Skip to contentSkip to conversation

Split *s*printf(3) from printf(3), and document the snprintf(,-1,) is UB

PATCH In review Alejandro Colomar 2 participants 7 messages Dec 5, 2025 – Dec 6, 2025 Man-pages: manual pages for Linux -- Sections 2, 3, 4, 5, and 7

lore
Share & tools
lore

Alejandro Colomar

author maintainer
#0 lore raw

cover Split *s*printf(3) from printf(3), and document the snprintf(,-1,) is UB

Hi!

Serge and I were reviewing some code, and we realized that it seems quite easy to induce a buffer overflow using snprintf(3).

We solved our problems by checking that the size is non-negative in a wrapper inline function.

	// vstprintf - va_list string truncate print formatted
	inline int
	vstprintf(char *restrict s, ssize_t size, const char *restrict fmt, va_list ap)
	{
		int  len;

		if (size <= 0 || size > INT_MAX) {
			errno = EOVERFLOW;
			return -1;
		}
		len = vsnprintf(s, size, fmt, ap);
		if (len == -1) {
			if (errno == EOVERFLOW)
				errno = E2BIG;
			return -1;
		}
		if (len >= size) {
			errno = E2BIG;
			return -1;
		}

		return len;
	}

[v]snprintf(3) seems to be even worse than I thought it was. Let's document this.

Alejandro Colomar (2):
man/man3/*printf.3: Split *s*printf(3) from printf(3) man/man3/snprintf.3: BUGS: snprintf(3) with a negative size is
essentially UB

 man/man3/printf.3    | 223 +---------------------------
 man/man3/snprintf.3  | 337 ++++++++++++++++++++++++++++++++++++++++++-
 man/man3/sprintf.3   |   2 +-
 man/man3/vsnprintf.3 |   2 +-
 man/man3/vsprintf.3  |   2 +-
 5 files changed, 346 insertions(+), 220 deletions(-)

Range-diff:

-:  --------- > 1:  a654dcfc2 man/man3/*printf.3: Split *s*printf(3) from printf(3)
-:  --------- > 2:  d00ed9434 man/man3/snprintf.3: BUGS: snprintf(3) with a negative size is essentially UB

base-commit: 2a5edb7e59ae024d1969e7379c078a58b5f50b80

Signature · 1 line
2.51.0
author maintainer
#1 lore raw

PATCH 1/2 man/man3/*printf.3: Split *s*printf(3) from printf(3)

