2007-06-12 21:07:21 +08:00
|
|
|
/*
|
|
|
|
* Copyright (C) 2007 Oracle. All rights reserved.
|
|
|
|
*
|
|
|
|
* This program is free software; you can redistribute it and/or
|
|
|
|
* modify it under the terms of the GNU General Public
|
|
|
|
* License v2 as published by the Free Software Foundation.
|
|
|
|
*
|
|
|
|
* This program is distributed in the hope that it will be useful,
|
|
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
|
* General Public License for more details.
|
|
|
|
*
|
|
|
|
* You should have received a copy of the GNU General Public
|
|
|
|
* License along with this program; if not, write to the
|
|
|
|
* Free Software Foundation, Inc., 59 Temple Place - Suite 330,
|
|
|
|
* Boston, MA 021110-1307, USA.
|
|
|
|
*/
|
|
|
|
|
2008-02-21 01:07:25 +08:00
|
|
|
#include <linux/bio.h>
|
include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h
percpu.h is included by sched.h and module.h and thus ends up being
included when building most .c files. percpu.h includes slab.h which
in turn includes gfp.h making everything defined by the two files
universally available and complicating inclusion dependencies.
percpu.h -> slab.h dependency is about to be removed. Prepare for
this change by updating users of gfp and slab facilities include those
headers directly instead of assuming availability. As this conversion
needs to touch large number of source files, the following script is
used as the basis of conversion.
http://userweb.kernel.org/~tj/misc/slabh-sweep.py
The script does the followings.
* Scan files for gfp and slab usages and update includes such that
only the necessary includes are there. ie. if only gfp is used,
gfp.h, if slab is used, slab.h.
* When the script inserts a new include, it looks at the include
blocks and try to put the new include such that its order conforms
to its surrounding. It's put in the include block which contains
core kernel includes, in the same order that the rest are ordered -
alphabetical, Christmas tree, rev-Xmas-tree or at the end if there
doesn't seem to be any matching order.
* If the script can't find a place to put a new include (mostly
because the file doesn't have fitting include block), it prints out
an error message indicating which .h file needs to be added to the
file.
The conversion was done in the following steps.
1. The initial automatic conversion of all .c files updated slightly
over 4000 files, deleting around 700 includes and adding ~480 gfp.h
and ~3000 slab.h inclusions. The script emitted errors for ~400
files.
2. Each error was manually checked. Some didn't need the inclusion,
some needed manual addition while adding it to implementation .h or
embedding .c file was more appropriate for others. This step added
inclusions to around 150 files.
3. The script was run again and the output was compared to the edits
from #2 to make sure no file was left behind.
4. Several build tests were done and a couple of problems were fixed.
e.g. lib/decompress_*.c used malloc/free() wrappers around slab
APIs requiring slab.h to be added manually.
5. The script was run on all .h files but without automatically
editing them as sprinkling gfp.h and slab.h inclusions around .h
files could easily lead to inclusion dependency hell. Most gfp.h
inclusion directives were ignored as stuff from gfp.h was usually
wildly available and often used in preprocessor macros. Each
slab.h inclusion directive was examined and added manually as
necessary.
6. percpu.h was updated not to include slab.h.
7. Build test were done on the following configurations and failures
were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my
distributed build env didn't work with gcov compiles) and a few
more options had to be turned off depending on archs to make things
build (like ipr on powerpc/64 which failed due to missing writeq).
* x86 and x86_64 UP and SMP allmodconfig and a custom test config.
* powerpc and powerpc64 SMP allmodconfig
* sparc and sparc64 SMP allmodconfig
* ia64 SMP allmodconfig
* s390 SMP allmodconfig
* alpha SMP allmodconfig
* um on x86_64 SMP allmodconfig
8. percpu.h modifications were reverted so that it could be applied as
a separate patch and serve as bisection point.
Given the fact that I had only a couple of failures from tests on step
6, I'm fairly confident about the coverage of this conversion patch.
If there is a breakage, it's likely to be something in one of the arch
headers which should be easily discoverable easily on most builds of
the specific arch.
Signed-off-by: Tejun Heo <tj@kernel.org>
Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
2010-03-24 16:04:11 +08:00
|
|
|
#include <linux/slab.h>
|
2008-02-21 01:07:25 +08:00
|
|
|
#include <linux/pagemap.h>
|
|
|
|
#include <linux/highmem.h>
|
2007-03-16 07:03:33 +08:00
|
|
|
#include "ctree.h"
|
2007-03-27 04:00:06 +08:00
|
|
|
#include "disk-io.h"
|
2007-03-21 02:38:32 +08:00
|
|
|
#include "transaction.h"
|
2013-07-25 19:22:34 +08:00
|
|
|
#include "volumes.h"
|
2007-05-30 03:17:08 +08:00
|
|
|
#include "print-tree.h"
|
2007-03-16 07:03:33 +08:00
|
|
|
|
2012-08-13 16:52:38 +08:00
|
|
|
#define __MAX_CSUM_ITEMS(r, size) ((unsigned long)(((BTRFS_LEAF_DATA_SIZE(r) - \
|
2008-12-02 20:17:45 +08:00
|
|
|
sizeof(struct btrfs_item) * 2) / \
|
|
|
|
size) - 1))
|
2009-01-07 00:42:00 +08:00
|
|
|
|
2012-09-21 04:33:00 +08:00
|
|
|
#define MAX_CSUM_ITEMS(r, size) (min_t(u32, __MAX_CSUM_ITEMS(r, size), \
|
|
|
|
PAGE_CACHE_SIZE))
|
2012-02-01 09:19:02 +08:00
|
|
|
|
2009-01-07 00:42:00 +08:00
|
|
|
#define MAX_ORDERED_SUM_BYTES(r) ((PAGE_SIZE - \
|
|
|
|
sizeof(struct btrfs_ordered_sum)) / \
|
Btrfs: remove btrfs_sector_sum structure
Using the structure btrfs_sector_sum to keep the checksum value is
unnecessary, because the extents that btrfs_sector_sum points to are
continuous, we can find out the expected checksums by btrfs_ordered_sum's
bytenr and the offset, so we can remove btrfs_sector_sum's bytenr. After
removing bytenr, there is only one member in the structure, so it makes
no sense to keep the structure, just remove it, and use a u32 array to
store the checksum value.
By this change, we don't use the while loop to get the checksums one by
one. Now, we can get several checksum value at one time, it improved the
performance by ~74% on my SSD (31MB/s -> 54MB/s).
test command:
# dd if=/dev/zero of=/mnt/btrfs/file0 bs=1M count=1024 oflag=sync
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2013-06-19 10:36:09 +08:00
|
|
|
sizeof(u32) * (r)->sectorsize)
|
2009-01-07 00:42:00 +08:00
|
|
|
|
2007-04-18 01:26:50 +08:00
|
|
|
int btrfs_insert_file_extent(struct btrfs_trans_handle *trans,
|
2008-05-03 02:43:14 +08:00
|
|
|
struct btrfs_root *root,
|
|
|
|
u64 objectid, u64 pos,
|
|
|
|
u64 disk_offset, u64 disk_num_bytes,
|
Btrfs: Add zlib compression support
This is a large change for adding compression on reading and writing,
both for inline and regular extents. It does some fairly large
surgery to the writeback paths.
Compression is off by default and enabled by mount -o compress. Even
when the -o compress mount option is not used, it is possible to read
compressed extents off the disk.
If compression for a given set of pages fails to make them smaller, the
file is flagged to avoid future compression attempts later.
* While finding delalloc extents, the pages are locked before being sent down
to the delalloc handler. This allows the delalloc handler to do complex things
such as cleaning the pages, marking them writeback and starting IO on their
behalf.
* Inline extents are inserted at delalloc time now. This allows us to compress
the data before inserting the inline extent, and it allows us to insert
an inline extent that spans multiple pages.
* All of the in-memory extent representations (extent_map.c, ordered-data.c etc)
are changed to record both an in-memory size and an on disk size, as well
as a flag for compression.
From a disk format point of view, the extent pointers in the file are changed
to record the on disk size of a given extent and some encoding flags.
Space in the disk format is allocated for compression encoding, as well
as encryption and a generic 'other' field. Neither the encryption or the
'other' field are currently used.
In order to limit the amount of data read for a single random read in the
file, the size of a compressed extent is limited to 128k. This is a
software only limit, the disk format supports u64 sized compressed extents.
In order to limit the ram consumed while processing extents, the uncompressed
size of a compressed extent is limited to 256k. This is a software only limit
and will be subject to tuning later.
Checksumming is still done on compressed extents, and it is done on the
uncompressed version of the data. This way additional encodings can be
layered on without having to figure out which encoding to checksum.
Compression happens at delalloc time, which is basically singled threaded because
it is usually done by a single pdflush thread. This makes it tricky to
spread the compression load across all the cpus on the box. We'll have to
look at parallel pdflush walks of dirty inodes at a later time.
Decompression is hooked into readpages and it does spread across CPUs nicely.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-30 02:49:59 +08:00
|
|
|
u64 num_bytes, u64 offset, u64 ram_bytes,
|
|
|
|
u8 compression, u8 encryption, u16 other_encoding)
|
2007-03-21 02:38:32 +08:00
|
|
|
{
|
2007-03-27 04:00:06 +08:00
|
|
|
int ret = 0;
|
|
|
|
struct btrfs_file_extent_item *item;
|
|
|
|
struct btrfs_key file_key;
|
2007-04-02 23:20:42 +08:00
|
|
|
struct btrfs_path *path;
|
2007-10-16 04:14:19 +08:00
|
|
|
struct extent_buffer *leaf;
|
2007-03-27 04:00:06 +08:00
|
|
|
|
2007-04-02 23:20:42 +08:00
|
|
|
path = btrfs_alloc_path();
|
2011-03-23 16:14:16 +08:00
|
|
|
if (!path)
|
|
|
|
return -ENOMEM;
|
2007-03-27 04:00:06 +08:00
|
|
|
file_key.objectid = objectid;
|
2007-04-18 01:26:50 +08:00
|
|
|
file_key.offset = pos;
|
2007-03-27 04:00:06 +08:00
|
|
|
btrfs_set_key_type(&file_key, BTRFS_EXTENT_DATA_KEY);
|
|
|
|
|
2009-03-13 23:00:37 +08:00
|
|
|
path->leave_spinning = 1;
|
2007-04-02 23:20:42 +08:00
|
|
|
ret = btrfs_insert_empty_item(trans, root, path, &file_key,
|
2007-03-27 04:00:06 +08:00
|
|
|
sizeof(*item));
|
2007-06-23 02:16:25 +08:00
|
|
|
if (ret < 0)
|
|
|
|
goto out;
|
2012-03-12 23:03:00 +08:00
|
|
|
BUG_ON(ret); /* Can't happen */
|
2007-10-16 04:14:19 +08:00
|
|
|
leaf = path->nodes[0];
|
|
|
|
item = btrfs_item_ptr(leaf, path->slots[0],
|
2007-03-27 04:00:06 +08:00
|
|
|
struct btrfs_file_extent_item);
|
2008-05-03 02:43:14 +08:00
|
|
|
btrfs_set_file_extent_disk_bytenr(leaf, item, disk_offset);
|
2007-10-16 04:15:53 +08:00
|
|
|
btrfs_set_file_extent_disk_num_bytes(leaf, item, disk_num_bytes);
|
2008-05-03 02:43:14 +08:00
|
|
|
btrfs_set_file_extent_offset(leaf, item, offset);
|
2007-10-16 04:15:53 +08:00
|
|
|
btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
|
Btrfs: Add zlib compression support
This is a large change for adding compression on reading and writing,
both for inline and regular extents. It does some fairly large
surgery to the writeback paths.
Compression is off by default and enabled by mount -o compress. Even
when the -o compress mount option is not used, it is possible to read
compressed extents off the disk.
If compression for a given set of pages fails to make them smaller, the
file is flagged to avoid future compression attempts later.
* While finding delalloc extents, the pages are locked before being sent down
to the delalloc handler. This allows the delalloc handler to do complex things
such as cleaning the pages, marking them writeback and starting IO on their
behalf.
* Inline extents are inserted at delalloc time now. This allows us to compress
the data before inserting the inline extent, and it allows us to insert
an inline extent that spans multiple pages.
* All of the in-memory extent representations (extent_map.c, ordered-data.c etc)
are changed to record both an in-memory size and an on disk size, as well
as a flag for compression.
From a disk format point of view, the extent pointers in the file are changed
to record the on disk size of a given extent and some encoding flags.
Space in the disk format is allocated for compression encoding, as well
as encryption and a generic 'other' field. Neither the encryption or the
'other' field are currently used.
In order to limit the amount of data read for a single random read in the
file, the size of a compressed extent is limited to 128k. This is a
software only limit, the disk format supports u64 sized compressed extents.
In order to limit the ram consumed while processing extents, the uncompressed
size of a compressed extent is limited to 256k. This is a software only limit
and will be subject to tuning later.
Checksumming is still done on compressed extents, and it is done on the
uncompressed version of the data. This way additional encodings can be
layered on without having to figure out which encoding to checksum.
Compression happens at delalloc time, which is basically singled threaded because
it is usually done by a single pdflush thread. This makes it tricky to
spread the compression load across all the cpus on the box. We'll have to
look at parallel pdflush walks of dirty inodes at a later time.
Decompression is hooked into readpages and it does spread across CPUs nicely.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-30 02:49:59 +08:00
|
|
|
btrfs_set_file_extent_ram_bytes(leaf, item, ram_bytes);
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_set_file_extent_generation(leaf, item, trans->transid);
|
|
|
|
btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
|
Btrfs: Add zlib compression support
This is a large change for adding compression on reading and writing,
both for inline and regular extents. It does some fairly large
surgery to the writeback paths.
Compression is off by default and enabled by mount -o compress. Even
when the -o compress mount option is not used, it is possible to read
compressed extents off the disk.
If compression for a given set of pages fails to make them smaller, the
file is flagged to avoid future compression attempts later.
* While finding delalloc extents, the pages are locked before being sent down
to the delalloc handler. This allows the delalloc handler to do complex things
such as cleaning the pages, marking them writeback and starting IO on their
behalf.
* Inline extents are inserted at delalloc time now. This allows us to compress
the data before inserting the inline extent, and it allows us to insert
an inline extent that spans multiple pages.
* All of the in-memory extent representations (extent_map.c, ordered-data.c etc)
are changed to record both an in-memory size and an on disk size, as well
as a flag for compression.
From a disk format point of view, the extent pointers in the file are changed
to record the on disk size of a given extent and some encoding flags.
Space in the disk format is allocated for compression encoding, as well
as encryption and a generic 'other' field. Neither the encryption or the
'other' field are currently used.
In order to limit the amount of data read for a single random read in the
file, the size of a compressed extent is limited to 128k. This is a
software only limit, the disk format supports u64 sized compressed extents.
In order to limit the ram consumed while processing extents, the uncompressed
size of a compressed extent is limited to 256k. This is a software only limit
and will be subject to tuning later.
Checksumming is still done on compressed extents, and it is done on the
uncompressed version of the data. This way additional encodings can be
layered on without having to figure out which encoding to checksum.
Compression happens at delalloc time, which is basically singled threaded because
it is usually done by a single pdflush thread. This makes it tricky to
spread the compression load across all the cpus on the box. We'll have to
look at parallel pdflush walks of dirty inodes at a later time.
