2
0
mirror of https://gitlab.isc.org/isc-projects/kea synced 2025-09-06 00:45:23 +00:00
Files
kea/src/lib/dns/cpp/buffer.h
JINMEI Tatuya 16e0f58b08 Added a small note about outputbuffer.getdata() in the detailed class
description.


git-svn-id: svn://bind10.isc.org/svn/bind10/branches/jinmei-dnsmessageapi@427 e5f2f494-b856-4b98-b285-d166d9295462
2010-01-04 20:31:05 +00:00

402 lines
16 KiB
C++

// Copyright (C) 2009 Internet Systems Consortium, Inc. ("ISC")
//
// Permission to use, copy, modify, and/or distribute this software for any
// purpose with or without fee is hereby granted, provided that the above
// copyright notice and this permission notice appear in all copies.
//
// THE SOFTWARE IS PROVIDED "AS IS" AND ISC DISCLAIMS ALL WARRANTIES WITH
// REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
// AND FITNESS. IN NO EVENT SHALL ISC BE LIABLE FOR ANY SPECIAL, DIRECT,
// INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
// LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
// OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
// PERFORMANCE OF THIS SOFTWARE.
// $Id$
#ifndef __BUFFER_H
#define __BUFFER_H 1
#include <vector>
#include <string.h>
#include <stdint.h>
#include "exceptions.h"
namespace isc {
namespace dns {
///
/// \brief A standard DNS module exception that is thrown if an out-of-range
/// buffer operation is being performed.
///
class InvalidBufferPosition : public Exception {
public:
InvalidBufferPosition(const char* file, size_t line, const char* what) :
isc::dns::Exception(file, line, what) {}
};
///\brief The \c InputBuffer class is a buffer abstraction for manipulating
/// read-only data.
///
/// The main purpose of this class is to provide a safe placeholder for
/// examining wire-format data received from a network.
///
/// Applications normally use this class only in a limited situation: as an
/// interface between legacy I/O operation (such as receiving data from a BSD
/// socket) and the rest of the BIND10 DNS library. One common usage of this
/// class for an application would therefore be something like this:
///
/// \code unsigned char buf[1024];
/// struct sockaddr addr;
/// socklen_t addrlen = sizeof(addr);
/// int cc = recvfrom(s, buf, sizeof(buf), 0, &addr, &addrlen);
/// InputBuffer buffer(buf, cc);
/// // pass the buffer to a DNS message object to parse the message \endcode
///
/// Other BIND10 DNS classes will then use methods of this class to get access
/// to the data, but the application normally doesn't have to care about the
/// details.
///
/// An \c InputBuffer object internally holds a reference to the given data,
/// rather than make a local copy of the data. Also, it does not have an
/// ownership of the given data. It is application's responsibility to ensure
/// the data remains valid throughout the lifetime of the \c InputBuffer
/// object. Likewise, this object generally assumes the data isn't modified
/// throughout its lifetime; if the application modifies the data while this
/// object retains a reference to it, the result is undefined. The application
/// will also be responsible for releasing the data when it's not needed if it
/// was dynamically acquired.
///
/// This is a deliberate design choice: although it's safer to make a local
/// copy of the given data on construction, it would cause unacceptable
/// performance overhead, especially considering that a DNS message can be
/// as large as a few KB. Alternatively, we could allow the object to allocate
/// memory internally and expose it to the application to store network data
/// in it. This is also a bad design, however, in that we would effectively
/// break the abstraction employed in the class, and do so by publishing
/// "read-only" stuff as a writable memory region. Since there doesn't seem to
/// be a perfect solution, we have adopted what we thought a "least bad" one.
///
/// Methods for reading data from the buffer generally work like an input
/// stream: it begins with the head of the data, and once some length of data
/// is read from the buffer, the next read operation will take place from the
/// head of the unread data. An object of this class internally holds (a
/// notion of) where the next read operation should start. We call it the
/// <em>read position</em> in this document.
class InputBuffer {
public:
///
/// \name Constructors and Destructor
//@{
/// \brief Constructor from variable length of data.
///
/// It is caller's responsibility to ensure that the data is valid as long
/// as the buffer exists.
/// \param data A pointer to the data stored in the buffer.
/// \param len The length of the data in bytes.
InputBuffer(const void* data, size_t len) :
position_(0), data_(static_cast<const uint8_t*>(data)), len_(len) {}
//@}
///
/// \name Getter Methods
//@{
/// \brief Return the length of the data stored in the buffer.
size_t getLength() const { return (len_); }
/// \brief Return the current read position.
size_t getPosition() const { return (position_); }
//@}
///
/// \name Setter Methods
///
//@{
/// \brief Set the read position of the buffer to the given value.
///
/// The new position must be in the valid range of the buffer; otherwise
/// an exception of class \c isc::dns::InvalidBufferPosition will be thrown.
