On 2025/9/28 16:02, ChengyuZhu6 wrote:
From: Chengyu Zhu <[email protected]>
Add support for combining tarindex files with remote OCI blobs
through a new source mechanism. This enables local metadata
storage while keeping blob data in remote registries.
e.g.:
mount.erofs -t erofs.nbd -o \
'oci.blob=13b7e9...,oci.platform=linux/amd64,\
oci.tarindex=ubuntu.erofs,oci.zinfo=ubuntu.zinfo' \
ubuntu:20.04 /mnt
Signed-off-by: Chengyu Zhu <[email protected]>
---
lib/liberofs_oci.h | 2 +
lib/remotes/oci.c | 20 ---
mount/main.c | 397 ++++++++++++++++++++++++++++++++++++++++++++-
3 files changed, 392 insertions(+), 27 deletions(-)
diff --git a/lib/liberofs_oci.h b/lib/liberofs_oci.h
index 71c8879..5298f18 100644
--- a/lib/liberofs_oci.h
+++ b/lib/liberofs_oci.h
@@ -35,6 +35,8 @@ struct ocierofs_config {
char *password;
char *blob_digest;
int layer_index;
+ char *tarindex_path;
+ char *zinfo_path;
};
struct ocierofs_layer_info {
diff --git a/lib/remotes/oci.c b/lib/remotes/oci.c
index b2f1f59..b25e0b2 100644
--- a/lib/remotes/oci.c
+++ b/lib/remotes/oci.c
@@ -1461,19 +1461,6 @@ static void ocierofs_io_close(struct erofs_vfile *vfile)
*(struct ocierofs_iostream **)vfile->payload = NULL;
}
-static int ocierofs_is_erofs_native_image(struct ocierofs_ctx *ctx)
-{
- if (ctx->layer_count > 0 && ctx->layers[0] &&
- ctx->layers[0]->media_type) {
- const char *media_type = ctx->layers[0]->media_type;
- size_t len = strlen(media_type);
-
- if (len >= 6 && strcmp(media_type + len - 6, ".erofs") == 0)
- return 0;
- }
- return -ENOENT;
-}
-
static struct erofs_vfops ocierofs_io_vfops = {
.pread = ocierofs_io_pread,
.read = ocierofs_io_read,
@@ -1497,13 +1484,6 @@ int ocierofs_io_open(struct erofs_vfile *vfile, const
struct ocierofs_config *cf
return err;
}
- err = ocierofs_is_erofs_native_image(ctx);
- if (err) {
- ocierofs_ctx_cleanup(ctx);
- free(ctx);
- return err;
- }
-
oci_iostream = calloc(1, sizeof(*oci_iostream));
if (!oci_iostream) {
ocierofs_ctx_cleanup(ctx);
diff --git a/mount/main.c b/mount/main.c
index eb0dd01..619bf13 100644
--- a/mount/main.c
+++ b/mount/main.c
@@ -16,6 +16,8 @@
#include "erofs/io.h"
#include "../lib/liberofs_nbd.h"
#include "../lib/liberofs_oci.h"
+#include "../lib/liberofs_gzran.h"
+
#ifdef HAVE_LINUX_LOOP_H
#include <linux/loop.h>
#else
@@ -35,6 +37,9 @@ struct loop_info {
#include <sys/sysmacros.h>
#endif
+/* Device boundary probe */
+#define EROFS_IS_DEVICE_PROBE(off) ((off) >= (INT64_MAX >> 9))
I meant just add a macro
#define EROFSMOUNT_NBD_DISK_SIZE (INT64_MAX >> 9)
+
enum erofs_backend_drv {
EROFSAUTO,
EROFSLOCAL,
@@ -141,7 +146,25 @@ static int erofsmount_parse_oci_option(const char *option)
if (!oci_cfg->password)
return -ENOMEM;
} else {
- return -EINVAL;
+ p = strstr(option,
"oci.tarindex=");
+ if (p != NULL) {
+ p +=
strlen("oci.tarindex=");
+
free(oci_cfg->tarindex_path);
