]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
5f23f6d0 DH |
2 | /* |
3 | * Tests x86 Memory Protection Keys (see Documentation/x86/protection-keys.txt) | |
4 | * | |
5 | * There are examples in here of: | |
6 | * * how to set protection keys on memory | |
7 | * * how to set/clear bits in PKRU (the rights register) | |
8 | * * how to handle SEGV_PKRU signals and extract pkey-relevant | |
9 | * information from the siginfo | |
10 | * | |
11 | * Things to add: | |
12 | * make sure KSM and KSM COW breaking works | |
13 | * prefault pages in at malloc, or not | |
14 | * protect MPX bounds tables with protection keys? | |
15 | * make sure VMA splitting/merging is working correctly | |
16 | * OOMs can destroy mm->mmap (see exit_mmap()), so make sure it is immune to pkeys | |
17 | * look for pkey "leaks" where it is still set on a VMA but "freed" back to the kernel | |
18 | * do a plain mprotect() to a mprotect_pkey() area and make sure the pkey sticks | |
19 | * | |
20 | * Compile like this: | |
21 | * gcc -o protection_keys -O2 -g -std=gnu99 -pthread -Wall protection_keys.c -lrt -ldl -lm | |
22 | * gcc -m32 -o protection_keys_32 -O2 -g -std=gnu99 -pthread -Wall protection_keys.c -lrt -ldl -lm | |
23 | */ | |
24 | #define _GNU_SOURCE | |
25 | #include <errno.h> | |
26 | #include <linux/futex.h> | |
27 | #include <sys/time.h> | |
28 | #include <sys/syscall.h> | |
29 | #include <string.h> | |
30 | #include <stdio.h> | |
31 | #include <stdint.h> | |
32 | #include <stdbool.h> | |
33 | #include <signal.h> | |
34 | #include <assert.h> | |
35 | #include <stdlib.h> | |
36 | #include <ucontext.h> | |
37 | #include <sys/mman.h> | |
38 | #include <sys/types.h> | |
39 | #include <sys/wait.h> | |
40 | #include <sys/stat.h> | |
41 | #include <fcntl.h> | |
42 | #include <unistd.h> | |
43 | #include <sys/ptrace.h> | |
44 | #include <setjmp.h> | |
45 | ||
46 | #include "pkey-helpers.h" | |
47 | ||
48 | int iteration_nr = 1; | |
49 | int test_nr; | |
50 | ||
51 | unsigned int shadow_pkru; | |
52 | ||
53 | #define HPAGE_SIZE (1UL<<21) | |
54 | #define ARRAY_SIZE(x) (sizeof(x) / sizeof(*(x))) | |
55 | #define ALIGN_UP(x, align_to) (((x) + ((align_to)-1)) & ~((align_to)-1)) | |
56 | #define ALIGN_DOWN(x, align_to) ((x) & ~((align_to)-1)) | |
57 | #define ALIGN_PTR_UP(p, ptr_align_to) ((typeof(p))ALIGN_UP((unsigned long)(p), ptr_align_to)) | |
58 | #define ALIGN_PTR_DOWN(p, ptr_align_to) ((typeof(p))ALIGN_DOWN((unsigned long)(p), ptr_align_to)) | |
59 | #define __stringify_1(x...) #x | |
60 | #define __stringify(x...) __stringify_1(x) | |
61 | ||
62 | #define PTR_ERR_ENOTSUP ((void *)-ENOTSUP) | |
63 | ||
64 | int dprint_in_signal; | |
65 | char dprint_in_signal_buffer[DPRINT_IN_SIGNAL_BUF_SIZE]; | |
66 | ||
67 | extern void abort_hooks(void); | |
68 | #define pkey_assert(condition) do { \ | |
69 | if (!(condition)) { \ | |
70 | dprintf0("assert() at %s::%d test_nr: %d iteration: %d\n", \ | |
71 | __FILE__, __LINE__, \ | |
72 | test_nr, iteration_nr); \ | |
73 | dprintf0("errno at assert: %d", errno); \ | |
74 | abort_hooks(); \ | |
75 | assert(condition); \ | |
76 | } \ | |
77 | } while (0) | |
78 | #define raw_assert(cond) assert(cond) | |
79 | ||
80 | void cat_into_file(char *str, char *file) | |
81 | { | |
82 | int fd = open(file, O_RDWR); | |
83 | int ret; | |
84 | ||
85 | dprintf2("%s(): writing '%s' to '%s'\n", __func__, str, file); | |
86 | /* | |
87 | * these need to be raw because they are called under | |
88 | * pkey_assert() | |
89 | */ | |
90 | raw_assert(fd >= 0); | |
91 | ret = write(fd, str, strlen(str)); | |
92 | if (ret != strlen(str)) { | |
93 | perror("write to file failed"); | |
94 | fprintf(stderr, "filename: '%s' str: '%s'\n", file, str); | |
95 | raw_assert(0); | |
96 | } | |
97 | close(fd); | |
98 | } | |
99 | ||
100 | #if CONTROL_TRACING > 0 | |
101 | static int warned_tracing; | |
102 | int tracing_root_ok(void) | |
103 | { | |
104 | if (geteuid() != 0) { | |
105 | if (!warned_tracing) | |
106 | fprintf(stderr, "WARNING: not run as root, " | |
107 | "can not do tracing control\n"); | |
108 | warned_tracing = 1; | |
109 | return 0; | |
110 | } | |
111 | return 1; | |
112 | } | |
113 | #endif | |
114 | ||
115 | void tracing_on(void) | |
116 | { | |
117 | #if CONTROL_TRACING > 0 | |
118 | #define TRACEDIR "/sys/kernel/debug/tracing" | |
119 | char pidstr[32]; | |
120 | ||
121 | if (!tracing_root_ok()) | |
122 | return; | |
123 | ||
124 | sprintf(pidstr, "%d", getpid()); | |
125 | cat_into_file("0", TRACEDIR "/tracing_on"); | |
126 | cat_into_file("\n", TRACEDIR "/trace"); | |
127 | if (1) { | |
128 | cat_into_file("function_graph", TRACEDIR "/current_tracer"); | |
129 | cat_into_file("1", TRACEDIR "/options/funcgraph-proc"); | |
130 | } else { | |
131 | cat_into_file("nop", TRACEDIR "/current_tracer"); | |
132 | } | |
133 | cat_into_file(pidstr, TRACEDIR "/set_ftrace_pid"); | |
134 | cat_into_file("1", TRACEDIR "/tracing_on"); | |
135 | dprintf1("enabled tracing\n"); | |
136 | #endif | |
137 | } | |
138 | ||
139 | void tracing_off(void) | |
140 | { | |
141 | #if CONTROL_TRACING > 0 | |
142 | if (!tracing_root_ok()) | |
143 | return; | |
144 | cat_into_file("0", "/sys/kernel/debug/tracing/tracing_on"); | |
145 | #endif | |
146 | } | |
147 | ||
148 | void abort_hooks(void) | |
149 | { | |
150 | fprintf(stderr, "running %s()...\n", __func__); | |
151 | tracing_off(); | |
152 | #ifdef SLEEP_ON_ABORT | |
153 | sleep(SLEEP_ON_ABORT); | |
154 | #endif | |
155 | } | |
156 | ||
157 | static inline void __page_o_noops(void) | |
158 | { | |
159 | /* 8-bytes of instruction * 512 bytes = 1 page */ | |
160 | asm(".rept 512 ; nopl 0x7eeeeeee(%eax) ; .endr"); | |
161 | } | |
162 | ||
163 | /* | |
164 | * This attempts to have roughly a page of instructions followed by a few | |
165 | * instructions that do a write, and another page of instructions. That | |
166 | * way, we are pretty sure that the write is in the second page of | |
167 | * instructions and has at least a page of padding behind it. | |
168 | * | |
169 | * *That* lets us be sure to madvise() away the write instruction, which | |
170 | * will then fault, which makes sure that the fault code handles | |
