Hi Gao, 

Understood your concern.

Can we do something like :

1) Allocate one buf of size EROFS_BLKSIZ
2) Read one page at a time into buf(memcpy) .call crc32c for it.

In this way we won't be writing directly into page data and will not do large allocation.
What do you think?

--Pratik

On Tue, 22 Oct, 2019, 10:04 PM Gao Xiang, <hsiangkao@aol.com> wrote:
Hi Pratik,

Some comments as below...

On Tue, Oct 22, 2019 at 09:09:56PM +0530, Pratik Shinde wrote:
> Patch for kernel side changes of checksum feature.Used kernel's
> crc32c library for calculating the checksum.
>
> Signed-off-by: Pratik Shinde <pratikshinde320@gmail.com>
> ---
>  fs/erofs/erofs_fs.h |  5 +++--
>  fs/erofs/internal.h |  3 ++-
>  fs/erofs/super.c    | 33 +++++++++++++++++++++++++++++++++
>  3 files changed, 38 insertions(+), 3 deletions(-)
>
> diff --git a/fs/erofs/erofs_fs.h b/fs/erofs/erofs_fs.h
> index b1ee565..4d8097a 100644
> --- a/fs/erofs/erofs_fs.h
> +++ b/fs/erofs/erofs_fs.h
> @@ -17,6 +17,7 @@
>   */
>  #define EROFS_FEATURE_INCOMPAT_LZ4_0PADDING  0x00000001
>  #define EROFS_ALL_FEATURE_INCOMPAT           EROFS_FEATURE_INCOMPAT_LZ4_0PADDING
> +#define EROFS_FEATURE_COMPAT_SB_CHKSUM               0x00000001

>  /* 128-byte erofs on-disk super block */
>  struct erofs_super_block {
> @@ -37,8 +38,8 @@ struct erofs_super_block {
>       __u8 uuid[16];          /* 128-bit uuid for volume */
>       __u8 volume_name[16];   /* volume name */
>       __le32 feature_incompat;
> -
> -     __u8 reserved2[44];
> +     __le32 chksum_blocks;   /* number of blocks used for checksum */
> +     __u8 reserved2[40];
>  };

>  /*
> diff --git a/fs/erofs/internal.h b/fs/erofs/internal.h
> index 544a453..cec27ca 100644
> --- a/fs/erofs/internal.h
> +++ b/fs/erofs/internal.h
> @@ -86,7 +86,7 @@ struct erofs_sb_info {
>       u8 uuid[16];                    /* 128-bit uuid for volume */
>       u8 volume_name[16];             /* volume name */
>       u32 feature_incompat;
> -
> +     u32 feature_compat;
>       unsigned int mount_opt;
>  };

> @@ -426,6 +426,7 @@ static inline void z_erofs_exit_zip_subsystem(void) {}
>  #endif       /* !CONFIG_EROFS_FS_ZIP */

>  #define EFSCORRUPTED    EUCLEAN         /* Filesystem is corrupted */
> +#define EFSBADCRC    EBADMSG         /* Bad crc found */

>  #endif       /* __EROFS_INTERNAL_H */

> diff --git a/fs/erofs/super.c b/fs/erofs/super.c
> index 0e36949..9cda72d 100644
> --- a/fs/erofs/super.c
> +++ b/fs/erofs/super.c
> @@ -9,6 +9,7 @@
>  #include <linux/statfs.h>
>  #include <linux/parser.h>
>  #include <linux/seq_file.h>
> +#include <linux/crc32c.h>
>  #include "xattr.h"

>  #define CREATE_TRACE_POINTS
> @@ -46,6 +47,31 @@ void _erofs_info(struct super_block *sb, const char *function,
>       va_end(args);
>  }

> +static int erofs_validate_sb_chksum(struct erofs_super_block *dsb,
> +                                    struct super_block *sb)
> +{
> +     u32 disk_chksum, nblocks, crc = 0;
> +     void *kaddr;
> +     struct page *page;
> +     int i;
> +
> +     disk_chksum = le32_to_cpu(dsb->checksum);
> +     nblocks = le32_to_cpu(dsb->chksum_blocks);

We cannot write the page data directly since the page cache should be kept in
sync with ondisk data (or for read-write fs, if it's claimed as uptodated, and
it is modified later,  you should mark it dirty, and do writeback then, but
that is not the erofs case.)

> +     dsb->checksum = 0;
> +     for (i = 0; i < nblocks; i++) {
> +             page = erofs_get_meta_page(sb, i);
> +             if (IS_ERR(page))
> +                     return PTR_ERR(page);
> +             kaddr = kmap(page);

Here kmap_atomic(page) is better. what I mean is kmap_atomic() in the caller
erofs_read_superblock(), it should be replaced to kmap() instead.

> +             crc = crc32c(crc, kaddr, EROFS_BLKSIZ);
> +             kunmap(page);
> +             unlock_page(page);

need
                put_page(page);


I'm not sure whether I explained quite well, but this patch needs something
to do. I'm now working on demonstrating new XZ algorithm and releasing
erofs-utils v1.0.

You can give more tries or I will help later. :-)

Thanks,
Gao Xiang


> +     }
> +     if (crc != disk_chksum)
> +             return -EFSBADCRC;
> +     return 0;
> +}
> +
>  static void erofs_inode_init_once(void *ptr)
>  {
>       struct erofs_inode *vi = ptr;
> @@ -121,6 +147,13 @@ static int erofs_read_superblock(struct super_block *sb)
>               goto out;
>       }

> +     if (dsb->feature_compat & EROFS_FEATURE_COMPAT_SB_CHKSUM) {
> +             ret = erofs_validate_sb_chksum(dsb, sb);
> +             if (ret < 0) {
> +                     erofs_err(sb, "super block checksum incorrect");
> +                     goto out;
> +             }
> +     }
>       blkszbits = dsb->blkszbits;
>       /* 9(512 bytes) + LOG_SECTORS_PER_BLOCK == LOG_BLOCK_SIZE */
>       if (blkszbits != LOG_BLOCK_SIZE) {
> --
> 2.9.3
>