Blame SOURCES/gdb-rhbz1842691-corefile-mem-access-14of15.patch

6543d1
From FEDORA_PATCHES Mon Sep 17 00:00:00 2001
6543d1
From: Keith Seitz <keiths@redhat.com>
6543d1
Date: Tue, 28 Jul 2020 09:46:44 -0400
6543d1
Subject: gdb-rhbz1842691-corefile-mem-access-14of15.patch
6543d1
6543d1
;; New core file tests with mappings over existing program memory
6543d1
;; Kevin Buettner, RH BZ 1842961
6543d1
6543d1
   Author: Kevin Buettner <kevinb@redhat.com>
6543d1
   Date:   Wed Jun 17 19:25:47 2020 -0700
6543d1
6543d1
    New core file tests with mappings over existing program memory
6543d1
6543d1
    This test case was inspired by Pedro's demonstration of a problem
6543d1
    with my v2 patches.  It can be found here:
6543d1
6543d1
        https://sourceware.org/pipermail/gdb-patches/2020-May/168826.html
6543d1
6543d1
    In a nutshell, my earlier patches could not handle the case in
6543d1
    which a read-only mapping created with mmap() was created at
6543d1
    an address used by other file-backed read-only memory in use by
6543d1
    the process.
6543d1
6543d1
    This problem has been fixed (for Linux, anyway) by the commit "Use
6543d1
    NT_FILE note section for reading core target memory".
6543d1
6543d1
    When I run this test without any of my recent corefile patches,
6543d1
    I see these failures:
6543d1
6543d1
    FAIL: gdb.base/corefile2.exp: kernel core: print/x mbuf_ro[0]@4
6543d1
    FAIL: gdb.base/corefile2.exp: kernel core: print/x mbuf_ro[pagesize-4]@4
6543d1
    FAIL: gdb.base/corefile2.exp: kernel core: print/x mbuf_ro[-3]@6
6543d1
    FAIL: gdb.base/corefile2.exp: kernel core: print/x mbuf_rw[pagesize-3]@6
6543d1
    FAIL: gdb.base/corefile2.exp: kernel core: print/x mbuf_ro[pagesize-3]@6
6543d1
    FAIL: gdb.base/corefile2.exp: maint print core-file-backed-mappings
6543d1
    FAIL: gdb.base/corefile2.exp: gcore core: print/x mbuf_ro[-3]@6
6543d1
6543d1
    The ones involving mbuf_ro will almost certainly fail when run on
6543d1
    non-Linux systems; I've used setup_xfail on those tests to prevent
6543d1
    them from outright FAILing when not run on Linux.  For a time, I
6543d1
    had considered skipping these tests altogether when not run on
6543d1
    Linux, but I changed my mind due to this failure...
6543d1
6543d1
    FAIL: gdb.base/corefile2.exp: print/x mbuf_rw[pagesize-3]@6
6543d1
6543d1
    I think it *should* pass without my recent corefile patches.  The fact
6543d1
    that it doesn't is likely due to a bug in GDB.  The following
6543d1
    interaction with GDB demonstrates the problem:
6543d1
6543d1
    (gdb) print/x mbuf_rw[pagesize-3]@6
6543d1
    $1 = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0}
6543d1
    (gdb) print/x mbuf_rw[pagesize]@3
6543d1
    $2 = {0x6b, 0x6b, 0x6b}
6543d1
6543d1
    The last three values in display of $1 should be the same as those
6543d1
    shown by $2.  Like this...
6543d1
6543d1
    (gdb) print/x mbuf_rw[pagesize-3]@6
6543d1
    $1 = {0x0, 0x0, 0x0, 0x6b, 0x6b, 0x6b}
6543d1
    (gdb) print/x mbuf_rw[pagesize]@3
6543d1
    $2 = {0x6b, 0x6b, 0x6b}
6543d1
6543d1
    That latter output was obtained with the use of all of my current
6543d1
    corefile patches.  I see no failures on Linux when running this test
6543d1
    with my current set of corefile patches.  I tested 3 architectures:
6543d1
    x86_64, s390x, and aarch64.
6543d1
6543d1
    I also tested on FreeBSD 12.1-RELEASE.  I see the following results
6543d1
    both with and without the current set of core file patches:
6543d1
6543d1
        # of expected passes		26
6543d1
        # of expected failures		8
6543d1
6543d1
    Of particular interest is that I did *not* see the problematic mbuf_rw
6543d1
    failure noted earlier (both with and without the core file patches).
