1 // SPDX-License-Identifier: GPL-2.0
3 * fs/proc/kcore.c kernel ELF core dumper
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>
13 #include <linux/crash_core.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>
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>
38 #define CORE_STR "CORE"
40 #ifndef ELF_CORE_EFLAGS
41 #define ELF_CORE_EFLAGS 0
44 static struct proc_dir_entry
*proc_root_kcore
;
47 #ifndef kc_vaddr_to_offset
48 #define kc_vaddr_to_offset(v) ((v) - PAGE_OFFSET)
50 #ifndef kc_offset_to_vaddr
51 #define kc_offset_to_vaddr(o) ((o) + PAGE_OFFSET)
54 static LIST_HEAD(kclist_head
);
55 static DECLARE_RWSEM(kclist_lock
);
56 static int kcore_need_update
= 1;
59 * Returns > 0 for RAM pages, 0 for non-RAM pages, < 0 on error
60 * Same as oldmem_pfn_is_ram in vmcore
62 static int (*mem_pfn_is_ram
)(unsigned long pfn
);
64 int __init
register_mem_pfn_is_ram(int (*fn
)(unsigned long pfn
))
72 static int pfn_is_ram(unsigned long pfn
)
75 return mem_pfn_is_ram(pfn
);
80 /* This doesn't grab kclist_lock, so it should only be used at init time. */
81 void __init
kclist_add(struct kcore_list
*new, void *addr
, size_t size
,
84 new->addr
= (unsigned long)addr
;
88 list_add_tail(&new->list
, &kclist_head
);
91 static size_t get_kcore_size(int *nphdr
, size_t *phdrs_len
, size_t *notes_len
,
97 *nphdr
= 1; /* PT_NOTE */
100 list_for_each_entry(m
, &kclist_head
, list
) {
101 try = kc_vaddr_to_offset((size_t)m
->addr
+ m
->size
);
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
+
117 return *data_offset
+ size
;
120 #ifdef CONFIG_HIGHMEM
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.)
126 static int kcore_ram_list(struct list_head
*head
)
128 struct kcore_list
*ent
;
130 ent
= kmalloc(sizeof(*ent
), GFP_KERNEL
);
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
);
140 #else /* !CONFIG_HIGHMEM */
142 #ifdef CONFIG_SPARSEMEM_VMEMMAP
143 /* calculate vmemmap's address from given system ram pfn and register it */
145 get_sparsemem_vmemmap_info(struct kcore_list
*ent
, struct list_head
*head
)
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
;
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
)
160 if (start
< tmp
->addr
+ tmp
->size
)
165 vmm
= kmalloc(sizeof(*vmm
), GFP_KERNEL
);
169 vmm
->size
= end
- start
;
170 vmm
->type
= KCORE_VMEMMAP
;
171 list_add_tail(&vmm
->list
, head
);
178 get_sparsemem_vmemmap_info(struct kcore_list
*ent
, struct list_head
*head
)
186 kclist_add_private(unsigned long pfn
, unsigned long nr_pages
, void *arg
)
188 struct list_head
*head
= (struct list_head
*)arg
;
189 struct kcore_list
*ent
;
195 p
= pfn_to_page(pfn
);
196 if (!memmap_valid_within(pfn
, p
, page_zone(p
)))
199 ent
= kmalloc(sizeof(*ent
), GFP_KERNEL
);
202 ent
->addr
= (unsigned long)page_to_virt(p
);
203 ent
->size
= nr_pages
<< PAGE_SHIFT
;
205 if (!virt_addr_valid(ent
