[LLVM][Support] check for error return from dladdr.

[jeremyd2019]

In case of an error, the DL_info struct may have been left uninitialized, so it is not safe to use its members.

In one error case, initialize dli_sname to nullptr explicitly, so that the later check against nullptr is guaranteed to be safe.

[llvmbot]

@llvm/pr-subscribers-llvm-support

Author: None (jeremyd2019)

Changes

In case of an error, the DL_info struct may have been left uninitialized, so it is not safe to use its members.

In one error case, initialize dli_sname to nullptr explicitly, so that the later check against nullptr is guaranteed to be safe.

---

Full diff: https://github.com/llvm/llvm-project/pull/138369.diff

1 Files Affected:

• (modified) llvm/lib/Support/Unix/Signals.inc (+20-13)

diff --git a/llvm/lib/Support/Unix/Signals.inc b/llvm/lib/Support/Unix/Signals.inc
index 691e1014f18e8..b2bc76a1121d4 100644
--- a/llvm/lib/Support/Unix/Signals.inc
+++ b/llvm/lib/Support/Unix/Signals.inc
@@ -826,14 +826,17 @@ void llvm::sys::PrintStackTrace(raw_ostream &OS, int Depth) {
   int width = 0;
   for (int i = 0; i < depth; ++i) {
     Dl_info dlinfo;
-    dladdr(StackTrace[i], &dlinfo);
-    const char *name = strrchr(dlinfo.dli_fname, '/');
-
     int nwidth;
-    if (!name)
-      nwidth = strlen(dlinfo.dli_fname);
-    else
-      nwidth = strlen(name) - 1;
+    if (dladdr(StackTrace[i], &dlinfo) == 0) {
+      nwidth = 7; // "(error)"
+    } else {
+      const char *name = strrchr(dlinfo.dli_fname, '/');
+
+      if (!name)
+        nwidth = strlen(dlinfo.dli_fname);
+      else
+        nwidth = strlen(name) - 1;
+    }
 
     if (nwidth > width)
       width = nwidth;
@@ -841,15 +844,19 @@ void llvm::sys::PrintStackTrace(raw_ostream &OS, int Depth) {
 
   for (int i = 0; i < depth; ++i) {
     Dl_info dlinfo;
-    dladdr(StackTrace[i], &dlinfo);
 
     OS << format("%-2d", i);
 
-    const char *name = strrchr(dlinfo.dli_fname, '/');
-    if (!name)
-      OS << format(" %-*s", width, static_cast<const char *>(dlinfo.dli_fname));
-    else
-      OS << format(" %-*s", width, name + 1);
+    if (dladdr(StackTrace[i], &dlinfo) == 0) {
+      OS << format(" %-*s", width, "(error)");
+      dlinfo.dli_sname = nullptr;
+    } else {
+      const char *name = strrchr(dlinfo.dli_fname, '/');
+      if (!name)
+        OS << format(" %-*s", width, static_cast<const char *>(dlinfo.dli_fname));
+      else
+        OS << format(" %-*s", width, name + 1);
+    }
 
     OS << format(" %#0*lx", (int)(sizeof(void *) * 2) + 2,
                  (unsigned long)StackTrace[i]);

[jeremyd2019]

As promised to @mstorsjo in #138117 (comment).

On success, these functions return a nonzero value.

[github-actions]

✅ With the latest revision this PR passed the C/C++ code formatter.

[mstorsjo]

LGTM

[jeremyd2019]

do I need to do something to get CI to run again? It doesn't seem to have failed due to anything in my control

[mstorsjo]

do I need to do something to get CI to run again? It doesn't seem to have failed due to anything in my control

Yeah the failure seems to be unrelated. Some people have access to trigger retries on builkite, but those jobs should be switched over to github actions quite soon (where it should be easier to restart them), so that's mostly a moot point.

Anyway, I didn't mean to hold this one up, I just seem to have lost track of it and forgotten to merge it.

[mstorsjo]

Btw, unrelated to this particular patch; it seems like you don't have a realname set on your github account. When merging patches via the "squash and merge" method, the patches get rewritten to use the name from the account, not the (proper) name as it was in the individual git comits - so the patches end up with just Author: jeremyd2019 <[email protected]>.