/// \param position The new position (offset from the beginning of the
/// buffer).
void setPosition(size_t position)
{
if (position > len_)
dns_throw(InvalidBufferPosition, "position is too large");
position_ = position;
}
//@}
///
/// \name Methods for reading data from the buffer.
//@{
/// \brief Read an unsigned 8-bit integer from the buffer and return it.
///
/// If the remaining length of the buffer is smaller than 8-bit, an
/// exception of class \c isc::dns::InvalidBufferPosition will be thrown.
uint8_t readUint8()
{
if (position_ + sizeof(uint8_t) > len_) {
dns_throw(InvalidBufferPosition, "read beyond end of buffer");
}
return (data_[position_++]);
}
/// \brief Read an unsigned 16-bit integer in network byte order from the
/// buffer, convert it to host byte order, and return it.
///
/// If the remaining length of the buffer is smaller than 16-bit, an
/// exception of class \c isc::dns::InvalidBufferPosition will be thrown.
uint16_t readUint16()
{
uint16_t data;
const uint8_t* cp;
if (position_ + sizeof(data) > len_) {
dns_throw(InvalidBufferPosition, "read beyond end of buffer");
}
cp = &data_[position_];
data = ((unsigned int)(cp[0])) << 8;
data |= ((unsigned int)(cp[1]));
position_ += sizeof(data);
return (data);
}
/// \brief Read an unsigned 32-bit integer in network byte order from the
/// buffer, convert it to host byte order, and return it.
///
/// If the remaining length of the buffer is smaller than 32-bit, an
/// exception of class \c isc::dns::InvalidBufferPosition will be thrown.
uint32_t readUint32()
{
uint32_t data;
const uint8_t* cp;
if (position_ + sizeof(data) > len_) {
dns_throw(InvalidBufferPosition, "read beyond end of buffer");
}
cp = &data_[position_];
data = ((unsigned int)(cp[0])) << 24;
data |= ((unsigned int)(cp[1])) << 16;
data |= ((unsigned int)(cp[2])) << 8;
data |= ((unsigned int)(cp[3]));
position_ += sizeof(data);
return (data);
}
/// \brief Read data of the specified length from the buffer and copy it to
/// the caller supplied buffer.
///
/// The data is copied as stored in the buffer; no conversion is performed.
/// If the remaining length of the buffer is smaller than the specified
/// length, an exception of class \c isc::dns::InvalidBufferPosition will
/// be thrown.
void readData(void* data, size_t len)
{
if (position_ + len > len_) {
dns_throw(InvalidBufferPosition, "read beyond end of buffer");
}
memcpy(data, &data_[position_], len);
position_ += len;
}
//@}
private:
size_t position_;
const uint8_t* data_;
size_t len_;
};
///
///\brief The \c OutputBuffer class is a buffer abstraction for manipulating
/// mutable data.
///
/// The main purpose of this class is to provide a safe workplace for
/// constructing wire-format data to be sent out to a network. Here,
/// <em>safe</em> means that it automatically allocates necessary memory and
/// avoid buffer overrun.
///
/// Like for the \c InputBuffer class, applications normally use this class only
/// in a limited situation. One common usage of this class for an application
/// would be something like this:
///
/// \code OutputBuffer buffer(4096); // give a sufficiently large initial size
/// // pass the buffer to a DNS message object to construct a wire-format
/// // DNS message.
/// struct sockaddr to;
/// sendto(s, buffer.getData(), buffer.getLength(), 0, &to, sizeof(to));
/// \endcode
///
/// where the \c getData() method gives a reference to the internal memory
/// region stored in the \c buffer object. This is a suboptimal design in that
/// it exposes an encapsulated "handle" of an object to its user.
/// Unfortunately, there is no easy way to avoid this without involving
/// expensive data copy if we want to use this object with a legacy API such as
/// a BSD socket interface. And, indeed, this is one major purpose for this
/// object. Applications should use this method only under such a special
/// circumstance. It should also be noted that the memory region returned by
/// \c getData() may be invalidated after a subsequent write operation.
///
/// An \c OutputBuffer class object automatically extends its memory region when
/// data is written beyond the end of the current buffer. However, it will
/// involve performance overhead such as reallocating more memory and copying
/// data. It is therefore recommended to construct the buffer object with a
/// sufficiently large initial size.
/// The \c getCapacity() method provides the current maximum size of data
/// (including the portion already written) that can be written into the buffer
/// without causing memory reallocation.
///
/// Methods for writing data into the buffer generally work like an output
/// stream: it begins with the head of the buffer, and once some length of data
/// is written into the buffer, the next write operation will take place from
/// the end of the buffer. Other methods to emulate "random access" are also
/// provided (e.g., \c writeUint16At()). The normal write operations are
/// normally exception-free as this class automatically extends the buffer
/// when necessary. However, in extreme cases such as an attempt of writing
/// multi-GB data, a separate exception (e.g., \c std::bad_alloc) may be thrown
/// by the system. This also applies to the constructor with a very large
/// initial size.