6543d1
    I still don't have an explanation for why this failure occurred on
6543d1
    Linux.  Prior to running the tests, I had hypothesized that I'd see
6543d1
    this failure on FreeBSD too, but testing shows that this is not the
6543d1
    case.
6543d1
6543d1
    Also of importance is that we see no FAILs with this test on FreeBSD
6543d1
    which indicates that I XFAILed the correct tests.
6543d1
6543d1
    This version runs the interesting tests twice, once with a kernel
6543d1
    created core file and another time with a gcore created core file.
6543d1
6543d1
    It also does a very minimal test of the new command "maint print
6543d1
    core-file-backed-mappings".
6543d1
6543d1
    gdb/testsuite/ChangeLog:
6543d1
6543d1
    	* gdb.base/corefile2.exp: New file.
6543d1
    	* gdb.base/coremaker2.exp: New file.
6543d1
6543d1
diff --git a/gdb/testsuite/gdb.base/corefile2.exp b/gdb/testsuite/gdb.base/corefile2.exp
6543d1
new file mode 100644
6543d1
--- /dev/null
6543d1
+++ b/gdb/testsuite/gdb.base/corefile2.exp
6543d1
@@ -0,0 +1,185 @@
6543d1
+# Copyright 2020 Free Software Foundation, Inc.
6543d1
+
6543d1
+# This program is free software; you can redistribute it and/or modify
6543d1
+# it under the terms of the GNU General Public License as published by
6543d1
+# the Free Software Foundation; either version 3 of the License, or
6543d1
+# (at your option) any later version.
6543d1
+#
6543d1
+# This program is distributed in the hope that it will be useful,
6543d1
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
6543d1
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
6543d1
+# GNU General Public License for more details.
6543d1
+#
6543d1
+# You should have received a copy of the GNU General Public License
6543d1
+# along with this program.  If not, see <http://www.gnu.org/licenses/>.
6543d1
+
6543d1
+# Tests of core file memory accesses when mmap() has been used to
6543d1
+# create a "hole" of zeroes over pre-existing memory regions.  See
6543d1
+# coremaker2.c for details.
6543d1
+
6543d1
+# are we on a target board
6543d1
+if ![isnative] then {
6543d1
+    return
6543d1
+}
6543d1
+
6543d1
+# Some of these tests will only work on GNU/Linux due to the
6543d1
+# fact that Linux core files includes a section describing
6543d1
+# memory address to file mappings.  We'll use set_up_xfail for the
6543d1
+# affected tests.  As other targets become supported, the condition
6543d1
+# can be changed accordingly.
6543d1
+
6543d1
+set xfail 0
6543d1
+if { ![istarget *-linux*] } {
6543d1
+    set xfail 1
6543d1
+}
6543d1
+
6543d1
+standard_testfile coremaker2.c
6543d1
+
6543d1
+if {[build_executable $testfile.exp $testfile $srcfile debug] == -1} {
6543d1
+    untested "failed to compile"
6543d1
+    return -1
6543d1
+}
6543d1
+
6543d1
+set corefile [core_find $binfile {}]
6543d1
+if {$corefile == ""} {
6543d1
+    return 0
6543d1
+}
6543d1
+
6543d1
+gdb_start
6543d1
+gdb_reinitialize_dir $srcdir/$subdir
6543d1
+gdb_load ${binfile}
6543d1
+
6543d1
+# Attempt to load the core file.
6543d1
+
6543d1
+gdb_test_multiple "core-file $corefile" "core-file command" {
6543d1
+    -re ".* program is being debugged already.*y or n. $" {
6543d1
+	# gdb_load may connect us to a gdbserver.
6543d1
+	send_gdb "y\n"
6543d1
+	exp_continue
6543d1
+    }
6543d1
+    -re "Core was generated by .*corefile.*\r\n\#0  .*\(\).*\r\n$gdb_prompt $" {
6543d1
+	pass "core-file command"
6543d1
+    }
6543d1
+    -re "Core was generated by .*\r\n\#0  .*\(\).*\r\n$gdb_prompt $" {
6543d1
+	pass "core-file command (with bad program name)"
6543d1
+    }
6543d1
+    -re ".*registers from core file: File in wrong format.* $" {
6543d1
+	fail "core-file command (could not read registers from core file)"
6543d1
+    }
6543d1
+}
6543d1
+
6543d1
+# Perform the "interesting" tests which check the contents of certain
6543d1
+# memory regions.
6543d1
+
6543d1
+proc do_tests { } {
6543d1
+    global xfail
6543d1
+
6543d1
+    # Check contents of beginning of buf_rw and buf_ro.
