| 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 <[email protected]> |
| 7 |
* ELF version written by David Howells <[email protected]> |
| 8 |
* Modified and incorporated into 2.3.x by Tigran Aivazian <[email protected]> |
| 9 |
* Support to dump vmalloc'd areas (ELF only), Tigran Aivazian <[email protected]> |
| 10 |
* Safe accesses to vmalloc/direct-mapped discontiguous areas, Kanoj Sarcar <[email protected]> |
| 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 |
#ifndef ELF_CORE_EFLAGS |
| 38 |
#define ELF_CORE_EFLAGS 0 |
| 39 |
#endif |
| 40 |
|
| 41 |
static struct proc_dir_entry *proc_root_kcore; |
| 42 |
|
| 43 |
|
| 44 |
#ifndef kc_vaddr_to_offset |
| 45 |
#define kc_vaddr_to_offset(v) ((v) - PAGE_OFFSET) |
| 46 |
#endif |
| 47 |
#ifndef kc_offset_to_vaddr |
| 48 |
#define kc_offset_to_vaddr(o) ((o) + PAGE_OFFSET) |
| 49 |
#endif |
| 50 |
|
| 51 |
#ifndef kc_xlate_dev_mem_ptr |
| 52 |
#define kc_xlate_dev_mem_ptr kc_xlate_dev_mem_ptr |
| 53 |
static inline void *kc_xlate_dev_mem_ptr(phys_addr_t phys) |
| 54 |
{ |
| 55 |
return __va(phys); |
| 56 |
} |
| 57 |
#endif |
| 58 |
#ifndef kc_unxlate_dev_mem_ptr |
| 59 |
#define kc_unxlate_dev_mem_ptr kc_unxlate_dev_mem_ptr |
| 60 |
static inline void kc_unxlate_dev_mem_ptr(phys_addr_t phys, void *virt) |
| 61 |
{ |
| 62 |
} |
| 63 |
#endif |
| 64 |
|
| 65 |
static LIST_HEAD(kclist_head); |
| 66 |
static int kcore_nphdr; |
| 67 |
static size_t kcore_phdrs_len; |
| 68 |
static size_t kcore_notes_len; |
| 69 |
static size_t kcore_data_offset; |
| 70 |
DEFINE_STATIC_PERCPU_RWSEM(kclist_lock); |
| 71 |
static int kcore_need_update = 1; |
| 72 |
|
| 73 |
/* |
| 74 |
* Returns > 0 for RAM pages, 0 for non-RAM pages, < 0 on error |
| 75 |
* Same as oldmem_pfn_is_ram in vmcore |
| 76 |
*/ |
| 77 |
static int (*mem_pfn_is_ram)(unsigned long pfn); |
| 78 |
|
| 79 |
int __init register_mem_pfn_is_ram(int (*fn)(unsigned long pfn)) |
| 80 |
{ |
| 81 |
if (mem_pfn_is_ram) |
| 82 |
return -EBUSY; |
| 83 |
mem_pfn_is_ram = fn; |
| 84 |
return 0; |
| 85 |
} |
| 86 |
|
| 87 |
static int pfn_is_ram(unsigned long pfn) |
| 88 |
{ |
| 89 |
if (mem_pfn_is_ram) |
| 90 |
return mem_pfn_is_ram(pfn); |
| 91 |
else |
| 92 |
return 1; |
| 93 |
} |
| 94 |
|
| 95 |
/* This doesn't grab kclist_lock, so it should only be used at init time. */ |
| 96 |
void __init kclist_add(struct kcore_list *new, void *addr, size_t size, |
| 97 |
int type) |
| 98 |
{ |
| 99 |
new->addr = (unsigned long)addr; |
| 100 |
new->size = size; |
| 101 |
new->type = type; |
| 102 |
|
| 103 |
list_add_tail(&new->list, &kclist_head); |
| 104 |
} |
| 105 |
|
| 106 |
static void update_kcore_size(void) |
| 107 |
{ |
| 108 |
size_t try, size; |
| 109 |
struct kcore_list *m; |
| 110 |
|
| 111 |
kcore_nphdr = 1; /* PT_NOTE */ |
| 112 |
size = 0; |
| 113 |
|
| 114 |
list_for_each_entry(m, &kclist_head, list) { |
