1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright Intel Corporation, 2023
4 *
5 * Author: Chao Peng <chao.p.peng@linux.intel.com>
6 */
7 #include <stdlib.h>
8 #include <string.h>
9 #include <unistd.h>
10 #include <errno.h>
11 #include <stdio.h>
12 #include <fcntl.h>
13
14 #include <linux/bitmap.h>
15 #include <linux/falloc.h>
16 #include <linux/sizes.h>
17 #include <sys/mman.h>
18 #include <sys/types.h>
19 #include <sys/stat.h>
20
21 #include "kvm_util.h"
22 #include "test_util.h"
23 #include "ucall_common.h"
24
25 static size_t page_size;
26
test_file_read_write(int fd,size_t total_size)27 static void test_file_read_write(int fd, size_t total_size)
28 {
29 char buf[64];
30
31 TEST_ASSERT(read(fd, buf, sizeof(buf)) < 0,
32 "read on a guest_mem fd should fail");
33 TEST_ASSERT(write(fd, buf, sizeof(buf)) < 0,
34 "write on a guest_mem fd should fail");
35 TEST_ASSERT(pread(fd, buf, sizeof(buf), 0) < 0,
36 "pread on a guest_mem fd should fail");
37 TEST_ASSERT(pwrite(fd, buf, sizeof(buf), 0) < 0,
38 "pwrite on a guest_mem fd should fail");
39 }
40
test_mmap_cow(int fd,size_t size)41 static void test_mmap_cow(int fd, size_t size)
42 {
43 void *mem;
44
45 mem = mmap(NULL, size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
46 TEST_ASSERT(mem == MAP_FAILED, "Copy-on-write not allowed by guest_memfd.");
47 }
48
test_mmap_supported(int fd,size_t total_size)49 static void test_mmap_supported(int fd, size_t total_size)
50 {
51 const char val = 0xaa;
52 char *mem;
53 size_t i;
54 int ret;
55
56 mem = kvm_mmap(total_size, PROT_READ | PROT_WRITE, MAP_SHARED, fd);
57
58 memset(mem, val, total_size);
59 for (i = 0; i < total_size; i++)
60 TEST_ASSERT_EQ(READ_ONCE(mem[i]), val);
61
62 ret = fallocate(fd, FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE, 0,
63 page_size);
64 TEST_ASSERT(!ret, "fallocate the first page should succeed.");
65
66 for (i = 0; i < page_size; i++)
67 TEST_ASSERT_EQ(READ_ONCE(mem[i]), 0x00);
68 for (; i < total_size; i++)
69 TEST_ASSERT_EQ(READ_ONCE(mem[i]), val);
70
71 memset(mem, val, page_size);
72 for (i = 0; i < total_size; i++)
73 TEST_ASSERT_EQ(READ_ONCE(mem[i]), val);
74
75 kvm_munmap(mem, total_size);
76 }
77
test_fault_sigbus(int fd,size_t accessible_size,size_t map_size)78 static void test_fault_sigbus(int fd, size_t accessible_size, size_t map_size)
79 {
80 const char val = 0xaa;
81 char *mem;
82 size_t i;
83
84 mem = kvm_mmap(map_size, PROT_READ | PROT_WRITE, MAP_SHARED, fd);
85
86 TEST_EXPECT_SIGBUS(memset(mem, val, map_size));
87 TEST_EXPECT_SIGBUS((void)READ_ONCE(mem[accessible_size]));
88
89 for (i = 0; i < accessible_size; i++)
90 TEST_ASSERT_EQ(READ_ONCE(mem[i]), val);
91
92 kvm_munmap(mem, map_size);
93 }
94
test_fault_overflow(int fd,size_t total_size)95 static void test_fault_overflow(int fd, size_t total_size)
96 {
97 test_fault_sigbus(fd, total_size, total_size * 4);
98 }
99
test_fault_private(int fd,size_t total_size)100 static void test_fault_private(int fd, size_t total_size)
101 {
102 test_fault_sigbus(fd, 0, total_size);
103 }
104
test_mmap_not_supported(int fd,size_t total_size)105 static void test_mmap_not_supported(int fd, size_t total_size)
106 {
107 char *mem;
108
109 mem = mmap(NULL, page_size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
110 TEST_ASSERT_EQ(mem, MAP_FAILED);
111
112 mem = mmap(NULL, total_size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
113 TEST_ASSERT_EQ(mem, MAP_FAILED);
114 }
115
test_file_size(int fd,size_t total_size)116 static void test_file_size(int fd, size_t total_size)
117 {
