Page History

Many common operating systems such as Windows and UNIX support file links, including hard links, symbolic (soft) links, and virtual drives. Hard links can be created in UNIX with the ln command or in Windows operating systems by calling the CreateHardLink() function. Symbolic links can be created in UNIX using the ln -s command or in Windows by using directory junctions in NTFS or the Linkd.exe (Win 2K resource kit) or "junction" freeware. Virtual drives can also be created in Windows using the subst command.

...

When creating new files, it may be possible to use functions that only create a new file where only where a file does not already exist. This prevents the application from overwriting an existing file during file creation. (see See FIO03-C. Do not make assumptions about fopen() and file creation.).

In rare cases, it is necessary to check for the existence of symbolic or hard links to ensure that a program is reading from an intended file and not a different file in another directory. In these cases, avoid creating a race condition when checking for the existence of symbolic links. (see See POS35-C. Avoid race conditions while checking for the existence of a symbolic link.).

Noncompliant Code Example

This noncompliant code example opens the file specified by the string file_name for read/write access and then writes user-supplied data to the file.:

char *file_name = /* something */;
char *userbuf = /* something */;
unsigned int userlen = /* length of userbuf string */;

int fd = open(file_name, O_RDWR);
if (fd == -1) {
   /* handle error */
}
write(fd, userbuf, userlen);

If the process is running with elevated privileges, an attacker can exploit this code, for example, by replacing the file with a symbolic link to the /etc/passwd authentication file. The attacker can then overwrite data stored in the password file to create a new root account with no password. As a result, this attack can be used to gain root privileges on a vulnerable system.

Compliant Solution (Linux 2.1.126+, FreeBSD, Solaris 10, POSIX.1-2008 O_NOFOLLOW)

Wiki MarkupSome systems provide the O_NOFOLLOW flag to help mitigate this problem. The flag will be required by the forthcoming The flag is required by the POSIX.1-2008 standard , and so will become more portable over time \ [[Austin Group 08|AA. C References#Austin Group 08]\]. If the flag is set and the supplied {{file_name}} is a symbolic link, then the open will Open Group 2008]. If the flag is set and the supplied file_name is a symbolic link, then the open will fail.

char *file_name = /* something */;
char *userbuf = /* something */;
unsigned int userlen = /* length of userbuf string */;

int fd = open(file_name, O_RDWR | O_NOFOLLOW);
if (fd == -1) {
  /* handle error */
}
write(fd, userbuf, userlen);

...

This compliant solution uses the lstat-fopen-fstat idiom illustrated in FIO05-C. Identify files using multiple file attributes.:

char *file_name = /* some value */;

struct stat orig_st;
if (lstat( file_name, &orig_st) != 0) {
  /* handle error */
}

if (!S_ISREG( orig_st.st_mode)) {
  /* file is irregular or symlink */
}

int fd = open(file_name, O_RDWR);
if (fd == -1) {
  /* handle error */
}

struct stat new_st;
if (fstat(fd, &new_st) != 0) {
  /* handle error */
}

if (orig_st.st_dev != new_st.st_dev ||
    orig_st.st_ino != new_st.st_ino) {
  /* file was tampered with during race window */
}

/* ... file is good, operate on fd ... */

This code is still subject to a time-of-check, time-of-use (TOCTOU) race condition, but before doing any operation on the file, it verifies that the file opened is the same file as was previously checked (by checking the file's device and i-node.) As a result, the code will recognize if an attacker has tampered with the file during the race window , and can operate accordingly.

...

One way to deal with hard links is simply to disallow opening of any file with two or more hard links. The following code snippet, when inserted into the previous example, will identify if a file has multiple hard links.:

if (orig_st.st_nlink > 1) {
  /* file has multiple hard links */
}

Because a hard link may not be created if the link and the linked-to file are on different devices, many platforms will place system-critical files on a different device than userdevice from the one where user-editable files are kept. For instance, the / directory, which contains critical system files like /etc/passwd, would live on one hard drive, while the /home directory, which contains user-editable files, would reside on a separate hard drive. This prevents users, for example, from creating hard links to /etc/passwd.

...

Automated Detection

...

...

...

...

Related Vulnerabilities

Search for vulnerabilities resulting from the violation of this rule on the CERT website.

References

\[[Austin Group 08|AA. C References#Austin Group 08]\]
\[[MITRE 07|AA. C References#MITRE 07]\] [CWE ID 59|http://cwe.mitre.org/data/definitions/59.html], "Failure to Resolve Links Before File Access (aka 'Link Following')"
\[[Open Group 04|AA. C References#Open Group 04]\] [{{open()}}|http://www.opengroup.org/onlinepubs/009695399/functions/open.html]
\[[Seacord 05|AA. C References#Seacord 05]\] Chapter 7, "File I/O"

Related Guidelines

Bibliography

...

Image Added Image Added POS00-C. Avoid race conditions with multiple threads      50. POSIX (POS)       Image Modified