]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - arch/x86/kernel/kexec-bzimage64.c
Merge tag 'ktest-v5.13' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt...
[mirror_ubuntu-jammy-kernel.git] / arch / x86 / kernel / kexec-bzimage64.c
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Kexec bzImage loader
4 *
5 * Copyright (C) 2014 Red Hat Inc.
6 * Authors:
7 * Vivek Goyal <vgoyal@redhat.com>
8 */
9
10 #define pr_fmt(fmt) "kexec-bzImage64: " fmt
11
12 #include <linux/string.h>
13 #include <linux/printk.h>
14 #include <linux/errno.h>
15 #include <linux/slab.h>
16 #include <linux/kexec.h>
17 #include <linux/kernel.h>
18 #include <linux/mm.h>
19 #include <linux/efi.h>
20 #include <linux/verification.h>
21
22 #include <asm/bootparam.h>
23 #include <asm/setup.h>
24 #include <asm/crash.h>
25 #include <asm/efi.h>
26 #include <asm/e820/api.h>
27 #include <asm/kexec-bzimage64.h>
28
29 #define MAX_ELFCOREHDR_STR_LEN 30 /* elfcorehdr=0x<64bit-value> */
30
31 /*
32 * Defines lowest physical address for various segments. Not sure where
33 * exactly these limits came from. Current bzimage64 loader in kexec-tools
34 * uses these so I am retaining it. It can be changed over time as we gain
35 * more insight.
36 */
37 #define MIN_PURGATORY_ADDR 0x3000
38 #define MIN_BOOTPARAM_ADDR 0x3000
39 #define MIN_KERNEL_LOAD_ADDR 0x100000
40 #define MIN_INITRD_LOAD_ADDR 0x1000000
41
42 /*
43 * This is a place holder for all boot loader specific data structure which
44 * gets allocated in one call but gets freed much later during cleanup
45 * time. Right now there is only one field but it can grow as need be.
46 */
47 struct bzimage64_data {
48 /*
49 * Temporary buffer to hold bootparams buffer. This should be
50 * freed once the bootparam segment has been loaded.
51 */
52 void *bootparams_buf;
53 };
54
55 static int setup_initrd(struct boot_params *params,
56 unsigned long initrd_load_addr, unsigned long initrd_len)
57 {
58 params->hdr.ramdisk_image = initrd_load_addr & 0xffffffffUL;
59 params->hdr.ramdisk_size = initrd_len & 0xffffffffUL;
60
61 params->ext_ramdisk_image = initrd_load_addr >> 32;
62 params->ext_ramdisk_size = initrd_len >> 32;
63
64 return 0;
65 }
66
67 static int setup_cmdline(struct kimage *image, struct boot_params *params,
68 unsigned long bootparams_load_addr,
69 unsigned long cmdline_offset, char *cmdline,
70 unsigned long cmdline_len)
71 {
72 char *cmdline_ptr = ((char *)params) + cmdline_offset;
73 unsigned long cmdline_ptr_phys, len = 0;
74 uint32_t cmdline_low_32, cmdline_ext_32;
75
76 if (image->type == KEXEC_TYPE_CRASH) {
77 len = sprintf(cmdline_ptr,
78 "elfcorehdr=0x%lx ", image->elf_load_addr);
79 }
80 memcpy(cmdline_ptr + len, cmdline, cmdline_len);
81 cmdline_len += len;
82
83 cmdline_ptr[cmdline_len - 1] = '\0';
84
85 pr_debug("Final command line is: %s\n", cmdline_ptr);
86 cmdline_ptr_phys = bootparams_load_addr + cmdline_offset;
87 cmdline_low_32 = cmdline_ptr_phys & 0xffffffffUL;
88 cmdline_ext_32 = cmdline_ptr_phys >> 32;
89
90 params->hdr.cmd_line_ptr = cmdline_low_32;
91 if (cmdline_ext_32)
92 params->ext_cmd_line_ptr = cmdline_ext_32;
93
94 return 0;
95 }
96
97 static int setup_e820_entries(struct boot_params *params)
98 {
99 unsigned int nr_e820_entries;
100
101 nr_e820_entries = e820_table_kexec->nr_entries;
102
103 /* TODO: Pass entries more than E820_MAX_ENTRIES_ZEROPAGE in bootparams setup data */
104 if (nr_e820_entries > E820_MAX_ENTRIES_ZEROPAGE)
105 nr_e820_entries = E820_MAX_ENTRIES_ZEROPAGE;
