Add virtio-fs support (alternative to virtio-9p) #1016

stefanha · 2018-12-12T14:23:44Z

These patches add several configuration.toml options for the virtio-fs shared file system. For more information about virtio-fs, see https://virtio-fs.gitlab.io/.

This pull request is a preview. virtio-fs has not been merged in Linux or QEMU yet.

To use virtio-fs, set enable_virtio_fs true and set virtio_fs_daemon to the path of your passthrough_ll file system daemon (code available at https://gitlab.com/virtio-fs/libfuse). A full howto based on Kata 1.4.0 and a Fedora initramfs is here: https://virtio-fs.gitlab.io/howto-kata.html.

Tweaking the cache size, cache mode, and enabling shared version metadata is mostly for development at this point. The default settings are recommended.

grahamwhaley · 2018-12-12T14:35:29Z

Marking as WIP/DNM to show the preview status.
I'm presuming you know already @stefanha that the govmm file change for QEMU additions will need to go via the govmm repo and then get re-vendored back into here - so, you may want to post and progress that one separately (as it may need to land first).

devimc · 2018-12-12T14:35:47Z

thanks @stefanha, could you please file an issue and add fixes in your commit message, for example
fixes #99999

stefanha · 2018-12-13T09:24:27Z

thanks @stefanha, could you please file an issue and add fixes in your commit message, for example
fixes #99999

@devimc is the rationale for requiring an issue for every pull request documented somewhere? This PR is Do Not Merge and I can think of other situations where it doesn't seem useful to create issues. I'd like to understand why every PR needs to have an associated issue (even if it's just a formality/placeholder).

stefanha · 2018-12-13T09:27:06Z

Marking as WIP/DNM to show the preview status.
I'm presuming you know already @stefanha that the govmm file change for QEMU additions will need to go via the govmm repo and then get re-vendored back into here - so, you may want to post and progress that one separately (as it may need to land first).

Sure, will do that once there is an indication that the Linux kernel community is okay with virtio-fs.

jodh-intel

Thanks for raising @stefanha!

It would be useful if you could also update kata-env.go to show the new config options. This helps users since the runtime can be configured from multiple files so kata-runtime kata-env summaries the settings actually in use. For a minimal example of updating that command, see #937.

jodh-intel · 2018-12-13T09:17:45Z

virtcontainers/device/config/config.go

+
+	// These are only meaningful for vhost user fs devices
+	Tag string
+        CacheSize uint32


Nit: Indent incorrect.

jodh-intel · 2018-12-13T09:21:01Z

virtcontainers/qemu.go

@@ -549,6 +551,34 @@ func (q *qemu) startSandbox() error {
 		}
 	}()

+	if q.config.VirtioFS {
+                var cmd *exec.Cmd


Nit: indent.

gofmt could help

jodh-intel · 2018-12-13T09:41:50Z

virtcontainers/qemu.go

+			}
+			id := hex.EncodeToString(randBytes)
+
+			// TODO make this unique so multiple volumes can be added!


Please could you raise an issue for this and then change this comment to:

// FIXME: https://github.com/kata-containers/runtime/issues/...

Aside: which reminds me, we really need to re-enable that FIXME/TODO checker... 😄

CI: re-enable TODO/FIXME checker tests#469

devimc · 2018-12-13T15:01:26Z

@stefanha https://github.com/kata-containers/community/blob/master/CONTRIBUTING.md#patch-format

ghost · 2018-12-18T09:32:10Z

@stefanha Any updates here?

stefanha · 2018-12-18T13:23:15Z

On Tue, Dec 18, 2018 at 9:32 AM Yash D Jain ***@***.***> wrote: @stefanha <https://github.com/stefanha> Any updates here?

