Line data Source code
1 : /* SPDX-License-Identifier: BSD-3-Clause
2 : * Copyright (C) 2017 Intel Corporation.
3 : * All rights reserved.
4 : */
5 :
6 : #include "spdk/stdinc.h"
7 :
8 : #include "env_internal.h"
9 :
10 : #include "spdk/version.h"
11 : #include "spdk/env_dpdk.h"
12 : #include "spdk/log.h"
13 : #include "spdk/config.h"
14 :
15 : #include <openssl/ssl.h>
16 : #include <openssl/err.h>
17 :
18 : #include <rte_config.h>
19 : #include <rte_eal.h>
20 : #include <rte_errno.h>
21 : #include <rte_vfio.h>
22 :
23 : #define SPDK_ENV_DPDK_DEFAULT_NAME "spdk"
24 : #define SPDK_ENV_DPDK_DEFAULT_SHM_ID -1
25 : #define SPDK_ENV_DPDK_DEFAULT_MEM_SIZE -1
26 : #define SPDK_ENV_DPDK_DEFAULT_MAIN_CORE -1
27 : #define SPDK_ENV_DPDK_DEFAULT_MEM_CHANNEL -1
28 : #define SPDK_ENV_DPDK_DEFAULT_CORE_MASK "0x1"
29 : #define SPDK_ENV_DPDK_DEFAULT_BASE_VIRTADDR 0x200000000000
30 :
31 : #define DPDK_ALLOW_PARAM "--allow"
32 : #define DPDK_BLOCK_PARAM "--block"
33 : #define DPDK_MAIN_CORE_PARAM "--main-lcore"
34 :
35 : static char **g_eal_cmdline;
36 : static int g_eal_cmdline_argcount;
37 : static bool g_external_init = true;
38 :
39 : static char *
40 0 : _sprintf_alloc(const char *format, ...)
41 : {
42 0 : va_list args;
43 0 : va_list args_copy;
44 : char *buf;
45 : size_t bufsize;
46 : int rc;
47 :
48 0 : va_start(args, format);
49 :
50 : /* Try with a small buffer first. */
51 0 : bufsize = 32;
52 :
53 : /* Limit maximum buffer size to something reasonable so we don't loop forever. */
54 0 : while (bufsize <= 1024 * 1024) {
55 0 : buf = malloc(bufsize);
56 0 : if (buf == NULL) {
57 0 : va_end(args);
58 0 : return NULL;
59 : }
60 :
61 0 : va_copy(args_copy, args);
62 0 : rc = vsnprintf(buf, bufsize, format, args_copy);
63 0 : va_end(args_copy);
64 :
65 : /*
66 : * If vsnprintf() returned a count within our current buffer size, we are done.
67 : * The count does not include the \0 terminator, so rc == bufsize is not OK.
68 : */
69 0 : if (rc >= 0 && (size_t)rc < bufsize) {
70 0 : va_end(args);
71 0 : return buf;
72 : }
73 :
74 : /*
75 : * vsnprintf() should return the required space, but some libc versions do not
76 : * implement this correctly, so just double the buffer size and try again.
77 : *
78 : * We don't need the data in buf, so rather than realloc(), use free() and malloc()
79 : * again to avoid a copy.
80 : */
81 0 : free(buf);
82 0 : bufsize *= 2;
83 : }
84 :
85 0 : va_end(args);
86 0 : return NULL;
87 : }
88 :
89 : void
90 0 : spdk_env_opts_init_ext(struct spdk_env_opts *opts, size_t opts_size)
91 : {
92 0 : if (!opts) {
93 0 : return;
94 : }
95 :
96 0 : memset(opts, 0, opts_size);
97 0 : opts->opts_size = opts_size;
98 :
99 : /* These fields were all valid before this structure was ABI
100 : * versioned, so we just set them without checking opts_size.
