]> git.ipfire.org Git - thirdparty/linux.git/blob - mm/process_vm_access.c
Merge tag 'libnvdimm-for-5.7' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdimm...
[thirdparty/linux.git] / mm / process_vm_access.c
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * linux/mm/process_vm_access.c
4 *
5 * Copyright (C) 2010-2011 Christopher Yeoh <cyeoh@au1.ibm.com>, IBM Corp.
6 */
7
8 #include <linux/mm.h>
9 #include <linux/uio.h>
10 #include <linux/sched.h>
11 #include <linux/sched/mm.h>
12 #include <linux/highmem.h>
13 #include <linux/ptrace.h>
14 #include <linux/slab.h>
15 #include <linux/syscalls.h>
16
17 #ifdef CONFIG_COMPAT
18 #include <linux/compat.h>
19 #endif
20
21 /**
22 * process_vm_rw_pages - read/write pages from task specified
23 * @pages: array of pointers to pages we want to copy
24 * @offset: offset in page to start copying from/to
25 * @len: number of bytes to copy
26 * @iter: where to copy to/from locally
27 * @vm_write: 0 means copy from, 1 means copy to
28 * Returns 0 on success, error code otherwise
29 */
30 static int process_vm_rw_pages(struct page **pages,
31 unsigned offset,
32 size_t len,
33 struct iov_iter *iter,
34 int vm_write)
35 {
36 /* Do the copy for each page */
37 while (len && iov_iter_count(iter)) {
38 struct page *page = *pages++;
39 size_t copy = PAGE_SIZE - offset;
40 size_t copied;
41
42 if (copy > len)
43 copy = len;
44
45 if (vm_write)
46 copied = copy_page_from_iter(page, offset, copy, iter);
47 else
48 copied = copy_page_to_iter(page, offset, copy, iter);
49
50 len -= copied;
51 if (copied < copy && iov_iter_count(iter))
52 return -EFAULT;
53 offset = 0;
54 }
55 return 0;
56 }
57
58 /* Maximum number of pages kmalloc'd to hold struct page's during copy */
59 #define PVM_MAX_KMALLOC_PAGES (PAGE_SIZE * 2)
60
61 /**
62 * process_vm_rw_single_vec - read/write pages from task specified
63 * @addr: start memory address of target process
64 * @len: size of area to copy to/from
65 * @iter: where to copy to/from locally
66 * @process_pages: struct pages area that can store at least
67 * nr_pages_to_copy struct page pointers
68 * @mm: mm for task
69 * @task: task to read/write from
70 * @vm_write: 0 means copy from, 1 means copy to
71 * Returns 0 on success or on failure error code
72 */
73 static int process_vm_rw_single_vec(unsigned long addr,
74 unsigned long len,
75 struct iov_iter *iter,
76 struct page **process_pages,
77 struct mm_struct *mm,
78 struct task_struct *task,
79 int vm_write)
80 {
81 unsigned long pa = addr & PAGE_MASK;
82 unsigned long start_offset = addr - pa;
83 unsigned long nr_pages;
84 ssize_t rc = 0;
85 unsigned long max_pages_per_loop = PVM_MAX_KMALLOC_PAGES
86 / sizeof(struct pages *);
87 unsigned int flags = 0;
88
89 /* Work out address and page range required */
90 if (len == 0)
91 return 0;
92 nr_pages = (addr + len - 1) / PAGE_SIZE - addr / PAGE_SIZE + 1;
93
94 if (vm_write)
95 flags |= FOLL_WRITE;
96
97 while (!rc && nr_pages && iov_iter_count(iter)) {
98 int pinned_pages = min(nr_pages, max_pages_per_loop);
99 int locked = 1;
100 size_t bytes;
101
102 /*
103 * Get the pages we're interested in. We must
104 * access remotely because task/mm might not
105 * current/current->mm
106 */
107 down_read(&mm->mmap_sem);
108 pinned_pages = pin_user_pages_remote(task, mm, pa, pinned_pages,
109 flags, process_pages,
110 NULL, &locked);
111 if (locked)
112 up_read(&mm->mmap_sem);
113 if (pinned_pages <= 0)
114 return -EFAULT;
115
116 bytes = pinned_pages * PAGE_SIZE - start_offset;
117 if (bytes > len)
118 bytes = len;
119
120 rc = process_vm_rw_pages(process_pages,
121 start_offset, bytes, iter,
122 vm_write);
123 len -= bytes;
124 start_offset = 0;
125 nr_pages -= pinned_pages;
126 pa += pinned_pages * PAGE_SIZE;
127
128 /* If vm_write is set, the pages need to be made dirty: */
129 unpin_user_pages_dirty_lock(process_pages, pinned_pages,
130 vm_write);
131 }
132
133 return rc;
134 }
135
136 /* Maximum number of entries for process pages array
137 which lives on stack */
138 #define PVM_MAX_PP_ARRAY_COUNT 16
139
140 /**
141 * process_vm_rw_core - core of reading/writing pages from task specified
142 * @pid: PID of process to read/write from/to
143 * @iter: where to copy to/from locally
144 * @rvec: iovec array specifying where to copy to/from in the other process
145 * @riovcnt: size of rvec array
146 * @flags: currently unused
147 * @vm_write: 0 if reading from other process, 1 if writing to other process
148 *
149 * Returns the number of bytes read/written or error code. May
150 * return less bytes than expected if an error occurs during the copying
151 * process.
