Re: [PATCH] ima: bound line scan in ima_read_policy() to fix OOB read

On Fri, 2026-08-14 at 10:54 +0200, Fabrice Derepas wrote:
> ima_read_policy() loads a policy file with kernel_read_file_from_path()
> and splits it into lines with
> 
> 	while (size > 0 && (p = strsep(&datap, "\n")))
> 
> kernel_read_file() allocates the destination with vmalloc(i_size) --
> exactly i_size bytes, and writes no NUL terminator. strsep()'s scan for
> the next '\n' is not bounded by @size, so when the last line has no
> trailing newline the scan runs off the end of the buffer (CWE-125). When
> i_size is a multiple of PAGE_SIZE the allocation has no slack and the
> read walks into the vmalloc guard page and faults.
> 
> Reproduced under KASAN in a VM: writing the path of a page-aligned
> policy file with no trailing newline to <securityfs>/ima/policy oopses:
> 
>   BUG: unable to handle page fault for address: ffffc90000032000
>   #PF: supervisor read access in kernel mode
>   RIP: 0010:strsep+0x7a/0xd0
>   Call Trace:
>    ima_write_policy+0x1f4/0x260
>    vfs_write+0x16a/0x6f0
>    ksys_write+0xcb/0x160
>    do_syscall_64+0xe0/0x5a0
> 
> This requires CAP_MAC_ADMIN (the policy file is mode 0200), but a policy
> file that does not end in a newline is an ordinary, non-malicious
> condition, so a legitimate policy load can crash the kernel.
> 
> Walk the buffer with memchr() bounded by the remaining size instead of
> strsep(): terminate each line in place at its newline, and parse a
> NUL-terminated copy of a final line that has none. The explicit per-line
> accounting replaces the old "size -= rc" step, whose off-by-one
> (ima_parse_add_rule() returns strlen() + 1) made a trailing line without a
> newline fail with -EINVAL; such a policy now loads. The loop now consumes
> the buffer exactly, so the trailing "if (size) return -EINVAL" is dropped.

Looks unnecessarily complicated. I would replicate instead the same
behavior of ima_write_policy() to allocate a buffer with an additional
byte for the terminator.

Read the inode size, vmalloc() size + 1, set the terminator, and pass
the buffer to kernel_read_file_from_path().

I would still pass NULL for file_size to save ourselves from rechecking
if it changed after the kernel_read_file_from_path() call.

Thanks

Roberto

> Fixes: 7429b092811f ("ima: load policy using path")
> Assisted-by: copilot-cli:claude-opus-4-6 frama-c
> Signed-off-by: Fabrice Derepas <[email protected]>
> ---
> Tested under KASAN (CONFIG_KASAN_GENERIC + CONFIG_KASAN_VMALLOC) in QEMU,
> loading a policy via "echo /path > <securityfs>/ima/policy":
> 
>   - page-aligned file, no trailing newline: unpatched -> guard-page oops
>     in strsep()/ima_read_policy() (trace above); patched -> no fault, the
>     load fails cleanly with -EINVAL on the (garbage) content.
>   - valid policy with a trailing newline: loads before and after.
>   - valid rule with no trailing newline: unpatched -> -EINVAL (the size
>     underflow); patched -> loads.
> 
> lib/string.o is not KASAN-instrumented, so the over-read is caught by the
> vmalloc guard page rather than a shadow report; the confirmation is the
> page-fault oops with strsep()/ima_write_policy() in the trace.
> 
>  security/integrity/ima/ima_fs.c | 46 ++++++++++++++++++++++++++-------
>  1 file changed, 36 insertions(+), 10 deletions(-)
> 
> diff --git a/security/integrity/ima/ima_fs.c b/security/integrity/ima/ima_fs.c
> index 174a94740..7b530b130 100644
> --- a/security/integrity/ima/ima_fs.c
> +++ b/security/integrity/ima/ima_fs.c
> @@ -526,12 +526,10 @@ static const struct file_operations ima_ascii_measurements_staged_ops = {
>  static ssize_t ima_read_policy(char *path)
>  {
>  	void *data = NULL;
> -	char *datap;
> -	size_t size;
> +	char *datap, *eol, *p;
> +	size_t size, linelen;
>  	int rc, pathlen = strlen(path);
>  
> -	char *p;
> -
>  	/* remove \n */
>  	datap = path;
>  	strsep(&datap, "\n");
> @@ -546,21 +544,49 @@ static ssize_t ima_read_policy(char *path)
>  	rc = 0;
>  
>  	datap = data;
> -	while (size > 0 && (p = strsep(&datap, "\n"))) {
> +	while (size > 0) {
> +		eol = memchr(datap, '\n', size);
> +		linelen = eol ? (size_t)(eol - datap) : size;
> +
> +		if (eol) {
> +			/* NUL-terminate the line in place, within bounds. */
> +			*eol = '\0';
> +			p = datap;
> +		} else {
> +			/*
> +			 * kernel_read_file_from_path() does not NUL-terminate
> +			 * the buffer, and it may be exactly i_size bytes long,
> +			 * so a string walk off the end is possible.  The final
> +			 * line without a trailing newline has no room for a
> +			 * terminator; parse a terminated copy instead.
> +			 */
> +			p = kmemdup_nul(datap, linelen, GFP_KERNEL);
> +			if (!p) {
> +				rc = -ENOMEM;
> +				break;
> +			}
> +		}
> +
>  		pr_debug("rule: %s\n", p);
>  		rc = ima_parse_add_rule(p);
> +		if (!eol)
> +			kfree(p);
>  		if (rc < 0)
>  			break;
> -		size -= rc;
> +		rc = 0;
> +
> +		datap += linelen;
> +		size -= linelen;
> +		if (eol) {
> +			datap++;	/* skip the newline */
> +			size--;
> +		}
>  	}
>  
>  	vfree(data);
>  	if (rc < 0)
>  		return rc;
> -	else if (size)
> -		return -EINVAL;
> -	else
> -		return pathlen;
> +	return pathlen;
>  }
>  
>  static ssize_t ima_write_policy(struct file *file, const char __user *buf,
> 
> base-commit: d58772d8520c7ef247c4b95c9bd76d3a25da9ff5