6543d1
+
6543d1
+    gdb_test {print/x buf_rw[0]@4} {\{0x6b, 0x6b, 0x6b, 0x6b\}}
6543d1
+    gdb_test {print/x buf_ro[0]@4} {\{0xc5, 0xc5, 0xc5, 0xc5\}}
6543d1
+
6543d1
+    # Check for correct contents at beginning of mbuf_rw and mbuf_ro.
6543d1
+
6543d1
+    gdb_test {print/x mbuf_rw[0]@4} {\{0x0, 0x0, 0x0, 0x0\}}
6543d1
+
6543d1
+    if { $xfail } { setup_xfail "*-*-*" }
6543d1
+    gdb_test {print/x mbuf_ro[0]@4} {\{0x0, 0x0, 0x0, 0x0\}}
6543d1
+
6543d1
+    # Check contents of mbuf_rw and mbuf_ro at the end of these regions.
6543d1
+
6543d1
+    gdb_test {print/x mbuf_rw[pagesize-4]@4} {\{0x0, 0x0, 0x0, 0x0\}}
6543d1
+
6543d1
+    if { $xfail } { setup_xfail "*-*-*" }
6543d1
+    gdb_test {print/x mbuf_ro[pagesize-4]@4} {\{0x0, 0x0, 0x0, 0x0\}}
6543d1
+
6543d1
+    # Check contents of mbuf_rw and mbuf_ro, right before the hole,
6543d1
+    # overlapping into the beginning of these mmap'd regions.
6543d1
+
6543d1
+    gdb_test {print/x mbuf_rw[-3]@6} {\{0x6b, 0x6b, 0x6b, 0x0, 0x0, 0x0\}}
6543d1
+
6543d1
+    if { $xfail } { setup_xfail "*-*-*" }
6543d1
+    gdb_test {print/x mbuf_ro[-3]@6} {\{0xc5, 0xc5, 0xc5, 0x0, 0x0, 0x0\}}
6543d1
+
6543d1
+    # Likewise, at the end of the mbuf_rw and mbuf_ro, with overlap.
6543d1
+
6543d1
+    # If this test FAILs, it's probably a genuine bug unrelated to whether
6543d1
+    # the core file includes a section describing memory address to file
6543d1
+    # mappings or not.  (So don't xfail it!)
6543d1
+    gdb_test {print/x mbuf_rw[pagesize-3]@6} {\{0x0, 0x0, 0x0, 0x6b, 0x6b, 0x6b\}}
6543d1
+
6543d1
+    if { $xfail } { setup_xfail "*-*-*" }
6543d1
+    gdb_test {print/x mbuf_ro[pagesize-3]@6} {\{0x0, 0x0, 0x0, 0xc5, 0xc5, 0xc5\}}
6543d1
+
6543d1
+    # Check contents of (what should be) buf_rw and buf_ro immediately after
6543d1
+    # mbuf_rw and mbuf_ro holes.
6543d1
+
6543d1
+    gdb_test {print/x mbuf_rw[pagesize]@4} {\{0x6b, 0x6b, 0x6b, 0x6b\}}
6543d1
+    gdb_test {print/x mbuf_ro[pagesize]@4} {\{0xc5, 0xc5, 0xc5, 0xc5\}}
6543d1
+
6543d1
+    # Check contents at ends of buf_rw and buf_rw.
6543d1
+
6543d1
+    gdb_test {print/x buf_rw[sizeof(buf_rw)-4]@4} {\{0x6b, 0x6b, 0x6b, 0x6b\}}
6543d1
+    gdb_test {print/x buf_ro[sizeof(buf_ro)-4]@4} {\{0xc5, 0xc5, 0xc5, 0xc5\}}
6543d1
+}
6543d1
+
6543d1
+# Run tests with kernel-produced core file.
6543d1
+
6543d1
+with_test_prefix "kernel core" {
6543d1
+    do_tests
6543d1
+}
6543d1
+
6543d1
+# Verify that "maint print core-file-backed-mappings" exists and does
6543d1
+# not crash GDB.  If it produces any output at all, make sure that
6543d1
+# that output at least mentions binfile.
6543d1
+
6543d1
+set test "maint print core-file-backed-mappings"
6543d1
+gdb_test_multiple $test "" {
6543d1
+    -re ".*$binfile.*$gdb_prompt $" {
6543d1
+	pass $test
6543d1
+    }
6543d1
+    -re "^$test\[\r\n\]*$gdb_prompt $" {
6543d1
+	pass "$test (no output)"
6543d1
+    }
6543d1
+}
6543d1
+
6543d1
+# Restart and run to the abort call.
