kprobe/dbi_kprobes_deps.c

   1 /*
   2  *  Dynamic Binary Instrumentation Module based on KProbes
   3  *  modules/kprobe/dbi_kprobes_deps.h
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License as published by
   7  * the Free Software Foundation; either version 2 of the License, or
   8  * (at your option) any later version.
   9  *
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License
  16  * along with this program; if not, write to the Free Software
  17  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  18  *
  19  * Copyright (C) Samsung Electronics, 2006-2010
  20  *
  21  * 2008-2009    Alexey Gerenkov <a.gerenkov@samsung.com> User-Space
  22  *              Probes initial implementation; Support x86/ARM/MIPS for both user and kernel spaces.
  23  * 2010         Ekaterina Gorelkina <e.gorelkina@samsung.com>: redesign module for separating core and arch parts
  24  *
  25
  26  */
  27
  28 #include <linux/module.h>
  29 #include <linux/sched.h>
  30
  31 #include <asm/pgtable.h>
  32
  33 #include "dbi_kprobes_deps.h"
  34 #include "dbi_kdebug.h"
  35
  36
  37 unsigned int *sched_addr;
  38 unsigned int *fork_addr;
  39
  40
  41 #define GUP_FLAGS_WRITE                  0x1
  42 #define GUP_FLAGS_WRITE                  0x1
  43 #define GUP_FLAGS_FORCE                  0x2
  44 #define GUP_FLAGS_IGNORE_VMA_PERMISSIONS 0x4
  45 #define GUP_FLAGS_IGNORE_SIGKILL         0x8
  46
  47 DECLARE_MOD_CB_DEP(kallsyms_search, unsigned long, const char *name);
  48
  49 DECLARE_MOD_FUNC_DEP(access_process_vm, int, struct task_struct * tsk, unsigned long addr, void *buf, int len, int write);
  50
  51 DECLARE_MOD_FUNC_DEP(find_extend_vma, struct vm_area_struct *, struct mm_struct * mm, unsigned long addr);
  52
  53 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 30)
  54 #if LINUX_VERSION_CODE > KERNEL_VERSION(2, 6, 18)
  55 DECLARE_MOD_FUNC_DEP(handle_mm_fault, int, struct mm_struct *mm, struct vm_area_struct *vma, unsigned long address, int write_access);
  56 #endif
  57 #else
  58 DECLARE_MOD_FUNC_DEP(handle_mm_fault, int, struct mm_struct *mm, struct vm_area_struct *vma, unsigned long address, unsigned int flags);
  59 #endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 30) */
  60
  61 DECLARE_MOD_FUNC_DEP(get_gate_vma, struct vm_area_struct *, struct task_struct *tsk);
  62 DECLARE_MOD_FUNC_DEP(in_gate_area_no_task, int, unsigned long addr);
  63 DECLARE_MOD_FUNC_DEP(follow_page, \
  64                 struct page *, struct vm_area_struct * vma, \
  65                 unsigned long address, unsigned int foll_flags);
  66 DECLARE_MOD_FUNC_DEP(__flush_anon_page, \
  67                 void, struct vm_area_struct *vma, struct page *page, \
  68                 unsigned long vmaddr);
  69 DECLARE_MOD_FUNC_DEP(vm_normal_page, \
  70                 struct page *, struct vm_area_struct *vma, \
  71                 unsigned long addr, pte_t pte);
  72
  73 DECLARE_MOD_FUNC_DEP(flush_ptrace_access, \
  74                 void, struct vm_area_struct *vma, struct page *page, \
  75                 unsigned long uaddr, void *kaddr, unsigned long len, int write);
  76
  77
  78 #if (LINUX_VERSION_CODE != KERNEL_VERSION(2, 6, 16))
  79 DECLARE_MOD_FUNC_DEP(put_task_struct, \
  80                 void, struct task_struct *tsk);
  81 #else
  82 DECLARE_MOD_FUNC_DEP(put_task_struct, \
  83                 void, struct rcu_head * rhp);
  84 #endif
  85
  86         DECLARE_MOD_DEP_WRAPPER(access_process_vm, int, struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
  87 IMP_MOD_DEP_WRAPPER (access_process_vm, tsk, addr, buf, len, write)
  88
  89         DECLARE_MOD_DEP_WRAPPER (find_extend_vma, struct vm_area_struct *, struct mm_struct * mm, unsigned long addr)
  90 IMP_MOD_DEP_WRAPPER (find_extend_vma, mm, addr)
  91
  92 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 30)
  93 #if LINUX_VERSION_CODE > KERNEL_VERSION(2, 6, 18)
