* block.c: Update for RVAL_DECODED.
Define BLKROTATIONAL and BLKZEROOUT.
(block_ioctl): Decode them. Use printnum_short, printnum_int,
printnum_long, printpair_int64, and umove_or_printaddr.
(print_blkpg_req, block_ioctl): Use umove_or_printaddr.
Decode as much data on entering syscall as possible.
* loop.c: Include <linux/ioctl.h> instead of <sys/ioctl.h>.
Update for RVAL_DECODED.
(decode_loop_info, decode_loop_info64): New functions.
(loop_ioctl): Use them. Decode LOOP_SET_STATUS, LOOP_SET_STATUS64,
LOOP_SET_FD, LOOP_CHANGE_FD, LOOP_CTL_ADD and LOOP_CTL_REMOVE on
entering syscall.
Print LOOP_SET_FD and LOOP_CHANGE_FD arguments using printfd.
* rtc.c: Update for RVAL_DECODED.
(print_rtc): Rename to print_rtc_time.
(decode_rtc_time, decode_rtc_wkalrm): New functions.
(rtc_ioctl): Use them.
[RTC_VL_READ]: Decode RTC_VL_READ.
Use printnum_long for RTC_IRQP_READ and RTC_EPOCH_READ decoding.
Add several generic integer pairs printing functions
Add functions to fetch and print pairs of integer types.
Note that these printers do not attempt to fetch data
in case of exiting(tcp) && syserror(tcp).
printnum_* printers will follow as soon as all callers are made ready
for this change.
* defs.h (printpair_int, printpair_long, printpair_int64): New
prototypes.
* util.c (DEF_PRINTPAIR): New macro.
(printpair_int, printpair_long, printpair_int64): New functions.
Add printnum_short and printnum_int64 in addition to already existing
printnum_int and printnum_long.
* defs.h (printnum_short, printnum_int64): New prototypes.
* util.c (DEF_PRINTNUM): New macro.
(printnum_int, printnum_long): Use DEF_PRINTNUM.
(printnum_short, printnum_int64): New functions.
printaddr is a simple function implementing "print NULL or address"
idiom.
umoven_or_printaddr is a wrapper around umoven that has the same return
value semantics as umoven but also prints the address when the data
is not going to be fetched (in case of exiting(tcp) && syserror(tcp))
or cannot be fetched (umoven fails).
umove_or_printaddr is a macro wrapper around umoven_or_printaddr that
mirrors umove wrapper around umoven.
* defs.h (printaddr, umoven_or_printaddr): New prototypes.
(umove_or_printaddr): New macro.
* util.c (printaddr, umoven_or_printaddr): New functions.
Implement a method for "on entering" parsers to return
"done with decoding" information to their callers.
If a syscall parser called from trace_syscall_entering returns a value
with RVAL_DECODED bit set, record this value in tcp->sys_func_rval
and use it in trace_syscall_exiting instead of calling the parser on
exiting syscall.
If an ioctl parser returned RVAL_DECODED, this would tell
SYS_FUNC(ioctl) that the decoding is finished but fallback printing of
the ioctl command argument is needed, while e.g. RVAL_DONE+1 would mean
that the decoding is finished and no fallback decoding is needed.
* defs.h (struct tcb): Add sys_func_rval.
(RVAL_DECODED): New macro.
* syscall.c (trace_syscall_entering): Initialize tcp->sys_func_rval
along with setting TCB_INSYSCALL flag.
(trace_syscall_exiting): Use tcp->sys_func_rval instead of calling
tcp->s_ent->sys_func when the former has RVAL_DECODED flag set.
Reset tcp->sys_func_rval along with clearing TCB_INSYSCALL flag.
* ioctl.c (sys_ioctl): Set RVAL_DECODED flag on exiting.
Print ioctl argument iff RVAL_DECODED is set and fallback decoding
is requested.
