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.
|
|
|
|
*/
|
|
|
|
|
2008-06-26 04:01:31 +08:00
|
|
|
#include <linux/version.h>
|
2007-03-23 00:13:20 +08:00
|
|
|
#include <linux/fs.h>
|
2007-03-29 01:57:48 +08:00
|
|
|
#include <linux/blkdev.h>
|
2007-03-29 07:44:27 +08:00
|
|
|
#include <linux/scatterlist.h>
|
2007-03-30 20:47:31 +08:00
|
|
|
#include <linux/swap.h>
|
2007-04-09 22:42:37 +08:00
|
|
|
#include <linux/radix-tree.h>
|
2007-05-03 03:53:43 +08:00
|
|
|
#include <linux/writeback.h>
|
2007-10-16 04:14:19 +08:00
|
|
|
#include <linux/buffer_head.h> // for block_sync_page
|
2008-04-10 04:28:12 +08:00
|
|
|
#include <linux/workqueue.h>
|
2008-06-26 04:01:31 +08:00
|
|
|
#include <linux/kthread.h>
|
|
|
|
# include <linux/freezer.h>
|
2008-04-12 03:45:51 +08:00
|
|
|
#include "crc32c.h"
|
2007-02-02 22:18:22 +08:00
|
|
|
#include "ctree.h"
|
|
|
|
#include "disk-io.h"
|
2007-03-17 04:20:31 +08:00
|
|
|
#include "transaction.h"
|
2007-04-09 22:42:37 +08:00
|
|
|
#include "btrfs_inode.h"
|
2008-03-25 03:01:56 +08:00
|
|
|
#include "volumes.h"
|
2007-10-16 04:15:53 +08:00
|
|
|
#include "print-tree.h"
|
2008-06-12 04:50:36 +08:00
|
|
|
#include "async-thread.h"
|
2008-06-26 04:01:30 +08:00
|
|
|
#include "locking.h"
|
2008-07-29 03:32:51 +08:00
|
|
|
#include "ref-cache.h"
|
2008-09-06 04:13:11 +08:00
|
|
|
#include "tree-log.h"
|
2007-02-02 22:18:22 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
#if 0
|
|
|
|
static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
|
2007-04-12 03:53:25 +08:00
|
|
|
{
|
2007-10-16 04:14:19 +08:00
|
|
|
if (extent_buffer_blocknr(buf) != btrfs_header_blocknr(buf)) {
|
|
|
|
printk(KERN_CRIT "buf blocknr(buf) is %llu, header is %llu\n",
|
|
|
|
(unsigned long long)extent_buffer_blocknr(buf),
|
|
|
|
(unsigned long long)btrfs_header_blocknr(buf));
|
2007-06-12 18:35:45 +08:00
|
|
|
return 1;
|
2007-03-29 01:57:48 +08:00
|
|
|
}
|
2007-02-23 21:38:36 +08:00
|
|
|
return 0;
|
2007-02-02 22:18:22 +08:00
|
|
|
}
|
2007-10-16 04:14:19 +08:00
|
|
|
#endif
|
2007-02-02 22:18:22 +08:00
|
|
|
|
2008-01-25 05:13:08 +08:00
|
|
|
static struct extent_io_ops btree_extent_io_ops;
|
2008-06-12 04:50:36 +08:00
|
|
|
static void end_workqueue_fn(struct btrfs_work *work);
|
2008-04-10 04:28:12 +08:00
|
|
|
|
2008-09-30 03:18:18 +08:00
|
|
|
/*
|
|
|
|
* end_io_wq structs are used to do processing in task context when an IO is
|
|
|
|
* complete. This is used during reads to verify checksums, and it is used
|
|
|
|
* by writes to insert metadata for new file extents after IO is complete.
|
|
|
|
*/
|
2008-04-10 04:28:12 +08:00
|
|
|
struct end_io_wq {
|
|
|
|
struct bio *bio;
|
|
|
|
bio_end_io_t *end_io;
|
|
|
|
void *private;
|
|
|
|
struct btrfs_fs_info *info;
|
|
|
|
int error;
|
2008-04-10 04:28:12 +08:00
|
|
|
int metadata;
|
2008-04-10 04:28:12 +08:00
|
|
|
struct list_head list;
|
2008-06-12 04:50:36 +08:00
|
|
|
struct btrfs_work work;
|
2008-04-10 04:28:12 +08:00
|
|
|
};
|
2007-11-08 10:08:01 +08:00
|
|
|
|
2008-09-30 03:18:18 +08:00
|
|
|
/*
|
|
|
|
* async submit bios are used to offload expensive checksumming
|
|
|
|
* onto the worker threads. They checksum file and metadata bios
|
|
|
|
* just before they are sent down the IO stack.
|
|
|
|
*/
|
2008-04-16 23:14:51 +08:00
|
|
|
struct async_submit_bio {
|
|
|
|
struct inode *inode;
|
|
|
|
struct bio *bio;
|
|
|
|
struct list_head list;
|
Btrfs: Add ordered async work queues
Btrfs uses kernel threads to create async work queues for cpu intensive
operations such as checksumming and decompression. These work well,
but they make it difficult to keep IO order intact.
A single writepages call from pdflush or fsync will turn into a number
of bios, and each bio is checksummed in parallel. Once the checksum is
computed, the bio is sent down to the disk, and since we don't control
the order in which the parallel operations happen, they might go down to
the disk in almost any order.
The code deals with this somewhat by having deep work queues for a single
kernel thread, making it very likely that a single thread will process all
the bios for a single inode.
This patch introduces an explicitly ordered work queue. As work structs
are placed into the queue they are put onto the tail of a list. They have
three callbacks:
->func (cpu intensive processing here)
->ordered_func (order sensitive processing here)
->ordered_free (free the work struct, all processing is done)
The work struct has three callbacks. The func callback does the cpu intensive
work, and when it completes the work struct is marked as done.
Every time a work struct completes, the list is checked to see if the head
is marked as done. If so the ordered_func callback is used to do the
order sensitive processing and the ordered_free callback is used to do
any cleanup. Then we loop back and check the head of the list again.
This patch also changes the checksumming code to use the ordered workqueues.
One a 4 drive array, it increases streaming writes from 280MB/s to 350MB/s.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-11-07 11:03:00 +08:00
|
|
|
extent_submit_bio_hook_t *submit_bio_start;
|
|
|
|
extent_submit_bio_hook_t *submit_bio_done;
|
2008-04-16 23:14:51 +08:00
|
|
|
int rw;
|
|
|
|
int mirror_num;
|
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
|
|
|
unsigned long bio_flags;
|
2008-06-12 04:50:36 +08:00
|
|
|
struct btrfs_work work;
|
2008-04-16 23:14:51 +08:00
|
|
|
};
|
|
|
|
|
2008-09-30 03:18:18 +08:00
|
|
|
/*
|
|
|
|
* extents on the btree inode are pretty simple, there's one extent
|
|
|
|
* that covers the entire device
|
|
|
|
*/
|
2007-10-16 04:14:19 +08:00
|
|
|
struct extent_map *btree_get_extent(struct inode *inode, struct page *page,
|
2008-01-25 05:13:08 +08:00
|
|
|
size_t page_offset, u64 start, u64 len,
|
2007-10-16 04:14:19 +08:00
|
|
|
int create)
|
2007-04-12 03:53:25 +08:00
|
|
|
{
|
2007-10-16 04:14:19 +08:00
|
|
|
struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
|
|
|
|
struct extent_map *em;
|
|
|
|
int ret;
|
|
|
|
|
2008-01-25 05:13:08 +08:00
|
|
|
spin_lock(&em_tree->lock);
|
|
|
|
em = lookup_extent_mapping(em_tree, start, len);
|
2008-05-07 23:43:44 +08:00
|
|
|
if (em) {
|
|
|
|
em->bdev =
|
|
|
|
BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
|
|
|
|
spin_unlock(&em_tree->lock);
|
2007-10-16 04:14:19 +08:00
|
|
|
goto out;
|
2008-05-07 23:43:44 +08:00
|
|
|
}
|
|
|
|
spin_unlock(&em_tree->lock);
|
2008-04-18 22:29:50 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
em = alloc_extent_map(GFP_NOFS);
|
|
|
|
if (!em) {
|
|
|
|
em = ERR_PTR(-ENOMEM);
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
em->start = 0;
|
2008-04-19 02:17:20 +08:00
|
|
|
em->len = (u64)-1;
|
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
|
|
|
em->block_len = (u64)-1;
|
2007-10-16 04:14:19 +08:00
|
|
|
em->block_start = 0;
|
2008-05-07 23:43:44 +08:00
|
|
|
em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
|
2008-01-25 05:13:08 +08:00
|
|
|
|
|
|
|
spin_lock(&em_tree->lock);
|
2007-10-16 04:14:19 +08:00
|
|
|
ret = add_extent_mapping(em_tree, em);
|
|
|
|
if (ret == -EEXIST) {
|
2008-04-19 02:17:20 +08:00
|
|
|
u64 failed_start = em->start;
|
|
|
|
u64 failed_len = em->len;
|
|
|
|
|
|
|
|
printk("failed to insert %Lu %Lu -> %Lu into tree\n",
|
|
|
|
em->start, em->len, em->block_start);
|
2007-10-16 04:14:19 +08:00
|
|
|
free_extent_map(em);
|
2008-04-18 22:29:50 +08:00
|
|
|
em = lookup_extent_mapping(em_tree, start, len);
|
2008-04-19 02:17:20 +08:00
|
|
|
if (em) {
|
|
|
|
printk("after failing, found %Lu %Lu %Lu\n",
|
|
|
|
em->start, em->len, em->block_start);
|
2008-04-18 22:29:50 +08:00
|
|
|
ret = 0;
|
2008-04-19 02:17:20 +08:00
|
|
|
} else {
|
|
|
|
em = lookup_extent_mapping(em_tree, failed_start,
|
|
|
|
failed_len);
|
|
|
|
if (em) {
|
|
|
|
printk("double failure lookup gives us "
|
|
|
|
"%Lu %Lu -> %Lu\n", em->start,
|
|
|
|
em->len, em->block_start);
|
|
|
|
free_extent_map(em);
|
|
|
|
}
|
2008-04-18 22:29:50 +08:00
|
|
|
ret = -EIO;
|
2008-04-19 02:17:20 +08:00
|
|
|
}
|
2007-10-16 04:14:19 +08:00
|
|
|
} else if (ret) {
|
2008-04-18 22:29:50 +08:00
|
|
|
free_extent_map(em);
|
|
|
|
em = NULL;
|
2007-10-16 04:14:19 +08:00
|
|
|
}
|
2008-04-18 22:29:50 +08:00
|
|
|
spin_unlock(&em_tree->lock);
|
|
|
|
|
|
|
|
if (ret)
|
|
|
|
em = ERR_PTR(ret);
|
2007-10-16 04:14:19 +08:00
|
|
|
out:
|
|
|
|
return em;
|
2007-04-12 03:53:25 +08:00
|
|
|
}
|
|
|
|
|
2007-10-16 04:19:22 +08:00
|
|
|
u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
|
|
|
|
{
|
2008-04-12 03:45:51 +08:00
|
|
|
return btrfs_crc32c(seed, data, len);
|
2007-10-16 04:19:22 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
void btrfs_csum_final(u32 crc, char *result)
|
|
|
|
{
|
|
|
|
*(__le32 *)result = ~cpu_to_le32(crc);
|
|
|
|
}
|
|
|
|
|
2008-09-30 03:18:18 +08:00
|
|
|
/*
|
|
|
|
* compute the csum for a btree block, and either verify it or write it
|
|
|
|
* into the csum field of the block.
|
|
|
|
*/
|
2007-10-16 04:19:22 +08:00
|
|
|
static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
|
|
|
|
int verify)
|
|
|
|
{
|
|
|
|
char result[BTRFS_CRC32_SIZE];
|
|
|
|
unsigned long len;
|
|
|
|
unsigned long cur_len;
|
|
|
|
unsigned long offset = BTRFS_CSUM_SIZE;
|
|
|
|
char *map_token = NULL;
|
|
|
|
char *kaddr;
|
|
|
|
unsigned long map_start;
|
|
|
|
unsigned long map_len;
|
|
|
|
int err;
|
|
|
|
u32 crc = ~(u32)0;
|
|
|
|
|
|
|
|
len = buf->len - offset;
|
|
|
|
while(len > 0) {
|
|
|
|
err = map_private_extent_buffer(buf, offset, 32,
|
|
|
|
&map_token, &kaddr,
|
|
|
|
&map_start, &map_len, KM_USER0);
|
|
|
|
if (err) {
|
|
|
|
printk("failed to map extent buffer! %lu\n",
|
|
|
|
offset);
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
cur_len = min(len, map_len - (offset - map_start));
|
|
|
|
crc = btrfs_csum_data(root, kaddr + offset - map_start,
|
|
|
|
crc, cur_len);
|
|
|
|
len -= cur_len;
|
|
|
|
offset += cur_len;
|
|
|
|
unmap_extent_buffer(buf, map_token, KM_USER0);
|
|
|
|
}
|
|
|
|
btrfs_csum_final(crc, result);
|
|
|
|
|
|
|
|
if (verify) {
|
2008-01-09 04:46:27 +08:00
|
|
|
/* FIXME, this is not good */
|
2008-04-01 23:21:32 +08:00
|
|
|
if (memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
|
2008-01-09 04:46:27 +08:00
|
|
|
u32 val;
|
|
|
|
u32 found = 0;
|
|
|
|
memcpy(&found, result, BTRFS_CRC32_SIZE);
|
|
|
|
|
|
|
|
read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
|
|
|
|
printk("btrfs: %s checksum verify failed on %llu "
|
2008-08-04 20:20:15 +08:00
|
|
|
"wanted %X found %X level %d\n",
|
2007-10-16 04:19:22 +08:00
|
|
|
root->fs_info->sb->s_id,
|
2008-08-04 20:20:15 +08:00
|
|
|
buf->start, val, found, btrfs_header_level(buf));
|
2007-10-16 04:19:22 +08:00
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
write_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE);
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2008-09-30 03:18:18 +08:00
|
|
|
/*
|
|
|
|
* we can't consider a given block up to date unless the transid of the
|
|
|
|
* block matches the transid in the parent node's pointer. This is how we
|
|
|
|
* detect blocks that either didn't get written at all or got written
|
|
|
|
* in the wrong place.
|
|
|
|
*/
|
2008-05-13 01:39:03 +08:00
|
|
|
static int verify_parent_transid(struct extent_io_tree *io_tree,
|
|
|
|
struct extent_buffer *eb, u64 parent_transid)
|
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
|
|
|
|
if (extent_buffer_uptodate(io_tree, eb) &&
|
|
|
|
btrfs_header_generation(eb) == parent_transid) {
|
|
|
|
ret = 0;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
printk("parent transid verify failed on %llu wanted %llu found %llu\n",
|
|
|
|
(unsigned long long)eb->start,
|
|
|
|
(unsigned long long)parent_transid,
|
|
|
|
(unsigned long long)btrfs_header_generation(eb));
|
|
|
|
ret = 1;
|
|
|
|
clear_extent_buffer_uptodate(io_tree, eb);
|
2008-07-30 22:29:12 +08:00
|
|
|
out:
|
2008-05-13 01:39:03 +08:00
|
|
|
unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
|
|
|
|
GFP_NOFS);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2008-09-30 03:18:18 +08:00
|
|
|
/*
|
|
|
|
* helper to read a given tree block, doing retries as required when
|
|
|
|
* the checksums don't match and we have alternate mirrors to try.