171 | * execute-only memory properly. | |
172 | */ | |
173 | __attribute__((__aligned__(PAGE_SIZE))) | |
174 | void lots_o_noops_around_write(int *write_to_me) | |
175 | { | |
176 | dprintf3("running %s()\n", __func__); | |
177 | __page_o_noops(); | |
178 | /* Assume this happens in the second page of instructions: */ | |
179 | *write_to_me = __LINE__; | |
180 | /* pad out by another page: */ | |
181 | __page_o_noops(); | |
182 | dprintf3("%s() done\n", __func__); | |
183 | } | |
184 | ||
185 | /* Define some kernel-like types */ | |
186 | #define u8 uint8_t | |
187 | #define u16 uint16_t | |
188 | #define u32 uint32_t | |
189 | #define u64 uint64_t | |
190 | ||
191 | #ifdef __i386__ | |
693cb558 AL |
192 | |
193 | #ifndef SYS_mprotect_key | |
194 | # define SYS_mprotect_key 380 | |
195 | #endif | |
196 | #ifndef SYS_pkey_alloc | |
197 | # define SYS_pkey_alloc 381 | |
198 | # define SYS_pkey_free 382 | |
199 | #endif | |
5f23f6d0 | 200 | #define REG_IP_IDX REG_EIP |
2195bff0 | 201 | #define si_pkey_offset 0x14 |
693cb558 | 202 | |
5f23f6d0 | 203 | #else |
693cb558 AL |
204 | |
205 | #ifndef SYS_mprotect_key | |
206 | # define SYS_mprotect_key 329 | |
207 | #endif | |
208 | #ifndef SYS_pkey_alloc | |
209 | # define SYS_pkey_alloc 330 | |
210 | # define SYS_pkey_free 331 | |
211 | #endif | |
5f23f6d0 DH |
212 | #define REG_IP_IDX REG_RIP |
213 | #define si_pkey_offset 0x20 | |
693cb558 | 214 | |
5f23f6d0 DH |
215 | #endif |
216 | ||
217 | void dump_mem(void *dumpme, int len_bytes) | |
218 | { | |
219 | char *c = (void *)dumpme; | |
220 | int i; | |
221 | ||
222 | for (i = 0; i < len_bytes; i += sizeof(u64)) { | |
223 | u64 *ptr = (u64 *)(c + i); | |
224 | dprintf1("dump[%03d][@%p]: %016jx\n", i, ptr, *ptr); | |
225 | } | |
226 | } | |
227 | ||
d12fe87e EB |
228 | #define SEGV_BNDERR 3 /* failed address bound checks */ |
229 | #define SEGV_PKUERR 4 | |
5f23f6d0 DH |
230 | |
231 | static char *si_code_str(int si_code) | |
232 | { | |
d12fe87e | 233 | if (si_code == SEGV_MAPERR) |
5f23f6d0 | 234 | return "SEGV_MAPERR"; |
d12fe87e | 235 | if (si_code == SEGV_ACCERR) |
5f23f6d0 | 236 | return "SEGV_ACCERR"; |
d12fe87e | 237 | if (si_code == SEGV_BNDERR) |
5f23f6d0 | 238 | return "SEGV_BNDERR"; |
d12fe87e | 239 | if (si_code == SEGV_PKUERR) |
5f23f6d0 DH |
240 | return "SEGV_PKUERR"; |
241 | return "UNKNOWN"; | |
242 | } | |
243 | ||
244 | int pkru_faults; | |
245 | int last_si_pkey = -1; | |
246 | void signal_handler(int signum, siginfo_t *si, void *vucontext) | |
247 | { | |
248 | ucontext_t *uctxt = vucontext; | |
249 | int trapno; | |
250 | unsigned long ip; | |
251 | char *fpregs; | |
252 | u32 *pkru_ptr; | |
91c49c2d | 253 | u64 siginfo_pkey; |
5f23f6d0 DH |
254 | u32 *si_pkey_ptr; |
255 | int pkru_offset; | |
256 | fpregset_t fpregset; | |
257 | ||
258 | dprint_in_signal = 1; | |
259 | dprintf1(">>>>===============SIGSEGV============================\n"); | |
260 | dprintf1("%s()::%d, pkru: 0x%x shadow: %x\n", __func__, __LINE__, | |
261 | __rdpkru(), shadow_pkru); | |
262 | ||
263 | trapno = uctxt->uc_mcontext.gregs[REG_TRAPNO]; | |
264 | ip = uctxt->uc_mcontext.gregs[REG_IP_IDX]; | |
265 | fpregset = uctxt->uc_mcontext.fpregs; | |
266 | fpregs = (void *)fpregset; | |
267 | ||
268 | dprintf2("%s() trapno: %d ip: 0x%lx info->si_code: %s/%d\n", __func__, | |
269 | trapno, ip, si_code_str(si->si_code), si->si_code); | |
270 | #ifdef __i386__ | |
271 | /* | |
272 | * 32-bit has some extra padding so that userspace can tell whether | |
273 | * the XSTATE header is present in addition to the "legacy" FPU | |
274 | * state. We just assume that it is here. | |
275 | */ | |
276 | fpregs += 0x70; | |
277 | #endif | |
278 | pkru_offset = pkru_xstate_offset(); | |
279 | pkru_ptr = (void *)(&fpregs[pkru_offset]); | |
280 | ||
281 | dprintf1("siginfo: %p\n", si); | |
282 | dprintf1(" fpregs: %p\n", fpregs); | |
283 | /* | |
284 | * If we got a PKRU fault, we *HAVE* to have at least one bit set in | |
285 | * here. | |
286 | */ | |
287 | dprintf1("pkru_xstate_offset: %d\n", pkru_xstate_offset()); | |
288 | if (DEBUG_LEVEL > 4) | |
289 | dump_mem(pkru_ptr - 128, 256); | |
290 | pkey_assert(*pkru_ptr); | |
291 | ||
292 | si_pkey_ptr = (u32 *)(((u8 *)si) + si_pkey_offset); | |
293 | dprintf1("si_pkey_ptr: %p\n", si_pkey_ptr); | |
294 | dump_mem(si_pkey_ptr - 8, 24); | |
91c49c2d DH |
295 | siginfo_pkey = *si_pkey_ptr; |
296 | pkey_assert(siginfo_pkey < NR_PKEYS); | |
297 | last_si_pkey = siginfo_pkey; | |
5f23f6d0 DH |
298 | |
299 | if ((si->si_code == SEGV_MAPERR) || | |
300 | (si->si_code == SEGV_ACCERR) || | |
301 | (si->si_code == SEGV_BNDERR)) { | |
302 | printf("non-PK si_code, exiting...\n"); | |
303 | exit(4); | |
304 | } | |
305 | ||
306 | dprintf1("signal pkru from xsave: %08x\n", *pkru_ptr); | |
307 | /* need __rdpkru() version so we do not do shadow_pkru checking */ | |
308 | dprintf1("signal pkru from pkru: %08x\n", __rdpkru()); | |
91c49c2d | 309 | dprintf1("pkey from siginfo: %jx\n", siginfo_pkey); |
5f23f6d0 DH |
310 | *(u64 *)pkru_ptr = 0x00000000; |
311 | dprintf1("WARNING: set PRKU=0 to allow faulting instruction to continue\n"); | |
312 | pkru_faults++; | |
313 | dprintf1("<<<<==================================================\n"); | |
314 | return; | |
315 | if (trapno == 14) { | |
316 | fprintf(stderr, | |
317 | "ERROR: In signal handler, page fault, trapno = %d, ip = %016lx\n", | |
318 | trapno, ip); | |
319 | fprintf(stderr, "si_addr %p\n", si->si_addr); | |
320 | fprintf(stderr, "REG_ERR: %lx\n", | |
321 | (unsigned long)uctxt->uc_mcontext.gregs[REG_ERR]); | |
322 | exit(1); | |
323 | } else { | |
324 | fprintf(stderr, "unexpected trap %d! at 0x%lx\n", trapno, ip); | |
325 | fprintf(stderr, "si_addr %p\n", si->si_addr); | |
326 | fprintf(stderr, "REG_ERR: %lx\n", | |
327 | (unsigned long)uctxt->uc_mcontext.gregs[REG_ERR]); | |
328 | exit(2); | |
329 | } | |
330 | dprint_in_signal = 0; | |
331 | } | |
332 | ||
333 | int wait_all_children(void) | |
334 | { | |
335 | int status; | |
336 | return waitpid(-1, &status, 0); | |
337 | } | |
338 | ||
339 | void sig_chld(int x) | |
340 | { | |
341 | dprint_in_signal = 1; | |
