From: Yang Sheng Date: Mon, 9 Mar 2015 15:05:52 +0000 (+0800) Subject: LU-5022 ldiskfs: enable support for RHEL7 X-Git-Tag: 2.7.51~48 X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=commitdiff_plain;h=eaf30e35002bf3936cdf7bf83685552d854f0f88 LU-5022 ldiskfs: enable support for RHEL7 This patch adds support for RHEL7.1 [3.10.0-229.el7] kernel. Signed-off-by: Bob Glossman Signed-off-by: Yang Sheng Change-Id: Ifbc294a53bd21eb35d373637d3326fc3c611c9f0 Reviewed-on: http://review.whamcloud.com/10249 Tested-by: Jenkins Tested-by: Maloo Reviewed-by: Andreas Dilger Reviewed-by: Oleg Drokin --- diff --git a/config/lustre-build-ldiskfs.m4 b/config/lustre-build-ldiskfs.m4 index 084e26b..a19ac4e 100644 --- a/config/lustre-build-ldiskfs.m4 +++ b/config/lustre-build-ldiskfs.m4 @@ -6,7 +6,7 @@ LDISKFS_SERIES= AC_MSG_CHECKING([which ldiskfs series to use]) AS_IF([test x$RHEL_KERNEL = xyes], [ case $RHEL_RELEASE_NO in - 70) LDISKFS_SERIES="3.10-rhel7.series" ;; + 71) LDISKFS_SERIES="3.10-rhel7.series" ;; 66) LDISKFS_SERIES="2.6-rhel6.6.series" ;; 65) LDISKFS_SERIES="2.6-rhel6.5.series" ;; 64) LDISKFS_SERIES="2.6-rhel6.4.series" ;; diff --git a/ldiskfs/kernel_patches/patches/rhel7/ext4-data-in-dirent.patch b/ldiskfs/kernel_patches/patches/rhel7/ext4-data-in-dirent.patch new file mode 100644 index 0000000..2e73eb3 --- /dev/null +++ b/ldiskfs/kernel_patches/patches/rhel7/ext4-data-in-dirent.patch @@ -0,0 +1,730 @@ +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. + +Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/dir.c +=================================================================== +--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/dir.c ++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/dir.c +@@ -71,11 +71,11 @@ int __ext4_check_dir_entry(const char *f + const int rlen = ext4_rec_len_from_disk(de->rec_len, + dir->i_sb->s_blocksize); + +- if (unlikely(rlen < EXT4_DIR_REC_LEN(1))) ++ if (unlikely(rlen < __EXT4_DIR_REC_LEN(1))) + 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_REC_LEN(de))) + error_msg = "rec_len is too small for name_len"; + else if (unlikely(((char *) de - buf) + rlen > size)) + error_msg = "directory entry across range"; +@@ -208,7 +208,7 @@ revalidate: + * 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); +@@ -438,12 +438,17 @@ 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) + dirent->name_len + 1; ++ if (dirent->file_type & EXT4_DIRENT_LUFID) ++ extra_data = ext4_get_dirent_data_len(dirent); ++ ++ len = sizeof(struct fname) + dirent->name_len + extra_data + 1; ++ + new_fn = kzalloc(len, GFP_KERNEL); + if (!new_fn) + return -ENOMEM; +@@ -452,7 +457,7 @@ int ext4_htree_store_dirent(struct file + new_fn->inode = le32_to_cpu(dirent->inode); + new_fn->name_len = dirent->name_len; + new_fn->file_type = dirent->file_type; +- memcpy(new_fn->name, dirent->name, dirent->name_len); ++ memcpy(new_fn->name, dirent->name, dirent->name_len + extra_data); + new_fn->name[dirent->name_len] = 0; + + while (*p) { +@@ -452,7 +457,7 @@ int ext4_htree_store_dirent(struct file + if (ext4_check_dir_entry(dir, NULL, de, bh, + buf, buf_size, offset)) + return -EIO; +- nlen = EXT4_DIR_REC_LEN(de->name_len); ++ nlen = EXT4_DIR_REC_LEN(de); + rlen = ext4_rec_len_from_disk(de->rec_len, buf_size); + de = (struct ext4_dir_entry_2 *)((char *)de + rlen); + offset += rlen; +Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/ext4.h +=================================================================== +--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/ext4.h ++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/ext4.h +@@ -952,6 +952,7 @@ struct ext4_inode_info { + #define EXT4_MOUNT_ERRORS_MASK 0x00070 + #define EXT4_MOUNT_MINIX_DF 0x00080 /* Mimics the Minix statfs */ + #define EXT4_MOUNT_NOLOAD 0x00100 /* Don't use existing journal*/ ++#define EXT4_MOUNT_DIRDATA 0x00200 /* Data in directory entries*/ + #define EXT4_MOUNT_DATA_FLAGS 0x00C00 /* Mode for data writes: */ + #define EXT4_MOUNT_JOURNAL_DATA 0x00400 /* Write data to journal */ + #define EXT4_MOUNT_ORDERED_DATA 0x00800 /* Flush data before commit */ +@@ -1534,6 +1535,7 @@ static inline void ext4_clear_state_flag + EXT4_FEATURE_INCOMPAT_64BIT| \ + EXT4_FEATURE_INCOMPAT_FLEX_BG| \ + EXT4_FEATURE_INCOMPAT_MMP | \ ++ EXT4_FEATURE_INCOMPAT_DIRDATA| \ + EXT4_FEATURE_INCOMPAT_INLINE_DATA) + #define EXT4_FEATURE_RO_COMPAT_SUPP (EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER| \ + EXT4_FEATURE_RO_COMPAT_LARGE_FILE| \ +@@ -1640,6 +1642,43 @@ 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 */ ++} __attribute__((packed)); ++ ++static inline unsigned char *ext4_dentry_get_data(struct super_block *sb, ++ struct ext4_dentry_param* p) ++ ++{ ++ if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_DIRDATA)) ++ return NULL; ++ if (p && p->edp_magic == EXT4_LUFID_MAGIC) ++ return &p->edp_len; ++ else ++ return NULL; ++} + + #define EXT4_FT_DIR_CSUM 0xDE + +@@ -1650,8 +1689,11 @@ 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_REC_LEN(de) (__EXT4_DIR_REC_LEN((de)->name_len +\ ++ ext4_get_dirent_data_len(de))) ++ + #define EXT4_MAX_REC_LEN ((1<<16)-1) + + /* +@@ -1987,11 +2029,11 @@ extern int ext4_find_dest_de(struct inod + struct buffer_head *bh, + void *buf, int buf_size, + const char *name, int namelen, +- 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, +- const char *name, int namelen); ++ const char *name, int namelen, void *data); + static inline void ext4_update_dx_flag(struct inode *inode) + { + if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb, +@@ -2004,11 +2046,18 @@ static unsigned char ext4_filetype_table + + static inline unsigned char get_dtype(struct super_block *sb, int filetype) + { ++ int fl_index = filetype & EXT4_FT_MASK; ++ + if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE) || +- (filetype >= EXT4_FT_MAX)) ++ (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); ++ + } + + /* fsync.c */ +@@ -2157,7 +2206,7 @@ extern struct buffer_head * ext4_find_en + struct ext4_dir_entry_2 ** res_dir, + int *inlined); + extern int ext4_add_dot_dotdot(handle_t *handle, struct inode *dir, +- struct inode *inode); ++ struct inode *inode, const void *, const void *); + extern int search_dir(struct buffer_head *bh, + char *search_buf, + int buf_size, +@@ -2761,6 +2810,28 @@ extern struct mutex ext4__aio_mutex[EXT4 + extern int ext4_resize_begin(struct super_block *sb); + extern void ext4_resize_end(struct super_block *sb); + ++/* ++ * 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; ++ ++ while (extra_data_flags) { ++ if (extra_data_flags & 1) { ++ dlen += *len + (dlen == 0); ++ len += *len; ++ } ++ extra_data_flags >>= 1; ++ } ++ return dlen; ++} ++ + #endif /* __KERNEL__ */ + + #endif /* _EXT4_H */ +Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/namei.c +=================================================================== +--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/namei.c ++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/namei.c +@@ -239,7 +239,8 @@ 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(const struct qstr *d_name, + struct inode *dir, +@@ -504,11 +505,12 @@ ext4_next_entry(struct ext4_dir_entry_2 + */ + 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)); ++ BUG_ON(de->name_len != 1); ++ /* get dotdot first */ ++ de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(de)); + +- /* dx root info is after dotdot entry */ +- de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(2)); ++ /* dx root info is after dotdot entry */ ++ de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(de)); + + return (struct dx_root_info *) de; + } +@@ -553,10 +555,16 @@ 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_REC_LEN(dot_de) - ++ EXT4_DIR_REC_LEN(dotdot_de) - infosize; + + if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb, + EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) +@@ -566,7 +574,7 @@ static inline unsigned dx_root_limit(str + + static inline unsigned dx_node_limit(struct inode *dir) + { +- unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0); ++ unsigned entry_space = dir->i_sb->s_blocksize - __EXT4_DIR_REC_LEN(0); + + if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb, + EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) +@@ -617,7 +625,7 @@ static struct stats dx_show_leaf(struct + printk(":%x.%u ", h.hash, + (unsigned) ((char *) de - base)); + } +- space += EXT4_DIR_REC_LEN(de->name_len); ++ space += EXT4_DIR_REC_LEN(de); + names++; + } + de = ext4_next_entry(de, size); +@@ -723,6 +731,7 @@ dx_probe(const struct qstr *d_name, stru + entries = (struct dx_entry *) (((char *)info) + info->info_length); + + if (dx_get_limit(entries) != dx_root_limit(dir, ++ (struct ext4_dir_entry_2*)bh->b_data, + info->info_length)) { + ext4_warning(dir->i_sb, "dx entry: limit != root limit"); + brelse(bh); +@@ -916,7 +925,7 @@ static int htree_dirblock_to_tree(struct + de = (struct ext4_dir_entry_2 *) bh->b_data; + top = (struct ext4_dir_entry_2 *) ((char *) de + + dir->i_sb->s_blocksize - +- EXT4_DIR_REC_LEN(0)); ++ __EXT4_DIR_REC_LEN(0)); + for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) { + if (ext4_check_dir_entry(dir, NULL, de, bh, + bh->b_data, bh->b_size, +@@ -1508,7 +1517,7 @@ 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_REC_LEN(de); + memcpy (to, de, rec_len); + ((struct ext4_dir_entry_2 *) to)->rec_len = + ext4_rec_len_to_disk(rec_len, blocksize); +@@ -1532,7 +1541,7 @@ 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_REC_LEN(de); + if (de > to) + memmove(to, de, rec_len); + to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize); +@@ -1664,14 +1673,15 @@ int ext4_find_dest_de(struct inode *dir, + struct buffer_head *bh, + void *buf, int buf_size, + const char *name, int namelen, +- 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(namelen); ++ unsigned short reclen = __EXT4_DIR_REC_LEN(namelen) + (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) { +@@ -1680,10 +1690,26 @@ int ext4_find_dest_de(struct inode *dir, + return -EIO; + if (ext4_match(namelen, name, de)) + return -EEXIST; +- nlen = EXT4_DIR_REC_LEN(de->name_len); ++ nlen = EXT4_DIR_REC_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 (namelen == 2 && memcmp(name, "..", 2) == 0) { ++ if ((de->inode ? rlen - nlen : rlen) >= ++ __EXT4_DIR_REC_LEN(namelen)) { ++ /* 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; + } +@@ -1697,12 +1723,12 @@ int ext4_find_dest_de(struct inode *dir, + void ext4_insert_dentry(struct inode *inode, + struct ext4_dir_entry_2 *de, + int buf_size, +- const char *name, int namelen) ++ const char *name, int namelen, void *data) + { + + int nlen, rlen; + +- nlen = EXT4_DIR_REC_LEN(de->name_len); ++ nlen = EXT4_DIR_REC_LEN(de); + rlen = ext4_rec_len_from_disk(de->rec_len, buf_size); + if (de->inode) { + struct ext4_dir_entry_2 *de1 = +@@ -1716,6 +1742,11 @@ void ext4_insert_dentry(struct inode *in + ext4_set_de_type(inode->i_sb, de, inode->i_mode); + de->name_len = namelen; + memcpy(de->name, name, namelen); ++ if (data) { ++ de->name[namelen] = 0; ++ memcpy(&de->name[namelen + 1], data, *(char *)data); ++ de->file_type |= EXT4_DIRENT_LUFID; ++ } + } + /* + * Add a new entry into a directory (leaf) block. If de is non-NULL, +@@ -1734,15 +1765,20 @@ static int add_dirent_to_buf(handle_t *h + int namelen = dentry->d_name.len; + unsigned int blocksize = dir->i_sb->s_blocksize; + int csum_size = 0; +- int err; ++ int err, dlen = 0; ++ unsigned char *data; + ++ data = ext4_dentry_get_data(inode->i_sb, (struct ext4_dentry_param *) ++ dentry->d_fsdata); + 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, +- name, namelen, &de); ++ name, namelen, &de, &dlen); + if (err) + return err; + } +@@ -1755,7 +1791,10 @@ static int add_dirent_to_buf(handle_t *h + } + + /* By now the buffer is marked for journaling */ +- ext4_insert_dentry(inode, de, blocksize, name, namelen); ++ /* If we're writing the short form of "dotdot", don't add the data section */ ++ if (dlen == 1) ++ data = NULL; ++ ext4_insert_dentry(inode, de, blocksize, name, namelen, data); + + /* + * XXX shouldn't update any times until successful +@@ -1866,7 +1905,8 @@ 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 */ + hinfo.hash_version = dx_info->hash_version; +@@ -1909,6 +1949,8 @@ 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); +@@ -1924,19 +1966,24 @@ static int ext4_update_dotdot(handle_t * + /* the first item must be "." */ + assert(de->name_len == 1 && de->name[0] == '.'); + len = le16_to_cpu(de->rec_len); +- assert(len >= EXT4_DIR_REC_LEN(1)); +- if (len > EXT4_DIR_REC_LEN(1)) { ++ assert(len >= __EXT4_DIR_REC_LEN(1)); ++ if (len > __EXT4_DIR_REC_LEN(1)) { + BUFFER_TRACE(dir_block, "get_write_access"); + err = ext4_journal_get_write_access(handle, dir_block); + if (err) + goto out_journal; + + journal = 1; +- de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(1)); ++ de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(de)); + } + +- len -= EXT4_DIR_REC_LEN(1); +- assert(len == 0 || len >= EXT4_DIR_REC_LEN(2)); ++ len -= EXT4_DIR_REC_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) { +@@ -1950,10 +1997,15 @@ static int ext4_update_dotdot(handle_t * + if (len > 0) + de->rec_len = cpu_to_le16(len); + else +- assert(le16_to_cpu(de->rec_len) >= EXT4_DIR_REC_LEN(2)); ++ 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) { +@@ -2428,30 +2480,61 @@ retry: + 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_REC_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); ++ EXT4_DIR_REC_LEN(de), blocksize); +- strcpy(de->name, ".."); +- ext4_set_de_type(inode->i_sb, de, S_IFDIR); + + return ext4_next_entry(de, blocksize); + } +@@ -2457,8 +2540,10 @@ struct ext4_dir_entry_2 *ext4_init_dot_d + } + + 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; + struct ext4_dir_entry_tail *t; +@@ -2488,7 +2573,11 @@ static int ext4_init_new_dir(handle_t *h + if (err) + goto out; + 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) { + t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize); +@@ -2508,7 +2597,8 @@ out: + /* 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) ++ struct inode *inode, ++ const void *data1, const void *data2) + { + if (IS_ERR(handle)) + return PTR_ERR(handle); +@@ -2518,7 +2608,7 @@ int ext4_add_dot_dotdot(handle_t *handle + + inode->i_op = &ext4_dir_inode_operations; + inode->i_fop = &ext4_dir_operations; +- return ext4_init_new_dir(handle, dir, inode); ++ return ext4_init_new_dir(handle, dir, inode, data1, data2); + } + EXPORT_SYMBOL(ext4_add_dot_dotdot); + +@@ -2546,7 +2636,7 @@ 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); +@@ -2598,7 +2688,7 @@ static int empty_dir(struct inode *inode + } + + sb = inode->i_sb; +- if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) { ++ if (inode->i_size < __EXT4_DIR_REC_LEN(1) + __EXT4_DIR_REC_LEN(2)) { + EXT4_ERROR_INODE(inode, "invalid size"); + return 1; + } +Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/inline.c +=================================================================== +--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/inline.c ++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/inline.c +@@ -988,7 +998,7 @@ static int ext4_add_dirent_to_inline(han + + err = ext4_find_dest_de(dir, inode, iloc->bh, + inline_start, inline_size, +- name, namelen, &de); ++ name, namelen, &de, NULL); + if (err) + return err; + +@@ -998,7 +998,7 @@ 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, name, namelen); ++ ext4_insert_dentry(inode, de, inline_size, name, namelen, NULL); + + ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size); + +@@ -1078,7 +1078,7 @@ static int ext4_update_inline_dir(handle + int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE; + int new_size = get_max_inline_xattr_value_size(dir, iloc); + +- if (new_size - old_size <= EXT4_DIR_REC_LEN(1)) ++ if (new_size - old_size <= __EXT4_DIR_REC_LEN(1)) + return -ENOSPC; + + ret = ext4_update_inline_data(handle, dir, +@@ -1348,7 +1348,7 @@ int htree_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_REC_LEN(&fake), + inline_size); + ext4_set_de_type(inode->i_sb, &fake, S_IFDIR); + de = &fake; +@@ -1358,7 +1358,7 @@ int htree_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_REC_LEN(&fake), + inline_size); + ext4_set_de_type(inode->i_sb, &fake, S_IFDIR); + de = &fake; +@@ -1455,8 +1455,8 @@ int ext4_read_inline_dir(struct file *fi + * So we will use extra_offset and extra_size to indicate them + * during the inline dir iteration. + */ +- dotdot_offset = EXT4_DIR_REC_LEN(1); +- dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2); ++ dotdot_offset = __EXT4_DIR_REC_LEN(1); ++ dotdot_size = dotdot_offset + __EXT4_DIR_REC_LEN(2); + extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE; + extra_size = extra_offset + inline_size; + +@@ -1493,7 +1493,7 @@ revalidate: + * failure will be detected in the + * dirent test below. */ + if (ext4_rec_len_from_disk(de->rec_len, +- extra_size) < EXT4_DIR_REC_LEN(1)) ++ extra_size) < __EXT4_DIR_REC_LEN(1)) + break; + i += ext4_rec_len_from_disk(de->rec_len, + extra_size); +Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/super.c +=================================================================== +--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/super.c ++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/super.c +@@ -1151,7 +1151,7 @@ enum { + Opt_data_err_abort, Opt_data_err_ignore, + 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_i_version, + Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit, + Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity, +@@ -1216,6 +1216,7 @@ static const match_table_t tokens = { + {Opt_stripe, "stripe=%u"}, + {Opt_delalloc, "delalloc"}, + {Opt_nodelalloc, "nodelalloc"}, ++ {Opt_dirdata, "dirdata"}, + {Opt_removed, "mblk_io_submit"}, + {Opt_removed, "nomblk_io_submit"}, + {Opt_block_validity, "block_validity"}, +@@ -1424,6 +1425,7 @@ 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}, diff --git a/ldiskfs/kernel_patches/patches/rhel7/ext4-disable-mb-cache.patch b/ldiskfs/kernel_patches/patches/rhel7/ext4-disable-mb-cache.patch new file mode 100644 index 0000000..38cfab6 --- /dev/null +++ b/ldiskfs/kernel_patches/patches/rhel7/ext4-disable-mb-cache.patch @@ -0,0 +1,157 @@ +mbcache provides absolutely no value for Lustre xattrs (because +they are unique and cannot be shared between files) and as we can +see it has a noticable overhead in some cases. In the past there +was a CONFIG_MBCACHE option that would allow it to be disabled, +but this was removed in newer kernels, so we will need to patch +ldiskfs to fix this. + +Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/ext4.h +=================================================================== +--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/ext4.h ++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/ext4.h +@@ -944,6 +944,7 @@ struct ext4_inode_info { + /* + * Mount flags set via mount options or defaults + */ ++#define EXT4_MOUNT_NO_MBCACHE 0x00001 /* Disable mbcache */ + #define EXT4_MOUNT_GRPID 0x00004 /* Create files with directory's group */ + #define EXT4_MOUNT_DEBUG 0x00008 /* Some debugging messages */ + #define EXT4_MOUNT_ERRORS_CONT 0x00010 /* Continue on errors */ +Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/super.c +=================================================================== +--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/super.c ++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/super.c +@@ -1157,6 +1157,7 @@ enum { + Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity, + Opt_inode_readahead_blks, Opt_journal_ioprio, + Opt_dioread_nolock, Opt_dioread_lock, ++ Opt_no_mbcache, + Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable, + Opt_max_dir_size_kb, + }; +@@ -1231,6 +1232,7 @@ static const match_table_t tokens = { + {Opt_discard, "discard"}, + {Opt_nodiscard, "nodiscard"}, + {Opt_init_itable, "init_itable=%u"}, ++ {Opt_no_mbcache, "no_mbcache"}, + {Opt_init_itable, "init_itable"}, + {Opt_noinit_itable, "noinit_itable"}, + {Opt_max_dir_size_kb, "max_dir_size_kb=%u"}, +@@ -1390,6 +1392,7 @@ static const struct mount_opts { + {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET}, + {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR}, + {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR}, ++ {Opt_no_mbcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET}, + {Opt_commit, 0, MOPT_GTE0}, + {Opt_max_batch_time, 0, MOPT_GTE0}, + {Opt_min_batch_time, 0, MOPT_GTE0}, +Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/xattr.c +=================================================================== +--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/xattr.c ++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/xattr.c +@@ -81,7 +81,8 @@ + # define ea_bdebug(bh, fmt, ...) no_printk(fmt, ##__VA_ARGS__) + #endif + +-static void ext4_xattr_cache_insert(struct buffer_head *); ++static void ext4_xattr_cache_insert(struct super_block *, ++ struct buffer_head *); + static struct buffer_head *ext4_xattr_cache_find(struct inode *, + struct ext4_xattr_header *, + struct mb_cache_entry **); +@@ -385,7 +386,7 @@ bad_block: + error = -EIO; + goto cleanup; + } +- ext4_xattr_cache_insert(bh); ++ ext4_xattr_cache_insert(inode->i_sb, bh); + entry = BFIRST(bh); + error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1, + inode); +@@ -546,7 +547,7 @@ ext4_xattr_block_list(struct dentry *den + error = -EIO; + goto cleanup; + } +- ext4_xattr_cache_insert(bh); ++ ext4_xattr_cache_insert(inode->i_sb, bh); + error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer, buffer_size); + + cleanup: +@@ -643,7 +644,9 @@ ext4_xattr_release_block(handle_t *handl + struct mb_cache_entry *ce = NULL; + int error = 0; + +- ce = mb_cache_entry_get(ext4_xattr_cache, bh->b_bdev, bh->b_blocknr); ++ if (!test_opt(inode->i_sb, NO_MBCACHE)) ++ ce = mb_cache_entry_get(ext4_xattr_cache, bh->b_bdev, ++ bh->b_blocknr); + BUFFER_TRACE(bh, "get_write_access"); + error = ext4_journal_get_write_access(handle, bh); + if (error) +@@ -1037,8 +1040,10 @@ ext4_xattr_block_set(handle_t *handle, s + #define header(x) ((struct ext4_xattr_header *)(x)) + + if (s->base) { +- ce = mb_cache_entry_get(ext4_xattr_cache, bs->bh->b_bdev, +- bs->bh->b_blocknr); ++ if (!test_opt(inode->i_sb, NO_MBCACHE)) ++ ce = mb_cache_entry_get(ext4_xattr_cache, ++ bs->bh->b_bdev, ++ bs->bh->b_blocknr); + BUFFER_TRACE(bs->bh, "get_write_access"); + error = ext4_journal_get_write_access(handle, bs->bh); + if (error) +@@ -1055,7 +1060,7 @@ ext4_xattr_block_set(handle_t *handle, s + if (!IS_LAST_ENTRY(s->first)) + ext4_xattr_rehash(header(s->base), + s->here); +- ext4_xattr_cache_insert(bs->bh); ++ ext4_xattr_cache_insert(sb, bs->bh); + } + unlock_buffer(bs->bh); + if (error == -EIO) +@@ -1138,7 +1143,8 @@ inserted: + if (error) + goto cleanup_dquot; + } +- mb_cache_entry_release(ce); ++ if (ce) ++ mb_cache_entry_release(ce); + ce = NULL; + } else if (bs->bh && s->base == bs->bh->b_data) { + /* We were modifying this block in-place. */ +@@ -1191,7 +1197,7 @@ getblk_failed: + memcpy(new_bh->b_data, s->base, new_bh->b_size); + set_buffer_uptodate(new_bh); + unlock_buffer(new_bh); +- ext4_xattr_cache_insert(new_bh); ++ ext4_xattr_cache_insert(sb, new_bh); + error = ext4_handle_dirty_xattr_block(handle, + inode, new_bh); + if (error) +@@ -1938,12 +1944,15 @@ ext4_xattr_put_super(struct super_block + * Returns 0, or a negative error number on failure. + */ + static void +-ext4_xattr_cache_insert(struct buffer_head *bh) ++ext4_xattr_cache_insert(struct super_block *sb, struct buffer_head *bh) + { + __u32 hash = le32_to_cpu(BHDR(bh)->h_hash); + struct mb_cache_entry *ce; + int error; + ++ if (test_opt(sb, NO_MBCACHE)) ++ return; ++ + ce = mb_cache_entry_alloc(ext4_xattr_cache, GFP_NOFS); + if (!ce) { + ea_bdebug(bh, "out of memory"); +@@ -2016,6 +2025,8 @@ ext4_xattr_cache_find(struct inode *inod + __u32 hash = le32_to_cpu(header->h_hash); + struct mb_cache_entry *ce; + ++ if (test_opt(inode->i_sb, NO_MBCACHE)) ++ return NULL; + if (!header->h_hash) + return NULL; /* never share */ + ea_idebug(inode, "looking for cached blocks [%x]", (int)hash); diff --git a/ldiskfs/kernel_patches/patches/rhel7/ext4-hash-indexed-dir-dotdot-update.patch b/ldiskfs/kernel_patches/patches/rhel7/ext4-hash-indexed-dir-dotdot-update.patch new file mode 100644 index 0000000..5a78064 --- /dev/null +++ b/ldiskfs/kernel_patches/patches/rhel7/ext4-hash-indexed-dir-dotdot-update.patch @@ -0,0 +1,87 @@ +Index: linux-3.10.0-123.9.3.el7.x86_64/fs/ext4/namei.c +=================================================================== +--- linux-3.10.0-123.9.3.el7.x86_64.orig/fs/ext4/namei.c ++++ linux-3.10.0-123.9.3.el7.x86_64/fs/ext4/namei.c +@@ -1894,6 +1894,72 @@ static int make_indexed_dir(handle_t *ha + return retval; + } + ++/* update ".." for hash-indexed directory, split the item "." if necessary */ ++static int ext4_update_dotdot(handle_t *handle, struct dentry *dentry, ++ struct inode *inode) ++{ ++ struct inode * dir = dentry->d_parent->d_inode; ++ struct buffer_head * dir_block; ++ struct ext4_dir_entry_2 * de; ++ int len, journal = 0, err = 0; ++ ++ if (IS_ERR(handle)) ++ return PTR_ERR(handle); ++ ++ if (IS_DIRSYNC(dir)) ++ handle->h_sync = 1; ++ ++ dir_block = ext4_bread(handle, dir, 0, 0, &err); ++ if (!dir_block) ++ goto out; ++ ++ de = (struct ext4_dir_entry_2 *)dir_block->b_data; ++ /* the first item must be "." */ ++ assert(de->name_len == 1 && de->name[0] == '.'); ++ len = le16_to_cpu(de->rec_len); ++ assert(len >= EXT4_DIR_REC_LEN(1)); ++ if (len > EXT4_DIR_REC_LEN(1)) { ++ BUFFER_TRACE(dir_block, "get_write_access"); ++ err = ext4_journal_get_write_access(handle, dir_block); ++ if (err) ++ goto out_journal; ++ ++ journal = 1; ++ de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(1)); ++ } ++ ++ len -= EXT4_DIR_REC_LEN(1); ++ assert(len == 0 || len >= EXT4_DIR_REC_LEN(2)); ++ de = (struct ext4_dir_entry_2 *) ++ ((char *) de + le16_to_cpu(de->rec_len)); ++ if (!journal) { ++ BUFFER_TRACE(dir_block, "get_write_access"); ++ err = ext4_journal_get_write_access(handle, dir_block); ++ if (err) ++ goto out_journal; ++ } ++ ++ de->inode = cpu_to_le32(inode->i_ino); ++ if (len > 0) ++ de->rec_len = cpu_to_le16(len); ++ else ++ 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); ++ ++out_journal: ++ if (journal) { ++ BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata"); ++ err = ext4_handle_dirty_dirent_node(handle, dir, dir_block); ++ ext4_mark_inode_dirty(handle, dir); ++ } ++ brelse (dir_block); ++ ++out: ++ return err; ++} ++ + /* + * ext4_add_entry() + * +@@ -1938,6 +2004,9 @@ int ext4_add_entry(handle_t *handle, str + } + + if (is_dx(dir)) { ++ 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, dentry, inode); + if (!retval || (retval != ERR_BAD_DX_DIR)) + return retval; diff --git a/ldiskfs/kernel_patches/patches/rhel7/ext4-inode-version.patch b/ldiskfs/kernel_patches/patches/rhel7/ext4-inode-version.patch new file mode 100644 index 0000000..dafdac7 --- /dev/null +++ b/ldiskfs/kernel_patches/patches/rhel7/ext4-inode-version.patch @@ -0,0 +1,56 @@ +Index: linux-3.10.0-123.el7.x86_64/fs/ext4/inode.c +=================================================================== +--- linux-3.10.0-123.el7.x86_64.orig/fs/ext4/inode.c ++++ linux-3.10.0-123.el7.x86_64/fs/ext4/inode.c +@@ -4286,10 +4286,10 @@ struct inode *ext4_iget(struct super_blo + EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); + + if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { +- inode->i_version = le32_to_cpu(raw_inode->i_disk_version); ++ ei->i_fs_version = le32_to_cpu(raw_inode->i_disk_version); + if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { + if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) +- inode->i_version |= ++ ei->i_fs_version |= + (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; + } + } +@@ -4506,11 +4506,11 @@ static int ext4_do_update_inode(handle_t + } + + if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { +- raw_inode->i_disk_version = cpu_to_le32(inode->i_version); ++ raw_inode->i_disk_version = cpu_to_le32(ei->i_fs_version); + if (ei->i_extra_isize) { + if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) + raw_inode->i_version_hi = +- cpu_to_le32(inode->i_version >> 32); ++ cpu_to_le32(ei->i_fs_version >> 32); + raw_inode->i_extra_isize = + cpu_to_le16(ei->i_extra_isize); + } +Index: linux-3.10.0-123.el7.x86_64/fs/ext4/ialloc.c +=================================================================== +--- linux-3.10.0-123.el7.x86_64.orig/fs/ext4/ialloc.c ++++ linux-3.10.0-123.el7.x86_64/fs/ext4/ialloc.c +@@ -898,6 +898,7 @@ got: + ei->i_dtime = 0; + ei->i_block_group = group; + ei->i_last_alloc_group = ~0; ++ ei->i_fs_version = 0; + + ext4_set_inode_flags(inode); + if (IS_DIRSYNC(inode)) +Index: linux-3.10.0-123.el7.x86_64/fs/ext4/ext4.h +=================================================================== +--- linux-3.10.0-123.el7.x86_64.orig/fs/ext4/ext4.h ++++ linux-3.10.0-123.el7.x86_64/fs/ext4/ext4.h +@@ -921,6 +921,8 @@ struct ext4_inode_info { + tid_t i_sync_tid; + tid_t i_datasync_tid; + ++ __u64 i_fs_version; ++ + /* Precomputed uuid+inum+igen checksum for seeding inode checksums */ + __u32 i_csum_seed; + }; diff --git a/ldiskfs/kernel_patches/patches/rhel7/ext4-kill-dx-root.patch b/ldiskfs/kernel_patches/patches/rhel7/ext4-kill-dx-root.patch new file mode 100644 index 0000000..f368040 --- /dev/null +++ b/ldiskfs/kernel_patches/patches/rhel7/ext4-kill-dx-root.patch @@ -0,0 +1,235 @@ +removes static definition of dx_root struct. so that "." and ".." dirent can +have extra data. This patch does not change any functionality but is required for +ext4_data_in_dirent patch. + +Index: linux-3.10.0-123.9.3.el7.x86_64/fs/ext4/namei.c +=================================================================== +--- linux-3.10.0-123.9.3.el7.x86_64.orig/fs/ext4/namei.c ++++ linux-3.10.0-123.9.3.el7.x86_64/fs/ext4/namei.c +@@ -193,22 +193,13 @@ 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[0]; ++ __le32 reserved_zero; ++ u8 hash_version; ++ u8 info_length; /* 8 */ ++ u8 indirect_levels; ++ u8 unused_flags; + }; + + struct dx_node +@@ -511,6 +502,16 @@ 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) + { +@@ -673,7 +674,7 @@ dx_probe(const struct qstr *d_name, stru + { + unsigned count, indirect; + struct dx_entry *at, *entries, *p, *q, *m; +- struct dx_root *root; ++ struct dx_root_info * info; + struct buffer_head *bh; + struct dx_frame *frame = frame_in; + u32 hash; +@@ -684,17 +685,18 @@ dx_probe(const struct qstr *d_name, stru + *err = PTR_ERR(bh); + goto fail; + } +- root = (struct dx_root *) 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) { ++ ++ info = dx_get_dx_info((struct ext4_dir_entry_2*)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", root->info.hash_version, dir->i_ino); ++ "#%lu", info->hash_version, dir->i_ino); + brelse(bh); + *err = ERR_BAD_DX_DIR; + goto fail; + } +- 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; +@@ -702,27 +704,26 @@ dx_probe(const struct qstr *d_name, stru + ext4fs_dirhash(d_name->name, d_name->len, hinfo); + hash = hinfo->hash; + +- if (root->info.unused_flags & 1) { ++ if (info->unused_flags & 1) { + ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x", +- root->info.unused_flags); ++ info->unused_flags); + brelse(bh); + *err = ERR_BAD_DX_DIR; + goto fail; + } + +- if ((indirect = root->info.indirect_levels) > 1) { ++ if ((indirect = info->indirect_levels) > 1) { + ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x", +- root->info.indirect_levels); ++ info->indirect_levels); + brelse(bh); + *err = ERR_BAD_DX_DIR; + 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(dir->i_sb, "dx entry: limit != root limit"); + brelse(bh); + *err = ERR_BAD_DX_DIR; +@@ -807,10 +808,12 @@ fail: + + static void dx_release (struct dx_frame *frames) + { ++ struct dx_root_info *info; + if (frames[0].bh == NULL) + return; + +- if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels) ++ info = dx_get_dx_info((struct ext4_dir_entry_2*)frames[0].bh->b_data); ++ if (info->indirect_levels) + brelse(frames[1].bh); + brelse(frames[0].bh); + } +@@ -1787,10 +1790,9 @@ static int make_indexed_dir(handle_t *ha + const char *name = dentry->d_name.name; + int namelen = dentry->d_name.len; + struct buffer_head *bh2; +- struct dx_root *root; + struct dx_frame frames[2], *frame; + struct dx_entry *entries; +- struct ext4_dir_entry_2 *de, *de2; ++ struct ext4_dir_entry_2 *de, *de2, *dot_de, *dotdot_de; + struct ext4_dir_entry_tail *t; + char *data1, *top; + unsigned len; +@@ -1798,7 +1800,7 @@ static int make_indexed_dir(handle_t *ha + unsigned blocksize; + struct dx_hash_info hinfo; + ext4_lblk_t block; +- struct fake_dirent *fde; ++ struct dx_root_info *dx_info; + int csum_size = 0; + + if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, +@@ -1813,18 +1815,19 @@ 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 -EIO; + } +- 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); +@@ -1850,19 +1853,23 @@ static int make_indexed_dir(handle_t *ha + } + + /* 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 */ +- hinfo.hash_version = root->info.hash_version; ++ hinfo.hash_version = dx_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; +@@ -2152,6 +2159,7 @@ static int ext4_dx_add_entry(handle_t *h + goto journal_error; + brelse (bh2); + } else { ++ struct dx_root_info * info; + dxtrace(printk(KERN_DEBUG + "Creating second level index...\n")); + memcpy((char *) entries2, (char *) entries, +@@ -2161,7 +2169,9 @@ static int ext4_dx_add_entry(handle_t *h + /* Set up root */ + dx_set_count(entries, 1); + dx_set_block(entries + 0, newblock); +- ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1; ++ info = dx_get_dx_info((struct ext4_dir_entry_2*) ++ frames[0].bh->b_data); ++ info->indirect_levels = 1; + + /* Add new access path frame */ + frame = frames + 1; diff --git a/ldiskfs/kernel_patches/patches/rhel7/ext4-large-eas.patch b/ldiskfs/kernel_patches/patches/rhel7/ext4-large-eas.patch new file mode 100644 index 0000000..da71240 --- /dev/null +++ b/ldiskfs/kernel_patches/patches/rhel7/ext4-large-eas.patch @@ -0,0 +1,1009 @@ +This patch implements the large EA support in ext4. If the size of +an EA value is larger than the blocksize, then the EA value would +not be saved in the external EA block, instead it would be saved +in an external EA inode. So, the patch also helps support a larger +number of EAs. + +Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/ext4.h +=================================================================== +--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/ext4.h ++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/ext4.h +@@ -1534,6 +1534,7 @@ static inline void ext4_clear_state_flag + EXT4_FEATURE_INCOMPAT_EXTENTS| \ + EXT4_FEATURE_INCOMPAT_64BIT| \ + EXT4_FEATURE_INCOMPAT_FLEX_BG| \ ++ EXT4_FEATURE_INCOMPAT_EA_INODE| \ + EXT4_FEATURE_INCOMPAT_MMP | \ + EXT4_FEATURE_INCOMPAT_DIRDATA| \ + EXT4_FEATURE_INCOMPAT_INLINE_DATA) +@@ -1934,6 +1935,12 @@ struct mmpd_data { + #define EXT4_MMP_MAX_CHECK_INTERVAL 300UL + + /* ++ * Maximum size of xattr attributes for FEATURE_INCOMPAT_EA_INODE 1Mb ++ * This limit is arbitrary, but is reasonable for the xattr API. ++ */ ++#define EXT4_XATTR_MAX_LARGE_EA_SIZE (1024 * 1024) ++ ++/* + * Function prototypes + */ + +@@ -1945,6 +1952,10 @@ struct mmpd_data { + # define ATTRIB_NORET __attribute__((noreturn)) + # define NORET_AND noreturn, + ++struct ext4_xattr_ino_array { ++ unsigned int xia_count; /* # of used item in the array */ ++ unsigned int xia_inodes[0]; ++}; + /* bitmap.c */ + extern unsigned int ext4_count_free(char *bitmap, unsigned numchars); + void ext4_inode_bitmap_csum_set(struct super_block *sb, ext4_group_t group, +@@ -2157,6 +2168,7 @@ extern void ext4_set_inode_flags(struct + extern void ext4_get_inode_flags(struct ext4_inode_info *); + extern int ext4_alloc_da_blocks(struct inode *inode); + extern void ext4_set_aops(struct inode *inode); ++extern int ext4_meta_trans_blocks(struct inode *, int nrblocks, int chunk); + extern int ext4_writepage_trans_blocks(struct inode *); + extern int ext4_chunk_trans_blocks(struct inode *, int nrblocks); + extern int ext4_discard_partial_page_buffers(handle_t *handle, +Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/inode.c +=================================================================== +--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/inode.c ++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/inode.c +@@ -133,8 +183,6 @@ void ext4_evict_inode(struct inode *inod + unsigned int length); + static int __ext4_journalled_writepage(struct page *page, unsigned int len); + static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh); +-static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, +- int pextents); + + /* + * Test whether an inode is a fast symlink. +@@ -183,6 +183,8 @@ void ext4_evict_inode(struct inode *inod + { + handle_t *handle; + int err; ++ int extra_credits = 3; ++ struct ext4_xattr_ino_array *lea_ino_array = NULL; + + trace_ext4_evict_inode(inode); + +@@ -235,8 +237,8 @@ void ext4_evict_inode(struct inode *inod + * protection against it + */ + sb_start_intwrite(inode->i_sb); +- handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, +- ext4_blocks_for_truncate(inode)+3); ++ ++ handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, extra_credits); + if (IS_ERR(handle)) { + ext4_std_error(inode->i_sb, PTR_ERR(handle)); + /* +@@ -248,9 +250,33 @@ void ext4_evict_inode(struct inode *inod + sb_end_intwrite(inode->i_sb); + goto no_delete; + } +- + if (IS_SYNC(inode)) + ext4_handle_sync(handle); ++ ++ /* ++ * Delete xattr inode before deleting the main inode. ++ */ ++ err = ext4_xattr_delete_inode(handle, inode, &lea_ino_array); ++ if (err) { ++ ext4_warning(inode->i_sb, ++ "couldn't delete inode's xattr (err %d)", err); ++ goto stop_handle; ++ } ++ ++ if (!ext4_handle_has_enough_credits(handle, ++ ext4_blocks_for_truncate(inode) + extra_credits)) { ++ err = ext4_journal_extend(handle, ++ ext4_blocks_for_truncate(inode) + extra_credits); ++ if (err > 0) ++ err = ext4_journal_restart(handle, ++ ext4_blocks_for_truncate(inode) + extra_credits); ++ if (err != 0) { ++ ext4_warning(inode->i_sb, ++ "couldn't extend journal (err %d)", err); ++ goto stop_handle; ++ } ++ } ++ + inode->i_size = 0; + err = ext4_mark_inode_dirty(handle, inode); + if (err) { +@@ -305,8 +331,12 @@ void ext4_evict_inode(struct inode *inod + ext4_clear_inode(inode); + else + ext4_free_inode(handle, inode); ++ + ext4_journal_stop(handle); + sb_end_intwrite(inode->i_sb); ++ ++ if (lea_ino_array != NULL) ++ ext4_xattr_inode_array_free(inode, lea_ino_array); + return; + no_delete: + ext4_clear_inode(inode); /* We must guarantee clearing of inode... */ +@@ -4841,7 +4871,7 @@ static int ext4_index_trans_blocks(struc + * + * Also account for superblock, inode, quota and xattr blocks + */ +-static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, ++int ext4_meta_trans_blocks(struct inode *inode, int lblocks, + int pextents) + { + ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); + int gdpblocks; +Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/xattr.c +=================================================================== +--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/xattr.c ++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/xattr.c +@@ -220,19 +220,26 @@ ext4_xattr_check_block(struct inode *ino + } + + static inline int +-ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size) ++ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size, ++ struct inode *inode) + { + size_t value_size = le32_to_cpu(entry->e_value_size); + +- if (entry->e_value_block != 0 || value_size > size || +- le16_to_cpu(entry->e_value_offs) + value_size > size) ++ if ((entry->e_value_inum == 0) && ++ (le16_to_cpu(entry->e_value_offs) + value_size > size)) ++ return -EIO; ++ if (entry->e_value_inum != 0 && ++ (le32_to_cpu(entry->e_value_inum) < EXT4_FIRST_INO(inode->i_sb) || ++ le32_to_cpu(entry->e_value_inum) > ++ le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_inodes_count))) + return -EIO; + return 0; + } + + static int + ext4_xattr_find_entry(struct ext4_xattr_entry **pentry, int name_index, +- const char *name, size_t size, int sorted) ++ const char *name, size_t size, int sorted, ++ struct inode *inode) + { + struct ext4_xattr_entry *entry; + size_t name_len; +@@ -252,11 +259,103 @@ ext4_xattr_find_entry(struct ext4_xattr_ + break; + } + *pentry = entry; +- if (!cmp && ext4_xattr_check_entry(entry, size)) ++ if (!cmp && ext4_xattr_check_entry(entry, size, inode)) + return -EIO; + return cmp ? -ENODATA : 0; + } + ++/* ++ * Read the EA value from an inode. ++ */ ++static int ++ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t *size) ++{ ++ unsigned long block = 0; ++ struct buffer_head *bh = NULL; ++ int err, blocksize; ++ size_t csize, ret_size = 0; ++ ++ if (*size == 0) ++ return 0; ++ ++ blocksize = ea_inode->i_sb->s_blocksize; ++ ++ while (ret_size < *size) { ++ csize = (*size - ret_size) > blocksize ? blocksize : ++ *size - ret_size; ++ bh = ext4_bread(NULL, ea_inode, block, 0, &err); ++ if (!bh) { ++ *size = ret_size; ++ return err; ++ } ++ memcpy(buf, bh->b_data, csize); ++ brelse(bh); ++ ++ buf += csize; ++ block += 1; ++ ret_size += csize; ++ } ++ ++ *size = ret_size; ++ ++ return err; ++} ++ ++struct inode *ext4_xattr_inode_iget(struct inode *parent, int ea_ino, int *err) ++{ ++ struct inode *ea_inode = NULL; ++ ++ ea_inode = ext4_iget(parent->i_sb, ea_ino); ++ if (IS_ERR(ea_inode) || is_bad_inode(ea_inode)) { ++ ext4_error(parent->i_sb, "error while reading EA inode %d", ++ ea_ino); ++ *err = -EIO; ++ return NULL; ++ } ++ ++ if (ea_inode->i_xattr_inode_parent != parent->i_ino || ++ ea_inode->i_generation != parent->i_generation) { ++ ext4_error(parent->i_sb, "Backpointer from EA inode %d " ++ "to parent invalid.", ea_ino); ++ *err = -EINVAL; ++ goto error; ++ } ++ ++ if (!(EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL)) { ++ ext4_error(parent->i_sb, "EA inode %d does not have " ++ "EXT4_EA_INODE_FL flag set.\n", ea_ino); ++ *err = -EINVAL; ++ goto error; ++ } ++ ++ *err = 0; ++ return ea_inode; ++ ++error: ++ iput(ea_inode); ++ return NULL; ++} ++ ++/* ++ * Read the value from the EA inode. ++ */ ++static int ++ext4_xattr_inode_get(struct inode *inode, int ea_ino, void *buffer, ++ size_t *size) ++{ ++ struct inode *ea_inode = NULL; ++ int err; ++ ++ ea_inode = ext4_xattr_inode_iget(inode, ea_ino, &err); ++ if (err) ++ return err; ++ ++ err = ext4_xattr_inode_read(ea_inode, buffer, size); ++ iput(ea_inode); ++ ++ return err; ++} ++ + static int + ext4_xattr_block_get(struct inode *inode, int name_index, const char *name, + void *buffer, size_t buffer_size) +@@ -288,7 +387,8 @@ bad_block: + } + ext4_xattr_cache_insert(bh); + entry = BFIRST(bh); +- error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1); ++ error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1, ++ inode); + if (error == -EIO) + goto bad_block; + if (error) +@@ -298,8 +398,16 @@ bad_block: + error = -ERANGE; + if (size > buffer_size) + goto cleanup; +- memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs), +- size); ++ if (entry->e_value_inum != 0) { ++ error = ext4_xattr_inode_get(inode, ++ le32_to_cpu(entry->e_value_inum), ++ buffer, &size); ++ if (error) ++ goto cleanup; ++ } else { ++ memcpy(buffer, bh->b_data + ++ le16_to_cpu(entry->e_value_offs), size); ++ } + } + error = size; + +@@ -333,7 +441,7 @@ ext4_xattr_ibody_get(struct inode *inode + if (error) + goto cleanup; + error = ext4_xattr_find_entry(&entry, name_index, name, +- end - (void *)entry, 0); ++ end - (void *)entry, 0, inode); + if (error) + goto cleanup; + size = le32_to_cpu(entry->e_value_size); +@@ -341,8 +449,16 @@ ext4_xattr_ibody_get(struct inode *inode + error = -ERANGE; + if (size > buffer_size) + goto cleanup; +- memcpy(buffer, (void *)IFIRST(header) + +- le16_to_cpu(entry->e_value_offs), size); ++ if (entry->e_value_inum != 0) { ++ error = ext4_xattr_inode_get(inode, ++ le32_to_cpu(entry->e_value_inum), ++ buffer, &size); ++ if (error) ++ goto cleanup; ++ } else { ++ memcpy(buffer, (void *)IFIRST(header) + ++ le16_to_cpu(entry->e_value_offs), size); ++ } + } + error = size; + +@@ -568,7 +684,7 @@ static size_t ext4_xattr_free_space(stru + *total += EXT4_XATTR_LEN(last->e_name_len); + { + for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) { +- if (!last->e_value_block && last->e_value_size) { ++ if (last->e_value_inum == 0 && last->e_value_size > 0) { + size_t offs = le16_to_cpu(last->e_value_offs); + if (offs < *min_offs) + *min_offs = offs; +@@ -577,16 +693,171 @@ static size_t ext4_xattr_free_space(stru + return (*min_offs - ((void *)last - base) - sizeof(__u32)); + } + ++/* ++ * Write the value of the EA in an inode. ++ */ ++static int ++ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode, ++ const void *buf, int bufsize) ++{ ++ struct buffer_head *bh = NULL; ++ unsigned long block = 0; ++ unsigned blocksize = ea_inode->i_sb->s_blocksize; ++ unsigned max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits; ++ int csize, wsize = 0; ++ int ret = 0; ++ int retries = 0; ++ ++retry: ++ while (ret >= 0 && ret < max_blocks) { ++ struct ext4_map_blocks map; ++ map.m_lblk = block += ret; ++ map.m_len = max_blocks -= ret; ++ ++ ret = ext4_map_blocks(handle, ea_inode, &map, EXT4_GET_BLOCKS_CREATE); ++ if (ret <= 0) { ++ ext4_mark_inode_dirty(handle, ea_inode); ++ if (ret == -ENOSPC && ++ ext4_should_retry_alloc(ea_inode->i_sb, &retries)) { ++ ret = 0; ++ goto retry; ++ } ++ break; ++ } ++ } ++ ++ if (ret < 0) ++ return ret; ++ ++ block = 0; ++ while (wsize < bufsize) { ++ if (bh != NULL) ++ brelse(bh); ++ csize = (bufsize - wsize) > blocksize ? blocksize : ++ bufsize - wsize; ++ bh = ext4_getblk(handle, ea_inode, block, 0, &ret); ++ if (!bh) ++ goto out; ++ ret = ext4_journal_get_write_access(handle, bh); ++ if (ret) ++ goto out; ++ ++ memcpy(bh->b_data, buf, csize); ++ set_buffer_uptodate(bh); ++ ext4_handle_dirty_metadata(handle, ea_inode, bh); ++ ++ buf += csize; ++ wsize += csize; ++ block += 1; ++ } ++ ++ i_size_write(ea_inode, wsize); ++ ext4_update_i_disksize(ea_inode, wsize); ++ ++ ext4_mark_inode_dirty(handle, ea_inode); ++ ++out: ++ brelse(bh); ++ ++ return ret; ++} ++ ++/* ++ * Create an inode to store the value of a large EA. ++ */ ++static struct inode * ++ext4_xattr_inode_create(handle_t *handle, struct inode *inode) ++{ ++ struct inode *ea_inode = NULL; ++ ++ /* ++ * Let the next inode be the goal, so we try and allocate the EA inode ++ * in the same group, or nearby one. ++ */ ++ ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode, ++ S_IFREG|0600, NULL, inode->i_ino + 1, NULL); ++ ++ if (!IS_ERR(ea_inode)) { ++ ea_inode->i_op = &ext4_file_inode_operations; ++ ea_inode->i_fop = &ext4_file_operations; ++ ext4_set_aops(ea_inode); ++ ea_inode->i_generation = inode->i_generation; ++ EXT4_I(ea_inode)->i_flags |= EXT4_EA_INODE_FL; ++ ++ /* ++ * A back-pointer from EA inode to parent inode will be useful ++ * for e2fsck. ++ */ ++ ea_inode->i_xattr_inode_parent = inode->i_ino; ++ unlock_new_inode(ea_inode); ++ } ++ ++ return ea_inode; ++} ++ ++/* ++ * Unlink the inode storing the value of the EA. ++ */ ++int ++ext4_xattr_inode_unlink(struct inode *inode, int ea_ino) ++{ ++ struct inode *ea_inode = NULL; ++ int err; ++ ++ ea_inode = ext4_xattr_inode_iget(inode, ea_ino, &err); ++ if (err) ++ return err; ++ ++ clear_nlink(ea_inode); ++ iput(ea_inode); ++ ++ return 0; ++} ++ ++/* ++ * Add value of the EA in an inode. ++ */ + static int +-ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s) ++ext4_xattr_inode_set(handle_t *handle, struct inode *inode, int *ea_ino, ++ const void *value, size_t value_len) ++{ ++ struct inode *ea_inode = NULL; ++ int err; ++ ++ /* Create an inode for the EA value */ ++ ea_inode = ext4_xattr_inode_create(handle, inode); ++ if (IS_ERR(ea_inode)) ++ return -1; ++ ++ err = ext4_xattr_inode_write(handle, ea_inode, value, value_len); ++ if (err) ++ clear_nlink(ea_inode); ++ else ++ *ea_ino = ea_inode->i_ino; ++ ++ iput(ea_inode); ++ ++ return err; ++} ++ ++static int ++ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s, ++ handle_t *handle, struct inode *inode) + { + struct ext4_xattr_entry *last; + size_t free, min_offs = s->end - s->base, name_len = strlen(i->name); ++ int in_inode = i->in_inode; ++ ++ if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb, ++ EXT4_FEATURE_INCOMPAT_EA_INODE) && ++ (EXT4_XATTR_SIZE(i->value_len) > ++ EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize))) ++ in_inode = 1; + + /* Compute min_offs and last. */ + last = s->first; + for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) { +- if (!last->e_value_block && last->e_value_size) { ++ if (last->e_value_inum == 0 && last->e_value_size > 0) { + size_t offs = le16_to_cpu(last->e_value_offs); + if (offs < min_offs) + min_offs = offs; +@@ -594,16 +865,21 @@ ext4_xattr_set_entry(struct ext4_xattr_i + } + free = min_offs - ((void *)last - s->base) - sizeof(__u32); + if (!s->not_found) { +- if (!s->here->e_value_block && s->here->e_value_size) { ++ if (!in_inode && s->here->e_value_inum == 0 && ++ s->here->e_value_size > 0) { + size_t size = le32_to_cpu(s->here->e_value_size); + free += EXT4_XATTR_SIZE(size); + } + free += EXT4_XATTR_LEN(name_len); + } + if (i->value) { +- if (free < EXT4_XATTR_SIZE(i->value_len) || +- free < EXT4_XATTR_LEN(name_len) + +- EXT4_XATTR_SIZE(i->value_len)) ++ size_t value_len = EXT4_XATTR_SIZE(i->value_len); ++ ++ if (in_inode) ++ value_len = 0; ++ ++ if (free < value_len || ++ free < EXT4_XATTR_LEN(name_len) + value_len) + return -ENOSPC; + } + +@@ -617,7 +893,8 @@ ext4_xattr_set_entry(struct ext4_xattr_i + s->here->e_name_len = name_len; + memcpy(s->here->e_name, i->name, name_len); + } else { +- if (!s->here->e_value_block && s->here->e_value_size) { ++ if (s->here->e_value_offs > 0 && s->here->e_value_inum == 0 && ++ s->here->e_value_size > 0) { + void *first_val = s->base + min_offs; + size_t offs = le16_to_cpu(s->here->e_value_offs); + void *val = s->base + offs; +@@ -651,13 +928,17 @@ ext4_xattr_set_entry(struct ext4_xattr_i + last = s->first; + while (!IS_LAST_ENTRY(last)) { + size_t o = le16_to_cpu(last->e_value_offs); +- if (!last->e_value_block && +- last->e_value_size && o < offs) ++ if (last->e_value_size > 0 && o < offs) + last->e_value_offs = + cpu_to_le16(o + size); + last = EXT4_XATTR_NEXT(last); + } + } ++ if (s->here->e_value_inum != 0) { ++ ext4_xattr_inode_unlink(inode, ++ le32_to_cpu(s->here->e_value_inum)); ++ s->here->e_value_inum = 0; ++ } + if (!i->value) { + /* Remove the old name. */ + size_t size = EXT4_XATTR_LEN(name_len); +@@ -671,10 +952,17 @@ ext4_xattr_set_entry(struct ext4_xattr_i + if (i->value) { + /* Insert the new value. */ + s->here->e_value_size = cpu_to_le32(i->value_len); +- if (i->value_len) { ++ if (in_inode) { ++ int ea_ino = le32_to_cpu(s->here->e_value_inum); ++ ext4_xattr_inode_set(handle, inode, &ea_ino, i->value, ++ i->value_len); ++ s->here->e_value_inum = cpu_to_le32(ea_ino); ++ s->here->e_value_offs = 0; ++ } else if (i->value_len) { + size_t size = EXT4_XATTR_SIZE(i->value_len); + void *val = s->base + min_offs - size; + s->here->e_value_offs = cpu_to_le16(min_offs - size); ++ s->here->e_value_inum = 0; + if (i->value == EXT4_ZERO_XATTR_VALUE) { + memset(val, 0, size); + } else { +@@ -724,7 +1012,7 @@ ext4_xattr_block_find(struct inode *inod + bs->s.end = bs->bh->b_data + bs->bh->b_size; + bs->s.here = bs->s.first; + error = ext4_xattr_find_entry(&bs->s.here, i->name_index, +- i->name, bs->bh->b_size, 1); ++ i->name, bs->bh->b_size, 1, inode); + if (error && error != -ENODATA) + goto cleanup; + bs->s.not_found = error; +@@ -748,8 +1036,6 @@ ext4_xattr_block_set(handle_t *handle, s + + #define header(x) ((struct ext4_xattr_header *)(x)) + +- if (i->value && i->value_len > sb->s_blocksize) +- return -ENOSPC; + if (s->base) { + ce = mb_cache_entry_get(ext4_xattr_cache, bs->bh->b_bdev, + bs->bh->b_blocknr); +@@ -764,7 +1050,7 @@ ext4_xattr_block_set(handle_t *handle, s + ce = NULL; + } + ea_bdebug(bs->bh, "modifying in-place"); +- error = ext4_xattr_set_entry(i, s); ++ error = ext4_xattr_set_entry(i, s, handle, inode); + if (!error) { + if (!IS_LAST_ENTRY(s->first)) + ext4_xattr_rehash(header(s->base), +@@ -815,7 +1101,7 @@ ext4_xattr_block_set(handle_t *handle, s + s->end = s->base + sb->s_blocksize; + } + +- error = ext4_xattr_set_entry(i, s); ++ error = ext4_xattr_set_entry(i, s, handle, inode); + if (error == -EIO) + goto bad_block; + if (error) +@@ -963,7 +1249,7 @@ int ext4_xattr_ibody_find(struct inode * + /* Find the named attribute. */ + error = ext4_xattr_find_entry(&is->s.here, i->name_index, + i->name, is->s.end - +- (void *)is->s.base, 0); ++ (void *)is->s.base, 0, inode); + if (error && error != -ENODATA) + return error; + is->s.not_found = error; +@@ -981,7 +1267,7 @@ int ext4_xattr_ibody_inline_set(handle_t + + if (EXT4_I(inode)->i_extra_isize == 0) + return -ENOSPC; +- error = ext4_xattr_set_entry(i, s); ++ error = ext4_xattr_set_entry(i, s, handle, inode); + if (error) { + if (error == -ENOSPC && + ext4_has_inline_data(inode)) { +@@ -993,7 +1279,7 @@ int ext4_xattr_ibody_inline_set(handle_t + error = ext4_xattr_ibody_find(inode, i, is); + if (error) + return error; +- error = ext4_xattr_set_entry(i, s); ++ error = ext4_xattr_set_entry(i, s, handle, inode); + } + if (error) + return error; +@@ -1019,7 +1305,7 @@ static int ext4_xattr_ibody_set(handle_t + + if (EXT4_I(inode)->i_extra_isize == 0) + return -ENOSPC; +- error = ext4_xattr_set_entry(i, s); ++ error = ext4_xattr_set_entry(i, s, handle, inode); + if (error) + return error; + header = IHDR(inode, ext4_raw_inode(&is->iloc)); +@@ -1055,7 +1341,7 @@ ext4_xattr_set_handle(handle_t *handle, + .name = name, + .value = value, + .value_len = value_len, +- ++ .in_inode = 0, + }; + struct ext4_xattr_ibody_find is = { + .s = { .not_found = -ENODATA, }, +@@ -1120,6 +1406,15 @@ ext4_xattr_set_handle(handle_t *handle, + goto cleanup; + } + error = ext4_xattr_block_set(handle, inode, &i, &bs); ++ if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb, ++ EXT4_FEATURE_INCOMPAT_EA_INODE) && ++ error == -ENOSPC) { ++ /* xattr not fit to block, store at external ++ * inode */ ++ i.in_inode = 1; ++ error = ext4_xattr_ibody_set(handle, inode, ++ &i, &is); ++ } + if (error) + goto cleanup; + if (!is.s.not_found) { +@@ -1166,9 +1461,22 @@ ext4_xattr_set(struct inode *inode, int + const void *value, size_t value_len, int flags) + { + handle_t *handle; ++ struct super_block *sb = inode->i_sb; + int error, retries = 0; + int credits = ext4_jbd2_credits_xattr(inode); + ++ if ((value_len >= EXT4_XATTR_MIN_LARGE_EA_SIZE(sb->s_blocksize)) && ++ EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EA_INODE)) { ++ int nrblocks = (value_len + sb->s_blocksize - 1) >> ++ sb->s_blocksize_bits; ++ ++ /* For new inode */ ++ credits += EXT4_SINGLEDATA_TRANS_BLOCKS(sb) + 3; ++ ++ /* For data blocks of EA inode */ ++ credits += ext4_meta_trans_blocks(inode, nrblocks, 0); ++ } ++ + retry: + handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits); + if (IS_ERR(handle)) { +@@ -1180,7 +1488,7 @@ retry: + value, value_len, flags); + error2 = ext4_journal_stop(handle); + if (error == -ENOSPC && +- ext4_should_retry_alloc(inode->i_sb, &retries)) ++ ext4_should_retry_alloc(sb, &retries)) + goto retry; + if (error == 0) + error = error2; +@@ -1202,7 +1510,7 @@ static void ext4_xattr_shift_entries(str + + /* Adjust the value offsets of the entries */ + for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) { +- if (!last->e_value_block && last->e_value_size) { ++ if (last->e_value_inum == 0 && last->e_value_size > 0) { + new_offs = le16_to_cpu(last->e_value_offs) + + value_offs_shift; + BUG_ON(new_offs + le32_to_cpu(last->e_value_size) +@@ -1441,19 +1749,91 @@ cleanup: + } + + +- ++#define EIA_INCR 16 /* must be 2^n */ ++#define EIA_MASK (EIA_INCR - 1) ++/* Add the large xattr @ino into @lea_ino_array for later deletion. ++ * If @lea_ino_array is new or full it will be grown and the old ++ * contents copied over. ++ */ ++static int ++ext4_expand_ino_array(struct ext4_xattr_ino_array **lea_ino_array, __u32 ino) ++{ ++ if (*lea_ino_array == NULL) { ++ /* ++ * Start with 15 inodes, so it fits into a power-of-two size. ++ * If *lea_ino_array is NULL, this is essentially offsetof() ++ */ ++ (*lea_ino_array) = ++ kmalloc(offsetof(struct ext4_xattr_ino_array, ++ xia_inodes[EIA_MASK]), ++ GFP_NOFS); ++ if (*lea_ino_array == NULL) ++ return -ENOMEM; ++ (*lea_ino_array)->xia_count = 0; ++ } else if (((*lea_ino_array)->xia_count & EIA_MASK) == EIA_MASK) { ++ /* expand the array once all 15 + n * 16 slots are full */ ++ struct ext4_xattr_ino_array *new_array = NULL; ++ int count = (*lea_ino_array)->xia_count; ++ ++ /* if new_array is NULL, this is essentially offsetof() */ ++ new_array = kmalloc( ++ offsetof(struct ext4_xattr_ino_array, ++ xia_inodes[count + EIA_INCR]), ++ GFP_NOFS); ++ if (new_array == NULL) ++ return -ENOMEM; ++ memcpy(new_array, *lea_ino_array, ++ offsetof(struct ext4_xattr_ino_array, ++ xia_inodes[count])); ++ kfree(*lea_ino_array); ++ *lea_ino_array = new_array; ++ } ++ (*lea_ino_array)->xia_inodes[(*lea_ino_array)->xia_count++] = ino; ++ return 0; ++} + /* + * ext4_xattr_delete_inode() + * +- * Free extended attribute resources associated with this inode. This ++ * Free extended attribute resources associated with this inode. Traverse ++ * all entries and unlink any xattr inodes associated with this inode. This + * is called immediately before an inode is freed. We have exclusive +- * access to the inode. ++ * access to the inode. If an orphan inode is deleted it will also delete any ++ * xattr block and all xattr inodes. They