kevin/linux
1
Fork 0
Code
master
linux/fs/ocfs2/move_extents.c
Heming Zhao 4eb7b93e03 ocfs2: improve write IO performance when fragmentation is high 
The group_search function ocfs2_cluster_group_search() should
bypass groups with insufficient space to avoid unnecessary
searches.

This patch is particularly useful when ocfs2 is handling huge
number small files, and volume fragmentation is very high.
In this case, ocfs2 is busy with looking up available la window
from //global_bitmap.

This patch introduces a new member in the Group Description (gd)
struct called 'bg_contig_free_bits', representing the max
contigous free bits in this gd. When ocfs2 allocates a new
la window from //global_bitmap, 'bg_contig_free_bits' helps
expedite the search process.

Let's image below path.

1. la state (->local_alloc_state) is set THROTTLED or DISABLED.

2. when user delete a large file and trigger
   ocfs2_local_alloc_seen_free_bits set osb->local_alloc_state
   unconditionally.

3. a write IOs thread run and trigger the worst performance path

```
ocfs2_reserve_clusters_with_limit
 ocfs2_reserve_local_alloc_bits
  ocfs2_local_alloc_slide_window //[1]
   + ocfs2_local_alloc_reserve_for_window //[2]
   + ocfs2_local_alloc_new_window //[3]
      ocfs2_recalc_la_window
```

[1]:
will be called when la window bits used up.

[2]:
under la state is ENABLED, and this func only check global_bitmap
free bits, it will succeed in general.

[3]:
will use the default la window size to search clusters then fail.
ocfs2_recalc_la_window attempts other la window sizes.
the timing complexity is O(n^4), resulting in a significant time
cost for scanning global bitmap. This leads to a dramatic slowdown
in write I/Os (e.g., user space 'dd').

i.e.
an ocfs2 partition size: 1.45TB, cluster size: 4KB,
la window default size: 106MB.
The partition is fragmentation by creating & deleting huge mount of
small files.

before this patch, the timing of [3] should be
(the number got from real world):
- la window size change order (size: MB):
  106, 53, 26.5, 13, 6.5, 3.25, 1.6, 0.8
  only 0.8MB succeed, 0.8MB also triggers la window to disable.
  ocfs2_local_alloc_new_window retries 8 times, first 7 times totally
  runs in worst case.
- group chain number: 242
  ocfs2_claim_suballoc_bits calls for-loop 242 times
- each chain has 49 block group
  ocfs2_search_chain calls while-loop 49 times
- each bg has 32256 blocks
  ocfs2_block_group_find_clear_bits calls while-loop for 32256 bits.
  for ocfs2_find_next_zero_bit uses ffz() to find zero bit, let's use
  (32256/64) (this is not worst value) for timing calucation.

the loop times: 7*242*49*(32256/64) = 41835024 (~42 million times)

In the worst case, user space writes 1MB data will trigger 42M scanning
times.

under this patch, the timing is '7*242*49 = 83006', reduced by three
orders of magnitude.

Link: https://lkml.kernel.org/r/20240328125203.20892-2-heming.zhao@suse.com
Signed-off-by: Heming Zhao <heming.zhao@suse.com>
Reviewed-by: Joseph Qi <joseph.qi@linux.alibaba.com>
Cc: Changwei Ge <gechangwei@live.cn>
Cc: Gang He <ghe@suse.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Jun Piao <piaojun@huawei.com>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Mark Fasheh <mark@fasheh.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2024-04-25 21:07:03 -07:00

