fs/proc/vmcore.c

   1 /*
   2  *      fs/proc/vmcore.c Interface for accessing the crash
   3  *                               dump from the system's previous life.
   4  *      Heavily borrowed from fs/proc/kcore.c
   5  *      Created by: Hariprasad Nellitheertha (hari@in.ibm.com)
   6  *      Copyright (C) IBM Corporation, 2004. All rights reserved
   7  *
   8  */
   9
  10 #include <linux/mm.h>
  11 #include <linux/kcore.h>
  12 #include <linux/user.h>
  13 #include <linux/elf.h>
  14 #include <linux/elfcore.h>
  15 #include <linux/export.h>
  16 #include <linux/slab.h>
  17 #include <linux/highmem.h>
  18 #include <linux/printk.h>
  19 #include <linux/bootmem.h>
  20 #include <linux/init.h>
  21 #include <linux/crash_dump.h>
  22 #include <linux/list.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/pagemap.h>
  25 #include <asm/uaccess.h>
  26 #include <asm/io.h>
  27 #include "internal.h"
  28
  29 /* List representing chunks of contiguous memory areas and their offsets in
  30  * vmcore file.
  31  */
  32 static LIST_HEAD(vmcore_list);
  33
  34 /* Stores the pointer to the buffer containing kernel elf core headers. */
  35 static char *elfcorebuf;
  36 static size_t elfcorebuf_sz;
  37 static size_t elfcorebuf_sz_orig;
  38
  39 static char *elfnotes_buf;
  40 static size_t elfnotes_sz;
  41
  42 /* Total size of vmcore file. */
  43 static u64 vmcore_size;
  44
  45 static struct proc_dir_entry *proc_vmcore;
  46
  47 /*
  48  * Returns > 0 for RAM pages, 0 for non-RAM pages, < 0 on error
  49  * The called function has to take care of module refcounting.
  50  */
  51 static int (*oldmem_pfn_is_ram)(unsigned long pfn);
  52
  53 int register_oldmem_pfn_is_ram(int (*fn)(unsigned long pfn))
  54 {
  55         if (oldmem_pfn_is_ram)
  56                 return -EBUSY;
  57         oldmem_pfn_is_ram = fn;
  58         return 0;
  59 }
  60 EXPORT_SYMBOL_GPL(register_oldmem_pfn_is_ram);
  61
  62 void unregister_oldmem_pfn_is_ram(void)
  63 {
  64         oldmem_pfn_is_ram = NULL;
  65         wmb();
  66 }
  67 EXPORT_SYMBOL_GPL(unregister_oldmem_pfn_is_ram);
  68
  69 static int pfn_is_ram(unsigned long pfn)
  70 {
  71         int (*fn)(unsigned long pfn);
  72         /* pfn is ram unless fn() checks pagetype */
  73         int ret = 1;
  74
  75         /*
  76          * Ask hypervisor if the pfn is really ram.
  77          * A ballooned page contains no data and reading from such a page
  78          * will cause high load in the hypervisor.
  79          */
  80         fn = oldmem_pfn_is_ram;
  81         if (fn)
  82                 ret = fn(pfn);
  83
  84         return ret;
  85 }
  86
  87 /* Reads a page from the oldmem device from given offset. */
  88 static ssize_t read_from_oldmem(char *buf, size_t count,
  89                                 u64 *ppos, int userbuf)
  90 {
  91         unsigned long pfn, offset;
  92         size_t nr_bytes;
  93         ssize_t read = 0, tmp;
  94
  95         if (!count)
  96                 return 0;
  97
  98         offset = (unsigned long)(*ppos % PAGE_SIZE);
  99         pfn = (unsigned long)(*ppos / PAGE_SIZE);
 100
 101         do {
 102                 if (count > (PAGE_SIZE - offset))
 103                         nr_bytes = PAGE_SIZE - offset;
 104                 else
 105                         nr_bytes = count;
 106
 107                 /* If pfn is not ram, return zeros for sparse dump files */
 108                 if (pfn_is_ram(pfn) == 0)
 109                         memset(buf, 0, nr_bytes);
 110                 else {
 111                         tmp = copy_oldmem_page(pfn, buf, nr_bytes,
 112                                                 offset, userbuf);
 113                         if (tmp < 0)
 114                                 return tmp;
 115                 }
 116                 *ppos += nr_bytes;
 117                 count -= nr_bytes;
 118                 buf += nr_bytes;
 119                 read += nr_bytes;
 120                 ++pfn;
 121                 offset = 0;
 122         } while (count);
 123
 124         return read;
 125 }
 126
 127 /*
 128  * Architectures may override this function to allocate ELF header in 2nd kernel
 129  */
 130 int __weak elfcorehdr_alloc(unsigned long long *addr, unsigned long long *size)
 131 {
 132         return 0;
 133 }
 134
 135 /*
 136  * Architectures may override this function to free header
 137  */
 138 void __weak elfcorehdr_free(unsigned long long addr)
 139 {}
 140
 141 /*
 142  * Architectures may override this function to read from ELF header
 143  */
 144 ssize_t __weak elfcorehdr_read(char *buf, size_t count, u64 *ppos)
 145 {
 146         return read_from_oldmem(buf, count, ppos, 0);
 147 }
 148
 149 /*
 150  * Architectures may override this function to read from notes sections
 151  */
 152 ssize_t __weak elfcorehdr_read_notes(char *buf, size_t count, u64 *ppos)
 153 {
 154         return read_from_oldmem(buf, count, ppos, 0);
 155 }
 156
 157 /*
 158  * Architectures may override this function to map oldmem
 159  */
 160 int __weak remap_oldmem_pfn_range(struct vm_area_struct *vma,
 161                                   unsigned long from, unsigned long pfn,
 162                                   unsigned long size, pgprot_t prot)
 163 {
 164         return remap_pfn_range(vma, from, pfn, size, prot);
 165 }
 166
 167 /*
 168  * Copy to either kernel or user space
 169  */
 170 static int copy_to(void *target, void *src, size_t size, int userbuf)
 171 {
 172         if (userbuf) {
 173                 if (copy_to_user((char __user *) target, src, size))
 174                         return -EFAULT;
 175         } else {
 176                 memcpy(target, src, size);
 177         }
 178         return 0;
 179 }
 180
 181 /* Read from the ELF header and then the crash dump. On error, negative value is
 182  * returned otherwise number of bytes read are returned.
