Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * fs/proc/kcore.c kernel ELF core dumper
4 *
5 * Modelled on fs/exec.c:aout_core_dump()
6 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
7 * ELF version written by David Howells <David.Howells@nexor.co.uk>
8 * Modified and incorporated into 2.3.x by Tigran Aivazian <tigran@veritas.com>
9 * Support to dump vmalloc'd areas (ELF only), Tigran Aivazian <tigran@veritas.com>
10 * Safe accesses to vmalloc/direct-mapped discontiguous areas, Kanoj Sarcar <kanoj@sgi.com>
11 */
12
13#include <linux/vmcore_info.h>
14#include <linux/mm.h>
15#include <linux/proc_fs.h>
16#include <linux/kcore.h>
17#include <linux/user.h>
18#include <linux/capability.h>
19#include <linux/elf.h>
20#include <linux/elfcore.h>
21#include <linux/vmalloc.h>
22#include <linux/highmem.h>
23#include <linux/printk.h>
24#include <linux/memblock.h>
25#include <linux/init.h>
26#include <linux/slab.h>
27#include <linux/uio.h>
28#include <asm/io.h>
29#include <linux/list.h>
30#include <linux/ioport.h>
31#include <linux/memory.h>
32#include <linux/sched/task.h>
33#include <linux/security.h>
34#include <asm/sections.h>
35#include "internal.h"
36
37#define CORE_STR "CORE"
38
39#ifndef ELF_CORE_EFLAGS
40#define ELF_CORE_EFLAGS 0
41#endif
42
43static struct proc_dir_entry *proc_root_kcore;
44
45
46#ifndef kc_vaddr_to_offset
47#define kc_vaddr_to_offset(v) ((v) - PAGE_OFFSET)
48#endif
49#ifndef kc_offset_to_vaddr
50#define kc_offset_to_vaddr(o) ((o) + PAGE_OFFSET)
51#endif
52
53#ifndef kc_xlate_dev_mem_ptr
54#define kc_xlate_dev_mem_ptr kc_xlate_dev_mem_ptr
55static inline void *kc_xlate_dev_mem_ptr(phys_addr_t phys)
56{
57 return __va(phys);
58}
59#endif
60#ifndef kc_unxlate_dev_mem_ptr
61#define kc_unxlate_dev_mem_ptr kc_unxlate_dev_mem_ptr
62static inline void kc_unxlate_dev_mem_ptr(phys_addr_t phys, void *virt)
63{
64}
65#endif
66
67static LIST_HEAD(kclist_head);
68static DECLARE_RWSEM(kclist_lock);
69static int kcore_need_update = 1;
70
71/*
72 * Returns > 0 for RAM pages, 0 for non-RAM pages, < 0 on error
73 * Same as oldmem_pfn_is_ram in vmcore
74 */
75static int (*mem_pfn_is_ram)(unsigned long pfn);
76
77int __init register_mem_pfn_is_ram(int (*fn)(unsigned long pfn))
78{
79 if (mem_pfn_is_ram)
80 return -EBUSY;
81 mem_pfn_is_ram = fn;
82 return 0;
83}
84
85static int pfn_is_ram(unsigned long pfn)
86{
87 if (mem_pfn_is_ram)
88 return mem_pfn_is_ram(pfn);
89 else
90 return 1;
91}
92
93/* This doesn't grab kclist_lock, so it should only be used at init time. */
94void __init kclist_add(struct kcore_list *new, void *addr, size_t size,
95 int type)
96{
97 new->addr = (unsigned long)addr;
98 new->size = size;
99 new->type = type;
100
101 list_add_tail(&new->list, &kclist_head);
102}
103
104static size_t get_kcore_size(int *nphdr, size_t *phdrs_len, size_t *notes_len,
105 size_t *data_offset)
106{
107 size_t try, size;
108 struct kcore_list *m;
109
110 *nphdr = 1; /* PT_NOTE */
111 size = 0;
112
113 list_for_each_entry(m, &kclist_head, list) {
114 try = kc_vaddr_to_offset((size_t)m->addr + m->size);
115 if (try > size)
116 size = try;
117 *nphdr = *nphdr + 1;
118 }
119
120 *phdrs_len = *nphdr * sizeof(struct elf_phdr);
121 *notes_len = (4 * sizeof(struct elf_note) +
122 3 * ALIGN(sizeof(CORE_STR), 4) +
123 VMCOREINFO_NOTE_NAME_BYTES +
124 ALIGN(sizeof(struct elf_prstatus), 4) +
125 ALIGN(sizeof(struct elf_prpsinfo), 4) +
126 ALIGN(arch_task_struct_size, 4) +
127 ALIGN(vmcoreinfo_size, 4));
128 *data_offset = PAGE_ALIGN(sizeof(struct elfhdr) + *phdrs_len +
129 *notes_len);
130 return *data_offset + size;
131}
132
133#ifdef CONFIG_HIGHMEM
134/*
135 * If no highmem, we can assume [0...max_low_pfn) continuous range of memory
136 * because memory hole is not as big as !HIGHMEM case.
137 * (HIGHMEM is special because part of memory is _invisible_ from the kernel.)
