Several potential vulnerabilities in the filesystem
Hi Das U-Boot developers,
I found several vulnerabilities in the u-boot filesysetm implementation,
1. in file fs/squashfs/sqfs_inode.c function sqfs_inode_size. The parameter blk_size is directly or indirectly from the storage data. Howver, without a sanity check, this value is directly used in the division operations, leading to a division-by-zero exception
2. in file fs/erofs/data.c, function z_erofs_read_one_data, the node data is read from the storage, however, without a proper check, the data can be corrupted. For example, the inode data is used in function z_erofs_read_data, map.m_llen will be calculated to a very large value, which means the length variable will be very large. It will cause a large memory clear with memset(buffer + end - offset, 0, length);
3. in file fs/squashfs/sqfs.c, function sqfs_frag_lookup, the header variable is read/calculated from the storage data, however, without a proper check, memcpy(entries, metadata, SQFS_METADATA_SIZE(header)); will cause a buffer over write when header cannot correctly clear the higher bits (SQFS_METADATA_SIZE(header)).
Could you please confirmware these vulnerabilities?
Best regards
On 2024/6/5 06:53, jianqiang wang wrote:
Hi Das U-Boot developers,
...
- in file fs/erofs/data.c, function z_erofs_read_one_data, the node
data is read from the storage, however, without a proper check, the data can be corrupted. For example, the inode data is used in function z_erofs_read_data, map.m_llen will be calculated to a very large value, which means the length variable will be very large. It will cause a large memory clear with memset(buffer + end - offset, 0, length);
Would you mind giving a reproducer or a crafted image to trigger this? Or it's your pure observation.
Thanks, Gao XIang
Hi,
I do have the crafted image.
payload_00500, payload_00763, payload_00846 can be used to reproduce 1,2,3 vulnerabilities respectively.
Each image is a hard drive file and the vulnerabilities can be triggered by performing the following operations:
struct udevice *dev; uclass_first_device_err(UCLASS_IDE, &dev); //detect the block device
fs_set_blk_dev("ide","0:1", 0); fs_ls("/"); //mount the first partition and list the root directory files
fs_set_blk_dev("ide","0:1", 0); char buf[10]; buf[0] = 0; buf[1] = 0; buf[2] = 0; buf[3] = 0; loff_t actread = 0; fs_read("/a.txt", (ulong)buf, 0, 5, &actread); printf("fd actread %lld %x %x %x\n",actread,buf[0],buf[1],buf[2]); read the /a.txt file
fs_set_blk_dev("ide","0:1", 0); fs_read("/a.txt.ln", (ulong)buf, 0, 5, &actread); printf("fd actread %lld %x %x %x\n",actread,buf[0],buf[1],buf[2]); read the /a.txt.ln symbol file
fs_set_blk_dev("ide","0:1", 0); fs_unlink("/a.txt.ln"); //unlink it
The second and third may not trigger a crash however, can be observed by inserting logging before the memset/memcpy function.
Best regards
Gao Xiang hsiangkao@linux.alibaba.com 于2024年6月5日周三 05:10写道：
On 2024/6/5 06:53, jianqiang wang wrote:
Hi Das U-Boot developers,
...
- in file fs/erofs/data.c, function z_erofs_read_one_data, the node
data is read from the storage, however, without a proper check, the data can be corrupted. For example, the inode data is used in function z_erofs_read_data, map.m_llen will be calculated to a very large value, which means the length variable will be very large. It will cause a large memory clear with memset(buffer + end - offset, 0, length);
Would you mind giving a reproducer or a crafted image to trigger this? Or it's your pure observation.
Thanks, Gao XIang
Hi Jianqiang,
On 2024/6/5 19:00, jianqiang wang wrote:
Hi,
I do have the crafted image.
payload_00500, payload_00763, payload_00846 can be used to reproduce 1,2,3 vulnerabilities respectively.
