[PATCH] libdwfl: add dwfl_local_find_debuginfo callback

[PATCH] libdwfl: add dwfl_local_find_debuginfo callback

[Nick Gasson]

This patch adds a variant of dwfl_standard_find_debuginfo that only
searches the local filesystem and does not fall back to debuginfod.

In my application I am using libdw to decode stack traces where some of
the loaded DSOs are generated on-the-fly by LLVM.  If these don't have
debug information I want the find_debuginfo callback to fail immediately
and not fall back to debuginfod which will never succeed.  Similarly if
the main executable is stripped.  I supposed I could unset the
DEBUGINFOD_URLS environment variable before calling into libdw, but that
seems a bit hacky and liable to break in the future if the debuginfod
client loads the URLs from a different place.

Signed-off-by: Nick Gasson <nick@nickg.me.uk>
---
 libdw/libdw.map          |  1 +
 libdwfl/find-debuginfo.c | 40 ++++++++++++++++++++++++++++++++--------
 libdwfl/libdwfl.h        |  9 ++++++++-
 libdwfl/libdwflP.h       |  1 +
 4 files changed, 42 insertions(+), 9 deletions(-)

diff --git a/libdw/libdw.map b/libdw/libdw.map
index 8ab0a2a062da..b8593f8ebf56 100644
--- a/libdw/libdw.map
+++ b/libdw/libdw.map
@@ -114,6 +114,7 @@ ELFUTILS_0.122 {
     dwfl_linux_proc_find_elf;
     dwfl_linux_proc_maps_report;
     dwfl_linux_proc_report;
+    dwfl_local_find_debuginfo;
     dwfl_module_addrdie;
     dwfl_module_addrname;
     dwfl_module_getdwarf;
diff --git a/libdwfl/find-debuginfo.c b/libdwfl/find-debuginfo.c
index 449df5a14b19..51f3fe622b07 100644
--- a/libdwfl/find-debuginfo.c
+++ b/libdwfl/find-debuginfo.c
@@ -346,14 +346,14 @@ fail_free:
 }
 
 int
-dwfl_standard_find_debuginfo (Dwfl_Module *mod,
-			      void **userdata __attribute__ ((unused)),
-			      const char *modname __attribute__ ((unused)),
-			      GElf_Addr base __attribute__ ((unused)),
-			      const char *file_name,
-			      const char *debuglink_file,
-			      GElf_Word debuglink_crc,
-			      char **debuginfo_file_name)
+dwfl_local_find_debuginfo (Dwfl_Module *mod,
+			   void **userdata __attribute__ ((unused)),
+			   const char *modname __attribute__ ((unused)),
+			   GElf_Addr base __attribute__ ((unused)),
+			   const char *file_name,
+			   const char *debuglink_file,
+			   GElf_Word debuglink_crc,
+			   char **debuginfo_file_name)
 {
   if (mod == NULL)
     return -1;
@@ -401,6 +401,28 @@ dwfl_standard_find_debuginfo (Dwfl_Module *mod,
       free (canon);
     }
 
+  return fd;
+}
+INTDEF (dwfl_local_find_debuginfo)
+
+int
+dwfl_standard_find_debuginfo (Dwfl_Module *mod,
+			      void **userdata,
+			      const char *modname,
+			      GElf_Addr base,
+			      const char *file_name,
+			      const char *debuglink_file,
+			      GElf_Word debuglink_crc,
+			      char **debuginfo_file_name)
+{
+  /* First try to find debug information in the local filesystem. */
+  int fd = INTUSE(dwfl_local_find_debuginfo) (mod, userdata,
+					      modname, base,
+					      file_name,
+					      debuglink_file,
+					      debuglink_crc,
+					      debuginfo_file_name);
+
 #ifdef ENABLE_LIBDEBUGINFOD
   /* Still nothing? Try if we can use the debuginfod client.
      But note that we might be looking for the alt file.
@@ -412,6 +434,8 @@ dwfl_standard_find_debuginfo (Dwfl_Module *mod,
      handles build-ids.  */
   if (fd < 0)
     {
+      const unsigned char *bits = NULL;
+      int bits_len = 0;
       if (mod->dw != NULL)
 	{
 	  const char *altname;
diff --git a/libdwfl/libdwfl.h b/libdwfl/libdwfl.h
index f98f1d525d94..9d3eb43729c7 100644
--- a/libdwfl/libdwfl.h
+++ b/libdwfl/libdwfl.h
@@ -292,12 +292,19 @@ extern int dwfl_build_id_find_debuginfo (Dwfl_Module *, void **,
    If there is no build ID or no valid debuginfo found by ID,
    it searches the debuginfo path by name, as described above.
    Any file found in the path is validated by build ID if possible,
-   or else by CRC32 checksum if enabled, and skipped if it does not match.  */
+   or else by CRC32 checksum if enabled, and skipped if it does not match.
+   The debuginfod client is consulted if no debug information is found
+   in the local filesystem. */
 extern int dwfl_standard_find_debuginfo (Dwfl_Module *, void **,
 					 const char *, Dwarf_Addr,
 					 const char *, const char *,
 					 GElf_Word, char **);
 
