--- /dev/null
+diff -ur a/fs/ext4/ext4.h b/fs/ext4/ext4.h
+--- a/fs/ext4/ext4.h 2021-12-02 15:38:37.084207460 -0700
++++ b/fs/ext4/ext4.h 2021-12-02 15:41:51.939182417 -0700
+@@ -1554,6 +1554,9 @@
+ unsigned int s_mb_min_to_scan;
+ unsigned int s_mb_stats;
+ unsigned int s_mb_order2_reqs;
++ ext4_fsblk_t s_mb_c1_blocks;
++ ext4_fsblk_t s_mb_c2_blocks;
++ ext4_fsblk_t s_mb_c3_blocks;
+ unsigned long *s_mb_prealloc_table;
+ unsigned int s_mb_group_prealloc;
+ unsigned int s_mb_max_inode_prealloc;
+@@ -1573,6 +1576,9 @@
+ atomic_t s_bal_goals; /* goal hits */
+ atomic_t s_bal_breaks; /* too long searches */
+ atomic_t s_bal_2orders; /* 2^order hits */
++ /* cX loop didn't find blocks */
++ atomic64_t s_bal_cX_failed[3];
++ atomic64_t s_bal_cX_skipped[3];
+ spinlock_t s_bal_lock;
+ unsigned long s_mb_buddies_generated;
+ unsigned long long s_mb_generation_time;
+@@ -2977,6 +2983,7 @@
+ /* 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 proc_ops ext4_mb_seq_alloc_fops;
+ extern const struct proc_ops ext4_seq_mb_last_group_fops;
+ extern int ext4_mb_seq_last_start_seq_show(struct seq_file *m, void *v);
+ extern long ext4_mb_stats;
+diff -ur a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c
+--- a/fs/ext4/mballoc.c 2021-12-02 15:38:37.044207688 -0700
++++ b/fs/ext4/mballoc.c 2021-12-02 15:41:51.943182397 -0700
+@@ -2281,6 +2281,20 @@
+ }
+ }
+
++static u64 available_blocks_count(struct ext4_sb_info *sbi)
++{
++ ext4_fsblk_t resv_blocks;
++ u64 bfree;
++ struct ext4_super_block *es = sbi->s_es;
++
++ resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
++ bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
++ percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
++
++ bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
++ return bfree - (ext4_r_blocks_count(es) + resv_blocks);
++}
++
+ static noinline_for_stack int
+ ext4_mb_regular_allocator(struct ext4_allocation_context *ac)
+ {
+@@ -2291,6 +2305,7 @@
+ struct ext4_sb_info *sbi;
+ struct super_block *sb;
+ struct ext4_buddy e4b;
++ ext4_fsblk_t avail_blocks;
+ int lost;
+
+ sb = ac->ac_sb;
+@@ -2344,6 +2359,21 @@
+
+ /* Let's just scan groups to find more-less suitable blocks */
+ cr = ac->ac_2order ? 0 : 1;
++
++ /* Choose what loop to pass based on disk fullness */
++ avail_blocks = available_blocks_count(sbi) ;
++
++ if (avail_blocks < sbi->s_mb_c3_blocks) {
++ cr = 3;
++ atomic64_inc(&sbi->s_bal_cX_skipped[2]);
++ } else if(avail_blocks < sbi->s_mb_c2_blocks) {
++ cr = 2;
++ atomic64_inc(&sbi->s_bal_cX_skipped[1]);
++ } else if(avail_blocks < sbi->s_mb_c1_blocks) {
++ cr = 1;
++ atomic64_inc(&sbi->s_bal_cX_skipped[0]);
++ }
++
+ /*
+ * cr == 0 try to get exact allocation,
+ * cr == 3 try to get anything
+@@ -2431,6 +2461,9 @@
+ if (ac->ac_status != AC_STATUS_CONTINUE)
+ break;
+ }
++ /* Processed all groups and haven't found blocks */
