Dmitry V. Levin [Mon, 4 Jun 2018 01:11:05 +0000 (01:11 +0000)]
bpf: enhance BPF_OBJ_GET_INFO_BY_FD decoder further
* bpf.c (print_ebpf_prog): Remove "decode" argument, all users updated.
(print_bpf_prog_info): Use the minimum of xlated_prog_len values on
entering and exiting syscall as the size of xlated_prog_insns array
to be decoded. Likewise, use the minimum of nr_map_ids values on entering
and exiting syscall as the size of map_ids array to be decoded.
* bpf.c (print_bpf_map_info): Print only those fields that are
present in the structure, as it has been growing over time.
(print_bpf_prog_info): Likewise. Print map_ids field as an array
of uint32_t instead of uint64_t; print "name" field.
Co-Authored-by: Dmitry V. Levin <ldv@altlinux.org>
bpf: move handling of the saved state to BPF_OBJ_GET_INFO_BY_FD decoder
As it is needed for proper info_len printing on exiting.
* bpf.c (print_bpf_obj_info_fn): Change return type to void, add "saved"
argument.
(struct obj_get_info_saved): New type definition.
(print_bpf_map_info, print_bpf_prog_info): Change return type to void,
add "saved" argument, do not handle private tcb data allocation and
obtainment.
(print_bpf_obj_info_addr): Change return type to void, add tcb argument,
print the address only on exiting.
(print_bpf_obj_info): Change return type to void, add "saved" argument.
(BEGIN_BPF_CMD_DECODER(BPF_OBJ_GET_INFO_BY_FD)): Handle private data
allocation/saving/restoring, use it for storing/printing info_len field.
Co-Authored-by: Dmitry V. Levin <ldv@altlinux.org>
bpf: print xlated_prog_insns only when xlated_prog_len had not been 0
If xlated_prog_len is zero on entering syscall and non-zero on exiting,
the program itself is not written into xlated_prog_insns.
* bpf.c (print_ebpf_prog): Add decode argument, print the address only
if decode is false.
(BEGIN_BPF_CMD_DECODER(BPF_PROG_LOAD)): Specify decode == true argument
to print_ebpf_prog.
(print_bpf_prog_info): Specify saved xlated_prog_len as decode argument
to print_ebpf_prog.
Co-Authored-by: Dmitry V. Levin <ldv@altlinux.org>
* bpf.c (print_ebpf_prog): Add static qualifier. Change the type
of "addr" argument from kernel_ulong_t to uint64_t, print it using
print_big_u64_addr.
(BEGIN_BPF_CMD_DECODER(BPF_PROG_LOAD), print_bpf_prog_info): Do not call
print_big_u64_addr before print_ebpf_prog.
Co-Authored-by: Dmitry V. Levin <ldv@altlinux.org>
tests: cast printf arguments to short when %h format qualifier is used
Reported by clang.
* tests/mknod.c (main): Cast printf argument for %#03ho format to short.
* tests/mknodat.c (main): Likewise.
* tests/net-sockaddr.c (main): Cast printf argument for %hu format
to short.
* tests/semop.c (main): Likewise.
configure.ac: check for linux/rtnetlink.h instead of linux/if_link.h for declarations
Historically, type definitions and declarations related to netlink route
interface attributes were in linux/rtnetlink.h, later they were moved
elsewhere (first linux/if.h, then linux/if_link.h) with the inclusion
of the corresponding header file in linux/rtnetlink.h.
* configure.ac: Use <linux/rtnetlink.h> instead of <linux/if_link.h>
in checks for interface netlink attributes-related declarations
and definitions.
tests: check decoding of new FRA_* netlink attributes
* configure.ac (AC_CHECK_TYPES): Check for struct fib_rule_port_range
in <linux/fib_rules.h>.
* tests/nlattr_fib_rule_hdr.c: Include <linux/in.h>.
(FRA_PROTOCOL, FRA_IP_PROTO, FRA_SPORT_RANGE, FRA_DPORT_RANGE): New
macro constants.
