Re: [PATCHv2] support ZSTD compression algorithm

[Mark Wielaard <mark@klomp.org>]

Hi Martin,

On Tue, Nov 29, 2022 at 01:05:45PM +0100, Martin Liška wrote:
> diff --git a/libelf/elf_compress.c b/libelf/elf_compress.c
> index d7f53af2..7a6e37a4 100644
> --- a/libelf/elf_compress.c
> +++ b/libelf/elf_compress.c
> @@ -39,6 +39,10 @@
>  #include <string.h>
>  #include <zlib.h>
> +#ifdef USE_ZSTD
> +#include <zstd.h>
> +#endif
> +
>  /* Cleanup and return result.  Don't leak memory.  */
>  static void *
>  do_deflate_cleanup (void *result, z_stream *z, void *out_buf,

OK.

> @@ -54,53 +58,14 @@ do_deflate_cleanup (void *result, z_stream *z, void *out_buf,
>  #define deflate_cleanup(result, cdata) \
>      do_deflate_cleanup(result, &z, out_buf, cdata)
> -/* Given a section, uses the (in-memory) Elf_Data to extract the
> -   original data size (including the given header size) and data
> -   alignment.  Returns a buffer that has at least hsize bytes (for the
> -   caller to fill in with a header) plus zlib compressed date.  Also
> -   returns the new buffer size in new_size (hsize + compressed data
> -   size).  Returns (void *) -1 when FORCE is false and the compressed
> -   data would be bigger than the original data.  */
>  void *
>  internal_function
> -__libelf_compress (Elf_Scn *scn, size_t hsize, int ei_data,
> -		   size_t *orig_size, size_t *orig_addralign,
> -		   size_t *new_size, bool force)
> +__libelf_compress_zlib (Elf_Scn *scn, size_t hsize, int ei_data,
> +			size_t *orig_size, size_t *orig_addralign,
> +			size_t *new_size, bool force,
> +			Elf_Data *data, Elf_Data *next_data,
> +			void *out_buf, size_t out_size, size_t block)
>  {
> -  /* The compressed data is the on-disk data.  We simplify the
> -     implementation a bit by asking for the (converted) in-memory
> -     data (which might be all there is if the user created it with
> -     elf_newdata) and then convert back to raw if needed before
> -     compressing.  Should be made a bit more clever to directly
> -     use raw if that is directly available.  */
> -  Elf_Data *data = elf_getdata (scn, NULL);
> -  if (data == NULL)
> -    return NULL;
> -
> -  /* When not forced and we immediately know we would use more data by
> -     compressing, because of the header plus zlib overhead (five bytes
> -     per 16 KB block, plus a one-time overhead of six bytes for the
> -     entire stream), don't do anything.  */
> -  Elf_Data *next_data = elf_getdata (scn, data);
> -  if (next_data == NULL && !force
> -      && data->d_size <= hsize + 5 + 6)
> -    return (void *) -1;
> -
> -  *orig_addralign = data->d_align;
> -  *orig_size = data->d_size;
> -
> -  /* Guess an output block size. 1/8th of the original Elf_Data plus
> -     hsize.  Make the first chunk twice that size (25%), then increase
> -     by a block (12.5%) when necessary.  */
> -  size_t block = (data->d_size / 8) + hsize;
> -  size_t out_size = 2 * block;
> -  void *out_buf = malloc (out_size);
> -  if (out_buf == NULL)
> -    {
> -      __libelf_seterrno (ELF_E_NOMEM);
> -      return NULL;
> -    }
> -
>    /* Caller gets to fill in the header at the start.  Just skip it here.  */
>    size_t used = hsize;

OK, all this is moved lower and then calls either
__libelf_compress_zlib or __libelf_compress_zstd.

