fs/proc/vmcore.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *      fs/proc/vmcore.c Interface for accessing the crash
   4  *                               dump from the system's previous life.
   5  *      Heavily borrowed from fs/proc/kcore.c
   6  *      Created by: Hariprasad Nellitheertha ([email protected])
   7  *      Copyright (C) IBM Corporation, 2004. All rights reserved
   8  *
   9  */
  10
  11 #include <linux/mm.h>
  12 #include <linux/kcore.h>
  13 #include <linux/user.h>
  14 #include <linux/elf.h>
  15 #include <linux/elfcore.h>
  16 #include <linux/export.h>
  17 #include <linux/slab.h>
  18 #include <linux/highmem.h>
  19 #include <linux/printk.h>
  20 #include <linux/memblock.h>
  21 #include <linux/init.h>
  22 #include <linux/crash_dump.h>
  23 #include <linux/list.h>
  24 #include <linux/moduleparam.h>
  25 #include <linux/mutex.h>
  26 #include <linux/vmalloc.h>
  27 #include <linux/pagemap.h>
  28 #include <linux/uaccess.h>
  29 #include <linux/cc_platform.h>
  30 #include <asm/io.h>
  31 #include "internal.h"
  32
  33 /* List representing chunks of contiguous memory areas and their offsets in
  34  * vmcore file.
  35  */
  36 static LIST_HEAD(vmcore_list);
  37
  38 /* Stores the pointer to the buffer containing kernel elf core headers. */
  39 static char *elfcorebuf;
  40 static size_t elfcorebuf_sz;
  41 static size_t elfcorebuf_sz_orig;
  42
  43 static char *elfnotes_buf;
  44 static size_t elfnotes_sz;
  45 /* Size of all notes minus the device dump notes */
  46 static size_t elfnotes_orig_sz;
  47
  48 /* Total size of vmcore file. */
  49 static u64 vmcore_size;
  50
  51 static struct proc_dir_entry *proc_vmcore;
  52
  53 #ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
  54 /* Device Dump list and mutex to synchronize access to list */
  55 static LIST_HEAD(vmcoredd_list);
  56 static DEFINE_MUTEX(vmcoredd_mutex);
  57
  58 static bool vmcoredd_disabled;
  59 core_param(novmcoredd, vmcoredd_disabled, bool, 0);
  60 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
  61
  62 /* Device Dump Size */
  63 static size_t vmcoredd_orig_sz;
  64
  65 static DEFINE_SPINLOCK(vmcore_cb_lock);
  66 DEFINE_STATIC_SRCU(vmcore_cb_srcu);
  67 /* List of registered vmcore callbacks. */
  68 static LIST_HEAD(vmcore_cb_list);
  69 /* Whether the vmcore has been opened once. */
  70 static bool vmcore_opened;
  71
  72 void register_vmcore_cb(struct vmcore_cb *cb)
  73 {
  74         INIT_LIST_HEAD(&cb->next);
  75         spin_lock(&vmcore_cb_lock);
  76         list_add_tail(&cb->next, &vmcore_cb_list);
  77         /*
  78          * Registering a vmcore callback after the vmcore was opened is
  79          * very unusual (e.g., manual driver loading).
  80          */
  81         if (vmcore_opened)
  82                 pr_warn_once("Unexpected vmcore callback registration\n");
  83         spin_unlock(&vmcore_cb_lock);
  84 }
  85 EXPORT_SYMBOL_GPL(register_vmcore_cb);
  86
  87 void unregister_vmcore_cb(struct vmcore_cb *cb)
  88 {
  89         spin_lock(&vmcore_cb_lock);
  90         list_del_rcu(&cb->next);
  91         /*
  92          * Unregistering a vmcore callback after the vmcore was opened is
  93          * very unusual (e.g., forced driver removal), but we cannot stop
  94          * unregistering.
  95          */
  96         if (vmcore_opened)
  97                 pr_warn_once("Unexpected vmcore callback unregistration\n");
  98         spin_unlock(&vmcore_cb_lock);
  99
 100         synchronize_srcu(&vmcore_cb_srcu);
 101 }
 102 EXPORT_SYMBOL_GPL(unregister_vmcore_cb);
 103
 104 static bool pfn_is_ram(unsigned long pfn)
 105 {
 106         struct vmcore_cb *cb;
 107         bool ret = true;
 108
 109         list_for_each_entry_srcu(cb, &vmcore_cb_list, next,
 110                                  srcu_read_lock_held(&vmcore_cb_srcu)) {
 111                 if (unlikely(!cb->pfn_is_ram))
 112                         continue;
 113                 ret = cb->pfn_is_ram(cb, pfn);
 114                 if (!ret)
 115                         break;
 116         }
 117
 118         return ret;
 119 }
 120
 121 static int open_vmcore(struct inode *inode, struct file *file)
 122 {
 123         spin_lock(&vmcore_cb_lock);
 124         vmcore_opened = true;
 125         spin_unlock(&vmcore_cb_lock);
 126
 127         return 0;
 128 }
 129
 130 /* Reads a page from the oldmem device from given offset. */
 131 ssize_t read_from_oldmem(char *buf, size_t count,
 132                          u64 *ppos, int userbuf,
 133                          bool encrypted)
 134 {
 135         unsigned long pfn, offset;
 136         size_t nr_bytes;
 137         ssize_t read = 0, tmp;
 138         int idx;
 139
 140         if (!count)
 141                 return 0;
 142
 143         offset = (unsigned long)(*ppos % PAGE_SIZE);
 144         pfn = (unsigned long)(*ppos / PAGE_SIZE);
 145
 146         idx = srcu_read_lock(&vmcore_cb_srcu);
 147         do {
 148                 if (count > (PAGE_SIZE - offset))
 149                         nr_bytes = PAGE_SIZE - offset;
 150                 else
 151                         nr_bytes = count;
 152
 153                 /* If pfn is not ram, return zeros for sparse dump files */
 154                 if (!pfn_is_ram(pfn)) {
 155                         tmp = 0;
 156                         if (!userbuf)
 157                                 memset(buf, 0, nr_bytes);
 158                         else if (clear_user(buf, nr_bytes))
 159                                 tmp = -EFAULT;
 160                 } else {
 161                         if (encrypted)
 162                                 tmp = copy_oldmem_page_encrypted(pfn, buf,
 163                                                                  nr_bytes,
 164                                                                  offset,
 165                                                                  userbuf);
 166                         else
 167                                 tmp = copy_oldmem_page(pfn, buf, nr_bytes,
 168                                                        offset, userbuf);
 169                 }
 170                 if (tmp < 0) {
 171                         srcu_read_unlock(&vmcore_cb_srcu, idx);
 172                         return tmp;
 173                 }
 174
 175                 *ppos += nr_bytes;
 176                 count -= nr_bytes;
 177                 buf += nr_bytes;
 178                 read += nr_bytes;
 179                 ++pfn;
 180                 offset = 0;
 181         } while (count);
 182         srcu_read_unlock(&vmcore_cb_srcu, idx);
 183
 184         return read;
 185 }
 186
 187 /*
 188  * Architectures may override this function to allocate ELF header in 2nd kernel
 189  */
 190 int __weak elfcorehdr_alloc(unsigned long long *addr, unsigned long long *size)
 191 {
 192         return 0;
 193 }
 194
 195 /*
 196  * Architectures may override this function to free header
 197  */
 198 void __weak elfcorehdr_free(unsigned long long addr)
 199 {}
 200
 201 /*
 202  * Architectures may override this function to read from ELF header
 203  */
 204 ssize_t __weak elfcorehdr_read(char *buf, size_t count, u64 *ppos)
 205 {
 206         return read_from_oldmem(buf, count, ppos, 0, false);
 207 }
 208
 209 /*
 210  * Architectures may override this function to read from notes sections
 211  */
 212 ssize_t __weak elfcorehdr_read_notes(char *buf, size_t count, u64 *ppos)
 213 {
 214         return read_from_oldmem(buf, count, ppos, 0, cc_platform_has(CC_ATTR_MEM_ENCRYPT));
 215 }
 216
 217 /*
 218  * Architectures may override this function to map oldmem
 219  */
 220 int __weak remap_oldmem_pfn_range(struct vm_area_struct *vma,
 221                                   unsigned long from, unsigned long pfn,
 222                                   unsigned long size, pgprot_t prot)
 223 {
 224         prot = pgprot_encrypted(prot);
 225         return remap_pfn_range(vma, from, pfn, size, prot);
 226 }
 227
 228 /*
 229  * Architectures which support memory encryption override this.