++ if (i == ngroups)
++ atomic64_inc(&sbi->s_bal_cX_failed[cr]);
+ }
+
+ if (ac->ac_b_ex.fe_len > 0 && ac->ac_status != AC_STATUS_FOUND &&
+@@ -2719,6 +2752,92 @@
+ .proc_write = ext4_mb_last_group_write,
+ };
+
++static int mb_seq_alloc_show(struct seq_file *seq, void *v)
++{
++ struct super_block *sb = seq->private;
++ struct ext4_sb_info *sbi = EXT4_SB(sb);
++
++ seq_printf(seq, "mballoc:\n");
++ seq_printf(seq, "\tblocks: %u\n", atomic_read(&sbi->s_bal_allocated));
++ seq_printf(seq, "\treqs: %u\n", atomic_read(&sbi->s_bal_reqs));
++ seq_printf(seq, "\tsuccess: %u\n", atomic_read(&sbi->s_bal_success));
++
++ seq_printf(seq, "\textents_scanned: %u\n",
++ atomic_read(&sbi->s_bal_ex_scanned));
++ seq_printf(seq, "\t\tgoal_hits: %u\n", atomic_read(&sbi->s_bal_goals));
++ seq_printf(seq, "\t\t2^n_hits: %u\n", atomic_read(&sbi->s_bal_2orders));
++ seq_printf(seq, "\t\tbreaks: %u\n", atomic_read(&sbi->s_bal_breaks));
++ seq_printf(seq, "\t\tlost: %u\n", atomic_read(&sbi->s_mb_lost_chunks));
++
++ seq_printf(seq, "\tuseless_c1_loops: %llu\n",
++ atomic64_read(&sbi->s_bal_cX_failed[0]));
++ seq_printf(seq, "\tuseless_c2_loops: %llu\n",
++ atomic64_read(&sbi->s_bal_cX_failed[1]));
++ seq_printf(seq, "\tuseless_c3_loops: %llu\n",
++ atomic64_read(&sbi->s_bal_cX_failed[2]));
++ seq_printf(seq, "\tskipped_c1_loops: %llu\n",
++ atomic64_read(&sbi->s_bal_cX_skipped[0]));
++ seq_printf(seq, "\tskipped_c2_loops: %llu\n",
++ atomic64_read(&sbi->s_bal_cX_skipped[1]));
++ seq_printf(seq, "\tskipped_c3_loops: %llu\n",
++ atomic64_read(&sbi->s_bal_cX_skipped[2]));
++ seq_printf(seq, "\tbuddies_generated: %lu\n",
++ sbi->s_mb_buddies_generated);
++ seq_printf(seq, "\tbuddies_time_used: %llu\n", sbi->s_mb_generation_time);
++ seq_printf(seq, "\tpreallocated: %u\n",
++ atomic_read(&sbi->s_mb_preallocated));
++ seq_printf(seq, "\tdiscarded: %u\n",
++ atomic_read(&sbi->s_mb_discarded));
++ return 0;
++}
++
++static ssize_t mb_seq_alloc_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)));
++
++ atomic_set(&sbi->s_bal_allocated, 0),
++ atomic_set(&sbi->s_bal_reqs, 0),
++ atomic_set(&sbi->s_bal_success, 0);
++
++ atomic_set(&sbi->s_bal_ex_scanned, 0),
++ atomic_set(&sbi->s_bal_goals, 0),
++ atomic_set(&sbi->s_bal_2orders, 0),
++ atomic_set(&sbi->s_bal_breaks, 0),
++ atomic_set(&sbi->s_mb_lost_chunks, 0);
++
++ atomic64_set(&sbi->s_bal_cX_failed[0], 0),
++ atomic64_set(&sbi->s_bal_cX_failed[1], 0),
++ atomic64_set(&sbi->s_bal_cX_failed[2], 0);
++
++ atomic64_set(&sbi->s_bal_cX_skipped[0], 0),
++ atomic64_set(&sbi->s_bal_cX_skipped[1], 0),
++ atomic64_set(&sbi->s_bal_cX_skipped[2], 0);
++
++
++ sbi->s_mb_buddies_generated = 0;
++ sbi->s_mb_generation_time = 0;
++
++ atomic_set(&sbi->s_mb_preallocated, 0),
++ atomic_set(&sbi->s_mb_discarded, 0);
++
++ return cnt;
++}
++
