thp: avoid dumping huge zero page
[linux-2.6/libata-dev.git] / mm / process_vm_access.c
blob926b466497492f3f8463ebc623adc4fbddf9547a
1 /*
2 * linux/mm/process_vm_access.c
4 * Copyright (C) 2010-2011 Christopher Yeoh <cyeoh@au1.ibm.com>, IBM Corp.
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.
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>
20 #ifdef CONFIG_COMPAT
21 #include <linux/compat.h>
22 #endif
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
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,
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)
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;
64 *bytes_copied = 0;
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);
73 if (pages_pinned != nr_pages_to_copy) {
74 rc = -EFAULT;
75 goto end;
78 /* Do the copy for each page */
79 for (pgs_copied = 0;
80 (pgs_copied < nr_pages_to_copy) && (*lvec_current < lvec_cnt);
81 pgs_copied++) {
82 /* Make sure we have a non zero length iovec */
83 while (*lvec_current < lvec_cnt
84 && lvec[*lvec_current].iov_len == 0)
85 (*lvec_current)++;
86 if (*lvec_current == lvec_cnt)
87 break;
90 * Will copy smallest of:
91 * - bytes remaining in page
92 * - bytes remaining in destination iovec
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);
100 target_kaddr = kmap(process_pages[pgs_copied]) + start_offset;
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;
118 *bytes_copied += bytes_to_copy;
119 *lvec_offset += bytes_to_copy;
120 if (*lvec_offset == lvec[*lvec_current].iov_len) {
122 * Need to copy remaining part of page into the
123 * next iovec if there are any bytes left in page
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;
136 end:
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]);
143 } else {
144 for (j = 0; j < pages_pinned; j++)
145 put_page(process_pages[j]);
148 return rc;
151 /* Maximum number of pages kmalloc'd to hold struct page's during copy */
152 #define PVM_MAX_KMALLOC_PAGES (PAGE_SIZE * 2)
155 * process_vm_rw_single_vec - read/write pages from task specified
156 * @addr: start memory address of target process
157 * @len: size of area to copy to/from
158 * @lvec: iovec array specifying where to copy to/from locally
159 * @lvec_cnt: number of elements in iovec array
160 * @lvec_current: index in iovec array we are up to
161 * @lvec_offset: offset in bytes from current iovec iov_base we are up to
162 * @process_pages: struct pages area that can store at least
163 * nr_pages_to_copy struct page pointers
164 * @mm: mm for task
165 * @task: task to read/write from
166 * @vm_write: 0 means copy from, 1 means copy to
167 * @bytes_copied: returns number of bytes successfully copied
168 * Returns 0 on success or on failure error code
170 static 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)
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 *);
192 *bytes_copied = 0;
194 /* Work out address and page range required */
195 if (len == 0)
196 return 0;
197 nr_pages = (addr + len - 1) / PAGE_SIZE - addr / PAGE_SIZE + 1;
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);
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;
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;
221 return rc;
224 /* Maximum number of entries for process pages array
225 which lives on stack */
226 #define PVM_MAX_PP_ARRAY_COUNT 16
229 * process_vm_rw_core - core of reading/writing pages from task specified
230 * @pid: PID of process to read/write from/to
231 * @lvec: iovec array specifying where to copy to/from locally
232 * @liovcnt: size of lvec array
233 * @rvec: iovec array specifying where to copy to/from in the other process
234 * @riovcnt: size of rvec array
235 * @flags: currently unused
236 * @vm_write: 0 if reading from other process, 1 if writing to other process
237 * Returns the number of bytes read/written or error code. May
238 * return less bytes than expected if an error occurs during the copying
239 * process.
