Revision fe2bd190a5b5a5b9c8ae224745a8f6b442202817 authored by Siying Dong on 26 March 2019, 17:43:22 UTC, committed by Facebook Github Bot on 26 March 2019, 17:53:19 UTC
Summary: Right now, BlobDB::Open() fails to put all trash files to delete scheduler, which causes some trash files permanently untracked. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5103 Differential Revision: D14606095 Pulled By: siying fbshipit-source-id: 41a9437a2948abb235c0ed85f9a04612d0e50183
1 parent 75133b1
io_posix.cc
// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
// This source code is licensed under both the GPLv2 (found in the
// COPYING file in the root directory) and Apache 2.0 License
// (found in the LICENSE.Apache file in the root directory).
//
// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file. See the AUTHORS file for names of contributors.
#ifdef ROCKSDB_LIB_IO_POSIX
#include "env/io_posix.h"
#include <errno.h>
#include <fcntl.h>
#include <algorithm>
#if defined(OS_LINUX)
#include <linux/fs.h>
#endif
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/ioctl.h>
#include <sys/mman.h>
#include <sys/stat.h>
#include <sys/types.h>
#ifdef OS_LINUX
#include <sys/statfs.h>
#include <sys/syscall.h>
#include <sys/sysmacros.h>
#endif
#include "env/posix_logger.h"
#include "monitoring/iostats_context_imp.h"
#include "port/port.h"
#include "rocksdb/slice.h"
#include "util/coding.h"
#include "util/string_util.h"
#include "util/sync_point.h"
#if defined(OS_LINUX) && !defined(F_SET_RW_HINT)
#define F_LINUX_SPECIFIC_BASE 1024
#define F_SET_RW_HINT (F_LINUX_SPECIFIC_BASE + 12)
#endif
namespace rocksdb {
// A wrapper for fadvise, if the platform doesn't support fadvise,
// it will simply return 0.
int Fadvise(int fd, off_t offset, size_t len, int advice) {
#ifdef OS_LINUX
return posix_fadvise(fd, offset, len, advice);
#else
(void)fd;
(void)offset;
(void)len;
(void)advice;
return 0; // simply do nothing.
#endif
}
namespace {
size_t GetLogicalBufferSize(int __attribute__((__unused__)) fd) {
#ifdef OS_LINUX
struct stat buf;
int result = fstat(fd, &buf);
if (result == -1) {
return kDefaultPageSize;
}
if (major(buf.st_dev) == 0) {
// Unnamed devices (e.g. non-device mounts), reserved as null device number.
// These don't have an entry in /sys/dev/block/. Return a sensible default.
return kDefaultPageSize;
}
// Reading queue/logical_block_size does not require special permissions.
const int kBufferSize = 100;
char path[kBufferSize];
char real_path[PATH_MAX + 1];
snprintf(path, kBufferSize, "/sys/dev/block/%u:%u", major(buf.st_dev),
minor(buf.st_dev));
if (realpath(path, real_path) == nullptr) {
return kDefaultPageSize;
}
std::string device_dir(real_path);
if (!device_dir.empty() && device_dir.back() == '/') {
device_dir.pop_back();
}
// NOTE: sda3 and nvme0n1p1 do not have a `queue/` subdir, only the parent sda
// and nvme0n1 have it.
