2018-04-04 01:23:33 +08:00
|
|
|
// SPDX-License-Identifier: GPL-2.0
|
2013-10-12 02:44:09 +08:00
|
|
|
/*
|
|
|
|
* Copyright (C) 2013 Fusion IO. All rights reserved.
|
|
|
|
*/
|
|
|
|
|
2016-06-01 19:18:25 +08:00
|
|
|
#include <linux/types.h>
|
2013-10-12 02:44:09 +08:00
|
|
|
#include "btrfs-tests.h"
|
|
|
|
#include "../ctree.h"
|
|
|
|
#include "../btrfs_inode.h"
|
|
|
|
#include "../disk-io.h"
|
|
|
|
#include "../extent_io.h"
|
|
|
|
#include "../volumes.h"
|
2016-03-10 17:26:59 +08:00
|
|
|
#include "../compression.h"
|
2022-10-19 22:51:00 +08:00
|
|
|
#include "../accessors.h"
|
2013-10-12 02:44:09 +08:00
|
|
|
|
|
|
|
static void insert_extent(struct btrfs_root *root, u64 start, u64 len,
|
|
|
|
u64 ram_bytes, u64 offset, u64 disk_bytenr,
|
|
|
|
u64 disk_len, u32 type, u8 compression, int slot)
|
|
|
|
{
|
|
|
|
struct btrfs_path path;
|
|
|
|
struct btrfs_file_extent_item *fi;
|
|
|
|
struct extent_buffer *leaf = root->node;
|
|
|
|
struct btrfs_key key;
|
|
|
|
u32 value_len = sizeof(struct btrfs_file_extent_item);
|
|
|
|
|
|
|
|
if (type == BTRFS_FILE_EXTENT_INLINE)
|
|
|
|
value_len += len;
|
|
|
|
memset(&path, 0, sizeof(path));
|
|
|
|
|
|
|
|
path.nodes[0] = leaf;
|
|
|
|
path.slots[0] = slot;
|
|
|
|
|
|
|
|
key.objectid = BTRFS_FIRST_FREE_OBJECTID;
|
|
|
|
key.type = BTRFS_EXTENT_DATA_KEY;
|
|
|
|
key.offset = start;
|
|
|
|
|
2023-09-12 20:04:29 +08:00
|
|
|
/*
|
|
|
|
* Passing a NULL trans handle is fine here, we have a dummy root eb
|
|
|
|
* and the tree is a single node (level 0).
|
|
|
|
*/
|
|
|
|
btrfs_setup_item_for_insert(NULL, root, &path, &key, value_len);
|
2013-10-12 02:44:09 +08:00
|
|
|
fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
|
|
|
|
btrfs_set_file_extent_generation(leaf, fi, 1);
|
|
|
|
btrfs_set_file_extent_type(leaf, fi, type);
|
|
|
|
btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
|
|
|
|
btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_len);
|
|
|
|
btrfs_set_file_extent_offset(leaf, fi, offset);
|
|
|
|
btrfs_set_file_extent_num_bytes(leaf, fi, len);
|
|
|
|
btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
|
|
|
|
btrfs_set_file_extent_compression(leaf, fi, compression);
|
|
|
|
btrfs_set_file_extent_encryption(leaf, fi, 0);
|
|
|
|
btrfs_set_file_extent_other_encoding(leaf, fi, 0);
|
|
|
|
}
|
|
|
|
|
2013-10-15 01:15:02 +08:00
|
|
|
static void insert_inode_item_key(struct btrfs_root *root)
|
|
|
|
{
|
|
|
|
struct btrfs_path path;
|
|
|
|
struct extent_buffer *leaf = root->node;
|
|
|
|
struct btrfs_key key;
|
|
|
|
u32 value_len = 0;
|
|
|
|
|
|
|
|
memset(&path, 0, sizeof(path));
|
|
|
|
|
|
|
|
path.nodes[0] = leaf;
|
|
|
|
path.slots[0] = 0;
|
|
|
|
|
|
|
|
key.objectid = BTRFS_INODE_ITEM_KEY;
|
|
|
|
key.type = BTRFS_INODE_ITEM_KEY;
|
|
|
|
key.offset = 0;
|
|
|
|
|
2023-09-12 20:04:29 +08:00
|
|
|
/*
|
|
|
|
* Passing a NULL trans handle is fine here, we have a dummy root eb
|
|
|
|
* and the tree is a single node (level 0).
|
|
|
|
*/
|
|
|
|
btrfs_setup_item_for_insert(NULL, root, &path, &key, value_len);
|
2013-10-15 01:15:02 +08:00
|
|
|
}
|
|
|
|
|
2013-10-12 02:44:09 +08:00
|
|
|
/*
|
|
|
|
* Build the most complicated map of extents the earth has ever seen. We want
|
|
|
|
* this so we can test all of the corner cases of btrfs_get_extent. Here is a
|
|
|
|
* diagram of how the extents will look though this may not be possible we still
|
|
|
|
* want to make sure everything acts normally (the last number is not inclusive)
|
|
|
|
*
|
2022-09-16 15:28:35 +08:00
|
|
|
* [0 - 6][ 6 - 4096 ][ 4096 - 4100][4100 - 8195][8195 - 12291]
|
|
|
|
* [inline][hole but no extent][ hole ][ regular ][regular1 split]
|
2013-10-12 02:44:09 +08:00
|
|
|
*
|
2016-06-01 19:18:25 +08:00
|
|
|
* [12291 - 16387][16387 - 24579][24579 - 28675][ 28675 - 32771][32771 - 36867 ]
|
|
|
|
* [ hole ][regular1 split][ prealloc ][ prealloc1 ][prealloc1 written]
|
2013-10-12 02:44:09 +08:00
|
|
|
*
|
2016-06-01 19:18:25 +08:00
|
|
|
* [36867 - 45059][45059 - 53251][53251 - 57347][57347 - 61443][61443- 69635]
|
|
|
|
* [ prealloc1 ][ compressed ][ compressed1 ][ regular ][ compressed1]
|
2013-10-12 02:44:09 +08:00
|
|
|
*
|
2016-06-01 19:18:25 +08:00
|
|
|
* [69635-73731][ 73731 - 86019 ][86019-90115]
|
|
|
|
* [ regular ][ hole but no extent][ regular ]
|
2013-10-12 02:44:09 +08:00
|
|
|
*/
|
2016-06-01 19:18:25 +08:00
|
|
|
static void setup_file_extents(struct btrfs_root *root, u32 sectorsize)
|
2013-10-12 02:44:09 +08:00
|
|
|
{
|
|
|
|
int slot = 0;
|
2015-12-15 00:42:10 +08:00
|
|
|
u64 disk_bytenr = SZ_1M;
|
2013-10-12 02:44:09 +08:00
|
|
|
u64 offset = 0;
|
|
|
|
|
|
|
|
/*
|
2022-09-16 15:28:35 +08:00
|
|
|
* Tree-checker has strict limits on inline extents that they can only
|
|
|
|
* exist at file offset 0, thus we can only have one inline file extent
|
|
|
|
* at most.
|
2013-10-12 02:44:09 +08:00
|
|
|
*/
|
2022-09-16 15:28:35 +08:00
|
|
|
insert_extent(root, offset, 6, 6, 0, 0, 0, BTRFS_FILE_EXTENT_INLINE, 0,
|
2013-10-12 02:44:09 +08:00
|
|
|
slot);
|
|
|
|
slot++;
|
2016-06-01 19:18:25 +08:00
|
|
|
offset = sectorsize;
|
2013-10-12 02:44:09 +08:00
|
|
|
|
|
|
|
/* Now another hole */
|
|
|
|
insert_extent(root, offset, 4, 4, 0, 0, 0, BTRFS_FILE_EXTENT_REG, 0,
|
|
|
|
slot);
|
|
|
|
slot++;
|
|
|
|
offset += 4;
|
|
|
|
|
|
|
|
/* Now for a regular extent */
|
2016-06-01 19:18:25 +08:00
|
|
|
insert_extent(root, offset, sectorsize - 1, sectorsize - 1, 0,
|
2024-04-30 06:23:03 +08:00
|
|
|
disk_bytenr, sectorsize - 1, BTRFS_FILE_EXTENT_REG, 0, slot);
|
2013-10-12 02:44:09 +08:00
|
|
|
slot++;
|
2016-06-01 19:18:25 +08:00
|
|
|
disk_bytenr += sectorsize;
|
|
|
|
offset += sectorsize - 1;
|
2013-10-12 02:44:09 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Now for 3 extents that were split from a hole punch so we test
|
|
|
|
* offsets properly.
|
|
|
|
*/
|
2016-06-01 19:18:25 +08:00
|
|
|
insert_extent(root, offset, sectorsize, 4 * sectorsize, 0, disk_bytenr,
|
|
|
|
4 * sectorsize, BTRFS_FILE_EXTENT_REG, 0, slot);
|
2013-10-12 02:44:09 +08:00
|
|
|
slot++;
|
2016-06-01 19:18:25 +08:00
|
|
|
offset += sectorsize;
|
|
|
|
insert_extent(root, offset, sectorsize, sectorsize, 0, 0, 0,
|
|
|
|
BTRFS_FILE_EXTENT_REG, 0, slot);
|
2013-10-12 02:44:09 +08:00
|
|
|
slot++;
|
2016-06-01 19:18:25 +08:00
|
|
|
offset += sectorsize;
|
|
|
|
insert_extent(root, offset, 2 * sectorsize, 4 * sectorsize,
|
|
|
|
2 * sectorsize, disk_bytenr, 4 * sectorsize,
|
2013-10-12 02:44:09 +08:00
|
|
|
BTRFS_FILE_EXTENT_REG, 0, slot);
|
|
|
|
slot++;
|
2016-06-01 19:18:25 +08:00
|
|
|
offset += 2 * sectorsize;
|
|
|
|
disk_bytenr += 4 * sectorsize;
|
2013-10-12 02:44:09 +08:00
|
|
|
|
|
|
|
/* Now for a unwritten prealloc extent */
|
2016-06-01 19:18:25 +08:00
|
|
|
insert_extent(root, offset, sectorsize, sectorsize, 0, disk_bytenr,
|
|
|
|
sectorsize, BTRFS_FILE_EXTENT_PREALLOC, 0, slot);
|
2013-10-12 02:44:09 +08:00
|
|
|
slot++;
|
2016-06-01 19:18:25 +08:00
|
|
|
offset += sectorsize;
|
2013-10-12 02:44:09 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* We want to jack up disk_bytenr a little more so the em stuff doesn't
|
|
|
|
* merge our records.
|
|
|
|
*/
|
2016-06-01 19:18:25 +08:00
|
|
|
disk_bytenr += 2 * sectorsize;
|
2013-10-12 02:44:09 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Now for a partially written prealloc extent, basically the same as
|
|
|
|
* the hole punch example above. Ram_bytes never changes when you mark
|
|
|
|
* extents written btw.
