linux/fs/btrfs/struct-funcs.c

167 lines
6.1 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0
/*
* Copyright (C) 2007 Oracle. All rights reserved.
*/
#include <asm/unaligned.h>
#include "ctree.h"
static bool check_setget_bounds(const struct extent_buffer *eb,
const void *ptr, unsigned off, int size)
{
const unsigned long member_offset = (unsigned long)ptr + off;
btrfs: remove redundant check in up check_setget_bounds There are two separate checks in the bounds checker, the first one being a special case of the second. As this function is performance critical due to checking access to any eb member, reducing the size can slightly improve performance. On a release build on x86_64 the helper is completely inlined so the function call overhead is also gone. There was a report of 5% performance drop on metadata heavy workload, that disappeared after disabling asserts. The most significant part of that is the bounds checker. https://lore.kernel.org/linux-btrfs/20200724164147.39925-1-josef@toxicpanda.com/ After the analysis, the optimized code removes the worst overhead which is the function call and the performance was restored. https://lore.kernel.org/linux-btrfs/20200730110943.GE3703@twin.jikos.cz/ 1. baseline, asserts on, setget check on run time: 46s run time with perf: 48s 2. asserts on, comment out setget check run time: 44s run time with perf: 47s So this is confirms the 5% difference 3. asserts on, optimized seget check run time: 44s run time with perf: 47s The optimizations are reducing the number of ifs to 1 and inlining the hot path. Low-level stuff, gets the performance back. Patch below. 4. asserts off, no setget check run time: 44s run time with perf: 45s This verifies that asserts other than the setget check have negligible impact on performance and it's not harmful to keep them on. Analysis where the performance is lost: * check_setget_bounds is short function, but it's still a function call, changing the flow of instructions and given how many times it's called the overhead adds up * there are two conditions, one to check if the range is completely outside (member_offset > eb->len) or partially inside (member_offset + size > eb->len) Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Signed-off-by: David Sterba <dsterba@suse.com>
2020-07-27 18:59:20 +00:00
if (unlikely(member_offset + size > eb->len)) {
btrfs_warn(eb->fs_info,
btrfs: remove redundant check in up check_setget_bounds There are two separate checks in the bounds checker, the first one being a special case of the second. As this function is performance critical due to checking access to any eb member, reducing the size can slightly improve performance. On a release build on x86_64 the helper is completely inlined so the function call overhead is also gone. There was a report of 5% performance drop on metadata heavy workload, that disappeared after disabling asserts. The most significant part of that is the bounds checker. https://lore.kernel.org/linux-btrfs/20200724164147.39925-1-josef@toxicpanda.com/ After the analysis, the optimized code removes the worst overhead which is the function call and the performance was restored. https://lore.kernel.org/linux-btrfs/20200730110943.GE3703@twin.jikos.cz/ 1. baseline, asserts on, setget check on run time: 46s run time with perf: 48s 2. asserts on, comment out setget check run time: 44s run time with perf: 47s So this is confirms the 5% difference 3. asserts on, optimized seget check run time: 44s run time with perf: 47s The optimizations are reducing the number of ifs to 1 and inlining the hot path. Low-level stuff, gets the performance back. Patch below. 4. asserts off, no setget check run time: 44s run time with perf: 45s This verifies that asserts other than the setget check have negligible impact on performance and it's not harmful to keep them on. Analysis where the performance is lost: * check_setget_bounds is short function, but it's still a function call, changing the flow of instructions and given how many times it's called the overhead adds up * there are two conditions, one to check if the range is completely outside (member_offset > eb->len) or partially inside (member_offset + size > eb->len) Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Signed-off-by: David Sterba <dsterba@suse.com>
2020-07-27 18:59:20 +00:00
"bad eb member %s: ptr 0x%lx start %llu member offset %lu size %d",
(member_offset > eb->len ? "start" : "end"),
(unsigned long)ptr, eb->start, member_offset, size);
return false;
}
return true;
}
/*
* Macro templates that define helpers to read/write extent buffer data of a
* given size, that are also used via ctree.h for access to item members by
* specialized helpers.
*
* Generic helpers:
* - btrfs_set_8 (for 8/16/32/64)
* - btrfs_get_8 (for 8/16/32/64)
*
* Generic helpers with a token (cached address of the most recently accessed
* page):
* - btrfs_set_token_8 (for 8/16/32/64)
* - btrfs_get_token_8 (for 8/16/32/64)
*
* The set/get functions handle data spanning two pages transparently, in case
* metadata block size is larger than page. Every pointer to metadata items is
* an offset into the extent buffer page array, cast to a specific type. This
* gives us all the type checking.