// $ ls -al '/sys/dev/block/8:3'
// lrwxrwxrwx. 1 root root 0 Jun 26 01:38 /sys/dev/block/8:3 ->
// ../../block/sda/sda3
// $ ls -al '/sys/dev/block/259:4'
// lrwxrwxrwx 1 root root 0 Jan 31 16:04 /sys/dev/block/259:4 ->
// ../../devices/pci0000:17/0000:17:00.0/0000:18:00.0/nvme/nvme0/nvme0n1/nvme0n1p1
size_t parent_end = device_dir.rfind('/', device_dir.length() - 1);
if (parent_end == std::string::npos) {
return kDefaultPageSize;
}
size_t parent_begin = device_dir.rfind('/', parent_end - 1);
if (parent_begin == std::string::npos) {
return kDefaultPageSize;
}
std::string parent =
device_dir.substr(parent_begin + 1, parent_end - parent_begin - 1);
std::string child = device_dir.substr(parent_end + 1, std::string::npos);
if (parent != "block" &&
(child.compare(0, 4, "nvme") || child.find('p') != std::string::npos)) {
device_dir = device_dir.substr(0, parent_end);
}
std::string fname = device_dir + "/queue/logical_block_size";
FILE* fp;
size_t size = 0;
fp = fopen(fname.c_str(), "r");
if (fp != nullptr) {
char* line = nullptr;
size_t len = 0;
if (getline(&line, &len, fp) != -1) {
sscanf(line, "%zu", &size);
}
free(line);
fclose(fp);
}
if (size != 0 && (size & (size - 1)) == 0) {
return size;
}
#endif
return kDefaultPageSize;
}
} // namespace
/*
* DirectIOHelper
*/
#ifndef NDEBUG
namespace {
bool IsSectorAligned(const size_t off, size_t sector_size) {
return off % sector_size == 0;
}
bool IsSectorAligned(const void* ptr, size_t sector_size) {
return uintptr_t(ptr) % sector_size == 0;
}
}
#endif
/*
* PosixSequentialFile
*/
PosixSequentialFile::PosixSequentialFile(const std::string& fname, FILE* file,
int fd, const EnvOptions& options)
: filename_(fname),
file_(file),
fd_(fd),
use_direct_io_(options.use_direct_reads),
logical_sector_size_(GetLogicalBufferSize(fd_)) {
assert(!options.use_direct_reads || !options.use_mmap_reads);
}
PosixSequentialFile::~PosixSequentialFile() {
if (!use_direct_io()) {
assert(file_);
fclose(file_);
} else {
assert(fd_);
close(fd_);
}
}
Status PosixSequentialFile::Read(size_t n, Slice* result, char* scratch) {
assert(result != nullptr && !use_direct_io());
Status s;
size_t r = 0;
do {
r = fread_unlocked(scratch, 1, n, file_);
} while (r == 0 && ferror(file_) && errno == EINTR);
*result = Slice(scratch, r);
if (r < n) {
if (feof(file_)) {
// We leave status as ok if we hit the end of the file
// We also clear the error so that the reads can continue
// if a new data is written to the file
clearerr(file_);
} else {
// A partial read with an error: return a non-ok status
s = IOError("While reading file sequentially", filename_, errno);
}
}
return s;
}
Status PosixSequentialFile::PositionedRead(uint64_t offset, size_t n,
Slice* result, char* scratch) {
assert(use_direct_io());
assert(IsSectorAligned(offset, GetRequiredBufferAlignment()));
assert(IsSectorAligned(n, GetRequiredBufferAlignment()));
assert(IsSectorAligned(scratch, GetRequiredBufferAlignment()));
Status s;
ssize_t r = -1;
size_t left = n;
char* ptr = scratch;
while (left > 0) {
r = pread(fd_, ptr, left, static_cast<off_t>(offset));
if (r <= 0) {
if (r == -1 && errno == EINTR) {
continue;
}
break;
}
ptr += r;
offset += r;
left -= r;
if (r % static_cast<ssize_t>(GetRequiredBufferAlignment()) != 0) {
// Bytes reads don't fill sectors. Should only happen at the end
// of the file.
