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/sched/mm.h>
16#include <linux/highmem.h>
17#include <linux/ptrace.h>
18#include <linux/slab.h>
19#include <linux/syscalls.h>
20
21#ifdef CONFIG_COMPAT
22#include <linux/compat.h>
23#endif
24
25/**
26 * process_vm_rw_pages - read/write pages from task specified
27 * @pages: array of pointers to pages we want to copy
28 * @offset: offset in page to start copying from/to
29 * @len: number of bytes to copy
30 * @iter: where to copy to/from locally
31 * @vm_write: 0 means copy from, 1 means copy to
32 * Returns 0 on success, error code otherwise
33 */
34static int process_vm_rw_pages(struct page **pages,
35 unsigned offset,
36 size_t len,
37 struct iov_iter *iter,
38 int vm_write)
39{
40 /* Do the copy for each page */
41 while (len && iov_iter_count(iter)) {
42 struct page *page = *pages++;
43 size_t copy = PAGE_SIZE - offset;
44 size_t copied;
45
46 if (copy > len)
47 copy = len;
48
49 if (vm_write) {
50 copied = copy_page_from_iter(page, offset, copy, iter);
51 set_page_dirty_lock(page);
52 } else {
53 copied = copy_page_to_iter(page, offset, copy, iter);
54 }
55 len -= copied;
56 if (copied < copy && iov_iter_count(iter))
57 return -EFAULT;
58 offset = 0;
59 }
60 return 0;
61}
62
63/* Maximum number of pages kmalloc'd to hold struct page's during copy */
64#define PVM_MAX_KMALLOC_PAGES (PAGE_SIZE * 2)
65
66/**
67 * process_vm_rw_single_vec - read/write pages from task specified
68 * @addr: start memory address of target process
69 * @len: size of area to copy to/from
70 * @iter: where to copy to/from locally
71 * @process_pages: struct pages area that can store at least
72 * nr_pages_to_copy struct page pointers
73 * @mm: mm for task
74 * @task: task to read/write from
75 * @vm_write: 0 means copy from, 1 means copy to
76 * Returns 0 on success or on failure error code
77 */
78static int process_vm_rw_single_vec(unsigned long addr,
79 unsigned long len,
80 struct iov_iter *iter,
81 struct page **process_pages,
82 struct mm_struct *mm,
83 struct task_struct *task,
84 int vm_write)
85{
86 unsigned long pa = addr & PAGE_MASK;
87 unsigned long start_offset = addr - pa;
88 unsigned long nr_pages;
89 ssize_t rc = 0;
90 unsigned long max_pages_per_loop = PVM_MAX_KMALLOC_PAGES
91 / sizeof(struct pages *);
92 unsigned int flags = 0;
93
94 /* Work out address and page range required */
95 if (len == 0)
96 return 0;
97 nr_pages = (addr + len - 1) / PAGE_SIZE - addr / PAGE_SIZE + 1;
98
99 if (vm_write)
100 flags |= FOLL_WRITE;
101
102 while (!rc && nr_pages && iov_iter_count(iter)) {
103 int pages = min(nr_pages, max_pages_per_loop);
104 int locked = 1;
105 size_t bytes;
106
107 /*
108 * Get the pages we're interested in. We must
109 * access remotely because task/mm might not
110 * current/current->mm
111 */
112 down_read(&mm->mmap_sem);
113 pages = get_user_pages_remote(task, mm, pa, pages, flags,
114 process_pages, NULL, &locked);
115 if (locked)
116 up_read(&mm->mmap_sem);
117 if (pages <= 0)
118 return -EFAULT;
119
120 bytes = pages * PAGE_SIZE - start_offset;
121 if (bytes > len)
122 bytes = len;
123
124 rc = process_vm_rw_pages(process_pages,
125 start_offset, bytes, iter,
126 vm_write);
127 len -= bytes;
128 start_offset = 0;
129 nr_pages -= pages;
130 pa += pages * PAGE_SIZE;
131 while (pages)
132 put_page(process_pages[--pages]);
133 }
134
135 return rc;
136}
137
138/* Maximum number of entries for process pages array
139 which lives on stack */
140#define PVM_MAX_PP_ARRAY_COUNT 16
141
142/**
143 * process_vm_rw_core - core of reading/writing pages from task specified
144 * @pid: PID of process to read/write from/to
145 * @iter: where to copy to/from locally
146 * @rvec: iovec array specifying where to copy to/from in the other process
147 * @riovcnt: size of rvec array
148 * @flags: currently unused
149 * @vm_write: 0 if reading from other process, 1 if writing to other process
150 *
151 * Returns the number of bytes read/written or error code. May
152 * return less bytes than expected if an error occurs during the copying
153 * process.
