How do I configure with ASAN? #22588

jkeenan · 2024-09-12T22:09:19Z

More than once, I have had to ask, "How do I invoke ./Configure so as to build with ASAN?" (Most recently: #22586 (comment)) But I never seem to succeed. Can we straighten this out so that more people can explore address sanitizers?

My environment:

$ uname -mrs
Linux 6.8.0-40-generic x86_64

$ clang --version
Ubuntu clang version 14.0.0-1ubuntu1.1
Target: x86_64-pc-linux-gnu
Thread model: posix
InstalledDir: /usr/bin

$ gcc --version
gcc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
Copyright (C) 2021 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

$ git describe
v5.41.3

My most recent attempt (from #22586 (comment)):

$ git clean -dfx; sh ./Configure -des -Dusedevel -Accflags=-fsanitize=address -Aldflags=-fsanitize=address -Dcc=clang
...

./Configure gets almost all the way through, then segfaults:

...
Run make depend now? [y]  
clang -c -DPERL_CORE -fsanitize=address -fno-strict-aliasing -pipe -fstack-protector-strong -I/usr/local/include -D_LARGEFILE_SOURCE -D_FILE_OFFSET_BITS=64 -O2 -Wall generate_uudmap.c
clang -o generate_uudmap -fsanitize=address -fstack-protector-strong -L/usr/local/lib generate_uudmap.o -lpthread -ldl -lm -lcrypt -lutil -lc 
./generate_uudmap uudmap.h bitcount.h mg_data.h
make: *** [Makefile:346: bitcount.h] Segmentation fault (core dumped)
...

I should note in passing that it feels awkward to me to delay calling -Dcc=clang until after I've called -Accflags and -Aldflags. But when I put -Dcc=clang where it feels more natural, ./Configure quickly collapses:

$ git clean -dfx; sh ./Configure -des -Dusedevel -Dcc=clang -Accflags=-fsanitize=address -Aldflags=-fsanitize=address
...
Let me guess what the preprocessor flags are...
Any additional ld flags (NOT including libraries)?
[ -fsanitize=address -fstack-protector-strong -L/usr/local/lib]  
Checking your choice of C compiler and flags for coherency...

I've tried to compile and run the following simple program:

#include <stdio.h>
int main() { printf("Ok\n"); return(0); }

I used the command:

	clang -o try -O2 -fsanitize=address -fno-strict-aliasing -pipe -fstack-protector-strong -I/usr/local/include -fsanitize=address -fstack-protector-strong -L/usr/local/lib try.c -lpthread -ldl -lm -lcrypt -lutil -lc
	 ./try

and I got the following output:

Segmentation fault (core dumped)
The program compiled OK, but exited with status 139.
You have a problem.  Shall I abort Configure [y]  
Ok.  Stopping Configure.

Next, let's try the above only with gcc instead of clang:

$ git clean -dfx; sh ./Configure -des -Dusedevel -Dcc=gcc -Accflags=-fsanitize=address -Aldflags=-fsanitize=address
...
AddressSanitizer:DEADLYSIGNAL
AddressSanitizer:DEADLYSIGNAL
AddressSanitizer:DEADLYSIGNAL
...
^C

I get what appears to be an infinite loop of those DEADLYSIGNALs.

And finally, putting gcc at the end of the invocation:

$ git clean -dfx; sh ./Configure -des -Dusedevel -Accflags=-fsanitize=address -Aldflags=-fsanitize=address -Dcc=gcc
...
AddressSanitizer:DEADLYSIGNAL
AddressSanitizer:DEADLYSIGNAL
AddressSanitizer:DEADLYSIGNAL
...
^C

I have a vague recollection that in the past I was able to complete ./Configure only to have my machine grind to a halt during make. But now I can't even get through ./Configure!

Guidance requested.

The text was updated successfully, but these errors were encountered:

khwilliamson · 2024-09-13T02:37:49Z

On 9/12/24 16:09, James E Keenan wrote: More than once, I have had to ask, "How do I invoke |./Configure| so as to build with ASAN?" (Most recently: #22586 (comment) <#22586 (comment)>) But I never seem to succeed. Can we straighten this out so that more people can explore address sanitizers? My environment: |$ uname -mrs Linux 6.8.0-40-generic x86_64 $ clang --version Ubuntu clang version 14.0.0-1ubuntu1.1 Target: x86_64-pc-linux-gnu Thread model: posix InstalledDir: /usr/bin $ gcc --version gcc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0 Copyright (C) 2021 Free Software Foundation, Inc. This is free software; see the source for copying conditions. There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. $ git describe v5.41.3 | My most recent attempt (from #22586 (comment) <#22586 (comment)>): |$ git clean -dfx; sh ./Configure -des -Dusedevel -Accflags=-fsanitize=address -Aldflags=-fsanitize=address -Dcc=clang ... | |./Configure| gets almost all the way through, then segfaults:

This is my precise invocation that worked just now bin/bash ./Configure -des -Uversiononly '-Dprefix=/home/khw/perl/devel' -Dusedevel '-Doptimize=-O0' -DDEBUGGING $'-A\'optimize=-g3\'' $'-A\'optimize=-ggdb3\'' $'-A\'optimize=-g3\'' $'-Accflags=\'-fsanitize=address\'' $'-Accflags=\'-fsanitize-address-use-after-scope\'' $'-Accflags=\'-fno-omit-frame-pointer\'' $'-Accflags=\'-fno-common\'' $'-Accflags=\'-ftrapv\'' $'-Accflags=\'-fsanitize-blacklist=/home/khw/perl/locales/straight/asan_ignore\'' $'-Accflags=\'-Wno-deprecated\'' $'-Accflags=\'-DNO_NL_LOCALE_NAME\'' $'-Dman1dir=\'none\'' $'-Dman3dir=\'none\'' $'-Aldflags=\'-fsanitize=address\'' $'-Alddlflags=\'-fsanitize=address\'' $'-Aldflags=\'-fsanitize-blacklist=/home/khw/perl/locales/straight/asan_ignore\'' $'-Alddlflags=\'-fsanitize-blacklist=/home/khw/perl/locales/straight/asan_ignore\'' $'-Alddlflags=\'-shared\'' '-Dcc=clang' -Dusecbacktrace -Dusethreads Some of these are obsolete or not generally applicable. asan_ignore is shipped with perl.

…

|... Run make depend now? [y] clang -c -DPERL_CORE -fsanitize=address -fno-strict-aliasing -pipe -fstack-protector-strong -I/usr/local/include -D_LARGEFILE_SOURCE -D_FILE_OFFSET_BITS=64 -O2 -Wall generate_uudmap.c clang -o generate_uudmap -fsanitize=address -fstack-protector-strong -L/usr/local/lib generate_uudmap.o -lpthread -ldl -lm -lcrypt -lutil -lc ./generate_uudmap uudmap.h bitcount.h mg_data.h make: *** [Makefile:346: bitcount.h] Segmentation fault (core dumped) ... | I should note in passing that it feels awkward to me to delay calling |-Dcc=clang| until *after* I've called |-Accflags| and |-Aldflags|. But when I put |-Dcc=clang| where it feels more natural, |./Configure| quickly collapses: |$ git clean -dfx; sh ./Configure -des -Dusedevel -Dcc=clang -Accflags=-fsanitize=address -Aldflags=-fsanitize=address ... Let me guess what the preprocessor flags are... Any additional ld flags (NOT including libraries)? [ -fsanitize=address -fstack-protector-strong -L/usr/local/lib] Checking your choice of C compiler and flags for coherency... I've tried to compile and run the following simple program: #include <stdio.h> int main() { printf("Ok\n"); return(0); } I used the command: clang -o try -O2 -fsanitize=address -fno-strict-aliasing -pipe -fstack-protector-strong -I/usr/local/include -fsanitize=address -fstack-protector-strong -L/usr/local/lib try.c -lpthread -ldl -lm -lcrypt -lutil -lc ./try and I got the following output: Segmentation fault (core dumped) The program compiled OK, but exited with status 139. You have a problem. Shall I abort Configure [y] Ok. Stopping Configure. | Next, let's try the above only with |gcc| instead of |clang|: |$ git clean -dfx; sh ./Configure -des -Dusedevel -Dcc=gcc -Accflags=-fsanitize=address -Aldflags=-fsanitize=address ... AddressSanitizer:DEADLYSIGNAL AddressSanitizer:DEADLYSIGNAL AddressSanitizer:DEADLYSIGNAL ... ^C | I get what appears to be an infinite loop of those |DEADLYSIGNAL|s. And finally, putting |gcc| at the end of the invocation: |$ git clean -dfx; sh ./Configure -des -Dusedevel -Accflags=-fsanitize=address -Aldflags=-fsanitize=address -Dcc=gcc ... AddressSanitizer:DEADLYSIGNAL AddressSanitizer:DEADLYSIGNAL AddressSanitizer:DEADLYSIGNAL ... ^C | I have a vague recollection that in the past I was able to complete |./Configure| only to have my machine grind to a halt during |make|. But now I can't even get through |./Configure|! Guidance requested. — Reply to this email directly, view it on GitHub <#22588>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAA2DH5PT3MK7LDFQTBQP6TZWIGKRAVCNFSM6AAAAABOEFI2BSVHI2DSMVQWIX3LMV43ASLTON2WKOZSGUZDGNBUGQ3DQMQ>. You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

mauke · 2024-09-13T03:06:49Z

I can't reproduce your segfaults. I've just built perl from a clean git directory using these exact commands:

./Configure -Dcc=clang -Accflags='-fsanitize=address' -Aldflags='-fsanitize=address' -Dusedevel -des
make

Unsurprisingly, gcc also works here:

./Configure -Dcc=gcc -Accflags='-fsanitize=address' -Aldflags='-fsanitize=address' -Dusedevel -des
make

In either case, I get a working perl:

$ ./perl -v

This is perl 5, version 41, subversion 4 (v5.41.4 (v5.41.3-102-gbd12780ddd)) built for x86_64-linux
...

