
211 | + section->offset_within_region + | ||
212 | + (iova - section->offset_within_address_space); | ||
213 | + DMAMap mem_region = { | ||
214 | + .translated_addr = (hwaddr)vaddr, | ||
215 | + .size = int128_get64(llsize) - 1, | ||
216 | + }; | ||
217 | + | ||
218 | + result = vhost_iova_tree_find_iova(v->iova_tree, &mem_region); | ||
219 | + iova = result->iova; | ||
220 | + vhost_iova_tree_remove(v->iova_tree, &mem_region); | ||
221 | + } | ||
222 | vhost_vdpa_iotlb_batch_begin_once(v); | ||
223 | ret = vhost_vdpa_dma_unmap(v, iova, int128_get64(llsize)); | ||
224 | if (ret) { | ||
225 | @@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_init_svq(struct vhost_dev *hdev, struct vhost_vdpa *v, | ||
226 | |||
227 | shadow_vqs = g_ptr_array_new_full(hdev->nvqs, vhost_svq_free); | ||
228 | for (unsigned n = 0; n < hdev->nvqs; ++n) { | ||
229 | - g_autoptr(VhostShadowVirtqueue) svq = vhost_svq_new(); | ||
230 | + g_autoptr(VhostShadowVirtqueue) svq = vhost_svq_new(v->iova_tree); | ||
231 | |||
232 | if (unlikely(!svq)) { | ||
233 | error_setg(errp, "Cannot create svq %u", n); | ||
234 | @@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_svq_set_fds(struct vhost_dev *dev, | ||
235 | /** | ||
236 | * Unmap a SVQ area in the device | ||
237 | */ | ||
238 | -static bool vhost_vdpa_svq_unmap_ring(struct vhost_vdpa *v, hwaddr iova, | ||
239 | - hwaddr size) | ||
240 | +static bool vhost_vdpa_svq_unmap_ring(struct vhost_vdpa *v, | ||
241 | + const DMAMap *needle) | ||
242 | { | ||
243 | + const DMAMap *result = vhost_iova_tree_find_iova(v->iova_tree, needle); | ||
244 | + hwaddr size; | ||
245 | int r; | ||
246 | |||
247 | - size = ROUND_UP(size, qemu_real_host_page_size); | ||
248 | - r = vhost_vdpa_dma_unmap(v, iova, size); | ||
249 | + if (unlikely(!result)) { | ||
250 | + error_report("Unable to find SVQ address to unmap"); | ||
251 | + return false; | ||
252 | + } | ||
253 | + | ||
254 | + size = ROUND_UP(result->size, qemu_real_host_page_size); | ||
255 | + r = vhost_vdpa_dma_unmap(v, result->iova, size); | ||
256 | return r == 0; | ||
40 | } | 257 | } |
41 | 258 | ||
42 | -Packet *packet_new(const void *data, int size) | 259 | static bool vhost_vdpa_svq_unmap_rings(struct vhost_dev *dev, |
43 | +Packet *packet_new(const void *data, int size, int vnet_hdr_len) | 260 | const VhostShadowVirtqueue *svq) |
44 | { | 261 | { |
45 | Packet *pkt = g_slice_new(Packet); | 262 | + DMAMap needle = {}; |
46 | 263 | struct vhost_vdpa *v = dev->opaque; | |
47 | pkt->data = g_memdup(data, size); | 264 | struct vhost_vring_addr svq_addr; |
48 | pkt->size = size; | 265 | - size_t device_size = vhost_svq_device_area_size(svq); |
49 | pkt->creation_ms = qemu_clock_get_ms(QEMU_CLOCK_HOST); | 266 | - size_t driver_size = vhost_svq_driver_area_size(svq); |
50 | + pkt->vnet_hdr_len = vnet_hdr_len; | 267 | bool ok; |
51 | 268 | ||
52 | return pkt; | 269 | vhost_svq_get_vring_addr(svq, &svq_addr); |
270 | |||
