X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/4c04d8355923f8323a1a3195fe3b73ac08b984ea..22c8f349c3475e4bee6df083cc1871f45da2aecb:/tools/tesh/run_context.c

diff --git a/tools/tesh/run_context.c b/tools/tesh/run_context.c
index 59f787e187..d8e21f3afd 100644
--- a/tools/tesh/run_context.c
+++ b/tools/tesh/run_context.c
@@ -8,6 +8,7 @@
 
 #include "tesh.h"
 
+#include <signal.h>
 #include <sys/types.h>
 #include <sys/wait.h>
 #include <sys/stat.h>
@@ -15,94 +16,165 @@
 
 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(tesh);
 
+int fg_job = 0;
 xbt_dynar_t bg_jobs = NULL;
 rctx_t armageddon_initiator = NULL;
 xbt_os_mutex_t armageddon_mutex = NULL;
+struct {
+  int num;
+  struct sigaction act;
+} oldact[3];                    /* SIGINT, SIGQUIT, SIGTERM */
+
+xbt_os_thread_t sigwaiter_thread;
+xbt_os_mutex_t sigwaiter_mutex;
+xbt_os_cond_t sigwaiter_cond;
+int armageddon_requested = 0;
+int caught_signum = 0;
 
 /*
  * Module management
  */
 
-static void kill_it(void *r)
+static void armageddon_sighandler(int signum)
 {
-  rctx_t rctx = *(rctx_t *) r;
+  xbt_os_mutex_acquire(sigwaiter_mutex);
+  caught_signum = signum;
+  armageddon_requested = 1;
+  xbt_os_cond_signal(sigwaiter_cond);
+  xbt_os_mutex_release(sigwaiter_mutex);
+}
+
+static void *armageddon_sigwaiter(_XBT_GNUC_UNUSED void *arg)
+{
+  xbt_os_mutex_acquire(sigwaiter_mutex);
+  /* Inform main thread that it started. */
+  xbt_os_cond_signal(sigwaiter_cond);
+  /* Wait for ending signal... */
+  xbt_os_cond_wait(sigwaiter_cond, sigwaiter_mutex);
+  if (armageddon_requested) {
+    ERROR2("Test suite `%s': caught signal %d", testsuite_name, caught_signum);
+    rctx_armageddon(rctx, 3);
+  }
+  xbt_os_mutex_release(sigwaiter_mutex);
+  return NULL;
+}
 
-  VERB2("Join thread %p which were running background cmd <%s>", rctx->runner,
-        rctx->filepos);
+static void wait_it(rctx_t rctx)
+{
+  VERB2("Join thread %p which were running background cmd <%s>",
+        rctx->runner, rctx->filepos);
   xbt_os_thread_join(rctx->runner, NULL);
+}
+
+static void kill_it(void *r)
+{
+  rctx_t rctx = *(rctx_t *) r;
+  wait_it(rctx);
   rctx_free(rctx);
 }
 
 void rctx_init(void)
 {
+  struct sigaction newact;
+  int i;
+  fg_job = 0;
   bg_jobs = xbt_dynar_new_sync(sizeof(rctx_t), kill_it);
   armageddon_mutex = xbt_os_mutex_init();
   armageddon_initiator = NULL;
+  sigwaiter_mutex = xbt_os_mutex_init();
+  sigwaiter_cond = xbt_os_cond_init();
+  xbt_os_mutex_acquire(sigwaiter_mutex);
+  sigwaiter_thread = xbt_os_thread_create("Armaggedon request waiter",
+                                          armageddon_sigwaiter, NULL);
+  /* Wait for thread to start... */
+  xbt_os_cond_wait(sigwaiter_cond, sigwaiter_mutex);
+  xbt_os_mutex_release(sigwaiter_mutex);
+  memset(&newact, 0, sizeof(newact));
+  newact.sa_handler = armageddon_sighandler;
+  oldact[0].num = SIGINT;
+  oldact[1].num = SIGQUIT;
+  oldact[2].num = SIGTERM;
+  for (i = 0; i < 3; i++)
+    sigaction(oldact[i].num, &newact, &oldact[i].act);
 }
 