1072 lines
25 KiB
C
Raw Permalink Blame History

// SPDX-License-Identifier: GPL-2.0-only
/*
* move_extents.c
*
* Copyright (C) 2011 Oracle. All rights reserved.
*/
#include <linux/fs.h>
#include <linux/types.h>
#include <linux/mount.h>
#include <linux/swap.h>
#include <cluster/masklog.h>
#include "ocfs2.h"
#include "ocfs2_ioctl.h"
#include "alloc.h"
#include "localalloc.h"
#include "aops.h"
#include "dlmglue.h"
#include "extent_map.h"
#include "inode.h"
#include "journal.h"
#include "suballoc.h"
#include "uptodate.h"
#include "super.h"
#include "dir.h"
#include "buffer_head_io.h"
#include "sysfile.h"
#include "refcounttree.h"
#include "move_extents.h"
struct ocfs2_move_extents_context {
struct inode *inode;
struct file *file;
int auto_defrag;
int partial;
int credits;
u32 new_phys_cpos;
u32 clusters_moved;
u64 refcount_loc;
struct ocfs2_move_extents *range;
struct ocfs2_extent_tree et;
struct ocfs2_alloc_context *meta_ac;
struct ocfs2_alloc_context *data_ac;
struct ocfs2_cached_dealloc_ctxt dealloc;
};
static int __ocfs2_move_extent(handle_t *handle,
struct ocfs2_move_extents_context *context,
u32 cpos, u32 len, u32 p_cpos, u32 new_p_cpos,
int ext_flags)
{
int ret = 0, index;
struct inode *inode = context->inode;
struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
struct ocfs2_extent_rec *rec, replace_rec;
struct ocfs2_path *path = NULL;
struct ocfs2_extent_list *el;
u64 ino = ocfs2_metadata_cache_owner(context->et.et_ci);
u64 old_blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cpos);
ret = ocfs2_duplicate_clusters_by_page(handle, inode, cpos,
p_cpos, new_p_cpos, len);
if (ret) {
mlog_errno(ret);
goto out;
}
memset(&replace_rec, 0, sizeof(replace_rec));
replace_rec.e_cpos = cpu_to_le32(cpos);
replace_rec.e_leaf_clusters = cpu_to_le16(len);
replace_rec.e_blkno = cpu_to_le64(ocfs2_clusters_to_blocks(inode->i_sb,
new_p_cpos));
path = ocfs2_new_path_from_et(&context->et);
if (!path) {
ret = -ENOMEM;
mlog_errno(ret);
goto out;
}
ret = ocfs2_find_path(INODE_CACHE(inode), path, cpos);
if (ret) {
mlog_errno(ret);
goto out;
}
el = path_leaf_el(path);
index = ocfs2_search_extent_list(el, cpos);
if (index == -1) {
ret = ocfs2_error(inode->i_sb,
"Inode %llu has an extent at cpos %u which can no longer be found\n",
(unsigned long long)ino, cpos);
goto out;
}
rec = &el->l_recs[index];
BUG_ON(ext_flags != rec->e_flags);
/*
* after moving/defraging to new location, the extent is not going
* to be refcounted anymore.
*/
replace_rec.e_flags = ext_flags & ~OCFS2_EXT_REFCOUNTED;
ret = ocfs2_split_extent(handle, &context->et, path, index,
&replace_rec, context->meta_ac,
&context->dealloc);
if (ret) {
mlog_errno(ret);
goto out;
}
context->new_phys_cpos = new_p_cpos;
/*
* need I to append truncate log for old clusters?
*/
if (old_blkno) {
if (ext_flags & OCFS2_EXT_REFCOUNTED)
ret = ocfs2_decrease_refcount(inode, handle,
ocfs2_blocks_to_clusters(osb->sb,
old_blkno),
len, context->meta_ac,
&context->dealloc, 1);
else
ret = ocfs2_truncate_log_append(osb, handle,
old_blkno, len);
}
ocfs2_update_inode_fsync_trans(handle, inode, 0);
out:
ocfs2_free_path(path);
return ret;
}
/*
* lock allocator, and reserve appropriate number of bits for
* meta blocks.
*/
static int ocfs2_lock_meta_allocator_move_extents(struct inode *inode,
struct ocfs2_extent_tree *et,
u32 clusters_to_move,
u32 extents_to_split,
