[linux] fix read_id_stat to handle unbalanced parens in process name

Fix issue #359 by changing the parse logic: find the first '(' and the last
')', the string between is the process name, the rest are the remaining
fields.

The old code tried to balance parentheses, which failed when the process
name contained unbalanced parens (e.g. 'nc ('), causing lsof to silently
skip the process.

Also add a test case with a helper that sets its comm.

Author: jiegec

.gitignore

@@ -138,6 +138,7 @@ lib/dialects/linux/tests/pipe
 lib/dialects/linux/tests/pty
 lib/dialects/linux/tests/ux
 lib/dialects/linux/tests/mmap
+lib/dialects/linux/tests/unbal_parens
 
 # automake
 Makefile.in

Makefile.am

@@ -164,7 +164,8 @@ LINUX_TESTS = lib/dialects/linux/tests/case-10-mqueue.bash \
 	lib/dialects/linux/tests/case-20-pipe-no-close-endpoint.bash \
 	lib/dialects/linux/tests/case-20-pty-endpoint.bash \
 	lib/dialects/linux/tests/case-20-ux-socket-endpoint.bash \
-	lib/dialects/linux/tests/case-20-ux-socket-endpoint-unaccepted.bash
+	lib/dialects/linux/tests/case-20-ux-socket-endpoint-unaccepted.bash \
+	lib/dialects/linux/tests/case-20-unbal-parens.bash
 EXTRA_DIST += $(LINUX_TESTS) lib/dialects/linux/tests/Makefile lib/dialects/linux/tests/case-00-linux-hello.bash
 if LINUX
 check_PROGRAMS += lib/dialects/linux/tests/epoll \
@@ -176,7 +177,8 @@ check_PROGRAMS += lib/dialects/linux/tests/epoll \
 	lib/dialects/linux/tests/pidfd \
 	lib/dialects/linux/tests/pipe \
 	lib/dialects/linux/tests/pty \
-	lib/dialects/linux/tests/ux
+	lib/dialects/linux/tests/ux \
+	lib/dialects/linux/tests/unbal_parens
 lib_dialects_linux_tests_mq_fork_LDADD = -lrt
 lib_dialects_linux_tests_mq_open_LDADD = -lrt
 

lib/dialects/linux/dproc.c

@@ -1666,120 +1666,109 @@ static int read_id_stat(struct lsof_context *ctx, /* context */
                         int *pgid) /* returned process group ID for PID
                                     * type */
 {
-    char buf[MAXPATHLEN], *cp, *cp1, **fp;
-    int ch, cx, es, pc;
+    char *cp, *cp1, *cp2, **fp;
+    int len = 0, n;
     char *cbf = (char *)NULL;
-    MALLOC_S cbfa = 0;
+    int cbfa = 0;
     FILE *fs;
+    int ret = 0;
     /*
      * Open the stat file path, assign a page size buffer to its stream,
-     * and read the file's first line.
+     * and read the entire file.
      */
     if (!(fs = open_proc_stream(ctx, p, "r", 0)))
         return (-1);
-    if (!(cp = fgets(buf, sizeof(buf), fs))) {
-
-    read_id_stat_exit:
-
-        CLEAN(cbf);
-        (void)fclose(fs);
-        return (-1);
+    for (;;) {
+        if (len + 1 >= cbfa) {
+            if (grow_array((void **)&cbf, &cbfa, sizeof(char), 256) != 0)
+                goto read_id_stat_exit;
+        }
+        n = fread(cbf + len, 1, cbfa - len - 1, fs);
+        if (n == 0) {
+            if (ferror(fs))
+                goto read_id_stat_exit;
+            break;
+        }
+        len += n;
     }
+    if (!cbf || !*cbf)
+        goto read_id_stat_exit;
+    cbf[len] = '\0';
+    (void)fclose(fs);
+    fs = NULL;
     /*
      * Skip to the first field, and make sure it is a matching ID.
      */
-    cp1 = cp;
+    cp = cbf;
     while (*cp && (*cp != ' ') && (*cp != '\t'))
         cp++;
     if (*cp)
         *cp = '\0';
-    if (atoi(cp1) != id)
+    if (atoi(cbf) != id)
         goto read_id_stat_exit;
     /*
      * The second field should contain the command, enclosed in parentheses.
-     * If it also has embedded '\n' characters, replace them with '?'
-     * characters, accumulating command characters until a closing parentheses
-     * appears.
-     *
+     * Find the first '(' and the last ')' in the file.  The string between
+     * them is the command name; the rest after the last ')' are the
+     * remaining fields.
      */
     for (++cp; *cp && (*cp == ' '); cp++)
         ;
     if (!cp || (*cp != '('))
         goto read_id_stat_exit;
-    cp++;
-    pc = 1; /* start the parenthesis balance count at 1 */
-
+    cp1 = ++cp;
+    cp2 = strrchr(cp, ')');
+    if (!cp2)
+        goto read_id_stat_exit;
     /*
-     * Enter the command characters safely.  Supply them from the initial read
-     * of the stat file line, a '\n' if the initial read didn't yield a ')'
-     * command closure, or by reading the rest of the command a character at
-     * a time from the stat file.  Count embedded '(' characters and balance
-     * them with embedded ')' characters.  The opening '(' starts the balance
-     * count at one.
+     * Replace embedded newlines in the command with '?'.
      */
-    for (cx = es = 0;;) {
-        if (!es)
-            ch = *cp++;
-        else {
-            if ((ch = fgetc(fs)) == EOF)
-                goto read_id_stat_exit;
-        }
-        if (ch == '(') /* a '(' advances the balance count */
-            pc++;
-        if (ch == ')') {
-
-            /*
-             * Balance parentheses when a closure is encountered.  When
-             * they are balanced, this is the end of the command.
-             */
-            pc--;
-            if (!pc)
-                break;
-        }
-        if ((cx + 2) > cbfa)
-            cbfa = alloc_cbf(ctx, (cx + 2), &cbf, cbfa);
-        cbf[cx] = ch;
-        cx++;
-        cbf[cx] = '\0';
-        if (!es && !*cp)
-            es = 1; /* Switch to fgetc() when a '\0' appears. */
+    for (cp = cp1; cp < cp2; cp++) {
+        if (*cp == '\n')
+            *cp = '?';
     }
-    *cmd = cbf;
-    cbf = NULL;
+    *cp2 = '\0';
+    *cmd = mkstrcpy(cp1, NULL);
+    if (!*cmd)
+        goto read_id_stat_exit;
     /*
-     * Read the remainder of the stat line if it was necessary to read command
-     * characters individually from the stat file.
-     *
-     * Separate the reminder into fields.
+     * Separate the remainder into fields.
      */
-    if (es)
-        cp = fgets(buf, sizeof(buf), fs);
-    (void)fclose(fs);
+    cp = cp2 + 1;
+    while (*cp && (*cp == ' '))
+        cp++;
     if (!cp || !*cp)
-        return (-1);
+        goto read_id_stat_exit;
     if (get_fields(ctx, cp, (char *)NULL, &fp, (int *)NULL, 0) < 3)
-        return (-1);
+        goto read_id_stat_exit;
     /*
-     * Convert and return parent process (fourth field) and process group (fifth
-     * field) IDs.
+     * Convert and return parent process (fourth field) and process group
+     * (fifth field) IDs.
      */
     if (fp[1] && *fp[1])
         *ppid = atoi(fp[1]);
     else
-        return (-1);
+        goto read_id_stat_exit;
     if (fp[2] && *fp[2])
         *pgid = atoi(fp[2]);
     else
-        return (-1);
+        goto read_id_stat_exit;
     /*
      * Check the state in the third field.  If it is 'Z', return that
      * indication.
      */
     if (fp[0] && !strcmp(fp[0], "Z"))
-        return (1);
+        ret = 1;
     else if (fp[0] && !strcmp(fp[0], "T"))
-        return (2);
-    return (0);
+        ret = 2;
+    CLEAN(cbf);
+    return ret;
+
+read_id_stat_exit:
+    CLEAN(cbf);
+    if (fs)
+        (void)fclose(fs);
+    return (-1);
 }
 
