Commit | Line | Data |
---|---|---|
fcf63409 CY |
1 | /* |
2 | * linux/mm/process_vm_access.c | |
3 | * | |
4 | * Copyright (C) 2010-2011 Christopher Yeoh <cyeoh@au1.ibm.com>, IBM Corp. | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or | |
7 | * modify it under the terms of the GNU General Public License | |
8 | * as published by the Free Software Foundation; either version | |
9 | * 2 of the License, or (at your option) any later version. | |
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 | * process_vm_rw_pages - read/write pages from task specified | |
26 | * @task: task to read/write from | |
27 | * @mm: mm for task | |
28 | * @process_pages: struct pages area that can store at least | |
29 | * nr_pages_to_copy struct page pointers | |
30 | * @pa: address of page in task to start copying from/to | |
31 | * @start_offset: offset in page to start copying from/to | |
32 | * @len: number of bytes to copy | |
33 | * @lvec: iovec array specifying where to copy to/from | |
34 | * @lvec_cnt: number of elements in iovec array | |
35 | * @lvec_current: index in iovec array we are up to | |
36 | * @lvec_offset: offset in bytes from current iovec iov_base we are up to | |
37 | * @vm_write: 0 means copy from, 1 means copy to | |
38 | * @nr_pages_to_copy: number of pages to copy | |
39 | * @bytes_copied: returns number of bytes successfully copied | |
40 | * Returns 0 on success, error code otherwise | |
41 | */ | |
42 | static 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, | |
c61c7038 | 48 | const struct iovec **iovp, |
1291afc1 | 49 | unsigned long *left, |
fcf63409 CY |
50 | size_t *lvec_offset, |
51 | int vm_write, | |
52 | unsigned int nr_pages_to_copy, | |
53 | ssize_t *bytes_copied) | |
54 | { | |
55 | int pages_pinned; | |
56 | void *target_kaddr; | |
57 | int pgs_copied = 0; | |
58 | int j; | |
59 | int ret; | |
60 | ssize_t bytes_to_copy; | |
61 | ssize_t rc = 0; | |
c61c7038 | 62 | const struct iovec *iov = *iovp; |
fcf63409 CY |
63 | |
64 | *bytes_copied = 0; | |
65 | ||
66 | /* Get the pages we're interested in */ | |
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 | /* Do the copy for each page */ | |
79 | for (pgs_copied = 0; | |
1291afc1 | 80 | (pgs_copied < nr_pages_to_copy) && *left; |
fcf63409 CY |
81 | pgs_copied++) { |
82 | /* Make sure we have a non zero length iovec */ | |
1291afc1 | 83 | while (*left && iov->iov_len == 0) { |
480402e1 | 84 | iov++; |
1291afc1 | 85 | (*left)--; |
480402e1 | 86 | } |
1291afc1 | 87 | if (!*left) |
fcf63409 CY |
88 | break; |
89 | ||
90 | /* | |
91 | * Will copy smallest of: | |
92 | * - bytes remaining in page | |
93 | * - bytes remaining in destination iovec | |
94 | */ | |
95 | bytes_to_copy = min_t(ssize_t, PAGE_SIZE - start_offset, | |
96 | len - *bytes_copied); | |
97 | bytes_to_copy = min_t(ssize_t, bytes_to_copy, | |
480402e1 | 98 | iov->iov_len |
fcf63409 CY |
99 | - *lvec_offset); |
100 | ||
101 | target_kaddr = kmap(process_pages[pgs_copied]) + start_offset; | |
102 | ||
103 | if (vm_write) | |
104 | ret = copy_from_user(target_kaddr, | |
480402e1 | 105 | iov->iov_base |
fcf63409 CY |
106 | + *lvec_offset, |
107 | bytes_to_copy); | |
108 | else | |
480402e1 | 109 | ret = copy_to_user(iov->iov_base |
fcf63409 CY |
110 | + *lvec_offset, |
111 | target_kaddr, bytes_to_copy); | |