154 */
155static ssize_t process_vm_rw_core(pid_t pid, struct iov_iter *iter,
156 const struct iovec *rvec,
157 unsigned long riovcnt,
158 unsigned long flags, int vm_write)
159{
160 struct task_struct *task;
161 struct page *pp_stack[PVM_MAX_PP_ARRAY_COUNT];
162 struct page **process_pages = pp_stack;
163 struct mm_struct *mm;
164 unsigned long i;
165 ssize_t rc = 0;
166 unsigned long nr_pages = 0;
167 unsigned long nr_pages_iov;
168 ssize_t iov_len;
169 size_t total_len = iov_iter_count(iter);
170
171 /*
172 * Work out how many pages of struct pages we're going to need
173 * when eventually calling get_user_pages
174 */
175 for (i = 0; i < riovcnt; i++) {
176 iov_len = rvec[i].iov_len;
177 if (iov_len > 0) {
178 nr_pages_iov = ((unsigned long)rvec[i].iov_base
179 + iov_len)
180 / PAGE_SIZE - (unsigned long)rvec[i].iov_base
181 / PAGE_SIZE + 1;
182 nr_pages = max(nr_pages, nr_pages_iov);
183 }
184 }
185
186 if (nr_pages == 0)
187 return 0;
188
189 if (nr_pages > PVM_MAX_PP_ARRAY_COUNT) {
190 /* For reliability don't try to kmalloc more than
191 2 pages worth */
192 process_pages = kmalloc(min_t(size_t, PVM_MAX_KMALLOC_PAGES,
193 sizeof(struct pages *)*nr_pages),
194 GFP_KERNEL);
195
196 if (!process_pages)
197 return -ENOMEM;
198 }
199
200 /* Get process information */
201 task = find_get_task_by_vpid(pid);
202 if (!task) {
203 rc = -ESRCH;
204 goto free_proc_pages;
205 }
206
207 mm = mm_access(task, PTRACE_MODE_ATTACH_REALCREDS);
208 if (!mm || IS_ERR(mm)) {
209 rc = IS_ERR(mm) ? PTR_ERR(mm) : -ESRCH;
210 /*
211 * Explicitly map EACCES to EPERM as EPERM is a more a
212 * appropriate error code for process_vw_readv/writev
213 */
214 if (rc == -EACCES)
215 rc = -EPERM;
216 goto put_task_struct;
217 }
218
219 for (i = 0; i < riovcnt && iov_iter_count(iter) && !rc; i++)
220 rc = process_vm_rw_single_vec(
221 (unsigned long)rvec[i].iov_base, rvec[i].iov_len,
222 iter, process_pages, mm, task, vm_write);
223
224 /* copied = space before - space after */
225 total_len -= iov_iter_count(iter);
226
227 /* If we have managed to copy any data at all then
228 we return the number of bytes copied. Otherwise
229 we return the error code */
230 if (total_len)
231 rc = total_len;
232
233 mmput(mm);
234
235put_task_struct:
236 put_task_struct(task);
237
238free_proc_pages:
239 if (process_pages != pp_stack)
240 kfree(process_pages);
241 return rc;
242}
243
244/**
245 * process_vm_rw - check iovecs before calling core routine
246 * @pid: PID of process to read/write from/to
247 * @lvec: iovec array specifying where to copy to/from locally
248 * @liovcnt: size of lvec array
249 * @rvec: iovec array specifying where to copy to/from in the other process
250 * @riovcnt: size of rvec array
251 * @flags: currently unused
252 * @vm_write: 0 if reading from other process, 1 if writing to other process
253 *
254 * Returns the number of bytes read/written or error code. May
255 * return less bytes than expected if an error occurs during the copying
256 * process.
