binutils-gdb/gdb/testsuite/gdb.base/catch-syscall.exp

623 lines
19 KiB
Plaintext
Raw Normal View History

# Copyright 1997-2016 Free Software Foundation, Inc.
# This program is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation; either version 3 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program. If not, see <http://www.gnu.org/licenses/>.
# This program tests the 'catch syscall' functionality.
#
# It was written by Sergio Durigan Junior <sergiodj@linux.vnet.ibm.com>
# on September/2008.
if { [is_remote target] || ![isnative] } then {
continue
}
# This shall be updated whenever 'catch syscall' is implemented
# on some architecture.
if { ![istarget "x86_64-*-linux*"] && ![istarget "i\[34567\]86-*-linux*"]
&& ![istarget "powerpc-*-linux*"] && ![istarget "powerpc64-*-linux*"]
&& ![istarget "sparc-*-linux*"] && ![istarget "sparc64-*-linux*"]
&& ![istarget "mips*-linux*"] && ![istarget "arm*-linux*"]
Support catch syscall on aarch64 linux Hi, This patch is to support catch syscall on aarch64 linux. We implement gdbarch method get_syscall_number for aarch64-linux, and add aarch64-linux.xml file, which looks straightforward, however the changes to test case doesn't. First of all, we enable catch-syscall.exp on aarch64-linux target, but skip the multi_arch testing on current stage. I plan to touch multi arch debugging on aarch64-linux later. Then, when I run catch-syscall.exp on aarch64-linux, gcc errors that SYS_pipe isn't defined. We find that aarch64 kernel only has pipe2 syscall and libc already convert pipe to pipe2. As a result, I change catch-syscall.c to use SYS_pipe if it is defined, otherwise use SYS_pipe2 instead. The vector all_syscalls in catch-syscall.exp can't be pre-determined, so I add a new proc setup_all_syscalls to fill it, according to the availability of SYS_pipe. Regression tested on {x86_64, aarch64}-linux x {native, gdbserver}. gdb: 2015-03-18 Yao Qi <yao.qi@linaro.org> PR tdep/18107 * aarch64-linux-tdep.c: Include xml-syscall.h (aarch64_linux_get_syscall_number): New function. (aarch64_linux_init_abi): Call set_gdbarch_get_syscall_number. * syscalls/aarch64-linux.xml: New file. gdb/testsuite: 2015-03-18 Yao Qi <yao.qi@linaro.org> PR tdep/18107 * gdb.base/catch-syscall.c [!SYS_pipe] (pipe2_syscall): New variable. * gdb.base/catch-syscall.exp: Don't skip it on aarch64*-*-linux* target. Remove elements in all_syscalls. (test_catch_syscall_multi_arch): Skip it on aarch64*-linux* target. (setup_all_syscalls): New proc.
2015-03-18 11:47:45 +01:00
&& ![istarget "s390*-linux*"] && ![istarget "aarch64*-*-linux*"] } {
continue
}
standard_testfile
Improve and fix catch-syscall.exp While fixing another bug, I found that the current gdb.base/catch-syscall.exp is kind of messy, could use some improvements, and is not correctly testing some things. I've made the following patch to address all the issues I found. On the organization side, it does a cleanup and removes unecessary imports of gdb_prompt, uses prepare_for_testing and clean_restart where needed, and fixes some comments. The testcase was also not correctly testing catching syscalls using only numbers, or catching many syscalls at once. I fixed that. The patch also uses a new method for obtaining the syscalls numbers: it relies on the C source file to get them, via <sys/syscall.h> and SYS_* macros. This makes the .exp file simpler because there is no need to include target conditionals there. I tested this on x86_64 Fedora 18. gdb/testsuite/ChangeLog: 2013-12-18 Sergio Durigan Junior <sergiodj@redhat.com> * gdb.base/catch-syscall.c: Include <sys/syscall.h>. (close_syscall, chroot_syscall, exit_group_syscall): New variables. * gdb.base/catch-syscall.exp: Replace gdb_compile by prepare_for_testing. Call fill_all_syscalls_numbers before starting. Replace gdb_exit, gdb_start, gdb_reinitialize_dir and gdb_load by clean_restart. (check_info_bp_any_syscall, check_info_bp_specific_syscall) (check_info_bp_many_syscalls): Remove global gdb_prompt. (check_call_to_syscall): Likewise. Add global decimal. Improve testing regex. (check_return_from_syscall): Likewise. (check_continue, insert_catch_syscall_with_arg): Remove global gdb_prompt. (insert_catch_syscall_with_many_args): Likewise. Add global decimal. Fix $filter_str. Improve testing regex. (check_for_program_end): Remove global gdb_prompt. (test_catch_syscall_without_args): Likewise. Add global decimal. Improve testing regex. (test_catch_syscall_with_args, test_catch_syscall_with_many_args) (test_catch_syscall_with_wrong_args) (test_catch_syscall_restarting_inferior) (test_catch_syscall_fail_nodatadir): Remove global gdb_prompt. (do_syscall_tests): Likewise. Remove global srcdir. (test_catch_syscall_without_args_noxml): Remove global gdb_prompt. Add global last_syscall_number. Test for the exact syscall number to be caught. (test_catch_syscall_with_args_noxml): Remove global gdb_prompt. Add global all_syscalls_numbers. Test each syscall number to be caught, instead of only testing "close". (test_catch_syscall_with_wrong_args_noxml): Remove global gdb_prompt. (do_syscall_tests_without_xml): Likewise. Remove global srcdir. Remove stale comment. (fill_all_syscalls_numbers): Add global last_syscall_number. Fill the correct syscall numbers using information from the inferior.
2013-12-18 23:19:01 +01:00
if { [prepare_for_testing ${testfile}.exp $testfile ${testfile}.c] } {
untested catch-syscall.exp
return -1
}
Support catch syscall on aarch64 linux Hi, This patch is to support catch syscall on aarch64 linux. We implement gdbarch method get_syscall_number for aarch64-linux, and add aarch64-linux.xml file, which looks straightforward, however the changes to test case doesn't. First of all, we enable catch-syscall.exp on aarch64-linux target, but skip the multi_arch testing on current stage. I plan to touch multi arch debugging on aarch64-linux later. Then, when I run catch-syscall.exp on aarch64-linux, gcc errors that SYS_pipe isn't defined. We find that aarch64 kernel only has pipe2 syscall and libc already convert pipe to pipe2. As a result, I change catch-syscall.c to use SYS_pipe if it is defined, otherwise use SYS_pipe2 instead. The vector all_syscalls in catch-syscall.exp can't be pre-determined, so I add a new proc setup_all_syscalls to fill it, according to the availability of SYS_pipe. Regression tested on {x86_64, aarch64}-linux x {native, gdbserver}. gdb: 2015-03-18 Yao Qi <yao.qi@linaro.org> PR tdep/18107 * aarch64-linux-tdep.c: Include xml-syscall.h (aarch64_linux_get_syscall_number): New function. (aarch64_linux_init_abi): Call set_gdbarch_get_syscall_number. * syscalls/aarch64-linux.xml: New file. gdb/testsuite: 2015-03-18 Yao Qi <yao.qi@linaro.org> PR tdep/18107 * gdb.base/catch-syscall.c [!SYS_pipe] (pipe2_syscall): New variable. * gdb.base/catch-syscall.exp: Don't skip it on aarch64*-*-linux* target. Remove elements in all_syscalls. (test_catch_syscall_multi_arch): Skip it on aarch64*-linux* target. (setup_all_syscalls): New proc.
2015-03-18 11:47:45 +01:00
# All (but the last) syscalls from the example code. It is filled in
# proc setup_all_syscalls.
set all_syscalls { }
set all_syscalls_numbers { }
Improve and fix catch-syscall.exp While fixing another bug, I found that the current gdb.base/catch-syscall.exp is kind of messy, could use some improvements, and is not correctly testing some things. I've made the following patch to address all the issues I found. On the organization side, it does a cleanup and removes unecessary imports of gdb_prompt, uses prepare_for_testing and clean_restart where needed, and fixes some comments. The testcase was also not correctly testing catching syscalls using only numbers, or catching many syscalls at once. I fixed that. The patch also uses a new method for obtaining the syscalls numbers: it relies on the C source file to get them, via <sys/syscall.h> and SYS_* macros. This makes the .exp file simpler because there is no need to include target conditionals there. I tested this on x86_64 Fedora 18. gdb/testsuite/ChangeLog: 2013-12-18 Sergio Durigan Junior <sergiodj@redhat.com> * gdb.base/catch-syscall.c: Include <sys/syscall.h>. (close_syscall, chroot_syscall, exit_group_syscall): New variables. * gdb.base/catch-syscall.exp: Replace gdb_compile by prepare_for_testing. Call fill_all_syscalls_numbers before starting. Replace gdb_exit, gdb_start, gdb_reinitialize_dir and gdb_load by clean_restart. (check_info_bp_any_syscall, check_info_bp_specific_syscall) (check_info_bp_many_syscalls): Remove global gdb_prompt. (check_call_to_syscall): Likewise. Add global decimal. Improve testing regex. (check_return_from_syscall): Likewise. (check_continue, insert_catch_syscall_with_arg): Remove global gdb_prompt. (insert_catch_syscall_with_many_args): Likewise. Add global decimal. Fix $filter_str. Improve testing regex. (check_for_program_end): Remove global gdb_prompt. (test_catch_syscall_without_args): Likewise. Add global decimal. Improve testing regex. (test_catch_syscall_with_args, test_catch_syscall_with_many_args) (test_catch_syscall_with_wrong_args) (test_catch_syscall_restarting_inferior) (test_catch_syscall_fail_nodatadir): Remove global gdb_prompt. (do_syscall_tests): Likewise. Remove global srcdir. (test_catch_syscall_without_args_noxml): Remove global gdb_prompt. Add global last_syscall_number. Test for the exact syscall number to be caught. (test_catch_syscall_with_args_noxml): Remove global gdb_prompt. Add global all_syscalls_numbers. Test each syscall number to be caught, instead of only testing "close". (test_catch_syscall_with_wrong_args_noxml): Remove global gdb_prompt. (do_syscall_tests_without_xml): Likewise. Remove global srcdir. Remove stale comment. (fill_all_syscalls_numbers): Add global last_syscall_number. Fill the correct syscall numbers using information from the inferior.
2013-12-18 23:19:01 +01:00
# The last syscall (exit()) does not return, so
# we cannot expect the catchpoint to be triggered
# twice. It is a special case.
set last_syscall "exit_group"
Improve and fix catch-syscall.exp While fixing another bug, I found that the current gdb.base/catch-syscall.exp is kind of messy, could use some improvements, and is not correctly testing some things. I've made the following patch to address all the issues I found. On the organization side, it does a cleanup and removes unecessary imports of gdb_prompt, uses prepare_for_testing and clean_restart where needed, and fixes some comments. The testcase was also not correctly testing catching syscalls using only numbers, or catching many syscalls at once. I fixed that. The patch also uses a new method for obtaining the syscalls numbers: it relies on the C source file to get them, via <sys/syscall.h> and SYS_* macros. This makes the .exp file simpler because there is no need to include target conditionals there. I tested this on x86_64 Fedora 18. gdb/testsuite/ChangeLog: 2013-12-18 Sergio Durigan Junior <sergiodj@redhat.com> * gdb.base/catch-syscall.c: Include <sys/syscall.h>. (close_syscall, chroot_syscall, exit_group_syscall): New variables. * gdb.base/catch-syscall.exp: Replace gdb_compile by prepare_for_testing. Call fill_all_syscalls_numbers before starting. Replace gdb_exit, gdb_start, gdb_reinitialize_dir and gdb_load by clean_restart. (check_info_bp_any_syscall, check_info_bp_specific_syscall) (check_info_bp_many_syscalls): Remove global gdb_prompt. (check_call_to_syscall): Likewise. Add global decimal. Improve testing regex. (check_return_from_syscall): Likewise. (check_continue, insert_catch_syscall_with_arg): Remove global gdb_prompt. (insert_catch_syscall_with_many_args): Likewise. Add global decimal. Fix $filter_str. Improve testing regex. (check_for_program_end): Remove global gdb_prompt. (test_catch_syscall_without_args): Likewise. Add global decimal. Improve testing regex. (test_catch_syscall_with_args, test_catch_syscall_with_many_args) (test_catch_syscall_with_wrong_args) (test_catch_syscall_restarting_inferior) (test_catch_syscall_fail_nodatadir): Remove global gdb_prompt. (do_syscall_tests): Likewise. Remove global srcdir. (test_catch_syscall_without_args_noxml): Remove global gdb_prompt. Add global last_syscall_number. Test for the exact syscall number to be caught. (test_catch_syscall_with_args_noxml): Remove global gdb_prompt. Add global all_syscalls_numbers. Test each syscall number to be caught, instead of only testing "close". (test_catch_syscall_with_wrong_args_noxml): Remove global gdb_prompt. (do_syscall_tests_without_xml): Likewise. Remove global srcdir. Remove stale comment. (fill_all_syscalls_numbers): Add global last_syscall_number. Fill the correct syscall numbers using information from the inferior.
2013-12-18 23:19:01 +01:00
set last_syscall_number { }
gdb: Improve syscall entry/return tracking on Linux The existing logic was simply to flip syscall entry/return state when a syscall trap was seen, and even then only with active 'catch syscall'. That can get out of sync if 'catch syscall' is toggled at odd times. This patch updates the entry/return state for all syscall traps, regardless of catching state, and also updates known syscall state for other kinds of traps. Almost all PTRACE_EVENT stops are delivered from the middle of a syscall, so this can act like an entry. Every other kind of ptrace stop is only delivered outside of syscall event pairs, so marking them ignored ensures the next syscall trap looks like an entry. Three new test scenarios are added to catch-syscall.exp: - Disable 'catch syscall' from an entry to deliberately miss the return event, then re-enable to make sure a new entry is recognized. - Enable 'catch syscall' for the first time from a vfork event, which is a PTRACE_EVENT_VFORK in the middle of the syscall. Make sure the next syscall event is recognized as the return. - Make sure entry and return are recognized for an ENOSYS syscall. This is to defeat a common x86 hack that uses the pre-filled ENOSYS return value as a sign of being on the entry side. gdb/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * linux-nat.c (linux_handle_syscall_trap): Always update entry/ return state, even when not actively catching syscalls at all. (linux_handle_extended_wait): Mark syscall_state like an entry. (wait_lwp): Set syscall_state ignored for other traps. (linux_nat_filter_event): Likewise. gdb/testsuite/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * gdb.base/catch-syscall.c: Include <sched.h>. (unknown_syscall): New variable. (main): Trigger a vfork and an unknown syscall. * gdb.base/catch-syscall.exp (vfork_syscalls): New variable. (unknown_syscall_number): Likewise. (check_call_to_syscall): Accept an optional syscall pattern. (check_return_from_syscall): Likewise. (check_continue): Likewise. (test_catch_syscall_without_args): Check for vfork and ENOSYS. (test_catch_syscall_skipping_return): New test toggling off 'catch syscall' to step over the syscall return, then toggling back on. (test_catch_syscall_mid_vfork): New test turning on 'catch syscall' during a PTRACE_EVENT_VFORK stop, in the middle of a vfork syscall. (do_syscall_tests): Call test_catch_syscall_without_args and test_catch_syscall_mid_vfork. (test_catch_syscall_without_args_noxml): Check for vfork and ENOSYS. (fill_all_syscalls_numbers): Initialize unknown_syscall_number.
2015-10-20 02:59:38 +02:00
set vfork_syscalls "(vfork|clone2?)"
set unknown_syscall_number { }
# Internal procedure used to check if, after issuing a 'catch syscall'
# command (without arguments), the 'info breakpoints' command displays
# that '"any syscall"' is to be caught.
proc check_info_bp_any_syscall {} {
# Verifying that the catchpoint appears in the 'info breakpoints'
# command, but with "<any syscall>".
set thistest "catch syscall appears in 'info breakpoints'"
gdb_test "info breakpoints" ".*catchpoint.*keep y.*syscall \"<any syscall>\".*" $thistest
}
# Internal procedure used to check if, after issuing a 'catch syscall X'
# command (with arguments), the 'info breakpoints' command displays
# that the syscall 'X' is to be caught.
proc check_info_bp_specific_syscall { syscall } {
set thistest "syscall(s) $syscall appears in 'info breakpoints'"
gdb_test "info breakpoints" ".*catchpoint.*keep y.*syscall(\[(\]s\[)\])? (.)?${syscall}(.)?.*" $thistest
}
# Internal procedure used to check if, after issuing a 'catch syscall X'
# command (with many arguments), the 'info breakpoints' command displays
# that the syscalls 'X' are to be caught.
proc check_info_bp_many_syscalls { syscalls } {
set filter_str ""
foreach name $syscalls {
set filter_str "${filter_str}${name}, "
}
set filter_str [ string trimright $filter_str ", " ]
set thistest "syscalls $filter_str appears in 'info breakpoints'"
gdb_test "info breakpoints" ".*catchpoint.*keep y.*syscalls (.)?${filter_str}(.)?.*" $thistest
}
gdb: Improve syscall entry/return tracking on Linux The existing logic was simply to flip syscall entry/return state when a syscall trap was seen, and even then only with active 'catch syscall'. That can get out of sync if 'catch syscall' is toggled at odd times. This patch updates the entry/return state for all syscall traps, regardless of catching state, and also updates known syscall state for other kinds of traps. Almost all PTRACE_EVENT stops are delivered from the middle of a syscall, so this can act like an entry. Every other kind of ptrace stop is only delivered outside of syscall event pairs, so marking them ignored ensures the next syscall trap looks like an entry. Three new test scenarios are added to catch-syscall.exp: - Disable 'catch syscall' from an entry to deliberately miss the return event, then re-enable to make sure a new entry is recognized. - Enable 'catch syscall' for the first time from a vfork event, which is a PTRACE_EVENT_VFORK in the middle of the syscall. Make sure the next syscall event is recognized as the return. - Make sure entry and return are recognized for an ENOSYS syscall. This is to defeat a common x86 hack that uses the pre-filled ENOSYS return value as a sign of being on the entry side. gdb/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * linux-nat.c (linux_handle_syscall_trap): Always update entry/ return state, even when not actively catching syscalls at all. (linux_handle_extended_wait): Mark syscall_state like an entry. (wait_lwp): Set syscall_state ignored for other traps. (linux_nat_filter_event): Likewise. gdb/testsuite/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * gdb.base/catch-syscall.c: Include <sched.h>. (unknown_syscall): New variable. (main): Trigger a vfork and an unknown syscall. * gdb.base/catch-syscall.exp (vfork_syscalls): New variable. (unknown_syscall_number): Likewise. (check_call_to_syscall): Accept an optional syscall pattern. (check_return_from_syscall): Likewise. (check_continue): Likewise. (test_catch_syscall_without_args): Check for vfork and ENOSYS. (test_catch_syscall_skipping_return): New test toggling off 'catch syscall' to step over the syscall return, then toggling back on. (test_catch_syscall_mid_vfork): New test turning on 'catch syscall' during a PTRACE_EVENT_VFORK stop, in the middle of a vfork syscall. (do_syscall_tests): Call test_catch_syscall_without_args and test_catch_syscall_mid_vfork. (test_catch_syscall_without_args_noxml): Check for vfork and ENOSYS. (fill_all_syscalls_numbers): Initialize unknown_syscall_number.
2015-10-20 02:59:38 +02:00
# This procedure checks if there was a call to a syscall. The optional
# pattern can match syscalls that vary in implementation, like vfork.
proc check_call_to_syscall { syscall { pattern "" } } {
Improve and fix catch-syscall.exp While fixing another bug, I found that the current gdb.base/catch-syscall.exp is kind of messy, could use some improvements, and is not correctly testing some things. I've made the following patch to address all the issues I found. On the organization side, it does a cleanup and removes unecessary imports of gdb_prompt, uses prepare_for_testing and clean_restart where needed, and fixes some comments. The testcase was also not correctly testing catching syscalls using only numbers, or catching many syscalls at once. I fixed that. The patch also uses a new method for obtaining the syscalls numbers: it relies on the C source file to get them, via <sys/syscall.h> and SYS_* macros. This makes the .exp file simpler because there is no need to include target conditionals there. I tested this on x86_64 Fedora 18. gdb/testsuite/ChangeLog: 2013-12-18 Sergio Durigan Junior <sergiodj@redhat.com> * gdb.base/catch-syscall.c: Include <sys/syscall.h>. (close_syscall, chroot_syscall, exit_group_syscall): New variables. * gdb.base/catch-syscall.exp: Replace gdb_compile by prepare_for_testing. Call fill_all_syscalls_numbers before starting. Replace gdb_exit, gdb_start, gdb_reinitialize_dir and gdb_load by clean_restart. (check_info_bp_any_syscall, check_info_bp_specific_syscall) (check_info_bp_many_syscalls): Remove global gdb_prompt. (check_call_to_syscall): Likewise. Add global decimal. Improve testing regex. (check_return_from_syscall): Likewise. (check_continue, insert_catch_syscall_with_arg): Remove global gdb_prompt. (insert_catch_syscall_with_many_args): Likewise. Add global decimal. Fix $filter_str. Improve testing regex. (check_for_program_end): Remove global gdb_prompt. (test_catch_syscall_without_args): Likewise. Add global decimal. Improve testing regex. (test_catch_syscall_with_args, test_catch_syscall_with_many_args) (test_catch_syscall_with_wrong_args) (test_catch_syscall_restarting_inferior) (test_catch_syscall_fail_nodatadir): Remove global gdb_prompt. (do_syscall_tests): Likewise. Remove global srcdir. (test_catch_syscall_without_args_noxml): Remove global gdb_prompt. Add global last_syscall_number. Test for the exact syscall number to be caught. (test_catch_syscall_with_args_noxml): Remove global gdb_prompt. Add global all_syscalls_numbers. Test each syscall number to be caught, instead of only testing "close". (test_catch_syscall_with_wrong_args_noxml): Remove global gdb_prompt. (do_syscall_tests_without_xml): Likewise. Remove global srcdir. Remove stale comment. (fill_all_syscalls_numbers): Add global last_syscall_number. Fill the correct syscall numbers using information from the inferior.