112 | kunmap(process_pages[pgs_copied]); | |
113 | if (ret) { | |
114 | *bytes_copied += bytes_to_copy - ret; | |
115 | pgs_copied++; | |
116 | rc = -EFAULT; | |
117 | goto end; | |
118 | } | |
119 | *bytes_copied += bytes_to_copy; | |
120 | *lvec_offset += bytes_to_copy; | |
480402e1 | 121 | if (*lvec_offset == iov->iov_len) { |
fcf63409 CY |
122 | /* |
123 | * Need to copy remaining part of page into the | |
124 | * next iovec if there are any bytes left in page | |
125 | */ | |
1291afc1 | 126 | (*left)--; |
480402e1 | 127 | iov++; |
fcf63409 CY |
128 | *lvec_offset = 0; |
129 | start_offset = (start_offset + bytes_to_copy) | |
130 | % PAGE_SIZE; | |
131 | if (start_offset) | |
132 | pgs_copied--; | |
133 | } else { | |
134 | start_offset = 0; | |
135 | } | |
136 | } | |
137 | ||
138 | end: | |
139 | if (vm_write) { | |
140 | for (j = 0; j < pages_pinned; j++) { | |
141 | if (j < pgs_copied) | |
142 | set_page_dirty_lock(process_pages[j]); | |
143 | put_page(process_pages[j]); | |
144 | } | |
145 | } else { | |
146 | for (j = 0; j < pages_pinned; j++) | |
147 | put_page(process_pages[j]); | |
148 | } | |
149 | ||
c61c7038 | 150 | *iovp = iov; |
fcf63409 CY |
151 | return rc; |
152 | } | |
153 | ||
154 | /* Maximum number of pages kmalloc'd to hold struct page's during copy */ | |
155 | #define PVM_MAX_KMALLOC_PAGES (PAGE_SIZE * 2) | |
156 | ||
157 | /** | |
158 | * process_vm_rw_single_vec - read/write pages from task specified | |
159 | * @addr: start memory address of target process | |
160 | * @len: size of area to copy to/from | |
161 | * @lvec: iovec array specifying where to copy to/from locally | |
162 | * @lvec_cnt: number of elements in iovec array | |
163 | * @lvec_current: index in iovec array we are up to | |
164 | * @lvec_offset: offset in bytes from current iovec iov_base we are up to | |
165 | * @process_pages: struct pages area that can store at least | |
166 | * nr_pages_to_copy struct page pointers | |
167 | * @mm: mm for task | |
168 | * @task: task to read/write from | |
169 | * @vm_write: 0 means copy from, 1 means copy to | |
170 | * @bytes_copied: returns number of bytes successfully copied | |
171 | * Returns 0 on success or on failure error code | |
172 | */ | |
173 | static int process_vm_rw_single_vec(unsigned long addr, | |
174 | unsigned long len, | |
12e3004e | 175 | const struct iovec **iovp, |
1291afc1 | 176 | unsigned long *left, |
fcf63409 CY |
177 | size_t *lvec_offset, |
178 | struct page **process_pages, | |
179 | struct mm_struct *mm, | |
180 | struct task_struct *task, | |
181 | int vm_write, | |
182 | ssize_t *bytes_copied) | |
183 | { | |
184 | unsigned long pa = addr & PAGE_MASK; | |
185 | unsigned long start_offset = addr - pa; | |
186 | unsigned long nr_pages; | |
187 | ssize_t bytes_copied_loop; | |
188 | ssize_t rc = 0; | |
189 | unsigned long nr_pages_copied = 0; | |
190 | unsigned long nr_pages_to_copy; | |
191 | unsigned long max_pages_per_loop = PVM_MAX_KMALLOC_PAGES | |
192 | / sizeof(struct pages *); | |
193 | ||
194 | *bytes_copied = 0; | |
195 | ||
196 | /* Work out address and page range required */ | |
197 | if (len == 0) | |
198 | return 0; | |
199 | nr_pages = (addr + len - 1) / PAGE_SIZE - addr / PAGE_SIZE + 1; | |
200 | ||
1291afc1 | 201 | while ((nr_pages_copied < nr_pages) && *left) { |
fcf63409 CY |
