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
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
|
/*
* Sample server to be tested with samples/client.c
*
* Copyright (c) 2020, Nutanix Inc. All rights reserved.
* Author: Thanos Makatos <thanos@nutanix.com>
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* * Neither the name of Nutanix nor the names of its contributors may be
* used to endorse or promote products derived from this software without
* specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL <COPYRIGHT HOLDER> BE LIABLE FOR ANY
* DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
* SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
* DAMAGE.
*
*/
#define _GNU_SOURCE
#include <stdio.h>
#include <err.h>
#include <stdlib.h>
#include <unistd.h>
#include <string.h>
#include <signal.h>
#include <errno.h>
#include <time.h>
#include <assert.h>
#include <openssl/md5.h>
#include <sys/mman.h>
#include <sys/param.h>
#include <sys/time.h>
#include "common.h"
#include "libvfio-user.h"
#include "tran_sock.h"
struct dma_regions {
uint64_t addr;
uint64_t len;
uint32_t prot;
};
#define NR_DMA_REGIONS 96
struct server_data {
time_t bar0;
void *bar1;
size_t bar1_size;
struct dma_regions regions[NR_DMA_REGIONS];
struct {
__u64 pending_bytes;
vfu_migr_state_t state;
} migration;
};
static void
_log(vfu_ctx_t *vfu_ctx UNUSED, UNUSED int level, char const *msg)
{
fprintf(stderr, "server: %s\n", msg);
}
static int
arm_timer(vfu_ctx_t *vfu_ctx, time_t t)
{
struct itimerval new = {.it_value.tv_sec = t - time(NULL) };
vfu_log(vfu_ctx, LOG_DEBUG, "arming timer to trigger in %ld seconds",
new.it_value.tv_sec);
if (setitimer(ITIMER_REAL, &new, NULL) != 0) {
vfu_log(vfu_ctx, LOG_ERR, "failed to arm timer: %m");
return -errno;
}
return 0;
}
ssize_t
bar0_access(vfu_ctx_t *vfu_ctx, char * const buf, size_t count, loff_t offset,
const bool is_write)
{
struct server_data *server_data = vfu_get_private(vfu_ctx);
if (count != sizeof(time_t) || offset != 0) {
vfu_log(vfu_ctx, LOG_ERR, "bad BAR0 access %#lx-%#lx",
offset, offset + count - 1);
errno = EINVAL;
return -1;
}
if (is_write) {
if (server_data->migration.state == VFU_MIGR_STATE_RUNNING) {
int ret = arm_timer(vfu_ctx, *(time_t*)buf);
if (ret < 0) {
return ret;
}
}
memcpy(&server_data->bar0, buf, count);
} else {
time_t delta = time(NULL) - server_data->bar0;
memcpy(buf, &delta, count);
}
return count;
}
ssize_t
bar1_access(vfu_ctx_t *vfu_ctx, char * const buf,
size_t count, loff_t offset,
const bool is_write)
{
struct server_data *server_data = vfu_get_private(vfu_ctx);
if (offset + count > server_data->bar1_size) {
vfu_log(vfu_ctx, LOG_ERR, "bad BAR1 access %#lx-%#lx",
offset, offset + count - 1);
errno = EINVAL;
return -1;
}
if (is_write) {
if (server_data->migration.state == VFU_MIGR_STATE_PRE_COPY) {
/* dirty the whole thing */
server_data->migration.pending_bytes = server_data->bar1_size;
}
memcpy(server_data->bar1 + offset, buf, count);
} else {
memcpy(buf, server_data->bar1, count);
}
return count;
}
bool irq_triggered = false;
static void _sa_handler(int signum)
{
int _errno = errno;
if (signum == SIGALRM) {
irq_triggered = true;
}
errno = _errno;
}
static void
