AS_VERSION_COMPARE([$LINUXRELEASE],[5.4.0],[],
[LDISKFS_SERIES="5.4.0-ml.series"],[
AS_VERSION_COMPARE([$LINUXRELEASE],[5.4.21],
- [LDISKFS_SERIES="5.4.0-ml.series"], # lt
- [LDISKFS_SERIES="5.4.21-ml.series"], # eq
- [LDISKFS_SERIES="5.4.21-ml.series"] # gt
+ [LDISKFS_SERIES="5.4.0-ml.series"], # lt
+ [LDISKFS_SERIES="5.4.21-ml.series"], # eq
+ [AS_VERSION_COMPARE([$LINUXRELEASE],[5.8.0],
+ [LDISKFS_SERIES="5.4.21-ml.series"], # lt
+ [LDISKFS_SERIES="5.8.0-ml.series"], # eq
+ [LDISKFS_SERIES="5.8.0-ml.series"], # gt
+ )]
)])
],
[])
linux_new_headers := htree_lock.h
trace_headers := $(wildcard @LINUX@/include/trace/events/ext4*.h)
-backfs_sources := $(filter-out %.mod.c,$(wildcard @EXT4_SRC_DIR@/*.c))
+backfs_sources := $(filter-out %.mod.c %/inode-test.c,$(wildcard @EXT4_SRC_DIR@/*.c))
new_sources := mmp.c htree_lock.c
new_headers :=
--- /dev/null
+Since we could skip corrupt block groups, this patch
+use ext4_warning() intead of ext4_error() to make FS not
+emount RO in default
+
+---
+ fs/ext4/balloc.c | 10 ++++----
+ fs/ext4/ialloc.c | 6 ++--
+ fs/ext4/mballoc.c | 66 +++++++++++++++++++++---------------------------------
+ 3 files changed, 34 insertions(+), 48 deletions(-)
+
+--- a/fs/ext4/balloc.c
++++ b/fs/ext4/balloc.c
+@@ -382,7 +382,7 @@ static int ext4_validate_block_bitmap(st
+ desc, bh) ||
+ ext4_simulate_fail(sb, EXT4_SIM_BBITMAP_CRC))) {
+ ext4_unlock_group(sb, block_group);
+- ext4_error(sb, "bg %u: bad block bitmap checksum", block_group);
++ ext4_warning(sb, "bg %u: bad block bitmap checksum", block_group);
+ ext4_mark_group_bitmap_corrupted(sb, block_group,
+ EXT4_GROUP_INFO_BBITMAP_CORRUPT);
+ return -EFSBADCRC;
+@@ -390,8 +390,8 @@ static int ext4_validate_block_bitmap(st
+ blk = ext4_valid_block_bitmap(sb, desc, block_group, bh);
+ if (unlikely(blk != 0)) {
+ ext4_unlock_group(sb, block_group);
+- ext4_error(sb, "bg %u: block %llu: invalid block bitmap",
+- block_group, blk);
++ ext4_warning(sb, "bg %u: block %llu: invalid block bitmap",
++ block_group, blk);
+ ext4_mark_group_bitmap_corrupted(sb, block_group,
+ EXT4_GROUP_INFO_BBITMAP_CORRUPT);
+ return -EFSCORRUPTED;
+@@ -467,8 +467,8 @@ ext4_read_block_bitmap_nowait(struct sup
+ ext4_unlock_group(sb, block_group);
+ unlock_buffer(bh);
+ if (err) {
+- ext4_error(sb, "Failed to init block bitmap for group "
+- "%u: %d", block_group, err);
++ ext4_warning(sb, "Failed to init block bitmap for group "
++ "%u: %d", block_group, err);
+ goto out;
+ }
+ goto verify;
+--- a/fs/ext4/ialloc.c
++++ b/fs/ext4/ialloc.c
+@@ -97,8 +97,8 @@ static int ext4_validate_inode_bitmap(st
+ EXT4_INODES_PER_GROUP(sb) / 8) ||
+ ext4_simulate_fail(sb, EXT4_SIM_IBITMAP_CRC)) {
+ ext4_unlock_group(sb, block_group);
+- ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
+- "inode_bitmap = %llu", block_group, blk);
++ ext4_warning(sb, "Corrupt inode bitmap - block_group = %u, "
++ "inode_bitmap = %llu", block_group, blk);
+ ext4_mark_group_bitmap_corrupted(sb, block_group,
+ EXT4_GROUP_INFO_IBITMAP_CORRUPT);
+ return -EFSBADCRC;
+@@ -345,7 +345,7 @@ out:
+ if (!fatal)
+ fatal = err;
+ } else {
+- ext4_error(sb, "bit already cleared for inode %lu", ino);
++ ext4_warning(sb, "bit already cleared for inode %lu", ino);
+ ext4_mark_group_bitmap_corrupted(sb, block_group,
+ EXT4_GROUP_INFO_IBITMAP_CORRUPT);
+ }
+--- a/fs/ext4/mballoc.c
++++ b/fs/ext4/mballoc.c
+@@ -802,10 +802,14 @@ int ext4_mb_generate_buddy(struct super_
+ grp->bb_fragments = fragments;
+
+ if (free != grp->bb_free) {
+- ext4_grp_locked_error(sb, group, 0, 0,
+- "block bitmap and bg descriptor "
+- "inconsistent: %u vs %u free clusters",
+- free, grp->bb_free);
++ struct ext4_group_desc *gdp;
++ gdp = ext4_get_group_desc(sb, group, NULL);
++ ext4_warning(sb, "group %lu: block bitmap and bg descriptor "
++ "inconsistent: %u vs %u free clusters "
++ "%u in gd, %lu pa's",
++ (long unsigned int)group, free, grp->bb_free,
++ ext4_free_group_clusters(sb, gdp),
++ grp->bb_prealloc_nr);
+ /*
+ * If we intend to continue, we consider group descriptor
+ * corrupt and update bb_free using bitmap value
+@@ -1168,7 +1172,7 @@ ext4_mb_load_buddy_gfp(struct super_bloc
+ int block;
+ int pnum;
+ int poff;
+- struct page *page;
++ struct page *page = NULL;
+ int ret;
+ struct ext4_group_info *grp;
+ struct ext4_sb_info *sbi = EXT4_SB(sb);
+@@ -1194,7 +1198,7 @@ ext4_mb_load_buddy_gfp(struct super_bloc
+ */
+ ret = ext4_mb_init_group(sb, group, gfp);
+ if (ret)
+- return ret;
++ goto err;
+ }
+
+ /*
+@@ -1297,6 +1301,7 @@ err:
+ put_page(e4b->bd_buddy_page);
+ e4b->bd_buddy = NULL;
+ e4b->bd_bitmap = NULL;
++ ext4_warning(sb, "Error loading buddy information for %u", group);
+ return ret;
+ }
+
+@@ -3782,9 +3787,11 @@ int ext4_mb_check_ondisk_bitmap(struct s
+ }
+
+ if (free != free_in_gdp) {
+- ext4_error(sb, "on-disk bitmap for group %d"
++ ext4_warning(sb, "on-disk bitmap for group %d"
+ "corrupted: %u blocks free in bitmap, %u - in gd\n",
+ group, free, free_in_gdp);
++ ext4_mark_group_bitmap_corrupted(sb, group,
++ EXT4_GROUP_INFO_BBITMAP_CORRUPT);
+ return -EIO;
+ }
+ return 0;
+@@ -4129,16 +4136,8 @@ ext4_mb_release_inode_pa(struct ext4_bud
+ /* "free < pa->pa_free" means we maybe double alloc the same blocks,
+ * otherwise maybe leave some free blocks unavailable, no need to BUG.*/
+ if ((free > pa->pa_free && !pa->pa_error) || (free < pa->pa_free)) {
+- ext4_error(sb, "pa free mismatch: [pa %p] "
+- "[phy %lu] [logic %lu] [len %u] [free %u] "
+- "[error %u] [inode %d] [freed %u]", pa,
+- (unsigned long)pa->pa_pstart,
+- (unsigned long)pa->pa_lstart,
+- pa->pa_len, (unsigned)pa->pa_free,
+- (unsigned)pa->pa_error, pa->pa_inode->i_ino,
+- free);
+ ext4_grp_locked_error(sb, group, 0, 0, "free %u, pa_free %u",
+- free, pa->pa_free);
++ free, pa->pa_free);
+ /*
+ * pa is already deleted so we use the value obtained
+ * from the bitmap and continue.
+@@ -4199,16 +4198,11 @@ ext4_mb_discard_group_preallocations(str
+ bitmap_bh = ext4_read_block_bitmap(sb, group);
+ if (IS_ERR(bitmap_bh)) {
+ err = PTR_ERR(bitmap_bh);
+- ext4_error_err(sb, -err,
+- "Error %d reading block bitmap for %u",
+- err, group);
+ goto out_dbg;
+ }
+
+ err = ext4_mb_load_buddy(sb, group, &e4b);
+ if (err) {
+- ext4_warning(sb, "Error %d loading buddy information for %u",
+- err, group);
+ put_bh(bitmap_bh);
+ goto out_dbg;
+ }
+@@ -4376,17 +4370,12 @@ repeat:
+
+ err = ext4_mb_load_buddy_gfp(sb, group, &e4b,
+ GFP_NOFS|__GFP_NOFAIL);
+- if (err) {
+- ext4_error_err(sb, -err, "Error %d loading buddy information for %u",
+- err, group);
++ if (err)
+ return;
+- }
+
+ bitmap_bh = ext4_read_block_bitmap(sb, group);
+ if (IS_ERR(bitmap_bh)) {
+ err = PTR_ERR(bitmap_bh);
+- ext4_error_err(sb, -err, "Error %d reading block bitmap for %u",
+- err, group);
+ ext4_mb_unload_buddy(&e4b);
+ continue;
+ }
+@@ -4681,11 +4670,8 @@ ext4_mb_discard_lg_preallocations(struct
+ group = ext4_get_group_number(sb, pa->pa_pstart);
+ err = ext4_mb_load_buddy_gfp(sb, group, &e4b,
+ GFP_NOFS|__GFP_NOFAIL);
+- if (err) {
+- ext4_error_err(sb, -err, "Error %d loading buddy information for %u",
+- err, group);
++ if (err)
+ continue;
+- }
+ ext4_lock_group(sb, group);
+ list_del(&pa->pa_group_list);
+ ext4_get_group_info(sb, group)->bb_prealloc_nr--;
+@@ -4974,7 +4960,7 @@ errout:
+ * been updated or not when fail case. So can
+ * not revert pa_free back, just mark pa_error*/
+ pa->pa_error++;
+- ext4_error(sb,
++ ext4_warning(sb,
+ "Updating bitmap error: [err %d] "
+ "[pa %p] [phy %lu] [logic %lu] "
+ "[len %u] [free %u] [error %u] "
+@@ -4985,6 +4971,7 @@ errout:
+ (unsigned)pa->pa_free,
+ (unsigned)pa->pa_error,
+ pa->pa_inode ? pa->pa_inode->i_ino : 0);
++ ext4_mark_group_bitmap_corrupted(sb, 0, 0);
+ }
+ }
+ ext4_mb_release_context(ac);
+@@ -5271,7 +5258,7 @@ do_more:
+ err = ext4_mb_load_buddy_gfp(sb, block_group, &e4b,
+ GFP_NOFS|__GFP_NOFAIL);
+ if (err)
+- goto error_return;
++ goto error_brelse;
+
+ /*
+ * We need to make sure we don't reuse the freed block until after the
+@@ -5362,8 +5349,9 @@ do_more:
+ goto do_more;
+ }
+ error_return:
+- brelse(bitmap_bh);
+ ext4_std_error(sb, err);
++error_brelse:
++ brelse(bitmap_bh);
+ return;
+ }
+
+@@ -5463,7 +5451,7 @@ int ext4_group_add_blocks(handle_t *hand
+
+ err = ext4_mb_load_buddy(sb, block_group, &e4b);
+ if (err)
+- goto error_return;
++ goto error_brelse;
+
+ /*
+ * need to update group_info->bb_free and bitmap
+@@ -5502,8 +5490,9 @@ int ext4_group_add_blocks(handle_t *hand
+ err = ret;
+
+ error_return:
+- brelse(bitmap_bh);
+ ext4_std_error(sb, err);
++error_brelse:
++ brelse(bitmap_bh);
+ return err;
+ }
+
+@@ -5578,11 +5567,8 @@ ext4_trim_all_free(struct super_block *s
+ trace_ext4_trim_all_free(sb, group, start, max);
+
+ ret = ext4_mb_load_buddy(sb, group, &e4b);
+- if (ret) {
+- ext4_warning(sb, "Error %d loading buddy information for %u",
+- ret, group);
++ if (ret)
+ return ret;
+- }
+ bitmap = e4b.bd_bitmap;
+
+ ext4_lock_group(sb, group);
--- /dev/null
+this patch implements feature which allows ext4 fs users (e.g. Lustre)
+to store data in ext4 dirent.
+data is stored in ext4 dirent after file-name, this space is accounted
+in de->rec_len. flag EXT4_DIRENT_LUFID added to d_type if extra data
+is present.
+
+make use of dentry->d_fsdata to pass fid to ext4. so no
+changes in ext4_add_entry() interface required.
+
+---
+ fs/ext4/dir.c | 13 ++-
+ fs/ext4/ext4.h | 97 +++++++++++++++++++++++++-
+ fs/ext4/inline.c | 8 +-
+ fs/ext4/namei.c | 201 +++++++++++++++++++++++++++++++++++++++++++------------
+ fs/ext4/super.c | 4 -
+ 5 files changed, 267 insertions(+), 56 deletions(-)
+
+--- a/fs/ext4/dir.c
++++ b/fs/ext4/dir.c
+@@ -78,7 +78,7 @@ int __ext4_check_dir_entry(const char *f
+ error_msg = "rec_len is smaller than minimal";
+ else if (unlikely(rlen % 4 != 0))
+ error_msg = "rec_len % 4 != 0";
+- else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len)))
++ else if (unlikely(rlen < EXT4_DIR_ENTRY_LEN(de)))
+ error_msg = "rec_len is too small for name_len";
+ else if (unlikely(next_offset > size))
+ error_msg = "directory entry overrun";
+@@ -226,7 +226,7 @@ static int ext4_readdir(struct file *fil
+ * failure will be detected in the
+ * dirent test below. */
+ if (ext4_rec_len_from_disk(de->rec_len,
+- sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
++ sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
+ break;
+ i += ext4_rec_len_from_disk(de->rec_len,
+ sb->s_blocksize);
+@@ -449,12 +449,17 @@ int ext4_htree_store_dirent(struct file
+ struct fname *fname, *new_fn;
+ struct dir_private_info *info;
+ int len;
++ int extra_data = 0;
+
+ info = dir_file->private_data;
+ p = &info->root.rb_node;
+
+ /* Create and allocate the fname structure */
+- len = sizeof(struct fname) + ent_name->len + 1;
++ if (dirent->file_type & EXT4_DIRENT_LUFID)
++ extra_data = ext4_get_dirent_data_len(dirent);
++
++ len = sizeof(struct fname) + ent_name->len + extra_data + 1;
++
+ new_fn = kzalloc(len, GFP_KERNEL);
+ if (!new_fn)
+ return -ENOMEM;
+@@ -463,7 +468,7 @@ int ext4_htree_store_dirent(struct file
+ new_fn->inode = le32_to_cpu(dirent->inode);
+ new_fn->name_len = ent_name->len;
+ new_fn->file_type = dirent->file_type;
+- memcpy(new_fn->name, ent_name->name, ent_name->len);
++ memcpy(new_fn->name, ent_name->name, ent_name->len + extra_data);
+
+ while (*p) {
+ parent = *p;
+--- a/fs/ext4/ext4.h
++++ b/fs/ext4/ext4.h
+@@ -1118,6 +1118,7 @@ struct ext4_inode_info {
+ __u32 i_csum_seed;
+
+ kprojid_t i_projid;
++ void *i_dirdata;
+ };
+
+ /*
+@@ -1161,6 +1162,7 @@ struct ext4_inode_info {
+ #define EXT4_MOUNT_POSIX_ACL 0x08000 /* POSIX Access Control Lists */
+ #define EXT4_MOUNT_NO_AUTO_DA_ALLOC 0x10000 /* No auto delalloc mapping */
+ #define EXT4_MOUNT_BARRIER 0x20000 /* Use block barriers */
++#define EXT4_MOUNT_DIRDATA 0x60000 /* Data in directory entries*/
+ #define EXT4_MOUNT_QUOTA 0x40000 /* Some quota option set */
+ #define EXT4_MOUNT_USRQUOTA 0x80000 /* "old" user quota,
+ * enable enforcement for hidden
+@@ -1955,6 +1957,7 @@ EXT4_FEATURE_INCOMPAT_FUNCS(casefold, C
+ EXT4_FEATURE_INCOMPAT_FLEX_BG| \
+ EXT4_FEATURE_INCOMPAT_EA_INODE| \
+ EXT4_FEATURE_INCOMPAT_MMP | \
++ EXT4_FEATURE_INCOMPAT_DIRDATA| \
+ EXT4_FEATURE_INCOMPAT_INLINE_DATA | \
+ EXT4_FEATURE_INCOMPAT_ENCRYPT | \
+ EXT4_FEATURE_INCOMPAT_CASEFOLD | \
+@@ -2133,6 +2136,43 @@ struct ext4_dir_entry_tail {
+ #define EXT4_FT_SYMLINK 7
+
+ #define EXT4_FT_MAX 8
++#define EXT4_FT_MASK 0xf
++
++#if EXT4_FT_MAX > EXT4_FT_MASK
++#error "conflicting EXT4_FT_MAX and EXT4_FT_MASK"
++#endif
++
++/*
++ * d_type has 4 unused bits, so it can hold four types data. these different
++ * type of data (e.g. lustre data, high 32 bits of 64-bit inode number) can be
++ * stored, in flag order, after file-name in ext4 dirent.