Dmitry V. Levin [Wed, 17 Jun 2015 20:58:33 +0000 (20:58 +0000)]
tests: skip stat32 test if struct stat is defined incorrectly
If stat.st_mode returned by syscall is 0, it means that the definition
of struct stat in <asm/stat.h> is incorrect. This is the case e.g. on
mips n32 where the only available stat syscall is 64-bit but at the same
time struct stat defined by <asm/stat.h> is 32-bit.
* tests/stat.c (main) [NR_stat]: Return 77 if st_mode is 0.
Dmitry V. Levin [Wed, 17 Jun 2015 20:09:13 +0000 (20:09 +0000)]
Consistenly use #ifdef to check for AC_DEFINE'd macros
* aio.c (print_common_flags): Use #ifdef in the check for
HAVE_STRUCT_IOCB_U_C_FLAGS.
* defs.h: Use #ifdef in the check for HAVE_LITTLE_ENDIAN_LONG_LONG.
* net.c: Use #ifdef in the check for HAVE_SENDMSG.
* syscall.c (dumpio): Likewise.
* printstat.h (DO_PRINTSTAT): Use #ifdef in checks for
HAVE_STRUCT_STAT_ST_* macros.
* util.c: Use #ifdef in checks for HAVE_SYS_XATTR_H.
Dmitry V. Levin [Tue, 16 Jun 2015 10:53:25 +0000 (10:53 +0000)]
mount: update parser to match kernel behaviour
* mount.c (sys_mount): Do not decode type and data strings for
MS_SHARED, MS_PRIVATE, MS_SLAVE, and MS_UNBINDABLE mount flags
that do not imply valid strings.
Dmitry V. Levin [Mon, 25 May 2015 20:41:02 +0000 (20:41 +0000)]
Introduce memory allocation wrappers
Introduce wrappers to the following functions that do memory allocation:
malloc, calloc, realloc, strdup.
This commit is a follow-up to the related discussions in strace-devel ML:
http://sourceforge.net/p/strace/mailman/message/33618180/
http://sourceforge.net/p/strace/mailman/message/33733470/
* defs.h (xmalloc, xcalloc, xreallocarray, xstrdup): New prototypes.
* xmalloc.c: New file.
* Makefile.am (strace_SOURCES): Add it.
* count.c (count_syscall, call_summary_pers): Use xcalloc.
* desc.c (decode_select): Use xmalloc.
* dirent.c (sys_getdents, sys_getdents64): Likewise.
* net.c (sys_recvmmsg): Use xstrdup.
* pathtrace.c (storepath): Use xreallocarray.
(pathtrace_match): Use xmalloc.
* strace.c (die_out_of_memory): Move to xmalloc.c.
(expand_tcbtab): Use xcalloc and xreallocarray.
(startup_child): Use xstrdup.
(init): Use xmalloc, xcalloc, and xstrdup.
* syscall.c (reallocate_qual): Use xreallocarray.
(qualify): Use xstrdup.
* unwind.c (unwind_tcb_init): Use xmalloc.
(build_mmap_cache): Use xcalloc, xreallocarray, and xstrdup.
(get_symbol_name): Use xreallocarray.
(stacktrace_walk, queue_put): Use xmalloc.
* util.c (printstr): Use xmalloc.
* vsprintf.c (strace_vfprintf): Likewise.
Dmitry V. Levin [Thu, 21 May 2015 16:19:40 +0000 (16:19 +0000)]
Fix ioctl entries on 32-bit architectures with 64-bit aligned structures
Some ioctl constants are defined to different values on those 32-bit
architectures where structures containing a 64-bit field are aligned on
a 8-byte boundary and have a size multiple of 8 bytes, and on other
32-bit architectures.
* configure.ac: Check the size of a structure containing a 64-bit field.
* linux/32/ioctls_inc.h: Rename to ...
* linux/32/ioctls_inc_align32.h: ... this.
* linux/32/ioctls_inc_align64.h: New file.
* linux/32/ioctls_inc.h: Include either 32/ioctls_inc_align32.h
or ioctls_inc_align64.h depending on SIZEOF_STRUCT_I64_I32.
