Skip to contentSkip to conversation

erofs-utils: mkfs: enable directory data in the metadata zone

PATCH Superseded Gao Xiang 2 participants 3 messages Jan 16 – Jan 17 Library code

lore
Share & tools
lore

Gao Xiang

author
#0 lore raw

PATCH 1/1 erofs-utils: mkfs: enable directory data in the metadata zone

It allows directory data and inode metadata to be kept as close as possible, significantly improving metadata performance for long-latency remote image use cases.

Usage:

 $ mkfs.erofs --MZ foo.erofs foo/
---
 include/erofs/importer.h |  1 +
 include/erofs/inode.h    |  3 +-
 include/erofs/internal.h |  8 ++++
 lib/cache.c              |  4 ++
 lib/inode.c              | 98 ++++++++++++++++++++++++++++------------
 lib/io.c                 |  5 +-
 lib/metabox.c            | 22 +++++----
 lib/remotes/s3.c         |  3 +-
 lib/super.c              |  4 +-
 lib/tar.c                |  2 +-
 mkfs/main.c              | 27 +++++++++--
 11 files changed, 129 insertions(+), 48 deletions(-)
include/erofs/importer.h +1 -0
index a525b474f1d5..60160d6bea05 100644
@@ -46,6 +46,7 @@ struct erofs_importer_params {
 	bool no_datainline;
 	/* Issue directory data (except inline data) separately from regular inodes */
 	bool grouped_dirdata;
+	bool dirdata_in_metazone;
 	bool hard_dereference;
 	bool ovlfs_strip;
 	bool dot_omitted;
include/erofs/inode.h +2 -1
index 89bd16aecc06..ba62ece9a7cc 100644
@@ -38,7 +38,8 @@ erofs_nid_t erofs_lookupnid(struct erofs_inode *inode);
 int erofs_iflush(struct erofs_inode *inode);
 struct erofs_dentry *erofs_d_alloc(struct erofs_inode *parent,
 				   const char *name);
-int erofs_allocate_inode_bh_data(struct erofs_inode *inode, erofs_blk_t nblocks);
+int erofs_allocate_inode_bh_data(struct erofs_inode *inode, erofs_blk_t nblocks,
+				 bool in_metazone);
 bool erofs_dentry_is_wht(struct erofs_sb_info *sbi, struct erofs_dentry *d);
 int __erofs_fill_inode(struct erofs_importer *im, struct erofs_inode *inode,
 		       struct stat *st, const char *path);
include/erofs/internal.h +8 -0
index 5798f10e89c2..2fe4514b3d23 100644
@@ -157,6 +157,7 @@ struct erofs_sb_info {
 	struct erofs_buffer_head *bh_devt;
 	bool useqpl;
 	bool sb_valid;
+	u32 metazone_startblk;
 };
 
 /* make sure that any user of the erofs headers has atleast 64bit off_t type */
@@ -205,6 +206,8 @@ struct erofs_diskbuf;
 #define EROFS_INODE_DATA_SOURCE_DISKBUF		2
 #define EROFS_INODE_DATA_SOURCE_RESVSP		3
 
+#define EROFS_I_BLKADDR_DEV_ID_BIT		48
+
 struct erofs_inode {
 	struct list_head i_hash, i_subdirs, i_xattrs;
 
@@ -308,6 +311,11 @@ static inline bool erofs_inode_in_metabox(struct erofs_inode *inode)
 	return inode->nid >> EROFS_DIRENT_NID_METABOX_BIT;
 }
 
+static inline erofs_blk_t erofs_inode_dev_baddr(struct erofs_inode *inode)
+{
+	return inode->u.i_blkaddr & (BIT(EROFS_I_BLKADDR_DEV_ID_BIT) - 1);
+}
+
 static inline erofs_off_t erofs_iloc(struct erofs_inode *inode)
 {
 	struct erofs_sb_info *sbi = inode->sbi;
lib/cache.c +4 -0
index a87575ad74d1..f23dbb06264a 100644
@@ -479,6 +479,10 @@ static int __erofs_bflush(struct erofs_bufmgr *bmgr,
 
 			/* flush and remove bh */
 			ret = bh->op->flush(bh);
+			if (__erofs_unlikely(ret == -EBUSY && !forget)) {
+				skip = true;
+				continue;
+			}
 			if (ret < 0)
 				return ret;
 		}
lib/inode.c +68 -30