Getting Kata patches merged is gated on first getting virtio-fs into Linux, QEMU, and the VIRTIO specification. We are working on that but it will take weeks, especially with the holiday/new year seasion. If the Kata maintainers would like to merge the virtio-fs patches *before* Linux, QEMU, and VIRTIO support this feature upstream, then I am happy to work on that. Just let me know what makes sense for Kata. The roadmap for virtio-fs is to optimize performance and to harden the file system daemon for production (security, jailing). We are also keen on feedback and discussion to see what should be improved for best use by Kata Containers.

jodh-intel · 2019-01-07T16:30:53Z

I guess we're still blocked on upstream. But it might be a good idea to keep this PR branch conflict free so we're ready to jump on it when the time is right?

jodh-intel · 2019-01-23T11:56:31Z

Hi @stefanha - could you replace the questions marks in the table below so we have some tracking information on virtio-fs support?

Dependency	Status	URL
kernel	?	?
virtio spec	?	?
qemu	?	?

stefanha · 2019-01-29T10:24:02Z

@jodh-intel

Dependency	Status	URL
kernel	?	https://lkml.org/lkml/2018/12/10/615
virtio spec	?	(draft) stefanha/virtio@`e1cac37`
qemu	?	https://patchwork.kernel.org/cover/10722021/
libfuse	?	?

Currently the VIRTIO spec changes are being worked on. They are the dependency for everything else.

jodh-intel · 2019-01-29T10:29:44Z

Thanks @stefanha!

raravena80 · 2019-02-05T20:18:16Z

@stefanha ping, any updates? Thx! This would be awesome to get in!

stefanha · 2019-02-06T01:44:41Z

@raravena80 Work is ongoing but the status hasn't changed.

jodh-intel

Thanks for updating @stefanha!

A few comments.

jodh-intel · 2019-02-11T10:45:05Z

virtcontainers/qemu.go

@@ -1230,7 +1260,36 @@ func (q *qemu) addDevice(devInfo interface{}, devType deviceType) error {

 	switch v := devInfo.(type) {
 	case types.Volume:
-		q.qemuConfig.Devices = q.arch.append9PVolume(q.qemuConfig.Devices, v)
+		if q.config.VirtioFS {
+			q.Logger().Info("Adding vhost-user-fs-pci volume")


Could you tweak this message (and the 9p one below) to use a structure format to make it more parseable - something like:

q.Logger().WithField("volume-type", "vhost-user-fs-pci").Info("adding volume") : q.Logger().WithField("volume-type", "virtio-9p").Info("adding volume")

jodh-intel · 2019-02-11T10:47:04Z

vendor/github.com/intel/govmm/qemu/qemu.go

@@ -92,6 +92,9 @@ const (
 	//VhostUserBlk represents a block vhostuser device type.
 	VhostUserBlk DeviceDriver = "vhost-user-blk-pci"

+	//VhostUserFS represents a virtio-fs vhostuser device type


Please get these changes into https://github.com/intel/govmm first, and then re-vendor the changes into the runtime (see https://github.com/kata-containers/community/blob/master/VENDORING.md).

@stefanha now that the change has been merged in govmm, please update the govmm version by modifying Gopkg.toml, and run dep ensure -update. Your commit should include all the changes implied by those actions, resulting in having:

Gopkg.toml

Gopkg.lock

vendor/github.com/intel/govmm/*

jodh-intel · 2019-02-11T10:48:36Z

virtcontainers/device/drivers/vhost_user_fs.go

+}
+
+func (device *VhostUserFSDevice) Detach(devReceiver api.DeviceReceiver) error {
+	skip, err := device.bumpAttachCount(true)


This looks wrong - the true parameter is for attach only, but this is a detach method...?

jodh-intel · 2019-02-11T10:53:34Z

pkg/katautils/config.go

@@ -102,6 +102,11 @@ type hypervisor struct {
 	DefaultBridges          uint32 `toml:"default_bridges"`
 	Msize9p                 uint32 `toml:"msize_9p"`
 	DisableBlockDeviceUse   bool   `toml:"disable_block_device_use"`
+	VirtioFS                bool   `toml:"enable_virtio_fs"`


It would be useful to see something in the output of kata-runtime kata-env to show that virtio-fs is active. This saves users from having to find the correct configuration file (the runtime support multiple locations).

For examples, see #702, #937.

stefanha · 2019-02-12T05:58:47Z

Thanks for the review, @jodh-intel. I'll make the changes in the next revision.

amshinde · 2019-03-11T22:12:36Z

virtcontainers/qemu.go

+			}
+			id := hex.EncodeToString(randBytes)
+
+			// TODO make this unique so multiple volumes can be added!