 183  */
 184 static ssize_t __read_vmcore(char *buffer, size_t buflen, loff_t *fpos,
 185                              int userbuf)
 186 {
 187         ssize_t acc = 0, tmp;
 188         size_t tsz;
 189         u64 start;
 190         struct vmcore *m = NULL;
 191
 192         if (buflen == 0 || *fpos >= vmcore_size)
 193                 return 0;
 194
 195         /* trim buflen to not go beyond EOF */
 196         if (buflen > vmcore_size - *fpos)
 197                 buflen = vmcore_size - *fpos;
 198
 199         /* Read ELF core header */
 200         if (*fpos < elfcorebuf_sz) {
 201                 tsz = min(elfcorebuf_sz - (size_t)*fpos, buflen);
 202                 if (copy_to(buffer, elfcorebuf + *fpos, tsz, userbuf))
 203                         return -EFAULT;
 204                 buflen -= tsz;
 205                 *fpos += tsz;
 206                 buffer += tsz;
 207                 acc += tsz;
 208
 209                 /* leave now if filled buffer already */
 210                 if (buflen == 0)
 211                         return acc;
 212         }
 213
 214         /* Read Elf note segment */
 215         if (*fpos < elfcorebuf_sz + elfnotes_sz) {
 216                 void *kaddr;
 217
 218                 tsz = min(elfcorebuf_sz + elfnotes_sz - (size_t)*fpos, buflen);
 219                 kaddr = elfnotes_buf + *fpos - elfcorebuf_sz;
 220                 if (copy_to(buffer, kaddr, tsz, userbuf))
 221                         return -EFAULT;
 222                 buflen -= tsz;
 223                 *fpos += tsz;
 224                 buffer += tsz;
 225                 acc += tsz;
 226
 227                 /* leave now if filled buffer already */
 228                 if (buflen == 0)
 229                         return acc;
 230         }
 231
 232         list_for_each_entry(m, &vmcore_list, list) {
 233                 if (*fpos < m->offset + m->size) {
 234                         tsz = min_t(size_t, m->offset + m->size - *fpos, buflen);
 235                         start = m->paddr + *fpos - m->offset;
 236                         tmp = read_from_oldmem(buffer, tsz, &start, userbuf);
 237                         if (tmp < 0)
 238                                 return tmp;
 239                         buflen -= tsz;
 240                         *fpos += tsz;
 241                         buffer += tsz;
 242                         acc += tsz;
 243
 244                         /* leave now if filled buffer already */
 245                         if (buflen == 0)
 246                                 return acc;
 247                 }
 248         }
 249
 250         return acc;
 251 }
 252
 253 static ssize_t read_vmcore(struct file *file, char __user *buffer,
 254                            size_t buflen, loff_t *fpos)
 255 {
 256         return __read_vmcore((__force char *) buffer, buflen, fpos, 1);
 257 }
 258
 259 /*
 260  * The vmcore fault handler uses the page cache and fills data using the
 261  * standard __vmcore_read() function.
 262  *
 263  * On s390 the fault handler is used for memory regions that can't be mapped
 264  * directly with remap_pfn_range().
 265  */
 266 static int mmap_vmcore_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 267 {
 268 #ifdef CONFIG_S390
 269         struct address_space *mapping = vma->vm_file->f_mapping;
 270         pgoff_t index = vmf->pgoff;
 271         struct page *page;
 272         loff_t offset;
 273         char *buf;
 274         int rc;
 275
 276         page = find_or_create_page(mapping, index, GFP_KERNEL);
 277         if (!page)
 278                 return VM_FAULT_OOM;
 279         if (!PageUptodate(page)) {
 280                 offset = (loff_t) index << PAGE_CACHE_SHIFT;
 281                 buf = __va((page_to_pfn(page) << PAGE_SHIFT));
 282                 rc = __read_vmcore(buf, PAGE_SIZE, &offset, 0);
 283                 if (rc < 0) {
 284                         unlock_page(page);
 285                         page_cache_release(page);
 286                         return (rc == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS;
 287                 }
 288                 SetPageUptodate(page);
 289         }
 290         unlock_page(page);
 291         vmf->page = page;
 292         return 0;
 293 #else
 294         return VM_FAULT_SIGBUS;
 295 #endif
 296 }
 297
 298 static const struct vm_operations_struct vmcore_mmap_ops = {
 299         .fault = mmap_vmcore_fault,
 300 };
 301
 302 /**
 303  * alloc_elfnotes_buf - allocate buffer for ELF note segment in
 304  *                      vmalloc memory
 305  *
 306  * @notes_sz: size of buffer
 307  *
 308  * If CONFIG_MMU is defined, use vmalloc_user() to allow users to mmap
 309  * the buffer to user-space by means of remap_vmalloc_range().
 310  *
 311  * If CONFIG_MMU is not defined, use vzalloc() since mmap_vmcore() is
 312  * disabled and there's no need to allow users to mmap the buffer.