200 lines not on this page · raw message

lib/io.c +4 -1
index 37a74f63c45e..0c5eb2c29989 100644
@@ -571,10 +571,13 @@ ssize_t erofs_io_write(struct erofs_vfile *vf, void *buf, size_t len)
 
 off_t erofs_io_lseek(struct erofs_vfile *vf, u64 offset, int whence)
 {
+	off_t ret;
+
 	if (vf->ops)
 		return vf->ops->lseek(vf, offset, whence);
 
-	return lseek(vf->fd, offset, whence);
+	ret = lseek(vf->fd, offset, whence);
+	return ret < 0 ? -errno : ret;
 }
 
 ssize_t erofs_io_sendfile(struct erofs_vfile *vout, struct erofs_vfile *vin,
lib/metabox.c +14 -8
index 37267ddb73cf..d6abd5123cc8 100644
@@ -54,7 +54,7 @@ int erofs_metadata_init(struct erofs_sb_info *sbi)
 	struct erofs_metamgr *m2gr;
 	int ret;
 
-	if (!sbi->m2gr && sbi->meta_blkaddr == EROFS_META_NEW_ADDR) {
+	if (!sbi->m2gr && sbi->metazone_startblk == EROFS_META_NEW_ADDR) {
 		m2gr = malloc(sizeof(*m2gr));
 		if (!m2gr)
 			return -ENOMEM;
@@ -62,6 +62,8 @@ int erofs_metadata_init(struct erofs_sb_info *sbi)
 		if (ret)
 			goto err_free;
 		sbi->m2gr = m2gr;
+		/* FIXME: sbi->meta_blkaddr should be 0 for 48-bit layouts */
+		sbi->meta_blkaddr = EROFS_META_NEW_ADDR;
 	}
 
 	if (!sbi->mxgr && erofs_sb_has_metabox(sbi)) {
@@ -124,20 +126,24 @@ int erofs_metazone_flush(struct erofs_sb_info *sbi)
 
 	if (!m2gr)
 		return 0;
-	m2bgr = m2gr->bmgr;
+	bh = erofs_balloc(sbi->bmgr, DATA, 0, 0);
+	if (!bh)
+		return PTR_ERR(bh);
+	erofs_mapbh(NULL, bh->block);
+	pos_out = erofs_btell(bh, false);
+	meta_blkaddr = pos_out >> sbi->blkszbits;
+	sbi->metazone_startblk = meta_blkaddr;
 
+	m2bgr = m2gr->bmgr;
 	ret = erofs_bflush(m2bgr, NULL);
 	if (ret)
 		return ret;
 
 	length = erofs_mapbh(m2bgr, NULL) << sbi->blkszbits;
-	bh = erofs_balloc(sbi->bmgr, DATA, length, 0);
-	if (!bh)
-		return PTR_ERR(bh);
+	ret = erofs_bh_balloon(bh, length);
+	if (ret < 0)
+		return ret;
 
-	erofs_mapbh(NULL, bh->block);
-	pos_out = erofs_btell(bh, false);
-	meta_blkaddr = pos_out >> sbi->blkszbits;
 	do {
 		count = min_t(erofs_off_t, length, INT_MAX);
 		ret = erofs_io_xcopy(sbi->bmgr->vf, pos_out,
lib/remotes/s3.c +2 -1
index 223c3e89d6fd..b0ca84b51afc 100644
@@ -1032,7 +1032,8 @@ static int s3erofs_remote_getobject(struct erofs_importer *im,
 		inode->datalayout = EROFS_INODE_FLAT_PLAIN;
 		inode->idata_size = 0;
 		ret = erofs_allocate_inode_bh_data(inode,
-				DIV_ROUND_UP(inode->i_size, 1U << sbi->blkszbits));
+				DIV_ROUND_UP(inode->i_size, 1U << sbi->blkszbits),
+				false);
 		if (ret)
 			return ret;
 		resp.vf = &sbi->bdev;
lib/super.c +2 -2
index a4837e5702ed..0180087e184e 100644
@@ -445,9 +445,9 @@ int erofs_mkfs_format_fs(struct erofs_sb_info *sbi, unsigned int blkszbits,
 	sbi->bmgr = bmgr;
 	bmgr->dsunit = dsunit;
 	if (metazone)
-		sbi->meta_blkaddr = EROFS_META_NEW_ADDR;
+		sbi->metazone_startblk = EROFS_META_NEW_ADDR;
 	else
-		sbi->meta_blkaddr = 0;
+		sbi->metazone_startblk = 0;
 	bh = erofs_reserve_sb(bmgr);
 	if (IS_ERR(bh))
 		return PTR_ERR(bh);
lib/tar.c +1 -1
index d5095169f9ba..1f3092566bd9 100644
@@ -632,7 +632,7 @@ static int tarerofs_write_uncompressed_file(struct erofs_inode *inode,
 	inode->datalayout = EROFS_INODE_FLAT_PLAIN;
 	nblocks = DIV_ROUND_UP(inode->i_size, 1U << sbi->blkszbits);
 
