erofs-utils: introduce extra xattr name prefix
cover erofs-utils: introduce extra xattr name prefix
As discussed in [1], erofs paring with overlayfs could satisfy the scenario which originally composefs targets. In this case, each file of erofs is tagged with overlay.metacopy and overlay.redirect xattr, which can be consumed by overlayfs.
The example xattr arrangement is like:
user.overlay.metacopy="" user.overlay.redirect="/d6/a6748c13e2ef146a9587c47cd51ef692a368b180fc3d32b1ee32df4bb99133"
In the current implementation of erofs, "user.overlay.redirect" xattr name will consume duplicate on-disk space. In other words, each file will allocate on-disk space for its "user.overlay.redirect" xattr name, which explodes the output erofs image as a consequence.
This patch set introduces extra xattr name prefix feature, in which user could specify customised xattr name prefix through mkfs.erofs. When matched with a user specified extra xattr name prefix, only the trailing part of the xattr name apart from the xattr name prefix will be stored on disk. Please refer to patch 3 for more details of the on-disk format.
This feature can significantly reduces the size of the output erofs image in the scenario where files share the same xattr name or prefix heavily.
Below is the test result of the size of the output erofs image with different option combinations of mkfs.erofs. This is tested on the rootfs givin in [2].
``` 4.2M large.erofs.T0.noxattr 7.4M large.erofs.T0.xattr 6.4M large.erofs.T0.xattr.share 5.7M large.erofs.noxattr 8.9M large.erofs.xattr 7.8M large.erofs.xattr.share ```
T0: "-T0" of mkfs.erofs, w/ this option, 32 bytes on-disk inode is used; wo/ this option, 64 byte on-disk inode is used instead
xattr: no extra option specified
noxattr: "-x -1", i.e. disable xattr
share: "--xattr-prefix=user.overlay.metacopy" and "--xattr-prefix=user.overlay.redirect" option of mkfs.erofs. w/ this option, the extra xattr name prefix feature is enabled.
It can be seen ~10% disk space is saved with this feature in the typical workload.
patch 1-2 are preparing patch.
[1] https://lore.kernel.org/all/CAOQ4uxgGc33_QVBXMbQTnmbpHio4amv=W7ax2vQ1UMet0k_KoA@mail.gmail.com/
[2] https://my.owndrive.com/index.php/s/irHJXRpZHtT3a5i
Jingbo Xu (6):
erofs-utils: declare prefix_len as u8
erofs-utils: extract packedfile API
erofs-utils: introduce on-disk format for extra xattr name prefix
erofs-utils: introduce init/cleanup routine for extra xattr name prefix
erofs-utils: build erofs_xattr_entry upon extra xattr name prefix
erofs-utils: mkfs.erofs: introduce --xattr-prefix option
include/erofs/config.h | 2 + include/erofs/fragments.h | 19 +++++-- include/erofs/inode.h | 2 + include/erofs/internal.h | 3 ++ include/erofs/xattr.h | 3 ++ include/erofs_fs.h | 7 ++- lib/fragments.c | 20 +++----- lib/xattr.c | 104 +++++++++++++++++++++++++++++++++++++- mkfs/main.c | 48 +++++++++++++++--- 9 files changed, 183 insertions(+), 25 deletions(-)
Signature · 1 line
2.19.1.6.gb485710b
PATCH 1/6 erofs-utils: declare prefix_len as u8
The on-disk erofs_xattr_entry.e_name_len is declared as u8, which implies the maximum length of the xattr name.
Let's also declare xattr_prefix.prefix_len as u8 to keep sync with the on-disk structure.
--- lib/xattr.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
lib/xattr.c +1 -1
@@ -41,7 +41,7 @@ static unsigned int shared_xattrs_count, shared_xattrs_size;static struct xattr_prefix {const char *prefix; -u16 prefix_len; +u8 prefix_len; } xattr_types[] = { [EROFS_XATTR_INDEX_USER ] = {XATTR_USER_PREFIX ,
Signature · 1 line
2.19.1.6.gb485710b
PATCH 2/6 erofs-utils: extract packedfile API
In reply to Jingbo XuLater packed_inode will be used in more scenarios other than fragments, thus extract the separate packedfile API for future use.