138 */
139static int kcore_ram_list(struct list_head *head)
140{
141 struct kcore_list *ent;
142
143 ent = kmalloc(sizeof(*ent), GFP_KERNEL);
144 if (!ent)
145 return -ENOMEM;
146 ent->addr = (unsigned long)__va(0);
147 ent->size = max_low_pfn << PAGE_SHIFT;
148 ent->type = KCORE_RAM;
149 list_add(&ent->list, head);
150 return 0;
151}
152
153#else /* !CONFIG_HIGHMEM */
154
155#ifdef CONFIG_SPARSEMEM_VMEMMAP
156/* calculate vmemmap's address from given system ram pfn and register it */
157static int
158get_sparsemem_vmemmap_info(struct kcore_list *ent, struct list_head *head)
159{
160 unsigned long pfn = __pa(ent->addr) >> PAGE_SHIFT;
161 unsigned long nr_pages = ent->size >> PAGE_SHIFT;
162 unsigned long start, end;
163 struct kcore_list *vmm, *tmp;
164
165
166 start = ((unsigned long)pfn_to_page(pfn)) & PAGE_MASK;
167 end = ((unsigned long)pfn_to_page(pfn + nr_pages)) - 1;
168 end = PAGE_ALIGN(end);
169 /* overlap check (because we have to align page */
170 list_for_each_entry(tmp, head, list) {
171 if (tmp->type != KCORE_VMEMMAP)
172 continue;
173 if (start < tmp->addr + tmp->size)
174 if (end > tmp->addr)
175 end = tmp->addr;
176 }
177 if (start < end) {
178 vmm = kmalloc(sizeof(*vmm), GFP_KERNEL);
179 if (!vmm)
180 return 0;
181 vmm->addr = start;
182 vmm->size = end - start;
183 vmm->type = KCORE_VMEMMAP;
184 list_add_tail(&vmm->list, head);
185 }
186 return 1;
187
188}
189#else
190static int
191get_sparsemem_vmemmap_info(struct kcore_list *ent, struct list_head *head)
192{
193 return 1;
194}
195
196#endif
197
198static int
199kclist_add_private(unsigned long pfn, unsigned long nr_pages, void *arg)
200{
201 struct list_head *head = (struct list_head *)arg;
202 struct kcore_list *ent;
203 struct page *p;
204
205 if (!pfn_valid(pfn))
206 return 1;
207
208 p = pfn_to_page(pfn);
209
210 ent = kmalloc(sizeof(*ent), GFP_KERNEL);
211 if (!ent)
212 return -ENOMEM;
213 ent->addr = (unsigned long)page_to_virt(p);
214 ent->size = nr_pages << PAGE_SHIFT;
215
216 if (!virt_addr_valid((void *)ent->addr))
217 goto free_out;
218
219 /* cut not-mapped area. ....from ppc-32 code. */
220 if (ULONG_MAX - ent->addr < ent->size)
221 ent->size = ULONG_MAX - ent->addr;
222
223 /*
224 * We've already checked virt_addr_valid so we know this address
225 * is a valid pointer, therefore we can check against it to determine
226 * if we need to trim
227 */
228 if (VMALLOC_START > ent->addr) {
229 if (VMALLOC_START - ent->addr < ent->size)
230 ent->size = VMALLOC_START - ent->addr;
231 }
232
233 ent->type = KCORE_RAM;
234 list_add_tail(&ent->list, head);
235
236 if (!get_sparsemem_vmemmap_info(ent, head)) {
237 list_del(&ent->list);
238 goto free_out;
239 }
240
241 return 0;
242free_out:
243 kfree(ent);
244 return 1;
245}
246
247static int kcore_ram_list(struct list_head *list)
248{
249 int nid, ret;
250 unsigned long end_pfn;
251
252 /* Not initialized....update now */
253 /* find out "max pfn" */
254 end_pfn = 0;
255 for_each_node_state(nid, N_MEMORY) {
256 unsigned long node_end;
257 node_end = node_end_pfn(nid);
258 if (end_pfn < node_end)
259 end_pfn = node_end;
260 }
261 /* scan 0 to max_pfn */
262 ret = walk_system_ram_range(0, end_pfn, list, kclist_add_private);
263 if (ret)
264 return -ENOMEM;
265 return 0;
266}
267#endif /* CONFIG_HIGHMEM */
268
269static int kcore_update_ram(void)
270{
271 LIST_HEAD(list);
272 LIST_HEAD(garbage);
273 int nphdr;
274 size_t phdrs_len, notes_len, data_offset;
275 struct kcore_list *tmp, *pos;
276 int ret = 0;
277
278 down_write(&kclist_lock);
279 if (!xchg(&kcore_need_update, 0))
280 goto out;
281
282 ret = kcore_ram_list(&list);
283 if (ret) {
284 /* Couldn't get the RAM list, try again next time. */
285 WRITE_ONCE(kcore_need_update, 1);
286 list_splice_tail(&list, &garbage);
287 goto out;
288 }
289
290 list_for_each_entry_safe(pos, tmp, &kclist_head, list) {
291 if (pos->type == KCORE_RAM || pos->type == KCORE_VMEMMAP)
292 list_move(&pos->list, &garbage);
293 }
294 list_splice_tail(&list, &kclist_head);
295
296 proc_root_kcore->size = get_kcore_size(&nphdr, &phdrs_len, ¬es_len,
297 &data_offset);
298
299out:
300 up_write(&kclist_lock);
301 list_for_each_entry_safe(pos, tmp, &garbage, list) {
302 list_del(&pos->list);
303 kfree(pos);
304 }
305 return ret;
306}
307
308static void append_kcore_note(char *notes, size_t *i, const char *name,
309 unsigned int type, const void *desc,
310 size_t descsz)
311{
312 struct elf_note *note = (struct elf_note *)¬es[*i];
313
314 note->n_namesz = strlen(name) + 1;
315 note->n_descsz = descsz;
316 note->n_type = type;
317 *i += sizeof(*note);
318 memcpy(¬es[*i], name, note->n_namesz);
319 *i = ALIGN(*i + note->n_namesz, 4);
320 memcpy(¬es[*i], desc, descsz);
321 *i = ALIGN(*i + descsz, 4);
322}
323
324static ssize_t read_kcore_iter(struct kiocb *iocb, struct iov_iter *iter)
325{
326 struct file *file = iocb->ki_filp;
327 char *buf = file->private_data;
328 loff_t *fpos = &iocb->ki_pos;
329 size_t phdrs_offset, notes_offset, data_offset;
330 size_t page_offline_frozen = 1;
331 size_t phdrs_len, notes_len;
332 struct kcore_list *m;
333 size_t tsz;
334 int nphdr;
335 unsigned long start;
336 size_t buflen = iov_iter_count(iter);
337 size_t orig_buflen = buflen;
338 int ret = 0;
339
340 down_read(&kclist_lock);
341 /*
342 * Don't race against drivers that set PageOffline() and expect no
343 * further page access.
