Sign Up
Log In
Log In
or
Sign Up
Places
All Projects
Status Monitor
Collapse sidebar
filesystems
lustre_2_15
0018-0005-LU-16354-ldiskfs-RHEL9.1-server-suppo...
Overview
Repositories
Revisions
Requests
Users
Attributes
Meta
File 0018-0005-LU-16354-ldiskfs-RHEL9.1-server-support.patch of Package lustre_2_15
From 391392e11743efcb0de815af8bb5a563e4b3cdc6 Mon Sep 17 00:00:00 2001 From: Shaun Tancheff <shaun.tancheff@hpe.com> Date: Sat, 21 Jan 2023 00:16:25 -0600 Subject: [PATCH] LU-16354 ldiskfs: RHEL9.1 server support ldiskfs patch series for RHEL9.1 Test-Parameters: trivial HPE-bug-id: LUS-11332 Signed-off-by: Shaun Tancheff <shaun.tancheff@hpe.com> Change-Id: Ia0757995ac7200eb50fadf5e106fe1d7b3dc0443 Reviewed-on: https://review.whamcloud.com/c/fs/lustre-release/+/49283 Tested-by: jenkins <devops@whamcloud.com> Tested-by: Maloo <maloo@whamcloud.com> Reviewed-by: Jian Yu <yujian@whamcloud.com> Reviewed-by: Oleg Drokin <green@whamcloud.com> Reviewed-by: Yang Sheng <ys@whamcloud.com> --- config/lustre-build-ldiskfs.m4 | 1 ldiskfs/kernel_patches/patches/linux-5.18/ext4-data-in-dirent.patch | 874 +++++++++ ldiskfs/kernel_patches/patches/linux-5.18/ext4-export-mb-stream-allocator-variables.patch | 111 + ldiskfs/kernel_patches/patches/linux-5.18/ext4-mballoc-extra-checks.patch | 316 +++ ldiskfs/kernel_patches/patches/linux-5.18/ext4-mballoc-pa-free-mismatch.patch | 110 + ldiskfs/kernel_patches/patches/linux-5.18/ext4-prealloc.patch | 403 ++++ ldiskfs/kernel_patches/patches/rhel9.1/ext4-enc-flag.patch | 70 ldiskfs/kernel_patches/patches/rhel9.1/ext4-pdirop.patch | 930 ++++++++++ ldiskfs/kernel_patches/series/ldiskfs-5.14-rhel9.1.series | 33 9 files changed, 2848 insertions(+) create mode 100644 ldiskfs/kernel_patches/patches/linux-5.18/ext4-data-in-dirent.patch create mode 100644 ldiskfs/kernel_patches/patches/linux-5.18/ext4-export-mb-stream-allocator-variables.patch create mode 100644 ldiskfs/kernel_patches/patches/linux-5.18/ext4-mballoc-extra-checks.patch create mode 100644 ldiskfs/kernel_patches/patches/linux-5.18/ext4-mballoc-pa-free-mismatch.patch create mode 100644 ldiskfs/kernel_patches/patches/linux-5.18/ext4-prealloc.patch create mode 100644 ldiskfs/kernel_patches/patches/rhel9.1/ext4-enc-flag.patch create mode 100644 ldiskfs/kernel_patches/patches/rhel9.1/ext4-pdirop.patch create mode 100644 ldiskfs/kernel_patches/series/ldiskfs-5.14-rhel9.1.series --- a/config/lustre-build-ldiskfs.m4 +++ b/config/lustre-build-ldiskfs.m4 @@ -13,6 +13,7 @@ esac AS_IF([test -z "$LDISKFS_SERIES"], [ AS_IF([test x$RHEL_KERNEL = xyes], [ case $RHEL_RELEASE_NO in + 91) LDISKFS_SERIES="5.14-rhel9.1.series" ;; 88) LDISKFS_SERIES="4.18-rhel8.8.series" ;; 87) LDISKFS_SERIES="4.18-rhel8.7.series" ;; 86) LDISKFS_SERIES="4.18-rhel8.6.series" ;; --- /dev/null +++ b/ldiskfs/kernel_patches/patches/linux-5.18/ext4-data-in-dirent.patch @@ -0,0 +1,874 @@ +Subject: [PATCH] linux-5.18/ext4-data-in-dirent + +--- + fs/ext4/dir.c | 9 +- + fs/ext4/ext4.h | 107 ++++++++++++++++-- + fs/ext4/fast_commit.c | 2 +- + fs/ext4/inline.c | 8 +- + fs/ext4/namei.c | 250 ++++++++++++++++++++++++++++++++---------- + fs/ext4/super.c | 4 +- + 6 files changed, 305 insertions(+), 75 deletions(-) + +diff --git a/fs/ext4/dir.c b/fs/ext4/dir.c +index a6bb86f..edd0540 100644 +--- a/fs/ext4/dir.c ++++ b/fs/ext4/dir.c +@@ -465,12 +465,17 @@ int ext4_htree_store_dirent(struct file *dir_file, __u32 hash, + struct fname *fname, *new_fn; + struct dir_private_info *info; + int len; ++ int extra_data = 0; + + info = dir_file->private_data; + p = &info->root.rb_node; + + /* Create and allocate the fname structure */ +- len = sizeof(struct fname) + ent_name->len + 1; ++ if (dirent->file_type & EXT4_DIRENT_LUFID) ++ extra_data = ext4_get_dirent_data_len(dirent); ++ ++ len = sizeof(struct fname) + ent_name->len + extra_data + 1; ++ + new_fn = kzalloc(len, GFP_KERNEL); + if (!new_fn) + return -ENOMEM; +@@ -479,7 +484,7 @@ int ext4_htree_store_dirent(struct file *dir_file, __u32 hash, + new_fn->inode = le32_to_cpu(dirent->inode); + new_fn->name_len = ent_name->len; + new_fn->file_type = dirent->file_type; +- memcpy(new_fn->name, ent_name->name, ent_name->len); ++ memcpy(new_fn->name, ent_name->name, ent_name->len + extra_data); + + while (*p) { + parent = *p; +diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h +index f4a1557..85556ce 100644 +--- a/fs/ext4/ext4.h ++++ b/fs/ext4/ext4.h +@@ -1173,6 +1173,7 @@ struct ext4_inode_info { + __u32 i_csum_seed; + + kprojid_t i_projid; ++ void *i_dirdata; + }; + + /* +@@ -1194,6 +1195,7 @@ struct ext4_inode_info { + * Mount flags set via mount options or defaults + */ + #define EXT4_MOUNT_NO_MBCACHE 0x00001 /* Do not use mbcache */ ++#define EXT4_MOUNT_DIRDATA 0x00002 /* Data in directory entries */ + #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 */ +@@ -2162,6 +2164,7 @@ EXT4_FEATURE_INCOMPAT_FUNCS(casefold, CASEFOLD) + EXT4_FEATURE_INCOMPAT_FLEX_BG| \ + EXT4_FEATURE_INCOMPAT_EA_INODE| \ + EXT4_FEATURE_INCOMPAT_MMP | \ ++ EXT4_FEATURE_INCOMPAT_DIRDATA| \ + EXT4_FEATURE_INCOMPAT_INLINE_DATA | \ + EXT4_FEATURE_INCOMPAT_ENCRYPT | \ + EXT4_FEATURE_INCOMPAT_CASEFOLD | \ +@@ -2373,6 +2376,42 @@ struct ext4_dir_entry_tail { + #define EXT4_FT_SYMLINK 7 + + #define EXT4_FT_MAX 8 ++#define EXT4_FT_MASK 0xf ++ ++#if EXT4_FT_MAX > EXT4_FT_MASK ++#error "conflicting EXT4_FT_MAX and EXT4_FT_MASK" ++#endif ++ ++/* ++ * d_type has 4 unused bits, so it can hold four types data. these different ++ * type of data (e.g. lustre data, high 32 bits of 64-bit inode number) can be ++ * stored, in flag order, after file-name in ext4 dirent. ++*/ ++/* ++ * this flag is added to d_type if ext4 dirent has extra data after ++ * filename. this data length is variable and length is stored in first byte ++ * of data. data start after filename NUL byte. ++ * This is used by Lustre FS. ++ */ ++#define EXT4_DIRENT_LUFID 0x10 ++ ++#define EXT4_LUFID_MAGIC 0xAD200907UL ++struct ext4_dentry_param { ++ __u32 edp_magic; /* EXT4_LUFID_MAGIC */ ++ char edp_len; /* size of edp_data in bytes */ ++ char edp_data[0]; /* packed array of data */ ++} __packed; ++ ++static inline unsigned char *ext4_dentry_get_data(struct super_block *sb, ++ struct ext4_dentry_param *p) ++{ ++ if (!ext4_has_feature_dirdata(sb)) ++ return NULL; ++ if (p && p->edp_magic == EXT4_LUFID_MAGIC) ++ return &p->edp_len; ++ else ++ return NULL; ++} + + #define EXT4_FT_DIR_CSUM 0xDE + +@@ -2384,6 +2423,17 @@ struct ext4_dir_entry_tail { + #define EXT4_DIR_PAD 4 + #define EXT4_DIR_ROUND (EXT4_DIR_PAD - 1) + #define EXT4_MAX_REC_LEN ((1<<16)-1) ++#define EXT4_DIR_REC_LEN_(name_len, i_dir) \ ++ ext4_dir_rec_len((name_len), (i_dir)) ++#define EXT4_DIR_ENTRY_LEN_(de, i_dir) \ ++ (EXT4_DIR_REC_LEN_((de)->name_len + ext4_get_dirent_data_len(de), \ ++ (i_dir))) ++/* ldiskfs */ ++#define EXT4_DIR_REC_LEN(name_len, i_dir) EXT4_DIR_REC_LEN_((name_len), (i_dir)) ++#define EXT4_DIR_ENTRY_LEN(de, i_dir) EXT4_DIR_ENTRY_LEN_((de), (i_dir)) ++/* lustre osd_handler compat -- ifdef LDISKFS_DIR_REC_LEN_WITH_DIR */ ++#define EXT4_DIR_REC_LEN_WITH_DIR 1 ++#define __EXT4_DIR_REC_LEN(name_len) EXT4_DIR_REC_LEN_((name_len), NULL) + + /* + * The rec_len is dependent on the type of directory. Directories that are +@@ -2391,10 +2441,10 @@ struct ext4_dir_entry_tail { + * ext4_extended_dir_entry_2. For all entries related to '.' or '..' you should + * pass NULL for dir, as those entries do not use the extra fields. + */ +-static inline unsigned int ext4_dir_rec_len(__u8 name_len, ++static inline unsigned int ext4_dir_rec_len(__u32 name_len, + const struct inode *dir) + { +- int rec_len = (name_len + 8 + EXT4_DIR_ROUND); ++ __u32 rec_len = (name_len + 8 + EXT4_DIR_ROUND); + + if (dir && ext4_hash_in_dirent(dir)) + rec_len += sizeof(struct ext4_dir_entry_hash); +@@ -2867,11 +2917,13 @@ extern int ext4_find_dest_de(struct inode *dir, struct inode *inode, + struct buffer_head *bh, + void *buf, int buf_size, + struct ext4_filename *fname, +- struct ext4_dir_entry_2 **dest_de); ++ struct ext4_dir_entry_2 **dest_de, ++ int *dlen); + void ext4_insert_dentry(struct inode *dir, struct inode *inode, + struct ext4_dir_entry_2 *de, + int buf_size, +- struct ext4_filename *fname); ++ struct ext4_filename *fname, ++ void *data); + static inline void ext4_update_dx_flag(struct inode *inode) + { + if (!ext4_has_feature_dir_index(inode->i_sb) && +@@ -2887,10 +2939,17 @@ static const unsigned char ext4_filetype_table[] = { + + static inline unsigned char get_dtype(struct super_block *sb, int filetype) + { +- if (!ext4_has_feature_filetype(sb) || filetype >= EXT4_FT_MAX) ++ int fl_index = filetype & EXT4_FT_MASK; ++ ++ if (!ext4_has_feature_filetype(sb) || fl_index >= EXT4_FT_MAX) + return DT_UNKNOWN; + +- return ext4_filetype_table[filetype]; ++ if (!test_opt(sb, DIRDATA)) ++ return ext4_filetype_table[fl_index]; ++ ++ return (ext4_filetype_table[fl_index]) | ++ (filetype & EXT4_DIRENT_LUFID); ++ + } + extern int ext4_check_all_de(struct inode *dir, struct buffer_head *bh, + void *buf, int buf_size); +@@ -3093,9 +3152,13 @@ extern int ext4_ind_migrate(struct inode *inode); + + /* namei.c */ + extern int ext4_init_new_dir(handle_t *handle, struct inode *dir, +- struct inode *inode); ++ struct inode *inode, ++ const void *data1, const void *data2); + extern int ext4_dirblock_csum_verify(struct inode *inode, + struct buffer_head *bh); ++extern int ext4_add_dot_dotdot(handle_t *handle, struct inode *dir, ++ struct inode *inode, ++ const void *data1, const void *data2); + 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, +@@ -3924,6 +3987,36 @@ static inline int ext4_buffer_uptodate(struct buffer_head *bh) + return buffer_uptodate(bh); + } + ++/* ++ * Compute the total directory entry data length. ++ * This includes the filename and an implicit NUL terminator (always present), ++ * and optional extensions. Each extension has a bit set in the high 4 bits of ++ * de->file_type, and the extension length is the first byte in each entry. ++ */ ++static inline int ext4_get_dirent_data_len(struct ext4_dir_entry_2 *de) ++{ ++ char *len = de->name + de->name_len + 1 /* NUL terminator */; ++ int dlen = 0; ++ __u8 extra_data_flags = (de->file_type & ~EXT4_FT_MASK) >> 4; ++ struct ext4_dir_entry_tail *t = (struct ext4_dir_entry_tail *)de; ++ ++ if (!t->det_reserved_zero1 && ++ le16_to_cpu(t->det_rec_len) == ++ sizeof(struct ext4_dir_entry_tail) && ++ !t->det_reserved_zero2 && ++ t->det_reserved_ft == EXT4_FT_DIR_CSUM) ++ return 0; ++ ++ while (extra_data_flags) { ++ if (extra_data_flags & 1) { ++ dlen += *len + (dlen == 0); ++ len += *len; ++ } ++ extra_data_flags >>= 1; ++ } ++ return dlen; ++} ++ + #endif /* __KERNEL__ */ + + #define EFSBADCRC EBADMSG /* Bad CRC detected */ +diff --git a/fs/ext4/fast_commit.c b/fs/ext4/fast_commit.c +index 3d72565..2331515 100644 +--- a/fs/ext4/fast_commit.c ++++ b/fs/ext4/fast_commit.c +@@ -1646,7 +1646,7 @@ static int ext4_fc_replay_create(struct super_block *sb, struct ext4_fc_tl *tl, + jbd_debug(1, "Dir %d not found.", darg.ino); + goto out; + } +- ret = ext4_init_new_dir(NULL, dir, inode); ++ ret = ext4_init_new_dir(NULL, dir, inode, NULL, NULL); + iput(dir); + if (ret) { + ret = 0; +diff --git a/fs/ext4/inline.c b/fs/ext4/inline.c +index e9ef5cf..a23fd25 100644 +--- a/fs/ext4/inline.c ++++ b/fs/ext4/inline.c +@@ -1031,7 +1031,7 @@ static int ext4_add_dirent_to_inline(handle_t *handle, + struct ext4_dir_entry_2 *de; + + err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start, +- inline_size, fname, &de); ++ inline_size, fname, &de, NULL); + if (err) + return err; + +@@ -1040,7 +1040,7 @@ static int ext4_add_dirent_to_inline(handle_t *handle, + EXT4_JTR_NONE); + if (err) + return err; +- ext4_insert_dentry(dir, inode, de, inline_size, fname); ++ ext4_insert_dentry(dir, inode, de, inline_size, fname, NULL); + + ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size); + +@@ -1398,7 +1398,7 @@ int ext4_inlinedir_to_tree(struct file *dir_file, + fake.name_len = 1; + strcpy(fake.name, "."); + fake.rec_len = ext4_rec_len_to_disk( +- ext4_dir_rec_len(fake.name_len, NULL), ++ EXT4_DIR_ENTRY_LEN(&fake, NULL), + inline_size); + ext4_set_de_type(inode->i_sb, &fake, S_IFDIR); + de = &fake; +@@ -1408,7 +1408,7 @@ int ext4_inlinedir_to_tree(struct file *dir_file, + fake.name_len = 2; + strcpy(fake.name, ".."); + fake.rec_len = ext4_rec_len_to_disk( +- ext4_dir_rec_len(fake.name_len, NULL), ++ EXT4_DIR_ENTRY_LEN(&fake, NULL), + inline_size); + ext4_set_de_type(inode->i_sb, &fake, S_IFDIR); + de = &fake; +diff --git a/fs/ext4/namei.c b/fs/ext4/namei.c +index 4d932a2..ee1a058 100644 +--- a/fs/ext4/namei.c ++++ b/fs/ext4/namei.c +@@ -262,13 +262,14 @@ static unsigned dx_get_count(struct dx_entry *entries); + static unsigned dx_get_limit(struct dx_entry *entries); + static void dx_set_count(struct dx_entry *entries, unsigned value); + static void dx_set_limit(struct dx_entry *entries, unsigned value); +-static unsigned dx_root_limit(struct inode *dir, unsigned infosize); ++static inline unsigned dx_root_limit(struct inode *dir, ++ struct ext4_dir_entry_2 *dot_de, unsigned infosize); + static unsigned dx_node_limit(struct inode *dir); + static struct dx_frame *dx_probe(struct ext4_filename *fname, + struct inode *dir, + struct dx_hash_info *hinfo, + struct dx_frame *frame); +-static void dx_release(struct dx_frame *frames); ++static void dx_release(struct dx_frame *frames, struct inode *dir); + static int dx_make_map(struct inode *dir, struct buffer_head *bh, + struct dx_hash_info *hinfo, + struct dx_map_entry *map_tail); +@@ -408,22 +409,23 @@ static struct dx_countlimit *get_dx_countlimit(struct inode *inode, + { + struct ext4_dir_entry *dp; + struct dx_root_info *root; +- int count_offset; ++ int count_offset, dot_rec_len, dotdot_rec_len; + + if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb)) + count_offset = 8; +- else if (le16_to_cpu(dirent->rec_len) == 12) { +- dp = (struct ext4_dir_entry *)(((void *)dirent) + 12); ++ else { ++ dot_rec_len = le16_to_cpu(dirent->rec_len); ++ dp = (struct ext4_dir_entry *)(((void *)dirent) + dot_rec_len); + if (le16_to_cpu(dp->rec_len) != +- EXT4_BLOCK_SIZE(inode->i_sb) - 12) ++ EXT4_BLOCK_SIZE(inode->i_sb) - dot_rec_len) + return NULL; +- root = (struct dx_root_info *)(((void *)dp + 12)); ++ dotdot_rec_len = EXT4_DIR_ENTRY_LEN((struct ext4_dir_entry_2 *)dp, NULL); ++ root = (struct dx_root_info *)(((void *)dp + dotdot_rec_len)); + if (root->reserved_zero || + root->info_length != sizeof(struct dx_root_info)) + return NULL; +- count_offset = 32; +- } else +- return NULL; ++ count_offset = 8 + dot_rec_len + dotdot_rec_len; ++ } + + if (offset) + *offset = count_offset; +@@ -526,13 +528,14 @@ ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize) + * 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) ++struct dx_root_info *dx_get_dx_info(struct ext4_dir_entry_2 *de, struct inode *i_dir) + { ++ BUG_ON(de->name_len != 1); + /* get dotdot first */ +- de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(1)); ++ de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_ENTRY_LEN(de, i_dir)); + + /* dx root info is after dotdot entry */ +- de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(2)); ++ de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_ENTRY_LEN(de, i_dir)); + + return (struct dx_root_info *)de; + } +@@ -577,11 +580,16 @@ static inline void dx_set_limit(struct dx_entry *entries, unsigned value) + ((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 int entry_space = dir->i_sb->s_blocksize - +- ext4_dir_rec_len(1, NULL) - +- ext4_dir_rec_len(2, NULL) - infosize; ++ struct ext4_dir_entry_2 *dotdot_de; ++ unsigned entry_space; ++ ++ BUG_ON(dot_de->name_len != 1); ++ dotdot_de = ext4_next_entry(dot_de, dir->i_sb->s_blocksize); ++ entry_space = dir->i_sb->s_blocksize - EXT4_DIR_ENTRY_LEN(dot_de, NULL) - ++ EXT4_DIR_ENTRY_LEN(dotdot_de, NULL) - infosize; + + if (ext4_has_metadata_csum(dir->i_sb)) + entry_space -= sizeof(struct dx_tail); +@@ -699,7 +707,7 @@ static struct stats dx_show_leaf(struct inode *dir, + (unsigned) ((char *) de - base)); + #endif + } +- space += ext4_dir_rec_len(de->name_len, dir); ++ space += EXT4_DIR_ENTRY_LEN(de, dir); + names++; + } + de = ext4_next_entry(de, size); +@@ -793,7 +801,7 @@ dx_probe(struct ext4_filename *fname, struct inode *dir, + if (IS_ERR(frame->bh)) + return (struct dx_frame *) frame->bh; + +- info = dx_get_dx_info((struct ext4_dir_entry_2 *)frame->bh->b_data); ++ info = dx_get_dx_info((struct ext4_dir_entry_2 *)frame->bh->b_data, dir); + if (info->hash_version != DX_HASH_TEA && + info->hash_version != DX_HASH_HALF_MD4 && + info->hash_version != DX_HASH_LEGACY && +@@ -849,11 +857,14 @@ dx_probe(struct ext4_filename *fname, struct inode *dir, + + entries = (struct dx_entry *)(((char *)info) + info->info_length); + +- if (dx_get_limit(entries) != dx_root_limit(dir, +- info->info_length)) { ++ if (dx_get_limit(entries) != ++ dx_root_limit(dir, (struct ext4_dir_entry_2 *)frame->bh->b_data, ++ info->info_length)) { + ext4_warning_inode(dir, "dx entry: limit %u != root limit %u", + dx_get_limit(entries), +- dx_root_limit(dir, info->info_length)); ++ dx_root_limit(dir, ++ (struct ext4_dir_entry_2 *)frame->bh->b_data, ++ info->info_length)); + goto fail; + } + +@@ -930,7 +941,7 @@ fail: + return ret_err; + } + +-static void dx_release(struct dx_frame *frames) ++static void dx_release(struct dx_frame *frames, struct inode *dir) + { + struct dx_root_info *info; + int i; +@@ -939,7 +950,7 @@ static void dx_release(struct dx_frame *frames) + if (frames[0].bh == NULL) + return; + +- info = dx_get_dx_info((struct ext4_dir_entry_2 *)frames[0].bh->b_data); ++ info = dx_get_dx_info((struct ext4_dir_entry_2 *)frames[0].bh->b_data, dir); + /* save local copy, "info" may be freed after brelse() */ + indirect_levels = info->indirect_levels; + for (i = 0; i <= indirect_levels; i++) { +@@ -1240,12 +1251,12 @@ int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash, + (count && ((hashval & 1) == 0))) + break; + } +- dx_release(frames); ++ dx_release(frames, dir); + dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, " + "next hash: %x\n", count, *next_hash)); + return count; + errout: +- dx_release(frames); ++ dx_release(frames, dir); + return (err); + } + +@@ -1778,7 +1789,7 @@ static struct buffer_head * ext4_dx_find_entry(struct inode *dir, + errout: + dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name)); + success: +- dx_release(frames); ++ dx_release(frames, dir); + return bh; + } + +@@ -1902,7 +1913,7 @@ dx_move_dirents(struct inode *dir, char *from, char *to, + 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, dir); ++ rec_len = EXT4_DIR_ENTRY_LEN(de, dir); + + memcpy (to, de, rec_len); + ((struct ext4_dir_entry_2 *) to)->rec_len = +@@ -1935,7 +1946,7 @@ static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base, + 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, dir); ++ rec_len = EXT4_DIR_ENTRY_LEN(de, dir); + if (de > to) + memmove(to, de, rec_len); + to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize); +@@ -2078,14 +2089,21 @@ int ext4_find_dest_de(struct inode *dir, struct inode *inode, + struct buffer_head *bh, + void *buf, int buf_size, + struct ext4_filename *fname, +- struct ext4_dir_entry_2 **dest_de) ++ struct ext4_dir_entry_2 **dest_de, ++ int *dlen) + { + struct ext4_dir_entry_2 *de; +- unsigned short reclen = ext4_dir_rec_len(fname_len(fname), dir); ++ unsigned short reclen; + int nlen, rlen; + unsigned int offset = 0; + char *top; + ++ if (dlen) { ++ reclen = ext4_dir_rec_len(fname_len(fname) + *dlen, dir); ++ *dlen = 0; ++ } else { ++ reclen = ext4_dir_rec_len(fname_len(fname), dir); ++ } + de = (struct ext4_dir_entry_2 *)buf; + top = buf + buf_size - reclen; + while ((char *) de <= top) { +@@ -2094,10 +2112,31 @@ int ext4_find_dest_de(struct inode *dir, struct inode *inode, + return -EFSCORRUPTED; + if (ext4_match(dir, fname, de)) + return -EEXIST; +- nlen = ext4_dir_rec_len(de->name_len, dir); ++ nlen = EXT4_DIR_ENTRY_LEN(de, dir); + rlen = ext4_rec_len_from_disk(de->rec_len, buf_size); + if ((de->inode ? rlen - nlen : rlen) >= reclen) + break; ++ ++ /* Then for dotdot entries, check for the smaller space ++ * required for just the entry, no FID ++ */ ++ if (fname_len(fname) == 2 && memcmp(fname_name(fname), "..", 2) == 0) { ++ if ((de->inode ? rlen - nlen : rlen) >= ++ ext4_dir_rec_len(fname_len(fname), dir)) { ++ /* set dlen = 1 to indicate not ++ * enough space store fid ++ */ ++ if (dlen) ++ *dlen = 1; ++ 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; + } +@@ -2112,12 +2151,13 @@ void ext4_insert_dentry(struct inode *dir, + struct inode *inode, + struct ext4_dir_entry_2 *de, + int buf_size, +- struct ext4_filename *fname) ++ struct ext4_filename *fname, ++ void *data) + { + + int nlen, rlen; + +- nlen = ext4_dir_rec_len(de->name_len, dir); ++ nlen = EXT4_DIR_ENTRY_LEN(de, dir); + rlen = ext4_rec_len_from_disk(de->rec_len, buf_size); + if (de->inode) { + struct ext4_dir_entry_2 *de1 = +@@ -2138,6 +2178,12 @@ void ext4_insert_dentry(struct inode *dir, + EXT4_DIRENT_HASHES(de)->minor_hash = + cpu_to_le32(hinfo->minor_hash); + } ++ if (data) { ++ de->name[fname_len(fname)] = 0; ++ memcpy(&de->name[fname_len(fname) + 1], data, *(char *)data); ++ de->file_type |= EXT4_DIRENT_LUFID; ++ } ++ + } + + /* +@@ -2155,14 +2201,19 @@ static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname, + { + unsigned int blocksize = dir->i_sb->s_blocksize; + int csum_size = 0; +- int err, err2; ++ int err, err2, dlen = 0; ++ unsigned char *data; + ++ data = ext4_dentry_get_data(inode->i_sb, (struct ext4_dentry_param *) ++ EXT4_I(inode)->i_dirdata); + if (ext4_has_metadata_csum(inode->i_sb)) + csum_size = sizeof(struct ext4_dir_entry_tail); + + if (!de) { ++ if (data) ++ dlen = (*data) + 1; + err = ext4_find_dest_de(dir, inode, bh, bh->b_data, +- blocksize - csum_size, fname, &de); ++ blocksize - csum_size, fname, &de, &dlen); + if (err) + return err; + } +@@ -2175,7 +2226,10 @@ static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname, + } + + /* By now the buffer is marked for journaling */ +- ext4_insert_dentry(dir, inode, de, blocksize, fname); ++ /* If writing the short form of "dotdot", don't add the data section */ ++ if (dlen == 1) ++ data = NULL; ++ ext4_insert_dentry(dir, inode, de, blocksize, fname, data); + + /* + * XXX shouldn't update any times until successful +@@ -2273,7 +2327,7 @@ static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname, + blocksize); + + /* initialize hashing info */ +- dx_info = dx_get_dx_info(dot_de); ++ dx_info = dx_get_dx_info(dot_de, dir); + memset(dx_info, 0, sizeof(*dx_info)); + dx_info->info_length = sizeof(*dx_info); + if (ext4_hash_in_dirent(dir)) +@@ -2284,7 +2338,8 @@ static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname, + 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(*dx_info))); ++ dx_set_limit(entries, dx_root_limit(dir, ++ dot_de, sizeof(*dx_info))); + + /* Initialize as for dx_probe */ + fname->hinfo.hash_version = dx_info->hash_version; +@@ -2325,7 +2380,7 @@ out_frames: + */ + if (retval) + ext4_mark_inode_dirty(handle, dir); +- dx_release(frames); ++ dx_release(frames, dir); + brelse(bh2); + return retval; + } +@@ -2338,6 +2393,8 @@ static int ext4_update_dotdot(handle_t *handle, struct dentry *dentry, + 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); +@@ -2353,21 +2410,26 @@ static int ext4_update_dotdot(handle_t *handle, struct dentry *dentry, + + de = (struct ext4_dir_entry_2 *)dir_block->b_data; + /* the first item must be "." */ +- assert(de->name_len == 1 && de->name[0] == '.'); ++ 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, dir)); ++ if (len > EXT4_DIR_REC_LEN(1, dir)) { + BUFFER_TRACE(dir_block, "get_write_access"); + err = ext4_journal_get_write_access(handle, dir->i_sb, dir_block, EXT4_JTR_NONE); + 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_ENTRY_LEN(de, dir)); + } + +- len -= EXT4_DIR_REC_LEN(1); +- assert(len == 0 || len >= EXT4_DIR_REC_LEN(2)); ++ len -= EXT4_DIR_ENTRY_LEN(de, NULL); ++ 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, dir)); ++ + de = (struct ext4_dir_entry_2 *) + ((char *) de + le16_to_cpu(de->rec_len)); + if (!journal) { +@@ -2381,10 +2443,15 @@ static int ext4_update_dotdot(handle_t *handle, struct dentry *dentry, + 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, dir)); + 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) { +@@ -2422,6 +2489,7 @@ static int ext4_add_entry(handle_t *handle, struct dentry *dentry, + ext4_lblk_t block, blocks; + int csum_size = 0; + ++ EXT4_I(inode)->i_dirdata = dentry->d_fsdata; + if (ext4_has_metadata_csum(inode->i_sb)) + csum_size = sizeof(struct ext4_dir_entry_tail); + +@@ -2664,7 +2732,7 @@ again: + dx_set_count(entries, 1); + dx_set_block(entries + 0, newblock); + info = dx_get_dx_info((struct ext4_dir_entry_2 *) +- frames[0].bh->b_data); ++ frames[0].bh->b_data, dir); + info->indirect_levels = 1; + dxtrace(printk(KERN_DEBUG + "Creating %d level index...\n", +@@ -2690,7 +2758,7 @@ journal_error: + ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */ + cleanup: + brelse(bh); +- dx_release(frames); ++ dx_release(frames, dir); + /* @restart is true means htree-path has been changed, we need to + * repeat dx_probe() to find out valid htree-path + */ +@@ -2993,38 +3061,73 @@ err_unlock_inode: + return err; + } + ++struct tp_block { ++ struct inode *inode; ++ void *data1; ++ void *data2; ++}; ++ + struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode, + struct ext4_dir_entry_2 *de, + int blocksize, int csum_size, + unsigned int parent_ino, int dotdot_real_len) + { ++ void *data1 = NULL, *data2 = NULL; ++ int dot_reclen = 0; ++ ++ if (dotdot_real_len == 10) { ++ struct tp_block *tpb = (struct tp_block *)inode; ++ data1 = tpb->data1; ++ data2 = tpb->data2; ++ inode = tpb->inode; ++ dotdot_real_len = 0; ++ } + de->inode = cpu_to_le32(inode->i_ino); + de->name_len = 1; +- de->rec_len = ext4_rec_len_to_disk(ext4_dir_rec_len(de->name_len, NULL), +- blocksize); + strcpy(de->name, "."); + ext4_set_de_type(inode->i_sb, de, S_IFDIR); + ++ /* get packed fid data*/ ++ data1 = ext4_dentry_get_data(inode->i_sb, ++ (struct ext4_dentry_param *) data1); ++ if (data1) { ++ de->name[1] = 0; ++ memcpy(&de->name[2], data1, *(char *) data1); ++ de->file_type |= EXT4_DIRENT_LUFID; ++ } ++ de->rec_len = cpu_to_le16(EXT4_DIR_ENTRY_LEN(de, NULL)); ++ ++ 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, NULL)), +- blocksize); ++ (csum_size + dot_reclen), blocksize); + else + de->rec_len = ext4_rec_len_to_disk( +- ext4_dir_rec_len(de->name_len, NULL), ++ EXT4_DIR_ENTRY_LEN(de, NULL), + blocksize); +- strcpy(de->name, ".."); +- ext4_set_de_type(inode->i_sb, de, S_IFDIR); + + return ext4_next_entry(de, blocksize); + } + + int ext4_init_new_dir(handle_t *handle, struct inode *dir, +- struct inode *inode) ++ struct inode *inode, ++ const void *data1, const void *data2) + { ++ struct tp_block param; + struct buffer_head *dir_block = NULL; + struct ext4_dir_entry_2 *de; + ext4_lblk_t block = 0; +@@ -3048,7 +3151,11 @@ int ext4_init_new_dir(handle_t *handle, struct inode *dir, + if (IS_ERR(dir_block)) + return PTR_ERR(dir_block); + de = (struct ext4_dir_entry_2 *)dir_block->b_data; +- ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0); ++ param.inode = inode; ++ param.data1 = (void *)data1; ++ param.data2 = (void *)data2; ++ ext4_init_dot_dotdot((struct inode *)(¶m), de, blocksize, ++ csum_size, dir->i_ino, 10); + set_nlink(inode, 2); + if (csum_size) + ext4_initialize_dirent_tail(dir_block, blocksize); +@@ -3063,6 +3170,29 @@ out: + return err; + } + ++/* Initialize @inode as a subdirectory of @dir, and add the ++ * "." and ".." entries into the first directory block. */ ++int ext4_add_dot_dotdot(handle_t *handle, struct inode *dir, ++ struct inode *inode, ++ const void *data1, const void *data2) ++{ ++ int rc; ++ ++ if (IS_ERR(handle)) ++ return PTR_ERR(handle); ++ ++ if (IS_DIRSYNC(dir)) ++ ext4_handle_sync(handle); ++ ++ inode->i_op = &ext4_dir_inode_operations; ++ inode->i_fop = &ext4_dir_operations; ++ rc = ext4_init_new_dir(handle, dir, inode, data1, data2); ++ if (!rc) ++ rc = ext4_mark_inode_dirty(handle, inode); ++ return rc; ++} ++EXPORT_SYMBOL(ext4_add_dot_dotdot); ++ + static int ext4_mkdir(struct user_namespace *mnt_userns, struct inode *dir, + struct dentry *dentry, umode_t mode) + { +@@ -3090,7 +3220,7 @@ 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); +diff --git a/fs/ext4/super.c b/fs/ext4/super.c +index 9a0af5a..917cca7 100644 +--- a/fs/ext4/super.c ++++ b/fs/ext4/super.c +@@ -1703,7 +1703,7 @@ enum { + Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption, + Opt_inlinecrypt, + Opt_usrjquota, Opt_grpjquota, Opt_quota, +- Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err, ++ Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err, Opt_dirdata, + Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version, + Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never, + Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error, +@@ -1819,6 +1819,7 @@ static const struct fs_parameter_spec ext4_param_specs[] = { + fsparam_u32 ("stripe", Opt_stripe), + fsparam_flag ("delalloc", Opt_delalloc), + fsparam_flag ("nodelalloc", Opt_nodelalloc), ++ fsparam_flag ("dirdata", Opt_dirdata), + fsparam_flag ("warn_on_error", Opt_warn_on_error), + fsparam_flag ("nowarn_on_error", Opt_nowarn_on_error), + fsparam_u32 ("debug_want_extra_isize", +@@ -1956,6 +1957,7 @@ static const struct mount_opts { + MOPT_CLEAR | MOPT_Q}, + {Opt_usrjquota, 0, MOPT_Q}, + {Opt_grpjquota, 0, MOPT_Q}, ++ {Opt_dirdata, EXT4_MOUNT_DIRDATA, MOPT_SET}, + {Opt_jqfmt, 0, MOPT_QFMT}, + {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET}, + {Opt_no_prefetch_block_bitmaps, EXT4_MOUNT_NO_PREFETCH_BLOCK_BITMAPS, +-- +2.34.1 + --- /dev/null +++ b/ldiskfs/kernel_patches/patches/linux-5.18/ext4-export-mb-stream-allocator-variables.patch @@ -0,0 +1,111 @@ +From 13afd00ca4268f5cc306bcf42ed4d7511bab4e67 Mon Sep 17 00:00:00 2001 +From: Shaun Tancheff <shaun.tancheff@hpe.com> +Date: Mon, 14 Nov 2022 01:26:25 -0600 +Subject: [PATCH] linux-5.18/ext4-export-mb-stream-allocator-variables + +--- + fs/ext4/ext4.h | 2 ++ + fs/ext4/mballoc.c | 58 +++++++++++++++++++++++++++++++++++++++++++++++ + fs/ext4/sysfs.c | 4 ++++ + 3 files changed, 64 insertions(+) + +diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h +index 378280b..558e824 100644 +--- a/fs/ext4/ext4.h ++++ b/fs/ext4/ext4.h +@@ -3106,6 +3106,8 @@ int ext4_fc_record_regions(struct super_block *sb, int ino, + + /* mballoc.c */ + extern const struct proc_ops ext4_seq_prealloc_table_fops; ++extern const struct proc_ops ext4_seq_mb_last_group_fops; ++extern int ext4_mb_seq_last_start_seq_show(struct seq_file *m, void *v); + extern const struct seq_operations ext4_mb_seq_groups_ops; + extern const struct seq_operations ext4_mb_seq_structs_summary_ops; + extern long ext4_mb_stats; +diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c +index c54dce8..9473536 100644 +--- a/fs/ext4/mballoc.c ++++ b/fs/ext4/mballoc.c +@@ -3189,6 +3189,64 @@ static struct kmem_cache *get_groupinfo_cache(int blocksize_bits) + return cachep; + } + ++#define EXT4_MB_MAX_INPUT_STRING_SIZE 32 ++ ++static ssize_t ext4_mb_last_group_write(struct file *file, ++ const char __user *buf, ++ size_t cnt, loff_t *pos) ++{ ++ char dummy[EXT4_MB_MAX_INPUT_STRING_SIZE + 1]; ++ struct super_block *sb = pde_data(file_inode(file)); ++ struct ext4_sb_info *sbi = EXT4_SB(sb); ++ unsigned long val; ++ char *end; ++ ++ if (cnt > EXT4_MB_MAX_INPUT_STRING_SIZE) ++ return -EINVAL; ++ if (copy_from_user(dummy, buf, cnt)) ++ return -EFAULT; ++ dummy[cnt] = '\0'; ++ val = simple_strtoul(dummy, &end, 0); ++ if (dummy == end) ++ return -EINVAL; ++ if (val >= ext4_get_groups_count(sb)) ++ return -ERANGE; ++ spin_lock(&sbi->s_md_lock); ++ sbi->s_mb_last_group = val; ++ sbi->s_mb_last_start = 0; ++ spin_unlock(&sbi->s_md_lock); ++ return cnt; ++} ++ ++static int ext4_mb_seq_last_group_seq_show(struct seq_file *m, void *v) ++{ ++ struct ext4_sb_info *sbi = EXT4_SB(m->private); ++ ++ seq_printf(m , "%ld\n", sbi->s_mb_last_group); ++ return 0; ++} ++ ++static int ext4_mb_seq_last_group_open(struct inode *inode, struct file *file) ++{ ++ return single_open(file, ext4_mb_seq_last_group_seq_show, pde_data(inode)); ++} ++ ++const struct proc_ops ext4_seq_mb_last_group_fops = { ++ .proc_open = ext4_mb_seq_last_group_open, ++ .proc_read = seq_read, ++ .proc_lseek = seq_lseek, ++ .proc_release = seq_release, ++ .proc_write = ext4_mb_last_group_write, ++}; ++ ++int ext4_mb_seq_last_start_seq_show(struct seq_file *m, void *v) ++{ ++ struct ext4_sb_info *sbi = EXT4_SB(m->private); ++ ++ seq_printf(m , "%ld\n", sbi->s_mb_last_start); ++ return 0; ++} ++ + /* + * Allocate the top-level s_group_info array for the specified number + * of groups +diff --git a/fs/ext4/sysfs.c b/fs/ext4/sysfs.c +index 459eb23..38bab55 100644 +--- a/fs/ext4/sysfs.c ++++ b/fs/ext4/sysfs.c +@@ -557,6 +557,10 @@ int ext4_register_sysfs(struct super_block *sb) + &ext4_mb_seq_groups_ops, sb); + proc_create_data("prealloc_table", S_IRUGO, sbi->s_proc, + &ext4_seq_prealloc_table_fops, sb); ++ proc_create_data("mb_last_group", S_IRUGO, sbi->s_proc, ++ &ext4_seq_mb_last_group_fops, sb); ++ proc_create_single_data("mb_last_start", S_IRUGO, sbi->s_proc, ++ ext4_mb_seq_last_start_seq_show, sb); + proc_create_single_data("mb_stats", 0444, sbi->s_proc, + ext4_seq_mb_stats_show, sb); + proc_create_seq_data("mb_structs_summary", 0444, sbi->s_proc, +-- +2.34.1 + --- /dev/null +++ b/ldiskfs/kernel_patches/patches/linux-5.18/ext4-mballoc-extra-checks.patch @@ -0,0 +1,316 @@ +From 26c43c25b39cab8071fa45c595844320979efd16 Mon Sep 17 00:00:00 2001 +From: Shaun Tancheff <shaun.tancheff@hpe.com> +Date: Sun, 13 Nov 2022 04:31:54 -0600 +Subject: [PATCH] P: linux-5.14/ext4-mballoc-extra-checks.patch + +--- + fs/ext4/ext4.h | 1 + + fs/ext4/mballoc.c | 103 ++++++++++++++++++++++++++++++++++++++++------ + fs/ext4/mballoc.h | 2 +- + 3 files changed, 93 insertions(+), 13 deletions(-) + +diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h +index d7ae1b1..f4a1557 100644 +--- a/fs/ext4/ext4.h ++++ b/fs/ext4/ext4.h +@@ -3480,6 +3480,7 @@ struct ext4_group_info { + ext4_grpblk_t bb_largest_free_order;/* order of largest frag in BG */ + ext4_group_t bb_group; /* Group number */ + struct list_head bb_prealloc_list; ++ unsigned long bb_prealloc_nr; + #ifdef DOUBLE_CHECK + void *bb_bitmap; + #endif +diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c +index 9952908..5dbd7e7 100644 +--- a/fs/ext4/mballoc.c ++++ b/fs/ext4/mballoc.c +@@ -399,7 +399,7 @@ static const char * const ext4_groupinfo_slab_names[NR_GRPINFO_CACHES] = { + "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); +@@ -1105,7 +1105,7 @@ mb_set_largest_free_order(struct super_block *sb, struct ext4_group_info *grp) + } + + 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); +@@ -1149,6 +1149,7 @@ void ext4_mb_generate_buddy(struct super_block *sb, + grp->bb_free = free; + ext4_mark_group_bitmap_corrupted(sb, group, + EXT4_GROUP_INFO_BBITMAP_CORRUPT); ++ return -EIO; + } + mb_set_largest_free_order(sb, grp); + +@@ -1158,6 +1159,8 @@ void ext4_mb_generate_buddy(struct super_block *sb, + atomic_inc(&sbi->s_mb_buddies_generated); + atomic64_add(period, &sbi->s_mb_generation_time); + mb_update_avg_fragment_size(sb, grp); ++ ++ return 0; + } + + /* The buddy information is attached the buddy cache inode +@@ -1260,7 +1263,7 @@ static int ext4_mb_init_cache(struct page *page, char *incore, gfp_t gfp) + } + + 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; +@@ -1304,7 +1307,7 @@ static int ext4_mb_init_cache(struct page *page, char *incore, gfp_t gfp) + 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 { +@@ -1319,7 +1322,7 @@ static int ext4_mb_init_cache(struct page *page, char *incore, gfp_t gfp) + 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); + +@@ -1329,7 +1332,8 @@ static int ext4_mb_init_cache(struct page *page, char *incore, gfp_t gfp) + incore = data; + } + } +- SetPageUptodate(page); ++ if (likely(err == 0)) ++ SetPageUptodate(page); + + out: + if (bh) { +@@ -2858,9 +2862,11 @@ static void *ext4_mb_seq_groups_next(struct seq_file *seq, void *v, loff_t *pos) + static int ext4_mb_seq_groups_show(struct seq_file *seq, void *v) + { + struct super_block *sb = pde_data(file_inode(seq->file)); ++ struct ext4_group_desc *gdp; + ext4_group_t group = (ext4_group_t) ((unsigned long) v); + int i; + int err, buddy_loaded = 0; ++ int free = 0; + struct ext4_buddy e4b; + struct ext4_group_info *grinfo; + unsigned char blocksize_bits = min_t(unsigned char, +@@ -2873,7 +2879,7 @@ static int ext4_mb_seq_groups_show(struct seq_file *seq, void *v) + + group--; + if (group == 0) +- seq_puts(seq, "#group: free frags first [" ++ seq_puts(seq, "#group: bfree gfree frags first pa [" + " 2^0 2^1 2^2 2^3 2^4 2^5 2^6 " + " 2^7 2^8 2^9 2^10 2^11 2^12 2^13 ]\n"); + +@@ -2891,13 +2897,19 @@ static int ext4_mb_seq_groups_show(struct seq_file *seq, void *v) + buddy_loaded = 1; + } + ++ gdp = ext4_get_group_desc(sb, group, NULL); ++ if (gdp != NULL) ++ free = ext4_free_group_clusters(sb, gdp); ++ + memcpy(&sg, ext4_get_group_info(sb, group), i); + + if (buddy_loaded) + ext4_mb_unload_buddy(&e4b); + +- seq_printf(seq, "#%-5u: %-5u %-5u %-5u [", group, sg.info.bb_free, +- sg.info.bb_fragments, sg.info.bb_first_free); ++ seq_printf(seq, "#%-5lu: %-5u %-5u %-5u %-5u %-5lu [", ++ (long unsigned int)group, sg.info.bb_free, free, ++ sg.info.bb_fragments, sg.info.bb_first_free, ++ sg.info.bb_prealloc_nr); + for (i = 0; i <= 13; i++) + seq_printf(seq, " %-5u", i <= blocksize_bits + 1 ? + sg.info.bb_counters[i] : 0); +@@ -4589,23 +4601,72 @@ static void ext4_mb_generate_from_freelist(struct super_block *sb, void *bitmap, + return; + } + ++/* ++ * check free blocks in bitmap match free block in group descriptor ++ * do this before taking preallocated blocks into account to be able ++ * to detect on-disk corruptions. The group lock should be hold by the ++ * caller. ++ */ ++int ext4_mb_check_ondisk_bitmap(struct super_block *sb, void *bitmap, ++ struct ext4_group_desc *gdp, int group) ++{ ++ unsigned short max = EXT4_CLUSTERS_PER_GROUP(sb); ++ unsigned short i, first, free = 0; ++ unsigned short free_in_gdp = ext4_free_group_clusters(sb, gdp); ++ ++ if (free_in_gdp == 0 && gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) ++ return 0; ++ ++ i = mb_find_next_zero_bit(bitmap, max, 0); ++ ++ while (i < max) { ++ first = i; ++ i = mb_find_next_bit(bitmap, max, i); ++ if (i > max) ++ i = max; ++ free += i - first; ++ if (i < max) ++ i = mb_find_next_zero_bit(bitmap, max, i); ++ } ++ ++ if (free != free_in_gdp) { ++ ext4_error(sb, "on-disk bitmap for group %d" ++ "corrupted: %u blocks free in bitmap, %u - in gd\n", ++ group, free, free_in_gdp); ++ return -EIO; ++ } ++ return 0; ++} ++ + /* + * the function goes through all preallocation in this group and marks them + * used in in-core bitmap. buddy must be generated from this bitmap + * Need to be called with ext4 group lock held + */ + static noinline_for_stack +-void ext4_mb_generate_from_pa(struct super_block *sb, void *bitmap, ++int ext4_mb_generate_from_pa(struct super_block *sb, void *bitmap, + ext4_group_t group) + { + struct ext4_group_info *grp = ext4_get_group_info(sb, group); + struct ext4_prealloc_space *pa; ++ struct ext4_group_desc *gdp; + struct list_head *cur; + ext4_group_t groupnr; + ext4_grpblk_t start; + int preallocated = 0; ++ int skip = 0, count = 0; ++ int err; + int len; + ++ gdp = ext4_get_group_desc(sb, group, NULL); ++ if (gdp == NULL) ++ return -EIO; ++ ++ /* before applying preallocations, check bitmap consistency */ ++ err = ext4_mb_check_ondisk_bitmap(sb, bitmap, gdp, group); ++ if (err) ++ return err; ++ + /* all form of preallocation discards first load group, + * so the only competing code is preallocation use. + * we don't need any locking here +@@ -4621,13 +4682,23 @@ void ext4_mb_generate_from_pa(struct super_block *sb, void *bitmap, + &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); + mb_set_bits(bitmap, start, len); + preallocated += len; ++ count++; ++ } ++ if (count + skip != grp->bb_prealloc_nr) { ++ ext4_error(sb, "lost preallocations: " ++ "count %d, bb_prealloc_nr %lu, skip %d\n", ++ count, grp->bb_prealloc_nr, skip); ++ return -EIO; + } + mb_debug(sb, "preallocated %d for group %u\n", preallocated, group); ++ return 0; + } + + static void ext4_mb_mark_pa_deleted(struct super_block *sb, +@@ -4711,6 +4782,7 @@ static void ext4_mb_put_pa(struct ext4_allocation_context *ac, + */ + 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); +@@ -4802,6 +4874,7 @@ ext4_mb_new_inode_pa(struct ext4_allocation_context *ac) + pa->pa_inode = ac->ac_inode; + + list_add(&pa->pa_group_list, &grp->bb_prealloc_list); ++ grp->bb_prealloc_nr++; + + spin_lock(pa->pa_obj_lock); + list_add_rcu(&pa->pa_inode_list, &ei->i_prealloc_list); +@@ -4857,6 +4930,7 @@ ext4_mb_new_group_pa(struct ext4_allocation_context *ac) + pa->pa_inode = NULL; + + list_add(&pa->pa_group_list, &grp->bb_prealloc_list); ++ grp->bb_prealloc_nr++; + + /* + * We will later add the new pa to the right bucket +@@ -5021,6 +5095,8 @@ ext4_mb_discard_group_preallocations(struct super_block *sb, + + 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); + } +@@ -5148,7 +5224,7 @@ repeat: + if (err) { + ext4_error_err(sb, -err, "Error %d loading buddy information for %u", + err, group); +- continue; ++ return; + } + + bitmap_bh = ext4_read_block_bitmap(sb, group); +@@ -5161,6 +5237,8 @@ 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); +@@ -5455,6 +5533,7 @@ ext4_mb_discard_lg_preallocations(struct super_block *sb, + } + 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); + +diff --git a/fs/ext4/mballoc.h b/fs/ext4/mballoc.h +index 39da92c..b6c4a30 100644 +--- a/fs/ext4/mballoc.h ++++ b/fs/ext4/mballoc.h +@@ -66,7 +66,7 @@ + /* + * for which requests use 2^N search using buddies + */ +-#define MB_DEFAULT_ORDER2_REQS 2 ++#define MB_DEFAULT_ORDER2_REQS 8 + + /* + * default group prealloc size 512 blocks +-- +2.34.1 + --- /dev/null +++ b/ldiskfs/kernel_patches/patches/linux-5.18/ext4-mballoc-pa-free-mismatch.patch @@ -0,0 +1,110 @@ +--- + fs/ext4/mballoc.c | 43 +++++++++++++++++++++++++++++++++++++------ + fs/ext4/mballoc.h | 1 + + 2 files changed, 38 insertions(+), 6 deletions(-) + +diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c +index 5dbd7e7..961732c 100644 +--- a/fs/ext4/mballoc.c ++++ b/fs/ext4/mballoc.c +@@ -4859,6 +4859,7 @@ ext4_mb_new_inode_pa(struct ext4_allocation_context *ac) + INIT_LIST_HEAD(&pa->pa_group_list); + pa->pa_deleted = 0; + pa->pa_type = MB_INODE_PA; ++ pa->pa_error = 0; + + mb_debug(sb, "new inode pa %p: %llu/%d for %u\n", pa, pa->pa_pstart, + pa->pa_len, pa->pa_lstart); +@@ -4914,6 +4915,7 @@ ext4_mb_new_group_pa(struct ext4_allocation_context *ac) + INIT_LIST_HEAD(&pa->pa_group_list); + pa->pa_deleted = 0; + pa->pa_type = MB_GROUP_PA; ++ pa->pa_error = 0; + + mb_debug(sb, "new group pa %p: %llu/%d for %u\n", pa, pa->pa_pstart, + pa->pa_len, pa->pa_lstart); +@@ -4967,7 +4969,9 @@ ext4_mb_release_inode_pa(struct ext4_buddy *e4b, struct buffer_head *bitmap_bh, + unsigned long long grp_blk_start; + int free = 0; + ++ assert_spin_locked(ext4_group_lock_ptr(sb, e4b->bd_group)); + BUG_ON(pa->pa_deleted == 0); ++ BUG_ON(pa->pa_inode == NULL); + ext4_get_group_no_and_offset(sb, pa->pa_pstart, &group, &bit); + grp_blk_start = pa->pa_pstart - EXT4_C2B(sbi, bit); + BUG_ON(group != e4b->bd_group && pa->pa_len != 0); +@@ -4990,12 +4994,18 @@ ext4_mb_release_inode_pa(struct ext4_buddy *e4b, struct buffer_head *bitmap_bh, + mb_free_blocks(pa->pa_inode, e4b, bit, next - bit); + bit = next + 1; + } +- if (free != pa->pa_free) { +- ext4_msg(e4b->bd_sb, KERN_CRIT, +- "pa %p: logic %lu, phys. %lu, len %d", +- pa, (unsigned long) pa->pa_lstart, +- (unsigned long) pa->pa_pstart, +- pa->pa_len); ++ ++ /* "free < pa->pa_free" means we maybe double alloc the same blocks, ++ * otherwise maybe leave some free blocks unavailable, no need to BUG.