thebestvibedev/tromcho.net
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Support xxhash 32-bit checksums (data_csum_type=xxh3_32)

Browse Source
This commit is contained in:
Vitaliy Filippov
2026-03-20 20:58:57 +03:00
parent 3ed7a0bc5e
commit 3aa307f034
19 changed files with 7672 additions and 48 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.gitea/workflows/test.yml
+18
View File
@@ -1278,6 +1278,24 @@ jobs:
echo ""
done
test_checksum_xxhash:
runs-on: ubuntu-latest
needs: build
container: ${{env.TEST_IMAGE}}:${{github.sha}}
steps:
- name: Run test
id: test
timeout-minutes: 3
run: TEST_NAME=xxhash OSD_ARGS="--data_csum_type xxh3_32" /root/vitastor/tests/test_checksum.sh
- name: Print logs
if: always() && steps.test.outcome == 'failure'
run: |
for i in /root/vitastor/testdata/*.log /root/vitastor/testdata/*.txt; do
echo "-------- $i --------"
cat $i
echo ""
done
test_old_checksum:
runs-on: ubuntu-latest
needs: build
src/blockstore/CMakeLists.txt
+1 -1
View File
@@ -4,7 +4,7 @@ project(vitastor)
# libvitastor_blk.a
add_library(vitastor_blk STATIC
../util/allocator.cpp ../util/crc32c.c ../util/ringloop.cpp
../util/allocator.cpp ../util/crc32c.c ../util/xxhash.c ../util/ringloop.cpp
multilist.cpp blockstore_heap.cpp blockstore_disk.cpp
blockstore.cpp blockstore_impl.cpp blockstore_init.cpp blockstore_open.cpp
blockstore_flush.cpp blockstore_read.cpp blockstore_stable.cpp blockstore_sync.cpp blockstore_write.cpp
src/blockstore/blockstore_disk.cpp
+5 -1
View File
@@ -83,13 +83,17 @@ void blockstore_disk_t::parse_config(std::map<std::string, std::string> & config
{
data_csum_type = BLOCKSTORE_CSUM_CRC32C;
}
else if (config["data_csum_type"] == "xxh3_32")
{
data_csum_type = BLOCKSTORE_CSUM_XXH3_32;
}
else if (config["data_csum_type"] == "" || config["data_csum_type"] == "none")
{
data_csum_type = BLOCKSTORE_CSUM_NONE;
}
else
{
throw std::runtime_error("data_csum_type="+config["data_csum_type"]+" is unsupported, only \"crc32c\" and \"none\" are supported");
throw std::runtime_error("data_csum_type="+config["data_csum_type"]+" is unsupported, only \"crc32c\", \"xxh3_32\" and \"none\" are supported");
}
csum_block_size = parse_size(config["csum_block_size"]);
discard_on_start = config.find("discard_on_start") != config.end() &&
src/blockstore/blockstore_disk.h
+1
View File
@@ -16,6 +16,7 @@
#define BLOCKSTORE_CSUM_NONE 0
// Lower byte of checksum type is its length
#define BLOCKSTORE_CSUM_CRC32C 0x104
#define BLOCKSTORE_CSUM_XXH3_32 0x204
#define MOCK_DATA_FD 1000
#define MOCK_META_FD 1001
src/blockstore/blockstore_heap.cpp
+94 -18
View File
@@ -12,6 +12,7 @@
#include "blockstore_heap.h"
#include "../util/allocator.h"
#include "../util/crc32c.h"
#include "../util/xxhash.h"
#include "../util/malloc_or_die.h"
#define BS_HEAP_FREE_MVCC 1
@@ -212,15 +213,24 @@ void heap_entry_t::set_big_location(blockstore_heap_t *heap, uint64_t location)
big().block_num = location / heap->dsk->data_block_size;
}
uint32_t heap_entry_t::calc_crc32c()
uint32_t heap_entry_t::calc_checksum(blockstore_disk_t *dsk)
{
auto old_crc32c = crc32c;
crc32c = 0;
