Denys Vlasenko [Fri, 15 Feb 2013 14:25:37 +0000 (15:25 +0100)]
Fix build error on Tile
* syscall.c (get_scno): [TILE] Remove TCB_WAITEXECVE check,
it is never true on Tile, and stopped compiling when
TCB_WAITEXECVE define was removed for Tile.
Denys Vlasenko [Fri, 15 Feb 2013 14:01:38 +0000 (15:01 +0100)]
x86: zero-extend 32-bit args in syscall entry instead of sign-extension
Zero-extension is slightly more common that sign-extension:
all pointers are zero-extended, and some other params are unsigned.
Whereas signed ones (fds, pids, etc) are often treated as
_32-bit ints_ even by kernel, so just unconditionally casting
such tcp->u_arg[N] to int works.
* syscall.c (get_syscall_args): [X86] Zero-extend 32-bit args
instead of sign-extension.
Denys Vlasenko [Fri, 15 Feb 2013 13:58:52 +0000 (14:58 +0100)]
Macroize conditional signed widening operation
* defs.h: Define widen_to_long() macro.
* signal.c (sys_kill): Use it instead of open-coding it.
(sys_tgkill): Use widen_to_long() on pids.
* resource.c (decode_rlimit): Formatting fix.
Denys Vlasenko [Fri, 15 Feb 2013 13:55:14 +0000 (14:55 +0100)]
A better handling of current_wordsize
On x86_64:
text data bss dec hex filename
435661 26628 47424 509713 7c711 strace_old
435501 26612 47440 509553 7c671 strace_new_clever_wordsize
On x32 and arm it should be even better, current_wordsize becomes
a constant there.
* defs.h: Declare current_wordsize as a variable if needed,
else declare as a constant define.
Remove declatation of personality_wordsize[].
* syscall.c: Make personality_wordsize[] static.
Declare current_wordsize as a variable if needed.
(set_personality): Set current_wordsize only if non-constant.
Denys Vlasenko [Thu, 14 Feb 2013 02:29:48 +0000 (03:29 +0100)]
[X86] Use ptrace(PTRACE_GETREGSET, NT_PRSTATUS) to get registers.
Unlike PTRACE_GETREGS, this new method detects 32-bit processes
reliably, without checking segment register values which
are undocumented and aren't part of any sort of API.
While at it, also fixed x32 detection to use __X32_SYSCALL_BIT,
as it should have been from the beginning.
* defs.h: Declare os_release and KERNEL_VERSION.
* strace.c: Make os_release non-static, remove KERNEL_VERSION define.
* syscall.c: New struct i386_user_regs_struct,
static union x86_regs_union and struct iovec x86_io.
(printcall): Use i386_regs or x86_64_regs depending on x86_io.iov_len.
(get_regs): On x86 and kernels 2.6.30+, use PTRACE_GETREGSET,
on earlier kernels fall back to old method.
(get_scno): [X86] Determine personality based on regset size
on scno & __X32_SYSCALL_BIT.
(syscall_fixup_on_sysenter): Use i386_regs or x86_64_regs depending
on x86_io.iov_len.
(get_syscall_args): Likewise.
(get_error): Likewise.
Denys Vlasenko [Wed, 13 Feb 2013 15:31:32 +0000 (16:31 +0100)]
Factor out code to check addr, fetch and print siginfo
* defs.h: Declare new function printsiginfo_at(tcp, addr).
* process.c (sys_waitid): Use printsiginfo_at().
(sys_ptrace): Likewise.
* signal.c: (printsiginfo_at): Implement this new function.
(sys_rt_sigsuspend): Use printsiginfo_at().
(sys_rt_sigtimedwait): Likewise.
Denys Vlasenko [Tue, 12 Feb 2013 12:06:51 +0000 (13:06 +0100)]
Remove hacks for old kernels for architectures which require new kernels
* defs.h: Do not define TCB_WAITEXECVE for AARCH64.
* util.c (change_syscall): For AARCH64 and X32, replace code
with dummy "return 0" and a comment explaining why that is ok
for these architectures.