  94         DECLARE_MOD_DEP_WRAPPER (handle_mm_fault, \
  95                         int, struct mm_struct *mm, struct vm_area_struct *vma, unsigned long address, int write_access)
  96 IMP_MOD_DEP_WRAPPER (handle_mm_fault, mm, vma, address, write_access)
  97 #endif
  98 #else
  99         DECLARE_MOD_DEP_WRAPPER (handle_mm_fault, \
 100                         int, struct mm_struct *mm, struct vm_area_struct *vma, unsigned long address, unsigned int flags)
 101 IMP_MOD_DEP_WRAPPER (handle_mm_fault, mm, vma, address, flags)
 102 #endif
 103
 104         DECLARE_MOD_DEP_WRAPPER (get_gate_vma, \
 105                         struct vm_area_struct *, struct task_struct *tsk)
 106 IMP_MOD_DEP_WRAPPER (get_gate_vma, tsk)
 107
 108         DECLARE_MOD_DEP_WRAPPER (in_gate_area_no_task, int, unsigned long addr)
 109 IMP_MOD_DEP_WRAPPER (in_gate_area_no_task, addr)
 110
 111         DECLARE_MOD_DEP_WRAPPER (follow_page, \
 112                         struct page *, struct vm_area_struct * vma, \
 113                         unsigned long address, unsigned int foll_flags)
 114 IMP_MOD_DEP_WRAPPER (follow_page, vma, address, foll_flags)
 115
 116         DECLARE_MOD_DEP_WRAPPER (__flush_anon_page, \
 117                         void, struct vm_area_struct *vma, \
 118                         struct page *page, unsigned long vmaddr)
 119 IMP_MOD_DEP_WRAPPER (__flush_anon_page, vma, page, vmaddr)
 120
 121         DECLARE_MOD_DEP_WRAPPER(vm_normal_page, \
 122                         struct page *, struct vm_area_struct *vma, \
 123                         unsigned long addr, pte_t pte)
 124 IMP_MOD_DEP_WRAPPER (vm_normal_page, vma, addr, pte)
 125
 126         DECLARE_MOD_DEP_WRAPPER (flush_ptrace_access, \
 127                         void, struct vm_area_struct *vma, struct page *page, \
 128                         unsigned long uaddr, void *kaddr, unsigned long len, int write)
 129 IMP_MOD_DEP_WRAPPER (flush_ptrace_access, vma, page, uaddr, kaddr, len, write)
 130
 131 int init_module_dependencies()
 132 {
 133 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 30)
 134 #if LINUX_VERSION_CODE > KERNEL_VERSION(2, 6, 18)
 135         INIT_MOD_DEP_VAR(handle_mm_fault, handle_mm_fault);
 136 #endif
 137 #endif
 138
 139         INIT_MOD_DEP_VAR(flush_ptrace_access, flush_ptrace_access);
 140         INIT_MOD_DEP_VAR(find_extend_vma, find_extend_vma);
 141         INIT_MOD_DEP_VAR(get_gate_vma, get_gate_vma);
 142         INIT_MOD_DEP_VAR(in_gate_area_no_task, in_gate_area_no_task);
 143         INIT_MOD_DEP_VAR(follow_page, follow_page);
 144         INIT_MOD_DEP_VAR(__flush_anon_page, __flush_anon_page);
 145         INIT_MOD_DEP_VAR(vm_normal_page, vm_normal_page);
 146         INIT_MOD_DEP_VAR(access_process_vm, access_process_vm);
 147
 148 #if (LINUX_VERSION_CODE != KERNEL_VERSION(2, 6, 16))
 149 # if (LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 11))
 150         INIT_MOD_DEP_VAR(put_task_struct, put_task_struct);
 151 # else
 152         INIT_MOD_DEP_VAR(put_task_struct, __put_task_struct);
 153 # endif
 154 #else /*2.6.16 */
 155         INIT_MOD_DEP_VAR(put_task_struct, __put_task_struct_cb);
 156 #endif
 157
 158         return 0;
 159 }
 160
 161 #define GUP_FLAGS_WRITE                  0x1
 162 #define GUP_FLAGS_FORCE                  0x2
 163 #define GUP_FLAGS_IGNORE_VMA_PERMISSIONS 0x4
 164 #define GUP_FLAGS_IGNORE_SIGKILL         0x8
 165
 166 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,18)
 167 static inline int use_zero_page(struct vm_area_struct *vma)
 168 {
 169         /*
 170          * We don't want to optimize FOLL_ANON for make_pages_present()
 171          * when it tries to page in a VM_LOCKED region. As to VM_SHARED,
 172          * we want to get the page from the page tables to make sure
 173          * that we serialize and update with any other user of that
 174          * mapping.