---
 man/man3/printf.3    | 223 +-----------------------------
 man/man3/snprintf.3  | 318 ++++++++++++++++++++++++++++++++++++++++++-
 man/man3/sprintf.3   |   2 +-
 man/man3/vsnprintf.3 |   2 +-
 man/man3/vsprintf.3  |   2 +-
 5 files changed, 327 insertions(+), 220 deletions(-)
man/man3/printf.3 +7 -216
index d8e165169..892d7682f 100644
@@ -5,8 +5,9 @@
 .\"
 .TH printf 3 (date) "Linux man-pages (unreleased)"
 .SH NAME
-printf, fprintf, dprintf, sprintf, snprintf, vprintf, vfprintf, vdprintf,
-vsprintf, vsnprintf \- formatted output conversion
+printf, fprintf, dprintf, vprintf, vfprintf, vdprintf,
+\-
+formatted output conversion
 .SH LIBRARY
 Standard C library
 .RI ( libc ,\~ \-lc )
@@ -19,22 +20,12 @@ .SH SYNOPSIS
 .BI "            const char *restrict " format ", ...);"
 .BI "int dprintf(int " fd ,
 .BI "            const char *restrict " format ", ...);"
-.BI "int sprintf(char *restrict " str ,
-.BI "            const char *restrict " format ", ...);"
-.BR "int snprintf(" "size_t size;"
-.BI "            char " str "[restrict " size "], size_t " size ,
-.BI "            const char *restrict " format ", ...);"
 .P
 .BI "int vprintf(const char *restrict " format ", va_list " ap );
 .BI "int vfprintf(FILE *restrict " stream ,
 .BI "            const char *restrict " format ", va_list " ap );
 .BI "int vdprintf(int " fd ,
 .BI "            const char *restrict " format ", va_list " ap );
-.BI "int vsprintf(char *restrict " str ,
-.BI "            const char *restrict " format ", va_list " ap );
-.BR "int vsnprintf(" "size_t size;"
-.BI "            char " str "[restrict " size "], size_t " size ,
-.BI "            const char *restrict " format ", va_list " ap );
 .fi
 .P
 .RS -4
@@ -42,13 +33,6 @@ .SH SYNOPSIS
 .BR feature_test_macros (7)):
 .RE
 .P
-.BR snprintf (),
-.BR vsnprintf ():
-.nf
-    _XOPEN_SOURCE >= 500 || _ISOC99_SOURCE
-        || /* glibc <= 2.19: */ _BSD_SOURCE
-.fi
-.P
 .BR dprintf (),
 .BR vdprintf ():
 .nf
@@ -74,14 +58,7 @@ .SH DESCRIPTION
 and
 .BR vfprintf ()
 write output to the given output
-.IR stream ;
-.BR sprintf (),
-.BR snprintf (),
-.BR vsprintf (),
-and
-.BR vsnprintf ()
-write to the character string
-.IR str .
+.IR stream .
 .P
 The function
 .BR dprintf ()
@@ -94,26 +71,13 @@ .SH DESCRIPTION
 stream.
 .P
 The functions
-.BR snprintf ()
-and
-.BR vsnprintf ()
-write at most
-.I size
-bytes (including the terminating null byte (\[aq]\[rs]0\[aq])) to
-.IR str .
-.P
-The functions
 .BR vprintf (),
 .BR vfprintf (),
-.BR vdprintf (),
-.BR vsprintf (),
-.BR vsnprintf ()
+.BR vdprintf ()
 are equivalent to the functions
 .BR printf (),
 .BR fprintf (),
 .BR dprintf (),
-.BR sprintf (),
-.BR snprintf (),
 respectively, except that they are called with a
 .I va_list
 instead of a variable number of arguments.
@@ -134,17 +98,6 @@ .SH DESCRIPTION
 the variable-length argument facilities of
 .BR stdarg (3))
 are converted for output.
-.P
-C99 and POSIX.1-2001 specify that the results are undefined if a call to
-.BR sprintf (),
-.BR snprintf (),
-.BR vsprintf (),
-or
-.BR vsnprintf ()
-would cause copying to take place between objects that overlap
-(e.g., if the target string array and one of the supplied input arguments
-refer to the same buffer).
-See CAVEATS.
 .SS Format of the format string
 The format string is a character string, beginning and ending
 in its initial shift state, if any.
@@ -913,22 +866,6 @@ .SH RETURN VALUE
 Upon successful return, these functions return the number of bytes
 printed (excluding the null byte used to end output to strings).
 .P
-The functions
-.BR snprintf ()
-and
-.BR vsnprintf ()
-do not write more than
-.I size
-bytes (including the terminating null byte (\[aq]\[rs]0\[aq])).
-If the output was truncated due to this limit, then the return value
-is the number of characters (excluding the terminating null byte)
-which would have been written to the final string if enough space
-had been available.
-Thus, a return value of
-.I size
-or more means that the output was truncated.
-(See also below under CAVEATS.)
-.P
 On error,
 a negative value is returned,
 and
@@ -966,12 +903,8 @@ .SH ATTRIBUTES
 .nh
 .BR printf (),
 .BR fprintf (),
-.BR sprintf (),
-.BR snprintf (),
 .BR vprintf (),
-.BR vfprintf (),
