X-Git-Url: http://plrg.eecs.uci.edu/git/?p=folly.git;a=blobdiff_plain;f=folly%2Fio%2FCursor.h;h=838d366bb52257222d89116bf468c472bde2a2ed;hp=79fd4cc789f0c4a3b2b70d92f1142f774cd1d42e;hb=afde52ff7655ba79759eafdbee14b3fe47428fce;hpb=820c91a0f9cb1d5c7938a6e9394e625781919772 diff --git a/folly/io/Cursor.h b/folly/io/Cursor.h index 79fd4cc7..838d366b 100644 --- a/folly/io/Cursor.h +++ b/folly/io/Cursor.h @@ -1,5 +1,5 @@ /* - * Copyright 2013 Facebook, Inc. + * Copyright 2013-present Facebook, Inc. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -13,20 +13,23 @@ * See the License for the specific language governing permissions and * limitations under the License. */ +#pragma once -#ifndef FOLLY_CURSOR_H -#define FOLLY_CURSOR_H - -#include +#include +#include +#include +#include #include -#include #include -#include -#include "folly/Bits.h" -#include "folly/io/IOBuf.h" -#include "folly/io/IOBufQueue.h" -#include "folly/Likely.h" +#include +#include +#include +#include +#include +#include +#include +#include /** * Cursor class for fast iteration over IOBuf chains. @@ -35,7 +38,7 @@ * * RWPrivateCursor - Read-write access, assumes private access to IOBuf chain * RWUnshareCursor - Read-write access, calls unshare on write (COW) - * Appender - Write access, assumes private access to IOBuf chian + * Appender - Write access, assumes private access to IOBuf chain * * Note that RW cursors write in the preallocated part of buffers (that is, * between the buffer's data() and tail()), while Appenders append to the end @@ -44,21 +47,152 @@ * Appender with a buffer chain; for this reason, Appenders assume private * access to the buffer (you need to call unshare() yourself if necessary). **/ -namespace folly { namespace io { +namespace folly { +namespace io { + namespace detail { -template +template class CursorBase { + // Make all the templated classes friends for copy constructor. + template friend class CursorBase; public: + explicit CursorBase(BufType* buf) : crtBuf_(buf), buffer_(buf) { + if (crtBuf_) { + crtPos_ = crtBegin_ = crtBuf_->data(); + crtEnd_ = crtBuf_->tail(); + } + } + + /** + * Copy constructor. + * + * This also allows constructing a CursorBase from other derived types. + * For instance, this allows constructing a Cursor from an RWPrivateCursor. + */ + template + explicit CursorBase(const CursorBase& cursor) + : crtBuf_(cursor.crtBuf_), + buffer_(cursor.buffer_), + crtBegin_(cursor.crtBegin_), + crtEnd_(cursor.crtEnd_), + crtPos_(cursor.crtPos_), + absolutePos_(cursor.absolutePos_) {} + + /** + * Reset cursor to point to a new buffer. + */ + void reset(BufType* buf) { + crtBuf_ = buf; + buffer_ = buf; + absolutePos_ = 0; + if (crtBuf_) { + crtPos_ = crtBegin_ = crtBuf_->data(); + crtEnd_ = crtBuf_->tail(); + } + } + + /** + * Get the current Cursor position relative to the head of IOBuf chain. + */ + size_t getCurrentPosition() const { + return (crtPos_ - crtBegin_) + absolutePos_; + } + const uint8_t* data() const { - return crtBuf_->data() + offset_; + return crtPos_; } - // Space available in the current IOBuf. May be 0; use peek() instead which - // will always point to a non-empty chunk of data or at the end of the - // chain. + /** + * Return the remaining space available in the current IOBuf. + * + * May return 0 if the cursor is at the end of an IOBuf. Use peekBytes() + * instead if you want to avoid this. peekBytes() will advance to the next + * non-empty IOBuf (up to the end of the chain) if the cursor is currently + * pointing at the end of a buffer. + */ size_t length() const { - return crtBuf_->length() - offset_; + return crtEnd_ - crtPos_; + } + + /** + * Return the space available until the end of the entire IOBuf chain. + */ + size_t totalLength() const { + if (crtBuf_ == buffer_) { + return crtBuf_->computeChainDataLength() - (crtPos_ - crtBegin_); + } + CursorBase end(buffer_->prev()); + end.crtPos_ = end.crtEnd_; + return end - *this; + } + + /** + * Return true if the cursor could advance the specified number of bytes + * from its current position. + * This is useful for applications that