*
* The extent buffer pages stored in the array pages do not form a contiguous
* phyusical range, but the API functions assume the linear offset to the range
* from 0 to metadata node size.
*/
#define DEFINE_BTRFS_SETGET_BITS(bits) \
u##bits btrfs_get_token_##bits(struct btrfs_map_token *token, \
const void *ptr, unsigned long off) \
{ \
const unsigned long member_offset = (unsigned long)ptr + off; \
btrfs: handle sectorsize < PAGE_SIZE case for extent buffer accessors To support sectorsize < PAGE_SIZE case, we need to take extra care of extent buffer accessors. Since sectorsize is smaller than PAGE_SIZE, one page can contain multiple tree blocks, we must use eb->start to determine the real offset to read/write for extent buffer accessors. This patch introduces two helpers to do this: - get_eb_page_index() This is to calculate the index to access extent_buffer::pages. It's just a simple wrapper around "start >> PAGE_SHIFT". For sectorsize == PAGE_SIZE case, nothing is changed. For sectorsize < PAGE_SIZE case, we always get index as 0, and the existing page shift also works. - get_eb_offset_in_page() This is to calculate the offset to access extent_buffer::pages. This needs to take extent_buffer::start into consideration. For sectorsize == PAGE_SIZE case, extent_buffer::start is always aligned to PAGE_SIZE, thus adding extent_buffer::start to offset_in_page() won't change the result. For sectorsize < PAGE_SIZE case, adding extent_buffer::start gives us the correct offset to access. This patch will touch the following parts to cover all extent buffer accessors: - BTRFS_SETGET_HEADER_FUNCS() - read_extent_buffer() - read_extent_buffer_to_user() - memcmp_extent_buffer() - write_extent_buffer_chunk_tree_uuid() - write_extent_buffer_fsid() - write_extent_buffer() - memzero_extent_buffer() - copy_extent_buffer_full() - copy_extent_buffer() - memcpy_extent_buffer() - memmove_extent_buffer() - btrfs_get_token_##bits() - btrfs_get_##bits() - btrfs_set_token_##bits() - btrfs_set_##bits() - generic_bin_search() Signed-off-by: Goldwyn Rodrigues <rgoldwyn@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2020-12-02 06:48:04 +00:00
const unsigned long idx = get_eb_page_index(member_offset); \
const unsigned long oip = get_eb_offset_in_page(token->eb, \
member_offset); \
const int size = sizeof(u##bits); \
u8 lebytes[sizeof(u##bits)]; \
const int part = PAGE_SIZE - oip; \
\
ASSERT(token); \
ASSERT(token->kaddr); \
ASSERT(check_setget_bounds(token->eb, ptr, off, size)); \
if (token->offset <= member_offset && \
member_offset + size <= token->offset + PAGE_SIZE) { \
return get_unaligned_le##bits(token->kaddr + oip); \
} \
token->kaddr = page_address(token->eb->pages[idx]); \
token->offset = idx << PAGE_SHIFT; \
if (INLINE_EXTENT_BUFFER_PAGES == 1 || oip + size <= PAGE_SIZE ) \
return get_unaligned_le##bits(token->kaddr + oip); \
\
memcpy(lebytes, token->kaddr + oip, part); \
token->kaddr = page_address(token->eb->pages[idx + 1]); \
token->offset = (idx + 1) << PAGE_SHIFT; \
memcpy(lebytes + part, token->kaddr, size - part); \
return get_unaligned_le##bits(lebytes); \
} \
u##bits btrfs_get_##bits(const struct extent_buffer *eb, \
const void *ptr, unsigned long off) \
{ \
const unsigned long member_offset = (unsigned long)ptr + off; \