-.BR vsprintf (),
-.BR vsnprintf ()
+.BR vfprintf ()
 T}	Thread safety	MT-Safe locale
 .TE
 .SH STANDARDS
@@ -980,17 +913,9 @@ .SH STANDARDS
 .TQ
 .BR printf ()
 .TQ
-.BR sprintf ()
-.TQ
 .BR vprintf ()
 .TQ
 .BR vfprintf ()
-.TQ
-.BR vsprintf ()
-.TQ
-.BR snprintf ()
-.TQ
-.BR vsnprintf ()
 C11, POSIX.1-2008.
 .TP
 .BR dprintf ()
@@ -1003,36 +928,11 @@ .SH HISTORY
 .TQ
 .BR printf ()
 .TQ
-.BR sprintf ()
-.TQ
 .BR vprintf ()
 .TQ
 .BR vfprintf ()
-.TQ
-.BR vsprintf ()
 C89, POSIX.1-2001.
 .TP
-.BR snprintf ()
-.TQ
-.BR vsnprintf ()
-SUSv2, C99, POSIX.1-2001.
-.IP
-Concerning the return value of
-.BR snprintf (),
-SUSv2 and C99 contradict each other: when
-.BR snprintf ()
-is called with
-.IR size =0
-then SUSv2 stipulates an unspecified return value less than 1,
-while C99 allows
-.I str
-to be NULL in this case, and gives the return value (as always)
-as the number of characters that would have been written in case
-the output string has been large enough.
-POSIX.1-2001 and later align their specification of
-.BR snprintf ()
-with C99.
-.TP
 .BR dprintf ()
 .TQ
 .BR vdprintf ()
@@ -1091,39 +991,6 @@ .SH HISTORY
 .B m
 conversion specifier, that is
 .IR %#m .
-.SH CAVEATS
-Some programs imprudently rely on code such as the following
-.P
-.in +4n
-.EX
-sprintf(buf, "%s some further text", buf);
-.EE
-.in
-.P
-to append text to
-.IR buf .
-However, the standards explicitly note that the results are undefined
-if source and destination buffers overlap when calling
-.BR sprintf (),
-.BR snprintf (),
-.BR vsprintf (),
-and
-.BR vsnprintf ().
-.\" http://sourceware.org/bugzilla/show_bug.cgi?id=7075
-Depending on the version of
-.BR gcc (1)
-used, and the compiler options employed, calls such as the above will
-.B not
-produce the expected results.
-.P
-The glibc implementation of the functions
-.BR snprintf ()
-and
-.BR vsnprintf ()
-conforms to the C99 standard, that is, behaves as described above,
-since glibc 2.1.
-Until glibc 2.0.6, they would return \-1
-when the output was truncated.
 .\" .SH HISTORY
 .\" UNIX V7 defines the three routines
 .\" .BR printf (),
@@ -1158,37 +1025,6 @@ .SH CAVEATS
 .\" .BR vdprintf ()
 .\" that print to a file descriptor instead of a stream.
 .SH BUGS
-Because
-.BR sprintf ()
-and
-.BR vsprintf ()
-assume an arbitrarily long string,
-callers must be careful not to overflow the actual space;
-this is often impossible to assure.
-Note that the length of the strings produced is locale-dependent
-and difficult to predict.
-Use
-.BR snprintf ()
-and
-.BR vsnprintf ()
-instead (or
-.BR asprintf (3)
-and
-.BR vasprintf (3)).
-.\" .P
-.\" Linux libc4.[45] does not have a
-.\" .BR snprintf (),
-.\" but provides a libbsd that contains an
-.\" .BR snprintf ()
-.\" equivalent to
-.\" .BR sprintf (),
-.\" that is, one that ignores the
-.\" .I size
-.\" argument.
-.\" Thus, the use of
-.\" .BR snprintf ()
-.\" with early libc4 leads to serious security problems.
-.P
 Code such as
 .BI printf( foo );
 often indicates a bug, since
@@ -1256,58 +1092,13 @@ .SH EXAMPLES
 .in
 .P
 one might obtain "Sonntag, 3.\& Juli, 10:02".
-.P
-To allocate a sufficiently large string and print into it
-(code correct for both glibc 2.0 and glibc 2.1):
-.P
-.EX
-#include <stdio.h>
-#include <stdlib.h>
-#include <stdarg.h>
-\&
-char *
-make_message(const char *fmt, ...)
-{
-    int n = 0;
-    size_t size = 0;
-    char *p = NULL;
-    va_list ap;
-\&
-    /* Determine required size.  */
-\&
-    va_start(ap, fmt);
-    n = vsnprintf(p, size, fmt, ap);
-    va_end(ap);
-\&
-    if (n < 0)
-        return NULL;
-\&
-    size = (size_t) n + 1;      /* One extra byte for \[aq]\[rs]0\[aq] */
-    p = malloc(size);
-    if (p == NULL)
-        return NULL;
-\&
-    va_start(ap, fmt);
-    n = vsnprintf(p, size, fmt, ap);
-    va_end(ap);
-\&
-    if (n < 0) {
-        free(p);
-        return NULL;
-    }
-\&
-    return p;
-}
-.EE
-.P
-If truncation occurs in glibc versions prior to glibc 2.0.6,
-this is treated as an error instead of being handled gracefully.
 .SH SEE ALSO
 .BR printf (1),
 .BR asprintf (3),
 .BR puts (3),
 .BR scanf (3),
 .BR setlocale (3),
+.BR snprintf (3),
 .BR strfromd (3),
 .BR wcrtomb (3),
 .BR wprintf (3),
man/man3/snprintf.3 +317 -1