Decompression is hooked into readpages and it does spread across CPUs nicely.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-30 02:49:59 +08:00
|
|
|
btrfs_set_file_extent_compression(leaf, item, compression);
|
|
|
|
btrfs_set_file_extent_encryption(leaf, item, encryption);
|
|
|
|
btrfs_set_file_extent_other_encoding(leaf, item, other_encoding);
|
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_mark_buffer_dirty(leaf);
|
2007-06-23 02:16:25 +08:00
|
|
|
out:
|
2007-04-02 23:20:42 +08:00
|
|
|
btrfs_free_path(path);
|
2007-06-23 02:16:25 +08:00
|
|
|
return ret;
|
2007-03-21 02:38:32 +08:00
|
|
|
}
|
2007-03-27 04:00:06 +08:00
|
|
|
|
2013-04-26 04:41:01 +08:00
|
|
|
static struct btrfs_csum_item *
|
|
|
|
btrfs_lookup_csum(struct btrfs_trans_handle *trans,
|
|
|
|
struct btrfs_root *root,
|
|
|
|
struct btrfs_path *path,
|
|
|
|
u64 bytenr, int cow)
|
2007-04-16 21:22:45 +08:00
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
struct btrfs_key file_key;
|
|
|
|
struct btrfs_key found_key;
|
|
|
|
struct btrfs_csum_item *item;
|
2007-10-16 04:14:19 +08:00
|
|
|
struct extent_buffer *leaf;
|
2007-04-16 21:22:45 +08:00
|
|
|
u64 csum_offset = 0;
|
2011-04-13 21:41:04 +08:00
|
|
|
u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
|
2007-04-19 04:15:28 +08:00
|
|
|
int csums_in_item;
|
2007-04-16 21:22:45 +08:00
|
|
|
|
Btrfs: move data checksumming into a dedicated tree
Btrfs stores checksums for each data block. Until now, they have
been stored in the subvolume trees, indexed by the inode that is
referencing the data block. This means that when we read the inode,
we've probably read in at least some checksums as well.
But, this has a few problems:
* The checksums are indexed by logical offset in the file. When
compression is on, this means we have to do the expensive checksumming
on the uncompressed data. It would be faster if we could checksum
the compressed data instead.
* If we implement encryption, we'll be checksumming the plain text and
storing that on disk. This is significantly less secure.
* For either compression or encryption, we have to get the plain text
back before we can verify the checksum as correct. This makes the raid
layer balancing and extent moving much more expensive.
* It makes the front end caching code more complex, as we have touch
the subvolume and inodes as we cache extents.
* There is potentitally one copy of the checksum in each subvolume
referencing an extent.
The solution used here is to store the extent checksums in a dedicated
tree. This allows us to index the checksums by phyiscal extent
start and length. It means:
* The checksum is against the data stored on disk, after any compression
or encryption is done.
* The checksum is stored in a central location, and can be verified without
following back references, or reading inodes.
This makes compression significantly faster by reducing the amount of
data that needs to be checksummed. It will also allow much faster
raid management code in general.
The checksums are indexed by a key with a fixed objectid (a magic value
in ctree.h) and offset set to the starting byte of the extent. This
allows us to copy the checksum items into the fsync log tree directly (or
any other tree), without having to invent a second format for them.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-09 05:58:54 +08:00
|
|
|
file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
|
|
|
|
file_key.offset = bytenr;
|
|
|
|
btrfs_set_key_type(&file_key, BTRFS_EXTENT_CSUM_KEY);
|
2007-04-18 01:26:50 +08:00
|
|
|
ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
|
2007-04-16 21:22:45 +08:00
|
|
|
if (ret < 0)
|
|
|
|
goto fail;
|
2007-10-16 04:14:19 +08:00
|
|
|
leaf = path->nodes[0];
|
2007-04-16 21:22:45 +08:00
|
|
|
if (ret > 0) {
|
|
|
|
ret = 1;
|
2007-04-18 03:39:32 +08:00
|
|
|
if (path->slots[0] == 0)
|
2007-04-16 21:22:45 +08:00
|
|
|
goto fail;
|
|
|
|
path->slots[0]--;
|
2007-10-16 04:14:19 +08:00
|
|
|
btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
|
Btrfs: move data checksumming into a dedicated tree
Btrfs stores checksums for each data block. Until now, they have
been stored in the subvolume trees, indexed by the inode that is
referencing the data block. This means that when we read the inode,
we've probably read in at least some checksums as well.
But, this has a few problems:
* The checksums are indexed by logical offset in the file. When
compression is on, this means we have to do the expensive checksumming
on the uncompressed data. It would be faster if we could checksum
the compressed data instead.
* If we implement encryption, we'll be checksumming the plain text and
storing that on disk. This is significantly less secure.
* For either compression or encryption, we have to get the plain text
back before we can verify the checksum as correct. This makes the raid
layer balancing and extent moving much more expensive.
* It makes the front end caching code more complex, as we have touch
the subvolume and inodes as we cache extents.
* There is potentitally one copy of the checksum in each subvolume
referencing an extent.
The solution used here is to store the extent checksums in a dedicated
tree. This allows us to index the checksums by phyiscal extent
start and length. It means:
* The checksum is against the data stored on disk, after any compression
or encryption is done.
* The checksum is stored in a central location, and can be verified without
following back references, or reading inodes.
This makes compression significantly faster by reducing the amount of
data that needs to be checksummed. It will also allow much faster
raid management code in general.
The checksums are indexed by a key with a fixed objectid (a magic value
in ctree.h) and offset set to the starting byte of the extent. This
allows us to copy the checksum items into the fsync log tree directly (or
any other tree), without having to invent a second format for them.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-09 05:58:54 +08:00
|
|
|
if (btrfs_key_type(&found_key) != BTRFS_EXTENT_CSUM_KEY)
|
2007-04-16 21:22:45 +08:00
|
|
|
goto fail;
|
Btrfs: move data checksumming into a dedicated tree
Btrfs stores checksums for each data block. Until now, they have
been stored in the subvolume trees, indexed by the inode that is
referencing the data block. This means that when we read the inode,
we've probably read in at least some checksums as well.
But, this has a few problems:
* The checksums are indexed by logical offset in the file. When
compression is on, this means we have to do the expensive checksumming
on the uncompressed data. It would be faster if we could checksum
the compressed data instead.
* If we implement encryption, we'll be checksumming the plain text and
storing that on disk. This is significantly less secure.
* For either compression or encryption, we have to get the plain text
back before we can verify the checksum as correct. This makes the raid
layer balancing and extent moving much more expensive.
* It makes the front end caching code more complex, as we have touch
the subvolume and inodes as we cache extents.
* There is potentitally one copy of the checksum in each subvolume
referencing an extent.
The solution used here is to store the extent checksums in a dedicated
tree. This allows us to index the checksums by phyiscal extent
start and length. It means:
* The checksum is against the data stored on disk, after any compression
or encryption is done.
* The checksum is stored in a central location, and can be verified without
following back references, or reading inodes.
This makes compression significantly faster by reducing the amount of
data that needs to be checksummed. It will also allow much faster
raid management code in general.
The checksums are indexed by a key with a fixed objectid (a magic value
in ctree.h) and offset set to the starting byte of the extent. This
allows us to copy the checksum items into the fsync log tree directly (or
any other tree), without having to invent a second format for them.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-09 05:58:54 +08:00
|
|
|
|
|
|
|
csum_offset = (bytenr - found_key.offset) >>
|
2007-04-16 21:22:45 +08:00
|
|
|
root->fs_info->sb->s_blocksize_bits;
|
2007-10-16 04:14:19 +08:00
|
|
|
csums_in_item = btrfs_item_size_nr(leaf, path->slots[0]);
|
2008-12-02 20:17:45 +08:00
|
|
|
csums_in_item /= csum_size;
|
2007-04-19 04:15:28 +08:00
|
|
|
|
2013-03-28 16:12:15 +08:00
|
|
|
if (csum_offset == csums_in_item) {
|
2007-04-19 04:15:28 +08:00
|
|
|
ret = -EFBIG;
|
2007-04-16 21:22:45 +08:00
|
|
|
goto fail;
|
2013-03-28 16:12:15 +08:00
|
|
|
} else if (csum_offset > csums_in_item) {
|
|
|
|
goto fail;
|
2007-04-16 21:22:45 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
|
2007-05-11 00:36:17 +08:00
|
|
|
item = (struct btrfs_csum_item *)((unsigned char *)item +
|
2008-12-02 20:17:45 +08:00
|
|
|
csum_offset * csum_size);
|
2007-04-16 21:22:45 +08:00
|
|
|
return item;
|
|
|
|
fail:
|
|
|
|
if (ret > 0)
|
2007-04-18 01:26:50 +08:00
|
|
|
ret = -ENOENT;
|
2007-04-16 21:22:45 +08:00
|
|
|
return ERR_PTR(ret);
|
|
|
|
}
|
|
|
|
|
2007-03-27 04:00:06 +08:00
|
|
|
int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
|
|
|
|
struct btrfs_root *root,
|
|
|
|
struct btrfs_path *path, u64 objectid,
|
2007-03-27 23:26:26 +08:00
|
|
|
u64 offset, int mod)
|
2007-03-27 04:00:06 +08:00
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
struct btrfs_key file_key;
|
|
|
|
int ins_len = mod < 0 ? -1 : 0;
|
|
|
|
int cow = mod != 0;
|
|
|
|
|
|
|
|
file_key.objectid = objectid;
|
2007-04-18 03:39:32 +08:00
|
|
|
file_key.offset = offset;
|
2007-03-27 04:00:06 +08:00
|
|
|
btrfs_set_key_type(&file_key, BTRFS_EXTENT_DATA_KEY);
|
|
|
|
ret = btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
|
|
|
|
return ret;
|
|
|
|
}
|
2007-03-30 03:15:27 +08:00
|
|
|
|
2013-07-25 19:22:34 +08:00
|
|
|
static void btrfs_io_bio_endio_readpage(struct btrfs_io_bio *bio, int err)
|
|
|
|
{
|
|
|
|
kfree(bio->csum_allocated);
|
|
|
|
}
|
|
|
|
|
2010-05-23 23:00:55 +08:00
|
|
|
static int __btrfs_lookup_bio_sums(struct btrfs_root *root,
|
|
|
|
struct inode *inode, struct bio *bio,
|
|
|
|
u64 logical_offset, u32 *dst, int dio)
|
2008-08-01 03:42:53 +08:00
|
|
|
{
|
|
|
|
struct bio_vec *bvec = bio->bi_io_vec;
|
2013-07-25 19:22:34 +08:00
|
|
|
struct btrfs_io_bio *btrfs_bio = btrfs_io_bio(bio);
|
|
|
|
struct btrfs_csum_item *item = NULL;
|
|
|
|
struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
|
|
|
|
struct btrfs_path *path;
|
|
|
|
u8 *csum;
|
2010-05-23 23:00:55 +08:00
|
|
|
u64 offset = 0;
|
2008-08-01 03:42:53 +08:00
|
|
|
u64 item_start_offset = 0;
|
|
|
|
u64 item_last_offset = 0;
|
Btrfs: move data checksumming into a dedicated tree
Btrfs stores checksums for each data block. Until now, they have
been stored in the subvolume trees, indexed by the inode that is
referencing the data block. This means that when we read the inode,
we've probably read in at least some checksums as well.
But, this has a few problems:
* The checksums are indexed by logical offset in the file. When
compression is on, this means we have to do the expensive checksumming
on the uncompressed data. It would be faster if we could checksum
the compressed data instead.
* If we implement encryption, we'll be checksumming the plain text and
storing that on disk. This is significantly less secure.
* For either compression or encryption, we have to get the plain text
back before we can verify the checksum as correct. This makes the raid
layer balancing and extent moving much more expensive.
* It makes the front end caching code more complex, as we have touch
the subvolume and inodes as we cache extents.
* There is potentitally one copy of the checksum in each subvolume
referencing an extent.
The solution used here is to store the extent checksums in a dedicated
tree. This allows us to index the checksums by phyiscal extent
start and length. It means:
* The checksum is against the data stored on disk, after any compression
or encryption is done.
* The checksum is stored in a central location, and can be verified without
following back references, or reading inodes.
This makes compression significantly faster by reducing the amount of
data that needs to be checksummed. It will also allow much faster
raid management code in general.
The checksums are indexed by a key with a fixed objectid (a magic value
in ctree.h) and offset set to the starting byte of the extent. This
allows us to copy the checksum items into the fsync log tree directly (or
any other tree), without having to invent a second format for them.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-09 05:58:54 +08:00
|
|
|
u64 disk_bytenr;
|
2008-08-01 03:42:53 +08:00
|
|
|
u32 diff;
|
2013-07-25 19:22:34 +08:00
|
|
|
int nblocks;
|
|
|
|
int bio_index = 0;
|
2013-04-05 15:20:56 +08:00
|
|
|
int count;
|
2013-07-25 19:22:34 +08:00
|
|
|
u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
|
2008-08-01 03:42:53 +08:00
|
|
|
|
|
|
|
path = btrfs_alloc_path();
|
2011-03-01 14:48:31 +08:00
|
|
|
if (!path)
|
|
|
|
return -ENOMEM;
|
2013-07-25 19:22:34 +08:00
|
|
|
|
2013-10-12 06:44:27 +08:00
|
|
|
nblocks = bio->bi_iter.bi_size >> inode->i_sb->s_blocksize_bits;
|
2013-07-25 19:22:34 +08:00
|
|
|
if (!dst) {
|
|
|
|
if (nblocks * csum_size > BTRFS_BIO_INLINE_CSUM_SIZE) {
|
|
|
|
btrfs_bio->csum_allocated = kmalloc(nblocks * csum_size,
|
|
|
|
GFP_NOFS);
|
|
|
|
if (!btrfs_bio->csum_allocated) {
|
|
|
|
btrfs_free_path(path);
|
|
|
|
return -ENOMEM;
|
|
|
|
}
|
|
|
|
btrfs_bio->csum = btrfs_bio->csum_allocated;
|
|
|
|
btrfs_bio->end_io = btrfs_io_bio_endio_readpage;
|
|
|
|
} else {
|
|
|
|
btrfs_bio->csum = btrfs_bio->csum_inline;
|
|
|
|
}
|
|
|
|
csum = btrfs_bio->csum;
|
|
|
|
} else {
|
|
|
|
csum = (u8 *)dst;
|
|
|
|
}
|
|
|
|
|
2013-10-12 06:44:27 +08:00
|
|
|
if (bio->bi_iter.bi_size > PAGE_CACHE_SIZE * 8)
|
2008-08-20 21:44:52 +08:00
|
|
|
path->reada = 2;
|
2008-08-01 03:42:53 +08:00
|
|
|
|
|
|
|
WARN_ON(bio->bi_vcnt <= 0);
|
|
|
|
|
2011-07-27 03:35:09 +08:00
|
|
|
/*
|
|
|
|
* the free space stuff is only read when it hasn't been
|
|
|
|
* updated in the current transaction. So, we can safely
|
|
|
|
* read from the commit root and sidestep a nasty deadlock
|
|
|
|
* between reading the free space cache and updating the csum tree.
|
|
|
|
*/
|
2012-07-10 19:28:39 +08:00
|
|
|
if (btrfs_is_free_space_inode(inode)) {
|
2011-07-27 03:35:09 +08:00
|
|
|
path->search_commit_root = 1;
|
2011-09-11 22:52:24 +08:00
|
|
|
path->skip_locking = 1;
|
|
|
|
}
|
2011-07-27 03:35:09 +08:00
|
|
|
|
2013-10-12 06:44:27 +08:00
|
|
|
disk_bytenr = (u64)bio->bi_iter.bi_sector << 9;
|
2010-05-23 23:00:55 +08:00
|
|
|
if (dio)
|
|
|
|
offset = logical_offset;
|
2009-01-06 10:25:51 +08:00
|
|
|
while (bio_index < bio->bi_vcnt) {
|
2010-05-23 23:00:55 +08:00
|
|
|
if (!dio)
|
|
|
|
offset = page_offset(bvec->bv_page) + bvec->bv_offset;
|
2013-07-25 19:22:34 +08:00
|
|
|
count = btrfs_find_ordered_sum(inode, offset, disk_bytenr,
|
|
|
|
(u32 *)csum, nblocks);
|
2013-04-05 15:20:56 +08:00
|
|
|
if (count)
|
2008-08-01 03:42:53 +08:00
|
|
|
goto found;
|
|
|
|
|
Btrfs: move data checksumming into a dedicated tree
Btrfs stores checksums for each data block. Until now, they have
been stored in the subvolume trees, indexed by the inode that is
referencing the data block. This means that when we read the inode,
we've probably read in at least some checksums as well.
But, this has a few problems:
* The checksums are indexed by logical offset in the file. When
compression is on, this means we have to do the expensive checksumming
on the uncompressed data. It would be faster if we could checksum
the compressed data instead.
* If we implement encryption, we'll be checksumming the plain text and
storing that on disk. This is significantly less secure.
* For either compression or encryption, we have to get the plain text
back before we can verify the checksum as correct. This makes the raid
layer balancing and extent moving much more expensive.
* It makes the front end caching code more complex, as we have touch
the subvolume and inodes as we cache extents.
* There is potentitally one copy of the checksum in each subvolume
referencing an extent.