++*/
++/*
++ * this flag is added to d_type if ext4 dirent has extra data after
++ * filename. this data length is variable and length is stored in first byte
++ * of data. data start after filename NUL byte.
++ * This is used by Lustre FS.
++ */
++#define EXT4_DIRENT_LUFID 0x10
++
++#define EXT4_LUFID_MAGIC 0xAD200907UL
++struct ext4_dentry_param {
++ __u32 edp_magic; /* EXT4_LUFID_MAGIC */
++ char edp_len; /* size of edp_data in bytes */
++ char edp_data[0]; /* packed array of data */
++} __packed;
++
++static inline unsigned char *ext4_dentry_get_data(struct super_block *sb,
++ struct ext4_dentry_param *p)
++
++{
++ if (!ext4_has_feature_dirdata(sb))
++ return NULL;
++ if (p && p->edp_magic == EXT4_LUFID_MAGIC)
++ return &p->edp_len;
++ else
++ return NULL;
++}
+
+ #define EXT4_FT_DIR_CSUM 0xDE
+
+@@ -2143,8 +2183,16 @@ struct ext4_dir_entry_tail {
+ */
+ #define EXT4_DIR_PAD 4
+ #define EXT4_DIR_ROUND (EXT4_DIR_PAD - 1)
+-#define EXT4_DIR_REC_LEN(name_len) (((name_len) + 8 + EXT4_DIR_ROUND) & \
++#define EXT4_DIR_REC_LEN_(name_len) (((name_len) + 8 + EXT4_DIR_ROUND) & \
+ ~EXT4_DIR_ROUND)
++#define EXT4_DIR_ENTRY_LEN_(de) (EXT4_DIR_REC_LEN_((de)->name_len +\
++ ext4_get_dirent_data_len(de)))
++/* ldiskfs */
++#define EXT4_DIR_REC_LEN(name_len) EXT4_DIR_REC_LEN_((name_len))
++#define EXT4_DIR_ENTRY_LEN(de) EXT4_DIR_ENTRY_LEN_((de))
++/* lustre osd_handler compat */
++#define __EXT4_DIR_REC_LEN(name_len) EXT4_DIR_REC_LEN_((name_len))
++
+ #define EXT4_MAX_REC_LEN ((1<<16)-1)
+
+ /*
+@@ -2604,11 +2652,11 @@ extern int ext4_find_dest_de(struct inod
+ struct buffer_head *bh,
+ void *buf, int buf_size,
+ struct ext4_filename *fname,
+- struct ext4_dir_entry_2 **dest_de);
++ struct ext4_dir_entry_2 **dest_de, int *dlen);
+ void ext4_insert_dentry(struct inode *inode,
+ struct ext4_dir_entry_2 *de,
+ int buf_size,
+- struct ext4_filename *fname);
++ struct ext4_filename *fname, void *data);
+ static inline void ext4_update_dx_flag(struct inode *inode)
+ {
+ if (!ext4_has_feature_dir_index(inode->i_sb)) {
+@@ -2623,10 +2671,17 @@ static const unsigned char ext4_filetype
+
+ static inline unsigned char get_dtype(struct super_block *sb, int filetype)
+ {
+- if (!ext4_has_feature_filetype(sb) || filetype >= EXT4_FT_MAX)
++ int fl_index = filetype & EXT4_FT_MASK;
++
++ if (!ext4_has_feature_filetype(sb) || fl_index >= EXT4_FT_MAX)
+ return DT_UNKNOWN;
+
+- return ext4_filetype_table[filetype];
++ if (!test_opt(sb, DIRDATA))
++ return ext4_filetype_table[fl_index];
++
++ return (ext4_filetype_table[fl_index]) |
++ (filetype & EXT4_DIRENT_LUFID);
++
+ }
+ extern int ext4_check_all_de(struct inode *dir, struct buffer_head *bh,
+ void *buf, int buf_size);
+@@ -2785,6 +2840,8 @@ extern struct inode *ext4_create_inode(h
+ extern int ext4_delete_entry(handle_t *handle, struct inode * dir,
+ struct ext4_dir_entry_2 *de_del,
+ struct buffer_head *bh);
++extern int ext4_add_dot_dotdot(handle_t *handle, struct inode *dir,
++ struct inode *inode, const void *, const void *);
+ extern int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
+ __u32 start_minor_hash, __u32 *next_hash);
+ extern int ext4_search_dir(struct buffer_head *bh,
+@@ -3559,6 +3616,36 @@ static inline int ext4_buffer_uptodate(s
+ return buffer_uptodate(bh);
+ }
+
++/*
++ * Compute the total directory entry data length.
++ * This includes the filename and an implicit NUL terminator (always present),
++ * and optional extensions. Each extension has a bit set in the high 4 bits of
++ * de->file_type, and the extension length is the first byte in each entry.
++ */
++static inline int ext4_get_dirent_data_len(struct ext4_dir_entry_2 *de)
++{
++ char *len = de->name + de->name_len + 1 /* NUL terminator */;
++ int dlen = 0;
++ __u8 extra_data_flags = (de->file_type & ~EXT4_FT_MASK) >> 4;
++ struct ext4_dir_entry_tail *t = (struct ext4_dir_entry_tail *)de;
++
++ if (!t->det_reserved_zero1 &&
++ le16_to_cpu(t->det_rec_len) ==
++ sizeof(struct ext4_dir_entry_tail) &&
++ !t->det_reserved_zero2 &&
++ t->det_reserved_ft == EXT4_FT_DIR_CSUM)
++ return 0;
++
++ while (extra_data_flags) {
++ if (extra_data_flags & 1) {
++ dlen += *len + (dlen == 0);
++ len += *len;
++ }
++ extra_data_flags >>= 1;
++ }
++ return dlen;
++}
++
+ #endif /* __KERNEL__ */
+
+ #define EFSBADCRC EBADMSG /* Bad CRC detected */
+--- a/fs/ext4/inline.c
++++ b/fs/ext4/inline.c
+@@ -1023,7 +1023,7 @@ static int ext4_add_dirent_to_inline(han
+ struct ext4_dir_entry_2 *de;
+
+ err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
+- inline_size, fname, &de);
++ inline_size, fname, &de, NULL);
+ if (err)
+ return err;
+
+@@ -1031,7 +1031,7 @@ static int ext4_add_dirent_to_inline(han
+ err = ext4_journal_get_write_access(handle, iloc->bh);
+ if (err)
+ return err;
+- ext4_insert_dentry(inode, de, inline_size, fname);
++ ext4_insert_dentry(inode, de, inline_size, fname, NULL);
+
+ ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
+
+@@ -1380,7 +1380,7 @@ int ext4_inlinedir_to_tree(struct file *
+ fake.name_len = 1;
+ strcpy(fake.name, ".");
+ fake.rec_len = ext4_rec_len_to_disk(
+- EXT4_DIR_REC_LEN(fake.name_len),
++ EXT4_DIR_ENTRY_LEN(&fake),
+ inline_size);
+ ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
+ de = &fake;
+@@ -1390,7 +1390,7 @@ int ext4_inlinedir_to_tree(struct file *
+ fake.name_len = 2;
+ strcpy(fake.name, "..");
+ fake.rec_len = ext4_rec_len_to_disk(
+- EXT4_DIR_REC_LEN(fake.name_len),
++ EXT4_DIR_ENTRY_LEN(&fake),
+ inline_size);
+ ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
+ de = &fake;
+--- a/fs/ext4/namei.c
++++ b/fs/ext4/namei.c
+@@ -265,7 +265,8 @@ static unsigned dx_get_count(struct dx_e
+ static unsigned dx_get_limit(struct dx_entry *entries);
+ static void dx_set_count(struct dx_entry *entries, unsigned value);
+ static void dx_set_limit(struct dx_entry *entries, unsigned value);
+-static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
++static inline unsigned dx_root_limit(struct inode *dir,
++ struct ext4_dir_entry_2 *dot_de, unsigned infosize);
+ static unsigned dx_node_limit(struct inode *dir);
+ static struct dx_frame *dx_probe(struct ext4_filename *fname,
+ struct inode *dir,
+@@ -409,22 +410,23 @@ static struct dx_countlimit *get_dx_coun
+ {
+ struct ext4_dir_entry *dp;
+ struct dx_root_info *root;
+- int count_offset;
++ int count_offset, dot_rec_len, dotdot_rec_len;
+
+ if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
+ count_offset = 8;
+- else if (le16_to_cpu(dirent->rec_len) == 12) {
+- dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
++ else {
++ dot_rec_len = le16_to_cpu(dirent->rec_len);
++ dp = (struct ext4_dir_entry *)(((void *)dirent) + dot_rec_len);
+ if (le16_to_cpu(dp->rec_len) !=
+- EXT4_BLOCK_SIZE(inode->i_sb) - 12)
++ EXT4_BLOCK_SIZE(inode->i_sb) - dot_rec_len)
+ return NULL;
+- root = (struct dx_root_info *)(((void *)dp + 12));
++ dotdot_rec_len = EXT4_DIR_ENTRY_LEN((struct ext4_dir_entry_2 *)dp);
++ root = (struct dx_root_info *)(((void *)dp + dotdot_rec_len));
+ if (root->reserved_zero ||
+ root->info_length != sizeof(struct dx_root_info))
+ return NULL;
+- count_offset = 32;
+- } else
+- return NULL;
++ count_offset = 8 + dot_rec_len + dotdot_rec_len;
++ }
+
+ if (offset)
+ *offset = count_offset;
+@@ -529,11 +531,12 @@ ext4_next_entry(struct ext4_dir_entry_2
+ */
+ struct dx_root_info *dx_get_dx_info(struct ext4_dir_entry_2 *de)
+ {
++ BUG_ON(de->name_len != 1);
+ /* get dotdot first */
+- de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(1));
++ de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_ENTRY_LEN(de));
+
+ /* dx root info is after dotdot entry */
+- de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(2));
++ de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_ENTRY_LEN(de));
+
+ return (struct dx_root_info *)de;
+ }
+@@ -578,10 +581,16 @@ static inline void dx_set_limit(struct d
+ ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
+ }
+
+-static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
++static inline unsigned dx_root_limit(struct inode *dir,
++ struct ext4_dir_entry_2 *dot_de, unsigned infosize)
+ {
+- unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
+- EXT4_DIR_REC_LEN(2) - infosize;
++ struct ext4_dir_entry_2 *dotdot_de;
++ unsigned entry_space;
++
++ BUG_ON(dot_de->name_len != 1);
++ dotdot_de = ext4_next_entry(dot_de, dir->i_sb->s_blocksize);
++ entry_space = dir->i_sb->s_blocksize - EXT4_DIR_ENTRY_LEN(dot_de) -
++ EXT4_DIR_ENTRY_LEN(dotdot_de) - infosize;
+
+ if (ext4_has_metadata_csum(dir->i_sb))
+ entry_space -= sizeof(struct dx_tail);
+@@ -702,7 +711,7 @@ static struct stats dx_show_leaf(struct
+ (unsigned) ((char *) de - base));
+ #endif
+ }
+- space += EXT4_DIR_REC_LEN(de->name_len);
++ space += EXT4_DIR_ENTRY_LEN(de);
+ names++;
+ }
+ de = ext4_next_entry(de, size);
+@@ -809,11 +818,14 @@ dx_probe(struct ext4_filename *fname, st
+
+ entries = (struct dx_entry *)(((char *)info) + info->info_length);
+
+- if (dx_get_limit(entries) != dx_root_limit(dir,
+- info->info_length)) {
++ if (dx_get_limit(entries) !=
++ dx_root_limit(dir, (struct ext4_dir_entry_2 *)frame->bh->b_data,
++ info->info_length)) {
+ ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
+ dx_get_limit(entries),
+- dx_root_limit(dir, info->info_length));
++ dx_root_limit(dir,
++ (struct ext4_dir_entry_2 *)frame->bh->b_data,
++ info->info_length));
+ goto fail;
+ }
+
+@@ -1799,7 +1811,7 @@ dx_move_dirents(char *from, char *to, st
+ while (count--) {
+ struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
+ (from + (map->offs<<2));
+- rec_len = EXT4_DIR_REC_LEN(de->name_len);
++ rec_len = EXT4_DIR_ENTRY_LEN(de);
+ memcpy (to, de, rec_len);
+ ((struct ext4_dir_entry_2 *) to)->rec_len =
+ ext4_rec_len_to_disk(rec_len, blocksize);
+@@ -1823,7 +1835,7 @@ static struct ext4_dir_entry_2* dx_pack_
+ while ((char*)de < base + blocksize) {
+ next = ext4_next_entry(de, blocksize);
+ if (de->inode && de->name_len) {
+- rec_len = EXT4_DIR_REC_LEN(de->name_len);
++ rec_len = EXT4_DIR_ENTRY_LEN(de);
+ if (de > to)
+ memmove(to, de, rec_len);
+ to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
+@@ -1950,14 +1962,16 @@ int ext4_find_dest_de(struct inode *dir,
+ struct buffer_head *bh,
+ void *buf, int buf_size,
+ struct ext4_filename *fname,
+- struct ext4_dir_entry_2 **dest_de)
++ struct ext4_dir_entry_2 **dest_de, int *dlen)
+ {
+ struct ext4_dir_entry_2 *de;
+- unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
++ unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname)) +
++ (dlen ? *dlen : 0);
+ int nlen, rlen;
+ unsigned int offset = 0;
+ char *top;
+
++ dlen ? *dlen = 0 : 0; /* default set to 0 */
+ de = (struct ext4_dir_entry_2 *)buf;
+ top = buf + buf_size - reclen;
+ while ((char *) de <= top) {
+@@ -1966,10 +1980,26 @@ int ext4_find_dest_de(struct inode *dir,
+ return -EFSCORRUPTED;
+ if (ext4_match(dir, fname, de))
+ return -EEXIST;
+- nlen = EXT4_DIR_REC_LEN(de->name_len);
++ nlen = EXT4_DIR_ENTRY_LEN(de);
+ rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
+ if ((de->inode ? rlen - nlen : rlen) >= reclen)
+ break;
++ /* Then for dotdot entries, check for the smaller space
++ * required for just the entry, no FID */
++ if (fname_len(fname) == 2 && memcmp(fname_name(fname), "..", 2) == 0) {
++ if ((de->inode ? rlen - nlen : rlen) >=
++ EXT4_DIR_REC_LEN(fname_len(fname))) {
++ /* set dlen=1 to indicate not
++ * enough space store fid */
++ dlen ? *dlen = 1 : 0;
++ break;
++ }
++ /* The new ".." entry must be written over the
++ * previous ".." entry, which is the first
++ * entry traversed by this scan. If it doesn't
++ * fit, something is badly wrong, so -EIO. */
++ return -EIO;
++ }
+ de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
+ offset += rlen;
+ }
+@@ -1983,12 +2013,12 @@ int ext4_find_dest_de(struct inode *dir,
+ void ext4_insert_dentry(struct inode *inode,
+ struct ext4_dir_entry_2 *de,
+ int buf_size,
+- struct ext4_filename *fname)
++ struct ext4_filename *fname, void *data)
+ {
+
+ int nlen, rlen;
+
+- nlen = EXT4_DIR_REC_LEN(de->name_len);
++ nlen = EXT4_DIR_ENTRY_LEN(de);
+ rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
+ if (de->inode) {
+ struct ext4_dir_entry_2 *de1 =