|
|
|
|
*/
|
2008-04-10 04:28:12 +08:00
|
|
|
static int btree_read_extent_buffer_pages(struct btrfs_root *root,
|
|
|
|
struct extent_buffer *eb,
|
2008-05-13 00:59:19 +08:00
|
|
|
u64 start, u64 parent_transid)
|
2008-04-10 04:28:12 +08:00
|
|
|
{
|
|
|
|
struct extent_io_tree *io_tree;
|
|
|
|
int ret;
|
|
|
|
int num_copies = 0;
|
|
|
|
int mirror_num = 0;
|
|
|
|
|
|
|
|
io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
|
|
|
|
while (1) {
|
|
|
|
ret = read_extent_buffer_pages(io_tree, eb, start, 1,
|
|
|
|
btree_get_extent, mirror_num);
|
2008-05-13 01:39:03 +08:00
|
|
|
if (!ret &&
|
|
|
|
!verify_parent_transid(io_tree, eb, parent_transid))
|
2008-04-10 04:28:12 +08:00
|
|
|
return ret;
|
2008-09-06 04:09:51 +08:00
|
|
|
printk("read extent buffer pages failed with ret %d mirror no %d\n", ret, mirror_num);
|
2008-04-10 04:28:12 +08:00
|
|
|
num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
|
|
|
|
eb->start, eb->len);
|
2008-04-29 04:40:52 +08:00
|
|
|
if (num_copies == 1)
|
2008-04-10 04:28:12 +08:00
|
|
|
return ret;
|
2008-04-29 04:40:52 +08:00
|
|
|
|
2008-04-10 04:28:12 +08:00
|
|
|
mirror_num++;
|
2008-04-29 04:40:52 +08:00
|
|
|
if (mirror_num > num_copies)
|
2008-04-10 04:28:12 +08:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
return -EIO;
|
|
|
|
}
|
2007-10-16 04:19:22 +08:00
|
|
|
|
2008-09-30 03:18:18 +08:00
|
|
|
/*
|
|
|
|
* checksum a dirty tree block before IO. This has extra checks to make
|
|
|
|
* sure we only fill in the checksum field in the first page of a multi-page block
|
|
|
|
*/
|
2007-10-16 04:19:22 +08:00
|
|
|
int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
|
|
|
|
{
|
2008-01-25 05:13:08 +08:00
|
|
|
struct extent_io_tree *tree;
|
2007-10-31 04:56:53 +08:00
|
|
|
u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
|
2007-10-16 04:19:22 +08:00
|
|
|
u64 found_start;
|
|
|
|
int found_level;
|
|
|
|
unsigned long len;
|
|
|
|
struct extent_buffer *eb;
|
2008-04-10 04:28:12 +08:00
|
|
|
int ret;
|
|
|
|
|
2008-01-25 05:13:08 +08:00
|
|
|
tree = &BTRFS_I(page->mapping->host)->io_tree;
|
2007-10-16 04:19:22 +08:00
|
|
|
|
|
|
|
if (page->private == EXTENT_PAGE_PRIVATE)
|
|
|
|
goto out;
|
|
|
|
if (!page->private)
|
|
|
|
goto out;
|
|
|
|
len = page->private >> 2;
|
|
|
|
if (len == 0) {
|
|
|
|
WARN_ON(1);
|
|
|
|
}
|
|
|
|
eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
|
2008-05-13 00:59:19 +08:00
|
|
|
ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
|
|
|
|
btrfs_header_generation(eb));
|
2008-04-10 04:28:12 +08:00
|
|
|
BUG_ON(ret);
|
2007-10-16 04:19:22 +08:00
|
|
|
found_start = btrfs_header_bytenr(eb);
|
|
|
|
if (found_start != start) {
|
|
|
|
printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
|
|
|
|
start, found_start, len);
|
2008-01-10 04:55:33 +08:00
|
|
|
WARN_ON(1);
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
if (eb->first_page != page) {
|
|
|
|
printk("bad first page %lu %lu\n", eb->first_page->index,
|
|
|
|
page->index);
|
|
|
|
WARN_ON(1);
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
if (!PageUptodate(page)) {
|
|
|
|
printk("csum not up to date page %lu\n", page->index);
|
|
|
|
WARN_ON(1);
|
|
|
|
goto err;
|
2007-10-16 04:19:22 +08:00
|
|
|
}
|
|
|
|
found_level = btrfs_header_level(eb);
|
2008-09-08 23:18:08 +08:00
|
|
|
|
2007-10-16 04:19:22 +08:00
|
|
|
csum_tree_block(root, eb, 0);
|
2008-01-10 04:55:33 +08:00
|
|
|
err:
|
2007-10-16 04:19:22 +08:00
|
|
|
free_extent_buffer(eb);
|
|
|
|
out:
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2008-04-10 04:28:12 +08:00
|
|
|
int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
|
|
|
|
struct extent_state *state)
|
|
|
|
{
|
|
|
|
struct extent_io_tree *tree;
|
|
|
|
u64 found_start;
|
|
|
|
int found_level;
|
|
|
|
unsigned long len;
|
|
|
|
struct extent_buffer *eb;
|
|
|
|
struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
|
2008-04-10 04:28:12 +08:00
|
|
|
int ret = 0;
|
2008-04-10 04:28:12 +08:00
|
|
|
|
|
|
|
tree = &BTRFS_I(page->mapping->host)->io_tree;
|
|
|
|
if (page->private == EXTENT_PAGE_PRIVATE)
|
|
|
|
goto out;
|
|
|
|
if (!page->private)
|
|
|
|
goto out;
|
|
|
|
len = page->private >> 2;
|
|
|
|
if (len == 0) {
|
|
|
|
WARN_ON(1);
|
|
|
|
}
|
|
|
|
eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
|
2008-04-10 04:28:12 +08:00
|
|
|
|
2008-04-10 04:28:12 +08:00
|
|
|
found_start = btrfs_header_bytenr(eb);
|
2008-09-12 20:57:47 +08:00
|
|
|
if (found_start != start) {
|
2008-09-06 04:09:51 +08:00
|
|
|
printk("bad tree block start %llu %llu\n",
|
|
|
|
(unsigned long long)found_start,
|
|
|
|
(unsigned long long)eb->start);
|
2008-04-10 04:28:12 +08:00
|
|
|
ret = -EIO;
|
2008-04-10 04:28:12 +08:00
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
if (eb->first_page != page) {
|
|
|
|
printk("bad first page %lu %lu\n", eb->first_page->index,
|
|
|
|
page->index);
|
|
|
|
WARN_ON(1);
|
2008-04-10 04:28:12 +08:00
|
|
|
ret = -EIO;
|
2008-04-10 04:28:12 +08:00
|
|
|
goto err;
|
|
|
|
}
|
2008-05-13 01:39:03 +08:00
|
|
|
if (memcmp_extent_buffer(eb, root->fs_info->fsid,
|
|
|
|
(unsigned long)btrfs_header_fsid(eb),
|
|
|
|
BTRFS_FSID_SIZE)) {
|
|
|
|
printk("bad fsid on block %Lu\n", eb->start);
|
|
|
|
ret = -EIO;
|
|
|
|
goto err;
|
|
|
|
}
|
2008-04-10 04:28:12 +08:00
|
|
|
found_level = btrfs_header_level(eb);
|
|
|
|
|
|
|
|
ret = csum_tree_block(root, eb, 1);
|
2008-04-10 04:28:12 +08:00
|
|
|
if (ret)
|
|
|
|
ret = -EIO;
|
2008-04-10 04:28:12 +08:00
|
|
|
|
|
|
|
end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
|
|
|
|
end = eb->start + end - 1;
|
|
|
|
err:
|
|
|
|
free_extent_buffer(eb);
|
|
|
|
out:
|
2008-04-10 04:28:12 +08:00
|
|
|
return ret;
|
2008-04-10 04:28:12 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static void end_workqueue_bio(struct bio *bio, int err)
|
|
|
|
{
|
|
|
|
struct end_io_wq *end_io_wq = bio->bi_private;
|
|
|
|
struct btrfs_fs_info *fs_info;
|
|
|
|
|
|
|
|
fs_info = end_io_wq->info;
|
|
|
|
end_io_wq->error = err;
|
2008-06-12 04:50:36 +08:00
|
|
|
end_io_wq->work.func = end_workqueue_fn;
|
|
|
|
end_io_wq->work.flags = 0;
|
2008-07-18 00:53:50 +08:00
|
|
|
if (bio->bi_rw & (1 << BIO_RW))
|
|
|
|
btrfs_queue_worker(&fs_info->endio_write_workers,
|
|
|
|
&end_io_wq->work);
|
|
|
|
else
|
|
|
|
btrfs_queue_worker(&fs_info->endio_workers, &end_io_wq->work);
|
2008-04-10 04:28:12 +08:00
|
|
|
}
|
|
|
|
|
2008-04-10 04:28:12 +08:00
|
|
|
int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
|
|
|
|
int metadata)
|
2008-03-25 03:01:56 +08:00
|
|
|
{
|
2008-04-10 04:28:12 +08:00
|
|
|
struct end_io_wq *end_io_wq;
|
|
|
|
end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
|
|
|
|
if (!end_io_wq)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
end_io_wq->private = bio->bi_private;
|
|
|
|
end_io_wq->end_io = bio->bi_end_io;
|
2008-04-10 04:28:12 +08:00
|
|
|
end_io_wq->info = info;
|
2008-04-10 04:28:12 +08:00
|
|
|
end_io_wq->error = 0;
|
|
|
|
end_io_wq->bio = bio;
|
2008-04-10 04:28:12 +08:00
|
|
|
end_io_wq->metadata = metadata;
|
2008-04-10 04:28:12 +08:00
|
|
|
|
|
|
|
bio->bi_private = end_io_wq;
|
|
|
|
bio->bi_end_io = end_workqueue_bio;
|
2008-04-10 04:28:12 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2008-08-21 01:39:41 +08:00
|
|
|
unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
|
2008-08-16 03:34:15 +08:00
|
|
|
{
|
2008-08-16 03:34:17 +08:00
|
|
|
unsigned long limit = min_t(unsigned long,
|
|
|
|
info->workers.max_workers,
|
|
|
|
info->fs_devices->open_devices);
|
|
|
|
return 256 * limit;
|
|
|
|
}
|
2008-08-16 03:34:15 +08:00
|
|
|
|
2008-08-16 03:34:17 +08:00
|
|
|
int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
|
|
|
|
{
|
2008-08-21 01:39:41 +08:00
|
|
|
return atomic_read(&info->nr_async_bios) >
|
|
|
|
btrfs_async_submit_limit(info);
|
2008-08-16 03:34:15 +08:00
|
|
|
}
|
|
|
|
|
Btrfs: Add ordered async work queues
Btrfs uses kernel threads to create async work queues for cpu intensive
operations such as checksumming and decompression. These work well,
but they make it difficult to keep IO order intact.
A single writepages call from pdflush or fsync will turn into a number
of bios, and each bio is checksummed in parallel. Once the checksum is
computed, the bio is sent down to the disk, and since we don't control
the order in which the parallel operations happen, they might go down to
the disk in almost any order.
The code deals with this somewhat by having deep work queues for a single
kernel thread, making it very likely that a single thread will process all
the bios for a single inode.
This patch introduces an explicitly ordered work queue. As work structs
are placed into the queue they are put onto the tail of a list. They have
three callbacks:
->func (cpu intensive processing here)
->ordered_func (order sensitive processing here)
->ordered_free (free the work struct, all processing is done)
The work struct has three callbacks. The func callback does the cpu intensive
work, and when it completes the work struct is marked as done.
Every time a work struct completes, the list is checked to see if the head
is marked as done. If so the ordered_func callback is used to do the
order sensitive processing and the ordered_free callback is used to do
any cleanup. Then we loop back and check the head of the list again.
This patch also changes the checksumming code to use the ordered workqueues.
One a 4 drive array, it increases streaming writes from 280MB/s to 350MB/s.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-11-07 11:03:00 +08:00
|
|
|
static void run_one_async_start(struct btrfs_work *work)
|
|
|
|
{
|
|
|
|
struct btrfs_fs_info *fs_info;
|
|
|
|
struct async_submit_bio *async;
|
|
|
|
|
|
|
|
async = container_of(work, struct async_submit_bio, work);
|
|
|
|
fs_info = BTRFS_I(async->inode)->root->fs_info;
|
|
|
|
async->submit_bio_start(async->inode, async->rw, async->bio,
|
|
|
|
async->mirror_num, async->bio_flags);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void run_one_async_done(struct btrfs_work *work)
|
2008-06-12 04:50:36 +08:00
|
|
|
{
|
|
|
|
struct btrfs_fs_info *fs_info;
|
|
|
|
struct async_submit_bio *async;
|
2008-08-16 03:34:17 +08:00
|
|
|
int limit;
|
2008-06-12 04:50:36 +08:00
|
|
|
|
|
|
|
async = container_of(work, struct async_submit_bio, work);
|
|
|
|
fs_info = BTRFS_I(async->inode)->root->fs_info;
|
2008-08-16 03:34:17 +08:00
|
|
|
|
2008-08-21 01:39:41 +08:00
|
|
|
limit = btrfs_async_submit_limit(fs_info);
|
2008-08-16 03:34:17 +08:00
|
|
|
limit = limit * 2 / 3;
|
|
|
|
|
2008-06-12 04:50:36 +08:00
|
|
|
atomic_dec(&fs_info->nr_async_submits);
|
2008-08-16 03:34:15 +08:00
|
|
|
|
2008-08-21 01:39:41 +08:00
|
|
|
if (atomic_read(&fs_info->nr_async_submits) < limit &&
|
|
|
|
waitqueue_active(&fs_info->async_submit_wait))
|
2008-08-16 03:34:17 +08:00
|
|
|
wake_up(&fs_info->async_submit_wait);
|
|
|
|
|
Btrfs: Add ordered async work queues
Btrfs uses kernel threads to create async work queues for cpu intensive
operations such as checksumming and decompression. These work well,
but they make it difficult to keep IO order intact.
A single writepages call from pdflush or fsync will turn into a number
of bios, and each bio is checksummed in parallel. Once the checksum is
computed, the bio is sent down to the disk, and since we don't control
the order in which the parallel operations happen, they might go down to
the disk in almost any order.
The code deals with this somewhat by having deep work queues for a single
kernel thread, making it very likely that a single thread will process all
the bios for a single inode.
This patch introduces an explicitly ordered work queue. As work structs
are placed into the queue they are put onto the tail of a list. They have
three callbacks:
->func (cpu intensive processing here)
->ordered_func (order sensitive processing here)
->ordered_free (free the work struct, all processing is done)
The work struct has three callbacks. The func callback does the cpu intensive
work, and when it completes the work struct is marked as done.
Every time a work struct completes, the list is checked to see if the head
is marked as done. If so the ordered_func callback is used to do the
order sensitive processing and the ordered_free callback is used to do
any cleanup. Then we loop back and check the head of the list again.
This patch also changes the checksumming code to use the ordered workqueues.
One a 4 drive array, it increases streaming writes from 280MB/s to 350MB/s.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-11-07 11:03:00 +08:00
|
|
|
async->submit_bio_done(async->inode, async->rw, async->bio,
|
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
|
|
|
async->mirror_num, async->bio_flags);
|
Btrfs: Add ordered async work queues
Btrfs uses kernel threads to create async work queues for cpu intensive
operations such as checksumming and decompression. These work well,
but they make it difficult to keep IO order intact.
A single writepages call from pdflush or fsync will turn into a number
of bios, and each bio is checksummed in parallel. Once the checksum is
computed, the bio is sent down to the disk, and since we don't control
the order in which the parallel operations happen, they might go down to
the disk in almost any order.
The code deals with this somewhat by having deep work queues for a single
kernel thread, making it very likely that a single thread will process all
the bios for a single inode.
This patch introduces an explicitly ordered work queue. As work structs
are placed into the queue they are put onto the tail of a list. They have
three callbacks:
->func (cpu intensive processing here)
->ordered_func (order sensitive processing here)
->ordered_free (free the work struct, all processing is done)
The work struct has three callbacks. The func callback does the cpu intensive
work, and when it completes the work struct is marked as done.
Every time a work struct completes, the list is checked to see if the head
is marked as done. If so the ordered_func callback is used to do the
order sensitive processing and the ordered_free callback is used to do
any cleanup. Then we loop back and check the head of the list again.
This patch also changes the checksumming code to use the ordered workqueues.
One a 4 drive array, it increases streaming writes from 280MB/s to 350MB/s.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-11-07 11:03:00 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static void run_one_async_free(struct btrfs_work *work)
|
|
|
|
{
|
|
|
|
struct async_submit_bio *async;
|
|
|
|
|
|
|
|
async = container_of(work, struct async_submit_bio, work);
|
2008-06-12 04:50:36 +08:00
|
|
|
kfree(async);
|
|
|
|
}
|
|
|
|
|
2008-04-16 23:14:51 +08:00
|
|
|
int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
|
|
|
|
int rw, struct bio *bio, int mirror_num,
|
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
|
|
|
unsigned long bio_flags,
|
Btrfs: Add ordered async work queues
Btrfs uses kernel threads to create async work queues for cpu intensive
operations such as checksumming and decompression. These work well,
but they make it difficult to keep IO order intact.
A single writepages call from pdflush or fsync will turn into a number
of bios, and each bio is checksummed in parallel. Once the checksum is
computed, the bio is sent down to the disk, and since we don't control
the order in which the parallel operations happen, they might go down to
the disk in almost any order.
The code deals with this somewhat by having deep work queues for a single
kernel thread, making it very likely that a single thread will process all
the bios for a single inode.
This patch introduces an explicitly ordered work queue. As work structs
are placed into the queue they are put onto the tail of a list. They have
three callbacks:
->func (cpu intensive processing here)
->ordered_func (order sensitive processing here)
->ordered_free (free the work struct, all processing is done)
The work struct has three callbacks. The func callback does the cpu intensive
work, and when it completes the work struct is marked as done.
Every time a work struct completes, the list is checked to see if the head
is marked as done. If so the ordered_func callback is used to do the
order sensitive processing and the ordered_free callback is used to do
any cleanup. Then we loop back and check the head of the list again.
This patch also changes the checksumming code to use the ordered workqueues.
One a 4 drive array, it increases streaming writes from 280MB/s to 350MB/s.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-11-07 11:03:00 +08:00
|
|
|
extent_submit_bio_hook_t *submit_bio_start,
|
|
|
|
extent_submit_bio_hook_t *submit_bio_done)
|
2008-04-16 23:14:51 +08:00
|
|
|
{
|
|
|
|
struct async_submit_bio *async;
|
2008-08-21 01:39:41 +08:00
|
|
|
int limit = btrfs_async_submit_limit(fs_info);
|
2008-04-16 23:14:51 +08:00
|
|
|
|
|
|
|
async = kmalloc(sizeof(*async), GFP_NOFS);
|
|
|
|
if (!async)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
async->inode = inode;
|
|
|
|
async->rw = rw;
|
|
|
|
async->bio = bio;
|
|
|
|
async->mirror_num = mirror_num;
|
Btrfs: Add ordered async work queues
Btrfs uses kernel threads to create async work queues for cpu intensive
operations such as checksumming and decompression. These work well,
but they make it difficult to keep IO order intact.
A single writepages call from pdflush or fsync will turn into a number
of bios, and each bio is checksummed in parallel. Once the checksum is
computed, the bio is sent down to the disk, and since we don't control
the order in which the parallel operations happen, they might go down to
the disk in almost any order.
The code deals with this somewhat by having deep work queues for a single
kernel thread, making it very likely that a single thread will process all
the bios for a single inode.
This patch introduces an explicitly ordered work queue. As work structs
are placed into the queue they are put onto the tail of a list. They have
three callbacks:
->func (cpu intensive processing here)
->ordered_func (order sensitive processing here)
->ordered_free (free the work struct, all processing is done)
The work struct has three callbacks. The func callback does the cpu intensive
work, and when it completes the work struct is marked as done.
Every time a work struct completes, the list is checked to see if the head
is marked as done. If so the ordered_func callback is used to do the
order sensitive processing and the ordered_free callback is used to do
any cleanup. Then we loop back and check the head of the list again.
This patch also changes the checksumming code to use the ordered workqueues.
One a 4 drive array, it increases streaming writes from 280MB/s to 350MB/s.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-11-07 11:03:00 +08:00
|
|
|
async->submit_bio_start = submit_bio_start;
|
|
|
|
async->submit_bio_done = submit_bio_done;
|
|
|
|
|
|
|
|
async->work.func = run_one_async_start;
|
|
|
|
async->work.ordered_func = run_one_async_done;
|
|
|
|
async->work.ordered_free = run_one_async_free;
|
|
|
|
|
2008-06-12 04:50:36 +08:00
|
|
|
async->work.flags = 0;
|
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
|
|
|
async->bio_flags = bio_flags;
|
2008-09-29 23:19:10 +08:00
|
|
|
|
|
|
|
while(atomic_read(&fs_info->async_submit_draining) &&
|
|
|
|
atomic_read(&fs_info->nr_async_submits)) {
|
|
|
|
wait_event(fs_info->async_submit_wait,
|
|
|
|
(atomic_read(&fs_info->nr_async_submits) == 0));
|
|
|
|
}
|
|
|
|
|
2008-05-16 04:15:45 +08:00
|
|
|
atomic_inc(&fs_info->nr_async_submits);
|
2008-06-12 04:50:36 +08:00
|
|
|
btrfs_queue_worker(&fs_info->workers, &async->work);
|
2008-08-16 03:34:17 +08:00
|
|
|
|
2008-08-28 18:15:24 +08:00
|
|
|
if (atomic_read(&fs_info->nr_async_submits) > limit) {
|
|
|
|
wait_event_timeout(fs_info->async_submit_wait,
|
2008-08-16 03:34:17 +08:00
|
|
|
(atomic_read(&fs_info->nr_async_submits) < limit),
|
|
|
|
HZ/10);
|
2008-08-28 18:15:24 +08:00
|
|
|
|
|
|
|
wait_event_timeout(fs_info->async_submit_wait,
|
|
|
|
(atomic_read(&fs_info->nr_async_bios) < limit),
|
|
|
|
HZ/10);
|
|
|
|
}
|
2008-11-07 11:02:51 +08:00
|
|
|
|
|
|
|
while(atomic_read(&fs_info->async_submit_draining) &&
|
|
|
|
atomic_read(&fs_info->nr_async_submits)) {
|
|
|
|
wait_event(fs_info->async_submit_wait,
|
|
|
|
(atomic_read(&fs_info->nr_async_submits) == 0));
|
|
|
|
}
|
|
|
|
|
2008-04-16 23:14:51 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2008-09-24 01:14:12 +08:00
|
|
|
static int btree_csum_one_bio(struct bio *bio)
|
|
|
|
{
|
|
|
|
struct bio_vec *bvec = bio->bi_io_vec;
|
|
|
|
int bio_index = 0;
|
|
|
|
struct btrfs_root *root;
|
|
|
|
|
|
|
|
WARN_ON(bio->bi_vcnt <= 0);
|
|
|
|
while(bio_index < bio->bi_vcnt) {
|
|
|
|
root = BTRFS_I(bvec->bv_page->mapping->host)->root;
|
|
|
|
csum_dirty_buffer(root, bvec->bv_page);
|
|
|
|
bio_index++;
|
|
|
|
bvec++;
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
Btrfs: Add ordered async work queues
Btrfs uses kernel threads to create async work queues for cpu intensive
operations such as checksumming and decompression. These work well,
but they make it difficult to keep IO order intact.