> @@ -205,9 +170,186 @@ __libelf_compress (Elf_Scn *scn, size_t hsize, int ei_data,
>    return out_buf;
>  }
> +#ifdef USE_ZSTD
> +/* Cleanup and return result.  Don't leak memory.  */
> +static void *
> +do_zstd_cleanup (void *result, ZSTD_CCtx * const cctx, void *out_buf,
> +		 Elf_Data *cdatap)
> +{
> +  ZSTD_freeCCtx (cctx);
> +  free (out_buf);
> +  if (cdatap != NULL)
> +    free (cdatap->d_buf);
> +  return result;
> +}
>
> +#define zstd_cleanup(result, cdata) \
> +    do_zstd_cleanup(result, cctx, out_buf, cdata)
> +

OK, mimics do_deflate_cleanup.

>  void *
>  internal_function
> -__libelf_decompress (void *buf_in, size_t size_in, size_t size_out)
> +__libelf_compress_zstd (Elf_Scn *scn, size_t hsize, int ei_data,
> +			size_t *orig_size, size_t *orig_addralign,
> +			size_t *new_size, bool force,
> +			Elf_Data *data, Elf_Data *next_data,
> +			void *out_buf, size_t out_size, size_t block)
> +{
> +  /* Caller gets to fill in the header at the start.  Just skip it here.  */
> +  size_t used = hsize;
> +
> +  ZSTD_CCtx* const cctx = ZSTD_createCCtx();
> +  Elf_Data cdata;
> +  cdata.d_buf = NULL;
> +
> +  /* Loop over data buffers.  */
> +  ZSTD_EndDirective mode = ZSTD_e_continue;
> +
> +  do
> +    {
> +      /* Convert to raw if different endianness.  */
> +      cdata = *data;
> +      bool convert = ei_data != MY_ELFDATA && data->d_size > 0;
> +      if (convert)
> +	{
> +	  /* Don't do this conversion in place, we might want to keep
> +	     the original data around, caller decides.  */
> +	  cdata.d_buf = malloc (data->d_size);
> +	  if (cdata.d_buf == NULL)
> +	    {
> +	      __libelf_seterrno (ELF_E_NOMEM);
> +	      return zstd_cleanup (NULL, NULL);
> +	    }
> +	  if (gelf_xlatetof (scn->elf, &cdata, data, ei_data) == NULL)
> +	    return zstd_cleanup (NULL, &cdata);
> +	}
> +
> +      ZSTD_inBuffer ib = { cdata.d_buf, cdata.d_size, 0 };
> +
> +      /* Get next buffer to see if this is the last one.  */
> +      data = next_data;
> +      if (data != NULL)
> +	{
> +	  *orig_addralign = MAX (*orig_addralign, data->d_align);
> +	  *orig_size += data->d_size;
> +	  next_data = elf_getdata (scn, data);
> +	}
> +      else
> +	mode = ZSTD_e_end;
> +
> +      /* Flush one data buffer.  */
> +      for (;;)
> +	{
> +	  ZSTD_outBuffer ob = { out_buf + used, out_size - used, 0 };
> +	  size_t ret = ZSTD_compressStream2 (cctx, &ob, &ib, mode);
> +	  if (ZSTD_isError (ret))
> +	    {
> +	      __libelf_seterrno (ELF_E_COMPRESS_ERROR);
> +	      return zstd_cleanup (NULL, convert ? &cdata : NULL);
> +	    }
> +	  used += ob.pos;
> +
> +	  /* Bail out if we are sure the user doesn't want the
> +	     compression forced and we are using more compressed data
> +	     than original data.  */
> +	  if (!force && mode == ZSTD_e_end && used >= *orig_size)
> +	    return zstd_cleanup ((void *) -1, convert ? &cdata : NULL);
> +
> +	  if (ret > 0)
> +	    {
> +	      void *bigger = realloc (out_buf, out_size + block);
> +	      if (bigger == NULL)
> +		{
> +		  __libelf_seterrno (ELF_E_NOMEM);
> +		  return zstd_cleanup (NULL, convert ? &cdata : NULL);
> +		}

OK, zstd_cleanup does also free out_buf.

> +	      out_buf = bigger;
> +	      out_size += block;
> +	    }
> +	  else
> +	    break;
> +	}
> +
> +      if (convert)
> +	{
> +	  free (cdata.d_buf);
> +	  cdata.d_buf = NULL;
> +	}
> +    }
> +  while (mode != ZSTD_e_end); /* More data blocks.  */
> +
> +  ZSTD_freeCCtx (cctx);
> +  *new_size = used;
> +  return out_buf;
> +}
> +#endif

Look good.

