erofs: support flattened block device for multi-blob images
authorJia Zhu <zhujia.zj@bytedance.com>
Thu, 2 Mar 2023 07:17:51 +0000 (15:17 +0800)
committerGao Xiang <hsiangkao@linux.alibaba.com>
Sun, 16 Apr 2023 17:15:46 +0000 (01:15 +0800)
In order to support mounting multi-blobs container image as a single
block device, add flattened block device feature for EROFS.

In this mode, all meta/data contents will be mapped into one block
space. User could compose a block device(by nbd/ublk/virtio-blk/
vhost-user-blk) from multiple sources and mount the block device by
EROFS directly. It can reduce the number of block devices used, and
it's also benefits in both VM file passthrough and distributed storage
scenarios.

You can test this using the method mentioned by:
https://github.com/dragonflyoss/image-service/pull/1139
1. Compose a (nbd)block device from multi-blobs.
2. Mount EROFS on mntdir/.
3. Compare the md5sum between source dir and mntdir/.

Later, we could also use it to refer original tar blobs.

Signed-off-by: Jia Zhu <zhujia.zj@bytedance.com>
Signed-off-by: Xin Yin <yinxin.x@bytedance.com>
Reviewed-by: Jingbo Xu <jefflexu@linux.alibaba.com>
Acked-by: Chao Yu <chao@kernel.org>
Tested-by: Jiang Liu <gerry@linux.alibaba.com>
Link: https://lore.kernel.org/r/20230302071751.48425-1-zhujia.zj@bytedance.com
[ Gao Xiang: refine commit message and use erofs_pos(). ]
Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com>
fs/erofs/data.c
fs/erofs/internal.h
fs/erofs/super.c

index 1c931e3..03c6ffd 100644 (file)
@@ -200,7 +200,6 @@ int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *map)
        struct erofs_device_info *dif;
        int id;
 
-       /* primary device by default */
        map->m_bdev = sb->s_bdev;
        map->m_daxdev = EROFS_SB(sb)->dax_dev;
        map->m_dax_part_off = EROFS_SB(sb)->dax_part_off;
@@ -213,12 +212,17 @@ int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *map)
                        up_read(&devs->rwsem);
                        return -ENODEV;
                }
+               if (devs->flatdev) {
+                       map->m_pa += erofs_pos(sb, dif->mapped_blkaddr);
+                       up_read(&devs->rwsem);
+                       return 0;
+               }
                map->m_bdev = dif->bdev;
                map->m_daxdev = dif->dax_dev;
                map->m_dax_part_off = dif->dax_part_off;
                map->m_fscache = dif->fscache;
                up_read(&devs->rwsem);
-       } else if (devs->extra_devices) {
+       } else if (devs->extra_devices && !devs->flatdev) {
                down_read(&devs->rwsem);
                idr_for_each_entry(&devs->tree, dif, id) {
                        erofs_off_t startoff, length;
index 9eff0c0..e30a4fd 100644 (file)
@@ -81,6 +81,7 @@ struct erofs_dev_context {
        struct rw_semaphore rwsem;
 
        unsigned int extra_devices;
+       bool flatdev;
 };
 
 struct erofs_fs_context {
index dbffcdd..9e56a6f 100644 (file)
@@ -252,7 +252,7 @@ static int erofs_init_device(struct erofs_buf *buf, struct super_block *sb,
                if (IS_ERR(fscache))
                        return PTR_ERR(fscache);
                dif->fscache = fscache;
-       } else {
+       } else if (!sbi->devs->flatdev) {
                bdev = blkdev_get_by_path(dif->path, FMODE_READ | FMODE_EXCL,
                                          sb->s_type);
                if (IS_ERR(bdev))
@@ -294,6 +294,9 @@ static int erofs_scan_devices(struct super_block *sb,
        if (!ondisk_extradevs)
                return 0;
 
+       if (!sbi->devs->extra_devices && !erofs_is_fscache_mode(sb))
+               sbi->devs->flatdev = true;
+
        sbi->device_id_mask = roundup_pow_of_two(ondisk_extradevs + 1) - 1;
        pos = le16_to_cpu(dsb->devt_slotoff) * EROFS_DEVT_SLOT_SIZE;
        down_read(&sbi->devs->rwsem);