[RFC] erofs-utils: on-disk extent format for blocks.

Gao Xiang gaoxiang25 at huawei.com
Mon Dec 30 17:49:11 AEDT 2019


On Mon, Dec 30, 2019 at 11:49:55AM +0530, Pratik Shinde wrote:
> Gao,
> 
> Okay, I will rename the 'size' variables accordingly.
> The 'S_SHIFT' gives the logical block number inside file. I am using it to
> assign logical block no.
> to extents.

S_SHIFT is used to transfer between i_mode and filetype, see:
https://elixir.bootlin.com/linux/latest/source/fs/ext4/ext4.h#L3196

IMO, which is not the above meaning, how about using EROFS_BLKSIZ
instead?

Thanks,
Gao Xiang

> Also, I will modify the code to record only data extents.
> 
> --Pratik.
> 
> On Sun, Dec 29, 2019 at 8:26 AM Gao Xiang <hsiangkao at aol.com> wrote:
> 
> > Hi Pratik,
> >
> > On Fri, Dec 27, 2019 at 09:13:48PM +0530, Pratik Shinde wrote:
> > > since this patch is quite different from previous patches I am treating
> > > this as new patch.
> > >
> > > 1) On disk extent format for erofs data blocks.
> > > 2) Detect holes inside files & skip allocation for hole blocks.
> > >
> > > Signed-off-by: Pratik Shinde <pratikshinde320 at gmail.com>
> > > ---
> > >  include/erofs/internal.h |  21 ++++++-
> > >  lib/inode.c              | 155
> > +++++++++++++++++++++++++++++++++++++++++------
> > >  2 files changed, 156 insertions(+), 20 deletions(-)
> > >
> > > diff --git a/include/erofs/internal.h b/include/erofs/internal.h
> > > index e13adda..128aa63 100644
> > > --- a/include/erofs/internal.h
> > > +++ b/include/erofs/internal.h
> > > @@ -63,7 +63,7 @@ struct erofs_sb_info {
> > >  extern struct erofs_sb_info sbi;
> > >
> > >  struct erofs_inode {
> > > -     struct list_head i_hash, i_subdirs, i_xattrs;
> > > +     struct list_head i_hash, i_subdirs, i_xattrs, i_extents;
> > >
> > >       unsigned int i_count;
> > >       struct erofs_inode *i_parent;
> > > @@ -93,6 +93,7 @@ struct erofs_inode {
> > >
> > >       unsigned int xattr_isize;
> > >       unsigned int extent_isize;
> > > +     unsigned int extent_meta_isize;
> >
> > maybe sparse_extent_isize is better...
> >
> > p.s. maybe we could send another patch rename extent_isize to
> > compress_extent_isize or some better name...
> >
> > >
> > >       erofs_nid_t nid;
> > >       struct erofs_buffer_head *bh;
> > > @@ -139,5 +140,23 @@ static inline const char *erofs_strerror(int err)
> > >       return msg;
> > >  }
> > >
> > > +#define HOLE_BLK     -1
> > > +/* on disk extent format */
> > > +struct erofs_extent {
> > > +     __le32 ee_lblk;
> > > +     __le32 ee_pblk;
> > > +     __le32 ee_len;
> > > +};
> > > +
> > > +struct erofs_extent_node {
> > > +     struct list_head next;
> > > +     erofs_blk_t lblk;
> > > +     erofs_blk_t pblk;
> > > +     u32 len;
> > > +};
> > > +
> > > +struct erofs_inline_extent_header {
> > > +     u32 count;
> > > +};
> > >  #endif
> > >
> > > diff --git a/lib/inode.c b/lib/inode.c
> > > index 0e19b11..a6af509 100644
> > > --- a/lib/inode.c
> > > +++ b/lib/inode.c
> > > @@ -38,6 +38,99 @@ static unsigned char erofs_type_by_mode[S_IFMT >>
> > S_SHIFT] = {
> > >
> > >  struct list_head inode_hashtable[NR_INODE_HASHTABLE];
> > >
> > > +
> > > +#define IS_HOLE(start, end) (roundup(start, EROFS_BLKSIZ) == start &&
> >       \
> > > +                          roundup(end, EROFS_BLKSIZ) == end &&       \
> > > +                         (end - start) % EROFS_BLKSIZ == 0)
> >
> > See below..
> >
> > > +
> > > +/* returns the number of holes present in the file */
> > > +unsigned int erofs_read_extents(struct erofs_inode *inode,
> > > +                             struct list_head *extents)
> > > +{
> > > +     int fd, st, en, dt;
> > > +     unsigned int nholes = 0;
> > > +     erofs_off_t data, hole, len, last_data;
> > > +     struct erofs_extent_node *e_hole, *e_data;
> > > +
> > > +     fd = open(inode->i_srcpath, O_RDONLY);
> > > +     if (fd < 0) {
> > > +             return -errno;
> > > +     }
> > > +     len = lseek(fd, 0, SEEK_END);
> > > +     if (lseek(fd, 0, SEEK_SET) == -1)
> > > +             return -errno;
> > > +     data = 0;
