public inbox for binutils@sourceware.org
 help / color / mirror / Atom feed
* RFC: Supporting SOURCE_DATE_EPOCH in ar
@ 2023-08-29  9:46 Nick Clifton
  2023-08-29 16:26 ` Fangrui Song
       [not found] ` <DS7PR12MB576518157026CC92239F6470CBE7A@DS7PR12MB5765.namprd12.prod.outlook.com>
  0 siblings, 2 replies; 3+ messages in thread
From: Nick Clifton @ 2023-08-29  9:46 UTC (permalink / raw)
  To: binutils

[-- Attachment #1: Type: text/plain, Size: 1320 bytes --]

Hi Guys,

  I have been looking improving our support for reproducible builds, and
  in particular how we might be able to use the SOURCE_DATE_EPOCH
  environment variable when creating and maintaining static archives.  I
  think that it can be done, and the attached patch is my proposal for
  implementing this.

  The patch makes several changes:

    * Adding files to a reproduicble archive, either one created with
      --enable-deterministic-archives or with SOURCE_DATA_EPOCH set is
      allowed.  If the update ('u') modifier is in effect it will be
      ignored and the new file(s) will always replace the old file(s).

    * Running "ar ruD" will no longer trigger an error, but instead give
      a warning message and then proceed as outlined above.

    * It adds a new function - bfd_get_current_time() - which can be
      used instead of calling time().  This will return
      SOURCE_DATE_EPOCH if it is defined.  The function can also be used
      to conditionally update the st_mtime field in a stat structure,
      changing it only if SOURCE_DATE_EPOCH is defined.

    * It adds three new tests to the binutils testsuite to check that
      the new behaviour is supported correctly.

  Any comments or suggestions ?
  
Cheers
  Nick
  
https://reproducible-builds.org/docs/source-date-epoch/


[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: ar-SOURCE_DATE_EPOCH.patch --]
[-- Type: text/x-patch, Size: 15225 bytes --]

diff --git a/bfd/archive.c b/bfd/archive.c
index 47b37bb6e37..1c7dc44ae8d 100644
--- a/bfd/archive.c
+++ b/bfd/archive.c
@@ -1893,7 +1893,7 @@ bfd_ar_hdr_from_filesystem (bfd *abfd, const char *filename, bfd *member)
     {
       /* Assume we just "made" the member, and fake it.  */
       struct bfd_in_memory *bim = (struct bfd_in_memory *) member->iostream;
-      time (&status.st_mtime);
+      status.st_mtime = bfd_get_current_time (0);
       status.st_uid = getuid ();
       status.st_gid = getgid ();
       status.st_mode = 0644;
@@ -1904,6 +1904,15 @@ bfd_ar_hdr_from_filesystem (bfd *abfd, const char *filename, bfd *member)
       bfd_set_error (bfd_error_system_call);
       return NULL;
     }
+  else
+    {
+      /* The call to stat() above has filled in the st_mtime field
+	 with the real time that the object was modified.  But if
+	 we are trying to generate deterministic archives based upon
+	 the SOURCE_DATE_EPOCH environment variable then we want to
+	 override that.  */
+      status.st_mtime = bfd_get_current_time (status.st_mtime);
+    }
 
   /* If the caller requested that the BFD generate deterministic output,
      fake values for modification time, UID, GID, and file mode.  */
@@ -2529,8 +2538,13 @@ _bfd_bsd_write_armap (bfd *arch,
       struct stat statbuf;
 
       if (stat (bfd_get_filename (arch), &statbuf) == 0)
-	bfd_ardata (arch)->armap_timestamp = (statbuf.st_mtime
-					      + ARMAP_TIME_OFFSET);
+	{
+	  /* If asked, replace the time with a deterministic value. */
+	  statbuf.st_mtime = bfd_get_current_time (statbuf.st_mtime);
+
+	  bfd_ardata (arch)->armap_timestamp = (statbuf.st_mtime
+						+ ARMAP_TIME_OFFSET);
+	}
       uid = getuid();
       gid = getgid();
     }
@@ -2617,7 +2631,8 @@ _bfd_bsd_write_armap (bfd *arch,
 }
 
 /* At the end of archive file handling, update the timestamp in the
-   file, so the linker will accept it.
+   file, so older linkers will accept it.  (This does not apply to
+   ld.bfd or ld.gold).
 
