ext4: Fix integer overflow in ext4fs_read_symlink()
Subsystems (2)
PATCH 1/2 ext4: Fix integer overflow in ext4fs_read_symlink()
While zalloc() takes a size_t type, adding 1 to the le32 variable will overflow.
A carefully crafted ext4 filesystem can exhibit an inode size of 0xffffffff and as consequence zalloc() will do a zero allocation.
Later in the function the inode size is again used for copying data. So an attacker can overwrite memory.
Avoid the overflow by using the __builtin_add_overflow() helper.
--- fs/ext4/ext4_common.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)
fs/ext4/ext4_common.c +6 -1
@@ -2183,13 +2183,18 @@ static char *ext4fs_read_symlink(struct ext2fs_node *node)struct ext2fs_node *diro = node;int status;loff_t actread; +size_t alloc_size;if (!diro->inode_read) { status =ext4fs_read_inode (diro->data, diro->ino, &diro->inode);if (status ==0 )return NULL ; } - symlink =zalloc (le32_to_cpu (diro->inode.size) +1 ); + +if (__builtin_add_overflow (le32_to_cpu (diro->inode.size),1 , &alloc_size)) +return NULL ; + + symlink =zalloc (alloc_size);if (!symlink)return NULL ;
Signature · 1 line
2.35.3
PATCH 2/2 ext4: Fix zalloc()
The zalloc() function suffers from two problems.
1. If memalign() fails it will return NULL and memset() will use a NULL pointer.
2. memalign() itself seems to crash when more than 2^32 bytes are requested.
So, check the return value of memalign() and allocate only of size is less than CONFIG_SYS_MALLOC_LEN.
---
FWIW, I didn't investigate further why memalign() fails for large sizes. Maybe this is an issue on it's own.
Thanks,
//richard
--- fs/ext4/ext4_common.h | 10 ++++++++-- 1 file changed, 8 insertions(+), 2 deletions(-)
fs/ext4/ext4_common.h +8 -2
@@ -43,8 +43,14 @@static inline void *zalloc (size_t size) { -void *p =memalign (ARCH_DMA_MINALIGN , size); -memset (p,0 , size); +void *p =NULL ; + +if (size <CONFIG_SYS_MALLOC_LEN ) + p =memalign (ARCH_DMA_MINALIGN , size); + +if (p) +memset (p,0 , size); +return p; }
Signature · 1 line
2.35.3
Re: [PATCH 2/2] ext4: Fix zalloc()
Quoted 28 lines · Richard Weinberger
The
zalloc()function suffers from two problems.
1. Ifmemalign()fails it will return NULL andmemset()will use a NULL pointer.
2.memalign()itself seems to crash when more than 2^32 bytes are requested.So, check the return value of
memalign()and allocate only of size is less thanCONFIG_SYS_MALLOC_LEN.Signed-off-by Richard Weinberger <[email protected]>---
FWIW, I didn't investigate further whymemalign()fails for large sizes. Maybe this is an issue on it's own.Thanks,
//richard--- fs/ext4/ext4_common.h | 10 ++++++++-- 1 file changed, 8 insertions(+), 2 deletions(-)fs/ext4/ext4_common.h +0
-2@@ -43,8 +43,14 @@static inline void *zalloc (size_t size) { -void *p =memalign (ARCH_DMA_MINALIGN , size); -memset (p,0 , size);
fs/ext4/ext4_common.h line 55
Quoted 10 lines
fs/ext4/ext4_common.h +8
-0+void *p =NULL ; + +if (size <CONFIG_SYS_MALLOC_LEN ) + p =memalign (ARCH_DMA_MINALIGN , size); + +if (p) +memset (p,0 , size); +return p; }
The problem here is that "zalloc" is inline and so this change causes about 1KiB of growth on platforms which enable ext4 and so at least mx6sabresd now overflows it's maximum size. Looking harder, I think the best solution here would be for ext4 to stop using its own wrapper and instead call our kzalloc compatibility function.
Signature · 1 line
Tom
Re: [PATCH 2/2] ext4: Fix zalloc()
Tom,
Am Donnerstag, 11. Juli 2024, 17:45:17 CEST schrieb Tom Rini:
Quoted 5 lines
The problem here is that "zalloc" is inline and so this change causes about 1KiB of growth on platforms which enable ext4 and so at least mx6sabresd now overflows it's maximum size. Looking harder, I think the best solution here would be for ext4 to stop using its own wrapper and instead call our kzalloc compatibility function.
As discussed on IRC yesterday, moving to kzalloc() is fine. But the crash around malloc() still needs a fix.
Last night I investigated further why u-boot's malloc() implementation crashes on my x86_64 test bed when ext4 tries to allocate a lot of memory.
It turned out that it's an integer overflow around malloc_extend_top() and sbrk().malloc_extend_top() uses a size_t to calculate the amount of required memory and sbrk() takes an ptrdiff_t type.
On x86_64, u-boot seems to use unsigned long for size_t but just an int for ptrdiff_t.