> > > +     last_data = 0;
> > > +     while (data < len) {
> > > +             hole = lseek(fd, data, SEEK_HOLE);
> > > +             if (hole == len)
> > > +                     break;
> > > +             data = lseek(fd, hole, SEEK_DATA);
> > > +             if (data < 0 || hole > data) {
> > > +                     return -EINVAL;
> > > +             }
> > > +             if (IS_HOLE(hole, data)) {
> > > +                     st = hole >> S_SHIFT;
> > > +                     en = data >> S_SHIFT;
> > > +                     dt = last_data >> S_SHIFT;
> >
> > Why using S_SHIFT here, some special meaning?
> >
> > > +                     last_data = data;
> > > +                     e_data = malloc(sizeof(struct erofs_extent_node));
> > > +                     if (e_data == NULL)
> > > +                             return -ENOMEM;
> > > +                     e_data->lblk = dt;
> > > +                     e_data->len = (st - dt);
> > > +                     list_add_tail(&e_data->next, extents);
> > > +                     e_hole = malloc(sizeof(struct erofs_extent_node));
> > > +                     if (e_hole == NULL)
> > > +                             return -ENOMEM;
> > > +                     e_hole->lblk = st;
> > > +                     e_hole->pblk = HOLE_BLK;
> > > +                     e_hole->len = (en - st);
> > > +                     list_add_tail(&e_hole->next, extents);
> > > +                     nholes += e_hole->len;
> >
> > Maybe we don't need to emit all HOLK extents since all data extents
> > are with _length_... It is easy to detect all holes between extents...
> >
> > If some block doesn't belong to any extent, it's a hole.
> >
> > Thanks,
> > Gao Xiang
> >
> > > +             }
> > > +     }
> > > +     /* rounddown to exclude tail-end data */
> > > +     if (last_data < len && (len - last_data) >= EROFS_BLKSIZ) {
> > > +             e_data = malloc(sizeof(struct erofs_extent_node));
> > > +             if (e_data == NULL)
> > > +                     return -ENOMEM;
> > > +             st = last_data >> S_SHIFT;
> > > +             e_data->lblk = st;
> > > +             e_data->len = rounddown((len - last_data), EROFS_BLKSIZ)
> > >> S_SHIFT;
> > > +             list_add_tail(&e_data->next, extents);
> > > +     }
> > > +     return nholes;
> > > +}
> > > +
> > > +char *erofs_create_extent_buffer(struct list_head *extents, unsigned
> > int size)
> > > +{
> > > +     struct erofs_extent_node *e_node;
> > > +     struct erofs_inline_extent_header *header;
> > > +     char *buf;
> > > +     unsigned int p = 0;
> > > +
> > > +     buf = malloc(size);
> > > +     if (buf == NULL)
> > > +             return ERR_PTR(-ENOMEM);
> > > +     header = (struct erofs_inline_extent_header *) buf;
> > > +     header->count = 0;
> > > +     p += sizeof(struct erofs_inline_extent_header);
> > > +     list_for_each_entry(e_node, extents, next) {
> > > +             const struct erofs_extent ee = {
> > > +                     .ee_lblk = cpu_to_le32(e_node->lblk),
> > > +                     .ee_pblk = cpu_to_le32(e_node->pblk),
> > > +                     .ee_len  = cpu_to_le32(e_node->len)
> > > +             };
> > > +             memcpy(buf + p, &ee, sizeof(struct erofs_extent));
> > > +             p += sizeof(struct erofs_extent);
> > > +             header->count++;
> > > +             list_del(&e_node->next);
> > > +             free(e_node);
> > > +     }
> > > +     return buf;
> > > +}
> > > +
> > >  void erofs_inode_manager_init(void)
> > >  {
> > >       unsigned int i;
> > > @@ -304,8 +397,9 @@ static bool erofs_file_is_compressible(struct
> > erofs_inode *inode)
> > >
> > >  int erofs_write_file(struct erofs_inode *inode)
> > >  {
> > > -     unsigned int nblocks, i;
> > > +     unsigned int nblocks, i, j, nholes;
> > >       int ret, fd;
> > > +     struct erofs_extent_node *e_node;
> > >
> > >       if (!inode->i_size) {
> > >               inode->datalayout = EROFS_INODE_FLAT_PLAIN;
> > > @@ -322,31 +416,43 @@ int erofs_write_file(struct erofs_inode *inode)
> > >       /* fallback to all data uncompressed */
