cifs: fix return code when failing to rename a file onto a directory

Cifs servers return ACCESS_DENIED when trying to rename onto a non-empty
directory. This is different from xfstest where we expect this to return
-EEXIST instead.

As a workaround, if we failed to rename the file and we got -EACCES
then try to open the target file and check the attributes if it is a
directory or not and if so remap the error to -EEXIST.

This makes us pass xfstest  generic/245

Signed-off-by: Ronnie Sahlberg <lsahlber@redhat.com>
---
 fs/cifs/smb2ops.c   | 40 ++++++++++++++++++++++++++++++++++++++++
 fs/cifs/smb2pdu.c   | 12 +++++++++++-
 fs/cifs/smb2proto.h |  2 ++
 3 files changed, 53 insertions(+), 1 deletion(-)

2017-11-09 15:52 GMT-08:00 Ronnie Sahlberg <lsahlber@redhat.com>:
> Cifs servers return ACCESS_DENIED when trying to rename onto a non-empty
> directory. This is different from xfstest where we expect this to return
> -EEXIST instead.
>
> As a workaround, if we failed to rename the file and we got -EACCES
> then try to open the target file and check the attributes if it is a
> directory or not and if so remap the error to -EEXIST.
>
> This makes us pass xfstest  generic/245
>
> Signed-off-by: Ronnie Sahlberg <lsahlber@redhat.com>
> ---
>  fs/cifs/smb2ops.c   | 40 ++++++++++++++++++++++++++++++++++++++++
>  fs/cifs/smb2pdu.c   | 12 +++++++++++-
>  fs/cifs/smb2proto.h |  2 ++
>  3 files changed, 53 insertions(+), 1 deletion(-)
>
> diff --git a/fs/cifs/smb2ops.c b/fs/cifs/smb2ops.c
> index bdb963d0ba32..5a620c71d91f 100644
> --- a/fs/cifs/smb2ops.c
> +++ b/fs/cifs/smb2ops.c
> @@ -426,6 +426,46 @@ smb2_query_file_info(const unsigned int xid, struct cifs_tcon *tcon,
>         return rc;
>  }
>
> +int
> +smb2_is_dir(const unsigned int xid, struct cifs_tcon *tcon,
> +           __le16 *target_file, int *is_dir)
> +{
> +       int rc;
> +       u8 oplock = SMB2_OPLOCK_LEVEL_NONE;
> +       struct cifs_open_parms oparms;
> +       struct cifs_fid fid;
> +       struct smb2_file_all_info *smb2_data;
> +
> +       smb2_data = kzalloc(sizeof(struct smb2_file_all_info) + PATH_MAX * 2,
> +                           GFP_KERNEL);
> +       if (smb2_data == NULL)
> +               return -ENOMEM;
> +
> +       oparms.tcon = tcon;
> +       oparms.desired_access = FILE_READ_ATTRIBUTES;
> +       oparms.disposition = FILE_OPEN;
> +       oparms.create_options = 0;
> +       oparms.fid = &fid;
> +       oparms.reconnect = false;
> +
> +       rc = SMB2_open(xid, &oparms, target_file, &oplock, NULL, NULL);
> +       if (rc)
> +               goto out;
> +
> +       rc = SMB2_query_info(xid, tcon, fid.persistent_fid, fid.volatile_fid,
> +                            smb2_data);

Can't we use FILE_DIRECTORY_FILE flag in CreateOptions of SMB2_CREATE
command and avoid querying attributes from the server? This will save
us 1 or 2 round trips.


> +       SMB2_close(xid, tcon, fid.persistent_fid, fid.volatile_fid);
> +       if (rc)
> +               goto out;
> +
> +       *is_dir = !!(le32_to_cpu(smb2_data->Attributes) &
> +                    FILE_ATTRIBUTE_DIRECTORY);
> +
> +out:
> +       kfree(smb2_data);
> +       return rc;
> +}
> +
>  #ifdef CONFIG_CIFS_XATTR
>  static ssize_t
>  move_smb2_ea_to_cifs(char *dst, size_t dst_size,
> diff --git a/fs/cifs/smb2pdu.c b/fs/cifs/smb2pdu.c
> index 553d574940b9..ec5de0a23378 100644
> --- a/fs/cifs/smb2pdu.c
> +++ b/fs/cifs/smb2pdu.c
> @@ -3144,7 +3144,7 @@ SMB2_rename(const unsigned int xid, struct cifs_tcon *tcon,
>         struct smb2_file_rename_info info;
>         void **data;
>         unsigned int size[2];
> -       int rc;
> +       int rc, is_dir;
>         int len = (2 * UniStrnlen((wchar_t *)target_file, PATH_MAX));
>
>         data = kmalloc(sizeof(void *) * 2, GFP_KERNEL);
> @@ -3165,6 +3165,16 @@ SMB2_rename(const unsigned int xid, struct cifs_tcon *tcon,
>         rc = send_set_info(xid, tcon, persistent_fid, volatile_fid,
>                 current->tgid, FILE_RENAME_INFORMATION, SMB2_O_INFO_FILE,
>                 0, 2, data, size);
> +
> +       /* SMB2 servers responds with ACCESS_DENIED when trying to rename
> +        * and replace onto a non-empty directory. Check for this and remap
> +        * to EEXIST.
> +        */
> +       if (rc == -EACCES) {
> +               if (!smb2_is_dir(xid, tcon, target_file, &is_dir) && is_dir)
> +                       rc = -EEXIST;
> +       }