[!HAVE_STRUCT_FIB_RULE_PORT_RANGE] (struct fib_rule_port_range): New
type.
(main): Check decoding of new FRA_* netlink attributes.
Co-Authored-by: Dmitry V. Levin <ldv@altlinux.org>
tests: use syscall() for ioctl() calls with oversized arguments
* tests/ioctl_inotify.c (sys_ioctl): New function.
(main): Use sys_ioctl for ioctl calls with oversized arguments.
* tests/ioctl_loop.c: Likewise.
* tests/ioctl_perf.c: Likewise.
Print interface indices according to the preferred xlat style settings.
* print_ifindex.c (INI_PFX, INI_SFX): New helper macros.
[HAVE_IF_INDEXTONAME] (get_ifname): New function.
[HAVE_IF_INDEXTONAME] (print_ifindex): Implement as a wrapper around
get_ifname and print_xlat_ex.
nlattr: print index names in netlink meminfo array
* xlat/netlink_sk_meminfo_indices.in: New file.
* nlattr.c: Include xlat/netlink_sk_meminfo_indices.h.
(decode_nla_meminfo): Use print_array_ex, pass
netlink_sk_meminfo_indices as index xlat table.
* tests/nlattr_inet_diag_msg.c: Update expected output.
print_array: add support for printing array indices
* defs.h (XLAT_STYLE_SPEC_BITS, XLAT_STYLE_MASK): New macro constants.
(tfetch_mem_fn, print_fn): New typedefs.
(enum print_array_flag_bits, enum print_array_flags): New enumerations.
(print_array_ex): Rename from print_array, add flags, index_xlat,
index_xlat_size, and index_dflt arguments.
(print_array): New static inline function, a thin wrapper around
print_array_ex.
util.c: Include "xlat.h".
(print_array): Rename to print_array_ex, add flags, index_xlat,
index_xlat_size, and index_dflt arguments. Print array indices
according to the style settings specified by flags if PAF_PRINT_INDICES
is set.
Dmitry V. Levin [Tue, 29 May 2018 01:15:19 +0000 (01:15 +0000)]
print_array: enhance printing of unfetchable object addresses
When umoven_func invocation fails to fetch data, it prints the faulty
address. If this happens to a subsequent umoven_func invocation,
the printed address may be undistinguishable from a valid data printed
by print_func, e.g. when the data is printed in a numeric form like
[0x1, 0x2, 0x3, 0xdefaced].
Fix this source of confusion by moving the printing of the faulty
address from umoven_func to print_array itself. This change renames
umoven_func to tfetch_mem_func and changes its semantics, so that
- tfetch_mem_func never prints anything;
- tfetch_mem_func returns true if the fetch succeeded,
and false otherwise.
Dmitry V. Levin [Mon, 28 May 2018 11:32:03 +0000 (11:32 +0000)]
Introduce tfetch_* family of functions and macros
These new functions are going to be used instead of umove_or_printaddr*
as umoven_func argument of print_array.
* defs.h (tfetch_mem64, tfetch_mem64_ignore_syserror): New
function prototypes.
(tfetch_mem, tfetch_mem_ignore_syserror): New static inline
wrappers.
(tfetch_obj): New macro wrapper around tfetch_mem.
* util.c (tfetch_mem64, tfetch_mem64_ignore_syserror): New
functions.
(umoven_or_printaddr64, umoven_or_printaddr64_ignore_syserror): Use
them.
bpf: decode bpf_attr.info field used by BPF_OBJ_GET_INFO_BY_FD command
* bpf_attr.h [!BPF_TAG_SIZE] (BPF_TAG_SIZE): Define.
[BPF_TAG_SIZE]: Check that BPF_TAG_SIZE is 8.
(struct bpf_map_info_struct, struct bpf_prog_info_struct): New
structures.
(bpf_map_info_struct_size, expected_bpf_map_info_struct_size,
bpf_prog_info_struct_size, expected_bpf_prog_info_struct_size): New
macro constants.