-	ret = erofs_allocate_inode_bh_data(inode, nblocks);
+	ret = erofs_allocate_inode_bh_data(inode, nblocks, false);
 	if (ret)
 		return ret;
 
mkfs/main.c +23 -4
index 620b1ed2b0c3..ffcb8cf75225 100644
@@ -100,7 +100,7 @@ static struct option long_options[] = {
 	{"oci", optional_argument, NULL, 534},
 #endif
 	{"zD", optional_argument, NULL, 536},
-	{"ZI", optional_argument, NULL, 537},
+	{"MZ", optional_argument, NULL, 537},
 	{"xattr-prefix", required_argument, NULL, 538},
 	{"xattr-inode-digest", required_argument, NULL, 539},
 	{0, 0, 0, 0},
@@ -178,7 +178,8 @@ static void usage(int argc, char **argv)
 		"    --mkfs-time         the timestamp is applied as build time only\n"
 		" -UX                    use a given filesystem UUID\n"
 		" --zD[=<0|1>]           specify directory compression: 0=disable [default], 1=enable\n"
-		" --ZI[=<0|1>]           specify the separate inode metadata zone availability: 0=disable [default], 1=enable\n"
+		" --MZ[=<0|[id]>]        put inode metadata ('i') and/or directory data ('d') into the separate metadata zone.\n"
+		"                        No argument enables both.  0=disable [default].\n"
 		" --all-root             make all files owned by root\n"
 #ifdef EROFS_MT_ENABLED
 		" --async-queue-limit=#  specify the maximum number of entries in the multi-threaded job queue\n"
@@ -1411,10 +1412,28 @@ static int mkfs_parse_options_cfg(struct erofs_importer_params *params,
 			}
 			break;
 		case 537:
-			if (!optarg || strcmp(optarg, "1"))
+			if (!optarg) {
 				mkfscfg.inode_metazone = true;
-			else
+				params->dirdata_in_metazone = true;
+			} else if (!strcmp(optarg, "0")) {
 				mkfscfg.inode_metazone = false;
+				params->dirdata_in_metazone = false;
+			} else {
+				for (i = 0; optarg[i]; ++i) {
+					if (optarg[i] == 'i') {
+						mkfscfg.inode_metazone = true;
+					} else if (optarg[i] == 'd') {
+						params->dirdata_in_metazone = true;
+					} else {
+						erofs_err("invalid metazone flags `%s`", optarg);
+						return -EINVAL;
+					}
+				}
+				if (params->dirdata_in_metazone && !mkfscfg.inode_metazone) {
+					erofs_err("inode metadata must be in the metadata zone if directory data is stored there");
+					return -EINVAL;
+				}
+			}
 			break;
 		case 538:
 			errno = 0;
Signature · 1 line
2.43.5
#1 lore raw

PATCH v2 1/1 erofs-utils: mkfs: enable directory data in the metadata zone

From: Gao Xiang <[email protected]>

It allows directory data and inode metadata to be kept as close as possible, significantly improving metadata performance for long-latency remote image use cases.

Usage:

 $ mkfs.erofs --MZ foo.erofs foo/

---
v2:
- fixup erofs_iflush() mapped i_blkaddr calculation.

