fs/proc/kcore.c

   1 /*
   2  *      fs/proc/kcore.c kernel ELF core dumper
   3  *
   4  *      Modelled on fs/exec.c:aout_core_dump()
   5  *      Jeremy Fitzhardinge <jeremy@sw.oz.au>
   6  *      ELF version written by David Howells <David.Howells@nexor.co.uk>
   7  *      Modified and incorporated into 2.3.x by Tigran Aivazian <tigran@veritas.com>
   8  *      Support to dump vmalloc'd areas (ELF only), Tigran Aivazian <tigran@veritas.com>
   9  *      Safe accesses to vmalloc/direct-mapped discontiguous areas, Kanoj Sarcar <kanoj@sgi.com>
  10  */
  11
  12 #include <linux/mm.h>
  13 #include <linux/proc_fs.h>
  14 #include <linux/user.h>
  15 #include <linux/capability.h>
  16 #include <linux/elf.h>
  17 #include <linux/elfcore.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/highmem.h>
  20 #include <linux/printk.h>
  21 #include <linux/bootmem.h>
  22 #include <linux/init.h>
  23 #include <linux/slab.h>
  24 #include <asm/uaccess.h>
  25 #include <asm/io.h>
  26 #include <linux/list.h>
  27 #include <linux/ioport.h>
  28 #include <linux/memory.h>
  29 #include <asm/sections.h>
  30
  31 #define CORE_STR "CORE"
  32
  33 #ifndef ELF_CORE_EFLAGS
  34 #define ELF_CORE_EFLAGS 0
  35 #endif
  36
  37 static struct proc_dir_entry *proc_root_kcore;
  38
  39
  40 #ifndef kc_vaddr_to_offset
  41 #define kc_vaddr_to_offset(v) ((v) - PAGE_OFFSET)
  42 #endif
  43 #ifndef kc_offset_to_vaddr
  44 #define kc_offset_to_vaddr(o) ((o) + PAGE_OFFSET)
  45 #endif
  46
  47 /* An ELF note in memory */
  48 struct memelfnote
  49 {
  50         const char *name;
  51         int type;
  52         unsigned int datasz;
  53         void *data;
  54 };
  55
  56 static LIST_HEAD(kclist_head);
  57 static DEFINE_RWLOCK(kclist_lock);
  58 static int kcore_need_update = 1;
  59
  60 void
  61 kclist_add(struct kcore_list *new, void *addr, size_t size, int type)
  62 {
  63         new->addr = (unsigned long)addr;
  64         new->size = size;
  65         new->type = type;
  66
  67         write_lock(&kclist_lock);
  68         list_add_tail(&new->list, &kclist_head);
  69         write_unlock(&kclist_lock);
  70 }
  71
  72 static size_t get_kcore_size(int *nphdr, size_t *elf_buflen)
  73 {
  74         size_t try, size;
  75         struct kcore_list *m;
  76
  77         *nphdr = 1; /* PT_NOTE */
  78         size = 0;
  79
  80         list_for_each_entry(m, &kclist_head, list) {
  81                 try = kc_vaddr_to_offset((size_t)m->addr + m->size);
  82                 if (try > size)
  83                         size = try;
  84                 *nphdr = *nphdr + 1;
  85         }
  86         *elf_buflen =   sizeof(struct elfhdr) +
  87                         (*nphdr + 2)*sizeof(struct elf_phdr) +
  88                         3 * ((sizeof(struct elf_note)) +
  89                              roundup(sizeof(CORE_STR), 4)) +
  90                         roundup(sizeof(struct elf_prstatus), 4) +
  91                         roundup(sizeof(struct elf_prpsinfo), 4) +
  92                         roundup(sizeof(struct task_struct), 4);
  93         *elf_buflen = PAGE_ALIGN(*elf_buflen);
  94         return size + *elf_buflen;
  95 }
  96
  97 static void free_kclist_ents(struct list_head *head)
  98 {
  99         struct kcore_list *tmp, *pos;
 100
 101         list_for_each_entry_safe(pos, tmp, head, list) {
 102                 list_del(&pos->list);
 103                 kfree(pos);
 104         }
 105 }
 106 /*
 107  * Replace all KCORE_RAM/KCORE_VMEMMAP information with passed list.
