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);
70 DECLARE_MOD_FUNC_DEP(in_gate_area_no_task, int, unsigned long addr);
71 DECLARE_MOD_FUNC_DEP(follow_page, \
72 struct page *, struct vm_area_struct * vma, \
73 unsigned long address, unsigned int foll_flags);
74 DECLARE_MOD_FUNC_DEP(__flush_anon_page, \
75 void, struct vm_area_struct *vma, struct page *page, \
76 unsigned long vmaddr);
77 DECLARE_MOD_FUNC_DEP(vm_normal_page, \
78 struct page *, struct vm_area_struct *vma, \
79 unsigned long addr, pte_t pte);
81 DECLARE_MOD_FUNC_DEP(flush_ptrace_access, \
82 void, struct vm_area_struct *vma, struct page *page, \
83 unsigned long uaddr, void *kaddr, unsigned long len, int write);
86 #if (LINUX_VERSION_CODE != KERNEL_VERSION(2, 6, 16))
87 DECLARE_MOD_FUNC_DEP(put_task_struct, \
88 void, struct task_struct *tsk);
90 DECLARE_MOD_FUNC_DEP(put_task_struct, \
91 void, struct rcu_head * rhp);
94 DECLARE_MOD_DEP_WRAPPER(access_process_vm, int, struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
95 IMP_MOD_DEP_WRAPPER (access_process_vm, tsk, addr, buf, len, write)
97 DECLARE_MOD_DEP_WRAPPER (find_extend_vma, struct vm_area_struct *, struct mm_struct * mm, unsigned long addr)
98 IMP_MOD_DEP_WRAPPER (find_extend_vma, mm, addr)
100 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 30)
101 #if LINUX_VERSION_CODE > KERNEL_VERSION(2, 6, 18)
102 DECLARE_MOD_DEP_WRAPPER (handle_mm_fault, \
103 int, struct mm_struct *mm, struct vm_area_struct *vma, unsigned long address, int write_access)
104 IMP_MOD_DEP_WRAPPER (handle_mm_fault, mm, vma, address, write_access)
107 DECLARE_MOD_DEP_WRAPPER (handle_mm_fault, \
108 int, struct mm_struct *mm, struct vm_area_struct *vma, unsigned long address, unsigned int flags)
109 IMP_MOD_DEP_WRAPPER (handle_mm_fault, mm, vma, address, flags)
112 DECLARE_MOD_DEP_WRAPPER (get_gate_vma, \
113 struct vm_area_struct *, struct task_struct *tsk)
114 IMP_MOD_DEP_WRAPPER (get_gate_vma, tsk)
116 DECLARE_MOD_DEP_WRAPPER (in_gate_area_no_task, int, unsigned long addr)
117 IMP_MOD_DEP_WRAPPER (in_gate_area_no_task, addr)
119 DECLARE_MOD_DEP_WRAPPER (follow_page, \
120 struct page *, struct vm_area_struct * vma, \
121 unsigned long address, unsigned int foll_flags)
122 IMP_MOD_DEP_WRAPPER (follow_page, vma, address, foll_flags)
124 DECLARE_MOD_DEP_WRAPPER (__flush_anon_page, \
125 void, struct vm_area_struct *vma, \
126 struct page *page, unsigned long vmaddr)
127 IMP_MOD_DEP_WRAPPER (__flush_anon_page, vma, page, vmaddr)
129 DECLARE_MOD_DEP_WRAPPER(vm_normal_page, \
130 struct page *, struct vm_area_struct *vma, \
131 unsigned long addr, pte_t pte)
132 IMP_MOD_DEP_WRAPPER (vm_normal_page, vma, addr, pte)
134 DECLARE_MOD_DEP_WRAPPER (flush_ptrace_access, \
135 void, struct vm_area_struct *vma, struct page *page, \
136 unsigned long uaddr, void *kaddr, unsigned long len, int write)
137 IMP_MOD_DEP_WRAPPER (flush_ptrace_access, vma, page, uaddr, kaddr, len, write)
139 int init_module_dependencies()
142 #if LINUX_VERSION_CODE > KERNEL_VERSION(2, 6, 29)
143 init_mm_ptr = (struct mm_struct*) kallsyms_search ("init_mm");
144 memcmp(init_mm_ptr, &init_mm, sizeof(struct mm_struct));
147 #if LINUX_VERSION_CODE > KERNEL_VERSION(2, 6, 18)
148 INIT_MOD_DEP_VAR(handle_mm_fault, handle_mm_fault);
151 INIT_MOD_DEP_VAR(flush_ptrace_access, flush_ptrace_access);
152 INIT_MOD_DEP_VAR(find_extend_vma, find_extend_vma);
153 INIT_MOD_DEP_VAR(get_gate_vma, get_gate_vma);
154 INIT_MOD_DEP_VAR(in_gate_area_no_task, in_gate_area_no_task);