318 lines not on this page · raw message

man/man3/sprintf.3 +1 -1
index 975530d90..48bccad4d 100644
@@ -1 +1 @@
-.so man3/printf.3
+.so man3/snprintf.3
man/man3/vsnprintf.3 +1 -1
index 975530d90..48bccad4d 100644
@@ -1 +1 @@
-.so man3/printf.3
+.so man3/snprintf.3
man/man3/vsprintf.3 +1 -1
index 975530d90..48bccad4d 100644
@@ -1 +1 @@
-.so man3/printf.3
+.so man3/snprintf.3
Signature · 1 line
2.51.0
author maintainer
#2 lore raw

PATCH 2/2 man/man3/snprintf.3: BUGS: snprintf(3) with a negative size is essentially UB

In reply to Alejandro Colomar
---
 man/man3/snprintf.3 | 21 ++++++++++++++++++++-
 1 file changed, 20 insertions(+), 1 deletion(-)
man/man3/snprintf.3 +20 -1
index dca9fc3ef..e6bb9547d 100644
@@ -54,7 +54,8 @@ .SH DESCRIPTION
 write at most
 .I size
 bytes (including the terminating null byte (\[aq]\[rs]0\[aq])) to
-.IR str .
+.I str
+(but see BUGS).
 .P
 The functions
 .BR vsprintf ()
@@ -242,6 +243,24 @@ .SS sprintf(), vsprintf()
 .BR asprintf (3)
 and
 .BR vasprintf (3)).
+.SS snprintf(), vsnprintf()
+If
+.I size
+is negative,
+these functions will behave
+as if the size was some unreasonably high value.
+This turns these functions essentially equivalent to
+.RB [ v ] sprintf (),
+and thus likely to overflow the buffer.
+It is the programmer's responsibility to pass a non-negative
+.IR size .
+.P
+For historic reasons,
+the
+.I size
+parameter is of type
+.IR int ,
+which makes it relatively easy to pass a negative value.
 .\" .P
 .\" Linux libc4.[45] does not have a
 .\" .BR snprintf (),
Signature · 1 line
2.51.0
author maintainer
#3 lore raw

Re: [PATCH 2/2] man/man3/snprintf.3: BUGS: snprintf(3) with a negative size is essentially UB

Hi,

Here's how this patch changes the formatted page:

$ MANWIDTH=72 diffman-git HEAD
--- HEAD^:man/man3/snprintf.3
+++ HEAD:man/man3/snprintf.3
@@ -33,7 +33,8 @@ DESCRIPTION
      to the character string str instead of a stream.

The functions snprintf() and vsnprintf() write at most size bytes

-     (including the terminating null byte ('\0')) to str.
+     (including the terminating null byte ('\0')) to str (but see
+     BUGS).