101 : */
102 0 : opts->name = SPDK_ENV_DPDK_DEFAULT_NAME;
103 0 : opts->core_mask = SPDK_ENV_DPDK_DEFAULT_CORE_MASK;
104 0 : opts->shm_id = SPDK_ENV_DPDK_DEFAULT_SHM_ID;
105 0 : opts->mem_size = SPDK_ENV_DPDK_DEFAULT_MEM_SIZE;
106 0 : opts->main_core = SPDK_ENV_DPDK_DEFAULT_MAIN_CORE;
107 0 : opts->mem_channel = SPDK_ENV_DPDK_DEFAULT_MEM_CHANNEL;
108 0 : opts->base_virtaddr = SPDK_ENV_DPDK_DEFAULT_BASE_VIRTADDR;
109 :
110 : #define SET_FIELD(field, value) \
111 : if (offsetof(struct spdk_env_opts, field) + sizeof(opts->field) <= opts_size) { \
112 : opts->field = value; \
113 : }
114 :
115 : #undef SET_FIELD
116 : }
117 :
118 0 : SPDK_LOG_DEPRECATION_REGISTER(spdk_env_opts_init, "spdk_env_opts_init()", "v25.05", 0);
119 :
120 : void
121 0 : spdk_env_opts_init(struct spdk_env_opts *opts)
122 : {
123 0 : struct spdk_env_opts _opts = {};
124 :
125 0 : SPDK_LOG_DEPRECATED(spdk_env_opts_init);
126 :
127 : /* This function predates the ABI versioning of spdk_env_opts, so
128 : * we just copy over the defaults for the size of the structure
129 : * when the ABI versioning was first introduced.
130 : */
131 0 : spdk_env_opts_init_ext(&_opts, sizeof(_opts));
132 0 : memcpy(opts, &_opts, offsetof(struct spdk_env_opts, opts_size));
133 0 : }
134 :
135 : static void
136 0 : free_args(char **args, int argcount)
137 : {
138 : int i;
139 :
140 0 : if (args == NULL) {
141 0 : return;
142 : }
143 :
144 0 : for (i = 0; i < argcount; i++) {
145 0 : free(args[i]);
146 : }
147 :
148 0 : if (argcount) {
149 0 : free(args);
150 : }
151 : }
152 :
153 : static char **
154 0 : push_arg(char *args[], int *argcount, char *arg)
155 : {
156 : char **tmp;
157 :
158 0 : if (arg == NULL) {
159 0 : SPDK_ERRLOG("%s: NULL arg supplied\n", __func__);
160 0 : free_args(args, *argcount);
161 0 : return NULL;
162 : }
163 :
164 0 : tmp = realloc(args, sizeof(char *) * (*argcount + 1));
165 0 : if (tmp == NULL) {
166 0 : free(arg);
167 0 : free_args(args, *argcount);
168 0 : return NULL;
169 : }
170 :
171 0 : tmp[*argcount] = arg;
172 0 : (*argcount)++;
173 :
174 0 : return tmp;
175 : }
176 :
177 : #if defined(__linux__) && defined(__x86_64__)
178 :
179 : /* TODO: Can likely get this value from rlimits in the future */
180 : #define SPDK_IOMMU_VA_REQUIRED_WIDTH 48
181 : #define VTD_CAP_MGAW_SHIFT 16
182 : #define VTD_CAP_MGAW_MASK (0x3F << VTD_CAP_MGAW_SHIFT)
183 : #define RD_AMD_CAP_VASIZE_SHIFT 15
184 : #define RD_AMD_CAP_VASIZE_MASK (0x7F << RD_AMD_CAP_VASIZE_SHIFT)
185 :
186 : static int
187 0 : get_iommu_width(void)
188 : {
189 0 : int width = 0;
190 0 : glob_t glob_results = {};
191 :
192 : /* Break * and / into separate strings to appease check_format.sh comment style check. */
193 0 : glob("/sys/devices/virtual/iommu/dmar*" "/intel-iommu/cap", 0, NULL, &glob_results);
194 0 : glob("/sys/class/iommu/ivhd*" "/amd-iommu/cap", GLOB_APPEND, NULL, &glob_results);