btrfs: handle sectorsize < PAGE_SIZE case for extent buffer accessors To support sectorsize < PAGE_SIZE case, we need to take extra care of extent buffer accessors. Since sectorsize is smaller than PAGE_SIZE, one page can contain multiple tree blocks, we must use eb->start to determine the real offset to read/write for extent buffer accessors. This patch introduces two helpers to do this: - get_eb_page_index() This is to calculate the index to access extent_buffer::pages. It's just a simple wrapper around "start >> PAGE_SHIFT". For sectorsize == PAGE_SIZE case, nothing is changed. For sectorsize < PAGE_SIZE case, we always get index as 0, and the existing page shift also works. - get_eb_offset_in_page() This is to calculate the offset to access extent_buffer::pages. This needs to take extent_buffer::start into consideration. For sectorsize == PAGE_SIZE case, extent_buffer::start is always aligned to PAGE_SIZE, thus adding extent_buffer::start to offset_in_page() won't change the result. For sectorsize < PAGE_SIZE case, adding extent_buffer::start gives us the correct offset to access. This patch will touch the following parts to cover all extent buffer accessors: - BTRFS_SETGET_HEADER_FUNCS() - read_extent_buffer() - read_extent_buffer_to_user() - memcmp_extent_buffer() - write_extent_buffer_chunk_tree_uuid() - write_extent_buffer_fsid() - write_extent_buffer() - memzero_extent_buffer() - copy_extent_buffer_full() - copy_extent_buffer() - memcpy_extent_buffer() - memmove_extent_buffer() - btrfs_get_token_##bits() - btrfs_get_##bits() - btrfs_set_token_##bits() - btrfs_set_##bits() - generic_bin_search() Signed-off-by: Goldwyn Rodrigues <rgoldwyn@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2020-12-02 06:48:04 +00:00
const unsigned long oip = get_eb_offset_in_page(eb, member_offset); \
const unsigned long idx = get_eb_page_index(member_offset); \
char *kaddr = page_address(eb->pages[idx]); \
const int size = sizeof(u##bits); \
const int part = PAGE_SIZE - oip; \
u8 lebytes[sizeof(u##bits)]; \
\
ASSERT(check_setget_bounds(eb, ptr, off, size)); \
if (INLINE_EXTENT_BUFFER_PAGES == 1 || oip + size <= PAGE_SIZE) \
return get_unaligned_le##bits(kaddr + oip); \
\
memcpy(lebytes, kaddr + oip, part); \
kaddr = page_address(eb->pages[idx + 1]); \
memcpy(lebytes + part, kaddr, size - part); \
return get_unaligned_le##bits(lebytes); \
} \
void btrfs_set_token_##bits(struct btrfs_map_token *token, \
const void *ptr, unsigned long off, \
u##bits val) \
{ \
const unsigned long member_offset = (unsigned long)ptr + off; \
btrfs: handle sectorsize < PAGE_SIZE case for extent buffer accessors To support sectorsize < PAGE_SIZE case, we need to take extra care of extent buffer accessors. Since sectorsize is smaller than PAGE_SIZE, one page can contain multiple tree blocks, we must use eb->start to determine the real offset to read/write for extent buffer accessors. This patch introduces two helpers to do this: - get_eb_page_index() This is to calculate the index to access extent_buffer::pages. It's just a simple wrapper around "start >> PAGE_SHIFT". For sectorsize == PAGE_SIZE case, nothing is changed. For sectorsize < PAGE_SIZE case, we always get index as 0, and the existing page shift also works. - get_eb_offset_in_page() This is to calculate the offset to access extent_buffer::pages. This needs to take extent_buffer::start into consideration. For sectorsize == PAGE_SIZE case, extent_buffer::start is always aligned to PAGE_SIZE, thus adding extent_buffer::start to offset_in_page() won't change the result. For sectorsize < PAGE_SIZE case, adding extent_buffer::start gives us the correct offset to access. This patch will touch the following parts to cover all extent buffer accessors: - BTRFS_SETGET_HEADER_FUNCS() - read_extent_buffer() - read_extent_buffer_to_user() - memcmp_extent_buffer() - write_extent_buffer_chunk_tree_uuid() - write_extent_buffer_fsid() - write_extent_buffer() - memzero_extent_buffer() - copy_extent_buffer_full() - copy_extent_buffer() - memcpy_extent_buffer() - memmove_extent_buffer() - btrfs_get_token_##bits() - btrfs_get_##bits() - btrfs_set_token_##bits() - btrfs_set_##bits() - generic_bin_search() Signed-off-by: Goldwyn Rodrigues <rgoldwyn@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2020-12-02 06:48:04 +00:00
const unsigned long idx = get_eb_page_index(member_offset); \
