Karol Lewandowski [Thu, 9 Aug 2018 06:58:38 +0000 (08:58 +0200)]
log_dump: Rework service file
This commit adjusts log_dump systemd service to
provide normal dbus-type unit.
Additionally, unused and erroneous "ENGINEER_MODE"
is removed.
Change-Id: I049952ab19d1e6a0b8ee8319cedb83226037814f
Karol Lewandowski [Tue, 31 Jul 2018 11:07:20 +0000 (13:07 +0200)]
Release 1.2.1
Bugfix release, includes:
* Fix off-by-one error in generating the so_info file
Change-Id: I4630757714e45db085980f2e002613df5375d9af
Mateusz Moscicki [Wed, 25 Jul 2018 07:18:00 +0000 (09:18 +0200)]
Fix off-by-one error in generating the so_info file
Change-Id: I65dba1b2be1237578df5679586f1449d4fa94023
Karol Lewandowski [Fri, 20 Jul 2018 10:01:28 +0000 (12:01 +0200)]
Release 1.2
This release brings following features:
* rewritten rpm interface - from using rpm command to direct librpm
usage - this offers considerable speed improvement
* waiting for /opt (or generally CRASH_ROOT_PATH) to appear -
useful for crashes occuring early during bootup
Minor fixes/debugging enchancements are provided.
Change-Id: Idb193bebc56c5f8d86fa360de1787cadb6a75583
Mateusz Moscicki [Tue, 17 Jul 2018 11:57:41 +0000 (13:57 +0200)]
Log message to dlog when wait_for_child returns non-zero value
Change-Id: Ib2b3df0939a13a553b05cc747e5b17a4e5fe580d
Mateusz Moscicki [Tue, 17 Jul 2018 11:43:39 +0000 (13:43 +0200)]
Add own implementation of getting rpm packages info
Calling the external rpm command is too slow, especially for binaries
with many shared object dependencies.
Change-Id: I48a86c0a0067bb2d37b7b7a12abae5d611c8b572
Mateusz Moscicki [Tue, 17 Jul 2018 08:30:31 +0000 (10:30 +0200)]
Copy /proc/<pid>/maps file to temporary directory
To save as much time as possible, we can parse maps file when coredump
is already saved and process is probably cleaned up.
Change-Id: I136d1c462fe2fe9c204deea09b1d0269cee32530
Mateusz Moscicki [Thu, 14 Jun 2018 11:18:22 +0000 (13:18 +0200)]
Add checking if CRASH_ROOT_PATH exists
When system booting, /opt may not be mounted yet. In this case if any
program will crash, the crash-manager will wait about 60 seconds for
/opt. After this time the the crash report will be lost.
Change-Id: I97e9541262d1a0f9844aa260ac23972f70e27638
Mateusz Moscicki [Mon, 2 Jul 2018 07:02:31 +0000 (09:02 +0200)]
Return exit code of the program invoked by run_command_write_fd_timeout()
Change-Id: Ib7f046aed8ca352463d6ba7559f947e487ad85b8
Mateusz Moscicki [Fri, 29 Jun 2018 11:31:38 +0000 (13:31 +0200)]
Fix reading /proc/<pid>/maps file
Macro is expanded by preprocessor, but sizeof() is evaluated at compile time.
Change-Id: I4ca20ff703cf89fba2137ec3eb24550d1013fdff
Karol Lewandowski [Mon, 18 Jun 2018 14:44:01 +0000 (16:44 +0200)]
packaging: Fix minicoredumper path
Change-Id: I5a57f9a3c0272ebed534aee7edd88d2c2f362b89
Karol Lewandowski [Mon, 18 Jun 2018 13:06:06 +0000 (15:06 +0200)]
Release crash-worker with minicoredumper
Change-Id: Ie72ab3379f50174e1c76c79da5af56cb8aaed5ad
Mateusz Moscicki [Thu, 22 Feb 2018 09:45:37 +0000 (10:45 +0100)]
crash-manager: add minicoredumper
minicoredumper strips unnecessary data from coredump
and makes use of sparse file mechanism, to quickly save
core files on disc.
Change-Id: Ic8e968b5e7a011f9a06794876c372063e968a354
Mateusz Moscicki [Tue, 20 Mar 2018 07:43:10 +0000 (08:43 +0100)]
Add *.so_info file to the report
*.so_info file contains paths, rpm package info (name, version, release,
architecture) and build IDs of mapped files that have a execute flag in
/proc/<pid>/maps.
Change-Id: I087a8d25efeb1f51a5e25d95cce5c60a8f2a607b
Karol Lewandowski [Fri, 15 Jun 2018 13:51:28 +0000 (13:51 +0000)]
Merge "Add timeout to run_command_write_fd and replace popen by execvpe" into tizen
Mateusz Moscicki [Thu, 14 Jun 2018 10:37:56 +0000 (12:37 +0200)]
Fix stack-buffer-overflow error
Lenghts in sscanf format string refer to actually read bytes, without
including space for null byte, E.g.:
...
char buff[5+1];
sscanf(other_buff, "%5s", buff);
...