 175          */
 176         if (vma->vm_flags & (VM_LOCKED | VM_SHARED))
 177                 return 0;
 178         /*
 179          * And if we have a fault routine, it's not an anonymous region.
 180          */
 181         return !vma->vm_ops || !vma->vm_ops->fault;
 182 }
 183
 184 int __get_user_pages_uprobe(struct task_struct *tsk, struct mm_struct *mm,
 185                      unsigned long start, int len, int flags,
 186                 struct page **pages, struct vm_area_struct **vmas)
 187 {
 188         int i;
 189         unsigned int vm_flags = 0;
 190         int write = !!(flags & GUP_FLAGS_WRITE);
 191         int force = !!(flags & GUP_FLAGS_FORCE);
 192         int ignore = !!(flags & GUP_FLAGS_IGNORE_VMA_PERMISSIONS);
 193         int ignore_sigkill = !!(flags & GUP_FLAGS_IGNORE_SIGKILL);
 194
 195         if (len <= 0)
 196                 return 0;
 197         /*
 198          * Require read or write permissions.
 199          * If 'force' is set, we only require the "MAY" flags.
 200          */
 201         vm_flags  = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
 202         vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
 203         i = 0;
 204
 205         do {
 206                 struct vm_area_struct *vma;
 207                 unsigned int foll_flags;
 208
 209                 //vma = find_extend_vma(mm, start);
 210                 vma = find_vma(mm, start);
 211                 if (!vma && in_gate_area(tsk, start)) {
 212                         unsigned long pg = start & PAGE_MASK;
 213                         struct vm_area_struct *gate_vma = get_gate_vma(tsk);
 214                         pgd_t *pgd;
 215                         pud_t *pud;
 216                         pmd_t *pmd;
 217                         pte_t *pte;
 218
 219                         /* user gate pages are read-only */
 220                         if (!ignore && write)
 221                                 return i ? : -EFAULT;
 222                         if (pg > TASK_SIZE)
 223                                 pgd = pgd_offset_k(pg);
 224                         else
 225                                 pgd = pgd_offset_gate(mm, pg);
 226                         BUG_ON(pgd_none(*pgd));
 227                         pud = pud_offset(pgd, pg);
 228                         BUG_ON(pud_none(*pud));
 229                         pmd = pmd_offset(pud, pg);
 230                         if (pmd_none(*pmd))
 231                                 return i ? : -EFAULT;
 232                         pte = pte_offset_map(pmd, pg);
 233                         if (pte_none(*pte)) {
 234                                 pte_unmap(pte);
 235                                 return i ? : -EFAULT;
 236                         }
 237                         if (pages) {
 238                                 struct page *page = vm_normal_page(gate_vma, start, *pte);
 239                                 pages[i] = page;
 240                                 if (page)
 241                                         get_page(page);
 242                         }
 243                         pte_unmap(pte);
 244                         if (vmas)
 245                                 vmas[i] = gate_vma;
 246                         i++;
 247                         start += PAGE_SIZE;
 248                         len--;
 249                         continue;
 250                 }
 251
 252                 if (!vma ||
 253                     (vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
 254                     (!ignore && !(vm_flags & vma->vm_flags)))
 255                         return i ? : -EFAULT;
 256
 257                 if (is_vm_hugetlb_page(vma)) {
 258 #if  LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 259                         i = follow_hugetlb_page(mm, vma, pages, vmas,
 260                                                 &start, &len, i);
 261 #else
 262                         i = follow_hugetlb_page(mm, vma, pages, vmas,
 263                                                 &start, &len, i, write);
 264 #endif
 265                         continue;
 266                 }
 267
 268                 foll_flags = FOLL_TOUCH;
 269                 if (pages)
 270                         foll_flags |= FOLL_GET;
 271
 272 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,18)
 273 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,30)
 274                 if (!write && use_zero_page(vma))
 275                   foll_flags |= FOLL_ANON;
 276 #endif
 277 #endif
 278
 279                 do {
 280                         struct page *page;
 281
 282 #if 0
 283                         /*
 284                          * If we have a pending SIGKILL, don't keep faulting
 285                          * pages and potentially allocating memory, unless
 286                          * current is handling munlock--e.g., on exit. In
 287                          * that case, we are not allocating memory.  Rather,
 288                          * we're only unlocking already resident/mapped pages.
