blob: e8136a72c13f31814b22fec3836011d9f0c93243 [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001// SPDX-License-Identifier: GPL-2.0
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
3 * linux/fs/read_write.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
Ingo Molnarb12fb7f2017-02-08 18:51:33 +01008#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -07009#include <linux/stat.h>
Ingo Molnarb12fb7f2017-02-08 18:51:33 +010010#include <linux/sched/xacct.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#include <linux/fcntl.h>
12#include <linux/file.h>
13#include <linux/uio.h>
Robert Love0eeca282005-07-12 17:06:03 -040014#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/security.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050016#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/syscalls.h>
Linus Torvaldse28cc712006-01-04 16:20:40 -080018#include <linux/pagemap.h>
Jens Axboed6b29d72007-06-04 09:59:47 +020019#include <linux/splice.h>
Al Viro561c6732013-02-24 10:52:26 -050020#include <linux/compat.h>
Zach Brown29732932015-11-10 16:53:30 -050021#include <linux/mount.h>
Wouter van Kesteren2feb55f2016-02-16 22:20:59 +010022#include <linux/fs.h>
Al Viro06ae43f2013-03-20 13:19:30 -040023#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080025#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <asm/unistd.h>
27
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -080028const struct file_operations generic_ro_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070029 .llseek = generic_file_llseek,
Al Viroaad4f8b2014-04-02 14:33:16 -040030 .read_iter = generic_file_read_iter,
Linus Torvalds1da177e2005-04-16 15:20:36 -070031 .mmap = generic_file_readonly_mmap,
Jens Axboe534f2aa2007-06-01 14:52:37 +020032 .splice_read = generic_file_splice_read,
Linus Torvalds1da177e2005-04-16 15:20:36 -070033};
34
35EXPORT_SYMBOL(generic_ro_fops);
36
Christoph Hellwigddef7ed2017-07-06 18:58:37 +020037static inline bool unsigned_offsets(struct file *file)
KAMEZAWA Hiroyuki4a3956c2010-10-01 14:20:22 -070038{
Al Virocccb5a12010-12-17 07:44:05 -050039 return file->f_mode & FMODE_UNSIGNED_OFFSET;
KAMEZAWA Hiroyuki4a3956c2010-10-01 14:20:22 -070040}
41
Jie Liu46a1c2c2013-06-25 12:02:13 +080042/**
43 * vfs_setpos - update the file offset for lseek
44 * @file: file structure in question
45 * @offset: file offset to seek to
46 * @maxsize: maximum file size
47 *
48 * This is a low-level filesystem helper for updating the file offset to
49 * the value specified by @offset if the given offset is valid and it is
50 * not equal to the current file offset.
51 *
52 * Return the specified offset on success and -EINVAL on invalid offset.
53 */
54loff_t vfs_setpos(struct file *file, loff_t offset, loff_t maxsize)
Andi Kleenef3d0fd2011-09-15 16:06:48 -070055{
56 if (offset < 0 && !unsigned_offsets(file))
57 return -EINVAL;
58 if (offset > maxsize)
59 return -EINVAL;
60
61 if (offset != file->f_pos) {
62 file->f_pos = offset;
63 file->f_version = 0;
64 }
65 return offset;
66}
Jie Liu46a1c2c2013-06-25 12:02:13 +080067EXPORT_SYMBOL(vfs_setpos);
Andi Kleenef3d0fd2011-09-15 16:06:48 -070068
Christoph Hellwig3a8cff42008-08-11 15:37:17 +020069/**
Andi Kleen57604952011-09-15 16:06:50 -070070 * generic_file_llseek_size - generic llseek implementation for regular files
Christoph Hellwig3a8cff42008-08-11 15:37:17 +020071 * @file: file structure to seek on
72 * @offset: file offset to seek to
Andrew Morton965c8e52012-12-17 15:59:39 -080073 * @whence: type of seek
Eric Sandeene8b96eb2012-04-30 13:11:29 -050074 * @size: max size of this file in file system
75 * @eof: offset used for SEEK_END position
Christoph Hellwig3a8cff42008-08-11 15:37:17 +020076 *
Andi Kleen57604952011-09-15 16:06:50 -070077 * This is a variant of generic_file_llseek that allows passing in a custom
Eric Sandeene8b96eb2012-04-30 13:11:29 -050078 * maximum file size and a custom EOF position, for e.g. hashed directories
Andi Kleenef3d0fd2011-09-15 16:06:48 -070079 *
80 * Synchronization:
Andi Kleen57604952011-09-15 16:06:50 -070081 * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
Andi Kleenef3d0fd2011-09-15 16:06:48 -070082 * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
83 * read/writes behave like SEEK_SET against seeks.
Christoph Hellwig3a8cff42008-08-11 15:37:17 +020084 */
Andi Kleen9465efc2008-06-27 11:05:24 +020085loff_t
Andrew Morton965c8e52012-12-17 15:59:39 -080086generic_file_llseek_size(struct file *file, loff_t offset, int whence,
Eric Sandeene8b96eb2012-04-30 13:11:29 -050087 loff_t maxsize, loff_t eof)
Linus Torvalds1da177e2005-04-16 15:20:36 -070088{
Andrew Morton965c8e52012-12-17 15:59:39 -080089 switch (whence) {
Christoph Hellwig3a8cff42008-08-11 15:37:17 +020090 case SEEK_END:
Eric Sandeene8b96eb2012-04-30 13:11:29 -050091 offset += eof;
Christoph Hellwig3a8cff42008-08-11 15:37:17 +020092 break;
93 case SEEK_CUR:
Alain Knaff5b6f1eb2008-11-10 17:08:08 -080094 /*
95 * Here we special-case the lseek(fd, 0, SEEK_CUR)
96 * position-querying operation. Avoid rewriting the "same"
97 * f_pos value back to the file because a concurrent read(),
98 * write() or lseek() might have altered it
99 */
100 if (offset == 0)
101 return file->f_pos;
Andi Kleenef3d0fd2011-09-15 16:06:48 -0700102 /*
103 * f_lock protects against read/modify/write race with other
104 * SEEK_CURs. Note that parallel writes and reads behave
105 * like SEEK_SET.
106 */
107 spin_lock(&file->f_lock);
Jie Liu46a1c2c2013-06-25 12:02:13 +0800108 offset = vfs_setpos(file, file->f_pos + offset, maxsize);
Andi Kleenef3d0fd2011-09-15 16:06:48 -0700109 spin_unlock(&file->f_lock);
110 return offset;
Josef Bacik982d8162011-07-18 13:21:35 -0400111 case SEEK_DATA:
112 /*
113 * In the generic case the entire file is data, so as long as
114 * offset isn't at the end of the file then the offset is data.
115 */
Andreas Gruenbacherfc468202017-09-25 12:23:03 +0200116 if ((unsigned long long)offset >= eof)
Josef Bacik982d8162011-07-18 13:21:35 -0400117 return -ENXIO;
118 break;
119 case SEEK_HOLE:
120 /*
121 * There is a virtual hole at the end of the file, so as long as
122 * offset isn't i_size or larger, return i_size.
123 */
Andreas Gruenbacherfc468202017-09-25 12:23:03 +0200124 if ((unsigned long long)offset >= eof)
Josef Bacik982d8162011-07-18 13:21:35 -0400125 return -ENXIO;
Eric Sandeene8b96eb2012-04-30 13:11:29 -0500126 offset = eof;
Josef Bacik982d8162011-07-18 13:21:35 -0400127 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 }
Christoph Hellwig3a8cff42008-08-11 15:37:17 +0200129
Jie Liu46a1c2c2013-06-25 12:02:13 +0800130 return vfs_setpos(file, offset, maxsize);
Andi Kleen57604952011-09-15 16:06:50 -0700131}
132EXPORT_SYMBOL(generic_file_llseek_size);
133
134/**
135 * generic_file_llseek - generic llseek implementation for regular files
136 * @file: file structure to seek on
137 * @offset: file offset to seek to
Andrew Morton965c8e52012-12-17 15:59:39 -0800138 * @whence: type of seek
Andi Kleen57604952011-09-15 16:06:50 -0700139 *
140 * This is a generic implemenation of ->llseek useable for all normal local
141 * filesystems. It just updates the file offset to the value specified by
Ming Lei546ae2d2013-04-29 15:06:07 -0700142 * @offset and @whence.
Andi Kleen57604952011-09-15 16:06:50 -0700143 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800144loff_t generic_file_llseek(struct file *file, loff_t offset, int whence)
Andi Kleen57604952011-09-15 16:06:50 -0700145{
146 struct inode *inode = file->f_mapping->host;
147
Andrew Morton965c8e52012-12-17 15:59:39 -0800148 return generic_file_llseek_size(file, offset, whence,
Eric Sandeene8b96eb2012-04-30 13:11:29 -0500149 inode->i_sb->s_maxbytes,
150 i_size_read(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151}
Andi Kleen9465efc2008-06-27 11:05:24 +0200152EXPORT_SYMBOL(generic_file_llseek);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153
jan Blunckae6afc32010-05-26 14:44:48 -0700154/**
Al Viro1bf9d142013-06-16 20:27:42 +0400155 * fixed_size_llseek - llseek implementation for fixed-sized devices
156 * @file: file structure to seek on
157 * @offset: file offset to seek to
158 * @whence: type of seek
159 * @size: size of the file
160 *
161 */
162loff_t fixed_size_llseek(struct file *file, loff_t offset, int whence, loff_t size)
163{
164 switch (whence) {
165 case SEEK_SET: case SEEK_CUR: case SEEK_END:
166 return generic_file_llseek_size(file, offset, whence,
167 size, size);
168 default:
169 return -EINVAL;
170 }
171}
172EXPORT_SYMBOL(fixed_size_llseek);
173
174/**
Al Virob25472f2015-12-05 22:04:48 -0500175 * no_seek_end_llseek - llseek implementation for fixed-sized devices
176 * @file: file structure to seek on
177 * @offset: file offset to seek to
178 * @whence: type of seek
179 *
180 */
181loff_t no_seek_end_llseek(struct file *file, loff_t offset, int whence)
182{
183 switch (whence) {
184 case SEEK_SET: case SEEK_CUR:
185 return generic_file_llseek_size(file, offset, whence,
Wouter van Kesteren2feb55f2016-02-16 22:20:59 +0100186 OFFSET_MAX, 0);
Al Virob25472f2015-12-05 22:04:48 -0500187 default:
188 return -EINVAL;
189 }
190}
191EXPORT_SYMBOL(no_seek_end_llseek);
192
193/**
194 * no_seek_end_llseek_size - llseek implementation for fixed-sized devices
195 * @file: file structure to seek on
196 * @offset: file offset to seek to
197 * @whence: type of seek
198 * @size: maximal offset allowed
199 *
200 */
201loff_t no_seek_end_llseek_size(struct file *file, loff_t offset, int whence, loff_t size)
202{
203 switch (whence) {
204 case SEEK_SET: case SEEK_CUR:
205 return generic_file_llseek_size(file, offset, whence,
206 size, 0);
207 default:
208 return -EINVAL;
209 }
210}
211EXPORT_SYMBOL(no_seek_end_llseek_size);
212
213/**
jan Blunckae6afc32010-05-26 14:44:48 -0700214 * noop_llseek - No Operation Performed llseek implementation
215 * @file: file structure to seek on
216 * @offset: file offset to seek to
Andrew Morton965c8e52012-12-17 15:59:39 -0800217 * @whence: type of seek
jan Blunckae6afc32010-05-26 14:44:48 -0700218 *
219 * This is an implementation of ->llseek useable for the rare special case when
220 * userspace expects the seek to succeed but the (device) file is actually not
221 * able to perform the seek. In this case you use noop_llseek() instead of
222 * falling back to the default implementation of ->llseek.
