blob: e80ddd70b746ce21bc79e8fcd94172a65290916c [file] [log] [blame]
/*
*
* Copyright (c) 2020-2021 Project CHIP Authors
* Copyright (c) 2016-2017 Nest Labs, Inc.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
/**
* @file
* This file defines the member functions and private data for
* the chip::System::PacketBuffer class, which provides the
* mechanisms for manipulating packets of octet-serialized
* data.
*/
// Include standard C library limit macros
#ifndef __STDC_LIMIT_MACROS
#define __STDC_LIMIT_MACROS
#endif
// Include module header
#include <system/SystemPacketBuffer.h>
// Include local headers
#include <lib/support/CodeUtils.h>
#include <lib/support/SafeInt.h>
#include <lib/support/logging/CHIPLogging.h>
#include <system/SystemFaultInjection.h>
#include <system/SystemMutex.h>
#include <system/SystemStats.h>
#include <stdint.h>
#include <limits.h>
#include <limits>
#include <stddef.h>
#include <stdlib.h>
#include <string.h>
#include <utility>
#if CHIP_SYSTEM_CONFIG_USE_LWIP
#include <lwip/mem.h>
#include <lwip/pbuf.h>
#endif // CHIP_SYSTEM_CONFIG_USE_LWIP
#if CHIP_SYSTEM_PACKETBUFFER_FROM_CHIP_HEAP
#include <lib/support/CHIPMem.h>
#endif
namespace chip {
namespace System {
#if CHIP_SYSTEM_PACKETBUFFER_FROM_CHIP_POOL
//
// Pool allocation for PacketBuffer objects.
//
PacketBuffer::BufferPoolElement PacketBuffer::sBufferPool[CHIP_SYSTEM_CONFIG_PACKETBUFFER_POOL_SIZE];
PacketBuffer * PacketBuffer::sFreeList = PacketBuffer::BuildFreeList();
#if !CHIP_SYSTEM_CONFIG_NO_LOCKING
static Mutex sBufferPoolMutex;
#define LOCK_BUF_POOL() \
do \
{ \
sBufferPoolMutex.Lock(); \
} while (0)
#define UNLOCK_BUF_POOL() \
do \
{ \
sBufferPoolMutex.Unlock(); \
} while (0)
#endif // !CHIP_SYSTEM_CONFIG_NO_LOCKING
PacketBuffer * PacketBuffer::BuildFreeList()
{
pbuf * lHead = nullptr;
for (int i = 0; i < CHIP_SYSTEM_CONFIG_PACKETBUFFER_POOL_SIZE; i++)
{
pbuf * lCursor = &sBufferPool[i].Header;
lCursor->next = lHead;
lCursor->ref = 0;
lHead = lCursor;
}
#if !CHIP_SYSTEM_CONFIG_NO_LOCKING
Mutex::Init(sBufferPoolMutex);
#endif // !CHIP_SYSTEM_CONFIG_NO_LOCKING
return static_cast<PacketBuffer *>(lHead);
}
#elif CHIP_SYSTEM_PACKETBUFFER_FROM_CHIP_HEAP
//
// Heap allocation for PacketBuffer objects.
//
#if CHIP_SYSTEM_PACKETBUFFER_HAS_CHECK
void PacketBuffer::InternalCheck(const PacketBuffer * buffer)
{
if (buffer)
{
VerifyOrDieWithMsg(::chip::Platform::MemoryDebugCheckPointer(buffer, buffer->alloc_size + kStructureSize), chipSystemLayer,
"invalid packet buffer pointer");
VerifyOrDieWithMsg(buffer->alloc_size >= buffer->ReservedSize() + buffer->len, chipSystemLayer,
"packet buffer overflow %u < %u+%u", buffer->alloc_size, buffer->ReservedSize(), buffer->len);
}
}
#endif // CHIP_SYSTEM_PACKETBUFFER_HAS_CHECK
// Number of unused bytes below which \c RightSize() won't bother reallocating.
constexpr uint16_t kRightSizingThreshold = 16;
void PacketBufferHandle::InternalRightSize()
{
// Require a single buffer with no other references.
if ((mBuffer == nullptr) || mBuffer->HasChainedBuffer() || (mBuffer->ref != 1))
{
return;
}
// Reallocate only if enough space will be saved.
