snprintf.diff

diff -ru ./src/mpid/ch3/channels/common/src/affinity/hwloc_bind.c ../mvapich2-2.3.7-2_PDL/src/mpid/ch3/channels/common/src/affinity/hwloc_bind.c
--- ./src/mpid/ch3/channels/common/src/affinity/hwloc_bind.c	2024-09-17 18:47:06.000000000 -0400
+++ ../mvapich2-2.3.7-2_PDL/src/mpid/ch3/channels/common/src/affinity/hwloc_bind.c	2025-09-04 14:44:37.451641000 -0400
@@ -3097,6 +3097,32 @@
     return;
 }
 
+/* 
+ * safely snprintf() at any offset in a buffer.  if the offset is
+ * past the end of the buffer (or before the beginning of the buffer)
+ * we discard the data.   returns #of chars that would be printed
+ * to a buffer w/o any size limits (same ret value as snprintf()).
+ */
+int snprintfoff(char *buf, int offset, size_t bufsiz, char *fmt, ...) {
+    va_list ap;
+    char *bp, dummy[4];
+    size_t resid;
+    int rv;
+
+    va_start(ap, fmt);
+    if (offset < 0 || offset >= bufsiz) {  /* out of bounds?  discard data */
+        resid = 0;
+        bp = dummy;
+    } else {                               /* offset is within our buffer */
+        resid = bufsiz - offset;
+        bp = buf + offset;
+    }
+
+    rv = vsnprintf(bp, resid, fmt, ap);
+
+    va_end(ap);
+    return(rv);
+}
 
 #undef FUNCNAME
 #define FUNCNAME mv2_generate_implicit_cpu_mapping
@@ -3190,11 +3216,11 @@
         for (i = 0; i < local_procs; i++) {
             curr = count;
             for (k = 0; k < num_app_threads; k++) {
-                j += snprintf (mapping+j, _POSIX2_LINE_MAX, "%d,", mv2_core_map[curr]);
+                j += snprintfoff(mapping, j, s_cpu_mapping_line_max, "%d,", mv2_core_map[curr]);
                 curr = (curr + 1) % num_pu;
             }
             mapping [--j] = '\0'; 
-            j += snprintf (mapping+j, _POSIX2_LINE_MAX, ":");
+            j += snprintfoff(mapping, j, s_cpu_mapping_line_max, ":");
             count = (count + hw_threads_per_core) % num_pu;
         }
     } else if (mv2_hybrid_binding_policy == HYBRID_LINEAR) {
@@ -3203,14 +3229,14 @@
          * resources  */
         for (i = 0; i < local_procs; i++) {
             for (k = 0; k < num_app_threads; k++) {
-                j += snprintf (mapping+j, _POSIX2_LINE_MAX, "%d,", mv2_core_map[curr]);
+                j += snprintfoff(mapping, j, s_cpu_mapping_line_max, "%d,", mv2_core_map[curr]);
 