> > >       inode->datalayout = EROFS_INODE_FLAT_INLINE;
> > >       nblocks = inode->i_size / EROFS_BLKSIZ;
> > > -
> > > -     ret = __allocate_inode_bh_data(inode, nblocks);
> > > +     nholes = erofs_read_extents(inode, &inode->i_extents);
> > > +     if (nholes < 0)
> > > +             return nholes;
> > > +     if (nblocks < 0)
> > > +             return nblocks;
> > > +     ret = __allocate_inode_bh_data(inode, nblocks - nholes);
> > >       if (ret)
> > >               return ret;
> > >
> > >       fd = open(inode->i_srcpath, O_RDONLY | O_BINARY);
> > >       if (fd < 0)
> > >               return -errno;
> > > -
> > > -     for (i = 0; i < nblocks; ++i) {
> > > -             char buf[EROFS_BLKSIZ];
> > > -
> > > -             ret = read(fd, buf, EROFS_BLKSIZ);
> > > -             if (ret != EROFS_BLKSIZ) {
> > > -                     if (ret < 0)
> > > -                             goto fail;
> > > -                     close(fd);
> > > -                     return -EAGAIN;
> > > +     i = inode->u.i_blkaddr;
> > > +     inode->extent_meta_isize = sizeof(struct
> > erofs_inline_extent_header);
> > > +     list_for_each_entry(e_node, &inode->i_extents, next) {
> > > +             inode->extent_meta_isize += sizeof(struct erofs_extent);
> > > +             if (e_node->pblk == HOLE_BLK) {
> > > +                     lseek(fd, e_node->len * EROFS_BLKSIZ, SEEK_CUR);
> > > +                     continue;
> > >               }
> > > +             e_node->pblk = i;
> > > +             i += e_node->len;
> > > +             for (j = 0; j < e_node->len; j++) {
> > > +                     char buf[EROFS_BLKSIZ];
> > > +                     ret = read(fd, buf, EROFS_BLKSIZ);
> > > +                     if (ret != EROFS_BLKSIZ) {
> > > +                             if (ret < 0)
> > > +                                     goto fail;
> > > +                             close(fd);
> > > +                             return -EAGAIN;
> > > +                     }
> > > +                     ret = blk_write(buf, e_node->pblk + j, 1);
> > > +                     if (ret)
> > > +                             goto fail;
> > >
> > > -             ret = blk_write(buf, inode->u.i_blkaddr + i, 1);
> > > -             if (ret)
> > > -                     goto fail;
> > > +             }
> > >       }
> > > -
> > >       /* read the tail-end data */
> > >       inode->idata_size = inode->i_size % EROFS_BLKSIZ;
> > >       if (inode->idata_size) {
> > > @@ -479,8 +585,19 @@ static bool erofs_bh_flush_write_inode(struct
> > erofs_buffer_head *bh)
> > >               if (ret)
> > >                       return false;
> > >               free(inode->compressmeta);
> > > +             off += inode->extent_isize;
> > >       }
> > >
> > > +     if (inode->extent_meta_isize) {
> > > +             char *extents =
> > erofs_create_extent_buffer(&inode->i_extents,
> > > +
> > inode->extent_meta_isize);
> > > +             if (IS_ERR(extents))
> > > +                     return false;
> > > +             ret = dev_write(extents, off, inode->extent_meta_isize);
> > > +             free(extents);
> > > +             if (ret)
> > > +                     return false;
> > > +     }
> > >       inode->bh = NULL;
> > >       erofs_iput(inode);
> > >       return erofs_bh_flush_generic_end(bh);
> > > @@ -737,10 +854,11 @@ struct erofs_inode *erofs_new_inode(void)
> > >
> > >       init_list_head(&inode->i_subdirs);
> > >       init_list_head(&inode->i_xattrs);
> > > +     init_list_head(&inode->i_extents);
> > >
> > >       inode->idata_size = 0;
> > >       inode->xattr_isize = 0;
> > > -     inode->extent_isize = 0;
> > > +     inode->extent_meta_isize = 0;
> > >
> > >       inode->bh = inode->bh_inline = inode->bh_data = NULL;
> > >       inode->idata = NULL;
> > > @@ -961,4 +1079,3 @@ struct erofs_inode
> > *erofs_mkfs_build_tree_from_path(struct erofs_inode *parent,
> > >
> > >       return erofs_mkfs_build_tree(inode);
> > >  }
> > > -
> > > --
> > > 2.9.3
> > >
> >


More information about the Linux-erofs mailing list