195 :
196 0 : for (size_t i = 0; i < glob_results.gl_pathc; i++) {
197 0 : const char *filename = glob_results.gl_pathv[0];
198 0 : FILE *file = fopen(filename, "r");
199 0 : uint64_t cap_reg = 0;
200 :
201 0 : if (file == NULL) {
202 0 : continue;
203 : }
204 :
205 0 : if (fscanf(file, "%" PRIx64, &cap_reg) == 1) {
206 0 : if (strstr(filename, "intel-iommu") != NULL) {
207 : /* We have an Intel IOMMU */
208 0 : int mgaw = ((cap_reg & VTD_CAP_MGAW_MASK) >> VTD_CAP_MGAW_SHIFT) + 1;
209 :
210 0 : if (width == 0 || (mgaw > 0 && mgaw < width)) {
211 0 : width = mgaw;
212 : }
213 0 : } else if (strstr(filename, "amd-iommu") != NULL) {
214 : /* We have an AMD IOMMU */
215 0 : int mgaw = ((cap_reg & RD_AMD_CAP_VASIZE_MASK) >> RD_AMD_CAP_VASIZE_SHIFT) + 1;
216 :
217 0 : if (width == 0 || (mgaw > 0 && mgaw < width)) {
218 0 : width = mgaw;
219 : }
220 : }
221 : }
222 :
223 0 : fclose(file);
224 : }
225 :
226 0 : globfree(&glob_results);
227 0 : return width;
228 : }
229 :
230 : #endif
231 :
232 : static int
233 0 : build_eal_cmdline(const struct spdk_env_opts *opts)
234 : {
235 0 : int argcount = 0;
236 : char **args;
237 : bool no_huge;
238 :
239 0 : args = NULL;
240 0 : no_huge = opts->no_huge || (opts->env_context && strstr(opts->env_context, "--no-huge") != NULL);
241 :
242 : /* set the program name */
243 0 : args = push_arg(args, &argcount, _sprintf_alloc("%s", opts->name));
244 0 : if (args == NULL) {
245 0 : return -1;
246 : }
247 :
248 : /* disable shared configuration files when in single process mode. This allows for cleaner shutdown */
249 0 : if (opts->shm_id < 0) {
250 0 : args = push_arg(args, &argcount, _sprintf_alloc("%s", "--no-shconf"));
251 0 : if (args == NULL) {
252 0 : return -1;
253 : }
254 : }
255 :
256 : /* Either lcore_map or core_mask must be set. If both, or none specified, fail */
257 0 : if ((opts->core_mask == NULL) == (opts->lcore_map == NULL)) {
258 0 : if (opts->core_mask && opts->lcore_map) {
259 0 : fprintf(stderr,
260 : "Both, lcore map and core mask are provided, while only one can be set\n");
261 : } else {
262 0 : fprintf(stderr, "Core mask or lcore map must be specified\n");
263 : }
264 0 : free_args(args, argcount);
265 0 : return -1;
266 : }
267 :
268 0 : if (opts->lcore_map) {
269 : /* If lcore list is set, generate --lcores parameter */
270 0 : args = push_arg(args, &argcount, _sprintf_alloc("--lcores=%s", opts->lcore_map));
271 0 : } else if (opts->core_mask[0] == '-') {
272 : /*
273 : * Set the coremask:
274 : *
275 : * - if it starts with '-', we presume it's literal EAL arguments such
276 : * as --lcores.
277 : *
278 : * - if it starts with '[', we presume it's a core list to use with the
279 : * -l option.
280 : *
281 : * - otherwise, it's a CPU mask of the form "0xff.." as expected by the
282 : * -c option.
