Update the secondary group cache (#1492)

* Use concurrent_flat_map to build a global cache rather than per
thread.
* Create global maintanence thread manager for clearing unused cache
entries.

libfuse/Makefile

@@ -56,7 +56,8 @@ SRC_CPP = \
 	lib/fuse_msgbuf.cpp \
 	lib/pin_threads.cpp \
 	lib/format.cpp \
-	lib/node.cpp
+	lib/node.cpp \
+	lib/maintenance_thread.cpp
 
 OBJS_C   = $(SRC_C:lib/%.c=build/%.o)
 OBJS_CPP = $(SRC_CPP:lib/%.cpp=build/%.o)

libfuse/include/fuse.h

@@ -688,8 +688,7 @@ int fuse_is_lib_option(const char *opt);
  */
 int fuse_main_real(int argc, char *argv[], const struct fuse_operations *op, size_t op_size);
 
-int  fuse_start_maintenance_thread(struct fuse *fuse);
-void fuse_stop_maintenance_thread(struct fuse *fuse);
+void fuse_populate_maintenance_thread(struct fuse *fuse);
 
 int  fuse_log_metrics_get(void);
 void fuse_log_metrics_set(int enabled);

libfuse/include/maintenance_thread.hpp

@@ -0,0 +1,13 @@
+#pragma once
+
+#include <functional>
+#include <vector>
+
+
+class MaintenanceThread
+{
+public:
+  static void setup();
+  static void stop();
+  static void push_job(const std::function<void(int)> &func);
+};

libfuse/lib/fuse.cpp

@@ -27,8 +27,10 @@
 #include "fuse_pollhandle.h"
 #include "fuse_msgbuf.hpp"
 
-#include <vector>
+#include "maintenance_thread.hpp"
+
 #include <string>
+#include <vector>
 
 #include <assert.h>
 #include <dlfcn.h>
@@ -141,7 +143,6 @@ struct fuse
   struct fuse_fs *fs;
   struct lock_queue_element *lockq;
 
-  pthread_t maintenance_thread;
   kvec_t(remembered_node_t) remembered_nodes;
 };
 
@@ -4032,49 +4033,26 @@ fuse_gc1()
   fuse_malloc_trim();
 }
 
-static
-void*
-fuse_maintenance_loop(void *fuse_)
+void
+fuse_populate_maintenance_thread(struct fuse *f_)
 {
-  int loops;
-  int sleep_time;
-  struct fuse *f = (struct fuse*)fuse_;
-
-  pthread_setname_np(pthread_self(),"fuse.maint");
-
-  loops = 0;
-  sleep_time = 60;
-  while(1)
-    {
-      if(remember_nodes(f))
-        fuse_prune_remembered_nodes(f);
-
-      if((loops % 15) == 0)
-        fuse_gc1();
-
-      if(g_LOG_METRICS)
-        metrics_log_nodes_info_to_tmp_dir(f);
-
-      loops++;
-      sleep(sleep_time);
-    }
-
-  return NULL;
-}
+  MaintenanceThread::push_job([=](int count_)
+  {
+    if(remember_nodes(f_))
+      fuse_prune_remembered_nodes(f_);
+  });
 
-int
-fuse_start_maintenance_thread(struct fuse *f_)
-{
-  return fuse_start_thread(&f_->maintenance_thread,fuse_maintenance_loop,f_);
-}
+  MaintenanceThread::push_job([](int count_)
+  {
+    if((count_ % 15) == 0)
+      fuse_gc1();
+  });
 
-void
-fuse_stop_maintenance_thread(struct fuse *f_)
-{
-  mutex_lock(&f_->lock);
-  pthread_cancel(f_->maintenance_thread);
-  mutex_unlock(&f_->lock);
-  pthread_join(f_->maintenance_thread,NULL);
+  MaintenanceThread::push_job([=](int count_)
+  {
+    if(g_LOG_METRICS)
+      metrics_log_nodes_info_to_tmp_dir(f_);
+  });
 }
 
 struct fuse*

libfuse/lib/fuse_loop.cpp

@@ -8,6 +8,7 @@
 #include "scope_guard.hpp"
 #include "thread_pool.hpp"
 #include "syslog.hpp"
+#include "maintenance_thread.hpp"
 
