diff mbox series

[v8,3/6] NFSD: add write_version to netlink command

Message ID 1036367642228283184f85715edc0e3227a8e3ae.1713209938.git.lorenzo@kernel.org (mailing list archive)
State Superseded
Headers show
Series convert write_threads, write_version and write_ports to netlink commands | expand

Checks

Context Check Description
netdev/series_format success Posting correctly formatted
netdev/tree_selection success Guessed tree name to be net-next
netdev/ynl success Generated files up to date; no warnings/errors; GEN HAS DIFF 2 files changed, 654 insertions(+);
netdev/fixes_present success Fixes tag not required for -next series
netdev/header_inline success No static functions without inline keyword in header files
netdev/build_32bit success Errors and warnings before: 928 this patch: 928
netdev/build_tools success Errors and warnings before: 0 this patch: 0
netdev/cc_maintainers warning 7 maintainers not CCed: pabeni@redhat.com trond.myklebust@hammerspace.com anna@kernel.org edumazet@google.com Dai.Ngo@oracle.com kolga@netapp.com tom@talpey.com
netdev/build_clang success Errors and warnings before: 937 this patch: 937
netdev/verify_signedoff success Signed-off-by tag matches author and committer
netdev/deprecated_api success None detected
netdev/check_selftest success No net selftest shell script
netdev/verify_fixes success No Fixes tag
netdev/build_allmodconfig_warn success Errors and warnings before: 939 this patch: 939
netdev/checkpatch warning CHECK: extern prototypes should be avoided in .h files WARNING: line length of 82 exceeds 80 columns WARNING: line length of 86 exceeds 80 columns WARNING: line length of 94 exceeds 80 columns
netdev/build_clang_rust success No Rust files in patch. Skipping build
netdev/kdoc fail Errors and warnings before: 9 this patch: 11
netdev/source_inline success Was 0 now: 0
netdev/contest success net-next-2024-04-16--15-00 (tests: 958)

Commit Message

Lorenzo Bianconi April 15, 2024, 7:44 p.m. UTC
Introduce write_version netlink command through a "declarative" interface.
This patch introduces a change in behavior since for version-set userspace
is expected to provide a NFS major/minor version list it wants to enable
while all the other ones will be disabled. (procfs write_version
command implements imperative interface where the admin writes +3/-3 to
enable/disable a single version.

Reviewed-by: Jeff Layton <jlayton@kernel.org>
Tested-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
---
 Documentation/netlink/specs/nfsd.yaml |  37 +++++++
 fs/nfsd/netlink.c                     |  24 +++++
 fs/nfsd/netlink.h                     |   5 +
 fs/nfsd/netns.h                       |   1 +
 fs/nfsd/nfsctl.c                      | 150 ++++++++++++++++++++++++++
 fs/nfsd/nfssvc.c                      |   3 +-
 include/uapi/linux/nfsd_netlink.h     |  18 ++++
 7 files changed, 236 insertions(+), 2 deletions(-)

Comments

NeilBrown April 16, 2024, 3:16 a.m. UTC | #1
On Tue, 16 Apr 2024, Lorenzo Bianconi wrote:
> Introduce write_version netlink command through a "declarative" interface.
> This patch introduces a change in behavior since for version-set userspace
> is expected to provide a NFS major/minor version list it wants to enable
> while all the other ones will be disabled. (procfs write_version
> command implements imperative interface where the admin writes +3/-3 to
> enable/disable a single version.

It seems a little weird to me that the interface always disables all
version, but then also allows individual versions to be disabled.

Would it be reasonable to simply ignore the "enabled" flag when setting
version, and just enable all versions listed??

Or maybe only enable those with the flag, and don't disable those
without the flag?

Those don't necessarily seem much better - but the current behaviour
still seems odd.

Also for getting the version, the doc says:

     doc: get nfs enabled versions

which I don't think it quite right.  The code reports all supported
versions, and indicates which of those are enabled.  So maybe:

     doc: get enabled status for all supported versions

I think that fact that it actually lists all supported versions is
useful and worth making explicit.

Thanks,
NeilBrown


> 
> Reviewed-by: Jeff Layton <jlayton@kernel.org>
> Tested-by: Jeff Layton <jlayton@kernel.org>
> Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
> ---
>  Documentation/netlink/specs/nfsd.yaml |  37 +++++++
>  fs/nfsd/netlink.c                     |  24 +++++
>  fs/nfsd/netlink.h                     |   5 +
>  fs/nfsd/netns.h                       |   1 +
>  fs/nfsd/nfsctl.c                      | 150 ++++++++++++++++++++++++++
>  fs/nfsd/nfssvc.c                      |   3 +-
>  include/uapi/linux/nfsd_netlink.h     |  18 ++++
>  7 files changed, 236 insertions(+), 2 deletions(-)
> 
> diff --git a/Documentation/netlink/specs/nfsd.yaml b/Documentation/netlink/specs/nfsd.yaml
> index cbe6c5fd6c4d..0396e8b3ea1f 100644
> --- a/Documentation/netlink/specs/nfsd.yaml
> +++ b/Documentation/netlink/specs/nfsd.yaml
> @@ -74,6 +74,26 @@ attribute-sets:
>        -
>          name: leasetime
>          type: u32
> +  -
> +    name: version
> +    attributes:
> +      -
> +        name: major
> +        type: u32
> +      -
> +        name: minor
> +        type: u32
> +      -
> +        name: enabled
> +        type: flag
> +  -
> +    name: server-proto
> +    attributes:
> +      -
> +        name: version
> +        type: nest
> +        nested-attributes: version
> +        multi-attr: true
>  
>  operations:
>    list:
> @@ -120,3 +140,20 @@ operations:
>              - threads
>              - gracetime
>              - leasetime
> +    -
> +      name: version-set
> +      doc: set nfs enabled versions
> +      attribute-set: server-proto
> +      flags: [ admin-perm ]
> +      do:
> +        request:
> +          attributes:
> +            - version
> +    -
> +      name: version-get
> +      doc: get nfs enabled versions
> +      attribute-set: server-proto
> +      do:
> +        reply:
> +          attributes:
> +            - version
> diff --git a/fs/nfsd/netlink.c b/fs/nfsd/netlink.c
> index 20a646af0324..bf5df9597288 100644
> --- a/fs/nfsd/netlink.c
> +++ b/fs/nfsd/netlink.c
> @@ -10,6 +10,13 @@
>  
>  #include <uapi/linux/nfsd_netlink.h>
>  
> +/* Common nested types */
> +const struct nla_policy nfsd_version_nl_policy[NFSD_A_VERSION_ENABLED + 1] = {
> +	[NFSD_A_VERSION_MAJOR] = { .type = NLA_U32, },
> +	[NFSD_A_VERSION_MINOR] = { .type = NLA_U32, },
> +	[NFSD_A_VERSION_ENABLED] = { .type = NLA_FLAG, },
> +};
> +
>  /* NFSD_CMD_THREADS_SET - do */
>  static const struct nla_policy nfsd_threads_set_nl_policy[NFSD_A_SERVER_WORKER_LEASETIME + 1] = {
>  	[NFSD_A_SERVER_WORKER_THREADS] = { .type = NLA_U32, },
> @@ -17,6 +24,11 @@ static const struct nla_policy nfsd_threads_set_nl_policy[NFSD_A_SERVER_WORKER_L
>  	[NFSD_A_SERVER_WORKER_LEASETIME] = { .type = NLA_U32, },
>  };
>  
> +/* NFSD_CMD_VERSION_SET - do */
> +static const struct nla_policy nfsd_version_set_nl_policy[NFSD_A_SERVER_PROTO_VERSION + 1] = {
> +	[NFSD_A_SERVER_PROTO_VERSION] = NLA_POLICY_NESTED(nfsd_version_nl_policy),
> +};
> +
>  /* Ops table for nfsd */
>  static const struct genl_split_ops nfsd_nl_ops[] = {
>  	{
> @@ -38,6 +50,18 @@ static const struct genl_split_ops nfsd_nl_ops[] = {
>  		.doit	= nfsd_nl_threads_get_doit,
>  		.flags	= GENL_CMD_CAP_DO,
>  	},
> +	{
> +		.cmd		= NFSD_CMD_VERSION_SET,
> +		.doit		= nfsd_nl_version_set_doit,
> +		.policy		= nfsd_version_set_nl_policy,
> +		.maxattr	= NFSD_A_SERVER_PROTO_VERSION,
> +		.flags		= GENL_ADMIN_PERM | GENL_CMD_CAP_DO,
> +	},
> +	{
> +		.cmd	= NFSD_CMD_VERSION_GET,
> +		.doit	= nfsd_nl_version_get_doit,
> +		.flags	= GENL_CMD_CAP_DO,
> +	},
>  };
>  
>  struct genl_family nfsd_nl_family __ro_after_init = {
> diff --git a/fs/nfsd/netlink.h b/fs/nfsd/netlink.h
> index 4137fac477e4..c7c0da275481 100644
> --- a/fs/nfsd/netlink.h
> +++ b/fs/nfsd/netlink.h
> @@ -11,6 +11,9 @@
>  
>  #include <uapi/linux/nfsd_netlink.h>
>  
> +/* Common nested types */
> +extern const struct nla_policy nfsd_version_nl_policy[NFSD_A_VERSION_ENABLED + 1];
> +
>  int nfsd_nl_rpc_status_get_start(struct netlink_callback *cb);
>  int nfsd_nl_rpc_status_get_done(struct netlink_callback *cb);
>  
> @@ -18,6 +21,8 @@ int nfsd_nl_rpc_status_get_dumpit(struct sk_buff *skb,
>  				  struct netlink_callback *cb);
>  int nfsd_nl_threads_set_doit(struct sk_buff *skb, struct genl_info *info);
>  int nfsd_nl_threads_get_doit(struct sk_buff *skb, struct genl_info *info);
> +int nfsd_nl_version_set_doit(struct sk_buff *skb, struct genl_info *info);
> +int nfsd_nl_version_get_doit(struct sk_buff *skb, struct genl_info *info);
>  
>  extern struct genl_family nfsd_nl_family;
>  
> diff --git a/fs/nfsd/netns.h b/fs/nfsd/netns.h
> index d4be519b5734..14ec15656320 100644
> --- a/fs/nfsd/netns.h
> +++ b/fs/nfsd/netns.h
> @@ -218,6 +218,7 @@ struct nfsd_net {
>  /* Simple check to find out if a given net was properly initialized */
>  #define nfsd_netns_ready(nn) ((nn)->sessionid_hashtbl)
>  
> +extern bool nfsd_support_version(int vers);
>  extern void nfsd_netns_free_versions(struct nfsd_net *nn);
>  
>  extern unsigned int nfsd_net_id;
> diff --git a/fs/nfsd/nfsctl.c b/fs/nfsd/nfsctl.c
> index 38a5df03981b..2c8929ef79e9 100644
> --- a/fs/nfsd/nfsctl.c
> +++ b/fs/nfsd/nfsctl.c
> @@ -1757,6 +1757,156 @@ int nfsd_nl_threads_get_doit(struct sk_buff *skb, struct genl_info *info)
>  	return err;
>  }
>  
> +/**
> + * nfsd_nl_version_set_doit - set the nfs enabled versions
> + * @skb: reply buffer
> + * @info: netlink metadata and command arguments
> + *
> + * Return 0 on success or a negative errno.
> + */
> +int nfsd_nl_version_set_doit(struct sk_buff *skb, struct genl_info *info)
> +{
> +	const struct nlattr *attr;
> +	struct nfsd_net *nn;
> +	int i, rem;
> +
> +	if (GENL_REQ_ATTR_CHECK(info, NFSD_A_SERVER_PROTO_VERSION))
> +		return -EINVAL;
> +
> +	mutex_lock(&nfsd_mutex);
> +
> +	nn = net_generic(genl_info_net(info), nfsd_net_id);
> +	if (nn->nfsd_serv) {
> +		mutex_unlock(&nfsd_mutex);
> +		return -EBUSY;
> +	}
> +
> +	/* clear current supported versions. */
> +	nfsd_vers(nn, 2, NFSD_CLEAR);
> +	nfsd_vers(nn, 3, NFSD_CLEAR);
> +	for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++)
> +		nfsd_minorversion(nn, i, NFSD_CLEAR);
> +
> +	nlmsg_for_each_attr(attr, info->nlhdr, GENL_HDRLEN, rem) {
> +		struct nlattr *tb[NFSD_A_VERSION_MAX + 1];
> +		u32 major, minor = 0;
> +		bool enabled;
> +
> +		if (nla_type(attr) != NFSD_A_SERVER_PROTO_VERSION)
> +			continue;
> +
> +		if (nla_parse_nested(tb, NFSD_A_VERSION_MAX, attr,
> +				     nfsd_version_nl_policy, info->extack) < 0)
> +			continue;
> +
> +		if (!tb[NFSD_A_VERSION_MAJOR])
> +			continue;
> +
> +		major = nla_get_u32(tb[NFSD_A_VERSION_MAJOR]);
> +		if (tb[NFSD_A_VERSION_MINOR])
> +			minor = nla_get_u32(tb[NFSD_A_VERSION_MINOR]);
> +
> +		enabled = nla_get_flag(tb[NFSD_A_VERSION_ENABLED]);
> +
> +		switch (major) {
> +		case 4:
> +			nfsd_minorversion(nn, minor, enabled ? NFSD_SET : NFSD_CLEAR);
> +			break;
> +		case 3:
> +		case 2:
> +			if (!minor)
> +				nfsd_vers(nn, major, enabled ? NFSD_SET : NFSD_CLEAR);
> +			break;
> +		default:
> +			break;
> +		}
> +	}
> +
> +	mutex_unlock(&nfsd_mutex);
> +
> +	return 0;
> +}
> +
> +/**
> + * nfsd_nl_version_get_doit - get the nfs enabled versions
> + * @skb: reply buffer
> + * @info: netlink metadata and command arguments
> + *
> + * Return 0 on success or a negative errno.
> + */
> +int nfsd_nl_version_get_doit(struct sk_buff *skb, struct genl_info *info)
> +{
> +	struct nfsd_net *nn;
> +	int i, err;
> +	void *hdr;
> +
> +	skb = genlmsg_new(GENLMSG_DEFAULT_SIZE, GFP_KERNEL);
> +	if (!skb)
> +		return -ENOMEM;
> +
> +	hdr = genlmsg_iput(skb, info);
> +	if (!hdr) {
> +		err = -EMSGSIZE;
> +		goto err_free_msg;
> +	}
> +
> +	mutex_lock(&nfsd_mutex);
> +	nn = net_generic(genl_info_net(info), nfsd_net_id);
> +
> +	for (i = 2; i <= 4; i++) {
> +		int j;
> +
> +		for (j = 0; j <= NFSD_SUPPORTED_MINOR_VERSION; j++) {
> +			struct nlattr *attr;
> +
> +			/* Don't record any versions the kernel doesn't have
> +			 * compiled in
> +			 */
> +			if (!nfsd_support_version(i))
> +				continue;
> +
> +			/* NFSv{2,3} does not support minor numbers */
> +			if (i < 4 && j)
> +				continue;
> +
> +			attr = nla_nest_start(skb,
> +					      NFSD_A_SERVER_PROTO_VERSION);
> +			if (!attr) {
> +				err = -EINVAL;
> +				goto err_nfsd_unlock;
> +			}
> +
> +			if (nla_put_u32(skb, NFSD_A_VERSION_MAJOR, i) ||
> +			    nla_put_u32(skb, NFSD_A_VERSION_MINOR, j)) {
> +				err = -EINVAL;
> +				goto err_nfsd_unlock;
> +			}
> +
> +			/* Set the enabled flag if the version is enabled */
> +			if (nfsd_vers(nn, i, NFSD_TEST) &&
> +			    (i < 4 || nfsd_minorversion(nn, j, NFSD_TEST)) &&
> +			    nla_put_flag(skb, NFSD_A_VERSION_ENABLED)) {
> +				err = -EINVAL;
> +				goto err_nfsd_unlock;
> +			}
> +
> +			nla_nest_end(skb, attr);
> +		}
> +	}
> +
> +	mutex_unlock(&nfsd_mutex);
> +	genlmsg_end(skb, hdr);
> +
> +	return genlmsg_reply(skb, info);
> +
> +err_nfsd_unlock:
> +	mutex_unlock(&nfsd_mutex);
> +err_free_msg:
> +	nlmsg_free(skb);
> +
> +	return err;
> +}
> +
>  /**
>   * nfsd_net_init - Prepare the nfsd_net portion of a new net namespace
>   * @net: a freshly-created network namespace
> diff --git a/fs/nfsd/nfssvc.c b/fs/nfsd/nfssvc.c
> index ca193f7ff0e1..4fc91f50138a 100644
> --- a/fs/nfsd/nfssvc.c
> +++ b/fs/nfsd/nfssvc.c
> @@ -133,8 +133,7 @@ struct svc_program		nfsd_program = {
>  	.pg_rpcbind_set		= nfsd_rpcbind_set,
>  };
>  
> -static bool
> -nfsd_support_version(int vers)
> +bool nfsd_support_version(int vers)
>  {
>  	if (vers >= NFSD_MINVERS && vers < NFSD_NRVERS)
>  		return nfsd_version[vers] != NULL;
> diff --git a/include/uapi/linux/nfsd_netlink.h b/include/uapi/linux/nfsd_netlink.h
> index ccc78a5ee650..8a0a2b344923 100644
> --- a/include/uapi/linux/nfsd_netlink.h
> +++ b/include/uapi/linux/nfsd_netlink.h
> @@ -38,10 +38,28 @@ enum {
>  	NFSD_A_SERVER_WORKER_MAX = (__NFSD_A_SERVER_WORKER_MAX - 1)
>  };
>  
> +enum {
> +	NFSD_A_VERSION_MAJOR = 1,
> +	NFSD_A_VERSION_MINOR,
> +	NFSD_A_VERSION_ENABLED,
> +
> +	__NFSD_A_VERSION_MAX,
> +	NFSD_A_VERSION_MAX = (__NFSD_A_VERSION_MAX - 1)
> +};
> +
> +enum {
> +	NFSD_A_SERVER_PROTO_VERSION = 1,
> +
> +	__NFSD_A_SERVER_PROTO_MAX,
> +	NFSD_A_SERVER_PROTO_MAX = (__NFSD_A_SERVER_PROTO_MAX - 1)
> +};
> +
>  enum {
>  	NFSD_CMD_RPC_STATUS_GET = 1,
>  	NFSD_CMD_THREADS_SET,
>  	NFSD_CMD_THREADS_GET,
> +	NFSD_CMD_VERSION_SET,
> +	NFSD_CMD_VERSION_GET,
>  
>  	__NFSD_CMD_MAX,
>  	NFSD_CMD_MAX = (__NFSD_CMD_MAX - 1)
> -- 
> 2.44.0
> 
>
Jeff Layton April 16, 2024, 10:26 a.m. UTC | #2
On Tue, 2024-04-16 at 13:16 +1000, NeilBrown wrote:
> On Tue, 16 Apr 2024, Lorenzo Bianconi wrote:
> > Introduce write_version netlink command through a "declarative" interface.
> > This patch introduces a change in behavior since for version-set userspace
> > is expected to provide a NFS major/minor version list it wants to enable
> > while all the other ones will be disabled. (procfs write_version
> > command implements imperative interface where the admin writes +3/-3 to
> > enable/disable a single version.
> 
> It seems a little weird to me that the interface always disables all
> version, but then also allows individual versions to be disabled.
> 
> Would it be reasonable to simply ignore the "enabled" flag when setting
> version, and just enable all versions listed??
>
> Or maybe only enable those with the flag, and don't disable those
> without the flag?
>
> Those don't necessarily seem much better - but the current behaviour
> still seems odd.
> 

I think it makes sense.

We disable all versions, and enable any that have the "enabled" flag set
in the call from userland. Userland technically needn't send down the
versions that are disabled in the call, but the current userland program
does.

I worry about imperative interfaces that might only say -- "enable v4.1,
but disable v3" and leave the others in their current state. That
requires that both the kernel and userland keep state about what
versions are currently enabled and disabled, and it's possible to get
that wrong.

My thinking was that by using a declarative interface like this, we
eliminate ambiguity in how these interfaces are supposed to work. The
client sends down an entire version map in one fell swoop, and we know
exactly what the result should look like.

Note that you can enable or disable just a single version with the
userland tool, but this way all of that complexity stays in userland.

> Also for getting the version, the doc says:
> 
>      doc: get nfs enabled versions
> 
> which I don't think it quite right.  The code reports all supported
> versions, and indicates which of those are enabled.  So maybe:
> 
>      doc: get enabled status for all supported versions
> 
>
> I think that fact that it actually lists all supported versions is
> useful and worth making explicit.
> 
> 

Agreed. We should make that change before merging anything.