map_dma(vfu_ctx_t *vfu_ctx, uint64_t iova, uint64_t len, uint32_t prot)
{
struct server_data *server_data = vfu_get_private(vfu_ctx);
int idx;
for (idx = 0; idx < NR_DMA_REGIONS; idx++) {
if (server_data->regions[idx].addr == 0 &&
server_data->regions[idx].len == 0)
break;
}
if (idx >= NR_DMA_REGIONS) {
errx(EXIT_FAILURE, "Failed to add dma region, slots full\n");
}
server_data->regions[idx].addr = iova;
server_data->regions[idx].len = len;
server_data->regions[idx].prot = prot;
}
static int
unmap_dma(vfu_ctx_t *vfu_ctx, uint64_t iova, uint64_t len)
{
struct server_data *server_data = vfu_get_private(vfu_ctx);
int idx;
for (idx = 0; idx < NR_DMA_REGIONS; idx++) {
if (server_data->regions[idx].addr == iova &&
server_data->regions[idx].len == len) {
server_data->regions[idx].addr = 0;
server_data->regions[idx].len = 0;
return 0;
}
}
return -EINVAL;
}
void get_md5sum(unsigned char *buf, int len, unsigned char *md5sum)
{
MD5_CTX ctx;
MD5_Init(&ctx);
MD5_Update(&ctx, buf, len);
MD5_Final(md5sum, &ctx);
return;
}
/*
* FIXME this function does DMA write/read using messages. This should be done
* on a region that is not memory mappable or an area of a region that is not
* sparsely memory mappable. We should also have a test where the server does
* DMA directly on the client memory.
*/
static void do_dma_io(vfu_ctx_t *vfu_ctx, struct server_data *server_data)
{
int count = 4096;
unsigned char buf[count];
unsigned char md5sum1[MD5_DIGEST_LENGTH], md5sum2[MD5_DIGEST_LENGTH];
int i, ret;
dma_sg_t sg;
assert(vfu_ctx != NULL);
ret = vfu_addr_to_sg(vfu_ctx, server_data->regions[0].addr, count, &sg,
1, PROT_WRITE);
if (ret < 0) {
errx(EXIT_FAILURE, "failed to map %#lx-%#lx: %s\n",
server_data->regions[0].addr,
server_data->regions[0].addr + count -1, strerror(-ret));
}
memset(buf, 'A', count);
get_md5sum(buf, count, md5sum1);
printf("%s: WRITE addr %#lx count %d\n", __func__,
server_data->regions[0].addr, count);
ret = vfu_dma_write(vfu_ctx, &sg, buf);
if (ret < 0) {
errx(EXIT_FAILURE, "vfu_dma_write failed: %s\n", strerror(-ret));
}
memset(buf, 0, count);
printf("%s: READ addr %#lx count %d\n", __func__,
server_data->regions[0].addr, count);
ret = vfu_dma_read(vfu_ctx, &sg, buf);
if (ret < 0) {
errx(EXIT_FAILURE, "vfu_dma_read failed: %s\n", strerror(-ret));
}
get_md5sum(buf, count, md5sum2);
for(i = 0; i < MD5_DIGEST_LENGTH; i++) {
if (md5sum2[i] != md5sum1[i]) {
errx(EXIT_FAILURE, "DMA write and DMA read mismatch\n");
}
}
}
static int device_reset(vfu_ctx_t *vfu_ctx UNUSED)
{
printf("device reset callback\n");
return 0;
}
static int
migration_device_state_transition(vfu_ctx_t *vfu_ctx, vfu_migr_state_t state)
{
struct server_data *server_data = vfu_get_private(vfu_ctx);
int ret;
struct itimerval new = { { 0 }, };
printf("migration: transition to device state %d\n", state);
switch (state) {
case VFU_MIGR_STATE_STOP_AND_COPY:
vfu_log(vfu_ctx, LOG_DEBUG, "disable timer");
if (setitimer(ITIMER_REAL, &new, NULL) != 0) {
err(EXIT_FAILURE, "failed to disable timer");
}
server_data->migration.pending_bytes = server_data->bar1_size + sizeof(time_t); /* FIXME BAR0 region size */
break;
case VFU_MIGR_STATE_PRE_COPY:
/* TODO must be less than size of data region in migration region */