202 | nr_pages_to_copy = min(nr_pages - nr_pages_copied, |
203 | max_pages_per_loop); | |
204 | ||
205 | rc = process_vm_rw_pages(task, mm, process_pages, pa, | |
206 | start_offset, len, | |
1291afc1 AV |
207 | iovp, left, |
208 | lvec_offset, | |
fcf63409 CY |
209 | vm_write, nr_pages_to_copy, |
210 | &bytes_copied_loop); | |
211 | start_offset = 0; | |
212 | *bytes_copied += bytes_copied_loop; | |
213 | ||
214 | if (rc < 0) { | |
215 | return rc; | |
216 | } else { | |
217 | len -= bytes_copied_loop; | |
218 | nr_pages_copied += nr_pages_to_copy; | |
219 | pa += nr_pages_to_copy * PAGE_SIZE; | |
220 | } | |
221 | } | |
222 | ||
223 | return rc; | |
224 | } | |
225 | ||
226 | /* Maximum number of entries for process pages array | |
227 | which lives on stack */ | |
228 | #define PVM_MAX_PP_ARRAY_COUNT 16 | |
229 | ||
230 | /** | |
231 | * process_vm_rw_core - core of reading/writing pages from task specified | |
232 | * @pid: PID of process to read/write from/to | |
233 | * @lvec: iovec array specifying where to copy to/from locally | |
234 | * @liovcnt: size of lvec array | |
235 | * @rvec: iovec array specifying where to copy to/from in the other process | |
236 | * @riovcnt: size of rvec array | |
237 | * @flags: currently unused | |
238 | * @vm_write: 0 if reading from other process, 1 if writing to other process | |
239 | * Returns the number of bytes read/written or error code. May | |
240 | * return less bytes than expected if an error occurs during the copying | |
241 | * process. | |
242 | */ | |
243 | static ssize_t process_vm_rw_core(pid_t pid, const struct iovec *lvec, | |
244 | unsigned long liovcnt, | |
245 | const struct iovec *rvec, | |
246 | unsigned long riovcnt, | |
247 | unsigned long flags, int vm_write) | |
248 | { | |
249 | struct task_struct *task; | |
250 | struct page *pp_stack[PVM_MAX_PP_ARRAY_COUNT]; | |
251 | struct page **process_pages = pp_stack; | |
252 | struct mm_struct *mm; | |
253 | unsigned long i; | |
254 | ssize_t rc = 0; | |
255 | ssize_t bytes_copied_loop; | |
256 | ssize_t bytes_copied = 0; | |
257 | unsigned long nr_pages = 0; | |
258 | unsigned long nr_pages_iov; | |
1291afc1 | 259 | unsigned long left = liovcnt; |
fcf63409 CY |
260 | size_t iov_l_curr_offset = 0; |
261 | ssize_t iov_len; | |
262 | ||
263 | /* | |
264 | * Work out how many pages of struct pages we're going to need | |
265 | * when eventually calling get_user_pages | |
266 | */ | |
267 | for (i = 0; i < riovcnt; i++) { | |
268 | iov_len = rvec[i].iov_len; | |
269 | if (iov_len > 0) { | |
270 | nr_pages_iov = ((unsigned long)rvec[i].iov_base | |
271 | + iov_len) | |
272 | / PAGE_SIZE - (unsigned long)rvec[i].iov_base | |
273 | / PAGE_SIZE + 1; | |
274 | nr_pages = max(nr_pages, nr_pages_iov); | |
275 | } | |
276 | } | |
277 | ||
278 | if (nr_pages == 0) | |
279 | return 0; | |
280 | ||
281 | if (nr_pages > PVM_MAX_PP_ARRAY_COUNT) { | |
282 | /* For reliability don't try to kmalloc more than | |
283 | 2 pages worth */ | |
284 | process_pages = kmalloc(min_t(size_t, PVM_MAX_KMALLOC_PAGES, | |
285 | sizeof(struct pages *)*nr_pages), | |
286 | GFP_KERNEL); | |
287 | ||
288 | if (!process_pages) | |
289 | return -ENOMEM; | |
290 | } | |
291 | ||
292 | /* Get process information */ | |
293 | rcu_read_lock(); | |
294 | task = find_task_by_vpid(pid); | |
295 | if (task) | |