152 */
153 static ssize_t process_vm_rw_core(pid_t pid, struct iov_iter *iter,
154 const struct iovec *rvec,
155 unsigned long riovcnt,
156 unsigned long flags, int vm_write)
157 {
158 struct task_struct *task;
159 struct page *pp_stack[PVM_MAX_PP_ARRAY_COUNT];
160 struct page **process_pages = pp_stack;
161 struct mm_struct *mm;
162 unsigned long i;
163 ssize_t rc = 0;
164 unsigned long nr_pages = 0;
165 unsigned long nr_pages_iov;
166 ssize_t iov_len;
167 size_t total_len = iov_iter_count(iter);
168
169 /*
170 * Work out how many pages of struct pages we're going to need
171 * when eventually calling get_user_pages
172 */
173 for (i = 0; i < riovcnt; i++) {
174 iov_len = rvec[i].iov_len;
175 if (iov_len > 0) {
176 nr_pages_iov = ((unsigned long)rvec[i].iov_base
177 + iov_len)
178 / PAGE_SIZE - (unsigned long)rvec[i].iov_base
179 / PAGE_SIZE + 1;
180 nr_pages = max(nr_pages, nr_pages_iov);
181 }
182 }
183
184 if (nr_pages == 0)
185 return 0;
186
187 if (nr_pages > PVM_MAX_PP_ARRAY_COUNT) {
188 /* For reliability don't try to kmalloc more than
189 2 pages worth */
190 process_pages = kmalloc(min_t(size_t, PVM_MAX_KMALLOC_PAGES,
191 sizeof(struct pages *)*nr_pages),
192 GFP_KERNEL);
193
194 if (!process_pages)
195 return -ENOMEM;
196 }
197
198 /* Get process information */
199 task = find_get_task_by_vpid(pid);
200 if (!task) {
201 rc = -ESRCH;
202 goto free_proc_pages;
203 }
204
205 mm = mm_access(task, PTRACE_MODE_ATTACH_REALCREDS);
206 if (!mm || IS_ERR(mm)) {
207 rc = IS_ERR(mm) ? PTR_ERR(mm) : -ESRCH;
208 /*
209 * Explicitly map EACCES to EPERM as EPERM is a more
210 * appropriate error code for process_vw_readv/writev
211 */
212 if (rc == -EACCES)
213 rc = -EPERM;
214 goto put_task_struct;
215 }
216
217 for (i = 0; i < riovcnt && iov_iter_count(iter) && !rc; i++)
218 rc = process_vm_rw_single_vec(
219 (unsigned long)rvec[i].iov_base, rvec[i].iov_len,
220 iter, process_pages, mm, task, vm_write);
221
222 /* copied = space before - space after */
223 total_len -= iov_iter_count(iter);
224
225 /* If we have managed to copy any data at all then
226 we return the number of bytes copied. Otherwise
227 we return the error code */
228 if (total_len)
229 rc = total_len;
230
231 mmput(mm);
232
233 put_task_struct:
234 put_task_struct(task);
235
236 free_proc_pages:
237 if (process_pages != pp_stack)
238 kfree(process_pages);
239 return rc;
240 }
241
242 /**
243 * process_vm_rw - check iovecs before calling core routine
244 * @pid: PID of process to read/write from/to
245 * @lvec: iovec array specifying where to copy to/from locally
246 * @liovcnt: size of lvec array
247 * @rvec: iovec array specifying where to copy to/from in the other process
248 * @riovcnt: size of rvec array
249 * @flags: currently unused
250 * @vm_write: 0 if reading from other process, 1 if writing to other process
251 *
252 * Returns the number of bytes read/written or error code. May
253 * return less bytes than expected if an error occurs during the copying
254 * process.