server_data->migration.pending_bytes = server_data->bar1_size;
break;
case VFU_MIGR_STATE_STOP:
/* FIXME should gracefully fail */
assert(server_data->migration.pending_bytes == 0);
break;
case VFU_MIGR_STATE_RESUME:
break;
case VFU_MIGR_STATE_RUNNING:
ret = arm_timer(vfu_ctx, server_data->bar0);
if (ret < 0) {
return ret;
}
break;
default:
assert(false); /* FIXME */
}
server_data->migration.state = state;
return 0;
}
static __u64
migration_get_pending_bytes(vfu_ctx_t *vfu_ctx)
{
struct server_data *server_data = vfu_get_private(vfu_ctx);
return server_data->migration.pending_bytes;
}
static int
migration_prepare_data(vfu_ctx_t *vfu_ctx, __u64 *offset, __u64 *size)
{
struct server_data *server_data = vfu_get_private(vfu_ctx);
*offset = 0;
if (size != NULL) {
*size = server_data->migration.pending_bytes;
}
return 0;
}
static ssize_t
migration_read_data(vfu_ctx_t *vfu_ctx, void *buf, __u64 size, __u64 offset)
{
struct server_data *server_data = vfu_get_private(vfu_ctx);
if (server_data->migration.state != VFU_MIGR_STATE_PRE_COPY &&
server_data->migration.state != VFU_MIGR_STATE_STOP_AND_COPY)
{
return size;
}
/*
* For ease of implementation we expect the client to read all migration
* data in one go; partial reads are not supported. This is allowed by VFIO
* however we don't yet support it. Similarly, when resuming, partial
* writes are supported by VFIO, however we don't in this sample.
*
* If in pre-copy state we copy BAR1, if in stop-and-copy state we copy
* both BAR1 and BAR0. Since we always copy BAR1 in the stop-and-copy state,
* copying BAR1 in the pre-copy state is pointless. Fixing this requires
* more complex state tracking which exceeds the scope of this sample.
*/
if (offset != 0 || size != server_data->migration.pending_bytes) {
return -EINVAL;
}
memcpy(buf, server_data->bar1, server_data->bar1_size);
if (server_data->migration.state == VFU_MIGR_STATE_STOP_AND_COPY) {
memcpy(buf + server_data->bar1_size, &server_data->bar0,
sizeof server_data->bar0);
}
server_data->migration.pending_bytes = 0;
return size;
}
static ssize_t
migration_write_data(vfu_ctx_t *vfu_ctx, void *data, __u64 size, __u64 offset)
{
struct server_data *server_data = vfu_get_private(vfu_ctx);
char *buf = data;
int ret;
assert(server_data != NULL);
assert(data != NULL);
if (offset != 0 || size < server_data->bar1_size) {
vfu_log(vfu_ctx, LOG_DEBUG, "XXX bad migration data write %#llx-%#llx",
offset, offset + size - 1);
return -EINVAL;
}
memcpy(server_data->bar1, buf, server_data->bar1_size);
buf += server_data->bar1_size;
size -= server_data->bar1_size;
if (size == 0) {
return 0;
}
if (size != sizeof server_data->bar0) {
return -EINVAL;
}
memcpy(&server_data->bar0, buf, sizeof server_data->bar0);
ret = bar0_access(vfu_ctx, buf, sizeof server_data->bar0, 0, true);
assert(ret == (int)size); /* FIXME */
return 0;
}
static int
migration_data_written(UNUSED vfu_ctx_t *vfu_ctx, UNUSED __u64 count)
{
/*
* We apply migration state directly in the migration_write_data callback,
* so we don't need to do anything here. We would have to apply migration
* state in this callback if the migration region was memory mappable, in
* which case we wouldn't know when the client wrote migration data.