271 | - ok = vhost_vdpa_svq_unmap_ring(v, svq_addr.desc_user_addr, driver_size); | ||
272 | + needle.translated_addr = svq_addr.desc_user_addr; | ||
273 | + ok = vhost_vdpa_svq_unmap_ring(v, &needle); | ||
274 | if (unlikely(!ok)) { | ||
275 | return false; | ||
276 | } | ||
277 | |||
278 | - return vhost_vdpa_svq_unmap_ring(v, svq_addr.used_user_addr, device_size); | ||
279 | + needle.translated_addr = svq_addr.used_user_addr; | ||
280 | + return vhost_vdpa_svq_unmap_ring(v, &needle); | ||
281 | +} | ||
282 | + | ||
283 | +/** | ||
284 | + * Map the SVQ area in the device | ||
285 | + * | ||
286 | + * @v: Vhost-vdpa device | ||
287 | + * @needle: The area to search iova | ||
288 | + * @errorp: Error pointer | ||
289 | + */ | ||
290 | +static bool vhost_vdpa_svq_map_ring(struct vhost_vdpa *v, DMAMap *needle, | ||
291 | + Error **errp) | ||
292 | +{ | ||
293 | + int r; | ||
294 | + | ||
295 | + r = vhost_iova_tree_map_alloc(v->iova_tree, needle); | ||
296 | + if (unlikely(r != IOVA_OK)) { | ||
297 | + error_setg(errp, "Cannot allocate iova (%d)", r); | ||
298 | + return false; | ||
299 | + } | ||
300 | + | ||
301 | + r = vhost_vdpa_dma_map(v, needle->iova, needle->size + 1, | ||
302 | + (void *)needle->translated_addr, | ||
303 | + needle->perm == IOMMU_RO); | ||
304 | + if (unlikely(r != 0)) { | ||
305 | + error_setg_errno(errp, -r, "Cannot map region to device"); | ||
306 | + vhost_iova_tree_remove(v->iova_tree, needle); | ||
307 | + } | ||
308 | + | ||
309 | + return r == 0; | ||
53 | } | 310 | } |
54 | diff --git a/net/colo.h b/net/colo.h | 311 | |
312 | /** | ||
313 | @@ -XXX,XX +XXX,XX @@ static bool vhost_vdpa_svq_map_rings(struct vhost_dev *dev, | ||
314 | struct vhost_vring_addr *addr, | ||
315 | Error **errp) | ||
316 | { | ||
317 | + DMAMap device_region, driver_region; | ||
318 | + struct vhost_vring_addr svq_addr; | ||
319 | struct vhost_vdpa *v = dev->opaque; | ||
320 | size_t device_size = vhost_svq_device_area_size(svq); | ||
321 | size_t driver_size = vhost_svq_driver_area_size(svq); | ||
322 | - int r; | ||
323 | + size_t avail_offset; | ||
324 | + bool ok; | ||
325 | |||
326 | ERRP_GUARD(); | ||
327 | - vhost_svq_get_vring_addr(svq, addr); | ||
328 | + vhost_svq_get_vring_addr(svq, &svq_addr); | ||
329 | |||
330 | - r = vhost_vdpa_dma_map(v, addr->desc_user_addr, driver_size, | ||
331 | - (void *)addr->desc_user_addr, true); | ||
332 | - if (unlikely(r != 0)) { | ||
333 | - error_setg_errno(errp, -r, "Cannot create vq driver region: "); | ||
334 | + driver_region = (DMAMap) { | ||
335 | + .translated_addr = svq_addr.desc_user_addr, | ||
336 | + .size = driver_size - 1, | ||
337 | + .perm = IOMMU_RO, | ||
338 | + }; | ||
339 | + ok = vhost_vdpa_svq_map_ring(v, &driver_region, errp); | ||
340 | + if (unlikely(!ok)) { | ||
341 | + error_prepend(errp, "Cannot create vq driver region: "); | ||
342 | return false; | ||
343 | } | ||
344 | + addr->desc_user_addr = driver_region.iova; | ||
345 | + avail_offset = svq_addr.avail_user_addr - svq_addr.desc_user_addr; | ||
