Compare commits

..
14 Commits
Author SHA1 Message Date
Patrick McCarty 913cbb1c1e Release v1.14.0
This release adds three new hardware metadata fields to include in
records created by libtelemetry: board_name, cpu_model, and
bios_version.

Signed-off-by: Patrick McCarty <patrick.mccarty@intel.com>
2017-09-08 07:32:13 -07:00
Patrick McCarty d3ce2f435c Run uncrustify on the source
Signed-off-by: Patrick McCarty <patrick.mccarty@intel.com>
2017-09-08 07:28:42 -07:00
avjarami d0a3e6bafa Including additional host metadata in headers
Additional headers added: board_name, cpu_model, and bios_version.

    * Board name is a combination of board_name and board_vendor from
    dmi file system.

    * CPU model is read from /proc/cpuinfo.

    * BIOS version is taken from dmi file system.
2017-09-08 07:26:01 -07:00
Patrick McCarty a9b16868f5 Release v1.13.0
This release introduces some changes in behavior for the crash probe and
journal probe:

* The crash probe now ensures core files are kept whenever backtraces
  are scrubbed from telemetry records, or when errors occur during runtime
  operation. This improves the developer experience when running ones own
  programs installed at arbitrary locations on the filesystem and also
  avoids data loss; either the crash backtrace is sent via telemetry, or
  when unable to send the backtrace, the core file is kept.

* The journal probe, by default, now only sends telemetry for journal
  log messages indicating systemd service failures. The previous behavior
  of sending any log messages above a certain log level is now wrapped
  behind the "opt-in-no-privacy-filters" config setting.

Bug fixes include:

* Fixing a segfault crash in the configuration code that occurred
  whenever an incomplete configuration file was encountered.

* Fixing a performance issue in 'telem-record-gen' to avoid extraneous
  strlen() calls in the classification input validation code.

Signed-off-by: Patrick McCarty <patrick.mccarty@intel.com>
2017-07-05 12:02:21 -07:00
Patrick McCarty cbbc579917 journal probe: conditionally enable log level filters
Because log messages from services contain arbitrary data, and sometimes
this data is privacy sensitive, only enable the log level filters when
the privacy filter override is in effect.

Signed-off-by: Patrick McCarty <patrick.mccarty@intel.com>
2017-06-30 14:33:31 -07:00
Patrick McCarty 7080839167 configuration: make config errors more actionable
The current behavior of the configuration code is to print a generic
error message and exit if configuration parsing fails.

To clarify the reason for exiting, and thus make the error message more
actionable, print an additional error message that describes which key
from the config is affected and the type of value it accepts.

Signed-off-by: Patrick McCarty <patrick.mccarty@intel.com>
2017-06-30 14:31:53 -07:00
Patrick McCarty 0d07e76955 configuration: fix regression in parsing that results in segfault
When a key from a INI file is missing, libnica returns NULL, so we need
to properly check for a non-NULL return value before passing it to
strdup().

Signed-off-by: Patrick McCarty <patrick.mccarty@intel.com>
2017-06-30 14:31:53 -07:00
Patrick McCarty 1c8af05cb9 telem-record-gen: avoid unnecessary strlen() calls
Since strlen(opt_class) is already called before this for loop, there is
no need to call it again. It also avoids repeatedly calling strlen() as
part of the for-loop conditional check.

Signed-off-by: Patrick McCarty <patrick.mccarty@intel.com>
2017-06-27 21:02:44 -07:00
Patrick McCarty cb4b951a1a crash probe: use a helper function for string prefix checks
Since the code repeats the same logic in several places, it is
convenient to split the routine into a helper function, startswith().

This also fixes a bug with the earlier port to libnica: in the
in_clr_build() function, strstr() was chosen as the replacement function
instead of strncmp(), which means a match will be found in any part of
the string. However, a match should only be found when it is a prefix.
To fix, switch the strstr() in that function to use startswith() as
well.