const uint8_t * const start = mBuffer->ReserveStart();
const uint8_t * const payload = mBuffer->Start();
const uint16_t usedSize = static_cast<uint16_t>(payload - start + mBuffer->len);
if (usedSize + kRightSizingThreshold > mBuffer->alloc_size)
{
return;
}
const size_t blockSize = usedSize + PacketBuffer::kStructureSize;
PacketBuffer * newBuffer = reinterpret_cast<PacketBuffer *>(chip::Platform::MemoryAlloc(blockSize));
if (newBuffer == nullptr)
{
ChipLogError(chipSystemLayer, "PacketBuffer: pool EMPTY.");
return;
}
uint8_t * const newStart = newBuffer->ReserveStart();
newBuffer->next = nullptr;
newBuffer->payload = newStart + (payload - start);
newBuffer->tot_len = mBuffer->tot_len;
newBuffer->len = mBuffer->len;
newBuffer->ref = 1;
newBuffer->alloc_size = static_cast<uint16_t>(usedSize);
memcpy(newStart, start, usedSize);
PacketBuffer::Free(mBuffer);
mBuffer = newBuffer;
}
#elif CHIP_SYSTEM_PACKETBUFFER_FROM_LWIP_CUSTOM_POOL
void PacketBufferHandle::InternalRightSize()
{
PacketBuffer * lNewPacket = static_cast<PacketBuffer *>(pbuf_rightsize((struct pbuf *) mBuffer, -1));
if (lNewPacket != mBuffer)
{
mBuffer = lNewPacket;
SYSTEM_STATS_UPDATE_LWIP_PBUF_COUNTS();
ChipLogDetail(chipSystemLayer, "PacketBuffer: RightSize Copied");
}
}
#endif
#ifndef LOCK_BUF_POOL
#define LOCK_BUF_POOL() \
do \
{ \
} while (0)
#endif // !defined(LOCK_BUF_POOL)
#ifndef UNLOCK_BUF_POOL
#define UNLOCK_BUF_POOL() \
do \
{ \
} while (0)
#endif // !defined(UNLOCK_BUF_POOL)
void PacketBuffer::SetStart(uint8_t * aNewStart)
{
uint8_t * const kStart = ReserveStart();
uint8_t * const kEnd = kStart + this->AllocSize();
if (aNewStart < kStart)
aNewStart = kStart;
else if (aNewStart > kEnd)
aNewStart = kEnd;
ptrdiff_t lDelta = aNewStart - static_cast<uint8_t *>(this->payload);
if (lDelta > this->len)
lDelta = this->len;
this->len = static_cast<uint16_t>(static_cast<ptrdiff_t>(this->len) - lDelta);
this->tot_len = static_cast<uint16_t>(static_cast<ptrdiff_t>(this->tot_len) - lDelta);
this->payload = aNewStart;
}
void PacketBuffer::SetDataLength(uint16_t aNewLen, PacketBuffer * aChainHead)
{
const uint16_t kMaxDataLen = this->MaxDataLength();
if (aNewLen > kMaxDataLen)
aNewLen = kMaxDataLen;
ptrdiff_t lDelta = static_cast<ptrdiff_t>(aNewLen) - static_cast<ptrdiff_t>(this->len);
this->len = aNewLen;
this->tot_len = static_cast<uint16_t>(this->tot_len + lDelta);
// SetDataLength is often called after a client finished writing to the buffer,
// so it's a good time to check for possible corruption.
Check(this);
while (aChainHead != nullptr && aChainHead != this)
{
Check(aChainHead);
aChainHead->tot_len = static_cast<uint16_t>(aChainHead->tot_len + lDelta);
aChainHead = aChainHead->ChainedBuffer();
}
}
uint16_t PacketBuffer::MaxDataLength() const
{
return static_cast<uint16_t>(AllocSize() - ReservedSize());
}
uint16_t PacketBuffer::AvailableDataLength() const
{
return static_cast<uint16_t>(this->MaxDataLength() - this->DataLength());
}
uint16_t PacketBuffer::ReservedSize() const
{
// Cast to uint16_t is safe because Start() always points to "after"
// ReserveStart(). At least when the payload is stored inline.
return static_cast<uint16_t>(Start() - ReserveStart());
}
uint8_t * PacketBuffer::ReserveStart()
{
return reinterpret_cast<uint8_t *>(this) + kStructureSize;
}
const uint8_t * PacketBuffer::ReserveStart() const
{
return reinterpret_cast<const uint8_t *>(this) + kStructureSize;
}
void PacketBuffer::AddToEnd(PacketBufferHandle && aPacketHandle)
{
// Ownership of aPacketHandle's buffer is transferred to the end of the chain.