break;
}
}
if (r < 0) {
// An error: return a non-ok status
s = IOError(
"While pread " + ToString(n) + " bytes from offset " + ToString(offset),
filename_, errno);
}
*result = Slice(scratch, (r < 0) ? 0 : n - left);
return s;
}
Status PosixSequentialFile::Skip(uint64_t n) {
if (fseek(file_, static_cast<long int>(n), SEEK_CUR)) {
return IOError("While fseek to skip " + ToString(n) + " bytes", filename_,
errno);
}
return Status::OK();
}
Status PosixSequentialFile::InvalidateCache(size_t offset, size_t length) {
#ifndef OS_LINUX
(void)offset;
(void)length;
return Status::OK();
#else
if (!use_direct_io()) {
// free OS pages
int ret = Fadvise(fd_, offset, length, POSIX_FADV_DONTNEED);
if (ret != 0) {
return IOError("While fadvise NotNeeded offset " + ToString(offset) +
" len " + ToString(length),
filename_, errno);
}
}
return Status::OK();
#endif
}
/*
* PosixRandomAccessFile
*/
#if defined(OS_LINUX)
size_t PosixHelper::GetUniqueIdFromFile(int fd, char* id, size_t max_size) {
if (max_size < kMaxVarint64Length * 3) {
return 0;
}
struct stat buf;
int result = fstat(fd, &buf);
if (result == -1) {
return 0;
}
long version = 0;
result = ioctl(fd, FS_IOC_GETVERSION, &version);
TEST_SYNC_POINT_CALLBACK("GetUniqueIdFromFile:FS_IOC_GETVERSION", &result);
if (result == -1) {
return 0;
}
uint64_t uversion = (uint64_t)version;
char* rid = id;
rid = EncodeVarint64(rid, buf.st_dev);
rid = EncodeVarint64(rid, buf.st_ino);
rid = EncodeVarint64(rid, uversion);
assert(rid >= id);
return static_cast<size_t>(rid - id);
}
#endif
#if defined(OS_MACOSX) || defined(OS_AIX)
size_t PosixHelper::GetUniqueIdFromFile(int fd, char* id, size_t max_size) {
if (max_size < kMaxVarint64Length * 3) {
return 0;
}
struct stat buf;
int result = fstat(fd, &buf);
if (result == -1) {
return 0;
}
char* rid = id;
rid = EncodeVarint64(rid, buf.st_dev);
rid = EncodeVarint64(rid, buf.st_ino);
rid = EncodeVarint64(rid, buf.st_gen);
assert(rid >= id);
return static_cast<size_t>(rid - id);
}
#endif
/*
* PosixRandomAccessFile
*
* pread() based random-access
*/
PosixRandomAccessFile::PosixRandomAccessFile(const std::string& fname, int fd,
const EnvOptions& options)
: filename_(fname),
fd_(fd),
use_direct_io_(options.use_direct_reads),
logical_sector_size_(GetLogicalBufferSize(fd_)) {
assert(!options.use_direct_reads || !options.use_mmap_reads);
assert(!options.use_mmap_reads || sizeof(void*) < 8);
}
PosixRandomAccessFile::~PosixRandomAccessFile() { close(fd_); }
Status PosixRandomAccessFile::Read(uint64_t offset, size_t n, Slice* result,
char* scratch) const {
if (use_direct_io()) {
assert(IsSectorAligned(offset, GetRequiredBufferAlignment()));
assert(IsSectorAligned(n, GetRequiredBufferAlignment()));
assert(IsSectorAligned(scratch, GetRequiredBufferAlignment()));
}
Status s;
ssize_t r = -1;
size_t left = n;
char* ptr = scratch;
while (left > 0) {
r = pread(fd_, ptr, left, static_cast<off_t>(offset));
if (r <= 0) {
if (r == -1 && errno == EINTR) {
continue;
}
break;
}
ptr += r;
offset += r;
left -= r;
if (use_direct_io() &&
r % static_cast<ssize_t>(GetRequiredBufferAlignment()) != 0) {
// Bytes reads don't fill sectors. Should only happen at the end
// of the file.