A single writepages call from pdflush or fsync will turn into a number
of bios, and each bio is checksummed in parallel. Once the checksum is
computed, the bio is sent down to the disk, and since we don't control
the order in which the parallel operations happen, they might go down to
the disk in almost any order.
The code deals with this somewhat by having deep work queues for a single
kernel thread, making it very likely that a single thread will process all
the bios for a single inode.
This patch introduces an explicitly ordered work queue. As work structs
are placed into the queue they are put onto the tail of a list. They have
three callbacks:
->func (cpu intensive processing here)
->ordered_func (order sensitive processing here)
->ordered_free (free the work struct, all processing is done)
The work struct has three callbacks. The func callback does the cpu intensive
work, and when it completes the work struct is marked as done.
Every time a work struct completes, the list is checked to see if the head
is marked as done. If so the ordered_func callback is used to do the
order sensitive processing and the ordered_free callback is used to do
any cleanup. Then we loop back and check the head of the list again.
This patch also changes the checksumming code to use the ordered workqueues.
One a 4 drive array, it increases streaming writes from 280MB/s to 350MB/s.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-11-07 11:03:00 +08:00
|
|
|
static int __btree_submit_bio_start(struct inode *inode, int rw,
|
|
|
|
struct bio *bio, int mirror_num,
|
|
|
|
unsigned long bio_flags)
|
2008-04-10 04:28:12 +08:00
|
|
|
{
|
2008-06-12 04:50:36 +08:00
|
|
|
/*
|
|
|
|
* when we're called for a write, we're already in the async
|
2008-08-16 03:34:16 +08:00
|
|
|
* submission context. Just jump into btrfs_map_bio
|
2008-06-12 04:50:36 +08:00
|
|
|
*/
|
Btrfs: Add ordered async work queues
Btrfs uses kernel threads to create async work queues for cpu intensive
operations such as checksumming and decompression. These work well,
but they make it difficult to keep IO order intact.
A single writepages call from pdflush or fsync will turn into a number
of bios, and each bio is checksummed in parallel. Once the checksum is
computed, the bio is sent down to the disk, and since we don't control
the order in which the parallel operations happen, they might go down to
the disk in almost any order.
The code deals with this somewhat by having deep work queues for a single
kernel thread, making it very likely that a single thread will process all
the bios for a single inode.
This patch introduces an explicitly ordered work queue. As work structs
are placed into the queue they are put onto the tail of a list. They have
three callbacks:
->func (cpu intensive processing here)
->ordered_func (order sensitive processing here)
->ordered_free (free the work struct, all processing is done)
The work struct has three callbacks. The func callback does the cpu intensive
work, and when it completes the work struct is marked as done.
Every time a work struct completes, the list is checked to see if the head
is marked as done. If so the ordered_func callback is used to do the
order sensitive processing and the ordered_free callback is used to do
any cleanup. Then we loop back and check the head of the list again.
This patch also changes the checksumming code to use the ordered workqueues.
One a 4 drive array, it increases streaming writes from 280MB/s to 350MB/s.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-11-07 11:03:00 +08:00
|
|
|
btree_csum_one_bio(bio);
|
|
|
|
return 0;
|
|
|
|
}
|
2008-04-10 04:28:12 +08:00
|
|
|
|
Btrfs: Add ordered async work queues
Btrfs uses kernel threads to create async work queues for cpu intensive
operations such as checksumming and decompression. These work well,
but they make it difficult to keep IO order intact.
A single writepages call from pdflush or fsync will turn into a number
of bios, and each bio is checksummed in parallel. Once the checksum is
computed, the bio is sent down to the disk, and since we don't control
the order in which the parallel operations happen, they might go down to
the disk in almost any order.
The code deals with this somewhat by having deep work queues for a single
kernel thread, making it very likely that a single thread will process all
the bios for a single inode.
This patch introduces an explicitly ordered work queue. As work structs
are placed into the queue they are put onto the tail of a list. They have
three callbacks:
->func (cpu intensive processing here)
->ordered_func (order sensitive processing here)
->ordered_free (free the work struct, all processing is done)
The work struct has three callbacks. The func callback does the cpu intensive
work, and when it completes the work struct is marked as done.
Every time a work struct completes, the list is checked to see if the head
is marked as done. If so the ordered_func callback is used to do the
order sensitive processing and the ordered_free callback is used to do
any cleanup. Then we loop back and check the head of the list again.
This patch also changes the checksumming code to use the ordered workqueues.
One a 4 drive array, it increases streaming writes from 280MB/s to 350MB/s.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-11-07 11:03:00 +08:00
|
|
|
static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
|
|
|
|
int mirror_num, unsigned long bio_flags)
|
|
|
|
{
|
2008-06-12 04:50:36 +08:00
|
|
|
/*
|
Btrfs: Add ordered async work queues
Btrfs uses kernel threads to create async work queues for cpu intensive
operations such as checksumming and decompression. These work well,
but they make it difficult to keep IO order intact.
A single writepages call from pdflush or fsync will turn into a number
of bios, and each bio is checksummed in parallel. Once the checksum is
computed, the bio is sent down to the disk, and since we don't control
the order in which the parallel operations happen, they might go down to
the disk in almost any order.
The code deals with this somewhat by having deep work queues for a single
kernel thread, making it very likely that a single thread will process all
the bios for a single inode.
This patch introduces an explicitly ordered work queue. As work structs
are placed into the queue they are put onto the tail of a list. They have
three callbacks:
->func (cpu intensive processing here)
->ordered_func (order sensitive processing here)
->ordered_free (free the work struct, all processing is done)
The work struct has three callbacks. The func callback does the cpu intensive
work, and when it completes the work struct is marked as done.
Every time a work struct completes, the list is checked to see if the head
is marked as done. If so the ordered_func callback is used to do the
order sensitive processing and the ordered_free callback is used to do
any cleanup. Then we loop back and check the head of the list again.
This patch also changes the checksumming code to use the ordered workqueues.
One a 4 drive array, it increases streaming writes from 280MB/s to 350MB/s.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-11-07 11:03:00 +08:00
|
|
|
* when we're called for a write, we're already in the async
|
|
|
|
* submission context. Just jump into btrfs_map_bio
|
2008-06-12 04:50:36 +08:00
|
|
|
*/
|
|
|
|
return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
|
2008-03-25 03:01:56 +08:00
|
|
|
}
|
|
|
|
|
2008-04-16 23:14:51 +08:00
|
|
|
static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
|
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
|
|
|
int mirror_num, unsigned long bio_flags)
|
2008-04-16 23:14:51 +08:00
|
|
|
{
|
2008-06-12 04:50:36 +08:00
|
|
|
/*
|
|
|
|
* kthread helpers are used to submit writes so that checksumming
|
|
|
|
* can happen in parallel across all CPUs
|
|
|
|
*/
|
2008-04-16 23:14:51 +08:00
|
|
|
if (!(rw & (1 << BIO_RW))) {
|
Btrfs: Add ordered async work queues
Btrfs uses kernel threads to create async work queues for cpu intensive
operations such as checksumming and decompression. These work well,
but they make it difficult to keep IO order intact.
A single writepages call from pdflush or fsync will turn into a number
of bios, and each bio is checksummed in parallel. Once the checksum is
computed, the bio is sent down to the disk, and since we don't control
the order in which the parallel operations happen, they might go down to
the disk in almost any order.
The code deals with this somewhat by having deep work queues for a single
kernel thread, making it very likely that a single thread will process all
the bios for a single inode.
This patch introduces an explicitly ordered work queue. As work structs
are placed into the queue they are put onto the tail of a list. They have
three callbacks:
->func (cpu intensive processing here)
->ordered_func (order sensitive processing here)
->ordered_free (free the work struct, all processing is done)
The work struct has three callbacks. The func callback does the cpu intensive
work, and when it completes the work struct is marked as done.
Every time a work struct completes, the list is checked to see if the head
is marked as done. If so the ordered_func callback is used to do the
order sensitive processing and the ordered_free callback is used to do
any cleanup. Then we loop back and check the head of the list again.
This patch also changes the checksumming code to use the ordered workqueues.
One a 4 drive array, it increases streaming writes from 280MB/s to 350MB/s.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-11-07 11:03:00 +08:00
|
|
|
int ret;
|
|
|
|
/*
|
|
|
|
* called for a read, do the setup so that checksum validation
|
|
|
|
* can happen in the async kernel threads
|
|
|
|
*/
|
|
|
|
ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
|
|
|
|
bio, 1);
|
|
|
|
BUG_ON(ret);
|
|
|
|
|
|
|
|
return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
|
2008-11-13 22:59:36 +08:00
|
|
|
mirror_num, 0);
|
2008-04-16 23:14:51 +08:00
|
|
|
}
|
|
|
|
return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
|
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
|
|
|
inode, rw, bio, mirror_num, 0,
|
Btrfs: Add ordered async work queues
Btrfs uses kernel threads to create async work queues for cpu intensive
operations such as checksumming and decompression. These work well,
but they make it difficult to keep IO order intact.
A single writepages call from pdflush or fsync will turn into a number
of bios, and each bio is checksummed in parallel. Once the checksum is
computed, the bio is sent down to the disk, and since we don't control
the order in which the parallel operations happen, they might go down to
the disk in almost any order.
The code deals with this somewhat by having deep work queues for a single
kernel thread, making it very likely that a single thread will process all
the bios for a single inode.
This patch introduces an explicitly ordered work queue. As work structs
are placed into the queue they are put onto the tail of a list. They have
three callbacks:
->func (cpu intensive processing here)
->ordered_func (order sensitive processing here)
->ordered_free (free the work struct, all processing is done)
The work struct has three callbacks. The func callback does the cpu intensive
work, and when it completes the work struct is marked as done.
Every time a work struct completes, the list is checked to see if the head
is marked as done. If so the ordered_func callback is used to do the
order sensitive processing and the ordered_free callback is used to do
any cleanup. Then we loop back and check the head of the list again.
This patch also changes the checksumming code to use the ordered workqueues.
One a 4 drive array, it increases streaming writes from 280MB/s to 350MB/s.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-11-07 11:03:00 +08:00
|
|
|
__btree_submit_bio_start,
|
|
|
|
__btree_submit_bio_done);
|
2008-04-16 23:14:51 +08:00
|
|
|
}
|
|
|
|
|
2007-11-08 10:08:01 +08:00
|
|
|
static int btree_writepage(struct page *page, struct writeback_control *wbc)
|
|
|
|
{
|
2008-01-25 05:13:08 +08:00
|
|
|
struct extent_io_tree *tree;
|
|
|
|
tree = &BTRFS_I(page->mapping->host)->io_tree;
|
2008-08-16 03:34:16 +08:00
|
|
|
|
|
|
|
if (current->flags & PF_MEMALLOC) {
|
|
|
|
redirty_page_for_writepage(wbc, page);
|
|
|
|
unlock_page(page);
|
|
|
|
return 0;
|
|
|
|
}
|
2007-10-16 04:14:19 +08:00
|
|
|
return extent_write_full_page(tree, page, btree_get_extent, wbc);
|
|
|
|
}
|
2007-11-08 10:08:01 +08:00
|
|
|
|
|
|
|
static int btree_writepages(struct address_space *mapping,
|
|
|
|
struct writeback_control *wbc)
|
|
|
|
{
|
2008-01-25 05:13:08 +08:00
|
|
|
struct extent_io_tree *tree;
|
|
|
|
tree = &BTRFS_I(mapping->host)->io_tree;
|
2007-12-12 01:42:00 +08:00
|
|
|
if (wbc->sync_mode == WB_SYNC_NONE) {
|
2007-11-27 08:34:41 +08:00
|
|
|
u64 num_dirty;
|
|
|
|
u64 start = 0;
|
2008-09-25 02:51:30 +08:00
|
|
|
unsigned long thresh = 32 * 1024 * 1024;
|
2007-11-27 23:52:01 +08:00
|
|
|
|
|
|
|
if (wbc->for_kupdate)
|
|
|
|
return 0;
|
|
|
|
|
2007-12-22 05:27:21 +08:00
|
|
|
num_dirty = count_range_bits(tree, &start, (u64)-1,
|
|
|
|
thresh, EXTENT_DIRTY);
|
2007-11-27 08:34:41 +08:00
|
|
|
if (num_dirty < thresh) {
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
}
|
2007-11-08 10:08:01 +08:00
|
|
|
return extent_writepages(tree, mapping, btree_get_extent, wbc);
|
|
|
|
}
|
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
int btree_readpage(struct file *file, struct page *page)
|
|
|
|
{
|
2008-01-25 05:13:08 +08:00
|
|
|
struct extent_io_tree *tree;
|
|
|
|
tree = &BTRFS_I(page->mapping->host)->io_tree;
|
2007-10-16 04:14:19 +08:00
|
|
|
return extent_read_full_page(tree, page, btree_get_extent);
|
|
|
|
}
|
2007-03-30 20:47:31 +08:00
|
|
|
|
2008-01-29 22:59:12 +08:00
|
|
|
static int btree_releasepage(struct page *page, gfp_t gfp_flags)
|
2007-10-16 04:14:19 +08:00
|
|
|
{
|
2008-01-25 05:13:08 +08:00
|
|
|
struct extent_io_tree *tree;
|
|
|
|
struct extent_map_tree *map;
|
2007-10-16 04:14:19 +08:00
|
|
|
int ret;
|
2007-03-29 01:57:48 +08:00
|
|
|
|
2008-09-12 03:51:43 +08:00
|
|
|
if (PageWriteback(page) || PageDirty(page))
|
|
|
|
return 0;
|
|
|
|
|
2008-01-25 05:13:08 +08:00
|
|
|
tree = &BTRFS_I(page->mapping->host)->io_tree;
|
|
|
|
map = &BTRFS_I(page->mapping->host)->extent_tree;
|
2008-07-22 23:18:07 +08:00
|
|
|
|
2008-04-18 22:29:50 +08:00
|
|
|
ret = try_release_extent_state(map, tree, page, gfp_flags);
|
2008-07-22 23:18:07 +08:00
|
|
|
if (!ret) {
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
ret = try_release_extent_buffer(tree, page);
|
2007-10-16 04:14:19 +08:00
|
|
|
if (ret == 1) {
|
|
|
|
ClearPagePrivate(page);
|
|
|
|
set_page_private(page, 0);
|
|
|
|
page_cache_release(page);
|
|
|
|
}
|
2008-07-22 23:18:07 +08:00
|
|
|
|
2007-03-29 01:57:48 +08:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
static void btree_invalidatepage(struct page *page, unsigned long offset)
|
2007-03-29 01:57:48 +08:00
|
|
|
{
|
2008-01-25 05:13:08 +08:00
|
|
|
struct extent_io_tree *tree;
|
|
|
|
tree = &BTRFS_I(page->mapping->host)->io_tree;
|
2007-10-16 04:14:19 +08:00
|
|
|
extent_invalidatepage(tree, page, offset);
|
|
|
|
btree_releasepage(page, GFP_NOFS);
|
2008-04-19 04:11:30 +08:00
|
|
|
if (PagePrivate(page)) {
|
2008-07-22 23:18:07 +08:00
|
|
|
printk("warning page private not zero on page %Lu\n",
|
|
|
|
page_offset(page));
|
2008-04-19 04:11:30 +08:00
|
|
|
ClearPagePrivate(page);
|
|
|
|
set_page_private(page, 0);
|
|
|
|
page_cache_release(page);
|
|
|
|
}
|
2007-03-29 01:57:48 +08:00
|
|
|
}
|
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
#if 0
|
2007-03-29 01:57:48 +08:00
|
|
|
static int btree_writepage(struct page *page, struct writeback_control *wbc)
|
2007-03-02 07:59:40 +08:00
|
|
|
{
|
2007-03-29 07:44:27 +08:00
|
|
|
struct buffer_head *bh;
|
2007-04-09 22:42:37 +08:00
|
|
|
struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
|
2007-03-29 07:44:27 +08:00
|
|
|
struct buffer_head *head;
|
|
|
|
if (!page_has_buffers(page)) {