255 */
256 static ssize_t process_vm_rw(pid_t pid,
257 const struct iovec __user *lvec,
258 unsigned long liovcnt,
259 const struct iovec __user *rvec,
260 unsigned long riovcnt,
261 unsigned long flags, int vm_write)
262 {
263 struct iovec iovstack_l[UIO_FASTIOV];
264 struct iovec iovstack_r[UIO_FASTIOV];
265 struct iovec *iov_l = iovstack_l;
266 struct iovec *iov_r = iovstack_r;
267 struct iov_iter iter;
268 ssize_t rc;
269 int dir = vm_write ? WRITE : READ;
270
271 if (flags != 0)
272 return -EINVAL;
273
274 /* Check iovecs */
275 rc = import_iovec(dir, lvec, liovcnt, UIO_FASTIOV, &iov_l, &iter);
276 if (rc < 0)
277 return rc;
278 if (!iov_iter_count(&iter))
279 goto free_iovecs;
280
281 rc = rw_copy_check_uvector(CHECK_IOVEC_ONLY, rvec, riovcnt, UIO_FASTIOV,
282 iovstack_r, &iov_r);
283 if (rc <= 0)
284 goto free_iovecs;
285
286 rc = process_vm_rw_core(pid, &iter, iov_r, riovcnt, flags, vm_write);
287
288 free_iovecs:
289 if (iov_r != iovstack_r)
290 kfree(iov_r);
291 kfree(iov_l);
292
293 return rc;
294 }
295
296 SYSCALL_DEFINE6(process_vm_readv, pid_t, pid, const struct iovec __user *, lvec,
297 unsigned long, liovcnt, const struct iovec __user *, rvec,
298 unsigned long, riovcnt, unsigned long, flags)
299 {
300 return process_vm_rw(pid, lvec, liovcnt, rvec, riovcnt, flags, 0);
301 }
302
303 SYSCALL_DEFINE6(process_vm_writev, pid_t, pid,
304 const struct iovec __user *, lvec,
305 unsigned long, liovcnt, const struct iovec __user *, rvec,
306 unsigned long, riovcnt, unsigned long, flags)
307 {
308 return process_vm_rw(pid, lvec, liovcnt, rvec, riovcnt, flags, 1);
309 }
310
311 #ifdef CONFIG_COMPAT
312
313 static ssize_t
314 compat_process_vm_rw(compat_pid_t pid,
315 const struct compat_iovec __user *lvec,
316 unsigned long liovcnt,
317 const struct compat_iovec __user *rvec,
318 unsigned long riovcnt,
319 unsigned long flags, int vm_write)
320 {
321 struct iovec iovstack_l[UIO_FASTIOV];
322 struct iovec iovstack_r[UIO_FASTIOV];
323 struct iovec *iov_l = iovstack_l;
324 struct iovec *iov_r = iovstack_r;
325 struct iov_iter iter;
326 ssize_t rc = -EFAULT;
327 int dir = vm_write ? WRITE : READ;
328
329 if (flags != 0)
330 return -EINVAL;
331
332 rc = compat_import_iovec(dir, lvec, liovcnt, UIO_FASTIOV, &iov_l, &iter);
333 if (rc < 0)
334 return rc;
335 if (!iov_iter_count(&iter))
336 goto free_iovecs;
337 rc = compat_rw_copy_check_uvector(CHECK_IOVEC_ONLY, rvec, riovcnt,
338 UIO_FASTIOV, iovstack_r,
339 &iov_r);
340 if (rc <= 0)
341 goto free_iovecs;
342
343 rc = process_vm_rw_core(pid, &iter, iov_r, riovcnt, flags, vm_write);
344
345 free_iovecs:
346 if (iov_r != iovstack_r)
347 kfree(iov_r);
348 kfree(iov_l);
349 return rc;
350 }
351
352 COMPAT_SYSCALL_DEFINE6(process_vm_readv, compat_pid_t, pid,
353 const struct compat_iovec __user *, lvec,
354 compat_ulong_t, liovcnt,
355 const struct compat_iovec __user *, rvec,
356 compat_ulong_t, riovcnt,
357 compat_ulong_t, flags)
358 {
359 return compat_process_vm_rw(pid, lvec, liovcnt, rvec,
360 riovcnt, flags, 0);
361 }
362
363 COMPAT_SYSCALL_DEFINE6(process_vm_writev, compat_pid_t, pid,
364 const struct compat_iovec __user *, lvec,
365 compat_ulong_t, liovcnt,
366 const struct compat_iovec __user *, rvec,
367 compat_ulong_t, riovcnt,
368 compat_ulong_t, flags)
369 {
370 return compat_process_vm_rw(pid, lvec, liovcnt, rvec,
371 riovcnt, flags, 1);
372 }
373
374 #endif