344 */
345 page_offline_freeze();
346
347 get_kcore_size(&nphdr, &phdrs_len, ¬es_len, &data_offset);
348 phdrs_offset = sizeof(struct elfhdr);
349 notes_offset = phdrs_offset + phdrs_len;
350
351 /* ELF file header. */
352 if (buflen && *fpos < sizeof(struct elfhdr)) {
353 struct elfhdr ehdr = {
354 .e_ident = {
355 [EI_MAG0] = ELFMAG0,
356 [EI_MAG1] = ELFMAG1,
357 [EI_MAG2] = ELFMAG2,
358 [EI_MAG3] = ELFMAG3,
359 [EI_CLASS] = ELF_CLASS,
360 [EI_DATA] = ELF_DATA,
361 [EI_VERSION] = EV_CURRENT,
362 [EI_OSABI] = ELF_OSABI,
363 },
364 .e_type = ET_CORE,
365 .e_machine = ELF_ARCH,
366 .e_version = EV_CURRENT,
367 .e_phoff = sizeof(struct elfhdr),
368 .e_flags = ELF_CORE_EFLAGS,
369 .e_ehsize = sizeof(struct elfhdr),
370 .e_phentsize = sizeof(struct elf_phdr),
371 .e_phnum = nphdr,
372 };
373
374 tsz = min_t(size_t, buflen, sizeof(struct elfhdr) - *fpos);
375 if (copy_to_iter((char *)&ehdr + *fpos, tsz, iter) != tsz) {
376 ret = -EFAULT;
377 goto out;
378 }
379
380 buflen -= tsz;
381 *fpos += tsz;
382 }
383
384 /* ELF program headers. */
385 if (buflen && *fpos < phdrs_offset + phdrs_len) {
386 struct elf_phdr *phdrs, *phdr;
387
388 phdrs = kzalloc(phdrs_len, GFP_KERNEL);
389 if (!phdrs) {
390 ret = -ENOMEM;
391 goto out;
392 }
393
394 phdrs[0].p_type = PT_NOTE;
395 phdrs[0].p_offset = notes_offset;
396 phdrs[0].p_filesz = notes_len;
397
398 phdr = &phdrs[1];
399 list_for_each_entry(m, &kclist_head, list) {
400 phdr->p_type = PT_LOAD;
401 phdr->p_flags = PF_R | PF_W | PF_X;
402 phdr->p_offset = kc_vaddr_to_offset(m->addr) + data_offset;
403 phdr->p_vaddr = (size_t)m->addr;
404 if (m->type == KCORE_RAM)
405 phdr->p_paddr = __pa(m->addr);
406 else if (m->type == KCORE_TEXT)
407 phdr->p_paddr = __pa_symbol(m->addr);
408 else
409 phdr->p_paddr = (elf_addr_t)-1;
410 phdr->p_filesz = phdr->p_memsz = m->size;
411 phdr->p_align = PAGE_SIZE;
412 phdr++;
413 }
414
415 tsz = min_t(size_t, buflen, phdrs_offset + phdrs_len - *fpos);
416 if (copy_to_iter((char *)phdrs + *fpos - phdrs_offset, tsz,
417 iter) != tsz) {
418 kfree(phdrs);
419 ret = -EFAULT;
420 goto out;
421 }
422 kfree(phdrs);
423
424 buflen -= tsz;
425 *fpos += tsz;
426 }
427
428 /* ELF note segment. */
429 if (buflen && *fpos < notes_offset + notes_len) {
430 struct elf_prstatus prstatus = {};
431 struct elf_prpsinfo prpsinfo = {
432 .pr_sname = 'R',
433 .pr_fname = "vmlinux",
434 };
435 char *notes;
436 size_t i = 0;
437
438 strscpy(prpsinfo.pr_psargs, saved_command_line,
439 sizeof(prpsinfo.pr_psargs));
440
441 notes = kzalloc(notes_len, GFP_KERNEL);
442 if (!notes) {
443 ret = -ENOMEM;
444 goto out;
445 }
446
447 append_kcore_note(notes, &i, CORE_STR, NT_PRSTATUS, &prstatus,
448 sizeof(prstatus));
449 append_kcore_note(notes, &i, CORE_STR, NT_PRPSINFO, &prpsinfo,
450 sizeof(prpsinfo));
451 append_kcore_note(notes, &i, CORE_STR, NT_TASKSTRUCT, current,
452 arch_task_struct_size);
453 /*
454 * vmcoreinfo_size is mostly constant after init time, but it
455 * can be changed by crash_save_vmcoreinfo(). Racing here with a
456 * panic on another CPU before the machine goes down is insanely
457 * unlikely, but it's better to not leave potential buffer
458 * overflows lying around, regardless.
459 */
460 append_kcore_note(notes, &i, VMCOREINFO_NOTE_NAME, 0,
461 vmcoreinfo_data,
462 min(vmcoreinfo_size, notes_len - i));
463
464 tsz = min_t(size_t, buflen, notes_offset + notes_len - *fpos);
465 if (copy_to_iter(notes + *fpos - notes_offset, tsz, iter) != tsz) {
466 kfree(notes);
467 ret = -EFAULT;
468 goto out;
469 }
470 kfree(notes);
471
472 buflen -= tsz;
473 *fpos += tsz;
474 }
475
476 /*
477 * Check to see if our file offset matches with any of
478 * the addresses in the elf_phdr on our list.
479 */
480 start = kc_offset_to_vaddr(*fpos - data_offset);
481 if ((tsz = (PAGE_SIZE - (start & ~PAGE_MASK))) > buflen)
482 tsz = buflen;
483
484 m = NULL;
485 while (buflen) {
486 struct page *page;
487 unsigned long pfn;
488 phys_addr_t phys;
489 void *__start;
490
491 /*
492 * If this is the first iteration or the address is not within
493 * the previous entry, search for a matching entry.
494 */
495 if (!m || start < m->addr || start >= m->addr + m->size) {
496 struct kcore_list *pos;
497
498 m = NULL;
499 list_for_each_entry(pos, &kclist_head, list) {
500 if (start >= pos->addr &&
501 start < pos->addr + pos->size) {
502 m = pos;
503 break;
504 }
505 }
506 }
507
508 if (page_offline_frozen++ % MAX_ORDER_NR_PAGES == 0) {
509 page_offline_thaw();
510 cond_resched();
511 page_offline_freeze();
512 }
513
514 if (!m) {
515 if (iov_iter_zero(tsz, iter) != tsz) {
516 ret = -EFAULT;
517 goto out;
518 }
519 goto skip;
520 }
521
522 switch (m->type) {
523 case KCORE_VMALLOC:
524 {
525 const char *src = (char *)start;
526 size_t read = 0, left = tsz;
527
528 /*
529 * vmalloc uses spinlocks, so we optimistically try to
530 * read memory. If this fails, fault pages in and try
531 * again until we are done.
