blob: 005d09cf3fa879fd2fab8fd6a95433fefa7abc4d [file] [log] [blame]
Jens Axboe5274f052006-03-30 15:15:30 +02001/*
2 * "splice": joining two ropes together by interweaving their strands.
3 *
4 * This is the "extended pipe" functionality, where a pipe is used as
5 * an arbitrary in-memory buffer. Think of a pipe as a small kernel
6 * buffer that you can use to transfer data from one end to the other.
7 *
8 * The traditional unix read/write is extended with a "splice()" operation
9 * that transfers data buffers to or from a pipe buffer.
10 *
11 * Named by Larry McVoy, original implementation from Linus, extended by
Jens Axboec2058e02006-04-11 13:56:34 +020012 * Jens to support splicing to files, network, direct splicing, etc and
13 * fixing lots of bugs.
Jens Axboe5274f052006-03-30 15:15:30 +020014 *
Jens Axboe0fe23472006-09-04 15:41:16 +020015 * Copyright (C) 2005-2006 Jens Axboe <axboe@kernel.dk>
Jens Axboec2058e02006-04-11 13:56:34 +020016 * Copyright (C) 2005-2006 Linus Torvalds <torvalds@osdl.org>
17 * Copyright (C) 2006 Ingo Molnar <mingo@elte.hu>
Jens Axboe5274f052006-03-30 15:15:30 +020018 *
19 */
Christoph Hellwigbe297962016-11-01 07:40:16 -060020#include <linux/bvec.h>
Jens Axboe5274f052006-03-30 15:15:30 +020021#include <linux/fs.h>
22#include <linux/file.h>
23#include <linux/pagemap.h>
Jens Axboed6b29d72007-06-04 09:59:47 +020024#include <linux/splice.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080025#include <linux/memcontrol.h>
Jens Axboe5274f052006-03-30 15:15:30 +020026#include <linux/mm_inline.h>
Jens Axboe5abc97a2006-03-30 15:16:46 +020027#include <linux/swap.h>
Jens Axboe4f6f0bd2006-04-02 23:04:46 +020028#include <linux/writeback.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050029#include <linux/export.h>
Jens Axboe4f6f0bd2006-04-02 23:04:46 +020030#include <linux/syscalls.h>
Jens Axboe912d35f2006-04-26 10:59:21 +020031#include <linux/uio.h>
James Morris29ce2052007-07-13 11:44:32 +020032#include <linux/security.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090033#include <linux/gfp.h>
Eric Dumazet35f9c092012-04-05 03:05:35 +000034#include <linux/socket.h>
Al Viro76b021d2013-03-02 10:19:56 -050035#include <linux/compat.h>
Ingo Molnar174cd4b2017-02-02 19:15:33 +010036#include <linux/sched/signal.h>
37
Al Viro06ae43f2013-03-20 13:19:30 -040038#include "internal.h"
Jens Axboe5274f052006-03-30 15:15:30 +020039
Jens Axboe83f91352006-04-02 23:05:09 +020040/*
41 * Attempt to steal a page from a pipe buffer. This should perhaps go into
42 * a vm helper function, it's already simplified quite a bit by the
43 * addition of remove_mapping(). If success is returned, the caller may
44 * attempt to reuse this page for another destination.
45 */
Jens Axboe76ad4d12006-05-03 10:41:33 +020046static int page_cache_pipe_buf_steal(struct pipe_inode_info *pipe,
Jens Axboe5abc97a2006-03-30 15:16:46 +020047 struct pipe_buffer *buf)
48{
49 struct page *page = buf->page;
Jens Axboe9e94cd42006-06-20 15:01:12 +020050 struct address_space *mapping;
Jens Axboe5abc97a2006-03-30 15:16:46 +020051
Jens Axboe9e0267c2006-04-19 15:57:31 +020052 lock_page(page);
53
Jens Axboe9e94cd42006-06-20 15:01:12 +020054 mapping = page_mapping(page);
55 if (mapping) {
56 WARN_ON(!PageUptodate(page));
Jens Axboe5abc97a2006-03-30 15:16:46 +020057
Jens Axboe9e94cd42006-06-20 15:01:12 +020058 /*
59 * At least for ext2 with nobh option, we need to wait on
60 * writeback completing on this page, since we'll remove it
61 * from the pagecache. Otherwise truncate wont wait on the
62 * page, allowing the disk blocks to be reused by someone else
63 * before we actually wrote our data to them. fs corruption
64 * ensues.
65 */
66 wait_on_page_writeback(page);
Jens Axboead8d6f02006-04-02 23:10:32 +020067
David Howells266cf652009-04-03 16:42:36 +010068 if (page_has_private(page) &&
69 !try_to_release_page(page, GFP_KERNEL))
Jens Axboeca39d652008-05-20 21:27:41 +020070 goto out_unlock;
Jens Axboe4f6f0bd2006-04-02 23:04:46 +020071
Jens Axboe9e94cd42006-06-20 15:01:12 +020072 /*
73 * If we succeeded in removing the mapping, set LRU flag
74 * and return good.
75 */
76 if (remove_mapping(mapping, page)) {
77 buf->flags |= PIPE_BUF_FLAG_LRU;
78 return 0;
79 }
Jens Axboe9e0267c2006-04-19 15:57:31 +020080 }
Jens Axboe5abc97a2006-03-30 15:16:46 +020081
Jens Axboe9e94cd42006-06-20 15:01:12 +020082 /*
83 * Raced with truncate or failed to remove page from current
84 * address space, unlock and return failure.
85 */
Jens Axboeca39d652008-05-20 21:27:41 +020086out_unlock:
Jens Axboe9e94cd42006-06-20 15:01:12 +020087 unlock_page(page);
88 return 1;
Jens Axboe5abc97a2006-03-30 15:16:46 +020089}
90
Jens Axboe76ad4d12006-05-03 10:41:33 +020091static void page_cache_pipe_buf_release(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +020092 struct pipe_buffer *buf)
93{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +030094 put_page(buf->page);
Jens Axboe14328732006-05-03 10:35:26 +020095 buf->flags &= ~PIPE_BUF_FLAG_LRU;
Jens Axboe5274f052006-03-30 15:15:30 +020096}
97
Jens Axboe08457182007-06-12 20:51:32 +020098/*
99 * Check whether the contents of buf is OK to access. Since the content
100 * is a page cache page, IO may be in flight.
101 */
Jens Axboecac36bb02007-06-14 13:10:48 +0200102static int page_cache_pipe_buf_confirm(struct pipe_inode_info *pipe,
103 struct pipe_buffer *buf)
Jens Axboe5274f052006-03-30 15:15:30 +0200104{
105 struct page *page = buf->page;
Jens Axboe49d0b212006-04-10 09:04:41 +0200106 int err;
Jens Axboe5274f052006-03-30 15:15:30 +0200107
108 if (!PageUptodate(page)) {
Jens Axboe49d0b212006-04-10 09:04:41 +0200109 lock_page(page);
110
111 /*
112 * Page got truncated/unhashed. This will cause a 0-byte
Ingo Molnar73d62d82006-04-11 13:57:21 +0200113 * splice, if this is the first page.
Jens Axboe49d0b212006-04-10 09:04:41 +0200114 */
115 if (!page->mapping) {
116 err = -ENODATA;
117 goto error;
118 }
119
120 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +0200121 * Uh oh, read-error from disk.
Jens Axboe49d0b212006-04-10 09:04:41 +0200122 */
123 if (!PageUptodate(page)) {
124 err = -EIO;
125 goto error;
126 }
127
128 /*
Jens Axboef84d7512006-05-01 19:59:03 +0200129 * Page is ok afterall, we are done.
Jens Axboe49d0b212006-04-10 09:04:41 +0200130 */
Jens Axboe5274f052006-03-30 15:15:30 +0200131 unlock_page(page);
Jens Axboe5274f052006-03-30 15:15:30 +0200132 }
133
Jens Axboef84d7512006-05-01 19:59:03 +0200134 return 0;
Jens Axboe49d0b212006-04-10 09:04:41 +0200135error:
136 unlock_page(page);
Jens Axboef84d7512006-05-01 19:59:03 +0200137 return err;
Jens Axboe70524492006-04-11 15:51:17 +0200138}
139
Hugh Dickins708e3502011-07-25 17:12:32 -0700140const struct pipe_buf_operations page_cache_pipe_buf_ops = {
Jens Axboe5274f052006-03-30 15:15:30 +0200141 .can_merge = 0,
Jens Axboecac36bb02007-06-14 13:10:48 +0200142 .confirm = page_cache_pipe_buf_confirm,
Jens Axboe5274f052006-03-30 15:15:30 +0200143 .release = page_cache_pipe_buf_release,
Jens Axboe5abc97a2006-03-30 15:16:46 +0200144 .steal = page_cache_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200145 .get = generic_pipe_buf_get,
Jens Axboe5274f052006-03-30 15:15:30 +0200146};
147
Jens Axboe912d35f2006-04-26 10:59:21 +0200148static int user_page_pipe_buf_steal(struct pipe_inode_info *pipe,
149 struct pipe_buffer *buf)
150{
Jens Axboe7afa6fd2006-05-01 20:02:33 +0200151 if (!(buf->flags & PIPE_BUF_FLAG_GIFT))
152 return 1;
153
Jens Axboe14328732006-05-03 10:35:26 +0200154 buf->flags |= PIPE_BUF_FLAG_LRU;
Jens Axboe330ab712006-05-02 15:29:57 +0200155 return generic_pipe_buf_steal(pipe, buf);
Jens Axboe912d35f2006-04-26 10:59:21 +0200156}
157
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800158static const struct pipe_buf_operations user_page_pipe_buf_ops = {
Jens Axboe912d35f2006-04-26 10:59:21 +0200159 .can_merge = 0,
Jens Axboecac36bb02007-06-14 13:10:48 +0200160 .confirm = generic_pipe_buf_confirm,
Jens Axboe912d35f2006-04-26 10:59:21 +0200161 .release = page_cache_pipe_buf_release,
162 .steal = user_page_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200163 .get = generic_pipe_buf_get,
Jens Axboe912d35f2006-04-26 10:59:21 +0200164};
165
Namhyung Kim825cdcb2011-05-23 19:58:53 +0200166static void wakeup_pipe_readers(struct pipe_inode_info *pipe)
167{
168 smp_mb();
169 if (waitqueue_active(&pipe->wait))
170 wake_up_interruptible(&pipe->wait);
171 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
172}
173
Jens Axboe932cc6d2007-06-21 13:10:21 +0200174/**
175 * splice_to_pipe - fill passed data into a pipe
176 * @pipe: pipe to fill
177 * @spd: data to fill
178 *
179 * Description:
Randy Dunlap79685b82007-07-27 08:08:51 +0200180 * @spd contains a map of pages and len/offset tuples, along with
Jens Axboe932cc6d2007-06-21 13:10:21 +0200181 * the struct pipe_buf_operations associated with these pages. This
182 * function will link that data to the pipe.