PacketBuffer * aPacket = std::move(aPacketHandle).UnsafeRelease();
#if CHIP_SYSTEM_CONFIG_USE_LWIP
pbuf_cat(this, aPacket);
#else // !CHIP_SYSTEM_CONFIG_USE_LWIP
PacketBuffer * lCursor = this;
while (true)
{
uint16_t old_total_length = lCursor->tot_len;
lCursor->tot_len = static_cast<uint16_t>(lCursor->tot_len + aPacket->tot_len);
VerifyOrDieWithMsg(lCursor->tot_len >= old_total_length, chipSystemLayer, "buffer chain too large");
if (!lCursor->HasChainedBuffer())
{
lCursor->next = aPacket;
break;
}
lCursor = lCursor->ChainedBuffer();
}
#endif // !CHIP_SYSTEM_CONFIG_USE_LWIP
}
void PacketBuffer::CompactHead()
{
uint8_t * const kStart = ReserveStart();
if (this->payload != kStart)
{
memmove(kStart, this->payload, this->len);
this->payload = kStart;
}
uint16_t lAvailLength = this->AvailableDataLength();
while (lAvailLength > 0 && HasChainedBuffer())
{
PacketBuffer & lNextPacket = *ChainedBuffer();
VerifyOrDieWithMsg(lNextPacket.ref == 1, chipSystemLayer, "next buffer %p is not exclusive to this chain", &lNextPacket);
uint16_t lMoveLength = lNextPacket.len;
if (lMoveLength > lAvailLength)
lMoveLength = lAvailLength;
memcpy(static_cast<uint8_t *>(this->payload) + this->len, lNextPacket.payload, lMoveLength);
lNextPacket.payload = static_cast<uint8_t *>(lNextPacket.payload) + lMoveLength;
this->len = static_cast<uint16_t>(this->len + lMoveLength);
lAvailLength = static_cast<uint16_t>(lAvailLength - lMoveLength);
lNextPacket.len = static_cast<uint16_t>(lNextPacket.len - lMoveLength);
lNextPacket.tot_len = static_cast<uint16_t>(lNextPacket.tot_len - lMoveLength);
if (lNextPacket.len == 0)
this->next = this->FreeHead(&lNextPacket);
}
}
void PacketBuffer::ConsumeHead(uint16_t aConsumeLength)
{
if (aConsumeLength > this->len)
aConsumeLength = this->len;
this->payload = static_cast<uint8_t *>(this->payload) + aConsumeLength;
this->len = static_cast<uint16_t>(this->len - aConsumeLength);
this->tot_len = static_cast<uint16_t>(this->tot_len - aConsumeLength);
}
/**
* Consume data in a chain of buffers.
*
* Consume data in a chain of buffers starting with the current buffer and proceeding through the remaining buffers in the
* chain. Each buffer that is completely consumed is freed and the function returns the first buffer (if any) containing the
* remaining data. The current buffer must be the head of the buffer chain.
*
* @param[in] aConsumeLength - number of bytes to consume from the current chain.
*
* @return the first buffer from the current chain that contains any remaining data. If no data remains, nullptr is returned.
*/
PacketBuffer * PacketBuffer::Consume(uint16_t aConsumeLength)
{
PacketBuffer * lPacket = this;
while (lPacket != nullptr && aConsumeLength > 0)
{
const uint16_t kLength = lPacket->DataLength();
if (aConsumeLength >= kLength)
{
lPacket = PacketBuffer::FreeHead(lPacket);
aConsumeLength = static_cast<uint16_t>(aConsumeLength - kLength);
}
else
{
lPacket->ConsumeHead(aConsumeLength);
break;
}
}
return lPacket;
}
CHIP_ERROR PacketBuffer::Read(uint8_t * aDestination, size_t aReadLength) const
{
const PacketBuffer * lPacket = this;
if (aReadLength > TotalLength())
{
return CHIP_ERROR_BUFFER_TOO_SMALL;
}
while (aReadLength > 0)
{
if (lPacket == nullptr)
{
// TotalLength() or an individual buffer's DataLength() must have been wrong.
