Implement diodmount --jobid that can start one server per job.

Fix some bugs that prevented 'service diodctl restart' from working properly.

ChangeLog

@@ -1,12 +1,28 @@
-2010-04-08 Jim Garlick <garlick@llnl.gov>
+2010-04-09 Jim Garlick <garlick@llnl.gov>
+
+	* diodctl/serv.c, diod/diod.c, common/diod_sock.c : Fix bug where
+	diodctl's listen socket remained open in its children, preventing
+	it from being independently restarted.
+
+	* scripts/diodctl.init.in : In service_restart, call service_stop
+	and service_start directly instead of $0 stop && $0 start.
+	For some reason $0 start did not work in this context.
+
+	* diodctl/serv.c : Use jobid in addition to uid to identify servers.
+	Set name of child diod to descriptive name like diod-shared or
+	diod-jobid-%s or diod-uid-%d.
 
-	* diodctl/serv.c : Instead of storing the uid in the server struct,
-	store a pointer to the Npuser so that we can easily get at the
-	munge payload later.
+	* diodctl/ops.c : Pass jobid, stored in transport, into
+	diodctl_serv_create () and diodctl_serv_getname ().
+
+	* common/diod_trans.c : Store jobid with munge info in transport.
+	Access with diod_trans_set_authuser() and diod_trans_get_jobid().
+
+2010-04-08 Jim Garlick <garlick@llnl.gov>
 
 	* common/diod_upool.c : If munge cred includes a payload, store it
 	with the Npuser struct, and allow it to be retrieved with
-	diod_upool_get_mungepayload.
+	diod_user_get_authinfo.
 
 	* diodmount/diodmount.c : Add -j (jobid) option, and embed string
 	in munge payload.

common/diod_sock.c

@@ -178,12 +178,12 @@ diod_sock_listen_hostport_list (List l, struct pollfd **fdsp, int *nfdsp,
     return ret;
 }
 
-/* Listen on the first nfds file descriptors, which are assumed to be
+/* Listen on the first [nfds] fds after [starting], which are assumed to be
  * open and bound to appropriate addresses.
  * Return 0 on failure, nonzero on success.
  */
 int
-diod_sock_listen_first_nfds (struct pollfd **fdsp, int *nfdsp, int nfds)
+diod_sock_listen_nfds (struct pollfd **fdsp, int *nfdsp, int nfds, int starting)
 {
     struct pollfd *fds;
     int i;
@@ -194,7 +194,7 @@ diod_sock_listen_first_nfds (struct pollfd **fdsp, int *nfdsp, int nfds)
         goto done;
     }
     for (i = 0; i < nfds; i++)
-        fds[i].fd = i;
+        fds[i].fd = starting + i;
     *nfdsp = nfds;
     *fdsp = fds;
 

common/diod_sock.h

@@ -3,7 +3,8 @@ enum {
     DIOD_SOCK_QUIET_EADDRINUSE=2,
 };
 
-int  diod_sock_listen_first_nfds  (struct pollfd **fdsp, int *nfdsp, int nfds);
+int  diod_sock_listen_nfds  (struct pollfd **fdsp, int *nfdsp, int nfds,
+                             int starting);
 
 int  diod_sock_listen_hostport_list (List l, struct pollfd **fdsp, int *nfdsp,
                                      char *nport, int flags);

common/diod_trans.c

@@ -50,6 +50,7 @@ typedef struct {
     int              magic;
     int              authenticated;
     uid_t            authuser;
+    char            *jobid;
 } DTrans;
 
 #define DIOD_TRANS_MAGIC    0xf00fbaaa
@@ -64,24 +65,21 @@ diod_trans_create (int fd, char *host, char *ip, char *svc)
     Nptrans *npt;
     DTrans *dt;
 
-    dt = malloc (sizeof(*dt));
-    if (!dt)
+    if (!(dt = malloc (sizeof(*dt))))
         return NULL;
     dt->magic = DIOD_TRANS_MAGIC;
     dt->fd = fd;
     dt->authenticated = 0;
-    dt->host = strdup (host);
-    if (!dt->host) {
+    dt->jobid = NULL;
+    if (!(dt->host = strdup (host))) {
         diod_trans_destroy (dt);
         return NULL;
     }
-    dt->ip = strdup (ip);
-    if (!dt->ip) {
+    if (!(dt->ip = strdup (ip))) {
         diod_trans_destroy (dt);
         return NULL;
     }
-    dt->svc = strdup (svc);
-    if (!dt->svc) {
+    if (!(dt->svc = strdup (svc))) {
         diod_trans_destroy (dt);
         return NULL;
     }
@@ -111,6 +109,8 @@ diod_trans_destroy (void *a)
         free (dt->ip);
     if (dt->svc)
         free (dt->svc);
+    if (dt->jobid)
+        free (dt->jobid);
 
