IF YOU WOULD LIKE TO GET AN ACCOUNT, please write an
email to Administrator. User accounts are meant only to access repo
and report issues and/or generate pull requests.
This is a purpose-specific Git hosting for
BaseALT
projects. Thank you for your understanding!
Только зарегистрированные пользователи имеют доступ к сервису!
Для получения аккаунта, обратитесь к администратору.
Introduce "flags" field to struct inject_data and use it instead
of magic rval and signo constants.
Due to layout of struct inject_data, this new field does not change
sizeof(struct inject_data).
* defs.h (INJECT_F_RETVAL, INJECT_F_SIGNAL): New macros.
(INJECT_OPTS_RVAL_DEFAULT): Remove macro.
(struct inject_data): Add "flags" field.
* filter_qualify.c (parse_inject_token, qualify_inject_common): Check
struct inject_data.flags instead of inject_data.rval
and inject_data.signo, do not initialize inject_opts.data.
* syscall.c (tamper_with_syscall_entering): Check struct
inject_data.flags instead of inject_data.rval and inject_data.signo.
* defs.h (inject_data): New structure.
(struct inject_opts): Replace "signo" and "rval" fields with "data"
field of type "struct inject_data".
* filter_qualify.c (parse_inject_token, qualify_inject_common): Update
for the new layout of struct inject_opts.
* syscall.c (tamper_with_syscall_entering, tamper_with_syscall_exiting):
Likewise.
Co-authored-by: Victor Krapivensky <krapivenskiy.va@phystech.edu>
I see no reason for not having these definitions for SPARC64 and RISC-V.
Also, borrow the check template from supported_personalities.h.
* syscall.c (update_personality) [defined RISCV, defined SPARC64]
<PERSONALITY_NAMES>: Define it for these architectures.
Co-authored-by: Dmitry V. Levin <ldv@altlinux.org>
As Chris Metcalf, the author of v4.7-40-g0b99a8a, told in private, there
is no strong reason for having tile's personality name different from
personality names on other architectures, so let's use spaces instead of
dashes here. And, while we are here, remove duplication of personality
name definitions for various architectures.
* syscall.c (update_personality) [defined TILE]: Change dashes to spaces
in personality names.
* number_set.h (number_slot_t, struct number_set): Move to number_set.c.
(struct number_set): Add forward declaration.
(read_set, write_set, signal_set): Change prototypes from objects
to pointers.
* filter_qualify.c (read_set, write_set, signal_set): Change definitions
from objects to pointers.
(abbrev_set, inject_set, raw_set, trace_set, verbose_set): Change
definitions from arrays to pointers.
(qualify_read): Initialize read_set before first use.
(qualify_write): Initialize write_set before first use.
(qualify_signals): Initialize signal_set before first use.
(qualify_trace): Initialize trace_set before first use.
(qualify_abbrev): Initialize abbrev_set before first use.
(qualify_verbose): Initialize verbose_set before first use.
(qualify_raw): Initialize raw_set before first use.
(qualify_inject_common): Initialize inject_set before first use.
* strace.c (print_signalled, print_stopped): Update signal_set usage.
* syscall.c (dumpio): Update usage of read_set and write_set.
Avoid proliferation of exactly the same definitions of shorthand
notations for macros defined in sysent.h by moving definitions
and undefs of these shorthand notations to separate files.
* sysent_shorthand_defs.h: New file.
* sysent_shorthand_undefs.h: Likewise.
* Makefile.am (strace_SOURCES): Add them.
* syscall.c: Use them.
Signed-off-by: Edgar Kaziakhmedov <edgar.kaziakhmedov@virtuozzo.com>
Signed-off-by: Dmitry V. Levin <ldv@altlinux.org>
Move is_negated_errno() to a separate new header file negated_errno.h
and include it just for architectures which require it.
is_negated_errno() is not used on those architectures
that have a dedicated register to signal a syscall error.
The issue was raised when compiling with clang, which is more
strict regarding semantics of unused static inline functions
defined in C files and will issue a -Wunused-function warrning
if they are not used anywhere.
* syscall.c (is_negated_errno): Move to ...
* negated_errno.h: ... new file.
* Makefile.am (strace_SOURCES): Add it.
* linux/aarch64/get_error.c: Include it.