532 */
533 while (true) {
534 read += vread_iter(iter, src, left);
535 if (read == tsz)
536 break;
537
538 src += read;
539 left -= read;
540
541 if (fault_in_iov_iter_writeable(iter, left)) {
542 ret = -EFAULT;
543 goto out;
544 }
545 }
546 break;
547 }
548 case KCORE_USER:
549 /* User page is handled prior to normal kernel page: */
550 if (copy_to_iter((char *)start, tsz, iter) != tsz) {
551 ret = -EFAULT;
552 goto out;
553 }
554 break;
555 case KCORE_RAM:
556 phys = __pa(start);
557 pfn = phys >> PAGE_SHIFT;
558 page = pfn_to_online_page(pfn);
559
560 /*
561 * Don't read offline sections, logically offline pages
562 * (e.g., inflated in a balloon), hwpoisoned pages,
563 * and explicitly excluded physical ranges.
564 */
565 if (!page || PageOffline(page) ||
566 is_page_hwpoison(page) || !pfn_is_ram(pfn) ||
567 pfn_is_unaccepted_memory(pfn)) {
568 if (iov_iter_zero(tsz, iter) != tsz) {
569 ret = -EFAULT;
570 goto out;
571 }
572 break;
573 }
574 fallthrough;
575 case KCORE_VMEMMAP:
576 case KCORE_TEXT:
577 if (m->type == KCORE_RAM) {
578 __start = kc_xlate_dev_mem_ptr(phys);
579 if (!__start) {
580 ret = -ENOMEM;
581 if (iov_iter_zero(tsz, iter) != tsz)
582 ret = -EFAULT;
583 goto out;
584 }
585 } else {
586 __start = (void *)start;
587 }
588
589 /*
590 * Sadly we must use a bounce buffer here to be able to
591 * make use of copy_from_kernel_nofault(), as these
592 * memory regions might not always be mapped on all
593 * architectures.
594 */
595 ret = copy_from_kernel_nofault(buf, __start, tsz);
596 if (m->type == KCORE_RAM)
597 kc_unxlate_dev_mem_ptr(phys, __start);
598 if (ret) {
599 if (iov_iter_zero(tsz, iter) != tsz) {
600 ret = -EFAULT;
601 goto out;
602 }
603 ret = 0;
604 /*
605 * We know the bounce buffer is safe to copy from, so
606 * use _copy_to_iter() directly.
607 */
608 } else if (_copy_to_iter(buf, tsz, iter) != tsz) {
609 ret = -EFAULT;
610 goto out;
611 }
612 break;
613 default:
614 pr_warn_once("Unhandled KCORE type: %d\n", m->type);
615 if (iov_iter_zero(tsz, iter) != tsz) {
616 ret = -EFAULT;
617 goto out;
618 }
619 }
620skip:
621 buflen -= tsz;
622 *fpos += tsz;
623 start += tsz;
624 tsz = (buflen > PAGE_SIZE ? PAGE_SIZE : buflen);
625 }
626
627out:
628 page_offline_thaw();
629 up_read(&kclist_lock);
630 if (ret)
631 return ret;
632 return orig_buflen - buflen;
633}
634
635static int open_kcore(struct inode *inode, struct file *filp)
636{
637 int ret = security_locked_down(LOCKDOWN_KCORE);
638
639 if (!capable(CAP_SYS_RAWIO))
640 return -EPERM;
641
642 if (ret)
643 return ret;
644
645 filp->private_data = kmalloc(PAGE_SIZE, GFP_KERNEL);
646 if (!filp->private_data)
647 return -ENOMEM;
648
649 if (kcore_need_update)
650 kcore_update_ram();
651 if (i_size_read(inode) != proc_root_kcore->size) {
652 inode_lock(inode);
653 i_size_write(inode, proc_root_kcore->size);
654 inode_unlock(inode);
655 }
656 return 0;
657}
658
659static int release_kcore(struct inode *inode, struct file *file)
660{
661 kfree(file->private_data);
662 return 0;
663}
664
665static const struct proc_ops kcore_proc_ops = {
666 .proc_read_iter = read_kcore_iter,
667 .proc_open = open_kcore,
668 .proc_release = release_kcore,
669 .proc_lseek = default_llseek,
670};
671
672/* just remember that we have to update kcore */
673static int __meminit kcore_callback(struct notifier_block *self,
674 unsigned long action, void *arg)
675{
676 switch (action) {
677 case MEM_ONLINE:
678 case MEM_OFFLINE:
679 kcore_need_update = 1;
680 break;
681 }
682 return NOTIFY_OK;
683}
684
685
686static struct kcore_list kcore_vmalloc;
687
688#ifdef CONFIG_ARCH_PROC_KCORE_TEXT
689static struct kcore_list kcore_text;
690/*
691 * If defined, special segment is used for mapping kernel text instead of
692 * direct-map area. We need to create special TEXT section.
693 */
694static void __init proc_kcore_text_init(void)
695{
696 kclist_add(&kcore_text, _text, _end - _text, KCORE_TEXT);
697}
698#else
699static void __init proc_kcore_text_init(void)
700{
701}
702#endif
703
704#if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
705/*
706 * MODULES_VADDR has no intersection with VMALLOC_ADDR.
