Also, provide fallback definitions for constants.
* xlat/mdb_flags.in (MDB_FLAGS_OFFLOAD): Add fallback definition.
(MDB_FLAGS_FAST_LEAVE): New constant, introduced by Linux commit
v5.4-rc1~131^2~321.
riscv64: remove dead code left after riscv64 mpers removal
Since there is no compat on RISCV64 so far, its dead remains are only
confusing.
* linux/riscv64/arch_get_personality.c: Remove.
* Makefile.am (EXTRA_DIST): Remove it.
Reported-by: Elvira Khabirova <lineprinter0@gmail.com>
Suggested-by: Dmitry V. Levin <ldv@altlinux.org>
Complements: v5.3~18 "riscv64: remove mpers support"
tests: try to avoid false positives in times sanity check
Checking clock_gettime(CLOCK_PROCESS_CPUTIME_ID) starting from the
second iteration may produce false positives sometimes, try to spend
more iterations before starting checking it.
* tests/times.c (NUM_USER_ITERS_SQRT): New enum entity.
(NUM_USER_ITERS): Define via NUM_USER_ITERS_SQRT.
(main): Check i against NUM_USER_ITERS_SQRT first before checking
the returned CPU time for sanity.
Fixed build on systems that do not provide necessary definitions
of AUDIT_ARCH_* constants.
* filter_seccomp.c [PERSONALITY0_AUDIT_ARCH]: Include
"xlat/elf_em.h" and "xlat/audit_arch.h" under XLAT_MACROS_ONLY.
* tests/filter_seccomp-flag.c: Likewise.
* NEWS: Mention this fix.
Fixes: v5.3~7 "Introduce seccomp-assisted syscall filtering"
As of now, despite of being stated that -D option runs strace as a "detached"
grandchild (and the option name being named after "daemon"), strace
still runs in the same process group and session, thus not being
"detached" in a common sense and being subjected to process group kill
and session termination kill. Quoting[1]:
I stumble upon unexpected behavior: if strace is used with option '-D'
(tracer as a detached grandchild) and process (leader) kills whole
process group, it will kill strace too.
It can be easily reproduced by `timeout` from "coreutils":
# timeout -s KILL 2 strace -D -o ./strace-inside.log /bin/sleep 10 &
Here we can see, that `strace` didn't finished its output (because it
was killed):
# tail -n 1 ./strace-inside.log
nanosleep({tv_sec=10, tv_nsec=0},
If `timeout` is not run in '--foreground' mode, it changes process group
and after "timeout" it sends two kills:
setpgid(0, 0) = 0
kill(37337, SIGKILL) = 0
kill(0, SIGKILL) = ?
The first kill is for the `sleep` and the second one is for the process
group (which is `strace` part of). PIDs and their relations are:
timeout pid=30595 [ppid=476 bash ] pgrp=30595
sleep pid=37337 [ppid=30595 timeout] pgrp=30595
strace pid=30603 [ppid=1 systemd] pgrp=30595
Here is "strace log" of `strace` inside `timeout`:
strace: Process 30603 attached
wait4(-1, <unfinished ...>) = ?
+++ killed by SIGKILL +++
I think that detached `strace` should not be killed like that -- it
should not be part of former grandparents' "job pipeline".
While this behaviour is not exactly intuitive, it is implemented this
way for quite some time, so it might be relied upon by some of strace
users. In order to address this issue, two new levels of
"daemonisation" are added, that put strace in a separate process group
and session, respectively.
[1] https://lists.strace.io/pipermail/strace-devel/2019-October/009160.html
* strace.1.in (.SH SYNOPSIS): Update.
(.SS Tracing): Document -DD and -DDD.
* strace.c (DAEMONIZE_NONE, DAEMONIZE_GRANDCHILD, DAEMONIZE_NEW_PGROUP,
* DAEMONIZE_NEW_SESSION, DAEMONIZE_OPTS_GUARD__, MAX_DAEMONIZE_OPTS):
* New enumeration entities.