 108  */
 109 static void __kcore_update_ram(struct list_head *list)
 110 {
 111         int nphdr;
 112         size_t size;
 113         struct kcore_list *tmp, *pos;
 114         LIST_HEAD(garbage);
 115
 116         write_lock(&kclist_lock);
 117         if (kcore_need_update) {
 118                 list_for_each_entry_safe(pos, tmp, &kclist_head, list) {
 119                         if (pos->type == KCORE_RAM
 120                                 || pos->type == KCORE_VMEMMAP)
 121                                 list_move(&pos->list, &garbage);
 122                 }
 123                 list_splice_tail(list, &kclist_head);
 124         } else
 125                 list_splice(list, &garbage);
 126         kcore_need_update = 0;
 127         proc_root_kcore->size = get_kcore_size(&nphdr, &size);
 128         write_unlock(&kclist_lock);
 129
 130         free_kclist_ents(&garbage);
 131 }
 132
 133
 134 #ifdef CONFIG_HIGHMEM
 135 /*
 136  * If no highmem, we can assume [0...max_low_pfn) continuous range of memory
 137  * because memory hole is not as big as !HIGHMEM case.
 138  * (HIGHMEM is special because part of memory is _invisible_ from the kernel.)
 139  */
 140 static int kcore_update_ram(void)
 141 {
 142         LIST_HEAD(head);
 143         struct kcore_list *ent;
 144         int ret = 0;
 145
 146         ent = kmalloc(sizeof(*ent), GFP_KERNEL);
 147         if (!ent)
 148                 return -ENOMEM;
 149         ent->addr = (unsigned long)__va(0);
 150         ent->size = max_low_pfn << PAGE_SHIFT;
 151         ent->type = KCORE_RAM;
 152         list_add(&ent->list, &head);
 153         __kcore_update_ram(&head);
 154         return ret;
 155 }
 156
 157 #else /* !CONFIG_HIGHMEM */
 158
 159 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 160 /* calculate vmemmap's address from given system ram pfn and register it */
 161 static int
 162 get_sparsemem_vmemmap_info(struct kcore_list *ent, struct list_head *head)
 163 {
 164         unsigned long pfn = __pa(ent->addr) >> PAGE_SHIFT;
 165         unsigned long nr_pages = ent->size >> PAGE_SHIFT;
 166         unsigned long start, end;
 167         struct kcore_list *vmm, *tmp;
 168
 169
 170         start = ((unsigned long)pfn_to_page(pfn)) & PAGE_MASK;
 171         end = ((unsigned long)pfn_to_page(pfn + nr_pages)) - 1;
 172         end = ALIGN(end, PAGE_SIZE);
 173         /* overlap check (because we have to align page */
 174         list_for_each_entry(tmp, head, list) {
 175                 if (tmp->type != KCORE_VMEMMAP)
 176                         continue;
 177                 if (start < tmp->addr + tmp->size)
 178                         if (end > tmp->addr)
 179                                 end = tmp->addr;
 180         }
 181         if (start < end) {
 182                 vmm = kmalloc(sizeof(*vmm), GFP_KERNEL);
 183                 if (!vmm)
 184                         return 0;
 185                 vmm->addr = start;
 186                 vmm->size = end - start;
 187                 vmm->type = KCORE_VMEMMAP;
 188                 list_add_tail(&vmm->list, head);
 189         }
 190         return 1;
 191
 192 }
 193 #else
 194 static int
 195 get_sparsemem_vmemmap_info(struct kcore_list *ent, struct list_head *head)
 196 {
 197         return 1;
 198 }
 199
 200 #endif
