8261242: [Linux] OSContainer::is_containerized() returns true when run outside a container

make/data/hotspot-symbols/symbols-linux

@@ -22,6 +22,7 @@
 #
 
 JVM_handle_linux_signal
+JVM_IsContainerized
 JVM_IsUseContainerSupport
 numa_error
 numa_warn

src/hotspot/os/linux/cgroupSubsystem_linux.cpp

@@ -63,7 +63,9 @@ CgroupSubsystem* CgroupSubsystemFactory::create() {
     // Construct the subsystem, free resources and return
     // Note: any index in cg_infos will do as the path is the same for
     //       all controllers.
-    CgroupController* unified = new CgroupV2Controller(cg_infos[MEMORY_IDX]._mount_path, cg_infos[MEMORY_IDX]._cgroup_path);
+    CgroupController* unified = new CgroupV2Controller(cg_infos[MEMORY_IDX]._mount_path,
+                                                       cg_infos[MEMORY_IDX]._cgroup_path,
+                                                       cg_infos[MEMORY_IDX]._read_only);
     log_debug(os, container)("Detected cgroups v2 unified hierarchy");
     cleanup(cg_infos);
     return new CgroupV2Subsystem(unified);
@@ -100,19 +102,19 @@ CgroupSubsystem* CgroupSubsystemFactory::create() {
     CgroupInfo info = cg_infos[i];
     if (info._data_complete) { // pids controller might have incomplete data
       if (strcmp(info._name, "memory") == 0) {
-        memory = new CgroupV1MemoryController(info._root_mount_path, info._mount_path);
+        memory = new CgroupV1MemoryController(info._root_mount_path, info._mount_path, info._read_only);
         memory->set_subsystem_path(info._cgroup_path);
       } else if (strcmp(info._name, "cpuset") == 0) {
-        cpuset = new CgroupV1Controller(info._root_mount_path, info._mount_path);
+        cpuset = new CgroupV1Controller(info._root_mount_path, info._mount_path, info._read_only);
         cpuset->set_subsystem_path(info._cgroup_path);
       } else if (strcmp(info._name, "cpu") == 0) {
-        cpu = new CgroupV1Controller(info._root_mount_path, info._mount_path);
+        cpu = new CgroupV1Controller(info._root_mount_path, info._mount_path, info._read_only);
         cpu->set_subsystem_path(info._cgroup_path);
       } else if (strcmp(info._name, "cpuacct") == 0) {
-        cpuacct = new CgroupV1Controller(info._root_mount_path, info._mount_path);
+        cpuacct = new CgroupV1Controller(info._root_mount_path, info._mount_path, info._read_only);
         cpuacct->set_subsystem_path(info._cgroup_path);
       } else if (strcmp(info._name, "pids") == 0) {
-        pids = new CgroupV1Controller(info._root_mount_path, info._mount_path);
+        pids = new CgroupV1Controller(info._root_mount_path, info._mount_path, info._read_only);
         pids->set_subsystem_path(info._cgroup_path);
       }
     } else {
@@ -127,7 +129,8 @@ void CgroupSubsystemFactory::set_controller_paths(CgroupInfo* cg_infos,
                                                   int controller,
                                                   const char* name,
                                                   char* mount_path,
-                                                  char* root_path) {
+                                                  char* root_path,
+                                                  bool read_only) {
   if (cg_infos[controller]._mount_path != nullptr) {
     // On some systems duplicate controllers get mounted in addition to
     // the main cgroup controllers most likely under /sys/fs/cgroup. In that
@@ -139,16 +142,74 @@ void CgroupSubsystemFactory::set_controller_paths(CgroupInfo* cg_infos,
       os::free(cg_infos[controller]._root_mount_path);
       cg_infos[controller]._mount_path = os::strdup(mount_path);
       cg_infos[controller]._root_mount_path = os::strdup(root_path);
+      cg_infos[controller]._read_only = read_only;
     } else {
       log_debug(os, container)("Duplicate %s controllers detected. Picking %s, skipping %s.",
                                name, cg_infos[controller]._mount_path, mount_path);
     }
   } else {
     cg_infos[controller]._mount_path = os::strdup(mount_path);
     cg_infos[controller]._root_mount_path = os::strdup(root_path);
+    cg_infos[controller]._read_only = read_only;
   }
 }
 