* bpf.c (print_bpf_obj_info_fn): New type.
(print_bpf_map_info, print_bpf_prog_info, fetch_bpf_obj_info,
print_bpf_obj_info_addr, print_bpf_obj_info): New functions.
(BEGIN_BPF_CMD_DECODER(BPF_OBJ_GET_INFO_BY_FD)): Print bpf_fd and info_len
fields only on entering, call print_bpf_obj_info for printing info field.
* tests/bpf.c (BPF_OBJ_GET_INFO_BY_FD_checks): Print info field
as a pointer.
Co-Authored-by: Dmitry V. Levin <ldv@altlinux.org>
bpf: add support for checking structures outside union bpf_attr
struct bpf_prog_info and bpf_map_info need essentially the same handling
as union bpf_attr.
* gen_bpf_attr_check.sh: Derive type_name from $struct if it doesn't
start with "BPF_", derive TYPE_NAME from type_name, use them in code
generation.
* m4/gen_bpf_attr_m4.sh: Rewrite parsing/generation code into awk,
add support for structures outside union bpf_attr.
Co-Authored-by: Dmitry V. Levin <ldv@altlinux.org>
Dmitry V. Levin [Sat, 26 May 2018 10:32:54 +0000 (10:32 +0000)]
bpf: remove bpf_priv_data
As various bpf commands need different data to save between entering
and exiting of the syscall, do not emulate union bpf_attr approach.
* bpf.c (struct bpf_priv_data): Remove.
(DECL_BPF_CMD_DECODER): Remove "struct bpf_priv_data *" argument.
(BEGIN_BPF_CMD_DECODER(BPF_PROG_QUERY)): Use set_tcb_priv_ulong
and get_tcb_priv_ulong to keep bpf_attr.query.prog_cnt field.
(SYS_FUNC(bpf)): Remove priv variable, do not initialize it,
and do not pass it to bpf_cmd_decoders.
tests: add union bpf_attr.query.prog_ids field decoding checks
* tests/bpf-success-v.c: New file.
* tests/bpf-success.c: Likewise.
* tests/bpf-success-v.test: New test.
* tests/bpf-success.test: Likewise.
* tests/.gitignore: Add bpf-success and bpf-success-v.
* tests/Makefile.am (check_PROGRAMS): Likewise.
(DECODER_TESTS): Add bpf-success.test and bpf-success-v.test.
* tests/bpf.c [INJECT_RETVAL] (sys_bpf): Check that retuned value
is equivalent to the injected one; append "(INJECTED)" string to errstr.
(prog_load_ids, prog_load_ids_ptr): New variables.
(init_BPF_PROG_QUERY_attr4, print_BPF_PROG_QUERY_attr4,
init_BPF_PROG_QUERY_attr5, print_BPF_PROG_QUERY_attr5): New functions.
(BPF_PROG_QUERY_checks): Drop "const" qualifier, add new checks.
* bpf.c (BEGIN_BPF_CMD_DECODER(BPF_PROG_QUERY)): Use
print_uint32_array_member instead of print_uint64_array_member for
printing prog_ids elements (and change the type of prog_id_buf from
uint64_t to uint32_t), as they are 32-bit; use common trick with
print_big_u64_addr for handling possible ambiguity with prog_ids
address.
* tests/bpf.c (BIG_ADDR_MAYBE): New macro.
(BPF_PROG_QUERY_checks): Update expected output.
* xlat/ebpf_regs.in: New file.
* bpf.c (struct ebpf_insn, struct ebpf_insns_data): New type definitions.
(print_ebpf_insn, print_ebpf_prog): New functions.
(DEF_BPF_CMD_DECODER(BPF_PROG_LOAD)): Use them.
* xlat/inet_protocols.in (IPPROTO_GGP, IPPROTO_HELLO, IPPROTO_ND):
Remove, not defined in Linux.
(IPPROTO_BEETPH): New constant, introduced by Linux commit
v2.6.19-rc1~33^2.