 void rctx_exit(void)
 {
-  if (bg_jobs) {
-    /* Do not use xbt_dynar_free or it will lock the dynar, preventing armageddon from working */
-    while (xbt_dynar_length(bg_jobs)) {
-      rctx_t rctx;
-      xbt_dynar_pop(bg_jobs, &rctx);
-      kill_it(&rctx);
-    }
-    xbt_dynar_free(&bg_jobs);
-  }
+  int i;
+  for (i = 0; i < 3; i++)
+    sigaction(oldact[i].num, &oldact[i].act, NULL);
+  xbt_os_cond_signal(sigwaiter_cond);
+  xbt_os_thread_join(sigwaiter_thread, NULL);
+  xbt_dynar_free(&bg_jobs);
+  xbt_os_cond_destroy(sigwaiter_cond);
+  xbt_os_mutex_destroy(sigwaiter_mutex);
   xbt_os_mutex_destroy(armageddon_mutex);
 }
 
 void rctx_wait_bg(void)
 {
-  if (bg_jobs) {
-    /* Do not use xbt_dynar_free or it will lock the dynar, preventing armageddon from working */
-    while (xbt_dynar_length(bg_jobs)) {
-      rctx_t rctx;
-      xbt_dynar_pop(bg_jobs, &rctx);
-      kill_it(&rctx);
+  /* Do not use xbt_dynar_free or it will lock the dynar, preventing armageddon
+   * from working */
+  while (xbt_dynar_length(bg_jobs)) {
+    rctx_t rctx = xbt_dynar_getlast_as(bg_jobs, rctx_t);
+    wait_it(rctx);
+    xbt_dynar_pop(bg_jobs, &rctx);
+    rctx_free(rctx);
+  }
+  xbt_dynar_reset(bg_jobs);
+}
+
+static void rctx_armageddon_kill_one(rctx_t initiator, const char *filepos,
+                                     rctx_t rctx)
+{
+  if (rctx != initiator) {
+    INFO2("Kill <%s> because <%s> failed", rctx->filepos, filepos);
+    xbt_os_mutex_acquire(rctx->interruption);
+    if (!rctx->reader_done) {
+      rctx->interrupted = 1;
+      kill(rctx->pid, SIGTERM);
+      usleep(100);
+      kill(rctx->pid, SIGKILL);
     }
-    xbt_dynar_free(&bg_jobs);
+    xbt_os_mutex_release(rctx->interruption);
   }
-  bg_jobs = xbt_dynar_new_sync(sizeof(rctx_t), kill_it);
 }
 
 void rctx_armageddon(rctx_t initiator, int exitcode)
 {
-  rctx_t rctx;
+  unsigned int cursor;
+  rctx_t job;
+  const char *filepos = initiator && initiator->filepos ?
+      initiator->filepos : "(master)";
 
-  DEBUG2("Armageddon request by <%s> (exit=%d)", initiator->filepos,
-         exitcode);
+  DEBUG2("Armageddon request by <%s> (exit=%d)", filepos, exitcode);
   xbt_os_mutex_acquire(armageddon_mutex);
   if (armageddon_initiator != NULL) {
     VERB0("Armageddon already started. Let it go");
-    xbt_os_mutex_release(initiator->interruption);
     xbt_os_mutex_release(armageddon_mutex);
     return;
   }
   DEBUG1("Armageddon request by <%s> got the lock. Let's go amok",
-         initiator->filepos);
+         filepos);
   armageddon_initiator = initiator;
   xbt_os_mutex_release(armageddon_mutex);
 
