[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] [edk2-devel] [PATCH 05/11] OvmfPkg/XenBusDxe: Construct paths without allocation



On 09/13/19 16:50, Anthony PERARD wrote:
> When doing an action with a path and subpath in the xenstore,
> XenStoreJoin is called to generate "$path/$subpath". But this function
> do an allocation of memory which isn't necessary. Instead we will
> construct the path with WRITE_REQUEST and data used to generate the
> path will be copied directly to the xenstore shared ring.
> 
> Also change WRITE_REQUEST.Len type, it only contain sizes and doesn't
> need to be exactly 32bits.
> 
> Ref: https://bugzilla.tianocore.org/show_bug.cgi?id=2190
> Signed-off-by: Anthony PERARD <anthony.perard@xxxxxxxxxx>
> ---
>  OvmfPkg/XenBusDxe/XenStore.c | 78 +++++++++++++++++++++---------------
>  1 file changed, 46 insertions(+), 32 deletions(-)
> 
> diff --git a/OvmfPkg/XenBusDxe/XenStore.c b/OvmfPkg/XenBusDxe/XenStore.c
> index 7b71dc156d..ca7be12d68 100644
> --- a/OvmfPkg/XenBusDxe/XenStore.c
> +++ b/OvmfPkg/XenBusDxe/XenStore.c
> @@ -53,7 +53,7 @@
>  
>  typedef struct {
>    CONST VOID  *Data;
> -  UINT32      Len;
> +  UINTN       Len;
>  } WRITE_REQUEST;
>  
>  /* Register callback to watch subtree (node) in the XenStore. */
> @@ -260,6 +260,35 @@ XenStoreFindWatch (
>    return NULL;
>  }
>  
> +/**
> +  Fill the first three slots of a WRITE_REQUEST array.
> +
> +  When those three slots are concatenated to generate a string, the resulting
> +  string will be "$Path\0" or "$Path/$SubPath\0" if SubPath is provided.
> +**/
> +STATIC
> +VOID
> +XenStorePrepareWriteRequest (
> +  IN OUT WRITE_REQUEST *WriteRequest,
> +  IN     CONST CHAR8   *Path,
> +  IN     CONST CHAR8   *SubPath OPTIONAL
> +  )
> +{
> +  SetMem(WriteRequest, 3 * sizeof (WRITE_REQUEST), 0);

(1) ZeroMem() is more idiomatic. Also, please insert a space before the
opening paren.

> +  WriteRequest[0].Data = Path;
> +  WriteRequest[0].Len = AsciiStrSize (Path);
> +  if (SubPath != NULL && SubPath[0] != '\0') {
> +    //
> +    // Remove the \0 from the first part of the request.
> +    //
> +    WriteRequest[0].Len--;
> +    WriteRequest[1].Data = "/";
> +    WriteRequest[1].Len = 1;
> +    WriteRequest[2].Data = SubPath;
> +    WriteRequest[2].Len = AsciiStrSize (SubPath);
> +  }
> +}
> +

So this suggests that only the last element in the array should point to
a NUL-terminated string. Strings pointed-to by earlier elements in the
array should not be NUL-terminated. Is that correct?

>  //
>  // Public Utility Functions
>  // API comments for these methods can be found in XenStore.h
> @@ -842,6 +871,7 @@ XenStoreTalkv (
>    @param Transaction    The transaction to use for this request.
>    @param RequestType    The type of message to send.
>    @param Body           The body of the request.
> +  @param SubPath        If !NULL and not "", "/$SubPath" is append to Body.
>    @param LenPtr         The returned length of the reply.
>    @param Result         The returned body of the reply.
>  
> @@ -854,16 +884,16 @@ XenStoreSingle (
>    IN  CONST XENSTORE_TRANSACTION *Transaction,
>    IN  enum xsd_sockmsg_type   RequestType,
>    IN  CONST CHAR8             *Body,
> +  IN  CONST CHAR8             *SubPath OPTIONAL,
>    OUT UINT32                  *LenPtr OPTIONAL,
>    OUT VOID                    **Result OPTIONAL
>    )
>  {
> -  WRITE_REQUEST WriteRequest;
> +  WRITE_REQUEST   WriteRequest[3];
>  
> -  WriteRequest.Data = (VOID *) Body;
> -  WriteRequest.Len = (UINT32)AsciiStrSize (Body);
> +  XenStorePrepareWriteRequest (WriteRequest, Body, SubPath);
>  
> -  return XenStoreTalkv (Transaction, RequestType, &WriteRequest, 1,
> +  return XenStoreTalkv (Transaction, RequestType, WriteRequest, 3,
>                          LenPtr, Result);

(2) It would be slightly more idiomatic to pass

  ARRAY_SIZE (WriteRequest)

in place of the naked 3.