 313  */
 314 static inline char *alloc_elfnotes_buf(size_t notes_sz)
 315 {
 316 #ifdef CONFIG_MMU
 317         return vmalloc_user(notes_sz);
 318 #else
 319         return vzalloc(notes_sz);
 320 #endif
 321 }
 322
 323 /*
 324  * Disable mmap_vmcore() if CONFIG_MMU is not defined. MMU is
 325  * essential for mmap_vmcore() in order to map physically
 326  * non-contiguous objects (ELF header, ELF note segment and memory
 327  * regions in the 1st kernel pointed to by PT_LOAD entries) into
 328  * virtually contiguous user-space in ELF layout.
 329  */
 330 #ifdef CONFIG_MMU
 331 static int mmap_vmcore(struct file *file, struct vm_area_struct *vma)
 332 {
 333         size_t size = vma->vm_end - vma->vm_start;
 334         u64 start, end, len, tsz;
 335         struct vmcore *m;
 336
 337         start = (u64)vma->vm_pgoff << PAGE_SHIFT;
 338         end = start + size;
 339
 340         if (size > vmcore_size || end > vmcore_size)
 341                 return -EINVAL;
 342
 343         if (vma->vm_flags & (VM_WRITE | VM_EXEC))
 344                 return -EPERM;
 345
 346         vma->vm_flags &= ~(VM_MAYWRITE | VM_MAYEXEC);
 347         vma->vm_flags |= VM_MIXEDMAP;
 348         vma->vm_ops = &vmcore_mmap_ops;
 349
 350         len = 0;
 351
 352         if (start < elfcorebuf_sz) {
 353                 u64 pfn;
 354
 355                 tsz = min(elfcorebuf_sz - (size_t)start, size);
 356                 pfn = __pa(elfcorebuf + start) >> PAGE_SHIFT;
 357                 if (remap_pfn_range(vma, vma->vm_start, pfn, tsz,
 358                                     vma->vm_page_prot))
 359                         return -EAGAIN;
 360                 size -= tsz;
 361                 start += tsz;
 362                 len += tsz;
 363
 364                 if (size == 0)
 365                         return 0;
 366         }
 367
 368         if (start < elfcorebuf_sz + elfnotes_sz) {
 369                 void *kaddr;
 370
 371                 tsz = min(elfcorebuf_sz + elfnotes_sz - (size_t)start, size);
 372                 kaddr = elfnotes_buf + start - elfcorebuf_sz;
 373                 if (remap_vmalloc_range_partial(vma, vma->vm_start + len,
 374                                                 kaddr, tsz))
 375                         goto fail;
 376                 size -= tsz;
 377                 start += tsz;
 378                 len += tsz;
 379
 380                 if (size == 0)
 381                         return 0;
 382         }
 383
 384         list_for_each_entry(m, &vmcore_list, list) {
 385                 if (start < m->offset + m->size) {
 386                         u64 paddr = 0;
 387
 388                         tsz = min_t(size_t, m->offset + m->size - start, size);
 389                         paddr = m->paddr + start - m->offset;
 390                         if (remap_oldmem_pfn_range(vma, vma->vm_start + len,
 391                                                    paddr >> PAGE_SHIFT, tsz,
 392                                                    vma->vm_page_prot))
 393                                 goto fail;
 394                         size -= tsz;
 395                         start += tsz;
 396                         len += tsz;
 397
 398                         if (size == 0)
 399                                 return 0;
 400                 }
 401         }
 402
 403         return 0;
 404 fail:
 405         do_munmap(vma->vm_mm, vma->vm_start, len);
 406         return -EAGAIN;
 407 }
 408 #else
 409 static int mmap_vmcore(struct file *file, struct vm_area_struct *vma)
 410 {
 411         return -ENOSYS;
 412 }
 413 #endif
 414
 415 static const struct file_operations proc_vmcore_operations = {
 416         .read           = read_vmcore,
 417         .llseek         = default_llseek,
 418         .mmap           = mmap_vmcore,
 419 };
 420
 421 static struct vmcore* __init get_new_element(void)
 422 {
 423         return kzalloc(sizeof(struct vmcore), GFP_KERNEL);
 424 }
 425
 426 static u64 __init get_vmcore_size(size_t elfsz, size_t elfnotesegsz,
 427                                   struct list_head *vc_list)
 428 {
 429         u64 size;
 430         struct vmcore *m;
 431
 432         size = elfsz + elfnotesegsz;
 433         list_for_each_entry(m, vc_list, list) {
 434                 size += m->size;
 435         }
 436         return size;
 437 }
 438
 439 /**
 440  * update_note_header_size_elf64 - update p_memsz member of each PT_NOTE entry
 441  *
 442  * @ehdr_ptr: ELF header
 443  *
 444  * This function updates p_memsz member of each PT_NOTE entry in the
 445  * program header table pointed to by @ehdr_ptr to real size of ELF
 446  * note segment.
