Skip to contentSkip to conversation

exec/binfmt_script: trip zero bytes from the buffer

PATCH In review Andrei Vagin 2 participants 5 messages Jun 15, 2021 – Jun 25, 2021

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 exec/binfmt_script: trip zero bytes from the buffer

Without this fix, if we try to run a script that contains only the interpreter line, the interpreter is executed with one extra empty argument.

The code is written so that i_end has to be set to the end of valuable data in the buffer.

Fixes ccbb18b67323 ("exec/binfmt_script: Don't modify bprm->buf and then return -ENOEXEC")
Cc "Eric W. Biederman" <[email protected]>
---
 fs/binfmt_script.c | 3 +++
 1 file changed, 3 insertions(+)
fs/binfmt_script.c +3 -0
index 1b6625e95958..e242680f96e1 100644
@@ -68,6 +68,9 @@ static int load_script(struct linux_binprm *bprm)
 		if (!next_terminator(i_end, buf_end))
 			return -ENOEXEC;
 		i_end = buf_end;
+		/* Trim zero bytes from i_end */
+		while (i_end[-1] == 0)
+			i_end--;
 	}
 	/* Trim any trailing spaces/tabs from i_end */
 	while (spacetab(i_end[-1]))
Signature · 1 line
2.31.1
#1 lore raw
Quoted 6 lines · Andrei Vagin

Without this fix, if we try to run a script that contains only the interpreter line, the interpreter is executed with one extra empty argument.

The code is written so that i_end has to be set to the end of valuable data in the buffer.

Out of curiosity how did you spot this change in behavior?

Quoted 21 lines
Fixes ccbb18b67323 ("exec/binfmt_script: Don't modify bprm->buf and then return -ENOEXEC")
Cc "Eric W. Biederman" <[email protected]>
---
 fs/binfmt_script.c | 3 +++
 1 file changed, 3 insertions(+)
fs/binfmt_script.c +3 -0
index 1b6625e95958..e242680f96e1 100644
@@ -68,6 +68,9 @@ static int load_script(struct linux_binprm *bprm)
 		if (!next_terminator(i_end, buf_end))
 			return -ENOEXEC;
 		i_end = buf_end;
+		/* Trim zero bytes from i_end */
+		while (i_end[-1] == 0)
+			i_end--;
 	}
 	/* Trim any trailing spaces/tabs from i_end */
 	while (spacetab(i_end[-1]))
#2 lore raw
Quoted 11 lines · Eric W. Biederman
Quoted 6 lines · Andrei Vagin

Without this fix, if we try to run a script that contains only the interpreter line, the interpreter is executed with one extra empty argument.

The code is written so that i_end has to be set to the end of valuable data in the buffer.

Out of curiosity how did you spot this change in behavior?

gVisor tests started failing with this change:
https://github.com/google/gvisor/blob/5e05950c1c520724e2e03963850868befb95efeb/test/syscalls/linux/exec.cc#L307

We run these tests on Ubuntu 20.04 and this is the reason why we caught this issue just a few days ago.

Quoted 22 lines
Quoted 22 lines
Fixes ccbb18b67323 ("exec/binfmt_script: Don't modify bprm->buf and then return -ENOEXEC")
Cc "Eric W. Biederman" <[email protected]>
---
 fs/binfmt_script.c | 3 +++
 1 file changed, 3 insertions(+)
fs/binfmt_script.c +3 -0
index 1b6625e95958..e242680f96e1 100644
@@ -68,6 +68,9 @@ static int load_script(struct linux_binprm *bprm)
              if (!next_terminator(i_end, buf_end))
                      return -ENOEXEC;
              i_end = buf_end;
+             /* Trim zero bytes from i_end */
+             while (i_end[-1] == 0)
+                     i_end--;
      }
      /* Trim any trailing spaces/tabs from i_end */
      while (spacetab(i_end[-1]))
#3 lore raw
Quoted 12 lines · Andrei Vagin
Quoted 10 lines · Eric W. Biederman
Quoted 6 lines · Andrei Vagin

Without this fix, if we try to run a script that contains only the interpreter line, the interpreter is executed with one extra empty argument.