6543d1
+
6543d1
+clean_restart $binfile
6543d1
+
6543d1
+if ![runto_main] then {
6543d1
+    fail "can't run to main"
6543d1
+    return
6543d1
+}
6543d1
+
6543d1
+gdb_breakpoint [gdb_get_line_number "abort"]
6543d1
+gdb_continue_to_breakpoint "at abort"
6543d1
+
6543d1
+# Do not execute abort call; instead, invoke gcore command to make a
6543d1
+# gdb-produced core file.
6543d1
+
6543d1
+set corefile [standard_output_file gcore.test]
6543d1
+set core_supported [gdb_gcore_cmd "$corefile" "save a corefile"]
6543d1
+if {!$core_supported} {
6543d1
+  return
6543d1
+}
6543d1
+
6543d1
+# maint print-core-file-backed-mappings shouldn't produce any output
6543d1
+# when not debugging a core file.
6543d1
+
6543d1
+gdb_test_no_output "maint print core-file-backed-mappings" \
6543d1
+    "maint print core-file-backed-mapping with no core file"
6543d1
+
6543d1
+clean_restart $binfile
6543d1
+
6543d1
+set core_loaded [gdb_core_cmd "$corefile" "re-load generated corefile"]
6543d1
+if { $core_loaded == -1 } {
6543d1
+    # No use proceeding from here.
6543d1
+    return
6543d1
+}
6543d1
+
6543d1
+# Run tests using gcore-produced core file.
6543d1
+
6543d1
+with_test_prefix "gcore core" {
6543d1
+    do_tests
6543d1
+}
6543d1
diff --git a/gdb/testsuite/gdb.base/coremaker2.c b/gdb/testsuite/gdb.base/coremaker2.c
6543d1
new file mode 100644
6543d1
--- /dev/null
6543d1
+++ b/gdb/testsuite/gdb.base/coremaker2.c
6543d1
@@ -0,0 +1,150 @@
6543d1
+/* Copyright 1992-2020 Free Software Foundation, Inc.
6543d1
+
6543d1
+   This file is part of GDB.
6543d1
+
6543d1
+   This program is free software; you can redistribute it and/or modify
6543d1
+   it under the terms of the GNU General Public License as published by
6543d1
+   the Free Software Foundation; either version 3 of the License, or
6543d1
+   (at your option) any later version.
6543d1
+
6543d1
+   This program is distributed in the hope that it will be useful,
6543d1
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
6543d1
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
6543d1
+   GNU General Public License for more details.
6543d1
+
6543d1
+   You should have received a copy of the GNU General Public License
6543d1
+   along with this program.  If not, see <http://www.gnu.org/licenses/>.  */
6543d1
+
6543d1
+/*  This test has two large memory areas buf_rw and buf_ro. 
6543d1
+
6543d1
+    buf_rw is written to by the program while buf_ro is initialized at
6543d1
+    compile / load time.  Thus, when a core file is created, buf_rw's
6543d1
+    memory should reside in the core file, but buf_ro probably won't be.
6543d1
+    Instead, the contents of buf_ro are available from the executable.
6543d1
+
6543d1
+    Now, for the wrinkle:  We create a one page read-only mapping over
6543d1
+    both of these areas.  This will create a one page "hole" of all
6543d1
+    zeros in each area.
