[PATCH v1] Fix out-of-bounds access in pg_bsd_indent's parser stack

Started by Vadim Shakirov22 days ago4 messageshackers
Jump to latest
#1Vadim Shakirov
vadimsakirov5@gmail.com

parser_state stores parser state in p_stack, il, and cstk, which
share the tos index. The existing overflow check used p_stack, the
largest array, allowing out-of-bounds accesses to cstk and il. Size
all three arrays equally.

Also widen the overflow check by one element: the lbrace case pushes
two entries (lbrace and then stmt), so there must be room for two
more elements when parse() is entered.
---
src/tools/pg_bsd_indent/indent_globs.h | 4 ++--
src/tools/pg_bsd_indent/parse.c | 2 +-
2 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/src/tools/pg_bsd_indent/indent_globs.h b/src/tools/pg_bsd_indent/indent_globs.h
index 917961bd3d6..e88dd52c891 100644
--- a/src/tools/pg_bsd_indent/indent_globs.h
+++ b/src/tools/pg_bsd_indent/indent_globs.h
@@ -234,8 +234,8 @@ extern int   ifdef_level;
 struct parser_state {
     int         last_token;
     int         p_stack[256];	/* this is the parsers stack */
-    int         il[64];		/* this stack stores indentation levels */
-    float       cstk[32];	/* used to store case stmt indentation levels */
+    int         il[256];		/* this stack stores indentation levels */
+    float       cstk[256];	/* used to store case stmt indentation levels */
     int         box_com;	/* set to true when we are in a "boxed"
 				 * comment. In that case, the first non-blank
 				 * char should be lined up with the / in / followed by * */
diff --git a/src/tools/pg_bsd_indent/parse.c b/src/tools/pg_bsd_indent/parse.c
index 94cea724393..f50a03193e1 100644
--- a/src/tools/pg_bsd_indent/parse.c
+++ b/src/tools/pg_bsd_indent/parse.c
@@ -203,7 +203,7 @@ parse(int tk) /* tk: the code for the construct scanned */

} /* end of switch */

-    if (ps.tos >= nitems(ps.p_stack) - 1)
+    if (ps.tos >= nitems(ps.p_stack) - 2)
 	errx(1, "Parser stack overflow");

reduce(); /* see if any reduction can be done */
--
2.43.0

#2Daniel Gustafsson
daniel@yesql.se
In reply to: Vadim Shakirov (#1)
Re: [PATCH v1] Fix out-of-bounds access in pg_bsd_indent's parser stack

On 30 Sep 2026, at 12:51, Vadim Shakirov <vadimsakirov5@gmail.com> wrote:

parser_state stores parser state in p_stack, il, and cstk, which
share the tos index. The existing overflow check used p_stack, the
largest array, allowing out-of-bounds accesses to cstk and il. Size
all three arrays equally.

Also widen the overflow check by one element: the lbrace case pushes
two entries (lbrace and then stmt), so there must be room for two
more elements when parse() is entered.

I haven't read the patch, but reproducers of bugs are good to have. Is there a
pattern which hits this out-of-bounds access?

--
Daniel Gustafsson

#3Alvaro Herrera
alvherre@2ndquadrant.com
In reply to: Daniel Gustafsson (#2)
Re: [PATCH v1] Fix out-of-bounds access in pg_bsd_indent's parser stack

On 2026-Sep-30, Daniel Gustafsson wrote:

I haven't read the patch, but reproducers of bugs are good to have. Is there a
pattern which hits this out-of-bounds access?

Fun. I did this

(echo "int main() {" ; for i in `seq 1 256`; do echo "if ($i) "; done ; echo "switch (argc) {case 1: break;}}" ) > pgindent.c

which produced a smallish file:
$ ls -l pgindent.c
-rw-rw-r-- 1 alvherre alvherre 2497 Sep 30 13:52 pgindent.c

then ran pg_bsd_indent on it, which caused a segmentation fault and
ended with a somewhat larger file

$ ls -l pgindent.c
-rw-rw-r-- 1 alvherre alvherre 599314432 Sep 30 13:53 pgindent.c

That size was the entirety of the free disk space in that partition.

After applying this fix, it doesn't crash anymore, and merely dies with
a "Parser stack overflow" after having written up to "if (251)".

--
Álvaro Herrera PostgreSQL Developer — https://www.EnterpriseDB.com/
“Cuando no hay humildad las personas se degradan” (A. Christie)

#4Daniel Gustafsson
daniel@yesql.se
In reply to: Alvaro Herrera (#3)
Re: [PATCH v1] Fix out-of-bounds access in pg_bsd_indent's parser stack

On 30 Sep 2026, at 13:59, Álvaro Herrera <alvherre@kurilemu.de> wrote:

On 2026-Sep-30, Daniel Gustafsson wrote:

I haven't read the patch, but reproducers of bugs are good to have. Is there a
pattern which hits this out-of-bounds access?

Fun. I did this

(echo "int main() {" ; for i in `seq 1 256`; do echo "if ($i) "; done ; echo "switch (argc) {case 1: break;}}" ) > pgindent.c

Neat, thanks!

--
Daniel Gustafsson