--- include/erofs/config.h | 1 + include/erofs/fragments.h | 19 ++++++++++++++++--- include/erofs/inode.h | 2 ++ lib/fragments.c | 20 ++++++++------------ mkfs/main.c | 32 ++++++++++++++++++++++++-------- 5 files changed, 51 insertions(+), 23 deletions(-)
include/erofs/config.h +1 -0
@@ -52,6 +52,7 @@ struct erofs_configure {bool c_dedupe;bool c_ignore_mtime;bool c_showprogress; +bool c_packedfile;#ifdef HAVE_LIBSELINUX struct selabel_handle *sehnd;
include/erofs/fragments.h +16 -3
@@ -12,15 +12,28 @@ extern "C"#include "erofs/internal.h" -extern const char *frags_packedname; -#define EROFS_PACKED_INODE frags_packedname +extern FILE *packedfile; +int erofs_packedfile_init (void ); +void erofs_packedfile_exit (void ); +struct erofs_inode *erofs_mkfs_build_packedfile (void ); + +static inline int erofs_packedfile_is_empty (void ) +{ +#ifdef HAVE_FTELLO64 +off64_t offset =ftello64 (packedfile); +#else +off_t offset =ftello (packedfile); +#endif +if (offset <0 ) +return -errno; +return offset ==0 ; +}int z_erofs_fragments_dedupe (struct erofs_inode *inode,int fd,u32 *tofcrc);int z_erofs_pack_file_from_fd (struct erofs_inode *inode,int fd,u32 tofcrc);int z_erofs_pack_fragments (struct erofs_inode *inode,void *data,unsigned int len,u32 tofcrc);void z_erofs_fragments_commit (struct erofs_inode *inode); -struct erofs_inode *erofs_mkfs_build_fragments (void );int erofs_fragments_init (void );void erofs_fragments_exit (void );
include/erofs/inode.h +2 -0
@@ -21,6 +21,8 @@ void erofs_inode_manager_init(void);unsigned int erofs_iput (struct erofs_inode *inode);erofs_nid_t erofs_lookupnid (struct erofs_inode *inode);struct erofs_inode *erofs_mkfs_build_tree_from_path (const char *path); + +#define EROFS_PACKED_INODE ("packed_file")struct erofs_inode *erofs_mkfs_build_special_from_fd (int fd,const char *name);#ifdef __cplusplus
lib/fragments.c +8 -12
@@ -39,8 +39,7 @@ struct erofs_fragment_dedupe_item {static struct list_head dupli_frags[FRAGMENT_HASHSIZE ]; -static FILE *packedfile; -const char *frags_packedname ="packed_file" ; +FILE *packedfile;#ifndef HAVE_LSEEK64 #define erofs_lseek64 lseek
@@ -195,15 +194,16 @@ static int z_erofs_fragments_dedupe_insert(void *data, unsigned int len,return 0 ; } -static void z_erofs_fragments_dedupe_init (void ) +int erofs_fragments_init (void ) {unsigned int i;for (i =0 ; i <FRAGMENT_HASHSIZE ; ++i)init_list_head (&dupli_frags[i]); +return 0 ; } -static void z_erofs_fragments_dedupe_exit (void ) +void erofs_fragments_exit (void ) {struct erofs_fragment_dedupe_item *di, *n;struct list_head *head;
@@ -324,23 +324,21 @@ int z_erofs_pack_fragments(struct erofs_inode *inode, void *data,return len; } -struct erofs_inode *erofs_mkfs_build_fragments (void ) +struct erofs_inode *erofs_mkfs_build_packedfile (void ) {fflush (packedfile);return erofs_mkfs_build_special_from_fd (fileno (packedfile), - frags_packedname); +EROFS_PACKED_INODE ); } -void erofs_fragments_exit (void ) +void erofs_packedfile_exit (void ) {if (packedfile)fclose (packedfile); - -z_erofs_fragments_dedupe_exit (); } -int erofs_fragments_init (void ) +int erofs_packedfile_init (void ) {#ifdef HAVE_TMPFILE64 packedfile =tmpfile64 ();