223 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800224loff_t noop_llseek(struct file *file, loff_t offset, int whence)
jan Blunckae6afc32010-05-26 14:44:48 -0700225{
226 return file->f_pos;
227}
228EXPORT_SYMBOL(noop_llseek);
229
Andrew Morton965c8e52012-12-17 15:59:39 -0800230loff_t no_llseek(struct file *file, loff_t offset, int whence)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231{
232 return -ESPIPE;
233}
234EXPORT_SYMBOL(no_llseek);
235
Andrew Morton965c8e52012-12-17 15:59:39 -0800236loff_t default_llseek(struct file *file, loff_t offset, int whence)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237{
Al Viro496ad9a2013-01-23 17:07:38 -0500238 struct inode *inode = file_inode(file);
David Sterba16abef02008-04-22 15:09:22 +0200239 loff_t retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240
Al Viro59551022016-01-22 15:40:57 -0500241 inode_lock(inode);
Andrew Morton965c8e52012-12-17 15:59:39 -0800242 switch (whence) {
Chris Snook7b8e8922007-05-08 00:24:13 -0700243 case SEEK_END:
Josef Bacik982d8162011-07-18 13:21:35 -0400244 offset += i_size_read(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 break;
Chris Snook7b8e8922007-05-08 00:24:13 -0700246 case SEEK_CUR:
Alain Knaff5b6f1eb2008-11-10 17:08:08 -0800247 if (offset == 0) {
248 retval = file->f_pos;
249 goto out;
250 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 offset += file->f_pos;
Josef Bacik982d8162011-07-18 13:21:35 -0400252 break;
253 case SEEK_DATA:
254 /*
255 * In the generic case the entire file is data, so as
256 * long as offset isn't at the end of the file then the
257 * offset is data.
258 */
Dan Carpenterbacb2d82011-07-26 17:25:20 +0300259 if (offset >= inode->i_size) {
260 retval = -ENXIO;
261 goto out;
262 }
Josef Bacik982d8162011-07-18 13:21:35 -0400263 break;
264 case SEEK_HOLE:
265 /*
266 * There is a virtual hole at the end of the file, so
267 * as long as offset isn't i_size or larger, return
268 * i_size.
269 */
Dan Carpenterbacb2d82011-07-26 17:25:20 +0300270 if (offset >= inode->i_size) {
271 retval = -ENXIO;
272 goto out;
273 }
Josef Bacik982d8162011-07-18 13:21:35 -0400274 offset = inode->i_size;
275 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 }
277 retval = -EINVAL;
Al Virocccb5a12010-12-17 07:44:05 -0500278 if (offset >= 0 || unsigned_offsets(file)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 if (offset != file->f_pos) {
280 file->f_pos = offset;
281 file->f_version = 0;
282 }
283 retval = offset;
284 }
Alain Knaff5b6f1eb2008-11-10 17:08:08 -0800285out:
Al Viro59551022016-01-22 15:40:57 -0500286 inode_unlock(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 return retval;
288}
289EXPORT_SYMBOL(default_llseek);
290
Andrew Morton965c8e52012-12-17 15:59:39 -0800291loff_t vfs_llseek(struct file *file, loff_t offset, int whence)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292{
293 loff_t (*fn)(struct file *, loff_t, int);
294
295 fn = no_llseek;
296 if (file->f_mode & FMODE_LSEEK) {
Al Viro72c2d532013-09-22 16:27:52 -0400297 if (file->f_op->llseek)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 fn = file->f_op->llseek;
299 }
Andrew Morton965c8e52012-12-17 15:59:39 -0800300 return fn(file, offset, whence);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301}
302EXPORT_SYMBOL(vfs_llseek);
303
Andrew Morton965c8e52012-12-17 15:59:39 -0800304SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305{
306 off_t retval;
Linus Torvalds9c225f22014-03-03 09:36:58 -0800307 struct fd f = fdget_pos(fd);
Al Viro2903ff02012-08-28 12:52:22 -0400308 if (!f.file)
309 return -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310
311 retval = -EINVAL;
Andrew Morton965c8e52012-12-17 15:59:39 -0800312 if (whence <= SEEK_MAX) {
313 loff_t res = vfs_llseek(f.file, offset, whence);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 retval = res;
315 if (res != (loff_t)retval)
316 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
317 }
Linus Torvalds9c225f22014-03-03 09:36:58 -0800318 fdput_pos(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 return retval;
320}
321
Al Viro561c6732013-02-24 10:52:26 -0500322#ifdef CONFIG_COMPAT
323COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
324{
325 return sys_lseek(fd, offset, whence);
326}
327#endif
328
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329#ifdef __ARCH_WANT_SYS_LLSEEK
Heiko Carstens003d7ab2009-01-14 14:14:21 +0100330SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
331 unsigned long, offset_low, loff_t __user *, result,
Andrew Morton965c8e52012-12-17 15:59:39 -0800332 unsigned int, whence)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333{
334 int retval;
Eric Biggersd7a15f82014-03-16 14:24:08 -0500335 struct fd f = fdget_pos(fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 loff_t offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337
Al Viro2903ff02012-08-28 12:52:22 -0400338 if (!f.file)
339 return -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340
341 retval = -EINVAL;
Andrew Morton965c8e52012-12-17 15:59:39 -0800342 if (whence > SEEK_MAX)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 goto out_putf;
344
Al Viro2903ff02012-08-28 12:52:22 -0400345 offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
Andrew Morton965c8e52012-12-17 15:59:39 -0800346 whence);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347
348 retval = (int)offset;
349 if (offset >= 0) {
350 retval = -EFAULT;
351 if (!copy_to_user(result, &offset, sizeof(offset)))
352 retval = 0;
353 }
354out_putf:
Eric Biggersd7a15f82014-03-16 14:24:08 -0500355 fdput_pos(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 return retval;
357}
358#endif
359
Al Viro68d70d02013-06-19 15:26:04 +0400360int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361{
362 struct inode *inode;
363 loff_t pos;
James Morrisc43e2592008-01-12 22:05:48 +1100364 int retval = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365
Al Viro496ad9a2013-01-23 17:07:38 -0500366 inode = file_inode(file);
Linus Torvaldse28cc712006-01-04 16:20:40 -0800367 if (unlikely((ssize_t) count < 0))
James Morrisc43e2592008-01-12 22:05:48 +1100368 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 pos = *ppos;
Al Virocccb5a12010-12-17 07:44:05 -0500370 if (unlikely(pos < 0)) {
371 if (!unsigned_offsets(file))
372 return retval;
373 if (count >= -pos) /* both values are in 0..LLONG_MAX */
374 return -EOVERFLOW;
375 } else if (unlikely((loff_t) (pos + count) < 0)) {
376 if (!unsigned_offsets(file))
KAMEZAWA Hiroyuki4a3956c2010-10-01 14:20:22 -0700377 return retval;
378 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500380 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
Christoph Hellwigacc15572015-12-03 12:59:49 +0100381 retval = locks_mandatory_area(inode, file, pos, pos + count - 1,
382 read_write == READ ? F_RDLCK : F_WRLCK);
Linus Torvaldse28cc712006-01-04 16:20:40 -0800383 if (retval < 0)
384 return retval;
385 }
Al Virobc613842016-03-31 21:48:20 -0400386 return security_file_permission(file,
James Morrisc43e2592008-01-12 22:05:48 +1100387 read_write == READ ? MAY_READ : MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388}
389
Al Viro5d5d5682015-04-03 15:41:18 -0400390static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
Al Viro293bc982014-02-11 18:37:41 -0500391{
392 struct iovec iov = { .iov_base = buf, .iov_len = len };
393 struct kiocb kiocb;
394 struct iov_iter iter;
395 ssize_t ret;
396
397 init_sync_kiocb(&kiocb, filp);
398 kiocb.ki_pos = *ppos;
Al Viro293bc982014-02-11 18:37:41 -0500399 iov_iter_init(&iter, READ, &iov, 1, len);
400
Miklos Szeredibb7462b2017-02-20 16:51:23 +0100401 ret = call_read_iter(filp, &kiocb, &iter);
Christoph Hellwig599bd192015-02-11 19:59:44 +0100402 BUG_ON(ret == -EIOCBQUEUED);
Al Viro293bc982014-02-11 18:37:41 -0500403 *ppos = kiocb.ki_pos;
404 return ret;
405}
406
Dmitry Kasatkin6fb50322014-11-05 17:01:17 +0200407ssize_t __vfs_read(struct file *file, char __user *buf, size_t count,
408 loff_t *pos)
409{
Dmitry Kasatkin6fb50322014-11-05 17:01:17 +0200410 if (file->f_op->read)
Al Viro3d04c8a2015-04-03 15:09:18 -0400411 return file->f_op->read(file, buf, count, pos);
Dmitry Kasatkin6fb50322014-11-05 17:01:17 +0200412 else if (file->f_op->read_iter)
Al Viro3d04c8a2015-04-03 15:09:18 -0400413 return new_sync_read(file, buf, count, pos);
Dmitry Kasatkin6fb50322014-11-05 17:01:17 +0200414 else
Al Viro3d04c8a2015-04-03 15:09:18 -0400415 return -EINVAL;
Dmitry Kasatkin6fb50322014-11-05 17:01:17 +0200416}
417
Christoph Hellwigbdd1d2d2017-09-01 17:39:13 +0200418ssize_t kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
Christoph Hellwigc41fbad2017-09-01 17:39:12 +0200419{
420 mm_segment_t old_fs;
Christoph Hellwigbdd1d2d2017-09-01 17:39:13 +0200421 ssize_t result;
Christoph Hellwigc41fbad2017-09-01 17:39:12 +0200422
423 old_fs = get_fs();
424 set_fs(get_ds());
425 /* The cast to a user pointer is valid due to the set_fs() */
Christoph Hellwigbdd1d2d2017-09-01 17:39:13 +0200426 result = vfs_read(file, (void __user *)buf, count, pos);
Christoph Hellwigc41fbad2017-09-01 17:39:12 +0200427 set_fs(old_fs);
428 return result;
429}
430EXPORT_SYMBOL(kernel_read);
Al Viro293bc982014-02-11 18:37:41 -0500431
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
433{
434 ssize_t ret;
435
436 if (!(file->f_mode & FMODE_READ))
437 return -EBADF;
Al Viro7f7f25e2014-02-11 17:49:24 -0500438 if (!(file->f_mode & FMODE_CAN_READ))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 return -EINVAL;
440 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
441 return -EFAULT;
442
443 ret = rw_verify_area(READ, file, pos, count);
Al Virobc613842016-03-31 21:48:20 -0400444 if (!ret) {
445 if (count > MAX_RW_COUNT)
446 count = MAX_RW_COUNT;
Dmitry Kasatkin6fb50322014-11-05 17:01:17 +0200447 ret = __vfs_read(file, buf, count, pos);
James Morrisc43e2592008-01-12 22:05:48 +1100448 if (ret > 0) {
Eric Paris2a12a9d2009-12-17 21:24:21 -0500449 fsnotify_access(file);
James Morrisc43e2592008-01-12 22:05:48 +1100450 add_rchar(current, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 }
James Morrisc43e2592008-01-12 22:05:48 +1100452 inc_syscr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 }
454
455 return ret;
456}
457
Al Viro5d5d5682015-04-03 15:41:18 -0400458static 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 -0500459{
460 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
461 struct kiocb kiocb;
462 struct iov_iter iter;
463 ssize_t ret;
464
465 init_sync_kiocb(&kiocb, filp);
466 kiocb.ki_pos = *ppos;
Al Viro293bc982014-02-11 18:37:41 -0500467 iov_iter_init(&iter, WRITE, &iov, 1, len);
468
Miklos Szeredibb7462b2017-02-20 16:51:23 +0100469 ret = call_write_iter(filp, &kiocb, &iter);
Christoph Hellwig599bd192015-02-11 19:59:44 +0100470 BUG_ON(ret == -EIOCBQUEUED);
Al Virof765b132015-04-06 20:50:38 -0400471 if (ret > 0)
