Skip to contentSkip to conversation

reiserfs : fix improper free in reiserfs_get_block

PATCH In review [email protected] 3 participants 5 messages Jul 14, 2020 – Jul 16, 2020

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 reiserfs : fix improper free in reiserfs_get_block

From: Tom Rix <[email protected]>

clang static analysis flags this error

inode.c:1083:5: warning: Argument to kfree() is the address of the
local variable 'unf_single', which is not memory allocated by malloc() [unix.Malloc]

                                kfree(un);
                                ^~~~~~~~~

Assignment of 'un'

	/*
	 * We use this in case we need to allocate
	 * only one block which is a fastpath
	 */
	unp_t unf_single = 0;

	...

	if (blocks_needed == 1) {
		un = &unf_single;
	} else {
		un = kcalloc(min(blocks_needed, max_to_insert),
			     UNFM_P_SIZE, GFP_NOFS);
		if (!un) {
			un = &unf_single;
			blocks_needed = 1;
			max_to_insert = 0;
		}
	}

The logic to free 'un'

	if (blocks_needed != 1)
		kfree(un);

Because the kcalloc failure falls back to using unf_single, the if-check for the free is wrong.

So improve the check.

---
 fs/reiserfs/inode.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/reiserfs/inode.c +1 -1
index 1509775da040..4d62148e43e6 100644
@@ -1079,7 +1079,7 @@ int reiserfs_get_block(struct inode *inode, sector_t block,
 						     UNFM_P_SIZE *
 						     blocks_needed);
 
-			if (blocks_needed != 1)
+			if (un != &unf_single)
 				kfree(un);
 
 			if (retval) {
Signature · 1 line
2.18.1
#1 lore raw
Quoted 31 lines · On Tue, Jul 14, 2020 at 06:05:09AM -0700, [email protected] wrote:

From: Tom Rix <[email protected]>

clang static analysis flags this error

inode.c:1083:5: warning: Argument to kfree() is the address of the
local variable 'unf_single', which is not memory allocated by malloc() [unix.Malloc]

                                kfree(un);
                                ^~~~~~~~~

Assignment of 'un'

	/*
	 * We use this in case we need to allocate
	 * only one block which is a fastpath
	 */
	unp_t unf_single = 0;

	...

	if (blocks_needed == 1) {
		un = &unf_single;
	} else {
		un = kcalloc(min(blocks_needed, max_to_insert),
			     UNFM_P_SIZE, GFP_NOFS);
		if (!un) {
			un = &unf_single;
			blocks_needed = 1;
			max_to_insert = 0;
		}
	}
Quoted 7 lines

The logic to free 'un'

	if (blocks_needed != 1)
		kfree(un);

Because the kcalloc failure falls back to using unf_single, the if-check for the free is wrong.

I think you mean "Because clang's static analysis is limited, it warns incorrectly about this". There's no path to get to the kfree with blocks_needed != 1 and un being equal to &unf_single.

Quoted 12 lines

So improve the check.

---
 fs/reiserfs/inode.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/reiserfs/inode.c +0 -0
index 1509775da040..4d62148e43e6 100644

fs/reiserfs/inode.c int reiserfs_get_block(struct inode *inode, sector_t block, · line 1083

Quoted 6 lines
fs/reiserfs/inode.c +1 -1
@@ -1079,7 +1079,7 @@ int reiserfs_get_block(struct inode *inode, sector_t block,
 						     UNFM_P_SIZE *
 						     blocks_needed);
 
-			if (blocks_needed != 1)
+			if (un != &unf_single)
 				kfree(un);

I don't actually object to this patch, but your analysis of clang's analysis is wrong.

author
#2 lore raw
Quoted 34 lines · Matthew Wilcox
Quoted 33 lines · On Tue, Jul 14, 2020 at 06:05:09AM -0700, [email protected] wrote:

From: Tom Rix <[email protected]>

clang static analysis flags this error

inode.c:1083:5: warning: Argument to kfree() is the address of the
local variable 'unf_single', which is not memory allocated by malloc() [unix.Malloc]

                                kfree(un);
                                ^~~~~~~~~

Assignment of 'un'

	/*
	 * We use this in case we need to allocate
	 * only one block which is a fastpath
	 */
	unp_t unf_single = 0;

	...

	if (blocks_needed == 1) {
		un = &unf_single;
	} else {
		un = kcalloc(min(blocks_needed, max_to_insert),
			     UNFM_P_SIZE, GFP_NOFS);
		if (!un) {
			un = &unf_single;
			blocks_needed = 1;
			max_to_insert = 0;
		}
	}

The logic to free 'un'

Quoted 8 lines
Quoted 5 lines
	if (blocks_needed != 1)
		kfree(un);

Because the kcalloc failure falls back to using unf_single, the if-check for the free is wrong.