 include/erofs/importer.h |   1 +
 include/erofs/inode.h    |   3 +-
 include/erofs/internal.h |   8 +++
 lib/cache.c              |   4 ++
 lib/inode.c              | 109 ++++++++++++++++++++++++++++-----------
 lib/io.c                 |   5 +-
 lib/metabox.c            |  22 +++++---
 lib/remotes/s3.c         |   3 +-
 lib/super.c              |   4 +-
 lib/tar.c                |   2 +-
 mkfs/main.c              |  26 ++++++++--
 11 files changed, 140 insertions(+), 47 deletions(-)
include/erofs/importer.h +1 -0
index a525b47..60160d6 100644
@@ -46,6 +46,7 @@ struct erofs_importer_params {
 	bool no_datainline;
 	/* Issue directory data (except inline data) separately from regular inodes */
 	bool grouped_dirdata;
+	bool dirdata_in_metazone;
 	bool hard_dereference;
 	bool ovlfs_strip;
 	bool dot_omitted;
include/erofs/inode.h +2 -1
index 89bd16a..ba62ece 100644
@@ -38,7 +38,8 @@ erofs_nid_t erofs_lookupnid(struct erofs_inode *inode);
 int erofs_iflush(struct erofs_inode *inode);
 struct erofs_dentry *erofs_d_alloc(struct erofs_inode *parent,
 				   const char *name);
-int erofs_allocate_inode_bh_data(struct erofs_inode *inode, erofs_blk_t nblocks);
+int erofs_allocate_inode_bh_data(struct erofs_inode *inode, erofs_blk_t nblocks,
+				 bool in_metazone);
 bool erofs_dentry_is_wht(struct erofs_sb_info *sbi, struct erofs_dentry *d);
 int __erofs_fill_inode(struct erofs_importer *im, struct erofs_inode *inode,
 		       struct stat *st, const char *path);
include/erofs/internal.h +8 -0
index 62594b8..db6014b 100644
@@ -156,6 +156,7 @@ struct erofs_sb_info {
 	struct erofs_buffer_head *bh_devt;
 	bool useqpl;
 	bool sb_valid;
+	u32 metazone_startblk;
 };
 
 /* make sure that any user of the erofs headers has atleast 64bit off_t type */
@@ -203,6 +204,8 @@ struct erofs_diskbuf;
 #define EROFS_INODE_DATA_SOURCE_DISKBUF		2
 #define EROFS_INODE_DATA_SOURCE_RESVSP		3
 
+#define EROFS_I_BLKADDR_DEV_ID_BIT		48
+
 struct erofs_inode {
 	struct list_head i_hash, i_subdirs, i_xattrs;
 
@@ -306,6 +309,11 @@ static inline bool erofs_inode_in_metabox(struct erofs_inode *inode)
 	return inode->nid >> EROFS_DIRENT_NID_METABOX_BIT;
 }
 
+static inline erofs_blk_t erofs_inode_dev_baddr(struct erofs_inode *inode)
+{
+	return inode->u.i_blkaddr & (BIT(EROFS_I_BLKADDR_DEV_ID_BIT) - 1);
+}
+
 static inline erofs_off_t erofs_iloc(struct erofs_inode *inode)
 {
 	struct erofs_sb_info *sbi = inode->sbi;
lib/cache.c +4 -0
index a87575a..f23dbb0 100644
@@ -479,6 +479,10 @@ static int __erofs_bflush(struct erofs_bufmgr *bmgr,
 
 			/* flush and remove bh */
 			ret = bh->op->flush(bh);
+			if (__erofs_unlikely(ret == -EBUSY && !forget)) {
+				skip = true;
+				continue;
+			}
 			if (ret < 0)
 				return ret;
 		}
lib/inode.c +80 -29

214 lines not on this page · raw message

lib/io.c +4 -1
index 37a74f6..0c5eb2c 100644
@@ -571,10 +571,13 @@ ssize_t erofs_io_write(struct erofs_vfile *vf, void *buf, size_t len)
 
 off_t erofs_io_lseek(struct erofs_vfile *vf, u64 offset, int whence)
 {
+	off_t ret;
+
 	if (vf->ops)
 		return vf->ops->lseek(vf, offset, whence);
 
-	return lseek(vf->fd, offset, whence);
+	ret = lseek(vf->fd, offset, whence);
+	return ret < 0 ? -errno : ret;
 }
 
 ssize_t erofs_io_sendfile(struct erofs_vfile *vout, struct erofs_vfile *vin,
lib/metabox.c +14 -8
index 37267dd..d6abd51 100644
@@ -54,7 +54,7 @@ int erofs_metadata_init(struct erofs_sb_info *sbi)
 	struct erofs_metamgr *m2gr;
 	int ret;
 