 447  */
 448 static int __init update_note_header_size_elf64(const Elf64_Ehdr *ehdr_ptr)
 449 {
 450         int i, rc=0;
 451         Elf64_Phdr *phdr_ptr;
 452         Elf64_Nhdr *nhdr_ptr;
 453
 454         phdr_ptr = (Elf64_Phdr *)(ehdr_ptr + 1);
 455         for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
 456                 void *notes_section;
 457                 u64 offset, max_sz, sz, real_sz = 0;
 458                 if (phdr_ptr->p_type != PT_NOTE)
 459                         continue;
 460                 max_sz = phdr_ptr->p_memsz;
 461                 offset = phdr_ptr->p_offset;
 462                 notes_section = kmalloc(max_sz, GFP_KERNEL);
 463                 if (!notes_section)
 464                         return -ENOMEM;
 465                 rc = elfcorehdr_read_notes(notes_section, max_sz, &offset);
 466                 if (rc < 0) {
 467                         kfree(notes_section);
 468                         return rc;
 469                 }
 470                 nhdr_ptr = notes_section;
 471                 while (nhdr_ptr->n_namesz != 0) {
 472                         sz = sizeof(Elf64_Nhdr) +
 473                                 ((nhdr_ptr->n_namesz + 3) & ~3) +
 474                                 ((nhdr_ptr->n_descsz + 3) & ~3);
 475                         if ((real_sz + sz) > max_sz) {
 476                                 pr_warn("Warning: Exceeded p_memsz, dropping PT_NOTE entry n_namesz=0x%x, n_descsz=0x%x\n",
 477                                         nhdr_ptr->n_namesz, nhdr_ptr->n_descsz);
 478                                 break;
 479                         }
 480                         real_sz += sz;
 481                         nhdr_ptr = (Elf64_Nhdr*)((char*)nhdr_ptr + sz);
 482                 }
 483                 kfree(notes_section);
 484                 phdr_ptr->p_memsz = real_sz;
 485                 if (real_sz == 0) {
 486                         pr_warn("Warning: Zero PT_NOTE entries found\n");
 487                 }
 488         }
 489
 490         return 0;
 491 }
 492
 493 /**
 494  * get_note_number_and_size_elf64 - get the number of PT_NOTE program
 495  * headers and sum of real size of their ELF note segment headers and
 496  * data.
 497  *
 498  * @ehdr_ptr: ELF header
 499  * @nr_ptnote: buffer for the number of PT_NOTE program headers
 500  * @sz_ptnote: buffer for size of unique PT_NOTE program header
 501  *
 502  * This function is used to merge multiple PT_NOTE program headers
 503  * into a unique single one. The resulting unique entry will have
 504  * @sz_ptnote in its phdr->p_mem.
 505  *
 506  * It is assumed that program headers with PT_NOTE type pointed to by
 507  * @ehdr_ptr has already been updated by update_note_header_size_elf64
 508  * and each of PT_NOTE program headers has actual ELF note segment
 509  * size in its p_memsz member.
 510  */
 511 static int __init get_note_number_and_size_elf64(const Elf64_Ehdr *ehdr_ptr,
 512                                                  int *nr_ptnote, u64 *sz_ptnote)
 513 {
 514         int i;
 515         Elf64_Phdr *phdr_ptr;
 516
 517         *nr_ptnote = *sz_ptnote = 0;
 518
 519         phdr_ptr = (Elf64_Phdr *)(ehdr_ptr + 1);
 520         for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
 521                 if (phdr_ptr->p_type != PT_NOTE)
 522                         continue;
 523                 *nr_ptnote += 1;
 524                 *sz_ptnote += phdr_ptr->p_memsz;
 525         }
 526
 527         return 0;
 528 }
 529
 530 /**
 531  * copy_notes_elf64 - copy ELF note segments in a given buffer
 532  *
 533  * @ehdr_ptr: ELF header
 534  * @notes_buf: buffer into which ELF note segments are copied
 535  *
 536  * This function is used to copy ELF note segment in the 1st kernel
 537  * into the buffer @notes_buf in the 2nd kernel. It is assumed that
 538  * size of the buffer @notes_buf is equal to or larger than sum of the
 539  * real ELF note segment headers and data.
 540  *
 541  * It is assumed that program headers with PT_NOTE type pointed to by
 542  * @ehdr_ptr has already been updated by update_note_header_size_elf64
 543  * and each of PT_NOTE program headers has actual ELF note segment
 544  * size in its p_memsz member.
 545  */
 546 static int __init copy_notes_elf64(const Elf64_Ehdr *ehdr_ptr, char *notes_buf)
 547 {
 548         int i, rc=0;
 549         Elf64_Phdr *phdr_ptr;
 550
 551         phdr_ptr = (Elf64_Phdr*)(ehdr_ptr + 1);
 552
 553         for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
 554                 u64 offset;
 555                 if (phdr_ptr->p_type != PT_NOTE)
 556                         continue;
 557                 offset = phdr_ptr->p_offset;
 558                 rc = elfcorehdr_read_notes(notes_buf, phdr_ptr->p_memsz,
 559                                            &offset);
 560                 if (rc < 0)
 561                         return rc;
 562                 notes_buf += phdr_ptr->p_memsz;
 563         }
 564
 565         return 0;
 566 }
 567
 568 /* Merges all the PT_NOTE headers into one. */
 569 static int __init merge_note_headers_elf64(char *elfptr, size_t *elfsz,
 570                                            char **notes_buf, size_t *notes_sz)
 571 {
 572         int i, nr_ptnote=0, rc=0;
 573         char *tmp;
 574         Elf64_Ehdr *ehdr_ptr;
 575         Elf64_Phdr phdr;
 576         u64 phdr_sz = 0, note_off;
 577
 578         ehdr_ptr = (Elf64_Ehdr *)elfptr;
 579
 580         rc = update_note_header_size_elf64(ehdr_ptr);
 581         if (rc < 0)
 582                 return rc;
 583
 584         rc = get_note_number_and_size_elf64(ehdr_ptr, &nr_ptnote, &phdr_sz);
 585         if (rc < 0)
 586                 return rc;
