* xlat/ipccalls.in: New file.
* ipc.c: New file.
* Makefile.am (libstrace_a_SOURCES): Add it.
* linux/dummy.h (sys_ipc): Remove stub alias.
* syscall.c (decode_ipc_subcall): Treat 1st argument of ipc syscall
as "unsigned int".
[S390 || S390X]: Skip ipc cubcalls that have non-zero version.
[SPARC64]: Likewise, for the native personality.
Save ipc cubcall version for later use by specific ipc parsers.
* ipc_msg.c (SYS_FUNC(msgrcv)): Handle non-zero ipc subcall version.
[SPARC64]: Handle non-ipc_kludge case for the native personality.
* linux/subcall.h (msgrcv): Change nargs from 4 to 5.
* linux/s390/syscallent.h (ipc): Change nargs from 6 to 5.
* linux/s390x/syscallent.h (ipc): Likewise.
* tests/netlink_protocol.c: Check that HAVE_SYS_XATTR_H is defined.
Include <stdlib.h> and <sys/xattr.h>.
(main): Check that system.sockprotoname attribute is supported.
x32, mips n32: make printargs print full 64-bit syscall arguments
As most of x32 syscalls are regular x86_64 syscalls, they should be
printed by default as 64-bit syscalls, and syscall parsers at their
discretion might print syscalls differently. This is already
implemented in 64-bit strace tracing x32 personality processes,
and now x32 strace will also behave this way.
Similar reasoning applies to mips n32.
* util.c (printargs) [HAVE_STRUCT_TCB_EXT_ARG]: Print syscall arguments
from ext_arg array using %llx format.
* xlat/socketcalls.in: New file.
* socketcall.c: New file.
* Makefile.am (libstrace_a_SOURCES): Add it.
* linux/dummy.h (sys_socketcall): Remove stub alias.
* syscall.c (decode_socket_subcall): Treat 1st argument of socketcall
as "int". Do not substitute syscall until all socketcall arguments
have been fetched successfully.
* mmsghdr.c (decode_mmsgvec): Clear syserror for the time of printing
struct mmsghdr array in !use_msg_len case.
(SYS_FUNC(sendmmsg)): As failed sendmmsg syscall does not
set msg_len fields, tell this to decode_mmsgvec.
* tests/mmsg_name.c (test_mmsg_name): Check it.
* defs.h (decode_mmsgvec): Add int argument.
* mmsghdr.c (print_struct_mmsghdr_config): Add msg_len_vlen field.
(print_struct_mmsghdr): Print struct mmsghdr.msg_len only for
msg_len_vlen first elements of struct mmsghdr array.
(decode_mmsgvec): Add msg_len_vlen argument.
(SYS_FUNC(sendmmsg), SYS_FUNC(recvmmsg)): Pass to decode_mmsgvec
the number of elements of struct mmsghdr array that contain meaningful
struct mmsghdr.msg_len.
* tests/mmsg_name.c (test_mmsg_name): Check it.
Fix decoding of msg_name* fields of recvmmsg syscall's msghdr array argument
As msg_namelen fields of recvmmsg syscall's struct msghdr array have
the same read-write semantics as struct msghdr argument of recvmsg
syscall, parser of recvmmsg syscall needs a similar fix.
* mmsghdr.c (print_struct_mmsghdr): Increment p_user_msg_namelen field.
(mmsgvec_data): New structure.
(free_mmsgvec_data, save_mmsgvec_namelen): New functions.
(decode_mmsgvec): Take into account the data saved
by save_mmsgvec_namelen function.
(SYS_FUNC(recvmmsg)): Use save_mmsgvec_namelen.
mmsghdr.c: rewrite decode_mmsgvec using print_array
* fetch_struct_mmsghdr.c (sizeof_struct_mmsghdr): New mpers printer.
* mmsghdr.c: Include <limits.h>.
(print_struct_mmsghdr_config): New structure.
(fetch_struct_mmsghdr_or_printaddr, print_struct_mmsghdr): New functions.
(decode_mmsgvec): Rewrite using them and print_array.
tests: add recv_mmsg and send_mmsg functions to libtests
* tests/tests.h (recv_mmsg, send_mmsg): New prototype.
* tests/libmmsg.c: New file.
* tests/Makefile.am (libtests_a_SOURCES): Add it.
* tests/mmsg.c Do not check for __NR_sendmmsg, __NR_recvmmsg,
HAVE_SENDMMSG, and HAVE_RECVMMSG. Do not include unused headers.
(recv_mmsg, send_mmsg): Remove.
As we've got a proper mechanism that parsers can use for storing private
data between entering and exiting stages, all cases of struct tcb.auxstr
abuse should go.
* btrfs.c (btrfs_ioctl): Use set_tcb_priv_ulong and get_tcb_priv_ulong
instead of abusing tcp->auxstr.
