2018-11-30 07:49:58

by Zheng Bin

[permalink] [raw]
Subject: [PATCH] nfsd: Return EPERM on utime/utimes/lutimes calls instead of EACCESS in some cases

As the man(3) page for utime/utimes/lutimes, EPERM is returned
when the second parameter of utime/utimes/lutimes is not NULL,
the caller's effective UID does not match the owner of the file,
and the caller is not privileged.

However, in a NFS directory, it will return EACCESS(nfsd_setattr->
fh_verify->nfsd_permission), This patch fix this.

Signed-off-by: zhengbin <[email protected]>
---
fs/nfsd/vfs.c | 17 +++++++++++++++--
1 file changed, 15 insertions(+), 2 deletions(-)

diff --git a/fs/nfsd/vfs.c b/fs/nfsd/vfs.c
index eb67098..9824e32 100644
--- a/fs/nfsd/vfs.c
+++ b/fs/nfsd/vfs.c
@@ -396,10 +396,23 @@ nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
bool get_write_count;
bool size_change = (iap->ia_valid & ATTR_SIZE);

- if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
+ if (iap->ia_valid & ATTR_SIZE) {
accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
- if (iap->ia_valid & ATTR_SIZE)
ftype = S_IFREG;
+ }
+
+ /*
+ * If utimes(2) and friends are called with times not NULL, we should
+ * not set NFSD_MAY_WRITE bit. Otherwise fh_verify->nfsd_permission
+ * will return EACCESS, when the caller's effective UID does not match
+ * the owner of the file, and the caller is not privileged. In this
+ * situation, we should return EPERM(notify_change will return this).
+ */
+ if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME)) {
+ accmode |= NFSD_MAY_OWNER_OVERRIDE;
+ if (!(iap->ia_valid & (ATTR_ATIME_SET | ATTR_MTIME_SET)))
+ accmode |= NFSD_MAY_WRITE;
+ }

/* Callers that do fh_verify should do the fh_want_write: */
get_write_count = !fhp->fh_dentry;
--
2.7.4



2018-12-03 08:36:40

by Zheng Bin

[permalink] [raw]
Subject: Re: [PATCH] nfsd: Return EPERM on utime/utimes/lutimes calls instead of EACCESS in some cases

hi, J. Bruce Fields, Jeff Layton

When you have time, confirm it? thanks

On 2018/11/30 15:51, zhengbin wrote:
> As the man(3) page for utime/utimes/lutimes, EPERM is returned
> when the second parameter of utime/utimes/lutimes is not NULL,
> the caller's effective UID does not match the owner of the file,
> and the caller is not privileged.
>
> However, in a NFS directory, it will return EACCESS(nfsd_setattr->
> fh_verify->nfsd_permission), This patch fix this.
>
> Signed-off-by: zhengbin <[email protected]>
> ---
> fs/nfsd/vfs.c | 17 +++++++++++++++--
> 1 file changed, 15 insertions(+), 2 deletions(-)
>
> diff --git a/fs/nfsd/vfs.c b/fs/nfsd/vfs.c
> index eb67098..9824e32 100644
> --- a/fs/nfsd/vfs.c
> +++ b/fs/nfsd/vfs.c
> @@ -396,10 +396,23 @@ nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
> bool get_write_count;
> bool size_change = (iap->ia_valid & ATTR_SIZE);
>
> - if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
> + if (iap->ia_valid & ATTR_SIZE) {
> accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
> - if (iap->ia_valid & ATTR_SIZE)
> ftype = S_IFREG;
> + }
> +
> + /*
> + * If utimes(2) and friends are called with times not NULL, we should
> + * not set NFSD_MAY_WRITE bit. Otherwise fh_verify->nfsd_permission
> + * will return EACCESS, when the caller's effective UID does not match
> + * the owner of the file, and the caller is not privileged. In this
> + * situation, we should return EPERM(notify_change will return this).
> + */
> + if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME)) {
> + accmode |= NFSD_MAY_OWNER_OVERRIDE;
> + if (!(iap->ia_valid & (ATTR_ATIME_SET | ATTR_MTIME_SET)))
> + accmode |= NFSD_MAY_WRITE;
> + }
>
> /* Callers that do fh_verify should do the fh_want_write: */
> get_write_count = !fhp->fh_dentry;
> --
> 2.7.4
>
>
> .
>


2018-12-04 20:13:59

by J. Bruce Fields

[permalink] [raw]
Subject: Re: [PATCH] nfsd: Return EPERM on utime/utimes/lutimes calls instead of EACCESS in some cases

On Mon, Dec 03, 2018 at 04:36:27PM +0800, zhengbin (A) wrote:
> hi, J. Bruce Fields, Jeff Layton
>
> When you have time, confirm it? thanks

Sorry for the delay.

The patch looks good to me.