118 struct stat sb;
119 int ret;
120
121 ret = fstat(fd, &sb);
122 TEST_ASSERT(!ret, "fstat should succeed");
123 TEST_ASSERT_EQ(sb.st_size, total_size);
124 TEST_ASSERT_EQ(sb.st_blksize, page_size);
125 }
126
test_fallocate(int fd,size_t total_size)127 static void test_fallocate(int fd, size_t total_size)
128 {
129 int ret;
130
131 ret = fallocate(fd, FALLOC_FL_KEEP_SIZE, 0, total_size);
132 TEST_ASSERT(!ret, "fallocate with aligned offset and size should succeed");
133
134 ret = fallocate(fd, FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE,
135 page_size - 1, page_size);
136 TEST_ASSERT(ret, "fallocate with unaligned offset should fail");
137
138 ret = fallocate(fd, FALLOC_FL_KEEP_SIZE, total_size, page_size);
139 TEST_ASSERT(ret, "fallocate beginning at total_size should fail");
140
141 ret = fallocate(fd, FALLOC_FL_KEEP_SIZE, total_size + page_size, page_size);
142 TEST_ASSERT(ret, "fallocate beginning after total_size should fail");
143
144 ret = fallocate(fd, FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE,
145 total_size, page_size);
146 TEST_ASSERT(!ret, "fallocate(PUNCH_HOLE) at total_size should succeed");
147
148 ret = fallocate(fd, FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE,
149 total_size + page_size, page_size);
150 TEST_ASSERT(!ret, "fallocate(PUNCH_HOLE) after total_size should succeed");
151
152 ret = fallocate(fd, FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE,
153 page_size, page_size - 1);
154 TEST_ASSERT(ret, "fallocate with unaligned size should fail");
155
156 ret = fallocate(fd, FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE,
157 page_size, page_size);
158 TEST_ASSERT(!ret, "fallocate(PUNCH_HOLE) with aligned offset and size should succeed");
159
160 ret = fallocate(fd, FALLOC_FL_KEEP_SIZE, page_size, page_size);
161 TEST_ASSERT(!ret, "fallocate to restore punched hole should succeed");
162 }
163
test_invalid_punch_hole(int fd,size_t total_size)164 static void test_invalid_punch_hole(int fd, size_t total_size)
165 {
166 struct {
167 off_t offset;
168 off_t len;
169 } testcases[] = {
170 {0, 1},
171 {0, page_size - 1},
172 {0, page_size + 1},
173
174 {1, 1},
175 {1, page_size - 1},
176 {1, page_size},
177 {1, page_size + 1},
178
179 {page_size, 1},
180 {page_size, page_size - 1},
181 {page_size, page_size + 1},
182 };
183 int ret, i;
184
185 for (i = 0; i < ARRAY_SIZE(testcases); i++) {
186 ret = fallocate(fd, FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE,
187 testcases[i].offset, testcases[i].len);
188 TEST_ASSERT(ret == -1 && errno == EINVAL,
189 "PUNCH_HOLE with !PAGE_SIZE offset (%lx) and/or length (%lx) should fail",
190 testcases[i].offset, testcases[i].len);
191 }
192 }
193
test_create_guest_memfd_invalid_sizes(struct kvm_vm * vm,uint64_t guest_memfd_flags)194 static void test_create_guest_memfd_invalid_sizes(struct kvm_vm *vm,
195 uint64_t guest_memfd_flags)
196 {
197 size_t size;
198 int fd;
199
200 for (size = 1; size < page_size; size++) {
201 fd = __vm_create_guest_memfd(vm, size, guest_memfd_flags);
202 TEST_ASSERT(fd < 0 && errno == EINVAL,
203 "guest_memfd() with non-page-aligned page size '0x%lx' should fail with EINVAL",
204 size);
205 }
206 }
207
test_create_guest_memfd_multiple(struct kvm_vm * vm)208 static void test_create_guest_memfd_multiple(struct kvm_vm *vm)
209 {
210 int fd1, fd2, ret;
211 struct stat st1, st2;
212
213 fd1 = __vm_create_guest_memfd(vm, page_size, 0);
214 TEST_ASSERT(fd1 != -1, "memfd creation should succeed");
215
216 ret = fstat(fd1, &st1);
217 TEST_ASSERT(ret != -1, "memfd fstat should succeed");