struct ocfs2_alloc_context **meta_ac,
int extra_blocks,
int *credits)
{
int ret, num_free_extents;
unsigned int max_recs_needed = 2 * extents_to_split + clusters_to_move;
struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
num_free_extents = ocfs2_num_free_extents(et);
if (num_free_extents < 0) {
ret = num_free_extents;
mlog_errno(ret);
goto out;
}
if (!num_free_extents ||
(ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed))
extra_blocks += ocfs2_extend_meta_needed(et->et_root_el);
ret = ocfs2_reserve_new_metadata_blocks(osb, extra_blocks, meta_ac);
if (ret) {
mlog_errno(ret);
goto out;
}
*credits += ocfs2_calc_extend_credits(osb->sb, et->et_root_el);
mlog(0, "reserve metadata_blocks: %d, data_clusters: %u, credits: %d\n",
extra_blocks, clusters_to_move, *credits);
out:
if (ret) {
if (*meta_ac) {
ocfs2_free_alloc_context(*meta_ac);
*meta_ac = NULL;
}
}
return ret;
}
/*
* Using one journal handle to guarantee the data consistency in case
* crash happens anywhere.
*
* XXX: defrag can end up with finishing partial extent as requested,
* due to not enough contiguous clusters can be found in allocator.
*/
static int ocfs2_defrag_extent(struct ocfs2_move_extents_context *context,
u32 cpos, u32 phys_cpos, u32 *len, int ext_flags)
{
int ret, credits = 0, extra_blocks = 0, partial = context->partial;
handle_t *handle;
struct inode *inode = context->inode;
struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
struct inode *tl_inode = osb->osb_tl_inode;
struct ocfs2_refcount_tree *ref_tree = NULL;
u32 new_phys_cpos, new_len;
u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
int need_free = 0;
if ((ext_flags & OCFS2_EXT_REFCOUNTED) && *len) {
BUG_ON(!ocfs2_is_refcount_inode(inode));
BUG_ON(!context->refcount_loc);
ret = ocfs2_lock_refcount_tree(osb, context->refcount_loc, 1,
&ref_tree, NULL);
if (ret) {
mlog_errno(ret);
return ret;
}
ret = ocfs2_prepare_refcount_change_for_del(inode,
context->refcount_loc,
phys_blkno,
*len,
&credits,
&extra_blocks);
if (ret) {
mlog_errno(ret);
goto out;
}
}
ret = ocfs2_lock_meta_allocator_move_extents(inode, &context->et,
*len, 1,
&context->meta_ac,
extra_blocks, &credits);
if (ret) {
mlog_errno(ret);
goto out;
}
/*
* should be using allocation reservation strategy there?
*
* if (context->data_ac)
* context->data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv;
*/
inode_lock(tl_inode);
if (ocfs2_truncate_log_needs_flush(osb)) {
ret = __ocfs2_flush_truncate_log(osb);
if (ret < 0) {
mlog_errno(ret);
goto out_unlock_mutex;
}
}
/*
* Make sure ocfs2_reserve_cluster is called after
* __ocfs2_flush_truncate_log, otherwise, dead lock may happen.
*
* If ocfs2_reserve_cluster is called
* before __ocfs2_flush_truncate_log, dead lock on global bitmap
* may happen.
*
*/
ret = ocfs2_reserve_clusters(osb, *len, &context->data_ac);
if (ret) {
mlog_errno(ret);
goto out_unlock_mutex;
}
handle = ocfs2_start_trans(osb, credits);
if (IS_ERR(handle)) {
ret = PTR_ERR(handle);
mlog_errno(ret);
goto out_unlock_mutex;
}
ret = __ocfs2_claim_clusters(handle, context->data_ac, 1, *len,
&new_phys_cpos, &new_len);
if (ret) {
mlog_errno(ret);
goto out_commit;
}
/*
* allowing partial extent moving is kind of 'pros and cons', it makes
* whole defragmentation less likely to fail, on the contrary, the bad
* thing is it may make the fs even more fragmented after moving, let