return CHIP_ERROR_INTERNAL;
}
size_t lToReadFromCurrentBuf = lPacket->DataLength();
if (aReadLength < lToReadFromCurrentBuf)
{
lToReadFromCurrentBuf = aReadLength;
}
memcpy(aDestination, lPacket->Start(), lToReadFromCurrentBuf);
aDestination += lToReadFromCurrentBuf;
aReadLength -= lToReadFromCurrentBuf;
lPacket = lPacket->ChainedBuffer();
}
return CHIP_NO_ERROR;
}
bool PacketBuffer::EnsureReservedSize(uint16_t aReservedSize)
{
const uint16_t kCurrentReservedSize = this->ReservedSize();
if (aReservedSize <= kCurrentReservedSize)
return true;
if ((aReservedSize + this->len) > this->AllocSize())
return false;
// Cast is safe because aReservedSize > kCurrentReservedSize.
const uint16_t kMoveLength = static_cast<uint16_t>(aReservedSize - kCurrentReservedSize);
memmove(static_cast<uint8_t *>(this->payload) + kMoveLength, this->payload, this->len);
payload = static_cast<uint8_t *>(this->payload) + kMoveLength;
return true;
}
bool PacketBuffer::AlignPayload(uint16_t aAlignBytes)
{
if (aAlignBytes == 0)
return false;
const uint16_t kPayloadOffset = static_cast<uint16_t>(reinterpret_cast<uintptr_t>(this->payload) % aAlignBytes);
if (kPayloadOffset == 0)
return true;
// Cast is safe because by construction kPayloadOffset < aAlignBytes.
const uint16_t kPayloadShift = static_cast<uint16_t>(aAlignBytes - kPayloadOffset);
if (!CanCastTo<uint16_t>(this->ReservedSize() + kPayloadShift))
{
return false;
}
return (this->EnsureReservedSize(static_cast<uint16_t>(this->ReservedSize() + kPayloadShift)));
}
/**
* Increment the reference count of the current buffer.
*/
void PacketBuffer::AddRef()
{
#if CHIP_SYSTEM_CONFIG_USE_LWIP
pbuf_ref(this);
#else // !CHIP_SYSTEM_CONFIG_USE_LWIP
LOCK_BUF_POOL();
VerifyOrDieWithMsg(this->ref < std::numeric_limits<decltype(this->ref)>::max(), chipSystemLayer,
"packet buffer refcount overflow");
++this->ref;
UNLOCK_BUF_POOL();
#endif // !CHIP_SYSTEM_CONFIG_USE_LWIP
}
PacketBufferHandle PacketBufferHandle::New(size_t aAvailableSize, uint16_t aReservedSize)
{
// Adding three 16-bit-int sized numbers together will never overflow
// assuming int is at least 32 bits.
static_assert(INT_MAX >= INT32_MAX, "int is not big enough");
static_assert(PacketBuffer::kStructureSize == sizeof(PacketBuffer), "PacketBuffer size mismatch");
static_assert(PacketBuffer::kStructureSize < UINT16_MAX, "Check for overflow more carefully");
static_assert(SIZE_MAX >= INT_MAX, "Our additions might not fit in size_t");
static_assert(PacketBuffer::kMaxSizeWithoutReserve <= UINT16_MAX, "PacketBuffer may have size not fitting uint16_t");
// When `aAvailableSize` fits in uint16_t (as tested below) and size_t is at least 32 bits (as asserted above),
// these additions will not overflow.