The code is written so that i_end has to be set to the end of valuable data in the buffer.

Quoted 7 lines
Quoted 1 line

Out of curiosity how did you spot this change in behavior?

gVisor tests started failing with this change:
https://github.com/google/gvisor/blob/5e05950c1c520724e2e03963850868befb95efeb/test/syscalls/linux/exec.cc#L307

We run these tests on Ubuntu 20.04 and this is the reason why we caught this issue just a few days ago.

I like where you are going, but starting at the end of the buffer there is the potential to skip deliberately embedded '\0' characters.

While looking at this I realized that your patch should not have made a difference but there is a subtle bug in the logic of next_non_spacetab, that allowed your code to make it that far.

Can you test my patch below?

I think I have simplified the logic enough to prevent bugs from getting in.

Eric

fs/binfmt_script.c +22 -18
index 1b6625e95958..7d204693326c 100644
@@ -26,7 +26,7 @@ static inline const char *next_non_spacetab(const char *first, const char *last)
 static inline const char *next_terminator(const char *first, const char *last)
 {
 	for (; first <= last; first++)
-		if (spacetab(*first) || !*first)
+		if (spacetab(*first))
 			return first;
 	return NULL;
 }
@@ -44,9 +44,9 @@ static int load_script(struct linux_binprm *bprm)
 	/*
 	 * This section handles parsing the #! line into separate
 	 * interpreter path and argument strings. We must be careful
-	 * because bprm->buf is not yet guaranteed to be NUL-terminated
-	 * (though the buffer will have trailing NUL padding when the
-	 * file size was smaller than the buffer size).
+	 * because bprm->buf is not guaranteed to be NUL-terminated
+	 * (the buffer will have trailing NUL padding when the file
+	 * size was smaller than the buffer size).
 	 *
 	 * We do not want to exec a truncated interpreter path, so either
 	 * we find a newline (which indicates nothing is truncated), or
@@ -57,33 +57,37 @@ static int load_script(struct linux_binprm *bprm)
 	 */
 	buf_end = bprm->buf + sizeof(bprm->buf) - 1;
 	i_end = strnchr(bprm->buf, sizeof(bprm->buf), '\n');
-	if (!i_end) {
-		i_end = next_non_spacetab(bprm->buf + 2, buf_end);
-		if (!i_end)
-			return -ENOEXEC; /* Entire buf is spaces/tabs */
-		/*
-		 * If there is no later space/tab/NUL we must assume the
-		 * interpreter path is truncated.
-		 */
-		if (!next_terminator(i_end, buf_end))
-			return -ENOEXEC;
-		i_end = buf_end;
+	if (i_end) {
+		/* Hide the trailing newline */
+		i_end = i_end - 1;
+	} else {
+		/* Find the end of the text */
+		i_end = memchr(bprm->buf + 2, '\0', sizeof(bprm->buf));
+		i_end = i_end ? i_end - 1 : buf_end;
 	}
+
 	/* Trim any trailing spaces/tabs from i_end */
-	while (spacetab(i_end[-1]))
+	while (spacetab(i_end[0]))
 		i_end--;
 
 	/* Skip over leading spaces/tabs */
 	i_name = next_non_spacetab(bprm->buf+2, i_end);
-	if (!i_name || (i_name == i_end))
+	if (!i_name)
 		return -ENOEXEC; /* No interpreter name found */
 
 	/* Is there an optional argument? */
 	i_arg = NULL;
 	i_sep = next_terminator(i_name, i_end);
-	if (i_sep && (*i_sep != '\0'))
+	if (i_sep)
 		i_arg = next_non_spacetab(i_sep, i_end);
 
+	/*
+	 * If there is no space/tab/NUL after the interpreter we must
+	 * assume the interpreter path is truncated.
+	 */
+	if (!i_sep && (i_end == buf_end))
+		return -ENOEXEC;
+
 	/*
 	 * If the script filename will be inaccessible after exec, typically
 	 * because it is a "/dev/fd/<fd>/.." path against an O_CLOEXEC fd, give
#4 lore raw
Quoted 70 lines · Eric W. Biederman
Quoted 12 lines · Andrei Vagin
Quoted 10 lines · Eric W. Biederman
Quoted 6 lines · Andrei Vagin

Without this fix, if we try to run a script that contains only the interpreter line, the interpreter is executed with one extra empty argument.

