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 2
57
58/* Maximum number of pages that can be stored at a time */
59#define PVM_MAX_USER_PAGES (PVM_MAX_KMALLOC_PAGES * PAGE_SIZE / sizeof(struct page *))
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 */
73static 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 int flags = 0;
86
87 /* Work out address and page range required */
88 if (len == 0)
89 return 0;
90 nr_pages = (addr + len - 1) / PAGE_SIZE - addr / PAGE_SIZE + 1;
91
92 if (vm_write)
93 flags |= FOLL_WRITE;
94
95 while (!rc && nr_pages && iov_iter_count(i: iter)) {
96 int pinned_pages = min_t(unsigned long, nr_pages, PVM_MAX_USER_PAGES);
97 int locked = 1;
98 size_t bytes;
99
100 /*
101 * Get the pages we're interested in. We must
102 * access remotely because task/mm might not
103 * current/current->mm
104 */
105 mmap_read_lock(mm);
106 pinned_pages = pin_user_pages_remote(mm, start: pa, nr_pages: pinned_pages,
107 gup_flags: flags, pages: process_pages,
108 locked: &locked);
109 if (locked)
110 mmap_read_unlock(mm);
111 if (pinned_pages <= 0)
112 return -EFAULT;
113
114 bytes = pinned_pages * PAGE_SIZE - start_offset;
115 if (bytes > len)
116 bytes = len;
117
118 rc = process_vm_rw_pages(pages: process_pages,
119 offset: start_offset, len: bytes, iter,
120 vm_write);
121 len -= bytes;
122 start_offset = 0;
123 nr_pages -= pinned_pages;
124 pa += pinned_pages * PAGE_SIZE;
125
126 /* If vm_write is set, the pages need to be made dirty: */
127 unpin_user_pages_dirty_lock(pages: process_pages, npages: pinned_pages,
128 make_dirty: vm_write);
129 }
130
131 return rc;
132}
133
134/* Maximum number of entries for process pages array
135 which lives on stack */
136#define PVM_MAX_PP_ARRAY_COUNT 16
137
138/**
139 * process_vm_rw_core - core of reading/writing pages from task specified
140 * @pid: PID of process to read/write from/to
141 * @iter: where to copy to/from locally
142 * @rvec: iovec array specifying where to copy to/from in the other process
143 * @riovcnt: size of rvec array
144 * @flags: currently unused
145 * @vm_write: 0 if reading from other process, 1 if writing to other process
146 *
147 * Returns the number of bytes read/written or error code. May
148 * return less bytes than expected if an error occurs during the copying
149 * process.
150 */
151static ssize_t process_vm_rw_core(pid_t pid, struct iov_iter *iter,
152 const struct iovec *rvec,
153 unsigned long riovcnt,
154 unsigned long flags, int vm_write)
155{
156 struct task_struct *task;
157 struct page *pp_stack[PVM_MAX_PP_ARRAY_COUNT];
158 struct page **process_pages = pp_stack;
159 struct mm_struct *mm;
160 unsigned long i;
161 ssize_t rc = 0;
162 unsigned long nr_pages = 0;
163 unsigned long nr_pages_iov;
164 ssize_t iov_len;
165 size_t total_len = iov_iter_count(i: iter);
166
167 /*
168 * Work out how many pages of struct pages we're going to need
169 * when eventually calling get_user_pages
170 */
171 for (i = 0; i < riovcnt; i++) {
172 iov_len = rvec[i].iov_len;
173 if (iov_len > 0) {
174 nr_pages_iov = ((unsigned long)rvec[i].iov_base
175 + iov_len - 1)
176 / PAGE_SIZE - (unsigned long)rvec[i].iov_base
177 / PAGE_SIZE + 1;
178 nr_pages = max(nr_pages, nr_pages_iov);
179 }
180 }
181
182 if (nr_pages == 0)
183 return 0;
184
185 if (nr_pages > PVM_MAX_PP_ARRAY_COUNT) {
186 /* For reliability don't try to kmalloc more than
187 2 pages worth */
188 process_pages = kmalloc(min_t(size_t, PVM_MAX_KMALLOC_PAGES * PAGE_SIZE,
189 sizeof(struct page *)*nr_pages),
190 GFP_KERNEL);