    Return TRUE if the timestamp was OK, or an unusual problem happened.
    Return FALSE if we updated the timestamp.  */
@@ -2642,10 +2657,17 @@ _bfd_archive_bsd_update_armap_timestamp (bfd *arch)
       /* Can't read mod time for some reason.  */
       return true;
     }
+
   if (((long) archstat.st_mtime) <= bfd_ardata (arch)->armap_timestamp)
     /* OK by the linker's rules.  */
     return true;
 
+  if (getenv ("SOURCE_DATE_EPOCH") != NULL
+      && bfd_ardata (arch)->armap_timestamp == bfd_get_current_time (0) + ARMAP_TIME_OFFSET)
+    /* If the archive's timestamp has been set to SOURCE_DATE_EPOCH
+       then leave it as-is.  */
+    return true;
+  
   /* Update the timestamp.  */
   bfd_ardata (arch)->armap_timestamp = archstat.st_mtime + ARMAP_TIME_OFFSET;
 
diff --git a/bfd/archive64.c b/bfd/archive64.c
index 63d2393ccfd..b80f3e8c4e5 100644
--- a/bfd/archive64.c
+++ b/bfd/archive64.c
@@ -186,8 +186,16 @@ _bfd_archive_64_bit_write_armap (bfd *arch,
   memcpy (hdr.ar_name, "/SYM64/", strlen ("/SYM64/"));
   if (!_bfd_ar_sizepad (hdr.ar_size, sizeof (hdr.ar_size), mapsize))
     return false;
-  _bfd_ar_spacepad (hdr.ar_date, sizeof (hdr.ar_date), "%ld",
-		    time (NULL));
+
+  time_t date;
+
+  if (arch->flags & BFD_DETERMINISTIC_OUTPUT)
+    date = 0;
+  else
+    date = bfd_get_current_time (0);
+
+  _bfd_ar_spacepad (hdr.ar_date, sizeof (hdr.ar_date), "%ld", (long) date);
+  
   /* This, at least, is what Intel coff sets the values to.: */
   _bfd_ar_spacepad (hdr.ar_uid, sizeof (hdr.ar_uid), "%ld", 0);
   _bfd_ar_spacepad (hdr.ar_gid, sizeof (hdr.ar_gid), "%ld", 0);
diff --git a/bfd/bfd-in2.h b/bfd/bfd-in2.h
index 1c4f75ae244..eddb9902f5e 100644
--- a/bfd/bfd-in2.h
+++ b/bfd/bfd-in2.h
@@ -2758,6 +2758,8 @@ void *bfd_mmap (bfd *abfd, void *addr, bfd_size_type len,
     void **map_addr, bfd_size_type *map_len)
 ATTRIBUTE_WARN_UNUSED_RESULT;
 
+time_t bfd_get_current_time (time_t now);
+
 /* Extracted from bfdwin.c.  */
 struct _bfd_window_internal;
 
