1
2
3
4
5
6
7
8
9
10
11
12#include <linux/mm.h>
13#include <linux/uio.h>
14#include <linux/sched.h>
15#include <linux/highmem.h>
16#include <linux/ptrace.h>
17#include <linux/slab.h>
18#include <linux/syscalls.h>
19
20#ifdef CONFIG_COMPAT
21#include <linux/compat.h>
22#endif
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42static int process_vm_rw_pages(struct task_struct *task,
43 struct mm_struct *mm,
44 struct page **process_pages,
45 unsigned long pa,
46 unsigned long start_offset,
47 unsigned long len,
48 const struct iovec *lvec,
49 unsigned long lvec_cnt,
50 unsigned long *lvec_current,
51 size_t *lvec_offset,
52 int vm_write,
53 unsigned int nr_pages_to_copy,
54 ssize_t *bytes_copied)
55{
56 int pages_pinned;
57 void *target_kaddr;
58 int pgs_copied = 0;
59 int j;
60 int ret;
61 ssize_t bytes_to_copy;
62 ssize_t rc = 0;
63
64 *bytes_copied = 0;
65
66
67 down_read(&mm->mmap_sem);
68 pages_pinned = get_user_pages(task, mm, pa,
69 nr_pages_to_copy,
70 vm_write, 0, process_pages, NULL);
71 up_read(&mm->mmap_sem);
72
73 if (pages_pinned != nr_pages_to_copy) {
74 rc = -EFAULT;
75 goto end;
76 }
77
78
79 for (pgs_copied = 0;
80 (pgs_copied < nr_pages_to_copy) && (*lvec_current < lvec_cnt);
81 pgs_copied++) {
82
83 while (*lvec_current < lvec_cnt
84 && lvec[*lvec_current].iov_len == 0)
85 (*lvec_current)++;
86 if (*lvec_current == lvec_cnt)
87 break;
88
89
90
91
92
93
94 bytes_to_copy = min_t(ssize_t, PAGE_SIZE - start_offset,
95 len - *bytes_copied);
96 bytes_to_copy = min_t(ssize_t, bytes_to_copy,
97 lvec[*lvec_current].iov_len
98 - *lvec_offset);
99
100 target_kaddr = kmap(process_pages[pgs_copied]) + start_offset;
101
102 if (vm_write)
103 ret = copy_from_user(target_kaddr,
104 lvec[*lvec_current].iov_base
105 + *lvec_offset,
106 bytes_to_copy);
107 else
108 ret = copy_to_user(lvec[*lvec_current].iov_base
109 + *lvec_offset,
110 target_kaddr, bytes_to_copy);
111 kunmap(process_pages[pgs_copied]);
112 if (ret) {
113 *bytes_copied += bytes_to_copy - ret;
114 pgs_copied++;
115 rc = -EFAULT;
116 goto end;
117 }
118 *bytes_copied += bytes_to_copy;
119 *lvec_offset += bytes_to_copy;
120 if (*lvec_offset == lvec[*lvec_current].iov_len) {
121
122
123
124
125 (*lvec_current)++;
126 *lvec_offset = 0;
127 start_offset = (start_offset + bytes_to_copy)
128 % PAGE_SIZE;
129 if (start_offset)
130 pgs_copied--;
131 } else {
132 start_offset = 0;
133 }
134 }
135
136end:
137 if (vm_write) {
138 for (j = 0; j < pages_pinned; j++) {
139 if (j < pgs_copied)
140 set_page_dirty_lock(process_pages[j]);
141 put_page(process_pages[j]);
142 }
143 } else {
144 for (j = 0; j < pages_pinned; j++)
145 put_page(process_pages[j]);
146 }
147
148 return rc;
149}
150
151
152#define PVM_MAX_KMALLOC_PAGES (PAGE_SIZE * 2)
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170static int process_vm_rw_single_vec(unsigned long addr,
171 unsigned long len,
172 const struct iovec *lvec,
173 unsigned long lvec_cnt,
174 unsigned long *lvec_current,
175 size_t *lvec_offset,
176 struct page **process_pages,
177 struct mm_struct *mm,
178 struct task_struct *task,
179 int vm_write,
180 ssize_t *bytes_copied)
181{
182 unsigned long pa = addr & PAGE_MASK;
183 unsigned long start_offset = addr - pa;
184 unsigned long nr_pages;
185 ssize_t bytes_copied_loop;
186 ssize_t rc = 0;
187 unsigned long nr_pages_copied = 0;