346 | + addr->avail_user_addr = driver_region.iova + avail_offset; | ||
347 | |||
348 | - r = vhost_vdpa_dma_map(v, addr->used_user_addr, device_size, | ||
349 | - (void *)addr->used_user_addr, false); | ||
350 | - if (unlikely(r != 0)) { | ||
351 | - error_setg_errno(errp, -r, "Cannot create vq device region: "); | ||
352 | + device_region = (DMAMap) { | ||
353 | + .translated_addr = svq_addr.used_user_addr, | ||
354 | + .size = device_size - 1, | ||
355 | + .perm = IOMMU_RW, | ||
356 | + }; | ||
357 | + ok = vhost_vdpa_svq_map_ring(v, &device_region, errp); | ||
358 | + if (unlikely(!ok)) { | ||
359 | + error_prepend(errp, "Cannot create vq device region: "); | ||
360 | + vhost_vdpa_svq_unmap_ring(v, &driver_region); | ||
361 | } | ||
362 | + addr->used_user_addr = device_region.iova; | ||
363 | |||
364 | - return r == 0; | ||
365 | + return ok; | ||
366 | } | ||
367 | |||
368 | static bool vhost_vdpa_svq_setup(struct vhost_dev *dev, | ||
369 | diff --git a/include/hw/virtio/vhost-vdpa.h b/include/hw/virtio/vhost-vdpa.h | ||
55 | index XXXXXXX..XXXXXXX 100644 | 370 | index XXXXXXX..XXXXXXX 100644 |
56 | --- a/net/colo.h | 371 | --- a/include/hw/virtio/vhost-vdpa.h |
57 | +++ b/net/colo.h | 372 | +++ b/include/hw/virtio/vhost-vdpa.h |
58 | @@ -XXX,XX +XXX,XX @@ typedef struct Packet { | 373 | @@ -XXX,XX +XXX,XX @@ |
59 | int size; | 374 | |
60 | /* Time of packet creation, in wall clock ms */ | 375 | #include <gmodule.h> |
61 | int64_t creation_ms; | 376 | |
62 | + /* Get vnet_hdr_len from filter */ | 377 | +#include "hw/virtio/vhost-iova-tree.h" |
63 | + uint32_t vnet_hdr_len; | 378 | #include "hw/virtio/virtio.h" |
64 | } Packet; | 379 | #include "standard-headers/linux/vhost_types.h" |
65 | 380 | ||
66 | typedef struct ConnectionKey { | 381 | @@ -XXX,XX +XXX,XX @@ typedef struct vhost_vdpa { |
67 | @@ -XXX,XX +XXX,XX @@ Connection *connection_get(GHashTable *connection_track_table, | 382 | MemoryListener listener; |
68 | ConnectionKey *key, | 383 | struct vhost_vdpa_iova_range iova_range; |
69 | GQueue *conn_list); | 384 | bool shadow_vqs_enabled; |
70 | void connection_hashtable_reset(GHashTable *connection_track_table); | 385 | + /* IOVA mapping used by the Shadow Virtqueue */ |
71 | -Packet *packet_new(const void *data, int size); | 386 | + VhostIOVATree *iova_tree; |
72 | +Packet *packet_new(const void *data, int size, int vnet_hdr_len); | 387 | GPtrArray *shadow_vqs; |
73 | void packet_destroy(void *opaque, void *user_data); | 388 | struct vhost_dev *dev; |
74 | 389 | VhostVDPAHostNotifier notifier[VIRTIO_QUEUE_MAX]; | |
75 | #endif /* QEMU_COLO_PROXY_H */ | ||
76 | diff --git a/net/filter-rewriter.c b/net/filter-rewriter.c | ||
77 | index XXXXXXX..XXXXXXX 100644 | ||
78 | --- a/net/filter-rewriter.c | ||
79 | +++ b/net/filter-rewriter.c | ||
80 | @@ -XXX,XX +XXX,XX @@ static ssize_t colo_rewriter_receive_iov(NetFilterState *nf, | ||
81 | char *buf = g_malloc0(size); | ||
82 | |||