707 */
708static struct kcore_list kcore_modules;
709static void __init add_modules_range(void)
710{
711 if (MODULES_VADDR != VMALLOC_START && MODULES_END != VMALLOC_END) {
712 kclist_add(&kcore_modules, (void *)MODULES_VADDR,
713 MODULES_END - MODULES_VADDR, KCORE_VMALLOC);
714 }
715}
716#else
717static void __init add_modules_range(void)
718{
719}
720#endif
721
722static int __init proc_kcore_init(void)
723{
724 proc_root_kcore = proc_create("kcore", S_IRUSR, NULL, &kcore_proc_ops);
725 if (!proc_root_kcore) {
726 pr_err("couldn't create /proc/kcore\n");
727 return 0; /* Always returns 0. */
728 }
729 /* Store text area if it's special */
730 proc_kcore_text_init();
731 /* Store vmalloc area */
732 kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
733 VMALLOC_END - VMALLOC_START, KCORE_VMALLOC);
734 add_modules_range();
735 /* Store direct-map area from physical memory map */
736 kcore_update_ram();
737 hotplug_memory_notifier(kcore_callback, DEFAULT_CALLBACK_PRI);
738
739 return 0;
740}
741fs_initcall(proc_kcore_init);
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * fs/proc/kcore.c kernel ELF core dumper
4 *
5 * Modelled on fs/exec.c:aout_core_dump()
6 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
7 * ELF version written by David Howells <David.Howells@nexor.co.uk>
8 * Modified and incorporated into 2.3.x by Tigran Aivazian <tigran@veritas.com>
9 * Support to dump vmalloc'd areas (ELF only), Tigran Aivazian <tigran@veritas.com>
10 * Safe accesses to vmalloc/direct-mapped discontiguous areas, Kanoj Sarcar <kanoj@sgi.com>
11 */
12
13#include <linux/crash_core.h>
14#include <linux/mm.h>
15#include <linux/proc_fs.h>
16#include <linux/kcore.h>
17#include <linux/user.h>
18#include <linux/capability.h>
19#include <linux/elf.h>
20#include <linux/elfcore.h>
21#include <linux/notifier.h>
22#include <linux/vmalloc.h>
23#include <linux/highmem.h>
24#include <linux/printk.h>
25#include <linux/memblock.h>
26#include <linux/init.h>
27#include <linux/slab.h>
28#include <linux/uaccess.h>
29#include <asm/io.h>
30#include <linux/list.h>
31#include <linux/ioport.h>
32#include <linux/memory.h>
33#include <linux/sched/task.h>
34#include <linux/security.h>
35#include <asm/sections.h>
36#include "internal.h"
37
38#define CORE_STR "CORE"
39
40#ifndef ELF_CORE_EFLAGS
41#define ELF_CORE_EFLAGS 0
42#endif
43
44static struct proc_dir_entry *proc_root_kcore;
45
46
47#ifndef kc_vaddr_to_offset
48#define kc_vaddr_to_offset(v) ((v) - PAGE_OFFSET)
49#endif
50#ifndef kc_offset_to_vaddr
51#define kc_offset_to_vaddr(o) ((o) + PAGE_OFFSET)
52#endif
53
54static LIST_HEAD(kclist_head);
55static DECLARE_RWSEM(kclist_lock);
56static int kcore_need_update = 1;
57
58/*
59 * Returns > 0 for RAM pages, 0 for non-RAM pages, < 0 on error
60 * Same as oldmem_pfn_is_ram in vmcore
61 */
62static int (*mem_pfn_is_ram)(unsigned long pfn);
63
64int __init register_mem_pfn_is_ram(int (*fn)(unsigned long pfn))
65{
66 if (mem_pfn_is_ram)
67 return -EBUSY;
68 mem_pfn_is_ram = fn;
69 return 0;
70}
71
72static int pfn_is_ram(unsigned long pfn)
73{
74 if (mem_pfn_is_ram)
75 return mem_pfn_is_ram(pfn);
76 else
77 return 1;
78}
79
80/* This doesn't grab kclist_lock, so it should only be used at init time. */
81void __init kclist_add(struct kcore_list *new, void *addr, size_t size,
82 int type)
83{
84 new->addr = (unsigned long)addr;
85 new->size = size;
86 new->type = type;
87
88 list_add_tail(&new->list, &kclist_head);
89}
90
91static size_t get_kcore_size(int *nphdr, size_t *phdrs_len, size_t *notes_len,
92 size_t *data_offset)
93{
94 size_t try, size;
95 struct kcore_list *m;
96
97 *nphdr = 1; /* PT_NOTE */
98 size = 0;
99
100 list_for_each_entry(m, &kclist_head, list) {
101 try = kc_vaddr_to_offset((size_t)m->addr + m->size);
102 if (try > size)
103 size = try;
104 *nphdr = *nphdr + 1;
105 }
106
107 *phdrs_len = *nphdr * sizeof(struct elf_phdr);
108 *notes_len = (4 * sizeof(struct elf_note) +
109 3 * ALIGN(sizeof(CORE_STR), 4) +
110 VMCOREINFO_NOTE_NAME_BYTES +
111 ALIGN(sizeof(struct elf_prstatus), 4) +
112 ALIGN(sizeof(struct elf_prpsinfo), 4) +
113 ALIGN(arch_task_struct_size, 4) +
114 ALIGN(vmcoreinfo_size, 4));
115 *data_offset = PAGE_ALIGN(sizeof(struct elfhdr) + *phdrs_len +
116 *notes_len);
117 return *data_offset + size;
118}
119
120#ifdef CONFIG_HIGHMEM
121/*
122 * If no highmem, we can assume [0...max_low_pfn) continuous range of memory
123 * because memory hole is not as big as !HIGHMEM case.
124 * (HIGHMEM is special because part of memory is _invisible_ from the kernel.)
