1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * crash.c - kernel crash support code.
4 * Copyright (C) 2002-2004 Eric Biederman <ebiederm@xmission.com>
5 */
6
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8
9 #include <linux/buildid.h>
10 #include <linux/init.h>
11 #include <linux/utsname.h>
12 #include <linux/vmalloc.h>
13 #include <linux/sizes.h>
14 #include <linux/kexec.h>
15 #include <linux/memory.h>
16 #include <linux/mm.h>
17 #include <linux/cpuhotplug.h>
18 #include <linux/memblock.h>
19 #include <linux/kmemleak.h>
20 #include <linux/crash_core.h>
21 #include <linux/reboot.h>
22 #include <linux/btf.h>
23 #include <linux/objtool.h>
24
25 #include <asm/page.h>
26 #include <asm/sections.h>
27
28 #include <crypto/sha1.h>
29
30 #include "kallsyms_internal.h"
31 #include "kexec_internal.h"
32
33 /* Per cpu memory for storing cpu states in case of system crash. */
34 note_buf_t __percpu *crash_notes;
35
36 #ifdef CONFIG_CRASH_DUMP
37
38 int kimage_crash_copy_vmcoreinfo(struct kimage *image)
39 {
40 struct page *vmcoreinfo_page;
41 void *safecopy;
42
43 if (!IS_ENABLED(CONFIG_CRASH_DUMP))
44 return 0;
45 if (image->type != KEXEC_TYPE_CRASH)
46 return 0;
47
48 /*
49 * For kdump, allocate one vmcoreinfo safe copy from the
50 * crash memory. as we have arch_kexec_protect_crashkres()
51 * after kexec syscall, we naturally protect it from write
52 * (even read) access under kernel direct mapping. But on
53 * the other hand, we still need to operate it when crash
54 * happens to generate vmcoreinfo note, hereby we rely on
55 * vmap for this purpose.
56 */
57 vmcoreinfo_page = kimage_alloc_control_pages(image, 0);
58 if (!vmcoreinfo_page) {
59 pr_warn("Could not allocate vmcoreinfo buffer\n");
60 return -ENOMEM;
61 }
62 safecopy = vmap(&vmcoreinfo_page, 1, VM_MAP, PAGE_KERNEL);
63 if (!safecopy) {
64 pr_warn("Could not vmap vmcoreinfo buffer\n");
65 return -ENOMEM;
66 }
67
68 image->vmcoreinfo_data_copy = safecopy;
69 crash_update_vmcoreinfo_safecopy(safecopy);
70
71 return 0;
72 }
73
74
75
76 int kexec_should_crash(struct task_struct *p)
77 {
78 /*
79 * If crash_kexec_post_notifiers is enabled, don't run
80 * crash_kexec() here yet, which must be run after panic
81 * notifiers in panic().
82 */
83 if (crash_kexec_post_notifiers)
84 return 0;
85 /*
86 * There are 4 panic() calls in make_task_dead() path, each of which
87 * corresponds to each of these 4 conditions.
88 */
89 if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops)
90 return 1;
91 return 0;
92 }
93
94 int kexec_crash_loaded(void)
95 {
96 return !!kexec_crash_image;
97 }
98 EXPORT_SYMBOL_GPL(kexec_crash_loaded);
99
100 /*
101 * No panic_cpu check version of crash_kexec(). This function is called
102 * only when panic_cpu holds the current CPU number; this is the only CPU
103 * which processes crash_kexec routines.
104 */
105 void __noclone __crash_kexec(struct pt_regs *regs)
106 {
107 /* Take the kexec_lock here to prevent sys_kexec_load
108 * running on one cpu from replacing the crash kernel
109 * we are using after a panic on a different cpu.
110 *
111 * If the crash kernel was not located in a fixed area
112 * of memory the xchg(&kexec_crash_image) would be
113 * sufficient. But since I reuse the memory...