* file_handle.c (SYS_FUNC(name_to_handle_at)): Likewise.
* msghdr.c (SYS_FUNC(recvmsg)): Likewise.
* net.c (decode_sockname, SYS_FUNC(recvfrom)): Likewise.
* v4l2.c (print_v4l2_queryctrl): Likewise.
Fix race condition in decoding timeout argument of recvmmsg syscall
As the value returned by sprint_timespec points to a static area and may
be overwritten by subsequent calls to sprint_timespec, it is not safe to
save this value on entering syscall and use it later on exiting.
* mmsghdr.c (SYS_FUNC(recvmmsg)): On entering syscall, copy the value
returned by sprint_timespec to a dynamically allocated memory, and save
the pointer using set_tcb_priv_data. On exiting, restore it using
get_tcb_priv_data.
Fix race condition in decoding rt_sigtimedwait's timeout argument
As the value returned by sprint_timespec points to a static area and may
be overwritten by subsequent calls to sprint_timespec, it is not safe to
save this value on entering syscall and use it later on exiting.
* signal.c (SYS_FUNC(rt_sigtimedwait)): On entering syscall, copy the
value returned by sprint_timespec to a dynamically allocated memory,
and save the pointer using set_tcb_priv_data. On exiting, restore it
using get_tcb_priv_data.
Patrik Jakobsson [Mon, 24 Aug 2015 12:42:47 +0000 (14:42 +0200)]
Add private data field to struct tcb
We need to be able to store private data in the struct tcb across it's
lifetime. To ensure proper deallocation of this stored data a callback
must be provided along with the data. The callback is executed
automatically on exiting syscall, and when the life of the tcb ends.
* defs.h (struct tcb): Add _priv_data and _free_priv_data fields.
(get_tcb_priv_data, set_tcb_priv_data, free_tcb_priv_data):
New prototypes.
(get_tcb_priv_ulong, set_tcb_priv_ulong): New static inline functions.
* strace.c (get_tcb_priv_data, set_tcb_priv_data, free_tcb_priv_data):
New functions.
(droptcb): Call free_tcb_priv_data.
* syscall.c (trace_syscall_exiting): Call free_tcb_priv_data
when clearing TCB_INSYSCALL flag.
Signed-off-by: Patrik Jakobsson <patrik.jakobsson@linux.intel.com> Signed-off-by: Dmitry V. Levin <ldv@altlinux.org>
Fix decoding of struct msghdr.msg_name* arguments of recvmsg syscall
As the msghdr.msg_namelen argument of recvmsg syscall has the same
read-write semantics as the address length argument of recvfrom syscall,
parser of recvmsg syscall needs a similar fix.
* defs.h (fetch_msghdr_namelen): New prototype.
(decode_msghdr): Add "int *" argument.
* msghdr.c (print_msghdr): Likewise. Treat it as a pointer to struct
msghdr.msg_namelen passed to the kernel. Pass to decode_sockaddr
the size of socket address actually returned by the kernel.
Print both user and kernel address lengths when the value changes.
(decode_msghdr, decode_mmsghdr): Add "int *" argument,
forward it to print_msghdr.
(decode_mmsgvec): Update decode_mmsghdr invocation.
(fetch_msghdr_namelen): New function.
* net.c (SYS_FUNC(sendmsg)): Update decode_msghdr invocation.
(SYS_FUNC(recvmsg)): Use fetch_msghdr_namelen on entering to save
struct msghdr.msg_namelen. On exiting, pass the saved value
to decode_msghdr.
* linux/64/ioctls_inc.h: Update from linux v4.7 using ioctls_gen.sh.
* linux/32/ioctls_inc_align32.h: Likewise.
* linux/32/ioctls_inc_align64.h: Likewise.
Fix decoding of sockaddr related arguments of recvfrom syscall
* net.c (SYS_FUNC(recvfrom)): Use fetch_socklen. Save the address
length passed to the kernel and print it on error. Do not print more
bytes of the socket address than actually returned by the kernel.
Print both user and kernel address lengths when the returned address
is truncated.
Print both values of changing address length in sockname family syscalls
* net.c (decode_sockname): Print both user and kernel address lengths
when the value changes.
* tests/net-y-unix.c (main): Update expected output.
* tests/net-yy-unix.c (main): Likewise.
* tests/net.expected: Likewise.
Enhance diagnostics of address truncation in sockname family syscalls
* net.c (decode_sockname): Print both user and kernel address lengths
when the returned address is truncated.
* tests/sun_path.expected: Update expected output.
Fix decoding of getsockname, getpeername, accept, and accept4 syscalls
* net.c (fetch_socklen): New function.
(decode_sockname): Use it. On entering syscall, decode early
and return RVAL_DECODED when possible. On error, print the address
length passed to the kernel. Do not print more bytes of the socket
address than actually returned by the kernel.
