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/config.h>
  13 #include <linux/mm.h>
  14 #include <linux/proc_fs.h>
  15 #include <linux/user.h>
  16 #include <linux/a.out.h>
  17 #include <linux/elf.h>
  18 #include <linux/elfcore.h>
  19 #include <linux/vmalloc.h>
  20 #include <linux/highmem.h>
  21 #include <asm/uaccess.h>
  22 #include <asm/io.h>
  23
  24
  25 static int open_kcore(struct inode * inode, struct file * filp)
  26 {
  27         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
  28 }
  29
  30 static ssize_t read_kcore(struct file *, char __user *, size_t, loff_t *);
  31
  32 struct file_operations proc_kcore_operations = {
  33         .read           = read_kcore,
  34         .open           = open_kcore,
  35 };
  36
  37 #ifndef kc_vaddr_to_offset
  38 #define kc_vaddr_to_offset(v) ((v) - PAGE_OFFSET)
  39 #endif
  40 #ifndef kc_offset_to_vaddr
  41 #define kc_offset_to_vaddr(o) ((o) + PAGE_OFFSET)
  42 #endif
  43
  44 #define roundup(x, y)  ((((x)+((y)-1))/(y))*(y))
  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 struct kcore_list *kclist;
  56 static rwlock_t kclist_lock = RW_LOCK_UNLOCKED;
  57
  58 void
  59 kclist_add(struct kcore_list *new, void *addr, size_t size)
  60 {
  61         new->addr = (unsigned long)addr;
  62         new->size = size;
  63
  64         write_lock(&kclist_lock);
  65         new->next = kclist;
  66         kclist = new;
  67         write_unlock(&kclist_lock);
  68 }
  69
  70 struct kcore_list *
  71 kclist_del(void *addr)
  72 {
  73         struct kcore_list *m, **p = &kclist;
  74
  75         write_lock(&kclist_lock);
  76         for (m = *p; m; p = &m->next) {
  77                 if (m->addr == (unsigned long)addr) {
  78                         *p = m->next;
  79                         write_unlock(&kclist_lock);
  80                         return m;
  81                 }
  82         }
  83         write_unlock(&kclist_lock);
  84         return 0;
  85 }
  86
  87 extern char saved_command_line[];
  88
  89 static size_t get_kcore_size(int *nphdr, size_t *elf_buflen)
  90 {
  91         size_t try, size;
  92         struct kcore_list *m;
  93
  94         *nphdr = 1; /* PT_NOTE */
  95         size = 0;
  96
  97         for (m=kclist; m; m=m->next) {
  98                 try = kc_vaddr_to_offset((size_t)m->addr + m->size);
  99                 if (try > size)
 100                         size = try;
 101                 *nphdr = *nphdr + 1;
 102         }
 103         *elf_buflen =   sizeof(struct elfhdr) +
 104                         (*nphdr + 2)*sizeof(struct elf_phdr) +
 105                         3 * sizeof(struct memelfnote) +
 106                         sizeof(struct elf_prstatus) +
 107                         sizeof(struct elf_prpsinfo) +
 108                         sizeof(struct task_struct);
 109         *elf_buflen = PAGE_ALIGN(*elf_buflen);
 110         return size + *elf_buflen;
 111 }
 112
 113
 114 /*****************************************************************************/
 115 /*
 116  * determine size of ELF note
 117  */
 118 static int notesize(struct memelfnote *en)
 119 {
 120         int sz;
 121
 122         sz = sizeof(struct elf_note);
 123         sz += roundup(strlen(en->name), 4);
 124         sz += roundup(en->datasz, 4);
 125
 126         return sz;
 127 } /* end notesize() */
 128
 129 /*****************************************************************************/
 130 /*
 131  * store a note in the header buffer
 132  */
 133 static char *storenote(struct memelfnote *men, char *bufp)