uint32_t res = ::crc32c(0, (uint8_t*)this, size);
crc32c = old_crc32c;
auto old_checksum = checksum;
checksum = 0;
uint32_t res = 0;
if (dsk->data_csum_type == BLOCKSTORE_CSUM_XXH3_32)
res = (uint32_t)XXH3_64bits(this, size);
else
res = ::crc32c(0, (uint8_t*)this, size);
checksum = old_checksum;
return res;
}
uint32_t heap_entry_t::calc_checksum(blockstore_heap_t *heap)
{
return calc_checksum(heap->dsk);
}
uint64_t blockstore_heap_t::get_pg_id(inode_t inode, uint64_t stripe)
{
uint64_t pg_num = 0;
@@ -385,12 +395,12 @@ corrupted_object:
goto corrupted_object;
}
// Verify crc
uint32_t expected_crc32c = wr->calc_crc32c();
if (wr->crc32c != expected_crc32c)
uint32_t expected_checksum = wr->calc_checksum(this);
if (wr->checksum != expected_checksum)
{
fprintf(stderr, "Error: entry %jx:%jx v%ju l%ju in metadata block %u at %u is corrupt (crc32c mismatch: expected %08x, got %08x). ",
fprintf(stderr, "Error: entry %jx:%jx v%ju l%ju in metadata block %u at %u is corrupt (checksum mismatch: expected %08x, got %08x). ",
wr->inode, wr->stripe, wr->version, wr->lsn,
block_num, block_offset, expected_crc32c, wr->crc32c);
block_num, block_offset, expected_checksum, wr->checksum);
goto corrupted_object;
}
// Verify offset & len
@@ -878,7 +888,11 @@ bool blockstore_heap_t::calc_checksums(heap_entry_t *wr, uint8_t *data, bool set
len = wr->big_intent().len;
else
assert(0);
uint32_t real_csum = crc32c(0, data, len);
uint32_t real_csum = 0;
if (dsk->data_csum_type == BLOCKSTORE_CSUM_XXH3_32)
real_csum = (uint32_t)XXH3_64bits(data, len);
else
real_csum = crc32c(0, data, len);
if (set)
{
*wr_csum = real_csum;
@@ -928,11 +942,26 @@ static uint32_t crc32c_iter(uint32_t prev_crc, const std::function<uint8_t*(uint
return prev_crc;
}
static void xxh3_iter(XXH3_state_t* xxh3_state, const std::function<uint8_t*(uint32_t start, uint32_t & len)> & next, uint32_t pos, uint32_t size)
{
uint32_t cur_len = 0;
while (size > 0)
{
uint8_t *data = next(pos, cur_len);
assert(data);
cur_len = (cur_len < size ? cur_len : size);
XXH3_64bits_update(xxh3_state, data, cur_len);
pos += cur_len;
size -= cur_len;
}
}
bool blockstore_heap_t::calc_block_checksums(uint32_t *block_csums, uint8_t *bitmap,
uint32_t start, uint32_t end, std::function<uint8_t*(uint32_t start, uint32_t & len)> next,
bool set, std::function<void(uint32_t, uint32_t, uint32_t)> bad_block_cb)
{
bool res = true;
XXH3_state_t* xxh3_state = NULL;
uint32_t pos = start;
uint32_t block_end = (start/dsk->csum_block_size + 1)*dsk->csum_block_size;
uint32_t block_crc = 0;
@@ -949,42 +978,89 @@ bool blockstore_heap_t::calc_block_checksums(uint32_t *block_csums, uint8_t *bit
pos += dsk->bitmap_granularity;
// zero padding at the beginning or at the end of the block is not counted
if (pos > prev && prev > 0 && pos < block_end)
block_crc = crc32c_pad(block_crc, NULL, 0, pos-prev, 0);
{
if (dsk->data_csum_type == BLOCKSTORE_CSUM_XXH3_32)
{
if (!xxh3_state)
{
xxh3_state = XXH3_createState();
XXH3_64bits_reset(xxh3_state);
}
uint32_t zeropad = pos-prev;
while (zeropad > 0)
{
uint32_t zerolen = zeropad > 4096 ? 4096 : zeropad;