* userspace make a good decision here.
*/
if (new_len != *len) {
mlog(0, "len_claimed: %u, len: %u\n", new_len, *len);
if (!partial) {
context->range->me_flags &= ~OCFS2_MOVE_EXT_FL_COMPLETE;
ret = -ENOSPC;
need_free = 1;
goto out_commit;
}
}
mlog(0, "cpos: %u, phys_cpos: %u, new_phys_cpos: %u\n", cpos,
phys_cpos, new_phys_cpos);
ret = __ocfs2_move_extent(handle, context, cpos, new_len, phys_cpos,
new_phys_cpos, ext_flags);
if (ret)
mlog_errno(ret);
if (partial && (new_len != *len))
*len = new_len;
/*
* Here we should write the new page out first if we are
* in write-back mode.
*/
ret = ocfs2_cow_sync_writeback(inode->i_sb, context->inode, cpos, *len);
if (ret)
mlog_errno(ret);
out_commit:
if (need_free && context->data_ac) {
struct ocfs2_alloc_context *data_ac = context->data_ac;
if (context->data_ac->ac_which == OCFS2_AC_USE_LOCAL)
ocfs2_free_local_alloc_bits(osb, handle, data_ac,
new_phys_cpos, new_len);
else
ocfs2_free_clusters(handle,
data_ac->ac_inode,
data_ac->ac_bh,
ocfs2_clusters_to_blocks(osb->sb, new_phys_cpos),
new_len);
}
ocfs2_commit_trans(osb, handle);
out_unlock_mutex:
inode_unlock(tl_inode);
if (context->data_ac) {
ocfs2_free_alloc_context(context->data_ac);
context->data_ac = NULL;
}
if (context->meta_ac) {
ocfs2_free_alloc_context(context->meta_ac);
context->meta_ac = NULL;
}
out:
if (ref_tree)
ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
return ret;
}
/*
* find the victim alloc group, where #blkno fits.
*/
static int ocfs2_find_victim_alloc_group(struct inode *inode,
u64 vict_blkno,
int type, int slot,
int *vict_bit,
struct buffer_head **ret_bh)
{
int ret, i, bits_per_unit = 0;
u64 blkno;
char namebuf[40];
struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
struct buffer_head *ac_bh = NULL, *gd_bh = NULL;
struct ocfs2_chain_list *cl;
struct ocfs2_chain_rec *rec;
struct ocfs2_dinode *ac_dinode;
struct ocfs2_group_desc *bg;
ocfs2_sprintf_system_inode_name(namebuf, sizeof(namebuf), type, slot);
ret = ocfs2_lookup_ino_from_name(osb->sys_root_inode, namebuf,
strlen(namebuf), &blkno);
if (ret) {
ret = -ENOENT;
goto out;
}
ret = ocfs2_read_blocks_sync(osb, blkno, 1, &ac_bh);
if (ret) {
mlog_errno(ret);
goto out;
}
ac_dinode = (struct ocfs2_dinode *)ac_bh->b_data;
cl = &(ac_dinode->id2.i_chain);
rec = &(cl->cl_recs[0]);
if (type == GLOBAL_BITMAP_SYSTEM_INODE)
bits_per_unit = osb->s_clustersize_bits -
inode->i_sb->s_blocksize_bits;
/*
* 'vict_blkno' was out of the valid range.
*/
if ((vict_blkno < le64_to_cpu(rec->c_blkno)) ||
(vict_blkno >= ((u64)le32_to_cpu(ac_dinode->id1.bitmap1.i_total) <<
bits_per_unit))) {
ret = -EINVAL;
goto out;
}
for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i++) {
rec = &(cl->cl_recs[i]);
if (!rec)
continue;
bg = NULL;
do {
if (!bg)
blkno = le64_to_cpu(rec->c_blkno);
else
blkno = le64_to_cpu(bg->bg_next_group);
if (gd_bh) {
brelse(gd_bh);
gd_bh = NULL;
}
ret = ocfs2_read_blocks_sync(osb, blkno, 1, &gd_bh);
if (ret) {
mlog_errno(ret);
goto out;
}
bg = (struct ocfs2_group_desc *)gd_bh->b_data;
if (vict_blkno < (le64_to_cpu(bg->bg_blkno) +
(le16_to_cpu(bg->bg_bits) << bits_per_unit))) {
*ret_bh = gd_bh;
*vict_bit = (vict_blkno - blkno) >>
bits_per_unit;
mlog(0, "find the victim group: #%llu, "
"total_bits: %u, vict_bit: %u\n",
blkno, le16_to_cpu(bg->bg_bits),
*vict_bit);
goto out;
}
} while (le64_to_cpu(bg->bg_next_group));
}
ret = -EINVAL;
out:
brelse(ac_bh);
/*
* caller has to release the gd_bh properly.
*/
return ret;
}
/*
* XXX: helper to validate and adjust moving goal.
*/
static int ocfs2_validate_and_adjust_move_goal(struct inode *inode,
struct ocfs2_move_extents *range)
{
int ret, goal_bit = 0;
struct buffer_head *gd_bh = NULL;
struct ocfs2_group_desc *bg;
struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
int c_to_b = 1 << (osb->s_clustersize_bits -
inode->i_sb->s_blocksize_bits);
/*
* make goal become cluster aligned.
*/
range->me_goal = ocfs2_block_to_cluster_start(inode->i_sb,
range->me_goal);
/*
* validate goal sits within global_bitmap, and return the victim
* group desc
*/
ret = ocfs2_find_victim_alloc_group(inode, range->me_goal,
GLOBAL_BITMAP_SYSTEM_INODE,
OCFS2_INVALID_SLOT,
&goal_bit, &gd_bh);
if (ret)
goto out;
bg = (struct ocfs2_group_desc *)gd_bh->b_data;
/*
* moving goal is not allowd to start with a group desc blok(#0 blk)
* let's compromise to the latter cluster.
*/
if (range->me_goal == le64_to_cpu(bg->bg_blkno))
range->me_goal += c_to_b;
/*
* movement is not gonna cross two groups.
*/
if ((le16_to_cpu(bg->bg_bits) - goal_bit) * osb->s_clustersize <
range->me_len) {
ret = -EINVAL;
goto out;
}
/*
* more exact validations/adjustments will be performed later during
* moving operation for each extent range.
*/
mlog(0, "extents get ready to be moved to #%llu block\n",
range->me_goal);
out:
brelse(gd_bh);
return ret;
}
static void ocfs2_probe_alloc_group(struct inode *inode, struct buffer_head *bh,
int *goal_bit, u32 move_len, u32 max_hop,
u32 *phys_cpos)
{
int i, used, last_free_bits = 0, base_bit = *goal_bit;
struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
u32 base_cpos = ocfs2_blocks_to_clusters(inode->i_sb,
le64_to_cpu(gd->bg_blkno));
for (i = base_bit; i < le16_to_cpu(gd->bg_bits); i++) {
used = ocfs2_test_bit(i, (unsigned long *)gd->bg_bitmap);
if (used) {
/*
* we even tried searching the free chunk by jumping
* a 'max_hop' distance, but still failed.
*/
if ((i - base_bit) > max_hop) {
*phys_cpos = 0;
break;
}
if (last_free_bits)
last_free_bits = 0;
continue;
} else
last_free_bits++;
if (last_free_bits == move_len) {
i -= move_len;
*goal_bit = i;
*phys_cpos = base_cpos + i;
break;
}
}
mlog(0, "found phys_cpos: %u to fit the wanted moving.\n", *phys_cpos);
}
static int ocfs2_move_extent(struct ocfs2_move_extents_context *context,
u32 cpos, u32 phys_cpos, u32 *new_phys_cpos,
u32 len, int ext_flags)
{
int ret, credits = 0, extra_blocks = 0, goal_bit = 0;
handle_t *handle;
struct inode *inode = context->inode;
struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
struct inode *tl_inode = osb->osb_tl_inode;
struct inode *gb_inode = NULL;
struct buffer_head *gb_bh = NULL;
struct buffer_head *gd_bh = NULL;
struct ocfs2_group_desc *gd;
struct ocfs2_refcount_tree *ref_tree = NULL;
u32 move_max_hop = ocfs2_blocks_to_clusters(inode->i_sb,
context->range->me_threshold);
u64 phys_blkno, new_phys_blkno;
phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
if ((ext_flags & OCFS2_EXT_REFCOUNTED) && len) {
BUG_ON(!ocfs2_is_refcount_inode(inode));
BUG_ON(!context->refcount_loc);
ret = ocfs2_lock_refcount_tree(osb, context->refcount_loc, 1,
&ref_tree, NULL);
if (ret) {
mlog_errno(ret);
return ret;
}
ret = ocfs2_prepare_refcount_change_for_del(inode,
context->refcount_loc,
phys_blkno,
len,
&credits,