+/* As dwfl_standard_find_debuginfo, but does not fall back to debuginfod. */
+extern int dwfl_local_find_debuginfo (Dwfl_Module *, void **,
+				      const char *, Dwarf_Addr,
+				      const char *, const char *,
+				      GElf_Word, char **);
 
 /* This callback must be used when using dwfl_offline_* to report modules,
    if ET_REL is to be supported.  */
diff --git a/libdwfl/libdwflP.h b/libdwfl/libdwflP.h
index 4344e356b452..8be4f341bbab 100644
--- a/libdwfl/libdwflP.h
+++ b/libdwfl/libdwflP.h
@@ -760,6 +760,7 @@ INTDECL (dwfl_report_end)
 INTDECL (dwfl_build_id_find_elf)
 INTDECL (dwfl_build_id_find_debuginfo)
 INTDECL (dwfl_standard_find_debuginfo)
+INTDECL (dwfl_local_find_debuginfo)
 INTDECL (dwfl_link_map_report)
 INTDECL (dwfl_linux_kernel_find_elf)
 INTDECL (dwfl_linux_kernel_module_section_address)
-- 
2.30.2

Re: [PATCH] libdwfl: add dwfl_local_find_debuginfo callback

[Mark Wielaard]

Hi Nick,

Sorry for the late reply.

On Sun, Jun 13, 2021 at 11:27:02AM +0800, Nick Gasson via Elfutils-devel wrote:
> This patch adds a variant of dwfl_standard_find_debuginfo that only
> searches the local filesystem and does not fall back to debuginfod.
> 
> In my application I am using libdw to decode stack traces where some of
> the loaded DSOs are generated on-the-fly by LLVM.  If these don't have
> debug information I want the find_debuginfo callback to fail immediately
> and not fall back to debuginfod which will never succeed.  Similarly if
> the main executable is stripped.  I supposed I could unset the
> DEBUGINFOD_URLS environment variable before calling into libdw, but that
> seems a bit hacky and liable to break in the future if the debuginfod
> client loads the URLs from a different place.

I don't think the (un)setting of DEBUGINFOD_URLS will ever change. It
is API now. But I can see a need for this.

I wonder if we can use the Dwfl_Callbacks debuginfo_path for this
instead of introducing a dwfl_standard_find_debuginfo variant. That
might make things a bit more configurable.

Something like the following maybe:

- If the last path element is "@" then the debuginfo client won't be used.
- If the last path element is "@{URLs}" (space separated) then the debuginfo client
  will be used with to given URLs.
- If there is no path element starting with @ then the default debuginfo client
  will be used after all local searches failed (or not when DEBUGINFOD_URLS isn't set).

Does that sound helpful? Does it make sense to make things configurable like this?

Thanks,

Mark

Re: [PATCH] libdwfl: add dwfl_local_find_debuginfo callback

[Nick Gasson]

Hi Mark,

On 03/07/21 05:59 am, Mark Wielaard wrote:
>
> I wonder if we can use the Dwfl_Callbacks debuginfo_path for this
> instead of introducing a dwfl_standard_find_debuginfo variant. That
> might make things a bit more configurable.
>
> Something like the following maybe:
>
> - If the last path element is "@" then the debuginfo client won't be used.
> - If the last path element is "@{URLs}" (space separated) then the debuginfo client
>   will be used with to given URLs.
> - If there is no path element starting with @ then the default debuginfo client
>   will be used after all local searches failed (or not when DEBUGINFOD_URLS isn't set).
>
> Does that sound helpful? Does it make sense to make things configurable like this?
>

Thanks for the reply.  Yes I think this will work fine for my case.

--
Nick