83 | iov_to_buf(iov, iovcnt, 0, buf, size); | ||
84 | - pkt = packet_new(buf, size); | ||
85 | + pkt = packet_new(buf, size, 0); | ||
86 | g_free(buf); | ||
87 | |||
88 | /* | ||
89 | -- | 390 | -- |
90 | 2.7.4 | 391 | 2.7.4 |
91 | 392 | ||
92 | 393 | diff view generated by jsdifflib |
1 | From: Zhang Chen <zhangchen.fnst@cn.fujitsu.com> | 1 | From: Eugenio Pérez <eperezma@redhat.com> |
---|---|---|---|
2 | 2 | ||
3 | This patch change the filter_send() parameter from CharBackend to MirrorState, | 3 | This is needed to achieve migration, so the destination can restore its |
4 | we can get more information like vnet_hdr(We use it to support packet with vnet_header). | 4 | index. |
5 | 5 | ||
6 | Signed-off-by: Zhang Chen <zhangchen.fnst@cn.fujitsu.com> | 6 | Setting base as last used idx, so destination will see as available all |
7 | the entries that the device did not use, including the in-flight | ||
8 | processing ones. | ||
9 | |||
10 | This is ok for networking, but other kinds of devices might have | ||
11 | problems with these retransmissions. | ||
12 | |||
13 | Acked-by: Michael S. Tsirkin <mst@redhat.com> | ||
14 | Signed-off-by: Eugenio Pérez <eperezma@redhat.com> | ||
7 | Signed-off-by: Jason Wang <jasowang@redhat.com> | 15 | Signed-off-by: Jason Wang <jasowang@redhat.com> |
8 | --- | 16 | --- |
9 | net/filter-mirror.c | 10 +++++----- | 17 | hw/virtio/vhost-vdpa.c | 17 +++++++++++++++++ |
10 | 1 file changed, 5 insertions(+), 5 deletions(-) | 18 | 1 file changed, 17 insertions(+) |
11 | 19 | ||
12 | diff --git a/net/filter-mirror.c b/net/filter-mirror.c | 20 | diff --git a/hw/virtio/vhost-vdpa.c b/hw/virtio/vhost-vdpa.c |
13 | index XXXXXXX..XXXXXXX 100644 | 21 | index XXXXXXX..XXXXXXX 100644 |
14 | --- a/net/filter-mirror.c | 22 | --- a/hw/virtio/vhost-vdpa.c |
15 | +++ b/net/filter-mirror.c | 23 | +++ b/hw/virtio/vhost-vdpa.c |
16 | @@ -XXX,XX +XXX,XX @@ typedef struct MirrorState { | 24 | @@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_set_vring_base(struct vhost_dev *dev, |
17 | SocketReadState rs; | 25 | static int vhost_vdpa_get_vring_base(struct vhost_dev *dev, |
18 | } MirrorState; | 26 | struct vhost_vring_state *ring) |
19 | |||
20 | -static int filter_send(CharBackend *chr_out, | ||
21 | +static int filter_send(MirrorState *s, | ||
22 | const struct iovec *iov, | ||
23 | int iovcnt) | ||
24 | { | 27 | { |
25 | @@ -XXX,XX +XXX,XX @@ static int filter_send(CharBackend *chr_out, | 28 | + struct vhost_vdpa *v = dev->opaque; |
26 | } | ||
27 | |||
28 | len = htonl(size); | ||
29 | - ret = qemu_chr_fe_write_all(chr_out, (uint8_t *)&len, sizeof(len)); | ||
30 | + ret = qemu_chr_fe_write_all(&s->chr_out, (uint8_t *)&len, sizeof(len)); | ||
31 | if (ret != sizeof(len)) { | ||
32 | goto err; | ||
33 | } | ||
34 | |||
35 | buf = g_malloc(size); | ||
36 | iov_to_buf(iov, iovcnt, 0, buf, size); | ||
37 | - ret = qemu_chr_fe_write_all(chr_out, (uint8_t *)buf, size); | ||
38 | + ret = qemu_chr_fe_write_all(&s->chr_out, (uint8_t *)buf, size); | ||