Each image is a hard drive file and the vulnerabilities can be triggered by performing the following operations:
struct udevice *dev; uclass_first_device_err(UCLASS_IDE, &dev); //detect the block device fs_set_blk_dev("ide","0:1", 0); fs_ls("/"); //mount the first partition and list the root directory files fs_set_blk_dev("ide","0:1", 0); char buf[10]; buf[0] = 0; buf[1] = 0; buf[2] = 0; buf[3] = 0; loff_t actread = 0; fs_read("/a.txt", (ulong)buf, 0, 5, &actread); printf("fd actread %lld %x %x %x\n",actread,buf[0],buf[1],buf[2]);read the /a.txt file
fs_set_blk_dev("ide","0:1", 0); fs_read("/a.txt.ln", (ulong)buf, 0, 5, &actread); printf("fd actread %lld %x %x %x\n",actread,buf[0],buf[1],buf[2]);read the /a.txt.ln symbol file
fs_set_blk_dev("ide","0:1", 0); fs_unlink("/a.txt.ln"); //unlink itThe second and third may not trigger a crash however, can be observed by inserting logging before the memset/memcpy function.
Sorry, I just found that this issue was already fixed in erofs-utils: https://git.kernel.org/pub/scm/linux/kernel/git/xiang/erofs-utils.git/commit...
Would you mind checking if the patch above fixes the issue?
Hi Jianan, Would you mind backporting this patch for u-boot?
Thanks, Gao Xiang
Best regards
Gao Xiang hsiangkao@linux.alibaba.com 于2024年6月5日周三 05:10写道：
On 2024/6/5 06:53, jianqiang wang wrote:
Hi Das U-Boot developers,
...
- in file fs/erofs/data.c, function z_erofs_read_one_data, the node
data is read from the storage, however, without a proper check, the data can be corrupted. For example, the inode data is used in function z_erofs_read_data, map.m_llen will be calculated to a very large value, which means the length variable will be very large. It will cause a large memory clear with memset(buffer + end - offset, 0, length);
Would you mind giving a reproducer or a crafted image to trigger this? Or it's your pure observation.
Thanks, Gao XIang
Hi Xiang,
I just checked the second crash, the patch can solve this issue. Did you also look into the other two issues?
Best Jianqiang
Gao Xiang hsiangkao@linux.alibaba.com 于2024年6月5日周三 13:18写道：
Hi Jianqiang,
On 2024/6/5 19:00, jianqiang wang wrote:
Hi,
I do have the crafted image.
payload_00500, payload_00763, payload_00846 can be used to reproduce 1,2,3 vulnerabilities respectively.
Each image is a hard drive file and the vulnerabilities can be triggered by performing the following operations:
struct udevice *dev; uclass_first_device_err(UCLASS_IDE, &dev); //detect the block device fs_set_blk_dev("ide","0:1", 0); fs_ls("/"); //mount the first partition and list the root directory files fs_set_blk_dev("ide","0:1", 0); char buf[10]; buf[0] = 0; buf[1] = 0; buf[2] = 0; buf[3] = 0; loff_t actread = 0; fs_read("/a.txt", (ulong)buf, 0, 5, &actread); printf("fd actread %lld %x %x %x\n",actread,buf[0],buf[1],buf[2]);read the /a.txt file
fs_set_blk_dev("ide","0:1", 0); fs_read("/a.txt.ln", (ulong)buf, 0, 5, &actread); printf("fd actread %lld %x %x %x\n",actread,buf[0],buf[1],buf[2]);read the /a.txt.ln symbol file
fs_set_blk_dev("ide","0:1", 0); fs_unlink("/a.txt.ln"); //unlink itThe second and third may not trigger a crash however, can be observed by inserting logging before the memset/memcpy function.
Sorry, I just found that this issue was already fixed in erofs-utils: https://git.kernel.org/pub/scm/linux/kernel/git/xiang/erofs-utils.git/commit...
Would you mind checking if the patch above fixes the issue?
Hi Jianan, Would you mind backporting this patch for u-boot?
Thanks, Gao Xiang
Best regards
Gao Xiang hsiangkao@linux.alibaba.com 于2024年6月5日周三 05:10写道：
On 2024/6/5 06:53, jianqiang wang wrote:
Hi Das U-Boot developers,
...
- in file fs/erofs/data.c, function z_erofs_read_one_data, the node
data is read from the storage, however, without a proper check, the data can be corrupted. For example, the inode data is used in function z_erofs_read_data, map.m_llen will be calculated to a very large value, which means the length variable will be very large. It will cause a large memory clear with memset(buffer + end - offset, 0, length);
Would you mind giving a reproducer or a crafted image to trigger this? Or it's your pure observation.