> > 
> > Reviewed-by: Jeff Layton <jlayton@kernel.org>
> > Tested-by: Jeff Layton <jlayton@kernel.org>
> > Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
> > ---
> >  Documentation/netlink/specs/nfsd.yaml |  37 +++++++
> >  fs/nfsd/netlink.c                     |  24 +++++
> >  fs/nfsd/netlink.h                     |   5 +
> >  fs/nfsd/netns.h                       |   1 +
> >  fs/nfsd/nfsctl.c                      | 150 ++++++++++++++++++++++++++
> >  fs/nfsd/nfssvc.c                      |   3 +-
> >  include/uapi/linux/nfsd_netlink.h     |  18 ++++
> >  7 files changed, 236 insertions(+), 2 deletions(-)
> > 
> > diff --git a/Documentation/netlink/specs/nfsd.yaml b/Documentation/netlink/specs/nfsd.yaml
> > index cbe6c5fd6c4d..0396e8b3ea1f 100644
> > --- a/Documentation/netlink/specs/nfsd.yaml
> > +++ b/Documentation/netlink/specs/nfsd.yaml
> > @@ -74,6 +74,26 @@ attribute-sets:
> >        -
> >          name: leasetime
> >          type: u32
> > +  -
> > +    name: version
> > +    attributes:
> > +      -
> > +        name: major
> > +        type: u32
> > +      -
> > +        name: minor
> > +        type: u32
> > +      -
> > +        name: enabled
> > +        type: flag
> > +  -
> > +    name: server-proto
> > +    attributes:
> > +      -
> > +        name: version
> > +        type: nest
> > +        nested-attributes: version
> > +        multi-attr: true
> >  
> >  operations:
> >    list:
> > @@ -120,3 +140,20 @@ operations:
> >              - threads
> >              - gracetime
> >              - leasetime
> > +    -
> > +      name: version-set
> > +      doc: set nfs enabled versions
> > +      attribute-set: server-proto
> > +      flags: [ admin-perm ]
> > +      do:
> > +        request:
> > +          attributes:
> > +            - version
> > +    -
> > +      name: version-get
> > +      doc: get nfs enabled versions
> > +      attribute-set: server-proto
> > +      do:
> > +        reply:
> > +          attributes:
> > +            - version
> > diff --git a/fs/nfsd/netlink.c b/fs/nfsd/netlink.c
> > index 20a646af0324..bf5df9597288 100644
> > --- a/fs/nfsd/netlink.c
> > +++ b/fs/nfsd/netlink.c
> > @@ -10,6 +10,13 @@
> >  
> >  #include <uapi/linux/nfsd_netlink.h>
> >  
> > +/* Common nested types */
> > +const struct nla_policy nfsd_version_nl_policy[NFSD_A_VERSION_ENABLED + 1] = {
> > +	[NFSD_A_VERSION_MAJOR] = { .type = NLA_U32, },
> > +	[NFSD_A_VERSION_MINOR] = { .type = NLA_U32, },
> > +	[NFSD_A_VERSION_ENABLED] = { .type = NLA_FLAG, },
> > +};
> > +
> >  /* NFSD_CMD_THREADS_SET - do */
> >  static const struct nla_policy nfsd_threads_set_nl_policy[NFSD_A_SERVER_WORKER_LEASETIME + 1] = {
> >  	[NFSD_A_SERVER_WORKER_THREADS] = { .type = NLA_U32, },
> > @@ -17,6 +24,11 @@ static const struct nla_policy nfsd_threads_set_nl_policy[NFSD_A_SERVER_WORKER_L
> >  	[NFSD_A_SERVER_WORKER_LEASETIME] = { .type = NLA_U32, },
> >  };
> >  
> > +/* NFSD_CMD_VERSION_SET - do */
> > +static const struct nla_policy nfsd_version_set_nl_policy[NFSD_A_SERVER_PROTO_VERSION + 1] = {
> > +	[NFSD_A_SERVER_PROTO_VERSION] = NLA_POLICY_NESTED(nfsd_version_nl_policy),
> > +};
> > +
> >  /* Ops table for nfsd */
> >  static const struct genl_split_ops nfsd_nl_ops[] = {
> >  	{
> > @@ -38,6 +50,18 @@ static const struct genl_split_ops nfsd_nl_ops[] = {
> >  		.doit	= nfsd_nl_threads_get_doit,
> >  		.flags	= GENL_CMD_CAP_DO,
> >  	},
> > +	{
> > +		.cmd		= NFSD_CMD_VERSION_SET,
> > +		.doit		= nfsd_nl_version_set_doit,
> > +		.policy		= nfsd_version_set_nl_policy,
> > +		.maxattr	= NFSD_A_SERVER_PROTO_VERSION,
> > +		.flags		= GENL_ADMIN_PERM | GENL_CMD_CAP_DO,
> > +	},
> > +	{
> > +		.cmd	= NFSD_CMD_VERSION_GET,
> > +		.doit	= nfsd_nl_version_get_doit,
> > +		.flags	= GENL_CMD_CAP_DO,
> > +	},
> >  };
> >  
> >  struct genl_family nfsd_nl_family __ro_after_init = {
> > diff --git a/fs/nfsd/netlink.h b/fs/nfsd/netlink.h
> > index 4137fac477e4..c7c0da275481 100644
> > --- a/fs/nfsd/netlink.h
> > +++ b/fs/nfsd/netlink.h
> > @@ -11,6 +11,9 @@
> >  
> >  #include <uapi/linux/nfsd_netlink.h>
> >  
> > +/* Common nested types */
> > +extern const struct nla_policy nfsd_version_nl_policy[NFSD_A_VERSION_ENABLED + 1];
> > +
> >  int nfsd_nl_rpc_status_get_start(struct netlink_callback *cb);
> >  int nfsd_nl_rpc_status_get_done(struct netlink_callback *cb);
> >  
> > @@ -18,6 +21,8 @@ int nfsd_nl_rpc_status_get_dumpit(struct sk_buff *skb,
> >  				  struct netlink_callback *cb);
> >  int nfsd_nl_threads_set_doit(struct sk_buff *skb, struct genl_info *info);
> >  int nfsd_nl_threads_get_doit(struct sk_buff *skb, struct genl_info *info);
> > +int nfsd_nl_version_set_doit(struct sk_buff *skb, struct genl_info *info);
> > +int nfsd_nl_version_get_doit(struct sk_buff *skb, struct genl_info *info);
> >  
> >  extern struct genl_family nfsd_nl_family;
> >  
> > diff --git a/fs/nfsd/netns.h b/fs/nfsd/netns.h
> > index d4be519b5734..14ec15656320 100644
> > --- a/fs/nfsd/netns.h
> > +++ b/fs/nfsd/netns.h
> > @@ -218,6 +218,7 @@ struct nfsd_net {
> >  /* Simple check to find out if a given net was properly initialized */
> >  #define nfsd_netns_ready(nn) ((nn)->sessionid_hashtbl)
> >  
> > +extern bool nfsd_support_version(int vers);
> >  extern void nfsd_netns_free_versions(struct nfsd_net *nn);
> >  
> >  extern unsigned int nfsd_net_id;
> > diff --git a/fs/nfsd/nfsctl.c b/fs/nfsd/nfsctl.c
> > index 38a5df03981b..2c8929ef79e9 100644
> > --- a/fs/nfsd/nfsctl.c
> > +++ b/fs/nfsd/nfsctl.c
> > @@ -1757,6 +1757,156 @@ int nfsd_nl_threads_get_doit(struct sk_buff *skb, struct genl_info *info)
> >  	return err;
> >  }
> >  
> > +/**
> > + * nfsd_nl_version_set_doit - set the nfs enabled versions
> > + * @skb: reply buffer
> > + * @info: netlink metadata and command arguments
> > + *
> > + * Return 0 on success or a negative errno.
> > + */
> > +int nfsd_nl_version_set_doit(struct sk_buff *skb, struct genl_info *info)
> > +{
> > +	const struct nlattr *attr;
> > +	struct nfsd_net *nn;
> > +	int i, rem;
> > +
> > +	if (GENL_REQ_ATTR_CHECK(info, NFSD_A_SERVER_PROTO_VERSION))
> > +		return -EINVAL;
> > +
> > +	mutex_lock(&nfsd_mutex);
> > +
> > +	nn = net_generic(genl_info_net(info), nfsd_net_id);
> > +	if (nn->nfsd_serv) {
> > +		mutex_unlock(&nfsd_mutex);
> > +		return -EBUSY;
> > +	}
> > +
> > +	/* clear current supported versions. */
> > +	nfsd_vers(nn, 2, NFSD_CLEAR);
> > +	nfsd_vers(nn, 3, NFSD_CLEAR);
> > +	for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++)
> > +		nfsd_minorversion(nn, i, NFSD_CLEAR);
> > +
> > +	nlmsg_for_each_attr(attr, info->nlhdr, GENL_HDRLEN, rem) {
> > +		struct nlattr *tb[NFSD_A_VERSION_MAX + 1];
> > +		u32 major, minor = 0;
> > +		bool enabled;
> > +
> > +		if (nla_type(attr) != NFSD_A_SERVER_PROTO_VERSION)
> > +			continue;
> > +
> > +		if (nla_parse_nested(tb, NFSD_A_VERSION_MAX, attr,
> > +				     nfsd_version_nl_policy, info->extack) < 0)
> > +			continue;
> > +
> > +		if (!tb[NFSD_A_VERSION_MAJOR])
> > +			continue;
> > +
> > +		major = nla_get_u32(tb[NFSD_A_VERSION_MAJOR]);
> > +		if (tb[NFSD_A_VERSION_MINOR])
> > +			minor = nla_get_u32(tb[NFSD_A_VERSION_MINOR]);
> > +
> > +		enabled = nla_get_flag(tb[NFSD_A_VERSION_ENABLED]);
> > +
> > +		switch (major) {
> > +		case 4:
> > +			nfsd_minorversion(nn, minor, enabled ? NFSD_SET : NFSD_CLEAR);
> > +			break;
> > +		case 3:
> > +		case 2:
> > +			if (!minor)
> > +				nfsd_vers(nn, major, enabled ? NFSD_SET : NFSD_CLEAR);
> > +			break;
> > +		default:
> > +			break;
> > +		}
> > +	}
> > +
> > +	mutex_unlock(&nfsd_mutex);
> > +
> > +	return 0;
> > +}
> > +
> > +/**
> > + * nfsd_nl_version_get_doit - get the nfs enabled versions
> > + * @skb: reply buffer
> > + * @info: netlink metadata and command arguments
> > + *
> > + * Return 0 on success or a negative errno.
> > + */
> > +int nfsd_nl_version_get_doit(struct sk_buff *skb, struct genl_info *info)
> > +{
> > +	struct nfsd_net *nn;
> > +	int i, err;
> > +	void *hdr;
> > +
> > +	skb = genlmsg_new(GENLMSG_DEFAULT_SIZE, GFP_KERNEL);
> > +	if (!skb)
> > +		return -ENOMEM;
> > +
> > +	hdr = genlmsg_iput(skb, info);
> > +	if (!hdr) {
> > +		err = -EMSGSIZE;
> > +		goto err_free_msg;
> > +	}
> > +
> > +	mutex_lock(&nfsd_mutex);
> > +	nn = net_generic(genl_info_net(info), nfsd_net_id);
> > +
> > +	for (i = 2; i <= 4; i++) {
> > +		int j;
> > +
> > +		for (j = 0; j <= NFSD_SUPPORTED_MINOR_VERSION; j++) {
> > +			struct nlattr *attr;
> > +
> > +			/* Don't record any versions the kernel doesn't have
> > +			 * compiled in
> > +			 */
> > +			if (!nfsd_support_version(i))
> > +				continue;
> > +
> > +			/* NFSv{2,3} does not support minor numbers */
> > +			if (i < 4 && j)
> > +				continue;
> > +
> > +			attr = nla_nest_start(skb,
> > +					      NFSD_A_SERVER_PROTO_VERSION);
> > +			if (!attr) {
> > +				err = -EINVAL;
> > +				goto err_nfsd_unlock;
> > +			}
> > +
> > +			if (nla_put_u32(skb, NFSD_A_VERSION_MAJOR, i) ||
> > +			    nla_put_u32(skb, NFSD_A_VERSION_MINOR, j)) {
> > +				err = -EINVAL;
> > +				goto err_nfsd_unlock;
> > +			}
> > +
> > +			/* Set the enabled flag if the version is enabled */
> > +			if (nfsd_vers(nn, i, NFSD_TEST) &&
> > +			    (i < 4 || nfsd_minorversion(nn, j, NFSD_TEST)) &&
> > +			    nla_put_flag(skb, NFSD_A_VERSION_ENABLED)) {
> > +				err = -EINVAL;
> > +				goto err_nfsd_unlock;
> > +			}
> > +
> > +			nla_nest_end(skb, attr);
> > +		}
> > +	}
> > +
> > +	mutex_unlock(&nfsd_mutex);
> > +	genlmsg_end(skb, hdr);
> > +
> > +	return genlmsg_reply(skb, info);
> > +
> > +err_nfsd_unlock:
> > +	mutex_unlock(&nfsd_mutex);
> > +err_free_msg:
> > +	nlmsg_free(skb);
> > +
> > +	return err;
> > +}
> > +
> >  /**
> >   * nfsd_net_init - Prepare the nfsd_net portion of a new net namespace
> >   * @net: a freshly-created network namespace
> > diff --git a/fs/nfsd/nfssvc.c b/fs/nfsd/nfssvc.c
> > index ca193f7ff0e1..4fc91f50138a 100644
> > --- a/fs/nfsd/nfssvc.c
> > +++ b/fs/nfsd/nfssvc.c
> > @@ -133,8 +133,7 @@ struct svc_program		nfsd_program = {
> >  	.pg_rpcbind_set		= nfsd_rpcbind_set,
> >  };
> >  
> > -static bool
> > -nfsd_support_version(int vers)
> > +bool nfsd_support_version(int vers)
> >  {
> >  	if (vers >= NFSD_MINVERS && vers < NFSD_NRVERS)
> >  		return nfsd_version[vers] != NULL;
> > diff --git a/include/uapi/linux/nfsd_netlink.h b/include/uapi/linux/nfsd_netlink.h
> > index ccc78a5ee650..8a0a2b344923 100644
> > --- a/include/uapi/linux/nfsd_netlink.h
> > +++ b/include/uapi/linux/nfsd_netlink.h
> > @@ -38,10 +38,28 @@ enum {
> >  	NFSD_A_SERVER_WORKER_MAX = (__NFSD_A_SERVER_WORKER_MAX - 1)
> >  };
> >  
> > +enum {
> > +	NFSD_A_VERSION_MAJOR = 1,
> > +	NFSD_A_VERSION_MINOR,
> > +	NFSD_A_VERSION_ENABLED,
> > +
> > +	__NFSD_A_VERSION_MAX,
> > +	NFSD_A_VERSION_MAX = (__NFSD_A_VERSION_MAX - 1)
> > +};
> > +
> > +enum {
> > +	NFSD_A_SERVER_PROTO_VERSION = 1,
> > +
> > +	__NFSD_A_SERVER_PROTO_MAX,
> > +	NFSD_A_SERVER_PROTO_MAX = (__NFSD_A_SERVER_PROTO_MAX - 1)
> > +};
> > +
> >  enum {
> >  	NFSD_CMD_RPC_STATUS_GET = 1,
> >  	NFSD_CMD_THREADS_SET,
> >  	NFSD_CMD_THREADS_GET,
> > +	NFSD_CMD_VERSION_SET,
> > +	NFSD_CMD_VERSION_GET,
> >  
> >  	__NFSD_CMD_MAX,
> >  	NFSD_CMD_MAX = (__NFSD_CMD_MAX - 1)
> > -- 
> > 2.44.0
> > 
> > 
>
NeilBrown April 16, 2024, 9:48 p.m. UTC | #3
On Tue, 16 Apr 2024, Jeff Layton wrote:
> On Tue, 2024-04-16 at 13:16 +1000, NeilBrown wrote:
> > On Tue, 16 Apr 2024, Lorenzo Bianconi wrote:
> > > Introduce write_version netlink command through a "declarative" interface.
> > > This patch introduces a change in behavior since for version-set userspace
> > > is expected to provide a NFS major/minor version list it wants to enable
> > > while all the other ones will be disabled. (procfs write_version
> > > command implements imperative interface where the admin writes +3/-3 to
> > > enable/disable a single version.
> > 
> > It seems a little weird to me that the interface always disables all
> > version, but then also allows individual versions to be disabled.
> > 
> > Would it be reasonable to simply ignore the "enabled" flag when setting
> > version, and just enable all versions listed??
> >
> > Or maybe only enable those with the flag, and don't disable those
> > without the flag?
> >
> > Those don't necessarily seem much better - but the current behaviour
> > still seems odd.
> > 
> 
> I think it makes sense.
> 
> We disable all versions, and enable any that have the "enabled" flag set
> in the call from userland. Userland technically needn't send down the
> versions that are disabled in the call, but the current userland program
> does.
> 
> I worry about imperative interfaces that might only say -- "enable v4.1,
> but disable v3" and leave the others in their current state. That
> requires that both the kernel and userland keep state about what
> versions are currently enabled and disabled, and it's possible to get
> that wrong.

I understand and support your aversion for imperative interfaces.
But this interface, as currently implemented, looks somewhat imperative.
The message sent to the kernel could enable some interfaces and then
disable them.  I know that isn't the intent, but it is what the code
supports.  Hence "weird".

We could add code to make that sort of thing impossible, but there isn't
much point.  Better to make it syntactically impossible.

Realistically there will never be NFSv4.3 as there is no need - new
features can be added incrementally.  So we could just pass an array of
5 active flags: 2,3,4.0,4.1,4.2.  I suspect you wouldn't like that and
I'm not sure that I do either.  A "read" would return the same array
with 3 possible states: unavailable, disabled, enabled.  (Maybe the
array could be variable length so 5.0 could be added one day...).

I haven't managed to come up with anything *better* than the current
proposal and I don't want to stand in its way, but I wanted to highlight
that it looks weird - as much imperative as declarative - in case
someone else might be able to come up with a better alternative.

Thanks,
NeilBrown


> 
> My thinking was that by using a declarative interface like this, we
> eliminate ambiguity in how these interfaces are supposed to work. The
> client sends down an entire version map in one fell swoop, and we know
> exactly what the result should look like.
> 
> Note that you can enable or disable just a single version with the
> userland tool, but this way all of that complexity stays in userland.
> 
> > Also for getting the version, the doc says:
> > 
> >      doc: get nfs enabled versions
> > 
> > which I don't think it quite right.  The code reports all supported
> > versions, and indicates which of those are enabled.  So maybe:
> > 
> >      doc: get enabled status for all supported versions
> > 
> >
> > I think that fact that it actually lists all supported versions is
> > useful and worth making explicit.
> > 
> > 
> 
> Agreed. We should make that change before merging anything.
> 
> > > 
> > > Reviewed-by: Jeff Layton <jlayton@kernel.org>
> > > Tested-by: Jeff Layton <jlayton@kernel.org>
> > > Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
> > > ---
> > >  Documentation/netlink/specs/nfsd.yaml |  37 +++++++
> > >  fs/nfsd/netlink.c                     |  24 +++++
> > >  fs/nfsd/netlink.h                     |   5 +
> > >  fs/nfsd/netns.h                       |   1 +
> > >  fs/nfsd/nfsctl.c                      | 150 ++++++++++++++++++++++++++
> > >  fs/nfsd/nfssvc.c                      |   3 +-
> > >  include/uapi/linux/nfsd_netlink.h     |  18 ++++
> > >  7 files changed, 236 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/Documentation/netlink/specs/nfsd.yaml b/Documentation/netlink/specs/nfsd.yaml
> > > index cbe6c5fd6c4d..0396e8b3ea1f 100644
> > > --- a/Documentation/netlink/specs/nfsd.yaml
> > > +++ b/Documentation/netlink/specs/nfsd.yaml
> > > @@ -74,6 +74,26 @@ attribute-sets:
> > >        -
> > >          name: leasetime
> > >          type: u32
> > > +  -
> > > +    name: version
> > > +    attributes:
> > > +      -
> > > +        name: major
> > > +        type: u32
> > > +      -
> > > +        name: minor
> > > +        type: u32
> > > +      -
> > > +        name: enabled
> > > +        type: flag
> > > +  -
> > > +    name: server-proto
> > > +    attributes:
> > > +      -
> > > +        name: version
> > > +        type: nest
> > > +        nested-attributes: version
> > > +        multi-attr: true
> > >  
> > >  operations:
> > >    list:
> > > @@ -120,3 +140,20 @@ operations:
> > >              - threads
> > >              - gracetime
> > >              - leasetime
> > > +    -
> > > +      name: version-set
> > > +      doc: set nfs enabled versions
> > > +      attribute-set: server-proto
> > > +      flags: [ admin-perm ]
> > > +      do:
> > > +        request:
> > > +          attributes:
> > > +            - version
> > > +    -
> > > +      name: version-get
> > > +      doc: get nfs enabled versions
> > > +      attribute-set: server-proto
> > > +      do:
> > > +        reply:
> > > +          attributes:
> > > +            - version
> > > diff --git a/fs/nfsd/netlink.c b/fs/nfsd/netlink.c
> > > index 20a646af0324..bf5df9597288 100644
> > > --- a/fs/nfsd/netlink.c
> > > +++ b/fs/nfsd/netlink.c
> > > @@ -10,6 +10,13 @@
> > >  
> > >  #include <uapi/linux/nfsd_netlink.h>
> > >  
> > > +/* Common nested types */
> > > +const struct nla_policy nfsd_version_nl_policy[NFSD_A_VERSION_ENABLED + 1] = {
> > > +	[NFSD_A_VERSION_MAJOR] = { .type = NLA_U32, },
> > > +	[NFSD_A_VERSION_MINOR] = { .type = NLA_U32, },
> > > +	[NFSD_A_VERSION_ENABLED] = { .type = NLA_FLAG, },
> > > +};
> > > +
> > >  /* NFSD_CMD_THREADS_SET - do */
> > >  static const struct nla_policy nfsd_threads_set_nl_policy[NFSD_A_SERVER_WORKER_LEASETIME + 1] = {
> > >  	[NFSD_A_SERVER_WORKER_THREADS] = { .type = NLA_U32, },
> > > @@ -17,6 +24,11 @@ static const struct nla_policy nfsd_threads_set_nl_policy[NFSD_A_SERVER_WORKER_L
> > >  	[NFSD_A_SERVER_WORKER_LEASETIME] = { .type = NLA_U32, },
> > >  };
> > >  
> > > +/* NFSD_CMD_VERSION_SET - do */
> > > +static const struct nla_policy nfsd_version_set_nl_policy[NFSD_A_SERVER_PROTO_VERSION + 1] = {
> > > +	[NFSD_A_SERVER_PROTO_VERSION] = NLA_POLICY_NESTED(nfsd_version_nl_policy),
> > > +};
> > > +
> > >  /* Ops table for nfsd */
> > >  static const struct genl_split_ops nfsd_nl_ops[] = {
> > >  	{
> > > @@ -38,6 +50,18 @@ static const struct genl_split_ops nfsd_nl_ops[] = {
> > >  		.doit	= nfsd_nl_threads_get_doit,
> > >  		.flags	= GENL_CMD_CAP_DO,
> > >  	},
> > > +	{
> > > +		.cmd		= NFSD_CMD_VERSION_SET,
> > > +		.doit		= nfsd_nl_version_set_doit,
> > > +		.policy		= nfsd_version_set_nl_policy,
> > > +		.maxattr	= NFSD_A_SERVER_PROTO_VERSION,
> > > +		.flags		= GENL_ADMIN_PERM | GENL_CMD_CAP_DO,
> > > +	},
> > > +	{
> > > +		.cmd	= NFSD_CMD_VERSION_GET,
> > > +		.doit	= nfsd_nl_version_get_doit,
> > > +		.flags	= GENL_CMD_CAP_DO,
> > > +	},
> > >  };
> > >  
> > >  struct genl_family nfsd_nl_family __ro_after_init = {
> > > diff --git a/fs/nfsd/netlink.h b/fs/nfsd/netlink.h
> > > index 4137fac477e4..c7c0da275481 100644
> > > --- a/fs/nfsd/netlink.h
> > > +++ b/fs/nfsd/netlink.h
> > > @@ -11,6 +11,9 @@
> > >  
> > >  #include <uapi/linux/nfsd_netlink.h>
> > >  
> > > +/* Common nested types */
> > > +extern const struct nla_policy nfsd_version_nl_policy[NFSD_A_VERSION_ENABLED + 1];
> > > +
> > >  int nfsd_nl_rpc_status_get_start(struct netlink_callback *cb);
> > >  int nfsd_nl_rpc_status_get_done(struct netlink_callback *cb);
> > >  
> > > @@ -18,6 +21,8 @@ int nfsd_nl_rpc_status_get_dumpit(struct sk_buff *skb,
> > >  				  struct netlink_callback *cb);
> > >  int nfsd_nl_threads_set_doit(struct sk_buff *skb, struct genl_info *info);
> > >  int nfsd_nl_threads_get_doit(struct sk_buff *skb, struct genl_info *info);
> > > +int nfsd_nl_version_set_doit(struct sk_buff *skb, struct genl_info *info);
> > > +int nfsd_nl_version_get_doit(struct sk_buff *skb, struct genl_info *info);
> > >  
> > >  extern struct genl_family nfsd_nl_family;
> > >  
> > > diff --git a/fs/nfsd/netns.h b/fs/nfsd/netns.h
> > > index d4be519b5734..14ec15656320 100644
> > > --- a/fs/nfsd/netns.h
> > > +++ b/fs/nfsd/netns.h
> > > @@ -218,6 +218,7 @@ struct nfsd_net {
> > >  /* Simple check to find out if a given net was properly initialized */
> > >  #define nfsd_netns_ready(nn) ((nn)->sessionid_hashtbl)
> > >  
> > > +extern bool nfsd_support_version(int vers);
> > >  extern void nfsd_netns_free_versions(struct nfsd_net *nn);
> > >  
> > >  extern unsigned int nfsd_net_id;
> > > diff --git a/fs/nfsd/nfsctl.c b/fs/nfsd/nfsctl.c
> > > index 38a5df03981b..2c8929ef79e9 100644
> > > --- a/fs/nfsd/nfsctl.c
> > > +++ b/fs/nfsd/nfsctl.c
> > > @@ -1757,6 +1757,156 @@ int nfsd_nl_threads_get_doit(struct sk_buff *skb, struct genl_info *info)
> > >  	return err;
> > >  }
> > >  
> > > +/**
> > > + * nfsd_nl_version_set_doit - set the nfs enabled versions
> > > + * @skb: reply buffer
> > > + * @info: netlink metadata and command arguments
> > > + *
> > > + * Return 0 on success or a negative errno.
> > > + */
> > > +int nfsd_nl_version_set_doit(struct sk_buff *skb, struct genl_info *info)
> > > +{
> > > +	const struct nlattr *attr;
> > > +	struct nfsd_net *nn;
> > > +	int i, rem;
> > > +
> > > +	if (GENL_REQ_ATTR_CHECK(info, NFSD_A_SERVER_PROTO_VERSION))
> > > +		return -EINVAL;
> > > +
> > > +	mutex_lock(&nfsd_mutex);
> > > +
> > > +	nn = net_generic(genl_info_net(info), nfsd_net_id);
> > > +	if (nn->nfsd_serv) {
> > > +		mutex_unlock(&nfsd_mutex);
> > > +		return -EBUSY;
> > > +	}
> > > +
> > > +	/* clear current supported versions. */
> > > +	nfsd_vers(nn, 2, NFSD_CLEAR);
> > > +	nfsd_vers(nn, 3, NFSD_CLEAR);
> > > +	for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++)
> > > +		nfsd_minorversion(nn, i, NFSD_CLEAR);
> > > +
> > > +	nlmsg_for_each_attr(attr, info->nlhdr, GENL_HDRLEN, rem) {
> > > +		struct nlattr *tb[NFSD_A_VERSION_MAX + 1];
> > > +		u32 major, minor = 0;
> > > +		bool enabled;
> > > +
> > > +		if (nla_type(attr) != NFSD_A_SERVER_PROTO_VERSION)
> > > +			continue;
> > > +
> > > +		if (nla_parse_nested(tb, NFSD_A_VERSION_MAX, attr,
> > > +				     nfsd_version_nl_policy, info->extack) < 0)
> > > +			continue;
> > > +
> > > +		if (!tb[NFSD_A_VERSION_MAJOR])
> > > +			continue;
> > > +
> > > +		major = nla_get_u32(tb[NFSD_A_VERSION_MAJOR]);
> > > +		if (tb[NFSD_A_VERSION_MINOR])
> > > +			minor = nla_get_u32(tb[NFSD_A_VERSION_MINOR]);
> > > +
> > > +		enabled = nla_get_flag(tb[NFSD_A_VERSION_ENABLED]);
> > > +
> > > +		switch (major) {
> > > +		case 4:
> > > +			nfsd_minorversion(nn, minor, enabled ? NFSD_SET : NFSD_CLEAR);
> > > +			break;
> > > +		case 3:
> > > +		case 2:
> > > +			if (!minor)
> > > +				nfsd_vers(nn, major, enabled ? NFSD_SET : NFSD_CLEAR);
> > > +			break;
> > > +		default:
> > > +			break;
> > > +		}
> > > +	}
> > > +
> > > +	mutex_unlock(&nfsd_mutex);
> > > +
> > > +	return 0;
> > > +}
> > > +
> > > +/**
> > > + * nfsd_nl_version_get_doit - get the nfs enabled versions
> > > + * @skb: reply buffer
> > > + * @info: netlink metadata and command arguments
> > > + *
> > > + * Return 0 on success or a negative errno.
> > > + */
> > > +int nfsd_nl_version_get_doit(struct sk_buff *skb, struct genl_info *info)
> > > +{
> > > +	struct nfsd_net *nn;
> > > +	int i, err;
> > > +	void *hdr;
> > > +
> > > +	skb = genlmsg_new(GENLMSG_DEFAULT_SIZE, GFP_KERNEL);
> > > +	if (!skb)
> > > +		return -ENOMEM;
> > > +
> > > +	hdr = genlmsg_iput(skb, info);
> > > +	if (!hdr) {
> > > +		err = -EMSGSIZE;
> > > +		goto err_free_msg;
> > > +	}
> > > +
> > > +	mutex_lock(&nfsd_mutex);
> > > +	nn = net_generic(genl_info_net(info), nfsd_net_id);
> > > +
> > > +	for (i = 2; i <= 4; i++) {
> > > +		int j;
> > > +
> > > +		for (j = 0; j <= NFSD_SUPPORTED_MINOR_VERSION; j++) {
> > > +			struct nlattr *attr;
> > > +
> > > +			/* Don't record any versions the kernel doesn't have
> > > +			 * compiled in
> > > +			 */
> > > +			if (!nfsd_support_version(i))
> > > +				continue;
> > > +
> > > +			/* NFSv{2,3} does not support minor numbers */
> > > +			if (i < 4 && j)
> > > +				continue;
> > > +
> > > +			attr = nla_nest_start(skb,
> > > +					      NFSD_A_SERVER_PROTO_VERSION);
> > > +			if (!attr) {
> > > +				err = -EINVAL;
> > > +				goto err_nfsd_unlock;
> > > +			}
> > > +
> > > +			if (nla_put_u32(skb, NFSD_A_VERSION_MAJOR, i) ||
> > > +			    nla_put_u32(skb, NFSD_A_VERSION_MINOR, j)) {
> > > +				err = -EINVAL;
> > > +				goto err_nfsd_unlock;
> > > +			}
> > > +
> > > +			/* Set the enabled flag if the version is enabled */
> > > +			if (nfsd_vers(nn, i, NFSD_TEST) &&
> > > +			    (i < 4 || nfsd_minorversion(nn, j, NFSD_TEST)) &&
> > > +			    nla_put_flag(skb, NFSD_A_VERSION_ENABLED)) {
> > > +				err = -EINVAL;
> > > +				goto err_nfsd_unlock;
> > > +			}
> > > +
> > > +			nla_nest_end(skb, attr);
> > > +		}
> > > +	}
> > > +
> > > +	mutex_unlock(&nfsd_mutex);
> > > +	genlmsg_end(skb, hdr);
> > > +
> > > +	return genlmsg_reply(skb, info);
> > > +
> > > +err_nfsd_unlock:
> > > +	mutex_unlock(&nfsd_mutex);
> > > +err_free_msg:
> > > +	nlmsg_free(skb);
> > > +
> > > +	return err;
> > > +}
> > > +
> > >  /**
> > >   * nfsd_net_init - Prepare the nfsd_net portion of a new net namespace
> > >   * @net: a freshly-created network namespace
> > > diff --git a/fs/nfsd/nfssvc.c b/fs/nfsd/nfssvc.c
> > > index ca193f7ff0e1..4fc91f50138a 100644
> > > --- a/fs/nfsd/nfssvc.c
> > > +++ b/fs/nfsd/nfssvc.c
> > > @@ -133,8 +133,7 @@ struct svc_program		nfsd_program = {
> > >  	.pg_rpcbind_set		= nfsd_rpcbind_set,
> > >  };
> > >  
> > > -static bool
> > > -nfsd_support_version(int vers)
> > > +bool nfsd_support_version(int vers)
> > >  {
> > >  	if (vers >= NFSD_MINVERS && vers < NFSD_NRVERS)
> > >  		return nfsd_version[vers] != NULL;
> > > diff --git a/include/uapi/linux/nfsd_netlink.h b/include/uapi/linux/nfsd_netlink.h
> > > index ccc78a5ee650..8a0a2b344923 100644
> > > --- a/include/uapi/linux/nfsd_netlink.h
> > > +++ b/include/uapi/linux/nfsd_netlink.h
> > > @@ -38,10 +38,28 @@ enum {
> > >  	NFSD_A_SERVER_WORKER_MAX = (__NFSD_A_SERVER_WORKER_MAX - 1)
> > >  };
> > >  
> > > +enum {
> > > +	NFSD_A_VERSION_MAJOR = 1,
> > > +	NFSD_A_VERSION_MINOR,
> > > +	NFSD_A_VERSION_ENABLED,
> > > +
> > > +	__NFSD_A_VERSION_MAX,
> > > +	NFSD_A_VERSION_MAX = (__NFSD_A_VERSION_MAX - 1)
> > > +};
> > > +
> > > +enum {
> > > +	NFSD_A_SERVER_PROTO_VERSION = 1,
> > > +
> > > +	__NFSD_A_SERVER_PROTO_MAX,
> > > +	NFSD_A_SERVER_PROTO_MAX = (__NFSD_A_SERVER_PROTO_MAX - 1)
> > > +};
> > > +
> > >  enum {
> > >  	NFSD_CMD_RPC_STATUS_GET = 1,
> > >  	NFSD_CMD_THREADS_SET,
> > >  	NFSD_CMD_THREADS_GET,
> > > +	NFSD_CMD_VERSION_SET,
> > > +	NFSD_CMD_VERSION_GET,
> > >  
> > >  	__NFSD_CMD_MAX,
> > >  	NFSD_CMD_MAX = (__NFSD_CMD_MAX - 1)
> > > -- 
> > > 2.44.0
> > > 
> > > 
> > 
> 
> -- 
> Jeff Layton <jlayton@kernel.org>
>
Jeff Layton April 16, 2024, 10:33 p.m. UTC | #4
On Wed, 2024-04-17 at 07:48 +1000, NeilBrown wrote:
> On Tue, 16 Apr 2024, Jeff Layton wrote:
> > On Tue, 2024-04-16 at 13:16 +1000, NeilBrown wrote:
> > > On Tue, 16 Apr 2024, Lorenzo Bianconi wrote:
> > > > Introduce write_version netlink command through a "declarative" interface.
> > > > This patch introduces a change in behavior since for version-set userspace
> > > > is expected to provide a NFS major/minor version list it wants to enable
> > > > while all the other ones will be disabled. (procfs write_version
> > > > command implements imperative interface where the admin writes +3/-3 to
> > > > enable/disable a single version.
> > > 
> > > It seems a little weird to me that the interface always disables all
> > > version, but then also allows individual versions to be disabled.
> > > 
> > > Would it be reasonable to simply ignore the "enabled" flag when setting
> > > version, and just enable all versions listed??
> > > 
> > > Or maybe only enable those with the flag, and don't disable those
> > > without the flag?
> > > 
> > > Those don't necessarily seem much better - but the current behaviour
> > > still seems odd.
> > > 
> > 
> > I think it makes sense.
> > 
> > We disable all versions, and enable any that have the "enabled" flag set
> > in the call from userland. Userland technically needn't send down the
> > versions that are disabled in the call, but the current userland program
> > does.
> > 
> > I worry about imperative interfaces that might only say -- "enable v4.1,
> > but disable v3" and leave the others in their current state. That
> > requires that both the kernel and userland keep state about what
> > versions are currently enabled and disabled, and it's possible to get
> > that wrong.
> 
> I understand and support your aversion for imperative interfaces.
> But this interface, as currently implemented, looks somewhat imperative.
> The message sent to the kernel could enable some interfaces and then
> disable them.  I know that isn't the intent, but it is what the code
> supports.  Hence "weird".
> 
> We could add code to make that sort of thing impossible, but there isn't
> much point.  Better to make it syntactically impossible.
> 
> Realistically there will never be NFSv4.3 as there is no need - new
> features can be added incrementally. 
> 