 #include "fuse_i.h"
 #include "fuse_kernel.h"
@@ -325,22 +326,23 @@ fuse_session_loop_mt(struct fuse_session *se_,
 }
 
 int
-fuse_loop_mt(struct fuse *f)
+fuse_loop_mt(struct fuse *f_)
 {
-  if(f == NULL)
-    return -1;
+  int res;
 
-  int res = fuse_start_maintenance_thread(f);
-  if(res)
+  if(f_ == NULL)
     return -1;
 
-  res = fuse_session_loop_mt(fuse_get_session(f),
+  MaintenanceThread::setup();
+  fuse_populate_maintenance_thread(f_);
+
+  res = fuse_session_loop_mt(fuse_get_session(f_),
                              fuse_config_get_read_thread_count(),
                              fuse_config_get_process_thread_count(),
                              fuse_config_get_process_thread_queue_depth(),
                              fuse_config_get_pin_threads());
 
-  fuse_stop_maintenance_thread(f);
+  MaintenanceThread::stop();
 
   return res;
 }

libfuse/lib/maintenance_thread.cpp

@@ -0,0 +1,65 @@
+#include "maintenance_thread.hpp"
+
+#include "fmt/core.h"
+
+#include <cassert>
+#include <mutex>
+#include <vector>
+
+#include <pthread.h>
+#include <unistd.h>
+
+
+pthread_t g_thread;
+std::vector<std::function<void(int)>> g_funcs;
+std::mutex g_mutex;
+
+static
+void*
+_thread_loop(void *)
+{
+  int count;
+
+  pthread_setname_np(pthread_self(),"fuse.maint");
+
+  count = 0;
+  while(true)
+    {
+      pthread_setcancelstate(PTHREAD_CANCEL_DISABLE,NULL);
+      {
+        std::lock_guard<std::mutex> lg(g_mutex);
+        for(auto &func : g_funcs)
+          func(count);
+      }
+      pthread_setcancelstate(PTHREAD_CANCEL_ENABLE,NULL);
+
+      count++;
+      ::sleep(60);
+    }
+
+  return NULL;
+}
+
+void
+MaintenanceThread::setup()
+{
+  int rv;
+
+  rv = pthread_create(&g_thread,NULL,_thread_loop,NULL);
+  assert((rv == 0) && "pthread_create failed");
+}
+
+void
+MaintenanceThread::push_job(const std::function<void(int)> &func_)
+{
+  std::lock_guard<std::mutex> lg(g_mutex);
+
+  g_funcs.emplace_back(func_);
+}
+
+void
+MaintenanceThread::stop()
+{
+  pthread_cancel(g_thread);
+  pthread_join(g_thread,NULL);
+}

mkdocs/docs/config/options.md

@@ -8,7 +8,7 @@ background](../intro_to_filesystems.md) in such things is recommended
 for more advanced configurations.
 
 These option names and values are the same regardless of whether you
-use them with the `mergerfs` commandline program, in fstab, or in a
+use them with the `mergerfs` command line program, in fstab, or in a
 config file.
 
 
@@ -204,6 +204,12 @@ config file.
 * **[passthrough](passthrough.md)**: Enable [FUSE IO
   passthrough](https://kernelnewbies.org/Linux_6.9#Faster_FUSE_I.2FO)
   if available. (default: off)
+* **gid-cache-expire-timeout**: Number of seconds till supplemental
+  group data is refreshed in the [GID
+  cache](../known_issues_bugs.md#supplemental-user-groups). (default:
+  3600)
+* **gid-cache-remove-timeout**: Number of seconds to wait till cached
+  data is removed due to lack of usage. (default: 43200)
 