188 unsigned long nr_pages_to_copy;
189 unsigned long max_pages_per_loop = PVM_MAX_KMALLOC_PAGES
190 / sizeof(struct pages *);
191
192 *bytes_copied = 0;
193
194
195 if (len == 0)
196 return 0;
197 nr_pages = (addr + len - 1) / PAGE_SIZE - addr / PAGE_SIZE + 1;
198
199 while ((nr_pages_copied < nr_pages) && (*lvec_current < lvec_cnt)) {
200 nr_pages_to_copy = min(nr_pages - nr_pages_copied,
201 max_pages_per_loop);
202
203 rc = process_vm_rw_pages(task, mm, process_pages, pa,
204 start_offset, len,
205 lvec, lvec_cnt,
206 lvec_current, lvec_offset,
207 vm_write, nr_pages_to_copy,
208 &bytes_copied_loop);
209 start_offset = 0;
210 *bytes_copied += bytes_copied_loop;
211
212 if (rc < 0) {
213 return rc;
214 } else {
215 len -= bytes_copied_loop;
216 nr_pages_copied += nr_pages_to_copy;
217 pa += nr_pages_to_copy * PAGE_SIZE;
218 }
219 }
220
221 return rc;
222}
223
224
225
226#define PVM_MAX_PP_ARRAY_COUNT 16
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241static ssize_t process_vm_rw_core(pid_t pid, const struct iovec *lvec,
242 unsigned long liovcnt,
243 const struct iovec *rvec,
244 unsigned long riovcnt,
245 unsigned long flags, int vm_write)
246{
247 struct task_struct *task;
248 struct page *pp_stack[PVM_MAX_PP_ARRAY_COUNT];
249 struct page **process_pages = pp_stack;
250 struct mm_struct *mm;
251 unsigned long i;
252 ssize_t rc = 0;
253 ssize_t bytes_copied_loop;
254 ssize_t bytes_copied = 0;
255 unsigned long nr_pages = 0;
256 unsigned long nr_pages_iov;
257 unsigned long iov_l_curr_idx = 0;
258 size_t iov_l_curr_offset = 0;
259 ssize_t iov_len;
260
261
262
263
264
265 for (i = 0; i < riovcnt; i++) {
266 iov_len = rvec[i].iov_len;
267 if (iov_len > 0) {
268 nr_pages_iov = ((unsigned long)rvec[i].iov_base
269 + iov_len)
270 / PAGE_SIZE - (unsigned long)rvec[i].iov_base
271 / PAGE_SIZE + 1;
272 nr_pages = max(nr_pages, nr_pages_iov);
273 }
274 }
275
276 if (nr_pages == 0)
277 return 0;
278
279 if (nr_pages > PVM_MAX_PP_ARRAY_COUNT) {
280
281
282 process_pages = kmalloc(min_t(size_t, PVM_MAX_KMALLOC_PAGES,
283 sizeof(struct pages *)*nr_pages),
284 GFP_KERNEL);
285
286 if (!process_pages)
287 return -ENOMEM;
288 }
289
290
291 rcu_read_lock();
292 task = find_task_by_vpid(pid);
293 if (task)
294 get_task_struct(task);
295 rcu_read_unlock();
296 if (!task) {
297 rc = -ESRCH;
298 goto free_proc_pages;
299 }
300
301 mm = mm_access(task, PTRACE_MODE_ATTACH);
302 if (!mm || IS_ERR(mm)) {
303 rc = IS_ERR(mm) ? PTR_ERR(mm) : -ESRCH;
304
305
306
307
308 if (rc == -EACCES)
309 rc = -EPERM;
310 goto put_task_struct;
311 }
312
313 for (i = 0; i < riovcnt && iov_l_curr_idx < liovcnt; i++) {
314 rc = process_vm_rw_single_vec(
315 (unsigned long)rvec[i].iov_base, rvec[i].iov_len,
316 lvec, liovcnt, &iov_l_curr_idx, &iov_l_curr_offset,
317 process_pages, mm, task, vm_write, &bytes_copied_loop);
318 bytes_copied += bytes_copied_loop;
319 if (rc != 0) {
320
321
322
323 if (bytes_copied)
324 rc = bytes_copied;
325 goto put_mm;
326 }
327 }
328
329 rc = bytes_copied;
330put_mm:
331 mmput(mm);
332
333put_task_struct:
334 put_task_struct(task);
335
336free_proc_pages:
337 if (process_pages != pp_stack)
338 kfree(process_pages);
339 return rc;
340}
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355static ssize_t process_vm_rw(pid_t pid,
356 const struct iovec __user *lvec,
357 unsigned long liovcnt,
358 const struct iovec __user *rvec,