+@@ -2002,6 +2032,11 @@ void ext4_insert_dentry(struct inode *in
+ ext4_set_de_type(inode->i_sb, de, inode->i_mode);
+ de->name_len = fname_len(fname);
+ memcpy(de->name, fname_name(fname), fname_len(fname));
++ if (data) {
++ de->name[fname_len(fname)] = 0;
++ memcpy(&de->name[fname_len(fname) + 1], data, *(char *)data);
++ de->file_type |= EXT4_DIRENT_LUFID;
++ }
+ }
+
+ /*
+@@ -2019,14 +2054,19 @@ static int add_dirent_to_buf(handle_t *h
+ {
+ unsigned int blocksize = dir->i_sb->s_blocksize;
+ int csum_size = 0;
+- int err, err2;
++ int err, err2, dlen = 0;
++ unsigned char *data;
+
++ data = ext4_dentry_get_data(inode->i_sb, (struct ext4_dentry_param *)
++ EXT4_I(inode)->i_dirdata);
+ if (ext4_has_metadata_csum(inode->i_sb))
+ csum_size = sizeof(struct ext4_dir_entry_tail);
+
+ if (!de) {
++ if (data)
++ dlen = (*data) + 1;
+ err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
+- blocksize - csum_size, fname, &de);
++ blocksize - csum_size, fname, &de, &dlen);
+ if (err)
+ return err;
+ }
+@@ -2038,7 +2078,10 @@ static int add_dirent_to_buf(handle_t *h
+ }
+
+ /* By now the buffer is marked for journaling */
+- ext4_insert_dentry(inode, de, blocksize, fname);
++ /* If writing the short form of "dotdot", don't add the data section */
++ if (dlen == 1)
++ data = NULL;
++ ext4_insert_dentry(inode, de, blocksize, fname, data);
+
+ /*
+ * XXX shouldn't update any times until successful
+@@ -2143,7 +2186,8 @@ static int make_indexed_dir(handle_t *ha
+
+ dx_set_block(entries, 1);
+ dx_set_count(entries, 1);
+- dx_set_limit(entries, dx_root_limit(dir, sizeof(*dx_info)));
++ dx_set_limit(entries, dx_root_limit(dir,
++ dot_de, sizeof(*dx_info)));
+
+ /* Initialize as for dx_probe */
+ fname->hinfo.hash_version = dx_info->hash_version;
+@@ -2193,6 +2237,8 @@ static int ext4_update_dotdot(handle_t *
+ struct buffer_head *dir_block;
+ struct ext4_dir_entry_2 *de;
+ int len, journal = 0, err = 0;
++ int dlen = 0;
++ char *data;
+
+ if (IS_ERR(handle))
+ return PTR_ERR(handle);
+@@ -2218,11 +2264,16 @@ static int ext4_update_dotdot(handle_t *
+ goto out_journal;
+
+ journal = 1;
+- de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(1));
++ de->rec_len = cpu_to_le16(EXT4_DIR_ENTRY_LEN(de));
+ }
+
+- len -= EXT4_DIR_REC_LEN(1);
+- assert(len == 0 || len >= EXT4_DIR_REC_LEN(2));
++ len -= EXT4_DIR_ENTRY_LEN(de);
++ data = ext4_dentry_get_data(dir->i_sb,
++ (struct ext4_dentry_param *)dentry->d_fsdata);
++ if (data)
++ dlen = *data + 1;
++ assert(len == 0 || len >= EXT4_DIR_REC_LEN(2 + dlen));
++
+ de = (struct ext4_dir_entry_2 *)
+ ((char *) de + le16_to_cpu(de->rec_len));
+ if (!journal) {
+@@ -2239,7 +2290,12 @@ static int ext4_update_dotdot(handle_t *
+ assert(le16_to_cpu(de->rec_len) >= EXT4_DIR_REC_LEN(2));
+ de->name_len = 2;
+ strcpy(de->name, "..");
+- ext4_set_de_type(dir->i_sb, de, S_IFDIR);
++ if (data != NULL && ext4_get_dirent_data_len(de) >= dlen) {
++ de->name[2] = 0;
++ memcpy(&de->name[2 + 1], data, *data);
++ ext4_set_de_type(dir->i_sb, de, S_IFDIR);
++ de->file_type |= EXT4_DIRENT_LUFID;
++ }
+
+ out_journal:
+ if (journal) {
+@@ -2280,6 +2336,7 @@ static int ext4_add_entry(handle_t *hand
+ ext4_lblk_t block, blocks;
+ int csum_size = 0;
+
++ EXT4_I(inode)->i_dirdata = dentry->d_fsdata;
+ if (ext4_has_metadata_csum(inode->i_sb))
+ csum_size = sizeof(struct ext4_dir_entry_tail);
+
+@@ -2832,37 +2889,70 @@ err_unlock_inode:
+ return err;
+ }
+
++struct tp_block {
++ struct inode *inode;
++ void *data1;
++ void *data2;
++};
++
+ struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
+ struct ext4_dir_entry_2 *de,
+ int blocksize, int csum_size,
+ unsigned int parent_ino, int dotdot_real_len)
+ {
++ void *data1 = NULL, *data2 = NULL;
++ int dot_reclen = 0;
++
++ if (dotdot_real_len == 10) {
++ struct tp_block *tpb = (struct tp_block *)inode;
++ data1 = tpb->data1;
++ data2 = tpb->data2;
++ inode = tpb->inode;
++ dotdot_real_len = 0;
++ }
+ de->inode = cpu_to_le32(inode->i_ino);
+ de->name_len = 1;
+- de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
+- blocksize);
+ strcpy(de->name, ".");
+ ext4_set_de_type(inode->i_sb, de, S_IFDIR);
+
++ /* get packed fid data*/
++ data1 = ext4_dentry_get_data(inode->i_sb,
++ (struct ext4_dentry_param *) data1);
++ if (data1) {
++ de->name[1] = 0;
++ memcpy(&de->name[2], data1, *(char *) data1);
++ de->file_type |= EXT4_DIRENT_LUFID;
++ }
++ de->rec_len = cpu_to_le16(EXT4_DIR_ENTRY_LEN(de));
++ dot_reclen = cpu_to_le16(de->rec_len);
+ de = ext4_next_entry(de, blocksize);
+ de->inode = cpu_to_le32(parent_ino);
+ de->name_len = 2;
++ strcpy(de->name, "..");
++ ext4_set_de_type(inode->i_sb, de, S_IFDIR);
++ data2 = ext4_dentry_get_data(inode->i_sb,
++ (struct ext4_dentry_param *) data2);
++ if (data2) {
++ de->name[2] = 0;
++ memcpy(&de->name[3], data2, *(char *) data2);
++ de->file_type |= EXT4_DIRENT_LUFID;
++ }
+ if (!dotdot_real_len)
+ de->rec_len = ext4_rec_len_to_disk(blocksize -
+- (csum_size + EXT4_DIR_REC_LEN(1)),
++ (csum_size + dot_reclen),
+ blocksize);
+ else
+ de->rec_len = ext4_rec_len_to_disk(
+- EXT4_DIR_REC_LEN(de->name_len), blocksize);
+- strcpy(de->name, "..");
+- ext4_set_de_type(inode->i_sb, de, S_IFDIR);
++ EXT4_DIR_ENTRY_LEN(de), blocksize);
+
+ return ext4_next_entry(de, blocksize);
+ }
+
+ static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
+- struct inode *inode)
++ struct inode *inode,
++ const void *data1, const void *data2)
+ {
++ struct tp_block param;
+ struct buffer_head *dir_block = NULL;
+ struct ext4_dir_entry_2 *de;
+ ext4_lblk_t block = 0;
+@@ -2886,7 +2976,11 @@ static int ext4_init_new_dir(handle_t *h
+ if (IS_ERR(dir_block))
+ return PTR_ERR(dir_block);
+ de = (struct ext4_dir_entry_2 *)dir_block->b_data;
+- ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
++ param.inode = inode;
++ param.data1 = (void *)data1;
++ param.data2 = (void *)data2;
++ ext4_init_dot_dotdot((struct inode *)(¶m), de, blocksize,
++ csum_size, dir->i_ino, 10);
+ set_nlink(inode, 2);
+ if (csum_size)
+ ext4_initialize_dirent_tail(dir_block, blocksize);
+@@ -2901,6 +2995,29 @@ out:
+ return err;
+ }
+
++/* Initialize @inode as a subdirectory of @dir, and add the
++ * "." and ".." entries into the first directory block. */
++int ext4_add_dot_dotdot(handle_t *handle, struct inode *dir,
++ struct inode *inode,
++ const void *data1, const void *data2)
++{
++ int rc;
++
++ if (IS_ERR(handle))
++ return PTR_ERR(handle);
++
++ if (IS_DIRSYNC(dir))
++ ext4_handle_sync(handle);
++
++ inode->i_op = &ext4_dir_inode_operations;
++ inode->i_fop = &ext4_dir_operations;
++ rc = ext4_init_new_dir(handle, dir, inode, data1, data2);
++ if (!rc)
++ rc = ext4_mark_inode_dirty(handle, inode);
++ return rc;
++}
++EXPORT_SYMBOL(ext4_add_dot_dotdot);
++
+ static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
+ {
+ handle_t *handle;
+@@ -2927,7 +3044,7 @@ retry:
+
+ inode->i_op = &ext4_dir_inode_operations;
+ inode->i_fop = &ext4_dir_operations;
+- err = ext4_init_new_dir(handle, dir, inode);
++ err = ext4_init_new_dir(handle, dir, inode, NULL, NULL);
+ if (err)
+ goto out_clear_inode;
+ err = ext4_mark_inode_dirty(handle, inode);
+--- a/fs/ext4/super.c
++++ b/fs/ext4/super.c
+@@ -1510,7 +1510,7 @@ enum {
+ Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
+ Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
+ Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
+- Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
++ Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err, Opt_dirdata,
+ Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version,
+ Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never,
+ Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
+@@ -1590,6 +1590,7 @@ static const match_table_t tokens = {
+ {Opt_nolazytime, "nolazytime"},
+ {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
+ {Opt_nodelalloc, "nodelalloc"},
++ {Opt_dirdata, "dirdata"},
+ {Opt_removed, "mblk_io_submit"},
+ {Opt_removed, "nomblk_io_submit"},
+ {Opt_block_validity, "block_validity"},
+@@ -1822,6 +1823,7 @@ static const struct mount_opts {
+ {Opt_usrjquota, 0, MOPT_Q},
+ {Opt_grpjquota, 0, MOPT_Q},
+ {Opt_offusrjquota, 0, MOPT_Q},
++ {Opt_dirdata, EXT4_MOUNT_DIRDATA, MOPT_SET},
+ {Opt_offgrpjquota, 0, MOPT_Q},
+ {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
+ {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
--- /dev/null
+---
+ fs/ext4/ext4.h | 2 +
+ fs/ext4/mballoc.c | 58 ++++++++++++++++++++++++++++++++++++++++++++++++++++++
+ fs/ext4/sysfs.c | 4 +++
+ 3 files changed, 64 insertions(+)
+
+--- a/fs/ext4/ext4.h
++++ b/fs/ext4/ext4.h
+@@ -2808,6 +2808,8 @@ extern void ext4_end_bitmap_read(struct
+ /* 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_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;
+ extern long ext4_mb_max_to_scan;
+ extern int ext4_mb_init(struct super_block *);
+--- a/fs/ext4/mballoc.c
++++ b/fs/ext4/mballoc.c
+@@ -2574,6 +2574,64 @@ static struct kmem_cache *get_groupinfo_
+ return cachep;
+ }
+
++#define EXT4_MB_MAX_INPUT_STRING_SIZE 32
++
++static ssize_t ext4_mb_last_group_write(struct file *file,
++ const char __user *buf,
++ size_t cnt, loff_t *pos)
++{
++ char dummy[EXT4_MB_MAX_INPUT_STRING_SIZE + 1];
++ struct super_block *sb = PDE_DATA(file_inode(file));
++ struct ext4_sb_info *sbi = EXT4_SB(sb);
++ unsigned long val;
++ char *end;
++
++ if (cnt > EXT4_MB_MAX_INPUT_STRING_SIZE)
++ return -EINVAL;
++ if (copy_from_user(dummy, buf, cnt))
++ return -EFAULT;
++ dummy[cnt] = '\0';
++ val = simple_strtoul(dummy, &end, 0);
++ if (dummy == end)
++ return -EINVAL;
++ if (val >= ext4_get_groups_count(sb))
++ return -ERANGE;
++ spin_lock(&sbi->s_md_lock);
++ sbi->s_mb_last_group = val;
++ sbi->s_mb_last_start = 0;
++ spin_unlock(&sbi->s_md_lock);
++ return cnt;
++}
++
++static int ext4_mb_seq_last_group_seq_show(struct seq_file *m, void *v)
++{
++ struct ext4_sb_info *sbi = EXT4_SB(m->private);
++
++ seq_printf(m , "%ld\n", sbi->s_mb_last_group);
++ return 0;
++}
++
++static int ext4_mb_seq_last_group_open(struct inode *inode, struct file *file)
++{
++ return single_open(file, ext4_mb_seq_last_group_seq_show, PDE_DATA(inode));
++}
++
++const struct proc_ops ext4_seq_mb_last_group_fops = {
++ .proc_open = ext4_mb_seq_last_group_open,
++ .proc_read = seq_read,
++ .proc_lseek = seq_lseek,
++ .proc_release = seq_release,
++ .proc_write = ext4_mb_last_group_write,
++};
++
++int ext4_mb_seq_last_start_seq_show(struct seq_file *m, void *v)
++{
++ struct ext4_sb_info *sbi = EXT4_SB(m->private);
++
++ seq_printf(m , "%ld\n", sbi->s_mb_last_start);
++ return 0;
++}
++
+ /*
+ * Allocate the top-level s_group_info array for the specified number
+ * of groups
+--- a/fs/ext4/sysfs.c
++++ b/fs/ext4/sysfs.c
+@@ -524,6 +524,10 @@ int ext4_register_sysfs(struct super_blo
+ &ext4_mb_seq_groups_ops, sb);
+ proc_create_data("prealloc_table", S_IRUGO, sbi->s_proc,
+ &ext4_seq_prealloc_table_fops, sb);
++ proc_create_data("mb_last_group", S_IRUGO, sbi->s_proc,
++ &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);
+ }
+ return 0;
+ }
--- /dev/null
+From aa282f628e4ad75eea7f8ee1b26dea920c238241 Mon Sep 17 00:00:00 2001
+From: Shaun Tancheff <stancheff@cray.com>
+Date: Tue, 6 Aug 2019 17:00:55 -0500
+Subject: [PATCH] + linux-5.3/ext4-kill-dx-root
+
+---
+ fs/ext4/namei.c | 111 +++++++++++++++++++++++++++++---------------------------
+ 1 file changed, 58 insertions(+), 53 deletions(-)
+
+--- a/fs/ext4/namei.c
++++ b/fs/ext4/namei.c
+@@ -219,22 +219,13 @@ struct dx_entry
+ * hash version mod 4 should never be 0. Sincerely, the paranoia department.
+ */
+
+-struct dx_root
++struct dx_root_info
+ {
+- struct fake_dirent dot;
+- char dot_name[4];
+- struct fake_dirent dotdot;
+- char dotdot_name[4];
+- struct dx_root_info
+- {
+- __le32 reserved_zero;
+- u8 hash_version;
+- u8 info_length; /* 8 */
+- u8 indirect_levels;
+- u8 unused_flags;
+- }
+- info;
+- struct dx_entry entries[];
++ __le32 reserved_zero;
++ u8 hash_version;
++ u8 info_length; /* 8 */
++ u8 indirect_levels;
++ u8 unused_flags;
+ };
+
+ struct dx_node
+@@ -536,6 +527,16 @@ ext4_next_entry(struct ext4_dir_entry_2
+ * Future: use high four bits of block for coalesce-on-delete flags
+ * Mask them off for now.