125 */
126static int kcore_ram_list(struct list_head *head)
127{
128 struct kcore_list *ent;
129
130 ent = kmalloc(sizeof(*ent), GFP_KERNEL);
131 if (!ent)
132 return -ENOMEM;
133 ent->addr = (unsigned long)__va(0);
134 ent->size = max_low_pfn << PAGE_SHIFT;
135 ent->type = KCORE_RAM;
136 list_add(&ent->list, head);
137 return 0;
138}
139
140#else /* !CONFIG_HIGHMEM */
141
142#ifdef CONFIG_SPARSEMEM_VMEMMAP
143/* calculate vmemmap's address from given system ram pfn and register it */
144static int
145get_sparsemem_vmemmap_info(struct kcore_list *ent, struct list_head *head)
146{
147 unsigned long pfn = __pa(ent->addr) >> PAGE_SHIFT;
148 unsigned long nr_pages = ent->size >> PAGE_SHIFT;
149 unsigned long start, end;
150 struct kcore_list *vmm, *tmp;
151
152
153 start = ((unsigned long)pfn_to_page(pfn)) & PAGE_MASK;
154 end = ((unsigned long)pfn_to_page(pfn + nr_pages)) - 1;
155 end = PAGE_ALIGN(end);
156 /* overlap check (because we have to align page */
157 list_for_each_entry(tmp, head, list) {
158 if (tmp->type != KCORE_VMEMMAP)
159 continue;
160 if (start < tmp->addr + tmp->size)
161 if (end > tmp->addr)
162 end = tmp->addr;
163 }
164 if (start < end) {
165 vmm = kmalloc(sizeof(*vmm), GFP_KERNEL);
166 if (!vmm)
167 return 0;
168 vmm->addr = start;
169 vmm->size = end - start;
170 vmm->type = KCORE_VMEMMAP;
171 list_add_tail(&vmm->list, head);
172 }
173 return 1;
174
175}
176#else
177static int
178get_sparsemem_vmemmap_info(struct kcore_list *ent, struct list_head *head)
179{
180 return 1;
181}
182
183#endif
184
185static int
186kclist_add_private(unsigned long pfn, unsigned long nr_pages, void *arg)
187{
188 struct list_head *head = (struct list_head *)arg;
189 struct kcore_list *ent;
190 struct page *p;
191
192 if (!pfn_valid(pfn))
193 return 1;
194
195 p = pfn_to_page(pfn);
196
197 ent = kmalloc(sizeof(*ent), GFP_KERNEL);
198 if (!ent)
199 return -ENOMEM;
200 ent->addr = (unsigned long)page_to_virt(p);
201 ent->size = nr_pages << PAGE_SHIFT;
202
203 if (!virt_addr_valid(ent->addr))
204 goto free_out;
205
206 /* cut not-mapped area. ....from ppc-32 code. */
207 if (ULONG_MAX - ent->addr < ent->size)
208 ent->size = ULONG_MAX - ent->addr;
209
210 /*
211 * We've already checked virt_addr_valid so we know this address
212 * is a valid pointer, therefore we can check against it to determine
213 * if we need to trim
214 */
215 if (VMALLOC_START > ent->addr) {
216 if (VMALLOC_START - ent->addr < ent->size)
217 ent->size = VMALLOC_START - ent->addr;
218 }
219
220 ent->type = KCORE_RAM;
221 list_add_tail(&ent->list, head);
222
223 if (!get_sparsemem_vmemmap_info(ent, head)) {
224 list_del(&ent->list);
225 goto free_out;
226 }
227
228 return 0;
229free_out:
230 kfree(ent);
231 return 1;
232}
233
234static int kcore_ram_list(struct list_head *list)
235{
236 int nid, ret;
237 unsigned long end_pfn;
238
239 /* Not inialized....update now */
240 /* find out "max pfn" */
241 end_pfn = 0;
242 for_each_node_state(nid, N_MEMORY) {
243 unsigned long node_end;
244 node_end = node_end_pfn(nid);
245 if (end_pfn < node_end)
246 end_pfn = node_end;
247 }
248 /* scan 0 to max_pfn */
249 ret = walk_system_ram_range(0, end_pfn, list, kclist_add_private);
250 if (ret)
251 return -ENOMEM;
252 return 0;
253}
254#endif /* CONFIG_HIGHMEM */
255
256static int kcore_update_ram(void)
257{
258 LIST_HEAD(list);
259 LIST_HEAD(garbage);
260 int nphdr;
261 size_t phdrs_len, notes_len, data_offset;
262 struct kcore_list *tmp, *pos;
263 int ret = 0;
264
265 down_write(&kclist_lock);
266 if (!xchg(&kcore_need_update, 0))
267 goto out;
268
269 ret = kcore_ram_list(&list);
270 if (ret) {
271 /* Couldn't get the RAM list, try again next time. */
272 WRITE_ONCE(kcore_need_update, 1);
273 list_splice_tail(&list, &garbage);
274 goto out;
275 }
276
277 list_for_each_entry_safe(pos, tmp, &kclist_head, list) {
278 if (pos->type == KCORE_RAM || pos->type == KCORE_VMEMMAP)
279 list_move(&pos->list, &garbage);
280 }
281 list_splice_tail(&list, &kclist_head);
282
283 proc_root_kcore->size = get_kcore_size(&nphdr, &phdrs_len, ¬es_len,
284 &data_offset);
285
286out:
287 up_write(&kclist_lock);
288 list_for_each_entry_safe(pos, tmp, &garbage, list) {
289 list_del(&pos->list);
290 kfree(pos);
291 }
292 return ret;
293}
294
295static void append_kcore_note(char *notes, size_t *i, const char *name,
296 unsigned int type, const void *desc,
297 size_t descsz)
298{
299 struct elf_note *note = (struct elf_note *)¬es[*i];
300
301 note->n_namesz = strlen(name) + 1;
302 note->n_descsz = descsz;
303 note->n_type = type;
304 *i += sizeof(*note);
305 memcpy(¬es[*i], name, note->n_namesz);
306 *i = ALIGN(*i + note->n_namesz, 4);
307 memcpy(¬es[*i], desc, descsz);
308 *i = ALIGN(*i + descsz, 4);
309}
310
311static ssize_t
312read_kcore(struct file *file, char __user *buffer, size_t buflen, loff_t *fpos)
313{
314 char *buf = file->private_data;
315 size_t phdrs_offset, notes_offset, data_offset;
316 size_t page_offline_frozen = 1;
317 size_t phdrs_len, notes_len;
318 struct kcore_list *m;
319 size_t tsz;
320 int nphdr;
321 unsigned long start;
322 size_t orig_buflen = buflen;
323 int ret = 0;
324
325 down_read(&kclist_lock);
326 /*
327 * Don't race against drivers that set PageOffline() and expect no
328 * further page access.