472 *ppos = kiocb.ki_pos;
Al Viro293bc982014-02-11 18:37:41 -0500473 return ret;
474}
475
Al Viro493c84c2015-04-03 15:06:43 -0400476ssize_t __vfs_write(struct file *file, const char __user *p, size_t count,
477 loff_t *pos)
478{
479 if (file->f_op->write)
480 return file->f_op->write(file, p, count, pos);
Al Viro493c84c2015-04-03 15:06:43 -0400481 else if (file->f_op->write_iter)
482 return new_sync_write(file, p, count, pos);
483 else
484 return -EINVAL;
485}
Al Viro493c84c2015-04-03 15:06:43 -0400486
Christoph Hellwig73e18f72017-09-01 17:39:15 +0200487ssize_t __kernel_write(struct file *file, const void *buf, size_t count, loff_t *pos)
Al Viro06ae43f2013-03-20 13:19:30 -0400488{
489 mm_segment_t old_fs;
490 const char __user *p;
491 ssize_t ret;
492
Al Viro7f7f25e2014-02-11 17:49:24 -0500493 if (!(file->f_mode & FMODE_CAN_WRITE))
Al Viro3e84f482013-03-27 15:20:30 +0000494 return -EINVAL;
495
Al Viro06ae43f2013-03-20 13:19:30 -0400496 old_fs = get_fs();
497 set_fs(get_ds());
498 p = (__force const char __user *)buf;
499 if (count > MAX_RW_COUNT)
500 count = MAX_RW_COUNT;
Al Viro493c84c2015-04-03 15:06:43 -0400501 ret = __vfs_write(file, p, count, pos);
Al Viro06ae43f2013-03-20 13:19:30 -0400502 set_fs(old_fs);
503 if (ret > 0) {
504 fsnotify_modify(file);
505 add_wchar(current, ret);
506 }
507 inc_syscw(current);
508 return ret;
509}
Al Viro2ec3a122014-08-19 11:48:09 -0400510EXPORT_SYMBOL(__kernel_write);
511
Christoph Hellwige13ec932017-09-01 17:39:14 +0200512ssize_t kernel_write(struct file *file, const void *buf, size_t count,
513 loff_t *pos)
Christoph Hellwigac452ac2017-09-01 17:39:11 +0200514{
515 mm_segment_t old_fs;
516 ssize_t res;
517
518 old_fs = get_fs();
519 set_fs(get_ds());
520 /* The cast to a user pointer is valid due to the set_fs() */
Christoph Hellwige13ec932017-09-01 17:39:14 +0200521 res = vfs_write(file, (__force const char __user *)buf, count, pos);
Christoph Hellwigac452ac2017-09-01 17:39:11 +0200522 set_fs(old_fs);
523
524 return res;
525}
526EXPORT_SYMBOL(kernel_write);
527
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
529{
530 ssize_t ret;
531
532 if (!(file->f_mode & FMODE_WRITE))
533 return -EBADF;
Al Viro7f7f25e2014-02-11 17:49:24 -0500534 if (!(file->f_mode & FMODE_CAN_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 return -EINVAL;
536 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
537 return -EFAULT;
538
539 ret = rw_verify_area(WRITE, file, pos, count);
Al Virobc613842016-03-31 21:48:20 -0400540 if (!ret) {
541 if (count > MAX_RW_COUNT)
542 count = MAX_RW_COUNT;
Al Viro03d95eb2013-03-20 13:04:20 -0400543 file_start_write(file);
Al Viro493c84c2015-04-03 15:06:43 -0400544 ret = __vfs_write(file, buf, count, pos);
James Morrisc43e2592008-01-12 22:05:48 +1100545 if (ret > 0) {
Eric Paris2a12a9d2009-12-17 21:24:21 -0500546 fsnotify_modify(file);
James Morrisc43e2592008-01-12 22:05:48 +1100547 add_wchar(current, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 }
James Morrisc43e2592008-01-12 22:05:48 +1100549 inc_syscw(current);
Al Viro03d95eb2013-03-20 13:04:20 -0400550 file_end_write(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 }
552
553 return ret;
554}
555
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556static inline loff_t file_pos_read(struct file *file)
557{
Kirill Smelkovb673f992019-03-26 22:20:43 +0000558 return file->f_mode & FMODE_STREAM ? 0 : file->f_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559}
560
561static inline void file_pos_write(struct file *file, loff_t pos)
562{
Kirill Smelkovb673f992019-03-26 22:20:43 +0000563 if ((file->f_mode & FMODE_STREAM) == 0)
564 file->f_pos = pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565}
566
Heiko Carstens3cdad422009-01-14 14:14:22 +0100567SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568{
Linus Torvalds9c225f22014-03-03 09:36:58 -0800569 struct fd f = fdget_pos(fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 ssize_t ret = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571
Al Viro2903ff02012-08-28 12:52:22 -0400572 if (f.file) {
573 loff_t pos = file_pos_read(f.file);
574 ret = vfs_read(f.file, buf, count, &pos);
Al Viro5faf1532013-06-15 05:49:36 +0400575 if (ret >= 0)
576 file_pos_write(f.file, pos);
Linus Torvalds9c225f22014-03-03 09:36:58 -0800577 fdput_pos(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 return ret;
580}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581
Heiko Carstens3cdad422009-01-14 14:14:22 +0100582SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
583 size_t, count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584{
Linus Torvalds9c225f22014-03-03 09:36:58 -0800585 struct fd f = fdget_pos(fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 ssize_t ret = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587
Al Viro2903ff02012-08-28 12:52:22 -0400588 if (f.file) {
589 loff_t pos = file_pos_read(f.file);
590 ret = vfs_write(f.file, buf, count, &pos);
Al Viro5faf1532013-06-15 05:49:36 +0400591 if (ret >= 0)
592 file_pos_write(f.file, pos);
Linus Torvalds9c225f22014-03-03 09:36:58 -0800593 fdput_pos(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 }
595
596 return ret;
597}
598
Al Viro4a0fd5b2013-01-21 15:16:58 -0500599SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
600 size_t, count, loff_t, pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601{
Al Viro2903ff02012-08-28 12:52:22 -0400602 struct fd f;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 ssize_t ret = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604
605 if (pos < 0)
606 return -EINVAL;
607
Al Viro2903ff02012-08-28 12:52:22 -0400608 f = fdget(fd);
609 if (f.file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 ret = -ESPIPE;
Al Viro2903ff02012-08-28 12:52:22 -0400611 if (f.file->f_mode & FMODE_PREAD)
612 ret = vfs_read(f.file, buf, count, &pos);
613 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 }
615
616 return ret;
617}
618
Al Viro4a0fd5b2013-01-21 15:16:58 -0500619SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
620 size_t, count, loff_t, pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621{
Al Viro2903ff02012-08-28 12:52:22 -0400622 struct fd f;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 ssize_t ret = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624
625 if (pos < 0)
626 return -EINVAL;
627
Al Viro2903ff02012-08-28 12:52:22 -0400628 f = fdget(fd);
629 if (f.file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 ret = -ESPIPE;
Al Viro2903ff02012-08-28 12:52:22 -0400631 if (f.file->f_mode & FMODE_PWRITE)
632 ret = vfs_write(f.file, buf, count, &pos);
633 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 }
635
636 return ret;
637}
638
639/*
640 * Reduce an iovec's length in-place. Return the resulting number of segments
641 */
642unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
643{
644 unsigned long seg = 0;
645 size_t len = 0;
646
647 while (seg < nr_segs) {
648 seg++;
649 if (len + iov->iov_len >= to) {
650 iov->iov_len = to - len;
651 break;
652 }
653 len += iov->iov_len;
654 iov++;
655 }
656 return seg;
657}
Eric Sandeen19295522008-01-28 23:58:27 -0500658EXPORT_SYMBOL(iov_shorten);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659
Al Viroac15ac02015-03-20 20:10:21 -0400660static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
Christoph Hellwigddef7ed2017-07-06 18:58:37 +0200661 loff_t *ppos, int type, rwf_t flags)
Al Viro293bc982014-02-11 18:37:41 -0500662{
663 struct kiocb kiocb;
Al Viro293bc982014-02-11 18:37:41 -0500664 ssize_t ret;
665
666 init_sync_kiocb(&kiocb, filp);
Goldwyn Rodriguesfdd2f5b2017-06-20 07:05:40 -0500667 ret = kiocb_set_rw_flags(&kiocb, flags);
668 if (ret)
669 return ret;
Al Viro293bc982014-02-11 18:37:41 -0500670 kiocb.ki_pos = *ppos;
Al Viro293bc982014-02-11 18:37:41 -0500671
Miklos Szeredi0f78d062017-02-20 16:51:23 +0100672 if (type == READ)
Miklos Szeredibb7462b2017-02-20 16:51:23 +0100673 ret = call_read_iter(filp, &kiocb, iter);
Miklos Szeredi0f78d062017-02-20 16:51:23 +0100674 else
Miklos Szeredibb7462b2017-02-20 16:51:23 +0100675 ret = call_write_iter(filp, &kiocb, iter);
Christoph Hellwig599bd192015-02-11 19:59:44 +0100676 BUG_ON(ret == -EIOCBQUEUED);
Al Viro293bc982014-02-11 18:37:41 -0500677 *ppos = kiocb.ki_pos;
678 return ret;
679}
680
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700681/* Do it by hand, with file-ops */
Al Viroac15ac02015-03-20 20:10:21 -0400682static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
Christoph Hellwigddef7ed2017-07-06 18:58:37 +0200683 loff_t *ppos, int type, rwf_t flags)
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700684{
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700685 ssize_t ret = 0;
686
Christoph Hellwig97be7eb2016-03-03 16:04:01 +0100687 if (flags & ~RWF_HIPRI)
Christoph Hellwig793b80e2016-03-03 16:03:58 +0100688 return -EOPNOTSUPP;
689
Al Viroac15ac02015-03-20 20:10:21 -0400690 while (iov_iter_count(iter)) {
691 struct iovec iovec = iov_iter_iovec(iter);
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700692 ssize_t nr;
693
Miklos Szeredi0f78d062017-02-20 16:51:23 +0100694 if (type == READ) {
695 nr = filp->f_op->read(filp, iovec.iov_base,
696 iovec.iov_len, ppos);
697 } else {
698 nr = filp->f_op->write(filp, iovec.iov_base,
699 iovec.iov_len, ppos);
700 }
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700701
702 if (nr < 0) {
703 if (!ret)
704 ret = nr;
705 break;
706 }
707 ret += nr;
Al Viroac15ac02015-03-20 20:10:21 -0400708 if (nr != iovec.iov_len)
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700709 break;
Al Viroac15ac02015-03-20 20:10:21 -0400710 iov_iter_advance(iter, nr);
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700711 }
712
713 return ret;
714}
715
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716/* A write operation does a read from user space and vice versa */
717#define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
718
Vegard Nossumffecee42016-10-08 11:18:07 +0200719/**
720 * rw_copy_check_uvector() - Copy an array of &struct iovec from userspace
721 * into the kernel and check that it is valid.
722 *
723 * @type: One of %CHECK_IOVEC_ONLY, %READ, or %WRITE.
724 * @uvector: Pointer to the userspace array.
725 * @nr_segs: Number of elements in userspace array.
726 * @fast_segs: Number of elements in @fast_pointer.
727 * @fast_pointer: Pointer to (usually small on-stack) kernel array.
728 * @ret_pointer: (output parameter) Pointer to a variable that will point to
729 * either @fast_pointer, a newly allocated kernel array, or NULL,
730 * depending on which array was used.
731 *
732 * This function copies an array of &struct iovec of @nr_segs from
733 * userspace into the kernel and checks that each element is valid (e.g.
734 * it does not point to a kernel address or cause overflow by being too
735 * large, etc.).