-	if (!sbi->m2gr && sbi->meta_blkaddr == EROFS_META_NEW_ADDR) {
+	if (!sbi->m2gr && sbi->metazone_startblk == EROFS_META_NEW_ADDR) {
 		m2gr = malloc(sizeof(*m2gr));
 		if (!m2gr)
 			return -ENOMEM;
@@ -62,6 +62,8 @@ int erofs_metadata_init(struct erofs_sb_info *sbi)
 		if (ret)
 			goto err_free;
 		sbi->m2gr = m2gr;
+		/* FIXME: sbi->meta_blkaddr should be 0 for 48-bit layouts */
+		sbi->meta_blkaddr = EROFS_META_NEW_ADDR;
 	}
 
 	if (!sbi->mxgr && erofs_sb_has_metabox(sbi)) {
@@ -124,20 +126,24 @@ int erofs_metazone_flush(struct erofs_sb_info *sbi)
 
 	if (!m2gr)
 		return 0;
-	m2bgr = m2gr->bmgr;
+	bh = erofs_balloc(sbi->bmgr, DATA, 0, 0);
+	if (!bh)
+		return PTR_ERR(bh);
+	erofs_mapbh(NULL, bh->block);
+	pos_out = erofs_btell(bh, false);
+	meta_blkaddr = pos_out >> sbi->blkszbits;
+	sbi->metazone_startblk = meta_blkaddr;
 
+	m2bgr = m2gr->bmgr;
 	ret = erofs_bflush(m2bgr, NULL);
 	if (ret)
 		return ret;
 
 	length = erofs_mapbh(m2bgr, NULL) << sbi->blkszbits;
-	bh = erofs_balloc(sbi->bmgr, DATA, length, 0);
-	if (!bh)
-		return PTR_ERR(bh);
+	ret = erofs_bh_balloon(bh, length);
+	if (ret < 0)
+		return ret;
 
-	erofs_mapbh(NULL, bh->block);
-	pos_out = erofs_btell(bh, false);
-	meta_blkaddr = pos_out >> sbi->blkszbits;
 	do {
 		count = min_t(erofs_off_t, length, INT_MAX);
 		ret = erofs_io_xcopy(sbi->bmgr->vf, pos_out,
lib/remotes/s3.c +2 -1
index 223c3e8..b0ca84b 100644
@@ -1032,7 +1032,8 @@ static int s3erofs_remote_getobject(struct erofs_importer *im,
 		inode->datalayout = EROFS_INODE_FLAT_PLAIN;
 		inode->idata_size = 0;
 		ret = erofs_allocate_inode_bh_data(inode,
-				DIV_ROUND_UP(inode->i_size, 1U << sbi->blkszbits));
+				DIV_ROUND_UP(inode->i_size, 1U << sbi->blkszbits),
+				false);
 		if (ret)
 			return ret;
 		resp.vf = &sbi->bdev;
lib/super.c +2 -2
index e54aff2..a203f96 100644
@@ -436,9 +436,9 @@ int erofs_mkfs_format_fs(struct erofs_sb_info *sbi, unsigned int blkszbits,
 	sbi->bmgr = bmgr;
 	bmgr->dsunit = dsunit;
 	if (metazone)
-		sbi->meta_blkaddr = EROFS_META_NEW_ADDR;
+		sbi->metazone_startblk = EROFS_META_NEW_ADDR;
 	else
-		sbi->meta_blkaddr = 0;
+		sbi->metazone_startblk = 0;
 	bh = erofs_reserve_sb(bmgr);
 	if (IS_ERR(bh))
 		return PTR_ERR(bh);
lib/tar.c +1 -1
index d509516..1f30925 100644
@@ -632,7 +632,7 @@ static int tarerofs_write_uncompressed_file(struct erofs_inode *inode,
 	inode->datalayout = EROFS_INODE_FLAT_PLAIN;
 	nblocks = DIV_ROUND_UP(inode->i_size, 1U << sbi->blkszbits);
 
-	ret = erofs_allocate_inode_bh_data(inode, nblocks);
+	ret = erofs_allocate_inode_bh_data(inode, nblocks, false);
 	if (ret)
 		return ret;
 
mkfs/main.c +22 -4
index e29eb99..70d08b5 100644
@@ -101,7 +101,7 @@ static struct option long_options[] = {
 	{"oci", optional_argument, NULL, 534},
 #endif
 	{"zD", optional_argument, NULL, 536},
-	{"ZI", optional_argument, NULL, 537},
+	{"MZ", optional_argument, NULL, 537},
 	{0, 0, 0, 0},
 };
 