|
|
|
|
*/
|
2016-06-01 19:18:25 +08:00
|
|
|
insert_extent(root, offset, sectorsize, 4 * sectorsize, 0, disk_bytenr,
|
|
|
|
4 * sectorsize, BTRFS_FILE_EXTENT_PREALLOC, 0, slot);
|
2013-10-12 02:44:09 +08:00
|
|
|
slot++;
|
2016-06-01 19:18:25 +08:00
|
|
|
offset += sectorsize;
|
|
|
|
insert_extent(root, offset, sectorsize, 4 * sectorsize, sectorsize,
|
|
|
|
disk_bytenr, 4 * sectorsize, BTRFS_FILE_EXTENT_REG, 0,
|
|
|
|
slot);
|
2013-10-12 02:44:09 +08:00
|
|
|
slot++;
|
2016-06-01 19:18:25 +08:00
|
|
|
offset += sectorsize;
|
|
|
|
insert_extent(root, offset, 2 * sectorsize, 4 * sectorsize,
|
|
|
|
2 * sectorsize, disk_bytenr, 4 * sectorsize,
|
2013-10-12 02:44:09 +08:00
|
|
|
BTRFS_FILE_EXTENT_PREALLOC, 0, slot);
|
|
|
|
slot++;
|
2016-06-01 19:18:25 +08:00
|
|
|
offset += 2 * sectorsize;
|
|
|
|
disk_bytenr += 4 * sectorsize;
|
2013-10-12 02:44:09 +08:00
|
|
|
|
|
|
|
/* Now a normal compressed extent */
|
2016-06-01 19:18:25 +08:00
|
|
|
insert_extent(root, offset, 2 * sectorsize, 2 * sectorsize, 0,
|
|
|
|
disk_bytenr, sectorsize, BTRFS_FILE_EXTENT_REG,
|
|
|
|
BTRFS_COMPRESS_ZLIB, slot);
|
2013-10-12 02:44:09 +08:00
|
|
|
slot++;
|
2016-06-01 19:18:25 +08:00
|
|
|
offset += 2 * sectorsize;
|
2013-10-12 02:44:09 +08:00
|
|
|
/* No merges */
|
2016-06-01 19:18:25 +08:00
|
|
|
disk_bytenr += 2 * sectorsize;
|
2013-10-12 02:44:09 +08:00
|
|
|
|
|
|
|
/* Now a split compressed extent */
|
2016-06-01 19:18:25 +08:00
|
|
|
insert_extent(root, offset, sectorsize, 4 * sectorsize, 0, disk_bytenr,
|
|
|
|
sectorsize, BTRFS_FILE_EXTENT_REG,
|
|
|
|
BTRFS_COMPRESS_ZLIB, slot);
|
2013-10-12 02:44:09 +08:00
|
|
|
slot++;
|
2016-06-01 19:18:25 +08:00
|
|
|
offset += sectorsize;
|
|
|
|
insert_extent(root, offset, sectorsize, sectorsize, 0,
|
|
|
|
disk_bytenr + sectorsize, sectorsize,
|
2013-10-12 02:44:09 +08:00
|
|
|
BTRFS_FILE_EXTENT_REG, 0, slot);
|
|
|
|
slot++;
|
2016-06-01 19:18:25 +08:00
|
|
|
offset += sectorsize;
|
|
|
|
insert_extent(root, offset, 2 * sectorsize, 4 * sectorsize,
|
|
|
|
2 * sectorsize, disk_bytenr, sectorsize,
|
2013-10-12 02:44:09 +08:00
|
|
|
BTRFS_FILE_EXTENT_REG, BTRFS_COMPRESS_ZLIB, slot);
|
|
|
|
slot++;
|
2016-06-01 19:18:25 +08:00
|
|
|
offset += 2 * sectorsize;
|
|
|
|
disk_bytenr += 2 * sectorsize;
|
2013-10-12 02:44:09 +08:00
|
|
|
|
|
|
|
/* Now extents that have a hole but no hole extent */
|
2016-06-01 19:18:25 +08:00
|
|
|
insert_extent(root, offset, sectorsize, sectorsize, 0, disk_bytenr,
|
|
|
|
sectorsize, BTRFS_FILE_EXTENT_REG, 0, slot);
|
2013-10-12 02:44:09 +08:00
|
|
|
slot++;
|
2016-06-01 19:18:25 +08:00
|
|
|
offset += 4 * sectorsize;
|
|
|
|
disk_bytenr += sectorsize;
|
|
|
|
insert_extent(root, offset, sectorsize, sectorsize, 0, disk_bytenr,
|
|
|
|
sectorsize, BTRFS_FILE_EXTENT_REG, 0, slot);
|
2013-10-12 02:44:09 +08:00
|
|
|
}
|
|
|
|
|
btrfs: use the flags of an extent map to identify the compression type
Currently, in struct extent_map, we use an unsigned int (32 bits) to
identify the compression type of an extent and an unsigned long (64 bits
on a 64 bits platform, 32 bits otherwise) for flags. We are only using
6 different flags, so an unsigned long is excessive and we can use flags
to identify the compression type instead of using a dedicated 32 bits
field.
We can easily have tens or hundreds of thousands (or more) of extent maps
on busy and large filesystems, specially with compression enabled or many
or large files with tons of small extents. So it's convenient to have the
extent_map structure as small as possible in order to use less memory.
So remove the compression type field from struct extent_map, use flags
to identify the compression type and shorten the flags field from an
unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and
reduces the size of the structure from 136 bytes down to 128 bytes, using
now only two cache lines, and increases the number of extent maps we can
have per 4K page from 30 to 32. By using a u32 for the flags instead of
an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(),
but that level of atomicity is not needed as most flags are never cleared
once set (before adding an extent map to the tree), and the ones that can
be cleared or set after an extent map is added to the tree, are always
performed while holding the write lock on the extent map tree, while the
reader holds a lock on the tree or tests for a flag that never changes
once the extent map is in the tree (such as compression flags).
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-05 00:20:33 +08:00
|
|
|
static u32 prealloc_only = 0;
|
|
|
|
static u32 compressed_only = 0;
|
|
|
|
static u32 vacancy_only = 0;
|
2013-10-12 02:44:09 +08:00
|
|
|
|
2016-06-01 19:18:25 +08:00
|
|
|
static noinline int test_btrfs_get_extent(u32 sectorsize, u32 nodesize)
|
2013-10-12 02:44:09 +08:00
|
|
|
{
|
2016-06-21 02:14:09 +08:00
|
|
|
struct btrfs_fs_info *fs_info = NULL;
|
2013-10-12 02:44:09 +08:00
|
|
|
struct inode *inode = NULL;
|
|
|
|
struct btrfs_root *root = NULL;
|
|
|
|
struct extent_map *em = NULL;
|
|
|
|
u64 orig_start;
|
|
|
|
u64 disk_bytenr;
|
|
|
|
u64 offset;
|
|
|
|
int ret = -ENOMEM;
|
|
|
|
|
2019-03-18 21:06:55 +08:00
|
|
|
test_msg("running btrfs_get_extent tests");
|
|
|
|
|
2013-10-12 02:44:09 +08:00
|
|
|
inode = btrfs_new_test_inode();
|
|
|
|
if (!inode) {
|
2019-03-16 00:28:46 +08:00
|
|
|
test_std_err(TEST_ALLOC_INODE);
|
2013-10-12 02:44:09 +08:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2016-06-15 21:22:56 +08:00
|
|
|
fs_info = btrfs_alloc_dummy_fs_info(nodesize, sectorsize);
|
2016-06-21 02:14:09 +08:00
|
|
|
if (!fs_info) {
|
2019-03-16 00:28:46 +08:00
|
|
|
test_std_err(TEST_ALLOC_FS_INFO);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2016-06-15 21:22:56 +08:00
|
|
|
root = btrfs_alloc_dummy_root(fs_info);
|
2016-06-21 02:14:09 +08:00
|
|
|
if (IS_ERR(root)) {
|
2019-03-16 00:28:46 +08:00
|
|
|
test_std_err(TEST_ALLOC_ROOT);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2016-06-15 21:22:56 +08:00
|
|
|
root->node = alloc_dummy_extent_buffer(fs_info, nodesize);
|
2013-10-12 02:44:09 +08:00
|
|
|
if (!root->node) {
|
2019-03-16 00:28:46 +08:00
|
|
|
test_std_err(TEST_ALLOC_ROOT);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
btrfs_set_header_nritems(root->node, 0);
|
|
|
|
btrfs_set_header_level(root->node, 0);
|
|
|
|
ret = -EINVAL;
|
|
|
|
|
|
|
|
/* First with no extents */
|
|
|
|
BTRFS_I(inode)->root = root;
|
2024-01-08 11:20:20 +08:00
|
|
|
em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, sectorsize);
|
2013-10-12 02:44:09 +08:00
|
|
|
if (IS_ERR(em)) {
|
|
|
|
em = NULL;
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("got an error when we shouldn't have");
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:06 +08:00
|
|
|
if (em->disk_bytenr != EXTENT_MAP_HOLE) {
|
|
|
|
test_err("expected a hole, got %llu", em->disk_bytenr);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
free_extent_map(em);
|
2022-09-19 22:06:29 +08:00
|
|
|
btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
|
2013-10-12 02:44:09 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* All of the magic numbers are based on the mapping setup in
|
|
|
|
* setup_file_extents, so if you change anything there you need to
|
|
|
|
* update the comment and update the expected values below.
|
|
|
|
*/
|
2016-06-01 19:18:25 +08:00
|
|
|
setup_file_extents(root, sectorsize);
|
2013-10-12 02:44:09 +08:00
|
|
|
|
2024-01-08 11:20:20 +08:00
|
|
|
em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, (u64)-1);
|
2013-10-12 02:44:09 +08:00
|
|
|
if (IS_ERR(em)) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("got an error when we shouldn't have");
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:06 +08:00
|
|
|
if (em->disk_bytenr != EXTENT_MAP_INLINE) {
|
|
|
|
test_err("expected an inline, got %llu", em->disk_bytenr);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2016-06-01 19:18:25 +08:00
|
|
|
|
2022-09-16 15:28:35 +08:00
|
|
|
/*
|
|
|
|
* For inline extent, we always round up the em to sectorsize, as
|
|
|
|
* they are either:
|
|
|
|
*
|
|
|
|
* a) a hidden hole
|
|
|
|
* The range will be zeroed at inline extent read time.
|
|
|
|
*
|
|
|
|
* b) a file extent with unaligned bytenr
|
|
|
|
* Tree checker will reject it.
|
|
|
|
*/
|
|
|
|
if (em->start != 0 || em->len != sectorsize) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err(
|
2022-09-16 15:28:35 +08:00
|
|
|
"unexpected extent wanted start 0 len %u, got start %llu len %llu",
|
|
|
|
sectorsize, em->start, em->len);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
if (em->flags != 0) {
|
btrfs: use the flags of an extent map to identify the compression type
Currently, in struct extent_map, we use an unsigned int (32 bits) to
identify the compression type of an extent and an unsigned long (64 bits
on a 64 bits platform, 32 bits otherwise) for flags. We are only using
6 different flags, so an unsigned long is excessive and we can use flags
to identify the compression type instead of using a dedicated 32 bits
field.
We can easily have tens or hundreds of thousands (or more) of extent maps
on busy and large filesystems, specially with compression enabled or many
or large files with tons of small extents. So it's convenient to have the
extent_map structure as small as possible in order to use less memory.
So remove the compression type field from struct extent_map, use flags
to identify the compression type and shorten the flags field from an
unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and
reduces the size of the structure from 136 bytes down to 128 bytes, using
now only two cache lines, and increases the number of extent maps we can
have per 4K page from 30 to 32. By using a u32 for the flags instead of
an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(),
but that level of atomicity is not needed as most flags are never cleared
once set (before adding an extent map to the tree), and the ones that can
be cleared or set after an extent map is added to the tree, are always
performed while holding the write lock on the extent map tree, while the
reader holds a lock on the tree or tests for a flag that never changes
once the extent map is in the tree (such as compression flags).
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-05 00:20:33 +08:00
|
|
|
test_err("unexpected flags set, want 0 have %u", em->flags);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
/*
|
|
|
|
* We don't test anything else for inline since it doesn't get set
|
|
|
|
* unless we have a page for it to write into. Maybe we should change
|
|
|
|
* this?
|
|
|
|
*/
|
|
|
|
offset = em->start + em->len;
|
|
|
|
free_extent_map(em);
|
|
|
|
|
2024-01-08 11:20:20 +08:00
|
|
|
em = btrfs_get_extent(BTRFS_I(inode), NULL, offset, sectorsize);
|
2013-10-12 02:44:09 +08:00
|
|
|
if (IS_ERR(em)) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("got an error when we shouldn't have");
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:06 +08:00
|
|
|
if (em->disk_bytenr != EXTENT_MAP_HOLE) {
|
|
|
|
test_err("expected a hole, got %llu", em->disk_bytenr);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
if (em->start != offset || em->len != 4) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err(
|
|
|
|
"unexpected extent wanted start %llu len 4, got start %llu len %llu",
|
|
|
|
offset, em->start, em->len);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
if (em->flags != 0) {
|
btrfs: use the flags of an extent map to identify the compression type
Currently, in struct extent_map, we use an unsigned int (32 bits) to
identify the compression type of an extent and an unsigned long (64 bits
on a 64 bits platform, 32 bits otherwise) for flags. We are only using
6 different flags, so an unsigned long is excessive and we can use flags
to identify the compression type instead of using a dedicated 32 bits
field.
We can easily have tens or hundreds of thousands (or more) of extent maps
on busy and large filesystems, specially with compression enabled or many
or large files with tons of small extents. So it's convenient to have the
extent_map structure as small as possible in order to use less memory.
So remove the compression type field from struct extent_map, use flags
to identify the compression type and shorten the flags field from an
unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and
reduces the size of the structure from 136 bytes down to 128 bytes, using
now only two cache lines, and increases the number of extent maps we can
have per 4K page from 30 to 32. By using a u32 for the flags instead of
an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(),
but that level of atomicity is not needed as most flags are never cleared
once set (before adding an extent map to the tree), and the ones that can
be cleared or set after an extent map is added to the tree, are always
performed while holding the write lock on the extent map tree, while the
reader holds a lock on the tree or tests for a flag that never changes
once the extent map is in the tree (such as compression flags).
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-05 00:20:33 +08:00
|
|
|
test_err("unexpected flags set, want 0 have %u", em->flags);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
offset = em->start + em->len;
|
|
|
|
free_extent_map(em);
|
|
|
|
|
|
|
|
/* Regular extent */
|
2024-01-08 11:20:20 +08:00
|
|
|
em = btrfs_get_extent(BTRFS_I(inode), NULL, offset, sectorsize);
|
2013-10-12 02:44:09 +08:00
|
|
|
if (IS_ERR(em)) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("got an error when we shouldn't have");
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:06 +08:00
|
|
|
if (em->disk_bytenr >= EXTENT_MAP_LAST_BYTE) {
|
|
|
|
test_err("expected a real extent, got %llu", em->disk_bytenr);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2016-06-01 19:18:25 +08:00
|
|
|
if (em->start != offset || em->len != sectorsize - 1) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err(
|
|
|
|
"unexpected extent wanted start %llu len 4095, got start %llu len %llu",
|
|
|
|
offset, em->start, em->len);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
if (em->flags != 0) {
|
btrfs: use the flags of an extent map to identify the compression type
Currently, in struct extent_map, we use an unsigned int (32 bits) to
identify the compression type of an extent and an unsigned long (64 bits
on a 64 bits platform, 32 bits otherwise) for flags. We are only using
6 different flags, so an unsigned long is excessive and we can use flags
to identify the compression type instead of using a dedicated 32 bits
field.