diff --git a/bfd/bfdio.c b/bfd/bfdio.c
index 457562f8c7c..7b8608b45dc 100644
--- a/bfd/bfdio.c
+++ b/bfd/bfdio.c
@@ -828,3 +828,47 @@ const struct bfd_iovec _bfd_memory_iovec =
   &memory_bread, &memory_bwrite, &memory_btell, &memory_bseek,
   &memory_bclose, &memory_bflush, &memory_bstat, &memory_bmmap
 };
+
+/*
+FUNCTION
+	bfd_get_current_time
+
+SYNOPSIS
+	time_t bfd_get_current_time (time_t now);
+
+DESCRIPTION
+	Returns the current time.
+
+	If the environment variable SOURCE_DATE_EPOCH is defined
+	then this is parsed and its value is returned.  Otherwise
+	if the paramter NOW is non-zero, then that is returned.
+	Otherwise the result of the system call "time(NULL)" is
+	returned.
+*/
+
+time_t
+bfd_get_current_time (time_t now)
+{
+  char *source_date_epoch;
+  unsigned long long epoch;
+
+  /* FIXME: We could probably cache this lookup,
+     and the parsing of its value below.  */
+  source_date_epoch = getenv ("SOURCE_DATE_EPOCH");
+
+  if (source_date_epoch == NULL)
+    {
+      if (now)
+	return now;
+      return time (NULL);
+    }
+
+  epoch = strtoull (source_date_epoch, NULL, 0);
+
+  /* If epoch == 0 then something is wrong with SOURCE_DATE_EPOCH,
+     but we do not have an easy way to report it.  Since the presence
+     of the environment variable implies that the user wants
+     deterministic behaviour we just accept the 0 value.  */
+
+  return (time_t) epoch;
+}
diff --git a/binutils/ar.c b/binutils/ar.c
index fe2d971962a..b09a3e04b71 100644
--- a/binutils/ar.c
+++ b/binutils/ar.c
@@ -807,7 +807,7 @@ main (int argc, char **argv)
 	fatal (_("`u' is only meaningful with the `r' option."));
 
       if (newer_only && deterministic > 0)
-        fatal (_("`u' is not meaningful with the `D' option."));
+        non_fatal (_("`u' is not meaningful with the `D' option - replacement will always happen."));
 
       if (newer_only && deterministic < 0 && DEFAULT_AR_DETERMINISTIC)
         non_fatal (_("\
@@ -1482,6 +1482,7 @@ replace_members (bfd *arch, char **files_to_move, bool quick)
 		  && current->arelt_data != NULL)
 		{
 		  bool replaced;
+
 		  if (newer_only)
 		    {
 		      struct stat fsbuf, asbuf;
@@ -1492,12 +1493,33 @@ replace_members (bfd *arch, char **files_to_move, bool quick)
 			    bfd_fatal (*files_to_move);
 			  goto next_file;
 			}
+
 		      if (bfd_stat_arch_elt (current, &asbuf) != 0)
 			/* xgettext:c-format */
 			fatal (_("internal stat error on %s"),
 			       bfd_get_filename (current));
 
 		      if (fsbuf.st_mtime <= asbuf.st_mtime)
+			/* A note about deterministic timestamps:  In an
+			   archive created in a determistic manner the
+			   individual elements will either have a timestamp
+			   of 0 or SOURCE_DATE_EPOCH, depending upon the
+			   method used.  This will be the value retrieved
+			   by bfd_stat_arch_elt().
+
+			   The timestamp in fsbuf.st_mtime however will
+			   definitely be greater than 0, and it is unlikely
+			   to be less than SOURCE_DATE_EPOCH.  (FIXME:
+			   should we test for this case case and issue an
+			   error message ?)
+
+			   So in either case fsbuf.st_mtime > asbuf.st_time
+			   and hence the incoming file will replace the
+			   current file.  Which is what should be expected to
+			   happen.  Deterministic archives have no real sense
+			   of the time/date when their elements were created,
+			   and so any updates to the archive should always
+			   result in replaced files.  */
 			goto next_file;
 		    }
 
diff --git a/binutils/doc/binutils.texi b/binutils/doc/binutils.texi
index 9f80f398c9d..4f9fb1cde9c 100644
--- a/binutils/doc/binutils.texi
+++ b/binutils/doc/binutils.texi
@@ -521,6 +521,10 @@ operation @samp{r} (replace).  In particular, the combination @samp{qu} is
 not allowed, since checking the timestamps would lose any speed
 advantage from the operation @samp{q}.
 