break;
}
}
if (r < 0) {
// An error: return a non-ok status
s = IOError(
"While pread offset " + ToString(offset) + " len " + ToString(n),
filename_, errno);
}
*result = Slice(scratch, (r < 0) ? 0 : n - left);
return s;
}
Status PosixRandomAccessFile::Prefetch(uint64_t offset, size_t n) {
Status s;
if (!use_direct_io()) {
ssize_t r = 0;
#ifdef OS_LINUX
r = readahead(fd_, offset, n);
#endif
#ifdef OS_MACOSX
radvisory advice;
advice.ra_offset = static_cast<off_t>(offset);
advice.ra_count = static_cast<int>(n);
r = fcntl(fd_, F_RDADVISE, &advice);
#endif
if (r == -1) {
s = IOError("While prefetching offset " + ToString(offset) + " len " +
ToString(n),
filename_, errno);
}
}
return s;
}
#if defined(OS_LINUX) || defined(OS_MACOSX) || defined(OS_AIX)
size_t PosixRandomAccessFile::GetUniqueId(char* id, size_t max_size) const {
return PosixHelper::GetUniqueIdFromFile(fd_, id, max_size);
}
#endif
void PosixRandomAccessFile::Hint(AccessPattern pattern) {
if (use_direct_io()) {
return;
}
switch (pattern) {
case NORMAL:
Fadvise(fd_, 0, 0, POSIX_FADV_NORMAL);
break;
case RANDOM:
Fadvise(fd_, 0, 0, POSIX_FADV_RANDOM);
break;
case SEQUENTIAL:
Fadvise(fd_, 0, 0, POSIX_FADV_SEQUENTIAL);
break;
case WILLNEED:
Fadvise(fd_, 0, 0, POSIX_FADV_WILLNEED);
break;
case DONTNEED:
Fadvise(fd_, 0, 0, POSIX_FADV_DONTNEED);
break;
default:
assert(false);
break;
}
}
Status PosixRandomAccessFile::InvalidateCache(size_t offset, size_t length) {
if (use_direct_io()) {
return Status::OK();
}
#ifndef OS_LINUX
(void)offset;
(void)length;
return Status::OK();
#else
// free OS pages
int ret = Fadvise(fd_, offset, length, POSIX_FADV_DONTNEED);
if (ret == 0) {
return Status::OK();
}
return IOError("While fadvise NotNeeded offset " + ToString(offset) +
" len " + ToString(length),
filename_, errno);
#endif
}
/*
* PosixMmapReadableFile
*
* mmap() based random-access
*/
// base[0,length-1] contains the mmapped contents of the file.
PosixMmapReadableFile::PosixMmapReadableFile(const int fd,
const std::string& fname,
void* base, size_t length,
const EnvOptions& options)
: fd_(fd), filename_(fname), mmapped_region_(base), length_(length) {
#ifdef NDEBUG
(void)options;
#endif
fd_ = fd_ + 0; // suppress the warning for used variables
assert(options.use_mmap_reads);
assert(!options.use_direct_reads);
}
PosixMmapReadableFile::~PosixMmapReadableFile() {
int ret = munmap(mmapped_region_, length_);
if (ret != 0) {
fprintf(stdout, "failed to munmap %p length %" ROCKSDB_PRIszt " \n",
mmapped_region_, length_);
}
close(fd_);
}
Status PosixMmapReadableFile::Read(uint64_t offset, size_t n, Slice* result,
char* /*scratch*/) const {
Status s;
if (offset > length_) {
*result = Slice();
return IOError("While mmap read offset " + ToString(offset) +
" larger than file length " + ToString(length_),
filename_, EINVAL);
} else if (offset + n > length_) {
n = static_cast<size_t>(length_ - offset);
}
*result = Slice(reinterpret_cast<char*>(mmapped_region_) + offset, n);
return s;
}
Status PosixMmapReadableFile::InvalidateCache(size_t offset, size_t length) {
#ifndef OS_LINUX
(void)offset;
(void)length;
return Status::OK();
#else
// free OS pages
int ret = Fadvise(fd_, offset, length, POSIX_FADV_DONTNEED);
if (ret == 0) {
return Status::OK();
}
return IOError("While fadvise not needed. Offset " + ToString(offset) +
" len" + ToString(length),
filename_, errno);
#endif
}
/*
* PosixMmapFile
*
* We preallocate up to an extra megabyte and use memcpy to append new
* data to the file. This is safe since we either properly close the
* file before reading from it, or for log files, the reading code
* knows enough to skip zero suffixes.