We can easily have tens or hundreds of thousands (or more) of extent maps
on busy and large filesystems, specially with compression enabled or many
or large files with tons of small extents. So it's convenient to have the
extent_map structure as small as possible in order to use less memory.
So remove the compression type field from struct extent_map, use flags
to identify the compression type and shorten the flags field from an
unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and
reduces the size of the structure from 136 bytes down to 128 bytes, using
now only two cache lines, and increases the number of extent maps we can
have per 4K page from 30 to 32. By using a u32 for the flags instead of
an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(),
but that level of atomicity is not needed as most flags are never cleared
once set (before adding an extent map to the tree), and the ones that can
be cleared or set after an extent map is added to the tree, are always
performed while holding the write lock on the extent map tree, while the
reader holds a lock on the tree or tests for a flag that never changes
once the extent map is in the tree (such as compression flags).
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-05 00:20:33 +08:00
|
|
|
test_err("unexpected flags set, want 0 have %u", em->flags);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:04 +08:00
|
|
|
if (em->offset != 0) {
|
|
|
|
test_err("wrong offset, want 0, have %llu", em->offset);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
offset = em->start + em->len;
|
|
|
|
free_extent_map(em);
|
|
|
|
|
|
|
|
/* The next 3 are split extents */
|
2024-01-08 11:20:20 +08:00
|
|
|
em = btrfs_get_extent(BTRFS_I(inode), NULL, offset, sectorsize);
|
2013-10-12 02:44:09 +08:00
|
|
|
if (IS_ERR(em)) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("got an error when we shouldn't have");
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:06 +08:00
|
|
|
if (em->disk_bytenr >= EXTENT_MAP_LAST_BYTE) {
|
|
|
|
test_err("expected a real extent, got %llu", em->disk_bytenr);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2016-06-01 19:18:25 +08:00
|
|
|
if (em->start != offset || em->len != sectorsize) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err(
|
|
|
|
"unexpected extent start %llu len %u, got start %llu len %llu",
|
2016-06-01 19:18:25 +08:00
|
|
|
offset, sectorsize, em->start, em->len);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
if (em->flags != 0) {
|
btrfs: use the flags of an extent map to identify the compression type
Currently, in struct extent_map, we use an unsigned int (32 bits) to
identify the compression type of an extent and an unsigned long (64 bits
on a 64 bits platform, 32 bits otherwise) for flags. We are only using
6 different flags, so an unsigned long is excessive and we can use flags
to identify the compression type instead of using a dedicated 32 bits
field.
We can easily have tens or hundreds of thousands (or more) of extent maps
on busy and large filesystems, specially with compression enabled or many
or large files with tons of small extents. So it's convenient to have the
extent_map structure as small as possible in order to use less memory.
So remove the compression type field from struct extent_map, use flags
to identify the compression type and shorten the flags field from an
unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and
reduces the size of the structure from 136 bytes down to 128 bytes, using
now only two cache lines, and increases the number of extent maps we can
have per 4K page from 30 to 32. By using a u32 for the flags instead of
an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(),
but that level of atomicity is not needed as most flags are never cleared
once set (before adding an extent map to the tree), and the ones that can
be cleared or set after an extent map is added to the tree, are always
performed while holding the write lock on the extent map tree, while the
reader holds a lock on the tree or tests for a flag that never changes
once the extent map is in the tree (such as compression flags).
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-05 00:20:33 +08:00
|
|
|
test_err("unexpected flags set, want 0 have %u", em->flags);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:04 +08:00
|
|
|
if (em->offset != 0) {
|
|
|
|
test_err("wrong offset, want 0, have %llu", em->offset);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:06 +08:00
|
|
|
disk_bytenr = extent_map_block_start(em);
|
2013-10-12 02:44:09 +08:00
|
|
|
orig_start = em->start;
|
|
|
|
offset = em->start + em->len;
|
|
|
|
free_extent_map(em);
|
|
|
|
|
2024-01-08 11:20:20 +08:00
|
|
|
em = btrfs_get_extent(BTRFS_I(inode), NULL, offset, sectorsize);
|
2013-10-12 02:44:09 +08:00
|
|
|
if (IS_ERR(em)) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("got an error when we shouldn't have");
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:06 +08:00
|
|
|
if (em->disk_bytenr != EXTENT_MAP_HOLE) {
|
|
|
|
test_err("expected a hole, got %llu", em->disk_bytenr);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2016-06-01 19:18:25 +08:00
|
|
|
if (em->start != offset || em->len != sectorsize) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err(
|
|
|
|
"unexpected extent wanted start %llu len %u, got start %llu len %llu",
|
2016-06-01 19:18:25 +08:00
|
|
|
offset, sectorsize, em->start, em->len);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
if (em->flags != 0) {
|
btrfs: use the flags of an extent map to identify the compression type
Currently, in struct extent_map, we use an unsigned int (32 bits) to
identify the compression type of an extent and an unsigned long (64 bits
on a 64 bits platform, 32 bits otherwise) for flags. We are only using
6 different flags, so an unsigned long is excessive and we can use flags
to identify the compression type instead of using a dedicated 32 bits
field.
We can easily have tens or hundreds of thousands (or more) of extent maps
on busy and large filesystems, specially with compression enabled or many
or large files with tons of small extents. So it's convenient to have the
extent_map structure as small as possible in order to use less memory.
So remove the compression type field from struct extent_map, use flags
to identify the compression type and shorten the flags field from an
unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and
reduces the size of the structure from 136 bytes down to 128 bytes, using
now only two cache lines, and increases the number of extent maps we can
have per 4K page from 30 to 32. By using a u32 for the flags instead of
an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(),
but that level of atomicity is not needed as most flags are never cleared
once set (before adding an extent map to the tree), and the ones that can
be cleared or set after an extent map is added to the tree, are always
performed while holding the write lock on the extent map tree, while the
reader holds a lock on the tree or tests for a flag that never changes
once the extent map is in the tree (such as compression flags).
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-05 00:20:33 +08:00
|
|
|
test_err("unexpected flags set, want 0 have %u", em->flags);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
offset = em->start + em->len;
|
|
|
|
free_extent_map(em);
|
|
|
|
|
2024-01-08 11:20:20 +08:00
|
|
|
em = btrfs_get_extent(BTRFS_I(inode), NULL, offset, sectorsize);
|
2013-10-12 02:44:09 +08:00
|
|
|
if (IS_ERR(em)) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("got an error when we shouldn't have");
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:06 +08:00
|
|
|
if (em->disk_bytenr >= EXTENT_MAP_LAST_BYTE) {
|
|
|
|
test_err("expected a real extent, got %llu", em->disk_bytenr);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2016-06-01 19:18:25 +08:00
|
|
|
if (em->start != offset || em->len != 2 * sectorsize) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err(
|
|
|
|
"unexpected extent wanted start %llu len %u, got start %llu len %llu",
|
2016-06-01 19:18:25 +08:00
|
|
|
offset, 2 * sectorsize, em->start, em->len);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
if (em->flags != 0) {
|
btrfs: use the flags of an extent map to identify the compression type
Currently, in struct extent_map, we use an unsigned int (32 bits) to
identify the compression type of an extent and an unsigned long (64 bits
on a 64 bits platform, 32 bits otherwise) for flags. We are only using
6 different flags, so an unsigned long is excessive and we can use flags
to identify the compression type instead of using a dedicated 32 bits
field.
We can easily have tens or hundreds of thousands (or more) of extent maps
on busy and large filesystems, specially with compression enabled or many
or large files with tons of small extents. So it's convenient to have the
extent_map structure as small as possible in order to use less memory.
So remove the compression type field from struct extent_map, use flags
to identify the compression type and shorten the flags field from an
unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and
reduces the size of the structure from 136 bytes down to 128 bytes, using
now only two cache lines, and increases the number of extent maps we can
have per 4K page from 30 to 32. By using a u32 for the flags instead of
an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(),
but that level of atomicity is not needed as most flags are never cleared
once set (before adding an extent map to the tree), and the ones that can
be cleared or set after an extent map is added to the tree, are always
performed while holding the write lock on the extent map tree, while the
reader holds a lock on the tree or tests for a flag that never changes
once the extent map is in the tree (such as compression flags).
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-05 00:20:33 +08:00
|
|
|
test_err("unexpected flags set, want 0 have %u", em->flags);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:04 +08:00
|
|
|
if (em->start - em->offset != orig_start) {
|
|
|
|
test_err("wrong offset, em->start=%llu em->offset=%llu orig_start=%llu",
|
|
|
|
em->start, em->offset, orig_start);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
disk_bytenr += (em->start - orig_start);
|
2024-04-30 06:23:06 +08:00
|
|
|
if (extent_map_block_start(em) != disk_bytenr) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("wrong block start, want %llu, have %llu",
|
2024-04-30 06:23:06 +08:00
|
|
|
disk_bytenr, extent_map_block_start(em));
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
offset = em->start + em->len;
|
|
|
|
free_extent_map(em);
|
|
|
|
|
|
|
|
/* Prealloc extent */
|
2024-01-08 11:20:20 +08:00
|
|
|
em = btrfs_get_extent(BTRFS_I(inode), NULL, offset, sectorsize);
|
2013-10-12 02:44:09 +08:00
|
|
|
if (IS_ERR(em)) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("got an error when we shouldn't have");
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:06 +08:00
|
|
|
if (em->disk_bytenr >= EXTENT_MAP_LAST_BYTE) {
|
|
|
|
test_err("expected a real extent, got %llu", em->disk_bytenr);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2016-06-01 19:18:25 +08:00
|
|
|
if (em->start != offset || em->len != sectorsize) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err(
|
|
|
|
"unexpected extent wanted start %llu len %u, got start %llu len %llu",
|
2016-06-01 19:18:25 +08:00
|
|
|
offset, sectorsize, em->start, em->len);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
if (em->flags != prealloc_only) {
|
btrfs: use the flags of an extent map to identify the compression type
Currently, in struct extent_map, we use an unsigned int (32 bits) to
identify the compression type of an extent and an unsigned long (64 bits
on a 64 bits platform, 32 bits otherwise) for flags. We are only using
6 different flags, so an unsigned long is excessive and we can use flags
to identify the compression type instead of using a dedicated 32 bits
field.
We can easily have tens or hundreds of thousands (or more) of extent maps
on busy and large filesystems, specially with compression enabled or many
or large files with tons of small extents. So it's convenient to have the
extent_map structure as small as possible in order to use less memory.
So remove the compression type field from struct extent_map, use flags
to identify the compression type and shorten the flags field from an
unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and
reduces the size of the structure from 136 bytes down to 128 bytes, using
now only two cache lines, and increases the number of extent maps we can
have per 4K page from 30 to 32. By using a u32 for the flags instead of
an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(),
but that level of atomicity is not needed as most flags are never cleared
once set (before adding an extent map to the tree), and the ones that can
be cleared or set after an extent map is added to the tree, are always
performed while holding the write lock on the extent map tree, while the
reader holds a lock on the tree or tests for a flag that never changes
once the extent map is in the tree (such as compression flags).
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-05 00:20:33 +08:00
|
|
|
test_err("unexpected flags set, want %u have %u",
|
2013-10-12 02:44:09 +08:00
|
|
|
prealloc_only, em->flags);
|
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:04 +08:00
|
|
|
if (em->offset != 0) {
|
|
|
|
test_err("wrong offset, want 0, have %llu", em->offset);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
offset = em->start + em->len;
|
|
|
|
free_extent_map(em);
|
|
|
|
|
|
|
|
/* The next 3 are a half written prealloc extent */
|
2024-01-08 11:20:20 +08:00
|
|
|
em = btrfs_get_extent(BTRFS_I(inode), NULL, offset, sectorsize);
|
2013-10-12 02:44:09 +08:00
|
|
|
if (IS_ERR(em)) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("got an error when we shouldn't have");
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:06 +08:00
|
|
|
if (em->disk_bytenr >= EXTENT_MAP_LAST_BYTE) {
|
|
|
|
test_err("expected a real extent, got %llu", em->disk_bytenr);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2016-06-01 19:18:25 +08:00
|
|
|
if (em->start != offset || em->len != sectorsize) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err(
|
|
|
|
"unexpected extent wanted start %llu len %u, got start %llu len %llu",
|
2016-06-01 19:18:25 +08:00
|
|
|
offset, sectorsize, em->start, em->len);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
if (em->flags != prealloc_only) {
|
btrfs: use the flags of an extent map to identify the compression type
Currently, in struct extent_map, we use an unsigned int (32 bits) to
identify the compression type of an extent and an unsigned long (64 bits
on a 64 bits platform, 32 bits otherwise) for flags. We are only using
6 different flags, so an unsigned long is excessive and we can use flags
to identify the compression type instead of using a dedicated 32 bits
field.