|
|
|
|
create_empty_buffers(page, root->fs_info->sb->s_blocksize,
|
|
|
|
(1 << BH_Dirty)|(1 << BH_Uptodate));
|
|
|
|
}
|
|
|
|
head = page_buffers(page);
|
|
|
|
bh = head;
|
|
|
|
do {
|
|
|
|
if (buffer_dirty(bh))
|
|
|
|
csum_tree_block(root, bh, 0);
|
|
|
|
bh = bh->b_this_page;
|
|
|
|
} while (bh != head);
|
2007-03-29 01:57:48 +08:00
|
|
|
return block_write_full_page(page, btree_get_block, wbc);
|
2007-03-02 07:59:40 +08:00
|
|
|
}
|
2007-10-16 04:14:19 +08:00
|
|
|
#endif
|
2007-02-02 22:18:22 +08:00
|
|
|
|
2007-03-29 01:57:48 +08:00
|
|
|
static struct address_space_operations btree_aops = {
|
|
|
|
.readpage = btree_readpage,
|
|
|
|
.writepage = btree_writepage,
|
2007-11-08 10:08:01 +08:00
|
|
|
.writepages = btree_writepages,
|
2007-10-16 04:14:19 +08:00
|
|
|
.releasepage = btree_releasepage,
|
|
|
|
.invalidatepage = btree_invalidatepage,
|
2007-03-29 01:57:48 +08:00
|
|
|
.sync_page = block_sync_page,
|
|
|
|
};
|
|
|
|
|
2008-05-13 00:59:19 +08:00
|
|
|
int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
|
|
|
|
u64 parent_transid)
|
2007-05-01 20:53:32 +08:00
|
|
|
{
|
2007-10-16 04:14:19 +08:00
|
|
|
struct extent_buffer *buf = NULL;
|
|
|
|
struct inode *btree_inode = root->fs_info->btree_inode;
|
2007-05-19 01:28:27 +08:00
|
|
|
int ret = 0;
|
2007-05-01 20:53:32 +08:00
|
|
|
|
2007-10-16 04:15:53 +08:00
|
|
|
buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
|
2007-10-16 04:14:19 +08:00
|
|
|
if (!buf)
|
2007-05-01 20:53:32 +08:00
|
|
|
return 0;
|
2008-01-25 05:13:08 +08:00
|
|
|
read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
|
2008-04-10 04:28:12 +08:00
|
|
|
buf, 0, 0, btree_get_extent, 0);
|
2007-10-16 04:14:19 +08:00
|
|
|
free_extent_buffer(buf);
|
2007-05-19 01:28:27 +08:00
|
|
|
return ret;
|
2007-05-01 20:53:32 +08:00
|
|
|
}
|
|
|
|
|
2008-04-02 01:48:14 +08:00
|
|
|
struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
|
|
|
|
u64 bytenr, u32 blocksize)
|
|
|
|
{
|
|
|
|
struct inode *btree_inode = root->fs_info->btree_inode;
|
|
|
|
struct extent_buffer *eb;
|
|
|
|
eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
|
|
|
|
bytenr, blocksize, GFP_NOFS);
|
|
|
|
return eb;
|
|
|
|
}
|
|
|
|
|
|
|
|
struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
|
|
|
|
u64 bytenr, u32 blocksize)
|
|
|
|
{
|
|
|
|
struct inode *btree_inode = root->fs_info->btree_inode;
|
|
|
|
struct extent_buffer *eb;
|
|
|
|
|
|
|
|
eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
|
|
|
|
bytenr, blocksize, NULL, GFP_NOFS);
|
|
|
|
return eb;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2008-09-06 04:13:11 +08:00
|
|
|
int btrfs_write_tree_block(struct extent_buffer *buf)
|
|
|
|
{
|
|
|
|
return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
|
2008-09-25 02:51:30 +08:00
|
|
|
buf->start + buf->len - 1, WB_SYNC_ALL);
|
2008-09-06 04:13:11 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
|
|
|
|
{
|
|
|
|
return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
|
|
|
|
buf->start, buf->start + buf->len -1);
|
|
|
|
}
|
|
|
|
|
2008-04-02 01:48:14 +08:00
|
|
|
struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
|
2008-05-13 00:59:19 +08:00
|
|
|
u32 blocksize, u64 parent_transid)
|
2008-04-02 01:48:14 +08:00
|
|
|
{
|
|
|
|
struct extent_buffer *buf = NULL;
|
|
|
|
struct inode *btree_inode = root->fs_info->btree_inode;
|
|
|
|
struct extent_io_tree *io_tree;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
io_tree = &BTRFS_I(btree_inode)->io_tree;
|
|
|
|
|
|
|
|
buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
|
|
|
|
if (!buf)
|
|
|
|
return NULL;
|
|
|
|
|
2008-05-13 00:59:19 +08:00
|
|
|
ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
|
2008-04-10 04:28:12 +08:00
|
|
|
|
|
|
|
if (ret == 0) {
|
|
|
|
buf->flags |= EXTENT_UPTODATE;
|
2008-09-06 04:09:51 +08:00
|
|
|
} else {
|
|
|
|
WARN_ON(1);
|
2008-04-10 04:28:12 +08:00
|
|
|
}
|
2007-10-16 04:14:19 +08:00
|
|
|
return buf;
|
2008-04-10 04:28:12 +08:00
|
|
|
|
2007-02-02 22:18:22 +08:00
|
|
|
}
|
|
|
|
|
2007-03-17 04:20:31 +08:00
|
|
|
int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
|
2007-10-16 04:14:19 +08:00
|
|
|
struct extent_buffer *buf)
|
2007-03-02 07:59:40 +08:00
|
|
|
{
|
2007-10-16 04:14:19 +08:00
|
|
|
struct inode *btree_inode = root->fs_info->btree_inode;
|
2008-01-10 04:55:33 +08:00
|
|
|
if (btrfs_header_generation(buf) ==
|
2008-06-26 04:01:30 +08:00
|
|
|
root->fs_info->running_transaction->transid) {
|
|
|
|
WARN_ON(!btrfs_tree_locked(buf));
|
2008-01-25 05:13:08 +08:00
|
|
|
clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
|
2008-01-10 04:55:33 +08:00
|
|
|
buf);
|
2008-06-26 04:01:30 +08:00
|
|
|
}
|
2007-10-16 04:14:19 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2007-10-16 04:15:53 +08:00
|
|
|
static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
|
2007-12-01 00:30:34 +08:00
|
|
|
u32 stripesize, struct btrfs_root *root,
|
2007-03-21 02:38:32 +08:00
|
|
|
struct btrfs_fs_info *fs_info,
|
2007-03-23 00:13:20 +08:00
|
|
|
u64 objectid)
|
2007-02-21 05:40:44 +08:00
|
|
|
{
|
2007-02-22 06:04:57 +08:00
|
|
|
root->node = NULL;
|
2007-04-09 22:42:37 +08:00
|
|
|
root->inode = NULL;
|
2007-03-07 09:08:01 +08:00
|
|
|
root->commit_root = NULL;
|
2008-07-29 03:32:19 +08:00
|
|
|
root->ref_tree = NULL;
|
2007-10-16 04:15:53 +08:00
|
|
|
root->sectorsize = sectorsize;
|
|
|
|
root->nodesize = nodesize;
|
|
|
|
root->leafsize = leafsize;
|
2007-12-01 00:30:34 +08:00
|
|
|
root->stripesize = stripesize;
|
2007-03-15 02:14:43 +08:00
|
|
|
root->ref_cows = 0;
|
2008-03-25 03:01:56 +08:00
|
|
|
root->track_dirty = 0;
|
|
|
|
|
2007-03-21 02:38:32 +08:00
|
|
|
root->fs_info = fs_info;
|
2007-04-09 22:42:37 +08:00
|
|
|
root->objectid = objectid;
|
|
|
|
root->last_trans = 0;
|
2007-04-11 00:13:09 +08:00
|
|
|
root->highest_inode = 0;
|
|
|
|
root->last_inode_alloc = 0;
|
2007-08-30 03:47:34 +08:00
|
|
|
root->name = NULL;
|
2008-01-03 22:08:48 +08:00
|
|
|
root->in_sysfs = 0;
|
2008-03-25 03:01:56 +08:00
|
|
|
|
|
|
|
INIT_LIST_HEAD(&root->dirty_list);
|
2008-07-25 00:17:14 +08:00
|
|
|
INIT_LIST_HEAD(&root->orphan_list);
|
2008-07-31 04:29:20 +08:00
|
|
|
INIT_LIST_HEAD(&root->dead_list);
|
2008-06-26 04:01:30 +08:00
|
|
|
spin_lock_init(&root->node_lock);
|
2008-07-31 04:29:20 +08:00
|
|
|
spin_lock_init(&root->list_lock);
|
2008-06-26 04:01:30 +08:00
|
|
|
mutex_init(&root->objectid_mutex);
|
2008-09-06 04:13:11 +08:00
|
|
|
mutex_init(&root->log_mutex);
|
2008-09-12 04:17:57 +08:00
|
|
|
extent_io_tree_init(&root->dirty_log_pages,
|
|
|
|
fs_info->btree_inode->i_mapping, GFP_NOFS);
|
2008-07-29 03:32:51 +08:00
|
|
|
|
|
|
|
btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
|
|
|
|
root->ref_tree = &root->ref_tree_struct;
|
|
|
|
|
2007-03-14 04:47:54 +08:00
|
|
|
memset(&root->root_key, 0, sizeof(root->root_key));
|
|
|
|
memset(&root->root_item, 0, sizeof(root->root_item));
|
2007-08-08 04:15:09 +08:00
|
|
|
memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
|
2007-08-30 03:47:34 +08:00
|
|
|
memset(&root->root_kobj, 0, sizeof(root->root_kobj));
|
2008-06-26 04:01:31 +08:00
|
|
|
root->defrag_trans_start = fs_info->generation;
|
2007-08-30 03:47:34 +08:00
|
|
|
init_completion(&root->kobj_unregister);
|
2007-08-08 04:15:09 +08:00
|
|
|
root->defrag_running = 0;
|
|
|
|
root->defrag_level = 0;
|
2007-04-21 08:23:12 +08:00
|
|
|
root->root_key.objectid = objectid;
|
2007-03-14 04:47:54 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2007-10-16 04:15:53 +08:00
|
|
|
static int find_and_setup_root(struct btrfs_root *tree_root,
|
2007-03-21 02:38:32 +08:00
|
|
|
struct btrfs_fs_info *fs_info,
|
|
|
|
u64 objectid,
|
2007-03-23 00:13:20 +08:00
|
|
|
struct btrfs_root *root)
|
2007-03-14 04:47:54 +08:00
|
|
|
{
|
|
|
|
int ret;
|
2007-10-16 04:15:53 +08:00
|
|
|
u32 blocksize;
|
2008-10-30 02:49:05 +08:00
|
|
|
u64 generation;
|
2007-03-14 04:47:54 +08:00
|
|
|
|
2007-10-16 04:15:53 +08:00
|
|
|
__setup_root(tree_root->nodesize, tree_root->leafsize,
|
2007-12-01 00:30:34 +08:00
|
|
|
tree_root->sectorsize, tree_root->stripesize,
|
|
|
|
root, fs_info, objectid);
|
2007-03-14 04:47:54 +08:00
|
|
|
ret = btrfs_find_last_root(tree_root, objectid,
|
|
|
|
&root->root_item, &root->root_key);
|
|
|
|
BUG_ON(ret);
|
|
|
|
|
2008-10-30 02:49:05 +08:00
|
|
|
generation = btrfs_root_generation(&root->root_item);
|
2007-10-16 04:15:53 +08:00
|
|
|
blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
|
|
|
|
root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
|
2008-10-30 02:49:05 +08:00
|
|
|
blocksize, generation);
|
2007-03-14 04:47:54 +08:00
|
|
|
BUG_ON(!root->node);
|
2007-02-21 05:40:44 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2008-09-06 04:13:11 +08:00
|
|
|
int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
|
|
|
|
struct btrfs_fs_info *fs_info)
|
|
|
|
{
|
|
|
|
struct extent_buffer *eb;
|
2008-09-12 04:17:57 +08:00
|
|
|
struct btrfs_root *log_root_tree = fs_info->log_root_tree;
|
|
|
|
u64 start = 0;
|
|
|
|
u64 end = 0;
|
2008-09-06 04:13:11 +08:00
|
|
|
int ret;
|
|
|
|
|
2008-09-12 04:17:57 +08:00
|
|
|
if (!log_root_tree)
|
2008-09-06 04:13:11 +08:00
|
|
|
return 0;
|
|
|
|
|
2008-09-12 04:17:57 +08:00
|
|
|
while(1) {
|
|
|
|
ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
|
|
|
|
0, &start, &end, EXTENT_DIRTY);
|
|
|
|
if (ret)
|
|
|
|
break;
|
|
|
|
|
|
|
|
clear_extent_dirty(&log_root_tree->dirty_log_pages,
|
|
|
|
start, end, GFP_NOFS);
|
|
|
|
}
|
2008-09-06 04:13:11 +08:00
|
|
|
eb = fs_info->log_root_tree->node;
|
|
|
|
|
|
|
|
WARN_ON(btrfs_header_level(eb) != 0);
|
|
|
|
WARN_ON(btrfs_header_nritems(eb) != 0);
|
|
|
|
|
2008-09-12 03:54:42 +08:00
|
|
|
ret = btrfs_free_reserved_extent(fs_info->tree_root,
|
|
|
|
eb->start, eb->len);
|
2008-09-06 04:13:11 +08:00
|
|
|
BUG_ON(ret);
|
|
|
|
|
|
|
|
free_extent_buffer(eb);
|
|
|
|
kfree(fs_info->log_root_tree);
|
|
|
|
fs_info->log_root_tree = NULL;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
|
|
|
|
struct btrfs_fs_info *fs_info)
|
2007-04-09 22:42:37 +08:00
|
|
|
{
|
|
|
|
struct btrfs_root *root;
|
|
|
|
struct btrfs_root *tree_root = fs_info->tree_root;
|
2008-09-06 04:13:11 +08:00
|
|
|
|
|
|
|
root = kzalloc(sizeof(*root), GFP_NOFS);
|
|
|
|
if (!root)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
__setup_root(tree_root->nodesize, tree_root->leafsize,
|
|
|
|
tree_root->sectorsize, tree_root->stripesize,
|
|
|
|
root, fs_info, BTRFS_TREE_LOG_OBJECTID);
|
|
|
|
|
|
|
|
root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
|
|
|
|
root->root_key.type = BTRFS_ROOT_ITEM_KEY;
|
|
|
|
root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
|
|
|
|
root->ref_cows = 0;
|
|
|
|
|
|
|
|
root->node = btrfs_alloc_free_block(trans, root, root->leafsize,
|
2008-09-24 01:14:14 +08:00
|
|
|
0, BTRFS_TREE_LOG_OBJECTID,
|
|
|
|
trans->transid, 0, 0, 0);
|
2008-09-06 04:13:11 +08:00
|
|
|
|
|
|
|
btrfs_set_header_nritems(root->node, 0);
|
|
|
|
btrfs_set_header_level(root->node, 0);
|
|
|
|
btrfs_set_header_bytenr(root->node, root->node->start);
|
|
|
|
btrfs_set_header_generation(root->node, trans->transid);
|
|
|
|
btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
|
|
|
|
|
|
|
|
write_extent_buffer(root->node, root->fs_info->fsid,
|
|
|
|
(unsigned long)btrfs_header_fsid(root->node),
|
|
|
|
BTRFS_FSID_SIZE);
|
|
|
|
btrfs_mark_buffer_dirty(root->node);
|
|
|
|
btrfs_tree_unlock(root->node);
|
|
|
|
fs_info->log_root_tree = root;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
|
|
|
|
struct btrfs_key *location)
|
|
|
|
{
|
|
|
|
struct btrfs_root *root;
|
|
|
|
struct btrfs_fs_info *fs_info = tree_root->fs_info;
|
2007-04-09 22:42:37 +08:00
|
|
|
struct btrfs_path *path;
|
2007-10-16 04:14:19 +08:00
|
|
|
struct extent_buffer *l;
|
2007-04-11 00:13:09 +08:00
|
|
|
u64 highest_inode;
|
2008-10-30 02:49:05 +08:00
|
|
|
u64 generation;
|
2007-10-16 04:15:53 +08:00
|
|
|
u32 blocksize;
|
2007-04-09 22:42:37 +08:00
|
|
|
int ret = 0;
|
|
|
|
|
2007-06-23 02:16:25 +08:00
|
|
|
root = kzalloc(sizeof(*root), GFP_NOFS);
|
2007-06-09 21:22:25 +08:00
|
|
|
if (!root)
|
2007-04-09 22:42:37 +08:00
|
|
|
return ERR_PTR(-ENOMEM);
|
|
|
|
if (location->offset == (u64)-1) {
|
2007-10-16 04:15:53 +08:00
|
|
|
ret = find_and_setup_root(tree_root, fs_info,
|
2007-04-09 22:42:37 +08:00
|
|
|
location->objectid, root);
|
|
|
|
if (ret) {
|
|
|
|
kfree(root);
|
|
|
|
return ERR_PTR(ret);
|
|
|
|
}
|
|
|
|
goto insert;
|
|
|
|
}
|
|
|
|
|
2007-10-16 04:15:53 +08:00
|
|
|
__setup_root(tree_root->nodesize, tree_root->leafsize,
|
2007-12-01 00:30:34 +08:00
|
|
|
tree_root->sectorsize, tree_root->stripesize,
|
|
|
|
root, fs_info, location->objectid);
|
2007-04-09 22:42:37 +08:00
|
|
|
|
|
|
|
path = btrfs_alloc_path();
|
|
|
|
BUG_ON(!path);
|
|
|
|
ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
|
|
|
|
if (ret != 0) {
|
|
|
|
if (ret > 0)
|
|
|
|
ret = -ENOENT;
|
|
|
|
goto out;
|
|
|
|
}
|
2007-10-16 04:14:19 +08:00
|
|
|
l = path->nodes[0];
|
|
|
|
read_extent_buffer(l, &root->root_item,
|
|
|
|
btrfs_item_ptr_offset(l, path->slots[0]),
|
2007-04-09 22:42:37 +08:00
|
|
|
sizeof(root->root_item));
|
2007-10-19 21:22:56 +08:00
|
|
|
memcpy(&root->root_key, location, sizeof(*location));
|
2007-04-09 22:42:37 +08:00
|
|
|
ret = 0;
|
|
|
|
out:
|
|
|
|
btrfs_release_path(root, path);
|
|
|
|
btrfs_free_path(path);
|
|
|
|
if (ret) {
|
|
|
|
kfree(root);
|
|
|
|
return ERR_PTR(ret);
|
|
|
|
}
|
2008-10-30 02:49:05 +08:00
|
|
|
generation = btrfs_root_generation(&root->root_item);
|
2007-10-16 04:15:53 +08:00
|
|
|
blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
|
|
|
|
root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
|
2008-10-30 02:49:05 +08:00
|
|
|
blocksize, generation);
|
2007-04-09 22:42:37 +08:00
|
|
|
BUG_ON(!root->node);
|
|
|
|
insert:
|
2008-09-06 04:13:11 +08:00
|
|
|
if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
|
|
|
|
root->ref_cows = 1;
|
|
|
|
ret = btrfs_find_highest_inode(root, &highest_inode);
|
|
|
|
if (ret == 0) {
|
|
|
|
root->highest_inode = highest_inode;
|
|
|
|
root->last_inode_alloc = highest_inode;
|
|
|
|
}
|
2007-06-23 02:16:25 +08:00
|
|
|
}
|
|
|
|
return root;
|
|
|
|
}
|
|
|
|
|
2008-01-09 04:46:30 +08:00
|
|
|
struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
|
|
|
|
u64 root_objectid)
|
|
|
|
{
|
|
|
|
struct btrfs_root *root;
|
|
|
|
|
|
|
|
if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
|
|
|
|
return fs_info->tree_root;
|
|
|
|
if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
|
|
|
|
return fs_info->extent_root;
|
|
|
|
|
|
|
|
root = radix_tree_lookup(&fs_info->fs_roots_radix,
|
|
|
|
(unsigned long)root_objectid);
|
|
|
|
return root;
|
|
|
|
}
|
|
|
|
|
2007-12-22 05:27:24 +08:00
|
|
|
struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
|
|
|
|
struct btrfs_key *location)
|
2007-06-23 02:16:25 +08:00
|
|
|
{
|
|
|
|
struct btrfs_root *root;
|
|
|
|
int ret;
|
|
|
|
|
2007-12-22 05:27:24 +08:00
|
|
|
if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
|
|
|
|
return fs_info->tree_root;
|
|
|
|
if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
|
|
|
|
return fs_info->extent_root;
|
2008-04-26 04:53:30 +08:00
|
|
|
if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
|
|
|
|
return fs_info->chunk_root;
|
|
|
|
if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
|
|
|
|
return fs_info->dev_root;
|
2007-12-22 05:27:24 +08:00
|
|
|
|
2007-06-23 02:16:25 +08:00
|
|
|
root = radix_tree_lookup(&fs_info->fs_roots_radix,
|
|
|
|
(unsigned long)location->objectid);
|
|
|
|
if (root)
|
|
|
|
return root;
|
|
|
|
|
2008-09-06 04:13:11 +08:00
|
|
|
root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
|
2007-06-23 02:16:25 +08:00
|
|
|
if (IS_ERR(root))
|
|
|
|
return root;
|
2007-04-11 04:58:11 +08:00
|
|
|
ret = radix_tree_insert(&fs_info->fs_roots_radix,
|
|
|
|
(unsigned long)root->root_key.objectid,
|
2007-04-09 22:42:37 +08:00
|
|
|
root);
|
|
|
|
if (ret) {
|
2007-10-16 04:14:19 +08:00
|
|
|
free_extent_buffer(root->node);
|
2007-04-09 22:42:37 +08:00
|
|
|
kfree(root);
|
|
|
|
return ERR_PTR(ret);