Change-Id: I203e1bc04ba1e352029849e5bd29a7a6ca8a5763
Sunmin Lee [Mon, 18 Sep 2017 04:27:50 +0000 (13:27 +0900)]
Fix vulnerabilities
Change-Id: Ia7d8565ef86cfef9d8b61e2e11020b6f1b49ed1d
Signed-off-by: Sunmin Lee <sunm.lee@samsung.com>
Mateusz Moscicki [Wed, 16 May 2018 06:23:00 +0000 (08:23 +0200)]
Add timeout to run_command_write_fd and replace popen by execvpe
dump_systemstate runs many external commands. When one hungs,
dump_systemstate will block the crash-worker.
execvpe replaced popen because popen passes command to /bin/sh, and this
can potentially allow to run the malicious command if attacker can
modify the fragment of cmd, e.g:
void read_file(char *filename)
{
...
sprintf(buff, "cat %s", filename);
popen(buff, "r");
...
}
main()
{
char filename[] = "/etc/passwd;rm -rf /";
read_file(filename);
}
Change-Id: Id7b37c058869c27d3c4d282d9d2dd30d5b9ec80c
Mateusz Moscicki [Wed, 30 May 2018 11:33:20 +0000 (13:33 +0200)]
crash-pipe: Fix write the coredump to the file
splice returns EINVAL if (loff_t)(len + out.pos) is negative number.
This was the case on aarch64 and x86_64.
Change-Id: Ic52d08efce3fca555a83319c8161c3656b553cc1
Mateusz Moscicki [Wed, 30 May 2018 11:31:21 +0000 (13:31 +0200)]
Revert "crash-pipe: Fix write the coredump to the file"
This reverts commit
70c1372b5a85a64d5ea7c5bb20b99429213942be.
Change-Id: I553b4c6ef2e5aa9cdc80e4d58ecd318dc0a2d532
Mateusz Moscicki [Fri, 20 Apr 2018 06:35:39 +0000 (08:35 +0200)]
Stop setting the PR_SET_DUMPABLE flag for crash-manager processes
prctl(PR_SET_DUMPABLE, 0) is not neccessary. Kernel runs the
crash-manager and sets RLIMIT_CORE to 1 for the process. This is special
value that prevents from running crash-manager recursively.
Change-Id: I33df7c28c6ce939f2903d02af673734d473b649e
Karol Lewandowski [Tue, 29 May 2018 08:09:53 +0000 (08:09 +0000)]
Merge "crash-pipe: Fix write the coredump to the file" into tizen
Mateusz Moscicki [Mon, 28 May 2018 10:46:32 +0000 (12:46 +0200)]
crash-pipe: Fix write the coredump to the file
splice len parameter is size_t type. On some architectures (eg. aarch64)
splice returned EINVAL when the parameter len was ssize_t variable and
was equal SSIZE_MAX
Change-Id: Id84038662dcd2969a33af1e276a3b4500a780f29
Mateusz Moscicki [Fri, 11 May 2018 13:34:42 +0000 (15:34 +0200)]
Set the default DBus policy for send and own to deny
Change-Id: I49230b5c711cf7c4158f6de2fb448638035d750b
Sunmin Lee [Thu, 28 Sep 2017 10:05:07 +0000 (19:05 +0900)]
Fix potentially undefined behavior
The documentation does not describe the behavior of the access()
function for a NULL path argument, which can be returned by mkdtemp()
Change-Id: I9c5c08eb424f5e9bdab248682b00e9cb895c7afa
Signed-off-by: Sunmin Lee <sunm.lee@samsung.com>
[ Cherry-picked from tizen_4.0 branch, extended commit message ]
Signed-off-by: Mateusz Moscicki <m.moscicki2@partner.samsung.com>
Mateusz Moscicki [Mon, 26 Feb 2018 12:22:47 +0000 (13:22 +0100)]
crash-manager: don't store in zip archive files referred by symlinks
minicoredump (will be introduced in next commit) has an option to save
/proc/<pid>/ files in report. As /proc/<pid>/fd/* contains symlinks, the
created archive will contain the contents of these fds. This is
undesirable as files might be quite big and contain private (sensitive)
data.
This commit adds -y option to zip, which will allow storing symlinks to
actual files instead of their contents (as it was available originally
under /proc/<pid>/)"
Change-Id: I0a7c9211eab1acd39da7db3930e967ddab5eceef
Mateusz Moscicki [Thu, 22 Feb 2018 09:40:41 +0000 (10:40 +0100)]
crash-manager: fix get cmd_path
This commit fixes issue with report not being generated for processes
started using a relative path (e.g.: ./test-app).
In above case cmdline contained only './test-app' string and based on
that it was impossible to determine the full path to a binary.
Consequently, crash-manager assumed crashed program binary did not
exist and did not generate the report.
This commit changes the logic as follows:
cinfo->cmd_name is read from /proc/<pid>/cmdline
cinfo->cmd_path is read from /proc/<pid>/exe
Change-Id: If1b7557b53417d2703a81644939692fee5c801ee
Mateusz Moscicki [Mon, 26 Feb 2018 07:17:53 +0000 (08:17 +0100)]
Fixed code style
Code has been changed to avoid messages:
(CODE_SMELL/Warning) Remove this unused "ret" local variable.
