* [PATCH RFC 1/2] erofs-utils: mkfs: add file-delta framework
2026-09-16 9:10 [PATCH RFC 0/2] erofs-utils: add dm-persistent file delta support Yifan Zhao
@ 2026-09-16 9:10 ` Yifan Zhao
2026-09-16 9:10 ` [PATCH RFC 2/2] erofs-utils: mkfs: support dm-persistent file deltas Yifan Zhao
1 sibling, 0 replies; 3+ messages in thread
From: Yifan Zhao @ 2026-09-16 9:10 UTC (permalink / raw)
To: linux-erofs
Add source operations for reading logical updates and layout operations for
inheriting parent mappings, writing updates and publishing output indexes.
Keep the parent read view separate from the output inode and its superblock.
Implement the Chunk layout backend with independent parent, update and
output granularities. Split inherited large chunks without copying their
data, retain historical device references and allow parent chunk merging.
Store changed nonzero data in the output primary and zero updates as holes.
Match targets by parent inode, including hard-linked files.
Reject unsupported 48-bit layouts and block ranges.
Assisted-by: Codex:gpt-5.6-sol
Signed-off-by: Yifan Zhao <yifan.yfzhao@foxmail.com>
---
include/erofs/internal.h | 2 +
lib/Makefile.am | 4 +-
lib/blobchunk.c | 25 +++++
lib/file_delta.c | 214 ++++++++++++++++++++++++++++++++++++++
lib/file_delta_chunk.c | 203 ++++++++++++++++++++++++++++++++++++
lib/liberofs_chunk.h | 2 +
lib/liberofs_file_delta.h | 81 +++++++++++++++
lib/rebuild.c | 34 +++++-
man/mkfs.erofs.1 | 56 ++++++++++
mkfs/main.c | 111 +++++++++++++++++++-
10 files changed, 725 insertions(+), 7 deletions(-)
create mode 100644 lib/file_delta.c
create mode 100644 lib/file_delta_chunk.c
create mode 100644 lib/liberofs_file_delta.h
diff --git a/include/erofs/internal.h b/include/erofs/internal.h
index 973bb96..6251a3e 100644
--- a/include/erofs/internal.h
+++ b/include/erofs/internal.h
@@ -94,6 +94,7 @@ struct erofs_xattrmgr;
struct z_erofs_mgr;
struct erofs_metamgr;
struct erofs_chunkmgr;
+struct erofs_file_delta_manager;
struct erofs_sb_info {
struct erofs_sb_lz4_info lz4;
@@ -154,6 +155,7 @@ struct erofs_sb_info {
struct erofs_xattrmgr *xamgr;
struct z_erofs_mgr *zmgr;
struct erofs_chunkmgr *chunkmgr;
+ struct erofs_file_delta_manager *file_delta_mgr;
struct erofs_metamgr *m2gr, *mxgr;
struct erofs_packed_inode *packedinode;
struct erofs_buffer_head *bh_sb;
diff --git a/lib/Makefile.am b/lib/Makefile.am
index 0a82915..3d562f4 100644
--- a/lib/Makefile.am
+++ b/lib/Makefile.am
@@ -24,6 +24,7 @@ noinst_HEADERS = $(top_srcdir)/include/erofs_fs.h \
$(top_srcdir)/lib/liberofs_cache.h \
$(top_srcdir)/lib/liberofs_chunk.h \
$(top_srcdir)/lib/liberofs_compress.h \
+ $(top_srcdir)/lib/liberofs_file_delta.h \
$(top_srcdir)/lib/liberofs_fragments.h \
$(top_srcdir)/lib/liberofs_private.h \
$(top_srcdir)/lib/liberofs_rebuild.h \
@@ -44,7 +45,8 @@ liberofs_la_SOURCES = config.c io.c cache.c super.c inode.c xattr.c exclude.c \
compress_hints.c sha256.c blobchunk.c dir.c \
fragments.c dedupe.c uuid_unparse.c uuid.c tar.c \
block_list.c rebuild.c diskbuf.c bitops.c dedupe_ext.c \
- vmdk.c metabox.c global.c importer.c base64.c
+ vmdk.c metabox.c global.c importer.c base64.c file_delta.c \
+ file_delta_chunk.c
liberofs_la_CFLAGS = -Wall ${libuuid_CFLAGS} -I$(top_srcdir)/include
liberofs_la_LIBADD = $(LIBEROFS_LIBS)
diff --git a/lib/blobchunk.c b/lib/blobchunk.c
index 8f43dd0..8150fd4 100644
--- a/lib/blobchunk.c
+++ b/lib/blobchunk.c
@@ -161,6 +161,31 @@ static struct erofs_chunkitem *erofs_get_chunk(struct erofs_sb_info *sbi,
return chunk;
}
+struct erofs_chunkitem *erofs_blob_write_chunk(struct erofs_sb_info *sbi,
+ const void *buf, erofs_off_t chunksize, int device_id)
+{
+ struct erofs_chunkmgr *cmgr = sbi->chunkmgr;
+ struct erofs_chunkitem *chunk;
+
+ if (!cmgr || !chunksize || chunksize > INT_MAX || device_id < 0)
+ return ERR_PTR(-EINVAL);
+ if (device_id &&
+ (device_id > sbi->extra_devices ||
+ !sbi->devs[device_id - 1].bmgr))
+ return ERR_PTR(-EINVAL);
+ if (!device_id && !sbi->bmgr)
+ return ERR_PTR(-EINVAL);
+ chunk = erofs_get_chunk(sbi, device_id, (u8 *)buf, chunksize);
+ if (IS_ERR(chunk))
+ return chunk;
+ if (chunk->blkaddr >= UINT32_MAX ||
+ BLK_ROUND_UP(sbi, chunksize) > UINT32_MAX - chunk->blkaddr) {
+ erofs_err("file-delta output exceeds the 32-bit block range");
+ return ERR_PTR(-EFBIG);
+ }
+ return chunk;
+}
+
void erofs_inode_fixup_chunkformat(struct erofs_inode *inode)
{
unsigned int unit, src;
diff --git a/lib/file_delta.c b/lib/file_delta.c
new file mode 100644
index 0000000..2d7653d
--- /dev/null
+++ b/lib/file_delta.c
@@ -0,0 +1,214 @@
+// SPDX-License-Identifier: GPL-2.0+ OR MIT
+#define _GNU_SOURCE
+#include <errno.h>
+#include <stdlib.h>
+#include <string.h>
+#include <config.h>
+#include "erofs/print.h"
+#include "liberofs_file_delta.h"
+
+static void erofs_file_delta_free(struct erofs_file_delta *delta)
+{
+ delta->ops->close(delta);
+ free(delta->target);
+ free(delta->source_path);
+ free(delta);
+}
+
+int erofs_file_delta_parse_spec(struct list_head *deltas, const char *arg,
+ const struct erofs_file_delta_source_ops * const *backends)
+{
+ const struct erofs_file_delta_source_ops *ops;
+ struct erofs_file_delta *delta;
+ const char *first, *second;
+ size_t target_len, namelen;
+ int ret;
+
+ first = strchr(arg, ':');
+ if (!first || first == arg)
+ return -EINVAL;
+ namelen = first - arg;
+ while ((ops = *backends++)) {
+ if (strlen(ops->name) == namelen &&
+ !memcmp(ops->name, arg, namelen))
+ break;
+ }
+ if (!ops)
+ return -EOPNOTSUPP;
+ second = strchr(first + 1, ':');
+ if (!second || !second[1])
+ return -EINVAL;
+ target_len = second - (first + 1);
+ if (!target_len)
+ return -EINVAL;
+
+ delta = calloc(1, sizeof(*delta));
+ if (!delta)
+ return -ENOMEM;
+ delta->ops = ops;