+  /* Kill foreground command */
+  if (fg_job)
+    rctx_armageddon_kill_one(initiator, filepos, rctx);
+
   /* Kill any background commands */
-  while (xbt_dynar_length(bg_jobs)) {
-    xbt_dynar_pop(bg_jobs, &rctx);
-    if (rctx != initiator) {
-      INFO2("Kill <%s> because <%s> failed", rctx->filepos,
-            initiator->filepos);
-      xbt_os_mutex_acquire(rctx->interruption);
-      rctx->interrupted = 1;
-      xbt_os_mutex_release(rctx->interruption);
-      if (!rctx->reader_done) {
-        kill(rctx->pid, SIGTERM);
-        usleep(100);
-        kill(rctx->pid, SIGKILL);
-      }
-    }
+  xbt_dynar_foreach(bg_jobs, cursor, job) {
+    rctx_armageddon_kill_one(initiator, filepos, job);
   }
 
+  /* Give runner threads a chance to acknowledge the processes deaths */
+  usleep(10000);
+  /* Ensure that nobody is running rctx_wait on exit */
+  if (fg_job)
+    xbt_os_mutex_acquire(rctx->interruption);
+  xbt_dynar_foreach(bg_jobs, cursor, job)
+    xbt_os_mutex_acquire(job->interruption);
   VERB0("Shut everything down!");
   exit(exitcode);
 }
@@ -114,23 +186,27 @@ void rctx_armageddon(rctx_t initiator, int exitcode)
 void rctx_empty(rctx_t rc)
 {
   int i;
-  char **env_it = environ;
+  char **env_it;
+  void *filepos;
 
   if (rc->cmd)
     free(rc->cmd);
   rc->cmd = NULL;
-  if (rc->filepos)
-    free(rc->filepos);
-  if (rc->env)
+  /* avoid race with rctx_armageddon log messages */
+  filepos = rc->filepos;
+  rc->filepos = NULL;
+  if (filepos)
+    free(filepos);
+  for (i = 0, env_it = environ; *env_it; i++, env_it++);
+  if (rc->env) {
+    for (env_it = rctx->env + i; *env_it; env_it++)
+      free(*env_it);
     free(rc->env);
+  }
+  rc->env_size = i + 1;
+  rc->env = malloc(rc->env_size * sizeof(char *));
+  memcpy(rc->env, environ, rc->env_size * sizeof(char *));
 
-  for (i = 0; *env_it; i++, env_it++);
-  i++;
-  rc->env_size = i;
-  rc->env = malloc(i * sizeof(char *));
-  memcpy(rc->env, environ, i * sizeof(char *));
-
-  rc->filepos = NULL;
   rc->is_empty = 1;
   rc->is_background = 0;
   rc->is_stoppable = 0;
@@ -167,8 +243,14 @@ void rctx_free(rctx_t rctx)
     free(rctx->cmd);
   if (rctx->filepos)
     free(rctx->filepos);
-  if (rctx->env)
+  if (rctx->env) {
+    int i;
+    char **env_it;
+    for (i = 0, env_it = environ; *env_it; i++, env_it++);
+    for (env_it = rctx->env + i; *env_it; env_it++)
+      free(*env_it);
     free(rctx->env);
+  }
   xbt_os_mutex_destroy(rctx->interruption);
   xbt_strbuff_free(rctx->input);
   xbt_strbuff_free(rctx->output_got);
@@ -201,9 +283,9 @@ void rctx_pushline(const char *filepos, char kind, char *line)
     if (rctx->cmd) {
       if (!rctx->is_empty) {
         ERROR2
-          ("[%s] More than one command in this chunk of lines (previous: %s).\n"
-           " Cannot guess which input/output belongs to which command.",
-           filepos, rctx->cmd);
+            ("[%s] More than one command in this chunk of lines (previous: %s).\n"
+             " Cannot guess which input/output belongs to which command.",
+             filepos, rctx->cmd);
         ERROR1("Test suite `%s': NOK (syntax error)", testsuite_name);
         rctx_armageddon(rctx, 1);
         return;
@@ -270,6 +352,7 @@ void rctx_pushline(const char *filepos, char kind, char *line)
       char *eq = strchr(line + len, '=');
       char *key = bprintf("%.*s", (int) (eq - line - len), line + len);
       xbt_dict_set(env, key, xbt_strdup(eq + 1), xbt_free_f);
+      free(key);
 
