2 * Dynamic Binary Instrumentation Module based on KProbes
3 * modules/kprobe/dbi_kprobes_deps.h
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.
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.
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.
19 * Copyright (C) Samsung Electronics, 2006-2010
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
28 #include <linux/module.h>
29 #include <linux/sched.h>
31 #include <asm/pgtable.h>
33 #include "dbi_kprobes_deps.h"
34 #include "dbi_kdebug.h"
37 #include <linux/slab.h>
39 unsigned int *sched_addr;
40 unsigned int *fork_addr;
43 #define GUP_FLAGS_WRITE 0x1
44 #define GUP_FLAGS_WRITE 0x1
45 #define GUP_FLAGS_FORCE 0x2
46 #define GUP_FLAGS_IGNORE_VMA_PERMISSIONS 0x4
47 #define GUP_FLAGS_IGNORE_SIGKILL 0x8
49 #if LINUX_VERSION_CODE > KERNEL_VERSION(2, 6, 29)
50 struct mm_struct* init_mm_ptr;
51 struct mm_struct init_mm;
55 DECLARE_MOD_CB_DEP(kallsyms_search, unsigned long, const char *name);
57 DECLARE_MOD_FUNC_DEP(access_process_vm, int, struct task_struct * tsk, unsigned long addr, void *buf, int len, int write);
59 DECLARE_MOD_FUNC_DEP(find_extend_vma, struct vm_area_struct *, struct mm_struct * mm, unsigned long addr);
61 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 30)
62 #if LINUX_VERSION_CODE > KERNEL_VERSION(2, 6, 18)
63 DECLARE_MOD_FUNC_DEP(handle_mm_fault, int, struct mm_struct *mm, struct vm_area_struct *vma, unsigned long address, int write_access);
66 DECLARE_MOD_FUNC_DEP(handle_mm_fault, int, struct mm_struct *mm, struct vm_area_struct *vma, unsigned long address, unsigned int flags);
67 #endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 30) */
69 DECLARE_MOD_FUNC_DEP(get_gate_vma, struct vm_area_struct *, struct task_struct *tsk);
71 #ifdef CONFIG_HUGETLB_PAGE
72 DECLARE_MOD_FUNC_DEP(follow_hugetlb_page, int, struct mm_struct *mm, struct vm_area_struct *vma, struct page **pages, struct vm_area_struct **vmas, unsigned long *position, int *length, int i, int write);
75 #ifdef __HAVE_ARCH_GATE_AREA
76 DECLARE_MOD_FUNC_DEP(in_gate_area, int, struct task_struct *tsk,unsigned long addr);
78 DECLARE_MOD_FUNC_DEP(in_gate_area_no_task, int, unsigned long addr);
80 DECLARE_MOD_FUNC_DEP(follow_page, \
81 struct page *, struct vm_area_struct * vma, \
82 unsigned long address, unsigned int foll_flags);
83 DECLARE_MOD_FUNC_DEP(__flush_anon_page, \
84 void, struct vm_area_struct *vma, struct page *page, \
85 unsigned long vmaddr);
86 DECLARE_MOD_FUNC_DEP(vm_normal_page, \
87 struct page *, struct vm_area_struct *vma, \
88 unsigned long addr, pte_t pte);
90 DECLARE_MOD_FUNC_DEP(flush_ptrace_access, \
91 void, struct vm_area_struct *vma, struct page *page, \
92 unsigned long uaddr, void *kaddr, unsigned long len, int write);
95 #if (LINUX_VERSION_CODE != KERNEL_VERSION(2, 6, 16))
96 DECLARE_MOD_FUNC_DEP(put_task_struct, \
97 void, struct task_struct *tsk);
99 DECLARE_MOD_FUNC_DEP(put_task_struct, \
100 void, struct rcu_head * rhp);
103 DECLARE_MOD_DEP_WRAPPER(access_process_vm, int, struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
104 IMP_MOD_DEP_WRAPPER (access_process_vm, tsk, addr, buf, len, write)
106 DECLARE_MOD_DEP_WRAPPER (find_extend_vma, struct vm_area_struct *, struct mm_struct * mm, unsigned long addr)