are checked by ext4_xattr_inode_iget() ++ * to ensure they belong to the parent inode and were not deleted already. + */ +-void +-ext4_xattr_delete_inode(handle_t *handle, struct inode *inode) ++int ++ext4_xattr_delete_inode(handle_t *handle, struct inode *inode, ++ struct ext4_xattr_ino_array **lea_ino_array) + { + struct buffer_head *bh = NULL; ++ struct ext4_xattr_ibody_header *header; ++ struct ext4_inode *raw_inode; ++ struct ext4_iloc iloc; ++ struct ext4_xattr_entry *entry; ++ int error = 0; ++ ++ if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR)) ++ goto delete_external_ea; ++ ++ error = ext4_get_inode_loc(inode, &iloc); ++ if (error) ++ goto cleanup; ++ raw_inode = ext4_raw_inode(&iloc); ++ header = IHDR(inode, raw_inode); ++ entry = IFIRST(header); ++ for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) { ++ if (entry->e_value_inum == 0) ++ continue; ++ if (ext4_expand_ino_array(lea_ino_array, ++ entry->e_value_inum) != 0) { ++ brelse(iloc.bh); ++ goto cleanup; ++ } ++ entry->e_value_inum = 0; ++ } ++ brelse(iloc.bh); + ++delete_external_ea: + if (!EXT4_I(inode)->i_file_acl) + goto cleanup; + bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl); +@@ -1468,11 +1848,74 @@ ext4_xattr_delete_inode(handle_t *handle + EXT4_I(inode)->i_file_acl); + goto cleanup; + } ++ ++ entry = BFIRST(bh); ++ for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) { ++ if (entry->e_value_inum == 0) ++ continue; ++ if (ext4_expand_ino_array(lea_ino_array, ++ entry->e_value_inum) != 0) ++ goto cleanup; ++ entry->e_value_inum = 0; ++ } ++ ++ /* add xattr inode to orphan list */ ++ if (*lea_ino_array != NULL) { ++ struct inode *ea_inode = NULL; ++ int idx = 0; ++ ++ for (; idx < (*lea_ino_array)->xia_count; ++idx) { ++ if (!ext4_handle_has_enough_credits(handle, 3)) { ++ error = ext4_journal_extend(handle, 3); ++ if (error > 0) ++ error = ext4_journal_restart(handle, 3); ++ if (error != 0) { ++ ext4_warning(inode->i_sb, ++ "couldn't extend journal " ++ "(err %d)", error); ++ goto cleanup; ++ } ++ } ++ ea_inode = ext4_xattr_inode_iget(inode, ++ (*lea_ino_array)->xia_inodes[idx], &error); ++ if (error) ++ continue; ++ ext4_orphan_add(handle, ea_inode); ++ /* the inode's i_count will be released by caller */ ++ } ++ } ++ + ext4_xattr_release_block(handle, inode, bh); + EXT4_I(inode)->i_file_acl = 0; + + cleanup: + brelse(bh); ++ ++ return error; ++} ++ ++void ++ext4_xattr_inode_array_free(struct inode *inode, ++ struct ext4_xattr_ino_array *lea_ino_array) ++{ ++ struct inode *ea_inode = NULL; ++ int idx = 0; ++ int err; ++ ++ if (lea_ino_array == NULL) ++ return; ++ ++ for (; idx < lea_ino_array->xia_count; ++idx) { ++ ea_inode = ext4_xattr_inode_iget(inode, ++ lea_ino_array->xia_inodes[idx], &err); ++ if (err) ++ continue; ++ clear_nlink(ea_inode); ++ iput(ea_inode); ++ /* for inode's i_count get from ext4_xattr_delete_inode */ ++ iput(ea_inode); ++ } ++ kfree(lea_ino_array); + } + + /* +@@ -1542,10 +1985,9 @@ ext4_xattr_cmp(struct ext4_xattr_header + entry1->e_name_index != entry2->e_name_index || + entry1->e_name_len != entry2->e_name_len || + entry1->e_value_size != entry2->e_value_size || ++ entry1->e_value_inum != entry2->e_value_inum || + memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len)) + return 1; +- if (entry1->e_value_block != 0 || entry2->e_value_block != 0) +- return -EIO; + if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs), + (char *)header2 + le16_to_cpu(entry2->e_value_offs), + le32_to_cpu(entry1->e_value_size))) +@@ -1629,7 +2071,7 @@ static inline void ext4_xattr_hash_entry + *name++; + } + +- if (entry->e_value_block == 0 && entry->e_value_size != 0) { ++ if (entry->e_value_inum == 0 && entry->e_value_size != 0) { + __le32 *value = (__le32 *)((char *)header + + le16_to_cpu(entry->e_value_offs)); + for (n = (le32_to_cpu(entry->e_value_size) + +Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/xattr.h +=================================================================== +--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/xattr.h ++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/xattr.h +@@ -42,7 +42,7 @@ struct ext4_xattr_entry { + __u8 e_name_len; /* length of name */ + __u8 e_name_index; /* attribute name index */ + __le16 e_value_offs; /* offset in disk block of value */ +- __le32 e_value_block; /* disk block attribute is stored on (n/i) */ ++ __le32 e_value_inum; /* inode in which the value is stored */ + __le32 e_value_size; /* size of attribute value */ + __le32 e_hash; /* hash value of name and value */ + char e_name[0]; /* attribute name */ +@@ -67,6 +67,15 @@ struct ext4_xattr_entry { + EXT4_I(inode)->i_extra_isize)) + #define IFIRST(hdr) ((struct ext4_xattr_entry *)((hdr)+1)) + ++#define i_xattr_inode_parent i_mtime.tv_sec ++ ++/* ++ * The minimum size of EA value when you start storing it in an external inode ++ * size of block - size of header - size of 1 entry - 4 null bytes ++*/ ++#define EXT4_XATTR_MIN_LARGE_EA_SIZE(b) \ ++ ((b) - EXT4_XATTR_LEN(3) - sizeof(struct ext4_xattr_header) - 4) ++ + #define BHDR(bh) ((struct ext4_xattr_header *)((bh)->b_data)) + #define ENTRY(ptr) ((struct ext4_xattr_entry *)(ptr)) + #define BFIRST(bh) ENTRY(BHDR(bh)+1) +@@ -75,10 +84,11 @@ struct ext4_xattr_entry { + #define EXT4_ZERO_XATTR_VALUE ((void *)-1) + + struct ext4_xattr_info { +- int name_index; + const char *name; + const void *value; + size_t value_len; ++ int name_index; ++ int in_inode; + }; + + struct ext4_xattr_search { +@@ -106,7 +116,13 @@ extern int ext4_xattr_get(struct inode * + extern int ext4_xattr_set(struct inode *, int, const char *, const void *, size_t, int); + extern int ext4_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int); + +-extern void ext4_xattr_delete_inode(handle_t *, struct inode *); ++extern struct inode *ext4_xattr_inode_iget(struct inode *parent, int ea_ino, ++ int *err); ++extern int ext4_xattr_inode_unlink(struct inode *inode, int ea_ino); ++extern int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode, ++ struct ext4_xattr_ino_array **array); ++extern void ext4_xattr_inode_array_free(struct inode *inode, ++ struct ext4_xattr_ino_array *array); + extern void ext4_xattr_put_super(struct super_block *); + + extern int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize, +Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/ialloc.c +=================================================================== +--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/ialloc.c ++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/ialloc.c +@@ -250,7 +250,6 @@ void ext4_free_inode(handle_t *handle, s + * as writing the quota to disk may need the lock as well. + */ + dquot_initialize(inode); +- ext4_xattr_delete_inode(handle, inode); + dquot_free_inode(inode); + dquot_drop(inode); + +Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/inline.c +=================================================================== +--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/inline.c ++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/inline.c +@@ -59,7 +59,7 @@ static int get_max_inline_xattr_value_si + + /* Compute min_offs. */ + for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) { +- if (!entry->e_value_block && entry->e_value_size) { ++ if (!entry->e_value_inum && entry->e_value_size) { + size_t offs = le16_to_cpu(entry->e_value_offs); + if (offs < min_offs) + min_offs = offs; diff --git a/ldiskfs/kernel_patches/patches/rhel7/ext4-lookup-dotdot.patch b/ldiskfs/kernel_patches/patches/rhel7/ext4-lookup-dotdot.patch new file mode 100644 index 0000000..f4318c5 --- /dev/null +++ b/ldiskfs/kernel_patches/patches/rhel7/ext4-lookup-dotdot.patch @@ -0,0 +1,37 @@ +Index: linux-3.10.9-200.fc17.x86_64/fs/ext4/namei.c +=================================================================== +--- linux-3.10.9-200.fc17.x86_64.orig/fs/ext4/namei.c ++++ linux-3.10.9-200.fc17.x86_64/fs/ext4/namei.c +@@ -1438,6 +1438,32 @@ static struct dentry *ext4_lookup(struct + return ERR_PTR(-EIO); + } + } ++ /* ".." shouldn't go into dcache to preserve dcache hierarchy ++ * otherwise we'll get parent being a child of actual child. ++ * see bug 10458 for details -bzzz */ ++ if (inode && (dentry->d_name.name[0] == '.' && (dentry->d_name.len == 1 || ++ (dentry->d_name.len == 2 && dentry->d_name.name[1] == '.')))) { ++ struct dentry *goal = NULL; ++ ++ /* first, look for an existing dentry - any one is good */ ++ goal = d_find_any_alias(inode); ++ if (goal == NULL) { ++ spin_lock(&dentry->d_lock); ++ /* there is no alias, we need to make current dentry: ++ * a) inaccessible for __d_lookup() ++ * b) inaccessible for iopen */ ++ J_ASSERT(hlist_unhashed(&dentry->d_alias)); ++ dentry->d_flags |= DCACHE_NFSFS_RENAMED; ++ /* this is d_instantiate() ... */ ++ hlist_add_head(&dentry->d_alias, &inode->i_dentry); ++ dentry->d_inode = inode; ++ spin_unlock(&dentry->d_lock); ++ } ++ if (goal) ++ iput(inode); ++ return goal; ++ } ++ + return d_splice_alias(inode, dentry); + } + diff --git a/ldiskfs/kernel_patches/patches/rhel7/ext4-max-dir-size.patch b/ldiskfs/kernel_patches/patches/rhel7/ext4-max-dir-size.patch new file mode 100644 index 0000000..9c8fc00 --- /dev/null +++ b/ldiskfs/kernel_patches/patches/rhel7/ext4-max-dir-size.patch @@ -0,0 +1,46 @@ +Add a proc interface for max_dir_size. + +Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/super.c +=================================================================== +--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/super.c ++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/super.c +@@ -2476,8 +2476,11 @@ static ssize_t sbi_ui_show(struct ext4_a + struct ext4_sb_info *sbi, char *buf) + { + unsigned int *ui = (unsigned int *) (((char *) sbi) + a->u.offset); ++ unsigned int v = *ui; + +- return snprintf(buf, PAGE_SIZE, "%u\n", *ui); ++ if (strcmp("max_dir_size", a->attr.name) == 0) ++ v <<= 10; ++ return snprintf(buf, PAGE_SIZE, "%u\n", v); + } + + static ssize_t sbi_ui_store(struct ext4_attr *a, +@@ -2491,6 +2494,8 @@ static ssize_t sbi_ui_store(struct ext4_ + ret = kstrtoul(skip_spaces(buf), 0, &t); + if (ret) + return ret; ++ if (strcmp("max_dir_size", a->attr.name) == 0) ++ t >>= 10; + *ui = t; + return count; + } +@@ -2557,6 +2562,8 @@ EXT4_RW_ATTR(reserved_clusters); + EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show, + inode_readahead_blks_store, 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); +@@ -2581,6 +2588,8 @@ 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), diff --git a/ldiskfs/kernel_patches/patches/rhel7/ext4-mballoc-extra-checks.patch b/ldiskfs/kernel_patches/patches/rhel7/ext4-mballoc-extra-checks.patch new file mode 100644 index 0000000..1716625 --- /dev/null +++ b/ldiskfs/kernel_patches/patches/rhel7/ext4-mballoc-extra-checks.patch @@ -0,0 +1,320 @@ +Index: linux-3.10.0-123.el7.x86_64/fs/ext4/ext4.h +=================================================================== +--- linux-3.10.0-123.el7.x86_64.orig/fs/ext4/ext4.h ++++ linux-3.10.0-123.el7.x86_64/fs/ext4/ext4.h +@@ -2391,6 +2391,7 @@ 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 +Index: linux-3.10.0-123.el7.x86_64/fs/ext4/mballoc.c +=================================================================== +--- linux-3.10.0-123.el7.x86_64.orig/fs/ext4/mballoc.c ++++ linux-3.10.0-123.el7.x86_64/fs/ext4/mballoc.c +@@ -362,7 +362,7 @@ static const char *ext4_groupinfo_slab_n + "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); +@@ -718,7 +718,7 @@ 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); +@@ -750,19 +750,13 @@ void 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); +- /* +- * If we intend to continue, we consider group descriptor +- * corrupt and update bb_free using bitmap value +- */ +- grp->bb_free = free; +- if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp)) +- percpu_counter_sub(&sbi->s_freeclusters_counter, +- grp->bb_free); +- set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state); ++ struct ext4_group_desc *gdp; ++ gdp = ext4_get_group_desc (sb, group, NULL); ++ ext4_error(sb, "group %lu: %u blocks in bitmap, %u in bb, " ++ "%u in gd, %lu pa's\n", (long unsigned int)group, ++ free, grp->bb_free, ext4_free_group_clusters(sb, gdp), ++ grp->bb_prealloc_nr); ++ return -EIO; + } + mb_set_largest_free_order(sb, grp); + +@@ -768,6 +767,8 @@ void ext4_mb_generate_buddy(struct super + EXT4_SB(sb)->s_mb_buddies_generated++; + EXT4_SB(sb)->s_mb_generation_time += period; + spin_unlock(&EXT4_SB(sb)->s_bal_lock); ++ ++ return 0; + } + + static void mb_regenerate_buddy(struct ext4_buddy *e4b) +@@ -883,7 +884,7 @@ 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; +@@ -922,7 +923,7 @@ 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 { +@@ -937,7 +938,7 @@ 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); + +@@ -947,7 +948,8 @@ static int ext4_mb_init_cache(struct pag + incore = data; + } + } +- SetPageUptodate(page); ++ if (likely(err == 0)) ++ SetPageUptodate(page); + + out: + if (bh) { +@@ -2224,9 +2226,11 @@ 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 = seq->private; ++ 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; + struct sg { +@@ -2236,10 +2240,10 @@ static int ext4_mb_seq_groups_show(struc + + group--; + if (group == 0) +- seq_printf(seq, "#%-5s: %-5s %-5s %-5s " ++ seq_printf(seq, "#%-5s: %-5s %-5s %-5s %-5s %-5s" + "[ %-5s %-5s %-5s %-5s %-5s %-5s %-5s " + "%-5s %-5s %-5s %-5s %-5s %-5s %-5s ]\n", +- "group", "free", "frags", "first", ++ "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"); + +@@ -2256,13 +2260,19 @@ 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 <= sb->s_blocksize_bits + 1 ? + sg.info.bb_counters[i] : 0); +@@ -3507,22 +3517,67 @@ 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_BLOCKS_PER_GROUP(sb); ++ unsigned short i, first, free = 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 != ext4_free_group_clusters(sb, gdp)) { ++ ext4_error(sb, "on-disk bitmap for group %d" ++ "corrupted: %u blocks free in bitmap, %u - in gd\n", ++ group, free, ext4_free_group_clusters(sb, 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 +@@ -3538,13 +3593,23 @@ 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(1, "prellocated %u for group %u\n", preallocated, group); ++ return 0; + } + + static void ext4_mb_pa_callback(struct rcu_head *head) +@@ -3603,6 +3668,7 @@ 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); +@@ -3697,6 +3763,7 @@ ext4_mb_new_inode_pa(struct ext4_allocat + + ext4_lock_group(sb, ac->ac_b_ex.fe_group); + list_add(&pa->pa_group_list, &grp->bb_prealloc_list); ++ grp->bb_prealloc_nr++; + ext4_unlock_group(sb, ac->ac_b_ex.fe_group); + + spin_lock(pa->pa_obj_lock); +@@ -3758,6 +3825,7 @@ ext4_mb_new_group_pa(struct ext4_allocat + + ext4_lock_group(sb, ac->ac_b_ex.fe_group); + list_add(&pa->pa_group_list, &grp->bb_prealloc_list); ++ grp->bb_prealloc_nr++; + ext4_unlock_group(sb, ac->ac_b_ex.fe_group); + + /* +@@ -3927,6 +3995,8 @@ 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); + } +@@ -4056,7 +4126,7 @@ repeat: + if (err) { + ext4_error(sb, "Error loading buddy information for %u", + group); +- continue; ++ return; + } + + bitmap_bh = ext4_read_block_bitmap(sb, group); +@@ -4068,6 +4138,8 @@ 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); +@@ -4328,6 +4400,7 @@ 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); + +Index: linux-3.10.0-123.el7.x86_64/fs/ext4/mballoc.h +=================================================================== +--- linux-3.10.0-123.el7.x86_64.orig/fs/ext4/mballoc.h ++++ linux-3.10.0-123.el7.x86_64/fs/ext4/mballoc.h +@@ -82,7 +82,7 @@ extern ushort ext4_mballoc_debug; + /* + * 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 diff --git a/ldiskfs/kernel_patches/patches/rhel7/ext4-mballoc-pa-free-mismatch.patch b/ldiskfs/kernel_patches/patches/rhel7/ext4-mballoc-pa-free-mismatch.patch new file mode 100644 index 0000000..a8aad4a --- /dev/null +++ b/ldiskfs/kernel_patches/patches/rhel7/ext4-mballoc-pa-free-mismatch.patch @@ -0,0 +1,110 @@ +Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/mballoc.c +=================================================================== +--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/mballoc.c ++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/mballoc.c +@@ -3747,6 +3747,7 @@ 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(1, "new inode pa %p: %llu/%u for %u\n", pa, + pa->pa_pstart, pa->pa_len, pa->pa_lstart); +@@ -3808,6 +3809,7 @@ 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(1, "new group pa %p: %llu/%u for %u\n", pa, + pa->pa_pstart, pa->pa_len, pa->pa_lstart); +@@ -3868,7 +3870,9 @@ ext4_mb_release_inode_pa(struct ext4_bud + int err = 0; + 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); +@@ -3891,12 +3895,18 @@ 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 %lu", +- pa, (unsigned long) pa->pa_lstart, +- (unsigned long) pa->pa_pstart, +- (unsigned long) 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 %lu] [freed %u]", 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->i_ino, ++ free); + ext4_grp_locked_error(sb, group, 0, 0, "free %u, pa_free %u", + free, pa->pa_free); + /* +@@ -3904,6 +3914,8 @@ 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 err; +@@ -4661,6 +4673,25 @@ 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: +Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/mballoc.h +=================================================================== +--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/mballoc.h ++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/mballoc.h +@@ -19,6 +19,7 @@ + #include + #include + #include ++#include + #include "ext4_jbd2.h" + #include "ext4.h" + +@@ -125,6 +126,7 @@ 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 */ + }; diff --git a/ldiskfs/kernel_patches/patches/rhel7/ext4-misc.patch b/ldiskfs/kernel_patches/patches/rhel7/ext4-misc.patch new file mode 100644 index 0000000..a7acd6d --- /dev/null +++ b/ldiskfs/kernel_patches/patches/rhel7/ext4-misc.patch @@ -0,0 +1,130 @@ +Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/ext4.h +=================================================================== +--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/ext4.h ++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/ext4.h +@@ -1427,6 +1427,8 @@ static inline void ext4_clear_state_flag + + #define NEXT_ORPHAN(inode) EXT4_I(inode)->i_dtime + ++#define JOURNAL_START_HAS_3ARGS 1 ++ + /* + * Codes for operating systems + */ +@@ -2612,6 +2614,11 @@ struct ext4_extent; + + extern int ext4_ext_tree_init(handle_t *handle, struct inode *); + extern int ext4_ext_writepage_trans_blocks(struct inode *, int); ++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); + int chunk); + extern int ext4_ext_map_blocks(handle_t *handle, struct inode *inode, +Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/namei.c +=================================================================== +--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/namei.c ++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/namei.c +@@ -48,7 +48,7 @@ + #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) + { +@@ -155,6 +155,7 @@ static struct buffer_head *__ext4_read_d + } + return bh; + } ++EXPORT_SYMBOL(ext4_append); + + #ifndef assert + #define assert(test) J_ASSERT(test) +@@ -2210,7 +2211,7 @@ out: + * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2, + * since this indicates that nlinks count was previously 1. + */ +-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 > 1) { +@@ -2222,16 +2223,18 @@ static void ext4_inc_count(handle_t *han + } + } + } ++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); + + + static int ext4_add_nondir(handle_t *handle, +Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/ialloc.c +=================================================================== +--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/ialloc.c ++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/ialloc.c +@@ -111,7 +111,7 @@ void ext4_end_bitmap_read(struct buffer_ + * + * Return buffer_head of bitmap on success or NULL. + */ +-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; +@@ -191,6 +191,7 @@ verify: + set_buffer_verified(bh); + return bh; + } ++EXPORT_SYMBOL(ext4_read_inode_bitmap); + + /* + * NOTE! When we get the inode, we're the only people +Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/inode.c +=================================================================== +--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/inode.c ++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/inode.c +@@ -5281,3 +5281,17 @@ out: + sb_end_pagefault(inode->i_sb); + 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); +Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/mballoc.c +=================================================================== +--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/mballoc.c ++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/mballoc.c +@@ -5281,7 +5281,6 @@ out: + void *buddy, void *bitmap, ext4_group_t group) + { + struct ext4_group_info *grp = ext4_get_group_info(sb, group); +- struct ext4_sb_info *sbi = EXT4_SB(sb); + ext4_grpblk_t max = EXT4_CLUSTERS_PER_GROUP(sb); + ext4_grpblk_t i = 0; + ext4_grpblk_t first; diff --git a/ldiskfs/kernel_patches/patches/rhel7/ext4-nocmtime.patch b/ldiskfs/kernel_patches/patches/rhel7/ext4-nocmtime.patch new file mode 100644 index 0000000..403627a --- /dev/null +++ b/ldiskfs/kernel_patches/patches/rhel7/ext4-nocmtime.patch @@ -0,0 +1,17 @@ +We won't change i_xtime in ldiskfs code path. But also +need keep normal function out of Lustre. So we using +S_NOCMTIME to indicate invoked from Lustre. + +Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/ext4.h +=================================================================== +--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/ext4.h ++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/ext4.h +@@ -1336,6 +1336,8 @@ static inline struct ext4_inode_info *EX + + static inline struct timespec ext4_current_time(struct inode *inode) + { ++ if (IS_NOCMTIME(inode)) ++ return inode->i_ctime; + return (inode->i_sb->s_time_gran < NSEC_PER_SEC) ? + current_fs_time(inode->i_sb) : CURRENT_TIME_SEC; + } diff --git a/ldiskfs/kernel_patches/patches/rhel7/ext4-osd-iop-common.patch b/ldiskfs/kernel_patches/patches/rhel7/ext4-osd-iop-common.patch new file mode 100644 index 0000000..d00452f --- /dev/null +++ b/ldiskfs/kernel_patches/patches/rhel7/ext4-osd-iop-common.patch @@ -0,0 +1,135 @@ +Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/ext4.h +=================================================================== +--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/ext4.h ++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/ext4.h +@@ -2145,6 +2145,19 @@ extern int ext4_orphan_add(handle_t *, s + extern int ext4_orphan_del(handle_t *, struct inode *); + extern int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash, + __u32 start_minor_hash, __u32 *next_hash); ++extern struct inode *ext4_create_inode(handle_t *handle, ++ struct inode * dir, int mode); ++extern int ext4_add_entry(handle_t *handle, struct dentry *dentry, ++ struct inode *inode); ++extern int ext4_delete_entry(handle_t *handle, struct inode * dir, ++ struct ext4_dir_entry_2 * de_del, ++ struct buffer_head * bh); ++extern struct buffer_head * ext4_find_entry(struct inode *dir, ++ const struct qstr *d_name, ++ struct ext4_dir_entry_2 ** res_dir, ++ int *inlined); ++extern int ext4_add_dot_dotdot(handle_t *handle, struct inode *dir, ++ struct inode *inode); + extern int search_dir(struct buffer_head *bh, + char *search_buf, + int buf_size, +Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/namei.c +=================================================================== +--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/namei.c ++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/namei.c +@@ -1211,7 +1211,7 @@ static int is_dx_internal_node(struct in + * The returned buffer_head has ->b_count elevated. The caller is expected + * to brelse() it when appropriate. + */ +-static struct buffer_head * ext4_find_entry (struct inode *dir, ++struct buffer_head * ext4_find_entry(struct inode *dir, + const struct qstr *d_name, + struct ext4_dir_entry_2 **res_dir, + int *inlined) +@@ -1355,6 +1355,7 @@ cleanup_and_exit: + brelse(bh_use[ra_ptr]); + return ret; + } ++EXPORT_SYMBOL(ext4_find_entry); + + static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name, + struct ext4_dir_entry_2 **res_dir, int *err) +@@ -1903,8 +1904,8 @@ static int make_indexed_dir(handle_t *ha + * 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(handle_t *handle, struct dentry *dentry, ++ struct inode *inode) + { + struct inode *dir = dentry->d_parent->d_inode; + struct buffer_head *bh; +@@ -1979,6 +1980,7 @@ static int ext4_add_entry(handle_t *hand + ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY); + return retval; + } ++EXPORT_SYMBOL(ext4_add_entry); + + /* + * Returns 0 for success, or a negative error value +@@ -2165,7 +2167,7 @@ int ext4_generic_delete_entry(handle_t * + return -ENOENT; + } + +-static int ext4_delete_entry(handle_t *handle, ++int ext4_delete_entry(handle_t *handle, + struct inode *dir, + struct ext4_dir_entry_2 *de_del, + struct buffer_head *bh) +@@ -2206,7 +2208,7 @@ out: + ext4_std_error(dir->i_sb, err); + return err; + } +- ++EXPORT_SYMBOL(ext4_delete_entry); + /* + * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2, + * since this indicates that nlinks count was previously 1. +@@ -2253,6 +2255,28 @@ static int ext4_add_nondir(handle_t *han + return err; + } + ++ /* Return locked inode, then the caller can modify the inode's states/flags ++ * before others finding it. The caller should unlock the inode by itself. */ ++struct inode * ext4_create_inode(handle_t *handle, struct inode * dir, int mode) ++{ ++ struct inode *inode; ++ ++ inode = ext4_new_inode(handle, dir, mode, NULL, 0, NULL); ++ if (!IS_ERR(inode)) { ++ if (S_ISCHR(mode) || S_ISBLK(mode) || S_ISFIFO(mode)) { ++#ifdef CONFIG_LDISKFS_FS_XATTR ++ inode->i_op = &ext4_special_inode_operations; ++#endif ++ } else { ++ inode->i_op = &ext4_file_inode_operations; ++ inode->i_fop = &ext4_file_operations; ++ ext4_set_aops(inode); ++ } ++ } ++ return inode; ++} ++EXPORT_SYMBOL(ext4_create_inode); ++ + /* + * By the time this is called, we already have created + * the directory cache entry for the new file, but it +@@ -2402,6 +2426,23 @@ 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) ++{ ++ if (IS_ERR(handle)) ++ return PTR_ERR(handle); ++ ++ if (IS_DIRSYNC(dir)) ++ ext4_handle_sync(handle); ++ ++ inode->i_op = &ext4_dir_inode_operations.ops; ++ inode->i_fop = &ext4_dir_operations; ++ return ext4_init_new_dir(handle, dir, inode); ++} ++EXPORT_SYMBOL(ext4_add_dot_dotdot); ++ + static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) + { + handle_t *handle; diff --git a/ldiskfs/kernel_patches/patches/rhel7/ext4-pdirop.patch b/ldiskfs/kernel_patches/patches/rhel7/ext4-pdirop.patch new file mode 100644 index 0000000..7d613aa --- /dev/null +++ b/ldiskfs/kernel_patches/patches/rhel7/ext4-pdirop.patch @@ -0,0 +1,2338 @@ +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 + - N-level htree directory + - integrate with osd-ldiskfs + +Index: linux-3.10.0-123.13.2.el7.x86_64/include/linux/htree_lock.h +=================================================================== +--- /dev/null ++++ linux-3.10.0-123.13.2.el7.x86_64/include/linux/htree_lock.h +@@ -0,0 +1,187 @@ ++/* ++ * include/linux/htree_lock.h ++ * ++ * Copyright (c) 2011, 2012, Intel Corporation. ++ * ++ * Author: Liang Zhen ++ */ ++ ++/* ++ * htree lock ++ * ++ * htree_lock is an advanced lock, it can support five lock modes (concept is ++ * taken from DLM) and it's a sleeping lock. ++ * ++ * most common use case is: ++ * - create a htree_lock_head for data ++ * - each thread (contender) creates it's own htree_lock ++ * - contender needs to call htree_lock(lock_node, mode) to protect data and ++ * call htree_unlock to release lock ++ * ++ * Also, there is advanced use-case which is more complex, user can have ++ * PW/PR lock on particular key, it's mostly used while user holding shared ++ * lock on the htree (CW, CR) ++ * ++ * htree_lock(lock_node, HTREE_LOCK_CR); lock the htree with CR ++ * htree_node_lock(lock_node, HTREE_LOCK_PR, key...); lock @key with PR ++ * ... ++ * htree_node_unlock(lock_node);; unlock the key ++ * ++ * Another tip is, we can have N-levels of this kind of keys, all we need to ++ * do is specifying N-levels while creating htree_lock_head, then we can ++ * lock/unlock a specific level by: ++ * htree_node_lock(lock_node, mode1, key1, level1...); ++ * do something; ++ * htree_node_lock(lock_node, mode1, key2, level2...); ++ * do something; ++ * htree_node_unlock(lock_node, level2); ++ * htree_node_unlock(lock_node, level1); ++ * ++ * NB: for multi-level, should be careful about locking order to avoid deadlock ++ */ ++ ++#ifndef _LINUX_HTREE_LOCK_H ++#define _LINUX_HTREE_LOCK_H ++ ++#include ++#include ++#include ++ ++/* ++ * Lock Modes ++ * more details can be found here: ++ * http://en.wikipedia.org/wiki/Distributed_lock_manager ++ */ ++typedef enum { ++ HTREE_LOCK_EX = 0, /* exclusive lock: incompatible with all others */ ++ HTREE_LOCK_PW, /* protected write: allows only CR users */ ++ HTREE_LOCK_PR, /* protected read: allow PR, CR users */ ++ HTREE_LOCK_CW, /* concurrent write: allow CR, CW users */ ++ HTREE_LOCK_CR, /* concurrent read: allow all but EX users */ ++ HTREE_LOCK_MAX, /* number of lock modes */ ++} htree_lock_mode_t; ++ ++#define HTREE_LOCK_NL HTREE_LOCK_MAX ++#define HTREE_LOCK_INVAL 0xdead10c ++ ++enum { ++ HTREE_HBITS_MIN = 2, ++ HTREE_HBITS_DEF = 14, ++ HTREE_HBITS_MAX = 32, ++}; ++ ++enum { ++ HTREE_EVENT_DISABLE = (0), ++ HTREE_EVENT_RD = (1 << HTREE_LOCK_PR), ++ HTREE_EVENT_WR = (1 << HTREE_LOCK_PW), ++ HTREE_EVENT_RDWR = (HTREE_EVENT_RD | HTREE_EVENT_WR), ++}; ++ ++struct htree_lock; ++ ++typedef void (*htree_event_cb_t)(void *target, void *event); ++ ++struct htree_lock_child { ++ struct list_head lc_list; /* granted list */ ++ htree_event_cb_t lc_callback; /* event callback */ ++ unsigned lc_events; /* event types */ ++}; ++ ++struct htree_lock_head { ++ unsigned long lh_lock; /* bits lock */ ++ /* blocked lock list (htree_lock) */ ++ struct list_head lh_blocked_list; ++ /* # key levels */ ++ u16 lh_depth; ++ /* hash bits for key and limit number of locks */ ++ u16 lh_hbits; ++ /* counters for blocked locks */ ++ u16 lh_nblocked[HTREE_LOCK_MAX]; ++ /* counters for granted locks */ ++ u16 lh_ngranted[HTREE_LOCK_MAX]; ++ /* private data */ ++ void *lh_private; ++ /* array of children locks */ ++ struct htree_lock_child lh_children[0]; ++}; ++ ++/* htree_lock_node_t is child-lock for a specific key (ln_value) */ ++struct htree_lock_node { ++ htree_lock_mode_t ln_mode; ++ /* major hash key */ ++ u16 ln_major_key; ++ /* minor hash key */ ++ u16 ln_minor_key; ++ struct list_head ln_major_list; ++ struct list_head ln_minor_list; ++ /* alive list, all locks (granted, blocked, listening) are on it */ ++ struct list_head ln_alive_list; ++ /* blocked list */ ++ struct list_head ln_blocked_list; ++ /* granted list */ ++ struct list_head ln_granted_list; ++ void *ln_ev_target; ++}; ++ ++struct htree_lock { ++ struct task_struct *lk_task; ++ struct htree_lock_head *lk_head; ++ void *lk_private; ++ unsigned lk_depth; ++ htree_lock_mode_t lk_mode; ++ struct list_head lk_blocked_list; ++ struct htree_lock_node lk_nodes[0]; ++}; ++ ++/* create a lock head, which stands for a resource */ ++struct htree_lock_head *htree_lock_head_alloc(unsigned depth, ++ unsigned hbits, unsigned priv); ++/* free a lock head */ ++void htree_lock_head_free(struct htree_lock_head *lhead); ++/* register event callback for child lock at level @depth */ ++void htree_lock_event_attach(struct htree_lock_head *lhead, unsigned depth, ++ unsigned events, htree_event_cb_t callback); ++/* create a lock handle, which stands for a thread */ ++struct htree_lock *htree_lock_alloc(unsigned depth, unsigned pbytes); ++/* free a lock handle */ ++void htree_lock_free(struct htree_lock *lck); ++/* lock htree, when @wait is true, 0 is returned if the lock can't ++ * be granted immediately */ ++int htree_lock_try(struct htree_lock *lck, struct htree_lock_head *lhead, ++ htree_lock_mode_t mode, int wait); ++/* unlock htree */ ++void htree_unlock(struct htree_lock *lck); ++/* unlock and relock htree with @new_mode */ ++int htree_change_lock_try(struct htree_lock *lck, ++ htree_lock_mode_t new_mode, int wait); ++void htree_change_mode(struct htree_lock *lck, htree_lock_mode_t mode); ++/* require child lock (key) of htree at level @dep, @event will be sent to all ++ * listeners on this @key while lock being granted */ ++int htree_node_lock_try(struct htree_lock *lck, htree_lock_mode_t mode, ++ u32 key, unsigned dep, int wait, void *event); ++/* release child lock at level @dep, this lock will listen on it's key ++ * if @event isn't NULL, event_cb will be called against @lck while granting ++ * any other lock at level @dep with the same key */ ++void htree_node_unlock(struct htree_lock *lck, unsigned dep, void *event); ++/* stop listening on child lock at level @dep */ ++void htree_node_stop_listen(struct htree_lock *lck, unsigned dep); ++/* for debug */ ++void htree_lock_stat_print(int depth); ++void htree_lock_stat_reset(void); ++ ++#define htree_lock(lck, lh, mode) htree_lock_try(lck, lh, mode, 1) ++#define htree_change_lock(lck, mode) htree_change_lock_try(lck, mode, 1) ++ ++#define htree_lock_mode(lck) ((lck)->lk_mode) ++ ++#define htree_node_lock(lck, mode, key, dep) \ ++ htree_node_lock_try(lck, mode, key, dep, 1, NULL) ++/* this is only safe in thread context of lock owner */ ++#define htree_node_is_granted(lck, dep) \ ++ ((lck)->lk_nodes[dep].ln_mode != HTREE_LOCK_INVAL && \ ++ (lck)->lk_nodes[dep].ln_mode != HTREE_LOCK_NL) ++/* this is only safe in thread context of lock owner */ ++#define htree_node_is_listening(lck, dep) \ ++ ((lck)->lk_nodes[dep].ln_mode == HTREE_LOCK_NL) ++ ++#endif +Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/htree_lock.c +=================================================================== +--- /dev/null ++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/htree_lock.c +@@ -0,0 +1,880 @@ ++/* ++ * fs/ext4/htree_lock.c ++ * ++ * Copyright (c) 2011, 2012, Intel Corporation. ++ * ++ * Author: Liang Zhen ++ */ ++#include ++#include ++#include ++#include ++ ++enum { ++ HTREE_LOCK_BIT_EX = (1 << HTREE_LOCK_EX), ++ HTREE_LOCK_BIT_PW = (1 << HTREE_LOCK_PW), ++ HTREE_LOCK_BIT_PR = (1 << HTREE_LOCK_PR), ++ HTREE_LOCK_BIT_CW = (1 << HTREE_LOCK_CW), ++ HTREE_LOCK_BIT_CR = (1 << HTREE_LOCK_CR), ++}; ++ ++enum { ++ HTREE_LOCK_COMPAT_EX = 0, ++ HTREE_LOCK_COMPAT_PW = HTREE_LOCK_COMPAT_EX | HTREE_LOCK_BIT_CR, ++ HTREE_LOCK_COMPAT_PR = HTREE_LOCK_COMPAT_PW | HTREE_LOCK_BIT_PR, ++ HTREE_LOCK_COMPAT_CW = HTREE_LOCK_COMPAT_PW | HTREE_LOCK_BIT_CW, ++ HTREE_LOCK_COMPAT_CR = HTREE_LOCK_COMPAT_CW | HTREE_LOCK_BIT_PR | ++ HTREE_LOCK_BIT_PW, ++}; ++ ++static int htree_lock_compat[] = { ++ [HTREE_LOCK_EX] HTREE_LOCK_COMPAT_EX, ++ [HTREE_LOCK_PW] HTREE_LOCK_COMPAT_PW, ++ [HTREE_LOCK_PR] HTREE_LOCK_COMPAT_PR, ++ [HTREE_LOCK_CW] HTREE_LOCK_COMPAT_CW, ++ [HTREE_LOCK_CR] HTREE_LOCK_COMPAT_CR, ++}; ++ ++/* max allowed htree-lock depth. ++ * We only need depth=3 for ext4 although user can have higher value. */ ++#define HTREE_LOCK_DEP_MAX 16 ++ ++#ifdef HTREE_LOCK_DEBUG ++ ++static char *hl_name[] = { ++ [HTREE_LOCK_EX] "EX", ++ [HTREE_LOCK_PW] "PW", ++ [HTREE_LOCK_PR] "PR", ++ [HTREE_LOCK_CW] "CW", ++ [HTREE_LOCK_CR] "CR", ++}; ++ ++/* lock stats */ ++struct htree_lock_node_stats { ++ unsigned long long blocked[HTREE_LOCK_MAX]; ++ unsigned long long granted[HTREE_LOCK_MAX]; ++ unsigned long long retried[HTREE_LOCK_MAX]; ++ unsigned long long events; ++}; ++ ++struct htree_lock_stats { ++ struct htree_lock_node_stats nodes[HTREE_LOCK_DEP_MAX]; ++ unsigned long long granted[HTREE_LOCK_MAX]; ++ unsigned long long blocked[HTREE_LOCK_MAX]; ++}; ++ ++static struct htree_lock_stats hl_stats; ++ ++void htree_lock_stat_reset(void) ++{ ++ memset(&hl_stats, 0, sizeof(hl_stats)); ++} ++ ++void htree_lock_stat_print(int depth) ++{ ++ int i; ++ int j; ++ ++ printk(KERN_DEBUG "HTREE LOCK STATS:\n"); ++ for (i = 0; i < HTREE_LOCK_MAX; i++) { ++ printk(KERN_DEBUG "[%s]: G [%10llu], B [%10llu]\n", ++ hl_name[i], hl_stats.granted[i], hl_stats.blocked[i]); ++ } ++ for (i = 0; i < depth; i++) { ++ printk(KERN_DEBUG "HTREE CHILD [%d] STATS:\n", i); ++ for (j = 0; j < HTREE_LOCK_MAX; j++) { ++ printk(KERN_DEBUG ++ "[%s]: G [%10llu], B [%10llu], R [%10llu]\n", ++ hl_name[j], hl_stats.nodes[i].granted[j], ++ hl_stats.nodes[i].blocked[j], ++ hl_stats.nodes[i].retried[j]); ++ } ++ } ++} ++ ++#define lk_grant_inc(m) do { hl_stats.granted[m]++; } while (0) ++#define lk_block_inc(m) do { hl_stats.blocked[m]++; } while (0) ++#define ln_grant_inc(d, m) do { hl_stats.nodes[d].granted[m]++; } while (0) ++#define ln_block_inc(d, m) do { hl_stats.nodes[d].blocked[m]++; } while (0) ++#define ln_retry_inc(d, m) do { hl_stats.nodes[d].retried[m]++; } while (0) ++#define ln_event_inc(d) do { hl_stats.nodes[d].events++; } while (0) ++ ++#else /* !DEBUG */ ++ ++void htree_lock_stat_reset(void) {} ++void htree_lock_stat_print(int depth) {} ++ ++#define lk_grant_inc(m) do {} while (0) ++#define lk_block_inc(m) do {} while (0) ++#define ln_grant_inc(d, m) do {} while (0) ++#define ln_block_inc(d, m) do {} while (0) ++#define ln_retry_inc(d, m) do {} while (0) ++#define ln_event_inc(d) do {} while (0) ++ ++#endif /* DEBUG */ ++ ++EXPORT_SYMBOL(htree_lock_stat_reset); ++EXPORT_SYMBOL(htree_lock_stat_print); ++ ++#define HTREE_DEP_ROOT (-1) ++ ++#define htree_spin_lock(lhead, dep) \ ++ bit_spin_lock((dep) + 1, &(lhead)->lh_lock) ++#define htree_spin_unlock(lhead, dep) \ ++ bit_spin_unlock((dep) + 1, &(lhead)->lh_lock) ++ ++#define htree_key_event_ignore(child, ln) \ ++ (!