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/mm.h> 14 #include <linux/proc_fs.h> 15 #include <linux/kcore.h> 16 #include <linux/user.h> 17 #include <linux/capability.h> 18 #include <linux/elf.h> 19 #include <linux/elfcore.h> 20 #include <linux/notifier.h> 21 #include <linux/vmalloc.h> 22 #include <linux/highmem.h> 23 #include <linux/printk.h> 24 #include <linux/bootmem.h> 25 #include <linux/init.h> 26 #include <linux/slab.h> 27 #include <linux/uaccess.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 <asm/sections.h> 34 #include "internal.h" 35 36 #define CORE_STR "CORE" 37 38 #ifndef ELF_CORE_EFLAGS 39 #define ELF_CORE_EFLAGS 0 40 #endif 41 42 static struct proc_dir_entry *proc_root_kcore; 43 44 45 #ifndef kc_vaddr_to_offset 46 #define kc_vaddr_to_offset(v) ((v) - PAGE_OFFSET) 47 #endif 48 #ifndef kc_offset_to_vaddr 49 #define kc_offset_to_vaddr(o) ((o) + PAGE_OFFSET) 50 #endif 51 52 /* An ELF note in memory */ 53 struct memelfnote 54 { 55 const char *name; 56 int type; 57 unsigned int datasz; 58 void *data; 59 }; 60 61 static LIST_HEAD(kclist_head); 62 static DEFINE_RWLOCK(kclist_lock); 63 static int kcore_need_update = 1; 64 65 void 66 kclist_add(struct kcore_list *new, void *addr, size_t size, int type) 67 { 68 new->addr = (unsigned long)addr; 69 new->size = size; 70 new->type = type; 71 72 write_lock(&kclist_lock); 73 list_add_tail(&new->list, &kclist_head); 74 write_unlock(&kclist_lock); 75 } 76 77 static size_t get_kcore_size(int *nphdr, size_t *elf_buflen) 78 { 79 size_t try, size; 80 struct kcore_list *m; 81 82 *nphdr = 1; /* PT_NOTE */ 83 size = 0; 84 85 list_for_each_entry(m, &kclist_head, list) { 86 try = kc_vaddr_to_offset((size_t)m->addr + m->size); 87 if (try > size) 88 size = try; 89 *nphdr = *nphdr + 1; 90 } 91 *elf_buflen = sizeof(struct elfhdr) + 92 (*nphdr + 2)*sizeof(struct elf_phdr) + 93 3 * ((sizeof(struct elf_note)) + 94 roundup(sizeof(CORE_STR), 4)) + 95 roundup(sizeof(struct elf_prstatus), 4) + 96 roundup(sizeof(struct elf_prpsinfo), 4) + 97 roundup(arch_task_struct_size, 4); 98 *elf_buflen = PAGE_ALIGN(*elf_buflen); 99 return size + *elf_buflen; 100 } 101 102 static void free_kclist_ents(struct list_head *head) 103 { 104 struct kcore_list *tmp, *pos; 105 106 list_for_each_entry_safe(pos, tmp, head, list) { 107 list_del(&pos->list); 108 kfree(pos); 109 } 110 } 111 /* 112 * Replace all KCORE_RAM/KCORE_VMEMMAP information with passed list. 113 */ 114 static void __kcore_update_ram(struct list_head *list) 115 { 116 int nphdr; 117 size_t size; 118 struct kcore_list *tmp, *pos; 119 LIST_HEAD(garbage); 120 121 write_lock(&kclist_lock); 122 if (kcore_need_update) { 123 list_for_each_entry_safe(pos, tmp, &kclist_head, list) { 124 if (pos->type == KCORE_RAM 125 || pos->type == KCORE_VMEMMAP) 126 list_move(&pos->list, &garbage); 127 } 128 list_splice_tail(list, &kclist_head); 129 } else 130 list_splice(list, &garbage); 131 kcore_need_update = 0; 132 proc_root_kcore->size = get_kcore_size(&nphdr, &size); 133 write_unlock(&kclist_lock); 134 135 free_kclist_ents(&garbage); 136 } 137 138 139 #ifdef CONFIG_HIGHMEM 140 /* 141 * If no highmem, we can assume [0...max_low_pfn) continuous range of memory 142 * because memory hole is not as big as !HIGHMEM case. 143 * (HIGHMEM is special because part of memory is _invisible_ from the kernel.) 