191
192 if (!process_pages)
193 return -ENOMEM;
194 }
195
196 /* Get process information */
197 task = find_get_task_by_vpid(nr: pid);
198 if (!task) {
199 rc = -ESRCH;
200 goto free_proc_pages;
201 }
202
203 mm = mm_access(task, PTRACE_MODE_ATTACH_REALCREDS);
204 if (IS_ERR(ptr: mm)) {
205 rc = PTR_ERR(ptr: mm);
206 /*
207 * Explicitly map EACCES to EPERM as EPERM is a more
208 * appropriate error code for process_vw_readv/writev
209 */
210 if (rc == -EACCES)
211 rc = -EPERM;
212 goto put_task_struct;
213 }
214
215 for (i = 0; i < riovcnt && iov_iter_count(i: iter) && !rc; i++)
216 rc = process_vm_rw_single_vec(
217 addr: (unsigned long)rvec[i].iov_base, len: rvec[i].iov_len,
218 iter, process_pages, mm, task, vm_write);
219
220 /* copied = space before - space after */
221 total_len -= iov_iter_count(i: iter);
222
223 /* If we have managed to copy any data at all then
224 we return the number of bytes copied. Otherwise
225 we return the error code */
226 if (total_len)
227 rc = total_len;
228
229 mmput(mm);
230
231put_task_struct:
232 put_task_struct(t: task);
233
234free_proc_pages:
235 if (process_pages != pp_stack)
236 kfree(objp: process_pages);
237 return rc;
238}
239
240/**
241 * process_vm_rw - check iovecs before calling core routine
242 * @pid: PID of process to read/write from/to
243 * @lvec: iovec array specifying where to copy to/from locally
244 * @liovcnt: size of lvec array
245 * @rvec: iovec array specifying where to copy to/from in the other process
246 * @riovcnt: size of rvec array
247 * @flags: currently unused
248 * @vm_write: 0 if reading from other process, 1 if writing to other process
249 *
250 * Returns the number of bytes read/written or error code. May
251 * return less bytes than expected if an error occurs during the copying
252 * process.
253 */
254static ssize_t process_vm_rw(pid_t pid,
255 const struct iovec __user *lvec,
256 unsigned long liovcnt,
257 const struct iovec __user *rvec,
258 unsigned long riovcnt,
259 unsigned long flags, int vm_write)
260{
261 struct iovec iovstack_l[UIO_FASTIOV];
262 struct iovec iovstack_r[UIO_FASTIOV];
263 struct iovec *iov_l = iovstack_l;
264 struct iovec *iov_r;
265 struct iov_iter iter;
266 ssize_t rc;
267 int dir = vm_write ? ITER_SOURCE : ITER_DEST;
268
269 if (flags != 0)
270 return -EINVAL;
271
272 /* Check iovecs */
273 rc = import_iovec(type: dir, uvec: lvec, nr_segs: liovcnt, UIO_FASTIOV, iovp: &iov_l, i: &iter);
274 if (rc < 0)
275 return rc;
276 if (!iov_iter_count(i: &iter))
277 goto free_iov_l;
278 iov_r = iovec_from_user(uvector: rvec, nr_segs: riovcnt, UIO_FASTIOV, fast_iov: iovstack_r,
279 in_compat_syscall());
280 if (IS_ERR(ptr: iov_r)) {
281 rc = PTR_ERR(ptr: iov_r);
282 goto free_iov_l;
283 }
284 rc = process_vm_rw_core(pid, iter: &iter, rvec: iov_r, riovcnt, flags, vm_write);
285 if (iov_r != iovstack_r)
286 kfree(objp: iov_r);
287free_iov_l:
288 kfree(objp: iov_l);
289 return rc;
290}
291
292SYSCALL_DEFINE6(process_vm_readv, pid_t, pid, const struct iovec __user *, lvec,
293 unsigned long, liovcnt, const struct iovec __user *, rvec,
294 unsigned long, riovcnt, unsigned long, flags)
295{
296 return process_vm_rw(pid, lvec, liovcnt, rvec, riovcnt, flags, vm_write: 0);
297}
298
299SYSCALL_DEFINE6(process_vm_writev, pid_t, pid,
300 const struct iovec __user *, lvec,
301 unsigned long, liovcnt, const struct iovec __user *, rvec,
302 unsigned long, riovcnt, unsigned long, flags)
303{
304 return process_vm_rw(pid, lvec, liovcnt, rvec, riovcnt, flags, vm_write: 1);
305}
306