(daemonized_tracer): Change type to unsigned int.
(usage): Document -DD and -DDD.
(startup_attach) <daemonized_tracer == DAEMONIZE_NEW_PGROUP>: Call
setpgid.
<daemonized_tracer == DAEMONIZE_NEW_SESSION>: Call setsid.
(init) <case 'D'>: Increase daemonized_tracer instead of setting to 1.
(init): Bail out if too many -D's are given.
* NEWS: Mention this improvement.
* tests/options-syntax.test: Add checks for -D option usage.
Co-authored-by: Eugene Syromyatnikov <evgsyr@gmail.com>
Co-authored-by: Dmitry V. Levin <ldv@altlinux.org>
Fix -b execve when --seccomp-bpf option is specified
As --seccomp-bpf does not support detaching, explicitly turn off
this option when -b execve is specified.
* strace.c (init): Turn off --seccomp-bpf when -b execve is specified.
* NEWS: Mention this fix.
* tests/bexecve.test: Check it.
Fixes: v5.3~7 "Introduce seccomp-assisted syscall filtering"
Avoid unsupported fork() call on no-MMU Linux systems to fix
the following link error:
ld: strace-filter_seccomp.o: in function `check_seccomp_filter':
filter_seccomp.c:(.text+0x39a): undefined reference to `fork'
collect2: error: ld returned 1 exit status
* filter_seccomp.c (__gcov_flush, check_seccomp_order_do_child,
check_seccomp_order_tracer): Move under HAVE_FORK guard.
(check_seccomp_order): Move fork code under HAVE_FORK guard.
(check_seccomp_filter_properties): Do not check for NOMMU_SYSTEM.
* NEWS: Mention this fix.
Reported-and-tested-by: Baruch Siach <baruch@tkos.co.il>
Fixes: v5.3~7 "Introduce seccomp-assisted syscall filtering"
Rename it to FLAG and implement via newly added BIT helper macro,
in preparation for general use
* defs.h: Use FLAG instead of FLAG_.
(FLAG_): Move it...
* macros.h (FLAG): Here, implement using BIT macro.
[!BIT] (BIT): New macro.
util: avoid double printing of ellipsis in print_array
When PAF_ARRAY_TRUNCATED flag is provided to print_array_ex.
* util.c (print_array_ex): Add truncated flag, set it when the output
is truncated, avoid explicit ellipsis printing when it is set.
Fixes: v5.3~14 "Add PAF_ARRAY_TRUNCATED flag for print_array_ex"
tests/sigaction: workaround odd libcs on alpha and mips
Apparently, some libcs define SA_RESTORER on alpha and mips
despite of the absence of sa_restorer field. Workaround this
to match the logic implemented in decode_old_sigaction().
* tests/sigaction.c (main) [ALPHA || MIPS]: Do not check decoding
of sa_restorer field.
Distribute copyright-year-gen, file-date-gen, and git-version-gen
These scripts have to be distributed along with configure.ac since
the latter uses them.
* Makefile.am (EXTRA_DIST): Add copyright-year-gen, file-date-gen,
and git-version-gen.
* tests/trace_lstat.in (lstat): Change -a argument from 32 to 31.
* tests/trace_stat.in (stat): Change -a argument from 32 to 30.
* tests/gen_tests.in (lstat): Change -a argument from 32 to 31.
(stat): Change -a argument from 32 to 30.
This fixes Debian bug #929715.
Option descriptions generally maintain lexicographical order.
* strace.1.in (.SS Miscellaneous): Move --seccomp-bpf option description
down in order to maintain lexicographical option description order.
Commits v5.3~74 and v5.3~73 have introduced an extended syntax
for time interval sizes specification, but the relevant descriotion
was lacking. Fix it by adding the relevant section to the man page
and reference to it in the descriptions of the respective options.
* strace.1.in (.SH OPTIONS): Rewrite descriptions of -O,
-e inject=delay_enter, and -e inject=delay_exit values, refer to section
"Time specification format description".
(.SS "Time specification format description"): New section.