       rctx->env = realloc(rctx->env, ++(rctx->env_size) * sizeof(char *));
       rctx->env[rctx->env_size - 2] = xbt_strdup(line + len);
@@ -301,8 +384,8 @@ static void *thread_writer(void *r)
     int got;
     DEBUG1("Still %d chars to write", rctx->input->used - posw);
     got =
-      write(rctx->child_to, rctx->input->data + posw,
-            rctx->input->used - posw);
+        write(rctx->child_to, rctx->input->data + posw,
+              rctx->input->used - posw);
     if (got > 0)
       posw += got;
     if (got < 0) {
@@ -353,8 +436,10 @@ static void *thread_reader(void *r)
   /* let this thread wait for the child so that the main thread can detect the timeout without blocking on the wait */
   got_pid = waitpid(rctx->pid, &rctx->status, 0);
   if (got_pid != rctx->pid) {
-    perror(bprintf("(%s) Cannot wait for the child %s (got pid %d where pid %d were expected;rctx=%p;status=%d)",
-		   xbt_thread_self_name(), rctx->cmd, (int)got_pid, (int)rctx->pid,rctx,rctx->status));
+    perror(bprintf
+           ("(%s) Cannot wait for the child %s (got pid %d where pid %d were expected;status=%d)",
+            xbt_thread_self_name(), rctx->cmd, (int) got_pid,
+            (int) rctx->pid, rctx->status));
     ERROR1("Test suite `%s': NOK (system error)", testsuite_name);
     rctx_armageddon(rctx, 4);
     return NULL;
@@ -364,42 +449,57 @@ static void *thread_reader(void *r)
   return NULL;
 }
 
-/* Special command: mkfile is a building creating a file with the input data as content */
+/* Special command: mkfile is a built-in creating a file with the input data as content */
 static void rctx_mkfile(void)
 {
   char *filename = xbt_strdup(rctx->cmd + strlen("mkfile "));
   FILE *OUT;
+  int err;
   xbt_str_trim(filename, NULL);
   OUT = fopen(filename, "w");
   if (!OUT) {
-    free(filename);
     THROW3(system_error, errno, "%s: Cannot create file %s: %s",
            rctx->filepos, filename, strerror(errno));
   }
-  fprintf(OUT, "%s", rctx->input->data);
-  fclose(OUT);
+  err = (fprintf(OUT, "%s", rctx->input->data) < 0);
+  err = (fclose(OUT) == -1) || err;
+  if (err) {
+    THROW3(system_error, errno, "%s: Cannot write file %s: %s",
+           rctx->filepos, filename, strerror(errno));
+  }
+  free(filename);
 }
 
 /* function to be called from the child to start the actual process */
 static void start_command(rctx_t rctx)
 {
-  xbt_dynar_t cmd = xbt_str_split_quoted(rctx->cmd);
+  xbt_dynar_t cmd;
   char *binary_name = NULL;
   unsigned int it;
   char *str;
-  xbt_dynar_get_cpy(cmd, 0, &binary_name);
-  char **args = xbt_new(char *, xbt_dynar_length(cmd) + 1);
+  char **args;
   int errcode;
 
   if (!strncmp(rctx->cmd, "mkfile ", strlen("mkfile "))) {
     rctx_mkfile();
+    /* Valgrind detects memory leaks here.
+     * To correct those leaks, we must free objects allocated in main() or in
+     * handle_suite(), but we have no more reference to them at this point.
+     * A quick and dirty hack to make valgrind happy it to uncomment the
+     * following line.
+     */
+    /* execlp("true", "true", (const char *)0); */
     exit(0);                    /* end the working child */
   }
 