155 INIT_MOD_DEP_VAR(follow_page, follow_page);
156 INIT_MOD_DEP_VAR(__flush_anon_page, __flush_anon_page);
157 INIT_MOD_DEP_VAR(vm_normal_page, vm_normal_page);
158 INIT_MOD_DEP_VAR(access_process_vm, access_process_vm);
160 #if (LINUX_VERSION_CODE != KERNEL_VERSION(2, 6, 16))
161 # if (LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 11))
162 INIT_MOD_DEP_VAR(put_task_struct, put_task_struct);
164 INIT_MOD_DEP_VAR(put_task_struct, __put_task_struct);
167 INIT_MOD_DEP_VAR(put_task_struct, __put_task_struct_cb);
173 #define GUP_FLAGS_WRITE 0x1
174 #define GUP_FLAGS_FORCE 0x2
175 #define GUP_FLAGS_IGNORE_VMA_PERMISSIONS 0x4
176 #define GUP_FLAGS_IGNORE_SIGKILL 0x8
178 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,18)
179 static inline int use_zero_page(struct vm_area_struct *vma)
182 * We don't want to optimize FOLL_ANON for make_pages_present()
183 * when it tries to page in a VM_LOCKED region. As to VM_SHARED,
184 * we want to get the page from the page tables to make sure
185 * that we serialize and update with any other user of that
188 if (vma->vm_flags & (VM_LOCKED | VM_SHARED))
191 * And if we have a fault routine, it's not an anonymous region.
193 return !vma->vm_ops || !vma->vm_ops->fault;
196 int __get_user_pages_uprobe(struct task_struct *tsk, struct mm_struct *mm,
197 unsigned long start, int len, int flags,
198 struct page **pages, struct vm_area_struct **vmas)
201 unsigned int vm_flags = 0;
202 int write = !!(flags & GUP_FLAGS_WRITE);
203 int force = !!(flags & GUP_FLAGS_FORCE);
204 int ignore = !!(flags & GUP_FLAGS_IGNORE_VMA_PERMISSIONS);
205 int ignore_sigkill = !!(flags & GUP_FLAGS_IGNORE_SIGKILL);
210 * Require read or write permissions.
211 * If 'force' is set, we only require the "MAY" flags.
213 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
214 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
218 struct vm_area_struct *vma;
219 unsigned int foll_flags;
221 //vma = find_extend_vma(mm, start);
222 vma = find_vma(mm, start);
223 if (!vma && in_gate_area(tsk, start)) {
224 unsigned long pg = start & PAGE_MASK;
225 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
231 /* user gate pages are read-only */
232 if (!ignore && write)
233 return i ? : -EFAULT;
235 pgd = pgd_offset_k(pg);
237 pgd = pgd_offset_gate(mm, pg);
238 BUG_ON(pgd_none(*pgd));
239 pud = pud_offset(pgd, pg);
240 BUG_ON(pud_none(*pud));
241 pmd = pmd_offset(pud, pg);
243 return i ? : -EFAULT;
244 pte = pte_offset_map(pmd, pg);
245 if (pte_none(*pte)) {
247 return i ? : -EFAULT;
250 struct page *page = vm_normal_page(gate_vma, start, *pte);
265 (vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
266 (!ignore && !(vm_flags & vma->vm_flags)))
267 return i ? : -EFAULT;
269 if (is_vm_hugetlb_page(vma)) {
270 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
271 i = follow_hugetlb_page(mm, vma, pages, vmas,
274 i = follow_hugetlb_page(mm, vma, pages, vmas,
275 &start, &len, i, write);
280 foll_flags = FOLL_TOUCH;
282 foll_flags |= FOLL_GET;
284 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,18)
285 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,30)
286 if (!write && use_zero_page(vma))
287 foll_flags |= FOLL_ANON;
296 * If we have a pending SIGKILL, don't keep faulting
297 * pages and potentially allocating memory, unless
298 * current is handling munlock--e.g., on exit. In
299 * that case, we are not allocating memory. Rather,
300 * we're only unlocking already resident/mapped pages.