183 *
Jens Axboe83f91352006-04-02 23:05:09 +0200184 */
Jens Axboed6b29d72007-06-04 09:59:47 +0200185ssize_t splice_to_pipe(struct pipe_inode_info *pipe,
186 struct splice_pipe_desc *spd)
Jens Axboe5274f052006-03-30 15:15:30 +0200187{
Jens Axboe00de00b2007-06-15 13:14:22 +0200188 unsigned int spd_pages = spd->nr_pages;
Al Viro8924fef2016-09-17 20:44:45 -0400189 int ret = 0, page_nr = 0;
Jens Axboe5274f052006-03-30 15:15:30 +0200190
Rabin Vincentd6785d92016-03-10 21:19:06 +0100191 if (!spd_pages)
192 return 0;
193
Al Viro8924fef2016-09-17 20:44:45 -0400194 if (unlikely(!pipe->readers)) {
195 send_sig(SIGPIPE, current, 0);
196 ret = -EPIPE;
197 goto out;
Jens Axboe5274f052006-03-30 15:15:30 +0200198 }
199
Al Viro8924fef2016-09-17 20:44:45 -0400200 while (pipe->nrbufs < pipe->buffers) {
201 int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
202 struct pipe_buffer *buf = pipe->bufs + newbuf;
Jens Axboe5274f052006-03-30 15:15:30 +0200203
Al Viro8924fef2016-09-17 20:44:45 -0400204 buf->page = spd->pages[page_nr];
205 buf->offset = spd->partial[page_nr].offset;
206 buf->len = spd->partial[page_nr].len;
207 buf->private = spd->partial[page_nr].private;
208 buf->ops = spd->ops;
Miklos Szeredi5a81e6a2017-02-16 17:49:02 +0100209 buf->flags = 0;
Jens Axboe5274f052006-03-30 15:15:30 +0200210
Al Viro8924fef2016-09-17 20:44:45 -0400211 pipe->nrbufs++;
212 page_nr++;
213 ret += buf->len;
214
215 if (!--spd->nr_pages)
216 break;
217 }
218
219 if (!ret)
220 ret = -EAGAIN;
221
222out:
Jens Axboe00de00b2007-06-15 13:14:22 +0200223 while (page_nr < spd_pages)
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800224 spd->spd_release(spd, page_nr++);
Jens Axboe5274f052006-03-30 15:15:30 +0200225
226 return ret;
227}
Hannes Frederic Sowa2b514572015-05-21 17:00:01 +0200228EXPORT_SYMBOL_GPL(splice_to_pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200229
Al Viro79fddc42016-09-17 22:38:20 -0400230ssize_t add_to_pipe(struct pipe_inode_info *pipe, struct pipe_buffer *buf)
231{
232 int ret;
233
234 if (unlikely(!pipe->readers)) {
235 send_sig(SIGPIPE, current, 0);
236 ret = -EPIPE;
237 } else if (pipe->nrbufs == pipe->buffers) {
238 ret = -EAGAIN;
239 } else {
240 int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
241 pipe->bufs[newbuf] = *buf;
242 pipe->nrbufs++;
243 return buf->len;
244 }
Miklos Szeredia7796382016-09-27 10:45:12 +0200245 pipe_buf_release(pipe, buf);
Al Viro79fddc42016-09-17 22:38:20 -0400246 return ret;
247}
248EXPORT_SYMBOL(add_to_pipe);
249
Jens Axboe35f3d142010-05-20 10:43:18 +0200250/*
251 * Check if we need to grow the arrays holding pages and partial page
252 * descriptions.
253 */
Eric Dumazet047fe362012-06-12 15:24:40 +0200254int splice_grow_spd(const struct pipe_inode_info *pipe, struct splice_pipe_desc *spd)
Jens Axboe35f3d142010-05-20 10:43:18 +0200255{
Mark Rutland6aa7de02017-10-23 14:07:29 -0700256 unsigned int buffers = READ_ONCE(pipe->buffers);
Eric Dumazet047fe362012-06-12 15:24:40 +0200257
258 spd->nr_pages_max = buffers;
259 if (buffers <= PIPE_DEF_BUFFERS)
Jens Axboe35f3d142010-05-20 10:43:18 +0200260 return 0;
261
Eric Dumazet047fe362012-06-12 15:24:40 +0200262 spd->pages = kmalloc(buffers * sizeof(struct page *), GFP_KERNEL);
263 spd->partial = kmalloc(buffers * sizeof(struct partial_page), GFP_KERNEL);
Jens Axboe35f3d142010-05-20 10:43:18 +0200264
265 if (spd->pages && spd->partial)
266 return 0;
267
268 kfree(spd->pages);
269 kfree(spd->partial);
270 return -ENOMEM;
271}
272
Eric Dumazet047fe362012-06-12 15:24:40 +0200273void splice_shrink_spd(struct splice_pipe_desc *spd)
Jens Axboe35f3d142010-05-20 10:43:18 +0200274{
Eric Dumazet047fe362012-06-12 15:24:40 +0200275 if (spd->nr_pages_max <= PIPE_DEF_BUFFERS)
Jens Axboe35f3d142010-05-20 10:43:18 +0200276 return;
277
278 kfree(spd->pages);
279 kfree(spd->partial);
280}
281
Jens Axboe83f91352006-04-02 23:05:09 +0200282/**
283 * generic_file_splice_read - splice data from file to a pipe
284 * @in: file to splice from
Jens Axboe932cc6d2007-06-21 13:10:21 +0200285 * @ppos: position in @in
Jens Axboe83f91352006-04-02 23:05:09 +0200286 * @pipe: pipe to splice to
287 * @len: number of bytes to splice
288 * @flags: splice modifier flags
289 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200290 * Description:
291 * Will read pages from given file and fill them into a pipe. Can be
Al Viro82c156f2016-09-22 23:35:42 -0400292 * used as long as it has more or less sane ->read_iter().
Jens Axboe932cc6d2007-06-21 13:10:21 +0200293 *
Jens Axboe83f91352006-04-02 23:05:09 +0200294 */
Jens Axboecbb7e572006-04-11 14:57:50 +0200295ssize_t generic_file_splice_read(struct file *in, loff_t *ppos,
296 struct pipe_inode_info *pipe, size_t len,
297 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200298{
Al Viro82c156f2016-09-22 23:35:42 -0400299 struct iov_iter to;
300 struct kiocb kiocb;
Al Viro82c156f2016-09-22 23:35:42 -0400301 int idx, ret;
Boaz Harroshbe64f882015-04-15 16:15:17 -0700302
Al Viro82c156f2016-09-22 23:35:42 -0400303 iov_iter_pipe(&to, ITER_PIPE | READ, pipe, len);
304 idx = to.idx;
305 init_sync_kiocb(&kiocb, in);
306 kiocb.ki_pos = *ppos;
Miklos Szeredibb7462b2017-02-20 16:51:23 +0100307 ret = call_read_iter(in, &kiocb, &to);
Miklos Szeredi723590e2009-08-15 08:43:22 +0200308 if (ret > 0) {
Al Viro82c156f2016-09-22 23:35:42 -0400309 *ppos = kiocb.ki_pos;
Miklos Szeredi723590e2009-08-15 08:43:22 +0200310 file_accessed(in);
Al Viro82c156f2016-09-22 23:35:42 -0400311 } else if (ret < 0) {
Al Viroc3a69022016-10-10 13:26:27 -0400312 to.idx = idx;
313 to.iov_offset = 0;
314 iov_iter_advance(&to, 0); /* to free what was emitted */
Al Viro82c156f2016-09-22 23:35:42 -0400315 /*
316 * callers of ->splice_read() expect -EAGAIN on
317 * "can't put anything in there", rather than -EFAULT.
318 */
319 if (ret == -EFAULT)
320 ret = -EAGAIN;
Miklos Szeredi723590e2009-08-15 08:43:22 +0200321 }
Jens Axboe5274f052006-03-30 15:15:30 +0200322
323 return ret;
324}
Jens Axboe059a8f32006-04-02 23:06:05 +0200325EXPORT_SYMBOL(generic_file_splice_read);
326
Al Viro241699c2016-09-22 16:33:12 -0400327const struct pipe_buf_operations default_pipe_buf_ops = {
Miklos Szeredi68181732009-05-07 15:37:36 +0200328 .can_merge = 0,
Miklos Szeredi68181732009-05-07 15:37:36 +0200329 .confirm = generic_pipe_buf_confirm,
330 .release = generic_pipe_buf_release,
331 .steal = generic_pipe_buf_steal,
332 .get = generic_pipe_buf_get,
333};
334
Miklos Szeredi28a625c2014-01-22 19:36:57 +0100335static int generic_pipe_buf_nosteal(struct pipe_inode_info *pipe,
336 struct pipe_buffer *buf)
337{
338 return 1;
339}
340
341/* Pipe buffer operations for a socket and similar. */
342const struct pipe_buf_operations nosteal_pipe_buf_ops = {
343 .can_merge = 0,
Miklos Szeredi28a625c2014-01-22 19:36:57 +0100344 .confirm = generic_pipe_buf_confirm,
345 .release = generic_pipe_buf_release,
346 .steal = generic_pipe_buf_nosteal,
347 .get = generic_pipe_buf_get,
348};
349EXPORT_SYMBOL(nosteal_pipe_buf_ops);
350
Al Viro523ac9a2016-09-23 15:34:57 -0400351static ssize_t kernel_readv(struct file *file, const struct kvec *vec,
Miklos Szeredi68181732009-05-07 15:37:36 +0200352 unsigned long vlen, loff_t offset)
353{
354 mm_segment_t old_fs;
355 loff_t pos = offset;
356 ssize_t res;
357
358 old_fs = get_fs();
359 set_fs(get_ds());
360 /* The cast to a user pointer is valid due to the set_fs() */
Christoph Hellwig793b80e2016-03-03 16:03:58 +0100361 res = vfs_readv(file, (const struct iovec __user *)vec, vlen, &pos, 0);
Miklos Szeredi68181732009-05-07 15:37:36 +0200362 set_fs(old_fs);
363
364 return res;
365}
366
Al Viro82c156f2016-09-22 23:35:42 -0400367static ssize_t default_file_splice_read(struct file *in, loff_t *ppos,
Miklos Szeredi68181732009-05-07 15:37:36 +0200368 struct pipe_inode_info *pipe, size_t len,
369 unsigned int flags)
370{
Al Viro523ac9a2016-09-23 15:34:57 -0400371 struct kvec *vec, __vec[PIPE_DEF_BUFFERS];
372 struct iov_iter to;
373 struct page **pages;
Miklos Szeredi68181732009-05-07 15:37:36 +0200374 unsigned int nr_pages;
Al Viro13c0f522016-12-10 13:20:53 -0500375 size_t offset, base, copied = 0;
Miklos Szeredi68181732009-05-07 15:37:36 +0200376 ssize_t res;
Miklos Szeredi68181732009-05-07 15:37:36 +0200377 int i;
Miklos Szeredi68181732009-05-07 15:37:36 +0200378
Al Viro523ac9a2016-09-23 15:34:57 -0400379 if (pipe->nrbufs == pipe->buffers)
380 return -EAGAIN;
381
382 /*
383 * Try to keep page boundaries matching to source pagecache ones -
384 * it probably won't be much help, but...