283 : */
284 0 : args = push_arg(args, &argcount, _sprintf_alloc("%s", opts->core_mask));
285 0 : } else if (opts->core_mask[0] == '[') {
286 0 : char *l_arg = _sprintf_alloc("-l %s", opts->core_mask + 1);
287 :
288 0 : if (l_arg != NULL) {
289 0 : int len = strlen(l_arg);
290 :
291 0 : if (l_arg[len - 1] == ']') {
292 0 : l_arg[len - 1] = '\0';
293 : }
294 : }
295 0 : args = push_arg(args, &argcount, l_arg);
296 : } else {
297 0 : args = push_arg(args, &argcount, _sprintf_alloc("-c %s", opts->core_mask));
298 : }
299 :
300 0 : if (args == NULL) {
301 0 : return -1;
302 : }
303 :
304 : /* set the memory channel number */
305 0 : if (opts->mem_channel > 0) {
306 0 : args = push_arg(args, &argcount, _sprintf_alloc("-n %d", opts->mem_channel));
307 0 : if (args == NULL) {
308 0 : return -1;
309 : }
310 : }
311 :
312 : /* set the memory size */
313 0 : if (opts->mem_size >= 0) {
314 0 : args = push_arg(args, &argcount, _sprintf_alloc("-m %d", opts->mem_size));
315 0 : if (args == NULL) {
316 0 : return -1;
317 : }
318 : }
319 :
320 : /* set no huge pages */
321 0 : if (opts->no_huge) {
322 0 : mem_disable_huge_pages();
323 : }
324 :
325 : /* set the main core */
326 0 : if (opts->main_core > 0) {
327 0 : args = push_arg(args, &argcount, _sprintf_alloc("%s=%d",
328 0 : DPDK_MAIN_CORE_PARAM, opts->main_core));
329 0 : if (args == NULL) {
330 0 : return -1;
331 : }
332 : }
333 :
334 : /* set no pci if enabled */
335 0 : if (opts->no_pci) {
336 0 : args = push_arg(args, &argcount, _sprintf_alloc("--no-pci"));
337 0 : if (args == NULL) {
338 0 : return -1;
339 : }
340 : }
341 :
342 0 : if (no_huge) {
343 0 : if (opts->hugepage_single_segments || opts->unlink_hugepage || opts->hugedir) {
344 0 : fprintf(stderr, "--no-huge invalid with other hugepage options\n");
345 0 : free_args(args, argcount);
346 0 : return -1;
347 : }
348 :
349 0 : if (opts->mem_size < 0) {
350 0 : fprintf(stderr,
351 : "Disabling hugepages requires specifying how much memory "
352 : "will be allocated using -s parameter\n");
353 0 : free_args(args, argcount);
354 0 : return -1;
355 : }
356 :
357 : /* iova-mode=pa is incompatible with no_huge */
358 0 : if (opts->iova_mode &&
359 0 : (strcmp(opts->iova_mode, "pa") == 0)) {
360 0 : fprintf(stderr, "iova-mode=pa is incompatible with specified "
361 : "no-huge parameter\n");
362 0 : free_args(args, argcount);
363 0 : return -1;
364 : }
365 :
366 0 : args = push_arg(args, &argcount, _sprintf_alloc("--no-huge"));
367 0 : args = push_arg(args, &argcount, _sprintf_alloc("--iova-mode=va"));
368 :
369 : } else {
370 : /* create just one hugetlbfs file */
371 0 : if (opts->hugepage_single_segments) {
372 0 : args = push_arg(args, &argcount, _sprintf_alloc("--single-file-segments"));
373 0 : if (args == NULL) {
374 0 : return -1;
375 : }
376 : }
377 :
378 : /* unlink hugepages after initialization */
379 : /* Note: Automatically unlink hugepage when shm_id < 0, since it means we're not using
380 : * multi-process so we don't need the hugepage links anymore. But we need to make sure
381 : * we don't specify --huge-unlink implicitly if --single-file-segments was specified since
382 : * DPDK doesn't support that.
383 : */
384 0 : if (opts->unlink_hugepage ||
385 0 : (opts->shm_id < 0 && !opts->hugepage_single_segments)) {
386 0 : args = push_arg(args, &argcount, _sprintf_alloc("--huge-unlink"));
387 0 : if (args == NULL) {
388 0 : return -1;
389 : }
390 : }
391 :
392 : /* use a specific hugetlbfs mount */
393 0 : if (opts->hugedir) {
394 0 : args = push_arg(args, &argcount, _sprintf_alloc("--huge-dir=%s", opts->hugedir));
395 0 : if (args == NULL) {
396 0 : return -1;
397 : }
398 : }
399 : }
400 :
401 0 : if (opts->num_pci_addr) {
402 : size_t i;
403 0 : char bdf[32];
404 0 : struct spdk_pci_addr *pci_addr =
405 0 : opts->pci_blocked ? opts->pci_blocked : opts->pci_allowed;
406 :
407 0 : for (i = 0; i < opts->num_pci_addr; i++) {
408 0 : spdk_pci_addr_fmt(bdf, 32, &pci_addr[i]);
409 0 : args = push_arg(args, &argcount, _sprintf_alloc("%s=%s",
410 0 : (opts->pci_blocked ? DPDK_BLOCK_PARAM : DPDK_ALLOW_PARAM),
411 : bdf));
412 0 : if (args == NULL) {
413 0 : return -1;
414 : }
415 : }
416 : }
417 :
418 : /* Disable DPDK telemetry information by default, can be modified with env_context.