| 115 |
try = kc_vaddr_to_offset((size_t)m->addr + m->size); |
| 116 |
if (try > size) |
| 117 |
size = try; |
| 118 |
kcore_nphdr++; |
| 119 |
} |
| 120 |
|
| 121 |
kcore_phdrs_len = kcore_nphdr * sizeof(struct elf_phdr); |
| 122 |
kcore_notes_len = (4 * sizeof(struct elf_note) + |
| 123 |
ALIGN(sizeof(NN_PRSTATUS), 4) + |
| 124 |
ALIGN(sizeof(NN_PRPSINFO), 4) + |
| 125 |
ALIGN(sizeof(NN_TASKSTRUCT), 4) + |
| 126 |
VMCOREINFO_NOTE_NAME_BYTES + |
| 127 |
ALIGN(sizeof(struct elf_prstatus), 4) + |
| 128 |
ALIGN(sizeof(struct elf_prpsinfo), 4) + |
| 129 |
ALIGN(arch_task_struct_size, 4) + |
| 130 |
ALIGN(vmcoreinfo_size, 4)); |
| 131 |
kcore_data_offset = PAGE_ALIGN(sizeof(struct elfhdr) + kcore_phdrs_len + |
| 132 |
kcore_notes_len); |
| 133 |
proc_root_kcore->size = kcore_data_offset + size; |
| 134 |
} |
| 135 |
|
| 136 |
#ifdef CONFIG_HIGHMEM |
| 137 |
/* |
| 138 |
* If no highmem, we can assume [0...max_low_pfn) continuous range of memory |
| 139 |
* because memory hole is not as big as !HIGHMEM case. |
| 140 |
* (HIGHMEM is special because part of memory is _invisible_ from the kernel.) |
| 141 |
*/ |
| 142 |
static int kcore_ram_list(struct list_head *head) |
| 143 |
{ |
| 144 |
struct kcore_list *ent; |
| 145 |
|
| 146 |
ent = kmalloc_obj(*ent); |
| 147 |
if (!ent) |
| 148 |
return -ENOMEM; |
| 149 |
ent->addr = (unsigned long)__va(0); |
| 150 |
ent->size = max_low_pfn << PAGE_SHIFT; |
| 151 |
ent->type = KCORE_RAM; |
| 152 |
list_add(&ent->list, head); |
| 153 |
return 0; |
| 154 |
} |
| 155 |
|
| 156 |
#else /* !CONFIG_HIGHMEM */ |
| 157 |
|
| 158 |
#ifdef CONFIG_SPARSEMEM_VMEMMAP |
| 159 |
/* calculate vmemmap's address from given system ram pfn and register it */ |
| 160 |
static int |
| 161 |
get_sparsemem_vmemmap_info(struct kcore_list *ent, struct list_head *head) |
| 162 |
{ |
| 163 |
unsigned long pfn = __pa(ent->addr) >> PAGE_SHIFT; |
| 164 |
unsigned long nr_pages = ent->size >> PAGE_SHIFT; |
| 165 |
unsigned long start, end; |
| 166 |
struct kcore_list *vmm, *tmp; |
| 167 |
|
| 168 |
|
| 169 |
start = ((unsigned long)pfn_to_page(pfn)) & PAGE_MASK; |
| 170 |
end = ((unsigned long)pfn_to_page(pfn + nr_pages)) - 1; |
| 171 |
end = PAGE_ALIGN(end); |
| 172 |
/* overlap check (because we have to align page */ |
| 173 |
list_for_each_entry(tmp, head, list) { |
| 174 |
if (tmp->type != KCORE_VMEMMAP) |
| 175 |
continue; |
| 176 |
if (start < tmp->addr + tmp->size) |
| 177 |
if (end > tmp->addr) |
| 178 |
end = tmp->addr; |
| 179 |
} |
| 180 |
if (start < end) { |
| 181 |
vmm = kmalloc_obj(*vmm); |
| 182 |
if (!vmm) |
| 183 |
return 0; |
| 184 |
vmm->addr = start; |
| 185 |
vmm->size = end - start; |
| 186 |
vmm->type = KCORE_VMEMMAP; |
| 187 |
list_add_tail(&vmm->list, head); |
| 188 |
} |
| 189 |
return 1; |
| 190 |
|
| 191 |
} |
| 192 |
#else |
| 193 |
static int |