*/
Status PosixMmapFile::UnmapCurrentRegion() {
TEST_KILL_RANDOM("PosixMmapFile::UnmapCurrentRegion:0", rocksdb_kill_odds);
if (base_ != nullptr) {
int munmap_status = munmap(base_, limit_ - base_);
if (munmap_status != 0) {
return IOError("While munmap", filename_, munmap_status);
}
file_offset_ += limit_ - base_;
base_ = nullptr;
limit_ = nullptr;
last_sync_ = nullptr;
dst_ = nullptr;
// Increase the amount we map the next time, but capped at 1MB
if (map_size_ < (1 << 20)) {
map_size_ *= 2;
}
}
return Status::OK();
}
Status PosixMmapFile::MapNewRegion() {
#ifdef ROCKSDB_FALLOCATE_PRESENT
assert(base_ == nullptr);
TEST_KILL_RANDOM("PosixMmapFile::UnmapCurrentRegion:0", rocksdb_kill_odds);
// we can't fallocate with FALLOC_FL_KEEP_SIZE here
if (allow_fallocate_) {
IOSTATS_TIMER_GUARD(allocate_nanos);
int alloc_status = fallocate(fd_, 0, file_offset_, map_size_);
if (alloc_status != 0) {
// fallback to posix_fallocate
alloc_status = posix_fallocate(fd_, file_offset_, map_size_);
}
if (alloc_status != 0) {
return Status::IOError("Error allocating space to file : " + filename_ +
"Error : " + strerror(alloc_status));
}
}
TEST_KILL_RANDOM("PosixMmapFile::Append:1", rocksdb_kill_odds);
void* ptr = mmap(nullptr, map_size_, PROT_READ | PROT_WRITE, MAP_SHARED, fd_,
file_offset_);
if (ptr == MAP_FAILED) {
return Status::IOError("MMap failed on " + filename_);
}
TEST_KILL_RANDOM("PosixMmapFile::Append:2", rocksdb_kill_odds);
base_ = reinterpret_cast<char*>(ptr);
limit_ = base_ + map_size_;
dst_ = base_;
last_sync_ = base_;
return Status::OK();
#else
return Status::NotSupported("This platform doesn't support fallocate()");
#endif
}
Status PosixMmapFile::Msync() {
if (dst_ == last_sync_) {
return Status::OK();
}
// Find the beginnings of the pages that contain the first and last
// bytes to be synced.
size_t p1 = TruncateToPageBoundary(last_sync_ - base_);
size_t p2 = TruncateToPageBoundary(dst_ - base_ - 1);
last_sync_ = dst_;
TEST_KILL_RANDOM("PosixMmapFile::Msync:0", rocksdb_kill_odds);
if (msync(base_ + p1, p2 - p1 + page_size_, MS_SYNC) < 0) {
return IOError("While msync", filename_, errno);
}
return Status::OK();
}
PosixMmapFile::PosixMmapFile(const std::string& fname, int fd, size_t page_size,
const EnvOptions& options)
: filename_(fname),
fd_(fd),
page_size_(page_size),
map_size_(Roundup(65536, page_size)),
base_(nullptr),
limit_(nullptr),
dst_(nullptr),
last_sync_(nullptr),
file_offset_(0) {
#ifdef ROCKSDB_FALLOCATE_PRESENT
allow_fallocate_ = options.allow_fallocate;
fallocate_with_keep_size_ = options.fallocate_with_keep_size;
#else
(void)options;
#endif
assert((page_size & (page_size - 1)) == 0);
assert(options.use_mmap_writes);
assert(!options.use_direct_writes);
}
PosixMmapFile::~PosixMmapFile() {
if (fd_ >= 0) {
PosixMmapFile::Close();
}
}
Status PosixMmapFile::Append(const Slice& data) {
const char* src = data.data();
size_t left = data.size();
while (left > 0) {
assert(base_ <= dst_);
assert(dst_ <= limit_);
size_t avail = limit_ - dst_;
if (avail == 0) {
Status s = UnmapCurrentRegion();
if (!s.ok()) {
return s;
}
s = MapNewRegion();
if (!s.ok()) {
return s;
}
TEST_KILL_RANDOM("PosixMmapFile::Append:0", rocksdb_kill_odds);
}
size_t n = (left <= avail) ? left : avail;
assert(dst_);
memcpy(dst_, src, n);
dst_ += n;
src += n;
left -= n;
}
return Status::OK();
}
Status PosixMmapFile::Close() {
Status s;
size_t unused = limit_ - dst_;
s = UnmapCurrentRegion();
if (!s.ok()) {
s = IOError("While closing mmapped file", filename_, errno);
} else if (unused > 0) {
// Trim the extra space at the end of the file
if (ftruncate(fd_, file_offset_ - unused) < 0) {
s = IOError("While ftruncating mmaped file", filename_, errno);
}
}
if (close(fd_) < 0) {
if (s.ok()) {
s = IOError("While closing mmapped file", filename_, errno);
}
}
fd_ = -1;
base_ = nullptr;
limit_ = nullptr;
return s;
}
Status PosixMmapFile::Flush() { return Status::OK(); }
Status PosixMmapFile::Sync() {
if (fdatasync(fd_) < 0) {
return IOError("While fdatasync mmapped file", filename_, errno);
}
return Msync();
}
/**
* Flush data as well as metadata to stable storage.