 230  */
 231 ssize_t __weak
 232 copy_oldmem_page_encrypted(unsigned long pfn, char *buf, size_t csize,
 233                            unsigned long offset, int userbuf)
 234 {
 235         return copy_oldmem_page(pfn, buf, csize, offset, userbuf);
 236 }
 237
 238 /*
 239  * Copy to either kernel or user space
 240  */
 241 static int copy_to(void *target, void *src, size_t size, int userbuf)
 242 {
 243         if (userbuf) {
 244                 if (copy_to_user((char __user *) target, src, size))
 245                         return -EFAULT;
 246         } else {
 247                 memcpy(target, src, size);
 248         }
 249         return 0;
 250 }
 251
 252 #ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
 253 static int vmcoredd_copy_dumps(void *dst, u64 start, size_t size, int userbuf)
 254 {
 255         struct vmcoredd_node *dump;
 256         u64 offset = 0;
 257         int ret = 0;
 258         size_t tsz;
 259         char *buf;
 260
 261         mutex_lock(&vmcoredd_mutex);
 262         list_for_each_entry(dump, &vmcoredd_list, list) {
 263                 if (start < offset + dump->size) {
 264                         tsz = min(offset + (u64)dump->size - start, (u64)size);
 265                         buf = dump->buf + start - offset;
 266                         if (copy_to(dst, buf, tsz, userbuf)) {
 267                                 ret = -EFAULT;
 268                                 goto out_unlock;
 269                         }
 270
 271                         size -= tsz;
 272                         start += tsz;
 273                         dst += tsz;
 274
 275                         /* Leave now if buffer filled already */
 276                         if (!size)
 277                                 goto out_unlock;
 278                 }
 279                 offset += dump->size;
 280         }
 281
 282 out_unlock:
 283         mutex_unlock(&vmcoredd_mutex);
 284         return ret;
 285 }
 286
 287 #ifdef CONFIG_MMU
 288 static int vmcoredd_mmap_dumps(struct vm_area_struct *vma, unsigned long dst,
 289                                u64 start, size_t size)
 290 {
 291         struct vmcoredd_node *dump;
 292         u64 offset = 0;
 293         int ret = 0;
 294         size_t tsz;
 295         char *buf;
 296
 297         mutex_lock(&vmcoredd_mutex);
 298         list_for_each_entry(dump, &vmcoredd_list, list) {
 299                 if (start < offset + dump->size) {
 300                         tsz = min(offset + (u64)dump->size - start, (u64)size);
 301                         buf = dump->buf + start - offset;
 302                         if (remap_vmalloc_range_partial(vma, dst, buf, 0,
 303                                                         tsz)) {
 304                                 ret = -EFAULT;
 305                                 goto out_unlock;
 306                         }
 307
 308                         size -= tsz;
 309                         start += tsz;
 310                         dst += tsz;
 311
 312                         /* Leave now if buffer filled already */
 313                         if (!size)
 314                                 goto out_unlock;
 315                 }
 316                 offset += dump->size;
 317         }
 318
 319 out_unlock:
 320         mutex_unlock(&vmcoredd_mutex);
 321         return ret;
 322 }
 323 #endif /* CONFIG_MMU */
 324 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
 325
 326 /* Read from the ELF header and then the crash dump. On error, negative value is
 327  * returned otherwise number of bytes read are returned.
 328  */
 329 static ssize_t __read_vmcore(char *buffer, size_t buflen, loff_t *fpos,
 330                              int userbuf)
 331 {
 332         ssize_t acc = 0, tmp;
 333         size_t tsz;
 334         u64 start;
 335         struct vmcore *m = NULL;
 336
 337         if (buflen == 0 || *fpos >= vmcore_size)
 338                 return 0;
 339
 340         /* trim buflen to not go beyond EOF */
 341         if (buflen > vmcore_size - *fpos)
 342                 buflen = vmcore_size - *fpos;
 343
 344         /* Read ELF core header */
 345         if (*fpos < elfcorebuf_sz) {
 346                 tsz = min(elfcorebuf_sz - (size_t)*fpos, buflen);
 347                 if (copy_to(buffer, elfcorebuf + *fpos, tsz, userbuf))
 348                         return -EFAULT;
 349                 buflen -= tsz;
 350                 *fpos += tsz;
 351                 buffer += tsz;
 352                 acc += tsz;
 353
 354                 /* leave now if filled buffer already */
 355                 if (buflen == 0)
 356                         return acc;
 357         }
 358
 359         /* Read Elf note segment */
 360         if (*fpos < elfcorebuf_sz + elfnotes_sz) {
 361                 void *kaddr;
 362
 363                 /* We add device dumps before other elf notes because the
 364                  * other elf notes may not fill the elf notes buffer
 365                  * completely and we will end up with zero-filled data
 366                  * between the elf notes and the device dumps. Tools will
 367                  * then try to decode this zero-filled data as valid notes
 368                  * and we don't want that. Hence, adding device dumps before
 369                  * the other elf notes ensure that zero-filled data can be
 370                  * avoided.
 371                  */
 372 #ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
 373                 /* Read device dumps */
 374                 if (*fpos < elfcorebuf_sz + vmcoredd_orig_sz) {
 375                         tsz = min(elfcorebuf_sz + vmcoredd_orig_sz -
 376                                   (size_t)*fpos, buflen);
 377                         start = *fpos - elfcorebuf_sz;
 378                         if (vmcoredd_copy_dumps(buffer, start, tsz, userbuf))
 379                                 return -EFAULT;
 380
 381                         buflen -= tsz;
 382                         *fpos += tsz;
 383                         buffer += tsz;
 384                         acc += tsz;
 385
 386                         /* leave now if filled buffer already */
 387                         if (!buflen)
 388                                 return acc;
 389                 }
 390 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
 391
 392                 /* Read remaining elf notes */
 393                 tsz = min(elfcorebuf_sz + elfnotes_sz - (size_t)*fpos, buflen);
 394                 kaddr = elfnotes_buf + *fpos - elfcorebuf_sz - vmcoredd_orig_sz;
 395                 if (copy_to(buffer, kaddr, tsz, userbuf))
 396                         return -EFAULT;
 397
 398                 buflen -= tsz;
 399                 *fpos += tsz;
 400                 buffer += tsz;
 401                 acc += tsz;
 402
 403                 /* leave now if filled buffer already */
 404                 if (buflen == 0)
 405                         return acc;
 406         }
 407
 408         list_for_each_entry(m, &vmcore_list, list) {
 409                 if (*fpos < m->offset + m->size) {
 410                         tsz = (size_t)min_t(unsigned long long,
 411                                             m->offset + m->size - *fpos,
 412                                             buflen);
 413                         start = m->paddr + *fpos - m->offset;
 414                         tmp = read_from_oldmem(buffer, tsz, &start,
 415                                                userbuf, cc_platform_has(CC_ATTR_MEM_ENCRYPT));
 416                         if (tmp < 0)
 417                                 return tmp;
 418                         buflen -= tsz;
 419                         *fpos += tsz;
 420                         buffer += tsz;
 421                         acc += tsz;
 422
 423                         /* leave now if filled buffer already */
 424                         if (buflen == 0)
 425                                 return acc;
 426                 }
 427         }
 428
 429         return acc;
 430 }
 431
 432 static ssize_t read_vmcore(struct file *file, char __user *buffer,
 433                            size_t buflen, loff_t *fpos)
 434 {
 435         return __read_vmcore((__force char *) buffer, buflen, fpos, 1);
 436 }
 437
 438 /*
 439  * The vmcore fault handler uses the page cache and fills data using the
 440  * standard __vmcore_read() function.
 441  *
 442  * On s390 the fault handler is used for memory regions that can't be mapped
 443  * directly with remap_pfn_range().