| 194 |
get_sparsemem_vmemmap_info(struct kcore_list *ent, struct list_head *head) |
| 195 |
{ |
| 196 |
return 1; |
| 197 |
} |
| 198 |
|
| 199 |
#endif |
| 200 |
|
| 201 |
static int |
| 202 |
kclist_add_private(unsigned long pfn, unsigned long nr_pages, void *arg) |
| 203 |
{ |
| 204 |
struct list_head *head = (struct list_head *)arg; |
| 205 |
struct kcore_list *ent; |
| 206 |
struct page *p; |
| 207 |
|
| 208 |
if (!pfn_valid(pfn)) |
| 209 |
return 1; |
| 210 |
|
| 211 |
p = pfn_to_page(pfn); |
| 212 |
|
| 213 |
ent = kmalloc_obj(*ent); |
| 214 |
if (!ent) |
| 215 |
return -ENOMEM; |
| 216 |
ent->addr = (unsigned long)page_to_virt(p); |
| 217 |
ent->size = nr_pages << PAGE_SHIFT; |
| 218 |
|
| 219 |
if (!virt_addr_valid((void *)ent->addr)) |
| 220 |
goto free_out; |
| 221 |
|
| 222 |
/* cut not-mapped area. ....from ppc-32 code. */ |
| 223 |
if (ULONG_MAX - ent->addr < ent->size) |
| 224 |
ent->size = ULONG_MAX - ent->addr; |
| 225 |
|
| 226 |
/* |
| 227 |
* We've already checked virt_addr_valid so we know this address |
| 228 |
* is a valid pointer, therefore we can check against it to determine |
| 229 |
* if we need to trim |
| 230 |
*/ |
| 231 |
if (VMALLOC_START > ent->addr) { |
| 232 |
if (VMALLOC_START - ent->addr < ent->size) |
| 233 |
ent->size = VMALLOC_START - ent->addr; |
| 234 |
} |
| 235 |
|
| 236 |
ent->type = KCORE_RAM; |
| 237 |
list_add_tail(&ent->list, head); |
| 238 |
|
| 239 |
if (!get_sparsemem_vmemmap_info(ent, head)) { |
| 240 |
list_del(&ent->list); |
| 241 |
goto free_out; |
| 242 |
} |
| 243 |
|
| 244 |
return 0; |
| 245 |
free_out: |
| 246 |
kfree(ent); |
| 247 |
return 1; |
| 248 |
} |
| 249 |
|
| 250 |
static int kcore_ram_list(struct list_head *list) |
| 251 |
{ |
| 252 |
int nid, ret; |
| 253 |
unsigned long end_pfn; |
| 254 |
|
| 255 |
/* Not initialized....update now */ |
| 256 |
/* find out "max pfn" */ |
| 257 |
end_pfn = 0; |
| 258 |
for_each_node_state(nid, N_MEMORY) { |
| 259 |
unsigned long node_end; |
| 260 |
node_end = node_end_pfn(nid); |
| 261 |
if (end_pfn < node_end) |
| 262 |
end_pfn = node_end; |
| 263 |
} |
| 264 |
/* scan 0 to max_pfn */ |
| 265 |
ret = walk_system_ram_range(0, end_pfn, list, kclist_add_private); |
| 266 |
if (ret) |
| 267 |
return -ENOMEM; |
| 268 |
return 0; |
| 269 |
} |
| 270 |
#endif /* CONFIG_HIGHMEM */ |
| 271 |
|
| 272 |
static int kcore_update_ram(void) |
| 273 |
{ |
| 274 |
LIST_HEAD(list); |
| 275 |
LIST_HEAD(garbage); |
| 276 |
struct kcore_list *tmp, *pos; |
| 277 |
int ret = 0; |
| 278 |
|
| 279 |
percpu_down_write(&kclist_lock); |
| 280 |
if (!xchg(&kcore_need_update, 0)) |
| 281 |
goto out; |
| 282 |
|
| 283 |
ret = kcore_ram_list(&list); |
| 284 |
if (ret) { |
| 285 |
/* Couldn't get the RAM list, try again next time. */ |
| 286 |
WRITE_ONCE(kcore_need_update, 1); |
| 287 |
list_splice_tail(&list, &garbage); |