736 *
737 * As an optimization, the caller may provide a pointer to a small
738 * on-stack array in @fast_pointer, typically %UIO_FASTIOV elements long
739 * (the size of this array, or 0 if unused, should be given in @fast_segs).
740 *
741 * @ret_pointer will always point to the array that was used, so the
742 * caller must take care not to call kfree() on it e.g. in case the
743 * @fast_pointer array was used and it was allocated on the stack.
744 *
745 * Return: The total number of bytes covered by the iovec array on success
746 * or a negative error code on error.
747 */
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700748ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
749 unsigned long nr_segs, unsigned long fast_segs,
750 struct iovec *fast_pointer,
Christopher Yeohac34ebb2012-05-31 16:26:42 -0700751 struct iovec **ret_pointer)
Linus Torvalds435f49a2010-10-29 10:36:49 -0700752{
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700753 unsigned long seg;
Linus Torvalds435f49a2010-10-29 10:36:49 -0700754 ssize_t ret;
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700755 struct iovec *iov = fast_pointer;
756
Linus Torvalds435f49a2010-10-29 10:36:49 -0700757 /*
758 * SuS says "The readv() function *may* fail if the iovcnt argument
759 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
760 * traditionally returned zero for zero segments, so...
761 */
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700762 if (nr_segs == 0) {
763 ret = 0;
Linus Torvalds435f49a2010-10-29 10:36:49 -0700764 goto out;
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700765 }
766
Linus Torvalds435f49a2010-10-29 10:36:49 -0700767 /*
768 * First get the "struct iovec" from user memory and
769 * verify all the pointers
770 */
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700771 if (nr_segs > UIO_MAXIOV) {
772 ret = -EINVAL;
Linus Torvalds435f49a2010-10-29 10:36:49 -0700773 goto out;
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700774 }
775 if (nr_segs > fast_segs) {
Linus Torvalds435f49a2010-10-29 10:36:49 -0700776 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700777 if (iov == NULL) {
778 ret = -ENOMEM;
Linus Torvalds435f49a2010-10-29 10:36:49 -0700779 goto out;
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700780 }
Linus Torvalds435f49a2010-10-29 10:36:49 -0700781 }
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700782 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
783 ret = -EFAULT;
Linus Torvalds435f49a2010-10-29 10:36:49 -0700784 goto out;
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700785 }
786
Linus Torvalds435f49a2010-10-29 10:36:49 -0700787 /*
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700788 * According to the Single Unix Specification we should return EINVAL
789 * if an element length is < 0 when cast to ssize_t or if the
790 * total length would overflow the ssize_t return value of the
791 * system call.
Linus Torvalds435f49a2010-10-29 10:36:49 -0700792 *
793 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
794 * overflow case.
795 */
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700796 ret = 0;
Linus Torvalds435f49a2010-10-29 10:36:49 -0700797 for (seg = 0; seg < nr_segs; seg++) {
798 void __user *buf = iov[seg].iov_base;
799 ssize_t len = (ssize_t)iov[seg].iov_len;
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700800
801 /* see if we we're about to use an invalid len or if
802 * it's about to overflow ssize_t */
Linus Torvalds435f49a2010-10-29 10:36:49 -0700803 if (len < 0) {
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700804 ret = -EINVAL;
Linus Torvalds435f49a2010-10-29 10:36:49 -0700805 goto out;
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700806 }
Christopher Yeohac34ebb2012-05-31 16:26:42 -0700807 if (type >= 0
Christopher Yeohfcf63402011-10-31 17:06:39 -0700808 && unlikely(!access_ok(vrfy_dir(type), buf, len))) {
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700809 ret = -EFAULT;
Linus Torvalds435f49a2010-10-29 10:36:49 -0700810 goto out;
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700811 }
Linus Torvalds435f49a2010-10-29 10:36:49 -0700812 if (len > MAX_RW_COUNT - ret) {
813 len = MAX_RW_COUNT - ret;
814 iov[seg].iov_len = len;
815 }
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700816 ret += len;
Linus Torvalds435f49a2010-10-29 10:36:49 -0700817 }
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700818out:
819 *ret_pointer = iov;
820 return ret;
821}
822
Al Virof5029852017-04-08 18:18:48 -0400823#ifdef CONFIG_COMPAT
824ssize_t compat_rw_copy_check_uvector(int type,
825 const struct compat_iovec __user *uvector, unsigned long nr_segs,
826 unsigned long fast_segs, struct iovec *fast_pointer,
827 struct iovec **ret_pointer)
828{
829 compat_ssize_t tot_len;
830 struct iovec *iov = *ret_pointer = fast_pointer;
831 ssize_t ret = 0;
832 int seg;
833
834 /*
835 * SuS says "The readv() function *may* fail if the iovcnt argument
836 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
837 * traditionally returned zero for zero segments, so...
838 */
839 if (nr_segs == 0)
840 goto out;
841
842 ret = -EINVAL;
843 if (nr_segs > UIO_MAXIOV)
844 goto out;
845 if (nr_segs > fast_segs) {
846 ret = -ENOMEM;
847 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
848 if (iov == NULL)
849 goto out;
850 }
851 *ret_pointer = iov;
852
853 ret = -EFAULT;
854 if (!access_ok(VERIFY_READ, uvector, nr_segs*sizeof(*uvector)))
855 goto out;
856
857 /*
858 * Single unix specification:
859 * We should -EINVAL if an element length is not >= 0 and fitting an
860 * ssize_t.
861 *
862 * In Linux, the total length is limited to MAX_RW_COUNT, there is
863 * no overflow possibility.
864 */
865 tot_len = 0;
866 ret = -EINVAL;
867 for (seg = 0; seg < nr_segs; seg++) {
868 compat_uptr_t buf;
869 compat_ssize_t len;
870
871 if (__get_user(len, &uvector->iov_len) ||
872 __get_user(buf, &uvector->iov_base)) {
873 ret = -EFAULT;
874 goto out;
875 }
876 if (len < 0) /* size_t not fitting in compat_ssize_t .. */
877 goto out;
878 if (type >= 0 &&
879 !access_ok(vrfy_dir(type), compat_ptr(buf), len)) {
880 ret = -EFAULT;
881 goto out;
882 }
883 if (len > MAX_RW_COUNT - tot_len)
884 len = MAX_RW_COUNT - tot_len;
885 tot_len += len;
886 iov->iov_base = compat_ptr(buf);
887 iov->iov_len = (compat_size_t) len;
888 uvector++;
889 iov++;
890 }
891 ret = tot_len;
892
893out:
894 return ret;
895}
896#endif
897
Christoph Hellwig19c73582017-05-27 11:16:48 +0300898static ssize_t do_iter_read(struct file *file, struct iov_iter *iter,
Christoph Hellwigddef7ed2017-07-06 18:58:37 +0200899 loff_t *pos, rwf_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 size_t tot_len;
Miklos Szeredi7687a7a2017-02-20 16:51:23 +0100902 ssize_t ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903
Christoph Hellwigedab5fe2017-05-27 11:16:49 +0300904 if (!(file->f_mode & FMODE_READ))
905 return -EBADF;
906 if (!(file->f_mode & FMODE_CAN_READ))
907 return -EINVAL;
908
Miklos Szeredi7687a7a2017-02-20 16:51:23 +0100909 tot_len = iov_iter_count(iter);
Al Viro0504c072015-03-21 19:40:11 -0400910 if (!tot_len)
911 goto out;
Christoph Hellwig19c73582017-05-27 11:16:48 +0300912 ret = rw_verify_area(READ, file, pos, tot_len);
Linus Torvaldse28cc712006-01-04 16:20:40 -0800913 if (ret < 0)
Christoph Hellwig19c73582017-05-27 11:16:48 +0300914 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915
Christoph Hellwig19c73582017-05-27 11:16:48 +0300916 if (file->f_op->read_iter)
917 ret = do_iter_readv_writev(file, iter, pos, READ, flags);
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700918 else
Christoph Hellwig19c73582017-05-27 11:16:48 +0300919 ret = do_loop_readv_writev(file, iter, pos, READ, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920out:
Christoph Hellwig19c73582017-05-27 11:16:48 +0300921 if (ret >= 0)
922 fsnotify_access(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924}
925
Christoph Hellwig18e97102017-05-27 11:16:51 +0300926ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos,
Christoph Hellwigddef7ed2017-07-06 18:58:37 +0200927 rwf_t flags)
Christoph Hellwig18e97102017-05-27 11:16:51 +0300928{
929 if (!file->f_op->read_iter)
930 return -EINVAL;
931 return do_iter_read(file, iter, ppos, flags);
932}
933EXPORT_SYMBOL(vfs_iter_read);
934
Christoph Hellwig19c73582017-05-27 11:16:48 +0300935static ssize_t do_iter_write(struct file *file, struct iov_iter *iter,
Christoph Hellwigddef7ed2017-07-06 18:58:37 +0200936 loff_t *pos, rwf_t flags)
Christoph Hellwig19c73582017-05-27 11:16:48 +0300937{
938 size_t tot_len;
939 ssize_t ret = 0;
940
Christoph Hellwigedab5fe2017-05-27 11:16:49 +0300941 if (!(file->f_mode & FMODE_WRITE))
942 return -EBADF;
943 if (!(file->f_mode & FMODE_CAN_WRITE))
944 return -EINVAL;
945
Christoph Hellwig19c73582017-05-27 11:16:48 +0300946 tot_len = iov_iter_count(iter);
947 if (!tot_len)
948 return 0;
949 ret = rw_verify_area(WRITE, file, pos, tot_len);
950 if (ret < 0)
951 return ret;
952
Christoph Hellwig19c73582017-05-27 11:16:48 +0300953 if (file->f_op->write_iter)
954 ret = do_iter_readv_writev(file, iter, pos, WRITE, flags);
955 else
956 ret = do_loop_readv_writev(file, iter, pos, WRITE, flags);
Christoph Hellwig19c73582017-05-27 11:16:48 +0300957 if (ret > 0)
958 fsnotify_modify(file);
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700959 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960}
961
Christoph Hellwigabbb65892017-05-27 11:16:52 +0300962ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos,
Christoph Hellwigddef7ed2017-07-06 18:58:37 +0200963 rwf_t flags)
Christoph Hellwigabbb65892017-05-27 11:16:52 +0300964{
965 if (!file->f_op->write_iter)
966 return -EINVAL;
967 return do_iter_write(file, iter, ppos, flags);
968}
969EXPORT_SYMBOL(vfs_iter_write);
970
Christoph Hellwig251b42a2017-05-27 11:16:46 +0300971ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
Christoph Hellwigddef7ed2017-07-06 18:58:37 +0200972 unsigned long vlen, loff_t *pos, rwf_t flags)
Miklos Szeredi7687a7a2017-02-20 16:51:23 +0100973{
974 struct iovec iovstack[UIO_FASTIOV];
975 struct iovec *iov = iovstack;
976 struct iov_iter iter;
977 ssize_t ret;
978
Christoph Hellwig251b42a2017-05-27 11:16:46 +0300979 ret = import_iovec(READ, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
Christoph Hellwigedab5fe2017-05-27 11:16:49 +0300980 if (ret >= 0) {
981 ret = do_iter_read(file, &iter, pos, flags);
982 kfree(iov);
983 }
Miklos Szeredi7687a7a2017-02-20 16:51:23 +0100984
985 return ret;
986}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987
Christoph Hellwig9725d4c2017-09-01 17:39:25 +0200988static ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
Christoph Hellwigddef7ed2017-07-06 18:58:37 +0200989 unsigned long vlen, loff_t *pos, rwf_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990{
Christoph Hellwig251b42a2017-05-27 11:16:46 +0300991 struct iovec iovstack[UIO_FASTIOV];
992 struct iovec *iov = iovstack;
993 struct iov_iter iter;
994 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995
Christoph Hellwig251b42a2017-05-27 11:16:46 +0300996 ret = import_iovec(WRITE, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
Christoph Hellwigedab5fe2017-05-27 11:16:49 +0300997 if (ret >= 0) {
Al Viro62473a22017-07-06 09:15:47 -0400998 file_start_write(file);
Christoph Hellwigedab5fe2017-05-27 11:16:49 +0300999 ret = do_iter_write(file, &iter, pos, flags);
Al Viro62473a22017-07-06 09:15:47 -04001000 file_end_write(file);
Christoph Hellwigedab5fe2017-05-27 11:16:49 +03001001 kfree(iov);
1002 }