114 */
115 if (kexec_trylock()) {
116 if (kexec_crash_image) {
117 struct pt_regs fixed_regs;
118
119 crash_setup_regs(&fixed_regs, regs);
120 crash_save_vmcoreinfo();
121 machine_crash_shutdown(&fixed_regs);
122 machine_kexec(kexec_crash_image);
123 }
124 kexec_unlock();
125 }
126 }
127 STACK_FRAME_NON_STANDARD(__crash_kexec);
128
129 __bpf_kfunc void crash_kexec(struct pt_regs *regs)
130 {
131 int old_cpu, this_cpu;
132
133 /*
134 * Only one CPU is allowed to execute the crash_kexec() code as with
135 * panic(). Otherwise parallel calls of panic() and crash_kexec()
136 * may stop each other. To exclude them, we use panic_cpu here too.
137 */
138 old_cpu = PANIC_CPU_INVALID;
139 this_cpu = raw_smp_processor_id();
140
141 if (atomic_try_cmpxchg(&panic_cpu, &old_cpu, this_cpu)) {
142 /* This is the 1st CPU which comes here, so go ahead. */
143 __crash_kexec(regs);
144
145 /*
146 * Reset panic_cpu to allow another panic()/crash_kexec()
147 * call.
148 */
149 atomic_set(&panic_cpu, PANIC_CPU_INVALID);
150 }
151 }
152
153 static inline resource_size_t crash_resource_size(const struct resource *res)
154 {
155 return !res->end ? 0 : resource_size(res);
156 }
157
158
159
160
161 int crash_prepare_elf64_headers(struct crash_mem *mem, int need_kernel_map,
162 void **addr, unsigned long *sz)
163 {
164 Elf64_Ehdr *ehdr;
165 Elf64_Phdr *phdr;
166 unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz;
167 unsigned char *buf;
168 unsigned int cpu, i;
169 unsigned long long notes_addr;
170 unsigned long mstart, mend;
171
172 /* extra phdr for vmcoreinfo ELF note */
173 nr_phdr = nr_cpus + 1;
174 nr_phdr += mem->nr_ranges;
175
176 /*
177 * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping
178 * area (for example, ffffffff80000000 - ffffffffa0000000 on x86_64).
179 * I think this is required by tools like gdb. So same physical
180 * memory will be mapped in two ELF headers. One will contain kernel
181 * text virtual addresses and other will have __va(physical) addresses.
182 */
183
184 nr_phdr++;
185 elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr);
186 elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN);
187
188 buf = vzalloc(elf_sz);
189 if (!buf)
190 return -ENOMEM;
191
192 ehdr = (Elf64_Ehdr *)buf;
193 phdr = (Elf64_Phdr *)(ehdr + 1);
194 memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
195 ehdr->e_ident[EI_CLASS] = ELFCLASS64;
196 ehdr->e_ident[EI_DATA] = ELFDATA2LSB;
197 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
198 ehdr->e_ident[EI_OSABI] = ELF_OSABI;
199 memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
200 ehdr->e_type = ET_CORE;
201 ehdr->e_machine = ELF_ARCH;
202 ehdr->e_version = EV_CURRENT;
203 ehdr->e_phoff = sizeof(Elf64_Ehdr);
204 ehdr->e_ehsize = sizeof(Elf64_Ehdr);
205 ehdr->e_phentsize = sizeof(Elf64_Phdr);
206
207 /* Prepare one phdr of type PT_NOTE for each possible CPU */
208 for_each_possible_cpu(cpu) {
209 phdr->p_type = PT_NOTE;
210 notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu));
211 phdr->p_offset = phdr->p_paddr = notes_addr;
212 phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t);
213 (ehdr->e_phnum)++;
214 phdr++;
215 }
216
217 /* Prepare one PT_NOTE header for vmcoreinfo */
218 phdr->p_type = PT_NOTE;
219 phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
220 phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE;