&extra_blocks);
if (ret) {
mlog_errno(ret);
goto out;
}
}
ret = ocfs2_lock_meta_allocator_move_extents(inode, &context->et,
len, 1,
&context->meta_ac,
extra_blocks, &credits);
if (ret) {
mlog_errno(ret);
goto out;
}
/*
* need to count 2 extra credits for global_bitmap inode and
* group descriptor.
*/
credits += OCFS2_INODE_UPDATE_CREDITS + 1;
/*
* ocfs2_move_extent() didn't reserve any clusters in lock_allocators()
* logic, while we still need to lock the global_bitmap.
*/
gb_inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
OCFS2_INVALID_SLOT);
if (!gb_inode) {
mlog(ML_ERROR, "unable to get global_bitmap inode\n");
ret = -EIO;
goto out;
}
inode_lock(gb_inode);
ret = ocfs2_inode_lock(gb_inode, &gb_bh, 1);
if (ret) {
mlog_errno(ret);
goto out_unlock_gb_mutex;
}
inode_lock(tl_inode);
handle = ocfs2_start_trans(osb, credits);
if (IS_ERR(handle)) {
ret = PTR_ERR(handle);
mlog_errno(ret);
goto out_unlock_tl_inode;
}
new_phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, *new_phys_cpos);
ret = ocfs2_find_victim_alloc_group(inode, new_phys_blkno,
GLOBAL_BITMAP_SYSTEM_INODE,
OCFS2_INVALID_SLOT,
&goal_bit, &gd_bh);
if (ret) {
mlog_errno(ret);
goto out_commit;
}
/*
* probe the victim cluster group to find a proper
* region to fit wanted movement, it even will perfrom
* a best-effort attempt by compromising to a threshold
* around the goal.
*/
ocfs2_probe_alloc_group(inode, gd_bh, &goal_bit, len, move_max_hop,
new_phys_cpos);
if (!*new_phys_cpos) {
ret = -ENOSPC;
goto out_commit;
}
ret = __ocfs2_move_extent(handle, context, cpos, len, phys_cpos,
*new_phys_cpos, ext_flags);
if (ret) {
mlog_errno(ret);
goto out_commit;
}
gd = (struct ocfs2_group_desc *)gd_bh->b_data;
ret = ocfs2_alloc_dinode_update_counts(gb_inode, handle, gb_bh, len,
le16_to_cpu(gd->bg_chain));
if (ret) {
mlog_errno(ret);
goto out_commit;
}
ret = ocfs2_block_group_set_bits(handle, gb_inode, gd, gd_bh,
goal_bit, len, 0, 0);
if (ret) {
ocfs2_rollback_alloc_dinode_counts(gb_inode, gb_bh, len,
le16_to_cpu(gd->bg_chain));
mlog_errno(ret);
}
/*
* Here we should write the new page out first if we are
* in write-back mode.
*/
ret = ocfs2_cow_sync_writeback(inode->i_sb, context->inode, cpos, len);
if (ret)
mlog_errno(ret);
out_commit:
ocfs2_commit_trans(osb, handle);
brelse(gd_bh);
out_unlock_tl_inode:
inode_unlock(tl_inode);
ocfs2_inode_unlock(gb_inode, 1);
out_unlock_gb_mutex:
inode_unlock(gb_inode);
brelse(gb_bh);
iput(gb_inode);
out:
if (context->meta_ac) {
ocfs2_free_alloc_context(context->meta_ac);
context->meta_ac = NULL;
}
if (ref_tree)
ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
return ret;
}
/*
* Helper to calculate the defraging length in one run according to threshold.
*/
static void ocfs2_calc_extent_defrag_len(u32 *alloc_size, u32 *len_defraged,
u32 threshold, int *skip)
{
if ((*alloc_size + *len_defraged) < threshold) {
/*
* proceed defragmentation until we meet the thresh
*/
*len_defraged += *alloc_size;
} else if (*len_defraged == 0) {
/*
* XXX: skip a large extent.
*/
*skip = 1;
} else {
/*
* split this extent to coalesce with former pieces as
* to reach the threshold.
*
* we're done here with one cycle of defragmentation
* in a size of 'thresh', resetting 'len_defraged'
* forces a new defragmentation.
*/
*alloc_size = threshold - *len_defraged;
*len_defraged = 0;
}
}
static int __ocfs2_move_extents_range(struct buffer_head *di_bh,