107 IMP_MOD_DEP_WRAPPER (find_extend_vma, mm, addr)
109 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 30)
110 #if LINUX_VERSION_CODE > KERNEL_VERSION(2, 6, 18)
111 DECLARE_MOD_DEP_WRAPPER (handle_mm_fault, \
112 int, struct mm_struct *mm, struct vm_area_struct *vma, unsigned long address, int write_access)
113 IMP_MOD_DEP_WRAPPER (handle_mm_fault, mm, vma, address, write_access)
116 DECLARE_MOD_DEP_WRAPPER (handle_mm_fault, \
117 int, struct mm_struct *mm, struct vm_area_struct *vma, unsigned long address, unsigned int flags)
118 IMP_MOD_DEP_WRAPPER (handle_mm_fault, mm, vma, address, flags)
121 DECLARE_MOD_DEP_WRAPPER (get_gate_vma, \
122 struct vm_area_struct *, struct task_struct *tsk)
123 IMP_MOD_DEP_WRAPPER (get_gate_vma, tsk)
125 #ifdef CONFIG_HUGETLB_PAGE
126 DECLARE_MOD_DEP_WRAPPER (follow_hugetlb_page, int, struct mm_struct *mm, struct vm_area_struct *vma, struct page **pages, struct vm_area_struct **vmas, unsigned long *position, int *length, int i, int write)
127 IMP_MOD_DEP_WRAPPER (follow_hugetlb_page, mm, vma, pages, vmas, position, length, i, write)
130 #ifdef __HAVE_ARCH_GATE_AREA
131 DECLARE_MOD_DEP_WRAPPER (in_gate_area, int, struct task_struct *tsk, unsigned long addr)
132 IMP_MOD_DEP_WRAPPER (in_gate_area, tsk, addr)
134 DECLARE_MOD_DEP_WRAPPER (in_gate_area_no_task, int, unsigned long addr)
135 IMP_MOD_DEP_WRAPPER (in_gate_area_no_task, addr)
138 #if (LINUX_VERSION_CODE != KERNEL_VERSION(2, 6, 11))
139 DECLARE_MOD_DEP_WRAPPER (follow_page, \
140 struct page *, struct vm_area_struct * vma, \
141 unsigned long address, unsigned int foll_flags)
142 IMP_MOD_DEP_WRAPPER (follow_page, vma, address, foll_flags)
144 DECLARE_MOD_DEP_WRAPPER (__flush_anon_page, \
145 void, struct vm_area_struct *vma, \
146 struct page *page, unsigned long vmaddr)
147 IMP_MOD_DEP_WRAPPER (__flush_anon_page, vma, page, vmaddr)
149 DECLARE_MOD_DEP_WRAPPER(vm_normal_page, \
150 struct page *, struct vm_area_struct *vma, \
151 unsigned long addr, pte_t pte)
152 IMP_MOD_DEP_WRAPPER (vm_normal_page, vma, addr, pte)
154 DECLARE_MOD_DEP_WRAPPER (flush_ptrace_access, \
155 void, struct vm_area_struct *vma, struct page *page, \
156 unsigned long uaddr, void *kaddr, unsigned long len, int write)
157 IMP_MOD_DEP_WRAPPER (flush_ptrace_access, vma, page, uaddr, kaddr, len, write)
159 int init_module_dependencies()
162 #if LINUX_VERSION_CODE > KERNEL_VERSION(2, 6, 29)
163 init_mm_ptr = (struct mm_struct*) kallsyms_search ("init_mm");
164 memcmp(init_mm_ptr, &init_mm, sizeof(struct mm_struct));
167 #if LINUX_VERSION_CODE > KERNEL_VERSION(2, 6, 18)
168 INIT_MOD_DEP_VAR(handle_mm_fault, handle_mm_fault);
171 INIT_MOD_DEP_VAR(flush_ptrace_access, flush_ptrace_access);
172 INIT_MOD_DEP_VAR(find_extend_vma, find_extend_vma);
173 INIT_MOD_DEP_VAR(get_gate_vma, get_gate_vma);
175 #ifdef CONFIG_HUGETLB_PAGE
176 INIT_MOD_DEP_VAR(follow_hugetlb_page, follow_hugetlb_page);
179 #ifdef __HAVE_ARCH_GATE_AREA
180 INIT_MOD_DEP_VAR(in_gate_area, in_gate_area);
182 INIT_MOD_DEP_VAR(in_gate_area_no_task, in_gate_area_no_task);
184 INIT_MOD_DEP_VAR(follow_page, follow_page);
186 INIT_MOD_DEP_VAR(__flush_anon_page, __flush_anon_page);
187 INIT_MOD_DEP_VAR(vm_normal_page, vm_normal_page);
188 INIT_MOD_DEP_VAR(access_process_vm, access_process_vm);