The functions vsprintf() and vsnprintf() are equivalent to the functions sprintf() and snprintf(), respectively, except that they
@@ -132,6 +133,16 @@ BUGS
Use snprintf() and vsnprintf() instead (or asprintf(3) and vasprintf(3)).

+ snprintf(), vsnprintf()

+     If size is negative, these functions will behave as if the size
+     was some unreasonably high value.  This turns these functions es‐
+     sentially equivalent to [v]sprintf(), and thus likely to overflow
+     the buffer.  It is the programmer’s responsibility to pass a non‐
+     negative size.

+
+ For historic reasons, the size parameter is of type int, which
+ makes it relatively easy to pass a negative value.
+
EXAMPLES
To allocate a sufficiently large string and print into it (code correct for both glibc 2.0 and glibc 2.1):

Cheers,
Alex

Quoted 48 lines · Alejandro Colomar
---
 man/man3/snprintf.3 | 21 ++++++++++++++++++++-
 1 file changed, 20 insertions(+), 1 deletion(-)
man/man3/snprintf.3 +20 -1
index dca9fc3ef..e6bb9547d 100644
@@ -54,7 +54,8 @@ .SH DESCRIPTION
 write at most
 .I size
 bytes (including the terminating null byte (\[aq]\[rs]0\[aq])) to
-.IR str .
+.I str
+(but see BUGS).
 .P
 The functions
 .BR vsprintf ()
@@ -242,6 +243,24 @@ .SS sprintf(), vsprintf()
 .BR asprintf (3)
 and
 .BR vasprintf (3)).
+.SS snprintf(), vsnprintf()
+If
+.I size
+is negative,
+these functions will behave
+as if the size was some unreasonably high value.
+This turns these functions essentially equivalent to
+.RB [ v ] sprintf (),
+and thus likely to overflow the buffer.
+It is the programmer's responsibility to pass a non-negative
+.IR size .
+.P
+For historic reasons,
+the
+.I size
+parameter is of type
+.IR int ,
+which makes it relatively easy to pass a negative value.
 .\" .P
 .\" Linux libc4.[45] does not have a
 .\" .BR snprintf (),
Signature · 1 line
2.51.0
Signature · 1 line
<https://www.alejandro-colomar.es>
author maintainer
#4 lore raw

Re: [PATCH 2/2] man/man3/snprintf.3: BUGS: snprintf(3) with a negative size is essentially UB

In reply to Alejandro Colomar

Hi,

Quoted 32 lines · Alejandro Colomar
---
 man/man3/snprintf.3 | 21 ++++++++++++++++++++-
 1 file changed, 20 insertions(+), 1 deletion(-)
man/man3/snprintf.3 +10 -1
index dca9fc3ef..e6bb9547d 100644
@@ -54,7 +54,8 @@ .SH DESCRIPTION
 write at most
 .I size
 bytes (including the terminating null byte (\[aq]\[rs]0\[aq])) to
-.IR str .
+.I str
+(but see BUGS).
 .P
 The functions
 .BR vsprintf ()
@@ -242,6 +243,24 @@ .SS sprintf(), vsprintf()
 .BR asprintf (3)
 and
 .BR vasprintf (3)).
+.SS snprintf(), vsnprintf()
+If
+.I size
+is negative,
+these functions will behave
+as if the size was some unreasonably high value.
+This turns these functions essentially equivalent to
+.RB [ v ] sprintf (),

man/man3/snprintf.3 .SS sprintf(), vsprintf() · line 263

Quoted 10 lines
man/man3/snprintf.3 +10 -0
+and thus likely to overflow the buffer.
+It is the programmer's responsibility to pass a non-negative
+.IR size .
+.P
+For historic reasons,
+the
+.I size
+parameter is of type
+.IR int ,
+which makes it relatively easy to pass a negative value.

Actually, I was completely wrong. The parameter is of type size_t. Only the return value is of type int.