144 */ 145 static int kcore_update_ram(void) 146 { 147 LIST_HEAD(head); 148 struct kcore_list *ent; 149 int ret = 0; 150 151 ent = kmalloc(sizeof(*ent), GFP_KERNEL); 152 if (!ent) 153 return -ENOMEM; 154 ent->addr = (unsigned long)__va(0); 155 ent->size = max_low_pfn << PAGE_SHIFT; 156 ent->type = KCORE_RAM; 157 list_add(&ent->list, &head); 158 __kcore_update_ram(&head); 159 return ret; 160 } 161 162 #else /* !CONFIG_HIGHMEM */ 163 164 #ifdef CONFIG_SPARSEMEM_VMEMMAP 165 /* calculate vmemmap's address from given system ram pfn and register it */ 166 static int 167 get_sparsemem_vmemmap_info(struct kcore_list *ent, struct list_head *head) 168 { 169 unsigned long pfn = __pa(ent->addr) >> PAGE_SHIFT; 170 unsigned long nr_pages = ent->size >> PAGE_SHIFT; 171 unsigned long start, end; 172 struct kcore_list *vmm, *tmp; 173 174 175 start = ((unsigned long)pfn_to_page(pfn)) & PAGE_MASK; 176 end = ((unsigned long)pfn_to_page(pfn + nr_pages)) - 1; 177 end = PAGE_ALIGN(end); 178 /* overlap check (because we have to align page */ 179 list_for_each_entry(tmp, head, list) { 180 if (tmp->type != KCORE_VMEMMAP) 181 continue; 182 if (start < tmp->addr + tmp->size) 183 if (end > tmp->addr) 184 end = tmp->addr; 185 } 186 if (start < end) { 187 vmm = kmalloc(sizeof(*vmm), GFP_KERNEL); 188 if (!vmm) 189 return 0; 190 vmm->addr = start; 191 vmm->size = end - start; 192 vmm->type = KCORE_VMEMMAP; 193 list_add_tail(&vmm->list, head); 194 } 195 return 1; 196 197 } 198 #else 199 static int 200 get_sparsemem_vmemmap_info(struct kcore_list *ent, struct list_head *head) 201 { 202 return 1; 203 } 204 205 #endif 206 207 static int 208 kclist_add_private(unsigned long pfn, unsigned long nr_pages, void *arg) 209 { 210 struct list_head *head = (struct list_head *)arg; 211 struct kcore_list *ent; 212 struct page *p; 213 214 if (!pfn_valid(pfn)) 215 return 1; 216 217 p = pfn_to_page(pfn); 218 if (!memmap_valid_within(pfn, p, page_zone(p))) 219 return 1; 220 221 ent = kmalloc(sizeof(*ent), GFP_KERNEL); 222 if (!ent) 223 return -ENOMEM; 224 ent->addr = (unsigned long)page_to_virt(p); 225 ent->size = nr_pages << PAGE_SHIFT; 226 227 if (!virt_addr_valid(ent->addr)) 228 goto free_out; 229 230 /* cut not-mapped area. ....from ppc-32 code. */ 231 if (ULONG_MAX - ent->addr < ent->size) 232 ent->size = ULONG_MAX - ent->addr; 233 234 /* 235 * We've already checked virt_addr_valid so we know this address 236 * is a valid pointer, therefore we can check against it to determine 237 * if we need to trim 238 */ 239 if (VMALLOC_START > ent->addr) { 240 if (VMALLOC_START - ent->addr < ent->size) 241 ent->size = VMALLOC_START - ent->addr; 242 } 243 244 ent->type = KCORE_RAM; 245 list_add_tail(&ent->list, head); 246 247 if (!get_sparsemem_vmemmap_info(ent, head)) { 248 list_del(&ent->list); 249 goto free_out; 250 } 251 252 return 0; 253 free_out: 254 kfree(ent); 255 return 1; 256 } 257 258 static int kcore_update_ram(void) 259 { 260 int nid, ret; 261 unsigned long end_pfn; 262 LIST_HEAD(head); 263 264 /* Not inialized....update now */ 265 /* find out "max pfn" */ 266 end_pfn = 0; 267 for_each_node_state(nid, N_MEMORY) { 268 unsigned long node_end; 269 node_end = node_end_pfn(nid); 270 if (end_pfn < node_end) 271 end_pfn = node_end; 272 } 273 /* scan 0 to max_pfn */ 274 ret = walk_system_ram_range(0, end_pfn, &head, kclist_add_private); 275 if (ret) { 276 free_kclist_ents(&head); 277 return -ENOMEM; 278 } 279 __kcore_update_ram(&head); 280 return ret; 281 } 282 #endif /* CONFIG_HIGHMEM */ 283 284 /*****************************************************************************/ 285 /* 286 * determine size of ELF note 287 */ 288 static int notesize(struct memelfnote *en) 289 { 290 int sz; 291 292 sz = sizeof(struct elf_note); 293 sz += roundup((strlen(en->name) + 1), 4); 294 sz += roundup(en->datasz, 4); 295 296 return sz; 297 } /* end notesize() */ 298 299 /*****************************************************************************/ 300 /* 301 * store a note in the header buffer 302 */ 303 static char *storenote(struct memelfnote *men, char *bufp) 304 { 305 struct elf_note en; 306 307 #define DUMP_WRITE(addr,nr) do { memcpy(bufp,addr,nr); bufp += nr; } while(0) 308 309 en.n_namesz = strlen(men->name) + 1; 310 en.n_descsz = men->datasz; 311 en.n_type = men->type; 312 313 DUMP_WRITE(&en, sizeof(en)); 314 DUMP_WRITE(men->name, en.n_namesz); 315 316 /* XXX - cast from long long to long to avoid need for libgcc.a */ 317 bufp = (char*) roundup((unsigned long)bufp,4); 318 DUMP_WRITE(men->data, men->datasz); 319 bufp = (char*) roundup((unsigned long)bufp,4); 320 321 #undef DUMP_WRITE 322 323 return bufp; 324 } /* end storenote() */ 325 326 /* 327 * store an ELF coredump header in the supplied buffer 328 * nphdr is the number of elf_phdr to insert 329 */ 330 static void elf_kcore_store_hdr(char *bufp, int nphdr, int dataoff) 331 { 332 struct elf_prstatus prstatus; /* NT_PRSTATUS */ 333 struct elf_prpsinfo prpsinfo; /* NT_PRPSINFO */ 334 struct elf_phdr *nhdr, *phdr; 335 struct elfhdr *elf; 336 struct memelfnote notes[3]; 337 off_t offset = 0; 338 struct kcore_list *m; 339 340 /* setup ELF header */ 341 elf = (struct elfhdr *) bufp; 342 bufp += sizeof(struct elfhdr); 343 offset += sizeof(struct elfhdr); 344 memcpy(elf->e_ident, ELFMAG, SELFMAG); 345 elf->e_ident[EI_CLASS] = ELF_CLASS; 346 elf->e_ident[EI_DATA] = ELF_DATA; 347 elf->e_ident[EI_VERSION]= EV_CURRENT; 348 elf->e_ident[EI_OSABI] = ELF_OSABI; 349 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD); 350 elf->e_type = ET_CORE; 351 elf->e_machine = ELF_ARCH; 352 elf->e_version = EV_CURRENT; 353 elf->e_entry = 0; 354 elf->e_phoff = sizeof(struct elfhdr); 355 elf->e_shoff = 0; 356 elf->e_flags = ELF_CORE_EFLAGS; 357 elf->e_ehsize = sizeof(struct elfhdr); 358 elf->e_phentsize= sizeof(struct elf_phdr); 359 elf->e_phnum = nphdr; 360 elf->e_shentsize= 0; 361 elf->e_shnum = 0; 362 elf->e_shstrndx = 0; 363 364 /* setup ELF PT_NOTE program header */ 365 nhdr = (struct elf_phdr *) bufp; 366 bufp += sizeof(struct elf_phdr); 367 offset += sizeof(struct elf_phdr); 368 nhdr->p_type = PT_NOTE; 369 nhdr->p_offset = 0; 370 nhdr->p_vaddr = 0; 371 nhdr->p_paddr = 0; 372 nhdr->p_filesz = 0; 373 nhdr->p_memsz = 0; 374 nhdr->p_flags = 0; 375 nhdr->p_align = 0; 376 377 /* setup ELF PT_LOAD program header for every area */ 378 list_for_each_entry(m, &kclist_head, list) { 379 phdr = (struct elf_phdr *) bufp; 380 bufp += sizeof(struct elf_phdr); 381 offset += sizeof(struct elf_phdr); 382 383 phdr->p_type = PT_LOAD; 384 phdr->p_flags = PF_R|PF_W|PF_X; 385 phdr->p_offset = kc_vaddr_to_offset(m->addr) + dataoff; 386 if (m->type == KCORE_REMAP) 387 phdr->p_vaddr = (size_t)m->vaddr; 388 else 389 phdr->p_vaddr = (size_t)m->addr; 390 if (m->type == KCORE_RAM || m->type == KCORE_TEXT || m->type == KCORE_REMAP) 391 phdr->p_paddr = __pa(m->addr); 392 else 393 phdr->p_paddr = (elf_addr_t)-1; 394 phdr->p_filesz = phdr->p_memsz = m->size; 395 phdr->p_align = PAGE_SIZE; 396 } 397 398 /* 399 * Set up the notes in similar form to SVR4 core dumps made 400 * with info from their /proc. 401 */ 402 nhdr->p_offset = offset; 403 404 /* set up the process status */ 405 notes[0].name = CORE_STR; 406 notes[0].type = NT_PRSTATUS; 407 notes[0].datasz = sizeof(struct elf_prstatus); 408 notes[0].data = &prstatus; 409 410 memset(&prstatus, 0, sizeof(struct elf_prstatus)); 411 412 nhdr->p_filesz = notesize(¬es[0]); 413 bufp = storenote(¬es[0], bufp); 414 415 /* set up the process info */ 416 notes[1].name = CORE_STR; 417 notes[1].type = NT_PRPSINFO; 418 notes[1].datasz = sizeof(struct elf_prpsinfo); 419 notes[1].data = &prpsinfo; 420 421 memset(&prpsinfo, 0, sizeof(struct elf_prpsinfo)); 422 prpsinfo.pr_state = 0; 423 prpsinfo.pr_sname = 'R'; 424 prpsinfo.pr_zomb = 0; 425 426 strcpy(prpsinfo.pr_fname, "vmlinux"); 427 strlcpy(prpsinfo.pr_psargs, saved_command_line, sizeof(prpsinfo.pr_psargs)); 428 429 nhdr->p_filesz += notesize(¬es[1]); 430 bufp = storenote(¬es[1], bufp); 431 432 /* set up the task structure */ 433 notes[2].name = CORE_STR; 434 notes[2].type = NT_TASKSTRUCT; 435 notes[2].datasz = arch_task_struct_size; 436 notes[2].data = current; 437 438 nhdr->p_filesz += notesize(¬es[2]); 439 bufp = storenote(¬es[2], bufp); 440 441 } /* end elf_kcore_store_hdr() */ 442 443 /*****************************************************************************/ 444 /* 445 * read from the ELF header and then kernel memory 446 */ 447 static ssize_t 448 read_kcore(struct file *file, char __user *buffer, size_t buflen, loff_t *fpos) 449 { 450 char *buf = file->private_data; 451 ssize_t acc = 0; 452 size_t size, tsz; 453 size_t elf_buflen; 454 int nphdr; 455 unsigned long start; 456 457 read_lock(&kclist_lock); 458 size = get_kcore_size(&nphdr, &elf_buflen); 459 460 if (buflen == 0 || *fpos >= size) { 461 read_unlock(&kclist_lock); 462 return 0; 463 } 464 465 /* trim buflen to not go beyond EOF */ 466 if (buflen > size - *fpos) 467 buflen = size - *fpos; 468 469 /* construct an ELF core header if we'll need some of it */ 470 if (*fpos < elf_buflen) { 471 char * elf_buf; 472 473 tsz = elf_buflen - *fpos; 474 if (buflen < tsz) 475 tsz = buflen; 476 elf_buf = kzalloc(elf_buflen, GFP_ATOMIC); 477 if (!elf_buf) { 478 read_unlock(&kclist_lock); 479 return -ENOMEM; 480 } 481 elf_kcore_store_hdr(elf_buf, nphdr, elf_buflen); 482 read_unlock(&kclist_lock); 483 if (copy_to_user(buffer, elf_buf + *fpos, tsz)) { 484 kfree(elf_buf); 485 return -EFAULT; 486 } 487 kfree(elf_buf); 488 buflen -= tsz; 489 *fpos += tsz; 490 buffer += tsz; 491 acc += tsz; 492 493 /* leave now if filled buffer already */ 494 if (buflen == 0) 495 return acc; 496 } else 497 read_unlock(&kclist_lock); 498 499 /* 500 * Check to see if our file offset matches with any of 501 * the addresses in the elf_phdr on our list. 502 */ 503 start = kc_offset_to_vaddr(*fpos - elf_buflen); 504 if ((tsz = (PAGE_SIZE - (start & ~PAGE_MASK))) > buflen) 505 tsz = buflen; 506 507 while (buflen) { 508 struct kcore_list *m; 509 510 read_lock(&kclist_lock); 511 list_for_each_entry(m, &kclist_head, list) { 512 if (start >= m->addr && start < (m->addr+m->size)) 513 break; 514 } 515 read_unlock(&kclist_lock); 516 517 if (&m->list == &kclist_head) { 518 if (clear_user(buffer, tsz)) 519 return -EFAULT; 520 } else if (m->type == KCORE_VMALLOC) { 521 vread(buf, (char *)start, tsz); 522 /* we have to zero-fill user buffer even if no read */ 523 if (copy_to_user(buffer, buf, tsz)) 524 return -EFAULT; 525 } else if (m->type == KCORE_USER) { 526 /* User page is handled prior to normal kernel page: */ 527 if (copy_to_user(buffer, (char *)start, tsz)) 528 return -EFAULT; 529 } else { 530 if (kern_addr_valid(start)) { 531 /* 532 * Using bounce buffer to bypass the 533 * hardened user copy kernel text checks. 