What if the target is directory but the permissions are read-only?

> +
>         kfree(data);
>         return rc;
>  }
> diff --git a/fs/cifs/smb2proto.h b/fs/cifs/smb2proto.h
> index e9ab5227e7a8..a7528c0941c6 100644
> --- a/fs/cifs/smb2proto.h
> +++ b/fs/cifs/smb2proto.h
> @@ -203,4 +203,6 @@ extern int smb3_validate_negotiate(const unsigned int, struct cifs_tcon *);
>
>  extern enum securityEnum smb2_select_sectype(struct TCP_Server_Info *,
>                                         enum securityEnum);
> +extern int smb2_is_dir(const unsigned int xid, struct cifs_tcon *tcon,
> +                      __le16 *target_file, int *is_dir);
>  #endif                 /* _SMB2PROTO_H */
> --
> 2.13.3
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-cifs" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html


--
Best regards,
Pavel Shilovsky
--
To unsubscribe from this list: send the line "unsubscribe linux-cifs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

On Thu, Nov 16, 2017 at 5:31 PM, Pavel Shilovsky <piastryyy@gmail.com> wrote:
> 2017-11-09 15:52 GMT-08:00 Ronnie Sahlberg <lsahlber@redhat.com>:
>> Cifs servers return ACCESS_DENIED when trying to rename onto a non-empty
>> directory. This is different from xfstest where we expect this to return
>> -EEXIST instead.
>>
>> As a workaround, if we failed to rename the file and we got -EACCES
>> then try to open the target file and check the attributes if it is a
>> directory or not and if so remap the error to -EEXIST.
>>
>> This makes us pass xfstest  generic/245
>>
>> Signed-off-by: Ronnie Sahlberg <lsahlber@redhat.com>
>> ---
>>  fs/cifs/smb2ops.c   | 40 ++++++++++++++++++++++++++++++++++++++++
>>  fs/cifs/smb2pdu.c   | 12 +++++++++++-
>>  fs/cifs/smb2proto.h |  2 ++
>>  3 files changed, 53 insertions(+), 1 deletion(-)
>>
>> diff --git a/fs/cifs/smb2ops.c b/fs/cifs/smb2ops.c
>> index bdb963d0ba32..5a620c71d91f 100644
>> --- a/fs/cifs/smb2ops.c
>> +++ b/fs/cifs/smb2ops.c
>> @@ -426,6 +426,46 @@ smb2_query_file_info(const unsigned int xid, struct cifs_tcon *tcon,
>>         return rc;
>>  }
>>
>> +int
>> +smb2_is_dir(const unsigned int xid, struct cifs_tcon *tcon,
>> +           __le16 *target_file, int *is_dir)
>> +{
>> +       int rc;
>> +       u8 oplock = SMB2_OPLOCK_LEVEL_NONE;
>> +       struct cifs_open_parms oparms;
>> +       struct cifs_fid fid;
>> +       struct smb2_file_all_info *smb2_data;
>> +
>> +       smb2_data = kzalloc(sizeof(struct smb2_file_all_info) + PATH_MAX * 2,
>> +                           GFP_KERNEL);
>> +       if (smb2_data == NULL)
>> +               return -ENOMEM;
>> +
>> +       oparms.tcon = tcon;
>> +       oparms.desired_access = FILE_READ_ATTRIBUTES;
>> +       oparms.disposition = FILE_OPEN;
>> +       oparms.create_options = 0;
>> +       oparms.fid = &fid;
>> +       oparms.reconnect = false;
>> +
>> +       rc = SMB2_open(xid, &oparms, target_file, &oplock, NULL, NULL);
>> +       if (rc)
>> +               goto out;
>> +
>> +       rc = SMB2_query_info(xid, tcon, fid.persistent_fid, fid.volatile_fid,
>> +                            smb2_data);
>
> Can't we use FILE_DIRECTORY_FILE flag in CreateOptions of SMB2_CREATE
> command and avoid querying attributes from the server? This will save
> us 1 or 2 round trips.
>
>
>> +       SMB2_close(xid, tcon, fid.persistent_fid, fid.volatile_fid);
>> +       if (rc)
>> +               goto out;
>> +
>> +       *is_dir = !!(le32_to_cpu(smb2_data->Attributes) &
>> +                    FILE_ATTRIBUTE_DIRECTORY);
>> +
>> +out:
>> +       kfree(smb2_data);