+ delta->target = strndup(first + 1, target_len);
+ delta->source_path = strdup(second + 1);
+ if (!delta->target || !delta->source_path) {
+ ret = -ENOMEM;
+ goto err_free_delta;
+ }
+ list_add_tail(&delta->list, deltas);
+ return 0;
+
+err_free_delta:
+ erofs_file_delta_free(delta);
+ return ret;
+}
+
+void erofs_file_delta_cleanup(struct list_head *deltas)
+{
+ struct erofs_file_delta *delta, *n;
+
+ list_for_each_entry_safe(delta, n, deltas, list) {
+ list_del(&delta->list);
+ erofs_file_delta_free(delta);
+ }
+}
+
+int erofs_file_delta_manager_init(struct erofs_file_delta_manager *mgr,
+ struct erofs_sb_info *parent, erofs_off_t unit_size)
+{
+ struct erofs_file_delta *delta, *pos;
+ int ret;
+
+ list_for_each_entry(delta, &mgr->deltas, list) {
+ delta->lower_inode = (struct erofs_inode) { .sbi = parent };
+ ret = erofs_ilookup(delta->target, &delta->lower_inode);
+ if (ret) {
+ erofs_err("file-delta TARGET %s cannot be read: %s",
+ delta->target, erofs_strerror(ret));
+ return ret;
+ }
+ list_for_each_entry(pos, &mgr->deltas, list) {
+ if (pos == delta)
+ break;
+ if (pos->lower_inode.nid == delta->lower_inode.nid) {
+ erofs_err("file-delta TARGET %s refers to the same file as %s",
+ delta->target, pos->target);
+ return -EEXIST;
+ }
+ }
+ if (!S_ISREG(delta->lower_inode.i_mode)) {
+ erofs_err("file-delta TARGET %s is not a regular file",
+ delta->target);
+ return -EINVAL;
+ }
+ if (!delta->lower_inode.i_size) {
+ erofs_err("file-delta TARGET %s is empty", delta->target);
+ return -EINVAL;
+ }
+
+ delta->unit_size = unit_size;
+ ret = delta->ops->open(delta, delta->lower_inode.i_size,
+ unit_size);
+ if (ret) {
+ erofs_err("file-delta %s source %s failed validation: %s",
+ delta->ops->name, delta->source_path,
+ erofs_strerror(ret));
+ return ret;
+ }
+ }
+ return 0;
+}
+
+struct erofs_file_delta *erofs_file_delta_lookup_target(
+ struct erofs_file_delta_manager *mgr, const struct erofs_inode *inode)
+{
+ struct erofs_file_delta *delta;
+
+ if (!mgr)
+ return NULL;
+ list_for_each_entry(delta, &mgr->deltas, list)
+ if (delta->lower_inode.sbi == inode->sbi &&
+ delta->lower_inode.nid == inode->nid)
+ return delta;
+ return NULL;
+}
+
+static int erofs_file_delta_read_update(const struct erofs_file_delta *delta,
+ size_t index, erofs_off_t size, void *buf,
+ struct erofs_file_delta_update *update)
+{
+ u64 unit_index;
+ int ret;
+
+ ret = delta->ops->read(delta, index, &unit_index, buf);
+ if (ret)
+ return ret;
+ if (__builtin_mul_overflow(unit_index, delta->unit_size,
+ &update->offset) || update->offset >= size)
+ return -EINVAL;
+ update->length = min_t(erofs_off_t, delta->unit_size,
+ size - update->offset);
+ update->data = buf;
+ return 0;
+}
+
+int erofs_file_delta_apply(struct erofs_file_delta *delta,
+ const struct erofs_file_delta_layout_ops *ops,
+ const struct erofs_inode *parent, struct erofs_inode *inode)
+{
+ struct erofs_file_delta_apply_ctx ctx = {
+ .parent = parent,
+ .inode = inode,
+ };
+ void *buf = NULL;
+ size_t i, nr_changes;
+ int ret;
+
+ DBG_BUGON(parent == inode || parent->i_size != inode->i_size);
+ DBG_BUGON(delta && (delta->applied || !delta->unit_size ||
+ delta->unit_size > SIZE_MAX));
+
+ nr_changes = delta ? delta->ops->count(delta) : 0;
+ if (nr_changes)
+ ctx.update_unit = delta->unit_size;
+ ret = ops->prepare(&ctx);
+ if (ret)
+ goto out;
+ if (nr_changes) {
+ buf = malloc(delta->unit_size);
+ if (!buf) {
+ ret = -ENOMEM;
+ goto out;
+ }
+ }
+ for (i = 0; i < nr_changes; ++i) {
+ struct erofs_file_delta_update update;
+
+ ret = erofs_file_delta_read_update(delta, i, parent->i_size,
+ buf, &update);
+ if (ret)
+ goto out;
+ ret = ops->update(&ctx, &update);
+ if (ret)
+ goto out;
+ }
+ ret = ops->finish(&ctx);
+ if (!ret && delta)
+ delta->applied = true;
+out:
+ ops->release(&ctx);
+ free(buf);
+ return ret;
+}
+
+int erofs_file_delta_manager_finish(struct erofs_file_delta_manager *mgr)
+{
+ struct erofs_file_delta *delta;
+
+ if (!mgr)
+ return 0;
+ list_for_each_entry(delta, &mgr->deltas, list) {
+ if (!delta->applied) {
+ erofs_err("file-delta TARGET %s was not applied", delta->target);
+ return -ENOENT;
+ }
+ }
+ return 0;
+}
diff --git a/lib/file_delta_chunk.c b/lib/file_delta_chunk.c
new file mode 100644
index 0000000..d43edb4
--- /dev/null
+++ b/lib/file_delta_chunk.c
@@ -0,0 +1,203 @@
+// SPDX-License-Identifier: GPL-2.0+ OR MIT
+#include <config.h>
+#include <stdlib.h>
+#include "erofs/print.h"
+#include "liberofs_chunk.h"
+#include "liberofs_file_delta.h"
+
+struct erofs_file_delta_chunk {
+ struct erofs_inode parent_cursor;
+ struct erofs_chunkitem **indexes;
+ size_t count;
+ unsigned int chunkbits;
+};
+
+static struct erofs_chunkitem *erofs_file_delta_chunk_import(
+ struct erofs_file_delta_apply_ctx *ctx, erofs_off_t offset)
+{
+ struct erofs_file_delta_chunk *state = ctx->private;
+ struct erofs_inode *parent = &state->parent_cursor;
+ struct erofs_sb_info *src = parent->sbi, *dst = ctx->inode->sbi;
+ struct erofs_map_blocks map = { .buf = __EROFS_BUF_INITIALIZER };
+ struct erofs_map_dev mdev;
+ struct erofs_device_info *dif;
+ erofs_off_t within, length;
+ erofs_blk_t startblk, nblocks;
+ unsigned int device_id;
+ int ret;
+
+ map.m_la = offset;
+ ret = erofs_map_blocks(parent, &map, 0);
+ if (ret)
+ return ERR_PTR(ret);
+ length = min_t(erofs_off_t, 1ULL << state->chunkbits,
+ parent->i_size - offset);
+ if (offset < map.m_la)
+ return ERR_PTR(-EFSCORRUPTED);
+ within = offset - map.m_la;
+ if (within >= map.m_llen || length > map.m_llen - within)
+ return ERR_PTR(-EFSCORRUPTED);
+ if (!(map.m_flags & EROFS_MAP_MAPPED))
+ return erofs_get_unhashed_chunk(dst, 0, EROFS_NULL_ADDR, 0);
+
+ DBG_BUGON(erofs_blkoff(src, map.m_pa));
+ mdev = (struct erofs_map_dev) {
+ .m_deviceid = map.m_deviceid,