385 */
386 offset = *ppos & ~PAGE_MASK;
387
388 iov_iter_pipe(&to, ITER_PIPE | READ, pipe, len + offset);
389
Al Viro13c0f522016-12-10 13:20:53 -0500390 res = iov_iter_get_pages_alloc(&to, &pages, len + offset, &base);
Al Viro523ac9a2016-09-23 15:34:57 -0400391 if (res <= 0)
Jens Axboe35f3d142010-05-20 10:43:18 +0200392 return -ENOMEM;
393
Al Viro13c0f522016-12-10 13:20:53 -0500394 nr_pages = DIV_ROUND_UP(res + base, PAGE_SIZE);
Al Viro523ac9a2016-09-23 15:34:57 -0400395
Jens Axboe35f3d142010-05-20 10:43:18 +0200396 vec = __vec;
Al Viro523ac9a2016-09-23 15:34:57 -0400397 if (nr_pages > PIPE_DEF_BUFFERS) {
398 vec = kmalloc(nr_pages * sizeof(struct kvec), GFP_KERNEL);
399 if (unlikely(!vec)) {
400 res = -ENOMEM;
401 goto out;
402 }
Jens Axboe35f3d142010-05-20 10:43:18 +0200403 }
404
Al Viro523ac9a2016-09-23 15:34:57 -0400405 pipe->bufs[to.idx].offset = offset;
406 pipe->bufs[to.idx].len -= offset;
Miklos Szeredi68181732009-05-07 15:37:36 +0200407
Al Viro523ac9a2016-09-23 15:34:57 -0400408 for (i = 0; i < nr_pages; i++) {
409 size_t this_len = min_t(size_t, len, PAGE_SIZE - offset);
410 vec[i].iov_base = page_address(pages[i]) + offset;
Miklos Szeredi68181732009-05-07 15:37:36 +0200411 vec[i].iov_len = this_len;
Miklos Szeredi68181732009-05-07 15:37:36 +0200412 len -= this_len;
413 offset = 0;
414 }
415
Al Viro523ac9a2016-09-23 15:34:57 -0400416 res = kernel_readv(in, vec, nr_pages, *ppos);
417 if (res > 0) {
418 copied = res;
Miklos Szeredi68181732009-05-07 15:37:36 +0200419 *ppos += res;
Al Viro523ac9a2016-09-23 15:34:57 -0400420 }
Miklos Szeredi68181732009-05-07 15:37:36 +0200421
Jens Axboe35f3d142010-05-20 10:43:18 +0200422 if (vec != __vec)
423 kfree(vec);
Al Viro523ac9a2016-09-23 15:34:57 -0400424out:
425 for (i = 0; i < nr_pages; i++)
426 put_page(pages[i]);
427 kvfree(pages);
428 iov_iter_advance(&to, copied); /* truncates and discards */
Miklos Szeredi68181732009-05-07 15:37:36 +0200429 return res;
Miklos Szeredi68181732009-05-07 15:37:36 +0200430}
Miklos Szeredi68181732009-05-07 15:37:36 +0200431
Jens Axboe5274f052006-03-30 15:15:30 +0200432/*
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200433 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
Jens Axboe016b6612006-04-25 15:42:00 +0200434 * using sendpage(). Return the number of bytes sent.
Jens Axboe5274f052006-03-30 15:15:30 +0200435 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200436static int pipe_to_sendpage(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +0200437 struct pipe_buffer *buf, struct splice_desc *sd)
438{
Jens Axboe6a14b902007-06-14 13:08:55 +0200439 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200440 loff_t pos = sd->pos;
Michał Mirosława8adbe32010-12-17 08:56:44 +0100441 int more;
Jens Axboe5274f052006-03-30 15:15:30 +0200442
Al Viro72c2d532013-09-22 16:27:52 -0400443 if (!likely(file->f_op->sendpage))
Michał Mirosława8adbe32010-12-17 08:56:44 +0100444 return -EINVAL;
Jens Axboe5274f052006-03-30 15:15:30 +0200445
Eric Dumazet35f9c092012-04-05 03:05:35 +0000446 more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0;
Eric Dumazetae62ca72013-01-06 18:21:49 +0000447
448 if (sd->len < sd->total_len && pipe->nrbufs > 1)
Eric Dumazet35f9c092012-04-05 03:05:35 +0000449 more |= MSG_SENDPAGE_NOTLAST;
Eric Dumazetae62ca72013-01-06 18:21:49 +0000450
Michał Mirosława8adbe32010-12-17 08:56:44 +0100451 return file->f_op->sendpage(file, buf->page, buf->offset,
452 sd->len, &pos, more);
Jens Axboe5274f052006-03-30 15:15:30 +0200453}
454
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200455static void wakeup_pipe_writers(struct pipe_inode_info *pipe)
456{
457 smp_mb();
458 if (waitqueue_active(&pipe->wait))
459 wake_up_interruptible(&pipe->wait);
460 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
461}
462
463/**
464 * splice_from_pipe_feed - feed available data from a pipe to a file
465 * @pipe: pipe to splice from
466 * @sd: information to @actor
467 * @actor: handler that splices the data
468 *
469 * Description:
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200470 * This function loops over the pipe and calls @actor to do the
471 * actual moving of a single struct pipe_buffer to the desired
472 * destination. It returns when there's no more buffers left in
473 * the pipe or if the requested number of bytes (@sd->total_len)
474 * have been copied. It returns a positive number (one) if the
475 * pipe needs to be filled with more data, zero if the required
476 * number of bytes have been copied and -errno on error.
477 *
478 * This, together with splice_from_pipe_{begin,end,next}, may be
479 * used to implement the functionality of __splice_from_pipe() when
480 * locking is required around copying the pipe buffers to the
481 * destination.
482 */
Al Viro96f9bc82014-04-05 04:35:49 -0400483static int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd,
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200484 splice_actor *actor)
485{
486 int ret;
487
488 while (pipe->nrbufs) {
489 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200490
491 sd->len = buf->len;
492 if (sd->len > sd->total_len)
493 sd->len = sd->total_len;
494
Miklos Szeredifba597d2016-09-27 10:45:12 +0200495 ret = pipe_buf_confirm(pipe, buf);
Michał Mirosława8adbe32010-12-17 08:56:44 +0100496 if (unlikely(ret)) {
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200497 if (ret == -ENODATA)
498 ret = 0;
499 return ret;
500 }
Michał Mirosława8adbe32010-12-17 08:56:44 +0100501
502 ret = actor(pipe, buf, sd);
503 if (ret <= 0)
504 return ret;
505
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200506 buf->offset += ret;
507 buf->len -= ret;
508
509 sd->num_spliced += ret;
510 sd->len -= ret;
511 sd->pos += ret;
512 sd->total_len -= ret;
513
514 if (!buf->len) {
Miklos Szeredia7796382016-09-27 10:45:12 +0200515 pipe_buf_release(pipe, buf);
Jens Axboe35f3d142010-05-20 10:43:18 +0200516 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200517 pipe->nrbufs--;
Al Viro6447a3c2013-03-21 11:01:38 -0400518 if (pipe->files)
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200519 sd->need_wakeup = true;
520 }
521
522 if (!sd->total_len)
523 return 0;
524 }
525
526 return 1;
527}
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200528
529/**
530 * splice_from_pipe_next - wait for some data to splice from
531 * @pipe: pipe to splice from
532 * @sd: information about the splice operation
533 *
534 * Description:
535 * This function will wait for some data and return a positive
536 * value (one) if pipe buffers are available. It will return zero
537 * or -errno if no more data needs to be spliced.
538 */
Al Viro96f9bc82014-04-05 04:35:49 -0400539static int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd)
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200540{
Jan Karac725bfc2015-11-23 13:09:50 +0100541 /*
542 * Check for signal early to make process killable when there are
543 * always buffers available
544 */
545 if (signal_pending(current))
546 return -ERESTARTSYS;
547
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200548 while (!pipe->nrbufs) {
549 if (!pipe->writers)
550 return 0;
551
552 if (!pipe->waiting_writers && sd->num_spliced)
553 return 0;
554
555 if (sd->flags & SPLICE_F_NONBLOCK)
556 return -EAGAIN;
557
558 if (signal_pending(current))
559 return -ERESTARTSYS;
560
561 if (sd->need_wakeup) {
562 wakeup_pipe_writers(pipe);
563 sd->need_wakeup = false;
564 }
565
566 pipe_wait(pipe);
567 }
568
569 return 1;
570}
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200571
572/**
573 * splice_from_pipe_begin - start splicing from pipe
Randy Dunlapb80901b2009-04-16 19:09:55 -0700574 * @sd: information about the splice operation
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200575 *
576 * Description:
577 * This function should be called before a loop containing
578 * splice_from_pipe_next() and splice_from_pipe_feed() to
579 * initialize the necessary fields of @sd.