+ oci_cfg->tarindex_path
= strdup(p);
+ if
(!oci_cfg->tarindex_path)
+ return -ENOMEM;
+ } else {
+ p = strstr(option,
"oci.zinfo=");
+ if (p != NULL) {
+ p +=
strlen("oci.zinfo=");
+
free(oci_cfg->zinfo_path);
+
oci_cfg->zinfo_path = strdup(p);
+ if
(!oci_cfg->zinfo_path)
+ return
-ENOMEM;
+ } else {
+ return -EINVAL;
+ }
+ }
}
}
}
@@ -332,11 +355,284 @@ static int erofsmount_fuse(const char *source, const
char *mountpoint,
return 0;
}
+struct erofsmount_tarindex_source {
+ struct erofs_vfile *tarindex_vf;
+ struct erofs_vfile *zinfo_vf;
+ u64 tarindex_size;
+};
+
struct erofsmount_nbd_ctx {
struct erofs_vfile vd; /* virtual device */
struct erofs_vfile sk; /* socket file */
};
+static ssize_t erofs_tarindex_pread(struct erofs_vfile *vf, void *buf,
+ size_t count, u64 offset)
+{
+ struct erofsmount_tarindex_source *hs;
+ ssize_t local_read, remote_read;
+ u64 local_part, remote_part, remote_offset;
index_part and tardata_part may be clearer.
+
+ hs = *(struct erofsmount_tarindex_source **)vf->payload;
+ if (!hs)
+ return -EINVAL;
+
+ /* Handle device boundary probe requests */
+ if (EROFS_IS_DEVICE_PROBE(offset))
+ return 0;
+
+ if (offset >= hs->tarindex_size) {
+ remote_offset = offset - hs->tarindex_size;
+
+ return hs->zinfo_vf->ops->pread(hs->zinfo_vf, buf, count,
remote_offset);
+ }
+
+ if (offset + count <= hs->tarindex_size)
+ return erofs_io_pread(hs->tarindex_vf, buf, count, offset);
+
+ local_part = hs->tarindex_size - offset;
+ remote_part = count - local_part;
+
+ local_read = erofs_io_pread(hs->tarindex_vf, buf, local_part, offset);
+ if (local_read < 0)
+ return local_read;
+
+ remote_read = hs->zinfo_vf->ops->pread(hs->zinfo_vf,
+ (char *)buf + local_read,
+ remote_part, 0);
+ if (remote_read < 0)
+ return remote_read;
+ return local_read + remote_read;
+}
+
+static void erofs_tarindex_close(struct erofs_vfile *vf)
+{
+ struct erofsmount_tarindex_source *hs;
+
+ if (!vf)
+ return;
+
+ hs = *(struct erofsmount_tarindex_source **)vf->payload;
+ if (!hs)
+ return;
+
+ if (hs->tarindex_size > 0) {
+ erofs_io_close(hs->tarindex_vf);
+ free(hs->tarindex_vf);
+ }
+
+ if (hs->zinfo_vf)
+ erofs_io_close(hs->zinfo_vf);
+
+ free(hs);
+}
+
+static int load_file_to_buf(const char *path, void **out, unsigned int
*out_len)
+{
+ FILE *fp = NULL;
+ void *buf = NULL;
+ int ret = 0;
+ long sz;
+ size_t num;
+
+ fp = fopen(path, "rb");
+ if (!fp)
+ return -errno;
+
+ if (fseek(fp, 0, SEEK_END) != 0) {
+ ret = -errno;
+ goto out;
+ }
+ sz = ftell(fp);
+ if (sz < 0) {
+ ret = -errno;
+ goto out;
+ }
+ if (fseek(fp, 0, SEEK_SET) != 0) {
+ ret = -errno;
+ goto out;
+ }
+ if (sz == 0) {
+ ret = -EINVAL;
+ goto out;
+ }
+
+ buf = malloc((size_t)sz);
+ if (!buf) {
+ ret = -ENOMEM;
+ goto out;
+ }
+
+ num = fread(buf, 1, (size_t)sz, fp);
+ if (num != (size_t)sz) {
+ ret = -EIO;
+ goto out;
+ }
+