221 (ehdr->e_phnum)++;
222 phdr++;
223
224 /* Prepare PT_LOAD type program header for kernel text region */
225 if (need_kernel_map) {
226 phdr->p_type = PT_LOAD;
227 phdr->p_flags = PF_R|PF_W|PF_X;
228 phdr->p_vaddr = (unsigned long) _text;
229 phdr->p_filesz = phdr->p_memsz = _end - _text;
230 phdr->p_offset = phdr->p_paddr = __pa_symbol(_text);
231 ehdr->e_phnum++;
232 phdr++;
233 }
234
235 /* Go through all the ranges in mem->ranges[] and prepare phdr */
236 for (i = 0; i < mem->nr_ranges; i++) {
237 mstart = mem->ranges[i].start;
238 mend = mem->ranges[i].end;
239
240 phdr->p_type = PT_LOAD;
241 phdr->p_flags = PF_R|PF_W|PF_X;
242 phdr->p_offset = mstart;
243
244 phdr->p_paddr = mstart;
245 phdr->p_vaddr = (unsigned long) __va(mstart);
246 phdr->p_filesz = phdr->p_memsz = mend - mstart + 1;
247 phdr->p_align = 0;
248 ehdr->e_phnum++;
249 #ifdef CONFIG_KEXEC_FILE
250 kexec_dprintk("Crash PT_LOAD ELF header. phdr=%p vaddr=0x%llx, paddr=0x%llx, sz=0x%llx e_phnum=%d p_offset=0x%llx\n",
251 phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz,
252 ehdr->e_phnum, phdr->p_offset);
253 #endif
254 phdr++;
255 }
256
257 *addr = buf;
258 *sz = elf_sz;
259 return 0;
260 }
261
262 int crash_exclude_mem_range(struct crash_mem *mem,
263 unsigned long long mstart, unsigned long long mend)
264 {
265 int i;
266 unsigned long long start, end, p_start, p_end;
267
268 for (i = 0; i < mem->nr_ranges; i++) {
269 start = mem->ranges[i].start;
270 end = mem->ranges[i].end;
271 p_start = mstart;
272 p_end = mend;
273
274 if (p_start > end)
275 continue;
276
277 /*
278 * Because the memory ranges in mem->ranges are stored in
279 * ascending order, when we detect `p_end < start`, we can
280 * immediately exit the for loop, as the subsequent memory
281 * ranges will definitely be outside the range we are looking
282 * for.
283 */
284 if (p_end < start)
285 break;
286
287 /* Truncate any area outside of range */
288 if (p_start < start)
289 p_start = start;
290 if (p_end > end)
291 p_end = end;
292
293 /* Found completely overlapping range */
294 if (p_start == start && p_end == end) {
295 memmove(&mem->ranges[i], &mem->ranges[i + 1],
296 (mem->nr_ranges - (i + 1)) * sizeof(mem->ranges[i]));
297 i--;
298 mem->nr_ranges--;
299 } else if (p_start > start && p_end < end) {
300 /* Split original range */
301 if (mem->nr_ranges >= mem->max_nr_ranges)
302 return -ENOMEM;
303
304 memmove(&mem->ranges[i + 2], &mem->ranges[i + 1],
305 (mem->nr_ranges - (i + 1)) * sizeof(mem->ranges[i]));
306
307 mem->ranges[i].end = p_start - 1;
308 mem->ranges[i + 1].start = p_end + 1;
309 mem->ranges[i + 1].end = end;
310
311 i++;
312 mem->nr_ranges++;
313 } else if (p_start != start)
314 mem->ranges[i].end = p_start - 1;
315 else
316 mem->ranges[i].start = p_end + 1;
317 }
318
319 return 0;
320 }
321
322 ssize_t crash_get_memory_size(void)
323 {
324 ssize_t size = 0;
325
326 if (!kexec_trylock())
327 return -EBUSY;
328
329 size += crash_resource_size(&crashk_res);
330 size += crash_resource_size(&crashk_low_res);
331
332 kexec_unlock();
333 return size;
334 }
335
336 static int __crash_shrink_memory(struct resource *old_res,
337 unsigned long new_size)
338 {
339 struct resource *ram_res;
340
341 ram_res = kzalloc(sizeof(*ram_res), GFP_KERNEL);
342 if (!ram_res)
343 return -ENOMEM;
344
