Skip to contentSkip to conversation

btrfs-progs: fix dropped 'const' qualifiers exposed by the latest GCC (15.2.1)

PATCH In review Qu Wenruo 2 participants 6 messages Feb 22 – Mar 17 Btrfs file system

lore
Share & tools
lore

Qu Wenruo

author
#0 lore raw

cover btrfs-progs: fix dropped 'const' qualifiers exposed by the latest GCC (15.2.1)

It looks lile the latest gcc has taken the 'const' qualifier checks one step further, a char pointer that points into a string that has 'const' qualifier should also has 'const'.

This will expose unexpected modification like the following:

void my_func(const char *options)

 {
         char *dot;

         dot = strchr(options, '.');
         if (!dot)
                 *dot = '\0';

}

In above example, @dot is either NULL or points to a location inside @options. For the later case, since @dot itself is not const, we can modify the content, resulting modification of the content of @options. The latest GCC is able to detect such proxy modification and gives us warning on them.

And in fact, btrfs-progs has exactly such proxied modification in bconf_save_param(), fixed in the last patch by dropping the 'const' quailifer.

Other than that, most are just false alerts and we can fix them by adding a const quailifer.

Qu Wenruo (3):
btrfs-progs: enhance find_option()
btrfs-progs: constify the @dots variable inside parse_range_u64()
btrfs-progs: drop the 'const' qualifier from bconf_save_param()

 common/parse-utils.c |  4 ++--
 common/utils.c       | 24 ++++++++++++++++++------
 common/utils.h       |  2 +-
 3 files changed, 21 insertions(+), 9 deletions(-)

--
2.53.0

author
#1 lore raw

PATCH 1/3 btrfs-progs: enhance find_option()

There are two minor problems in that function:

- Latest GCC reports dropped 'const' qualifier

The latest GCC (15.2.1) will report the 'const' qualifier is dropped for the strstr() call.

It looks like the latest GCC has taken the 'const' qualifier checks one step further, that any pointer that points into a 'const' pointer should also have a 'const' qualifier, to make sure there is no modification into the original pointed memory.

In our case, @options parameter has a 'const' qualifier, thus the returned @tmp pointer should also have a 'const' qualifier, or we can modify the @options through @tmp.

- No error handling if strdup() failed

Enhance them by:

- Add the 'const' qualifier for @tmp
This also means we can not use @tmp pointer to modify the newly duplicated string.
Instead use a new @i index to access the new string.

- Return NULL if strdup() failed
And make the caller check errno for memory allocation failure.

---
 common/utils.c | 22 +++++++++++++++++-----
 1 file changed, 17 insertions(+), 5 deletions(-)
common/utils.c +17 -5
index 23dbd9d16781..f183705edb8f 100644
@@ -375,19 +375,27 @@ struct mnt_entry {
 /*
  * Find first occurrence of up an option string (as "option=") in @options,
  * separated by comma. Return allocated string as "option=value"
+ *
+ * If NULL is returned, the caller needs to check errno to make sure it's not
+ * caused by memory allocation failure.
  */
 static char *find_option(const char *options, const char *option)
 {
-	char *tmp, *ret;
+	const char *tmp;
+	char *ret;
+	int i;
 
+	errno = 0;
 	tmp = strstr(options, option);
 	if (!tmp)
 		return NULL;
 	ret = strdup(tmp);
-	tmp = ret;
-	while (*tmp && *tmp != ',')
-		tmp++;
-	*tmp = 0;
+	if (!ret)
+		return NULL;
+	i = 0;
+	while (*(ret + i) && *(ret + i) != ',')
+		i++;
+	*(ret + i) = '\0';
 	return ret;
 }
 
@@ -600,6 +608,10 @@ int find_mount_fsroot(const char *subvol, const char *subvolid, char **mount)
 			 * requested by the caller
 			 */
 			opt = find_option(ent.options2, "subvolid=");
+			if (!opt && errno) {
+				ret = -errno;
+				goto out;
+			}
 			if (!opt)
 				goto nextline;
 			value = opt + strlen("subvolid=");
Signature · 1 line
2.53.0
#2 lore raw

Re: [PATCH 1/3] btrfs-progs: enhance find_option()

Quoted 55 lines · Qu Wenruo

There are two minor problems in that function:

- Latest GCC reports dropped 'const' qualifier

The latest GCC (15.2.1) will report the 'const' qualifier is dropped for the strstr() call.

It looks like the latest GCC has taken the 'const' qualifier checks one step further, that any pointer that points into a 'const' pointer should also have a 'const' qualifier, to make sure there is no modification into the original pointed memory.

In our case, @options parameter has a 'const' qualifier, thus the returned @tmp pointer should also have a 'const' qualifier, or we can modify the @options through @tmp.

- No error handling if strdup() failed

Enhance them by:

- Add the 'const' qualifier for @tmp
This also means we can not use @tmp pointer to modify the newly duplicated string.
Instead use a new @i index to access the new string.

- Return NULL if strdup() failed
And make the caller check errno for memory allocation failure.