XXH3_64bits_update(xxh3_state, zero_page, zerolen);
zeropad -= zerolen;
}
}
else
block_crc = crc32c_pad(block_crc, NULL, 0, pos-prev, 0);
}
prev = pos;
while (pos < end && pos < block_end && (bitmap[pos/dsk->bitmap_granularity/8] & (1 << ((pos/dsk->bitmap_granularity) % 8))))
pos += dsk->bitmap_granularity;
if (pos > prev)
{
isset = true;
block_crc = crc32c_iter(block_crc, next, prev, pos-prev);
if (dsk->data_csum_type == BLOCKSTORE_CSUM_XXH3_32)
{
if (!xxh3_state)
{
xxh3_state = XXH3_createState();
XXH3_64bits_reset(xxh3_state);
}
xxh3_iter(xxh3_state, next, prev, pos-prev);
}
else
block_crc = crc32c_iter(block_crc, next, prev, pos-prev);
}
prev = pos;
}
}
else
{
block_crc = crc32c_iter(block_crc, next, pos, (end > block_end ? block_end : end)-pos);
if (dsk->data_csum_type == BLOCKSTORE_CSUM_XXH3_32)
{
if (!xxh3_state)
{
xxh3_state = XXH3_createState();
XXH3_64bits_reset(xxh3_state);
}
xxh3_iter(xxh3_state, next, pos, (end > block_end ? block_end : end)-pos);
}
else
block_crc = crc32c_iter(block_crc, next, pos, (end > block_end ? block_end : end)-pos);
pos = (end > block_end ? block_end : end);
isset = true;
}
if (dsk->data_csum_type == BLOCKSTORE_CSUM_XXH3_32 && xxh3_state)
{
block_crc = (uint32_t)XXH3_64bits_digest(xxh3_state);
XXH3_64bits_reset(xxh3_state);
}
if (set)
{
*block_csums = block_crc;
}
else if (isset && block_crc != *block_csums)
{
res = false;
if (bad_block_cb)
{
bad_block_cb(blk_start, *block_csums, block_crc);
res = false;
}
else
return false;
break;
}
block_end += dsk->csum_block_size;
block_crc = 0;
block_csums++;
}
if (dsk->data_csum_type == BLOCKSTORE_CSUM_XXH3_32 && xxh3_state)
{
block_crc = (uint32_t)XXH3_64bits_digest(xxh3_state);
XXH3_freeState(xxh3_state);
xxh3_state = NULL;
}
return res;
}
@@ -1336,7 +1412,7 @@ int blockstore_heap_t::add_entry(uint32_t wr_size, uint32_t *modified_block,
insert_list_item(li);
li->block_num = block_num;
new_wr->size = wr_size;
new_wr->crc32c = new_wr->calc_crc32c();
new_wr->checksum = new_wr->calc_checksum(this);
return 0;
}
src/blockstore/blockstore_heap.h
+5 -4
View File
@@ -43,7 +43,7 @@ struct __attribute__((__packed__)) heap_entry_t
{
uint16_t size;
uint16_t entry_type;
uint32_t crc32c;
uint32_t checksum;
uint64_t lsn;
uint64_t inode;
uint64_t stripe;
@@ -69,7 +69,8 @@ struct __attribute__((__packed__)) heap_entry_t
uint32_t *get_checksum(blockstore_heap_t *heap);
uint64_t big_location(blockstore_heap_t *heap);
void set_big_location(blockstore_heap_t *heap, uint64_t location);
uint32_t calc_crc32c();
uint32_t calc_checksum(blockstore_heap_t *heap);
uint32_t calc_checksum(blockstore_disk_t *dsk);
};
struct __attribute__((__packed__)) heap_small_write_t
@@ -80,7 +81,7 @@ struct __attribute__((__packed__)) heap_small_write_t
uint32_t offset;
uint32_t len;
// Also includes 1 bitmap and 1 crc32c after the bitmap if checksums are disabled
// Also includes 1 bitmap and 1 checksum after the bitmap if block checksums are disabled
};
struct __attribute__((__packed__)) heap_big_write_t
@@ -98,7 +99,7 @@ struct __attribute__((__packed__)) heap_big_intent_t
uint32_t offset;
uint32_t len;
