STATS: Dynamically mapped shared memory segment

Move from using a hash to a vector with offsets into shared memory.
Limit exposure of VPP data structures and include files to
external stats library and applications.

Change-Id: Ic06129f12d10cf4c4946a86d9bc734eacff2c7da
Signed-off-by: Ole Troan <ot@cisco.com>

3 files changed, 242 insertions, 231 deletions

diff --git a/src/vpp-api/client/libvppapiclient.map b/src/vpp-api/client/libvppapiclient.map
index f51b92403da..00a26fbc90e 100644
--- a/src/vpp-api/client/libvppapiclient.map
+++ b/src/vpp-api/client/libvppapiclient.map
@@ -15,8 +15,6 @@ VPPAPICLIENT_18.10 {
 	api_main;
 	stat_segment_connect;
 	stat_segment_disconnect;
-	stat_segment_register;
-	stat_segment_collect;
 	stat_segment_ls;
 	stat_segment_dump;
 	stat_segment_data_free;
diff --git a/src/vpp-api/client/stat_client.c b/src/vpp-api/client/stat_client.c
index ad7078ea745..b5aab74de8a 100644
--- a/src/vpp-api/client/stat_client.c
+++ b/src/vpp-api/client/stat_client.c
@@ -17,72 +17,122 @@
  *------------------------------------------------------------------
  */
 
-#include <vlib/vlib.h>
-#include <vppinfra/socket.h>
-#include <svm/ssvm.h>
-#include <vpp/stats/stats.h>
+#include <stdio.h>
+#include <errno.h>
+#include <sys/types.h>
+#include <sys/socket.h>
+#include <sys/un.h>
+#include <stdbool.h>
+#include <sys/stat.h>
 #include <regex.h>
+#include <assert.h>
+#include <vppinfra/vec.h>
+#include <vppinfra/lock.h>
 #include "stat_client.h"
+#include <stdatomic.h>
 
 typedef struct
 {
-  u64 current_epoch;
-  volatile int segment_ready;
-  ssvm_private_t stat_segment;	/* mapped stats segment object */
-  ssvm_shared_header_t *shared_header;
-  clib_spinlock_t *stat_segment_lockp;	/* Spinlock for the stats segment */
-  uword *counter_vector_by_name;
-  u64 *error_base;
+  uint64_t current_epoch;
+  stat_segment_shared_header_t *shared_header;
+  stat_segment_directory_entry_t *directory_vector;
+  ssize_t memory_size;
 } stat_client_main_t;
 
 stat_client_main_t stat_client_main;
 
+static int
+recv_fd (int sock)
+{
+  struct msghdr msg = { 0 };
+  struct cmsghdr *cmsg;
+  int fd = -1;
+  char iobuf[1];
+  struct iovec io = {.iov_base = iobuf,.iov_len = sizeof (iobuf) };
+  union
+  {
+    char buf[CMSG_SPACE (sizeof (fd))];
+    struct cmsghdr align;
+  } u;
+  msg.msg_iov = &io;
+  msg.msg_iovlen = 1;
+  msg.msg_control = u.buf;
+  msg.msg_controllen = sizeof (u.buf);
+
+  ssize_t size;
+  if ((size = recvmsg (sock, &msg, 0)) < 0)
+    {
+      perror ("recvmsg failed");
+      return -1;
+    }
+  cmsg = CMSG_FIRSTHDR (&msg);
+  if (cmsg && cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS)
+    {
+      memmove (&fd, CMSG_DATA (cmsg), sizeof (fd));
+    }
+  return fd;
+}
+
+static stat_segment_directory_entry_t *
+get_stat_vector (void)
+{
+  stat_client_main_t *sm = &stat_client_main;
+  ASSERT (sm->shared_header);
+  return stat_segment_pointer (sm->shared_header,
+			       sm->shared_header->directory_offset);
+}
+
 int
 stat_segment_connect (char *socket_name)
 {
   stat_client_main_t *sm = &stat_client_main;
-  ssvm_private_t *ssvmp = &sm->stat_segment;
-  clib_socket_t s = { 0 };
-  clib_error_t *err;
-  int fd = -1, retval;
+  int mfd = -1;
+  int sock;
 
   memset (sm, 0, sizeof (*sm));
-  s.config = socket_name;
-  s.flags = CLIB_SOCKET_F_IS_CLIENT | CLIB_SOCKET_F_SEQPACKET;
-  err = clib_socket_init (&s);
-  if (err)
+  if ((sock = socket (AF_UNIX, SOCK_SEQPACKET, 0)) < 0)
     {
-      clib_error_report (err);
+      perror ("Couldn't open socket");
       return -1;
     }
-  err = clib_socket_recvmsg (&s, 0, 0, &fd, 1);
-  if (err)
+
+  struct sockaddr_un un = { 0 };
+  un.sun_family = AF_UNIX;
+  strncpy ((char *) un.sun_path, socket_name, sizeof (un.sun_path) - 1);
+  if (connect (sock, (struct sockaddr *) &un, sizeof (struct sockaddr_un)) <
+      0)
     {
-      clib_error_report (err);
+      perror ("connect");
       return -1;
     }
-  clib_socket_close (&s);
 