*/
Status PosixMmapFile::Fsync() {
if (fsync(fd_) < 0) {
return IOError("While fsync mmaped file", filename_, errno);
}
return Msync();
}
/**
* Get the size of valid data in the file. This will not match the
* size that is returned from the filesystem because we use mmap
* to extend file by map_size every time.
*/
uint64_t PosixMmapFile::GetFileSize() {
size_t used = dst_ - base_;
return file_offset_ + used;
}
Status PosixMmapFile::InvalidateCache(size_t offset, size_t length) {
#ifndef OS_LINUX
(void)offset;
(void)length;
return Status::OK();
#else
// free OS pages
int ret = Fadvise(fd_, offset, length, POSIX_FADV_DONTNEED);
if (ret == 0) {
return Status::OK();
}
return IOError("While fadvise NotNeeded mmapped file", filename_, errno);
#endif
}
#ifdef ROCKSDB_FALLOCATE_PRESENT
Status PosixMmapFile::Allocate(uint64_t offset, uint64_t len) {
assert(offset <= std::numeric_limits<off_t>::max());
assert(len <= std::numeric_limits<off_t>::max());
TEST_KILL_RANDOM("PosixMmapFile::Allocate:0", rocksdb_kill_odds);
int alloc_status = 0;
if (allow_fallocate_) {
alloc_status = fallocate(
fd_, fallocate_with_keep_size_ ? FALLOC_FL_KEEP_SIZE : 0,
static_cast<off_t>(offset), static_cast<off_t>(len));
}
if (alloc_status == 0) {
return Status::OK();
} else {
return IOError(
"While fallocate offset " + ToString(offset) + " len " + ToString(len),
filename_, errno);
}
}
#endif
/*
* PosixWritableFile
*
* Use posix write to write data to a file.
*/
PosixWritableFile::PosixWritableFile(const std::string& fname, int fd,
const EnvOptions& options)
: filename_(fname),
use_direct_io_(options.use_direct_writes),
fd_(fd),
filesize_(0),
logical_sector_size_(GetLogicalBufferSize(fd_)) {
#ifdef ROCKSDB_FALLOCATE_PRESENT
allow_fallocate_ = options.allow_fallocate;
fallocate_with_keep_size_ = options.fallocate_with_keep_size;
#endif
assert(!options.use_mmap_writes);
}
PosixWritableFile::~PosixWritableFile() {
if (fd_ >= 0) {
PosixWritableFile::Close();
}
}
Status PosixWritableFile::Append(const Slice& data) {
if (use_direct_io()) {
assert(IsSectorAligned(data.size(), GetRequiredBufferAlignment()));
assert(IsSectorAligned(data.data(), GetRequiredBufferAlignment()));
}
const char* src = data.data();
size_t left = data.size();
while (left != 0) {
ssize_t done = write(fd_, src, left);
if (done < 0) {
if (errno == EINTR) {
continue;
}
return IOError("While appending to file", filename_, errno);
}
left -= done;
src += done;
}
filesize_ += data.size();
return Status::OK();
}
Status PosixWritableFile::PositionedAppend(const Slice& data, uint64_t offset) {
if (use_direct_io()) {
assert(IsSectorAligned(offset, GetRequiredBufferAlignment()));
assert(IsSectorAligned(data.size(), GetRequiredBufferAlignment()));
assert(IsSectorAligned(data.data(), GetRequiredBufferAlignment()));
}
assert(offset <= std::numeric_limits<off_t>::max());
const char* src = data.data();
size_t left = data.size();
while (left != 0) {
ssize_t done = pwrite(fd_, src, left, static_cast<off_t>(offset));
if (done < 0) {
if (errno == EINTR) {
continue;
}
return IOError("While pwrite to file at offset " + ToString(offset),
filename_, errno);
}
left -= done;
offset += done;
src += done;
}
filesize_ = offset;
return Status::OK();
}
Status PosixWritableFile::Truncate(uint64_t size) {
Status s;
int r = ftruncate(fd_, size);
if (r < 0) {
s = IOError("While ftruncate file to size " + ToString(size), filename_,
errno);
} else {
filesize_ = size;
}
return s;
}
Status PosixWritableFile::Close() {
Status s;
size_t block_size;
size_t last_allocated_block;
GetPreallocationStatus(&block_size, &last_allocated_block);
if (last_allocated_block > 0) {
// trim the extra space preallocated at the end of the file
// NOTE(ljin): we probably don't want to surface failure as an IOError,
// but it will be nice to log these errors.