The solution used here is to store the extent checksums in a dedicated
tree. This allows us to index the checksums by phyiscal extent
start and length. It means:
* The checksum is against the data stored on disk, after any compression
or encryption is done.
* The checksum is stored in a central location, and can be verified without
following back references, or reading inodes.
This makes compression significantly faster by reducing the amount of
data that needs to be checksummed. It will also allow much faster
raid management code in general.
The checksums are indexed by a key with a fixed objectid (a magic value
in ctree.h) and offset set to the starting byte of the extent. This
allows us to copy the checksum items into the fsync log tree directly (or
any other tree), without having to invent a second format for them.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-09 05:58:54 +08:00
|
|
|
if (!item || disk_bytenr < item_start_offset ||
|
|
|
|
disk_bytenr >= item_last_offset) {
|
2008-08-01 03:42:53 +08:00
|
|
|
struct btrfs_key found_key;
|
|
|
|
u32 item_size;
|
|
|
|
|
|
|
|
if (item)
|
2011-04-21 07:20:15 +08:00
|
|
|
btrfs_release_path(path);
|
Btrfs: move data checksumming into a dedicated tree
Btrfs stores checksums for each data block. Until now, they have
been stored in the subvolume trees, indexed by the inode that is
referencing the data block. This means that when we read the inode,
we've probably read in at least some checksums as well.
But, this has a few problems:
* The checksums are indexed by logical offset in the file. When
compression is on, this means we have to do the expensive checksumming
on the uncompressed data. It would be faster if we could checksum
the compressed data instead.
* If we implement encryption, we'll be checksumming the plain text and
storing that on disk. This is significantly less secure.
* For either compression or encryption, we have to get the plain text
back before we can verify the checksum as correct. This makes the raid
layer balancing and extent moving much more expensive.
* It makes the front end caching code more complex, as we have touch
the subvolume and inodes as we cache extents.
* There is potentitally one copy of the checksum in each subvolume
referencing an extent.
The solution used here is to store the extent checksums in a dedicated
tree. This allows us to index the checksums by phyiscal extent
start and length. It means:
* The checksum is against the data stored on disk, after any compression
or encryption is done.
* The checksum is stored in a central location, and can be verified without
following back references, or reading inodes.
This makes compression significantly faster by reducing the amount of
data that needs to be checksummed. It will also allow much faster
raid management code in general.
The checksums are indexed by a key with a fixed objectid (a magic value
in ctree.h) and offset set to the starting byte of the extent. This
allows us to copy the checksum items into the fsync log tree directly (or
any other tree), without having to invent a second format for them.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-09 05:58:54 +08:00
|
|
|
item = btrfs_lookup_csum(NULL, root->fs_info->csum_root,
|
|
|
|
path, disk_bytenr, 0);
|
2008-08-01 03:42:53 +08:00
|
|
|
if (IS_ERR(item)) {
|
2013-04-05 15:20:56 +08:00
|
|
|
count = 1;
|
2013-07-25 19:22:34 +08:00
|
|
|
memset(csum, 0, csum_size);
|
2008-12-12 23:03:38 +08:00
|
|
|
if (BTRFS_I(inode)->root->root_key.objectid ==
|
|
|
|
BTRFS_DATA_RELOC_TREE_OBJECTID) {
|
|
|
|
set_extent_bits(io_tree, offset,
|
|
|
|
offset + bvec->bv_len - 1,
|
|
|
|
EXTENT_NODATASUM, GFP_NOFS);
|
|
|
|
} else {
|
2013-12-21 00:37:06 +08:00
|
|
|
btrfs_info(BTRFS_I(inode)->root->fs_info,
|
|
|
|
"no csum found for inode %llu start %llu",
|
2013-08-20 19:20:07 +08:00
|
|
|
btrfs_ino(inode), offset);
|
2008-12-12 23:03:38 +08:00
|
|
|
}
|
2008-08-04 20:35:53 +08:00
|
|
|
item = NULL;
|
2011-04-21 07:20:15 +08:00
|
|
|
btrfs_release_path(path);
|
2008-08-01 03:42:53 +08:00
|
|
|
goto found;
|
|
|
|
}
|
|
|
|
btrfs_item_key_to_cpu(path->nodes[0], &found_key,
|
|
|
|
path->slots[0]);
|
|
|
|
|
|
|
|
item_start_offset = found_key.offset;
|
|
|
|
item_size = btrfs_item_size_nr(path->nodes[0],
|
|
|
|
path->slots[0]);
|
|
|
|
item_last_offset = item_start_offset +
|
2008-12-02 20:17:45 +08:00
|
|
|
(item_size / csum_size) *
|
2008-08-01 03:42:53 +08:00
|
|
|
root->sectorsize;
|
|
|
|
item = btrfs_item_ptr(path->nodes[0], path->slots[0],
|
|
|
|
struct btrfs_csum_item);
|
|
|
|
}
|
|
|
|
/*
|
|
|
|
* this byte range must be able to fit inside
|
|
|
|
* a single leaf so it will also fit inside a u32
|
|
|
|
*/
|
Btrfs: move data checksumming into a dedicated tree
Btrfs stores checksums for each data block. Until now, they have
been stored in the subvolume trees, indexed by the inode that is
referencing the data block. This means that when we read the inode,
we've probably read in at least some checksums as well.
But, this has a few problems:
* The checksums are indexed by logical offset in the file. When
compression is on, this means we have to do the expensive checksumming
on the uncompressed data. It would be faster if we could checksum
the compressed data instead.
* If we implement encryption, we'll be checksumming the plain text and
storing that on disk. This is significantly less secure.
* For either compression or encryption, we have to get the plain text
back before we can verify the checksum as correct. This makes the raid
layer balancing and extent moving much more expensive.
* It makes the front end caching code more complex, as we have touch
the subvolume and inodes as we cache extents.
* There is potentitally one copy of the checksum in each subvolume
referencing an extent.
The solution used here is to store the extent checksums in a dedicated
tree. This allows us to index the checksums by phyiscal extent
start and length. It means:
* The checksum is against the data stored on disk, after any compression
or encryption is done.
* The checksum is stored in a central location, and can be verified without
following back references, or reading inodes.
This makes compression significantly faster by reducing the amount of
data that needs to be checksummed. It will also allow much faster
raid management code in general.
The checksums are indexed by a key with a fixed objectid (a magic value
in ctree.h) and offset set to the starting byte of the extent. This
allows us to copy the checksum items into the fsync log tree directly (or
any other tree), without having to invent a second format for them.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-09 05:58:54 +08:00
|
|
|
diff = disk_bytenr - item_start_offset;
|
2008-08-01 03:42:53 +08:00
|
|
|
diff = diff / root->sectorsize;
|
2008-12-02 20:17:45 +08:00
|
|
|
diff = diff * csum_size;
|
2013-07-25 19:22:34 +08:00
|
|
|
count = min_t(int, nblocks, (item_last_offset - disk_bytenr) >>
|
|
|
|
inode->i_sb->s_blocksize_bits);
|
|
|
|
read_extent_buffer(path->nodes[0], csum,
|
2008-08-05 11:17:27 +08:00
|
|
|
((unsigned long)item) + diff,
|
2013-04-05 15:20:56 +08:00
|
|
|
csum_size * count);
|
2008-08-01 03:42:53 +08:00
|
|
|
found:
|
2013-07-25 19:22:34 +08:00
|
|
|
csum += count * csum_size;
|
|
|
|
nblocks -= count;
|
2014-04-29 13:07:58 +08:00
|
|
|
bio_index += count;
|
2013-04-05 15:20:56 +08:00
|
|
|
while (count--) {
|
|
|
|
disk_bytenr += bvec->bv_len;
|
|
|
|
offset += bvec->bv_len;
|
|
|
|
bvec++;
|
|
|
|
}
|
2008-08-01 03:42:53 +08:00
|
|
|
}
|
|
|
|
btrfs_free_path(path);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2010-05-23 23:00:55 +08:00
|
|
|
int btrfs_lookup_bio_sums(struct btrfs_root *root, struct inode *inode,
|
|
|
|
struct bio *bio, u32 *dst)
|
|
|
|
{
|
|
|
|
return __btrfs_lookup_bio_sums(root, inode, bio, 0, dst, 0);
|
|
|
|
}
|
|
|
|
|
|
|
|
int btrfs_lookup_bio_sums_dio(struct btrfs_root *root, struct inode *inode,
|
2013-07-25 19:22:34 +08:00
|
|
|
struct btrfs_dio_private *dip, struct bio *bio,
|
|
|
|
u64 offset)
|
2010-05-23 23:00:55 +08:00
|
|
|
{
|
2013-10-12 06:44:27 +08:00
|
|
|
int len = (bio->bi_iter.bi_sector << 9) - dip->disk_bytenr;
|
2013-07-25 19:22:34 +08:00
|
|
|
u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
len >>= inode->i_sb->s_blocksize_bits;
|
|
|
|
len *= csum_size;
|
|
|
|
|
|
|
|
ret = __btrfs_lookup_bio_sums(root, inode, bio, offset,
|
|
|
|
(u32 *)(dip->csum + len), 1);
|
|
|
|
return ret;
|
2010-05-23 23:00:55 +08:00
|
|
|
}
|
|
|
|
|
2008-12-12 23:03:38 +08:00
|
|
|
int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
|
2011-03-08 21:14:00 +08:00
|
|
|
struct list_head *list, int search_commit)
|
2008-12-12 23:03:38 +08:00
|
|
|
{
|
|
|
|
struct btrfs_key key;
|
|
|
|
struct btrfs_path *path;
|
|
|
|
struct extent_buffer *leaf;
|
|
|
|
struct btrfs_ordered_sum *sums;
|
|
|
|
struct btrfs_csum_item *item;
|
2011-08-06 06:46:16 +08:00
|
|
|
LIST_HEAD(tmplist);
|
2008-12-12 23:03:38 +08:00
|
|
|
unsigned long offset;
|
|
|
|
int ret;
|
|
|
|
size_t size;
|
|
|
|
u64 csum_end;
|
2011-04-13 21:41:04 +08:00
|
|
|
u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
|
2008-12-12 23:03:38 +08:00
|
|
|
|
2013-10-15 21:36:40 +08:00
|
|
|
ASSERT(start == ALIGN(start, root->sectorsize) &&
|
|
|
|
(end + 1) == ALIGN(end + 1, root->sectorsize));
|
|
|
|
|
2008-12-12 23:03:38 +08:00
|
|
|
path = btrfs_alloc_path();
|
btrfs: don't BUG_ON btrfs_alloc_path() errors
This patch fixes many callers of btrfs_alloc_path() which BUG_ON allocation
failure. All the sites that are fixed in this patch were checked by me to
be fairly trivial to fix because of at least one of two criteria:
- Callers of the function catch errors from it already so bubbling the
error up will be handled.
- Callers of the function might BUG_ON any nonzero return code in which
case there is no behavior changed (but we still got to remove a BUG_ON)
The following functions were updated:
btrfs_lookup_extent, alloc_reserved_tree_block, btrfs_remove_block_group,
btrfs_lookup_csums_range, btrfs_csum_file_blocks, btrfs_mark_extent_written,
btrfs_inode_by_name, btrfs_new_inode, btrfs_symlink,
insert_reserved_file_extent, and run_delalloc_nocow
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
2011-07-14 01:38:47 +08:00
|
|
|
if (!path)
|
|
|
|
return -ENOMEM;
|
2008-12-12 23:03:38 +08:00
|
|
|
|
2011-03-08 21:14:00 +08:00
|
|
|
if (search_commit) {
|
|
|
|
path->skip_locking = 1;
|
|
|
|
path->reada = 2;
|
|
|
|
path->search_commit_root = 1;
|
|
|
|
}
|
|
|
|
|
2008-12-12 23:03:38 +08:00
|
|
|
key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
|
|
|
|
key.offset = start;
|
|
|
|
key.type = BTRFS_EXTENT_CSUM_KEY;
|
|
|
|
|
2009-01-07 00:42:00 +08:00
|
|
|
ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
|
2008-12-12 23:03:38 +08:00
|
|
|
if (ret < 0)
|
|
|
|
goto fail;
|
|
|
|
if (ret > 0 && path->slots[0] > 0) {
|
|
|
|
leaf = path->nodes[0];
|
|
|
|
btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
|
|
|
|
if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
|
|
|
|
key.type == BTRFS_EXTENT_CSUM_KEY) {
|
|
|
|
offset = (start - key.offset) >>
|
|
|
|
root->fs_info->sb->s_blocksize_bits;
|
|
|
|
if (offset * csum_size <
|
|
|
|
btrfs_item_size_nr(leaf, path->slots[0] - 1))
|
|
|
|
path->slots[0]--;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
while (start <= end) {
|
|
|
|
leaf = path->nodes[0];
|
|
|
|
if (path->slots[0] >= btrfs_header_nritems(leaf)) {
|
2009-01-07 00:42:00 +08:00
|
|
|
ret = btrfs_next_leaf(root, path);
|
2008-12-12 23:03:38 +08:00
|
|
|
if (ret < 0)
|
|
|
|
goto fail;
|
|
|
|
if (ret > 0)
|
|
|
|
break;
|
|
|
|
leaf = path->nodes[0];
|
|
|
|
}
|
|
|
|
|
|
|
|
btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
|
|
|
|
if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
|
2013-03-18 17:18:09 +08:00
|
|
|
key.type != BTRFS_EXTENT_CSUM_KEY ||
|
|
|
|
key.offset > end)
|
2008-12-12 23:03:38 +08:00
|
|
|
break;
|
|
|
|
|
|
|
|
if (key.offset > start)
|
|
|
|
start = key.offset;
|
|
|
|
|
|
|
|
size = btrfs_item_size_nr(leaf, path->slots[0]);
|
|
|
|
csum_end = key.offset + (size / csum_size) * root->sectorsize;
|
2008-12-17 23:21:48 +08:00
|
|
|
if (csum_end <= start) {
|
|
|
|
path->slots[0]++;
|
|
|
|
continue;
|
|
|
|
}
|
2008-12-12 23:03:38 +08:00
|
|
|
|
2009-01-07 00:42:00 +08:00
|
|
|
csum_end = min(csum_end, end + 1);
|
2008-12-12 23:03:38 +08:00
|
|
|
item = btrfs_item_ptr(path->nodes[0], path->slots[0],
|
|
|
|
struct btrfs_csum_item);
|
2009-01-07 00:42:00 +08:00
|
|
|
while (start < csum_end) {
|
|
|
|
size = min_t(size_t, csum_end - start,
|
Btrfs: remove btrfs_sector_sum structure
Using the structure btrfs_sector_sum to keep the checksum value is
unnecessary, because the extents that btrfs_sector_sum points to are
continuous, we can find out the expected checksums by btrfs_ordered_sum's
bytenr and the offset, so we can remove btrfs_sector_sum's bytenr. After
removing bytenr, there is only one member in the structure, so it makes
no sense to keep the structure, just remove it, and use a u32 array to
store the checksum value.
By this change, we don't use the while loop to get the checksums one by
one. Now, we can get several checksum value at one time, it improved the
performance by ~74% on my SSD (31MB/s -> 54MB/s).
test command:
# dd if=/dev/zero of=/mnt/btrfs/file0 bs=1M count=1024 oflag=sync
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2013-06-19 10:36:09 +08:00
|
|
|
MAX_ORDERED_SUM_BYTES(root));
|
2009-01-07 00:42:00 +08:00
|
|
|
sums = kzalloc(btrfs_ordered_sum_size(root, size),
|
Btrfs: remove btrfs_sector_sum structure
Using the structure btrfs_sector_sum to keep the checksum value is
unnecessary, because the extents that btrfs_sector_sum points to are
continuous, we can find out the expected checksums by btrfs_ordered_sum's
bytenr and the offset, so we can remove btrfs_sector_sum's bytenr. After
removing bytenr, there is only one member in the structure, so it makes
no sense to keep the structure, just remove it, and use a u32 array to
store the checksum value.