 444  */
 445 static vm_fault_t mmap_vmcore_fault(struct vm_fault *vmf)
 446 {
 447 #ifdef CONFIG_S390
 448         struct address_space *mapping = vmf->vma->vm_file->f_mapping;
 449         pgoff_t index = vmf->pgoff;
 450         struct page *page;
 451         loff_t offset;
 452         char *buf;
 453         int rc;
 454
 455         page = find_or_create_page(mapping, index, GFP_KERNEL);
 456         if (!page)
 457                 return VM_FAULT_OOM;
 458         if (!PageUptodate(page)) {
 459                 offset = (loff_t) index << PAGE_SHIFT;
 460                 buf = __va((page_to_pfn(page) << PAGE_SHIFT));
 461                 rc = __read_vmcore(buf, PAGE_SIZE, &offset, 0);
 462                 if (rc < 0) {
 463                         unlock_page(page);
 464                         put_page(page);
 465                         return vmf_error(rc);
 466                 }
 467                 SetPageUptodate(page);
 468         }
 469         unlock_page(page);
 470         vmf->page = page;
 471         return 0;
 472 #else
 473         return VM_FAULT_SIGBUS;
 474 #endif
 475 }
 476
 477 static const struct vm_operations_struct vmcore_mmap_ops = {
 478         .fault = mmap_vmcore_fault,
 479 };
 480
 481 /**
 482  * vmcore_alloc_buf - allocate buffer in vmalloc memory
 483  * @size: size of buffer
 484  *
 485  * If CONFIG_MMU is defined, use vmalloc_user() to allow users to mmap
 486  * the buffer to user-space by means of remap_vmalloc_range().
 487  *
 488  * If CONFIG_MMU is not defined, use vzalloc() since mmap_vmcore() is
 489  * disabled and there's no need to allow users to mmap the buffer.
 490  */
 491 static inline char *vmcore_alloc_buf(size_t size)
 492 {
 493 #ifdef CONFIG_MMU
 494         return vmalloc_user(size);
 495 #else
 496         return vzalloc(size);
 497 #endif
 498 }
 499
 500 /*
 501  * Disable mmap_vmcore() if CONFIG_MMU is not defined. MMU is
 502  * essential for mmap_vmcore() in order to map physically
 503  * non-contiguous objects (ELF header, ELF note segment and memory
 504  * regions in the 1st kernel pointed to by PT_LOAD entries) into
 505  * virtually contiguous user-space in ELF layout.
 506  */
 507 #ifdef CONFIG_MMU
 508 /*
 509  * remap_oldmem_pfn_checked - do remap_oldmem_pfn_range replacing all pages
 510  * reported as not being ram with the zero page.
 511  *
 512  * @vma: vm_area_struct describing requested mapping
 513  * @from: start remapping from
 514  * @pfn: page frame number to start remapping to
 515  * @size: remapping size
 516  * @prot: protection bits
 517  *
 518  * Returns zero on success, -EAGAIN on failure.
 519  */
 520 static int remap_oldmem_pfn_checked(struct vm_area_struct *vma,
 521                                     unsigned long from, unsigned long pfn,
 522                                     unsigned long size, pgprot_t prot)
 523 {
 524         unsigned long map_size;
 525         unsigned long pos_start, pos_end, pos;
 526         unsigned long zeropage_pfn = my_zero_pfn(0);
 527         size_t len = 0;
 528
 529         pos_start = pfn;
 530         pos_end = pfn + (size >> PAGE_SHIFT);
 531
 532         for (pos = pos_start; pos < pos_end; ++pos) {
 533                 if (!pfn_is_ram(pos)) {
 534                         /*
 535                          * We hit a page which is not ram. Remap the continuous
 536                          * region between pos_start and pos-1 and replace
 537                          * the non-ram page at pos with the zero page.
 538                          */
 539                         if (pos > pos_start) {
 540                                 /* Remap continuous region */
 541                                 map_size = (pos - pos_start) << PAGE_SHIFT;
 542                                 if (remap_oldmem_pfn_range(vma, from + len,
 543                                                            pos_start, map_size,
 544                                                            prot))
 545                                         goto fail;
 546                                 len += map_size;
 547                         }
 548                         /* Remap the zero page */
 549                         if (remap_oldmem_pfn_range(vma, from + len,
 550                                                    zeropage_pfn,
 551                                                    PAGE_SIZE, prot))
 552                                 goto fail;
 553                         len += PAGE_SIZE;
 554                         pos_start = pos + 1;
 555                 }
 556         }
 557         if (pos > pos_start) {
 558                 /* Remap the rest */
 559                 map_size = (pos - pos_start) << PAGE_SHIFT;
 560                 if (remap_oldmem_pfn_range(vma, from + len, pos_start,
 561                                            map_size, prot))
 562                         goto fail;
 563         }
 564         return 0;
 565 fail:
 566         do_munmap(vma->vm_mm, from, len, NULL);
 567         return -EAGAIN;
 568 }
 569
 570 static int vmcore_remap_oldmem_pfn(struct vm_area_struct *vma,
 571                             unsigned long from, unsigned long pfn,
 572                             unsigned long size, pgprot_t prot)
 573 {
 574         int ret, idx;
 575
 576         /*
 577          * Check if a callback was registered to avoid looping over all
 578          * pages without a reason.
 579          */
 580         idx = srcu_read_lock(&vmcore_cb_srcu);
 581         if (!list_empty(&vmcore_cb_list))
 582                 ret = remap_oldmem_pfn_checked(vma, from, pfn, size, prot);
 583         else
 584                 ret = remap_oldmem_pfn_range(vma, from, pfn, size, prot);
 585         srcu_read_unlock(&vmcore_cb_srcu, idx);
 586         return ret;
 587 }
 588
 589 static int mmap_vmcore(struct file *file, struct vm_area_struct *vma)
 590 {
 591         size_t size = vma->vm_end - vma->vm_start;
 592         u64 start, end, len, tsz;
 593         struct vmcore *m;
 594
 595         start = (u64)vma->vm_pgoff << PAGE_SHIFT;
 596         end = start + size;
 597
 598         if (size > vmcore_size || end > vmcore_size)
 599                 return -EINVAL;
 600
 601         if (vma->vm_flags & (VM_WRITE | VM_EXEC))
 602                 return -EPERM;
 603
 604         vma->vm_flags &= ~(VM_MAYWRITE | VM_MAYEXEC);
 605         vma->vm_flags |= VM_MIXEDMAP;
 606         vma->vm_ops = &vmcore_mmap_ops;
 607
 608         len = 0;
 609
 610         if (start < elfcorebuf_sz) {
 611                 u64 pfn;
 612
 613                 tsz = min(elfcorebuf_sz - (size_t)start, size);
 614                 pfn = __pa(elfcorebuf + start) >> PAGE_SHIFT;
 615                 if (remap_pfn_range(vma, vma->vm_start, pfn, tsz,
 616                                     vma->vm_page_prot))
 617                         return -EAGAIN;
 618                 size -= tsz;
 619                 start += tsz;
 620                 len += tsz;
 621
 622                 if (size == 0)
 623                         return 0;
 624         }
 625
 626         if (start < elfcorebuf_sz + elfnotes_sz) {
 627                 void *kaddr;
 628
 629                 /* We add device dumps before other elf notes because the
 630                  * other elf notes may not fill the elf notes buffer
 631                  * completely and we will end up with zero-filled data
 632                  * between the elf notes and the device dumps. Tools will
 633                  * then try to decode this zero-filled data as valid notes
 634                  * and we don't want that. Hence, adding device dumps before
 635                  * the other elf notes ensure that zero-filled data can be
 636                  * avoided. This also ensures that the device dumps and
 637                  * other elf notes can be properly mmaped at page aligned
 638                  * address.