*/
return 0;
}
size_t
nr_pages(size_t size)
{
return (size / sysconf(_SC_PAGE_SIZE) + (size % sysconf(_SC_PAGE_SIZE) > 1));
}
size_t
page_align(size_t size) {
return nr_pages(size) * sysconf(_SC_PAGE_SIZE);
}
int main(int argc, char *argv[])
{
int ret;
bool verbose = false;
char opt;
struct sigaction act = {.sa_handler = _sa_handler};
const size_t bar1_size = 0x3000;
size_t migr_regs_size, migr_data_size, migr_size;
struct server_data server_data = {
.migration = {
.state = VFU_MIGR_STATE_RUNNING
}
};
vfu_ctx_t *vfu_ctx;
FILE *bar1_fp, *migr_fp;
const vfu_migration_callbacks_t migr_callbacks = {
.version = VFU_MIGR_CALLBACKS_VERS,
.transition = &migration_device_state_transition,
.get_pending_bytes = &migration_get_pending_bytes,
.prepare_data = &migration_prepare_data,
.read_data = &migration_read_data,
.data_written = &migration_data_written,
.write_data = &migration_write_data
};
while ((opt = getopt(argc, argv, "v")) != -1) {
switch (opt) {
case 'v':
verbose = true;
break;
default: /* '?' */
errx(EXIT_FAILURE, "Usage: %s [-v] <socketpath>\n", argv[0]);
}
}
if (optind >= argc) {
errx(EXIT_FAILURE, "missing vfio-user socket path");
}
sigemptyset(&act.sa_mask);
if (sigaction(SIGALRM, &act, NULL) == -1) {
err(EXIT_FAILURE, "failed to register signal handler");
}
vfu_ctx = vfu_create_ctx(VFU_TRANS_SOCK, argv[optind], 0, &server_data,
VFU_DEV_TYPE_PCI);
if (vfu_ctx == NULL) {
err(EXIT_FAILURE, "failed to initialize device emulation\n");
}
ret = vfu_setup_log(vfu_ctx, _log, verbose ? LOG_DEBUG : LOG_ERR);
if (ret < 0) {
err(EXIT_FAILURE, "failed to setup log");
}
ret = vfu_pci_init(vfu_ctx, VFU_PCI_TYPE_CONVENTIONAL,
PCI_HEADER_TYPE_NORMAL, 0);
if (ret < 0) {
err(EXIT_FAILURE, "vfu_pci_init() failed") ;
}
vfu_pci_set_id(vfu_ctx, 0xdead, 0xbeef, 0xcafe, 0xbabe);
ret = vfu_setup_region(vfu_ctx, VFU_PCI_DEV_BAR0_REGION_IDX, sizeof(time_t),
&bar0_access, VFU_REGION_FLAG_RW, NULL, 0, -1);
if (ret < 0) {
err(EXIT_FAILURE, "failed to setup BAR0 region");
}
/*
* Setup BAR1 to be 3 pages in size where only the first and the last pages
* are mappable. The client can still mmap the 2nd page, we can't prohibit
* this under Linux. If we really want to prohibit it we have to use
* separate files for the same region.
*/
if ((bar1_fp = tmpfile()) == NULL) {
err(EXIT_FAILURE, "failed to create BAR1 file");
}
server_data.bar1_size = bar1_size;
if (ftruncate(fileno(bar1_fp), server_data.bar1_size) == -1) {
err(EXIT_FAILURE, "failed to truncate BAR1 file");
}
server_data.bar1 = mmap(NULL, server_data.bar1_size, PROT_READ | PROT_WRITE,
MAP_SHARED, fileno(bar1_fp), 0);
if (server_data.bar1 == MAP_FAILED) {
err(EXIT_FAILURE, "failed to mmap BAR1");
}
struct iovec bar1_mmap_areas[] = {
{ .iov_base = (void*)0, .iov_len = 0x1000 },
{ .iov_base = (void*)0x2000, .iov_len = 0x1000 }
};
ret = vfu_setup_region(vfu_ctx, VFU_PCI_DEV_BAR1_REGION_IDX,
server_data.bar1_size, &bar1_access,
VFU_REGION_FLAG_RW, bar1_mmap_areas, 2,
fileno(bar1_fp));
if (ret < 0) {
err(EXIT_FAILURE, "failed to setup BAR1 region");
}
ret = vfu_setup_device_reset_cb(vfu_ctx, &device_reset);
if (ret < 0) {
err(EXIT_FAILURE, "failed to setup device reset callbacks");
}
ret = vfu_setup_device_dma_cb(vfu_ctx, &map_dma, &unmap_dma);
if (ret < 0) {
err(EXIT_FAILURE, "failed to setup device DMA callbacks");
}
ret = vfu_setup_device_nr_irqs(vfu_ctx, VFU_DEV_INTX_IRQ, 1);
if (ret < 0) {
err(EXIT_FAILURE, "failed to setup irq counts");
}
/* setup migration */
/*
* The migration registers aren't memory mappable, so in order to make the
* rest of the migration region memory mappable we must effectively reserve
* an entire page.