Denys Vlasenko [Tue, 12 Feb 2013 11:50:10 +0000 (12:50 +0100)]
Remove old kernel hacks for architectures which require new kernels
* defs.h: Do not define TCB_WAITEXECVE for AVR32, BFIN and TILE.
* util.c (change_syscall): For AVR32, BFIN and TILE, replace code
with dummy "return 0" and a comment explaining why that is ok
for these architectures.
Denys Vlasenko [Tue, 12 Feb 2013 10:57:48 +0000 (11:57 +0100)]
Handle recv[m]msg for non-native 32-bit personality syscalls
* net.c (printmsghdr): If current_wordsize is 4 and long is wider than it,
read 32-bit struct msghdr and expand it into a native one before using it.
(printmmsghdr): Likewise for struct mmsghdr.
Denys Vlasenko [Tue, 12 Feb 2013 10:43:46 +0000 (11:43 +0100)]
Remove unused / ambiguously used defines
We sometimes use LINUXSPARC and sometimes (more often)
use "defined(SPARC) || defined(SPARC64)". Better to use
one construct consistently.
LINUX_MIPS64 is altogether unused.
* defs.h: Remove LINUXSPARC and LINUX_MIPS64 defines.
Move PTRACE_xxx compat defines up, before arch-specific
machinery. Use defined(SPARC) || defined(SPARC64)
instead of LINUXSPARC.
* file.c: Use defined(SPARC) || defined(SPARC64) instead of LINUXSPARC.
* signal.c: Likewise.
Denys Vlasenko [Mon, 11 Feb 2013 11:29:36 +0000 (12:29 +0100)]
Preparatory patch for "new" x86 personality detection
* syscall.c: Move PT_FLAGS_COMPAT define to its only user, get_scno().
Rename arm_regs_union's fields to names less likely to collide with
system defines.
(get_regs): Use sizeof(arm_regs_union) instead of sizeof(aarch64_regs).
This should be the same, but htis way it's cleaner.
Remove __X32_SYSCALL_MASK and use __X32_SYSCALL_BIT instead.
Explain 64-bit check in X32 build better.
Mike Frysinger [Sat, 9 Feb 2013 00:10:07 +0000 (19:10 -0500)]
Fix decoding of sysctl() when oldval fields are NULL
If you call glibc's syscall wrapper like so:
static int name[] = { CTL_NET, NET_IPV4, NET_IPV4_LOCAL_PORT_RANGE };
int buffer[2] = { 32768, 61000 };
size_t size = sizeof(buffer);
sysctl(name, 3, 0, 0, buffer, size);
(note that oldval/oldlenp are NULL).
The current strace code complains like so:
_sysctl({{CTL_NET, NET_IPV4, NET_IPV4_LOCAL_PORT_RANGE, 38}, 3, process_vm_readv: Bad address
(nil), 0, 0x7fffe23c3960, 8}) = -1 EACCES (Permission denied)
Since passing NULL for the old values is valid, handle that explicitly.
This also simplifies the code a bit by splitting up the handling of the
new and old args so that we only handle the new args once.
Now the output looks like:
_sysctl({{CTL_NET, NET_IPV4, NET_IPV4_LOCAL_PORT_RANGE, 38}, 3, NULL, 0, 0x7fff8c0c91b0, 8) = -1 EACCES (Permission denied)
* system.c (sys_sysctl): Check if info.oldval is NULL first. Move the
processing of oldlen/info.newval/info.newlen out so they always get
executed. Fix the format strings so we use %lu for unsigned long rather
than a mix of %ld and %lu.
Denys Vlasenko [Fri, 8 Feb 2013 14:34:46 +0000 (15:34 +0100)]
Dying suddenly with abort() is rude, avoid if possible
* file.c (sys_utime): Don't call abort() if wordsize is strange.
Instead, warn user about it.
* desc.c (printflock): Use the same message string as in sys_utime.
Denys Vlasenko [Fri, 8 Feb 2013 11:59:13 +0000 (12:59 +0100)]
Remove vestigial hacks around non-Linux struct sigactions
* signal.c: Stop using __sighandler_t glibc'ism. Remove SA_HANDLER macro.