(CODE_SMELL/Warning) Remove this unused "command" local variable.
In addition, the code has been modified to remove unnecessary bracket.
Change-Id: I5bf9820d16f7b7947844e078eba9a1bab11ddd6c
Mateusz Moscicki [Wed, 14 Feb 2018 07:17:02 +0000 (08:17 +0100)]
crash-stack: fixed copying data from bufferfile to outputfile
Reading the last chunk of data sets the feof indicator.
Change-Id: I396a3d6e1c398c7201ef8bc7dfa1ec9c1b7ddb8a
Michal Bloch [Thu, 28 Sep 2017 14:38:52 +0000 (16:38 +0200)]
Improve error handling
* handle reallocation failures in scan dump.
* generate a log if unlinking a failed dump file also fails
* fix a memory leak when passing duplicate `--save-core` parameter to crash-pipe; generate a log in that case
* explicitly free a resource on exit
Change-Id: I71b22b3375c6c9ab9a43dab41e80eadc3fa599f6
Signed-off-by: Michal Bloch <m.bloch@samsung.com>
Karol Lewandowski [Thu, 28 Sep 2017 14:59:53 +0000 (14:59 +0000)]
Merge "svace fixes: insufficient error handling" into tizen
Karol Lewandowski [Tue, 26 Sep 2017 11:29:32 +0000 (13:29 +0200)]
svace fixes: insufficient error handling
This commit brings fixes for folowing SVACE-reported issues:
* UNCHECKED_FUNC_RES.FREAD: Return value of fread function was compared to zero at crash-stack.c:836, but ferror/feof wasn't called.
Result was compared with zero at /home/abuild/rpmbuild/BUILD/crash-worker-1.0.1/src/crash-stack/crash-stack.c:836
* HANDLE_LEAK: The handle 'fd' was created at crash-manager.c:508 by calling function 'open' and lost at crash-manager.c:515.
[acquire] Call of open at /home/abuild/rpmbuild/BUILD/crash-worker-1.0.1/src/crash-manager/crash-manager.c:508
[leaked] leaked at /home/abuild/rpmbuild/BUILD/crash-worker-1.0.1/src/crash-manager/crash-manager.c:515
* DEREF_OF_NULL.CONST: Pointer '&tail->next', which was assigned NULL value at crash-stack.c:642, is dereferenced at crash-stack.c:692.
[dereference] Variable '&tail->next' is dereferenced at /home/abuild/rpmbuild/BUILD/crash-worker-1.0.1/src/crash-stack/crash-stack.c:692
[null] Assign null at /home/abuild/rpmbuild/BUILD/crash-worker-1.0.1/src/crash-stack/crash-stack.c:642
* RACE.NO_UMASK: Function 'umask(077)' needs to be called before 'mkstemp' at crash-stack.c:964, to prevent a potential race condition vulnerability.
function call at /home/abuild/rpmbuild/BUILD/crash-worker-1.0.1/src/crash-stack/crash-stack.c:964
Change-Id: Ief6dd93ec8d795fccffbc2d823a6af8fcf63c965
Kichan Kwon [Thu, 28 Sep 2017 01:46:21 +0000 (10:46 +0900)]
dump_systemstate : add journal log
- Log only latest 1024 logs from last boot
Change-Id: I535402a6807d2c4a21fb08a2ba874473f49045f6
Signed-off-by: Kichan Kwon <k_c.kwon@samsung.com>
Sunmin Lee [Wed, 13 Sep 2017 05:36:18 +0000 (14:36 +0900)]
Fix bugs
- Fix misspelling for m4 condition
- Do not install unused services: tizen-debug-on/off
Change-Id: I09296e1365407c93ebb6dd9897ccc5b9e346bddd
Signed-off-by: Sunmin Lee <sunm.lee@samsung.com>
Sunmin Lee [Tue, 12 Sep 2017 07:28:52 +0000 (16:28 +0900)]
Remove unused file
Remove unused file
Change-Id: Iabe9ee9e49cf9f4272c3a3691f48f3c75ff0a281
Signed-off-by: Sunmin Lee <sunm.lee@samsung.com>
ByungSoo Kim [Mon, 21 Aug 2017 14:24:00 +0000 (23:24 +0900)]
log_dump: assign CAP_SYS_PTRACE capability
When log_dump has system_fw permission,
memps which is forked from /usr/bin/dump_systemstate can't read
/proc/self/maps and /proc/self/smaps.
It requires CAP_SYS_PTRACE capability.
So, it should have CAP_SYS_PTRACE capability and inheritance attribute.