There is no need _so_far_. Who knows what the future will bring? Maybe
we'll need v4.3 in order to add some needed feature?

>  So we could just pass an array of
> 5 active flags: 2,3,4.0,4.1,4.2.  I suspect you wouldn't like that and
> I'm not sure that I do either.  A "read" would return the same array
> with 3 possible states: unavailable, disabled, enabled.  (Maybe the
> array could be variable length so 5.0 could be added one day...).
> 

A set of flags is basically what this interface is. They just happen to
also be labeled with the major and minorversion. I think that's a good
thing.

>
> I haven't managed to come up with anything *better* than the current
> proposal and I don't want to stand in its way, but I wanted to highlight
> that it looks weird - as much imperative as declarative - in case
> someone else might be able to come up with a better alternative.
> 

The intention was to create a symmetrical interface. We have 2
operations: a "get" that asks "what versions are currently supported and
are they enabled?", and a "set" that says "here is the new state of
version enablement".

The userland tool always sends down a complete set of versions. The
kernel is (currently) more forgiving, and treats omitted versions as
being disabled. The kernel could require that every supported version be
represented in the "set" operation, if that's more desirable.

> > 
> > My thinking was that by using a declarative interface like this, we
> > eliminate ambiguity in how these interfaces are supposed to work. The
> > client sends down an entire version map in one fell swoop, and we know
> > exactly what the result should look like.
> > 
> > Note that you can enable or disable just a single version with the
> > userland tool, but this way all of that complexity stays in userland.
> > 
> > > Also for getting the version, the doc says:
> > > 
> > >      doc: get nfs enabled versions
> > > 
> > > which I don't think it quite right.  The code reports all supported
> > > versions, and indicates which of those are enabled.  So maybe:
> > > 
> > >      doc: get enabled status for all supported versions
> > > 
> > > 
> > > I think that fact that it actually lists all supported versions is
> > > useful and worth making explicit.
> > > 
> > > 
> > 
> > Agreed. We should make that change before merging anything.
> > 
> > > > 
> > > > Reviewed-by: Jeff Layton <jlayton@kernel.org>
> > > > Tested-by: Jeff Layton <jlayton@kernel.org>
> > > > Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
> > > > ---
> > > >  Documentation/netlink/specs/nfsd.yaml |  37 +++++++
> > > >  fs/nfsd/netlink.c                     |  24 +++++
> > > >  fs/nfsd/netlink.h                     |   5 +
> > > >  fs/nfsd/netns.h                       |   1 +
> > > >  fs/nfsd/nfsctl.c                      | 150 ++++++++++++++++++++++++++
> > > >  fs/nfsd/nfssvc.c                      |   3 +-
> > > >  include/uapi/linux/nfsd_netlink.h     |  18 ++++
> > > >  7 files changed, 236 insertions(+), 2 deletions(-)
> > > > 
> > > > diff --git a/Documentation/netlink/specs/nfsd.yaml b/Documentation/netlink/specs/nfsd.yaml
> > > > index cbe6c5fd6c4d..0396e8b3ea1f 100644
> > > > --- a/Documentation/netlink/specs/nfsd.yaml
> > > > +++ b/Documentation/netlink/specs/nfsd.yaml
> > > > @@ -74,6 +74,26 @@ attribute-sets:
> > > >        -
> > > >          name: leasetime
> > > >          type: u32
> > > > +  -
> > > > +    name: version
> > > > +    attributes:
> > > > +      -
> > > > +        name: major
> > > > +        type: u32
> > > > +      -
> > > > +        name: minor
> > > > +        type: u32
> > > > +      -
> > > > +        name: enabled
> > > > +        type: flag
> > > > +  -
> > > > +    name: server-proto
> > > > +    attributes:
> > > > +      -
> > > > +        name: version
> > > > +        type: nest
> > > > +        nested-attributes: version
> > > > +        multi-attr: true
> > > >  
> > > >  operations:
> > > >    list:
> > > > @@ -120,3 +140,20 @@ operations:
> > > >              - threads
> > > >              - gracetime
> > > >              - leasetime
> > > > +    -
> > > > +      name: version-set
> > > > +      doc: set nfs enabled versions
> > > > +      attribute-set: server-proto
> > > > +      flags: [ admin-perm ]
> > > > +      do:
> > > > +        request:
> > > > +          attributes:
> > > > +            - version
> > > > +    -
> > > > +      name: version-get
> > > > +      doc: get nfs enabled versions
> > > > +      attribute-set: server-proto
> > > > +      do:
> > > > +        reply:
> > > > +          attributes:
> > > > +            - version
> > > > diff --git a/fs/nfsd/netlink.c b/fs/nfsd/netlink.c
> > > > index 20a646af0324..bf5df9597288 100644
> > > > --- a/fs/nfsd/netlink.c
> > > > +++ b/fs/nfsd/netlink.c
> > > > @@ -10,6 +10,13 @@
> > > >  
> > > >  #include <uapi/linux/nfsd_netlink.h>
> > > >  
> > > > +/* Common nested types */
> > > > +const struct nla_policy nfsd_version_nl_policy[NFSD_A_VERSION_ENABLED + 1] = {
> > > > +	[NFSD_A_VERSION_MAJOR] = { .type = NLA_U32, },
> > > > +	[NFSD_A_VERSION_MINOR] = { .type = NLA_U32, },
> > > > +	[NFSD_A_VERSION_ENABLED] = { .type = NLA_FLAG, },
> > > > +};
> > > > +
> > > >  /* NFSD_CMD_THREADS_SET - do */
> > > >  static const struct nla_policy nfsd_threads_set_nl_policy[NFSD_A_SERVER_WORKER_LEASETIME + 1] = {
> > > >  	[NFSD_A_SERVER_WORKER_THREADS] = { .type = NLA_U32, },
> > > > @@ -17,6 +24,11 @@ static const struct nla_policy nfsd_threads_set_nl_policy[NFSD_A_SERVER_WORKER_L
> > > >  	[NFSD_A_SERVER_WORKER_LEASETIME] = { .type = NLA_U32, },
> > > >  };
> > > >  
> > > > +/* NFSD_CMD_VERSION_SET - do */
> > > > +static const struct nla_policy nfsd_version_set_nl_policy[NFSD_A_SERVER_PROTO_VERSION + 1] = {
> > > > +	[NFSD_A_SERVER_PROTO_VERSION] = NLA_POLICY_NESTED(nfsd_version_nl_policy),
> > > > +};
> > > > +
> > > >  /* Ops table for nfsd */
> > > >  static const struct genl_split_ops nfsd_nl_ops[] = {
> > > >  	{
> > > > @@ -38,6 +50,18 @@ static const struct genl_split_ops nfsd_nl_ops[] = {
> > > >  		.doit	= nfsd_nl_threads_get_doit,
> > > >  		.flags	= GENL_CMD_CAP_DO,
> > > >  	},
> > > > +	{
> > > > +		.cmd		= NFSD_CMD_VERSION_SET,
> > > > +		.doit		= nfsd_nl_version_set_doit,
> > > > +		.policy		= nfsd_version_set_nl_policy,
> > > > +		.maxattr	= NFSD_A_SERVER_PROTO_VERSION,
> > > > +		.flags		= GENL_ADMIN_PERM | GENL_CMD_CAP_DO,
> > > > +	},
> > > > +	{
> > > > +		.cmd	= NFSD_CMD_VERSION_GET,
> > > > +		.doit	= nfsd_nl_version_get_doit,
> > > > +		.flags	= GENL_CMD_CAP_DO,
> > > > +	},
> > > >  };
> > > >  
> > > >  struct genl_family nfsd_nl_family __ro_after_init = {
> > > > diff --git a/fs/nfsd/netlink.h b/fs/nfsd/netlink.h
> > > > index 4137fac477e4..c7c0da275481 100644
> > > > --- a/fs/nfsd/netlink.h
> > > > +++ b/fs/nfsd/netlink.h
> > > > @@ -11,6 +11,9 @@
> > > >  
> > > >  #include <uapi/linux/nfsd_netlink.h>
> > > >  
> > > > +/* Common nested types */
> > > > +extern const struct nla_policy nfsd_version_nl_policy[NFSD_A_VERSION_ENABLED + 1];
> > > > +
> > > >  int nfsd_nl_rpc_status_get_start(struct netlink_callback *cb);
> > > >  int nfsd_nl_rpc_status_get_done(struct netlink_callback *cb);
> > > >  
> > > > @@ -18,6 +21,8 @@ int nfsd_nl_rpc_status_get_dumpit(struct sk_buff *skb,
> > > >  				  struct netlink_callback *cb);
> > > >  int nfsd_nl_threads_set_doit(struct sk_buff *skb, struct genl_info *info);
> > > >  int nfsd_nl_threads_get_doit(struct sk_buff *skb, struct genl_info *info);
> > > > +int nfsd_nl_version_set_doit(struct sk_buff *skb, struct genl_info *info);
> > > > +int nfsd_nl_version_get_doit(struct sk_buff *skb, struct genl_info *info);
> > > >  
> > > >  extern struct genl_family nfsd_nl_family;
> > > >  
> > > > diff --git a/fs/nfsd/netns.h b/fs/nfsd/netns.h
> > > > index d4be519b5734..14ec15656320 100644
> > > > --- a/fs/nfsd/netns.h
> > > > +++ b/fs/nfsd/netns.h
> > > > @@ -218,6 +218,7 @@ struct nfsd_net {
> > > >  /* Simple check to find out if a given net was properly initialized */
> > > >  #define nfsd_netns_ready(nn) ((nn)->sessionid_hashtbl)
> > > >  
> > > > +extern bool nfsd_support_version(int vers);
> > > >  extern void nfsd_netns_free_versions(struct nfsd_net *nn);
> > > >  
> > > >  extern unsigned int nfsd_net_id;
> > > > diff --git a/fs/nfsd/nfsctl.c b/fs/nfsd/nfsctl.c
> > > > index 38a5df03981b..2c8929ef79e9 100644
> > > > --- a/fs/nfsd/nfsctl.c
> > > > +++ b/fs/nfsd/nfsctl.c
> > > > @@ -1757,6 +1757,156 @@ int nfsd_nl_threads_get_doit(struct sk_buff *skb, struct genl_info *info)
> > > >  	return err;
> > > >  }
> > > >  
> > > > +/**
> > > > + * nfsd_nl_version_set_doit - set the nfs enabled versions
> > > > + * @skb: reply buffer
> > > > + * @info: netlink metadata and command arguments
> > > > + *
> > > > + * Return 0 on success or a negative errno.
> > > > + */
> > > > +int nfsd_nl_version_set_doit(struct sk_buff *skb, struct genl_info *info)
> > > > +{
> > > > +	const struct nlattr *attr;
> > > > +	struct nfsd_net *nn;
> > > > +	int i, rem;
> > > > +
> > > > +	if (GENL_REQ_ATTR_CHECK(info, NFSD_A_SERVER_PROTO_VERSION))
> > > > +		return -EINVAL;
> > > > +
> > > > +	mutex_lock(&nfsd_mutex);
> > > > +
> > > > +	nn = net_generic(genl_info_net(info), nfsd_net_id);
> > > > +	if (nn->nfsd_serv) {
> > > > +		mutex_unlock(&nfsd_mutex);
> > > > +		return -EBUSY;
> > > > +	}
> > > > +
> > > > +	/* clear current supported versions. */
> > > > +	nfsd_vers(nn, 2, NFSD_CLEAR);
> > > > +	nfsd_vers(nn, 3, NFSD_CLEAR);
> > > > +	for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++)
> > > > +		nfsd_minorversion(nn, i, NFSD_CLEAR);
> > > > +
> > > > +	nlmsg_for_each_attr(attr, info->nlhdr, GENL_HDRLEN, rem) {
> > > > +		struct nlattr *tb[NFSD_A_VERSION_MAX + 1];
> > > > +		u32 major, minor = 0;
> > > > +		bool enabled;
> > > > +
> > > > +		if (nla_type(attr) != NFSD_A_SERVER_PROTO_VERSION)
> > > > +			continue;
> > > > +
> > > > +		if (nla_parse_nested(tb, NFSD_A_VERSION_MAX, attr,
> > > > +				     nfsd_version_nl_policy, info->extack) < 0)
> > > > +			continue;
> > > > +
> > > > +		if (!tb[NFSD_A_VERSION_MAJOR])
> > > > +			continue;
> > > > +
> > > > +		major = nla_get_u32(tb[NFSD_A_VERSION_MAJOR]);
> > > > +		if (tb[NFSD_A_VERSION_MINOR])
> > > > +			minor = nla_get_u32(tb[NFSD_A_VERSION_MINOR]);
> > > > +
> > > > +		enabled = nla_get_flag(tb[NFSD_A_VERSION_ENABLED]);
> > > > +
> > > > +		switch (major) {
> > > > +		case 4:
> > > > +			nfsd_minorversion(nn, minor, enabled ? NFSD_SET : NFSD_CLEAR);
> > > > +			break;
> > > > +		case 3:
> > > > +		case 2:
> > > > +			if (!minor)
> > > > +				nfsd_vers(nn, major, enabled ? NFSD_SET : NFSD_CLEAR);
> > > > +			break;
> > > > +		default:
> > > > +			break;
> > > > +		}
> > > > +	}
> > > > +
> > > > +	mutex_unlock(&nfsd_mutex);
> > > > +
> > > > +	return 0;
> > > > +}
> > > > +
> > > > +/**
> > > > + * nfsd_nl_version_get_doit - get the nfs enabled versions
> > > > + * @skb: reply buffer
> > > > + * @info: netlink metadata and command arguments
> > > > + *
> > > > + * Return 0 on success or a negative errno.
> > > > + */
> > > > +int nfsd_nl_version_get_doit(struct sk_buff *skb, struct genl_info *info)
> > > > +{
> > > > +	struct nfsd_net *nn;
> > > > +	int i, err;
> > > > +	void *hdr;
> > > > +
> > > > +	skb = genlmsg_new(GENLMSG_DEFAULT_SIZE, GFP_KERNEL);
> > > > +	if (!skb)
> > > > +		return -ENOMEM;
> > > > +
> > > > +	hdr = genlmsg_iput(skb, info);
> > > > +	if (!hdr) {
> > > > +		err = -EMSGSIZE;
> > > > +		goto err_free_msg;
> > > > +	}
> > > > +
> > > > +	mutex_lock(&nfsd_mutex);
> > > > +	nn = net_generic(genl_info_net(info), nfsd_net_id);
> > > > +
> > > > +	for (i = 2; i <= 4; i++) {
> > > > +		int j;
> > > > +
> > > > +		for (j = 0; j <= NFSD_SUPPORTED_MINOR_VERSION; j++) {
> > > > +			struct nlattr *attr;
> > > > +
> > > > +			/* Don't record any versions the kernel doesn't have
> > > > +			 * compiled in
> > > > +			 */
> > > > +			if (!nfsd_support_version(i))
> > > > +				continue;
> > > > +
> > > > +			/* NFSv{2,3} does not support minor numbers */
> > > > +			if (i < 4 && j)
> > > > +				continue;
> > > > +
> > > > +			attr = nla_nest_start(skb,
> > > > +					      NFSD_A_SERVER_PROTO_VERSION);
> > > > +			if (!attr) {
> > > > +				err = -EINVAL;
> > > > +				goto err_nfsd_unlock;
> > > > +			}
> > > > +
> > > > +			if (nla_put_u32(skb, NFSD_A_VERSION_MAJOR, i) ||
> > > > +			    nla_put_u32(skb, NFSD_A_VERSION_MINOR, j)) {
> > > > +				err = -EINVAL;
> > > > +				goto err_nfsd_unlock;
> > > > +			}
> > > > +
> > > > +			/* Set the enabled flag if the version is enabled */
> > > > +			if (nfsd_vers(nn, i, NFSD_TEST) &&
> > > > +			    (i < 4 || nfsd_minorversion(nn, j, NFSD_TEST)) &&
> > > > +			    nla_put_flag(skb, NFSD_A_VERSION_ENABLED)) {
> > > > +				err = -EINVAL;
> > > > +				goto err_nfsd_unlock;
> > > > +			}
> > > > +
> > > > +			nla_nest_end(skb, attr);
> > > > +		}
> > > > +	}
> > > > +
> > > > +	mutex_unlock(&nfsd_mutex);
> > > > +	genlmsg_end(skb, hdr);
> > > > +
> > > > +	return genlmsg_reply(skb, info);
> > > > +
> > > > +err_nfsd_unlock:
> > > > +	mutex_unlock(&nfsd_mutex);
> > > > +err_free_msg:
> > > > +	nlmsg_free(skb);
> > > > +
> > > > +	return err;
> > > > +}
> > > > +
> > > >  /**
> > > >   * nfsd_net_init - Prepare the nfsd_net portion of a new net namespace
> > > >   * @net: a freshly-created network namespace
> > > > diff --git a/fs/nfsd/nfssvc.c b/fs/nfsd/nfssvc.c
> > > > index ca193f7ff0e1..4fc91f50138a 100644
> > > > --- a/fs/nfsd/nfssvc.c
> > > > +++ b/fs/nfsd/nfssvc.c
> > > > @@ -133,8 +133,7 @@ struct svc_program		nfsd_program = {
> > > >  	.pg_rpcbind_set		= nfsd_rpcbind_set,
> > > >  };
> > > >  
> > > > -static bool
> > > > -nfsd_support_version(int vers)
> > > > +bool nfsd_support_version(int vers)
> > > >  {
> > > >  	if (vers >= NFSD_MINVERS && vers < NFSD_NRVERS)
> > > >  		return nfsd_version[vers] != NULL;
> > > > diff --git a/include/uapi/linux/nfsd_netlink.h b/include/uapi/linux/nfsd_netlink.h
> > > > index ccc78a5ee650..8a0a2b344923 100644
> > > > --- a/include/uapi/linux/nfsd_netlink.h
> > > > +++ b/include/uapi/linux/nfsd_netlink.h
> > > > @@ -38,10 +38,28 @@ enum {
> > > >  	NFSD_A_SERVER_WORKER_MAX = (__NFSD_A_SERVER_WORKER_MAX - 1)
> > > >  };
> > > >  
> > > > +enum {
> > > > +	NFSD_A_VERSION_MAJOR = 1,
> > > > +	NFSD_A_VERSION_MINOR,
> > > > +	NFSD_A_VERSION_ENABLED,
> > > > +
> > > > +	__NFSD_A_VERSION_MAX,
> > > > +	NFSD_A_VERSION_MAX = (__NFSD_A_VERSION_MAX - 1)
> > > > +};
> > > > +
> > > > +enum {
> > > > +	NFSD_A_SERVER_PROTO_VERSION = 1,
> > > > +
> > > > +	__NFSD_A_SERVER_PROTO_MAX,
> > > > +	NFSD_A_SERVER_PROTO_MAX = (__NFSD_A_SERVER_PROTO_MAX - 1)
> > > > +};
> > > > +
> > > >  enum {
> > > >  	NFSD_CMD_RPC_STATUS_GET = 1,
> > > >  	NFSD_CMD_THREADS_SET,
> > > >  	NFSD_CMD_THREADS_GET,
> > > > +	NFSD_CMD_VERSION_SET,
> > > > +	NFSD_CMD_VERSION_GET,
> > > >  
> > > >  	__NFSD_CMD_MAX,
> > > >  	NFSD_CMD_MAX = (__NFSD_CMD_MAX - 1)
> > > > -- 
> > > > 2.44.0
> > > > 
> > > > 
> > > 
> > 
> > -- 
> > Jeff Layton <jlayton@kernel.org>
> > 
>
NeilBrown April 16, 2024, 11:05 p.m. UTC | #5
On Wed, 17 Apr 2024, Jeff Layton wrote:
> On Wed, 2024-04-17 at 07:48 +1000, NeilBrown wrote:
> > On Tue, 16 Apr 2024, Jeff Layton wrote:
> > > On Tue, 2024-04-16 at 13:16 +1000, NeilBrown wrote:
> > > > On Tue, 16 Apr 2024, Lorenzo Bianconi wrote:
> > > > > Introduce write_version netlink command through a "declarative" interface.
> > > > > This patch introduces a change in behavior since for version-set userspace
> > > > > is expected to provide a NFS major/minor version list it wants to enable
> > > > > while all the other ones will be disabled. (procfs write_version
> > > > > command implements imperative interface where the admin writes +3/-3 to
> > > > > enable/disable a single version.
> > > > 
> > > > It seems a little weird to me that the interface always disables all
> > > > version, but then also allows individual versions to be disabled.
> > > > 
> > > > Would it be reasonable to simply ignore the "enabled" flag when setting
> > > > version, and just enable all versions listed??
> > > > 
> > > > Or maybe only enable those with the flag, and don't disable those
> > > > without the flag?
> > > > 
> > > > Those don't necessarily seem much better - but the current behaviour
> > > > still seems odd.
> > > > 
> > > 
> > > I think it makes sense.
> > > 
> > > We disable all versions, and enable any that have the "enabled" flag set
> > > in the call from userland. Userland technically needn't send down the
> > > versions that are disabled in the call, but the current userland program
> > > does.
> > > 
> > > I worry about imperative interfaces that might only say -- "enable v4.1,
> > > but disable v3" and leave the others in their current state. That
> > > requires that both the kernel and userland keep state about what
> > > versions are currently enabled and disabled, and it's possible to get
> > > that wrong.
> > 
> > I understand and support your aversion for imperative interfaces.
> > But this interface, as currently implemented, looks somewhat imperative.
> > The message sent to the kernel could enable some interfaces and then
> > disable them.  I know that isn't the intent, but it is what the code
> > supports.  Hence "weird".
> > 
> > We could add code to make that sort of thing impossible, but there isn't
> > much point.  Better to make it syntactically impossible.
> > 
> > Realistically there will never be NFSv4.3 as there is no need - new
> > features can be added incrementally. 
> > 
> 
> There is no need _so_far_. Who knows what the future will bring? Maybe
> we'll need v4.3 in order to add some needed feature?
> 
> >  So we could just pass an array of
> > 5 active flags: 2,3,4.0,4.1,4.2.  I suspect you wouldn't like that and
> > I'm not sure that I do either.  A "read" would return the same array
> > with 3 possible states: unavailable, disabled, enabled.  (Maybe the
> > array could be variable length so 5.0 could be added one day...).
> > 
> 
> A set of flags is basically what this interface is. They just happen to
> also be labeled with the major and minorversion. I think that's a good
> thing.