 201
 202 static int
 203 kclist_add_private(unsigned long pfn, unsigned long nr_pages, void *arg)
 204 {
 205         struct list_head *head = (struct list_head *)arg;
 206         struct kcore_list *ent;
 207
 208         ent = kmalloc(sizeof(*ent), GFP_KERNEL);
 209         if (!ent)
 210                 return -ENOMEM;
 211         ent->addr = (unsigned long)__va((pfn << PAGE_SHIFT));
 212         ent->size = nr_pages << PAGE_SHIFT;
 213
 214         /* Sanity check: Can happen in 32bit arch...maybe */
 215         if (ent->addr < (unsigned long) __va(0))
 216                 goto free_out;
 217
 218         /* cut not-mapped area. ....from ppc-32 code. */
 219         if (ULONG_MAX - ent->addr < ent->size)
 220                 ent->size = ULONG_MAX - ent->addr;
 221
 222         /* cut when vmalloc() area is higher than direct-map area */
 223         if (VMALLOC_START > (unsigned long)__va(0)) {
 224                 if (ent->addr > VMALLOC_START)
 225                         goto free_out;
 226                 if (VMALLOC_START - ent->addr < ent->size)
 227                         ent->size = VMALLOC_START - ent->addr;
 228         }
 229
 230         ent->type = KCORE_RAM;
 231         list_add_tail(&ent->list, head);
 232
 233         if (!get_sparsemem_vmemmap_info(ent, head)) {
 234                 list_del(&ent->list);
 235                 goto free_out;
 236         }
 237
 238         return 0;
 239 free_out:
 240         kfree(ent);
 241         return 1;
 242 }
 243
 244 static int kcore_update_ram(void)
 245 {
 246         int nid, ret;
 247         unsigned long end_pfn;
 248         LIST_HEAD(head);
 249
 250         /* Not inialized....update now */
 251         /* find out "max pfn" */
 252         end_pfn = 0;
 253         for_each_node_state(nid, N_MEMORY) {
 254                 unsigned long node_end;
 255                 node_end  = NODE_DATA(nid)->node_start_pfn +
 256                         NODE_DATA(nid)->node_spanned_pages;
 257                 if (end_pfn < node_end)
 258                         end_pfn = node_end;
 259         }
 260         /* scan 0 to max_pfn */
 261         ret = walk_system_ram_range(0, end_pfn, &head, kclist_add_private);
 262         if (ret) {
 263                 free_kclist_ents(&head);
 264                 return -ENOMEM;
 265         }
 266         __kcore_update_ram(&head);
 267         return ret;
 268 }
 269 #endif /* CONFIG_HIGHMEM */
 270
 271 /*****************************************************************************/
 272 /*
 273  * determine size of ELF note
 274  */
 275 static int notesize(struct memelfnote *en)
 276 {
 277         int sz;
 278
 279         sz = sizeof(struct elf_note);
 280         sz += roundup((strlen(en->name) + 1), 4);
 281         sz += roundup(en->datasz, 4);
 282
 283         return sz;
 284 } /* end notesize() */
 285
 286 /*****************************************************************************/
 287 /*
 288  * store a note in the header buffer
 289  */
 290 static char *storenote(struct memelfnote *men, char *bufp)
 291 {
 292         struct elf_note en;
 293
 294 #define DUMP_WRITE(addr,nr) do { memcpy(bufp,addr,nr); bufp += nr; } while(0)
 295
 296         en.n_namesz = strlen(men->name) + 1;
 297         en.n_descsz = men->datasz;
 298         en.n_type = men->type;