My environment:

$ uname -mrs
Linux 6.8.0-105041-tuxedo x86_64

$ clang --version
Ubuntu clang version 18.1.3 (1ubuntu1)
Target: x86_64-pc-linux-gnu
Thread model: posix
InstalledDir: /usr/bin

$ gcc --version
gcc (Ubuntu 13.2.0-23ubuntu4) 13.2.0
Copyright (C) 2023 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

$ git describe 
v5.41.3-102-gbd12780ddd

jkeenan · 2024-09-16T20:15:16Z

On my Ubuntu Linux laptop (where the original complaint) was run, I tried @mauke's invocation and got the same failures as before.

Much to my surprise, I got farther on two servers to which I have access.

Debian 11

$ uname -mrs
Linux 6.1.0-23-amd64 x86_64

$ gcc --version | head -n 1
gcc (Debian 12.2.0-14) 12.2.0

$ clang --version
Debian clang version 14.0.6
Target: x86_64-pc-linux-gnu
Thread model: posix
InstalledDir: /usr/bin

$ ./Configure -Dcc=gcc -Accflags='-fsanitize=address' -Aldflags='-fsanitize=address' -Dusedevel -des && make test_prep
[# Completed successfully ]
$ make test_harness
...
Test Summary Report
-------------------
re/pat.t                                                           (Wstat: 9 (Signal: KILL) Tests: 276 Failed: 0)
  Non-zero wait status: 9
  Parse errors: Bad plan.  You planned 1295 tests but ran 276.
re/pat_psycho.t                                                    (Wstat: 9 (Signal: KILL) Tests: 1 Failed: 0)
  Non-zero wait status: 9
  Parse errors: Bad plan.  You planned 15 tests but ran 1.
Files=2889, Tests=1184831, 1088 wallclock secs (149.93 usr 40.12 sys + 1170.84 cusr 369.65 csys = 1730.54 CPU)
Result: FAIL

ISTR that at times in the past we've seen failures in t/re/pat.t and t/re/pat_psycho.t under ASAN.

FreeBSD 13

$ uname -mrs
FreeBSD 13.3-RELEASE-p6 amd64

$ gcc --version | head -n 1
gcc (FreeBSD Ports Collection) 13.2.0

$ clang --version
FreeBSD clang version 17.0.6 (https://github.com/llvm/llvm-project.git llvmorg-17.0.6-0-g6009708b4367)
Target: x86_64-unknown-freebsd13.3
Thread model: posix
InstalledDir: /usr/bin

$ ./Configure -Dcc=clang -Accflags='-fsanitize=address' -Aldflags='-fsanitize=address' -Dusedevel -des && make
...
	Everything is up to date. Type 'make test' to run test suite.

$ make test_prep
...
$ make test_harness
...
All tests successful.
Files=2900, Tests=1186258, 823 wallclock secs (264.85 usr 46.89 sys + 2030.75 cusr 373.27 csys = 2715.77 CPU)
Result: PASS

I have a suspicion that this is related to these problems reported to the 'actions' project on GitHub:

actions/runner-images#9491

actions/runner-images#9524

... but the issues discussed there are way outside my area of expertise.

iabyn · 2024-09-17T10:17:50Z

On Mon, Sep 16, 2024 at 01:15:46PM -0700, James E Keenan wrote: On my Ubuntu Linux laptop (where the original complaint) was run, I tried @mauke's invocation and got the same failures as before. Much to my surprise, I got farther on two servers to which I have access.

Note that ASAN requires a *lot* more memory than a normal build. For example on my 16Gb laptop I normally build+test with -j16, but under ASAN I run with -j4 - any more, and I see tests start to fail.

…

-- The optimist believes that he lives in the best of all possible worlds. As does the pessimist.

tonycoz · 2024-09-23T01:04:48Z

I've tried to compile and run the following simple program:

#include <stdio.h>
int main() { printf("Ok\n"); return(0); }

I used the command:

	clang -o try -O2 -fsanitize=address -fno-strict-aliasing -pipe -fstack-protector-strong -I/usr/local/include -fsanitize=address -fstack-protector-strong -L/usr/local/lib try.c -lpthread -ldl -lm -lcrypt -lutil -lc
	 ./try

What do you see if you try copying the code to a file and compiling/running it yourself? Does it differ if you remove all the -l arguments?