(IPPROTO_L2TP): New constant, introduced by Linux commit
v2.6.35-rc1~473^2~594.
(IPPROTO_MH): New constant, introduced by Linux commit
v2.6.19-rc1~1272^2~157.
Add default values to constants, sort the file.
There's little reason not to decode error code for syscalls printed in
raw; moreover, it creates inconsistencies for unknown syscall printing.
Another issue was with error numbers unknown to strace: previously, they
were printed as "... = -1 1234 (Unknown error 1234)" which looked kinda
weird.
* syscall.c (print_err_ret): New function.
(syscall_exiting_trace): Remove u_error variable, use tcp->u_error
instead. Remove u_error_str variable.
(syscall_exiting_trace) <if (raw(tcp))>: Call print_err_ret if u_error
is non-zero.
(syscall_exiting_trace) <if (!(sys_res & RVAL_NONE) && tcp->u_error)>:
Use print_err_ret for default error printing.
* tests/nsyscalls.c [!LINUX_MIPSO32] (test_syscall): Update expected
output.
* tests/qual_fault.c (invoke) <if (is_raw)>: Likewise.
* strace.1.in (.SH DESCIPTION): Update unknown syscall example.
Dmitry V. Levin [Thu, 24 May 2018 12:23:39 +0000 (12:23 +0000)]
tests: further cleanup bpf big address checks
Replace BIG_ADDR_IS_64BIT and BIG_ADDR_IS_32BIT with BIG_ADDR().
* tests/bpf.c (BIG_ADDR): New macro.
(BIG_ADDR_IS_64BIT, BIG_ADDR_IS_32BIT): Remove.
(BPF_PROG_LOAD_checks, BPF_OBJ_PIN_checks,
BPF_RAW_TRACEPOINT_OPEN_checks): Use BIG_ADDR() instead of
BIG_ADDR_IS_64BIT and BIG_ADDR_IS_32BIT.
Introduce print_uint32_array_member helper function
Analogous to print_int32_array_member, this helper is going to be used
in various decoders, including BPF_PROG_QUERY, PERF_EVENT_IOC_*,
and netlink meminfo.
* defs.h (print_uint32_array_member): New declaration.
* util.c (print_uint32_array_member): New function.
sockaddr: add support for sockaddr_l2 without l2_bdaddr_type
Apparently, struct sockaddr_l2 changed over time: the l2_bdaddr_type
field was added only by commit v3.5-rc1~35^2~3^2~18^2^2~39.
* configure.ac (AC_CHECK_HEADERS): Move bluetooth/bluetooth.h check into
a separate check.
(AC_CHECK_HEADERS([bluetooth/bluetooth.h])): New check, also checks for
struct sockaddr_l2.l2_bdaddr_type member in case of success.
* sockaddr.c (print_sockaddr_data_bt) <case offsetof(struct sockaddr_l2,
l2_bdaddr_type)>: Add to sizeof(struct sockaddr_l2) case. Print
l2_bdaddr_type field only if the provided sockaddr is big enough.
* tests/net-sockaddr.c (check_l2): Put l2_bdaddr_type usage/printing
under HAVE_STRUCT_SOCKADDR_L2_L2_BDADDR_TYPE ifdef, add a check for the
sockaddr_l2 structure without the l2_bdaddr_type field.
sock: guess ifr_hwaddr size in SIOCSIFHWADDR/SIOCGIFHWADDR
And make the output more structured.
* sock.c: Include xlat/arp_hardware_types.h in XLAT_MACROS_ONLY mode.
(print_ifreq) <case SIOCSIFHWADDR, case SIOCGIFHWADDR>: Introduce
hwaddr_sizes array, print ifr_hwaddr.sa_family, print ifr_hwaddr.sa_data
using hwaddr_sizes and PRINT_FIELD_MAC_SZ.
strace.1.in: display -k option description only if it is enabled
Leverage the fact that manpage is generated and add condition statements
around mentions of -k option.