Isnt this unique since you are using a random string(id) as part of the path ?

This comment is outdated. Multiple vhost-user-fs devices can be added but only when SharedVersions is false. I'll fix this in the next revision.

amshinde · 2019-03-11T22:15:47Z

virtcontainers/qemu.go

+
+			vhostDev := config.VhostUserDeviceAttrs{
+				Tag: v.MountTag,
+				Type: config.VhostUserFS,


Indent seems to be off here. @stefanha Can you format the files using gofmt?

Thanks, will fix in the next revision.

amshinde · 2019-03-11T22:26:25Z

cli/config/configuration-qemu.toml.in

@@ -121,6 +121,9 @@ virtio_fs_cache_size = @DEFVIRTIOFSCACHESIZE@
 #    Metadata, data, and pathname lookup are cached in guest and never expire.
 virtio_fs_cache = "@DEFVIRTIOFSCACHE@"

+# Use shared version metadata
+virtio_fs_shared_versions = @DEFVIRTIOFSSHAREDVERSIONS@


@stefanha Can you elaborate more on what this option does, pros and cons of using this?

This is an experimental feature and it's not ready for use. I will remove it for the time being since it relies on an external daemon (ireg) that isn't managed by kata-runtime yet. Once this feature is working it will make the cache=always mode coherent across shared volumes.

Agreed. I think the first step is to get virtio-fs early supported with the things that are reliable in virtio-fs for now. We can definitely add the ireg support with a later follow up PR :)

amshinde · 2019-03-11T22:32:53Z

@stefanha Can you take a look at the comments here? Although virtio-fs has not merged upstream, we are planning to add early support for virtio-fs. It would be great if you could submit a patch for VhostUserFs device to the govmm repo as well.

stefanha · 2019-03-12T08:13:46Z

@stefanha It would be great if you could submit a patch for VhostUserFs device to the govmm repo as well.

The VhostUserFs device has been merged into govmm:
kata-containers/govmm#86

sboeuf · 2019-03-14T17:33:09Z

@jodh-intel @devimc @grahamwhaley @jcvenegas @bergwolf @WeiZhang555 Please take a look as we're trying to move forward with the early support of virtio-fs for Kata. Think about this PR as a high prio one since we don't want to have it pending for months :)

jodh-intel

Hi @stefanha - comments added.

jodh-intel · 2019-03-14T17:53:19Z

virtcontainers/device/drivers/vhost_user_fs.go

+
+	defer func() {
+		if err == nil {
+			device.AttachCount = 1


I'm rather confused by this: the function calls device.bumpAttachCount(true) to increase AttachCount and then explicitly sets AttachCount = 1 if no error. I see the same logic in reverse in Detach(). It seems therefore that the only valid AttachCount values are 0 and 1 but can't that be checked for rather than overriding what bumpAttachCount() is doing?

Maybe a few comments explaining the intent would help here?

@jodh-intel I think I am the one to be blamed 😿

The logic is confusing as you said but it's right.

AttachCount == 0: do real attach, and set AttachCound = 1;

AttachCount==1: AttachCount++, but skip the real attact.

AttachCount==IntMax: error

I'll create a new PR to improve the readability.

@WeiZhang555

I'll create a new PR to improve the readability.

Great! Please open an issue for this and post the link here so that this comment does not get lost.

Here it is: #1392

It's assigned to me, I'll handle it today.

PR #1394 for this

devimc · 2019-03-14T18:14:43Z

virtcontainers/qemu.go

+		if q.config.VirtioFSSharedVersions {
+			args = append(args, "-o", "shared")
+		}
+		args = append(args, "/")


Q. why / ? are you given access to the whole rootfs? can you please add a comment?

This is removed in the next revision of this PR. The "/" was a dummy argument that has now been removed from virtiofsd. The shared directory is specified by the virtiofsd -o source path/to/dir option.