We can easily have tens or hundreds of thousands (or more) of extent maps
on busy and large filesystems, specially with compression enabled or many
or large files with tons of small extents. So it's convenient to have the
extent_map structure as small as possible in order to use less memory.
So remove the compression type field from struct extent_map, use flags
to identify the compression type and shorten the flags field from an
unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and
reduces the size of the structure from 136 bytes down to 128 bytes, using
now only two cache lines, and increases the number of extent maps we can
have per 4K page from 30 to 32. By using a u32 for the flags instead of
an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(),
but that level of atomicity is not needed as most flags are never cleared
once set (before adding an extent map to the tree), and the ones that can
be cleared or set after an extent map is added to the tree, are always
performed while holding the write lock on the extent map tree, while the
reader holds a lock on the tree or tests for a flag that never changes
once the extent map is in the tree (such as compression flags).
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-05 00:20:33 +08:00
|
|
|
test_err("unexpected flags set, want %u have %u",
|
2013-10-12 02:44:09 +08:00
|
|
|
prealloc_only, em->flags);
|
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:04 +08:00
|
|
|
if (em->offset != 0) {
|
|
|
|
test_err("wrong offset, want 0, have %llu", em->offset);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:06 +08:00
|
|
|
disk_bytenr = extent_map_block_start(em);
|
2013-10-12 02:44:09 +08:00
|
|
|
orig_start = em->start;
|
|
|
|
offset = em->start + em->len;
|
|
|
|
free_extent_map(em);
|
|
|
|
|
2024-01-08 11:20:20 +08:00
|
|
|
em = btrfs_get_extent(BTRFS_I(inode), NULL, offset, sectorsize);
|
2013-10-12 02:44:09 +08:00
|
|
|
if (IS_ERR(em)) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("got an error when we shouldn't have");
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:06 +08:00
|
|
|
if (em->disk_bytenr >= EXTENT_MAP_HOLE) {
|
|
|
|
test_err("expected a real extent, got %llu", em->disk_bytenr);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2016-06-01 19:18:25 +08:00
|
|
|
if (em->start != offset || em->len != sectorsize) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err(
|
|
|
|
"unexpected extent wanted start %llu len %u, got start %llu len %llu",
|
2016-06-01 19:18:25 +08:00
|
|
|
offset, sectorsize, em->start, em->len);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
if (em->flags != 0) {
|
btrfs: use the flags of an extent map to identify the compression type
Currently, in struct extent_map, we use an unsigned int (32 bits) to
identify the compression type of an extent and an unsigned long (64 bits
on a 64 bits platform, 32 bits otherwise) for flags. We are only using
6 different flags, so an unsigned long is excessive and we can use flags
to identify the compression type instead of using a dedicated 32 bits
field.
We can easily have tens or hundreds of thousands (or more) of extent maps
on busy and large filesystems, specially with compression enabled or many
or large files with tons of small extents. So it's convenient to have the
extent_map structure as small as possible in order to use less memory.
So remove the compression type field from struct extent_map, use flags
to identify the compression type and shorten the flags field from an
unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and
reduces the size of the structure from 136 bytes down to 128 bytes, using
now only two cache lines, and increases the number of extent maps we can
have per 4K page from 30 to 32. By using a u32 for the flags instead of
an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(),
but that level of atomicity is not needed as most flags are never cleared
once set (before adding an extent map to the tree), and the ones that can
be cleared or set after an extent map is added to the tree, are always
performed while holding the write lock on the extent map tree, while the
reader holds a lock on the tree or tests for a flag that never changes
once the extent map is in the tree (such as compression flags).
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-05 00:20:33 +08:00
|
|
|
test_err("unexpected flags set, want 0 have %u", em->flags);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:04 +08:00
|
|
|
if (em->start - em->offset != orig_start) {
|
|
|
|
test_err("unexpected offset, wanted %llu, have %llu",
|
|
|
|
em->start - orig_start, em->offset);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:06 +08:00
|
|
|
if (extent_map_block_start(em) != disk_bytenr + em->offset) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("unexpected block start, wanted %llu, have %llu",
|
2024-04-30 06:23:06 +08:00
|
|
|
disk_bytenr + em->offset, extent_map_block_start(em));
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
offset = em->start + em->len;
|
|
|
|
free_extent_map(em);
|
|
|
|
|
2024-01-08 11:20:20 +08:00
|
|
|
em = btrfs_get_extent(BTRFS_I(inode), NULL, offset, sectorsize);
|
2013-10-12 02:44:09 +08:00
|
|
|
if (IS_ERR(em)) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("got an error when we shouldn't have");
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:06 +08:00
|
|
|
if (em->disk_bytenr >= EXTENT_MAP_LAST_BYTE) {
|
|
|
|
test_err("expected a real extent, got %llu", em->disk_bytenr);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2016-06-01 19:18:25 +08:00
|
|
|
if (em->start != offset || em->len != 2 * sectorsize) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err(
|
|
|
|
"unexpected extent wanted start %llu len %u, got start %llu len %llu",
|
2016-06-01 19:18:25 +08:00
|
|
|
offset, 2 * sectorsize, em->start, em->len);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
if (em->flags != prealloc_only) {
|
btrfs: use the flags of an extent map to identify the compression type
Currently, in struct extent_map, we use an unsigned int (32 bits) to
identify the compression type of an extent and an unsigned long (64 bits
on a 64 bits platform, 32 bits otherwise) for flags. We are only using
6 different flags, so an unsigned long is excessive and we can use flags
to identify the compression type instead of using a dedicated 32 bits
field.
We can easily have tens or hundreds of thousands (or more) of extent maps
on busy and large filesystems, specially with compression enabled or many
or large files with tons of small extents. So it's convenient to have the
extent_map structure as small as possible in order to use less memory.
So remove the compression type field from struct extent_map, use flags
to identify the compression type and shorten the flags field from an
unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and
reduces the size of the structure from 136 bytes down to 128 bytes, using
now only two cache lines, and increases the number of extent maps we can
have per 4K page from 30 to 32. By using a u32 for the flags instead of
an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(),
but that level of atomicity is not needed as most flags are never cleared
once set (before adding an extent map to the tree), and the ones that can
be cleared or set after an extent map is added to the tree, are always
performed while holding the write lock on the extent map tree, while the
reader holds a lock on the tree or tests for a flag that never changes
once the extent map is in the tree (such as compression flags).
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-05 00:20:33 +08:00
|
|
|
test_err("unexpected flags set, want %u have %u",
|
2013-10-12 02:44:09 +08:00
|
|
|
prealloc_only, em->flags);
|
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:04 +08:00
|
|
|
if (em->start - em->offset != orig_start) {
|
|
|
|
test_err("wrong offset, em->start=%llu em->offset=%llu orig_start=%llu",
|
|
|
|
em->start, em->offset, orig_start);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:06 +08:00
|
|
|
if (extent_map_block_start(em) != disk_bytenr + em->offset) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("unexpected block start, wanted %llu, have %llu",
|
2024-04-30 06:23:06 +08:00
|
|
|
disk_bytenr + em->offset, extent_map_block_start(em));
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
offset = em->start + em->len;
|
|
|
|
free_extent_map(em);
|
|
|
|
|
|
|
|
/* Now for the compressed extent */
|
2024-01-08 11:20:20 +08:00
|
|
|
em = btrfs_get_extent(BTRFS_I(inode), NULL, offset, sectorsize);
|
2013-10-12 02:44:09 +08:00
|
|
|
if (IS_ERR(em)) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("got an error when we shouldn't have");
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:06 +08:00
|
|
|
if (em->disk_bytenr >= EXTENT_MAP_LAST_BYTE) {
|
|
|
|
test_err("expected a real extent, got %llu", em->disk_bytenr);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2016-06-01 19:18:25 +08:00
|
|
|
if (em->start != offset || em->len != 2 * sectorsize) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err(
|
|
|
|
"unexpected extent wanted start %llu len %u, got start %llu len %llu",
|
2016-06-01 19:18:25 +08:00
|
|
|
offset, 2 * sectorsize, em->start, em->len);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
if (em->flags != compressed_only) {
|
btrfs: use the flags of an extent map to identify the compression type
Currently, in struct extent_map, we use an unsigned int (32 bits) to
identify the compression type of an extent and an unsigned long (64 bits
on a 64 bits platform, 32 bits otherwise) for flags. We are only using
6 different flags, so an unsigned long is excessive and we can use flags
to identify the compression type instead of using a dedicated 32 bits
field.
We can easily have tens or hundreds of thousands (or more) of extent maps
on busy and large filesystems, specially with compression enabled or many
or large files with tons of small extents. So it's convenient to have the
extent_map structure as small as possible in order to use less memory.
So remove the compression type field from struct extent_map, use flags
to identify the compression type and shorten the flags field from an
unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and
reduces the size of the structure from 136 bytes down to 128 bytes, using
now only two cache lines, and increases the number of extent maps we can
have per 4K page from 30 to 32. By using a u32 for the flags instead of
an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(),
but that level of atomicity is not needed as most flags are never cleared
once set (before adding an extent map to the tree), and the ones that can
be cleared or set after an extent map is added to the tree, are always
performed while holding the write lock on the extent map tree, while the
reader holds a lock on the tree or tests for a flag that never changes
once the extent map is in the tree (such as compression flags).
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-05 00:20:33 +08:00
|
|
|
test_err("unexpected flags set, want %u have %u",
|
2013-10-12 02:44:09 +08:00
|
|
|
compressed_only, em->flags);
|
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:04 +08:00
|
|
|
if (em->offset != 0) {
|
|
|
|
test_err("wrong offset, want 0, have %llu", em->offset);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
btrfs: use the flags of an extent map to identify the compression type
Currently, in struct extent_map, we use an unsigned int (32 bits) to
identify the compression type of an extent and an unsigned long (64 bits
on a 64 bits platform, 32 bits otherwise) for flags. We are only using
6 different flags, so an unsigned long is excessive and we can use flags
to identify the compression type instead of using a dedicated 32 bits
field.
We can easily have tens or hundreds of thousands (or more) of extent maps
on busy and large filesystems, specially with compression enabled or many
or large files with tons of small extents. So it's convenient to have the
extent_map structure as small as possible in order to use less memory.
So remove the compression type field from struct extent_map, use flags
to identify the compression type and shorten the flags field from an
unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and
reduces the size of the structure from 136 bytes down to 128 bytes, using
now only two cache lines, and increases the number of extent maps we can
have per 4K page from 30 to 32. By using a u32 for the flags instead of
an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(),
but that level of atomicity is not needed as most flags are never cleared
once set (before adding an extent map to the tree), and the ones that can
be cleared or set after an extent map is added to the tree, are always
performed while holding the write lock on the extent map tree, while the
reader holds a lock on the tree or tests for a flag that never changes
once the extent map is in the tree (such as compression flags).
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-05 00:20:33 +08:00
|
|
|
if (extent_map_compression(em) != BTRFS_COMPRESS_ZLIB) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("unexpected compress type, wanted %d, got %d",
|
btrfs: use the flags of an extent map to identify the compression type
Currently, in struct extent_map, we use an unsigned int (32 bits) to
identify the compression type of an extent and an unsigned long (64 bits
on a 64 bits platform, 32 bits otherwise) for flags. We are only using
6 different flags, so an unsigned long is excessive and we can use flags
to identify the compression type instead of using a dedicated 32 bits
field.
We can easily have tens or hundreds of thousands (or more) of extent maps
on busy and large filesystems, specially with compression enabled or many
or large files with tons of small extents. So it's convenient to have the
extent_map structure as small as possible in order to use less memory.
So remove the compression type field from struct extent_map, use flags
to identify the compression type and shorten the flags field from an
unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and
reduces the size of the structure from 136 bytes down to 128 bytes, using
now only two cache lines, and increases the number of extent maps we can
have per 4K page from 30 to 32. By using a u32 for the flags instead of
an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(),
but that level of atomicity is not needed as most flags are never cleared
once set (before adding an extent map to the tree), and the ones that can
be cleared or set after an extent map is added to the tree, are always
performed while holding the write lock on the extent map tree, while the
reader holds a lock on the tree or tests for a flag that never changes
once the extent map is in the tree (such as compression flags).
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-05 00:20:33 +08:00
|
|
|
BTRFS_COMPRESS_ZLIB, extent_map_compression(em));
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
offset = em->start + em->len;
|
|
|
|
free_extent_map(em);
|
|
|
|
|
|
|
|
/* Split compressed extent */
|
2024-01-08 11:20:20 +08:00
|
|
|
em = btrfs_get_extent(BTRFS_I(inode), NULL, offset, sectorsize);
|
2013-10-12 02:44:09 +08:00
|
|
|
if (IS_ERR(em)) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("got an error when we shouldn't have");
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:06 +08:00
|
|
|
if (em->disk_bytenr >= EXTENT_MAP_LAST_BYTE) {
|
|
|
|
test_err("expected a real extent, got %llu", em->disk_bytenr);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2016-06-01 19:18:25 +08:00
|
|
|
if (em->start != offset || em->len != sectorsize) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err(
|
|
|
|
"unexpected extent wanted start %llu len %u, got start %llu len %llu",
|
2016-06-01 19:18:25 +08:00
|
|
|
offset, sectorsize, em->start, em->len);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
if (em->flags != compressed_only) {
|
btrfs: use the flags of an extent map to identify the compression type
Currently, in struct extent_map, we use an unsigned int (32 bits) to
identify the compression type of an extent and an unsigned long (64 bits
on a 64 bits platform, 32 bits otherwise) for flags. We are only using
6 different flags, so an unsigned long is excessive and we can use flags
to identify the compression type instead of using a dedicated 32 bits
field.