329 */
330 page_offline_freeze();
331
332 get_kcore_size(&nphdr, &phdrs_len, ¬es_len, &data_offset);
333 phdrs_offset = sizeof(struct elfhdr);
334 notes_offset = phdrs_offset + phdrs_len;
335
336 /* ELF file header. */
337 if (buflen && *fpos < sizeof(struct elfhdr)) {
338 struct elfhdr ehdr = {
339 .e_ident = {
340 [EI_MAG0] = ELFMAG0,
341 [EI_MAG1] = ELFMAG1,
342 [EI_MAG2] = ELFMAG2,
343 [EI_MAG3] = ELFMAG3,
344 [EI_CLASS] = ELF_CLASS,
345 [EI_DATA] = ELF_DATA,
346 [EI_VERSION] = EV_CURRENT,
347 [EI_OSABI] = ELF_OSABI,
348 },
349 .e_type = ET_CORE,
350 .e_machine = ELF_ARCH,
351 .e_version = EV_CURRENT,
352 .e_phoff = sizeof(struct elfhdr),
353 .e_flags = ELF_CORE_EFLAGS,
354 .e_ehsize = sizeof(struct elfhdr),
355 .e_phentsize = sizeof(struct elf_phdr),
356 .e_phnum = nphdr,
357 };
358
359 tsz = min_t(size_t, buflen, sizeof(struct elfhdr) - *fpos);
360 if (copy_to_user(buffer, (char *)&ehdr + *fpos, tsz)) {
361 ret = -EFAULT;
362 goto out;
363 }
364
365 buffer += tsz;
366 buflen -= tsz;
367 *fpos += tsz;
368 }
369
370 /* ELF program headers. */
371 if (buflen && *fpos < phdrs_offset + phdrs_len) {
372 struct elf_phdr *phdrs, *phdr;
373
374 phdrs = kzalloc(phdrs_len, GFP_KERNEL);
375 if (!phdrs) {
376 ret = -ENOMEM;
377 goto out;
378 }
379
380 phdrs[0].p_type = PT_NOTE;
381 phdrs[0].p_offset = notes_offset;
382 phdrs[0].p_filesz = notes_len;
383
384 phdr = &phdrs[1];
385 list_for_each_entry(m, &kclist_head, list) {
386 phdr->p_type = PT_LOAD;
387 phdr->p_flags = PF_R | PF_W | PF_X;
388 phdr->p_offset = kc_vaddr_to_offset(m->addr) + data_offset;
389 phdr->p_vaddr = (size_t)m->addr;
390 if (m->type == KCORE_RAM)
391 phdr->p_paddr = __pa(m->addr);
392 else if (m->type == KCORE_TEXT)
393 phdr->p_paddr = __pa_symbol(m->addr);
394 else
395 phdr->p_paddr = (elf_addr_t)-1;
396 phdr->p_filesz = phdr->p_memsz = m->size;
397 phdr->p_align = PAGE_SIZE;
398 phdr++;
399 }
400
401 tsz = min_t(size_t, buflen, phdrs_offset + phdrs_len - *fpos);
402 if (copy_to_user(buffer, (char *)phdrs + *fpos - phdrs_offset,
403 tsz)) {
404 kfree(phdrs);
405 ret = -EFAULT;
406 goto out;
407 }
408 kfree(phdrs);
409
410 buffer += tsz;
411 buflen -= tsz;
412 *fpos += tsz;
413 }
414
415 /* ELF note segment. */
416 if (buflen && *fpos < notes_offset + notes_len) {
417 struct elf_prstatus prstatus = {};
418 struct elf_prpsinfo prpsinfo = {
419 .pr_sname = 'R',
420 .pr_fname = "vmlinux",
421 };
422 char *notes;
423 size_t i = 0;
424
425 strlcpy(prpsinfo.pr_psargs, saved_command_line,
426 sizeof(prpsinfo.pr_psargs));
427
428 notes = kzalloc(notes_len, GFP_KERNEL);
429 if (!notes) {
430 ret = -ENOMEM;
431 goto out;
432 }
433
434 append_kcore_note(notes, &i, CORE_STR, NT_PRSTATUS, &prstatus,
435 sizeof(prstatus));
436 append_kcore_note(notes, &i, CORE_STR, NT_PRPSINFO, &prpsinfo,
437 sizeof(prpsinfo));
438 append_kcore_note(notes, &i, CORE_STR, NT_TASKSTRUCT, current,
439 arch_task_struct_size);
440 /*
441 * vmcoreinfo_size is mostly constant after init time, but it
442 * can be changed by crash_save_vmcoreinfo(). Racing here with a
443 * panic on another CPU before the machine goes down is insanely
444 * unlikely, but it's better to not leave potential buffer
445 * overflows lying around, regardless.
446 */
447 append_kcore_note(notes, &i, VMCOREINFO_NOTE_NAME, 0,
448 vmcoreinfo_data,
449 min(vmcoreinfo_size, notes_len - i));
450
451 tsz = min_t(size_t, buflen, notes_offset + notes_len - *fpos);
452 if (copy_to_user(buffer, notes + *fpos - notes_offset, tsz)) {
453 kfree(notes);
454 ret = -EFAULT;
455 goto out;
456 }
457 kfree(notes);
458
459 buffer += tsz;
460 buflen -= tsz;
461 *fpos += tsz;
462 }
463
464 /*
465 * Check to see if our file offset matches with any of
466 * the addresses in the elf_phdr on our list.
467 */
468 start = kc_offset_to_vaddr(*fpos - data_offset);
469 if ((tsz = (PAGE_SIZE - (start & ~PAGE_MASK))) > buflen)
470 tsz = buflen;
471
472 m = NULL;
473 while (buflen) {
474 struct page *page;
475 unsigned long pfn;
476
477 /*
478 * If this is the first iteration or the address is not within
479 * the previous entry, search for a matching entry.
