On Mon, 8 Jun 2026 at 20:22, <[email protected]> wrote:
> From: Sunil Dora <[email protected]>
>
> Many kernels enforce a per-string length limit on argv and envp
> strings passed to execve(). On Linux, MAX_ARG_STRLEN limits each
> string to 32 * PAGE_SIZE (~128KB); Windows limits individual
> environment variables to 32767 characters. When the gcc driver
> composes a COLLECT_GCC_OPTIONS value that exceeds such a limit,
> execve() of cc1, collect2 or lto-wrapper fails with E2BIG.
>
> When the assembled value would exceed COLLECT_GCC_OPTIONS_MAX_LENGTH
> (default 1024, host-overridable via defaults.h), the driver writes
> the option list to a temporary response file via writeargv() and
> exports "COLLECT_GCC_OPTIONS=@<path>" instead. collect2,
> lto-wrapper and lto-plugin transparently expand the @file form using
> existing expandargv() infrastructure, so the change is invisible to
> normal builds.
>
> Bootstrapped and regression tested on x86_64-pc-linux-gnu.
>
> PR driver/111527
>
> gcc/ChangeLog:
>
> * defaults.h (COLLECT_GCC_OPTIONS_MAX_LENGTH): New macro.
> * collect-utils.cc (requote_one_arg): New function.
> (read_collect_gcc_options): New function.
> * collect-utils.h (read_collect_gcc_options): Declare.
> * collect2.cc (main): Use read_collect_gcc_options instead
> of getenv.
> * doc/invoke.texi (Environment Variables): Document the
> @file form of COLLECT_GCC_OPTIONS.
> * gcc.cc (xsetenv_collect_gcc_options): New function.
> (set_collect_gcc_options): Use xsetenv_collect_gcc_options.
> * lto-wrapper.cc (run_gcc): Use read_collect_gcc_options
> instead of getenv.
>
> gcc/testsuite/ChangeLog:
>
> * gcc.misc-tests/pr111527.exp: New test.
>
> lto-plugin/ChangeLog:
>
> * lto-plugin.c (append_quoted_to_buf): New function.
> (read_collect_gcc_options): New function.
> (onload): Use read_collect_gcc_options instead of getenv.
>
> Signed-off-by: Sunil Dora <[email protected]>
> ---
> v1 -> v2:
> - Make spill unconditional, drop configure.ac/configure/config.in
> changes
> - Add COLLECT_GCC_OPTIONS_MAX_LENGTH (default 1024) to defaults.h
> - Rename to xsetenv_collect_gcc_options(), set env internally
> - Fix strchr comparison to use != nullptr
> - Fix NULL to nullptr in collect-utils.cc
> - Use XRESIZEVAR in lto-plugin.c
> - Update invoke.texi to be platform neutral
> - Remove Linux-only guard from test
>
> gcc/collect-utils.cc | 58 +++++++++++++++++
> gcc/collect-utils.h | 2 +-
> gcc/collect2.cc | 6 +-
> gcc/defaults.h | 6 ++
> gcc/doc/invoke.texi | 8 +++
> gcc/gcc.cc | 48 +++++++++++++-
> gcc/lto-wrapper.cc | 4 +-
> gcc/testsuite/gcc.misc-tests/pr111527.exp | 78 +++++++++++++++++++++++
> lto-plugin/lto-plugin.c | 68 +++++++++++++++++++-
> 9 files changed, 270 insertions(+), 8 deletions(-)
> create mode 100644 gcc/testsuite/gcc.misc-tests/pr111527.exp
>
> diff --git a/gcc/collect-utils.cc b/gcc/collect-utils.cc
> index ad37e7a4905..b623088c5c5 100644
> --- a/gcc/collect-utils.cc
> +++ b/gcc/collect-utils.cc
> @@ -269,3 +269,61 @@ utils_cleanup (bool from_signal)
>
> tool_cleanup (from_signal);
> }
> +
> +/* Append OPT to OB in COLLECT_GCC_OPTIONS shell-quoted form.
