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