Signed-off-by: Patrick McCarty <patrick.mccarty@intel.com>
2017-06-27 21:02:44 -07:00
Patrick McCarty bb018a158f Document condition under which core files are removed
Signed-off-by: Patrick McCarty <patrick.mccarty@intel.com>
2017-06-26 17:14:50 -07:00
Patrick McCarty 577484cdb2 crash probe: also keep core files when errors occur
During the processing of a core file or sending telemetry, any numbers
of errors may occur. Make sure the core file is not unlinked under these
conditions.

Signed-off-by: Patrick McCarty <patrick.mccarty@intel.com>
2017-06-26 17:14:50 -07:00
Patrick McCarty 049c9a35b9 crash probe: keep core files when backtraces are scrubbed
When the path filters for privacy are in effect, backtraces are scrubbed
from records, and this results in the core files being unlinked.

However, this is not friendly behavior for the developer. A common
situation that triggers the path filters is installing custom binaries
on the system (say, under /usr/local/bin or /opt/bin) for testing
purposes. To better enable developers to debug their programs, having
the core files available to process with gdb is very valuable.

Eventually, I would like to add an opt-in to keep all core files, but
I'll wait until the configuration code is refactored.

Signed-off-by: Patrick McCarty <patrick.mccarty@intel.com>
2017-06-26 17:14:50 -07:00
Patrick McCarty 56a6a4777b Release v1.12.4
This release includes a change that results in lower memory consumption
for telemd when the service is idle.

Signed-off-by: Patrick McCarty <patrick.mccarty@intel.com>
2017-06-23 11:16:07 -07:00
Patrick McCarty d2ecf76c70 Init/deinit libcurl global environment for each POST
When the daemon is sitting idle, we are seeing libcurl consume around
2MB of memory, which had been previously allocated on-the-fly for its
global environment.

To have more control over this memory consumption, making sure the
daemon uses as little memory as possible when doing no work, explicitly
allocate the libcurl global environment before each POST and deallocate
it afterwards.