// Also includes 2 bitmaps and 1 crc32c if checksums are disabled
// Also includes 2 bitmaps and 1 checksums if block checksums are disabled
};
struct __attribute__((__packed__)) heap_list_item_t
src/disk_tool/CMakeLists.txt
+1 -1
View File
@@ -7,7 +7,7 @@ add_executable(vitastor-disk
disk_tool.cpp disk_simple_offsets.cpp
disk_tool_discard.cpp disk_tool_journal.cpp disk_tool_meta.cpp disk_tool_prepare.cpp disk_tool_resize.cpp
disk_tool_resize_auto.cpp disk_tool_udev.cpp disk_tool_utils.cpp disk_tool_upgrade.cpp
../util/crc32c.c ../util/str_util.cpp ../util/json_util.cpp ../../json11/json11.cpp ../util/rw_blocking.cpp ../util/allocator.cpp ../util/ringloop.cpp
../util/crc32c.c ../util/xxhash.c ../util/str_util.cpp ../util/json_util.cpp ../../json11/json11.cpp ../util/rw_blocking.cpp ../util/allocator.cpp ../util/ringloop.cpp
../blockstore/blockstore_disk.cpp ../blockstore/blockstore_heap.cpp ../blockstore/multilist.cpp
)
target_link_libraries(vitastor-disk
src/disk_tool/disk_tool.h
+2 -2
View File
@@ -136,8 +136,8 @@ struct disk_tool_t
void choose_journal_block(uint32_t je_size);
int resize_rebuild_journal();
int resize_write_new_journal();
void remap_big_write(heap_entry_t *wr);
void remap_small_write(heap_entry_t *wr);
void remap_big_write(blockstore_heap_t *heap, heap_entry_t *wr);
void remap_small_write(blockstore_heap_t *heap, heap_entry_t *wr);
void fill_old_clean_entry(blockstore_heap_t *heap, heap_entry_t *big_wr);
void fill_old_journal_entry(blockstore_heap_t *heap, heap_entry_t *wr);
int resize_rebuild_meta();
src/disk_tool/disk_tool_meta.cpp
+3 -3
View File
@@ -751,7 +751,7 @@ close_err0:
{
*wr->get_checksum(&heap) = sscanf_json("%jx", write_entry["data_crc32c"]);
}
wr->crc32c = wr->calc_crc32c();
wr->checksum = wr->calc_checksum(&heap);
assert((uint8_t*)wr + wr->size == new_meta_buf + meta_offset + used_space);
}
}
@@ -794,7 +794,7 @@ close_err:
fromhexstr(meta_entry["ext_bitmap"].string_value(), new_clean_entry_bitmap_size, wr->get_ext_bitmap(&heap));
if (new_meta_hdr->data_csum_type != 0)
fromhexstr(meta_entry["data_csum"].string_value(), new_data_csum_size, wr->get_checksums(&heap));
wr->crc32c = wr->calc_crc32c();
wr->checksum = wr->calc_checksum(&heap);
assert((uint8_t*)wr + wr->size == new_meta_buf + meta_offset + used_space);
auto j_it = journal_by_object.find(oid);
if (j_it != journal_by_object.end())
@@ -860,7 +860,7 @@ close_err:
assert(0);
}
wr->size = wr->get_size(&heap);
wr->crc32c = wr->calc_crc32c();
wr->checksum = wr->calc_checksum(&heap);
assert((uint8_t*)wr + wr->size == new_meta_buf + meta_offset + used_space);
}
}
src/disk_tool/disk_tool_resize.cpp
+7 -7
View File
@@ -531,7 +531,7 @@ int disk_tool_t::resize_write_new_journal()
return 0;
}
void disk_tool_t::remap_big_write(heap_entry_t *wr)
void disk_tool_t::remap_big_write(blockstore_heap_t *heap, heap_entry_t *wr)
{
uint64_t block_num = wr->big().block_num;
auto remap_it = data_remap.find(block_num);
@@ -544,10 +544,10 @@ void disk_tool_t::remap_big_write(heap_entry_t *wr)
}
block_num += data_idx_diff;
wr->big().block_num = block_num;
wr->crc32c = wr->calc_crc32c();