 639                  */
 640 #ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
 641                 /* Read device dumps */
 642                 if (start < elfcorebuf_sz + vmcoredd_orig_sz) {
 643                         u64 start_off;
 644
 645                         tsz = min(elfcorebuf_sz + vmcoredd_orig_sz -
 646                                   (size_t)start, size);
 647                         start_off = start - elfcorebuf_sz;
 648                         if (vmcoredd_mmap_dumps(vma, vma->vm_start + len,
 649                                                 start_off, tsz))
 650                                 goto fail;
 651
 652                         size -= tsz;
 653                         start += tsz;
 654                         len += tsz;
 655
 656                         /* leave now if filled buffer already */
 657                         if (!size)
 658                                 return 0;
 659                 }
 660 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
 661
 662                 /* Read remaining elf notes */
 663                 tsz = min(elfcorebuf_sz + elfnotes_sz - (size_t)start, size);
 664                 kaddr = elfnotes_buf + start - elfcorebuf_sz - vmcoredd_orig_sz;
 665                 if (remap_vmalloc_range_partial(vma, vma->vm_start + len,
 666                                                 kaddr, 0, tsz))
 667                         goto fail;
 668
 669                 size -= tsz;
 670                 start += tsz;
 671                 len += tsz;
 672
 673                 if (size == 0)
 674                         return 0;
 675         }
 676
 677         list_for_each_entry(m, &vmcore_list, list) {
 678                 if (start < m->offset + m->size) {
 679                         u64 paddr = 0;
 680
 681                         tsz = (size_t)min_t(unsigned long long,
 682                                             m->offset + m->size - start, size);
 683                         paddr = m->paddr + start - m->offset;
 684                         if (vmcore_remap_oldmem_pfn(vma, vma->vm_start + len,
 685                                                     paddr >> PAGE_SHIFT, tsz,
 686                                                     vma->vm_page_prot))
 687                                 goto fail;
 688                         size -= tsz;
 689                         start += tsz;
 690                         len += tsz;
 691
 692                         if (size == 0)
 693                                 return 0;
 694                 }
 695         }
 696
 697         return 0;
 698 fail:
 699         do_munmap(vma->vm_mm, vma->vm_start, len, NULL);
 700         return -EAGAIN;
 701 }
 702 #else
 703 static int mmap_vmcore(struct file *file, struct vm_area_struct *vma)
 704 {
 705         return -ENOSYS;
 706 }
 707 #endif
 708
 709 static const struct proc_ops vmcore_proc_ops = {
 710         .proc_open      = open_vmcore,
 711         .proc_read      = read_vmcore,
 712         .proc_lseek     = default_llseek,
 713         .proc_mmap      = mmap_vmcore,
 714 };
 715
 716 static struct vmcore* __init get_new_element(void)
 717 {
 718         return kzalloc(sizeof(struct vmcore), GFP_KERNEL);
 719 }
 720
 721 static u64 get_vmcore_size(size_t elfsz, size_t elfnotesegsz,
 722                            struct list_head *vc_list)
 723 {
 724         u64 size;
 725         struct vmcore *m;
 726
 727         size = elfsz + elfnotesegsz;
 728         list_for_each_entry(m, vc_list, list) {
 729                 size += m->size;
 730         }
 731         return size;
 732 }
 733
 734 /**
 735  * update_note_header_size_elf64 - update p_memsz member of each PT_NOTE entry
 736  *
 737  * @ehdr_ptr: ELF header
 738  *
 739  * This function updates p_memsz member of each PT_NOTE entry in the
 740  * program header table pointed to by @ehdr_ptr to real size of ELF
 741  * note segment.
 742  */
 743 static int __init update_note_header_size_elf64(const Elf64_Ehdr *ehdr_ptr)
 744 {
 745         int i, rc=0;
 746         Elf64_Phdr *phdr_ptr;
 747         Elf64_Nhdr *nhdr_ptr;
 748
 749         phdr_ptr = (Elf64_Phdr *)(ehdr_ptr + 1);
 750         for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
 751                 void *notes_section;
 752                 u64 offset, max_sz, sz, real_sz = 0;
 753                 if (phdr_ptr->p_type != PT_NOTE)
 754                         continue;
 755                 max_sz = phdr_ptr->p_memsz;
 756                 offset = phdr_ptr->p_offset;
 757                 notes_section = kmalloc(max_sz, GFP_KERNEL);
 758                 if (!notes_section)
 759                         return -ENOMEM;
 760                 rc = elfcorehdr_read_notes(notes_section, max_sz, &offset);
 761                 if (rc < 0) {
 762                         kfree(notes_section);
 763                         return rc;
 764                 }
 765                 nhdr_ptr = notes_section;
 766                 while (nhdr_ptr->n_namesz != 0) {
 767                         sz = sizeof(Elf64_Nhdr) +
 768                                 (((u64)nhdr_ptr->n_namesz + 3) & ~3) +
 769                                 (((u64)nhdr_ptr->n_descsz + 3) & ~3);
 770                         if ((real_sz + sz) > max_sz) {
 771                                 pr_warn("Warning: Exceeded p_memsz, dropping PT_NOTE entry n_namesz=0x%x, n_descsz=0x%x\n",
 772                                         nhdr_ptr->n_namesz, nhdr_ptr->n_descsz);
 773                                 break;
 774                         }
 775                         real_sz += sz;
 776                         nhdr_ptr = (Elf64_Nhdr*)((char*)nhdr_ptr + sz);
 777                 }
 778                 kfree(notes_section);
 779                 phdr_ptr->p_memsz = real_sz;
 780                 if (real_sz == 0) {
 781                         pr_warn("Warning: Zero PT_NOTE entries found\n");
 782                 }
 783         }
 784
 785         return 0;
 786 }
 787
 788 /**
 789  * get_note_number_and_size_elf64 - get the number of PT_NOTE program
 790  * headers and sum of real size of their ELF note segment headers and
 791  * data.
 792  *
 793  * @ehdr_ptr: ELF header
 794  * @nr_ptnote: buffer for the number of PT_NOTE program headers
 795  * @sz_ptnote: buffer for size of unique PT_NOTE program header
 796  *
 797  * This function is used to merge multiple PT_NOTE program headers
 798  * into a unique single one. The resulting unique entry will have
 799  * @sz_ptnote in its phdr->p_mem.
 800  *
 801  * It is assumed that program headers with PT_NOTE type pointed to by
 802  * @ehdr_ptr has already been updated by update_note_header_size_elf64
 803  * and each of PT_NOTE program headers has actual ELF note segment
 804  * size in its p_memsz member.
 805  */
 806 static int __init get_note_number_and_size_elf64(const Elf64_Ehdr *ehdr_ptr,
 807                                                  int *nr_ptnote, u64 *sz_ptnote)
 808 {
 809         int i;
 810         Elf64_Phdr *phdr_ptr;
 811
 812         *nr_ptnote = *sz_ptnote = 0;
 813
 814         phdr_ptr = (Elf64_Phdr *)(ehdr_ptr + 1);
 815         for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
 816                 if (phdr_ptr->p_type != PT_NOTE)
 817                         continue;
 818                 *nr_ptnote += 1;
 819                 *sz_ptnote += phdr_ptr->p_memsz;
 820         }
 821
 822         return 0;
 823 }
 824
 825 /**
 826  * copy_notes_elf64 - copy ELF note segments in a given buffer
 827  *
 828  * @ehdr_ptr: ELF header
 829  * @notes_buf: buffer into which ELF note segments are copied
 830  *
 831  * This function is used to copy ELF note segment in the 1st kernel
 832  * into the buffer @notes_buf in the 2nd kernel. It is assumed that
 833  * size of the buffer @notes_buf is equal to or larger than sum of the
 834  * real ELF note segment headers and data.
 835  *
 836  * It is assumed that program headers with PT_NOTE type pointed to by
 837  * @ehdr_ptr has already been updated by update_note_header_size_elf64
 838  * and each of PT_NOTE program headers has actual ELF note segment
 839  * size in its p_memsz member.