Christoph Hellwig251b42a2017-05-27 11:16:46 +03001003 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001006static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
Christoph Hellwigddef7ed2017-07-06 18:58:37 +02001007 unsigned long vlen, rwf_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008{
Linus Torvalds9c225f22014-03-03 09:36:58 -08001009 struct fd f = fdget_pos(fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 ssize_t ret = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011
Al Viro2903ff02012-08-28 12:52:22 -04001012 if (f.file) {
1013 loff_t pos = file_pos_read(f.file);
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001014 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
Al Viro5faf1532013-06-15 05:49:36 +04001015 if (ret >= 0)
1016 file_pos_write(f.file, pos);
Linus Torvalds9c225f22014-03-03 09:36:58 -08001017 fdput_pos(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 }
1019
1020 if (ret > 0)
Alexey Dobriyan4b98d112007-02-10 01:46:45 -08001021 add_rchar(current, ret);
1022 inc_syscr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 return ret;
1024}
1025
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001026static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
Christoph Hellwigddef7ed2017-07-06 18:58:37 +02001027 unsigned long vlen, rwf_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028{
Linus Torvalds9c225f22014-03-03 09:36:58 -08001029 struct fd f = fdget_pos(fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 ssize_t ret = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031
Al Viro2903ff02012-08-28 12:52:22 -04001032 if (f.file) {
1033 loff_t pos = file_pos_read(f.file);
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001034 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
Al Viro5faf1532013-06-15 05:49:36 +04001035 if (ret >= 0)
1036 file_pos_write(f.file, pos);
Linus Torvalds9c225f22014-03-03 09:36:58 -08001037 fdput_pos(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 }
1039
1040 if (ret > 0)
Alexey Dobriyan4b98d112007-02-10 01:46:45 -08001041 add_wchar(current, ret);
1042 inc_syscw(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 return ret;
1044}
1045
Linus Torvalds601cc112009-04-03 08:03:22 -07001046static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
Gerd Hoffmannf3554f42009-04-02 16:59:23 -07001047{
Linus Torvalds601cc112009-04-03 08:03:22 -07001048#define HALF_LONG_BITS (BITS_PER_LONG / 2)
1049 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
1050}
1051
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001052static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
Christoph Hellwigddef7ed2017-07-06 18:58:37 +02001053 unsigned long vlen, loff_t pos, rwf_t flags)
Linus Torvalds601cc112009-04-03 08:03:22 -07001054{
Al Viro2903ff02012-08-28 12:52:22 -04001055 struct fd f;
Gerd Hoffmannf3554f42009-04-02 16:59:23 -07001056 ssize_t ret = -EBADF;
Gerd Hoffmannf3554f42009-04-02 16:59:23 -07001057
1058 if (pos < 0)
1059 return -EINVAL;
1060
Al Viro2903ff02012-08-28 12:52:22 -04001061 f = fdget(fd);
1062 if (f.file) {
Gerd Hoffmannf3554f42009-04-02 16:59:23 -07001063 ret = -ESPIPE;
Al Viro2903ff02012-08-28 12:52:22 -04001064 if (f.file->f_mode & FMODE_PREAD)
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001065 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
Al Viro2903ff02012-08-28 12:52:22 -04001066 fdput(f);
Gerd Hoffmannf3554f42009-04-02 16:59:23 -07001067 }
1068
1069 if (ret > 0)
1070 add_rchar(current, ret);
1071 inc_syscr(current);
1072 return ret;
1073}
1074
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001075static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
Christoph Hellwigddef7ed2017-07-06 18:58:37 +02001076 unsigned long vlen, loff_t pos, rwf_t flags)
Gerd Hoffmannf3554f42009-04-02 16:59:23 -07001077{
Al Viro2903ff02012-08-28 12:52:22 -04001078 struct fd f;
Gerd Hoffmannf3554f42009-04-02 16:59:23 -07001079 ssize_t ret = -EBADF;
Gerd Hoffmannf3554f42009-04-02 16:59:23 -07001080
1081 if (pos < 0)
1082 return -EINVAL;
1083
Al Viro2903ff02012-08-28 12:52:22 -04001084 f = fdget(fd);
1085 if (f.file) {
Gerd Hoffmannf3554f42009-04-02 16:59:23 -07001086 ret = -ESPIPE;
Al Viro2903ff02012-08-28 12:52:22 -04001087 if (f.file->f_mode & FMODE_PWRITE)
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001088 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
Al Viro2903ff02012-08-28 12:52:22 -04001089 fdput(f);
Gerd Hoffmannf3554f42009-04-02 16:59:23 -07001090 }
1091
1092 if (ret > 0)
1093 add_wchar(current, ret);
1094 inc_syscw(current);
1095 return ret;
1096}
1097
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001098SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
1099 unsigned long, vlen)
1100{
1101 return do_readv(fd, vec, vlen, 0);
1102}
1103
1104SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
1105 unsigned long, vlen)
1106{
1107 return do_writev(fd, vec, vlen, 0);
1108}
1109
1110SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
1111 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1112{
1113 loff_t pos = pos_from_hilo(pos_h, pos_l);
1114
1115 return do_preadv(fd, vec, vlen, pos, 0);
1116}
1117
1118SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1119 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
Christoph Hellwigddef7ed2017-07-06 18:58:37 +02001120 rwf_t, flags)
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001121{
1122 loff_t pos = pos_from_hilo(pos_h, pos_l);
1123
1124 if (pos == -1)
1125 return do_readv(fd, vec, vlen, flags);
1126
1127 return do_preadv(fd, vec, vlen, pos, flags);
1128}
1129
1130SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1131 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1132{
1133 loff_t pos = pos_from_hilo(pos_h, pos_l);
1134
1135 return do_pwritev(fd, vec, vlen, pos, 0);
1136}
1137
1138SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1139 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
Christoph Hellwigddef7ed2017-07-06 18:58:37 +02001140 rwf_t, flags)
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001141{
1142 loff_t pos = pos_from_hilo(pos_h, pos_l);
1143
1144 if (pos == -1)
1145 return do_writev(fd, vec, vlen, flags);
1146
1147 return do_pwritev(fd, vec, vlen, pos, flags);
1148}
1149
Al Viro72ec3512013-03-20 10:42:10 -04001150#ifdef CONFIG_COMPAT
Al Viro72ec3512013-03-20 10:42:10 -04001151static size_t compat_readv(struct file *file,
1152 const struct compat_iovec __user *vec,
Christoph Hellwigddef7ed2017-07-06 18:58:37 +02001153 unsigned long vlen, loff_t *pos, rwf_t flags)
Al Viro72ec3512013-03-20 10:42:10 -04001154{
1155 struct iovec iovstack[UIO_FASTIOV];
1156 struct iovec *iov = iovstack;
1157 struct iov_iter iter;
1158 ssize_t ret;
1159
Christoph Hellwig26c87fb2017-05-27 11:16:47 +03001160 ret = compat_import_iovec(READ, vec, vlen, UIO_FASTIOV, &iov, &iter);
Christoph Hellwigedab5fe2017-05-27 11:16:49 +03001161 if (ret >= 0) {
1162 ret = do_iter_read(file, &iter, pos, flags);
1163 kfree(iov);
1164 }
Al Viro72ec3512013-03-20 10:42:10 -04001165 if (ret > 0)
1166 add_rchar(current, ret);
1167 inc_syscr(current);
1168 return ret;
1169}
1170
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001171static size_t do_compat_readv(compat_ulong_t fd,
1172 const struct compat_iovec __user *vec,
Christoph Hellwigddef7ed2017-07-06 18:58:37 +02001173 compat_ulong_t vlen, rwf_t flags)
Al Viro72ec3512013-03-20 10:42:10 -04001174{
Linus Torvalds9c225f22014-03-03 09:36:58 -08001175 struct fd f = fdget_pos(fd);
Al Viro72ec3512013-03-20 10:42:10 -04001176 ssize_t ret;
1177 loff_t pos;
1178
1179 if (!f.file)
1180 return -EBADF;
1181 pos = f.file->f_pos;
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001182 ret = compat_readv(f.file, vec, vlen, &pos, flags);
Al Viro5faf1532013-06-15 05:49:36 +04001183 if (ret >= 0)
1184 f.file->f_pos = pos;
Linus Torvalds9c225f22014-03-03 09:36:58 -08001185 fdput_pos(f);
Al Viro72ec3512013-03-20 10:42:10 -04001186 return ret;
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001187
Al Viro72ec3512013-03-20 10:42:10 -04001188}
1189
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001190COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1191 const struct compat_iovec __user *,vec,
1192 compat_ulong_t, vlen)
1193{
1194 return do_compat_readv(fd, vec, vlen, 0);
1195}
1196
1197static long do_compat_preadv64(unsigned long fd,
Heiko Carstens378a10f2014-03-05 10:43:51 +01001198 const struct compat_iovec __user *vec,
Christoph Hellwigddef7ed2017-07-06 18:58:37 +02001199 unsigned long vlen, loff_t pos, rwf_t flags)
Al Viro72ec3512013-03-20 10:42:10 -04001200{
1201 struct fd f;
1202 ssize_t ret;
1203
1204 if (pos < 0)
1205 return -EINVAL;
1206 f = fdget(fd);
1207 if (!f.file)
1208 return -EBADF;
1209 ret = -ESPIPE;
1210 if (f.file->f_mode & FMODE_PREAD)
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001211 ret = compat_readv(f.file, vec, vlen, &pos, flags);
Al Viro72ec3512013-03-20 10:42:10 -04001212 fdput(f);
1213 return ret;
1214}
1215
Heiko Carstens378a10f2014-03-05 10:43:51 +01001216#ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1217COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1218 const struct compat_iovec __user *,vec,
1219 unsigned long, vlen, loff_t, pos)
1220{
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001221 return do_compat_preadv64(fd, vec, vlen, pos, 0);
Heiko Carstens378a10f2014-03-05 10:43:51 +01001222}
1223#endif
1224
Heiko Carstensdfd948e2014-01-29 14:05:44 -08001225COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
Al Viro72ec3512013-03-20 10:42:10 -04001226 const struct compat_iovec __user *,vec,
Heiko Carstensdfd948e2014-01-29 14:05:44 -08001227 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
Al Viro72ec3512013-03-20 10:42:10 -04001228{
1229 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
Heiko Carstens378a10f2014-03-05 10:43:51 +01001230
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001231 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1232}
1233
H.J. Lu3ebfd812016-07-14 12:31:53 -07001234#ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1235COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd,
1236 const struct compat_iovec __user *,vec,
Christoph Hellwigddef7ed2017-07-06 18:58:37 +02001237 unsigned long, vlen, loff_t, pos, rwf_t, flags)
H.J. Lu3ebfd812016-07-14 12:31:53 -07001238{
Aurelien Jarno58c90c42018-12-06 20:05:34 +01001239 if (pos == -1)
1240 return do_compat_readv(fd, vec, vlen, flags);
1241
H.J. Lu3ebfd812016-07-14 12:31:53 -07001242 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1243}
1244#endif
1245
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001246COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1247 const struct compat_iovec __user *,vec,
1248 compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
Christoph Hellwigddef7ed2017-07-06 18:58:37 +02001249 rwf_t, flags)
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001250{
1251 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1252
1253 if (pos == -1)
1254 return do_compat_readv(fd, vec, vlen, flags);
1255
1256 return do_compat_preadv64(fd, vec, vlen, pos, flags);
Al Viro72ec3512013-03-20 10:42:10 -04001257}
1258
1259static size_t compat_writev(struct file *file,
1260 const struct compat_iovec __user *vec,
Christoph Hellwigddef7ed2017-07-06 18:58:37 +02001261 unsigned long vlen, loff_t *pos, rwf_t flags)
Al Viro72ec3512013-03-20 10:42:10 -04001262{
Christoph Hellwig26c87fb2017-05-27 11:16:47 +03001263 struct iovec iovstack[UIO_FASTIOV];
1264 struct iovec *iov = iovstack;
1265 struct iov_iter iter;
Christoph Hellwigedab5fe2017-05-27 11:16:49 +03001266 ssize_t ret;
Al Viro72ec3512013-03-20 10:42:10 -04001267