->addr
))
208 /* cut not-mapped area. ....from ppc-32 code. */
209 if (ULONG_MAX
- ent
->addr
< ent
->size
)
210 ent
->size
= ULONG_MAX
- ent
->addr
;
213 * We've already checked virt_addr_valid so we know this address
214 * is a valid pointer, therefore we can check against it to determine
217 if (VMALLOC_START
> ent
->addr
) {
218 if (VMALLOC_START
- ent
->addr
< ent
->size
)
219 ent
->size
= VMALLOC_START
- ent
->addr
;
222 ent
->type
= KCORE_RAM
;
223 list_add_tail(&ent
->list
, head
);
225 if (!get_sparsemem_vmemmap_info(ent
, head
)) {
226 list_del(&ent
->list
);
236 static int kcore_ram_list(struct list_head
*list
)
239 unsigned long end_pfn
;
241 /* Not inialized....update now */
242 /* find out "max pfn" */
244 for_each_node_state(nid
, N_MEMORY
) {
245 unsigned long node_end
;
246 node_end
= node_end_pfn(nid
);
247 if (end_pfn
< node_end
)
250 /* scan 0 to max_pfn */
251 ret
= walk_system_ram_range(0, end_pfn
, list
, kclist_add_private
);
256 #endif /* CONFIG_HIGHMEM */
258 static int kcore_update_ram(void)
263 size_t phdrs_len
, notes_len
, data_offset
;
264 struct kcore_list
*tmp
, *pos
;
267 down_write(&kclist_lock
);
268 if (!xchg(&kcore_need_update
, 0))
271 ret
= kcore_ram_list(&list
);
273 /* Couldn't get the RAM list, try again next time. */
274 WRITE_ONCE(kcore_need_update
, 1);
275 list_splice_tail(&list
, &garbage
);
279 list_for_each_entry_safe(pos
, tmp
, &kclist_head
, list
) {
280 if (pos
->type
== KCORE_RAM
|| pos
->type
== KCORE_VMEMMAP
)
281 list_move(&pos
->list
, &garbage
);
283 list_splice_tail(&list
, &kclist_head
);
285 proc_root_kcore
->size
= get_kcore_size(&nphdr
, &phdrs_len
, ¬es_len
,
289 up_write(&kclist_lock
);
290 list_for_each_entry_safe(pos
, tmp
, &garbage
, list
) {
291 list_del(&pos
->list
);
297 static void append_kcore_note(char *notes
, size_t *i
, const char *name
,
298 unsigned int type
, const void *desc
,
301 struct elf_note
*note
= (struct elf_note
*)¬es
[*i
];
303 note
->n_namesz
= strlen(name
) + 1;
304 note
->n_descsz
= descsz
;
307 memcpy(¬es
[*i
], name
, note
->n_namesz
);
308 *i
= ALIGN(*i
+ note
->n_namesz
, 4);
309 memcpy(¬es
[*i
], desc
, descsz
);
310 *i
= ALIGN(*i
+ descsz
, 4);
314 read_kcore(struct file
*file
, char __user
*buffer
, size_t buflen
, loff_t
*fpos
)
316 char *buf
= file
->private_data
;
317 size_t phdrs_offset
, notes_offset
, data_offset
;
318 size_t phdrs_len
, notes_len
;
319 struct kcore_list
*m
;
323 size_t orig_buflen
= buflen
;
326 down_read(&kclist_lock
);
328 get_kcore_size(&nphdr
, &phdrs_len
, ¬es_len
, &data_offset
);
329 phdrs_offset
= sizeof(struct elfhdr
);
330 notes_offset
= phdrs_offset
+ phdrs_len
;
332 /* ELF file header. */
333 if (buflen
&& *fpos
< sizeof(struct elfhdr
)) {
334 struct elfhdr ehdr
= {
340 [EI_CLASS
] = ELF_CLASS
,
341 [EI_DATA
] = ELF_DATA
,
342 [EI_VERSION
] = EV_CURRENT
,
343 [EI_OSABI