As far as I can tell, we've done this forever. And I'm a little worried
about changing such long-standing behavior, but I think you're right.

So, applying for for 4.21.

--b.

> On 2018/11/30 15:51, zhengbin wrote:
> > As the man(3) page for utime/utimes/lutimes, EPERM is returned
> > when the second parameter of utime/utimes/lutimes is not NULL,
> > the caller's effective UID does not match the owner of the file,
> > and the caller is not privileged.
> >
> > However, in a NFS directory, it will return EACCESS(nfsd_setattr->
> > fh_verify->nfsd_permission), This patch fix this.
> >
> > Signed-off-by: zhengbin <[email protected]>
> > ---
> > fs/nfsd/vfs.c | 17 +++++++++++++++--
> > 1 file changed, 15 insertions(+), 2 deletions(-)
> >
> > diff --git a/fs/nfsd/vfs.c b/fs/nfsd/vfs.c
> > index eb67098..9824e32 100644
> > --- a/fs/nfsd/vfs.c
> > +++ b/fs/nfsd/vfs.c
> > @@ -396,10 +396,23 @@ nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
> > bool get_write_count;
> > bool size_change = (iap->ia_valid & ATTR_SIZE);
> >
> > - if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
> > + if (iap->ia_valid & ATTR_SIZE) {
> > accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
> > - if (iap->ia_valid & ATTR_SIZE)
> > ftype = S_IFREG;
> > + }
> > +
> > + /*
> > + * If utimes(2) and friends are called with times not NULL, we should
> > + * not set NFSD_MAY_WRITE bit. Otherwise fh_verify->nfsd_permission
> > + * will return EACCESS, when the caller's effective UID does not match
> > + * the owner of the file, and the caller is not privileged. In this
> > + * situation, we should return EPERM(notify_change will return this).
> > + */
> > + if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME)) {
> > + accmode |= NFSD_MAY_OWNER_OVERRIDE;
> > + if (!(iap->ia_valid & (ATTR_ATIME_SET | ATTR_MTIME_SET)))
> > + accmode |= NFSD_MAY_WRITE;
> > + }
> >
> > /* Callers that do fh_verify should do the fh_want_write: */
> > get_write_count = !fhp->fh_dentry;
> > --
> > 2.7.4
> >
> >
> > .
> >

2018-12-05 01:04:35

by Zheng Bin

[permalink] [raw]
Subject: Re: [PATCH] nfsd: Return EPERM on utime/utimes/lutimes calls instead of EACCESS in some cases

thanks,
>>> As the man(3) page for utime/utimes/lutimes, EPERM is returned

-------->this should be As the man page for utime/utimes/lutimes, my little mistake. utime/utimes are man(2), lutimes is man(3)

I send v2 patch?Or when you apply, help modify it?

On 2018/12/5 4:13, J. Bruce Fields wrote:
> On Mon, Dec 03, 2018 at 04:36:27PM +0800, zhengbin (A) wrote:
>> hi, J. Bruce Fields, Jeff Layton
>>
>> When you have time, confirm it? thanks
>
> Sorry for the delay.
>
> The patch looks good to me.
>
> As far as I can tell, we've done this forever. And I'm a little worried
> about changing such long-standing behavior, but I think you're right.
>
> So, applying for for 4.21.
>
> --b.
>
>> On 2018/11/30 15:51, zhengbin wrote:
>>> As the man(3) page for utime/utimes/lutimes, EPERM is returned
>>> when the second parameter of utime/utimes/lutimes is not NULL,
>>> the caller's effective UID does not match the owner of the file,
>>> and the caller is not privileged.
>>>
>>> However, in a NFS directory, it will return EACCESS(nfsd_setattr->
>>> fh_verify->nfsd_permission), This patch fix this.
>>>
>>> Signed-off-by: zhengbin <[email protected]>
>>> ---
>>> fs/nfsd/vfs.c | 17 +++++++++++++++--
>>> 1 file changed, 15 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/fs/nfsd/vfs.c b/fs/nfsd/vfs.c
>>> index eb67098..9824e32 100644
>>> --- a/fs/nfsd/vfs.c
>>> +++ b/fs/nfsd/vfs.c
>>> @@ -396,10 +396,23 @@ nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
>>> bool get_write_count;
>>> bool size_change = (iap->ia_valid & ATTR_SIZE);
>>>
>>> - if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
>>> + if (iap->ia_valid & ATTR_SIZE) {
>>> accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
>>> - if (iap->ia_valid & ATTR_SIZE)
>>> ftype = S_IFREG;
>>> + }
>>> +
>>> + /*
>>> + * If utimes(2) and friends are called with times not NULL, we should
>>> + * not set NFSD_MAY_WRITE bit. Otherwise fh_verify->nfsd_permission
>>> + * will return EACCESS, when the caller's effective UID does not match
>>> + * the owner of the file, and the caller is not privileged. In this
>>> + * situation, we should return EPERM(notify_change will return this).
>>> + */
>>> + if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME)) {
>>> + accmode |= NFSD_MAY_OWNER_OVERRIDE;
>>> + if (!(iap->ia_valid & (ATTR_ATIME_SET | ATTR_MTIME_SET)))
>>> + accmode |= NFSD_MAY_WRITE;
>>> + }
>>>
>>> /* Callers that do fh_verify should do the fh_want_write: */
>>> get_write_count = !fhp->fh_dentry;
>>> --
>>> 2.7.4
>>>
>>>
>>> .
>>>
>
> .
>