By this change, we don't use the while loop to get the checksums one by
one. Now, we can get several checksum value at one time, it improved the
performance by ~74% on my SSD (31MB/s -> 54MB/s).
test command:
# dd if=/dev/zero of=/mnt/btrfs/file0 bs=1M count=1024 oflag=sync
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2013-06-19 10:36:09 +08:00
|
|
|
GFP_NOFS);
|
2011-08-06 06:46:16 +08:00
|
|
|
if (!sums) {
|
|
|
|
ret = -ENOMEM;
|
|
|
|
goto fail;
|
|
|
|
}
|
2008-12-12 23:03:38 +08:00
|
|
|
|
2009-01-07 00:42:00 +08:00
|
|
|
sums->bytenr = start;
|
Btrfs: remove btrfs_sector_sum structure
Using the structure btrfs_sector_sum to keep the checksum value is
unnecessary, because the extents that btrfs_sector_sum points to are
continuous, we can find out the expected checksums by btrfs_ordered_sum's
bytenr and the offset, so we can remove btrfs_sector_sum's bytenr. After
removing bytenr, there is only one member in the structure, so it makes
no sense to keep the structure, just remove it, and use a u32 array to
store the checksum value.
By this change, we don't use the while loop to get the checksums one by
one. Now, we can get several checksum value at one time, it improved the
performance by ~74% on my SSD (31MB/s -> 54MB/s).
test command:
# dd if=/dev/zero of=/mnt/btrfs/file0 bs=1M count=1024 oflag=sync
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2013-06-19 10:36:09 +08:00
|
|
|
sums->len = (int)size;
|
2009-01-07 00:42:00 +08:00
|
|
|
|
|
|
|
offset = (start - key.offset) >>
|
|
|
|
root->fs_info->sb->s_blocksize_bits;
|
|
|
|
offset *= csum_size;
|
Btrfs: remove btrfs_sector_sum structure
Using the structure btrfs_sector_sum to keep the checksum value is
unnecessary, because the extents that btrfs_sector_sum points to are
continuous, we can find out the expected checksums by btrfs_ordered_sum's
bytenr and the offset, so we can remove btrfs_sector_sum's bytenr. After
removing bytenr, there is only one member in the structure, so it makes
no sense to keep the structure, just remove it, and use a u32 array to
store the checksum value.
By this change, we don't use the while loop to get the checksums one by
one. Now, we can get several checksum value at one time, it improved the
performance by ~74% on my SSD (31MB/s -> 54MB/s).
test command:
# dd if=/dev/zero of=/mnt/btrfs/file0 bs=1M count=1024 oflag=sync
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2013-06-19 10:36:09 +08:00
|
|
|
size >>= root->fs_info->sb->s_blocksize_bits;
|
2009-01-07 00:42:00 +08:00
|
|
|
|
Btrfs: remove btrfs_sector_sum structure
Using the structure btrfs_sector_sum to keep the checksum value is
unnecessary, because the extents that btrfs_sector_sum points to are
continuous, we can find out the expected checksums by btrfs_ordered_sum's
bytenr and the offset, so we can remove btrfs_sector_sum's bytenr. After
removing bytenr, there is only one member in the structure, so it makes
no sense to keep the structure, just remove it, and use a u32 array to
store the checksum value.
By this change, we don't use the while loop to get the checksums one by
one. Now, we can get several checksum value at one time, it improved the
performance by ~74% on my SSD (31MB/s -> 54MB/s).
test command:
# dd if=/dev/zero of=/mnt/btrfs/file0 bs=1M count=1024 oflag=sync
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2013-06-19 10:36:09 +08:00
|
|
|
read_extent_buffer(path->nodes[0],
|
|
|
|
sums->sums,
|
|
|
|
((unsigned long)item) + offset,
|
|
|
|
csum_size * size);
|
|
|
|
|
|
|
|
start += root->sectorsize * size;
|
2011-08-06 06:46:16 +08:00
|
|
|
list_add_tail(&sums->list, &tmplist);
|
2009-01-07 00:42:00 +08:00
|
|
|
}
|
2008-12-12 23:03:38 +08:00
|
|
|
path->slots[0]++;
|
|
|
|
}
|
|
|
|
ret = 0;
|
|
|
|
fail:
|
2011-08-06 06:46:16 +08:00
|
|
|
while (ret < 0 && !list_empty(&tmplist)) {
|
|
|
|
sums = list_entry(&tmplist, struct btrfs_ordered_sum, list);
|
|
|
|
list_del(&sums->list);
|
|
|
|
kfree(sums);
|
|
|
|
}
|
|
|
|
list_splice_tail(&tmplist, list);
|
|
|
|
|
2008-12-12 23:03:38 +08:00
|
|
|
btrfs_free_path(path);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2008-07-18 18:17:13 +08:00
|
|
|
int btrfs_csum_one_bio(struct btrfs_root *root, struct inode *inode,
|
Btrfs: move data checksumming into a dedicated tree
Btrfs stores checksums for each data block. Until now, they have
been stored in the subvolume trees, indexed by the inode that is
referencing the data block. This means that when we read the inode,
we've probably read in at least some checksums as well.
But, this has a few problems:
* The checksums are indexed by logical offset in the file. When
compression is on, this means we have to do the expensive checksumming
on the uncompressed data. It would be faster if we could checksum
the compressed data instead.
* If we implement encryption, we'll be checksumming the plain text and
storing that on disk. This is significantly less secure.
* For either compression or encryption, we have to get the plain text
back before we can verify the checksum as correct. This makes the raid
layer balancing and extent moving much more expensive.
* It makes the front end caching code more complex, as we have touch
the subvolume and inodes as we cache extents.
* There is potentitally one copy of the checksum in each subvolume
referencing an extent.
The solution used here is to store the extent checksums in a dedicated
tree. This allows us to index the checksums by phyiscal extent
start and length. It means:
* The checksum is against the data stored on disk, after any compression
or encryption is done.
* The checksum is stored in a central location, and can be verified without
following back references, or reading inodes.
This makes compression significantly faster by reducing the amount of
data that needs to be checksummed. It will also allow much faster
raid management code in general.
The checksums are indexed by a key with a fixed objectid (a magic value
in ctree.h) and offset set to the starting byte of the extent. This
allows us to copy the checksum items into the fsync log tree directly (or
any other tree), without having to invent a second format for them.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-09 05:58:54 +08:00
|
|
|
struct bio *bio, u64 file_start, int contig)
|
2008-04-16 23:15:20 +08:00
|
|
|
{
|
2008-07-18 00:53:50 +08:00
|
|
|
struct btrfs_ordered_sum *sums;
|
2008-07-18 18:17:13 +08:00
|
|
|
struct btrfs_ordered_extent *ordered;
|
2008-04-16 23:15:20 +08:00
|
|
|
char *data;
|
|
|
|
struct bio_vec *bvec = bio->bi_io_vec;
|
|
|
|
int bio_index = 0;
|
Btrfs: remove btrfs_sector_sum structure
Using the structure btrfs_sector_sum to keep the checksum value is
unnecessary, because the extents that btrfs_sector_sum points to are
continuous, we can find out the expected checksums by btrfs_ordered_sum's
bytenr and the offset, so we can remove btrfs_sector_sum's bytenr. After
removing bytenr, there is only one member in the structure, so it makes
no sense to keep the structure, just remove it, and use a u32 array to
store the checksum value.
By this change, we don't use the while loop to get the checksums one by
one. Now, we can get several checksum value at one time, it improved the
performance by ~74% on my SSD (31MB/s -> 54MB/s).
test command:
# dd if=/dev/zero of=/mnt/btrfs/file0 bs=1M count=1024 oflag=sync
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2013-06-19 10:36:09 +08:00
|
|
|
int index;
|
2008-07-18 18:17:13 +08:00
|
|
|
unsigned long total_bytes = 0;
|
|
|
|
unsigned long this_sum_bytes = 0;
|
|
|
|
u64 offset;
|
2008-04-16 23:15:20 +08:00
|
|
|
|
2008-07-18 00:53:50 +08:00
|
|
|
WARN_ON(bio->bi_vcnt <= 0);
|
2013-10-12 06:44:27 +08:00
|
|
|
sums = kzalloc(btrfs_ordered_sum_size(root, bio->bi_iter.bi_size),
|
|
|
|
GFP_NOFS);
|
2008-04-16 23:15:20 +08:00
|
|
|
if (!sums)
|
|
|
|
return -ENOMEM;
|
2008-07-18 18:17:13 +08:00
|
|
|
|
2013-10-12 06:44:27 +08:00
|
|
|
sums->len = bio->bi_iter.bi_size;
|
2008-07-18 00:53:50 +08:00
|
|
|
INIT_LIST_HEAD(&sums->list);
|
Btrfs: move data checksumming into a dedicated tree
Btrfs stores checksums for each data block. Until now, they have
been stored in the subvolume trees, indexed by the inode that is
referencing the data block. This means that when we read the inode,
we've probably read in at least some checksums as well.
But, this has a few problems:
* The checksums are indexed by logical offset in the file. When
compression is on, this means we have to do the expensive checksumming
on the uncompressed data. It would be faster if we could checksum
the compressed data instead.
* If we implement encryption, we'll be checksumming the plain text and
storing that on disk. This is significantly less secure.
* For either compression or encryption, we have to get the plain text
back before we can verify the checksum as correct. This makes the raid
layer balancing and extent moving much more expensive.
* It makes the front end caching code more complex, as we have touch
the subvolume and inodes as we cache extents.
* There is potentitally one copy of the checksum in each subvolume
referencing an extent.
The solution used here is to store the extent checksums in a dedicated
tree. This allows us to index the checksums by phyiscal extent
start and length. It means:
* The checksum is against the data stored on disk, after any compression
or encryption is done.
* The checksum is stored in a central location, and can be verified without
following back references, or reading inodes.
This makes compression significantly faster by reducing the amount of
data that needs to be checksummed. It will also allow much faster
raid management code in general.
The checksums are indexed by a key with a fixed objectid (a magic value
in ctree.h) and offset set to the starting byte of the extent. This
allows us to copy the checksum items into the fsync log tree directly (or
any other tree), without having to invent a second format for them.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-09 05:58:54 +08:00
|
|
|
|
|
|
|
if (contig)
|
|
|
|
offset = file_start;
|
|
|
|
else
|
|
|
|
offset = page_offset(bvec->bv_page) + bvec->bv_offset;
|
|
|
|
|
|
|
|
ordered = btrfs_lookup_ordered_extent(inode, offset);
|
2012-03-12 23:03:00 +08:00
|
|
|
BUG_ON(!ordered); /* Logic error */
|
2013-10-12 06:44:27 +08:00
|
|
|
sums->bytenr = (u64)bio->bi_iter.bi_sector << 9;
|
Btrfs: remove btrfs_sector_sum structure
Using the structure btrfs_sector_sum to keep the checksum value is
unnecessary, because the extents that btrfs_sector_sum points to are
continuous, we can find out the expected checksums by btrfs_ordered_sum's
bytenr and the offset, so we can remove btrfs_sector_sum's bytenr. After
removing bytenr, there is only one member in the structure, so it makes
no sense to keep the structure, just remove it, and use a u32 array to
store the checksum value.
By this change, we don't use the while loop to get the checksums one by
one. Now, we can get several checksum value at one time, it improved the
performance by ~74% on my SSD (31MB/s -> 54MB/s).
test command:
# dd if=/dev/zero of=/mnt/btrfs/file0 bs=1M count=1024 oflag=sync
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2013-06-19 10:36:09 +08:00
|
|
|
index = 0;
|
2008-04-16 23:15:20 +08:00
|
|
|
|
2009-01-06 10:25:51 +08:00
|
|
|
while (bio_index < bio->bi_vcnt) {
|
Btrfs: move data checksumming into a dedicated tree
Btrfs stores checksums for each data block. Until now, they have
been stored in the subvolume trees, indexed by the inode that is
referencing the data block. This means that when we read the inode,
we've probably read in at least some checksums as well.
But, this has a few problems:
* The checksums are indexed by logical offset in the file. When
compression is on, this means we have to do the expensive checksumming
on the uncompressed data. It would be faster if we could checksum
the compressed data instead.
* If we implement encryption, we'll be checksumming the plain text and
storing that on disk. This is significantly less secure.
* For either compression or encryption, we have to get the plain text
back before we can verify the checksum as correct. This makes the raid
layer balancing and extent moving much more expensive.
* It makes the front end caching code more complex, as we have touch
the subvolume and inodes as we cache extents.
* There is potentitally one copy of the checksum in each subvolume
referencing an extent.
The solution used here is to store the extent checksums in a dedicated
tree. This allows us to index the checksums by phyiscal extent
start and length. It means:
* The checksum is against the data stored on disk, after any compression
or encryption is done.
* The checksum is stored in a central location, and can be verified without
following back references, or reading inodes.
This makes compression significantly faster by reducing the amount of
data that needs to be checksummed. It will also allow much faster
raid management code in general.
The checksums are indexed by a key with a fixed objectid (a magic value
in ctree.h) and offset set to the starting byte of the extent. This
allows us to copy the checksum items into the fsync log tree directly (or
any other tree), without having to invent a second format for them.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-09 05:58:54 +08:00
|
|
|
if (!contig)
|
|
|
|
offset = page_offset(bvec->bv_page) + bvec->bv_offset;
|
|
|
|
|
2013-01-23 04:43:09 +08:00
|
|
|
if (offset >= ordered->file_offset + ordered->len ||
|
|
|
|
offset < ordered->file_offset) {
|
2008-07-18 18:17:13 +08:00
|
|
|
unsigned long bytes_left;
|
|
|
|
sums->len = this_sum_bytes;
|
|
|
|
this_sum_bytes = 0;
|
|
|
|
btrfs_add_ordered_sum(inode, ordered, sums);
|
|
|
|
btrfs_put_ordered_extent(ordered);
|
|
|
|
|
2013-10-12 06:44:27 +08:00
|
|
|
bytes_left = bio->bi_iter.bi_size - total_bytes;
|
2008-07-18 18:17:13 +08:00
|
|
|
|
|
|
|
sums = kzalloc(btrfs_ordered_sum_size(root, bytes_left),
|
|
|
|
GFP_NOFS);
|
2012-03-12 23:03:00 +08:00
|
|
|
BUG_ON(!sums); /* -ENOMEM */
|
2008-07-18 18:17:13 +08:00
|
|
|
sums->len = bytes_left;
|
Btrfs: move data checksumming into a dedicated tree
Btrfs stores checksums for each data block. Until now, they have
been stored in the subvolume trees, indexed by the inode that is
referencing the data block. This means that when we read the inode,
we've probably read in at least some checksums as well.
But, this has a few problems:
* The checksums are indexed by logical offset in the file. When
compression is on, this means we have to do the expensive checksumming
on the uncompressed data. It would be faster if we could checksum
the compressed data instead.
* If we implement encryption, we'll be checksumming the plain text and
storing that on disk. This is significantly less secure.
* For either compression or encryption, we have to get the plain text
back before we can verify the checksum as correct. This makes the raid
layer balancing and extent moving much more expensive.
* It makes the front end caching code more complex, as we have touch
the subvolume and inodes as we cache extents.
* There is potentitally one copy of the checksum in each subvolume
referencing an extent.
The solution used here is to store the extent checksums in a dedicated
tree. This allows us to index the checksums by phyiscal extent
start and length. It means:
* The checksum is against the data stored on disk, after any compression
or encryption is done.
* The checksum is stored in a central location, and can be verified without
following back references, or reading inodes.
This makes compression significantly faster by reducing the amount of
data that needs to be checksummed. It will also allow much faster
raid management code in general.
The checksums are indexed by a key with a fixed objectid (a magic value
in ctree.h) and offset set to the starting byte of the extent. This
allows us to copy the checksum items into the fsync log tree directly (or
any other tree), without having to invent a second format for them.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-09 05:58:54 +08:00
|
|
|
ordered = btrfs_lookup_ordered_extent(inode, offset);
|
2012-03-12 23:03:00 +08:00
|
|
|
BUG_ON(!ordered); /* Logic error */
|
2013-10-12 06:44:27 +08:00
|
|
|
sums->bytenr = ((u64)bio->bi_iter.bi_sector << 9) +
|
Btrfs: remove btrfs_sector_sum structure
Using the structure btrfs_sector_sum to keep the checksum value is
unnecessary, because the extents that btrfs_sector_sum points to are
continuous, we can find out the expected checksums by btrfs_ordered_sum's
bytenr and the offset, so we can remove btrfs_sector_sum's bytenr. After
removing bytenr, there is only one member in the structure, so it makes
no sense to keep the structure, just remove it, and use a u32 array to
store the checksum value.