+/*
+ * Determine whether or not the mount options, which are comma separated,
+ * contain the 'ro' string.
+ */
+static bool find_ro_opt(char* mount_opts) {
+  char* token;
+  char* mo_ptr = mount_opts;
+  // mount options are comma-separated (man proc).
+  while ((token = strsep(&mo_ptr, ",")) != NULL) {
+    if (strcmp(token, "ro") == 0) {
+      return true;
+    }
+  }
+  return false;
+}
+
+/*
+ * Read values of a /proc/self/mountinfo line into variables. For cgroups v1
+ * super options are needed. On cgroups v2 super options are not used.
+ *
+ * The scanning of a single mountinfo line entry is as follows:
+ *
+ * 36  35  98:0      /mnt1 /mnt2 rw,noatime master:1 - ext3 /dev/root rw,errors=continue
+ * (1) (2) (3):(4)   (5)   (6)      (7)      (8)   (9) (10)   (11)         (12)
+ *
+ * The numbers in parentheses are labels for the descriptions below:
+ *
+ *  (1)   mount ID:        matched with '%*d' and discarded
+ *  (2)   parent ID:       matched with '%*d' and discarded
+ *  (3)   major:           ---,---> major, minor separated by ':'. matched with '%*d:%*d' and discarded
+ *  (4)   minor:           ---'
+ *  (5)   root:            matched with '%s' and captured in 'tmproot'. Must be non-empty.
+ *  (6)   mount point:     matched with '%s' and captured in 'tmpmount'. Must be non-empty.
+ *  (7)   mount options:   matched with '%s' and captured in 'mount_opts'. Must be non-empty.
+ *  (8)   optional fields: ---,---> matched with '%*[^-]-'. Anything not a hyphen, followed by a hyphen
+ *  (9)   separator:       ---'     and discarded. Note: The discarded match is space characters if there
+ *                                  are no optionals. Otherwise it includes the optional fields as well.
+ * (10)   filesystem type: matched with '%s' and captured in 'tmp_fs_type'
+ * (11)   mount source:    matched with '%*s' and discarded
+ * (12)   super options:   matched with '%s' and captured in 'tmpcgroups'
+ */
+static inline bool match_mount_info_line(char* line,
+                                         char* tmproot,
+                                         char* tmpmount,
+                                         char* mount_opts,
+                                         char* tmp_fs_type,
+                                         char* tmpcgroups) {
+ return sscanf(line,
+               "%*d %*d %*d:%*d %s %s %s%*[^-]- %s %*s %s",
+               tmproot,
+               tmpmount,
+               mount_opts,
+               tmp_fs_type,
+               tmpcgroups) == 5;
+}
+
 bool CgroupSubsystemFactory::determine_type(CgroupInfo* cg_infos,
                                             const char* proc_cgroups,
                                             const char* proc_self_cgroup,
@@ -320,26 +381,40 @@ bool CgroupSubsystemFactory::determine_type(CgroupInfo* cg_infos,
     char tmproot[MAXPATHLEN+1];
     char tmpmount[MAXPATHLEN+1];
     char tmpcgroups[MAXPATHLEN+1];
+    char mount_opts[MAXPATHLEN+1];
     char *cptr = tmpcgroups;
     char *token;
 
-    // Cgroup v2 relevant info. We only look for the _mount_path iff is_cgroupsV2 so
-    // as to avoid memory stomping of the _mount_path pointer later on in the cgroup v1
-    // block in the hybrid case.
-    if (is_cgroupsV2 && sscanf(p, "%*d %*d %*d:%*d %s %s %*[^-]- %s %*s %*s", tmproot, tmpmount, tmp_fs_type) == 3) {
+    /* Cgroup v2 relevant info. We only look for the _mount_path iff is_cgroupsV2 so
+     * as to avoid memory stomping of the _mount_path pointer later on in the cgroup v1
+     * block in the hybrid case.
+     *
+     * We collect the read only mount option in the cgroup infos so as to have that
+     * info ready when determining is_containerized().
+     */
+    if (is_cgroupsV2 && match_mount_info_line(p,
+                                              tmproot,
+                                              tmpmount,
+                                              mount_opts,
+                                              tmp_fs_type,
+                                              tmpcgroups /* unused */)) {
       // we likely have an early match return (e.g. cgroup fs match), be sure we have cgroup2 as fstype
       if (strcmp("cgroup2", tmp_fs_type) == 0) {
         cgroupv2_mount_point_found = true;
         any_cgroup_mounts_found = true;
+        // For unified we only have a single line with cgroup2 fs type.
+        // Therefore use that option for all CG info structs.
+        bool ro_option = find_ro_opt(mount_opts);
         for (int i = 0; i < CG_INFO_LENGTH; i++) {
-          set_controller_paths(cg_infos, i, "(cg2, unified)", tmpmount, tmproot);
+          set_controller_paths(cg_infos, i, "(cg2, unified)", tmpmount, tmproot, ro_option);
         }
       }
     }
 