We can easily have tens or hundreds of thousands (or more) of extent maps
on busy and large filesystems, specially with compression enabled or many
or large files with tons of small extents. So it's convenient to have the
extent_map structure as small as possible in order to use less memory.
So remove the compression type field from struct extent_map, use flags
to identify the compression type and shorten the flags field from an
unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and
reduces the size of the structure from 136 bytes down to 128 bytes, using
now only two cache lines, and increases the number of extent maps we can
have per 4K page from 30 to 32. By using a u32 for the flags instead of
an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(),
but that level of atomicity is not needed as most flags are never cleared
once set (before adding an extent map to the tree), and the ones that can
be cleared or set after an extent map is added to the tree, are always
performed while holding the write lock on the extent map tree, while the
reader holds a lock on the tree or tests for a flag that never changes
once the extent map is in the tree (such as compression flags).
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-05 00:20:33 +08:00
|
|
|
test_err("unexpected flags set, want %u have %u",
|
2013-10-12 02:44:09 +08:00
|
|
|
compressed_only, em->flags);
|
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:04 +08:00
|
|
|
if (em->offset != 0) {
|
|
|
|
test_err("wrong offset, want 0, have %llu", em->offset);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
btrfs: use the flags of an extent map to identify the compression type
Currently, in struct extent_map, we use an unsigned int (32 bits) to
identify the compression type of an extent and an unsigned long (64 bits
on a 64 bits platform, 32 bits otherwise) for flags. We are only using
6 different flags, so an unsigned long is excessive and we can use flags
to identify the compression type instead of using a dedicated 32 bits
field.
We can easily have tens or hundreds of thousands (or more) of extent maps
on busy and large filesystems, specially with compression enabled or many
or large files with tons of small extents. So it's convenient to have the
extent_map structure as small as possible in order to use less memory.
So remove the compression type field from struct extent_map, use flags
to identify the compression type and shorten the flags field from an
unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and
reduces the size of the structure from 136 bytes down to 128 bytes, using
now only two cache lines, and increases the number of extent maps we can
have per 4K page from 30 to 32. By using a u32 for the flags instead of
an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(),
but that level of atomicity is not needed as most flags are never cleared
once set (before adding an extent map to the tree), and the ones that can
be cleared or set after an extent map is added to the tree, are always
performed while holding the write lock on the extent map tree, while the
reader holds a lock on the tree or tests for a flag that never changes
once the extent map is in the tree (such as compression flags).
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-05 00:20:33 +08:00
|
|
|
if (extent_map_compression(em) != BTRFS_COMPRESS_ZLIB) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("unexpected compress type, wanted %d, got %d",
|
btrfs: use the flags of an extent map to identify the compression type
Currently, in struct extent_map, we use an unsigned int (32 bits) to
identify the compression type of an extent and an unsigned long (64 bits
on a 64 bits platform, 32 bits otherwise) for flags. We are only using
6 different flags, so an unsigned long is excessive and we can use flags
to identify the compression type instead of using a dedicated 32 bits
field.
We can easily have tens or hundreds of thousands (or more) of extent maps
on busy and large filesystems, specially with compression enabled or many
or large files with tons of small extents. So it's convenient to have the
extent_map structure as small as possible in order to use less memory.
So remove the compression type field from struct extent_map, use flags
to identify the compression type and shorten the flags field from an
unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and
reduces the size of the structure from 136 bytes down to 128 bytes, using
now only two cache lines, and increases the number of extent maps we can
have per 4K page from 30 to 32. By using a u32 for the flags instead of
an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(),
but that level of atomicity is not needed as most flags are never cleared
once set (before adding an extent map to the tree), and the ones that can
be cleared or set after an extent map is added to the tree, are always
performed while holding the write lock on the extent map tree, while the
reader holds a lock on the tree or tests for a flag that never changes
once the extent map is in the tree (such as compression flags).
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-05 00:20:33 +08:00
|
|
|
BTRFS_COMPRESS_ZLIB, extent_map_compression(em));
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:06 +08:00
|
|
|
disk_bytenr = extent_map_block_start(em);
|
2013-10-12 02:44:09 +08:00
|
|
|
orig_start = em->start;
|
|
|
|
offset = em->start + em->len;
|
|
|
|
free_extent_map(em);
|
|
|
|
|
2024-01-08 11:20:20 +08:00
|
|
|
em = btrfs_get_extent(BTRFS_I(inode), NULL, offset, sectorsize);
|
2013-10-12 02:44:09 +08:00
|
|
|
if (IS_ERR(em)) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("got an error when we shouldn't have");
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:06 +08:00
|
|
|
if (em->disk_bytenr >= EXTENT_MAP_LAST_BYTE) {
|
|
|
|
test_err("expected a real extent, got %llu", em->disk_bytenr);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2016-06-01 19:18:25 +08:00
|
|
|
if (em->start != offset || em->len != sectorsize) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err(
|
|
|
|
"unexpected extent wanted start %llu len %u, got start %llu len %llu",
|
2016-06-01 19:18:25 +08:00
|
|
|
offset, sectorsize, em->start, em->len);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
if (em->flags != 0) {
|
btrfs: use the flags of an extent map to identify the compression type
Currently, in struct extent_map, we use an unsigned int (32 bits) to
identify the compression type of an extent and an unsigned long (64 bits
on a 64 bits platform, 32 bits otherwise) for flags. We are only using
6 different flags, so an unsigned long is excessive and we can use flags
to identify the compression type instead of using a dedicated 32 bits
field.
We can easily have tens or hundreds of thousands (or more) of extent maps
on busy and large filesystems, specially with compression enabled or many
or large files with tons of small extents. So it's convenient to have the
extent_map structure as small as possible in order to use less memory.
So remove the compression type field from struct extent_map, use flags
to identify the compression type and shorten the flags field from an
unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and
reduces the size of the structure from 136 bytes down to 128 bytes, using
now only two cache lines, and increases the number of extent maps we can
have per 4K page from 30 to 32. By using a u32 for the flags instead of
an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(),
but that level of atomicity is not needed as most flags are never cleared
once set (before adding an extent map to the tree), and the ones that can
be cleared or set after an extent map is added to the tree, are always
performed while holding the write lock on the extent map tree, while the
reader holds a lock on the tree or tests for a flag that never changes
once the extent map is in the tree (such as compression flags).
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-05 00:20:33 +08:00
|
|
|
test_err("unexpected flags set, want 0 have %u", em->flags);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:04 +08:00
|
|
|
if (em->offset != 0) {
|
|
|
|
test_err("wrong offset, want 0, have %llu", em->offset);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
offset = em->start + em->len;
|
|
|
|
free_extent_map(em);
|
|
|
|
|
2024-01-08 11:20:20 +08:00
|
|
|
em = btrfs_get_extent(BTRFS_I(inode), NULL, offset, sectorsize);
|
2013-10-12 02:44:09 +08:00
|
|
|
if (IS_ERR(em)) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("got an error when we shouldn't have");
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:06 +08:00
|
|
|
if (extent_map_block_start(em) != disk_bytenr) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("block start does not match, want %llu got %llu",
|
2024-04-30 06:23:06 +08:00
|
|
|
disk_bytenr, extent_map_block_start(em));
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2016-06-01 19:18:25 +08:00
|
|
|
if (em->start != offset || em->len != 2 * sectorsize) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err(
|
|
|
|
"unexpected extent wanted start %llu len %u, got start %llu len %llu",
|
2016-06-01 19:18:25 +08:00
|
|
|
offset, 2 * sectorsize, em->start, em->len);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
if (em->flags != compressed_only) {
|
btrfs: use the flags of an extent map to identify the compression type
Currently, in struct extent_map, we use an unsigned int (32 bits) to
identify the compression type of an extent and an unsigned long (64 bits
on a 64 bits platform, 32 bits otherwise) for flags. We are only using
6 different flags, so an unsigned long is excessive and we can use flags
to identify the compression type instead of using a dedicated 32 bits
field.
We can easily have tens or hundreds of thousands (or more) of extent maps
on busy and large filesystems, specially with compression enabled or many
or large files with tons of small extents. So it's convenient to have the
extent_map structure as small as possible in order to use less memory.
So remove the compression type field from struct extent_map, use flags
to identify the compression type and shorten the flags field from an
unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and
reduces the size of the structure from 136 bytes down to 128 bytes, using
now only two cache lines, and increases the number of extent maps we can
have per 4K page from 30 to 32. By using a u32 for the flags instead of
an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(),
but that level of atomicity is not needed as most flags are never cleared
once set (before adding an extent map to the tree), and the ones that can
be cleared or set after an extent map is added to the tree, are always
performed while holding the write lock on the extent map tree, while the
reader holds a lock on the tree or tests for a flag that never changes
once the extent map is in the tree (such as compression flags).
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-05 00:20:33 +08:00
|
|
|
test_err("unexpected flags set, want %u have %u",
|
2013-10-12 02:44:09 +08:00
|
|
|
compressed_only, em->flags);
|
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:04 +08:00
|
|
|
if (em->start - em->offset != orig_start) {
|
|
|
|
test_err("wrong offset, em->start=%llu em->offset=%llu orig_start=%llu",
|
|
|
|
em->start, em->offset, orig_start);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
btrfs: use the flags of an extent map to identify the compression type
Currently, in struct extent_map, we use an unsigned int (32 bits) to
identify the compression type of an extent and an unsigned long (64 bits
on a 64 bits platform, 32 bits otherwise) for flags. We are only using
6 different flags, so an unsigned long is excessive and we can use flags
to identify the compression type instead of using a dedicated 32 bits
field.
We can easily have tens or hundreds of thousands (or more) of extent maps
on busy and large filesystems, specially with compression enabled or many
or large files with tons of small extents. So it's convenient to have the
extent_map structure as small as possible in order to use less memory.
So remove the compression type field from struct extent_map, use flags
to identify the compression type and shorten the flags field from an
unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and
reduces the size of the structure from 136 bytes down to 128 bytes, using
now only two cache lines, and increases the number of extent maps we can
have per 4K page from 30 to 32. By using a u32 for the flags instead of
an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(),
but that level of atomicity is not needed as most flags are never cleared
once set (before adding an extent map to the tree), and the ones that can
be cleared or set after an extent map is added to the tree, are always
performed while holding the write lock on the extent map tree, while the
reader holds a lock on the tree or tests for a flag that never changes
once the extent map is in the tree (such as compression flags).
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-05 00:20:33 +08:00
|
|
|
if (extent_map_compression(em) != BTRFS_COMPRESS_ZLIB) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("unexpected compress type, wanted %d, got %d",
|
btrfs: use the flags of an extent map to identify the compression type
Currently, in struct extent_map, we use an unsigned int (32 bits) to
identify the compression type of an extent and an unsigned long (64 bits
on a 64 bits platform, 32 bits otherwise) for flags. We are only using
6 different flags, so an unsigned long is excessive and we can use flags
to identify the compression type instead of using a dedicated 32 bits
field.
We can easily have tens or hundreds of thousands (or more) of extent maps
on busy and large filesystems, specially with compression enabled or many
or large files with tons of small extents. So it's convenient to have the
extent_map structure as small as possible in order to use less memory.
So remove the compression type field from struct extent_map, use flags
to identify the compression type and shorten the flags field from an
unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and
reduces the size of the structure from 136 bytes down to 128 bytes, using
now only two cache lines, and increases the number of extent maps we can
have per 4K page from 30 to 32. By using a u32 for the flags instead of
an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(),
but that level of atomicity is not needed as most flags are never cleared
once set (before adding an extent map to the tree), and the ones that can
be cleared or set after an extent map is added to the tree, are always
performed while holding the write lock on the extent map tree, while the
reader holds a lock on the tree or tests for a flag that never changes
once the extent map is in the tree (such as compression flags).
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-05 00:20:33 +08:00
|
|
|
BTRFS_COMPRESS_ZLIB, extent_map_compression(em));
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
offset = em->start + em->len;
|
|
|
|
free_extent_map(em);
|
|
|
|
|
|
|
|
/* A hole between regular extents but no hole extent */
|
2024-01-08 11:20:20 +08:00
|
|
|
em = btrfs_get_extent(BTRFS_I(inode), NULL, offset + 6, sectorsize);
|
2013-10-12 02:44:09 +08:00
|
|
|
if (IS_ERR(em)) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("got an error when we shouldn't have");
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:06 +08:00
|
|
|
if (em->disk_bytenr >= EXTENT_MAP_LAST_BYTE) {
|
|
|
|
test_err("expected a real extent, got %llu", em->disk_bytenr);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2016-06-01 19:18:25 +08:00
|
|
|
if (em->start != offset || em->len != sectorsize) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err(
|
|
|
|
"unexpected extent wanted start %llu len %u, got start %llu len %llu",
|
2016-06-01 19:18:25 +08:00
|
|
|
offset, sectorsize, em->start, em->len);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
if (em->flags != 0) {
|
btrfs: use the flags of an extent map to identify the compression type
Currently, in struct extent_map, we use an unsigned int (32 bits) to
identify the compression type of an extent and an unsigned long (64 bits
on a 64 bits platform, 32 bits otherwise) for flags. We are only using
6 different flags, so an unsigned long is excessive and we can use flags
to identify the compression type instead of using a dedicated 32 bits
field.