Change-Id: If2bd16964dba8e616e4a4fcc5cd489feb4c40b21
Signed-off-by: ByungSoo Kim <bs1770.kim@samsung.com>
minsoo kim [Mon, 21 Aug 2017 05:49:32 +0000 (14:49 +0900)]
crash-worker: add dump_systemstate memps
Change-Id: I50fbdc296a0ab8dfab96cfde29b0c3afa94eb1e7
Signed-off-by: minsoo kim <ms203.kim@samsung.com>
minsoo kim [Mon, 3 Jul 2017 05:04:22 +0000 (14:04 +0900)]
crash-worker: add supplementarygroups log systemd-journal
To gather dlog, journal log, add supplementarygroups to service
But dbus activation not applied suppplementarygroup,
So rearrange service role
org.tizen.system.crash.service : dbus activation
log_dump.service : log_dump daemon
Change-Id: I93e2bb5142a83f6dfa9ad56197916855988501ee
Signed-off-by: minsoo kim <ms203.kim@samsung.com>
minsoo kim [Wed, 10 May 2017 02:09:09 +0000 (11:09 +0900)]
crash-worker: change permission to none root process
for reduce root permission, change to system_fw
Change-Id: I04fafa89d2a4e0a1e3bc5a6e0019b2aebb428235
Signed-off-by: minsoo kim <ms203.kim@samsung.com>
Sunmin Lee [Wed, 9 Aug 2017 10:38:41 +0000 (19:38 +0900)]
Apply ASLR to executables: crash-pipe, crash-stack
Change-Id: Ia4a0452832a705f6fbac8034d7da93297883e4bc
Signed-off-by: Sunmin Lee <sunm.lee@samsung.com>
Karol Lewandowski [Tue, 23 May 2017 13:01:36 +0000 (15:01 +0200)]
packaging: Disable sys-assert
Ptrace-based mechanism (crash-stack) should be used as default
and only mechanism for crash-report generation.
Change-Id: I102fc02e5538dcf9f21d7b72ff8892e1b3500265
Kunhoon Baik [Tue, 27 Jun 2017 07:07:40 +0000 (16:07 +0900)]
Preparation of debug & release package
Some vendor would like to have different crash management policy for debug and release image.
Thus, Tizen will provide crash-manager debug and release package.
This patch is the preparation for the package separation.
In case of crash-manager debug, following CFLAGS will be enabled.
export CFLAGS+=" -DTIZEN_ENGINEER_MODE"
export CFLAGS+=" -DTIZEN_ENABLE_COREDUMP"
In the future, export CFLAGS+=" -DTIZEN_DEBUG_ENABLE" may be added.
Basically, default Tizen image will include crash management policy for debug mode.
cf) .debugmode file will be used for just crash-popup control for 3.0 compatibility.
Change-Id: I892d1f0d5de65da0abb3e002fa235989f43def56
minsoo kim [Fri, 26 May 2017 09:58:35 +0000 (18:58 +0900)]
log_dump: selective mode for dbus activation
log_dump running root in debug mode, and none_root with normal mode.
Simplify org.tizen.system.crash.service conditional compile by using m4
Change-Id: I04b734f722266264434b27e9fe7b17cdc73263f1
Signed-off-by: minsoo kim <ms203.kim@samsung.com>
Signed-off-by: Sunmin Lee <sunm.lee@samsung.com>
Sunmin Lee [Fri, 9 Jun 2017 03:58:48 +0000 (12:58 +0900)]
Adjust buffer size for address
The addr gets string of address consisting of begin/end
address and '-', null character.
So the length of addrees buffer should be set to:
ADDR_LEN * 2 + 2 (at least)
Change-Id: I6cdb8530be50c4c29fbb51a5dcb5fa8e094e5352
Signed-off-by: Sunmin Lee <sunm.lee@samsung.com>
Sunmin Lee [Thu, 8 Jun 2017 05:05:57 +0000 (14:05 +0900)]
Fix sscanf vulnerability
Specify a limit on the input string length
and keep the last byte for null character.
Change-Id: I717ac2ae565f2627e5de26426ec24c6ccf772c5e
Signed-off-by: Sunmin Lee <sunm.lee@samsung.com>
Karol Lewandowski [Tue, 16 May 2017 13:25:06 +0000 (15:25 +0200)]
crash-stack: Fix Tizen Coding Style violations
Rules at: https://wiki.tizen.org/wiki/Native_Platform_Coding_Idiom_and_Style_Guide
This commit brings no functional changes.
Change-Id: Ic95fd2d1e5508d602fa24ea33933448c52ce625d
Karol Lewandowski [Thu, 11 May 2017 13:47:18 +0000 (15:47 +0200)]
crash-stack: Change "ThreadID" heuristic to work on aarch64
Commit
82e3caa86 ("crash-stack: Find crashed tid by wchan")
introduced heuristic method of finding correct thread id, if
it was not passed by the kernel as parameter to crash-manager.
This commit modifies this heuristic to also consider "pipe_wait"
function in kernel to signify crashed thread. This is what was
observed to be true on aarch64.
Change-Id: Iefa6d028f459ff9491f8e184433f913f39a096eb
Karol Lewandowski [Fri, 5 May 2017 13:03:20 +0000 (15:03 +0200)]
crash-worker: Generate core only if /opt/etc/.debugmode file is present
Additionally, this patch moves actual path definition to one place.
Change-Id: I984ac2c67d8a22ed749f2b1945a1cef237b88ace
Karol Lewandowski [Tue, 25 Apr 2017 06:48:19 +0000 (06:48 +0000)]
Merge "Removing dependency of tizen-debug-on/off service from sys-assert" into tizen
Karol Lewandowski [Mon, 24 Apr 2017 15:34:18 +0000 (17:34 +0200)]
Removing dependency of tizen-debug-on/off service from sys-assert
Tizen-debug-on/off service should control enabling/disabling coredump generation.