106
107 params->e820_entries = nr_e820_entries;
108 memcpy(&params->e820_table, &e820_table_kexec->entries, nr_e820_entries*sizeof(struct e820_entry));
109
110 return 0;
111 }
112
113 #ifdef CONFIG_EFI
114 static int setup_efi_info_memmap(struct boot_params *params,
115 unsigned long params_load_addr,
116 unsigned int efi_map_offset,
117 unsigned int efi_map_sz)
118 {
119 void *efi_map = (void *)params + efi_map_offset;
120 unsigned long efi_map_phys_addr = params_load_addr + efi_map_offset;
121 struct efi_info *ei = &params->efi_info;
122
123 if (!efi_map_sz)
124 return 0;
125
126 efi_runtime_map_copy(efi_map, efi_map_sz);
127
128 ei->efi_memmap = efi_map_phys_addr & 0xffffffff;
129 ei->efi_memmap_hi = efi_map_phys_addr >> 32;
130 ei->efi_memmap_size = efi_map_sz;
131
132 return 0;
133 }
134
135 static int
136 prepare_add_efi_setup_data(struct boot_params *params,
137 unsigned long params_load_addr,
138 unsigned int efi_setup_data_offset)
139 {
140 unsigned long setup_data_phys;
141 struct setup_data *sd = (void *)params + efi_setup_data_offset;
142 struct efi_setup_data *esd = (void *)sd + sizeof(struct setup_data);
143
144 esd->fw_vendor = efi_fw_vendor;
145 esd->tables = efi_config_table;
146 esd->smbios = efi.smbios;
147
148 sd->type = SETUP_EFI;
149 sd->len = sizeof(struct efi_setup_data);
150
151 /* Add setup data */
152 setup_data_phys = params_load_addr + efi_setup_data_offset;
153 sd->next = params->hdr.setup_data;
154 params->hdr.setup_data = setup_data_phys;
155
156 return 0;
157 }
158
159 static int
160 setup_efi_state(struct boot_params *params, unsigned long params_load_addr,
161 unsigned int efi_map_offset, unsigned int efi_map_sz,
162 unsigned int efi_setup_data_offset)
163 {
164 struct efi_info *current_ei = &boot_params.efi_info;
165 struct efi_info *ei = &params->efi_info;
166
167 if (!efi_enabled(EFI_RUNTIME_SERVICES))
168 return 0;
169
170 if (!current_ei->efi_memmap_size)
171 return 0;
172
173 params->secure_boot = boot_params.secure_boot;
174 ei->efi_loader_signature = current_ei->efi_loader_signature;
175 ei->efi_systab = current_ei->efi_systab;
176 ei->efi_systab_hi = current_ei->efi_systab_hi;
177
178 ei->efi_memdesc_version = current_ei->efi_memdesc_version;
179 ei->efi_memdesc_size = efi_get_runtime_map_desc_size();
180
181 setup_efi_info_memmap(params, params_load_addr, efi_map_offset,
182 efi_map_sz);
183 prepare_add_efi_setup_data(params, params_load_addr,
184 efi_setup_data_offset);
185 return 0;
186 }
187 #endif /* CONFIG_EFI */
188
189 static int
190 setup_boot_parameters(struct kimage *image, struct boot_params *params,
191 unsigned long params_load_addr,
192 unsigned int efi_map_offset, unsigned int efi_map_sz,
193 unsigned int efi_setup_data_offset)
194 {
195 unsigned int nr_e820_entries;
196 unsigned long long mem_k, start, end;
197 int i, ret = 0;
198
199 /* Get subarch from existing bootparams */
200 params->hdr.hardware_subarch = boot_params.hdr.hardware_subarch;
201
202 /* Copying screen_info will do? */
203 memcpy(&params->screen_info, &screen_info, sizeof(struct screen_info));
204
205 /* Fill in memsize later */
206 params->screen_info.ext_mem_k = 0;
207 params->alt_mem_k = 0;
208
209 /* Always fill in RSDP: it is either 0 or a valid value */
210 params->acpi_rsdp_addr = boot_params.acpi_rsdp_addr;
211
212 /* Default APM info */
213 memset(&params->apm_bios_info, 0, sizeof(params->apm_bios_info));
214
215 /* Default drive info */