> +/* Given a section, uses the (in-memory) Elf_Data to extract the
> +   original data size (including the given header size) and data
> +   alignment.  Returns a buffer that has at least hsize bytes (for the
> +   caller to fill in with a header) plus zlib compressed date.  Also
> +   returns the new buffer size in new_size (hsize + compressed data
> +   size).  Returns (void *) -1 when FORCE is false and the compressed
> +   data would be bigger than the original data.  */
> +void *
> +internal_function
> +__libelf_compress (Elf_Scn *scn, size_t hsize, int ei_data,
> +		   size_t *orig_size, size_t *orig_addralign,
> +		   size_t *new_size, bool force, bool use_zstd)
> +{
> +  /* The compressed data is the on-disk data.  We simplify the
> +     implementation a bit by asking for the (converted) in-memory
> +     data (which might be all there is if the user created it with
> +     elf_newdata) and then convert back to raw if needed before
> +     compressing.  Should be made a bit more clever to directly
> +     use raw if that is directly available.  */
> +  Elf_Data *data = elf_getdata (scn, NULL);
> +  if (data == NULL)
> +    return NULL;
> +
> +  /* When not forced and we immediately know we would use more data by
> +     compressing, because of the header plus zlib overhead (five bytes
> +     per 16 KB block, plus a one-time overhead of six bytes for the
> +     entire stream), don't do anything.
> +     Size estimation for ZSTD compression would be similar.  */
> +  Elf_Data *next_data = elf_getdata (scn, data);
> +  if (next_data == NULL && !force
> +      && data->d_size <= hsize + 5 + 6)
> +    return (void *) -1;
> +
> +  *orig_addralign = data->d_align;
> +  *orig_size = data->d_size;
> +
> +  /* Guess an output block size. 1/8th of the original Elf_Data plus
> +     hsize.  Make the first chunk twice that size (25%), then increase
> +     by a block (12.5%) when necessary.  */
> +  size_t block = (data->d_size / 8) + hsize;
> +  size_t out_size = 2 * block;
> +  void *out_buf = malloc (out_size);
> +  if (out_buf == NULL)
> +    {
> +      __libelf_seterrno (ELF_E_NOMEM);
> +      return NULL;
> +    }

OK, this is all just moved from earlier with an extra use_zstd flag.

> +  if (use_zstd)
> +#ifdef USE_ZSTD
> +    return __libelf_compress_zstd (scn, hsize, ei_data, orig_size,
> +				   orig_addralign, new_size, force,
> +				   data, next_data, out_buf, out_size,
> +				   block);
> +#else
> +    return NULL;

Should this also call __libelf_seterrno (ELF_E_UNKNOWN_COMPRESSION_TYPE) ?

> +#endif
> +  else
> +    return __libelf_compress_zlib (scn, hsize, ei_data, orig_size,
> +				   orig_addralign, new_size, force,
> +				   data, next_data, out_buf, out_size,
> +				   block);
> +}
> +
> +void *
> +internal_function
> +__libelf_decompress_zlib (void *buf_in, size_t size_in, size_t size_out)
>  {
>    /* Catch highly unlikely compression ratios so we don't allocate
>       some giant amount of memory for nothing. The max compression
> @@ -218,7 +360,7 @@ __libelf_decompress (void *buf_in, size_t size_in, size_t size_out)
>        return NULL;
>      }
> -  /* Malloc might return NULL when requestion zero size.  This is highly
> +  /* Malloc might return NULL when requesting zero size.  This is highly
>       unlikely, it would only happen when the compression was forced.
>       But we do need a non-NULL buffer to return and set as result.
>       Just make sure to always allocate at least 1 byte.  */

OK, typo fix.