Good for whom?  Labelling allows for labelling inconsistencies.

Maybe the kernel should be able to provide an ordered list of labels,
and separately an array of which labels are enabled/disabled.
Userspace could send down a new set of enabled/disabled flags based on
the agreed set of labels.

Here is a question that is a bit of a diversion, but might help us
understand the context more fully:

  Why would anyone disable v4.2 separately from v4.1 ??

I understand that v2, v3, v4.0, v4.1 are effectively different protocols
and you might want to ensure that only the approved/tested protocol is
used.  But v4.2 is just a few enhancements on v4.1.  Why would you want
to disable it?

The answer I can think of that there might be bugs (perish the
thought!!) in some of those features so you might want to avoid using
them.
But in that case, it is really the features that you want to suppress,
not the protocol version.

Maybe I might want to disable delegation - to just write delegation.
Can I do that?  What if I just want to disable server-side copy, but
keep fallocate and umask support?

i.e.  is a list of versions really the granularity that we want to use
for this interface?

NeilBrown

> 
> >
> > I haven't managed to come up with anything *better* than the current
> > proposal and I don't want to stand in its way, but I wanted to highlight
> > that it looks weird - as much imperative as declarative - in case
> > someone else might be able to come up with a better alternative.
> > 
> 
> The intention was to create a symmetrical interface. We have 2
> operations: a "get" that asks "what versions are currently supported and
> are they enabled?", and a "set" that says "here is the new state of
> version enablement".
> 
> The userland tool always sends down a complete set of versions. The
> kernel is (currently) more forgiving, and treats omitted versions as
> being disabled. The kernel could require that every supported version be
> represented in the "set" operation, if that's more desirable.
> 
> > > 
> > > My thinking was that by using a declarative interface like this, we
> > > eliminate ambiguity in how these interfaces are supposed to work. The
> > > client sends down an entire version map in one fell swoop, and we know
> > > exactly what the result should look like.
> > > 
> > > Note that you can enable or disable just a single version with the
> > > userland tool, but this way all of that complexity stays in userland.
> > > 
> > > > Also for getting the version, the doc says:
> > > > 
> > > >      doc: get nfs enabled versions
> > > > 
> > > > which I don't think it quite right.  The code reports all supported
> > > > versions, and indicates which of those are enabled.  So maybe:
> > > > 
> > > >      doc: get enabled status for all supported versions
> > > > 
> > > > 
> > > > I think that fact that it actually lists all supported versions is
> > > > useful and worth making explicit.
> > > > 
> > > > 
> > > 
> > > Agreed. We should make that change before merging anything.
> > > 
> > > > > 
> > > > > Reviewed-by: Jeff Layton <jlayton@kernel.org>
> > > > > Tested-by: Jeff Layton <jlayton@kernel.org>
> > > > > Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
> > > > > ---
> > > > >  Documentation/netlink/specs/nfsd.yaml |  37 +++++++
> > > > >  fs/nfsd/netlink.c                     |  24 +++++
> > > > >  fs/nfsd/netlink.h                     |   5 +
> > > > >  fs/nfsd/netns.h                       |   1 +
> > > > >  fs/nfsd/nfsctl.c                      | 150 ++++++++++++++++++++++++++
> > > > >  fs/nfsd/nfssvc.c                      |   3 +-
> > > > >  include/uapi/linux/nfsd_netlink.h     |  18 ++++
> > > > >  7 files changed, 236 insertions(+), 2 deletions(-)
> > > > > 
> > > > > diff --git a/Documentation/netlink/specs/nfsd.yaml b/Documentation/netlink/specs/nfsd.yaml
> > > > > index cbe6c5fd6c4d..0396e8b3ea1f 100644
> > > > > --- a/Documentation/netlink/specs/nfsd.yaml
> > > > > +++ b/Documentation/netlink/specs/nfsd.yaml
> > > > > @@ -74,6 +74,26 @@ attribute-sets:
> > > > >        -
> > > > >          name: leasetime
> > > > >          type: u32
> > > > > +  -
> > > > > +    name: version
> > > > > +    attributes:
> > > > > +      -
> > > > > +        name: major
> > > > > +        type: u32
> > > > > +      -
> > > > > +        name: minor
> > > > > +        type: u32
> > > > > +      -
> > > > > +        name: enabled
> > > > > +        type: flag
> > > > > +  -
> > > > > +    name: server-proto
> > > > > +    attributes:
> > > > > +      -
> > > > > +        name: version
> > > > > +        type: nest
> > > > > +        nested-attributes: version
> > > > > +        multi-attr: true
> > > > >  
> > > > >  operations:
> > > > >    list:
> > > > > @@ -120,3 +140,20 @@ operations:
> > > > >              - threads
> > > > >              - gracetime
> > > > >              - leasetime
> > > > > +    -
> > > > > +      name: version-set
> > > > > +      doc: set nfs enabled versions
> > > > > +      attribute-set: server-proto
> > > > > +      flags: [ admin-perm ]
> > > > > +      do:
> > > > > +        request:
> > > > > +          attributes:
> > > > > +            - version
> > > > > +    -
> > > > > +      name: version-get
> > > > > +      doc: get nfs enabled versions
> > > > > +      attribute-set: server-proto
> > > > > +      do:
> > > > > +        reply:
> > > > > +          attributes:
> > > > > +            - version
> > > > > diff --git a/fs/nfsd/netlink.c b/fs/nfsd/netlink.c
> > > > > index 20a646af0324..bf5df9597288 100644
> > > > > --- a/fs/nfsd/netlink.c
> > > > > +++ b/fs/nfsd/netlink.c
> > > > > @@ -10,6 +10,13 @@
> > > > >  
> > > > >  #include <uapi/linux/nfsd_netlink.h>
> > > > >  
> > > > > +/* Common nested types */
> > > > > +const struct nla_policy nfsd_version_nl_policy[NFSD_A_VERSION_ENABLED + 1] = {
> > > > > +	[NFSD_A_VERSION_MAJOR] = { .type = NLA_U32, },
> > > > > +	[NFSD_A_VERSION_MINOR] = { .type = NLA_U32, },
> > > > > +	[NFSD_A_VERSION_ENABLED] = { .type = NLA_FLAG, },
> > > > > +};
> > > > > +
> > > > >  /* NFSD_CMD_THREADS_SET - do */
> > > > >  static const struct nla_policy nfsd_threads_set_nl_policy[NFSD_A_SERVER_WORKER_LEASETIME + 1] = {
> > > > >  	[NFSD_A_SERVER_WORKER_THREADS] = { .type = NLA_U32, },
> > > > > @@ -17,6 +24,11 @@ static const struct nla_policy nfsd_threads_set_nl_policy[NFSD_A_SERVER_WORKER_L
> > > > >  	[NFSD_A_SERVER_WORKER_LEASETIME] = { .type = NLA_U32, },
> > > > >  };
> > > > >  
> > > > > +/* NFSD_CMD_VERSION_SET - do */
> > > > > +static const struct nla_policy nfsd_version_set_nl_policy[NFSD_A_SERVER_PROTO_VERSION + 1] = {
> > > > > +	[NFSD_A_SERVER_PROTO_VERSION] = NLA_POLICY_NESTED(nfsd_version_nl_policy),
> > > > > +};
> > > > > +
> > > > >  /* Ops table for nfsd */
> > > > >  static const struct genl_split_ops nfsd_nl_ops[] = {
> > > > >  	{
> > > > > @@ -38,6 +50,18 @@ static const struct genl_split_ops nfsd_nl_ops[] = {
> > > > >  		.doit	= nfsd_nl_threads_get_doit,
> > > > >  		.flags	= GENL_CMD_CAP_DO,
> > > > >  	},
> > > > > +	{
> > > > > +		.cmd		= NFSD_CMD_VERSION_SET,
> > > > > +		.doit		= nfsd_nl_version_set_doit,
> > > > > +		.policy		= nfsd_version_set_nl_policy,
> > > > > +		.maxattr	= NFSD_A_SERVER_PROTO_VERSION,
> > > > > +		.flags		= GENL_ADMIN_PERM | GENL_CMD_CAP_DO,
> > > > > +	},
> > > > > +	{
> > > > > +		.cmd	= NFSD_CMD_VERSION_GET,
> > > > > +		.doit	= nfsd_nl_version_get_doit,
> > > > > +		.flags	= GENL_CMD_CAP_DO,
> > > > > +	},
> > > > >  };
> > > > >  
> > > > >  struct genl_family nfsd_nl_family __ro_after_init = {
> > > > > diff --git a/fs/nfsd/netlink.h b/fs/nfsd/netlink.h
> > > > > index 4137fac477e4..c7c0da275481 100644
> > > > > --- a/fs/nfsd/netlink.h
> > > > > +++ b/fs/nfsd/netlink.h
> > > > > @@ -11,6 +11,9 @@
> > > > >  
> > > > >  #include <uapi/linux/nfsd_netlink.h>
> > > > >  
> > > > > +/* Common nested types */
> > > > > +extern const struct nla_policy nfsd_version_nl_policy[NFSD_A_VERSION_ENABLED + 1];
> > > > > +
> > > > >  int nfsd_nl_rpc_status_get_start(struct netlink_callback *cb);
> > > > >  int nfsd_nl_rpc_status_get_done(struct netlink_callback *cb);
> > > > >  
> > > > > @@ -18,6 +21,8 @@ int nfsd_nl_rpc_status_get_dumpit(struct sk_buff *skb,
> > > > >  				  struct netlink_callback *cb);
> > > > >  int nfsd_nl_threads_set_doit(struct sk_buff *skb, struct genl_info *info);
> > > > >  int nfsd_nl_threads_get_doit(struct sk_buff *skb, struct genl_info *info);
> > > > > +int nfsd_nl_version_set_doit(struct sk_buff *skb, struct genl_info *info);
> > > > > +int nfsd_nl_version_get_doit(struct sk_buff *skb, struct genl_info *info);
> > > > >  
> > > > >  extern struct genl_family nfsd_nl_family;
> > > > >  
> > > > > diff --git a/fs/nfsd/netns.h b/fs/nfsd/netns.h
> > > > > index d4be519b5734..14ec15656320 100644
> > > > > --- a/fs/nfsd/netns.h
> > > > > +++ b/fs/nfsd/netns.h
> > > > > @@ -218,6 +218,7 @@ struct nfsd_net {
> > > > >  /* Simple check to find out if a given net was properly initialized */
> > > > >  #define nfsd_netns_ready(nn) ((nn)->sessionid_hashtbl)
> > > > >  
> > > > > +extern bool nfsd_support_version(int vers);
> > > > >  extern void nfsd_netns_free_versions(struct nfsd_net *nn);
> > > > >  
> > > > >  extern unsigned int nfsd_net_id;
> > > > > diff --git a/fs/nfsd/nfsctl.c b/fs/nfsd/nfsctl.c
> > > > > index 38a5df03981b..2c8929ef79e9 100644
> > > > > --- a/fs/nfsd/nfsctl.c
> > > > > +++ b/fs/nfsd/nfsctl.c
> > > > > @@ -1757,6 +1757,156 @@ int nfsd_nl_threads_get_doit(struct sk_buff *skb, struct genl_info *info)
> > > > >  	return err;
> > > > >  }
> > > > >  
> > > > > +/**
> > > > > + * nfsd_nl_version_set_doit - set the nfs enabled versions
> > > > > + * @skb: reply buffer
> > > > > + * @info: netlink metadata and command arguments
> > > > > + *
> > > > > + * Return 0 on success or a negative errno.
> > > > > + */
> > > > > +int nfsd_nl_version_set_doit(struct sk_buff *skb, struct genl_info *info)
> > > > > +{
> > > > > +	const struct nlattr *attr;
> > > > > +	struct nfsd_net *nn;
> > > > > +	int i, rem;
> > > > > +
> > > > > +	if (GENL_REQ_ATTR_CHECK(info, NFSD_A_SERVER_PROTO_VERSION))
> > > > > +		return -EINVAL;
> > > > > +
> > > > > +	mutex_lock(&nfsd_mutex);
> > > > > +
> > > > > +	nn = net_generic(genl_info_net(info), nfsd_net_id);
> > > > > +	if (nn->nfsd_serv) {
> > > > > +		mutex_unlock(&nfsd_mutex);
> > > > > +		return -EBUSY;
> > > > > +	}
> > > > > +
> > > > > +	/* clear current supported versions. */
> > > > > +	nfsd_vers(nn, 2, NFSD_CLEAR);
> > > > > +	nfsd_vers(nn, 3, NFSD_CLEAR);
> > > > > +	for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++)
> > > > > +		nfsd_minorversion(nn, i, NFSD_CLEAR);
> > > > > +
> > > > > +	nlmsg_for_each_attr(attr, info->nlhdr, GENL_HDRLEN, rem) {
> > > > > +		struct nlattr *tb[NFSD_A_VERSION_MAX + 1];
> > > > > +		u32 major, minor = 0;
> > > > > +		bool enabled;
> > > > > +
> > > > > +		if (nla_type(attr) != NFSD_A_SERVER_PROTO_VERSION)
> > > > > +			continue;
> > > > > +
> > > > > +		if (nla_parse_nested(tb, NFSD_A_VERSION_MAX, attr,
> > > > > +				     nfsd_version_nl_policy, info->extack) < 0)
> > > > > +			continue;
> > > > > +
> > > > > +		if (!tb[NFSD_A_VERSION_MAJOR])
> > > > > +			continue;
> > > > > +
> > > > > +		major = nla_get_u32(tb[NFSD_A_VERSION_MAJOR]);
> > > > > +		if (tb[NFSD_A_VERSION_MINOR])
> > > > > +			minor = nla_get_u32(tb[NFSD_A_VERSION_MINOR]);
> > > > > +
> > > > > +		enabled = nla_get_flag(tb[NFSD_A_VERSION_ENABLED]);
> > > > > +
> > > > > +		switch (major) {
> > > > > +		case 4:
> > > > > +			nfsd_minorversion(nn, minor, enabled ? NFSD_SET : NFSD_CLEAR);
> > > > > +			break;
> > > > > +		case 3:
> > > > > +		case 2:
> > > > > +			if (!minor)
> > > > > +				nfsd_vers(nn, major, enabled ? NFSD_SET : NFSD_CLEAR);
> > > > > +			break;
> > > > > +		default:
> > > > > +			break;
> > > > > +		}
> > > > > +	}
> > > > > +
> > > > > +	mutex_unlock(&nfsd_mutex);
> > > > > +
> > > > > +	return 0;
> > > > > +}
> > > > > +
> > > > > +/**
> > > > > + * nfsd_nl_version_get_doit - get the nfs enabled versions
> > > > > + * @skb: reply buffer
> > > > > + * @info: netlink metadata and command arguments
> > > > > + *
> > > > > + * Return 0 on success or a negative errno.
> > > > > + */
> > > > > +int nfsd_nl_version_get_doit(struct sk_buff *skb, struct genl_info *info)
> > > > > +{
> > > > > +	struct nfsd_net *nn;
> > > > > +	int i, err;
> > > > > +	void *hdr;
> > > > > +
> > > > > +	skb = genlmsg_new(GENLMSG_DEFAULT_SIZE, GFP_KERNEL);
> > > > > +	if (!skb)
> > > > > +		return -ENOMEM;
> > > > > +
> > > > > +	hdr = genlmsg_iput(skb, info);
> > > > > +	if (!hdr) {
> > > > > +		err = -EMSGSIZE;
> > > > > +		goto err_free_msg;
> > > > > +	}
> > > > > +
> > > > > +	mutex_lock(&nfsd_mutex);
> > > > > +	nn = net_generic(genl_info_net(info), nfsd_net_id);
> > > > > +
> > > > > +	for (i = 2; i <= 4; i++) {
> > > > > +		int j;
> > > > > +
> > > > > +		for (j = 0; j <= NFSD_SUPPORTED_MINOR_VERSION; j++) {
> > > > > +			struct nlattr *attr;
> > > > > +
> > > > > +			/* Don't record any versions the kernel doesn't have
> > > > > +			 * compiled in
> > > > > +			 */
> > > > > +			if (!nfsd_support_version(i))
> > > > > +				continue;
> > > > > +
> > > > > +			/* NFSv{2,3} does not support minor numbers */
> > > > > +			if (i < 4 && j)
> > > > > +				continue;
> > > > > +
> > > > > +			attr = nla_nest_start(skb,
> > > > > +					      NFSD_A_SERVER_PROTO_VERSION);
> > > > > +			if (!attr) {
> > > > > +				err = -EINVAL;
> > > > > +				goto err_nfsd_unlock;
> > > > > +			}
> > > > > +
> > > > > +			if (nla_put_u32(skb, NFSD_A_VERSION_MAJOR, i) ||
> > > > > +			    nla_put_u32(skb, NFSD_A_VERSION_MINOR, j)) {
> > > > > +				err = -EINVAL;
> > > > > +				goto err_nfsd_unlock;
> > > > > +			}
> > > > > +
> > > > > +			/* Set the enabled flag if the version is enabled */
> > > > > +			if (nfsd_vers(nn, i, NFSD_TEST) &&
> > > > > +			    (i < 4 || nfsd_minorversion(nn, j, NFSD_TEST)) &&
> > > > > +			    nla_put_flag(skb, NFSD_A_VERSION_ENABLED)) {
> > > > > +				err = -EINVAL;
> > > > > +				goto err_nfsd_unlock;
> > > > > +			}
> > > > > +
> > > > > +			nla_nest_end(skb, attr);
> > > > > +		}
> > > > > +	}
> > > > > +
> > > > > +	mutex_unlock(&nfsd_mutex);
> > > > > +	genlmsg_end(skb, hdr);
> > > > > +
> > > > > +	return genlmsg_reply(skb, info);
> > > > > +
> > > > > +err_nfsd_unlock:
> > > > > +	mutex_unlock(&nfsd_mutex);
> > > > > +err_free_msg:
> > > > > +	nlmsg_free(skb);
> > > > > +
> > > > > +	return err;
> > > > > +}
> > > > > +
> > > > >  /**
> > > > >   * nfsd_net_init - Prepare the nfsd_net portion of a new net namespace
> > > > >   * @net: a freshly-created network namespace
> > > > > diff --git a/fs/nfsd/nfssvc.c b/fs/nfsd/nfssvc.c
> > > > > index ca193f7ff0e1..4fc91f50138a 100644
> > > > > --- a/fs/nfsd/nfssvc.c
> > > > > +++ b/fs/nfsd/nfssvc.c
> > > > > @@ -133,8 +133,7 @@ struct svc_program		nfsd_program = {
> > > > >  	.pg_rpcbind_set		= nfsd_rpcbind_set,
> > > > >  };
> > > > >  
> > > > > -static bool
> > > > > -nfsd_support_version(int vers)
> > > > > +bool nfsd_support_version(int vers)
> > > > >  {
> > > > >  	if (vers >= NFSD_MINVERS && vers < NFSD_NRVERS)
> > > > >  		return nfsd_version[vers] != NULL;
> > > > > diff --git a/include/uapi/linux/nfsd_netlink.h b/include/uapi/linux/nfsd_netlink.h
> > > > > index ccc78a5ee650..8a0a2b344923 100644
> > > > > --- a/include/uapi/linux/nfsd_netlink.h
> > > > > +++ b/include/uapi/linux/nfsd_netlink.h
> > > > > @@ -38,10 +38,28 @@ enum {
> > > > >  	NFSD_A_SERVER_WORKER_MAX = (__NFSD_A_SERVER_WORKER_MAX - 1)
> > > > >  };
> > > > >  
> > > > > +enum {
> > > > > +	NFSD_A_VERSION_MAJOR = 1,
> > > > > +	NFSD_A_VERSION_MINOR,
> > > > > +	NFSD_A_VERSION_ENABLED,
> > > > > +
> > > > > +	__NFSD_A_VERSION_MAX,
> > > > > +	NFSD_A_VERSION_MAX = (__NFSD_A_VERSION_MAX - 1)
> > > > > +};
> > > > > +
> > > > > +enum {
> > > > > +	NFSD_A_SERVER_PROTO_VERSION = 1,
> > > > > +
> > > > > +	__NFSD_A_SERVER_PROTO_MAX,
> > > > > +	NFSD_A_SERVER_PROTO_MAX = (__NFSD_A_SERVER_PROTO_MAX - 1)
> > > > > +};
> > > > > +
> > > > >  enum {
> > > > >  	NFSD_CMD_RPC_STATUS_GET = 1,
> > > > >  	NFSD_CMD_THREADS_SET,
> > > > >  	NFSD_CMD_THREADS_GET,
> > > > > +	NFSD_CMD_VERSION_SET,
> > > > > +	NFSD_CMD_VERSION_GET,
> > > > >  
> > > > >  	__NFSD_CMD_MAX,
> > > > >  	NFSD_CMD_MAX = (__NFSD_CMD_MAX - 1)
> > > > > -- 
> > > > > 2.44.0
> > > > > 
> > > > > 
> > > > 
> > > 
> > > -- 
> > > Jeff Layton <jlayton@kernel.org>
> > > 
> > 
> 
> -- 
> Jeff Layton <jlayton@kernel.org>
>
Jeff Layton April 17, 2024, 12:10 a.m. UTC | #6
On Wed, 2024-04-17 at 09:05 +1000, NeilBrown wrote:
> On Wed, 17 Apr 2024, Jeff Layton wrote:
> > On Wed, 2024-04-17 at 07:48 +1000, NeilBrown wrote:
> > > On Tue, 16 Apr 2024, Jeff Layton wrote:
> > > > On Tue, 2024-04-16 at 13:16 +1000, NeilBrown wrote:
> > > > > On Tue, 16 Apr 2024, Lorenzo Bianconi wrote:
> > > > > > Introduce write_version netlink command through a "declarative" interface.
> > > > > > This patch introduces a change in behavior since for version-set userspace
> > > > > > is expected to provide a NFS major/minor version list it wants to enable
> > > > > > while all the other ones will be disabled. (procfs write_version
> > > > > > command implements imperative interface where the admin writes +3/-3 to
> > > > > > enable/disable a single version.
> > > > > 
> > > > > It seems a little weird to me that the interface always disables all
> > > > > version, but then also allows individual versions to be disabled.
> > > > > 
> > > > > Would it be reasonable to simply ignore the "enabled" flag when setting
> > > > > version, and just enable all versions listed??
> > > > > 
> > > > > Or maybe only enable those with the flag, and don't disable those
> > > > > without the flag?
> > > > > 
> > > > > Those don't necessarily seem much better - but the current behaviour
> > > > > still seems odd.
> > > > > 
> > > > 
> > > > I think it makes sense.
> > > > 
> > > > We disable all versions, and enable any that have the "enabled" flag set
> > > > in the call from userland. Userland technically needn't send down the
> > > > versions that are disabled in the call, but the current userland program
> > > > does.
> > > > 
> > > > I worry about imperative interfaces that might only say -- "enable v4.1,
> > > > but disable v3" and leave the others in their current state. That
> > > > requires that both the kernel and userland keep state about what
> > > > versions are currently enabled and disabled, and it's possible to get
> > > > that wrong.
> > > 
> > > I understand and support your aversion for imperative interfaces.
> > > But this interface, as currently implemented, looks somewhat imperative.
> > > The message sent to the kernel could enable some interfaces and then
> > > disable them.  I know that isn't the intent, but it is what the code
> > > supports.  Hence "weird".
> > > 
> > > We could add code to make that sort of thing impossible, but there isn't
> > > much point.  Better to make it syntactically impossible.
> > > 
> > > Realistically there will never be NFSv4.3 as there is no need - new
> > > features can be added incrementally. 
> > > 
> > 
> > There is no need _so_far_. Who knows what the future will bring? Maybe
> > we'll need v4.3 in order to add some needed feature?
> > 
> > >  So we could just pass an array of
> > > 5 active flags: 2,3,4.0,4.1,4.2.  I suspect you wouldn't like that and
> > > I'm not sure that I do either.  A "read" would return the same array
> > > with 3 possible states: unavailable, disabled, enabled.  (Maybe the
> > > array could be variable length so 5.0 could be added one day...).
> > > 
> > 
> > A set of flags is basically what this interface is. They just happen to
> > also be labeled with the major and minorversion. I think that's a good
> > thing.
> 
> Good for whom?  Labelling allows for labelling inconsistencies.
> 