want to do checked reads instead of + * catching exceptions and is more efficient than using totalLength as it + * walks the minimal set of buffers in the chain to determine the result. + */ + bool canAdvance(size_t amount) const { + const IOBuf* nextBuf = crtBuf_; + size_t available = length(); + do { + if (available >= amount) { + return true; + } + amount -= available; + nextBuf = nextBuf->next(); + available = nextBuf->length(); + } while (nextBuf != buffer_); + return false; + } + + /* + * Return true if the cursor is at the end of the entire IOBuf chain. + */ + bool isAtEnd() const { + // Check for the simple cases first. + if (crtPos_ != crtEnd_) { + return false; + } + if (crtBuf_ == buffer_->prev()) { + return true; + } + // We are at the end of a buffer, but it isn't the last buffer. + // We might still be at the end if the remaining buffers in the chain are + // empty. + const IOBuf* buf = crtBuf_->next();; + while (buf != buffer_) { + if (buf->length() > 0) { + return false; + } + buf = buf->next(); + } + return true; + } + + /** + * Advances the cursor to the end of the entire IOBuf chain. + */ + void advanceToEnd() { + // Simple case, we're already in the last IOBuf. + if (crtBuf_ == buffer_->prev()) { + crtPos_ = crtEnd_; + return; + } + + auto* nextBuf = crtBuf_->next(); + while (nextBuf != buffer_) { + absolutePos_ += crtEnd_ - crtBegin_; + + crtBuf_ = nextBuf; + nextBuf = crtBuf_->next(); + crtBegin_ = crtBuf_->data(); + crtPos_ = crtEnd_ = crtBuf_->tail(); + + static_cast(this)->advanceDone(); + } } Derived& operator+=(size_t offset) { @@ -66,13 +200,86 @@ class CursorBase { p->skip(offset); return *p; } + Derived operator+(size_t offset) const { + Derived other(*this); + other.skip(offset); + return other; + } + + Derived& operator-=(size_t offset) { + Derived* p = static_cast(this); + p->retreat(offset); + return *p; + } + Derived operator-(size_t offset) const { + Derived other(*this); + other.retreat(offset); + return other; + } + + /** + * Compare cursors for equality/inequality. + * + * Two cursors are equal if they are pointing to the same location in the + * same IOBuf chain. + */ + bool operator==(const Derived& other) const { + const IOBuf* crtBuf = crtBuf_; + auto crtPos = crtPos_; + // We can be pointing to the end of a buffer chunk, find first non-empty. + while (crtPos == crtBuf->tail() && crtBuf != buffer_->prev()) { + crtBuf = crtBuf->next(); + crtPos = crtBuf->data(); + } + + const IOBuf* crtBufOther = other.crtBuf_; + auto crtPosOther = other.crtPos_; + // We can be pointing to the end of a buffer chunk, find first non-empty. + while (crtPosOther == crtBufOther->tail() && + crtBufOther != other.buffer_->prev()) { + crtBufOther = crtBufOther->next(); + crtPosOther = crtBufOther->data(); + } + return (crtPos == crtPosOther) && (crtBuf == crtBufOther); + } + bool operator!=(const Derived& other) const { + return !operator==(other); + } template - typename std::enable_if::value, T>::type - read() { - T val; - pull(&val, sizeof(T)); - return val; + typename std::enable_if::value, bool>::type tryRead( + T& val) { + if (LIKELY(crtPos_ + sizeof(T) <= crtEnd_)) { + val = loadUnaligned(data()); + crtPos_ += sizeof(T); + return true; + } + return pullAtMostSlow(&val, sizeof(T)) == sizeof(T); + } + + template + bool tryReadBE(T& val) { + const bool result = tryRead(val); + val = Endian::big(val); + return result; + } + + template + bool tryReadLE(T& val) { + const bool result = tryRead(val); + val = Endian::little(val); + return result; + } + + template + T read() { + if (LIKELY(crtPos_ + sizeof(T) <= crtEnd_)) { + T val = loadUnaligned(data()); + crtPos_ += sizeof(T); + return val; + } else { + return readSlow(); + } } template @@ -94,23 +301,14 @@ class CursorBase { */ std::string readFixedString(size_t len) { std::string str; - str.reserve(len); - for (;;) { - // Fast path: it all fits in one buffer. - size_t available = length(); - if (LIKELY(available >= len)) { - str.append(reinterpret_cast(data()), len); - offset_ += len; - return str; - } - - str.append(reinterpret_cast(data()), available); - if (UNLIKELY(!tryAdvanceBuffer())) { - throw std::out_of_range("string underflow"); - } - len -= available; + if (LIKELY(length() >= len)) { + str.append(reinterpret_cast(data()), len); + crtPos_ += len; + } else { + readFixedStringSlow(&str, len); } + return str; } /** @@ -122,143 +320,168 @@ class CursorBase { * vs. using pull(). */ std::string readTerminatedString( - char termChar = '\0', - size_t maxLength = std::numeric_limits::max()) { - std::string str; - - for (;;) { - const uint8_t* buf = data(); - size_t buflen = length(); + char termChar = '\0', + size_t maxLength = std::numeric_limits::max()); - size_t i = 0; - while (i < buflen && buf[i] != termChar) { - ++i; + /* + * Read all bytes until the specified predicate returns true. + * + * The predicate will be called on each byte in turn, until it returns false + * or until the end of the IOBuf chain is reached. + * + * Returns the result as a string. + */ + template + std::string readWhile(const Predicate& predicate); - // Do this check after incrementing 'i', as even though we start at the - // 0 byte, it still represents a single character - if (str.length() + i >= maxLength) { - throw std::length_error("string overflow"); - } - } + /* + * Read all bytes until the specified predicate returns true. + * + * This is a more generic version of readWhile() takes an arbitrary Output + * object, and calls Output::append() with each chunk of matching data. + */ + template + void readWhile(const Predicate& predicate, Output& out); - str.append(reinterpret_cast(buf), i); - if (i < buflen) { - skip(i + 1); - return str; - } + /* + * Skip all bytes until the specified predicate returns true. + * + * The predicate will be called on each byte in turn, until it returns false + * or until the end of the IOBuf chain is reached. + */ + template + void skipWhile(const Predicate& predicate); - skip(i); + size_t skipAtMost(size_t len) { + if (LIKELY(crtPos_ + len < crtEnd_)) { + crtPos_ += len; + return len; + } + return skipAtMostSlow(len); + } - if (UNLIKELY(!tryAdvanceBuffer())) { - throw std::out_of_range("string underflow"); - } + void skip(size_t len) { + if (LIKELY(crtPos_ + len < crtEnd_)) { + crtPos_ += len; + } else { + skipSlow(len); } } - explicit CursorBase(BufType* buf) - : crtBuf_(buf) - , offset_(0) - , buffer_(buf) {} + /** + * Skip bytes in the current IOBuf without advancing to the next one. + * Precondition: length() >= len + */ + void skipNoAdvance(size_t len) { + DCHECK_LE(len, length()); + crtPos_ += len; + } - // Make all the templated classes friends for copy constructor. - template friend class CursorBase; + size_t retreatAtMost(size_t len) { + if (len <= static_cast(crtPos_ - crtBegin_)) { + crtPos_ -= len; + return len; + } + return retreatAtMostSlow(len); + } - template - explicit CursorBase(const T& cursor) { - crtBuf_ = cursor.crtBuf_; - offset_ = cursor.offset_; - buffer_ = cursor.buffer_; + void retreat(size_t len) { + if (len <= static_cast(crtPos_ - crtBegin_)) { + crtPos_ -= len; + } else { + retreatSlow(len); + } } - // reset cursor to point to a new buffer. - void reset(BufType* buf) { - crtBuf_ = buf; - buffer_ = buf; - offset_ = 0; + size_t pullAtMost(void* buf, size_t len) { + // Fast path: it all fits in one buffer. + if (LIKELY(crtPos_ + len <= crtEnd_)) { + memcpy(buf, data(), len); + crtPos_ += len; + return len; + } + return pullAtMostSlow(buf, len); + } + + void pull(void* buf, size_t len) { + if (LIKELY(crtPos_ + len <= crtEnd_)) { + memcpy(buf, data(), len); + crtPos_ += len; + } else { + pullSlow(buf, len); + } } /** * Return the available data in the current buffer. * If you want to gather more data from the chain into a contiguous region - * (for hopefully zero-copy access), use gather() before peek(). + * (for hopefully zero-copy access), use gather() before peekBytes(). */ - std::pair peek() { + ByteRange peekBytes() { // Ensure that we're pointing to valid data size_t available = length(); while (UNLIKELY(available == 0 && tryAdvanceBuffer())) { available = length(); } - - return std::make_pair(data(), available); + return