|
|
|
|
}
|
2008-11-13 03:34:12 +08:00
|
|
|
if (!(fs_info->sb->s_flags & MS_RDONLY)) {
|
|
|
|
ret = btrfs_find_dead_roots(fs_info->tree_root,
|
|
|
|
root->root_key.objectid, root);
|
|
|
|
BUG_ON(ret);
|
|
|
|
btrfs_orphan_cleanup(root);
|
|
|
|
}
|
2007-12-22 05:27:24 +08:00
|
|
|
return root;
|
|
|
|
}
|
|
|
|
|
|
|
|
struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
|
|
|
|
struct btrfs_key *location,
|
|
|
|
const char *name, int namelen)
|
|
|
|
{
|
|
|
|
struct btrfs_root *root;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
root = btrfs_read_fs_root_no_name(fs_info, location);
|
|
|
|
if (!root)
|
|
|
|
return NULL;
|
2007-08-30 03:47:34 +08:00
|
|
|
|
2008-01-03 22:08:48 +08:00
|
|
|
if (root->in_sysfs)
|
|
|
|
return root;
|
|
|
|
|
2007-08-30 03:47:34 +08:00
|
|
|
ret = btrfs_set_root_name(root, name, namelen);
|
|
|
|
if (ret) {
|
2007-10-16 04:14:19 +08:00
|
|
|
free_extent_buffer(root->node);
|
2007-08-30 03:47:34 +08:00
|
|
|
kfree(root);
|
|
|
|
return ERR_PTR(ret);
|
|
|
|
}
|
|
|
|
|
|
|
|
ret = btrfs_sysfs_add_root(root);
|
|
|
|
if (ret) {
|
2007-10-16 04:14:19 +08:00
|
|
|
free_extent_buffer(root->node);
|
2007-08-30 03:47:34 +08:00
|
|
|
kfree(root->name);
|
|
|
|
kfree(root);
|
|
|
|
return ERR_PTR(ret);
|
|
|
|
}
|
2008-01-03 22:08:48 +08:00
|
|
|
root->in_sysfs = 1;
|
2007-04-09 22:42:37 +08:00
|
|
|
return root;
|
|
|
|
}
|
2007-10-16 04:19:22 +08:00
|
|
|
#if 0
|
|
|
|
static int add_hasher(struct btrfs_fs_info *info, char *type) {
|
|
|
|
struct btrfs_hasher *hasher;
|
|
|
|
|
|
|
|
hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
|
|
|
|
if (!hasher)
|
|
|
|
return -ENOMEM;
|
|
|
|
hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
|
|
|
|
if (!hasher->hash_tfm) {
|
|
|
|
kfree(hasher);
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
spin_lock(&info->hash_lock);
|
|
|
|
list_add(&hasher->list, &info->hashers);
|
|
|
|
spin_unlock(&info->hash_lock);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
#endif
|
2008-03-26 22:28:07 +08:00
|
|
|
|
|
|
|
static int btrfs_congested_fn(void *congested_data, int bdi_bits)
|
|
|
|
{
|
|
|
|
struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
|
|
|
|
int ret = 0;
|
|
|
|
struct list_head *cur;
|
|
|
|
struct btrfs_device *device;
|
|
|
|
struct backing_dev_info *bdi;
|
2008-11-13 22:59:36 +08:00
|
|
|
#if 0
|
2008-05-16 04:15:45 +08:00
|
|
|
if ((bdi_bits & (1 << BDI_write_congested)) &&
|
2008-08-16 03:34:15 +08:00
|
|
|
btrfs_congested_async(info, 0))
|
2008-05-16 04:15:45 +08:00
|
|
|
return 1;
|
2008-11-13 22:59:36 +08:00
|
|
|
#endif
|
2008-03-26 22:28:07 +08:00
|
|
|
list_for_each(cur, &info->fs_devices->devices) {
|
|
|
|
device = list_entry(cur, struct btrfs_device, dev_list);
|
2008-05-14 01:46:40 +08:00
|
|
|
if (!device->bdev)
|
|
|
|
continue;
|
2008-03-26 22:28:07 +08:00
|
|
|
bdi = blk_get_backing_dev_info(device->bdev);
|
|
|
|
if (bdi && bdi_congested(bdi, bdi_bits)) {
|
|
|
|
ret = 1;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2008-04-22 21:22:11 +08:00
|
|
|
/*
|
|
|
|
* this unplugs every device on the box, and it is only used when page
|
|
|
|
* is null
|
|
|
|
*/
|
|
|
|
static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
|
|
|
|
{
|
|
|
|
struct list_head *cur;
|
|
|
|
struct btrfs_device *device;
|
|
|
|
struct btrfs_fs_info *info;
|
|
|
|
|
|
|
|
info = (struct btrfs_fs_info *)bdi->unplug_io_data;
|
|
|
|
list_for_each(cur, &info->fs_devices->devices) {
|
|
|
|
device = list_entry(cur, struct btrfs_device, dev_list);
|
|
|
|
bdi = blk_get_backing_dev_info(device->bdev);
|
|
|
|
if (bdi->unplug_io_fn) {
|
|
|
|
bdi->unplug_io_fn(bdi, page);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2008-03-26 22:28:07 +08:00
|
|
|
void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
|
|
|
|
{
|
2008-04-22 21:22:11 +08:00
|
|
|
struct inode *inode;
|
2008-04-21 22:03:05 +08:00
|
|
|
struct extent_map_tree *em_tree;
|
|
|
|
struct extent_map *em;
|
2008-04-23 01:26:47 +08:00
|
|
|
struct address_space *mapping;
|
2008-04-22 21:22:11 +08:00
|
|
|
u64 offset;
|
|
|
|
|
2008-04-23 01:26:47 +08:00
|
|
|
/* the generic O_DIRECT read code does this */
|
2008-04-22 21:22:11 +08:00
|
|
|
if (!page) {
|
|
|
|
__unplug_io_fn(bdi, page);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2008-04-23 01:26:47 +08:00
|
|
|
/*
|
|
|
|
* page->mapping may change at any time. Get a consistent copy
|
|
|
|
* and use that for everything below
|
|
|
|
*/
|
|
|
|
smp_mb();
|
|
|
|
mapping = page->mapping;
|
|
|
|
if (!mapping)
|
|
|
|
return;
|
|
|
|
|
|
|
|
inode = mapping->host;
|
2008-11-11 02:08:31 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* don't do the expensive searching for a small number of
|
|
|
|
* devices
|
|
|
|
*/
|
|
|
|
if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
|
|
|
|
__unplug_io_fn(bdi, page);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2008-04-22 21:22:11 +08:00
|
|
|
offset = page_offset(page);
|
2008-03-26 22:28:07 +08:00
|
|
|
|
2008-04-21 22:03:05 +08:00
|
|
|
em_tree = &BTRFS_I(inode)->extent_tree;
|
|
|
|
spin_lock(&em_tree->lock);
|
|
|
|
em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
|
|
|
|
spin_unlock(&em_tree->lock);
|
2008-07-24 21:41:53 +08:00
|
|
|
if (!em) {
|
|
|
|
__unplug_io_fn(bdi, page);
|
2008-04-21 22:03:05 +08:00
|
|
|
return;
|
2008-07-24 21:41:53 +08:00
|
|
|
}
|
2008-04-21 22:03:05 +08:00
|
|
|
|
2008-07-24 21:41:53 +08:00
|
|
|
if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
|
|
|
|
free_extent_map(em);
|
|
|
|
__unplug_io_fn(bdi, page);
|
|
|
|
return;
|
|
|
|
}
|
2008-04-21 22:03:05 +08:00
|
|
|
offset = offset - em->start;
|
|
|
|
btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
|
|
|
|
em->block_start + offset, page);
|
|
|
|
free_extent_map(em);
|
2008-03-26 22:28:07 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
|
|
|
|
{
|
|
|
|
bdi_init(bdi);
|
2008-04-19 04:13:31 +08:00
|
|
|
bdi->ra_pages = default_backing_dev_info.ra_pages;
|
2008-03-26 22:28:07 +08:00
|
|
|
bdi->state = 0;
|
|
|
|
bdi->capabilities = default_backing_dev_info.capabilities;
|
|
|
|
bdi->unplug_io_fn = btrfs_unplug_io_fn;
|
|
|
|
bdi->unplug_io_data = info;
|
|
|
|
bdi->congested_fn = btrfs_congested_fn;
|
|
|
|
bdi->congested_data = info;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2008-04-10 04:28:12 +08:00
|
|
|
static int bio_ready_for_csum(struct bio *bio)
|
|
|
|
{
|
|
|
|
u64 length = 0;
|
|
|
|
u64 buf_len = 0;
|
|
|
|
u64 start = 0;
|
|
|
|
struct page *page;
|
|
|
|
struct extent_io_tree *io_tree = NULL;
|
|
|
|
struct btrfs_fs_info *info = NULL;
|
|
|
|
struct bio_vec *bvec;
|
|
|
|
int i;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
bio_for_each_segment(bvec, bio, i) {
|
|
|
|
page = bvec->bv_page;
|
|
|
|
if (page->private == EXTENT_PAGE_PRIVATE) {
|
|
|
|
length += bvec->bv_len;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
if (!page->private) {
|
|
|
|
length += bvec->bv_len;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
length = bvec->bv_len;
|
|
|
|
buf_len = page->private >> 2;
|
|
|
|
start = page_offset(page) + bvec->bv_offset;
|
|
|
|
io_tree = &BTRFS_I(page->mapping->host)->io_tree;
|
|
|
|
info = BTRFS_I(page->mapping->host)->root->fs_info;
|
|
|
|
}
|
|
|
|
/* are we fully contained in this bio? */
|
|
|
|
if (buf_len <= length)
|
|
|
|
return 1;
|
|
|
|
|
|
|
|
ret = extent_range_uptodate(io_tree, start + length,
|
|
|
|
start + buf_len - 1);
|
|
|
|
if (ret == 1)
|
|
|
|
return ret;
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2008-06-12 04:50:36 +08:00
|
|
|
/*
|
|
|
|
* called by the kthread helper functions to finally call the bio end_io
|
|
|
|
* functions. This is where read checksum verification actually happens
|
|
|
|
*/
|
|
|
|
static void end_workqueue_fn(struct btrfs_work *work)
|
2008-04-10 04:28:12 +08:00
|
|
|
{
|
|
|
|
struct bio *bio;
|
2008-06-12 04:50:36 +08:00
|
|
|
struct end_io_wq *end_io_wq;
|
|
|
|
struct btrfs_fs_info *fs_info;
|
2008-04-10 04:28:12 +08:00
|
|
|
int error;
|
|
|
|
|
2008-06-12 04:50:36 +08:00
|
|
|
end_io_wq = container_of(work, struct end_io_wq, work);
|
|
|
|
bio = end_io_wq->bio;
|
|
|
|
fs_info = end_io_wq->info;
|
2008-04-10 04:28:12 +08:00
|
|
|
|
2008-06-12 04:50:36 +08:00
|
|
|
/* metadata bios are special because the whole tree block must
|
|
|
|
* be checksummed at once. This makes sure the entire block is in
|
|
|
|
* ram and up to date before trying to verify things. For
|
|
|
|
* blocksize <= pagesize, it is basically a noop
|
|
|
|
*/
|
|
|
|
if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
|
|
|
|
btrfs_queue_worker(&fs_info->endio_workers,
|
|
|
|
&end_io_wq->work);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
error = end_io_wq->error;
|
|
|
|
bio->bi_private = end_io_wq->private;
|
|
|
|
bio->bi_end_io = end_io_wq->end_io;
|
|
|
|
kfree(end_io_wq);
|
|
|
|
bio_endio(bio, error);
|
2008-04-16 23:14:51 +08:00
|
|
|
}
|
|
|
|
|
2008-06-26 04:01:31 +08:00
|
|
|
static int cleaner_kthread(void *arg)
|
|
|
|
{
|
|
|
|
struct btrfs_root *root = arg;
|
|
|
|
|
|
|
|
do {
|
|
|
|
smp_mb();
|
|
|
|
if (root->fs_info->closing)
|
|
|
|
break;
|
|
|
|
|
|
|
|
vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
|
|
|
|
mutex_lock(&root->fs_info->cleaner_mutex);
|
|
|
|
btrfs_clean_old_snapshots(root);
|
|
|
|
mutex_unlock(&root->fs_info->cleaner_mutex);
|
|
|
|
|
|
|
|
if (freezing(current)) {
|
|
|
|
refrigerator();
|
|
|
|
} else {
|
|
|
|
smp_mb();
|
|
|
|
if (root->fs_info->closing)
|
|
|
|
break;
|
|
|
|
set_current_state(TASK_INTERRUPTIBLE);
|
|
|
|
schedule();
|
|
|
|
__set_current_state(TASK_RUNNING);
|
|
|
|
}
|
|
|
|
} while (!kthread_should_stop());
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int transaction_kthread(void *arg)
|
|
|
|
{
|
|
|
|
struct btrfs_root *root = arg;
|
|
|
|
struct btrfs_trans_handle *trans;
|
|
|
|
struct btrfs_transaction *cur;
|
|
|
|
unsigned long now;
|
|
|
|
unsigned long delay;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
do {
|
|
|
|
smp_mb();
|
|
|
|
if (root->fs_info->closing)
|
|
|
|
break;
|
|
|
|
|
|
|
|
delay = HZ * 30;
|
|
|
|
vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
|
|
|
|
mutex_lock(&root->fs_info->transaction_kthread_mutex);
|
|
|
|
|
2008-07-30 04:15:18 +08:00
|
|
|
if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
|
|
|
|
printk("btrfs: total reference cache size %Lu\n",
|
|
|
|
root->fs_info->total_ref_cache_size);
|
|
|
|
}
|
2008-07-29 03:32:19 +08:00
|
|
|
|
2008-06-26 04:01:31 +08:00
|
|
|
mutex_lock(&root->fs_info->trans_mutex);
|
|
|
|
cur = root->fs_info->running_transaction;
|
|
|
|
if (!cur) {
|
|
|
|
mutex_unlock(&root->fs_info->trans_mutex);
|
|
|
|
goto sleep;
|
|
|
|
}
|
2008-07-29 03:32:19 +08:00
|
|
|
|
2008-06-26 04:01:31 +08:00
|
|
|
now = get_seconds();
|
|
|
|
if (now < cur->start_time || now - cur->start_time < 30) {
|
|
|
|
mutex_unlock(&root->fs_info->trans_mutex);
|
|
|
|
delay = HZ * 5;
|
|
|
|
goto sleep;
|
|
|
|
}
|
|
|
|
mutex_unlock(&root->fs_info->trans_mutex);
|
|
|
|
trans = btrfs_start_transaction(root, 1);
|
|
|
|
ret = btrfs_commit_transaction(trans, root);
|
|
|
|
sleep:
|
|
|
|
wake_up_process(root->fs_info->cleaner_kthread);
|
|
|
|
mutex_unlock(&root->fs_info->transaction_kthread_mutex);
|
|
|
|
|
|
|
|
if (freezing(current)) {
|
|
|
|
refrigerator();
|
|
|
|
} else {
|
|
|
|
if (root->fs_info->closing)
|
|
|
|
break;
|
|
|
|
set_current_state(TASK_INTERRUPTIBLE);
|
|
|
|
schedule_timeout(delay);
|
|
|
|
__set_current_state(TASK_RUNNING);
|
|
|
|
}
|
|
|
|
} while (!kthread_should_stop());
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2008-03-25 03:02:07 +08:00
|
|
|
struct btrfs_root *open_ctree(struct super_block *sb,
|
2008-05-14 01:46:40 +08:00
|
|
|
struct btrfs_fs_devices *fs_devices,
|
|
|
|
char *options)
|
2007-03-21 23:12:56 +08:00
|
|
|
{
|
2007-10-16 04:15:53 +08:00
|
|
|
u32 sectorsize;
|
|
|
|
u32 nodesize;
|
|
|
|
u32 leafsize;
|
|
|
|
u32 blocksize;
|
2007-12-01 00:30:34 +08:00
|
|
|
u32 stripesize;
|
2008-10-30 02:49:05 +08:00
|
|
|
u64 generation;
|
2008-05-07 23:43:44 +08:00
|
|
|
struct buffer_head *bh;
|
2008-09-06 04:13:11 +08:00
|
|
|
struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
|
2007-03-23 00:13:20 +08:00
|
|
|
GFP_NOFS);
|
2008-09-06 04:13:11 +08:00
|
|
|
struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
|
2007-03-23 00:13:20 +08:00
|
|
|
GFP_NOFS);
|
2008-04-04 04:29:03 +08:00
|
|
|
struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
|
2007-03-23 00:13:20 +08:00
|
|
|
GFP_NOFS);
|
2008-09-06 04:13:11 +08:00
|
|
|
struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
|
2008-03-25 03:01:56 +08:00
|
|
|
GFP_NOFS);
|
2008-09-06 04:13:11 +08:00
|
|
|
struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
|
2008-03-25 03:01:56 +08:00
|
|
|
GFP_NOFS);
|
2008-09-06 04:13:11 +08:00
|
|
|
struct btrfs_root *log_tree_root;
|
|
|
|
|
2007-02-02 22:18:22 +08:00
|
|
|
int ret;
|
2008-04-01 23:21:34 +08:00
|
|
|
int err = -EINVAL;
|
2008-06-12 09:47:56 +08:00
|
|
|
|
2007-04-02 22:50:19 +08:00
|
|
|
struct btrfs_super_block *disk_super;
|
2008-04-04 04:29:03 +08:00
|
|
|
|
2008-10-02 07:09:04 +08:00
|
|
|
if (!extent_root || !tree_root || !fs_info ||
|
|
|
|
!chunk_root || !dev_root) {
|
2007-06-12 18:35:45 +08:00
|
|
|
err = -ENOMEM;
|
|
|
|
goto fail;
|
|
|
|
}
|
2007-04-09 22:42:37 +08:00
|
|
|
INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
|
2007-04-20 09:01:03 +08:00
|
|
|
INIT_LIST_HEAD(&fs_info->trans_list);
|
2007-06-09 06:11:48 +08:00
|
|
|
INIT_LIST_HEAD(&fs_info->dead_roots);
|
2007-10-16 04:19:22 +08:00
|
|
|
INIT_LIST_HEAD(&fs_info->hashers);
|
2008-08-05 11:17:27 +08:00
|
|
|
INIT_LIST_HEAD(&fs_info->delalloc_inodes);
|
2007-10-16 04:19:22 +08:00
|
|
|
spin_lock_init(&fs_info->hash_lock);
|
2007-12-22 05:27:21 +08:00
|
|
|
spin_lock_init(&fs_info->delalloc_lock);
|
2008-01-15 21:40:48 +08:00
|
|
|
spin_lock_init(&fs_info->new_trans_lock);
|
2008-07-29 03:32:19 +08:00
|
|
|
spin_lock_init(&fs_info->ref_cache_lock);
|
2007-10-16 04:19:22 +08:00
|
|
|
|
2007-08-30 03:47:34 +08:00
|
|
|
init_completion(&fs_info->kobj_unregister);
|
2007-03-21 02:38:32 +08:00
|
|
|
fs_info->tree_root = tree_root;
|
|
|
|
fs_info->extent_root = extent_root;
|
2008-03-25 03:01:56 +08:00
|
|
|
fs_info->chunk_root = chunk_root;
|
|
|
|
fs_info->dev_root = dev_root;
|
2008-03-25 03:02:07 +08:00
|
|
|
fs_info->fs_devices = fs_devices;
|
2008-03-25 03:01:56 +08:00
|
|
|
INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
|
2008-03-25 03:01:59 +08:00
|
|
|
INIT_LIST_HEAD(&fs_info->space_info);
|
2008-03-25 03:01:56 +08:00
|
|
|
btrfs_mapping_init(&fs_info->mapping_tree);
|
2008-05-16 04:15:45 +08:00
|
|
|
atomic_set(&fs_info->nr_async_submits, 0);
|
2008-11-07 11:02:51 +08:00
|
|
|
atomic_set(&fs_info->async_delalloc_pages, 0);
|
2008-09-29 23:19:10 +08:00
|
|
|
atomic_set(&fs_info->async_submit_draining, 0);
|
2008-08-16 03:34:15 +08:00
|
|
|
atomic_set(&fs_info->nr_async_bios, 0);
|
2008-06-26 04:01:30 +08:00
|
|
|
atomic_set(&fs_info->throttles, 0);
|
2008-07-30 04:15:18 +08:00
|
|
|
atomic_set(&fs_info->throttle_gen, 0);
|
2007-03-23 00:13:20 +08:00
|
|
|
fs_info->sb = sb;
|
2007-12-18 09:14:04 +08:00
|
|
|
fs_info->max_extent = (u64)-1;
|
2008-01-30 05:03:38 +08:00
|
|
|
fs_info->max_inline = 8192 * 1024;
|
2008-03-26 22:28:07 +08:00
|
|
|
setup_bdi(fs_info, &fs_info->bdi);
|
2007-03-29 01:57:48 +08:00
|
|
|
fs_info->btree_inode = new_inode(sb);
|
|
|
|
fs_info->btree_inode->i_ino = 1;
|
2007-04-02 22:50:19 +08:00
|
|
|
fs_info->btree_inode->i_nlink = 1;
|
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
|
|
|
|
2008-06-12 09:47:56 +08:00
|
|
|
fs_info->thread_pool_size = min(num_online_cpus() + 2, 8);
|
2008-04-19 02:17:20 +08:00
|
|
|
|
2008-07-24 23:57:52 +08:00
|
|
|
INIT_LIST_HEAD(&fs_info->ordered_extents);
|
|
|
|
spin_lock_init(&fs_info->ordered_extent_lock);
|
|
|
|
|
2008-05-07 23:43:44 +08:00
|
|
|
sb->s_blocksize = 4096;
|
|
|
|
sb->s_blocksize_bits = blksize_bits(4096);
|
|
|
|
|
2008-04-19 02:17:20 +08:00
|
|
|
/*
|
|
|
|
* we set the i_size on the btree inode to the max possible int.
