--- /dev/null
+commit d8d8fd9192a54c7b8caef8cca9b7a1eb5e5e3298
+Author: Alex Zhuravlev <alex.zhuravlev@sun.com>
+AuthorDate: Thu Oct 23 10:02:19 2008 +0000
+
+Subject: ext4: support for tunable preallocation window
+Add support for tunable preallocation window and new tunables
+for large/small requests.
+
+Bugzilla-ID: b=12800
+Signed-off-by: Alex Zhuravlev <alex.zhuravlev@sun.com>
+Reviewed-by: Kalpak Shah <kalpak@clusterfs.com>
+Reviewed-by: Andreas Dilger <andreas.dilger@sun.com>
+---
+ fs/ext4/ext4.h | 7 +-
+ fs/ext4/inode.c | 3 +
+ fs/ext4/mballoc.c | 219 +++++++++++++++++++++++++++++++++++-----------
+ fs/ext4/sysfs.c | 8 +-
+ 4 files changed, 182 insertions(+), 55 deletions(-)
+
+diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
+index b03dbd7..78d3176 100644
+--- a/fs/ext4/ext4.h
++++ b/fs/ext4/ext4.h
+@@ -1291,6 +1291,8 @@ extern void mb_set_bits(void *bm, int cur, int len);
+ #define EXT4_DFL_MAX_MNT_COUNT 20 /* Allow 20 mounts */
+ #define EXT4_DFL_CHECKINTERVAL 0 /* Don't use interval check */
+
++#define EXT4_MAX_PREALLOC_TABLE 64
++
+ /*
+ * Behaviour when detecting errors
+ */
+@@ -1597,11 +1599,13 @@ struct ext4_sb_info {
+ /* tunables */
+ unsigned long s_stripe;
+ unsigned int s_mb_max_linear_groups;
+- 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 long *s_mb_prealloc_table;
+ unsigned int s_mb_group_prealloc;
+ unsigned int s_mb_max_inode_prealloc;
+ unsigned int s_max_dir_size_kb;
+@@ -2940,6 +2944,7 @@ int ext4_fc_record_regions(struct super_block *sb, int ino,
+ int len, int replay);
+
+ /* mballoc.c */
++extern const struct proc_ops ext4_seq_prealloc_table_fops;
+ extern const struct seq_operations ext4_mb_seq_groups_ops;
+ extern const struct seq_operations ext4_mb_seq_structs_summary_ops;
+ extern long ext4_mb_stats;
+diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
+index 6723622..2ed00ae 100644
+--- a/fs/ext4/inode.c
++++ b/fs/ext4/inode.c
+@@ -2720,6 +2720,9 @@ static int ext4_writepages(struct address_space *mapping,
+ PAGE_SIZE >> inode->i_blkbits);
+ }
+
++ 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 8164480..94a299b 100644
+--- a/fs/ext4/mballoc.c
++++ b/fs/ext4/mballoc.c
+@@ -3070,6 +3070,99 @@ const struct seq_operations ext4_mb_seq_structs_summary_ops = {
+ .show = ext4_mb_seq_structs_summary_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++;
++ 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 proc_ops ext4_seq_prealloc_table_fops = {
++ .proc_open = mb_prealloc_table_seq_open,
++ .proc_read = seq_read,
++ .proc_lseek = seq_lseek,
++ .proc_release = single_release,
++ .proc_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;
+@@ -3382,7 +3475,7 @@ static void ext4_discard_work(struct work_struct *work)
+ 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;
+@@ -3454,7 +3547,6 @@ 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;
+ sbi->s_mb_max_inode_prealloc = MB_DEFAULT_MAX_INODE_PREALLOC;
+ /*
+@@ -3479,9 +3571,29 @@ 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 == NULL) {
++ 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);
+@@ -3515,6 +3627,7 @@ out_free_locality_groups:
+ out:
+ kfree(sbi->s_mb_largest_free_orders);
+ kfree(sbi->s_mb_largest_free_orders_locks);
++ kfree(sbi->s_mb_prealloc_table);
+ kfree(sbi->s_mb_offsets);
+ kfree(sbi->s_mb_maxs);
+ iput(sbi->s_buddy_cache);
+@@ -3783,7 +3896,6 @@ ext4_mb_mark_diskspace_used(struct ext4_allocation_context *ac,
+ 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);
+@@ -4026,13 +4138,14 @@ ext4_mb_normalize_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 */
+@@ -4061,51 +4174,46 @@ ext4_mb_normalize_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;
++
++ 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;
+
+- /* 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;
++ 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;
+
+ /*
+ * For tiny groups (smaller than 8MB) the chosen allocation
+@@ -4196,7 +4304,6 @@ ext4_mb_normalize_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 */
+
+@@ -5199,11 +5306,19 @@ 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/sysfs.c b/fs/ext4/sysfs.c
+index d233c24..9d2254f 100644
+--- a/fs/ext4/sysfs.c
++++ b/fs/ext4/sysfs.c
+@@ -212,7 +212,8 @@ EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
+ 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(mb_max_inode_prealloc, s_mb_max_inode_prealloc);
+ EXT4_RW_ATTR_SBI_UI(mb_max_linear_groups, s_mb_max_linear_groups);
+@@ -262,7 +263,8 @@ static struct attribute *ext4_attrs[] = {
+ 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(mb_max_inode_prealloc),
+ ATTR_LIST(mb_max_linear_groups),
+@@ -543,6 +545,8 @@ int ext4_register_sysfs(struct super_block *sb)
+ ext4_fc_info_show, 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);
+ proc_create_single_data("mb_stats", 0444, sbi->s_proc,
+ ext4_seq_mb_stats_show, sb);
+ proc_create_seq_data("mb_structs_summary", 0444, sbi->s_proc,
+--
+2.34.1
+
git://git.whamcloud.com / fs / lustre-release.git / commitdiff