Thanks, Gao XIang
On 2024/6/5 19:26, jianqiang wang wrote:
Hi Xiang,
I just checked the second crash, the patch can solve this issue. Did you also look into the other two issues?
I'm only responsible for the EROFS project.
Thanks, Gao Xiang
Best Jianqiang
Gao Xiang hsiangkao@linux.alibaba.com 于2024年6月5日周三 13:18写道：
Hi Jianqiang,
On 2024/6/5 19:00, jianqiang wang wrote:
Hi,
I do have the crafted image.
payload_00500, payload_00763, payload_00846 can be used to reproduce 1,2,3 vulnerabilities respectively.
Each image is a hard drive file and the vulnerabilities can be triggered by performing the following operations:
struct udevice *dev; uclass_first_device_err(UCLASS_IDE, &dev); //detect the block device fs_set_blk_dev("ide","0:1", 0); fs_ls("/"); //mount the first partition and list the root directory files fs_set_blk_dev("ide","0:1", 0); char buf[10]; buf[0] = 0; buf[1] = 0; buf[2] = 0; buf[3] = 0; loff_t actread = 0; fs_read("/a.txt", (ulong)buf, 0, 5, &actread); printf("fd actread %lld %x %x %x\n",actread,buf[0],buf[1],buf[2]);read the /a.txt file
fs_set_blk_dev("ide","0:1", 0); fs_read("/a.txt.ln", (ulong)buf, 0, 5, &actread); printf("fd actread %lld %x %x %x\n",actread,buf[0],buf[1],buf[2]);read the /a.txt.ln symbol file
fs_set_blk_dev("ide","0:1", 0); fs_unlink("/a.txt.ln"); //unlink itThe second and third may not trigger a crash however, can be observed by inserting logging before the memset/memcpy function.
Sorry, I just found that this issue was already fixed in erofs-utils: https://git.kernel.org/pub/scm/linux/kernel/git/xiang/erofs-utils.git/commit...
Would you mind checking if the patch above fixes the issue?
Hi Jianan, Would you mind backporting this patch for u-boot?
Thanks, Gao Xiang
Best regards
Gao Xiang hsiangkao@linux.alibaba.com 于2024年6月5日周三 05:10写道：
On 2024/6/5 06:53, jianqiang wang wrote:
Hi Das U-Boot developers,
...
- in file fs/erofs/data.c, function z_erofs_read_one_data, the node
data is read from the storage, however, without a proper check, the data can be corrupted. For example, the inode data is used in function z_erofs_read_data, map.m_llen will be calculated to a very large value, which means the length variable will be very large. It will cause a large memory clear with memset(buffer + end - offset, 0, length);
Would you mind giving a reproducer or a crafted image to trigger this? Or it's your pure observation.
Thanks, Gao XIang
Could you please forward the issues to whoever is responsible for them?
Gao Xiang hsiangkao@linux.alibaba.com 于2024年6月5日周三 13:35写道：
On 2024/6/5 19:26, jianqiang wang wrote:
Hi Xiang,
I just checked the second crash, the patch can solve this issue. Did you also look into the other two issues?
I'm only responsible for the EROFS project.
Thanks, Gao Xiang
Best Jianqiang
Gao Xiang hsiangkao@linux.alibaba.com 于2024年6月5日周三 13:18写道：
Hi Jianqiang,
On 2024/6/5 19:00, jianqiang wang wrote:
Hi,
I do have the crafted image.
payload_00500, payload_00763, payload_00846 can be used to reproduce 1,2,3 vulnerabilities respectively.
Each image is a hard drive file and the vulnerabilities can be triggered by performing the following operations:
struct udevice *dev; uclass_first_device_err(UCLASS_IDE, &dev); //detect the block device fs_set_blk_dev("ide","0:1", 0); fs_ls("/"); //mount the first partition and list the root directory files fs_set_blk_dev("ide","0:1", 0); char buf[10]; buf[0] = 0; buf[1] = 0; buf[2] = 0; buf[3] = 0; loff_t actread = 0; fs_read("/a.txt", (ulong)buf, 0, 5, &actread); printf("fd actread %lld %x %x %x\n",actread,buf[0],buf[1],buf[2]);read the /a.txt file
fs_set_blk_dev("ide","0:1", 0); fs_read("/a.txt.ln", (ulong)buf, 0, 5, &actread); printf("fd actread %lld %x %x %x\n",actread,buf[0],buf[1],buf[2]);read the /a.txt.ln symbol file
fs_set_blk_dev("ide","0:1", 0); fs_unlink("/a.txt.ln"); //unlink itThe second and third may not trigger a crash however, can be observed by inserting logging before the memset/memcpy function.