-  memset (ssvmp, 0, sizeof (*ssvmp));
-  ssvmp->fd = fd;
-
-  /* Note: this closes memfd.fd */
-  retval = ssvm_slave_init_memfd (ssvmp);
-  if (retval)
+  if ((mfd = recv_fd (sock)) < 0)
     {
-      fprintf (stderr, "WARNING: segment map returned %d\n", retval);
+      fprintf (stderr, "Receiving file descriptor failed\n");
       return -1;
     }
+  close (sock);
 
-  ASSERT (ssvmp && ssvmp->sh);
+  /* mmap shared memory segment. */
+  void *memaddr;
+  struct stat st = { 0 };
 
-  /* Pick up the segment lock from the shared memory header */
-  sm->shared_header = ssvmp->sh;
-  sm->stat_segment_lockp = (clib_spinlock_t *) (sm->shared_header->opaque[0]);
-  sm->segment_ready = 1;
+  if (fstat (mfd, &st) == -1)
+    {
+      perror ("mmap");
+      return -1;
+    }
+  if ((memaddr =
+       mmap (NULL, st.st_size, PROT_READ, MAP_SHARED, mfd, 0)) == MAP_FAILED)
+    {
+      perror ("mmap");
+      return -1;
+    }
 
-  sm->counter_vector_by_name =
-    (uword *) sm->shared_header->opaque[STAT_SEGMENT_OPAQUE_DIR];
+  sm->memory_size = st.st_size;
+  sm->shared_header = memaddr;
+  sm->directory_vector =
+    stat_segment_pointer (memaddr, sm->shared_header->directory_offset);
 
   return 0;
 }
@@ -91,146 +141,72 @@ void
 stat_segment_disconnect (void)
 {
   stat_client_main_t *sm = &stat_client_main;
-  ssvm_delete_memfd (&sm->stat_segment);
-  return;
-}
-
-/*
- * The application needs to register which counters it is interested
- * in.
- */
-stat_segment_cached_pointer_t *
-stat_segment_register (u8 * stats[])
-{
-  int i;
-  uword *p;
-  stat_client_main_t *sm = &stat_client_main;
-  stat_segment_cached_pointer_t *cp, *cached_pointer_vec = 0;
+  munmap (sm->shared_header, sm->memory_size);
 
-  for (i = 0; i < vec_len (stats); i++)
-    {
-      p = hash_get_mem (sm->counter_vector_by_name, stats[i]);
-      if (p == 0)
-	{
-	  fprintf (stderr, "WARN: %s not in directory!\n", stats[i]);
-	  continue;
-	}
-      vec_add2 (cached_pointer_vec, cp, 1);
-      cp->name = strdup ((char *) stats[i]);	// Point to p->key instead?
-    }
-  return cached_pointer_vec;
-}
-
-static u64 *
-get_error_base (u32 thread_index)
-{
-  u64 *error_base = 0;
-  uword *p;
-  stat_client_main_t *sm = &stat_client_main;
-  stat_segment_directory_entry_t *ep;
-
-  /* Special case /err/0/counter_vector */
-  p = hash_get_mem (sm->counter_vector_by_name,
-		    format (0, "/err/%d/counter_vector", thread_index));
-  if (p)
-    {
-      ep = (stat_segment_directory_entry_t *) (p[0]);
-      error_base = ep->value;
-    }
-  return error_base;
+  return;
 }
 