When Tizen-debug-off service is executed, only callstack should be generated.
Thus, although disabling sys-assert, the services should be remained.
TODO : For disabling sys-assert, the services should be changed properly.
Change-Id: I617356cd31b51dbe4f103cb3118fefe454e717db
Karol Lewandowski [Sat, 22 Apr 2017 10:47:35 +0000 (10:47 +0000)]
Merge "spec: change group(system -> root) on tizen-debug-off.[service|socket]" into tizen
INSUN PYO [Thu, 20 Apr 2017 05:09:37 +0000 (14:09 +0900)]
spec: change group(system -> root) on tizen-debug-off.[service|socket]
Signed-off-by: INSUN PYO <insun.pyo@samsung.com>
Change-Id: I32605d431b3ff16cce61ae2dde7be262a1eac56f
Karol Lewandowski [Wed, 15 Mar 2017 15:43:03 +0000 (16:43 +0100)]
crash-pipe: Drop all reporting functionality
Reporting is handled by crash-stack.
Change-Id: I6066c5a414fd2e98efbcc0ddc57f2f967cbfb859
Karol Lewandowski [Wed, 15 Mar 2017 15:39:51 +0000 (16:39 +0100)]
crash-pipe: Introduce option to copy core using splice(2)
Linux splice(2) avoids user-space copy and is effectively faster than
regular read(2) + write(2).
According to (ftrace) test below splice is about 20% faster.
sysctl -w kernel.core_pattern='|/usr/libexec/core-pipe --save-core /opt/usr/share/temp/core.%p'
cd /sys/fs/kernel/tracing/
echo do_coredump > set_ftrace_filter
echo function_graph > current_tracer
cat trace_pipe
/* splice */
2) $ 2183078 us | } /* do_coredump */
4) $ 2127946 us | } /* do_coredump */
1) $ 2377004 us | } /* do_coredump */
1) $ 2088114 us | } /* do_coredump */
/* read/write */
1) $ 3088297 us | } /* do_coredump */
1) $ 2953330 us | } /* do_coredump */
1) $ 2647784 us | } /* do_coredump */
3) $ 2271315 us | }
Change-Id: Id2c681a364ce96d98c70329abdf4a64f6d0b405d
Karol Lewandowski [Tue, 18 Apr 2017 13:00:30 +0000 (15:00 +0200)]
Merge branch 'tizen_3.0'
This commit brings followings fixes from tizen_3.0 branch:
- Fallback to ptrace call-stack unwinder if sys-assert failed
- libunwind: Fix off by one in _create_crash_stack
- sys-assert: Fix handle leak of mmap
Change-Id: I422c4ddff0fb4dfb72738d70a7170080304081dd
Karol Lewandowski [Fri, 14 Apr 2017 07:36:14 +0000 (09:36 +0200)]
crash-stack: Eliminate now-deprecated readdir_r
This commit fixes build break with new toolchain.
Glibc 2.24 deprecated readdir_r() - readdir() now recommended even
for multithreaded programs. This commit fixes following error:
/home/abuild/rpmbuild/BUILD/crash-worker-1.0.0/src/crash-stack/crash-stack.c: In function 'find_crash_tid':
/home/abuild/rpmbuild/BUILD/crash-worker-1.0.0/src/crash-stack/crash-stack.c:891:4: error: 'readdir_r' is deprecated [-Werror=deprecated-declarations]
while (readdir_r(dir, &entry, &dentry) == 0 && dentry) {
^~~~~
Change-Id: I99f22a0de87f2539988e1669ae2149dcac74a4df
Sunmin Lee [Wed, 5 Apr 2017 07:53:29 +0000 (16:53 +0900)]
sys-assert: Fix handle leak of mmap
Unmap the memory after using it if it was mmaped.
Change-Id: Ia567c165ac09308a57ae8bb5eac4a9a2a524ec15
Signed-off-by: Sunmin Lee <sunm.lee@samsung.com>
Sunmin Lee [Mon, 3 Apr 2017 01:27:39 +0000 (10:27 +0900)]
Revert "Separate logdump from crash-worker"
This reverts commit
c7826eefdccc5776eb0665cb0feb5d060e69fe96.
Change-Id: I398cf722d445b037d32d8116cc9c18a3a326b673
Karol Lewandowski [Tue, 7 Mar 2017 15:13:18 +0000 (16:13 +0100)]
Fallback to ptrace call-stack unwinder if sys-assert failed
This commit enables ptrace-based call stack unwinder
(TIZEN_FEATURE_PTRACE_CALLSTACK).
When both SYS_ASSERT and PTRACE flavors are enabled
crash-manager will default to SYS_ASSERT, and fallback
to PTRACE only if former failed to generate the report
(eg. client application overwriten libsysassert's
signal handlers).