 299
 300         DUMP_WRITE(&en, sizeof(en));
 301         DUMP_WRITE(men->name, en.n_namesz);
 302
 303         /* XXX - cast from long long to long to avoid need for libgcc.a */
 304         bufp = (char*) roundup((unsigned long)bufp,4);
 305         DUMP_WRITE(men->data, men->datasz);
 306         bufp = (char*) roundup((unsigned long)bufp,4);
 307
 308 #undef DUMP_WRITE
 309
 310         return bufp;
 311 } /* end storenote() */
 312
 313 /*
 314  * store an ELF coredump header in the supplied buffer
 315  * nphdr is the number of elf_phdr to insert
 316  */
 317 static void elf_kcore_store_hdr(char *bufp, int nphdr, int dataoff)
 318 {
 319         struct elf_prstatus prstatus;   /* NT_PRSTATUS */
 320         struct elf_prpsinfo prpsinfo;   /* NT_PRPSINFO */
 321         struct elf_phdr *nhdr, *phdr;
 322         struct elfhdr *elf;
 323         struct memelfnote notes[3];
 324         off_t offset = 0;
 325         struct kcore_list *m;
 326
 327         /* setup ELF header */
 328         elf = (struct elfhdr *) bufp;
 329         bufp += sizeof(struct elfhdr);
 330         offset += sizeof(struct elfhdr);
 331         memcpy(elf->e_ident, ELFMAG, SELFMAG);
 332         elf->e_ident[EI_CLASS]  = ELF_CLASS;
 333         elf->e_ident[EI_DATA]   = ELF_DATA;
 334         elf->e_ident[EI_VERSION]= EV_CURRENT;
 335         elf->e_ident[EI_OSABI] = ELF_OSABI;
 336         memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
 337         elf->e_type     = ET_CORE;
 338         elf->e_machine  = ELF_ARCH;
 339         elf->e_version  = EV_CURRENT;
 340         elf->e_entry    = 0;
 341         elf->e_phoff    = sizeof(struct elfhdr);
 342         elf->e_shoff    = 0;
 343         elf->e_flags    = ELF_CORE_EFLAGS;
 344         elf->e_ehsize   = sizeof(struct elfhdr);
 345         elf->e_phentsize= sizeof(struct elf_phdr);
 346         elf->e_phnum    = nphdr;
 347         elf->e_shentsize= 0;
 348         elf->e_shnum    = 0;
 349         elf->e_shstrndx = 0;
 350
 351         /* setup ELF PT_NOTE program header */
 352         nhdr = (struct elf_phdr *) bufp;
 353         bufp += sizeof(struct elf_phdr);
 354         offset += sizeof(struct elf_phdr);
 355         nhdr->p_type    = PT_NOTE;
 356         nhdr->p_offset  = 0;
 357         nhdr->p_vaddr   = 0;
 358         nhdr->p_paddr   = 0;
 359         nhdr->p_filesz  = 0;
 360         nhdr->p_memsz   = 0;
 361         nhdr->p_flags   = 0;
 362         nhdr->p_align   = 0;
 363
 364         /* setup ELF PT_LOAD program header for every area */
 365         list_for_each_entry(m, &kclist_head, list) {
 366                 phdr = (struct elf_phdr *) bufp;
 367                 bufp += sizeof(struct elf_phdr);
 368                 offset += sizeof(struct elf_phdr);
 369
 370                 phdr->p_type    = PT_LOAD;
 371                 phdr->p_flags   = PF_R|PF_W|PF_X;
 372                 phdr->p_offset  = kc_vaddr_to_offset(m->addr) + dataoff;
 373                 phdr->p_vaddr   = (size_t)m->addr;
 374                 phdr->p_paddr   = 0;
 375                 phdr->p_filesz  = phdr->p_memsz = m->size;
 376                 phdr->p_align   = PAGE_SIZE;
 377         }
 378
 379         /*
 380          * Set up the notes in similar form to SVR4 core dumps made
 381          * with info from their /proc.