| 288 |
goto out; |
| 289 |
} |
| 290 |
|
| 291 |
list_for_each_entry_safe(pos, tmp, &kclist_head, list) { |
| 292 |
if (pos->type == KCORE_RAM || pos->type == KCORE_VMEMMAP) |
| 293 |
list_move(&pos->list, &garbage); |
| 294 |
} |
| 295 |
list_splice_tail(&list, &kclist_head); |
| 296 |
|
| 297 |
update_kcore_size(); |
| 298 |
|
| 299 |
out: |
| 300 |
percpu_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 |
|
| 308 |
static 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 |
|
| 324 |
static 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; |
| 330 |
size_t page_offline_frozen = 1; |
| 331 |
struct kcore_list *m; |
| 332 |
size_t tsz; |
| 333 |
unsigned long start; |
| 334 |
size_t buflen = iov_iter_count(iter); |
| 335 |
size_t orig_buflen = buflen; |
| 336 |
int ret = 0; |
| 337 |
|
| 338 |
percpu_down_read(&kclist_lock); |
| 339 |
/* |
| 340 |
* Don't race against drivers that set PageOffline() and expect no |
| 341 |
* further page access. |
| 342 |
*/ |
| 343 |
page_offline_freeze(); |
| 344 |
|
| 345 |
phdrs_offset = sizeof(struct elfhdr); |
| 346 |
notes_offset = phdrs_offset + kcore_phdrs_len; |
| 347 |
|
| 348 |
/* ELF file header. */ |
| 349 |
if (buflen && *fpos < sizeof(struct elfhdr)) { |
| 350 |
struct elfhdr ehdr = { |
| 351 |
.e_ident = { |
| 352 |
[EI_MAG0] = ELFMAG0, |
| 353 |
[EI_MAG1] = ELFMAG1, |
| 354 |
[EI_MAG2] = ELFMAG2, |
| 355 |
[EI_MAG3] = ELFMAG3, |
| 356 |
[EI_CLASS] = ELF_CLASS, |
| 357 |
[EI_DATA] = ELF_DATA, |
| 358 |
[EI_VERSION] = EV_CURRENT, |
| 359 |
[EI_OSABI] = ELF_OSABI, |
| 360 |
}, |
| 361 |
.e_type = ET_CORE, |
| 362 |
.e_machine = ELF_ARCH, |
| 363 |
.e_version = EV_CURRENT, |
| 364 |
.e_phoff = sizeof(struct elfhdr), |
| 365 |
.e_flags = ELF_CORE_EFLAGS, |
| 366 |
.e_ehsize = sizeof(struct elfhdr), |
| 367 |
.e_phentsize = sizeof(struct elf_phdr), |
| 368 |
.e_phnum = kcore_nphdr, |
| 369 |
}; |
| 370 |
|
| 371 |
tsz = min_t(size_t, buflen, sizeof(struct elfhdr) - *fpos); |
| 372 |
if (copy_to_iter((char *)&ehdr + *fpos, tsz, iter) != tsz) { |
| 373 |
ret = -EFAULT; |
| 374 |
goto out; |
| 375 |
} |
| 376 |
|
| 377 |
buflen -= tsz; |
| 378 |
*fpos += tsz; |
| 379 |
} |
| 380 |
|
| 381 |
/* ELF program headers. */ |
| 382 |
if (buflen && *fpos < phdrs_offset + kcore_phdrs_len) { |
| 383 |
struct elf_phdr *phdrs, *phdr; |
| 384 |
|
| 385 |
phdrs = kzalloc(kcore_phdrs_len, GFP_KERNEL); |
| 386 |
if (!phdrs) { |
| 387 |
ret = -ENOMEM; |
| 388 |
goto out; |
| 389 |
} |
| 390 |
|
| 391 |
phdrs[0].p_type = PT_NOTE; |
| 392 |
phdrs[0].p_offset = notes_offset; |
| 393 |
phdrs[0].p_filesz = kcore_notes_len; |
| 394 |
|
| 395 |
phdr = &phdrs[1]; |
| 396 |
list_for_each_entry(m, &kclist_head, list) { |
| 397 |
phdr->p_type = PT_LOAD; |
| 398 |
phdr->p_flags = PF_R | PF_W | PF_X; |
| 399 |
phdr->p_offset = kc_vaddr_to_offset(m->addr) |
