https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127581
Bug ID: 127581
Summary: -Wanalyzer-fd-leak false positive for dup2
Product: gcc
Version: 16.2.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: analyzer
Assignee: dmalcolm at gcc dot gnu.org
Reporter: norbert.eicker at gmx dot de
Target Milestone: ---
Created attachment 65673
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=65673&action=edit
Simple reproducer
When compiling the reproducer I get:
> gcc-16 -fanalyzer -c reproducer.c
reproducer.c: In function ‘redirect’:
reproducer.c:9:8: warning: leak of file descriptor ‘dup2(fd, 1)’ [CWE-775]
[-Wanalyzer-fd-leak]
9 | if (dup2(fd, STDOUT_FILENO) == -1) return -1;
| ^
‘redirect’: events 1-3
7 | if (fd < 0) return -1;
| ^
| |
| (1) following ‘false’ branch (when ‘fd >= 0’)... ─>─┐
| │
| │
|┌───────────────────────────────────────────────────────────┘
8 |│
9 |│ if (dup2(fd, STDOUT_FILENO) == -1) return -1;
|│ ~~~~~~~~~~~~~~~~~~~~~~~
|│ |
|└───────>(2) ...to here
| (3) opened here
‘redirect’: event 4
9 | if (dup2(fd, STDOUT_FILENO) == -1) return -1;
| ^
| |
| (4) ⚠ ‘dup2(fd, 1)’ leaks here; was opened at (3)
I guess this is a false positive for various reasons:
- the new file descriptor is not only returned by dup2() but also forced to be
the second argument -- unless -1 is returned to indicate a failure
- if the second argument is in the range [0,2] (i.e. one of STDIN_FILENO,
STDOUT_FILENO, STDERR_FILENO) the new file descriptor shall not assumed to be
leaked
If the second argument has a value different from [0,2], it shall from now on
take the role of a file descriptor. I.e. at the end of the scope of the second
argument a leak of a file descriptor shall be indicated unless it was close()ed
before