 587
 588         *notes_sz = roundup(phdr_sz, PAGE_SIZE);
 589         *notes_buf = alloc_elfnotes_buf(*notes_sz);
 590         if (!*notes_buf)
 591                 return -ENOMEM;
 592
 593         rc = copy_notes_elf64(ehdr_ptr, *notes_buf);
 594         if (rc < 0)
 595                 return rc;
 596
 597         /* Prepare merged PT_NOTE program header. */
 598         phdr.p_type    = PT_NOTE;
 599         phdr.p_flags   = 0;
 600         note_off = sizeof(Elf64_Ehdr) +
 601                         (ehdr_ptr->e_phnum - nr_ptnote +1) * sizeof(Elf64_Phdr);
 602         phdr.p_offset  = roundup(note_off, PAGE_SIZE);
 603         phdr.p_vaddr   = phdr.p_paddr = 0;
 604         phdr.p_filesz  = phdr.p_memsz = phdr_sz;
 605         phdr.p_align   = 0;
 606
 607         /* Add merged PT_NOTE program header*/
 608         tmp = elfptr + sizeof(Elf64_Ehdr);
 609         memcpy(tmp, &phdr, sizeof(phdr));
 610         tmp += sizeof(phdr);
 611
 612         /* Remove unwanted PT_NOTE program headers. */
 613         i = (nr_ptnote - 1) * sizeof(Elf64_Phdr);
 614         *elfsz = *elfsz - i;
 615         memmove(tmp, tmp+i, ((*elfsz)-sizeof(Elf64_Ehdr)-sizeof(Elf64_Phdr)));
 616         memset(elfptr + *elfsz, 0, i);
 617         *elfsz = roundup(*elfsz, PAGE_SIZE);
 618
 619         /* Modify e_phnum to reflect merged headers. */
 620         ehdr_ptr->e_phnum = ehdr_ptr->e_phnum - nr_ptnote + 1;
 621
 622         return 0;
 623 }
 624
 625 /**
 626  * update_note_header_size_elf32 - update p_memsz member of each PT_NOTE entry
 627  *
 628  * @ehdr_ptr: ELF header
 629  *
 630  * This function updates p_memsz member of each PT_NOTE entry in the
 631  * program header table pointed to by @ehdr_ptr to real size of ELF
 632  * note segment.
 633  */
 634 static int __init update_note_header_size_elf32(const Elf32_Ehdr *ehdr_ptr)
 635 {
 636         int i, rc=0;
 637         Elf32_Phdr *phdr_ptr;
 638         Elf32_Nhdr *nhdr_ptr;
 639
 640         phdr_ptr = (Elf32_Phdr *)(ehdr_ptr + 1);
 641         for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
 642                 void *notes_section;
 643                 u64 offset, max_sz, sz, real_sz = 0;
 644                 if (phdr_ptr->p_type != PT_NOTE)
 645                         continue;
 646                 max_sz = phdr_ptr->p_memsz;
 647                 offset = phdr_ptr->p_offset;
 648                 notes_section = kmalloc(max_sz, GFP_KERNEL);
 649                 if (!notes_section)
 650                         return -ENOMEM;
 651                 rc = elfcorehdr_read_notes(notes_section, max_sz, &offset);
 652                 if (rc < 0) {
 653                         kfree(notes_section);
 654                         return rc;
 655                 }
 656                 nhdr_ptr = notes_section;
 657                 while (nhdr_ptr->n_namesz != 0) {
 658                         sz = sizeof(Elf32_Nhdr) +
 659                                 ((nhdr_ptr->n_namesz + 3) & ~3) +
 660                                 ((nhdr_ptr->n_descsz + 3) & ~3);
 661                         if ((real_sz + sz) > max_sz) {
 662                                 pr_warn("Warning: Exceeded p_memsz, dropping PT_NOTE entry n_namesz=0x%x, n_descsz=0x%x\n",
 663                                         nhdr_ptr->n_namesz, nhdr_ptr->n_descsz);
 664                                 break;
 665                         }
 666                         real_sz += sz;
 667                         nhdr_ptr = (Elf32_Nhdr*)((char*)nhdr_ptr + sz);
 668                 }
 669                 kfree(notes_section);
 670                 phdr_ptr->p_memsz = real_sz;
 671                 if (real_sz == 0) {
 672                         pr_warn("Warning: Zero PT_NOTE entries found\n");
 673                 }
 674         }
 675
 676         return 0;
 677 }
 678
 679 /**
 680  * get_note_number_and_size_elf32 - get the number of PT_NOTE program
 681  * headers and sum of real size of their ELF note segment headers and
 682  * data.
 683  *
 684  * @ehdr_ptr: ELF header
 685  * @nr_ptnote: buffer for the number of PT_NOTE program headers
 686  * @sz_ptnote: buffer for size of unique PT_NOTE program header
 687  *
 688  * This function is used to merge multiple PT_NOTE program headers
 689  * into a unique single one. The resulting unique entry will have
 690  * @sz_ptnote in its phdr->p_mem.
 691  *
 692  * It is assumed that program headers with PT_NOTE type pointed to by
 693  * @ehdr_ptr has already been updated by update_note_header_size_elf32
 694  * and each of PT_NOTE program headers has actual ELF note segment
 695  * size in its p_memsz member.
 696  */
 697 static int __init get_note_number_and_size_elf32(const Elf32_Ehdr *ehdr_ptr,
 698                                                  int *nr_ptnote, u64 *sz_ptnote)
 699 {
 700         int i;
 701         Elf32_Phdr *phdr_ptr;
 702
 703         *nr_ptnote = *sz_ptnote = 0;
 704
 705         phdr_ptr = (Elf32_Phdr *)(ehdr_ptr + 1);
 706         for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
 707                 if (phdr_ptr->p_type != PT_NOTE)
 708                         continue;
 709                 *nr_ptnote += 1;
 710                 *sz_ptnote += phdr_ptr->p_memsz;
 711         }
 712
 713         return 0;
 714 }
 715
 716 /**
 717  * copy_notes_elf32 - copy ELF note segments in a given buffer
 718  *
 719  * @ehdr_ptr: ELF header
 720  * @notes_buf: buffer into which ELF note segments are copied
 721  *
 722  * This function is used to copy ELF note segment in the 1st kernel
 723  * into the buffer @notes_buf in the 2nd kernel. It is assumed that
 724  * size of the buffer @notes_buf is equal to or larger than sum of the
 725  * real ELF note segment headers and data.