+ .m_pa = map.m_pa,
+ };
+ ret = erofs_map_dev(src, &mdev);
+ if (ret)
+ return ERR_PTR(ret);
+ /* The parent primary device precedes its inherited external devices. */
+ device_id = mdev.m_deviceid + 1;
+ DBG_BUGON(device_id > dst->extra_devices);
+ dif = mdev.m_deviceid ? &src->devs[mdev.m_deviceid - 1] : &src->dif0;
+ startblk = erofs_blknr(src, mdev.m_pa);
+ nblocks = BLK_ROUND_UP(src, map.m_plen);
+ if (startblk >= dif->blocks || nblocks > dif->blocks - startblk) {
+ if (mdev.m_deviceid)
+ erofs_err("%s: parent chunk range exceeds external device %u",
+ ctx->inode->i_srcpath, mdev.m_deviceid);
+ else
+ erofs_err("%s: parent chunk range exceeds the primary device",
+ ctx->inode->i_srcpath);
+ return ERR_PTR(-EFSCORRUPTED);
+ }
+ /* map describes the parent chunk, which may cover several output chunks. */
+ return erofs_get_unhashed_chunk(dst, device_id,
+ startblk + erofs_blknr(src, within), 0);
+}
+
+static int erofs_file_delta_chunk_prepare(struct erofs_file_delta_apply_ctx *ctx)
+{
+ const struct erofs_inode *parent = ctx->parent;
+ struct erofs_sb_info *sbi = ctx->inode->sbi;
+ struct erofs_file_delta_chunk *state;
+ unsigned int chunkbits;
+ erofs_off_t chunksize;
+ u64 count;
+ size_t i;
+ int ret;
+
+ DBG_BUGON(parent->datalayout != EROFS_INODE_CHUNK_BASED);
+ if (!(parent->u.chunkformat & EROFS_CHUNK_FORMAT_INDEXES)) {
+ erofs_err("%s: parent file has no explicit chunk indexes",
+ ctx->inode->i_srcpath);
+ return -EOPNOTSUPP;
+ }
+ if (parent->u.chunkformat & EROFS_CHUNK_FORMAT_48BIT) {
+ erofs_err("%s: file-delta does not support 48-bit chunk indexes",
+ ctx->inode->i_srcpath);
+ return -EOPNOTSUPP;
+ }
+ DBG_BUGON(parent->sbi->blkszbits != sbi->blkszbits);
+ chunkbits = parent->u.chunkbits;
+ if (ctx->update_unit) {
+ DBG_BUGON(ctx->update_unit & (ctx->update_unit - 1) ||
+ ctx->update_unit < erofs_blksiz(sbi));
+ chunkbits = min_t(unsigned int, chunkbits, ilog2(ctx->update_unit));
+ }
+ if (chunkbits < sbi->blkszbits ||
+ chunkbits - sbi->blkszbits > EROFS_CHUNK_FORMAT_BLKBITS_MASK)
+ return -EINVAL;
+ chunksize = 1ULL << chunkbits;
+ count = parent->i_size / chunksize + !!(parent->i_size % chunksize);
+ if (count > INT_MAX / sizeof(struct erofs_inode_chunk_index) ||
+ count > SIZE_MAX / sizeof(struct erofs_chunkitem *)) {
+ erofs_err("%s: too many chunk indexes", ctx->inode->i_srcpath);
+ return -E2BIG;
+ }
+ state = calloc(1, sizeof(*state));
+ if (!state)
+ return -ENOMEM;
+ ctx->private = state;
+ state->parent_cursor = *parent;
+ state->chunkbits = chunkbits;
+ state->count = count;
+ state->indexes = calloc(count, sizeof(*state->indexes));
+ if (!state->indexes)
+ return -ENOMEM;
+ ret = erofs_blob_init(sbi, 0);
+ if (ret)
+ return ret;
+ for (i = 0; i < state->count; ++i) {
+ struct erofs_chunkitem *chunk;
+
+ chunk = erofs_file_delta_chunk_import(ctx, (erofs_off_t)i << chunkbits);
+ if (IS_ERR(chunk))
+ return PTR_ERR(chunk);
+ state->indexes[i] = chunk;
+ }
+ return 0;
+}
+
+static bool erofs_file_delta_chunk_is_zero(const void *buf, size_t len)
+{
+ const u8 *p = buf;
+
+ while (len--)
+ if (*p++)
+ return false;
+ return true;
+}
+
+static int erofs_file_delta_chunk_update(struct erofs_file_delta_apply_ctx *ctx,
+ const struct erofs_file_delta_update *update)
+{
+ struct erofs_file_delta_chunk *state = ctx->private;
+ struct erofs_sb_info *sbi = ctx->inode->sbi;
+ const erofs_off_t chunksize = 1ULL << state->chunkbits;
+ const u8 *data = update->data;
+ size_t pos, len;
+
+ DBG_BUGON(update->offset & (chunksize - 1) ||
+ ((update->length & (chunksize - 1)) &&
+ update->offset + update->length != ctx->inode->i_size));
+ for (pos = 0; pos < update->length; pos += len) {
+ struct erofs_chunkitem *chunk;
+
+ len = min_t(erofs_off_t, chunksize, update->length - pos);
+ if (erofs_file_delta_chunk_is_zero(data + pos, len))
+ chunk = erofs_get_unhashed_chunk(sbi, 0, EROFS_NULL_ADDR, 0);
+ else
+ chunk = erofs_blob_write_chunk(sbi, data + pos, len, 0);
+ if (IS_ERR(chunk))
+ return PTR_ERR(chunk);
+ state->indexes[(update->offset + pos) >> state->chunkbits] = chunk;
+ }
+ return 0;
+}
+
+static int erofs_file_delta_chunk_finish(struct erofs_file_delta_apply_ctx *ctx)
+{
+ struct erofs_file_delta_chunk *state = ctx->private;
+ struct erofs_inode *inode = ctx->inode;
+
+ inode->chunkindexes = state->indexes;
+ inode->extent_isize = state->count * sizeof(struct erofs_inode_chunk_index);
+ inode->datalayout = EROFS_INODE_CHUNK_BASED;
+ inode->u.chunkbits = state->chunkbits;
+ inode->u.chunkformat = EROFS_CHUNK_FORMAT_INDEXES |
+ (state->chunkbits - inode->sbi->blkszbits);
+ state->indexes = NULL;
+ return 0;
+}
+
+static void erofs_file_delta_chunk_release(struct erofs_file_delta_apply_ctx *ctx)
+{
+ struct erofs_file_delta_chunk *state = ctx->private;
+
+ if (!state)
+ return;
+ /* Chunk objects belong to the destination chunk manager. */
+ free(state->indexes);
+ free(state);
+ ctx->private = NULL;
+}
+
+const struct erofs_file_delta_layout_ops erofs_file_delta_chunk_layout_ops = {
+ .prepare = erofs_file_delta_chunk_prepare,
+ .update = erofs_file_delta_chunk_update,
+ .finish = erofs_file_delta_chunk_finish,
+ .release = erofs_file_delta_chunk_release,
+};
diff --git a/lib/liberofs_chunk.h b/lib/liberofs_chunk.h
index 16f3d05..8dfeeb2 100644
--- a/lib/liberofs_chunk.h
+++ b/lib/liberofs_chunk.h
@@ -17,6 +17,8 @@ extern "C"
struct erofs_chunkitem *erofs_get_unhashed_chunk(struct erofs_sb_info *sbi,
unsigned int device_id, erofs_blk_t blkaddr,
erofs_off_t sourceoffset);
+struct erofs_chunkitem *erofs_blob_write_chunk(struct erofs_sb_info *sbi,
+ const void *buf, erofs_off_t chunksize, int device_id);
void erofs_inode_fixup_chunkformat(struct erofs_inode *inode);
int erofs_write_chunk_indexes(struct erofs_inode *inode, struct erofs_vfile *vf,