Change-Id: Ifc3626b7b46f47c3fd68efcb18d3900908ce3884
Karol Lewandowski [Wed, 8 Mar 2017 15:23:52 +0000 (16:23 +0100)]
libunwind: Fix off by one in _create_crash_stack
This fixes call stack missing last element (usually _start,
in the case of invalid PC - the PC itself).
Change-Id: Ib1dae12bd06f1a4dfa1339fa6ea704bff05926c9
Sunmin Lee [Tue, 28 Feb 2017 06:02:25 +0000 (15:02 +0900)]
Separate logdump from crash-worker
In Tizen 4.0, the log_dump tool is supposed to be managed
in separate package.
Change-Id: Ie45cfdd9ee0530ef2e520ac781e9971120c87aeb
Signed-off-by: Sunmin Lee <sunm.lee@samsung.com>
Sunmin Lee [Fri, 24 Feb 2017 02:27:58 +0000 (11:27 +0900)]
Merge branch 'tizen_3.0' into tizen
Change-Id: I56715923d0ac0257736598e711d2c113548372cb
Sunmin Lee [Tue, 21 Feb 2017 05:02:20 +0000 (14:02 +0900)]
Fix vulnerability
- Enforce read size of read() less than buffer
- Do null check about opened bufferfile
Change-Id: Ie7d43eb1b04e68e03e97d5e9fed75aee669bfd0a
Signed-off-by: Sunmin Lee <sunm.lee@samsung.com>
Sunmin Lee [Tue, 21 Feb 2017 01:07:11 +0000 (10:07 +0900)]
Execute dump_systemstate in parallel
Since rearrange the order of crash modules, coredump had made
too late. It has caused several issues such as missing coredump
when something happened during dump_systemstate.
Therefore, execute dump_systemstate in parallel.
In addition, to avoid confusing launch crash-popup right after
whole crash works are finished.
Change-Id: I73c10c7d60f04d218f313b760c0381d0b8d95de7
Signed-off-by: Sunmin Lee <sunm.lee@samsung.com>
Sunmin Lee [Mon, 20 Feb 2017 09:46:44 +0000 (18:46 +0900)]
crash-stack: Get executable name from cmdline
The executable name in crash-stack is different with sys-assert.
In case of some applications, "exe" has indicated real path of app
that makes it hard to know what the actual app name is.
It's better to refer to "cmdline" instead.
Change-Id: I4895dbce427d50cd20a593585f2640554a4268fc
Signed-off-by: Sunmin Lee <sunm.lee@samsung.com>
Sunmin Lee [Wed, 28 Dec 2016 08:03:36 +0000 (17:03 +0900)]
Print pre-ptrace info to buffer file
Some information had lost after ptrace.
However, the order of info also should be preserved.
To resolve this, that information are going to be printed
to temporal space first and will be merged in result file.
Change-Id: I523ded4ffc15b5edeac1944b62b5a42a5a44bc83
Signed-off-by: Sunmin Lee <sunm.lee@samsung.com>
Sunmin Lee [Fri, 6 Jan 2017 00:37:33 +0000 (09:37 +0900)]
crash-stack: Find crashed tid by wchan
If tid is not offered by kernel, crash-stack would
find crashed tid by checking wchan value of each thread.
Change-Id: Idfd6866fe17af4c7a266b3ae127e39087b0848f0
Signed-off-by: Sunmin Lee <sunm.lee@samsung.com>
Sunmin Lee [Mon, 9 Jan 2017 05:03:34 +0000 (14:03 +0900)]
crash-manager: Utilize cmdline instead of core_pattern
There are limitations on using core_pattern variables to get command info.
%e: *its length is limited (up to 16 byte)
%E: in case of some applications, it returns not actual app binary path
(it returns launchpad-loader)
Because of these reasons, use cmdline of /proc instead to get command info.
(It also makes a consensus with sys-assert)
* Refer to: http://lxr.free-electrons.com/source/include/linux/sched.h#L1663
Change-Id: Ie486785485e4bcee2c62fb1e4fa98329d0f01170
Signed-off-by: Sunmin Lee <sunm.lee@samsung.com>
Sunmin Lee [Wed, 15 Feb 2017 07:27:57 +0000 (16:27 +0900)]
Set suid_dumpable to debug (1)
As a platform policy, Tizen offers core dumping for
all process by default. However, until now some process
has got cleared dump flag by the system.
We modify it to these processes also could be dumpable.
Change-Id: I2f8fb6a76b891f4ce6af97349ab963ef5c8183fc
Signed-off-by: Sunmin Lee <sunm.lee@samsung.com>
Łukasz Stelmach [Wed, 1 Feb 2017 13:58:04 +0000 (14:58 +0100)]
crash-stack: Make output match sys-assert report
Change-Id: I7ecc9b045cc9b83d328d1d97bb08d4174c9d72ba
Adrian Szyndela [Mon, 2 Jan 2017 14:03:58 +0000 (15:03 +0100)]
crash-stack: remove the core dump handling remains
Hopefully, there is no more rotting flesh.