wr->checksum = wr->calc_checksum(heap);
}
void disk_tool_t::remap_small_write(heap_entry_t *wr)
void disk_tool_t::remap_small_write(blockstore_heap_t *heap, heap_entry_t *wr)
{
if (new_meta_format == BLOCKSTORE_META_FORMAT_HEAP && wr->small().len > 0)
{
@@ -559,7 +559,7 @@ void disk_tool_t::remap_small_write(heap_entry_t *wr)
memcpy(new_journal_ptr, buffer_area+wr->small().location, wr->small().len);
wr->small().location = new_journal_ptr-new_journal_buf;
new_journal_ptr += wr->small().len;
wr->crc32c = wr->calc_crc32c();
wr->checksum = wr->calc_checksum(heap);
}
}
@@ -673,11 +673,11 @@ int disk_tool_t::resize_rebuild_meta()
{
if (wr->type() == BS_HEAP_BIG_WRITE || wr->type() == BS_HEAP_BIG_INTENT)
{
remap_big_write(wr);
remap_big_write(heap, wr);
}
else if (wr->type() == BS_HEAP_SMALL_WRITE)
{
remap_small_write(wr);
remap_small_write(heap, wr);
}
else if (wr->type() != BS_HEAP_DELETE && new_meta_format != BLOCKSTORE_META_FORMAT_HEAP)
{
@@ -786,7 +786,7 @@ int disk_tool_t::resize_rebuild_meta()
memcpy(((uint8_t*)wr) + sizeof(heap_big_write_t) + new_clean_entry_bitmap_size, bitmap+new_clean_entry_bitmap_size, new_clean_entry_bitmap_size);
memcpy(((uint8_t*)wr) + sizeof(heap_big_write_t) + 2*new_clean_entry_bitmap_size, bitmap+2*new_clean_entry_bitmap_size, new_data_csum_size);
}
wr->crc32c = wr->calc_crc32c();
wr->checksum = wr->calc_checksum(&dsk);
new_meta_pos += wr->size;
}
else
src/test/CMakeLists.txt
+1
View File
@@ -41,6 +41,7 @@ add_executable(test_heap
../blockstore/multilist.cpp
../blockstore/blockstore_heap.cpp
../util/crc32c.c
../util/xxhash.c
../util/allocator.cpp
../blockstore/blockstore_disk.cpp
../util/str_util.cpp
src/test/test_heap.cpp
+1 -1
View File
@@ -1117,7 +1117,7 @@ void test_corruption()
blockstore_heap_t heap(&dsk, buffer_area.data());
auto entry = ((heap_entry_t*)tmp.data());
entry->size++;
entry->crc32c = entry->calc_crc32c();
entry->checksum = entry->calc_checksum(&heap);
uint64_t entries_loaded;
assert(heap.load_blocks(0, dsk.meta_block_size, tmp.data(), false, entries_loaded) == EDOM);
}
src/util/crc32c.c
+1 -1
View File
@@ -391,7 +391,7 @@ uint32_t crc32c(uint32_t crc, const void *buf, size_t len)
#endif
static uint8_t zero_page[4096] = {};
uint8_t zero_page[4096] = {};
uint32_t crc32c_pad(uint32_t prev_crc, const void *buf, size_t len, size_t left_pad, size_t right_pad)
{
src/util/crc32c.h
+1
View File
@@ -16,6 +16,7 @@ extern "C" {
uint32_t crc32c(uint32_t crc, const void *buf, size_t len);
uint32_t crc32c_pad(uint32_t prev_crc, const void *buf, size_t len, size_t left_pad, size_t right_pad);
uint32_t crc32c_nopad(uint32_t prev_crc, const void *buf, size_t len, size_t left_pad, size_t right_pad);
extern uint8_t zero_page[4096];
#ifdef __cplusplus
};
#endif
src/util/xxhash.c
+7492
View File
File diff suppressed because it is too large Load Diff
src/util/xxhash.h
+31
View File
@@ -0,0 +1,31 @@
// 7.5k LOC header is not a header
// This is a real header for xxhash3
#pragma once
#if defined(__cplusplus) && !defined(XXH_NO_EXTERNC_GUARD)
extern "C" {
#endif
typedef enum {
XXH_OK = 0,
XXH_ERROR
} XXH_errorcode;