erofs_off_t off);
diff --git a/lib/liberofs_file_delta.h b/lib/liberofs_file_delta.h
new file mode 100644
index 0000000..e03bd6c
--- /dev/null
+++ b/lib/liberofs_file_delta.h
@@ -0,0 +1,81 @@
+/* SPDX-License-Identifier: GPL-2.0+ OR MIT */
+#ifndef __EROFS_LIB_LIBEROFS_FILE_DELTA_H
+#define __EROFS_LIB_LIBEROFS_FILE_DELTA_H
+
+#include "erofs/internal.h"
+
+struct erofs_file_delta;
+
+struct erofs_file_delta_source_ops {
+ const char *name;
+ int (*open)(struct erofs_file_delta *delta, u64 target_size,
+ erofs_off_t unit_size);
+ void (*close)(struct erofs_file_delta *delta);
+ /* Updates are ordered by logical unit, without duplicates. */
+ size_t (*count)(const struct erofs_file_delta *delta);
+ /* Read one complete source unit into the caller's buffer. */
+ int (*read)(const struct erofs_file_delta *delta, size_t index,
+ u64 *unit_index, void *buf);
+};
+
+struct erofs_file_delta {
+ struct list_head list;
+ const struct erofs_file_delta_source_ops *ops;
+ char *target;
+ char *source_path;
+ struct erofs_inode lower_inode;
+ erofs_off_t unit_size;
+ void *private;
+ bool applied;
+};
+
+struct erofs_file_delta_manager {
+ struct list_head deltas;
+};
+
+struct erofs_file_delta_update {
+ erofs_off_t offset;
+ size_t length;
+ /* Borrowed until the layout's update callback returns. */
+ const void *data;
+};
+
+struct erofs_file_delta_apply_ctx {
+ const struct erofs_inode *parent;
+ /* A separate output inode, already bound to the destination sbi. */
+ struct erofs_inode *inode;
+ /* Logical source unit size; zero for inheritance without updates. */
+ erofs_off_t update_unit;
+ void *private;
+};
+
+struct erofs_file_delta_layout_ops {
+ int (*prepare)(struct erofs_file_delta_apply_ctx *ctx);
+ int (*update)(struct erofs_file_delta_apply_ctx *ctx,
+ const struct erofs_file_delta_update *update);
+ /* Transfer the completed layout to the output inode on success. */
+ int (*finish)(struct erofs_file_delta_apply_ctx *ctx);
+ /* Called after every prepare attempt, including partial failure. */
+ void (*release)(struct erofs_file_delta_apply_ctx *ctx);
+};
+
+extern const struct erofs_file_delta_layout_ops
+ erofs_file_delta_chunk_layout_ops;
+
+int erofs_file_delta_parse_spec(struct list_head *deltas, const char *arg,
+ const struct erofs_file_delta_source_ops * const *backends);
+void erofs_file_delta_cleanup(struct list_head *deltas);
+
+/* Prepare the collected tasks; the caller also cleans up on failure. */
+int erofs_file_delta_manager_init(struct erofs_file_delta_manager *mgr,
+ struct erofs_sb_info *parent, erofs_off_t unit_size);
+int erofs_file_delta_manager_finish(struct erofs_file_delta_manager *mgr);
+
+struct erofs_file_delta *erofs_file_delta_lookup_target(
+ struct erofs_file_delta_manager *mgr, const struct erofs_inode *inode);
+/* A NULL delta inherits the parent mapping without applying updates. */
+int erofs_file_delta_apply(struct erofs_file_delta *delta,
+ const struct erofs_file_delta_layout_ops *ops,
+ const struct erofs_inode *parent, struct erofs_inode *inode);
+
+#endif
diff --git a/lib/rebuild.c b/lib/rebuild.c
index 7d28c1e..9c38204 100644
--- a/lib/rebuild.c
+++ b/lib/rebuild.c
@@ -15,6 +15,7 @@
#include "erofs/internal.h"
#include "erofs/io.h"
#include "liberofs_chunk.h"
+#include "liberofs_file_delta.h"
#include "liberofs_rebuild.h"
#include "liberofs_uuid.h"
@@ -198,7 +199,7 @@ static int erofs_rebuild_load_chunk_index(struct erofs_sb_info *dst_sb,
.buf = __EROFS_BUF_INITIALIZER,
};
- map.m_la = i << chunkbits;
+ map.m_la = (erofs_off_t)i << chunkbits;
ret = erofs_map_blocks(inode, &map, 0);
if (ret)
goto err;
@@ -233,6 +234,28 @@ static int erofs_rebuild_write_blob_index(struct erofs_sb_info *dst_sb,
return erofs_rebuild_load_chunk_index(dst_sb, inode);
}
+static int erofs_rebuild_apply_file_delta(struct erofs_sb_info *dst_sb,
+ struct erofs_inode *inode)
+{
+ const struct erofs_file_delta_layout_ops *ops;
+ struct erofs_file_delta *delta;
+ struct erofs_inode parent = *inode;
+
+ switch (parent.datalayout) {
+ case EROFS_INODE_CHUNK_BASED:
+ ops = &erofs_file_delta_chunk_layout_ops;
+ break;
+ default:
+ erofs_err("%s: unsupported file-delta datalayout %u",
+ inode->i_srcpath, parent.datalayout);
+ return -EOPNOTSUPP;
+ }
+ delta = erofs_file_delta_lookup_target(dst_sb->file_delta_mgr, &parent);
+ inode->sbi = dst_sb;
+ return erofs_file_delta_apply(delta, ops,
+ delta ? &delta->lower_inode : &parent, inode);
+}
+
static int erofs_rebuild_write_full_data(struct erofs_inode *inode)
{
struct erofs_sb_info *src_sbi = inode->sbi;
@@ -339,9 +362,12 @@ static int erofs_rebuild_update_inode(struct erofs_sb_info *dst_sb,
inode->u.i_blkaddr = EROFS_NULL_ADDR;
break;
}
- if (datamode == EROFS_REBUILD_DATA_BLOB_INDEX)
- err = erofs_rebuild_write_blob_index(dst_sb, inode);
- else if (datamode == EROFS_REBUILD_DATA_RESVSP)
+ if (datamode == EROFS_REBUILD_DATA_BLOB_INDEX) {
+ if (dst_sb->file_delta_mgr)
+ err = erofs_rebuild_apply_file_delta(dst_sb, inode);
+ else
+ err = erofs_rebuild_write_blob_index(dst_sb, inode);
+ } else if (datamode == EROFS_REBUILD_DATA_RESVSP)
inode->datasource = EROFS_INODE_DATA_SOURCE_RESVSP;
else if (datamode == EROFS_REBUILD_DATA_FULL)
err = erofs_rebuild_write_full_data(inode);
diff --git a/man/mkfs.erofs.1 b/man/mkfs.erofs.1
index 65ec807..dcac6f0 100644
--- a/man/mkfs.erofs.1
+++ b/man/mkfs.erofs.1
@@ -339,6 +339,62 @@ Note that \fBrvsp\fR takes precedence over \fB--tar=i\fR or \fB--tar=headerball\
Only \fBrvsp\fR is supported.
.RE
.TP
+.BI "\-\-file-delta=" FORMAT:TARGET:SOURCE
+Rebuild exactly one parent image while applying a backend-specific delta
+\fISOURCE\fR to \fITARGET\fR. This option requires
+.BR \-Eforce-chunk-indexes ,
+.BR \-\-chunksize .