-f64
+double
 stat_segment_heartbeat (void)
 {
-  f64 *heartbeat = 0;
-  uword *p;
-  stat_client_main_t *sm = &stat_client_main;
-  stat_segment_directory_entry_t *ep;
-
-  /* Special case /err/0/counter_vector */
-  p = hash_get_mem (sm->counter_vector_by_name,
-		    format (0, "/sys/heartbeat%c", 0));
-  if (p)
-    {
-      ep = (stat_segment_directory_entry_t *) (p[0]);
-      heartbeat = ep->value;
-    }
-  return *heartbeat;
-}
-
-static void
-maybe_update_cached_pointers (stat_segment_cached_pointer_t * cached_pointers)
-{
   stat_client_main_t *sm = &stat_client_main;
-  stat_segment_cached_pointer_t *cp;
-  uword *p;
-  int i;
-
-  /* Cached pointers OK? */
-  if (sm->current_epoch ==
-      (u64) sm->shared_header->opaque[STAT_SEGMENT_OPAQUE_EPOCH])
-    return;
-
-  /* Special case /err/0/counter_vector */
-  sm->error_base = get_error_base (0);
-
-  /* Nope, fix them... */
-  for (i = 0; i < vec_len (cached_pointers); i++)
-    {
-      cp = &cached_pointers[i];
-
-      p = hash_get_mem (sm->counter_vector_by_name, cp->name);
-      if (p == 0)
-	{
-	  fprintf (stderr, "WARN: %s not in directory!\n", cp->name);
-	  continue;
-	}
-      cp->ep = (stat_segment_directory_entry_t *) (p[0]);
-    }
-
-  /* And remember that we did... */
-  sm->current_epoch =
-    (u64) sm->shared_header->opaque[STAT_SEGMENT_OPAQUE_EPOCH];
+  stat_segment_directory_entry_t *vec = get_stat_vector ();
+  double *hb = stat_segment_pointer (sm->shared_header, vec[4].offset);
+  return *hb;
 }
 
 stat_segment_data_t
-copy_data (stat_segment_directory_entry_t * ep, u64 * error_base, char *name)
+copy_data (stat_segment_directory_entry_t * ep)
 {
+  stat_client_main_t *sm = &stat_client_main;
   stat_segment_data_t result = { 0 };
-  u32 error_index;
   int i;
   vlib_counter_t **combined_c;	/* Combined counter */
   counter_t **simple_c;		/* Simple counter */
+  counter_t *error_base;
+  uint64_t *offset_vector;
+
+  assert (sm->shared_header);
+
   result.type = ep->type;
-  result.name = name;
+  result.name = strdup (ep->name);
   switch (ep->type)
     {
-    case STAT_DIR_TYPE_SCALAR_POINTER:
-      result.scalar_value = *(f64 *) ep->value;
-      break;
-
-    case STAT_DIR_TYPE_VECTOR_POINTER:
-      result.vector_pointer = ep->value;
+    case STAT_DIR_TYPE_SCALAR_INDEX:
+      result.scalar_value = ep->value;
       break;
 
     case STAT_DIR_TYPE_COUNTER_VECTOR_SIMPLE:
-      simple_c = ep->value;
+      simple_c = stat_segment_pointer (sm->shared_header, ep->offset);
       result.simple_counter_vec = vec_dup (simple_c);
+      offset_vector =
+	stat_segment_pointer (sm->shared_header, ep->offset_vector);
       for (i = 0; i < vec_len (simple_c); i++)
-	result.simple_counter_vec[i] = vec_dup (simple_c[i]);
+	{
+	  counter_t *cb =
+	    stat_segment_pointer (sm->shared_header, offset_vector[i]);
+	  result.simple_counter_vec[i] = vec_dup (cb);
+	}
       break;
 
     case STAT_DIR_TYPE_COUNTER_VECTOR_COMBINED:
-      combined_c = ep->value;
+      combined_c = stat_segment_pointer (sm->shared_header, ep->offset);
       result.combined_counter_vec = vec_dup (combined_c);
+      offset_vector =
+	stat_segment_pointer (sm->shared_header, ep->offset_vector);
       for (i = 0; i < vec_len (combined_c); i++)
-	result.combined_counter_vec[i] = vec_dup (combined_c[i]);
+	{
+	  vlib_counter_t *cb =
+	    stat_segment_pointer (sm->shared_header, offset_vector[i]);
+	  result.combined_counter_vec[i] = vec_dup (cb);
+	}
       break;
 
     case STAT_DIR_TYPE_ERROR_INDEX:
-      error_index = (uintptr_t) ep->value;
-      result.error_value = error_base[error_index];
+      error_base =
+	stat_segment_pointer (sm->shared_header,
+			      sm->shared_header->error_offset);
+      result.error_value = error_base[ep->index];
       break;
 
     default:
@@ -239,32 +215,6 @@ copy_data (stat_segment_directory_entry_t * ep, u64 * error_base, char *name)
   return result;
 }
 