>  }
>  
> @@ -1113,15 +1143,12 @@ XenStoreListDirectory (
>    OUT CONST CHAR8           ***DirectoryListPtr
>    )
>  {
> -  CHAR8 *Path;
>    CHAR8 *TempStr;
>    UINT32 Len = 0;
>    XENSTORE_STATUS Status;
>  
> -  Path = XenStoreJoin (DirectoryPath, Node);
> -  Status = XenStoreSingle (Transaction, XS_DIRECTORY, Path, &Len,
> +  Status = XenStoreSingle (Transaction, XS_DIRECTORY, DirectoryPath, Node, 
> &Len,
>                             (VOID **) &TempStr);
> -  FreePool (Path);
>    if (Status != XENSTORE_STATUS_SUCCESS) {
>      return Status;
>    }
> @@ -1160,13 +1187,11 @@ XenStoreRead (
>    OUT VOID                    **Result
>    )
>  {
> -  CHAR8 *Path;
>    VOID *Value;
>    XENSTORE_STATUS Status;
>  
> -  Path = XenStoreJoin (DirectoryPath, Node);
> -  Status = XenStoreSingle (Transaction, XS_READ, Path, LenPtr, &Value);
> -  FreePool (Path);
> +  Status = XenStoreSingle (Transaction, XS_READ, DirectoryPath, Node,
> +    LenPtr, &Value);

(3) Indentation.

>    if (Status != XENSTORE_STATUS_SUCCESS) {
>      return Status;
>    }
> @@ -1183,21 +1208,13 @@ XenStoreWrite (
>    IN CONST CHAR8           *Str
>    )
>  {
> -  CHAR8 *Path;
> -  WRITE_REQUEST WriteRequest[2];
> -  XENSTORE_STATUS Status;
> +  WRITE_REQUEST   WriteRequest[4];
>  
> -  Path = XenStoreJoin (DirectoryPath, Node);
> +  XenStorePrepareWriteRequest (WriteRequest, DirectoryPath, Node);
> +  WriteRequest[3].Data = Str;
> +  WriteRequest[3].Len = AsciiStrLen (Str);

Now we have two strings, pointed-to by elements in the array, that are
NUL-terminated: the element at offset 2, and the one at offset 3. Is
that intentional? Is that part of the message framing?

Hmmm... From the original code:

>  
> -  WriteRequest[0].Data = (VOID *) Path;
> -  WriteRequest[0].Len = (UINT32)AsciiStrSize (Path);
> -  WriteRequest[1].Data = (VOID *) Str;
> -  WriteRequest[1].Len = (UINT32)AsciiStrLen (Str);

That seems to be the case. I guess the first run (offsets 0 through 2)
is parsed until the first NUL is encountered, for "path", then the
second run (3 and onwards) is parsed until the second NUL for "data".
Sounds plausible; OK.

> -
> -  Status = XenStoreTalkv (Transaction, XS_WRITE, WriteRequest, 2, NULL, 
> NULL);
> -  FreePool (Path);
> -
> -  return Status;
> +  return XenStoreTalkv (Transaction, XS_WRITE, WriteRequest, 4, NULL, NULL);

(4) Please use ARRAY_SIZE(); it's more robust.

>  }
>  
>  XENSTORE_STATUS
> @@ -1207,12 +1224,9 @@ XenStoreRemove (
>    IN CONST CHAR8            *Node
>    )
>  {
> -  CHAR8 *Path;
>    XENSTORE_STATUS Status;
>  
> -  Path = XenStoreJoin (DirectoryPath, Node);
> -  Status = XenStoreSingle (Transaction, XS_RM, Path, NULL, NULL);
> -  FreePool (Path);
> +  Status = XenStoreSingle (Transaction, XS_RM, DirectoryPath, Node, NULL, 
> NULL);
>  
>    return Status;
>  }
> @@ -1226,7 +1240,7 @@ XenStoreTransactionStart (
>    XENSTORE_STATUS Status;
>  
>    Status = XenStoreSingle (XST_NIL, XS_TRANSACTION_START, "", NULL,
> -                           (VOID **) &IdStr);
> +    NULL, (VOID **) &IdStr);

(5) Indentation.

>    if (Status == XENSTORE_STATUS_SUCCESS) {
>      Transaction->Id = (UINT32)AsciiStrDecimalToUintn (IdStr);
>      FreePool (IdStr);
> @@ -1246,7 +1260,7 @@ XenStoreTransactionEnd (
>    AbortStr[0] = Abort ? 'F' : 'T';
>    AbortStr[1] = '\0';
>  
> -  return XenStoreSingle (Transaction, XS_TRANSACTION_END, AbortStr, NULL, 
> NULL);
> +  return XenStoreSingle (Transaction, XS_TRANSACTION_END, AbortStr, NULL, 
> NULL, NULL);
>  }
>  
>  XENSTORE_STATUS
> 

With the above addressed:

Reviewed-by: Laszlo Ersek <lersek@xxxxxxxxxx>

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel

 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.