Hi Pratik,
On Wed, Aug 21, 2019 at 10:08:08PM +0530, Pratik Shinde wrote:
> Hello Maintainers,
>
> After going through the recent mail thread between linux's filesystem folks
> on erofs channel, I felt erofs needs an interactive debug utility (like xfs_db)
> which can be used to examine erofs images & can also be used to inject errors OR
> fuzzing for testing purpose & dumping different erofs meta data structures
> for debugging.
> In order to demonstrate above I wrote an experimental patch that simply dumps
> the superblock of an image after mkfs completes.the full fletch utility will run
> independently and be able to seek / print / modify any byte of an erofs image,
> dump structures/lists/directory content of an image.
Yes, I think we really need that interactive tools, actually I'm stuggling in
modifing Guifu's erofs-fuse now, we need to add the parsing ability to "lib/"
first.
I mean, first, I will add a "fuse" field to "cfg". If it is false, it will
generate a image, or it will parse a image...
And then we need to add parsing logic into "lib/" as well, and use
"if (cfg.fuse)" to differnate whether it should read or write data.
That is my prelimitary thought.. I will work on this framework in this weekend.
and then we can work together on it. :)
p.s. Pratik, if you have some time, could you take some extra time adding the
super checksum calulation to EROFS? I mean we can add EROFS_FEATURE_SB_CHKSUM
to the compat superblock field ("features"), and do crc32_le on kernel and mkfs...
If you dont have time, I will do it later instead... (since we are using EROFS
on the top of dm-verity, but completing the superblock chksum is also a good idea.)
And then we can add block-based verification layer to EROFS, it can be seen
as a hash tree like dm-verity or just simply CRC32 arrays for user to choise.
Thanks,
Gao Xiang
>
> NOTE:This is an experimental patch just to demonstrate the purpose. The patch
> lacks a lot of things like coding standard, and new code runs in the context
> of mkfs itself.kindly ignore it.
>
> kindly provide your feedback on this.
>
> Signed-off-by: Pratik Shinde <pratikshinde320@gmail.com>
> ---
> include/erofs/io.h | 8 ++++++++
> lib/io.c | 27 +++++++++++++++++++++++++++
> mkfs/main.c | 36 ++++++++++++++++++++++++++++++++++++
> 3 files changed, 71 insertions(+)
>
> diff --git a/include/erofs/io.h b/include/erofs/io.h
> index 4b574bd..e91d6ee 100644
> --- a/include/erofs/io.h
> +++ b/include/erofs/io.h
> @@ -18,6 +18,7 @@
>
> int dev_open(const char *devname);
> void dev_close(void);
> +int dev_read(void *buf, u64 offset, size_t len);
> int dev_write(const void *buf, u64 offset, size_t len);
> int dev_fillzero(u64 offset, size_t len, bool padding);
> int dev_fsync(void);
> @@ -30,5 +31,12 @@ static inline int blk_write(const void *buf, erofs_blk_t blkaddr,
> blknr_to_addr(nblocks));
> }
>
> +static inline int blk_read(void *buf, erofs_blk_t blkaddr,
> + u32 nblocks)
> +{
> + return dev_read(buf, blknr_to_addr(blkaddr),
> + blknr_to_addr(nblocks));
> +}
> +
> #endif
>
> diff --git a/lib/io.c b/lib/io.c
> index 15c5a35..87d7d6c 100644
> --- a/lib/io.c
> +++ b/lib/io.c
> @@ -109,6 +109,33 @@ u64 dev_length(void)
> return erofs_devsz;
> }
>
> +int dev_read(void *buf, u64 offset, size_t len)
> +{
> + int ret;
> +
> + if (cfg.c_dry_run)
> + return 0;
> +
> + if (!buf) {
> + erofs_err("buf is NULL");
> + return -EINVAL;
> + }
> + if (offset >= erofs_devsz || len > erofs_devsz ||
> + offset > erofs_devsz - len) {
> + erofs_err("read posion[%" PRIu64 ", %zd] is too large beyond the end of device(%" PRIu64 ").",
> + offset, len, erofs_devsz);
> + return -EINVAL;
> + }
> +
> + ret = pread64(erofs_devfd, buf, len, (off64_t)offset);
> + if (ret != (int)len) {
> + erofs_err("Failed to read data from device - %s:[%" PRIu64 ", %zd].",
> + erofs_devname, offset, len);
> + return -errno;
> + }
> + return 0;
> +}
> +
> int dev_write(const void *buf, u64 offset, size_t len)
> {
> int ret;
> diff --git a/mkfs/main.c b/mkfs/main.c
> index f127fe1..109486e 100644
> --- a/mkfs/main.c
> +++ b/mkfs/main.c
> @@ -182,6 +182,41 @@ int erofs_mkfs_update_super_block(struct erofs_buffer_head *bh,
> return 0;
> }
>
> +void erofs_dump_super(char *img_path)
> +{
> + struct erofs_super_block *sb;
> + char buf[EROFS_BLKSIZ];
> + unsigned int blksz;
> + int ret = 0;
> +
> + if (img_path == NULL) {
> + erofs_err("image path cannot be null");
> + return;
> + }
> + ret = blk_read(buf, 0, 1);
> + if (ret) {
> + erofs_err("error reading super-block structure");
> + return;
> + }
> +
> + sb = (struct erofs_super_block *)((u8 *)buf + EROFS_SUPER_OFFSET);
> + if (le32_to_cpu(sb->magic) != EROFS_SUPER_MAGIC_V1) {
> + erofs_err("not a erofs image");
> + return;
> + }
> +
> + erofs_dump("magic: 0x%x\n", le32_to_cpu(sb->magic));
> + blksz = 1 << sb->blkszbits;
> + erofs_dump("block size: %d\n", blksz);
> + erofs_dump("root inode: %d\n", le32_to_cpu(sb->root_nid));
> + erofs_dump("inodes: %llu\n", le64_to_cpu(sb->inos));
> + erofs_dump("build time: %u\n", le32_to_cpu(sb->build_time));
> + erofs_dump("blocks: %u\n", le32_to_cpu(sb->blocks));
> + erofs_dump("meta block: %u\n", le32_to_cpu(sb->meta_blkaddr));
> + erofs_dump("xattr block: %u\n", le32_to_cpu(sb->xattr_blkaddr));
> + erofs_dump("requirements: 0x%x\n", le32_to_cpu(sb->requirements));
> +}
> +
> int main(int argc, char **argv)
> {
> int err = 0;
> @@ -268,6 +303,7 @@ int main(int argc, char **argv)
> err = -EIO;
> exit:
> z_erofs_compress_exit();
> + erofs_dump_super("dummy");
> dev_close();
> erofs_exit_configure();
>
> --
> 2.9.3
>