 382          */
 383         nhdr->p_offset  = offset;
 384
 385         /* set up the process status */
 386         notes[0].name = CORE_STR;
 387         notes[0].type = NT_PRSTATUS;
 388         notes[0].datasz = sizeof(struct elf_prstatus);
 389         notes[0].data = &prstatus;
 390
 391         memset(&prstatus, 0, sizeof(struct elf_prstatus));
 392
 393         nhdr->p_filesz  = notesize(&notes[0]);
 394         bufp = storenote(&notes[0], bufp);
 395
 396         /* set up the process info */
 397         notes[1].name   = CORE_STR;
 398         notes[1].type   = NT_PRPSINFO;
 399         notes[1].datasz = sizeof(struct elf_prpsinfo);
 400         notes[1].data   = &prpsinfo;
 401
 402         memset(&prpsinfo, 0, sizeof(struct elf_prpsinfo));
 403         prpsinfo.pr_state       = 0;
 404         prpsinfo.pr_sname       = 'R';
 405         prpsinfo.pr_zomb        = 0;
 406
 407         strcpy(prpsinfo.pr_fname, "vmlinux");
 408         strncpy(prpsinfo.pr_psargs, saved_command_line, ELF_PRARGSZ);
 409
 410         nhdr->p_filesz  += notesize(&notes[1]);
 411         bufp = storenote(&notes[1], bufp);
 412
 413         /* set up the task structure */
 414         notes[2].name   = CORE_STR;
 415         notes[2].type   = NT_TASKSTRUCT;
 416         notes[2].datasz = sizeof(struct task_struct);
 417         notes[2].data   = current;
 418
 419         nhdr->p_filesz  += notesize(&notes[2]);
 420         bufp = storenote(&notes[2], bufp);
 421
 422 } /* end elf_kcore_store_hdr() */
 423
 424 /*****************************************************************************/
 425 /*
 426  * read from the ELF header and then kernel memory
 427  */
 428 static ssize_t
 429 read_kcore(struct file *file, char __user *buffer, size_t buflen, loff_t *fpos)
 430 {
 431         ssize_t acc = 0;
 432         size_t size, tsz;
 433         size_t elf_buflen;
 434         int nphdr;
 435         unsigned long start;
 436
 437         read_lock(&kclist_lock);
 438         size = get_kcore_size(&nphdr, &elf_buflen);
 439
 440         if (buflen == 0 || *fpos >= size) {
 441                 read_unlock(&kclist_lock);
 442                 return 0;
 443         }
 444
 445         /* trim buflen to not go beyond EOF */
 446         if (buflen > size - *fpos)
 447                 buflen = size - *fpos;
 448
 449         /* construct an ELF core header if we'll need some of it */
 450         if (*fpos < elf_buflen) {
 451                 char * elf_buf;
 452
 453                 tsz = elf_buflen - *fpos;
 454                 if (buflen < tsz)
 455                         tsz = buflen;
 456                 elf_buf = kzalloc(elf_buflen, GFP_ATOMIC);
 457                 if (!elf_buf) {
 458                         read_unlock(&kclist_lock);
 459                         return -ENOMEM;
 460                 }
 461                 elf_kcore_store_hdr(elf_buf, nphdr, elf_buflen);
 462                 read_unlock(&kclist_lock);
 463                 if (copy_to_user(buffer, elf_buf + *fpos, tsz)) {
 464                         kfree(elf_buf);
 465                         return -EFAULT;
 466                 }
 467                 kfree(elf_buf);
 468                 buflen -= tsz;
 469                 *fpos += tsz;
 470                 buffer += tsz;
 471                 acc += tsz;
 472
 473                 /* leave now if filled buffer already */
 474                 if (buflen == 0)
 475                         return acc;
 476         } else
 477                 read_unlock(&kclist_lock);
 478
 479         /*
 480          * Check to see if our file offset matches with any of
 481          * the addresses in the elf_phdr on our list.
 482          */
 483         start = kc_offset_to_vaddr(*fpos - elf_buflen);
 484         if ((tsz = (PAGE_SIZE - (start & ~PAGE_MASK))) > buflen)
 485                 tsz = buflen;
 486
 487         while (buflen) {
 488                 struct kcore_list *m;
 489
 490                 read_lock(&kclist_lock);
 491                 list_for_each_entry(m, &kclist_head, list) {
 492                         if (start >= m->addr && start < (m->addr+m->size))
 493                                 break;
 494                 }
 495                 read_unlock(&kclist_lock);
 496
 497                 if (&m->list == &kclist_head) {
 498                         if (clear_user(buffer, tsz))
 499                                 return -EFAULT;
 500                 } else if (is_vmalloc_or_module_addr((void *)start)) {
 501                         char * elf_buf;
 502
 503                         elf_buf = kzalloc(tsz, GFP_KERNEL);
 504                         if (!elf_buf)
 505                                 return -ENOMEM;
 506                         vread(elf_buf, (char *)start, tsz);
 507                         /* we have to zero-fill user buffer even if no read */
 508                         if (copy_to_user(buffer, elf_buf, tsz)) {
 509                                 kfree(elf_buf);
 510                                 return -EFAULT;
 511                         }
 512                         kfree(elf_buf);
 513                 } else {
 514                         if (kern_addr_valid(start)) {
 515                                 unsigned long n;
 516
 517                                 n = copy_to_user(buffer, (char *)start, tsz);
 518                                 /*
 519                                  * We cannot distinguish between fault on source
 520                                  * and fault on destination. When this happens
 521                                  * we clear too and hope it will trigger the
 522                                  * EFAULT again.