     free (dt);
 }
@@ -166,14 +166,16 @@ diod_trans_get_svc (Nptrans *trans)
 }
 
 void
-diod_trans_set_authuser (Nptrans *trans, uid_t uid)
+diod_trans_set_authuser (Nptrans *trans, uid_t uid, char *jobid)
 {
     DTrans *dt = trans->aux;
 
     assert (dt->magic == DIOD_TRANS_MAGIC);
 
     dt->authuser = uid;
     dt->authenticated = 1;
+    dt->jobid = jobid;
+
 }
 
 int
@@ -192,6 +194,16 @@ diod_trans_get_authuser (Nptrans *trans, uid_t *uidp)
     return ret;
 }
 
+char *
+diod_trans_get_jobid (Nptrans *trans)
+{
+    DTrans *dt = trans->aux;
+
+    assert (dt->magic == DIOD_TRANS_MAGIC);
+
+    return dt->jobid;
+}
+
 /*
  * vi:tabstop=4 shiftwidth=4 expandtab
  */

common/diod_trans.h

@@ -26,8 +26,9 @@ void diod_trans_destroy (void *trans);
 char *diod_trans_get_host (Nptrans *trans);
 char *diod_trans_get_ip (Nptrans *trans);
 char *diod_trans_get_svc (Nptrans *trans);
-void diod_trans_set_authuser (Nptrans *trans, uid_t uid);
+void diod_trans_set_authuser (Nptrans *trans, uid_t uid, char *jobid);
 int diod_trans_get_authuser (Nptrans *trans, uid_t *uidp);
+char *diod_trans_get_jobid (Nptrans *trans);
 
 /*
  * vi:tabstop=4 shiftwidth=4 expandtab

common/diod_upool.c

@@ -101,7 +101,7 @@ static Npuserpool upool = {
 typedef struct {
     int         magic;
     int         munged;         /* user supplied a valid munge cred */
-    char       *payload;        /* munge payload (if any) */
+    char       *jobid;          /* munge payload (if any) */
     int         nsg;            /* number of supplementary groups */
     gid_t       sg[NGROUPS_MAX];/* supplementary gid array */
 } Duser;
@@ -143,8 +143,8 @@ diod_switch_user (Npuser *u)
 static void
 _free_duser (Duser *d)
 {
-    if (d->payload)
-        free (d->payload);
+    if (d->jobid)
+        free (d->jobid);
     free (d);
 }
 
@@ -177,15 +177,15 @@ _getsg (struct passwd *pwd, gid_t *gp, int *glenp)
 }
 
 static Duser *
-_alloc_duser (struct passwd *pwd, int munged, char *payload)
+_alloc_duser (struct passwd *pwd, int munged, char *jobid)
 {
     Duser *d = NULL;
 
     if (!(d = np_malloc (sizeof (*d))))
         goto done;
     d->magic = DUSER_MAGIC;
     d->munged = munged;
-    d->payload = payload;
+    d->jobid = jobid;
     if (!_getsg (pwd, d->sg, &d->nsg))
         np_uerror (errno);
 done:
@@ -207,12 +207,19 @@ diod_become_user (char *name, uid_t uid, int realtoo)
     char buf[PASSWD_BUFSIZE];
     int nsg;
     gid_t sg[NGROUPS_MAX];
+    char *endptr;
 
+    if (name) {
+        errno = 0;
+        uid = strtoul (name, &endptr, 10);
+        if (errno == 0 && *name != '\0' && *endptr == '\0')
+            name = NULL;
+    }
     if (name) {
         if ((err = getpwnam_r (name, &pw, buf, sizeof(buf), &pwd)) != 0)
-            errn_exit (err, "error looking up uid %d", uid);
+            errn_exit (err, "error looking up user %s", name);
         if (!pwd)
-            msg_exit ("error looking up uid %d", uid);
+            msg_exit ("error looking up user %s", name);
     } else {
         if ((err = getpwuid_r (uid, &pw, buf, sizeof(buf), &pwd)) != 0)
             errn_exit (err, "error looking up uid %d", uid);
@@ -231,15 +238,14 @@ diod_become_user (char *name, uid_t uid, int realtoo)
 
 
 static Npuser *
-_alloc_user (Npuserpool *up, struct passwd *pwd, int munged,
-             char *payload)
+_alloc_user (Npuserpool *up, struct passwd *pwd, int munged, char *jobid)
 {
     Npuser *u;
     int err;
 
     if (!(u = np_malloc (sizeof (*u))))
         goto done;
-    if (!(u->aux = _alloc_duser (pwd, munged, payload)))
+    if (!(u->aux = _alloc_duser (pwd, munged, jobid)))
         goto done;
     if ((err = pthread_mutex_init (&u->lock, NULL)) != 0) {
         np_uerror (err);
@@ -282,23 +288,24 @@ diod_udestroy (Npuserpool *up, Npuser *u)
 }
 
 int
-diod_user_has_mungecred (Npuser *u)
+diod_user_get_authinfo (Npuser *u, int *authenticated, char **jobidp)
 {
     Duser *d = u->aux;
+    char *cpy = NULL;
+    int res = 0;
 
     assert (d->magic == DUSER_MAGIC);
 