Prepare parsers of sockname family syscalls for the upcoming fix
* net.c (do_sockname): Rename to decode_sockname, drop second argument,
do not print sock_type_flags, return RVAL_DECODED on exiting.
(SYS_FUNC(accept), SYS_FUNC(getsockname)): Use decode_sockname instead
of do_sockname, take its return code into account.
(SYS_FUNC(accept4)): Likewise. Print sock_type_flags when
decode_sockname returns RVAL_DECODED.
* tests/netlink_protocol.c (send_query): Check that no part
of netlink message is fetched beyond the end of message.
Check decoding of zero address and zero length messages.
Check decoding of messages with nlmsg_len exceeding message length.
Introduce a general socket netlink parser which prints the header and
a string for the remaining part of the buffer. It doesn't handle all
the netlink flags and types yet because the parser needs more
information, this will be implemented later.
* net.c (decode_sockbuf): New function.
(SYS_FUNC(send), SYS_FUNC(sendto), SYS_FUNC(recv), SYS_FUNC(recvfrom)):
Use it instead of printstr.
* msghdr.c (print_msghdr): Pass IOV_DECODE_NETLINK to tprint_iov_upto()
for netlink sockets.
* netlink.c: New file.
* Makefile.am (strace_SOURCES): Add it.
* defs.h (decode_netlink, getfdproto): New prototypes.
(iov_decode): Add IOV_DECODER_NETLINK.
* io.c (print_iovec): Use decode_netlink().
* util.c (getfdproto): Remove static keyword.
* xlat/netlink_flags.in: New file.
* xlat/netlink_types.in: New file.
* xlat/fsmagic.in: Add NSFS_MAGIC, TRACEFS_MAGIC, BPF_FS_MAGIC,
and UDF_SUPER_MAGIC introduced by linux kernel commits
v3.19-rc1~53^2~7^2~1, v4.1-rc1~147^2~4, v4.4-rc1~141^2~17^2~1,
and v4.7-rc1~161^2, respectively.
msghdr.c: limit output when printing excessively large messages
Limit output of struct msghdr.msg_control when
struct msghdr.msg_controllen is greater than the maximum ancillary
buffer size specified in /proc/sys/net/core/optmem_max file.
* configure.ac (AC_CHECK_FUNCS): Add open64.
* defs.h (read_int_from_file): New prototype.
* util.c (read_int_from_file): New function.
* msghdr.c (get_optmem_max): New function based on read_int_from_file.
(decode_msg_control): Use it to check control_len argument.
* tests/msg_control.c (test_big_len): New function.
(main): Use it to test printing of excessively large messages.
msghdr.c: limit IP_RECVOPTS/IP_RETOPTS output in abbrev mode
* msghdr.c (print_cmsg_ip_opts): Do not print more than max_strlen
bytes of ip options in abbrev mode.
* tests/msg_control.c (print_ip_opts): Use DEFAULT_STRLEN.
(test_sol_ip): Test IP_RETOPTS output in abbrev mode.
* msghdr.c (print_scm_rights): Do not print more than max_strlen
descriptors in abbrev mode.
* tests/msg_control.c (DEFAULT_STRLEN): New macro.
(print_fds): Use it.
(test_scm_rights3): New function.
(main): Use it to test SCM_RIGHTS output in abbrev mode.
* tests/msg_control-v.c: New file.
* tests/msg_control-v.test: New test.
* tests/.gitignore: Add msg_control-v.
* tests/Makefile.am (check_PROGRAMS): Likewise.
(DECODER_TESTS): Add msg_control-v.test.
Dmitry V. Levin [Thu, 30 Jun 2016 22:39:02 +0000 (22:39 +0000)]
msghdr.c: merge print_cmsg_ip_ttl and print_cmsg_ip_checksum
* msghdr.c (print_cmsg_ip_ttl): Rename to print_cmsg_uint.
(print_cmsg_ip_checksum): Remove.
(print_cmsg_type_data): Replace print_cmsg_ip_ttl
and print_cmsg_ip_checksum with print_cmsg_uint.
Dmitry V. Levin [Mon, 27 Jun 2016 00:02:55 +0000 (00:02 +0000)]
net.c: move all msghdr and mmsghdr parsers to a separate file
* defs.h (msg_flags, socketlayers, decode_msghdr, decode_mmsg):
New prototypes.
* msghdr.c: New file.
* Makefile.am (strace_SOURCES): Add it.
* net.c: Move all msghdr and mmsghdr parsers to msghdr.c.
Dmitry V. Levin [Mon, 27 Jun 2016 00:02:50 +0000 (00:02 +0000)]
Mpersify fetchers of struct msghdr and struct mmsghdr
* fetch_struct_msghdr.c: New file.
* fetch_struct_mmsghdr.c: Likewise.