 289                          */
 290                         if (unlikely(!ignore_sigkill &&
 291                                         fatal_signal_pending(current)))
 292                                 return i ? i : -ERESTARTSYS;
 293 #endif
 294
 295                         if (write)
 296                                 foll_flags |= FOLL_WRITE;
 297
 298
 299                         //cond_resched();
 300
 301                         DBPRINTF ("pages = %p vma = %p\n", pages, vma);
 302                         while (!(page = follow_page(vma, start, foll_flags))) {
 303                                 int ret;
 304                                 ret = handle_mm_fault(mm, vma, start,
 305                                                 foll_flags & FOLL_WRITE);
 306
 307 #if  LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 308                                 if (ret & VM_FAULT_WRITE)
 309                                   foll_flags &= ~FOLL_WRITE;
 310
 311                                 switch (ret & ~VM_FAULT_WRITE) {
 312                                 case VM_FAULT_MINOR:
 313                                   tsk->min_flt++;
 314                                   break;
 315                                 case VM_FAULT_MAJOR:
 316                                   tsk->maj_flt++;
 317                                   break;
 318                                 case VM_FAULT_SIGBUS:
 319                                   return i ? i : -EFAULT;
 320                                 case VM_FAULT_OOM:
 321                                   return i ? i : -ENOMEM;
 322                                 default:
 323                                   BUG();
 324                                 }
 325
 326 #else
 327                                 if (ret & VM_FAULT_ERROR) {
 328                                   if (ret & VM_FAULT_OOM)
 329                                     return i ? i : -ENOMEM;
 330                                   else if (ret & VM_FAULT_SIGBUS)
 331                                     return i ? i : -EFAULT;
 332                                   BUG();
 333                                 }
 334                                 if (ret & VM_FAULT_MAJOR)
 335                                   tsk->maj_flt++;
 336                                 else
 337                                   tsk->min_flt++;
 338
 339                                 /*
 340                                  * The VM_FAULT_WRITE bit tells us that
 341                                  * do_wp_page has broken COW when necessary,
 342                                  * even if maybe_mkwrite decided not to set
 343                                  * pte_write. We can thus safely do subsequent
 344                                  * page lookups as if they were reads. But only
 345                                  * do so when looping for pte_write is futile:
 346                                  * in some cases userspace may also be wanting
 347                                  * to write to the gotten user page, which a
 348                                  * read fault here might prevent (a readonly
 349                                  * page might get reCOWed by userspace write).