*/
migr_regs_size = vfu_get_migr_register_area_size();
migr_data_size = page_align(bar1_size + sizeof(time_t));
migr_size = migr_regs_size + migr_data_size;
if ((migr_fp = tmpfile()) == NULL) {
err(EXIT_FAILURE, "failed to create migration file");
}
if (ftruncate(fileno(migr_fp), migr_size) == -1) {
err(EXIT_FAILURE, "failed to truncate migration file");
}
server_data.bar1 = mmap(NULL, server_data.bar1_size, PROT_READ | PROT_WRITE,
MAP_SHARED, fileno(bar1_fp), 0);
if (server_data.bar1 == MAP_FAILED) {
err(EXIT_FAILURE, "failed to mmap migration file");
}
struct iovec migr_mmap_areas[] = {
[0] = {
.iov_base = (void *)migr_regs_size,
.iov_len = migr_data_size
},
};
ret = vfu_setup_region(vfu_ctx, VFU_PCI_DEV_MIGR_REGION_IDX, migr_size,
NULL, VFU_REGION_FLAG_RW, migr_mmap_areas,
ARRAY_SIZE(migr_mmap_areas), fileno(migr_fp));
if (ret < 0) {
err(EXIT_FAILURE, "failed to setup migration region");
}
ret = vfu_setup_device_migration_callbacks(vfu_ctx, &migr_callbacks,
migr_regs_size);
if (ret < 0) {
err(EXIT_FAILURE, "failed to setup device migration");
}
ret = vfu_realize_ctx(vfu_ctx);
if (ret < 0) {
err(EXIT_FAILURE, "failed to realize device");
}
ret = vfu_attach_ctx(vfu_ctx);
if (ret < 0) {
err(EXIT_FAILURE, "failed to attach device");
}
do {
ret = vfu_run_ctx(vfu_ctx);
if (ret == -EINTR) {
if (irq_triggered) {
irq_triggered = false;
vfu_irq_trigger(vfu_ctx, 0);
/*
* Apart from triggering an IRQ via the eventfd, we also
* trigger an IRQ via a message, simply for demonstrating how
* it's done. The client expects this behavior from the server.
*/
ret = vfu_irq_message(vfu_ctx, 0);
if (ret < 0) {
err(EXIT_FAILURE, "vfu_irq_message() failed");
}
/*
* We also initiate some dummy DMA via an explicit message,
* again to show how DMA is done. This is used if the client's
* RAM isn't mappable or the server implementation prefers it
* this way. Again, the client expects the server to send DMA
* messages right after it has triggered the IRQs.
*/
do_dma_io(vfu_ctx, &server_data);
ret = 0;
}
}
} while (ret == 0);
if (ret != -ENOTCONN && ret != -EINTR && ret != -ESHUTDOWN) {
errx(EXIT_FAILURE, "failed to realize device emulation: %s\n",
strerror(-ret));
}
vfu_destroy_ctx(vfu_ctx);
return EXIT_SUCCESS;
}
/* ex: set tabstop=4 shiftwidth=4 softtabstop=4 expandtab: */
|