* Makefile.am (strace_SOURCES): Add them.
* net.c (struct msghdr32, struct mmsghdr32, copy_from_msghdr32,
fetch_msghdr, fetch_mmsghdr): Remove.
(decode_msghdr, dumpiov_in_msghdr): Use fetch_struct_msghdr instead
of fetch_msghdr.
(decode_mmsghdr): Change msg_len argument to use_msg_len.
Use fetch_struct_mmsghdr instead of fetch_mmsghdr.
Return fetch_struct_mmsghdr's return code.
(decode_mmsg): Rename to decode_mmsgvec. Take addr and len arguments.
Do not print vlen and flags. Check decode_mmsghdr's return code.
Print mmsghdr array using square brackets.
(dumpiov_in_mmsghdr): Use fetch_struct_mmsghdr instead of fetch_mmsghdr.
(SYS_FUNC(sendmmsg), SYS_FUNC(recvmmsg)): Use decode_mmsgvec instead
of decode_mmsg. Print vlen and flags.
* tests/mmsg.c (main): Update expected output.
Dmitry V. Levin [Mon, 27 Jun 2016 00:02:41 +0000 (00:02 +0000)]
net.c: move fallback definition of struct mmsghdr to a separate file.
* msghdr.h: New file.
* Makefile.am (strace_SOURCES): Add it.
* net.c: Include it. Move fallback definition of struct mmsghdr there.
* tests/mmsg.c: Include "msghdr.h". Remove fallback definition
of struct mmsghdr.
Dmitry V. Levin [Sun, 26 Jun 2016 23:57:29 +0000 (23:57 +0000)]
net.c: rename some internal functions for consistency and readability
* net.c (printcmsghdr): Rename to decode_msg_control.
(do_msghdr): Rename to print_msghdr.
(extractmsghdr): Rename to fetch_msghdr.
(extractmmsghdr): Rename to fetch_mmsghdr.
(printmsghdr): Rename to decode_msghdr.
(printmmsghdr): Rename to decode_mmsghdr.
All callers updated.
Dmitry V. Levin [Sun, 26 Jun 2016 23:57:18 +0000 (23:57 +0000)]
net.c: when printing struct msghdr.msg_control, also print its name
* net.c (printcmsghdr): Skip control messages of zero length.
Always print "msg_control=" before the control message.
(do_msghdr): Print control message before its length, not after.
* tests/inet-cmsg.c (main): Update expected output.
* tests/scm_rights-fd.test: Likewise.
Dmitry V. Levin [Thu, 23 Jun 2016 15:15:22 +0000 (15:15 +0000)]
net.c: skip details of too short socket addresses, part 2
* net.c (sa_printers): Add minimal address length for each socket family.
(print_sockaddr): Call socket family specific sockaddr printer
only when socket address length is greater or equal
the minimal address length for the socket family.
Dmitry V. Levin [Thu, 23 Jun 2016 21:06:54 +0000 (21:06 +0000)]
net.c: fix printing AF_BLUETOOTH socket addresses
* xlat/hci_channels.in: New file.
* net.c (print_sockaddr_data_raw): New function.
(print_sockaddr): Use it.
[HAVE_BLUETOOTH_BLUETOOTH_H]: Include "xlat/hci_channels.h".
[HAVE_BLUETOOTH_BLUETOOTH_H] (print_sockaddr_data_bt): Fix printing
sockaddr_hci, sockaddr_sco, sockaddr_rc, and sockaddr_l2 structures.
Dmitry V. Levin [Fri, 24 Jun 2016 11:15:51 +0000 (11:15 +0000)]
net.c: fix printing AF_IPX sockets addresses, part 2
* net.c (print_sockaddr_data_ipx): Print sockaddr_ipx.sipx_network
using %#08x format. Print elements of sockaddr_ipx.sipx_node array
and sockaddr_ipx.sipx_type using %#02x format.
AF_UNIX abstract socket address specified by sockaddr_un.sun_path
is not a NUL-terminated string.
* net.c (SIZEOF_SA_FAMILY): New macro.
(print_sockaddr_data_un): Use it. Specify actual address length
in print_quoted_string calls. Do not set QUOTE_0_TERMINATED flag
when printing an abstract socket address.
Dmitry V. Levin [Thu, 23 Jun 2016 09:32:24 +0000 (09:32 +0000)]
net.c: do not print sockaddr_in6.sin6_scope_id unnecessarily
Prior to RFC2553, struct sockaddr_in6 had no sin6_scope_id field.
As the kernel still accepts RFC2133 editions of struct sockaddr_in6,
print sockaddr_in6.sin6_scope_id only when it is specified.
* net.c (SIN6_MIN_LEN): New macro.
(print_sockaddr_data_in6): Print sockaddr_in6.sin6_scope_id only when
socket address length exceeds SIN6_MIN_LEN.