216 memset(&params->hd0_info, 0, sizeof(params->hd0_info));
217 memset(&params->hd1_info, 0, sizeof(params->hd1_info));
218
219 if (image->type == KEXEC_TYPE_CRASH) {
220 ret = crash_setup_memmap_entries(image, params);
221 if (ret)
222 return ret;
223 } else
224 setup_e820_entries(params);
225
226 nr_e820_entries = params->e820_entries;
227
228 for (i = 0; i < nr_e820_entries; i++) {
229 if (params->e820_table[i].type != E820_TYPE_RAM)
230 continue;
231 start = params->e820_table[i].addr;
232 end = params->e820_table[i].addr + params->e820_table[i].size - 1;
233
234 if ((start <= 0x100000) && end > 0x100000) {
235 mem_k = (end >> 10) - (0x100000 >> 10);
236 params->screen_info.ext_mem_k = mem_k;
237 params->alt_mem_k = mem_k;
238 if (mem_k > 0xfc00)
239 params->screen_info.ext_mem_k = 0xfc00; /* 64M*/
240 if (mem_k > 0xffffffff)
241 params->alt_mem_k = 0xffffffff;
242 }
243 }
244
245 #ifdef CONFIG_EFI
246 /* Setup EFI state */
247 setup_efi_state(params, params_load_addr, efi_map_offset, efi_map_sz,
248 efi_setup_data_offset);
249 #endif
250 /* Setup EDD info */
251 memcpy(params->eddbuf, boot_params.eddbuf,
252 EDDMAXNR * sizeof(struct edd_info));
253 params->eddbuf_entries = boot_params.eddbuf_entries;
254
255 memcpy(params->edd_mbr_sig_buffer, boot_params.edd_mbr_sig_buffer,
256 EDD_MBR_SIG_MAX * sizeof(unsigned int));
257
258 return ret;
259 }
260
261 static int bzImage64_probe(const char *buf, unsigned long len)
262 {
263 int ret = -ENOEXEC;
264 struct setup_header *header;
265
266 /* kernel should be at least two sectors long */
267 if (len < 2 * 512) {
268 pr_err("File is too short to be a bzImage\n");
269 return ret;
270 }
271
272 header = (struct setup_header *)(buf + offsetof(struct boot_params, hdr));
273 if (memcmp((char *)&header->header, "HdrS", 4) != 0) {
274 pr_err("Not a bzImage\n");
275 return ret;
276 }
277
278 if (header->boot_flag != 0xAA55) {
279 pr_err("No x86 boot sector present\n");
280 return ret;
281 }
282
283 if (header->version < 0x020C) {
284 pr_err("Must be at least protocol version 2.12\n");
285 return ret;
286 }
287
288 if (!(header->loadflags & LOADED_HIGH)) {
289 pr_err("zImage not a bzImage\n");
290 return ret;
291 }
292
293 if (!(header->xloadflags & XLF_KERNEL_64)) {
294 pr_err("Not a bzImage64. XLF_KERNEL_64 is not set.\n");
295 return ret;
296 }
297
298 if (!(header->xloadflags & XLF_CAN_BE_LOADED_ABOVE_4G)) {
299 pr_err("XLF_CAN_BE_LOADED_ABOVE_4G is not set.\n");
300 return ret;
301 }
302
303 /*
304 * Can't handle 32bit EFI as it does not allow loading kernel
305 * above 4G. This should be handled by 32bit bzImage loader
306 */
307 if (efi_enabled(EFI_RUNTIME_SERVICES) && !efi_enabled(EFI_64BIT)) {
308 pr_debug("EFI is 32 bit. Can't load kernel above 4G.\n");
309 return ret;
310 }
311
312 if (!(header->xloadflags & XLF_5LEVEL) && pgtable_l5_enabled()) {
313 pr_err("bzImage cannot handle 5-level paging mode.\n");
314 return ret;
315 }
316
317 /* I've got a bzImage */
318 pr_debug("It's a relocatable bzImage64\n");
319 ret = 0;
320
321 return ret;
322 }
323
324 static void *bzImage64_load(struct kimage *image, char *kernel,
325 unsigned long kernel_len, char *initrd,
326 unsigned long initrd_len, char *cmdline,
327 unsigned long cmdline_len)
328 {
329
330 struct setup_header *header;
331 int setup_sects, kern16_size, ret = 0;
332 unsigned long setup_header_size, params_cmdline_sz;
333 struct boot_params *params;
334 unsigned long bootparam_load_addr, kernel_load_addr, initrd_load_addr;