WeiZhang555 · 2019-03-18T09:59:30Z

Overall it looks good to me, nice work 😄

PR needs rebase, and I met some problem when test the PR locally, the filesystem keeps reporting "Busy" even a ls command can't get a result though the container is up successfully, still trying to figure out why...

sboeuf

Thanks @stefanha and @dagrh for this PR.
Overall LGTM, but I have a few comments. Hopefully we can merge it soon after you reworked it :)

One last comment about documentation, it would be very nice to have dedicated documentation for this new feature. And part of this documentation is to mention that it introduces a dependency on an external binary, the virtiofsd daemon. We will package it as part of Kata packages but it'd be nice to have it written/mentioned somewhere.

sboeuf · 2019-03-18T17:41:26Z

cli/config/configuration-qemu.toml.in

@@ -97,6 +97,9 @@ default_memory = @DEFMEMSZ@
 # 9pfs is used instead to pass the rootfs.
 disable_block_device_use = @DEFDISABLEBLOCK@

+# Use virtio-fs instead of 9p for shared filesystems.
+enable_virtio_fs = @DEFVIRTIOFS@


I think having a config option shared_fs with the default value virtio-9p would make more sense. And obviously in this case, we would need to switch it to virtio-fs if we wanted to use it instead of 9p.

Will fix in the next revision.

sboeuf · 2019-03-18T18:06:50Z

vendor/github.com/intel/govmm/qemu/qemu.go

@@ -92,6 +92,9 @@ const (
 	//VhostUserBlk represents a block vhostuser device type.
 	VhostUserBlk DeviceDriver = "vhost-user-blk-pci"

+	//VhostUserFS represents a virtio-fs vhostuser device type


@stefanha now that the change has been merged in govmm, please update the govmm version by modifying Gopkg.toml, and run dep ensure -update. Your commit should include all the changes implied by those actions, resulting in having:

Gopkg.toml

Gopkg.lock

vendor/github.com/intel/govmm/*

sboeuf · 2019-03-18T18:17:55Z

virtcontainers/device/config/config.go

@@ -174,6 +177,9 @@ type VhostUserDeviceAttrs struct {

 	// MacAddress is only meaningful for vhost user net device
 	MacAddress string
+
+	// These are only meaningful for vhost user fs devices


s/These are/Tag is/

Sorry, it's unclear from the context that additional fields are added by later patches:
// These are only meaningful for vhost user fs devices
Tag string
CacheSize uint32
Cache string
SharedVersions bool

sboeuf · 2019-03-18T18:23:34Z

virtcontainers/hypervisor.go

@@ -238,6 +238,9 @@ type HypervisorConfig struct {
 	// DisableBlockDeviceUse disallows a block device from being used.
 	DisableBlockDeviceUse bool

+	// VirtioFS enables virtio-fs instead of virtio-9p
+	VirtioFS bool


Following my other comment, I think we should define SharedFs as a string, as this would be more generic.

sboeuf · 2019-03-18T18:25:34Z

virtcontainers/device/drivers/vhost_user_fs.go

+
+	defer func() {
+		if err == nil {
+			device.AttachCount = 1


@WeiZhang555

I'll create a new PR to improve the readability.

Great! Please open an issue for this and post the link here so that this comment does not get lost.

sboeuf · 2019-03-18T19:19:30Z

virtcontainers/qemu.go

+		// of this child process.
+		sourcePath := filepath.Join(kataHostSharedDir, q.id)
+
+		cmd := exec.Command(q.config.VirtioFSDaemon, "-o", "virtio_socket=" + sockPath, "-o", "source=" + sourcePath, "/")


This part (spawning the virtiofsd binary) is not QEMU specific. It can be shared across hypervisors as long as they support virtio-fs. That's why it'd be better to have it implemented from hypervisor.go directly so that each hypervisor implementation could reuse the common code. Or, to invoke this from the generic sandbox.go code.

I looked into this but found sandbox.go and hypervisor.go don't offer convenient places to put this code. The issue is that qemu.go creats the vm directory so sandbox.go cannot launch the daemon before calling hypervisor.startSandbox(). And putting this function into hypervisor.go is awkward because it relies on paths that come from qemu.go (or another caller who has access to objects).