By this change, we don't use the while loop to get the checksums one by
one. Now, we can get several checksum value at one time, it improved the
performance by ~74% on my SSD (31MB/s -> 54MB/s).
test command:
# dd if=/dev/zero of=/mnt/btrfs/file0 bs=1M count=1024 oflag=sync
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2013-06-19 10:36:09 +08:00
|
|
|
total_bytes;
|
|
|
|
index = 0;
|
2008-07-18 18:17:13 +08:00
|
|
|
}
|
|
|
|
|
2011-11-25 23:14:28 +08:00
|
|
|
data = kmap_atomic(bvec->bv_page);
|
Btrfs: remove btrfs_sector_sum structure
Using the structure btrfs_sector_sum to keep the checksum value is
unnecessary, because the extents that btrfs_sector_sum points to are
continuous, we can find out the expected checksums by btrfs_ordered_sum's
bytenr and the offset, so we can remove btrfs_sector_sum's bytenr. After
removing bytenr, there is only one member in the structure, so it makes
no sense to keep the structure, just remove it, and use a u32 array to
store the checksum value.
By this change, we don't use the while loop to get the checksums one by
one. Now, we can get several checksum value at one time, it improved the
performance by ~74% on my SSD (31MB/s -> 54MB/s).
test command:
# dd if=/dev/zero of=/mnt/btrfs/file0 bs=1M count=1024 oflag=sync
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2013-06-19 10:36:09 +08:00
|
|
|
sums->sums[index] = ~(u32)0;
|
|
|
|
sums->sums[index] = btrfs_csum_data(data + bvec->bv_offset,
|
|
|
|
sums->sums[index],
|
|
|
|
bvec->bv_len);
|
2011-11-25 23:14:28 +08:00
|
|
|
kunmap_atomic(data);
|
Btrfs: remove btrfs_sector_sum structure
Using the structure btrfs_sector_sum to keep the checksum value is
unnecessary, because the extents that btrfs_sector_sum points to are
continuous, we can find out the expected checksums by btrfs_ordered_sum's
bytenr and the offset, so we can remove btrfs_sector_sum's bytenr. After
removing bytenr, there is only one member in the structure, so it makes
no sense to keep the structure, just remove it, and use a u32 array to
store the checksum value.
By this change, we don't use the while loop to get the checksums one by
one. Now, we can get several checksum value at one time, it improved the
performance by ~74% on my SSD (31MB/s -> 54MB/s).
test command:
# dd if=/dev/zero of=/mnt/btrfs/file0 bs=1M count=1024 oflag=sync
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2013-06-19 10:36:09 +08:00
|
|
|
btrfs_csum_final(sums->sums[index],
|
|
|
|
(char *)(sums->sums + index));
|
2008-07-23 11:06:42 +08:00
|
|
|
|
2008-04-16 23:15:20 +08:00
|
|
|
bio_index++;
|
Btrfs: remove btrfs_sector_sum structure
Using the structure btrfs_sector_sum to keep the checksum value is
unnecessary, because the extents that btrfs_sector_sum points to are
continuous, we can find out the expected checksums by btrfs_ordered_sum's
bytenr and the offset, so we can remove btrfs_sector_sum's bytenr. After
removing bytenr, there is only one member in the structure, so it makes
no sense to keep the structure, just remove it, and use a u32 array to
store the checksum value.
By this change, we don't use the while loop to get the checksums one by
one. Now, we can get several checksum value at one time, it improved the
performance by ~74% on my SSD (31MB/s -> 54MB/s).
test command:
# dd if=/dev/zero of=/mnt/btrfs/file0 bs=1M count=1024 oflag=sync
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2013-06-19 10:36:09 +08:00
|
|
|
index++;
|
2008-07-18 18:17:13 +08:00
|
|
|
total_bytes += bvec->bv_len;
|
|
|
|
this_sum_bytes += bvec->bv_len;
|
Btrfs: move data checksumming into a dedicated tree
Btrfs stores checksums for each data block. Until now, they have
been stored in the subvolume trees, indexed by the inode that is
referencing the data block. This means that when we read the inode,
we've probably read in at least some checksums as well.
But, this has a few problems:
* The checksums are indexed by logical offset in the file. When
compression is on, this means we have to do the expensive checksumming
on the uncompressed data. It would be faster if we could checksum
the compressed data instead.
* If we implement encryption, we'll be checksumming the plain text and
storing that on disk. This is significantly less secure.
* For either compression or encryption, we have to get the plain text
back before we can verify the checksum as correct. This makes the raid
layer balancing and extent moving much more expensive.
* It makes the front end caching code more complex, as we have touch
the subvolume and inodes as we cache extents.
* There is potentitally one copy of the checksum in each subvolume
referencing an extent.
The solution used here is to store the extent checksums in a dedicated
tree. This allows us to index the checksums by phyiscal extent
start and length. It means:
* The checksum is against the data stored on disk, after any compression
or encryption is done.
* The checksum is stored in a central location, and can be verified without
following back references, or reading inodes.
This makes compression significantly faster by reducing the amount of
data that needs to be checksummed. It will also allow much faster
raid management code in general.
The checksums are indexed by a key with a fixed objectid (a magic value
in ctree.h) and offset set to the starting byte of the extent. This
allows us to copy the checksum items into the fsync log tree directly (or
any other tree), without having to invent a second format for them.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-09 05:58:54 +08:00
|
|
|
offset += bvec->bv_len;
|
2008-04-16 23:15:20 +08:00
|
|
|
bvec++;
|
|
|
|
}
|
2008-07-23 11:06:42 +08:00
|
|
|
this_sum_bytes = 0;
|
2008-07-18 18:17:13 +08:00
|
|
|
btrfs_add_ordered_sum(inode, ordered, sums);
|
|
|
|
btrfs_put_ordered_extent(ordered);
|
2008-04-16 23:15:20 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2008-12-10 22:10:46 +08:00
|
|
|
/*
|
|
|
|
* helper function for csum removal, this expects the
|
|
|
|
* key to describe the csum pointed to by the path, and it expects
|
|
|
|
* the csum to overlap the range [bytenr, len]
|
|
|
|
*
|
|
|
|
* The csum should not be entirely contained in the range and the
|
|
|
|
* range should not be entirely contained in the csum.
|
|
|
|
*
|
|
|
|
* This calls btrfs_truncate_item with the correct args based on the
|
|
|
|
* overlap, and fixes up the key as required.
|
|
|
|
*/
|
2013-04-16 13:18:22 +08:00
|
|
|
static noinline void truncate_one_csum(struct btrfs_root *root,
|
2012-03-01 21:56:26 +08:00
|
|
|
struct btrfs_path *path,
|
|
|
|
struct btrfs_key *key,
|
|
|
|
u64 bytenr, u64 len)
|
2008-12-10 22:10:46 +08:00
|
|
|
{
|
|
|
|
struct extent_buffer *leaf;
|
2011-04-13 21:41:04 +08:00
|
|
|
u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
|
2008-12-10 22:10:46 +08:00
|
|
|
u64 csum_end;
|
|
|
|
u64 end_byte = bytenr + len;
|
|
|
|
u32 blocksize_bits = root->fs_info->sb->s_blocksize_bits;
|
|
|
|
|
|
|
|
leaf = path->nodes[0];
|
|
|
|
csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
|
|
|
|
csum_end <<= root->fs_info->sb->s_blocksize_bits;
|
|
|
|
csum_end += key->offset;
|
|
|
|
|
|
|
|
if (key->offset < bytenr && csum_end <= end_byte) {
|
|
|
|
/*
|
|
|
|
* [ bytenr - len ]
|
|
|
|
* [ ]
|
|
|
|
* [csum ]
|
|
|
|
* A simple truncate off the end of the item
|
|
|
|
*/
|
|
|
|
u32 new_size = (bytenr - key->offset) >> blocksize_bits;
|
|
|
|
new_size *= csum_size;
|
2013-04-16 13:18:22 +08:00
|
|
|
btrfs_truncate_item(root, path, new_size, 1);
|
2008-12-10 22:10:46 +08:00
|
|
|
} else if (key->offset >= bytenr && csum_end > end_byte &&
|
|
|
|
end_byte > key->offset) {
|
|
|
|
/*
|
|
|
|
* [ bytenr - len ]
|
|
|
|
* [ ]
|
|
|
|
* [csum ]
|
|
|
|
* we need to truncate from the beginning of the csum
|
|
|
|
*/
|
|
|
|
u32 new_size = (csum_end - end_byte) >> blocksize_bits;
|
|
|
|
new_size *= csum_size;
|
|
|
|
|
2013-04-16 13:18:22 +08:00
|
|
|
btrfs_truncate_item(root, path, new_size, 0);
|
2008-12-10 22:10:46 +08:00
|
|
|
|
|
|
|
key->offset = end_byte;
|
2013-04-16 13:18:22 +08:00
|
|
|
btrfs_set_item_key_safe(root, path, key);
|
2008-12-10 22:10:46 +08:00
|
|
|
} else {
|
|
|
|
BUG();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* deletes the csum items from the csum tree for a given
|
|
|
|
* range of bytes.
|
|
|
|
*/
|
|
|
|
int btrfs_del_csums(struct btrfs_trans_handle *trans,
|
|
|
|
struct btrfs_root *root, u64 bytenr, u64 len)
|
|
|
|
{
|
|
|
|
struct btrfs_path *path;
|
|
|
|
struct btrfs_key key;
|
|
|
|
u64 end_byte = bytenr + len;
|
|
|
|
u64 csum_end;
|
|
|
|
struct extent_buffer *leaf;
|
|
|
|
int ret;
|
2011-04-13 21:41:04 +08:00
|
|
|
u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
|
2008-12-10 22:10:46 +08:00
|
|
|
int blocksize_bits = root->fs_info->sb->s_blocksize_bits;
|
|
|
|
|
|
|
|
root = root->fs_info->csum_root;
|
|
|
|
|
|
|
|
path = btrfs_alloc_path();
|
2011-01-26 14:22:08 +08:00
|
|
|
if (!path)
|
|
|
|
return -ENOMEM;
|
2008-12-10 22:10:46 +08:00
|
|
|
|
2009-01-06 10:25:51 +08:00
|
|
|
while (1) {
|
2008-12-10 22:10:46 +08:00
|
|
|
key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
|
|
|
|
key.offset = end_byte - 1;
|
|
|
|
key.type = BTRFS_EXTENT_CSUM_KEY;
|
|
|
|
|
2009-03-13 23:00:37 +08:00
|
|
|
path->leave_spinning = 1;
|
2008-12-10 22:10:46 +08:00
|
|
|
ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
|
|
|
|
if (ret > 0) {
|
|
|
|
if (path->slots[0] == 0)
|
2011-05-19 12:37:44 +08:00
|
|
|
break;
|
2008-12-10 22:10:46 +08:00
|
|
|
path->slots[0]--;
|
2011-01-29 02:44:44 +08:00
|
|
|
} else if (ret < 0) {
|
2011-05-19 12:37:44 +08:00
|
|
|
break;
|
2008-12-10 22:10:46 +08:00
|
|
|
}
|
2011-01-29 02:44:44 +08:00
|
|
|
|
2008-12-10 22:10:46 +08:00
|
|
|
leaf = path->nodes[0];
|
|
|
|
btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
|
|
|
|
|
|
|
|
if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
|
|
|
|
key.type != BTRFS_EXTENT_CSUM_KEY) {
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (key.offset >= end_byte)
|
|
|
|
break;
|
|
|
|
|
|
|
|
csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
|
|
|
|
csum_end <<= blocksize_bits;
|
|
|
|
csum_end += key.offset;
|
|
|
|
|
|
|
|
/* this csum ends before we start, we're done */
|
|
|
|
if (csum_end <= bytenr)
|
|
|
|
break;
|
|
|
|
|
|
|
|
/* delete the entire item, it is inside our range */
|
|
|
|
if (key.offset >= bytenr && csum_end <= end_byte) {
|
|
|
|
ret = btrfs_del_item(trans, root, path);
|
2011-05-19 12:37:44 +08:00
|
|
|
if (ret)
|
|
|
|
goto out;
|
2008-12-17 02:51:01 +08:00
|
|
|
if (key.offset == bytenr)
|
|
|
|
break;
|
2008-12-10 22:10:46 +08:00
|
|
|
} else if (key.offset < bytenr && csum_end > end_byte) {
|
|
|
|
unsigned long offset;
|
|
|
|
unsigned long shift_len;
|
|
|
|
unsigned long item_offset;
|
|
|
|
/*
|
|
|
|
* [ bytenr - len ]
|
|
|
|
* [csum ]
|
|
|
|
*
|
|
|
|
* Our bytes are in the middle of the csum,
|
|
|
|
* we need to split this item and insert a new one.
|
|
|
|
*
|
|
|
|
* But we can't drop the path because the
|
|
|
|
* csum could change, get removed, extended etc.
|
|
|
|
*
|
|
|
|
* The trick here is the max size of a csum item leaves
|
|
|
|
* enough room in the tree block for a single
|
|
|
|
* item header. So, we split the item in place,
|
|
|
|
* adding a new header pointing to the existing
|
|
|
|
* bytes. Then we loop around again and we have
|
|
|
|
* a nicely formed csum item that we can neatly
|
|
|
|
* truncate.
|
|
|
|
*/
|
|
|
|
offset = (bytenr - key.offset) >> blocksize_bits;
|
|
|
|
offset *= csum_size;
|
|
|
|
|
|
|
|
shift_len = (len >> blocksize_bits) * csum_size;
|
|
|
|
|
|
|
|
item_offset = btrfs_item_ptr_offset(leaf,
|
|
|
|
path->slots[0]);
|
|
|
|
|
|
|
|
memset_extent_buffer(leaf, 0, item_offset + offset,
|
|
|
|
shift_len);
|
|
|
|
key.offset = bytenr;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* btrfs_split_item returns -EAGAIN when the
|
|
|
|
* item changed size or key
|
|
|
|
*/
|
|
|
|
ret = btrfs_split_item(trans, root, path, &key, offset);
|
2012-03-12 23:03:00 +08:00
|
|
|
if (ret && ret != -EAGAIN) {
|
|
|
|
btrfs_abort_transaction(trans, root, ret);
|
|
|
|
goto out;
|
|
|
|
}
|
2008-12-10 22:10:46 +08:00
|
|
|
|
|
|
|
key.offset = end_byte - 1;
|
|
|
|
} else {
|
2013-04-16 13:18:22 +08:00
|
|
|
truncate_one_csum(root, path, &key, bytenr, len);
|
2008-12-17 02:51:01 +08:00
|
|
|
if (key.offset < bytenr)
|
|
|
|
break;
|
2008-12-10 22:10:46 +08:00
|
|
|
}
|
2011-04-21 07:20:15 +08:00
|
|
|
btrfs_release_path(path);
|
2008-12-10 22:10:46 +08:00
|
|
|
}
|
2011-05-19 12:37:44 +08:00
|
|
|
ret = 0;
|
2008-12-10 22:10:46 +08:00
|
|
|
out:
|
|
|
|
btrfs_free_path(path);
|
2011-05-19 12:37:44 +08:00
|
|
|
return ret;
|
2008-12-10 22:10:46 +08:00
|
|
|
}
|
|
|
|
|
2008-02-21 01:07:25 +08:00
|
|
|
int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
|
Btrfs: move data checksumming into a dedicated tree
Btrfs stores checksums for each data block. Until now, they have
been stored in the subvolume trees, indexed by the inode that is
referencing the data block. This means that when we read the inode,
we've probably read in at least some checksums as well.
But, this has a few problems:
* The checksums are indexed by logical offset in the file. When
compression is on, this means we have to do the expensive checksumming
on the uncompressed data. It would be faster if we could checksum
the compressed data instead.
* If we implement encryption, we'll be checksumming the plain text and
storing that on disk. This is significantly less secure.
* For either compression or encryption, we have to get the plain text
back before we can verify the checksum as correct. This makes the raid
layer balancing and extent moving much more expensive.
* It makes the front end caching code more complex, as we have touch
the subvolume and inodes as we cache extents.
* There is potentitally one copy of the checksum in each subvolume
referencing an extent.
The solution used here is to store the extent checksums in a dedicated
tree. This allows us to index the checksums by phyiscal extent
start and length. It means:
* The checksum is against the data stored on disk, after any compression
or encryption is done.