Christoph Hellwig26c87fb2017-05-27 11:16:47 +03001268 ret = compat_import_iovec(WRITE, vec, vlen, UIO_FASTIOV, &iov, &iter);
Christoph Hellwigedab5fe2017-05-27 11:16:49 +03001269 if (ret >= 0) {
Al Viro62473a22017-07-06 09:15:47 -04001270 file_start_write(file);
Christoph Hellwigedab5fe2017-05-27 11:16:49 +03001271 ret = do_iter_write(file, &iter, pos, flags);
Al Viro62473a22017-07-06 09:15:47 -04001272 file_end_write(file);
Christoph Hellwigedab5fe2017-05-27 11:16:49 +03001273 kfree(iov);
1274 }
Al Viro72ec3512013-03-20 10:42:10 -04001275 if (ret > 0)
1276 add_wchar(current, ret);
1277 inc_syscw(current);
1278 return ret;
1279}
1280
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001281static size_t do_compat_writev(compat_ulong_t fd,
1282 const struct compat_iovec __user* vec,
Christoph Hellwigddef7ed2017-07-06 18:58:37 +02001283 compat_ulong_t vlen, rwf_t flags)
Al Viro72ec3512013-03-20 10:42:10 -04001284{
Linus Torvalds9c225f22014-03-03 09:36:58 -08001285 struct fd f = fdget_pos(fd);
Al Viro72ec3512013-03-20 10:42:10 -04001286 ssize_t ret;
1287 loff_t pos;
1288
1289 if (!f.file)
1290 return -EBADF;
1291 pos = f.file->f_pos;
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001292 ret = compat_writev(f.file, vec, vlen, &pos, flags);
Al Viro5faf1532013-06-15 05:49:36 +04001293 if (ret >= 0)
1294 f.file->f_pos = pos;
Linus Torvalds9c225f22014-03-03 09:36:58 -08001295 fdput_pos(f);
Al Viro72ec3512013-03-20 10:42:10 -04001296 return ret;
1297}
1298
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001299COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1300 const struct compat_iovec __user *, vec,
1301 compat_ulong_t, vlen)
1302{
1303 return do_compat_writev(fd, vec, vlen, 0);
1304}
1305
1306static long do_compat_pwritev64(unsigned long fd,
Heiko Carstens378a10f2014-03-05 10:43:51 +01001307 const struct compat_iovec __user *vec,
Christoph Hellwigddef7ed2017-07-06 18:58:37 +02001308 unsigned long vlen, loff_t pos, rwf_t flags)
Al Viro72ec3512013-03-20 10:42:10 -04001309{
1310 struct fd f;
1311 ssize_t ret;
1312
1313 if (pos < 0)
1314 return -EINVAL;
1315 f = fdget(fd);
1316 if (!f.file)
1317 return -EBADF;
1318 ret = -ESPIPE;
1319 if (f.file->f_mode & FMODE_PWRITE)
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001320 ret = compat_writev(f.file, vec, vlen, &pos, flags);
Al Viro72ec3512013-03-20 10:42:10 -04001321 fdput(f);
1322 return ret;
1323}
1324
Heiko Carstens378a10f2014-03-05 10:43:51 +01001325#ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1326COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1327 const struct compat_iovec __user *,vec,
1328 unsigned long, vlen, loff_t, pos)
1329{
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001330 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
Heiko Carstens378a10f2014-03-05 10:43:51 +01001331}
1332#endif
1333
Heiko Carstensdfd948e2014-01-29 14:05:44 -08001334COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
Al Viro72ec3512013-03-20 10:42:10 -04001335 const struct compat_iovec __user *,vec,
Heiko Carstensdfd948e2014-01-29 14:05:44 -08001336 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
Al Viro72ec3512013-03-20 10:42:10 -04001337{
1338 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
Heiko Carstens378a10f2014-03-05 10:43:51 +01001339
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001340 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
Al Viro72ec3512013-03-20 10:42:10 -04001341}
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001342
H.J. Lu3ebfd812016-07-14 12:31:53 -07001343#ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1344COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd,
1345 const struct compat_iovec __user *,vec,
Christoph Hellwigddef7ed2017-07-06 18:58:37 +02001346 unsigned long, vlen, loff_t, pos, rwf_t, flags)
H.J. Lu3ebfd812016-07-14 12:31:53 -07001347{
Aurelien Jarno58c90c42018-12-06 20:05:34 +01001348 if (pos == -1)
1349 return do_compat_writev(fd, vec, vlen, flags);
1350
H.J. Lu3ebfd812016-07-14 12:31:53 -07001351 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1352}
1353#endif
1354
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001355COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1356 const struct compat_iovec __user *,vec,
Christoph Hellwigddef7ed2017-07-06 18:58:37 +02001357 compat_ulong_t, vlen, u32, pos_low, u32, pos_high, rwf_t, flags)
Milosz Tanskif17d8b32016-03-03 16:03:59 +01001358{
1359 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1360
1361 if (pos == -1)
1362 return do_compat_writev(fd, vec, vlen, flags);
1363
1364 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1365}
1366
Al Viro72ec3512013-03-20 10:42:10 -04001367#endif
1368
Al Viro19f4fc32013-02-24 02:17:03 -05001369static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1370 size_t count, loff_t max)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371{
Al Viro2903ff02012-08-28 12:52:22 -04001372 struct fd in, out;
1373 struct inode *in_inode, *out_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 loff_t pos;
Al Viro7995bd22013-06-20 18:58:36 +04001375 loff_t out_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 ssize_t retval;
Al Viro2903ff02012-08-28 12:52:22 -04001377 int fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378
1379 /*
1380 * Get input file, and verify that it is ok..
1381 */
1382 retval = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001383 in = fdget(in_fd);
1384 if (!in.file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 goto out;
Al Viro2903ff02012-08-28 12:52:22 -04001386 if (!(in.file->f_mode & FMODE_READ))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 goto fput_in;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 retval = -ESPIPE;
Al Viro7995bd22013-06-20 18:58:36 +04001389 if (!ppos) {
1390 pos = in.file->f_pos;
1391 } else {
1392 pos = *ppos;
Al Viro2903ff02012-08-28 12:52:22 -04001393 if (!(in.file->f_mode & FMODE_PREAD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 goto fput_in;
Al Viro7995bd22013-06-20 18:58:36 +04001395 }
1396 retval = rw_verify_area(READ, in.file, &pos, count);
Linus Torvaldse28cc712006-01-04 16:20:40 -08001397 if (retval < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 goto fput_in;
Al Virobc613842016-03-31 21:48:20 -04001399 if (count > MAX_RW_COUNT)
1400 count = MAX_RW_COUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 /*
1403 * Get output file, and verify that it is ok..
1404 */
1405 retval = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001406 out = fdget(out_fd);
1407 if (!out.file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 goto fput_in;
Al Viro2903ff02012-08-28 12:52:22 -04001409 if (!(out.file->f_mode & FMODE_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 goto fput_out;
1411 retval = -EINVAL;
Al Viro496ad9a2013-01-23 17:07:38 -05001412 in_inode = file_inode(in.file);
1413 out_inode = file_inode(out.file);
Al Viro7995bd22013-06-20 18:58:36 +04001414 out_pos = out.file->f_pos;
1415 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
Linus Torvaldse28cc712006-01-04 16:20:40 -08001416 if (retval < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 goto fput_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 if (!max)
1420 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1421
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 if (unlikely(pos + count > max)) {
1423 retval = -EOVERFLOW;
1424 if (pos >= max)
1425 goto fput_out;
1426 count = max - pos;
1427 }
1428
Jens Axboed96e6e72007-06-11 12:18:52 +02001429 fl = 0;
Jens Axboe534f2aa2007-06-01 14:52:37 +02001430#if 0
Jens Axboed96e6e72007-06-11 12:18:52 +02001431 /*
1432 * We need to debate whether we can enable this or not. The
1433 * man page documents EAGAIN return for the output at least,
1434 * and the application is arguably buggy if it doesn't expect
1435 * EAGAIN on a non-blocking file descriptor.
1436 */
Al Viro2903ff02012-08-28 12:52:22 -04001437 if (in.file->f_flags & O_NONBLOCK)
Jens Axboed96e6e72007-06-11 12:18:52 +02001438 fl = SPLICE_F_NONBLOCK;
Jens Axboe534f2aa2007-06-01 14:52:37 +02001439#endif
Al Viro50cd2c52013-05-23 20:10:34 -04001440 file_start_write(out.file);
Al Viro7995bd22013-06-20 18:58:36 +04001441 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
Al Viro50cd2c52013-05-23 20:10:34 -04001442 file_end_write(out.file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443
1444 if (retval > 0) {
Alexey Dobriyan4b98d112007-02-10 01:46:45 -08001445 add_rchar(current, retval);
1446 add_wchar(current, retval);
Scott Wolchoka68c2f12012-12-20 15:05:52 -08001447 fsnotify_access(in.file);
1448 fsnotify_modify(out.file);
Al Viro7995bd22013-06-20 18:58:36 +04001449 out.file->f_pos = out_pos;
1450 if (ppos)
1451 *ppos = pos;
1452 else
1453 in.file->f_pos = pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455
Alexey Dobriyan4b98d112007-02-10 01:46:45 -08001456 inc_syscr(current);
1457 inc_syscw(current);
Al Viro7995bd22013-06-20 18:58:36 +04001458 if (pos > max)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 retval = -EOVERFLOW;
1460
1461fput_out:
Al Viro2903ff02012-08-28 12:52:22 -04001462 fdput(out);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463fput_in:
Al Viro2903ff02012-08-28 12:52:22 -04001464 fdput(in);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465out:
1466 return retval;
1467}
1468
Heiko Carstens002c8972009-01-14 14:14:18 +01001469SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470{
1471 loff_t pos;
1472 off_t off;
1473 ssize_t ret;
1474
1475 if (offset) {
1476 if (unlikely(get_user(off, offset)))
1477 return -EFAULT;
1478 pos = off;
1479 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1480 if (unlikely(put_user(pos, offset)))
1481 return -EFAULT;
1482 return ret;
1483 }
1484
1485 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1486}
1487
Heiko Carstens002c8972009-01-14 14:14:18 +01001488SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489{
1490 loff_t pos;
1491 ssize_t ret;
1492
1493 if (offset) {
1494 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1495 return -EFAULT;
1496 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1497 if (unlikely(put_user(pos, offset)))
1498 return -EFAULT;
1499 return ret;
1500 }
1501
1502 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1503}
Al Viro19f4fc32013-02-24 02:17:03 -05001504
1505#ifdef CONFIG_COMPAT
1506COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1507 compat_off_t __user *, offset, compat_size_t, count)
1508{
1509 loff_t pos;
1510 off_t off;
1511 ssize_t ret;
1512
1513 if (offset) {
1514 if (unlikely(get_user(off, offset)))
1515 return -EFAULT;
1516 pos = off;
1517 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1518 if (unlikely(put_user(pos, offset)))
1519 return -EFAULT;
1520 return ret;
1521 }
1522
1523 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1524}
1525
1526COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1527 compat_loff_t __user *, offset, compat_size_t, count)
1528{
1529 loff_t pos;
1530 ssize_t ret;
1531
1532 if (offset) {
1533 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1534 return -EFAULT;
1535 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1536 if (unlikely(put_user(pos, offset)))
1537 return -EFAULT;
1538 return ret;
1539 }
1540
1541 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1542}
1543#endif
Zach Brown29732932015-11-10 16:53:30 -05001544
1545/*
1546 * copy_file_range() differs from regular file read and write in that it
1547 * specifically allows return partial success. When it does so is up to
1548 * the copy_file_range method.