2018-12-05 01:37:18

by J. Bruce Fields

[permalink] [raw]
Subject: Re: [PATCH] nfsd: Return EPERM on utime/utimes/lutimes calls instead of EACCESS in some cases

On Wed, Dec 05, 2018 at 09:04:22AM +0800, zhengbin (A) wrote:
> thanks,
> >>> As the man(3) page for utime/utimes/lutimes, EPERM is returned
>
> -------->this should be As the man page for utime/utimes/lutimes, my little mistake. utime/utimes are man(2), lutimes is man(3)
>
> I send v2 patch?Or when you apply, help modify it?

Does this look OK?

--b.

commit eb6d67589750
Author: zhengbin <[email protected]>
Date: Fri Nov 30 16:04:25 2018 +0800

nfsd: Return EPERM, not EACCES, in some SETATTR cases

As the man(2) page for utime/utimes states, EPERM is returned when the
second parameter of utime/utimes/lutimes is not NULL, the caller's
effective UID does not match the owner of the file, and the caller is
not privileged.

However, in a NFS directory mounted from knfsd, it will return EACCES
(from nfsd_setattr-> fh_verify->nfsd_permission). This patch fixes
that.

Signed-off-by: zhengbin <[email protected]>
Signed-off-by: J. Bruce Fields <[email protected]>

diff --git a/fs/nfsd/vfs.c b/fs/nfsd/vfs.c
index eb67098117b4..9824e32b2f23 100644
--- a/fs/nfsd/vfs.c
+++ b/fs/nfsd/vfs.c
@@ -396,10 +396,23 @@ nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
bool get_write_count;
bool size_change = (iap->ia_valid & ATTR_SIZE);

- if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
+ if (iap->ia_valid & ATTR_SIZE) {
accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
- if (iap->ia_valid & ATTR_SIZE)
ftype = S_IFREG;
+ }
+
+ /*
+ * If utimes(2) and friends are called with times not NULL, we should
+ * not set NFSD_MAY_WRITE bit. Otherwise fh_verify->nfsd_permission
+ * will return EACCESS, when the caller's effective UID does not match
+ * the owner of the file, and the caller is not privileged. In this
+ * situation, we should return EPERM(notify_change will return this).
+ */
+ if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME)) {
+ accmode |= NFSD_MAY_OWNER_OVERRIDE;
+ if (!(iap->ia_valid & (ATTR_ATIME_SET | ATTR_MTIME_SET)))
+ accmode |= NFSD_MAY_WRITE;
+ }

/* Callers that do fh_verify should do the fh_want_write: */
get_write_count = !fhp->fh_dentry;

2018-12-05 01:43:20

by Zheng Bin

[permalink] [raw]
Subject: Re: [PATCH] nfsd: Return EPERM on utime/utimes/lutimes calls instead of EACCESS in some cases

second parameter of utime/utimes/lutimes is not NULL -------> maybe lutimes should be removed? I think "second parameter of utime/utimes is not NULL" would be better