 726  *
 727  * It is assumed that program headers with PT_NOTE type pointed to by
 728  * @ehdr_ptr has already been updated by update_note_header_size_elf32
 729  * and each of PT_NOTE program headers has actual ELF note segment
 730  * size in its p_memsz member.
 731  */
 732 static int __init copy_notes_elf32(const Elf32_Ehdr *ehdr_ptr, char *notes_buf)
 733 {
 734         int i, rc=0;
 735         Elf32_Phdr *phdr_ptr;
 736
 737         phdr_ptr = (Elf32_Phdr*)(ehdr_ptr + 1);
 738
 739         for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
 740                 u64 offset;
 741                 if (phdr_ptr->p_type != PT_NOTE)
 742                         continue;
 743                 offset = phdr_ptr->p_offset;
 744                 rc = elfcorehdr_read_notes(notes_buf, phdr_ptr->p_memsz,
 745                                            &offset);
 746                 if (rc < 0)
 747                         return rc;
 748                 notes_buf += phdr_ptr->p_memsz;
 749         }
 750
 751         return 0;
 752 }
 753
 754 /* Merges all the PT_NOTE headers into one. */
 755 static int __init merge_note_headers_elf32(char *elfptr, size_t *elfsz,
 756                                            char **notes_buf, size_t *notes_sz)
 757 {
 758         int i, nr_ptnote=0, rc=0;
 759         char *tmp;
 760         Elf32_Ehdr *ehdr_ptr;
 761         Elf32_Phdr phdr;
 762         u64 phdr_sz = 0, note_off;
 763
 764         ehdr_ptr = (Elf32_Ehdr *)elfptr;
 765
 766         rc = update_note_header_size_elf32(ehdr_ptr);
 767         if (rc < 0)
 768                 return rc;
 769
 770         rc = get_note_number_and_size_elf32(ehdr_ptr, &nr_ptnote, &phdr_sz);
 771         if (rc < 0)
 772                 return rc;
 773
 774         *notes_sz = roundup(phdr_sz, PAGE_SIZE);
 775         *notes_buf = alloc_elfnotes_buf(*notes_sz);
 776         if (!*notes_buf)
 777                 return -ENOMEM;
 778
 779         rc = copy_notes_elf32(ehdr_ptr, *notes_buf);
 780         if (rc < 0)
 781                 return rc;
 782
 783         /* Prepare merged PT_NOTE program header. */
 784         phdr.p_type    = PT_NOTE;
 785         phdr.p_flags   = 0;
 786         note_off = sizeof(Elf32_Ehdr) +
 787                         (ehdr_ptr->e_phnum - nr_ptnote +1) * sizeof(Elf32_Phdr);
 788         phdr.p_offset  = roundup(note_off, PAGE_SIZE);
 789         phdr.p_vaddr   = phdr.p_paddr = 0;
 790         phdr.p_filesz  = phdr.p_memsz = phdr_sz;
 791         phdr.p_align   = 0;
 792
 793         /* Add merged PT_NOTE program header*/
 794         tmp = elfptr + sizeof(Elf32_Ehdr);
 795         memcpy(tmp, &phdr, sizeof(phdr));
 796         tmp += sizeof(phdr);
 797
 798         /* Remove unwanted PT_NOTE program headers. */
 799         i = (nr_ptnote - 1) * sizeof(Elf32_Phdr);
 800         *elfsz = *elfsz - i;
 801         memmove(tmp, tmp+i, ((*elfsz)-sizeof(Elf32_Ehdr)-sizeof(Elf32_Phdr)));
 802         memset(elfptr + *elfsz, 0, i);
 803         *elfsz = roundup(*elfsz, PAGE_SIZE);
 804
 805         /* Modify e_phnum to reflect merged headers. */
 806         ehdr_ptr->e_phnum = ehdr_ptr->e_phnum - nr_ptnote + 1;
 807
 808         return 0;
 809 }
 810
 811 /* Add memory chunks represented by program headers to vmcore list. Also update
 812  * the new offset fields of exported program headers. */
 813 static int __init process_ptload_program_headers_elf64(char *elfptr,
 814                                                 size_t elfsz,
 815                                                 size_t elfnotes_sz,
 816                                                 struct list_head *vc_list)
 817 {
 818         int i;
 819         Elf64_Ehdr *ehdr_ptr;
 820         Elf64_Phdr *phdr_ptr;
 821         loff_t vmcore_off;
 822         struct vmcore *new;
 823
 824         ehdr_ptr = (Elf64_Ehdr *)elfptr;
 825         phdr_ptr = (Elf64_Phdr*)(elfptr + sizeof(Elf64_Ehdr)); /* PT_NOTE hdr */
 826
 827         /* Skip Elf header, program headers and Elf note segment. */
 828         vmcore_off = elfsz + elfnotes_sz;
 829
 830         for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
 831                 u64 paddr, start, end, size;
 832
 833                 if (phdr_ptr->p_type != PT_LOAD)
 834                         continue;
 835
 836                 paddr = phdr_ptr->p_offset;
 837                 start = rounddown(paddr, PAGE_SIZE);
 838                 end = roundup(paddr + phdr_ptr->p_memsz, PAGE_SIZE);
 839                 size = end - start;
 840
 841                 /* Add this contiguous chunk of memory to vmcore list.*/
 842                 new = get_new_element();
 843                 if (!new)
 844                         return -ENOMEM;
 845                 new->paddr = start;
 846                 new->size = size;
 847                 list_add_tail(&new->list, vc_list);
 848
 849                 /* Update the program header offset. */
 850                 phdr_ptr->p_offset = vmcore_off + (paddr - start);
 851                 vmcore_off = vmcore_off + size;