> @@ -260,6 +402,51 @@ __libelf_decompress (void *buf_in, size_t size_in, size_t size_out)
>    return buf_out;
>  }
> +#ifdef USE_ZSTD
> +void *
> +internal_function
> +__libelf_decompress_zstd (void *buf_in, size_t size_in, size_t size_out)
> +{
> +  /* Malloc might return NULL when requesting zero size.  This is highly
> +     unlikely, it would only happen when the compression was forced.
> +     But we do need a non-NULL buffer to return and set as result.
> +     Just make sure to always allocate at least 1 byte.  */
> +  void *buf_out = malloc (size_out ?: 1);
> +  if (unlikely (buf_out == NULL))
> +    {
> +      __libelf_seterrno (ELF_E_NOMEM);
> +      return NULL;
> +    }
> +
> +  size_t ret = ZSTD_decompress (buf_out, size_out, buf_in, size_in);
> +  if (ZSTD_isError (ret))
> +    {
> +      free (buf_out);
> +      __libelf_seterrno (ELF_E_DECOMPRESS_ERROR);
> +      return NULL;
> +    }
> +  else
> +    return buf_out;
> +}
> +#endif

OK.

> +void *
> +internal_function
> +__libelf_decompress (int chtype, void *buf_in, size_t size_in, size_t size_out)
> +{
> +  if (chtype == ELFCOMPRESS_ZLIB)
> +    return __libelf_decompress_zlib (buf_in, size_in, size_out);
> +  else
> +    {
> +#ifdef USE_ZSTD
> +    return __libelf_decompress_zstd (buf_in, size_in, size_out);
> +#else
> +    __libelf_seterrno (ELF_E_DECOMPRESS_ERROR);

Maybe ELF_E_UNKNOWN_COMPRESSION_TYPE ?

> +    return NULL;
> +#endif
> +    }
> +}
> +
>  void *
>  internal_function
>  __libelf_decompress_elf (Elf_Scn *scn, size_t *size_out, size_t *addralign)
> @@ -268,7 +455,7 @@ __libelf_decompress_elf (Elf_Scn *scn, size_t *size_out, size_t *addralign)
>    if (gelf_getchdr (scn, &chdr) == NULL)
>      return NULL;
> -  if (chdr.ch_type != ELFCOMPRESS_ZLIB)
> +  if (chdr.ch_type != ELFCOMPRESS_ZLIB && chdr.ch_type != ELFCOMPRESS_ZSTD)
>      {
>        __libelf_seterrno (ELF_E_UNKNOWN_COMPRESSION_TYPE);
>        return NULL;

Should the chdr.ch_type != ELFCOMPRESS_ZSTD be guarded by #ifdef USE_ZSTD ?

> @@ -295,7 +482,9 @@ __libelf_decompress_elf (Elf_Scn *scn, size_t *size_out, size_t *addralign)
>  		  ? sizeof (Elf32_Chdr) : sizeof (Elf64_Chdr));
>    size_t size_in = data->d_size - hsize;
>    void *buf_in = data->d_buf + hsize;
> -  void *buf_out = __libelf_decompress (buf_in, size_in, chdr.ch_size);
> +  void *buf_out
> +    = __libelf_decompress (chdr.ch_type, buf_in, size_in, chdr.ch_size);
> +
>    *size_out = chdr.ch_size;
>    *addralign = chdr.ch_addralign;
>    return buf_out;
> @@ -394,7 +583,7 @@ elf_compress (Elf_Scn *scn, int type, unsigned int flags)
>      }
>    int compressed = (sh_flags & SHF_COMPRESSED);
> -  if (type == ELFCOMPRESS_ZLIB)
> +  if (type == ELFCOMPRESS_ZLIB || type == ELFCOMPRESS_ZSTD)
>      {
>        /* Compress/Deflate.  */
>        if (compressed == 1)
> @@ -408,7 +597,8 @@ elf_compress (Elf_Scn *scn, int type, unsigned int flags)
>        size_t orig_size, orig_addralign, new_size;
>        void *out_buf = __libelf_compress (scn, hsize, elfdata,
>  					 &orig_size, &orig_addralign,
> -					 &new_size, force);
> +					 &new_size, force,
> +					 type == ELFCOMPRESS_ZSTD);
>        /* Compression would make section larger, don't change anything.  */
>        if (out_buf == (void *) -1)
> @@ -422,7 +612,7 @@ elf_compress (Elf_Scn *scn, int type, unsigned int flags)
>        if (elfclass == ELFCLASS32)
>  	{
>  	  Elf32_Chdr chdr;
> -	  chdr.ch_type = ELFCOMPRESS_ZLIB;
> +	  chdr.ch_type = type;
>  	  chdr.ch_size = orig_size;
>  	  chdr.ch_addralign = orig_addralign;
>  	  if (elfdata != MY_ELFDATA)
> @@ -436,7 +626,7 @@ elf_compress (Elf_Scn *scn, int type, unsigned int flags)
>        else
>  	{
>  	  Elf64_Chdr chdr;
> -	  chdr.ch_type = ELFCOMPRESS_ZLIB;
> +	  chdr.ch_type = type;
>  	  chdr.ch_reserved = 0;
>  	  chdr.ch_size = orig_size;
>  	  chdr.ch_addralign = sh_addralign;