                 curr = ((curr + hw_threads_per_core) >= num_pu) ?
                             ((curr + hw_threads_per_core+ ++step) % num_pu) :
                             (curr + hw_threads_per_core) % num_pu;
             }
             mapping [--j] = '\0';
-            j += snprintf (mapping+j, _POSIX2_LINE_MAX, ":");
+            j += snprintfoff(mapping, j, s_cpu_mapping_line_max, ":");
         }    
     } else if (mv2_hybrid_binding_policy == HYBRID_SPREAD) {
 #if defined(CHANNEL_MRAIL)
@@ -3232,12 +3258,12 @@
             for (i = 0; i < local_procs; i++) {
                  for (k = curr; k < curr+chunk; k++) {
                      for (l = 0; l < hw_threads_per_core; l++) {
-                        j += snprintf (mapping+j, _POSIX2_LINE_MAX, "%d,", 
+                        j += snprintfoff(mapping, j, s_cpu_mapping_line_max, "%d,", 
                                 mv2_core_map[k * hw_threads_per_core + l]);
                      }
                  }
                  mapping [--j] = '\0';
-                 j += snprintf (mapping+j, _POSIX2_LINE_MAX, ":");
+                 j += snprintfoff(mapping, j, s_cpu_mapping_line_max, ":");
                  curr = (curr + chunk) % size;
             } 
         } else {
@@ -3252,11 +3278,11 @@
             for (i = 0; i < num_sockets; i++) {
                 for (k = curr; k < curr+ranks_per_sock; k++) {
                     for (l = 0; l < hw_threads_per_core; l++) {
-                        j += snprintf (mapping+j, _POSIX2_LINE_MAX, "%d,",
+                        j += snprintfoff(mapping, j, s_cpu_mapping_line_max, "%d,",
                                 mv2_core_map[k * hw_threads_per_core + l]);
                     }
                     mapping [--j] = '\0';
-                    j += snprintf (mapping+j, _POSIX2_LINE_MAX, ":");
+                    j += snprintfoff(mapping, j, s_cpu_mapping_line_max, ":");
                 }
                 curr = (curr + ((num_pu_per_socket/hw_threads_per_core)  * chunk)) % size;
             }
@@ -3265,7 +3291,7 @@
         /* Bunch mapping: Bind each MPI rank to a single phyical core of first
          * socket followed by second secket */
         for (i = 0; i < local_procs; i++) {
-            j += snprintf (mapping+j, _POSIX2_LINE_MAX, "%d:", mv2_core_map[k]);
+            j += snprintfoff(mapping, j, s_cpu_mapping_line_max, "%d:", mv2_core_map[k]);
             k = (k + hw_threads_per_core) % size;
         } 
     } else if (mv2_hybrid_binding_policy == HYBRID_SCATTER) {
@@ -3280,10 +3306,13 @@
             PRINT_INFO((MPIDI_Process.my_pg_rank == 0), "WARNING: Scatter is not a valid policy "
                     "for single-socket systems. Please re-run with Bunch or any other "
                     "applicable policy\n");
+            MPIU_Free(mv2_core_map);
+            MPIU_Free(mv2_core_map_per_numa);
+            hwloc_bitmap_free(allowed_cpuset);
             return MPI_ERR_OTHER;
         }
         for (i = 0; i < local_procs; i++) {
-            j += snprintf (mapping+j, _POSIX2_LINE_MAX, "%d:", mv2_core_map[k]);
+            j += snprintfoff(mapping, j, s_cpu_mapping_line_max, "%d:", mv2_core_map[k]);
             k = (i % num_sockets == 0) ?
                     (k + num_pu_per_socket) % size :
                     (k + num_pu_per_socket + hw_threads_per_core) % size;
@@ -3315,10 +3344,10 @@
         /* NUMA mapping: Bind consecutive MPI ranks to different NUMA domains in
          * round-robin fashion. */
         for (i = 0; i < local_procs; i++) {
-            j += snprintf (mapping+j, _POSIX2_LINE_MAX, "%d,", 
+            j += snprintfoff(mapping, j, s_cpu_mapping_line_max, "%d,", 
                                mv2_core_map_per_numa[node_base_pu+node_offset]);
             mapping [--j] = '\0';
-            j += snprintf (mapping+j, _POSIX2_LINE_MAX, ":");
+            j += snprintfoff(mapping, j, s_cpu_mapping_line_max, ":");
             node_base_pu = (node_base_pu + num_pu_per_numanode) % size;
             node_offset = (node_base_pu == 0) ? 
                             (node_offset + ((hw_threads_per_core > 0) ? hw_threads_per_core : 1)) : 
@@ -3326,10 +3355,21 @@
         }
     }
 
+    /* make sure we did not overflow mapping[] */
+    if (j >= s_cpu_mapping_line_max) {
+        fprintf(stderr, "ERROR: mapping[] ran out of space!\n");
+        /* cleanup */
+        MPIU_Free(mv2_core_map);
+        MPIU_Free(mv2_core_map_per_numa);
+        hwloc_bitmap_free(allowed_cpuset); 
+        return MPI_ERR_OTHER;
+    }
+
     /* copy the generated mapping string to final mapping*/
+    /* note: we discard last char of mapping[] (should be an extra ':') */
     s_cpu_mapping = (char *) MPIU_Malloc (sizeof (char) * j);
     strncpy (s_cpu_mapping, mapping, j);
-    s_cpu_mapping[j-1] = '\0';
+    s_cpu_mapping[j-1] = '\0';    /* was the extra ':' we discard */
 
     if (MPIDI_Process.my_pg_rank == 0) {
         PRINT_DEBUG(DEBUG_INIT_verbose>0, "num_physical_cores_per_socket %d, mapping: %s\n",