39 | g_free(buf); | ||
40 | if (ret != size) { | ||
41 | goto err; | ||
42 | @@ -XXX,XX +XXX,XX @@ static ssize_t filter_mirror_receive_iov(NetFilterState *nf, | ||
43 | MirrorState *s = FILTER_MIRROR(nf); | ||
44 | int ret; | 29 | int ret; |
45 | 30 | ||
46 | - ret = filter_send(&s->chr_out, iov, iovcnt); | 31 | + if (v->shadow_vqs_enabled) { |
47 | + ret = filter_send(s, iov, iovcnt); | 32 | + VhostShadowVirtqueue *svq = g_ptr_array_index(v->shadow_vqs, |
48 | if (ret) { | 33 | + ring->index); |
49 | error_report("filter mirror send failed(%s)", strerror(-ret)); | 34 | + |
50 | } | 35 | + /* |
51 | @@ -XXX,XX +XXX,XX @@ static ssize_t filter_redirector_receive_iov(NetFilterState *nf, | 36 | + * Setting base as last used idx, so destination will see as available |
52 | int ret; | 37 | + * all the entries that the device did not use, including the in-flight |
53 | 38 | + * processing ones. | |
54 | if (qemu_chr_fe_backend_connected(&s->chr_out)) { | 39 | + * |
55 | - ret = filter_send(&s->chr_out, iov, iovcnt); | 40 | + * TODO: This is ok for networking, but other kinds of devices might |
56 | + ret = filter_send(s, iov, iovcnt); | 41 | + * have problems with these retransmissions. |
57 | if (ret) { | 42 | + */ |
58 | error_report("filter redirector send failed(%s)", strerror(-ret)); | 43 | + ring->num = svq->last_used_idx; |
59 | } | 44 | + return 0; |
45 | + } | ||
46 | + | ||
47 | ret = vhost_vdpa_call(dev, VHOST_GET_VRING_BASE, ring); | ||
48 | trace_vhost_vdpa_get_vring_base(dev, ring->index, ring->num); | ||
49 | return ret; | ||
60 | -- | 50 | -- |
61 | 2.7.4 | 51 | 2.7.4 |
62 | 52 | ||
63 | 53 | diff view generated by jsdifflib |
1 | From: Zhang Chen <zhangchen.fnst@cn.fujitsu.com> | 1 | From: Eugenio Pérez <eperezma@redhat.com> |
---|---|---|---|
2 | 2 | ||
3 | COLO-Proxy just focus on packet payload, so we skip vnet header. | 3 | Setting the log address would make the device start reporting invalid |
4 | dirty memory because the SVQ vrings are located in qemu's memory. | ||
4 | 5 | ||
5 | Signed-off-by: Zhang Chen <zhangchen.fnst@cn.fujitsu.com> | 6 | Acked-by: Michael S. Tsirkin <mst@redhat.com> |
7 | Signed-off-by: Eugenio Pérez <eperezma@redhat.com> | ||
6 | Signed-off-by: Jason Wang <jasowang@redhat.com> | 8 | Signed-off-by: Jason Wang <jasowang@redhat.com> |
7 | --- | 9 | --- |
8 | net/colo-compare.c | 8 ++++++-- | 10 | hw/virtio/vhost-vdpa.c | 3 ++- |
9 | 1 file changed, 6 insertions(+), 2 deletions(-) | 11 | 1 file changed, 2 insertions(+), 1 deletion(-) |
10 | 12 | ||
11 | diff --git a/net/colo-compare.c b/net/colo-compare.c | 13 | diff --git a/hw/virtio/vhost-vdpa.c b/hw/virtio/vhost-vdpa.c |
12 | index XXXXXXX..XXXXXXX 100644 | 14 | index XXXXXXX..XXXXXXX 100644 |
13 | --- a/net/colo-compare.c | 15 | --- a/hw/virtio/vhost-vdpa.c |
14 | +++ b/net/colo-compare.c | 16 | +++ b/hw/virtio/vhost-vdpa.c |