 134 {
 135         struct elf_note en;
 136
 137 #define DUMP_WRITE(addr,nr) do { memcpy(bufp,addr,nr); bufp += nr; } while(0)
 138
 139         en.n_namesz = strlen(men->name);
 140         en.n_descsz = men->datasz;
 141         en.n_type = men->type;
 142
 143         DUMP_WRITE(&en, sizeof(en));
 144         DUMP_WRITE(men->name, en.n_namesz);
 145
 146         /* XXX - cast from long long to long to avoid need for libgcc.a */
 147         bufp = (char*) roundup((unsigned long)bufp,4);
 148         DUMP_WRITE(men->data, men->datasz);
 149         bufp = (char*) roundup((unsigned long)bufp,4);
 150
 151 #undef DUMP_WRITE
 152
 153         return bufp;
 154 } /* end storenote() */
 155
 156 /*
 157  * store an ELF coredump header in the supplied buffer
 158  * nphdr is the number of elf_phdr to insert
 159  */
 160 static void elf_kcore_store_hdr(char *bufp, int nphdr, int dataoff)
 161 {
 162         struct elf_prstatus prstatus;   /* NT_PRSTATUS */
 163         struct elf_prpsinfo prpsinfo;   /* NT_PRPSINFO */
 164         struct elf_phdr *nhdr, *phdr;
 165         struct elfhdr *elf;
 166         struct memelfnote notes[3];
 167         off_t offset = 0;
 168         struct kcore_list *m;
 169
 170         /* setup ELF header */
 171         elf = (struct elfhdr *) bufp;
 172         bufp += sizeof(struct elfhdr);
 173         offset += sizeof(struct elfhdr);
 174         memcpy(elf->e_ident, ELFMAG, SELFMAG);
 175         elf->e_ident[EI_CLASS]  = ELF_CLASS;
 176         elf->e_ident[EI_DATA]   = ELF_DATA;
 177         elf->e_ident[EI_VERSION]= EV_CURRENT;
 178         elf->e_ident[EI_OSABI] = ELF_OSABI;
 179         memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
 180         elf->e_type     = ET_CORE;
 181         elf->e_machine  = ELF_ARCH;
 182         elf->e_version  = EV_CURRENT;
 183         elf->e_entry    = 0;
 184         elf->e_phoff    = sizeof(struct elfhdr);
 185         elf->e_shoff    = 0;
 186         elf->e_flags    = 0;
 187         elf->e_ehsize   = sizeof(struct elfhdr);
 188         elf->e_phentsize= sizeof(struct elf_phdr);
 189         elf->e_phnum    = nphdr;
 190         elf->e_shentsize= 0;
 191         elf->e_shnum    = 0;
 192         elf->e_shstrndx = 0;
 193
 194         /* setup ELF PT_NOTE program header */
 195         nhdr = (struct elf_phdr *) bufp;
 196         bufp += sizeof(struct elf_phdr);
 197         offset += sizeof(struct elf_phdr);
 198         nhdr->p_type    = PT_NOTE;
 199         nhdr->p_offset  = 0;
 200         nhdr->p_vaddr   = 0;
 201         nhdr->p_paddr   = 0;
 202         nhdr->p_filesz  = 0;
 203         nhdr->p_memsz   = 0;
 204         nhdr->p_flags   = 0;
 205         nhdr->p_align   = 0;
 206
 207         /* setup ELF PT_LOAD program header for every area */
 208         for (m=kclist; m; m=m->next) {
 209                 phdr = (struct elf_phdr *) bufp;
 210                 bufp += sizeof(struct elf_phdr);
 211                 offset += sizeof(struct elf_phdr);
 212
 213                 phdr->p_type    = PT_LOAD;
 214                 phdr->p_flags   = PF_R|PF_W|PF_X;
 215                 phdr->p_offset  = kc_vaddr_to_offset(m->addr) + dataoff;
 216                 phdr->p_vaddr   = (size_t)m->addr;
 217                 phdr->p_paddr   = 0;
 218                 phdr->p_filesz  = phdr->p_memsz = m->size;
 219                 phdr->p_align   = PAGE_SIZE;
 220         }
 221
 222         /*
 223          * Set up the notes in similar form to SVR4 core dumps made
 224          * with info from their /proc.