 /*

lib/dialects/linux/tests/Makefile

@@ -8,6 +8,7 @@ HELPERS = \
 	pty \
 	ux \
 	mmap \
+	unbal_parens \
 	\
 	open_with_flags \
 	\

lib/dialects/linux/tests/case-20-unbal-parens.bash

@@ -0,0 +1,70 @@
+#!/bin/bash
+source tests/common.bash
+
+TARGET=$tcasedir/unbal_parens
+
+r=0
+
+test_pattern() {
+    local pattern="$1"
+    local expected="$2"
+    local pid
+    local output
+    local comm
+    
+    $TARGET "$pattern" | {
+        read pid
+        if [ -z "$pid" ]; then
+            echo "FAIL: Failed to start helper process"
+            exit 1
+        fi
+        
+        # Give /proc time to update
+        sleep 0.2
+        
+        # Check that lsof finds the process
+        output=$($lsof -p $pid 2>/dev/null)
+        if ! grep -Fq "$pid" <<<"$output"; then
+            echo "FAIL: lsof did not find process with comm='$pattern'"
+            echo "pid: $pid"
+            kill $pid 2>/dev/null
+            exit 1
+        fi
+        
+        # Check that the reported command name matches expected
+        comm=$($lsof -p $pid -Fc 2>/dev/null | grep '^c' | head -1 | sed 's/^c//')
+        if [ "$comm" != "$expected" ]; then
+            echo "FAIL: comm mismatch for pattern='$pattern'"
+            echo "  expected: '$expected'"
+            echo "  got:      '$comm'"
+            kill $pid 2>/dev/null
+            exit 1
+        fi
+        
+        echo "PASS: pattern='$pattern' comm='$comm'"
+        kill $pid 2>/dev/null
+        exit 0
+    }
+    
+    if [ $? -ne 0 ]; then
+        r=1
+    fi
+}
+
+{
+    # Test various unbalanced parenthesis patterns
+    test_pattern "test(" "test(" || r=1
+    test_pattern "test)" "test)" || r=1
+    test_pattern "a(b)c" "a(b)c" || r=1
+    test_pattern "a(b)c)d)e" "a(b)c)d)e" || r=1
+    test_pattern "((" "((" || r=1
+    test_pattern "))" "))" || r=1
+    test_pattern "((()))" "((()))" || r=1
+    test_pattern "(((" "(((" || r=1
+    test_pattern ")))" ")))" || r=1
+
+    # Test newline in comm is replaced with '?'
+    test_pattern "NEWLINE" "a?b" || r=1
+    
+    exit $r
+} >> $report 2>&1

lib/dialects/linux/tests/unbal_parens.c

@@ -0,0 +1,22 @@
+#define _GNU_SOURCE
+#include <sys/prctl.h>
+#include <stdio.h>
+#include <string.h>
+#include <unistd.h>
+
+int main(int argc, char **argv) {
+    const char *name = (argc > 1) ? argv[1] : "test(";
+    
+    if (strcmp(name, "NEWLINE") == 0) {
+        name = "a\nb";
+    }
+    
+    if (prctl(PR_SET_NAME, name) < 0) {
+        perror("prctl");
+        return 1;
+    }
+    printf("%d\n", getpid());
+    fflush(stdout);
+    pause();
+    return 0;
+}