> + *FIRST_P controls the leading separator and is updated. */
> +
> +static void
> +requote_one_arg (struct obstack *ob, bool *first_p, const char *opt)
> +{
> + const char *p, *q = opt;
> + if (!*first_p)
> + obstack_grow (ob, " ", 1);
> + obstack_grow (ob, "'", 1);
> + while ((p = strchr (q, '\'')) != nullptr)
> + {
> + obstack_grow (ob, q, p - q);
> + obstack_grow (ob, "'\\''", 4);
> + q = ++p;
> + }
> + obstack_grow (ob, q, strlen (q));
> + obstack_grow (ob, "'", 1);
> + *first_p = false;
> +}
> +
> +/* Return COLLECT_GCC_OPTIONS, expanding an @file reference if present.
> + Returns nullptr if unset. Result is owned by an internal cache. */
> +
> +const char *
> +read_collect_gcc_options (void)
> +{
> + static char *cached;
> +
> + const char *raw = getenv ("COLLECT_GCC_OPTIONS");
> + if (raw == nullptr)
> + return nullptr;
> + if (raw[0] != '@')
> + return raw;
> +
> + /* Feed "@file" to expandargv via a minimal argv. */
> + int argc = 2;
> + char **argv = XCNEWVEC (char *, 3);
> + argv[0] = xstrdup (tool_name);
> + argv[1] = xstrdup (raw);
> + argv[2] = nullptr;
> + expandargv (&argc, &argv);
> +
> + struct obstack ob;
> + obstack_init (&ob);
> + bool first = true;
> + for (int i = 1; i < argc; i++)
> + requote_one_arg (&ob, &first, argv[i]);
> + obstack_1grow (&ob, '\0');
> + char *result = xstrdup ((const char *) XOBFINISH (&ob, char *));
> + obstack_free (&ob, nullptr);
> + freeargv (argv);
> +
> + free (cached);
> + cached = result;
> + return cached;
> +}
>
Hello,
I just have a question, there seems to be code duplication with
read_collect_gcc_options in this file and lto-plugin.c. Have you considered
moving them to a header file and including that?
Thanks,
Shivam
diff --git a/gcc/collect-utils.h b/gcc/collect-utils.h
> index 3ed80271fdb..50b4bba9a0b 100644
> --- a/gcc/collect-utils.h
> +++ b/gcc/collect-utils.h
> @@ -33,7 +33,7 @@ extern int collect_wait (const char *, struct pex_obj *);
> extern void do_wait (const char *, struct pex_obj *);
> extern void fork_execute (const char *, char **, bool, const char *);
> extern void utils_cleanup (bool);
> -
> +extern const char *read_collect_gcc_options (void);
>
> extern bool debug;
> extern bool verbose;
> diff --git a/gcc/collect2.cc b/gcc/collect2.cc
> index 6985e0b4d15..c7b0ad4321a 100644
> --- a/gcc/collect2.cc
> +++ b/gcc/collect2.cc
> @@ -1008,7 +1008,7 @@ main (int argc, char **argv)
> /* Now pick up any flags we want early from COLLECT_GCC_OPTIONS
> The LTO options are passed here as are other options that might
> be unsuitable for ld (e.g. -save-temps). */
> - p = getenv ("COLLECT_GCC_OPTIONS");
> + p = read_collect_gcc_options ();
> while (p && *p)
> {
> const char *q = extract_string (&p);
> @@ -1206,7 +1206,7 @@ main (int argc, char **argv)
> AIX support needs to know if -shared has been specified before
> parsing commandline arguments. */
>
> - p = getenv ("COLLECT_GCC_OPTIONS");
> + p = read_collect_gcc_options ();
> while (p && *p)
> {
> const char *q = extract_string (&p);
> @@ -1599,7 +1599,7 @@ main (int argc, char **argv)
> fprintf (stderr, "o_file = %s\n",
> (o_file ? o_file : "not found"));
>
> - ptr = getenv ("COLLECT_GCC_OPTIONS");
> + ptr = read_collect_gcc_options ();
> if (ptr)
> fprintf (stderr, "COLLECT_GCC_OPTIONS = %s\n", ptr);
>
> diff --git a/gcc/defaults.h b/gcc/defaults.h
> index 87f710697f0..1b579339451 100644
> --- a/gcc/defaults.h
> +++ b/gcc/defaults.h
> @@ -1463,4 +1463,10 @@ see the files COPYING3 and COPYING.RUNTIME
> respectively. If not, see
> typedef TARGET_UNIT target_unit;
> #endif
>
> +/* Maximum length of COLLECT_GCC_OPTIONS before the driver spills it
> + to a response file. Hosts with tighter limits may override this. */
> +#ifndef COLLECT_GCC_OPTIONS_MAX_LENGTH
> +#define COLLECT_GCC_OPTIONS_MAX_LENGTH 1024
> +#endif
> +
> #endif /* ! GCC_DEFAULTS_H */
> diff --git a/gcc/doc/invoke.texi b/gcc/doc/invoke.texi
> index 339d1d2c97a..d97186715c2 100644
> --- a/gcc/doc/invoke.texi
> +++ b/gcc/doc/invoke.texi
> @@ -37775,6 +37775,14 @@ set and it cannot connect to it.