We can easily have tens or hundreds of thousands (or more) of extent maps
on busy and large filesystems, specially with compression enabled or many
or large files with tons of small extents. So it's convenient to have the
extent_map structure as small as possible in order to use less memory.
So remove the compression type field from struct extent_map, use flags
to identify the compression type and shorten the flags field from an
unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and
reduces the size of the structure from 136 bytes down to 128 bytes, using
now only two cache lines, and increases the number of extent maps we can
have per 4K page from 30 to 32. By using a u32 for the flags instead of
an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(),
but that level of atomicity is not needed as most flags are never cleared
once set (before adding an extent map to the tree), and the ones that can
be cleared or set after an extent map is added to the tree, are always
performed while holding the write lock on the extent map tree, while the
reader holds a lock on the tree or tests for a flag that never changes
once the extent map is in the tree (such as compression flags).
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-05 00:20:33 +08:00
|
|
|
test_err("unexpected flags set, want 0 have %u", em->flags);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:04 +08:00
|
|
|
if (em->offset != 0) {
|
|
|
|
test_err("wrong offset, want 0, have %llu", em->offset);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
offset = em->start + em->len;
|
|
|
|
free_extent_map(em);
|
|
|
|
|
2024-01-08 11:20:20 +08:00
|
|
|
em = btrfs_get_extent(BTRFS_I(inode), NULL, offset, SZ_4M);
|
2013-10-12 02:44:09 +08:00
|
|
|
if (IS_ERR(em)) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("got an error when we shouldn't have");
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:06 +08:00
|
|
|
if (em->disk_bytenr != EXTENT_MAP_HOLE) {
|
|
|
|
test_err("expected a hole extent, got %llu", em->disk_bytenr);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
/*
|
|
|
|
* Currently we just return a length that we requested rather than the
|
|
|
|
* length of the actual hole, if this changes we'll have to change this
|
|
|
|
* test.
|
|
|
|
*/
|
2016-06-01 19:18:25 +08:00
|
|
|
if (em->start != offset || em->len != 3 * sectorsize) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err(
|
|
|
|
"unexpected extent wanted start %llu len %u, got start %llu len %llu",
|
2016-06-01 19:18:25 +08:00
|
|
|
offset, 3 * sectorsize, em->start, em->len);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
if (em->flags != vacancy_only) {
|
btrfs: use the flags of an extent map to identify the compression type
Currently, in struct extent_map, we use an unsigned int (32 bits) to
identify the compression type of an extent and an unsigned long (64 bits
on a 64 bits platform, 32 bits otherwise) for flags. We are only using
6 different flags, so an unsigned long is excessive and we can use flags
to identify the compression type instead of using a dedicated 32 bits
field.
We can easily have tens or hundreds of thousands (or more) of extent maps
on busy and large filesystems, specially with compression enabled or many
or large files with tons of small extents. So it's convenient to have the
extent_map structure as small as possible in order to use less memory.
So remove the compression type field from struct extent_map, use flags
to identify the compression type and shorten the flags field from an
unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and
reduces the size of the structure from 136 bytes down to 128 bytes, using
now only two cache lines, and increases the number of extent maps we can
have per 4K page from 30 to 32. By using a u32 for the flags instead of
an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(),
but that level of atomicity is not needed as most flags are never cleared
once set (before adding an extent map to the tree), and the ones that can
be cleared or set after an extent map is added to the tree, are always
performed while holding the write lock on the extent map tree, while the
reader holds a lock on the tree or tests for a flag that never changes
once the extent map is in the tree (such as compression flags).
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-05 00:20:33 +08:00
|
|
|
test_err("unexpected flags set, want %u have %u",
|
2013-10-12 02:44:09 +08:00
|
|
|
vacancy_only, em->flags);
|
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:04 +08:00
|
|
|
if (em->offset != 0) {
|
|
|
|
test_err("wrong offset, want 0, have %llu", em->offset);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
offset = em->start + em->len;
|
|
|
|
free_extent_map(em);
|
|
|
|
|
2024-01-08 11:20:20 +08:00
|
|
|
em = btrfs_get_extent(BTRFS_I(inode), NULL, offset, sectorsize);
|
2013-10-12 02:44:09 +08:00
|
|
|
if (IS_ERR(em)) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("got an error when we shouldn't have");
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:06 +08:00
|
|
|
if (em->disk_bytenr >= EXTENT_MAP_LAST_BYTE) {
|
|
|
|
test_err("expected a real extent, got %llu", em->disk_bytenr);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2016-06-01 19:18:25 +08:00
|
|
|
if (em->start != offset || em->len != sectorsize) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err(
|
|
|
|
"unexpected extent wanted start %llu len %u, got start %llu len %llu",
|
2016-06-01 19:18:25 +08:00
|
|
|
offset, sectorsize, em->start, em->len);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
if (em->flags != 0) {
|
btrfs: use the flags of an extent map to identify the compression type
Currently, in struct extent_map, we use an unsigned int (32 bits) to
identify the compression type of an extent and an unsigned long (64 bits
on a 64 bits platform, 32 bits otherwise) for flags. We are only using
6 different flags, so an unsigned long is excessive and we can use flags
to identify the compression type instead of using a dedicated 32 bits
field.
We can easily have tens or hundreds of thousands (or more) of extent maps
on busy and large filesystems, specially with compression enabled or many
or large files with tons of small extents. So it's convenient to have the
extent_map structure as small as possible in order to use less memory.
So remove the compression type field from struct extent_map, use flags
to identify the compression type and shorten the flags field from an
unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and
reduces the size of the structure from 136 bytes down to 128 bytes, using
now only two cache lines, and increases the number of extent maps we can
have per 4K page from 30 to 32. By using a u32 for the flags instead of
an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(),
but that level of atomicity is not needed as most flags are never cleared
once set (before adding an extent map to the tree), and the ones that can
be cleared or set after an extent map is added to the tree, are always
performed while holding the write lock on the extent map tree, while the
reader holds a lock on the tree or tests for a flag that never changes
once the extent map is in the tree (such as compression flags).
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-05 00:20:33 +08:00
|
|
|
test_err("unexpected flags set, want 0 have %u", em->flags);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:04 +08:00
|
|
|
if (em->offset != 0) {
|
|
|
|
test_err("wrong orig offset, want 0, have %llu", em->offset);
|
2013-10-12 02:44:09 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
ret = 0;
|
|
|
|
out:
|
|
|
|
if (!IS_ERR(em))
|
|
|
|
free_extent_map(em);
|
|
|
|
iput(inode);
|
2014-05-08 05:06:09 +08:00
|
|
|
btrfs_free_dummy_root(root);
|
2016-06-21 02:14:09 +08:00
|
|
|
btrfs_free_dummy_fs_info(fs_info);
|
2013-10-12 02:44:09 +08:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2016-06-01 19:18:25 +08:00
|
|
|
static int test_hole_first(u32 sectorsize, u32 nodesize)
|
2013-10-15 01:15:02 +08:00
|
|
|
{
|
2016-06-21 02:14:09 +08:00
|
|
|
struct btrfs_fs_info *fs_info = NULL;
|
2013-10-15 01:15:02 +08:00
|
|
|
struct inode *inode = NULL;
|
|
|
|
struct btrfs_root *root = NULL;
|
|
|
|
struct extent_map *em = NULL;
|
|
|
|
int ret = -ENOMEM;
|
|
|
|
|
2019-03-18 21:06:55 +08:00
|
|
|
test_msg("running hole first btrfs_get_extent test");
|
|
|
|
|
2013-10-15 01:15:02 +08:00
|
|
|
inode = btrfs_new_test_inode();
|
|
|
|
if (!inode) {
|
2019-03-16 00:28:46 +08:00
|
|
|
test_std_err(TEST_ALLOC_INODE);
|
2013-10-15 01:15:02 +08:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2016-06-15 21:22:56 +08:00
|
|
|
fs_info = btrfs_alloc_dummy_fs_info(nodesize, sectorsize);
|
2016-06-21 02:14:09 +08:00
|
|
|
if (!fs_info) {
|
2019-03-16 00:28:46 +08:00
|
|
|
test_std_err(TEST_ALLOC_FS_INFO);
|
2013-10-15 01:15:02 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2016-06-15 21:22:56 +08:00
|
|
|
root = btrfs_alloc_dummy_root(fs_info);
|
2016-06-21 02:14:09 +08:00
|
|
|
if (IS_ERR(root)) {
|
2019-03-16 00:28:46 +08:00
|
|
|
test_std_err(TEST_ALLOC_ROOT);
|
2013-10-15 01:15:02 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2016-06-15 21:22:56 +08:00
|
|
|
root->node = alloc_dummy_extent_buffer(fs_info, nodesize);
|
2013-10-15 01:15:02 +08:00
|
|
|
if (!root->node) {
|
2019-03-16 00:28:46 +08:00
|
|
|
test_std_err(TEST_ALLOC_ROOT);
|
2013-10-15 01:15:02 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
btrfs_set_header_nritems(root->node, 0);
|
|
|
|
btrfs_set_header_level(root->node, 0);
|
|
|
|
BTRFS_I(inode)->root = root;
|
|
|
|
ret = -EINVAL;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Need a blank inode item here just so we don't confuse
|
|
|
|
* btrfs_get_extent.
|
|
|
|
*/
|
|
|
|
insert_inode_item_key(root);
|
2016-06-01 19:18:25 +08:00
|
|
|
insert_extent(root, sectorsize, sectorsize, sectorsize, 0, sectorsize,
|
|
|
|
sectorsize, BTRFS_FILE_EXTENT_REG, 0, 1);
|
2024-01-08 11:20:20 +08:00
|
|
|
em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, 2 * sectorsize);
|
2013-10-15 01:15:02 +08:00
|
|
|
if (IS_ERR(em)) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("got an error when we shouldn't have");
|
2013-10-15 01:15:02 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:06 +08:00
|
|
|
if (em->disk_bytenr != EXTENT_MAP_HOLE) {
|
|
|
|
test_err("expected a hole, got %llu", em->disk_bytenr);
|
2013-10-15 01:15:02 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2016-06-01 19:18:25 +08:00
|
|
|
if (em->start != 0 || em->len != sectorsize) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err(
|
|
|
|
"unexpected extent wanted start 0 len %u, got start %llu len %llu",
|
2016-06-01 19:18:25 +08:00
|
|
|
sectorsize, em->start, em->len);
|
2013-10-15 01:15:02 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
if (em->flags != vacancy_only) {
|
btrfs: use the flags of an extent map to identify the compression type
Currently, in struct extent_map, we use an unsigned int (32 bits) to
identify the compression type of an extent and an unsigned long (64 bits
on a 64 bits platform, 32 bits otherwise) for flags. We are only using
6 different flags, so an unsigned long is excessive and we can use flags
to identify the compression type instead of using a dedicated 32 bits
field.
We can easily have tens or hundreds of thousands (or more) of extent maps
on busy and large filesystems, specially with compression enabled or many
or large files with tons of small extents. So it's convenient to have the
extent_map structure as small as possible in order to use less memory.
So remove the compression type field from struct extent_map, use flags
to identify the compression type and shorten the flags field from an
unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and
reduces the size of the structure from 136 bytes down to 128 bytes, using
now only two cache lines, and increases the number of extent maps we can
have per 4K page from 30 to 32. By using a u32 for the flags instead of
an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(),
but that level of atomicity is not needed as most flags are never cleared
once set (before adding an extent map to the tree), and the ones that can
be cleared or set after an extent map is added to the tree, are always
performed while holding the write lock on the extent map tree, while the
reader holds a lock on the tree or tests for a flag that never changes
once the extent map is in the tree (such as compression flags).