218 TEST_ASSERT(st1.st_size == page_size, "memfd st_size should match requested size");
219
220 fd2 = __vm_create_guest_memfd(vm, page_size * 2, 0);
221 TEST_ASSERT(fd2 != -1, "memfd creation should succeed");
222
223 ret = fstat(fd2, &st2);
224 TEST_ASSERT(ret != -1, "memfd fstat should succeed");
225 TEST_ASSERT(st2.st_size == page_size * 2, "second memfd st_size should match requested size");
226
227 ret = fstat(fd1, &st1);
228 TEST_ASSERT(ret != -1, "memfd fstat should succeed");
229 TEST_ASSERT(st1.st_size == page_size, "first memfd st_size should still match requested size");
230 TEST_ASSERT(st1.st_ino != st2.st_ino, "different memfd should have different inode numbers");
231
232 close(fd2);
233 close(fd1);
234 }
235
test_guest_memfd_flags(struct kvm_vm * vm)236 static void test_guest_memfd_flags(struct kvm_vm *vm)
237 {
238 uint64_t valid_flags = vm_check_cap(vm, KVM_CAP_GUEST_MEMFD_FLAGS);
239 uint64_t flag;
240 int fd;
241
242 for (flag = BIT(0); flag; flag <<= 1) {
243 fd = __vm_create_guest_memfd(vm, page_size, flag);
244 if (flag & valid_flags) {
245 TEST_ASSERT(fd >= 0,
246 "guest_memfd() with flag '0x%lx' should succeed",
247 flag);
248 close(fd);
249 } else {
250 TEST_ASSERT(fd < 0 && errno == EINVAL,
251 "guest_memfd() with flag '0x%lx' should fail with EINVAL",
252 flag);
253 }
254 }
255 }
256
257 #define gmem_test(__test, __vm, __flags) \
258 do { \
259 int fd = vm_create_guest_memfd(__vm, page_size * 4, __flags); \
260 \
261 test_##__test(fd, page_size * 4); \
262 close(fd); \
263 } while (0)
264
__test_guest_memfd(struct kvm_vm * vm,uint64_t flags)265 static void __test_guest_memfd(struct kvm_vm *vm, uint64_t flags)
266 {
267 test_create_guest_memfd_multiple(vm);
268 test_create_guest_memfd_invalid_sizes(vm, flags);
269
270 gmem_test(file_read_write, vm, flags);
271
272 if (flags & GUEST_MEMFD_FLAG_MMAP) {
273 if (flags & GUEST_MEMFD_FLAG_INIT_SHARED) {
274 gmem_test(mmap_supported, vm, flags);
275 gmem_test(fault_overflow, vm, flags);
276 } else {
277 gmem_test(fault_private, vm, flags);
278 }
279
280 gmem_test(mmap_cow, vm, flags);
281 } else {
282 gmem_test(mmap_not_supported, vm, flags);
283 }
284
285 gmem_test(file_size, vm, flags);
286 gmem_test(fallocate, vm, flags);
287 gmem_test(invalid_punch_hole, vm, flags);
288 }
289
test_guest_memfd(unsigned long vm_type)290 static void test_guest_memfd(unsigned long vm_type)
291 {
292 struct kvm_vm *vm = vm_create_barebones_type(vm_type);
293 uint64_t flags;
294
295 test_guest_memfd_flags(vm);
296
297 __test_guest_memfd(vm, 0);
298
299 flags = vm_check_cap(vm, KVM_CAP_GUEST_MEMFD_FLAGS);
300 if (flags & GUEST_MEMFD_FLAG_MMAP)
301 __test_guest_memfd(vm, GUEST_MEMFD_FLAG_MMAP);
302
303 /* MMAP should always be supported if INIT_SHARED is supported. */
304 if (flags & GUEST_MEMFD_FLAG_INIT_SHARED)
305 __test_guest_memfd(vm, GUEST_MEMFD_FLAG_MMAP |
306 GUEST_MEMFD_FLAG_INIT_SHARED);
307
308 kvm_vm_free(vm);
309 }
310
guest_code(uint8_t * mem,uint64_t size)311 static void guest_code(uint8_t *mem, uint64_t size)
312 {
313 size_t i;
314
315 for (i = 0; i < size; i++)
316 __GUEST_ASSERT(mem[i] == 0xaa,
317 "Guest expected 0xaa at offset %lu, got 0x%x", i, mem[i]);
318
319 memset(mem, 0xff, size);
320 GUEST_DONE();
321 }
322
test_guest_memfd_guest(void)323 static void test_guest_memfd_guest(void)
324 {
325 /*
326 * Skip the first 4gb and slot0. slot0 maps <1gb and is used to back
327 * the guest's code, stack, and page tables, and low memory contains
328 * the PCI hole and other MMIO regions that need to be avoided.