Signed-off-by: Patrick McCarty <patrick.mccarty@intel.com>
2017-06-20 19:33:06 -07:00
10 changed files with 278 additions and 56 deletions
+1 -1
View File
@@ -2,7 +2,7 @@
# Process this file with autoconf to produce a configure script.
AC_PREREQ([2.69])
AC_INIT([telemetrics-client], [1.12.3], [https://clearlinux.org/])
AC_INIT([telemetrics-client], [1.14.0], [https://clearlinux.org/])
AC_CONFIG_AUX_DIR([build-aux])
AM_INIT_AUTOMAKE([1.14 -Wall -Werror -Wno-extra-portability foreign subdir-objects])
AM_SILENT_RULES([yes])
+4 -1
View File
@@ -33,7 +33,10 @@ static const char *header_names[] = {
TM_SYSTEM_BUILD_STR,
TM_KERNEL_VERSION_STR,
TM_PAYLOAD_VERSION_STR,
TM_SYSTEM_NAME_STR
TM_SYSTEM_NAME_STR,
TM_BOARD_NAME_STR,
TM_CPU_MODEL_STR,
TM_BIOS_VERSION_STR
};
const char *get_header_name(int ind)
+7 -1
View File
@@ -35,6 +35,9 @@
#define TM_KERNEL_VERSION 8
#define TM_PAYLOAD_VERSION 9
#define TM_SYSTEM_NAME 10
#define TM_BOARD_NAME 11
#define TM_CPU_MODEL 12
#define TM_BIOS_VERSION 13
#define TM_RECORD_VERSION_STR "record_format_version"
#define TM_CLASSIFICATION_STR "classification"
@@ -47,8 +50,11 @@
#define TM_KERNEL_VERSION_STR "kernel_version"
#define TM_PAYLOAD_VERSION_STR "payload_format_version"
#define TM_SYSTEM_NAME_STR "system_name"
#define TM_BOARD_NAME_STR "board_name"
#define TM_CPU_MODEL_STR "cpu_model"
#define TM_BIOS_VERSION_STR "bios_version"
#define NUM_HEADERS 11
#define NUM_HEADERS 14
/* For internal library usage. Bump the version whenever we change the record
* structure (e.g. adding or removing a header field). Note that the value
+11 -2
View File
@@ -97,8 +97,15 @@ bool read_config_from_file(char *config_file, struct configuration *config)
return false;
} else {
for (int i = CONF_STR_MIN + 1; i < CONF_STR_MAX; i++) {
config->strValues[i] = strdup(nc_hashmap_get(nc_hashmap_get(keyfile, "settings"), config_key_str[i]));
if (config->strValues[i] == NULL) {
char *ptr;
ptr = nc_hashmap_get(nc_hashmap_get(keyfile, "settings"), config_key_str[i]);
if (ptr) {
config->strValues[i] = strdup(ptr);
if (config->strValues[i] == NULL) {
return false;
}
} else {
fprintf(stderr, "ERR: missing key with string value: %s\n", config_key_str[i]);
return false;
}
}
@@ -108,6 +115,7 @@ bool read_config_from_file(char *config_file, struct configuration *config)
if (ptr) {
config->intValues[i] = strtoll(ptr, NULL, 10);
} else {
fprintf(stderr, "ERR: missing key with integer value: %s\n", config_key_int[i]);
return false;
}
}
@@ -125,6 +133,7 @@ bool read_config_from_file(char *config_file, struct configuration *config)
config->boolValues[i] = true;
}
} else {
fprintf(stderr, "ERR: missing key with boolean value: %s\n", config_key_bool[i]);
return false;
}
}
+59 -38
View File
@@ -68,6 +68,9 @@ static char clr_build_class[40] = "org.clearlinux/crash/clr-build";
static char error_class[30] = "org.clearlinux/crash/error";
static char unknown_class[30] = "org.clearlinux/crash/unknown";
static char temp_core[] = "/tmp/corefile-XXXXXX";
static bool keep_core = false;
static const Dwfl_Callbacks cb =
{
.find_elf = dwfl_build_id_find_elf,
@@ -147,23 +150,14 @@ static int temp_core_file(void)
int tmp;
ssize_t ret;
char core[PATH_MAX] = "/tmp/corefile-XXXXXX";
/* mkstemp() opens the file with O_EXCL and 0600 permissions, so no need
* to change umask or manipulate the fd to meet those requirements.
*/
if ((tmp = mkstemp(core)) < 0) {
if ((tmp = mkstemp(temp_core)) < 0) {
telem_perror("Failed to create temp core file");
return -1;
}
#ifndef DEBUG
if (unlink(core) < 0) {