2013-12-18 23:19:01 +01:00
global decimal
gdb: Improve syscall entry/return tracking on Linux The existing logic was simply to flip syscall entry/return state when a syscall trap was seen, and even then only with active 'catch syscall'. That can get out of sync if 'catch syscall' is toggled at odd times. This patch updates the entry/return state for all syscall traps, regardless of catching state, and also updates known syscall state for other kinds of traps. Almost all PTRACE_EVENT stops are delivered from the middle of a syscall, so this can act like an entry. Every other kind of ptrace stop is only delivered outside of syscall event pairs, so marking them ignored ensures the next syscall trap looks like an entry. Three new test scenarios are added to catch-syscall.exp: - Disable 'catch syscall' from an entry to deliberately miss the return event, then re-enable to make sure a new entry is recognized. - Enable 'catch syscall' for the first time from a vfork event, which is a PTRACE_EVENT_VFORK in the middle of the syscall. Make sure the next syscall event is recognized as the return. - Make sure entry and return are recognized for an ENOSYS syscall. This is to defeat a common x86 hack that uses the pre-filled ENOSYS return value as a sign of being on the entry side. gdb/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * linux-nat.c (linux_handle_syscall_trap): Always update entry/ return state, even when not actively catching syscalls at all. (linux_handle_extended_wait): Mark syscall_state like an entry. (wait_lwp): Set syscall_state ignored for other traps. (linux_nat_filter_event): Likewise. gdb/testsuite/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * gdb.base/catch-syscall.c: Include <sched.h>. (unknown_syscall): New variable. (main): Trigger a vfork and an unknown syscall. * gdb.base/catch-syscall.exp (vfork_syscalls): New variable. (unknown_syscall_number): Likewise. (check_call_to_syscall): Accept an optional syscall pattern. (check_return_from_syscall): Likewise. (check_continue): Likewise. (test_catch_syscall_without_args): Check for vfork and ENOSYS. (test_catch_syscall_skipping_return): New test toggling off 'catch syscall' to step over the syscall return, then toggling back on. (test_catch_syscall_mid_vfork): New test turning on 'catch syscall' during a PTRACE_EVENT_VFORK stop, in the middle of a vfork syscall. (do_syscall_tests): Call test_catch_syscall_without_args and test_catch_syscall_mid_vfork. (test_catch_syscall_without_args_noxml): Check for vfork and ENOSYS. (fill_all_syscalls_numbers): Initialize unknown_syscall_number.
2015-10-20 02:59:38 +02:00
if { $pattern eq "" } {
set pattern "${syscall}"
}
set thistest "program has called $syscall"
gdb: Improve syscall entry/return tracking on Linux The existing logic was simply to flip syscall entry/return state when a syscall trap was seen, and even then only with active 'catch syscall'. That can get out of sync if 'catch syscall' is toggled at odd times. This patch updates the entry/return state for all syscall traps, regardless of catching state, and also updates known syscall state for other kinds of traps. Almost all PTRACE_EVENT stops are delivered from the middle of a syscall, so this can act like an entry. Every other kind of ptrace stop is only delivered outside of syscall event pairs, so marking them ignored ensures the next syscall trap looks like an entry. Three new test scenarios are added to catch-syscall.exp: - Disable 'catch syscall' from an entry to deliberately miss the return event, then re-enable to make sure a new entry is recognized. - Enable 'catch syscall' for the first time from a vfork event, which is a PTRACE_EVENT_VFORK in the middle of the syscall. Make sure the next syscall event is recognized as the return. - Make sure entry and return are recognized for an ENOSYS syscall. This is to defeat a common x86 hack that uses the pre-filled ENOSYS return value as a sign of being on the entry side. gdb/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * linux-nat.c (linux_handle_syscall_trap): Always update entry/ return state, even when not actively catching syscalls at all. (linux_handle_extended_wait): Mark syscall_state like an entry. (wait_lwp): Set syscall_state ignored for other traps. (linux_nat_filter_event): Likewise. gdb/testsuite/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * gdb.base/catch-syscall.c: Include <sched.h>. (unknown_syscall): New variable. (main): Trigger a vfork and an unknown syscall. * gdb.base/catch-syscall.exp (vfork_syscalls): New variable. (unknown_syscall_number): Likewise. (check_call_to_syscall): Accept an optional syscall pattern. (check_return_from_syscall): Likewise. (check_continue): Likewise. (test_catch_syscall_without_args): Check for vfork and ENOSYS. (test_catch_syscall_skipping_return): New test toggling off 'catch syscall' to step over the syscall return, then toggling back on. (test_catch_syscall_mid_vfork): New test turning on 'catch syscall' during a PTRACE_EVENT_VFORK stop, in the middle of a vfork syscall. (do_syscall_tests): Call test_catch_syscall_without_args and test_catch_syscall_mid_vfork. (test_catch_syscall_without_args_noxml): Check for vfork and ENOSYS. (fill_all_syscalls_numbers): Initialize unknown_syscall_number.
2015-10-20 02:59:38 +02:00
gdb_test "continue" "Catchpoint $decimal \\(call to syscall .?${pattern}.?\\).*" $thistest
}
gdb: Improve syscall entry/return tracking on Linux The existing logic was simply to flip syscall entry/return state when a syscall trap was seen, and even then only with active 'catch syscall'. That can get out of sync if 'catch syscall' is toggled at odd times. This patch updates the entry/return state for all syscall traps, regardless of catching state, and also updates known syscall state for other kinds of traps. Almost all PTRACE_EVENT stops are delivered from the middle of a syscall, so this can act like an entry. Every other kind of ptrace stop is only delivered outside of syscall event pairs, so marking them ignored ensures the next syscall trap looks like an entry. Three new test scenarios are added to catch-syscall.exp: - Disable 'catch syscall' from an entry to deliberately miss the return event, then re-enable to make sure a new entry is recognized. - Enable 'catch syscall' for the first time from a vfork event, which is a PTRACE_EVENT_VFORK in the middle of the syscall. Make sure the next syscall event is recognized as the return. - Make sure entry and return are recognized for an ENOSYS syscall. This is to defeat a common x86 hack that uses the pre-filled ENOSYS return value as a sign of being on the entry side. gdb/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * linux-nat.c (linux_handle_syscall_trap): Always update entry/ return state, even when not actively catching syscalls at all. (linux_handle_extended_wait): Mark syscall_state like an entry. (wait_lwp): Set syscall_state ignored for other traps. (linux_nat_filter_event): Likewise. gdb/testsuite/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * gdb.base/catch-syscall.c: Include <sched.h>. (unknown_syscall): New variable. (main): Trigger a vfork and an unknown syscall. * gdb.base/catch-syscall.exp (vfork_syscalls): New variable. (unknown_syscall_number): Likewise. (check_call_to_syscall): Accept an optional syscall pattern. (check_return_from_syscall): Likewise. (check_continue): Likewise. (test_catch_syscall_without_args): Check for vfork and ENOSYS. (test_catch_syscall_skipping_return): New test toggling off 'catch syscall' to step over the syscall return, then toggling back on. (test_catch_syscall_mid_vfork): New test turning on 'catch syscall' during a PTRACE_EVENT_VFORK stop, in the middle of a vfork syscall. (do_syscall_tests): Call test_catch_syscall_without_args and test_catch_syscall_mid_vfork. (test_catch_syscall_without_args_noxml): Check for vfork and ENOSYS. (fill_all_syscalls_numbers): Initialize unknown_syscall_number.
2015-10-20 02:59:38 +02:00
# This procedure checks if the syscall returned. The optional pattern
# can match syscalls that vary in implementation, like vfork.
proc check_return_from_syscall { syscall { pattern "" } } {
Improve and fix catch-syscall.exp While fixing another bug, I found that the current gdb.base/catch-syscall.exp is kind of messy, could use some improvements, and is not correctly testing some things. I've made the following patch to address all the issues I found. On the organization side, it does a cleanup and removes unecessary imports of gdb_prompt, uses prepare_for_testing and clean_restart where needed, and fixes some comments. The testcase was also not correctly testing catching syscalls using only numbers, or catching many syscalls at once. I fixed that. The patch also uses a new method for obtaining the syscalls numbers: it relies on the C source file to get them, via <sys/syscall.h> and SYS_* macros. This makes the .exp file simpler because there is no need to include target conditionals there. I tested this on x86_64 Fedora 18. gdb/testsuite/ChangeLog: 2013-12-18 Sergio Durigan Junior <sergiodj@redhat.com> * gdb.base/catch-syscall.c: Include <sys/syscall.h>. (close_syscall, chroot_syscall, exit_group_syscall): New variables. * gdb.base/catch-syscall.exp: Replace gdb_compile by prepare_for_testing. Call fill_all_syscalls_numbers before starting. Replace gdb_exit, gdb_start, gdb_reinitialize_dir and gdb_load by clean_restart. (check_info_bp_any_syscall, check_info_bp_specific_syscall) (check_info_bp_many_syscalls): Remove global gdb_prompt. (check_call_to_syscall): Likewise. Add global decimal. Improve testing regex. (check_return_from_syscall): Likewise. (check_continue, insert_catch_syscall_with_arg): Remove global gdb_prompt. (insert_catch_syscall_with_many_args): Likewise. Add global decimal. Fix $filter_str. Improve testing regex. (check_for_program_end): Remove global gdb_prompt. (test_catch_syscall_without_args): Likewise. Add global decimal. Improve testing regex. (test_catch_syscall_with_args, test_catch_syscall_with_many_args) (test_catch_syscall_with_wrong_args) (test_catch_syscall_restarting_inferior) (test_catch_syscall_fail_nodatadir): Remove global gdb_prompt. (do_syscall_tests): Likewise. Remove global srcdir. (test_catch_syscall_without_args_noxml): Remove global gdb_prompt. Add global last_syscall_number. Test for the exact syscall number to be caught. (test_catch_syscall_with_args_noxml): Remove global gdb_prompt. Add global all_syscalls_numbers. Test each syscall number to be caught, instead of only testing "close". (test_catch_syscall_with_wrong_args_noxml): Remove global gdb_prompt. (do_syscall_tests_without_xml): Likewise. Remove global srcdir. Remove stale comment. (fill_all_syscalls_numbers): Add global last_syscall_number. Fill the correct syscall numbers using information from the inferior.
2013-12-18 23:19:01 +01:00
global decimal
gdb: Improve syscall entry/return tracking on Linux The existing logic was simply to flip syscall entry/return state when a syscall trap was seen, and even then only with active 'catch syscall'. That can get out of sync if 'catch syscall' is toggled at odd times. This patch updates the entry/return state for all syscall traps, regardless of catching state, and also updates known syscall state for other kinds of traps. Almost all PTRACE_EVENT stops are delivered from the middle of a syscall, so this can act like an entry. Every other kind of ptrace stop is only delivered outside of syscall event pairs, so marking them ignored ensures the next syscall trap looks like an entry. Three new test scenarios are added to catch-syscall.exp: - Disable 'catch syscall' from an entry to deliberately miss the return event, then re-enable to make sure a new entry is recognized. - Enable 'catch syscall' for the first time from a vfork event, which is a PTRACE_EVENT_VFORK in the middle of the syscall. Make sure the next syscall event is recognized as the return. - Make sure entry and return are recognized for an ENOSYS syscall. This is to defeat a common x86 hack that uses the pre-filled ENOSYS return value as a sign of being on the entry side. gdb/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * linux-nat.c (linux_handle_syscall_trap): Always update entry/ return state, even when not actively catching syscalls at all. (linux_handle_extended_wait): Mark syscall_state like an entry. (wait_lwp): Set syscall_state ignored for other traps. (linux_nat_filter_event): Likewise. gdb/testsuite/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * gdb.base/catch-syscall.c: Include <sched.h>. (unknown_syscall): New variable. (main): Trigger a vfork and an unknown syscall. * gdb.base/catch-syscall.exp (vfork_syscalls): New variable. (unknown_syscall_number): Likewise. (check_call_to_syscall): Accept an optional syscall pattern. (check_return_from_syscall): Likewise. (check_continue): Likewise. (test_catch_syscall_without_args): Check for vfork and ENOSYS. (test_catch_syscall_skipping_return): New test toggling off 'catch syscall' to step over the syscall return, then toggling back on. (test_catch_syscall_mid_vfork): New test turning on 'catch syscall' during a PTRACE_EVENT_VFORK stop, in the middle of a vfork syscall. (do_syscall_tests): Call test_catch_syscall_without_args and test_catch_syscall_mid_vfork. (test_catch_syscall_without_args_noxml): Check for vfork and ENOSYS. (fill_all_syscalls_numbers): Initialize unknown_syscall_number.
2015-10-20 02:59:38 +02:00
if { $pattern eq "" } {
set pattern "${syscall}"
}
set thistest "syscall $syscall has returned"
gdb: Improve syscall entry/return tracking on Linux The existing logic was simply to flip syscall entry/return state when a syscall trap was seen, and even then only with active 'catch syscall'. That can get out of sync if 'catch syscall' is toggled at odd times. This patch updates the entry/return state for all syscall traps, regardless of catching state, and also updates known syscall state for other kinds of traps. Almost all PTRACE_EVENT stops are delivered from the middle of a syscall, so this can act like an entry. Every other kind of ptrace stop is only delivered outside of syscall event pairs, so marking them ignored ensures the next syscall trap looks like an entry. Three new test scenarios are added to catch-syscall.exp: - Disable 'catch syscall' from an entry to deliberately miss the return event, then re-enable to make sure a new entry is recognized. - Enable 'catch syscall' for the first time from a vfork event, which is a PTRACE_EVENT_VFORK in the middle of the syscall. Make sure the next syscall event is recognized as the return. - Make sure entry and return are recognized for an ENOSYS syscall. This is to defeat a common x86 hack that uses the pre-filled ENOSYS return value as a sign of being on the entry side. gdb/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * linux-nat.c (linux_handle_syscall_trap): Always update entry/ return state, even when not actively catching syscalls at all. (linux_handle_extended_wait): Mark syscall_state like an entry. (wait_lwp): Set syscall_state ignored for other traps. (linux_nat_filter_event): Likewise. gdb/testsuite/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * gdb.base/catch-syscall.c: Include <sched.h>. (unknown_syscall): New variable. (main): Trigger a vfork and an unknown syscall. * gdb.base/catch-syscall.exp (vfork_syscalls): New variable. (unknown_syscall_number): Likewise. (check_call_to_syscall): Accept an optional syscall pattern. (check_return_from_syscall): Likewise. (check_continue): Likewise. (test_catch_syscall_without_args): Check for vfork and ENOSYS. (test_catch_syscall_skipping_return): New test toggling off 'catch syscall' to step over the syscall return, then toggling back on. (test_catch_syscall_mid_vfork): New test turning on 'catch syscall' during a PTRACE_EVENT_VFORK stop, in the middle of a vfork syscall. (do_syscall_tests): Call test_catch_syscall_without_args and test_catch_syscall_mid_vfork. (test_catch_syscall_without_args_noxml): Check for vfork and ENOSYS. (fill_all_syscalls_numbers): Initialize unknown_syscall_number.
2015-10-20 02:59:38 +02:00
gdb_test "continue" "Catchpoint $decimal \\(returned from syscall ${pattern}\\).*" $thistest
}
# Internal procedure that performs two 'continue' commands and checks if
gdb: Improve syscall entry/return tracking on Linux The existing logic was simply to flip syscall entry/return state when a syscall trap was seen, and even then only with active 'catch syscall'. That can get out of sync if 'catch syscall' is toggled at odd times. This patch updates the entry/return state for all syscall traps, regardless of catching state, and also updates known syscall state for other kinds of traps. Almost all PTRACE_EVENT stops are delivered from the middle of a syscall, so this can act like an entry. Every other kind of ptrace stop is only delivered outside of syscall event pairs, so marking them ignored ensures the next syscall trap looks like an entry. Three new test scenarios are added to catch-syscall.exp: - Disable 'catch syscall' from an entry to deliberately miss the return event, then re-enable to make sure a new entry is recognized. - Enable 'catch syscall' for the first time from a vfork event, which is a PTRACE_EVENT_VFORK in the middle of the syscall. Make sure the next syscall event is recognized as the return. - Make sure entry and return are recognized for an ENOSYS syscall. This is to defeat a common x86 hack that uses the pre-filled ENOSYS return value as a sign of being on the entry side. gdb/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * linux-nat.c (linux_handle_syscall_trap): Always update entry/ return state, even when not actively catching syscalls at all. (linux_handle_extended_wait): Mark syscall_state like an entry. (wait_lwp): Set syscall_state ignored for other traps. (linux_nat_filter_event): Likewise. gdb/testsuite/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * gdb.base/catch-syscall.c: Include <sched.h>. (unknown_syscall): New variable. (main): Trigger a vfork and an unknown syscall. * gdb.base/catch-syscall.exp (vfork_syscalls): New variable. (unknown_syscall_number): Likewise. (check_call_to_syscall): Accept an optional syscall pattern. (check_return_from_syscall): Likewise. (check_continue): Likewise. (test_catch_syscall_without_args): Check for vfork and ENOSYS. (test_catch_syscall_skipping_return): New test toggling off 'catch syscall' to step over the syscall return, then toggling back on. (test_catch_syscall_mid_vfork): New test turning on 'catch syscall' during a PTRACE_EVENT_VFORK stop, in the middle of a vfork syscall. (do_syscall_tests): Call test_catch_syscall_without_args and test_catch_syscall_mid_vfork. (test_catch_syscall_without_args_noxml): Check for vfork and ENOSYS. (fill_all_syscalls_numbers): Initialize unknown_syscall_number.
2015-10-20 02:59:38 +02:00
# a syscall call AND return occur. The optional pattern can match
# syscalls that vary in implementation, like vfork.
proc check_continue { syscall { pattern "" } } {
# Testing if the 'continue' stops at the
# specified syscall_name. If it does, then it should
# first print that the infeior has called the syscall,
# and after print that the syscall has returned.
# Testing if the inferior has called the syscall.
gdb: Improve syscall entry/return tracking on Linux The existing logic was simply to flip syscall entry/return state when a syscall trap was seen, and even then only with active 'catch syscall'. That can get out of sync if 'catch syscall' is toggled at odd times. This patch updates the entry/return state for all syscall traps, regardless of catching state, and also updates known syscall state for other kinds of traps. Almost all PTRACE_EVENT stops are delivered from the middle of a syscall, so this can act like an entry. Every other kind of ptrace stop is only delivered outside of syscall event pairs, so marking them ignored ensures the next syscall trap looks like an entry. Three new test scenarios are added to catch-syscall.exp: - Disable 'catch syscall' from an entry to deliberately miss the return event, then re-enable to make sure a new entry is recognized. - Enable 'catch syscall' for the first time from a vfork event, which is a PTRACE_EVENT_VFORK in the middle of the syscall. Make sure the next syscall event is recognized as the return. - Make sure entry and return are recognized for an ENOSYS syscall. This is to defeat a common x86 hack that uses the pre-filled ENOSYS return value as a sign of being on the entry side. gdb/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * linux-nat.c (linux_handle_syscall_trap): Always update entry/ return state, even when not actively catching syscalls at all. (linux_handle_extended_wait): Mark syscall_state like an entry. (wait_lwp): Set syscall_state ignored for other traps. (linux_nat_filter_event): Likewise. gdb/testsuite/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * gdb.base/catch-syscall.c: Include <sched.h>. (unknown_syscall): New variable. (main): Trigger a vfork and an unknown syscall. * gdb.base/catch-syscall.exp (vfork_syscalls): New variable. (unknown_syscall_number): Likewise. (check_call_to_syscall): Accept an optional syscall pattern. (check_return_from_syscall): Likewise. (check_continue): Likewise. (test_catch_syscall_without_args): Check for vfork and ENOSYS. (test_catch_syscall_skipping_return): New test toggling off 'catch syscall' to step over the syscall return, then toggling back on. (test_catch_syscall_mid_vfork): New test turning on 'catch syscall' during a PTRACE_EVENT_VFORK stop, in the middle of a vfork syscall. (do_syscall_tests): Call test_catch_syscall_without_args and test_catch_syscall_mid_vfork. (test_catch_syscall_without_args_noxml): Check for vfork and ENOSYS. (fill_all_syscalls_numbers): Initialize unknown_syscall_number.