Now you're just being silly. You wanted a variable length array, but
what if the next slot is not v4.3 but 5.0? A positional interpretation
of a slot in an array is just as just as subject to problems.

> Maybe the kernel should be able to provide an ordered list of labels,
> and separately an array of which labels are enabled/disabled.
> Userspace could send down a new set of enabled/disabled flags based on
> the agreed set of labels.
> 

How is this better than what's been proposed? One strength of netlink is
that the data is structured. The already proposed interface takes
advantage of that.

> Here is a question that is a bit of a diversion, but might help us
> understand the context more fully:
> 
>   Why would anyone disable v4.2 separately from v4.1 ??
> 

Furthermore, what does it mean to disable v4.1 but leave v4.2 enabled?

> I understand that v2, v3, v4.0, v4.1 are effectively different protocols
> and you might want to ensure that only the approved/tested protocol is
> used.  But v4.2 is just a few enhancements on v4.1.  Why would you want
> to disable it?
> 
> The answer I can think of that there might be bugs (perish the
> thought!!) in some of those features so you might want to avoid using
> them.
> But in that case, it is really the features that you want to suppress,
> not the protocol version.
> 
> Maybe I might want to disable delegation - to just write delegation.
> Can I do that?  What if I just want to disable server-side copy, but
> keep fallocate and umask support?
> 
> i.e.  is a list of versions really the granularity that we want to use
> for this interface?
> 

Our current goal is to replace rpc.nfsd with a new program that works
via netlink. An important bit of what rpc.nfsd does is start the NFS
server with the settings in /etc/nfs.conf. Some of those settings are
vers3=, vers4.0=, etc. that govern how /proc/fs/nfsd/versions is set.
We have an immediate need to deal with those settings today, and
probably will for quite some time.

I'm not opposed to augmenting that with something more granular, but I
don't think we should block this interface and wait on that. We can
extend the interface at some point in the future to take a new feature
bitmask or something, and just declare that (e.g.) declaring vers4.2=n
disables some subset of those features.

> 
> > 
> > > 
> > > I haven't managed to come up with anything *better* than the current
> > > proposal and I don't want to stand in its way, but I wanted to highlight
> > > that it looks weird - as much imperative as declarative - in case
> > > someone else might be able to come up with a better alternative.
> > > 
> > 
> > The intention was to create a symmetrical interface. We have 2
> > operations: a "get" that asks "what versions are currently supported and
> > are they enabled?", and a "set" that says "here is the new state of
> > version enablement".
> > 
> > The userland tool always sends down a complete set of versions. The
> > kernel is (currently) more forgiving, and treats omitted versions as
> > being disabled. The kernel could require that every supported version be
> > represented in the "set" operation, if that's more desirable.
> > 
> > > > 
> > > > My thinking was that by using a declarative interface like this, we
> > > > eliminate ambiguity in how these interfaces are supposed to work. The
> > > > client sends down an entire version map in one fell swoop, and we know
> > > > exactly what the result should look like.
> > > > 
> > > > Note that you can enable or disable just a single version with the
> > > > userland tool, but this way all of that complexity stays in userland.
> > > > 
> > > > > Also for getting the version, the doc says:
> > > > > 
> > > > >      doc: get nfs enabled versions
> > > > > 
> > > > > which I don't think it quite right.  The code reports all supported
> > > > > versions, and indicates which of those are enabled.  So maybe:
> > > > > 
> > > > >      doc: get enabled status for all supported versions
> > > > > 
> > > > > 
> > > > > I think that fact that it actually lists all supported versions is
> > > > > useful and worth making explicit.
> > > > > 
> > > > > 
> > > > 
> > > > Agreed. We should make that change before merging anything.
> > > > 
> > > > > > 
> > > > > > Reviewed-by: Jeff Layton <jlayton@kernel.org>
> > > > > > Tested-by: Jeff Layton <jlayton@kernel.org>
> > > > > > Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
> > > > > > ---
> > > > > >  Documentation/netlink/specs/nfsd.yaml |  37 +++++++
> > > > > >  fs/nfsd/netlink.c                     |  24 +++++
> > > > > >  fs/nfsd/netlink.h                     |   5 +
> > > > > >  fs/nfsd/netns.h                       |   1 +
> > > > > >  fs/nfsd/nfsctl.c                      | 150 ++++++++++++++++++++++++++
> > > > > >  fs/nfsd/nfssvc.c                      |   3 +-
> > > > > >  include/uapi/linux/nfsd_netlink.h     |  18 ++++
> > > > > >  7 files changed, 236 insertions(+), 2 deletions(-)
> > > > > > 
> > > > > > diff --git a/Documentation/netlink/specs/nfsd.yaml b/Documentation/netlink/specs/nfsd.yaml
> > > > > > index cbe6c5fd6c4d..0396e8b3ea1f 100644
> > > > > > --- a/Documentation/netlink/specs/nfsd.yaml
> > > > > > +++ b/Documentation/netlink/specs/nfsd.yaml
> > > > > > @@ -74,6 +74,26 @@ attribute-sets:
> > > > > >        -
> > > > > >          name: leasetime
> > > > > >          type: u32
> > > > > > +  -
> > > > > > +    name: version
> > > > > > +    attributes:
> > > > > > +      -
> > > > > > +        name: major
> > > > > > +        type: u32
> > > > > > +      -
> > > > > > +        name: minor
> > > > > > +        type: u32
> > > > > > +      -
> > > > > > +        name: enabled
> > > > > > +        type: flag
> > > > > > +  -
> > > > > > +    name: server-proto
> > > > > > +    attributes:
> > > > > > +      -
> > > > > > +        name: version
> > > > > > +        type: nest
> > > > > > +        nested-attributes: version
> > > > > > +        multi-attr: true
> > > > > >  
> > > > > >  operations:
> > > > > >    list:
> > > > > > @@ -120,3 +140,20 @@ operations:
> > > > > >              - threads
> > > > > >              - gracetime
> > > > > >              - leasetime
> > > > > > +    -
> > > > > > +      name: version-set
> > > > > > +      doc: set nfs enabled versions
> > > > > > +      attribute-set: server-proto
> > > > > > +      flags: [ admin-perm ]
> > > > > > +      do:
> > > > > > +        request:
> > > > > > +          attributes:
> > > > > > +            - version
> > > > > > +    -
> > > > > > +      name: version-get
> > > > > > +      doc: get nfs enabled versions
> > > > > > +      attribute-set: server-proto
> > > > > > +      do:
> > > > > > +        reply:
> > > > > > +          attributes:
> > > > > > +            - version
> > > > > > diff --git a/fs/nfsd/netlink.c b/fs/nfsd/netlink.c
> > > > > > index 20a646af0324..bf5df9597288 100644
> > > > > > --- a/fs/nfsd/netlink.c
> > > > > > +++ b/fs/nfsd/netlink.c
> > > > > > @@ -10,6 +10,13 @@
> > > > > >  
> > > > > >  #include <uapi/linux/nfsd_netlink.h>
> > > > > >  
> > > > > > +/* Common nested types */
> > > > > > +const struct nla_policy nfsd_version_nl_policy[NFSD_A_VERSION_ENABLED + 1] = {
> > > > > > +	[NFSD_A_VERSION_MAJOR] = { .type = NLA_U32, },
> > > > > > +	[NFSD_A_VERSION_MINOR] = { .type = NLA_U32, },
> > > > > > +	[NFSD_A_VERSION_ENABLED] = { .type = NLA_FLAG, },
> > > > > > +};
> > > > > > +
> > > > > >  /* NFSD_CMD_THREADS_SET - do */
> > > > > >  static const struct nla_policy nfsd_threads_set_nl_policy[NFSD_A_SERVER_WORKER_LEASETIME + 1] = {
> > > > > >  	[NFSD_A_SERVER_WORKER_THREADS] = { .type = NLA_U32, },
> > > > > > @@ -17,6 +24,11 @@ static const struct nla_policy nfsd_threads_set_nl_policy[NFSD_A_SERVER_WORKER_L
> > > > > >  	[NFSD_A_SERVER_WORKER_LEASETIME] = { .type = NLA_U32, },
> > > > > >  };
> > > > > >  
> > > > > > +/* NFSD_CMD_VERSION_SET - do */
> > > > > > +static const struct nla_policy nfsd_version_set_nl_policy[NFSD_A_SERVER_PROTO_VERSION + 1] = {
> > > > > > +	[NFSD_A_SERVER_PROTO_VERSION] = NLA_POLICY_NESTED(nfsd_version_nl_policy),
> > > > > > +};
> > > > > > +
> > > > > >  /* Ops table for nfsd */
> > > > > >  static const struct genl_split_ops nfsd_nl_ops[] = {
> > > > > >  	{
> > > > > > @@ -38,6 +50,18 @@ static const struct genl_split_ops nfsd_nl_ops[] = {
> > > > > >  		.doit	= nfsd_nl_threads_get_doit,
> > > > > >  		.flags	= GENL_CMD_CAP_DO,
> > > > > >  	},
> > > > > > +	{
> > > > > > +		.cmd		= NFSD_CMD_VERSION_SET,
> > > > > > +		.doit		= nfsd_nl_version_set_doit,
> > > > > > +		.policy		= nfsd_version_set_nl_policy,
> > > > > > +		.maxattr	= NFSD_A_SERVER_PROTO_VERSION,
> > > > > > +		.flags		= GENL_ADMIN_PERM | GENL_CMD_CAP_DO,
> > > > > > +	},
> > > > > > +	{
> > > > > > +		.cmd	= NFSD_CMD_VERSION_GET,
> > > > > > +		.doit	= nfsd_nl_version_get_doit,
> > > > > > +		.flags	= GENL_CMD_CAP_DO,
> > > > > > +	},
> > > > > >  };
> > > > > >  
> > > > > >  struct genl_family nfsd_nl_family __ro_after_init = {
> > > > > > diff --git a/fs/nfsd/netlink.h b/fs/nfsd/netlink.h
> > > > > > index 4137fac477e4..c7c0da275481 100644
> > > > > > --- a/fs/nfsd/netlink.h
> > > > > > +++ b/fs/nfsd/netlink.h
> > > > > > @@ -11,6 +11,9 @@
> > > > > >  
> > > > > >  #include <uapi/linux/nfsd_netlink.h>
> > > > > >  
> > > > > > +/* Common nested types */
> > > > > > +extern const struct nla_policy nfsd_version_nl_policy[NFSD_A_VERSION_ENABLED + 1];
> > > > > > +
> > > > > >  int nfsd_nl_rpc_status_get_start(struct netlink_callback *cb);
> > > > > >  int nfsd_nl_rpc_status_get_done(struct netlink_callback *cb);
> > > > > >  
> > > > > > @@ -18,6 +21,8 @@ int nfsd_nl_rpc_status_get_dumpit(struct sk_buff *skb,
> > > > > >  				  struct netlink_callback *cb);
> > > > > >  int nfsd_nl_threads_set_doit(struct sk_buff *skb, struct genl_info *info);
> > > > > >  int nfsd_nl_threads_get_doit(struct sk_buff *skb, struct genl_info *info);
> > > > > > +int nfsd_nl_version_set_doit(struct sk_buff *skb, struct genl_info *info);
> > > > > > +int nfsd_nl_version_get_doit(struct sk_buff *skb, struct genl_info *info);
> > > > > >  
> > > > > >  extern struct genl_family nfsd_nl_family;
> > > > > >  
> > > > > > diff --git a/fs/nfsd/netns.h b/fs/nfsd/netns.h
> > > > > > index d4be519b5734..14ec15656320 100644
> > > > > > --- a/fs/nfsd/netns.h
> > > > > > +++ b/fs/nfsd/netns.h
> > > > > > @@ -218,6 +218,7 @@ struct nfsd_net {
> > > > > >  /* Simple check to find out if a given net was properly initialized */
> > > > > >  #define nfsd_netns_ready(nn) ((nn)->sessionid_hashtbl)
> > > > > >  
> > > > > > +extern bool nfsd_support_version(int vers);
> > > > > >  extern void nfsd_netns_free_versions(struct nfsd_net *nn);
> > > > > >  
> > > > > >  extern unsigned int nfsd_net_id;
> > > > > > diff --git a/fs/nfsd/nfsctl.c b/fs/nfsd/nfsctl.c
> > > > > > index 38a5df03981b..2c8929ef79e9 100644
> > > > > > --- a/fs/nfsd/nfsctl.c
> > > > > > +++ b/fs/nfsd/nfsctl.c
> > > > > > @@ -1757,6 +1757,156 @@ int nfsd_nl_threads_get_doit(struct sk_buff *skb, struct genl_info *info)
> > > > > >  	return err;
> > > > > >  }
> > > > > >  
> > > > > > +/**
> > > > > > + * nfsd_nl_version_set_doit - set the nfs enabled versions
> > > > > > + * @skb: reply buffer
> > > > > > + * @info: netlink metadata and command arguments
> > > > > > + *
> > > > > > + * Return 0 on success or a negative errno.
> > > > > > + */
> > > > > > +int nfsd_nl_version_set_doit(struct sk_buff *skb, struct genl_info *info)
> > > > > > +{
> > > > > > +	const struct nlattr *attr;
> > > > > > +	struct nfsd_net *nn;
> > > > > > +	int i, rem;
> > > > > > +
> > > > > > +	if (GENL_REQ_ATTR_CHECK(info, NFSD_A_SERVER_PROTO_VERSION))
> > > > > > +		return -EINVAL;
> > > > > > +
> > > > > > +	mutex_lock(&nfsd_mutex);
> > > > > > +
> > > > > > +	nn = net_generic(genl_info_net(info), nfsd_net_id);
> > > > > > +	if (nn->nfsd_serv) {
> > > > > > +		mutex_unlock(&nfsd_mutex);
> > > > > > +		return -EBUSY;
> > > > > > +	}
> > > > > > +
> > > > > > +	/* clear current supported versions. */
> > > > > > +	nfsd_vers(nn, 2, NFSD_CLEAR);
> > > > > > +	nfsd_vers(nn, 3, NFSD_CLEAR);
> > > > > > +	for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++)
> > > > > > +		nfsd_minorversion(nn, i, NFSD_CLEAR);
> > > > > > +
> > > > > > +	nlmsg_for_each_attr(attr, info->nlhdr, GENL_HDRLEN, rem) {
> > > > > > +		struct nlattr *tb[NFSD_A_VERSION_MAX + 1];
> > > > > > +		u32 major, minor = 0;
> > > > > > +		bool enabled;
> > > > > > +
> > > > > > +		if (nla_type(attr) != NFSD_A_SERVER_PROTO_VERSION)
> > > > > > +			continue;
> > > > > > +
> > > > > > +		if (nla_parse_nested(tb, NFSD_A_VERSION_MAX, attr,
> > > > > > +				     nfsd_version_nl_policy, info->extack) < 0)
> > > > > > +			continue;
> > > > > > +
> > > > > > +		if (!tb[NFSD_A_VERSION_MAJOR])
> > > > > > +			continue;
> > > > > > +
> > > > > > +		major = nla_get_u32(tb[NFSD_A_VERSION_MAJOR]);
> > > > > > +		if (tb[NFSD_A_VERSION_MINOR])
> > > > > > +			minor = nla_get_u32(tb[NFSD_A_VERSION_MINOR]);
> > > > > > +
> > > > > > +		enabled = nla_get_flag(tb[NFSD_A_VERSION_ENABLED]);
> > > > > > +
> > > > > > +		switch (major) {
> > > > > > +		case 4:
> > > > > > +			nfsd_minorversion(nn, minor, enabled ? NFSD_SET : NFSD_CLEAR);
> > > > > > +			break;
> > > > > > +		case 3:
> > > > > > +		case 2:
> > > > > > +			if (!minor)
> > > > > > +				nfsd_vers(nn, major, enabled ? NFSD_SET : NFSD_CLEAR);
> > > > > > +			break;
> > > > > > +		default:
> > > > > > +			break;
> > > > > > +		}
> > > > > > +	}
> > > > > > +
> > > > > > +	mutex_unlock(&nfsd_mutex);
> > > > > > +
> > > > > > +	return 0;
> > > > > > +}
> > > > > > +
> > > > > > +/**
> > > > > > + * nfsd_nl_version_get_doit - get the nfs enabled versions
> > > > > > + * @skb: reply buffer
> > > > > > + * @info: netlink metadata and command arguments
> > > > > > + *
> > > > > > + * Return 0 on success or a negative errno.
> > > > > > + */
> > > > > > +int nfsd_nl_version_get_doit(struct sk_buff *skb, struct genl_info *info)
> > > > > > +{
> > > > > > +	struct nfsd_net *nn;
> > > > > > +	int i, err;
> > > > > > +	void *hdr;
> > > > > > +
> > > > > > +	skb = genlmsg_new(GENLMSG_DEFAULT_SIZE, GFP_KERNEL);
> > > > > > +	if (!skb)
> > > > > > +		return -ENOMEM;
> > > > > > +
> > > > > > +	hdr = genlmsg_iput(skb, info);
> > > > > > +	if (!hdr) {
> > > > > > +		err = -EMSGSIZE;
> > > > > > +		goto err_free_msg;
> > > > > > +	}
> > > > > > +
> > > > > > +	mutex_lock(&nfsd_mutex);
> > > > > > +	nn = net_generic(genl_info_net(info), nfsd_net_id);
> > > > > > +
> > > > > > +	for (i = 2; i <= 4; i++) {
> > > > > > +		int j;
> > > > > > +
> > > > > > +		for (j = 0; j <= NFSD_SUPPORTED_MINOR_VERSION; j++) {
> > > > > > +			struct nlattr *attr;
> > > > > > +
> > > > > > +			/* Don't record any versions the kernel doesn't have
> > > > > > +			 * compiled in
> > > > > > +			 */
> > > > > > +			if (!nfsd_support_version(i))
> > > > > > +				continue;
> > > > > > +
> > > > > > +			/* NFSv{2,3} does not support minor numbers */
> > > > > > +			if (i < 4 && j)
> > > > > > +				continue;
> > > > > > +
> > > > > > +			attr = nla_nest_start(skb,
> > > > > > +					      NFSD_A_SERVER_PROTO_VERSION);
> > > > > > +			if (!attr) {
> > > > > > +				err = -EINVAL;
> > > > > > +				goto err_nfsd_unlock;
> > > > > > +			}
> > > > > > +
> > > > > > +			if (nla_put_u32(skb, NFSD_A_VERSION_MAJOR, i) ||
> > > > > > +			    nla_put_u32(skb, NFSD_A_VERSION_MINOR, j)) {
> > > > > > +				err = -EINVAL;
> > > > > > +				goto err_nfsd_unlock;
> > > > > > +			}
> > > > > > +
> > > > > > +			/* Set the enabled flag if the version is enabled */
> > > > > > +			if (nfsd_vers(nn, i, NFSD_TEST) &&
> > > > > > +			    (i < 4 || nfsd_minorversion(nn, j, NFSD_TEST)) &&
> > > > > > +			    nla_put_flag(skb, NFSD_A_VERSION_ENABLED)) {
> > > > > > +				err = -EINVAL;
> > > > > > +				goto err_nfsd_unlock;
> > > > > > +			}
> > > > > > +
> > > > > > +			nla_nest_end(skb, attr);
> > > > > > +		}
> > > > > > +	}
> > > > > > +
> > > > > > +	mutex_unlock(&nfsd_mutex);
> > > > > > +	genlmsg_end(skb, hdr);
> > > > > > +
> > > > > > +	return genlmsg_reply(skb, info);
> > > > > > +
> > > > > > +err_nfsd_unlock:
> > > > > > +	mutex_unlock(&nfsd_mutex);
> > > > > > +err_free_msg:
> > > > > > +	nlmsg_free(skb);
> > > > > > +
> > > > > > +	return err;
> > > > > > +}
> > > > > > +
> > > > > >  /**
> > > > > >   * nfsd_net_init - Prepare the nfsd_net portion of a new net namespace
> > > > > >   * @net: a freshly-created network namespace
> > > > > > diff --git a/fs/nfsd/nfssvc.c b/fs/nfsd/nfssvc.c
> > > > > > index ca193f7ff0e1..4fc91f50138a 100644
> > > > > > --- a/fs/nfsd/nfssvc.c
> > > > > > +++ b/fs/nfsd/nfssvc.c
> > > > > > @@ -133,8 +133,7 @@ struct svc_program		nfsd_program = {
> > > > > >  	.pg_rpcbind_set		= nfsd_rpcbind_set,
> > > > > >  };
> > > > > >  
> > > > > > -static bool
> > > > > > -nfsd_support_version(int vers)
> > > > > > +bool nfsd_support_version(int vers)
> > > > > >  {
> > > > > >  	if (vers >= NFSD_MINVERS && vers < NFSD_NRVERS)
> > > > > >  		return nfsd_version[vers] != NULL;
> > > > > > diff --git a/include/uapi/linux/nfsd_netlink.h b/include/uapi/linux/nfsd_netlink.h
> > > > > > index ccc78a5ee650..8a0a2b344923 100644
> > > > > > --- a/include/uapi/linux/nfsd_netlink.h
> > > > > > +++ b/include/uapi/linux/nfsd_netlink.h
> > > > > > @@ -38,10 +38,28 @@ enum {
> > > > > >  	NFSD_A_SERVER_WORKER_MAX = (__NFSD_A_SERVER_WORKER_MAX - 1)
> > > > > >  };
> > > > > >  
> > > > > > +enum {
> > > > > > +	NFSD_A_VERSION_MAJOR = 1,
> > > > > > +	NFSD_A_VERSION_MINOR,
> > > > > > +	NFSD_A_VERSION_ENABLED,
> > > > > > +
> > > > > > +	__NFSD_A_VERSION_MAX,
> > > > > > +	NFSD_A_VERSION_MAX = (__NFSD_A_VERSION_MAX - 1)
> > > > > > +};
> > > > > > +
> > > > > > +enum {
> > > > > > +	NFSD_A_SERVER_PROTO_VERSION = 1,
> > > > > > +
> > > > > > +	__NFSD_A_SERVER_PROTO_MAX,
> > > > > > +	NFSD_A_SERVER_PROTO_MAX = (__NFSD_A_SERVER_PROTO_MAX - 1)
> > > > > > +};
> > > > > > +
> > > > > >  enum {
> > > > > >  	NFSD_CMD_RPC_STATUS_GET = 1,
> > > > > >  	NFSD_CMD_THREADS_SET,
> > > > > >  	NFSD_CMD_THREADS_GET,
> > > > > > +	NFSD_CMD_VERSION_SET,
> > > > > > +	NFSD_CMD_VERSION_GET,
> > > > > >  
> > > > > >  	__NFSD_CMD_MAX,
> > > > > >  	NFSD_CMD_MAX = (__NFSD_CMD_MAX - 1)
> > > > > > -- 
> > > > > > 2.44.0
> > > > > > 
> > > > > > 
> > > > > 
> > > > 
> > > > -- 
> > > > Jeff Layton <jlayton@kernel.org>
> > > > 
> > > 
> > 
> > -- 
> > Jeff Layton <jlayton@kernel.org>
> > 
>
NeilBrown April 17, 2024, 12:43 a.m. UTC | #7
On Wed, 17 Apr 2024, Jeff Layton wrote:
> On Wed, 2024-04-17 at 09:05 +1000, NeilBrown wrote:
> > On Wed, 17 Apr 2024, Jeff Layton wrote:
> > > On Wed, 2024-04-17 at 07:48 +1000, NeilBrown wrote:
> > > > On Tue, 16 Apr 2024, Jeff Layton wrote:
> > > > > On Tue, 2024-04-16 at 13:16 +1000, NeilBrown wrote:
> > > > > > On Tue, 16 Apr 2024, Lorenzo Bianconi wrote:
> > > > > > > Introduce write_version netlink command through a "declarative" interface.
> > > > > > > This patch introduces a change in behavior since for version-set userspace
> > > > > > > is expected to provide a NFS major/minor version list it wants to enable
> > > > > > > while all the other ones will be disabled. (procfs write_version
> > > > > > > command implements imperative interface where the admin writes +3/-3 to
> > > > > > > enable/disable a single version.
> > > > > > 
> > > > > > It seems a little weird to me that the interface always disables all
> > > > > > version, but then also allows individual versions to be disabled.
> > > > > > 
> > > > > > Would it be reasonable to simply ignore the "enabled" flag when setting
> > > > > > version, and just enable all versions listed??
> > > > > > 
> > > > > > Or maybe only enable those with the flag, and don't disable those
> > > > > > without the flag?
> > > > > > 
> > > > > > Those don't necessarily seem much better - but the current behaviour
> > > > > > still seems odd.
> > > > > > 
> > > > > 
> > > > > I think it makes sense.
> > > > > 
> > > > > We disable all versions, and enable any that have the "enabled" flag set
> > > > > in the call from userland. Userland technically needn't send down the
> > > > > versions that are disabled in the call, but the current userland program
> > > > > does.
> > > > > 
> > > > > I worry about imperative interfaces that might only say -- "enable v4.1,
> > > > > but disable v3" and leave the others in their current state. That
> > > > > requires that both the kernel and userland keep state about what
> > > > > versions are currently enabled and disabled, and it's possible to get
> > > > > that wrong.
> > > > 
> > > > I understand and support your aversion for imperative interfaces.
> > > > But this interface, as currently implemented, looks somewhat imperative.
> > > > The message sent to the kernel could enable some interfaces and then
> > > > disable them.  I know that isn't the intent, but it is what the code
> > > > supports.  Hence "weird".
> > > > 
> > > > We could add code to make that sort of thing impossible, but there isn't
> > > > much point.  Better to make it syntactically impossible.
> > > > 
> > > > Realistically there will never be NFSv4.3 as there is no need - new
> > > > features can be added incrementally. 
> > > > 
> > > 
> > > There is no need _so_far_. Who knows what the future will bring? Maybe
> > > we'll need v4.3 in order to add some needed feature?
> > > 
> > > >  So we could just pass an array of
> > > > 5 active flags: 2,3,4.0,4.1,4.2.  I suspect you wouldn't like that and
> > > > I'm not sure that I do either.  A "read" would return the same array
> > > > with 3 possible states: unavailable, disabled, enabled.  (Maybe the
> > > > array could be variable length so 5.0 could be added one day...).
> > > > 
> > > 
> > > A set of flags is basically what this interface is. They just happen to
> > > also be labeled with the major and minorversion. I think that's a good
> > > thing.
> > 
> > Good for whom?  Labelling allows for labelling inconsistencies.
> > 
> 
> Now you're just being silly. You wanted a variable length array, but
> what if the next slot is not v4.3 but 5.0? A positional interpretation
> of a slot in an array is just as just as subject to problems.