*/ ++ if ((free > pa->pa_free && !pa->pa_error) || (free < pa->pa_free)) { ++ ext4_error(sb, "pa free mismatch: [pa %p] " ++ "[phy %lu] [logic %lu] [len %u] [free %u] " ++ "[error %u] [inode %d] [freed %u]", pa, ++ (unsigned long)pa->pa_pstart, ++ (unsigned long)pa->pa_lstart, ++ pa->pa_len, (unsigned)pa->pa_free, ++ (unsigned)pa->pa_error, pa->pa_inode->i_ino, ++ free); + ext4_grp_locked_error(sb, group, 0, 0, "free %u, pa_free %u", + free, pa->pa_free); + /* +@@ -5003,6 +5013,8 @@ ext4_mb_release_inode_pa(struct ext4_buddy *e4b, struct buffer_head *bitmap_bh, + * from the bitmap and continue. + */ + } ++ /* do not verify if the file system is being umounted */ ++ BUG_ON(atomic_read(&sb->s_active) > 0 && pa->pa_free != free); + atomic_add(free, &sbi->s_mb_discarded); + + return 0; +@@ -5858,6 +5870,25 @@ 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: +diff --git a/fs/ext4/mballoc.h b/fs/ext4/mballoc.h +index b6c4a30..f1a9152 100644 +--- a/fs/ext4/mballoc.h ++++ b/fs/ext4/mballoc.h +@@ -128,6 +128,7 @@ 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 */ + }; +-- +2.34.1 + --- /dev/null +++ b/ldiskfs/kernel_patches/patches/linux-5.18/ext4-prealloc.patch @@ -0,0 +1,403 @@ +From bdb2b47482f4e525ae0a7bc1334cbeca9115beac Mon Sep 17 00:00:00 2001 +From: Shaun Tancheff <shaun.tancheff@hpe.com> +Date: Mon, 14 Nov 2022 01:18:58 -0600 +Subject: [PATCH] linux-5.18/ext4-prealloc.patch + +--- + fs/ext4/ext4.h | 7 +- + fs/ext4/inode.c | 3 + + fs/ext4/mballoc.c | 219 +++++++++++++++++++++++++++++++++++----------- + fs/ext4/sysfs.c | 8 +- + 4 files changed, 182 insertions(+), 55 deletions(-) + +diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h +index b03dbd7..78d3176 100644 +--- a/fs/ext4/ext4.h ++++ b/fs/ext4/ext4.h +@@ -1291,6 +1291,8 @@ extern void mb_set_bits(void *bm, int cur, int len); + #define EXT4_DFL_MAX_MNT_COUNT 20 /* Allow 20 mounts */ + #define EXT4_DFL_CHECKINTERVAL 0 /* Don't use interval check */ + ++#define EXT4_MAX_PREALLOC_TABLE 64 ++ + /* + * Behaviour when detecting errors + */ +@@ -1597,11 +1599,13 @@ struct ext4_sb_info { + /* tunables */ + unsigned long s_stripe; + unsigned int s_mb_max_linear_groups; +- unsigned int s_mb_stream_request; ++ unsigned long s_mb_small_req; ++ unsigned long s_mb_large_req; + unsigned int s_mb_max_to_scan; + unsigned int s_mb_min_to_scan; + unsigned int s_mb_stats; + unsigned int s_mb_order2_reqs; ++ unsigned long *s_mb_prealloc_table; + unsigned int s_mb_group_prealloc; + unsigned int s_mb_max_inode_prealloc; + unsigned int s_max_dir_size_kb; +@@ -2940,6 +2944,7 @@ int ext4_fc_record_regions(struct super_block *sb, int ino, + int len, int replay); + + /* mballoc.c */ ++extern const struct proc_ops ext4_seq_prealloc_table_fops; + extern const struct seq_operations ext4_mb_seq_groups_ops; + extern const struct seq_operations ext4_mb_seq_structs_summary_ops; + extern long ext4_mb_stats; +diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c +index 6723622..2ed00ae 100644 +--- a/fs/ext4/inode.c ++++ b/fs/ext4/inode.c +@@ -2720,6 +2720,9 @@ static int ext4_writepages(struct address_space *mapping, + PAGE_SIZE >> inode->i_blkbits); + } + ++ if (wbc->nr_to_write < sbi->s_mb_small_req) ++ wbc->nr_to_write = sbi->s_mb_small_req; ++ + if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) + range_whole = 1; + +diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c +index 8164480..94a299b 100644 +--- a/fs/ext4/mballoc.c ++++ b/fs/ext4/mballoc.c +@@ -3070,6 +3070,99 @@ const struct seq_operations ext4_mb_seq_structs_summary_ops = { + .show = ext4_mb_seq_structs_summary_show, + }; + ++static int ext4_mb_check_and_update_prealloc(struct ext4_sb_info *sbi, ++ char *str, size_t cnt, ++ int update) ++{ ++ unsigned long value; ++ unsigned long prev = 0; ++ char *cur; ++ char *next; ++ char *end; ++ int num = 0; ++ ++ cur = str; ++ end = str + cnt; ++ while (cur < end) { ++ while ((cur < end) && (*cur == ' ')) cur++; ++ value = simple_strtol(cur, &next, 0); ++ if (value == 0) ++ break; ++ if (cur == next) ++ return -EINVAL; ++ ++ cur = next; ++ ++ if (value > (sbi->s_blocks_per_group - 1 - 1 - sbi->s_itb_per_group)) ++ return -EINVAL; ++ ++ /* they should add values in order */ ++ if (value <= prev) ++ return -EINVAL; ++ ++ if (update) ++ sbi->s_mb_prealloc_table[num] = value; ++ ++ prev = value; ++ num++; ++ } ++ ++ if (num > EXT4_MAX_PREALLOC_TABLE - 1) ++ return -EOVERFLOW; ++ ++ if (update) ++ sbi->s_mb_prealloc_table[num] = 0; ++ ++ return 0; ++} ++ ++static ssize_t ext4_mb_prealloc_table_proc_write(struct file *file, ++ const char __user *buf, ++ size_t cnt, loff_t *pos) ++{ ++ struct ext4_sb_info *sbi = EXT4_SB(pde_data(file_inode(file))); ++ char str[128]; ++ int rc; ++ ++ if (cnt >= sizeof(str)) ++ return -EINVAL; ++ if (copy_from_user(str, buf, cnt)) ++ return -EFAULT; ++ ++ rc = ext4_mb_check_and_update_prealloc(sbi, str, cnt, 0); ++ if (rc) ++ return rc; ++ ++ rc = ext4_mb_check_and_update_prealloc(sbi, str, cnt, 1); ++ return rc ? rc : cnt; ++} ++ ++static int mb_prealloc_table_seq_show(struct seq_file *m, void *v) ++{ ++ struct ext4_sb_info *sbi = EXT4_SB(m->private); ++ int i; ++ ++ for (i = 0; i < EXT4_MAX_PREALLOC_TABLE && ++ sbi->s_mb_prealloc_table[i] != 0; i++) ++ seq_printf(m, "%ld ", sbi->s_mb_prealloc_table[i]); ++ seq_printf(m, "\n"); ++ ++ return 0; ++} ++ ++static int mb_prealloc_table_seq_open(struct inode *inode, struct file *file) ++{ ++ return single_open(file, mb_prealloc_table_seq_show, pde_data(inode)); ++} ++ ++const struct proc_ops ext4_seq_prealloc_table_fops = { ++ .proc_open = mb_prealloc_table_seq_open, ++ .proc_read = seq_read, ++ .proc_lseek = seq_lseek, ++ .proc_release = single_release, ++ .proc_write = ext4_mb_prealloc_table_proc_write, ++}; ++ + static struct kmem_cache *get_groupinfo_cache(int blocksize_bits) + { + int cache_index = blocksize_bits - EXT4_MIN_BLOCK_LOG_SIZE; +@@ -3382,7 +3475,7 @@ static void ext4_discard_work(struct work_struct *work) + int ext4_mb_init(struct super_block *sb) + { + struct ext4_sb_info *sbi = EXT4_SB(sb); +- unsigned i, j; ++ unsigned i, j, k, l; + unsigned offset, offset_incr; + unsigned max; + int ret; +@@ -3454,7 +3547,6 @@ int ext4_mb_init(struct super_block *sb) + sbi->s_mb_max_to_scan = MB_DEFAULT_MAX_TO_SCAN; + sbi->s_mb_min_to_scan = MB_DEFAULT_MIN_TO_SCAN; + sbi->s_mb_stats = MB_DEFAULT_STATS; +- sbi->s_mb_stream_request = MB_DEFAULT_STREAM_THRESHOLD; + sbi->s_mb_order2_reqs = MB_DEFAULT_ORDER2_REQS; + sbi->s_mb_max_inode_prealloc = MB_DEFAULT_MAX_INODE_PREALLOC; + /* +@@ -3479,9 +3571,29 @@ int ext4_mb_init(struct super_block *sb) + * RAID stripe size so that preallocations don't fragment + * the stripes. + */ +- if (sbi->s_stripe > 1) { +- sbi->s_mb_group_prealloc = roundup( +- sbi->s_mb_group_prealloc, sbi->s_stripe); ++ ++ /* Allocate table once */ ++ sbi->s_mb_prealloc_table = kzalloc( ++ EXT4_MAX_PREALLOC_TABLE * sizeof(unsigned long), GFP_NOFS); ++ if (sbi->s_mb_prealloc_table == NULL) { ++ ret = -ENOMEM; ++ goto out; ++ } ++ ++ if (sbi->s_stripe == 0) { ++ for (k = 0, l = 4; k <= 9; ++k, l *= 2) ++ sbi->s_mb_prealloc_table[k] = l; ++ ++ sbi->s_mb_small_req = 256; ++ sbi->s_mb_large_req = 1024; ++ sbi->s_mb_group_prealloc = 512; ++ } else { ++ for (k = 0, l = sbi->s_stripe; k <= 2; ++k, l *= 2) ++ sbi->s_mb_prealloc_table[k] = l; ++ ++ sbi->s_mb_small_req = sbi->s_stripe; ++ sbi->s_mb_large_req = sbi->s_stripe * 8; ++ sbi->s_mb_group_prealloc = sbi->s_stripe * 4; + } + + sbi->s_locality_groups = alloc_percpu(struct ext4_locality_group); +@@ -3515,6 +3627,7 @@ out_free_locality_groups: + out: + kfree(sbi->s_mb_largest_free_orders); + kfree(sbi->s_mb_largest_free_orders_locks); ++ kfree(sbi->s_mb_prealloc_table); + kfree(sbi->s_mb_offsets); + sbi->s_mb_offsets = NULL; + kfree(sbi->s_mb_maxs); +@@ -3783,7 +3896,6 @@ ext4_mb_mark_diskspace_used(struct ext4_allocation_context *ac, + int err, len; + + BUG_ON(ac->ac_status != AC_STATUS_FOUND); +- BUG_ON(ac->ac_b_ex.fe_len <= 0); + + sb = ac->ac_sb; + sbi = EXT4_SB(sb); +@@ -4026,13 +4138,14 @@ ext4_mb_normalize_request(struct ext4_allocation_context *ac, + struct ext4_allocation_request *ar) + { + struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb); +- int bsbits, max; ++ int bsbits, i, wind; + ext4_lblk_t end; +- loff_t size, start_off; ++ loff_t size; + loff_t orig_size __maybe_unused; + ext4_lblk_t start; + struct ext4_inode_info *ei = EXT4_I(ac->ac_inode); + struct ext4_prealloc_space *pa; ++ unsigned long value, last_non_zero; + + /* do normalize only data requests, metadata requests + do not need preallocation */ +@@ -4061,51 +4174,46 @@ ext4_mb_normalize_request(struct ext4_allocation_context *ac, + size = size << bsbits; + if (size < i_size_read(ac->ac_inode)) + size = i_size_read(ac->ac_inode); +- orig_size = size; ++ size = (size + ac->ac_sb->s_blocksize - 1) >> bsbits; ++ ++ start = wind = 0; ++ value = last_non_zero = 0; ++ ++ /* let's choose preallocation window depending on file size */ ++ for (i = 0; i < EXT4_MAX_PREALLOC_TABLE; i++) { ++ value = sbi->s_mb_prealloc_table[i]; ++ if (value == 0) ++ break; ++ else ++ last_non_zero = value; + +- /* max size of free chunks */ +- max = 2 << bsbits; +- +-#define NRL_CHECK_SIZE(req, size, max, chunk_size) \ +- (req <= (size) || max <= (chunk_size)) +- +- /* first, try to predict filesize */ +- /* XXX: should this table be tunable? */ +- start_off = 0; +- if (size <= 16 * 1024) { +- size = 16 * 1024; +- } else if (size <= 32 * 1024) { +- size = 32 * 1024; +- } else if (size <= 64 * 1024) { +- size = 64 * 1024; +- } else if (size <= 128 * 1024) { +- size = 128 * 1024; +- } else if (size <= 256 * 1024) { +- size = 256 * 1024; +- } else if (size <= 512 * 1024) { +- size = 512 * 1024; +- } else if (size <= 1024 * 1024) { +- size = 1024 * 1024; +- } else if (NRL_CHECK_SIZE(size, 4 * 1024 * 1024, max, 2 * 1024)) { +- start_off = ((loff_t)ac->ac_o_ex.fe_logical >> +- (21 - bsbits)) << 21; +- size = 2 * 1024 * 1024; +- } else if (NRL_CHECK_SIZE(size, 8 * 1024 * 1024, max, 4 * 1024)) { +- start_off = ((loff_t)ac->ac_o_ex.fe_logical >> +- (22 - bsbits)) << 22; +- size = 4 * 1024 * 1024; +- } else if (NRL_CHECK_SIZE(ac->ac_o_ex.fe_len, +- (8<<20)>>bsbits, max, 8 * 1024)) { +- start_off = ((loff_t)ac->ac_o_ex.fe_logical >> +- (23 - bsbits)) << 23; +- size = 8 * 1024 * 1024; ++ if (size <= value) { ++ wind = value; ++ break; ++ } ++ } ++ ++ if (wind == 0) { ++ if (last_non_zero != 0) { ++ __u64 tstart, tend; ++ /* file is quite large, we now preallocate with ++ * the biggest configured window with regart to ++ * logical offset */ ++ wind = last_non_zero; ++ tstart = ac->ac_o_ex.fe_logical; ++ do_div(tstart, wind); ++ start = tstart * wind; ++ tend = ac->ac_o_ex.fe_logical + ac->ac_o_ex.fe_len - 1; ++ do_div(tend, wind); ++ tend = tend * wind + wind; ++ size = tend - start; ++ } + } else { +- start_off = (loff_t) ac->ac_o_ex.fe_logical << bsbits; +- size = (loff_t) EXT4_C2B(EXT4_SB(ac->ac_sb), +- ac->ac_o_ex.fe_len) << bsbits; ++ size = wind; + } +- size = size >> bsbits; +- start = start_off >> bsbits; ++ ++ ++ orig_size = size; + + /* + * For tiny groups (smaller than 8MB) the chosen allocation +@@ -4196,7 +4304,6 @@ ext4_mb_normalize_request(struct ext4_allocation_context *ac, + (unsigned long) ac->ac_o_ex.fe_logical); + BUG(); + } +- BUG_ON(size <= 0 || size > EXT4_BLOCKS_PER_GROUP(ac->ac_sb)); + + /* now prepare goal request */ + +@@ -5199,11 +5306,19 @@ static void ext4_mb_group_or_file(struct