2015-10-20 02:59:38 +02:00
check_call_to_syscall $syscall $pattern
# And now, that the syscall has returned.
gdb: Improve syscall entry/return tracking on Linux The existing logic was simply to flip syscall entry/return state when a syscall trap was seen, and even then only with active 'catch syscall'. That can get out of sync if 'catch syscall' is toggled at odd times. This patch updates the entry/return state for all syscall traps, regardless of catching state, and also updates known syscall state for other kinds of traps. Almost all PTRACE_EVENT stops are delivered from the middle of a syscall, so this can act like an entry. Every other kind of ptrace stop is only delivered outside of syscall event pairs, so marking them ignored ensures the next syscall trap looks like an entry. Three new test scenarios are added to catch-syscall.exp: - Disable 'catch syscall' from an entry to deliberately miss the return event, then re-enable to make sure a new entry is recognized. - Enable 'catch syscall' for the first time from a vfork event, which is a PTRACE_EVENT_VFORK in the middle of the syscall. Make sure the next syscall event is recognized as the return. - Make sure entry and return are recognized for an ENOSYS syscall. This is to defeat a common x86 hack that uses the pre-filled ENOSYS return value as a sign of being on the entry side. gdb/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * linux-nat.c (linux_handle_syscall_trap): Always update entry/ return state, even when not actively catching syscalls at all. (linux_handle_extended_wait): Mark syscall_state like an entry. (wait_lwp): Set syscall_state ignored for other traps. (linux_nat_filter_event): Likewise. gdb/testsuite/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * gdb.base/catch-syscall.c: Include <sched.h>. (unknown_syscall): New variable. (main): Trigger a vfork and an unknown syscall. * gdb.base/catch-syscall.exp (vfork_syscalls): New variable. (unknown_syscall_number): Likewise. (check_call_to_syscall): Accept an optional syscall pattern. (check_return_from_syscall): Likewise. (check_continue): Likewise. (test_catch_syscall_without_args): Check for vfork and ENOSYS. (test_catch_syscall_skipping_return): New test toggling off 'catch syscall' to step over the syscall return, then toggling back on. (test_catch_syscall_mid_vfork): New test turning on 'catch syscall' during a PTRACE_EVENT_VFORK stop, in the middle of a vfork syscall. (do_syscall_tests): Call test_catch_syscall_without_args and test_catch_syscall_mid_vfork. (test_catch_syscall_without_args_noxml): Check for vfork and ENOSYS. (fill_all_syscalls_numbers): Initialize unknown_syscall_number.
2015-10-20 02:59:38 +02:00
check_return_from_syscall $syscall $pattern
}
# Inserts a syscall catchpoint with an argument.
proc insert_catch_syscall_with_arg { syscall } {
Improve and fix catch-syscall.exp While fixing another bug, I found that the current gdb.base/catch-syscall.exp is kind of messy, could use some improvements, and is not correctly testing some things. I've made the following patch to address all the issues I found. On the organization side, it does a cleanup and removes unecessary imports of gdb_prompt, uses prepare_for_testing and clean_restart where needed, and fixes some comments. The testcase was also not correctly testing catching syscalls using only numbers, or catching many syscalls at once. I fixed that. The patch also uses a new method for obtaining the syscalls numbers: it relies on the C source file to get them, via <sys/syscall.h> and SYS_* macros. This makes the .exp file simpler because there is no need to include target conditionals there. I tested this on x86_64 Fedora 18. gdb/testsuite/ChangeLog: 2013-12-18 Sergio Durigan Junior <sergiodj@redhat.com> * gdb.base/catch-syscall.c: Include <sys/syscall.h>. (close_syscall, chroot_syscall, exit_group_syscall): New variables. * gdb.base/catch-syscall.exp: Replace gdb_compile by prepare_for_testing. Call fill_all_syscalls_numbers before starting. Replace gdb_exit, gdb_start, gdb_reinitialize_dir and gdb_load by clean_restart. (check_info_bp_any_syscall, check_info_bp_specific_syscall) (check_info_bp_many_syscalls): Remove global gdb_prompt. (check_call_to_syscall): Likewise. Add global decimal. Improve testing regex. (check_return_from_syscall): Likewise. (check_continue, insert_catch_syscall_with_arg): Remove global gdb_prompt. (insert_catch_syscall_with_many_args): Likewise. Add global decimal. Fix $filter_str. Improve testing regex. (check_for_program_end): Remove global gdb_prompt. (test_catch_syscall_without_args): Likewise. Add global decimal. Improve testing regex. (test_catch_syscall_with_args, test_catch_syscall_with_many_args) (test_catch_syscall_with_wrong_args) (test_catch_syscall_restarting_inferior) (test_catch_syscall_fail_nodatadir): Remove global gdb_prompt. (do_syscall_tests): Likewise. Remove global srcdir. (test_catch_syscall_without_args_noxml): Remove global gdb_prompt. Add global last_syscall_number. Test for the exact syscall number to be caught. (test_catch_syscall_with_args_noxml): Remove global gdb_prompt. Add global all_syscalls_numbers. Test each syscall number to be caught, instead of only testing "close". (test_catch_syscall_with_wrong_args_noxml): Remove global gdb_prompt. (do_syscall_tests_without_xml): Likewise. Remove global srcdir. Remove stale comment. (fill_all_syscalls_numbers): Add global last_syscall_number. Fill the correct syscall numbers using information from the inferior.
2013-12-18 23:19:01 +01:00
global decimal
# Trying to set the catchpoint
set thistest "catch syscall with arguments ($syscall)"
Improve and fix catch-syscall.exp While fixing another bug, I found that the current gdb.base/catch-syscall.exp is kind of messy, could use some improvements, and is not correctly testing some things. I've made the following patch to address all the issues I found. On the organization side, it does a cleanup and removes unecessary imports of gdb_prompt, uses prepare_for_testing and clean_restart where needed, and fixes some comments. The testcase was also not correctly testing catching syscalls using only numbers, or catching many syscalls at once. I fixed that. The patch also uses a new method for obtaining the syscalls numbers: it relies on the C source file to get them, via <sys/syscall.h> and SYS_* macros. This makes the .exp file simpler because there is no need to include target conditionals there. I tested this on x86_64 Fedora 18. gdb/testsuite/ChangeLog: 2013-12-18 Sergio Durigan Junior <sergiodj@redhat.com> * gdb.base/catch-syscall.c: Include <sys/syscall.h>. (close_syscall, chroot_syscall, exit_group_syscall): New variables. * gdb.base/catch-syscall.exp: Replace gdb_compile by prepare_for_testing. Call fill_all_syscalls_numbers before starting. Replace gdb_exit, gdb_start, gdb_reinitialize_dir and gdb_load by clean_restart. (check_info_bp_any_syscall, check_info_bp_specific_syscall) (check_info_bp_many_syscalls): Remove global gdb_prompt. (check_call_to_syscall): Likewise. Add global decimal. Improve testing regex. (check_return_from_syscall): Likewise. (check_continue, insert_catch_syscall_with_arg): Remove global gdb_prompt. (insert_catch_syscall_with_many_args): Likewise. Add global decimal. Fix $filter_str. Improve testing regex. (check_for_program_end): Remove global gdb_prompt. (test_catch_syscall_without_args): Likewise. Add global decimal. Improve testing regex. (test_catch_syscall_with_args, test_catch_syscall_with_many_args) (test_catch_syscall_with_wrong_args) (test_catch_syscall_restarting_inferior) (test_catch_syscall_fail_nodatadir): Remove global gdb_prompt. (do_syscall_tests): Likewise. Remove global srcdir. (test_catch_syscall_without_args_noxml): Remove global gdb_prompt. Add global last_syscall_number. Test for the exact syscall number to be caught. (test_catch_syscall_with_args_noxml): Remove global gdb_prompt. Add global all_syscalls_numbers. Test each syscall number to be caught, instead of only testing "close". (test_catch_syscall_with_wrong_args_noxml): Remove global gdb_prompt. (do_syscall_tests_without_xml): Likewise. Remove global srcdir. Remove stale comment. (fill_all_syscalls_numbers): Add global last_syscall_number. Fill the correct syscall numbers using information from the inferior.
2013-12-18 23:19:01 +01:00
gdb_test "catch syscall $syscall" "Catchpoint $decimal \\(syscall \'?${syscall}\'?( \[${decimal}\])?\\)" $thistest
check_info_bp_specific_syscall $syscall
}
# Inserts a syscall catchpoint with many arguments.
proc insert_catch_syscall_with_many_args { syscalls numbers } {
Improve and fix catch-syscall.exp While fixing another bug, I found that the current gdb.base/catch-syscall.exp is kind of messy, could use some improvements, and is not correctly testing some things. I've made the following patch to address all the issues I found. On the organization side, it does a cleanup and removes unecessary imports of gdb_prompt, uses prepare_for_testing and clean_restart where needed, and fixes some comments. The testcase was also not correctly testing catching syscalls using only numbers, or catching many syscalls at once. I fixed that. The patch also uses a new method for obtaining the syscalls numbers: it relies on the C source file to get them, via <sys/syscall.h> and SYS_* macros. This makes the .exp file simpler because there is no need to include target conditionals there. I tested this on x86_64 Fedora 18. gdb/testsuite/ChangeLog: 2013-12-18 Sergio Durigan Junior <sergiodj@redhat.com> * gdb.base/catch-syscall.c: Include <sys/syscall.h>. (close_syscall, chroot_syscall, exit_group_syscall): New variables. * gdb.base/catch-syscall.exp: Replace gdb_compile by prepare_for_testing. Call fill_all_syscalls_numbers before starting. Replace gdb_exit, gdb_start, gdb_reinitialize_dir and gdb_load by clean_restart. (check_info_bp_any_syscall, check_info_bp_specific_syscall) (check_info_bp_many_syscalls): Remove global gdb_prompt. (check_call_to_syscall): Likewise. Add global decimal. Improve testing regex. (check_return_from_syscall): Likewise. (check_continue, insert_catch_syscall_with_arg): Remove global gdb_prompt. (insert_catch_syscall_with_many_args): Likewise. Add global decimal. Fix $filter_str. Improve testing regex. (check_for_program_end): Remove global gdb_prompt. (test_catch_syscall_without_args): Likewise. Add global decimal. Improve testing regex. (test_catch_syscall_with_args, test_catch_syscall_with_many_args) (test_catch_syscall_with_wrong_args) (test_catch_syscall_restarting_inferior) (test_catch_syscall_fail_nodatadir): Remove global gdb_prompt. (do_syscall_tests): Likewise. Remove global srcdir. (test_catch_syscall_without_args_noxml): Remove global gdb_prompt. Add global last_syscall_number. Test for the exact syscall number to be caught. (test_catch_syscall_with_args_noxml): Remove global gdb_prompt. Add global all_syscalls_numbers. Test each syscall number to be caught, instead of only testing "close". (test_catch_syscall_with_wrong_args_noxml): Remove global gdb_prompt. (do_syscall_tests_without_xml): Likewise. Remove global srcdir. Remove stale comment. (fill_all_syscalls_numbers): Add global last_syscall_number. Fill the correct syscall numbers using information from the inferior.
2013-12-18 23:19:01 +01:00
global decimal
set catch [ join $syscalls " " ]
set filter_str ""
foreach name $syscalls number $numbers {
Improve and fix catch-syscall.exp While fixing another bug, I found that the current gdb.base/catch-syscall.exp is kind of messy, could use some improvements, and is not correctly testing some things. I've made the following patch to address all the issues I found. On the organization side, it does a cleanup and removes unecessary imports of gdb_prompt, uses prepare_for_testing and clean_restart where needed, and fixes some comments. The testcase was also not correctly testing catching syscalls using only numbers, or catching many syscalls at once. I fixed that. The patch also uses a new method for obtaining the syscalls numbers: it relies on the C source file to get them, via <sys/syscall.h> and SYS_* macros. This makes the .exp file simpler because there is no need to include target conditionals there. I tested this on x86_64 Fedora 18. gdb/testsuite/ChangeLog: 2013-12-18 Sergio Durigan Junior <sergiodj@redhat.com> * gdb.base/catch-syscall.c: Include <sys/syscall.h>. (close_syscall, chroot_syscall, exit_group_syscall): New variables. * gdb.base/catch-syscall.exp: Replace gdb_compile by prepare_for_testing. Call fill_all_syscalls_numbers before starting. Replace gdb_exit, gdb_start, gdb_reinitialize_dir and gdb_load by clean_restart. (check_info_bp_any_syscall, check_info_bp_specific_syscall) (check_info_bp_many_syscalls): Remove global gdb_prompt. (check_call_to_syscall): Likewise. Add global decimal. Improve testing regex. (check_return_from_syscall): Likewise. (check_continue, insert_catch_syscall_with_arg): Remove global gdb_prompt. (insert_catch_syscall_with_many_args): Likewise. Add global decimal. Fix $filter_str. Improve testing regex. (check_for_program_end): Remove global gdb_prompt. (test_catch_syscall_without_args): Likewise. Add global decimal. Improve testing regex. (test_catch_syscall_with_args, test_catch_syscall_with_many_args) (test_catch_syscall_with_wrong_args) (test_catch_syscall_restarting_inferior) (test_catch_syscall_fail_nodatadir): Remove global gdb_prompt. (do_syscall_tests): Likewise. Remove global srcdir. (test_catch_syscall_without_args_noxml): Remove global gdb_prompt. Add global last_syscall_number. Test for the exact syscall number to be caught. (test_catch_syscall_with_args_noxml): Remove global gdb_prompt. Add global all_syscalls_numbers. Test each syscall number to be caught, instead of only testing "close". (test_catch_syscall_with_wrong_args_noxml): Remove global gdb_prompt. (do_syscall_tests_without_xml): Likewise. Remove global srcdir. Remove stale comment. (fill_all_syscalls_numbers): Add global last_syscall_number. Fill the correct syscall numbers using information from the inferior.
2013-12-18 23:19:01 +01:00
set filter_str "${filter_str}'${name}' \\\[${number}\\\] "
}
set filter_str [ string trimright $filter_str " " ]
# Trying to set the catchpoint
set thistest "catch syscall with arguments ($filter_str)"
Improve and fix catch-syscall.exp While fixing another bug, I found that the current gdb.base/catch-syscall.exp is kind of messy, could use some improvements, and is not correctly testing some things. I've made the following patch to address all the issues I found. On the organization side, it does a cleanup and removes unecessary imports of gdb_prompt, uses prepare_for_testing and clean_restart where needed, and fixes some comments. The testcase was also not correctly testing catching syscalls using only numbers, or catching many syscalls at once. I fixed that. The patch also uses a new method for obtaining the syscalls numbers: it relies on the C source file to get them, via <sys/syscall.h> and SYS_* macros. This makes the .exp file simpler because there is no need to include target conditionals there. I tested this on x86_64 Fedora 18. gdb/testsuite/ChangeLog: 2013-12-18 Sergio Durigan Junior <sergiodj@redhat.com> * gdb.base/catch-syscall.c: Include <sys/syscall.h>. (close_syscall, chroot_syscall, exit_group_syscall): New variables. * gdb.base/catch-syscall.exp: Replace gdb_compile by prepare_for_testing. Call fill_all_syscalls_numbers before starting. Replace gdb_exit, gdb_start, gdb_reinitialize_dir and gdb_load by clean_restart. (check_info_bp_any_syscall, check_info_bp_specific_syscall) (check_info_bp_many_syscalls): Remove global gdb_prompt. (check_call_to_syscall): Likewise. Add global decimal. Improve testing regex. (check_return_from_syscall): Likewise. (check_continue, insert_catch_syscall_with_arg): Remove global gdb_prompt. (insert_catch_syscall_with_many_args): Likewise. Add global decimal. Fix $filter_str. Improve testing regex. (check_for_program_end): Remove global gdb_prompt. (test_catch_syscall_without_args): Likewise. Add global decimal. Improve testing regex. (test_catch_syscall_with_args, test_catch_syscall_with_many_args) (test_catch_syscall_with_wrong_args) (test_catch_syscall_restarting_inferior) (test_catch_syscall_fail_nodatadir): Remove global gdb_prompt. (do_syscall_tests): Likewise. Remove global srcdir. (test_catch_syscall_without_args_noxml): Remove global gdb_prompt. Add global last_syscall_number. Test for the exact syscall number to be caught. (test_catch_syscall_with_args_noxml): Remove global gdb_prompt. Add global all_syscalls_numbers. Test each syscall number to be caught, instead of only testing "close". (test_catch_syscall_with_wrong_args_noxml): Remove global gdb_prompt. (do_syscall_tests_without_xml): Likewise. Remove global srcdir. Remove stale comment. (fill_all_syscalls_numbers): Add global last_syscall_number. Fill the correct syscall numbers using information from the inferior.
2013-12-18 23:19:01 +01:00
gdb_test "catch syscall $catch" "Catchpoint $decimal \\(syscalls ${filter_str}\\).*" $thistest
check_info_bp_many_syscalls $syscalls
}
proc check_for_program_end {} {
# Deleting the catchpoints
delete_breakpoints
gdb * infrun.c (print_exited_reason): Include inferior id and pid in message. gdb/testsuite * lib/opencl.exp (skip_opencl_tests): Update for exit message change. * lib/mi-support.exp (mi_gdb_test): Update for exit message change. * lib/gdb.exp (gdb_test_multiple): Update comment. Update for exit message change. (skip_altivec_tests): Update for exit message change. (skip_vsx_tests): Likewise. (gdb_continue_to_end): Likewise. Add 'command' argument. * lib/cell.exp (skip_cell_tests): Update for exit message change. * gdb.threads/tls.exp: Update for exit message change. * gdb.threads/thread-unwindonsignal.exp: Use gdb_continue_to_end. * gdb.threads/step.exp (step_it): Update for exit message change. (continue_all): Likewise. * gdb.threads/print-threads.exp (test_all_threads): Update for exit message change. * gdb.threads/interrupted-hand-call.exp: Use gdb_continue_to_end. * gdb.threads/execl.exp: Use gdb_continue_to_end. * gdb.python/py-prettyprint.exp (run_lang_tests): Use gdb_continue_to_end. * gdb.hp/gdb.objdbg/objdbg02.exp: Use gdb_continue_to_end. * gdb.hp/gdb.objdbg/objdbg01.exp: Use gdb_continue_to_end. * gdb.hp/gdb.defects/solib-d.exp: Update for exit message change. * gdb.cp/method.exp: Update for exit message change. * gdb.cp/mb-templates.exp: Update for exit message change. * gdb.cp/mb-inline.exp: Use gdb_continue_to_end. * gdb.cp/annota3.exp: Update for exit message change. * gdb.cp/annota2.exp: Update for exit message change. * gdb.cell/fork.exp: Use gdb_continue_to_end. * gdb.base/term.exp: Update for exit message change. * gdb.base/step-test.exp (test_i): Update for exit message change. * gdb.base/sigstep.exp (advance): Update for exit message change. (advancei): Likewise. * gdb.base/siginfo.exp: Update for exit message change. * gdb.base/shlib-call.exp: Use gdb_continue_to_end. * gdb.base/reread.exp: Use gdb_continue_to_end. * gdb.base/langs.exp: Use gdb_continue_to_end. * gdb.base/interrupt.exp: Update for exit message change. * gdb.base/gdb1555.exp: Update for exit message change. * gdb.base/exe-lock.exp: Use gdb_continue_to_end. * gdb.base/ending-run.exp: Update for exit message change. * gdb.base/chng-syms.exp: Update for exit message change. * gdb.base/checkpoint.exp: Update for exit message change. * gdb.base/catch-syscall.exp (check_for_program_end): Use gdb_continue_to_end. (test_catch_syscall_with_wrong_args): Likewise. * gdb.base/call-signal-resume.exp: Use gdb_continue_to_end. * gdb.base/break-interp.exp (test_ld): Update for exit message change. * gdb.base/bang.exp: Update for exit message change. * gdb.base/attach.exp (do_attach_tests): Use gdb_continue_to_end. (do_call_attach_tests): Likewise. * gdb.base/a2-run.exp: Update for exit message change. * gdb.arch/ppc-dfp.exp: Update for exit message change. * gdb.ada/tasks.exp: Use gdb_continue_to_end. * gdb.ada/catch_ex.exp: Use gdb_continue_to_end.
