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