+Note - if an archive has been created in a deterministic manner, eg
+via the use of the @option{D} modifier, then replacement will always
+happen and the @option{u} modifier will be ineffective.
+
 @item U
 @cindex deterministic archives
 @kindex --enable-deterministic-archives
diff --git a/binutils/testsuite/binutils-all/ar.exp b/binutils/testsuite/binutils-all/ar.exp
index b9a325cff4b..a0fa54d866d 100644
--- a/binutils/testsuite/binutils-all/ar.exp
+++ b/binutils/testsuite/binutils-all/ar.exp
@@ -467,6 +467,225 @@ proc deterministic_archive { } {
     pass $testname
 }
 
+# Test replacing a member of a deterministic archive.
+
+proc replacing_deterministic_member { } {
+    global AR
+    global AS
+    global NM
+    global srcdir
+    global subdir
+    global obj
+
+    set testname "replacing deterministic member"
+
+    if [is_remote host] {
+	# The kind of filename trickery that we are about to
+	#  play is hard to do if we have to operate remotely.
+	unsupported $testname
+	return
+    }
+
+    file mkdir tmpdir/ar
+    
+    if ![binutils_assemble $srcdir/$subdir/bintest.s tmpdir/bintest.${obj}] {
+	unsupported $testname
+	return
+    }
+
+    # Wait a second and then build a second object file - with the same name
+    # as the first, but in a different directory.
+    sleep 1
+    if ![binutils_assemble $srcdir/$subdir/copytest.s tmpdir/ar/bintest.${obj}] {
+	unsupported $testname
+	return
+    }
+
+    set archive tmpdir/artest.a
+    set older_objfile tmpdir/bintest.${obj}
+    set newer_objfile tmpdir/ar/bintest.${obj}
+    set older_length [file size $older_objfile]
+    # set newer_length [file size $newer_objfile]
+    
+    remote_file build delete tmpdir/artest.a
+
+    # Build the archive with the *newer* object file.
+    
+    set got [binutils_run $AR "rcD $archive ${newer_objfile}"]
+    if ![string match "" $got] {
+	fail "$testname: (could not build archive)"
+	return
+    }
+
+    # Now replace the newer file with the older one.  On a normal
+    # archive this will not work, but one created to be deterministic
+    # should always replace its members.
+    
+    set got [binutils_run $AR "ruD $archive $older_objfile"]
+    # The archiver will warn that 'u' and 'D' do not work together
+    if ![string match "*not meaningful*" $got] {
+	fail "$testname: (failed to replace file)"
+	return
+    }
+
+    set got [binutils_run $AR "tvO $archive"]
+    if ![string match "rw-r--r-- 0/0 *${older_length} *bintest.${obj} 0x*" $got] {
+	fail "$testname (wrong size, expected: $older_length)"
+	return
+    }
+
+    pass $testname
+}
+
+# Test replacing a member of a non-deterministic archive.
+
+proc replacing_non_deterministic_member { } {
+    global AR
+    global AS
+    global NM
+    global srcdir
+    global subdir
+    global obj
+
+    set testname "replacing non-deterministic member"
+
+    if [is_remote host] {
+	# The kind of filename trickery that we are about to
+	#  play is hard to do if we have to operate remotely.
+	unsupported $testname
+	return
+    }
+
+    file mkdir tmpdir/ar
+    
+    if ![binutils_assemble $srcdir/$subdir/bintest.s tmpdir/bintest.${obj}] {
+	unsupported $testname
+	return
+    }
+
+    # Wait a second and then build a second object file - with the same name
+    # as the first, but in a different directory.
+    sleep 1
+    if ![binutils_assemble $srcdir/$subdir/copytest.s tmpdir/ar/bintest.${obj}] {
+	unsupported $testname
+	return
+    }
+
+    set archive tmpdir/artest.a
+    set older_objfile tmpdir/bintest.${obj}
+    set newer_objfile tmpdir/ar/bintest.${obj}
+    # set older_length [file size $older_objfile]
+    set newer_length [file size $newer_objfile]