I don't think it could look like a imperative interface, which the
current one does a bit.

> 
> > Maybe the kernel should be able to provide an ordered list of labels,
> > and separately an array of which labels are enabled/disabled.
> > Userspace could send down a new set of enabled/disabled flags based on
> > the agreed set of labels.
> > 
> 
> How is this better than what's been proposed? One strength of netlink is
> that the data is structured. The already proposed interface takes
> advantage of that.
> 
> > Here is a question that is a bit of a diversion, but might help us
> > understand the context more fully:
> > 
> >   Why would anyone disable v4.2 separately from v4.1 ??
> > 
> 
> Furthermore, what does it mean to disable v4.1 but leave v4.2 enabled?

Indeed!

> 
> > I understand that v2, v3, v4.0, v4.1 are effectively different protocols
> > and you might want to ensure that only the approved/tested protocol is
> > used.  But v4.2 is just a few enhancements on v4.1.  Why would you want
> > to disable it?
> > 
> > The answer I can think of that there might be bugs (perish the
> > thought!!) in some of those features so you might want to avoid using
> > them.
> > But in that case, it is really the features that you want to suppress,
> > not the protocol version.
> > 
> > Maybe I might want to disable delegation - to just write delegation.
> > Can I do that?  What if I just want to disable server-side copy, but
> > keep fallocate and umask support?
> > 
> > i.e.  is a list of versions really the granularity that we want to use
> > for this interface?
> > 
> 
> Our current goal is to replace rpc.nfsd with a new program that works
> via netlink. An important bit of what rpc.nfsd does is start the NFS
> server with the settings in /etc/nfs.conf. Some of those settings are
> vers3=, vers4.0=, etc. that govern how /proc/fs/nfsd/versions is set.
> We have an immediate need to deal with those settings today, and
> probably will for quite some time.
> 
> I'm not opposed to augmenting that with something more granular, but I
> don't think we should block this interface and wait on that. We can
> extend the interface at some point in the future to take a new feature
> bitmask or something, and just declare that (e.g.) declaring vers4.2=n
> disables some subset of those features.

I agree that we don't want to block "good" while we wait for "perfect".  I
just want to be sure that what we have is "good".

The current /proc interface uses strings like "v3" and "v4.1".
The proposed netlink interface uses pairs on u32s - "major" and "minor".
So we lose some easy paths for extensibility.  Are we comfortable with
that?

This isn't a big deal - certainly not a blocked.  I just don't feel
entirely comfortable about the current interface and I'm exploring to
see if there might be something better.