+  cmd = xbt_str_split_quoted(rctx->cmd);
+  xbt_dynar_get_cpy(cmd, 0, &binary_name);
+  args = xbt_new(char *, xbt_dynar_length(cmd) + 1);
   xbt_dynar_foreach(cmd, it, str) {
     args[it] = xbt_strdup(str);
   }
   args[it] = NULL;
+  xbt_dynar_free_container(&cmd);
 
   /* To search for the right executable path when not trivial */
   struct stat stat_buf;
@@ -456,9 +556,16 @@ void rctx_start(void)
   rctx->cmd = xbt_str_varsubst(rctx->cmd, env);
   VERB2("Start %s %s", rctx->cmd,
         (rctx->is_background ? "(background job)" : ""));
+  xbt_os_mutex_acquire(armageddon_mutex);
+  if (armageddon_initiator) {
+    VERB0("Armageddon in progress. Do not start job.");
+    xbt_os_mutex_release(armageddon_mutex);
+    return;
+  }
   if (pipe(child_in) || pipe(child_out)) {
     perror("Cannot open the pipes");
     ERROR1("Test suite `%s': NOK (system error)", testsuite_name);
+    xbt_os_mutex_release(armageddon_mutex);
     rctx_armageddon(rctx, 4);
   }
 
@@ -466,6 +573,7 @@ void rctx_start(void)
   if (rctx->pid < 0) {
     perror("Cannot fork the command");
     ERROR1("Test suite `%s': NOK (system error)", testsuite_name);
+    xbt_os_mutex_release(armageddon_mutex);
     rctx_armageddon(rctx, 4);
     return;
   }
@@ -484,12 +592,11 @@ void rctx_start(void)
 
     rctx->reader_done = 0;
     rctx->reader =
-      xbt_os_thread_create("reader", thread_reader, (void *) rctx);
+        xbt_os_thread_create("reader", thread_reader, (void *) rctx);
     rctx->writer =
-      xbt_os_thread_create("writer", thread_writer, (void *) rctx);
+        xbt_os_thread_create("writer", thread_writer, (void *) rctx);
 
   } else {                      /* child */
-
     close(child_in[1]);
     dup2(child_in[0], 0);
     close(child_in[0]);
@@ -505,7 +612,10 @@ void rctx_start(void)
   rctx->is_stoppable = 1;
 
   if (!rctx->is_background) {
+    fg_job = 1;
+    xbt_os_mutex_release(armageddon_mutex);
     rctx_wait(rctx);
+    fg_job = 0;
   } else {
     /* Damn. Copy the rctx and launch a thread to handle it */
     rctx_t old = rctx;
@@ -517,8 +627,10 @@ void rctx_start(void)
     DEBUG2("Launch a thread to wait for %s %d", old->cmd, old->pid);
     runner = xbt_os_thread_create(old->cmd, rctx_wait, (void *) old);
     old->runner = runner;
-    VERB3("Launched thread %p to wait for %s %d", runner, old->cmd, old->pid);
+    VERB3("Launched thread %p to wait for %s %d", runner, old->cmd,
+          old->pid);
     xbt_dynar_push(bg_jobs, &old);
+    xbt_os_mutex_release(armageddon_mutex);
   }
 }
 
@@ -540,7 +652,7 @@ void *rctx_wait(void *r)
            rctx->cmd);
 
   /* Wait for the child to die or the timeout to happen (or an armageddon to happen) */
-  while (!rctx->interrupted && !rctx->reader_done
+  while (!rctx->reader_done
          && (rctx->end_time < 0 || rctx->end_time >= now)) {
     usleep(100);
     now = time(NULL);
@@ -553,7 +665,6 @@ void *rctx_wait(void *r)
     kill(rctx->pid, SIGTERM);
     usleep(100);
     kill(rctx->pid, SIGKILL);
-    rctx->reader_done = 1;
   }
 
   /* Make sure helper threads die.
@@ -574,7 +685,8 @@ void *rctx_wait(void *r)
 