| 400 |
+ kcore_data_offset; |
| 401 |
phdr->p_vaddr = (size_t)m->addr; |
| 402 |
if (m->type == KCORE_RAM) |
| 403 |
phdr->p_paddr = __pa(m->addr); |
| 404 |
else if (m->type == KCORE_TEXT) |
| 405 |
phdr->p_paddr = __pa_symbol(m->addr); |
| 406 |
else |
| 407 |
phdr->p_paddr = (elf_addr_t)-1; |
| 408 |
phdr->p_filesz = phdr->p_memsz = m->size; |
| 409 |
phdr->p_align = PAGE_SIZE; |
| 410 |
phdr++; |
| 411 |
} |
| 412 |
|
| 413 |
tsz = min_t(size_t, buflen, |
| 414 |
phdrs_offset + kcore_phdrs_len - *fpos); |
| 415 |
if (copy_to_iter((char *)phdrs + *fpos - phdrs_offset, tsz, |
| 416 |
iter) != tsz) { |
| 417 |
kfree(phdrs); |
| 418 |
ret = -EFAULT; |
| 419 |
goto out; |
| 420 |
} |
| 421 |
kfree(phdrs); |
| 422 |
|
| 423 |
buflen -= tsz; |
| 424 |
*fpos += tsz; |
| 425 |
} |
| 426 |
|
| 427 |
/* ELF note segment. */ |
| 428 |
if (buflen && *fpos < notes_offset + kcore_notes_len) { |
| 429 |
struct elf_prstatus prstatus = {}; |
| 430 |
struct elf_prpsinfo prpsinfo = { |
| 431 |
.pr_sname = 'R', |
| 432 |
.pr_fname = "vmlinux", |
| 433 |
}; |
| 434 |
char *notes; |
| 435 |
size_t i = 0; |
| 436 |
|
| 437 |
strscpy(prpsinfo.pr_psargs, saved_command_line, |
| 438 |
sizeof(prpsinfo.pr_psargs)); |
| 439 |
|
| 440 |
notes = kzalloc(kcore_notes_len, GFP_KERNEL); |
| 441 |
if (!notes) { |
| 442 |
ret = -ENOMEM; |
| 443 |
goto out; |
| 444 |
} |
| 445 |
|
| 446 |
append_kcore_note(notes, &i, NN_PRSTATUS, NT_PRSTATUS, &prstatus, |
| 447 |
sizeof(prstatus)); |
| 448 |
append_kcore_note(notes, &i, NN_PRPSINFO, NT_PRPSINFO, &prpsinfo, |
| 449 |
sizeof(prpsinfo)); |
| 450 |
append_kcore_note(notes, &i, NN_TASKSTRUCT, NT_TASKSTRUCT, current, |
| 451 |
arch_task_struct_size); |
| 452 |
/* |
| 453 |
* vmcoreinfo_size is mostly constant after init time, but it |
| 454 |
* can be changed by crash_save_vmcoreinfo(). Racing here with a |
| 455 |
* panic on another CPU before the machine goes down is insanely |
| 456 |
* unlikely, but it's better to not leave potential buffer |
| 457 |
* overflows lying around, regardless. |
| 458 |
*/ |
| 459 |
append_kcore_note(notes, &i, VMCOREINFO_NOTE_NAME, 0, |
| 460 |
vmcoreinfo_data, |
| 461 |
min(vmcoreinfo_size, kcore_notes_len - i)); |
| 462 |
|
| 463 |
tsz = min_t(size_t, buflen, |
| 464 |
notes_offset + kcore_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 - kcore_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 |
} |
| 620 |
skip: |
| 621 |
buflen -= tsz; |
| 622 |
*fpos += tsz; |
| 623 |
start += tsz; |
| 624 |
tsz = (buflen > PAGE_SIZE ? PAGE_SIZE : buflen); |
| 625 |
} |
| 626 |
|
| 627 |
out: |
| 628 |
page_offline_thaw(); |
| 629 |
percpu_up_read(&kclist_lock); |
| 630 |
if (ret) |
| 631 |
return ret; |
| 632 |
return orig_buflen - buflen; |
| 633 |
} |
| 634 |
|
| 635 |
static 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 |
|
| 659 |