Complements: v5.3~74 "delay: use parse_ts for parsing delay value"
Complements: v5.3~73 "count: use parse_ts for parsing overhead value"
As inject option description is moved to a separate section, its
.TP macro now requires an argument.
* strace.1.in (.SS Tampering): Add indent size argument to the first
.TP macro instance.
Fixes: v5.3~99 "strace.1.in: reorder options"
Indent the C preprocessor directives to reflect their nesting
using the following script:
$ cppi -l $(git grep -El '^[[:space:]]*#[[:space:]]*(if|ifdef|ifndef|elif|else|endif|define|pragma)[[:space:]]' |grep -v '\.sh$') |while read f; do
cppi < "$f" > "$f".cppi; mv "$f".cppi "$f"
done
Test filter_seccomp-perf checks whether seccomp-filter is actually
enabled by comparing the number of syscalls performed in a time interval
when seccomp-filter is enabled vs. disabled. The number of syscalls
should be at least one order of magnitude higher when seccomp-filter
is enabled.
Test filter_seccomp-flag ensures the audit_arch_vec[].flag constants do
not conflict with syscall numbers. If this test fails, then the number
of syscalls grew high enough that the code for seccomp-filter needs to
be updated.
* tests/init.sh (test_prog_set): New function.
* tests/status-none-f.c: New file.
* tests/filter_seccomp.in: Likewise.
* tests/filter_seccomp.sh: Likewise.
* tests/filter_seccomp-perf.c: Likewise.
* tests/filter_seccomp-flag.c: Likewise.
* tests/filter_seccomp-perf.test: New test.
* tests/Makefile.am (EXTRA_DIST): Add filter_seccomp.in and
filter_seccomp.sh.
(MISC_TESTS): Add filter_seccomp-perf.test.
(check_PROGRAMS): Add filter_seccomp-perf and filter_seccomp-flag.
* tests/pure_executables.list: Add status-none-f.
* tests/.gitignore: Add status-none-f, filter_seccomp-perf, and
filter_seccomp-flag.
* tests/gen_tests.in (filter_seccomp, filter_seccomp-flag): New entries.
Co-authored-by: Paul Chaignon <paul.chaignon@gmail.com>
Co-Authored-by: Dmitry V. Levin <ldv@altlinux.org>
filter_seccomp: skip seccomp setup when there's nothing to filter
If the trace_set set is complete (no syscalls are filtered), seccomp
filtering is disabled. This patch adds a new is_complete_set_array
function to check whether all sets of a set array are complete.
* number_set.c (is_complete_set_array): New function.
* number_set.h (is_complete_set_array): New prototype.
* filter_seccomp.c (check_seccomp_filter): Skip seccomp setup if there is
nothing to filter.
Signed-off-by: Paul Chaignon <paul.chaignon@gmail.com>
With this patch, strace can rely on seccomp to only be stopped at syscalls
of interest, instead of stopping at all syscalls. The seccomp filtering
of syscalls is opt-in only; it must be enabled with the --seccomp-bpf
option. Kernel support is first checked with check_seccomp_filter(),
which also ensures the BPF program derived from the syscalls to filter
is not larger than the kernel's limit.
The --seccomp-bpf option implies -f, but a warning is emitted if -f is not
explicitly specified. Since a task's children inherit its seccomp
filters, we want to ensure all children are also traced to avoid their
syscalls failing with ENOSYS (cf. SECCOMP_RET_TRACE in seccomp man page).
Fork/vfork/clone children of traced processes are marked as not having a
seccomp filter until we receive a first seccomp-stop. They are therefore
stopped at every syscall entries and exits until that first seccomp-stop.
The current BPF program implements a simple linear match of the syscall
numbers. Contiguous sequences of syscall numbers are however matched as
an interval, with two instructions only. The algorithm can be improved
or replaced in the future without impacting user-observed behavior.
The behavior of SECCOMP_RET_TRACE changed between Linux 4.7 and 4.8
(cf. PTRACE_EVENT_SECCOMP in ptrace man page). This patch supports both
behaviors by checking the kernel's actual behavior before installing the
seccomp filter.