345 ram_res->start = old_res->start + new_size;
346 ram_res->end = old_res->end;
347 ram_res->flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM;
348 ram_res->name = "System RAM";
349
350 if (!new_size) {
351 release_resource(old_res);
352 old_res->start = 0;
353 old_res->end = 0;
354 } else {
355 crashk_res.end = ram_res->start - 1;
356 }
357
358 crash_free_reserved_phys_range(ram_res->start, ram_res->end);
359 insert_resource(&iomem_resource, ram_res);
360
361 return 0;
362 }
363
364 int crash_shrink_memory(unsigned long new_size)
365 {
366 int ret = 0;
367 unsigned long old_size, low_size;
368
369 if (!kexec_trylock())
370 return -EBUSY;
371
372 if (kexec_crash_image) {
373 ret = -ENOENT;
374 goto unlock;
375 }
376
377 low_size = crash_resource_size(&crashk_low_res);
378 old_size = crash_resource_size(&crashk_res) + low_size;
379 new_size = roundup(new_size, KEXEC_CRASH_MEM_ALIGN);
380 if (new_size >= old_size) {
381 ret = (new_size == old_size) ? 0 : -EINVAL;
382 goto unlock;
383 }
384
385 /*
386 * (low_size > new_size) implies that low_size is greater than zero.
387 * This also means that if low_size is zero, the else branch is taken.
388 *
389 * If low_size is greater than 0, (low_size > new_size) indicates that
390 * crashk_low_res also needs to be shrunken. Otherwise, only crashk_res
391 * needs to be shrunken.
392 */
393 if (low_size > new_size) {
394 ret = __crash_shrink_memory(&crashk_res, 0);
395 if (ret)
396 goto unlock;
397
398 ret = __crash_shrink_memory(&crashk_low_res, new_size);
399 } else {
400 ret = __crash_shrink_memory(&crashk_res, new_size - low_size);
401 }
402
403 /* Swap crashk_res and crashk_low_res if needed */
404 if (!crashk_res.end && crashk_low_res.end) {
405 crashk_res.start = crashk_low_res.start;
406 crashk_res.end = crashk_low_res.end;
407 release_resource(&crashk_low_res);
408 crashk_low_res.start = 0;
409 crashk_low_res.end = 0;
410 insert_resource(&iomem_resource, &crashk_res);
411 }
412
413 unlock:
414 kexec_unlock();
415 return ret;
416 }
417
418 void crash_save_cpu(struct pt_regs *regs, int cpu)
419 {
420 struct elf_prstatus prstatus;
421 u32 *buf;
422
423 if ((cpu < 0) || (cpu >= nr_cpu_ids))
424 return;
425
426 /* Using ELF notes here is opportunistic.
427 * I need a well defined structure format
428 * for the data I pass, and I need tags
429 * on the data to indicate what information I have
430 * squirrelled away. ELF notes happen to provide
431 * all of that, so there is no need to invent something new.
432 */
433 buf = (u32 *)per_cpu_ptr(crash_notes, cpu);
434 if (!buf)
435 return;
436 memset(&prstatus, 0, sizeof(prstatus));
437 prstatus.common.pr_pid = current->pid;
438 elf_core_copy_regs(&prstatus.pr_reg, regs);
439 buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
440 &prstatus, sizeof(prstatus));
441 final_note(buf);
442 }
443
444
445
446 static int __init crash_notes_memory_init(void)
447 {
448 /* Allocate memory for saving cpu registers. */
449 size_t size, align;
450
451 /*
452 * crash_notes could be allocated across 2 vmalloc pages when percpu
453 * is vmalloc based . vmalloc doesn't guarantee 2 continuous vmalloc
454 * pages are also on 2 continuous physical pages. In this case the
455 * 2nd part of crash_notes in 2nd page could be lost since only the
456 * starting address and size of crash_notes are exported through sysfs.
