tests: cleanup temporary files left by read-write test executable
This test executable is invoked by 19 tests, leaving about 100M
of data in temporary files.
In each of these tests read-write is invoked twice: the first time by
run_prog and the second time by run_strace. Since the temporary file
created during the first invocation is used by strace during the second
invocation, this file can only be cleaned up at the end of the second
invocation.
* tests/read-write.c (main): If the temporary file was not created,
unlink it before the exit.
With inclusion of device file and, later, pidfd decoding into -yy
semantics, it is clear that one cannot rely on stability of set
of file descriptor decoding features enabled by -yy. State that
formally by aliasing -yy to --decode-fds=all. Those who want only
specific FD decoding capabilities being enabled know that they need
it and can resort to --decode-fds=SET syntax.
* strace.1.in (.SS Output formatting): Change -yy alias
to --decode-fds=all, update the -yy option description.
* strace.c (usage): Likewise; expand --decode-fds description.
(init): Change yyflag_qual value to "all".
travis: remove build jobs covered by github actions
* .travis.yml: Remove all bionic build jobs on x86_64, x86, and x32 except
the first one which is probably the fasted and is kept for reference.
Remove all gcc-8 and gcc-9 jobs since the corresponding repositories
are broken.
Somewhat similar to Travis CI, this runs "make distcheck" on Ubuntu
18.04 using gcc-9, gcc-7, clang-9, and clang-6 on x86_64, x86, and x32
architectures.
Compared with Travis CI, GitHub Actions service currently provides
a significantly better parallelism as well as (unsurprisingly)
better integration with github.
However, GitHub Actions cannot replace Travis CI completely as
the latter can build on aarch64, s390x, and ppc64le architectures.
* .github/workflows/source-code-check.yml: Remove.
* .github/workflows/ci.yml: New file.
All invocations of poll-P executable must have a properly created
descriptor 9.
* tests/poll-P.test: Fix run_prog invocations.
Fixes: v4.21~122 "tests: check path tracing of poll syscall"
tests: remove poll-P and ppoll-P from pure_executables.list
Since these executables have to be invoked in a special way,
it was not correct to include them into pure_executables.list.
* tests/pure_executables.list: Move poll-P and ppoll-P ...
* tests/Makefile.am (check_PROGRAMS): ... here.
Fixes: v4.21~122 "tests: check path tracing of poll syscall"
Fixes: v4.21~121 "tests: check path tracing of ppoll syscall"
Silence the following incorrect compilation warning produced by clang:
tests/io_uring_register.c:271:10: error: format specifies type 'unsigned short' but the argument has type 'int' [-Werror,-Wformat]
probe->ops[i].flags & (typeof(probe->ops[i].flags)) ~1,
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
* tests/io_uring_register.c (main): Cast the whole
struct io_uring_probe_op.flags based expression
to the type of the first operand.
Complements: v5.5-183-g8b6487f9a "tests: fix clang compilation warning"
* .travis.yml [$CHECK == coverage]: Change dist from trusty to bionic,
change the compiler from gcc to gcc-9, change STACKTRACE from libunwind
to libdw.
* ci/run-build-and-tests.sh [$CHECK == coverage] (GCOV): New variable,
pass its value to ./codecov.bash.
As codecov-python, starting with commit
v2.0.15-44-g9f50cca7e5c5e0bb5e2859db238da08e3195e105 released almost
a month ago, became an utterly unusable pile of garbage, switch to
codecov-bash that still works.
* ci/install-dependencies.sh [$CHECK == coverage]: Do not install
lcov, python-pip, python-setuptools, and codecov. Download
https://codecov.io/bash and save it as codecov.bash.
* ci/run-build-and-tests.sh [$CHECK == coverage] (ac_cv_prog_LCOV,
ac_cv_prog_GENHTML): New variables, export them.
[[$CHECK == coverage]: Invoke codecov.bash instead of codecov.
Since many tests sleep a lot, running more of them in parallel
increases overall performance.
* ci/run-build-and-tests.sh (j2): New variable.
Use it instead of $j for "make check" and "make distcheck".
Silence the following compilation warning produced by clang:
tests/io_uring_register.c:271:10: error: format specifies type 'unsigned short' but the argument has type 'int' [-Werror,-Wformat]
probe->ops[i].flags & ~1,
^~~~~~~~~~~~~~~~~~~~~~~~
* tests/io_uring_register.c (main): Cast the second operand
of struct io_uring_probe_op.flags based expression to the type
of the first operand.
