[lustre-devel] [PATCH 03/22] ext4: prealloc table optimization
James Simmons
jsimmons at infradead.org
Sun Jul 21 18:23:32 PDT 2019
Optimize prealloc table
Signed-off-by: James Simmons <jsimmons at infradead.org>
---
fs/ext4/ext4.h | 7 +-
fs/ext4/inode.c | 3 +
fs/ext4/mballoc.c | 221 +++++++++++++++++++++++++++++++++++++++++-------------
fs/ext4/namei.c | 4 +-
fs/ext4/sysfs.c | 8 +-
5 files changed, 186 insertions(+), 57 deletions(-)
diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
index 8abbcab..423ab4d 100644
--- a/fs/ext4/ext4.h
+++ b/fs/ext4/ext4.h
@@ -1190,6 +1190,8 @@ struct ext4_inode_info {
/* Metadata checksum algorithm codes */
#define EXT4_CRC32C_CHKSUM 1
+#define EXT4_MAX_PREALLOC_TABLE 64
+
/*
* Structure of the super block
*/
@@ -1447,12 +1449,14 @@ struct ext4_sb_info {
/* tunables */
unsigned long s_stripe;
- unsigned int s_mb_stream_request;
+ unsigned long s_mb_small_req;
+ unsigned long s_mb_large_req;
unsigned int s_mb_max_to_scan;
unsigned int s_mb_min_to_scan;
unsigned int s_mb_stats;
unsigned int s_mb_order2_reqs;
unsigned int s_mb_group_prealloc;
+ unsigned long *s_mb_prealloc_table;
unsigned int s_max_dir_size_kb;
/* where last allocation was done - for stream allocation */
unsigned long s_mb_last_group;
@@ -2457,6 +2461,7 @@ extern int ext4_init_inode_table(struct super_block *sb,
extern void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate);
/* mballoc.c */
+extern const struct file_operations ext4_seq_prealloc_table_fops;
extern const struct seq_operations ext4_mb_seq_groups_ops;
extern long ext4_mb_stats;
extern long ext4_mb_max_to_scan;
diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
index 6e66175..c37418a 100644
--- a/fs/ext4/inode.c
+++ b/fs/ext4/inode.c
@@ -2796,6 +2796,9 @@ static int ext4_writepages(struct address_space *mapping,
ext4_journal_stop(handle);
}
+ if (wbc->nr_to_write < sbi->s_mb_small_req)
+ wbc->nr_to_write = sbi->s_mb_small_req;
+
if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
range_whole = 1;
diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c
index 99ba720..3be3bef 100644
--- a/fs/ext4/mballoc.c
+++ b/fs/ext4/mballoc.c
@@ -2339,6 +2339,101 @@ static void ext4_mb_seq_groups_stop(struct seq_file *seq, void *v)
.show = ext4_mb_seq_groups_show,
};
+static int ext4_mb_check_and_update_prealloc(struct ext4_sb_info *sbi,
+ char *str, size_t cnt,
+ int update)
+{
+ unsigned long value;
+ unsigned long prev = 0;
+ char *cur;
+ char *next;
+ char *end;
+ int num = 0;
+
+ cur = str;
+ end = str + cnt;
+ while (cur < end) {
+ while ((cur < end) && (*cur == ' ')) cur++;
+ /* Yuck - simple_strtol */
+ value = simple_strtol(cur, &next, 0);
+ if (value == 0)
+ break;
+ if (cur == next)
+ return -EINVAL;
+
+ cur = next;
+
+ if (value > (sbi->s_blocks_per_group - 1 - 1 - sbi->s_itb_per_group))
+ return -EINVAL;
+
+ /* they should add values in order */
+ if (value <= prev)
+ return -EINVAL;
+
+ if (update)
+ sbi->s_mb_prealloc_table[num] = value;
+
+ prev = value;
+ num++;
+ }
+
+ if (num > EXT4_MAX_PREALLOC_TABLE - 1)
+ return -EOVERFLOW;
+