342 | dprintf2("[%d] SIGCHLD: %d\n", getpid(), x); | |
343 | dprint_in_signal = 0; | |
344 | } | |
345 | ||
346 | void setup_sigsegv_handler(void) | |
347 | { | |
348 | int r, rs; | |
349 | struct sigaction newact; | |
350 | struct sigaction oldact; | |
351 | ||
352 | /* #PF is mapped to sigsegv */ | |
353 | int signum = SIGSEGV; | |
354 | ||
355 | newact.sa_handler = 0; | |
356 | newact.sa_sigaction = signal_handler; | |
357 | ||
358 | /*sigset_t - signals to block while in the handler */ | |
359 | /* get the old signal mask. */ | |
360 | rs = sigprocmask(SIG_SETMASK, 0, &newact.sa_mask); | |
361 | pkey_assert(rs == 0); | |
362 | ||
363 | /* call sa_sigaction, not sa_handler*/ | |
364 | newact.sa_flags = SA_SIGINFO; | |
365 | ||
366 | newact.sa_restorer = 0; /* void(*)(), obsolete */ | |
367 | r = sigaction(signum, &newact, &oldact); | |
368 | r = sigaction(SIGALRM, &newact, &oldact); | |
369 | pkey_assert(r == 0); | |
370 | } | |
371 | ||
372 | void setup_handlers(void) | |
373 | { | |
374 | signal(SIGCHLD, &sig_chld); | |
375 | setup_sigsegv_handler(); | |
376 | } | |
377 | ||
378 | pid_t fork_lazy_child(void) | |
379 | { | |
380 | pid_t forkret; | |
381 | ||
382 | forkret = fork(); | |
383 | pkey_assert(forkret >= 0); | |
384 | dprintf3("[%d] fork() ret: %d\n", getpid(), forkret); | |
385 | ||
386 | if (!forkret) { | |
387 | /* in the child */ | |
388 | while (1) { | |
389 | dprintf1("child sleeping...\n"); | |
390 | sleep(30); | |
391 | } | |
392 | } | |
393 | return forkret; | |
394 | } | |
395 | ||
396 | void davecmp(void *_a, void *_b, int len) | |
397 | { | |
398 | int i; | |
399 | unsigned long *a = _a; | |
400 | unsigned long *b = _b; | |
401 | ||
402 | for (i = 0; i < len / sizeof(*a); i++) { | |
403 | if (a[i] == b[i]) | |
404 | continue; | |
405 | ||
406 | dprintf3("[%3d]: a: %016lx b: %016lx\n", i, a[i], b[i]); | |
407 | } | |
408 | } | |
409 | ||
410 | void dumpit(char *f) | |
411 | { | |
412 | int fd = open(f, O_RDONLY); | |
413 | char buf[100]; | |
414 | int nr_read; | |
415 | ||
416 | dprintf2("maps fd: %d\n", fd); | |
417 | do { | |
418 | nr_read = read(fd, &buf[0], sizeof(buf)); | |
419 | write(1, buf, nr_read); | |
420 | } while (nr_read > 0); | |
421 | close(fd); | |
422 | } | |
423 | ||
424 | #define PKEY_DISABLE_ACCESS 0x1 | |
425 | #define PKEY_DISABLE_WRITE 0x2 | |
426 | ||
427 | u32 pkey_get(int pkey, unsigned long flags) | |
428 | { | |
429 | u32 mask = (PKEY_DISABLE_ACCESS|PKEY_DISABLE_WRITE); | |
430 | u32 pkru = __rdpkru(); | |
431 | u32 shifted_pkru; | |
432 | u32 masked_pkru; | |
433 | ||
434 | dprintf1("%s(pkey=%d, flags=%lx) = %x / %d\n", | |
435 | __func__, pkey, flags, 0, 0); | |
436 | dprintf2("%s() raw pkru: %x\n", __func__, pkru); | |
437 | ||
438 | shifted_pkru = (pkru >> (pkey * PKRU_BITS_PER_PKEY)); | |
439 | dprintf2("%s() shifted_pkru: %x\n", __func__, shifted_pkru); | |
440 | masked_pkru = shifted_pkru & mask; | |
441 | dprintf2("%s() masked pkru: %x\n", __func__, masked_pkru); | |
442 | /* | |
443 | * shift down the relevant bits to the lowest two, then | |
444 | * mask off all the other high bits. | |
445 | */ | |
446 | return masked_pkru; | |
447 | } | |
448 | ||
449 | int pkey_set(int pkey, unsigned long rights, unsigned long flags) | |
450 | { | |
451 | u32 mask = (PKEY_DISABLE_ACCESS|PKEY_DISABLE_WRITE); | |
452 | u32 old_pkru = __rdpkru(); | |
453 | u32 new_pkru; | |
454 | ||
455 | /* make sure that 'rights' only contains the bits we expect: */ | |
456 | assert(!(rights & ~mask)); | |
457 | ||
458 | /* copy old pkru */ | |
459 | new_pkru = old_pkru; | |
460 | /* mask out bits from pkey in old value: */ | |
461 | new_pkru &= ~(mask << (pkey * PKRU_BITS_PER_PKEY)); | |
462 | /* OR in new bits for pkey: */ | |
463 | new_pkru |= (rights << (pkey * PKRU_BITS_PER_PKEY)); | |
464 | ||
465 | __wrpkru(new_pkru); | |
466 | ||
467 | dprintf3("%s(pkey=%d, rights=%lx, flags=%lx) = %x pkru now: %x old_pkru: %x\n", | |
468 | __func__, pkey, rights, flags, 0, __rdpkru(), old_pkru); | |
469 | return 0; | |
470 | } | |
471 | ||
472 | void pkey_disable_set(int pkey, int flags) | |
473 | { | |
474 | unsigned long syscall_flags = 0; | |
475 | int ret; | |
476 | int pkey_rights; | |
16846c2d | 477 | u32 orig_pkru = rdpkru(); |
5f23f6d0 DH |
478 | |
479 | dprintf1("START->%s(%d, 0x%x)\n", __func__, | |
480 | pkey, flags); | |
481 | pkey_assert(flags & (PKEY_DISABLE_ACCESS | PKEY_DISABLE_WRITE)); | |
482 | ||
483 | pkey_rights = pkey_get(pkey, syscall_flags); | |
484 | ||
485 | dprintf1("%s(%d) pkey_get(%d): %x\n", __func__, | |
486 | pkey, pkey, pkey_rights); | |
487 | pkey_assert(pkey_rights >= 0); | |
488 | ||
489 | pkey_rights |= flags; | |
490 | ||
491 | ret = pkey_set(pkey, pkey_rights, syscall_flags); | |
492 | assert(!ret); | |
493 | /*pkru and flags have the same format */ | |
494 | shadow_pkru |= flags << (pkey * 2); | |
495 | dprintf1("%s(%d) shadow: 0x%x\n", __func__, pkey, shadow_pkru); | |
496 | ||
497 | pkey_assert(ret >= 0); | |
498 | ||
499 | pkey_rights = pkey_get(pkey, syscall_flags); | |
500 | dprintf1("%s(%d) pkey_get(%d): %x\n", __func__, | |
501 | pkey, pkey, pkey_rights); | |
502 | ||
503 | dprintf1("%s(%d) pkru: 0x%x\n", __func__, pkey, rdpkru()); | |
504 | if (flags) | |
505 | pkey_assert(rdpkru() > orig_pkru); | |
506 | dprintf1("END<---%s(%d, 0x%x)\n", __func__, | |
507 | pkey, flags); | |
508 | } | |
509 | ||
510 | void pkey_disable_clear(int pkey, int flags) | |
511 | { | |
512 | unsigned long syscall_flags = 0; | |
513 | int ret; | |
514 | int pkey_rights = pkey_get(pkey, syscall_flags); | |
515 | u32 orig_pkru = rdpkru(); | |
516 | ||
517 | pkey_assert(flags & (PKEY_DISABLE_ACCESS | PKEY_DISABLE_WRITE)); | |
518 | ||
519 | dprintf1("%s(%d) pkey_get(%d): %x\n", __func__, | |
520 | pkey, pkey, pkey_rights); | |
521 | pkey_assert(pkey_rights >= 0); | |
522 | ||
523 | pkey_rights |= flags; | |
524 | ||
525 | ret = pkey_set(pkey, pkey_rights, 0); | |
526 | /* pkru and flags have the same format */ | |
527 | shadow_pkru &= ~(flags << (pkey * 2)); | |
528 | pkey_assert(ret >= 0); | |
529 | ||
530 | pkey_rights = pkey_get(pkey, syscall_flags); | |
531 | dprintf1("%s(%d) pkey_get(%d): %x\n", __func__, | |