296 | get_task_struct(task); | |
297 | rcu_read_unlock(); | |
298 | if (!task) { | |
299 | rc = -ESRCH; | |
300 | goto free_proc_pages; | |
301 | } | |
302 | ||
8cdb878d CY |
303 | mm = mm_access(task, PTRACE_MODE_ATTACH); |
304 | if (!mm || IS_ERR(mm)) { | |
305 | rc = IS_ERR(mm) ? PTR_ERR(mm) : -ESRCH; | |
306 | /* | |
307 | * Explicitly map EACCES to EPERM as EPERM is a more a | |
308 | * appropriate error code for process_vw_readv/writev | |
309 | */ | |
310 | if (rc == -EACCES) | |
311 | rc = -EPERM; | |
fcf63409 CY |
312 | goto put_task_struct; |
313 | } | |
314 | ||
1291afc1 | 315 | for (i = 0; i < riovcnt && left; i++) { |
fcf63409 CY |
316 | rc = process_vm_rw_single_vec( |
317 | (unsigned long)rvec[i].iov_base, rvec[i].iov_len, | |
1291afc1 | 318 | &lvec, &left, &iov_l_curr_offset, |
fcf63409 CY |
319 | process_pages, mm, task, vm_write, &bytes_copied_loop); |
320 | bytes_copied += bytes_copied_loop; | |
321 | if (rc != 0) { | |
322 | /* If we have managed to copy any data at all then | |
323 | we return the number of bytes copied. Otherwise | |
324 | we return the error code */ | |
325 | if (bytes_copied) | |
326 | rc = bytes_copied; | |
327 | goto put_mm; | |
328 | } | |
329 | } | |
330 | ||
331 | rc = bytes_copied; | |
332 | put_mm: | |
333 | mmput(mm); | |
334 | ||
335 | put_task_struct: | |
336 | put_task_struct(task); | |
337 | ||
338 | free_proc_pages: | |
339 | if (process_pages != pp_stack) | |
340 | kfree(process_pages); | |
341 | return rc; | |
342 | } | |
343 | ||
344 | /** | |
345 | * process_vm_rw - check iovecs before calling core routine | |
346 | * @pid: PID of process to read/write from/to | |
347 | * @lvec: iovec array specifying where to copy to/from locally | |
348 | * @liovcnt: size of lvec array | |
349 | * @rvec: iovec array specifying where to copy to/from in the other process | |
350 | * @riovcnt: size of rvec array | |
351 | * @flags: currently unused | |
352 | * @vm_write: 0 if reading from other process, 1 if writing to other process | |
353 | * Returns the number of bytes read/written or error code. May | |
354 | * return less bytes than expected if an error occurs during the copying | |
355 | * process. | |
356 | */ | |
357 | static ssize_t process_vm_rw(pid_t pid, | |
358 | const struct iovec __user *lvec, | |
359 | unsigned long liovcnt, | |
360 | const struct iovec __user *rvec, | |
361 | unsigned long riovcnt, | |
362 | unsigned long flags, int vm_write) | |
363 | { | |
364 | struct iovec iovstack_l[UIO_FASTIOV]; | |
365 | struct iovec iovstack_r[UIO_FASTIOV]; | |
366 | struct iovec *iov_l = iovstack_l; | |
367 | struct iovec *iov_r = iovstack_r; | |
368 | ssize_t rc; | |
369 | ||
370 | if (flags != 0) | |
371 | return -EINVAL; | |
372 | ||
373 | /* Check iovecs */ | |
374 | if (vm_write) | |
375 | rc = rw_copy_check_uvector(WRITE, lvec, liovcnt, UIO_FASTIOV, | |
ac34ebb3 | 376 | iovstack_l, &iov_l); |
fcf63409 CY |
377 | else |
378 | rc = rw_copy_check_uvector(READ, lvec, liovcnt, UIO_FASTIOV, | |
ac34ebb3 | 379 | iovstack_l, &iov_l); |
fcf63409 CY |
380 | if (rc <= 0) |
381 | goto free_iovecs; | |
382 | ||
ac34ebb3 CY |
383 | rc = rw_copy_check_uvector(CHECK_IOVEC_ONLY, rvec, riovcnt, UIO_FASTIOV, |
384 | iovstack_r, &iov_r); | |
fcf63409 CY |
385 | if (rc <= 0) |
386 | goto free_iovecs; | |
387 | ||