302 if (unlikely(!ignore_sigkill &&
303 fatal_signal_pending(current)))
304 return i ? i : -ERESTARTSYS;
308 foll_flags |= FOLL_WRITE;
313 DBPRINTF ("pages = %p vma = %p\n", pages, vma);
314 while (!(page = follow_page(vma, start, foll_flags))) {
316 ret = handle_mm_fault(mm, vma, start,
317 foll_flags & FOLL_WRITE);
319 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
320 if (ret & VM_FAULT_WRITE)
321 foll_flags &= ~FOLL_WRITE;
323 switch (ret & ~VM_FAULT_WRITE) {
330 case VM_FAULT_SIGBUS:
331 return i ? i : -EFAULT;
333 return i ? i : -ENOMEM;
339 if (ret & VM_FAULT_ERROR) {
340 if (ret & VM_FAULT_OOM)
341 return i ? i : -ENOMEM;
342 else if (ret & VM_FAULT_SIGBUS)
343 return i ? i : -EFAULT;
346 if (ret & VM_FAULT_MAJOR)
352 * The VM_FAULT_WRITE bit tells us that
353 * do_wp_page has broken COW when necessary,
354 * even if maybe_mkwrite decided not to set
355 * pte_write. We can thus safely do subsequent
356 * page lookups as if they were reads. But only
357 * do so when looping for pte_write is futile:
358 * in some cases userspace may also be wanting
359 * to write to the gotten user page, which a
360 * read fault here might prevent (a readonly
361 * page might get reCOWed by userspace write).
363 if ((ret & VM_FAULT_WRITE) &&
364 !(vma->vm_flags & VM_WRITE))
365 foll_flags &= ~FOLL_WRITE;
373 return i ? i : PTR_ERR(page);
377 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
378 flush_anon_page(page, start);
380 flush_anon_page(vma, page, start);
382 flush_dcache_page(page);
389 } while (len && start < vma->vm_end);
396 int get_user_pages_uprobe(struct task_struct *tsk, struct mm_struct *mm,
397 unsigned long start, int len, int write, int force,
398 struct page **pages, struct vm_area_struct **vmas)
400 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29)
404 flags |= GUP_FLAGS_WRITE;
406 flags |= GUP_FLAGS_FORCE;
408 return __get_user_pages_uprobe(tsk, mm,
412 return get_user_pages(tsk, mm, start, len, write, force, pages, vmas);
416 int access_process_vm_atomic(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
418 struct mm_struct *mm;
419 struct vm_area_struct *vma;
422 mm = get_task_mm(tsk);
426 //down_read(&mm->mmap_sem);
427 /* ignore errors, just check how much was successfully transferred */
429 int bytes, ret, offset;
431 struct page *page = NULL;
433 ret = get_user_pages_uprobe(tsk, mm, addr, 1,
434 write, 1, &page, &vma);
437 * Check if this is a VM_IO | VM_PFNMAP VMA, which
438 * we can access using slightly different code.
440 #ifdef CONFIG_HAVE_IOREMAP_PROT
441 vma = find_vma(mm, addr);
444 if (vma->vm_ops && vma->vm_ops->access)
445 ret = vma->vm_ops->access(vma, addr, buf,
453 offset = addr & (PAGE_SIZE-1);
454 if (bytes > PAGE_SIZE-offset)
455 bytes = PAGE_SIZE-offset;
459 copy_to_user_page(vma, page, addr,
460 maddr + offset, buf, bytes);
461 set_page_dirty_lock(page);
463 copy_from_user_page(vma, page, addr,
464 buf, maddr + offset, bytes);
467 page_cache_release(page);
473 //up_read(&mm->mmap_sem);
476 return buf - old_buf;
479 int page_present (struct mm_struct *mm, unsigned long address)
487 pgd = pgd_offset(mm, address);
488 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
491 pud = pud_offset(pgd, address);
492 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
495 pmd = pmd_offset(pud, address);
496 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
499 ptep = pte_offset_map(pmd, address);
505 if (pte_present(pte)) {
507 if (pfn_valid(pfn)) {
517 EXPORT_SYMBOL_GPL (page_present);
518 EXPORT_SYMBOL_GPL (get_user_pages_uprobe);
519 EXPORT_SYMBOL_GPL (access_process_vm_atomic);