Skip to content

Add support for detecting docker container #2047

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
charliegracie merged 1 commit into from
Feb 27, 2018
Merged

Add support for detecting docker container #2047

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion port/common/omrport.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -261,7 +261,7 @@ static OMRPortLibrary MasterPortLibraryTable = {
omrsysinfo_cgroup_get_enabled_subsystems, /* sysinfo_cgroup_get_enabled_subsystems */
omrsysinfo_cgroup_enable_subsystems, /* sysinfo_cgroup_enable_subsystems */
omrsysinfo_cgroup_are_subsystems_enabled, /* sysinfo_cgroup_are_subsystems_enabled */
omrsysinfo_cgroup_get_memlimit, /* sysinfo_cgroup_get_memlimit */
omrsysinfo_cgroup_get_memlimit, /* sysinfo_cgroup_get_memlimit */
omrport_init_library, /* port_init_library */
omrport_startup_library, /* port_startup_library */
omrport_create_library, /* port_create_library */
Expand Down
1 change: 1 addition & 0 deletions port/common/omrsysinfo.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -902,6 +902,7 @@ omrsysinfo_cgroup_are_subsystems_enabled(struct OMRPortLibrary *portLibrary, uin
* omrsysinfo_cgroup_enable_limits() before calling this function.
* When the fuction returns OMRPORT_ERROR_SYSINFO_CGROUP_UNSUPPORTED_PLATFORM,
* value of *limits is unspecified.
* Note that 'limit' parameter must not be NULL.
*
* @param[in] portLibrary pointer to OMRPortLibrary
* @param[out] limit pointer to uint64_t which successful return contains memory limit imposed by cgroup
Expand Down
176 changes: 134 additions & 42 deletions port/unix/omrsysinfo.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -256,8 +256,17 @@ struct {
#if defined(LINUX)

#define OMR_CGROUP_V1_MOUNT_POINT "/sys/fs/cgroup"
#define ROOT_CGROUP "/"

/* Currently 12 subsystems/resource controllers are defined.
/* An entry in /proc/<pid>/cgroup is of following form:
* <hierarchy ID>:<subsystem>[,<subsystem>]*:<cgroup name>
*
* An example:
* 7:cpuacct,cpu:/mycgroup
*/
#define PROC_PID_CGROUP_ENTRY_FORMAT "%d:%[^:]:%s"

/* Currently 12 subsystems or resource controllers are defined.
*/
typedef enum OMRCgroupSubsystem {
INVALID_SUBSYSTEM = -1,
Expand Down Expand Up @@ -331,12 +340,14 @@ static uint16_t getPhysicalMemory();
#endif /* defined(OMRZTPF) */

#if defined(LINUX) && !defined(OMRZTPF)
static BOOLEAN isCgroupV1Available(struct OMRPortLibrary *portLibrary);
static void freeCgroupEntries(struct OMRPortLibrary *portLibrary, OMRCgroupEntry *cgEntryList);
static char * getCgroupNameForSubsystem(struct OMRPortLibrary *portLibrary, OMRCgroupEntry *cgEntryList, const char *subsystem);
static int32_t addCgroupEntry(struct OMRPortLibrary *portLibrary, OMRCgroupEntry **cgEntryList, int32_t hierId, const char *subsystem, const char *cgroupName);
static int32_t readCgroupFile(struct OMRPortLibrary *portLibrary, int pid, OMRCgroupEntry **cgroupEntryList, uint64_t *availableSubsystems);
static int32_t readCgroupFile(struct OMRPortLibrary *portLibrary, int pid, BOOLEAN inContainer, OMRCgroupEntry **cgroupEntryList, uint64_t *availableSubsystems);
static OMRCgroupSubsystem getCgroupSubsystemFromFlag(uint64_t subsystemFlag);
static int32_t readCgroupSubsystemFile(struct OMRPortLibrary *portLibrary, uint64_t subsystemFlag, const char *fileName, int32_t numItemsToRead, const char *format, ...);
static int32_t isRunningInContainer(struct OMRPortLibrary *portLibrary, BOOLEAN *inContainer);
#endif /* defined(LINUX) */


Expand Down Expand Up @@ -1701,7 +1712,7 @@ omrsysinfo_startup(struct OMRPortLibrary *portLibrary)
#if defined(LINUX) && !defined(OMRZTPF)
PPG_cgroupEntryList = NULL;
/* To handle the case where multiple port libraries are started and shutdown,
* as done by some fvtests (eg fvtest/porttest/j9portTest.cpp) that create fake portlibrary
* as done by some fvtests (eg fvtest/porttest/j9portTest.cpp) that create fake portlibrary
* to test its management and lifecycle,
* we need to ensure globals like cgroupEntryListMonitor are initialized and destroyed only once.
*/
Expand Down Expand Up @@ -3314,6 +3325,34 @@ omrsysinfo_os_kernel_info(struct OMRPortLibrary *portLibrary, struct OMROSKernel