419 : * Prevents creation of dpdk_telemetry socket and additional pthread for it.
420 : */
421 0 : args = push_arg(args, &argcount, _sprintf_alloc("--no-telemetry"));
422 0 : if (args == NULL) {
423 0 : return -1;
424 : }
425 :
426 : /* Lower default EAL loglevel to RTE_LOG_NOTICE - normal, but significant messages.
427 : * This can be overridden by specifying the same option in opts->env_context
428 : */
429 0 : args = push_arg(args, &argcount, strdup("--log-level=lib.eal:6"));
430 0 : if (args == NULL) {
431 0 : return -1;
432 : }
433 :
434 : /* Lower default CRYPTO loglevel to RTE_LOG_WARNING to avoid a ton of init msgs.
435 : * This can be overridden by specifying the same option in opts->env_context
436 : */
437 0 : args = push_arg(args, &argcount, strdup("--log-level=lib.cryptodev:5"));
438 0 : if (args == NULL) {
439 0 : return -1;
440 : }
441 :
442 : /* Lower default POWER loglevel to RTE_LOG_WARNING to avoid a ton of init msgs.
443 : * This can be overridden by specifying the same option in opts->env_context
444 : */
445 0 : args = push_arg(args, &argcount, strdup("--log-level=lib.power:5"));
446 0 : if (args == NULL) {
447 0 : return -1;
448 : }
449 :
450 : /* `user1` log type is used by rte_vhost, which prints an INFO log for each received
451 : * vhost user message. We don't want that. The same log type is also used by a couple
452 : * of other DPDK libs, but none of which we make use right now. If necessary, this can
453 : * be overridden via opts->env_context.
454 : */
455 0 : args = push_arg(args, &argcount, strdup("--log-level=user1:6"));
456 0 : if (args == NULL) {
457 0 : return -1;
458 : }
459 :
460 : #ifdef __linux__
461 :
462 0 : if (opts->iova_mode) {
463 : /* iova-mode=pa is incompatible with no_huge */
464 0 : args = push_arg(args, &argcount, _sprintf_alloc("--iova-mode=%s", opts->iova_mode));
465 0 : if (args == NULL) {
466 0 : return -1;
467 : }
468 : } else {
469 : /* When using vfio with enable_unsafe_noiommu_mode=Y, we need iova-mode=pa,
470 : * but DPDK guesses it should be iova-mode=va. Add a check and force
471 : * iova-mode=pa here. */
472 0 : if (!no_huge && rte_vfio_noiommu_is_enabled()) {
473 0 : args = push_arg(args, &argcount, _sprintf_alloc("--iova-mode=pa"));
474 0 : if (args == NULL) {
475 0 : return -1;
476 : }
477 : }
478 :
479 : #if defined(__x86_64__)
480 : /* DPDK by default guesses that it should be using iova-mode=va so that it can
481 : * support running as an unprivileged user. However, some systems (especially
482 : * virtual machines) don't have an IOMMU capable of handling the full virtual
483 : * address space and DPDK doesn't currently catch that. Add a check in SPDK
484 : * and force iova-mode=pa here. */
485 0 : if (!no_huge && get_iommu_width() < SPDK_IOMMU_VA_REQUIRED_WIDTH) {
486 0 : args = push_arg(args, &argcount, _sprintf_alloc("--iova-mode=pa"));
487 0 : if (args == NULL) {
488 0 : return -1;
489 : }
490 : }
491 : #elif defined(__PPC64__)
492 : /* On Linux + PowerPC, DPDK doesn't support VA mode at all. Unfortunately, it doesn't correctly
493 : * auto-detect at the moment, so we'll just force it here. */
494 : args = push_arg(args, &argcount, _sprintf_alloc("--iova-mode=pa"));
495 : if (args == NULL) {
496 : return -1;
497 : }
498 : #endif
499 : }
500 :
501 :
502 : /* Set the base virtual address - it must be an address that is not in the
503 : * ASAN shadow region, otherwise ASAN-enabled builds will ignore the
504 : * mmap hint.