329 */
330 const uint64_t gpa = SZ_4G;
331 const int slot = 1;
332
333 struct kvm_vcpu *vcpu;
334 struct kvm_vm *vm;
335 uint8_t *mem;
336 size_t size;
337 int fd, i;
338
339 if (!kvm_check_cap(KVM_CAP_GUEST_MEMFD_FLAGS))
340 return;
341
342 vm = __vm_create_shape_with_one_vcpu(VM_SHAPE_DEFAULT, &vcpu, 1, guest_code);
343
344 TEST_ASSERT(vm_check_cap(vm, KVM_CAP_GUEST_MEMFD_FLAGS) & GUEST_MEMFD_FLAG_MMAP,
345 "Default VM type should support MMAP, supported flags = 0x%x",
346 vm_check_cap(vm, KVM_CAP_GUEST_MEMFD_FLAGS));
347 TEST_ASSERT(vm_check_cap(vm, KVM_CAP_GUEST_MEMFD_FLAGS) & GUEST_MEMFD_FLAG_INIT_SHARED,
348 "Default VM type should support INIT_SHARED, supported flags = 0x%x",
349 vm_check_cap(vm, KVM_CAP_GUEST_MEMFD_FLAGS));
350
351 size = vm->page_size;
352 fd = vm_create_guest_memfd(vm, size, GUEST_MEMFD_FLAG_MMAP |
353 GUEST_MEMFD_FLAG_INIT_SHARED);
354 vm_set_user_memory_region2(vm, slot, KVM_MEM_GUEST_MEMFD, gpa, size, NULL, fd, 0);
355
356 mem = kvm_mmap(size, PROT_READ | PROT_WRITE, MAP_SHARED, fd);
357 memset(mem, 0xaa, size);
358 kvm_munmap(mem, size);
359
360 virt_pg_map(vm, gpa, gpa);
361 vcpu_args_set(vcpu, 2, gpa, size);
362 vcpu_run(vcpu);
363
364 TEST_ASSERT_EQ(get_ucall(vcpu, NULL), UCALL_DONE);
365
366 mem = kvm_mmap(size, PROT_READ | PROT_WRITE, MAP_SHARED, fd);
367 for (i = 0; i < size; i++)
368 TEST_ASSERT_EQ(mem[i], 0xff);
369
370 close(fd);
371 kvm_vm_free(vm);
372 }
373
main(int argc,char * argv[])374 int main(int argc, char *argv[])
375 {
376 unsigned long vm_types, vm_type;
377
378 TEST_REQUIRE(kvm_has_cap(KVM_CAP_GUEST_MEMFD));
379
380 page_size = getpagesize();
381
382 /*
383 * Not all architectures support KVM_CAP_VM_TYPES. However, those that
384 * support guest_memfd have that support for the default VM type.
385 */
386 vm_types = kvm_check_cap(KVM_CAP_VM_TYPES);
387 if (!vm_types)
388 vm_types = BIT(VM_TYPE_DEFAULT);
389
390 for_each_set_bit(vm_type, &vm_types, BITS_PER_TYPE(vm_types))
391 test_guest_memfd(vm_type);
392
393 test_guest_memfd_guest();
394 }
395