Explain why we can't use "sa_handler" as a field name.
(sys_sigaction): Use __sa_handler instead of SA_HANDLER macro.
(sys_rt_sigaction): Likewise.
Denys Vlasenko [Wed, 6 Feb 2013 17:24:39 +0000 (18:24 +0100)]
Rename ARM's regs structure to arm_regs
Compile-tested.
* defs.h: Rename regs structure to arm_regs.
* syscall.c: Likewise.
(printcall): Use new name instead of old one.
(get_regs): Likewise.
(get_scno): Likewise.
(get_syscall_args): Likewise.
(get_error): Likewise.
* signal.c (sys_sigreturn): Likewise.
Denys Vlasenko [Wed, 6 Feb 2013 12:18:42 +0000 (13:18 +0100)]
Shortcut tests for fork/exec syscalls
This change should speed up strace by a tiny bit.
More importantly, it makes it much more clear that
fork and exec fixups are not necessary for any reasonably
recent kernels. IOW: syscall_fixup_for_fork_exec() and its callees
are all dead code.
* defs.h: Declare new need_fork_exec_workarounds flag variable.
* strace.c: Define need_fork_exec_workarounds flag variable.
(test_ptrace_setoptions_followfork): Return 0/1 as success/fail indicator.
(test_ptrace_setoptions_for_all): Likewise.
(init): Set need_fork_exec_workarounds to TRUE if needed.
* syscall.c: Rename internal_syscall() to syscall_fixup_for_fork_exec().
(trace_syscall_entering): Call syscall_fixup_for_fork_exec() only if
need_fork_exec_workarounds == TRUE.
(trace_syscall_exiting): Likewise.
Chris Metcalf [Tue, 5 Feb 2013 16:48:33 +0000 (17:48 +0100)]
Add tilegx support to strace
tilegx support has been in the kernel since 3.0.
In addition, fix some issues with the tilepro support already
present in strace, primarily the decision to use the
<asm/unistd.h> numbering space for system calls.
* defs.h [TILE]: Include <asm/ptrace.h> and provide an extern
struct pt_regs tile_regs for efficiency. Provide compat 32-bit
personality via SUPPORTED_PERSONALITIES, PERSONALITY0_WORDSIZE,
PERSONALITY1_WORDSIZE, and DEFAULT_PERSONALITY.
* linux/tile/errnoent1.h: New file, includes linux/errnoent.h.
* linux/tile/ioctlent1.h: New file, includes linux/ioctlent.h.
* linux/tile/signalent1.h: New file, includes linux/signalent.h.
* linux/tile/syscallent.h: Update with new asm-generic syscalls.
The version previously committed was the from the first tile patch
to LKML, which subsequently was changed to use <asm-generic/unistd.h>.
* linux/tile/syscallent1.h: Copy from linux/tile/syscallent.h.
* mem.c (addtileflags) [TILE]: use %ld properly for a "long" variable.
* process.c [TILE]: Choose clone arguments correctly and properly
suppress all "struct user" related offsets in user_struct_offsets.
* signal.c [TILE]: Use tile_regs not upeek.
* syscall.c (update_personality) [TILE]: Print mode.
(PT_FLAGS_COMPAT) [TILE]: Provide if not in system headers.
(tile_regs) [TILE]: Define 'struct pt_regs' variable to hold state.
(get_regs) [TILE]: use PTRACE_GETREGS to set tile_regs rather than using upeek.
(get_scno) [TILE]: Set personality.
(get_syscall_args) [TILE]: Use tile_regs.
(get_syscall_result) [TILE]: Update tile_regs.
(get_error) [TILE]: Use tile_regs.
(printcall) [TILE]: Print pc.
(arg0_offset, arg1_offset, restore_arg0, restore_arg1) [TILE]:
Properly handle tile call semantics and support tilegx.
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com> Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
Analogous improvement in sys_sigreturn() is also implemented.
* defs.h: Declare extern struct pt_regs regs for SPARC[64] and ARM.
Declare clear_regs(), get_regs() and get_regs_error flag variable.