struct ocfs2_move_extents_context *context)
{
int ret = 0, flags, do_defrag, skip = 0;
u32 cpos, phys_cpos, move_start, len_to_move, alloc_size;
u32 len_defraged = 0, defrag_thresh = 0, new_phys_cpos = 0;
struct inode *inode = context->inode;
struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
struct ocfs2_move_extents *range = context->range;
struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
if ((i_size_read(inode) == 0) || (range->me_len == 0))
return 0;
if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
return 0;
context->refcount_loc = le64_to_cpu(di->i_refcount_loc);
ocfs2_init_dinode_extent_tree(&context->et, INODE_CACHE(inode), di_bh);
ocfs2_init_dealloc_ctxt(&context->dealloc);
/*
* TO-DO XXX:
*
* - xattr extents.
*/
do_defrag = context->auto_defrag;
/*
* extents moving happens in unit of clusters, for the sake
* of simplicity, we may ignore two clusters where 'byte_start'
* and 'byte_start + len' were within.
*/
move_start = ocfs2_clusters_for_bytes(osb->sb, range->me_start);
len_to_move = (range->me_start + range->me_len) >>
osb->s_clustersize_bits;
if (len_to_move >= move_start)
len_to_move -= move_start;
else
len_to_move = 0;
if (do_defrag) {
defrag_thresh = range->me_threshold >> osb->s_clustersize_bits;
if (defrag_thresh <= 1)
goto done;
} else
new_phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb,
range->me_goal);
mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, clen: %u, "
"thresh: %u\n",
(unsigned long long)OCFS2_I(inode)->ip_blkno,
(unsigned long long)range->me_start,
(unsigned long long)range->me_len,
move_start, len_to_move, defrag_thresh);
cpos = move_start;
while (len_to_move) {
ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &alloc_size,
&flags);
if (ret) {
mlog_errno(ret);
goto out;
}
if (alloc_size > len_to_move)
alloc_size = len_to_move;
/*
* XXX: how to deal with a hole:
*
* - skip the hole of course
* - force a new defragmentation
*/
if (!phys_cpos) {
if (do_defrag)
len_defraged = 0;
goto next;
}
if (do_defrag) {
ocfs2_calc_extent_defrag_len(&alloc_size, &len_defraged,
defrag_thresh, &skip);
/*
* skip large extents
*/
if (skip) {
skip = 0;
goto next;
}
mlog(0, "#Defrag: cpos: %u, phys_cpos: %u, "
"alloc_size: %u, len_defraged: %u\n",
cpos, phys_cpos, alloc_size, len_defraged);
ret = ocfs2_defrag_extent(context, cpos, phys_cpos,
&alloc_size, flags);
} else {
ret = ocfs2_move_extent(context, cpos, phys_cpos,
&new_phys_cpos, alloc_size,
flags);
new_phys_cpos += alloc_size;
}
if (ret < 0) {
mlog_errno(ret);
goto out;
}
context->clusters_moved += alloc_size;
next:
cpos += alloc_size;
len_to_move -= alloc_size;
}
done:
range->me_flags |= OCFS2_MOVE_EXT_FL_COMPLETE;
out:
range->me_moved_len = ocfs2_clusters_to_bytes(osb->sb,
context->clusters_moved);
range->me_new_offset = ocfs2_clusters_to_bytes(osb->sb,
context->new_phys_cpos);
ocfs2_schedule_truncate_log_flush(osb, 1);
ocfs2_run_deallocs(osb, &context->dealloc);
return ret;
}
static int ocfs2_move_extents(struct ocfs2_move_extents_context *context)
{
int status;
handle_t *handle;
struct inode *inode = context->inode;
struct ocfs2_dinode *di;
struct buffer_head *di_bh = NULL;
struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
return -EROFS;
inode_lock(inode);
/*
* This prevents concurrent writes from other nodes
*/
status = ocfs2_rw_lock(inode, 1);
if (status) {