359 unsigned long riovcnt,
360 unsigned long flags, int vm_write)
361{
362 struct iovec iovstack_l[UIO_FASTIOV];
363 struct iovec iovstack_r[UIO_FASTIOV];
364 struct iovec *iov_l = iovstack_l;
365 struct iovec *iov_r = iovstack_r;
366 ssize_t rc;
367
368 if (flags != 0)
369 return -EINVAL;
370
371
372 if (vm_write)
373 rc = rw_copy_check_uvector(WRITE, lvec, liovcnt, UIO_FASTIOV,
374 iovstack_l, &iov_l);
375 else
376 rc = rw_copy_check_uvector(READ, lvec, liovcnt, UIO_FASTIOV,
377 iovstack_l, &iov_l);
378 if (rc <= 0)
379 goto free_iovecs;
380
381 rc = rw_copy_check_uvector(CHECK_IOVEC_ONLY, rvec, riovcnt, UIO_FASTIOV,
382 iovstack_r, &iov_r);
383 if (rc <= 0)
384 goto free_iovecs;
385
386 rc = process_vm_rw_core(pid, iov_l, liovcnt, iov_r, riovcnt, flags,
387 vm_write);
388
389free_iovecs:
390 if (iov_r != iovstack_r)
391 kfree(iov_r);
392 if (iov_l != iovstack_l)
393 kfree(iov_l);
394
395 return rc;
396}
397
398SYSCALL_DEFINE6(process_vm_readv, pid_t, pid, const struct iovec __user *, lvec,
399 unsigned long, liovcnt, const struct iovec __user *, rvec,
400 unsigned long, riovcnt, unsigned long, flags)
401{
402 return process_vm_rw(pid, lvec, liovcnt, rvec, riovcnt, flags, 0);
403}
404
405SYSCALL_DEFINE6(process_vm_writev, pid_t, pid,
406 const struct iovec __user *, lvec,
407 unsigned long, liovcnt, const struct iovec __user *, rvec,
408 unsigned long, riovcnt, unsigned long, flags)
409{
410 return process_vm_rw(pid, lvec, liovcnt, rvec, riovcnt, flags, 1);
411}
412
413#ifdef CONFIG_COMPAT
414
415asmlinkage ssize_t
416compat_process_vm_rw(compat_pid_t pid,
417 const struct compat_iovec __user *lvec,
418 unsigned long liovcnt,
419 const struct compat_iovec __user *rvec,
420 unsigned long riovcnt,
421 unsigned long flags, int vm_write)
422{
423 struct iovec iovstack_l[UIO_FASTIOV];
424 struct iovec iovstack_r[UIO_FASTIOV];
425 struct iovec *iov_l = iovstack_l;
426 struct iovec *iov_r = iovstack_r;
427 ssize_t rc = -EFAULT;
428
429 if (flags != 0)
430 return -EINVAL;
431
432 if (vm_write)
433 rc = compat_rw_copy_check_uvector(WRITE, lvec, liovcnt,
434 UIO_FASTIOV, iovstack_l,
435 &iov_l);
436 else
437 rc = compat_rw_copy_check_uvector(READ, lvec, liovcnt,
438 UIO_FASTIOV, iovstack_l,
439 &iov_l);
440 if (rc <= 0)
441 goto free_iovecs;
442 rc = compat_rw_copy_check_uvector(CHECK_IOVEC_ONLY, rvec, riovcnt,
443 UIO_FASTIOV, iovstack_r,
444 &iov_r);
445 if (rc <= 0)
446 goto free_iovecs;
447
448 rc = process_vm_rw_core(pid, iov_l, liovcnt, iov_r, riovcnt, flags,
449 vm_write);
450
451free_iovecs:
452 if (iov_r != iovstack_r)
453 kfree(iov_r);
454 if (iov_l != iovstack_l)
455 kfree(iov_l);
456 return rc;
457}
458
459asmlinkage ssize_t
460compat_sys_process_vm_readv(compat_pid_t pid,
461 const struct compat_iovec __user *lvec,
462 unsigned long liovcnt,
463 const struct compat_iovec __user *rvec,
464 unsigned long riovcnt,
465 unsigned long flags)
466{
467 return compat_process_vm_rw(pid, lvec, liovcnt, rvec,
468 riovcnt, flags, 0);
469}
470
471asmlinkage ssize_t
472compat_sys_process_vm_writev(compat_pid_t pid,
473 const struct compat_iovec __user *lvec,
474 unsigned long liovcnt,
475 const struct compat_iovec __user *rvec,
476 unsigned long riovcnt,
477 unsigned long flags)
478{
479 return compat_process_vm_rw(pid, lvec, liovcnt, rvec,
480 riovcnt, flags, 1);
481}
482
483#endif
484