335 struct bzimage64_data *ldata;
336 struct kexec_entry64_regs regs64;
337 void *stack;
338 unsigned int setup_hdr_offset = offsetof(struct boot_params, hdr);
339 unsigned int efi_map_offset, efi_map_sz, efi_setup_data_offset;
340 struct kexec_buf kbuf = { .image = image, .buf_max = ULONG_MAX,
341 .top_down = true };
342 struct kexec_buf pbuf = { .image = image, .buf_min = MIN_PURGATORY_ADDR,
343 .buf_max = ULONG_MAX, .top_down = true };
344
345 header = (struct setup_header *)(kernel + setup_hdr_offset);
346 setup_sects = header->setup_sects;
347 if (setup_sects == 0)
348 setup_sects = 4;
349
350 kern16_size = (setup_sects + 1) * 512;
351 if (kernel_len < kern16_size) {
352 pr_err("bzImage truncated\n");
353 return ERR_PTR(-ENOEXEC);
354 }
355
356 if (cmdline_len > header->cmdline_size) {
357 pr_err("Kernel command line too long\n");
358 return ERR_PTR(-EINVAL);
359 }
360
361 /*
362 * In case of crash dump, we will append elfcorehdr=<addr> to
363 * command line. Make sure it does not overflow
364 */
365 if (cmdline_len + MAX_ELFCOREHDR_STR_LEN > header->cmdline_size) {
366 pr_debug("Appending elfcorehdr=<addr> to command line exceeds maximum allowed length\n");
367 return ERR_PTR(-EINVAL);
368 }
369
370 /* Allocate and load backup region */
371 if (image->type == KEXEC_TYPE_CRASH) {
372 ret = crash_load_segments(image);
373 if (ret)
374 return ERR_PTR(ret);
375 }
376
377 /*
378 * Load purgatory. For 64bit entry point, purgatory code can be
379 * anywhere.
380 */
381 ret = kexec_load_purgatory(image, &pbuf);
382 if (ret) {
383 pr_err("Loading purgatory failed\n");
384 return ERR_PTR(ret);
385 }
386
387 pr_debug("Loaded purgatory at 0x%lx\n", pbuf.mem);
388
389
390 /*
391 * Load Bootparams and cmdline and space for efi stuff.
392 *
393 * Allocate memory together for multiple data structures so
394 * that they all can go in single area/segment and we don't
395 * have to create separate segment for each. Keeps things
396 * little bit simple
397 */
398 efi_map_sz = efi_get_runtime_map_size();
399 params_cmdline_sz = sizeof(struct boot_params) + cmdline_len +
400 MAX_ELFCOREHDR_STR_LEN;
401 params_cmdline_sz = ALIGN(params_cmdline_sz, 16);
402 kbuf.bufsz = params_cmdline_sz + ALIGN(efi_map_sz, 16) +
403 sizeof(struct setup_data) +
404 sizeof(struct efi_setup_data);
405
406 params = kzalloc(kbuf.bufsz, GFP_KERNEL);
407 if (!params)
408 return ERR_PTR(-ENOMEM);
409 efi_map_offset = params_cmdline_sz;
410 efi_setup_data_offset = efi_map_offset + ALIGN(efi_map_sz, 16);
411
412 /* Copy setup header onto bootparams. Documentation/x86/boot.rst */
413 setup_header_size = 0x0202 + kernel[0x0201] - setup_hdr_offset;
414
415 /* Is there a limit on setup header size? */
416 memcpy(&params->hdr, (kernel + setup_hdr_offset), setup_header_size);
417
418 kbuf.buffer = params;
419 kbuf.memsz = kbuf.bufsz;
420 kbuf.buf_align = 16;
421 kbuf.buf_min = MIN_BOOTPARAM_ADDR;
422 ret = kexec_add_buffer(&kbuf);
423 if (ret)
424 goto out_free_params;
425 bootparam_load_addr = kbuf.mem;
426 pr_debug("Loaded boot_param, command line and misc at 0x%lx bufsz=0x%lx memsz=0x%lx\n",
427 bootparam_load_addr, kbuf.bufsz, kbuf.bufsz);
428
429 /* Load kernel */
430 kbuf.buffer = kernel + kern16_size;
431 kbuf.bufsz = kernel_len - kern16_size;
432 kbuf.memsz = PAGE_ALIGN(header->init_size);
433 kbuf.buf_align = header->kernel_alignment;
434 kbuf.buf_min = MIN_KERNEL_LOAD_ADDR;
435 kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