 523                                  */
 524                                 if (n) {
 525                                         if (clear_user(buffer + tsz - n,
 526                                                                 n))
 527                                                 return -EFAULT;
 528                                 }
 529                         } else {
 530                                 if (clear_user(buffer, tsz))
 531                                         return -EFAULT;
 532                         }
 533                 }
 534                 buflen -= tsz;
 535                 *fpos += tsz;
 536                 buffer += tsz;
 537                 acc += tsz;
 538                 start += tsz;
 539                 tsz = (buflen > PAGE_SIZE ? PAGE_SIZE : buflen);
 540         }
 541
 542         return acc;
 543 }
 544
 545
 546 static int open_kcore(struct inode *inode, struct file *filp)
 547 {
 548         if (!capable(CAP_SYS_RAWIO))
 549                 return -EPERM;
 550         if (kcore_need_update)
 551                 kcore_update_ram();
 552         if (i_size_read(inode) != proc_root_kcore->size) {
 553                 mutex_lock(&inode->i_mutex);
 554                 i_size_write(inode, proc_root_kcore->size);
 555                 mutex_unlock(&inode->i_mutex);
 556         }
 557         return 0;
 558 }
 559
 560
 561 static const struct file_operations proc_kcore_operations = {
 562         .read           = read_kcore,
 563         .open           = open_kcore,
 564         .llseek         = default_llseek,
 565 };
 566
 567 #ifdef CONFIG_MEMORY_HOTPLUG
 568 /* just remember that we have to update kcore */
 569 static int __meminit kcore_callback(struct notifier_block *self,
 570                                     unsigned long action, void *arg)
 571 {
 572         switch (action) {
 573         case MEM_ONLINE:
 574         case MEM_OFFLINE:
 575                 write_lock(&kclist_lock);
 576                 kcore_need_update = 1;
 577                 write_unlock(&kclist_lock);
 578         }
 579         return NOTIFY_OK;
 580 }
 581 #endif
 582
 583
 584 static struct kcore_list kcore_vmalloc;
 585
 586 #ifdef CONFIG_ARCH_PROC_KCORE_TEXT
 587 static struct kcore_list kcore_text;
 588 /*
 589  * If defined, special segment is used for mapping kernel text instead of
 590  * direct-map area. We need to create special TEXT section.
 591  */
 592 static void __init proc_kcore_text_init(void)
 593 {
 594         kclist_add(&kcore_text, _text, _end - _text, KCORE_TEXT);
 595 }
 596 #else
 597 static void __init proc_kcore_text_init(void)
 598 {
 599 }
 600 #endif
 601
 602 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
 603 /*
 604  * MODULES_VADDR has no intersection with VMALLOC_ADDR.
 605  */
 606 struct kcore_list kcore_modules;
 607 static void __init add_modules_range(void)
 608 {
 609         kclist_add(&kcore_modules, (void *)MODULES_VADDR,
 610                         MODULES_END - MODULES_VADDR, KCORE_VMALLOC);
 611 }
 612 #else
 613 static void __init add_modules_range(void)
 614 {
 615 }
 616 #endif
 617
 618 static int __init proc_kcore_init(void)
 619 {
 620         proc_root_kcore = proc_create("kcore", S_IRUSR, NULL,
 621                                       &proc_kcore_operations);
 622         if (!proc_root_kcore) {
 623                 pr_err("couldn't create /proc/kcore\n");
 624                 return 0; /* Always returns 0. */
 625         }
 626         /* Store text area if it's special */
 627         proc_kcore_text_init();
 628         /* Store vmalloc area */
 629         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
 630                 VMALLOC_END - VMALLOC_START, KCORE_VMALLOC);
 631         add_modules_range();
 632         /* Store direct-map area from physical memory map */
 633         kcore_update_ram();
 634         hotplug_memory_notifier(kcore_callback, 0);
 635
 636         return 0;
 637 }
 638 module_init(proc_kcore_init);