Let's leave it in qemu.go until there is a need to move it out for Firecracker.

sboeuf · 2019-03-18T19:23:45Z

Makefile

@@ -163,6 +163,8 @@ DEFENTROPYSOURCE := /dev/urandom
 DEFDISABLEBLOCK := false
 DEFVIRTIOFS := false
 DEFVIRTIOFSDAEMON :=
+# Default DAX mapping cache size in GiB


Maybe MiB would be more appropriate if we don't want a too large cache region by default.

Fixed in the next revision.

sboeuf · 2019-03-18T19:28:39Z

cli/config/configuration-qemu.toml.in

@@ -121,6 +121,9 @@ virtio_fs_cache_size = @DEFVIRTIOFSCACHESIZE@
 #    Metadata, data, and pathname lookup are cached in guest and never expire.
 virtio_fs_cache = "@DEFVIRTIOFSCACHE@"

+# Use shared version metadata
+virtio_fs_shared_versions = @DEFVIRTIOFSSHAREDVERSIONS@


Agreed. I think the first step is to get virtio-fs early supported with the things that are reliable in virtio-fs for now. We can definitely add the ireg support with a later follow up PR :)

sboeuf · 2019-04-19T20:38:29Z

@stefanha sorry about that! I've just removed them!

Shortlog: b3e7a9e Merge pull request kata-containers#91 from stefanha/virtio-fs-cache-size-mb 058cda0 qemu: use MiB instead of Gib for virtio-fs cache size Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

grahamwhaley · 2019-05-01T14:31:40Z

Looks like it has a rebase conflict with kata-env_test.go now.
What's the status here folks - feels like we are close to a merge of virtio-fs??
/cc @ganeshmaharaj who has been looking as well I think??

stefanha · 2019-05-01T14:35:26Z

Looks like it has a rebase conflict with kata-env_test.go now.

I'm rebasing right now and will push the resolved version.

Add a config option to select between virtio-9p and virtiofs. This option currently has no effect and will be used in a later patch. Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

Add a config option for the virtio-fs vhost-user daemon path. Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

bergwolf

Mostly looks good to me. Just two comments inline. Thanks!

bergwolf · 2019-05-05T07:20:49Z

cli/config/configuration-qemu.toml.in

+virtio_fs_daemon = "@DEFVIRTIOFSDAEMON@"
+
+# Default size of DAX cache in GiB
+virtio_fs_cache_size = @DEFVIRTIOFSCACHESIZE@


DEFVIRTIOFSCACHESIZE := 8192, it is in MiB instead?

Yes, thank you! Will fix in the next revision.

bergwolf · 2019-05-05T11:45:51Z

virtcontainers/qemu.go

+		go func() {
+			scanner := bufio.NewScanner(stderr)
+			for scanner.Scan() {
+				if strings.Contains(scanner.Text(), "Waiting for QEMU socket connection...") {


This is really hacky and may block other virtiofsd implementations. One possible fix is to check the existence of the unix socket before and after cmd.Start(). For now, please at least add a FIXME above so that we can revisit it later.

This is really hacky and may block other virtiofsd implementations. One possible fix is to check the existence of the unix socket before and after cmd.Start(). For now, please at least add a FIXME above so that we can revisit it later.

Reporting readiness from a daemon is common practice. For example, systemd has the sd_notify(3) interface where we would say "READY=1".

Polling for the existence of a UNIX domain socket has disadvantages:

A polling loop with a sleep() call adds latency. The solution I've posted immediately detects that the daemon is ready without adding latency or busy loops.

There is a race condition between the daemon's bind(2)+listen(2) calls and QEMU's connect(2). If QEMU is launched and attempts to connect between bind(2) and listen(2) the connect(2) call will fail. This complicates things: either QEMU needs to add reconnect logic or kata-runtime needs to connect and pass the fd to QEMU, or the daemon needs to allow unexpected disconnects (from kata-runtime).

The only hacky thing I see here is the "QEMU" in the string we are checking. virtiofsd works with other hypervisors and shouldn't know about QEMU. I will change the output of virtiofsd so this string is generic.