436 ret = kexec_add_buffer(&kbuf);
437 if (ret)
438 goto out_free_params;
439 kernel_load_addr = kbuf.mem;
440
441 pr_debug("Loaded 64bit kernel at 0x%lx bufsz=0x%lx memsz=0x%lx\n",
442 kernel_load_addr, kbuf.bufsz, kbuf.memsz);
443
444 /* Load initrd high */
445 if (initrd) {
446 kbuf.buffer = initrd;
447 kbuf.bufsz = kbuf.memsz = initrd_len;
448 kbuf.buf_align = PAGE_SIZE;
449 kbuf.buf_min = MIN_INITRD_LOAD_ADDR;
450 kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
451 ret = kexec_add_buffer(&kbuf);
452 if (ret)
453 goto out_free_params;
454 initrd_load_addr = kbuf.mem;
455
456 pr_debug("Loaded initrd at 0x%lx bufsz=0x%lx memsz=0x%lx\n",
457 initrd_load_addr, initrd_len, initrd_len);
458
459 setup_initrd(params, initrd_load_addr, initrd_len);
460 }
461
462 setup_cmdline(image, params, bootparam_load_addr,
463 sizeof(struct boot_params), cmdline, cmdline_len);
464
465 /* bootloader info. Do we need a separate ID for kexec kernel loader? */
466 params->hdr.type_of_loader = 0x0D << 4;
467 params->hdr.loadflags = 0;
468
469 /* Setup purgatory regs for entry */
470 ret = kexec_purgatory_get_set_symbol(image, "entry64_regs", &regs64,
471 sizeof(regs64), 1);
472 if (ret)
473 goto out_free_params;
474
475 regs64.rbx = 0; /* Bootstrap Processor */
476 regs64.rsi = bootparam_load_addr;
477 regs64.rip = kernel_load_addr + 0x200;
478 stack = kexec_purgatory_get_symbol_addr(image, "stack_end");
479 if (IS_ERR(stack)) {
480 pr_err("Could not find address of symbol stack_end\n");
481 ret = -EINVAL;
482 goto out_free_params;
483 }
484
485 regs64.rsp = (unsigned long)stack;
486 ret = kexec_purgatory_get_set_symbol(image, "entry64_regs", &regs64,
487 sizeof(regs64), 0);
488 if (ret)
489 goto out_free_params;
490
491 ret = setup_boot_parameters(image, params, bootparam_load_addr,
492 efi_map_offset, efi_map_sz,
493 efi_setup_data_offset);
494 if (ret)
495 goto out_free_params;
496
497 /* Allocate loader specific data */
498 ldata = kzalloc(sizeof(struct bzimage64_data), GFP_KERNEL);
499 if (!ldata) {
500 ret = -ENOMEM;
501 goto out_free_params;
502 }
503
504 /*
505 * Store pointer to params so that it could be freed after loading
506 * params segment has been loaded and contents have been copied
507 * somewhere else.
508 */
509 ldata->bootparams_buf = params;
510 return ldata;
511
512 out_free_params:
513 kfree(params);
514 return ERR_PTR(ret);
515 }
516
517 /* This cleanup function is called after various segments have been loaded */
518 static int bzImage64_cleanup(void *loader_data)
519 {
520 struct bzimage64_data *ldata = loader_data;
521
522 if (!ldata)
523 return 0;
524
525 kfree(ldata->bootparams_buf);
526 ldata->bootparams_buf = NULL;
527
528 return 0;
529 }
530
531 #ifdef CONFIG_KEXEC_BZIMAGE_VERIFY_SIG
532 static int bzImage64_verify_sig(const char *kernel, unsigned long kernel_len)
533 {
534 int ret;
535
536 ret = verify_pefile_signature(kernel, kernel_len,
537 VERIFY_USE_SECONDARY_KEYRING,
538 VERIFYING_KEXEC_PE_SIGNATURE);
539 if (ret == -ENOKEY && IS_ENABLED(CONFIG_INTEGRITY_PLATFORM_KEYRING)) {
540 ret = verify_pefile_signature(kernel, kernel_len,
541 VERIFY_USE_PLATFORM_KEYRING,
542 VERIFYING_KEXEC_PE_SIGNATURE);
543 }
544 return ret;
545 }
546 #endif
547
548 const struct kexec_file_ops kexec_bzImage64_ops = {
549 .probe = bzImage64_probe,
550 .load = bzImage64_load,
551 .cleanup = bzImage64_cleanup,
552 #ifdef CONFIG_KEXEC_BZIMAGE_VERIFY_SIG
553 .verify_sig = bzImage64_verify_sig,
554 #endif
555 };