If that's failing it seems like you have a toolchain problem, not a perl problem.

jkeenan · 2024-09-23T11:04:06Z

What do you see if you try copying the code to a file and compiling/running it yourself? Does it differ if you remove all the -l arguments?

If that's failing it seems like you have a toolchain problem, not a perl problem.

$ uname -mrs
Linux 6.8.0-45-generic x86_64
$ clang --version
Ubuntu clang version 14.0.0-1ubuntu1.1
Target: x86_64-pc-linux-gnu
Thread model: posix
InstalledDir: /usr/bin

$ cat gh-22588-try.c 
#include <stdio.h>
int main() { printf("Ok\n"); return(0); }

$ clang -o try -O2 -fsanitize=address -fno-strict-aliasing \
-pipe -fstack-protector-strong -I/usr/local/include \
-fsanitize=address -fstack-protector-strong \
-L/usr/local/lib $P5P_DIR/gh-22588-try.c \
-lpthread -ldl -lm -lcrypt -lutil -lc

$ ./try
Ok

$ rm ./try
$ clang -o try -O2 -fsanitize=address -fno-strict-aliasing \
-pipe -fstack-protector-strong -I/usr/local/include \
-fsanitize=address -fstack-protector-strong \
-L/usr/local/lib $P5P_DIR/gh-22588-try.c
$ ./try
Ok

tonycoz · 2024-09-23T23:06:58Z

Exactly which OS are you running this on? Ubuntu something, but which version?

jkeenan · 2024-09-24T01:32:57Z

Exactly which OS are you running this on? Ubuntu something, but which version?

22.04 LTS

$  uname -a
Linux tpadx1 6.8.0-45-generic #45~22.04.1-Ubuntu SMP PREEMPT_DYNAMIC Wed Sep 11 15:25:05 UTC 2 x86_64 x86_64 x86_64 GNU/Linux

I expect to upgrade to 24.04 within 2 months.

tonycoz · 2024-10-23T03:33:44Z

I suspect you have a damaged install, you could try:

sudo dpkg -V

though you'll probably see it complain about many config files.

jkeenan · 2024-10-23T10:54:38Z

I upgraded to Ubuntu 24.04 LTS last week, which brought along upgrades to clang and gcc.

$ uname -mrs
Linux 6.8.0-47-generic x86_64

$ lsb_release -a
...
Description:	Ubuntu 24.04.1 LTS
...

$ gcc --version
gcc (Ubuntu 13.2.0-23ubuntu4) 13.2.0
...

$ clang --version
Ubuntu clang version 18.1.3 (1ubuntu1)
...

I was able to configure and build (thru make test_prep) successfully with these commands:

$ git describe
v5.41.5-17-g161ff7336a

$ ./perl -Ilib -V:config_args
config_args='-des -Dusedevel -Dcc=clang -Accflags=-fsanitize=address -Aldflags=-fsanitize=address';

As expected, make test_harness experienced failures, likely due to memory, on programs like t/re/pat_psycho.t and slowed the machine to a mini-crawl, so I killed the harness.

jkeenan added the Needs Triage label Sep 12, 2024

jkeenan removed the Needs Triage label Sep 16, 2024

tonycoz self-assigned this Oct 22, 2024

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

How do I configure with ASAN? #22588

How do I configure with ASAN? #22588

jkeenan commented Sep 12, 2024 •

edited

Loading

khwilliamson commented Sep 13, 2024 via email

mauke commented Sep 13, 2024

jkeenan commented Sep 16, 2024

iabyn commented Sep 17, 2024 via email

tonycoz commented Sep 23, 2024

jkeenan commented Sep 23, 2024

tonycoz commented Sep 23, 2024

jkeenan commented Sep 24, 2024

tonycoz commented Oct 23, 2024

jkeenan commented Oct 23, 2024

How do I configure with ASAN? #22588

How do I configure with ASAN? #22588

Comments

jkeenan commented Sep 12, 2024 • edited Loading

khwilliamson commented Sep 13, 2024 via email

mauke commented Sep 13, 2024

jkeenan commented Sep 16, 2024

iabyn commented Sep 17, 2024 via email

tonycoz commented Sep 23, 2024

jkeenan commented Sep 23, 2024

tonycoz commented Sep 23, 2024

jkeenan commented Sep 24, 2024

tonycoz commented Oct 23, 2024

jkeenan commented Oct 23, 2024

jkeenan commented Sep 12, 2024 •

edited

Loading