Sorry, I just found that this issue was already fixed in erofs-utils: https://git.kernel.org/pub/scm/linux/kernel/git/xiang/erofs-utils.git/commit...
Would you mind checking if the patch above fixes the issue?
Hi Jianan, Would you mind backporting this patch for u-boot?
Thanks, Gao Xiang
Best regards
Gao Xiang hsiangkao@linux.alibaba.com 于2024年6月5日周三 05:10写道：
On 2024/6/5 06:53, jianqiang wang wrote:
Hi Das U-Boot developers,
...
- in file fs/erofs/data.c, function z_erofs_read_one_data, the node
data is read from the storage, however, without a proper check, the data can be corrupted. For example, the inode data is used in function z_erofs_read_data, map.m_llen will be calculated to a very large value, which means the length variable will be very large. It will cause a large memory clear with memset(buffer + end - offset, 0, length);
Would you mind giving a reproducer or a crafted image to trigger this? Or it's your pure observation.
Thanks, Gao XIang
On 2024/6/5 19:18, Gao Xiang wrote:
Hi Jianqiang,
On 2024/6/5 19:00, jianqiang wang wrote:
Hi,
I do have the crafted image.
payload_00500, payload_00763, payload_00846 can be used to reproduce 1,2,3 vulnerabilities respectively.
Each image is a hard drive file and the vulnerabilities can be triggered by performing the following operations:
struct udevice *dev; uclass_first_device_err(UCLASS_IDE, &dev); //detect the blockdevice
fs_set_blk_dev("ide","0:1", 0); fs_ls("/"); //mount the first partition and list the rootdirectory files
fs_set_blk_dev("ide","0:1", 0); char buf[10]; buf[0] = 0; buf[1] = 0; buf[2] = 0; buf[3] = 0; loff_t actread = 0; fs_read("/a.txt", (ulong)buf, 0, 5, &actread); printf("fd actread %lld %x %x %x\n",actread,buf[0],buf[1],buf[2]);read the /a.txt file
fs_set_blk_dev("ide","0:1", 0); fs_read("/a.txt.ln", (ulong)buf, 0, 5, &actread); printf("fd actread %lld %x %x %x\n",actread,buf[0],buf[1],buf[2]);read the /a.txt.ln symbol file
fs_set_blk_dev("ide","0:1", 0); fs_unlink("/a.txt.ln"); //unlink itThe second and third may not trigger a crash however, can be observed by inserting logging before the memset/memcpy function.
Sorry, I just found that this issue was already fixed in erofs-utils: https://git.kernel.org/pub/scm/linux/kernel/git/xiang/erofs-utils.git/commit...
Would you mind checking if the patch above fixes the issue?
Hi Jianan, Would you mind backporting this patch for u-boot?
https://lore.kernel.org/u-boot/20240605140554.1883-1-jnhuang95@gmail.com/T/#...
Thanks, Jianan
Thanks, Gao Xiang
Best regards
Gao Xiang hsiangkao@linux.alibaba.com 于2024年6月5日周三 05:10写道：
On 2024/6/5 06:53, jianqiang wang wrote:
Hi Das U-Boot developers,
...
- in file fs/erofs/data.c, function z_erofs_read_one_data, the node
data is read from the storage, however, without a proper check, the data can be corrupted. For example, the inode data is used in function z_erofs_read_data, map.m_llen will be calculated to a very large value, which means the length variable will be very large. It will cause a large memory clear with memset(buffer + end - offset, 0, length);
Would you mind giving a reproducer or a crafted image to trigger this? Or it's your pure observation.
Thanks, Gao XIang
participants (3)
-
Gao Xiang -
Jianan Huang -
jianqiang wang