#if defined(LINUX)

/**
* @internal
* Checks if cgroup v1 system is available
*
* @param[in] portLibrary pointer to OMRPortLibrary
*
* @return TRUE if cgroup v1 system is available, FALSE otherwise
*/
static BOOLEAN
isCgroupV1Available(struct OMRPortLibrary *portLibrary)
{
struct statfs buf = {0};
int32_t rc = 0;
BOOLEAN result = TRUE;

/* If tmpfs is mounted on /sys/fs/cgroup, then it indicates cgroup v1 system is available */
rc = statfs(OMR_CGROUP_V1_MOUNT_POINT, &buf);
if (0 != rc) {
portLibrary->error_set_last_error(portLibrary, errno, OMRPORT_ERROR_SYSINFO_SYS_FS_CGROUP_STATFS_FAILED);
result = FALSE;
} else if (TMPFS_MAGIC != buf.f_type) {
portLibrary->error_set_last_error_with_message_format(portLibrary, OMRPORT_ERROR_SYSINFO_SYS_FS_CGROUP_TMPFS_NOT_MOUNTED, "tmpfs is not mounted on " OMR_CGROUP_V1_MOUNT_POINT);
result = FALSE;
}

return result;
}

/**
* @internal
* Free resources allocated for OMRCgroupEntry
Expand Down Expand Up @@ -3420,21 +3459,20 @@ addCgroupEntry(struct OMRPortLibrary *portLibrary, OMRCgroupEntry **cgEntryList,
*
* @param[in] portLibrary pointer to OMRPortLibrary
* @param[in] pid process id
* @param[in] inContainer if set to TRUE then ignore cgroup in /proc/<pid>/cgroup and use ROOT_CGROUP instead
* @param[out] cgroupEntryList pointer to OMRCgroupEntry *. On successful return, *cgroupEntry
* points to a circular linked list. Each element of the list is populated based on the contents
* of /proc/<pid>/cgroup file.
* @param[out] availableSubsystems on successful return, contains bitwise-OR of flags of type OMR_CGROUP_SUBSYSTEMS_*
* @param[out] availableSubsystems on successful return, contains bitwise-OR of flags of type OMR_CGROUP_SUBSYSTEMS_*
* indicating the subsystems available for use
*
* returns 0 on success, negative code on error
*/
static int32_t
readCgroupFile(struct OMRPortLibrary *portLibrary, int pid, OMRCgroupEntry **cgroupEntryList, uint64_t *availableSubsystems)
readCgroupFile(struct OMRPortLibrary *portLibrary, int pid, BOOLEAN inContainer, OMRCgroupEntry **cgroupEntryList, uint64_t *availableSubsystems)
{
char cgroup[PATH_MAX];
char cgroupFilePath[PATH_MAX];
uintptr_t requiredSize = 0;
/* This array should be large enough to read names of all subsystems. 1024 should be enough. */
char subsystems[1024];
FILE *cgroupFile = NULL;
OMRCgroupEntry *cgEntryList = NULL;
uint64_t available = 0;
Expand All @@ -3444,34 +3482,29 @@ readCgroupFile(struct OMRPortLibrary *portLibrary, int pid, OMRCgroupEntry **cgr

requiredSize = portLibrary->str_printf(portLibrary, NULL, (uint32_t)-1, "/proc/%d/cgroup", pid);
Assert_PRT_true(requiredSize <= PATH_MAX);
portLibrary->str_printf(portLibrary, cgroup, sizeof(cgroup), "/proc/%d/cgroup", pid);
cgroupFile = fopen(cgroup, "r");
portLibrary->str_printf(portLibrary, cgroupFilePath, sizeof(cgroupFilePath), "/proc/%d/cgroup", pid);

/* Even if 'inContainer' is TRUE, we need to parse the cgroup file to get the list of subsystems */
cgroupFile = fopen(cgroupFilePath, "r");
if (NULL == cgroupFile) {
rc = portLibrary->error_set_last_error(portLibrary, errno, OMRPORT_ERROR_SYSINFO_PROCESS_CGROUP_FILE_FOPEN_FAILED);
goto _end;
}

while (0 == feof(cgroupFile)) {
char cgroup[PATH_MAX];
/* This array should be large enough to read names of all subsystems. 1024 should be enough based on current supported subsystems. */
char subsystems[1024];
char *cursor = NULL;
char *separator = NULL;
int32_t hierId = -1;