505 : *
506 : * Ref: https://github.com/google/sanitizers/wiki/AddressSanitizerAlgorithm
507 : */
508 0 : args = push_arg(args, &argcount, _sprintf_alloc("--base-virtaddr=0x%" PRIx64, opts->base_virtaddr));
509 0 : if (args == NULL) {
510 0 : return -1;
511 : }
512 :
513 : /* --match-allocation prevents DPDK from merging or splitting system memory allocations under the hood.
514 : * This is critical for RDMA when attempting to use an rte_mempool based buffer pool. If DPDK merges two
515 : * physically or IOVA contiguous memory regions, then when we go to allocate a buffer pool, it can split
516 : * the memory for a buffer over two allocations meaning the buffer will be split over a memory region.
517 : */
518 :
519 : /* --no-huge is incompatible with --match-allocations
520 : * Ref: https://doc.dpdk.org/guides/prog_guide/env_abstraction_layer.html#hugepage-allocation-matching
521 : */
522 0 : if (!no_huge &&
523 0 : (!opts->env_context || strstr(opts->env_context, "--legacy-mem") == NULL)) {
524 0 : args = push_arg(args, &argcount, _sprintf_alloc("%s", "--match-allocations"));
525 0 : if (args == NULL) {
526 0 : return -1;
527 : }
528 : }
529 :
530 0 : if (opts->shm_id < 0) {
531 0 : args = push_arg(args, &argcount, _sprintf_alloc("--file-prefix=spdk_pid%d",
532 : getpid()));
533 0 : if (args == NULL) {
534 0 : return -1;
535 : }
536 : } else {
537 0 : args = push_arg(args, &argcount, _sprintf_alloc("--file-prefix=spdk%d",
538 0 : opts->shm_id));
539 0 : if (args == NULL) {
540 0 : return -1;
541 : }
542 :
543 : /* set the process type */
544 0 : args = push_arg(args, &argcount, _sprintf_alloc("--proc-type=auto"));
545 0 : if (args == NULL) {
546 0 : return -1;
547 : }
548 : }
549 :
550 : /* --vfio-vf-token used for VF initialized by vfio_pci driver. */
551 0 : if (opts->vf_token) {
552 0 : args = push_arg(args, &argcount, _sprintf_alloc("--vfio-vf-token=%s",
553 0 : opts->vf_token));
554 0 : if (args == NULL) {
555 0 : return -1;
556 : }
557 : }
558 : #endif
559 :
560 0 : if (opts->env_context) {
561 0 : char *ptr = strdup(opts->env_context);
562 0 : char *tok = strtok(ptr, " \t");
563 :
564 : /* DPDK expects each argument as a separate string in the argv
565 : * array, so we need to tokenize here in case the caller
566 : * passed multiple arguments in the env_context string.
