[bug #63330] readdir() error in 4.4 on Solaris 8

[Paul D. Smith]

Update of bug #63330 (project make):

  Status:None => Not A Bug  
 Open/Closed:Open => Closed 

___

Follow-up Comment #12:

Ah!  Good news!  Thanks for letting us know.


___

Reply to this item at:

  

___
Message sent via Savannah
https://savannah.gnu.org/

[bug #63330] readdir() error in 4.4 on Solaris 8

[anonymous]

Follow-up Comment #11, bug #63330 (project make):

I was working from a git checkout and it appears that it somehow got into a
strange state.  A build of a fresh download of the 4.4 tarball is fine.  My
apologies for the confusion, this can be closed.


___

Reply to this item at:

  

___
Message sent via Savannah
https://savannah.gnu.org/

Re: [bug #63330] readdir() error in 4.4 on Solaris 8

[Jeffrey Walton]

On Tue, Nov 8, 2022 at 10:40 AM Dmitry Goncharov
 wrote:
>
> Follow-up Comment #9, bug #63330 (project make):
>
> Depending on _LARGEFILE64_SOURCE and _FILE_OFFSET_BITS dirent has d_ino and
> d_off fields 32 or 64 bits wide.
> Looks like in your case the remote machine has 64bit d_ino and d_off and make
> is built with 32bit d_ino and d_off.
>
> i don't see checks for _FILE_OFFSET_BITS in your config.log.
> Does make work if you build 64 bit or set -D_FILE_OFFSET_BITS=64?
> What is the difference, in regards to large file support, between your good
> working make-4.3 config.log and config.status and this one that fails?
>
> In regards to the error see
> https://docs.oracle.com/cd/E19455-01/806-1075/msgs-2215/index.html

If I recall correctly, Sun recommends building 64-bit binaries on
64-bit OSes, but AutoTools still defaults to 32-bit builds on the
platform.

>From https://docs.oracle.com/cd/E37838_01/html/E66175/features-1.html :

All new applications must be developed as 64-bit
applications and many of the existing 32-bit applications
might require conversion to 64-bit.

Jeff

[bug #63330] readdir() error in 4.4 on Solaris 8

[Paul D. Smith]

Follow-up Comment #10, bug #63330 (project make):

Wow I'm REALLY confused.  I've examined the config.log file you provided and I
can't make heads or tails of it.  There is no reference anywhere in that file,
to checking for largefile support.  This likely explains why it's not enabled
on this system.

Just to be clear, you did download the official source build from
https://ftp.gnu.org/gnu/make/make-4.4.tar.gz right?  Or, did you create the
package yourself from Git?

Can you run md5sum on the downloaded package and the "configure" script in the
directory; I get:

d7575a26a94ee8427130e9db23cdaa78  make-4.4.tar.gz
3d5221bb26db5cfa8b297738320f33b3  make-4.4/configure


When I look at your log file, I see this:

configure:6535: checking for ranlib
configure:6556: found /usr/ccs/bin/ranlib
configure:6567: result: ranlib
configure:6647: checking for size_t


Right in between the check for ranlib and the check for size_t, is where the
check for largefile support is supposed to be.  Not only is there no output
for it here, but the line numbers in the script are such that it doesn't seem
like there's even any space in the configure script where this might be run!

When I run configure from make-4.4 on my system, I get this:

configure:6506: checking for ranlib
configure:6527: found /usr/bin/ranlib
configure:6538: result: ranlib
configure:6572: checking for special C compiler options needed for large
files
configure:6620: result: no
configure:6626: checking for _FILE_OFFSET_BITS value needed for large files
configure:6652: gcc -c -g -O2  conftest.c >&5
configure:6652: $? = 0
configure:6687: result: no
configure:6962: checking for size_t


Note that it does do the check, and note that all the line numbers are a
little off but the size_t check is a full 300 lines different (presumably
where the largefile support check should be).


PS. from the config.log I note you invoked the configure script as just
"configure".  I recommend you use "./configure" explicitly, if building from
within the source directory, to avoid relying on the PATH settings.  I have no
reason to believe this is the problem since the config.log appears to have
been created from the GNU Make 4.4 configure, but I still recommend using a
specific path.


___

Reply to this item at:

  

___
Message sent via Savannah
https://savannah.gnu.org/

[bug #63330] readdir() error in 4.4 on Solaris 8

[Dmitry Goncharov]

Follow-up Comment #9, bug #63330 (project make):

Depending on _LARGEFILE64_SOURCE and _FILE_OFFSET_BITS dirent has d_ino and
d_off fields 32 or 64 bits wide.
Looks like in your case the remote machine has 64bit d_ino and d_off and make
is built with 32bit d_ino and d_off.

i don't see checks for _FILE_OFFSET_BITS in your config.log.
Does make work if you build 64 bit or set -D_FILE_OFFSET_BITS=64?
What is the difference, in regards to large file support, between your good
working make-4.3 config.log and config.status and this one that fails?