     /* Cgroup v1 relevant info
      *
-     * Find the cgroup mount point for memory, cpuset, cpu, cpuacct, pids
+     * Find the cgroup mount point for memory, cpuset, cpu, cpuacct, pids. For each controller
+     * determine whether or not they show up as mounted read only or not.
      *
      * Example for docker:
      * 219 214 0:29 /docker/7208cebd00fa5f2e342b1094f7bed87fa25661471a4637118e65f1c995be8a34 /sys/fs/cgroup/memory ro,nosuid,nodev,noexec,relatime - cgroup cgroup rw,memory
@@ -348,32 +423,38 @@ bool CgroupSubsystemFactory::determine_type(CgroupInfo* cg_infos,
      * 34 28 0:29 / /sys/fs/cgroup/memory rw,nosuid,nodev,noexec,relatime shared:16 - cgroup cgroup rw,memory
      *
      * 44 31 0:39 / /sys/fs/cgroup/pids rw,nosuid,nodev,noexec,relatime shared:23 - cgroup cgroup rw,pids
+     *
      */
-    if (sscanf(p, "%*d %*d %*d:%*d %s %s %*[^-]- %s %*s %s", tmproot, tmpmount, tmp_fs_type, tmpcgroups) == 4) {
+    if (match_mount_info_line(p, tmproot, tmpmount, mount_opts, tmp_fs_type, tmpcgroups)) {
       if (strcmp("cgroup", tmp_fs_type) != 0) {
         // Skip cgroup2 fs lines on hybrid or unified hierarchy.
         continue;
       }
       while ((token = strsep(&cptr, ",")) != nullptr) {
         if (strcmp(token, "memory") == 0) {
           any_cgroup_mounts_found = true;
-          set_controller_paths(cg_infos, MEMORY_IDX, token, tmpmount, tmproot);
+          bool ro_option = find_ro_opt(mount_opts);
+          set_controller_paths(cg_infos, MEMORY_IDX, token, tmpmount, tmproot, ro_option);
           cg_infos[MEMORY_IDX]._data_complete = true;
         } else if (strcmp(token, "cpuset") == 0) {
           any_cgroup_mounts_found = true;
-          set_controller_paths(cg_infos, CPUSET_IDX, token, tmpmount, tmproot);
+          bool ro_option = find_ro_opt(mount_opts);
+          set_controller_paths(cg_infos, CPUSET_IDX, token, tmpmount, tmproot, ro_option);
           cg_infos[CPUSET_IDX]._data_complete = true;
         } else if (strcmp(token, "cpu") == 0) {
           any_cgroup_mounts_found = true;
-          set_controller_paths(cg_infos, CPU_IDX, token, tmpmount, tmproot);
+          bool ro_option = find_ro_opt(mount_opts);
+          set_controller_paths(cg_infos, CPU_IDX, token, tmpmount, tmproot, ro_option);
           cg_infos[CPU_IDX]._data_complete = true;
         } else if (strcmp(token, "cpuacct") == 0) {
           any_cgroup_mounts_found = true;
-          set_controller_paths(cg_infos, CPUACCT_IDX, token, tmpmount, tmproot);
+          bool ro_option = find_ro_opt(mount_opts);
+          set_controller_paths(cg_infos, CPUACCT_IDX, token, tmpmount, tmproot, ro_option);
           cg_infos[CPUACCT_IDX]._data_complete = true;
         } else if (strcmp(token, "pids") == 0) {
           any_cgroup_mounts_found = true;
-          set_controller_paths(cg_infos, PIDS_IDX, token, tmpmount, tmproot);
+          bool ro_option = find_ro_opt(mount_opts);
+          set_controller_paths(cg_infos, PIDS_IDX, token, tmpmount, tmproot, ro_option);
           cg_infos[PIDS_IDX]._data_complete = true;
         }
       }

src/hotspot/os/linux/cgroupSubsystem_linux.hpp

@@ -105,6 +105,7 @@
 class CgroupController: public CHeapObj<mtInternal> {
   public:
     virtual char* subsystem_path() = 0;
+    virtual bool is_read_only() = 0;
 
     /* Read a numerical value as unsigned long
      *
@@ -208,6 +209,7 @@ class CgroupSubsystem: public CHeapObj<mtInternal> {
     virtual jlong memory_and_swap_limit_in_bytes() = 0;
     virtual jlong memory_soft_limit_in_bytes() = 0;
     virtual jlong memory_max_usage_in_bytes() = 0;
+    virtual bool is_containerized() = 0;
 