1549 */
1550ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1551 struct file *file_out, loff_t pos_out,
1552 size_t len, unsigned int flags)
1553{
1554 struct inode *inode_in = file_inode(file_in);
1555 struct inode *inode_out = file_inode(file_out);
1556 ssize_t ret;
1557
1558 if (flags != 0)
1559 return -EINVAL;
1560
Amir Goldstein11cbfb12017-01-31 10:34:56 +02001561 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1562 return -EISDIR;
1563 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1564 return -EINVAL;
1565
Zach Brown29732932015-11-10 16:53:30 -05001566 ret = rw_verify_area(READ, file_in, &pos_in, len);
Al Virobc613842016-03-31 21:48:20 -04001567 if (unlikely(ret))
1568 return ret;
1569
1570 ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1571 if (unlikely(ret))
Zach Brown29732932015-11-10 16:53:30 -05001572 return ret;
1573
1574 if (!(file_in->f_mode & FMODE_READ) ||
1575 !(file_out->f_mode & FMODE_WRITE) ||
Anna Schumakereac70052015-11-10 16:53:33 -05001576 (file_out->f_flags & O_APPEND))
Zach Brown29732932015-11-10 16:53:30 -05001577 return -EBADF;
1578
1579 /* this could be relaxed once a method supports cross-fs copies */
1580 if (inode_in->i_sb != inode_out->i_sb)
1581 return -EXDEV;
1582
1583 if (len == 0)
1584 return 0;
1585
Amir Goldsteinbfe219d2017-01-31 10:34:57 +02001586 file_start_write(file_out);
Zach Brown29732932015-11-10 16:53:30 -05001587
Christoph Hellwiga76b5b02016-12-09 16:17:19 -08001588 /*
1589 * Try cloning first, this is supported by more file systems, and
1590 * more efficient if both clone and copy are supported (e.g. NFS).
1591 */
1592 if (file_in->f_op->clone_file_range) {
1593 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1594 file_out, pos_out, len);
1595 if (ret == 0) {
1596 ret = len;
1597 goto done;
1598 }
1599 }
1600
1601 if (file_out->f_op->copy_file_range) {
Anna Schumakereac70052015-11-10 16:53:33 -05001602 ret = file_out->f_op->copy_file_range(file_in, pos_in, file_out,
1603 pos_out, len, flags);
Christoph Hellwiga76b5b02016-12-09 16:17:19 -08001604 if (ret != -EOPNOTSUPP)
1605 goto done;
1606 }
Anna Schumakereac70052015-11-10 16:53:33 -05001607
Christoph Hellwiga76b5b02016-12-09 16:17:19 -08001608 ret = do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1609 len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
1610
1611done:
Zach Brown29732932015-11-10 16:53:30 -05001612 if (ret > 0) {
1613 fsnotify_access(file_in);
1614 add_rchar(current, ret);
1615 fsnotify_modify(file_out);
1616 add_wchar(current, ret);
1617 }
Christoph Hellwiga76b5b02016-12-09 16:17:19 -08001618
Zach Brown29732932015-11-10 16:53:30 -05001619 inc_syscr(current);
1620 inc_syscw(current);
1621
Amir Goldsteinbfe219d2017-01-31 10:34:57 +02001622 file_end_write(file_out);
Zach Brown29732932015-11-10 16:53:30 -05001623
1624 return ret;
1625}
1626EXPORT_SYMBOL(vfs_copy_file_range);
1627
1628SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1629 int, fd_out, loff_t __user *, off_out,
1630 size_t, len, unsigned int, flags)
1631{
1632 loff_t pos_in;
1633 loff_t pos_out;
1634 struct fd f_in;
1635 struct fd f_out;
1636 ssize_t ret = -EBADF;
1637
1638 f_in = fdget(fd_in);
1639 if (!f_in.file)
1640 goto out2;
1641
1642 f_out = fdget(fd_out);
1643 if (!f_out.file)
1644 goto out1;
1645
1646 ret = -EFAULT;
1647 if (off_in) {
1648 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1649 goto out;
1650 } else {
1651 pos_in = f_in.file->f_pos;
1652 }
1653
1654 if (off_out) {
1655 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1656 goto out;
1657 } else {
1658 pos_out = f_out.file->f_pos;
1659 }
1660
1661 ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1662 flags);
1663 if (ret > 0) {
1664 pos_in += ret;
1665 pos_out += ret;
1666
1667 if (off_in) {
1668 if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1669 ret = -EFAULT;
1670 } else {
1671 f_in.file->f_pos = pos_in;
1672 }
1673
1674 if (off_out) {
1675 if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1676 ret = -EFAULT;
1677 } else {
1678 f_out.file->f_pos = pos_out;
1679 }
1680 }
1681
1682out:
1683 fdput(f_out);
1684out1:
1685 fdput(f_in);
1686out2:
1687 return ret;
1688}
Christoph Hellwig04b38d62015-12-03 12:59:50 +01001689
1690static int clone_verify_area(struct file *file, loff_t pos, u64 len, bool write)
1691{
1692 struct inode *inode = file_inode(file);
1693
1694 if (unlikely(pos < 0))
1695 return -EINVAL;
1696
1697 if (unlikely((loff_t) (pos + len) < 0))
1698 return -EINVAL;
1699
1700 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
1701 loff_t end = len ? pos + len - 1 : OFFSET_MAX;
1702 int retval;
1703
1704 retval = locks_mandatory_area(inode, file, pos, end,
1705 write ? F_WRLCK : F_RDLCK);
1706 if (retval < 0)
1707 return retval;
1708 }
1709
1710 return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
1711}
Darrick J. Wong850fc692018-10-30 10:40:55 +11001712/*
1713 * Ensure that we don't remap a partial EOF block in the middle of something
1714 * else. Assume that the offsets have already been checked for block
1715 * alignment.
1716 *
1717 * For deduplication we always scale down to the previous block because we
1718 * can't meaningfully compare post-EOF contents.
1719 *
1720 * For clone we only link a partial EOF block above the destination file's EOF.
1721 */
1722static int generic_remap_check_len(struct inode *inode_in,
1723 struct inode *inode_out,
1724 loff_t pos_out,
1725 u64 *len,
1726 bool is_dedupe)
1727{
1728 u64 blkmask = i_blocksize(inode_in) - 1;
1729
1730 if ((*len & blkmask) == 0)
1731 return 0;
1732
1733 if (is_dedupe)
1734 *len &= ~blkmask;
1735 else if (pos_out + *len < i_size_read(inode_out))
1736 return -EINVAL;
1737
1738 return 0;
1739}
Christoph Hellwig04b38d62015-12-03 12:59:50 +01001740
Darrick J. Wong876bec6f2016-12-09 16:18:30 -08001741/*
1742 * Check that the two inodes are eligible for cloning, the ranges make
1743 * sense, and then flush all dirty data. Caller must ensure that the
1744 * inodes have been locked against any other modifications.
Darrick J. Wong22725ce2016-12-19 15:13:26 -08001745 *
1746 * Returns: 0 for "nothing to clone", 1 for "something to clone", or
1747 * the usual negative error code.
Darrick J. Wong876bec6f2016-12-09 16:18:30 -08001748 */
1749int vfs_clone_file_prep_inodes(struct inode *inode_in, loff_t pos_in,
1750 struct inode *inode_out, loff_t pos_out,
1751 u64 *len, bool is_dedupe)
1752{
1753 loff_t bs = inode_out->i_sb->s_blocksize;
1754 loff_t blen;
1755 loff_t isize;
1756 bool same_inode = (inode_in == inode_out);
1757 int ret;
1758
1759 /* Don't touch certain kinds of inodes */
1760 if (IS_IMMUTABLE(inode_out))
1761 return -EPERM;
1762
1763 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
1764 return -ETXTBSY;
1765
1766 /* Don't reflink dirs, pipes, sockets... */
1767 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1768 return -EISDIR;
1769 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1770 return -EINVAL;
1771
1772 /* Are we going all the way to the end? */
1773 isize = i_size_read(inode_in);
Darrick J. Wong22725ce2016-12-19 15:13:26 -08001774 if (isize == 0)
Darrick J. Wong876bec6f2016-12-09 16:18:30 -08001775 return 0;
Darrick J. Wong876bec6f2016-12-09 16:18:30 -08001776
1777 /* Zero length dedupe exits immediately; reflink goes to EOF. */
1778 if (*len == 0) {
Darrick J. Wong22725ce2016-12-19 15:13:26 -08001779 if (is_dedupe || pos_in == isize)
Darrick J. Wong876bec6f2016-12-09 16:18:30 -08001780 return 0;
Darrick J. Wong22725ce2016-12-19 15:13:26 -08001781 if (pos_in > isize)
1782 return -EINVAL;
Darrick J. Wong876bec6f2016-12-09 16:18:30 -08001783 *len = isize - pos_in;
1784 }
1785
1786 /* Ensure offsets don't wrap and the input is inside i_size */
1787 if (pos_in + *len < pos_in || pos_out + *len < pos_out ||
1788 pos_in + *len > isize)
1789 return -EINVAL;
1790
1791 /* Don't allow dedupe past EOF in the dest file */
1792 if (is_dedupe) {
1793 loff_t disize;
1794
1795 disize = i_size_read(inode_out);
1796 if (pos_out >= disize || pos_out + *len > disize)
1797 return -EINVAL;
1798 }
1799
1800 /* If we're linking to EOF, continue to the block boundary. */
1801 if (pos_in + *len == isize)
1802 blen = ALIGN(isize, bs) - pos_in;
1803 else
1804 blen = *len;
1805
1806 /* Only reflink if we're aligned to block boundaries */
1807 if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_in + blen, bs) ||
1808 !IS_ALIGNED(pos_out, bs) || !IS_ALIGNED(pos_out + blen, bs))
1809 return -EINVAL;
1810
1811 /* Don't allow overlapped reflink within the same file */
1812 if (same_inode) {
1813 if (pos_out + blen > pos_in && pos_out < pos_in + blen)
1814 return -EINVAL;
1815 }
1816
1817 /* Wait for the completion of any pending IOs on both files */
1818 inode_dio_wait(inode_in);
1819 if (!same_inode)
1820 inode_dio_wait(inode_out);
1821
1822 ret = filemap_write_and_wait_range(inode_in->i_mapping,
1823 pos_in, pos_in + *len - 1);
1824 if (ret)
1825 return ret;
1826
1827 ret = filemap_write_and_wait_range(inode_out->i_mapping,
1828 pos_out, pos_out + *len - 1);
1829 if (ret)
1830 return ret;
1831
1832 /*
1833 * Check that the extents are the same.