* The checksum is stored in a central location, and can be verified without
following back references, or reading inodes.
This makes compression significantly faster by reducing the amount of
data that needs to be checksummed. It will also allow much faster
raid management code in general.
The checksums are indexed by a key with a fixed objectid (a magic value
in ctree.h) and offset set to the starting byte of the extent. This
allows us to copy the checksum items into the fsync log tree directly (or
any other tree), without having to invent a second format for them.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-09 05:58:54 +08:00
|
|
|
struct btrfs_root *root,
|
2008-07-18 00:53:50 +08:00
|
|
|
struct btrfs_ordered_sum *sums)
|
2007-03-30 03:15:27 +08:00
|
|
|
{
|
|
|
|
struct btrfs_key file_key;
|
2007-04-16 21:22:45 +08:00
|
|
|
struct btrfs_key found_key;
|
2007-04-02 23:20:42 +08:00
|
|
|
struct btrfs_path *path;
|
2007-03-30 03:15:27 +08:00
|
|
|
struct btrfs_csum_item *item;
|
2008-02-21 01:07:25 +08:00
|
|
|
struct btrfs_csum_item *item_end;
|
2007-10-16 04:22:25 +08:00
|
|
|
struct extent_buffer *leaf = NULL;
|
Btrfs: remove btrfs_sector_sum structure
Using the structure btrfs_sector_sum to keep the checksum value is
unnecessary, because the extents that btrfs_sector_sum points to are
continuous, we can find out the expected checksums by btrfs_ordered_sum's
bytenr and the offset, so we can remove btrfs_sector_sum's bytenr. After
removing bytenr, there is only one member in the structure, so it makes
no sense to keep the structure, just remove it, and use a u32 array to
store the checksum value.
By this change, we don't use the while loop to get the checksums one by
one. Now, we can get several checksum value at one time, it improved the
performance by ~74% on my SSD (31MB/s -> 54MB/s).
test command:
# dd if=/dev/zero of=/mnt/btrfs/file0 bs=1M count=1024 oflag=sync
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2013-06-19 10:36:09 +08:00
|
|
|
u64 next_offset;
|
|
|
|
u64 total_bytes = 0;
|
2007-04-16 21:22:45 +08:00
|
|
|
u64 csum_offset;
|
Btrfs: remove btrfs_sector_sum structure
Using the structure btrfs_sector_sum to keep the checksum value is
unnecessary, because the extents that btrfs_sector_sum points to are
continuous, we can find out the expected checksums by btrfs_ordered_sum's
bytenr and the offset, so we can remove btrfs_sector_sum's bytenr. After
removing bytenr, there is only one member in the structure, so it makes
no sense to keep the structure, just remove it, and use a u32 array to
store the checksum value.
By this change, we don't use the while loop to get the checksums one by
one. Now, we can get several checksum value at one time, it improved the
performance by ~74% on my SSD (31MB/s -> 54MB/s).
test command:
# dd if=/dev/zero of=/mnt/btrfs/file0 bs=1M count=1024 oflag=sync
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2013-06-19 10:36:09 +08:00
|
|
|
u64 bytenr;
|
2007-10-26 03:42:56 +08:00
|
|
|
u32 nritems;
|
|
|
|
u32 ins_size;
|
Btrfs: remove btrfs_sector_sum structure
Using the structure btrfs_sector_sum to keep the checksum value is
unnecessary, because the extents that btrfs_sector_sum points to are
continuous, we can find out the expected checksums by btrfs_ordered_sum's
bytenr and the offset, so we can remove btrfs_sector_sum's bytenr. After
removing bytenr, there is only one member in the structure, so it makes
no sense to keep the structure, just remove it, and use a u32 array to
store the checksum value.
By this change, we don't use the while loop to get the checksums one by
one. Now, we can get several checksum value at one time, it improved the
performance by ~74% on my SSD (31MB/s -> 54MB/s).
test command:
# dd if=/dev/zero of=/mnt/btrfs/file0 bs=1M count=1024 oflag=sync
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2013-06-19 10:36:09 +08:00
|
|
|
int index = 0;
|
|
|
|
int found_next;
|
|
|
|
int ret;
|
2011-04-13 21:41:04 +08:00
|
|
|
u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
|
2008-02-21 01:07:25 +08:00
|
|
|
|
2007-04-02 23:20:42 +08:00
|
|
|
path = btrfs_alloc_path();
|
btrfs: don't BUG_ON btrfs_alloc_path() errors
This patch fixes many callers of btrfs_alloc_path() which BUG_ON allocation
failure. All the sites that are fixed in this patch were checked by me to
be fairly trivial to fix because of at least one of two criteria:
- Callers of the function catch errors from it already so bubbling the
error up will be handled.
- Callers of the function might BUG_ON any nonzero return code in which
case there is no behavior changed (but we still got to remove a BUG_ON)
The following functions were updated:
btrfs_lookup_extent, alloc_reserved_tree_block, btrfs_remove_block_group,
btrfs_lookup_csums_range, btrfs_csum_file_blocks, btrfs_mark_extent_written,
btrfs_inode_by_name, btrfs_new_inode, btrfs_symlink,
insert_reserved_file_extent, and run_delalloc_nocow
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
2011-07-14 01:38:47 +08:00
|
|
|
if (!path)
|
|
|
|
return -ENOMEM;
|
2008-02-21 01:07:25 +08:00
|
|
|
again:
|
|
|
|
next_offset = (u64)-1;
|
|
|
|
found_next = 0;
|
Btrfs: remove btrfs_sector_sum structure
Using the structure btrfs_sector_sum to keep the checksum value is
unnecessary, because the extents that btrfs_sector_sum points to are
continuous, we can find out the expected checksums by btrfs_ordered_sum's
bytenr and the offset, so we can remove btrfs_sector_sum's bytenr. After
removing bytenr, there is only one member in the structure, so it makes
no sense to keep the structure, just remove it, and use a u32 array to
store the checksum value.
By this change, we don't use the while loop to get the checksums one by
one. Now, we can get several checksum value at one time, it improved the
performance by ~74% on my SSD (31MB/s -> 54MB/s).
test command:
# dd if=/dev/zero of=/mnt/btrfs/file0 bs=1M count=1024 oflag=sync
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2013-06-19 10:36:09 +08:00
|
|
|
bytenr = sums->bytenr + total_bytes;
|
Btrfs: move data checksumming into a dedicated tree
Btrfs stores checksums for each data block. Until now, they have
been stored in the subvolume trees, indexed by the inode that is
referencing the data block. This means that when we read the inode,
we've probably read in at least some checksums as well.
But, this has a few problems:
* The checksums are indexed by logical offset in the file. When
compression is on, this means we have to do the expensive checksumming
on the uncompressed data. It would be faster if we could checksum
the compressed data instead.
* If we implement encryption, we'll be checksumming the plain text and
storing that on disk. This is significantly less secure.
* For either compression or encryption, we have to get the plain text
back before we can verify the checksum as correct. This makes the raid
layer balancing and extent moving much more expensive.
* It makes the front end caching code more complex, as we have touch
the subvolume and inodes as we cache extents.
* There is potentitally one copy of the checksum in each subvolume
referencing an extent.
The solution used here is to store the extent checksums in a dedicated
tree. This allows us to index the checksums by phyiscal extent
start and length. It means:
* The checksum is against the data stored on disk, after any compression
or encryption is done.
* The checksum is stored in a central location, and can be verified without
following back references, or reading inodes.
This makes compression significantly faster by reducing the amount of
data that needs to be checksummed. It will also allow much faster
raid management code in general.
The checksums are indexed by a key with a fixed objectid (a magic value
in ctree.h) and offset set to the starting byte of the extent. This
allows us to copy the checksum items into the fsync log tree directly (or
any other tree), without having to invent a second format for them.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-09 05:58:54 +08:00
|
|
|
file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
|
Btrfs: remove btrfs_sector_sum structure
Using the structure btrfs_sector_sum to keep the checksum value is
unnecessary, because the extents that btrfs_sector_sum points to are
continuous, we can find out the expected checksums by btrfs_ordered_sum's
bytenr and the offset, so we can remove btrfs_sector_sum's bytenr. After
removing bytenr, there is only one member in the structure, so it makes
no sense to keep the structure, just remove it, and use a u32 array to
store the checksum value.
By this change, we don't use the while loop to get the checksums one by
one. Now, we can get several checksum value at one time, it improved the
performance by ~74% on my SSD (31MB/s -> 54MB/s).
test command:
# dd if=/dev/zero of=/mnt/btrfs/file0 bs=1M count=1024 oflag=sync
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2013-06-19 10:36:09 +08:00
|
|
|
file_key.offset = bytenr;
|
Btrfs: move data checksumming into a dedicated tree
Btrfs stores checksums for each data block. Until now, they have
been stored in the subvolume trees, indexed by the inode that is
referencing the data block. This means that when we read the inode,
we've probably read in at least some checksums as well.
But, this has a few problems:
* The checksums are indexed by logical offset in the file. When
compression is on, this means we have to do the expensive checksumming
on the uncompressed data. It would be faster if we could checksum
the compressed data instead.
* If we implement encryption, we'll be checksumming the plain text and
storing that on disk. This is significantly less secure.
* For either compression or encryption, we have to get the plain text
back before we can verify the checksum as correct. This makes the raid
layer balancing and extent moving much more expensive.
* It makes the front end caching code more complex, as we have touch
the subvolume and inodes as we cache extents.
* There is potentitally one copy of the checksum in each subvolume
referencing an extent.
The solution used here is to store the extent checksums in a dedicated
tree. This allows us to index the checksums by phyiscal extent
start and length. It means:
* The checksum is against the data stored on disk, after any compression
or encryption is done.
* The checksum is stored in a central location, and can be verified without
following back references, or reading inodes.
This makes compression significantly faster by reducing the amount of
data that needs to be checksummed. It will also allow much faster
raid management code in general.
The checksums are indexed by a key with a fixed objectid (a magic value
in ctree.h) and offset set to the starting byte of the extent. This
allows us to copy the checksum items into the fsync log tree directly (or
any other tree), without having to invent a second format for them.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-09 05:58:54 +08:00
|
|
|
btrfs_set_key_type(&file_key, BTRFS_EXTENT_CSUM_KEY);
|
2007-04-19 04:15:28 +08:00
|
|
|
|
Btrfs: remove btrfs_sector_sum structure
Using the structure btrfs_sector_sum to keep the checksum value is
unnecessary, because the extents that btrfs_sector_sum points to are
continuous, we can find out the expected checksums by btrfs_ordered_sum's
bytenr and the offset, so we can remove btrfs_sector_sum's bytenr. After
removing bytenr, there is only one member in the structure, so it makes
no sense to keep the structure, just remove it, and use a u32 array to
store the checksum value.
By this change, we don't use the while loop to get the checksums one by
one. Now, we can get several checksum value at one time, it improved the
performance by ~74% on my SSD (31MB/s -> 54MB/s).
test command:
# dd if=/dev/zero of=/mnt/btrfs/file0 bs=1M count=1024 oflag=sync
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2013-06-19 10:36:09 +08:00
|
|
|
item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
|
2007-10-16 04:22:25 +08:00
|
|
|
if (!IS_ERR(item)) {
|
2008-08-28 18:15:25 +08:00
|
|
|
ret = 0;
|
Btrfs: remove btrfs_sector_sum structure
Using the structure btrfs_sector_sum to keep the checksum value is
unnecessary, because the extents that btrfs_sector_sum points to are
continuous, we can find out the expected checksums by btrfs_ordered_sum's
bytenr and the offset, so we can remove btrfs_sector_sum's bytenr. After
removing bytenr, there is only one member in the structure, so it makes
no sense to keep the structure, just remove it, and use a u32 array to
store the checksum value.
By this change, we don't use the while loop to get the checksums one by
one. Now, we can get several checksum value at one time, it improved the
performance by ~74% on my SSD (31MB/s -> 54MB/s).
test command:
# dd if=/dev/zero of=/mnt/btrfs/file0 bs=1M count=1024 oflag=sync
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2013-06-19 10:36:09 +08:00
|
|
|
leaf = path->nodes[0];
|
|
|
|
item_end = btrfs_item_ptr(leaf, path->slots[0],
|
|
|
|
struct btrfs_csum_item);
|
|
|
|
item_end = (struct btrfs_csum_item *)((char *)item_end +
|
|
|
|
btrfs_item_size_nr(leaf, path->slots[0]));
|
2007-04-19 04:15:28 +08:00
|
|
|
goto found;
|
2007-10-16 04:22:25 +08:00
|
|
|
}
|
2007-04-19 04:15:28 +08:00
|
|
|
ret = PTR_ERR(item);
|
2010-05-16 22:49:59 +08:00
|
|
|
if (ret != -EFBIG && ret != -ENOENT)
|
|
|
|
goto fail_unlock;
|
|
|
|
|
2007-04-19 04:15:28 +08:00
|
|
|
if (ret == -EFBIG) {
|
|
|
|
u32 item_size;
|
|
|
|
/* we found one, but it isn't big enough yet */
|
2007-10-16 04:14:19 +08:00
|
|
|
leaf = path->nodes[0];
|
|
|
|
item_size = btrfs_item_size_nr(leaf, path->slots[0]);
|
2008-12-02 20:17:45 +08:00
|
|
|
if ((item_size / csum_size) >=
|
|
|
|
MAX_CSUM_ITEMS(root, csum_size)) {
|
2007-04-19 04:15:28 +08:00
|
|
|
/* already at max size, make a new one */
|
|
|
|
goto insert;
|
|
|
|
}
|
|
|
|
} else {
|
2007-10-26 03:42:56 +08:00
|
|
|
int slot = path->slots[0] + 1;
|
2007-04-19 04:15:28 +08:00
|
|
|
/* we didn't find a csum item, insert one */
|
2007-10-26 03:42:56 +08:00
|
|
|
nritems = btrfs_header_nritems(path->nodes[0]);
|
2014-04-09 21:38:34 +08:00
|
|
|
if (!nritems || (path->slots[0] >= nritems - 1)) {
|
2007-10-26 03:42:56 +08:00
|
|
|
ret = btrfs_next_leaf(root, path);
|
2007-10-30 00:01:05 +08:00
|
|
|
if (ret == 1)
|
2007-10-26 03:42:56 +08:00
|
|
|
found_next = 1;
|
2007-10-30 00:01:05 +08:00
|
|
|
if (ret != 0)
|
2007-10-26 03:42:56 +08:00
|
|
|
goto insert;
|
Btrfs: fix csum tree corruption, duplicate and outdated checksums
Under rare circumstances we can end up leaving 2 versions of a checksum
for the same file extent range.
The reason for this is that after calling btrfs_next_leaf we process
slot 0 of the leaf it returns, instead of processing the slot set in
path->slots[0]. Most of the time (by far) path->slots[0] is 0, but after
btrfs_next_leaf() releases the path and before it searches for the next
leaf, another task might cause a split of the next leaf, which migrates
some of its keys to the leaf we were processing before calling
btrfs_next_leaf(). In this case btrfs_next_leaf() returns again the
same leaf but with path->slots[0] having a slot number corresponding
to the first new key it got, that is, a slot number that didn't exist
before calling btrfs_next_leaf(), as the leaf now has more keys than
it had before. So we must really process the returned leaf starting at
path->slots[0] always, as it isn't always 0, and the key at slot 0 can
have an offset much lower than our search offset/bytenr.