15 | @@ -XXX,XX +XXX,XX @@ static int colo_packet_compare_common(Packet *ppkt, Packet *spkt, int offset) | 17 | @@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_dev_start(struct vhost_dev *dev, bool started) |
16 | sec_ip_src, sec_ip_dst); | 18 | static int vhost_vdpa_set_log_base(struct vhost_dev *dev, uint64_t base, |
19 | struct vhost_log *log) | ||
20 | { | ||
21 | - if (vhost_vdpa_one_time_request(dev)) { | ||
22 | + struct vhost_vdpa *v = dev->opaque; | ||
23 | + if (v->shadow_vqs_enabled || vhost_vdpa_one_time_request(dev)) { | ||
24 | return 0; | ||
17 | } | 25 | } |
18 | 26 | ||
19 | + offset = ppkt->vnet_hdr_len + offset; | ||
20 | + | ||
21 | if (ppkt->size == spkt->size) { | ||
22 | - return memcmp(ppkt->data + offset, spkt->data + offset, | ||
23 | + return memcmp(ppkt->data + offset, | ||
24 | + spkt->data + offset, | ||
25 | spkt->size - offset); | ||
26 | } else { | ||
27 | trace_colo_compare_main("Net packet size are not the same"); | ||
28 | @@ -XXX,XX +XXX,XX @@ static int colo_packet_compare_tcp(Packet *spkt, Packet *ppkt) | ||
29 | */ | ||
30 | if (ptcp->th_off > 5) { | ||
31 | ptrdiff_t tcp_offset; | ||
32 | + | ||
33 | tcp_offset = ppkt->transport_header - (uint8_t *)ppkt->data | ||
34 | - + (ptcp->th_off * 4); | ||
35 | + + (ptcp->th_off * 4) - ppkt->vnet_hdr_len; | ||
36 | res = colo_packet_compare_common(ppkt, spkt, tcp_offset); | ||
37 | } else if (ptcp->th_sum == stcp->th_sum) { | ||
38 | res = colo_packet_compare_common(ppkt, spkt, ETH_HLEN); | ||
39 | -- | 27 | -- |
40 | 2.7.4 | 28 | 2.7.4 |
41 | 29 | ||
42 | 30 | diff view generated by jsdifflib |
1 | From: Zhang Chen <zhangchen.fnst@cn.fujitsu.com> | 1 | From: Eugenio Pérez <eperezma@redhat.com> |
---|---|---|---|
2 | 2 | ||
3 | This patch change the compare_chr_send() parameter from CharBackend to CompareState, | 3 | SVQ is able to log the dirty bits by itself, so let's use it to not |
4 | we can get more information like vnet_hdr(We use it to support packet with vnet_header). | 4 | block migration. |
5 | 5 | ||
6 | Signed-off-by: Zhang Chen <zhangchen.fnst@cn.fujitsu.com> | 6 | Also, ignore set and clear of VHOST_F_LOG_ALL on set_features if SVQ is |
7 | enabled. Even if the device supports it, the reports would be nonsense | ||
8 | because SVQ memory is in the qemu region. | ||
9 | |||
10 | The log region is still allocated. Future changes might skip that, but | ||
11 | this series is already long enough. | ||
12 | |||
13 | Acked-by: Michael S. Tsirkin <mst@redhat.com> | ||
14 | Signed-off-by: Eugenio Pérez <eperezma@redhat.com> | ||
7 | Signed-off-by: Jason Wang <jasowang@redhat.com> | 15 | Signed-off-by: Jason Wang <jasowang@redhat.com> |
8 | --- | 16 | --- |
9 | net/colo-compare.c | 14 +++++++------- | 17 | hw/virtio/vhost-vdpa.c | 39 +++++++++++++++++++++++++++++++++++---- |
10 | 1 file changed, 7 insertions(+), 7 deletions(-) | 18 | include/hw/virtio/vhost-vdpa.h | 1 + |
19 | 2 files changed, 36 insertions(+), 4 deletions(-) | ||
11 | 20 | ||
12 | diff --git a/net/colo-compare.c b/net/colo-compare.c | 21 | diff --git a/hw/virtio/vhost-vdpa.c b/hw/virtio/vhost-vdpa.c |