Thanks,
NeilBrown


> 
> > 
> > > 
> > > > 
> > > > I haven't managed to come up with anything *better* than the current
> > > > proposal and I don't want to stand in its way, but I wanted to highlight
> > > > that it looks weird - as much imperative as declarative - in case
> > > > someone else might be able to come up with a better alternative.
> > > > 
> > > 
> > > The intention was to create a symmetrical interface. We have 2
> > > operations: a "get" that asks "what versions are currently supported and
> > > are they enabled?", and a "set" that says "here is the new state of
> > > version enablement".
> > > 
> > > The userland tool always sends down a complete set of versions. The
> > > kernel is (currently) more forgiving, and treats omitted versions as
> > > being disabled. The kernel could require that every supported version be
> > > represented in the "set" operation, if that's more desirable.
> > > 
> > > > > 
> > > > > My thinking was that by using a declarative interface like this, we
> > > > > eliminate ambiguity in how these interfaces are supposed to work. The
> > > > > client sends down an entire version map in one fell swoop, and we know
> > > > > exactly what the result should look like.
> > > > > 
> > > > > Note that you can enable or disable just a single version with the
> > > > > userland tool, but this way all of that complexity stays in userland.
> > > > > 
> > > > > > Also for getting the version, the doc says:
> > > > > > 
> > > > > >      doc: get nfs enabled versions
> > > > > > 
> > > > > > which I don't think it quite right.  The code reports all supported
> > > > > > versions, and indicates which of those are enabled.  So maybe:
> > > > > > 
> > > > > >      doc: get enabled status for all supported versions
> > > > > > 
> > > > > > 
> > > > > > I think that fact that it actually lists all supported versions is
> > > > > > useful and worth making explicit.
> > > > > > 
> > > > > > 
> > > > > 
> > > > > Agreed. We should make that change before merging anything.
> > > > > 
> > > > > > > 
> > > > > > > Reviewed-by: Jeff Layton <jlayton@kernel.org>
> > > > > > > Tested-by: Jeff Layton <jlayton@kernel.org>
> > > > > > > Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
> > > > > > > ---
> > > > > > >  Documentation/netlink/specs/nfsd.yaml |  37 +++++++
> > > > > > >  fs/nfsd/netlink.c                     |  24 +++++
> > > > > > >  fs/nfsd/netlink.h                     |   5 +
> > > > > > >  fs/nfsd/netns.h                       |   1 +
> > > > > > >  fs/nfsd/nfsctl.c                      | 150 ++++++++++++++++++++++++++
> > > > > > >  fs/nfsd/nfssvc.c                      |   3 +-
> > > > > > >  include/uapi/linux/nfsd_netlink.h     |  18 ++++
> > > > > > >  7 files changed, 236 insertions(+), 2 deletions(-)
> > > > > > > 
> > > > > > > diff --git a/Documentation/netlink/specs/nfsd.yaml b/Documentation/netlink/specs/nfsd.yaml
> > > > > > > index cbe6c5fd6c4d..0396e8b3ea1f 100644
> > > > > > > --- a/Documentation/netlink/specs/nfsd.yaml
> > > > > > > +++ b/Documentation/netlink/specs/nfsd.yaml
> > > > > > > @@ -74,6 +74,26 @@ attribute-sets:
> > > > > > >        -
> > > > > > >          name: leasetime
> > > > > > >          type: u32
> > > > > > > +  -
> > > > > > > +    name: version
> > > > > > > +    attributes:
> > > > > > > +      -
> > > > > > > +        name: major
> > > > > > > +        type: u32
> > > > > > > +      -
> > > > > > > +        name: minor
> > > > > > > +        type: u32
> > > > > > > +      -
> > > > > > > +        name: enabled
> > > > > > > +        type: flag
> > > > > > > +  -
> > > > > > > +    name: server-proto
> > > > > > > +    attributes:
> > > > > > > +      -
> > > > > > > +        name: version
> > > > > > > +        type: nest
> > > > > > > +        nested-attributes: version
> > > > > > > +        multi-attr: true
> > > > > > >  
> > > > > > >  operations:
> > > > > > >    list:
> > > > > > > @@ -120,3 +140,20 @@ operations:
> > > > > > >              - threads
> > > > > > >              - gracetime
> > > > > > >              - leasetime
> > > > > > > +    -
> > > > > > > +      name: version-set
> > > > > > > +      doc: set nfs enabled versions
> > > > > > > +      attribute-set: server-proto
> > > > > > > +      flags: [ admin-perm ]
> > > > > > > +      do:
> > > > > > > +        request:
> > > > > > > +          attributes:
> > > > > > > +            - version
> > > > > > > +    -
> > > > > > > +      name: version-get
> > > > > > > +      doc: get nfs enabled versions
> > > > > > > +      attribute-set: server-proto
> > > > > > > +      do:
> > > > > > > +        reply:
> > > > > > > +          attributes:
> > > > > > > +            - version
> > > > > > > diff --git a/fs/nfsd/netlink.c b/fs/nfsd/netlink.c
> > > > > > > index 20a646af0324..bf5df9597288 100644
> > > > > > > --- a/fs/nfsd/netlink.c
> > > > > > > +++ b/fs/nfsd/netlink.c
> > > > > > > @@ -10,6 +10,13 @@
> > > > > > >  
> > > > > > >  #include <uapi/linux/nfsd_netlink.h>
> > > > > > >  
> > > > > > > +/* Common nested types */
> > > > > > > +const struct nla_policy nfsd_version_nl_policy[NFSD_A_VERSION_ENABLED + 1] = {
> > > > > > > +	[NFSD_A_VERSION_MAJOR] = { .type = NLA_U32, },
> > > > > > > +	[NFSD_A_VERSION_MINOR] = { .type = NLA_U32, },
> > > > > > > +	[NFSD_A_VERSION_ENABLED] = { .type = NLA_FLAG, },
> > > > > > > +};
> > > > > > > +
> > > > > > >  /* NFSD_CMD_THREADS_SET - do */
> > > > > > >  static const struct nla_policy nfsd_threads_set_nl_policy[NFSD_A_SERVER_WORKER_LEASETIME + 1] = {
> > > > > > >  	[NFSD_A_SERVER_WORKER_THREADS] = { .type = NLA_U32, },
> > > > > > > @@ -17,6 +24,11 @@ static const struct nla_policy nfsd_threads_set_nl_policy[NFSD_A_SERVER_WORKER_L
> > > > > > >  	[NFSD_A_SERVER_WORKER_LEASETIME] = { .type = NLA_U32, },
> > > > > > >  };
> > > > > > >  
> > > > > > > +/* NFSD_CMD_VERSION_SET - do */
> > > > > > > +static const struct nla_policy nfsd_version_set_nl_policy[NFSD_A_SERVER_PROTO_VERSION + 1] = {
> > > > > > > +	[NFSD_A_SERVER_PROTO_VERSION] = NLA_POLICY_NESTED(nfsd_version_nl_policy),
> > > > > > > +};
> > > > > > > +
> > > > > > >  /* Ops table for nfsd */
> > > > > > >  static const struct genl_split_ops nfsd_nl_ops[] = {
> > > > > > >  	{
> > > > > > > @@ -38,6 +50,18 @@ static const struct genl_split_ops nfsd_nl_ops[] = {
> > > > > > >  		.doit	= nfsd_nl_threads_get_doit,
> > > > > > >  		.flags	= GENL_CMD_CAP_DO,
> > > > > > >  	},
> > > > > > > +	{
> > > > > > > +		.cmd		= NFSD_CMD_VERSION_SET,
> > > > > > > +		.doit		= nfsd_nl_version_set_doit,
> > > > > > > +		.policy		= nfsd_version_set_nl_policy,
> > > > > > > +		.maxattr	= NFSD_A_SERVER_PROTO_VERSION,
> > > > > > > +		.flags		= GENL_ADMIN_PERM | GENL_CMD_CAP_DO,
> > > > > > > +	},
> > > > > > > +	{
> > > > > > > +		.cmd	= NFSD_CMD_VERSION_GET,
> > > > > > > +		.doit	= nfsd_nl_version_get_doit,
> > > > > > > +		.flags	= GENL_CMD_CAP_DO,
> > > > > > > +	},
> > > > > > >  };
> > > > > > >  
> > > > > > >  struct genl_family nfsd_nl_family __ro_after_init = {
> > > > > > > diff --git a/fs/nfsd/netlink.h b/fs/nfsd/netlink.h
> > > > > > > index 4137fac477e4..c7c0da275481 100644
> > > > > > > --- a/fs/nfsd/netlink.h
> > > > > > > +++ b/fs/nfsd/netlink.h
> > > > > > > @@ -11,6 +11,9 @@
> > > > > > >  
> > > > > > >  #include <uapi/linux/nfsd_netlink.h>
> > > > > > >  
> > > > > > > +/* Common nested types */
> > > > > > > +extern const struct nla_policy nfsd_version_nl_policy[NFSD_A_VERSION_ENABLED + 1];
> > > > > > > +
> > > > > > >  int nfsd_nl_rpc_status_get_start(struct netlink_callback *cb);
> > > > > > >  int nfsd_nl_rpc_status_get_done(struct netlink_callback *cb);
> > > > > > >  
> > > > > > > @@ -18,6 +21,8 @@ int nfsd_nl_rpc_status_get_dumpit(struct sk_buff *skb,
> > > > > > >  				  struct netlink_callback *cb);
> > > > > > >  int nfsd_nl_threads_set_doit(struct sk_buff *skb, struct genl_info *info);
> > > > > > >  int nfsd_nl_threads_get_doit(struct sk_buff *skb, struct genl_info *info);
> > > > > > > +int nfsd_nl_version_set_doit(struct sk_buff *skb, struct genl_info *info);
> > > > > > > +int nfsd_nl_version_get_doit(struct sk_buff *skb, struct genl_info *info);
> > > > > > >  
> > > > > > >  extern struct genl_family nfsd_nl_family;
> > > > > > >  
> > > > > > > diff --git a/fs/nfsd/netns.h b/fs/nfsd/netns.h
> > > > > > > index d4be519b5734..14ec15656320 100644
> > > > > > > --- a/fs/nfsd/netns.h
> > > > > > > +++ b/fs/nfsd/netns.h
> > > > > > > @@ -218,6 +218,7 @@ struct nfsd_net {
> > > > > > >  /* Simple check to find out if a given net was properly initialized */
> > > > > > >  #define nfsd_netns_ready(nn) ((nn)->sessionid_hashtbl)
> > > > > > >  
> > > > > > > +extern bool nfsd_support_version(int vers);
> > > > > > >  extern void nfsd_netns_free_versions(struct nfsd_net *nn);
> > > > > > >  
> > > > > > >  extern unsigned int nfsd_net_id;
> > > > > > > diff --git a/fs/nfsd/nfsctl.c b/fs/nfsd/nfsctl.c
> > > > > > > index 38a5df03981b..2c8929ef79e9 100644
> > > > > > > --- a/fs/nfsd/nfsctl.c
> > > > > > > +++ b/fs/nfsd/nfsctl.c
> > > > > > > @@ -1757,6 +1757,156 @@ int nfsd_nl_threads_get_doit(struct sk_buff *skb, struct genl_info *info)
> > > > > > >  	return err;
> > > > > > >  }
> > > > > > >  
> > > > > > > +/**
> > > > > > > + * nfsd_nl_version_set_doit - set the nfs enabled versions
> > > > > > > + * @skb: reply buffer
> > > > > > > + * @info: netlink metadata and command arguments
> > > > > > > + *
> > > > > > > + * Return 0 on success or a negative errno.
> > > > > > > + */
> > > > > > > +int nfsd_nl_version_set_doit(struct sk_buff *skb, struct genl_info *info)
> > > > > > > +{
> > > > > > > +	const struct nlattr *attr;
> > > > > > > +	struct nfsd_net *nn;
> > > > > > > +	int i, rem;
> > > > > > > +
> > > > > > > +	if (GENL_REQ_ATTR_CHECK(info, NFSD_A_SERVER_PROTO_VERSION))
> > > > > > > +		return -EINVAL;
> > > > > > > +
> > > > > > > +	mutex_lock(&nfsd_mutex);
> > > > > > > +
> > > > > > > +	nn = net_generic(genl_info_net(info), nfsd_net_id);
> > > > > > > +	if (nn->nfsd_serv) {
> > > > > > > +		mutex_unlock(&nfsd_mutex);
> > > > > > > +		return -EBUSY;
> > > > > > > +	}
> > > > > > > +
> > > > > > > +	/* clear current supported versions. */
> > > > > > > +	nfsd_vers(nn, 2, NFSD_CLEAR);
> > > > > > > +	nfsd_vers(nn, 3, NFSD_CLEAR);
> > > > > > > +	for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++)
> > > > > > > +		nfsd_minorversion(nn, i, NFSD_CLEAR);
> > > > > > > +
> > > > > > > +	nlmsg_for_each_attr(attr, info->nlhdr, GENL_HDRLEN, rem) {
> > > > > > > +		struct nlattr *tb[NFSD_A_VERSION_MAX + 1];
> > > > > > > +		u32 major, minor = 0;
> > > > > > > +		bool enabled;
> > > > > > > +
> > > > > > > +		if (nla_type(attr) != NFSD_A_SERVER_PROTO_VERSION)
> > > > > > > +			continue;
> > > > > > > +
> > > > > > > +		if (nla_parse_nested(tb, NFSD_A_VERSION_MAX, attr,
> > > > > > > +				     nfsd_version_nl_policy, info->extack) < 0)
> > > > > > > +			continue;
> > > > > > > +
> > > > > > > +		if (!tb[NFSD_A_VERSION_MAJOR])
> > > > > > > +			continue;
> > > > > > > +
> > > > > > > +		major = nla_get_u32(tb[NFSD_A_VERSION_MAJOR]);
> > > > > > > +		if (tb[NFSD_A_VERSION_MINOR])
> > > > > > > +			minor = nla_get_u32(tb[NFSD_A_VERSION_MINOR]);
> > > > > > > +
> > > > > > > +		enabled = nla_get_flag(tb[NFSD_A_VERSION_ENABLED]);
> > > > > > > +
> > > > > > > +		switch (major) {
> > > > > > > +		case 4:
> > > > > > > +			nfsd_minorversion(nn, minor, enabled ? NFSD_SET : NFSD_CLEAR);
> > > > > > > +			break;
> > > > > > > +		case 3:
> > > > > > > +		case 2:
> > > > > > > +			if (!minor)
> > > > > > > +				nfsd_vers(nn, major, enabled ? NFSD_SET : NFSD_CLEAR);
> > > > > > > +			break;
> > > > > > > +		default:
> > > > > > > +			break;
> > > > > > > +		}
> > > > > > > +	}
> > > > > > > +
> > > > > > > +	mutex_unlock(&nfsd_mutex);
> > > > > > > +
> > > > > > > +	return 0;
> > > > > > > +}
> > > > > > > +
> > > > > > > +/**
> > > > > > > + * nfsd_nl_version_get_doit - get the nfs enabled versions
> > > > > > > + * @skb: reply buffer
> > > > > > > + * @info: netlink metadata and command arguments
> > > > > > > + *
> > > > > > > + * Return 0 on success or a negative errno.
> > > > > > > + */
> > > > > > > +int nfsd_nl_version_get_doit(struct sk_buff *skb, struct genl_info *info)
> > > > > > > +{
> > > > > > > +	struct nfsd_net *nn;
> > > > > > > +	int i, err;
> > > > > > > +	void *hdr;
> > > > > > > +
> > > > > > > +	skb = genlmsg_new(GENLMSG_DEFAULT_SIZE, GFP_KERNEL);
> > > > > > > +	if (!skb)
> > > > > > > +		return -ENOMEM;
> > > > > > > +
> > > > > > > +	hdr = genlmsg_iput(skb, info);
> > > > > > > +	if (!hdr) {
> > > > > > > +		err = -EMSGSIZE;
> > > > > > > +		goto err_free_msg;
> > > > > > > +	}
> > > > > > > +
> > > > > > > +	mutex_lock(&nfsd_mutex);
> > > > > > > +	nn = net_generic(genl_info_net(info), nfsd_net_id);
> > > > > > > +
> > > > > > > +	for (i = 2; i <= 4; i++) {
> > > > > > > +		int j;
> > > > > > > +
> > > > > > > +		for (j = 0; j <= NFSD_SUPPORTED_MINOR_VERSION; j++) {
> > > > > > > +			struct nlattr *attr;
> > > > > > > +
> > > > > > > +			/* Don't record any versions the kernel doesn't have
> > > > > > > +			 * compiled in
> > > > > > > +			 */
> > > > > > > +			if (!nfsd_support_version(i))
> > > > > > > +				continue;
> > > > > > > +
> > > > > > > +			/* NFSv{2,3} does not support minor numbers */
> > > > > > > +			if (i < 4 && j)
> > > > > > > +				continue;
> > > > > > > +
> > > > > > > +			attr = nla_nest_start(skb,
> > > > > > > +					      NFSD_A_SERVER_PROTO_VERSION);
> > > > > > > +			if (!attr) {
> > > > > > > +				err = -EINVAL;
> > > > > > > +				goto err_nfsd_unlock;
> > > > > > > +			}
> > > > > > > +
> > > > > > > +			if (nla_put_u32(skb, NFSD_A_VERSION_MAJOR, i) ||
> > > > > > > +			    nla_put_u32(skb, NFSD_A_VERSION_MINOR, j)) {
> > > > > > > +				err = -EINVAL;
> > > > > > > +				goto err_nfsd_unlock;
> > > > > > > +			}
> > > > > > > +
> > > > > > > +			/* Set the enabled flag if the version is enabled */
> > > > > > > +			if (nfsd_vers(nn, i, NFSD_TEST) &&
> > > > > > > +			    (i < 4 || nfsd_minorversion(nn, j, NFSD_TEST)) &&
> > > > > > > +			    nla_put_flag(skb, NFSD_A_VERSION_ENABLED)) {
> > > > > > > +				err = -EINVAL;
> > > > > > > +				goto err_nfsd_unlock;
> > > > > > > +			}
> > > > > > > +
> > > > > > > +			nla_nest_end(skb, attr);
> > > > > > > +		}
> > > > > > > +	}
> > > > > > > +
> > > > > > > +	mutex_unlock(&nfsd_mutex);
> > > > > > > +	genlmsg_end(skb, hdr);
> > > > > > > +
> > > > > > > +	return genlmsg_reply(skb, info);
> > > > > > > +
> > > > > > > +err_nfsd_unlock:
> > > > > > > +	mutex_unlock(&nfsd_mutex);
> > > > > > > +err_free_msg:
> > > > > > > +	nlmsg_free(skb);
> > > > > > > +
> > > > > > > +	return err;
> > > > > > > +}
> > > > > > > +
> > > > > > >  /**
> > > > > > >   * nfsd_net_init - Prepare the nfsd_net portion of a new net namespace
> > > > > > >   * @net: a freshly-created network namespace
> > > > > > > diff --git a/fs/nfsd/nfssvc.c b/fs/nfsd/nfssvc.c
> > > > > > > index ca193f7ff0e1..4fc91f50138a 100644
> > > > > > > --- a/fs/nfsd/nfssvc.c
> > > > > > > +++ b/fs/nfsd/nfssvc.c
> > > > > > > @@ -133,8 +133,7 @@ struct svc_program		nfsd_program = {
> > > > > > >  	.pg_rpcbind_set		= nfsd_rpcbind_set,
> > > > > > >  };
> > > > > > >  
> > > > > > > -static bool
> > > > > > > -nfsd_support_version(int vers)
> > > > > > > +bool nfsd_support_version(int vers)
> > > > > > >  {
> > > > > > >  	if (vers >= NFSD_MINVERS && vers < NFSD_NRVERS)
> > > > > > >  		return nfsd_version[vers] != NULL;
> > > > > > > diff --git a/include/uapi/linux/nfsd_netlink.h b/include/uapi/linux/nfsd_netlink.h
> > > > > > > index ccc78a5ee650..8a0a2b344923 100644
> > > > > > > --- a/include/uapi/linux/nfsd_netlink.h
> > > > > > > +++ b/include/uapi/linux/nfsd_netlink.h
> > > > > > > @@ -38,10 +38,28 @@ enum {
> > > > > > >  	NFSD_A_SERVER_WORKER_MAX = (__NFSD_A_SERVER_WORKER_MAX - 1)
> > > > > > >  };
> > > > > > >  
> > > > > > > +enum {
> > > > > > > +	NFSD_A_VERSION_MAJOR = 1,
> > > > > > > +	NFSD_A_VERSION_MINOR,
> > > > > > > +	NFSD_A_VERSION_ENABLED,
> > > > > > > +
> > > > > > > +	__NFSD_A_VERSION_MAX,
> > > > > > > +	NFSD_A_VERSION_MAX = (__NFSD_A_VERSION_MAX - 1)
> > > > > > > +};
> > > > > > > +
> > > > > > > +enum {
> > > > > > > +	NFSD_A_SERVER_PROTO_VERSION = 1,
> > > > > > > +
> > > > > > > +	__NFSD_A_SERVER_PROTO_MAX,
> > > > > > > +	NFSD_A_SERVER_PROTO_MAX = (__NFSD_A_SERVER_PROTO_MAX - 1)
> > > > > > > +};
> > > > > > > +
> > > > > > >  enum {
> > > > > > >  	NFSD_CMD_RPC_STATUS_GET = 1,
> > > > > > >  	NFSD_CMD_THREADS_SET,
> > > > > > >  	NFSD_CMD_THREADS_GET,
> > > > > > > +	NFSD_CMD_VERSION_SET,
> > > > > > > +	NFSD_CMD_VERSION_GET,
> > > > > > >  
> > > > > > >  	__NFSD_CMD_MAX,
> > > > > > >  	NFSD_CMD_MAX = (__NFSD_CMD_MAX - 1)
> > > > > > > -- 
> > > > > > > 2.44.0
> > > > > > > 
> > > > > > > 
> > > > > > 
> > > > > 
> > > > > -- 
> > > > > Jeff Layton <jlayton@kernel.org>
> > > > > 
> > > > 
> > > 
> > > -- 
> > > Jeff Layton <jlayton@kernel.org>
> > > 
> > 
> 
> -- 
> Jeff Layton <jlayton@kernel.org>
>
Jeff Layton April 17, 2024, 11:25 a.m. UTC | #8
On Wed, 2024-04-17 at 10:43 +1000, NeilBrown wrote:
> On Wed, 17 Apr 2024, Jeff Layton wrote:
> > On Wed, 2024-04-17 at 09:05 +1000, NeilBrown wrote:
> > > On Wed, 17 Apr 2024, Jeff Layton wrote:
> > > > On Wed, 2024-04-17 at 07:48 +1000, NeilBrown wrote:
> > > > > On Tue, 16 Apr 2024, Jeff Layton wrote:
> > > > > > On Tue, 2024-04-16 at 13:16 +1000, NeilBrown wrote:
> > > > > > > On Tue, 16 Apr 2024, Lorenzo Bianconi wrote:
> > > > > > > > Introduce write_version netlink command through a "declarative" interface.
> > > > > > > > This patch introduces a change in behavior since for version-set userspace
> > > > > > > > is expected to provide a NFS major/minor version list it wants to enable
> > > > > > > > while all the other ones will be disabled. (procfs write_version
> > > > > > > > command implements imperative interface where the admin writes +3/-3 to
> > > > > > > > enable/disable a single version.
> > > > > > > 
> > > > > > > It seems a little weird to me that the interface always disables all
> > > > > > > version, but then also allows individual versions to be disabled.
> > > > > > > 
> > > > > > > Would it be reasonable to simply ignore the "enabled" flag when setting
> > > > > > > version, and just enable all versions listed??
> > > > > > > 
> > > > > > > Or maybe only enable those with the flag, and don't disable those
> > > > > > > without the flag?
> > > > > > > 
> > > > > > > Those don't necessarily seem much better - but the current behaviour
> > > > > > > still seems odd.
> > > > > > > 
> > > > > > 
> > > > > > I think it makes sense.
> > > > > > 
> > > > > > We disable all versions, and enable any that have the "enabled" flag set
> > > > > > in the call from userland. Userland technically needn't send down the
> > > > > > versions that are disabled in the call, but the current userland program
> > > > > > does.
> > > > > > 
> > > > > > I worry about imperative interfaces that might only say -- "enable v4.1,
> > > > > > but disable v3" and leave the others in their current state. That
> > > > > > requires that both the kernel and userland keep state about what
> > > > > > versions are currently enabled and disabled, and it's possible to get
> > > > > > that wrong.
> > > > > 
> > > > > I understand and support your aversion for imperative interfaces.
> > > > > But this interface, as currently implemented, looks somewhat imperative.
> > > > > The message sent to the kernel could enable some interfaces and then
> > > > > disable them.  I know that isn't the intent, but it is what the code
> > > > > supports.  Hence "weird".
> > > > > 
> > > > > We could add code to make that sort of thing impossible, but there isn't
> > > > > much point.  Better to make it syntactically impossible.
> > > > > 
> > > > > Realistically there will never be NFSv4.3 as there is no need - new
> > > > > features can be added incrementally. 
> > > > > 
> > > > 
> > > > There is no need _so_far_. Who knows what the future will bring? Maybe
> > > > we'll need v4.3 in order to add some needed feature?
> > > > 
> > > > >  So we could just pass an array of
> > > > > 5 active flags: 2,3,4.0,4.1,4.2.  I suspect you wouldn't like that and
> > > > > I'm not sure that I do either.  A "read" would return the same array
> > > > > with 3 possible states: unavailable, disabled, enabled.  (Maybe the
> > > > > array could be variable length so 5.0 could be added one day...).
> > > > > 
> > > > 
> > > > A set of flags is basically what this interface is. They just happen to
> > > > also be labeled with the major and minorversion. I think that's a good
> > > > thing.
> > > 
> > > Good for whom?  Labelling allows for labelling inconsistencies.
> > > 
> > 
> > Now you're just being silly. You wanted a variable length array, but
> > what if the next slot is not v4.3 but 5.0? A positional interpretation
> > of a slot in an array is just as just as subject to problems.
> 
> I don't think it could look like a imperative interface, which the
> current one does a bit.
> 
> > 
> > > Maybe the kernel should be able to provide an ordered list of labels,
> > > and separately an array of which labels are enabled/disabled.
> > > Userspace could send down a new set of enabled/disabled flags based on
> > > the agreed set of labels.
> > > 
> > 
> > How is this better than what's been proposed? One strength of netlink is
> > that the data is structured. The already proposed interface takes
> > advantage of that.
> > 
> > > Here is a question that is a bit of a diversion, but might help us
> > > understand the context more fully:
> > > 
> > >   Why would anyone disable v4.2 separately from v4.1 ??
> > > 
> > 
> > Furthermore, what does it mean to disable v4.1 but leave v4.2 enabled?
> 
> Indeed!
> 
> > 
> > > I understand that v2, v3, v4.0, v4.1 are effectively different protocols
> > > and you might want to ensure that only the approved/tested protocol is
> > > used.  But v4.2 is just a few enhancements on v4.1.  Why would you want
> > > to disable it?
> > > 
> > > The answer I can think of that there might be bugs (perish the
> > > thought!!) in some of those features so you might want to avoid using
> > > them.
> > > But in that case, it is really the features that you want to suppress,
> > > not the protocol version.
> > > 
> > > Maybe I might want to disable delegation - to just write delegation.
> > > Can I do that?  What if I just want to disable server-side copy, but
> > > keep fallocate and umask support?
> > > 
> > > i.e.  is a list of versions really the granularity that we want to use
> > > for this interface?
> > > 
> > 
> > Our current goal is to replace rpc.nfsd with a new program that works
> > via netlink. An important bit of what rpc.nfsd does is start the NFS
> > server with the settings in /etc/nfs.conf. Some of those settings are
> > vers3=, vers4.0=, etc. that govern how /proc/fs/nfsd/versions is set.
> > We have an immediate need to deal with those settings today, and
> > probably will for quite some time.
> > 
> > I'm not opposed to augmenting that with something more granular, but I
> > don't think we should block this interface and wait on that. We can
> > extend the interface at some point in the future to take a new feature
> > bitmask or something, and just declare that (e.g.) declaring vers4.2=n
> > disables some subset of those features.
> 
> I agree that we don't want to block "good" while we wait for "perfect".  I
> just want to be sure that what we have is "good".
> 
> The current /proc interface uses strings like "v3" and "v4.1".
> The proposed netlink interface uses pairs on u32s - "major" and "minor".
> So we lose some easy paths for extensibility.  Are we comfortable with
> that?
> 
> This isn't a big deal - certainly not a blocked.  I just don't feel
> entirely comfortable about the current interface and I'm exploring to
> see if there might be something better.
> 
> 

Ok, I'm not convinced that anything you've proposed so far is better
than what we have, but I'm open-minded.

At this point we have these to-dos:

1) make the threads interface accept an array of integers rather than a
singleton. This is needed when sunrpc.pool_mode is not global.

2) have the interface pass down the scope string in the THREADS_SET
instead of making userland change the UTS namespace before starting
threads. This should just mean adding a new optional string attribute to
the threads interfaces.

...anything else we're missing that needs doing here? We'd really like
to see this in -next soon, so we can possibly make v6.10 with this.

Thanks,
NeilBrown April 22, 2024, 3:36 a.m. UTC | #9
On Wed, 17 Apr 2024, Jeff Layton wrote:
> On Wed, 2024-04-17 at 10:43 +1000, NeilBrown wrote:
> > On Wed, 17 Apr 2024, Jeff Layton wrote:
> > > On Wed, 2024-04-17 at 09:05 +1000, NeilBrown wrote:
> > > > On Wed, 17 Apr 2024, Jeff Layton wrote:
> > > > > On Wed, 2024-04-17 at 07:48 +1000, NeilBrown wrote:
> > > > > > On Tue, 16 Apr 2024, Jeff Layton wrote:
> > > > > > > On Tue, 2024-04-16 at 13:16 +1000, NeilBrown wrote:
> > > > > > > > On Tue, 16 Apr 2024, Lorenzo Bianconi wrote:
> > > > > > > > > Introduce write_version netlink command through a "declarative" interface.
> > > > > > > > > This patch introduces a change in behavior since for version-set userspace
> > > > > > > > > is expected to provide a NFS major/minor version list it wants to enable
> > > > > > > > > while all the other ones will be disabled. (procfs write_version
> > > > > > > > > command implements imperative interface where the admin writes +3/-3 to
> > > > > > > > > enable/disable a single version.
> > > > > > > > 
> > > > > > > > It seems a little weird to me that the interface always disables all
> > > > > > > > version, but then also allows individual versions to be disabled.
> > > > > > > > 
> > > > > > > > Would it be reasonable to simply ignore the "enabled" flag when setting
> > > > > > > > version, and just enable all versions listed??
> > > > > > > > 
> > > > > > > > Or maybe only enable those with the flag, and don't disable those
> > > > > > > > without the flag?
> > > > > > > > 
> > > > > > > > Those don't necessarily seem much better - but the current behaviour
> > > > > > > > still seems odd.
> > > > > > > > 
> > > > > > > 
> > > > > > > I think it makes sense.
> > > > > > > 
> > > > > > > We disable all versions, and enable any that have the "enabled" flag set
> > > > > > > in the call from userland. Userland technically needn't send down the
> > > > > > > versions that are disabled in the call, but the current userland program
> > > > > > > does.
> > > > > > > 
> > > > > > > I worry about imperative interfaces that might only say -- "enable v4.1,
> > > > > > > but disable v3" and leave the others in their current state. That
> > > > > > > requires that both the kernel and userland keep state about what
> > > > > > > versions are currently enabled and disabled, and it's possible to get
> > > > > > > that wrong.
> > > > > > 
> > > > > > I understand and support your aversion for imperative interfaces.
> > > > > > But this interface, as currently implemented, looks somewhat imperative.
> > > > > > The message sent to the kernel could enable some interfaces and then
> > > > > > disable them.  I know that isn't the intent, but it is what the code
> > > > > > supports.  Hence "weird".
> > > > > > 
> > > > > > We could add code to make that sort of thing impossible, but there isn't
> > > > > > much point.  Better to make it syntactically impossible.
> > > > > > 
> > > > > > Realistically there will never be NFSv4.3 as there is no need - new
> > > > > > features can be added incrementally. 
> > > > > > 
> > > > > 
> > > > > There is no need _so_far_. Who knows what the future will bring? Maybe
> > > > > we'll need v4.3 in order to add some needed feature?
> > > > > 
> > > > > >  So we could just pass an array of
> > > > > > 5 active flags: 2,3,4.0,4.1,4.2.  I suspect you wouldn't like that and
> > > > > > I'm not sure that I do either.  A "read" would return the same array
> > > > > > with 3 possible states: unavailable, disabled, enabled.  (Maybe the
> > > > > > array could be variable length so 5.0 could be added one day...).
> > > > > > 
> > > > > 
> > > > > A set of flags is basically what this interface is. They just happen to
> > > > > also be labeled with the major and minorversion. I think that's a good
> > > > > thing.
> > > > 
> > > > Good for whom?  Labelling allows for labelling inconsistencies.
> > > > 
> > > 
> > > Now you're just being silly. You wanted a variable length array, but
> > > what if the next slot is not v4.3 but 5.0? A positional interpretation
> > > of a slot in an array is just as just as subject to problems.
> > 
> > I don't think it could look like a imperative interface, which the
> > current one does a bit.
> > 
> > > 
> > > > Maybe the kernel should be able to provide an ordered list of labels,
> > > > and separately an array of which labels are enabled/disabled.
> > > > Userspace could send down a new set of enabled/disabled flags based on
> > > > the agreed set of labels.
> > > > 
> > > 
> > > How is this better than what's been proposed? One strength of netlink is
> > > that the data is structured. The already proposed interface takes
> > > advantage of that.
> > > 
> > > > Here is a question that is a bit of a diversion, but might help us
> > > > understand the context more fully:
> > > > 
> > > >   Why would anyone disable v4.2 separately from v4.1 ??
> > > > 
> > > 
> > > Furthermore, what does it mean to disable v4.1 but leave v4.2 enabled?
> > 
> > Indeed!
> > 
> > > 
> > > > I understand that v2, v3, v4.0, v4.1 are effectively different protocols
> > > > and you might want to ensure that only the approved/tested protocol is
> > > > used.  But v4.2 is just a few enhancements on v4.1.  Why would you want
> > > > to disable it?
> > > > 
> > > > The answer I can think of that there might be bugs (perish the
> > > > thought!!) in some of those features so you might want to avoid using
> > > > them.
> > > > But in that case, it is really the features that you want to suppress,
> > > > not the protocol version.
> > > > 
> > > > Maybe I might want to disable delegation - to just write delegation.
> > > > Can I do that?  What if I just want to disable server-side copy, but
> > > > keep fallocate and umask support?
> > > > 
> > > > i.e.  is a list of versions really the granularity that we want to use
> > > > for this interface?
> > > > 
> > > 
> > > Our current goal is to replace rpc.nfsd with a new program that works
> > > via netlink. An important bit of what rpc.nfsd does is start the NFS
> > > server with the settings in /etc/nfs.conf. Some of those settings are
> > > vers3=, vers4.0=, etc. that govern how /proc/fs/nfsd/versions is set.
> > > We have an immediate need to deal with those settings today, and
> > > probably will for quite some time.
> > > 
> > > I'm not opposed to augmenting that with something more granular, but I
> > > don't think we should block this interface and wait on that. We can
> > > extend the interface at some point in the future to take a new feature
> > > bitmask or something, and just declare that (e.g.) declaring vers4.2=n
> > > disables some subset of those features.
> > 
> > I agree that we don't want to block "good" while we wait for "perfect".  I
> > just want to be sure that what we have is "good".
> > 
> > The current /proc interface uses strings like "v3" and "v4.1".
> > The proposed netlink interface uses pairs on u32s - "major" and "minor".
> > So we lose some easy paths for extensibility.  Are we comfortable with
> > that?
> > 
> > This isn't a big deal - certainly not a blocked.  I just don't feel
> > entirely comfortable about the current interface and I'm exploring to
> > see if there might be something better.
> > 
> > 
> 
> Ok, I'm not convinced that anything you've proposed so far is better
> than what we have, but I'm open-minded.

Thanks.  I admit that I don't think anything I've come up with is better
either.