 350                                  */
 351                                 if ((ret & VM_FAULT_WRITE) &&
 352                                     !(vma->vm_flags & VM_WRITE))
 353                                   foll_flags &= ~FOLL_WRITE;
 354
 355                                 //cond_resched();
 356 #endif
 357
 358                         }
 359
 360                         if (IS_ERR(page))
 361                                 return i ? i : PTR_ERR(page);
 362                         if (pages) {
 363                                 pages[i] = page;
 364
 365 #if  LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 366                                 flush_anon_page(page, start);
 367 #else
 368                                 flush_anon_page(vma, page, start);
 369 #endif
 370                                 flush_dcache_page(page);
 371                         }
 372                         if (vmas)
 373                                 vmas[i] = vma;
 374                         i++;
 375                         start += PAGE_SIZE;
 376                         len--;
 377                 } while (len && start < vma->vm_end);
 378         } while (len);
 379         return i;
 380 }
 381 #endif
 382
 383
 384 int get_user_pages_uprobe(struct task_struct *tsk, struct mm_struct *mm,
 385                 unsigned long start, int len, int write, int force,
 386                 struct page **pages, struct vm_area_struct **vmas)
 387 {
 388 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29)
 389         int flags = 0;
 390
 391         if (write)
 392                 flags |= GUP_FLAGS_WRITE;
 393         if (force)
 394                 flags |= GUP_FLAGS_FORCE;
 395
 396         return __get_user_pages_uprobe(tsk, mm,
 397                                 start, len, flags,
 398                                 pages, vmas);
 399 #else
 400         return get_user_pages(tsk, mm, start, len, write, force, pages, vmas);
 401 #endif
 402 }
 403
 404 int access_process_vm_atomic(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
 405 {
 406         struct mm_struct *mm;
 407         struct vm_area_struct *vma;
 408         void *old_buf = buf;
 409
 410         mm = get_task_mm(tsk);
 411         if (!mm)
 412                 return 0;
 413
 414         down_read(&mm->mmap_sem);
 415         /* ignore errors, just check how much was successfully transferred */
 416         while (len) {
 417                 int bytes, ret, offset;
 418                 void *maddr;
 419                 struct page *page = NULL;
 420
 421                 ret = get_user_pages_uprobe(tsk, mm, addr, 1,
 422                                 write, 1, &page, &vma);
 423                 if (ret <= 0) {
 424                         /*
 425                          * Check if this is a VM_IO | VM_PFNMAP VMA, which
 426                          * we can access using slightly different code.
 427                          */
 428 #ifdef CONFIG_HAVE_IOREMAP_PROT
 429                         vma = find_vma(mm, addr);
 430                         if (!vma)
 431                                 break;
 432                         if (vma->vm_ops && vma->vm_ops->access)
 433                                 ret = vma->vm_ops->access(vma, addr, buf,
 434                                                           len, write);
 435                         if (ret <= 0)
 436 #endif
 437                                 break;
 438                         bytes = ret;
 439                 } else {
 440                         bytes = len;
 441                         offset = addr & (PAGE_SIZE-1);
 442                         if (bytes > PAGE_SIZE-offset)
 443                                 bytes = PAGE_SIZE-offset;
 444
 445                         maddr = kmap(page);
 446                         if (write) {
 447                                 copy_to_user_page(vma, page, addr,
 448                                                   maddr + offset, buf, bytes);
 449                                 set_page_dirty_lock(page);
 450                         } else {
 451                                 copy_from_user_page(vma, page, addr,
 452                                                     buf, maddr + offset, bytes);
 453                         }
 454                         kunmap(page);
 455                         page_cache_release(page);
 456                 }
 457                 len -= bytes;
 458                 buf += bytes;
 459                 addr += bytes;
 460         }
 461         up_read(&mm->mmap_sem);
 462         mmput(mm);
 463
 464         return buf - old_buf;
 465 }
 466
 467 int page_present (struct mm_struct *mm, unsigned long address)
 468 {
 469         pgd_t *pgd;
 470         pud_t *pud;
 471         pmd_t *pmd;
 472         pte_t *ptep, pte;
 473         unsigned long pfn;
 474
 475         pgd = pgd_offset(mm, address);
 476         if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
 477                 goto out;
 478
 479         pud = pud_offset(pgd, address);
 480         if (pud_none(*pud) || unlikely(pud_bad(*pud)))
 481                 goto out;
 482
 483         pmd = pmd_offset(pud, address);
 484         if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
 485                 goto out;
 486
 487         ptep = pte_offset_map(pmd, address);
 488         if (!ptep)
 489                 goto out;
 490
 491         pte = *ptep;
 492         pte_unmap(ptep);
 493         if (pte_present(pte)) {
 494                 pfn = pte_pfn(pte);
 495                 if (pfn_valid(pfn)) {
 496                         return 1;
 497                 }
 498         }
 499
 500 out:
 501         return 0;
 502 }
 503
 504
 505 EXPORT_SYMBOL_GPL (page_present);
 506 EXPORT_SYMBOL_GPL (get_user_pages_uprobe);
 507 EXPORT_SYMBOL_GPL (access_process_vm_atomic);
 508