13 | index XXXXXXX..XXXXXXX 100644 | 22 | index XXXXXXX..XXXXXXX 100644 |
14 | --- a/net/colo-compare.c | 23 | --- a/hw/virtio/vhost-vdpa.c |
15 | +++ b/net/colo-compare.c | 24 | +++ b/hw/virtio/vhost-vdpa.c |
16 | @@ -XXX,XX +XXX,XX @@ enum { | 25 | @@ -XXX,XX +XXX,XX @@ static bool vhost_vdpa_one_time_request(struct vhost_dev *dev) |
17 | SECONDARY_IN, | 26 | return v->index != 0; |
18 | }; | ||
19 | |||
20 | -static int compare_chr_send(CharBackend *out, | ||
21 | +static int compare_chr_send(CompareState *s, | ||
22 | const uint8_t *buf, | ||
23 | uint32_t size); | ||
24 | |||
25 | @@ -XXX,XX +XXX,XX @@ static void colo_compare_connection(void *opaque, void *user_data) | ||
26 | } | ||
27 | |||
28 | if (result) { | ||
29 | - ret = compare_chr_send(&s->chr_out, pkt->data, pkt->size); | ||
30 | + ret = compare_chr_send(s, pkt->data, pkt->size); | ||
31 | if (ret < 0) { | ||
32 | error_report("colo_send_primary_packet failed"); | ||
33 | } | ||
34 | @@ -XXX,XX +XXX,XX @@ static void colo_compare_connection(void *opaque, void *user_data) | ||
35 | } | ||
36 | } | 27 | } |
37 | 28 | ||
38 | -static int compare_chr_send(CharBackend *out, | 29 | +static int vhost_vdpa_get_dev_features(struct vhost_dev *dev, |
39 | +static int compare_chr_send(CompareState *s, | 30 | + uint64_t *features) |
40 | const uint8_t *buf, | 31 | +{ |
41 | uint32_t size) | 32 | + int ret; |
33 | + | ||
34 | + ret = vhost_vdpa_call(dev, VHOST_GET_FEATURES, features); | ||
35 | + trace_vhost_vdpa_get_features(dev, *features); | ||
36 | + return ret; | ||
37 | +} | ||
38 | + | ||
39 | static int vhost_vdpa_init_svq(struct vhost_dev *hdev, struct vhost_vdpa *v, | ||
40 | Error **errp) | ||
42 | { | 41 | { |
43 | @@ -XXX,XX +XXX,XX @@ static int compare_chr_send(CharBackend *out, | 42 | @@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_init_svq(struct vhost_dev *hdev, struct vhost_vdpa *v, |
44 | return 0; | 43 | return 0; |
45 | } | 44 | } |
46 | 45 | ||
47 | - ret = qemu_chr_fe_write_all(out, (uint8_t *)&len, sizeof(len)); | 46 | - r = hdev->vhost_ops->vhost_get_features(hdev, &dev_features); |
48 | + ret = qemu_chr_fe_write_all(&s->chr_out, (uint8_t *)&len, sizeof(len)); | 47 | + r = vhost_vdpa_get_dev_features(hdev, &dev_features); |
49 | if (ret != sizeof(len)) { | 48 | if (r != 0) { |
50 | goto err; | 49 | error_setg_errno(errp, -r, "Can't get vdpa device features"); |
50 | return r; | ||
51 | @@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_set_mem_table(struct vhost_dev *dev, | ||
52 | static int vhost_vdpa_set_features(struct vhost_dev *dev, | ||
53 | uint64_t features) | ||
54 | { | ||
55 | + struct vhost_vdpa *v = dev->opaque; | ||
56 | int ret; | ||
57 | |||
58 | if (vhost_vdpa_one_time_request(dev)) { | ||
59 | return 0; | ||
51 | } | 60 | } |
52 | 61 | ||
53 | - ret = qemu_chr_fe_write_all(out, (uint8_t *)buf, size); | 62 | + if (v->shadow_vqs_enabled) { |
