xlat: provide fallback definitions for CLOCK_* constants
Since new values have been added gradually over various kernel versions,
it's better to define them explicitly in order to avoid situations when
strace built with older kernel headers cannot decode some recently
defined values.
* tests/quotactl.h (fill_memory_ex): Add period parameter,
use it as a divisor in non-constant part of value.
(fill_memory): Specify period of 0x80 to fill_memory_ex call.
* tests/quotactl-xfs.c (main): Likewise.
keyctl: use kernel_ulong_t and getarg_ull instead of long
This potentially fixes keyctl decoder for x32 personality.
* keyctl.c (keyctl_update_key, keyctl_read_key, keyctl_instantiate_key,
keyctl_instantiate_key_iov, keyctl_dh_compute): Change addr and len
arguments to kernel_ulong_t. Print len using %llu format.
(keyctl_keyring_search): Change addr1 and addr2 arguments
to kernel_ulong_t.
(print_dh_params): Change addr argument to kernel_ulong_t.
(SYS_FUNC(keyctl)): Retrieve arguments via getarg_ull, pass them
to the appropriate handlers.
* keyctl.c (keyctl_update_key): Change addr and len arguments
to unsigned.
(keyctl_read_key): Likewise.
(keyctl_instantiate_key): Likewise.
(keyctl_instantiate_key_iov): Likewise.
(keyctl_dh_compute): Likewise.
(keyctl_keyring_search): Change addr1 and addr2 arguments to unsigned.
(keyctl_chown_key): Change user and group arguments to unsigned.
(keyctl_get_persistent): Change uid argument to unsigned.
keyctl_read_key had subtle bug by treating out buffer in KEYCTL_READ
as NUL-terminated, which is not true. We fix it by adding parameter to
keyctl_read_key signalising whether buffer is NUL-terminated and using
printstr_ex for printing (expectedly) NUL-terminated strings.
* keyctl.c (keyctl_read_key): Add has_nul parameter. Do not use -1 as
string len. Use printstr_ex for buffer output with user style depending
on has_nul value.
(SYS_FUNC(keyctl)): Specify has_nul parameter to keyctl_read_key
by comparing cmd value with KEYCTL_READ.
Add printstr_ex which allows for providing user quotation style
This is useful for providing QUOTE_OMIT_TRAILING_0 and maybe other
flags.
* defs.h (printstr_ex): New prototype.
(printstr): Change to a wrapper around printstr_ex with zero user style.
* util.c (printstr): Rename to ...
(printstr_ex) ... new function, add user_style argument which is or'ed
with computed style.
util: add quote_string flag signalising that string is NUL-terminated
It is useful in cases strings with size provided are expected to be
NUL-terminated but are not trustworthy enough to call just plain
printstr(str, -1).
* defs.h (QUOTE_OMIT_TRAILING_0): New constant definition.
* util.c (string_quote): Swallow terminating NUL if
QUOTE_OMIT_TRAILING_0 is set.
* keyctl.c (struct keyctl_dh_params): New structure.
(print_dh_params, keyctl_dh_compute): New functions.
(SYS_FUNC(keyctl)): Add support for KEYCTL_DH_COMPUTE cmd value.
Dmitry V. Levin [Mon, 3 Oct 2016 11:59:39 +0000 (11:59 +0000)]
Use err_name, print unrecognized errno values as numbers
* syscall.c (trace_syscall_exiting): Use err_name() instead
of open-coding it. Print unrecognized errno values using %lu format
instead of ERRNO_%lu as the latter prodices an invalid constant.
Dmitry V. Levin [Mon, 3 Oct 2016 11:48:55 +0000 (11:48 +0000)]
struct tcb: change the type of u_error field from int to unsigned long
This is the type actually used for the error code on architectures
that use a dedicated register.
* defs.h (struct tcb): Change the type of u_error to unsigned long.
* syscall.c (trace_syscall_exiting): Change the type of u_error variable
to unsigned long, print it using %lu format, drop no longer needed
explicit cast to unsigned long.
(saved_u_error): Change type to unsigned long.
keyctl: do not print comma for KEYCTL_SESSION_TO_PARENT command
Since this command doesn't have any additional arguments, the comma does
not needed. Since this is the only command which lacks additional
arguments, it's better to add special case for it rather than add
printing of comma to all other commands.
* keyctl.c (SYS_FUNC(keyctl)): Add check for command not being
KEYCTL_SESSION_TO_PARENT when printing comma dividing cmd argument
from the rest.
UID/GID are unsigned except special -1 value (which is also special in
context of specific keyctl commands), so special printing function
should be used.
* keyctl.c (keyctl_chown_key, keyctl_get_persistent): Use printuid
instead of printf with "%d" conversion for printing UID/GID.
Enhance -e abbrev=set, -e raw=set, and -e verbose=set
Enhance abbrev=, raw=, and verbose= to accept the same syntax as trace=.
For example, this allows such syntax as -e verbose=file.
* syscall.c (lookup_class): Define before qual_syscall.
(qualify): Move the loop based on lookup_class ...
(qual_syscall): ... here.