+    
+    remote_file build delete tmpdir/artest.a
+
+    # Build the archive with the *newer* object file.
+    
+    set got [binutils_run $AR "rc $archive ${newer_objfile}"]
+    if ![string match "" $got] {
+	fail "$testname: (could not build archive)"
+	return
+    }
+
+    # Now try to replace the newer file with the older one.  This should not work.
+    
+    set got [binutils_run $AR "ru $archive $older_objfile"]
+    if ![string match "" $got] {
+	fail "$testname: (failed to replace file)"
+	return
+    }
+
+    # Since this archive is non-deterministic, we do not know what the
+    # user or group ids will be, so we have to use */* to match them.
+    set got [binutils_run $AR "tvO $archive"]
+    if ![string match "rw-r--r-- */* *${newer_length} *bintest.${obj} 0x*" $got] {
+	fail "$testname (wrong size, expected: $newer_length)"
+	return
+    }
+
+    pass $testname
+}
+
+# Test replacing a member of deterministic archive created by using SOURCE_DATE_EPOCH.
+
+proc replacing_sde_deterministic_member { } {
+    global AR
+    global AS
+    global NM
+    global srcdir
+    global subdir
+    global obj
+
+    set testname "replacing SOURCE_DATE_EPOCH deterministic member"
+
+    if [is_remote host] {
+	# The kind of filename trickery that we are about to
+	#  play is hard to do if we have to operate remotely.
+	unsupported $testname
+	return
+    }
+
+    file mkdir tmpdir/ar
+    
+    if ![binutils_assemble $srcdir/$subdir/bintest.s tmpdir/bintest.${obj}] {
+	unsupported $testname
+	return
+    }
+
+    # Wait a second and then build a second object file - with the same name
+    # as the first, but in a different directory.
+    sleep 1
+    if ![binutils_assemble $srcdir/$subdir/copytest.s tmpdir/ar/bintest.${obj}] {
+	unsupported $testname
+	return
+    }
+
+    set archive tmpdir/artest.a
+    set older_objfile tmpdir/bintest.${obj}
+    set newer_objfile tmpdir/ar/bintest.${obj}
+    set older_length [file size $older_objfile]
+    # set newer_length [file size $newer_objfile]
+    
+    remote_file build delete tmpdir/artest.a
+
+    # Build the archive with the *newer* object file.
+    setenv SOURCE_DATE_EPOCH "1000"
+    
+    set got [binutils_run $AR "rc $archive ${newer_objfile}"]
+    if ![string match "" $got] {
+	fail "$testname: (could not build archive)"
+	unsetenv SOURCE_DATE_EPOCH
+	return
+    }
+
+    # Now replace the newer file with the older one.  On a normal
+    # archive this will not work, but one created to be deterministic
+    # should always replace its members.
+    
+    set got [binutils_run $AR "ru $archive $older_objfile"]
+    if ![string match "" $got] {
+	fail "$testname: (failed to replace file)"
+	unsetenv SOURCE_DATE_EPOCH
+	return
+    }
+
+    # Since this archive has fixed source dates, but non-deterministic
+    # uid and gid values we have to use */* to match them.
+    set got [binutils_run $AR "tvO $archive"]
+    if ![string match "rw-r--r-- */* *${older_length} *bintest.${obj} 0x*" $got] {
+	fail "$testname (wrong size, expected: $older_length)"
+	unsetenv SOURCE_DATE_EPOCH
+	return
+    }
+
+    # FIXME - it would be nice if we could check to see that the time & date
+    # in the archive listing matches SOURCE_DATE_EPOCH.
+
+    unsetenv SOURCE_DATE_EPOCH
+    pass $testname
+}
+
+
 proc unique_symbol { } {
     global AR
     global AS
@@ -797,6 +1016,9 @@ if { [file exists $base_dir/bfdtest1] && [file exists $base_dir/bfdtest2] } {
 symbol_table
 argument_parsing
 deterministic_archive
+replacing_deterministic_member
+replacing_non_deterministic_member
+replacing_sde_deterministic_member
 delete_an_element
 move_an_element
 empty_archive