I've submitting a virtiofsd merge request to remove references to QEMU from the output:
https://gitlab.com/virtio-fs/qemu/merge_requests/4

@stefanha The exact virtiofsd output string becomes part of the ABI we rely on for virtiofsd readiness, -- that is the part I'm not very comfortable with. Systemd's use of sd_notify is standard because READY=1 etc state change is explicitly documented in man(3) sd_notify so it becomes part of the sd_notify API. If we decide to use the arbitrary string as virtiofsd readiness ABI, we need to declare it loud and clear so that any virtiofsd implementation knows it MUST print the exactly same string to stderr to announce its readiness. Right now it is hard-coded hidden in kata runtime code... To make it more acceptable, I think at least:

It needs to be defined as a static variable

We should document it explicitly somewhere obvious to virtiofsd implementers

OTOH, it seems that other vhost user daemons in qemu/contrib/ do not report their readiness. How do they handle the issue in practice?

Anyway, I'm OK to merge the PR as is since the feature is experimental. We can address the readiness issue (if necessary) in follow up PRs. @stefanha please fix the other comment and we can kick off the tests to see how CI says about it.

I have pushed the latest version with GiB -> MiB fixed and QEMU -> vhost-user in the stderr message.

I will take a look at implementing sd_notify(3) in virtiofsd. That way it will also work with systemd and we won't use stderr. A separate file descriptor is passed for sd_notify(3) and that's a bit cleaner. This will come as a separate pull request.

Thanks @stefanha!

When enable_virtio_fs is true, add a vhost-user-fs-pci for the kataShared volume instead of 9p. Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

Use virtio-fs instead of virtio-9p when virtio-fs is enabled. Fixes: kata-containers#1542 Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

Add VirtioFSCacheSize aka virtio_fs_cache_size option to set the size (in MiB) of the DAX cache. Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>

Several cache modes are supported by virtio-fs. They affect the performance and consistency characteristics of the file system. For the time being cache="none" is recommended, but the other modes can be experimented with. Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>

It's useful to know whether virtio-9p or virtio-fs is being used. Add the status to the kata-env output. Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

bergwolf · 2019-05-06T15:33:58Z

/test

This patches adds virtio-fs capability to the kata kernel along with config changes to enable the same on kata by default. The system will only be exercised when `shared_fs` is set to `virtio-fs` in the kata configuration file. the default still remains to be 9p Fixes: kata-containers#387 Depends-on: github.com/kata-containers/runtime#1016 Signed-off-by: Ganesh Maharaj Mahalingam <ganesh.mahalingam@intel.com>

Add configuration option to use virtio-fs. We will currently use nemu for testing the virtio-fs support. Depends-on: github.com/kata-containers/runtime#1016 Depends-on: github.com/kata-containers/packaging#484 Fixes: kata-containers#1536. Signed-off-by: Salvador Fuentes <salvador.fuentes@intel.com>

bergwolf · 2019-05-07T08:38:29Z

The failed cases are unrelated.

Add configuration option to use virtio-fs. We will currently use nemu for testing the virtio-fs support. Depends-on: github.com/kata-containers/runtime#1016 Depends-on: github.com/kata-containers/packaging#484 Fixes: kata-containers#1536. Signed-off-by: Salvador Fuentes <salvador.fuentes@intel.com>

sboeuf · 2019-05-07T18:33:32Z

@egernst 🍾 🍾 🍾

This patches adds virtio-fs capability to the kata kernel along with config changes to enable the same on kata by default. The system will only be exercised when `shared_fs` is set to `virtio-fs` in the kata configuration file. the default still remains to be 9p Fixes: kata-containers#387 Depends-on: github.com/kata-containers/runtime#1016 Signed-off-by: Ganesh Maharaj Mahalingam <ganesh.mahalingam@intel.com>

grahamwhaley added wip labels Dec 12, 2018

stefanha mentioned this pull request Dec 12, 2018

mount: add virtio-fs mount driver kata-containers/agent#434

Merged

jodh-intel reviewed Dec 13, 2018

View reviewed changes

stefanha force-pushed the virtio-fs-core branch from be9154d to e1546ce Compare February 11, 2019 09:03