I think you mean "Because clang's static analysis is limited, it warns incorrectly about this". There's no path to get to the kfree with blocks_needed != 1 and un being equal to &unf_single.

Ok.

Quoted 21 lines
Quoted 12 lines

So improve the check.

---
 fs/reiserfs/inode.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/reiserfs/inode.c +0 -0
index 1509775da040..4d62148e43e6 100644

fs/reiserfs/inode.c int reiserfs_get_block(struct inode *inode, sector_t block, · line 1083

Quoted 6 lines
fs/reiserfs/inode.c +1 -1
@@ -1079,7 +1079,7 @@ int reiserfs_get_block(struct inode *inode, sector_t block,
 						     UNFM_P_SIZE *
 						     blocks_needed);
 
-			if (blocks_needed != 1)
+			if (un != &unf_single)
 				kfree(un);

I don't actually object to this patch, but your analysis of clang's analysis is wrong.

#3 lore raw
Quoted 39 lines · Tom Rix
Quoted 37 lines · Matthew Wilcox
Quoted 36 lines · On Tue, Jul 14, 2020 at 06:05:09AM -0700, [email protected] wrote:

From: Tom Rix <[email protected]>

clang static analysis flags this error

inode.c:1083:5: warning: Argument to kfree() is the address of the
local variable 'unf_single', which is not memory allocated by malloc() [unix.Malloc]

                                kfree(un);
                                ^~~~~~~~~

Assignment of 'un'

	/*
	 * We use this in case we need to allocate
	 * only one block which is a fastpath
	 */
	unp_t unf_single = 0;

	...

	if (blocks_needed == 1) {
		un = &unf_single;
	} else {
		un = kcalloc(min(blocks_needed, max_to_insert),
			     UNFM_P_SIZE, GFP_NOFS);
		if (!un) {
			un = &unf_single;
			blocks_needed = 1;
			max_to_insert = 0;
		}
	}

The logic to free 'un'

	if (blocks_needed != 1)
		kfree(un);
Quoted 7 lines
Quoted 5 lines
Quoted 2 lines

Because the kcalloc failure falls back to using unf_single, the if-check for the free is wrong.

I think you mean "Because clang's static analysis is limited, it warns incorrectly about this". There's no path to get to the kfree with blocks_needed != 1 and un being equal to &unf_single.

Ok.

I agree with Matthew the patch will make the code more obviously correct so it's a sensible cleanup. But the changelog needs to redone to reflect this is just a cleanup before the patch can be merged.

Honza

Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
author
#4 lore raw
Quoted 44 lines · Jan Kara
Quoted 43 lines · Tom Rix
Quoted 42 lines · Matthew Wilcox
Quoted 31 lines · On Tue, Jul 14, 2020 at 06:05:09AM -0700, [email protected] wrote:

From: Tom Rix <[email protected]>

clang static analysis flags this error

inode.c:1083:5: warning: Argument to kfree() is the address of the
local variable 'unf_single', which is not memory allocated by malloc() [unix.Malloc]

                                kfree(un);
                                ^~~~~~~~~

Assignment of 'un'

	/*
	 * We use this in case we need to allocate
	 * only one block which is a fastpath
	 */
	unp_t unf_single = 0;

	...

	if (blocks_needed == 1) {
		un = &unf_single;
	} else {
		un = kcalloc(min(blocks_needed, max_to_insert),
			     UNFM_P_SIZE, GFP_NOFS);
		if (!un) {
			un = &unf_single;
			blocks_needed = 1;
			max_to_insert = 0;
		}
	}
Quoted 7 lines

The logic to free 'un'

	if (blocks_needed != 1)
		kfree(un);

Because the kcalloc failure falls back to using unf_single, the if-check for the free is wrong.

I think you mean "Because clang's static analysis is limited, it warns incorrectly about this". There's no path to get to the kfree with blocks_needed != 1 and un being equal to &unf_single.

Quoted 6 lines
Quoted 1 line

Ok.

I agree with Matthew the patch will make the code more obviously correct so it's a sensible cleanup. But the changelog needs to redone to reflect this is just a cleanup before the patch can be merged.

Honza

I am going to look into the problem with the analyzer because that is where the fix should go.

If the problem isn't resolvable, i will loop back to this clean up.

Tom