+ */
++struct dx_root_info *dx_get_dx_info(struct ext4_dir_entry_2 *de)
++{
++ /* get dotdot first */
++ de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(1));
++
++ /* dx root info is after dotdot entry */
++ de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(2));
++
++ return (struct dx_root_info *)de;
++}
+
+ static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
+ {
+@@ -760,7 +761,7 @@ dx_probe(struct ext4_filename *fname, st
+ {
+ unsigned count, indirect;
+ struct dx_entry *at, *entries, *p, *q, *m;
+- struct dx_root *root;
++ struct dx_root_info *info;
+ struct dx_frame *frame = frame_in;
+ struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
+ u32 hash;
+@@ -769,18 +770,17 @@ dx_probe(struct ext4_filename *fname, st
+ frame->bh = ext4_read_dirblock(dir, 0, INDEX);
+ if (IS_ERR(frame->bh))
+ return (struct dx_frame *) frame->bh;
+-
+- root = (struct dx_root *) frame->bh->b_data;
+- 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 for directory "
+- "%lu", root->info.hash_version, dir->i_ino);
++ info = dx_get_dx_info((struct ext4_dir_entry_2 *)frame->bh->b_data);
++ if (info->hash_version != DX_HASH_TEA &&
++ info->hash_version != DX_HASH_HALF_MD4 &&
++ info->hash_version != DX_HASH_LEGACY) {
++ ext4_warning(dir->i_sb, "Unrecognised inode hash code %d for directory "
++ "#%lu", info->hash_version, dir->i_ino);
+ goto fail;
+ }
+ if (fname)
+ hinfo = &fname->hinfo;
+- hinfo->hash_version = root->info.hash_version;
++ hinfo->hash_version = info->hash_version;
+ if (hinfo->hash_version <= DX_HASH_TEA)
+ hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
+ hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
+@@ -788,13 +788,13 @@ dx_probe(struct ext4_filename *fname, st
+ ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), hinfo);
+ hash = hinfo->hash;
+
+- if (root->info.unused_flags & 1) {
++ if (info->unused_flags & 1) {
+ ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
+- root->info.unused_flags);
++ info->unused_flags);
+ goto fail;
+ }
+
+- indirect = root->info.indirect_levels;
++ indirect = info->indirect_levels;
+ if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
+ ext4_warning(dir->i_sb,
+ "Directory (ino: %lu) htree depth %#06x exceed"
+@@ -807,14 +807,13 @@ dx_probe(struct ext4_filename *fname, st
+ goto fail;
+ }
+
+- entries = (struct dx_entry *)(((char *)&root->info) +
+- root->info.info_length);
++ entries = (struct dx_entry *)(((char *)info) + info->info_length);
+
+ if (dx_get_limit(entries) != dx_root_limit(dir,
+- root->info.info_length)) {
++ info->info_length)) {
+ ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
+ dx_get_limit(entries),
+- dx_root_limit(dir, root->info.info_length));
++ dx_root_limit(dir, info->info_length));
+ goto fail;
+ }
+
+@@ -899,7 +898,7 @@ static void dx_release(struct dx_frame *
+ if (frames[0].bh == NULL)
+ return;
+
+- info = &((struct dx_root *)frames[0].bh->b_data)->info;
++ info = dx_get_dx_info((struct ext4_dir_entry_2 *)frames[0].bh->b_data);
+ /* save local copy, "info" may be freed after brelse() */
+ indirect_levels = info->indirect_levels;
+ for (i = 0; i <= indirect_levels; i++) {
+@@ -2072,16 +2071,15 @@ static int make_indexed_dir(handle_t *ha
+ struct inode *inode, struct buffer_head *bh)
+ {
+ struct buffer_head *bh2;
+- struct dx_root *root;
+ struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
+ struct dx_entry *entries;
+- struct ext4_dir_entry_2 *de, *de2;
++ struct ext4_dir_entry_2 *de, *de2, *dot_de, *dotdot_de;
+ char *data2, *top;
+ unsigned len;
+ int retval;
+ unsigned blocksize;
+ ext4_lblk_t block;
+- struct fake_dirent *fde;
++ struct dx_root_info *dx_info;
+ int csum_size = 0;
+
+ if (ext4_has_metadata_csum(inode->i_sb))
+@@ -2096,18 +2094,19 @@ static int make_indexed_dir(handle_t *ha
+ brelse(bh);
+ return retval;
+ }
+- root = (struct dx_root *) bh->b_data;
++
++ dot_de = (struct ext4_dir_entry_2 *)bh->b_data;
++ dotdot_de = ext4_next_entry(dot_de, blocksize);
+
+ /* The 0th block becomes the root, move the dirents out */
+- fde = &root->dotdot;
+- de = (struct ext4_dir_entry_2 *)((char *)fde +
+- ext4_rec_len_from_disk(fde->rec_len, blocksize));
+- if ((char *) de >= (((char *) root) + blocksize)) {
++ de = (struct ext4_dir_entry_2 *)((char *)dotdot_de +
++ ext4_rec_len_from_disk(dotdot_de->rec_len, blocksize));
++ if ((char *)de >= (((char *)dot_de) + blocksize)) {
+ EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
+ brelse(bh);
+ return -EFSCORRUPTED;
+ }
+- len = ((char *) root) + (blocksize - csum_size) - (char *) de;
++ len = ((char *)dot_de) + (blocksize - csum_size) - (char *)de;
+
+ /* Allocate new block for the 0th block's dirents */
+ bh2 = ext4_append(handle, dir, &block);
+@@ -2130,19 +2129,24 @@ static int make_indexed_dir(handle_t *ha
+ ext4_initialize_dirent_tail(bh2, blocksize);
+
+ /* Initialize the root; the dot dirents already exist */
+- de = (struct ext4_dir_entry_2 *) (&root->dotdot);
+- de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
+- blocksize);
+- memset (&root->info, 0, sizeof(root->info));
+- root->info.info_length = sizeof(root->info);
+- root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
+- entries = root->entries;
++ dotdot_de->rec_len =
++ ext4_rec_len_to_disk(blocksize - le16_to_cpu(dot_de->rec_len),
++ blocksize);
++
++ /* initialize hashing info */
++ dx_info = dx_get_dx_info(dot_de);
++ memset(dx_info, 0, sizeof(*dx_info));
++ dx_info->info_length = sizeof(*dx_info);
++ dx_info->hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
++
++ entries = (void *)dx_info + sizeof(*dx_info);
++
+ dx_set_block(entries, 1);
+ dx_set_count(entries, 1);
+- dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
++ dx_set_limit(entries, dx_root_limit(dir, sizeof(*dx_info)));
+
+ /* Initialize as for dx_probe */
+- fname->hinfo.hash_version = root->info.hash_version;
++ fname->hinfo.hash_version = dx_info->hash_version;
+ if (fname->hinfo.hash_version <= DX_HASH_TEA)
+ fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
+ fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
+@@ -2506,7 +2510,7 @@ again:
+ goto journal_error;
+ }
+ } else {
+- struct dx_root *dxroot;
++ struct dx_root_info *info;
+ memcpy((char *) entries2, (char *) entries,
+ icount * sizeof(struct dx_entry));
+ dx_set_limit(entries2, dx_node_limit(dir));
+@@ -2514,8 +2518,9 @@ again:
+ /* Set up root */
+ dx_set_count(entries, 1);
+ dx_set_block(entries + 0, newblock);
+- dxroot = (struct dx_root *)frames[0].bh->b_data;
+- dxroot->info.indirect_levels += 1;
++ info = dx_get_dx_info((struct ext4_dir_entry_2 *)
++ frames[0].bh->b_data);
++ info->indirect_levels = 1;
+ dxtrace(printk(KERN_DEBUG
+ "Creating %d level index...\n",
+ dxroot->info.indirect_levels));
--- /dev/null
+Add a proc interface for max_dir_size.
+
+---
+ fs/ext4/sysfs.c | 10 +++++++++-
+ 1 file changed, 9 insertions(+), 1 deletion(-)
+
+--- a/fs/ext4/sysfs.c
++++ b/fs/ext4/sysfs.c
+@@ -209,6 +209,8 @@ EXT4_ATTR_FUNC(reserved_clusters, 0644);
+ EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, inode_readahead,
+ ext4_sb_info, s_inode_readahead_blks);
+ 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(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);
+@@ -252,6 +254,8 @@ static struct attribute *ext4_attrs[] =
+ ATTR_LIST(reserved_clusters),
+ ATTR_LIST(inode_readahead_blks),
+ ATTR_LIST(inode_goal),
++ ATTR_LIST(max_dir_size),
++ ATTR_LIST(max_dir_size_kb),
+ ATTR_LIST(mb_stats),
+ ATTR_LIST(mb_max_to_scan),
+ ATTR_LIST(mb_min_to_scan),
+@@ -376,7 +380,9 @@ static ssize_t ext4_attr_show(struct kob
+ le32_to_cpup(ptr));
+ else
+ return snprintf(buf, PAGE_SIZE, "%u\n",
+- *((unsigned int *) ptr));
++ strcmp("max_dir_size", a->attr.name) ?
++ *((unsigned int *) ptr) :
++ (*((unsigned int *) ptr)) << 10);
+ case attr_pointer_ul:
+ if (!ptr)
+ return 0;
+@@ -439,6 +445,8 @@ static ssize_t ext4_attr_store(struct ko
+ ret = kstrtoul(skip_spaces(buf), 0, &t);
+ if (ret)
+ return ret;
++ if (strcmp("max_dir_size", a->attr.name) == 0)
++ t >>= 10;
+ if (a->attr_ptr == ptr_ext4_super_block_offset)
+ *((__le32 *) ptr) = cpu_to_le32(t);
+ else
--- /dev/null
+---
+ fs/ext4/ext4.h | 1
+ fs/ext4/mballoc.c | 103 +++++++++++++++++++++++++++++++++++++++++++++++-------
+ fs/ext4/mballoc.h | 2 -
+ 3 files changed, 93 insertions(+), 13 deletions(-)
+
+--- a/fs/ext4/ext4.h
++++ b/fs/ext4/ext4.h
+@@ -3156,6 +3156,7 @@ struct ext4_group_info {
+ ext4_grpblk_t bb_fragments; /* nr of freespace fragments */
+ ext4_grpblk_t bb_largest_free_order;/* order of largest frag in BG */
+ struct list_head bb_prealloc_list;
++ unsigned long bb_prealloc_nr;
+ #ifdef DOUBLE_CHECK
+ void *bb_bitmap;
+ #endif
+--- a/fs/ext4/mballoc.c
++++ b/fs/ext4/mballoc.c
+@@ -345,7 +345,7 @@ static const char * const ext4_groupinfo
+ "ext4_groupinfo_64k", "ext4_groupinfo_128k"
+ };
+
+-static void ext4_mb_generate_from_pa(struct super_block *sb, void *bitmap,
++static int ext4_mb_generate_from_pa(struct super_block *sb, void *bitmap,
+ ext4_group_t group);
+ static void ext4_mb_generate_from_freelist(struct super_block *sb, void *bitmap,
+ ext4_group_t group);
+@@ -769,7 +769,7 @@ mb_set_largest_free_order(struct super_b
+ }
+
+ static noinline_for_stack
+-void ext4_mb_generate_buddy(struct super_block *sb,
++int ext4_mb_generate_buddy(struct super_block *sb,
+ void *buddy, void *bitmap, ext4_group_t group)
+ {
+ struct ext4_group_info *grp = ext4_get_group_info(sb, group);
+@@ -813,6 +813,7 @@ void ext4_mb_generate_buddy(struct super
+ grp->bb_free = free;
+ ext4_mark_group_bitmap_corrupted(sb, group,
+ EXT4_GROUP_INFO_BBITMAP_CORRUPT);
++ return -EIO;
+ }
+ mb_set_largest_free_order(sb, grp);
+
+@@ -823,6 +824,8 @@ void ext4_mb_generate_buddy(struct super
+ sbi->s_mb_buddies_generated++;
+ sbi->s_mb_generation_time += period;
+ spin_unlock(&sbi->s_bal_lock);
++
++ return 0;
+ }
+
+ static void mb_regenerate_buddy(struct ext4_buddy *e4b)
+@@ -943,7 +946,7 @@ static int ext4_mb_init_cache(struct pag
+ }
+
+ first_block = page->index * blocks_per_page;
+- for (i = 0; i < blocks_per_page; i++) {
++ for (i = 0; i < blocks_per_page && err == 0; i++) {
+ group = (first_block + i) >> 1;
+ if (group >= ngroups)
+ break;
+@@ -987,7 +990,7 @@ static int ext4_mb_init_cache(struct pag
+ ext4_lock_group(sb, group);
+ /* init the buddy */
+ memset(data, 0xff, blocksize);
+- ext4_mb_generate_buddy(sb, data, incore, group);
++ err = ext4_mb_generate_buddy(sb, data, incore, group);
+ ext4_unlock_group(sb, group);
+ incore = NULL;
+ } else {
+@@ -1002,7 +1005,7 @@ static int ext4_mb_init_cache(struct pag
+ memcpy(data, bitmap, blocksize);
+
+ /* mark all preallocated blks used in in-core bitmap */
+- ext4_mb_generate_from_pa(sb, data, group);
++ err = ext4_mb_generate_from_pa(sb, data, group);
+ ext4_mb_generate_from_freelist(sb, data, group);
+ ext4_unlock_group(sb, group);
+
+@@ -1012,7 +1015,8 @@ static int ext4_mb_init_cache(struct pag
+ incore = data;
+ }
+ }
+- SetPageUptodate(page);
++ if (likely(err == 0))
++ SetPageUptodate(page);
+
+ out:
+ if (bh) {
+@@ -2396,9 +2400,11 @@ static void *ext4_mb_seq_groups_next(str
+ static int ext4_mb_seq_groups_show(struct seq_file *seq, void *v)
+ {
+ struct super_block *sb = PDE_DATA(file_inode(seq->file));
++ struct ext4_group_desc *gdp;
+ ext4_group_t group = (ext4_group_t) ((unsigned long) v);
+ int i;
+ int err, buddy_loaded = 0;
++ int free = 0;
+ struct ext4_buddy e4b;
+ struct ext4_group_info *grinfo;
+ unsigned char blocksize_bits = min_t(unsigned char,
+@@ -2411,7 +2417,7 @@ static int ext4_mb_seq_groups_show(struc
+
+ group--;
+ if (group == 0)
+- seq_puts(seq, "#group: free frags first ["
++ seq_puts(seq, "#group: bfree gfree frags first pa ["
+ " 2^0 2^1 2^2 2^3 2^4 2^5 2^6 "
+ " 2^7 2^8 2^9 2^10 2^11 2^12 2^13 ]\n");
+
+@@ -2429,13 +2435,19 @@ static int ext4_mb_seq_groups_show(struc
+ buddy_loaded = 1;
+ }
+
++ gdp = ext4_get_group_desc(sb, group, NULL);
++ if (gdp != NULL)
++ free = ext4_free_group_clusters(sb, gdp);
++
+ memcpy(&sg, ext4_get_group_info(sb, group), i);
+
+ if (buddy_loaded)
+ ext4_mb_unload_buddy(&e4b);
+
+- seq_printf(seq, "#%-5u: %-5u %-5u %-5u [", group, sg.info.bb_free,
+- sg.info.bb_fragments, sg.info.bb_first_free);
++ seq_printf(seq, "#%-5lu: %-5u %-5u %-5u %-5u %-5lu [",
++ (long unsigned int)group, sg.info.bb_free, free,
++ sg.info.bb_fragments, sg.info.bb_first_free,
++ sg.info.bb_prealloc_nr);
+ for (i = 0; i <= 13; i++)
+ seq_printf(seq, " %-5u", i <= blocksize_bits + 1 ?
+ sg.info.bb_counters[i] : 0);
+@@ -3742,22 +3754,71 @@ static void ext4_mb_generate_from_freeli
+ }
+
+ /*
++ * check free blocks in bitmap match free block in group descriptor
++ * do this before taking preallocated blocks into account to be able
++ * to detect on-disk corruptions. The group lock should be hold by the
++ * caller.