++static int mb_seq_alloc_open(struct inode *inode, struct file *file)
++{
++ return single_open(file, mb_seq_alloc_show, PDE_DATA(inode));
++}
++
++const struct proc_ops ext4_mb_seq_alloc_fops = {
++ .proc_open = mb_seq_alloc_open,
++ .proc_read = seq_read,
++ .proc_lseek = seq_lseek,
++ .proc_release = single_release,
++ .proc_write = mb_seq_alloc_write,
++};
++
+ int ext4_mb_seq_last_start_seq_show(struct seq_file *m, void *v)
+ {
+ struct ext4_sb_info *sbi = EXT4_SB(m->private);
+@@ -2973,6 +3092,7 @@
+ return 0;
+ }
+
++#define THRESHOLD_BLOCKS(ts) (ext4_blocks_count(sbi->s_es) / 100 * ts)
+ int ext4_mb_init(struct super_block *sb)
+ {
+ struct ext4_sb_info *sbi = EXT4_SB(sb);
+@@ -3027,6 +3147,9 @@
+ sbi->s_mb_stats = MB_DEFAULT_STATS;
+ sbi->s_mb_order2_reqs = MB_DEFAULT_ORDER2_REQS;
+ sbi->s_mb_max_inode_prealloc = MB_DEFAULT_MAX_INODE_PREALLOC;
++ sbi->s_mb_c1_blocks = THRESHOLD_BLOCKS(MB_DEFAULT_C1_THRESHOLD);
++ sbi->s_mb_c2_blocks = THRESHOLD_BLOCKS(MB_DEFAULT_C2_THRESHOLD);
++ sbi->s_mb_c3_blocks = THRESHOLD_BLOCKS(MB_DEFAULT_C3_THRESHOLD);
+ /*
+ * The default group preallocation is 512, which for 4k block
+ * sizes translates to 2 megabytes. However for bigalloc file
+@@ -3166,6 +3289,16 @@
+ atomic_read(&sbi->s_bal_reqs),
+ atomic_read(&sbi->s_bal_success));
+ ext4_msg(sb, KERN_INFO,
++ "mballoc: (%llu, %llu, %llu) useless c(0,1,2) loops",
++ atomic64_read(&sbi->s_bal_cX_failed[0]),
++ atomic64_read(&sbi->s_bal_cX_failed[1]),
++ atomic64_read(&sbi->s_bal_cX_failed[2]));
++ ext4_msg(sb, KERN_INFO,
++ "mballoc: (%llu, %llu, %llu) skipped c(0,1,2) loops",
++ atomic64_read(&sbi->s_bal_cX_skipped[0]),
++ atomic64_read(&sbi->s_bal_cX_skipped[1]),
++ atomic64_read(&sbi->s_bal_cX_skipped[2]));
++ ext4_msg(sb, KERN_INFO,
+ "mballoc: %u extents scanned, %u goal hits, "
+ "%u 2^N hits, %u breaks, %u lost",
+ atomic_read(&sbi->s_bal_ex_scanned),
+diff -ur a/fs/ext4/mballoc.h b/fs/ext4/mballoc.h
+--- a/fs/ext4/mballoc.h 2021-12-02 15:38:36.772209242 -0700
++++ b/fs/ext4/mballoc.h 2021-12-02 15:41:51.943182397 -0700
+@@ -68,6 +68,9 @@
+ * for which requests use 2^N search using buddies
+ */
+ #define MB_DEFAULT_ORDER2_REQS 8
++#define MB_DEFAULT_C1_THRESHOLD 25
++#define MB_DEFAULT_C2_THRESHOLD 15
++#define MB_DEFAULT_C3_THRESHOLD 5
+
+ /*
+ * default group prealloc size 512 blocks
+diff -ur a/fs/ext4/sysfs.c b/fs/ext4/sysfs.c
+--- a/fs/ext4/sysfs.c 2021-12-02 15:38:37.044207688 -0700
++++ b/fs/ext4/sysfs.c 2021-12-02 15:43:17.050780832 -0700
+@@ -21,6 +21,9 @@
+ typedef enum {
+ attr_noop,
+ attr_delayed_allocation_blocks,
++ attr_mb_c1_threshold,
++ attr_mb_c2_threshold,
++ attr_mb_c3_threshold,
+ attr_session_write_kbytes,
+ attr_lifetime_write_kbytes,
+ attr_reserved_clusters,
+@@ -135,6 +138,32 @@
+ task_pid_vnr(sbi->s_journal->j_task));
+ }
+
++#define THRESHOLD_PERCENT(ts) (ts * 100 / ext4_blocks_count(sbi->s_es))