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: RFC: Supporting SOURCE_DATE_EPOCH in ar
  2023-08-29  9:46 RFC: Supporting SOURCE_DATE_EPOCH in ar Nick Clifton
@ 2023-08-29 16:26 ` Fangrui Song
       [not found] ` <DS7PR12MB576518157026CC92239F6470CBE7A@DS7PR12MB5765.namprd12.prod.outlook.com>
  1 sibling, 0 replies; 3+ messages in thread
From: Fangrui Song @ 2023-08-29 16:26 UTC (permalink / raw)
  To: Nick Clifton; +Cc: binutils

On Tue, Aug 29, 2023 at 2:46 AM Nick Clifton via Binutils
<binutils@sourceware.org> wrote:
>
> Hi Guys,
>
>   I have been looking improving our support for reproducible builds, and
>   in particular how we might be able to use the SOURCE_DATE_EPOCH
>   environment variable when creating and maintaining static archives.  I
>   think that it can be done, and the attached patch is my proposal for
>   implementing this.
>
>   The patch makes several changes:
>
>     * Adding files to a reproduicble archive, either one created with
>       --enable-deterministic-archives or with SOURCE_DATA_EPOCH set is
>       allowed.  If the update ('u') modifier is in effect it will be
>       ignored and the new file(s) will always replace the old file(s).
>
>     * Running "ar ruD" will no longer trigger an error, but instead give
>       a warning message and then proceed as outlined above.
>
>     * It adds a new function - bfd_get_current_time() - which can be
>       used instead of calling time().  This will return
>       SOURCE_DATE_EPOCH if it is defined.  The function can also be used
>       to conditionally update the st_mtime field in a stat structure,
>       changing it only if SOURCE_DATE_EPOCH is defined.
>
>     * It adds three new tests to the binutils testsuite to check that
>       the new behaviour is supported correctly.
>
>   Any comments or suggestions ?
>
> Cheers
>   Nick
>
> https://reproducible-builds.org/docs/source-date-epoch/
>

Hi Nick,

Thanks for improving reproducibility!

I wonder whether binutils/configure.ac can default to
default_ar_deterministic=1.
https://wiki.debian.org/ReproducibleBuilds/TimestampsInStaticLibraries
Debian "binutils is now built with --enable-deterministic-archives
since version 2.25-6."
https://bugzilla.redhat.com/show_bug.cgi?id=1124342

With these changes, do we need more work on a
--disable-deterministic-archives build of ar operating on
deterministic archives?

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: RFC: Supporting SOURCE_DATE_EPOCH in ar
       [not found] ` <DS7PR12MB576518157026CC92239F6470CBE7A@DS7PR12MB5765.namprd12.prod.outlook.com>
@ 2023-09-01 12:01   ` Nick Clifton
  0 siblings, 0 replies; 3+ messages in thread
From: Nick Clifton @ 2023-09-01 12:01 UTC (permalink / raw)
  To: Fangrui Song; +Cc: binutils

Hi Fangrui,

> Thanks for improving reproducibility!

You are welcome.  Since nobody else commented or complained, I have gone
ahead and committed my patch.


> I wonder whether binutils/configure.ac can default to
> default_ar_deterministic=1.

I think that this might be a bad idea, because it will confuse users
who are unaware the usefulness of reproducible builds.  Developers and
distributions that are aware of the feature will be able to configure
their tools correctly, so I think that the current status quo is best.

Cheers
   Nick





^ permalink raw reply	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2023-09-01 12:01 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2023-08-29  9:46 RFC: Supporting SOURCE_DATE_EPOCH in ar Nick Clifton
2023-08-29 16:26 ` Fangrui Song
     [not found] ` <DS7PR12MB576518157026CC92239F6470CBE7A@DS7PR12MB5765.namprd12.prod.outlook.com>
2023-09-01 12:01   ` Nick Clifton

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).