((child)->lc_events & (1 << (ln)->ln_mode))) ++ ++static int ++htree_key_list_empty(struct htree_lock_node *ln) ++{ ++ return list_empty(&ln->ln_major_list) && list_empty(&ln->ln_minor_list); ++} ++ ++static void ++htree_key_list_del_init(struct htree_lock_node *ln) ++{ ++ struct htree_lock_node *tmp = NULL; ++ ++ if (!list_empty(&ln->ln_minor_list)) { ++ tmp = list_entry(ln->ln_minor_list.next, ++ struct htree_lock_node, ln_minor_list); ++ list_del_init(&ln->ln_minor_list); ++ } ++ ++ if (list_empty(&ln->ln_major_list)) ++ return; ++ ++ if (tmp == NULL) { /* not on minor key list */ ++ list_del_init(&ln->ln_major_list); ++ } else { ++ BUG_ON(!list_empty(&tmp->ln_major_list)); ++ list_replace_init(&ln->ln_major_list, &tmp->ln_major_list); ++ } ++} ++ ++static void ++htree_key_list_replace_init(struct htree_lock_node *old, ++ struct htree_lock_node *new) ++{ ++ if (!list_empty(&old->ln_major_list)) ++ list_replace_init(&old->ln_major_list, &new->ln_major_list); ++ ++ if (!list_empty(&old->ln_minor_list)) ++ list_replace_init(&old->ln_minor_list, &new->ln_minor_list); ++} ++ ++static void ++htree_key_event_enqueue(struct htree_lock_child *child, ++ struct htree_lock_node *ln, int dep, void *event) ++{ ++ struct htree_lock_node *tmp; ++ ++ /* NB: ALWAYS called holding lhead::lh_lock(dep) */ ++ BUG_ON(ln->ln_mode == HTREE_LOCK_NL); ++ if (event == NULL || htree_key_event_ignore(child, ln)) ++ return; ++ ++ /* shouldn't be a very long list */ ++ list_for_each_entry(tmp, &ln->ln_alive_list, ln_alive_list) { ++ if (tmp->ln_mode == HTREE_LOCK_NL) { ++ ln_event_inc(dep); ++ if (child->lc_callback != NULL) ++ child->lc_callback(tmp->ln_ev_target, event); ++ } ++ } ++} ++ ++static int ++htree_node_lock_enqueue(struct htree_lock *newlk, struct htree_lock *curlk, ++ unsigned dep, int wait, void *event) ++{ ++ struct htree_lock_child *child = &newlk->lk_head->lh_children[dep]; ++ struct htree_lock_node *newln = &newlk->lk_nodes[dep]; ++ struct htree_lock_node *curln = &curlk->lk_nodes[dep]; ++ ++ /* NB: ALWAYS called holding lhead::lh_lock(dep) */ ++ /* NB: we only expect PR/PW lock mode at here, only these two modes are ++ * allowed for htree_node_lock(asserted in htree_node_lock_internal), ++ * NL is only used for listener, user can't directly require NL mode */ ++ if ((curln->ln_mode == HTREE_LOCK_NL) || ++ (curln->ln_mode != HTREE_LOCK_PW && ++ newln->ln_mode != HTREE_LOCK_PW)) { ++ /* no conflict, attach it on granted list of @curlk */ ++ if (curln->ln_mode != HTREE_LOCK_NL) { ++ list_add(&newln->ln_granted_list, ++ &curln->ln_granted_list); ++ } else { ++ /* replace key owner */ ++ htree_key_list_replace_init(curln, newln); ++ } ++ ++ list_add(&newln->ln_alive_list, &curln->ln_alive_list); ++ htree_key_event_enqueue(child, newln, dep, event); ++ ln_grant_inc(dep, newln->ln_mode); ++ return 1; /* still hold lh_lock */ ++ } ++ ++ if (!wait) { /* can't grant and don't want to wait */ ++ ln_retry_inc(dep, newln->ln_mode); ++ newln->ln_mode = HTREE_LOCK_INVAL; ++ return -1; /* don't wait and just return -1 */ ++ } ++ ++ newlk->lk_task = current; ++ set_current_state(TASK_UNINTERRUPTIBLE); ++ /* conflict, attach it on blocked list of curlk */ ++ list_add_tail(&newln->ln_blocked_list, &curln->ln_blocked_list); ++ list_add(&newln->ln_alive_list, &curln->ln_alive_list); ++ ln_block_inc(dep, newln->ln_mode); ++ ++ htree_spin_unlock(newlk->lk_head, dep); ++ /* wait to be given the lock */ ++ if (newlk->lk_task != NULL) ++ schedule(); ++ /* granted, no doubt, wake up will set me RUNNING */ ++ if (event == NULL || htree_key_event_ignore(child, newln)) ++ return 0; /* granted without lh_lock */ ++ ++ htree_spin_lock(newlk->lk_head, dep); ++ htree_key_event_enqueue(child, newln, dep, event); ++ return 1; /* still hold lh_lock */ ++} ++ ++/* ++ * get PR/PW access to particular tree-node according to @dep and @key, ++ * it will return -1 if @wait is false and can't immediately grant this lock. ++ * All listeners(HTREE_LOCK_NL) on @dep and with the same @key will get ++ * @event if it's not NULL. ++ * NB: ALWAYS called holding lhead::lh_lock ++ */ ++static int ++htree_node_lock_internal(struct htree_lock_head *lhead, struct htree_lock *lck, ++ htree_lock_mode_t mode, u32 key, unsigned dep, ++ int wait, void *event) ++{ ++ LIST_HEAD (list); ++ struct htree_lock *tmp; ++ struct htree_lock *tmp2; ++ u16 major; ++ u16 minor; ++ u8 reverse; ++ u8 ma_bits; ++ u8 mi_bits; ++ ++ BUG_ON(mode != HTREE_LOCK_PW && mode != HTREE_LOCK_PR); ++ BUG_ON(htree_node_is_granted(lck, dep)); ++ ++ key = hash_long(key, lhead->lh_hbits); ++ ++ mi_bits = lhead->lh_hbits >> 1; ++ ma_bits = lhead->lh_hbits - mi_bits; ++ ++ lck->lk_nodes[dep].ln_major_key = major = key & ((1U << ma_bits) - 1); ++ lck->lk_nodes[dep].ln_minor_key = minor = key >> ma_bits; ++ lck->lk_nodes[dep].ln_mode = mode; ++ ++ /* ++ * The major key list is an ordered list, so searches are started ++ * at the end of the list that is numerically closer to major_key, ++ * so at most half of the list will be walked (for well-distributed ++ * keys). The list traversal aborts early if the expected key ++ * location is passed. ++ */ ++ reverse = (major >= (1 << (ma_bits - 1))); ++ ++ if (reverse) { ++ list_for_each_entry_reverse(tmp, ++ &lhead->lh_children[dep].lc_list, ++ lk_nodes[dep].ln_major_list) { ++ if (tmp->lk_nodes[dep].ln_major_key == major) { ++ goto search_minor; ++ ++ } else if (tmp->lk_nodes[dep].ln_major_key < major) { ++ /* attach _after_ @tmp */ ++ list_add(&lck->lk_nodes[dep].ln_major_list, ++ &tmp->lk_nodes[dep].ln_major_list); ++ goto out_grant_major; ++ } ++ } ++ ++ list_add(&lck->lk_nodes[dep].ln_major_list, ++ &lhead->lh_children[dep].lc_list); ++ goto out_grant_major; ++ ++ } else { ++ list_for_each_entry(tmp, &lhead->lh_children[dep].lc_list, ++ lk_nodes[dep].ln_major_list) { ++ if (tmp->lk_nodes[dep].ln_major_key == major) { ++ goto search_minor; ++ ++ } else if (tmp->lk_nodes[dep].ln_major_key > major) { ++ /* insert _before_ @tmp */ ++ list_add_tail(&lck->lk_nodes[dep].ln_major_list, ++ &tmp->lk_nodes[dep].ln_major_list); ++ goto out_grant_major; ++ } ++ } ++ ++ list_add_tail(&lck->lk_nodes[dep].ln_major_list, ++ &lhead->lh_children[dep].lc_list); ++ goto out_grant_major; ++ } ++ ++ search_minor: ++ /* ++ * NB: minor_key list doesn't have a "head", @list is just a ++ * temporary stub for helping list searching, make sure it's removed ++ * after searching. ++ * minor_key list is an ordered list too. ++ */ ++ list_add_tail(&list, &tmp->lk_nodes[dep].ln_minor_list); ++ ++ reverse = (minor >= (1 << (mi_bits - 1))); ++ ++ if (reverse) { ++ list_for_each_entry_reverse(tmp2, &list, ++ lk_nodes[dep].ln_minor_list) { ++ if (tmp2->lk_nodes[dep].ln_minor_key == minor) { ++ goto out_enqueue; ++ ++ } else if (tmp2->lk_nodes[dep].ln_minor_key < minor) { ++ /* attach _after_ @tmp2 */ ++ list_add(&lck->lk_nodes[dep].ln_minor_list, ++ &tmp2->lk_nodes[dep].ln_minor_list); ++ goto out_grant_minor; ++ } ++ } ++ ++ list_add(&lck->lk_nodes[dep].ln_minor_list, &list); ++ ++ } else { ++ list_for_each_entry(tmp2, &list, ++ lk_nodes[dep].ln_minor_list) { ++ if (tmp2->lk_nodes[dep].ln_minor_key == minor) { ++ goto out_enqueue; ++ ++ } else if (tmp2->lk_nodes[dep].ln_minor_key > minor) { ++ /* insert _before_ @tmp2 */ ++ list_add_tail(&lck->lk_nodes[dep].ln_minor_list, ++ &tmp2->lk_nodes[dep].ln_minor_list); ++ goto out_grant_minor; ++ } ++ } ++ ++ list_add_tail(&lck->lk_nodes[dep].ln_minor_list, &list); ++ } ++ ++ out_grant_minor: ++ if (list.next == &lck->lk_nodes[dep].ln_minor_list) { ++ /* new lock @lck is the first one on minor_key list, which ++ * means it has the smallest minor_key and it should ++ * replace @tmp as minor_key owner */ ++ list_replace_init(&tmp->lk_nodes[dep].ln_major_list, ++ &lck->lk_nodes[dep].ln_major_list); ++ } ++ /* remove the temporary head */ ++ list_del(&list); ++ ++ out_grant_major: ++ ln_grant_inc(dep, lck->lk_nodes[dep].ln_mode); ++ return 1; /* granted with holding lh_lock */ ++ ++ out_enqueue: ++ list_del(&list); /* remove temprary head */ ++ return htree_node_lock_enqueue(lck, tmp2, dep, wait, event); ++} ++ ++/* ++ * release the key of @lck at level @dep, and grant any blocked locks. ++ * caller will still listen on @key if @event is not NULL, which means ++ * caller can see a event (by event_cb) while granting any lock with ++ * the same key at level @dep. ++ * NB: ALWAYS called holding lhead::lh_lock ++ * NB: listener will not block anyone because listening mode is HTREE_LOCK_NL ++ */ ++static void ++htree_node_unlock_internal(struct htree_lock_head *lhead, ++ struct htree_lock *curlk, unsigned dep, void *event) ++{ ++ struct htree_lock_node *curln = &curlk->lk_nodes[dep]; ++ struct htree_lock *grtlk = NULL; ++ struct htree_lock_node *grtln; ++ struct htree_lock *poslk; ++ struct htree_lock *tmplk; ++ ++ if (!htree_node_is_granted(curlk, dep)) ++ return; ++ ++ if (!list_empty(&curln->ln_granted_list)) { ++ /* there is another granted lock */ ++ grtlk = list_entry(curln->ln_granted_list.next, ++ struct htree_lock, ++ lk_nodes[dep].ln_granted_list); ++ list_del_init(&curln->ln_granted_list); ++ } ++ ++ if (grtlk == NULL && !list_empty(&curln->ln_blocked_list)) { ++ /* ++ * @curlk is the only granted lock, so we confirmed: ++ * a) curln is key owner (attached on major/minor_list), ++ * so if there is any blocked lock, it should be attached ++ * on curln->ln_blocked_list ++ * b) we always can grant the first blocked lock ++ */ ++ grtlk = list_entry(curln->ln_blocked_list.next, ++ struct htree_lock, ++ lk_nodes[dep].ln_blocked_list); ++ BUG_ON(grtlk->lk_task == NULL); ++ wake_up_process(grtlk->lk_task); ++ } ++ ++ if (event != NULL && ++ lhead->lh_children[dep].lc_events != HTREE_EVENT_DISABLE) { ++ curln->ln_ev_target = event; ++ curln->ln_mode = HTREE_LOCK_NL; /* listen! */ ++ } else { ++ curln->ln_mode = HTREE_LOCK_INVAL; ++ } ++ ++ if (grtlk == NULL) { /* I must be the only one locking this key */ ++ struct htree_lock_node *tmpln; ++ ++ BUG_ON(htree_key_list_empty(curln)); ++ ++ if (curln->ln_mode == HTREE_LOCK_NL) /* listening */ ++ return; ++ ++ /* not listening */ ++ if (list_empty(&curln->ln_alive_list)) { /* no more listener */ ++ htree_key_list_del_init(curln); ++ return; ++ } ++ ++ tmpln = list_entry(curln->ln_alive_list.next, ++ struct htree_lock_node, ln_alive_list); ++ ++ BUG_ON(tmpln->ln_mode != HTREE_LOCK_NL); ++ ++ htree_key_list_replace_init(curln, tmpln); ++ list_del_init(&curln->ln_alive_list); ++ ++ return; ++ } ++ ++ /* have a granted lock */ ++ grtln = &grtlk->lk_nodes[dep]; ++ if (!list_empty(&curln->ln_blocked_list)) { ++ /* only key owner can be on both lists */ ++ BUG_ON(htree_key_list_empty(curln)); ++ ++ if (list_empty(&grtln->ln_blocked_list)) { ++ list_add(&grtln->ln_blocked_list, ++ &curln->ln_blocked_list); ++ } ++ list_del_init(&curln->ln_blocked_list); ++ } ++ /* ++ * NB: this is the tricky part: ++ * We have only two modes for child-lock (PR and PW), also, ++ * only owner of the key (attached on major/minor_list) can be on ++ * both blocked_list and granted_list, so @grtlk must be one ++ * of these two cases: ++ * ++ * a) @grtlk is taken from granted_list, which means we've granted ++ * more than one lock so @grtlk has to be PR, the first blocked ++ * lock must be PW and we can't grant it at all. ++ * So even @grtlk is not owner of the key (empty blocked_list), ++ * we don't care because we can't grant any lock. ++ * b) we just grant a new lock which is taken from head of blocked ++ * list, and it should be the first granted lock, and it should ++ * be the first one linked on blocked_list. ++ * ++ * Either way, we can get correct result by iterating blocked_list ++ * of @grtlk, and don't have to bother on how to find out ++ * owner of current key. ++ */ ++ list_for_each_entry_safe(poslk, tmplk, &grtln->ln_blocked_list, ++ lk_nodes[dep].ln_blocked_list) { ++ if (grtlk->lk_nodes[dep].ln_mode == HTREE_LOCK_PW || ++ poslk->lk_nodes[dep].ln_mode == HTREE_LOCK_PW) ++ break; ++ /* grant all readers */ ++ list_del_init(&poslk->lk_nodes[dep].ln_blocked_list); ++ list_add(&poslk->lk_nodes[dep].ln_granted_list, ++ &grtln->ln_granted_list); ++ ++ BUG_ON(poslk->lk_task == NULL); ++ wake_up_process(poslk->lk_task); ++ } ++ ++ /* if @curln is the owner of this key, replace it with @grtln */ ++ if (!htree_key_list_empty(curln)) ++ htree_key_list_replace_init(curln, grtln); ++ ++ if (curln->ln_mode == HTREE_LOCK_INVAL) ++ list_del_init(&curln->ln_alive_list); ++} ++ ++/* ++ * it's just wrapper of htree_node_lock_internal, it returns 1 on granted ++ * and 0 only if @wait is false and can't grant it immediately ++ */ ++int ++htree_node_lock_try(struct htree_lock *lck, htree_lock_mode_t mode, ++ u32 key, unsigned dep, int wait, void *event) ++{ ++ struct htree_lock_head *lhead = lck->lk_head; ++ int rc; ++ ++ BUG_ON(dep >= lck->lk_depth); ++ BUG_ON(lck->lk_mode == HTREE_LOCK_INVAL); ++ ++ htree_spin_lock(lhead, dep); ++ rc = htree_node_lock_internal(lhead, lck, mode, key, dep, wait, event); ++ if (rc != 0) ++ htree_spin_unlock(lhead, dep); ++ return rc >= 0; ++} ++EXPORT_SYMBOL(htree_node_lock_try); ++ ++/* it's wrapper of htree_node_unlock_internal */ ++void ++htree_node_unlock(struct htree_lock *lck, unsigned dep, void *event) ++{ ++ struct htree_lock_head *lhead = lck->lk_head; ++ ++ BUG_ON(dep >= lck->lk_depth); ++ BUG_ON(lck->lk_mode == HTREE_LOCK_INVAL); ++ ++ htree_spin_lock(lhead, dep); ++ htree_node_unlock_internal(lhead, lck, dep, event); ++ htree_spin_unlock(lhead, dep); ++} ++EXPORT_SYMBOL(htree_node_unlock); ++ ++/* stop listening on child-lock level @dep */ ++void ++htree_node_stop_listen(struct htree_lock *lck, unsigned dep) ++{ ++ struct htree_lock_node *ln = &lck->lk_nodes[dep]; ++ struct htree_lock_node *tmp; ++ ++ BUG_ON(htree_node_is_granted(lck, dep)); ++ BUG_ON(!list_empty(&ln->ln_blocked_list)); ++ BUG_ON(!list_empty(&ln->ln_granted_list)); ++ ++ if (!htree_node_is_listening(lck, dep)) ++ return; ++ ++ htree_spin_lock(lck->lk_head, dep); ++ ln->ln_mode = HTREE_LOCK_INVAL; ++ ln->ln_ev_target = NULL; ++ ++ if (htree_key_list_empty(ln)) { /* not owner */ ++ list_del_init(&ln->ln_alive_list); ++ goto out; ++ } ++ ++ /* I'm the owner... */ ++ if (list_empty(&ln->ln_alive_list)) { /* no more listener */ ++ htree_key_list_del_init(ln); ++ goto out; ++ } ++ ++ tmp = list_entry(ln->ln_alive_list.next, ++ struct htree_lock_node, ln_alive_list); ++ ++ BUG_ON(tmp->ln_mode != HTREE_LOCK_NL); ++ htree_key_list_replace_init(ln, tmp); ++ list_del_init(&ln->ln_alive_list); ++ out: ++ htree_spin_unlock(lck->lk_head, dep); ++} ++EXPORT_SYMBOL(htree_node_stop_listen); ++ ++/* release all child-locks if we have any */ ++static void ++htree_node_release_all(struct htree_lock *lck) ++{ ++ int i; ++ ++ for (i = 0; i < lck->lk_depth; i++) { ++ if (htree_node_is_granted(lck, i)) ++ htree_node_unlock(lck, i, NULL); ++ else if (htree_node_is_listening(lck, i)) ++ htree_node_stop_listen(lck, i); ++ } ++} ++ ++/* ++ * obtain htree lock, it could be blocked inside if there's conflict ++ * with any granted or blocked lock and @wait is true. ++ * NB: ALWAYS called holding lhead::lh_lock ++ */ ++static int ++htree_lock_internal(struct htree_lock *lck, int wait) ++{ ++ struct htree_lock_head *lhead = lck->lk_head; ++ int granted = 0; ++ int blocked = 0; ++ int i; ++ ++ for (i = 0; i < HTREE_LOCK_MAX; i++) { ++ if (lhead->lh_ngranted[i] != 0) ++ granted |= 1 << i; ++ if (lhead->lh_nblocked[i] != 0) ++ blocked |= 1 << i; ++ } ++ if ((htree_lock_compat[lck->lk_mode] & granted) != granted || ++ (htree_lock_compat[lck->lk_mode] & blocked) != blocked) { ++ /* will block current lock even it just conflicts with any ++ * other blocked lock, so lock like EX wouldn't starve */ ++ if (!wait) ++ return -1; ++ lhead->lh_nblocked[lck->lk_mode]++; ++ lk_block_inc(lck->lk_mode); ++ ++ lck->lk_task = current; ++ list_add_tail(&lck->lk_blocked_list, &lhead->lh_blocked_list); ++ ++ set_current_state(TASK_UNINTERRUPTIBLE); ++ htree_spin_unlock(lhead, HTREE_DEP_ROOT); ++ /* wait to be given the lock */ ++ if (lck->lk_task != NULL) ++ schedule(); ++ /* granted, no doubt. wake up will set me RUNNING */ ++ return 0; /* without lh_lock */ ++ } ++ lhead->lh_ngranted[lck->lk_mode]++; ++ lk_grant_inc(lck->lk_mode); ++ return 1; ++} ++ ++/* release htree lock. NB: ALWAYS called holding lhead::lh_lock */ ++static void ++htree_unlock_internal(struct htree_lock *lck) ++{ ++ struct htree_lock_head *lhead = lck->lk_head; ++ struct htree_lock *tmp; ++ struct htree_lock *tmp2; ++ int granted = 0; ++ int i; ++ ++ BUG_ON(lhead->lh_ngranted[lck->lk_mode] == 0); ++ ++ lhead->lh_ngranted[lck->lk_mode]--; ++ lck->lk_mode = HTREE_LOCK_INVAL; ++ ++ for (i = 0; i < HTREE_LOCK_MAX; i++) { ++ if (lhead->lh_ngranted[i] != 0) ++ granted |= 1 << i; ++ } ++ list_for_each_entry_safe(tmp, tmp2, ++ &lhead->lh_blocked_list, lk_blocked_list) { ++ /* conflict with any granted lock? */ ++ if ((htree_lock_compat[tmp->lk_mode] & granted) != granted) ++ break; ++ ++ list_del_init(&tmp->lk_blocked_list); ++ ++ BUG_ON(lhead->lh_nblocked[tmp->lk_mode] == 0); ++ ++ lhead->lh_nblocked[tmp->lk_mode]--; ++ lhead->lh_ngranted[tmp->lk_mode]++; ++ granted |= 1 << tmp->lk_mode; ++ ++ BUG_ON(tmp->lk_task == NULL); ++ wake_up_process(tmp->lk_task); ++ } ++} ++ ++/* it's wrapper of htree_lock_internal and exported interface. ++ * It always return 1 with granted lock if @wait is true, it can return 0 ++ * if @wait is false and locking request can't be granted immediately */ ++int ++htree_lock_try(struct htree_lock *lck, struct htree_lock_head *lhead, ++ htree_lock_mode_t mode, int wait) ++{ ++ int rc; ++ ++ BUG_ON(lck->lk_depth > lhead->lh_depth); ++ BUG_ON(lck->lk_head != NULL); ++ BUG_ON(lck->lk_task != NULL); ++ ++ lck->lk_head = lhead; ++ lck->lk_mode = mode; ++ ++ htree_spin_lock(lhead, HTREE_DEP_ROOT); ++ rc = htree_lock_internal(lck, wait); ++ if (rc != 0) ++ htree_spin_unlock(lhead, HTREE_DEP_ROOT); ++ return rc >= 0; ++} ++EXPORT_SYMBOL(htree_lock_try); ++ ++/* it's wrapper of htree_unlock_internal and exported interface. ++ * It will release all htree_node_locks and htree_lock */ ++void ++htree_unlock(struct htree_lock *lck) ++{ ++ BUG_ON(lck->lk_head == NULL); ++ BUG_ON(lck->lk_mode == HTREE_LOCK_INVAL); ++ ++ htree_node_release_all(lck); ++ ++ htree_spin_lock(lck->lk_head, HTREE_DEP_ROOT); ++ htree_unlock_internal(lck); ++ htree_spin_unlock(lck->lk_head, HTREE_DEP_ROOT); ++ lck->lk_head = NULL; ++ lck->lk_task = NULL; ++} ++EXPORT_SYMBOL(htree_unlock); ++ ++/* change lock mode */ ++void ++htree_change_mode(struct htree_lock *lck, htree_lock_mode_t mode) ++{ ++ BUG_ON(lck->lk_mode == HTREE_LOCK_INVAL); ++ lck->lk_mode = mode; ++} ++EXPORT_SYMBOL(htree_change_mode); ++ ++/* release htree lock, and lock it again with new mode. ++ * This function will first release all htree_node_locks and htree_lock, ++ * then try to gain htree_lock with new @mode. ++ * It always return 1 with granted lock if @wait is true, it can return 0 ++ * if @wait is false and locking request can't be granted immediately */ ++int ++htree_change_lock_try(struct htree_lock *lck, htree_lock_mode_t mode, int wait) ++{ ++ struct htree_lock_head *lhead = lck->lk_head; ++ int rc; ++ ++ BUG_ON(lhead == NULL); ++ BUG_ON(lck->lk_mode == mode); ++ BUG_ON(lck->lk_mode == HTREE_LOCK_INVAL || mode == HTREE_LOCK_INVAL); ++ ++ htree_node_release_all(lck); ++ ++ htree_spin_lock(lhead, HTREE_DEP_ROOT); ++ htree_unlock_internal(lck); ++ lck->lk_mode = mode; ++ rc = htree_lock_internal(lck, wait); ++ if (rc != 0) ++ htree_spin_unlock(lhead, HTREE_DEP_ROOT); ++ return rc >= 0; ++} ++EXPORT_SYMBOL(htree_change_lock_try); ++ ++/* create a htree_lock head with @depth levels (number of child-locks), ++ * it is a per resoruce structure */ ++struct htree_lock_head * ++htree_lock_head_alloc(unsigned depth, unsigned hbits, unsigned priv) ++{ ++ struct htree_lock_head *lhead; ++ int i; ++ ++ if (depth > HTREE_LOCK_DEP_MAX) { ++ printk(KERN_ERR "%d is larger than max htree_lock depth %d\n", ++ depth, HTREE_LOCK_DEP_MAX); ++ return NULL; ++ } ++ ++ lhead = kzalloc(offsetof(struct htree_lock_head, ++ lh_children[depth]) + priv, GFP_NOFS); ++ if (lhead == NULL) ++ return NULL; ++ ++ if (hbits < HTREE_HBITS_MIN) ++ lhead->lh_hbits = HTREE_HBITS_MIN; ++ else if (hbits > HTREE_HBITS_MAX) ++ lhead->lh_hbits = HTREE_HBITS_MAX; ++ ++ lhead->lh_lock = 0; ++ lhead->lh_depth = depth; ++ INIT_LIST_HEAD(&lhead->lh_blocked_list); ++ if (priv > 0) { ++ lhead->lh_private = (void *)lhead + ++ offsetof(struct htree_lock_head, lh_children[depth]); ++ } ++ ++ for (i = 0; i < depth; i++) { ++ INIT_LIST_HEAD(&lhead->lh_children[i].lc_list); ++ lhead->lh_children[i].lc_events = HTREE_EVENT_DISABLE; ++ } ++ return lhead; ++} ++EXPORT_SYMBOL(htree_lock_head_alloc); ++ ++/* free the htree_lock head */ ++void ++htree_lock_head_free(struct htree_lock_head *lhead) ++{ ++ int i; ++ ++ BUG_ON(!list_empty(&lhead->lh_blocked_list)); ++ for (i = 0; i < lhead->lh_depth; i++) ++ BUG_ON(!list_empty(&lhead->lh_children[i].lc_list)); ++ kfree(lhead); ++} ++EXPORT_SYMBOL(htree_lock_head_free); ++ ++/* register event callback for @events of child-lock at level @dep */ ++void ++htree_lock_event_attach(struct htree_lock_head *lhead, unsigned dep, ++ unsigned events, htree_event_cb_t callback) ++{ ++ BUG_ON(lhead->lh_depth <= dep); ++ lhead->lh_children[dep].lc_events = events; ++ lhead->lh_children[dep].lc_callback = callback; ++} ++EXPORT_SYMBOL(htree_lock_event_attach); ++ ++/* allocate a htree_lock, which is per-thread structure, @pbytes is some ++ * extra-bytes as private data for caller */ ++struct htree_lock * ++htree_lock_alloc(unsigned depth, unsigned pbytes) ++{ ++ struct htree_lock *lck; ++ int i = offsetof(struct htree_lock, lk_nodes[depth]); ++ ++ if (depth > HTREE_LOCK_DEP_MAX) { ++ printk(KERN_ERR "%d is larger than max htree_lock depth %d\n", ++ depth, HTREE_LOCK_DEP_MAX); ++ return NULL; ++ } ++ lck = kzalloc(i + pbytes, GFP_NOFS); ++ if (lck == NULL) ++ return NULL; ++ ++ if (pbytes != 0) ++ lck->lk_private = (void *)lck + i; ++ lck->lk_mode = HTREE_LOCK_INVAL; ++ lck->lk_depth = depth; ++ INIT_LIST_HEAD(&lck->lk_blocked_list); ++ ++ for (i = 0; i < depth; i++) { ++ struct htree_lock_node *node = &lck->lk_nodes[i]; ++ ++ node->ln_mode = HTREE_LOCK_INVAL; ++ INIT_LIST_HEAD(&node->ln_major_list); ++ INIT_LIST_HEAD(&node->ln_minor_list); ++ INIT_LIST_HEAD(&node->ln_alive_list); ++ INIT_LIST_HEAD(&node->ln_blocked_list); ++ INIT_LIST_HEAD(&node->ln_granted_list); ++ } ++ ++ return lck; ++} ++EXPORT_SYMBOL(htree_lock_alloc); ++ ++/* free htree_lock node */ ++void ++htree_lock_free(struct htree_lock *lck) ++{ ++ BUG_ON(lck->lk_mode != HTREE_LOCK_INVAL); ++ kfree(lck); ++} ++EXPORT_SYMBOL(htree_lock_free); +Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/ext4.h +=================================================================== +--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/ext4.h ++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/ext4.h +@@ -27,6 +27,7 @@ + #include + #include + #include ++#include + #include + #include + #include +@@ -810,6 +811,9 @@ 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, +@@ -1536,6 +1540,7 @@ static inline void ext4_clear_state_flag + EXT4_FEATURE_INCOMPAT_META_BG| \ + EXT4_FEATURE_INCOMPAT_EXTENTS| \ + EXT4_FEATURE_INCOMPAT_64BIT| \ ++ EXT4_FEATURE_INCOMPAT_LARGEDIR|\ + EXT4_FEATURE_INCOMPAT_FLEX_BG| \ + EXT4_FEATURE_INCOMPAT_EA_INODE| \ + EXT4_FEATURE_INCOMPAT_MMP | \ +@@ -1954,6 +1959,76 @@ struct mmpd_data { + # define NORET_TYPE /**/ + # define ATTRIB_NORET __attribute__((noreturn)) + # define NORET_AND noreturn, ++/* htree levels for ext4 */ ++#define EXT4_HTREE_LEVEL_COMPAT 2 ++#define EXT4_HTREE_LEVEL 3 ++ ++static inline int ++ext4_dir_htree_level(struct super_block *sb) ++{ ++ return EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_LARGEDIR) ? ++ EXT4_HTREE_LEVEL : EXT4_HTREE_LEVEL_COMPAT; ++} ++ ++/* 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) ++ + + struct ext4_xattr_ino_array { + unsigned int xia_count; /* # of used item in the array */ +@@ -2050,9 +2125,17 @@ void ext4_insert_dentry(struct inode *in + const char *name, int namelen, 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_COMPAT_FEATURE(inode->i_sb, + EXT4_FEATURE_COMPAT_DIR_INDEX)) + ext4_clear_inode_flag(inode, EXT4_INODE_INDEX); ++#endif + } + static unsigned char ext4_filetype_table[] = { + DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK +@@ -2212,14 +2295,14 @@ extern int ext4_htree_fill_tree(struct f + extern struct inode *ext4_create_inode(handle_t *handle, + struct inode * dir, int mode); + extern int ext4_add_entry(handle_t *handle, struct dentry *dentry, +- struct inode *inode); ++ struct inode *inode, struct htree_lock *lck); + extern int ext4_delete_entry(handle_t *handle, struct inode * dir, + struct ext4_dir_entry_2 * de_del, + struct buffer_head * bh); + extern struct buffer_head * ext4_find_entry(struct inode *dir, + const struct qstr *d_name, + struct ext4_dir_entry_2 ** res_dir, +- int *inlined); ++ int *inlined, struct htree_lock *lck); + extern int ext4_add_dot_dotdot(handle_t *handle, struct inode *dir, + struct inode *inode, const void *, const void *); + extern int search_dir(struct buffer_head *bh, +@@ -2382,13 +2465,15 @@ static inline void ext4_r_blocks_count_s + es->s_r_blocks_count_hi = cpu_to_le32(blk >> 32); + } + +-static inline loff_t ext4_isize(struct ext4_inode *raw_inode) ++static inline loff_t ext4_isize(struct super_block *sb, ++ struct ext4_inode *raw_inode) + { +- if (S_ISREG(le16_to_cpu(raw_inode->i_mode))) ++ if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_LARGEDIR) || ++ S_ISREG(le16_to_cpu(raw_inode->i_mode))) + return ((loff_t)le32_to_cpu(raw_inode->i_size_high) << 32) | + le32_to_cpu(raw_inode->i_size_lo); +- else +- return (loff_t) le32_to_cpu(raw_inode->i_size_lo); ++ ++ return (loff_t) le32_to_cpu(raw_inode->i_size_lo); + } + + static inline void ext4_isize_set(struct ext4_inode *raw_inode, loff_t i_size) +Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/namei.c +=================================================================== +--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/namei.c ++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/namei.c +@@ -53,6 +53,7 @@ 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 = 0; + + if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb && +@@ -60,15 +61,22 @@ 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, 1, &err); +- if (!bh) ++ if (!bh) { ++ up(&ei->i_append_sem); + return ERR_PTR(err); ++ } + 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); +@@ -246,7 +254,7 @@ static struct dx_frame *dx_probe(const s + struct inode *dir, + struct dx_hash_info *hinfo, + struct dx_frame *frame, +- int *err); ++ struct htree_lock *lck, int *err); + static void dx_release(struct dx_frame *frames); + static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize, + struct dx_hash_info *hinfo, struct dx_map_entry map[]); +@@ -259,13 +267,13 @@ 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, + const struct qstr *d_name, + struct ext4_dir_entry_2 **res_dir, +- int *err); ++ struct htree_lock *lck, int *err); + static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry, +- struct inode *inode); ++ struct inode *inode, struct htree_lock *lck); + + /* checksumming functions */ + void initialize_dirent_tail(struct ext4_dir_entry_tail *t, +@@ -517,7 +525,7 @@ struct dx_root_info * dx_get_dx_info(str + + static inline ext4_lblk_t dx_get_block(struct dx_entry *entry) + { +- return le32_to_cpu(entry->block) & 0x00ffffff; ++ return le32_to_cpu(entry->block) & 0x0fffffff; + } + + static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value) +@@ -667,6 +675,223 @@ 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) ++ ++/* 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 (!(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. + * +@@ -678,16 +903,17 @@ struct stats dx_show_entries(struct dx_h + */ + static struct dx_frame * + dx_probe(const struct qstr *d_name, struct inode *dir, +- struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err) ++ struct dx_hash_info *hinfo, struct dx_frame *frame_in, ++ struct htree_lock *lck, int *err) + { + 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 buffer_head *bh; + struct dx_frame *frame = frame_in; + u32 hash; + +- frame->bh = NULL; ++ memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0])); + bh = ext4_read_dirblock(dir, 0, INDEX); + if (IS_ERR(bh)) { + *err = PTR_ERR(bh); +@@ -720,9 +946,16 @@ dx_probe(const struct qstr *d_name, stru + goto fail; + } + +- if ((indirect = info->indirect_levels) > 1) { +- ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x", +- 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 " ++ "supported value", dir->i_ino, ++ ext4_dir_htree_level(dir->i_sb)); ++ if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) { ++ ext4_warning(dir->i_sb, "Enable large directory " ++ "feature to access it"); ++ } + brelse(bh); + *err = ERR_BAD_DX_DIR; + goto fail; +@@ -742,8 +975,15 @@ dx_probe(const struct qstr *d_name, stru + 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(dir->i_sb, + "dx entry: no count or count > limit"); + brelse(bh); +@@ -784,7 +1024,70 @@ dx_probe(const struct qstr *d_name, stru + frame->bh = bh; + frame->entries = entries; + frame->at = at; +- if (!indirect--) return frame; ++ ++ 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--; + bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX); + if (IS_ERR(bh)) { + *err = PTR_ERR(bh); +@@ -818,13 +1121,18 @@ fail: + static void dx_release (struct dx_frame *frames) + { + struct dx_root_info *info; ++ int i; ++ + if (frames[0].bh == NULL) + return; + + info = dx_get_dx_info((struct ext4_dir_entry_2*)frames[0].bh->b_data); +- if (info->indirect_levels) +- brelse(frames[1].bh); +- brelse(frames[0].bh); ++ for (i = 0; i <= info->indirect_levels; i++) { ++ if (frames[i].bh == NULL) ++ break; ++ brelse(frames[i].bh); ++ frames[i].bh = NULL; ++ } + } + + /* +@@ -847,7 +1155,7 @@ 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; +@@ -862,12 +1170,22 @@ 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--; + } + +@@ -890,6 +1208,13 @@ 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); +@@ -898,6 +1223,7 @@ 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; + } + +@@ -966,7 +1292,7 @@ int ext4_htree_fill_tree(struct file *di + { + struct dx_hash_info hinfo; + struct ext4_dir_entry_2 *de; +- struct dx_frame frames[2], *frame; ++ struct dx_frame frames[EXT4_HTREE_LEVEL], *frame; + struct inode *dir; + ext4_lblk_t block; + int count = 0; +@@ -1000,10 +1326,10 @@ int ext4_htree_fill_tree(struct file *di + } + hinfo.hash = start_hash; + hinfo.minor_hash = 0; +- frame = dx_probe(NULL, dir, &hinfo, frames, &err); ++ /* assume it's PR locked */ ++ frame = dx_probe(NULL, dir, &hinfo, frames, NULL, &err); + if (!frame) + return err; +- + /* Add '.' and '..' from the htree header */ + if (!start_hash && !start_minor_hash) { + de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data; +@@ -1030,7 +1356,7 @@ 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; +@@ -1226,7 +1552,7 @@ static int is_dx_internal_node(struct in + struct buffer_head * ext4_find_entry(struct inode *dir, + const struct qstr *d_name, + 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]; +@@ -1270,7 +1596,7 @@ struct buffer_head * ext4_find_entry(str + goto restart; + } + if (is_dx(dir)) { +- bh = ext4_dx_find_entry(dir, d_name, res_dir, &err); ++ bh = ext4_dx_find_entry(dir, d_name, res_dir, lck, &err); + /* + * On success, or if the error was file not found, + * return. Otherwise, fall back to doing a search the +@@ -1280,6 +1606,7 @@ struct buffer_head * ext4_find_entry(str + return bh; + dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, " + "falling back\n")); ++ ext4_htree_safe_relock(lck); + } + nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb); + start = EXT4_I(dir)->i_dir_start_lookup; +@@ -1369,17 +1696,19 @@ cleanup_and_exit: + } + EXPORT_SYMBOL(ext4_find_entry); + +-static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name, +- struct ext4_dir_entry_2 **res_dir, int *err) ++static struct buffer_head * ext4_dx_find_entry(struct inode *dir, ++ const struct qstr *d_name, ++ struct ext4_dir_entry_2 **res_dir, ++ struct htree_lock *lck, int *err) + { + struct super_block * sb = dir->i_sb; + struct dx_hash_info hinfo; +- struct dx_frame frames[2], *frame; ++ struct dx_frame frames[EXT4_HTREE_LEVEL], *frame; + struct buffer_head *bh; + ext4_lblk_t block; + int retval; + +- if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err))) ++ if (!(frame = dx_probe(d_name, dir, &hinfo, frames, lck, err))) + return NULL; + do { + block = dx_get_block(frame->at); +@@ -1403,7 +1732,7 @@ static struct buffer_head * ext4_dx_find + + /* Check to see if we should continue to search */ + retval = ext4_htree_next_block(dir, hinfo.hash, frame, +- frames, NULL); ++ frames, NULL, lck); + if (retval < 0) { + ext4_warning(sb, + "error reading index page in directory #%lu", +@@ -1429,7 +1758,7 @@ static struct dentry *ext4_lookup(struct + if (dentry->d_name.len > EXT4_NAME_LEN) + return ERR_PTR(-ENAMETOOLONG); + +- bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL); ++ bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL, NULL); + if (IS_ERR(bh)) + return (struct dentry *) bh; + inode = NULL; +@@ -1489,7 +1818,7 @@ struct dentry *ext4_get_parent(struct de + struct ext4_dir_entry_2 * de; + struct buffer_head *bh; + +- bh = ext4_find_entry(child->d_inode, &dotdot, &de, NULL); ++ bh = ext4_find_entry(child->d_inode, &dotdot, &de, NULL, NULL); + if (IS_ERR(bh)) + return (struct dentry *) bh; + if (!bh) +@@ -1559,8 +1888,9 @@ 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, int *error) ++ struct buffer_head **bh, struct dx_frame *frames, ++ struct dx_frame *frame, struct dx_hash_info *hinfo, ++ struct htree_lock *lck, int *error) + { + unsigned blocksize = dir->i_sb->s_blocksize; + unsigned count, continued; +@@ -1624,7 +1954,14 @@ 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, +@@ -1643,13 +1980,21 @@ static struct ext4_dir_entry_2 *do_split + dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1)); + dxtrace(dx_show_leaf (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_dirent_node(handle, dir, bh2); + if (err) + goto journal_error; +@@ -1809,7 +2154,7 @@ static int add_dirent_to_buf(handle_t *h + */ + dir->i_mtime = dir->i_ctime = ext4_current_time(dir); + ext4_update_dx_flag(dir); +- dir->i_version++; ++ inode_inc_iversion(dir); + ext4_mark_inode_dirty(handle, dir); + BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); + err = ext4_handle_dirty_dirent_node(handle, dir, bh); +@@ -1829,7 +2174,7 @@ static int make_indexed_dir(handle_t *ha + const char *name = dentry->d_name.name; + int namelen = dentry->d_name.len; + struct buffer_head *bh2; +- struct dx_frame frames[2], *frame; ++ struct dx_frame frames[EXT4_HTREE_LEVEL], *frame; + struct dx_entry *entries; + struct ext4_dir_entry_2 *de, *de2, *dot_de, *dotdot_de; + struct ext4_dir_entry_tail *t; +@@ -1923,7 +2268,7 @@ static int make_indexed_dir(handle_t *ha + ext4_handle_dirty_dx_node(handle, dir, frame->bh); + ext4_handle_dirty_dirent_node(handle, dir, bh); + +- de = do_split(handle,dir, &bh, frame, &hinfo, &retval); ++ de = do_split(handle,dir, &bh, frames, frame, &hinfo, NULL, &retval); + if (!de) { + /* + * Even if the block split failed, we have to properly write +@@ -2030,7 +2375,7 @@ out: + * the entry, as someone else might have used it while you slept. + */ + int ext4_add_entry(handle_t *handle, struct dentry *dentry, +- struct inode *inode) ++ struct inode *inode, struct htree_lock *lck) + { + struct inode *dir = dentry->d_parent->d_inode; + struct buffer_head *bh; +@@ -2066,9 +2411,10 @@ int ext4_add_entry(handle_t *handle, str + 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, dentry, inode); ++ retval = ext4_dx_add_entry(handle, dentry, inode, lck); + if (!retval || (retval != ERR_BAD_DX_DIR)) + return retval; ++ ext4_htree_safe_relock(lck); + ext4_clear_inode_flag(dir, EXT4_INODE_INDEX); + dx_fallback++; + ext4_mark_inode_dirty(handle, dir); +@@ -2114,18 +2460,21 @@ EXPORT_SYMBOL(ext4_add_entry); + * Returns 0 for success, or a negative error value + */ + static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry, +- struct inode *inode) ++ struct inode *inode, struct htree_lock *lck) + { +- struct dx_frame frames[2], *frame; ++ struct dx_frame frames[EXT4_HTREE_LEVEL], *frame; + struct dx_entry *entries, *at; + struct dx_hash_info hinfo; + struct buffer_head *bh; + struct inode *dir = dentry->d_parent->d_inode; + struct super_block *sb = dir->i_sb; + struct ext4_dir_entry_2 *de; ++ int restart; + int err; + +- frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err); ++again: ++ restart = 0; ++ frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, lck, &err); + if (!frame) + return err; + entries = frame->entries; +@@ -2137,33 +2486,53 @@ static int ext4_dx_add_entry(handle_t *h + goto cleanup; + } + +- BUFFER_TRACE(bh, "get_write_access"); +- err = ext4_journal_get_write_access(handle, bh); +- if (err) +- goto journal_error; +- + err = add_dirent_to_buf(handle, dentry, inode, NULL, bh); + if (err != -ENOSPC) + goto cleanup; + ++ err = 0; + /* Block full, should compress but for now just split */ + dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n", + dx_get_count(entries), dx_get_limit(entries))); + /* Need to split index? */ + if (dx_get_count(entries) == dx_get_limit(entries)) { + ext4_lblk_t newblock; +- unsigned icount = dx_get_count(entries); +- int levels = frame - frames; ++ int levels = frame - frames + 1; ++ unsigned icount; ++ int add_level = 1; + struct dx_entry *entries2; + struct dx_node *node2; + struct buffer_head *bh2; + +- if (levels && (dx_get_count(frames->entries) == +- dx_get_limit(frames->entries))) { +- ext4_warning(sb, "Directory index full!"); ++ 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)) { ++ add_level = 0; ++ break; ++ } ++ frame--; /* split higher index block */ ++ at = frame->at; ++ entries = frame->entries; ++ restart = 1; ++ } ++ if (add_level && levels == ext4_dir_htree_level(sb)) { ++ ext4_warning(sb, "Directory (ino: %lu) index full, " ++ "reach max htree level :%d", ++ dir->i_ino, levels); ++ if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) { ++ ext4_warning(sb, "Large directory feature is" ++ "not enabled on this " ++ "filesystem"); ++ } + err = -ENOSPC; + goto cleanup; + } ++ icount = dx_get_count(entries); + bh2 = ext4_append(handle, dir, &newblock); + if (IS_ERR(bh2)) { + err = PTR_ERR(bh2); +@@ -2178,7 +2547,7 @@ static int ext4_dx_add_entry(handle_t *h + err = ext4_journal_get_write_access(handle, frame->bh); + if (err) + goto journal_error; +- if (levels) { ++ if (!add_level) { + unsigned icount1 = icount/2, icount2 = icount - icount1; + unsigned hash2 = dx_get_hash(entries + icount1); + dxtrace(printk(KERN_DEBUG "Split index %i/%i\n", +@@ -2186,7 +2555,7 @@ static int ext4_dx_add_entry(handle_t *h + + BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */ + err = ext4_journal_get_write_access(handle, +- frames[0].bh); ++ (frame - 1)->bh); + if (err) + goto journal_error; + +@@ -2202,18 +2571,24 @@ static int ext4_dx_add_entry(handle_t *h + frame->entries = entries = entries2; + swap(frame->bh, bh2); + } +- dx_insert_block(frames + 0, hash2, newblock); +- dxtrace(dx_show_index("node", frames[1].entries)); ++ dx_insert_block((frame - 1), hash2, newblock); ++ dxtrace(dx_show_index("node", frame->entries)); + dxtrace(dx_show_index("node", + ((struct dx_node *) bh2->b_data)->entries)); + err = ext4_handle_dirty_dx_node(handle, dir, bh2); + if (err) + goto journal_error; + brelse (bh2); ++ ext4_handle_dirty_metadata(handle, inode, ++ (frame - 1)->bh); ++ if (restart) { ++ ext4_handle_dirty_metadata(handle, inode, ++ frame->bh); ++ goto cleanup; ++ } + } else { + struct dx_root_info * info; +- dxtrace(printk(KERN_DEBUG +- "Creating second level index...