532 | pkey, pkey, pkey_rights); | |
533 | ||
534 | dprintf1("%s(%d) pkru: 0x%x\n", __func__, pkey, rdpkru()); | |
535 | if (flags) | |
536 | assert(rdpkru() > orig_pkru); | |
537 | } | |
538 | ||
539 | void pkey_write_allow(int pkey) | |
540 | { | |
541 | pkey_disable_clear(pkey, PKEY_DISABLE_WRITE); | |
542 | } | |
543 | void pkey_write_deny(int pkey) | |
544 | { | |
545 | pkey_disable_set(pkey, PKEY_DISABLE_WRITE); | |
546 | } | |
547 | void pkey_access_allow(int pkey) | |
548 | { | |
549 | pkey_disable_clear(pkey, PKEY_DISABLE_ACCESS); | |
550 | } | |
551 | void pkey_access_deny(int pkey) | |
552 | { | |
553 | pkey_disable_set(pkey, PKEY_DISABLE_ACCESS); | |
554 | } | |
555 | ||
556 | int sys_mprotect_pkey(void *ptr, size_t size, unsigned long orig_prot, | |
557 | unsigned long pkey) | |
558 | { | |
559 | int sret; | |
560 | ||
561 | dprintf2("%s(0x%p, %zx, prot=%lx, pkey=%lx)\n", __func__, | |
562 | ptr, size, orig_prot, pkey); | |
563 | ||
564 | errno = 0; | |
565 | sret = syscall(SYS_mprotect_key, ptr, size, orig_prot, pkey); | |
566 | if (errno) { | |
567 | dprintf2("SYS_mprotect_key sret: %d\n", sret); | |
568 | dprintf2("SYS_mprotect_key prot: 0x%lx\n", orig_prot); | |
569 | dprintf2("SYS_mprotect_key failed, errno: %d\n", errno); | |
570 | if (DEBUG_LEVEL >= 2) | |
571 | perror("SYS_mprotect_pkey"); | |
572 | } | |
573 | return sret; | |
574 | } | |
575 | ||
576 | int sys_pkey_alloc(unsigned long flags, unsigned long init_val) | |
577 | { | |
578 | int ret = syscall(SYS_pkey_alloc, flags, init_val); | |
579 | dprintf1("%s(flags=%lx, init_val=%lx) syscall ret: %d errno: %d\n", | |
580 | __func__, flags, init_val, ret, errno); | |
581 | return ret; | |
582 | } | |
583 | ||
584 | int alloc_pkey(void) | |
585 | { | |
586 | int ret; | |
587 | unsigned long init_val = 0x0; | |
588 | ||
589 | dprintf1("alloc_pkey()::%d, pkru: 0x%x shadow: %x\n", | |
590 | __LINE__, __rdpkru(), shadow_pkru); | |
591 | ret = sys_pkey_alloc(0, init_val); | |
592 | /* | |
593 | * pkey_alloc() sets PKRU, so we need to reflect it in | |
594 | * shadow_pkru: | |
595 | */ | |
596 | dprintf4("alloc_pkey()::%d, ret: %d pkru: 0x%x shadow: 0x%x\n", | |
597 | __LINE__, ret, __rdpkru(), shadow_pkru); | |
598 | if (ret) { | |
599 | /* clear both the bits: */ | |
600 | shadow_pkru &= ~(0x3 << (ret * 2)); | |
601 | dprintf4("alloc_pkey()::%d, ret: %d pkru: 0x%x shadow: 0x%x\n", | |
602 | __LINE__, ret, __rdpkru(), shadow_pkru); | |
603 | /* | |
604 | * move the new state in from init_val | |
605 | * (remember, we cheated and init_val == pkru format) | |
606 | */ | |
607 | shadow_pkru |= (init_val << (ret * 2)); | |
608 | } | |
609 | dprintf4("alloc_pkey()::%d, ret: %d pkru: 0x%x shadow: 0x%x\n", | |
610 | __LINE__, ret, __rdpkru(), shadow_pkru); | |
611 | dprintf1("alloc_pkey()::%d errno: %d\n", __LINE__, errno); | |
612 | /* for shadow checking: */ | |
613 | rdpkru(); | |
614 | dprintf4("alloc_pkey()::%d, ret: %d pkru: 0x%x shadow: 0x%x\n", | |
615 | __LINE__, ret, __rdpkru(), shadow_pkru); | |
616 | return ret; | |
617 | } | |
618 | ||
619 | int sys_pkey_free(unsigned long pkey) | |
620 | { | |
621 | int ret = syscall(SYS_pkey_free, pkey); | |
622 | dprintf1("%s(pkey=%ld) syscall ret: %d\n", __func__, pkey, ret); | |
623 | return ret; | |
624 | } | |
625 | ||
626 | /* | |
627 | * I had a bug where pkey bits could be set by mprotect() but | |
628 | * not cleared. This ensures we get lots of random bit sets | |
629 | * and clears on the vma and pte pkey bits. | |
630 | */ | |
631 | int alloc_random_pkey(void) | |
632 | { | |
633 | int max_nr_pkey_allocs; | |
634 | int ret; | |
635 | int i; | |
636 | int alloced_pkeys[NR_PKEYS]; | |
637 | int nr_alloced = 0; | |
638 | int random_index; | |
639 | memset(alloced_pkeys, 0, sizeof(alloced_pkeys)); | |
640 | ||
641 | /* allocate every possible key and make a note of which ones we got */ | |
642 | max_nr_pkey_allocs = NR_PKEYS; | |
643 | max_nr_pkey_allocs = 1; | |
644 | for (i = 0; i < max_nr_pkey_allocs; i++) { | |
645 | int new_pkey = alloc_pkey(); | |
646 | if (new_pkey < 0) | |
647 | break; | |
648 | alloced_pkeys[nr_alloced++] = new_pkey; | |
649 | } | |
650 | ||
651 | pkey_assert(nr_alloced > 0); | |
652 | /* select a random one out of the allocated ones */ | |
653 | random_index = rand() % nr_alloced; | |
654 | ret = alloced_pkeys[random_index]; | |
655 | /* now zero it out so we don't free it next */ | |
656 | alloced_pkeys[random_index] = 0; | |
657 | ||
658 | /* go through the allocated ones that we did not want and free them */ | |
659 | for (i = 0; i < nr_alloced; i++) { | |
660 | int free_ret; | |
661 | if (!alloced_pkeys[i]) | |
662 | continue; | |
663 | free_ret = sys_pkey_free(alloced_pkeys[i]); | |
664 | pkey_assert(!free_ret); | |
665 | } | |
666 | dprintf1("%s()::%d, ret: %d pkru: 0x%x shadow: 0x%x\n", __func__, | |
667 | __LINE__, ret, __rdpkru(), shadow_pkru); | |
668 | return ret; | |
669 | } | |
670 | ||
671 | int mprotect_pkey(void *ptr, size_t size, unsigned long orig_prot, | |
672 | unsigned long pkey) | |
673 | { | |
674 | int nr_iterations = random() % 100; | |
675 | int ret; | |
676 | ||
677 | while (0) { | |
678 | int rpkey = alloc_random_pkey(); | |
679 | ret = sys_mprotect_pkey(ptr, size, orig_prot, pkey); | |
680 | dprintf1("sys_mprotect_pkey(%p, %zx, prot=0x%lx, pkey=%ld) ret: %d\n", | |
681 | ptr, size, orig_prot, pkey, ret); | |
682 | if (nr_iterations-- < 0) | |
683 | break; | |
684 | ||
685 | dprintf1("%s()::%d, ret: %d pkru: 0x%x shadow: 0x%x\n", __func__, | |
686 | __LINE__, ret, __rdpkru(), shadow_pkru); | |
687 | sys_pkey_free(rpkey); | |
688 | dprintf1("%s()::%d, ret: %d pkru: 0x%x shadow: 0x%x\n", __func__, | |
689 | __LINE__, ret, __rdpkru(), shadow_pkru); | |
690 | } | |
691 | pkey_assert(pkey < NR_PKEYS); | |
692 | ||
693 | ret = sys_mprotect_pkey(ptr, size, orig_prot, pkey); | |
694 | dprintf1("mprotect_pkey(%p, %zx, prot=0x%lx, pkey=%ld) ret: %d\n", | |
695 | ptr, size, orig_prot, pkey, ret); | |
696 | pkey_assert(!ret); | |
697 | dprintf1("%s()::%d, ret: %d pkru: 0x%x shadow: 0x%x\n", __func__, | |
698 | __LINE__, ret, __rdpkru(), shadow_pkru); | |
699 | return ret; | |
700 | } | |