---
 common/utils.c | 22 +++++++++++++++++-----
 1 file changed, 17 insertions(+), 5 deletions(-)
common/utils.c +7 -1
index 23dbd9d16781..f183705edb8f 100644
@@ -375,19 +375,27 @@ struct mnt_entry {
 /*
  * Find first occurrence of up an option string (as "option=") in @options,
  * separated by comma. Return allocated string as "option=value"
+ *
+ * If NULL is returned, the caller needs to check errno to make sure it's not
+ * caused by memory allocation failure.
  */
 static char *find_option(const char *options, const char *option)
 {
-	char *tmp, *ret;
+	const char *tmp;
+	char *ret;
+	int i;
 
+	errno = 0;
 	tmp = strstr(options, option);
 	if (!tmp)

common/utils.c struct mnt_entry { · line 396

Quoted 10 lines
common/utils.c +4 -4
 		return NULL;
 	ret = strdup(tmp);
-	tmp = ret;
-	while (*tmp && *tmp != ',')
-		tmp++;
-	*tmp = 0;
+	if (!ret)
+		return NULL;
+	i = 0;
+	while (*(ret + i) && *(ret + i) != ',')

Is there a reason for not using ret[i] ?

author
#3 lore raw

PATCH 2/3 btrfs-progs: constify the @dots variable inside parse_range_u64()

In reply to Qu Wenruo

[WARNING]
The latest compiler (GCC 15.2.1) along with the latest glibc (2.43) will report the following warning:

common/parse-utils.c: In function 'parse_range_u64':
common/parse-utils.c:82:14: warning: assignment discards 'const' qualifier from pointer target type [-Wdiscarded-qualifiers]

    82 |         dots = strstr(range, "..");
       |              ^

[CAUSE]
Although the strstr() definition in man page doesn't have a 'const' qualifier, it looks like the latest GCC has taken the 'const' qualifier checks one step further, that any pointer that points into a 'const' pointer should also have a 'const' qualifier, to make sure there is no modification into the original pointed memory.

[FIX]
Just add a 'const' qualifier to @dots.

---
 common/parse-utils.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
common/parse-utils.c +2 -2
index 9e88aa844b44..4fc9ec865572 100644
@@ -74,9 +74,9 @@ int parse_u64(const char *str, u64 *result)
  */
 int parse_range_u64(const char *range, u64 *start, u64 *end)
 {
-	char *dots;
-	char *endptr;
+	const char *dots;
 	const char *rest;
+	char *endptr;
 	int skipped = 0;
 
 	dots = strstr(range, "..");
Signature · 1 line
2.53.0
author
#4 lore raw

PATCH 3/3 btrfs-progs: drop the 'const' qualifier from bconf_save_param()

In reply to Qu Wenruo

[BUG]
The latest GCC (15.2.1) will report the following warning:

common/utils.c: In function 'bconf_save_param':
common/utils.c:975:13: warning: assignment discards 'const' qualifier from pointer target type [-Wdiscarded-qualifiers]
975 | tmp = strchr(str, '=');
| ^

[CAUSE]
I believe the latest GCC has a more strict const qualifier checks, that any returned pointer that points into a const pointer should also have a 'const' qualifier.

But in this particular case, we're even modifying the content of the original const string.

This means we shouldn't really have 'const' qualifier for bconf_save_param() in the first place.

[FIX]
Drop the 'const' qualifier from @str parameter, as we're going to modify the string in that function.

---
 common/utils.c | 2 +-
 common/utils.h | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)
common/utils.c +1 -1
index f183705edb8f..59e318282918 100644
@@ -968,7 +968,7 @@ const char *bconf_param_value(const char *key)
 	return NULL;
 }
 
-void bconf_save_param(const char *str)
+void bconf_save_param(char *str)
 {
 	char *tmp;
 
common/utils.h +1 -1
index 1419850df387..d0bd062b75a5 100644
@@ -118,7 +118,7 @@ void btrfs_config_init(void);
 void bconf_be_verbose(void);
 void bconf_be_quiet(void);
 void bconf_add_param(const char *key, const char *value);
-void bconf_save_param(const char *str);
+void bconf_save_param(char *str);
 void bconf_set_dry_run(void);
 bool bconf_is_dry_run(void);
 const char *bconf_param_value(const char *key);
Signature · 1 line
2.53.0
#5 lore raw
In reply to Qu Wenruo
Quoted 25 lines · Qu Wenruo

It looks lile the latest gcc has taken the 'const' qualifier checks one step further, a char pointer that points into a string that has 'const' qualifier should also has 'const'.

This will expose unexpected modification like the following:

void my_func(const char *options)

 {
         char *dot;

         dot = strchr(options, '.');
         if (!dot)
                 *dot = '\0';

}

In above example, @dot is either NULL or points to a location inside @options. For the later case, since @dot itself is not const, we can modify the content, resulting modification of the content of @options. The latest GCC is able to detect such proxy modification and gives us warning on them.

And in fact, btrfs-progs has exactly such proxied modification in bconf_save_param(), fixed in the last patch by dropping the 'const' quailifer.

Quoted 7 lines

Other than that, most are just false alerts and we can fix them by adding a const quailifer.

Qu Wenruo (3):
btrfs-progs: enhance find_option()
btrfs-progs: constify the @dots variable inside parse_range_u64()
btrfs-progs: drop the 'const' qualifier from bconf_save_param()

Added to devel, thanks.