2011-03-07 17:03:04 +01:00
gdb_continue_to_end
}
proc test_catch_syscall_without_args {} {
gdb: Improve syscall entry/return tracking on Linux The existing logic was simply to flip syscall entry/return state when a syscall trap was seen, and even then only with active 'catch syscall'. That can get out of sync if 'catch syscall' is toggled at odd times. This patch updates the entry/return state for all syscall traps, regardless of catching state, and also updates known syscall state for other kinds of traps. Almost all PTRACE_EVENT stops are delivered from the middle of a syscall, so this can act like an entry. Every other kind of ptrace stop is only delivered outside of syscall event pairs, so marking them ignored ensures the next syscall trap looks like an entry. Three new test scenarios are added to catch-syscall.exp: - Disable 'catch syscall' from an entry to deliberately miss the return event, then re-enable to make sure a new entry is recognized. - Enable 'catch syscall' for the first time from a vfork event, which is a PTRACE_EVENT_VFORK in the middle of the syscall. Make sure the next syscall event is recognized as the return. - Make sure entry and return are recognized for an ENOSYS syscall. This is to defeat a common x86 hack that uses the pre-filled ENOSYS return value as a sign of being on the entry side. gdb/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * linux-nat.c (linux_handle_syscall_trap): Always update entry/ return state, even when not actively catching syscalls at all. (linux_handle_extended_wait): Mark syscall_state like an entry. (wait_lwp): Set syscall_state ignored for other traps. (linux_nat_filter_event): Likewise. gdb/testsuite/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * gdb.base/catch-syscall.c: Include <sched.h>. (unknown_syscall): New variable. (main): Trigger a vfork and an unknown syscall. * gdb.base/catch-syscall.exp (vfork_syscalls): New variable. (unknown_syscall_number): Likewise. (check_call_to_syscall): Accept an optional syscall pattern. (check_return_from_syscall): Likewise. (check_continue): Likewise. (test_catch_syscall_without_args): Check for vfork and ENOSYS. (test_catch_syscall_skipping_return): New test toggling off 'catch syscall' to step over the syscall return, then toggling back on. (test_catch_syscall_mid_vfork): New test turning on 'catch syscall' during a PTRACE_EVENT_VFORK stop, in the middle of a vfork syscall. (do_syscall_tests): Call test_catch_syscall_without_args and test_catch_syscall_mid_vfork. (test_catch_syscall_without_args_noxml): Check for vfork and ENOSYS. (fill_all_syscalls_numbers): Initialize unknown_syscall_number.
2015-10-20 02:59:38 +02:00
global all_syscalls last_syscall vfork_syscalls unknown_syscall_number decimal
with_test_prefix "without arguments" {
# Trying to set the syscall.
Improve and fix catch-syscall.exp While fixing another bug, I found that the current gdb.base/catch-syscall.exp is kind of messy, could use some improvements, and is not correctly testing some things. I've made the following patch to address all the issues I found. On the organization side, it does a cleanup and removes unecessary imports of gdb_prompt, uses prepare_for_testing and clean_restart where needed, and fixes some comments. The testcase was also not correctly testing catching syscalls using only numbers, or catching many syscalls at once. I fixed that. The patch also uses a new method for obtaining the syscalls numbers: it relies on the C source file to get them, via <sys/syscall.h> and SYS_* macros. This makes the .exp file simpler because there is no need to include target conditionals there. I tested this on x86_64 Fedora 18. gdb/testsuite/ChangeLog: 2013-12-18 Sergio Durigan Junior <sergiodj@redhat.com> * gdb.base/catch-syscall.c: Include <sys/syscall.h>. (close_syscall, chroot_syscall, exit_group_syscall): New variables. * gdb.base/catch-syscall.exp: Replace gdb_compile by prepare_for_testing. Call fill_all_syscalls_numbers before starting. Replace gdb_exit, gdb_start, gdb_reinitialize_dir and gdb_load by clean_restart. (check_info_bp_any_syscall, check_info_bp_specific_syscall) (check_info_bp_many_syscalls): Remove global gdb_prompt. (check_call_to_syscall): Likewise. Add global decimal. Improve testing regex. (check_return_from_syscall): Likewise. (check_continue, insert_catch_syscall_with_arg): Remove global gdb_prompt. (insert_catch_syscall_with_many_args): Likewise. Add global decimal. Fix $filter_str. Improve testing regex. (check_for_program_end): Remove global gdb_prompt. (test_catch_syscall_without_args): Likewise. Add global decimal. Improve testing regex. (test_catch_syscall_with_args, test_catch_syscall_with_many_args) (test_catch_syscall_with_wrong_args) (test_catch_syscall_restarting_inferior) (test_catch_syscall_fail_nodatadir): Remove global gdb_prompt. (do_syscall_tests): Likewise. Remove global srcdir. (test_catch_syscall_without_args_noxml): Remove global gdb_prompt. Add global last_syscall_number. Test for the exact syscall number to be caught. (test_catch_syscall_with_args_noxml): Remove global gdb_prompt. Add global all_syscalls_numbers. Test each syscall number to be caught, instead of only testing "close". (test_catch_syscall_with_wrong_args_noxml): Remove global gdb_prompt. (do_syscall_tests_without_xml): Likewise. Remove global srcdir. Remove stale comment. (fill_all_syscalls_numbers): Add global last_syscall_number. Fill the correct syscall numbers using information from the inferior.
2013-12-18 23:19:01 +01:00
gdb_test "catch syscall" "Catchpoint $decimal \\(any syscall\\)"
check_info_bp_any_syscall
# We have to check every syscall.
foreach name $all_syscalls {
check_continue $name
}
gdb: Improve syscall entry/return tracking on Linux The existing logic was simply to flip syscall entry/return state when a syscall trap was seen, and even then only with active 'catch syscall'. That can get out of sync if 'catch syscall' is toggled at odd times. This patch updates the entry/return state for all syscall traps, regardless of catching state, and also updates known syscall state for other kinds of traps. Almost all PTRACE_EVENT stops are delivered from the middle of a syscall, so this can act like an entry. Every other kind of ptrace stop is only delivered outside of syscall event pairs, so marking them ignored ensures the next syscall trap looks like an entry. Three new test scenarios are added to catch-syscall.exp: - Disable 'catch syscall' from an entry to deliberately miss the return event, then re-enable to make sure a new entry is recognized. - Enable 'catch syscall' for the first time from a vfork event, which is a PTRACE_EVENT_VFORK in the middle of the syscall. Make sure the next syscall event is recognized as the return. - Make sure entry and return are recognized for an ENOSYS syscall. This is to defeat a common x86 hack that uses the pre-filled ENOSYS return value as a sign of being on the entry side. gdb/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * linux-nat.c (linux_handle_syscall_trap): Always update entry/ return state, even when not actively catching syscalls at all. (linux_handle_extended_wait): Mark syscall_state like an entry. (wait_lwp): Set syscall_state ignored for other traps. (linux_nat_filter_event): Likewise. gdb/testsuite/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * gdb.base/catch-syscall.c: Include <sched.h>. (unknown_syscall): New variable. (main): Trigger a vfork and an unknown syscall. * gdb.base/catch-syscall.exp (vfork_syscalls): New variable. (unknown_syscall_number): Likewise. (check_call_to_syscall): Accept an optional syscall pattern. (check_return_from_syscall): Likewise. (check_continue): Likewise. (test_catch_syscall_without_args): Check for vfork and ENOSYS. (test_catch_syscall_skipping_return): New test toggling off 'catch syscall' to step over the syscall return, then toggling back on. (test_catch_syscall_mid_vfork): New test turning on 'catch syscall' during a PTRACE_EVENT_VFORK stop, in the middle of a vfork syscall. (do_syscall_tests): Call test_catch_syscall_without_args and test_catch_syscall_mid_vfork. (test_catch_syscall_without_args_noxml): Check for vfork and ENOSYS. (fill_all_syscalls_numbers): Initialize unknown_syscall_number.
2015-10-20 02:59:38 +02:00
check_continue "vfork" $vfork_syscalls
with_test_prefix "ENOSYS" {
check_continue $unknown_syscall_number
}
# At last but not least, we check if the inferior has called
# the last (exit) syscall.
check_call_to_syscall $last_syscall
# Now let's see if the inferior correctly finishes.
check_for_program_end
}
}
proc test_catch_syscall_with_args {} {
with_test_prefix "with arguments" {
set syscall_name "close"
insert_catch_syscall_with_arg $syscall_name
# Can we continue until we catch the syscall?
check_continue $syscall_name
# Now let's see if the inferior correctly finishes.
check_for_program_end
}
}
proc test_catch_syscall_with_many_args {} {
with_test_prefix "with many arguments" {
Improve and fix catch-syscall.exp While fixing another bug, I found that the current gdb.base/catch-syscall.exp is kind of messy, could use some improvements, and is not correctly testing some things. I've made the following patch to address all the issues I found. On the organization side, it does a cleanup and removes unecessary imports of gdb_prompt, uses prepare_for_testing and clean_restart where needed, and fixes some comments. The testcase was also not correctly testing catching syscalls using only numbers, or catching many syscalls at once. I fixed that. The patch also uses a new method for obtaining the syscalls numbers: it relies on the C source file to get them, via <sys/syscall.h> and SYS_* macros. This makes the .exp file simpler because there is no need to include target conditionals there. I tested this on x86_64 Fedora 18. gdb/testsuite/ChangeLog: 2013-12-18 Sergio Durigan Junior <sergiodj@redhat.com> * gdb.base/catch-syscall.c: Include <sys/syscall.h>. (close_syscall, chroot_syscall, exit_group_syscall): New variables. * gdb.base/catch-syscall.exp: Replace gdb_compile by prepare_for_testing. Call fill_all_syscalls_numbers before starting. Replace gdb_exit, gdb_start, gdb_reinitialize_dir and gdb_load by clean_restart. (check_info_bp_any_syscall, check_info_bp_specific_syscall) (check_info_bp_many_syscalls): Remove global gdb_prompt. (check_call_to_syscall): Likewise. Add global decimal. Improve testing regex. (check_return_from_syscall): Likewise. (check_continue, insert_catch_syscall_with_arg): Remove global gdb_prompt. (insert_catch_syscall_with_many_args): Likewise. Add global decimal. Fix $filter_str. Improve testing regex. (check_for_program_end): Remove global gdb_prompt. (test_catch_syscall_without_args): Likewise. Add global decimal. Improve testing regex. (test_catch_syscall_with_args, test_catch_syscall_with_many_args) (test_catch_syscall_with_wrong_args) (test_catch_syscall_restarting_inferior) (test_catch_syscall_fail_nodatadir): Remove global gdb_prompt. (do_syscall_tests): Likewise. Remove global srcdir. (test_catch_syscall_without_args_noxml): Remove global gdb_prompt. Add global last_syscall_number. Test for the exact syscall number to be caught. (test_catch_syscall_with_args_noxml): Remove global gdb_prompt. Add global all_syscalls_numbers. Test each syscall number to be caught, instead of only testing "close". (test_catch_syscall_with_wrong_args_noxml): Remove global gdb_prompt. (do_syscall_tests_without_xml): Likewise. Remove global srcdir. Remove stale comment. (fill_all_syscalls_numbers): Add global last_syscall_number. Fill the correct syscall numbers using information from the inferior.
2013-12-18 23:19:01 +01:00
global all_syscalls all_syscalls_numbers
insert_catch_syscall_with_many_args $all_syscalls $all_syscalls_numbers
# Can we continue until we catch the syscalls?
foreach name $all_syscalls {
check_continue $name
}
# Now let's see if the inferior correctly finishes.
check_for_program_end
}
}
proc test_catch_syscall_with_wrong_args {} {
with_test_prefix "wrong args" {
# mlock is not called from the source
set syscall_name "mlock"
insert_catch_syscall_with_arg $syscall_name
# Now, we must verify if the program stops with a continue.
# If it doesn't, everything is right (since we don't have
# a syscall named "mlock" in it). Otherwise, this is a failure.
set thistest "catch syscall with unused syscall ($syscall_name)"
gdb_continue_to_end $thistest
}
}
proc test_catch_syscall_restarting_inferior {} {
with_test_prefix "restarting inferior" {
set syscall_name "chroot"
with_test_prefix "entry" {
insert_catch_syscall_with_arg $syscall_name
# Let's first reach the entry of the syscall.
check_call_to_syscall $syscall_name
}
with_test_prefix "entry/return" {
# Now, restart the program.
rerun_to_main
# And check for entry/return.
check_continue $syscall_name
# Can we finish?
check_for_program_end
}
}
}
gdb: Improve syscall entry/return tracking on Linux The existing logic was simply to flip syscall entry/return state when a syscall trap was seen, and even then only with active 'catch syscall'. That can get out of sync if 'catch syscall' is toggled at odd times. This patch updates the entry/return state for all syscall traps, regardless of catching state, and also updates known syscall state for other kinds of traps. Almost all PTRACE_EVENT stops are delivered from the middle of a syscall, so this can act like an entry. Every other kind of ptrace stop is only delivered outside of syscall event pairs, so marking them ignored ensures the next syscall trap looks like an entry. Three new test scenarios are added to catch-syscall.exp: - Disable 'catch syscall' from an entry to deliberately miss the return event, then re-enable to make sure a new entry is recognized. - Enable 'catch syscall' for the first time from a vfork event, which is a PTRACE_EVENT_VFORK in the middle of the syscall. Make sure the next syscall event is recognized as the return. - Make sure entry and return are recognized for an ENOSYS syscall. This is to defeat a common x86 hack that uses the pre-filled ENOSYS return value as a sign of being on the entry side. gdb/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * linux-nat.c (linux_handle_syscall_trap): Always update entry/ return state, even when not actively catching syscalls at all. (linux_handle_extended_wait): Mark syscall_state like an entry. (wait_lwp): Set syscall_state ignored for other traps. (linux_nat_filter_event): Likewise. gdb/testsuite/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * gdb.base/catch-syscall.c: Include <sched.h>. (unknown_syscall): New variable. (main): Trigger a vfork and an unknown syscall. * gdb.base/catch-syscall.exp (vfork_syscalls): New variable. (unknown_syscall_number): Likewise. (check_call_to_syscall): Accept an optional syscall pattern. (check_return_from_syscall): Likewise. (check_continue): Likewise. (test_catch_syscall_without_args): Check for vfork and ENOSYS. (test_catch_syscall_skipping_return): New test toggling off 'catch syscall' to step over the syscall return, then toggling back on. (test_catch_syscall_mid_vfork): New test turning on 'catch syscall' during a PTRACE_EVENT_VFORK stop, in the middle of a vfork syscall. (do_syscall_tests): Call test_catch_syscall_without_args and test_catch_syscall_mid_vfork. (test_catch_syscall_without_args_noxml): Check for vfork and ENOSYS. (fill_all_syscalls_numbers): Initialize unknown_syscall_number.
2015-10-20 02:59:38 +02:00
proc test_catch_syscall_skipping_return {} {
with_test_prefix "skipping return" {
with_test_prefix "entry" {
set syscall_name "write"
insert_catch_syscall_with_arg $syscall_name
# Let's first reach the entry of the syscall.
check_call_to_syscall $syscall_name
# Now purposely skip the syscall return.
delete_breakpoints
gdb_test "stepi" ".*" "step over syscall return"
}
# With a naive entry/return toggle, gdb will still think
# the target is due for a syscall return.
with_test_prefix "entry/return" {
set syscall_name "read"
insert_catch_syscall_with_arg $syscall_name
# Check for entry first, then return.
check_continue $syscall_name
# Can we finish?
check_for_program_end
}
}
}
proc test_catch_syscall_mid_vfork {} {
global gdb_prompt decimal vfork_syscalls
with_test_prefix "mid-vfork" {
# Verify that the system supports "catch vfork".
gdb_test "catch vfork" "Catchpoint $decimal \\(vfork\\)" "insert first vfork catchpoint"
gdb_test_multiple "continue" "continue to first vfork catchpoint" {
-re ".*Your system does not support this type\r\nof catchpoint.*$gdb_prompt $" {
unsupported "continue to first vfork catchpoint"
return
}
-re ".*Catchpoint $decimal \\(vforked process $decimal\\).*$gdb_prompt $" {
pass "continue to first vfork catchpoint"
}
}
# Check that we now reach vfork return only.
# (The actual syscall used varies by architecture.)
gdb_test "catch syscall" "Catchpoint $decimal \\(any syscall\\)"
check_return_from_syscall "vfork" $vfork_syscalls
# Can we finish?
check_for_program_end
}
}
proc test_catch_syscall_fail_nodatadir {} {
with_test_prefix "fail no datadir" {
# Sanitizing.
delete_breakpoints
# Make sure GDB doesn't load the syscalls xml from the system
# data directory.
gdb_test "set data-directory /the/path/to/nowhere" \
"Warning: /the/path/to/nowhere: .*"
# Testing to see if we receive a warning when calling "catch
# syscall" without XML support (without datadir).
set thistest "catch syscall displays a warning when there is no XML support"
gdb_test "catch syscall" \
"warning: Could not load the syscall XML file.*warning: GDB will not be able to display syscall names nor to verify if.*any provided syscall numbers are valid.*Catchpoint .*(syscall).*" \
$thistest
# Since the catchpoint was set, we must check if it's present
# in "info breakpoints" output.
check_info_bp_any_syscall
# Sanitizing.
delete_breakpoints
}
}
proc do_syscall_tests {} {
# NOTE: We don't have to point gdb at the correct data-directory.
# For the build tree that is handled by INTERNAL_GDBFLAGS.
# Verify that the 'catch syscall' help is available
set thistest "help catch syscall"
gdb_test "help catch syscall" "Catch system calls.*" $thistest
# Try to set a catchpoint to a nonsense syscall
set thistest "catch syscall to a nonsense syscall is prohibited"
gdb_test "catch syscall nonsense_syscall" "Unknown syscall name .*" $thistest
# Regression test for syscall completer bug.
gdb_test "complete catch syscall close chroo" \
"catch syscall close chroot" \
"complete catch syscall with multiple words"
# Testing the 'catch syscall' command without arguments.
# This test should catch any syscalls.
if [runto_main] then { test_catch_syscall_without_args }
# Testing the 'catch syscall' command with arguments.
# This test should only catch the specified syscall.
if [runto_main] then { test_catch_syscall_with_args }
# Testing the 'catch syscall' command with many arguments.
# This test should catch $all_syscalls.
if [runto_main] then { test_catch_syscall_with_many_args }
# Testing the 'catch syscall' command with WRONG arguments.
# This test should not trigger any catchpoints.
if [runto_main] then { test_catch_syscall_with_wrong_args }
gdb: Improve syscall entry/return tracking on Linux The existing logic was simply to flip syscall entry/return state when a syscall trap was seen, and even then only with active 'catch syscall'. That can get out of sync if 'catch syscall' is toggled at odd times. This patch updates the entry/return state for all syscall traps, regardless of catching state, and also updates known syscall state for other kinds of traps. Almost all PTRACE_EVENT stops are delivered from the middle of a syscall, so this can act like an entry. Every other kind of ptrace stop is only delivered outside of syscall event pairs, so marking them ignored ensures the next syscall trap looks like an entry. Three new test scenarios are added to catch-syscall.exp: - Disable 'catch syscall' from an entry to deliberately miss the return event, then re-enable to make sure a new entry is recognized. - Enable 'catch syscall' for the first time from a vfork event, which is a PTRACE_EVENT_VFORK in the middle of the syscall. Make sure the next syscall event is recognized as the return. - Make sure entry and return are recognized for an ENOSYS syscall. This is to defeat a common x86 hack that uses the pre-filled ENOSYS return value as a sign of being on the entry side. gdb/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * linux-nat.c (linux_handle_syscall_trap): Always update entry/ return state, even when not actively catching syscalls at all. (linux_handle_extended_wait): Mark syscall_state like an entry. (wait_lwp): Set syscall_state ignored for other traps. (linux_nat_filter_event): Likewise. gdb/testsuite/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * gdb.base/catch-syscall.c: Include <sched.h>. (unknown_syscall): New variable. (main): Trigger a vfork and an unknown syscall. * gdb.base/catch-syscall.exp (vfork_syscalls): New variable. (unknown_syscall_number): Likewise. (check_call_to_syscall): Accept an optional syscall pattern. (check_return_from_syscall): Likewise. (check_continue): Likewise. (test_catch_syscall_without_args): Check for vfork and ENOSYS. (test_catch_syscall_skipping_return): New test toggling off 'catch syscall' to step over the syscall return, then toggling back on. (test_catch_syscall_mid_vfork): New test turning on 'catch syscall' during a PTRACE_EVENT_VFORK stop, in the middle of a vfork syscall. (do_syscall_tests): Call test_catch_syscall_without_args and test_catch_syscall_mid_vfork. (test_catch_syscall_without_args_noxml): Check for vfork and ENOSYS. (fill_all_syscalls_numbers): Initialize unknown_syscall_number.