++ */
++int ext4_mb_check_ondisk_bitmap(struct super_block *sb, void *bitmap,
++ struct ext4_group_desc *gdp, int group)
++{
++ unsigned short max = EXT4_CLUSTERS_PER_GROUP(sb);
++ unsigned short i, first, free = 0;
++ unsigned short free_in_gdp = ext4_free_group_clusters(sb, gdp);
++
++ if (free_in_gdp == 0 && gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))
++ return 0;
++
++ i = mb_find_next_zero_bit(bitmap, max, 0);
++
++ while (i < max) {
++ first = i;
++ i = mb_find_next_bit(bitmap, max, i);
++ if (i > max)
++ i = max;
++ free += i - first;
++ if (i < max)
++ i = mb_find_next_zero_bit(bitmap, max, i);
++ }
++
++ if (free != free_in_gdp) {
++ ext4_error(sb, "on-disk bitmap for group %d"
++ "corrupted: %u blocks free in bitmap, %u - in gd\n",
++ group, free, free_in_gdp);
++ return -EIO;
++ }
++ return 0;
++}
++
++/*
+ * the function goes through all preallocation in this group and marks them
+ * used in in-core bitmap. buddy must be generated from this bitmap
+ * Need to be called with ext4 group lock held
+ */
+ static noinline_for_stack
+-void ext4_mb_generate_from_pa(struct super_block *sb, void *bitmap,
++int ext4_mb_generate_from_pa(struct super_block *sb, void *bitmap,
+ ext4_group_t group)
+ {
+ struct ext4_group_info *grp = ext4_get_group_info(sb, group);
+ struct ext4_prealloc_space *pa;
++ struct ext4_group_desc *gdp;
+ struct list_head *cur;
+ ext4_group_t groupnr;
+ ext4_grpblk_t start;
+ int preallocated = 0;
++ int skip = 0, count = 0;
++ int err;
+ int len;
+
++ gdp = ext4_get_group_desc(sb, group, NULL);
++ if (gdp == NULL)
++ return -EIO;
++
++ /* before applying preallocations, check bitmap consistency */
++ err = ext4_mb_check_ondisk_bitmap(sb, bitmap, gdp, group);
++ if (err)
++ return err;
++
+ /* all form of preallocation discards first load group,
+ * so the only competing code is preallocation use.
+ * we don't need any locking here
+@@ -3773,13 +3834,23 @@ void ext4_mb_generate_from_pa(struct sup
+ &groupnr, &start);
+ len = pa->pa_len;
+ spin_unlock(&pa->pa_lock);
+- if (unlikely(len == 0))
++ if (unlikely(len == 0)) {
++ skip++;
+ continue;
++ }
+ BUG_ON(groupnr != group);
+ ext4_set_bits(bitmap, start, len);
+ preallocated += len;
++ count++;
++ }
++ if (count + skip != grp->bb_prealloc_nr) {
++ ext4_error(sb, "lost preallocations: "
++ "count %d, bb_prealloc_nr %lu, skip %d\n",
++ count, grp->bb_prealloc_nr, skip);
++ return -EIO;
+ }
+ mb_debug(sb, "preallocated %d for group %u\n", preallocated, group);
++ return 0;
+ }
+
+ static void ext4_mb_pa_callback(struct rcu_head *head)
+@@ -3843,6 +3914,7 @@ static void ext4_mb_put_pa(struct ext4_a
+ */
+ ext4_lock_group(sb, grp);
+ list_del(&pa->pa_group_list);
++ ext4_get_group_info(sb, grp)->bb_prealloc_nr--;
+ ext4_unlock_group(sb, grp);
+
+ spin_lock(pa->pa_obj_lock);
+@@ -3934,6 +4006,7 @@ ext4_mb_new_inode_pa(struct ext4_allocat
+ pa->pa_inode = ac->ac_inode;
+
+ list_add(&pa->pa_group_list, &grp->bb_prealloc_list);
++ grp->bb_prealloc_nr++;
+
+ spin_lock(pa->pa_obj_lock);
+ list_add_rcu(&pa->pa_inode_list, &ei->i_prealloc_list);
+@@ -3988,6 +4061,7 @@ ext4_mb_new_group_pa(struct ext4_allocat
+ pa->pa_inode = NULL;
+
+ list_add(&pa->pa_group_list, &grp->bb_prealloc_list);
++ grp->bb_prealloc_nr++;
+
+ /*
+ * We will later add the new pa to the right bucket
+@@ -4155,6 +4229,8 @@ repeat:
+
+ spin_unlock(&pa->pa_lock);
+
++ BUG_ON(grp->bb_prealloc_nr == 0);
++ grp->bb_prealloc_nr--;
+ list_del(&pa->pa_group_list);
+ list_add(&pa->u.pa_tmp_list, &list);
+ }
+@@ -4291,7 +4367,7 @@ repeat:
+ if (err) {
+ ext4_error_err(sb, -err, "Error %d loading buddy information for %u",
+ err, group);
+- continue;
++ return;
+ }
+
+ bitmap_bh = ext4_read_block_bitmap(sb, group);
+@@ -4304,6 +4380,8 @@ repeat:
+ }
+
+ ext4_lock_group(sb, group);
++ BUG_ON(e4b.bd_info->bb_prealloc_nr == 0);
++ e4b.bd_info->bb_prealloc_nr--;
+ list_del(&pa->pa_group_list);
+ ext4_mb_release_inode_pa(&e4b, bitmap_bh, pa);
+ ext4_unlock_group(sb, group);
+@@ -4598,6 +4676,7 @@ ext4_mb_discard_lg_preallocations(struct
+ }
+ ext4_lock_group(sb, group);
+ list_del(&pa->pa_group_list);
++ ext4_get_group_info(sb, group)->bb_prealloc_nr--;
+ ext4_mb_release_group_pa(&e4b, pa);
+ ext4_unlock_group(sb, group);
+
+--- a/fs/ext4/mballoc.h
++++ b/fs/ext4/mballoc.h
+@@ -66,7 +66,7 @@
+ /*
+ * for which requests use 2^N search using buddies
+ */
+-#define MB_DEFAULT_ORDER2_REQS 2
++#define MB_DEFAULT_ORDER2_REQS 8
+
+ /*
+ * default group prealloc size 512 blocks
--- /dev/null
+---
+ fs/ext4/mballoc.c | 43 +++++++++++++++++++++++++++++++++++++------
+ fs/ext4/mballoc.h | 2 ++
+ 2 files changed, 39 insertions(+), 6 deletions(-)
+
+--- a/fs/ext4/mballoc.c
++++ b/fs/ext4/mballoc.c
+@@ -3992,6 +3992,7 @@ ext4_mb_new_inode_pa(struct ext4_allocat
+ INIT_LIST_HEAD(&pa->pa_group_list);
+ pa->pa_deleted = 0;
+ pa->pa_type = MB_INODE_PA;
++ pa->pa_error = 0;
+
+ mb_debug(sb, "new inode pa %p: %llu/%d for %u\n", pa, pa->pa_pstart,
+ pa->pa_len, pa->pa_lstart);
+@@ -4046,6 +4047,7 @@ ext4_mb_new_group_pa(struct ext4_allocat
+ INIT_LIST_HEAD(&pa->pa_group_list);
+ pa->pa_deleted = 0;
+ pa->pa_type = MB_GROUP_PA;
++ pa->pa_error = 0;
+
+ mb_debug(sb, "new group pa %p: %llu/%d for %u\n", pa, pa->pa_pstart,
+ pa->pa_len, pa->pa_lstart);
+@@ -4098,7 +4100,9 @@ ext4_mb_release_inode_pa(struct ext4_bud
+ unsigned long long grp_blk_start;
+ int free = 0;
+
++ assert_spin_locked(ext4_group_lock_ptr(sb, e4b->bd_group));
+ BUG_ON(pa->pa_deleted == 0);
++ BUG_ON(pa->pa_inode == NULL);
+ ext4_get_group_no_and_offset(sb, pa->pa_pstart, &group, &bit);
+ grp_blk_start = pa->pa_pstart - EXT4_C2B(sbi, bit);
+ BUG_ON(group != e4b->bd_group && pa->pa_len != 0);
+@@ -4121,12 +4125,18 @@ ext4_mb_release_inode_pa(struct ext4_bud
+ mb_free_blocks(pa->pa_inode, e4b, bit, next - bit);
+ bit = next + 1;
+ }
+- if (free != pa->pa_free) {
+- ext4_msg(e4b->bd_sb, KERN_CRIT,
+- "pa %p: logic %lu, phys. %lu, len %d",
+- pa, (unsigned long) pa->pa_lstart,
+- (unsigned long) pa->pa_pstart,
+- pa->pa_len);
++
++ /* "free < pa->pa_free" means we maybe double alloc the same blocks,
++ * otherwise maybe leave some free blocks unavailable, no need to BUG.*/
++ if ((free > pa->pa_free && !pa->pa_error) || (free < pa->pa_free)) {
++ ext4_error(sb, "pa free mismatch: [pa %p] "
++ "[phy %lu] [logic %lu] [len %u] [free %u] "
++ "[error %u] [inode %d] [freed %u]", pa,
++ (unsigned long)pa->pa_pstart,
++ (unsigned long)pa->pa_lstart,
++ pa->pa_len, (unsigned)pa->pa_free,
++ (unsigned)pa->pa_error, pa->pa_inode->i_ino,
++ free);
+ ext4_grp_locked_error(sb, group, 0, 0, "free %u, pa_free %u",
+ free, pa->pa_free);
+ /*
+@@ -4134,6 +4144,8 @@ ext4_mb_release_inode_pa(struct ext4_bud
+ * from the bitmap and continue.
+ */
+ }
++ /* do not verify if the file system is being umounted */
++ BUG_ON(atomic_read(&sb->s_active) > 0 && pa->pa_free != free);
+ atomic_add(free, &sbi->s_mb_discarded);
+
+ return 0;
+@@ -4955,6 +4967,25 @@ errout:
+ ac->ac_b_ex.fe_len = 0;
+ ar->len = 0;
+ ext4_mb_show_ac(ac);
++ if (ac->ac_pa) {
++ struct ext4_prealloc_space *pa = ac->ac_pa;
++
++ /* We can not make sure whether the bitmap has
++ * been updated or not when fail case. So can
++ * not revert pa_free back, just mark pa_error*/
++ pa->pa_error++;
++ ext4_error(sb,
++ "Updating bitmap error: [err %d] "
++ "[pa %p] [phy %lu] [logic %lu] "
++ "[len %u] [free %u] [error %u] "
++ "[inode %lu]", *errp, pa,
++ (unsigned long)pa->pa_pstart,
++ (unsigned long)pa->pa_lstart,
++ (unsigned)pa->pa_len,
++ (unsigned)pa->pa_free,
++ (unsigned)pa->pa_error,
++ pa->pa_inode ? pa->pa_inode->i_ino : 0);
++ }
+ }
+ ext4_mb_release_context(ac);
+ out:
+--- a/fs/ext4/mballoc.h
++++ b/fs/ext4/mballoc.h
+@@ -20,6 +20,7 @@
+ #include <linux/seq_file.h>
+ #include <linux/blkdev.h>
+ #include <linux/mutex.h>
++#include <linux/genhd.h>
+ #include "ext4_jbd2.h"
+ #include "ext4.h"
+
+@@ -107,6 +108,7 @@ struct ext4_prealloc_space {
+ ext4_grpblk_t pa_len; /* len of preallocated chunk */
+ ext4_grpblk_t pa_free; /* how many blocks are free */
+ unsigned short pa_type; /* pa type. inode or group */
++ unsigned short pa_error;
+ spinlock_t *pa_obj_lock;
+ struct inode *pa_inode; /* hack, for history only */
+ };
--- /dev/null
+---
+ fs/ext4/ext4.h | 23 ++++++++++++++++++++++-
+ fs/ext4/ialloc.c | 3 ++-
+ fs/ext4/inode.c | 15 +++++++++++++++
+ fs/ext4/namei.c | 9 ++++++---
+ fs/ext4/super.c | 10 ++--------
+ 5 files changed, 47 insertions(+), 13 deletions(-)
+
+--- a/fs/ext4/ext4.h
++++ b/fs/ext4/ext4.h
+@@ -1759,6 +1759,8 @@ static inline bool ext4_verity_in_progre
+
+ #define NEXT_ORPHAN(inode) EXT4_I(inode)->i_dtime
+
++#define JOURNAL_START_HAS_3ARGS 1
++
+ /*
+ * Codes for operating systems
+ */
+@@ -1990,7 +1992,21 @@ static inline bool ext4_has_unknown_ext#
+
+ EXTN_FEATURE_FUNCS(2)
+ EXTN_FEATURE_FUNCS(3)
+-EXTN_FEATURE_FUNCS(4)
++static inline bool ext4_has_unknown_ext4_compat_features(struct super_block *sb)
++{
++ return ((EXT4_SB(sb)->s_es->s_feature_compat &
++ cpu_to_le32(~EXT4_FEATURE_COMPAT_SUPP)) != 0);
++}
++static inline bool ext4_has_unknown_ext4_ro_compat_features(struct super_block *sb)
++{
++ return ((EXT4_SB(sb)->s_es->s_feature_ro_compat &
++ cpu_to_le32(~EXT4_FEATURE_RO_COMPAT_SUPP)) != 0);
++}
++static inline bool ext4_has_unknown_ext4_incompat_features(struct super_block *sb)
++{
++ return ((EXT4_SB(sb)->s_es->s_feature_incompat &
++ cpu_to_le32(~EXT4_FEATURE_INCOMPAT_SUPP)) != 0);
++}
+
+ static inline bool ext4_has_compat_features(struct super_block *sb)
+ {
+@@ -3393,6 +3409,11 @@ struct ext4_extent;
+ #define EXT_MAX_BLOCKS 0xffffffff
+
+ extern void ext4_ext_tree_init(handle_t *handle, struct inode *inode);
++extern struct buffer_head *ext4_read_inode_bitmap(struct super_block *sb,
++ ext4_group_t block_group);
++extern struct buffer_head *ext4_append(handle_t *handle,
++ struct inode *inode,
++ ext4_lblk_t *block);
+ extern int ext4_ext_index_trans_blocks(struct inode *inode, int extents);
+ extern int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
+ struct ext4_map_blocks *map, int flags);
+--- a/fs/ext4/ialloc.c
++++ b/fs/ext4/ialloc.c
+@@ -115,7 +115,7 @@ verified:
+ *
+ * Return buffer_head of bitmap on success, or an ERR_PTR on error.
+ */
+-static struct buffer_head *
++struct buffer_head *
+ ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
+ {
+ struct ext4_group_desc *desc;
+@@ -213,6 +213,7 @@ out:
+ put_bh(bh);
+ return ERR_PTR(err);
+ }
++EXPORT_SYMBOL(ext4_read_inode_bitmap);
+
+ /*
+ * NOTE! When we get the inode, we're the only people
+--- a/fs/ext4/inode.c
++++ b/fs/ext4/inode.c
+@@ -6065,3 +6065,18 @@ vm_fault_t ext4_filemap_fault(struct vm_
+
+ return ret;
+ }
++EXPORT_SYMBOL(ext4_map_blocks);
++EXPORT_SYMBOL(ext4_truncate);
++EXPORT_SYMBOL(ext4_iget);
++EXPORT_SYMBOL(ext4_bread);
++EXPORT_SYMBOL(ext4_itable_unused_count);
++EXPORT_SYMBOL(ext4_force_commit);
++EXPORT_SYMBOL(__ext4_mark_inode_dirty);
++EXPORT_SYMBOL(ext4_get_group_desc);
++EXPORT_SYMBOL(__ext4_journal_get_write_access);
++EXPORT_SYMBOL(__ext4_journal_start_sb);
++EXPORT_SYMBOL(__ext4_journal_stop);
++EXPORT_SYMBOL(__ext4_handle_dirty_metadata);
++EXPORT_SYMBOL(__ext4_std_error);
++EXPORT_SYMBOL(ext4fs_dirhash);
++EXPORT_SYMBOL(ext4_get_inode_loc);
+--- a/fs/ext4/namei.c
++++ b/fs/ext4/namei.c
+@@ -50,7 +50,7 @@
+ #define NAMEI_RA_BLOCKS 4
+ #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
+
+-static struct buffer_head *ext4_append(handle_t *handle,
++struct buffer_head *ext4_append(handle_t *handle,
+ struct inode *inode,
+ ext4_lblk_t *block)
+ {
+@@ -181,6 +181,7 @@ static struct buffer_head *__ext4_read_d
+ }
+ return bh;
+ }
++EXPORT_SYMBOL(ext4_append);
+
+ #ifndef assert
+ #define assert(test) J_ASSERT(test)
+@@ -2572,23 +2573,25 @@ EXPORT_SYMBOL(ext4_delete_entry);
+ * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
+ * on regular files) and to avoid creating huge/slow non-HTREE directories.
