The underlay block storage could work in a stripe-like manner to improve
performance and space efficiency.

EROFS on-disk layout is flexible enough for such use cases.

Cc: Changpeng Liu <changpe...@tencent.com>
Signed-off-by: Gao Xiang <hsiang...@linux.alibaba.com>
---
 include/erofs/cache.h |  3 +++
 lib/cache.c           | 25 ++++++++++++++++++++-----
 man/mkfs.erofs.1      |  3 +++
 mkfs/main.c           | 11 +++++++++++
 4 files changed, 37 insertions(+), 5 deletions(-)

diff --git a/include/erofs/cache.h b/include/erofs/cache.h
index d8559a8..6ff80ab 100644
--- a/include/erofs/cache.h
+++ b/include/erofs/cache.h
@@ -67,6 +67,9 @@ struct erofs_bufmgr {
 
        /* last mapped buffer block to accelerate erofs_mapbh() */
        struct erofs_buffer_block *last_mapped_block;
+
+       /* align data block addresses to multiples of `dsunit` */
+       unsigned int dsunit;
 };
 
 static inline const int get_alignsize(struct erofs_sb_info *sbi, int type,
diff --git a/lib/cache.c b/lib/cache.c
index cb05466..d3e2e62 100644
--- a/lib/cache.c
+++ b/lib/cache.c
@@ -156,10 +156,15 @@ static int erofs_bfind_for_attach(struct erofs_bufmgr 
*bmgr,
        bb = NULL;
 
        /* try to find a most-fit mapped buffer block first */
-       if (size + inline_ext >= blksiz)
+
+       if (__erofs_unlikely(bmgr->dsunit > 1)) {
+               used_before = blksiz - alignsize;
+       } else if (size + inline_ext >= blksiz) {
                goto skip_mapped;
+       } else {
+               used_before = rounddown(blksiz - (size + inline_ext), 
alignsize);
+       }
 
-       used_before = rounddown(blksiz - (size + inline_ext), alignsize);
        for (; used_before; --used_before) {
                struct list_head *bt = bmgr->mapped_buckets[type] + used_before;
 
@@ -181,7 +186,7 @@ static int erofs_bfind_for_attach(struct erofs_bufmgr *bmgr,
                ret = __erofs_battach(cur, NULL, size, alignsize,
                                      inline_ext, true);
                if (ret < 0) {
-                       DBG_BUGON(1);
+                       DBG_BUGON(!(bmgr->dsunit > 1));
                        continue;
                }
 
@@ -324,10 +329,20 @@ struct erofs_buffer_head *erofs_battach(struct 
erofs_buffer_head *bh,
 static void __erofs_mapbh(struct erofs_buffer_block *bb)
 {
        struct erofs_bufmgr *bmgr = bb->buffers.fsprivate;
-       erofs_blk_t blkaddr;
+       erofs_blk_t blkaddr = bmgr->tail_blkaddr;
 
        if (bb->blkaddr == NULL_ADDR) {
-               bb->blkaddr = bmgr->tail_blkaddr;
+               bb->blkaddr = blkaddr;
+               if (__erofs_unlikely(bmgr->dsunit > 1) && bb->type == DATA) {
+                       struct erofs_buffer_block *pb = list_prev_entry(bb, 
list);
+
+                       bb->blkaddr = roundup(blkaddr, bmgr->dsunit);
+                       if (pb != &bmgr->blkh &&
+                           pb->blkaddr + pb->buffers.nblocks >= blkaddr) {
+                               DBG_BUGON(pb->blkaddr + pb->buffers.nblocks > 
blkaddr);
+                               pb->buffers.nblocks = bb->blkaddr - pb->blkaddr;
+                       }
+               }
                bmgr->last_mapped_block = bb;
                erofs_bupdate_mapped(bb);
        }
diff --git a/man/mkfs.erofs.1 b/man/mkfs.erofs.1
index 0093839..ee84163 100644
--- a/man/mkfs.erofs.1
+++ b/man/mkfs.erofs.1
@@ -149,6 +149,9 @@ the given primary algorithm, alternative algorithms can be 
specified with
 are extended regular expressions, matched against absolute paths within
 the output filesystem, with no leading /.
 .TP
+.BI "\-\-dsunit=" #
+Align all data block addresses to multiples of #.
+.TP
 .BI "\-\-exclude-path=" path
 Ignore file that matches the exact literal path.
 You may give multiple
diff --git a/mkfs/main.c b/mkfs/main.c
index 9ca7dad..1f4b7c6 100644
--- a/mkfs/main.c
+++ b/mkfs/main.c
@@ -86,6 +86,7 @@ static struct option long_options[] = {
        {"all-time", no_argument, NULL, 526},
        {"sort", required_argument, NULL, 527},
        {"hard-dereference", no_argument, NULL, 528},
+       {"dsunit", required_argument, NULL, 529},
        {0, 0, 0, 0},
 };
 
@@ -162,6 +163,7 @@ static void usage(int argc, char **argv)
                " --blobdev=X           specify an extra device X to store 
chunked data\n"
                " --chunksize=#         generate chunk-based files with #-byte 
chunks\n"
                " --clean=X             run full clean build (default) or:\n"
+               " --dsunit=#            align all data block addresses to 
multiples of #\n"
                " --incremental=X       run incremental build\n"
                "                       (X = data|rvsp; data=full data, 
rvsp=space is allocated\n"
                "                                       and filled with 
zeroes)\n"
@@ -243,6 +245,7 @@ static unsigned int rebuild_src_count;
 static LIST_HEAD(rebuild_src_list);
 static u8 fixeduuid[16];
 static bool valid_fixeduuid;
+static unsigned int dsunit;
 
 static int erofs_mkfs_feat_set_legacy_compress(bool en, const char *val,
                                               unsigned int vallen)
@@ -856,6 +859,13 @@ static int mkfs_parse_options_cfg(int argc, char *argv[])
                case 528:
                        cfg.c_hard_dereference = true;
                        break;
+               case 529:
+                       dsunit = strtoul(optarg, &endptr, 0);
+                       if (*endptr != '\0') {
+                               erofs_err("invalid dsunit %s", optarg);
+                               return -EINVAL;
+                       }
+                       break;
                case 'V':
                        version();
                        exit(0);
@@ -1318,6 +1328,7 @@ int main(int argc, char **argv)
                }
                sb_bh = NULL;
        }
+       g_sbi.bmgr->dsunit = dsunit;
 
        /* Use the user-defined UUID or generate one for clean builds */
        if (valid_fixeduuid)
-- 
2.43.5

Reply via email to