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-05 00:20:33 +08:00
|
|
|
test_err("wrong flags, wanted %u, have %u", vacancy_only,
|
2013-10-15 01:15:02 +08:00
|
|
|
em->flags);
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
free_extent_map(em);
|
|
|
|
|
2024-01-08 11:20:20 +08:00
|
|
|
em = btrfs_get_extent(BTRFS_I(inode), NULL, sectorsize, 2 * sectorsize);
|
2013-10-15 01:15:02 +08:00
|
|
|
if (IS_ERR(em)) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("got an error when we shouldn't have");
|
2013-10-15 01:15:02 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2024-04-30 06:23:06 +08:00
|
|
|
if (extent_map_block_start(em) != sectorsize) {
|
|
|
|
test_err("expected a real extent, got %llu", extent_map_block_start(em));
|
2013-10-15 01:15:02 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2016-06-01 19:18:25 +08:00
|
|
|
if (em->start != sectorsize || em->len != sectorsize) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err(
|
|
|
|
"unexpected extent wanted start %u len %u, got start %llu len %llu",
|
2016-06-01 19:18:25 +08:00
|
|
|
sectorsize, sectorsize, em->start, em->len);
|
2013-10-15 01:15:02 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
if (em->flags != 0) {
|
btrfs: use the flags of an extent map to identify the compression type
Currently, in struct extent_map, we use an unsigned int (32 bits) to
identify the compression type of an extent and an unsigned long (64 bits
on a 64 bits platform, 32 bits otherwise) for flags. We are only using
6 different flags, so an unsigned long is excessive and we can use flags
to identify the compression type instead of using a dedicated 32 bits
field.
We can easily have tens or hundreds of thousands (or more) of extent maps
on busy and large filesystems, specially with compression enabled or many
or large files with tons of small extents. So it's convenient to have the
extent_map structure as small as possible in order to use less memory.
So remove the compression type field from struct extent_map, use flags
to identify the compression type and shorten the flags field from an
unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and
reduces the size of the structure from 136 bytes down to 128 bytes, using
now only two cache lines, and increases the number of extent maps we can
have per 4K page from 30 to 32. By using a u32 for the flags instead of
an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(),
but that level of atomicity is not needed as most flags are never cleared
once set (before adding an extent map to the tree), and the ones that can
be cleared or set after an extent map is added to the tree, are always
performed while holding the write lock on the extent map tree, while the
reader holds a lock on the tree or tests for a flag that never changes
once the extent map is in the tree (such as compression flags).
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-05 00:20:33 +08:00
|
|
|
test_err("unexpected flags set, wanted 0 got %u",
|
2013-10-15 01:15:02 +08:00
|
|
|
em->flags);
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
ret = 0;
|
|
|
|
out:
|
|
|
|
if (!IS_ERR(em))
|
|
|
|
free_extent_map(em);
|
|
|
|
iput(inode);
|
2014-05-08 05:06:09 +08:00
|
|
|
btrfs_free_dummy_root(root);
|
2016-06-21 02:14:09 +08:00
|
|
|
btrfs_free_dummy_fs_info(fs_info);
|
2013-10-15 01:15:02 +08:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2016-06-01 19:18:25 +08:00
|
|
|
static int test_extent_accounting(u32 sectorsize, u32 nodesize)
|
2015-03-17 05:38:52 +08:00
|
|
|
{
|
2016-06-21 02:14:09 +08:00
|
|
|
struct btrfs_fs_info *fs_info = NULL;
|
2015-03-17 05:38:52 +08:00
|
|
|
struct inode *inode = NULL;
|
|
|
|
struct btrfs_root *root = NULL;
|
|
|
|
int ret = -ENOMEM;
|
|
|
|
|
2019-03-18 21:06:55 +08:00
|
|
|
test_msg("running outstanding_extents tests");
|
|
|
|
|
2015-03-17 05:38:52 +08:00
|
|
|
inode = btrfs_new_test_inode();
|
|
|
|
if (!inode) {
|
2019-03-16 00:28:46 +08:00
|
|
|
test_std_err(TEST_ALLOC_INODE);
|
2015-03-17 05:38:52 +08:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2016-06-15 21:22:56 +08:00
|
|
|
fs_info = btrfs_alloc_dummy_fs_info(nodesize, sectorsize);
|
2016-06-21 02:14:09 +08:00
|
|
|
if (!fs_info) {
|
2019-03-16 00:28:46 +08:00
|
|
|
test_std_err(TEST_ALLOC_FS_INFO);
|
2015-03-17 05:38:52 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2016-06-15 21:22:56 +08:00
|
|
|
root = btrfs_alloc_dummy_root(fs_info);
|
2016-06-21 02:14:09 +08:00
|
|
|
if (IS_ERR(root)) {
|
2019-03-16 00:28:46 +08:00
|
|
|
test_std_err(TEST_ALLOC_ROOT);
|
2015-03-17 05:38:52 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
BTRFS_I(inode)->root = root;
|
|
|
|
|
|
|
|
/* [BTRFS_MAX_EXTENT_SIZE] */
|
2020-06-03 13:55:35 +08:00
|
|
|
ret = btrfs_set_extent_delalloc(BTRFS_I(inode), 0,
|
|
|
|
BTRFS_MAX_EXTENT_SIZE - 1, 0, NULL);
|
2015-03-17 05:38:52 +08:00
|
|
|
if (ret) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("btrfs_set_extent_delalloc returned %d", ret);
|
2015-03-17 05:38:52 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
if (BTRFS_I(inode)->outstanding_extents != 1) {
|
|
|
|
ret = -EINVAL;
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("miscount, wanted 1, got %u",
|
2015-03-17 05:38:52 +08:00
|
|
|
BTRFS_I(inode)->outstanding_extents);
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2016-06-01 19:18:25 +08:00
|
|
|
/* [BTRFS_MAX_EXTENT_SIZE][sectorsize] */
|
2020-06-03 13:55:35 +08:00
|
|
|
ret = btrfs_set_extent_delalloc(BTRFS_I(inode), BTRFS_MAX_EXTENT_SIZE,
|
2016-06-01 19:18:25 +08:00
|
|
|
BTRFS_MAX_EXTENT_SIZE + sectorsize - 1,
|
2019-07-17 21:18:17 +08:00
|
|
|
0, NULL);
|
2015-03-17 05:38:52 +08:00
|
|
|
if (ret) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("btrfs_set_extent_delalloc returned %d", ret);
|
2015-03-17 05:38:52 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
if (BTRFS_I(inode)->outstanding_extents != 2) {
|
|
|
|
ret = -EINVAL;
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("miscount, wanted 2, got %u",
|
2015-03-17 05:38:52 +08:00
|
|
|
BTRFS_I(inode)->outstanding_extents);
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2016-06-01 19:18:25 +08:00
|
|
|
/* [BTRFS_MAX_EXTENT_SIZE/2][sectorsize HOLE][the rest] */
|
2015-03-17 05:38:52 +08:00
|
|
|
ret = clear_extent_bit(&BTRFS_I(inode)->io_tree,
|
|
|
|
BTRFS_MAX_EXTENT_SIZE >> 1,
|
2016-06-01 19:18:25 +08:00
|
|
|
(BTRFS_MAX_EXTENT_SIZE >> 1) + sectorsize - 1,
|
btrfs: fix missing delalloc new bit for new delalloc ranges
When doing a buffered write, through one of the write family syscalls, we
look for ranges which currently don't have allocated extents and set the
'delalloc new' bit on them, so that we can report a correct number of used
blocks to the stat(2) syscall until delalloc is flushed and ordered extents
complete.
However there are a few other places where we can do a buffered write
against a range that is mapped to a hole (no extent allocated) and where
we do not set the 'new delalloc' bit. Those places are:
- Doing a memory mapped write against a hole;
- Cloning an inline extent into a hole starting at file offset 0;
- Calling btrfs_cont_expand() when the i_size of the file is not aligned
to the sector size and is located in a hole. For example when cloning
to a destination offset beyond EOF.
So after such cases, until the corresponding delalloc range is flushed and
the respective ordered extents complete, we can report an incorrect number
of blocks used through the stat(2) syscall.
In some cases we can end up reporting 0 used blocks to stat(2), which is a
particular bad value to report as it may mislead tools to think a file is
completely sparse when its i_size is not zero, making them skip reading
any data, an undesired consequence for tools such as archivers and other
backup tools, as reported a long time ago in the following thread (and
other past threads):
https://lists.gnu.org/archive/html/bug-tar/2016-07/msg00001.html
Example reproducer:
$ cat reproducer.sh
#!/bin/bash
MNT=/mnt/sdi
DEV=/dev/sdi
mkfs.btrfs -f $DEV > /dev/null
# mkfs.xfs -f $DEV > /dev/null
# mkfs.ext4 -F $DEV > /dev/null
# mkfs.f2fs -f $DEV > /dev/null
mount $DEV $MNT
xfs_io -f -c "truncate 64K" \
-c "mmap -w 0 64K" \
-c "mwrite -S 0xab 0 64K" \
-c "munmap" \
$MNT/foo
blocks_used=$(stat -c %b $MNT/foo)
echo "blocks used: $blocks_used"
if [ $blocks_used -eq 0 ]; then
echo "ERROR: blocks used is 0"
fi
umount $DEV
$ ./reproducer.sh
blocks used: 0
ERROR: blocks used is 0
So move the logic that decides to set the 'delalloc bit' bit into the
function btrfs_set_extent_delalloc(), since that is what we use for all
those missing cases as well as for the cases that currently work well.
This change is also preparatory work for an upcoming patch that fixes
other problems related to tracking and reporting the number of bytes used
by an inode.
CC: stable@vger.kernel.org # 4.19+
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2020-11-04 19:07:31 +08:00
|
|
|
EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
|
2022-09-10 05:53:47 +08:00
|
|
|
EXTENT_UPTODATE, NULL);
|
2015-03-17 05:38:52 +08:00
|
|
|
if (ret) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("clear_extent_bit returned %d", ret);
|
2015-03-17 05:38:52 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
if (BTRFS_I(inode)->outstanding_extents != 2) {
|
|
|
|
ret = -EINVAL;
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("miscount, wanted 2, got %u",
|
2015-03-17 05:38:52 +08:00
|
|
|
BTRFS_I(inode)->outstanding_extents);
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2016-06-01 19:18:25 +08:00
|
|
|
/* [BTRFS_MAX_EXTENT_SIZE][sectorsize] */
|
2020-06-03 13:55:35 +08:00
|
|
|
ret = btrfs_set_extent_delalloc(BTRFS_I(inode), BTRFS_MAX_EXTENT_SIZE >> 1,
|
2016-06-01 19:18:25 +08:00
|
|
|
(BTRFS_MAX_EXTENT_SIZE >> 1)
|
|
|
|
+ sectorsize - 1,
|
2019-07-17 21:18:17 +08:00
|
|
|
0, NULL);
|
2015-03-17 05:38:52 +08:00
|
|
|
if (ret) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("btrfs_set_extent_delalloc returned %d", ret);
|
2015-03-17 05:38:52 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
if (BTRFS_I(inode)->outstanding_extents != 2) {
|
|
|
|
ret = -EINVAL;
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("miscount, wanted 2, got %u",
|
2015-03-17 05:38:52 +08:00
|
|
|
BTRFS_I(inode)->outstanding_extents);
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
2016-06-01 19:18:25 +08:00
|
|
|
* [BTRFS_MAX_EXTENT_SIZE+sectorsize][sectorsize HOLE][BTRFS_MAX_EXTENT_SIZE+sectorsize]
|
2015-03-17 05:38:52 +08:00
|
|
|
*/
|
2020-06-03 13:55:35 +08:00
|
|
|
ret = btrfs_set_extent_delalloc(BTRFS_I(inode),
|
2016-06-01 19:18:25 +08:00
|
|
|
BTRFS_MAX_EXTENT_SIZE + 2 * sectorsize,
|
|
|
|
(BTRFS_MAX_EXTENT_SIZE << 1) + 3 * sectorsize - 1,
|
2019-07-17 21:18:17 +08:00
|
|
|
0, NULL);
|
2015-03-17 05:38:52 +08:00
|
|
|
if (ret) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("btrfs_set_extent_delalloc returned %d", ret);
|
2015-03-17 05:38:52 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
if (BTRFS_I(inode)->outstanding_extents != 4) {
|
|
|
|
ret = -EINVAL;
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("miscount, wanted 4, got %u",
|
2015-03-17 05:38:52 +08:00
|
|
|
BTRFS_I(inode)->outstanding_extents);
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2016-06-01 19:18:25 +08:00
|
|
|
/*
|
|
|
|
* [BTRFS_MAX_EXTENT_SIZE+sectorsize][sectorsize][BTRFS_MAX_EXTENT_SIZE+sectorsize]
|
|
|
|
*/
|
2020-06-03 13:55:35 +08:00
|
|
|
ret = btrfs_set_extent_delalloc(BTRFS_I(inode),
|
2016-06-01 19:18:25 +08:00
|
|
|
BTRFS_MAX_EXTENT_SIZE + sectorsize,
|
2019-07-17 21:18:17 +08:00
|
|
|
BTRFS_MAX_EXTENT_SIZE + 2 * sectorsize - 1, 0, NULL);
|
2015-03-17 05:38:52 +08:00
|
|
|
if (ret) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("btrfs_set_extent_delalloc returned %d", ret);
|
2015-03-17 05:38:52 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
if (BTRFS_I(inode)->outstanding_extents != 3) {
|
|
|
|
ret = -EINVAL;
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("miscount, wanted 3, got %u",
|
2015-03-17 05:38:52 +08:00
|
|
|
BTRFS_I(inode)->outstanding_extents);
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* [BTRFS_MAX_EXTENT_SIZE+4k][4K HOLE][BTRFS_MAX_EXTENT_SIZE+4k] */
|
|
|
|
ret = clear_extent_bit(&BTRFS_I(inode)->io_tree,
|
2016-06-01 19:18:25 +08:00
|
|
|
BTRFS_MAX_EXTENT_SIZE + sectorsize,
|
|
|
|
BTRFS_MAX_EXTENT_SIZE + 2 * sectorsize - 1,
|
btrfs: fix missing delalloc new bit for new delalloc ranges
When doing a buffered write, through one of the write family syscalls, we
look for ranges which currently don't have allocated extents and set the
'delalloc new' bit on them, so that we can report a correct number of used
blocks to the stat(2) syscall until delalloc is flushed and ordered extents
complete.