+ if (update)
+ sbi->s_mb_prealloc_table[num] = 0;
+
+ return 0;
+}
+
+static ssize_t ext4_mb_prealloc_table_proc_write(struct file *file,
+ const char __user *buf,
+ size_t cnt, loff_t *pos)
+{
+ struct ext4_sb_info *sbi = EXT4_SB(PDE_DATA(file_inode(file)));
+ char str[128];
+ int rc;
+
+ if (cnt >= sizeof(str))
+ return -EINVAL;
+ if (copy_from_user(str, buf, cnt))
+ return -EFAULT;
+
+ rc = ext4_mb_check_and_update_prealloc(sbi, str, cnt, 0);
+ if (rc)
+ return rc;
+
+ rc = ext4_mb_check_and_update_prealloc(sbi, str, cnt, 1);
+ return rc ? rc : cnt;
+}
+
+static int mb_prealloc_table_seq_show(struct seq_file *m, void *v)
+{
+ struct ext4_sb_info *sbi = EXT4_SB(m->private);
+ int i;
+
+ for (i = 0; i < EXT4_MAX_PREALLOC_TABLE &&
+ sbi->s_mb_prealloc_table[i] != 0; i++)
+ seq_printf(m, "%ld ", sbi->s_mb_prealloc_table[i]);
+ seq_printf(m, "\n");
+
+ return 0;
+}
+
+static int mb_prealloc_table_seq_open(struct inode *inode, struct file *file)
+{
+ return single_open(file, mb_prealloc_table_seq_show, PDE_DATA(inode));
+}
+
+const struct file_operations ext4_seq_prealloc_table_fops = {
+ .owner = THIS_MODULE,
+ .open = mb_prealloc_table_seq_open,
+ .read = seq_read,
+ .llseek = seq_lseek,
+ .release = single_release,
+ .write = ext4_mb_prealloc_table_proc_write,
+};
+
static struct kmem_cache *get_groupinfo_cache(int blocksize_bits)
{
int cache_index = blocksize_bits - EXT4_MIN_BLOCK_LOG_SIZE;
@@ -2567,7 +2662,7 @@ static int ext4_groupinfo_create_slab(size_t size)
int ext4_mb_init(struct super_block *sb)
{
struct ext4_sb_info *sbi = EXT4_SB(sb);
- unsigned i, j;
+ unsigned i, j, k, l;
unsigned offset, offset_incr;
unsigned max;
int ret;
@@ -2616,7 +2711,6 @@ int ext4_mb_init(struct super_block *sb)
sbi->s_mb_max_to_scan = MB_DEFAULT_MAX_TO_SCAN;
sbi->s_mb_min_to_scan = MB_DEFAULT_MIN_TO_SCAN;
sbi->s_mb_stats = MB_DEFAULT_STATS;
- sbi->s_mb_stream_request = MB_DEFAULT_STREAM_THRESHOLD;
sbi->s_mb_order2_reqs = MB_DEFAULT_ORDER2_REQS;
/*
* The default group preallocation is 512, which for 4k block
@@ -2640,9 +2734,28 @@ int ext4_mb_init(struct super_block *sb)
* RAID stripe size so that preallocations don't fragment
* the stripes.
*/
- if (sbi->s_stripe > 1) {
- sbi->s_mb_group_prealloc = roundup(
- sbi->s_mb_group_prealloc, sbi->s_stripe);
+ /* Allocate table once */
+ sbi->s_mb_prealloc_table = kzalloc(
+ EXT4_MAX_PREALLOC_TABLE * sizeof(unsigned long), GFP_NOFS);
+ if (!sbi->s_mb_prealloc_table) {
+ ret = -ENOMEM;
+ goto out;
+ }
+
+ if (sbi->s_stripe == 0) {
+ for (k = 0, l = 4; k <= 9; ++k, l *= 2)
+ sbi->s_mb_prealloc_table[k] = l;
+
+ sbi->s_mb_small_req = 256;
+ sbi->s_mb_large_req = 1024;
+ sbi->s_mb_group_prealloc = 512;
+ } else {
+ for (k = 0, l = sbi->s_stripe; k <= 2; ++k, l *= 2)
+ sbi->s_mb_prealloc_table[k] = l;
+
+ sbi->s_mb_small_req = sbi->s_stripe;
+ sbi->s_mb_large_req = sbi->s_stripe * 8;
+ sbi->s_mb_group_prealloc = sbi->s_stripe * 4;
}
sbi->s_locality_groups = alloc_percpu(struct ext4_locality_group);
@@ -2670,6 +2783,7 @@ int ext4_mb_init(struct super_block *sb)
free_percpu(sbi->s_locality_groups);