257 */
258static ssize_t process_vm_rw(pid_t pid,
259 const struct iovec __user *lvec,
260 unsigned long liovcnt,
261 const struct iovec __user *rvec,
262 unsigned long riovcnt,
263 unsigned long flags, int vm_write)
264{
265 struct iovec iovstack_l[UIO_FASTIOV];
266 struct iovec iovstack_r[UIO_FASTIOV];
267 struct iovec *iov_l = iovstack_l;
268 struct iovec *iov_r = iovstack_r;
269 struct iov_iter iter;
270 ssize_t rc;
271 int dir = vm_write ? WRITE : READ;
272
273 if (flags != 0)
274 return -EINVAL;
275
276 /* Check iovecs */
277 rc = import_iovec(dir, lvec, liovcnt, UIO_FASTIOV, &iov_l, &iter);
278 if (rc < 0)
279 return rc;
280 if (!iov_iter_count(&iter))
281 goto free_iovecs;
282
283 rc = rw_copy_check_uvector(CHECK_IOVEC_ONLY, rvec, riovcnt, UIO_FASTIOV,
284 iovstack_r, &iov_r);
285 if (rc <= 0)
286 goto free_iovecs;
287
288 rc = process_vm_rw_core(pid, &iter, iov_r, riovcnt, flags, vm_write);
289
290free_iovecs:
291 if (iov_r != iovstack_r)
292 kfree(iov_r);
293 kfree(iov_l);
294
295 return rc;
296}
297
298SYSCALL_DEFINE6(process_vm_readv, pid_t, pid, const struct iovec __user *, lvec,
299 unsigned long, liovcnt, const struct iovec __user *, rvec,
300 unsigned long, riovcnt, unsigned long, flags)
301{
302 return process_vm_rw(pid, lvec, liovcnt, rvec, riovcnt, flags, 0);
303}
304
305SYSCALL_DEFINE6(process_vm_writev, pid_t, pid,
306 const struct iovec __user *, lvec,
307 unsigned long, liovcnt, const struct iovec __user *, rvec,
308 unsigned long, riovcnt, unsigned long, flags)
309{
310 return process_vm_rw(pid, lvec, liovcnt, rvec, riovcnt, flags, 1);
311}
312
313#ifdef CONFIG_COMPAT
314
315static ssize_t
316compat_process_vm_rw(compat_pid_t pid,
317 const struct compat_iovec __user *lvec,
318 unsigned long liovcnt,
319 const struct compat_iovec __user *rvec,
320 unsigned long riovcnt,
321 unsigned long flags, int vm_write)
322{
323 struct iovec iovstack_l[UIO_FASTIOV];
324 struct iovec iovstack_r[UIO_FASTIOV];
325 struct iovec *iov_l = iovstack_l;
326 struct iovec *iov_r = iovstack_r;
327 struct iov_iter iter;
328 ssize_t rc = -EFAULT;
329 int dir = vm_write ? WRITE : READ;
330
331 if (flags != 0)
332 return -EINVAL;
333
334 rc = compat_import_iovec(dir, lvec, liovcnt, UIO_FASTIOV, &iov_l, &iter);
335 if (rc < 0)
336 return rc;
337 if (!iov_iter_count(&iter))
338 goto free_iovecs;
339 rc = compat_rw_copy_check_uvector(CHECK_IOVEC_ONLY, rvec, riovcnt,
340 UIO_FASTIOV, iovstack_r,
341 &iov_r);
342 if (rc <= 0)
343 goto free_iovecs;
344
345 rc = process_vm_rw_core(pid, &iter, iov_r, riovcnt, flags, vm_write);
346
347free_iovecs:
348 if (iov_r != iovstack_r)
349 kfree(iov_r);
350 kfree(iov_l);
351 return rc;
352}
353
354COMPAT_SYSCALL_DEFINE6(process_vm_readv, compat_pid_t, pid,
355 const struct compat_iovec __user *, lvec,
356 compat_ulong_t, liovcnt,
357 const struct compat_iovec __user *, rvec,
358 compat_ulong_t, riovcnt,
359 compat_ulong_t, flags)
360{
361 return compat_process_vm_rw(pid, lvec, liovcnt, rvec,
362 riovcnt, flags, 0);
363}
364
365COMPAT_SYSCALL_DEFINE6(process_vm_writev, compat_pid_t, pid,
366 const struct compat_iovec __user *, lvec,
367 compat_ulong_t, liovcnt,
368 const struct compat_iovec __user *, rvec,
369 compat_ulong_t, riovcnt,
370 compat_ulong_t, flags)
371{
372 return compat_process_vm_rw(pid, lvec, liovcnt, rvec,
373 riovcnt, flags, 1);
374}
375
376#endif
377