+It cannot be used with
+.BR \-\-incremental
+or non-default rebuild data modes, and it cannot be combined with
+.BR \-\-blobdev .
+It also rejects
+.BR \-\-xattr-inode-digest
+and parents carrying inode-digest xattrs because historical image devices are
+not opened to recompute full-file digests.
+
+\fITARGET\fR is resolved from the root of the parent image. It must name
+a non-empty regular file using \fBCHUNK_BASED\fR layout with
+explicit chunk indexes.
+Updating a hard-linked target updates all links to that inode.
+\fB\-\-chunksize\fR specifies the logical update unit of \fISOURCE\fR;
+the parent may use a different chunk size, including merged larger chunks.
+A target file can be specified only once, even through different path spellings.
+TARGET cannot contain a colon.
+\fIFORMAT\fR determines how \fISOURCE\fR is validated and read.
+
+File-delta currently supports only the non-48-bit format. Parents and
+chunk indexes using 48-bit features are rejected, as are options enabling
+those features. The output must fit within the 32-bit block range;
+file-delta does not automatically enable 48-bit support.
+
+Each delta record replaces a logical byte range. The output uses the smaller
+of the parent chunk size and the source update unit for the entire target
+inode. Inherited mappings are split as needed without reading or rewriting
+their data. Files without updates retain their parent chunk size.
+\fB\-Eforce-chunk-indexes\fR selects explicit indexes and allows contiguous
+chunks to be merged when building a parent image.
+Non-zero payload is stored compactly in the output image's primary device;
+identical chunks in this new
+layer can share one physical location. The output inode inherits every clean
+chunk index from the parent and points changed non-zero chunks to its own
+primary device. Changed zero chunks become holes and consume no data space.
+
+The parent's primary device becomes external device 1. Its existing external
+devices are shifted up by one slot, and every inherited chunk device ID is
+incremented accordingly. Historical images are therefore ordered newest
+first: the direct parent is device 1, its parent is device 2, and so on. The
+output image depends on all of them. Each inherited mapping must fit within
+its source device, and the output is limited to 256 external devices.
+\fISOURCE\fR is build input only and is not referenced by the output image.
+
+The filesystem block size is derived from the parent, so sub-4096-byte
+parent and delta chunk geometries do not require a redundant \fB-b\fR option.
+
+The source specification is the remainder after the second colon, so it may
+itself contain colons. Repeated file-delta options must refer to distinct target
+files. A source may be reused when the selected backend permits it.
+.TP
.BI "\-\-max-extent-bytes=" #
Specify maximum decompressed extent size in bytes. The default is unlimited.
.TP
diff --git a/mkfs/main.c b/mkfs/main.c
index f0d3d8c..7598f0a 100644
--- a/mkfs/main.c
+++ b/mkfs/main.c
@@ -26,6 +26,7 @@
#include "erofs/compress_hints.h"
#include "../lib/compressor.h"
#include "../lib/liberofs_chunk.h"
+#include "../lib/liberofs_file_delta.h"
#include "../lib/liberofs_gzran.h"
#include "../lib/liberofs_metabox.h"
#include "../lib/liberofs_oci.h"
@@ -105,6 +106,7 @@ static struct option long_options[] = {
{"MZ", optional_argument, NULL, 537},
{"xattr-prefix", required_argument, NULL, 538},
{"xattr-inode-digest", optional_argument, NULL, 539},
+ {"file-delta", required_argument, NULL, 540},
{0, 0, 0, 0},
};
@@ -201,6 +203,7 @@ static void usage(int argc, char **argv)
" --dsunit=# align all data block addresses to multiples of #\n"
" --exclude-path=X avoid including file X (X = exact literal path)\n"
" --exclude-regex=X avoid including files that match X (X = regular expression)\n"
+ " --file-delta=X rebuild TARGET with X=FORMAT:TARGET:SOURCE into the output image\n"
#ifdef HAVE_LIBSELINUX
" --file-contexts=X specify a file contexts file to setup selinux labels\n"
#endif
@@ -328,6 +331,12 @@ static enum {
static unsigned int rebuild_src_count;
static LIST_HEAD(rebuild_src_list);
+static struct erofs_file_delta_manager file_delta_mgr = {
+ .deltas = LIST_HEAD_INIT(file_delta_mgr.deltas),
+};
+static const struct erofs_file_delta_source_ops * const file_delta_backends[] = {
+ NULL,
+};
static u8 fixeduuid[16];
static bool valid_fixeduuid;
static unsigned int dsunit;
@@ -1497,6 +1506,15 @@ static int mkfs_parse_options_cfg(struct erofs_importer_params *params,
return err;
}
break;
+ case 540:
+ err = erofs_file_delta_parse_spec(&file_delta_mgr.deltas, optarg,
+ file_delta_backends);
+ if (err) {
+ erofs_err("invalid --file-delta=%s: %s", optarg,
+ erofs_strerror(err));
+ return err;
+ }
+ break;
case 'V':
version();
exit(0);
@@ -1552,6 +1570,26 @@ static int mkfs_parse_options_cfg(struct erofs_importer_params *params,
return err;
}
+ if (!list_empty(&file_delta_mgr.deltas)) {
+ if (erofs_sb_has_48bit(&g_sbi) || mkfs_dot_omitted) {
+ erofs_err("--file-delta does not support 48-bit features");
+ return -EOPNOTSUPP;
+ }
+ if (incremental_mode ||
+ source_mode != EROFS_MKFS_SOURCE_REBUILD ||
+ rebuild_src_count != 1 || !mkfscfg.chunkbits ||
+ cfg.c_force_chunkformat != FORCE_INODE_CHUNK_INDEXES ||
+ dataimport_mode != EROFS_MKFS_DATA_IMPORT_DEFAULT) {
+ erofs_err("--file-delta requires one rebuild parent, --chunksize, "
+ "-Eforce-chunk-indexes and the default rebuild data mode");
+ return -EINVAL;
+ }
+ if (erofs_sb_has_ishare_xattrs(&g_sbi)) {
+ erofs_err("--xattr-inode-digest is unsupported with --file-delta");
+ return -EOPNOTSUPP;
+ }
+ }
+
if (quiet) {
cfg.c_dbg_lvl = EROFS_ERR;
cfg.c_showprogress = false;
@@ -1567,7 +1605,8 @@ static int mkfs_parse_options_cfg(struct erofs_importer_params *params,