However there are a few other places where we can do a buffered write
against a range that is mapped to a hole (no extent allocated) and where
we do not set the 'new delalloc' bit. Those places are:
- Doing a memory mapped write against a hole;
- Cloning an inline extent into a hole starting at file offset 0;
- Calling btrfs_cont_expand() when the i_size of the file is not aligned
to the sector size and is located in a hole. For example when cloning
to a destination offset beyond EOF.
So after such cases, until the corresponding delalloc range is flushed and
the respective ordered extents complete, we can report an incorrect number
of blocks used through the stat(2) syscall.
In some cases we can end up reporting 0 used blocks to stat(2), which is a
particular bad value to report as it may mislead tools to think a file is
completely sparse when its i_size is not zero, making them skip reading
any data, an undesired consequence for tools such as archivers and other
backup tools, as reported a long time ago in the following thread (and
other past threads):
https://lists.gnu.org/archive/html/bug-tar/2016-07/msg00001.html
Example reproducer:
$ cat reproducer.sh
#!/bin/bash
MNT=/mnt/sdi
DEV=/dev/sdi
mkfs.btrfs -f $DEV > /dev/null
# mkfs.xfs -f $DEV > /dev/null
# mkfs.ext4 -F $DEV > /dev/null
# mkfs.f2fs -f $DEV > /dev/null
mount $DEV $MNT
xfs_io -f -c "truncate 64K" \
-c "mmap -w 0 64K" \
-c "mwrite -S 0xab 0 64K" \
-c "munmap" \
$MNT/foo
blocks_used=$(stat -c %b $MNT/foo)
echo "blocks used: $blocks_used"
if [ $blocks_used -eq 0 ]; then
echo "ERROR: blocks used is 0"
fi
umount $DEV
$ ./reproducer.sh
blocks used: 0
ERROR: blocks used is 0
So move the logic that decides to set the 'delalloc bit' bit into the
function btrfs_set_extent_delalloc(), since that is what we use for all
those missing cases as well as for the cases that currently work well.
This change is also preparatory work for an upcoming patch that fixes
other problems related to tracking and reporting the number of bytes used
by an inode.
CC: stable@vger.kernel.org # 4.19+
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2020-11-04 19:07:31 +08:00
|
|
|
EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
|
2022-09-10 05:53:47 +08:00
|
|
|
EXTENT_UPTODATE, NULL);
|
2015-03-17 05:38:52 +08:00
|
|
|
if (ret) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("clear_extent_bit returned %d", ret);
|
2015-03-17 05:38:52 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
if (BTRFS_I(inode)->outstanding_extents != 4) {
|
|
|
|
ret = -EINVAL;
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("miscount, wanted 4, got %u",
|
2015-03-17 05:38:52 +08:00
|
|
|
BTRFS_I(inode)->outstanding_extents);
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Refill the hole again just for good measure, because I thought it
|
|
|
|
* might fail and I'd rather satisfy my paranoia at this point.
|
|
|
|
*/
|
2020-06-03 13:55:35 +08:00
|
|
|
ret = btrfs_set_extent_delalloc(BTRFS_I(inode),
|
2016-06-01 19:18:25 +08:00
|
|
|
BTRFS_MAX_EXTENT_SIZE + sectorsize,
|
2019-07-17 21:18:17 +08:00
|
|
|
BTRFS_MAX_EXTENT_SIZE + 2 * sectorsize - 1, 0, NULL);
|
2015-03-17 05:38:52 +08:00
|
|
|
if (ret) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("btrfs_set_extent_delalloc returned %d", ret);
|
2015-03-17 05:38:52 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
if (BTRFS_I(inode)->outstanding_extents != 3) {
|
|
|
|
ret = -EINVAL;
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("miscount, wanted 3, got %u",
|
2015-03-17 05:38:52 +08:00
|
|
|
BTRFS_I(inode)->outstanding_extents);
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Empty */
|
|
|
|
ret = clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
|
btrfs: fix missing delalloc new bit for new delalloc ranges
When doing a buffered write, through one of the write family syscalls, we
look for ranges which currently don't have allocated extents and set the
'delalloc new' bit on them, so that we can report a correct number of used
blocks to the stat(2) syscall until delalloc is flushed and ordered extents
complete.
However there are a few other places where we can do a buffered write
against a range that is mapped to a hole (no extent allocated) and where
we do not set the 'new delalloc' bit. Those places are:
- Doing a memory mapped write against a hole;
- Cloning an inline extent into a hole starting at file offset 0;
- Calling btrfs_cont_expand() when the i_size of the file is not aligned
to the sector size and is located in a hole. For example when cloning
to a destination offset beyond EOF.
So after such cases, until the corresponding delalloc range is flushed and
the respective ordered extents complete, we can report an incorrect number
of blocks used through the stat(2) syscall.
In some cases we can end up reporting 0 used blocks to stat(2), which is a
particular bad value to report as it may mislead tools to think a file is
completely sparse when its i_size is not zero, making them skip reading
any data, an undesired consequence for tools such as archivers and other
backup tools, as reported a long time ago in the following thread (and
other past threads):
https://lists.gnu.org/archive/html/bug-tar/2016-07/msg00001.html
Example reproducer:
$ cat reproducer.sh
#!/bin/bash
MNT=/mnt/sdi
DEV=/dev/sdi
mkfs.btrfs -f $DEV > /dev/null
# mkfs.xfs -f $DEV > /dev/null
# mkfs.ext4 -F $DEV > /dev/null
# mkfs.f2fs -f $DEV > /dev/null
mount $DEV $MNT
xfs_io -f -c "truncate 64K" \
-c "mmap -w 0 64K" \
-c "mwrite -S 0xab 0 64K" \
-c "munmap" \
$MNT/foo
blocks_used=$(stat -c %b $MNT/foo)
echo "blocks used: $blocks_used"
if [ $blocks_used -eq 0 ]; then
echo "ERROR: blocks used is 0"
fi
umount $DEV
$ ./reproducer.sh
blocks used: 0
ERROR: blocks used is 0
So move the logic that decides to set the 'delalloc bit' bit into the
function btrfs_set_extent_delalloc(), since that is what we use for all
those missing cases as well as for the cases that currently work well.
This change is also preparatory work for an upcoming patch that fixes
other problems related to tracking and reporting the number of bytes used
by an inode.
CC: stable@vger.kernel.org # 4.19+
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2020-11-04 19:07:31 +08:00
|
|
|
EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
|
2022-09-10 05:53:47 +08:00
|
|
|
EXTENT_UPTODATE, NULL);
|
2015-03-17 05:38:52 +08:00
|
|
|
if (ret) {
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("clear_extent_bit returned %d", ret);
|
2015-03-17 05:38:52 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
if (BTRFS_I(inode)->outstanding_extents) {
|
|
|
|
ret = -EINVAL;
|
2018-05-17 06:00:42 +08:00
|
|
|
test_err("miscount, wanted 0, got %u",
|
2015-03-17 05:38:52 +08:00
|
|
|
BTRFS_I(inode)->outstanding_extents);
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
ret = 0;
|
|
|
|
out:
|
|
|
|
if (ret)
|
|
|
|
clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
|
btrfs: fix missing delalloc new bit for new delalloc ranges
When doing a buffered write, through one of the write family syscalls, we
look for ranges which currently don't have allocated extents and set the
'delalloc new' bit on them, so that we can report a correct number of used
blocks to the stat(2) syscall until delalloc is flushed and ordered extents
complete.
However there are a few other places where we can do a buffered write
against a range that is mapped to a hole (no extent allocated) and where
we do not set the 'new delalloc' bit. Those places are:
- Doing a memory mapped write against a hole;
- Cloning an inline extent into a hole starting at file offset 0;
- Calling btrfs_cont_expand() when the i_size of the file is not aligned
to the sector size and is located in a hole. For example when cloning
to a destination offset beyond EOF.
So after such cases, until the corresponding delalloc range is flushed and
the respective ordered extents complete, we can report an incorrect number
of blocks used through the stat(2) syscall.
In some cases we can end up reporting 0 used blocks to stat(2), which is a
particular bad value to report as it may mislead tools to think a file is
completely sparse when its i_size is not zero, making them skip reading
any data, an undesired consequence for tools such as archivers and other
backup tools, as reported a long time ago in the following thread (and
other past threads):
https://lists.gnu.org/archive/html/bug-tar/2016-07/msg00001.html
Example reproducer:
$ cat reproducer.sh
#!/bin/bash
MNT=/mnt/sdi
DEV=/dev/sdi
mkfs.btrfs -f $DEV > /dev/null
# mkfs.xfs -f $DEV > /dev/null
# mkfs.ext4 -F $DEV > /dev/null
# mkfs.f2fs -f $DEV > /dev/null
mount $DEV $MNT
xfs_io -f -c "truncate 64K" \
-c "mmap -w 0 64K" \
-c "mwrite -S 0xab 0 64K" \
-c "munmap" \
$MNT/foo
blocks_used=$(stat -c %b $MNT/foo)
echo "blocks used: $blocks_used"
if [ $blocks_used -eq 0 ]; then
echo "ERROR: blocks used is 0"
fi
umount $DEV
$ ./reproducer.sh
blocks used: 0
ERROR: blocks used is 0
So move the logic that decides to set the 'delalloc bit' bit into the
function btrfs_set_extent_delalloc(), since that is what we use for all
those missing cases as well as for the cases that currently work well.
This change is also preparatory work for an upcoming patch that fixes
other problems related to tracking and reporting the number of bytes used
by an inode.
CC: stable@vger.kernel.org # 4.19+
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2020-11-04 19:07:31 +08:00
|
|
|
EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
|
2022-09-10 05:53:47 +08:00
|
|
|
EXTENT_UPTODATE, NULL);
|
2015-03-17 05:38:52 +08:00
|
|
|
iput(inode);
|
|
|
|
btrfs_free_dummy_root(root);
|
2016-06-21 02:14:09 +08:00
|
|
|
btrfs_free_dummy_fs_info(fs_info);
|
2015-03-17 05:38:52 +08:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2016-06-01 19:18:25 +08:00
|
|
|
int btrfs_test_inodes(u32 sectorsize, u32 nodesize)
|
2013-10-12 02:44:09 +08:00
|
|
|
{
|
2013-10-15 01:15:02 +08:00
|
|
|
int ret;
|
|
|
|
|
2019-03-18 21:06:55 +08:00
|
|
|
test_msg("running inode tests");
|
|
|
|
|
btrfs: use the flags of an extent map to identify the compression type
Currently, in struct extent_map, we use an unsigned int (32 bits) to
identify the compression type of an extent and an unsigned long (64 bits
on a 64 bits platform, 32 bits otherwise) for flags. We are only using
6 different flags, so an unsigned long is excessive and we can use flags
to identify the compression type instead of using a dedicated 32 bits
field.
We can easily have tens or hundreds of thousands (or more) of extent maps
on busy and large filesystems, specially with compression enabled or many
or large files with tons of small extents. So it's convenient to have the
extent_map structure as small as possible in order to use less memory.
So remove the compression type field from struct extent_map, use flags
to identify the compression type and shorten the flags field from an
unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and
reduces the size of the structure from 136 bytes down to 128 bytes, using
now only two cache lines, and increases the number of extent maps we can
have per 4K page from 30 to 32. By using a u32 for the flags instead of
an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(),
but that level of atomicity is not needed as most flags are never cleared
once set (before adding an extent map to the tree), and the ones that can
be cleared or set after an extent map is added to the tree, are always
performed while holding the write lock on the extent map tree, while the
reader holds a lock on the tree or tests for a flag that never changes
once the extent map is in the tree (such as compression flags).
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2023-12-05 00:20:33 +08:00
|
|
|
compressed_only |= EXTENT_FLAG_COMPRESS_ZLIB;
|
|
|
|
prealloc_only |= EXTENT_FLAG_PREALLOC;
|
2013-10-15 01:15:02 +08:00
|
|
|
|
2016-06-01 19:18:25 +08:00
|
|
|
ret = test_btrfs_get_extent(sectorsize, nodesize);
|
2013-10-15 01:15:02 +08:00
|
|
|
if (ret)
|
|
|
|
return ret;
|
2016-06-01 19:18:25 +08:00
|
|
|
ret = test_hole_first(sectorsize, nodesize);
|
2015-03-17 05:38:52 +08:00
|
|
|
if (ret)
|
|
|
|
return ret;
|
2016-06-01 19:18:25 +08:00
|
|
|
return test_extent_accounting(sectorsize, nodesize);
|
2013-10-12 02:44:09 +08:00
|
|
|
}
|