 225          */
 226         nhdr->p_offset  = offset;
 227
 228         /* set up the process status */
 229         notes[0].name = "CORE";
 230         notes[0].type = NT_PRSTATUS;
 231         notes[0].datasz = sizeof(struct elf_prstatus);
 232         notes[0].data = &prstatus;
 233
 234         memset(&prstatus, 0, sizeof(struct elf_prstatus));
 235
 236         nhdr->p_filesz  = notesize(&notes[0]);
 237         bufp = storenote(&notes[0], bufp);
 238
 239         /* set up the process info */
 240         notes[1].name   = "CORE";
 241         notes[1].type   = NT_PRPSINFO;
 242         notes[1].datasz = sizeof(struct elf_prpsinfo);
 243         notes[1].data   = &prpsinfo;
 244
 245         memset(&prpsinfo, 0, sizeof(struct elf_prpsinfo));
 246         prpsinfo.pr_state       = 0;
 247         prpsinfo.pr_sname       = 'R';
 248         prpsinfo.pr_zomb        = 0;
 249
 250         strcpy(prpsinfo.pr_fname, "vmlinux");
 251         strncpy(prpsinfo.pr_psargs, saved_command_line, ELF_PRARGSZ);
 252
 253         nhdr->p_filesz  += notesize(&notes[1]);
 254         bufp = storenote(&notes[1], bufp);
 255
 256         /* set up the task structure */
 257         notes[2].name   = "CORE";
 258         notes[2].type   = NT_TASKSTRUCT;
 259         notes[2].datasz = sizeof(struct task_struct);
 260         notes[2].data   = current;
 261
 262         nhdr->p_filesz  += notesize(&notes[2]);
 263         bufp = storenote(&notes[2], bufp);
 264
 265 } /* end elf_kcore_store_hdr() */
 266
 267 /*****************************************************************************/
 268 /*
 269  * read from the ELF header and then kernel memory
 270  */
 271 static ssize_t
 272 read_kcore(struct file *file, char __user *buffer, size_t buflen, loff_t *fpos)
 273 {
 274         ssize_t acc = 0;
 275         size_t size, tsz;
 276         size_t elf_buflen;
 277         int nphdr;
 278         unsigned long start;
 279
 280         read_lock(&kclist_lock);
 281         tsz =  get_kcore_size(&nphdr, &elf_buflen);
 282         proc_root_kcore->size = size = tsz + elf_buflen;
 283         if (buflen == 0 || *fpos >= size) {
 284                 read_unlock(&kclist_lock);
 285                 return 0;
 286         }
 287
 288         /* trim buflen to not go beyond EOF */
 289         if (buflen > size - *fpos)
 290                 buflen = size - *fpos;
 291
 292         /* construct an ELF core header if we'll need some of it */
 293         if (*fpos < elf_buflen) {
 294                 char * elf_buf;
 295
 296                 tsz = elf_buflen - *fpos;
 297                 if (buflen < tsz)
 298                         tsz = buflen;
 299                 elf_buf = kmalloc(elf_buflen, GFP_ATOMIC);
 300                 if (!elf_buf) {
 301                         read_unlock(&kclist_lock);
 302                         return -ENOMEM;
 303                 }
 304                 memset(elf_buf, 0, elf_buflen);
 305                 elf_kcore_store_hdr(elf_buf, nphdr, elf_buflen);
 306                 read_unlock(&kclist_lock);
 307                 if (copy_to_user(buffer, elf_buf + *fpos, tsz)) {
 308                         kfree(elf_buf);
 309                         return -EFAULT;
 310                 }
 311                 kfree(elf_buf);
 312                 buflen -= tsz;
 313                 *fpos += tsz;
 314                 buffer += tsz;
 315                 acc += tsz;
 316
 317                 /* leave now if filled buffer already */
 318                 if (buflen == 0)
 319                         return acc;
 320         } else
 321                 read_unlock(&kclist_lock);
 322
 323         /*
 324          * Check to see if our file offset matches with any of
 325          * the addresses in the elf_phdr on our list.