int dummy __attribute__((__unused__));
dummy = ftruncate(fd_, filesize_);
#if defined(ROCKSDB_FALLOCATE_PRESENT) && defined(FALLOC_FL_PUNCH_HOLE) && \
!defined(TRAVIS)
// in some file systems, ftruncate only trims trailing space if the
// new file size is smaller than the current size. Calling fallocate
// with FALLOC_FL_PUNCH_HOLE flag to explicitly release these unused
// blocks. FALLOC_FL_PUNCH_HOLE is supported on at least the following
// filesystems:
// XFS (since Linux 2.6.38)
// ext4 (since Linux 3.0)
// Btrfs (since Linux 3.7)
// tmpfs (since Linux 3.5)
// We ignore error since failure of this operation does not affect
// correctness.
// TRAVIS - this code does not work on TRAVIS filesystems.
// the FALLOC_FL_KEEP_SIZE option is expected to not change the size
// of the file, but it does. Simple strace report will show that.
// While we work with Travis-CI team to figure out if this is a
// quirk of Docker/AUFS, we will comment this out.
struct stat file_stats;
int result = fstat(fd_, &file_stats);
// After ftruncate, we check whether ftruncate has the correct behavior.
// If not, we should hack it with FALLOC_FL_PUNCH_HOLE
if (result == 0 &&
(file_stats.st_size + file_stats.st_blksize - 1) /
file_stats.st_blksize !=
file_stats.st_blocks / (file_stats.st_blksize / 512)) {
IOSTATS_TIMER_GUARD(allocate_nanos);
if (allow_fallocate_) {
fallocate(fd_, FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE, filesize_,
block_size * last_allocated_block - filesize_);
}
}
#endif
}
if (close(fd_) < 0) {
s = IOError("While closing file after writing", filename_, errno);
}
fd_ = -1;
return s;
}
// write out the cached data to the OS cache
Status PosixWritableFile::Flush() { return Status::OK(); }
Status PosixWritableFile::Sync() {
if (fdatasync(fd_) < 0) {
return IOError("While fdatasync", filename_, errno);
}
return Status::OK();
}
Status PosixWritableFile::Fsync() {
if (fsync(fd_) < 0) {
return IOError("While fsync", filename_, errno);
}
return Status::OK();
}
bool PosixWritableFile::IsSyncThreadSafe() const { return true; }
uint64_t PosixWritableFile::GetFileSize() { return filesize_; }
void PosixWritableFile::SetWriteLifeTimeHint(Env::WriteLifeTimeHint hint) {
#ifdef OS_LINUX
// Suppress Valgrind "Unimplemented functionality" error.