This is causing the trouble.
How about this?
arch/x86/include/asm/posix_types.h +2 -1
@@ -20,11 +20,12 @@ typedef unsigned short __kernel_gid_t;#if defined(__x86_64__)typedef unsigned long __kernel_size_t ;typedef long __kernel_ssize_t ; +typedef long __kernel_ptrdiff_t ;#else typedef unsigned int __kernel_size_t ;typedef int __kernel_ssize_t ; -#endif typedef int __kernel_ptrdiff_t ; +#endif typedef long __kernel_time_t ;typedef long __kernel_suseconds_t ;typedef long __kernel_clock_t ;
Thanks,
//richard
Signature · 2 lines
sigma star gmbh | Eduard-Bodem-Gasse 6, 6020 Innsbruck, AUT UID/VAT Nr: ATU 66964118 | FN: 374287y
Re: [PATCH 2/2] ext4: Fix zalloc()
In reply to Richard WeinbergerQuoted 23 lines · Richard Weinberger
The
zalloc()function suffers from two problems.
1. Ifmemalign()fails it will return NULL andmemset()will use a NULL pointer.
2.memalign()itself seems to crash when more than 2^32 bytes are requested.So, check the return value of
memalign()and allocate only of size is less thanCONFIG_SYS_MALLOC_LEN.Signed-off-by Richard Weinberger <[email protected]>---
FWIW, I didn't investigate further whymemalign()fails for large sizes. Maybe this is an issue on it's own.Thanks,
//richard--- fs/ext4/ext4_common.h | 10 ++++++++-- 1 file changed, 8 insertions(+), 2 deletions(-)fs/ext4/ext4_common.h +0
-0
fs/ext4/ext4_common.h line 49
Quoted 9 lines
fs/ext4/ext4_common.h +4
-2@@ -43,8 +43,14 @@static inline void *zalloc (size_t size) { -void *p =memalign (ARCH_DMA_MINALIGN , size); -memset (p,0 , size); +void *p =NULL ; + +if (size <CONFIG_SYS_MALLOC_LEN ) + p =memalign (ARCH_DMA_MINALIGN , size);
Memalign() is called in many code locations.
If memalign() has a bug, it needs to be fixed in memalign. We should not try to work around it in all callers.
Best regards
Heinrich
Quoted 7 lines
fs/ext4/ext4_common.h +4
-0+ +if (p) +memset (p,0 , size); +return p; }
Re: [PATCH 2/2] ext4: Fix zalloc()
Am Freitag, 12. Juli 2024, 13:15:56 CEST schrieb 'Heinrich Schuchardt' via upstream:
Quoted 4 lines
Memalign()is called in many code locations.If
memalign()has a bug, it needs to be fixed in memalign. We should not try to work around it in all callers.
Agreed, I did already:
See https://lists.denx.de/pipermail/u-boot/2024-July/558477.html
Thanks,
//richard
Signature · 2 lines
sigma star gmbh | Eduard-Bodem-Gasse 6, 6020 Innsbruck, AUT UID/VAT Nr: ATU 66964118 | FN: 374287y
Quoted 23 lines · Richard Weinberger
While
zalloc()takes asize_ttype, adding 1 to the le32 variable will overflow.
A carefully crafted ext4 filesystem can exhibit an inode size of0xffffffffand as consequencezalloc()will do a zero allocation.Later in the function the inode size is again used for copying data. So an attacker can overwrite memory.
Avoid the overflow by using the
__builtin_add_overflow()helper.Signed-off-by Richard Weinberger <[email protected]>--- fs/ext4/ext4_common.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)fs/ext4/ext4_common.c +0
-0@@ -2183,13 +2183,18 @@ static char *ext4fs_read_symlink(struct ext2fs_node *node)struct ext2fs_node *diro = node;int status;loff_t actread;
fs/ext4/ext4_common.c static char *ext4fs_read_symlink(struct ext2fs_node *node) · line 2194
Quoted 10 lines
fs/ext4/ext4_common.c +3
-1+size_t alloc_size;if (!diro->inode_read) { status =ext4fs_read_inode (diro->data, diro->ino, &diro->inode);if (status ==0 )return NULL ; } - symlink =zalloc (le32_to_cpu (diro->inode.size) +1 ); + +if (__builtin_add_overflow (le32_to_cpu (diro->inode.size),1 , &alloc_size))
U-Boot is freestanding code. You cannot use built-ins.
Best regards
Heinrich
Quoted 6 lines
fs/ext4/ext4_common.c +3
-0+return NULL ; + + symlink =zalloc (alloc_size);if (!symlink)return NULL ;
Am Freitag, 12. Juli 2024, 13:10:12 CEST schrieb 'Heinrich Schuchardt' via upstream:
Quoted 26 lines · Richard Weinberger
Quoted 25 lines
While
zalloc()takes asize_ttype, adding 1 to the le32 variable will overflow.