54 | + ret = qemu_chr_fe_write_all(&s->chr_out, (uint8_t *)buf, size); | 63 | + if ((v->acked_features ^ features) == BIT_ULL(VHOST_F_LOG_ALL)) { |
55 | if (ret != size) { | 64 | + /* |
56 | goto err; | 65 | + * QEMU is just trying to enable or disable logging. SVQ handles |
57 | } | 66 | + * this sepparately, so no need to forward this. |
58 | @@ -XXX,XX +XXX,XX @@ static void compare_pri_rs_finalize(SocketReadState *pri_rs) | 67 | + */ |
59 | 68 | + v->acked_features = features; | |
60 | if (packet_enqueue(s, PRIMARY_IN)) { | 69 | + return 0; |
61 | trace_colo_compare_main("primary: unsupported packet in"); | 70 | + } |
62 | - compare_chr_send(&s->chr_out, pri_rs->buf, pri_rs->packet_len); | 71 | + |
63 | + compare_chr_send(s, pri_rs->buf, pri_rs->packet_len); | 72 | + v->acked_features = features; |
64 | } else { | 73 | + |
65 | /* compare connection */ | 74 | + /* We must not ack _F_LOG if SVQ is enabled */ |
66 | g_queue_foreach(&s->conn_list, colo_compare_connection, s); | 75 | + features &= ~BIT_ULL(VHOST_F_LOG_ALL); |
67 | @@ -XXX,XX +XXX,XX @@ static void colo_flush_packets(void *opaque, void *user_data) | 76 | + } |
68 | 77 | + | |
69 | while (!g_queue_is_empty(&conn->primary_list)) { | 78 | trace_vhost_vdpa_set_features(dev, features); |
70 | pkt = g_queue_pop_head(&conn->primary_list); | 79 | ret = vhost_vdpa_call(dev, VHOST_SET_FEATURES, &features); |
71 | - compare_chr_send(&s->chr_out, pkt->data, pkt->size); | 80 | if (ret) { |
72 | + compare_chr_send(s, pkt->data, pkt->size); | 81 | @@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_set_vring_call(struct vhost_dev *dev, |
73 | packet_destroy(pkt, NULL); | 82 | static int vhost_vdpa_get_features(struct vhost_dev *dev, |
74 | } | 83 | uint64_t *features) |
75 | while (!g_queue_is_empty(&conn->secondary_list)) { | 84 | { |
85 | - int ret; | ||
86 | + struct vhost_vdpa *v = dev->opaque; | ||
87 | + int ret = vhost_vdpa_get_dev_features(dev, features); | ||
88 | + | ||
89 | + if (ret == 0 && v->shadow_vqs_enabled) { | ||
90 | + /* Add SVQ logging capabilities */ | ||
91 | + *features |= BIT_ULL(VHOST_F_LOG_ALL); | ||
92 | + } | ||
93 | |||
94 | - ret = vhost_vdpa_call(dev, VHOST_GET_FEATURES, features); | ||
95 | - trace_vhost_vdpa_get_features(dev, *features); | ||
96 | return ret; | ||
97 | } | ||
98 | |||
99 | diff --git a/include/hw/virtio/vhost-vdpa.h b/include/hw/virtio/vhost-vdpa.h | ||
100 | index XXXXXXX..XXXXXXX 100644 | ||
101 | --- a/include/hw/virtio/vhost-vdpa.h | ||
102 | +++ b/include/hw/virtio/vhost-vdpa.h | ||
103 | @@ -XXX,XX +XXX,XX @@ typedef struct vhost_vdpa { | ||
104 | bool iotlb_batch_begin_sent; | ||
105 | MemoryListener listener; | ||
106 | struct vhost_vdpa_iova_range iova_range; | ||
107 | + uint64_t acked_features; | ||
108 | bool shadow_vqs_enabled; | ||
109 | /* IOVA mapping used by the Shadow Virtqueue */ | ||
110 | VhostIOVATree *iova_tree; | ||
76 | -- | 111 | -- |
77 | 2.7.4 | 112 | 2.7.4 |
78 | 113 | ||
79 | 114 | diff view generated by jsdifflib |