241 static 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)
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;
262 * Work out how many pages of struct pages we're going to need
263 * when eventually calling get_user_pages
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);
276 if (nr_pages == 0)
277 return 0;
279 if (nr_pages > PVM_MAX_PP_ARRAY_COUNT) {
280 /* For reliability don't try to kmalloc more than
281 2 pages worth */
282 process_pages = kmalloc(min_t(size_t, PVM_MAX_KMALLOC_PAGES,
283 sizeof(struct pages *)*nr_pages),
284 GFP_KERNEL);
286 if (!process_pages)
287 return -ENOMEM;
290 /* Get process information */
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;
301 mm = mm_access(task, PTRACE_MODE_ATTACH);
302 if (!mm || IS_ERR(mm)) {
303 rc = IS_ERR(mm) ? PTR_ERR(mm) : -ESRCH;
305 * Explicitly map EACCES to EPERM as EPERM is a more a
306 * appropriate error code for process_vw_readv/writev
308 if (rc == -EACCES)
309 rc = -EPERM;
310 goto put_task_struct;
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 /* If we have managed to copy any data at all then
321 we return the number of bytes copied. Otherwise
322 we return the error code */
323 if (bytes_copied)
324 rc = bytes_copied;
325 goto put_mm;
329 rc = bytes_copied;
330 put_mm:
331 mmput(mm);
333 put_task_struct:
334 put_task_struct(task);
336 free_proc_pages:
337 if (process_pages != pp_stack)
338 kfree(process_pages);
339 return rc;
343 * process_vm_rw - check iovecs before calling core routine
344 * @pid: PID of process to read/write from/to
345 * @lvec: iovec array specifying where to copy to/from locally
346 * @liovcnt: size of lvec array
347 * @rvec: iovec array specifying where to copy to/from in the other process
348 * @riovcnt: size of rvec array
349 * @flags: currently unused
350 * @vm_write: 0 if reading from other process, 1 if writing to other process
351 * Returns the number of bytes read/written or error code. May
352 * return less bytes than expected if an error occurs during the copying
353 * process.
355 static 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)
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;
368 if (flags != 0)
369 return -EINVAL;
371 /* Check iovecs */
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;
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;
386 rc = process_vm_rw_core(pid, iov_l, liovcnt, iov_r, riovcnt, flags,
387 vm_write);
389 free_iovecs:
390 if (iov_r != iovstack_r)
391 kfree(iov_r);
392 if (iov_l != iovstack_l)
393 kfree(iov_l);
395 return rc;
398 SYSCALL_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)
402 return process_vm_rw(pid, lvec, liovcnt, rvec, riovcnt, flags, 0);
405 SYSCALL_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)
410 return process_vm_rw(pid, lvec, liovcnt, rvec, riovcnt, flags, 1);
413 #ifdef CONFIG_COMPAT
415 asmlinkage ssize_t
416 compat_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)
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;
429 if (flags != 0)
430 return -EINVAL;
432 if (!access_ok(VERIFY_READ, lvec, liovcnt * sizeof(*lvec)))
433 goto out;
435 if (!access_ok(VERIFY_READ, rvec, riovcnt * sizeof(*rvec)))
436 goto out;
438 if (vm_write)
439 rc = compat_rw_copy_check_uvector(WRITE, lvec, liovcnt,
440 UIO_FASTIOV, iovstack_l,
441 &iov_l);
442 else
443 rc = compat_rw_copy_check_uvector(READ, lvec, liovcnt,
444 UIO_FASTIOV, iovstack_l,
445 &iov_l);
446 if (rc <= 0)
447 goto free_iovecs;
448 rc = compat_rw_copy_check_uvector(CHECK_IOVEC_ONLY, rvec, riovcnt,
449 UIO_FASTIOV, iovstack_r,
450 &iov_r);
451 if (rc <= 0)
452 goto free_iovecs;
454 rc = process_vm_rw_core(pid, iov_l, liovcnt, iov_r, riovcnt, flags,
455 vm_write);
457 free_iovecs:
458 if (iov_r != iovstack_r)
459 kfree(iov_r);
460 if (iov_l != iovstack_l)
461 kfree(iov_l);
463 out:
464 return rc;
467 asmlinkage ssize_t
468 compat_sys_process_vm_readv(compat_pid_t pid,
469 const struct compat_iovec __user *lvec,
470 unsigned long liovcnt,
471 const struct compat_iovec __user *rvec,
472 unsigned long riovcnt,
473 unsigned long flags)
475 return compat_process_vm_rw(pid, lvec, liovcnt, rvec,
476 riovcnt, flags, 0);
479 asmlinkage ssize_t
480 compat_sys_process_vm_writev(compat_pid_t pid,
481 const struct compat_iovec __user *lvec,
482 unsigned long liovcnt,
483 const struct compat_iovec __user *rvec,
484 unsigned long riovcnt,
485 unsigned long flags)
487 return compat_process_vm_rw(pid, lvec, liovcnt, rvec,
488 riovcnt, flags, 1);
491 #endif