580 */
Al Viro96f9bc82014-04-05 04:35:49 -0400581static void splice_from_pipe_begin(struct splice_desc *sd)
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200582{
583 sd->num_spliced = 0;
584 sd->need_wakeup = false;
585}
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200586
587/**
588 * splice_from_pipe_end - finish splicing from pipe
589 * @pipe: pipe to splice from
590 * @sd: information about the splice operation
591 *
592 * Description:
593 * This function will wake up pipe writers if necessary. It should
594 * be called after a loop containing splice_from_pipe_next() and
595 * splice_from_pipe_feed().
596 */
Al Viro96f9bc82014-04-05 04:35:49 -0400597static void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd)
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200598{
599 if (sd->need_wakeup)
600 wakeup_pipe_writers(pipe);
601}
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200602
Jens Axboe932cc6d2007-06-21 13:10:21 +0200603/**
604 * __splice_from_pipe - splice data from a pipe to given actor
605 * @pipe: pipe to splice from
606 * @sd: information to @actor
607 * @actor: handler that splices the data
608 *
609 * Description:
610 * This function does little more than loop over the pipe and call
611 * @actor to do the actual moving of a single struct pipe_buffer to
612 * the desired destination. See pipe_to_file, pipe_to_sendpage, or
613 * pipe_to_user.
614 *
Jens Axboe83f91352006-04-02 23:05:09 +0200615 */
Jens Axboec66ab6f2007-06-12 21:17:17 +0200616ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd,
617 splice_actor *actor)
Jens Axboe5274f052006-03-30 15:15:30 +0200618{
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200619 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200620
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200621 splice_from_pipe_begin(sd);
622 do {
Jan Karac2489e02015-11-23 13:09:51 +0100623 cond_resched();
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200624 ret = splice_from_pipe_next(pipe, sd);
625 if (ret > 0)
626 ret = splice_from_pipe_feed(pipe, sd, actor);
627 } while (ret > 0);
628 splice_from_pipe_end(pipe, sd);
Jens Axboe5274f052006-03-30 15:15:30 +0200629
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200630 return sd->num_spliced ? sd->num_spliced : ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200631}
Mark Fasheh40bee44e2007-03-21 13:11:02 +0100632EXPORT_SYMBOL(__splice_from_pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200633
Jens Axboe932cc6d2007-06-21 13:10:21 +0200634/**
635 * splice_from_pipe - splice data from a pipe to a file
636 * @pipe: pipe to splice from
637 * @out: file to splice to
638 * @ppos: position in @out
639 * @len: how many bytes to splice
640 * @flags: splice modifier flags
641 * @actor: handler that splices the data
642 *
643 * Description:
Miklos Szeredi29339702009-04-14 19:48:37 +0200644 * See __splice_from_pipe. This function locks the pipe inode,
Jens Axboe932cc6d2007-06-21 13:10:21 +0200645 * otherwise it's identical to __splice_from_pipe().
646 *
647 */
Mark Fasheh6da61802006-10-17 18:43:07 +0200648ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out,
649 loff_t *ppos, size_t len, unsigned int flags,
650 splice_actor *actor)
651{
652 ssize_t ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200653 struct splice_desc sd = {
654 .total_len = len,
655 .flags = flags,
656 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +0200657 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +0200658 };
Mark Fasheh6da61802006-10-17 18:43:07 +0200659
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200660 pipe_lock(pipe);
Jens Axboec66ab6f2007-06-12 21:17:17 +0200661 ret = __splice_from_pipe(pipe, &sd, actor);
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200662 pipe_unlock(pipe);
Mark Fasheh6da61802006-10-17 18:43:07 +0200663
664 return ret;
665}
666
667/**
Al Viro8d020762014-04-05 04:27:08 -0400668 * iter_file_splice_write - splice data from a pipe to a file
669 * @pipe: pipe info
670 * @out: file to write to
671 * @ppos: position in @out
672 * @len: number of bytes to splice
673 * @flags: splice modifier flags
674 *
675 * Description:
676 * Will either move or copy pages (determined by @flags options) from
677 * the given pipe inode to the given file.
678 * This one is ->write_iter-based.
679 *
680 */
681ssize_t
682iter_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
683 loff_t *ppos, size_t len, unsigned int flags)
684{
685 struct splice_desc sd = {
686 .total_len = len,
687 .flags = flags,
688 .pos = *ppos,
689 .u.file = out,
690 };
691 int nbufs = pipe->buffers;
692 struct bio_vec *array = kcalloc(nbufs, sizeof(struct bio_vec),
693 GFP_KERNEL);
694 ssize_t ret;
695
696 if (unlikely(!array))
697 return -ENOMEM;
698
699 pipe_lock(pipe);
700
701 splice_from_pipe_begin(&sd);
702 while (sd.total_len) {
703 struct iov_iter from;
Al Viro8d020762014-04-05 04:27:08 -0400704 size_t left;
705 int n, idx;
706
707 ret = splice_from_pipe_next(pipe, &sd);
708 if (ret <= 0)
709 break;
710
711 if (unlikely(nbufs < pipe->buffers)) {
712 kfree(array);
713 nbufs = pipe->buffers;
714 array = kcalloc(nbufs, sizeof(struct bio_vec),
715 GFP_KERNEL);
716 if (!array) {
717 ret = -ENOMEM;
718 break;
719 }
720 }
721
722 /* build the vector */
723 left = sd.total_len;
724 for (n = 0, idx = pipe->curbuf; left && n < pipe->nrbufs; n++, idx++) {
725 struct pipe_buffer *buf = pipe->bufs + idx;
726 size_t this_len = buf->len;
727
728 if (this_len > left)
729 this_len = left;
730
731 if (idx == pipe->buffers - 1)
732 idx = -1;
733
Miklos Szeredifba597d2016-09-27 10:45:12 +0200734 ret = pipe_buf_confirm(pipe, buf);
Al Viro8d020762014-04-05 04:27:08 -0400735 if (unlikely(ret)) {
736 if (ret == -ENODATA)
737 ret = 0;
738 goto done;
739 }
740
741 array[n].bv_page = buf->page;
742 array[n].bv_len = this_len;
743 array[n].bv_offset = buf->offset;
744 left -= this_len;
745 }
746
Al Viro05afcb72015-01-23 01:08:07 -0500747 iov_iter_bvec(&from, ITER_BVEC | WRITE, array, n,
748 sd.total_len - left);
Christoph Hellwigabbb65892017-05-27 11:16:52 +0300749 ret = vfs_iter_write(out, &from, &sd.pos, 0);
Al Viro8d020762014-04-05 04:27:08 -0400750 if (ret <= 0)
751 break;
752
753 sd.num_spliced += ret;
754 sd.total_len -= ret;
Christoph Hellwigdbe4e192015-01-25 21:11:59 +0100755 *ppos = sd.pos;
Al Viro8d020762014-04-05 04:27:08 -0400756
757 /* dismiss the fully eaten buffers, adjust the partial one */
758 while (ret) {
759 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
760 if (ret >= buf->len) {
Al Viro8d020762014-04-05 04:27:08 -0400761 ret -= buf->len;
762 buf->len = 0;
Miklos Szeredia7796382016-09-27 10:45:12 +0200763 pipe_buf_release(pipe, buf);
Al Viro8d020762014-04-05 04:27:08 -0400764 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
765 pipe->nrbufs--;
766 if (pipe->files)
767 sd.need_wakeup = true;
768 } else {
769 buf->offset += ret;
770 buf->len -= ret;
771 ret = 0;
772 }
773 }
774 }
775done:
776 kfree(array);
777 splice_from_pipe_end(pipe, &sd);
778
779 pipe_unlock(pipe);
780
781 if (sd.num_spliced)
782 ret = sd.num_spliced;
783
784 return ret;
785}
786
787EXPORT_SYMBOL(iter_file_splice_write);
788
Miklos Szeredib2858d72009-05-19 11:37:46 +0200789static int write_pipe_buf(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
790 struct splice_desc *sd)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200791{
Miklos Szeredib2858d72009-05-19 11:37:46 +0200792 int ret;
793 void *data;
Al Viro06ae43f2013-03-20 13:19:30 -0400794 loff_t tmp = sd->pos;
Miklos Szeredib2858d72009-05-19 11:37:46 +0200795
Al Virofbb32752014-02-02 21:09:54 -0500796 data = kmap(buf->page);
Al Viro06ae43f2013-03-20 13:19:30 -0400797 ret = __kernel_write(sd->u.file, data + buf->offset, sd->len, &tmp);
Al Virofbb32752014-02-02 21:09:54 -0500798 kunmap(buf->page);
Miklos Szeredib2858d72009-05-19 11:37:46 +0200799
800 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200801}
802
803static ssize_t default_file_splice_write(struct pipe_inode_info *pipe,
804 struct file *out, loff_t *ppos,
805 size_t len, unsigned int flags)
806{
Miklos Szeredib2858d72009-05-19 11:37:46 +0200807 ssize_t ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200808
Miklos Szeredib2858d72009-05-19 11:37:46 +0200809 ret = splice_from_pipe(pipe, out, ppos, len, flags, write_pipe_buf);
810 if (ret > 0)
811 *ppos += ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200812
Miklos Szeredib2858d72009-05-19 11:37:46 +0200813 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200814}
815
Jens Axboe83f91352006-04-02 23:05:09 +0200816/**
817 * generic_splice_sendpage - splice data from a pipe to a socket
Jens Axboe932cc6d2007-06-21 13:10:21 +0200818 * @pipe: pipe to splice from
Jens Axboe83f91352006-04-02 23:05:09 +0200819 * @out: socket to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +0200820 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +0200821 * @len: number of bytes to splice
822 * @flags: splice modifier flags
823 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200824 * Description:
825 * Will send @len bytes from the pipe to a network socket. No data copying
826 * is involved.
Jens Axboe83f91352006-04-02 23:05:09 +0200827 *
828 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200829ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200830 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200831{
Jens Axboe00522fb2006-04-26 14:39:29 +0200832 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage);
Jens Axboe5274f052006-03-30 15:15:30 +0200833}
834
Jens Axboe059a8f32006-04-02 23:06:05 +0200835EXPORT_SYMBOL(generic_splice_sendpage);
Jeff Garzika0f06782006-03-30 23:06:13 -0500836
Jens Axboe83f91352006-04-02 23:05:09 +0200837/*
838 * Attempt to initiate a splice from pipe to file.