@@ -177,7 +177,7 @@ static void usage(int argc, char **argv)
 		"    --mkfs-time        the timestamp is applied as build time only\n"
 		" -UX                   use a given filesystem UUID\n"
 		" --zD[=<0|1>]          specify directory compression: 0=disable [default], 1=enable\n"
-		" --ZI[=<0|1>]          specify the separate inode metadata zone availability: 0=disable [default], 1=enable\n"
+		" --MZ[=<0|[id]>]       put inode metadata ('i') and/or directory data ('d') into the separate metadata zone.\n"
 		" --all-root            make all files owned by root\n"
 #ifdef EROFS_MT_ENABLED
 		" --async-queue-limit=# specify the maximum number of entries in the multi-threaded job queue\n"
@@ -1419,10 +1419,28 @@ static int mkfs_parse_options_cfg(struct erofs_importer_params *params,
 			}
 			break;
 		case 537:
-			if (!optarg || strcmp(optarg, "1"))
+			if (!optarg) {
 				mkfscfg.inode_metazone = true;
-			else
+				params->dirdata_in_metazone = true;
+			} else if (!strcmp(optarg, "0")) {
 				mkfscfg.inode_metazone = false;
+				params->dirdata_in_metazone = false;
+			} else {
+				for (i = 0; optarg[i]; ++i) {
+					if (optarg[i] == 'i') {
+						mkfscfg.inode_metazone = true;
+					} else if (optarg[i] == 'd') {
+						params->dirdata_in_metazone = true;
+					} else {
+						erofs_err("invalid metazone flags `%s`", optarg);
+						return -EINVAL;
+					}
+				}
+				if (params->dirdata_in_metazone && !mkfscfg.inode_metazone) {
+					erofs_err("inode metadata must be in the metadata zone if directory data is stored there");
+					return -EINVAL;
+				}
+			}
 			break;
 		case 'V':
 			version();
Signature · 1 line
2.47.3
author
#2 lore raw

PATCH v3 1/1 erofs-utils: mkfs: enable directory data in the metadata zone

It allows directory data and inode metadata to be kept as close as possible, significantly improving metadata performance for long-latency remote image use cases.

Usage:

 $ mkfs.erofs --MZ foo.erofs foo/

---
v3:
- use correct vfile according to corresponding bufmgr in

    erofs_write_tail_end().
 include/erofs/importer.h |   1 +
 include/erofs/inode.h    |   3 +-
 include/erofs/internal.h |   8 +++
 lib/cache.c              |   4 ++
 lib/inode.c              | 114 ++++++++++++++++++++++++++++-----------
 lib/io.c                 |   5 +-
 lib/metabox.c            |  22 +++++---
 lib/remotes/s3.c         |   3 +-
 lib/super.c              |   4 +-
 lib/tar.c                |   2 +-
 mkfs/main.c              |  26 +++++++--
 11 files changed, 144 insertions(+), 48 deletions(-)
include/erofs/importer.h +1 -0
index a525b474f1d5..60160d6bea05 100644
@@ -46,6 +46,7 @@ struct erofs_importer_params {
 	bool no_datainline;
 	/* Issue directory data (except inline data) separately from regular inodes */
 	bool grouped_dirdata;
+	bool dirdata_in_metazone;
 	bool hard_dereference;
 	bool ovlfs_strip;
 	bool dot_omitted;
include/erofs/inode.h +2 -1
index 89bd16aecc06..ba62ece9a7cc 100644
@@ -38,7 +38,8 @@ erofs_nid_t erofs_lookupnid(struct erofs_inode *inode);
 int erofs_iflush(struct erofs_inode *inode);
 struct erofs_dentry *erofs_d_alloc(struct erofs_inode *parent,
 				   const char *name);
-int erofs_allocate_inode_bh_data(struct erofs_inode *inode, erofs_blk_t nblocks);
+int erofs_allocate_inode_bh_data(struct erofs_inode *inode, erofs_blk_t nblocks,
+				 bool in_metazone);
 bool erofs_dentry_is_wht(struct erofs_sb_info *sbi, struct erofs_dentry *d);
 int __erofs_fill_inode(struct erofs_importer *im, struct erofs_inode *inode,
 		       struct stat *st, const char *path);
include/erofs/internal.h +8 -0
index 62594b877151..db6014bce62f 100644
@@ -156,6 +156,7 @@ struct erofs_sb_info {
 	struct erofs_buffer_head *bh_devt;
 	bool useqpl;
 	bool sb_valid;
+	u32 metazone_startblk;
 };
 
 /* make sure that any user of the erofs headers has atleast 64bit off_t type */
@@ -203,6 +204,8 @@ struct erofs_diskbuf;
 #define EROFS_INODE_DATA_SOURCE_DISKBUF		2
 #define EROFS_INODE_DATA_SOURCE_RESVSP		3
 