* strace.c (trace): Call get_regs(pid) as soon as we know the tcb
and that it is stopped.
* syscall.c (get_regs): New function. Used to fetch registers early,
just after tracee has stopped.
(printcall): Move it here from util.c. Use global regs.REG data,
if available on the arch, instead of re-fetching it.
(get_scno): Use global regs.REG data.
(get_syscall_result): Likewise.
* signal.c (sys_sigreturn): Likewise.
* util.c (printcall): Moved to syscall.c.
Stanislav Brabec [Mon, 10 Dec 2012 19:18:49 +0000 (20:18 +0100)]
Fix sys_semtimedop decoding on s390x
The s390 and s390x pass semtimedop arguments differently from other
architectures. sys_semtimedop parser was fixed for s390 by commit v4.6-177-ge0f5fd8, and s390x requires the same fix.
* linux/ipc.c (sys_semtimedop): Fix timespec decoding on s390x.
sys_semtimedop: fix timeval argument index in wrapped call
Looking at the implementation of wrapped semtimedop() call inside glibc
and kernel, I started to believe that timeval should be located in
tcp->u_arg[4] and not tcp->u_arg[5]. Fortunately, tcp->u_arg[5] now
works correctly as well, due to side effects of decode_ipc_subcall().
declaration in header:
int semtimedop(semid, *sops, nsops, *timeout);
0 1 2 3
sys_ipc arguments in glibc on all patforms except s390*:
semid, (int) nsops, 0, CHECK_N (sops, nsops), timeout
0 1 2 3 4
We have to use indexes: 0 3 1 4
sys_ipc arguments on s390*:
semid, (int) nsops, timeout, sops
0 1 2 3
We have to use indexes: 0 3 1 2
* ipc.c (sys_semtimedop) [!S390]: Fix timeval argument index in
indirect_ipccall case.
James Hogan [Thu, 29 Nov 2012 17:37:37 +0000 (17:37 +0000)]
Add state argument to change_syscall and fix SPARC
Add a state argument to change_syscall() so that SPARC can modify that
instead of read-modify-writing the whole register set. This function is
always called within an arg_setup/arg_finish_change sequence which on
certain architectures like SPARC will also be doing a read-modify-write.
This prevents the second write (from arg_finish_change) from undoing the
effects of the change_syscall call.
* util.c (change_syscall): Move below definition of arg_setup_state.
Add state argument.
[SPARC || SPARC64] Change to set state->u_regs[U_REG_G1] rather than
read-modify-writing it with PTRACE_GETREGS and PTRACE_SETREGS.
(setbpt, clearbpt): Pass state argument to change_syscall.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Steve McIntyre [Sat, 10 Nov 2012 11:24:48 +0000 (11:24 +0000)]
Add support for tracing 32-bit ARM EABI binaries on AArch64
* defs.h [AARCH64]: Copy in the definition of arm_pt_regs and the
accessor macros, so it's possible to build on AArch64 without
ARM system headers. Set SUPPORTED_PERSONALITIES to 2.
Define PERSONALITY0_WORDSIZE and PERSONALITY1_WORDSIZE.
Set DEFAULT_PERSONALITY to 1.
* linux/aarch64/errnoent1.h: New file, includes generic errnoent.h.
* linux/aarch64/ioctlent1.h: New file, includes generic ioctlent.h.
* linux/aarch64/signalent1.h: New file, includes generic signalent.h.
* linux/aarch64/syscallent1.h: Rename from linux/aarch64/syscallent.h.
* linux/aarch64/syscallent.h: New file, includes arm/syscallent.h.
* syscall.c [AARCH64]: Define aarch64_regs.
(update_personality) [AARCH64]: Add debug output.
(get_scno) [AARCH64]: Determine if we're in ARM or AArch64 mode by
checking the size of the returned uio structure from PTRACE_GETREGSET
and interpret the structure accordingly.
(get_syscall_result): Likewise.
(get_syscall_args): Merge the AArch64 and ARM sections so that on
AArch64 we can fall back to supporting the ARM personality.
(get_error): Likewise.