 **NOTE:** Options are evaluated in the order listed so if the options
 are **func.rmdir=rand,category.action=ff** the **action** category

mkdocs/docs/known_issues_bugs.md

@@ -4,25 +4,26 @@
 
 ### Supplemental user groups
 
+#### Supplemental group caching
+
 Due to the overhead of
 [getgroups/setgroups](http://linux.die.net/man/2/setgroups) mergerfs
-utilizes a cache. This cache is opportunistic and per thread. Each
-thread will query the supplemental groups for a user when that
-particular thread needs to change credentials and will keep that data
-for the lifetime of the thread. This means that if a user is added to
-a group it may not be picked up without the restart of
-mergerfs. In the future this may be improved to allow a periodic or
-manual clearing of the cache.
+utilizes a cache. As necessary the supplemental group information will
+be queried and cached. That cached list of groups will be used to set
+the supplement groups as necessary. Due to the high cost of querying
+the group list the default expiry for said data is 1 hour and after 12
+hours of no usage will be removed from the cache all together.
+
+#### Host vs Container identity
 
 While not a bug some users have found when using containers that
 supplemental groups defined inside the container don't work as
 expected. Since mergerfs lives outside the container it is querying
-the host's group database. Effectively containers have their own user
-and group definitions unless setup otherwise just as different systems
-would.
+the host's group database. Containers have their own user and group
+definitions unless setup otherwise just as different systems would.
 
 Users should mount in the host group file into the containers or use a
-standard shared user & groups technology like NIS or LDAP.
+standard shared user and groups technology like NIS or LDAP.
 
 
 ### directory mtime is not being updated

src/config.cpp

@@ -106,6 +106,8 @@ Config::Config()
     fsname(),
     func(),
     fuse_msg_size("1M"),
+    gid_cache_expire_timeout(60 * 60),
+    gid_cache_remove_timeout(60 * 60 * 12),
     ignorepponrename(false),
     inodecalc("hybrid-hash"),
     lazy_umount_mountpoint(false),
@@ -184,6 +186,8 @@ Config::Config()
   _map["func.unlink"]            = &func.unlink;
   _map["func.utimens"]           = &func.utimens;
   _map["fuse_msg_size"]          = &fuse_msg_size;
+  _map["gid-cache-expire-timeout"] = &gid_cache_expire_timeout;
+  _map["gid-cache-remove-timeout"] = &gid_cache_remove_timeout;
   _map["handle-killpriv"]        = &handle_killpriv;
   _map["handle-killpriv-v2"]     = &handle_killpriv_v2;
   _map["ignorepponrename"]       = &ignorepponrename;

src/config.hpp

@@ -21,6 +21,7 @@
 #include "config_cachefiles.hpp"
 #include "config_flushonclose.hpp"
 #include "config_follow_symlinks.hpp"
+#include "config_gidcache.hpp"
 #include "config_inodecalc.hpp"
 #include "config_link_exdev.hpp"
 #include "config_log_metrics.hpp"
@@ -130,6 +131,8 @@ public:
   ConfigSTR      fsname;
   Funcs          func;
   ConfigPageSize fuse_msg_size;
+  GIDCacheExpireTimeout gid_cache_expire_timeout;
+  GIDCacheRemoveTimeout gid_cache_remove_timeout;
   ConfigBOOL     handle_killpriv = true;
   ConfigBOOL     handle_killpriv_v2 = true;
   ConfigBOOL     ignorepponrename;

src/config_gidcache.cpp

@@ -0,0 +1,65 @@
+#include "config_gidcache.hpp"
+
+#include "gidcache.hpp"
+
+#include "to_string.hpp"
+#include "from_string.hpp"
+
+
+GIDCacheExpireTimeout::GIDCacheExpireTimeout(const int i_)
+{
+  GIDCache::expire_timeout = i_;
+}
+
+GIDCacheExpireTimeout::GIDCacheExpireTimeout(const std::string &s_)
+{
+  from_string(s_);
+}
+
+std::string
+GIDCacheExpireTimeout::to_string(void) const
+{
+  return str::to(GIDCache::expire_timeout);
+}
+
+int
+GIDCacheExpireTimeout::from_string(const std::string &s_)
+{
+  int rv;
+
+  rv = str::from(s_,&GIDCache::expire_timeout);
+  if(rv < 0)
+    return rv;
+
+  return 0;
+}
+
+
+GIDCacheRemoveTimeout::GIDCacheRemoveTimeout(const int i_)
+{
+  GIDCache::remove_timeout = i_;
+}
+
+
+GIDCacheRemoveTimeout::GIDCacheRemoveTimeout(const std::string &s_)
+{
+  from_string(s_);
+}
+
+std::string
+GIDCacheRemoveTimeout::to_string(void) const
+{
+  return str::to(GIDCache::remove_timeout);
+}
+
+int
+GIDCacheRemoveTimeout::from_string(const std::string &s_)
+{
+  int rv;
+
+  rv = str::from(s_,&GIDCache::remove_timeout);
+  if(rv < 0)
+    return rv;
+
+  return 0;
+}