>> +       return rc;
>> +}
>> +
>>  #ifdef CONFIG_CIFS_XATTR
>>  static ssize_t
>>  move_smb2_ea_to_cifs(char *dst, size_t dst_size,
>> diff --git a/fs/cifs/smb2pdu.c b/fs/cifs/smb2pdu.c
>> index 553d574940b9..ec5de0a23378 100644
>> --- a/fs/cifs/smb2pdu.c
>> +++ b/fs/cifs/smb2pdu.c
>> @@ -3144,7 +3144,7 @@ SMB2_rename(const unsigned int xid, struct cifs_tcon *tcon,
>>         struct smb2_file_rename_info info;
>>         void **data;
>>         unsigned int size[2];
>> -       int rc;
>> +       int rc, is_dir;
>>         int len = (2 * UniStrnlen((wchar_t *)target_file, PATH_MAX));
>>
>>         data = kmalloc(sizeof(void *) * 2, GFP_KERNEL);
>> @@ -3165,6 +3165,16 @@ SMB2_rename(const unsigned int xid, struct cifs_tcon *tcon,
>>         rc = send_set_info(xid, tcon, persistent_fid, volatile_fid,
>>                 current->tgid, FILE_RENAME_INFORMATION, SMB2_O_INFO_FILE,
>>                 0, 2, data, size);
>> +
>> +       /* SMB2 servers responds with ACCESS_DENIED when trying to rename
>> +        * and replace onto a non-empty directory. Check for this and remap
>> +        * to EEXIST.
>> +        */
>> +       if (rc == -EACCES) {
>> +               if (!smb2_is_dir(xid, tcon, target_file, &is_dir) && is_dir)
>> +                       rc = -EEXIST;
>> +       }
>
> What if the target is directory but the permissions are read-only?
>
>> +
>>         kfree(data);
>>         return rc;
>>  }
>> diff --git a/fs/cifs/smb2proto.h b/fs/cifs/smb2proto.h
>> index e9ab5227e7a8..a7528c0941c6 100644
>> --- a/fs/cifs/smb2proto.h
>> +++ b/fs/cifs/smb2proto.h
>> @@ -203,4 +203,6 @@ extern int smb3_validate_negotiate(const unsigned int, struct cifs_tcon *);
>>
>>  extern enum securityEnum smb2_select_sectype(struct TCP_Server_Info *,
>>                                         enum securityEnum);
>> +extern int smb2_is_dir(const unsigned int xid, struct cifs_tcon *tcon,
>> +                      __le16 *target_file, int *is_dir);
>>  #endif                 /* _SMB2PROTO_H */
>> --
>> 2.13.3
>>

Pavel makes good points ...

Pavel Shilovsky <piastryyy@gmail.com> writes:
>
> What if the target is directory but the permissions are read-only?

I get the same resulsts on a local filesystem (xfs) and on a mounted
smb2 windows2016 share (both with and without the patch applied):

overwrite dir b with file a (with access to b)
=======================
rename("test/a", "test/b")              = -1 EISDIR (Is a directory)

overwrite dir b with file a (without access to b)
=======================
rename("test/a", "test/b")              = -1 EISDIR (Is a directory)

overwrite file b with file a (with access to b)
=======================
rename("test/a", "test/b")              = 0

overwrite file b with file a (without access to b)
=======================
rename("test/a", "test/b")              = 0

So.. I'm not sure what the xfstest test is doing.

The only difference between the local and remote fs is that for cifs you
can't delete test if you can't access test/b. (I am deleting "test"
between my tests).

That is on cifs:

  mkdir -p test/b
  chmod a-rw test/b
  rm -rf test

Will fail, but works on xfs. You you strace rm you will see:

  unlinkat(4, "b", AT_REMOVEDIR) = -1 EACCES (Permission denied)

I think this is due to the fundamental difference in how file removal
permission works in Windows vs POSIX. In POSIX you need +w on the parent
dir ("test") to be able to delete a file. On Windows there is a special
delete permission on the object itself. Not sure if this is "fixable".

Cheers,