[Bug runtime/24935] New: Reading userland static data always leads to read faults on Fedora 29 x86_64

[Bug runtime/24935] New: Reading userland static data always leads to read faults on Fedora 29 x86_64

[agentzh at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=24935

            Bug ID: 24935
           Summary: Reading userland static data always leads to read
                    faults on Fedora 29 x86_64
           Product: systemtap
           Version: unspecified
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: runtime
          Assignee: systemtap at sourceware dot org
          Reporter: agentzh at gmail dot com
  Target Milestone: ---

I've noted that reading userland C static data always fails with read faults
(-14) on Fedora 29 x86_64. Below is a minimal example that can demonstrate this
issue:

File test.c:

    static const char *s = "hi";

    int main(void) {
        return 0;
    }

File test.stp:

    probe process.function("main") {
        printf("s = %s\n", user_string(@var("s")));
    }

And then compile the test.c C program:

    gcc -g test.c

And then run the test.stp script:

    $ stap test.stp -c './a.out'
    ERROR: user string copy fault -14 at 00000000a25ef291 [man error::fault]
near identifier 'user_string_n' at
/opt/stap/share/systemtap/tapset/uconversions.stp:114:10
    WARNING: Number of errors: 1, skipped probes: 0
    WARNING: /opt/stap/bin/staprun exited with status: 1
    Pass 5: run failed.  [man error::pass5]

I'm using the latest stap master branch (commit 8ffab23ff4):

    $ /opt/stap/bin/stap -V
    Systemtap translator/driver (version 4.2/0.176, commit
release-4.1-72-g8ffab23ff4a1)
    Copyright (C) 2005-2019 Red Hat, Inc. and others
    This is free software; see the source for copying conditions.
    tested kernel versions: 2.6.18 ... 5.1-rc2
    enabled features: AVAHI BPF PYTHON2 LIBSQLITE3 LIBXML2 NLS NSS READLINE

The system is Fedora 29:

    $ cat /etc/redhat-release
    Fedora release 29 (Twenty Nine)


More info:

    $ gcc --version
    gcc (GCC) 8.3.1 20190223 (Red Hat 8.3.1-2)
    Copyright (C) 2018 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.

    $ uname -a
    Linux glass 5.1.20-200.fc29.x86_64 #1 SMP Fri Jul 26 15:15:46 UTC 2019
x86_64 x86_64 x86_64 GNU/Linux

The same test example works flawlessly on CentOS 7:

    $ stap -c './a.out' test.stp
    s = hi

Using gcc options -fpic -pie to compile the test.c program makes no difference:

    ERROR: user string copy fault -14 at 000000004e89a066 [man error::fault]
near identifier 'user_string_n' at
/opt/stap/share/systemtap/tapset/uconversions.stp:114:10
    WARNING: Number of errors: 1, skipped probes: 0
    WARNING: /opt/stap/bin/staprun exited with status: 1
    Pass 5: run failed.  [man error::pass5]

-- 
You are receiving this mail because:
You are the assignee for the bug.

[Bug runtime/24935] Reading userland static data always leads to read faults on Fedora 29 x86_64

[wcohen at redhat dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=24935

William Cohen <wcohen at redhat dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
                 CC|                            |wcohen at redhat dot com
         Resolution|---                         |NOTABUG

--- Comment #1 from William Cohen <wcohen at redhat dot com> ---
One could have the probe fire after the static string has been accessed by
test.c, for example in the process("function").return for a function that
accesses s. However, in this test.c example for this PR nothing is going to
ever going to cause a page fault to pull in that static page that s points at.

Systemtap can only access pages that are already mapped in.  SystemTap can't
trigger a pagefault to map in a page that is not currently mapped in. A similar
situation was observed on a number of the testsuite/systemtap.syscall tests
where a string was stored on a page that was not mapped in on syscall entry and
the syscall would page fault the string in.  The work around in the syscall
tests was to use mlockall(MCL_CURRENT) to ensure that the string in .rodata
section was mapped in.  One example fix of this is git commit
e67e4b19fd73953fbc8a23c91ae22f9d35a7c817:

Author: William Cohen <wcohen@redhat.com>  2023-11-15 17:01:14
Committer: William Cohen <wcohen@redhat.com>  2023-11-15 17:01:14
Parent: 114de05acfdab36295c9c6beee9715e986810d11 (PR29076: Additional syscall
test fixes for .rodata on x86_64)
Child:  130af146398ca5788be92d258eabe1cee662037f (Tweak
testsuite/semok/target_addr.stp to work with linux 5.14 and newer.)
Branches: master, remotes/origin/master, remotes/origin/mcermak-pr30321,
wcohen/nfs
Follows: release-5.0a
Precedes: 

    PR29076: syscall test fixes for .rodata on x86_64 for pwritev.c and sysfs.c

-- 
You are receiving this mail because:
You are the assignee for the bug.