] = ELF_OSABI
,
346 .e_machine
= ELF_ARCH
,
347 .e_version
= EV_CURRENT
,
348 .e_phoff
= sizeof(struct elfhdr
),
349 .e_flags
= ELF_CORE_EFLAGS
,
350 .e_ehsize
= sizeof(struct elfhdr
),
351 .e_phentsize
= sizeof(struct elf_phdr
),
355 tsz
= min_t(size_t, buflen
, sizeof(struct elfhdr
) - *fpos
);
356 if (copy_to_user(buffer
, (char *)&ehdr
+ *fpos
, tsz
)) {
366 /* ELF program headers. */
367 if (buflen
&& *fpos
< phdrs_offset
+ phdrs_len
) {
368 struct elf_phdr
*phdrs
, *phdr
;
370 phdrs
= kzalloc(phdrs_len
, GFP_KERNEL
);
376 phdrs
[0].p_type
= PT_NOTE
;
377 phdrs
[0].p_offset
= notes_offset
;
378 phdrs
[0].p_filesz
= notes_len
;
381 list_for_each_entry(m
, &kclist_head
, list
) {
382 phdr
->p_type
= PT_LOAD
;
383 phdr
->p_flags
= PF_R
| PF_W
| PF_X
;
384 phdr
->p_offset
= kc_vaddr_to_offset(m
->addr
) + data_offset
;
385 if (m
->type
== KCORE_REMAP
)
386 phdr
->p_vaddr
= (size_t)m
->vaddr
;
388 phdr
->p_vaddr
= (size_t)m
->addr
;
389 if (m
->type
== KCORE_RAM
|| m
->type
== KCORE_REMAP
)
390 phdr
->p_paddr
= __pa(m
->addr
);
391 else if (m
->type
== KCORE_TEXT
)
392 phdr
->p_paddr
= __pa_symbol(m
->addr
);
394 phdr
->p_paddr
= (elf_addr_t
)-1;
395 phdr
->p_filesz
= phdr
->p_memsz
= m
->size
;
396 phdr
->p_align
= PAGE_SIZE
;
400 tsz
= min_t(size_t, buflen
, phdrs_offset
+ phdrs_len
- *fpos
);
401 if (copy_to_user(buffer
, (char *)phdrs
+ *fpos
- phdrs_offset
,
414 /* ELF note segment. */
415 if (buflen
&& *fpos
< notes_offset
+ notes_len
) {
416 struct elf_prstatus prstatus
= {};
417 struct elf_prpsinfo prpsinfo
= {
419 .pr_fname
= "vmlinux",
424 strlcpy(prpsinfo
.pr_psargs
, saved_command_line
,
425 sizeof(prpsinfo
.pr_psargs
));
427 notes
= kzalloc(notes_len
, GFP_KERNEL
);
433 append_kcore_note(notes
, &i
, CORE_STR
, NT_PRSTATUS
, &prstatus
,
435 append_kcore_note(notes
, &i
, CORE_STR
, NT_PRPSINFO
, &prpsinfo
,
437 append_kcore_note(notes
, &i
, CORE_STR
, NT_TASKSTRUCT
, current
,
438 arch_task_struct_size
);
440 * vmcoreinfo_size is mostly constant after init time, but it
441 * can be changed by crash_save_vmcoreinfo(). Racing here with a
442 * panic on another CPU before the machine goes down is insanely
443 * unlikely, but it's better to not leave potential buffer
444 * overflows lying around, regardless.
446 append_kcore_note(notes
, &i
, VMCOREINFO_NOTE_NAME
, 0,
448 min(vmcoreinfo_size
, notes_len
- i
));
450 tsz
= min_t(size_t, buflen
, notes_offset
+ notes_len
- *fpos
);
451 if (copy_to_user(buffer
, notes
+ *fpos
- notes_offset
, tsz
)) {
464 * Check to see if our file offset matches with any of
465 * the addresses in the elf_phdr on our list.
467 start
= kc_offset_to_vaddr(*fpos
- data_offset
);
468 if ((tsz
= (PAGE_SIZE
- (start
& ~PAGE_MASK
))) > buflen
)
474 * If this is the first iteration or the address is not within
475 * the previous entry, search for a matching entry.