src/config_gidcache.hpp

@@ -0,0 +1,25 @@
+#pragma once
+
+#include "tofrom_string.hpp"
+
+class GIDCacheExpireTimeout : public ToFromString
+{
+public:
+  GIDCacheExpireTimeout(const int i = 60 * 60);
+  GIDCacheExpireTimeout(const std::string &);
+
+public:
+  std::string to_string(void) const final;
+  int from_string(const std::string &) final;
+};
+
+class GIDCacheRemoveTimeout : public ToFromString
+{
+public:
+  GIDCacheRemoveTimeout(const int = 60 * 60 * 12);
+  GIDCacheRemoveTimeout(const std::string &);
+
+public:
+  std::string to_string(void) const final;
+  int from_string(const std::string &) final;
+};

src/fuse_ioctl.cpp

@@ -336,7 +336,7 @@ namespace l
         fuse_invalidate_all_nodes();
         return 0;
       case IOCTL_INVALIDATE_GID_CACHE:
-        GIDCache::invalidate_all_caches();
+        GIDCache::invalidate_all();
         break;
       }
 

src/gidcache.cpp

@@ -14,7 +14,7 @@
   OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
 */
 
-#include "rnd.hpp"
+#include "gidcache.hpp"
 
 #include <grp.h>
 #include <pwd.h>
@@ -28,94 +28,9 @@
 # include <sys/param.h>
 #endif
 
-#include <cassert>
-#include <cstdlib>
-#include <algorithm>
-#include <unordered_map>
-#include <mutex>
-
-#include "gidcache.hpp"
-
-std::mutex                              g_REGISTERED_CACHES_MUTEX;
-std::unordered_map<pthread_t,GIDCache*> g_REGISTERED_CACHES;
-
-
-inline
-bool
-GIDRecord::operator<(const struct GIDRecord &b) const
-{
-  return uid < b.uid;
-}
-
-GIDCache::GIDCache()
-  : invalidate(false),
-    size(0),
-    recs()
-{
-  std::lock_guard<std::mutex> guard(g_REGISTERED_CACHES_MUTEX);
-
-  bool inserted;
-  pthread_t pthid;
-
-  pthid = pthread_self();
-  inserted = g_REGISTERED_CACHES.emplace(pthid,this).second;
-
-  assert(inserted == true);
-}
-
-inline
-GIDRecord *
-GIDCache::begin(void)
-{
-  return &recs[0];
-}
-
-inline
-GIDRecord *
-GIDCache::end(void)
-{
-  return &recs[size];
-}
-
-inline
-GIDRecord *
-GIDCache::allocrec(void)
-{
-  if(size == MAXRECS)
-    return &recs[RND::rand64(MAXRECS)];
-  else
-    return &recs[size++];
-}
-
-inline
-GIDRecord *
-GIDCache::lower_bound(GIDRecord   *begin,
-                      GIDRecord   *end,
-                      const uid_t  uid)
-{
-  int step;
-  int count;
-  GIDRecord *iter;
-
-  count = std::distance(begin,end);
-  while(count > 0)
-    {
-      iter = begin;
-      step = count / 2;
-      std::advance(iter,step);
-      if(iter->uid < uid)
-        {
-          begin  = ++iter;
-          count -= step + 1;
-        }
-      else
-        {
-          count = step;
-        }
-    }
-
-  return begin;
-}
+int GIDCache::expire_timeout = (60 * 60);
+int GIDCache::remove_timeout = (60 * 60 * 12);
+boost::concurrent_flat_map<uid_t,GIDRecord> GIDCache::_records;
 
