Let the F-Stack tools find a vdev / file_prefix stack by itself (#419)

lib/ff_dpdk_if.c

@@ -24,9 +24,13 @@
  *
  */
 #include <assert.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
 #include <unistd.h>
 #include <sys/mman.h>
 #include <errno.h>
+#include <fcntl.h>
 #include <time.h>
 
 #include <rte_common.h>
@@ -103,6 +107,9 @@ typedef char ff_reload_ring_name_fit_check[
      + 2 + 1                              /* "_g" + generation */
      <= (size_t)(RTE_RING_NAMESIZE - 1)) ? 1 : -1];
 
+/* also read by tools/compat/ff_ipc.c */
+#define FF_PRIMARY_RECORD "/var/run/ff_primary.conf"
+
 #ifdef FF_KNI
 #define KNI_MBUF_MAX 2048
 #define KNI_QUEUE_SIZE KNI_MBUF_MAX
@@ -2001,6 +2008,51 @@ fdir_add_tcp_flow(uint16_t port_id, uint16_t queue, uint16_t dir,
 
 #endif
 
+/* lets the F-Stack tools find this primary's EAL file prefix
+ * (tools/compat/ff_ipc.c reads it back). Best effort: a missing record
+ * only means the tools fall back to FF_FILE_PREFIX or the default. */
+static void
+ff_primary_record_write(void)
+{
+    char path[64], buf[1024], tmp[sizeof(FF_PRIMARY_RECORD) + 24], *p;
+    unsigned long long start;
+    int fd, i, rc;
+    ssize_t n;
+
+    snprintf(path, sizeof(path), "/proc/%d/stat", (int)getpid());
+    fd = open(path, O_RDONLY);
+    if (fd < 0)
+        return;
+    n = read(fd, buf, sizeof(buf) - 1);
+    close(fd);
+    if (n <= 0)
+        return;
+    buf[n] = '\0';
+
+    /* field 22 (starttime), counted after the ")" that ends comm */
+    p = strrchr(buf, ')');
+    if (p == NULL || p[1] != ' ')
+        return;
+    p += 2;
+    for (i = 0; i < 19; i++) {
+        p = strchr(p, ' ');
+        if (p == NULL)
+            return;
+        p++;
+    }
+    start = strtoull(p, NULL, 10);
+
+    snprintf(tmp, sizeof(tmp), FF_PRIMARY_RECORD ".%d.tmp", (int)getpid());
+    fd = open(tmp, O_WRONLY | O_CREAT | O_TRUNC | O_NOFOLLOW, 0644);
+    if (fd < 0)
+        return;
+    rc = dprintf(fd, "pid=%d\nstart_time=%llu\nruntime_dir=%s\n",
+        (int)getpid(), start, rte_eal_get_runtime_dir());
+    close(fd);
+    if (rc <= 0 || rename(tmp, FF_PRIMARY_RECORD) != 0)
+        unlink(tmp);
+}
+
 int
 ff_dpdk_init(int argc, char **argv)
 {
@@ -2019,6 +2071,10 @@ ff_dpdk_init(int argc, char **argv)
         rte_exit(EXIT_FAILURE, "Error with EAL initialization\n");
     }
 
+    if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
+        ff_primary_record_write();
+    }
+
     if (ff_global_cfg.log.level) {
         ff_log_open_set();
     }

tools/README.md

@@ -31,8 +31,9 @@ Each tool add an option `-p`(Which F-Stack process to communicate with, default
 The tools attach to the stack as DPDK secondary processes, so they must use the stack's EAL file prefix (the directory name under `/var/run/dpdk/`):
 
 - with `graceful_reload=1` the prefix is read from the resident slim primary's record `/var/run/ff_slim_primary.conf` (only while that primary is alive);
+- otherwise from the record every F-Stack primary writes, `/var/run/ff_primary.conf` (only while that primary is alive; with several stacks on one host it names the one started last);
 - otherwise the DPDK default `rte` is used;
-- `FF_FILE_PREFIX=<name>` overrides both. Note that `[dpdk] file_prefix=<v>` in config.ini gives the directory `container-<v>`, and a config with `[vdevN]` but no file_prefix gives `container`, e.g. `FF_FILE_PREFIX=container-<v> ff_top`.
+- `FF_FILE_PREFIX=<name>` overrides all of them, e.g. to pick one of several stacks. Note that `[dpdk] file_prefix=<v>` in config.ini gives the directory `container-<v>`, and a config with `[vdevN]` but no file_prefix gives `container`, e.g. `FF_FILE_PREFIX=container-<v> ff_top`.
 
 The tools do not probe PCI devices (`--no-pci`). `ff_top` and `ff_traffic` follow graceful reloads while sampling: a sample taken right after the target switched generations is not printed, because the new generation's counters start from zero.
 

tools/compat/ff_ipc.c

@@ -49,6 +49,8 @@ static int ff_ipc_exited;
 /* EAL file prefix override, the directory name under /var/run/dpdk */
 #define FF_IPC_PREFIX_ENV       "FF_FILE_PREFIX"
 #define FF_IPC_PRIMARY_RECORD   "/var/run/ff_slim_primary.conf"
+/* written by every F-Stack primary (lib/ff_dpdk_if.c) */
+#define FF_IPC_FSTACK_RECORD    "/var/run/ff_primary.conf"
 #define FF_IPC_RUNTIME_BASE     "/var/run/dpdk/"
 
 static struct rte_mempool *message_pool;
@@ -408,8 +410,9 @@ ff_ipc_proc_start(long pid, unsigned long long *start)
     return 0;
 }
 
-/* EAL file prefix of the runtime directory recorded by a live resident
- * slim primary (graceful_reload=1). No heap use: the EAL is not up yet. */
+/* EAL file prefix of the runtime directory recorded by a live primary:
+ * the resident slim primary (graceful_reload=1) or any F-Stack primary.
+ * No heap use: the EAL is not up yet. */
 static int
 ff_ipc_record_prefix(const char *record, char *prefix, size_t len)
 {
@@ -450,8 +453,9 @@ ff_ipc_record_prefix(const char *record, char *prefix, size_t len)
     return 0;
 }
 
-/* FF_FILE_PREFIX wins, then the resident primary's record; otherwise
- * the EAL default prefix is used. */
+/* FF_FILE_PREFIX wins, then the resident primary's record; ff_ipc_init()
+ * then tries the F-Stack primary's record, otherwise the EAL default
+ * prefix is used. */
 static const char *
 ff_ipc_file_prefix(const char *record, char *prefix, size_t len)
 {
@@ -500,6 +504,9 @@ ff_ipc_init(void)
 
     argc = (int)(sizeof(dpdk_argv) / sizeof(dpdk_argv[0])) - 2;
     p = ff_ipc_file_prefix(FF_IPC_PRIMARY_RECORD, prefix, sizeof(prefix));
+    if (p == NULL && ff_ipc_record_prefix(FF_IPC_FSTACK_RECORD, prefix,
+        sizeof(prefix)) == 0)
+        p = prefix;
     if (p != NULL) {
         snprintf(prefix_arg, sizeof(prefix_arg), "--file-prefix=%s", p);
         dpdk_argv[argc++] = prefix_arg;