+ */
+-static void ext4_inc_count(handle_t *handle, struct inode *inode)
++void ext4_inc_count(handle_t *handle, struct inode *inode)
+ {
+ inc_nlink(inode);
+ if (is_dx(inode) &&
+ (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
+ set_nlink(inode, 1);
+ }
++EXPORT_SYMBOL(ext4_inc_count);
+
+ /*
+ * If a directory had nlink == 1, then we should let it be 1. This indicates
+ * directory has >EXT4_LINK_MAX subdirs.
+ */
+-static void ext4_dec_count(handle_t *handle, struct inode *inode)
++void ext4_dec_count(handle_t *handle, struct inode *inode)
+ {
+ if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
+ drop_nlink(inode);
+ }
++EXPORT_SYMBOL(ext4_dec_count);
+
+
+ /*
+--- a/fs/ext4/super.c
++++ b/fs/ext4/super.c
+@@ -347,7 +347,7 @@ static void __save_error_info(struct sup
+ return;
+ es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
+ ext4_update_tstamp(es, s_last_error_time);
+- strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
++ strlcpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
+ es->s_last_error_line = cpu_to_le32(line);
+ es->s_last_error_ino = cpu_to_le32(ino);
+ es->s_last_error_block = cpu_to_le64(block);
+@@ -408,7 +408,7 @@ static void __save_error_info(struct sup
+ if (!es->s_first_error_time) {
+ es->s_first_error_time = es->s_last_error_time;
+ es->s_first_error_time_hi = es->s_last_error_time_hi;
+- strncpy(es->s_first_error_func, func,
++ strlcpy(es->s_first_error_func, func,
+ sizeof(es->s_first_error_func));
+ es->s_first_error_line = cpu_to_le32(line);
+ es->s_first_error_ino = es->s_last_error_ino;
+@@ -6315,16 +6315,12 @@ static int __init ext4_init_fs(void)
+ err = init_inodecache();
+ if (err)
+ goto out1;
+- register_as_ext3();
+- register_as_ext2();
+ err = register_filesystem(&ext4_fs_type);
+ if (err)
+ goto out;
+
+ return 0;
+ out:
+- unregister_as_ext2();
+- unregister_as_ext3();
+ destroy_inodecache();
+ out1:
+ ext4_exit_mballoc();
+@@ -6347,8 +6343,6 @@ out7:
+ static void __exit ext4_exit_fs(void)
+ {
+ ext4_destroy_lazyinit_thread();
+- unregister_as_ext2();
+- unregister_as_ext3();
+ unregister_filesystem(&ext4_fs_type);
+ destroy_inodecache();
+ ext4_exit_mballoc();
--- /dev/null
+---
+ fs/ext4/ext4.h | 1 +
+ fs/ext4/namei.c | 6 ++++--
+ 2 files changed, 5 insertions(+), 2 deletions(-)
+
+--- a/fs/ext4/ext4.h
++++ b/fs/ext4/ext4.h
+@@ -1716,6 +1716,7 @@ enum {
+ EXT4_STATE_NO_EXPAND, /* No space for expansion */
+ EXT4_STATE_DA_ALLOC_CLOSE, /* Alloc DA blks on close */
+ EXT4_STATE_EXT_MIGRATE, /* Inode is migrating */
++ EXT4_STATE_IAM, /* Lustre IAM objects */
+ EXT4_STATE_NEWENTRY, /* File just added to dir */
+ EXT4_STATE_MAY_INLINE_DATA, /* may have in-inode data */
+ EXT4_STATE_EXT_PRECACHED, /* extents have been precached */
+--- a/fs/ext4/namei.c
++++ b/fs/ext4/namei.c
+@@ -60,8 +60,10 @@ struct buffer_head *ext4_append(handle_t
+
+ if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
+ ((inode->i_size >> 10) >=
+- EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
+- return ERR_PTR(-ENOSPC);
++ EXT4_SB(inode->i_sb)->s_max_dir_size_kb))) {
++ if (!ext4_test_inode_state(inode, EXT4_STATE_IAM))
++ return ERR_PTR(-ENOSPC);
++ }
+
+ /* with parallel dir operations all appends
+ * have to be serialized -bzzz */
--- /dev/null
+From 1a0f7f0b9c13ef0aa86e125f350b6733bff8db3c Mon Sep 17 00:00:00 2001
+From: Shaun Tancheff <stancheff@cray.com>
+Date: Wed, 15 Jan 2020 07:35:13 -0600
+Subject: [PATCH] Single directory performance is a critical for HPC workloads.
+ In a typical use case an application creates a separate output file for each
+ node and task in a job. As nodes and tasks increase, hundreds of thousands of
+ files may be created in a single directory within a short window of time.
+ Today, both filename lookup and file system modifying operations (such as
+ create and unlink) are protected with a single lock for an entire ldiskfs
+ directory. PDO project will remove this bottleneck by introducing a parallel
+ locking mechanism for entire ldiskfs directories. This work will enable
+ multiple application threads to simultaneously lookup, create and unlink in
+ parallel.
+
+This patch contains:
+ - pdirops support for ldiskfs
+ - integrate with osd-ldiskfs
+---
+ fs/ext4/Makefile | 1
+ fs/ext4/ext4.h | 78 +++++++++
+ fs/ext4/namei.c | 454 ++++++++++++++++++++++++++++++++++++++++++++++++++-----
+ fs/ext4/super.c | 1
+ 4 files changed, 494 insertions(+), 40 deletions(-)
+ create mode 100644 fs/ext4/htree_lock.c
+ create mode 100644 include/linux/htree_lock.h
+
+--- a/fs/ext4/Makefile
++++ b/fs/ext4/Makefile
+@@ -7,6 +7,7 @@ obj-$(CONFIG_EXT4_FS) += ext4.o
+
+ ext4-y := balloc.o bitmap.o block_validity.o dir.o ext4_jbd2.o extents.o \
+ extents_status.o file.o fsmap.o fsync.o hash.o ialloc.o \
++ htree_lock.o \
+ indirect.o inline.o inode.o ioctl.o mballoc.o migrate.o \
+ mmp.o move_extent.o namei.o page-io.o readpage.o resize.o \
+ super.o symlink.o sysfs.o xattr.o xattr_hurd.o xattr_trusted.o \
+--- a/fs/ext4/ext4.h
++++ b/fs/ext4/ext4.h
+@@ -29,6 +29,7 @@
+ #include <linux/timer.h>
+ #include <linux/version.h>
+ #include <linux/wait.h>
++#include <linux/htree_lock.h>
+ #include <linux/sched/signal.h>
+ #include <linux/blockgroup_lock.h>
+ #include <linux/percpu_counter.h>
+@@ -987,6 +988,9 @@ struct ext4_inode_info {
+ __u32 i_dtime;
+ ext4_fsblk_t i_file_acl;
+
++ /* following fields for parallel directory operations -bzzz */
++ struct semaphore i_append_sem;
++
+ /*
+ * i_block_group is the number of the block group which contains
+ * this file's inode. Constant across the lifetime of the inode,
+@@ -2299,6 +2303,72 @@ struct dx_hash_info
+ */
+ #define HASH_NB_ALWAYS 1
+
++/* assume name-hash is protected by upper layer */
++#define EXT4_HTREE_LOCK_HASH 0
++
++enum ext4_pdo_lk_types {
++#if EXT4_HTREE_LOCK_HASH
++ EXT4_LK_HASH,
++#endif
++ EXT4_LK_DX, /* index block */
++ EXT4_LK_DE, /* directory entry block */
++ EXT4_LK_SPIN, /* spinlock */
++ EXT4_LK_MAX,
++};
++
++/* read-only bit */
++#define EXT4_LB_RO(b) (1 << (b))
++/* read + write, high bits for writer */
++#define EXT4_LB_RW(b) ((1 << (b)) | (1 << (EXT4_LK_MAX + (b))))
++
++enum ext4_pdo_lock_bits {
++ /* DX lock bits */
++ EXT4_LB_DX_RO = EXT4_LB_RO(EXT4_LK_DX),
++ EXT4_LB_DX = EXT4_LB_RW(EXT4_LK_DX),
++ /* DE lock bits */
++ EXT4_LB_DE_RO = EXT4_LB_RO(EXT4_LK_DE),
++ EXT4_LB_DE = EXT4_LB_RW(EXT4_LK_DE),
++ /* DX spinlock bits */
++ EXT4_LB_SPIN_RO = EXT4_LB_RO(EXT4_LK_SPIN),
++ EXT4_LB_SPIN = EXT4_LB_RW(EXT4_LK_SPIN),
++ /* accurate searching */
++ EXT4_LB_EXACT = EXT4_LB_RO(EXT4_LK_MAX << 1),
++};
++
++enum ext4_pdo_lock_opc {
++ /* external */
++ EXT4_HLOCK_READDIR = (EXT4_LB_DE_RO | EXT4_LB_DX_RO),
++ EXT4_HLOCK_LOOKUP = (EXT4_LB_DE_RO | EXT4_LB_SPIN_RO |
++ EXT4_LB_EXACT),
++ EXT4_HLOCK_DEL = (EXT4_LB_DE | EXT4_LB_SPIN_RO |
++ EXT4_LB_EXACT),
++ EXT4_HLOCK_ADD = (EXT4_LB_DE | EXT4_LB_SPIN_RO),
++
++ /* internal */
++ EXT4_HLOCK_LOOKUP_SAFE = (EXT4_LB_DE_RO | EXT4_LB_DX_RO |
++ EXT4_LB_EXACT),
++ EXT4_HLOCK_DEL_SAFE = (EXT4_LB_DE | EXT4_LB_DX_RO | EXT4_LB_EXACT),
++ EXT4_HLOCK_SPLIT = (EXT4_LB_DE | EXT4_LB_DX | EXT4_LB_SPIN),
++};
++
++extern struct htree_lock_head *ext4_htree_lock_head_alloc(unsigned hbits);
++#define ext4_htree_lock_head_free(lhead) htree_lock_head_free(lhead)
++
++extern struct htree_lock *ext4_htree_lock_alloc(void);
++#define ext4_htree_lock_free(lck) htree_lock_free(lck)
++
++extern void ext4_htree_lock(struct htree_lock *lck,
++ struct htree_lock_head *lhead,
++ struct inode *dir, unsigned flags);
++#define ext4_htree_unlock(lck) htree_unlock(lck)
++
++extern struct buffer_head *ext4_find_entry_locked(struct inode *dir,
++ const struct qstr *d_name,
++ struct ext4_dir_entry_2 **res_dir,
++ int *inlined, struct htree_lock *lck);
++extern int ext4_add_entry_locked(handle_t *handle, struct dentry *dentry,
++ struct inode *inode, struct htree_lock *lck);
++
+ struct ext4_filename {
+ const struct qstr *usr_fname;
+ struct fscrypt_str disk_name;
+@@ -2666,11 +2736,19 @@ void ext4_insert_dentry(struct inode *in
+ struct ext4_filename *fname, void *data);
+ static inline void ext4_update_dx_flag(struct inode *inode)
+ {
++ /* Disable it for ldiskfs, because going from a DX directory to
++ * a non-DX directory while it is in use will completely break
++ * the htree-locking.
++ * If we really want to support this operation in the future,
++ * we need to exclusively lock the directory at here which will
++ * increase complexity of code */
++#if 0
+ if (!ext4_has_feature_dir_index(inode->i_sb)) {
+ /* ext4_iget() should have caught this... */
+ WARN_ON_ONCE(ext4_has_feature_metadata_csum(inode->i_sb));
+ ext4_clear_inode_flag(inode, EXT4_INODE_INDEX);
+ }
++#endif
+ }
+ static const unsigned char ext4_filetype_table[] = {
+ DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
+--- a/fs/ext4/namei.c
++++ b/fs/ext4/namei.c
+@@ -55,6 +55,7 @@ struct buffer_head *ext4_append(handle_t
+ ext4_lblk_t *block)
+ {
+ struct buffer_head *bh;
++ struct ext4_inode_info *ei = EXT4_I(inode);
+ int err;
+
+ if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
+@@ -62,15 +63,22 @@ struct buffer_head *ext4_append(handle_t
+ EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
+ return ERR_PTR(-ENOSPC);
+
++ /* with parallel dir operations all appends
++ * have to be serialized -bzzz */
++ down(&ei->i_append_sem);
++
+ *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
+
+ bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
+- if (IS_ERR(bh))
++ if (IS_ERR(bh)) {
++ up(&ei->i_append_sem);
+ return bh;
++ }
+ inode->i_size += inode->i_sb->s_blocksize;
+ EXT4_I(inode)->i_disksize = inode->i_size;
+ BUFFER_TRACE(bh, "get_write_access");
+ err = ext4_journal_get_write_access(handle, bh);
++ up(&ei->i_append_sem);
+ if (err) {
+ brelse(bh);
+ ext4_std_error(inode->i_sb, err);
+@@ -271,7 +279,8 @@ static unsigned dx_node_limit(struct ino
+ static struct dx_frame *dx_probe(struct ext4_filename *fname,
+ struct inode *dir,
+ struct dx_hash_info *hinfo,
+- struct dx_frame *frame);
++ struct dx_frame *frame,
++ struct htree_lock *lck);
+ static void dx_release(struct dx_frame *frames);
+ static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
+ unsigned blocksize, struct dx_hash_info *hinfo,
+@@ -285,12 +294,13 @@ static void dx_insert_block(struct dx_fr
+ static int ext4_htree_next_block(struct inode *dir, __u32 hash,
+ struct dx_frame *frame,
+ struct dx_frame *frames,
+- __u32 *start_hash);
++ __u32 *start_hash, struct htree_lock *lck);
+ static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
+ struct ext4_filename *fname,
+- struct ext4_dir_entry_2 **res_dir);
++ struct ext4_dir_entry_2 **res_dir, struct htree_lock *lck);
+ static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
+- struct inode *dir, struct inode *inode);
++ struct inode *dir, struct inode *inode,
++ struct htree_lock *lck);
+
+ /* checksumming functions */
+ void ext4_initialize_dirent_tail(struct buffer_head *bh,
+@@ -755,6 +765,227 @@ struct stats dx_show_entries(struct dx_h
+ }
+ #endif /* DX_DEBUG */
+
++/* private data for htree_lock */
++struct ext4_dir_lock_data {
++ unsigned ld_flags; /* bits-map for lock types */
++ unsigned ld_count; /* # entries of the last DX block */
++ struct dx_entry ld_at_entry; /* copy of leaf dx_entry */
++ struct dx_entry *ld_at; /* position of leaf dx_entry */
++};
++
++#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)
++
++static void ext4_htree_event_cb(void *target, void *event)
++{
++ u64 *block = (u64 *)target;
++
++ if (*block == dx_get_block((struct dx_entry *)event))
++ *block = EXT4_HTREE_NODE_CHANGED;
++}
++
++struct htree_lock_head *ext4_htree_lock_head_alloc(unsigned hbits)
++{
++ struct htree_lock_head *lhead;
++
++ lhead = htree_lock_head_alloc(EXT4_LK_MAX, hbits, 0);
++ if (lhead != NULL) {
++ htree_lock_event_attach(lhead, EXT4_LK_SPIN, HTREE_EVENT_WR,
++ ext4_htree_event_cb);
++ }
++ return lhead;
++}
++EXPORT_SYMBOL(ext4_htree_lock_head_alloc);
++
++struct htree_lock *ext4_htree_lock_alloc(void)
++{
++ return htree_lock_alloc(EXT4_LK_MAX,
++ sizeof(struct ext4_dir_lock_data));
++}
++EXPORT_SYMBOL(ext4_htree_lock_alloc);
++
++static htree_lock_mode_t ext4_htree_mode(unsigned flags)
++{
++ switch (flags) {
++ default: /* 0 or unknown flags require EX lock */
++ return HTREE_LOCK_EX;
++ case EXT4_HLOCK_READDIR:
++ return HTREE_LOCK_PR;
++ case EXT4_HLOCK_LOOKUP:
++ return HTREE_LOCK_CR;
++ case EXT4_HLOCK_DEL:
++ case EXT4_HLOCK_ADD:
++ return HTREE_LOCK_CW;
++ }
++}
++
++/* return PR for read-only operations, otherwise return EX */
++static inline htree_lock_mode_t ext4_htree_safe_mode(unsigned flags)
++{
++ int writer = (flags & EXT4_LB_DE) == EXT4_LB_DE;
++
++ /* 0 requires EX lock */
++ return (flags == 0 || writer) ? HTREE_LOCK_EX : HTREE_LOCK_PR;
++}
++
++static int ext4_htree_safe_locked(struct htree_lock *lck)
++{
++ int writer;
++
++ if (lck == NULL || lck->lk_mode == HTREE_LOCK_EX)
++ return 1;
++
++ writer = (ext4_htree_lock_data(lck)->ld_flags & EXT4_LB_DE) ==
++ EXT4_LB_DE;
++ if (writer) /* all readers & writers are excluded? */
++ return lck->lk_mode == HTREE_LOCK_EX;
++
++ /* all writers are excluded? */
++ return lck->lk_mode == HTREE_LOCK_PR ||
++ lck->lk_mode == HTREE_LOCK_PW ||
++ lck->lk_mode == HTREE_LOCK_EX;
++}
++
++/* relock htree_lock with EX mode if it's change operation, otherwise
++ * relock it with PR mode. It's noop if PDO is disabled. */
++static void ext4_htree_safe_relock(struct htree_lock *lck)
++{
++ if (!ext4_htree_safe_locked(lck)) {
++ unsigned flags = ext4_htree_lock_data(lck)->ld_flags;
++
++ htree_change_lock(lck, ext4_htree_safe_mode(flags));
++ }
++}
++
++void ext4_htree_lock(struct htree_lock *lck, struct htree_lock_head *lhead,
++ struct inode *dir, unsigned flags)
++{
++ htree_lock_mode_t mode = is_dx(dir) ? ext4_htree_mode(flags) :
++ ext4_htree_safe_mode(flags);
++
++ ext4_htree_lock_data(lck)->ld_flags = flags;
++ htree_lock(lck, lhead, mode);
++ if (!is_dx(dir))
++ ext4_htree_safe_relock(lck); /* make sure it's safe locked */
++}
++EXPORT_SYMBOL(ext4_htree_lock);
++
++static int ext4_htree_node_lock(struct htree_lock *lck, struct dx_entry *at,
++ unsigned lmask, int wait, void *ev)
++{
++ u32 key = (at == NULL) ? 0 : dx_get_block(at);
++ u32 mode;
++
++ /* NOOP if htree is well protected or caller doesn't require the lock */
++ if (ext4_htree_safe_locked(lck) ||
++ !(ext4_htree_lock_data(lck)->ld_flags & lmask))
++ return 1;
++
++ mode = (ext4_htree_lock_data(lck)->ld_flags & lmask) == lmask ?
