blob: 5816d4c4cab09c22274c2cdc40f4176f16426f0a [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/read_write.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7#include <linux/slab.h>
8#include <linux/stat.h>
9#include <linux/fcntl.h>
10#include <linux/file.h>
11#include <linux/uio.h>
Robert Love0eeca282005-07-12 17:06:03 -040012#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/security.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050014#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/syscalls.h>
Linus Torvaldse28cc712006-01-04 16:20:40 -080016#include <linux/pagemap.h>
Jens Axboed6b29d72007-06-04 09:59:47 +020017#include <linux/splice.h>
Al Viro561c6732013-02-24 10:52:26 -050018#include <linux/compat.h>
Zach Brown29732932015-11-10 16:53:30 -050019#include <linux/mount.h>
Wouter van Kesteren2feb55f2016-02-16 22:20:59 +010020#include <linux/fs.h>
Al Viro06ae43f2013-03-20 13:19:30 -040021#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070022
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080023#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <asm/unistd.h>
25
Al Viroc0bd14af2013-05-04 15:00:54 -040026typedef ssize_t (*io_fn_t)(struct file *, char __user *, size_t, loff_t *);
Al Viro293bc982014-02-11 18:37:41 -050027typedef ssize_t (*iter_fn_t)(struct kiocb *, struct iov_iter *);
Al Viroc0bd14af2013-05-04 15:00:54 -040028
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -080029const struct file_operations generic_ro_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070030 .llseek = generic_file_llseek,
Al Viroaad4f8b2014-04-02 14:33:16 -040031 .read_iter = generic_file_read_iter,
Linus Torvalds1da177e2005-04-16 15:20:36 -070032 .mmap = generic_file_readonly_mmap,
Jens Axboe534f2aa2007-06-01 14:52:37 +020033 .splice_read = generic_file_splice_read,
Linus Torvalds1da177e2005-04-16 15:20:36 -070034};
35
36EXPORT_SYMBOL(generic_ro_fops);
37
Al Virocccb5a12010-12-17 07:44:05 -050038static inline int unsigned_offsets(struct file *file)
KAMEZAWA Hiroyuki4a3956c2010-10-01 14:20:22 -070039{
Al Virocccb5a12010-12-17 07:44:05 -050040 return file->f_mode & FMODE_UNSIGNED_OFFSET;
KAMEZAWA Hiroyuki4a3956c2010-10-01 14:20:22 -070041}
42
Jie Liu46a1c2c2013-06-25 12:02:13 +080043/**
44 * vfs_setpos - update the file offset for lseek
45 * @file: file structure in question
46 * @offset: file offset to seek to
47 * @maxsize: maximum file size
48 *
49 * This is a low-level filesystem helper for updating the file offset to
50 * the value specified by @offset if the given offset is valid and it is
51 * not equal to the current file offset.
52 *
53 * Return the specified offset on success and -EINVAL on invalid offset.
54 */
55loff_t vfs_setpos(struct file *file, loff_t offset, loff_t maxsize)
Andi Kleenef3d0fd2011-09-15 16:06:48 -070056{
57 if (offset < 0 && !unsigned_offsets(file))
58 return -EINVAL;
59 if (offset > maxsize)
60 return -EINVAL;
61
62 if (offset != file->f_pos) {
63 file->f_pos = offset;
64 file->f_version = 0;
65 }
66 return offset;
67}
Jie Liu46a1c2c2013-06-25 12:02:13 +080068EXPORT_SYMBOL(vfs_setpos);
Andi Kleenef3d0fd2011-09-15 16:06:48 -070069
Christoph Hellwig3a8cff42008-08-11 15:37:17 +020070/**
Andi Kleen57604952011-09-15 16:06:50 -070071 * generic_file_llseek_size - generic llseek implementation for regular files
Christoph Hellwig3a8cff42008-08-11 15:37:17 +020072 * @file: file structure to seek on
73 * @offset: file offset to seek to
Andrew Morton965c8e52012-12-17 15:59:39 -080074 * @whence: type of seek
Eric Sandeene8b96eb2012-04-30 13:11:29 -050075 * @size: max size of this file in file system
76 * @eof: offset used for SEEK_END position
Christoph Hellwig3a8cff42008-08-11 15:37:17 +020077 *
Andi Kleen57604952011-09-15 16:06:50 -070078 * This is a variant of generic_file_llseek that allows passing in a custom
Eric Sandeene8b96eb2012-04-30 13:11:29 -050079 * maximum file size and a custom EOF position, for e.g. hashed directories
Andi Kleenef3d0fd2011-09-15 16:06:48 -070080 *
81 * Synchronization:
Andi Kleen57604952011-09-15 16:06:50 -070082 * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
Andi Kleenef3d0fd2011-09-15 16:06:48 -070083 * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
84 * read/writes behave like SEEK_SET against seeks.
Christoph Hellwig3a8cff42008-08-11 15:37:17 +020085 */
Andi Kleen9465efc2008-06-27 11:05:24 +020086loff_t
Andrew Morton965c8e52012-12-17 15:59:39 -080087generic_file_llseek_size(struct file *file, loff_t offset, int whence,
Eric Sandeene8b96eb2012-04-30 13:11:29 -050088 loff_t maxsize, loff_t eof)
Linus Torvalds1da177e2005-04-16 15:20:36 -070089{
Andrew Morton965c8e52012-12-17 15:59:39 -080090 switch (whence) {
Christoph Hellwig3a8cff42008-08-11 15:37:17 +020091 case SEEK_END:
Eric Sandeene8b96eb2012-04-30 13:11:29 -050092 offset += eof;
Christoph Hellwig3a8cff42008-08-11 15:37:17 +020093 break;
94 case SEEK_CUR:
Alain Knaff5b6f1eb2008-11-10 17:08:08 -080095 /*
96 * Here we special-case the lseek(fd, 0, SEEK_CUR)
97 * position-querying operation. Avoid rewriting the "same"
98 * f_pos value back to the file because a concurrent read(),
99 * write() or lseek() might have altered it
100 */
101 if (offset == 0)
102 return file->f_pos;
Andi Kleenef3d0fd2011-09-15 16:06:48 -0700103 /*
104 * f_lock protects against read/modify/write race with other
105 * SEEK_CURs. Note that parallel writes and reads behave
106 * like SEEK_SET.
107 */
108 spin_lock(&file->f_lock);
Jie Liu46a1c2c2013-06-25 12:02:13 +0800109 offset = vfs_setpos(file, file->f_pos + offset, maxsize);
Andi Kleenef3d0fd2011-09-15 16:06:48 -0700110 spin_unlock(&file->f_lock);
111 return offset;
Josef Bacik982d8162011-07-18 13:21:35 -0400112 case SEEK_DATA:
113 /*
114 * In the generic case the entire file is data, so as long as
115 * offset isn't at the end of the file then the offset is data.
116 */
Eric Sandeene8b96eb2012-04-30 13:11:29 -0500117 if (offset >= eof)
Josef Bacik982d8162011-07-18 13:21:35 -0400118 return -ENXIO;
119 break;
120 case SEEK_HOLE:
121 /*
122 * There is a virtual hole at the end of the file, so as long as
123 * offset isn't i_size or larger, return i_size.
124 */
Eric Sandeene8b96eb2012-04-30 13:11:29 -0500125 if (offset >= eof)
Josef Bacik982d8162011-07-18 13:21:35 -0400126 return -ENXIO;
Eric Sandeene8b96eb2012-04-30 13:11:29 -0500127 offset = eof;
Josef Bacik982d8162011-07-18 13:21:35 -0400128 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 }
Christoph Hellwig3a8cff42008-08-11 15:37:17 +0200130
Jie Liu46a1c2c2013-06-25 12:02:13 +0800131 return vfs_setpos(file, offset, maxsize);
Andi Kleen57604952011-09-15 16:06:50 -0700132}
133EXPORT_SYMBOL(generic_file_llseek_size);
134
135/**
136 * generic_file_llseek - generic llseek implementation for regular files
137 * @file: file structure to seek on
138 * @offset: file offset to seek to
Andrew Morton965c8e52012-12-17 15:59:39 -0800139 * @whence: type of seek
Andi Kleen57604952011-09-15 16:06:50 -0700140 *
141 * This is a generic implemenation of ->llseek useable for all normal local
142 * filesystems. It just updates the file offset to the value specified by
Ming Lei546ae2d2013-04-29 15:06:07 -0700143 * @offset and @whence.
Andi Kleen57604952011-09-15 16:06:50 -0700144 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800145loff_t generic_file_llseek(struct file *file, loff_t offset, int whence)
Andi Kleen57604952011-09-15 16:06:50 -0700146{
147 struct inode *inode = file->f_mapping->host;
148
Andrew Morton965c8e52012-12-17 15:59:39 -0800149 return generic_file_llseek_size(file, offset, whence,
Eric Sandeene8b96eb2012-04-30 13:11:29 -0500150 inode->i_sb->s_maxbytes,
151 i_size_read(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152}
Andi Kleen9465efc2008-06-27 11:05:24 +0200153EXPORT_SYMBOL(generic_file_llseek);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154
jan Blunckae6afc32010-05-26 14:44:48 -0700155/**
Al Viro1bf9d142013-06-16 20:27:42 +0400156 * fixed_size_llseek - llseek implementation for fixed-sized devices
157 * @file: file structure to seek on
158 * @offset: file offset to seek to
159 * @whence: type of seek
160 * @size: size of the file
161 *
162 */
163loff_t fixed_size_llseek(struct file *file, loff_t offset, int whence, loff_t size)
164{
165 switch (whence) {
166 case SEEK_SET: case SEEK_CUR: case SEEK_END:
167 return generic_file_llseek_size(file, offset, whence,
168 size, size);
169 default:
170 return -EINVAL;
171 }
172}
173EXPORT_SYMBOL(fixed_size_llseek);
174
175/**
Al Virob25472f2015-12-05 22:04:48 -0500176 * no_seek_end_llseek - llseek implementation for fixed-sized devices
177 * @file: file structure to seek on
178 * @offset: file offset to seek to
179 * @whence: type of seek
180 *
181 */
182loff_t no_seek_end_llseek(struct file *file, loff_t offset, int whence)
183{
184 switch (whence) {
185 case SEEK_SET: case SEEK_CUR:
186 return generic_file_llseek_size(file, offset, whence,
Wouter van Kesteren2feb55f2016-02-16 22:20:59 +0100187 OFFSET_MAX, 0);
Al Virob25472f2015-12-05 22:04:48 -0500188 default:
189 return -EINVAL;
190 }
191}
192EXPORT_SYMBOL(no_seek_end_llseek);
193
194/**
195 * no_seek_end_llseek_size - llseek implementation for fixed-sized devices
196 * @file: file structure to seek on
197 * @offset: file offset to seek to
198 * @whence: type of seek
199 * @size: maximal offset allowed
200 *
201 */
202loff_t no_seek_end_llseek_size(struct file *file, loff_t offset, int whence, loff_t size)
203{
204 switch (whence) {
205 case SEEK_SET: case SEEK_CUR:
206 return generic_file_llseek_size(file, offset, whence,
207 size, 0);
208 default:
209 return -EINVAL;
210 }
211}
212EXPORT_SYMBOL(no_seek_end_llseek_size);
213
214/**
jan Blunckae6afc32010-05-26 14:44:48 -0700215 * noop_llseek - No Operation Performed llseek implementation
216 * @file: file structure to seek on
217 * @offset: file offset to seek to
Andrew Morton965c8e52012-12-17 15:59:39 -0800218 * @whence: type of seek
jan Blunckae6afc32010-05-26 14:44:48 -0700219 *
220 * This is an implementation of ->llseek useable for the rare special case when
221 * userspace expects the seek to succeed but the (device) file is actually not
222 * able to perform the seek. In this case you use noop_llseek() instead of
223 * falling back to the default implementation of ->llseek.
224 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800225loff_t noop_llseek(struct file *file, loff_t offset, int whence)
jan Blunckae6afc32010-05-26 14:44:48 -0700226{
227 return file->f_pos;
228}
229EXPORT_SYMBOL(noop_llseek);
230
Andrew Morton965c8e52012-12-17 15:59:39 -0800231loff_t no_llseek(struct file *file, loff_t offset, int whence)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232{
233 return -ESPIPE;
234}
235EXPORT_SYMBOL(no_llseek);
236
Andrew Morton965c8e52012-12-17 15:59:39 -0800237loff_t default_llseek(struct file *file, loff_t offset, int whence)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238{
Al Viro496ad9a2013-01-23 17:07:38 -0500239 struct inode *inode = file_inode(file);
David Sterba16abef02008-04-22 15:09:22 +0200240 loff_t retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241
Al Viro59551022016-01-22 15:40:57 -0500242 inode_lock(inode);
Andrew Morton965c8e52012-12-17 15:59:39 -0800243 switch (whence) {
Chris Snook7b8e8922007-05-08 00:24:13 -0700244 case SEEK_END:
Josef Bacik982d8162011-07-18 13:21:35 -0400245 offset += i_size_read(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 break;
Chris Snook7b8e8922007-05-08 00:24:13 -0700247 case SEEK_CUR:
Alain Knaff5b6f1eb2008-11-10 17:08:08 -0800248 if (offset == 0) {
249 retval = file->f_pos;
250 goto out;
251 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 offset += file->f_pos;
Josef Bacik982d8162011-07-18 13:21:35 -0400253 break;
254 case SEEK_DATA:
255 /*
256 * In the generic case the entire file is data, so as
257 * long as offset isn't at the end of the file then the
258 * offset is data.
259 */
Dan Carpenterbacb2d82011-07-26 17:25:20 +0300260 if (offset >= inode->i_size) {
261 retval = -ENXIO;
262 goto out;
263 }
Josef Bacik982d8162011-07-18 13:21:35 -0400264 break;
265 case SEEK_HOLE:
266 /*
267 * There is a virtual hole at the end of the file, so
268 * as long as offset isn't i_size or larger, return
269 * i_size.
270 */
Dan Carpenterbacb2d82011-07-26 17:25:20 +0300271 if (offset >= inode->i_size) {
272 retval = -ENXIO;
273 goto out;
274 }
Josef Bacik982d8162011-07-18 13:21:35 -0400275 offset = inode->i_size;
276 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 }
278 retval = -EINVAL;
Al Virocccb5a12010-12-17 07:44:05 -0500279 if (offset >= 0 || unsigned_offsets(file)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 if (offset != file->f_pos) {
281 file->f_pos = offset;
282 file->f_version = 0;
283 }
284 retval = offset;
285 }
Alain Knaff5b6f1eb2008-11-10 17:08:08 -0800286out:
Al Viro59551022016-01-22 15:40:57 -0500287 inode_unlock(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 return retval;
289}
290EXPORT_SYMBOL(default_llseek);
291
Andrew Morton965c8e52012-12-17 15:59:39 -0800292loff_t vfs_llseek(struct file *file, loff_t offset, int whence)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293{
294 loff_t (*fn)(struct file *, loff_t, int);
295
296 fn = no_llseek;
297 if (file->f_mode & FMODE_LSEEK) {
Al Viro72c2d532013-09-22 16:27:52 -0400298 if (file->f_op->llseek)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 fn = file->f_op->llseek;
300 }
Andrew Morton965c8e52012-12-17 15:59:39 -0800301 return fn(file, offset, whence);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302}
303EXPORT_SYMBOL(vfs_llseek);
304
Andrew Morton965c8e52012-12-17 15:59:39 -0800305SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306{
307 off_t retval;
Linus Torvalds9c225f22014-03-03 09:36:58 -0800308 struct fd f = fdget_pos(fd);
Al Viro2903ff02012-08-28 12:52:22 -0400309 if (!f.file)
310 return -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311
312 retval = -EINVAL;
Andrew Morton965c8e52012-12-17 15:59:39 -0800313 if (whence <= SEEK_MAX) {
314 loff_t res = vfs_llseek(f.file, offset, whence);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 retval = res;
316 if (res != (loff_t)retval)
317 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
318 }
Linus Torvalds9c225f22014-03-03 09:36:58 -0800319 fdput_pos(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 return retval;
321}
322
Al Viro561c6732013-02-24 10:52:26 -0500323#ifdef CONFIG_COMPAT
324COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
325{
326 return sys_lseek(fd, offset, whence);
327}
328#endif
329
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330#ifdef __ARCH_WANT_SYS_LLSEEK
Heiko Carstens003d7ab2009-01-14 14:14:21 +0100331SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
332 unsigned long, offset_low, loff_t __user *, result,
Andrew Morton965c8e52012-12-17 15:59:39 -0800333 unsigned int, whence)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334{
335 int retval;
Eric Biggersd7a15f82014-03-16 14:24:08 -0500336 struct fd f = fdget_pos(fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 loff_t offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338
Al Viro2903ff02012-08-28 12:52:22 -0400339 if (!f.file)
340 return -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341
342 retval = -EINVAL;
Andrew Morton965c8e52012-12-17 15:59:39 -0800343 if (whence > SEEK_MAX)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 goto out_putf;
345
Al Viro2903ff02012-08-28 12:52:22 -0400346 offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
Andrew Morton965c8e52012-12-17 15:59:39 -0800347 whence);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348
349 retval = (int)offset;
350 if (offset >= 0) {
351 retval = -EFAULT;
352 if (!copy_to_user(result, &offset, sizeof(offset)))
353 retval = 0;
354 }
355out_putf:
Eric Biggersd7a15f82014-03-16 14:24:08 -0500356 fdput_pos(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 return retval;
358}
359#endif
360
Christoph Hellwigdbe4e192015-01-25 21:11:59 +0100361ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos)
362{
363 struct kiocb kiocb;
364 ssize_t ret;
365
366 if (!file->f_op->read_iter)
367 return -EINVAL;
368
369 init_sync_kiocb(&kiocb, file);
370 kiocb.ki_pos = *ppos;
Christoph Hellwigdbe4e192015-01-25 21:11:59 +0100371
372 iter->type |= READ;
373 ret = file->f_op->read_iter(&kiocb, iter);
Christoph Hellwig599bd192015-02-11 19:59:44 +0100374 BUG_ON(ret == -EIOCBQUEUED);
Christoph Hellwigdbe4e192015-01-25 21:11:59 +0100375 if (ret > 0)
376 *ppos = kiocb.ki_pos;
377 return ret;
378}
379EXPORT_SYMBOL(vfs_iter_read);
380
381ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos)
382{
383 struct kiocb kiocb;
384 ssize_t ret;
385
386 if (!file->f_op->write_iter)
387 return -EINVAL;
388
389 init_sync_kiocb(&kiocb, file);
390 kiocb.ki_pos = *ppos;
Christoph Hellwigdbe4e192015-01-25 21:11:59 +0100391
392 iter->type |= WRITE;
393 ret = file->f_op->write_iter(&kiocb, iter);
Christoph Hellwig599bd192015-02-11 19:59:44 +0100394 BUG_ON(ret == -EIOCBQUEUED);
Christoph Hellwigdbe4e192015-01-25 21:11:59 +0100395 if (ret > 0)
396 *ppos = kiocb.ki_pos;
397 return ret;
398}
399EXPORT_SYMBOL(vfs_iter_write);
400
Al Viro68d70d02013-06-19 15:26:04 +0400401int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402{
403 struct inode *inode;
404 loff_t pos;
James Morrisc43e2592008-01-12 22:05:48 +1100405 int retval = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
Al Viro496ad9a2013-01-23 17:07:38 -0500407 inode = file_inode(file);
Linus Torvaldse28cc712006-01-04 16:20:40 -0800408 if (unlikely((ssize_t) count < 0))
James Morrisc43e2592008-01-12 22:05:48 +1100409 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 pos = *ppos;
Al Virocccb5a12010-12-17 07:44:05 -0500411 if (unlikely(pos < 0)) {
412 if (!unsigned_offsets(file))
413 return retval;
414 if (count >= -pos) /* both values are in 0..LLONG_MAX */
415 return -EOVERFLOW;
416 } else if (unlikely((loff_t) (pos + count) < 0)) {
417 if (!unsigned_offsets(file))
KAMEZAWA Hiroyuki4a3956c2010-10-01 14:20:22 -0700418 return retval;
419 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500421 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
Christoph Hellwigacc15572015-12-03 12:59:49 +0100422 retval = locks_mandatory_area(inode, file, pos, pos + count - 1,
423 read_write == READ ? F_RDLCK : F_WRLCK);
Linus Torvaldse28cc712006-01-04 16:20:40 -0800424 if (retval < 0)
425 return retval;
426 }
Al Virobc613842016-03-31 21:48:20 -0400427 return security_file_permission(file,
James Morrisc43e2592008-01-12 22:05:48 +1100428 read_write == READ ? MAY_READ : MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429}
430
Al Viro5d5d5682015-04-03 15:41:18 -0400431static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
Al Viro293bc982014-02-11 18:37:41 -0500432{
433 struct iovec iov = { .iov_base = buf, .iov_len = len };
434 struct kiocb kiocb;
435 struct iov_iter iter;
436 ssize_t ret;
437
438 init_sync_kiocb(&kiocb, filp);
439 kiocb.ki_pos = *ppos;
Al Viro293bc982014-02-11 18:37:41 -0500440 iov_iter_init(&iter, READ, &iov, 1, len);
441
442 ret = filp->f_op->read_iter(&kiocb, &iter);
Christoph Hellwig599bd192015-02-11 19:59:44 +0100443 BUG_ON(ret == -EIOCBQUEUED);
Al Viro293bc982014-02-11 18:37:41 -0500444 *ppos = kiocb.ki_pos;
445 return ret;
446}
447
Dmitry Kasatkin6fb50322014-11-05 17:01:17 +0200448ssize_t __vfs_read(struct file *file, char __user *buf, size_t count,
449 loff_t *pos)
450{
Dmitry Kasatkin6fb50322014-11-05 17:01:17 +0200451 if (file->f_op->read)
Al Viro3d04c8a2015-04-03 15:09:18 -0400452 return file->f_op->read(file, buf, count, pos);
Dmitry Kasatkin6fb50322014-11-05 17:01:17 +0200453 else if (file->f_op->read_iter)
Al Viro3d04c8a2015-04-03 15:09:18 -0400454 return new_sync_read(file, buf, count, pos);
Dmitry Kasatkin6fb50322014-11-05 17:01:17 +0200455 else
Al Viro3d04c8a2015-04-03 15:09:18 -0400456 return -EINVAL;
Dmitry Kasatkin6fb50322014-11-05 17:01:17 +0200457}
Al Viro3d04c8a2015-04-03 15:09:18 -0400458EXPORT_SYMBOL(__vfs_read);
Dmitry Kasatkin6fb50322014-11-05 17:01:17 +0200459
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
461{
462 ssize_t ret;
463
464 if (!(file->f_mode & FMODE_READ))
465 return -EBADF;
Al Viro7f7f25e2014-02-11 17:49:24 -0500466 if (!(file->f_mode & FMODE_CAN_READ))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 return -EINVAL;
468 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
469 return -EFAULT;
470
471 ret = rw_verify_area(READ, file, pos, count);
Al Virobc613842016-03-31 21:48:20 -0400472 if (!ret) {
473 if (count > MAX_RW_COUNT)
474 count = MAX_RW_COUNT;
Dmitry Kasatkin6fb50322014-11-05 17:01:17 +0200475 ret = __vfs_read(file, buf, count, pos);
James Morrisc43e2592008-01-12 22:05:48 +1100476 if (ret > 0) {
Eric Paris2a12a9d2009-12-17 21:24:21 -0500477 fsnotify_access(file);
James Morrisc43e2592008-01-12 22:05:48 +1100478 add_rchar(current, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 }
James Morrisc43e2592008-01-12 22:05:48 +1100480 inc_syscr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 }
482
483 return ret;
484}
485
486EXPORT_SYMBOL(vfs_read);
487
Al Viro5d5d5682015-04-03 15:41:18 -0400488static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
Al Viro293bc982014-02-11 18:37:41 -0500489{
490 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
491 struct kiocb kiocb;
492 struct iov_iter iter;
493 ssize_t ret;
494
495 init_sync_kiocb(&kiocb, filp);
496 kiocb.ki_pos = *ppos;
Al Viro293bc982014-02-11 18:37:41 -0500497 iov_iter_init(&iter, WRITE, &iov, 1, len);
498
499 ret = filp->f_op->write_iter(&kiocb, &iter);
Christoph Hellwig599bd192015-02-11 19:59:44 +0100500 BUG_ON(ret == -EIOCBQUEUED);
Al Virof765b132015-04-06 20:50:38 -0400501 if (ret > 0)
502 *ppos = kiocb.ki_pos;
Al Viro293bc982014-02-11 18:37:41 -0500503 return ret;
504}
505
Al Viro493c84c2015-04-03 15:06:43 -0400506ssize_t __vfs_write(struct file *file, const char __user *p, size_t count,
507 loff_t *pos)
508{
509 if (file->f_op->write)
510 return file->f_op->write(file, p, count, pos);
Al Viro493c84c2015-04-03 15:06:43 -0400511 else if (file->f_op->write_iter)
512 return new_sync_write(file, p, count, pos);
513 else
514 return -EINVAL;
515}
516EXPORT_SYMBOL(__vfs_write);
517
Al Viro06ae43f2013-03-20 13:19:30 -0400518ssize_t __kernel_write(struct file *file, const char *buf, size_t count, loff_t *pos)
519{
520 mm_segment_t old_fs;
521 const char __user *p;
522 ssize_t ret;
523
Al Viro7f7f25e2014-02-11 17:49:24 -0500524 if (!(file->f_mode & FMODE_CAN_WRITE))
Al Viro3e84f482013-03-27 15:20:30 +0000525 return -EINVAL;
526
Al Viro06ae43f2013-03-20 13:19:30 -0400527 old_fs = get_fs();
528 set_fs(get_ds());
529 p = (__force const char __user *)buf;
530 if (count > MAX_RW_COUNT)
531 count = MAX_RW_COUNT;
Al Viro493c84c2015-04-03 15:06:43 -0400532 ret = __vfs_write(file, p, count, pos);
Al Viro06ae43f2013-03-20 13:19:30 -0400533 set_fs(old_fs);
534 if (ret > 0) {
535 fsnotify_modify(file);
536 add_wchar(current, ret);
537 }
538 inc_syscw(current);
539 return ret;
540}
541
Al Viro2ec3a122014-08-19 11:48:09 -0400542EXPORT_SYMBOL(__kernel_write);
543
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
545{
546 ssize_t ret;
547
548 if (!(file->f_mode & FMODE_WRITE))
549 return -EBADF;
Al Viro7f7f25e2014-02-11 17:49:24 -0500550 if (!(file->f_mode & FMODE_CAN_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 return -EINVAL;
552 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
553 return -EFAULT;
554
555 ret = rw_verify_area(WRITE, file, pos, count);
Al Virobc613842016-03-31 21:48:20 -0400556 if (!ret) {
557 if (count > MAX_RW_COUNT)
558 count = MAX_RW_COUNT;
Al Viro03d95eb2013-03-20 13:04:20 -0400559 file_start_write(file);
Al Viro493c84c2015-04-03 15:06:43 -0400560 ret = __vfs_write(file, buf, count, pos);
James Morrisc43e2592008-01-12 22:05:48 +1100561 if (ret > 0) {
Eric Paris2a12a9d2009-12-17 21:24:21 -0500562 fsnotify_modify(file);
James Morrisc43e2592008-01-12 22:05:48 +1100563 add_wchar(current, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 }
James Morrisc43e2592008-01-12 22:05:48 +1100565 inc_syscw(current);
Al Viro03d95eb2013-03-20 13:04:20 -0400566 file_end_write(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 }
568
569 return ret;
570}
571
572EXPORT_SYMBOL(vfs_write);
573
574static inline loff_t file_pos_read(struct file *file)
575{
576 return file->f_pos;
577}
578
579static inline void file_pos_write(struct file *file, loff_t pos)
580{
581 file->f_pos = pos;
582}
583
Heiko Carstens3cdad422009-01-14 14:14:22 +0100584SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585{
Linus Torvalds9c225f22014-03-03 09:36:58 -0800586 struct fd f = fdget_pos(fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 ssize_t ret = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588
Al Viro2903ff02012-08-28 12:52:22 -0400589 if (f.file) {
590 loff_t pos = file_pos_read(f.file);
591 ret = vfs_read(f.file, buf, count, &pos);
Al Viro5faf1532013-06-15 05:49:36 +0400592 if (ret >= 0)
593 file_pos_write(f.file, pos);
Linus Torvalds9c225f22014-03-03 09:36:58 -0800594 fdput_pos(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 return ret;
597}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598
Heiko Carstens3cdad422009-01-14 14:14:22 +0100599SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
600 size_t, count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601{
Linus Torvalds9c225f22014-03-03 09:36:58 -0800602 struct fd f = fdget_pos(fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 ssize_t ret = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604
Al Viro2903ff02012-08-28 12:52:22 -0400605 if (f.file) {
606 loff_t pos = file_pos_read(f.file);
607 ret = vfs_write(f.file, buf, count, &pos);
Al Viro5faf1532013-06-15 05:49:36 +0400608 if (ret >= 0)
609 file_pos_write(f.file, pos);
Linus Torvalds9c225f22014-03-03 09:36:58 -0800610 fdput_pos(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 }
612
613 return ret;
614}
615
Al Viro4a0fd5b2013-01-21 15:16:58 -0500616SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
617 size_t, count, loff_t, pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618{
Al Viro2903ff02012-08-28 12:52:22 -0400619 struct fd f;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 ssize_t ret = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621
622 if (pos < 0)
623 return -EINVAL;
624
Al Viro2903ff02012-08-28 12:52:22 -0400625 f = fdget(fd);
626 if (f.file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 ret = -ESPIPE;
Al Viro2903ff02012-08-28 12:52:22 -0400628 if (f.file->f_mode & FMODE_PREAD)
629 ret = vfs_read(f.file, buf, count, &pos);
630 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 }
632
633 return ret;
634}
635
Al Viro4a0fd5b2013-01-21 15:16:58 -0500636SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
637 size_t, count, loff_t, pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638{
Al Viro2903ff02012-08-28 12:52:22 -0400639 struct fd f;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 ssize_t ret = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641
642 if (pos < 0)
643 return -EINVAL;
644
Al Viro2903ff02012-08-28 12:52:22 -0400645 f = fdget(fd);
646 if (f.file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 ret = -ESPIPE;
Al Viro2903ff02012-08-28 12:52:22 -0400648 if (f.file->f_mode & FMODE_PWRITE)
649 ret = vfs_write(f.file, buf, count, &pos);
650 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 }
652
653 return ret;
654}
655
656/*
657 * Reduce an iovec's length in-place. Return the resulting number of segments
658 */
659unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
660{
661 unsigned long seg = 0;
662 size_t len = 0;
663
664 while (seg < nr_segs) {
665 seg++;
666 if (len + iov->iov_len >= to) {
667 iov->iov_len = to - len;
668 break;
669 }
670 len += iov->iov_len;
671 iov++;
672 }
673 return seg;
674}
Eric Sandeen19295522008-01-28 23:58:27 -0500675EXPORT_SYMBOL(iov_shorten);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676
Al Viroac15ac02015-03-20 20:10:21 -0400677static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
Christoph Hellwig793b80e2016-03-03 16:03:58 +0100678 loff_t *ppos, iter_fn_t fn, int flags)
Al Viro293bc982014-02-11 18:37:41 -0500679{
680 struct kiocb kiocb;
Al Viro293bc982014-02-11 18:37:41 -0500681 ssize_t ret;
682
Christoph Hellwige864f392016-04-07 08:52:03 -0700683 if (flags & ~(RWF_HIPRI | RWF_DSYNC | RWF_SYNC))
Christoph Hellwig793b80e2016-03-03 16:03:58 +0100684 return -EOPNOTSUPP;
685
Al Viro293bc982014-02-11 18:37:41 -0500686 init_sync_kiocb(&kiocb, filp);
Christoph Hellwig97be7eb2016-03-03 16:04:01 +0100687 if (flags & RWF_HIPRI)
688 kiocb.ki_flags |= IOCB_HIPRI;
Christoph Hellwige864f392016-04-07 08:52:03 -0700689 if (flags & RWF_DSYNC)
690 kiocb.ki_flags |= IOCB_DSYNC;
691 if (flags & RWF_SYNC)
692 kiocb.ki_flags |= (IOCB_DSYNC | IOCB_SYNC);
Al Viro293bc982014-02-11 18:37:41 -0500693 kiocb.ki_pos = *ppos;
Al Viro293bc982014-02-11 18:37:41 -0500694
Al Viroac15ac02015-03-20 20:10:21 -0400695 ret = fn(&kiocb, iter);
Christoph Hellwig599bd192015-02-11 19:59:44 +0100696 BUG_ON(ret == -EIOCBQUEUED);
Al Viro293bc982014-02-11 18:37:41 -0500697 *ppos = kiocb.ki_pos;
698 return ret;
699}
700
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700701/* Do it by hand, with file-ops */
Al Viroac15ac02015-03-20 20:10:21 -0400702static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
Christoph Hellwig793b80e2016-03-03 16:03:58 +0100703 loff_t *ppos, io_fn_t fn, int flags)
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700704{
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700705 ssize_t ret = 0;
706
Christoph Hellwig97be7eb2016-03-03 16:04:01 +0100707 if (flags & ~RWF_HIPRI)
Christoph Hellwig793b80e2016-03-03 16:03:58 +0100708 return -EOPNOTSUPP;
709
Al Viroac15ac02015-03-20 20:10:21 -0400710 while (iov_iter_count(iter)) {
711 struct iovec iovec = iov_iter_iovec(iter);
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700712 ssize_t nr;
713
Al Viroac15ac02015-03-20 20:10:21 -0400714 nr = fn(filp, iovec.iov_base, iovec.iov_len, ppos);
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700715
716 if (nr < 0) {
717 if (!ret)
718 ret = nr;
719 break;
720 }
721 ret += nr;
Al Viroac15ac02015-03-20 20:10:21 -0400722 if (nr != iovec.iov_len)
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700723 break;
Al Viroac15ac02015-03-20 20:10:21 -0400724 iov_iter_advance(iter, nr);
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700725 }
726
727 return ret;
728}
729
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730/* A write operation does a read from user space and vice versa */
731#define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
732
Vegard Nossumffecee42016-10-08 11:18:07 +0200733/**
734 * rw_copy_check_uvector() - Copy an array of &struct iovec from userspace
735 * into the kernel and check that it is valid.
736 *
737 * @type: One of %CHECK_IOVEC_ONLY, %READ, or %WRITE.
738 * @uvector: Pointer to the userspace array.
739 * @nr_segs: Number of elements in userspace array.
740 * @fast_segs: Number of elements in @fast_pointer.
741 * @fast_pointer: Pointer to (usually small on-stack) kernel array.
742 * @ret_pointer: (output parameter) Pointer to a variable that will point to
743 * either @fast_pointer, a newly allocated kernel array, or NULL,
744 * depending on which array was used.
745 *
746 * This function copies an array of &struct iovec of @nr_segs from
747 * userspace into the kernel and checks that each element is valid (e.g.
748 * it does not point to a kernel address or cause overflow by being too
749 * large, etc.).
750 *
751 * As an optimization, the caller may provide a pointer to a small
752 * on-stack array in @fast_pointer, typically %UIO_FASTIOV elements long
753 * (the size of this array, or 0 if unused, should be given in @fast_segs).
754 *
755 * @ret_pointer will always point to the array that was used, so the
756 * caller must take care not to call kfree() on it e.g. in case the
757 * @fast_pointer array was used and it was allocated on the stack.
758 *
759 * Return: The total number of bytes covered by the iovec array on success
760 * or a negative error code on error.
761 */
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700762ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
763 unsigned long nr_segs, unsigned long fast_segs,
764 struct iovec *fast_pointer,
Christopher Yeohac34ebb2012-05-31 16:26:42 -0700765 struct iovec **ret_pointer)
Linus Torvalds435f49a2010-10-29 10:36:49 -0700766{
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700767 unsigned long seg;
Linus Torvalds435f49a2010-10-29 10:36:49 -0700768 ssize_t ret;
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700769 struct iovec *iov = fast_pointer;
770
Linus Torvalds435f49a2010-10-29 10:36:49 -0700771 /*
772 * SuS says "The readv() function *may* fail if the iovcnt argument
773 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
774 * traditionally returned zero for zero segments, so...
775 */
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700776 if (nr_segs == 0) {
777 ret = 0;
Linus Torvalds435f49a2010-10-29 10:36:49 -0700778 goto out;
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700779 }
780
Linus Torvalds435f49a2010-10-29 10:36:49 -0700781 /*
782 * First get the "struct iovec" from user memory and
783 * verify all the pointers
784 */
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700785 if (nr_segs > UIO_MAXIOV) {
786 ret = -EINVAL;
Linus Torvalds435f49a2010-10-29 10:36:49 -0700787 goto out;
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700788 }
789 if (nr_segs > fast_segs) {
Linus Torvalds435f49a2010-10-29 10:36:49 -0700790 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700791 if (iov == NULL) {
792 ret = -ENOMEM;
Linus Torvalds435f49a2010-10-29 10:36:49 -0700793 goto out;
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700794 }
Linus Torvalds435f49a2010-10-29 10:36:49 -0700795 }
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700796 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
797 ret = -EFAULT;
Linus Torvalds435f49a2010-10-29 10:36:49 -0700798 goto out;
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700799 }
800
Linus Torvalds435f49a2010-10-29 10:36:49 -0700801 /*
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700802 * According to the Single Unix Specification we should return EINVAL
803 * if an element length is < 0 when cast to ssize_t or if the
804 * total length would overflow the ssize_t return value of the
805 * system call.
Linus Torvalds435f49a2010-10-29 10:36:49 -0700806 *
807 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
808 * overflow case.
809 */
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700810 ret = 0;
Linus Torvalds435f49a2010-10-29 10:36:49 -0700811 for (seg = 0; seg < nr_segs; seg++) {
812 void __user *buf = iov[seg].iov_base;
813 ssize_t len = (ssize_t)iov[seg].iov_len;
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700814
815 /* see if we we're about to use an invalid len or if
816 * it's about to overflow ssize_t */
Linus Torvalds435f49a2010-10-29 10:36:49 -0700817 if (len < 0) {
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700818 ret = -EINVAL;
Linus Torvalds435f49a2010-10-29 10:36:49 -0700819 goto out;
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700820 }
Christopher Yeohac34ebb2012-05-31 16:26:42 -0700821 if (type >= 0
Christopher Yeohfcf63402011-10-31 17:06:39 -0700822 && unlikely(!access_ok(vrfy_dir(type), buf, len))) {
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700823 ret = -EFAULT;
Linus Torvalds435f49a2010-10-29 10:36:49 -0700824 goto out;
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700825 }
Linus Torvalds435f49a2010-10-29 10:36:49 -0700826 if (len > MAX_RW_COUNT - ret) {
827 len = MAX_RW_COUNT - ret;
828 iov[seg].iov_len = len;
829 }
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700830 ret += len;
Linus Torvalds435f49a2010-10-29 10:36:49 -0700831 }
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700832out:
833 *ret_pointer = iov;
834 return ret;
835}
836
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837static ssize_t do_readv_writev(int type, struct file *file,
838 const struct iovec __user * uvector,
Christoph Hellwig793b80e2016-03-03 16:03:58 +0100839 unsigned long nr_segs, loff_t *pos,
840 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 size_t tot_len;
843 struct iovec iovstack[UIO_FASTIOV];
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700844 struct iovec *iov = iovstack;
Al Viroac15ac02015-03-20 20:10:21 -0400845 struct iov_iter iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 io_fn_t fn;
Al Viro293bc982014-02-11 18:37:41 -0500848 iter_fn_t iter_fn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849
Al Viro0504c072015-03-21 19:40:11 -0400850 ret = import_iovec(type, uvector, nr_segs,
851 ARRAY_SIZE(iovstack), &iov, &iter);
852 if (ret < 0)
853 return ret;
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700854
Al Viro0504c072015-03-21 19:40:11 -0400855 tot_len = iov_iter_count(&iter);
856 if (!tot_len)
857 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 ret = rw_verify_area(type, file, pos, tot_len);
Linus Torvaldse28cc712006-01-04 16:20:40 -0800859 if (ret < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 goto out;
861
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 if (type == READ) {
863 fn = file->f_op->read;
Al Viro293bc982014-02-11 18:37:41 -0500864 iter_fn = file->f_op->read_iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 } else {
866 fn = (io_fn_t)file->f_op->write;
Al Viro293bc982014-02-11 18:37:41 -0500867 iter_fn = file->f_op->write_iter;
Al Viro03d95eb2013-03-20 13:04:20 -0400868 file_start_write(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 }
870
Al Viro293bc982014-02-11 18:37:41 -0500871 if (iter_fn)
Christoph Hellwig793b80e2016-03-03 16:03:58 +0100872 ret = do_iter_readv_writev(file, &iter, pos, iter_fn, flags);
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700873 else
Christoph Hellwig793b80e2016-03-03 16:03:58 +0100874 ret = do_loop_readv_writev(file, &iter, pos, fn, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875
Al Viro03d95eb2013-03-20 13:04:20 -0400876 if (type != READ)
877 file_end_write(file);
878
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879out:
Al Viro0504c072015-03-21 19:40:11 -0400880 kfree(iov);
Robert Love0eeca282005-07-12 17:06:03 -0400881 if ((ret + (type == READ)) > 0) {
882 if (type == READ)
Eric Paris2a12a9d2009-12-17 21:24:21 -0500883 fsnotify_access(file);
Robert Love0eeca282005-07-12 17:06:03 -0400884 else
Eric Paris2a12a9d2009-12-17 21:24:21 -0500885 fsnotify_modify(file);
Robert Love0eeca282005-07-12 17:06:03 -0400886 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888}
889
890ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
Christoph Hellwig793b80e2016-03-03 16:03:58 +0100891 unsigned long vlen, loff_t *pos, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892{
893 if (!(file->f_mode & FMODE_READ))
894 return -EBADF;
Al Viro7f7f25e2014-02-11 17:49:24 -0500895 if (!(file->f_mode & FMODE_CAN_READ))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 return -EINVAL;
897
Christoph Hellwig793b80e2016-03-03 16:03:58 +0100898 return do_readv_writev(READ, file, vec, vlen, pos, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899}
900
901EXPORT_SYMBOL(vfs_readv);
902
903ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
Christoph Hellwig793b80e2016-03-03 16:03:58 +0100904 unsigned long vlen, loff_t *pos, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905{
906 if (!(file->f_mode & FMODE_WRITE))
907 return -EBADF;
Al Viro7f7f25e2014-02-11 17:49:24 -0500908 if (!(file->f_mode & FMODE_CAN_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 return -EINVAL;
910
Christoph Hellwig793b80e2016-03-03 16:03:58 +0100911 return do_readv_writev(WRITE, file, vec, vlen, pos, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912}
913
914EXPORT_SYMBOL(vfs_writev);
915
Milosz Tanskif17d8b32016-03-03 16:03:59 +0100916static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
917 unsigned long vlen, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918{
Linus Torvalds9c225f22014-03-03 09:36:58 -0800919 struct fd f = fdget_pos(fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 ssize_t ret = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921
Al Viro2903ff02012-08-28 12:52:22 -0400922 if (f.file) {
923 loff_t pos = file_pos_read(f.file);
Milosz Tanskif17d8b32016-03-03 16:03:59 +0100924 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
Al Viro5faf1532013-06-15 05:49:36 +0400925 if (ret >= 0)
926 file_pos_write(f.file, pos);
Linus Torvalds9c225f22014-03-03 09:36:58 -0800927 fdput_pos(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 }
929
930 if (ret > 0)
Alexey Dobriyan4b98d112007-02-10 01:46:45 -0800931 add_rchar(current, ret);
932 inc_syscr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 return ret;
934}
935
Milosz Tanskif17d8b32016-03-03 16:03:59 +0100936static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
937 unsigned long vlen, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938{
Linus Torvalds9c225f22014-03-03 09:36:58 -0800939 struct fd f = fdget_pos(fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 ssize_t ret = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941
Al Viro2903ff02012-08-28 12:52:22 -0400942 if (f.file) {
943 loff_t pos = file_pos_read(f.file);
Milosz Tanskif17d8b32016-03-03 16:03:59 +0100944 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
Al Viro5faf1532013-06-15 05:49:36 +0400945 if (ret >= 0)
946 file_pos_write(f.file, pos);
Linus Torvalds9c225f22014-03-03 09:36:58 -0800947 fdput_pos(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 }
949
950 if (ret > 0)
Alexey Dobriyan4b98d112007-02-10 01:46:45 -0800951 add_wchar(current, ret);
952 inc_syscw(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 return ret;
954}
955
Linus Torvalds601cc112009-04-03 08:03:22 -0700956static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
Gerd Hoffmannf3554f42009-04-02 16:59:23 -0700957{
Linus Torvalds601cc112009-04-03 08:03:22 -0700958#define HALF_LONG_BITS (BITS_PER_LONG / 2)
959 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
960}
961
Milosz Tanskif17d8b32016-03-03 16:03:59 +0100962static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
963 unsigned long vlen, loff_t pos, int flags)
Linus Torvalds601cc112009-04-03 08:03:22 -0700964{
Al Viro2903ff02012-08-28 12:52:22 -0400965 struct fd f;
Gerd Hoffmannf3554f42009-04-02 16:59:23 -0700966 ssize_t ret = -EBADF;
Gerd Hoffmannf3554f42009-04-02 16:59:23 -0700967
968 if (pos < 0)
969 return -EINVAL;
970
Al Viro2903ff02012-08-28 12:52:22 -0400971 f = fdget(fd);
972 if (f.file) {
Gerd Hoffmannf3554f42009-04-02 16:59:23 -0700973 ret = -ESPIPE;
Al Viro2903ff02012-08-28 12:52:22 -0400974 if (f.file->f_mode & FMODE_PREAD)
Milosz Tanskif17d8b32016-03-03 16:03:59 +0100975 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
Al Viro2903ff02012-08-28 12:52:22 -0400976 fdput(f);
Gerd Hoffmannf3554f42009-04-02 16:59:23 -0700977 }
978
979 if (ret > 0)
980 add_rchar(current, ret);
981 inc_syscr(current);
982 return ret;
983}
984
Milosz Tanskif17d8b32016-03-03 16:03:59 +0100985static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
986 unsigned long vlen, loff_t pos, int flags)
Gerd Hoffmannf3554f42009-04-02 16:59:23 -0700987{
Al Viro2903ff02012-08-28 12:52:22 -0400988 struct fd f;
Gerd Hoffmannf3554f42009-04-02 16:59:23 -0700989 ssize_t ret = -EBADF;
Gerd Hoffmannf3554f42009-04-02 16:59:23 -0700990
991 if (pos < 0)
992 return -EINVAL;
993
Al Viro2903ff02012-08-28 12:52:22 -0400994 f = fdget(fd);
995 if (f.file) {
Gerd Hoffmannf3554f42009-04-02 16:59:23 -0700996 ret = -ESPIPE;
Al Viro2903ff02012-08-28 12:52:22 -0400997 if (f.file->f_mode & FMODE_PWRITE)
Milosz Tanskif17d8b32016-03-03 16:03:59 +0100998 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
Al Viro2903ff02012-08-28 12:52:22 -0400999 fdput(f);
Gerd Hoffmannf3554f42009-04-02 16:59:23 -07001000 }
1001
1002 if (ret > 0)
1003 add_wchar(current, ret);
1004 inc_syscw(current);
1005 return ret;
1006}
1007
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001008SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
1009 unsigned long, vlen)
1010{
1011 return do_readv(fd, vec, vlen, 0);
1012}
1013
1014SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
1015 unsigned long, vlen)
1016{
1017 return do_writev(fd, vec, vlen, 0);
1018}
1019
1020SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
1021 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1022{
1023 loff_t pos = pos_from_hilo(pos_h, pos_l);
1024
1025 return do_preadv(fd, vec, vlen, pos, 0);
1026}
1027
1028SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1029 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1030 int, flags)
1031{
1032 loff_t pos = pos_from_hilo(pos_h, pos_l);
1033
1034 if (pos == -1)
1035 return do_readv(fd, vec, vlen, flags);
1036
1037 return do_preadv(fd, vec, vlen, pos, flags);
1038}
1039
1040SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1041 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1042{
1043 loff_t pos = pos_from_hilo(pos_h, pos_l);
1044
1045 return do_pwritev(fd, vec, vlen, pos, 0);
1046}
1047
1048SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1049 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1050 int, flags)
1051{
1052 loff_t pos = pos_from_hilo(pos_h, pos_l);
1053
1054 if (pos == -1)
1055 return do_writev(fd, vec, vlen, flags);
1056
1057 return do_pwritev(fd, vec, vlen, pos, flags);
1058}
1059
Al Viro72ec3512013-03-20 10:42:10 -04001060#ifdef CONFIG_COMPAT
1061
1062static ssize_t compat_do_readv_writev(int type, struct file *file,
1063 const struct compat_iovec __user *uvector,
Christoph Hellwig793b80e2016-03-03 16:03:58 +01001064 unsigned long nr_segs, loff_t *pos,
1065 int flags)
Al Viro72ec3512013-03-20 10:42:10 -04001066{
1067 compat_ssize_t tot_len;
1068 struct iovec iovstack[UIO_FASTIOV];
1069 struct iovec *iov = iovstack;
Al Viroac15ac02015-03-20 20:10:21 -04001070 struct iov_iter iter;
Al Viro72ec3512013-03-20 10:42:10 -04001071 ssize_t ret;
1072 io_fn_t fn;
Al Viro293bc982014-02-11 18:37:41 -05001073 iter_fn_t iter_fn;
Al Viro72ec3512013-03-20 10:42:10 -04001074
Al Viro0504c072015-03-21 19:40:11 -04001075 ret = compat_import_iovec(type, uvector, nr_segs,
1076 UIO_FASTIOV, &iov, &iter);
1077 if (ret < 0)
1078 return ret;
Al Viro72ec3512013-03-20 10:42:10 -04001079
Al Viro0504c072015-03-21 19:40:11 -04001080 tot_len = iov_iter_count(&iter);
1081 if (!tot_len)
1082 goto out;
Al Viro72ec3512013-03-20 10:42:10 -04001083 ret = rw_verify_area(type, file, pos, tot_len);
1084 if (ret < 0)
1085 goto out;
1086
Al Viro72ec3512013-03-20 10:42:10 -04001087 if (type == READ) {
1088 fn = file->f_op->read;
Al Viro293bc982014-02-11 18:37:41 -05001089 iter_fn = file->f_op->read_iter;
Al Viro72ec3512013-03-20 10:42:10 -04001090 } else {
1091 fn = (io_fn_t)file->f_op->write;
Al Viro293bc982014-02-11 18:37:41 -05001092 iter_fn = file->f_op->write_iter;
Al Viro03d95eb2013-03-20 13:04:20 -04001093 file_start_write(file);
Al Viro72ec3512013-03-20 10:42:10 -04001094 }
1095
Al Viro293bc982014-02-11 18:37:41 -05001096 if (iter_fn)
Christoph Hellwig793b80e2016-03-03 16:03:58 +01001097 ret = do_iter_readv_writev(file, &iter, pos, iter_fn, flags);
Al Viro03d95eb2013-03-20 13:04:20 -04001098 else
Christoph Hellwig793b80e2016-03-03 16:03:58 +01001099 ret = do_loop_readv_writev(file, &iter, pos, fn, flags);
Al Viro72ec3512013-03-20 10:42:10 -04001100
Al Viro03d95eb2013-03-20 13:04:20 -04001101 if (type != READ)
1102 file_end_write(file);
1103
Al Viro72ec3512013-03-20 10:42:10 -04001104out:
Al Viro0504c072015-03-21 19:40:11 -04001105 kfree(iov);
Al Viro72ec3512013-03-20 10:42:10 -04001106 if ((ret + (type == READ)) > 0) {
1107 if (type == READ)
1108 fsnotify_access(file);
1109 else
1110 fsnotify_modify(file);
1111 }
1112 return ret;
1113}
1114
1115static size_t compat_readv(struct file *file,
1116 const struct compat_iovec __user *vec,
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001117 unsigned long vlen, loff_t *pos, int flags)
Al Viro72ec3512013-03-20 10:42:10 -04001118{
1119 ssize_t ret = -EBADF;
1120
1121 if (!(file->f_mode & FMODE_READ))
1122 goto out;
1123
1124 ret = -EINVAL;
Al Viro7f7f25e2014-02-11 17:49:24 -05001125 if (!(file->f_mode & FMODE_CAN_READ))
Al Viro72ec3512013-03-20 10:42:10 -04001126 goto out;
1127
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001128 ret = compat_do_readv_writev(READ, file, vec, vlen, pos, flags);
Al Viro72ec3512013-03-20 10:42:10 -04001129
1130out:
1131 if (ret > 0)
1132 add_rchar(current, ret);
1133 inc_syscr(current);
1134 return ret;
1135}
1136
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001137static size_t do_compat_readv(compat_ulong_t fd,
1138 const struct compat_iovec __user *vec,
1139 compat_ulong_t vlen, int flags)
Al Viro72ec3512013-03-20 10:42:10 -04001140{
Linus Torvalds9c225f22014-03-03 09:36:58 -08001141 struct fd f = fdget_pos(fd);
Al Viro72ec3512013-03-20 10:42:10 -04001142 ssize_t ret;
1143 loff_t pos;
1144
1145 if (!f.file)
1146 return -EBADF;
1147 pos = f.file->f_pos;
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001148 ret = compat_readv(f.file, vec, vlen, &pos, flags);
Al Viro5faf1532013-06-15 05:49:36 +04001149 if (ret >= 0)
1150 f.file->f_pos = pos;
Linus Torvalds9c225f22014-03-03 09:36:58 -08001151 fdput_pos(f);
Al Viro72ec3512013-03-20 10:42:10 -04001152 return ret;
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001153
Al Viro72ec3512013-03-20 10:42:10 -04001154}
1155
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001156COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1157 const struct compat_iovec __user *,vec,
1158 compat_ulong_t, vlen)
1159{
1160 return do_compat_readv(fd, vec, vlen, 0);
1161}
1162
1163static long do_compat_preadv64(unsigned long fd,
Heiko Carstens378a10f2014-03-05 10:43:51 +01001164 const struct compat_iovec __user *vec,
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001165 unsigned long vlen, loff_t pos, int flags)
Al Viro72ec3512013-03-20 10:42:10 -04001166{
1167 struct fd f;
1168 ssize_t ret;
1169
1170 if (pos < 0)
1171 return -EINVAL;
1172 f = fdget(fd);
1173 if (!f.file)
1174 return -EBADF;
1175 ret = -ESPIPE;
1176 if (f.file->f_mode & FMODE_PREAD)
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001177 ret = compat_readv(f.file, vec, vlen, &pos, flags);
Al Viro72ec3512013-03-20 10:42:10 -04001178 fdput(f);
1179 return ret;
1180}
1181
Heiko Carstens378a10f2014-03-05 10:43:51 +01001182#ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1183COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1184 const struct compat_iovec __user *,vec,
1185 unsigned long, vlen, loff_t, pos)
1186{
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001187 return do_compat_preadv64(fd, vec, vlen, pos, 0);
Heiko Carstens378a10f2014-03-05 10:43:51 +01001188}
1189#endif
1190
Heiko Carstensdfd948e2014-01-29 14:05:44 -08001191COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
Al Viro72ec3512013-03-20 10:42:10 -04001192 const struct compat_iovec __user *,vec,
Heiko Carstensdfd948e2014-01-29 14:05:44 -08001193 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
Al Viro72ec3512013-03-20 10:42:10 -04001194{
1195 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
Heiko Carstens378a10f2014-03-05 10:43:51 +01001196
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001197 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1198}
1199
H.J. Lu3ebfd812016-07-14 12:31:53 -07001200#ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1201COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd,
1202 const struct compat_iovec __user *,vec,
1203 unsigned long, vlen, loff_t, pos, int, flags)
1204{
1205 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1206}
1207#endif
1208
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001209COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1210 const struct compat_iovec __user *,vec,
1211 compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
1212 int, flags)
1213{
1214 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1215
1216 if (pos == -1)
1217 return do_compat_readv(fd, vec, vlen, flags);
1218
1219 return do_compat_preadv64(fd, vec, vlen, pos, flags);
Al Viro72ec3512013-03-20 10:42:10 -04001220}
1221
1222static size_t compat_writev(struct file *file,
1223 const struct compat_iovec __user *vec,
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001224 unsigned long vlen, loff_t *pos, int flags)
Al Viro72ec3512013-03-20 10:42:10 -04001225{
1226 ssize_t ret = -EBADF;
1227
1228 if (!(file->f_mode & FMODE_WRITE))
1229 goto out;
1230
1231 ret = -EINVAL;
Al Viro7f7f25e2014-02-11 17:49:24 -05001232 if (!(file->f_mode & FMODE_CAN_WRITE))
Al Viro72ec3512013-03-20 10:42:10 -04001233 goto out;
1234
Christoph Hellwig793b80e2016-03-03 16:03:58 +01001235 ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos, 0);
Al Viro72ec3512013-03-20 10:42:10 -04001236
1237out:
1238 if (ret > 0)
1239 add_wchar(current, ret);
1240 inc_syscw(current);
1241 return ret;
1242}
1243
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001244static size_t do_compat_writev(compat_ulong_t fd,
1245 const struct compat_iovec __user* vec,
1246 compat_ulong_t vlen, int flags)
Al Viro72ec3512013-03-20 10:42:10 -04001247{
Linus Torvalds9c225f22014-03-03 09:36:58 -08001248 struct fd f = fdget_pos(fd);
Al Viro72ec3512013-03-20 10:42:10 -04001249 ssize_t ret;
1250 loff_t pos;
1251
1252 if (!f.file)
1253 return -EBADF;
1254 pos = f.file->f_pos;
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001255 ret = compat_writev(f.file, vec, vlen, &pos, flags);
Al Viro5faf1532013-06-15 05:49:36 +04001256 if (ret >= 0)
1257 f.file->f_pos = pos;
Linus Torvalds9c225f22014-03-03 09:36:58 -08001258 fdput_pos(f);
Al Viro72ec3512013-03-20 10:42:10 -04001259 return ret;
1260}
1261
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001262COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1263 const struct compat_iovec __user *, vec,
1264 compat_ulong_t, vlen)
1265{
1266 return do_compat_writev(fd, vec, vlen, 0);
1267}
1268
1269static long do_compat_pwritev64(unsigned long fd,
Heiko Carstens378a10f2014-03-05 10:43:51 +01001270 const struct compat_iovec __user *vec,
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001271 unsigned long vlen, loff_t pos, int flags)
Al Viro72ec3512013-03-20 10:42:10 -04001272{
1273 struct fd f;
1274 ssize_t ret;
1275
1276 if (pos < 0)
1277 return -EINVAL;
1278 f = fdget(fd);
1279 if (!f.file)
1280 return -EBADF;
1281 ret = -ESPIPE;
1282 if (f.file->f_mode & FMODE_PWRITE)
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001283 ret = compat_writev(f.file, vec, vlen, &pos, flags);
Al Viro72ec3512013-03-20 10:42:10 -04001284 fdput(f);
1285 return ret;
1286}
1287
Heiko Carstens378a10f2014-03-05 10:43:51 +01001288#ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1289COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1290 const struct compat_iovec __user *,vec,
1291 unsigned long, vlen, loff_t, pos)
1292{
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001293 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
Heiko Carstens378a10f2014-03-05 10:43:51 +01001294}
1295#endif
1296
Heiko Carstensdfd948e2014-01-29 14:05:44 -08001297COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
Al Viro72ec3512013-03-20 10:42:10 -04001298 const struct compat_iovec __user *,vec,
Heiko Carstensdfd948e2014-01-29 14:05:44 -08001299 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
Al Viro72ec3512013-03-20 10:42:10 -04001300{
1301 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
Heiko Carstens378a10f2014-03-05 10:43:51 +01001302
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001303 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
Al Viro72ec3512013-03-20 10:42:10 -04001304}
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001305
H.J. Lu3ebfd812016-07-14 12:31:53 -07001306#ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1307COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd,
1308 const struct compat_iovec __user *,vec,
1309 unsigned long, vlen, loff_t, pos, int, flags)
1310{
1311 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1312}
1313#endif
1314
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001315COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1316 const struct compat_iovec __user *,vec,
1317 compat_ulong_t, vlen, u32, pos_low, u32, pos_high, int, flags)
1318{
1319 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1320
1321 if (pos == -1)
1322 return do_compat_writev(fd, vec, vlen, flags);
1323
1324 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1325}
1326
Al Viro72ec3512013-03-20 10:42:10 -04001327#endif
1328
Al Viro19f4fc32013-02-24 02:17:03 -05001329static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1330 size_t count, loff_t max)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331{
Al Viro2903ff02012-08-28 12:52:22 -04001332 struct fd in, out;
1333 struct inode *in_inode, *out_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 loff_t pos;
Al Viro7995bd22013-06-20 18:58:36 +04001335 loff_t out_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 ssize_t retval;
Al Viro2903ff02012-08-28 12:52:22 -04001337 int fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338
1339 /*
1340 * Get input file, and verify that it is ok..
1341 */
1342 retval = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001343 in = fdget(in_fd);
1344 if (!in.file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 goto out;
Al Viro2903ff02012-08-28 12:52:22 -04001346 if (!(in.file->f_mode & FMODE_READ))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 goto fput_in;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 retval = -ESPIPE;
Al Viro7995bd22013-06-20 18:58:36 +04001349 if (!ppos) {
1350 pos = in.file->f_pos;
1351 } else {
1352 pos = *ppos;
Al Viro2903ff02012-08-28 12:52:22 -04001353 if (!(in.file->f_mode & FMODE_PREAD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 goto fput_in;
Al Viro7995bd22013-06-20 18:58:36 +04001355 }
1356 retval = rw_verify_area(READ, in.file, &pos, count);
Linus Torvaldse28cc712006-01-04 16:20:40 -08001357 if (retval < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 goto fput_in;
Al Virobc613842016-03-31 21:48:20 -04001359 if (count > MAX_RW_COUNT)
1360 count = MAX_RW_COUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 /*
1363 * Get output file, and verify that it is ok..
1364 */
1365 retval = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001366 out = fdget(out_fd);
1367 if (!out.file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 goto fput_in;
Al Viro2903ff02012-08-28 12:52:22 -04001369 if (!(out.file->f_mode & FMODE_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 goto fput_out;
1371 retval = -EINVAL;
Al Viro496ad9a2013-01-23 17:07:38 -05001372 in_inode = file_inode(in.file);
1373 out_inode = file_inode(out.file);
Al Viro7995bd22013-06-20 18:58:36 +04001374 out_pos = out.file->f_pos;
1375 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
Linus Torvaldse28cc712006-01-04 16:20:40 -08001376 if (retval < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 goto fput_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 if (!max)
1380 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1381
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 if (unlikely(pos + count > max)) {
1383 retval = -EOVERFLOW;
1384 if (pos >= max)
1385 goto fput_out;
1386 count = max - pos;
1387 }
1388
Jens Axboed96e6e72007-06-11 12:18:52 +02001389 fl = 0;
Jens Axboe534f2aa2007-06-01 14:52:37 +02001390#if 0
Jens Axboed96e6e72007-06-11 12:18:52 +02001391 /*
1392 * We need to debate whether we can enable this or not. The
1393 * man page documents EAGAIN return for the output at least,
1394 * and the application is arguably buggy if it doesn't expect
1395 * EAGAIN on a non-blocking file descriptor.
1396 */
Al Viro2903ff02012-08-28 12:52:22 -04001397 if (in.file->f_flags & O_NONBLOCK)
Jens Axboed96e6e72007-06-11 12:18:52 +02001398 fl = SPLICE_F_NONBLOCK;
Jens Axboe534f2aa2007-06-01 14:52:37 +02001399#endif
Al Viro50cd2c52013-05-23 20:10:34 -04001400 file_start_write(out.file);
Al Viro7995bd22013-06-20 18:58:36 +04001401 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
Al Viro50cd2c52013-05-23 20:10:34 -04001402 file_end_write(out.file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403
1404 if (retval > 0) {
Alexey Dobriyan4b98d112007-02-10 01:46:45 -08001405 add_rchar(current, retval);
1406 add_wchar(current, retval);
Scott Wolchoka68c2f12012-12-20 15:05:52 -08001407 fsnotify_access(in.file);
1408 fsnotify_modify(out.file);
Al Viro7995bd22013-06-20 18:58:36 +04001409 out.file->f_pos = out_pos;
1410 if (ppos)
1411 *ppos = pos;
1412 else
1413 in.file->f_pos = pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415
Alexey Dobriyan4b98d112007-02-10 01:46:45 -08001416 inc_syscr(current);
1417 inc_syscw(current);
Al Viro7995bd22013-06-20 18:58:36 +04001418 if (pos > max)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 retval = -EOVERFLOW;
1420
1421fput_out:
Al Viro2903ff02012-08-28 12:52:22 -04001422 fdput(out);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423fput_in:
Al Viro2903ff02012-08-28 12:52:22 -04001424 fdput(in);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425out:
1426 return retval;
1427}
1428
Heiko Carstens002c8972009-01-14 14:14:18 +01001429SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430{
1431 loff_t pos;
1432 off_t off;
1433 ssize_t ret;
1434
1435 if (offset) {
1436 if (unlikely(get_user(off, offset)))
1437 return -EFAULT;
1438 pos = off;
1439 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1440 if (unlikely(put_user(pos, offset)))
1441 return -EFAULT;
1442 return ret;
1443 }
1444
1445 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1446}
1447
Heiko Carstens002c8972009-01-14 14:14:18 +01001448SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449{
1450 loff_t pos;
1451 ssize_t ret;
1452
1453 if (offset) {
1454 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1455 return -EFAULT;
1456 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1457 if (unlikely(put_user(pos, offset)))
1458 return -EFAULT;
1459 return ret;
1460 }
1461
1462 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1463}
Al Viro19f4fc32013-02-24 02:17:03 -05001464
1465#ifdef CONFIG_COMPAT
1466COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1467 compat_off_t __user *, offset, compat_size_t, count)
1468{
1469 loff_t pos;
1470 off_t off;
1471 ssize_t ret;
1472
1473 if (offset) {
1474 if (unlikely(get_user(off, offset)))
1475 return -EFAULT;
1476 pos = off;
1477 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1478 if (unlikely(put_user(pos, offset)))
1479 return -EFAULT;
1480 return ret;
1481 }
1482
1483 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1484}
1485
1486COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1487 compat_loff_t __user *, offset, compat_size_t, count)
1488{
1489 loff_t pos;
1490 ssize_t ret;
1491
1492 if (offset) {
1493 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1494 return -EFAULT;
1495 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1496 if (unlikely(put_user(pos, offset)))
1497 return -EFAULT;
1498 return ret;
1499 }
1500
1501 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1502}
1503#endif
Zach Brown29732932015-11-10 16:53:30 -05001504
1505/*
1506 * copy_file_range() differs from regular file read and write in that it
1507 * specifically allows return partial success. When it does so is up to
1508 * the copy_file_range method.
1509 */
1510ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1511 struct file *file_out, loff_t pos_out,
1512 size_t len, unsigned int flags)
1513{
1514 struct inode *inode_in = file_inode(file_in);
1515 struct inode *inode_out = file_inode(file_out);
1516 ssize_t ret;
1517
1518 if (flags != 0)
1519 return -EINVAL;
1520
Zach Brown29732932015-11-10 16:53:30 -05001521 ret = rw_verify_area(READ, file_in, &pos_in, len);
Al Virobc613842016-03-31 21:48:20 -04001522 if (unlikely(ret))
1523 return ret;
1524
1525 ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1526 if (unlikely(ret))
Zach Brown29732932015-11-10 16:53:30 -05001527 return ret;
1528
1529 if (!(file_in->f_mode & FMODE_READ) ||
1530 !(file_out->f_mode & FMODE_WRITE) ||
Anna Schumakereac70052015-11-10 16:53:33 -05001531 (file_out->f_flags & O_APPEND))
Zach Brown29732932015-11-10 16:53:30 -05001532 return -EBADF;
1533
1534 /* this could be relaxed once a method supports cross-fs copies */
1535 if (inode_in->i_sb != inode_out->i_sb)
1536 return -EXDEV;
1537
1538 if (len == 0)
1539 return 0;
1540
Miklos Szeredi36161192016-12-16 11:02:54 +01001541 sb_start_write(inode_out->i_sb);
Zach Brown29732932015-11-10 16:53:30 -05001542
Christoph Hellwiga76b5b02016-12-09 16:17:19 -08001543 /*
1544 * Try cloning first, this is supported by more file systems, and
1545 * more efficient if both clone and copy are supported (e.g. NFS).
1546 */
1547 if (file_in->f_op->clone_file_range) {
1548 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1549 file_out, pos_out, len);
1550 if (ret == 0) {
1551 ret = len;
1552 goto done;
1553 }
1554 }
1555
1556 if (file_out->f_op->copy_file_range) {
Anna Schumakereac70052015-11-10 16:53:33 -05001557 ret = file_out->f_op->copy_file_range(file_in, pos_in, file_out,
1558 pos_out, len, flags);
Christoph Hellwiga76b5b02016-12-09 16:17:19 -08001559 if (ret != -EOPNOTSUPP)
1560 goto done;
1561 }
Anna Schumakereac70052015-11-10 16:53:33 -05001562
Christoph Hellwiga76b5b02016-12-09 16:17:19 -08001563 ret = do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1564 len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
1565
1566done:
Zach Brown29732932015-11-10 16:53:30 -05001567 if (ret > 0) {
1568 fsnotify_access(file_in);
1569 add_rchar(current, ret);
1570 fsnotify_modify(file_out);
1571 add_wchar(current, ret);
1572 }
Christoph Hellwiga76b5b02016-12-09 16:17:19 -08001573
Zach Brown29732932015-11-10 16:53:30 -05001574 inc_syscr(current);
1575 inc_syscw(current);
1576
Miklos Szeredi36161192016-12-16 11:02:54 +01001577 sb_end_write(inode_out->i_sb);
Zach Brown29732932015-11-10 16:53:30 -05001578
1579 return ret;
1580}
1581EXPORT_SYMBOL(vfs_copy_file_range);
1582
1583SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1584 int, fd_out, loff_t __user *, off_out,
1585 size_t, len, unsigned int, flags)
1586{
1587 loff_t pos_in;
1588 loff_t pos_out;
1589 struct fd f_in;
1590 struct fd f_out;
1591 ssize_t ret = -EBADF;
1592
1593 f_in = fdget(fd_in);
1594 if (!f_in.file)
1595 goto out2;
1596
1597 f_out = fdget(fd_out);
1598 if (!f_out.file)
1599 goto out1;
1600
1601 ret = -EFAULT;
1602 if (off_in) {
1603 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1604 goto out;
1605 } else {
1606 pos_in = f_in.file->f_pos;
1607 }
1608
1609 if (off_out) {
1610 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1611 goto out;
1612 } else {
1613 pos_out = f_out.file->f_pos;
1614 }
1615
1616 ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1617 flags);
1618 if (ret > 0) {
1619 pos_in += ret;
1620 pos_out += ret;
1621
1622 if (off_in) {
1623 if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1624 ret = -EFAULT;
1625 } else {
1626 f_in.file->f_pos = pos_in;
1627 }
1628
1629 if (off_out) {
1630 if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1631 ret = -EFAULT;
1632 } else {
1633 f_out.file->f_pos = pos_out;
1634 }
1635 }
1636
1637out:
1638 fdput(f_out);
1639out1:
1640 fdput(f_in);
1641out2:
1642 return ret;
1643}
Christoph Hellwig04b38d62015-12-03 12:59:50 +01001644
1645static int clone_verify_area(struct file *file, loff_t pos, u64 len, bool write)
1646{
1647 struct inode *inode = file_inode(file);
1648
1649 if (unlikely(pos < 0))
1650 return -EINVAL;
1651
1652 if (unlikely((loff_t) (pos + len) < 0))
1653 return -EINVAL;
1654
1655 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
1656 loff_t end = len ? pos + len - 1 : OFFSET_MAX;
1657 int retval;
1658
1659 retval = locks_mandatory_area(inode, file, pos, end,
1660 write ? F_WRLCK : F_RDLCK);
1661 if (retval < 0)
1662 return retval;
1663 }
1664
1665 return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
1666}
1667
Darrick J. Wong876bec6f2016-12-09 16:18:30 -08001668/*
1669 * Check that the two inodes are eligible for cloning, the ranges make
1670 * sense, and then flush all dirty data. Caller must ensure that the
1671 * inodes have been locked against any other modifications.
Darrick J. Wong22725ce2016-12-19 15:13:26 -08001672 *
1673 * Returns: 0 for "nothing to clone", 1 for "something to clone", or
1674 * the usual negative error code.
Darrick J. Wong876bec6f2016-12-09 16:18:30 -08001675 */
1676int vfs_clone_file_prep_inodes(struct inode *inode_in, loff_t pos_in,
1677 struct inode *inode_out, loff_t pos_out,
1678 u64 *len, bool is_dedupe)
1679{
1680 loff_t bs = inode_out->i_sb->s_blocksize;
1681 loff_t blen;
1682 loff_t isize;
1683 bool same_inode = (inode_in == inode_out);
1684 int ret;
1685
1686 /* Don't touch certain kinds of inodes */
1687 if (IS_IMMUTABLE(inode_out))
1688 return -EPERM;
1689
1690 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
1691 return -ETXTBSY;
1692
1693 /* Don't reflink dirs, pipes, sockets... */
1694 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1695 return -EISDIR;
1696 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1697 return -EINVAL;
1698
1699 /* Are we going all the way to the end? */
1700 isize = i_size_read(inode_in);
Darrick J. Wong22725ce2016-12-19 15:13:26 -08001701 if (isize == 0)
Darrick J. Wong876bec6f2016-12-09 16:18:30 -08001702 return 0;
Darrick J. Wong876bec6f2016-12-09 16:18:30 -08001703
1704 /* Zero length dedupe exits immediately; reflink goes to EOF. */
1705 if (*len == 0) {
Darrick J. Wong22725ce2016-12-19 15:13:26 -08001706 if (is_dedupe || pos_in == isize)
Darrick J. Wong876bec6f2016-12-09 16:18:30 -08001707 return 0;
Darrick J. Wong22725ce2016-12-19 15:13:26 -08001708 if (pos_in > isize)
1709 return -EINVAL;
Darrick J. Wong876bec6f2016-12-09 16:18:30 -08001710 *len = isize - pos_in;
1711 }
1712
1713 /* Ensure offsets don't wrap and the input is inside i_size */
1714 if (pos_in + *len < pos_in || pos_out + *len < pos_out ||
1715 pos_in + *len > isize)
1716 return -EINVAL;
1717
1718 /* Don't allow dedupe past EOF in the dest file */
1719 if (is_dedupe) {
1720 loff_t disize;
1721
1722 disize = i_size_read(inode_out);
1723 if (pos_out >= disize || pos_out + *len > disize)
1724 return -EINVAL;
1725 }
1726
1727 /* If we're linking to EOF, continue to the block boundary. */
1728 if (pos_in + *len == isize)
1729 blen = ALIGN(isize, bs) - pos_in;
1730 else
1731 blen = *len;
1732
1733 /* Only reflink if we're aligned to block boundaries */
1734 if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_in + blen, bs) ||
1735 !IS_ALIGNED(pos_out, bs) || !IS_ALIGNED(pos_out + blen, bs))
1736 return -EINVAL;
1737
1738 /* Don't allow overlapped reflink within the same file */
1739 if (same_inode) {
1740 if (pos_out + blen > pos_in && pos_out < pos_in + blen)
1741 return -EINVAL;
1742 }
1743
1744 /* Wait for the completion of any pending IOs on both files */
1745 inode_dio_wait(inode_in);
1746 if (!same_inode)
1747 inode_dio_wait(inode_out);
1748
1749 ret = filemap_write_and_wait_range(inode_in->i_mapping,
1750 pos_in, pos_in + *len - 1);
1751 if (ret)
1752 return ret;
1753
1754 ret = filemap_write_and_wait_range(inode_out->i_mapping,
1755 pos_out, pos_out + *len - 1);
1756 if (ret)
1757 return ret;
1758
1759 /*
1760 * Check that the extents are the same.
1761 */
1762 if (is_dedupe) {
1763 bool is_same = false;
1764
1765 ret = vfs_dedupe_file_range_compare(inode_in, pos_in,
1766 inode_out, pos_out, *len, &is_same);
1767 if (ret)
1768 return ret;
1769 if (!is_same)
1770 return -EBADE;
1771 }
1772
Darrick J. Wong22725ce2016-12-19 15:13:26 -08001773 return 1;
Darrick J. Wong876bec6f2016-12-09 16:18:30 -08001774}
1775EXPORT_SYMBOL(vfs_clone_file_prep_inodes);
1776
Christoph Hellwig04b38d62015-12-03 12:59:50 +01001777int vfs_clone_file_range(struct file *file_in, loff_t pos_in,
1778 struct file *file_out, loff_t pos_out, u64 len)
1779{
1780 struct inode *inode_in = file_inode(file_in);
1781 struct inode *inode_out = file_inode(file_out);
1782 int ret;
1783
Amir Goldsteinb335e9d2016-10-26 22:34:01 +03001784 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1785 return -EISDIR;
1786 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1787 return -EINVAL;
1788
Amir Goldstein913b86e2016-09-23 11:38:10 +03001789 /*
1790 * FICLONE/FICLONERANGE ioctls enforce that src and dest files are on
1791 * the same mount. Practically, they only need to be on the same file
1792 * system.
1793 */
1794 if (inode_in->i_sb != inode_out->i_sb)
Christoph Hellwig04b38d62015-12-03 12:59:50 +01001795 return -EXDEV;
1796
Christoph Hellwig04b38d62015-12-03 12:59:50 +01001797 if (!(file_in->f_mode & FMODE_READ) ||
1798 !(file_out->f_mode & FMODE_WRITE) ||
Christoph Hellwig0fcbf992016-02-26 18:53:12 +01001799 (file_out->f_flags & O_APPEND))
Christoph Hellwig04b38d62015-12-03 12:59:50 +01001800 return -EBADF;
1801
Christoph Hellwig0fcbf992016-02-26 18:53:12 +01001802 if (!file_in->f_op->clone_file_range)
1803 return -EOPNOTSUPP;
1804
Christoph Hellwig04b38d62015-12-03 12:59:50 +01001805 ret = clone_verify_area(file_in, pos_in, len, false);
1806 if (ret)
1807 return ret;
1808
1809 ret = clone_verify_area(file_out, pos_out, len, true);
1810 if (ret)
1811 return ret;
1812
1813 if (pos_in + len > i_size_read(inode_in))
1814 return -EINVAL;
1815
Christoph Hellwig04b38d62015-12-03 12:59:50 +01001816 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1817 file_out, pos_out, len);
1818 if (!ret) {
1819 fsnotify_access(file_in);
1820 fsnotify_modify(file_out);
1821 }
1822
Christoph Hellwig04b38d62015-12-03 12:59:50 +01001823 return ret;
1824}
1825EXPORT_SYMBOL(vfs_clone_file_range);
Darrick J. Wong54dbc152015-12-19 00:55:59 -08001826
Darrick J. Wong876bec6f2016-12-09 16:18:30 -08001827/*
1828 * Read a page's worth of file data into the page cache. Return the page
1829 * locked.
1830 */
1831static struct page *vfs_dedupe_get_page(struct inode *inode, loff_t offset)
1832{
1833 struct address_space *mapping;
1834 struct page *page;
1835 pgoff_t n;
1836
1837 n = offset >> PAGE_SHIFT;
1838 mapping = inode->i_mapping;
1839 page = read_mapping_page(mapping, n, NULL);
1840 if (IS_ERR(page))
1841 return page;
1842 if (!PageUptodate(page)) {
1843 put_page(page);
1844 return ERR_PTR(-EIO);
1845 }
1846 lock_page(page);
1847 return page;
1848}
1849
1850/*
1851 * Compare extents of two files to see if they are the same.
1852 * Caller must have locked both inodes to prevent write races.
1853 */
1854int vfs_dedupe_file_range_compare(struct inode *src, loff_t srcoff,
1855 struct inode *dest, loff_t destoff,
1856 loff_t len, bool *is_same)
1857{
1858 loff_t src_poff;
1859 loff_t dest_poff;
1860 void *src_addr;
1861 void *dest_addr;
1862 struct page *src_page;
1863 struct page *dest_page;
1864 loff_t cmp_len;
1865 bool same;
1866 int error;
1867
1868 error = -EINVAL;
1869 same = true;
1870 while (len) {
1871 src_poff = srcoff & (PAGE_SIZE - 1);
1872 dest_poff = destoff & (PAGE_SIZE - 1);
1873 cmp_len = min(PAGE_SIZE - src_poff,
1874 PAGE_SIZE - dest_poff);
1875 cmp_len = min(cmp_len, len);
1876 if (cmp_len <= 0)
1877 goto out_error;
1878
1879 src_page = vfs_dedupe_get_page(src, srcoff);
1880 if (IS_ERR(src_page)) {
1881 error = PTR_ERR(src_page);
1882 goto out_error;
1883 }
1884 dest_page = vfs_dedupe_get_page(dest, destoff);
1885 if (IS_ERR(dest_page)) {
1886 error = PTR_ERR(dest_page);
1887 unlock_page(src_page);
1888 put_page(src_page);
1889 goto out_error;
1890 }
1891 src_addr = kmap_atomic(src_page);
1892 dest_addr = kmap_atomic(dest_page);
1893
1894 flush_dcache_page(src_page);
1895 flush_dcache_page(dest_page);
1896
1897 if (memcmp(src_addr + src_poff, dest_addr + dest_poff, cmp_len))
1898 same = false;
1899
1900 kunmap_atomic(dest_addr);
1901 kunmap_atomic(src_addr);
1902 unlock_page(dest_page);
1903 unlock_page(src_page);
1904 put_page(dest_page);
1905 put_page(src_page);
1906
1907 if (!same)
1908 break;
1909
1910 srcoff += cmp_len;
1911 destoff += cmp_len;
1912 len -= cmp_len;
1913 }
1914
1915 *is_same = same;
1916 return 0;
1917
1918out_error:
1919 return error;
1920}
1921EXPORT_SYMBOL(vfs_dedupe_file_range_compare);
1922
Darrick J. Wong54dbc152015-12-19 00:55:59 -08001923int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
1924{
1925 struct file_dedupe_range_info *info;
1926 struct inode *src = file_inode(file);
1927 u64 off;
1928 u64 len;
1929 int i;
1930 int ret;
1931 bool is_admin = capable(CAP_SYS_ADMIN);
1932 u16 count = same->dest_count;
1933 struct file *dst_file;
1934 loff_t dst_off;
1935 ssize_t deduped;
1936
1937 if (!(file->f_mode & FMODE_READ))
1938 return -EINVAL;
1939
1940 if (same->reserved1 || same->reserved2)
1941 return -EINVAL;
1942
1943 off = same->src_offset;
1944 len = same->src_length;
1945
1946 ret = -EISDIR;
1947 if (S_ISDIR(src->i_mode))
1948 goto out;
1949
1950 ret = -EINVAL;
1951 if (!S_ISREG(src->i_mode))
1952 goto out;
1953
1954 ret = clone_verify_area(file, off, len, false);
1955 if (ret < 0)
1956 goto out;
1957 ret = 0;
1958
Darrick J. Wong22725ce2016-12-19 15:13:26 -08001959 if (off + len > i_size_read(src))
1960 return -EINVAL;
1961
Darrick J. Wong54dbc152015-12-19 00:55:59 -08001962 /* pre-format output fields to sane values */
1963 for (i = 0; i < count; i++) {
1964 same->info[i].bytes_deduped = 0ULL;
1965 same->info[i].status = FILE_DEDUPE_RANGE_SAME;
1966 }
1967
1968 for (i = 0, info = same->info; i < count; i++, info++) {
1969 struct inode *dst;
1970 struct fd dst_fd = fdget(info->dest_fd);
1971
1972 dst_file = dst_fd.file;
1973 if (!dst_file) {
1974 info->status = -EBADF;
1975 goto next_loop;
1976 }
1977 dst = file_inode(dst_file);
1978
1979 ret = mnt_want_write_file(dst_file);
1980 if (ret) {
1981 info->status = ret;
1982 goto next_loop;
1983 }
1984
1985 dst_off = info->dest_offset;
1986 ret = clone_verify_area(dst_file, dst_off, len, true);
1987 if (ret < 0) {
1988 info->status = ret;
1989 goto next_file;
1990 }
1991 ret = 0;
1992
1993 if (info->reserved) {
1994 info->status = -EINVAL;
1995 } else if (!(is_admin || (dst_file->f_mode & FMODE_WRITE))) {
1996 info->status = -EINVAL;
1997 } else if (file->f_path.mnt != dst_file->f_path.mnt) {
1998 info->status = -EXDEV;
1999 } else if (S_ISDIR(dst->i_mode)) {
2000 info->status = -EISDIR;
2001 } else if (dst_file->f_op->dedupe_file_range == NULL) {
2002 info->status = -EINVAL;
2003 } else {
2004 deduped = dst_file->f_op->dedupe_file_range(file, off,
2005 len, dst_file,
2006 info->dest_offset);
2007 if (deduped == -EBADE)
2008 info->status = FILE_DEDUPE_RANGE_DIFFERS;
2009 else if (deduped < 0)
2010 info->status = deduped;
2011 else
2012 info->bytes_deduped += deduped;
2013 }
2014
2015next_file:
2016 mnt_drop_write_file(dst_file);
2017next_loop:
2018 fdput(dst_fd);
Darrick J. Wonge62e5602016-01-22 16:58:28 -08002019
2020 if (fatal_signal_pending(current))
2021 goto out;
Darrick J. Wong54dbc152015-12-19 00:55:59 -08002022 }
2023
2024out:
2025 return ret;
2026}
2027EXPORT_SYMBOL(vfs_dedupe_file_range);