--- /dev/null
+---
+ fs/ext4/ext4.h | 1
+ fs/ext4/namei.c | 68 ++++++++++++++++++++++++++++++++++++++++++++++++++++++--
+ fs/ext4/super.c | 2 +
+ fs/ext4/sysfs.c | 2 +
+ 4 files changed, 71 insertions(+), 2 deletions(-)
+
+--- a/fs/ext4/ext4.h
++++ b/fs/ext4/ext4.h
+@@ -1547,6 +1547,7 @@ struct ext4_sb_info {
+ unsigned int s_mb_group_prealloc;
+ unsigned int s_mb_max_inode_prealloc;
+ unsigned int s_max_dir_size_kb;
++ unsigned long s_warning_dir_size;
+ /* where last allocation was done - for stream allocation */
+ unsigned long s_mb_last_group;
+ unsigned long s_mb_last_start;
+--- a/fs/ext4/namei.c
++++ b/fs/ext4/namei.c
+@@ -775,12 +775,20 @@ struct ext4_dir_lock_data {
+ #define ext4_htree_lock_data(l) ((struct ext4_dir_lock_data *)(l)->lk_private)
+ #define ext4_find_entry(dir, name, dirent, inline) \
+ ext4_find_entry_locked(dir, name, dirent, inline, NULL)
+-#define ext4_add_entry(handle, dentry, inode) \
+- ext4_add_entry_locked(handle, dentry, inode, NULL)
+
+ /* NB: ext4_lblk_t is 32 bits so we use high bits to identify invalid blk */
+ #define EXT4_HTREE_NODE_CHANGED (0xcafeULL << 32)
+
++inline int ext4_add_entry(handle_t *handle, struct dentry *dentry,
++ struct inode *inode)
++{
++ int ret = ext4_add_entry_locked(handle, dentry, inode, NULL);
++
++ if (ret == -ENOBUFS)
++ ret = 0;
++ return ret;
++}
++
+ static void ext4_htree_event_cb(void *target, void *event)
+ {
+ u64 *block = (u64 *)target;
+@@ -2657,6 +2665,54 @@ out:
+ return err;
+ }
+
++static unsigned long __ext4_max_dir_size(struct dx_frame *frames,
++ struct dx_frame *frame, struct inode *dir)
++{
++ unsigned long max_dir_size;
++
++ if (EXT4_SB(dir->i_sb)->s_max_dir_size_kb) {
++ max_dir_size = EXT4_SB(dir->i_sb)->s_max_dir_size_kb << 10;
++ } else {
++ max_dir_size = EXT4_BLOCK_SIZE(dir->i_sb);
++ while (frame >= frames) {
++ max_dir_size *= dx_get_limit(frame->entries);
++ if (frame == frames)
++ break;
++ frame--;
++ }
++ /* use 75% of max dir size in average */
++ max_dir_size = max_dir_size / 4 * 3;
++ }
++ return max_dir_size;
++}
++
++/*
++ * With hash tree growing, it is easy to hit ENOSPC, but it is hard
++ * to predict when it will happen. let's give administrators warning
++ * when reaching 3/5 and 2/3 of limit
++ */
++static inline bool dir_size_in_warning_range(struct dx_frame *frames,
++ struct dx_frame *frame,
++ struct inode *dir)
++{
++ unsigned long size1, size2;
++ struct super_block *sb = dir->i_sb;
++
++ if (unlikely(!EXT4_SB(sb)->s_warning_dir_size))
++ EXT4_SB(sb)->s_warning_dir_size =
++ __ext4_max_dir_size(frames, frame, dir);
++
++ size1 = EXT4_SB(sb)->s_warning_dir_size / 16 * 10;
++ size1 = size1 & ~(EXT4_BLOCK_SIZE(sb) - 1);
++ size2 = EXT4_SB(sb)->s_warning_dir_size / 16 * 11;
++ size2 = size2 & ~(EXT4_BLOCK_SIZE(sb) - 1);
++ if (in_range(dir->i_size, size1, EXT4_BLOCK_SIZE(sb)) ||
++ in_range(dir->i_size, size2, EXT4_BLOCK_SIZE(sb)))
++ return true;
++
++ return false;
++}
++
+ /*
+ * ext4_add_entry()
+ *
+@@ -2796,6 +2852,7 @@ static int ext4_dx_add_entry(handle_t *h
+ struct ext4_dir_entry_2 *de;
+ int restart;
+ int err;
++ bool ret_warn = false;
+
+ again:
+ restart = 0;
+@@ -2824,6 +2881,11 @@ again:
+ /* Block full, should compress but for now just split */
+ dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
+ dx_get_count(entries), dx_get_limit(entries)));
++
++ if (frame - frames + 1 >= ext4_dir_htree_level(sb) ||
++ EXT4_SB(sb)->s_warning_dir_size)
++ ret_warn = dir_size_in_warning_range(frames, frame, dir);
++
+ /* Need to split index? */
+ if (dx_get_count(entries) == dx_get_limit(entries)) {
+ ext4_lblk_t newblock;
+@@ -2987,6 +3049,8 @@ cleanup:
+ */
+ if (restart && err == 0)
+ goto again;
++ if (err == 0 && ret_warn)
++ err = -ENOBUFS;
+ return err;
+ }
+
+--- a/fs/ext4/super.c
++++ b/fs/ext4/super.c
+@@ -2243,6 +2243,8 @@ static int handle_mount_opt(struct super
+ sbi->s_li_wait_mult = arg;
+ } else if (token == Opt_max_dir_size_kb) {
+ sbi->s_max_dir_size_kb = arg;
++ /* reset s_warning_dir_size and make it re-calculated */
++ sbi->s_warning_dir_size = 0;
+ #ifdef CONFIG_EXT4_DEBUG
+ } else if (token == Opt_fc_debug_max_replay) {
+ sbi->s_fc_debug_max_replay = arg;
+--- a/fs/ext4/sysfs.c
++++ b/fs/ext4/sysfs.c
+@@ -214,6 +214,7 @@ EXT4_ATTR_OFFSET(inode_readahead_blks, 0
+ EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
+ EXT4_RW_ATTR_SBI_UI(max_dir_size, s_max_dir_size_kb);
+ EXT4_RW_ATTR_SBI_UI(max_dir_size_kb, s_max_dir_size_kb);
++EXT4_RW_ATTR_SBI_UI(warning_dir_size, s_warning_dir_size);
+ 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);
+@@ -264,6 +265,7 @@ static struct attribute *ext4_attrs[] =
+ ATTR_LIST(inode_goal),
+ ATTR_LIST(max_dir_size),
+ ATTR_LIST(max_dir_size_kb),
++ ATTR_LIST(warning_dir_size),
+ ATTR_LIST(mb_stats),
+ ATTR_LIST(mb_max_to_scan),
+ ATTR_LIST(mb_min_to_scan),
git://git.whamcloud.com / fs / lustre-release.git / blobdiff