* xlat/bpf_commands.in (BPF_MAP_LOOKUP_BATCH): New constant introduced
by Linux commit v5.6-rc1~151^2~46^2~23^2~7.
(BPF_MAP_LOOKUP_AND_DELETE_BATCH): New constant introduced by Linux
commit v5.6-rc1~151^2~46^2~23^2~4.
(BPF_MAP_UPDATE_BATCH, BPF_MAP_DELETE_BATCH): New constants introduced
by Linux commit v5.6-rc1~151^2~46^2~23^2~6.
* bpf.c (BEGIN_BPF_CMD_DECODER(BPF_MAP_LOOKUP_BATCH),
BEGIN_BPF_CMD_DECODER(BPF_MAP_UPDATE_BATCH),
BEGIN_BPF_CMD_DECODER(BPF_MAP_DELETE_BATCH)): Decode the new commands.
(decode_BPF_MAP_LOOKUP_AND_DELETE_BATCH): New macro.
(SYS_FUNC(bpf)): Decode the new commands.
* bpf_attr.h (BPF_MAP_LOOKUP_BATCH_struct,
BPF_MAP_LOOKUP_AND_DELETE_BATCH_struct, BPF_MAP_UPDATE_BATCH_struct,
BPF_MAP_DELETE_BATCH_struct): New structs introduced by Linux commit
v5.6-rc1~151^2~46^2~23^2~7.
(BPF_MAP_LOOKUP_BATCH_struct_size,
BPF_MAP_LOOKUP_AND_DELETE_BATCH_struct_size,
BPF_MAP_UPDATE_BATCH_struct_size, BPF_MAP_DELETE_BATCH_struct_size): New
macros.
* NEWS: Mention this.
* tests/bpf.c (BPF_MAP_LOOKUP_BATCH_checks, BPF_MAP_UPDATE_BATCH_checks,
BPF_MAP_DELETE_BATCH_checks): Tests for the new commands.
Signed-off-by: Paul Chaignon <paul.chaignon@gmail.com>
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
libdw's address-to-symbol logic is quite slow. Since when stracing a
process, it is normal to see the same call sites over and over again,
let's address that by adding a straightforward N-way associative cache
in front of libdw.
The cache parameters could probably use some fine-tuning,
but ought to be good enough for now.
The resulting performance improvement:
$ time ./strace-orig -o /dev/null -k bash -c 'for i in {0..10000}; do echo; done' >/dev/null
real 0m12,014s
user 0m11,512s
sys 0m0,575s
$ time ./strace-cache -o /dev/null -k bash -c 'for i in {0..10000}; do echo; done' >/dev/null
real 0m0,765s
user 0m0,451s
sys 0m0,371s
Reviewed-By: Masatake YAMATO <yamato@redhat.com>
* unwind-libdw.c: Include "static_assert.h".
(STRACE_UW_CACHE_SIZE, STRACE_UW_CACHE_ASSOC): New macros.
(struct cache_entry): New type.
(struct ctx): Add "cache" field, extend the type of "last_proc_updating"
field.
(mapping_generation): Extend the type, initialize to 1.
(uwcache_clock): New variable.
(tcb_init): Initialize struct ctx.cache.
(truct frame_user_data): Add "ctx" field.
(find_bucket): New function.
(frame_callback): Use it.
(tcb_walk): Initialize user_data.ctx.
... because Linux kernel arch/sparc/kernel/syscalls/syscall.tbl
still lists it.
* linux/sparc/syscallent.h [267]: Add vserver entry.
* linux/sparc64/syscallent.h [267]: Add vserver entry.
hdio: factor out the HDIO_GETGEO handler into a separate function
And update it up to the current coding style.
* hdio.c: Include "print_fields.h".
(print_hdio_getgeo): New function.
(hdio_ioctl) <case HDIO_GETGEO>: Call print_hdio_getgeo instead
of decoding in place.
linux/32: move wait4 and [gs]ettimeofday from syscallent.h to syscallent-time32.h
According to Linux kernel include/uapi/asm-generic/unistd.h, starting
with commit v5.1-rc1~160^2~1^2~3 __NR_wait4, __NR_gettimeofday, and
__NR_settimeofday are placed under __ARCH_WANT_TIME32_SYSCALLS and
therefore should belong to syscallent-time32.h along with other 32-bit
time_t based syscalls.