-    return d->munged;
-}
-
-char *
-diod_user_get_mungepayload (Npuser *u)
-{
-    Duser *d = u->aux;
-
-    assert (d->magic == DUSER_MAGIC);
-
-    return d->payload;
+    if (d->munged && jobidp) {
+        if (d->jobid && !(cpy = strdup (d->jobid))) {
+            res = -1;
+            goto done;
+        }
+        *jobidp = cpy;
+    }
+    *authenticated = d->munged;
+done:
+    return res;
 }
 
 static int

common/diod_upool.h

@@ -22,10 +22,9 @@
  *****************************************************************************/
 
 int diod_switch_user (Npuser *u);
-int diod_user_has_mungecred (Npuser *u);
-char *diod_user_get_mungepayload (Npuser *u);
+char *diod_user_get_jobid (Npuser *u);
 int diod_switch_user (Npuser *u);
 void diod_become_user (char *name, uid_t uid, int realtoo);
-
+int diod_user_get_authinfo (Npuser *u, int *authenticated, char **jobidp);
 
 extern Npuserpool *diod_upool;

diod/diod.c

@@ -209,7 +209,7 @@ main(int argc, char **argv)
     if (!srv)
         msg_exit ("out of memory");
     if (Fopt) {
-        if (!diod_sock_listen_first_nfds (&fds, &nfds, Fopt))
+        if (!diod_sock_listen_nfds (&fds, &nfds, Fopt, 3))
             msg_exit ("failed to set up listen ports");
     } else {
         hplist = diod_conf_get_diodlisten ();

diod/ops.c

@@ -577,8 +577,11 @@ diod_attach (Npfid *fid, Npfid *nafid, Npstr *uname, Npstr *aname)
      * By the time we get here, invalid munge creds have already been rejected.
      */
     if (diod_conf_get_munge ()) {
-        if (diod_user_has_mungecred (fid->user)) {
-            diod_trans_set_authuser (fid->conn->trans, fid->user->uid);
+        int authenticated;
+
+        (void) diod_user_get_authinfo (fid->user, &authenticated, NULL);
+        if (authenticated) {
+            diod_trans_set_authuser (fid->conn->trans, fid->user->uid, NULL);
         } else {
             if (diod_trans_get_authuser (fid->conn->trans, &auid) < 0) {
                 np_uerror (EPERM);

diodctl/ops.c

@@ -96,8 +96,16 @@ _ctl_attach (Npfid *fid, Npfid *nafid, Npstr *uname, Npstr *aname)
      * By the time we get here, invalid munge creds have already been rejected.
      */
     if (diod_conf_get_munge ()) {
-        if (diod_user_has_mungecred (fid->user)) {
-            diod_trans_set_authuser (fid->conn->trans, fid->user->uid);
+        char *jobid;
+        int authenticated;
+
+        if (diod_user_get_authinfo (fid->user, &authenticated, &jobid) < 0) {
+            np_uerror (ENOMEM);
+            msg ("diodctl_attach: out of memory");
+            goto done;
+        }
+        if (authenticated) {
+            diod_trans_set_authuser (fid->conn->trans, fid->user->uid, jobid);
         } else {
             if (diod_trans_get_authuser (fid->conn->trans, &auid) < 0) {
                 np_uerror (EPERM);
@@ -181,8 +189,9 @@ static int
 _server_read (Npfilefid* file, u64 offset, u32 count, u8* data, Npreq *req)
 {
     Npfid *fid = file->fid;
+    char *jobid = diod_trans_get_jobid (fid->conn->trans);
 
-    return diodctl_serv_getname (fid->user, offset, count, data);
+    return diodctl_serv_getname (fid->user, jobid, offset, count, data);
 }
 
 /* Handle a write to the 'ctl' file.
@@ -192,9 +201,10 @@ static int
 _ctl_write (Npfilefid* file, u64 offset, u32 count, u8* data, Npreq *req)
 {
     Npfid *fid = file->fid;
+    char *jobid = diod_trans_get_jobid (fid->conn->trans);
     int ret = 0;
 
-    if (diodctl_serv_create (fid->user))
+    if (diodctl_serv_create (fid->user, jobid))
         ret = count;
     return ret;
 }