ext4_allocation_context *ac) + + /* don't use group allocation for large files */ + size = max(size, isize); +- if (size > sbi->s_mb_stream_request) { ++ if ((ac->ac_o_ex.fe_len >= sbi->s_mb_small_req) || ++ (size >= sbi->s_mb_large_req)) { + ac->ac_flags |= EXT4_MB_STREAM_ALLOC; + return; + } + ++ /* ++ * request is so large that we don't care about ++ * streaming - it overweights any possible seek ++ */ ++ if (ac->ac_o_ex.fe_len >= sbi->s_mb_large_req) ++ return; ++ + BUG_ON(ac->ac_lg != NULL); + /* + * locality group prealloc space are per cpu. The reason for having +diff --git a/fs/ext4/sysfs.c b/fs/ext4/sysfs.c +index d233c24..9d2254f 100644 +--- a/fs/ext4/sysfs.c ++++ b/fs/ext4/sysfs.c +@@ -212,7 +212,8 @@ EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats); + EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan); + EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan); + EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs); +-EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request); ++EXT4_RW_ATTR_SBI_UI(mb_small_req, s_mb_small_req); ++EXT4_RW_ATTR_SBI_UI(mb_large_req, s_mb_large_req); + EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc); + EXT4_RW_ATTR_SBI_UI(mb_max_inode_prealloc, s_mb_max_inode_prealloc); + EXT4_RW_ATTR_SBI_UI(mb_max_linear_groups, s_mb_max_linear_groups); +@@ -262,7 +263,8 @@ static struct attribute *ext4_attrs[] = { + ATTR_LIST(mb_max_to_scan), + ATTR_LIST(mb_min_to_scan), + ATTR_LIST(mb_order2_req), +- ATTR_LIST(mb_stream_req), ++ ATTR_LIST(mb_small_req), ++ ATTR_LIST(mb_large_req), + ATTR_LIST(mb_group_prealloc), + ATTR_LIST(mb_max_inode_prealloc), + ATTR_LIST(mb_max_linear_groups), +@@ -543,6 +545,8 @@ int ext4_register_sysfs(struct super_block *sb) + ext4_fc_info_show, sb); + proc_create_seq_data("mb_groups", S_IRUGO, sbi->s_proc, + &ext4_mb_seq_groups_ops, sb); ++ proc_create_data("prealloc_table", S_IRUGO, sbi->s_proc, ++ &ext4_seq_prealloc_table_fops, sb); + proc_create_single_data("mb_stats", 0444, sbi->s_proc, + ext4_seq_mb_stats_show, sb); + proc_create_seq_data("mb_structs_summary", 0444, sbi->s_proc, +-- +2.34.1 + --- /dev/null +++ b/ldiskfs/kernel_patches/patches/rhel9.1/ext4-enc-flag.patch @@ -0,0 +1,70 @@ +From 5bca121ff1fa1baa45516b76869de70254ef80f6 Mon Sep 17 00:00:00 2001 +From: Shaun Tancheff <shaun.tancheff@hpe.com> +Date: Fri, 20 Jan 2023 09:14:45 -0600 +Subject: [PATCH] rhel9.1/ext4-enc-flag + +--- + fs/ext4/ext4.h | 3 +++ + fs/ext4/inode.c | 6 ++++-- + fs/ext4/xattr.c | 7 +++++++ + 3 files changed, 14 insertions(+), 2 deletions(-) + +diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h +index fd40157..276a4f5 100644 +--- a/fs/ext4/ext4.h ++++ b/fs/ext4/ext4.h +@@ -1253,6 +1253,9 @@ struct ext4_inode_info { + #define EXT4_MOUNT_DISCARD 0x40000000 /* Issue DISCARD requests */ + #define EXT4_MOUNT_INIT_INODE_TABLE 0x80000000 /* Initialize uninitialized itables */ + ++/* we know this is a Lustre mount thanks to the DIRDATA flag */ ++#define IS_LUSTRE_MOUNT(sb) test_opt((sb), DIRDATA) ++ + /* + * Mount flags set either automatically (could not be set by mount option) + * based on per file system feature or property or in special cases such as +diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c +index d66a142..0aa15d1 100644 +--- a/fs/ext4/inode.c ++++ b/fs/ext4/inode.c +@@ -4631,7 +4631,8 @@ void ext4_set_inode_flags(struct inode *inode, bool init) + if (init && ext4_should_enable_dax(inode)) + new_fl |= S_DAX; + +- if (flags & EXT4_ENCRYPT_FL) ++ if (flags & EXT4_ENCRYPT_FL && ++ unlikely(!IS_LUSTRE_MOUNT(inode->i_sb))) + new_fl |= S_ENCRYPTED; + if (flags & EXT4_CASEFOLD_FL) + new_fl |= S_CASEFOLD; +@@ -5546,7 +5547,8 @@ int ext4_getattr(struct user_namespace *mnt_userns, const struct path *path, + stat->attributes |= STATX_ATTR_APPEND; + if (flags & EXT4_COMPR_FL) + stat->attributes |= STATX_ATTR_COMPRESSED; +- if (flags & EXT4_ENCRYPT_FL) ++ if (flags & EXT4_ENCRYPT_FL && ++ unlikely(!IS_LUSTRE_MOUNT(inode->i_sb))) + stat->attributes |= STATX_ATTR_ENCRYPTED; + if (flags & EXT4_IMMUTABLE_FL) + stat->attributes |= STATX_ATTR_IMMUTABLE; +diff --git a/fs/ext4/xattr.c b/fs/ext4/xattr.c +index 5d92078..6f83ad0 100644 +--- a/fs/ext4/xattr.c ++++ b/fs/ext4/xattr.c +@@ -2483,6 +2483,13 @@ retry_inode: + if (IS_SYNC(inode)) + ext4_handle_sync(handle); + } ++ ++ if (!error && name_index == EXT4_XATTR_INDEX_ENCRYPTION && ++ strcmp(name, "c") == 0) { ++ EXT4_I(inode)->i_flags |= EXT4_ENCRYPT_FL; ++ mark_inode_dirty(inode); ++ } ++ + ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, handle); + + cleanup: +-- +2.34.1 + --- /dev/null +++ b/ldiskfs/kernel_patches/patches/rhel9.1/ext4-pdirop.patch @@ -0,0 +1,930 @@ +Subject: [PATCH] ext4-pdirop + + In a typical use case an application creates a separate output file for each + node and task in a job. As nodes and tasks increase, hundreds of thousands of + files may be created in a single directory within a short window of time. + Today, both filename lookup and file system modifying operations (such as + create and unlink) are protected with a single lock for an entire ldiskfs + directory. PDO project will remove this bottleneck by introducing a parallel + locking mechanism for entire ldiskfs directories. This work will enable + multiple application threads to simultaneously lookup, create and unlink in + parallel. + +This patch contains: + - pdirops support for ldiskfs + - integrate with osd-ldiskfs +Signed-off-by: Liang Zhen <liang.zhen@intel.com> +Change-Id: I269c0e3112e68f3acd79e860dab052a68c7d7aaa +--- + fs/ext4/Makefile | 1 + + fs/ext4/ext4.h | 78 ++++++++ + fs/ext4/namei.c | 465 ++++++++++++++++++++++++++++++++++++++++++----- + fs/ext4/super.c | 1 + + 4 files changed, 504 insertions(+), 41 deletions(-) + +diff --git a/fs/ext4/Makefile b/fs/ext4/Makefile +index 7d89142..e0dea46 100644 +--- a/fs/ext4/Makefile ++++ b/fs/ext4/Makefile +@@ -7,6 +7,7 @@ obj-$(CONFIG_EXT4_FS) += ext4.o + + ext4-y := balloc.o bitmap.o block_validity.o dir.o ext4_jbd2.o extents.o \ + extents_status.o file.o fsmap.o fsync.o hash.o ialloc.o \ ++ htree_lock.o \ + indirect.o inline.o inode.o ioctl.o mballoc.o migrate.o \ + mmp.o move_extent.o namei.o page-io.o readpage.o resize.o \ + super.o symlink.o sysfs.o xattr.o xattr_hurd.o xattr_trusted.o \ +diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h +index 73f594d..febe183 100644 +--- a/fs/ext4/ext4.h ++++ b/fs/ext4/ext4.h +@@ -29,6 +29,7 @@ + #include <linux/mutex.h> + #include <linux/timer.h> + #include <linux/wait.h> ++#include <linux/htree_lock.h> + #include <linux/sched/signal.h> + #include <linux/blockgroup_lock.h> + #include <linux/percpu_counter.h> +@@ -1019,6 +1020,9 @@ 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, +@@ -2554,6 +2558,72 @@ struct dx_hash_info + */ + #define HASH_NB_ALWAYS 1 + ++/* assume name-hash is protected by upper layer */ ++#define EXT4_HTREE_LOCK_HASH 0 ++ ++enum ext4_pdo_lk_types { ++#if EXT4_HTREE_LOCK_HASH ++ EXT4_LK_HASH, ++#endif ++ EXT4_LK_DX, /* index block */ ++ EXT4_LK_DE, /* directory entry block */ ++ EXT4_LK_SPIN, /* spinlock */ ++ EXT4_LK_MAX, ++}; ++ ++/* read-only bit */ ++#define EXT4_LB_RO(b) (1 << (b)) ++/* read + write, high bits for writer */ ++#define EXT4_LB_RW(b) ((1 << (b)) | (1 << (EXT4_LK_MAX + (b)))) ++ ++enum ext4_pdo_lock_bits { ++ /* DX lock bits */ ++ EXT4_LB_DX_RO = EXT4_LB_RO(EXT4_LK_DX), ++ EXT4_LB_DX = EXT4_LB_RW(EXT4_LK_DX), ++ /* DE lock bits */ ++ EXT4_LB_DE_RO = EXT4_LB_RO(EXT4_LK_DE), ++ EXT4_LB_DE = EXT4_LB_RW(EXT4_LK_DE), ++ /* DX spinlock bits */ ++ EXT4_LB_SPIN_RO = EXT4_LB_RO(EXT4_LK_SPIN), ++ EXT4_LB_SPIN = EXT4_LB_RW(EXT4_LK_SPIN), ++ /* accurate searching */ ++ EXT4_LB_EXACT = EXT4_LB_RO(EXT4_LK_MAX << 1), ++}; ++ ++enum ext4_pdo_lock_opc { ++ /* external */ ++ EXT4_HLOCK_READDIR = (EXT4_LB_DE_RO | EXT4_LB_DX_RO), ++ EXT4_HLOCK_LOOKUP = (EXT4_LB_DE_RO | EXT4_LB_SPIN_RO | ++ EXT4_LB_EXACT), ++ EXT4_HLOCK_DEL = (EXT4_LB_DE | EXT4_LB_SPIN_RO | ++ EXT4_LB_EXACT), ++ EXT4_HLOCK_ADD = (EXT4_LB_DE | EXT4_LB_SPIN_RO), ++ ++ /* internal */ ++ EXT4_HLOCK_LOOKUP_SAFE = (EXT4_LB_DE_RO | EXT4_LB_DX_RO | ++ EXT4_LB_EXACT), ++ EXT4_HLOCK_DEL_SAFE = (EXT4_LB_DE | EXT4_LB_DX_RO | EXT4_LB_EXACT), ++ EXT4_HLOCK_SPLIT = (EXT4_LB_DE | EXT4_LB_DX | EXT4_LB_SPIN), ++}; ++ ++extern struct htree_lock_head *ext4_htree_lock_head_alloc(unsigned hbits); ++#define ext4_htree_lock_head_free(lhead) htree_lock_head_free(lhead) ++ ++extern struct htree_lock *ext4_htree_lock_alloc(void); ++#define ext4_htree_lock_free(lck) htree_lock_free(lck) ++ ++extern void ext4_htree_lock(struct htree_lock *lck, ++ struct htree_lock_head *lhead, ++ struct inode *dir, unsigned flags); ++#define ext4_htree_unlock(lck) htree_unlock(lck) ++ ++extern struct buffer_head *ext4_find_entry_locked(struct inode *dir, ++ const struct qstr *d_name, ++ struct ext4_dir_entry_2 **res_dir, ++ int *inlined, struct htree_lock *lck); ++extern int ext4_add_entry_locked(handle_t *handle, struct dentry *dentry, ++ struct inode *inode, struct htree_lock *lck); ++ + struct ext4_filename { + const struct qstr *usr_fname; + struct fscrypt_str disk_name; +@@ -2932,12 +3002,20 @@ void ext4_insert_dentry(struct inode *dir, struct inode *inode, + void *data); + static inline void ext4_update_dx_flag(struct inode *inode) + { ++ /* Disable it for ldiskfs, because going from a DX directory to ++ * a non-DX directory while it is in use will completely break ++ * the htree-locking. ++ * If we really want to support this operation in the future, ++ * we need to exclusively lock the directory at here which will ++ * increase complexity of code */ ++#if 0 + if (!ext4_has_feature_dir_index(inode->i_sb) && + ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) { + /* ext4_iget() should have caught this... */ + WARN_ON_ONCE(ext4_has_feature_metadata_csum(inode->i_sb)); + ext4_clear_inode_flag(inode, EXT4_INODE_INDEX); + } ++#endif + } + static const unsigned char ext4_filetype_table[] = { + DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK +diff --git a/fs/ext4/namei.c b/fs/ext4/namei.c +index 2760dc6..2d14bd2 100644 +--- a/fs/ext4/namei.c ++++ b/fs/ext4/namei.c +@@ -56,6 +56,7 @@ struct buffer_head *ext4_append(handle_t *handle, + { + struct ext4_map_blocks map; + struct buffer_head *bh; ++ struct ext4_inode_info *ei = EXT4_I(inode); + int err; + + if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb && +@@ -63,6 +64,10 @@ struct buffer_head *ext4_append(handle_t *handle, + 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; + map.m_lblk = *block; + map.m_len = 1; +@@ -73,21 +78,27 @@ struct buffer_head *ext4_append(handle_t *handle, + * directory. + */ + err = ext4_map_blocks(NULL, inode, &map, 0); +- if (err < 0) ++ if (err < 0) { ++ up(&ei->i_append_sem); + return ERR_PTR(err); ++ } + if (err) { ++ up(&ei->i_append_sem); + EXT4_ERROR_INODE(inode, "Logical block already allocated"); + return ERR_PTR(-EFSCORRUPTED); + } + + bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE); +- if (IS_ERR(bh)) ++ if (IS_ERR(bh)) { ++ up(&ei->i_append_sem); + return bh; ++ } + inode->i_size += inode->i_sb->s_blocksize; + EXT4_I(inode)->i_disksize = inode->i_size; + BUFFER_TRACE(bh, "get_write_access"); + err = ext4_journal_get_write_access(handle, inode->i_sb, bh, + EXT4_JTR_NONE); ++ up(&ei->i_append_sem); + if (err) { + brelse(bh); + ext4_std_error(inode->i_sb, err); +@@ -291,7 +302,8 @@ static unsigned dx_node_limit(struct inode *dir); + static struct dx_frame *dx_probe(struct ext4_filename *fname, + struct inode *dir, + struct dx_hash_info *hinfo, +- struct dx_frame *frame); ++ struct dx_frame *frame, ++ struct htree_lock *lck); + static void dx_release(struct dx_frame *frames, struct inode *dir); + static int dx_make_map(struct inode *dir, struct buffer_head *bh, + struct dx_hash_info *hinfo, +@@ -307,12 +319,13 @@ static void dx_insert_block(struct dx_frame *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); + static struct buffer_head * ext4_dx_find_entry(struct inode *dir, + struct ext4_filename *fname, +- struct ext4_dir_entry_2 **res_dir); ++ struct ext4_dir_entry_2 **res_dir, struct htree_lock *lck); + static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname, +- struct inode *dir, struct inode *inode); ++ struct inode *dir, struct inode *inode, ++ struct htree_lock *lck); + + /* checksumming functions */ + void ext4_initialize_dirent_tail(struct buffer_head *bh, +@@ -797,6 +810,227 @@ static inline void htree_rep_invariant_check(struct dx_entry *at, + } + #endif /* DX_DEBUG */ + ++/* private data for htree_lock */ ++struct ext4_dir_lock_data { ++ unsigned ld_flags; /* bits-map for lock types */ ++ unsigned ld_count; /* # entries of the last DX block */ ++ struct dx_entry ld_at_entry; /* copy of leaf dx_entry */ ++ struct dx_entry *ld_at; /* position of leaf dx_entry */ ++}; ++ ++#define ext4_htree_lock_data(l) ((struct ext4_dir_lock_data *)(l)->lk_private) ++#define ext4_find_entry(dir, name, dirent, inline) \ ++ ext4_find_entry_locked(dir, name, dirent, inline, NULL) ++#define ext4_add_entry(handle, dentry, inode) \ ++ ext4_add_entry_locked(handle, dentry, inode, NULL) ++ ++/* NB: ext4_lblk_t is 32 bits so we use high bits to identify invalid blk */ ++#define EXT4_HTREE_NODE_CHANGED (0xcafeULL << 32) ++ ++static void ext4_htree_event_cb(void *target, void *event) ++{ ++ u64 *block = (u64 *)target; ++ ++ if (*block == dx_get_block((struct dx_entry *)event)) ++ *block = EXT4_HTREE_NODE_CHANGED; ++} ++ ++struct htree_lock_head *ext4_htree_lock_head_alloc(unsigned hbits) ++{ ++ struct htree_lock_head *lhead; ++ ++ lhead = htree_lock_head_alloc(EXT4_LK_MAX, hbits, 0); ++ if (lhead != NULL) { ++ htree_lock_event_attach(lhead, EXT4_LK_SPIN, HTREE_EVENT_WR, ++ ext4_htree_event_cb); ++ } ++ return lhead; ++} ++EXPORT_SYMBOL(ext4_htree_lock_head_alloc); ++ ++struct htree_lock *ext4_htree_lock_alloc(void) ++{ ++ return htree_lock_alloc(EXT4_LK_MAX, ++ sizeof(struct ext4_dir_lock_data)); ++} ++EXPORT_SYMBOL(ext4_htree_lock_alloc); ++ ++static htree_lock_mode_t ext4_htree_mode(unsigned flags) ++{ ++ switch (flags) { ++ default: /* 0 or unknown flags require EX lock */ ++ return HTREE_LOCK_EX; ++ case EXT4_HLOCK_READDIR: ++ return HTREE_LOCK_PR; ++ case EXT4_HLOCK_LOOKUP: ++ return HTREE_LOCK_CR; ++ case EXT4_HLOCK_DEL: ++ case EXT4_HLOCK_ADD: ++ return HTREE_LOCK_CW; ++ } ++} ++ ++/* return PR for read-only operations, otherwise return EX */ ++static inline htree_lock_mode_t ext4_htree_safe_mode(unsigned flags) ++{ ++ int writer = (flags & EXT4_LB_DE) == EXT4_LB_DE; ++ ++ /* 0 requires EX lock */ ++ return (flags == 0 || writer) ? HTREE_LOCK_EX : HTREE_LOCK_PR; ++} ++ ++static int ext4_htree_safe_locked(struct htree_lock *lck) ++{ ++ int writer; ++ ++ if (lck == NULL || lck->lk_mode == HTREE_LOCK_EX) ++ return 1; ++ ++ writer = (ext4_htree_lock_data(lck)->ld_flags & EXT4_LB_DE) == ++ EXT4_LB_DE; ++ if (writer) /* all readers & writers are excluded? */ ++ return lck->lk_mode == HTREE_LOCK_EX; ++ ++ /* all writers are excluded? */ ++ return lck->lk_mode == HTREE_LOCK_PR || ++ lck->lk_mode == HTREE_LOCK_PW || ++ lck->lk_mode == HTREE_LOCK_EX; ++} ++ ++/* relock htree_lock with EX mode if it's change operation, otherwise ++ * relock it with PR mode. It's noop if PDO is disabled. */ ++static void ext4_htree_safe_relock(struct htree_lock *lck) ++{ ++ if (!ext4_htree_safe_locked(lck)) { ++ unsigned flags = ext4_htree_lock_data(lck)->ld_flags; ++ ++ htree_change_lock(lck, ext4_htree_safe_mode(flags)); ++ } ++} ++ ++void ext4_htree_lock(struct htree_lock *lck, struct htree_lock_head *lhead, ++ struct inode *dir, unsigned flags) ++{ ++ htree_lock_mode_t mode = is_dx(dir) ? ext4_htree_mode(flags) : ++ ext4_htree_safe_mode(flags); ++ ++ ext4_htree_lock_data(lck)->ld_flags = flags; ++ htree_lock(lck, lhead, mode); ++ if (!is_dx(dir)) ++ ext4_htree_safe_relock(lck); /* make sure it's safe locked */ ++} ++EXPORT_SYMBOL(ext4_htree_lock); ++ ++static int ext4_htree_node_lock(struct htree_lock *lck, struct dx_entry *at, ++ unsigned lmask, int wait, void *ev) ++{ ++ u32 key = (at == NULL) ? 0 : dx_get_block(at); ++ u32 mode; ++ ++ /* NOOP if htree is well protected or caller doesn't require the lock */ ++ if (ext4_htree_safe_locked(lck) || ++ !(ext4_htree_lock_data(lck)->ld_flags & lmask)) ++ return 1; ++ ++ mode = (ext4_htree_lock_data(lck)->ld_flags & lmask) == lmask ? ++ HTREE_LOCK_PW : HTREE_LOCK_PR; ++ while (1) { ++ if (htree_node_lock_try(lck, mode, key, ffz(~lmask), wait, ev)) ++ return 1; ++ if (!(lmask & EXT4_LB_SPIN)) /* not a spinlock */ ++ return 0; ++ cpu_relax(); /* spin until granted */ ++ } ++} ++ ++static int ext4_htree_node_locked(struct htree_lock *lck, unsigned lmask) ++{ ++ return ext4_htree_safe_locked(lck) || ++ htree_node_is_granted(lck, ffz(~lmask)); ++} ++ ++static void ext4_htree_node_unlock(struct htree_lock *lck, ++ unsigned lmask, void *buf) ++{ ++ /* NB: it's safe to call mutiple times or even it's not locked */ ++ if (!ext4_htree_safe_locked(lck) && ++ htree_node_is_granted(lck, ffz(~lmask))) ++ htree_node_unlock(lck, ffz(~lmask), buf); ++} ++ ++#define ext4_htree_dx_lock(lck, key) \ ++ ext4_htree_node_lock(lck, key, EXT4_LB_DX, 1, NULL) ++#define ext4_htree_dx_lock_try(lck, key) \ ++ ext4_htree_node_lock(lck, key, EXT4_LB_DX, 0, NULL) ++#define ext4_htree_dx_unlock(lck) \ ++ ext4_htree_node_unlock(lck, EXT4_LB_DX, NULL) ++#define ext4_htree_dx_locked(lck) \ ++ ext4_htree_node_locked(lck, EXT4_LB_DX) ++ ++static void ext4_htree_dx_need_lock(struct htree_lock *lck) ++{ ++ struct ext4_dir_lock_data *ld; ++ ++ if (ext4_htree_safe_locked(lck)) ++ return; ++ ++ ld = ext4_htree_lock_data(lck); ++ switch (ld->ld_flags) { ++ default: ++ return; ++ case EXT4_HLOCK_LOOKUP: ++ ld->ld_flags = EXT4_HLOCK_LOOKUP_SAFE; ++ return; ++ case EXT4_HLOCK_DEL: ++ ld->ld_flags = EXT4_HLOCK_DEL_SAFE; ++ return; ++ case EXT4_HLOCK_ADD: ++ ld->ld_flags = EXT4_HLOCK_SPLIT; ++ return; ++ } ++} ++ ++#define ext4_htree_de_lock(lck, key) \ ++ ext4_htree_node_lock(lck, key, EXT4_LB_DE, 1, NULL) ++#define ext4_htree_de_unlock(lck) \ ++ ext4_htree_node_unlock(lck, EXT4_LB_DE, NULL) ++ ++#define ext4_htree_spin_lock(lck, key, event) \ ++ ext4_htree_node_lock(lck, key, EXT4_LB_SPIN, 0, event) ++#define ext4_htree_spin_unlock(lck) \ ++ ext4_htree_node_unlock(lck, EXT4_LB_SPIN, NULL) ++#define ext4_htree_spin_unlock_listen(lck, p) \ ++ ext4_htree_node_unlock(lck, EXT4_LB_SPIN, p) ++ ++static void ext4_htree_spin_stop_listen(struct htree_lock *lck) ++{ ++ if (!ext4_htree_safe_locked(lck) && ++ htree_node_is_listening(lck, ffz(~EXT4_LB_SPIN))) ++ htree_node_stop_listen(lck, ffz(~EXT4_LB_SPIN)); ++} ++ ++enum { ++ DX_HASH_COL_IGNORE, /* ignore collision while probing frames */ ++ DX_HASH_COL_YES, /* there is collision and it does matter */ ++ DX_HASH_COL_NO, /* there is no collision */ ++}; ++ ++static int dx_probe_hash_collision(struct htree_lock *lck, ++ struct dx_entry *entries, ++ struct dx_entry *at, u32 hash) ++{ ++ if (!(lck && ext4_htree_lock_data(lck)->ld_flags & EXT4_LB_EXACT)) { ++ return DX_HASH_COL_IGNORE; /* don't care about collision */ ++ ++ } else if (at == entries + dx_get_count(entries) - 1) { ++ return DX_HASH_COL_IGNORE; /* not in any leaf of this DX */ ++ ++ } else { /* hash collision? */ ++ return ((dx_get_hash(at + 1) & ~1) == hash) ? ++ DX_HASH_COL_YES : DX_HASH_COL_NO; ++ } ++} ++ + /* + * Probe for a directory leaf block to search. + * +@@ -808,10 +1042,11 @@ static inline void htree_rep_invariant_check(struct dx_entry *at, + */ + static struct dx_frame * + dx_probe(struct ext4_filename *fname, struct inode *dir, +- struct dx_hash_info *hinfo, struct dx_frame *frame_in) ++ struct dx_hash_info *hinfo, struct dx_frame *frame_in, ++ struct htree_lock *lck) + { + unsigned count, indirect, level, i; +- struct dx_entry *at, *entries, *p, *q, *m; ++ struct dx_entry *at, *entries, *p, *q, *m, *dx = NULL; + struct dx_root_info *info; + struct dx_frame *frame = frame_in; + struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR); +@@ -895,8 +1130,16 @@ dx_probe(struct ext4_filename *fname, struct inode *dir, + level = 0; + blocks[0] = 0; + while (1) { ++ if (indirect == level) { /* 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)) { ++ ext4_htree_spin_unlock(lck); /* release spin */ + ext4_warning_inode(dir, + "dx entry: count %u beyond limit %u", + count, dx_get_limit(entries)); +@@ -923,6 +1166,74 @@ dx_probe(struct ext4_filename *fname, struct inode *dir, + frame->entries = entries; + frame->at = at; + ++ if (indirect == level) { /* 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; ++ + block = dx_get_block(at); + for (i = 0; i <= level; i++) { + if (blocks[i] == block) { +@@ -932,8 +1243,7 @@ dx_probe(struct ext4_filename *fname, struct inode *dir, + goto fail; + } + } +- if (++level > indirect) +- return frame; ++ ++level; + blocks[level] = block; + frame++; + frame->bh = ext4_read_dirblock(dir, block, INDEX); +@@ -1004,7 +1314,7 @@ static void dx_release(struct dx_frame *frames, struct inode *dir) + 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; +@@ -1019,12 +1329,22 @@ static int ext4_htree_next_block(struct inode *dir, __u32 hash, + * 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--; + } + +@@ -1047,6 +1367,13 @@ static int ext4_htree_next_block(struct inode *dir, __u32 hash, + * 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); +@@ -1055,6 +1382,7 @@ static int ext4_htree_next_block(struct inode *dir, __u32 hash, + p->bh = bh; + p->at = p->entries = ((struct dx_node *) bh->b_data)->entries; + } ++ ext4_htree_de_lock(lck, p->at); + return 1; + } + +@@ -1216,10 +1544,10 @@ int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash, + } + hinfo.hash = start_hash; + hinfo.minor_hash = 0; +- frame = dx_probe(NULL, dir, &hinfo, frames); ++ /* assume it's PR locked */ ++ frame = dx_probe(NULL, dir, &hinfo, frames, NULL); + if (IS_ERR(frame)) + return PTR_ERR(frame); +- + /* Add '.' and '..' from the htree header */ + if (!start_hash && !start_minor_hash) { + de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data; +@@ -1259,7 +1587,7 @@ int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash, + 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; +@@ -1579,7 +1907,7 @@ static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block, + static