* Makefile.am (EXTRA_DIST): Add linux/32/ioctls_inc_align32.h and
* linux/32/ioctls_inc_align64.h.
* tests/ioctl.c (main): Test for VIDIOC_ENUMINPUT.
* tests/ioctl.expected: Check VIDIOC_ENUMINPUT decoding.
tests: workaround old gawk versions that do not provide @include support
Some of our tests use GNU awk @include extension, but not all currently
used versions of GNU awk support it. Skip these tests if gawk does not
provide @include support.
* tests/init.sh (check_gawk): New function.
(match_awk): Use it.
Dmitry V. Levin [Tue, 31 Mar 2015 19:45:08 +0000 (19:45 +0000)]
umovestr: read chunks of memory up to pagesize at a time
* defs.h (get_pagesize): New prototype.
* mem.c (get_pagesize) Make global.
* util.c (PAGMASK): Remove.
(vm_read_mem): New process_vm_readv proxy function.
(umoven, umovestr): Use it.
(umovestr): Read chunks up to pagesize at a time.
Dmitry V. Levin [Mon, 30 Mar 2015 15:21:55 +0000 (15:21 +0000)]
When process_vm_readv fails with EPERM, try PTRACE_PEEKDATA
process_vm_readv() and ptrace(PTRACE_PEEKDATA) have inconsistent access
control rules wrt traced processes: process_vm_readv() is more likely to
fail with EPERM than ptrace(PTRACE_PEEKDATA) when tracing a process that
has execve'd a privileged executable.
* util.c (umoven, umovestr): If process_vm_readv returned EPERM,
fall back to ptrace(PTRACE_PEEKDATA).
Reported-by: Andrew Guertin <lists@dolphinling.net>
Dmitry V. Levin [Fri, 27 Mar 2015 23:28:15 +0000 (23:28 +0000)]
signal.c: move siginfo_t parsers to a separate file
* printsiginfo.c: New file.
* Makefile.am (strace_SOURCES): Add it.
* defs.h (printsiginfo): Change second argument's type from int to bool.
* signal.c: Stop defining siginfo_t related constants.
Move inclusion of xlat/sig*_codes.h files to printsiginfo.c
(printsigsource, printsigval, printsiginfo_at): Move to printsiginfo.c.
(printsiginfo): Change second argument's type from int to bool,
split, and move to printsiginfo.c.
Denys Vlasenko [Wed, 25 Mar 2015 13:16:08 +0000 (14:16 +0100)]
sparc: delete personality 1
Personality 1 for sparc and sparc64 appears to be an old attempt
to support stracing Solaris binaries.
It stalled after the only syscall, solaris_open, was covered:
all other solaris_foo's are printargs.
This change deletes personality 1. Now sparc is an one-personality
arch, and sparc64 is a two-personality one.
For clarity, pure renaming of personality 2 files to personality 1 for sparc64
is performed in the next commit - this prevents a state where syscallent1.h
seems to indergo a lot of changes, where in fact in is deleted, and then
replaced by syscallent2.h
Fix this by updating ARM syscall sanity check.
In particular, get_scno() should not set TCB_INSYSCALL flag because
other code assumes that s_ent is properly initialized when this flag
is set.
* linux/arm/get_scno.c: Check syscall number after fetching.
Do not apply the check to SCNO_IN_RANGE syscalls.
Do not set TCB_INSYSCALL flag.
Extend diagnostics and move it to debug level.
Dmitry V. Levin [Sun, 22 Mar 2015 22:13:55 +0000 (22:13 +0000)]
syscall.c: split arch specific code into separate arch files
Split code that use arch-specific registers to separate arch files.
* syscall.c: Move definitions of variables containing fetched registers
to linux/*/arch_regs.c files.
[HAVE_GETRVAL2] (getrval2): Move arch-specific code
to linux/*/arch_getrval2.c, include "arch_getrval2.c".
(print_pc): Move arch-specific code to linux/*/print_pc.c files,
include "print_pc.c".