 840  */
 841 static int __init copy_notes_elf64(const Elf64_Ehdr *ehdr_ptr, char *notes_buf)
 842 {
 843         int i, rc=0;
 844         Elf64_Phdr *phdr_ptr;
 845
 846         phdr_ptr = (Elf64_Phdr*)(ehdr_ptr + 1);
 847
 848         for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
 849                 u64 offset;
 850                 if (phdr_ptr->p_type != PT_NOTE)
 851                         continue;
 852                 offset = phdr_ptr->p_offset;
 853                 rc = elfcorehdr_read_notes(notes_buf, phdr_ptr->p_memsz,
 854                                            &offset);
 855                 if (rc < 0)
 856                         return rc;
 857                 notes_buf += phdr_ptr->p_memsz;
 858         }
 859
 860         return 0;
 861 }
 862
 863 /* Merges all the PT_NOTE headers into one. */
 864 static int __init merge_note_headers_elf64(char *elfptr, size_t *elfsz,
 865                                            char **notes_buf, size_t *notes_sz)
 866 {
 867         int i, nr_ptnote=0, rc=0;
 868         char *tmp;
 869         Elf64_Ehdr *ehdr_ptr;
 870         Elf64_Phdr phdr;
 871         u64 phdr_sz = 0, note_off;
 872
 873         ehdr_ptr = (Elf64_Ehdr *)elfptr;
 874
 875         rc = update_note_header_size_elf64(ehdr_ptr);
 876         if (rc < 0)
 877                 return rc;
 878
 879         rc = get_note_number_and_size_elf64(ehdr_ptr, &nr_ptnote, &phdr_sz);
 880         if (rc < 0)
 881                 return rc;
 882
 883         *notes_sz = roundup(phdr_sz, PAGE_SIZE);
 884         *notes_buf = vmcore_alloc_buf(*notes_sz);
 885         if (!*notes_buf)
 886                 return -ENOMEM;
 887
 888         rc = copy_notes_elf64(ehdr_ptr, *notes_buf);
 889         if (rc < 0)
 890                 return rc;
 891
 892         /* Prepare merged PT_NOTE program header. */
 893         phdr.p_type    = PT_NOTE;
 894         phdr.p_flags   = 0;
 895         note_off = sizeof(Elf64_Ehdr) +
 896                         (ehdr_ptr->e_phnum - nr_ptnote +1) * sizeof(Elf64_Phdr);
 897         phdr.p_offset  = roundup(note_off, PAGE_SIZE);
 898         phdr.p_vaddr   = phdr.p_paddr = 0;
 899         phdr.p_filesz  = phdr.p_memsz = phdr_sz;
 900         phdr.p_align   = 0;
 901
 902         /* Add merged PT_NOTE program header*/
 903         tmp = elfptr + sizeof(Elf64_Ehdr);
 904         memcpy(tmp, &phdr, sizeof(phdr));
 905         tmp += sizeof(phdr);
 906
 907         /* Remove unwanted PT_NOTE program headers. */
 908         i = (nr_ptnote - 1) * sizeof(Elf64_Phdr);
 909         *elfsz = *elfsz - i;
 910         memmove(tmp, tmp+i, ((*elfsz)-sizeof(Elf64_Ehdr)-sizeof(Elf64_Phdr)));
 911         memset(elfptr + *elfsz, 0, i);
 912         *elfsz = roundup(*elfsz, PAGE_SIZE);
 913
 914         /* Modify e_phnum to reflect merged headers. */
 915         ehdr_ptr->e_phnum = ehdr_ptr->e_phnum - nr_ptnote + 1;
 916
 917         /* Store the size of all notes.  We need this to update the note
 918          * header when the device dumps will be added.
 919          */
 920         elfnotes_orig_sz = phdr.p_memsz;
 921
 922         return 0;
 923 }
 924
 925 /**
 926  * update_note_header_size_elf32 - update p_memsz member of each PT_NOTE entry
 927  *
 928  * @ehdr_ptr: ELF header
 929  *
 930  * This function updates p_memsz member of each PT_NOTE entry in the
 931  * program header table pointed to by @ehdr_ptr to real size of ELF
 932  * note segment.
 933  */
 934 static int __init update_note_header_size_elf32(const Elf32_Ehdr *ehdr_ptr)
 935 {
 936         int i, rc=0;
 937         Elf32_Phdr *phdr_ptr;
 938         Elf32_Nhdr *nhdr_ptr;
 939
 940         phdr_ptr = (Elf32_Phdr *)(ehdr_ptr + 1);
 941         for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
 942                 void *notes_section;
 943                 u64 offset, max_sz, sz, real_sz = 0;
 944                 if (phdr_ptr->p_type != PT_NOTE)
 945                         continue;
 946                 max_sz = phdr_ptr->p_memsz;
 947                 offset = phdr_ptr->p_offset;
 948                 notes_section = kmalloc(max_sz, GFP_KERNEL);
 949                 if (!notes_section)
 950                         return -ENOMEM;
 951                 rc = elfcorehdr_read_notes(notes_section, max_sz, &offset);
 952                 if (rc < 0) {
 953                         kfree(notes_section);
 954                         return rc;
 955                 }
 956                 nhdr_ptr = notes_section;
 957                 while (nhdr_ptr->n_namesz != 0) {
 958                         sz = sizeof(Elf32_Nhdr) +
 959                                 (((u64)nhdr_ptr->n_namesz + 3) & ~3) +
 960                                 (((u64)nhdr_ptr->n_descsz + 3) & ~3);
 961                         if ((real_sz + sz) > max_sz) {
 962                                 pr_warn("Warning: Exceeded p_memsz, dropping PT_NOTE entry n_namesz=0x%x, n_descsz=0x%x\n",
 963                                         nhdr_ptr->n_namesz, nhdr_ptr->n_descsz);
 964                                 break;
 965                         }
 966                         real_sz += sz;
 967                         nhdr_ptr = (Elf32_Nhdr*)((char*)nhdr_ptr + sz);
 968                 }
 969                 kfree(notes_section);
 970                 phdr_ptr->p_memsz = real_sz;
 971                 if (real_sz == 0) {
 972                         pr_warn("Warning: Zero PT_NOTE entries found\n");
 973                 }
 974         }
 975
 976         return 0;
 977 }
 978
 979 /**
 980  * get_note_number_and_size_elf32 - get the number of PT_NOTE program
 981  * headers and sum of real size of their ELF note segment headers and
 982  * data.
 983  *
 984  * @ehdr_ptr: ELF header
 985  * @nr_ptnote: buffer for the number of PT_NOTE program headers
 986  * @sz_ptnote: buffer for size of unique PT_NOTE program header
 987  *
 988  * This function is used to merge multiple PT_NOTE program headers
 989  * into a unique single one. The resulting unique entry will have
 990  * @sz_ptnote in its phdr->p_mem.
 991  *
 992  * It is assumed that program headers with PT_NOTE type pointed to by
 993  * @ehdr_ptr has already been updated by update_note_header_size_elf32
 994  * and each of PT_NOTE program headers has actual ELF note segment
 995  * size in its p_memsz member.
 996  */
 997 static int __init get_note_number_and_size_elf32(const Elf32_Ehdr *ehdr_ptr,
 998                                                  int *nr_ptnote, u64 *sz_ptnote)
 999 {
1000         int i;
1001         Elf32_Phdr *phdr_ptr;
1002
1003         *nr_ptnote = *sz_ptnote = 0;
1004
1005         phdr_ptr = (Elf32_Phdr *)(ehdr_ptr + 1);
1006         for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
1007                 if (phdr_ptr->p_type != PT_NOTE)
1008                         continue;
1009                 *nr_ptnote += 1;
1010                 *sz_ptnote += phdr_ptr->p_memsz;
1011         }
1012
1013         return 0;
1014 }
1015
1016 /**
1017  * copy_notes_elf32 - copy ELF note segments in a given buffer
1018  *
1019  * @ehdr_ptr: ELF header
1020  * @notes_buf: buffer into which ELF note segments are copied
1021  *
1022  * This function is used to copy ELF note segment in the 1st kernel
1023  * into the buffer @notes_buf in the 2nd kernel. It is assumed that
1024  * size of the buffer @notes_buf is equal to or larger than sum of the
1025  * real ELF note segment headers and data.
1026  *
1027  * It is assumed that program headers with PT_NOTE type pointed to by
1028  * @ehdr_ptr has already been updated by update_note_header_size_elf32
1029  * and each of PT_NOTE program headers has actual ELF note segment
1030  * size in its p_memsz member.