Have a lovely day!
Alex

Quoted 6 lines
man/man3/snprintf.3 +0 -0
 .\" .P
 .\" Linux libc4.[45] does not have a
 .\" .BR snprintf (),
Signature · 1 line
2.51.0
Signature · 1 line
<https://www.alejandro-colomar.es>
#5 lore raw

Re: [PATCH 2/2] man/man3/snprintf.3: BUGS: snprintf(3) with a negative size is essentially UB

Dec 6, 2025 07:32:46 Alejandro Colomar <[email protected]>:

Quoted 51 lines

Hi,

Quoted 32 lines · Alejandro Colomar

---
man/man3/snprintf.3 | 21 ++++++++++++++++++++-

1 file changed, 20 insertions(+), 1 deletion(-)
man/man3/snprintf.3 +0 -0
index dca9fc3ef..e6bb9547d 100644
@@ -54,7 +54,8 @@ .SH DESCRIPTION

write at most
.I size
bytes (including the terminating null byte (\[aq]\[rs]0\[aq])) to -.IR str .
+.I str
+(but see BUGS).
.P
The functions
.BR vsprintf ()

man/man3/snprintf.3 +0 -0
@@ -242,6 +243,24 @@ .SS sprintf(), vsprintf()
 .BR asprintf (3)

and
.BR vasprintf (3)).
+.SS snprintf(), vsnprintf()

man/man3/snprintf.3 +7 -0
+If
+.I size
+is negative,
+these functions will behave
+as if the size was some unreasonably high value.
+This turns these functions essentially equivalent to
+.RB [ v ] sprintf (),

man/man3/snprintf.3 .SS sprintf(), vsprintf() · line 263

Quoted 10 lines
man/man3/snprintf.3 +10 -0
+and thus likely to overflow the buffer.
+It is the programmer's responsibility to pass a non-negative
+.IR size .
+.P
+For historic reasons,
+the
+.I size
+parameter is of type
+.IR int ,
+which makes it relatively easy to pass a negative value.

Actually, I was completely wrong.  The parameter is of type size_t. Only the return value is of type int.

Quoted 3 lines

Have a lovely day!
Alex

Right,  that's actually the danger. It's easy to pass in an int or ssize_t, and end up overflowing the buffer, even if you (the caller) check for > buflen in advance, by passing in a negative value. By the time you check the return value,  it's too late.

"You just need to know (and not do that)", but if you are reviewing a patch and don't  notice the extra s in ssize_t...

author maintainer
#6 lore raw

Re: [PATCH 2/2] man/man3/snprintf.3: BUGS: snprintf(3) with a negative size is essentially UB

Hi Serge,

Quoted 4 lines · Serge Hallyn
Quoted 4 lines

Actually, I was completely wrong.  The parameter is of type size_t. Only the return value is of type int.

Quoted 8 lines
Quoted 3 lines

Have a lovely day!
Alex

Right,  that's actually the danger. It's easy to pass in an int or ssize_t, and end up overflowing the buffer, even if you (the caller) check for > buflen in advance, by passing in a negative value. By the time you check the return value,  it's too late.

ssize_t should be safe, as it has the same width as size_t. It should be impossible to have a negative value in ssize_t, because objects can't grow past PTRDIFF_MAX, which is --at least in all platforms we care about-- exactly the same as SSIZE_MAX.

So, to pass a negative value, you would have to either write it literally, or do something really bad. malloc(3) can't possibly create an object that would result in a negative ssize_t.

The problem would only exist with int, because int is easy to overflow.

That is, the following code is fine:

	ssize_t  x = _Countof(buf);
	snprintf(buf, x, ...);

And it behaves exactly like this:

	size_t  x = _Countof(buf);
	snprintf(buf, x, ...);

However, the following is likely to result in a buffer overflow:

	int  x = _Countof(buf);
	snprintf(buf, x, ...);

Have a lovely day!
Alex

Quoted 1 line

"You just need to know (and not do that)", but if you are reviewing a patch and don't  notice the extra s in ssize_t...

Signature · 1 line
<https://www.alejandro-colomar.es>