/* Following is the description of /proc/<pid>/cgroup copied from 'man' page for proc:
*
* Each entry in /proc/<pid>/cgroup is of type:
* 5:cpuacct,cpu,cpuset:/daemons
* The colon-separated fields are, from left to right:
* 1. hierarchy ID number
* 2. set of subsystems bound to the hierarchy
* 3. control group in the hierarchy to which the process belongs
*/
rc = fscanf(cgroupFile, "%d:%[^:]:%s", &hierId, subsystems, cgroup);
rc = fscanf(cgroupFile, PROC_PID_CGROUP_ENTRY_FORMAT, &hierId, subsystems, cgroup);
/* Ensure we didn't overflow */
Assert_PRT_true(strlen(subsystems) < 1024);
Assert_PRT_true(strlen(cgroup) < PATH_MAX);

if (EOF == rc) {
rc = 0;
break;
} else if (3 != rc) {
rc = portLibrary->error_set_last_error_with_message_format(portLibrary, OMRPORT_ERROR_SYSINFO_PROCESS_CGROUP_FILE_READ_FAILED, "unexpcted format of /proc/%d/cgroup file", pid);
Expand All @@ -3490,7 +3523,12 @@ readCgroupFile(struct OMRPortLibrary *portLibrary, int pid, OMRCgroupEntry **cgr
if (OMR_ARE_NO_BITS_SET(available, supportedSubsystems[i].flag)
&& !strcmp(cursor, supportedSubsystems[i].name)
) {
rc = addCgroupEntry(portLibrary, &cgEntryList, hierId, cursor, cgroup);
const char *cgroupToUse = cgroup;

if (TRUE == inContainer) {
cgroupToUse = ROOT_CGROUP;
}
rc = addCgroupEntry(portLibrary, &cgEntryList, hierId, cursor, cgroupToUse);
if (0 != rc) {
goto _end;
}
Expand Down Expand Up @@ -3617,6 +3655,63 @@ readCgroupSubsystemFile(struct OMRPortLibrary *portLibrary, uint64_t subsystemFl
return rc;
}

/**
* Checks if the process is running inside container
*
* @param[in] portLibrary pointer to OMRPortLibrary
* @param[out] inContainer pointer to BOOLEAN which on successful return indicates if the process is running in container or not
*
* @return 0 on success, otherwise negative error code
*/
static int32_t
isRunningInContainer(struct OMRPortLibrary *portLibrary, BOOLEAN *inContainer)
{
int32_t rc = 0;

/* Assume we are not in container */
*inContainer = FALSE;

if (isCgroupV1Available(portLibrary)) {
/* Read PID 1's cgroup file /proc/1/cgroup and check cgroup name for each subsystem.
* If cgroup name for each subsystem points to the root cgroup "/",
* then the process is not running in a container.
* For any other cgroup name, assume we are in a container.
*/
FILE *cgroupFile = fopen("/proc/1/cgroup", "r");

if (NULL == cgroupFile) {
rc = portLibrary->error_set_last_error(portLibrary, errno, OMRPORT_ERROR_SYSINFO_PROCESS_CGROUP_FILE_FOPEN_FAILED);
goto _end;
}

while (0 == feof(cgroupFile)) {
char cgroup[PATH_MAX];
char subsystems[1024];
int32_t hierId = -1;

rc = fscanf(cgroupFile, PROC_PID_CGROUP_ENTRY_FORMAT, &hierId, subsystems, cgroup);
/* Ensure we didn't overflow */
Assert_PRT_true(strlen(subsystems) < 1024);
Assert_PRT_true(strlen(cgroup) < PATH_MAX);

if (EOF == rc) {
break;
} else if (3 != rc) {
rc = portLibrary->error_set_last_error_with_message_format(portLibrary, OMRPORT_ERROR_SYSINFO_PROCESS_CGROUP_FILE_READ_FAILED, "unexpected format of /proc/1/cgroup file");
goto _end;
Copy link
Contributor

@DanHeidinga DanHeidinga Feb 20, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@return 0 on success, otherwise negative error code

This may exit the method with a positive, non-zero return value contrary to the function contract.

Copy link
Contributor Author

@ashu-mehra ashu-mehra Feb 20, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

rc is reassigned return value of error_set_last_error_with_message_format which is negative error code.

Copy link
Contributor

@DanHeidinga DanHeidinga Feb 20, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I missed that - sorry.