1834 */
1835 if (is_dedupe) {
1836 bool is_same = false;
1837
1838 ret = vfs_dedupe_file_range_compare(inode_in, pos_in,
1839 inode_out, pos_out, *len, &is_same);
1840 if (ret)
1841 return ret;
1842 if (!is_same)
1843 return -EBADE;
1844 }
1845
Darrick J. Wong850fc692018-10-30 10:40:55 +11001846 ret = generic_remap_check_len(inode_in, inode_out, pos_out, len,
1847 is_dedupe);
1848 if (ret)
1849 return ret;
1850
Darrick J. Wong22725ce2016-12-19 15:13:26 -08001851 return 1;
Darrick J. Wong876bec6f2016-12-09 16:18:30 -08001852}
1853EXPORT_SYMBOL(vfs_clone_file_prep_inodes);
1854
Amir Goldstein8a6cee32018-10-22 20:56:46 +03001855int do_clone_file_range(struct file *file_in, loff_t pos_in,
1856 struct file *file_out, loff_t pos_out, u64 len)
Christoph Hellwig04b38d62015-12-03 12:59:50 +01001857{
1858 struct inode *inode_in = file_inode(file_in);
1859 struct inode *inode_out = file_inode(file_out);
1860 int ret;
1861
Amir Goldsteinb335e9d2016-10-26 22:34:01 +03001862 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1863 return -EISDIR;
1864 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1865 return -EINVAL;
1866
Amir Goldstein913b86e92016-09-23 11:38:10 +03001867 /*
1868 * FICLONE/FICLONERANGE ioctls enforce that src and dest files are on
1869 * the same mount. Practically, they only need to be on the same file
1870 * system.
1871 */
1872 if (inode_in->i_sb != inode_out->i_sb)
Christoph Hellwig04b38d62015-12-03 12:59:50 +01001873 return -EXDEV;
1874
Christoph Hellwig04b38d62015-12-03 12:59:50 +01001875 if (!(file_in->f_mode & FMODE_READ) ||
1876 !(file_out->f_mode & FMODE_WRITE) ||
Christoph Hellwig0fcbf992016-02-26 18:53:12 +01001877 (file_out->f_flags & O_APPEND))
Christoph Hellwig04b38d62015-12-03 12:59:50 +01001878 return -EBADF;
1879
Christoph Hellwig0fcbf992016-02-26 18:53:12 +01001880 if (!file_in->f_op->clone_file_range)
1881 return -EOPNOTSUPP;
1882
Christoph Hellwig04b38d62015-12-03 12:59:50 +01001883 ret = clone_verify_area(file_in, pos_in, len, false);
1884 if (ret)
1885 return ret;
1886
1887 ret = clone_verify_area(file_out, pos_out, len, true);
1888 if (ret)
1889 return ret;
1890
1891 if (pos_in + len > i_size_read(inode_in))
1892 return -EINVAL;
1893
Christoph Hellwig04b38d62015-12-03 12:59:50 +01001894 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1895 file_out, pos_out, len);
1896 if (!ret) {
1897 fsnotify_access(file_in);
1898 fsnotify_modify(file_out);
1899 }
1900
Christoph Hellwig04b38d62015-12-03 12:59:50 +01001901 return ret;
1902}
Amir Goldstein8a6cee32018-10-22 20:56:46 +03001903EXPORT_SYMBOL(do_clone_file_range);
1904
1905int vfs_clone_file_range(struct file *file_in, loff_t pos_in,
1906 struct file *file_out, loff_t pos_out, u64 len)
1907{
1908 int ret;
1909
1910 file_start_write(file_out);
1911 ret = do_clone_file_range(file_in, pos_in, file_out, pos_out, len);
1912 file_end_write(file_out);
1913
1914 return ret;
1915}
Christoph Hellwig04b38d62015-12-03 12:59:50 +01001916EXPORT_SYMBOL(vfs_clone_file_range);
Darrick J. Wong54dbc152015-12-19 00:55:59 -08001917
Darrick J. Wong1aca2002019-08-11 15:52:25 -07001918/* Read a page's worth of file data into the page cache. */
Darrick J. Wong876bec6f2016-12-09 16:18:30 -08001919static struct page *vfs_dedupe_get_page(struct inode *inode, loff_t offset)
1920{
1921 struct address_space *mapping;
1922 struct page *page;
1923 pgoff_t n;
1924
1925 n = offset >> PAGE_SHIFT;
1926 mapping = inode->i_mapping;
1927 page = read_mapping_page(mapping, n, NULL);
1928 if (IS_ERR(page))
1929 return page;
1930 if (!PageUptodate(page)) {
1931 put_page(page);
1932 return ERR_PTR(-EIO);
1933 }
Darrick J. Wong876bec6f2016-12-09 16:18:30 -08001934 return page;
1935}
1936
1937/*
Darrick J. Wong1aca2002019-08-11 15:52:25 -07001938 * Lock two pages, ensuring that we lock in offset order if the pages are from
1939 * the same file.
1940 */
1941static void vfs_lock_two_pages(struct page *page1, struct page *page2)
1942{
1943 /* Always lock in order of increasing index. */
1944 if (page1->index > page2->index)
1945 swap(page1, page2);
1946
1947 lock_page(page1);
1948 if (page1 != page2)
1949 lock_page(page2);
1950}
1951
1952/* Unlock two pages, being careful not to unlock the same page twice. */
1953static void vfs_unlock_two_pages(struct page *page1, struct page *page2)
1954{
1955 unlock_page(page1);
1956 if (page1 != page2)
1957 unlock_page(page2);
1958}
1959
1960/*
Darrick J. Wong876bec6f2016-12-09 16:18:30 -08001961 * Compare extents of two files to see if they are the same.
1962 * Caller must have locked both inodes to prevent write races.
1963 */
1964int vfs_dedupe_file_range_compare(struct inode *src, loff_t srcoff,
1965 struct inode *dest, loff_t destoff,
1966 loff_t len, bool *is_same)
1967{
1968 loff_t src_poff;
1969 loff_t dest_poff;
1970 void *src_addr;
1971 void *dest_addr;
1972 struct page *src_page;
1973 struct page *dest_page;
1974 loff_t cmp_len;
1975 bool same;
1976 int error;
1977
1978 error = -EINVAL;
1979 same = true;
1980 while (len) {
1981 src_poff = srcoff & (PAGE_SIZE - 1);
1982 dest_poff = destoff & (PAGE_SIZE - 1);
1983 cmp_len = min(PAGE_SIZE - src_poff,
1984 PAGE_SIZE - dest_poff);
1985 cmp_len = min(cmp_len, len);
1986 if (cmp_len <= 0)
1987 goto out_error;
1988
1989 src_page = vfs_dedupe_get_page(src, srcoff);
1990 if (IS_ERR(src_page)) {
1991 error = PTR_ERR(src_page);
1992 goto out_error;
1993 }
1994 dest_page = vfs_dedupe_get_page(dest, destoff);
1995 if (IS_ERR(dest_page)) {
1996 error = PTR_ERR(dest_page);
Darrick J. Wong876bec6f2016-12-09 16:18:30 -08001997 put_page(src_page);
1998 goto out_error;
1999 }
Darrick J. Wong1aca2002019-08-11 15:52:25 -07002000
2001 vfs_lock_two_pages(src_page, dest_page);
2002
2003 /*
2004 * Now that we've locked both pages, make sure they're still
2005 * mapped to the file data we're interested in. If not,
2006 * someone is invalidating pages on us and we lose.
2007 */
2008 if (!PageUptodate(src_page) || !PageUptodate(dest_page) ||
2009 src_page->mapping != src->i_mapping ||
2010 dest_page->mapping != dest->i_mapping) {
2011 same = false;
2012 goto unlock;
2013 }
2014
Darrick J. Wong876bec6f2016-12-09 16:18:30 -08002015 src_addr = kmap_atomic(src_page);
2016 dest_addr = kmap_atomic(dest_page);
2017
2018 flush_dcache_page(src_page);
2019 flush_dcache_page(dest_page);
2020
2021 if (memcmp(src_addr + src_poff, dest_addr + dest_poff, cmp_len))
2022 same = false;
2023
2024 kunmap_atomic(dest_addr);
2025 kunmap_atomic(src_addr);
Darrick J. Wong1aca2002019-08-11 15:52:25 -07002026unlock:
2027 vfs_unlock_two_pages(src_page, dest_page);
Darrick J. Wong876bec6f2016-12-09 16:18:30 -08002028 put_page(dest_page);
2029 put_page(src_page);
2030
2031 if (!same)
2032 break;
2033
2034 srcoff += cmp_len;
2035 destoff += cmp_len;
2036 len -= cmp_len;
2037 }
2038
2039 *is_same = same;
2040 return 0;
2041
2042out_error:
2043 return error;
2044}
2045EXPORT_SYMBOL(vfs_dedupe_file_range_compare);
2046
Darrick J. Wong54dbc152015-12-19 00:55:59 -08002047int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
2048{
2049 struct file_dedupe_range_info *info;
2050 struct inode *src = file_inode(file);
2051 u64 off;
2052 u64 len;
2053 int i;
2054 int ret;
2055 bool is_admin = capable(CAP_SYS_ADMIN);
2056 u16 count = same->dest_count;
2057 struct file *dst_file;
2058 loff_t dst_off;
2059 ssize_t deduped;
2060
2061 if (!(file->f_mode & FMODE_READ))
2062 return -EINVAL;
2063
2064 if (same->reserved1 || same->reserved2)
2065 return -EINVAL;
2066
2067 off = same->src_offset;
2068 len = same->src_length;
2069
2070 ret = -EISDIR;
2071 if (S_ISDIR(src->i_mode))
2072 goto out;
2073
2074 ret = -EINVAL;
2075 if (!S_ISREG(src->i_mode))
2076 goto out;
2077
2078 ret = clone_verify_area(file, off, len, false);
2079 if (ret < 0)
2080 goto out;
2081 ret = 0;
2082
Darrick J. Wong22725ce2016-12-19 15:13:26 -08002083 if (off + len > i_size_read(src))
2084 return -EINVAL;
2085
Darrick J. Wong54dbc152015-12-19 00:55:59 -08002086 /* pre-format output fields to sane values */
2087 for (i = 0; i < count; i++) {
2088 same->info[i].bytes_deduped = 0ULL;
2089 same->info[i].status = FILE_DEDUPE_RANGE_SAME;
2090 }
2091
2092 for (i = 0, info = same->info; i < count; i++, info++) {
2093 struct inode *dst;
2094 struct fd dst_fd = fdget(info->dest_fd);
2095
2096 dst_file = dst_fd.file;
2097 if (!dst_file) {
2098 info->status = -EBADF;
2099 goto next_loop;
2100 }
2101 dst = file_inode(dst_file);
2102
2103 ret = mnt_want_write_file(dst_file);
2104 if (ret) {
2105 info->status = ret;
2106 goto next_loop;
2107 }
2108
2109 dst_off = info->dest_offset;
2110 ret = clone_verify_area(dst_file, dst_off, len, true);
2111 if (ret < 0) {
2112 info->status = ret;
2113 goto next_file;
2114 }
2115 ret = 0;
2116
2117 if (info->reserved) {
2118 info->status = -EINVAL;
2119 } else if (!(is_admin || (dst_file->f_mode & FMODE_WRITE))) {
2120 info->status = -EINVAL;
2121 } else if (file->f_path.mnt != dst_file->f_path.mnt) {
2122 info->status = -EXDEV;
2123 } else if (S_ISDIR(dst->i_mode)) {
2124 info->status = -EISDIR;
2125 } else if (dst_file->f_op->dedupe_file_range == NULL) {
2126 info->status = -EINVAL;
2127 } else {
2128 deduped = dst_file->f_op->dedupe_file_range(file, off,
2129 len, dst_file,
2130 info->dest_offset);
2131 if (deduped == -EBADE)
2132 info->status = FILE_DEDUPE_RANGE_DIFFERS;
2133 else if (deduped < 0)
2134 info->status = deduped;
2135 else
2136 info->bytes_deduped += deduped;
2137 }
2138
2139next_file:
2140 mnt_drop_write_file(dst_file);
2141next_loop:
2142 fdput(dst_fd);
Darrick J. Wonge62e5602016-01-22 16:58:28 -08002143
2144 if (fatal_signal_pending(current))
2145 goto out;
Darrick J. Wong54dbc152015-12-19 00:55:59 -08002146 }
2147
2148out:
2149 return ret;
2150}
2151EXPORT_SYMBOL(vfs_dedupe_file_range);