const size_t lAllocSize = aReservedSize + aAvailableSize;
const size_t lBlockSize = PacketBuffer::kStructureSize + lAllocSize;
PacketBuffer * lPacket;
CHIP_SYSTEM_FAULT_INJECT(FaultInjection::kFault_PacketBufferNew, return PacketBufferHandle());
if (aAvailableSize > UINT16_MAX || lAllocSize > PacketBuffer::kMaxSizeWithoutReserve || lBlockSize > UINT16_MAX)
{
ChipLogError(chipSystemLayer, "PacketBuffer: allocation too large.");
return PacketBufferHandle();
}
#if CHIP_SYSTEM_CONFIG_USE_LWIP
lPacket = static_cast<PacketBuffer *>(
pbuf_alloc(PBUF_RAW, static_cast<uint16_t>(lAllocSize), CHIP_SYSTEM_CONFIG_PACKETBUFFER_LWIP_PBUF_TYPE));
SYSTEM_STATS_UPDATE_LWIP_PBUF_COUNTS();
#elif CHIP_SYSTEM_PACKETBUFFER_FROM_CHIP_POOL
static_cast<void>(lBlockSize);
#if !CHIP_SYSTEM_CONFIG_NO_LOCKING && CHIP_SYSTEM_CONFIG_FREERTOS_LOCKING
if (!sBufferPoolMutex.isInitialized())
{
Mutex::Init(sBufferPoolMutex);
}
#endif
LOCK_BUF_POOL();
lPacket = PacketBuffer::sFreeList;
if (lPacket != nullptr)
{
PacketBuffer::sFreeList = lPacket->ChainedBuffer();
SYSTEM_STATS_INCREMENT(chip::System::Stats::kSystemLayer_NumPacketBufs);
}
UNLOCK_BUF_POOL();
#elif CHIP_SYSTEM_PACKETBUFFER_FROM_CHIP_HEAP
lPacket = reinterpret_cast<PacketBuffer *>(chip::Platform::MemoryAlloc(lBlockSize));
SYSTEM_STATS_INCREMENT(chip::System::Stats::kSystemLayer_NumPacketBufs);
#else
#error "Unimplemented PacketBuffer storage case"
#endif
if (lPacket == nullptr)
{
ChipLogError(chipSystemLayer, "PacketBuffer: pool EMPTY.");
return PacketBufferHandle();
}
lPacket->payload = lPacket->ReserveStart() + aReservedSize;
lPacket->len = lPacket->tot_len = 0;
lPacket->next = nullptr;
lPacket->ref = 1;
#if CHIP_SYSTEM_PACKETBUFFER_FROM_CHIP_HEAP
lPacket->alloc_size = static_cast<uint16_t>(lAllocSize);
#endif
return PacketBufferHandle(lPacket);
}
PacketBufferHandle PacketBufferHandle::NewWithData(const void * aData, size_t aDataSize, uint16_t aAdditionalSize,
uint16_t aReservedSize)
{
if (aDataSize > UINT16_MAX)
{
ChipLogError(chipSystemLayer, "PacketBuffer: allocation too large.");
return PacketBufferHandle();
}
// Since `aDataSize` fits in uint16_t, the sum `aDataSize + aAdditionalSize` will not overflow.
// `New()` will only return a non-null buffer if the total allocation size does not overflow.
PacketBufferHandle buffer = New(aDataSize + aAdditionalSize, aReservedSize);
if (buffer.mBuffer != nullptr)
{
memcpy(buffer.mBuffer->payload, aData, aDataSize);
buffer.mBuffer->len = buffer.mBuffer->tot_len = static_cast<uint16_t>(aDataSize);
}
return buffer;
}
/**
* Free all packet buffers in a chain.
*
* Decrement the reference count to all the buffers in the current chain. If the reference count reaches 0, the respective buffers
* are freed or returned to allocation pools as appropriate. As a rule, users should treat this method as an equivalent of
* `free()` function and not use the argument after the call.
*
* @param[in] aPacket - packet buffer to be freed.
*/
void PacketBuffer::Free(PacketBuffer * aPacket)
{
#if CHIP_SYSTEM_CONFIG_USE_LWIP
if (aPacket != nullptr)
{
pbuf_free(aPacket);
SYSTEM_STATS_UPDATE_LWIP_PBUF_COUNTS();
}
#elif CHIP_SYSTEM_PACKETBUFFER_FROM_CHIP_HEAP || CHIP_SYSTEM_PACKETBUFFER_FROM_CHIP_POOL
LOCK_BUF_POOL();
while (aPacket != nullptr)
{
PacketBuffer * lNextPacket = aPacket->ChainedBuffer();
VerifyOrDieWithMsg(aPacket->ref > 0, chipSystemLayer, "SystemPacketBuffer::Free: aPacket->ref = 0");
aPacket->ref--;
if (aPacket->ref == 0)
{
SYSTEM_STATS_DECREMENT(chip::System::Stats::kSystemLayer_NumPacketBufs);
#if CHIP_SYSTEM_PACKETBUFFER_FROM_CHIP_HEAP
::chip::Platform::MemoryDebugCheckPointer(aPacket, aPacket->alloc_size + kStructureSize);
#endif
aPacket->Clear();
#if CHIP_SYSTEM_PACKETBUFFER_FROM_CHIP_POOL
aPacket->next = sFreeList;
sFreeList = aPacket;
#elif CHIP_SYSTEM_PACKETBUFFER_FROM_CHIP_HEAP
chip::Platform::MemoryFree(aPacket);
#endif
aPacket = lNextPacket;
}
else
{
aPacket = nullptr;
}
}
UNLOCK_BUF_POOL();
#else
#error "Unimplemented PacketBuffer storage case"
#endif
}
/**
* Clear content of the packet buffer.