* linux/arc/get_error.c: Likewise.
* linux/arm/get_error.c: Likewise.
* linux/avr32/get_error.c: Likewise.
* linux/bfin/get_error.c: Likewise.
* linux/crisv10/get_error.c: Likewise.
* linux/hppa/get_error.c: Likewise.
* linux/i386/get_error.c: Likewise.
* linux/ia64/get_error.c: Likewise.
* linux/m68k/get_error.c: Likewise.
* linux/metag/get_error.c: Likewise.
* linux/microblaze/get_error.c: Likewise.
* linux/or1k/get_error.c: Likewise.
* linux/riscv/get_error.c: Likewise.
* linux/s390/get_error.c: Likewise.
* linux/sh/get_error.c: Likewise.
* linux/sh64/get_error.c: Likewise.
* linux/tile/get_error.c: Likewise.
* linux/x86_64/get_error.c: Likewise.
* linux/xtensa/get_error.c: Likewise.
This change removes the trace_syscall function. Now, the code that uses
syscall.c trace functions is expected to check whether it is a syscall
entry or exit (with entering(tcp)/exiting(tcp)) itself, and then make
an appropriate sequence of function calls.
* defs.h: Update comment on TCB_INSYSCALL.
(trace_syscall): Remove prototype.
(syscall_entering_decode, syscall_entering_trace,
syscall_entering_finish, syscall_exiting_decode, syscall_exiting_trace,
syscall_exiting_finish): New prototypes.
* strace.c (trace_syscall): New static replacement for old trace_syscall.
* syscall.c (trace_syscall): Remove.
(trace_syscall_entering): Split into ...
(syscall_entering_decode, syscall_entering_trace,
syscall_entering_finish): ... new functions.
(trace_syscall_exiting): Split into ...
(syscall_exiting_decode, syscall_exiting_trace,
syscall_exiting_finish): ... new functions.
This saves up to 25% of ptrace syscalls in case of trace filtering.
* syscall.c (trace_syscall_exiting): Do not call gettimeofday
and get_regs for filtered syscalls.
* NEWS: Mention this change.
* tests/get_regs.test: New test.
* tests/Makefile.am (MISC_TESTS): Add it.
strace used to call get_regs on every stop of a tracee, including cases
when the result is going to be discarded. Prepare to change this lavish
practice by delaying get_regs invocation until its result is actually
needed.
* defs.h (get_regs): Remove.
* strace.c (trace): Do not call get_regs.
* syscall.c (get_regs): Add static qualifier.
(trace_syscall_exiting, print_pc, get_scno): Call get_regs.
At this moment there is only one call to get_regs,
but there is going to be more of them, so a guard is needed.
This change also adds a clear_regs call before get_regs,
which essentially reverts commit v4.9-288-ge9bfff6.
* strace.c (trace): Call clear_regs before get_regs.
* syscall.c (get_regs): Skip if get_regs_error != -1.
* syscall.c (trace_syscall_entering, trace_syscall_exiting): Move
definitions of automatic variables closer to their first use. This
change slightly improves readability.
Drop use of INJECT_OPTS_RVAL_DISABLE special inject_opts.rval value,
INJECT_OPTS_RVAL_DEFAULT is enough.
* defs.h (INJECT_OPTS_RVAL_DISABLE): Remove.
* qualify.c (qualify_inject_common): Do not replace
INJECT_OPTS_RVAL_DEFAULT with INJECT_OPTS_RVAL_DISABLE
in inject_opts.rval.
* syscall.c (tamper_with_syscall_entering): Check
for INJECT_OPTS_RVAL_DEFAULT instead of INJECT_OPTS_RVAL_DISABLE.
As -e fault= injection syntax has been extended to configure various
kinds of injections besides fault injection, the original option name
bacame too narrow.
This change renames internal constants, types, and functions from
"fault" to more generic inject/tamper ones.
* defs.h (fault_opts): Rename to inject_opts.
(FAULT_OPTS_RVAL_DEFAULT): Rename to INJECT_OPTS_RVAL_DEFAULT.
(FAULT_OPTS_RVAL_DISABLE): Rename to INJECT_OPTS_RVAL_DISABLE.
(struct tcb): Rename fault_vec field to inject_vec.