static int release_kcore(struct inode *inode, struct file *file) |
| 660 |
{ |
| 661 |
kfree(file->private_data); |
| 662 |
return 0; |
| 663 |
} |
| 664 |
|
| 665 |
static const struct proc_ops kcore_proc_ops = { |
| 666 |
.proc_flags = PROC_ENTRY_PERMANENT, |
| 667 |
.proc_read_iter = read_kcore_iter, |
| 668 |
.proc_open = open_kcore, |
| 669 |
.proc_release = release_kcore, |
| 670 |
.proc_lseek = default_llseek, |
| 671 |
}; |
| 672 |
|
| 673 |
/* just remember that we have to update kcore */ |
| 674 |
static int __meminit kcore_callback(struct notifier_block *self, |
| 675 |
unsigned long action, void *arg) |
| 676 |
{ |
| 677 |
switch (action) { |
| 678 |
case MEM_ONLINE: |
| 679 |
case MEM_OFFLINE: |
| 680 |
kcore_need_update = 1; |
| 681 |
break; |
| 682 |
} |
| 683 |
return NOTIFY_OK; |
| 684 |
} |
| 685 |
|
| 686 |
|
| 687 |
static struct kcore_list kcore_vmalloc; |
| 688 |
|
| 689 |
#ifdef CONFIG_ARCH_PROC_KCORE_TEXT |
| 690 |
static struct kcore_list kcore_text; |
| 691 |
/* |
| 692 |
* If defined, special segment is used for mapping kernel text instead of |
| 693 |
* direct-map area. We need to create special TEXT section. |
| 694 |
*/ |
| 695 |
static void __init proc_kcore_text_init(void) |
| 696 |
{ |
| 697 |
kclist_add(&kcore_text, _text, _end - _text, KCORE_TEXT); |
| 698 |
} |
| 699 |
#else |
| 700 |
static void __init proc_kcore_text_init(void) |
| 701 |
{ |
| 702 |
} |
| 703 |
#endif |
| 704 |
|
| 705 |
#if defined(CONFIG_MODULES) && defined(MODULES_VADDR) |
| 706 |
/* |
| 707 |
* MODULES_VADDR has no intersection with VMALLOC_ADDR. |
| 708 |
*/ |
| 709 |
static struct kcore_list kcore_modules; |
| 710 |
static void __init add_modules_range(void) |
| 711 |
{ |
| 712 |
if (MODULES_VADDR != VMALLOC_START && MODULES_END != VMALLOC_END) { |
| 713 |
kclist_add(&kcore_modules, (void *)MODULES_VADDR, |
| 714 |
MODULES_END - MODULES_VADDR, KCORE_VMALLOC); |
| 715 |
} |
| 716 |
} |
| 717 |
#else |
| 718 |
static void __init add_modules_range(void) |
| 719 |
{ |
| 720 |
} |
| 721 |
#endif |
| 722 |
|
| 723 |
static int __init proc_kcore_init(void) |
| 724 |
{ |
| 725 |
proc_root_kcore = proc_create("kcore", S_IRUSR, NULL, &kcore_proc_ops); |
| 726 |
if (!proc_root_kcore) { |
| 727 |
pr_err("couldn't create /proc/kcore\n"); |
| 728 |
return 0; /* Always returns 0. */ |
| 729 |
} |
| 730 |
/* Store text area if it's special */ |
| 731 |
proc_kcore_text_init(); |
| 732 |
/* Store vmalloc area */ |
| 733 |
kclist_add(&kcore_vmalloc, (void *)VMALLOC_START, |
| 734 |
VMALLOC_END - VMALLOC_START, KCORE_VMALLOC); |
| 735 |
add_modules_range(); |
| 736 |
/* Store direct-map area from physical memory map */ |
| 737 |
kcore_update_ram(); |
| 738 |
hotplug_memory_notifier(kcore_callback, DEFAULT_CALLBACK_PRI); |
| 739 |
|
| 740 |
return 0; |
| 741 |
} |
| 742 |
fs_initcall(proc_kcore_init); |