477 if (!m
|| start
< m
->addr
|| start
>= m
->addr
+ m
->size
) {
478 list_for_each_entry(m
, &kclist_head
, list
) {
479 if (start
>= m
->addr
&&
480 start
< m
->addr
+ m
->size
)
485 if (&m
->list
== &kclist_head
) {
486 if (clear_user(buffer
, tsz
)) {
490 m
= NULL
; /* skip the list anchor */
491 } else if (!pfn_is_ram(__pa(start
) >> PAGE_SHIFT
)) {
492 if (clear_user(buffer
, tsz
)) {
496 } else if (m
->type
== KCORE_VMALLOC
) {
497 vread(buf
, (char *)start
, tsz
);
498 /* we have to zero-fill user buffer even if no read */
499 if (copy_to_user(buffer
, buf
, tsz
)) {
503 } else if (m
->type
== KCORE_USER
) {
504 /* User page is handled prior to normal kernel page: */
505 if (copy_to_user(buffer
, (char *)start
, tsz
)) {
510 if (kern_addr_valid(start
)) {
512 * Using bounce buffer to bypass the
513 * hardened user copy kernel text checks.
515 if (probe_kernel_read(buf
, (void *) start
, tsz
)) {
516 if (clear_user(buffer
, tsz
)) {
521 if (copy_to_user(buffer
, buf
, tsz
)) {
527 if (clear_user(buffer
, tsz
)) {
537 tsz
= (buflen
> PAGE_SIZE
? PAGE_SIZE
: buflen
);
541 up_read(&kclist_lock
);
544 return orig_buflen
- buflen
;
547 static int open_kcore(struct inode
*inode
, struct file
*filp
)
549 int ret
= security_locked_down(LOCKDOWN_KCORE
);
551 if (!capable(CAP_SYS_RAWIO
))
557 filp
->private_data
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
558 if (!filp
->private_data
)
561 if (kcore_need_update
)
563 if (i_size_read(inode
) != proc_root_kcore
->size
) {
565 i_size_write(inode
, proc_root_kcore
->size
);
571 static int release_kcore(struct inode
*inode
, struct file
*file
)
573 kfree(file
->private_data
);
577 static const struct proc_ops kcore_proc_ops
= {
578 .proc_read
= read_kcore
,
579 .proc_open
= open_kcore
,
580 .proc_release
= release_kcore
,
581 .proc_lseek
= default_llseek
,
584 /* just remember that we have to update kcore */
585 static int __meminit
kcore_callback(struct notifier_block
*self
,
586 unsigned long action
, void *arg
)
591 kcore_need_update
= 1;
597 static struct notifier_block kcore_callback_nb __meminitdata
= {
598 .notifier_call
= kcore_callback
,
602 static struct kcore_list kcore_vmalloc
;
604 #ifdef CONFIG_ARCH_PROC_KCORE_TEXT
605 static struct kcore_list kcore_text
;
607 * If defined, special segment is used for mapping kernel text instead of
608 * direct-map area. We need to create special TEXT section.
610 static void __init
proc_kcore_text_init(void)
612 kclist_add(&kcore_text
, _text
, _end
- _text
, KCORE_TEXT
);
615 static void __init
proc_kcore_text_init(void)
620 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
622 * MODULES_VADDR has no intersection with VMALLOC_ADDR.
624 static struct kcore_list kcore_modules
;
625 static void __init
add_modules_range(void)
627 if (MODULES_VADDR
!= VMALLOC_START
&& MODULES_END
!= VMALLOC_END
) {
628 kclist_add(&kcore_modules
, (void *)MODULES_VADDR
,
629 MODULES_END
- MODULES_VADDR
, KCORE_VMALLOC
);
633 static void __init
add_modules_range(void)
638 static int __init
proc_kcore_init(void)
640 proc_root_kcore
= proc_create("kcore", S_IRUSR
, NULL
, &kcore_proc_ops
);
641 if (!proc_root_kcore
) {
642 pr_err("couldn't create /proc/kcore\n");
643 return 0; /* Always returns 0. */
645 /* Store text area if it's special */
646 proc_kcore_text_init();
647 /* Store vmalloc area */
648 kclist_add(&kcore_vmalloc
, (void *)VMALLOC_START
,
649 VMALLOC_END
- VMALLOC_START
, KCORE_VMALLOC
);
651 /* Store direct-map area from physical memory map */
653 register_hotmemory_notifier(&kcore_callback_nb
);
657 fs_initcall(proc_kcore_init
);