Signed-off-by: Steve McIntyre <steve.mcintyre@linaro.org>
Steve McIntyre [Wed, 24 Oct 2012 16:58:16 +0000 (17:58 +0100)]
Add AArch64 support to strace
AArch64 has been included in linux from 3.7 onwards.
Add support for AArch64 in strace, tested on linux in a simulator.
* configure.ac: Support AArch64.
* defs.h [AARCH64]: Include <sys/ptrace.h>, define TCB_WAITEXECVE.
* ipc.c (indirect_ipccall): Support AArch64.
* process.c (struct_user_offsets): Likewise.
* syscall.c [AARCH64]: Include <asm/ptrace.h>, <sys/uio.h>, and
<elf.h>. Define struct user_pt_regs regs.
(get_scno, get_syscall_result): Support AArch64 using PTRACE_GETREGSET.
(get_syscall_args, get_error): Support AArch64.
* linux/aarch64/ioctlent.h.in: New file.
* linux/aarch64/syscallent.h: New file, based on linux 3.7 version of
asm-generic/unistd.h.
Signed-off-by: Steve McIntyre <steve.mcintyre@linaro.org>
Mike Frysinger [Thu, 27 Sep 2012 05:12:39 +0000 (01:12 -0400)]
x32: add 64bit annotation too
Since someone can invoke these entry points directly with syscall(),
at least decode their name and show that they're 64bit versions rather
than just showing syscall_###.
* linux/x32/syscallent.h: Sync all missing entries below 312 with x86_64.
strace used to honor fflush(3) return value in trace_syscall_entering
which resulted to tracees not being PTRACE_SYSCALL'ed which in turn
caused nasty hangups like this one:
$ strace -o'|:' pwd
|:: Broken pipe
There is little strace can do in case of fflush(3) returning EOF, and
hangup is certainly not the best solution for the issue.
Mike Frysinger [Wed, 22 Aug 2012 15:56:15 +0000 (11:56 -0400)]
x32: update {g,s}etsockopt syscall numbers
Starting with linux 3.6 (and backported to earlier kernels), these two
syscalls have changed numbers (moving from native to compat entry points).
Update the strace syscall list accordingly.
* linux/x32/syscallent.h: Move setsockopt from 54 to 541, and move
getsockopt from 55 to 542.
* defs.h: Define USE_SEIZE to 1. Remove PTRACE_SEIZE_DEVEL
and PTRACE_EVENT_STOP1.
* strace.c (ptrace_attach_or_seize): Replace PTRACE_SEIZE_DEVEL
with 0.
(trace): Do not check for PTRACE_EVENT_STOP1.
Denys Vlasenko [Fri, 18 May 2012 00:03:24 +0000 (02:03 +0200)]
Merge adjacent printing operations in a few places
* file.c (sys_readahead): Merge tprints() with following printllval().
(sys_ftruncate64): Likewise.
(sys_fadvise64): Likewise.
(sys_fadvise64_64): Likewise.
(sys_fallocate): Merge tprints() with following tprintf().
Denys Vlasenko [Wed, 16 May 2012 10:23:58 +0000 (12:23 +0200)]
Fix a few goofs in sys_sysctl()
* system.c (sys_sysctl): Cast pointer to long, not size_t,
when we intend to use it as an address. Set oldlen to 0 prior
to reading into it - we want to have deterministic result
if read fails.
Denys Vlasenko [Tue, 15 May 2012 13:35:52 +0000 (15:35 +0200)]
Remove outdated comment about suspending new tracees
We no longer track parent/child relationship between tracees.
Therefore, we no longer suspend new tracee until parent is seen
exiting form [v]fork/clone. The comment is obsolete.
Denys Vlasenko [Tue, 15 May 2012 12:28:56 +0000 (14:28 +0200)]
Make sure current_wordsize and PERSONALITY0_WORDSIZE are ints in all arches
On 64bit systems with a single personality, they used to be sizeof(long),
which has type "long", not "int", which complicates printf formats.
* defs.h: Ensure that PERSONALITY0_WORDSIZE;s tyoe is int.
This in turn makes sure current_wordsize is also an int.