For example, consider the following scenario, where we have:
sums->bytenr: 40157184, sums->len: 16384, sums end: 40173568
four 4kb file data blocks with offsets 40157184, 40161280, 40165376, 40169472
Leaf N:
slot = 0 slot = btrfs_header_nritems() - 1
|-------------------------------------------------------------------|
| [(CSUM CSUM 39239680), size 8] ... [(CSUM CSUM 40116224), size 4] |
|-------------------------------------------------------------------|
Leaf N + 1:
slot = 0 slot = btrfs_header_nritems() - 1
|--------------------------------------------------------------------|
| [(CSUM CSUM 40161280), size 32] ... [((CSUM CSUM 40615936), size 8 |
|--------------------------------------------------------------------|
Because we are at the last slot of leaf N, we call btrfs_next_leaf() to
find the next highest key, which releases the current path and then searches
for that next key. However after releasing the path and before finding that
next key, the item at slot 0 of leaf N + 1 gets moved to leaf N, due to a call
to ctree.c:push_leaf_left() (via ctree.c:split_leaf()), and therefore
btrfs_next_leaf() will returns us a path again with leaf N but with the slot
pointing to its new last key (CSUM CSUM 40161280). This new version of leaf N
is then:
slot = 0 slot = btrfs_header_nritems() - 2 slot = btrfs_header_nritems() - 1
|----------------------------------------------------------------------------------------------------|
| [(CSUM CSUM 39239680), size 8] ... [(CSUM CSUM 40116224), size 4] [(CSUM CSUM 40161280), size 32] |
|----------------------------------------------------------------------------------------------------|
And incorrecly using slot 0, makes us set next_offset to 39239680 and we jump
into the "insert:" label, which will set tmp to:
tmp = min((sums->len - total_bytes) >> blocksize_bits,
(next_offset - file_key.offset) >> blocksize_bits) =
min((16384 - 0) >> 12, (39239680 - 40157184) >> 12) =
min(4, (u64)-917504 = 18446744073708634112 >> 12) = 4
and
ins_size = csum_size * tmp = 4 * 4 = 16 bytes.
In other words, we insert a new csum item in the tree with key
(CSUM_OBJECTID CSUM_KEY 40157184 = sums->bytenr) that contains the checksums
for all the data (4 blocks of 4096 bytes each = sums->len). Which is wrong,
because the item with key (CSUM CSUM 40161280) (the one that was moved from
leaf N + 1 to the end of leaf N) contains the old checksums of the last 12288
bytes of our data and won't get those old checksums removed.
So this leaves us 2 different checksums for 3 4kb blocks of data in the tree,
and breaks the logical rule:
Key_N+1.offset >= Key_N.offset + length_of_data_its_checksums_cover
An obvious bad effect of this is that a subsequent csum tree lookup to get
the checksum of any of the blocks with logical offset of 40161280, 40165376
or 40169472 (the last 3 4kb blocks of file data), will get the old checksums.
Cc: stable@vger.kernel.org
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
2014-08-10 04:22:27 +08:00
|
|
|
slot = path->slots[0];
|
2007-10-26 03:42:56 +08:00
|
|
|
}
|
|
|
|
btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
|
Btrfs: move data checksumming into a dedicated tree
Btrfs stores checksums for each data block. Until now, they have
been stored in the subvolume trees, indexed by the inode that is
referencing the data block. This means that when we read the inode,
we've probably read in at least some checksums as well.
But, this has a few problems:
* The checksums are indexed by logical offset in the file. When
compression is on, this means we have to do the expensive checksumming
on the uncompressed data. It would be faster if we could checksum
the compressed data instead.
* If we implement encryption, we'll be checksumming the plain text and
storing that on disk. This is significantly less secure.
* For either compression or encryption, we have to get the plain text
back before we can verify the checksum as correct. This makes the raid
layer balancing and extent moving much more expensive.
* It makes the front end caching code more complex, as we have touch
the subvolume and inodes as we cache extents.
* There is potentitally one copy of the checksum in each subvolume
referencing an extent.
The solution used here is to store the extent checksums in a dedicated
tree. This allows us to index the checksums by phyiscal extent
start and length. It means:
* The checksum is against the data stored on disk, after any compression
or encryption is done.
* The checksum is stored in a central location, and can be verified without
following back references, or reading inodes.
This makes compression significantly faster by reducing the amount of
data that needs to be checksummed. It will also allow much faster
raid management code in general.
The checksums are indexed by a key with a fixed objectid (a magic value
in ctree.h) and offset set to the starting byte of the extent. This
allows us to copy the checksum items into the fsync log tree directly (or
any other tree), without having to invent a second format for them.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-09 05:58:54 +08:00
|
|
|
if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
|
|
|
|
found_key.type != BTRFS_EXTENT_CSUM_KEY) {
|
2007-10-26 03:42:56 +08:00
|
|
|
found_next = 1;
|
|
|
|
goto insert;
|
|
|
|
}
|
|
|
|
next_offset = found_key.offset;
|
|
|
|
found_next = 1;
|
2007-04-19 04:15:28 +08:00
|
|
|
goto insert;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* at this point, we know the tree has an item, but it isn't big
|
|
|
|
* enough yet to put our csum in. Grow it
|
|
|
|
*/
|
2011-04-21 07:20:15 +08:00
|
|
|
btrfs_release_path(path);
|
2007-04-16 21:22:45 +08:00
|
|
|
ret = btrfs_search_slot(trans, root, &file_key, path,
|
2008-12-02 20:17:45 +08:00
|
|
|
csum_size, 1);
|
2007-04-16 21:22:45 +08:00
|
|
|
if (ret < 0)
|
2008-08-16 03:34:18 +08:00
|
|
|
goto fail_unlock;
|
2008-12-10 22:10:46 +08:00
|
|
|
|
|
|
|
if (ret > 0) {
|
|
|
|
if (path->slots[0] == 0)
|
|
|
|
goto insert;
|
|
|
|
path->slots[0]--;
|
2007-04-16 21:22:45 +08:00
|
|
|
}
|
2008-12-10 22:10:46 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
leaf = path->nodes[0];
|
|
|
|
btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
|
Btrfs: move data checksumming into a dedicated tree
Btrfs stores checksums for each data block. Until now, they have
been stored in the subvolume trees, indexed by the inode that is
referencing the data block. This means that when we read the inode,
we've probably read in at least some checksums as well.
But, this has a few problems:
* The checksums are indexed by logical offset in the file. When
compression is on, this means we have to do the expensive checksumming
on the uncompressed data. It would be faster if we could checksum
the compressed data instead.
* If we implement encryption, we'll be checksumming the plain text and
storing that on disk. This is significantly less secure.
* For either compression or encryption, we have to get the plain text
back before we can verify the checksum as correct. This makes the raid
layer balancing and extent moving much more expensive.
* It makes the front end caching code more complex, as we have touch
the subvolume and inodes as we cache extents.
* There is potentitally one copy of the checksum in each subvolume
referencing an extent.
The solution used here is to store the extent checksums in a dedicated
tree. This allows us to index the checksums by phyiscal extent
start and length. It means:
* The checksum is against the data stored on disk, after any compression
or encryption is done.
* The checksum is stored in a central location, and can be verified without
following back references, or reading inodes.
This makes compression significantly faster by reducing the amount of
data that needs to be checksummed. It will also allow much faster
raid management code in general.
The checksums are indexed by a key with a fixed objectid (a magic value
in ctree.h) and offset set to the starting byte of the extent. This
allows us to copy the checksum items into the fsync log tree directly (or
any other tree), without having to invent a second format for them.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-09 05:58:54 +08:00
|
|
|
csum_offset = (bytenr - found_key.offset) >>
|
2007-04-16 21:22:45 +08:00
|
|
|
root->fs_info->sb->s_blocksize_bits;
|
2008-12-10 22:10:46 +08:00
|
|
|
|
Btrfs: move data checksumming into a dedicated tree
Btrfs stores checksums for each data block. Until now, they have
been stored in the subvolume trees, indexed by the inode that is
referencing the data block. This means that when we read the inode,
we've probably read in at least some checksums as well.
But, this has a few problems:
* The checksums are indexed by logical offset in the file. When
compression is on, this means we have to do the expensive checksumming
on the uncompressed data. It would be faster if we could checksum
the compressed data instead.
* If we implement encryption, we'll be checksumming the plain text and
storing that on disk. This is significantly less secure.
* For either compression or encryption, we have to get the plain text
back before we can verify the checksum as correct. This makes the raid
layer balancing and extent moving much more expensive.
* It makes the front end caching code more complex, as we have touch
the subvolume and inodes as we cache extents.
* There is potentitally one copy of the checksum in each subvolume
referencing an extent.
The solution used here is to store the extent checksums in a dedicated
tree. This allows us to index the checksums by phyiscal extent
start and length. It means:
* The checksum is against the data stored on disk, after any compression
or encryption is done.
* The checksum is stored in a central location, and can be verified without
following back references, or reading inodes.
This makes compression significantly faster by reducing the amount of
data that needs to be checksummed. It will also allow much faster
raid management code in general.
The checksums are indexed by a key with a fixed objectid (a magic value
in ctree.h) and offset set to the starting byte of the extent. This
allows us to copy the checksum items into the fsync log tree directly (or
any other tree), without having to invent a second format for them.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-09 05:58:54 +08:00
|
|
|
if (btrfs_key_type(&found_key) != BTRFS_EXTENT_CSUM_KEY ||
|
|
|
|
found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
|
2008-12-02 20:17:45 +08:00
|
|
|
csum_offset >= MAX_CSUM_ITEMS(root, csum_size)) {
|
2007-04-16 21:22:45 +08:00
|
|
|
goto insert;
|
|
|
|
}
|
2008-12-10 22:10:46 +08:00
|
|
|
|
Btrfs: extend the checksum item as much as possible
For write, we also reserve some space for COW blocks during updating
the checksum tree, and we calculate the number of blocks by checking
if the number of bytes outstanding that are going to need csums needs
one more block for csum.
When we add these checksum into the checksum tree, we use ordered sums
list.
Every ordered sum contains csums for each sector, and we'll first try
to look up an existing csum item,
a) if we don't yet have a proper csum item, then we need to insert one,
b) or if we find one but the csum item is not big enough, then we need
to extend it.
The point is we'll unlock the whole path and then insert or extend.
So others can hack in and update the tree.
Each insert or extend needs update the tree with COW on, and we may need
to insert/extend for many times.
That means what we've reserved for updating checksum tree is NOT enough
indeed.
The case is even more serious with having several write threads at the
same time, it can end up eating our reserved space quickly and starting
eating globle reserve pool instead.
I don't yet come up with a way to calculate the worse case for updating
csum, but extending the checksum item as much as possible can be helpful
in my test.
The idea behind is that it can reduce the times we insert/extend so that
it saves us precious reserved space.
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2013-02-04 21:12:18 +08:00
|
|
|
if (csum_offset == btrfs_item_size_nr(leaf, path->slots[0]) /
|
2008-12-02 20:17:45 +08:00
|
|
|
csum_size) {
|
Btrfs: extend the checksum item as much as possible
For write, we also reserve some space for COW blocks during updating
the checksum tree, and we calculate the number of blocks by checking
if the number of bytes outstanding that are going to need csums needs
one more block for csum.
When we add these checksum into the checksum tree, we use ordered sums
list.
Every ordered sum contains csums for each sector, and we'll first try
to look up an existing csum item,
a) if we don't yet have a proper csum item, then we need to insert one,
b) or if we find one but the csum item is not big enough, then we need
to extend it.
The point is we'll unlock the whole path and then insert or extend.
So others can hack in and update the tree.
Each insert or extend needs update the tree with COW on, and we may need
to insert/extend for many times.
That means what we've reserved for updating checksum tree is NOT enough
indeed.
The case is even more serious with having several write threads at the
same time, it can end up eating our reserved space quickly and starting
eating globle reserve pool instead.
I don't yet come up with a way to calculate the worse case for updating
csum, but extending the checksum item as much as possible can be helpful
in my test.
The idea behind is that it can reduce the times we insert/extend so that
it saves us precious reserved space.
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2013-02-04 21:12:18 +08:00
|
|
|
int extend_nr;
|
|
|
|
u64 tmp;
|
|
|
|
u32 diff;
|
|
|
|
u32 free_space;
|
2008-12-10 22:10:46 +08:00
|
|
|
|
Btrfs: extend the checksum item as much as possible
For write, we also reserve some space for COW blocks during updating
the checksum tree, and we calculate the number of blocks by checking
if the number of bytes outstanding that are going to need csums needs
one more block for csum.
When we add these checksum into the checksum tree, we use ordered sums
list.
Every ordered sum contains csums for each sector, and we'll first try
to look up an existing csum item,
a) if we don't yet have a proper csum item, then we need to insert one,
b) or if we find one but the csum item is not big enough, then we need
to extend it.
The point is we'll unlock the whole path and then insert or extend.
So others can hack in and update the tree.
Each insert or extend needs update the tree with COW on, and we may need
to insert/extend for many times.
That means what we've reserved for updating checksum tree is NOT enough
indeed.
The case is even more serious with having several write threads at the
same time, it can end up eating our reserved space quickly and starting
eating globle reserve pool instead.
I don't yet come up with a way to calculate the worse case for updating
csum, but extending the checksum item as much as possible can be helpful
in my test.
The idea behind is that it can reduce the times we insert/extend so that
it saves us precious reserved space.
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2013-02-04 21:12:18 +08:00
|
|
|
if (btrfs_leaf_free_space(root, leaf) <
|
|
|
|
sizeof(struct btrfs_item) + csum_size * 2)
|
|
|
|
goto insert;
|
|
|
|
|
|
|
|
free_space = btrfs_leaf_free_space(root, leaf) -
|
|
|
|
sizeof(struct btrfs_item) - csum_size;
|
Btrfs: remove btrfs_sector_sum structure
Using the structure btrfs_sector_sum to keep the checksum value is
unnecessary, because the extents that btrfs_sector_sum points to are
continuous, we can find out the expected checksums by btrfs_ordered_sum's
bytenr and the offset, so we can remove btrfs_sector_sum's bytenr. After
removing bytenr, there is only one member in the structure, so it makes
no sense to keep the structure, just remove it, and use a u32 array to
store the checksum value.
By this change, we don't use the while loop to get the checksums one by
one. Now, we can get several checksum value at one time, it improved the
performance by ~74% on my SSD (31MB/s -> 54MB/s).
test command:
# dd if=/dev/zero of=/mnt/btrfs/file0 bs=1M count=1024 oflag=sync
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2013-06-19 10:36:09 +08:00
|
|
|
tmp = sums->len - total_bytes;
|
Btrfs: extend the checksum item as much as possible
For write, we also reserve some space for COW blocks during updating
the checksum tree, and we calculate the number of blocks by checking
if the number of bytes outstanding that are going to need csums needs
one more block for csum.
When we add these checksum into the checksum tree, we use ordered sums
list.
Every ordered sum contains csums for each sector, and we'll first try
to look up an existing csum item,
a) if we don't yet have a proper csum item, then we need to insert one,
b) or if we find one but the csum item is not big enough, then we need
to extend it.
The point is we'll unlock the whole path and then insert or extend.
So others can hack in and update the tree.
Each insert or extend needs update the tree with COW on, and we may need
to insert/extend for many times.
That means what we've reserved for updating checksum tree is NOT enough
indeed.
The case is even more serious with having several write threads at the
same time, it can end up eating our reserved space quickly and starting
eating globle reserve pool instead.
I don't yet come up with a way to calculate the worse case for updating
csum, but extending the checksum item as much as possible can be helpful
in my test.
The idea behind is that it can reduce the times we insert/extend so that
it saves us precious reserved space.
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2013-02-04 21:12:18 +08:00
|
|
|
tmp >>= root->fs_info->sb->s_blocksize_bits;
|
|
|
|
WARN_ON(tmp < 1);
|
|
|
|
|
|
|
|
extend_nr = max_t(int, 1, (int)tmp);
|
|
|
|
diff = (csum_offset + extend_nr) * csum_size;
|
|
|
|
diff = min(diff, MAX_CSUM_ITEMS(root, csum_size) * csum_size);
|
2008-12-10 22:10:46 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
|
Btrfs: extend the checksum item as much as possible
For write, we also reserve some space for COW blocks during updating
the checksum tree, and we calculate the number of blocks by checking
if the number of bytes outstanding that are going to need csums needs
one more block for csum.
When we add these checksum into the checksum tree, we use ordered sums
list.
Every ordered sum contains csums for each sector, and we'll first try
to look up an existing csum item,
a) if we don't yet have a proper csum item, then we need to insert one,
b) or if we find one but the csum item is not big enough, then we need
to extend it.
The point is we'll unlock the whole path and then insert or extend.
So others can hack in and update the tree.
Each insert or extend needs update the tree with COW on, and we may need
to insert/extend for many times.
That means what we've reserved for updating checksum tree is NOT enough
indeed.
The case is even more serious with having several write threads at the
same time, it can end up eating our reserved space quickly and starting
eating globle reserve pool instead.
I don't yet come up with a way to calculate the worse case for updating
csum, but extending the checksum item as much as possible can be helpful
in my test.
The idea behind is that it can reduce the times we insert/extend so that
it saves us precious reserved space.