>
> This feature is experimental and subject to change or removal without
> notice.
> +
> +@vindex COLLECT_GCC_OPTIONS
> +@item COLLECT_GCC_OPTIONS
> +Set by the driver and read by @command{collect2}, @command{lto-wrapper},
> +and the LTO linker plugin to pass the driver's option list. If
> +the list would exceed @code{COLLECT_GCC_OPTIONS_MAX_LENGTH} bytes,
> +the driver writes it to a temporary file and sets this variable to
> +@samp{@@@var{path}} instead.
> @end table
>
> @noindent
> diff --git a/gcc/gcc.cc b/gcc/gcc.cc
> index 0ed2cd96be1..b03e84dd9f5 100644
> --- a/gcc/gcc.cc
> +++ b/gcc/gcc.cc
> @@ -5659,6 +5659,52 @@ process_command (unsigned int decoded_options_count,
> infiles[n_infiles].name = 0;
> }
>
> +/* Set COLLECT_GCC_OPTIONS in the environment. If the value would
> + exceed COLLECT_GCC_OPTIONS_MAX_LENGTH, spill it to a temporary
> + response file and set the variable to @<path> instead. */
> +
> +static void
> +xsetenv_collect_gcc_options (char *string)
> +{
> + if (strlen (string) <= COLLECT_GCC_OPTIONS_MAX_LENGTH)
> + {
> + xputenv (string);
> + return;
> + }
> +
> + static const char prefix[] = "COLLECT_GCC_OPTIONS=";
> + gcc_assert (startswith (string, prefix));
> +
> + /* parse_options_from_collect_gcc_options expects argc to start
> + at 1, so push a placeholder argv[0]. */
> + struct obstack argv_obstack;
> + obstack_init (&argv_obstack);
> + obstack_ptr_grow (&argv_obstack, const_cast<char *> (progname));
> + int argc;
> + parse_options_from_collect_gcc_options (string + sizeof (prefix) - 1,
> + &argv_obstack, &argc);
> + char **argv = XOBFINISH (&argv_obstack, char **);
> +
> + char *temp_file = make_temp_file ("");
> + FILE *f = fopen (temp_file, "wb");
> + if (f == nullptr)
> + fatal_error (input_location,
> + "cannot open response file %qs: %m", temp_file);
> + /* writeargv walks until NULL; skip our placeholder argv[0]. */
> + if (writeargv (argv + 1, f) != 0)
> + fatal_error (input_location,
> + "cannot write response file %qs: %m", temp_file);
> + if (fclose (f) != 0)
> + fatal_error (input_location,
> + "cannot close response file %qs: %m", temp_file);
> +
> + char *env_val = concat (prefix, "@", temp_file, nullptr);
> + /* Delete on both success and failure unless -save-temps. */
> + record_temp_file (temp_file, !save_temps_flag, !save_temps_flag);
> + obstack_free (&argv_obstack, nullptr);
> + xputenv (env_val);
> +}
> +
> /* Store switches not filtered out by %<S in spec in COLLECT_GCC_OPTIONS
> and place that in the environment. */
>
> @@ -5737,7 +5783,7 @@ set_collect_gcc_options (void)
> }
>
> obstack_grow (&collect_obstack, "\0", 1);
> - xputenv (XOBFINISH (&collect_obstack, char *));
> + xsetenv_collect_gcc_options (XOBFINISH (&collect_obstack, char *));
> }
>
> /* Process a spec string, accumulating and running commands. */