839 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200840static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200841 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200842{
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200843 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *,
844 loff_t *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +0200845
Al Viro72c2d532013-09-22 16:27:52 -0400846 if (out->f_op->splice_write)
Changli Gaocc56f7d2009-11-04 09:09:52 +0100847 splice_write = out->f_op->splice_write;
848 else
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200849 splice_write = default_file_splice_write;
850
Al Viro500368f2013-05-23 20:07:11 -0400851 return splice_write(pipe, out, ppos, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +0200852}
853
Jens Axboe83f91352006-04-02 23:05:09 +0200854/*
855 * Attempt to initiate a splice from a file to a pipe.
856 */
Jens Axboecbb7e572006-04-11 14:57:50 +0200857static long do_splice_to(struct file *in, loff_t *ppos,
858 struct pipe_inode_info *pipe, size_t len,
859 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200860{
Miklos Szeredi68181732009-05-07 15:37:36 +0200861 ssize_t (*splice_read)(struct file *, loff_t *,
862 struct pipe_inode_info *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +0200863 int ret;
864
Jens Axboe49570e92006-04-11 13:56:09 +0200865 if (unlikely(!(in->f_mode & FMODE_READ)))
Jens Axboe5274f052006-03-30 15:15:30 +0200866 return -EBADF;
867
Jens Axboecbb7e572006-04-11 14:57:50 +0200868 ret = rw_verify_area(READ, in, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +0200869 if (unlikely(ret < 0))
870 return ret;
871
Al Viro03cc0782016-04-02 14:56:58 -0400872 if (unlikely(len > MAX_RW_COUNT))
873 len = MAX_RW_COUNT;
874
Al Viro72c2d532013-09-22 16:27:52 -0400875 if (in->f_op->splice_read)
Changli Gaocc56f7d2009-11-04 09:09:52 +0100876 splice_read = in->f_op->splice_read;
877 else
Miklos Szeredi68181732009-05-07 15:37:36 +0200878 splice_read = default_file_splice_read;
879
880 return splice_read(in, ppos, pipe, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +0200881}
882
Jens Axboe932cc6d2007-06-21 13:10:21 +0200883/**
884 * splice_direct_to_actor - splices data directly between two non-pipes
885 * @in: file to splice from
886 * @sd: actor information on where to splice to
887 * @actor: handles the data splicing
888 *
889 * Description:
890 * This is a special case helper to splice directly between two
891 * points, without requiring an explicit pipe. Internally an allocated
Randy Dunlap79685b82007-07-27 08:08:51 +0200892 * pipe is cached in the process, and reused during the lifetime of
Jens Axboe932cc6d2007-06-21 13:10:21 +0200893 * that process.
894 *
Jens Axboec66ab6f2007-06-12 21:17:17 +0200895 */
896ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
897 splice_direct_actor *actor)
Jens Axboeb92ce552006-04-11 13:52:07 +0200898{
899 struct pipe_inode_info *pipe;
900 long ret, bytes;
901 umode_t i_mode;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200902 size_t len;
Christophe Leroy0ff28d92015-05-06 17:26:47 +0200903 int i, flags, more;
Jens Axboeb92ce552006-04-11 13:52:07 +0200904
905 /*
906 * We require the input being a regular file, as we don't want to
907 * randomly drop data for eg socket -> socket splicing. Use the
908 * piped splicing for that!
909 */
Al Viro496ad9a2013-01-23 17:07:38 -0500910 i_mode = file_inode(in)->i_mode;
Jens Axboeb92ce552006-04-11 13:52:07 +0200911 if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
912 return -EINVAL;
913
914 /*
915 * neither in nor out is a pipe, setup an internal pipe attached to
916 * 'out' and transfer the wanted data from 'in' to 'out' through that
917 */
918 pipe = current->splice_pipe;
Jens Axboe49570e92006-04-11 13:56:09 +0200919 if (unlikely(!pipe)) {
Al Viro7bee1302013-03-21 11:04:15 -0400920 pipe = alloc_pipe_info();
Jens Axboeb92ce552006-04-11 13:52:07 +0200921 if (!pipe)
922 return -ENOMEM;
923
924 /*
925 * We don't have an immediate reader, but we'll read the stuff
Jens Axboe00522fb2006-04-26 14:39:29 +0200926 * out of the pipe right after the splice_to_pipe(). So set
Jens Axboeb92ce552006-04-11 13:52:07 +0200927 * PIPE_READERS appropriately.
928 */
929 pipe->readers = 1;
930
931 current->splice_pipe = pipe;
932 }
933
934 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +0200935 * Do the splice.
Jens Axboeb92ce552006-04-11 13:52:07 +0200936 */
937 ret = 0;
938 bytes = 0;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200939 len = sd->total_len;
940 flags = sd->flags;
941
942 /*
943 * Don't block on output, we have to drain the direct pipe.
944 */
945 sd->flags &= ~SPLICE_F_NONBLOCK;
Christophe Leroy0ff28d92015-05-06 17:26:47 +0200946 more = sd->flags & SPLICE_F_MORE;
Jens Axboeb92ce552006-04-11 13:52:07 +0200947
948 while (len) {
Jens Axboe51a92c02007-07-13 14:11:43 +0200949 size_t read_len;
Tom Zanussia82c53a2008-05-09 13:28:36 +0200950 loff_t pos = sd->pos, prev_pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +0200951
Jens Axboebcd4f3a2007-07-16 14:41:49 +0200952 ret = do_splice_to(in, &pos, pipe, len, flags);
Jens Axboe51a92c02007-07-13 14:11:43 +0200953 if (unlikely(ret <= 0))
Jens Axboeb92ce552006-04-11 13:52:07 +0200954 goto out_release;
955
956 read_len = ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200957 sd->total_len = read_len;
Jens Axboeb92ce552006-04-11 13:52:07 +0200958
959 /*
Christophe Leroy0ff28d92015-05-06 17:26:47 +0200960 * If more data is pending, set SPLICE_F_MORE
961 * If this is the last data and SPLICE_F_MORE was not set
962 * initially, clears it.
963 */
964 if (read_len < len)
965 sd->flags |= SPLICE_F_MORE;
966 else if (!more)
967 sd->flags &= ~SPLICE_F_MORE;
968 /*
Jens Axboeb92ce552006-04-11 13:52:07 +0200969 * NOTE: nonblocking mode only applies to the input. We
970 * must not do the output in nonblocking mode as then we
971 * could get stuck data in the internal pipe:
972 */
Jens Axboec66ab6f2007-06-12 21:17:17 +0200973 ret = actor(pipe, sd);
Tom Zanussia82c53a2008-05-09 13:28:36 +0200974 if (unlikely(ret <= 0)) {
975 sd->pos = prev_pos;
Jens Axboeb92ce552006-04-11 13:52:07 +0200976 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +0200977 }
Jens Axboeb92ce552006-04-11 13:52:07 +0200978
979 bytes += ret;
980 len -= ret;
Jens Axboebcd4f3a2007-07-16 14:41:49 +0200981 sd->pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +0200982
Tom Zanussia82c53a2008-05-09 13:28:36 +0200983 if (ret < read_len) {
984 sd->pos = prev_pos + ret;
Jens Axboe51a92c02007-07-13 14:11:43 +0200985 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +0200986 }
Jens Axboeb92ce552006-04-11 13:52:07 +0200987 }
988
Jens Axboe9e971982008-01-29 21:05:57 +0100989done:
Jens Axboeb92ce552006-04-11 13:52:07 +0200990 pipe->nrbufs = pipe->curbuf = 0;
Jens Axboe80848702008-01-30 12:24:48 +0100991 file_accessed(in);
Jens Axboeb92ce552006-04-11 13:52:07 +0200992 return bytes;
993
994out_release:
995 /*
996 * If we did an incomplete transfer we must release
997 * the pipe buffers in question:
998 */
Jens Axboe35f3d142010-05-20 10:43:18 +0200999 for (i = 0; i < pipe->buffers; i++) {
Jens Axboeb92ce552006-04-11 13:52:07 +02001000 struct pipe_buffer *buf = pipe->bufs + i;
1001
Miklos Szeredia7796382016-09-27 10:45:12 +02001002 if (buf->ops)
1003 pipe_buf_release(pipe, buf);
Jens Axboeb92ce552006-04-11 13:52:07 +02001004 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001005
Jens Axboe9e971982008-01-29 21:05:57 +01001006 if (!bytes)
1007 bytes = ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001008
Jens Axboe9e971982008-01-29 21:05:57 +01001009 goto done;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001010}
1011EXPORT_SYMBOL(splice_direct_to_actor);
1012
1013static int direct_splice_actor(struct pipe_inode_info *pipe,
1014 struct splice_desc *sd)
1015{
Jens Axboe6a14b902007-06-14 13:08:55 +02001016 struct file *file = sd->u.file;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001017
Al Viro7995bd22013-06-20 18:58:36 +04001018 return do_splice_from(pipe, file, sd->opos, sd->total_len,
Changli Gao2cb4b052010-06-29 13:09:18 +02001019 sd->flags);
Jens Axboec66ab6f2007-06-12 21:17:17 +02001020}
1021
Jens Axboe932cc6d2007-06-21 13:10:21 +02001022/**
1023 * do_splice_direct - splices data directly between two files
1024 * @in: file to splice from
1025 * @ppos: input file offset
1026 * @out: file to splice to
Randy Dunlapacdb37c2013-06-22 19:44:08 -07001027 * @opos: output file offset
Jens Axboe932cc6d2007-06-21 13:10:21 +02001028 * @len: number of bytes to splice
1029 * @flags: splice modifier flags
1030 *
1031 * Description:
1032 * For use by do_sendfile(). splice can easily emulate sendfile, but
1033 * doing it in the application would incur an extra system call
1034 * (splice in + splice out, as compared to just sendfile()). So this helper
1035 * can splice directly through a process-private pipe.