In regards to the error see
https://docs.oracle.com/cd/E19455-01/806-1075/msgs-2215/index.html




___

Reply to this item at:

  

___
Message sent via Savannah
https://savannah.gnu.org/

Re: [bug #63330] readdir() error in 4.4 on Solaris 8

[Jeffrey Walton]

On Tue, Nov 8, 2022 at 8:16 AM anonymous  wrote:
>
> Follow-up Comment #8, bug #63330 (project make):
>
> Just checked and this also affects Solaris 10.  The issue looks identical, let
> me know if you want details from that system.

For testing, gcc210 at the Compile Farm is Solaris 10. See
https://cfarm.tetaneutral.net/machines/list/ .

Jeff

[bug #63330] readdir() error in 4.4 on Solaris 8

[anonymous]

Follow-up Comment #8, bug #63330 (project make):

Just checked and this also affects Solaris 10.  The issue looks identical, let
me know if you want details from that system.


___

Reply to this item at:

  

___
Message sent via Savannah
https://savannah.gnu.org/

[bug #63330] readdir() error in 4.4 on Solaris 8

[anonymous]

Follow-up Comment #7, bug #63330 (project make):

Oh and yes, 4.3 works just fine on NFS mounts.  It's what I've been using to
build 4.4.  The NFS mount is definitely over 4GB but this is the first time
I've run into issues with that.


___

Reply to this item at:

  

___
Message sent via Savannah
https://savannah.gnu.org/

[bug #63330] readdir() error in 4.4 on Solaris 8

[anonymous]

Follow-up Comment #6, bug #63330 (project make):

I definitely don't expect everything to work perfectly on a system this old,
and if it does turn out to be a Solaris bug I'll certainly find a way to deal
with it locally.  That being said I find that compiling on these older systems
sometimes turns up interesting corner cases.

Attached are config.log and config.status from a build configured with GCC
4.7.4.  Note that I had to manually remove -Wformat-signedness and
-Wduplicated-cond from the generated Makefile because GCC 4.7.4 doesn't
understand them, and disable -Werror for src/getopt.c, src/misc.c, and
src/posixos.c all for harmless/unrelated warnings.

(file #53940, file #53941)

___

Additional Item Attachment:

File name: config.log Size:185 KB


File name: config.status  Size:46 KB




___

Reply to this item at:

  

___
Message sent via Savannah
https://savannah.gnu.org/

[bug #63330] readdir() error in 4.4 on Solaris 8

[Paul D. Smith]

Follow-up Comment #5, bug #63330 (project make):

That's certainly disturbing and I would very much like to figure out what's
wrong.  Just to be sure I built and ran the GNU Make regression tests on an
NFS-mounted partition and it worked fine, but this was a GNU/Linux client and
a NetApp NFS server so very different.  I unfortunately don't have access to a
Solaris system.

It's hard to understand how a simple readdir() on an NFS directory would
return EOVERFLOW.  The only thing I can think of is that your version of GNU
make was not built with large file support enabled, and your NFS partition has
sizes >4G, but that should not be the case as we include the gnulib largefile
module.  Solaris 8 was released 22 years ago and went EOL >10 years ago, but
presumably the largefile support will work there regardless.

Just to be clear, when you said that GNU Make 4.3 worked OK did you mean it
worked on this same NFS partition where GNU Make 4.4 fails?

Can you either attach to this bug, or email to me directly at psm...@gnu.org,
the config.log file and config.status file from the GNU Make 4.4 configure
step?


___

Reply to this item at:

  

___
Message sent via Savannah
https://savannah.gnu.org/

Re: [bug #63330] readdir() error in 4.4 on Solaris 8

[Philip Guenther]

Is make being built with -D_FILE_OFFSET_BITS=64 to use the large-file
APIs?  If not, then maybe the problem is an inode number that doesn't fit
in the old 32-bit ino_t that some struct dirent's were stuck with.  Check
the output of "ls -l -i" (note the -i option!) to see if there's an inode
number larger than 2billion in the output, like the '..' directory entry
from the NFS parent filesystem...

(You have less than 16 years of use on that Solaris 8 box, as it still has
a 32-bit time_t, which wraps in Jan 2038.  Type fast!)

