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

source code of linux/mm/process_vm_access.c