1036 *
1037 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001038long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
Al Viro7995bd22013-06-20 18:58:36 +04001039 loff_t *opos, size_t len, unsigned int flags)
Jens Axboec66ab6f2007-06-12 21:17:17 +02001040{
1041 struct splice_desc sd = {
1042 .len = len,
1043 .total_len = len,
1044 .flags = flags,
1045 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +02001046 .u.file = out,
Al Viro7995bd22013-06-20 18:58:36 +04001047 .opos = opos,
Jens Axboec66ab6f2007-06-12 21:17:17 +02001048 };
Jens Axboe51a92c02007-07-13 14:11:43 +02001049 long ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001050
Al Viro18c67cb2013-06-19 15:41:54 +04001051 if (unlikely(!(out->f_mode & FMODE_WRITE)))
1052 return -EBADF;
1053
1054 if (unlikely(out->f_flags & O_APPEND))
1055 return -EINVAL;
1056
1057 ret = rw_verify_area(WRITE, out, opos, len);
1058 if (unlikely(ret < 0))
1059 return ret;
1060
Jens Axboec66ab6f2007-06-12 21:17:17 +02001061 ret = splice_direct_to_actor(in, &sd, direct_splice_actor);
Jens Axboe51a92c02007-07-13 14:11:43 +02001062 if (ret > 0)
Tom Zanussia82c53a2008-05-09 13:28:36 +02001063 *ppos = sd.pos;
Jens Axboe51a92c02007-07-13 14:11:43 +02001064
Jens Axboec66ab6f2007-06-12 21:17:17 +02001065 return ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001066}
Miklos Szeredi1c118592014-10-24 00:14:35 +02001067EXPORT_SYMBOL(do_splice_direct);
Jens Axboeb92ce552006-04-11 13:52:07 +02001068
Al Viro8924fef2016-09-17 20:44:45 -04001069static int wait_for_space(struct pipe_inode_info *pipe, unsigned flags)
1070{
Linus Torvalds52bce912016-12-21 10:59:34 -08001071 for (;;) {
1072 if (unlikely(!pipe->readers)) {
1073 send_sig(SIGPIPE, current, 0);
1074 return -EPIPE;
1075 }
1076 if (pipe->nrbufs != pipe->buffers)
1077 return 0;
Al Viro8924fef2016-09-17 20:44:45 -04001078 if (flags & SPLICE_F_NONBLOCK)
1079 return -EAGAIN;
1080 if (signal_pending(current))
1081 return -ERESTARTSYS;
1082 pipe->waiting_writers++;
1083 pipe_wait(pipe);
1084 pipe->waiting_writers--;
1085 }
Al Viro8924fef2016-09-17 20:44:45 -04001086}
1087
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001088static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1089 struct pipe_inode_info *opipe,
1090 size_t len, unsigned int flags);
Jens Axboeddac0d32006-11-04 12:49:32 +01001091
1092/*
Jens Axboe83f91352006-04-02 23:05:09 +02001093 * Determine where to splice to/from.
1094 */
Ingo Molnar529565d2006-04-10 15:18:58 +02001095static long do_splice(struct file *in, loff_t __user *off_in,
1096 struct file *out, loff_t __user *off_out,
1097 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001098{
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001099 struct pipe_inode_info *ipipe;
1100 struct pipe_inode_info *opipe;
Al Viro7995bd22013-06-20 18:58:36 +04001101 loff_t offset;
Jens Axboea4514eb2006-04-19 15:57:05 +02001102 long ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001103
Linus Torvalds71993e62010-11-28 13:56:09 -08001104 ipipe = get_pipe_info(in);
1105 opipe = get_pipe_info(out);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001106
1107 if (ipipe && opipe) {
1108 if (off_in || off_out)
1109 return -ESPIPE;
1110
1111 if (!(in->f_mode & FMODE_READ))
1112 return -EBADF;
1113
1114 if (!(out->f_mode & FMODE_WRITE))
1115 return -EBADF;
1116
1117 /* Splicing to self would be fun, but... */
1118 if (ipipe == opipe)
1119 return -EINVAL;
1120
1121 return splice_pipe_to_pipe(ipipe, opipe, len, flags);
1122 }
1123
1124 if (ipipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001125 if (off_in)
1126 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001127 if (off_out) {
Changli Gao19c9a492010-06-29 13:10:36 +02001128 if (!(out->f_mode & FMODE_PWRITE))
Jens Axboeb92ce552006-04-11 13:52:07 +02001129 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001130 if (copy_from_user(&offset, off_out, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001131 return -EFAULT;
Al Viro7995bd22013-06-20 18:58:36 +04001132 } else {
1133 offset = out->f_pos;
1134 }
Ingo Molnar529565d2006-04-10 15:18:58 +02001135
Al Viro18c67cb2013-06-19 15:41:54 +04001136 if (unlikely(!(out->f_mode & FMODE_WRITE)))
1137 return -EBADF;
1138
1139 if (unlikely(out->f_flags & O_APPEND))
1140 return -EINVAL;
1141
1142 ret = rw_verify_area(WRITE, out, &offset, len);
1143 if (unlikely(ret < 0))
1144 return ret;
1145
Al Viro500368f2013-05-23 20:07:11 -04001146 file_start_write(out);
Al Viro7995bd22013-06-20 18:58:36 +04001147 ret = do_splice_from(ipipe, out, &offset, len, flags);
Al Viro500368f2013-05-23 20:07:11 -04001148 file_end_write(out);
Jens Axboea4514eb2006-04-19 15:57:05 +02001149
Al Viro7995bd22013-06-20 18:58:36 +04001150 if (!off_out)
1151 out->f_pos = offset;
1152 else if (copy_to_user(off_out, &offset, sizeof(loff_t)))
Jens Axboea4514eb2006-04-19 15:57:05 +02001153 ret = -EFAULT;
1154
1155 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001156 }
Jens Axboe5274f052006-03-30 15:15:30 +02001157
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001158 if (opipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001159 if (off_out)
1160 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001161 if (off_in) {
Changli Gao19c9a492010-06-29 13:10:36 +02001162 if (!(in->f_mode & FMODE_PREAD))
Jens Axboeb92ce552006-04-11 13:52:07 +02001163 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001164 if (copy_from_user(&offset, off_in, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001165 return -EFAULT;
Al Viro7995bd22013-06-20 18:58:36 +04001166 } else {
1167 offset = in->f_pos;
1168 }
Ingo Molnar529565d2006-04-10 15:18:58 +02001169
Al Viro8924fef2016-09-17 20:44:45 -04001170 pipe_lock(opipe);
1171 ret = wait_for_space(opipe, flags);
1172 if (!ret)
1173 ret = do_splice_to(in, &offset, opipe, len, flags);
1174 pipe_unlock(opipe);
1175 if (ret > 0)
1176 wakeup_pipe_readers(opipe);
Al Viro7995bd22013-06-20 18:58:36 +04001177 if (!off_in)
1178 in->f_pos = offset;
1179 else if (copy_to_user(off_in, &offset, sizeof(loff_t)))
Jens Axboea4514eb2006-04-19 15:57:05 +02001180 ret = -EFAULT;
1181
1182 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001183 }
Jens Axboe5274f052006-03-30 15:15:30 +02001184
1185 return -EINVAL;
1186}
1187
Al Viro79fddc42016-09-17 22:38:20 -04001188static int iter_to_pipe(struct iov_iter *from,
1189 struct pipe_inode_info *pipe,
1190 unsigned flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001191{
Al Viro79fddc42016-09-17 22:38:20 -04001192 struct pipe_buffer buf = {
1193 .ops = &user_page_pipe_buf_ops,
1194 .flags = flags
1195 };
1196 size_t total = 0;
1197 int ret = 0;
1198 bool failed = false;
1199
1200 while (iov_iter_count(from) && !failed) {
1201 struct page *pages[16];
Al Virodb85a9e2016-09-17 20:25:06 -04001202 ssize_t copied;
1203 size_t start;
Al Viro79fddc42016-09-17 22:38:20 -04001204 int n;
Jens Axboe912d35f2006-04-26 10:59:21 +02001205
Al Viro79fddc42016-09-17 22:38:20 -04001206 copied = iov_iter_get_pages(from, pages, ~0UL, 16, &start);
1207 if (copied <= 0) {
1208 ret = copied;
1209 break;
1210 }
Jens Axboe912d35f2006-04-26 10:59:21 +02001211
Al Viro79fddc42016-09-17 22:38:20 -04001212 for (n = 0; copied; n++, start = 0) {
Al Virodb85a9e2016-09-17 20:25:06 -04001213 int size = min_t(int, copied, PAGE_SIZE - start);
Al Viro79fddc42016-09-17 22:38:20 -04001214 if (!failed) {
1215 buf.page = pages[n];
1216 buf.offset = start;
1217 buf.len = size;
1218 ret = add_to_pipe(pipe, &buf);
1219 if (unlikely(ret < 0)) {
1220 failed = true;
1221 } else {
1222 iov_iter_advance(from, ret);
1223 total += ret;
1224 }
1225 } else {
1226 put_page(pages[n]);
1227 }
Al Virodb85a9e2016-09-17 20:25:06 -04001228 copied -= size;
Jens Axboe912d35f2006-04-26 10:59:21 +02001229 }
Jens Axboe912d35f2006-04-26 10:59:21 +02001230 }
Al Viro79fddc42016-09-17 22:38:20 -04001231 return total ? total : ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001232}
1233
Jens Axboe6a14b902007-06-14 13:08:55 +02001234static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1235 struct splice_desc *sd)
1236{
Al Viro6130f532014-02-03 18:19:51 -05001237 int n = copy_page_to_iter(buf->page, buf->offset, sd->len, sd->u.data);
1238 return n == sd->len ? n : -EFAULT;
Jens Axboe6a14b902007-06-14 13:08:55 +02001239}
1240
1241/*
1242 * For lack of a better implementation, implement vmsplice() to userspace
1243 * as a simple copy of the pipes pages to the user iov.
1244 */
Al Viro6130f532014-02-03 18:19:51 -05001245static long vmsplice_to_user(struct file *file, const struct iovec __user *uiov,
Jens Axboe6a14b902007-06-14 13:08:55 +02001246 unsigned long nr_segs, unsigned int flags)
1247{
1248 struct pipe_inode_info *pipe;
1249 struct splice_desc sd;
Jens Axboe6a14b902007-06-14 13:08:55 +02001250 long ret;
Al Viro6130f532014-02-03 18:19:51 -05001251 struct iovec iovstack[UIO_FASTIOV];
1252 struct iovec *iov = iovstack;
1253 struct iov_iter iter;
Jens Axboe6a14b902007-06-14 13:08:55 +02001254
Linus Torvalds71993e62010-11-28 13:56:09 -08001255 pipe = get_pipe_info(file);
Jens Axboe6a14b902007-06-14 13:08:55 +02001256 if (!pipe)
1257 return -EBADF;
1258
Al Viro345995f2015-03-21 19:17:55 -04001259 ret = import_iovec(READ, uiov, nr_segs,
1260 ARRAY_SIZE(iovstack), &iov, &iter);
1261 if (ret < 0)
1262 return ret;
Al Viro6130f532014-02-03 18:19:51 -05001263
Al Viro345995f2015-03-21 19:17:55 -04001264 sd.total_len = iov_iter_count(&iter);
Al Viro6130f532014-02-03 18:19:51 -05001265 sd.len = 0;
Al Viro6130f532014-02-03 18:19:51 -05001266 sd.flags = flags;
1267 sd.u.data = &iter;
1268 sd.pos = 0;
1269
Al Viro345995f2015-03-21 19:17:55 -04001270 if (sd.total_len) {
1271 pipe_lock(pipe);
1272 ret = __splice_from_pipe(pipe, &sd, pipe_to_user);
1273 pipe_unlock(pipe);
1274 }
Jens Axboe6a14b902007-06-14 13:08:55 +02001275
Al Viro345995f2015-03-21 19:17:55 -04001276 kfree(iov);
Jens Axboe6a14b902007-06-14 13:08:55 +02001277 return ret;
1278}
1279
Jens Axboe912d35f2006-04-26 10:59:21 +02001280/*
1281 * vmsplice splices a user address range into a pipe. It can be thought of
1282 * as splice-from-memory, where the regular splice is splice-from-file (or
1283 * to file). In both cases the output is a pipe, naturally.