OK.

> diff --git a/libelf/elf_compress_gnu.c b/libelf/elf_compress_gnu.c
> index 3d2977e7..8e20b30e 100644
> --- a/libelf/elf_compress_gnu.c
> +++ b/libelf/elf_compress_gnu.c
> @@ -103,7 +103,8 @@ elf_compress_gnu (Elf_Scn *scn, int inflate, unsigned int flags)
>        size_t orig_size, new_size, orig_addralign;
>        void *out_buf = __libelf_compress (scn, hsize, elfdata,
>  					 &orig_size, &orig_addralign,
> -					 &new_size, force);
> +					 &new_size, force,
> +					 /* use_zstd */ false);
>        /* Compression would make section larger, don't change anything.  */
>        if (out_buf == (void *) -1)

OK.

> @@ -178,7 +179,7 @@ elf_compress_gnu (Elf_Scn *scn, int inflate, unsigned int flags)
>        size_t size = gsize;
>        size_t size_in = data->d_size - hsize;
>        void *buf_in = data->d_buf + hsize;
> -      void *buf_out = __libelf_decompress (buf_in, size_in, size);
> +      void *buf_out = __libelf_decompress (ELFCOMPRESS_ZLIB, buf_in, size_in, size);
>        if (buf_out == NULL)
>  	return -1;

OK.

> diff --git a/libelf/libelfP.h b/libelf/libelfP.h
> index d88a613c..6624f38a 100644
> --- a/libelf/libelfP.h
> +++ b/libelf/libelfP.h
> @@ -574,10 +574,10 @@ extern uint32_t __libelf_crc32 (uint32_t crc, unsigned char *buf, size_t len)
>  extern void * __libelf_compress (Elf_Scn *scn, size_t hsize, int ei_data,
>  				 size_t *orig_size, size_t *orig_addralign,
> -				 size_t *size, bool force)
> +				 size_t *size, bool force, bool use_zstd)
>       internal_function;
> -extern void * __libelf_decompress (void *buf_in, size_t size_in,
> +extern void * __libelf_decompress (int chtype, void *buf_in, size_t size_in,
>  				   size_t size_out) internal_function;
>  extern void * __libelf_decompress_elf (Elf_Scn *scn,
>  				       size_t *size_out, size_t *addralign)

OK.

> diff --git a/src/elfcompress.c b/src/elfcompress.c
> index eff765e8..b0f1677c 100644
> --- a/src/elfcompress.c
> +++ b/src/elfcompress.c
> @@ -55,9 +55,10 @@ enum ch_type
>    UNSET = -1,
>    NONE,
>    ZLIB,
> +  ZSTD,
>    /* Maximal supported ch_type.  */
> -  MAXIMAL_CH_TYPE = ZLIB,
> +  MAXIMAL_CH_TYPE = ZSTD,
>    ZLIB_GNU = 1 << 16
>  };

OK.

> @@ -139,6 +140,12 @@ parse_opt (int key, char *arg __attribute__ ((unused)),
>  	type = ZLIB;
>        else if (strcmp ("zlib-gnu", arg) == 0 || strcmp ("gnu", arg) == 0)
>  	type = ZLIB_GNU;
> +      else if (strcmp ("zstd", arg) == 0)
> +#ifdef USE_ZSTD
> +	type = ZSTD;
> +#else
> +	argp_error (state, N_("ZSTD support is not enabled"));
> +#endif
>        else
>  	argp_error (state, N_("unknown compression type '%s'"), arg);
>        break;

Nice error handling.