388 | rc = process_vm_rw_core(pid, iov_l, liovcnt, iov_r, riovcnt, flags, | |
389 | vm_write); | |
390 | ||
391 | free_iovecs: | |
392 | if (iov_r != iovstack_r) | |
393 | kfree(iov_r); | |
394 | if (iov_l != iovstack_l) | |
395 | kfree(iov_l); | |
396 | ||
397 | return rc; | |
398 | } | |
399 | ||
400 | SYSCALL_DEFINE6(process_vm_readv, pid_t, pid, const struct iovec __user *, lvec, | |
401 | unsigned long, liovcnt, const struct iovec __user *, rvec, | |
402 | unsigned long, riovcnt, unsigned long, flags) | |
403 | { | |
404 | return process_vm_rw(pid, lvec, liovcnt, rvec, riovcnt, flags, 0); | |
405 | } | |
406 | ||
407 | SYSCALL_DEFINE6(process_vm_writev, pid_t, pid, | |
408 | const struct iovec __user *, lvec, | |
409 | unsigned long, liovcnt, const struct iovec __user *, rvec, | |
410 | unsigned long, riovcnt, unsigned long, flags) | |
411 | { | |
412 | return process_vm_rw(pid, lvec, liovcnt, rvec, riovcnt, flags, 1); | |
413 | } | |
414 | ||
415 | #ifdef CONFIG_COMPAT | |
416 | ||
417 | asmlinkage ssize_t | |
418 | compat_process_vm_rw(compat_pid_t pid, | |
419 | const struct compat_iovec __user *lvec, | |
420 | unsigned long liovcnt, | |
421 | const struct compat_iovec __user *rvec, | |
422 | unsigned long riovcnt, | |
423 | unsigned long flags, int vm_write) | |
424 | { | |
425 | struct iovec iovstack_l[UIO_FASTIOV]; | |
426 | struct iovec iovstack_r[UIO_FASTIOV]; | |
427 | struct iovec *iov_l = iovstack_l; | |
428 | struct iovec *iov_r = iovstack_r; | |
429 | ssize_t rc = -EFAULT; | |
430 | ||
431 | if (flags != 0) | |
432 | return -EINVAL; | |
433 | ||
fcf63409 CY |
434 | if (vm_write) |
435 | rc = compat_rw_copy_check_uvector(WRITE, lvec, liovcnt, | |
436 | UIO_FASTIOV, iovstack_l, | |
ac34ebb3 | 437 | &iov_l); |
fcf63409 CY |
438 | else |
439 | rc = compat_rw_copy_check_uvector(READ, lvec, liovcnt, | |
440 | UIO_FASTIOV, iovstack_l, | |
ac34ebb3 | 441 | &iov_l); |
fcf63409 CY |
442 | if (rc <= 0) |
443 | goto free_iovecs; | |
ac34ebb3 | 444 | rc = compat_rw_copy_check_uvector(CHECK_IOVEC_ONLY, rvec, riovcnt, |
fcf63409 | 445 | UIO_FASTIOV, iovstack_r, |
ac34ebb3 | 446 | &iov_r); |
fcf63409 CY |
447 | if (rc <= 0) |
448 | goto free_iovecs; | |
449 | ||
450 | rc = process_vm_rw_core(pid, iov_l, liovcnt, iov_r, riovcnt, flags, | |
451 | vm_write); | |
452 | ||
453 | free_iovecs: | |
454 | if (iov_r != iovstack_r) | |
455 | kfree(iov_r); | |
456 | if (iov_l != iovstack_l) | |
457 | kfree(iov_l); | |
fcf63409 CY |
458 | return rc; |
459 | } | |
460 | ||
461 | asmlinkage ssize_t | |
462 | compat_sys_process_vm_readv(compat_pid_t pid, | |
463 | const struct compat_iovec __user *lvec, | |
464 | unsigned long liovcnt, | |
465 | const struct compat_iovec __user *rvec, | |
466 | unsigned long riovcnt, | |
467 | unsigned long flags) | |
468 | { | |
469 | return compat_process_vm_rw(pid, lvec, liovcnt, rvec, | |
470 | riovcnt, flags, 0); | |
471 | } | |
472 | ||
473 | asmlinkage ssize_t | |
474 | compat_sys_process_vm_writev(compat_pid_t pid, | |
475 | const struct compat_iovec __user *lvec, | |
476 | unsigned long liovcnt, | |
477 | const struct compat_iovec __user *rvec, | |
478 | unsigned long riovcnt, | |
479 | unsigned long flags) | |
480 | { | |
481 | return compat_process_vm_rw(pid, lvec, liovcnt, rvec, | |
482 | riovcnt, flags, 1); | |
483 | } | |
484 | ||
485 | #endif |