190 #if (LINUX_VERSION_CODE != KERNEL_VERSION(2, 6, 16))
191 # if (LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 11))
192 INIT_MOD_DEP_VAR(put_task_struct, put_task_struct);
194 INIT_MOD_DEP_VAR(put_task_struct, __put_task_struct);
197 INIT_MOD_DEP_VAR(put_task_struct, __put_task_struct_cb);
203 #define GUP_FLAGS_WRITE 0x1
204 #define GUP_FLAGS_FORCE 0x2
205 #define GUP_FLAGS_IGNORE_VMA_PERMISSIONS 0x4
206 #define GUP_FLAGS_IGNORE_SIGKILL 0x8
208 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,18)
209 static inline int use_zero_page(struct vm_area_struct *vma)
212 * We don't want to optimize FOLL_ANON for make_pages_present()
213 * when it tries to page in a VM_LOCKED region. As to VM_SHARED,
214 * we want to get the page from the page tables to make sure
215 * that we serialize and update with any other user of that
218 if (vma->vm_flags & (VM_LOCKED | VM_SHARED))
221 * And if we have a fault routine, it's not an anonymous region.
223 return !vma->vm_ops || !vma->vm_ops->fault;
226 int __get_user_pages_uprobe(struct task_struct *tsk, struct mm_struct *mm,
227 unsigned long start, int len, int flags,
228 struct page **pages, struct vm_area_struct **vmas)
231 unsigned int vm_flags = 0;
232 int write = !!(flags & GUP_FLAGS_WRITE);
233 int force = !!(flags & GUP_FLAGS_FORCE);
234 int ignore = !!(flags & GUP_FLAGS_IGNORE_VMA_PERMISSIONS);
235 int ignore_sigkill = !!(flags & GUP_FLAGS_IGNORE_SIGKILL);
240 * Require read or write permissions.
241 * If 'force' is set, we only require the "MAY" flags.
243 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
244 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
248 struct vm_area_struct *vma;
249 unsigned int foll_flags;
251 //vma = find_extend_vma(mm, start);
252 vma = find_vma(mm, start);
253 if (!vma && in_gate_area(tsk, start)) {
254 unsigned long pg = start & PAGE_MASK;
255 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
261 /* user gate pages are read-only */
262 if (!ignore && write)
263 return i ? : -EFAULT;
265 pgd = pgd_offset_k(pg);
267 pgd = pgd_offset_gate(mm, pg);
268 BUG_ON(pgd_none(*pgd));
269 pud = pud_offset(pgd, pg);
270 BUG_ON(pud_none(*pud));
271 pmd = pmd_offset(pud, pg);
273 return i ? : -EFAULT;
274 pte = pte_offset_map(pmd, pg);
275 if (pte_none(*pte)) {
277 return i ? : -EFAULT;
280 struct page *page = vm_normal_page(gate_vma, start, *pte);
295 (vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
296 (!ignore && !(vm_flags & vma->vm_flags)))
297 return i ? : -EFAULT;
299 if (is_vm_hugetlb_page(vma)) {
300 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
301 i = follow_hugetlb_page(mm, vma, pages, vmas,
304 i = follow_hugetlb_page(mm, vma, pages, vmas,
305 &start, &len, i, write);
310 foll_flags = FOLL_TOUCH;
312 foll_flags |= FOLL_GET;
314 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,18)
315 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,30)
316 if (!write && use_zero_page(vma))
317 foll_flags |= FOLL_ANON;
326 * If we have a pending SIGKILL, don't keep faulting
327 * pages and potentially allocating memory, unless
328 * current is handling munlock--e.g., on exit. In
329 * that case, we are not allocating memory. Rather,
330 * we're only unlocking already resident/mapped pages.