> diff --git a/gcc/lto-wrapper.cc b/gcc/lto-wrapper.cc
> index 9610fa5b7a8..9c90377099d 100644
> --- a/gcc/lto-wrapper.cc
> +++ b/gcc/lto-wrapper.cc
> @@ -1418,7 +1418,7 @@ run_gcc (unsigned argc, char *argv[])
> char *list_option_full = NULL;
> const char *linker_output = NULL;
> const char *collect_gcc;
> - char *collect_gcc_options;
> + const char *collect_gcc_options;
> int parallel = 0;
> int jobserver = 0;
> bool jobserver_requested = false;
> @@ -1452,7 +1452,7 @@ run_gcc (unsigned argc, char *argv[])
> if (!collect_gcc)
> fatal_error (input_location,
> "environment variable %<COLLECT_GCC%> must be set");
> - collect_gcc_options = getenv ("COLLECT_GCC_OPTIONS");
> + collect_gcc_options = read_collect_gcc_options ();
> if (!collect_gcc_options)
> fatal_error (input_location,
> "environment variable %<COLLECT_GCC_OPTIONS%> must be
> set");
> diff --git a/gcc/testsuite/gcc.misc-tests/pr111527.exp
> b/gcc/testsuite/gcc.misc-tests/pr111527.exp
> new file mode 100644
> index 00000000000..bf335871996
> --- /dev/null
> +++ b/gcc/testsuite/gcc.misc-tests/pr111527.exp
> @@ -0,0 +1,78 @@
> +# Copyright (C) 2026 Free Software Foundation, Inc.
> +# This program is free software; you can redistribute it and/or modify
> +# it under the terms of the GNU General Public License as published by
> +# the Free Software Foundation; either version 3 of the License, or
> +# (at your option) any later version.
> +#
> +# This program is distributed in the hope that it will be useful,
> +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> +# GNU General Public License for more details.
> +#
> +# You should have received a copy of the GNU General Public License
> +# along with GCC; see the file COPYING3. If not see
> +# <http://www.gnu.org/licenses/>.
> +
> +# PR driver/111527 - very long COLLECT_GCC_OPTIONS exceeding
> +# COLLECT_GCC_OPTIONS_MAX_LENGTH. The driver should spill
> +# the option list to a response file when it would not fit.
> +
> +load_lib gcc-defs.exp
> +load_lib target-supports.exp
> +
> +if { [is_remote host] } {
> + return
> +}
> +
> +global GCC_UNDER_TEST
> +if { ![info exists GCC_UNDER_TEST] } {
> + set GCC_UNDER_TEST [find_gcc]
> +}
> +
> +# Use @file rather than additional_flags: DejaGnu collapses very
> +# long flag strings. The driver still expands @file into argv and
> +# builds COLLECT_GCC_OPTIONS from it, exercising the spill path.
> +set work_dir [pwd]
> +set src [file join $work_dir "pr111527.c"]
> +set rsp [file join $work_dir "pr111527.rsp"]
> +set obj [file join $work_dir "pr111527.o"]
> +
> +set f [open $src w]
> +puts $f "int main (void) { return 0; }"
> +close $f
> +
> +set f [open $rsp w]
> +for { set i 0 } { $i < 50 } { incr i } {
> + puts $f "-DPR111527_PADDING_$i=1"
> +}
> +close $f
> +
> +# (1) Build must succeed.