@@ -349,7 +347,5 @@ int erofs_fragments_init(void)#endif if (!packedfile)return -ENOMEM ; - -z_erofs_fragments_dedupe_init ();return 0 ; }
mkfs/main.c +24 -8
@@ -555,6 +555,8 @@ static int mkfs_parse_options_cfg(int argc, char *argv[]) } cfg.c_pclusterblks_packed = pclustersize_packed >> sbi.blkszbits; } +if (cfg.c_fragments) + cfg.c_packedfile = true;return 0 ; }
@@ -770,10 +772,17 @@ int main(int argc, char **argv) }#endif erofs_show_config (); -if (cfg.c_fragments) { +if (cfg.c_packedfile) {if (!cfg.c_pclusterblks_packed) cfg.c_pclusterblks_packed = cfg.c_pclusterblks_def; + err =erofs_packedfile_init (); +if (err) { +erofs_err ("failed to initialize packedfile" ); +return 1 ; + } + } +if (cfg.c_fragments) { err =erofs_fragments_init ();if (err) {erofs_err ("failed to initialize fragments" );
@@ -880,15 +889,20 @@ int main(int argc, char **argv) } packed_nid =0 ; -if (cfg.c_fragments &&erofs_sb_has_fragments ()) { -erofs_update_progressinfo ("Handling packed_file ..." ); - packed_inode =erofs_mkfs_build_fragments (); -if (IS_ERR (packed_inode)) { - err =PTR_ERR (packed_inode); +if (cfg.c_packedfile) { + err =erofs_packedfile_is_empty (); +if (err <0 ) goto exit; +if (!err) { +erofs_update_progressinfo ("Handling packed_file ..." ); + packed_inode =erofs_mkfs_build_packedfile (); +if (IS_ERR (packed_inode)) { + err =PTR_ERR (packed_inode); + goto exit; + } + packed_nid =erofs_lookupnid (packed_inode); +erofs_iput (packed_inode); } - packed_nid =erofs_lookupnid (packed_inode); -erofs_iput (packed_inode); } err = erofs_mkfs_update_super_block(sb_bh, root_nid, &nblocks,
@@ -917,6 +931,8 @@ exit:erofs_blob_exit ();if (cfg.c_fragments)erofs_fragments_exit (); +if (cfg.c_packedfile) +erofs_packedfile_exit ();erofs_exit_configure ();if (err) {
Signature · 1 line
2.19.1.6.gb485710b
PATCH 3/6 erofs-utils: introduce on-disk format for extra xattr name prefix
In reply to Jingbo XuBesides the predefined xattr name prefixes, introduces extra xattr name prefixes, which works similarly as the predefined xattr name prefixees, except that they are user specified.
When matched with a user specified extra xattr name prefix, only the trailing part of the xattr name except the xattr name prefix will be stored in erofs_xattr_entry.e_name. e_name is empty if the xattr name matches exactly as the extra xattr name prefix.
All extra xattr name prefixes are stored in the packed file. For each extra xattr name prefix, the on-disk format in packed file can be shown as (u8 prefix_len, prefix string), where the first u8 represents the length of the prefix string (excluding '\0'), followed by the content of the prefix string (including '\0').
Two fields are introduced to the on-disk superblock, where ea_prefix_count represents the total number of the extra xattr name prefixes recorded in the packed file, and ea_prefix_off represents the start offset of recorded name prefixes in the packed file.
When referring to an extra xattr name prefix, the highest bit (bit 7) of erofs_xattr_entry.e_name_index is set, while the lower bits (bit 0-6) as a whole represents the index of the referred name prefix among all extra xattr name prefixes in the packed file.