params->pclusterblks_max = pclustersize_max >> mkfs_blkszbits;
params->pclusterblks_def = params->pclusterblks_max;
}
- if (mkfscfg.chunkbits && mkfscfg.chunkbits < mkfs_blkszbits) {
+ if (list_empty(&file_delta_mgr.deltas) && mkfscfg.chunkbits &&
+ mkfscfg.chunkbits < mkfs_blkszbits) {
erofs_err("chunksize %u must be larger than block size",
1u << mkfscfg.chunkbits);
return -EINVAL;
@@ -1683,6 +1722,39 @@ static int erofs_mkfs_rebuild_load_trees(struct erofs_inode *root)
return -EINVAL;
}
+ if (!list_empty(&file_delta_mgr.deltas)) {
+ unsigned int output_devices;
+
+ src = list_first_entry(&rebuild_src_list,
+ struct erofs_sb_info, list);
+ DBG_BUGON(datamode != EROFS_REBUILD_DATA_BLOB_INDEX ||
+ rebuild_src_count != 1);
+
+ output_devices = src->extra_devices + 1;
+ ret = erofs_mkfs_init_devices(&g_sbi, output_devices);
+ if (ret)
+ return ret;
+ g_sbi.devs[0].blocks = src->dif0.blocks;
+ erofs_uuid_unparse_as_tag(src->uuid,
+ (char *)g_sbi.devs[0].tag);
+ for (idx = 0; idx < src->extra_devices; ++idx) {
+ g_sbi.devs[idx + 1].blocks = src->devs[idx].blocks;
+ memcpy(g_sbi.devs[idx + 1].tag, src->devs[idx].tag,
+ sizeof(g_sbi.devs[idx + 1].tag));
+ }
+ ret = erofs_file_delta_manager_init(&file_delta_mgr, src,
+ 1ULL << mkfscfg.chunkbits);
+ if (ret)
+ return ret;
+ root->sbi->file_delta_mgr = &file_delta_mgr;
+ src->xamgr = g_sbi.xamgr;
+ ret = erofs_rebuild_load_tree(root, src, datamode);
+ src->xamgr = NULL;
+ if (ret)
+ return ret;
+ return erofs_file_delta_manager_finish(&file_delta_mgr);
+ }
+
list_for_each_entry(src, &rebuild_src_list, list) {
src->xamgr = g_sbi.xamgr;
ret = erofs_rebuild_load_tree(root, src, datamode);
@@ -1856,6 +1928,31 @@ int main(int argc, char **argv)
goto exit;
}
mkfs_blkszbits = src->blkszbits;
+ if (!list_empty(&file_delta_mgr.deltas)) {
+ if (erofs_sb_has_ishare_xattrs(src)) {
+ erofs_err("file-delta parent uses inode digest xattrs");
+ err = -EOPNOTSUPP;
+ goto exit;
+ }
+ if (mkfscfg.chunkbits < mkfs_blkszbits) {
+ erofs_err("chunksize %u must be at least parent block size %u",
+ 1U << mkfscfg.chunkbits,
+ 1U << mkfs_blkszbits);
+ err = -EINVAL;
+ goto exit;
+ }
+ if (erofs_sb_has_48bit(src)) {
+ erofs_err("file-delta does not support 48-bit parents");
+ err = -EOPNOTSUPP;
+ goto exit;
+ }
+ if (src->extra_devices >= EROFS_MAX_BLOB_DEVS) {
+ erofs_err("file-delta supports at most %u external devices",
+ EROFS_MAX_BLOB_DEVS);
+ err = -E2BIG;
+ goto exit;
+ }
+ }
} else if (mkfs_oci_tarindex_mode) {
mkfs_blkszbits = 9;
tar_index_512b = true;
@@ -1942,7 +2039,8 @@ int main(int argc, char **argv)
importer_params.ddev_id_def = mkfscfg.blobdev_path ? 1 : 0;
}
- if (tar_index_512b || mkfscfg.chunkbits) {
+ if ((tar_index_512b || mkfscfg.chunkbits) &&
+ list_empty(&file_delta_mgr.deltas)) {
err = erofs_blob_init(&g_sbi, mkfscfg.chunkbits);
if (err)
goto exit;
@@ -2059,6 +2157,13 @@ int main(int argc, char **argv)
erofs_iput(root);
root = NULL;
+ if (g_sbi.file_delta_mgr &&
+ (erofs_sb_has_48bit(&g_sbi) || g_sbi.dif0.blocks > UINT32_MAX)) {
+ erofs_err("file-delta output requires 48-bit support");
+ err = -EFBIG;
+ goto exit;
+ }
+
err = erofs_writesb(&g_sbi);
if (err)
goto exit;
@@ -2077,6 +2182,8 @@ int main(int argc, char **argv)
exit:
if (root)
erofs_iput(root);
+ g_sbi.file_delta_mgr = NULL;
+ erofs_file_delta_cleanup(&file_delta_mgr.deltas);
z_erofs_dedupe_exit();
blklst = erofs_blocklist_close();
if (blklst)
--
2.55.0
^ permalink raw reply related [flat|nested] 3+ messages in thread* [PATCH RFC 2/2] erofs-utils: mkfs: support dm-persistent file deltas
2026-09-16 9:10 [PATCH RFC 0/2] erofs-utils: add dm-persistent file delta support Yifan Zhao
2026-09-16 9:10 ` [PATCH RFC 1/2] erofs-utils: mkfs: add file-delta framework Yifan Zhao
@ 2026-09-16 9:10 ` Yifan Zhao
1 sibling, 0 replies; 3+ messages in thread
From: Yifan Zhao @ 2026-09-16 9:10 UTC (permalink / raw)
To: linux-erofs
Parse and validate version 1 dm-persistent snapshot exception stores, then
expose their logical update units through the file-delta source operations.
The source unit size is independent of the parent chunk geometry.
Assisted-by: Codex:gpt-5.6-sol
Signed-off-by: Yifan Zhao <yifan.yfzhao@foxmail.com>
---
lib/Makefile.am | 3 +-
lib/dm_persistent.c | 396 +++++++++++++++++++++++++++++++++++
lib/liberofs_dm_persistent.h | 9 +
man/mkfs.erofs.1 | 15 ++
mkfs/main.c | 2 +
5 files changed, 424 insertions(+), 1 deletion(-)
create mode 100644 lib/dm_persistent.c
create mode 100644 lib/liberofs_dm_persistent.h
diff --git a/lib/Makefile.am b/lib/Makefile.am
index 3d562f4..24e7d0b 100644
--- a/lib/Makefile.am
+++ b/lib/Makefile.am
@@ -24,6 +24,7 @@ noinst_HEADERS = $(top_srcdir)/include/erofs_fs.h \
$(top_srcdir)/lib/liberofs_cache.h \
$(top_srcdir)/lib/liberofs_chunk.h \
$(top_srcdir)/lib/liberofs_compress.h \
+ $(top_srcdir)/lib/liberofs_dm_persistent.h \
$(top_srcdir)/lib/liberofs_file_delta.h \
$(top_srcdir)/lib/liberofs_fragments.h \
$(top_srcdir)/lib/liberofs_private.h \
@@ -46,7 +47,7 @@ liberofs_la_SOURCES = config.c io.c cache.c super.c inode.c xattr.c exclude.c \
fragments.c dedupe.c uuid_unparse.c uuid.c tar.c \
block_list.c rebuild.c diskbuf.c bitops.c dedupe_ext.c \
vmdk.c metabox.c global.c importer.c base64.c file_delta.c \
- file_delta_chunk.c
+ file_delta_chunk.c dm_persistent.c
liberofs_la_CFLAGS = -Wall ${libuuid_CFLAGS} -I$(top_srcdir)/include
liberofs_la_LIBADD = $(LIBEROFS_LIBS)
diff --git a/lib/dm_persistent.c b/lib/dm_persistent.c
new file mode 100644
index 0000000..0b0cfeb
--- /dev/null
+++ b/lib/dm_persistent.c
@@ -0,0 +1,396 @@
+// SPDX-License-Identifier: GPL-2.0+ OR MIT
+#define _GNU_SOURCE
+#include <errno.h>
+#include <fcntl.h>
+#include <limits.h>
+#include <stdlib.h>
+#include <sys/ioctl.h>
+#include <sys/stat.h>
+#include <unistd.h>
+#include <config.h>
