ebl Try to cancel read() with pthread_kill(TIMEOUT) to fix #1166

git-svn-id: https://bacula.svn.sourceforge.net/svnroot/bacula/trunk@7734 91ce42f0-d328-0410-95d8-f526ca767f89

diff --git a/bacula/patches/testing/bug_1166_cancel_read.patch b/bacula/patches/testing/bug_1166_cancel_read.patch
new file mode 100644 (file)
index 0000000..c25e1d0
--- /dev/null
+++ b/bacula/patches/testing/bug_1166_cancel_read.patch
@@ -0,0 +1,118 @@
+Index: src/dird/backup.c
+===================================================================
+--- src/dird/backup.c  (revision 7725)
++++ src/dird/backup.c  (working copy)
+@@ -193,6 +193,7 @@
+    Jmsg(jcr, M_INFO, 0, _("Start Backup JobId %s, Job=%s\n"),
+         edit_uint64(jcr->JobId, ed1), jcr->Job);
+ 
++   jcr->set_owner();            /* we are responsible for this job */
+    set_jcr_job_status(jcr, JS_Running);
+    Dmsg2(100, "JobId=%d JobLevel=%c\n", jcr->jr.JobId, jcr->jr.JobLevel);
+    if (!db_update_job_start_record(jcr, jcr->db, &jcr->jr)) {
+@@ -361,8 +362,10 @@
+       if (timeout) {
+          tid = start_bsock_timer(fd, timeout); /* TODO: New timeout directive??? */
+       }
++      Dmsg0(1, "======== Wait for a message\n");
+       /* Wait for Client to terminate */
+       while ((n = bget_dirmsg(fd)) >= 0) {
++         Dmsg1(1, "======= Get client message=%s\n", fd->msg);
+          if (!fd_ok && 
+              (sscanf(fd->msg, EndJob, &jcr->FDJobStatus, &JobFiles,
+                      &ReadBytes, &JobBytes, &Errors, &VSS, &Encrypt) == 7 ||
+Index: src/dird/job.c
+===================================================================
+--- src/dird/job.c     (revision 7731)
++++ src/dird/job.c     (working copy)
+@@ -390,6 +390,14 @@
+       jobq_remove(&job_queue, jcr); /* attempt to remove it from queue */
+       return true;
+ 
++   case JS_Running:
++      if (jcr->get_JobType() == JT_BACKUP && jcr->file_bsock) {
++         /* When in JS_Running state, the main thread can wait for
++          * EndJob message from the Client. We send a signal to the job
++          * thread to cancel the read() 
++          */
++         jcr->send_signal_to_owner(TIMEOUT_SIGNAL);
++      }
+    default:
+       /* Cancel File daemon */
+       if (jcr->file_bsock) {
+Index: src/jcr.h
+===================================================================
+--- src/jcr.h  (revision 7725)
++++ src/jcr.h  (working copy)
+@@ -180,6 +180,8 @@
+    bool is_job_canceled() {return job_canceled(this); };
+    int32_t get_JobType() { return m_JobType; };
+    int32_t get_JobLevel() { return m_JobLevel; };
++   void set_owner();
++   int send_signal_to_owner(int signal);
+ 
+    const char *get_OperationName();    /* in lib/jcr.c */
+    const char *get_ActionName(bool past); /* in lib/jcr.c */
+Index: src/lib/jcr.c
+===================================================================
+--- src/lib/jcr.c      (revision 7725)
++++ src/lib/jcr.c      (working copy)
+@@ -303,6 +303,31 @@
+ }
+ 
+ /*
++ * Update the my_thread_id variable with the current thread id
++ */
++void JCR::set_owner()
++{
++   Dmsg1(0, "JCR::set_owner(%p)\n", pthread_self());
++   this->my_thread_id = pthread_self();
++}
++
++/*
++ * Send a signal to the JCR thread owner. (to break a system call)
++ * (Must have use set_owner() before)
++ */
++int JCR::send_signal_to_owner(int sig)
++{
++   int ret=0;
++   Dmsg2(0, "Sending TIMEOUT ?? to %p by %p\n", this->my_thread_id, pthread_self());
++
++   if (!pthread_equal(this->my_thread_id, pthread_self())) {
++      Dmsg2(0, "Sending TIMEOUT to %p by %p\n", this->my_thread_id, pthread_self());
++      ret = pthread_kill(this->my_thread_id, sig);
++   }
++   return ret;
++}
++
++/*
+  * Push a subroutine address into the job end callback stack
+  */
+ void job_end_push(JCR *jcr, void job_end_cb(JCR *jcr,void *), void *ctx)
+@@ -719,7 +744,7 @@
+     bool set_waittime = false;
+     int oldJobStatus = jcr->JobStatus;
+ 
+-    Dmsg2(800, "set_jcr_job_status(%s, %c)\n", jcr->Job, JobStatus);
++    Dmsg2(2, "set_jcr_job_status(%s, %c)\n", jcr->Job, JobStatus);
+     /* if wait state is new, we keep current time for watchdog MaxWaitTime */
+     switch (JobStatus) {
+        case JS_WaitFD:
+@@ -740,7 +765,7 @@
+     * For a set of errors, ... keep the current status
+     *   so it isn't lost. For all others, set it.
+     */
+-   Dmsg3(300, "jid=%u OnEntry JobStatus=%c set=%c\n", (uint32_t)jcr->JobId,
++   Dmsg3(2, "jid=%u OnEntry JobStatus=%c set=%c\n", (uint32_t)jcr->JobId,
+          jcr->JobStatus, JobStatus);
+    switch (jcr->JobStatus) {
+    case JS_ErrorTerminated:
+@@ -781,7 +806,7 @@
+       }
+    }
+    if (oldJobStatus != jcr->JobStatus) {
+-      Dmsg3(200, "jid=%u leave set_old_job_status=%c new_set=%c\n", (uint32_t)jcr->JobId,
++      Dmsg3(2, "jid=%u leave set_old_job_status=%c new_set=%c\n", (uint32_t)jcr->JobId,
+          oldJobStatus, JobStatus);
+ //    generate_plugin_event(jcr, bEventStatusChange, NULL);
+    }