--- include/erofs_fs.h | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)
include/erofs_fs.h +6 -1
@@ -77,7 +77,9 @@ struct erofs_super_block { }__packed u1;__le16 extra_devices;/* # of devices besides the primary device */ __le16 devt_slotoff;/* startoff = devt_slotoff * devt_slotsize */ -__u8 reserved[6 ]; +__u8 reserved; +__u8 ea_prefix_count;/* # of extra xattr name prefix */ +__le32 ea_prefix_off;/* start offset of extra prefixes in packed inode */ __le64 packed_nid;/* nid of the special packed inode */ __u8 reserved2[24 ]; };
@@ -228,6 +230,9 @@ struct erofs_xattr_ibody_header {#define EROFS_XATTR_INDEX_LUSTRE 5#define EROFS_XATTR_INDEX_SECURITY 6 +#define EROFS_XATTR_EA_FLAG (1 << 7) +#define EROFS_XATTR_EA_MASK (EROFS_XATTR_EINDEX_FLAG - 1) +/* xattr entry (for both inline & shared xattrs) */ struct erofs_xattr_entry {__u8 e_name_len;/* length of name */
Signature · 1 line
2.19.1.6.gb485710b
Re: [PATCH 3/6] erofs-utils: introduce on-disk format for extra xattr name prefix
Quoted 3 lines · Jingbo Xu
Besides the predefined xattr name prefixes, introduces extra xattr name prefixes, which works similarly as the predefined xattr name prefixees, except that they are user specified.
Besides the built-in xattr name prefixes, let's introduce user-defined extra xattr name prefixes, which works similarly as the pre-defined xattr name prefixees, except that they are user specified.
Quoted 5 lines
When matched with a user specified extra xattr name prefix, only the trailing part of the xattr name except the xattr name prefix will be stored in erofs_xattr_entry.e_name.
e_nameis empty if the xattr name matches exactly as the extra xattr name prefix.
When a user-specified extra xattr name prefix is matched, only the trailing part of the xattr name which already eliminates the given prefix will be stored in erofs_xattr_entry.e_name. e_name can be empty if the xattr full name matches exactly as the name prefix.
Quoted 6 lines
All extra xattr name prefixes are stored in the packed file. For each extra xattr name prefix, the on-disk format in packed file can be shown as (u8
prefix_len, prefix string), where the first u8 represents the length of the prefix string (excluding '\0'), followed by the content of the prefix string (including '\0').
Why we have to introduce a prefix_len here?
Quoted 10 lines
Two fields are introduced to the on-disk superblock, where
ea_prefix_countrepresents the total number of the extra xattr name prefixes recorded in the packed file, andea_prefix_offrepresents the start offset of recorded name prefixes in the packed file.When referring to an extra xattr name prefix, the highest bit (bit 7) of erofs_xattr_entry.e_name_index is set, while the lower bits (bit 0-6) as a whole represents the index of the referred name prefix among all extra xattr name prefixes in the packed file.
When an extra xattr name prefix is referred, the highest bit (bit 7) of erofs_xattr_entry.e_name_index is set, while the lower bits (bit 0-6) as a whole represents the index of the referred name prefix among all extra xattr name prefixes in the packed file.
Quoted 22 lines
Signed-off-by Jingbo Xu <[email protected]>--- include/erofs_fs.h | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)include/erofs_fs.h +3
-1@@ -77,7 +77,9 @@ struct erofs_super_block { }__packed u1;__le16 extra_devices;/* # of devices besides the primary device */ __le16 devt_slotoff;/* startoff = devt_slotoff * devt_slotsize */ -__u8 reserved[6 ]; +__u8 reserved; +__u8 ea_prefix_count;/* # of extra xattr name prefix */ +__le32 ea_prefix_off;/* start offset of extra prefixes in packed inode */ __le64 packed_nid;/* nid of the special packed inode */ __u8 reserved2[24 ]; };
include/erofs_fs.h struct erofs_xattr_ibody_header { · line 234
Quoted 5 lines
include/erofs_fs.h +2
-0@@ -228,6 +230,9 @@ struct erofs_xattr_ibody_header {#define EROFS_XATTR_INDEX_LUSTRE 5#define EROFS_XATTR_INDEX_SECURITY 6 +#define EROFS_XATTR_EA_FLAG (1 << 7) +#define EROFS_XATTR_EA_MASK (EROFS_XATTR_EINDEX_FLAG - 1)
EROFS_EXTRA_XATTR_FLAG 0x80EROFS_EXTRA_XATTR_MASK 0x7F
?