* linux/32/syscallent.h [169, 170, 260]: Move entries ...
* linux/32/syscallent-time32.h: ... here.
* time.c (SYS_FUNC(gettimeofday), SYS_FUNC(settimeofday)): Place under
[HAVE_ARCH_TIME32_SYSCALLS || HAVE_ARCH_OLD_TIME64_SYSCALLS] guard.
* wait.c (SYS_FUNC(wait4)): Likewise.
Complements: v5.5-39-gcc7e89011 "linux/32: Split 32-bit time_t based syscalls into a separate header"
tests: fix another race condition in strace-DD.test
Apparently, timeout(1) utility first sends its signal to the child
process, then installs a signal handler, and finally sends its signal to
the process group. This creates a race between timeout(1) and strace,
the latter can manage to handle tracee syscall exit trap before a signal
to the process group is delivered.
* tests/setpgrp-exec.c: New file.
* tests/strace-DD.test: Rewrite without use of timeout(1).
* tests/strace-DD.expected: Remove.
* tests/Makefile.am (check_PROGRAMS): Add setpgrp-exec.
(EXTRA_DIST): Remove strace-DD.expected.
Fixes: v5.4~106 "tests: check -DD and -DDD options"
tests: add a workaround for trace_creds test on mips o32
Since every mips o32 executable used by trace_creds test invokes
prctl(PR_GET_FP_MODE) on their own and these calls are not filtered out
by %creds, they have to be filtered out manually before comparison.
* tests/init.sh (test_pure_prog_set) [$STRACE_ARCH == mips &&
$MIPS_ABI == o32 && $NAME == *creds]: Filter out prctl(PR_GET_FP_MODE)
from the strace log before comparison.
* strace.1.in (.SH OPTIONS) <--seccomp-bpf>: Remove the claim that this
option implies -f, document that --seccomp-bpf has no effect unless
-f/--follow-forks is also specified.
Complements: v5.5-38-g2a7ba2a97 "strace: disable seccomp-bpf instead of enabling following forks"
When Linux kernel merge window opens, changes of all sorts start to pour
down, breaking UAPI in various ways. This does not work for strace,
we need a stable UAPI.
* .gitlab-ci.yml (variables): Add KBRANCH.
* .travis.yml (env:global): Likewise.
Apparently, mips executables may invoke prctl(PR_GET_FP_MODE) on their
own. Since these calls are not filtered out by %creds, they have to be
dealt with before comparison.
* tests/getuid.test [$STRACE_ARCH == mips]: Filter out
prctl(PR_GET_FP_MODE) from the log of %creds tests.
Apparently, $SLEEP_A_BIT might be insufficient for the tracer
to complete.
* tests/strace-DD.test: Find out the tracer pid and wait for the tracer
completion before checking its log.
Suggested-by: Eugene Syromyatnikov <evgsyr@gmail.com>
Fixes: v5.4~106 "tests: check -DD and -DDD options"
getxxid used to test alpha-specific getxgid, getxuid, and getxgid
syscalls. Now each of these syscalls has its own test.
This hopefully fixes trace_creds test on alpha.
* tests/getxxid.c: Split into getxpid.c, getxuid.c, and getxgid.c.
* tests/gen_tests.in (getxxid): Remove.
(getxgid, getxpid, getxuid): New tests.
* tests/pure_executables.list: Replace getxxid with getxgid, getxpid,
and getxuid.
* tests/.gitignore: Likewise.
ia64: rename getpagesize syscall entry to old_getpagesize
... to match the Linux kernel that introduced __NR_old_getpagesize
by commit v5.0-rc1~164^2~4.
* linux/ia64/syscallent.h [BASE_NR + 147]: Rename "getpagesize"
to "old_getpagesize".
tests: disable umovestr_cached.test on mips o32 unless PTRACE_GET_SYSCALL_INFO works
If PTRACE_GET_SYSCALL_INFO does not work, then the test has to be
skipped on mips o32 because only the first four syscall arguments are
obtained from registers on that architecture, remaining syscall
arguments have to be fetched from the tracee memory.