+#ifdef HAVE_LINUX_FS_H
+#include <linux/fs.h>
+#endif
+#include "erofs/print.h"
+#include "liberofs_dm_persistent.h"
+
+#ifndef EFSCORRUPTED
+#ifdef EUCLEAN
+#define EFSCORRUPTED EUCLEAN
+#else
+#define EFSCORRUPTED EIO
+#endif
+#endif
+
+#ifndef O_CLOEXEC
+#define O_CLOEXEC 0
+#endif
+#ifndef O_BINARY
+#define O_BINARY 0
+#endif
+
+#define EROFS_DM_SNAP_MAGIC 0x70416e53U
+#define EROFS_DM_DISK_VERSION 1U
+#define EROFS_DM_SECTOR_SIZE 512U
+
+struct erofs_dm_exception {
+ u64 old_chunk;
+ u64 new_chunk;
+};
+
+struct erofs_dm_delta {
+ int fd;
+ u64 store_size;
+ u64 target_size;
+ u64 chunk_bytes;
+ u64 exceptions_per_area;
+ size_t nr_exceptions;
+ struct erofs_dm_exception *exceptions;
+};
+
+static void erofs_dm_persistent_close(struct erofs_dm_delta *delta);
+
+struct erofs_dm_disk_header {
+ __le32 magic;
+ __le32 valid;
+ __le32 version;
+ __le32 chunk_size;
+} __packed;
+
+struct erofs_dm_disk_exception {
+ __le64 old_chunk;
+ __le64 new_chunk;
+} __packed;
+
+static int erofs_dm_get_store_size(int fd, const struct stat *st, u64 *size)
+{
+ if (S_ISREG(st->st_mode)) {
+ if (st->st_size < 0)
+ return -EOVERFLOW;
+ *size = st->st_size;
+ return 0;
+ }
+ if (!S_ISBLK(st->st_mode))
+ return -ENOTSUP;
+
+#ifdef BLKGETSIZE64
+ if (!ioctl(fd, BLKGETSIZE64, size))
+ return 0;
+#endif
+#ifdef BLKGETSIZE
+ {
+ unsigned long sectors;
+
+ if (!ioctl(fd, BLKGETSIZE, §ors)) {
+ if ((u64)sectors > UINT64_MAX >> 9)
+ return -EOVERFLOW;
+ *size = (u64)sectors << 9;
+ return 0;
+ }
+ }
+#endif
+#if defined(BLKGETSIZE64) || defined(BLKGETSIZE)
+ return -errno;
+#else
+ return -ENOTSUP;
+#endif
+}
+
+static int erofs_dm_pread_exact(int fd, void *buf, size_t len, u64 offset)
+{
+ size_t done = 0;
+
+ if (offset > INT64_MAX || len > INT64_MAX - offset)
+ return -EOVERFLOW;
+ while (done < len) {
+ ssize_t ret;
+
+#ifdef HAVE_PREAD64
+ ret = pread64(fd, (char *)buf + done, len - done,
+ (off64_t)(offset + done));
+#else
+ ret = pread(fd, (char *)buf + done, len - done,
+ (off_t)(offset + done));
+#endif
+ if (ret < 0) {
+ if (errno == EINTR)
+ continue;
+ return -errno;
+ }
+ if (!ret)
+ return -EIO;
+ done += ret;
+ }
+ return 0;
+}
+
+static int erofs_dm_exception_cmp(const void *a, const void *b)
+{
+ const struct erofs_dm_exception *ea = a, *eb = b;
+
+ return (ea->old_chunk > eb->old_chunk) -
+ (ea->old_chunk < eb->old_chunk);
+}
+
+static int erofs_dm_append_exception(struct erofs_dm_delta *delta,
+ size_t *capacity, u64 max_target_chunks,
+ u64 old_chunk, u64 new_chunk)
+{
+ struct erofs_dm_exception *p;
+ size_t new_capacity, bytes;
+
+ if (delta->nr_exceptions >= max_target_chunks)
+ return -EINVAL;
+ if (delta->nr_exceptions < *capacity)
+ goto add;
+
+ new_capacity = *capacity ? *capacity << 1 : 64;
+ if (new_capacity < *capacity)
+ return -E2BIG;
+ if (new_capacity > max_target_chunks)
+ new_capacity = max_target_chunks;
+ if (new_capacity <= *capacity ||
+ __builtin_mul_overflow(new_capacity, sizeof(*p), &bytes))
+ return -E2BIG;
+ p = realloc(delta->exceptions, bytes);
+ if (!p)
+ return -ENOMEM;
+ delta->exceptions = p;
+ *capacity = new_capacity;
+add:
+ delta->exceptions[delta->nr_exceptions++] =
+ (struct erofs_dm_exception) {
+ .old_chunk = old_chunk,
+ .new_chunk = new_chunk,
+ };
+ return 0;
+}
+
+static int erofs_dm_parse_exceptions(struct erofs_dm_delta *delta)
+{
+ u64 max_target_chunks = delta->target_size / delta->chunk_bytes;
+ struct erofs_dm_disk_exception *area_buf;
+ size_t capacity = 0;
+ u64 area, stride;
+ int ret = 0;
+ bool partial = delta->target_size % delta->chunk_bytes;
+
+ stride = delta->exceptions_per_area + 1;
+ max_target_chunks += partial;
+
+ area_buf = malloc(delta->chunk_bytes);
+ if (!area_buf)
+ return -ENOMEM;
+
+ for (area = 0; ; ++area) {
+ u64 metadata_chunk, metadata_offset, metadata_end;
+ u64 i;
+
+ if (__builtin_mul_overflow(stride, area, &metadata_chunk) ||
+ check_add_overflow(metadata_chunk, (u64)1, &metadata_chunk) ||
+ __builtin_mul_overflow(metadata_chunk, delta->chunk_bytes,
+ &metadata_offset) ||
+ check_add_overflow(metadata_offset, delta->chunk_bytes,
+ &metadata_end) ||
+ metadata_end > delta->store_size) {
+ ret = -EFSCORRUPTED;
+ break;
+ }
+ ret = erofs_dm_pread_exact(delta->fd, area_buf,
+ delta->chunk_bytes, metadata_offset);
+ if (ret) {
+ if (ret == -EIO)
+ ret = -EFSCORRUPTED;
+ break;
+ }
+
+ for (i = 0; i < delta->exceptions_per_area; ++i) {
+ u64 old_chunk = le64_to_cpu(area_buf[i].old_chunk);
+ u64 new_chunk = le64_to_cpu(area_buf[i].new_chunk);
+ u64 payload_offset, payload_end;
+
+ if (!new_chunk)
+ goto terminated;
+ if (new_chunk % stride == 1 ||
+ __builtin_mul_overflow(new_chunk,
+ delta->chunk_bytes,
+ &payload_offset) ||
+ check_add_overflow(payload_offset, delta->chunk_bytes,
+ &payload_end) ||
+ payload_end > delta->store_size) {
+ ret = -EFSCORRUPTED;
+ goto out;
+ }
+ if (old_chunk >= max_target_chunks) {
+ ret = -EINVAL;
+ goto out;
+ }
+ ret = erofs_dm_append_exception(delta, &capacity,
+ max_target_chunks, old_chunk, new_chunk);
+ if (ret)
+ goto out;
+ }
+ }
+ goto out;
+
+terminated:
+ if (delta->nr_exceptions > 1)
+ qsort(delta->exceptions, delta->nr_exceptions,
+ sizeof(*delta->exceptions), erofs_dm_exception_cmp);
+ for (area = 1; area < delta->nr_exceptions; ++area) {
+ if (delta->exceptions[area - 1].old_chunk ==
+ delta->exceptions[area].old_chunk) {
+ ret = -EINVAL;
+ break;
+ }
+ }
+out:
+ free(area_buf);
+ return ret;
+}
+
+static int erofs_dm_persistent_open(struct erofs_dm_delta *delta,
+ const char *path, u64 target_size,
+ erofs_off_t unit_size)
+{
+ struct erofs_dm_disk_header header;
+ struct stat st;
+ u32 sectors;
+ u64 chunk_bytes;
+ int flags, ret;
+
+ BUILD_BUG_ON(sizeof(header) != 16);
+ BUILD_BUG_ON(sizeof(struct erofs_dm_disk_exception) != 16);
+ *delta = (struct erofs_dm_delta) { .fd = -1 };