Change-Id: I59003e8a38dd2817f78d23319848273615a0a5be
Adrian Szyndela [Mon, 2 Jan 2017 13:37:54 +0000 (14:37 +0100)]
crash-stack: add file offset to output
Change-Id: Ib9f77b61d98a60e192877fad965a30ad8730af2f
Rafal Pietruch [Fri, 30 Dec 2016 07:38:29 +0000 (08:38 +0100)]
crash-stack: disable unw_get_proc_name prediction
libunwind unw_get_proc_name function may return the name of
a label or a preceeding, nearby procedure; although the value
of the returned name plus the returned offset will
be equal to the instruction-pointer of the stack frame,
developers can find a exact symbol with related code position
in debug packages
Change-Id: I2471c2a9483b577844110f25128a6d539d39e5ad
Rafal Pietruch [Wed, 21 Dec 2016 16:39:01 +0000 (17:39 +0100)]
crash-stack: integration test added
Change-Id: Iaac051671b89018dda37c212bb1c9da756879ed4
Sunmin Lee [Wed, 18 Jan 2017 05:23:37 +0000 (14:23 +0900)]
(Modified) Deal with Toolchain upgrade
(Revert "Deal with Toolchain upgrade")
This patch substitutes deprecated function "readdir_r" with "readdir"
rather than ignoring warning.
Change-Id: I3b4d3f9c28cc60c851f30332aa88780078471085
Signed-off-by: Sunmin Lee <sunm.lee@samsung.com>
Sunmin Lee [Mon, 2 Jan 2017 04:26:55 +0000 (13:26 +0900)]
Deal with Toolchain upgrade
According to Tizen toolchain will be upgraded in 4.0,
several build issues should be resolved.
This patch deals with "Language rules stricting".
Change-Id: I1993fb3b2a285c341f8b8b92ec32a428704ddc5a
Signed-off-by: Sunmin Lee <sunm.lee@samsung.com>
Sunmin Lee [Fri, 30 Dec 2016 00:59:56 +0000 (09:59 +0900)]
Merge branch 'tizen_3.0' into tizen
Change-Id: I62eb5bc520f06ea0daa13a7de4c3364d0a12695d
Sunmin Lee [Wed, 28 Dec 2016 05:35:13 +0000 (14:35 +0900)]
Change internal modules work
- crash-pipe: create core file only
- crash-stack: create .info file (call-stack)
Change-Id: I71ba012f426e5d055f3acfc2b2820b5fcb2e17ec
Signed-off-by: Sunmin Lee <sunm.lee@samsung.com>
Sunmin Lee [Wed, 28 Dec 2016 01:20:32 +0000 (10:20 +0900)]
Fix vulnerability
Check range of signo before use it
Change-Id: Icae63a7185a897ba6688b45715b208c2d92df1b7
Signed-off-by: Sunmin Lee <sunm.lee@samsung.com>
Kunhoon Baik [Tue, 27 Dec 2016 13:00:35 +0000 (22:00 +0900)]
crash-stack: libunwind used on aarch64
Make use of "ptrace: Add support for GETREGSET" patch to libunwind
https://lists.nongnu.org/archive/html/libunwind-devel/2016-07/msg00001.html
Depends-On: I2e720e6697fbc1facf1d7547b398f5665b17731e
Change-Id: I0ce3886d271c9b78d2a44b99a628ccc502e6e400
Sunmin Lee [Tue, 27 Dec 2016 05:49:41 +0000 (14:49 +0900)]
crash-stack: Simplify signal info
It is hard to get signal info which caused dump using ptrace.
Therefore, crash-stack prints signal number only
which came from core_pattern.
Change-Id: I6cd865a89d3e2c5dd83cfe134bb8cddec4f7f81f
Signed-off-by: Sunmin Lee <sunm.lee@samsung.com>
Sunmin Lee [Tue, 27 Dec 2016 11:15:52 +0000 (20:15 +0900)]
Remove PD License
Change-Id: Iaa3daf6bf96c5ae59ad77515a0f942f4d4a71586
Signed-off-by: Sunmin Lee <sunm.lee@samsung.com>
Sunmin Lee [Tue, 27 Dec 2016 10:33:55 +0000 (19:33 +0900)]
Adjust smack label
In response to security issue, restrict smack label of crash-manager.
In addition, adjust crash directories label to accord with
work of crash-worker.
Change-Id: I97390d50337fd36a746e260f3f6ebc29c6990526
Signed-off-by: Sunmin Lee <sunm.lee@samsung.com>
Sunmin Lee [Tue, 27 Dec 2016 10:48:51 +0000 (02:48 -0800)]
Merge "log_dump: Restrict dbus permission" into tizen_3.0
Sunmin Lee [Tue, 27 Dec 2016 07:44:49 +0000 (16:44 +0900)]
Separate package dependencies of crash-stack
According to experimental variable, exclude unused package
dependencies from crash-worker.spec.
+ Bug fix:
"crash-stack: Make ptrace callstack feature selective"
Change-Id: If8746a74fb937f7929538a995d6f1cf392e19bd4
Signed-off-by: Sunmin Lee <sunm.lee@samsung.com>
Sunmin Lee [Tue, 27 Dec 2016 02:25:12 +0000 (11:25 +0900)]
log_dump: Restrict dbus permission
log_dump can be activated using dbus method call.