[X86_64] (x86_64_getregs_old): Rename to getregs_old, move to
linux/x86_64/getregs_old.c, include "getregs_old.c".
[POWERPC] (powerpc_getregs_old): Rename to getregs_old, move to
linux/powerpc/getregs_old.c, include "getregs_old.c".
(get_regs) [X86_64, POWERPC]: Update callers.
(get_scno): Move arch-specific code to linux/*/get_scno.c,
include "get_scno.c".
(get_syscall_args): Move arch-specific code
to linux/*/get_syscall_args.c, include "get_syscall_args.c".
(get_error): Move arch-specific code to linux/*/get_error.c,
include "get_error.c".
(get_syscall_result): Move arch-specific code
to linux/*/get_syscall_result.c, include "get_syscall_result.c".
* Makefile.am (EXTRA_DIST): Add new linux/*/*.c files.
Dmitry V. Levin [Mon, 23 Mar 2015 18:48:32 +0000 (18:48 +0000)]
Move get_regs error check from trace_syscall_entering to get_scno
Starting with commit v4.10-25-g8497b62, get_scno() can be called outside
trace_syscall_entering(), so move the get_regs_error check from
trace_syscall_entering() to get_scno().
Denys Vlasenko [Sat, 21 Mar 2015 16:51:52 +0000 (17:51 +0100)]
Show the syscall name in "resuming interrupted call" message
When signal is received, or if we have attached to a process,
current syscall (if process is in one) gets restarted.
Some syscalls are restarted via "restart_syscall()" mechanism.
On such sycalls, we don't show _which_ syscall gets restarted.
IOW: users want to see "resuming interrupted nanosleep"
instead of "resuming interrupted call" when they attach to "sleep 999".
Kernel does expose this information. The only thing we need is
to fetch syscall# on attach, and save it.
This patch does this. It adds tcp->s_prev_ent, which is
a pointer to struct_sysent of the previous syscall of this tracee.
It can be NULL.
sys_restart_syscall() is made to use it when the message is generated.
To similarly handle restart_syscall() *after signals*, not just
on attach, on each syscall exit patch saves exited syscall's data
in the same member (tcp->s_prev_ent).
Dmitry V. Levin [Thu, 19 Mar 2015 22:58:22 +0000 (22:58 +0000)]
sparc, sparc64: fix decoding of mmap2
* linux/sparc/syscallent.h (mmap2): Decode with sys_mmap_4koff,
not sys_mmap.
* linux/sparc64/syscallent2.h (mmap2): Decode with sys_mmap_pgoff,
not sys_mmap.
Dmitry V. Levin [Wed, 18 Mar 2015 16:32:04 +0000 (16:32 +0000)]
aarch64, arm: decode extra padded compat struct statfs64
According to arch/arm/kernel/sys_oabi-compat.c,
struct statfs64 has extra padding with EABI.
* statfs.c [AARCH64 || defined ARM] (COMPAT_STATFS64_PADDED_SIZE):
Define.
(do_statfs64_fstatfs64): New function, factored out from sys_statfs64.
[COMPAT_STATFS64_PADDED_SIZE]: Check it in addition to
sizeof(struct compat_statfs64).
(sys_statfs64, sys_fstatfs64): Use do_statfs64_fstatfs64.
Dmitry V. Levin [Mon, 16 Mar 2015 18:10:21 +0000 (18:10 +0000)]
tests/uid*: use fchown* instead of chown*
Newer architectures have no chown syscall, so use fchown* syscalls
for testing printuid.
* tests/uid.test: Use fchown instead of chown.
* tests/uid.c: Test __NR_fchown instead of __NR_chown.
(main): Use __NR_fchown instead of __NR_chown.
* tests/uid32.c: Test __NR_fchown32 instead of __NR_chown32.
(main): Use __NR_fchown32 instead of __NR_chown32.
* tests/uid16.c: Test __NR_fchown and __NR_fchown32 instead
of __NR_chown and __NR_chown32.
(main): Use __NR_fchown instead of __NR_chown.
* tests/uid.awk: Update regexp.
projects / strace / log