+#define EROFS_I_BLKADDR_DEV_ID_BIT		48
+
 struct erofs_inode {
 	struct list_head i_hash, i_subdirs, i_xattrs;
 
@@ -306,6 +309,11 @@ static inline bool erofs_inode_in_metabox(struct erofs_inode *inode)
 	return inode->nid >> EROFS_DIRENT_NID_METABOX_BIT;
 }
 
+static inline erofs_blk_t erofs_inode_dev_baddr(struct erofs_inode *inode)
+{
+	return inode->u.i_blkaddr & (BIT(EROFS_I_BLKADDR_DEV_ID_BIT) - 1);
+}
+
 static inline erofs_off_t erofs_iloc(struct erofs_inode *inode)
 {
 	struct erofs_sb_info *sbi = inode->sbi;
lib/cache.c +4 -0
index a87575ad74d1..f23dbb06264a 100644
@@ -479,6 +479,10 @@ static int __erofs_bflush(struct erofs_bufmgr *bmgr,
 
 			/* flush and remove bh */
 			ret = bh->op->flush(bh);
+			if (__erofs_unlikely(ret == -EBUSY && !forget)) {
+				skip = true;
+				continue;
+			}
 			if (ret < 0)
 				return ret;
 		}
lib/inode.c +84 -30

237 lines not on this page · raw message

lib/io.c +4 -1
index 37a74f63c45e..0c5eb2c29989 100644
@@ -571,10 +571,13 @@ ssize_t erofs_io_write(struct erofs_vfile *vf, void *buf, size_t len)
 
 off_t erofs_io_lseek(struct erofs_vfile *vf, u64 offset, int whence)
 {
+	off_t ret;
+
 	if (vf->ops)
 		return vf->ops->lseek(vf, offset, whence);
 
-	return lseek(vf->fd, offset, whence);
+	ret = lseek(vf->fd, offset, whence);
+	return ret < 0 ? -errno : ret;
 }
 
 ssize_t erofs_io_sendfile(struct erofs_vfile *vout, struct erofs_vfile *vin,
lib/metabox.c +14 -8
index 37267ddb73cf..d6abd5123cc8 100644
@@ -54,7 +54,7 @@ int erofs_metadata_init(struct erofs_sb_info *sbi)
 	struct erofs_metamgr *m2gr;
 	int ret;
 
-	if (!sbi->m2gr && sbi->meta_blkaddr == EROFS_META_NEW_ADDR) {
+	if (!sbi->m2gr && sbi->metazone_startblk == EROFS_META_NEW_ADDR) {
 		m2gr = malloc(sizeof(*m2gr));
 		if (!m2gr)
 			return -ENOMEM;
@@ -62,6 +62,8 @@ int erofs_metadata_init(struct erofs_sb_info *sbi)
 		if (ret)
 			goto err_free;
 		sbi->m2gr = m2gr;
+		/* FIXME: sbi->meta_blkaddr should be 0 for 48-bit layouts */
+		sbi->meta_blkaddr = EROFS_META_NEW_ADDR;
 	}
 
 	if (!sbi->mxgr && erofs_sb_has_metabox(sbi)) {
@@ -124,20 +126,24 @@ int erofs_metazone_flush(struct erofs_sb_info *sbi)
 
 	if (!m2gr)
 		return 0;
-	m2bgr = m2gr->bmgr;
+	bh = erofs_balloc(sbi->bmgr, DATA, 0, 0);
+	if (!bh)
+		return PTR_ERR(bh);
+	erofs_mapbh(NULL, bh->block);
+	pos_out = erofs_btell(bh, false);
+	meta_blkaddr = pos_out >> sbi->blkszbits;
+	sbi->metazone_startblk = meta_blkaddr;
 
+	m2bgr = m2gr->bmgr;
 	ret = erofs_bflush(m2bgr, NULL);
 	if (ret)
 		return ret;
 
 	length = erofs_mapbh(m2bgr, NULL) << sbi->blkszbits;
-	bh = erofs_balloc(sbi->bmgr, DATA, length, 0);
-	if (!bh)
-		return PTR_ERR(bh);
+	ret = erofs_bh_balloon(bh, length);
+	if (ret < 0)
+		return ret;
 