|
|
|
|
* the real end of the address space is determined by all of
|
|
|
|
* the devices in the system
|
|
|
|
*/
|
|
|
|
fs_info->btree_inode->i_size = OFFSET_MAX;
|
2007-03-29 01:57:48 +08:00
|
|
|
fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
|
2008-03-26 22:28:07 +08:00
|
|
|
fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
|
|
|
|
|
2008-01-25 05:13:08 +08:00
|
|
|
extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
|
2007-10-16 04:14:19 +08:00
|
|
|
fs_info->btree_inode->i_mapping,
|
|
|
|
GFP_NOFS);
|
2008-01-25 05:13:08 +08:00
|
|
|
extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
|
|
|
|
GFP_NOFS);
|
|
|
|
|
|
|
|
BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
|
2007-11-08 10:08:01 +08:00
|
|
|
|
Btrfs: free space accounting redo
1) replace the per fs_info extent_io_tree that tracked free space with two
rb-trees per block group to track free space areas via offset and size. The
reason to do this is because most allocations come with a hint byte where to
start, so we can usually find a chunk of free space at that hint byte to satisfy
the allocation and get good space packing. If we cannot find free space at or
after the given offset we fall back on looking for a chunk of the given size as
close to that given offset as possible. When we fall back on the size search we
also try to find a slot as close to the size we want as possible, to avoid
breaking small chunks off of huge areas if possible.
2) remove the extent_io_tree that tracked the block group cache from fs_info and
replaced it with an rb-tree thats tracks block group cache via offset. also
added a per space_info list that tracks the block group cache for the particular
space so we can lookup related block groups easily.
3) cleaned up the allocation code to make it a little easier to read and a
little less complicated. Basically there are 3 steps, first look from our
provided hint. If we couldn't find from that given hint, start back at our
original search start and look for space from there. If that fails try to
allocate space if we can and start looking again. If not we're screwed and need
to start over again.
4) small fixes. there were some issues in volumes.c where we wouldn't allocate
the rest of the disk. fixed cow_file_range to actually pass the alloc_hint,
which has helped a good bit in making the fs_mark test I run have semi-normal
results as we run out of space. Generally with data allocations we don't track
where we last allocated from, so everytime we did a data allocation we'd search
through every block group that we have looking for free space. Now searching a
block group with no free space isn't terribly time consuming, it was causing a
slight degradation as we got more data block groups. The alloc_hint has fixed
this slight degredation and made things semi-normal.
There is still one nagging problem I'm working on where we will get ENOSPC when
there is definitely plenty of space. This only happens with metadata
allocations, and only when we are almost full. So you generally hit the 85%
mark first, but sometimes you'll hit the BUG before you hit the 85% wall. I'm
still tracking it down, but until then this seems to be pretty stable and make a
significant performance gain.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-09-24 01:14:11 +08:00
|
|
|
spin_lock_init(&fs_info->block_group_cache_lock);
|
|
|
|
fs_info->block_group_cache_tree.rb_node = NULL;
|
|
|
|
|
2008-01-25 05:13:08 +08:00
|
|
|
extent_io_tree_init(&fs_info->pinned_extents,
|
2007-10-16 04:15:26 +08:00
|
|
|
fs_info->btree_inode->i_mapping, GFP_NOFS);
|
2008-01-25 05:13:08 +08:00
|
|
|
extent_io_tree_init(&fs_info->pending_del,
|
2007-10-16 04:15:26 +08:00
|
|
|
fs_info->btree_inode->i_mapping, GFP_NOFS);
|
2008-01-25 05:13:08 +08:00
|
|
|
extent_io_tree_init(&fs_info->extent_ins,
|
2007-10-16 04:15:26 +08:00
|
|
|
fs_info->btree_inode->i_mapping, GFP_NOFS);
|
2007-04-21 01:16:02 +08:00
|
|
|
fs_info->do_barriers = 1;
|
2008-01-18 23:54:22 +08:00
|
|
|
|
Btrfs: update space balancing code
This patch updates the space balancing code to utilize the new
backref format. Before, btrfs-vol -b would break any COW links
on data blocks or metadata. This was slow and caused the amount
of space used to explode if a large number of snapshots were present.
The new code can keeps the sharing of all data extents and
most of the tree blocks.
To maintain the sharing of data extents, the space balance code uses
a seperate inode hold data extent pointers, then updates the references
to point to the new location.
To maintain the sharing of tree blocks, the space balance code uses
reloc trees to relocate tree blocks in reference counted roots.
There is one reloc tree for each subvol, and all reloc trees share
same root key objectid. Reloc trees are snapshots of the latest
committed roots of subvols (root->commit_root).
To relocate a tree block referenced by a subvol, there are two steps.
COW the block through subvol's reloc tree, then update block pointer in
the subvol to point to the new block. Since all reloc trees share
same root key objectid, doing special handing for tree blocks
owned by them is easy. Once a tree block has been COWed in one
reloc tree, we can use the resulting new block directly when the
same block is required to COW again through other reloc trees.
In this way, relocated tree blocks are shared between reloc trees,
so they are also shared between subvols.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-09-26 22:09:34 +08:00
|
|
|
INIT_LIST_HEAD(&fs_info->dead_reloc_roots);
|
|
|
|
btrfs_leaf_ref_tree_init(&fs_info->reloc_ref_tree);
|
2008-09-26 22:04:53 +08:00
|
|
|
btrfs_leaf_ref_tree_init(&fs_info->shared_ref_tree);
|
|
|
|
|
2007-04-09 22:42:37 +08:00
|
|
|
BTRFS_I(fs_info->btree_inode)->root = tree_root;
|
|
|
|
memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
|
|
|
|
sizeof(struct btrfs_key));
|
2007-03-30 20:47:31 +08:00
|
|
|
insert_inode_hash(fs_info->btree_inode);
|
2007-06-12 18:35:45 +08:00
|
|
|
|
2007-03-23 03:59:16 +08:00
|
|
|
mutex_init(&fs_info->trans_mutex);
|
2008-09-06 04:13:11 +08:00
|
|
|
mutex_init(&fs_info->tree_log_mutex);
|
2008-06-26 04:01:30 +08:00
|
|
|
mutex_init(&fs_info->drop_mutex);
|
Btrfs: nuke fs wide allocation mutex V2
This patch removes the giant fs_info->alloc_mutex and replaces it with a bunch
of little locks.
There is now a pinned_mutex, which is used when messing with the pinned_extents
extent io tree, and the extent_ins_mutex which is used with the pending_del and
extent_ins extent io trees.
The locking for the extent tree stuff was inspired by a patch that Yan Zheng
wrote to fix a race condition, I cleaned it up some and changed the locking
around a little bit, but the idea remains the same. Basically instead of
holding the extent_ins_mutex throughout the processing of an extent on the
extent_ins or pending_del trees, we just hold it while we're searching and when
we clear the bits on those trees, and lock the extent for the duration of the
operations on the extent.
Also to keep from getting hung up waiting to lock an extent, I've added a
try_lock_extent so if we cannot lock the extent, move on to the next one in the
tree and we'll come back to that one. I have tested this heavily and it does
not appear to break anything. This has to be applied on top of my
find_free_extent redo patch.
I tested this patch on top of Yan's space reblancing code and it worked fine.
The only thing that has changed since the last version is I pulled out all my
debugging stuff, apparently I forgot to run guilt refresh before I sent the
last patch out. Thank you,
Signed-off-by: Josef Bacik <jbacik@redhat.com>
2008-10-30 02:49:05 +08:00
|
|
|
mutex_init(&fs_info->extent_ins_mutex);
|
|
|
|
mutex_init(&fs_info->pinned_mutex);
|
2008-06-26 04:01:30 +08:00
|
|
|
mutex_init(&fs_info->chunk_mutex);
|
2008-06-26 04:01:31 +08:00
|
|
|
mutex_init(&fs_info->transaction_kthread_mutex);
|
|
|
|
mutex_init(&fs_info->cleaner_mutex);
|
2008-07-09 02:19:17 +08:00
|
|
|
mutex_init(&fs_info->volume_mutex);
|
Btrfs: update space balancing code
This patch updates the space balancing code to utilize the new
backref format. Before, btrfs-vol -b would break any COW links
on data blocks or metadata. This was slow and caused the amount
of space used to explode if a large number of snapshots were present.
The new code can keeps the sharing of all data extents and
most of the tree blocks.
To maintain the sharing of data extents, the space balance code uses
a seperate inode hold data extent pointers, then updates the references
to point to the new location.
To maintain the sharing of tree blocks, the space balance code uses
reloc trees to relocate tree blocks in reference counted roots.
There is one reloc tree for each subvol, and all reloc trees share
same root key objectid. Reloc trees are snapshots of the latest
committed roots of subvols (root->commit_root).
To relocate a tree block referenced by a subvol, there are two steps.
COW the block through subvol's reloc tree, then update block pointer in
the subvol to point to the new block. Since all reloc trees share
same root key objectid, doing special handing for tree blocks
owned by them is easy. Once a tree block has been COWed in one
reloc tree, we can use the resulting new block directly when the
same block is required to COW again through other reloc trees.
In this way, relocated tree blocks are shared between reloc trees,
so they are also shared between subvols.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-09-26 22:09:34 +08:00
|
|
|
mutex_init(&fs_info->tree_reloc_mutex);
|
2008-07-18 00:53:50 +08:00
|
|
|
init_waitqueue_head(&fs_info->transaction_throttle);
|
2008-07-18 00:54:14 +08:00
|
|
|
init_waitqueue_head(&fs_info->transaction_wait);
|
2008-08-16 03:34:17 +08:00
|
|
|
init_waitqueue_head(&fs_info->async_submit_wait);
|
2008-09-06 04:13:11 +08:00
|
|
|
init_waitqueue_head(&fs_info->tree_log_wait);
|
|
|
|
atomic_set(&fs_info->tree_log_commit, 0);
|
|
|
|
atomic_set(&fs_info->tree_log_writers, 0);
|
|
|
|
fs_info->tree_log_transid = 0;
|
2007-03-14 04:47:54 +08:00
|
|
|
|
2007-10-16 04:19:22 +08:00
|
|
|
#if 0
|
|
|
|
ret = add_hasher(fs_info, "crc32c");
|
|
|
|
if (ret) {
|
|
|
|
printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
|
|
|
|
err = -ENOMEM;
|
|
|
|
goto fail_iput;
|
|
|
|
}
|
|
|
|
#endif
|
2008-03-25 03:01:56 +08:00
|
|
|
__setup_root(4096, 4096, 4096, 4096, tree_root,
|
2007-04-02 22:50:19 +08:00
|
|
|
fs_info, BTRFS_ROOT_TREE_OBJECTID);
|
2007-04-12 03:53:25 +08:00
|
|
|
|
2007-03-29 01:57:48 +08:00
|
|
|
|
2008-05-07 23:43:44 +08:00
|
|
|
bh = __bread(fs_devices->latest_bdev,
|
|
|
|
BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
|
|
|
|
if (!bh)
|
2007-06-12 18:35:45 +08:00
|
|
|
goto fail_iput;
|
|
|
|
|
2008-05-07 23:43:44 +08:00
|
|
|
memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
|
|
|
|
brelse(bh);
|
2007-10-16 04:14:19 +08:00
|
|
|
|
2008-05-07 23:43:44 +08:00
|
|
|
memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
|
2008-03-25 03:01:56 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
disk_super = &fs_info->super_copy;
|
2007-04-09 22:42:37 +08:00
|
|
|
if (!btrfs_super_root(disk_super))
|
2007-06-12 18:35:45 +08:00
|
|
|
goto fail_sb_buffer;
|
2007-04-09 22:42:37 +08:00
|
|
|
|
2008-06-10 22:40:29 +08:00
|
|
|
err = btrfs_parse_options(tree_root, options);
|
|
|
|
if (err)
|
|
|
|
goto fail_sb_buffer;
|
2008-05-14 01:46:40 +08:00
|
|
|
|
2008-06-12 09:47:56 +08:00
|
|
|
/*
|
|
|
|
* we need to start all the end_io workers up front because the
|
|
|
|
* queue work function gets called at interrupt time, and so it
|
|
|
|
* cannot dynamically grow.
|
|
|
|
*/
|
2008-08-16 03:34:16 +08:00
|
|
|
btrfs_init_workers(&fs_info->workers, "worker",
|
|
|
|
fs_info->thread_pool_size);
|
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
|
|
|
|
2008-11-07 11:02:51 +08:00
|
|
|
btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
|
|
|
|
fs_info->thread_pool_size);
|
|
|
|
|
2008-08-16 03:34:16 +08:00
|
|
|
btrfs_init_workers(&fs_info->submit_workers, "submit",
|
2008-08-16 03:34:14 +08:00
|
|
|
min_t(u64, fs_devices->num_devices,
|
|
|
|
fs_info->thread_pool_size));
|
2008-08-01 03:42:53 +08:00
|
|
|
|
|
|
|
/* a higher idle thresh on the submit workers makes it much more
|
|
|
|
* likely that bios will be send down in a sane order to the
|
|
|
|
* devices
|
|
|
|
*/
|
|
|
|
fs_info->submit_workers.idle_thresh = 64;
|
2008-08-16 03:34:18 +08:00
|
|
|
|
2008-11-07 11:02:51 +08:00
|
|
|
fs_info->workers.idle_thresh = 16;
|
Btrfs: Add ordered async work queues
Btrfs uses kernel threads to create async work queues for cpu intensive
operations such as checksumming and decompression. These work well,
but they make it difficult to keep IO order intact.
A single writepages call from pdflush or fsync will turn into a number
of bios, and each bio is checksummed in parallel. Once the checksum is
computed, the bio is sent down to the disk, and since we don't control
the order in which the parallel operations happen, they might go down to
the disk in almost any order.
The code deals with this somewhat by having deep work queues for a single
kernel thread, making it very likely that a single thread will process all
the bios for a single inode.
This patch introduces an explicitly ordered work queue. As work structs
are placed into the queue they are put onto the tail of a list. They have
three callbacks:
->func (cpu intensive processing here)
->ordered_func (order sensitive processing here)
->ordered_free (free the work struct, all processing is done)
The work struct has three callbacks. The func callback does the cpu intensive
work, and when it completes the work struct is marked as done.
Every time a work struct completes, the list is checked to see if the head
is marked as done. If so the ordered_func callback is used to do the
order sensitive processing and the ordered_free callback is used to do
any cleanup. Then we loop back and check the head of the list again.
This patch also changes the checksumming code to use the ordered workqueues.