///
/// Note to developers: it may make more sense to introduce an abstract base
/// class for the \c OutputBuffer and define the simple implementation as a
/// a concrete derived class. That way we can provide flexibility for future
/// extension such as more efficient buffer implementation or allowing users
/// to have their own customized version without modifying the source code.
/// We in fact considered that option, but at the moment chose the simpler
/// approach with a single concrete class because it may make the
/// implementation unnecessarily complicated while we were still not certain
/// if we really want that flexibility. We may revisit the class design as
/// we see more applications of the class. The same considerations apply to
/// the \c InputBuffer and \c MessageRenderer classes.
class OutputBuffer {
public:
///
/// \name Constructors and Destructor
///
//@{
/// \brief Constructor from the initial size of the buffer.
///
/// \param len The initial length of the buffer in bytes.
OutputBuffer(size_t len) { data_.reserve(len); }
//@}
///
/// \name Getter Methods
///
//@{
/// \brief Return the current capacity of the buffer.
size_t getCapacity() const { return (data_.capacity()); }
/// \brief Return a pointer to the head of the data stored in the buffer.
///
/// The caller can assume that the subsequent \c getLength() bytes are
/// identical to the stored data of the buffer.
///
/// Note: The pointer returned by this method may be invalidated after a
/// subsequent write operation.
const void* getData() const { return (&data_[0]); }
/// \brief Return the length of data written in the buffer.
size_t getLength() const { return (data_.size()); }
/// \brief Return the value of the buffer at the specified position.
///
/// \c pos must specify the valid position of the buffer; otherwise an
/// exception class of \c InvalidBufferPosition will be thrown.
///
/// \param pos The position in the buffer to be returned.
uint8_t operator[](size_t pos) const
{
if (pos >= data_.size()) {
dns_throw(InvalidBufferPosition, "read at invalid position");
}
return (data_[pos]);
}
//@}
///
/// \name Methods for writing data into the buffer.
///
//@{
/// \brief Insert a specified length of gap at the end of the buffer.
///
/// The caller should not assume any particular value to be inserted.
/// This method is provided as a shortcut to make a hole in the buffer
/// that is to be filled in later, e.g, by \ref writeUint16At().
/// \param len The length of the gap to be inserted in bytes.
void skip(size_t len) { data_.insert(data_.end(), len, 0); }
/// \brief Clear buffer content.
///
/// This method can be used to re-initialize and reuse the buffer without
/// constructing a new one.
void clear() { data_.clear(); }
/// \brief Write an unsigned 8-bit integer into the buffer.
///
/// \param data The 8-bit integer to be written into the buffer.
void writeUint8(uint8_t data) { data_.push_back(data); }
/// \brief Write an unsigned 16-bit integer in host byte order into the
/// buffer in network byte order.
///
/// \param data The 16-bit integer to be written into the buffer.
void writeUint16(uint16_t data)
{
data_.push_back(static_cast<uint8_t>((data & 0xff00U) >> 8));
data_.push_back(static_cast<uint8_t>(data & 0x00ffU));
}
/// \brief Write an unsigned 16-bit integer in host byte order at the
/// specified position of the buffer in network byte order.
///
/// The buffer must have a sufficient room to store the given data at the
/// given position, that is, <code>pos + 2 < getLength()</code>;
/// otherwise an exception of class \c isc::dns::InvalidBufferPosition will
/// be thrown.
/// Note also that this method never extends the buffer.
///
/// \param data The 16-bit integer to be written into the buffer.
/// \param pos The beginning position in the buffer to write the data.
void writeUint16At(uint16_t data, size_t pos)
{
if (pos + sizeof(data) >= data_.size()) {
dns_throw(InvalidBufferPosition, "write at invalid position");
}
data_[pos] = static_cast<uint8_t>((data & 0xff00U) >> 8);
data_[pos + 1] = static_cast<uint8_t>(data & 0x00ffU);
}
/// \brief Write an unsigned 32-bit integer in host byte order
/// into the buffer in network byte order.
///
/// \param data The 32-bit integer to be written into the buffer.
void writeUint32(uint32_t data)
{
data_.push_back(static_cast<uint8_t>((data & 0xff000000) >> 24));
data_.push_back(static_cast<uint8_t>((data & 0x00ff0000) >> 16));
data_.push_back(static_cast<uint8_t>((data & 0x0000ff00) >> 8));
data_.push_back(static_cast<uint8_t>(data & 0x000000ff));
}
/// \brief Copy an arbitrary length of data into the buffer.
///
/// No conversion on the copied data is performed.
///
/// \param data A pointer to the data to be copied into the buffer.
/// \param len The length of the data in bytes.
void writeData(const void *data, size_t len)
{
const uint8_t* cp = static_cast<const uint8_t*>(data);
data_.insert(data_.end(), cp, cp + len);
}
//@}
private:
std::vector<uint8_t> data_;
};
}
}
#endif // __BUFFER_H
// Local Variables:
// mode: c++
// End: