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