1031  */
1032 static int __init copy_notes_elf32(const Elf32_Ehdr *ehdr_ptr, char *notes_buf)
1033 {
1034         int i, rc=0;
1035         Elf32_Phdr *phdr_ptr;
1036
1037         phdr_ptr = (Elf32_Phdr*)(ehdr_ptr + 1);
1038
1039         for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
1040                 u64 offset;
1041                 if (phdr_ptr->p_type != PT_NOTE)
1042                         continue;
1043                 offset = phdr_ptr->p_offset;
1044                 rc = elfcorehdr_read_notes(notes_buf, phdr_ptr->p_memsz,
1045                                            &offset);
1046                 if (rc < 0)
1047                         return rc;
1048                 notes_buf += phdr_ptr->p_memsz;
1049         }
1050
1051         return 0;
1052 }
1053
1054 /* Merges all the PT_NOTE headers into one. */
1055 static int __init merge_note_headers_elf32(char *elfptr, size_t *elfsz,
1056                                            char **notes_buf, size_t *notes_sz)
1057 {
1058         int i, nr_ptnote=0, rc=0;
1059         char *tmp;
1060         Elf32_Ehdr *ehdr_ptr;
1061         Elf32_Phdr phdr;
1062         u64 phdr_sz = 0, note_off;
1063
1064         ehdr_ptr = (Elf32_Ehdr *)elfptr;
1065
1066         rc = update_note_header_size_elf32(ehdr_ptr);
1067         if (rc < 0)
1068                 return rc;
1069
1070         rc = get_note_number_and_size_elf32(ehdr_ptr, &nr_ptnote, &phdr_sz);
1071         if (rc < 0)
1072                 return rc;
1073
1074         *notes_sz = roundup(phdr_sz, PAGE_SIZE);
1075         *notes_buf = vmcore_alloc_buf(*notes_sz);
1076         if (!*notes_buf)
1077                 return -ENOMEM;
1078
1079         rc = copy_notes_elf32(ehdr_ptr, *notes_buf);
1080         if (rc < 0)
1081                 return rc;
1082
1083         /* Prepare merged PT_NOTE program header. */
1084         phdr.p_type    = PT_NOTE;
1085         phdr.p_flags   = 0;
1086         note_off = sizeof(Elf32_Ehdr) +
1087                         (ehdr_ptr->e_phnum - nr_ptnote +1) * sizeof(Elf32_Phdr);
1088         phdr.p_offset  = roundup(note_off, PAGE_SIZE);
1089         phdr.p_vaddr   = phdr.p_paddr = 0;
1090         phdr.p_filesz  = phdr.p_memsz = phdr_sz;
1091         phdr.p_align   = 0;
1092
1093         /* Add merged PT_NOTE program header*/
1094         tmp = elfptr + sizeof(Elf32_Ehdr);
1095         memcpy(tmp, &phdr, sizeof(phdr));
1096         tmp += sizeof(phdr);
1097
1098         /* Remove unwanted PT_NOTE program headers. */
1099         i = (nr_ptnote - 1) * sizeof(Elf32_Phdr);
1100         *elfsz = *elfsz - i;
1101         memmove(tmp, tmp+i, ((*elfsz)-sizeof(Elf32_Ehdr)-sizeof(Elf32_Phdr)));
1102         memset(elfptr + *elfsz, 0, i);
1103         *elfsz = roundup(*elfsz, PAGE_SIZE);
1104
1105         /* Modify e_phnum to reflect merged headers. */
1106         ehdr_ptr->e_phnum = ehdr_ptr->e_phnum - nr_ptnote + 1;
1107
1108         /* Store the size of all notes.  We need this to update the note
1109          * header when the device dumps will be added.
1110          */
1111         elfnotes_orig_sz = phdr.p_memsz;
1112
1113         return 0;
1114 }
1115
1116 /* Add memory chunks represented by program headers to vmcore list. Also update
1117  * the new offset fields of exported program headers. */
1118 static int __init process_ptload_program_headers_elf64(char *elfptr,
1119                                                 size_t elfsz,
1120                                                 size_t elfnotes_sz,
1121                                                 struct list_head *vc_list)
1122 {
1123         int i;
1124         Elf64_Ehdr *ehdr_ptr;
1125         Elf64_Phdr *phdr_ptr;
1126         loff_t vmcore_off;
1127         struct vmcore *new;
1128
1129         ehdr_ptr = (Elf64_Ehdr *)elfptr;
1130         phdr_ptr = (Elf64_Phdr*)(elfptr + sizeof(Elf64_Ehdr)); /* PT_NOTE hdr */
1131
1132         /* Skip Elf header, program headers and Elf note segment. */
1133         vmcore_off = elfsz + elfnotes_sz;
1134
1135         for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
1136                 u64 paddr, start, end, size;
1137
1138                 if (phdr_ptr->p_type != PT_LOAD)
1139                         continue;
1140
1141                 paddr = phdr_ptr->p_offset;
1142                 start = rounddown(paddr, PAGE_SIZE);
1143                 end = roundup(paddr + phdr_ptr->p_memsz, PAGE_SIZE);
1144                 size = end - start;
1145
1146                 /* Add this contiguous chunk of memory to vmcore list.*/
1147                 new = get_new_element();
1148                 if (!new)
1149                         return -ENOMEM;
1150                 new->paddr = start;
1151                 new->size = size;
1152                 list_add_tail(&new->list, vc_list);
1153
1154                 /* Update the program header offset. */
1155                 phdr_ptr->p_offset = vmcore_off + (paddr - start);
1156                 vmcore_off = vmcore_off + size;
1157         }
1158         return 0;
1159 }
1160
1161 static int __init process_ptload_program_headers_elf32(char *elfptr,
1162                                                 size_t elfsz,
1163                                                 size_t elfnotes_sz,
1164                                                 struct list_head *vc_list)
1165 {
1166         int i;
1167         Elf32_Ehdr *ehdr_ptr;
1168         Elf32_Phdr *phdr_ptr;
1169         loff_t vmcore_off;
1170         struct vmcore *new;
1171
1172         ehdr_ptr = (Elf32_Ehdr *)elfptr;
1173         phdr_ptr = (Elf32_Phdr*)(elfptr + sizeof(Elf32_Ehdr)); /* PT_NOTE hdr */
1174
1175         /* Skip Elf header, program headers and Elf note segment. */
1176         vmcore_off = elfsz + elfnotes_sz;
1177
1178         for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
1179                 u64 paddr, start, end, size;
1180
1181                 if (phdr_ptr->p_type != PT_LOAD)
1182                         continue;
1183
1184                 paddr = phdr_ptr->p_offset;
1185                 start = rounddown(paddr, PAGE_SIZE);
1186                 end = roundup(paddr + phdr_ptr->p_memsz, PAGE_SIZE);
1187                 size = end - start;
1188
1189                 /* Add this contiguous chunk of memory to vmcore list.*/
1190                 new = get_new_element();
1191                 if (!new)
1192                         return -ENOMEM;
1193                 new->paddr = start;
1194                 new->size = size;
1195                 list_add_tail(&new->list, vc_list);
1196
1197                 /* Update the program header offset */
1198                 phdr_ptr->p_offset = vmcore_off + (paddr - start);
1199                 vmcore_off = vmcore_off + size;
1200         }
1201         return 0;
1202 }
1203
1204 /* Sets offset fields of vmcore elements. */
1205 static void set_vmcore_list_offsets(size_t elfsz, size_t elfnotes_sz,
1206                                     struct list_head *vc_list)
1207 {
1208         loff_t vmcore_off;
1209         struct vmcore *m;
1210
1211         /* Skip Elf header, program headers and Elf note segment. */
1212         vmcore_off = elfsz + elfnotes_sz;
1213
1214         list_for_each_entry(m, vc_list, list) {
1215                 m->offset = vmcore_off;
1216                 vmcore_off += m->size;
1217         }
1218 }
1219
1220 static void free_elfcorebuf(void)
1221 {