++ HTREE_LOCK_PW : HTREE_LOCK_PR;
++ while (1) {
++ if (htree_node_lock_try(lck, mode, key, ffz(~lmask), wait, ev))
++ return 1;
++ if (!(lmask & EXT4_LB_SPIN)) /* not a spinlock */
++ return 0;
++ cpu_relax(); /* spin until granted */
++ }
++}
++
++static int ext4_htree_node_locked(struct htree_lock *lck, unsigned lmask)
++{
++ return ext4_htree_safe_locked(lck) ||
++ htree_node_is_granted(lck, ffz(~lmask));
++}
++
++static void ext4_htree_node_unlock(struct htree_lock *lck,
++ unsigned lmask, void *buf)
++{
++ /* NB: it's safe to call mutiple times or even it's not locked */
++ if (!ext4_htree_safe_locked(lck) &&
++ htree_node_is_granted(lck, ffz(~lmask)))
++ htree_node_unlock(lck, ffz(~lmask), buf);
++}
++
++#define ext4_htree_dx_lock(lck, key) \
++ ext4_htree_node_lock(lck, key, EXT4_LB_DX, 1, NULL)
++#define ext4_htree_dx_lock_try(lck, key) \
++ ext4_htree_node_lock(lck, key, EXT4_LB_DX, 0, NULL)
++#define ext4_htree_dx_unlock(lck) \
++ ext4_htree_node_unlock(lck, EXT4_LB_DX, NULL)
++#define ext4_htree_dx_locked(lck) \
++ ext4_htree_node_locked(lck, EXT4_LB_DX)
++
++static void ext4_htree_dx_need_lock(struct htree_lock *lck)
++{
++ struct ext4_dir_lock_data *ld;
++
++ if (ext4_htree_safe_locked(lck))
++ return;
++
++ ld = ext4_htree_lock_data(lck);
++ switch (ld->ld_flags) {
++ default:
++ return;
++ case EXT4_HLOCK_LOOKUP:
++ ld->ld_flags = EXT4_HLOCK_LOOKUP_SAFE;
++ return;
++ case EXT4_HLOCK_DEL:
++ ld->ld_flags = EXT4_HLOCK_DEL_SAFE;
++ return;
++ case EXT4_HLOCK_ADD:
++ ld->ld_flags = EXT4_HLOCK_SPLIT;
++ return;
++ }
++}
++
++#define ext4_htree_de_lock(lck, key) \
++ ext4_htree_node_lock(lck, key, EXT4_LB_DE, 1, NULL)
++#define ext4_htree_de_unlock(lck) \
++ ext4_htree_node_unlock(lck, EXT4_LB_DE, NULL)
++
++#define ext4_htree_spin_lock(lck, key, event) \
++ ext4_htree_node_lock(lck, key, EXT4_LB_SPIN, 0, event)
++#define ext4_htree_spin_unlock(lck) \
++ ext4_htree_node_unlock(lck, EXT4_LB_SPIN, NULL)
++#define ext4_htree_spin_unlock_listen(lck, p) \
++ ext4_htree_node_unlock(lck, EXT4_LB_SPIN, p)
++
++static void ext4_htree_spin_stop_listen(struct htree_lock *lck)
++{
++ if (!ext4_htree_safe_locked(lck) &&
++ htree_node_is_listening(lck, ffz(~EXT4_LB_SPIN)))
++ htree_node_stop_listen(lck, ffz(~EXT4_LB_SPIN));
++}
++
++enum {
++ DX_HASH_COL_IGNORE, /* ignore collision while probing frames */
++ DX_HASH_COL_YES, /* there is collision and it does matter */
++ DX_HASH_COL_NO, /* there is no collision */
++};
++
++static int dx_probe_hash_collision(struct htree_lock *lck,
++ struct dx_entry *entries,
++ struct dx_entry *at, u32 hash)
++{
++ if (!(lck && ext4_htree_lock_data(lck)->ld_flags & EXT4_LB_EXACT)) {
++ return DX_HASH_COL_IGNORE; /* don't care about collision */
++
++ } else if (at == entries + dx_get_count(entries) - 1) {
++ return DX_HASH_COL_IGNORE; /* not in any leaf of this DX */
++
++ } else { /* hash collision? */
++ return ((dx_get_hash(at + 1) & ~1) == hash) ?
++ DX_HASH_COL_YES : DX_HASH_COL_NO;
++ }
++}
++
+ /*
+ * Probe for a directory leaf block to search.
+ *
+@@ -766,10 +997,11 @@ struct stats dx_show_entries(struct dx_h
+ */
+ static struct dx_frame *
+ dx_probe(struct ext4_filename *fname, struct inode *dir,
+- struct dx_hash_info *hinfo, struct dx_frame *frame_in)
++ struct dx_hash_info *hinfo, struct dx_frame *frame_in,
++ struct htree_lock *lck)
+ {
+ unsigned count, indirect;
+- struct dx_entry *at, *entries, *p, *q, *m;
++ struct dx_entry *at, *entries, *p, *q, *m, *dx = NULL;
+ struct dx_root_info *info;
+ struct dx_frame *frame = frame_in;
+ struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
+@@ -831,8 +1063,15 @@ dx_probe(struct ext4_filename *fname, st
+
+ dxtrace(printk("Look up %x", hash));
+ while (1) {
++ if (indirect == 0) { /* the last index level */
++ /* NB: ext4_htree_dx_lock() could be noop if
++ * DX-lock flag is not set for current operation */
++ ext4_htree_dx_lock(lck, dx);
++ ext4_htree_spin_lock(lck, dx, NULL);
++ }
+ count = dx_get_count(entries);
+- if (!count || count > dx_get_limit(entries)) {
++ if (count == 0 || count > dx_get_limit(entries)) {
++ ext4_htree_spin_unlock(lck); /* release spin */
+ ext4_warning_inode(dir,
+ "dx entry: count %u beyond limit %u",
+ count, dx_get_limit(entries));
+@@ -871,8 +1110,70 @@ dx_probe(struct ext4_filename *fname, st
+ dx_get_block(at)));
+ frame->entries = entries;
+ frame->at = at;
+- if (!indirect--)
++
++ if (indirect == 0) { /* the last index level */
++ struct ext4_dir_lock_data *ld;
++ u64 myblock;
++
++ /* By default we only lock DE-block, however, we will
++ * also lock the last level DX-block if:
++ * a) there is hash collision
++ * we will set DX-lock flag (a few lines below)
++ * and redo to lock DX-block
++ * see detail in dx_probe_hash_collision()
++ * b) it's a retry from splitting
++ * we need to lock the last level DX-block so nobody
++ * else can split any leaf blocks under the same
++ * DX-block, see detail in ext4_dx_add_entry()
++ */
++ if (ext4_htree_dx_locked(lck)) {
++ /* DX-block is locked, just lock DE-block
++ * and return */
++ ext4_htree_spin_unlock(lck);
++ if (!ext4_htree_safe_locked(lck))
++ ext4_htree_de_lock(lck, frame->at);
++ return frame;
++ }
++ /* it's pdirop and no DX lock */
++ if (dx_probe_hash_collision(lck, entries, at, hash) ==
++ DX_HASH_COL_YES) {
++ /* found hash collision, set DX-lock flag
++ * and retry to abtain DX-lock */
++ ext4_htree_spin_unlock(lck);
++ ext4_htree_dx_need_lock(lck);
++ continue;
++ }
++ ld = ext4_htree_lock_data(lck);
++ /* because I don't lock DX, so @at can't be trusted
++ * after I release spinlock so I have to save it */
++ ld->ld_at = at;
++ ld->ld_at_entry = *at;
++ ld->ld_count = dx_get_count(entries);
++
++ frame->at = &ld->ld_at_entry;
++ myblock = dx_get_block(at);
++
++ /* NB: ordering locking */
++ ext4_htree_spin_unlock_listen(lck, &myblock);
++ /* other thread can split this DE-block because:
++ * a) I don't have lock for the DE-block yet
++ * b) I released spinlock on DX-block
++ * if it happened I can detect it by listening
++ * splitting event on this DE-block */
++ ext4_htree_de_lock(lck, frame->at);
++ ext4_htree_spin_stop_listen(lck);
++
++ if (myblock == EXT4_HTREE_NODE_CHANGED) {
++ /* someone split this DE-block before
++ * I locked it, I need to retry and lock
++ * valid DE-block */
++ ext4_htree_de_unlock(lck);
++ continue;
++ }
+ return frame;
++ }
++ dx = at;
++ indirect--;
+ frame++;
+ frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
+ if (IS_ERR(frame->bh)) {
+@@ -941,7 +1242,7 @@ static void dx_release(struct dx_frame *
+ static int ext4_htree_next_block(struct inode *dir, __u32 hash,
+ struct dx_frame *frame,
+ struct dx_frame *frames,
+- __u32 *start_hash)
++ __u32 *start_hash, struct htree_lock *lck)
+ {
+ struct dx_frame *p;
+ struct buffer_head *bh;
+@@ -956,12 +1257,22 @@ static int ext4_htree_next_block(struct
+ * this loop, num_frames indicates the number of interior
+ * nodes need to be read.
+ */
++ ext4_htree_de_unlock(lck);
+ while (1) {
+- if (++(p->at) < p->entries + dx_get_count(p->entries))
+- break;
++ if (num_frames > 0 || ext4_htree_dx_locked(lck)) {
++ /* num_frames > 0 :
++ * DX block
++ * ext4_htree_dx_locked:
++ * frame->at is reliable pointer returned by dx_probe,
++ * otherwise dx_probe already knew no collision */
++ if (++(p->at) < p->entries + dx_get_count(p->entries))
++ break;
++ }
+ if (p == frames)
+ return 0;
+ num_frames++;
++ if (num_frames == 1)
++ ext4_htree_dx_unlock(lck);
+ p--;
+ }
+
+@@ -984,6 +1295,13 @@ static int ext4_htree_next_block(struct
+ * block so no check is necessary
+ */
+ while (num_frames--) {
++ if (num_frames == 0) {
++ /* it's not always necessary, we just don't want to
++ * detect hash collision again */
++ ext4_htree_dx_need_lock(lck);
++ ext4_htree_dx_lock(lck, p->at);
++ }
++
+ bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
+ if (IS_ERR(bh))
+ return PTR_ERR(bh);
+@@ -992,6 +1310,7 @@ static int ext4_htree_next_block(struct
+ p->bh = bh;
+ p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
+ }
++ ext4_htree_de_lock(lck, p->at);
+ return 1;
+ }
+
+@@ -1136,10 +1455,10 @@ int ext4_htree_fill_tree(struct file *di
+ }
+ hinfo.hash = start_hash;
+ hinfo.minor_hash = 0;
+- frame = dx_probe(NULL, dir, &hinfo, frames);
++ /* assume it's PR locked */
++ frame = dx_probe(NULL, dir, &hinfo, frames, NULL);
+ if (IS_ERR(frame))
+ return PTR_ERR(frame);
+-
+ /* Add '.' and '..' from the htree header */
+ if (!start_hash && !start_minor_hash) {
+ de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
+@@ -1179,7 +1498,7 @@ int ext4_htree_fill_tree(struct file *di
+ count += ret;
+ hashval = ~0;
+ ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
+- frame, frames, &hashval);
++ frame, frames, &hashval, NULL);
+ *next_hash = hashval;
+ if (ret < 0) {
+ err = ret;
+@@ -1455,7 +1774,7 @@ static int is_dx_internal_node(struct in
+ static struct buffer_head *__ext4_find_entry(struct inode *dir,
+ struct ext4_filename *fname,
+ struct ext4_dir_entry_2 **res_dir,
+- int *inlined)
++ int *inlined, struct htree_lock *lck)
+ {
+ struct super_block *sb;
+ struct buffer_head *bh_use[NAMEI_RA_SIZE];
+@@ -1497,7 +1816,7 @@ static struct buffer_head *__ext4_find_e
+ goto restart;
+ }
+ if (is_dx(dir)) {
+- ret = ext4_dx_find_entry(dir, fname, res_dir);
++ ret = ext4_dx_find_entry(dir, fname, res_dir, lck);
+ /*
+ * On success, or if the error was file not found,
+ * return. Otherwise, fall back to doing a search the
+@@ -1507,6 +1826,7 @@ static struct buffer_head *__ext4_find_e
+ goto cleanup_and_exit;
+ dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
+ "falling back\n"));
++ ext4_htree_safe_relock(lck);
+ ret = NULL;
+ }
+ nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
+@@ -1597,10 +1917,10 @@ cleanup_and_exit:
+ return ret;
+ }
+
+-static struct buffer_head *ext4_find_entry(struct inode *dir,
++struct buffer_head *ext4_find_entry_locked(struct inode *dir,
+ const struct qstr *d_name,
+ struct ext4_dir_entry_2 **res_dir,
+- int *inlined)
++ int *inlined, struct htree_lock *lck)
+ {
+ int err;
+ struct ext4_filename fname;
+@@ -1612,12 +1932,14 @@ static struct buffer_head *ext4_find_ent
+ if (err)
+ return ERR_PTR(err);
+
+- bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
++ bh = __ext4_find_entry(dir, &fname, res_dir, inlined, lck);
+
+ ext4_fname_free_filename(&fname);
+ return bh;
+ }
+
++EXPORT_SYMBOL(ext4_find_entry_locked);
++
+ static struct buffer_head *ext4_lookup_entry(struct inode *dir,
+ struct dentry *dentry,
+ struct ext4_dir_entry_2 **res_dir)
+@@ -1632,7 +1954,7 @@ static struct buffer_head *ext4_lookup_e
+ if (err)
+ return ERR_PTR(err);
+
+- bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
++ bh = __ext4_find_entry(dir, &fname, res_dir, NULL, NULL);
+
+ ext4_fname_free_filename(&fname);
+ return bh;
+@@ -1640,7 +1962,8 @@ static struct buffer_head *ext4_lookup_e
+
+ static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
+ struct ext4_filename *fname,
+- struct ext4_dir_entry_2 **res_dir)
++ struct ext4_dir_entry_2 **res_dir,
++ struct htree_lock *lck)
+ {
+ struct super_block * sb = dir->i_sb;
+ struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
+@@ -1651,7 +1974,7 @@ static struct buffer_head * ext4_dx_find
+ #ifdef CONFIG_FS_ENCRYPTION
+ *res_dir = NULL;
+ #endif
+- frame = dx_probe(fname, dir, NULL, frames);
++ frame = dx_probe(fname, dir, NULL, frames, lck);
+ if (IS_ERR(frame))
+ return (struct buffer_head *) frame;
+ do {
+@@ -1673,7 +1996,7 @@ static struct buffer_head * ext4_dx_find
+
+ /* Check to see if we should continue to search */
+ retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
+- frames, NULL);
++ frames, NULL, lck);
+ if (retval < 0) {
+ ext4_warning_inode(dir,
+ "error %d reading directory index block",
+@@ -1853,8 +2176,9 @@ static struct ext4_dir_entry_2* dx_pack_
+ * Returns pointer to de in block into which the new entry will be inserted.