mlog_errno(status);
goto out;
}
status = ocfs2_inode_lock(inode, &di_bh, 1);
if (status) {
mlog_errno(status);
goto out_rw_unlock;
}
/*
* rememer ip_xattr_sem also needs to be held if necessary
*/
down_write(&OCFS2_I(inode)->ip_alloc_sem);
status = __ocfs2_move_extents_range(di_bh, context);
up_write(&OCFS2_I(inode)->ip_alloc_sem);
if (status) {
mlog_errno(status);
goto out_inode_unlock;
}
/*
* We update ctime for these changes
*/
handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
if (IS_ERR(handle)) {
status = PTR_ERR(handle);
mlog_errno(status);
goto out_inode_unlock;
}
status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
OCFS2_JOURNAL_ACCESS_WRITE);
if (status) {
mlog_errno(status);
goto out_commit;
}
di = (struct ocfs2_dinode *)di_bh->b_data;
inode_set_ctime_current(inode);
di->i_ctime = cpu_to_le64(inode_get_ctime_sec(inode));
di->i_ctime_nsec = cpu_to_le32(inode_get_ctime_nsec(inode));
ocfs2_update_inode_fsync_trans(handle, inode, 0);
ocfs2_journal_dirty(handle, di_bh);
out_commit:
ocfs2_commit_trans(osb, handle);
out_inode_unlock:
brelse(di_bh);
ocfs2_inode_unlock(inode, 1);
out_rw_unlock:
ocfs2_rw_unlock(inode, 1);
out:
inode_unlock(inode);
return status;
}
int ocfs2_ioctl_move_extents(struct file *filp, void __user *argp)
{
int status;
struct inode *inode = file_inode(filp);
struct ocfs2_move_extents range;
struct ocfs2_move_extents_context *context;
if (!argp)
return -EINVAL;
status = mnt_want_write_file(filp);
if (status)
return status;
if ((!S_ISREG(inode->i_mode)) || !(filp->f_mode & FMODE_WRITE)) {
status = -EPERM;
goto out_drop;
}
if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
status = -EPERM;
goto out_drop;
}
context = kzalloc(sizeof(struct ocfs2_move_extents_context), GFP_NOFS);
if (!context) {
status = -ENOMEM;
mlog_errno(status);
goto out_drop;
}
context->inode = inode;
context->file = filp;
if (copy_from_user(&range, argp, sizeof(range))) {
status = -EFAULT;
goto out_free;
}
if (range.me_start > i_size_read(inode)) {
status = -EINVAL;
goto out_free;
}
if (range.me_start + range.me_len > i_size_read(inode))
range.me_len = i_size_read(inode) - range.me_start;
context->range = &range;
/*
* ok, the default theshold for the defragmentation
* is 1M, since our maximum clustersize was 1M also.
* any thought?
*/
if (!range.me_threshold)
range.me_threshold = 1024 * 1024;
if (range.me_threshold > i_size_read(inode))
range.me_threshold = i_size_read(inode);
if (range.me_flags & OCFS2_MOVE_EXT_FL_AUTO_DEFRAG) {
context->auto_defrag = 1;
if (range.me_flags & OCFS2_MOVE_EXT_FL_PART_DEFRAG)
context->partial = 1;
} else {
/*
* first best-effort attempt to validate and adjust the goal
* (physical address in block), while it can't guarantee later
* operation can succeed all the time since global_bitmap may
* change a bit over time.
*/
status = ocfs2_validate_and_adjust_move_goal(inode, &range);
if (status)
goto out_copy;
}
status = ocfs2_move_extents(context);
if (status)
mlog_errno(status);
out_copy:
/*
* movement/defragmentation may end up being partially completed,
* that's the reason why we need to return userspace the finished
* length and new_offset even if failure happens somewhere.
*/
if (copy_to_user(argp, &range, sizeof(range)))
status = -EFAULT;
out_free:
kfree(context);
out_drop:
mnt_drop_write_file(filp);
return status;
}
View Git Blame Copy Permalink