Thanks,
Gao Xiang
Quoted 4 lines
include/erofs_fs.h +1
-0+/* xattr entry (for both inline & shared xattrs) */ struct erofs_xattr_entry {__u8 e_name_len;/* length of name */
PATCH 4/6 erofs-utils: introduce init/cleanup routine for extra xattr name prefix
In reply to Jingbo XuIntroduce helpers for registering and cleaning up extra xattr name prefix.
--- include/erofs/xattr.h | 3 +++ lib/xattr.c | 58 +++++++++++++++++++++++++++++++++++++++++++ 2 files changed, 61 insertions(+)
include/erofs/xattr.h +3 -0
@@ -70,6 +70,9 @@ int erofs_prepare_xattr_ibody(struct erofs_inode *inode);char *erofs_export_xattr_ibody (struct list_head *ixattrs,unsigned int size);int erofs_build_shared_xattrs_from_path (const char *path); +int erofs_insert_ea_type (const char *prefix); +void erofs_cleanup_ea_type (void ); + #ifdef __cplusplus }#endif
lib/xattr.c +58 -0
67 lines not on this page · raw message
Signature · 1 line
2.19.1.6.gb485710b
PATCH 5/6 erofs-utils: build erofs_xattr_entry upon extra xattr name prefix
In reply to Jingbo XuBuild the xattr entry also considering the extra xattr name prefix. The user specified extra xattr name prefix takes precedence over the predefined xattr name prefix.
--- include/erofs/internal.h | 3 +++ lib/xattr.c | 44 ++++++++++++++++++++++++++++++++++++++++ mkfs/main.c | 2 ++ 3 files changed, 49 insertions(+)
include/erofs/internal.h +3 -0
@@ -102,6 +102,9 @@ struct erofs_sb_info {u16 device_id_mask;/* used for others */ };erofs_nid_t packed_nid; + +u32 ea_prefix_off; +u8 ea_prefix_count; };/* make sure that any user of the erofs headers has atleast 64bit off_t type */
lib/xattr.c +44 -0
69 lines not on this page · raw message
mkfs/main.c +2 -0
@@ -574,6 +574,8 @@ int erofs_mkfs_update_super_block(struct erofs_buffer_head *bh, .blocks = 0, .meta_blkaddr = sbi.meta_blkaddr, .xattr_blkaddr = sbi.xattr_blkaddr, + .ea_prefix_count = sbi.ea_prefix_count, + .ea_prefix_off = cpu_to_le32(sbi.ea_prefix_off), .feature_incompat = cpu_to_le32(sbi.feature_incompat), .feature_compat =cpu_to_le32 (sbi.feature_compat & ~EROFS_FEATURE_COMPAT_SB_CHKSUM),
Signature · 1 line
2.19.1.6.gb485710b
PATCH 6/6 erofs-utils: mkfs.erofs: introduce --xattr-prefix option
In reply to Jingbo XuIntroduce --xattr-prefix option to make user capable of specifying customised extra xattr name prefix.
--- include/erofs/config.h | 1 + mkfs/main.c | 16 +++++++++++++++- 2 files changed, 16 insertions(+), 1 deletion(-)
include/erofs/config.h +1 -0
@@ -53,6 +53,7 @@ struct erofs_configure {bool c_ignore_mtime;bool c_showprogress;bool c_packedfile; +bool c_ea_prefix;#ifdef HAVE_LIBSELINUX struct selabel_handle *sehnd;
mkfs/main.c +15 -1
@@ -56,6 +56,7 @@ static struct option long_options[] = { {"preserve-mtime", no_argument, NULL, 15}, {"uid-offset", required_argument, NULL, 16}, {"gid-offset", required_argument, NULL, 17}, + {"xattr-prefix", required_argument, NULL, 19}, {"mount-point", required_argument, NULL, 512}, #ifdef WITH_ANDROID {"product-out", required_argument, NULL, 513},
@@ -116,6 +117,7 @@ static void usage(void)" --random-pclusterblks randomize pclusterblks for big pcluster (debugging only)\n" " --random-algorithms randomize per-file algorithms (debugging only)\n"#endif +" --xattr-prefix=X X=extra xattr name prefix\n" " --mount-point=X X=prefix of target fs path (default: /)\n" #ifdef WITH_ANDROID"\nwith following android-specific options:\n"