701 | ||
702 | struct pkey_malloc_record { | |
703 | void *ptr; | |
704 | long size; | |
705 | }; | |
706 | struct pkey_malloc_record *pkey_malloc_records; | |
707 | long nr_pkey_malloc_records; | |
708 | void record_pkey_malloc(void *ptr, long size) | |
709 | { | |
710 | long i; | |
711 | struct pkey_malloc_record *rec = NULL; | |
712 | ||
713 | for (i = 0; i < nr_pkey_malloc_records; i++) { | |
714 | rec = &pkey_malloc_records[i]; | |
715 | /* find a free record */ | |
716 | if (rec) | |
717 | break; | |
718 | } | |
719 | if (!rec) { | |
720 | /* every record is full */ | |
721 | size_t old_nr_records = nr_pkey_malloc_records; | |
722 | size_t new_nr_records = (nr_pkey_malloc_records * 2 + 1); | |
723 | size_t new_size = new_nr_records * sizeof(struct pkey_malloc_record); | |
724 | dprintf2("new_nr_records: %zd\n", new_nr_records); | |
725 | dprintf2("new_size: %zd\n", new_size); | |
726 | pkey_malloc_records = realloc(pkey_malloc_records, new_size); | |
727 | pkey_assert(pkey_malloc_records != NULL); | |
728 | rec = &pkey_malloc_records[nr_pkey_malloc_records]; | |
729 | /* | |
730 | * realloc() does not initialize memory, so zero it from | |
731 | * the first new record all the way to the end. | |
732 | */ | |
733 | for (i = 0; i < new_nr_records - old_nr_records; i++) | |
734 | memset(rec + i, 0, sizeof(*rec)); | |
735 | } | |
736 | dprintf3("filling malloc record[%d/%p]: {%p, %ld}\n", | |
737 | (int)(rec - pkey_malloc_records), rec, ptr, size); | |
738 | rec->ptr = ptr; | |
739 | rec->size = size; | |
740 | nr_pkey_malloc_records++; | |
741 | } | |
742 | ||
743 | void free_pkey_malloc(void *ptr) | |
744 | { | |
745 | long i; | |
746 | int ret; | |
747 | dprintf3("%s(%p)\n", __func__, ptr); | |
748 | for (i = 0; i < nr_pkey_malloc_records; i++) { | |
749 | struct pkey_malloc_record *rec = &pkey_malloc_records[i]; | |
750 | dprintf4("looking for ptr %p at record[%ld/%p]: {%p, %ld}\n", | |
751 | ptr, i, rec, rec->ptr, rec->size); | |
752 | if ((ptr < rec->ptr) || | |
753 | (ptr >= rec->ptr + rec->size)) | |
754 | continue; | |
755 | ||
756 | dprintf3("found ptr %p at record[%ld/%p]: {%p, %ld}\n", | |
757 | ptr, i, rec, rec->ptr, rec->size); | |
758 | nr_pkey_malloc_records--; | |
759 | ret = munmap(rec->ptr, rec->size); | |
760 | dprintf3("munmap ret: %d\n", ret); | |
761 | pkey_assert(!ret); | |
762 | dprintf3("clearing rec->ptr, rec: %p\n", rec); | |
763 | rec->ptr = NULL; | |
764 | dprintf3("done clearing rec->ptr, rec: %p\n", rec); | |
765 | return; | |
766 | } | |
767 | pkey_assert(false); | |
768 | } | |
769 | ||
770 | ||
771 | void *malloc_pkey_with_mprotect(long size, int prot, u16 pkey) | |
772 | { | |
773 | void *ptr; | |
774 | int ret; | |
775 | ||
776 | rdpkru(); | |
777 | dprintf1("doing %s(size=%ld, prot=0x%x, pkey=%d)\n", __func__, | |
778 | size, prot, pkey); | |
779 | pkey_assert(pkey < NR_PKEYS); | |
780 | ptr = mmap(NULL, size, prot, MAP_ANONYMOUS|MAP_PRIVATE, -1, 0); | |
781 | pkey_assert(ptr != (void *)-1); | |
782 | ret = mprotect_pkey((void *)ptr, PAGE_SIZE, prot, pkey); | |
783 | pkey_assert(!ret); | |
784 | record_pkey_malloc(ptr, size); | |
785 | rdpkru(); | |
786 | ||
787 | dprintf1("%s() for pkey %d @ %p\n", __func__, pkey, ptr); | |
788 | return ptr; | |
789 | } | |
790 | ||
791 | void *malloc_pkey_anon_huge(long size, int prot, u16 pkey) | |
792 | { | |
793 | int ret; | |
794 | void *ptr; | |
795 | ||
796 | dprintf1("doing %s(size=%ld, prot=0x%x, pkey=%d)\n", __func__, | |
797 | size, prot, pkey); | |
798 | /* | |
799 | * Guarantee we can fit at least one huge page in the resulting | |
800 | * allocation by allocating space for 2: | |
801 | */ | |
802 | size = ALIGN_UP(size, HPAGE_SIZE * 2); | |
803 | ptr = mmap(NULL, size, PROT_NONE, MAP_ANONYMOUS|MAP_PRIVATE, -1, 0); | |
804 | pkey_assert(ptr != (void *)-1); | |
805 | record_pkey_malloc(ptr, size); | |
806 | mprotect_pkey(ptr, size, prot, pkey); | |
807 | ||
808 | dprintf1("unaligned ptr: %p\n", ptr); | |
809 | ptr = ALIGN_PTR_UP(ptr, HPAGE_SIZE); | |
810 | dprintf1(" aligned ptr: %p\n", ptr); | |
811 | ret = madvise(ptr, HPAGE_SIZE, MADV_HUGEPAGE); | |
812 | dprintf1("MADV_HUGEPAGE ret: %d\n", ret); | |
813 | ret = madvise(ptr, HPAGE_SIZE, MADV_WILLNEED); | |
814 | dprintf1("MADV_WILLNEED ret: %d\n", ret); | |
815 | memset(ptr, 0, HPAGE_SIZE); | |
816 | ||
817 | dprintf1("mmap()'d thp for pkey %d @ %p\n", pkey, ptr); | |
818 | return ptr; | |
819 | } | |
820 | ||
821 | int hugetlb_setup_ok; | |
822 | #define GET_NR_HUGE_PAGES 10 | |
823 | void setup_hugetlbfs(void) | |
824 | { | |
825 | int err; | |
826 | int fd; | |
5f23f6d0 DH |
827 | char buf[] = "123"; |
828 | ||
829 | if (geteuid() != 0) { | |
830 | fprintf(stderr, "WARNING: not run as root, can not do hugetlb test\n"); | |
831 | return; | |
832 | } | |
833 | ||
834 | cat_into_file(__stringify(GET_NR_HUGE_PAGES), "/proc/sys/vm/nr_hugepages"); | |
835 | ||
836 | /* | |
837 | * Now go make sure that we got the pages and that they | |
838 | * are 2M pages. Someone might have made 1G the default. | |
839 | */ | |
840 | fd = open("/sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages", O_RDONLY); | |
841 | if (fd < 0) { | |
842 | perror("opening sysfs 2M hugetlb config"); | |
843 | return; | |
844 | } | |
845 | ||
846 | /* -1 to guarantee leaving the trailing \0 */ | |
847 | err = read(fd, buf, sizeof(buf)-1); | |
848 | close(fd); | |
849 | if (err <= 0) { | |
850 | perror("reading sysfs 2M hugetlb config"); | |
851 | return; | |
852 | } | |
853 | ||
854 | if (atoi(buf) != GET_NR_HUGE_PAGES) { | |
855 | fprintf(stderr, "could not confirm 2M pages, got: '%s' expected %d\n", | |
856 | buf, GET_NR_HUGE_PAGES); | |
857 | return; | |
858 | } | |
859 | ||
860 | hugetlb_setup_ok = 1; | |
861 | } | |
862 | ||
863 | void *malloc_pkey_hugetlb(long size, int prot, u16 pkey) | |
864 | { | |
865 | void *ptr; | |
866 | int flags = MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB; | |
867 | ||
868 | if (!hugetlb_setup_ok) | |
869 | return PTR_ERR_ENOTSUP; | |
870 | ||
871 | dprintf1("doing %s(%ld, %x, %x)\n", __func__, size, prot, pkey); | |
872 | size = ALIGN_UP(size, HPAGE_SIZE * 2); | |
873 | pkey_assert(pkey < NR_PKEYS); | |