332 if (unlikely(!ignore_sigkill &&
333 fatal_signal_pending(current)))
334 return i ? i : -ERESTARTSYS;
338 foll_flags |= FOLL_WRITE;
343 DBPRINTF ("pages = %p vma = %p\n", pages, vma);
344 while (!(page = follow_page(vma, start, foll_flags))) {
346 ret = handle_mm_fault(mm, vma, start,
347 foll_flags & FOLL_WRITE);
349 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
350 if (ret & VM_FAULT_WRITE)
351 foll_flags &= ~FOLL_WRITE;
353 switch (ret & ~VM_FAULT_WRITE) {
360 case VM_FAULT_SIGBUS:
361 return i ? i : -EFAULT;
363 return i ? i : -ENOMEM;
369 if (ret & VM_FAULT_ERROR) {
370 if (ret & VM_FAULT_OOM)
371 return i ? i : -ENOMEM;
372 else if (ret & VM_FAULT_SIGBUS)
373 return i ? i : -EFAULT;
376 if (ret & VM_FAULT_MAJOR)
382 * The VM_FAULT_WRITE bit tells us that
383 * do_wp_page has broken COW when necessary,
384 * even if maybe_mkwrite decided not to set
385 * pte_write. We can thus safely do subsequent
386 * page lookups as if they were reads. But only
387 * do so when looping for pte_write is futile:
388 * in some cases userspace may also be wanting
389 * to write to the gotten user page, which a
390 * read fault here might prevent (a readonly
391 * page might get reCOWed by userspace write).
393 if ((ret & VM_FAULT_WRITE) &&
394 !(vma->vm_flags & VM_WRITE))
395 foll_flags &= ~FOLL_WRITE;
403 return i ? i : PTR_ERR(page);
407 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
408 flush_anon_page(page, start);
410 flush_anon_page(vma, page, start);
412 flush_dcache_page(page);
419 } while (len && start < vma->vm_end);
426 int get_user_pages_uprobe(struct task_struct *tsk, struct mm_struct *mm,
427 unsigned long start, int len, int write, int force,
428 struct page **pages, struct vm_area_struct **vmas)
430 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29)
434 flags |= GUP_FLAGS_WRITE;
436 flags |= GUP_FLAGS_FORCE;
438 return __get_user_pages_uprobe(tsk, mm,
442 return get_user_pages(tsk, mm, start, len, write, force, pages, vmas);
447 int access_process_vm_atomic(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
449 struct mm_struct *mm;
450 struct vm_area_struct *vma;
452 unsigned long addr1 = addr;
453 unsigned int* inst_buf = (unsigned int*)buf;
456 mm = get_task_mm(tsk);
460 down_read(&mm->mmap_sem);
461 /* ignore errors, just check how much was successfully transferred */
463 int bytes, ret, offset;
465 struct page *page = NULL;
467 ret = get_user_pages_uprobe(tsk, mm, addr, 1,
468 write, 1, &page, &vma);
472 * Check if this is a VM_IO | VM_PFNMAP VMA, which
473 * we can access using slightly different code.
475 #ifdef CONFIG_HAVE_IOREMAP_PROT
476 vma = find_vma(mm, addr);
479 if (vma->vm_ops && vma->vm_ops->access)
480 ret = vma->vm_ops->access(vma, addr, buf,
488 offset = addr & (PAGE_SIZE-1);
489 if (bytes > PAGE_SIZE-offset)
490 bytes = PAGE_SIZE-offset;
494 copy_to_user_page(vma, page, addr,
495 maddr + offset, buf, bytes);
496 set_page_dirty_lock(page);
498 copy_from_user_page(vma, page, addr,
499 buf, maddr + offset, bytes);
502 page_cache_release(page);
508 up_read(&mm->mmap_sem);
512 if(addr1 == 0xad327238)
514 printk(">>>>> %s\n", tsk->comm);
515 printk(">>>>> %x\n", inst_buf[0]);
519 return buf - old_buf;
522 int page_present (struct mm_struct *mm, unsigned long address)
530 pgd = pgd_offset(mm, address);
531 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
534 pud = pud_offset(pgd, address);
535 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
538 pmd = pmd_offset(pud, address);
539 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
542 ptep = pte_offset_map(pmd, address);
548 if (pte_present(pte)) {
550 if (pfn_valid(pfn)) {
560 EXPORT_SYMBOL_GPL (page_present);
561 EXPORT_SYMBOL_GPL (get_user_pages_uprobe);
562 EXPORT_SYMBOL_GPL (access_process_vm_atomic);