 static
 int
@@ -131,85 +46,111 @@ _getgrouplist(const char  *user,
 #endif
 }
 
-GIDRecord *
-GIDCache::cache(const uid_t uid,
-                const gid_t gid)
-{
-  int rv;
-  char buf[4096];
-  struct passwd pwd;
-  struct passwd *pwdrv;
-  GIDRecord *rec;
-
-  rec = allocrec();
-
-  rec->uid = uid;
-  rv = ::getpwuid_r(uid,&pwd,buf,sizeof(buf),&pwdrv);
-  if(pwdrv != NULL && rv == 0)
-    {
-      rec->size = 0;
-      ::_getgrouplist(pwd.pw_name,gid,NULL,&rec->size);
-      rec->size = std::min(MAXGIDS,rec->size);
-      rv = ::_getgrouplist(pwd.pw_name,gid,rec->gids,&rec->size);
-      if(rv == -1)
-        {
-          rec->gids[0] = gid;
-          rec->size    = 1;
-        }
-    }
-
-  return rec;
-}
-
 static
 inline
 int
-setgroups(const GIDRecord *rec)
+_setgroups(const std::vector<gid_t> gids_)
 {
 #if defined __linux__ and UGID_USE_RWLOCK == 0
 # if defined SYS_setgroups32
-  return ::syscall(SYS_setgroups32,rec->size,rec->gids);
+  return ::syscall(SYS_setgroups32,gids_.size(),gids_.data());
 # else
-  return ::syscall(SYS_setgroups,rec->size,rec->gids);
+  return ::syscall(SYS_setgroups,gids_.size(),gids_.data());
 # endif
 #else
-  return ::setgroups(rec->size,rec->gids);
+  return ::setgroups(gids_.size(),gids_.data());
 #endif
 }
 
-int
-GIDCache::initgroups(const uid_t uid,
-                     const gid_t gid)
+static
+void
+_getgroups(const uid_t         uid_,
+           const gid_t         gid_,
+           std::vector<gid_t> &gids_)
 {
   int rv;
-  GIDRecord *rec;
+  int ngroups;
+  char buf[4096];
+  struct passwd pwd;
+  struct passwd *pwdrv;
 
-  if(invalidate)
-    {
-      size       = 0;
-      invalidate = false;
-    }
+  gids_.clear();
+  rv = ::getpwuid_r(uid_,&pwd,buf,sizeof(buf),&pwdrv);
+  if((rv == -1) || (pwdrv == NULL))
+    goto error;
+
+  ngroups = 0;
+  rv = ::_getgrouplist(pwd.pw_name,gid_,NULL,&ngroups);
+  gids_.resize(ngroups);
+
+  rv = ::_getgrouplist(pwd.pw_name,gid_,gids_.data(),&ngroups);
+  if((size_t)ngroups < gids_.size())
+    gids_.resize(ngroups);
 
-  rec = lower_bound(begin(),end(),uid);
-  if(rec == end() || rec->uid != uid)
+  return;
+
+ error:
+  gids_.clear();
+  gids_.push_back(gid_);
+}
+
+
+int
+GIDCache::initgroups(const uid_t uid_,
+                     const gid_t gid_)
+{
+  auto first_func =
+    [=](auto &x)
     {
-      rec = cache(uid,gid);
-      rv = ::setgroups(rec);
-      std::sort(begin(),end());
-    }
-  else
+      x.second.last_update = ::time(NULL);
+      ::_getgroups(uid_,gid_,x.second.gids);
+      ::_setgroups(x.second.gids);
+    };
+  auto exists_func =
+    [=](auto &x)
     {
-      rv = ::setgroups(rec);
-    }
+      time_t now;
+
+      now = ::time(NULL);
+      if((now - x.second.last_update) > GIDCache::expire_timeout)
+        {
+          ::_getgroups(uid_,gid_,x.second.gids);
+          x.second.last_update = now;
+        }
+      ::_setgroups(x.second.gids);
+    };
+
+  _records.try_emplace_and_visit(uid_,
+                                 first_func,
+                                 exists_func);
 