Jens Axboe912d35f2006-04-26 10:59:21 +02001284 */
Al Virodb85a9e2016-09-17 20:25:06 -04001285static long vmsplice_to_pipe(struct file *file, const struct iovec __user *uiov,
Jens Axboe6a14b902007-06-14 13:08:55 +02001286 unsigned long nr_segs, unsigned int flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001287{
Jens Axboeddac0d32006-11-04 12:49:32 +01001288 struct pipe_inode_info *pipe;
Al Virodb85a9e2016-09-17 20:25:06 -04001289 struct iovec iovstack[UIO_FASTIOV];
1290 struct iovec *iov = iovstack;
1291 struct iov_iter from;
Jens Axboe35f3d142010-05-20 10:43:18 +02001292 long ret;
Al Viro79fddc42016-09-17 22:38:20 -04001293 unsigned buf_flag = 0;
1294
1295 if (flags & SPLICE_F_GIFT)
1296 buf_flag = PIPE_BUF_FLAG_GIFT;
Jens Axboe912d35f2006-04-26 10:59:21 +02001297
Linus Torvalds71993e62010-11-28 13:56:09 -08001298 pipe = get_pipe_info(file);
Jens Axboeddac0d32006-11-04 12:49:32 +01001299 if (!pipe)
Jens Axboe912d35f2006-04-26 10:59:21 +02001300 return -EBADF;
Jens Axboe912d35f2006-04-26 10:59:21 +02001301
Al Virodb85a9e2016-09-17 20:25:06 -04001302 ret = import_iovec(WRITE, uiov, nr_segs,
1303 ARRAY_SIZE(iovstack), &iov, &from);
1304 if (ret < 0)
1305 return ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001306
Al Viro8924fef2016-09-17 20:44:45 -04001307 pipe_lock(pipe);
1308 ret = wait_for_space(pipe, flags);
Al Viro79fddc42016-09-17 22:38:20 -04001309 if (!ret)
1310 ret = iter_to_pipe(&from, pipe, buf_flag);
Al Viro8924fef2016-09-17 20:44:45 -04001311 pipe_unlock(pipe);
1312 if (ret > 0)
1313 wakeup_pipe_readers(pipe);
Al Virodb85a9e2016-09-17 20:25:06 -04001314 kfree(iov);
Jens Axboe35f3d142010-05-20 10:43:18 +02001315 return ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001316}
1317
Jens Axboe6a14b902007-06-14 13:08:55 +02001318/*
1319 * Note that vmsplice only really supports true splicing _from_ user memory
1320 * to a pipe, not the other way around. Splicing from user memory is a simple
1321 * operation that can be supported without any funky alignment restrictions
1322 * or nasty vm tricks. We simply map in the user memory and fill them into
1323 * a pipe. The reverse isn't quite as easy, though. There are two possible
1324 * solutions for that:
1325 *
1326 * - memcpy() the data internally, at which point we might as well just
1327 * do a regular read() on the buffer anyway.
1328 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1329 * has restriction limitations on both ends of the pipe).
1330 *
1331 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1332 *
1333 */
Dominik Brodowski30cfe4e2018-03-17 15:00:24 +01001334static long do_vmsplice(int fd, const struct iovec __user *iov,
1335 unsigned long nr_segs, unsigned int flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001336{
Al Viro2903ff02012-08-28 12:52:22 -04001337 struct fd f;
Jens Axboe912d35f2006-04-26 10:59:21 +02001338 long error;
Jens Axboe912d35f2006-04-26 10:59:21 +02001339
Al Viro3d6ea292016-12-10 13:17:32 -05001340 if (unlikely(flags & ~SPLICE_F_ALL))
1341 return -EINVAL;
Jens Axboe6a14b902007-06-14 13:08:55 +02001342 if (unlikely(nr_segs > UIO_MAXIOV))
1343 return -EINVAL;
1344 else if (unlikely(!nr_segs))
1345 return 0;
1346
Jens Axboe912d35f2006-04-26 10:59:21 +02001347 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001348 f = fdget(fd);
1349 if (f.file) {
1350 if (f.file->f_mode & FMODE_WRITE)
1351 error = vmsplice_to_pipe(f.file, iov, nr_segs, flags);
1352 else if (f.file->f_mode & FMODE_READ)
1353 error = vmsplice_to_user(f.file, iov, nr_segs, flags);
Jens Axboe912d35f2006-04-26 10:59:21 +02001354
Al Viro2903ff02012-08-28 12:52:22 -04001355 fdput(f);
Jens Axboe912d35f2006-04-26 10:59:21 +02001356 }
1357
1358 return error;
1359}
1360
Dominik Brodowski30cfe4e2018-03-17 15:00:24 +01001361SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, iov,
1362 unsigned long, nr_segs, unsigned int, flags)
1363{
1364 return do_vmsplice(fd, iov, nr_segs, flags);
1365}
1366
Al Viro76b021d2013-03-02 10:19:56 -05001367#ifdef CONFIG_COMPAT
1368COMPAT_SYSCALL_DEFINE4(vmsplice, int, fd, const struct compat_iovec __user *, iov32,
1369 unsigned int, nr_segs, unsigned int, flags)
1370{
1371 unsigned i;
1372 struct iovec __user *iov;
1373 if (nr_segs > UIO_MAXIOV)
1374 return -EINVAL;
1375 iov = compat_alloc_user_space(nr_segs * sizeof(struct iovec));
1376 for (i = 0; i < nr_segs; i++) {
1377 struct compat_iovec v;
1378 if (get_user(v.iov_base, &iov32[i].iov_base) ||
1379 get_user(v.iov_len, &iov32[i].iov_len) ||
1380 put_user(compat_ptr(v.iov_base), &iov[i].iov_base) ||
1381 put_user(v.iov_len, &iov[i].iov_len))
1382 return -EFAULT;
1383 }
Dominik Brodowski30cfe4e2018-03-17 15:00:24 +01001384 return do_vmsplice(fd, iov, nr_segs, flags);
Al Viro76b021d2013-03-02 10:19:56 -05001385}
1386#endif
1387
Heiko Carstens836f92a2009-01-14 14:14:33 +01001388SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in,
1389 int, fd_out, loff_t __user *, off_out,
1390 size_t, len, unsigned int, flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001391{
Al Viro2903ff02012-08-28 12:52:22 -04001392 struct fd in, out;
Jens Axboe5274f052006-03-30 15:15:30 +02001393 long error;
Jens Axboe5274f052006-03-30 15:15:30 +02001394
1395 if (unlikely(!len))
1396 return 0;
1397
Al Viro3d6ea292016-12-10 13:17:32 -05001398 if (unlikely(flags & ~SPLICE_F_ALL))
1399 return -EINVAL;
1400
Jens Axboe5274f052006-03-30 15:15:30 +02001401 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001402 in = fdget(fd_in);
1403 if (in.file) {
1404 if (in.file->f_mode & FMODE_READ) {
1405 out = fdget(fd_out);
1406 if (out.file) {
1407 if (out.file->f_mode & FMODE_WRITE)
1408 error = do_splice(in.file, off_in,
1409 out.file, off_out,
Ingo Molnar529565d2006-04-10 15:18:58 +02001410 len, flags);
Al Viro2903ff02012-08-28 12:52:22 -04001411 fdput(out);
Jens Axboe5274f052006-03-30 15:15:30 +02001412 }
1413 }
Al Viro2903ff02012-08-28 12:52:22 -04001414 fdput(in);
Jens Axboe5274f052006-03-30 15:15:30 +02001415 }
Jens Axboe5274f052006-03-30 15:15:30 +02001416 return error;
1417}
Jens Axboe70524492006-04-11 15:51:17 +02001418
1419/*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001420 * Make sure there's data to read. Wait for input if we can, otherwise
1421 * return an appropriate error.
1422 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001423static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001424{
1425 int ret;
1426
1427 /*
1428 * Check ->nrbufs without the inode lock first. This function
1429 * is speculative anyways, so missing one is ok.
1430 */
1431 if (pipe->nrbufs)
1432 return 0;
1433
1434 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001435 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001436
1437 while (!pipe->nrbufs) {
1438 if (signal_pending(current)) {
1439 ret = -ERESTARTSYS;
1440 break;
1441 }
1442 if (!pipe->writers)
1443 break;
1444 if (!pipe->waiting_writers) {
1445 if (flags & SPLICE_F_NONBLOCK) {
1446 ret = -EAGAIN;
1447 break;
1448 }
1449 }
1450 pipe_wait(pipe);
1451 }
1452
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001453 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001454 return ret;
1455}
1456
1457/*
1458 * Make sure there's writeable room. Wait for room if we can, otherwise
1459 * return an appropriate error.
1460 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001461static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001462{
1463 int ret;
1464
1465 /*
1466 * Check ->nrbufs without the inode lock first. This function
1467 * is speculative anyways, so missing one is ok.