}

if (0 != strcmp(ROOT_CGROUP, cgroup)) {
*inContainer = TRUE;
break;
}
}
rc = 0;
}
_end:
return rc;
}

#endif /* defined(LINUX) */

BOOLEAN
Expand All @@ -3627,25 +3722,21 @@ omrsysinfo_cgroup_is_system_available(struct OMRPortLibrary *portLibrary)
int32_t rc = OMRPORT_ERROR_SYSINFO_CGROUP_UNSUPPORTED_PLATFORM;

if (NULL == PPG_cgroupEntryList) {
struct statfs buf = {0};
if (isCgroupV1Available(portLibrary)) {
BOOLEAN inContainer = FALSE;

/* If tmpfs is mounted on /sys/fs/cgroup, then it indicates cgroup v1 system is available */
rc = statfs(OMR_CGROUP_V1_MOUNT_POINT, &buf);
if (0 != rc) {
rc = portLibrary->error_set_last_error(portLibrary, errno, OMRPORT_ERROR_SYSINFO_SYS_FS_CGROUP_STATFS_FAILED);
goto _end;
} else if (TMPFS_MAGIC != buf.f_type) {
rc = portLibrary->error_set_last_error_with_message_format(portLibrary, OMRPORT_ERROR_SYSINFO_SYS_FS_CGROUP_TMPFS_NOT_MOUNTED, "tmpfs is not mounted on " OMR_CGROUP_V1_MOUNT_POINT);
goto _end;
}

omrthread_monitor_enter(cgroupEntryListMonitor);
if (NULL == PPG_cgroupEntryList) {
rc = readCgroupFile(portLibrary, getpid(), &PPG_cgroupEntryList, &PPG_cgroupSubsystemsAvailable);
}
omrthread_monitor_exit(cgroupEntryListMonitor);
if (0 != rc) {
goto _end;
rc = isRunningInContainer(portLibrary, &inContainer);
Copy link
Contributor

@DanHeidinga DanHeidinga Feb 20, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

isRunningInContainer immediately calls isCgroupV1Available so this ends up doing the isCgroupV1Available work twice.

Can this be commoned into a single call? I hesitate to suggest a isCgroupV1AndInContainer(portLibrary, &isCgroupV1, &inContainer) call as we should be able to come up with a better solution.

Copy link
Contributor Author

@ashu-mehra ashu-mehra Feb 21, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I noticed that but didn't bother as isCgroupV1Available just adds an overhead of one syscall.
But if we want to common it out, how about calling the new function checkCgroupPrereqs(&cgroupVersion, &inContainer) where cgroupVersion would be an enum { V1, V2 } (currently the code only supports V1, but sooner than later we would have to consider V2 as well).

Copy link
Contributor

@DanHeidinga DanHeidinga Feb 22, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Since we'll have to touch this code again to add V2 support, I'm OK with leaving it as is now.

if (0 != rc) {
goto _end;
Copy link
Contributor

@DanHeidinga DanHeidinga Feb 20, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can you add a level 1 tracepoint here? Assuming the error condition is rare, we should capture some diagnostics to know why we failed to determine if we were in a container or not.

Copy link
Contributor Author

@ashu-mehra ashu-mehra Feb 21, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

we should capture some diagnostics to know why we failed to determine if we were in a container or not.

I agree. I actually want to add tracepoints in the code at various error conditions and thought of taking that up it in a separate PR. Is that ok?

Copy link
Contributor

@DanHeidinga DanHeidinga Feb 21, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Works for me.

}
omrthread_monitor_enter(cgroupEntryListMonitor);
if (NULL == PPG_cgroupEntryList) {
rc = readCgroupFile(portLibrary, getpid(), inContainer, &PPG_cgroupEntryList, &PPG_cgroupSubsystemsAvailable);
}
omrthread_monitor_exit(cgroupEntryListMonitor);
if (0 != rc) {
goto _end;
}
}
} else {
rc = 0;
Expand Down Expand Up @@ -3720,13 +3811,14 @@ int32_t
omrsysinfo_cgroup_get_memlimit(struct OMRPortLibrary *portLibrary, uint64_t *limit)
{
int32_t rc = OMRPORT_ERROR_SYSINFO_CGROUP_UNSUPPORTED_PLATFORM;

Assert_PRT_true(NULL != limit);
Copy link
Contributor

@DanHeidinga DanHeidinga Feb 16, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Please update the function prototype to indicate limit cannot be null.

Copy link
Contributor Author

@ashu-mehra ashu-mehra Feb 19, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Done


#if defined(LINUX) && !defined(OMRZTPF)
uint64_t cgroupMemLimit = 0;
uint64_t physicalMemLimit = 0;
int32_t numItemsToRead = 1; /* memory.limit_in_bytes file contains only one integer value */

Assert_PRT_true(NULL != limit);

rc = readCgroupSubsystemFile(portLibrary, OMR_CGROUP_SUBSYSTEM_MEMORY, "memory.limit_in_bytes", numItemsToRead, "%lu", &cgroupMemLimit);

if (0 != rc) {
Expand Down