telem_perror("Failed to unlink temp core file");
return -1;
}
#endif
while (true) {
// Use Linux-specific splice(2) here;
// simplifies copying data from pipe->file
@@ -416,6 +410,18 @@ fail:
return -1;
}
static bool startswith(const char *full, const char *prefix)
{
while (*prefix) {
if (*prefix != *full) {
return false;
}
full++;
prefix++;
}
return true;
}
static bool in_clr_build(char *fullpath)
{
// Global override for privacy filters
@@ -427,8 +433,8 @@ static bool in_clr_build(char *fullpath)
* The build environment for Clear Linux packages is set up by 'mock',
* and the chroot in which rpmbuild builds the packages has this prefix.
*/
if ((strstr(fullpath, "/builddir/build/BUILD/")) ||
(strstr(fullpath, "!builddir!build!BUILD!"))) {
if (startswith(fullpath, "/builddir/build/BUILD/") ||
startswith(fullpath, "!builddir!build!BUILD!")) {
return true;
}
@@ -444,13 +450,13 @@ static bool is_banned_path(char *fullpath)
// Anything outside of /usr/, or in /usr/local/, we consider third-party
if ((strncmp(fullpath, "/usr/", 5) != 0) &&
(strncmp(fullpath, "!usr!", 5) != 0)) {
if (!startswith(fullpath, "/usr/") &&
!startswith(fullpath, "!usr!")) {
return true;
}
if ((strncmp(fullpath, "/usr/local/", 11) == 0) ||
(strncmp(fullpath, "!usr!local!", 11) == 0)) {
if (startswith(fullpath, "/usr/local/") ||
startswith(fullpath, "!usr!local!")) {
return true;
}
@@ -563,28 +569,6 @@ int main(int argc, char **argv)
exit(EXIT_FAILURE);
}
if (proc_path && in_clr_build(proc_path)) {
telem_log(LOG_NOTICE, "Ignoring core (from mock build)\n");
backtrace = nc_string_dup("Crash from Clear package build\n");
if (!send_data(&backtrace, unknown_severity, clr_build_class)) {
goto fail;
}
goto success;
}
if (proc_path && is_banned_path(proc_path)) {
telem_log(LOG_NOTICE, "Ignoring core (third-party binary)\n");
backtrace = nc_string_dup("Crash from third party\n");
if (!send_data(&backtrace, unknown_severity, unknown_class)) {
goto fail;
}
goto success;
}
if (core_file) {
core_fd = open(core_file, O_RDONLY);
if (core_fd == -1) {
@@ -617,6 +601,32 @@ int main(int argc, char **argv)
}
}
if (proc_path && in_clr_build(proc_path)) {
telem_log(LOG_NOTICE, "Ignoring core (from mock build)\n");
backtrace = nc_string_dup("Crash from Clear package build\n");
keep_core = true;
if (!send_data(&backtrace, unknown_severity, clr_build_class)) {
goto fail;
}
goto success;
}
if (proc_path && is_banned_path(proc_path)) {
telem_log(LOG_NOTICE, "Ignoring core (third-party binary)\n");
backtrace = nc_string_dup("Crash from third party\n");
keep_core = true;
if (!send_data(&backtrace, unknown_severity, unknown_class)) {
goto fail;
}
goto success;
}
elf_version(EV_CURRENT);
if (prepare_corefile(&e_core, core_fd) < 0) {
@@ -667,6 +677,11 @@ success:
ret = EXIT_SUCCESS;
fail:
// Do not remove the core file if any errors occur
if (ret == EXIT_FAILURE) {
keep_core = true;
}
free(core_file);
free(proc_name);
free(proc_path);
@@ -695,6 +710,12 @@ fail:
close(core_fd);
}
// Remove the core file by default, except when the --core-file option
// is specified, or when keep_core is overridden to true.
if (!core_file && !keep_core) {
unlink(temp_core);
}
return ret;
}
+22 -8
View File
@@ -38,6 +38,7 @@
#include "config.h"
#include "log.h"
#include "telemetry.h"
#include "probe.h"
#include "nica/nc-string.h"
#define BOOT_ID_LEN 33
@@ -227,9 +228,14 @@ static bool add_filters(sd_journal *journal)
/* The semantics of how journal entry matching works is described in
* detail in sd_journal_add_match(3).
*