2015-10-20 02:59:38 +02:00
# Testing the 'catch syscall' command during a restart of
# the inferior.
if [runto_main] then { test_catch_syscall_restarting_inferior }
Partial fix for PR breakpoints/10737: Make syscall info be per-arch instead of global This patch intends to partially fix PR breakpoints/10737, which is about making the syscall information (for the "catch syscall" command) be per-arch, instead of global. This is not a full fix because of the other issues pointed by Pedro here: <https://sourceware.org/bugzilla/show_bug.cgi?id=10737#c5> However, I consider it a good step towards the real fix. It will also help me fix <https://sourceware.org/bugzilla/show_bug.cgi?id=17402>. What this patch does, basically, is move the "syscalls_info" struct to gdbarch. Currently, the syscall information is stored in a global variable inside gdb/xml-syscall.c, which means that there is no easy way to correlate this info with the current target or architecture being used, for example. This causes strange behaviors, because the syscall info is not re-read when the arch changes. For example, if you put a syscall catchpoint in syscall 5 on i386 (syscall open), and then load a x86_64 program on GDB and put the same syscall 5 there (fstat on x86_64), you will still see that GDB tells you that it is catching "open", even though it is not. With this patch, GDB correctly says that it will be catching fstat syscalls. (gdb) set architecture i386 The target architecture is assumed to be i386 (gdb) catch syscall 5 Catchpoint 1 (syscall 'open' [5]) (gdb) set architecture i386:x86-64 The target architecture is assumed to be i386:x86-64 (gdb) catch syscall 5 Catchpoint 2 (syscall 'open' [5]) But with the patch: (gdb) set architecture i386 The target architecture is assumed to be i386 (gdb) catch syscall 5 Catchpoint 1 (syscall 'open' [5]) (gdb) set architecture i386:x86-64 The target architecture is assumed to be i386:x86-64 (gdb) catch syscall 5 Catchpoint 2 (syscall 'fstat' [5]) As I said, there are still some problems on the "catch syscall" mechanism, because (for example) the user should be able to "catch syscall open" on i386, and then expect "open" to be caught also on x86_64. Currently, it doesn't work. I intend to work on this later. gdb/ 2014-11-20 Sergio Durigan Junior <sergiodj@redhat.com> PR breakpoints/10737 * amd64-linux-tdep.c (amd64_linux_init_abi_common): Adjust call to set_xml_syscall_file_name to provide gdbarch. * arm-linux-tdep.c (arm_linux_init_abi): Likewise. * bfin-linux-tdep.c (bfin_linux_init_abi): Likewise. * breakpoint.c (print_it_catch_syscall): Adjust call to get_syscall_by_number to provide gdbarch. (print_one_catch_syscall): Likewise. (print_mention_catch_syscall): Likewise. (print_recreate_catch_syscall): Likewise. (catch_syscall_split_args): Adjust calls to get_syscall_by_number and get_syscall_by_name to provide gdbarch. (catch_syscall_completer): Adjust call to get_syscall_names to provide gdbarch. * gdbarch.c: Regenerate. * gdbarch.h: Likewise. * gdbarch.sh: Forward declare "struct syscalls_info". (xml_syscall_file): New variable. (syscalls_info): Likewise. * i386-linux-tdep.c (i386_linux_init_abi): Adjust call to set_xml_syscall_file_name to provide gdbarch. * mips-linux-tdep.c (mips_linux_init_abi): Likewise. * ppc-linux-tdep.c (ppc_linux_init_abi): Likewise. * s390-linux-tdep.c (s390_gdbarch_init): Likewise. * sparc-linux-tdep.c (sparc32_linux_init_abi): Likewise. * sparc64-linux-tdep.c (sparc64_linux_init_abi): Likewise. * xml-syscall.c: Include gdbarch.h. (set_xml_syscall_file_name): Accept gdbarch parameter. (get_syscall_by_number): Likewise. (get_syscall_by_name): Likewise. (get_syscall_names): Likewise. (my_gdb_datadir): Delete global variable. (struct syscalls_info) <my_gdb_datadir>: New variable. (struct syscalls_info) <sysinfo>: Rename variable to "syscalls_info". (sysinfo): Delete global variable. (have_initialized_sysinfo): Likewise. (xml_syscall_file): Likewise. (sysinfo_free_syscalls_desc): Rename to... (syscalls_info_free_syscalls_desc): ... this. (free_syscalls_info): Rename "sysinfo" to "syscalls_info". Adjust code to the new layout of "struct syscalls_info". (make_cleanup_free_syscalls_info): Rename parameter "sysinfo" to "syscalls_info". (syscall_create_syscall_desc): Likewise. (syscall_start_syscall): Likewise. (syscall_parse_xml): Likewise. (xml_init_syscalls_info): Likewise. Drop "const" from return value. (init_sysinfo): Rename to... (init_syscalls_info): ...this. Add gdbarch as a parameter. Adjust function to deal with gdbarch. (xml_get_syscall_number): Delete parameter sysinfo. Accept gdbarch as a parameter. Adjust code. (xml_get_syscall_name): Likewise. (xml_list_of_syscalls): Likewise. (set_xml_syscall_file_name): Accept gdbarch as parameter. (get_syscall_by_number): Likewise. (get_syscall_by_name): Likewise. (get_syscall_names): Likewise. * xml-syscall.h (set_xml_syscall_file_name): Likewise. (get_syscall_by_number): Likewise. (get_syscall_by_name): Likewise. (get_syscall_names): Likewise. gdb/testsuite/ 2014-11-20 Sergio Durigan Junior <sergiodj@redhat.com> PR breakpoints/10737 * gdb.base/catch-syscall.exp (do_syscall_tests): Call test_catch_syscall_multi_arch. (test_catch_syscall_multi_arch): New function.
2014-11-20 18:28:18 +01:00
gdb: Improve syscall entry/return tracking on Linux The existing logic was simply to flip syscall entry/return state when a syscall trap was seen, and even then only with active 'catch syscall'. That can get out of sync if 'catch syscall' is toggled at odd times. This patch updates the entry/return state for all syscall traps, regardless of catching state, and also updates known syscall state for other kinds of traps. Almost all PTRACE_EVENT stops are delivered from the middle of a syscall, so this can act like an entry. Every other kind of ptrace stop is only delivered outside of syscall event pairs, so marking them ignored ensures the next syscall trap looks like an entry. Three new test scenarios are added to catch-syscall.exp: - Disable 'catch syscall' from an entry to deliberately miss the return event, then re-enable to make sure a new entry is recognized. - Enable 'catch syscall' for the first time from a vfork event, which is a PTRACE_EVENT_VFORK in the middle of the syscall. Make sure the next syscall event is recognized as the return. - Make sure entry and return are recognized for an ENOSYS syscall. This is to defeat a common x86 hack that uses the pre-filled ENOSYS return value as a sign of being on the entry side. gdb/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * linux-nat.c (linux_handle_syscall_trap): Always update entry/ return state, even when not actively catching syscalls at all. (linux_handle_extended_wait): Mark syscall_state like an entry. (wait_lwp): Set syscall_state ignored for other traps. (linux_nat_filter_event): Likewise. gdb/testsuite/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * gdb.base/catch-syscall.c: Include <sched.h>. (unknown_syscall): New variable. (main): Trigger a vfork and an unknown syscall. * gdb.base/catch-syscall.exp (vfork_syscalls): New variable. (unknown_syscall_number): Likewise. (check_call_to_syscall): Accept an optional syscall pattern. (check_return_from_syscall): Likewise. (check_continue): Likewise. (test_catch_syscall_without_args): Check for vfork and ENOSYS. (test_catch_syscall_skipping_return): New test toggling off 'catch syscall' to step over the syscall return, then toggling back on. (test_catch_syscall_mid_vfork): New test turning on 'catch syscall' during a PTRACE_EVENT_VFORK stop, in the middle of a vfork syscall. (do_syscall_tests): Call test_catch_syscall_without_args and test_catch_syscall_mid_vfork. (test_catch_syscall_without_args_noxml): Check for vfork and ENOSYS. (fill_all_syscalls_numbers): Initialize unknown_syscall_number.
2015-10-20 02:59:38 +02:00
# Testing the 'catch syscall' command toggling off past a
# syscall return, then resuming entry/return as normal.
if [runto_main] then { test_catch_syscall_skipping_return }
# Testing the 'catch syscall' command starting mid-vfork.
if [runto_main] then { test_catch_syscall_mid_vfork }
Partial fix for PR breakpoints/10737: Make syscall info be per-arch instead of global This patch intends to partially fix PR breakpoints/10737, which is about making the syscall information (for the "catch syscall" command) be per-arch, instead of global. This is not a full fix because of the other issues pointed by Pedro here: <https://sourceware.org/bugzilla/show_bug.cgi?id=10737#c5> However, I consider it a good step towards the real fix. It will also help me fix <https://sourceware.org/bugzilla/show_bug.cgi?id=17402>. What this patch does, basically, is move the "syscalls_info" struct to gdbarch. Currently, the syscall information is stored in a global variable inside gdb/xml-syscall.c, which means that there is no easy way to correlate this info with the current target or architecture being used, for example. This causes strange behaviors, because the syscall info is not re-read when the arch changes. For example, if you put a syscall catchpoint in syscall 5 on i386 (syscall open), and then load a x86_64 program on GDB and put the same syscall 5 there (fstat on x86_64), you will still see that GDB tells you that it is catching "open", even though it is not. With this patch, GDB correctly says that it will be catching fstat syscalls. (gdb) set architecture i386 The target architecture is assumed to be i386 (gdb) catch syscall 5 Catchpoint 1 (syscall 'open' [5]) (gdb) set architecture i386:x86-64 The target architecture is assumed to be i386:x86-64 (gdb) catch syscall 5 Catchpoint 2 (syscall 'open' [5]) But with the patch: (gdb) set architecture i386 The target architecture is assumed to be i386 (gdb) catch syscall 5 Catchpoint 1 (syscall 'open' [5]) (gdb) set architecture i386:x86-64 The target architecture is assumed to be i386:x86-64 (gdb) catch syscall 5 Catchpoint 2 (syscall 'fstat' [5]) As I said, there are still some problems on the "catch syscall" mechanism, because (for example) the user should be able to "catch syscall open" on i386, and then expect "open" to be caught also on x86_64. Currently, it doesn't work. I intend to work on this later. gdb/ 2014-11-20 Sergio Durigan Junior <sergiodj@redhat.com> PR breakpoints/10737 * amd64-linux-tdep.c (amd64_linux_init_abi_common): Adjust call to set_xml_syscall_file_name to provide gdbarch. * arm-linux-tdep.c (arm_linux_init_abi): Likewise. * bfin-linux-tdep.c (bfin_linux_init_abi): Likewise. * breakpoint.c (print_it_catch_syscall): Adjust call to get_syscall_by_number to provide gdbarch. (print_one_catch_syscall): Likewise. (print_mention_catch_syscall): Likewise. (print_recreate_catch_syscall): Likewise. (catch_syscall_split_args): Adjust calls to get_syscall_by_number and get_syscall_by_name to provide gdbarch. (catch_syscall_completer): Adjust call to get_syscall_names to provide gdbarch. * gdbarch.c: Regenerate. * gdbarch.h: Likewise. * gdbarch.sh: Forward declare "struct syscalls_info". (xml_syscall_file): New variable. (syscalls_info): Likewise. * i386-linux-tdep.c (i386_linux_init_abi): Adjust call to set_xml_syscall_file_name to provide gdbarch. * mips-linux-tdep.c (mips_linux_init_abi): Likewise. * ppc-linux-tdep.c (ppc_linux_init_abi): Likewise. * s390-linux-tdep.c (s390_gdbarch_init): Likewise. * sparc-linux-tdep.c (sparc32_linux_init_abi): Likewise. * sparc64-linux-tdep.c (sparc64_linux_init_abi): Likewise. * xml-syscall.c: Include gdbarch.h. (set_xml_syscall_file_name): Accept gdbarch parameter. (get_syscall_by_number): Likewise. (get_syscall_by_name): Likewise. (get_syscall_names): Likewise. (my_gdb_datadir): Delete global variable. (struct syscalls_info) <my_gdb_datadir>: New variable. (struct syscalls_info) <sysinfo>: Rename variable to "syscalls_info". (sysinfo): Delete global variable. (have_initialized_sysinfo): Likewise. (xml_syscall_file): Likewise. (sysinfo_free_syscalls_desc): Rename to... (syscalls_info_free_syscalls_desc): ... this. (free_syscalls_info): Rename "sysinfo" to "syscalls_info". Adjust code to the new layout of "struct syscalls_info". (make_cleanup_free_syscalls_info): Rename parameter "sysinfo" to "syscalls_info". (syscall_create_syscall_desc): Likewise. (syscall_start_syscall): Likewise. (syscall_parse_xml): Likewise. (xml_init_syscalls_info): Likewise. Drop "const" from return value. (init_sysinfo): Rename to... (init_syscalls_info): ...this. Add gdbarch as a parameter. Adjust function to deal with gdbarch. (xml_get_syscall_number): Delete parameter sysinfo. Accept gdbarch as a parameter. Adjust code. (xml_get_syscall_name): Likewise. (xml_list_of_syscalls): Likewise. (set_xml_syscall_file_name): Accept gdbarch as parameter. (get_syscall_by_number): Likewise. (get_syscall_by_name): Likewise. (get_syscall_names): Likewise. * xml-syscall.h (set_xml_syscall_file_name): Likewise. (get_syscall_by_number): Likewise. (get_syscall_by_name): Likewise. (get_syscall_names): Likewise. gdb/testsuite/ 2014-11-20 Sergio Durigan Junior <sergiodj@redhat.com> PR breakpoints/10737 * gdb.base/catch-syscall.exp (do_syscall_tests): Call test_catch_syscall_multi_arch. (test_catch_syscall_multi_arch): New function.
2014-11-20 18:28:18 +01:00
# Testing if the 'catch syscall' command works when switching to
# different architectures on-the-fly (PR gdb/10737).
if [runto_main] then { test_catch_syscall_multi_arch }
}
proc test_catch_syscall_without_args_noxml {} {
with_test_prefix "without args noxml" {
# We will need the syscall names even not using it because we
# need to know know many syscalls are in the example file.
gdb: Improve syscall entry/return tracking on Linux The existing logic was simply to flip syscall entry/return state when a syscall trap was seen, and even then only with active 'catch syscall'. That can get out of sync if 'catch syscall' is toggled at odd times. This patch updates the entry/return state for all syscall traps, regardless of catching state, and also updates known syscall state for other kinds of traps. Almost all PTRACE_EVENT stops are delivered from the middle of a syscall, so this can act like an entry. Every other kind of ptrace stop is only delivered outside of syscall event pairs, so marking them ignored ensures the next syscall trap looks like an entry. Three new test scenarios are added to catch-syscall.exp: - Disable 'catch syscall' from an entry to deliberately miss the return event, then re-enable to make sure a new entry is recognized. - Enable 'catch syscall' for the first time from a vfork event, which is a PTRACE_EVENT_VFORK in the middle of the syscall. Make sure the next syscall event is recognized as the return. - Make sure entry and return are recognized for an ENOSYS syscall. This is to defeat a common x86 hack that uses the pre-filled ENOSYS return value as a sign of being on the entry side. gdb/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * linux-nat.c (linux_handle_syscall_trap): Always update entry/ return state, even when not actively catching syscalls at all. (linux_handle_extended_wait): Mark syscall_state like an entry. (wait_lwp): Set syscall_state ignored for other traps. (linux_nat_filter_event): Likewise. gdb/testsuite/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * gdb.base/catch-syscall.c: Include <sched.h>. (unknown_syscall): New variable. (main): Trigger a vfork and an unknown syscall. * gdb.base/catch-syscall.exp (vfork_syscalls): New variable. (unknown_syscall_number): Likewise. (check_call_to_syscall): Accept an optional syscall pattern. (check_return_from_syscall): Likewise. (check_continue): Likewise. (test_catch_syscall_without_args): Check for vfork and ENOSYS. (test_catch_syscall_skipping_return): New test toggling off 'catch syscall' to step over the syscall return, then toggling back on. (test_catch_syscall_mid_vfork): New test turning on 'catch syscall' during a PTRACE_EVENT_VFORK stop, in the middle of a vfork syscall. (do_syscall_tests): Call test_catch_syscall_without_args and test_catch_syscall_mid_vfork. (test_catch_syscall_without_args_noxml): Check for vfork and ENOSYS. (fill_all_syscalls_numbers): Initialize unknown_syscall_number.
2015-10-20 02:59:38 +02:00
global decimal all_syscalls last_syscall_number unknown_syscall_number all_syscalls_numbers
delete_breakpoints
gdb_test "catch syscall" "Catchpoint .*(syscall).*"
# Now, we should be able to set a catchpoint, and GDB shall
# not display the warning anymore.
Improve and fix catch-syscall.exp While fixing another bug, I found that the current gdb.base/catch-syscall.exp is kind of messy, could use some improvements, and is not correctly testing some things. I've made the following patch to address all the issues I found. On the organization side, it does a cleanup and removes unecessary imports of gdb_prompt, uses prepare_for_testing and clean_restart where needed, and fixes some comments. The testcase was also not correctly testing catching syscalls using only numbers, or catching many syscalls at once. I fixed that. The patch also uses a new method for obtaining the syscalls numbers: it relies on the C source file to get them, via <sys/syscall.h> and SYS_* macros. This makes the .exp file simpler because there is no need to include target conditionals there. I tested this on x86_64 Fedora 18. gdb/testsuite/ChangeLog: 2013-12-18 Sergio Durigan Junior <sergiodj@redhat.com> * gdb.base/catch-syscall.c: Include <sys/syscall.h>. (close_syscall, chroot_syscall, exit_group_syscall): New variables. * gdb.base/catch-syscall.exp: Replace gdb_compile by prepare_for_testing. Call fill_all_syscalls_numbers before starting. Replace gdb_exit, gdb_start, gdb_reinitialize_dir and gdb_load by clean_restart. (check_info_bp_any_syscall, check_info_bp_specific_syscall) (check_info_bp_many_syscalls): Remove global gdb_prompt. (check_call_to_syscall): Likewise. Add global decimal. Improve testing regex. (check_return_from_syscall): Likewise. (check_continue, insert_catch_syscall_with_arg): Remove global gdb_prompt. (insert_catch_syscall_with_many_args): Likewise. Add global decimal. Fix $filter_str. Improve testing regex. (check_for_program_end): Remove global gdb_prompt. (test_catch_syscall_without_args): Likewise. Add global decimal. Improve testing regex. (test_catch_syscall_with_args, test_catch_syscall_with_many_args) (test_catch_syscall_with_wrong_args) (test_catch_syscall_restarting_inferior) (test_catch_syscall_fail_nodatadir): Remove global gdb_prompt. (do_syscall_tests): Likewise. Remove global srcdir. (test_catch_syscall_without_args_noxml): Remove global gdb_prompt. Add global last_syscall_number. Test for the exact syscall number to be caught. (test_catch_syscall_with_args_noxml): Remove global gdb_prompt. Add global all_syscalls_numbers. Test each syscall number to be caught, instead of only testing "close". (test_catch_syscall_with_wrong_args_noxml): Remove global gdb_prompt. (do_syscall_tests_without_xml): Likewise. Remove global srcdir. Remove stale comment. (fill_all_syscalls_numbers): Add global last_syscall_number. Fill the correct syscall numbers using information from the inferior.
2013-12-18 23:19:01 +01:00
foreach name $all_syscalls number $all_syscalls_numbers {
with_test_prefix "$name" {
Improve and fix catch-syscall.exp While fixing another bug, I found that the current gdb.base/catch-syscall.exp is kind of messy, could use some improvements, and is not correctly testing some things. I've made the following patch to address all the issues I found. On the organization side, it does a cleanup and removes unecessary imports of gdb_prompt, uses prepare_for_testing and clean_restart where needed, and fixes some comments. The testcase was also not correctly testing catching syscalls using only numbers, or catching many syscalls at once. I fixed that. The patch also uses a new method for obtaining the syscalls numbers: it relies on the C source file to get them, via <sys/syscall.h> and SYS_* macros. This makes the .exp file simpler because there is no need to include target conditionals there. I tested this on x86_64 Fedora 18. gdb/testsuite/ChangeLog: 2013-12-18 Sergio Durigan Junior <sergiodj@redhat.com> * gdb.base/catch-syscall.c: Include <sys/syscall.h>. (close_syscall, chroot_syscall, exit_group_syscall): New variables. * gdb.base/catch-syscall.exp: Replace gdb_compile by prepare_for_testing. Call fill_all_syscalls_numbers before starting. Replace gdb_exit, gdb_start, gdb_reinitialize_dir and gdb_load by clean_restart. (check_info_bp_any_syscall, check_info_bp_specific_syscall) (check_info_bp_many_syscalls): Remove global gdb_prompt. (check_call_to_syscall): Likewise. Add global decimal. Improve testing regex. (check_return_from_syscall): Likewise. (check_continue, insert_catch_syscall_with_arg): Remove global gdb_prompt. (insert_catch_syscall_with_many_args): Likewise. Add global decimal. Fix $filter_str. Improve testing regex. (check_for_program_end): Remove global gdb_prompt. (test_catch_syscall_without_args): Likewise. Add global decimal. Improve testing regex. (test_catch_syscall_with_args, test_catch_syscall_with_many_args) (test_catch_syscall_with_wrong_args) (test_catch_syscall_restarting_inferior) (test_catch_syscall_fail_nodatadir): Remove global gdb_prompt. (do_syscall_tests): Likewise. Remove global srcdir. (test_catch_syscall_without_args_noxml): Remove global gdb_prompt. Add global last_syscall_number. Test for the exact syscall number to be caught. (test_catch_syscall_with_args_noxml): Remove global gdb_prompt. Add global all_syscalls_numbers. Test each syscall number to be caught, instead of only testing "close". (test_catch_syscall_with_wrong_args_noxml): Remove global gdb_prompt. (do_syscall_tests_without_xml): Likewise. Remove global srcdir. Remove stale comment. (fill_all_syscalls_numbers): Add global last_syscall_number. Fill the correct syscall numbers using information from the inferior.