Because this feature should be used by administrator only,
restrict the dbus activation permission.
Change-Id: Ib63ea10fd5101af146e35cf9ef47e96f91a3229f
Signed-off-by: Sunmin Lee <sunm.lee@samsung.com>
Sunmin Lee [Mon, 26 Dec 2016 11:31:43 +0000 (20:31 +0900)]
crash-stack: Make ptrace callstack feature selective
In Tizen 3.0, the crash-stack using ptrace to generate callstack
will be offered as a choice.
Developers can enable this feature through setting a variable as on:
TIZEN_FEATURE_PTRACE_CALLSTACK
Change-Id: I7dfbf563144174abcd5e50677eb8943cdb8d8ebe
Signed-off-by: Sunmin Lee <sunm.lee@samsung.com>
Kunhoon Baik [Mon, 4 Jul 2016 11:01:31 +0000 (20:01 +0900)]
Remove SIGSEGV handling
SIGSEGV (signal number - 11) may occur double crash when sys-assert tries to unwind backtrace.
like that
If the double crash is generated, it causes useless backtrace of corefile (double crashed memory dump)
Thus, sys-assert does not handle SIGSEGV.
Change-Id: I826ee2bab796fc87b2b4a003683315514af96339
Sunmin Lee [Thu, 22 Dec 2016 08:58:58 +0000 (17:58 +0900)]
Fix handle leak
Fix handle leak of file pointer at crash-stack
Change-Id: I2d68e04ad84ea7263e3c45cc91ef295557532bde
Signed-off-by: Sunmin Lee <sunm.lee@samsung.com>
Karol Lewandowski [Tue, 20 Dec 2016 17:33:58 +0000 (09:33 -0800)]
Merge "crash-stack: use libunwind for i686 and x86_64" into tizen_3.0
Karol Lewandowski [Tue, 20 Dec 2016 16:23:33 +0000 (08:23 -0800)]
Merge "crash-stack: fixed attaching to thread" into tizen_3.0
Rafal Pietruch [Thu, 15 Dec 2016 15:28:17 +0000 (16:28 +0100)]
crash-stack: use libunwind for i686 and x86_64
Change-Id: I0022cc3ac772ba26ed67f7d64d45c3ab4e22db88
Adrian Szyndela [Fri, 16 Dec 2016 12:27:35 +0000 (13:27 +0100)]
crash-stack: fixed attaching to thread
It is not necessary to attach to parent process main thread to
resolve stack in another thread. This commit also solves hanging up
on 'D' (and 'Z') processes - main thread is not attachable,
when a crash occurs in another thread.
Change-Id: Ic3a085951acb4057fcdd4a476693954d1d3a522b
Sunmin Lee [Mon, 19 Dec 2016 06:26:30 +0000 (15:26 +0900)]
crash-manager: consider backward compatibility for crash-stack
Though the crash-stack is available about tid option,
previous kernel has not supported this feature.
Therefore the crash-manager should call crash-stack
considering backward compatibility.
Change-Id: I779783ed46760b88a6d454a05710b29cd4418a44
Signed-off-by: Sunmin Lee <sunm.lee@samsung.com>
Sunmin Lee [Tue, 20 Dec 2016 09:41:00 +0000 (18:41 +0900)]
crash-stack: deprecate corefile option
crash-stack should be able to generate call stack
using pid only.
(Generating corefile causes sluggish issue)
Change-Id: Ie9e74494b966838d1dd6548424d51bfd2df4c1e0
Sunmin Lee [Thu, 15 Dec 2016 02:26:38 +0000 (11:26 +0900)]
Revert "engineer mode: make .debugmode on engineer image"
.debugmode is moved to other package
This reverts commit
b153665a12cda6a29f9ba9d12db7f2682ade9af5.
Change-Id: Idfe9cae63b8f4e4122c74abb144250eea45d9f07
Adrian Szyndela [Thu, 15 Dec 2016 12:03:21 +0000 (13:03 +0100)]
crash-stack: get registers from tid instead of pid
Change-Id: Id4ea3f57808420ec6ac0941c2612a5af51136c96
Adrian Szyndela [Tue, 13 Dec 2016 13:43:06 +0000 (14:43 +0100)]
crash-stack: work around hanging up on 'D' processes
Processes in uninterruptible sleep state ('D' status in ps)
cannot be ptraced until they leave the state. To prevent
hanging up of crash-stack, we check the target process state.
If it is 'D', then we give up ptracing it, and instead
print the information from /proc/<pid>/wchan, /proc/<pid>/syscall
and /proc/<pid>/stack.
Change-Id: I82644051a5499c42db2a8e51a7af98693e8da363
Karol Lewandowski [Wed, 14 Dec 2016 16:53:05 +0000 (17:53 +0100)]
crash-stack: Drop duplicate call stack footer
Change-Id: I938f2ead95d312450edb853740cce47becf796fc
Łukasz Stelmach [Wed, 14 Dec 2016 16:30:54 +0000 (17:30 +0100)]
crash-stack: print memory information
Print information from /proc/meminfo and /proc/PID/status
Change-Id: I6c66e9ee6c8eca20a8925e9ecc23e09c0dabcb25