-	erofs_mapbh(NULL, bh->block);
-	pos_out = erofs_btell(bh, false);
-	meta_blkaddr = pos_out >> sbi->blkszbits;
 	do {
 		count = min_t(erofs_off_t, length, INT_MAX);
 		ret = erofs_io_xcopy(sbi->bmgr->vf, pos_out,
lib/remotes/s3.c +2 -1
index 223c3e89d6fd..b0ca84b51afc 100644
@@ -1032,7 +1032,8 @@ static int s3erofs_remote_getobject(struct erofs_importer *im,
 		inode->datalayout = EROFS_INODE_FLAT_PLAIN;
 		inode->idata_size = 0;
 		ret = erofs_allocate_inode_bh_data(inode,
-				DIV_ROUND_UP(inode->i_size, 1U << sbi->blkszbits));
+				DIV_ROUND_UP(inode->i_size, 1U << sbi->blkszbits),
+				false);
 		if (ret)
 			return ret;
 		resp.vf = &sbi->bdev;
lib/super.c +2 -2
index e54aff2d4ab7..a203f96e6996 100644
@@ -436,9 +436,9 @@ int erofs_mkfs_format_fs(struct erofs_sb_info *sbi, unsigned int blkszbits,
 	sbi->bmgr = bmgr;
 	bmgr->dsunit = dsunit;
 	if (metazone)
-		sbi->meta_blkaddr = EROFS_META_NEW_ADDR;
+		sbi->metazone_startblk = EROFS_META_NEW_ADDR;
 	else
-		sbi->meta_blkaddr = 0;
+		sbi->metazone_startblk = 0;
 	bh = erofs_reserve_sb(bmgr);
 	if (IS_ERR(bh))
 		return PTR_ERR(bh);
lib/tar.c +1 -1
index d5095169f9ba..1f3092566bd9 100644
@@ -632,7 +632,7 @@ static int tarerofs_write_uncompressed_file(struct erofs_inode *inode,
 	inode->datalayout = EROFS_INODE_FLAT_PLAIN;
 	nblocks = DIV_ROUND_UP(inode->i_size, 1U << sbi->blkszbits);
 
-	ret = erofs_allocate_inode_bh_data(inode, nblocks);
+	ret = erofs_allocate_inode_bh_data(inode, nblocks, false);
 	if (ret)
 		return ret;
 
mkfs/main.c +22 -4
index e29eb99cca08..70d08b5a8fd3 100644
@@ -101,7 +101,7 @@ static struct option long_options[] = {
 	{"oci", optional_argument, NULL, 534},
 #endif
 	{"zD", optional_argument, NULL, 536},
-	{"ZI", optional_argument, NULL, 537},
+	{"MZ", optional_argument, NULL, 537},
 	{0, 0, 0, 0},
 };
 
@@ -177,7 +177,7 @@ static void usage(int argc, char **argv)
 		"    --mkfs-time        the timestamp is applied as build time only\n"
 		" -UX                   use a given filesystem UUID\n"
 		" --zD[=<0|1>]          specify directory compression: 0=disable [default], 1=enable\n"
-		" --ZI[=<0|1>]          specify the separate inode metadata zone availability: 0=disable [default], 1=enable\n"
+		" --MZ[=<0|[id]>]       put inode metadata ('i') and/or directory data ('d') into the separate metadata zone.\n"
 		" --all-root            make all files owned by root\n"
 #ifdef EROFS_MT_ENABLED
 		" --async-queue-limit=# specify the maximum number of entries in the multi-threaded job queue\n"
@@ -1419,10 +1419,28 @@ static int mkfs_parse_options_cfg(struct erofs_importer_params *params,
 			}
 			break;
 		case 537:
-			if (!optarg || strcmp(optarg, "1"))
+			if (!optarg) {
 				mkfscfg.inode_metazone = true;
-			else
+				params->dirdata_in_metazone = true;
+			} else if (!strcmp(optarg, "0")) {
 				mkfscfg.inode_metazone = false;
+				params->dirdata_in_metazone = false;
+			} else {
+				for (i = 0; optarg[i]; ++i) {
+					if (optarg[i] == 'i') {
+						mkfscfg.inode_metazone = true;
+					} else if (optarg[i] == 'd') {
+						params->dirdata_in_metazone = true;
+					} else {
+						erofs_err("invalid metazone flags `%s`", optarg);
+						return -EINVAL;
+					}
+				}
+				if (params->dirdata_in_metazone && !mkfscfg.inode_metazone) {
+					erofs_err("inode metadata must be in the metadata zone if directory data is stored there");
+					return -EINVAL;
+				}
+			}
 			break;
 		case 'V':
 			version();
Signature · 1 line
2.43.5