457 * Here round up the size of crash_notes to the nearest power of two
458 * and pass it to __alloc_percpu as align value. This can make sure
459 * crash_notes is allocated inside one physical page.
460 */
461 size = sizeof(note_buf_t);
462 align = min(roundup_pow_of_two(sizeof(note_buf_t)), PAGE_SIZE);
463
464 /*
465 * Break compile if size is bigger than PAGE_SIZE since crash_notes
466 * definitely will be in 2 pages with that.
467 */
468 BUILD_BUG_ON(size > PAGE_SIZE);
469
470 crash_notes = __alloc_percpu(size, align);
471 if (!crash_notes) {
472 pr_warn("Memory allocation for saving cpu register states failed\n");
473 return -ENOMEM;
474 }
475 return 0;
476 }
477 subsys_initcall(crash_notes_memory_init);
478
479 #endif /*CONFIG_CRASH_DUMP*/
480
481 #ifdef CONFIG_CRASH_HOTPLUG
482 #undef pr_fmt
483 #define pr_fmt(fmt) "crash hp: " fmt
484
485 /*
486 * Different than kexec/kdump loading/unloading/jumping/shrinking which
487 * usually rarely happen, there will be many crash hotplug events notified
488 * during one short period, e.g one memory board is hot added and memory
489 * regions are online. So mutex lock __crash_hotplug_lock is used to
490 * serialize the crash hotplug handling specifically.
491 */
492 static DEFINE_MUTEX(__crash_hotplug_lock);
493 #define crash_hotplug_lock() mutex_lock(&__crash_hotplug_lock)
494 #define crash_hotplug_unlock() mutex_unlock(&__crash_hotplug_lock)
495
496 /*
497 * This routine utilized when the crash_hotplug sysfs node is read.
498 * It reflects the kernel's ability/permission to update the kdump
499 * image directly.
500 */
501 int crash_check_hotplug_support(void)
502 {
503 int rc = 0;
504
505 crash_hotplug_lock();
506 /* Obtain lock while reading crash information */
507 if (!kexec_trylock()) {
508 pr_info("kexec_trylock() failed, elfcorehdr may be inaccurate\n");
509 crash_hotplug_unlock();
510 return 0;
511 }
512 if (kexec_crash_image) {
513 rc = kexec_crash_image->hotplug_support;
514 }
515 /* Release lock now that update complete */
516 kexec_unlock();
517 crash_hotplug_unlock();
518
519 return rc;
520 }
521
522 /*
523 * To accurately reflect hot un/plug changes of cpu and memory resources
524 * (including onling and offlining of those resources), the elfcorehdr
525 * (which is passed to the crash kernel via the elfcorehdr= parameter)
526 * must be updated with the new list of CPUs and memories.
527 *
528 * In order to make changes to elfcorehdr, two conditions are needed:
529 * First, the segment containing the elfcorehdr must be large enough
530 * to permit a growing number of resources; the elfcorehdr memory size
531 * is based on NR_CPUS_DEFAULT and CRASH_MAX_MEMORY_RANGES.
532 * Second, purgatory must explicitly exclude the elfcorehdr from the
533 * list of segments it checks (since the elfcorehdr changes and thus
534 * would require an update to purgatory itself to update the digest).
535 */
536 static void crash_handle_hotplug_event(unsigned int hp_action, unsigned int cpu, void *arg)
537 {
538 struct kimage *image;
539
540 crash_hotplug_lock();
541 /* Obtain lock while changing crash information */
542 if (!kexec_trylock()) {
543 pr_info("kexec_trylock() failed, elfcorehdr may be inaccurate\n");
544 crash_hotplug_unlock();
545 return;
546 }
547
548 /* Check kdump is not loaded */
549 if (!kexec_crash_image)
550 goto out;
551
552 image = kexec_crash_image;
553
554 /* Check that kexec segments update is permitted */
555 if (!image->hotplug_support)
556 goto out;
557
558 if (hp_action == KEXEC_CRASH_HP_ADD_CPU ||
559 hp_action == KEXEC_CRASH_HP_REMOVE_CPU)
560 pr_debug("hp_action %u, cpu %u\n", hp_action, cpu);
561 else
562 pr_debug("hp_action %u\n", hp_action);
563
564 /*
565 * The elfcorehdr_index is set to -1 when the struct kimage
566 * is allocated. Find the segment containing the elfcorehdr,
567 * if not already found.