> @@ -281,6 +288,44 @@ get_sections (unsigned int *sections, size_t shnum)
>    return s;
>  }
> +/* Return compression type of a given section SHDR.  */
> +
> +static enum ch_type
> +get_section_chtype (Elf_Scn *scn, GElf_Shdr *shdr, const char *sname,
> +		    size_t ndx)
> +{
> +  enum ch_type chtype = UNSET;
> +  if ((shdr->sh_flags & SHF_COMPRESSED) != 0)
> +    {
> +      GElf_Chdr chdr;
> +      if (gelf_getchdr (scn, &chdr) != NULL)
> +	{
> +	  chtype = (enum ch_type)chdr.ch_type;
> +	  if (chtype == NONE)
> +	    {
> +	      error (0, 0, "Compression type for section %zd"
> +		     " can't be zero ", ndx);
> +	      chtype = UNSET;
> +	    }
> +	  else if (chtype > MAXIMAL_CH_TYPE)
> +	    {
> +	      error (0, 0, "Compression type (%d) for section %zd"
> +		     " is unsupported ", chtype, ndx);
> +	      chtype = UNSET;
> +	    }
> +	}
> +      else
> +	error (0, 0, "Couldn't get chdr for section %zd", ndx);

Shouldn't this error be fatal?

> +    }
> +  /* Set ZLIB_GNU compression manually for .zdebug* sections.  */
> +  else if (startswith (sname, ".zdebug"))
> +    chtype = ZLIB_GNU;
> +  else
> +    chtype = NONE;
> +
> +  return chtype;
> +}
> +
>  static int
>  process_file (const char *fname)
>  {
> @@ -461,26 +506,29 @@ process_file (const char *fname)
>        if (section_name_matches (sname))
>  	{
> -	  if (!force && type == NONE
> -	      && (shdr->sh_flags & SHF_COMPRESSED) == 0
> -	      && !startswith (sname, ".zdebug"))
> -	    {
> -	      if (verbose > 0)
> -		printf ("[%zd] %s already decompressed\n", ndx, sname);
> -	    }
> -	  else if (!force && type == ZLIB
> -		   && (shdr->sh_flags & SHF_COMPRESSED) != 0)
> -	    {
> -	      if (verbose > 0)
> -		printf ("[%zd] %s already compressed\n", ndx, sname);
> -	    }
> -	  else if (!force && type == ZLIB_GNU
> -		   && startswith (sname, ".zdebug"))
> +	  enum ch_type schtype = get_section_chtype (scn, shdr, sname, ndx);
> +	  if (!force && verbose > 0)
>  	    {
> -	      if (verbose > 0)
> -		printf ("[%zd] %s already GNU compressed\n", ndx, sname);
> +	      /* The current compression matches the final one.  */
> +	      if (type == schtype)
> +		switch (type)
> +		  {
> +		  case NONE:
> +		    printf ("[%zd] %s already decompressed\n", ndx, sname);
> +		    break;
> +		  case ZLIB:
> +		  case ZSTD:
> +		    printf ("[%zd] %s already compressed\n", ndx, sname);
> +		    break;
> +		  case ZLIB_GNU:
> +		    printf ("[%zd] %s already GNU compressed\n", ndx, sname);
> +		    break;
> +		  default:
> +		    abort ();
> +		  }
>  	    }
> -	  else if (shdr->sh_type != SHT_NOBITS
> +
> +	  if (shdr->sh_type != SHT_NOBITS
>  	      && (shdr->sh_flags & SHF_ALLOC) == 0)
>  	    {
>  	      set_section (sections, ndx);
> @@ -692,37 +740,12 @@ process_file (const char *fname)
>  	     (de)compressed, invalidating the string pointers.  */
>  	  sname = xstrdup (sname);
> +
>  	  /* Detect source compression that is how is the section compressed
>  	     now.  */
> -	  GElf_Chdr chdr;
> -	  enum ch_type schtype = NONE;
> -	  if ((shdr->sh_flags & SHF_COMPRESSED) != 0)
> -	    {
> -	      if (gelf_getchdr (scn, &chdr) != NULL)
> -		{
> -		  schtype = (enum ch_type)chdr.ch_type;
> -		  if (schtype == NONE)
> -		    {
> -		      error (0, 0, "Compression type for section %zd"
> -			     " can't be zero ", ndx);
> -		      goto cleanup;
> -		    }
> -		  else if (schtype > MAXIMAL_CH_TYPE)
> -		    {
> -		      error (0, 0, "Compression type (%d) for section %zd"
> -			     " is unsupported ", schtype, ndx);
> -		      goto cleanup;
> -		    }
> -		}
> -	      else
> -		{
> -		  error (0, 0, "Couldn't get chdr for section %zd", ndx);
> -		  goto cleanup;
> -		}
> -	    }
> -	  /* Set ZLIB compression manually for .zdebug* sections.  */
> -	  else if (startswith (sname, ".zdebug"))
> -	    schtype = ZLIB_GNU;
> +	  enum ch_type schtype = get_section_chtype (scn, shdr, sname, ndx);
> +	  if (schtype == UNSET)
> +	    goto cleanup;
>  	  /* We might want to decompress (and rename), but not
>  	     compress during this pass since we might need the section
> @@ -754,7 +777,7 @@ process_file (const char *fname)
>  	    case ZLIB_GNU:
>  	      if (startswith (sname, ".debug"))
>  		{
> -		  if (schtype == ZLIB)
> +		  if (schtype == ZLIB || schtype == ZSTD)
>  		    {
>  		      /* First decompress to recompress GNU style.
>  			 Don't report even when verbose.  */