-stat_segment_data_t *
-stat_segment_collect (stat_segment_cached_pointer_t * cached_pointers)
-{
-  stat_client_main_t *sm = &stat_client_main;
-  stat_segment_data_t *res = 0;
-  int i;
-
-  /* Grab the stats segment lock */
-  clib_spinlock_lock (sm->stat_segment_lockp);
-
-  /* see if we need to update cached pointers */
-  maybe_update_cached_pointers (cached_pointers);
-
-  for (i = 0; i < vec_len (cached_pointers); i++)
-    {
-      vec_add1 (res,
-		copy_data (cached_pointers[i].ep, sm->error_base,
-			   cached_pointers[i].name));
-    }
-
-  /* Drop the lock */
-  clib_spinlock_unlock (sm->stat_segment_lockp);
-
-  return res;
-}
-
 void
 stat_segment_data_free (stat_segment_data_t * res)
 {
@@ -286,19 +236,48 @@ stat_segment_data_free (stat_segment_data_t * res)
 	default:
 	  ;
 	}
+      free (res[i].name);
     }
   vec_free (res);
 }
 
-u8 **
-stat_segment_ls (u8 ** patterns)
+
+typedef struct
+{
+  uint64_t epoch;
+} stat_segment_access_t;
+
+static void
+stat_segment_access_start (stat_segment_access_t * sa)
 {
   stat_client_main_t *sm = &stat_client_main;
-  hash_pair_t *p;
-  u8 **dir = 0;
+  stat_segment_shared_header_t *shared_header = sm->shared_header;
+  sa->epoch = shared_header->epoch;
+  while (shared_header->in_progress != 0)
+    ;
+}
+
+static bool
+stat_segment_access_end (stat_segment_access_t * sa)
+{
+  stat_client_main_t *sm = &stat_client_main;
+  stat_segment_shared_header_t *shared_header = sm->shared_header;
+
+  if (shared_header->epoch != sa->epoch || shared_header->in_progress)
+    return false;
+  return true;
+}
+
+uint32_t *
+stat_segment_ls (uint8_t ** patterns)
+{
+  stat_client_main_t *sm = &stat_client_main;
+  stat_segment_access_t sa;
+
+  uint32_t *dir = 0;
   regex_t regex[vec_len (patterns)];
 
-  int i;
+  int i, j;
   for (i = 0; i < vec_len (patterns); i++)
     {
       int rv = regcomp (&regex[i], (char *) patterns[i], 0);
@@ -309,58 +288,67 @@ stat_segment_ls (u8 ** patterns)
 	}
     }
 
-  clib_spinlock_lock (sm->stat_segment_lockp);
-
-  /* *INDENT-OFF* */
-  hash_foreach_pair (p, sm->counter_vector_by_name,
-  ({
-    for (i = 0; i < vec_len(patterns); i++) {
-      int rv = regexec(&regex[i], (char *)p->key, 0, NULL, 0);
-      if (rv == 0) {
-	vec_add1 (dir, (u8 *)p->key);
-	break;
-      }
-    }
-    if (vec_len(patterns) == 0)
-      vec_add1 (dir, (u8 *)p->key);
-  }));
-  /* *INDENT-ON* */
+  stat_segment_access_start (&sa);
 
-  clib_spinlock_unlock (sm->stat_segment_lockp);
+  stat_segment_directory_entry_t *counter_vec = get_stat_vector ();
+  for (j = 0; j < vec_len (counter_vec); j++)
+    {
+      for (i = 0; i < vec_len (patterns); i++)
+	{
+	  int rv = regexec (&regex[i], counter_vec[j].name, 0, NULL, 0);
+	  if (rv == 0)
+	    {
+	      vec_add1 (dir, j);
+	      break;
+	    }
+	}
+      if (vec_len (patterns) == 0)
+	vec_add1 (dir, j);
+    }
 
   for (i = 0; i < vec_len (patterns); i++)
     regfree (&regex[i]);
 
+  if (!stat_segment_access_end (&sa))
+    {
+      /* Failed, clean up */
+      vec_free (dir);
+      return 0;
+
+    }
+
+  /* Update last version */
+  sm->current_epoch = sa.epoch;
   return dir;
 }
 
 stat_segment_data_t *
-stat_segment_dump (u8 * stats[])
+stat_segment_dump (uint32_t * stats)
 {
   int i;
-  uword *p;
   stat_client_main_t *sm = &stat_client_main;
   stat_segment_directory_entry_t *ep;
   stat_segment_data_t *res = 0;
+  stat_segment_access_t sa;
 
-  clib_spinlock_lock (sm->stat_segment_lockp);
+  /* Has directory been update? */
+  if (sm->shared_header->epoch != sm->current_epoch)
+    return 0;
 