2013-12-18 23:19:01 +01:00
check_continue $number
}
}
gdb: Improve syscall entry/return tracking on Linux The existing logic was simply to flip syscall entry/return state when a syscall trap was seen, and even then only with active 'catch syscall'. That can get out of sync if 'catch syscall' is toggled at odd times. This patch updates the entry/return state for all syscall traps, regardless of catching state, and also updates known syscall state for other kinds of traps. Almost all PTRACE_EVENT stops are delivered from the middle of a syscall, so this can act like an entry. Every other kind of ptrace stop is only delivered outside of syscall event pairs, so marking them ignored ensures the next syscall trap looks like an entry. Three new test scenarios are added to catch-syscall.exp: - Disable 'catch syscall' from an entry to deliberately miss the return event, then re-enable to make sure a new entry is recognized. - Enable 'catch syscall' for the first time from a vfork event, which is a PTRACE_EVENT_VFORK in the middle of the syscall. Make sure the next syscall event is recognized as the return. - Make sure entry and return are recognized for an ENOSYS syscall. This is to defeat a common x86 hack that uses the pre-filled ENOSYS return value as a sign of being on the entry side. gdb/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * linux-nat.c (linux_handle_syscall_trap): Always update entry/ return state, even when not actively catching syscalls at all. (linux_handle_extended_wait): Mark syscall_state like an entry. (wait_lwp): Set syscall_state ignored for other traps. (linux_nat_filter_event): Likewise. gdb/testsuite/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * gdb.base/catch-syscall.c: Include <sched.h>. (unknown_syscall): New variable. (main): Trigger a vfork and an unknown syscall. * gdb.base/catch-syscall.exp (vfork_syscalls): New variable. (unknown_syscall_number): Likewise. (check_call_to_syscall): Accept an optional syscall pattern. (check_return_from_syscall): Likewise. (check_continue): Likewise. (test_catch_syscall_without_args): Check for vfork and ENOSYS. (test_catch_syscall_skipping_return): New test toggling off 'catch syscall' to step over the syscall return, then toggling back on. (test_catch_syscall_mid_vfork): New test turning on 'catch syscall' during a PTRACE_EVENT_VFORK stop, in the middle of a vfork syscall. (do_syscall_tests): Call test_catch_syscall_without_args and test_catch_syscall_mid_vfork. (test_catch_syscall_without_args_noxml): Check for vfork and ENOSYS. (fill_all_syscalls_numbers): Initialize unknown_syscall_number.
2015-10-20 02:59:38 +02:00
check_continue "vfork" $decimal
with_test_prefix "ENOSYS" {
check_continue $unknown_syscall_number
}
# At last but not least, we check if the inferior has called
# the last (exit) syscall.
Improve and fix catch-syscall.exp While fixing another bug, I found that the current gdb.base/catch-syscall.exp is kind of messy, could use some improvements, and is not correctly testing some things. I've made the following patch to address all the issues I found. On the organization side, it does a cleanup and removes unecessary imports of gdb_prompt, uses prepare_for_testing and clean_restart where needed, and fixes some comments. The testcase was also not correctly testing catching syscalls using only numbers, or catching many syscalls at once. I fixed that. The patch also uses a new method for obtaining the syscalls numbers: it relies on the C source file to get them, via <sys/syscall.h> and SYS_* macros. This makes the .exp file simpler because there is no need to include target conditionals there. I tested this on x86_64 Fedora 18. gdb/testsuite/ChangeLog: 2013-12-18 Sergio Durigan Junior <sergiodj@redhat.com> * gdb.base/catch-syscall.c: Include <sys/syscall.h>. (close_syscall, chroot_syscall, exit_group_syscall): New variables. * gdb.base/catch-syscall.exp: Replace gdb_compile by prepare_for_testing. Call fill_all_syscalls_numbers before starting. Replace gdb_exit, gdb_start, gdb_reinitialize_dir and gdb_load by clean_restart. (check_info_bp_any_syscall, check_info_bp_specific_syscall) (check_info_bp_many_syscalls): Remove global gdb_prompt. (check_call_to_syscall): Likewise. Add global decimal. Improve testing regex. (check_return_from_syscall): Likewise. (check_continue, insert_catch_syscall_with_arg): Remove global gdb_prompt. (insert_catch_syscall_with_many_args): Likewise. Add global decimal. Fix $filter_str. Improve testing regex. (check_for_program_end): Remove global gdb_prompt. (test_catch_syscall_without_args): Likewise. Add global decimal. Improve testing regex. (test_catch_syscall_with_args, test_catch_syscall_with_many_args) (test_catch_syscall_with_wrong_args) (test_catch_syscall_restarting_inferior) (test_catch_syscall_fail_nodatadir): Remove global gdb_prompt. (do_syscall_tests): Likewise. Remove global srcdir. (test_catch_syscall_without_args_noxml): Remove global gdb_prompt. Add global last_syscall_number. Test for the exact syscall number to be caught. (test_catch_syscall_with_args_noxml): Remove global gdb_prompt. Add global all_syscalls_numbers. Test each syscall number to be caught, instead of only testing "close". (test_catch_syscall_with_wrong_args_noxml): Remove global gdb_prompt. (do_syscall_tests_without_xml): Likewise. Remove global srcdir. Remove stale comment. (fill_all_syscalls_numbers): Add global last_syscall_number. Fill the correct syscall numbers using information from the inferior.
2013-12-18 23:19:01 +01:00
check_call_to_syscall $last_syscall_number
delete_breakpoints
}
}
proc test_catch_syscall_with_args_noxml {} {
with_test_prefix "with args noxml" {
Improve and fix catch-syscall.exp While fixing another bug, I found that the current gdb.base/catch-syscall.exp is kind of messy, could use some improvements, and is not correctly testing some things. I've made the following patch to address all the issues I found. On the organization side, it does a cleanup and removes unecessary imports of gdb_prompt, uses prepare_for_testing and clean_restart where needed, and fixes some comments. The testcase was also not correctly testing catching syscalls using only numbers, or catching many syscalls at once. I fixed that. The patch also uses a new method for obtaining the syscalls numbers: it relies on the C source file to get them, via <sys/syscall.h> and SYS_* macros. This makes the .exp file simpler because there is no need to include target conditionals there. I tested this on x86_64 Fedora 18. gdb/testsuite/ChangeLog: 2013-12-18 Sergio Durigan Junior <sergiodj@redhat.com> * gdb.base/catch-syscall.c: Include <sys/syscall.h>. (close_syscall, chroot_syscall, exit_group_syscall): New variables. * gdb.base/catch-syscall.exp: Replace gdb_compile by prepare_for_testing. Call fill_all_syscalls_numbers before starting. Replace gdb_exit, gdb_start, gdb_reinitialize_dir and gdb_load by clean_restart. (check_info_bp_any_syscall, check_info_bp_specific_syscall) (check_info_bp_many_syscalls): Remove global gdb_prompt. (check_call_to_syscall): Likewise. Add global decimal. Improve testing regex. (check_return_from_syscall): Likewise. (check_continue, insert_catch_syscall_with_arg): Remove global gdb_prompt. (insert_catch_syscall_with_many_args): Likewise. Add global decimal. Fix $filter_str. Improve testing regex. (check_for_program_end): Remove global gdb_prompt. (test_catch_syscall_without_args): Likewise. Add global decimal. Improve testing regex. (test_catch_syscall_with_args, test_catch_syscall_with_many_args) (test_catch_syscall_with_wrong_args) (test_catch_syscall_restarting_inferior) (test_catch_syscall_fail_nodatadir): Remove global gdb_prompt. (do_syscall_tests): Likewise. Remove global srcdir. (test_catch_syscall_without_args_noxml): Remove global gdb_prompt. Add global last_syscall_number. Test for the exact syscall number to be caught. (test_catch_syscall_with_args_noxml): Remove global gdb_prompt. Add global all_syscalls_numbers. Test each syscall number to be caught, instead of only testing "close". (test_catch_syscall_with_wrong_args_noxml): Remove global gdb_prompt. (do_syscall_tests_without_xml): Likewise. Remove global srcdir. Remove stale comment. (fill_all_syscalls_numbers): Add global last_syscall_number. Fill the correct syscall numbers using information from the inferior.
2013-12-18 23:19:01 +01:00
global all_syscalls_numbers
delete_breakpoints
Improve and fix catch-syscall.exp While fixing another bug, I found that the current gdb.base/catch-syscall.exp is kind of messy, could use some improvements, and is not correctly testing some things. I've made the following patch to address all the issues I found. On the organization side, it does a cleanup and removes unecessary imports of gdb_prompt, uses prepare_for_testing and clean_restart where needed, and fixes some comments. The testcase was also not correctly testing catching syscalls using only numbers, or catching many syscalls at once. I fixed that. The patch also uses a new method for obtaining the syscalls numbers: it relies on the C source file to get them, via <sys/syscall.h> and SYS_* macros. This makes the .exp file simpler because there is no need to include target conditionals there. I tested this on x86_64 Fedora 18. gdb/testsuite/ChangeLog: 2013-12-18 Sergio Durigan Junior <sergiodj@redhat.com> * gdb.base/catch-syscall.c: Include <sys/syscall.h>. (close_syscall, chroot_syscall, exit_group_syscall): New variables. * gdb.base/catch-syscall.exp: Replace gdb_compile by prepare_for_testing. Call fill_all_syscalls_numbers before starting. Replace gdb_exit, gdb_start, gdb_reinitialize_dir and gdb_load by clean_restart. (check_info_bp_any_syscall, check_info_bp_specific_syscall) (check_info_bp_many_syscalls): Remove global gdb_prompt. (check_call_to_syscall): Likewise. Add global decimal. Improve testing regex. (check_return_from_syscall): Likewise. (check_continue, insert_catch_syscall_with_arg): Remove global gdb_prompt. (insert_catch_syscall_with_many_args): Likewise. Add global decimal. Fix $filter_str. Improve testing regex. (check_for_program_end): Remove global gdb_prompt. (test_catch_syscall_without_args): Likewise. Add global decimal. Improve testing regex. (test_catch_syscall_with_args, test_catch_syscall_with_many_args) (test_catch_syscall_with_wrong_args) (test_catch_syscall_restarting_inferior) (test_catch_syscall_fail_nodatadir): Remove global gdb_prompt. (do_syscall_tests): Likewise. Remove global srcdir. (test_catch_syscall_without_args_noxml): Remove global gdb_prompt. Add global last_syscall_number. Test for the exact syscall number to be caught. (test_catch_syscall_with_args_noxml): Remove global gdb_prompt. Add global all_syscalls_numbers. Test each syscall number to be caught, instead of only testing "close". (test_catch_syscall_with_wrong_args_noxml): Remove global gdb_prompt. (do_syscall_tests_without_xml): Likewise. Remove global srcdir. Remove stale comment. (fill_all_syscalls_numbers): Add global last_syscall_number. Fill the correct syscall numbers using information from the inferior.
2013-12-18 23:19:01 +01:00
# Inserting all syscalls numbers to be caught
foreach syscall_number $all_syscalls_numbers {
insert_catch_syscall_with_arg $syscall_number
}
Improve and fix catch-syscall.exp While fixing another bug, I found that the current gdb.base/catch-syscall.exp is kind of messy, could use some improvements, and is not correctly testing some things. I've made the following patch to address all the issues I found. On the organization side, it does a cleanup and removes unecessary imports of gdb_prompt, uses prepare_for_testing and clean_restart where needed, and fixes some comments. The testcase was also not correctly testing catching syscalls using only numbers, or catching many syscalls at once. I fixed that. The patch also uses a new method for obtaining the syscalls numbers: it relies on the C source file to get them, via <sys/syscall.h> and SYS_* macros. This makes the .exp file simpler because there is no need to include target conditionals there. I tested this on x86_64 Fedora 18. gdb/testsuite/ChangeLog: 2013-12-18 Sergio Durigan Junior <sergiodj@redhat.com> * gdb.base/catch-syscall.c: Include <sys/syscall.h>. (close_syscall, chroot_syscall, exit_group_syscall): New variables. * gdb.base/catch-syscall.exp: Replace gdb_compile by prepare_for_testing. Call fill_all_syscalls_numbers before starting. Replace gdb_exit, gdb_start, gdb_reinitialize_dir and gdb_load by clean_restart. (check_info_bp_any_syscall, check_info_bp_specific_syscall) (check_info_bp_many_syscalls): Remove global gdb_prompt. (check_call_to_syscall): Likewise. Add global decimal. Improve testing regex. (check_return_from_syscall): Likewise. (check_continue, insert_catch_syscall_with_arg): Remove global gdb_prompt. (insert_catch_syscall_with_many_args): Likewise. Add global decimal. Fix $filter_str. Improve testing regex. (check_for_program_end): Remove global gdb_prompt. (test_catch_syscall_without_args): Likewise. Add global decimal. Improve testing regex. (test_catch_syscall_with_args, test_catch_syscall_with_many_args) (test_catch_syscall_with_wrong_args) (test_catch_syscall_restarting_inferior) (test_catch_syscall_fail_nodatadir): Remove global gdb_prompt. (do_syscall_tests): Likewise. Remove global srcdir. (test_catch_syscall_without_args_noxml): Remove global gdb_prompt. Add global last_syscall_number. Test for the exact syscall number to be caught. (test_catch_syscall_with_args_noxml): Remove global gdb_prompt. Add global all_syscalls_numbers. Test each syscall number to be caught, instead of only testing "close". (test_catch_syscall_with_wrong_args_noxml): Remove global gdb_prompt. (do_syscall_tests_without_xml): Likewise. Remove global srcdir. Remove stale comment. (fill_all_syscalls_numbers): Add global last_syscall_number. Fill the correct syscall numbers using information from the inferior.
2013-12-18 23:19:01 +01:00
# Checking that all syscalls are caught.
foreach syscall_number $all_syscalls_numbers {
check_continue $syscall_number
}
delete_breakpoints
}
}
proc test_catch_syscall_with_wrong_args_noxml {} {
with_test_prefix "with wrong args noxml" {
delete_breakpoints
# Even without XML support, GDB should not accept unknown
# syscall names for the catchpoint.
gdb_test "catch syscall nonsense_syscall" \
"Unknown syscall name .nonsense_syscall.*"
delete_breakpoints
}
}
Partial fix for PR breakpoints/10737: Make syscall info be per-arch instead of global This patch intends to partially fix PR breakpoints/10737, which is about making the syscall information (for the "catch syscall" command) be per-arch, instead of global. This is not a full fix because of the other issues pointed by Pedro here: <https://sourceware.org/bugzilla/show_bug.cgi?id=10737#c5> However, I consider it a good step towards the real fix. It will also help me fix <https://sourceware.org/bugzilla/show_bug.cgi?id=17402>. What this patch does, basically, is move the "syscalls_info" struct to gdbarch. Currently, the syscall information is stored in a global variable inside gdb/xml-syscall.c, which means that there is no easy way to correlate this info with the current target or architecture being used, for example. This causes strange behaviors, because the syscall info is not re-read when the arch changes. For example, if you put a syscall catchpoint in syscall 5 on i386 (syscall open), and then load a x86_64 program on GDB and put the same syscall 5 there (fstat on x86_64), you will still see that GDB tells you that it is catching "open", even though it is not. With this patch, GDB correctly says that it will be catching fstat syscalls. (gdb) set architecture i386 The target architecture is assumed to be i386 (gdb) catch syscall 5 Catchpoint 1 (syscall 'open' [5]) (gdb) set architecture i386:x86-64 The target architecture is assumed to be i386:x86-64 (gdb) catch syscall 5 Catchpoint 2 (syscall 'open' [5]) But with the patch: (gdb) set architecture i386 The target architecture is assumed to be i386 (gdb) catch syscall 5 Catchpoint 1 (syscall 'open' [5]) (gdb) set architecture i386:x86-64 The target architecture is assumed to be i386:x86-64 (gdb) catch syscall 5 Catchpoint 2 (syscall 'fstat' [5]) As I said, there are still some problems on the "catch syscall" mechanism, because (for example) the user should be able to "catch syscall open" on i386, and then expect "open" to be caught also on x86_64. Currently, it doesn't work. I intend to work on this later. gdb/ 2014-11-20 Sergio Durigan Junior <sergiodj@redhat.com> PR breakpoints/10737 * amd64-linux-tdep.c (amd64_linux_init_abi_common): Adjust call to set_xml_syscall_file_name to provide gdbarch. * arm-linux-tdep.c (arm_linux_init_abi): Likewise. * bfin-linux-tdep.c (bfin_linux_init_abi): Likewise. * breakpoint.c (print_it_catch_syscall): Adjust call to get_syscall_by_number to provide gdbarch. (print_one_catch_syscall): Likewise. (print_mention_catch_syscall): Likewise. (print_recreate_catch_syscall): Likewise. (catch_syscall_split_args): Adjust calls to get_syscall_by_number and get_syscall_by_name to provide gdbarch. (catch_syscall_completer): Adjust call to get_syscall_names to provide gdbarch. * gdbarch.c: Regenerate. * gdbarch.h: Likewise. * gdbarch.sh: Forward declare "struct syscalls_info". (xml_syscall_file): New variable. (syscalls_info): Likewise. * i386-linux-tdep.c (i386_linux_init_abi): Adjust call to set_xml_syscall_file_name to provide gdbarch. * mips-linux-tdep.c (mips_linux_init_abi): Likewise. * ppc-linux-tdep.c (ppc_linux_init_abi): Likewise. * s390-linux-tdep.c (s390_gdbarch_init): Likewise. * sparc-linux-tdep.c (sparc32_linux_init_abi): Likewise. * sparc64-linux-tdep.c (sparc64_linux_init_abi): Likewise. * xml-syscall.c: Include gdbarch.h. (set_xml_syscall_file_name): Accept gdbarch parameter. (get_syscall_by_number): Likewise. (get_syscall_by_name): Likewise. (get_syscall_names): Likewise. (my_gdb_datadir): Delete global variable. (struct syscalls_info) <my_gdb_datadir>: New variable. (struct syscalls_info) <sysinfo>: Rename variable to "syscalls_info". (sysinfo): Delete global variable. (have_initialized_sysinfo): Likewise. (xml_syscall_file): Likewise. (sysinfo_free_syscalls_desc): Rename to... (syscalls_info_free_syscalls_desc): ... this. (free_syscalls_info): Rename "sysinfo" to "syscalls_info". Adjust code to the new layout of "struct syscalls_info". (make_cleanup_free_syscalls_info): Rename parameter "sysinfo" to "syscalls_info". (syscall_create_syscall_desc): Likewise. (syscall_start_syscall): Likewise. (syscall_parse_xml): Likewise. (xml_init_syscalls_info): Likewise. Drop "const" from return value. (init_sysinfo): Rename to... (init_syscalls_info): ...this. Add gdbarch as a parameter. Adjust function to deal with gdbarch. (xml_get_syscall_number): Delete parameter sysinfo. Accept gdbarch as a parameter. Adjust code. (xml_get_syscall_name): Likewise. (xml_list_of_syscalls): Likewise. (set_xml_syscall_file_name): Accept gdbarch as parameter. (get_syscall_by_number): Likewise. (get_syscall_by_name): Likewise. (get_syscall_names): Likewise. * xml-syscall.h (set_xml_syscall_file_name): Likewise. (get_syscall_by_number): Likewise. (get_syscall_by_name): Likewise. (get_syscall_names): Likewise. gdb/testsuite/ 2014-11-20 Sergio Durigan Junior <sergiodj@redhat.com> PR breakpoints/10737 * gdb.base/catch-syscall.exp (do_syscall_tests): Call test_catch_syscall_multi_arch. (test_catch_syscall_multi_arch): New function.