(TCB_FAULT_INJ): Rename to TCB_TAMPERED.
(QUAL_FAULT): Rename to QUAL_INJECT.
(fault_vec): Rename to inject_vec.
All users changed.
* qualify.c (fault_set): Rename to inject_set.
(parse_fault_token): Rename to parse_inject_token.
(parse_fault_expression): Rename to parse_inject_expression.
All callers changed.
* syscall.c (fault_vec): Rename to inject_vec.
(tcb_fault_opts): Rename to tcb_inject_opts.
(inject_syscall_fault_entering): Rename to tamper_with_syscall_entering.
(update_syscall_fault_exiting): Rename to tamper_with_syscall_exiting.
(syscall_fault_injected): Rename to syscall_tampered.
All callers changed.
Limit the number of files where <asm/unistd.h> is included
to only those few that need it.
* defs.h: Do not include <asm/unistd.h>.
* clone.c: Include <asm/unistd.h>.
* strace.c: Likewise.
* syscall.c: Likewise.
* util.c: Likewise.
This extends the fault injection capability with :signal=SIG option
which injects a signal on entering each syscall from the specified set.
:signal and :error options are complementary, if they are both specified
the syscall will be fault injected as usual and the specified signal
will be delivered to the tracee.
* defs.h (struct fault_opts): Change the type of err field to int16_t,
add signo field.
(trace_syscall): Add a pointer argument.
* qualify.c: Include "nsig.h".
(parse_fault_token): Handle signal= option.
(qualify_fault): Update default fault_opts.
* strace.c (trace): Forward signal number from trace_syscall
to ptrace_restart(PTRACE_SYSCALL).
* syscall.c (inject_syscall_fault_entering): Add pointer argument,
save there the signal number specified by fault options. Do not inject
a syscall fault unless instructed by fault options.
(update_syscall_fault_exiting): Update the error code injection check.
(trace_syscall_entering): Add pointer argument, forward it to
inject_syscall_fault_entering.
(trace_syscall): Add pointer argument, forward it to
trace_syscall_entering.
Signed-off-by: Seraphime Kirkovski <kirkseraph@gmail.com>
Signed-off-by: Dmitry V. Levin <ldv@altlinux.org>
Remove temporary macros created for transition from long
to kernel_ulong_t.
Automatically replace PRI_kr[dux] with PRI_kl[dux] using
$ git grep -l 'PRI_kr[dux]' | xargs sed -ri 's/PRI_kr([dux])/PRI_kl\1/g'
* defs.h (PRI_krd, PRI_kru, PRI_krx): Remove. All users updated.
Remove temporary types created for transition from long
to kernel_ulong_t.
Automatically replace kernel_scno_t and kernel_ureg_t with
kernel_ulong_t using
$ git grep -El 'kernel_(scno|ureg)_t' |
xargs sed -ri 's/kernel_(scno|ureg)_t/kernel_ulong_t/g'
* kernel_types.h (kernel_scno_t, kernel_ureg_t): Remove.
All users updated.
Prepare for the change of kernel_ureg_t and kernel_scno_t from
unsigned long to kernel_ulong_t.
* defs.h (PRI_krd, PRI_kru, PRI_krx): New temporary macros.
All users updated.
* defs.h (upeek, upoke): Change offset argument type from long
to unsigned long. Change res argument type from long to kernel_ureg_t.
* upeek.c (upeek): Likewise.
* upoke.c (upoke.c): Likewise.
* syscall.c (print_pc) [ARCH_PC_PEEK_ADDR]: Change pc type from long
to kernel_ureg_t.
* linux/aarch64/arch_regs.c (arm_sp_ptr): Remove redundant cast.
* linux/metag/get_syscall_args.c (get_syscall_args): Likewise.
* linux/sh/get_syscall_result.c (get_syscall_result_regs): Likewise.
* linux/sh64/get_syscall_result.c (get_syscall_result_regs): Likewise.
* linux/powerpc/getregs_old.c (getregs_old): Remove redundant casts.
* linux/alpha/arch_getrval2.c (getrval2): Change r20 type from long
to unsigned long.
* linux/alpha/arch_regs.c (alpha_r0, alpha_a3): Change type from long
to unsigned long.
* linux/bfin/arch_regs.c (bfin_r0): Likewise.