const unsigned long oip = get_eb_offset_in_page(token->eb, \
member_offset); \
const int size = sizeof(u##bits); \
u8 lebytes[sizeof(u##bits)]; \
const int part = PAGE_SIZE - oip; \
\
ASSERT(token); \
ASSERT(token->kaddr); \
ASSERT(check_setget_bounds(token->eb, ptr, off, size)); \
if (token->offset <= member_offset && \
member_offset + size <= token->offset + PAGE_SIZE) { \
put_unaligned_le##bits(val, token->kaddr + oip); \
return; \
} \
token->kaddr = page_address(token->eb->pages[idx]); \
token->offset = idx << PAGE_SHIFT; \
if (INLINE_EXTENT_BUFFER_PAGES == 1 || oip + size <= PAGE_SIZE) { \
put_unaligned_le##bits(val, token->kaddr + oip); \
return; \
} \
put_unaligned_le##bits(val, lebytes); \
memcpy(token->kaddr + oip, lebytes, part); \
token->kaddr = page_address(token->eb->pages[idx + 1]); \
token->offset = (idx + 1) << PAGE_SHIFT; \
memcpy(token->kaddr, lebytes + part, size - part); \
} \
void btrfs_set_##bits(const struct extent_buffer *eb, void *ptr, \
unsigned long off, u##bits val) \
{ \
const unsigned long member_offset = (unsigned long)ptr + off; \
btrfs: handle sectorsize < PAGE_SIZE case for extent buffer accessors To support sectorsize < PAGE_SIZE case, we need to take extra care of extent buffer accessors. Since sectorsize is smaller than PAGE_SIZE, one page can contain multiple tree blocks, we must use eb->start to determine the real offset to read/write for extent buffer accessors. This patch introduces two helpers to do this: - get_eb_page_index() This is to calculate the index to access extent_buffer::pages. It's just a simple wrapper around "start >> PAGE_SHIFT". For sectorsize == PAGE_SIZE case, nothing is changed. For sectorsize < PAGE_SIZE case, we always get index as 0, and the existing page shift also works. - get_eb_offset_in_page() This is to calculate the offset to access extent_buffer::pages. This needs to take extent_buffer::start into consideration. For sectorsize == PAGE_SIZE case, extent_buffer::start is always aligned to PAGE_SIZE, thus adding extent_buffer::start to offset_in_page() won't change the result. For sectorsize < PAGE_SIZE case, adding extent_buffer::start gives us the correct offset to access. This patch will touch the following parts to cover all extent buffer accessors: - BTRFS_SETGET_HEADER_FUNCS() - read_extent_buffer() - read_extent_buffer_to_user() - memcmp_extent_buffer() - write_extent_buffer_chunk_tree_uuid() - write_extent_buffer_fsid() - write_extent_buffer() - memzero_extent_buffer() - copy_extent_buffer_full() - copy_extent_buffer() - memcpy_extent_buffer() - memmove_extent_buffer() - btrfs_get_token_##bits() - btrfs_get_##bits() - btrfs_set_token_##bits() - btrfs_set_##bits() - generic_bin_search() Signed-off-by: Goldwyn Rodrigues <rgoldwyn@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2020-12-02 06:48:04 +00:00
const unsigned long oip = get_eb_offset_in_page(eb, member_offset); \
const unsigned long idx = get_eb_page_index(member_offset); \
char *kaddr = page_address(eb->pages[idx]); \
const int size = sizeof(u##bits); \
const int part = PAGE_SIZE - oip; \
u8 lebytes[sizeof(u##bits)]; \
\
ASSERT(check_setget_bounds(eb, ptr, off, size)); \
if (INLINE_EXTENT_BUFFER_PAGES == 1 || oip + size <= PAGE_SIZE) { \
put_unaligned_le##bits(val, kaddr + oip); \
return; \
} \
\
put_unaligned_le##bits(val, lebytes); \
memcpy(kaddr + oip, lebytes, part); \
kaddr = page_address(eb->pages[idx + 1]); \
memcpy(kaddr, lebytes + part, size - part); \
}
DEFINE_BTRFS_SETGET_BITS(8)
DEFINE_BTRFS_SETGET_BITS(16)
DEFINE_BTRFS_SETGET_BITS(32)
DEFINE_BTRFS_SETGET_BITS(64)
void btrfs_node_key(const struct extent_buffer *eb,
struct btrfs_disk_key *disk_key, int nr)
{
unsigned long ptr = btrfs_node_key_ptr_offset(nr);
read_eb_member(eb, (struct btrfs_key_ptr *)ptr,
struct btrfs_key_ptr, key, disk_key);
}