567 : */
568 0 : while (tok != NULL) {
569 0 : args = push_arg(args, &argcount, strdup(tok));
570 0 : tok = strtok(NULL, " \t");
571 : }
572 :
573 0 : free(ptr);
574 : }
575 :
576 0 : g_eal_cmdline = args;
577 0 : g_eal_cmdline_argcount = argcount;
578 0 : return argcount;
579 : }
580 :
581 : int
582 0 : spdk_env_dpdk_post_init(bool legacy_mem)
583 : {
584 : int rc;
585 :
586 0 : rc = pci_env_init();
587 0 : if (rc < 0) {
588 0 : SPDK_ERRLOG("pci_env_init() failed\n");
589 0 : return rc;
590 : }
591 :
592 0 : rc = mem_map_init(legacy_mem);
593 0 : if (rc < 0) {
594 0 : SPDK_ERRLOG("Failed to allocate mem_map\n");
595 0 : return rc;
596 : }
597 :
598 0 : rc = vtophys_init();
599 0 : if (rc < 0) {
600 0 : SPDK_ERRLOG("Failed to initialize vtophys\n");
601 0 : return rc;
602 : }
603 :
604 0 : return 0;
605 : }
606 :
607 : void
608 0 : spdk_env_dpdk_post_fini(void)
609 : {
610 0 : pci_env_fini();
611 :
612 0 : free_args(g_eal_cmdline, g_eal_cmdline_argcount);
613 0 : g_eal_cmdline = NULL;
614 0 : g_eal_cmdline_argcount = 0;
615 0 : }
616 :
617 : static void
618 0 : env_copy_opts(struct spdk_env_opts *opts, const struct spdk_env_opts *opts_user, size_t opts_size)
619 : {
620 0 : spdk_env_opts_init_ext(opts, sizeof(*opts));
621 0 : memcpy(opts, opts_user, offsetof(struct spdk_env_opts, opts_size));
622 0 : opts->opts_size = opts_size;
623 :
624 : #define SET_FIELD(field) \
625 : if (offsetof(struct spdk_env_opts, field) + sizeof(opts->field) <= opts_size) { \
626 : opts->field = opts_user->field; \
627 : }
628 :
629 : #undef SET_FIELD
630 0 : }
631 :
632 : int
633 0 : spdk_env_init_ext(const struct spdk_env_opts *opts_user)
634 : {
635 0 : struct spdk_env_opts opts_local = {};
636 0 : struct spdk_env_opts *opts = &opts_local;
637 0 : char **dpdk_args = NULL;
638 0 : char *args_print = NULL, *args_tmp = NULL;
639 : OPENSSL_INIT_SETTINGS *settings;
640 : int i, rc;
641 : int orig_optind;
642 : bool legacy_mem;
643 :
644 : /* If SPDK env has been initialized before, then only pci env requires
645 : * reinitialization.
646 : */
647 0 : if (g_external_init == false) {
648 0 : if (opts_user != NULL) {
649 0 : fprintf(stderr, "Invalid arguments to reinitialize SPDK env\n");
650 0 : return -EINVAL;
651 : }
652 :
653 0 : printf("Starting %s / %s reinitialization...\n", SPDK_VERSION_STRING, rte_version());
654 0 : pci_env_reinit();
655 :
656 0 : return 0;
657 : }
658 :
659 0 : if (opts_user == NULL) {
660 0 : fprintf(stderr, "NULL arguments to initialize DPDK\n");
661 0 : return -EINVAL;
662 : }
663 :
664 0 : if (opts_user->opts_size < offsetof(struct spdk_env_opts, opts_size) + sizeof(opts->opts_size)) {
665 0 : fprintf(stderr, "Invalid opts->opts_size\n");
666 0 : return -EINVAL;
667 : }
668 :
669 0 : env_copy_opts(opts, opts_user, opts_user->opts_size);
670 :
671 0 : settings = OPENSSL_INIT_new();
672 0 : if (!settings) {
673 0 : fprintf(stderr, "Failed to create openssl settings object\n");
674 0 : ERR_print_errors_fp(stderr);
675 0 : return -ENOMEM;
676 : }
677 :
678 : #if OPENSSL_VERSION_NUMBER >= 0x30000000 /* OPENSSL 3.0.0 */
679 0 : OPENSSL_INIT_set_config_file_flags(settings, 0);
680 : #endif
681 0 : rc = OPENSSL_init_ssl(OPENSSL_INIT_LOAD_CONFIG, settings);
682 0 : if (rc != 1) {
683 0 : fprintf(stderr, "Failed to initialize OpenSSL\n");
684 0 : ERR_print_errors_fp(stderr);
685 0 : return -EINVAL;
686 : }
687 0 : OPENSSL_INIT_free(settings);
688 :
689 0 : rc = build_eal_cmdline(opts);
690 0 : if (rc < 0) {
691 0 : SPDK_ERRLOG("Invalid arguments to initialize DPDK\n");
692 0 : return -EINVAL;
693 : }
694 :
695 0 : SPDK_PRINTF("Starting %s / %s initialization...\n", SPDK_VERSION_STRING, rte_version());
696 :
697 0 : args_print = _sprintf_alloc("[ DPDK EAL parameters: ");
698 0 : if (args_print == NULL) {
699 0 : return -ENOMEM;
700 : }
701 0 : for (i = 0; i < g_eal_cmdline_argcount; i++) {
702 0 : args_tmp = args_print;
703 0 : args_print = _sprintf_alloc("%s%s ", args_tmp, g_eal_cmdline[i]);
704 0 : if (args_print == NULL) {
705 0 : free(args_tmp);
706 0 : return -ENOMEM;
707 : }
708 0 : free(args_tmp);
709 : }
710 0 : SPDK_PRINTF("%s]\n", args_print);
711 0 : free(args_print);
712 :
713 : /* DPDK rearranges the array we pass to it, so make a copy
714 : * before passing so we can still free the individual strings
715 : * correctly.