* linux/crisv10/arch_regs.c (cris_r10): Likewise.
* linux/hppa/arch_regs.c (hppa_r28): Likewise.
* linux/ia64/arch_regs.c (IA64_PSR_IS): Likewise.
* linux/microblaze/arch_regs.c (microblaze_r3): Likewise.
* linux/sh/arch_regs.c (sh_r0): Likewise.
* linux/sh64/arch_regs.c (sh64_r9): Likewise.
* linux/xtensa/arch_regs.c (xtensa_a2): Likewise.
* linux/alpha/arch_sigreturn.c (arch_sigreturn): Change addr type
from long to unsigned long.
* linux/microblaze/arch_sigreturn.c (arch_sigreturn): Likewise.
* linux/alpha/get_scno.c (arch_get_scno): Update for the change
of signedness.
* linux/arc/get_syscall_args.c (get_syscall_args): Change arc_args type
from pointer to long to pointer to unsigned long.
* linux/arm/arch_regs.c (arm_sp_ptr): Change type from pointer to long
to pointer to unsigned long.
* linux/arm/arch_regs.h (arm_sp_ptr): Likewise.
* linux/i386/arch_regs.c (i386_esp_ptr): Likewise.
* linux/i386/arch_regs.h (i386_esp_ptr): Likewise.
* linux/m68k/arch_regs.c (m68k_usp_ptr): Likewise.
* linux/m68k/arch_regs.h (m68k_usp_ptr): Likewise.
* linux/ia64/get_syscall_args.c (get_syscall_args): Use umove
instead of umoven.
* linux/sh/arch_getrval2.c (getrval2): Change val type from long
to unsigned long.
* defs.h (dumpiov_in_msghdr): Change data_size argument type
from unsigned long to kernel_ureg_t.
* msghdr.h (print_struct_msghdr): Likewise.
* msghdr.c (print_struct_msghdr): Likewise. Cast msghdr.msg_iovlen
and msghdr.msg_controllen to kernel_ureg_t instead of unsigned long.
(decode_msghdr, dumpiov_in_msghdr): Change data_size argument type
from unsigned long to kernel_ureg_t.
(SYS_FUNC(sendmsg)): Pass -1 instead of -1UL as data_size argument
to decode_msghdr.
* mmsghdr.c (print_struct_mmsghdr): Pass (kernel_ureg_t) -1 instead
of -1UL as data_size argument to print_struct_msghdr.
(dumpiov_in_mmsghdr): Cast msghdr.msg_iovlen to kernel_ureg_t instead
of unsigned long.
* syscall.c (dumpio): Pass -1 instead of -1UL as data_size argument
to dumpiov_in_msghdr
x32 syscalls starting with number 512 are compat syscalls that operate
on compat types. Set COMPAT_SYSCALL_TYPES flag to these syscall entries
and make get_syscall_args truncate arguments of syscalls that have this
flag set, to avoid unpleasant alternative of adding multiple
widen_to_ulong invocations to individual syscall parsers.
preadv, pwritev, preadv2, and pwritev2 are exceptions: while these
syscalls operate on compat types, the offset argument has a regular
64-bit type, so the automatic argument truncation cannot be applied.
* defs.h (COMPAT_SYSCALL_TYPES): New macro.
* syscall.c (CST): New macro, defined to COMPAT_SYSCALL_TYPES.
* linux/ia64/syscallent.h: Do not include "../i386/syscallent.h"
unless [CST > 0].
* linux/x32/syscallent.h [512..533, 536..545]: Add CST flag.
* linux/x86_64/get_syscall_args.c (get_syscall_args): Truncate arguments
of compat syscalls to 32-bit values.
* tests/ksysent.c (CST): New macro, defined to 0.
* tests/nsyscalls.c (CST): Likewise.
current_klongsize checks are more universal, therefore they are usually
shorter and easier to comprehend.
* desc.c (umove_kulong_array_or_printaddr): Check for current_klongsize
instead of current_personality and current_wordsize.
* io.c (print_lld_from_low_high_val): Likewise.
* lseek.c (SYS_FUNC(lseek)): Likewise.
* resource.c (decode_rlimit): Likewise.
* syscall.c (is_negated_errno): Likewise.
* util.c (getllval, getarg_klu): Likewise.