534 */ 535 if (probe_kernel_read(buf, (void *) start, tsz)) { 536 if (clear_user(buffer, tsz)) 537 return -EFAULT; 538 } else { 539 if (copy_to_user(buffer, buf, tsz)) 540 return -EFAULT; 541 } 542 } else { 543 if (clear_user(buffer, tsz)) 544 return -EFAULT; 545 } 546 } 547 buflen -= tsz; 548 *fpos += tsz; 549 buffer += tsz; 550 acc += tsz; 551 start += tsz; 552 tsz = (buflen > PAGE_SIZE ? PAGE_SIZE : buflen); 553 } 554 555 return acc; 556 } 557 558 559 static int open_kcore(struct inode *inode, struct file *filp) 560 { 561 if (!capable(CAP_SYS_RAWIO)) 562 return -EPERM; 563 564 filp->private_data = kmalloc(PAGE_SIZE, GFP_KERNEL); 565 if (!filp->private_data) 566 return -ENOMEM; 567 568 if (kcore_need_update) 569 kcore_update_ram(); 570 if (i_size_read(inode) != proc_root_kcore->size) { 571 inode_lock(inode); 572 i_size_write(inode, proc_root_kcore->size); 573 inode_unlock(inode); 574 } 575 return 0; 576 } 577 578 static int release_kcore(struct inode *inode, struct file *file) 579 { 580 kfree(file->private_data); 581 return 0; 582 } 583 584 static const struct file_operations proc_kcore_operations = { 585 .read = read_kcore, 586 .open = open_kcore, 587 .release = release_kcore, 588 .llseek = default_llseek, 589 }; 590 591 /* just remember that we have to update kcore */ 592 static int __meminit kcore_callback(struct notifier_block *self, 593 unsigned long action, void *arg) 594 { 595 switch (action) { 596 case MEM_ONLINE: 597 case MEM_OFFLINE: 598 write_lock(&kclist_lock); 599 kcore_need_update = 1; 600 write_unlock(&kclist_lock); 601 } 602 return NOTIFY_OK; 603 } 604 605 static struct notifier_block kcore_callback_nb __meminitdata = { 606 .notifier_call = kcore_callback, 607 .priority = 0, 608 }; 609 610 static struct kcore_list kcore_vmalloc; 611 612 #ifdef CONFIG_ARCH_PROC_KCORE_TEXT 613 static struct kcore_list kcore_text; 614 /* 615 * If defined, special segment is used for mapping kernel text instead of 616 * direct-map area. We need to create special TEXT section. 617 */ 618 static void __init proc_kcore_text_init(void) 619 { 620 kclist_add(&kcore_text, _text, _end - _text, KCORE_TEXT); 621 } 622 #else 623 static void __init proc_kcore_text_init(void) 624 { 625 } 626 #endif 627 628 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR) 629 /* 630 * MODULES_VADDR has no intersection with VMALLOC_ADDR. 631 */ 632 struct kcore_list kcore_modules; 633 static void __init add_modules_range(void) 634 { 635 if (MODULES_VADDR != VMALLOC_START && MODULES_END != VMALLOC_END) { 636 kclist_add(&kcore_modules, (void *)MODULES_VADDR, 637 MODULES_END - MODULES_VADDR, KCORE_VMALLOC); 638 } 639 } 640 #else 641 static void __init add_modules_range(void) 642 { 643 } 644 #endif 645 646 static int __init proc_kcore_init(void) 647 { 648 proc_root_kcore = proc_create("kcore", S_IRUSR, NULL, 649 &proc_kcore_operations); 650 if (!proc_root_kcore) { 651 pr_err("couldn't create /proc/kcore\n"); 652 return 0; /* Always returns 0. */ 653 } 654 /* Store text area if it's special */ 655 proc_kcore_text_init(); 656 /* Store vmalloc area */ 657 kclist_add(&kcore_vmalloc, (void *)VMALLOC_START, 658 VMALLOC_END - VMALLOC_START, KCORE_VMALLOC); 659 add_modules_range(); 660 /* Store direct-map area from physical memory map */ 661 kcore_update_ram(); 662 register_hotmemory_notifier(&kcore_callback_nb); 663 664 return 0; 665 } 666 fs_initcall(proc_kcore_init); 667