* count.c (call_summary): Revert the change which added cast to int.
Mike Frysinger [Fri, 4 May 2012 23:37:29 +0000 (19:37 -0400)]
util: fix building when glibc has a stub process_vm_readv
If you have a newer glibc which provides process_vm_readv, but it is built
against older kernel headers which lack __NR_process_vm_readv, the library
will contain a stub implementation that just returns ENOSYS. Autoconf
checks for this case explicitly and will declare it as unavailable. So we
end up in a case where the headers provide the prototype, but autoconf has
not defined HAVE_PROCESS_VM_READV, so we hit the same build failure again:
util.c:738:16: error: static declaration of 'process_vm_readv' follows non-static declaration
/usr/include/bits/uio.h:58:16: note: previous declaration of 'process_vm_readv' was here
So rename our local function to something unique, and add a define so the
callers all hit the right place.
* util.c (strace_process_vm_readv): Rename from process_vm_readv.
(process_vm_readv): Define to strace_process_vm_readv.
Dmitry V. Levin [Wed, 2 May 2012 10:21:49 +0000 (10:21 +0000)]
Fix build with <linux/loop.h> from 2.6.18 kernel headers
* configure.ac: Check for LO_FLAGS_AUTOCLEAR and LO_FLAGS_PARTSCAN
declarations.
* loop.c (loop_flags_options): Use LO_FLAGS_AUTOCLEAR and
LO_FLAGS_PARTSCAN only when appropriate declarations are available.
(loop_ioctl): Use LOOP_SET_CAPACITY only when it is defined.
Dmitry V. Levin [Tue, 1 May 2012 21:04:18 +0000 (21:04 +0000)]
NEWS: update for release
* NEWS (Improvements): Mention recent recvmsg/recvmmsg decoders
enhancements.
(Portability): Add a recommendation for the minimum Linux kernel
version to use.
Dmitry V. Levin [Tue, 1 May 2012 20:41:40 +0000 (20:41 +0000)]
Fix recvmmsg decode: do not show more data than actually returned
This change complements recent fix for recvmsg decoding.
* net.c (printmmsghdr): Add msg_len parameter to pass down to do_msghdr.
When this parameter is zero, pass mmsghdr.msg_len to do_msghdr instead.
(decode_mmsg): Add msg_len parameter, pass it down to printmmsghdr.
(sys_sendmmsg): Call decode_mmsg with msg_len == (unsigned long) -1L.
(sys_recvmmsg): Call decode_mmsg with msg_len == 0.
We often pass syscall params and other long-sized values
as printstr(len). Truncating them to int may be a bad thing.
* defs.h: Change len parameter's type from int to long in
string_quote and printstr function declarations.
* util.c (string_quote): Special-case only len==-1, not all len<0.
(printstr): Likewise.
Fix recvmsg decode: do not show more data than actually returned
I noticed that "hostname -d" talks over netlink and gets 20 bytes
of response, but we show entire 1024 bytes of iov.
This changes fixes that.
* defs.h: New function tprint_iov_upto.
* io.c (tprint_iov_upto): Definition of this function.
(tprint_iov): Call tprint_iov_upto.
* net.c (do_msghdr): Add data_size parameter, pass it down to tprint_iov_upto.
(printmsghdr): Add data_size parameter, pass it down to do_msghdr.
(printmmsghdr): Call do_msghdr with data_size==ULONG_MAX.
(sys_sendmsg): Call printmsghdr with data_size==ULONG_MAX.
(sys_recvmsg): Call printmsghdr with data_size==tcp->u_rval.
Mike Frysinger [Fri, 27 Apr 2012 22:58:20 +0000 (18:58 -0400)]
Cast current_wordsize to an int
On 64bit systems with a single personality, we see:
count.c: In function 'call_summary':
count.c:223:5: warning: format '%u' expects type 'unsigned int',
but argument 3 has type 'long unsigned int'
Since on multi-personality systems this is an array of ints, cast
the multiplication to an int and update the printf format.
* count.c (call_summary): Change %u to %d and cast first argument to int.
projects / strace / log