typedef uint64_t XXH64_hash_t;
XXH64_hash_t XXH3_64bits(const void* input, size_t length);
__attribute__((__pure__)) XXH64_hash_t XXH3_64bits( const void* input, size_t length);
__attribute__((__pure__)) XXH64_hash_t XXH3_64bits_withSeed( const void* input, size_t length, XXH64_hash_t seed);
__attribute__((__pure__)) XXH64_hash_t XXH3_64bits_withSecret( const void* data, size_t len, const void* secret, size_t secretSize);
typedef struct XXH3_state_s XXH3_state_t;
__attribute__((__malloc__)) XXH3_state_t* XXH3_createState(void);
XXH_errorcode XXH3_freeState(XXH3_state_t* statePtr);
void XXH3_copyState( XXH3_state_t* dst_state, const XXH3_state_t* src_state);
XXH_errorcode XXH3_64bits_reset( XXH3_state_t* statePtr);
XXH_errorcode XXH3_64bits_reset_withSeed( XXH3_state_t* statePtr, XXH64_hash_t seed);
XXH_errorcode XXH3_64bits_reset_withSecret( XXH3_state_t* statePtr, const void* secret, size_t secretSize);
XXH_errorcode XXH3_64bits_update ( XXH3_state_t* statePtr, const void* input, size_t length);
__attribute__((__pure__)) XXH64_hash_t XXH3_64bits_digest ( const XXH3_state_t* statePtr);
#if defined (__cplusplus) && !defined(XXH_NO_EXTERNC_GUARD)
} /* extern "C" */
#endif
tests/common.sh
+1 -1
View File
@@ -120,7 +120,7 @@ wait_condition()
done
}
VITASTOR_CFG='"etcd_address":"'$ETCD_URL'"'
VITASTOR_CFG='"etcd_address":"'$ETCD_URL'"'"$VITASTOR_CFG"
if [[ "$ETCD_SCHEME" = "https" ]]; then
VITASTOR_CFG="$VITASTOR_CFG"',"etcd_ca":"'$(pwd)'/testdata/etcd.crt"'
fi
tests/run_tests.sh
+1
View File
@@ -101,6 +101,7 @@ SCHEME=ec ./test_heal.sh
ANTIETCD=1 ./test_heal.sh
./test_checksum.sh
TEST_NAME=xxhash OSD_ARGS="--data_csum_type xxh3_32" ./test_checksum.sh
OLD=1 ./test_checksum.sh
./test_corrupt_all.sh
OLD=1 ./test_corrupt_all.sh
tests/test_checksum.sh
+6 -8
View File
@@ -2,26 +2,24 @@
OSD_ARGS="--data_csum_type crc32c --csum_block_size 32k --inmemory_journal false $OSD_ARGS"
OFFSET_ARGS="--data_csum_type crc32c --csum_block_size 32k --inmemory_journal false $OFFSET_ARGS"
PG_COUNT=${PG_COUNT:-64}
PG_COUNT=${PG_COUNT:-1}
. `dirname $0`/run_3osds.sh
check_qemu
IMG_SIZE=128
PRIMARY=$($ETCDCTL get --print-value-only /vitastor/pg/config | jq -r '.items["1"]["1"].primary')
$ETCDCTL put /vitastor/config/inode/1/1 '{"name":"testimg","size":'$((IMG_SIZE*1024*1024))'}'
# Write
$VITASTOR_FIO -bs=1M -direct=1 -iodepth=4 \
-mirror_file=./testdata/bin/mirror.bin -end_fsync=1 -rw=write -image=testimg -runtime=10
# Intentionally corrupt OSD data and restart it
kill $OSD1_PID
# Intentionally corrupt primary OSD data
data_offset=$(build/src/disk_tool/vitastor-disk simple-offsets ./testdata/bin/test_osd1.bin $OFFSET_ARGS | grep data_offset | awk '{print $2}')
truncate -s $data_offset ./testdata/bin/test_osd1.bin
dd if=/dev/zero of=./testdata/bin/test_osd1.bin bs=1024 count=1 seek=$((OSD_SIZE*1024-1))
start_osd 1
# FIXME: corrupt the journal WHEN OSD IS RUNNING and check reads too
truncate -s $data_offset ./testdata/bin/test_osd$PRIMARY.bin
dd if=/dev/zero of=./testdata/bin/test_osd$PRIMARY.bin bs=1024 count=1 seek=$((OSD_SIZE*1024-1))
# Wait until start
wait_up 10