 326          */
 327         start = kc_offset_to_vaddr(*fpos - elf_buflen);
 328         if ((tsz = (PAGE_SIZE - (start & ~PAGE_MASK))) > buflen)
 329                 tsz = buflen;
 330
 331         while (buflen) {
 332                 struct kcore_list *m;
 333
 334                 read_lock(&kclist_lock);
 335                 for (m=kclist; m; m=m->next) {
 336                         if (start >= m->addr && start < (m->addr+m->size))
 337                                 break;
 338                 }
 339                 read_unlock(&kclist_lock);
 340
 341                 if (m == NULL) {
 342                         if (clear_user(buffer, tsz))
 343                                 return -EFAULT;
 344                 } else if ((start >= VMALLOC_START) && (start < VMALLOC_END)) {
 345                         char * elf_buf;
 346                         struct vm_struct *m;
 347                         unsigned long curstart = start;
 348                         unsigned long cursize = tsz;
 349
 350                         elf_buf = kmalloc(tsz, GFP_KERNEL);
 351                         if (!elf_buf)
 352                                 return -ENOMEM;
 353                         memset(elf_buf, 0, tsz);
 354
 355                         read_lock(&vmlist_lock);
 356                         for (m=vmlist; m && cursize; m=m->next) {
 357                                 unsigned long vmstart;
 358                                 unsigned long vmsize;
 359                                 unsigned long msize = m->size - PAGE_SIZE;
 360
 361                                 if (((unsigned long)m->addr + msize) <
 362                                                                 curstart)
 363                                         continue;
 364                                 if ((unsigned long)m->addr > (curstart +
 365                                                                 cursize))
 366                                         break;
 367                                 vmstart = (curstart < (unsigned long)m->addr ?
 368                                         (unsigned long)m->addr : curstart);
 369                                 if (((unsigned long)m->addr + msize) >
 370                                                         (curstart + cursize))
 371                                         vmsize = curstart + cursize - vmstart;
 372                                 else
 373                                         vmsize = (unsigned long)m->addr +
 374                                                         msize - vmstart;
 375                                 curstart = vmstart + vmsize;
 376                                 cursize -= vmsize;
 377                                 /* don't dump ioremap'd stuff! (TA) */
 378                                 if (m->flags & VM_IOREMAP)
 379                                         continue;
 380                                 memcpy(elf_buf + (vmstart - start),
 381                                         (char *)vmstart, vmsize);
 382                         }
 383                         read_unlock(&vmlist_lock);
 384                         if (copy_to_user(buffer, elf_buf, tsz)) {
 385                                 kfree(elf_buf);
 386                                 return -EFAULT;
 387                         }
 388                         kfree(elf_buf);
 389                 } else {
 390                         if (kern_addr_valid(start)) {
 391                                 unsigned long n;
 392
 393                                 n = copy_to_user(buffer, (char *)start, tsz);
 394                                 /*
 395                                  * We cannot distingush between fault on source
 396                                  * and fault on destination. When this happens
 397                                  * we clear too and hope it will trigger the
 398                                  * EFAULT again.
 399                                  */
 400                                 if (n) {
 401                                         if (clear_user(buffer + tsz - n,
 402                                                                 tsz - n))
 403                                                 return -EFAULT;
 404                                 }
 405                         } else {
 406                                 if (clear_user(buffer, tsz))
 407                                         return -EFAULT;
 408                         }
 409                 }
 410                 buflen -= tsz;
 411                 *fpos += tsz;
 412                 buffer += tsz;
 413                 acc += tsz;
 414                 start += tsz;
 415                 tsz = (buflen > PAGE_SIZE ? PAGE_SIZE : buflen);
 416         }
 417
 418         return acc;
 419 }