One a 4 drive array, it increases streaming writes from 280MB/s to 350MB/s.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-11-07 11:03:00 +08:00
|
|
|
fs_info->workers.ordered = 1;
|
2008-08-01 03:42:53 +08:00
|
|
|
|
2008-11-07 11:02:51 +08:00
|
|
|
fs_info->delalloc_workers.idle_thresh = 2;
|
|
|
|
fs_info->delalloc_workers.ordered = 1;
|
|
|
|
|
2008-08-16 03:34:16 +08:00
|
|
|
btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
|
|
|
|
btrfs_init_workers(&fs_info->endio_workers, "endio",
|
|
|
|
fs_info->thread_pool_size);
|
|
|
|
btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
|
2008-07-18 00:53:50 +08:00
|
|
|
fs_info->thread_pool_size);
|
2008-08-01 03:42:53 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* endios are largely parallel and should have a very
|
|
|
|
* low idle thresh
|
|
|
|
*/
|
|
|
|
fs_info->endio_workers.idle_thresh = 4;
|
2008-09-12 03:53:37 +08:00
|
|
|
fs_info->endio_write_workers.idle_thresh = 64;
|
2008-08-01 03:42:53 +08:00
|
|
|
|
2008-06-12 09:47:56 +08:00
|
|
|
btrfs_start_workers(&fs_info->workers, 1);
|
2008-06-13 02:46:17 +08:00
|
|
|
btrfs_start_workers(&fs_info->submit_workers, 1);
|
2008-11-07 11:02:51 +08:00
|
|
|
btrfs_start_workers(&fs_info->delalloc_workers, 1);
|
2008-07-18 00:53:51 +08:00
|
|
|
btrfs_start_workers(&fs_info->fixup_workers, 1);
|
2008-06-12 09:47:56 +08:00
|
|
|
btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
|
2008-07-18 00:53:50 +08:00
|
|
|
btrfs_start_workers(&fs_info->endio_write_workers,
|
|
|
|
fs_info->thread_pool_size);
|
2008-06-12 09:47:56 +08:00
|
|
|
|
2008-06-10 22:40:29 +08:00
|
|
|
err = -EINVAL;
|
2008-05-14 04:03:06 +08:00
|
|
|
if (btrfs_super_num_devices(disk_super) > fs_devices->open_devices) {
|
2008-03-25 03:02:07 +08:00
|
|
|
printk("Btrfs: wanted %llu devices, but found %llu\n",
|
|
|
|
(unsigned long long)btrfs_super_num_devices(disk_super),
|
2008-05-14 04:03:06 +08:00
|
|
|
(unsigned long long)fs_devices->open_devices);
|
2008-05-14 01:46:40 +08:00
|
|
|
if (btrfs_test_opt(tree_root, DEGRADED))
|
|
|
|
printk("continuing in degraded mode\n");
|
|
|
|
else {
|
|
|
|
goto fail_sb_buffer;
|
|
|
|
}
|
2008-03-25 03:02:07 +08:00
|
|
|
}
|
2008-05-14 01:46:40 +08:00
|
|
|
|
2008-04-19 04:13:31 +08:00
|
|
|
fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
|
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
|
|
|
fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
|
|
|
|
4 * 1024 * 1024 / PAGE_CACHE_SIZE);
|
2008-04-19 04:13:31 +08:00
|
|
|
|
2007-10-16 04:15:53 +08:00
|
|
|
nodesize = btrfs_super_nodesize(disk_super);
|
|
|
|
leafsize = btrfs_super_leafsize(disk_super);
|
|
|
|
sectorsize = btrfs_super_sectorsize(disk_super);
|
2007-12-01 00:30:34 +08:00
|
|
|
stripesize = btrfs_super_stripesize(disk_super);
|
2007-10-16 04:15:53 +08:00
|
|
|
tree_root->nodesize = nodesize;
|
|
|
|
tree_root->leafsize = leafsize;
|
|
|
|
tree_root->sectorsize = sectorsize;
|
2007-12-01 00:30:34 +08:00
|
|
|
tree_root->stripesize = stripesize;
|
2008-05-07 23:43:44 +08:00
|
|
|
|
|
|
|
sb->s_blocksize = sectorsize;
|
|
|
|
sb->s_blocksize_bits = blksize_bits(sectorsize);
|
2007-10-16 04:15:53 +08:00
|
|
|
|
2007-06-12 18:35:45 +08:00
|
|
|
if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
|
|
|
|
sizeof(disk_super->magic))) {
|
|
|
|
printk("btrfs: valid FS not found on %s\n", sb->s_id);
|
|
|
|
goto fail_sb_buffer;
|
|
|
|
}
|
2007-10-16 04:19:22 +08:00
|
|
|
|
2008-06-26 04:01:30 +08:00
|
|
|
mutex_lock(&fs_info->chunk_mutex);
|
2008-03-25 03:01:56 +08:00
|
|
|
ret = btrfs_read_sys_array(tree_root);
|
2008-06-26 04:01:30 +08:00
|
|
|
mutex_unlock(&fs_info->chunk_mutex);
|
2008-04-25 21:04:37 +08:00
|
|
|
if (ret) {
|
|
|
|
printk("btrfs: failed to read the system array on %s\n",
|
|
|
|
sb->s_id);
|
|
|
|
goto fail_sys_array;
|
|
|
|
}
|
2008-03-25 03:01:56 +08:00
|
|
|
|
|
|
|
blocksize = btrfs_level_size(tree_root,
|
|
|
|
btrfs_super_chunk_root_level(disk_super));
|
2008-10-30 02:49:05 +08:00
|
|
|
generation = btrfs_super_chunk_root_generation(disk_super);
|
2008-03-25 03:01:56 +08:00
|
|
|
|
|
|
|
__setup_root(nodesize, leafsize, sectorsize, stripesize,
|
|
|
|
chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
|
|
|
|
|
|
|
|
chunk_root->node = read_tree_block(chunk_root,
|
|
|
|
btrfs_super_chunk_root(disk_super),
|
2008-10-30 02:49:05 +08:00
|
|
|
blocksize, generation);
|
2008-03-25 03:01:56 +08:00
|
|
|
BUG_ON(!chunk_root->node);
|
|
|
|
|
2008-04-16 03:41:47 +08:00
|
|
|
read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
|
|
|
|
(unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
|
|
|
|
BTRFS_UUID_SIZE);
|
|
|
|
|
2008-06-26 04:01:30 +08:00
|
|
|
mutex_lock(&fs_info->chunk_mutex);
|
2008-03-25 03:01:56 +08:00
|
|
|
ret = btrfs_read_chunk_tree(chunk_root);
|
2008-06-26 04:01:30 +08:00
|
|
|
mutex_unlock(&fs_info->chunk_mutex);
|
2008-03-25 03:01:56 +08:00
|
|
|
BUG_ON(ret);
|
|
|
|
|
2008-05-14 01:46:40 +08:00
|
|
|
btrfs_close_extra_devices(fs_devices);
|
|
|
|
|
2007-10-16 04:15:53 +08:00
|
|
|
blocksize = btrfs_level_size(tree_root,
|
|
|
|
btrfs_super_root_level(disk_super));
|
2008-10-30 02:49:05 +08:00
|
|
|
generation = btrfs_super_generation(disk_super);
|
2008-03-25 03:01:56 +08:00
|
|
|
|
2007-03-23 00:13:20 +08:00
|
|
|
tree_root->node = read_tree_block(tree_root,
|
2007-10-16 04:15:53 +08:00
|
|
|
btrfs_super_root(disk_super),
|
2008-10-30 02:49:05 +08:00
|
|
|
blocksize, generation);
|
2007-06-12 18:35:45 +08:00
|
|
|
if (!tree_root->node)
|
|
|
|
goto fail_sb_buffer;
|
2007-03-14 04:47:54 +08:00
|
|
|
|
2007-10-16 04:15:53 +08:00
|
|
|
|
|
|
|
ret = find_and_setup_root(tree_root, fs_info,
|
2007-03-23 00:13:20 +08:00
|
|
|
BTRFS_EXTENT_TREE_OBJECTID, extent_root);
|
2008-03-25 03:01:56 +08:00
|
|
|
if (ret)
|
2007-06-12 18:35:45 +08:00
|
|
|
goto fail_tree_root;
|
2008-03-25 03:01:56 +08:00
|
|
|
extent_root->track_dirty = 1;
|
|
|
|
|
|
|
|
ret = find_and_setup_root(tree_root, fs_info,
|
|
|
|
BTRFS_DEV_TREE_OBJECTID, dev_root);
|
|
|
|
dev_root->track_dirty = 1;
|
|
|
|
|
|
|
|
if (ret)
|
|
|
|
goto fail_extent_root;
|
2007-03-14 04:47:54 +08:00
|
|
|
|
2007-04-27 04:46:15 +08:00
|
|
|
btrfs_read_block_groups(extent_root);
|
|
|
|
|
2008-11-13 03:34:12 +08:00
|
|
|
fs_info->generation = generation + 1;
|
|
|
|
fs_info->last_trans_committed = generation;
|
2008-04-05 03:40:00 +08:00
|
|
|
fs_info->data_alloc_profile = (u64)-1;
|
|
|
|
fs_info->metadata_alloc_profile = (u64)-1;
|
|
|
|
fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
|
2008-06-26 04:01:31 +08:00
|
|
|
fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
|
|
|
|
"btrfs-cleaner");
|
|
|
|
if (!fs_info->cleaner_kthread)
|
|
|
|
goto fail_extent_root;
|
|
|
|
|
|
|
|
fs_info->transaction_kthread = kthread_run(transaction_kthread,
|
|
|
|
tree_root,
|
|
|
|
"btrfs-transaction");
|
|
|
|
if (!fs_info->transaction_kthread)
|
2008-06-26 04:01:31 +08:00
|
|
|
goto fail_cleaner;
|
2008-06-26 04:01:31 +08:00
|
|
|
|
2008-11-13 03:34:12 +08:00
|
|
|
if (sb->s_flags & MS_RDONLY)
|
|
|
|
return tree_root;
|
|
|
|
|
2008-09-06 04:13:11 +08:00
|
|
|
if (btrfs_super_log_root(disk_super) != 0) {
|
|
|
|
u32 blocksize;
|
|
|
|
u64 bytenr = btrfs_super_log_root(disk_super);
|
|
|
|
|
|
|
|
blocksize =
|
|
|
|
btrfs_level_size(tree_root,
|
|
|
|
btrfs_super_log_root_level(disk_super));
|
2008-04-05 03:40:00 +08:00
|
|
|
|
2008-09-06 04:13:11 +08:00
|
|
|
log_tree_root = kzalloc(sizeof(struct btrfs_root),
|
|
|
|
GFP_NOFS);
|
|
|
|
|
|
|
|
__setup_root(nodesize, leafsize, sectorsize, stripesize,
|
|
|
|
log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
|
|
|
|
|
|
|
|
log_tree_root->node = read_tree_block(tree_root, bytenr,
|
2008-10-30 02:49:05 +08:00
|
|
|
blocksize,
|
|
|
|
generation + 1);
|
2008-09-06 04:13:11 +08:00
|
|
|
ret = btrfs_recover_log_trees(log_tree_root);
|
|
|
|
BUG_ON(ret);
|
|
|
|
}
|
Btrfs: update space balancing code
This patch updates the space balancing code to utilize the new
backref format. Before, btrfs-vol -b would break any COW links
on data blocks or metadata. This was slow and caused the amount
of space used to explode if a large number of snapshots were present.
The new code can keeps the sharing of all data extents and
most of the tree blocks.
To maintain the sharing of data extents, the space balance code uses
a seperate inode hold data extent pointers, then updates the references
to point to the new location.
To maintain the sharing of tree blocks, the space balance code uses
reloc trees to relocate tree blocks in reference counted roots.
There is one reloc tree for each subvol, and all reloc trees share
same root key objectid. Reloc trees are snapshots of the latest
committed roots of subvols (root->commit_root).
To relocate a tree block referenced by a subvol, there are two steps.
COW the block through subvol's reloc tree, then update block pointer in
the subvol to point to the new block. Since all reloc trees share
same root key objectid, doing special handing for tree blocks
owned by them is easy. Once a tree block has been COWed in one
reloc tree, we can use the resulting new block directly when the
same block is required to COW again through other reloc trees.
In this way, relocated tree blocks are shared between reloc trees,
so they are also shared between subvols.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-09-26 22:09:34 +08:00
|
|
|
|
|
|
|
ret = btrfs_cleanup_reloc_trees(tree_root);
|
|
|
|
BUG_ON(ret);
|
|
|
|
|
2007-04-09 22:42:37 +08:00
|
|
|
return tree_root;
|
2007-06-12 18:35:45 +08:00
|
|
|
|
2008-06-26 04:01:31 +08:00
|
|
|
fail_cleaner:
|
2008-06-26 04:01:31 +08:00
|
|
|
kthread_stop(fs_info->cleaner_kthread);
|
2008-03-25 03:01:56 +08:00
|
|
|
fail_extent_root:
|
|
|
|
free_extent_buffer(extent_root->node);
|
2007-06-12 18:35:45 +08:00
|
|
|
fail_tree_root:
|
2007-10-16 04:14:19 +08:00
|
|
|
free_extent_buffer(tree_root->node);
|
2008-04-25 21:04:37 +08:00
|
|
|
fail_sys_array:
|
2007-06-12 18:35:45 +08:00
|
|
|
fail_sb_buffer:
|
2008-07-18 00:53:51 +08:00
|
|
|
btrfs_stop_workers(&fs_info->fixup_workers);
|
2008-11-07 11:02:51 +08:00
|
|
|
btrfs_stop_workers(&fs_info->delalloc_workers);
|
2008-06-12 04:50:36 +08:00
|
|
|
btrfs_stop_workers(&fs_info->workers);
|
|
|
|
btrfs_stop_workers(&fs_info->endio_workers);
|
2008-07-18 00:53:50 +08:00
|
|
|
btrfs_stop_workers(&fs_info->endio_write_workers);
|
2008-06-13 02:46:17 +08:00
|
|
|
btrfs_stop_workers(&fs_info->submit_workers);
|
2008-06-12 09:47:56 +08:00
|
|
|
fail_iput:
|
|
|
|
iput(fs_info->btree_inode);
|
2007-06-12 18:35:45 +08:00
|
|
|
fail:
|
2008-05-14 01:46:40 +08:00
|
|
|
btrfs_close_devices(fs_info->fs_devices);
|
2008-04-25 21:04:37 +08:00
|
|
|
btrfs_mapping_tree_free(&fs_info->mapping_tree);
|
|
|
|
|
2007-06-12 18:35:45 +08:00
|
|
|
kfree(extent_root);
|
|
|
|
kfree(tree_root);
|
2008-03-27 04:24:23 +08:00
|
|
|
bdi_destroy(&fs_info->bdi);
|
2007-06-12 18:35:45 +08:00
|
|
|
kfree(fs_info);
|
2008-10-02 07:09:51 +08:00
|
|
|
kfree(chunk_root);
|
|
|
|
kfree(dev_root);
|
2007-06-12 18:35:45 +08:00
|
|
|
return ERR_PTR(err);
|
2007-02-02 22:18:22 +08:00
|
|
|
}
|
|
|
|
|
2008-04-11 04:19:33 +08:00
|
|
|
static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
|
|
|
|
{
|
|
|
|
char b[BDEVNAME_SIZE];
|
|
|
|
|
|
|
|
if (uptodate) {
|
|
|
|
set_buffer_uptodate(bh);
|
|
|
|
} else {
|
|
|
|
if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
|
|
|
|
printk(KERN_WARNING "lost page write due to "
|
|
|
|
"I/O error on %s\n",
|
|
|
|
bdevname(bh->b_bdev, b));
|
|
|
|
}
|
2008-05-13 01:39:03 +08:00
|
|
|
/* note, we dont' set_buffer_write_io_error because we have
|
|
|
|
* our own ways of dealing with the IO errors
|
|
|
|
*/
|
2008-04-11 04:19:33 +08:00
|
|
|
clear_buffer_uptodate(bh);
|
|
|
|
}
|
|
|
|
unlock_buffer(bh);
|
|
|
|
put_bh(bh);
|
|
|
|
}
|
|
|
|
|
|
|
|
int write_all_supers(struct btrfs_root *root)
|
|
|
|
{
|
|
|
|
struct list_head *cur;
|
|
|
|
struct list_head *head = &root->fs_info->fs_devices->devices;
|
|
|
|
struct btrfs_device *dev;
|
2008-05-07 23:43:44 +08:00
|
|
|
struct btrfs_super_block *sb;
|
2008-04-11 04:19:33 +08:00
|
|
|
struct btrfs_dev_item *dev_item;
|
|
|
|
struct buffer_head *bh;
|
|
|
|
int ret;
|
|
|
|
int do_barriers;
|
2008-04-29 21:38:00 +08:00
|
|
|
int max_errors;
|
|
|
|
int total_errors = 0;
|
2008-05-07 23:43:44 +08:00
|
|
|
u32 crc;
|
|
|
|
u64 flags;
|
2008-04-11 04:19:33 +08:00
|
|
|
|
2008-04-29 21:38:00 +08:00
|
|
|
max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
|
2008-04-11 04:19:33 +08:00
|
|
|
do_barriers = !btrfs_test_opt(root, NOBARRIER);
|
|
|
|
|
2008-05-07 23:43:44 +08:00
|
|
|
sb = &root->fs_info->super_for_commit;
|
|
|
|
dev_item = &sb->dev_item;
|
2008-04-11 04:19:33 +08:00
|
|
|
list_for_each(cur, head) {
|
|
|
|
dev = list_entry(cur, struct btrfs_device, dev_list);
|
2008-05-14 01:46:40 +08:00
|
|
|
if (!dev->bdev) {
|
|
|
|
total_errors++;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
if (!dev->in_fs_metadata)
|
|
|
|
continue;
|
|
|
|
|
2008-05-07 23:43:44 +08:00
|
|
|
btrfs_set_stack_device_type(dev_item, dev->type);
|
|
|
|
btrfs_set_stack_device_id(dev_item, dev->devid);
|
|
|
|
btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
|
|
|
|
btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
|
|
|
|
btrfs_set_stack_device_io_align(dev_item, dev->io_align);
|
|
|
|
btrfs_set_stack_device_io_width(dev_item, dev->io_width);
|
|
|
|
btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
|
|
|
|
memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
|
|
|
|
flags = btrfs_super_flags(sb);
|
|
|
|
btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
|
|
|
|
|
|
|
|
|
|
|
|
crc = ~(u32)0;
|
|
|
|
crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
|
|
|
|
BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
|
|
|
|
btrfs_csum_final(crc, sb->csum);
|
|
|
|
|
|
|
|
bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
|
2008-04-11 04:19:33 +08:00
|
|
|
BTRFS_SUPER_INFO_SIZE);
|
|
|
|
|
2008-05-07 23:43:44 +08:00
|
|
|
memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
|
2008-04-11 04:19:33 +08:00
|
|
|
dev->pending_io = bh;
|
|
|
|
|