* tests/umovestr_cached.test [$STRACE_ARCH == mips && $MIPS_ABI == o32]:
Skip the test if PTRACE_GET_SYSCALL_INFO does not work.
tests: re-enable umovestr_cached.test on ia64 if PTRACE_GET_SYSCALL_INFO works
If PTRACE_GET_SYSCALL_INFO works, then the exception made for ia64
earlier is not relevant.
* tests/umovestr_cached.test [$STRACE_ARCH == ia64]: Do not skip
the test if PTRACE_GET_SYSCALL_INFO works.
Complements: v5.3~23 "tests: disable umovestr_cached.test on ia64"
Introduced by Linux commit v5.6-rc1~122^2~4 and wired up in Linux commit
v5.6-rc1~122^2~3.
* pidfd_getfd.c: New file.
* Makefile.am (strace_SOURCES): Add it.
* linux/syscallent-common.h ([BASE_NR + 438]): Wire up pidfd_getfd.
* tests/pidfd_getfd.c: New file.
* tests/pidfd_getfd-y.c: Likewise.
* tests/.gitignore: Add pidfd_getfd and pidfd_getfd-y.
* tests/pure_executables.list: Likewise.
* tests/gen_tests.in: Add pidfd_getfd and pidfd_getfd-y tests.
And rename it to print_pid_fd in an attempt to minimise confusion with
the kernel's pidfd concept.
* kcmp (PRINT_FIELD_PIDFD, SYS_FUNC(kcmp)): Change printpidfd calls to
print_pid_fd.
(printpidfd): Move it...
* util.c (print_pid_fd): ...here. Drop the static qualifier, update
the comment describing why this is a stub.
* defs.h (print_pid_fd): New declaration.
Introduced by Linux commit v5.6-rc1~133^2~2.
* configure.ac (AC_CHECK_HEADERS): Check for linux/openat2.h.
* defs.h (print_numeric_umode_t_64): New declaration.
* linux/syscallent-common.h ([BASE_NR + 437]): Wire up openat2.
* types/openat2.h: New file.
* open.c: Include "print_fields.h", "types/openat2.h",
and "xlat/open_resolve_flags.h".
(sprint_open_modes64): Rename from sprint_open_modes, change argument
type to uint64_t.
(sprint_open_modes): New function, a wrapper for sprint_open_modes64
with argument type of unsigned int.
(tprint_open_modes64): Rename from tprint_open_modes, change argument
type to uint64_t, call sprint_open_modes64 instead of sprint_open_modes.
(tprint_open_modes): New function, a wrapper for tprint_open_modes64
with an argument of type unsigned int.
(print_open_how): New function.
(SYS_FUNC(openat2)): New decoder.
* printmode.c (print_numeric_umode_t_64): New function.
* tests/openat2.c: New file.
* tests/openat2-Xabbrev.c: Likewise.
* tests/openat2-Xraw.c: Likewise.
* tests/openat2-Xverbose.c: Likewise.
* tests/openat2-v.c: Likewise.
* tests/openat2-v-y.c: Likewise.
* tests/openat2-v-y-Xabbrev.c: Likewise.
* tests/openat2-v-y-Xraw.c: Likewise.
* tests/openat2-v-y-Xverbose.c: Likewise.
* tests/openat2-y.c: Likewise.
* tests/.gitignore: Add openat2, openat2-Xabbrev, openat2-Xraw,
openat2-Xverbose, openat2-v, openat2-v-y, openat2-v-y-Xabbrev,
openat2-v-y-Xraw, openat2-v-y-Xverbose, openat2-y.
* tests/pure_executables.list: Likewise.
* tests/gen_tests.in: Add openat2, openat2-Xabbrev, openat2-Xraw,
openat2-Xverbose, openat2-v, openat2-v-y, openat2-v-y-Xabbrev,
openat2-v-y-Xraw, openat2-v-y-Xverbose, openat2-y tests.
* xlat/open_resolve_flags.in: New file.
As it is to be used elsewhere.
* clone.c (print_nonzero_bytes): Move it...
* util.c (print_nonzero_bytes): ...here. Remove static qualifier.
* defs.h (print_nonzero_bytes): New declaration.
Dmitry V. Levin
Eugene Syromyatnikov
Paul Chaignon
Jann Horn
Gleb Fotengauer-Malinovskiy