\n")); ++ + memcpy((char *) entries2, (char *) entries, + icount * sizeof(struct dx_entry)); + dx_set_limit(entries2, dx_node_limit(dir)); +@@ -2223,35 +2598,63 @@ static int ext4_dx_add_entry(handle_t *h + dx_set_block(entries + 0, newblock); + info = dx_get_dx_info((struct ext4_dir_entry_2*) + frames[0].bh->b_data); +- info->indirect_levels = 1; ++ info->indirect_levels += 1; ++ dxtrace(printk(KERN_DEBUG ++ "Creating %d level index...\n", ++ info->indirect_levels)); ++ ext4_handle_dirty_metadata(handle, inode, frame->bh); ++ ext4_handle_dirty_metadata(handle, inode, bh2); ++ brelse(bh2); ++ restart = 1; ++ goto cleanup; ++ } ++ } else if (!ext4_htree_dx_locked(lck)) { ++ struct ext4_dir_lock_data *ld = ext4_htree_lock_data(lck); + +- /* Add new access path frame */ +- frame = frames + 1; +- frame->at = at = at - entries + entries2; +- frame->entries = entries = entries2; +- frame->bh = bh2; +- err = ext4_journal_get_write_access(handle, +- frame->bh); +- if (err) +- goto journal_error; ++ /* 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; + } +- err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh); ++ /* OK, I've got DX lock and nothing changed */ ++ frame->at = ld->ld_at; + if (err) { + ext4_std_error(inode->i_sb, err); + goto cleanup; + } + } +- de = do_split(handle, dir, &bh, frame, &hinfo, &err); ++ de = do_split(handle, dir, &bh, frames, frame, &hinfo, lck, &err); + if (!de) + goto cleanup; ++ + err = add_dirent_to_buf(handle, dentry, inode, de, bh); + goto cleanup; + + journal_error: + ext4_std_error(dir->i_sb, err); + 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 ++ * repeat dx_probe() to find out valid htree-path */ ++ if (restart && err == 0) ++ goto again; + return err; + } + +@@ -2288,7 +2691,7 @@ int ext4_generic_delete_entry(handle_t * + blocksize); + else + de->inode = 0; +- dir->i_version++; ++ inode_inc_iversion(dir); + return 0; + } + i += ext4_rec_len_from_disk(de->rec_len, blocksize); +@@ -2373,7 +2776,7 @@ EXPORT_SYMBOL(ext4_dec_count); + static int ext4_add_nondir(handle_t *handle, + struct dentry *dentry, struct inode *inode) + { +- int err = ext4_add_entry(handle, dentry, inode); ++ int err = ext4_add_entry(handle, dentry, inode, NULL); + if (!err) { + ext4_mark_inode_dirty(handle, inode); + unlock_new_inode(inode); +@@ -2641,7 +3044,7 @@ retry: + goto out_clear_inode; + err = ext4_mark_inode_dirty(handle, inode); + if (!err) +- err = ext4_add_entry(handle, dentry, inode); ++ err = ext4_add_entry(handle, dentry, inode, NULL); + if (err) { + out_clear_inode: + clear_nlink(inode); +@@ -2907,7 +3310,7 @@ static int ext4_rmdir(struct inode *dir, + dquot_initialize(dentry->d_inode); + + retval = -ENOENT; +- bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL); ++ bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL, NULL); + if (IS_ERR(bh)) + return PTR_ERR(bh); + if (!bh) +@@ -2974,7 +3377,7 @@ static int ext4_unlink(struct inode *dir + dquot_initialize(dentry->d_inode); + + retval = -ENOENT; +- bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL); ++ bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL, NULL); + if (IS_ERR(bh)) + return PTR_ERR(bh); + if (!bh) +@@ -3153,7 +3556,7 @@ retry: + ext4_inc_count(handle, inode); + ihold(inode); + +- err = ext4_add_entry(handle, dentry, inode); ++ err = ext4_add_entry(handle, dentry, inode, NULL); + if (!err) { + ext4_mark_inode_dirty(handle, inode); + d_instantiate(dentry, inode); +@@ -3183,7 +3556,7 @@ retry: + struct buffer_head *bh; + struct ext4_dir_entry_2 *de; + +- bh = ext4_find_entry(dir, d_name, &de, NULL); ++ bh = ext4_find_entry(dir, d_name, &de, NULL, NULL); + if (IS_ERR(bh)) + return PTR_ERR(bh); + if (bh) { +@@ -3230,7 +3633,7 @@ static int ext4_rename(struct inode *old + if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir)) + ext4_handle_sync(handle); + +- old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL); ++ old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL, NULL); + if (IS_ERR(old.bh)) + return PTR_ERR(old.bh); + /* +@@ -3244,7 +3647,7 @@ static int ext4_rename(struct inode *old + + new_inode = new_dentry->d_inode; + new.bh = ext4_find_entry(new.dir, &new.dentry->d_name, +- &new.de, &new.inlined); ++ &new.de, &new.inlined, NULL); + if (IS_ERR(new.bh)) { + if (!new_inode) { + brelse(new_bh); +@@ -3275,7 +3678,7 @@ static int ext4_rename(struct inode *old + goto end_rename; + } + if (!new.bh) { +- retval = ext4_add_entry(handle, new.dentry, old.inode); ++ retval = ext4_add_entry(handle, new.dentry, old.inode, NULL); + if (retval) + goto end_rename; + } else { +@@ -3375,7 +3678,7 @@ static int ext4_rename(struct inode *old + dquot_initialize(new.dir); + + old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, +- &old.de, &old.inlined); ++ &old.de, &old.inlined, NULL); + /* + * Check for inode number is _not_ due to possible IO errors. + * We might rmdir the source, keep it as pwd of some process +@@ -3475,7 +3678,7 @@ static int ext4_rename(struct inode *old + goto end_rename; + + new.bh = ext4_find_entry(new.dir, &new.dentry->d_name, +- &new.de, &new.inlined); ++ &new.de, &new.inlined, NULL); + + /* RENAME_EXCHANGE case: old *and* new must both exist */ + if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino) +Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/inode.c +=================================================================== +--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/inode.c ++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/inode.c +@@ -4264,7 +4264,7 @@ struct inode *ext4_iget(struct super_blo + if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) + ei->i_file_acl |= + ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; +- inode->i_size = ext4_isize(raw_inode); ++ inode->i_size = ext4_isize(sb, raw_inode); + ei->i_disksize = inode->i_size; + #ifdef CONFIG_QUOTA + ei->i_reserved_quota = 0; +@@ -4499,7 +4499,7 @@ static int ext4_do_update_inode(handle_t + raw_inode->i_file_acl_high = + cpu_to_le16(ei->i_file_acl >> 32); + raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); +- if (ei->i_disksize != ext4_isize(raw_inode)) { ++ if (ei->i_disksize != ext4_isize(inode->i_sb, raw_inode)) { + ext4_isize_set(raw_inode, ei->i_disksize); + need_datasync = 1; + } +Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/Makefile +=================================================================== +--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/Makefile ++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/Makefile +@@ -8,7 +8,7 @@ ext4-y := balloc.o bitmap.o dir.o file.o + ioctl.o namei.o super.o symlink.o hash.o resize.o extents.o \ + ext4_jbd2.o migrate.o mballoc.o block_validity.o move_extent.o \ + mmp.o indirect.o extents_status.o xattr.o xattr_user.o \ +- xattr_trusted.o inline.o ++ xattr_trusted.o inline.o htree_lock.o + + ext4-$(CONFIG_EXT4_FS_POSIX_ACL) += acl.o + ext4-$(CONFIG_EXT4_FS_SECURITY) += xattr_security.o +Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/super.c +=================================================================== +--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/super.c ++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/super.c +@@ -872,6 +872,7 @@ static struct inode *ext4_alloc_inode(st + + ei->vfs_inode.i_version = 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); diff --git a/ldiskfs/kernel_patches/patches/rhel7/ext4-prealloc.patch b/ldiskfs/kernel_patches/patches/rhel7/ext4-prealloc.patch new file mode 100644 index 0000000..63f8837 --- /dev/null +++ b/ldiskfs/kernel_patches/patches/rhel7/ext4-prealloc.patch @@ -0,0 +1,388 @@ +Index: linux-3.10.0-123.el7.x86_64/fs/ext4/ext4.h +=================================================================== +--- linux-3.10.0-123.el7.x86_64.orig/fs/ext4/ext4.h ++++ linux-3.10.0-123.el7.x86_64/fs/ext4/ext4.h +@@ -1243,11 +1243,14 @@ struct ext4_sb_info { + + /* tunables */ + unsigned long s_stripe; +- unsigned int s_mb_stream_request; ++ unsigned long s_mb_small_req; ++ unsigned long s_mb_large_req; + unsigned int s_mb_max_to_scan; + unsigned int s_mb_min_to_scan; + unsigned int s_mb_stats; + unsigned int s_mb_order2_reqs; ++ unsigned long *s_mb_prealloc_table; ++ unsigned long s_mb_prealloc_table_size; + unsigned int s_mb_group_prealloc; + unsigned int s_max_writeback_mb_bump; + unsigned int s_max_dir_size_kb; +Index: linux-3.10.0-123.el7.x86_64/fs/ext4/mballoc.c +=================================================================== +--- linux-3.10.0-123.el7.x86_64.orig/fs/ext4/mballoc.c ++++ linux-3.10.0-123.el7.x86_64/fs/ext4/mballoc.c +@@ -1828,6 +1828,25 @@ int ext4_mb_find_by_goal(struct ext4_all + return 0; + } + ++static void ext4_mb_prealloc_table_add(struct ext4_sb_info *sbi, int value) ++{ ++ int i; ++ ++ if (value > (sbi->s_blocks_per_group - 1 - 1 - sbi->s_itb_per_group)) ++ return; ++ ++ for (i = 0; i < sbi->s_mb_prealloc_table_size; i++) { ++ if (sbi->s_mb_prealloc_table[i] == 0) { ++ sbi->s_mb_prealloc_table[i] = value; ++ return; ++ } ++ ++ /* they should add values in order */ ++ if (value <= sbi->s_mb_prealloc_table[i]) ++ return; ++ } ++} ++ + /* + * The routine scans buddy structures (not bitmap!) from given order + * to max order and tries to find big enough chunk to satisfy the req +@@ -2263,6 +2282,86 @@ static const struct seq_operations ext4_ + .show = ext4_mb_seq_groups_show, + }; + ++#define EXT4_MB_PREALLOC_TABLE "prealloc_table" ++ ++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))); ++ unsigned long value; ++ unsigned long prev = 0; ++ char str[128]; ++ char *cur; ++ char *end; ++ unsigned long *new_table; ++ int num = 0; ++ int i = 0; ++ ++ if (cnt >= sizeof(str)) ++ return -EINVAL; ++ if (copy_from_user(str, buf, cnt)) ++ return -EFAULT; ++ ++ num = 0; ++ cur = str; ++ end = str + cnt; ++ while (cur < end) { ++ while ((cur < end) && (*cur == ' ')) cur++; ++ value = simple_strtol(cur, &cur, 0); ++ if (value == 0) ++ break; ++ if (value <= prev) ++ return -EINVAL; ++ prev = value; ++ num++; ++ } ++ ++ new_table = kmalloc(num * sizeof(*new_table), GFP_KERNEL); ++ if (new_table == NULL) ++ return -ENOMEM; ++ kfree(sbi->s_mb_prealloc_table); ++ memset(new_table, 0, num * sizeof(*new_table)); ++ sbi->s_mb_prealloc_table = new_table; ++ sbi->s_mb_prealloc_table_size = num; ++ cur = str; ++ end = str + cnt; ++ while (cur < end && i < num) { ++ while ((cur < end) && (*cur == ' ')) cur++; ++ value = simple_strtol(cur, &cur, 0); ++ ext4_mb_prealloc_table_add(sbi, value); ++ i++; ++ } ++ ++ return 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 < sbi->s_mb_prealloc_table_size; 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)); ++} ++ ++struct file_operations ext4_mb_prealloc_seq_fops = { ++ .owner = THIS_MODULE, ++ .open = mb_prealloc_table_seq_open, ++ .read = seq_read, ++ .llseek = seq_lseek, ++ .release = single_release, ++ .write = ext4_mb_prealloc_table_proc_write, ++}; ++ + static int ext4_mb_seq_groups_open(struct inode *inode, struct file *file) + { + struct super_block *sb = PDE_DATA(inode); +@@ -2557,7 +2656,6 @@ 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 +@@ -2581,9 +2679,48 @@ 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); ++ ++ if (sbi->s_stripe == 0) { ++ sbi->s_mb_prealloc_table_size = 10; ++ i = sbi->s_mb_prealloc_table_size * sizeof(unsigned long); ++ sbi->s_mb_prealloc_table = kmalloc(i, GFP_NOFS); ++ if (sbi->s_mb_prealloc_table == NULL) { ++ ret = -ENOMEM; ++ goto out; ++ } ++ memset(sbi->s_mb_prealloc_table, 0, i); ++ ++ ext4_mb_prealloc_table_add(sbi, 4); ++ ext4_mb_prealloc_table_add(sbi, 8); ++ ext4_mb_prealloc_table_add(sbi, 16); ++ ext4_mb_prealloc_table_add(sbi, 32); ++ ext4_mb_prealloc_table_add(sbi, 64); ++ ext4_mb_prealloc_table_add(sbi, 128); ++ ext4_mb_prealloc_table_add(sbi, 256); ++ ext4_mb_prealloc_table_add(sbi, 512); ++ ext4_mb_prealloc_table_add(sbi, 1024); ++ ext4_mb_prealloc_table_add(sbi, 2048); ++ ++ sbi->s_mb_small_req = 256; ++ sbi->s_mb_large_req = 1024; ++ sbi->s_mb_group_prealloc = 512; ++ } else { ++ sbi->s_mb_prealloc_table_size = 3; ++ i = sbi->s_mb_prealloc_table_size * sizeof(unsigned long); ++ sbi->s_mb_prealloc_table = kmalloc(i, GFP_NOFS); ++ if (sbi->s_mb_prealloc_table == NULL) { ++ ret = -ENOMEM; ++ goto out; ++ } ++ memset(sbi->s_mb_prealloc_table, 0, i); ++ ++ ext4_mb_prealloc_table_add(sbi, sbi->s_stripe); ++ ext4_mb_prealloc_table_add(sbi, sbi->s_stripe * 2); ++ ext4_mb_prealloc_table_add(sbi, sbi->s_stripe * 4); ++ ++ 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); +@@ -2605,9 +2742,13 @@ int ext4_mb_init(struct super_block *sb) + if (ret != 0) + goto out_free_locality_groups; + +- if (sbi->s_proc) ++ if (sbi->s_proc) { + proc_create_data("mb_groups", S_IRUGO, sbi->s_proc, + &ext4_mb_seq_groups_fops, sb); ++ proc_create_data(EXT4_MB_PREALLOC_TABLE, S_IFREG | S_IRUGO | ++ S_IWUSR, sbi->s_proc, ++ &ext4_mb_prealloc_seq_fops, sb); ++ } + + return 0; + +@@ -2615,6 +2756,7 @@ 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); + + +@@ -2651,8 +2793,10 @@ int ext4_mb_release(struct super_block * + struct ext4_sb_info *sbi = EXT4_SB(sb); + struct kmem_cache *cachep = get_groupinfo_cache(sb->s_blocksize_bits); + +- if (sbi->s_proc) ++ if (sbi->s_proc) { + remove_proc_entry("mb_groups", sbi->s_proc); ++ remove_proc_entry(EXT4_MB_PREALLOC_TABLE, sbi->s_proc); ++ } + + if (sbi->s_group_info) { + for (i = 0; i < ngroups; i++) { +@@ -2963,9 +3107,9 @@ 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); +@@ -2998,51 +3142,34 @@ 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; + +- /* max size of free chunks */ +- max = 2 << bsbits; ++ start = wind = 0; + +-#define NRL_CHECK_SIZE(req, size, max, chunk_size) \ +- (req <= (size) || max <= (chunk_size)) ++ /* let's choose preallocation window depending on file size */ ++ for (i = 0; i < sbi->s_mb_prealloc_table_size; i++) { ++ if (size <= sbi->s_mb_prealloc_table[i]) { ++ wind = sbi->s_mb_prealloc_table[i]; ++ break; ++ } ++ } ++ size = wind; + +- /* 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; +- } 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; ++ if (wind == 0) { ++ __u64 tstart, tend; ++ /* file is quite large, we now preallocate with ++ * the biggest configured window with regart to ++ * logical offset */ ++ wind = sbi->s_mb_prealloc_table[i - 1]; ++ 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; + } +- 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) { +@@ -3117,7 +3245,6 @@ ext4_mb_normalize_request(struct ext4_al + BUG_ON(start + size <= ac->ac_o_ex.fe_logical && + start > ac->ac_o_ex.fe_logical); + } +- BUG_ON(size <= 0 || size > EXT4_BLOCKS_PER_GROUP(ac->ac_sb)); + + /* now prepare goal request */ + +@@ -4056,11 +4183,19 @@ 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 +Index: linux-3.10.0-123.el7.x86_64/fs/ext4/super.c +=================================================================== +--- linux-3.10.0-123.el7.x86_64.orig/fs/ext4/super.c ++++ linux-3.10.0-123.el7.x86_64/fs/ext4/super.c +@@ -2555,7 +2555,8 @@ 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(max_writeback_mb_bump, s_max_writeback_mb_bump); + EXT4_RW_ATTR_SBI_UI(extent_max_zeroout_kb, s_extent_max_zeroout_kb); +@@ -2578,7 +2579,8 @@ 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), +Index: linux-3.10.0-123.el7.x86_64/fs/ext4/inode.c +=================================================================== +--- linux-3.10.0-123.el7.x86_64.orig/fs/ext4/inode.c ++++ linux-3.10.0-123.el7.x86_64/fs/ext4/inode.c +@@ -2476,6 +2476,10 @@ static int ext4_da_writepages(struct add + if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) + return -EROFS; + ++ if (wbc->nr_to_write < sbi->s_mb_small_req) { ++ wbc->nr_to_write = sbi->s_mb_small_req; ++ } ++ + if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) + range_whole = 1; + diff --git a/ldiskfs/kernel_patches/patches/rhel7/ext4-remove-truncate-warning.patch b/ldiskfs/kernel_patches/patches/rhel7/ext4-remove-truncate-warning.patch new file mode 100644 index 0000000..999d86a --- /dev/null +++ b/ldiskfs/kernel_patches/patches/rhel7/ext4-remove-truncate-warning.patch @@ -0,0 +1,17 @@ +There will cause a deadlock if invoke ext4_truncate with i_mutex locked +in lustre. Since lustre has own lock to provide protect so we don't +need this check at all. + +Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/inode.c +=================================================================== +--- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/inode.c ++++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/inode.c +@@ -3847,8 +3847,6 @@ void ext4_truncate(struct inode *inode) + * or it completely new indode. In those cases we might not + * have i_mutex locked because it's not necessary. + */ +- if (!(inode->i_state & (I_NEW|I_FREEING))) +- WARN_ON(!mutex_is_locked(&inode->i_mutex)); + trace_ext4_truncate_enter(inode); + + if (!ext4_can_truncate(inode)) diff --git a/ldiskfs/kernel_patches/series/ldiskfs-3.10-rhel7.series b/ldiskfs/kernel_patches/series/ldiskfs-3.10-rhel7.series new file mode 100644 index 0000000..8437fdb --- /dev/null +++ b/ldiskfs/kernel_patches/series/ldiskfs-3.10-rhel7.series @@ -0,0 +1,17 @@ +rhel7/ext4-inode-version.patch +rhel7/ext4-lookup-dotdot.patch +rhel6.3/ext4-print-inum-in-htree-warning.patch +rhel7/ext4-prealloc.patch +rhel7/ext4-mballoc-extra-checks.patch +rhel7/ext4-misc.patch +rhel7/ext4-osd-iop-common.patch +rhel7/ext4-hash-indexed-dir-dotdot-update.patch +rhel7/ext4-kill-dx-root.patch +rhel7/ext4-mballoc-pa-free-mismatch.patch +rhel7/ext4-data-in-dirent.patch +rhel7/ext4-large-eas.patch +rhel7/ext4-disable-mb-cache.patch +rhel7/ext4-nocmtime.patch +rhel7/ext4-pdirop.patch +rhel7/ext4-max-dir-size.patch +rhel7/ext4-remove-truncate-warning.patch diff --git a/lustre/osd-ldiskfs/osd_handler.c b/lustre/osd-ldiskfs/osd_handler.c index b2cf6e3..369914a 100644 --- a/lustre/osd-ldiskfs/osd_handler.c +++ b/lustre/osd-ldiskfs/osd_handler.c @@ -54,6 +54,9 @@ /* XATTR_{REPLACE,CREATE} */ #include +#include +#include +#undef ENTRY /* * struct OBD_{ALLOC,FREE}*() * OBD_FAIL_CHECK @@ -72,7 +75,6 @@ #include #include -#include #include int ldiskfs_pdo = 1; diff --git a/lustre/osd-ldiskfs/osd_iam.c b/lustre/osd-ldiskfs/osd_iam.c index 8a1fdf6..0251e5b 100644 --- a/lustre/osd-ldiskfs/osd_iam.c +++ b/lustre/osd-ldiskfs/osd_iam.c @@ -147,10 +147,14 @@ #include #include #include + +#include +#include +#undef ENTRY + #include "osd_internal.h" -#include "xattr.h" -#include "acl.h" +#include /* * List of all registered formats.