+ delta->target_size = target_size;
+
+ delta->fd = open(path, O_RDONLY | O_BINARY | O_CLOEXEC | O_NONBLOCK);
+ if (delta->fd < 0) {
+ ret = -errno;
+ goto err;
+ }
+ if (fstat(delta->fd, &st)) {
+ ret = -errno;
+ goto err;
+ }
+ ret = erofs_dm_get_store_size(delta->fd, &st, &delta->store_size);
+ if (ret)
+ goto err;
+ flags = fcntl(delta->fd, F_GETFL);
+ if (flags < 0 || fcntl(delta->fd, F_SETFL, flags & ~O_NONBLOCK)) {
+ ret = -errno;
+ goto err;
+ }
+ ret = erofs_dm_pread_exact(delta->fd, &header, sizeof(header), 0);
+ if (ret) {
+ if (ret == -EIO)
+ ret = -EFSCORRUPTED;
+ goto err;
+ }
+ if (le32_to_cpu(header.magic) != EROFS_DM_SNAP_MAGIC ||
+ le32_to_cpu(header.version) != EROFS_DM_DISK_VERSION) {
+ ret = -EFSCORRUPTED;
+ goto err;
+ }
+ if (le32_to_cpu(header.valid) != 1) {
+ ret = -EINVAL;
+ goto err;
+ }
+
+ sectors = le32_to_cpu(header.chunk_size);
+ if (!sectors || (sectors & (sectors - 1)) ||
+ sectors > (INT_MAX >> 9)) {
+ ret = -EFSCORRUPTED;
+ goto err;
+ }
+ chunk_bytes = (u64)sectors * EROFS_DM_SECTOR_SIZE;
+ if (chunk_bytes != unit_size) {
+ erofs_err("file-delta source %s chunk size %llu does not match %llu",
+ path, chunk_bytes | 0ULL, unit_size | 0ULL);
+ ret = -EINVAL;
+ goto err;
+ }
+ delta->chunk_bytes = chunk_bytes;
+ delta->exceptions_per_area = chunk_bytes /
+ sizeof(struct erofs_dm_disk_exception);
+ ret = erofs_dm_parse_exceptions(delta);
+ if (ret)
+ goto err;
+ return 0;
+err:
+ erofs_dm_persistent_close(delta);
+ return ret;
+}
+
+static void erofs_dm_persistent_close(struct erofs_dm_delta *delta)
+{
+ if (delta->fd >= 0)
+ close(delta->fd);
+ free(delta->exceptions);
+ *delta = (struct erofs_dm_delta) { .fd = -1 };
+}
+
+static int erofs_dm_persistent_read_payload(const struct erofs_dm_delta *delta,
+ size_t exception_index, void *buf)
+{
+ u64 offset = delta->exceptions[exception_index].new_chunk *
+ delta->chunk_bytes;
+
+ return erofs_dm_pread_exact(delta->fd, buf, delta->chunk_bytes, offset);
+}
+
+static int erofs_file_delta_dm_open(struct erofs_file_delta *delta,
+ u64 target_size,
+ erofs_off_t unit_size)
+{
+ struct erofs_dm_delta *dm;
+ int ret;
+
+ dm = malloc(sizeof(*dm));
+ if (!dm)
+ return -ENOMEM;
+ ret = erofs_dm_persistent_open(dm, delta->source_path, target_size,
+ unit_size);
+ if (ret) {
+ free(dm);
+ return ret;
+ }
+ delta->private = dm;
+ return 0;
+}
+
+static void erofs_file_delta_dm_close(struct erofs_file_delta *delta)
+{
+ struct erofs_dm_delta *dm = delta->private;
+
+ if (!dm)
+ return;
+ erofs_dm_persistent_close(dm);
+ free(dm);
+ delta->private = NULL;
+}
+
+static size_t erofs_file_delta_dm_count(const struct erofs_file_delta *delta)
+{
+ const struct erofs_dm_delta *dm = delta->private;
+
+ return dm->nr_exceptions;
+}
+
+static int erofs_file_delta_dm_read(const struct erofs_file_delta *delta,
+ size_t index, u64 *chunk, void *buf)
+{
+ const struct erofs_dm_delta *dm = delta->private;
+
+ DBG_BUGON(index >= dm->nr_exceptions || !chunk || !buf);
+ *chunk = dm->exceptions[index].old_chunk;
+ return erofs_dm_persistent_read_payload(dm, index, buf);
+}
+
+const struct erofs_file_delta_source_ops erofs_file_delta_dm_persistent_ops = {
+ .name = "dm-persistent",
+ .open = erofs_file_delta_dm_open,
+ .close = erofs_file_delta_dm_close,
+ .count = erofs_file_delta_dm_count,
+ .read = erofs_file_delta_dm_read,
+};
diff --git a/lib/liberofs_dm_persistent.h b/lib/liberofs_dm_persistent.h
new file mode 100644
index 0000000..1a0f476
--- /dev/null
+++ b/lib/liberofs_dm_persistent.h
@@ -0,0 +1,9 @@
+/* SPDX-License-Identifier: GPL-2.0+ OR MIT */
+#ifndef __EROFS_LIB_LIBEROFS_DM_PERSISTENT_H
+#define __EROFS_LIB_LIBEROFS_DM_PERSISTENT_H
+
+#include "liberofs_file_delta.h"
+
+extern const struct erofs_file_delta_source_ops erofs_file_delta_dm_persistent_ops;
+
+#endif
diff --git a/man/mkfs.erofs.1 b/man/mkfs.erofs.1
index dcac6f0..e38355d 100644
--- a/man/mkfs.erofs.1
+++ b/man/mkfs.erofs.1
@@ -394,6 +394,21 @@ parent and delta chunk geometries do not require a redundant \fB-b\fR option.
The source specification is the remainder after the second colon, so it may
itself contain colons. Repeated file-delta options must refer to distinct target
files. A source may be reused when the selected backend permits it.
+.RS
+.TP
+.B dm-persistent
+Treat \fISOURCE\fR as a Linux dm-persistent snapshot exception store version
+1. Each exception replaces the logical update unit identified by
+\fBold_chunk\fR with the payload at \fBnew_chunk\fR. The dm chunk size and
+\fB\-\-chunksize\fR must be equal. An update to the last partial unit is
+clipped to the target file size.
+
+The source can be a regular file or block device and must remain unchanged for
+the complete build. Its header must have magic \fB0x70416e53\fR,
+\fBvalid=1\fR, version 1, and a power-of-two chunk size.
+FIFOs and other unsupported source object types are rejected without waiting
+for a peer to open them.
+.RE
.TP
.BI "\-\-max-extent-bytes=" #
Specify maximum decompressed extent size in bytes. The default is unlimited.
diff --git a/mkfs/main.c b/mkfs/main.c
index 7598f0a..d6d3e9d 100644
--- a/mkfs/main.c
+++ b/mkfs/main.c
@@ -26,6 +26,7 @@
#include "erofs/compress_hints.h"
#include "../lib/compressor.h"
#include "../lib/liberofs_chunk.h"
+#include "../lib/liberofs_dm_persistent.h"
#include "../lib/liberofs_file_delta.h"
#include "../lib/liberofs_gzran.h"
#include "../lib/liberofs_metabox.h"
@@ -335,6 +336,7 @@ static struct erofs_file_delta_manager file_delta_mgr = {
.deltas = LIST_HEAD_INIT(file_delta_mgr.deltas),
};
static const struct erofs_file_delta_source_ops * const file_delta_backends[] = {
+ &erofs_file_delta_dm_persistent_ops,
NULL,
};
static u8 fixeduuid[16];
--
2.55.0
^ permalink raw reply related [flat|nested] 3+ messages in thread