6543d1
+
6543d1
+    Will GDB be able to correctly read memory from each of the four
6543d1
+    (or six, if you count the regions on the other side of each hole)
6543d1
+    memory regions?  */
6543d1
+
6543d1
+#include <stdio.h>
6543d1
+#include <sys/types.h>
6543d1
+#include <fcntl.h>
6543d1
+#include <sys/mman.h>
6543d1
+#include <signal.h>
6543d1
+#include <stdlib.h>
6543d1
+#include <unistd.h>
6543d1
+#include <string.h>
6543d1
+#include <errno.h>
6543d1
+
6543d1
+/* These are globals so that we can find them easily when debugging
6543d1
+   the core file.  */
6543d1
+long pagesize;
6543d1
+unsigned long long addr;
6543d1
+char *mbuf_ro;
6543d1
+char *mbuf_rw;
6543d1
+
6543d1
+/* 24 KiB buffer.  */
6543d1
+char buf_rw[24 * 1024];
6543d1
+
6543d1
+/* 24 KiB worth of data.  For this test case, we can't allocate a
6543d1
+   buffer and then fill it; we want GDB to have to read this data
6543d1
+   from the executable; it should NOT find it in the core file.  */
6543d1
+
6543d1
+#define C5_16 \
6543d1
+  0xc5, 0xc5, 0xc5, 0xc5, \
6543d1
+  0xc5, 0xc5, 0xc5, 0xc5, \
6543d1
+  0xc5, 0xc5, 0xc5, 0xc5, \
6543d1
+  0xc5, 0xc5, 0xc5, 0xc5
6543d1
+
6543d1
+#define C5_256 \
6543d1
+  C5_16, C5_16, C5_16, C5_16, \
6543d1
+  C5_16, C5_16, C5_16, C5_16, \
6543d1
+  C5_16, C5_16, C5_16, C5_16, \
6543d1
+  C5_16, C5_16, C5_16, C5_16
6543d1
+
6543d1
+#define C5_1k \
6543d1
+  C5_256, C5_256, C5_256, C5_256
6543d1
+
6543d1
+#define C5_24k \
6543d1
+  C5_1k, C5_1k, C5_1k, C5_1k, \
6543d1
+  C5_1k, C5_1k, C5_1k, C5_1k, \
6543d1
+  C5_1k, C5_1k, C5_1k, C5_1k, \
6543d1
+  C5_1k, C5_1k, C5_1k, C5_1k, \
6543d1
+  C5_1k, C5_1k, C5_1k, C5_1k, \
6543d1
+  C5_1k, C5_1k, C5_1k, C5_1k
6543d1
+
6543d1
+const char buf_ro[] = { C5_24k };
6543d1
+
6543d1
+int
6543d1
+main (int argc, char **argv)
6543d1
+{
6543d1
+  int i, bitcount;
6543d1
+
6543d1
+#ifdef _SC_PAGESIZE
6543d1
+  pagesize = sysconf (_SC_PAGESIZE);
6543d1
+#else
6543d1
+  pagesize = 8192;
6543d1
+#endif
6543d1
+
6543d1
+  /* Verify that pagesize is a power of 2.  */
6543d1
+  bitcount = 0;
6543d1
+  for (i = 0; i < 4 * sizeof (pagesize); i++)
6543d1
+    if (pagesize & (1 << i))
6543d1
+      bitcount++;
6543d1
+
6543d1
+  if (bitcount != 1)
6543d1
+    {
6543d1
+      fprintf (stderr, "pagesize is not a power of 2.\n");
6543d1
+      exit (1);
6543d1
+    }
6543d1
+
6543d1
+  /* Compute an address that should be within buf_ro.  Complain if not.  */
6543d1
+  addr = ((unsigned long long) buf_ro + pagesize) & ~(pagesize - 1);
6543d1
+
6543d1
+  if (addr <= (unsigned long long) buf_ro
6543d1
+      || addr >= (unsigned long long) buf_ro + sizeof (buf_ro))
6543d1
+    {
6543d1
+      fprintf (stderr, "Unable to compute a suitable address within buf_ro.\n");
6543d1
+      exit (1);
6543d1
+    }
6543d1
+
6543d1
+  mbuf_ro = mmap ((void *) addr, pagesize, PROT_READ,
6543d1
+               MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0);
6543d1
+
6543d1
+  if (mbuf_ro == MAP_FAILED)
6543d1
+    {
6543d1
+      fprintf (stderr, "mmap #1 failed: %s.\n", strerror (errno));
6543d1
+      exit (1);
6543d1
+    }
6543d1
+
6543d1
+  /* Write (and fill) the R/W region.  */
6543d1
+  for (i = 0; i < sizeof (buf_rw); i++)
6543d1
+    buf_rw[i] = 0x6b;
6543d1
+
6543d1
+  /* Compute an mmap address within buf_rw.  Complain if it's somewhere
6543d1
+     else.  */
6543d1
+  addr = ((unsigned long long) buf_rw + pagesize) & ~(pagesize - 1);
6543d1
+
6543d1
+  if (addr <= (unsigned long long) buf_rw
6543d1
+      || addr >= (unsigned long long) buf_rw + sizeof (buf_rw))
6543d1
+    {
6543d1
+      fprintf (stderr, "Unable to compute a suitable address within buf_rw.\n");
6543d1
+      exit (1);
6543d1
+    }
6543d1
+
6543d1
+  mbuf_rw = mmap ((void *) addr, pagesize, PROT_READ,
6543d1
+               MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0);
6543d1
+
6543d1
+  if (mbuf_rw == MAP_FAILED)
6543d1
+    {
6543d1
+      fprintf (stderr, "mmap #2 failed: %s.\n", strerror (errno));
6543d1
+      exit (1);
6543d1
+    }
6543d1
+
6543d1
+  /* With correct ulimit, etc. this should cause a core dump.  */
6543d1
+  abort ();
6543d1
+}