@@ -475,6 +477,16 @@ static int mkfs_parse_options_cfg(int argc, char *argv[])return -EINVAL ; }break ; +case 19 : + errno =0 ; + opt =erofs_insert_ea_type (optarg); +if (opt) { +erofs_err ("failed to parse extra xattr prefix: %s" , +erofs_strerror (opt)); +return opt; + } + cfg.c_ea_prefix = true; +break ;case 1 :usage ();exit (0 );
@@ -555,7 +567,7 @@ static int mkfs_parse_options_cfg(int argc, char *argv[]) } cfg.c_pclusterblks_packed = pclustersize_packed >> sbi.blkszbits; } -if (cfg.c_fragments) +if (cfg.c_fragments || cfg.c_ea_prefix) cfg.c_packedfile = true;return 0 ; }
@@ -935,6 +947,8 @@ exit:erofs_fragments_exit ();if (cfg.c_packedfile)erofs_packedfile_exit (); +if (cfg.c_ea_prefix) +erofs_cleanup_ea_type ();erofs_exit_configure ();if (err) {
Signature · 1 line
2.19.1.6.gb485710b
PATCH 1/1 tmp debug
In reply to Jingbo Xu--- lib/inode.c | 7 ++++++- lib/xattr.c | 3 +++ mkfs/main.c | 4 ++++ 3 files changed, 13 insertions(+), 1 deletion(-)
lib/inode.c +6 -1
@@ -272,8 +272,11 @@ erofs_nid_t erofs_lookupnid(struct erofs_inode *inode)struct erofs_buffer_head *const bh = inode->bh;erofs_off_t off, meta_offset; -if (!bh || (long long )inode->nid >0 ) +if (!bh || (long long )inode->nid >0 ) { +erofs_dbg (" nid %llu to file %s, bh %p" , + inode->nid, inode->i_srcpath, inode->bh);return inode->nid; + }erofs_mapbh (bh->block); off =erofs_btell (bh, false);
@@ -1266,6 +1269,8 @@ struct erofs_inode *erofs_mkfs_build_special_from_fd(int fd, const char *name)if (name ==EROFS_PACKED_INODE ) { sbi.packed_nid =EROFS_PACKED_NID_UNALLOCATED ; inode->nid = sbi.packed_nid; +erofs_dbg ("erofs_mkfs_build_special_from_fd: nid %llu to file %s, bh %p" , + inode->nid, inode->i_srcpath, inode->bh); } ret =erofs_write_compressed_file (inode, fd);
lib/xattr.c +3 -0
@@ -141,11 +141,14 @@ static bool match_prefix(const char *key, u8 *index, u16 *len)struct xattr_prefix *p;struct extra_xattr_type_node *tnode; +printf ("[match_prefix]: enter extra_xattr_types_count++ is %u\n" , extra_xattr_types_count);list_for_each_entry (tnode, &extra_xattr_types, list) { p = &tnode->type; +printf ("[iterating]: key %s, prefix %s\n" , key, p->prefix);if (p->prefix && !strncmp (p->prefix, key, p->prefix_len)) { *len = p->prefix_len; *index = tnode->index; +printf ("match_prefix: key %s, matches prefix %s, index %u\n" , key, p->prefix, tnode->index);return true; } }
mkfs/main.c +4 -0
@@ -911,13 +911,17 @@ int main(int argc, char **argv)erofs_update_progressinfo ("Handling packed_file ..." ); packed_inode =erofs_mkfs_build_packedfile ();if (IS_ERR (packed_inode)) { +printf ("[%s %d] erofs_mkfs_build_packedfile failed\n" , __func__, __LINE__); err =PTR_ERR (packed_inode); goto exit; } +printf ("[%s %d] start erofs_lookupnid...\n" , __func__, __LINE__); packed_nid =erofs_lookupnid (packed_inode); +printf ("[%s %d] packed_nid %lu\n" , __func__, __LINE__, packed_nid);erofs_iput (packed_inode); } } +printf ("[%s %d] packed_nid %lu\n" , __func__, __LINE__, packed_nid); err =erofs_mkfs_update_super_block (sb_bh, root_nid, &nblocks, packed_nid);
Signature · 1 line
2.19.1.6.gb485710b