1222         free_pages((unsigned long)elfcorebuf, get_order(elfcorebuf_sz_orig));
1223         elfcorebuf = NULL;
1224         vfree(elfnotes_buf);
1225         elfnotes_buf = NULL;
1226 }
1227
1228 static int __init parse_crash_elf64_headers(void)
1229 {
1230         int rc=0;
1231         Elf64_Ehdr ehdr;
1232         u64 addr;
1233
1234         addr = elfcorehdr_addr;
1235
1236         /* Read Elf header */
1237         rc = elfcorehdr_read((char *)&ehdr, sizeof(Elf64_Ehdr), &addr);
1238         if (rc < 0)
1239                 return rc;
1240
1241         /* Do some basic Verification. */
1242         if (memcmp(ehdr.e_ident, ELFMAG, SELFMAG) != 0 ||
1243                 (ehdr.e_type != ET_CORE) ||
1244                 !vmcore_elf64_check_arch(&ehdr) ||
1245                 ehdr.e_ident[EI_CLASS] != ELFCLASS64 ||
1246                 ehdr.e_ident[EI_VERSION] != EV_CURRENT ||
1247                 ehdr.e_version != EV_CURRENT ||
1248                 ehdr.e_ehsize != sizeof(Elf64_Ehdr) ||
1249                 ehdr.e_phentsize != sizeof(Elf64_Phdr) ||
1250                 ehdr.e_phnum == 0) {
1251                 pr_warn("Warning: Core image elf header is not sane\n");
1252                 return -EINVAL;
1253         }
1254
1255         /* Read in all elf headers. */
1256         elfcorebuf_sz_orig = sizeof(Elf64_Ehdr) +
1257                                 ehdr.e_phnum * sizeof(Elf64_Phdr);
1258         elfcorebuf_sz = elfcorebuf_sz_orig;
1259         elfcorebuf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1260                                               get_order(elfcorebuf_sz_orig));
1261         if (!elfcorebuf)
1262                 return -ENOMEM;
1263         addr = elfcorehdr_addr;
1264         rc = elfcorehdr_read(elfcorebuf, elfcorebuf_sz_orig, &addr);
1265         if (rc < 0)
1266                 goto fail;
1267
1268         /* Merge all PT_NOTE headers into one. */
1269         rc = merge_note_headers_elf64(elfcorebuf, &elfcorebuf_sz,
1270                                       &elfnotes_buf, &elfnotes_sz);
1271         if (rc)
1272                 goto fail;
1273         rc = process_ptload_program_headers_elf64(elfcorebuf, elfcorebuf_sz,
1274                                                   elfnotes_sz, &vmcore_list);
1275         if (rc)
1276                 goto fail;
1277         set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
1278         return 0;
1279 fail:
1280         free_elfcorebuf();
1281         return rc;
1282 }
1283
1284 static int __init parse_crash_elf32_headers(void)
1285 {
1286         int rc=0;
1287         Elf32_Ehdr ehdr;
1288         u64 addr;
1289
1290         addr = elfcorehdr_addr;
1291
1292         /* Read Elf header */
1293         rc = elfcorehdr_read((char *)&ehdr, sizeof(Elf32_Ehdr), &addr);
1294         if (rc < 0)
1295                 return rc;
1296
1297         /* Do some basic Verification. */
1298         if (memcmp(ehdr.e_ident, ELFMAG, SELFMAG) != 0 ||
1299                 (ehdr.e_type != ET_CORE) ||
1300                 !vmcore_elf32_check_arch(&ehdr) ||
1301                 ehdr.e_ident[EI_CLASS] != ELFCLASS32||
1302                 ehdr.e_ident[EI_VERSION] != EV_CURRENT ||
1303                 ehdr.e_version != EV_CURRENT ||
1304                 ehdr.e_ehsize != sizeof(Elf32_Ehdr) ||
1305                 ehdr.e_phentsize != sizeof(Elf32_Phdr) ||
1306                 ehdr.e_phnum == 0) {
1307                 pr_warn("Warning: Core image elf header is not sane\n");
1308                 return -EINVAL;
1309         }
1310
1311         /* Read in all elf headers. */
1312         elfcorebuf_sz_orig = sizeof(Elf32_Ehdr) + ehdr.e_phnum * sizeof(Elf32_Phdr);
1313         elfcorebuf_sz = elfcorebuf_sz_orig;
1314         elfcorebuf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1315                                               get_order(elfcorebuf_sz_orig));
1316         if (!elfcorebuf)
1317                 return -ENOMEM;
1318         addr = elfcorehdr_addr;
1319         rc = elfcorehdr_read(elfcorebuf, elfcorebuf_sz_orig, &addr);
1320         if (rc < 0)
1321                 goto fail;
1322
1323         /* Merge all PT_NOTE headers into one. */
1324         rc = merge_note_headers_elf32(elfcorebuf, &elfcorebuf_sz,
1325                                       &elfnotes_buf, &elfnotes_sz);
1326         if (rc)
1327                 goto fail;
1328         rc = process_ptload_program_headers_elf32(elfcorebuf, elfcorebuf_sz,
1329                                                   elfnotes_sz, &vmcore_list);
1330         if (rc)
1331                 goto fail;
1332         set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
1333         return 0;
1334 fail:
1335         free_elfcorebuf();
1336         return rc;
1337 }
1338
1339 static int __init parse_crash_elf_headers(void)
1340 {
1341         unsigned char e_ident[EI_NIDENT];
1342         u64 addr;
1343         int rc=0;
1344
1345         addr = elfcorehdr_addr;
1346         rc = elfcorehdr_read(e_ident, EI_NIDENT, &addr);
1347         if (rc < 0)
1348                 return rc;
1349         if (memcmp(e_ident, ELFMAG, SELFMAG) != 0) {
1350                 pr_warn("Warning: Core image elf header not found\n");
1351                 return -EINVAL;
1352         }
1353
1354         if (e_ident[EI_CLASS] == ELFCLASS64) {
1355                 rc = parse_crash_elf64_headers();
1356                 if (rc)
1357                         return rc;
1358         } else if (e_ident[EI_CLASS] == ELFCLASS32) {
1359                 rc = parse_crash_elf32_headers();
1360                 if (rc)
1361                         return rc;
1362         } else {
1363                 pr_warn("Warning: Core image elf header is not sane\n");
1364                 return -EINVAL;
1365         }
1366
1367         /* Determine vmcore size. */
1368         vmcore_size = get_vmcore_size(elfcorebuf_sz, elfnotes_sz,
1369                                       &vmcore_list);
1370
1371         return 0;
1372 }
1373
1374 #ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
1375 /**
1376  * vmcoredd_write_header - Write vmcore device dump header at the
1377  * beginning of the dump's buffer.
1378  * @buf: Output buffer where the note is written
1379  * @data: Dump info
1380  * @size: Size of the dump
1381  *
1382  * Fills beginning of the dump's buffer with vmcore device dump header.
1383  */
1384 static void vmcoredd_write_header(void *buf, struct vmcoredd_data *data,
1385                                   u32 size)
1386 {
1387         struct vmcoredd_header *vdd_hdr = (struct vmcoredd_header *)buf;
1388
1389         vdd_hdr->n_namesz = sizeof(vdd_hdr->name);
1390         vdd_hdr->n_descsz = size + sizeof(vdd_hdr->dump_name);
1391         vdd_hdr->n_type = NT_VMCOREDD;
1392
1393         strncpy((char *)vdd_hdr->name, VMCOREDD_NOTE_NAME,
1394                 sizeof(vdd_hdr->name));
1395         memcpy(vdd_hdr->dump_name, data->dump_name, sizeof(vdd_hdr->dump_name));
1396 }
1397
1398 /**
1399  * vmcoredd_update_program_headers - Update all Elf program headers
1400  * @elfptr: Pointer to elf header
1401  * @elfnotesz: Size of elf notes aligned to page size
1402  * @vmcoreddsz: Size of device dumps to be added to elf note header
1403  *
1404  * Determine type of Elf header (Elf64 or Elf32) and update the elf note size.
1405  * Also update the offsets of all the program headers after the elf note header.