stefanha requested a review from a team as a code owner February 11, 2019 09:03

jodh-intel reviewed Feb 11, 2019

View reviewed changes

stefanha force-pushed the virtio-fs-core branch from e1546ce to 97cea12 Compare February 20, 2019 14:48

amshinde reviewed Mar 11, 2019

View reviewed changes

jodh-intel reviewed Mar 14, 2019

View reviewed changes

devimc reviewed Mar 14, 2019

View reviewed changes

sboeuf reviewed Mar 18, 2019

View reviewed changes

sboeuf removed do-not-merge labels Apr 19, 2019

vendor: update govmm

0217077

Shortlog: b3e7a9e Merge pull request kata-containers#91 from stefanha/virtio-fs-cache-size-mb 058cda0 qemu: use MiB instead of Gib for virtio-fs cache size Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

stefanhaRH added 2 commits May 1, 2019 10:55

config: add shared_fs option

9e87fa2

Add a config option to select between virtio-9p and virtiofs. This option currently has no effect and will be used in a later patch. Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

config: add virtio_fs_daemon string

d690dff

Add a config option for the virtio-fs vhost-user daemon path. Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

stefanha force-pushed the virtio-fs-core branch from fae27a6 to 5555e70 Compare May 1, 2019 14:56

bergwolf reviewed May 5, 2019

View reviewed changes

stefanhaRH and others added 5 commits May 5, 2019 11:32

qemu: add vhost-user-fs-pci device instead of 9p

9480978

When enable_virtio_fs is true, add a vhost-user-fs-pci for the kataShared volume instead of 9p. Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

kata_agent: use virtio-fs shared dir in CreateSandbox

82d1a9d

Use virtio-fs instead of virtio-9p when virtio-fs is enabled. Fixes: kata-containers#1542 Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

virtiofs: Add cache size option

6767c1a

Add VirtioFSCacheSize aka virtio_fs_cache_size option to set the size (in MiB) of the DAX cache. Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>

kata-env: add SharedFS option to output

0a69eb8

It's useful to know whether virtio-9p or virtio-fs is being used. Add the status to the kata-env output. Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

stefanha force-pushed the virtio-fs-core branch from 5555e70 to 0a69eb8 Compare May 6, 2019 11:18

chavafg mentioned this pull request May 6, 2019

ci: Add virtio-fs support kata-containers/tests#1537

Merged

bergwolf approved these changes May 7, 2019

View reviewed changes

ganeshmaharaj approved these changes May 7, 2019

View reviewed changes

egernst merged commit 9a27ac2 into kata-containers:master May 7, 2019

minzcmu mentioned this pull request Jun 19, 2019

Evaluate Kata Containers over HyperD screwdriver-cd/screwdriver#818

Closed

Add virtio-fs support (alternative to virtio-9p) #1016

Add virtio-fs support (alternative to virtio-9p) #1016

Conversation

stefanha commented Dec 12, 2018

grahamwhaley commented Dec 12, 2018

devimc commented Dec 12, 2018

stefanha commented Dec 13, 2018 • edited Loading

stefanha commented Dec 13, 2018

jodh-intel left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

devimc commented Dec 13, 2018

ghost commented Dec 18, 2018

stefanha commented Dec 18, 2018 via email

jodh-intel commented Jan 7, 2019

jodh-intel commented Jan 23, 2019

stefanha commented Jan 29, 2019 • edited Loading

jodh-intel commented Jan 29, 2019

raravena80 commented Feb 5, 2019

stefanha commented Feb 6, 2019

jodh-intel left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

stefanha commented Feb 12, 2019

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

amshinde commented Mar 11, 2019

stefanha commented Mar 12, 2019

sboeuf commented Mar 14, 2019

jodh-intel left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

WeiZhang555 commented Mar 18, 2019 • edited Loading

sboeuf left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

sboeuf commented Apr 19, 2019 • edited Loading

grahamwhaley commented May 1, 2019

stefanha commented May 1, 2019

bergwolf left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

stefanha commented Dec 13, 2018 •

edited

Loading

stefanha commented Jan 29, 2019 •

edited

Loading

WeiZhang555 commented Mar 18, 2019 •

edited

Loading

sboeuf commented Apr 19, 2019 •

edited

Loading