> +set cmd "$GCC_UNDER_TEST -c $src -o $obj @$rsp"
> +verbose -log "Test 1: $cmd"
> +set status [remote_exec host $cmd]
> +set rc [lindex $status 0]
> +set out [lindex $status 1]
> +if { $rc == 0 } {
> + pass "PR111527: build with very long COLLECT_GCC_OPTIONS"
> +} else {
> + fail "PR111527: build with very long COLLECT_GCC_OPTIONS"
> + verbose -log "compiler output: $out"
> +}
> +file delete -force $obj
> +
> +# (2) Confirm spill path engaged.
> +set cmd "$GCC_UNDER_TEST -v -c $src -o $obj @$rsp"
> +verbose -log "Test 2: $cmd"
> +set status [remote_exec host $cmd]
> +set out [lindex $status 1]
> +if { [regexp {COLLECT_GCC_OPTIONS=@[^[:space:]]+} $out] } {
> + pass "PR111527: driver spilled to @file"
> +} else {
> + fail "PR111527: driver spilled to @file"
> + verbose -log "compiler output: $out"
> +}
> +
> +file delete -force $obj
> +file delete -force $rsp
> +file delete -force $src
> diff --git a/lto-plugin/lto-plugin.c b/lto-plugin/lto-plugin.c
> index ffa4fe1552f..569ebe539d1 100644
> --- a/lto-plugin/lto-plugin.c
> +++ b/lto-plugin/lto-plugin.c
> @@ -1506,6 +1506,72 @@ negotiate_api_version (void)
> }
> }
>
> +/* Append ARG to *BUFP in COLLECT_GCC_OPTIONS shell-quoted form
> + (each arg in single quotes, embedded ' becomes '\''), growing
> + the buffer as needed. */
> +
> +static void
> +append_quoted_to_buf (char **bufp, size_t *posp, size_t *capp,
> + const char *arg, bool first)
> +{
> + /* 4 bytes per char worst case, plus quotes and separator. */
> + size_t need = strlen (arg) * 4 + 4;
> + if (*posp + need + 1 > *capp)
> + {
> + *capp = (*posp + need + 1) * 2;
> + *bufp = XRESIZEVAR (char, *bufp, *capp);
> + }
> + char *p = *bufp + *posp;
> + if (!first)
> + *p++ = ' ';
> + *p++ = '\'';
> + for (const char *q = arg; *q; q++)
> + {
> + if (*q == '\'')
> + {
> + *p++ = '\''; *p++ = '\\'; *p++ = '\''; *p++ = '\'';
> + }
> + else
> + *p++ = *q;
> + }
> + *p++ = '\'';
> + *posp = p - *bufp;
> +}
> +
> +/* Return COLLECT_GCC_OPTIONS, expanding @file into the shell-quoted
> + text expected by onload. Result owned by an internal cache. */
> +
> +static const char *
> +read_collect_gcc_options (void)
> +{
> + static char *cached;
> +
> + const char *raw = getenv ("COLLECT_GCC_OPTIONS");
> + if (raw == NULL)
> + return NULL;
> + if (raw[0] != '@')
> + return raw;
> +
> + int argc = 2;
> + char **argv = (char **) xcalloc (3, sizeof (char *));
> + argv[0] = xstrdup ("lto-plugin");
> + argv[1] = xstrdup (raw);
> + argv[2] = NULL;
> + expandargv (&argc, &argv);
> +
> + size_t cap = 256, pos = 0;
> + char *buf = (char *) xmalloc (cap);
> + buf[0] = '\0';
> + for (int i = 1; i < argc; i++)
> + append_quoted_to_buf (&buf, &pos, &cap, argv[i], i == 1);
> + buf[pos] = '\0';
> + freeargv (argv);
> +
> + free (cached);
> + cached = buf;
> + return cached;
> +}
> +
> /* Called by a linker after loading the plugin. TV is the transfer
> vector. */
>
> enum ld_plugin_status
> @@ -1617,7 +1683,7 @@ onload (struct ld_plugin_tv *tv)
> "could not register the all_symbols_read callback");
> }
>
> - char *collect_gcc_options = getenv ("COLLECT_GCC_OPTIONS");
> + const char *collect_gcc_options = read_collect_gcc_options ();
> if (collect_gcc_options)
> {
> /* Support -fno-use-linker-plugin by failing to load the plugin
> --
> 2.43.0
>
>