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2013-02-04 21:12:18 +08:00
|
|
|
diff = min(free_space, diff);
|
|
|
|
diff /= csum_size;
|
|
|
|
diff *= csum_size;
|
2008-12-10 22:10:46 +08:00
|
|
|
|
2013-04-16 13:18:49 +08:00
|
|
|
btrfs_extend_item(root, path, diff);
|
Btrfs: remove btrfs_sector_sum structure
Using the structure btrfs_sector_sum to keep the checksum value is
unnecessary, because the extents that btrfs_sector_sum points to are
continuous, we can find out the expected checksums by btrfs_ordered_sum's
bytenr and the offset, so we can remove btrfs_sector_sum's bytenr. After
removing bytenr, there is only one member in the structure, so it makes
no sense to keep the structure, just remove it, and use a u32 array to
store the checksum value.
By this change, we don't use the while loop to get the checksums one by
one. Now, we can get several checksum value at one time, it improved the
performance by ~74% on my SSD (31MB/s -> 54MB/s).
test command:
# dd if=/dev/zero of=/mnt/btrfs/file0 bs=1M count=1024 oflag=sync
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2013-06-19 10:36:09 +08:00
|
|
|
ret = 0;
|
2007-04-16 21:22:45 +08:00
|
|
|
goto csum;
|
|
|
|
}
|
|
|
|
|
|
|
|
insert:
|
2011-04-21 07:20:15 +08:00
|
|
|
btrfs_release_path(path);
|
2007-04-16 21:22:45 +08:00
|
|
|
csum_offset = 0;
|
2007-10-26 03:42:56 +08:00
|
|
|
if (found_next) {
|
Btrfs: extend the checksum item as much as possible
For write, we also reserve some space for COW blocks during updating
the checksum tree, and we calculate the number of blocks by checking
if the number of bytes outstanding that are going to need csums needs
one more block for csum.
When we add these checksum into the checksum tree, we use ordered sums
list.
Every ordered sum contains csums for each sector, and we'll first try
to look up an existing csum item,
a) if we don't yet have a proper csum item, then we need to insert one,
b) or if we find one but the csum item is not big enough, then we need
to extend it.
The point is we'll unlock the whole path and then insert or extend.
So others can hack in and update the tree.
Each insert or extend needs update the tree with COW on, and we may need
to insert/extend for many times.
That means what we've reserved for updating checksum tree is NOT enough
indeed.
The case is even more serious with having several write threads at the
same time, it can end up eating our reserved space quickly and starting
eating globle reserve pool instead.
I don't yet come up with a way to calculate the worse case for updating
csum, but extending the checksum item as much as possible can be helpful
in my test.
The idea behind is that it can reduce the times we insert/extend so that
it saves us precious reserved space.
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2013-02-04 21:12:18 +08:00
|
|
|
u64 tmp;
|
Btrfs: move data checksumming into a dedicated tree
Btrfs stores checksums for each data block. Until now, they have
been stored in the subvolume trees, indexed by the inode that is
referencing the data block. This means that when we read the inode,
we've probably read in at least some checksums as well.
But, this has a few problems:
* The checksums are indexed by logical offset in the file. When
compression is on, this means we have to do the expensive checksumming
on the uncompressed data. It would be faster if we could checksum
the compressed data instead.
* If we implement encryption, we'll be checksumming the plain text and
storing that on disk. This is significantly less secure.
* For either compression or encryption, we have to get the plain text
back before we can verify the checksum as correct. This makes the raid
layer balancing and extent moving much more expensive.
* It makes the front end caching code more complex, as we have touch
the subvolume and inodes as we cache extents.
* There is potentitally one copy of the checksum in each subvolume
referencing an extent.
The solution used here is to store the extent checksums in a dedicated
tree. This allows us to index the checksums by phyiscal extent
start and length. It means:
* The checksum is against the data stored on disk, after any compression
or encryption is done.
* The checksum is stored in a central location, and can be verified without
following back references, or reading inodes.
This makes compression significantly faster by reducing the amount of
data that needs to be checksummed. It will also allow much faster
raid management code in general.
The checksums are indexed by a key with a fixed objectid (a magic value
in ctree.h) and offset set to the starting byte of the extent. This
allows us to copy the checksum items into the fsync log tree directly (or
any other tree), without having to invent a second format for them.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-09 05:58:54 +08:00
|
|
|
|
Btrfs: remove btrfs_sector_sum structure
Using the structure btrfs_sector_sum to keep the checksum value is
unnecessary, because the extents that btrfs_sector_sum points to are
continuous, we can find out the expected checksums by btrfs_ordered_sum's
bytenr and the offset, so we can remove btrfs_sector_sum's bytenr. After
removing bytenr, there is only one member in the structure, so it makes
no sense to keep the structure, just remove it, and use a u32 array to
store the checksum value.
By this change, we don't use the while loop to get the checksums one by
one. Now, we can get several checksum value at one time, it improved the
performance by ~74% on my SSD (31MB/s -> 54MB/s).
test command:
# dd if=/dev/zero of=/mnt/btrfs/file0 bs=1M count=1024 oflag=sync
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2013-06-19 10:36:09 +08:00
|
|
|
tmp = sums->len - total_bytes;
|
2007-10-26 03:42:56 +08:00
|
|
|
tmp >>= root->fs_info->sb->s_blocksize_bits;
|
Btrfs: extend the checksum item as much as possible
For write, we also reserve some space for COW blocks during updating
the checksum tree, and we calculate the number of blocks by checking
if the number of bytes outstanding that are going to need csums needs
one more block for csum.
When we add these checksum into the checksum tree, we use ordered sums
list.
Every ordered sum contains csums for each sector, and we'll first try
to look up an existing csum item,
a) if we don't yet have a proper csum item, then we need to insert one,
b) or if we find one but the csum item is not big enough, then we need
to extend it.
The point is we'll unlock the whole path and then insert or extend.
So others can hack in and update the tree.
Each insert or extend needs update the tree with COW on, and we may need
to insert/extend for many times.
That means what we've reserved for updating checksum tree is NOT enough
indeed.
The case is even more serious with having several write threads at the
same time, it can end up eating our reserved space quickly and starting
eating globle reserve pool instead.
I don't yet come up with a way to calculate the worse case for updating
csum, but extending the checksum item as much as possible can be helpful
in my test.
The idea behind is that it can reduce the times we insert/extend so that
it saves us precious reserved space.
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2013-02-04 21:12:18 +08:00
|
|
|
tmp = min(tmp, (next_offset - file_key.offset) >>
|
|
|
|
root->fs_info->sb->s_blocksize_bits);
|
|
|
|
|
2007-10-26 03:42:56 +08:00
|
|
|
tmp = max((u64)1, tmp);
|
2008-12-02 20:17:45 +08:00
|
|
|
tmp = min(tmp, (u64)MAX_CSUM_ITEMS(root, csum_size));
|
|
|
|
ins_size = csum_size * tmp;
|
2007-10-26 03:42:56 +08:00
|
|
|
} else {
|
2008-12-02 20:17:45 +08:00
|
|
|
ins_size = csum_size;
|
2007-10-26 03:42:56 +08:00
|
|
|
}
|
2009-03-13 23:00:37 +08:00
|
|
|
path->leave_spinning = 1;
|
2007-04-02 23:20:42 +08:00
|
|
|
ret = btrfs_insert_empty_item(trans, root, path, &file_key,
|
2007-10-26 03:42:56 +08:00
|
|
|
ins_size);
|
2009-03-13 23:00:37 +08:00
|
|
|
path->leave_spinning = 0;
|
2007-06-23 02:16:25 +08:00
|
|
|
if (ret < 0)
|
2008-08-16 03:34:18 +08:00
|
|
|
goto fail_unlock;
|
2013-10-31 13:00:08 +08:00
|
|
|
if (WARN_ON(ret != 0))
|
2008-08-16 03:34:18 +08:00
|
|
|
goto fail_unlock;
|
2007-10-16 04:14:19 +08:00
|
|
|
leaf = path->nodes[0];
|
Btrfs: remove btrfs_sector_sum structure
Using the structure btrfs_sector_sum to keep the checksum value is
unnecessary, because the extents that btrfs_sector_sum points to are
continuous, we can find out the expected checksums by btrfs_ordered_sum's
bytenr and the offset, so we can remove btrfs_sector_sum's bytenr. After
removing bytenr, there is only one member in the structure, so it makes
no sense to keep the structure, just remove it, and use a u32 array to
store the checksum value.
By this change, we don't use the while loop to get the checksums one by
one. Now, we can get several checksum value at one time, it improved the
performance by ~74% on my SSD (31MB/s -> 54MB/s).
test command:
# dd if=/dev/zero of=/mnt/btrfs/file0 bs=1M count=1024 oflag=sync
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2013-06-19 10:36:09 +08:00
|
|
|
csum:
|
2007-10-16 04:14:19 +08:00
|
|
|
item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
|
Btrfs: remove btrfs_sector_sum structure
Using the structure btrfs_sector_sum to keep the checksum value is
unnecessary, because the extents that btrfs_sector_sum points to are
continuous, we can find out the expected checksums by btrfs_ordered_sum's
bytenr and the offset, so we can remove btrfs_sector_sum's bytenr. After
removing bytenr, there is only one member in the structure, so it makes
no sense to keep the structure, just remove it, and use a u32 array to
store the checksum value.
By this change, we don't use the while loop to get the checksums one by
one. Now, we can get several checksum value at one time, it improved the
performance by ~74% on my SSD (31MB/s -> 54MB/s).
test command:
# dd if=/dev/zero of=/mnt/btrfs/file0 bs=1M count=1024 oflag=sync
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2013-06-19 10:36:09 +08:00
|
|
|
item_end = (struct btrfs_csum_item *)((unsigned char *)item +
|
|
|
|
btrfs_item_size_nr(leaf, path->slots[0]));
|
2007-05-11 00:36:17 +08:00
|
|
|
item = (struct btrfs_csum_item *)((unsigned char *)item +
|
2008-12-02 20:17:45 +08:00
|
|
|
csum_offset * csum_size);
|
2007-04-18 01:26:50 +08:00
|
|
|
found:
|
Btrfs: remove btrfs_sector_sum structure
Using the structure btrfs_sector_sum to keep the checksum value is
unnecessary, because the extents that btrfs_sector_sum points to are
continuous, we can find out the expected checksums by btrfs_ordered_sum's
bytenr and the offset, so we can remove btrfs_sector_sum's bytenr. After
removing bytenr, there is only one member in the structure, so it makes
no sense to keep the structure, just remove it, and use a u32 array to
store the checksum value.
By this change, we don't use the while loop to get the checksums one by
one. Now, we can get several checksum value at one time, it improved the
performance by ~74% on my SSD (31MB/s -> 54MB/s).
test command:
# dd if=/dev/zero of=/mnt/btrfs/file0 bs=1M count=1024 oflag=sync
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
2013-06-19 10:36:09 +08:00
|
|
|
ins_size = (u32)(sums->len - total_bytes) >>
|
|
|
|
root->fs_info->sb->s_blocksize_bits;
|
|
|
|
ins_size *= csum_size;
|
|
|
|
ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
|
|
|
|
ins_size);
|
|
|
|
write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
|
|
|
|
ins_size);
|
|
|
|
|
|
|
|
ins_size /= csum_size;
|
|
|
|
total_bytes += ins_size * root->sectorsize;
|
|
|
|
index += ins_size;
|
2011-07-20 00:04:14 +08:00
|
|
|
|
2007-04-02 23:20:42 +08:00
|
|
|
btrfs_mark_buffer_dirty(path->nodes[0]);
|
2008-07-18 00:53:50 +08:00
|
|
|
if (total_bytes < sums->len) {
|
2011-04-21 07:20:15 +08:00
|
|
|
btrfs_release_path(path);
|
2009-03-13 23:00:37 +08:00
|
|
|
cond_resched();
|
2008-02-21 01:07:25 +08:00
|
|
|
goto again;
|
|
|
|
}
|
2008-08-16 03:34:18 +08:00
|
|
|
out:
|
2007-04-02 23:20:42 +08:00
|
|
|
btrfs_free_path(path);
|
2007-03-30 03:15:27 +08:00
|
|
|
return ret;
|
2008-08-16 03:34:18 +08:00
|
|
|
|
|
|
|
fail_unlock:
|
|
|
|
goto out;
|
2007-03-30 03:15:27 +08:00
|
|
|
}
|
2014-06-09 10:48:05 +08:00
|
|
|
|
|
|
|
void btrfs_extent_item_to_extent_map(struct inode *inode,
|
|
|
|
const struct btrfs_path *path,
|
|
|
|
struct btrfs_file_extent_item *fi,
|
|
|
|
const bool new_inline,
|
|
|
|
struct extent_map *em)
|
|
|
|
{
|
|
|
|
struct btrfs_root *root = BTRFS_I(inode)->root;
|
|
|
|
struct extent_buffer *leaf = path->nodes[0];
|
|
|
|
const int slot = path->slots[0];
|
|
|
|
struct btrfs_key key;
|
|
|
|
u64 extent_start, extent_end;
|
|
|
|
u64 bytenr;
|
|
|
|
u8 type = btrfs_file_extent_type(leaf, fi);
|
|
|
|
int compress_type = btrfs_file_extent_compression(leaf, fi);
|
|
|
|
|
|
|
|
em->bdev = root->fs_info->fs_devices->latest_bdev;
|
|
|
|
btrfs_item_key_to_cpu(leaf, &key, slot);
|
|
|
|
extent_start = key.offset;
|
|
|
|
|
|
|
|
if (type == BTRFS_FILE_EXTENT_REG ||
|
|
|
|
type == BTRFS_FILE_EXTENT_PREALLOC) {
|
|
|
|
extent_end = extent_start +
|
|
|
|
btrfs_file_extent_num_bytes(leaf, fi);
|
|
|
|
} else if (type == BTRFS_FILE_EXTENT_INLINE) {
|
|
|
|
size_t size;
|
|
|
|
size = btrfs_file_extent_inline_len(leaf, slot, fi);
|
|
|
|
extent_end = ALIGN(extent_start + size, root->sectorsize);
|
|
|
|
}
|
|
|
|
|
|
|
|
em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
|
|
|
|
if (type == BTRFS_FILE_EXTENT_REG ||
|
|
|
|
type == BTRFS_FILE_EXTENT_PREALLOC) {
|
|
|
|
em->start = extent_start;
|
|
|
|
em->len = extent_end - extent_start;
|
|
|
|
em->orig_start = extent_start -
|
|
|
|
btrfs_file_extent_offset(leaf, fi);
|
|
|
|
em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
|
|
|
|
bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
|
|
|
|
if (bytenr == 0) {
|
|
|
|
em->block_start = EXTENT_MAP_HOLE;
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
if (compress_type != BTRFS_COMPRESS_NONE) {
|
|
|
|
set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
|
|
|
|
em->compress_type = compress_type;
|
|
|
|
em->block_start = bytenr;
|
|
|
|
em->block_len = em->orig_block_len;
|
|
|
|
} else {
|
|
|
|
bytenr += btrfs_file_extent_offset(leaf, fi);
|
|
|
|
em->block_start = bytenr;
|
|
|
|
em->block_len = em->len;
|
|
|
|
if (type == BTRFS_FILE_EXTENT_PREALLOC)
|
|
|
|
set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
|
|
|
|
}
|
|
|
|
} else if (type == BTRFS_FILE_EXTENT_INLINE) {
|
|
|
|
em->block_start = EXTENT_MAP_INLINE;
|
|
|
|
em->start = extent_start;
|
|
|
|
em->len = extent_end - extent_start;
|
|
|
|
/*
|
|
|
|
* Initialize orig_start and block_len with the same values
|
|
|
|
* as in inode.c:btrfs_get_extent().
|
|
|
|
*/
|
|
|
|
em->orig_start = EXTENT_MAP_HOLE;
|
|
|
|
em->block_len = (u64)-1;
|
|
|
|
if (!new_inline && compress_type != BTRFS_COMPRESS_NONE) {
|
|
|
|
set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
|
|
|
|
em->compress_type = compress_type;
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
btrfs_err(root->fs_info,
|
|
|
|
"unknown file extent item type %d, inode %llu, offset %llu, root %llu",
|
|
|
|
type, btrfs_ino(inode), extent_start,
|
|
|
|
root->root_key.objectid);
|
|
|
|
}
|
|
|
|
}
|