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