-  return rv;
+  return 0;
 }
 
 void
-GIDCache::invalidate_all_caches()
+GIDCache::invalidate_all()
 {
-  std::lock_guard<std::mutex> guard(g_REGISTERED_CACHES_MUTEX);
+  _records.visit_all([](auto &x)
+  {
+    x.second.last_update = 0;
+  });
+}
+
+void
+GIDCache::clear_all()
+{
+  _records.clear();
+}
+
+void
+GIDCache::clear_unused()
+{
+  time_t now = ::time(NULL);
+  auto erase_func =
+    [=](auto &x)
+    {
+      return ((now - x.second.last_update) > GIDCache::remove_timeout);
+    };
 
-  for(auto &p : g_REGISTERED_CACHES)
-    p.second->invalidate = true;
+  _records.erase_if(erase_func);
 }

src/gidcache.hpp

@@ -16,10 +16,13 @@
 
 #pragma once
 
+#include "boost/unordered/concurrent_flat_map.hpp"
+
 #include <sys/types.h>
 #include <unistd.h>
 
 #include <array>
+#include <vector>
 
 #define MAXGIDS 32
 #define MAXRECS 256
@@ -32,42 +35,28 @@
 // told to invalidate the cache on demand. A second instance on the
 // same thread will cause an assert to be triggered.
 
-
 struct GIDRecord
 {
-  uid_t uid;
-  int   size;
-  gid_t gids[MAXGIDS];
-
-  bool
-  operator<(const struct GIDRecord &b) const;
+  std::vector<gid_t> gids;
+  time_t last_update;
 };
 
 struct GIDCache
 {
 public:
-  GIDCache();
-
-public:
-  bool   invalidate;
-  size_t size;
-  std::array<GIDRecord,MAXRECS> recs;
-
-private:
-  GIDRecord *begin(void);
-  GIDRecord *end(void);
-  GIDRecord *allocrec(void);
-  GIDRecord *lower_bound(GIDRecord   *begin,
-                         GIDRecord   *end,
-                         const uid_t  uid);
-  GIDRecord *cache(const uid_t uid,
-                   const gid_t gid);
-
-public:
+  static
   int
   initgroups(const uid_t uid,
              const gid_t gid);
 
+  static void invalidate_all();
+  static void clear_all();
+  static void clear_unused();
+
 public:
-  static void invalidate_all_caches();
+  static int expire_timeout;
+  static int remove_timeout;
+
+private:
+  static boost::concurrent_flat_map<uid_t,GIDRecord> _records;
 };

src/mergerfs.cpp

@@ -29,6 +29,7 @@
 #include "strvec.hpp"
 #include "syslog.hpp"
 #include "version.hpp"
+#include "maintenance_thread.hpp"
 
 #include "fuse_access.hpp"
 #include "fuse_bmap.hpp"
@@ -246,7 +247,7 @@ namespace l
   {
     SysLog::info("Received SIGUSR2 - triggering thorough gc");
     fuse_gc();
-    GIDCache::invalidate_all_caches();
+    GIDCache::clear_all();
   }
 
   static
@@ -310,6 +311,11 @@ namespace l
 
     l::warn_if_not_root();
 
+    MaintenanceThread::push_job([](int count_)
+    {
+      if((count_ % 60) == 0)
+        GIDCache::clear_unused();
+    });
     l::setup_resources(cfg->scheduling_priority);
     l::setup_signal_handlers();
     l::get_fuse_operations(ops,cfg->nullrw);

src/ugid.cpp

@@ -28,8 +28,6 @@ namespace ugid
   initgroups(const uid_t uid_,
              const gid_t gid_)
   {
-    static thread_local GIDCache cache;
-
-    cache.initgroups(uid_,gid_);
+    GIDCache::initgroups(uid_,gid_);
   }
 }