874 | ptr = mmap(NULL, size, PROT_NONE, flags, -1, 0); | |
875 | pkey_assert(ptr != (void *)-1); | |
876 | mprotect_pkey(ptr, size, prot, pkey); | |
877 | ||
878 | record_pkey_malloc(ptr, size); | |
879 | ||
880 | dprintf1("mmap()'d hugetlbfs for pkey %d @ %p\n", pkey, ptr); | |
881 | return ptr; | |
882 | } | |
883 | ||
884 | void *malloc_pkey_mmap_dax(long size, int prot, u16 pkey) | |
885 | { | |
886 | void *ptr; | |
887 | int fd; | |
888 | ||
889 | dprintf1("doing %s(size=%ld, prot=0x%x, pkey=%d)\n", __func__, | |
890 | size, prot, pkey); | |
891 | pkey_assert(pkey < NR_PKEYS); | |
892 | fd = open("/dax/foo", O_RDWR); | |
893 | pkey_assert(fd >= 0); | |
894 | ||
895 | ptr = mmap(0, size, prot, MAP_SHARED, fd, 0); | |
896 | pkey_assert(ptr != (void *)-1); | |
897 | ||
898 | mprotect_pkey(ptr, size, prot, pkey); | |
899 | ||
900 | record_pkey_malloc(ptr, size); | |
901 | ||
902 | dprintf1("mmap()'d for pkey %d @ %p\n", pkey, ptr); | |
903 | close(fd); | |
904 | return ptr; | |
905 | } | |
906 | ||
907 | void *(*pkey_malloc[])(long size, int prot, u16 pkey) = { | |
908 | ||
909 | malloc_pkey_with_mprotect, | |
910 | malloc_pkey_anon_huge, | |
911 | malloc_pkey_hugetlb | |
912 | /* can not do direct with the pkey_mprotect() API: | |
913 | malloc_pkey_mmap_direct, | |
914 | malloc_pkey_mmap_dax, | |
915 | */ | |
916 | }; | |
917 | ||
918 | void *malloc_pkey(long size, int prot, u16 pkey) | |
919 | { | |
920 | void *ret; | |
921 | static int malloc_type; | |
922 | int nr_malloc_types = ARRAY_SIZE(pkey_malloc); | |
923 | ||
924 | pkey_assert(pkey < NR_PKEYS); | |
925 | ||
926 | while (1) { | |
927 | pkey_assert(malloc_type < nr_malloc_types); | |
928 | ||
929 | ret = pkey_malloc[malloc_type](size, prot, pkey); | |
930 | pkey_assert(ret != (void *)-1); | |
931 | ||
932 | malloc_type++; | |
933 | if (malloc_type >= nr_malloc_types) | |
934 | malloc_type = (random()%nr_malloc_types); | |
935 | ||
936 | /* try again if the malloc_type we tried is unsupported */ | |
937 | if (ret == PTR_ERR_ENOTSUP) | |
938 | continue; | |
939 | ||
940 | break; | |
941 | } | |
942 | ||
943 | dprintf3("%s(%ld, prot=%x, pkey=%x) returning: %p\n", __func__, | |
944 | size, prot, pkey, ret); | |
945 | return ret; | |
946 | } | |
947 | ||
948 | int last_pkru_faults; | |
949 | void expected_pk_fault(int pkey) | |
950 | { | |
951 | dprintf2("%s(): last_pkru_faults: %d pkru_faults: %d\n", | |
952 | __func__, last_pkru_faults, pkru_faults); | |
953 | dprintf2("%s(%d): last_si_pkey: %d\n", __func__, pkey, last_si_pkey); | |
954 | pkey_assert(last_pkru_faults + 1 == pkru_faults); | |
955 | pkey_assert(last_si_pkey == pkey); | |
956 | /* | |
957 | * The signal handler shold have cleared out PKRU to let the | |
958 | * test program continue. We now have to restore it. | |
959 | */ | |
960 | if (__rdpkru() != 0) | |
961 | pkey_assert(0); | |
962 | ||
963 | __wrpkru(shadow_pkru); | |
964 | dprintf1("%s() set PKRU=%x to restore state after signal nuked it\n", | |
965 | __func__, shadow_pkru); | |
966 | last_pkru_faults = pkru_faults; | |
967 | last_si_pkey = -1; | |
968 | } | |
969 | ||
970 | void do_not_expect_pk_fault(void) | |
971 | { | |
972 | pkey_assert(last_pkru_faults == pkru_faults); | |
973 | } | |
974 | ||
975 | int test_fds[10] = { -1 }; | |
976 | int nr_test_fds; | |
977 | void __save_test_fd(int fd) | |
978 | { | |
979 | pkey_assert(fd >= 0); | |
980 | pkey_assert(nr_test_fds < ARRAY_SIZE(test_fds)); | |
981 | test_fds[nr_test_fds] = fd; | |
982 | nr_test_fds++; | |
983 | } | |
984 | ||
985 | int get_test_read_fd(void) | |
986 | { | |
987 | int test_fd = open("/etc/passwd", O_RDONLY); | |
988 | __save_test_fd(test_fd); | |
989 | return test_fd; | |
990 | } | |
991 | ||
992 | void close_test_fds(void) | |
993 | { | |
994 | int i; | |
995 | ||
996 | for (i = 0; i < nr_test_fds; i++) { | |
997 | if (test_fds[i] < 0) | |
998 | continue; | |
999 | close(test_fds[i]); | |
1000 | test_fds[i] = -1; | |
1001 | } | |
1002 | nr_test_fds = 0; | |
1003 | } | |
1004 | ||
1005 | #define barrier() __asm__ __volatile__("": : :"memory") | |
1006 | __attribute__((noinline)) int read_ptr(int *ptr) | |
1007 | { | |
1008 | /* | |
1009 | * Keep GCC from optimizing this away somehow | |
1010 | */ | |
1011 | barrier(); | |
1012 | return *ptr; | |
1013 | } | |
1014 | ||
1015 | void test_read_of_write_disabled_region(int *ptr, u16 pkey) | |
1016 | { | |
1017 | int ptr_contents; | |
1018 | ||
1019 | dprintf1("disabling write access to PKEY[1], doing read\n"); | |
1020 | pkey_write_deny(pkey); | |
1021 | ptr_contents = read_ptr(ptr); | |
1022 | dprintf1("*ptr: %d\n", ptr_contents); | |
1023 | dprintf1("\n"); | |
1024 | } | |
1025 | void test_read_of_access_disabled_region(int *ptr, u16 pkey) | |
1026 | { | |
1027 | int ptr_contents; | |
1028 | ||
1029 | dprintf1("disabling access to PKEY[%02d], doing read @ %p\n", pkey, ptr); | |
1030 | rdpkru(); | |
1031 | pkey_access_deny(pkey); | |
1032 | ptr_contents = read_ptr(ptr); | |
1033 | dprintf1("*ptr: %d\n", ptr_contents); | |
1034 | expected_pk_fault(pkey); | |
1035 | } | |
1036 | void test_write_of_write_disabled_region(int *ptr, u16 pkey) | |
1037 | { | |
1038 | dprintf1("disabling write access to PKEY[%02d], doing write\n", pkey); | |
1039 | pkey_write_deny(pkey); | |
1040 | *ptr = __LINE__; | |
1041 | expected_pk_fault(pkey); | |
1042 | } | |
1043 | void test_write_of_access_disabled_region(int *ptr, u16 pkey) | |
1044 | { | |
1045 | dprintf1("disabling access to PKEY[%02d], doing write\n", pkey); | |
1046 | pkey_access_deny(pkey); | |
1047 | *ptr = __LINE__; | |
1048 | expected_pk_fault(pkey); | |
1049 | } | |
1050 | void test_kernel_write_of_access_disabled_region(int *ptr, u16 pkey) | |
1051 | { | |
1052 | int ret; | |
1053 | int test_fd = get_test_read_fd(); | |
1054 | ||
1055 | dprintf1("disabling access to PKEY[%02d], " | |
1056 | "having kernel read() to buffer\n", pkey); | |
1057 | pkey_access_deny(pkey); | |
1058 | ret = read(test_fd, ptr, 1); | |
1059 | dprintf1("read ret: %d\n", ret); | |
1060 | pkey_assert(ret); | |
1061 | } | |
1062 | void test_kernel_write_of_write_disabled_region(int *ptr, u16 pkey) | |
1063 | { | |