OK.

> @@ -818,19 +841,22 @@ process_file (const char *fname)
>  	      break;
>  	    case ZLIB:
> -	      if ((shdr->sh_flags & SHF_COMPRESSED) == 0)
> +	    case ZSTD:
> +	      if (schtype != type)
>  		{
> -		  if (schtype == ZLIB_GNU)
> +		  if (schtype != NONE)
>  		    {
> -		      /* First decompress to recompress zlib style.
> -			 Don't report even when verbose.  */
> +		      /* Decompress first.  */
>  		      if (compress_section (scn, size, sname, NULL, ndx,
>  					    schtype, NONE, false) < 0)
>  			goto cleanup;
> -		      snamebuf[0] = '.';
> -		      strcpy (&snamebuf[1], &sname[2]);
> -		      newname = snamebuf;
> +		      if (schtype == ZLIB_GNU)
> +			{
> +			  snamebuf[0] = '.';
> +			  strcpy (&snamebuf[1], &sname[2]);
> +			  newname = snamebuf;
> +			}
>  		    }
>  		  if (skip_compress_section)

OK.

> @@ -838,7 +864,7 @@ process_file (const char *fname)
>  		      if (ndx == shdrstrndx)
>  			{
>  			  shstrtab_size = size;
> -			  shstrtab_compressed = ZLIB;
> +			  shstrtab_compressed = type;
>  			  if (shstrtab_name != NULL
>  			      || shstrtab_newname != NULL)
>  			    {
> @@ -855,7 +881,7 @@ process_file (const char *fname)
>  		      else
>  			{
>  			  symtab_size = size;
> -			  symtab_compressed = ZLIB;
> +			  symtab_compressed = type;
>  			  symtab_name = xstrdup (sname);
>  			  symtab_newname = (newname == NULL
>  					    ? NULL : xstrdup (newname));

OK.