1406  */
1407 static void vmcoredd_update_program_headers(char *elfptr, size_t elfnotesz,
1408                                             size_t vmcoreddsz)
1409 {
1410         unsigned char *e_ident = (unsigned char *)elfptr;
1411         u64 start, end, size;
1412         loff_t vmcore_off;
1413         u32 i;
1414
1415         vmcore_off = elfcorebuf_sz + elfnotesz;
1416
1417         if (e_ident[EI_CLASS] == ELFCLASS64) {
1418                 Elf64_Ehdr *ehdr = (Elf64_Ehdr *)elfptr;
1419                 Elf64_Phdr *phdr = (Elf64_Phdr *)(elfptr + sizeof(Elf64_Ehdr));
1420
1421                 /* Update all program headers */
1422                 for (i = 0; i < ehdr->e_phnum; i++, phdr++) {
1423                         if (phdr->p_type == PT_NOTE) {
1424                                 /* Update note size */
1425                                 phdr->p_memsz = elfnotes_orig_sz + vmcoreddsz;
1426                                 phdr->p_filesz = phdr->p_memsz;
1427                                 continue;
1428                         }
1429
1430                         start = rounddown(phdr->p_offset, PAGE_SIZE);
1431                         end = roundup(phdr->p_offset + phdr->p_memsz,
1432                                       PAGE_SIZE);
1433                         size = end - start;
1434                         phdr->p_offset = vmcore_off + (phdr->p_offset - start);
1435                         vmcore_off += size;
1436                 }
1437         } else {
1438                 Elf32_Ehdr *ehdr = (Elf32_Ehdr *)elfptr;
1439                 Elf32_Phdr *phdr = (Elf32_Phdr *)(elfptr + sizeof(Elf32_Ehdr));
1440
1441                 /* Update all program headers */
1442                 for (i = 0; i < ehdr->e_phnum; i++, phdr++) {
1443                         if (phdr->p_type == PT_NOTE) {
1444                                 /* Update note size */
1445                                 phdr->p_memsz = elfnotes_orig_sz + vmcoreddsz;
1446                                 phdr->p_filesz = phdr->p_memsz;
1447                                 continue;
1448                         }
1449
1450                         start = rounddown(phdr->p_offset, PAGE_SIZE);
1451                         end = roundup(phdr->p_offset + phdr->p_memsz,
1452                                       PAGE_SIZE);
1453                         size = end - start;
1454                         phdr->p_offset = vmcore_off + (phdr->p_offset - start);
1455                         vmcore_off += size;
1456                 }
1457         }
1458 }
1459
1460 /**
1461  * vmcoredd_update_size - Update the total size of the device dumps and update
1462  * Elf header
1463  * @dump_size: Size of the current device dump to be added to total size
1464  *
1465  * Update the total size of all the device dumps and update the Elf program
1466  * headers. Calculate the new offsets for the vmcore list and update the
1467  * total vmcore size.
1468  */
1469 static void vmcoredd_update_size(size_t dump_size)
1470 {
1471         vmcoredd_orig_sz += dump_size;
1472         elfnotes_sz = roundup(elfnotes_orig_sz, PAGE_SIZE) + vmcoredd_orig_sz;
1473         vmcoredd_update_program_headers(elfcorebuf, elfnotes_sz,
1474                                         vmcoredd_orig_sz);
1475
1476         /* Update vmcore list offsets */
1477         set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
1478
1479         vmcore_size = get_vmcore_size(elfcorebuf_sz, elfnotes_sz,
1480                                       &vmcore_list);
1481         proc_vmcore->size = vmcore_size;
1482 }
1483
1484 /**
1485  * vmcore_add_device_dump - Add a buffer containing device dump to vmcore
1486  * @data: dump info.
1487  *
1488  * Allocate a buffer and invoke the calling driver's dump collect routine.
1489  * Write Elf note at the beginning of the buffer to indicate vmcore device
1490  * dump and add the dump to global list.
1491  */
1492 int vmcore_add_device_dump(struct vmcoredd_data *data)
1493 {
1494         struct vmcoredd_node *dump;
1495         void *buf = NULL;
1496         size_t data_size;
1497         int ret;
1498
1499         if (vmcoredd_disabled) {
1500                 pr_err_once("Device dump is disabled\n");
1501                 return -EINVAL;
1502         }
1503
1504         if (!data || !strlen(data->dump_name) ||
1505             !data->vmcoredd_callback || !data->size)
1506                 return -EINVAL;
1507
1508         dump = vzalloc(sizeof(*dump));
1509         if (!dump) {
1510                 ret = -ENOMEM;
1511                 goto out_err;
1512         }
1513
1514         /* Keep size of the buffer page aligned so that it can be mmaped */
1515         data_size = roundup(sizeof(struct vmcoredd_header) + data->size,
1516                             PAGE_SIZE);
1517
1518         /* Allocate buffer for driver's to write their dumps */
1519         buf = vmcore_alloc_buf(data_size);
1520         if (!buf) {
1521                 ret = -ENOMEM;
1522                 goto out_err;
1523         }
1524
1525         vmcoredd_write_header(buf, data, data_size -
1526                               sizeof(struct vmcoredd_header));
1527
1528         /* Invoke the driver's dump collection routing */
1529         ret = data->vmcoredd_callback(data, buf +
1530                                       sizeof(struct vmcoredd_header));
1531         if (ret)
1532                 goto out_err;
1533
1534         dump->buf = buf;
1535         dump->size = data_size;
1536
1537         /* Add the dump to driver sysfs list */
1538         mutex_lock(&vmcoredd_mutex);
1539         list_add_tail(&dump->list, &vmcoredd_list);
1540         mutex_unlock(&vmcoredd_mutex);
1541
1542         vmcoredd_update_size(data_size);
1543         return 0;
1544
1545 out_err:
1546         vfree(buf);
1547         vfree(dump);
1548
1549         return ret;
1550 }
1551 EXPORT_SYMBOL(vmcore_add_device_dump);
1552 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
1553
1554 /* Free all dumps in vmcore device dump list */
1555 static void vmcore_free_device_dumps(void)
1556 {
1557 #ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
1558         mutex_lock(&vmcoredd_mutex);
1559         while (!list_empty(&vmcoredd_list)) {
1560                 struct vmcoredd_node *dump;
1561
1562                 dump = list_first_entry(&vmcoredd_list, struct vmcoredd_node,
1563                                         list);
1564                 list_del(&dump->list);
1565                 vfree(dump->buf);
1566                 vfree(dump);
1567         }
1568         mutex_unlock(&vmcoredd_mutex);
1569 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
1570 }
1571
1572 /* Init function for vmcore module. */
1573 static int __init vmcore_init(void)
1574 {
1575         int rc = 0;
1576
1577         /* Allow architectures to allocate ELF header in 2nd kernel */
1578         rc = elfcorehdr_alloc(&elfcorehdr_addr, &elfcorehdr_size);
1579         if (rc)
1580                 return rc;
1581         /*
1582          * If elfcorehdr= has been passed in cmdline or created in 2nd kernel,
1583          * then capture the dump.
1584          */
1585         if (!(is_vmcore_usable()))
1586                 return rc;
1587         rc = parse_crash_elf_headers();
1588         if (rc) {
1589                 pr_warn("Kdump: vmcore not initialized\n");
1590                 return rc;
1591         }
1592         elfcorehdr_free(elfcorehdr_addr);
1593         elfcorehdr_addr = ELFCORE_ADDR_ERR;
1594
1595         proc_vmcore = proc_create("vmcore", S_IRUSR, NULL, &vmcore_proc_ops);
1596         if (proc_vmcore)
1597                 proc_vmcore->size = vmcore_size;
1598         return 0;
1599 }
1600 fs_initcall(vmcore_init);
1601
1602 /* Cleanup function for vmcore module. */
1603 void vmcore_cleanup(void)
1604 {
1605         if (proc_vmcore) {
1606                 proc_remove(proc_vmcore);
1607                 proc_vmcore = NULL;
1608         }
1609
1610         /* clear the vmcore list. */
1611         while (!list_empty(&vmcore_list)) {
1612                 struct vmcore *m;
1613
1614                 m = list_first_entry(&vmcore_list, struct vmcore, list);
1615                 list_del(&m->list);
1616                 kfree(m);
1617         }
1618         free_elfcorebuf();
1619
1620         /* clear vmcore device dump list */
1621         vmcore_free_device_dumps();
1622 }