> 
> At this point we have these to-dos:
> 
> 1) make the threads interface accept an array of integers rather than a
> singleton. This is needed when sunrpc.pool_mode is not global.
> 

While this I think probably still makes sense, I'm feeling less
enthusiastic about it and probably wouldn't complain if it didn't
happen.

I don't like that fact that we need to configure a number of threads.  I
would much rather it grow/shrink dynamically.  I've got more patches
that are heading in this direction but they aren't ready yet.

If we do land these and they prove to be effective, then configuring
per-pool threads would become extremely uninteresting.  The demand on
each pool would adjust each pool separately.

So if use an array here turns out to be problematic for some reason, I
won't complain.

> 2) have the interface pass down the scope string in the THREADS_SET
> instead of making userland change the UTS namespace before starting
> threads. This should just mean adding a new optional string attribute to
> the threads interfaces.
> 
> ...anything else we're missing that needs doing here? We'd really like
> to see this in -next soon, so we can possibly make v6.10 with this.

I haven't yet found any obvious gaps.  I agree that we can and should
move forward promptly.

Thanks,
NeilBrown


> 
> Thanks,
> -- 
> Jeff Layton <jlayton@kernel.org>
>
Jeff Layton April 23, 2024, 12:42 p.m. UTC | #10
On Mon, 2024-04-22 at 13:36 +1000, NeilBrown wrote:
> On Wed, 17 Apr 2024, Jeff Layton wrote:
> > On Wed, 2024-04-17 at 10:43 +1000, NeilBrown wrote:
> > > On Wed, 17 Apr 2024, Jeff Layton wrote:
> > > > On Wed, 2024-04-17 at 09:05 +1000, NeilBrown wrote:
> > > > > On Wed, 17 Apr 2024, Jeff Layton wrote:
> > > > > > On Wed, 2024-04-17 at 07:48 +1000, NeilBrown wrote:
> > > > > > > On Tue, 16 Apr 2024, Jeff Layton wrote:
> > > > > > > > On Tue, 2024-04-16 at 13:16 +1000, NeilBrown wrote:
> > > > > > > > > On Tue, 16 Apr 2024, Lorenzo Bianconi wrote:
> > > > > > > > > > Introduce write_version netlink command through a "declarative" interface.
> > > > > > > > > > This patch introduces a change in behavior since for version-set userspace
> > > > > > > > > > is expected to provide a NFS major/minor version list it wants to enable
> > > > > > > > > > while all the other ones will be disabled. (procfs write_version
> > > > > > > > > > command implements imperative interface where the admin writes +3/-3 to
> > > > > > > > > > enable/disable a single version.
> > > > > > > > > 
> > > > > > > > > It seems a little weird to me that the interface always disables all
> > > > > > > > > version, but then also allows individual versions to be disabled.
> > > > > > > > > 
> > > > > > > > > Would it be reasonable to simply ignore the "enabled" flag when setting
> > > > > > > > > version, and just enable all versions listed??
> > > > > > > > > 
> > > > > > > > > Or maybe only enable those with the flag, and don't disable those
> > > > > > > > > without the flag?
> > > > > > > > > 
> > > > > > > > > Those don't necessarily seem much better - but the current behaviour
> > > > > > > > > still seems odd.
> > > > > > > > > 
> > > > > > > > 
> > > > > > > > I think it makes sense.
> > > > > > > > 
> > > > > > > > We disable all versions, and enable any that have the "enabled" flag set
> > > > > > > > in the call from userland. Userland technically needn't send down the
> > > > > > > > versions that are disabled in the call, but the current userland program
> > > > > > > > does.
> > > > > > > > 
> > > > > > > > I worry about imperative interfaces that might only say -- "enable v4.1,
> > > > > > > > but disable v3" and leave the others in their current state. That
> > > > > > > > requires that both the kernel and userland keep state about what
> > > > > > > > versions are currently enabled and disabled, and it's possible to get
> > > > > > > > that wrong.
> > > > > > > 
> > > > > > > I understand and support your aversion for imperative interfaces.
> > > > > > > But this interface, as currently implemented, looks somewhat imperative.
> > > > > > > The message sent to the kernel could enable some interfaces and then
> > > > > > > disable them.  I know that isn't the intent, but it is what the code
> > > > > > > supports.  Hence "weird".
> > > > > > > 
> > > > > > > We could add code to make that sort of thing impossible, but there isn't
> > > > > > > much point.  Better to make it syntactically impossible.
> > > > > > > 
> > > > > > > Realistically there will never be NFSv4.3 as there is no need - new
> > > > > > > features can be added incrementally. 
> > > > > > > 
> > > > > > 
> > > > > > There is no need _so_far_. Who knows what the future will bring? Maybe
> > > > > > we'll need v4.3 in order to add some needed feature?
> > > > > > 
> > > > > > >  So we could just pass an array of
> > > > > > > 5 active flags: 2,3,4.0,4.1,4.2.  I suspect you wouldn't like that and
> > > > > > > I'm not sure that I do either.  A "read" would return the same array
> > > > > > > with 3 possible states: unavailable, disabled, enabled.  (Maybe the
> > > > > > > array could be variable length so 5.0 could be added one day...).
> > > > > > > 
> > > > > > 
> > > > > > A set of flags is basically what this interface is. They just happen to
> > > > > > also be labeled with the major and minorversion. I think that's a good
> > > > > > thing.
> > > > > 
> > > > > Good for whom?  Labelling allows for labelling inconsistencies.
> > > > > 
> > > > 
> > > > Now you're just being silly. You wanted a variable length array, but
> > > > what if the next slot is not v4.3 but 5.0? A positional interpretation
> > > > of a slot in an array is just as just as subject to problems.
> > > 
> > > I don't think it could look like a imperative interface, which the
> > > current one does a bit.
> > > 
> > > > 
> > > > > Maybe the kernel should be able to provide an ordered list of labels,
> > > > > and separately an array of which labels are enabled/disabled.
> > > > > Userspace could send down a new set of enabled/disabled flags based on
> > > > > the agreed set of labels.
> > > > > 
> > > > 
> > > > How is this better than what's been proposed? One strength of netlink is
> > > > that the data is structured. The already proposed interface takes
> > > > advantage of that.
> > > > 
> > > > > Here is a question that is a bit of a diversion, but might help us
> > > > > understand the context more fully:
> > > > > 
> > > > >   Why would anyone disable v4.2 separately from v4.1 ??
> > > > > 
> > > > 
> > > > Furthermore, what does it mean to disable v4.1 but leave v4.2 enabled?
> > > 
> > > Indeed!
> > > 
> > > > 
> > > > > I understand that v2, v3, v4.0, v4.1 are effectively different protocols
> > > > > and you might want to ensure that only the approved/tested protocol is
> > > > > used.  But v4.2 is just a few enhancements on v4.1.  Why would you want
> > > > > to disable it?
> > > > > 
> > > > > The answer I can think of that there might be bugs (perish the
> > > > > thought!!) in some of those features so you might want to avoid using
> > > > > them.
> > > > > But in that case, it is really the features that you want to suppress,
> > > > > not the protocol version.
> > > > > 
> > > > > Maybe I might want to disable delegation - to just write delegation.
> > > > > Can I do that?  What if I just want to disable server-side copy, but
> > > > > keep fallocate and umask support?
> > > > > 
> > > > > i.e.  is a list of versions really the granularity that we want to use
> > > > > for this interface?
> > > > > 
> > > > 
> > > > Our current goal is to replace rpc.nfsd with a new program that works
> > > > via netlink. An important bit of what rpc.nfsd does is start the NFS
> > > > server with the settings in /etc/nfs.conf. Some of those settings are
> > > > vers3=, vers4.0=, etc. that govern how /proc/fs/nfsd/versions is set.
> > > > We have an immediate need to deal with those settings today, and
> > > > probably will for quite some time.
> > > > 
> > > > I'm not opposed to augmenting that with something more granular, but I
> > > > don't think we should block this interface and wait on that. We can
> > > > extend the interface at some point in the future to take a new feature
> > > > bitmask or something, and just declare that (e.g.) declaring vers4.2=n
> > > > disables some subset of those features.
> > > 
> > > I agree that we don't want to block "good" while we wait for "perfect".  I
> > > just want to be sure that what we have is "good".
> > > 
> > > The current /proc interface uses strings like "v3" and "v4.1".
> > > The proposed netlink interface uses pairs on u32s - "major" and "minor".
> > > So we lose some easy paths for extensibility.  Are we comfortable with
> > > that?
> > > 
> > > This isn't a big deal - certainly not a blocked.  I just don't feel
> > > entirely comfortable about the current interface and I'm exploring to
> > > see if there might be something better.
> > > 
> > > 
> > 
> > Ok, I'm not convinced that anything you've proposed so far is better
> > than what we have, but I'm open-minded.
> 
> Thanks.  I admit that I don't think anything I've come up with is better
> either.
> 
> > 
> > At this point we have these to-dos:
> > 
> > 1) make the threads interface accept an array of integers rather than a
> > singleton. This is needed when sunrpc.pool_mode is not global.
> > 
> 
> While this I think probably still makes sense, I'm feeling less
> enthusiastic about it and probably wouldn't complain if it didn't
> happen.
> 

Ok. I think making it an array is fine. For now, it will just error out
when you send down more than one value, but I think we can make it work
the way you'd expect. Regardlness, cleaning up and unifying the pooled
vs. non-pooled server handling seems like a good thing to do. The
existing one is a long-standing kludge.

> I don't like that fact that we need to configure a number of threads.  I
> would much rather it grow/shrink dynamically.  I've got more patches
> that are heading in this direction but they aren't ready yet.
> 
> If we do land these and they prove to be effective, then configuring
> per-pool threads would become extremely uninteresting.  The demand on
> each pool would adjust each pool separately.
> 
> So if use an array here turns out to be problematic for some reason, I
> won't complain.
> 

+1 on making the thread pool sizing dynamic. I had started looking at
what heuristics we should use, but I haven't gotten very far yet.

Some general thoughts:

nfsd threads spend most of their time waiting. They are only very rarely
cpu-bound, so the max size of the thread pools should mostly scale with
the amount of memory in the host.

I have a patch that adds a new percpu counter to count how often we go
to wake a thread and don't find one, and have to queue it. It clearly
hits a lot more when there are fewer running nfsd threads.

I don't think that's sufficient to know when to spawn a new thread
though. For that, we probably ought to be looking at how long RPCs are
sitting and waiting for a thread to pick them up. When that starts
growing then we probably need to bring in more threads.

Conversely, when a thread is sitting idle for a long time (10s? 60s?),
we can probably spin it down. Maybe we can have a new LRU for the
threads and have a periodic thread reaper job that checks for those.


> > 2) have the interface pass down the scope string in the THREADS_SET
> > instead of making userland change the UTS namespace before starting
> > threads. This should just mean adding a new optional string attribute to
> > the threads interfaces.
> > 
> > ...anything else we're missing that needs doing here? We'd really like
> > to see this in -next soon, so we can possibly make v6.10 with this.
> 
> I haven't yet found any obvious gaps.  I agree that we can and should
> move forward promptly.
> 

Great. I think Lorenzo is planning to send another set soon that should
hopefully be reasonable for merge.

> 
> > 
> > Thanks,
> > -- 
> > Jeff Layton <jlayton@kernel.org>
> > 
>
diff mbox series

Patch

diff --git a/Documentation/netlink/specs/nfsd.yaml b/Documentation/netlink/specs/nfsd.yaml
index cbe6c5fd6c4d..0396e8b3ea1f 100644
--- a/Documentation/netlink/specs/nfsd.yaml
+++ b/Documentation/netlink/specs/nfsd.yaml
@@ -74,6 +74,26 @@  attribute-sets:
       -
         name: leasetime
         type: u32
+  -
+    name: version
+    attributes:
+      -
+        name: major
+        type: u32
+      -
+        name: minor
+        type: u32
+      -
+        name: enabled
+        type: flag
+  -
+    name: server-proto
+    attributes:
+      -
+        name: version
+        type: nest
+        nested-attributes: version
+        multi-attr: true
 
 operations:
   list:
@@ -120,3 +140,20 @@  operations:
             - threads
             - gracetime
             - leasetime
+    -
+      name: version-set
+      doc: set nfs enabled versions
+      attribute-set: server-proto
+      flags: [ admin-perm ]
+      do:
+        request:
+          attributes:
+            - version
+    -
+      name: version-get
+      doc: get nfs enabled versions
+      attribute-set: server-proto
+      do:
+        reply:
+          attributes:
+            - version
diff --git a/fs/nfsd/netlink.c b/fs/nfsd/netlink.c
index 20a646af0324..bf5df9597288 100644
--- a/fs/nfsd/netlink.c
+++ b/fs/nfsd/netlink.c
@@ -10,6 +10,13 @@ 
 
 #include <uapi/linux/nfsd_netlink.h>
 
+/* Common nested types */
+const struct nla_policy nfsd_version_nl_policy[NFSD_A_VERSION_ENABLED + 1] = {
+	[NFSD_A_VERSION_MAJOR] = { .type = NLA_U32, },
+	[NFSD_A_VERSION_MINOR] = { .type = NLA_U32, },
+	[NFSD_A_VERSION_ENABLED] = { .type = NLA_FLAG, },
+};
+
 /* NFSD_CMD_THREADS_SET - do */
 static const struct nla_policy nfsd_threads_set_nl_policy[NFSD_A_SERVER_WORKER_LEASETIME + 1] = {
 	[NFSD_A_SERVER_WORKER_THREADS] = { .type = NLA_U32, },
@@ -17,6 +24,11 @@  static const struct nla_policy nfsd_threads_set_nl_policy[NFSD_A_SERVER_WORKER_L
 	[NFSD_A_SERVER_WORKER_LEASETIME] = { .type = NLA_U32, },
 };
 
+/* NFSD_CMD_VERSION_SET - do */
+static const struct nla_policy nfsd_version_set_nl_policy[NFSD_A_SERVER_PROTO_VERSION + 1] = {
+	[NFSD_A_SERVER_PROTO_VERSION] = NLA_POLICY_NESTED(nfsd_version_nl_policy),
+};
+
 /* Ops table for nfsd */
 static const struct genl_split_ops nfsd_nl_ops[] = {
 	{
@@ -38,6 +50,18 @@  static const struct genl_split_ops nfsd_nl_ops[] = {
 		.doit	= nfsd_nl_threads_get_doit,
 		.flags	= GENL_CMD_CAP_DO,
 	},
+	{
+		.cmd		= NFSD_CMD_VERSION_SET,
+		.doit		= nfsd_nl_version_set_doit,
+		.policy		= nfsd_version_set_nl_policy,
+		.maxattr	= NFSD_A_SERVER_PROTO_VERSION,
+		.flags		= GENL_ADMIN_PERM | GENL_CMD_CAP_DO,
+	},
+	{
+		.cmd	= NFSD_CMD_VERSION_GET,
+		.doit	= nfsd_nl_version_get_doit,
+		.flags	= GENL_CMD_CAP_DO,
+	},
 };
 
 struct genl_family nfsd_nl_family __ro_after_init = {
diff --git a/fs/nfsd/netlink.h b/fs/nfsd/netlink.h
index 4137fac477e4..c7c0da275481 100644
--- a/fs/nfsd/netlink.h
+++ b/fs/nfsd/netlink.h
@@ -11,6 +11,9 @@ 
 
 #include <uapi/linux/nfsd_netlink.h>
 
+/* Common nested types */
+extern const struct nla_policy nfsd_version_nl_policy[NFSD_A_VERSION_ENABLED + 1];
+
 int nfsd_nl_rpc_status_get_start(struct netlink_callback *cb);
 int nfsd_nl_rpc_status_get_done(struct netlink_callback *cb);
 
@@ -18,6 +21,8 @@  int nfsd_nl_rpc_status_get_dumpit(struct sk_buff *skb,
 				  struct netlink_callback *cb);
 int nfsd_nl_threads_set_doit(struct sk_buff *skb, struct genl_info *info);
 int nfsd_nl_threads_get_doit(struct sk_buff *skb, struct genl_info *info);
+int nfsd_nl_version_set_doit(struct sk_buff *skb, struct genl_info *info);
+int nfsd_nl_version_get_doit(struct sk_buff *skb, struct genl_info *info);
 
 extern struct genl_family nfsd_nl_family;
 
diff --git a/fs/nfsd/netns.h b/fs/nfsd/netns.h
index d4be519b5734..14ec15656320 100644
--- a/fs/nfsd/netns.h
+++ b/fs/nfsd/netns.h
@@ -218,6 +218,7 @@  struct nfsd_net {
 /* Simple check to find out if a given net was properly initialized */
 #define nfsd_netns_ready(nn) ((nn)->sessionid_hashtbl)
 
+extern bool nfsd_support_version(int vers);
 extern void nfsd_netns_free_versions(struct nfsd_net *nn);
 
 extern unsigned int nfsd_net_id;
diff --git a/fs/nfsd/nfsctl.c b/fs/nfsd/nfsctl.c
index 38a5df03981b..2c8929ef79e9 100644
--- a/fs/nfsd/nfsctl.c
+++ b/fs/nfsd/nfsctl.c
@@ -1757,6 +1757,156 @@  int nfsd_nl_threads_get_doit(struct sk_buff *skb, struct genl_info *info)
 	return err;
 }
 
+/**
+ * nfsd_nl_version_set_doit - set the nfs enabled versions
+ * @skb: reply buffer
+ * @info: netlink metadata and command arguments
+ *
+ * Return 0 on success or a negative errno.
+ */
+int nfsd_nl_version_set_doit(struct sk_buff *skb, struct genl_info *info)
+{
+	const struct nlattr *attr;
+	struct nfsd_net *nn;
+	int i, rem;
+
+	if (GENL_REQ_ATTR_CHECK(info, NFSD_A_SERVER_PROTO_VERSION))
+		return -EINVAL;
+
+	mutex_lock(&nfsd_mutex);
+
+	nn = net_generic(genl_info_net(info), nfsd_net_id);
+	if (nn->nfsd_serv) {
+		mutex_unlock(&nfsd_mutex);
+		return -EBUSY;
+	}
+
+	/* clear current supported versions. */
+	nfsd_vers(nn, 2, NFSD_CLEAR);
+	nfsd_vers(nn, 3, NFSD_CLEAR);
+	for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++)
+		nfsd_minorversion(nn, i, NFSD_CLEAR);
+
+	nlmsg_for_each_attr(attr, info->nlhdr, GENL_HDRLEN, rem) {
+		struct nlattr *tb[NFSD_A_VERSION_MAX + 1];
+		u32 major, minor = 0;
+		bool enabled;
+
+		if (nla_type(attr) != NFSD_A_SERVER_PROTO_VERSION)
+			continue;
+
+		if (nla_parse_nested(tb, NFSD_A_VERSION_MAX, attr,
+				     nfsd_version_nl_policy, info->extack) < 0)
+			continue;
+
+		if (!tb[NFSD_A_VERSION_MAJOR])
+			continue;
+
+		major = nla_get_u32(tb[NFSD_A_VERSION_MAJOR]);
+		if (tb[NFSD_A_VERSION_MINOR])
+			minor = nla_get_u32(tb[NFSD_A_VERSION_MINOR]);
+
+		enabled = nla_get_flag(tb[NFSD_A_VERSION_ENABLED]);
+
+		switch (major) {
+		case 4:
+			nfsd_minorversion(nn, minor, enabled ? NFSD_SET : NFSD_CLEAR);
+			break;
+		case 3:
+		case 2:
+			if (!minor)
+				nfsd_vers(nn, major, enabled ? NFSD_SET : NFSD_CLEAR);
+			break;
+		default:
+			break;
+		}
+	}
+
+	mutex_unlock(&nfsd_mutex);
+
+	return 0;
+}
+
+/**
+ * nfsd_nl_version_get_doit - get the nfs enabled versions
+ * @skb: reply buffer
+ * @info: netlink metadata and command arguments
+ *
+ * Return 0 on success or a negative errno.
+ */
+int nfsd_nl_version_get_doit(struct sk_buff *skb, struct genl_info *info)
+{
+	struct nfsd_net *nn;
+	int i, err;
+	void *hdr;
+
+	skb = genlmsg_new(GENLMSG_DEFAULT_SIZE, GFP_KERNEL);
+	if (!skb)
+		return -ENOMEM;
+
+	hdr = genlmsg_iput(skb, info);
+	if (!hdr) {
+		err = -EMSGSIZE;
+		goto err_free_msg;
+	}
+
+	mutex_lock(&nfsd_mutex);
+	nn = net_generic(genl_info_net(info), nfsd_net_id);
+
+	for (i = 2; i <= 4; i++) {
+		int j;
+
+		for (j = 0; j <= NFSD_SUPPORTED_MINOR_VERSION; j++) {
+			struct nlattr *attr;
+
+			/* Don't record any versions the kernel doesn't have
+			 * compiled in
+			 */
+			if (!nfsd_support_version(i))
+				continue;
+
+			/* NFSv{2,3} does not support minor numbers */
+			if (i < 4 && j)
+				continue;
+
+			attr = nla_nest_start(skb,
+					      NFSD_A_SERVER_PROTO_VERSION);
+			if (!attr) {
+				err = -EINVAL;
+				goto err_nfsd_unlock;
+			}
+
+			if (nla_put_u32(skb, NFSD_A_VERSION_MAJOR, i) ||
+			    nla_put_u32(skb, NFSD_A_VERSION_MINOR, j)) {
+				err = -EINVAL;
+				goto err_nfsd_unlock;
+			}
+
+			/* Set the enabled flag if the version is enabled */
+			if (nfsd_vers(nn, i, NFSD_TEST) &&
+			    (i < 4 || nfsd_minorversion(nn, j, NFSD_TEST)) &&
+			    nla_put_flag(skb, NFSD_A_VERSION_ENABLED)) {
+				err = -EINVAL;
+				goto err_nfsd_unlock;
+			}
+
+			nla_nest_end(skb, attr);
+		}
+	}
+
+	mutex_unlock(&nfsd_mutex);
+	genlmsg_end(skb, hdr);
+
+	return genlmsg_reply(skb, info);
+
+err_nfsd_unlock:
+	mutex_unlock(&nfsd_mutex);
+err_free_msg:
+	nlmsg_free(skb);
+
+	return err;
+}
+
 /**
  * nfsd_net_init - Prepare the nfsd_net portion of a new net namespace
  * @net: a freshly-created network namespace
diff --git a/fs/nfsd/nfssvc.c b/fs/nfsd/nfssvc.c
index ca193f7ff0e1..4fc91f50138a 100644
--- a/fs/nfsd/nfssvc.c
+++ b/fs/nfsd/nfssvc.c
@@ -133,8 +133,7 @@  struct svc_program		nfsd_program = {
 	.pg_rpcbind_set		= nfsd_rpcbind_set,
 };
 
-static bool
-nfsd_support_version(int vers)
+bool nfsd_support_version(int vers)
 {
 	if (vers >= NFSD_MINVERS && vers < NFSD_NRVERS)
 		return nfsd_version[vers] != NULL;
diff --git a/include/uapi/linux/nfsd_netlink.h b/include/uapi/linux/nfsd_netlink.h
index ccc78a5ee650..8a0a2b344923 100644
--- a/include/uapi/linux/nfsd_netlink.h
+++ b/include/uapi/linux/nfsd_netlink.h
@@ -38,10 +38,28 @@  enum {
 	NFSD_A_SERVER_WORKER_MAX = (__NFSD_A_SERVER_WORKER_MAX - 1)
 };
 
+enum {
+	NFSD_A_VERSION_MAJOR = 1,
+	NFSD_A_VERSION_MINOR,
+	NFSD_A_VERSION_ENABLED,
+
+	__NFSD_A_VERSION_MAX,
+	NFSD_A_VERSION_MAX = (__NFSD_A_VERSION_MAX - 1)
+};
+
+enum {
+	NFSD_A_SERVER_PROTO_VERSION = 1,
+
+	__NFSD_A_SERVER_PROTO_MAX,
+	NFSD_A_SERVER_PROTO_MAX = (__NFSD_A_SERVER_PROTO_MAX - 1)
+};
+
 enum {
 	NFSD_CMD_RPC_STATUS_GET = 1,
 	NFSD_CMD_THREADS_SET,
 	NFSD_CMD_THREADS_GET,
+	NFSD_CMD_VERSION_SET,
+	NFSD_CMD_VERSION_GET,
 
 	__NFSD_CMD_MAX,
 	NFSD_CMD_MAX = (__NFSD_CMD_MAX - 1)