ByteRange{data(), available}; } - void pull(void* buf, size_t len) { - if (UNLIKELY(pullAtMost(buf, len) != len)) { - throw std::out_of_range("underflow"); - } + /** + * Alternate version of peekBytes() that returns a std::pair + * instead of a ByteRange. (This method pre-dates ByteRange.) + * + * This function will eventually be deprecated. + */ + std::pair peek() { + auto bytes = peekBytes(); + return std::make_pair(bytes.data(), bytes.size()); } void clone(std::unique_ptr& buf, size_t len) { if (UNLIKELY(cloneAtMost(buf, len) != len)) { - throw std::out_of_range("underflow"); - } - } - - void skip(size_t len) { - if (UNLIKELY(skipAtMost(len) != len)) { - throw std::out_of_range("underflow"); + std::__throw_out_of_range("underflow"); } } - size_t pullAtMost(void* buf, size_t len) { - uint8_t* p = reinterpret_cast(buf); - size_t copied = 0; - for (;;) { - // Fast path: it all fits in one buffer. - size_t available = length(); - if (LIKELY(available >= len)) { - memcpy(p, data(), len); - offset_ += len; - return copied + len; - } - - memcpy(p, data(), available); - copied += available; - if (UNLIKELY(!tryAdvanceBuffer())) { - return copied; - } - p += available; - len -= available; + void clone(folly::IOBuf& buf, size_t len) { + if (UNLIKELY(cloneAtMost(buf, len) != len)) { + std::__throw_out_of_range("underflow"); } } - size_t cloneAtMost(std::unique_ptr& buf, size_t len) { - buf.reset(nullptr); + size_t cloneAtMost(folly::IOBuf& buf, size_t len) { + // We might be at the end of buffer. + advanceBufferIfEmpty(); std::unique_ptr tmp; size_t copied = 0; - for (;;) { + for (int loopCount = 0; true; ++loopCount) { // Fast path: it all fits in one buffer. size_t available = length(); if (LIKELY(available >= len)) { - tmp = crtBuf_->cloneOne(); - tmp->trimStart(offset_); - tmp->trimEnd(tmp->length() - len); - offset_ += len; - if (!buf) { - buf = std::move(tmp); + if (loopCount == 0) { + crtBuf_->cloneOneInto(buf); + buf.trimStart(crtPos_ - crtBegin_); + buf.trimEnd(buf.length() - len); } else { - buf->prependChain(std::move(tmp)); + tmp = crtBuf_->cloneOne(); + tmp->trimStart(crtPos_ - crtBegin_); + tmp->trimEnd(tmp->length() - len); + buf.prependChain(std::move(tmp)); } + + crtPos_ += len; + advanceBufferIfEmpty(); return copied + len; } - tmp = crtBuf_->cloneOne(); - tmp->trimStart(offset_); - if (!buf) { - buf = std::move(tmp); + if (loopCount == 0) { + crtBuf_->cloneOneInto(buf); + buf.trimStart(crtPos_ - crtBegin_); } else { - buf->prependChain(std::move(tmp)); + tmp = crtBuf_->cloneOne(); + tmp->trimStart(crtPos_ - crtBegin_); + buf.prependChain(std::move(tmp)); } copied += available; @@ -269,22 +492,11 @@ class CursorBase { } } - size_t skipAtMost(size_t len) { - size_t skipped = 0; - for (;;) { - // Fast path: it all fits in one buffer. - size_t available = length(); - if (LIKELY(available >= len)) { - offset_ += len; - return skipped + len; - } - - skipped += available; - if (UNLIKELY(!tryAdvanceBuffer())) { - return skipped; - } - len -= available; + size_t cloneAtMost(std::unique_ptr& buf, size_t len) { + if (!buf) { + buf = std::make_unique(); } + return cloneAtMost(*buf, len); } /** @@ -295,7 +507,7 @@ class CursorBase { size_t len = 0; if (otherBuf != crtBuf_) { - len += otherBuf->length() - other.offset_; + len += other.crtEnd_ - other.crtPos_; for (otherBuf = otherBuf->next(); otherBuf != crtBuf_ && otherBuf != other.buffer_; @@ -304,16 +516,16 @@ class CursorBase { } if (otherBuf == other.buffer_) { - throw std::out_of_range("wrap-around"); + std::__throw_out_of_range("wrap-around"); } - len += offset_; + len += crtPos_ - crtBegin_; } else { - if (offset_ < other.offset_) { - throw std::out_of_range("underflow"); + if (crtPos_ < other.crtPos_) { + std::__throw_out_of_range("underflow"); } - len += offset_ - other.offset_; + len += crtPos_ - other.crtPos_; } return len; @@ -325,85 +537,249 @@ class CursorBase { size_t operator-(const BufType* buf) const { size_t len = 0; - BufType *curBuf = buf; + const BufType* curBuf = buf; while (curBuf != crtBuf_) { len += curBuf->length(); curBuf = curBuf->next(); if (curBuf == buf || curBuf == buffer_) { - throw std::out_of_range("wrap-around"); + std::__throw_out_of_range("wrap-around"); } } - len += offset_; + len += crtPos_ - crtBegin_; return len; } protected: - BufType* crtBuf_; - size_t offset_; + ~CursorBase() { } - ~CursorBase(){} + BufType* head() { + return buffer_; + } bool tryAdvanceBuffer() { BufType* nextBuf = crtBuf_->next(); if (UNLIKELY(nextBuf == buffer_)) { - offset_ = crtBuf_->length(); + crtPos_ = crtEnd_; return false; } - offset_ = 0; + absolutePos_ += crtEnd_ - crtBegin_; crtBuf_ = nextBuf; + crtPos_ = crtBegin_ = crtBuf_->data(); + crtEnd_ = crtBuf_->tail(); + static_cast(this)->advanceDone(); + return true; + } + + bool tryRetreatBuffer() { + if (UNLIKELY(crtBuf_ == buffer_)) { + crtPos_ = crtBegin_; + return false; + } + crtBuf_ = crtBuf_->prev(); + crtBegin_ = crtBuf_->data(); + crtPos_ = crtEnd_ = crtBuf_->tail(); + absolutePos_ -= crtEnd_ - crtBegin_; static_cast(this)->advanceDone(); return true; } + void advanceBufferIfEmpty() { + if (crtPos_ == crtEnd_) { + tryAdvanceBuffer(); + } + } + + BufType* crtBuf_; + BufType* buffer_; + const uint8_t* crtBegin_{nullptr}; + const uint8_t* crtEnd_{nullptr}; + const uint8_t* crtPos_{nullptr}; + size_t absolutePos_{0}; + private: + template + FOLLY_NOINLINE T readSlow() { + T val; + pullSlow(&val, sizeof(T)); + return val; + } + + void readFixedStringSlow(std::string* str, size_t len) { + for (size_t available; (available = length()) < len; ) { + str->append(reinterpret_cast(data()), available); + if (UNLIKELY(!tryAdvanceBuffer())) { + std::__throw_out_of_range("string underflow"); + } + len -= available; + } + str->append(reinterpret_cast(data()), len); + crtPos_ += len; + advanceBufferIfEmpty(); + } + + size_t pullAtMostSlow(void* buf, size_t len) { + uint8_t* p = reinterpret_cast(buf); + size_t copied = 0; + for (size_t available; (available = length()) < len; ) { + memcpy(p, data(), available); + copied += available; + if (UNLIKELY(!tryAdvanceBuffer())) { + return copied; + } + p += available; + len -= available; + } + memcpy(p, data(), len); + crtPos_ += len; + advanceBufferIfEmpty(); + return copied + len; + } + + void pullSlow(void* buf, size_t len) { + if (UNLIKELY(pullAtMostSlow(buf, len) != len)) { + std::__throw_out_of_range("underflow"); + } + } + + size_t skipAtMostSlow(size_t len) { + size_t skipped = 0; + for (size_t available; (available = length()) < len; ) { + skipped += available; + if (UNLIKELY(!tryAdvanceBuffer())) { + return skipped; + } + len -= available; + } + crtPos_ += len; + advanceBufferIfEmpty(); + return skipped + len; + } + + void skipSlow(size_t len) { + if (UNLIKELY(skipAtMostSlow(len) != len)) { + std::__throw_out_of_range("underflow"); + } + } + + size_t retreatAtMostSlow(size_t len) { + size_t retreated = 0; + for (size_t available; (available = crtPos_ - crtBegin_) < len;) { + retreated += available; + if (UNLIKELY(!tryRetreatBuffer())) { + return retreated; + } + len -= available; + } + crtPos_ -= len; + return retreated + len; + } + + void retreatSlow(size_t len) { + if (UNLIKELY(retreatAtMostSlow(len) != len)) { + std::__throw_out_of_range("underflow"); + } + } + void advanceDone() { } +}; - BufType* buffer_; +} // namespace detail + +class Cursor : public detail::CursorBase { + public: + explicit Cursor(const IOBuf* buf) + : detail::CursorBase(buf) {} + + template + explicit Cursor(const detail::CursorBase& cursor) + : detail::CursorBase(cursor) {} }; +namespace detail { + template class Writable { public: template - typename std::enable_if::value>::type + typename std::enable_if::value>::type write(T value) { const uint8_t* u8 = reinterpret_cast(&value); - push(u8, sizeof(T)); + Derived* d = static_cast(this); + d->push(u8, sizeof(T)); } template void writeBE(T value) { - write(Endian::big(value)); + Derived* d = static_cast(this); + d->write(Endian::big(value)); } template void writeLE(T value) { - write(Endian::little(value)); + Derived* d = static_cast(this); + d->write(Endian::little(value)); } void push(const uint8_t* buf, size_t