++
++static int save_threshold_percent(struct ext4_sb_info *sbi, const char *buf,
++ ext4_fsblk_t *blocks)
++{
++ unsigned long long val;
++
++ int ret;
++
++ ret = kstrtoull(skip_spaces(buf), 0, &val);
++ if (ret || val > 100)
++ return -EINVAL;
++
++ *blocks = val * ext4_blocks_count(sbi->s_es) / 100;
++ return 0;
++}
++
++static ssize_t mb_threshold_store(struct ext4_sb_info *sbi,
++ const char *buf, size_t count,
++ ext4_fsblk_t *blocks)
++{
++ int ret = save_threshold_percent(sbi, buf, blocks);
++
++ return ret ?: count;
++}
++
+ #define EXT4_ATTR(_name,_mode,_id) \
+ static struct ext4_attr ext4_attr_##_name = { \
+ .attr = {.name = __stringify(_name), .mode = _mode }, \
+@@ -204,6 +233,9 @@
+ EXT4_ATTR_FUNC(lifetime_write_kbytes, 0444);
+ EXT4_ATTR_FUNC(reserved_clusters, 0644);
+ EXT4_ATTR_FUNC(sra_exceeded_retry_limit, 0444);
++EXT4_ATTR_FUNC(mb_c1_threshold, 0644);
++EXT4_ATTR_FUNC(mb_c2_threshold, 0644);
++EXT4_ATTR_FUNC(mb_c3_threshold, 0644);
+
+ EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, inode_readahead,
+ ext4_sb_info, s_inode_readahead_blks);
+@@ -258,6 +290,9 @@
+ ATTR_LIST(lifetime_write_kbytes),
+ ATTR_LIST(reserved_clusters),
+ ATTR_LIST(sra_exceeded_retry_limit),
++ ATTR_LIST(mb_c1_threshold),
++ ATTR_LIST(mb_c2_threshold),
++ ATTR_LIST(mb_c3_threshold),
+ ATTR_LIST(inode_readahead_blks),
+ ATTR_LIST(inode_goal),
+ ATTR_LIST(max_dir_size),
+@@ -377,6 +412,15 @@
+ return snprintf(buf, PAGE_SIZE, "%llu\n",
+ (s64) EXT4_C2B(sbi,
+ percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
++ case attr_mb_c1_threshold:
++ return scnprintf(buf, PAGE_SIZE, "%llu\n",
++ THRESHOLD_PERCENT(sbi->s_mb_c1_blocks));
++ case attr_mb_c2_threshold:
++ return scnprintf(buf, PAGE_SIZE, "%llu\n",
++ THRESHOLD_PERCENT(sbi->s_mb_c2_blocks));
++ case attr_mb_c3_threshold:
++ return scnprintf(buf, PAGE_SIZE, "%llu\n",
++ THRESHOLD_PERCENT(sbi->s_mb_c3_blocks));
+ case attr_session_write_kbytes:
+ return session_write_kbytes_show(sbi, buf);
+ case attr_lifetime_write_kbytes:
+@@ -482,6 +526,12 @@
+ return inode_readahead_blks_store(sbi, buf, len);
+ case attr_trigger_test_error:
+ return trigger_test_error(sbi, buf, len);
++ case attr_mb_c1_threshold:
++ return mb_threshold_store(sbi, buf, len, &sbi->s_mb_c1_blocks);
++ case attr_mb_c2_threshold:
++ return mb_threshold_store(sbi, buf, len, &sbi->s_mb_c2_blocks);
++ case attr_mb_c3_threshold:
++ return mb_threshold_store(sbi, buf, len, &sbi->s_mb_c3_blocks);
+ }
+ return 0;
+ }
+@@ -546,6 +596,8 @@
+ &ext4_seq_mb_last_group_fops, sb);
+ proc_create_single_data("mb_last_start", S_IRUGO, sbi->s_proc,
+ ext4_mb_seq_last_start_seq_show, sb);
++ proc_create_data("mb_alloc_stats", S_IFREG | S_IRUGO | S_IWUSR,
++ sbi->s_proc, &ext4_mb_seq_alloc_fops, sb);
+ }
+ return 0;
+ }
git://git.whamcloud.com / fs / lustre-release.git / blobdiff