sbi->s_locality_groups = NULL;
out:
+ kfree(sbi->s_mb_prealloc_table);
kfree(sbi->s_mb_offsets);
sbi->s_mb_offsets = NULL;
kfree(sbi->s_mb_maxs);
@@ -2932,7 +3046,6 @@ void ext4_exit_mballoc(void)
int err, len;
BUG_ON(ac->ac_status != AC_STATUS_FOUND);
- BUG_ON(ac->ac_b_ex.fe_len <= 0);
sb = ac->ac_sb;
sbi = EXT4_SB(sb);
@@ -3062,13 +3175,14 @@ static void ext4_mb_normalize_group_request(struct ext4_allocation_context *ac)
struct ext4_allocation_request *ar)
{
struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
- int bsbits, max;
+ int bsbits, i, wind;
ext4_lblk_t end;
- loff_t size, start_off;
+ loff_t size;
loff_t orig_size __maybe_unused;
ext4_lblk_t start;
struct ext4_inode_info *ei = EXT4_I(ac->ac_inode);
struct ext4_prealloc_space *pa;
+ unsigned long value, last_non_zero;
/* do normalize only data requests, metadata requests
do not need preallocation */
@@ -3097,51 +3211,47 @@ static void ext4_mb_normalize_group_request(struct ext4_allocation_context *ac)
size = size << bsbits;
if (size < i_size_read(ac->ac_inode))
size = i_size_read(ac->ac_inode);
- orig_size = size;
+ size = (size + ac->ac_sb->s_blocksize - 1) >> bsbits;
- /* max size of free chunks */
- max = 2 << bsbits;
-
-#define NRL_CHECK_SIZE(req, size, max, chunk_size) \
- (req <= (size) || max <= (chunk_size))
-
- /* first, try to predict filesize */
- /* XXX: should this table be tunable? */
- start_off = 0;
- if (size <= 16 * 1024) {
- size = 16 * 1024;
- } else if (size <= 32 * 1024) {
- size = 32 * 1024;
- } else if (size <= 64 * 1024) {
- size = 64 * 1024;
- } else if (size <= 128 * 1024) {
- size = 128 * 1024;
- } else if (size <= 256 * 1024) {
- size = 256 * 1024;
- } else if (size <= 512 * 1024) {
- size = 512 * 1024;
- } else if (size <= 1024 * 1024) {
- size = 1024 * 1024;
- } else if (NRL_CHECK_SIZE(size, 4 * 1024 * 1024, max, 2 * 1024)) {
- start_off = ((loff_t)ac->ac_o_ex.fe_logical >>
- (21 - bsbits)) << 21;
- size = 2 * 1024 * 1024;
- } else if (NRL_CHECK_SIZE(size, 8 * 1024 * 1024, max, 4 * 1024)) {
- start_off = ((loff_t)ac->ac_o_ex.fe_logical >>
- (22 - bsbits)) << 22;
- size = 4 * 1024 * 1024;
- } else if (NRL_CHECK_SIZE(ac->ac_o_ex.fe_len,
- (8<<20)>>bsbits, max, 8 * 1024)) {
- start_off = ((loff_t)ac->ac_o_ex.fe_logical >>
- (23 - bsbits)) << 23;
- size = 8 * 1024 * 1024;
+ start = wind = 0;
+ value = last_non_zero = 0;
+
+ /* let's choose preallocation window depending on file size */
+ for (i = 0; i < EXT4_MAX_PREALLOC_TABLE; i++) {
+ value = sbi->s_mb_prealloc_table[i];
+ if (value == 0)
+ break;
+ else
+ last_non_zero = value;
+
+ if (size <= value) {
+ wind = value;
+ break;
+ }
+ }
+
+ if (wind == 0) {
+ if (last_non_zero != 0) {
+ u64 tstart, tend;
+
+ /* file is quite large, we now preallocate with
+ * the biggest configured window with regart to
+ * logical offset
+ */
+ wind = last_non_zero;
+ tstart = ac->ac_o_ex.fe_logical;
+ do_div(tstart, wind);
+ start = tstart * wind;
+ tend = ac->ac_o_ex.fe_logical + ac->ac_o_ex.fe_len - 1;
+ do_div(tend, wind);
+ tend = tend * wind + wind;
+ size = tend - start;