* tests/qual_syscall.test: Check it.
fadvise: change printing of len argument to unsigned in fadvise64 syscall
The kernel declares fadvise64 as
long sys_fadvise64(int fd, loff_t offset, size_t len, int advice);
* fadvise.c (SYS_FUNC(fadvise64)): Change conversion specifier from
"%ld" to "%lu" for printing len argument since kernel expects argument
of type size_t.
fallocate: change print format of offset and len arguments to signed
Since types of these arguments are off_t and kernel actually expects
signed values in order to fail when negative values are provided,
lets display these values as signed.
* fallocate.c (SYS_FUNC(fallocate)): Change conversion specifier for
printing "offset" and "len" syscall arguments from %llu to %lld.
quota: remove legacy subcommand decoding support code
Support for these quota subcommands by linux kernels has been dropped
long time ago (between 2.5.16 and 2.5.17), so lets drop it in order
to minimize amount of code which should be subjected to testing and
structured conversion.
quota: add packed attribute to struct if_dqblk definition
Since the only difference between 32-bit and 64-bit environments
regarding this structure is its padding, lets just add packed attribute
to its definition instead of going full mpers.
Q_XQUOTASYNC is no-op, but it does not require id/addr arguments anyway.
* quota.c (decode_cmd_data): Add Q_SYNC and Q_XQUOTASYNC to subcommand
dispatch switch since id/addr arguments for these subcommands are known
to be ignored.
quota: avoid printing id argument for those commands that ignore it
As id argument is not used for Q_QUOTAOFF, Q_GETFMT, Q_V2_GETINFO,
Q_GETINFO, Q_V2_SEETINFO, Q_SETINFO, Q_SYNC, Q_XQUOTAON, Q_XQUOTAOFF,
Q_XQUOTARM, Q_XGETQSTAT, Q_XGETQSTATV, Q_XQUOTASYNC, Q_V1_GETSTATS, and
Q_V2_GETSTATS subcommands, do not print it for these subcommands.
* quota.c (decode_cmd_data): Add id argument parameter, add printing
of id argument for Q_GETQUOTA, Q_SETQUOTA, Q_GETNEXTQUOTA,
Q_V1_GETQUOTA, Q_V1_SETQUOTA, Q_V2_GETQUOTA, Q_V2_SETQUOTA,
Q_XGETQUOTA, Q_XGETNEXTQUOTA, Q_XSETQLIM, and unknown subcommands.
(SYS_FUNC(quotactl)): Remove printing of id argument, pass it
to decode_cmd_data.
* quota.c (SYS_FUNC(quotactl)): Since id call argument is used for
user/group/project ID which all have special semantics for -1,
print it the same way as UIDs are printed.
decode_open: print the mode argument when O_TMPFILE flag is set
O_TMPFILE reqires the mode argument (just like O_CREAT), so print it.
* open.c (STRACE_O_TMPFILE): New macro.
(decode_open): Print the mode argument when O_TMPFILE flag is set.
* tests/open.c (main): Check it.
Fixes RH#1377846.
* xlat/fsmagic.in: Add BFS_MAGIC, GFS2_MAGIC, and ROMFS_MAGIC constants
defined in linux/bfs_fs.h, linux/gfs2_ondisk.h, and linux/romfs_fs.h,
respectively.
* tests/ipc_msg.c: Additional msgget (parameter format) and msgctl
(parameter format, decoding of struct msqid_ds in IPC_SET/IPC_STAT
commands) checks.
* tests/ipc_sem.c: Additional semget and semctl checks.
* tests/ipc_shm.c: Additional shmget and shmctl checks.
* tests/semop.c: Additional semop checks. Add checks for semtimedop.
* tests/semop.test: Add explicit -e parameter in order to trace both
semop and semtimedop.
* tests/shmxt.c: Additional shmat and shmdt checks.
* linux/64/ioctls_inc.h: Update from linux v4.8 using ioctls_gen.sh.
* linux/32/ioctls_inc_align32.h: Likewise.
* linux/32/ioctls_inc_align64.h: Likewise.
ipc: fix printing of nsops argument of semop and semtimedop syscalls
According to POSIX, nsops argument of semop syscall has type size_t,
but the kernel treats nsops argument of semop and semtimedop syscalls
as unsigned int.
* ipc_sem.c (tprint_sembuf_array): Change type of "count" argument
from unsigned long to unsigned int, print it using %u format.
Mark io_setup and io_destroy as memory mapping related syscalls
As io_setup syscall allocates some memory using do_mmap_pgoff, and
io_destroy deallocates this memory using vm_munmap, set TRACE_MEMORY
flag for all sysentries of io_setup and io_destroy using the following
oneliner:
sed -ri '/io_setup|io_destroy/ s/0,/TM,/' linux/*/syscallent*.h
* linux/*/syscallent*.h (io_setup, io_destroy): Change sys_flags to TM.
It was incorrectly assumed that __*64 types are long long on all
platforms, despite strace having specially crafted macros in order
to handle precisely this architecture discrepancy.
The commit fixes this oversight.
* tests/aio.c (main): Use PRI__*64 macros for correct format conversion
specifiers for __*64-typed values.