+ */
+ static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
+- struct buffer_head **bh,struct dx_frame *frame,
+- struct dx_hash_info *hinfo)
++ struct buffer_head **bh, struct dx_frame *frames,
++ struct dx_frame *frame, struct dx_hash_info *hinfo,
++ struct htree_lock *lck)
+ {
+ unsigned blocksize = dir->i_sb->s_blocksize;
+ unsigned count, continued;
+@@ -1915,8 +2239,14 @@ static struct ext4_dir_entry_2 *do_split
+ hash2, split, count-split));
+
+ /* Fancy dance to stay within two buffers */
+- de2 = dx_move_dirents(data1, data2, map + split, count - split,
+- blocksize);
++ if (hinfo->hash < hash2) {
++ de2 = dx_move_dirents(data1, data2, map + split,
++ count - split, blocksize);
++ } else {
++ /* make sure we will add entry to the same block which
++ * we have already locked */
++ de2 = dx_move_dirents(data1, data2, map, split, blocksize);
++ }
+ de = dx_pack_dirents(data1, blocksize);
+ de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
+ (char *) de,
+@@ -1934,12 +2264,21 @@ static struct ext4_dir_entry_2 *do_split
+ dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
+ blocksize, 1));
+
+- /* Which block gets the new entry? */
+- if (hinfo->hash >= hash2) {
+- swap(*bh, bh2);
+- de = de2;
++ ext4_htree_spin_lock(lck, frame > frames ? (frame - 1)->at : NULL,
++ frame->at); /* notify block is being split */
++ if (hinfo->hash < hash2) {
++ dx_insert_block(frame, hash2 + continued, newblock);
++
++ } else {
++ /* switch block number */
++ dx_insert_block(frame, hash2 + continued,
++ dx_get_block(frame->at));
++ dx_set_block(frame->at, newblock);
++ (frame->at)++;
+ }
+- dx_insert_block(frame, hash2 + continued, newblock);
++ ext4_htree_spin_unlock(lck);
++ ext4_htree_dx_unlock(lck);
++
+ err = ext4_handle_dirty_dirblock(handle, dir, bh2);
+ if (err)
+ goto journal_error;
+@@ -2209,7 +2548,7 @@ static int make_indexed_dir(handle_t *ha
+ if (retval)
+ goto out_frames;
+
+- de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
++ de = do_split(handle, dir, &bh2, frames, frame, &fname->hinfo, NULL);
+ if (IS_ERR(de)) {
+ retval = PTR_ERR(de);
+ goto out_frames;
+@@ -2319,8 +2658,8 @@ out:
+ * may not sleep between calling this and putting something into
+ * the entry, as someone else might have used it while you slept.
+ */
+-static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
+- struct inode *inode)
++int ext4_add_entry_locked(handle_t *handle, struct dentry *dentry,
++ struct inode *inode, struct htree_lock *lck)
+ {
+ struct inode *dir = d_inode(dentry->d_parent);
+ struct buffer_head *bh = NULL;
+@@ -2370,9 +2709,10 @@ static int ext4_add_entry(handle_t *hand
+ if (dentry->d_name.len == 2 &&
+ memcmp(dentry->d_name.name, "..", 2) == 0)
+ return ext4_update_dotdot(handle, dentry, inode);
+- retval = ext4_dx_add_entry(handle, &fname, dir, inode);
++ retval = ext4_dx_add_entry(handle, &fname, dir, inode, lck);
+ if (!retval || (retval != ERR_BAD_DX_DIR))
+ goto out;
++ ext4_htree_safe_relock(lck);
+ /* Can we just ignore htree data? */
+ if (ext4_has_metadata_csum(sb)) {
+ EXT4_ERROR_INODE(dir,
+@@ -2435,12 +2775,14 @@ out:
+ ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
+ return retval;
+ }
++EXPORT_SYMBOL(ext4_add_entry_locked);
+
+ /*
+ * Returns 0 for success, or a negative error value
+ */
+ static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
+- struct inode *dir, struct inode *inode)
++ struct inode *dir, struct inode *inode,
++ struct htree_lock *lck)
+ {
+ struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
+ struct dx_entry *entries, *at;
+@@ -2452,7 +2794,7 @@ static int ext4_dx_add_entry(handle_t *h
+
+ again:
+ restart = 0;
+- frame = dx_probe(fname, dir, NULL, frames);
++ frame = dx_probe(fname, dir, NULL, frames, lck);
+ if (IS_ERR(frame))
+ return PTR_ERR(frame);
+ entries = frame->entries;
+@@ -2487,6 +2829,12 @@ again:
+ struct dx_node *node2;
+ struct buffer_head *bh2;
+
++ if (!ext4_htree_safe_locked(lck)) { /* retry with EX lock */
++ ext4_htree_safe_relock(lck);
++ restart = 1;
++ goto cleanup;
++ }
++
+ while (frame > frames) {
+ if (dx_get_count((frame - 1)->entries) <
+ dx_get_limit((frame - 1)->entries)) {
+@@ -2589,8 +2937,32 @@ again:
+ restart = 1;
+ goto journal_error;
+ }
++ } else if (!ext4_htree_dx_locked(lck)) {
++ struct ext4_dir_lock_data *ld = ext4_htree_lock_data(lck);
++
++ /* not well protected, require DX lock */
++ ext4_htree_dx_need_lock(lck);
++ at = frame > frames ? (frame - 1)->at : NULL;
++
++ /* NB: no risk of deadlock because it's just a try.
++ *
++ * NB: we check ld_count for twice, the first time before
++ * having DX lock, the second time after holding DX lock.
++ *
++ * NB: We never free blocks for directory so far, which
++ * means value returned by dx_get_count() should equal to
++ * ld->ld_count if nobody split any DE-block under @at,
++ * and ld->ld_at still points to valid dx_entry. */
++ if ((ld->ld_count != dx_get_count(entries)) ||
++ !ext4_htree_dx_lock_try(lck, at) ||
++ (ld->ld_count != dx_get_count(entries))) {
++ restart = 1;
++ goto cleanup;
++ }
++ /* OK, I've got DX lock and nothing changed */
++ frame->at = ld->ld_at;
+ }
+- de = do_split(handle, dir, &bh, frame, &fname->hinfo);
++ de = do_split(handle, dir, &bh, frames, frame, &fname->hinfo, lck);
+ if (IS_ERR(de)) {
+ err = PTR_ERR(de);
+ goto cleanup;
+@@ -2601,6 +2973,8 @@ again:
+ journal_error:
+ ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
+ cleanup:
++ ext4_htree_dx_unlock(lck);
++ ext4_htree_de_unlock(lck);
+ brelse(bh);
+ dx_release(frames);
+ /* @restart is true means htree-path has been changed, we need to
+--- a/fs/ext4/super.c
++++ b/fs/ext4/super.c
+@@ -1122,6 +1122,7 @@ static struct inode *ext4_alloc_inode(st
+
+ inode_set_iversion(&ei->vfs_inode, 1);
+ spin_lock_init(&ei->i_raw_lock);
++ sema_init(&ei->i_append_sem, 1);
+ INIT_LIST_HEAD(&ei->i_prealloc_list);
+ spin_lock_init(&ei->i_prealloc_lock);
+ ext4_es_init_tree(&ei->i_es_tree);
--- /dev/null
+---
+ fs/ext4/ext4.h | 7 +
+ fs/ext4/inode.c | 3
+ fs/ext4/mballoc.c | 215 +++++++++++++++++++++++++++++++++++++++++-------------
+ fs/ext4/sysfs.c | 8 +-
+ 4 files changed, 180 insertions(+), 53 deletions(-)
+
+--- a/fs/ext4/ext4.h
++++ b/fs/ext4/ext4.h
+@@ -1241,6 +1241,8 @@ extern void ext4_set_bits(void *bm, int
+ /* Metadata checksum algorithm codes */
+ #define EXT4_CRC32C_CHKSUM 1
+
++#define EXT4_MAX_PREALLOC_TABLE 64
++
+ /*
+ * Structure of the super block
+ */
+@@ -1497,11 +1499,13 @@ 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 long *s_mb_prealloc_table;
+ unsigned int s_mb_group_prealloc;
+ unsigned int s_max_dir_size_kb;
+ /* where last allocation was done - for stream allocation */
+@@ -2645,6 +2649,7 @@ extern int ext4_init_inode_table(struct
+ extern void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate);
+
+ /* mballoc.c */
++extern const struct proc_ops 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;
+--- a/fs/ext4/inode.c
++++ b/fs/ext4/inode.c
+@@ -2698,6 +2698,9 @@ static int ext4_writepages(struct addres
+ 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;
+
+--- a/fs/ext4/mballoc.c
++++ b/fs/ext4/mballoc.c
+@@ -2455,6 +2455,99 @@ const struct seq_operations ext4_mb_seq_
+ .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++;
++ 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;
+@@ -2685,7 +2778,7 @@ static int ext4_groupinfo_create_slab(si
+ 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;
+@@ -2734,7 +2827,6 @@ 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
+@@ -2758,9 +2850,29 @@ 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);
+@@ -2788,6 +2900,7 @@ out_free_locality_groups:
+ 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);
+@@ -3057,7 +3170,6 @@ ext4_mb_mark_diskspace_used(struct ext4_
+ 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);
+@@ -3187,13 +3299,14 @@ ext4_mb_normalize_request(struct ext4_al
+ 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 */
+@@ -3222,51 +3335,46 @@ ext4_mb_normalize_request(struct ext4_al
+ 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;
+
+- /* max size of free chunks */
+- max = 2 << bsbits;
++ /* 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;
+
+-#define NRL_CHECK_SIZE(req, size, max, chunk_size) \
+- (req <= (size) || max <= (chunk_size))
++ if (size <= value) {
++ wind = value;
++ break;
++ }
++ }
+
+- /* 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 (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) {
+@@ -3348,7 +3456,6 @@ ext4_mb_normalize_request(struct ext4_al
+ (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 */
+
+@@ -4341,11 +4448,19 @@ static void ext4_mb_group_or_file(struct
+
+ /* 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
+--- a/fs/ext4/sysfs.c
++++ b/fs/ext4/sysfs.c
+@@ -213,7 +213,8 @@ 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(extent_max_zeroout_kb, s_extent_max_zeroout_kb);
+ EXT4_ATTR(trigger_fs_error, 0200, trigger_test_error);
+@@ -255,7 +256,8 @@ 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(max_writeback_mb_bump),
+ ATTR_LIST(extent_max_zeroout_kb),
+@@ -510,6 +512,8 @@ int ext4_register_sysfs(struct super_blo
+ 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;
+ }
--- /dev/null
+---
+ fs/ext4/ext4.h | 7 ++
+ fs/ext4/mballoc.c | 133 ++++++++++++++++++++++++++++++++++++++++++++++++++++++
+ fs/ext4/mballoc.h | 3 +
+ fs/ext4/sysfs.c | 52 +++++++++++++++++++++
+ 4 files changed, 195 insertions(+)
+
+--- a/fs/ext4/ext4.h
++++ b/fs/ext4/ext4.h
+@@ -1518,6 +1518,9 @@ struct ext4_sb_info {
+ 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_max_dir_size_kb;
+@@ -1534,6 +1537,9 @@ struct ext4_sb_info {
+ 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;
+@@ -2813,6 +2819,7 @@ ext4_read_inode_bitmap(struct super_bloc
+ /* 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;
+--- a/fs/ext4/mballoc.c
++++ b/fs/ext4/mballoc.c
+@@ -2218,6 +2218,20 @@ out:
+ return ret;
+ }
+
++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)
+ {
+@@ -2227,6 +2241,7 @@ ext4_mb_regular_allocator(struct ext4_al
+ struct ext4_sb_info *sbi;
+ struct super_block *sb;
+ struct ext4_buddy e4b;
++ ext4_fsblk_t avail_blocks;
+
+ sb = ac->ac_sb;
+ sbi = EXT4_SB(sb);
+@@ -2279,6 +2294,21 @@ ext4_mb_regular_allocator(struct ext4_al
+
+ /* 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
+@@ -2342,6 +2372,9 @@ repeat:
+ 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 &&
+@@ -2624,6 +2657,92 @@ const struct proc_ops ext4_seq_mb_last_g
+ .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);
+@@ -2850,6 +2969,7 @@ static int ext4_groupinfo_create_slab(si
+ 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);
+@@ -2903,6 +3023,9 @@ int ext4_mb_init(struct super_block *sb)
+ sbi->s_mb_min_to_scan = MB_DEFAULT_MIN_TO_SCAN;
+ sbi->s_mb_stats = MB_DEFAULT_STATS;
+ sbi->s_mb_order2_reqs = MB_DEFAULT_ORDER2_REQS;
++ 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
+@@ -3042,6 +3165,16 @@ int ext4_mb_release(struct super_block *
+ 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),
+--- a/fs/ext4/mballoc.h
++++ b/fs/ext4/mballoc.h
+@@ -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
+--- a/fs/ext4/sysfs.c
++++ b/fs/ext4/sysfs.c
+@@ -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,
+@@ -140,6 +143,32 @@ static ssize_t journal_task_show(struct
+ 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 }, \
+@@ -205,6 +234,9 @@ EXT4_ATTR_FUNC(delayed_allocation_blocks
+ EXT4_ATTR_FUNC(session_write_kbytes, 0444);
+ EXT4_ATTR_FUNC(lifetime_write_kbytes, 0444);
+ EXT4_ATTR_FUNC(reserved_clusters, 0644);
++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);
+@@ -253,6 +285,9 @@ static struct attribute *ext4_attrs[] =
+ ATTR_LIST(session_write_kbytes),
+ ATTR_LIST(lifetime_write_kbytes),
+ ATTR_LIST(reserved_clusters),
++ 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),
+@@ -365,6 +400,15 @@ static ssize_t ext4_attr_show(struct kob
+ 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:
+@@ -466,6 +510,12 @@ static ssize_t ext4_attr_store(struct ko
+ 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;
+ }
+@@ -528,6 +578,8 @@ int ext4_register_sysfs(struct super_blo
+ &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;
+ }
--- /dev/null
+rhel8/ext4-inode-version.patch
+linux-5.4/ext4-lookup-dotdot.patch
+suse15/ext4-print-inum-in-htree-warning.patch
+linux-5.8/ext4-prealloc.patch
+ubuntu18/ext4-osd-iop-common.patch
+linux-5.8/ext4-misc.patch
+linux-5.8/ext4-mballoc-extra-checks.patch
+linux-5.4/ext4-hash-indexed-dir-dotdot-update.patch
+linux-5.8/ext4-kill-dx-root.patch
+linux-5.8/ext4-mballoc-pa-free-mismatch.patch
+linux-5.8/ext4-data-in-dirent.patch
+rhel8/ext4-nocmtime.patch
+base/ext4-htree-lock.patch
+linux-5.8/ext4-pdirop.patch
+linux-5.8/ext4-max-dir-size.patch
+linux-5.8/ext4-corrupted-inode-block-bitmaps-handling-patches.patch
+linux-5.4/ext4-give-warning-with-dir-htree-growing.patch
+ubuntu18/ext4-jcb-optimization.patch
+linux-5.4/ext4-attach-jinode-in-writepages.patch
+rhel8/ext4-dont-check-before-replay.patch
+rhel7.6/ext4-use-GFP_NOFS-in-ext4_inode_attach_jinode.patch
+rhel7.6/ext4-export-orphan-add.patch
+linux-5.8/ext4-export-mb-stream-allocator-variables.patch
+ubuntu19/ext4-iget-with-flags.patch
+linux-5.4/export-ext4fs-dirhash-helper.patch
+linux-5.4/ext4-misc.patch
+linux-5.8/ext4-simple-blockalloc.patch
+linux-5.8/ext4-no-max-dir-size-limit-for-iam-objects.patch