#ifndef ROCKSDB_VALGRIND_RUN
if (hint == write_hint_) {
return;
}
if (fcntl(fd_, F_SET_RW_HINT, &hint) == 0) {
write_hint_ = hint;
}
#else
(void)hint;
#endif // ROCKSDB_VALGRIND_RUN
#else
(void)hint;
#endif // OS_LINUX
}
Status PosixWritableFile::InvalidateCache(size_t offset, size_t length) {
if (use_direct_io()) {
return Status::OK();
}
#ifndef OS_LINUX
(void)offset;
(void)length;
return Status::OK();
#else
// free OS pages
int ret = Fadvise(fd_, offset, length, POSIX_FADV_DONTNEED);
if (ret == 0) {
return Status::OK();
}
return IOError("While fadvise NotNeeded", filename_, errno);
#endif
}
#ifdef ROCKSDB_FALLOCATE_PRESENT
Status PosixWritableFile::Allocate(uint64_t offset, uint64_t len) {
assert(offset <= std::numeric_limits<off_t>::max());
assert(len <= std::numeric_limits<off_t>::max());
TEST_KILL_RANDOM("PosixWritableFile::Allocate:0", rocksdb_kill_odds);
IOSTATS_TIMER_GUARD(allocate_nanos);
int alloc_status = 0;
if (allow_fallocate_) {
alloc_status = fallocate(
fd_, fallocate_with_keep_size_ ? FALLOC_FL_KEEP_SIZE : 0,
static_cast<off_t>(offset), static_cast<off_t>(len));
}
if (alloc_status == 0) {
return Status::OK();
} else {
return IOError(
"While fallocate offset " + ToString(offset) + " len " + ToString(len),
filename_, errno);
}
}
#endif
#ifdef ROCKSDB_RANGESYNC_PRESENT
Status PosixWritableFile::RangeSync(uint64_t offset, uint64_t nbytes) {
assert(offset <= std::numeric_limits<off_t>::max());
assert(nbytes <= std::numeric_limits<off_t>::max());
if (sync_file_range(fd_, static_cast<off_t>(offset),
static_cast<off_t>(nbytes), SYNC_FILE_RANGE_WRITE) == 0) {
return Status::OK();
} else {
return IOError("While sync_file_range offset " + ToString(offset) +
" bytes " + ToString(nbytes),
filename_, errno);
}
}
#endif
#ifdef OS_LINUX
size_t PosixWritableFile::GetUniqueId(char* id, size_t max_size) const {
return PosixHelper::GetUniqueIdFromFile(fd_, id, max_size);
}
#endif
/*
* PosixRandomRWFile
*/
PosixRandomRWFile::PosixRandomRWFile(const std::string& fname, int fd,
const EnvOptions& /*options*/)
: filename_(fname), fd_(fd) {}
PosixRandomRWFile::~PosixRandomRWFile() {
if (fd_ >= 0) {
Close();
}
}
Status PosixRandomRWFile::Write(uint64_t offset, const Slice& data) {
const char* src = data.data();
size_t left = data.size();
while (left != 0) {
ssize_t done = pwrite(fd_, src, left, offset);
if (done < 0) {
// error while writing to file
if (errno == EINTR) {
// write was interrupted, try again.
continue;
}
return IOError(
"While write random read/write file at offset " + ToString(offset),
filename_, errno);
}
// Wrote `done` bytes
left -= done;
offset += done;
src += done;
}
return Status::OK();
}
Status PosixRandomRWFile::Read(uint64_t offset, size_t n, Slice* result,
char* scratch) const {
size_t left = n;
char* ptr = scratch;
while (left > 0) {
ssize_t done = pread(fd_, ptr, left, offset);
if (done < 0) {
// error while reading from file
if (errno == EINTR) {
// read was interrupted, try again.
continue;
}
return IOError("While reading random read/write file offset " +
ToString(offset) + " len " + ToString(n),
filename_, errno);
} else if (done == 0) {
// Nothing more to read
break;
}
// Read `done` bytes
ptr += done;
offset += done;
left -= done;
}
*result = Slice(scratch, n - left);
return Status::OK();
}
Status PosixRandomRWFile::Flush() { return Status::OK(); }
Status PosixRandomRWFile::Sync() {
if (fdatasync(fd_) < 0) {
return IOError("While fdatasync random read/write file", filename_, errno);
}
return Status::OK();
}
Status PosixRandomRWFile::Fsync() {
if (fsync(fd_) < 0) {
return IOError("While fsync random read/write file", filename_, errno);
}
return Status::OK();
}
Status PosixRandomRWFile::Close() {
if (close(fd_) < 0) {
return IOError("While close random read/write file", filename_, errno);
}
fd_ = -1;
return Status::OK();
}
PosixMemoryMappedFileBuffer::~PosixMemoryMappedFileBuffer() {
// TODO should have error handling though not much we can do...
munmap(this->base_, length_);
}
/*
* PosixDirectory
*/
PosixDirectory::~PosixDirectory() { close(fd_); }
Status PosixDirectory::Fsync() {
#ifndef OS_AIX
if (fsync(fd_) == -1) {
return IOError("While fsync", "a directory", errno);
}
#endif
return Status::OK();
}
} // namespace rocksdb
#endif
![swh spinner](/static/img/swh-spinner.gif)
Computing file changes ...