716 : */
717 0 : dpdk_args = calloc(g_eal_cmdline_argcount, sizeof(char *));
718 0 : if (dpdk_args == NULL) {
719 0 : SPDK_ERRLOG("Failed to allocate dpdk_args\n");
720 0 : return -ENOMEM;
721 : }
722 0 : memcpy(dpdk_args, g_eal_cmdline, sizeof(char *) * g_eal_cmdline_argcount);
723 :
724 0 : fflush(stdout);
725 0 : orig_optind = optind;
726 0 : optind = 1;
727 0 : rc = rte_eal_init(g_eal_cmdline_argcount, dpdk_args);
728 0 : optind = orig_optind;
729 :
730 0 : free(dpdk_args);
731 :
732 0 : if (rc < 0) {
733 0 : if (rte_errno == EALREADY) {
734 0 : SPDK_ERRLOG("DPDK already initialized\n");
735 : } else {
736 0 : SPDK_ERRLOG("Failed to initialize DPDK\n");
737 : }
738 0 : return -rte_errno;
739 : }
740 :
741 0 : legacy_mem = false;
742 0 : if (opts->env_context && strstr(opts->env_context, "--legacy-mem") != NULL) {
743 0 : legacy_mem = true;
744 : }
745 :
746 0 : rc = spdk_env_dpdk_post_init(legacy_mem);
747 0 : if (rc == 0) {
748 0 : g_external_init = false;
749 : }
750 :
751 0 : return rc;
752 : }
753 :
754 0 : SPDK_LOG_DEPRECATION_REGISTER(spdk_env_init, "spdk_env_init()", "v25.05", 0);
755 :
756 : int
757 0 : spdk_env_init(const struct spdk_env_opts *opts_user)
758 : {
759 0 : SPDK_LOG_DEPRECATED(spdk_env_init);
760 :
761 : /* This is an old API, prior to having a size-versions spdk_env_opts
762 : * struct. This means we do not know the size of the _opts parameter,
763 : * we must assume it has size matching the structure when the
764 : * versioning was first introduced. So allocate latest version of
765 : * this struct on the stack and use env_copy_opts to copy over the
766 : * bytes we know must be valid.
767 : */
768 0 : struct spdk_env_opts opts = {};
769 :
770 0 : env_copy_opts(&opts, opts_user, offsetof(struct spdk_env_opts, opts_size) + sizeof(opts.opts_size));
771 0 : return spdk_env_init_ext(&opts);
772 : }
773 :
774 : /* We use priority 101 which is the highest priority level available
775 : * to applications (the toolchains reserve 1 to 100 for internal usage).
776 : * This ensures this destructor runs last, after any other destructors
777 : * that might still need the environment up and running.
778 : */
779 : __attribute__((destructor(101))) static void
780 0 : dpdk_cleanup(void)
781 : {
782 : /* Only call rte_eal_cleanup if the SPDK env library called rte_eal_init. */
783 0 : if (!g_external_init) {
784 0 : rte_eal_cleanup();
785 : }
786 0 : }
787 :
788 : void
789 0 : spdk_env_fini(void)
790 : {
791 0 : spdk_env_dpdk_post_fini();
792 0 : }
793 :
794 : bool
795 0 : spdk_env_dpdk_external_init(void)
796 : {
797 0 : return g_external_init;
798 : }
|