480 */
481 if (!m || start < m->addr || start >= m->addr + m->size) {
482 list_for_each_entry(m, &kclist_head, list) {
483 if (start >= m->addr &&
484 start < m->addr + m->size)
485 break;
486 }
487 }
488
489 if (page_offline_frozen++ % MAX_ORDER_NR_PAGES == 0) {
490 page_offline_thaw();
491 cond_resched();
492 page_offline_freeze();
493 }
494
495 if (&m->list == &kclist_head) {
496 if (clear_user(buffer, tsz)) {
497 ret = -EFAULT;
498 goto out;
499 }
500 m = NULL; /* skip the list anchor */
501 goto skip;
502 }
503
504 switch (m->type) {
505 case KCORE_VMALLOC:
506 vread(buf, (char *)start, tsz);
507 /* we have to zero-fill user buffer even if no read */
508 if (copy_to_user(buffer, buf, tsz)) {
509 ret = -EFAULT;
510 goto out;
511 }
512 break;
513 case KCORE_USER:
514 /* User page is handled prior to normal kernel page: */
515 if (copy_to_user(buffer, (char *)start, tsz)) {
516 ret = -EFAULT;
517 goto out;
518 }
519 break;
520 case KCORE_RAM:
521 pfn = __pa(start) >> PAGE_SHIFT;
522 page = pfn_to_online_page(pfn);
523
524 /*
525 * Don't read offline sections, logically offline pages
526 * (e.g., inflated in a balloon), hwpoisoned pages,
527 * and explicitly excluded physical ranges.
528 */
529 if (!page || PageOffline(page) ||
530 is_page_hwpoison(page) || !pfn_is_ram(pfn)) {
531 if (clear_user(buffer, tsz)) {
532 ret = -EFAULT;
533 goto out;
534 }
535 break;
536 }
537 fallthrough;
538 case KCORE_VMEMMAP:
539 case KCORE_TEXT:
540 if (kern_addr_valid(start)) {
541 /*
542 * Using bounce buffer to bypass the
543 * hardened user copy kernel text checks.
544 */
545 if (copy_from_kernel_nofault(buf, (void *)start,
546 tsz)) {
547 if (clear_user(buffer, tsz)) {
548 ret = -EFAULT;
549 goto out;
550 }
551 } else {
552 if (copy_to_user(buffer, buf, tsz)) {
553 ret = -EFAULT;
554 goto out;
555 }
556 }
557 } else {
558 if (clear_user(buffer, tsz)) {
559 ret = -EFAULT;
560 goto out;
561 }
562 }
563 break;
564 default:
565 pr_warn_once("Unhandled KCORE type: %d\n", m->type);
566 if (clear_user(buffer, tsz)) {
567 ret = -EFAULT;
568 goto out;
569 }
570 }
571skip:
572 buflen -= tsz;
573 *fpos += tsz;
574 buffer += tsz;
575 start += tsz;
576 tsz = (buflen > PAGE_SIZE ? PAGE_SIZE : buflen);
577 }
578
579out:
580 page_offline_thaw();
581 up_read(&kclist_lock);
582 if (ret)
583 return ret;
584 return orig_buflen - buflen;
585}
586
587static int open_kcore(struct inode *inode, struct file *filp)
588{
589 int ret = security_locked_down(LOCKDOWN_KCORE);
590
591 if (!capable(CAP_SYS_RAWIO))
592 return -EPERM;
593
594 if (ret)
595 return ret;
596
597 filp->private_data = kmalloc(PAGE_SIZE, GFP_KERNEL);
598 if (!filp->private_data)
599 return -ENOMEM;
600
601 if (kcore_need_update)
602 kcore_update_ram();
603 if (i_size_read(inode) != proc_root_kcore->size) {
604 inode_lock(inode);
605 i_size_write(inode, proc_root_kcore->size);
606 inode_unlock(inode);
607 }
608 return 0;
609}
610
611static int release_kcore(struct inode *inode, struct file *file)
612{
613 kfree(file->private_data);
614 return 0;
615}
616
617static const struct proc_ops kcore_proc_ops = {
618 .proc_read = read_kcore,
619 .proc_open = open_kcore,
620 .proc_release = release_kcore,
621 .proc_lseek = default_llseek,
622};
623
624/* just remember that we have to update kcore */
625static int __meminit kcore_callback(struct notifier_block *self,
626 unsigned long action, void *arg)
627{
628 switch (action) {
629 case MEM_ONLINE:
630 case MEM_OFFLINE:
631 kcore_need_update = 1;
632 break;
633 }
634 return NOTIFY_OK;
635}
636
637static struct notifier_block kcore_callback_nb __meminitdata = {
638 .notifier_call = kcore_callback,
639 .priority = 0,
640};
641
642static struct kcore_list kcore_vmalloc;
643
644#ifdef CONFIG_ARCH_PROC_KCORE_TEXT
645static struct kcore_list kcore_text;
646/*
647 * If defined, special segment is used for mapping kernel text instead of
648 * direct-map area. We need to create special TEXT section.
649 */
650static void __init proc_kcore_text_init(void)
651{
652 kclist_add(&kcore_text, _text, _end - _text, KCORE_TEXT);
653}
654#else
655static void __init proc_kcore_text_init(void)
656{
657}
658#endif
659
660#if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
661/*
662 * MODULES_VADDR has no intersection with VMALLOC_ADDR.
663 */
664static struct kcore_list kcore_modules;
665static void __init add_modules_range(void)
666{
667 if (MODULES_VADDR != VMALLOC_START && MODULES_END != VMALLOC_END) {
668 kclist_add(&kcore_modules, (void *)MODULES_VADDR,
669 MODULES_END - MODULES_VADDR, KCORE_VMALLOC);
670 }
671}
672#else
673static void __init add_modules_range(void)
674{
675}
676#endif
677
678static int __init proc_kcore_init(void)
679{
680 proc_root_kcore = proc_create("kcore", S_IRUSR, NULL, &kcore_proc_ops);
681 if (!proc_root_kcore) {
682 pr_err("couldn't create /proc/kcore\n");
683 return 0; /* Always returns 0. */
684 }
685 /* Store text area if it's special */
686 proc_kcore_text_init();
687 /* Store vmalloc area */
688 kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
689 VMALLOC_END - VMALLOC_START, KCORE_VMALLOC);
690 add_modules_range();
691 /* Store direct-map area from physical memory map */
692 kcore_update_ram();
693 register_hotmemory_notifier(&kcore_callback_nb);
694
695 return 0;
696}
697fs_initcall(proc_kcore_init);