1468 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001469 if (pipe->nrbufs < pipe->buffers)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001470 return 0;
1471
1472 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001473 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001474
Jens Axboe35f3d142010-05-20 10:43:18 +02001475 while (pipe->nrbufs >= pipe->buffers) {
Jens Axboeaadd06e2006-07-10 11:00:01 +02001476 if (!pipe->readers) {
1477 send_sig(SIGPIPE, current, 0);
1478 ret = -EPIPE;
1479 break;
1480 }
1481 if (flags & SPLICE_F_NONBLOCK) {
1482 ret = -EAGAIN;
1483 break;
1484 }
1485 if (signal_pending(current)) {
1486 ret = -ERESTARTSYS;
1487 break;
1488 }
1489 pipe->waiting_writers++;
1490 pipe_wait(pipe);
1491 pipe->waiting_writers--;
1492 }
1493
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001494 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001495 return ret;
1496}
1497
1498/*
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001499 * Splice contents of ipipe to opipe.
1500 */
1501static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1502 struct pipe_inode_info *opipe,
1503 size_t len, unsigned int flags)
1504{
1505 struct pipe_buffer *ibuf, *obuf;
1506 int ret = 0, nbuf;
1507 bool input_wakeup = false;
1508
1509
1510retry:
1511 ret = ipipe_prep(ipipe, flags);
1512 if (ret)
1513 return ret;
1514
1515 ret = opipe_prep(opipe, flags);
1516 if (ret)
1517 return ret;
1518
1519 /*
1520 * Potential ABBA deadlock, work around it by ordering lock
1521 * grabbing by pipe info address. Otherwise two different processes
1522 * could deadlock (one doing tee from A -> B, the other from B -> A).
1523 */
1524 pipe_double_lock(ipipe, opipe);
1525
1526 do {
1527 if (!opipe->readers) {
1528 send_sig(SIGPIPE, current, 0);
1529 if (!ret)
1530 ret = -EPIPE;
1531 break;
1532 }
1533
1534 if (!ipipe->nrbufs && !ipipe->writers)
1535 break;
1536
1537 /*
1538 * Cannot make any progress, because either the input
1539 * pipe is empty or the output pipe is full.
1540 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001541 if (!ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001542 /* Already processed some buffers, break */
1543 if (ret)
1544 break;
1545
1546 if (flags & SPLICE_F_NONBLOCK) {
1547 ret = -EAGAIN;
1548 break;
1549 }
1550
1551 /*
1552 * We raced with another reader/writer and haven't
1553 * managed to process any buffers. A zero return
1554 * value means EOF, so retry instead.
1555 */
1556 pipe_unlock(ipipe);
1557 pipe_unlock(opipe);
1558 goto retry;
1559 }
1560
1561 ibuf = ipipe->bufs + ipipe->curbuf;
Jens Axboe35f3d142010-05-20 10:43:18 +02001562 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001563 obuf = opipe->bufs + nbuf;
1564
1565 if (len >= ibuf->len) {
1566 /*
1567 * Simply move the whole buffer from ipipe to opipe
1568 */
1569 *obuf = *ibuf;
1570 ibuf->ops = NULL;
1571 opipe->nrbufs++;
Jens Axboe35f3d142010-05-20 10:43:18 +02001572 ipipe->curbuf = (ipipe->curbuf + 1) & (ipipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001573 ipipe->nrbufs--;
1574 input_wakeup = true;
1575 } else {
1576 /*
1577 * Get a reference to this pipe buffer,
1578 * so we can copy the contents over.
1579 */
Miklos Szeredi7bf2d1d2016-09-27 10:45:12 +02001580 pipe_buf_get(ipipe, ibuf);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001581 *obuf = *ibuf;
1582
1583 /*
1584 * Don't inherit the gift flag, we need to
1585 * prevent multiple steals of this page.
1586 */
1587 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1588
1589 obuf->len = len;
1590 opipe->nrbufs++;
1591 ibuf->offset += obuf->len;
1592 ibuf->len -= obuf->len;
1593 }
1594 ret += obuf->len;
1595 len -= obuf->len;
1596 } while (len);
1597
1598 pipe_unlock(ipipe);
1599 pipe_unlock(opipe);
1600
1601 /*
1602 * If we put data in the output pipe, wakeup any potential readers.
1603 */
Namhyung Kim825cdcb2011-05-23 19:58:53 +02001604 if (ret > 0)
1605 wakeup_pipe_readers(opipe);
1606
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001607 if (input_wakeup)
1608 wakeup_pipe_writers(ipipe);
1609
1610 return ret;
1611}
1612
1613/*
Jens Axboe70524492006-04-11 15:51:17 +02001614 * Link contents of ipipe to opipe.
1615 */
1616static int link_pipe(struct pipe_inode_info *ipipe,
1617 struct pipe_inode_info *opipe,
1618 size_t len, unsigned int flags)
1619{
1620 struct pipe_buffer *ibuf, *obuf;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001621 int ret = 0, i = 0, nbuf;
Jens Axboe70524492006-04-11 15:51:17 +02001622
1623 /*
1624 * Potential ABBA deadlock, work around it by ordering lock
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001625 * grabbing by pipe info address. Otherwise two different processes
Jens Axboe70524492006-04-11 15:51:17 +02001626 * could deadlock (one doing tee from A -> B, the other from B -> A).
1627 */
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001628 pipe_double_lock(ipipe, opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001629
Jens Axboeaadd06e2006-07-10 11:00:01 +02001630 do {
Jens Axboe70524492006-04-11 15:51:17 +02001631 if (!opipe->readers) {
1632 send_sig(SIGPIPE, current, 0);
1633 if (!ret)
1634 ret = -EPIPE;
1635 break;
1636 }
Jens Axboe70524492006-04-11 15:51:17 +02001637
1638 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001639 * If we have iterated all input buffers or ran out of
1640 * output room, break.
Jens Axboe70524492006-04-11 15:51:17 +02001641 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001642 if (i >= ipipe->nrbufs || opipe->nrbufs >= opipe->buffers)
Jens Axboe70524492006-04-11 15:51:17 +02001643 break;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001644
Jens Axboe35f3d142010-05-20 10:43:18 +02001645 ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (ipipe->buffers-1));
1646 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001647
Jens Axboe2a27250e2006-04-19 15:56:40 +02001648 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001649 * Get a reference to this pipe buffer,
1650 * so we can copy the contents over.
Jens Axboe2a27250e2006-04-19 15:56:40 +02001651 */
Miklos Szeredi7bf2d1d2016-09-27 10:45:12 +02001652 pipe_buf_get(ipipe, ibuf);
Jens Axboe70524492006-04-11 15:51:17 +02001653
Jens Axboeaadd06e2006-07-10 11:00:01 +02001654 obuf = opipe->bufs + nbuf;
1655 *obuf = *ibuf;
Jens Axboe70524492006-04-11 15:51:17 +02001656
Jens Axboeaadd06e2006-07-10 11:00:01 +02001657 /*
1658 * Don't inherit the gift flag, we need to
1659 * prevent multiple steals of this page.
1660 */
1661 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1662
1663 if (obuf->len > len)
1664 obuf->len = len;
1665
1666 opipe->nrbufs++;
1667 ret += obuf->len;
1668 len -= obuf->len;
1669 i++;
1670 } while (len);
Jens Axboe70524492006-04-11 15:51:17 +02001671
Jens Axboe02cf01a2008-02-20 10:34:51 +01001672 /*
1673 * return EAGAIN if we have the potential of some data in the
1674 * future, otherwise just return 0
1675 */
1676 if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK))
1677 ret = -EAGAIN;
1678
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001679 pipe_unlock(ipipe);
1680 pipe_unlock(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001681
Jens Axboeaadd06e2006-07-10 11:00:01 +02001682 /*
1683 * If we put data in the output pipe, wakeup any potential readers.
1684 */
Namhyung Kim825cdcb2011-05-23 19:58:53 +02001685 if (ret > 0)
1686 wakeup_pipe_readers(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001687
1688 return ret;
1689}
1690
1691/*
1692 * This is a tee(1) implementation that works on pipes. It doesn't copy
1693 * any data, it simply references the 'in' pages on the 'out' pipe.
1694 * The 'flags' used are the SPLICE_F_* variants, currently the only
1695 * applicable one is SPLICE_F_NONBLOCK.
1696 */
1697static long do_tee(struct file *in, struct file *out, size_t len,
1698 unsigned int flags)
1699{
Linus Torvalds71993e62010-11-28 13:56:09 -08001700 struct pipe_inode_info *ipipe = get_pipe_info(in);
1701 struct pipe_inode_info *opipe = get_pipe_info(out);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001702 int ret = -EINVAL;
Jens Axboe70524492006-04-11 15:51:17 +02001703
1704 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001705 * Duplicate the contents of ipipe to opipe without actually
1706 * copying the data.
Jens Axboe70524492006-04-11 15:51:17 +02001707 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001708 if (ipipe && opipe && ipipe != opipe) {
1709 /*
1710 * Keep going, unless we encounter an error. The ipipe/opipe
1711 * ordering doesn't really matter.
1712 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001713 ret = ipipe_prep(ipipe, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001714 if (!ret) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001715 ret = opipe_prep(opipe, flags);
Jens Axboe02cf01a2008-02-20 10:34:51 +01001716 if (!ret)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001717 ret = link_pipe(ipipe, opipe, len, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001718 }
1719 }
Jens Axboe70524492006-04-11 15:51:17 +02001720
Jens Axboeaadd06e2006-07-10 11:00:01 +02001721 return ret;
Jens Axboe70524492006-04-11 15:51:17 +02001722}
1723
Heiko Carstens836f92a2009-01-14 14:14:33 +01001724SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags)
Jens Axboe70524492006-04-11 15:51:17 +02001725{
Al Viro2903ff02012-08-28 12:52:22 -04001726 struct fd in;
1727 int error;
Jens Axboe70524492006-04-11 15:51:17 +02001728
Al Viro3d6ea292016-12-10 13:17:32 -05001729 if (unlikely(flags & ~SPLICE_F_ALL))
1730 return -EINVAL;
1731
Jens Axboe70524492006-04-11 15:51:17 +02001732 if (unlikely(!len))
1733 return 0;
1734
1735 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001736 in = fdget(fdin);
1737 if (in.file) {
1738 if (in.file->f_mode & FMODE_READ) {
1739 struct fd out = fdget(fdout);
1740 if (out.file) {
1741 if (out.file->f_mode & FMODE_WRITE)
1742 error = do_tee(in.file, out.file,
1743 len, flags);
1744 fdput(out);
Jens Axboe70524492006-04-11 15:51:17 +02001745 }
1746 }
Al Viro2903ff02012-08-28 12:52:22 -04001747 fdput(in);
Jens Axboe70524492006-04-11 15:51:17 +02001748 }
1749
1750 return error;
1751}