On 2018/12/5 9:37, J. Bruce Fields wrote:
> On Wed, Dec 05, 2018 at 09:04:22AM +0800, zhengbin (A) wrote:
>> thanks,
>>>>> As the man(3) page for utime/utimes/lutimes, EPERM is returned
>>
>> -------->this should be As the man page for utime/utimes/lutimes, my little mistake. utime/utimes are man(2), lutimes is man(3)
>>
>> I send v2 patch?Or when you apply, help modify it?
>
> Does this look OK?
>
> --b.
>
> commit eb6d67589750
> Author: zhengbin <[email protected]>
> Date: Fri Nov 30 16:04:25 2018 +0800
>
> nfsd: Return EPERM, not EACCES, in some SETATTR cases
>
> As the man(2) page for utime/utimes states, EPERM is returned when the
> second parameter of utime/utimes/lutimes is not NULL, the caller's
> effective UID does not match the owner of the file, and the caller is
> not privileged.
>
> However, in a NFS directory mounted from knfsd, it will return EACCES
> (from nfsd_setattr-> fh_verify->nfsd_permission). This patch fixes
> that.
>
> Signed-off-by: zhengbin <[email protected]>
> Signed-off-by: J. Bruce Fields <[email protected]>
>
> diff --git a/fs/nfsd/vfs.c b/fs/nfsd/vfs.c
> index eb67098117b4..9824e32b2f23 100644
> --- a/fs/nfsd/vfs.c
> +++ b/fs/nfsd/vfs.c
> @@ -396,10 +396,23 @@ nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
> bool get_write_count;
> bool size_change = (iap->ia_valid & ATTR_SIZE);
>
> - if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
> + if (iap->ia_valid & ATTR_SIZE) {
> accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
> - if (iap->ia_valid & ATTR_SIZE)
> ftype = S_IFREG;
> + }
> +
> + /*
> + * If utimes(2) and friends are called with times not NULL, we should
> + * not set NFSD_MAY_WRITE bit. Otherwise fh_verify->nfsd_permission
> + * will return EACCESS, when the caller's effective UID does not match
> + * the owner of the file, and the caller is not privileged. In this
> + * situation, we should return EPERM(notify_change will return this).
> + */
> + if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME)) {
> + accmode |= NFSD_MAY_OWNER_OVERRIDE;
> + if (!(iap->ia_valid & (ATTR_ATIME_SET | ATTR_MTIME_SET)))
> + accmode |= NFSD_MAY_WRITE;
> + }
>
> /* Callers that do fh_verify should do the fh_want_write: */
> get_write_count = !fhp->fh_dentry;
>
> .
>


2018-12-05 01:49:02

by J. Bruce Fields

[permalink] [raw]
Subject: Re: [PATCH] nfsd: Return EPERM on utime/utimes/lutimes calls instead of EACCESS in some cases

On Wed, Dec 05, 2018 at 09:43:06AM +0800, zhengbin (A) wrote:
> second parameter of utime/utimes/lutimes is not NULL -------> maybe lutimes should be removed? I think "second parameter of utime/utimes is not NULL" would be better

Also fixed, thanks.--b.

>
> On 2018/12/5 9:37, J. Bruce Fields wrote:
> > On Wed, Dec 05, 2018 at 09:04:22AM +0800, zhengbin (A) wrote:
> >> thanks,
> >>>>> As the man(3) page for utime/utimes/lutimes, EPERM is returned
> >>
> >> -------->this should be As the man page for utime/utimes/lutimes, my little mistake. utime/utimes are man(2), lutimes is man(3)
> >>
> >> I send v2 patch?Or when you apply, help modify it?
> >
> > Does this look OK?
> >
> > --b.
> >
> > commit eb6d67589750
> > Author: zhengbin <[email protected]>
> > Date: Fri Nov 30 16:04:25 2018 +0800
> >
> > nfsd: Return EPERM, not EACCES, in some SETATTR cases
> >
> > As the man(2) page for utime/utimes states, EPERM is returned when the
> > second parameter of utime/utimes/lutimes is not NULL, the caller's
> > effective UID does not match the owner of the file, and the caller is
> > not privileged.
> >
> > However, in a NFS directory mounted from knfsd, it will return EACCES
> > (from nfsd_setattr-> fh_verify->nfsd_permission). This patch fixes
> > that.
> >
> > Signed-off-by: zhengbin <[email protected]>
> > Signed-off-by: J. Bruce Fields <[email protected]>
> >
> > diff --git a/fs/nfsd/vfs.c b/fs/nfsd/vfs.c
> > index eb67098117b4..9824e32b2f23 100644
> > --- a/fs/nfsd/vfs.c
> > +++ b/fs/nfsd/vfs.c
> > @@ -396,10 +396,23 @@ nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
> > bool get_write_count;
> > bool size_change = (iap->ia_valid & ATTR_SIZE);
> >
> > - if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
> > + if (iap->ia_valid & ATTR_SIZE) {
> > accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
> > - if (iap->ia_valid & ATTR_SIZE)
> > ftype = S_IFREG;
> > + }
> > +
> > + /*
> > + * If utimes(2) and friends are called with times not NULL, we should
> > + * not set NFSD_MAY_WRITE bit. Otherwise fh_verify->nfsd_permission
> > + * will return EACCESS, when the caller's effective UID does not match
> > + * the owner of the file, and the caller is not privileged. In this
> > + * situation, we should return EPERM(notify_change will return this).
> > + */
> > + if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME)) {
> > + accmode |= NFSD_MAY_OWNER_OVERRIDE;
> > + if (!(iap->ia_valid & (ATTR_ATIME_SET | ATTR_MTIME_SET)))
> > + accmode |= NFSD_MAY_WRITE;
> > + }
> >
> > /* Callers that do fh_verify should do the fh_want_write: */
> > get_write_count = !fhp->fh_dentry;
> >
> > .
> >