 852         }
 853         return 0;
 854 }
 855
 856 static int __init process_ptload_program_headers_elf32(char *elfptr,
 857                                                 size_t elfsz,
 858                                                 size_t elfnotes_sz,
 859                                                 struct list_head *vc_list)
 860 {
 861         int i;
 862         Elf32_Ehdr *ehdr_ptr;
 863         Elf32_Phdr *phdr_ptr;
 864         loff_t vmcore_off;
 865         struct vmcore *new;
 866
 867         ehdr_ptr = (Elf32_Ehdr *)elfptr;
 868         phdr_ptr = (Elf32_Phdr*)(elfptr + sizeof(Elf32_Ehdr)); /* PT_NOTE hdr */
 869
 870         /* Skip Elf header, program headers and Elf note segment. */
 871         vmcore_off = elfsz + elfnotes_sz;
 872
 873         for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
 874                 u64 paddr, start, end, size;
 875
 876                 if (phdr_ptr->p_type != PT_LOAD)
 877                         continue;
 878
 879                 paddr = phdr_ptr->p_offset;
 880                 start = rounddown(paddr, PAGE_SIZE);
 881                 end = roundup(paddr + phdr_ptr->p_memsz, PAGE_SIZE);
 882                 size = end - start;
 883
 884                 /* Add this contiguous chunk of memory to vmcore list.*/
 885                 new = get_new_element();
 886                 if (!new)
 887                         return -ENOMEM;
 888                 new->paddr = start;
 889                 new->size = size;
 890                 list_add_tail(&new->list, vc_list);
 891
 892                 /* Update the program header offset */
 893                 phdr_ptr->p_offset = vmcore_off + (paddr - start);
 894                 vmcore_off = vmcore_off + size;
 895         }
 896         return 0;
 897 }
 898
 899 /* Sets offset fields of vmcore elements. */
 900 static void __init set_vmcore_list_offsets(size_t elfsz, size_t elfnotes_sz,
 901                                            struct list_head *vc_list)
 902 {
 903         loff_t vmcore_off;
 904         struct vmcore *m;
 905
 906         /* Skip Elf header, program headers and Elf note segment. */
 907         vmcore_off = elfsz + elfnotes_sz;
 908
 909         list_for_each_entry(m, vc_list, list) {
 910                 m->offset = vmcore_off;
 911                 vmcore_off += m->size;
 912         }
 913 }
 914
 915 static void free_elfcorebuf(void)
 916 {
 917         free_pages((unsigned long)elfcorebuf, get_order(elfcorebuf_sz_orig));
 918         elfcorebuf = NULL;
 919         vfree(elfnotes_buf);
 920         elfnotes_buf = NULL;
 921 }
 922
 923 static int __init parse_crash_elf64_headers(void)
 924 {
 925         int rc=0;
 926         Elf64_Ehdr ehdr;
 927         u64 addr;
 928
 929         addr = elfcorehdr_addr;
 930
 931         /* Read Elf header */
 932         rc = elfcorehdr_read((char *)&ehdr, sizeof(Elf64_Ehdr), &addr);
 933         if (rc < 0)
 934                 return rc;
 935
 936         /* Do some basic Verification. */
 937         if (memcmp(ehdr.e_ident, ELFMAG, SELFMAG) != 0 ||
 938                 (ehdr.e_type != ET_CORE) ||
 939                 !vmcore_elf64_check_arch(&ehdr) ||
 940                 ehdr.e_ident[EI_CLASS] != ELFCLASS64 ||
 941                 ehdr.e_ident[EI_VERSION] != EV_CURRENT ||
 942                 ehdr.e_version != EV_CURRENT ||
 943                 ehdr.e_ehsize != sizeof(Elf64_Ehdr) ||
 944                 ehdr.e_phentsize != sizeof(Elf64_Phdr) ||
 945                 ehdr.e_phnum == 0) {
 946                 pr_warn("Warning: Core image elf header is not sane\n");
 947                 return -EINVAL;
 948         }
 949
 950         /* Read in all elf headers. */
 951         elfcorebuf_sz_orig = sizeof(Elf64_Ehdr) +
 952                                 ehdr.e_phnum * sizeof(Elf64_Phdr);
 953         elfcorebuf_sz = elfcorebuf_sz_orig;
 954         elfcorebuf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
 955                                               get_order(elfcorebuf_sz_orig));
 956         if (!elfcorebuf)
 957                 return -ENOMEM;
 958         addr = elfcorehdr_addr;
 959         rc = elfcorehdr_read(elfcorebuf, elfcorebuf_sz_orig, &addr);
 960         if (rc < 0)
 961                 goto fail;
 962
 963         /* Merge all PT_NOTE headers into one. */
 964         rc = merge_note_headers_elf64(elfcorebuf, &elfcorebuf_sz,
 965                                       &elfnotes_buf, &elfnotes_sz);
 966         if (rc)
 967                 goto fail;
 968         rc = process_ptload_program_headers_elf64(elfcorebuf, elfcorebuf_sz,
 969                                                   elfnotes_sz, &vmcore_list);
 970         if (rc)
 971                 goto fail;
 972         set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
 973         return 0;
 974 fail:
 975         free_elfcorebuf();
 976         return rc;
 977 }
 978
 979 static int __init parse_crash_elf32_headers(void)
 980 {
 981         int rc=0;
 982         Elf32_Ehdr ehdr;