568 */
569 if (image->elfcorehdr_index < 0) {
570 unsigned long mem;
571 unsigned char *ptr;
572 unsigned int n;
573
574 for (n = 0; n < image->nr_segments; n++) {
575 mem = image->segment[n].mem;
576 ptr = kmap_local_page(pfn_to_page(mem >> PAGE_SHIFT));
577 if (ptr) {
578 /* The segment containing elfcorehdr */
579 if (memcmp(ptr, ELFMAG, SELFMAG) == 0)
580 image->elfcorehdr_index = (int)n;
581 kunmap_local(ptr);
582 }
583 }
584 }
585
586 if (image->elfcorehdr_index < 0) {
587 pr_err("unable to locate elfcorehdr segment");
588 goto out;
589 }
590
591 /* Needed in order for the segments to be updated */
592 arch_kexec_unprotect_crashkres();
593
594 /* Differentiate between normal load and hotplug update */
595 image->hp_action = hp_action;
596
597 /* Now invoke arch-specific update handler */
598 arch_crash_handle_hotplug_event(image, arg);
599
600 /* No longer handling a hotplug event */
601 image->hp_action = KEXEC_CRASH_HP_NONE;
602 image->elfcorehdr_updated = true;
603
604 /* Change back to read-only */
605 arch_kexec_protect_crashkres();
606
607 /* Errors in the callback is not a reason to rollback state */
608 out:
609 /* Release lock now that update complete */
610 kexec_unlock();
611 crash_hotplug_unlock();
612 }
613
614 static int crash_memhp_notifier(struct notifier_block *nb, unsigned long val, void *arg)
615 {
616 switch (val) {
617 case MEM_ONLINE:
618 crash_handle_hotplug_event(KEXEC_CRASH_HP_ADD_MEMORY,
619 KEXEC_CRASH_HP_INVALID_CPU, arg);
620 break;
621
622 case MEM_OFFLINE:
623 crash_handle_hotplug_event(KEXEC_CRASH_HP_REMOVE_MEMORY,
624 KEXEC_CRASH_HP_INVALID_CPU, arg);
625 break;
626 }
627 return NOTIFY_OK;
628 }
629
630 static struct notifier_block crash_memhp_nb = {
631 .notifier_call = crash_memhp_notifier,
632 .priority = 0
633 };
634
635 static int crash_cpuhp_online(unsigned int cpu)
636 {
637 crash_handle_hotplug_event(KEXEC_CRASH_HP_ADD_CPU, cpu, NULL);
638 return 0;
639 }
640
641 static int crash_cpuhp_offline(unsigned int cpu)
642 {
643 crash_handle_hotplug_event(KEXEC_CRASH_HP_REMOVE_CPU, cpu, NULL);
644 return 0;
645 }
646
647 static int __init crash_hotplug_init(void)
648 {
649 int result = 0;
650
651 if (IS_ENABLED(CONFIG_MEMORY_HOTPLUG))
652 register_memory_notifier(&crash_memhp_nb);
653
654 if (IS_ENABLED(CONFIG_HOTPLUG_CPU)) {
655 result = cpuhp_setup_state_nocalls(CPUHP_BP_PREPARE_DYN,
656 "crash/cpuhp", crash_cpuhp_online, crash_cpuhp_offline);
657 }
658
659 return result;
660 }
661
662 subsys_initcall(crash_hotplug_init);
663 #endif
664
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.