len) { Derived* d = static_cast(this); if (d->pushAtMost(buf, len) != len) { - throw std::out_of_range("overflow"); + std::__throw_out_of_range("overflow"); } } -}; -} // namespace detail + void push(ByteRange buf) { + if (this->pushAtMost(buf) != buf.size()) { + std::__throw_out_of_range("overflow"); + } + } -class Cursor : public detail::CursorBase { - public: - explicit Cursor(const IOBuf* buf) - : detail::CursorBase(buf) {} + size_t pushAtMost(ByteRange buf) { + Derived* d = static_cast(this); + return d->pushAtMost(buf.data(), buf.size()); + } - template - explicit Cursor(CursorType& cursor) - : detail::CursorBase(cursor) {} + /** + * push len bytes of data from input cursor, data could be in an IOBuf chain. + * If input cursor contains less than len bytes, or this cursor has less than + * len bytes writable space, an out_of_range exception will be thrown. + */ + void push(Cursor cursor, size_t len) { + if (this->pushAtMost(cursor, len) != len) { + std::__throw_out_of_range("overflow"); + } + } + + size_t pushAtMost(Cursor cursor, size_t len) { + size_t written = 0; + for(;;) { + auto currentBuffer = cursor.peekBytes(); + const uint8_t* crtData = currentBuffer.data(); + size_t available = currentBuffer.size(); + if (available == 0) { + // end of buffer chain + return written; + } + // all data is in current buffer + if (available >= len) { + this->push(crtData, len); + cursor.skip(len); + return written + len; + } + + // write the whole current IOBuf + this->push(crtData, available); + cursor.skip(available); + written += available; + len -= available; + } + } }; +} // namespace detail + enum class CursorAccess { PRIVATE, UNSHARE @@ -419,8 +795,8 @@ class RWCursor : detail::CursorBase, IOBuf>(buf), maybeShared_(true) {} - template - explicit RWCursor(CursorType& cursor) + template + explicit RWCursor(const detail::CursorBase& cursor) : detail::CursorBase, IOBuf>(cursor), maybeShared_(true) {} /** @@ -428,10 +804,42 @@ class RWCursor * by coalescing subsequent buffers from the chain as necessary. */ void gather(size_t n) { - this->crtBuf_->gather(this->offset_ + n); + // Forbid attempts to gather beyond the end of this IOBuf chain. + // Otherwise we could try to coalesce the head of the chain and end up + // accidentally freeing it, invalidating the pointer owned by external + // code. + // + // If crtBuf_ == head() then IOBuf::gather() will perform all necessary + // checking. We only have to perform an explicit check here when calling + // gather() on a non-head element. + if (this->crtBuf_ != this->head() && this->totalLength() < n) { + throw std::overflow_error("cannot gather() past the end of the chain"); + } + size_t offset = this->crtPos_ - this->crtBegin_; + this->crtBuf_->gather(offset + n); + this->crtBegin_ = this->crtBuf_->data(); + this->crtEnd_ = this->crtBuf_->tail(); + this->crtPos_ = this->crtBegin_ + offset; + } + void gatherAtMost(size_t n) { + size_t size = std::min(n, this->totalLength()); + size_t offset = this->crtPos_ - this->crtBegin_; + this->crtBuf_->gather(offset + size); + this->crtBegin_ = this->crtBuf_->data(); + this->crtEnd_ = this->crtBuf_->tail(); + this->crtPos_ = this->crtBegin_ + offset; } + using detail::Writable>::pushAtMost; size_t pushAtMost(const uint8_t* buf, size_t len) { + // We have to explicitly check for an input length of 0. + // We support buf being nullptr in this case, but we need to avoid calling + // memcpy() with a null source pointer, since that is undefined behavior + // even if the length is 0. + if (len == 0) { + return 0; + } + size_t copied = 0; for (;;) { // Fast path: the current buffer is big enough. @@ -441,7 +849,7 @@ class RWCursor maybeUnshare(); } memcpy(writableData(), buf, len); - this->offset_ += len; + this->crtPos_ += len; return copied + len; } @@ -459,17 +867,17 @@ class RWCursor } void insert(std::unique_ptr buf) { - folly::IOBuf* nextBuf; - if (this->offset_ == 0) { + this->absolutePos_ += buf->computeChainDataLength(); + if (this->crtPos_ == this->crtBegin_ && this->crtBuf_ != this->buffer_) { // Can just prepend - nextBuf = buf.get(); this->crtBuf_->prependChain(std::move(buf)); } else { + IOBuf* nextBuf; std::unique_ptr remaining; - if (this->crtBuf_->length() - this->offset_ > 0) { + if (this->crtPos_ != this->crtEnd_) { // Need to split current IOBuf in two. remaining = this->crtBuf_->cloneOne(); - remaining->trimStart(this->offset_); + remaining->trimStart(this->crtPos_ - this->crtBegin_); nextBuf = remaining.get(); buf->prependChain(std::move(remaining)); } else { @@ -477,21 +885,28 @@ class RWCursor nextBuf = this->crtBuf_->next(); } this->crtBuf_->trimEnd(this->length()); + this->absolutePos_ += this->crtPos_ - this->crtBegin_; this->crtBuf_->appendChain(std::move(buf)); + + // Jump past the new links + this->crtBuf_ = nextBuf; + this->crtPos_ = this->crtBegin_ = this->crtBuf_->data(); + this->crtEnd_ = this->crtBuf_->tail(); } - // Jump past the new links - this->offset_ = 0; - this->crtBuf_ = nextBuf; } uint8_t* writableData() { - return this->crtBuf_->writableData() + this->offset_; + return this->crtBuf_->writableData() + (this->crtPos_ - this->crtBegin_); } private: void maybeUnshare() { if (UNLIKELY(maybeShared_)) { + size_t offset = this->crtPos_ - this->crtBegin_; this->crtBuf_->unshareOne(); + this->crtBegin_ = this->crtBuf_->data(); + this->crtEnd_ = this->crtBuf_->tail(); + this->crtPos_ = this->crtBegin_ + offset; maybeShared_ = false; } } @@ -516,7 +931,7 @@ typedef RWCursor RWUnshareCursor; */ class Appender : public detail::Writable { public: - Appender(IOBuf* buf, uint32_t growth) + Appender(IOBuf* buf, uint64_t growth) : buffer_(buf), crtBuf_(buf->prev()), growth_(growth) { @@ -542,7 +957,7 @@ class Appender : public detail::Writable { * Ensure at least n contiguous bytes available to write. * Postcondition: length() >= n. */ - void ensure(uint32_t n) { + void ensure(uint64_t n) { if (LIKELY(length() >= n)) { return; } @@ -550,7 +965,7 @@ class Appender : public detail::Writable { // Waste the rest of the current buffer and allocate a new one. // Don't make it too small, either. if (growth_ == 0) { - throw std::out_of_range("can't grow buffer chain"); + std::__throw_out_of_range("can't grow buffer chain"); } n = std::max(n, growth_); @@ -558,7 +973,16 @@ class Appender : public detail::Writable { crtBuf_ = buffer_->prev(); } + using detail::Writable::pushAtMost; size_t pushAtMost(const uint8_t* buf, size_t len) { + // We have to explicitly check for an input length of 0. + // We support buf being nullptr in this case, but we need to avoid calling + // memcpy() with a null source pointer, since that is undefined behavior + // even if the length is 0. + if (len == 0) { + return 0; + } + size_t copied = 0; for (;;) { // Fast path: it all fits in one buffer. @@ -580,6 +1004,42 @@ class Appender : public detail::Writable { } } + /* + * Append to the end of this buffer, using a printf() style + * format specifier. + * + * Note that folly/Format.h provides nicer and more type-safe mechanisms + * for formatting strings, which should generally be preferred over + * printf-style formatting. Appender objects can be used directly as an + * output argument for Formatter objects. For example: + * + * Appender app(&iobuf); + * format("{} {}", "hello", "world")(app); + * + * However, printf-style strings are still needed when dealing with existing + * third-party code in some cases. + * + * This will always add a nul-terminating character after the end + * of the output. However, the buffer data length will only be updated to + * include the data itself. The nul terminator will be the first byte in the + * buffer tailroom. + * + * This method may throw exceptions on error. + */ + void printf(FOLLY_PRINTF_FORMAT const char* fmt, ...) + FOLLY_PRINTF_FORMAT_ATTR(2, 3); + + void vprintf(const char* fmt, va_list ap); + + /* + * Calling an Appender object with a StringPiece will append the string + * piece. This allows Appender objects to be used directly with + * Formatter. + */ + void operator()(StringPiece sp) { + push(ByteRange(sp)); + } + private: bool tryGrowChain() { assert(crtBuf_->next() == buffer_); @@ -594,7 +1054,7 @@ class Appender : public detail::Writable { IOBuf* buffer_; IOBuf* crtBuf_; - uint32_t growth_; + uint64_t growth_; }; class QueueAppender : public detail::Writable { @@ -603,92 +1063,99 @@ class QueueAppender : public detail::Writable { * Create an Appender that writes to a IOBufQueue. When we allocate * space in the queue, we grow no more than growth bytes at once * (unless you call ensure() with a bigger value yourself). - * Throw if we ever need to allocate more than maxTotalGrowth. */ - QueueAppender(IOBufQueue* queue, - uint32_t growth, - size_t maxTotalGrowth = std::numeric_limits::max()) { - reset(queue, growth, maxTotalGrowth); - } + QueueAppender(IOBufQueue* queue, uint64_t growth) + : queueCache_(queue), growth_(growth) {} - void reset(IOBufQueue* queue, - uint32_t growth, - size_t maxTotalGrowth = std::numeric_limits::max()) { - queue_ = queue; + void reset(IOBufQueue* queue, uint64_t growth) { + queueCache_.reset(queue); growth_ = growth; - remainingGrowth_ = maxTotalGrowth; - next_ = nullptr; - available_ = 0; } - uint8_t* writableData() { return next_; } + uint8_t* writableData() { + return queueCache_.writableData(); + } - size_t length() const { return available_; } + size_t length() { + return queueCache_.length(); + } void append(size_t n) { - assert(n <= available_); - assert(n <= remainingGrowth_); - queue_->postallocate(n); - next_ += n; - available_ -= n; - remainingGrowth_ -= n; + queueCache_.append(n); } // Ensure at least n contiguous; can go above growth_, throws if // not enough room. - void ensure(uint32_t n) { - if (UNLIKELY(n > remainingGrowth_)) { - throw std::out_of_range("overflow"); + void ensure(size_t n) { + if (length() < n) { + ensureSlow(n); } + } - if (LIKELY(length() >= n)) { - return; + template + typename std::enable_if::value>::type write(T value) { + // We can't fail. + if (length() >= sizeof(T)) { + storeUnaligned(queueCache_.writableData(), value); + queueCache_.appendUnsafe(sizeof(T)); + } else { + writeSlow(value); } - - size_t desired = std::min(growth_, remainingGrowth_ - n); - - // Grab some more. - auto p = queue_->preallocate(n, desired); - - next_ = static_cast(p.first); - available_ = p.second; } + using detail::Writable::pushAtMost; size_t pushAtMost(const uint8_t* buf, size_t len) { - if (UNLIKELY(len > remainingGrowth_)) { - len = remainingGrowth_; + // Fill the current buffer + const size_t copyLength = std::min(len, length()); + if (copyLength != 0) { + memcpy(writableData(), buf, copyLength); + queueCache_.appendUnsafe(copyLength); + buf += copyLength; } - - size_t remaining = len; + size_t remaining = len - copyLength; + // Allocate more buffers as necessary while (remaining != 0) { - ensure(std::min(remaining, growth_)); - size_t n = std::min(remaining, available_); - memcpy(next_, buf, n); - buf += n; - remaining -= n; - append(n); + auto p = queueCache_.queue()->preallocate( + std::min(remaining, growth_), growth_, remaining); + memcpy(p.first, buf, p.second); + queueCache_.queue()->postallocate(p.second); + buf += p.second; + remaining -= p.second; } - return len; } - // insert doesn't count towards maxTotalGrowth void insert(std::unique_ptr buf) { if (buf) { - queue_->append(std::move(buf), true); - next_ = nullptr; - available_ = 0; + queueCache_.queue()->append(std::move(buf), true); } } + void insert(const folly::IOBuf& buf) { + insert(buf.clone()); + } + private: - folly::IOBufQueue* queue_; - size_t growth_; - size_t remainingGrowth_; - uint8_t* next_; - size_t available_; + folly::IOBufQueue::WritableRangeCache queueCache_{nullptr}; + size_t growth_{0}; + + FOLLY_NOINLINE void ensureSlow(size_t n) { + queueCache_.queue()->preallocate(n, growth_); + queueCache_.fillCache(); + } + + template + typename std::enable_if::value>::type FOLLY_NOINLINE + writeSlow(T value) { + queueCache_.queue()->preallocate(sizeof(T), growth_); + queueCache_.fillCache(); + + storeUnaligned(queueCache_.writableData(), value); + queueCache_.appendUnsafe(sizeof(T)); + } }; -}} // folly::io +} // namespace io +} // namespace folly -#endif // FOLLY_CURSOR_H +#include