* filter_seccomp.c: New file.
* filter_seccomp.h: New file.
* Makefile.am (strace_SOURCES): Add filter_seccomp.c and
filter_seccomp.h.
* linux/aarch64/arch_defs_.h (PERSONALITY0_AUDIT_ARCH,
PERSONALITY1_AUDIT_ARCH): Define for aarch64.
* linux/powerpc64/arch_defs_.h (PERSONALITY0_AUDIT_ARCH,
PERSONALITY1_AUDIT_ARCH): Likewise for powerpc64.
* linux/s390x/arch_defs_.h (PERSONALITY0_AUDIT_ARCH,
* linux/sparc64/arch_defs_.h (PERSONALITY0_AUDIT_ARCH,
PERSONALITY1_AUDIT_ARCH): Likewise for sparc64.
PERSONALITY1_AUDIT_ARCH): Likewise for s390x.
* linux/tile/arch_defs_.h (PERSONALITY0_AUDIT_ARCH,
PERSONALITY1_AUDIT_ARCH): Likewise for tile.
* linux/x32/arch_defs_.h (PERSONALITY0_AUDIT_ARCH,
PERSONALITY1_AUDIT_ARCH): Likewise for x32.
* linux/x86_64/arch_defs_.h (PERSONALITY0_AUDIT_ARCH,
PERSONALITY1_AUDIT_ARCH, PERSONALITY2_AUDIT_ARCH): Likewise for x86_64.
* linux/ia64/arch_defs_.h (PERSONALITY0_AUDIT_ARCH): Likewise for IA64.
* strace.c (usage): Document --seccomp-bpf option.
(startup_child): Mark process has having seccomp filter.
(exec_or_die): Initialize seccomp filtering if requested.
(init): Handle --seccomp-bpf option and check that seccomp can be
enabled.
(print_debug_info): Handle PTRACE_EVENT_SECCOMP.
(next_event): Capture PTRACE_EVENT_SECCOMP event.
(dispatch_event): Handle PTRACE_EVENT_SECCOMP event.
* trace_event.h (trace_event): New enumeration entity.
* strace.1.in: Document new --seccomp-bpf option.
* NEWS: Mention this change.
Co-authored-by: Paul Chaignon <paul.chaignon@gmail.com>
Co-Authored-by: Dmitry V. Levin <ldv@altlinux.org>
Recognize --help and --version options as aliases to -h and -V options,
respectively.
* strace.c: Include <getopt.h>.
(init): Move short options to optstring, add longopts array, use
getopt_long instead of getopt.
(usage): Document --help and --version options.
* strace.1.in: Likewise.
* tests/strace-V.test: Check that "strace --version" output is the same
as "strace -V" output.
Figure out whether the ioctl code is decoded inside a comment and adjust
printflags/printxval calls accordingly.
* ioctl.c (ioctl_print_code, evdev_decode_number): Add abbrev variable,
set it to true if xlat style is not XLAT_STYLE_VERBOSE, do not provide
dflt and set xlat style to XLAT_STYLE_ABBREV in printflags/printxval
calls (that are now changed to printflags_ex/printxval_ex to accomodate
the change).
Use null-terminating array instead of providing array size in the first
element.
* tests/ioctl_evdev-success.c (print_mtslots, main): Update
mtslots_str/invalid_mtslot_str array usage.
PAF_ARRAY_TRUNCATED allows enforcing the fact that an array
is truncated, which is useful for arrays in local memory that are known
as being truncated.
* defs.h (xlat_style_private_flag_bits): Add PAF_ARRAY_TRUNCATED_BIT.
(xlat_style_private_flags): Add PAF_ARRAY_TRUNCATED.
* util.c (print_array_ex): Handle PAF_ARRAY_TRUNCATED in flags.
Eugene Syromyatnikov
Dmitry V. Levin
Fanda Uchytil
Andreas Schwab
Chen Jingpiao
Paul Chaignon