struct buffer_head *__ext4_find_entry(struct inode *dir, + struct ext4_filename *fname, + struct ext4_dir_entry_2 **res_dir, +- int *inlined) ++ int *inlined, struct htree_lock *lck) + { + struct super_block *sb; + struct buffer_head *bh_use[NAMEI_RA_SIZE]; +@@ -1621,7 +1949,7 @@ static struct buffer_head *__ext4_find_entry(struct inode *dir, + goto restart; + } + if (is_dx(dir)) { +- ret = ext4_dx_find_entry(dir, fname, res_dir); ++ ret = ext4_dx_find_entry(dir, fname, res_dir, lck); + /* + * On success, or if the error was file not found, + * return. Otherwise, fall back to doing a search the +@@ -1631,6 +1959,7 @@ static struct buffer_head *__ext4_find_entry(struct inode *dir, + goto cleanup_and_exit; + dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, " + "falling back\n")); ++ ext4_htree_safe_relock(lck); + ret = NULL; + } + nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb); +@@ -1721,10 +2050,10 @@ cleanup_and_exit: + return ret; + } + +-static struct buffer_head *ext4_find_entry(struct inode *dir, ++struct buffer_head *ext4_find_entry_locked(struct inode *dir, + const struct qstr *d_name, + struct ext4_dir_entry_2 **res_dir, +- int *inlined) ++ int *inlined, struct htree_lock *lck) + { + int err; + struct ext4_filename fname; +@@ -1736,12 +2065,14 @@ static struct buffer_head *ext4_find_entry(struct inode *dir, + if (err) + return ERR_PTR(err); + +- bh = __ext4_find_entry(dir, &fname, res_dir, inlined); ++ bh = __ext4_find_entry(dir, &fname, res_dir, inlined, lck); + + ext4_fname_free_filename(&fname); + return bh; + } + ++EXPORT_SYMBOL(ext4_find_entry_locked); ++ + static struct buffer_head *ext4_lookup_entry(struct inode *dir, + struct dentry *dentry, + struct ext4_dir_entry_2 **res_dir) +@@ -1757,7 +2088,7 @@ static struct buffer_head *ext4_lookup_entry(struct inode *dir, + if (err) + return ERR_PTR(err); + +- bh = __ext4_find_entry(dir, &fname, res_dir, NULL); ++ bh = __ext4_find_entry(dir, &fname, res_dir, NULL, NULL); + + ext4_fname_free_filename(&fname); + return bh; +@@ -1765,7 +2096,8 @@ static struct buffer_head *ext4_lookup_entry(struct inode *dir, + + static struct buffer_head * ext4_dx_find_entry(struct inode *dir, + struct ext4_filename *fname, +- struct ext4_dir_entry_2 **res_dir) ++ struct ext4_dir_entry_2 **res_dir, ++ struct htree_lock *lck) + { + struct super_block * sb = dir->i_sb; + struct dx_frame frames[EXT4_HTREE_LEVEL], *frame; +@@ -1776,7 +2108,7 @@ static struct buffer_head * ext4_dx_find_entry(struct inode *dir, + #ifdef CONFIG_FS_ENCRYPTION + *res_dir = NULL; + #endif +- frame = dx_probe(fname, dir, NULL, frames); ++ frame = dx_probe(fname, dir, NULL, frames, lck); + if (IS_ERR(frame)) + return (struct buffer_head *) frame; + do { +@@ -1798,7 +2130,7 @@ static struct buffer_head * ext4_dx_find_entry(struct inode *dir, + + /* Check to see if we should continue to search */ + retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame, +- frames, NULL); ++ frames, NULL, lck); + if (retval < 0) { + ext4_warning_inode(dir, + "error %d reading directory index block", +@@ -1987,8 +2319,9 @@ static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base, + * Returns pointer to de in block into which the new entry will be inserted. + */ + static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir, +- struct buffer_head **bh,struct dx_frame *frame, +- struct dx_hash_info *hinfo) ++ struct buffer_head **bh, struct dx_frame *frames, ++ struct dx_frame *frame, struct dx_hash_info *hinfo, ++ struct htree_lock *lck) + { + unsigned blocksize = dir->i_sb->s_blocksize; + unsigned continued; +@@ -2065,8 +2398,14 @@ static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir, + hash2, split, count-split)); + + /* Fancy dance to stay within two buffers */ +- de2 = dx_move_dirents(dir, data1, data2, map + split, count - split, +- blocksize); ++ if (hinfo->hash < hash2) { ++ de2 = dx_move_dirents(dir, 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(dir, data1, data2, map, split, blocksize); ++ } + de = dx_pack_dirents(dir, data1, blocksize); + de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) - + (char *) de, +@@ -2084,12 +2423,21 @@ static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir, + dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2, + blocksize, 1)); + +- /* Which block gets the new entry? */ +- if (hinfo->hash >= hash2) { +- swap(*bh, bh2); +- de = de2; ++ ext4_htree_spin_lock(lck, frame > frames ? (frame - 1)->at : NULL, ++ frame->at); /* notify block is being split */ ++ if (hinfo->hash < hash2) { ++ dx_insert_block(frame, hash2 + continued, newblock); ++ ++ } else { ++ /* switch block number */ ++ dx_insert_block(frame, hash2 + continued, ++ dx_get_block(frame->at)); ++ dx_set_block(frame->at, newblock); ++ (frame->at)++; + } +- dx_insert_block(frame, hash2 + continued, newblock); ++ ext4_htree_spin_unlock(lck); ++ ext4_htree_dx_unlock(lck); ++ + err = ext4_handle_dirty_dirblock(handle, dir, bh2); + if (err) + goto journal_error; +@@ -2388,7 +2736,7 @@ static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname, + if (retval) + goto out_frames; + +- de = do_split(handle,dir, &bh2, frame, &fname->hinfo); ++ de = do_split(handle, dir, &bh2, frames, frame, &fname->hinfo, NULL); + if (IS_ERR(de)) { + retval = PTR_ERR(de); + goto out_frames; +@@ -2498,8 +2846,8 @@ out: + * may not sleep between calling this and putting something into + * the entry, as someone else might have used it while you slept. + */ +-static int ext4_add_entry(handle_t *handle, struct dentry *dentry, +- struct inode *inode) ++int ext4_add_entry_locked(handle_t *handle, struct dentry *dentry, ++ struct inode *inode, struct htree_lock *lck) + { + struct inode *dir = d_inode(dentry->d_parent); + struct buffer_head *bh = NULL; +@@ -2548,9 +2896,10 @@ static int ext4_add_entry(handle_t *handle, struct dentry *dentry, + if (dentry->d_name.len == 2 && + memcmp(dentry->d_name.name, "..", 2) == 0) + return ext4_update_dotdot(handle, dentry, inode); +- retval = ext4_dx_add_entry(handle, &fname, dir, inode); ++ retval = ext4_dx_add_entry(handle, &fname, dir, inode, lck); + if (!retval || (retval != ERR_BAD_DX_DIR)) + goto out; ++ ext4_htree_safe_relock(lck); + /* Can we just ignore htree data? */ + if (ext4_has_metadata_csum(sb)) { + EXT4_ERROR_INODE(dir, +@@ -2613,12 +2962,14 @@ out: + ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY); + return retval; + } ++EXPORT_SYMBOL(ext4_add_entry_locked); + + /* + * Returns 0 for success, or a negative error value + */ + static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname, +- struct inode *dir, struct inode *inode) ++ struct inode *dir, struct inode *inode, ++ struct htree_lock *lck) + { + struct dx_frame frames[EXT4_HTREE_LEVEL], *frame; + struct dx_entry *entries, *at; +@@ -2630,7 +2981,7 @@ static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname, + + again: + restart = 0; +- frame = dx_probe(fname, dir, NULL, frames); ++ frame = dx_probe(fname, dir, NULL, frames, lck); + if (IS_ERR(frame)) + return PTR_ERR(frame); + entries = frame->entries; +@@ -2665,6 +3016,12 @@ again: + struct dx_node *node2; + struct buffer_head *bh2; + ++ if (!ext4_htree_safe_locked(lck)) { /* retry with EX lock */ ++ ext4_htree_safe_relock(lck); ++ restart = 1; ++ goto cleanup; ++ } ++ + while (frame > frames) { + if (dx_get_count((frame - 1)->entries) < + dx_get_limit((frame - 1)->entries)) { +@@ -2768,8 +3125,32 @@ again: + restart = 1; + goto journal_error; + } ++ } else if (!ext4_htree_dx_locked(lck)) { ++ struct ext4_dir_lock_data *ld = ext4_htree_lock_data(lck); ++ ++ /* not well protected, require DX lock */ ++ ext4_htree_dx_need_lock(lck); ++ at = frame > frames ? (frame - 1)->at : NULL; ++ ++ /* NB: no risk of deadlock because it's just a try. ++ * ++ * NB: we check ld_count for twice, the first time before ++ * having DX lock, the second time after holding DX lock. ++ * ++ * NB: We never free blocks for directory so far, which ++ * means value returned by dx_get_count() should equal to ++ * ld->ld_count if nobody split any DE-block under @at, ++ * and ld->ld_at still points to valid dx_entry. */ ++ if ((ld->ld_count != dx_get_count(entries)) || ++ !ext4_htree_dx_lock_try(lck, at) || ++ (ld->ld_count != dx_get_count(entries))) { ++ restart = 1; ++ goto cleanup; ++ } ++ /* OK, I've got DX lock and nothing changed */ ++ frame->at = ld->ld_at; + } +- de = do_split(handle, dir, &bh, frame, &fname->hinfo); ++ de = do_split(handle, dir, &bh, frames, frame, &fname->hinfo, lck); + if (IS_ERR(de)) { + err = PTR_ERR(de); + goto cleanup; +@@ -2780,6 +3161,8 @@ again: + journal_error: + ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */ + cleanup: ++ ext4_htree_dx_unlock(lck); ++ ext4_htree_de_unlock(lck); + brelse(bh); + dx_release(frames, dir); + /* @restart is true means htree-path has been changed, we need to +diff --git a/fs/ext4/super.c b/fs/ext4/super.c +index b271e1f..cda88f8 100644 +--- a/fs/ext4/super.c ++++ b/fs/ext4/super.c +@@ -1320,6 +1320,7 @@ static struct inode *ext4_alloc_inode(struct super_block *sb) + + inode_set_iversion(&ei->vfs_inode, 1); + spin_lock_init(&ei->i_raw_lock); ++ sema_init(&ei->i_append_sem, 1); + INIT_LIST_HEAD(&ei->i_prealloc_list); + atomic_set(&ei->i_prealloc_active, 0); + spin_lock_init(&ei->i_prealloc_lock); +-- +2.34.1 + --- /dev/null +++ b/ldiskfs/kernel_patches/series/ldiskfs-5.14-rhel9.1.series @@ -0,0 +1,33 @@ +linux-5.16/ext4-inode-version.patch +linux-5.4/ext4-lookup-dotdot.patch +linux-5.14/ext4-print-inum-in-htree-warning.patch +linux-5.18/ext4-prealloc.patch +linux-5.16/ext4-osd-iop-common.patch +linux-5.16/ext4-misc.patch +linux-5.18/ext4-mballoc-extra-checks.patch +sles15sp4/ext4-hash-indexed-dir-dotdot-update.patch +linux-5.14/ext4-kill-dx-root.patch +linux-5.18/ext4-mballoc-pa-free-mismatch.patch +linux-5.18/ext4-data-in-dirent.patch +rhel8/ext4-nocmtime.patch +base/ext4-htree-lock.patch +rhel9.1/ext4-pdirop.patch +rhel9/ext4-max-dir-size.patch +linux-5.16/ext4-corrupted-inode-block-bitmaps-handling-patches.patch +rhel9/ext4-give-warning-with-dir-htree-growing.patch +ubuntu18/ext4-jcb-optimization.patch +linux-5.10/ext4-attach-jinode-in-writepages.patch +rhel8/ext4-dont-check-before-replay.patch +rhel7.6/ext4-use-GFP_NOFS-in-ext4_inode_attach_jinode.patch +rhel7.6/ext4-export-orphan-add.patch +linux-5.18/ext4-export-mb-stream-allocator-variables.patch +ubuntu19/ext4-iget-with-flags.patch +linux-5.14/export-ext4fs-dirhash-helper.patch +linux-5.8/ext4-no-max-dir-size-limit-for-iam-objects.patch +linux-5.14/ext4-ialloc-uid-gid-and-pass-owner-down.patch +linux-5.14/ext4-projid-xattrs.patch +base/ext4-delayed-iput.patch +linux-5.14/ext4-xattr-disable-credits-check.patch +linux-5.10/ext4-fiemap-kernel-data.patch +rhel8/ext4-old_ea_inodes_handling_fix.patch +rhel9.1/ext4-enc-flag.patch
Locations
Projects
Search
Status Monitor
Help
OpenBuildService.org
Documentation
API Documentation
Code of Conduct
Contact
Support
@OBShq
Terms
openSUSE Build Service is sponsored by
The Open Build Service is an
openSUSE project
.
Sign Up
Log In
Places
Places
All Projects
Status Monitor