2014-11-20 18:28:18 +01:00
proc test_catch_syscall_multi_arch {} {
global decimal binfile
if { [istarget "i*86-*-*"] || [istarget "x86_64-*-*"] } {
set arch1 "i386"
set arch2 "i386:x86-64"
set syscall1_name "exit"
set syscall2_name "write"
set syscall_number 1
} elseif { [istarget "powerpc-*-linux*"] \
|| [istarget "powerpc64-*-linux*"] } {
set arch1 "powerpc:common"
set arch2 "powerpc:common64"
set syscall1_name "openat"
set syscall2_name "unlinkat"
set syscall_number 286
} elseif { [istarget "sparc-*-linux*"] \
|| [istarget "sparc64-*-linux*"] } {
set arch1 "sparc"
set arch2 "sparc:v9"
set syscall1_name "setresuid32"
set syscall2_name "setresuid"
set syscall_number 108
} elseif { [istarget "mips*-linux*"] } {
# MIPS does not use the same numbers for syscalls on 32 and 64
# bits.
verbose "Not testing MIPS for multi-arch syscall support"
return
} elseif { [istarget "arm*-linux*"] } {
# catch syscall supports only 32-bit ARM for now.
verbose "Not testing ARM for multi-arch syscall support"
return
Support catch syscall on aarch64 linux Hi, This patch is to support catch syscall on aarch64 linux. We implement gdbarch method get_syscall_number for aarch64-linux, and add aarch64-linux.xml file, which looks straightforward, however the changes to test case doesn't. First of all, we enable catch-syscall.exp on aarch64-linux target, but skip the multi_arch testing on current stage. I plan to touch multi arch debugging on aarch64-linux later. Then, when I run catch-syscall.exp on aarch64-linux, gcc errors that SYS_pipe isn't defined. We find that aarch64 kernel only has pipe2 syscall and libc already convert pipe to pipe2. As a result, I change catch-syscall.c to use SYS_pipe if it is defined, otherwise use SYS_pipe2 instead. The vector all_syscalls in catch-syscall.exp can't be pre-determined, so I add a new proc setup_all_syscalls to fill it, according to the availability of SYS_pipe. Regression tested on {x86_64, aarch64}-linux x {native, gdbserver}. gdb: 2015-03-18 Yao Qi <yao.qi@linaro.org> PR tdep/18107 * aarch64-linux-tdep.c: Include xml-syscall.h (aarch64_linux_get_syscall_number): New function. (aarch64_linux_init_abi): Call set_gdbarch_get_syscall_number. * syscalls/aarch64-linux.xml: New file. gdb/testsuite: 2015-03-18 Yao Qi <yao.qi@linaro.org> PR tdep/18107 * gdb.base/catch-syscall.c [!SYS_pipe] (pipe2_syscall): New variable. * gdb.base/catch-syscall.exp: Don't skip it on aarch64*-*-linux* target. Remove elements in all_syscalls. (test_catch_syscall_multi_arch): Skip it on aarch64*-linux* target. (setup_all_syscalls): New proc.
2015-03-18 11:47:45 +01:00
} elseif { [istarget "aarch64*-linux*"] } {
set arch1 "aarch64"
set arch2 "arm"
set syscall1_name "reboot"
set syscall2_name "_newselect"
set syscall_number 142
Partial fix for PR breakpoints/10737: Make syscall info be per-arch instead of global This patch intends to partially fix PR breakpoints/10737, which is about making the syscall information (for the "catch syscall" command) be per-arch, instead of global. This is not a full fix because of the other issues pointed by Pedro here: <https://sourceware.org/bugzilla/show_bug.cgi?id=10737#c5> However, I consider it a good step towards the real fix. It will also help me fix <https://sourceware.org/bugzilla/show_bug.cgi?id=17402>. What this patch does, basically, is move the "syscalls_info" struct to gdbarch. Currently, the syscall information is stored in a global variable inside gdb/xml-syscall.c, which means that there is no easy way to correlate this info with the current target or architecture being used, for example. This causes strange behaviors, because the syscall info is not re-read when the arch changes. For example, if you put a syscall catchpoint in syscall 5 on i386 (syscall open), and then load a x86_64 program on GDB and put the same syscall 5 there (fstat on x86_64), you will still see that GDB tells you that it is catching "open", even though it is not. With this patch, GDB correctly says that it will be catching fstat syscalls. (gdb) set architecture i386 The target architecture is assumed to be i386 (gdb) catch syscall 5 Catchpoint 1 (syscall 'open' [5]) (gdb) set architecture i386:x86-64 The target architecture is assumed to be i386:x86-64 (gdb) catch syscall 5 Catchpoint 2 (syscall 'open' [5]) But with the patch: (gdb) set architecture i386 The target architecture is assumed to be i386 (gdb) catch syscall 5 Catchpoint 1 (syscall 'open' [5]) (gdb) set architecture i386:x86-64 The target architecture is assumed to be i386:x86-64 (gdb) catch syscall 5 Catchpoint 2 (syscall 'fstat' [5]) As I said, there are still some problems on the "catch syscall" mechanism, because (for example) the user should be able to "catch syscall open" on i386, and then expect "open" to be caught also on x86_64. Currently, it doesn't work. I intend to work on this later. gdb/ 2014-11-20 Sergio Durigan Junior <sergiodj@redhat.com> PR breakpoints/10737 * amd64-linux-tdep.c (amd64_linux_init_abi_common): Adjust call to set_xml_syscall_file_name to provide gdbarch. * arm-linux-tdep.c (arm_linux_init_abi): Likewise. * bfin-linux-tdep.c (bfin_linux_init_abi): Likewise. * breakpoint.c (print_it_catch_syscall): Adjust call to get_syscall_by_number to provide gdbarch. (print_one_catch_syscall): Likewise. (print_mention_catch_syscall): Likewise. (print_recreate_catch_syscall): Likewise. (catch_syscall_split_args): Adjust calls to get_syscall_by_number and get_syscall_by_name to provide gdbarch. (catch_syscall_completer): Adjust call to get_syscall_names to provide gdbarch. * gdbarch.c: Regenerate. * gdbarch.h: Likewise. * gdbarch.sh: Forward declare "struct syscalls_info". (xml_syscall_file): New variable. (syscalls_info): Likewise. * i386-linux-tdep.c (i386_linux_init_abi): Adjust call to set_xml_syscall_file_name to provide gdbarch. * mips-linux-tdep.c (mips_linux_init_abi): Likewise. * ppc-linux-tdep.c (ppc_linux_init_abi): Likewise. * s390-linux-tdep.c (s390_gdbarch_init): Likewise. * sparc-linux-tdep.c (sparc32_linux_init_abi): Likewise. * sparc64-linux-tdep.c (sparc64_linux_init_abi): Likewise. * xml-syscall.c: Include gdbarch.h. (set_xml_syscall_file_name): Accept gdbarch parameter. (get_syscall_by_number): Likewise. (get_syscall_by_name): Likewise. (get_syscall_names): Likewise. (my_gdb_datadir): Delete global variable. (struct syscalls_info) <my_gdb_datadir>: New variable. (struct syscalls_info) <sysinfo>: Rename variable to "syscalls_info". (sysinfo): Delete global variable. (have_initialized_sysinfo): Likewise. (xml_syscall_file): Likewise. (sysinfo_free_syscalls_desc): Rename to... (syscalls_info_free_syscalls_desc): ... this. (free_syscalls_info): Rename "sysinfo" to "syscalls_info". Adjust code to the new layout of "struct syscalls_info". (make_cleanup_free_syscalls_info): Rename parameter "sysinfo" to "syscalls_info". (syscall_create_syscall_desc): Likewise. (syscall_start_syscall): Likewise. (syscall_parse_xml): Likewise. (xml_init_syscalls_info): Likewise. Drop "const" from return value. (init_sysinfo): Rename to... (init_syscalls_info): ...this. Add gdbarch as a parameter. Adjust function to deal with gdbarch. (xml_get_syscall_number): Delete parameter sysinfo. Accept gdbarch as a parameter. Adjust code. (xml_get_syscall_name): Likewise. (xml_list_of_syscalls): Likewise. (set_xml_syscall_file_name): Accept gdbarch as parameter. (get_syscall_by_number): Likewise. (get_syscall_by_name): Likewise. (get_syscall_names): Likewise. * xml-syscall.h (set_xml_syscall_file_name): Likewise. (get_syscall_by_number): Likewise. (get_syscall_by_name): Likewise. (get_syscall_names): Likewise. gdb/testsuite/ 2014-11-20 Sergio Durigan Junior <sergiodj@redhat.com> PR breakpoints/10737 * gdb.base/catch-syscall.exp (do_syscall_tests): Call test_catch_syscall_multi_arch. (test_catch_syscall_multi_arch): New function.
2014-11-20 18:28:18 +01:00
} elseif { [istarget "s390*-linux*"] } {
set arch1 "s390:31-bit"
Partial fix for PR breakpoints/10737: Make syscall info be per-arch instead of global This patch intends to partially fix PR breakpoints/10737, which is about making the syscall information (for the "catch syscall" command) be per-arch, instead of global. This is not a full fix because of the other issues pointed by Pedro here: <https://sourceware.org/bugzilla/show_bug.cgi?id=10737#c5> However, I consider it a good step towards the real fix. It will also help me fix <https://sourceware.org/bugzilla/show_bug.cgi?id=17402>. What this patch does, basically, is move the "syscalls_info" struct to gdbarch. Currently, the syscall information is stored in a global variable inside gdb/xml-syscall.c, which means that there is no easy way to correlate this info with the current target or architecture being used, for example. This causes strange behaviors, because the syscall info is not re-read when the arch changes. For example, if you put a syscall catchpoint in syscall 5 on i386 (syscall open), and then load a x86_64 program on GDB and put the same syscall 5 there (fstat on x86_64), you will still see that GDB tells you that it is catching "open", even though it is not. With this patch, GDB correctly says that it will be catching fstat syscalls. (gdb) set architecture i386 The target architecture is assumed to be i386 (gdb) catch syscall 5 Catchpoint 1 (syscall 'open' [5]) (gdb) set architecture i386:x86-64 The target architecture is assumed to be i386:x86-64 (gdb) catch syscall 5 Catchpoint 2 (syscall 'open' [5]) But with the patch: (gdb) set architecture i386 The target architecture is assumed to be i386 (gdb) catch syscall 5 Catchpoint 1 (syscall 'open' [5]) (gdb) set architecture i386:x86-64 The target architecture is assumed to be i386:x86-64 (gdb) catch syscall 5 Catchpoint 2 (syscall 'fstat' [5]) As I said, there are still some problems on the "catch syscall" mechanism, because (for example) the user should be able to "catch syscall open" on i386, and then expect "open" to be caught also on x86_64. Currently, it doesn't work. I intend to work on this later. gdb/ 2014-11-20 Sergio Durigan Junior <sergiodj@redhat.com> PR breakpoints/10737 * amd64-linux-tdep.c (amd64_linux_init_abi_common): Adjust call to set_xml_syscall_file_name to provide gdbarch. * arm-linux-tdep.c (arm_linux_init_abi): Likewise. * bfin-linux-tdep.c (bfin_linux_init_abi): Likewise. * breakpoint.c (print_it_catch_syscall): Adjust call to get_syscall_by_number to provide gdbarch. (print_one_catch_syscall): Likewise. (print_mention_catch_syscall): Likewise. (print_recreate_catch_syscall): Likewise. (catch_syscall_split_args): Adjust calls to get_syscall_by_number and get_syscall_by_name to provide gdbarch. (catch_syscall_completer): Adjust call to get_syscall_names to provide gdbarch. * gdbarch.c: Regenerate. * gdbarch.h: Likewise. * gdbarch.sh: Forward declare "struct syscalls_info". (xml_syscall_file): New variable. (syscalls_info): Likewise. * i386-linux-tdep.c (i386_linux_init_abi): Adjust call to set_xml_syscall_file_name to provide gdbarch. * mips-linux-tdep.c (mips_linux_init_abi): Likewise. * ppc-linux-tdep.c (ppc_linux_init_abi): Likewise. * s390-linux-tdep.c (s390_gdbarch_init): Likewise. * sparc-linux-tdep.c (sparc32_linux_init_abi): Likewise. * sparc64-linux-tdep.c (sparc64_linux_init_abi): Likewise. * xml-syscall.c: Include gdbarch.h. (set_xml_syscall_file_name): Accept gdbarch parameter. (get_syscall_by_number): Likewise. (get_syscall_by_name): Likewise. (get_syscall_names): Likewise. (my_gdb_datadir): Delete global variable. (struct syscalls_info) <my_gdb_datadir>: New variable. (struct syscalls_info) <sysinfo>: Rename variable to "syscalls_info". (sysinfo): Delete global variable. (have_initialized_sysinfo): Likewise. (xml_syscall_file): Likewise. (sysinfo_free_syscalls_desc): Rename to... (syscalls_info_free_syscalls_desc): ... this. (free_syscalls_info): Rename "sysinfo" to "syscalls_info". Adjust code to the new layout of "struct syscalls_info". (make_cleanup_free_syscalls_info): Rename parameter "sysinfo" to "syscalls_info". (syscall_create_syscall_desc): Likewise. (syscall_start_syscall): Likewise. (syscall_parse_xml): Likewise. (xml_init_syscalls_info): Likewise. Drop "const" from return value. (init_sysinfo): Rename to... (init_syscalls_info): ...this. Add gdbarch as a parameter. Adjust function to deal with gdbarch. (xml_get_syscall_number): Delete parameter sysinfo. Accept gdbarch as a parameter. Adjust code. (xml_get_syscall_name): Likewise. (xml_list_of_syscalls): Likewise. (set_xml_syscall_file_name): Accept gdbarch as parameter. (get_syscall_by_number): Likewise. (get_syscall_by_name): Likewise. (get_syscall_names): Likewise. * xml-syscall.h (set_xml_syscall_file_name): Likewise. (get_syscall_by_number): Likewise. (get_syscall_by_name): Likewise. (get_syscall_names): Likewise. gdb/testsuite/ 2014-11-20 Sergio Durigan Junior <sergiodj@redhat.com> PR breakpoints/10737 * gdb.base/catch-syscall.exp (do_syscall_tests): Call test_catch_syscall_multi_arch. (test_catch_syscall_multi_arch): New function.
2014-11-20 18:28:18 +01:00
set arch2 "s390:64-bit"
set syscall1_name "_newselect"
set syscall2_name "select"
set syscall_number 142
}
with_test_prefix "multiple targets" {
# We are not interested in loading any binary here, and in
# some systems (PowerPC, for example), if we load a binary
# there is no way to set other architecture.
gdb_exit
gdb_start
gdb_test "set architecture $arch1" \
"The target architecture is assumed to be $arch1" \
"set arch to $arch1"
gdb_test "catch syscall $syscall_number" \
"Catchpoint $decimal \\(syscall .${syscall1_name}. \\\[${syscall_number}\\\]\\)" \
"insert catch syscall on syscall $syscall_number -- $syscall1_name on $arch1"
gdb_test "set architecture $arch2" \
"The target architecture is assumed to be $arch2" \
"set arch to $arch2"
gdb_test "catch syscall $syscall_number" \
"Catchpoint $decimal \\(syscall .${syscall2_name}. \\\[${syscall_number}\\\]\\)" \
"insert catch syscall on syscall $syscall_number -- $syscall2_name on $arch2"
clean_restart $binfile
}
}
proc do_syscall_tests_without_xml {} {
# Make sure GDB doesn't load the syscalls xml from the system data
# directory.
gdb_test "set data-directory /the/path/to/nowhere" \
"Warning: /the/path/to/nowhere: .*"
# Let's test if we can catch syscalls without XML support.
# We should succeed, but GDB is not supposed to print syscall names.
if [runto_main] then { test_catch_syscall_without_args_noxml }
# The only valid argument "catch syscall" should accept is the
# syscall number, and not the name (since it can't translate a
# name to a number).
if [runto_main] then { test_catch_syscall_with_args_noxml }
# Now, we'll try to provide a syscall name (valid or not) to the command,
# and expect it to fail.
if [runto_main] then { test_catch_syscall_with_wrong_args_noxml }
}
# This procedure fills the vector "all_syscalls_numbers" with the proper
# numbers for the used syscalls according to the architecture.
proc fill_all_syscalls_numbers {} {
gdb: Improve syscall entry/return tracking on Linux The existing logic was simply to flip syscall entry/return state when a syscall trap was seen, and even then only with active 'catch syscall'. That can get out of sync if 'catch syscall' is toggled at odd times. This patch updates the entry/return state for all syscall traps, regardless of catching state, and also updates known syscall state for other kinds of traps. Almost all PTRACE_EVENT stops are delivered from the middle of a syscall, so this can act like an entry. Every other kind of ptrace stop is only delivered outside of syscall event pairs, so marking them ignored ensures the next syscall trap looks like an entry. Three new test scenarios are added to catch-syscall.exp: - Disable 'catch syscall' from an entry to deliberately miss the return event, then re-enable to make sure a new entry is recognized. - Enable 'catch syscall' for the first time from a vfork event, which is a PTRACE_EVENT_VFORK in the middle of the syscall. Make sure the next syscall event is recognized as the return. - Make sure entry and return are recognized for an ENOSYS syscall. This is to defeat a common x86 hack that uses the pre-filled ENOSYS return value as a sign of being on the entry side. gdb/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * linux-nat.c (linux_handle_syscall_trap): Always update entry/ return state, even when not actively catching syscalls at all. (linux_handle_extended_wait): Mark syscall_state like an entry. (wait_lwp): Set syscall_state ignored for other traps. (linux_nat_filter_event): Likewise. gdb/testsuite/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * gdb.base/catch-syscall.c: Include <sched.h>. (unknown_syscall): New variable. (main): Trigger a vfork and an unknown syscall. * gdb.base/catch-syscall.exp (vfork_syscalls): New variable. (unknown_syscall_number): Likewise. (check_call_to_syscall): Accept an optional syscall pattern. (check_return_from_syscall): Likewise. (check_continue): Likewise. (test_catch_syscall_without_args): Check for vfork and ENOSYS. (test_catch_syscall_skipping_return): New test toggling off 'catch syscall' to step over the syscall return, then toggling back on. (test_catch_syscall_mid_vfork): New test turning on 'catch syscall' during a PTRACE_EVENT_VFORK stop, in the middle of a vfork syscall. (do_syscall_tests): Call test_catch_syscall_without_args and test_catch_syscall_mid_vfork. (test_catch_syscall_without_args_noxml): Check for vfork and ENOSYS. (fill_all_syscalls_numbers): Initialize unknown_syscall_number.
2015-10-20 02:59:38 +02:00
global all_syscalls_numbers last_syscall_number unknown_syscall_number all_syscalls
Fix PR breakpoints/16297: catch syscall with syscall 0 Code rationale ============== by: Gabriel Krisman Bertazi This is a fix for bug 16297. The problem occurs when the user attempts to catch any syscall 0 (such as syscall read on Linux/x86_64). GDB was not able to catch the syscall and was missing the breakpoint. Now, breakpoint_hit_catch_syscall returns immediately when it finds the correct syscall number, avoiding a following check for the end of the search vector, that returns a no hit if the syscall number was zero. Testcase rationale ================== by: Sergio Durigan Junior This testcase is a little difficult to write. By doing a quick inspection at the Linux source, one can see that, in many targets, the syscall number 0 is restart_syscall, which is forbidden to be called from userspace. Therefore, on many targets, there's just no way to test this safely. My decision was to take the simpler route and just adds the "read" syscall on the default test. Its number on x86_64 is zero, which is "good enough" since many people here do their tests on x86_64 anyway and it is a popular architecture. However, there was another little gotcha. When using "read" passing 0 as the third parameter (i.e., asking it to read 0 bytes), current libc implementations could choose not to effectively call the syscall. Therefore, the best solution was to create a temporary pipe, write 1 byte into it, and then read this byte from it. gdb/ChangeLog 2013-12-19 Gabriel Krisman Bertazi <gabriel@krisman.be> PR breakpoints/16297 * breakpoint.c (breakpoint_hit_catch_syscall): Return immediately when expected syscall is hit. gdb/testsuite/ChangeLog 2013-12-19 Sergio Durigan Junior <sergiodj@redhat.com> PR breakpoints/16297 * gdb.base/catch-syscall.c (read_syscall, pipe_syscall) (write_syscall): New variables. (main): Create a pipe, write 1 byte in it, and read 1 byte from it. * gdb.base/catch-syscall.exp (all_syscalls): Include "pipe, "write" and "read" syscalls. (fill_all_syscalls_numbers): Improve the way to obtain syscalls numbers.