* strace.1.in (.SH SYNOPSIS): Put option list inside an
".if '@ENABLE_STACKTRACE_TRUE@'#' .ig end_unwind_opt ... .end_unwind_opt"
condition; add a second list of options without -k and put it inside
".if '@ENABLE_STACKTRACE_FALSE@'#' .ig end_no_unwind_opt ... .end_no_unwind_opt"
condition.
(.SS Output format) <-k>: Put it inside
".if '@ENABLE_STACKTRACE_TRUE@'#' .ig end_unwind ... .end_unwind"
condition, remove the mention of --enable-stacktrace configuration
option requirement.
xlat: add support for xlat_styles in printxval_dispatch
* defs.h (printxval_dispatch_ex): Rename from printxval_dispatch, add
style argument.
(printxval_dispatch): New static inline function, a thin wrapper around
printxval_dispatch_ex.
* xlat.c (printxval_dispatch): Rename to printxval_dispatch_ex, add
style argument, handle it.
Co-Authored-by: Dmitry V. Levin <ldv@altlinux.org>
util.c: make memory allocation error messages more informative
As those are pretty unusual, let's add some additional information
that can be reported by users.
* util.c (print_quoted_string_ex, dumpstr): Add reasons for memory
allocation errors.
(sizeof_iov): Change type from size_t to unsigned int.
(dumpiov_upto): Rewrite size initialisation similarly to the way it's
done in print_quoted_string_ex, add reasons for memory allocation errors.
Co-Authored-by: Dmitry V. Levin <ldv@altlinux.org>
The current way of designation of 64-bit specific syscalls in x32
personality not only conflicts with fault injection specification
syntax, but also makes it impossible to specify such syscalls at all
(since everything that starts from a digit is considered a syscall
number specification). So, let's change prefix to suffix and use
octothorpe instead of colon as a separator.
* xlat/bpf_attach_type.in (BPF_SK_MSG_VERDICT): New constant,
introduced by Linux commit v4.17-rc1~148^2~156^2~1^2~13.
(BPF_CGROUP_INET4_BIND, BPF_CGROUP_INET6_BIND): New constants,
introduced by Linux commit v4.17-rc1~148^2~19^2^2~6.
(BPF_CGROUP_INET4_CONNECT, BPF_CGROUP_INET6_CONNECT): New constants,
introduced by Linux commit v4.17-rc1~148^2~19^2^2~3.
(BPF_CGROUP_INET4_POST_BIND, BPF_CGROUP_INET6_POST_BIND): New constants,
introduced by Linux commit v4.17-rc1~148^2~19^2^2~1.
* tests/bpf.c (BPF_PROG_QUERY_checks): Update expected output.
Andre McCurdy [Tue, 15 May 2018 22:34:39 +0000 (15:34 -0700)]
linux/arm/raw_syscall.h: avoid r7 specified register variables with Thumb
If Thumb code is being generated and frame pointers are enabled, the
Thumb frame pointer in r7 clashes with any local variable which may
need to be assigned to r7 (e.g. the syscall NR when making a raw
syscall).
With gcc, the double use of r7 results in a build error, e.g.
strace-4.22/tests/inject-nf.c:86:1: error: r7 cannot be used in asm here
With clang, the double use of r7 can result in the compiler silently
generating broken code which crashes at run time due to frame pointer
corruption:
https://bugs.llvm.org/show_bug.cgi?id=34165
In most cases the problem isn't visible as frame pointers will be
disabled automatically due to optimisation level. However to handle
cases where frame pointers are enabled (e.g. when CFLAGS etc are set
to support a debug build, etc) provide a version of raw_syscall_0
which manually saves and restores the frame pointer value in r7
to a temporary register before setting up the syscall NR in r7
and invoking the syscall.
* linux/arm/raw_syscall.h (raw_syscall_0) [__thumb__]: Provide
an alternative version.
Signed-off-by: Andre McCurdy <armccurdy@gmail.com>