Philip

On Mon, Nov 7, 2022 at 8:17 PM anonymous  wrote:

> Follow-up Comment #4, bug #63330 (project make):
>
> It looks like this is somehow related to NFS.  I was running the
> newly-generated make in the root of the source tree, and on an NFS mounted
> directory it fails immediately.  If I copy the tree to a local directory I
> have no problems.
>
>
> ___
>
> Reply to this item at:
>
>   
>
> ___
> Message sent via Savannah
> https://savannah.gnu.org/
>
>
>

[bug #63330] readdir() error in 4.4 on Solaris 8

[anonymous]

Follow-up Comment #4, bug #63330 (project make):

It looks like this is somehow related to NFS.  I was running the
newly-generated make in the root of the source tree, and on an NFS mounted
directory it fails immediately.  If I copy the tree to a local directory I
have no problems.


___

Reply to this item at:

  

___
Message sent via Savannah
https://savannah.gnu.org/

[bug #63330] readdir() error in 4.4 on Solaris 8

[Paul D. Smith]

Follow-up Comment #3, bug #63330 (project make):

It's very strange because the code in 4.4 dir.c was changed in many places (to
introduce directory cache invalidation) but this specific code related to
readdir() is the same as in 4.3.

I think we will need to either get a repro case (even if it only repros on
Solaris 8 just seeing the makefile that fails, and the "make -d" debug output,
could be enlightening), or else access to a system where we can log in to
reproduce the problem, or else someone willing to serve as "remote debugger"
for us to track it down.


___

Reply to this item at:

  

___
Message sent via Savannah
https://savannah.gnu.org/

[bug #63330] readdir() error in 4.4 on Solaris 8

[Paul D. Smith]

Follow-up Comment #2, bug #63330 (project make):

Oh, actually that can't be it.  I forgot that ENULLLOOP() sets errno=0 before
it tries the command.

Hm.  Well, POSIX does allow EOVERFLOW to be returned by readdir() (that's the
error you see).  So maybe there's something about the directory it's trying to
read, that is problematic.

You may need to enhance the error message to show the name of the directory
being parsed, then if you use "ls -al" on it do you see anything strange
there?

For more debugging you could also show each filename as this readdir() reads
it in, then perhaps you'd get an idea of which one is causing the problem
(maybe).


___

Reply to this item at:

  

___
Message sent via Savannah
https://savannah.gnu.org/

[bug #63330] readdir() error in 4.4 on Solaris 8

[Paul D. Smith]

Follow-up Comment #1, bug #63330 (project make):

Hm.  Maybe on Solaris readdir() doesn't reset errno?  In src/dir.c you'll find
this code:


  ENULLLOOP (d, readdir (dir->dirstream));
  if (d == 0)
{
  if (errno)
pfatal_with_name ("INTERNAL: readdir");
  break;
}


Try changing it to set errno first:


  errno = 0; /* add this line */
  ENULLLOOP (d, readdir (dir->dirstream));
  if (d == 0)
{
  if (errno)
pfatal_with_name ("INTERNAL: readdir");
  break;
}



___

Reply to this item at:

  

___
Message sent via Savannah
https://savannah.gnu.org/

[bug #63330] readdir() error in 4.4 on Solaris 8

[anonymous]

URL:
  

 Summary: readdir() error in 4.4 on Solaris 8
 Project: make
   Submitter: None
   Submitted: Mon 07 Nov 2022 03:37:45 PM UTC
Severity: 3 - Normal
  Item Group: Bug
  Status: None
 Privacy: Public
 Assigned to: None
 Open/Closed: Open
 Discussion Lock: Any
   Component Version: 4.4
Operating System: POSIX-Based
   Fixed Release: None
   Triage Status: None


___

Follow-up Comments:


---
Date: Mon 07 Nov 2022 03:37:45 PM UTC By: Anonymous
I'm getting an error in readdir() on Solaris 8 with 4.4 compiled with both gcc
and Sun Studio 11:

make: *** INTERNAL: readdir: Value too large for defined data type.  Stop.

4.3 was fine.  Everything looks OK from a cursory glance, I'm not sure what's
happening here.

Starting program: /scratch/make/make

Breakpoint 1, 0xff2cc484 in readdir () from /usr/lib/libc.so.1
(gdb) where
#0  0xff2cc484 in readdir () from /usr/lib/libc.so.1
#1  0x00019ac8 in dir_contents_file_exists_p (dir=0x6e988, filename=0x44db0
"GNUmakefile") at src/dir.c:726
#2  0x00019fd0 in dir_file_exists_p (filename=0x44db0 "GNUmakefile",
dirname=0x40740 ".") at src/dir.c:805
#3  file_exists_p (name=0x44db0 "GNUmakefile") at src/dir.c:854
#4  0x000324b0 in read_all_makefiles (makefiles=0x0) at src/read.c:256
#5  0x00015b94 in main (argc=, argv=,
envp=) at src/main.c:2069








___

Reply to this item at:

  

___
Message sent via Savannah
https://savannah.gnu.org/