     virtual char * cpu_cpuset_cpus() = 0;
     virtual char * cpu_cpuset_memory_nodes() = 0;
@@ -230,6 +232,7 @@ class CgroupInfo : public StackObj {
     char* _name;
     int _hierarchy_id;
     bool _enabled;
+    bool _read_only;            // whether or not the mount path is mounted read-only
     bool _data_complete;    // indicating cgroup v1 data is complete for this controller
     char* _cgroup_path;     // cgroup controller path from /proc/self/cgroup
     char* _root_mount_path; // root mount path from /proc/self/mountinfo. Unused for cgroup v2
@@ -240,6 +243,7 @@ class CgroupInfo : public StackObj {
       _name = nullptr;
       _hierarchy_id = -1;
       _enabled = false;
+      _read_only = false;
       _data_complete = false;
       _cgroup_path = nullptr;
       _root_mount_path = nullptr;
@@ -271,7 +275,8 @@ class CgroupSubsystemFactory: AllStatic {
                                      int controller,
                                      const char* name,
                                      char* mount_path,
-                                     char* root_path);
+                                     char* root_path,
+                                     bool read_only);
     // Determine the cgroup type (version 1 or version 2), given
     // relevant paths to files. Sets 'flags' accordingly.
     static bool determine_type(CgroupInfo* cg_infos,

src/hotspot/os/linux/cgroupV1Subsystem_linux.cpp

@@ -203,6 +203,17 @@ jlong CgroupV1Subsystem::memory_soft_limit_in_bytes() {
   }
 }
 
+bool CgroupV1Subsystem::is_containerized() {
+  // containerized iff all required controllers are mounted
+  // read-only. See OSContainer::is_containerized() for
+  // the full logic.
+  //
+  return _memory->controller()->is_read_only() &&
+         _cpu->controller()->is_read_only() &&
+         _cpuacct->is_read_only() &&
+         _cpuset->is_read_only();
+}
+
 /* memory_usage_in_bytes
  *
  * Return the amount of used memory for this process.

src/hotspot/os/linux/cgroupV1Subsystem_linux.hpp

@@ -36,19 +36,22 @@ class CgroupV1Controller: public CgroupController {
     /* mountinfo contents */
     char *_root;
     char *_mount_point;
+    bool _read_only;
 
     /* Constructed subsystem directory */
     char *_path;
 
   public:
-    CgroupV1Controller(char *root, char *mountpoint) {
+    CgroupV1Controller(char *root, char *mountpoint, bool ro) {
       _root = os::strdup(root);
       _mount_point = os::strdup(mountpoint);
       _path = nullptr;
+      _read_only = ro;
     }
 
     virtual void set_subsystem_path(char *cgroup_path);
     char *subsystem_path() { return _path; }
+    bool is_read_only() { return _read_only; }
 };
 
 class CgroupV1MemoryController: public CgroupV1Controller {
@@ -65,7 +68,7 @@ class CgroupV1MemoryController: public CgroupV1Controller {
     void set_hierarchical(bool value) { _uses_mem_hierarchy = value; }
 
   public:
-    CgroupV1MemoryController(char *root, char *mountpoint) : CgroupV1Controller(root, mountpoint) {
+    CgroupV1MemoryController(char *root, char *mountpoint, bool ro) : CgroupV1Controller(root, mountpoint, ro) {
       _uses_mem_hierarchy = false;
     }
 
@@ -94,6 +97,7 @@ class CgroupV1Subsystem: public CgroupSubsystem {
 
     jlong pids_max();
     jlong pids_current();
+    bool is_containerized();
 
     void print_version_specific_info(outputStream* st);
 

src/hotspot/os/linux/cgroupV2Subsystem_linux.cpp

@@ -94,6 +94,10 @@ int CgroupV2Subsystem::cpu_quota() {
   return limit;
 }
 
+bool CgroupV2Subsystem::is_containerized() {
+  return _unified->is_read_only();
+}
+
 char* CgroupV2Subsystem::cpu_cpuset_cpus() {
   char cpus[1024];
   CONTAINER_READ_STRING_CHECKED(_unified, "/cpuset.cpus", "cpuset.cpus", cpus, 1024);

src/hotspot/os/linux/cgroupV2Subsystem_linux.hpp

@@ -33,19 +33,22 @@ class CgroupV2Controller: public CgroupController {
     char *_mount_path;
     /* The cgroup path for the controller */
     char *_cgroup_path;
+    bool _read_only;
 
     /* Constructed full path to the subsystem directory */
     char *_path;
     static char* construct_path(char* mount_path, char *cgroup_path);
 
   public:
-    CgroupV2Controller(char * mount_path, char *cgroup_path) {
+    CgroupV2Controller(char * mount_path, char *cgroup_path, bool ro) {
       _mount_path = mount_path;
       _cgroup_path = os::strdup(cgroup_path);
       _path = construct_path(mount_path, cgroup_path);
+      _read_only = ro;
     }
 
     char *subsystem_path() { return _path; }
+    bool is_read_only() { return _read_only; }
 };
 
 class CgroupV2Subsystem: public CgroupSubsystem {
@@ -80,6 +83,7 @@ class CgroupV2Subsystem: public CgroupSubsystem {
     jlong pids_max();
     jlong pids_current();
 
+    bool is_containerized();
     void print_version_specific_info(outputStream* st);
 
     const char * container_type() {