* The matches declared here correspond to the logical expression:
* When the privacy filter override is enabled, the matches declared
* here correspond to this logical expression:
*
* BOOTID && ((P0 || P1 || P2 || P3) || EXITED)
* BOOTID && ((P0 || P1 || P2 || P3) || EXITED)
*
* Otherwise, the expression is:
*
* BOOTID && EXITED
*
* BOOTID is short for _BOOT_ID=VAL, where VAL is the boot ID for the
* current boot. P0, P1, etc stand for PRIORITY=0, etc. And EXITED is
@@ -239,12 +245,20 @@ static bool add_filters(sd_journal *journal)
JOURNAL_MATCH(data);
free(data);
JOURNAL_AND;
// The four highest log levels, all indicating errors
JOURNAL_MATCH("PRIORITY=0");
JOURNAL_MATCH("PRIORITY=1");
JOURNAL_MATCH("PRIORITY=2");
JOURNAL_MATCH("PRIORITY=3");
JOURNAL_OR;
// Filter messages with the four highest log levels, all indicating
// errors, but only when the privacy filters override is in effect;
// because the log messages contain arbitrary strings, and this probe
// does not yet keep a whitelist of allowed patterns or a blacklist of
// banned patterns.
if (access(TM_PRIVACY_FILTERS_OVERRIDE, F_OK) == 0) {
JOURNAL_MATCH("PRIORITY=0");
JOURNAL_MATCH("PRIORITY=1");
JOURNAL_MATCH("PRIORITY=2");
JOURNAL_MATCH("PRIORITY=3");
JOURNAL_OR;
}
// Only set for service-level error conditions
JOURNAL_MATCH("EXIT_CODE=exited");
+4 -4
View File
@@ -150,7 +150,7 @@ fail:
int validate_opts(void)
{
size_t i;
size_t len;
int ret = 0;
/* classification */
@@ -159,15 +159,15 @@ int validate_opts(void)
return ret;
}
i = strlen(opt_class);
len = strlen(opt_class);
if ((i == 0) || (i > 120)) {
if ((len == 0) || (len > 120)) {
fprintf(stderr, "Error: Valid size for classification "
"is 1-120 chars\n");
return ret;
}
for (int c = 0; c < strlen(opt_class); c++) {
for (int c = 0; c < len; c++) {
if (isascii(opt_class[c]) == 0) {
fprintf(stderr, "Error: Non-ascii characters detected "
"in classification - aborting\n");
+8
View File
@@ -462,6 +462,11 @@ bool post_record_http(char *headers[], char *body, bool spool)
const char *cert_file = get_cainfo_config();
const char *tid_header = get_tidheader_config();
// Initialize the libcurl global environment once per POST. This lets us
// clean up the environment after each POST so that when the daemon is
// sitting idle, it will be consuming as little memory as possible.
curl_global_init(CURL_GLOBAL_ALL);
curl = curl_easy_init();
if (!curl) {
telem_log(LOG_ERR, "curl_easy_init(): Unable to start libcurl"
@@ -532,6 +537,9 @@ bool post_record_http(char *headers[], char *body, bool spool)
curl_slist_free_all(custom_headers);
curl_easy_cleanup(curl);
curl_global_cleanup();
return res ? false : true;
}
+158
View File
@@ -400,6 +400,62 @@ static int set_timestamp_header(struct telem_ref *t_ref)
return status;
}
/**
* Sets cpu model for telemetry record. The information from cpu is extracted
* from /proc/cpuinfo, specifically "model name" attribute.
*
* @param t_ref Telemetry Record reference obtained from tm_create_record.
*
* @return 0 if successful, or a negative errno-style value if not.
*
*/
static int set_cpu_model_header(struct telem_ref *t_ref)
{
FILE *fs = NULL;
char buf[SMALL_LINE_BUF] = { 0 };
char *model_name = NULL;
const char *attr_name = "model name";
int status = 0;
size_t attr_len = strlen(attr_name);
size_t model_str_len = 0;
fs = fopen("/proc/cpuinfo", "r");
if (fs != NULL) {
while (fgets(buf, SMALL_LINE_BUF, fs)) {
if (strncmp(attr_name, buf, attr_len) == 0) {
model_name = strchr(buf, ':');
break;
}
}
fclose(fs);
if (model_name != NULL) {
model_str_len = strlen(model_name);