-  sm->error_base = get_error_base (0);
+  stat_segment_access_start (&sa);
   for (i = 0; i < vec_len (stats); i++)
     {
-      p = hash_get_mem (sm->counter_vector_by_name, stats[i]);
-      if (p == 0)
-	{
-	  fprintf (stderr, "WARN: %s not in directory!\n", stats[i]);
-	  continue;
-	}
       /* Collect counter */
-      ep = (stat_segment_directory_entry_t *) (p[0]);
-      vec_add1 (res, copy_data (ep, sm->error_base, (char *) stats[i]));
+      ep = vec_elt_at_index (sm->directory_vector, stats[i]);
+      vec_add1 (res, copy_data (ep));
     }
-  clib_spinlock_unlock (sm->stat_segment_lockp);
 
-  return res;
+  if (stat_segment_access_end (&sa))
+    return res;
+
+  fprintf (stderr, "Epoch changed while reading, invalid results\n");
+  // TODO increase counter
+  return 0;
 }
 
 /* Wrapper for accessing vectors from other languages */
@@ -380,6 +368,36 @@ stat_segment_string_vector (u8 ** string_vector, char *string)
   return string_vector;
 }
 
+stat_segment_data_t *
+stat_segment_dump_entry (uint32_t index)
+{
+  stat_client_main_t *sm = &stat_client_main;
+  stat_segment_directory_entry_t *ep;
+  stat_segment_data_t *res = 0;
+  stat_segment_access_t sa;
+
+  stat_segment_access_start (&sa);
+
+  /* Collect counter */
+  ep = vec_elt_at_index (sm->directory_vector, index);
+  vec_add1 (res, copy_data (ep));
+
+  if (stat_segment_access_end (&sa))
+    return res;
+  return 0;
+}
+
+char *
+stat_segment_index_to_name (uint32_t index)
+{
+  char *name;
+  stat_segment_directory_entry_t *counter_vec = get_stat_vector ();
+  stat_segment_directory_entry_t *ep;
+  ep = vec_elt_at_index (counter_vec, index);
+  name = strdup (ep->name);
+  return name;
+}
+
 /*
  * fd.io coding-style-patch-verification: ON
  *
diff --git a/src/vpp-api/client/stat_client.h b/src/vpp-api/client/stat_client.h
index 73e91b1881b..c1a0ecf4753 100644
--- a/src/vpp-api/client/stat_client.h
+++ b/src/vpp-api/client/stat_client.h
@@ -17,8 +17,9 @@
 #ifndef included_stat_client_h
 #define included_stat_client_h
 
-#include <vlib/vlib.h>
-#include <vpp/stats/stats.h>
+#include <stdint.h>
+#include <vpp/stats/stat_segment.h>
+#include <vlib/counter_types.h>
 
 typedef struct
 {
@@ -26,31 +27,25 @@ typedef struct
   stat_directory_type_t type;
   union
   {
-    f64 scalar_value;
-    u64 error_value;
-    u64 *vector_pointer;
+    double scalar_value;
+    uint64_t error_value;
     counter_t **simple_counter_vec;
     vlib_counter_t **combined_counter_vec;
   };
 } stat_segment_data_t;
 
-typedef struct
-{
-  char *name;
-  stat_segment_directory_entry_t *ep;
-} stat_segment_cached_pointer_t;
-
 int stat_segment_connect (char *socket_name);
 void stat_segment_disconnect (void);
-u8 **stat_segment_ls (u8 ** pattern);
-stat_segment_data_t *stat_segment_dump (u8 ** counter_vec);
-stat_segment_cached_pointer_t *stat_segment_register (u8 ** counter_vec);
-/* Collects registered counters */
-stat_segment_data_t *stat_segment_collect (stat_segment_cached_pointer_t *);
-void stat_segment_data_free (stat_segment_data_t * res);
-f64 stat_segment_heartbeat (void);
-u8 **stat_segment_string_vector (u8 ** string_vector, char *string);
+uint8_t **stat_segment_string_vector (uint8_t ** string_vector, char *string);
 int stat_segment_vec_len (void *vec);
+uint32_t *stat_segment_ls (uint8_t ** pattern);
+stat_segment_data_t *stat_segment_dump (uint32_t * counter_vec);
+stat_segment_data_t *stat_segment_dump_entry (uint32_t index);
+void stat_segment_data_free (stat_segment_data_t * res);
+
+double stat_segment_heartbeat (void);
+
+char *stat_segment_index_to_name (uint32_t index);
 
 #endif /* included_stat_client_h */