   /* Check for broken pipe */
   if (rctx->brokenpipe)
-    VERB0("Warning: Child did not consume all its input (I got broken pipe)");
+    VERB0
+        ("Warning: Child did not consume all its input (I got broken pipe)");
 
   /* Check for timeouts */
   if (rctx->timeout) {
@@ -587,6 +699,7 @@ void *rctx_wait(void *r)
            testsuite_name, rctx->filepos, timeout_value);
     DEBUG2("<%s> Interrupted = %d", rctx->filepos, rctx->interrupted);
     if (!rctx->interrupted) {
+      xbt_os_mutex_release(rctx->interruption);
       rctx_armageddon(rctx, 3);
       return NULL;
     }
@@ -622,9 +735,10 @@ void *rctx_wait(void *r)
     if (WIFEXITED(rctx->status)
         && WEXITSTATUS(rctx->status) != rctx->expected_return) {
       if (rctx->expected_return)
-        ERROR4("Test suite `%s': NOK (<%s> returned code %d instead of %d)",
-               testsuite_name, rctx->filepos,
-               WEXITSTATUS(rctx->status), rctx->expected_return);
+        ERROR4
+            ("Test suite `%s': NOK (<%s> returned code %d instead of %d)",
+             testsuite_name, rctx->filepos, WEXITSTATUS(rctx->status),
+             rctx->expected_return);
       else
         ERROR3("Test suite `%s': NOK (<%s> returned code %d)",
                testsuite_name, rctx->filepos, WEXITSTATUS(rctx->status));
@@ -653,12 +767,20 @@ void *rctx_wait(void *r)
     errcode = 1;
   }
 
-  if (rctx->output == e_output_check
-      && (rctx->output_got->used != rctx->output_wanted->used
-          || strcmp(rctx->output_got->data, rctx->output_wanted->data))) {
+  if ((errcode && errcode != 1) || rctx->interrupted) {
+    /* checking output, and matching */
+    xbt_dynar_t a = xbt_str_split(rctx->output_got->data, "\n");
+    char *out = xbt_str_join(a, "\n||");
+    xbt_dynar_free(&a);
+    INFO2("Output of <%s> so far: \n||%s", rctx->filepos, out);
+    free(out);
+  } else if (rctx->output == e_output_check
+             && (rctx->output_got->used != rctx->output_wanted->used
+                 || strcmp(rctx->output_got->data,
+                           rctx->output_wanted->data))) {
     if (XBT_LOG_ISENABLED(tesh, xbt_log_priority_info)) {
       char *diff =
-        xbt_str_diff(rctx->output_wanted->data, rctx->output_got->data);
+          xbt_str_diff(rctx->output_wanted->data, rctx->output_got->data);
       ERROR2("Output of <%s> mismatch:\n%s", rctx->filepos, diff);
       free(diff);
     }
@@ -674,20 +796,18 @@ void *rctx_wait(void *r)
     xbt_dynar_free(&a);
     INFO1("Here is the (ignored) command output: \n||%s", out);
     free(out);
-  } else if ((errcode && errcode != 1) || rctx->interrupted) {
-    /* checking output, and matching */
-    xbt_dynar_t a = xbt_str_split(rctx->output_got->data, "\n");
-    char *out = xbt_str_join(a, "\n||");
-    xbt_dynar_free(&a);
-    INFO2("Output of <%s> so far: \n||%s", rctx->filepos, out);
-    free(out);
   }
 
   if (!rctx->is_background) {
-    rctx_empty(rctx);
+    xbt_os_mutex_acquire(armageddon_mutex);
+    /* Don't touch rctx if armageddon is in progress. */
+    if (!armageddon_initiator)
+      rctx_empty(rctx);
+    xbt_os_mutex_release(armageddon_mutex);
   }
   if (errcode) {
     if (!rctx->interrupted) {
+      xbt_os_mutex_release(rctx->interruption);
       rctx_armageddon(rctx, errcode);
       return NULL;
     }