*
* This method is called by Free(), before the buffer is released to the free buffer pool.
*/
void PacketBuffer::Clear()
{
tot_len = 0;
len = 0;
#if CHIP_SYSTEM_PACKETBUFFER_FROM_CHIP_HEAP
alloc_size = 0;
#endif
}
/**
* Free the first buffer in a chain, returning a pointer to the remaining buffers.
`*
* @note When the buffer chain is referenced by multiple callers, `FreeHead()` will detach the head, but will not forcibly
* deallocate the head buffer.
*
* @param[in] aPacket - buffer chain.
*
* @return packet buffer chain consisting of the tail of the input buffer (may be \c nullptr).
*/
PacketBuffer * PacketBuffer::FreeHead(PacketBuffer * aPacket)
{
PacketBuffer * lNextPacket = aPacket->ChainedBuffer();
aPacket->next = nullptr;
PacketBuffer::Free(aPacket);
return lNextPacket;
}
PacketBufferHandle PacketBufferHandle::PopHead()
{
PacketBuffer * head = mBuffer;
// This takes ownership from the `next` link.
mBuffer = mBuffer->ChainedBuffer();
head->next = nullptr;
head->tot_len = head->len;
// The returned handle takes ownership from this.
return PacketBufferHandle(head);
}
PacketBufferHandle PacketBufferHandle::CloneData() const
{
PacketBufferHandle cloneHead;
for (PacketBuffer * original = mBuffer; original != nullptr; original = original->ChainedBuffer())
{
uint16_t originalDataSize = original->MaxDataLength();
uint16_t originalReservedSize = original->ReservedSize();
if (originalDataSize + originalReservedSize > PacketBuffer::kMaxSizeWithoutReserve)
{
// The original memory allocation may have provided a larger block than requested (e.g. when using a shared pool),
// and in particular may have provided a larger block than we are able to request from PackBufferHandle::New().
// It is a genuine error if that extra space has been used.
if (originalReservedSize + original->DataLength() > PacketBuffer::kMaxSizeWithoutReserve)
{
return PacketBufferHandle();
}
// Otherwise, reduce the requested data size. This subtraction can not underflow because the above test
// guarantees originalReservedSize <= PacketBuffer::kMaxSizeWithoutReserve.
originalDataSize = static_cast<uint16_t>(PacketBuffer::kMaxSizeWithoutReserve - originalReservedSize);
}
PacketBufferHandle clone = PacketBufferHandle::New(originalDataSize, originalReservedSize);
if (clone.IsNull())
{
return PacketBufferHandle();
}
clone.mBuffer->tot_len = clone.mBuffer->len = original->len;
memcpy(clone->ReserveStart(), original->ReserveStart(), originalDataSize + originalReservedSize);
if (cloneHead.IsNull())
{
cloneHead = std::move(clone);
}
else
{
cloneHead->AddToEnd(std::move(clone));
}
}
return cloneHead;
}
} // namespace System
namespace Encoding {
System::PacketBufferHandle PacketBufferWriterUtil::Finalize(BufferWriter & aBufferWriter, System::PacketBufferHandle & aPacket)
{
if (!aPacket.IsNull() && aBufferWriter.Fit())
{
// Since mPacket was successfully allocated to hold the maximum length,
// we know that the actual length fits in a uint16_t.
aPacket->SetDataLength(static_cast<uint16_t>(aBufferWriter.Needed()));
}
else
{
aPacket = nullptr;
}
aBufferWriter = Encoding::BufferWriter(nullptr, 0);
return std::move(aPacket);
}
} // namespace Encoding
} // namespace chip