> @@ -1378,7 +1404,7 @@ main (int argc, char **argv)
>  	N_("Place (de)compressed output into FILE"),
>  	0 },
>        { "type", 't', "TYPE", 0,
> -	N_("What type of compression to apply. TYPE can be 'none' (decompress), 'zlib' (ELF ZLIB compression, the default, 'zlib-gabi' is an alias) or 'zlib-gnu' (.zdebug GNU style compression, 'gnu' is an alias)"),
> +	N_("What type of compression to apply. TYPE can be 'none' (decompress), 'zlib' (ELF ZLIB compression, the default, 'zlib-gabi' is an alias), 'zlib-gnu' (.zdebug GNU style compression, 'gnu' is an alias) or 'zstd'"),
>  	0 },
>        { "name", 'n', "SECTION", 0,
>  	N_("SECTION name to (de)compress, SECTION is an extended wildcard pattern (defaults to '.?(z)debug*')"),

I would say or 'zstd' (ELF ZSTD compression)" to match the 'zlib; type
description.

> diff --git a/src/readelf.c b/src/readelf.c
> index cc3e0229..451f8400 100644
> --- a/src/readelf.c
> +++ b/src/readelf.c
> @@ -1238,13 +1238,17 @@ get_visibility_type (int value)
>  static const char *
>  elf_ch_type_name (unsigned int code)
>  {
> -  if (code == 0)
> -    return "NONE";
> -
> -  if (code == ELFCOMPRESS_ZLIB)
> -    return "ZLIB";
> -
> -  return "UNKNOWN";
> +  switch (code)
> +    {
> +    case 0:
> +      return "NONE";
> +    case ELFCOMPRESS_ZLIB:
> +      return "ZLIB";
> +    case ELFCOMPRESS_ZSTD:
> +      return "ZSTD";
> +    default:
> +      return "UNKNOWN";
> +    }
>  }
>  /* Print the section headers.  */

OK.

> diff --git a/tests/run-compress-test.sh b/tests/run-compress-test.sh
> index a6a298f5..3f9c990e 100755
> --- a/tests/run-compress-test.sh
> +++ b/tests/run-compress-test.sh
> @@ -61,6 +61,30 @@ testrun_elfcompress_file()
>      echo "uncompress $elfcompressedfile -> $elfuncompressedfile"
>      testrun ${abs_top_builddir}/src/elfcompress -v -t none -o ${elfuncompressedfile} ${elfcompressedfile}
>      testrun ${abs_top_builddir}/src/elfcmp ${uncompressedfile} ${elfuncompressedfile}
> +
> +    outputfile="${infile}.gabi.zstd"
> +    tempfiles "$outputfile"
> +    echo "zstd compress $elfcompressedfile -> $outputfile"
> +    testrun ${abs_top_builddir}/src/elfcompress -v -t zstd -o ${outputfile} ${elfcompressedfile}
> +    testrun ${abs_top_builddir}/src/elfcmp ${uncompressedfile} ${outputfile}
> +    echo "checking compressed section header" $outputfile
> +    testrun ${abs_top_builddir}/src/readelf -Sz ${outputfile} | grep "ELF ZSTD" >/dev/null
> +
> +    zstdfile="${infile}.zstd"
> +    tempfiles "$zstdfile"
> +    echo "zstd compress $uncompressedfile -> $zstdfile"
> +    testrun ${abs_top_builddir}/src/elfcompress -v -t zstd -o ${zstdfile} ${elfuncompressedfile}
> +    testrun ${abs_top_builddir}/src/elfcmp ${uncompressedfile} ${zstdfile}
> +    echo "checking compressed section header" $zstdfile
> +    testrun ${abs_top_builddir}/src/readelf -Sz ${zstdfile} | grep "ELF ZSTD" >/dev/null
> +
> +    zstdgnufile="${infile}.zstd.gnu"
> +    tempfiles "$zstdgnufile"
> +    echo "zstd re-compress to GNU ZLIB $zstdfile -> $zstdgnufile"
> +    testrun ${abs_top_builddir}/src/elfcompress -v -t zlib-gnu -o ${zstdgnufile} ${zstdfile}
> +    testrun ${abs_top_builddir}/src/elfcmp ${uncompressedfile} ${zstdgnufile}
> +    echo "checking .zdebug section name" $zstdgnufile
> +    testrun ${abs_top_builddir}/src/readelf -S ${zstdgnufile} | grep ".zdebug" >/dev/null
>  }
>  testrun_elfcompress()

You might add these to a separate run test file or pass some ZSTD flag
through the test environment to conditionally run these new tests.

Cheers,

Mark