+ *out = buf;
+ *out_len = (unsigned int)sz;
+ buf = NULL;
+
+out:
+ if (fp)
+ fclose(fp);
+ if (ret < 0 && buf)
+ free(buf);
+ return ret;
+}
+
+static int erofsmount_init_gzran(struct erofs_vfile **zinfo_vf,
+ const struct ocierofs_config *oci_cfg,
+ const char *zinfo_path)
+{
+ int err = 0;
+ void *zinfo_data = NULL;
+ unsigned int zinfo_len = 0;
+ struct erofs_vfile *oci_vf = NULL;
+
+ oci_vf = malloc(sizeof(*oci_vf));
+ if (!oci_vf) {
+ err = -ENOMEM;
+ goto cleanup;
+ }
+
+ err = ocierofs_io_open(oci_vf, oci_cfg);
+ if (err) {
+ free(oci_vf);
+ goto cleanup;
+ }
+
+ /* If no zinfo_path, return oci_vf directly for tar format */
+ if (!zinfo_path) {
+ *zinfo_vf = oci_vf;
+ return 0;
+ }
+
+ err = load_file_to_buf(zinfo_path, &zinfo_data, &zinfo_len);
+ if (err) {
+ erofs_io_close(oci_vf);
+ free(oci_vf);
+ return err;
+ }
+
+ *zinfo_vf = erofs_gzran_zinfo_open(oci_vf, zinfo_data, zinfo_len);
+ if (IS_ERR(*zinfo_vf)) {
+ err = PTR_ERR(*zinfo_vf);
+ *zinfo_vf = NULL;
+ erofs_io_close(oci_vf);
+ free(oci_vf);
+ goto cleanup;
+ }
+
+ free(zinfo_data);
+ return 0;
+
+cleanup:
+ if (zinfo_data)
+ free(zinfo_data);
+ return err;
+}
+
+static ssize_t erofs_tarindex_sendfile(struct erofs_vfile *vout, struct
erofs_vfile *vin,
+ off_t *pos, size_t count)
+{
+ static char buf[32768];
+ ssize_t total_written = 0, ret = 0, written;
+ size_t to_read;
+ u64 read_offset;
+
+ while (count > 0) {
+ to_read = min_t(size_t, count, sizeof(buf));
+ read_offset = pos ? *pos : 0;
+
+ ret = erofs_tarindex_pread(vin, buf, to_read, read_offset);
+ if (ret <= 0) {
+ if (ret < 0 && total_written == 0)
+ return ret;
+ break;
+ }
+
+ written = __erofs_io_write(vout->fd, buf, ret);
+ if (written < 0) {
+ ret = -errno;
+ break;
+ }
+ if (written != ret)
+ ret = written;
+
+ total_written += ret;
+ count -= ret;
+ if (pos)
+ *pos += ret;
+ }
+ return count;
+}
+
+static struct erofs_vfops tarindex_vfile_ops = {
+ .pread = erofs_tarindex_pread,
+ .sendfile = erofs_tarindex_sendfile,
+ .close = erofs_tarindex_close,
+};
+
+/*
+ * Create tarindex source for gzran+oci hybrid mode with three scenarios:
+ * 1. tarindex + zinfo: Remote data is tar.gzip format
+ * 2. tarindex only: Remote data is tar format
+ */
+static int erofs_create_tarindex_source(struct erofs_vfile *out_vf,
+ const struct ocierofs_config *oci_cfg,
+ const char *tarindex_path,
+ const char *zinfo_path)
+{
+ struct erofsmount_tarindex_source *hs;
+ int err;
+ struct stat st;
+
+ hs = calloc(1, sizeof(*hs));
+ if (!hs)
+ return -ENOMEM;
+
+ if (tarindex_path) {
+ hs->tarindex_vf = malloc(sizeof(*hs->tarindex_vf));
+ if (!hs->tarindex_vf) {
+ err = -ENOMEM;
+ goto cleanup;
+ }
+
+ hs->tarindex_vf->fd = open(tarindex_path, O_RDONLY);
Let's avoid `hs->tarindex_vf->fd`.
I suggest using a local variable vf directly.
Otherwise it looks good to me.
Thanks,
Gao Xiang