2013-12-19 20:01:49 +01:00
foreach syscall $all_syscalls {
lappend all_syscalls_numbers [get_integer_valueof "${syscall}_syscall" -1]
}
Improve and fix catch-syscall.exp While fixing another bug, I found that the current gdb.base/catch-syscall.exp is kind of messy, could use some improvements, and is not correctly testing some things. I've made the following patch to address all the issues I found. On the organization side, it does a cleanup and removes unecessary imports of gdb_prompt, uses prepare_for_testing and clean_restart where needed, and fixes some comments. The testcase was also not correctly testing catching syscalls using only numbers, or catching many syscalls at once. I fixed that. The patch also uses a new method for obtaining the syscalls numbers: it relies on the C source file to get them, via <sys/syscall.h> and SYS_* macros. This makes the .exp file simpler because there is no need to include target conditionals there. I tested this on x86_64 Fedora 18. gdb/testsuite/ChangeLog: 2013-12-18 Sergio Durigan Junior <sergiodj@redhat.com> * gdb.base/catch-syscall.c: Include <sys/syscall.h>. (close_syscall, chroot_syscall, exit_group_syscall): New variables. * gdb.base/catch-syscall.exp: Replace gdb_compile by prepare_for_testing. Call fill_all_syscalls_numbers before starting. Replace gdb_exit, gdb_start, gdb_reinitialize_dir and gdb_load by clean_restart. (check_info_bp_any_syscall, check_info_bp_specific_syscall) (check_info_bp_many_syscalls): Remove global gdb_prompt. (check_call_to_syscall): Likewise. Add global decimal. Improve testing regex. (check_return_from_syscall): Likewise. (check_continue, insert_catch_syscall_with_arg): Remove global gdb_prompt. (insert_catch_syscall_with_many_args): Likewise. Add global decimal. Fix $filter_str. Improve testing regex. (check_for_program_end): Remove global gdb_prompt. (test_catch_syscall_without_args): Likewise. Add global decimal. Improve testing regex. (test_catch_syscall_with_args, test_catch_syscall_with_many_args) (test_catch_syscall_with_wrong_args) (test_catch_syscall_restarting_inferior) (test_catch_syscall_fail_nodatadir): Remove global gdb_prompt. (do_syscall_tests): Likewise. Remove global srcdir. (test_catch_syscall_without_args_noxml): Remove global gdb_prompt. Add global last_syscall_number. Test for the exact syscall number to be caught. (test_catch_syscall_with_args_noxml): Remove global gdb_prompt. Add global all_syscalls_numbers. Test each syscall number to be caught, instead of only testing "close". (test_catch_syscall_with_wrong_args_noxml): Remove global gdb_prompt. (do_syscall_tests_without_xml): Likewise. Remove global srcdir. Remove stale comment. (fill_all_syscalls_numbers): Add global last_syscall_number. Fill the correct syscall numbers using information from the inferior.
2013-12-18 23:19:01 +01:00
set last_syscall_number [get_integer_valueof "exit_group_syscall" -1]
gdb: Improve syscall entry/return tracking on Linux The existing logic was simply to flip syscall entry/return state when a syscall trap was seen, and even then only with active 'catch syscall'. That can get out of sync if 'catch syscall' is toggled at odd times. This patch updates the entry/return state for all syscall traps, regardless of catching state, and also updates known syscall state for other kinds of traps. Almost all PTRACE_EVENT stops are delivered from the middle of a syscall, so this can act like an entry. Every other kind of ptrace stop is only delivered outside of syscall event pairs, so marking them ignored ensures the next syscall trap looks like an entry. Three new test scenarios are added to catch-syscall.exp: - Disable 'catch syscall' from an entry to deliberately miss the return event, then re-enable to make sure a new entry is recognized. - Enable 'catch syscall' for the first time from a vfork event, which is a PTRACE_EVENT_VFORK in the middle of the syscall. Make sure the next syscall event is recognized as the return. - Make sure entry and return are recognized for an ENOSYS syscall. This is to defeat a common x86 hack that uses the pre-filled ENOSYS return value as a sign of being on the entry side. gdb/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * linux-nat.c (linux_handle_syscall_trap): Always update entry/ return state, even when not actively catching syscalls at all. (linux_handle_extended_wait): Mark syscall_state like an entry. (wait_lwp): Set syscall_state ignored for other traps. (linux_nat_filter_event): Likewise. gdb/testsuite/ChangeLog: 2015-10-19 Josh Stone <jistone@redhat.com> * gdb.base/catch-syscall.c: Include <sched.h>. (unknown_syscall): New variable. (main): Trigger a vfork and an unknown syscall. * gdb.base/catch-syscall.exp (vfork_syscalls): New variable. (unknown_syscall_number): Likewise. (check_call_to_syscall): Accept an optional syscall pattern. (check_return_from_syscall): Likewise. (check_continue): Likewise. (test_catch_syscall_without_args): Check for vfork and ENOSYS. (test_catch_syscall_skipping_return): New test toggling off 'catch syscall' to step over the syscall return, then toggling back on. (test_catch_syscall_mid_vfork): New test turning on 'catch syscall' during a PTRACE_EVENT_VFORK stop, in the middle of a vfork syscall. (do_syscall_tests): Call test_catch_syscall_without_args and test_catch_syscall_mid_vfork. (test_catch_syscall_without_args_noxml): Check for vfork and ENOSYS. (fill_all_syscalls_numbers): Initialize unknown_syscall_number.
2015-10-20 02:59:38 +02:00
set unknown_syscall_number [get_integer_valueof "unknown_syscall" -1]
Improve and fix catch-syscall.exp While fixing another bug, I found that the current gdb.base/catch-syscall.exp is kind of messy, could use some improvements, and is not correctly testing some things. I've made the following patch to address all the issues I found. On the organization side, it does a cleanup and removes unecessary imports of gdb_prompt, uses prepare_for_testing and clean_restart where needed, and fixes some comments. The testcase was also not correctly testing catching syscalls using only numbers, or catching many syscalls at once. I fixed that. The patch also uses a new method for obtaining the syscalls numbers: it relies on the C source file to get them, via <sys/syscall.h> and SYS_* macros. This makes the .exp file simpler because there is no need to include target conditionals there. I tested this on x86_64 Fedora 18. gdb/testsuite/ChangeLog: 2013-12-18 Sergio Durigan Junior <sergiodj@redhat.com> * gdb.base/catch-syscall.c: Include <sys/syscall.h>. (close_syscall, chroot_syscall, exit_group_syscall): New variables. * gdb.base/catch-syscall.exp: Replace gdb_compile by prepare_for_testing. Call fill_all_syscalls_numbers before starting. Replace gdb_exit, gdb_start, gdb_reinitialize_dir and gdb_load by clean_restart. (check_info_bp_any_syscall, check_info_bp_specific_syscall) (check_info_bp_many_syscalls): Remove global gdb_prompt. (check_call_to_syscall): Likewise. Add global decimal. Improve testing regex. (check_return_from_syscall): Likewise. (check_continue, insert_catch_syscall_with_arg): Remove global gdb_prompt. (insert_catch_syscall_with_many_args): Likewise. Add global decimal. Fix $filter_str. Improve testing regex. (check_for_program_end): Remove global gdb_prompt. (test_catch_syscall_without_args): Likewise. Add global decimal. Improve testing regex. (test_catch_syscall_with_args, test_catch_syscall_with_many_args) (test_catch_syscall_with_wrong_args) (test_catch_syscall_restarting_inferior) (test_catch_syscall_fail_nodatadir): Remove global gdb_prompt. (do_syscall_tests): Likewise. Remove global srcdir. (test_catch_syscall_without_args_noxml): Remove global gdb_prompt. Add global last_syscall_number. Test for the exact syscall number to be caught. (test_catch_syscall_with_args_noxml): Remove global gdb_prompt. Add global all_syscalls_numbers. Test each syscall number to be caught, instead of only testing "close". (test_catch_syscall_with_wrong_args_noxml): Remove global gdb_prompt. (do_syscall_tests_without_xml): Likewise. Remove global srcdir. Remove stale comment. (fill_all_syscalls_numbers): Add global last_syscall_number. Fill the correct syscall numbers using information from the inferior.
2013-12-18 23:19:01 +01:00
}
Support catch syscall on aarch64 linux Hi, This patch is to support catch syscall on aarch64 linux. We implement gdbarch method get_syscall_number for aarch64-linux, and add aarch64-linux.xml file, which looks straightforward, however the changes to test case doesn't. First of all, we enable catch-syscall.exp on aarch64-linux target, but skip the multi_arch testing on current stage. I plan to touch multi arch debugging on aarch64-linux later. Then, when I run catch-syscall.exp on aarch64-linux, gcc errors that SYS_pipe isn't defined. We find that aarch64 kernel only has pipe2 syscall and libc already convert pipe to pipe2. As a result, I change catch-syscall.c to use SYS_pipe if it is defined, otherwise use SYS_pipe2 instead. The vector all_syscalls in catch-syscall.exp can't be pre-determined, so I add a new proc setup_all_syscalls to fill it, according to the availability of SYS_pipe. Regression tested on {x86_64, aarch64}-linux x {native, gdbserver}. gdb: 2015-03-18 Yao Qi <yao.qi@linaro.org> PR tdep/18107 * aarch64-linux-tdep.c: Include xml-syscall.h (aarch64_linux_get_syscall_number): New function. (aarch64_linux_init_abi): Call set_gdbarch_get_syscall_number. * syscalls/aarch64-linux.xml: New file. gdb/testsuite: 2015-03-18 Yao Qi <yao.qi@linaro.org> PR tdep/18107 * gdb.base/catch-syscall.c [!SYS_pipe] (pipe2_syscall): New variable. * gdb.base/catch-syscall.exp: Don't skip it on aarch64*-*-linux* target. Remove elements in all_syscalls. (test_catch_syscall_multi_arch): Skip it on aarch64*-linux* target. (setup_all_syscalls): New proc.
2015-03-18 11:47:45 +01:00
# Set up the vector all_syscalls.
proc setup_all_syscalls {} {
global all_syscalls
global gdb_prompt
# They are ordered according to the file, so do not change this.
lappend all_syscalls "close"
lappend all_syscalls "chroot"
# SYS_pipe doesn't exist on aarch64 kernel.
set test "check SYS_pipe"
gdb_test_multiple "p pipe_syscall" $test {
-re " = .*$gdb_prompt $" {
pass $test
lappend all_syscalls "pipe"
}
-re "No symbol .*$gdb_prompt $" {
pass $test
# SYS_pipe isn't defined, use SYS_pipe2 instead.
lappend all_syscalls "pipe2"
}
}
lappend all_syscalls "write"
lappend all_syscalls "read"
}
setup_all_syscalls
Improve and fix catch-syscall.exp While fixing another bug, I found that the current gdb.base/catch-syscall.exp is kind of messy, could use some improvements, and is not correctly testing some things. I've made the following patch to address all the issues I found. On the organization side, it does a cleanup and removes unecessary imports of gdb_prompt, uses prepare_for_testing and clean_restart where needed, and fixes some comments. The testcase was also not correctly testing catching syscalls using only numbers, or catching many syscalls at once. I fixed that. The patch also uses a new method for obtaining the syscalls numbers: it relies on the C source file to get them, via <sys/syscall.h> and SYS_* macros. This makes the .exp file simpler because there is no need to include target conditionals there. I tested this on x86_64 Fedora 18. gdb/testsuite/ChangeLog: 2013-12-18 Sergio Durigan Junior <sergiodj@redhat.com> * gdb.base/catch-syscall.c: Include <sys/syscall.h>. (close_syscall, chroot_syscall, exit_group_syscall): New variables. * gdb.base/catch-syscall.exp: Replace gdb_compile by prepare_for_testing. Call fill_all_syscalls_numbers before starting. Replace gdb_exit, gdb_start, gdb_reinitialize_dir and gdb_load by clean_restart. (check_info_bp_any_syscall, check_info_bp_specific_syscall) (check_info_bp_many_syscalls): Remove global gdb_prompt. (check_call_to_syscall): Likewise. Add global decimal. Improve testing regex. (check_return_from_syscall): Likewise. (check_continue, insert_catch_syscall_with_arg): Remove global gdb_prompt. (insert_catch_syscall_with_many_args): Likewise. Add global decimal. Fix $filter_str. Improve testing regex. (check_for_program_end): Remove global gdb_prompt. (test_catch_syscall_without_args): Likewise. Add global decimal. Improve testing regex. (test_catch_syscall_with_args, test_catch_syscall_with_many_args) (test_catch_syscall_with_wrong_args) (test_catch_syscall_restarting_inferior) (test_catch_syscall_fail_nodatadir): Remove global gdb_prompt. (do_syscall_tests): Likewise. Remove global srcdir. (test_catch_syscall_without_args_noxml): Remove global gdb_prompt. Add global last_syscall_number. Test for the exact syscall number to be caught. (test_catch_syscall_with_args_noxml): Remove global gdb_prompt. Add global all_syscalls_numbers. Test each syscall number to be caught, instead of only testing "close". (test_catch_syscall_with_wrong_args_noxml): Remove global gdb_prompt. (do_syscall_tests_without_xml): Likewise. Remove global srcdir. Remove stale comment. (fill_all_syscalls_numbers): Add global last_syscall_number. Fill the correct syscall numbers using information from the inferior.
2013-12-18 23:19:01 +01:00
# Fill all the syscalls numbers before starting anything.
fill_all_syscalls_numbers
# Execute the tests, using XML support
gdb_exit
Improve and fix catch-syscall.exp While fixing another bug, I found that the current gdb.base/catch-syscall.exp is kind of messy, could use some improvements, and is not correctly testing some things. I've made the following patch to address all the issues I found. On the organization side, it does a cleanup and removes unecessary imports of gdb_prompt, uses prepare_for_testing and clean_restart where needed, and fixes some comments. The testcase was also not correctly testing catching syscalls using only numbers, or catching many syscalls at once. I fixed that. The patch also uses a new method for obtaining the syscalls numbers: it relies on the C source file to get them, via <sys/syscall.h> and SYS_* macros. This makes the .exp file simpler because there is no need to include target conditionals there. I tested this on x86_64 Fedora 18. gdb/testsuite/ChangeLog: 2013-12-18 Sergio Durigan Junior <sergiodj@redhat.com> * gdb.base/catch-syscall.c: Include <sys/syscall.h>. (close_syscall, chroot_syscall, exit_group_syscall): New variables. * gdb.base/catch-syscall.exp: Replace gdb_compile by prepare_for_testing. Call fill_all_syscalls_numbers before starting. Replace gdb_exit, gdb_start, gdb_reinitialize_dir and gdb_load by clean_restart. (check_info_bp_any_syscall, check_info_bp_specific_syscall) (check_info_bp_many_syscalls): Remove global gdb_prompt. (check_call_to_syscall): Likewise. Add global decimal. Improve testing regex. (check_return_from_syscall): Likewise. (check_continue, insert_catch_syscall_with_arg): Remove global gdb_prompt. (insert_catch_syscall_with_many_args): Likewise. Add global decimal. Fix $filter_str. Improve testing regex. (check_for_program_end): Remove global gdb_prompt. (test_catch_syscall_without_args): Likewise. Add global decimal. Improve testing regex. (test_catch_syscall_with_args, test_catch_syscall_with_many_args) (test_catch_syscall_with_wrong_args) (test_catch_syscall_restarting_inferior) (test_catch_syscall_fail_nodatadir): Remove global gdb_prompt. (do_syscall_tests): Likewise. Remove global srcdir. (test_catch_syscall_without_args_noxml): Remove global gdb_prompt. Add global last_syscall_number. Test for the exact syscall number to be caught. (test_catch_syscall_with_args_noxml): Remove global gdb_prompt. Add global all_syscalls_numbers. Test each syscall number to be caught, instead of only testing "close". (test_catch_syscall_with_wrong_args_noxml): Remove global gdb_prompt. (do_syscall_tests_without_xml): Likewise. Remove global srcdir. Remove stale comment. (fill_all_syscalls_numbers): Add global last_syscall_number. Fill the correct syscall numbers using information from the inferior.
2013-12-18 23:19:01 +01:00
if { ![gdb_skip_xml_test] } {
clean_restart $binfile
do_syscall_tests
# Now, we have to see if GDB displays a warning when we
# don't set the data-directory but try to use catch syscall
# anyway. For that, we must restart GDB first.
Improve and fix catch-syscall.exp While fixing another bug, I found that the current gdb.base/catch-syscall.exp is kind of messy, could use some improvements, and is not correctly testing some things. I've made the following patch to address all the issues I found. On the organization side, it does a cleanup and removes unecessary imports of gdb_prompt, uses prepare_for_testing and clean_restart where needed, and fixes some comments. The testcase was also not correctly testing catching syscalls using only numbers, or catching many syscalls at once. I fixed that. The patch also uses a new method for obtaining the syscalls numbers: it relies on the C source file to get them, via <sys/syscall.h> and SYS_* macros. This makes the .exp file simpler because there is no need to include target conditionals there. I tested this on x86_64 Fedora 18. gdb/testsuite/ChangeLog: 2013-12-18 Sergio Durigan Junior <sergiodj@redhat.com> * gdb.base/catch-syscall.c: Include <sys/syscall.h>. (close_syscall, chroot_syscall, exit_group_syscall): New variables. * gdb.base/catch-syscall.exp: Replace gdb_compile by prepare_for_testing. Call fill_all_syscalls_numbers before starting. Replace gdb_exit, gdb_start, gdb_reinitialize_dir and gdb_load by clean_restart. (check_info_bp_any_syscall, check_info_bp_specific_syscall) (check_info_bp_many_syscalls): Remove global gdb_prompt. (check_call_to_syscall): Likewise. Add global decimal. Improve testing regex. (check_return_from_syscall): Likewise. (check_continue, insert_catch_syscall_with_arg): Remove global gdb_prompt. (insert_catch_syscall_with_many_args): Likewise. Add global decimal. Fix $filter_str. Improve testing regex. (check_for_program_end): Remove global gdb_prompt. (test_catch_syscall_without_args): Likewise. Add global decimal. Improve testing regex. (test_catch_syscall_with_args, test_catch_syscall_with_many_args) (test_catch_syscall_with_wrong_args) (test_catch_syscall_restarting_inferior) (test_catch_syscall_fail_nodatadir): Remove global gdb_prompt. (do_syscall_tests): Likewise. Remove global srcdir. (test_catch_syscall_without_args_noxml): Remove global gdb_prompt. Add global last_syscall_number. Test for the exact syscall number to be caught. (test_catch_syscall_with_args_noxml): Remove global gdb_prompt. Add global all_syscalls_numbers. Test each syscall number to be caught, instead of only testing "close". (test_catch_syscall_with_wrong_args_noxml): Remove global gdb_prompt. (do_syscall_tests_without_xml): Likewise. Remove global srcdir. Remove stale comment. (fill_all_syscalls_numbers): Add global last_syscall_number. Fill the correct syscall numbers using information from the inferior.
2013-12-18 23:19:01 +01:00
clean_restart $binfile
test_catch_syscall_fail_nodatadir
}
# Restart gdb
Improve and fix catch-syscall.exp While fixing another bug, I found that the current gdb.base/catch-syscall.exp is kind of messy, could use some improvements, and is not correctly testing some things. I've made the following patch to address all the issues I found. On the organization side, it does a cleanup and removes unecessary imports of gdb_prompt, uses prepare_for_testing and clean_restart where needed, and fixes some comments. The testcase was also not correctly testing catching syscalls using only numbers, or catching many syscalls at once. I fixed that. The patch also uses a new method for obtaining the syscalls numbers: it relies on the C source file to get them, via <sys/syscall.h> and SYS_* macros. This makes the .exp file simpler because there is no need to include target conditionals there. I tested this on x86_64 Fedora 18. gdb/testsuite/ChangeLog: 2013-12-18 Sergio Durigan Junior <sergiodj@redhat.com> * gdb.base/catch-syscall.c: Include <sys/syscall.h>. (close_syscall, chroot_syscall, exit_group_syscall): New variables. * gdb.base/catch-syscall.exp: Replace gdb_compile by prepare_for_testing. Call fill_all_syscalls_numbers before starting. Replace gdb_exit, gdb_start, gdb_reinitialize_dir and gdb_load by clean_restart. (check_info_bp_any_syscall, check_info_bp_specific_syscall) (check_info_bp_many_syscalls): Remove global gdb_prompt. (check_call_to_syscall): Likewise. Add global decimal. Improve testing regex. (check_return_from_syscall): Likewise. (check_continue, insert_catch_syscall_with_arg): Remove global gdb_prompt. (insert_catch_syscall_with_many_args): Likewise. Add global decimal. Fix $filter_str. Improve testing regex. (check_for_program_end): Remove global gdb_prompt. (test_catch_syscall_without_args): Likewise. Add global decimal. Improve testing regex. (test_catch_syscall_with_args, test_catch_syscall_with_many_args) (test_catch_syscall_with_wrong_args) (test_catch_syscall_restarting_inferior) (test_catch_syscall_fail_nodatadir): Remove global gdb_prompt. (do_syscall_tests): Likewise. Remove global srcdir. (test_catch_syscall_without_args_noxml): Remove global gdb_prompt. Add global last_syscall_number. Test for the exact syscall number to be caught. (test_catch_syscall_with_args_noxml): Remove global gdb_prompt. Add global all_syscalls_numbers. Test each syscall number to be caught, instead of only testing "close". (test_catch_syscall_with_wrong_args_noxml): Remove global gdb_prompt. (do_syscall_tests_without_xml): Likewise. Remove global srcdir. Remove stale comment. (fill_all_syscalls_numbers): Add global last_syscall_number. Fill the correct syscall numbers using information from the inferior.
2013-12-18 23:19:01 +01:00
clean_restart $binfile
# Execute the tests, without XML support. In this case, GDB will
# only display syscall numbers, and not syscall names.
do_syscall_tests_without_xml