The code is written so that i_end has to be set to the end of valuable data in the buffer.

Quoted 7 lines
Quoted 1 line

Out of curiosity how did you spot this change in behavior?

gVisor tests started failing with this change:
https://github.com/google/gvisor/blob/5e05950c1c520724e2e03963850868befb95efeb/test/syscalls/linux/exec.cc#L307

We run these tests on Ubuntu 20.04 and this is the reason why we caught this issue just a few days ago.

I like where you are going, but starting at the end of the buffer there is the potential to skip deliberately embedded '\0' characters.

While looking at this I realized that your patch should not have made a difference but there is a subtle bug in the logic of next_non_spacetab, that allowed your code to make it that far.

Can you test my patch below?

I think I have simplified the logic enough to prevent bugs from getting in.

Eric

fs/binfmt_script.c +4 -8
index 1b6625e95958..7d204693326c 100644
@@ -26,7 +26,7 @@ static inline const char *next_non_spacetab(const char *first, const char *last)
 static inline const char *next_terminator(const char *first, const char *last)
 {
 	for (; first <= last; first++)
-		if (spacetab(*first) || !*first)
+		if (spacetab(*first))
 			return first;
 	return NULL;
 }
@@ -44,9 +44,9 @@ static int load_script(struct linux_binprm *bprm)
 	/*
 	 * This section handles parsing the #! line into separate
 	 * interpreter path and argument strings. We must be careful
-	 * because bprm->buf is not yet guaranteed to be NUL-terminated
-	 * (though the buffer will have trailing NUL padding when the
-	 * file size was smaller than the buffer size).
+	 * because bprm->buf is not guaranteed to be NUL-terminated
+	 * (the buffer will have trailing NUL padding when the file
+	 * size was smaller than the buffer size).
 	 *
 	 * We do not want to exec a truncated interpreter path, so either
 	 * we find a newline (which indicates nothing is truncated), or
@@ -57,33 +57,37 @@ static int load_script(struct linux_binprm *bprm)
 	 */
 	buf_end = bprm->buf + sizeof(bprm->buf) - 1;
 	i_end = strnchr(bprm->buf, sizeof(bprm->buf), '\n');
-	if (!i_end) {
-		i_end = next_non_spacetab(bprm->buf + 2, buf_end);
-		if (!i_end)
-			return -ENOEXEC; /* Entire buf is spaces/tabs */

fs/binfmt_script.c static int load_script(struct linux_binprm *bprm) · line 62

Quoted 10 lines
fs/binfmt_script.c +3 -7
-		/*
-		 * If there is no later space/tab/NUL we must assume the
-		 * interpreter path is truncated.
-		 */
-		if (!next_terminator(i_end, buf_end))
-			return -ENOEXEC;
-		i_end = buf_end;
+	if (i_end) {
+		/* Hide the trailing newline */
+		i_end = i_end - 1;

Your patch changes the meaning of i_end. Now it points to the last symbol, but this function contains the line:

	*((char *)i_end) = '\0';

and it drops the last meaningful symbol. With the following tiny fix, my test passes:

@@ -114,7 +115,7 @@ static int load_script(struct linux_binprm *bprm)

        if (retval < 0)
                return retval;
        bprm->argc++;
-       *((char *)i_end) = '\0';
+       *((char *)(i_end + 1)) = '\0';
        if (i_arg) {
                *((char *)i_sep) = '\0';
                retval = copy_string_kernel(i_arg, bprm);

Thanks,
Andrei