 983         u64 addr;
 984
 985         addr = elfcorehdr_addr;
 986
 987         /* Read Elf header */
 988         rc = elfcorehdr_read((char *)&ehdr, sizeof(Elf32_Ehdr), &addr);
 989         if (rc < 0)
 990                 return rc;
 991
 992         /* Do some basic Verification. */
 993         if (memcmp(ehdr.e_ident, ELFMAG, SELFMAG) != 0 ||
 994                 (ehdr.e_type != ET_CORE) ||
 995                 !elf_check_arch(&ehdr) ||
 996                 ehdr.e_ident[EI_CLASS] != ELFCLASS32||
 997                 ehdr.e_ident[EI_VERSION] != EV_CURRENT ||
 998                 ehdr.e_version != EV_CURRENT ||
 999                 ehdr.e_ehsize != sizeof(Elf32_Ehdr) ||
1000                 ehdr.e_phentsize != sizeof(Elf32_Phdr) ||
1001                 ehdr.e_phnum == 0) {
1002                 pr_warn("Warning: Core image elf header is not sane\n");
1003                 return -EINVAL;
1004         }
1005
1006         /* Read in all elf headers. */
1007         elfcorebuf_sz_orig = sizeof(Elf32_Ehdr) + ehdr.e_phnum * sizeof(Elf32_Phdr);
1008         elfcorebuf_sz = elfcorebuf_sz_orig;
1009         elfcorebuf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1010                                               get_order(elfcorebuf_sz_orig));
1011         if (!elfcorebuf)
1012                 return -ENOMEM;
1013         addr = elfcorehdr_addr;
1014         rc = elfcorehdr_read(elfcorebuf, elfcorebuf_sz_orig, &addr);
1015         if (rc < 0)
1016                 goto fail;
1017
1018         /* Merge all PT_NOTE headers into one. */
1019         rc = merge_note_headers_elf32(elfcorebuf, &elfcorebuf_sz,
1020                                       &elfnotes_buf, &elfnotes_sz);
1021         if (rc)
1022                 goto fail;
1023         rc = process_ptload_program_headers_elf32(elfcorebuf, elfcorebuf_sz,
1024                                                   elfnotes_sz, &vmcore_list);
1025         if (rc)
1026                 goto fail;
1027         set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
1028         return 0;
1029 fail:
1030         free_elfcorebuf();
1031         return rc;
1032 }
1033
1034 static int __init parse_crash_elf_headers(void)
1035 {
1036         unsigned char e_ident[EI_NIDENT];
1037         u64 addr;
1038         int rc=0;
1039
1040         addr = elfcorehdr_addr;
1041         rc = elfcorehdr_read(e_ident, EI_NIDENT, &addr);
1042         if (rc < 0)
1043                 return rc;
1044         if (memcmp(e_ident, ELFMAG, SELFMAG) != 0) {
1045                 pr_warn("Warning: Core image elf header not found\n");
1046                 return -EINVAL;
1047         }
1048
1049         if (e_ident[EI_CLASS] == ELFCLASS64) {
1050                 rc = parse_crash_elf64_headers();
1051                 if (rc)
1052                         return rc;
1053         } else if (e_ident[EI_CLASS] == ELFCLASS32) {
1054                 rc = parse_crash_elf32_headers();
1055                 if (rc)
1056                         return rc;
1057         } else {
1058                 pr_warn("Warning: Core image elf header is not sane\n");
1059                 return -EINVAL;
1060         }
1061
1062         /* Determine vmcore size. */
1063         vmcore_size = get_vmcore_size(elfcorebuf_sz, elfnotes_sz,
1064                                       &vmcore_list);
1065
1066         return 0;
1067 }
1068
1069 /* Init function for vmcore module. */
1070 static int __init vmcore_init(void)
1071 {
1072         int rc = 0;
1073
1074         /* Allow architectures to allocate ELF header in 2nd kernel */
1075         rc = elfcorehdr_alloc(&elfcorehdr_addr, &elfcorehdr_size);
1076         if (rc)
1077                 return rc;
1078         /*
1079          * If elfcorehdr= has been passed in cmdline or created in 2nd kernel,
1080          * then capture the dump.
1081          */
1082         if (!(is_vmcore_usable()))
1083                 return rc;
1084         rc = parse_crash_elf_headers();
1085         if (rc) {
1086                 pr_warn("Kdump: vmcore not initialized\n");
1087                 return rc;
1088         }
1089         elfcorehdr_free(elfcorehdr_addr);
1090         elfcorehdr_addr = ELFCORE_ADDR_ERR;
1091
1092         proc_vmcore = proc_create("vmcore", S_IRUSR, NULL, &proc_vmcore_operations);
1093         if (proc_vmcore)
1094                 proc_vmcore->size = vmcore_size;
1095         return 0;
1096 }
1097 fs_initcall(vmcore_init);
1098
1099 /* Cleanup function for vmcore module. */
1100 void vmcore_cleanup(void)
1101 {
1102         struct list_head *pos, *next;
1103
1104         if (proc_vmcore) {
1105                 proc_remove(proc_vmcore);
1106                 proc_vmcore = NULL;
1107         }
1108
1109         /* clear the vmcore list. */
1110         list_for_each_safe(pos, next, &vmcore_list) {
1111                 struct vmcore *m;
1112
1113                 m = list_entry(pos, struct vmcore, list);
1114                 list_del(&m->list);
1115                 kfree(m);
1116         }
1117         free_elfcorebuf();
1118 }