|
|
|
get_bh(bh);
|
|
|
|
set_buffer_uptodate(bh);
|
|
|
|
lock_buffer(bh);
|
|
|
|
bh->b_end_io = btrfs_end_buffer_write_sync;
|
|
|
|
|
|
|
|
if (do_barriers && dev->barriers) {
|
|
|
|
ret = submit_bh(WRITE_BARRIER, bh);
|
|
|
|
if (ret == -EOPNOTSUPP) {
|
|
|
|
printk("btrfs: disabling barriers on dev %s\n",
|
|
|
|
dev->name);
|
|
|
|
set_buffer_uptodate(bh);
|
|
|
|
dev->barriers = 0;
|
|
|
|
get_bh(bh);
|
|
|
|
lock_buffer(bh);
|
|
|
|
ret = submit_bh(WRITE, bh);
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
ret = submit_bh(WRITE, bh);
|
|
|
|
}
|
2008-04-29 21:38:00 +08:00
|
|
|
if (ret)
|
|
|
|
total_errors++;
|
2008-04-11 04:19:33 +08:00
|
|
|
}
|
2008-04-29 21:38:00 +08:00
|
|
|
if (total_errors > max_errors) {
|
|
|
|
printk("btrfs: %d errors while writing supers\n", total_errors);
|
|
|
|
BUG();
|
|
|
|
}
|
|
|
|
total_errors = 0;
|
2008-04-11 04:19:33 +08:00
|
|
|
|
|
|
|
list_for_each(cur, head) {
|
|
|
|
dev = list_entry(cur, struct btrfs_device, dev_list);
|
2008-05-14 01:46:40 +08:00
|
|
|
if (!dev->bdev)
|
|
|
|
continue;
|
|
|
|
if (!dev->in_fs_metadata)
|
|
|
|
continue;
|
|
|
|
|
2008-04-11 04:19:33 +08:00
|
|
|
BUG_ON(!dev->pending_io);
|
|
|
|
bh = dev->pending_io;
|
|
|
|
wait_on_buffer(bh);
|
|
|
|
if (!buffer_uptodate(dev->pending_io)) {
|
|
|
|
if (do_barriers && dev->barriers) {
|
|
|
|
printk("btrfs: disabling barriers on dev %s\n",
|
|
|
|
dev->name);
|
|
|
|
set_buffer_uptodate(bh);
|
|
|
|
get_bh(bh);
|
|
|
|
lock_buffer(bh);
|
|
|
|
dev->barriers = 0;
|
|
|
|
ret = submit_bh(WRITE, bh);
|
|
|
|
BUG_ON(ret);
|
|
|
|
wait_on_buffer(bh);
|
2008-05-13 01:39:03 +08:00
|
|
|
if (!buffer_uptodate(bh))
|
|
|
|
total_errors++;
|
2008-04-11 04:19:33 +08:00
|
|
|
} else {
|
2008-04-29 21:38:00 +08:00
|
|
|
total_errors++;
|
2008-04-11 04:19:33 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
}
|
|
|
|
dev->pending_io = NULL;
|
|
|
|
brelse(bh);
|
|
|
|
}
|
2008-04-29 21:38:00 +08:00
|
|
|
if (total_errors > max_errors) {
|
|
|
|
printk("btrfs: %d errors while writing supers\n", total_errors);
|
|
|
|
BUG();
|
|
|
|
}
|
2008-04-11 04:19:33 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2007-03-17 04:20:31 +08:00
|
|
|
int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
|
2007-03-23 03:59:16 +08:00
|
|
|
*root)
|
2007-02-02 22:18:22 +08:00
|
|
|
{
|
2007-04-21 01:16:02 +08:00
|
|
|
int ret;
|
2007-10-16 04:14:19 +08:00
|
|
|
|
2008-04-11 04:19:33 +08:00
|
|
|
ret = write_all_supers(root);
|
2007-10-16 04:14:19 +08:00
|
|
|
return ret;
|
2007-02-22 06:04:57 +08:00
|
|
|
}
|
|
|
|
|
2007-06-23 02:16:25 +08:00
|
|
|
int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
|
2007-04-11 04:58:11 +08:00
|
|
|
{
|
|
|
|
radix_tree_delete(&fs_info->fs_roots_radix,
|
|
|
|
(unsigned long)root->root_key.objectid);
|
2008-01-15 03:41:16 +08:00
|
|
|
if (root->in_sysfs)
|
|
|
|
btrfs_sysfs_del_root(root);
|
2007-04-11 04:58:11 +08:00
|
|
|
if (root->inode)
|
|
|
|
iput(root->inode);
|
|
|
|
if (root->node)
|
2007-10-16 04:14:19 +08:00
|
|
|
free_extent_buffer(root->node);
|
2007-04-11 04:58:11 +08:00
|
|
|
if (root->commit_root)
|
2007-10-16 04:14:19 +08:00
|
|
|
free_extent_buffer(root->commit_root);
|
2007-08-30 03:47:34 +08:00
|
|
|
if (root->name)
|
|
|
|
kfree(root->name);
|
2007-04-11 04:58:11 +08:00
|
|
|
kfree(root);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2007-05-03 03:53:43 +08:00
|
|
|
static int del_fs_roots(struct btrfs_fs_info *fs_info)
|
2007-04-09 22:42:37 +08:00
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
struct btrfs_root *gang[8];
|
|
|
|
int i;
|
|
|
|
|
|
|
|
while(1) {
|
|
|
|
ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
|
|
|
|
(void **)gang, 0,
|
|
|
|
ARRAY_SIZE(gang));
|
|
|
|
if (!ret)
|
|
|
|
break;
|
2007-04-11 04:58:11 +08:00
|
|
|
for (i = 0; i < ret; i++)
|
2007-06-23 02:16:25 +08:00
|
|
|
btrfs_free_fs_root(fs_info, gang[i]);
|
2007-04-09 22:42:37 +08:00
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
2007-04-13 00:14:00 +08:00
|
|
|
|
2008-11-13 03:34:12 +08:00
|
|
|
int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
|
2007-02-22 06:04:57 +08:00
|
|
|
{
|
2008-11-13 03:34:12 +08:00
|
|
|
u64 root_objectid = 0;
|
|
|
|
struct btrfs_root *gang[8];
|
|
|
|
int i;
|
2007-03-14 04:47:54 +08:00
|
|
|
int ret;
|
2007-03-17 04:20:31 +08:00
|
|
|
|
2008-11-13 03:34:12 +08:00
|
|
|
while (1) {
|
|
|
|
ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
|
|
|
|
(void **)gang, root_objectid,
|
|
|
|
ARRAY_SIZE(gang));
|
|
|
|
if (!ret)
|
|
|
|
break;
|
|
|
|
for (i = 0; i < ret; i++) {
|
|
|
|
root_objectid = gang[i]->root_key.objectid;
|
|
|
|
ret = btrfs_find_dead_roots(fs_info->tree_root,
|
|
|
|
root_objectid, gang[i]);
|
|
|
|
BUG_ON(ret);
|
|
|
|
btrfs_orphan_cleanup(gang[i]);
|
|
|
|
}
|
|
|
|
root_objectid++;
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
2008-06-26 04:01:30 +08:00
|
|
|
|
2008-11-13 03:34:12 +08:00
|
|
|
int btrfs_commit_super(struct btrfs_root *root)
|
|
|
|
{
|
|
|
|
struct btrfs_trans_handle *trans;
|
|
|
|
int ret;
|
2008-06-26 04:01:31 +08:00
|
|
|
|
2008-11-13 03:34:12 +08:00
|
|
|
mutex_lock(&root->fs_info->cleaner_mutex);
|
2008-06-26 04:01:31 +08:00
|
|
|
btrfs_clean_old_snapshots(root);
|
2008-11-13 03:34:12 +08:00
|
|
|
mutex_unlock(&root->fs_info->cleaner_mutex);
|
2007-03-23 03:59:16 +08:00
|
|
|
trans = btrfs_start_transaction(root, 1);
|
2007-06-23 02:16:25 +08:00
|
|
|
ret = btrfs_commit_transaction(trans, root);
|
2008-11-13 03:34:12 +08:00
|
|
|
BUG_ON(ret);
|
|
|
|
/* run commit again to drop the original snapshot */
|
2007-03-23 03:59:16 +08:00
|
|
|
trans = btrfs_start_transaction(root, 1);
|
|
|
|
btrfs_commit_transaction(trans, root);
|
|
|
|
ret = btrfs_write_and_wait_transaction(NULL, root);
|
2007-03-14 04:47:54 +08:00
|
|
|
BUG_ON(ret);
|
2008-05-01 01:59:35 +08:00
|
|
|
|
2008-11-13 03:34:12 +08:00
|
|
|
ret = write_ctree_super(NULL, root);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
int close_ctree(struct btrfs_root *root)
|
|
|
|
{
|
|
|
|
struct btrfs_fs_info *fs_info = root->fs_info;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
fs_info->closing = 1;
|
|
|
|
smp_mb();
|
|
|
|
|
|
|
|
kthread_stop(root->fs_info->transaction_kthread);
|
|
|
|
kthread_stop(root->fs_info->cleaner_kthread);
|
|
|
|
|
|
|
|
if (!(fs_info->sb->s_flags & MS_RDONLY)) {
|
|
|
|
ret = btrfs_commit_super(root);
|
|
|
|
if (ret) {
|
|
|
|
printk("btrfs: commit super returns %d\n", ret);
|
|
|
|
}
|
|
|
|
}
|
2007-04-09 22:42:37 +08:00
|
|
|
|
2008-02-01 00:05:37 +08:00
|
|
|
if (fs_info->delalloc_bytes) {
|
|
|
|
printk("btrfs: at unmount delalloc count %Lu\n",
|
|
|
|
fs_info->delalloc_bytes);
|
|
|
|
}
|
2008-07-29 03:32:19 +08:00
|
|
|
if (fs_info->total_ref_cache_size) {
|
|
|
|
printk("btrfs: at umount reference cache size %Lu\n",
|
|
|
|
fs_info->total_ref_cache_size);
|
|
|
|
}
|
2008-07-31 04:29:20 +08:00
|
|
|
|
2007-04-09 22:42:37 +08:00
|
|
|
if (fs_info->extent_root->node)
|
2007-10-16 04:14:19 +08:00
|
|
|
free_extent_buffer(fs_info->extent_root->node);
|
2007-10-16 04:14:48 +08:00
|
|
|
|
2007-04-09 22:42:37 +08:00
|
|
|
if (fs_info->tree_root->node)
|
2007-10-16 04:14:19 +08:00
|
|
|
free_extent_buffer(fs_info->tree_root->node);
|
2007-10-16 04:14:48 +08:00
|
|
|
|
2008-03-25 03:01:56 +08:00
|
|
|
if (root->fs_info->chunk_root->node);
|
|
|
|
free_extent_buffer(root->fs_info->chunk_root->node);
|
|
|
|
|
|
|
|
if (root->fs_info->dev_root->node);
|
|
|
|
free_extent_buffer(root->fs_info->dev_root->node);
|
|
|
|
|
2007-04-27 04:46:15 +08:00
|
|
|
btrfs_free_block_groups(root->fs_info);
|
2007-12-18 09:14:04 +08:00
|
|
|
|
2008-11-13 03:34:12 +08:00
|
|
|
del_fs_roots(fs_info);
|
2007-12-18 09:14:04 +08:00
|
|
|
|
2008-11-13 03:34:12 +08:00
|
|
|
iput(fs_info->btree_inode);
|
2008-04-19 04:11:30 +08:00
|
|
|
|
2008-07-18 00:53:51 +08:00
|
|
|
btrfs_stop_workers(&fs_info->fixup_workers);
|
2008-11-07 11:02:51 +08:00
|
|
|
btrfs_stop_workers(&fs_info->delalloc_workers);
|
2008-06-12 04:50:36 +08:00
|
|
|
btrfs_stop_workers(&fs_info->workers);
|
|
|
|
btrfs_stop_workers(&fs_info->endio_workers);
|
2008-07-18 00:53:50 +08:00
|
|
|
btrfs_stop_workers(&fs_info->endio_write_workers);
|
2008-06-13 02:46:17 +08:00
|
|
|
btrfs_stop_workers(&fs_info->submit_workers);
|
2008-05-01 01:59:35 +08:00
|
|
|
|
2007-10-16 04:19:22 +08:00
|
|
|
#if 0
|
|
|
|
while(!list_empty(&fs_info->hashers)) {
|
|
|
|
struct btrfs_hasher *hasher;
|
|
|
|
hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
|
|
|
|
hashers);
|
|
|
|
list_del(&hasher->hashers);
|
|
|
|
crypto_free_hash(&fs_info->hash_tfm);
|
|
|
|
kfree(hasher);
|
|
|
|
}
|
|
|
|
#endif
|
2008-05-14 01:46:40 +08:00
|
|
|
btrfs_close_devices(fs_info->fs_devices);
|
2008-03-25 03:01:56 +08:00
|
|
|
btrfs_mapping_tree_free(&fs_info->mapping_tree);
|
2008-04-14 21:48:18 +08:00
|
|
|
|
2008-03-26 22:28:07 +08:00
|
|
|
bdi_destroy(&fs_info->bdi);
|
2008-03-25 03:01:56 +08:00
|
|
|
|
2007-04-09 22:42:37 +08:00
|
|
|
kfree(fs_info->extent_root);
|
|
|
|
kfree(fs_info->tree_root);
|
2008-03-25 03:01:56 +08:00
|
|
|
kfree(fs_info->chunk_root);
|
|
|
|
kfree(fs_info->dev_root);
|
2007-02-02 22:18:22 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2008-05-13 01:39:03 +08:00
|
|
|
int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
|
2007-10-16 04:14:19 +08:00
|
|
|
{
|
2008-05-13 01:39:03 +08:00
|
|
|
int ret;
|
2007-10-16 04:18:55 +08:00
|
|
|
struct inode *btree_inode = buf->first_page->mapping->host;
|
2008-05-13 01:39:03 +08:00
|
|
|
|
|
|
|
ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
|
|
|
|
if (!ret)
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
|
|
|
|
parent_transid);
|
|
|
|
return !ret;
|
2007-10-16 04:14:19 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
|
2007-06-29 03:57:36 +08:00
|
|
|
{
|
2007-10-16 04:18:55 +08:00
|
|
|
struct inode *btree_inode = buf->first_page->mapping->host;
|
2008-01-25 05:13:08 +08:00
|
|
|
return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
|
2007-10-16 04:14:19 +08:00
|
|
|
buf);
|
|
|
|
}
|
2007-08-08 04:15:09 +08:00
|
|
|
|
2007-10-16 04:14:19 +08:00
|
|
|
void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
|
|
|
|
{
|
2007-10-16 04:18:55 +08:00
|
|
|
struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
|
2007-10-16 04:14:19 +08:00
|
|
|
u64 transid = btrfs_header_generation(buf);
|
|
|
|
struct inode *btree_inode = root->fs_info->btree_inode;
|
2007-08-08 04:15:09 +08:00
|
|
|
|
2008-06-26 04:01:30 +08:00
|
|
|
WARN_ON(!btrfs_tree_locked(buf));
|
2007-06-29 03:57:36 +08:00
|
|
|
if (transid != root->fs_info->generation) {
|
|
|
|
printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
|
2007-10-16 04:15:53 +08:00
|
|
|
(unsigned long long)buf->start,
|
2007-06-29 03:57:36 +08:00
|
|
|
transid, root->fs_info->generation);
|
|
|
|
WARN_ON(1);
|
|
|
|
}
|
2008-01-25 05:13:08 +08:00
|
|
|
set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
|
2007-02-02 22:18:22 +08:00
|
|
|
}
|
|
|
|
|
2007-09-17 22:58:06 +08:00
|
|
|
void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
|
2007-05-03 03:53:43 +08:00
|
|
|
{
|
2008-05-09 23:52:25 +08:00
|
|
|
/*
|
|
|
|
* looks as though older kernels can get into trouble with
|
|
|
|
* this code, they end up stuck in balance_dirty_pages forever
|
|
|
|
*/
|
2008-05-01 01:59:35 +08:00
|
|
|
struct extent_io_tree *tree;
|
|
|
|
u64 num_dirty;
|
|
|
|
u64 start = 0;
|
2008-11-07 11:02:51 +08:00
|
|
|
unsigned long thresh = 32 * 1024 * 1024;
|
2008-05-01 01:59:35 +08:00
|
|
|
tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
|
|
|
|
|
2008-08-21 01:39:41 +08:00
|
|
|
if (current_is_pdflush() || current->flags & PF_MEMALLOC)
|
2008-05-01 01:59:35 +08:00
|
|
|
return;
|
|
|
|
|
|
|
|
num_dirty = count_range_bits(tree, &start, (u64)-1,
|
|
|
|
thresh, EXTENT_DIRTY);
|
|
|
|
if (num_dirty > thresh) {
|
|
|
|
balance_dirty_pages_ratelimited_nr(
|
2008-02-19 01:12:38 +08:00
|
|
|
root->fs_info->btree_inode->i_mapping, 1);
|
2008-05-01 01:59:35 +08:00
|
|
|
}
|
2008-05-09 23:52:25 +08:00
|
|
|
return;
|
2007-05-03 03:53:43 +08:00
|
|
|
}
|
2007-10-16 04:17:34 +08:00
|
|
|
|
2008-05-13 00:59:19 +08:00
|
|
|
int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
|
2007-10-16 04:17:34 +08:00
|
|
|
{
|
2007-10-16 04:18:55 +08:00
|
|
|
struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
|
2008-04-10 04:28:12 +08:00
|
|
|
int ret;
|
2008-05-13 00:59:19 +08:00
|
|
|
ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
|
2008-04-10 04:28:12 +08:00
|
|
|
if (ret == 0) {
|
|
|
|
buf->flags |= EXTENT_UPTODATE;
|
|
|
|
}
|
|
|
|
return ret;
|
2007-10-16 04:17:34 +08:00
|
|
|
}
|
2007-11-08 10:08:01 +08:00
|
|
|
|
2008-09-08 23:18:08 +08:00
|
|
|
int btree_lock_page_hook(struct page *page)
|
|
|
|
{
|
|
|
|
struct inode *inode = page->mapping->host;
|
|
|
|
struct btrfs_root *root = BTRFS_I(inode)->root;
|
|
|
|
struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
|
|
|
|
struct extent_buffer *eb;
|
|
|
|
unsigned long len;
|
|
|
|
u64 bytenr = page_offset(page);
|
|
|
|
|
|
|
|
if (page->private == EXTENT_PAGE_PRIVATE)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
len = page->private >> 2;
|
|
|
|
eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
|
|
|
|
if (!eb)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
btrfs_tree_lock(eb);
|
|
|
|
spin_lock(&root->fs_info->hash_lock);
|
|
|
|
btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
|
|
|
|
spin_unlock(&root->fs_info->hash_lock);
|
|
|
|
btrfs_tree_unlock(eb);
|
|
|
|
free_extent_buffer(eb);
|
|
|
|
out:
|
|
|
|
lock_page(page);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2008-01-25 05:13:08 +08:00
|
|
|
static struct extent_io_ops btree_extent_io_ops = {
|
2008-09-08 23:18:08 +08:00
|
|
|
.write_cache_pages_lock_hook = btree_lock_page_hook,
|
2008-04-10 04:28:12 +08:00
|
|
|
.readpage_end_io_hook = btree_readpage_end_io_hook,
|
2008-03-25 03:01:56 +08:00
|
|
|
.submit_bio_hook = btree_submit_bio_hook,
|
2008-03-25 03:02:07 +08:00
|
|
|
/* note we're sharing with inode.c for the merge bio hook */
|
|
|
|
.merge_bio_hook = btrfs_merge_bio_hook,
|
2007-11-08 10:08:01 +08:00
|
|
|
};
|