if (model_str_len > 2) {
model_name = (char *)model_name + 2 * sizeof(char);
model_name[model_str_len - 2] = '\0';
} else {
model_name = "blank";
}
} else {
model_name = "blank";
fprintf(stderr, "NOTICE: Unable to find attribute:%s\n", attr_name);
}
status = set_header(
&(t_ref->record->headers[TM_CPU_MODEL]),
TM_CPU_MODEL_STR, model_name,
&(t_ref->record->header_size));
} else {
#ifdef DEBUG
fprint(stderr, "NOTICE: Unable to open /proc/cpuinfo\n");
#endif
status = -1;
}
return status;
}
/**
* A healper function for set_host_type_header that reads the first line out of
* a file and chomps the newline if necessary. If the file does not exist the
@@ -497,6 +553,90 @@ static int get_dmi_value(const char *source, const char *key, char **buf)
return ret;
}
/**
* Sets the board name for telemetry record, this record is a combination of
* board name and board vendor. This information is read from dmi filesystem
* Board Name (board_name) and Board Vendor (board_vendor).
*
* @param t_ref Telemetry Record reference obtained from tm_create_record.
*
* @return 0 if successful, or a negative errno-style value if not.
*
*/
static int set_board_name_header(struct telem_ref *t_ref)
{
int status = 0;
int rc;
char *buf = NULL;
char *bn = NULL;
char *bv = NULL;
rc = get_dmi_value("/sys/class/dmi/id/board_name", "bn", &bn);
if (rc < 0) {
status = rc;
goto cleanup;
}
rc = get_dmi_value("/sys/class/dmi/id/board_vendor", "bv", &bv);
if (rc < 0) {
status = rc;
goto cleanup;
}
rc = asprintf(&buf, "%s|%s", bn, bv);
if (rc < 0) {
status = -ENOMEM;
goto cleanup;
} else {
status = set_header(
&(t_ref->record->headers[TM_BOARD_NAME]),
TM_BOARD_NAME_STR, buf,
&(t_ref->record->header_size));
free(buf);
}
cleanup:
if (bn != NULL) {
free(bn);
}
if (bv != NULL) {
free(bv);
}
return status;
}
/**
* Sets BIOS version header for telemetry record, this information is
* read from dmi filesystem BIOS Version (bios_version).
*
* @param t_ref Telemetry Record reference obtained from tm_create_record.
*
* @return 0 if successful, or a negative errno-style value if not.
*
*/
static int set_bios_version_header(struct telem_ref *t_ref)
{
int status = 0;
int rc = 0;
char *bios_version = NULL;
rc = get_dmi_value("/sys/class/dmi/id/bios_version", "bv", &bios_version);
if (rc < 0) {
status = rc;
} else {
status = set_header(
&(t_ref->record->headers[TM_BIOS_VERSION]),
TM_BIOS_VERSION_STR, bios_version,
&(t_ref->record->header_size));
free(bios_version);
}
return status;
}
/**
* Sets the hosttype header, which is a tuple of three values looked for
* in the dmi filesystem. System Vendor (sys_vendor), Product Name
@@ -730,6 +870,24 @@ int allocate_header(struct telem_ref *t_ref, uint32_t severity,
goto free_and_fail;
}
i++;
if ((ret = set_board_name_header(t_ref)) < 0) {
goto free_and_fail;
}
i++;
if ((ret = set_cpu_model_header(t_ref)) < 0) {
goto free_and_fail;
}
i++;
if ((ret = set_bios_version_header(t_ref)) < 0) {
goto free_and_fail;
}
i++; /* Not necessary, but including for future expansion */
return ret;
+4 -1
View File
@@ -295,7 +295,10 @@ START_TEST(check_process_record_with_correct_size_and_data)
"machine_id: 1234\ncreation_timestamp: 1418672344\narch:x86_64\n"
"host_type: macbookpro\nbuild: 200\nkernel_version: 3.15\n"
"payload_format_version: 1\n"
"system_name: clear-linux-os\n";
"system_name: clear-linux-os\n"
"board_name: Qemu|Intel\n"
"cpu_model: Intel(R) Core(TM) i7-5650U CPU @ 2.20GHz\n"
"bios_version: Qemu\n";
char *post_body = "test message";
set_up_socket_pair(&client_fd, &server_fd);