A carefully crafted ext4 filesystem can exhibit an inode size of0xffffffffand as consequencezalloc()will do a zero allocation.Later in the function the inode size is again used for copying data. So an attacker can overwrite memory.
Avoid the overflow by using the
__builtin_add_overflow()helper.Signed-off-by Richard Weinberger <[email protected]>--- fs/ext4/ext4_common.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)fs/ext4/ext4_common.c +1
-0@@ -2183,13 +2183,18 @@ static char *ext4fs_read_symlink(struct ext2fs_node *node)struct ext2fs_node *diro = node;int status;loff_t actread; +size_t alloc_size;
Quoted 10 lines
fs/ext4/ext4_common.c static char *ext4fs_read_symlink(struct ext2fs_node *node) · line 2194
Quoted 8 lines
fs/ext4/ext4_common.c +2
-1if (!diro->inode_read) { status =ext4fs_read_inode (diro->data, diro->ino, &diro->inode);if (status ==0 )return NULL ; } - symlink =zalloc (le32_to_cpu (diro->inode.size) +1 ); + +if (__builtin_add_overflow (le32_to_cpu (diro->inode.size),1 , &alloc_size))U-Boot is freestanding code. You cannot use built-ins.
Hm, I see man built-ins in the U-Boot source.
Why is this one special?
Thanks,
//richard
Signature · 2 lines
sigma star gmbh | Eduard-Bodem-Gasse 6, 6020 Innsbruck, AUT UID/VAT Nr: ATU 66964118 | FN: 374287y
Quoted 38 lines · Richard Weinberger
Am Freitag, 12. Juli 2024, 13:10:12 CEST schrieb 'Heinrich Schuchardt' via upstream:
Quoted 26 lines · Richard Weinberger
Quoted 25 lines
While
zalloc()takes asize_ttype, adding 1 to the le32 variable will overflow.
A carefully crafted ext4 filesystem can exhibit an inode size of0xffffffffand as consequencezalloc()will do a zero allocation.Later in the function the inode size is again used for copying data. So an attacker can overwrite memory.
Avoid the overflow by using the
__builtin_add_overflow()helper.Signed-off-by Richard Weinberger <[email protected]>--- fs/ext4/ext4_common.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)fs/ext4/ext4_common.c +1
-0@@ -2183,13 +2183,18 @@ static char *ext4fs_read_symlink(struct ext2fs_node *node)struct ext2fs_node *diro = node;int status;loff_t actread; +size_t alloc_size;Quoted 10 lines
fs/ext4/ext4_common.c static char *ext4fs_read_symlink(struct ext2fs_node *node) · line 2194
Quoted 8 lines
fs/ext4/ext4_common.c +2
-1if (!diro->inode_read) { status =ext4fs_read_inode (diro->data, diro->ino, &diro->inode);if (status ==0 )return NULL ; } - symlink =zalloc (le32_to_cpu (diro->inode.size) +1 ); + +if (__builtin_add_overflow (le32_to_cpu (diro->inode.size),1 , &alloc_size))U-Boot is freestanding code. You cannot use built-ins.
Quoted 2 lines
Hm, I see man built-ins in the U-Boot source.
Why is this one special?
See the definition of COMPILER_HAS_GENERIC_BUILTIN_OVERFLOW in include/linux/compiler-clang.h.
Best regards
Heinrich
Am Freitag, 12. Juli 2024, 13:19:32 CEST schrieb Heinrich Schuchardt:
Quoted 5 lines
Quoted 2 lines
Hm, I see man built-ins in the U-Boot source.
Why is this one special?See the definition of
COMPILER_HAS_GENERIC_BUILTIN_OVERFLOWininclude/linux/compiler-clang.h.
So I can't use __builtin_add_overflow() because
u-boot is still supporting outdated compilers that don't offer __builtin_add_overflow()?
Maybe this is the ideal moment to raise the bar? __builtin_add_overflow() is supported by gcc >= 5.1 and clang >= 8.
Thanks,
//richard
Signature · 2 lines
sigma star gmbh | Eduard-Bodem-Gasse 6, 6020 Innsbruck, AUT UID/VAT Nr: ATU 66964118 | FN: 374287y
Quoted 4 lines · Richard Weinberger
Am Freitag, 12. Juli 2024, 13:19:32 CEST schrieb Heinrich Schuchardt:
Quoted 3 lines
Quoted 2 lines
Hm, I see man built-ins in the U-Boot source.
Why is this one special?
Quoted 8 lines
Quoted 2 lines
See the definition of
COMPILER_HAS_GENERIC_BUILTIN_OVERFLOWininclude/linux/compiler-clang.h.So I can't use
__builtin_add_overflow()because
u-boot is still supporting outdated compilers that don't offer__builtin_add_overflow()?Maybe this is the ideal moment to raise the bar?
__builtin_add_overflow()is supported by gcc >= 5.1 and clang >= 8.
We already fail on gcc older than 6, for ARM. And we don't have a minimum clang specifically, but it's probably newer than 8. So this should be fine.
Signature · 1 line
Tom