1064 | int ret; | |
1065 | int test_fd = get_test_read_fd(); | |
1066 | ||
1067 | pkey_write_deny(pkey); | |
1068 | ret = read(test_fd, ptr, 100); | |
1069 | dprintf1("read ret: %d\n", ret); | |
1070 | if (ret < 0 && (DEBUG_LEVEL > 0)) | |
1071 | perror("verbose read result (OK for this to be bad)"); | |
1072 | pkey_assert(ret); | |
1073 | } | |
1074 | ||
1075 | void test_kernel_gup_of_access_disabled_region(int *ptr, u16 pkey) | |
1076 | { | |
1077 | int pipe_ret, vmsplice_ret; | |
1078 | struct iovec iov; | |
1079 | int pipe_fds[2]; | |
1080 | ||
1081 | pipe_ret = pipe(pipe_fds); | |
1082 | ||
1083 | pkey_assert(pipe_ret == 0); | |
1084 | dprintf1("disabling access to PKEY[%02d], " | |
1085 | "having kernel vmsplice from buffer\n", pkey); | |
1086 | pkey_access_deny(pkey); | |
1087 | iov.iov_base = ptr; | |
1088 | iov.iov_len = PAGE_SIZE; | |
1089 | vmsplice_ret = vmsplice(pipe_fds[1], &iov, 1, SPLICE_F_GIFT); | |
1090 | dprintf1("vmsplice() ret: %d\n", vmsplice_ret); | |
1091 | pkey_assert(vmsplice_ret == -1); | |
1092 | ||
1093 | close(pipe_fds[0]); | |
1094 | close(pipe_fds[1]); | |
1095 | } | |
1096 | ||
1097 | void test_kernel_gup_write_to_write_disabled_region(int *ptr, u16 pkey) | |
1098 | { | |
1099 | int ignored = 0xdada; | |
1100 | int futex_ret; | |
1101 | int some_int = __LINE__; | |
1102 | ||
1103 | dprintf1("disabling write to PKEY[%02d], " | |
1104 | "doing futex gunk in buffer\n", pkey); | |
1105 | *ptr = some_int; | |
1106 | pkey_write_deny(pkey); | |
1107 | futex_ret = syscall(SYS_futex, ptr, FUTEX_WAIT, some_int-1, NULL, | |
1108 | &ignored, ignored); | |
1109 | if (DEBUG_LEVEL > 0) | |
1110 | perror("futex"); | |
1111 | dprintf1("futex() ret: %d\n", futex_ret); | |
1112 | } | |
1113 | ||
1114 | /* Assumes that all pkeys other than 'pkey' are unallocated */ | |
1115 | void test_pkey_syscalls_on_non_allocated_pkey(int *ptr, u16 pkey) | |
1116 | { | |
1117 | int err; | |
1118 | int i; | |
1119 | ||
1120 | /* Note: 0 is the default pkey, so don't mess with it */ | |
1121 | for (i = 1; i < NR_PKEYS; i++) { | |
1122 | if (pkey == i) | |
1123 | continue; | |
1124 | ||
1125 | dprintf1("trying get/set/free to non-allocated pkey: %2d\n", i); | |
1126 | err = sys_pkey_free(i); | |
1127 | pkey_assert(err); | |
1128 | ||
5f23f6d0 DH |
1129 | err = sys_pkey_free(i); |
1130 | pkey_assert(err); | |
1131 | ||
1132 | err = sys_mprotect_pkey(ptr, PAGE_SIZE, PROT_READ, i); | |
1133 | pkey_assert(err); | |
1134 | } | |
1135 | } | |
1136 | ||
1137 | /* Assumes that all pkeys other than 'pkey' are unallocated */ | |
1138 | void test_pkey_syscalls_bad_args(int *ptr, u16 pkey) | |
1139 | { | |
1140 | int err; | |
5f23f6d0 DH |
1141 | int bad_pkey = NR_PKEYS+99; |
1142 | ||
5f23f6d0 DH |
1143 | /* pass a known-invalid pkey in: */ |
1144 | err = sys_mprotect_pkey(ptr, PAGE_SIZE, PROT_READ, bad_pkey); | |
1145 | pkey_assert(err); | |
1146 | } | |
1147 | ||
1148 | /* Assumes that all pkeys other than 'pkey' are unallocated */ | |
1149 | void test_pkey_alloc_exhaust(int *ptr, u16 pkey) | |
1150 | { | |
5f23f6d0 DH |
1151 | int err; |
1152 | int allocated_pkeys[NR_PKEYS] = {0}; | |
1153 | int nr_allocated_pkeys = 0; | |
1154 | int i; | |
1155 | ||
1156 | for (i = 0; i < NR_PKEYS*2; i++) { | |
1157 | int new_pkey; | |
1158 | dprintf1("%s() alloc loop: %d\n", __func__, i); | |
1159 | new_pkey = alloc_pkey(); | |
1160 | dprintf4("%s()::%d, err: %d pkru: 0x%x shadow: 0x%x\n", __func__, | |
1161 | __LINE__, err, __rdpkru(), shadow_pkru); | |
1162 | rdpkru(); /* for shadow checking */ | |
1163 | dprintf2("%s() errno: %d ENOSPC: %d\n", __func__, errno, ENOSPC); | |
1164 | if ((new_pkey == -1) && (errno == ENOSPC)) { | |
1165 | dprintf2("%s() failed to allocate pkey after %d tries\n", | |
1166 | __func__, nr_allocated_pkeys); | |
1167 | break; | |
1168 | } | |
1169 | pkey_assert(nr_allocated_pkeys < NR_PKEYS); | |
1170 | allocated_pkeys[nr_allocated_pkeys++] = new_pkey; | |
1171 | } | |
1172 | ||
1173 | dprintf3("%s()::%d\n", __func__, __LINE__); | |
1174 | ||
1175 | /* | |
1176 | * ensure it did not reach the end of the loop without | |
1177 | * failure: | |
1178 | */ | |
1179 | pkey_assert(i < NR_PKEYS*2); | |
1180 | ||
1181 | /* | |
1182 | * There are 16 pkeys supported in hardware. One is taken | |
1183 | * up for the default (0) and another can be taken up by | |
1184 | * an execute-only mapping. Ensure that we can allocate | |
1185 | * at least 14 (16-2). | |
1186 | */ | |
1187 | pkey_assert(i >= NR_PKEYS-2); | |
1188 | ||
1189 | for (i = 0; i < nr_allocated_pkeys; i++) { | |
1190 | err = sys_pkey_free(allocated_pkeys[i]); | |
1191 | pkey_assert(!err); | |
1192 | rdpkru(); /* for shadow checking */ | |
1193 | } | |
1194 | } | |
1195 | ||
1196 | void test_ptrace_of_child(int *ptr, u16 pkey) | |
1197 | { | |
1198 | __attribute__((__unused__)) int peek_result; | |
1199 | pid_t child_pid; | |
1200 | void *ignored = 0; | |
1201 | long ret; | |
1202 | int status; | |
1203 | /* | |
1204 | * This is the "control" for our little expermient. Make sure | |
1205 | * we can always access it when ptracing. | |
1206 | */ | |
1207 | int *plain_ptr_unaligned = malloc(HPAGE_SIZE); | |
1208 | int *plain_ptr = ALIGN_PTR_UP(plain_ptr_unaligned, PAGE_SIZE); | |
1209 | ||
1210 | /* | |
1211 | * Fork a child which is an exact copy of this process, of course. | |
1212 | * That means we can do all of our tests via ptrace() and then plain | |
1213 | * memory access and ensure they work differently. | |
1214 | */ | |
1215 | child_pid = fork_lazy_child(); | |
1216 | dprintf1("[%d] child pid: %d\n", getpid(), child_pid); | |
1217 | ||
1218 | ret = ptrace(PTRACE_ATTACH, child_pid, ignored, ignored); | |
1219 | if (ret) | |
1220 | perror("attach"); | |
1221 | dprintf1("[%d] attach ret: %ld %d\n", getpid(), ret, __LINE__); | |
1222 | pkey_assert(ret != -1); | |
1223 | ret = waitpid(child_pid, &status, WUNTRACED); | |
1224 | if ((ret != child_pid) || !(WIFSTOPPED(status))) { | |
1225 | fprintf(stderr, "weird waitpid result %ld stat %x\n", | |
1226 | ret, status); | |
1227 | pkey_assert(0); | |
1228 | } | |
1229 | dprintf2("waitpid ret: %ld\n", ret); | |
1230 | dprintf2("waitpid status: %d\n", status); | |