+ }
} else {
- start_off = (loff_t) ac->ac_o_ex.fe_logical << bsbits;
- size = (loff_t) EXT4_C2B(EXT4_SB(ac->ac_sb),
- ac->ac_o_ex.fe_len) << bsbits;
+ size = wind;
}
- size = size >> bsbits;
- start = start_off >> bsbits;
+
+ orig_size = size;
/* don't cover already allocated blocks in selected range */
if (ar->pleft && start <= ar->lleft) {
@@ -3223,7 +3333,6 @@ static void ext4_mb_normalize_group_request(struct ext4_allocation_context *ac)
(unsigned long) ac->ac_o_ex.fe_logical);
BUG();
}
- BUG_ON(size <= 0 || size > EXT4_BLOCKS_PER_GROUP(ac->ac_sb));
/* now prepare goal request */
@@ -4191,11 +4300,19 @@ static void ext4_mb_group_or_file(struct ext4_allocation_context *ac)
/* don't use group allocation for large files */
size = max(size, isize);
- if (size > sbi->s_mb_stream_request) {
+ if ((ac->ac_o_ex.fe_len >= sbi->s_mb_small_req) ||
+ (size >= sbi->s_mb_large_req)) {
ac->ac_flags |= EXT4_MB_STREAM_ALLOC;
return;
}
+ /*
+ * request is so large that we don't care about
+ * streaming - it overweights any possible seek
+ */
+ if (ac->ac_o_ex.fe_len >= sbi->s_mb_large_req)
+ return;
+
BUG_ON(ac->ac_lg != NULL);
/*
* locality group prealloc space are per cpu. The reason for having
diff --git a/fs/ext4/namei.c b/fs/ext4/namei.c
index a616f58..a52b311 100644
--- a/fs/ext4/namei.c
+++ b/fs/ext4/namei.c
@@ -752,8 +752,8 @@ struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
if (root->info.hash_version != DX_HASH_TEA &&
root->info.hash_version != DX_HASH_HALF_MD4 &&
root->info.hash_version != DX_HASH_LEGACY) {
- ext4_warning_inode(dir, "Unrecognised inode hash code %u",
- root->info.hash_version);
+ ext4_warning_inode(dir, "Unrecognised inode hash code %u for directory %lu",
+ root->info.hash_version, dir->i_ino);
goto fail;
}
if (fname)
diff --git a/fs/ext4/sysfs.c b/fs/ext4/sysfs.c
index 04b4f53..1375815 100644
--- a/fs/ext4/sysfs.c
+++ b/fs/ext4/sysfs.c
@@ -184,7 +184,8 @@ static ssize_t journal_task_show(struct ext4_sb_info *sbi, char *buf)
EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
-EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
+EXT4_RW_ATTR_SBI_UI(mb_small_req, s_mb_small_req);
+EXT4_RW_ATTR_SBI_UI(mb_large_req, s_mb_large_req);
EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
EXT4_RW_ATTR_SBI_UI(extent_max_zeroout_kb, s_extent_max_zeroout_kb);
EXT4_ATTR(trigger_fs_error, 0200, trigger_test_error);
@@ -213,7 +214,8 @@ static ssize_t journal_task_show(struct ext4_sb_info *sbi, char *buf)
ATTR_LIST(mb_max_to_scan),
ATTR_LIST(mb_min_to_scan),
ATTR_LIST(mb_order2_req),
- ATTR_LIST(mb_stream_req),
+ ATTR_LIST(mb_small_req),
+ ATTR_LIST(mb_large_req),
ATTR_LIST(mb_group_prealloc),
ATTR_LIST(max_writeback_mb_bump),
ATTR_LIST(extent_max_zeroout_kb),
@@ -413,6 +415,8 @@ int ext4_register_sysfs(struct super_block *sb)
sb);
proc_create_seq_data("mb_groups", S_IRUGO, sbi->s_proc,
&ext4_mb_seq_groups_ops, sb);
+ proc_create_data("prealloc_table", S_IRUGO, sbi->s_proc,
+ &ext4_seq_prealloc_table_fops, sb);
}
return 0;
}
--
1.8.3.1
More information about the lustre-devel
mailing list