1231 | ||
1232 | pkey_access_deny(pkey); | |
1233 | pkey_write_deny(pkey); | |
1234 | ||
1235 | /* Write access, untested for now: | |
1236 | ret = ptrace(PTRACE_POKEDATA, child_pid, peek_at, data); | |
1237 | pkey_assert(ret != -1); | |
1238 | dprintf1("poke at %p: %ld\n", peek_at, ret); | |
1239 | */ | |
1240 | ||
1241 | /* | |
1242 | * Try to access the pkey-protected "ptr" via ptrace: | |
1243 | */ | |
1244 | ret = ptrace(PTRACE_PEEKDATA, child_pid, ptr, ignored); | |
1245 | /* expect it to work, without an error: */ | |
1246 | pkey_assert(ret != -1); | |
1247 | /* Now access from the current task, and expect an exception: */ | |
1248 | peek_result = read_ptr(ptr); | |
1249 | expected_pk_fault(pkey); | |
1250 | ||
1251 | /* | |
1252 | * Try to access the NON-pkey-protected "plain_ptr" via ptrace: | |
1253 | */ | |
1254 | ret = ptrace(PTRACE_PEEKDATA, child_pid, plain_ptr, ignored); | |
1255 | /* expect it to work, without an error: */ | |
1256 | pkey_assert(ret != -1); | |
1257 | /* Now access from the current task, and expect NO exception: */ | |
1258 | peek_result = read_ptr(plain_ptr); | |
1259 | do_not_expect_pk_fault(); | |
1260 | ||
1261 | ret = ptrace(PTRACE_DETACH, child_pid, ignored, 0); | |
1262 | pkey_assert(ret != -1); | |
1263 | ||
1264 | ret = kill(child_pid, SIGKILL); | |
1265 | pkey_assert(ret != -1); | |
1266 | ||
1267 | wait(&status); | |
1268 | ||
1269 | free(plain_ptr_unaligned); | |
1270 | } | |
1271 | ||
1272 | void test_executing_on_unreadable_memory(int *ptr, u16 pkey) | |
1273 | { | |
1274 | void *p1; | |
1275 | int scratch; | |
1276 | int ptr_contents; | |
1277 | int ret; | |
1278 | ||
1279 | p1 = ALIGN_PTR_UP(&lots_o_noops_around_write, PAGE_SIZE); | |
1280 | dprintf3("&lots_o_noops: %p\n", &lots_o_noops_around_write); | |
1281 | /* lots_o_noops_around_write should be page-aligned already */ | |
1282 | assert(p1 == &lots_o_noops_around_write); | |
1283 | ||
1284 | /* Point 'p1' at the *second* page of the function: */ | |
1285 | p1 += PAGE_SIZE; | |
1286 | ||
1287 | madvise(p1, PAGE_SIZE, MADV_DONTNEED); | |
1288 | lots_o_noops_around_write(&scratch); | |
1289 | ptr_contents = read_ptr(p1); | |
1290 | dprintf2("ptr (%p) contents@%d: %x\n", p1, __LINE__, ptr_contents); | |
1291 | ||
1292 | ret = mprotect_pkey(p1, PAGE_SIZE, PROT_EXEC, (u64)pkey); | |
1293 | pkey_assert(!ret); | |
1294 | pkey_access_deny(pkey); | |
1295 | ||
1296 | dprintf2("pkru: %x\n", rdpkru()); | |
1297 | ||
1298 | /* | |
1299 | * Make sure this is an *instruction* fault | |
1300 | */ | |
1301 | madvise(p1, PAGE_SIZE, MADV_DONTNEED); | |
1302 | lots_o_noops_around_write(&scratch); | |
1303 | do_not_expect_pk_fault(); | |
1304 | ptr_contents = read_ptr(p1); | |
1305 | dprintf2("ptr (%p) contents@%d: %x\n", p1, __LINE__, ptr_contents); | |
1306 | expected_pk_fault(pkey); | |
1307 | } | |
1308 | ||
1309 | void test_mprotect_pkey_on_unsupported_cpu(int *ptr, u16 pkey) | |
1310 | { | |
1311 | int size = PAGE_SIZE; | |
1312 | int sret; | |
1313 | ||
1314 | if (cpu_has_pku()) { | |
1315 | dprintf1("SKIP: %s: no CPU support\n", __func__); | |
1316 | return; | |
1317 | } | |
1318 | ||
1319 | sret = syscall(SYS_mprotect_key, ptr, size, PROT_READ, pkey); | |
1320 | pkey_assert(sret < 0); | |
1321 | } | |
1322 | ||
1323 | void (*pkey_tests[])(int *ptr, u16 pkey) = { | |
1324 | test_read_of_write_disabled_region, | |
1325 | test_read_of_access_disabled_region, | |
1326 | test_write_of_write_disabled_region, | |
1327 | test_write_of_access_disabled_region, | |
1328 | test_kernel_write_of_access_disabled_region, | |
1329 | test_kernel_write_of_write_disabled_region, | |
1330 | test_kernel_gup_of_access_disabled_region, | |
1331 | test_kernel_gup_write_to_write_disabled_region, | |
1332 | test_executing_on_unreadable_memory, | |
1333 | test_ptrace_of_child, | |
1334 | test_pkey_syscalls_on_non_allocated_pkey, | |
1335 | test_pkey_syscalls_bad_args, | |
1336 | test_pkey_alloc_exhaust, | |
1337 | }; | |
1338 | ||
1339 | void run_tests_once(void) | |
1340 | { | |
1341 | int *ptr; | |
1342 | int prot = PROT_READ|PROT_WRITE; | |
1343 | ||
1344 | for (test_nr = 0; test_nr < ARRAY_SIZE(pkey_tests); test_nr++) { | |
1345 | int pkey; | |
1346 | int orig_pkru_faults = pkru_faults; | |
1347 | ||
1348 | dprintf1("======================\n"); | |
1349 | dprintf1("test %d preparing...\n", test_nr); | |
1350 | ||
1351 | tracing_on(); | |
1352 | pkey = alloc_random_pkey(); | |
1353 | dprintf1("test %d starting with pkey: %d\n", test_nr, pkey); | |
1354 | ptr = malloc_pkey(PAGE_SIZE, prot, pkey); | |
1355 | dprintf1("test %d starting...\n", test_nr); | |
1356 | pkey_tests[test_nr](ptr, pkey); | |
1357 | dprintf1("freeing test memory: %p\n", ptr); | |
1358 | free_pkey_malloc(ptr); | |
1359 | sys_pkey_free(pkey); | |
1360 | ||
1361 | dprintf1("pkru_faults: %d\n", pkru_faults); | |
1362 | dprintf1("orig_pkru_faults: %d\n", orig_pkru_faults); | |
1363 | ||
1364 | tracing_off(); | |
1365 | close_test_fds(); | |
1366 | ||
7738789f | 1367 | printf("test %2d PASSED (iteration %d)\n", test_nr, iteration_nr); |
5f23f6d0 DH |
1368 | dprintf1("======================\n\n"); |
1369 | } | |
1370 | iteration_nr++; | |
1371 | } | |
1372 | ||
1373 | void pkey_setup_shadow(void) | |
1374 | { | |
1375 | shadow_pkru = __rdpkru(); | |
1376 | } | |
1377 | ||
1378 | int main(void) | |
1379 | { | |
1380 | int nr_iterations = 22; | |
1381 | ||
1382 | setup_handlers(); | |
1383 | ||
1384 | printf("has pku: %d\n", cpu_has_pku()); | |
1385 | ||
1386 | if (!cpu_has_pku()) { | |
1387 | int size = PAGE_SIZE; | |
1388 | int *ptr; | |
1389 | ||
1390 | printf("running PKEY tests for unsupported CPU/OS\n"); | |
1391 | ||
1392 | ptr = mmap(NULL, size, PROT_NONE, MAP_ANONYMOUS|MAP_PRIVATE, -1, 0); | |
1393 | assert(ptr != (void *)-1); | |
1394 | test_mprotect_pkey_on_unsupported_cpu(ptr, 1); | |
1395 | exit(0); | |
1396 | } | |
1397 | ||
1398 | pkey_setup_shadow(); | |
1399 | printf("startup pkru: %x\n", rdpkru()); | |
1400 | setup_hugetlbfs(); | |
1401 | ||
1402 | while (nr_iterations-- > 0) | |
1403 | run_tests_once(); | |
1404 | ||
1405 | printf("done (all tests OK)\n"); | |
1406 | return 0; | |
1407 | } |